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>
94 #include <linux/prefetch.h>
96 #include <net/net_namespace.h>
97 #include <net/protocol.h>
99 #include <net/route.h>
100 #include <net/inetpeer.h>
101 #include <net/sock.h>
102 #include <net/ip_fib.h>
105 #include <net/icmp.h>
106 #include <net/xfrm.h>
107 #include <net/netevent.h>
108 #include <net/rtnetlink.h>
110 #include <linux/sysctl.h>
112 #include <net/atmclip.h>
113 #include <net/secure_seq.h>
115 #define RT_FL_TOS(oldflp4) \
116 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
118 #define IP_MAX_MTU 0xFFF0
120 #define RT_GC_TIMEOUT (300*HZ)
122 static int ip_rt_max_size
;
123 static int ip_rt_gc_timeout __read_mostly
= RT_GC_TIMEOUT
;
124 static int ip_rt_gc_interval __read_mostly
= 60 * HZ
;
125 static int ip_rt_gc_min_interval __read_mostly
= HZ
/ 2;
126 static int ip_rt_redirect_number __read_mostly
= 9;
127 static int ip_rt_redirect_load __read_mostly
= HZ
/ 50;
128 static int ip_rt_redirect_silence __read_mostly
= ((HZ
/ 50) << (9 + 1));
129 static int ip_rt_error_cost __read_mostly
= HZ
;
130 static int ip_rt_error_burst __read_mostly
= 5 * HZ
;
131 static int ip_rt_gc_elasticity __read_mostly
= 8;
132 static int ip_rt_mtu_expires __read_mostly
= 10 * 60 * HZ
;
133 static int ip_rt_min_pmtu __read_mostly
= 512 + 20 + 20;
134 static int ip_rt_min_advmss __read_mostly
= 256;
135 static int rt_chain_length_max __read_mostly
= 20;
136 static int redirect_genid
;
138 static struct delayed_work expires_work
;
139 static unsigned long expires_ljiffies
;
142 * Interface to generic destination cache.
145 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
);
146 static unsigned int ipv4_default_advmss(const struct dst_entry
*dst
);
147 static unsigned int ipv4_default_mtu(const struct dst_entry
*dst
);
148 static void ipv4_dst_destroy(struct dst_entry
*dst
);
149 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
);
150 static void ipv4_link_failure(struct sk_buff
*skb
);
151 static void ip_rt_update_pmtu(struct dst_entry
*dst
, u32 mtu
);
152 static int rt_garbage_collect(struct dst_ops
*ops
);
154 static void ipv4_dst_ifdown(struct dst_entry
*dst
, struct net_device
*dev
,
159 static u32
*ipv4_cow_metrics(struct dst_entry
*dst
, unsigned long old
)
161 struct rtable
*rt
= (struct rtable
*) dst
;
162 struct inet_peer
*peer
;
166 rt_bind_peer(rt
, rt
->rt_dst
, 1);
170 u32
*old_p
= __DST_METRICS_PTR(old
);
171 unsigned long prev
, new;
174 if (inet_metrics_new(peer
))
175 memcpy(p
, old_p
, sizeof(u32
) * RTAX_MAX
);
177 new = (unsigned long) p
;
178 prev
= cmpxchg(&dst
->_metrics
, old
, new);
181 p
= __DST_METRICS_PTR(prev
);
182 if (prev
& DST_METRICS_READ_ONLY
)
186 fib_info_put(rt
->fi
);
194 static struct neighbour
*ipv4_neigh_lookup(const struct dst_entry
*dst
, const void *daddr
);
196 static struct dst_ops ipv4_dst_ops
= {
198 .protocol
= cpu_to_be16(ETH_P_IP
),
199 .gc
= rt_garbage_collect
,
200 .check
= ipv4_dst_check
,
201 .default_advmss
= ipv4_default_advmss
,
202 .default_mtu
= ipv4_default_mtu
,
203 .cow_metrics
= ipv4_cow_metrics
,
204 .destroy
= ipv4_dst_destroy
,
205 .ifdown
= ipv4_dst_ifdown
,
206 .negative_advice
= ipv4_negative_advice
,
207 .link_failure
= ipv4_link_failure
,
208 .update_pmtu
= ip_rt_update_pmtu
,
209 .local_out
= __ip_local_out
,
210 .neigh_lookup
= ipv4_neigh_lookup
,
213 #define ECN_OR_COST(class) TC_PRIO_##class
215 const __u8 ip_tos2prio
[16] = {
217 ECN_OR_COST(BESTEFFORT
),
219 ECN_OR_COST(BESTEFFORT
),
225 ECN_OR_COST(INTERACTIVE
),
227 ECN_OR_COST(INTERACTIVE
),
228 TC_PRIO_INTERACTIVE_BULK
,
229 ECN_OR_COST(INTERACTIVE_BULK
),
230 TC_PRIO_INTERACTIVE_BULK
,
231 ECN_OR_COST(INTERACTIVE_BULK
)
239 /* The locking scheme is rather straight forward:
241 * 1) Read-Copy Update protects the buckets of the central route hash.
242 * 2) Only writers remove entries, and they hold the lock
243 * as they look at rtable reference counts.
244 * 3) Only readers acquire references to rtable entries,
245 * they do so with atomic increments and with the
249 struct rt_hash_bucket
{
250 struct rtable __rcu
*chain
;
253 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
254 defined(CONFIG_PROVE_LOCKING)
256 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
257 * The size of this table is a power of two and depends on the number of CPUS.
258 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
260 #ifdef CONFIG_LOCKDEP
261 # define RT_HASH_LOCK_SZ 256
264 # define RT_HASH_LOCK_SZ 4096
266 # define RT_HASH_LOCK_SZ 2048
268 # define RT_HASH_LOCK_SZ 1024
270 # define RT_HASH_LOCK_SZ 512
272 # define RT_HASH_LOCK_SZ 256
276 static spinlock_t
*rt_hash_locks
;
277 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
279 static __init
void rt_hash_lock_init(void)
283 rt_hash_locks
= kmalloc(sizeof(spinlock_t
) * RT_HASH_LOCK_SZ
,
286 panic("IP: failed to allocate rt_hash_locks\n");
288 for (i
= 0; i
< RT_HASH_LOCK_SZ
; i
++)
289 spin_lock_init(&rt_hash_locks
[i
]);
292 # define rt_hash_lock_addr(slot) NULL
294 static inline void rt_hash_lock_init(void)
299 static struct rt_hash_bucket
*rt_hash_table __read_mostly
;
300 static unsigned rt_hash_mask __read_mostly
;
301 static unsigned int rt_hash_log __read_mostly
;
303 static DEFINE_PER_CPU(struct rt_cache_stat
, rt_cache_stat
);
304 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
306 static inline unsigned int rt_hash(__be32 daddr
, __be32 saddr
, int idx
,
309 return jhash_3words((__force u32
)daddr
, (__force u32
)saddr
,
314 static inline int rt_genid(struct net
*net
)
316 return atomic_read(&net
->ipv4
.rt_genid
);
319 #ifdef CONFIG_PROC_FS
320 struct rt_cache_iter_state
{
321 struct seq_net_private p
;
326 static struct rtable
*rt_cache_get_first(struct seq_file
*seq
)
328 struct rt_cache_iter_state
*st
= seq
->private;
329 struct rtable
*r
= NULL
;
331 for (st
->bucket
= rt_hash_mask
; st
->bucket
>= 0; --st
->bucket
) {
332 if (!rcu_dereference_raw(rt_hash_table
[st
->bucket
].chain
))
335 r
= rcu_dereference_bh(rt_hash_table
[st
->bucket
].chain
);
337 if (dev_net(r
->dst
.dev
) == seq_file_net(seq
) &&
338 r
->rt_genid
== st
->genid
)
340 r
= rcu_dereference_bh(r
->dst
.rt_next
);
342 rcu_read_unlock_bh();
347 static struct rtable
*__rt_cache_get_next(struct seq_file
*seq
,
350 struct rt_cache_iter_state
*st
= seq
->private;
352 r
= rcu_dereference_bh(r
->dst
.rt_next
);
354 rcu_read_unlock_bh();
356 if (--st
->bucket
< 0)
358 } while (!rcu_dereference_raw(rt_hash_table
[st
->bucket
].chain
));
360 r
= rcu_dereference_bh(rt_hash_table
[st
->bucket
].chain
);
365 static struct rtable
*rt_cache_get_next(struct seq_file
*seq
,
368 struct rt_cache_iter_state
*st
= seq
->private;
369 while ((r
= __rt_cache_get_next(seq
, r
)) != NULL
) {
370 if (dev_net(r
->dst
.dev
) != seq_file_net(seq
))
372 if (r
->rt_genid
== st
->genid
)
378 static struct rtable
*rt_cache_get_idx(struct seq_file
*seq
, loff_t pos
)
380 struct rtable
*r
= rt_cache_get_first(seq
);
383 while (pos
&& (r
= rt_cache_get_next(seq
, r
)))
385 return pos
? NULL
: r
;
388 static void *rt_cache_seq_start(struct seq_file
*seq
, loff_t
*pos
)
390 struct rt_cache_iter_state
*st
= seq
->private;
392 return rt_cache_get_idx(seq
, *pos
- 1);
393 st
->genid
= rt_genid(seq_file_net(seq
));
394 return SEQ_START_TOKEN
;
397 static void *rt_cache_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
401 if (v
== SEQ_START_TOKEN
)
402 r
= rt_cache_get_first(seq
);
404 r
= rt_cache_get_next(seq
, v
);
409 static void rt_cache_seq_stop(struct seq_file
*seq
, void *v
)
411 if (v
&& v
!= SEQ_START_TOKEN
)
412 rcu_read_unlock_bh();
415 static int rt_cache_seq_show(struct seq_file
*seq
, void *v
)
417 if (v
== SEQ_START_TOKEN
)
418 seq_printf(seq
, "%-127s\n",
419 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
420 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
423 struct rtable
*r
= v
;
428 n
= dst_get_neighbour(&r
->dst
);
429 HHUptod
= (n
&& (n
->nud_state
& NUD_CONNECTED
)) ? 1 : 0;
432 seq_printf(seq
, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
433 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
434 r
->dst
.dev
? r
->dst
.dev
->name
: "*",
435 (__force u32
)r
->rt_dst
,
436 (__force u32
)r
->rt_gateway
,
437 r
->rt_flags
, atomic_read(&r
->dst
.__refcnt
),
438 r
->dst
.__use
, 0, (__force u32
)r
->rt_src
,
439 dst_metric_advmss(&r
->dst
) + 40,
440 dst_metric(&r
->dst
, RTAX_WINDOW
),
441 (int)((dst_metric(&r
->dst
, RTAX_RTT
) >> 3) +
442 dst_metric(&r
->dst
, RTAX_RTTVAR
)),
446 r
->rt_spec_dst
, &len
);
448 seq_printf(seq
, "%*s\n", 127 - len
, "");
453 static const struct seq_operations rt_cache_seq_ops
= {
454 .start
= rt_cache_seq_start
,
455 .next
= rt_cache_seq_next
,
456 .stop
= rt_cache_seq_stop
,
457 .show
= rt_cache_seq_show
,
460 static int rt_cache_seq_open(struct inode
*inode
, struct file
*file
)
462 return seq_open_net(inode
, file
, &rt_cache_seq_ops
,
463 sizeof(struct rt_cache_iter_state
));
466 static const struct file_operations rt_cache_seq_fops
= {
467 .owner
= THIS_MODULE
,
468 .open
= rt_cache_seq_open
,
471 .release
= seq_release_net
,
475 static void *rt_cpu_seq_start(struct seq_file
*seq
, loff_t
*pos
)
480 return SEQ_START_TOKEN
;
482 for (cpu
= *pos
-1; cpu
< nr_cpu_ids
; ++cpu
) {
483 if (!cpu_possible(cpu
))
486 return &per_cpu(rt_cache_stat
, cpu
);
491 static void *rt_cpu_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
495 for (cpu
= *pos
; cpu
< nr_cpu_ids
; ++cpu
) {
496 if (!cpu_possible(cpu
))
499 return &per_cpu(rt_cache_stat
, cpu
);
505 static void rt_cpu_seq_stop(struct seq_file
*seq
, void *v
)
510 static int rt_cpu_seq_show(struct seq_file
*seq
, void *v
)
512 struct rt_cache_stat
*st
= v
;
514 if (v
== SEQ_START_TOKEN
) {
515 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");
519 seq_printf(seq
,"%08x %08x %08x %08x %08x %08x %08x %08x "
520 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
521 dst_entries_get_slow(&ipv4_dst_ops
),
544 static const struct seq_operations rt_cpu_seq_ops
= {
545 .start
= rt_cpu_seq_start
,
546 .next
= rt_cpu_seq_next
,
547 .stop
= rt_cpu_seq_stop
,
548 .show
= rt_cpu_seq_show
,
552 static int rt_cpu_seq_open(struct inode
*inode
, struct file
*file
)
554 return seq_open(file
, &rt_cpu_seq_ops
);
557 static const struct file_operations rt_cpu_seq_fops
= {
558 .owner
= THIS_MODULE
,
559 .open
= rt_cpu_seq_open
,
562 .release
= seq_release
,
565 #ifdef CONFIG_IP_ROUTE_CLASSID
566 static int rt_acct_proc_show(struct seq_file
*m
, void *v
)
568 struct ip_rt_acct
*dst
, *src
;
571 dst
= kcalloc(256, sizeof(struct ip_rt_acct
), GFP_KERNEL
);
575 for_each_possible_cpu(i
) {
576 src
= (struct ip_rt_acct
*)per_cpu_ptr(ip_rt_acct
, i
);
577 for (j
= 0; j
< 256; j
++) {
578 dst
[j
].o_bytes
+= src
[j
].o_bytes
;
579 dst
[j
].o_packets
+= src
[j
].o_packets
;
580 dst
[j
].i_bytes
+= src
[j
].i_bytes
;
581 dst
[j
].i_packets
+= src
[j
].i_packets
;
585 seq_write(m
, dst
, 256 * sizeof(struct ip_rt_acct
));
590 static int rt_acct_proc_open(struct inode
*inode
, struct file
*file
)
592 return single_open(file
, rt_acct_proc_show
, NULL
);
595 static const struct file_operations rt_acct_proc_fops
= {
596 .owner
= THIS_MODULE
,
597 .open
= rt_acct_proc_open
,
600 .release
= single_release
,
604 static int __net_init
ip_rt_do_proc_init(struct net
*net
)
606 struct proc_dir_entry
*pde
;
608 pde
= proc_net_fops_create(net
, "rt_cache", S_IRUGO
,
613 pde
= proc_create("rt_cache", S_IRUGO
,
614 net
->proc_net_stat
, &rt_cpu_seq_fops
);
618 #ifdef CONFIG_IP_ROUTE_CLASSID
619 pde
= proc_create("rt_acct", 0, net
->proc_net
, &rt_acct_proc_fops
);
625 #ifdef CONFIG_IP_ROUTE_CLASSID
627 remove_proc_entry("rt_cache", net
->proc_net_stat
);
630 remove_proc_entry("rt_cache", net
->proc_net
);
635 static void __net_exit
ip_rt_do_proc_exit(struct net
*net
)
637 remove_proc_entry("rt_cache", net
->proc_net_stat
);
638 remove_proc_entry("rt_cache", net
->proc_net
);
639 #ifdef CONFIG_IP_ROUTE_CLASSID
640 remove_proc_entry("rt_acct", net
->proc_net
);
644 static struct pernet_operations ip_rt_proc_ops __net_initdata
= {
645 .init
= ip_rt_do_proc_init
,
646 .exit
= ip_rt_do_proc_exit
,
649 static int __init
ip_rt_proc_init(void)
651 return register_pernet_subsys(&ip_rt_proc_ops
);
655 static inline int ip_rt_proc_init(void)
659 #endif /* CONFIG_PROC_FS */
661 static inline void rt_free(struct rtable
*rt
)
663 call_rcu_bh(&rt
->dst
.rcu_head
, dst_rcu_free
);
666 static inline void rt_drop(struct rtable
*rt
)
669 call_rcu_bh(&rt
->dst
.rcu_head
, dst_rcu_free
);
672 static inline int rt_fast_clean(struct rtable
*rth
)
674 /* Kill broadcast/multicast entries very aggresively, if they
675 collide in hash table with more useful entries */
676 return (rth
->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) &&
677 rt_is_input_route(rth
) && rth
->dst
.rt_next
;
680 static inline int rt_valuable(struct rtable
*rth
)
682 return (rth
->rt_flags
& (RTCF_REDIRECTED
| RTCF_NOTIFY
)) ||
683 (rth
->peer
&& rth
->peer
->pmtu_expires
);
686 static int rt_may_expire(struct rtable
*rth
, unsigned long tmo1
, unsigned long tmo2
)
691 if (atomic_read(&rth
->dst
.__refcnt
))
694 age
= jiffies
- rth
->dst
.lastuse
;
695 if ((age
<= tmo1
&& !rt_fast_clean(rth
)) ||
696 (age
<= tmo2
&& rt_valuable(rth
)))
702 /* Bits of score are:
704 * 30: not quite useless
705 * 29..0: usage counter
707 static inline u32
rt_score(struct rtable
*rt
)
709 u32 score
= jiffies
- rt
->dst
.lastuse
;
711 score
= ~score
& ~(3<<30);
716 if (rt_is_output_route(rt
) ||
717 !(rt
->rt_flags
& (RTCF_BROADCAST
|RTCF_MULTICAST
|RTCF_LOCAL
)))
723 static inline bool rt_caching(const struct net
*net
)
725 return net
->ipv4
.current_rt_cache_rebuild_count
<=
726 net
->ipv4
.sysctl_rt_cache_rebuild_count
;
729 static inline bool compare_hash_inputs(const struct rtable
*rt1
,
730 const struct rtable
*rt2
)
732 return ((((__force u32
)rt1
->rt_key_dst
^ (__force u32
)rt2
->rt_key_dst
) |
733 ((__force u32
)rt1
->rt_key_src
^ (__force u32
)rt2
->rt_key_src
) |
734 (rt1
->rt_route_iif
^ rt2
->rt_route_iif
)) == 0);
737 static inline int compare_keys(struct rtable
*rt1
, struct rtable
*rt2
)
739 return (((__force u32
)rt1
->rt_key_dst
^ (__force u32
)rt2
->rt_key_dst
) |
740 ((__force u32
)rt1
->rt_key_src
^ (__force u32
)rt2
->rt_key_src
) |
741 (rt1
->rt_mark
^ rt2
->rt_mark
) |
742 (rt1
->rt_key_tos
^ rt2
->rt_key_tos
) |
743 (rt1
->rt_route_iif
^ rt2
->rt_route_iif
) |
744 (rt1
->rt_oif
^ rt2
->rt_oif
)) == 0;
747 static inline int compare_netns(struct rtable
*rt1
, struct rtable
*rt2
)
749 return net_eq(dev_net(rt1
->dst
.dev
), dev_net(rt2
->dst
.dev
));
752 static inline int rt_is_expired(struct rtable
*rth
)
754 return rth
->rt_genid
!= rt_genid(dev_net(rth
->dst
.dev
));
758 * Perform a full scan of hash table and free all entries.
759 * Can be called by a softirq or a process.
760 * In the later case, we want to be reschedule if necessary
762 static void rt_do_flush(struct net
*net
, int process_context
)
765 struct rtable
*rth
, *next
;
767 for (i
= 0; i
<= rt_hash_mask
; i
++) {
768 struct rtable __rcu
**pprev
;
771 if (process_context
&& need_resched())
773 rth
= rcu_dereference_raw(rt_hash_table
[i
].chain
);
777 spin_lock_bh(rt_hash_lock_addr(i
));
780 pprev
= &rt_hash_table
[i
].chain
;
781 rth
= rcu_dereference_protected(*pprev
,
782 lockdep_is_held(rt_hash_lock_addr(i
)));
785 next
= rcu_dereference_protected(rth
->dst
.rt_next
,
786 lockdep_is_held(rt_hash_lock_addr(i
)));
789 net_eq(dev_net(rth
->dst
.dev
), net
)) {
790 rcu_assign_pointer(*pprev
, next
);
791 rcu_assign_pointer(rth
->dst
.rt_next
, list
);
794 pprev
= &rth
->dst
.rt_next
;
799 spin_unlock_bh(rt_hash_lock_addr(i
));
801 for (; list
; list
= next
) {
802 next
= rcu_dereference_protected(list
->dst
.rt_next
, 1);
809 * While freeing expired entries, we compute average chain length
810 * and standard deviation, using fixed-point arithmetic.
811 * This to have an estimation of rt_chain_length_max
812 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
813 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
817 #define ONE (1UL << FRACT_BITS)
820 * Given a hash chain and an item in this hash chain,
821 * find if a previous entry has the same hash_inputs
822 * (but differs on tos, mark or oif)
823 * Returns 0 if an alias is found.
824 * Returns ONE if rth has no alias before itself.
826 static int has_noalias(const struct rtable
*head
, const struct rtable
*rth
)
828 const struct rtable
*aux
= head
;
831 if (compare_hash_inputs(aux
, rth
))
833 aux
= rcu_dereference_protected(aux
->dst
.rt_next
, 1);
838 static void rt_check_expire(void)
840 static unsigned int rover
;
841 unsigned int i
= rover
, goal
;
843 struct rtable __rcu
**rthp
;
844 unsigned long samples
= 0;
845 unsigned long sum
= 0, sum2
= 0;
849 delta
= jiffies
- expires_ljiffies
;
850 expires_ljiffies
= jiffies
;
851 mult
= ((u64
)delta
) << rt_hash_log
;
852 if (ip_rt_gc_timeout
> 1)
853 do_div(mult
, ip_rt_gc_timeout
);
854 goal
= (unsigned int)mult
;
855 if (goal
> rt_hash_mask
)
856 goal
= rt_hash_mask
+ 1;
857 for (; goal
> 0; goal
--) {
858 unsigned long tmo
= ip_rt_gc_timeout
;
859 unsigned long length
;
861 i
= (i
+ 1) & rt_hash_mask
;
862 rthp
= &rt_hash_table
[i
].chain
;
869 if (rcu_dereference_raw(*rthp
) == NULL
)
872 spin_lock_bh(rt_hash_lock_addr(i
));
873 while ((rth
= rcu_dereference_protected(*rthp
,
874 lockdep_is_held(rt_hash_lock_addr(i
)))) != NULL
) {
875 prefetch(rth
->dst
.rt_next
);
876 if (rt_is_expired(rth
)) {
877 *rthp
= rth
->dst
.rt_next
;
881 if (rth
->dst
.expires
) {
882 /* Entry is expired even if it is in use */
883 if (time_before_eq(jiffies
, rth
->dst
.expires
)) {
886 rthp
= &rth
->dst
.rt_next
;
888 * We only count entries on
889 * a chain with equal hash inputs once
890 * so that entries for different QOS
891 * levels, and other non-hash input
892 * attributes don't unfairly skew
893 * the length computation
895 length
+= has_noalias(rt_hash_table
[i
].chain
, rth
);
898 } else if (!rt_may_expire(rth
, tmo
, ip_rt_gc_timeout
))
901 /* Cleanup aged off entries. */
902 *rthp
= rth
->dst
.rt_next
;
905 spin_unlock_bh(rt_hash_lock_addr(i
));
907 sum2
+= length
*length
;
910 unsigned long avg
= sum
/ samples
;
911 unsigned long sd
= int_sqrt(sum2
/ samples
- avg
*avg
);
912 rt_chain_length_max
= max_t(unsigned long,
914 (avg
+ 4*sd
) >> FRACT_BITS
);
920 * rt_worker_func() is run in process context.
921 * we call rt_check_expire() to scan part of the hash table
923 static void rt_worker_func(struct work_struct
*work
)
926 schedule_delayed_work(&expires_work
, ip_rt_gc_interval
);
930 * Perturbation of rt_genid by a small quantity [1..256]
931 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
932 * many times (2^24) without giving recent rt_genid.
933 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
935 static void rt_cache_invalidate(struct net
*net
)
937 unsigned char shuffle
;
939 get_random_bytes(&shuffle
, sizeof(shuffle
));
940 atomic_add(shuffle
+ 1U, &net
->ipv4
.rt_genid
);
945 * delay < 0 : invalidate cache (fast : entries will be deleted later)
946 * delay >= 0 : invalidate & flush cache (can be long)
948 void rt_cache_flush(struct net
*net
, int delay
)
950 rt_cache_invalidate(net
);
952 rt_do_flush(net
, !in_softirq());
955 /* Flush previous cache invalidated entries from the cache */
956 void rt_cache_flush_batch(struct net
*net
)
958 rt_do_flush(net
, !in_softirq());
961 static void rt_emergency_hash_rebuild(struct net
*net
)
964 printk(KERN_WARNING
"Route hash chain too long!\n");
965 rt_cache_invalidate(net
);
969 Short description of GC goals.
971 We want to build algorithm, which will keep routing cache
972 at some equilibrium point, when number of aged off entries
973 is kept approximately equal to newly generated ones.
975 Current expiration strength is variable "expire".
976 We try to adjust it dynamically, so that if networking
977 is idle expires is large enough to keep enough of warm entries,
978 and when load increases it reduces to limit cache size.
981 static int rt_garbage_collect(struct dst_ops
*ops
)
983 static unsigned long expire
= RT_GC_TIMEOUT
;
984 static unsigned long last_gc
;
986 static int equilibrium
;
988 struct rtable __rcu
**rthp
;
989 unsigned long now
= jiffies
;
991 int entries
= dst_entries_get_fast(&ipv4_dst_ops
);
994 * Garbage collection is pretty expensive,
995 * do not make it too frequently.
998 RT_CACHE_STAT_INC(gc_total
);
1000 if (now
- last_gc
< ip_rt_gc_min_interval
&&
1001 entries
< ip_rt_max_size
) {
1002 RT_CACHE_STAT_INC(gc_ignored
);
1006 entries
= dst_entries_get_slow(&ipv4_dst_ops
);
1007 /* Calculate number of entries, which we want to expire now. */
1008 goal
= entries
- (ip_rt_gc_elasticity
<< rt_hash_log
);
1010 if (equilibrium
< ipv4_dst_ops
.gc_thresh
)
1011 equilibrium
= ipv4_dst_ops
.gc_thresh
;
1012 goal
= entries
- equilibrium
;
1014 equilibrium
+= min_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
1015 goal
= entries
- equilibrium
;
1018 /* We are in dangerous area. Try to reduce cache really
1021 goal
= max_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
1022 equilibrium
= entries
- goal
;
1025 if (now
- last_gc
>= ip_rt_gc_min_interval
)
1029 equilibrium
+= goal
;
1036 for (i
= rt_hash_mask
, k
= rover
; i
>= 0; i
--) {
1037 unsigned long tmo
= expire
;
1039 k
= (k
+ 1) & rt_hash_mask
;
1040 rthp
= &rt_hash_table
[k
].chain
;
1041 spin_lock_bh(rt_hash_lock_addr(k
));
1042 while ((rth
= rcu_dereference_protected(*rthp
,
1043 lockdep_is_held(rt_hash_lock_addr(k
)))) != NULL
) {
1044 if (!rt_is_expired(rth
) &&
1045 !rt_may_expire(rth
, tmo
, expire
)) {
1047 rthp
= &rth
->dst
.rt_next
;
1050 *rthp
= rth
->dst
.rt_next
;
1054 spin_unlock_bh(rt_hash_lock_addr(k
));
1063 /* Goal is not achieved. We stop process if:
1065 - if expire reduced to zero. Otherwise, expire is halfed.
1066 - if table is not full.
1067 - if we are called from interrupt.
1068 - jiffies check is just fallback/debug loop breaker.
1069 We will not spin here for long time in any case.
1072 RT_CACHE_STAT_INC(gc_goal_miss
);
1079 if (dst_entries_get_fast(&ipv4_dst_ops
) < ip_rt_max_size
)
1081 } while (!in_softirq() && time_before_eq(jiffies
, now
));
1083 if (dst_entries_get_fast(&ipv4_dst_ops
) < ip_rt_max_size
)
1085 if (dst_entries_get_slow(&ipv4_dst_ops
) < ip_rt_max_size
)
1087 if (net_ratelimit())
1088 printk(KERN_WARNING
"dst cache overflow\n");
1089 RT_CACHE_STAT_INC(gc_dst_overflow
);
1093 expire
+= ip_rt_gc_min_interval
;
1094 if (expire
> ip_rt_gc_timeout
||
1095 dst_entries_get_fast(&ipv4_dst_ops
) < ipv4_dst_ops
.gc_thresh
||
1096 dst_entries_get_slow(&ipv4_dst_ops
) < ipv4_dst_ops
.gc_thresh
)
1097 expire
= ip_rt_gc_timeout
;
1102 * Returns number of entries in a hash chain that have different hash_inputs
1104 static int slow_chain_length(const struct rtable
*head
)
1107 const struct rtable
*rth
= head
;
1110 length
+= has_noalias(head
, rth
);
1111 rth
= rcu_dereference_protected(rth
->dst
.rt_next
, 1);
1113 return length
>> FRACT_BITS
;
1116 static struct neighbour
*ipv4_neigh_lookup(const struct dst_entry
*dst
, const void *daddr
)
1118 struct neigh_table
*tbl
= &arp_tbl
;
1119 static const __be32 inaddr_any
= 0;
1120 struct net_device
*dev
= dst
->dev
;
1121 const __be32
*pkey
= daddr
;
1122 struct neighbour
*n
;
1124 #if defined(CONFIG_ATM_CLIP) || defined(CONFIG_ATM_CLIP_MODULE)
1125 if (dev
->type
== ARPHRD_ATM
)
1126 tbl
= clip_tbl_hook
;
1128 if (dev
->flags
& (IFF_LOOPBACK
| IFF_POINTOPOINT
))
1131 n
= __ipv4_neigh_lookup(tbl
, dev
, *(__force u32
*)pkey
);
1134 return neigh_create(tbl
, pkey
, dev
);
1137 static int rt_bind_neighbour(struct rtable
*rt
)
1139 struct neighbour
*n
= ipv4_neigh_lookup(&rt
->dst
, &rt
->rt_gateway
);
1142 dst_set_neighbour(&rt
->dst
, n
);
1147 static struct rtable
*rt_intern_hash(unsigned hash
, struct rtable
*rt
,
1148 struct sk_buff
*skb
, int ifindex
)
1150 struct rtable
*rth
, *cand
;
1151 struct rtable __rcu
**rthp
, **candp
;
1155 int attempts
= !in_softirq();
1159 min_score
= ~(u32
)0;
1164 if (!rt_caching(dev_net(rt
->dst
.dev
))) {
1166 * If we're not caching, just tell the caller we
1167 * were successful and don't touch the route. The
1168 * caller hold the sole reference to the cache entry, and
1169 * it will be released when the caller is done with it.
1170 * If we drop it here, the callers have no way to resolve routes
1171 * when we're not caching. Instead, just point *rp at rt, so
1172 * the caller gets a single use out of the route
1173 * Note that we do rt_free on this new route entry, so that
1174 * once its refcount hits zero, we are still able to reap it
1176 * Note: To avoid expensive rcu stuff for this uncached dst,
1177 * we set DST_NOCACHE so that dst_release() can free dst without
1178 * waiting a grace period.
1181 rt
->dst
.flags
|= DST_NOCACHE
;
1182 if (rt
->rt_type
== RTN_UNICAST
|| rt_is_output_route(rt
)) {
1183 int err
= rt_bind_neighbour(rt
);
1185 if (net_ratelimit())
1187 "Neighbour table failure & not caching routes.\n");
1189 return ERR_PTR(err
);
1196 rthp
= &rt_hash_table
[hash
].chain
;
1198 spin_lock_bh(rt_hash_lock_addr(hash
));
1199 while ((rth
= rcu_dereference_protected(*rthp
,
1200 lockdep_is_held(rt_hash_lock_addr(hash
)))) != NULL
) {
1201 if (rt_is_expired(rth
)) {
1202 *rthp
= rth
->dst
.rt_next
;
1206 if (compare_keys(rth
, rt
) && compare_netns(rth
, rt
)) {
1208 *rthp
= rth
->dst
.rt_next
;
1210 * Since lookup is lockfree, the deletion
1211 * must be visible to another weakly ordered CPU before
1212 * the insertion at the start of the hash chain.
1214 rcu_assign_pointer(rth
->dst
.rt_next
,
1215 rt_hash_table
[hash
].chain
);
1217 * Since lookup is lockfree, the update writes
1218 * must be ordered for consistency on SMP.
1220 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rth
);
1222 dst_use(&rth
->dst
, now
);
1223 spin_unlock_bh(rt_hash_lock_addr(hash
));
1227 skb_dst_set(skb
, &rth
->dst
);
1231 if (!atomic_read(&rth
->dst
.__refcnt
)) {
1232 u32 score
= rt_score(rth
);
1234 if (score
<= min_score
) {
1243 rthp
= &rth
->dst
.rt_next
;
1247 /* ip_rt_gc_elasticity used to be average length of chain
1248 * length, when exceeded gc becomes really aggressive.
1250 * The second limit is less certain. At the moment it allows
1251 * only 2 entries per bucket. We will see.
1253 if (chain_length
> ip_rt_gc_elasticity
) {
1254 *candp
= cand
->dst
.rt_next
;
1258 if (chain_length
> rt_chain_length_max
&&
1259 slow_chain_length(rt_hash_table
[hash
].chain
) > rt_chain_length_max
) {
1260 struct net
*net
= dev_net(rt
->dst
.dev
);
1261 int num
= ++net
->ipv4
.current_rt_cache_rebuild_count
;
1262 if (!rt_caching(net
)) {
1263 printk(KERN_WARNING
"%s: %d rebuilds is over limit, route caching disabled\n",
1264 rt
->dst
.dev
->name
, num
);
1266 rt_emergency_hash_rebuild(net
);
1267 spin_unlock_bh(rt_hash_lock_addr(hash
));
1269 hash
= rt_hash(rt
->rt_key_dst
, rt
->rt_key_src
,
1270 ifindex
, rt_genid(net
));
1275 /* Try to bind route to arp only if it is output
1276 route or unicast forwarding path.
1278 if (rt
->rt_type
== RTN_UNICAST
|| rt_is_output_route(rt
)) {
1279 int err
= rt_bind_neighbour(rt
);
1281 spin_unlock_bh(rt_hash_lock_addr(hash
));
1283 if (err
!= -ENOBUFS
) {
1285 return ERR_PTR(err
);
1288 /* Neighbour tables are full and nothing
1289 can be released. Try to shrink route cache,
1290 it is most likely it holds some neighbour records.
1292 if (attempts
-- > 0) {
1293 int saved_elasticity
= ip_rt_gc_elasticity
;
1294 int saved_int
= ip_rt_gc_min_interval
;
1295 ip_rt_gc_elasticity
= 1;
1296 ip_rt_gc_min_interval
= 0;
1297 rt_garbage_collect(&ipv4_dst_ops
);
1298 ip_rt_gc_min_interval
= saved_int
;
1299 ip_rt_gc_elasticity
= saved_elasticity
;
1303 if (net_ratelimit())
1304 printk(KERN_WARNING
"ipv4: Neighbour table overflow.\n");
1306 return ERR_PTR(-ENOBUFS
);
1310 rt
->dst
.rt_next
= rt_hash_table
[hash
].chain
;
1313 * Since lookup is lockfree, we must make sure
1314 * previous writes to rt are committed to memory
1315 * before making rt visible to other CPUS.
1317 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rt
);
1319 spin_unlock_bh(rt_hash_lock_addr(hash
));
1323 skb_dst_set(skb
, &rt
->dst
);
1327 static atomic_t __rt_peer_genid
= ATOMIC_INIT(0);
1329 static u32
rt_peer_genid(void)
1331 return atomic_read(&__rt_peer_genid
);
1334 void rt_bind_peer(struct rtable
*rt
, __be32 daddr
, int create
)
1336 struct inet_peer
*peer
;
1338 peer
= inet_getpeer_v4(daddr
, create
);
1340 if (peer
&& cmpxchg(&rt
->peer
, NULL
, peer
) != NULL
)
1343 rt
->rt_peer_genid
= rt_peer_genid();
1347 * Peer allocation may fail only in serious out-of-memory conditions. However
1348 * we still can generate some output.
1349 * Random ID selection looks a bit dangerous because we have no chances to
1350 * select ID being unique in a reasonable period of time.
1351 * But broken packet identifier may be better than no packet at all.
1353 static void ip_select_fb_ident(struct iphdr
*iph
)
1355 static DEFINE_SPINLOCK(ip_fb_id_lock
);
1356 static u32 ip_fallback_id
;
1359 spin_lock_bh(&ip_fb_id_lock
);
1360 salt
= secure_ip_id((__force __be32
)ip_fallback_id
^ iph
->daddr
);
1361 iph
->id
= htons(salt
& 0xFFFF);
1362 ip_fallback_id
= salt
;
1363 spin_unlock_bh(&ip_fb_id_lock
);
1366 void __ip_select_ident(struct iphdr
*iph
, struct dst_entry
*dst
, int more
)
1368 struct rtable
*rt
= (struct rtable
*) dst
;
1370 if (rt
&& !(rt
->dst
.flags
& DST_NOPEER
)) {
1371 if (rt
->peer
== NULL
)
1372 rt_bind_peer(rt
, rt
->rt_dst
, 1);
1374 /* If peer is attached to destination, it is never detached,
1375 so that we need not to grab a lock to dereference it.
1378 iph
->id
= htons(inet_getid(rt
->peer
, more
));
1382 printk(KERN_DEBUG
"rt_bind_peer(0) @%p\n",
1383 __builtin_return_address(0));
1385 ip_select_fb_ident(iph
);
1387 EXPORT_SYMBOL(__ip_select_ident
);
1389 static void rt_del(unsigned hash
, struct rtable
*rt
)
1391 struct rtable __rcu
**rthp
;
1394 rthp
= &rt_hash_table
[hash
].chain
;
1395 spin_lock_bh(rt_hash_lock_addr(hash
));
1397 while ((aux
= rcu_dereference_protected(*rthp
,
1398 lockdep_is_held(rt_hash_lock_addr(hash
)))) != NULL
) {
1399 if (aux
== rt
|| rt_is_expired(aux
)) {
1400 *rthp
= aux
->dst
.rt_next
;
1404 rthp
= &aux
->dst
.rt_next
;
1406 spin_unlock_bh(rt_hash_lock_addr(hash
));
1409 static void check_peer_redir(struct dst_entry
*dst
, struct inet_peer
*peer
)
1411 struct rtable
*rt
= (struct rtable
*) dst
;
1412 __be32 orig_gw
= rt
->rt_gateway
;
1413 struct neighbour
*n
, *old_n
;
1415 dst_confirm(&rt
->dst
);
1417 rt
->rt_gateway
= peer
->redirect_learned
.a4
;
1419 n
= ipv4_neigh_lookup(&rt
->dst
, &rt
->rt_gateway
);
1421 rt
->rt_gateway
= orig_gw
;
1424 old_n
= xchg(&rt
->dst
._neighbour
, n
);
1426 neigh_release(old_n
);
1427 if (!(n
->nud_state
& NUD_VALID
)) {
1428 neigh_event_send(n
, NULL
);
1430 rt
->rt_flags
|= RTCF_REDIRECTED
;
1431 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE
, n
);
1435 /* called in rcu_read_lock() section */
1436 void ip_rt_redirect(__be32 old_gw
, __be32 daddr
, __be32 new_gw
,
1437 __be32 saddr
, struct net_device
*dev
)
1440 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
1441 __be32 skeys
[2] = { saddr
, 0 };
1442 int ikeys
[2] = { dev
->ifindex
, 0 };
1443 struct inet_peer
*peer
;
1450 if (new_gw
== old_gw
|| !IN_DEV_RX_REDIRECTS(in_dev
) ||
1451 ipv4_is_multicast(new_gw
) || ipv4_is_lbcast(new_gw
) ||
1452 ipv4_is_zeronet(new_gw
))
1453 goto reject_redirect
;
1455 if (!IN_DEV_SHARED_MEDIA(in_dev
)) {
1456 if (!inet_addr_onlink(in_dev
, new_gw
, old_gw
))
1457 goto reject_redirect
;
1458 if (IN_DEV_SEC_REDIRECTS(in_dev
) && ip_fib_check_default(new_gw
, dev
))
1459 goto reject_redirect
;
1461 if (inet_addr_type(net
, new_gw
) != RTN_UNICAST
)
1462 goto reject_redirect
;
1465 for (s
= 0; s
< 2; s
++) {
1466 for (i
= 0; i
< 2; i
++) {
1468 struct rtable __rcu
**rthp
;
1471 hash
= rt_hash(daddr
, skeys
[s
], ikeys
[i
], rt_genid(net
));
1473 rthp
= &rt_hash_table
[hash
].chain
;
1475 while ((rt
= rcu_dereference(*rthp
)) != NULL
) {
1476 rthp
= &rt
->dst
.rt_next
;
1478 if (rt
->rt_key_dst
!= daddr
||
1479 rt
->rt_key_src
!= skeys
[s
] ||
1480 rt
->rt_oif
!= ikeys
[i
] ||
1481 rt_is_input_route(rt
) ||
1482 rt_is_expired(rt
) ||
1483 !net_eq(dev_net(rt
->dst
.dev
), net
) ||
1485 rt
->dst
.dev
!= dev
||
1486 rt
->rt_gateway
!= old_gw
)
1490 rt_bind_peer(rt
, rt
->rt_dst
, 1);
1494 if (peer
->redirect_learned
.a4
!= new_gw
||
1495 peer
->redirect_genid
!= redirect_genid
) {
1496 peer
->redirect_learned
.a4
= new_gw
;
1497 peer
->redirect_genid
= redirect_genid
;
1498 atomic_inc(&__rt_peer_genid
);
1500 check_peer_redir(&rt
->dst
, peer
);
1508 #ifdef CONFIG_IP_ROUTE_VERBOSE
1509 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit())
1510 printk(KERN_INFO
"Redirect from %pI4 on %s about %pI4 ignored.\n"
1511 " Advised path = %pI4 -> %pI4\n",
1512 &old_gw
, dev
->name
, &new_gw
,
1518 static bool peer_pmtu_expired(struct inet_peer
*peer
)
1520 unsigned long orig
= ACCESS_ONCE(peer
->pmtu_expires
);
1523 time_after_eq(jiffies
, orig
) &&
1524 cmpxchg(&peer
->pmtu_expires
, orig
, 0) == orig
;
1527 static bool peer_pmtu_cleaned(struct inet_peer
*peer
)
1529 unsigned long orig
= ACCESS_ONCE(peer
->pmtu_expires
);
1532 cmpxchg(&peer
->pmtu_expires
, orig
, 0) == orig
;
1535 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
)
1537 struct rtable
*rt
= (struct rtable
*)dst
;
1538 struct dst_entry
*ret
= dst
;
1541 if (dst
->obsolete
> 0) {
1544 } else if (rt
->rt_flags
& RTCF_REDIRECTED
) {
1545 unsigned hash
= rt_hash(rt
->rt_key_dst
, rt
->rt_key_src
,
1547 rt_genid(dev_net(dst
->dev
)));
1550 } else if (rt
->peer
&& peer_pmtu_expired(rt
->peer
)) {
1551 dst_metric_set(dst
, RTAX_MTU
, rt
->peer
->pmtu_orig
);
1559 * 1. The first ip_rt_redirect_number redirects are sent
1560 * with exponential backoff, then we stop sending them at all,
1561 * assuming that the host ignores our redirects.
1562 * 2. If we did not see packets requiring redirects
1563 * during ip_rt_redirect_silence, we assume that the host
1564 * forgot redirected route and start to send redirects again.
1566 * This algorithm is much cheaper and more intelligent than dumb load limiting
1569 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1570 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1573 void ip_rt_send_redirect(struct sk_buff
*skb
)
1575 struct rtable
*rt
= skb_rtable(skb
);
1576 struct in_device
*in_dev
;
1577 struct inet_peer
*peer
;
1581 in_dev
= __in_dev_get_rcu(rt
->dst
.dev
);
1582 if (!in_dev
|| !IN_DEV_TX_REDIRECTS(in_dev
)) {
1586 log_martians
= IN_DEV_LOG_MARTIANS(in_dev
);
1590 rt_bind_peer(rt
, rt
->rt_dst
, 1);
1593 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
, rt
->rt_gateway
);
1597 /* No redirected packets during ip_rt_redirect_silence;
1598 * reset the algorithm.
1600 if (time_after(jiffies
, peer
->rate_last
+ ip_rt_redirect_silence
))
1601 peer
->rate_tokens
= 0;
1603 /* Too many ignored redirects; do not send anything
1604 * set dst.rate_last to the last seen redirected packet.
1606 if (peer
->rate_tokens
>= ip_rt_redirect_number
) {
1607 peer
->rate_last
= jiffies
;
1611 /* Check for load limit; set rate_last to the latest sent
1614 if (peer
->rate_tokens
== 0 ||
1617 (ip_rt_redirect_load
<< peer
->rate_tokens
)))) {
1618 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
, rt
->rt_gateway
);
1619 peer
->rate_last
= jiffies
;
1620 ++peer
->rate_tokens
;
1621 #ifdef CONFIG_IP_ROUTE_VERBOSE
1623 peer
->rate_tokens
== ip_rt_redirect_number
&&
1625 printk(KERN_WARNING
"host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1626 &ip_hdr(skb
)->saddr
, rt
->rt_iif
,
1627 &rt
->rt_dst
, &rt
->rt_gateway
);
1632 static int ip_error(struct sk_buff
*skb
)
1634 struct rtable
*rt
= skb_rtable(skb
);
1635 struct inet_peer
*peer
;
1640 switch (rt
->dst
.error
) {
1645 code
= ICMP_HOST_UNREACH
;
1648 code
= ICMP_NET_UNREACH
;
1649 IP_INC_STATS_BH(dev_net(rt
->dst
.dev
),
1650 IPSTATS_MIB_INNOROUTES
);
1653 code
= ICMP_PKT_FILTERED
;
1658 rt_bind_peer(rt
, rt
->rt_dst
, 1);
1664 peer
->rate_tokens
+= now
- peer
->rate_last
;
1665 if (peer
->rate_tokens
> ip_rt_error_burst
)
1666 peer
->rate_tokens
= ip_rt_error_burst
;
1667 peer
->rate_last
= now
;
1668 if (peer
->rate_tokens
>= ip_rt_error_cost
)
1669 peer
->rate_tokens
-= ip_rt_error_cost
;
1674 icmp_send(skb
, ICMP_DEST_UNREACH
, code
, 0);
1676 out
: kfree_skb(skb
);
1681 * The last two values are not from the RFC but
1682 * are needed for AMPRnet AX.25 paths.
1685 static const unsigned short mtu_plateau
[] =
1686 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1688 static inline unsigned short guess_mtu(unsigned short old_mtu
)
1692 for (i
= 0; i
< ARRAY_SIZE(mtu_plateau
); i
++)
1693 if (old_mtu
> mtu_plateau
[i
])
1694 return mtu_plateau
[i
];
1698 unsigned short ip_rt_frag_needed(struct net
*net
, const struct iphdr
*iph
,
1699 unsigned short new_mtu
,
1700 struct net_device
*dev
)
1702 unsigned short old_mtu
= ntohs(iph
->tot_len
);
1703 unsigned short est_mtu
= 0;
1704 struct inet_peer
*peer
;
1706 peer
= inet_getpeer_v4(iph
->daddr
, 1);
1708 unsigned short mtu
= new_mtu
;
1710 if (new_mtu
< 68 || new_mtu
>= old_mtu
) {
1711 /* BSD 4.2 derived systems incorrectly adjust
1712 * tot_len by the IP header length, and report
1713 * a zero MTU in the ICMP message.
1716 old_mtu
>= 68 + (iph
->ihl
<< 2))
1717 old_mtu
-= iph
->ihl
<< 2;
1718 mtu
= guess_mtu(old_mtu
);
1721 if (mtu
< ip_rt_min_pmtu
)
1722 mtu
= ip_rt_min_pmtu
;
1723 if (!peer
->pmtu_expires
|| mtu
< peer
->pmtu_learned
) {
1724 unsigned long pmtu_expires
;
1726 pmtu_expires
= jiffies
+ ip_rt_mtu_expires
;
1731 peer
->pmtu_learned
= mtu
;
1732 peer
->pmtu_expires
= pmtu_expires
;
1733 atomic_inc(&__rt_peer_genid
);
1738 return est_mtu
? : new_mtu
;
1741 static void check_peer_pmtu(struct dst_entry
*dst
, struct inet_peer
*peer
)
1743 unsigned long expires
= ACCESS_ONCE(peer
->pmtu_expires
);
1747 if (time_before(jiffies
, expires
)) {
1748 u32 orig_dst_mtu
= dst_mtu(dst
);
1749 if (peer
->pmtu_learned
< orig_dst_mtu
) {
1750 if (!peer
->pmtu_orig
)
1751 peer
->pmtu_orig
= dst_metric_raw(dst
, RTAX_MTU
);
1752 dst_metric_set(dst
, RTAX_MTU
, peer
->pmtu_learned
);
1754 } else if (cmpxchg(&peer
->pmtu_expires
, expires
, 0) == expires
)
1755 dst_metric_set(dst
, RTAX_MTU
, peer
->pmtu_orig
);
1758 static void ip_rt_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
1760 struct rtable
*rt
= (struct rtable
*) dst
;
1761 struct inet_peer
*peer
;
1766 rt_bind_peer(rt
, rt
->rt_dst
, 1);
1769 unsigned long pmtu_expires
= ACCESS_ONCE(peer
->pmtu_expires
);
1771 if (mtu
< ip_rt_min_pmtu
)
1772 mtu
= ip_rt_min_pmtu
;
1773 if (!pmtu_expires
|| mtu
< peer
->pmtu_learned
) {
1775 pmtu_expires
= jiffies
+ ip_rt_mtu_expires
;
1779 peer
->pmtu_learned
= mtu
;
1780 peer
->pmtu_expires
= pmtu_expires
;
1782 atomic_inc(&__rt_peer_genid
);
1783 rt
->rt_peer_genid
= rt_peer_genid();
1785 check_peer_pmtu(dst
, peer
);
1790 static void ipv4_validate_peer(struct rtable
*rt
)
1792 if (rt
->rt_peer_genid
!= rt_peer_genid()) {
1793 struct inet_peer
*peer
;
1796 rt_bind_peer(rt
, rt
->rt_dst
, 0);
1800 check_peer_pmtu(&rt
->dst
, peer
);
1802 if (peer
->redirect_genid
!= redirect_genid
)
1803 peer
->redirect_learned
.a4
= 0;
1804 if (peer
->redirect_learned
.a4
&&
1805 peer
->redirect_learned
.a4
!= rt
->rt_gateway
)
1806 check_peer_redir(&rt
->dst
, peer
);
1809 rt
->rt_peer_genid
= rt_peer_genid();
1813 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
)
1815 struct rtable
*rt
= (struct rtable
*) dst
;
1817 if (rt_is_expired(rt
))
1819 ipv4_validate_peer(rt
);
1823 static void ipv4_dst_destroy(struct dst_entry
*dst
)
1825 struct rtable
*rt
= (struct rtable
*) dst
;
1826 struct inet_peer
*peer
= rt
->peer
;
1829 fib_info_put(rt
->fi
);
1839 static void ipv4_link_failure(struct sk_buff
*skb
)
1843 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_HOST_UNREACH
, 0);
1845 rt
= skb_rtable(skb
);
1846 if (rt
&& rt
->peer
&& peer_pmtu_cleaned(rt
->peer
))
1847 dst_metric_set(&rt
->dst
, RTAX_MTU
, rt
->peer
->pmtu_orig
);
1850 static int ip_rt_bug(struct sk_buff
*skb
)
1852 printk(KERN_DEBUG
"ip_rt_bug: %pI4 -> %pI4, %s\n",
1853 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1854 skb
->dev
? skb
->dev
->name
: "?");
1861 We do not cache source address of outgoing interface,
1862 because it is used only by IP RR, TS and SRR options,
1863 so that it out of fast path.
1865 BTW remember: "addr" is allowed to be not aligned
1869 void ip_rt_get_source(u8
*addr
, struct sk_buff
*skb
, struct rtable
*rt
)
1873 if (rt_is_output_route(rt
))
1874 src
= ip_hdr(skb
)->saddr
;
1876 struct fib_result res
;
1882 memset(&fl4
, 0, sizeof(fl4
));
1883 fl4
.daddr
= iph
->daddr
;
1884 fl4
.saddr
= iph
->saddr
;
1885 fl4
.flowi4_tos
= RT_TOS(iph
->tos
);
1886 fl4
.flowi4_oif
= rt
->dst
.dev
->ifindex
;
1887 fl4
.flowi4_iif
= skb
->dev
->ifindex
;
1888 fl4
.flowi4_mark
= skb
->mark
;
1891 if (fib_lookup(dev_net(rt
->dst
.dev
), &fl4
, &res
) == 0)
1892 src
= FIB_RES_PREFSRC(dev_net(rt
->dst
.dev
), res
);
1894 src
= inet_select_addr(rt
->dst
.dev
, rt
->rt_gateway
,
1898 memcpy(addr
, &src
, 4);
1901 #ifdef CONFIG_IP_ROUTE_CLASSID
1902 static void set_class_tag(struct rtable
*rt
, u32 tag
)
1904 if (!(rt
->dst
.tclassid
& 0xFFFF))
1905 rt
->dst
.tclassid
|= tag
& 0xFFFF;
1906 if (!(rt
->dst
.tclassid
& 0xFFFF0000))
1907 rt
->dst
.tclassid
|= tag
& 0xFFFF0000;
1911 static unsigned int ipv4_default_advmss(const struct dst_entry
*dst
)
1913 unsigned int advmss
= dst_metric_raw(dst
, RTAX_ADVMSS
);
1916 advmss
= max_t(unsigned int, dst
->dev
->mtu
- 40,
1918 if (advmss
> 65535 - 40)
1919 advmss
= 65535 - 40;
1924 static unsigned int ipv4_default_mtu(const struct dst_entry
*dst
)
1926 unsigned int mtu
= dst
->dev
->mtu
;
1928 if (unlikely(dst_metric_locked(dst
, RTAX_MTU
))) {
1929 const struct rtable
*rt
= (const struct rtable
*) dst
;
1931 if (rt
->rt_gateway
!= rt
->rt_dst
&& mtu
> 576)
1935 if (mtu
> IP_MAX_MTU
)
1941 static void rt_init_metrics(struct rtable
*rt
, const struct flowi4
*fl4
,
1942 struct fib_info
*fi
)
1944 struct inet_peer
*peer
;
1947 /* If a peer entry exists for this destination, we must hook
1948 * it up in order to get at cached metrics.
1950 if (fl4
&& (fl4
->flowi4_flags
& FLOWI_FLAG_PRECOW_METRICS
))
1953 rt
->peer
= peer
= inet_getpeer_v4(rt
->rt_dst
, create
);
1955 rt
->rt_peer_genid
= rt_peer_genid();
1956 if (inet_metrics_new(peer
))
1957 memcpy(peer
->metrics
, fi
->fib_metrics
,
1958 sizeof(u32
) * RTAX_MAX
);
1959 dst_init_metrics(&rt
->dst
, peer
->metrics
, false);
1961 check_peer_pmtu(&rt
->dst
, peer
);
1962 if (peer
->redirect_genid
!= redirect_genid
)
1963 peer
->redirect_learned
.a4
= 0;
1964 if (peer
->redirect_learned
.a4
&&
1965 peer
->redirect_learned
.a4
!= rt
->rt_gateway
) {
1966 rt
->rt_gateway
= peer
->redirect_learned
.a4
;
1967 rt
->rt_flags
|= RTCF_REDIRECTED
;
1970 if (fi
->fib_metrics
!= (u32
*) dst_default_metrics
) {
1972 atomic_inc(&fi
->fib_clntref
);
1974 dst_init_metrics(&rt
->dst
, fi
->fib_metrics
, true);
1978 static void rt_set_nexthop(struct rtable
*rt
, const struct flowi4
*fl4
,
1979 const struct fib_result
*res
,
1980 struct fib_info
*fi
, u16 type
, u32 itag
)
1982 struct dst_entry
*dst
= &rt
->dst
;
1985 if (FIB_RES_GW(*res
) &&
1986 FIB_RES_NH(*res
).nh_scope
== RT_SCOPE_LINK
)
1987 rt
->rt_gateway
= FIB_RES_GW(*res
);
1988 rt_init_metrics(rt
, fl4
, fi
);
1989 #ifdef CONFIG_IP_ROUTE_CLASSID
1990 dst
->tclassid
= FIB_RES_NH(*res
).nh_tclassid
;
1994 if (dst_mtu(dst
) > IP_MAX_MTU
)
1995 dst_metric_set(dst
, RTAX_MTU
, IP_MAX_MTU
);
1996 if (dst_metric_raw(dst
, RTAX_ADVMSS
) > 65535 - 40)
1997 dst_metric_set(dst
, RTAX_ADVMSS
, 65535 - 40);
1999 #ifdef CONFIG_IP_ROUTE_CLASSID
2000 #ifdef CONFIG_IP_MULTIPLE_TABLES
2001 set_class_tag(rt
, fib_rules_tclass(res
));
2003 set_class_tag(rt
, itag
);
2007 static struct rtable
*rt_dst_alloc(struct net_device
*dev
,
2008 bool nopolicy
, bool noxfrm
)
2010 return dst_alloc(&ipv4_dst_ops
, dev
, 1, -1,
2012 (nopolicy
? DST_NOPOLICY
: 0) |
2013 (noxfrm
? DST_NOXFRM
: 0));
2016 /* called in rcu_read_lock() section */
2017 static int ip_route_input_mc(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2018 u8 tos
, struct net_device
*dev
, int our
)
2023 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
2027 /* Primary sanity checks. */
2032 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
2033 ipv4_is_loopback(saddr
) || skb
->protocol
!= htons(ETH_P_IP
))
2036 if (ipv4_is_zeronet(saddr
)) {
2037 if (!ipv4_is_local_multicast(daddr
))
2039 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
2041 err
= fib_validate_source(skb
, saddr
, 0, tos
, 0, dev
, &spec_dst
,
2046 rth
= rt_dst_alloc(init_net
.loopback_dev
,
2047 IN_DEV_CONF_GET(in_dev
, NOPOLICY
), false);
2051 #ifdef CONFIG_IP_ROUTE_CLASSID
2052 rth
->dst
.tclassid
= itag
;
2054 rth
->dst
.output
= ip_rt_bug
;
2056 rth
->rt_key_dst
= daddr
;
2057 rth
->rt_key_src
= saddr
;
2058 rth
->rt_genid
= rt_genid(dev_net(dev
));
2059 rth
->rt_flags
= RTCF_MULTICAST
;
2060 rth
->rt_type
= RTN_MULTICAST
;
2061 rth
->rt_key_tos
= tos
;
2062 rth
->rt_dst
= daddr
;
2063 rth
->rt_src
= saddr
;
2064 rth
->rt_route_iif
= dev
->ifindex
;
2065 rth
->rt_iif
= dev
->ifindex
;
2067 rth
->rt_mark
= skb
->mark
;
2068 rth
->rt_gateway
= daddr
;
2069 rth
->rt_spec_dst
= spec_dst
;
2070 rth
->rt_peer_genid
= 0;
2074 rth
->dst
.input
= ip_local_deliver
;
2075 rth
->rt_flags
|= RTCF_LOCAL
;
2078 #ifdef CONFIG_IP_MROUTE
2079 if (!ipv4_is_local_multicast(daddr
) && IN_DEV_MFORWARD(in_dev
))
2080 rth
->dst
.input
= ip_mr_input
;
2082 RT_CACHE_STAT_INC(in_slow_mc
);
2084 hash
= rt_hash(daddr
, saddr
, dev
->ifindex
, rt_genid(dev_net(dev
)));
2085 rth
= rt_intern_hash(hash
, rth
, skb
, dev
->ifindex
);
2086 return IS_ERR(rth
) ? PTR_ERR(rth
) : 0;
2097 static void ip_handle_martian_source(struct net_device
*dev
,
2098 struct in_device
*in_dev
,
2099 struct sk_buff
*skb
,
2103 RT_CACHE_STAT_INC(in_martian_src
);
2104 #ifdef CONFIG_IP_ROUTE_VERBOSE
2105 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit()) {
2107 * RFC1812 recommendation, if source is martian,
2108 * the only hint is MAC header.
2110 printk(KERN_WARNING
"martian source %pI4 from %pI4, on dev %s\n",
2111 &daddr
, &saddr
, dev
->name
);
2112 if (dev
->hard_header_len
&& skb_mac_header_was_set(skb
)) {
2114 const unsigned char *p
= skb_mac_header(skb
);
2115 printk(KERN_WARNING
"ll header: ");
2116 for (i
= 0; i
< dev
->hard_header_len
; i
++, p
++) {
2118 if (i
< (dev
->hard_header_len
- 1))
2127 /* called in rcu_read_lock() section */
2128 static int __mkroute_input(struct sk_buff
*skb
,
2129 const struct fib_result
*res
,
2130 struct in_device
*in_dev
,
2131 __be32 daddr
, __be32 saddr
, u32 tos
,
2132 struct rtable
**result
)
2136 struct in_device
*out_dev
;
2137 unsigned int flags
= 0;
2141 /* get a working reference to the output device */
2142 out_dev
= __in_dev_get_rcu(FIB_RES_DEV(*res
));
2143 if (out_dev
== NULL
) {
2144 if (net_ratelimit())
2145 printk(KERN_CRIT
"Bug in ip_route_input" \
2146 "_slow(). Please, report\n");
2151 err
= fib_validate_source(skb
, saddr
, daddr
, tos
, FIB_RES_OIF(*res
),
2152 in_dev
->dev
, &spec_dst
, &itag
);
2154 ip_handle_martian_source(in_dev
->dev
, in_dev
, skb
, daddr
,
2161 flags
|= RTCF_DIRECTSRC
;
2163 if (out_dev
== in_dev
&& err
&&
2164 (IN_DEV_SHARED_MEDIA(out_dev
) ||
2165 inet_addr_onlink(out_dev
, saddr
, FIB_RES_GW(*res
))))
2166 flags
|= RTCF_DOREDIRECT
;
2168 if (skb
->protocol
!= htons(ETH_P_IP
)) {
2169 /* Not IP (i.e. ARP). Do not create route, if it is
2170 * invalid for proxy arp. DNAT routes are always valid.
2172 * Proxy arp feature have been extended to allow, ARP
2173 * replies back to the same interface, to support
2174 * Private VLAN switch technologies. See arp.c.
2176 if (out_dev
== in_dev
&&
2177 IN_DEV_PROXY_ARP_PVLAN(in_dev
) == 0) {
2183 rth
= rt_dst_alloc(out_dev
->dev
,
2184 IN_DEV_CONF_GET(in_dev
, NOPOLICY
),
2185 IN_DEV_CONF_GET(out_dev
, NOXFRM
));
2191 rth
->rt_key_dst
= daddr
;
2192 rth
->rt_key_src
= saddr
;
2193 rth
->rt_genid
= rt_genid(dev_net(rth
->dst
.dev
));
2194 rth
->rt_flags
= flags
;
2195 rth
->rt_type
= res
->type
;
2196 rth
->rt_key_tos
= tos
;
2197 rth
->rt_dst
= daddr
;
2198 rth
->rt_src
= saddr
;
2199 rth
->rt_route_iif
= in_dev
->dev
->ifindex
;
2200 rth
->rt_iif
= in_dev
->dev
->ifindex
;
2202 rth
->rt_mark
= skb
->mark
;
2203 rth
->rt_gateway
= daddr
;
2204 rth
->rt_spec_dst
= spec_dst
;
2205 rth
->rt_peer_genid
= 0;
2209 rth
->dst
.input
= ip_forward
;
2210 rth
->dst
.output
= ip_output
;
2212 rt_set_nexthop(rth
, NULL
, res
, res
->fi
, res
->type
, itag
);
2220 static int ip_mkroute_input(struct sk_buff
*skb
,
2221 struct fib_result
*res
,
2222 const struct flowi4
*fl4
,
2223 struct in_device
*in_dev
,
2224 __be32 daddr
, __be32 saddr
, u32 tos
)
2226 struct rtable
* rth
= NULL
;
2230 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2231 if (res
->fi
&& res
->fi
->fib_nhs
> 1)
2232 fib_select_multipath(res
);
2235 /* create a routing cache entry */
2236 err
= __mkroute_input(skb
, res
, in_dev
, daddr
, saddr
, tos
, &rth
);
2240 /* put it into the cache */
2241 hash
= rt_hash(daddr
, saddr
, fl4
->flowi4_iif
,
2242 rt_genid(dev_net(rth
->dst
.dev
)));
2243 rth
= rt_intern_hash(hash
, rth
, skb
, fl4
->flowi4_iif
);
2245 return PTR_ERR(rth
);
2250 * NOTE. We drop all the packets that has local source
2251 * addresses, because every properly looped back packet
2252 * must have correct destination already attached by output routine.
2254 * Such approach solves two big problems:
2255 * 1. Not simplex devices are handled properly.
2256 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2257 * called with rcu_read_lock()
2260 static int ip_route_input_slow(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2261 u8 tos
, struct net_device
*dev
)
2263 struct fib_result res
;
2264 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
2268 struct rtable
* rth
;
2272 struct net
* net
= dev_net(dev
);
2274 /* IP on this device is disabled. */
2279 /* Check for the most weird martians, which can be not detected
2283 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
2284 ipv4_is_loopback(saddr
))
2285 goto martian_source
;
2287 if (ipv4_is_lbcast(daddr
) || (saddr
== 0 && daddr
== 0))
2290 /* Accept zero addresses only to limited broadcast;
2291 * I even do not know to fix it or not. Waiting for complains :-)
2293 if (ipv4_is_zeronet(saddr
))
2294 goto martian_source
;
2296 if (ipv4_is_zeronet(daddr
) || ipv4_is_loopback(daddr
))
2297 goto martian_destination
;
2300 * Now we are ready to route packet.
2303 fl4
.flowi4_iif
= dev
->ifindex
;
2304 fl4
.flowi4_mark
= skb
->mark
;
2305 fl4
.flowi4_tos
= tos
;
2306 fl4
.flowi4_scope
= RT_SCOPE_UNIVERSE
;
2309 err
= fib_lookup(net
, &fl4
, &res
);
2311 if (!IN_DEV_FORWARD(in_dev
))
2316 RT_CACHE_STAT_INC(in_slow_tot
);
2318 if (res
.type
== RTN_BROADCAST
)
2321 if (res
.type
== RTN_LOCAL
) {
2322 err
= fib_validate_source(skb
, saddr
, daddr
, tos
,
2323 net
->loopback_dev
->ifindex
,
2324 dev
, &spec_dst
, &itag
);
2326 goto martian_source_keep_err
;
2328 flags
|= RTCF_DIRECTSRC
;
2333 if (!IN_DEV_FORWARD(in_dev
))
2335 if (res
.type
!= RTN_UNICAST
)
2336 goto martian_destination
;
2338 err
= ip_mkroute_input(skb
, &res
, &fl4
, in_dev
, daddr
, saddr
, tos
);
2342 if (skb
->protocol
!= htons(ETH_P_IP
))
2345 if (ipv4_is_zeronet(saddr
))
2346 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
2348 err
= fib_validate_source(skb
, saddr
, 0, tos
, 0, dev
, &spec_dst
,
2351 goto martian_source_keep_err
;
2353 flags
|= RTCF_DIRECTSRC
;
2355 flags
|= RTCF_BROADCAST
;
2356 res
.type
= RTN_BROADCAST
;
2357 RT_CACHE_STAT_INC(in_brd
);
2360 rth
= rt_dst_alloc(net
->loopback_dev
,
2361 IN_DEV_CONF_GET(in_dev
, NOPOLICY
), false);
2365 rth
->dst
.input
= ip_local_deliver
;
2366 rth
->dst
.output
= ip_rt_bug
;
2367 #ifdef CONFIG_IP_ROUTE_CLASSID
2368 rth
->dst
.tclassid
= itag
;
2371 rth
->rt_key_dst
= daddr
;
2372 rth
->rt_key_src
= saddr
;
2373 rth
->rt_genid
= rt_genid(net
);
2374 rth
->rt_flags
= flags
|RTCF_LOCAL
;
2375 rth
->rt_type
= res
.type
;
2376 rth
->rt_key_tos
= tos
;
2377 rth
->rt_dst
= daddr
;
2378 rth
->rt_src
= saddr
;
2379 #ifdef CONFIG_IP_ROUTE_CLASSID
2380 rth
->dst
.tclassid
= itag
;
2382 rth
->rt_route_iif
= dev
->ifindex
;
2383 rth
->rt_iif
= dev
->ifindex
;
2385 rth
->rt_mark
= skb
->mark
;
2386 rth
->rt_gateway
= daddr
;
2387 rth
->rt_spec_dst
= spec_dst
;
2388 rth
->rt_peer_genid
= 0;
2391 if (res
.type
== RTN_UNREACHABLE
) {
2392 rth
->dst
.input
= ip_error
;
2393 rth
->dst
.error
= -err
;
2394 rth
->rt_flags
&= ~RTCF_LOCAL
;
2396 hash
= rt_hash(daddr
, saddr
, fl4
.flowi4_iif
, rt_genid(net
));
2397 rth
= rt_intern_hash(hash
, rth
, skb
, fl4
.flowi4_iif
);
2404 RT_CACHE_STAT_INC(in_no_route
);
2405 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_UNIVERSE
);
2406 res
.type
= RTN_UNREACHABLE
;
2412 * Do not cache martian addresses: they should be logged (RFC1812)
2414 martian_destination
:
2415 RT_CACHE_STAT_INC(in_martian_dst
);
2416 #ifdef CONFIG_IP_ROUTE_VERBOSE
2417 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit())
2418 printk(KERN_WARNING
"martian destination %pI4 from %pI4, dev %s\n",
2419 &daddr
, &saddr
, dev
->name
);
2423 err
= -EHOSTUNREACH
;
2436 martian_source_keep_err
:
2437 ip_handle_martian_source(dev
, in_dev
, skb
, daddr
, saddr
);
2441 int ip_route_input_common(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2442 u8 tos
, struct net_device
*dev
, bool noref
)
2444 struct rtable
* rth
;
2446 int iif
= dev
->ifindex
;
2454 if (!rt_caching(net
))
2457 tos
&= IPTOS_RT_MASK
;
2458 hash
= rt_hash(daddr
, saddr
, iif
, rt_genid(net
));
2460 for (rth
= rcu_dereference(rt_hash_table
[hash
].chain
); rth
;
2461 rth
= rcu_dereference(rth
->dst
.rt_next
)) {
2462 if ((((__force u32
)rth
->rt_key_dst
^ (__force u32
)daddr
) |
2463 ((__force u32
)rth
->rt_key_src
^ (__force u32
)saddr
) |
2464 (rth
->rt_route_iif
^ iif
) |
2465 (rth
->rt_key_tos
^ tos
)) == 0 &&
2466 rth
->rt_mark
== skb
->mark
&&
2467 net_eq(dev_net(rth
->dst
.dev
), net
) &&
2468 !rt_is_expired(rth
)) {
2469 ipv4_validate_peer(rth
);
2471 dst_use_noref(&rth
->dst
, jiffies
);
2472 skb_dst_set_noref(skb
, &rth
->dst
);
2474 dst_use(&rth
->dst
, jiffies
);
2475 skb_dst_set(skb
, &rth
->dst
);
2477 RT_CACHE_STAT_INC(in_hit
);
2481 RT_CACHE_STAT_INC(in_hlist_search
);
2485 /* Multicast recognition logic is moved from route cache to here.
2486 The problem was that too many Ethernet cards have broken/missing
2487 hardware multicast filters :-( As result the host on multicasting
2488 network acquires a lot of useless route cache entries, sort of
2489 SDR messages from all the world. Now we try to get rid of them.
2490 Really, provided software IP multicast filter is organized
2491 reasonably (at least, hashed), it does not result in a slowdown
2492 comparing with route cache reject entries.
2493 Note, that multicast routers are not affected, because
2494 route cache entry is created eventually.
2496 if (ipv4_is_multicast(daddr
)) {
2497 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
2500 int our
= ip_check_mc_rcu(in_dev
, daddr
, saddr
,
2501 ip_hdr(skb
)->protocol
);
2503 #ifdef CONFIG_IP_MROUTE
2505 (!ipv4_is_local_multicast(daddr
) &&
2506 IN_DEV_MFORWARD(in_dev
))
2509 int res
= ip_route_input_mc(skb
, daddr
, saddr
,
2518 res
= ip_route_input_slow(skb
, daddr
, saddr
, tos
, dev
);
2522 EXPORT_SYMBOL(ip_route_input_common
);
2524 /* called with rcu_read_lock() */
2525 static struct rtable
*__mkroute_output(const struct fib_result
*res
,
2526 const struct flowi4
*fl4
,
2527 __be32 orig_daddr
, __be32 orig_saddr
,
2528 int orig_oif
, __u8 orig_rtos
,
2529 struct net_device
*dev_out
,
2532 struct fib_info
*fi
= res
->fi
;
2533 struct in_device
*in_dev
;
2534 u16 type
= res
->type
;
2537 if (ipv4_is_loopback(fl4
->saddr
) && !(dev_out
->flags
& IFF_LOOPBACK
))
2538 return ERR_PTR(-EINVAL
);
2540 if (ipv4_is_lbcast(fl4
->daddr
))
2541 type
= RTN_BROADCAST
;
2542 else if (ipv4_is_multicast(fl4
->daddr
))
2543 type
= RTN_MULTICAST
;
2544 else if (ipv4_is_zeronet(fl4
->daddr
))
2545 return ERR_PTR(-EINVAL
);
2547 if (dev_out
->flags
& IFF_LOOPBACK
)
2548 flags
|= RTCF_LOCAL
;
2550 in_dev
= __in_dev_get_rcu(dev_out
);
2552 return ERR_PTR(-EINVAL
);
2554 if (type
== RTN_BROADCAST
) {
2555 flags
|= RTCF_BROADCAST
| RTCF_LOCAL
;
2557 } else if (type
== RTN_MULTICAST
) {
2558 flags
|= RTCF_MULTICAST
| RTCF_LOCAL
;
2559 if (!ip_check_mc_rcu(in_dev
, fl4
->daddr
, fl4
->saddr
,
2561 flags
&= ~RTCF_LOCAL
;
2562 /* If multicast route do not exist use
2563 * default one, but do not gateway in this case.
2566 if (fi
&& res
->prefixlen
< 4)
2570 rth
= rt_dst_alloc(dev_out
,
2571 IN_DEV_CONF_GET(in_dev
, NOPOLICY
),
2572 IN_DEV_CONF_GET(in_dev
, NOXFRM
));
2574 return ERR_PTR(-ENOBUFS
);
2576 rth
->dst
.output
= ip_output
;
2578 rth
->rt_key_dst
= orig_daddr
;
2579 rth
->rt_key_src
= orig_saddr
;
2580 rth
->rt_genid
= rt_genid(dev_net(dev_out
));
2581 rth
->rt_flags
= flags
;
2582 rth
->rt_type
= type
;
2583 rth
->rt_key_tos
= orig_rtos
;
2584 rth
->rt_dst
= fl4
->daddr
;
2585 rth
->rt_src
= fl4
->saddr
;
2586 rth
->rt_route_iif
= 0;
2587 rth
->rt_iif
= orig_oif
? : dev_out
->ifindex
;
2588 rth
->rt_oif
= orig_oif
;
2589 rth
->rt_mark
= fl4
->flowi4_mark
;
2590 rth
->rt_gateway
= fl4
->daddr
;
2591 rth
->rt_spec_dst
= fl4
->saddr
;
2592 rth
->rt_peer_genid
= 0;
2596 RT_CACHE_STAT_INC(out_slow_tot
);
2598 if (flags
& RTCF_LOCAL
) {
2599 rth
->dst
.input
= ip_local_deliver
;
2600 rth
->rt_spec_dst
= fl4
->daddr
;
2602 if (flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) {
2603 rth
->rt_spec_dst
= fl4
->saddr
;
2604 if (flags
& RTCF_LOCAL
&&
2605 !(dev_out
->flags
& IFF_LOOPBACK
)) {
2606 rth
->dst
.output
= ip_mc_output
;
2607 RT_CACHE_STAT_INC(out_slow_mc
);
2609 #ifdef CONFIG_IP_MROUTE
2610 if (type
== RTN_MULTICAST
) {
2611 if (IN_DEV_MFORWARD(in_dev
) &&
2612 !ipv4_is_local_multicast(fl4
->daddr
)) {
2613 rth
->dst
.input
= ip_mr_input
;
2614 rth
->dst
.output
= ip_mc_output
;
2620 rt_set_nexthop(rth
, fl4
, res
, fi
, type
, 0);
2626 * Major route resolver routine.
2627 * called with rcu_read_lock();
2630 static struct rtable
*ip_route_output_slow(struct net
*net
, struct flowi4
*fl4
)
2632 struct net_device
*dev_out
= NULL
;
2633 __u8 tos
= RT_FL_TOS(fl4
);
2634 unsigned int flags
= 0;
2635 struct fib_result res
;
2642 #ifdef CONFIG_IP_MULTIPLE_TABLES
2646 orig_daddr
= fl4
->daddr
;
2647 orig_saddr
= fl4
->saddr
;
2648 orig_oif
= fl4
->flowi4_oif
;
2650 fl4
->flowi4_iif
= net
->loopback_dev
->ifindex
;
2651 fl4
->flowi4_tos
= tos
& IPTOS_RT_MASK
;
2652 fl4
->flowi4_scope
= ((tos
& RTO_ONLINK
) ?
2653 RT_SCOPE_LINK
: RT_SCOPE_UNIVERSE
);
2657 rth
= ERR_PTR(-EINVAL
);
2658 if (ipv4_is_multicast(fl4
->saddr
) ||
2659 ipv4_is_lbcast(fl4
->saddr
) ||
2660 ipv4_is_zeronet(fl4
->saddr
))
2663 /* I removed check for oif == dev_out->oif here.
2664 It was wrong for two reasons:
2665 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2666 is assigned to multiple interfaces.
2667 2. Moreover, we are allowed to send packets with saddr
2668 of another iface. --ANK
2671 if (fl4
->flowi4_oif
== 0 &&
2672 (ipv4_is_multicast(fl4
->daddr
) ||
2673 ipv4_is_lbcast(fl4
->daddr
))) {
2674 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2675 dev_out
= __ip_dev_find(net
, fl4
->saddr
, false);
2676 if (dev_out
== NULL
)
2679 /* Special hack: user can direct multicasts
2680 and limited broadcast via necessary interface
2681 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2682 This hack is not just for fun, it allows
2683 vic,vat and friends to work.
2684 They bind socket to loopback, set ttl to zero
2685 and expect that it will work.
2686 From the viewpoint of routing cache they are broken,
2687 because we are not allowed to build multicast path
2688 with loopback source addr (look, routing cache
2689 cannot know, that ttl is zero, so that packet
2690 will not leave this host and route is valid).
2691 Luckily, this hack is good workaround.
2694 fl4
->flowi4_oif
= dev_out
->ifindex
;
2698 if (!(fl4
->flowi4_flags
& FLOWI_FLAG_ANYSRC
)) {
2699 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2700 if (!__ip_dev_find(net
, fl4
->saddr
, false))
2706 if (fl4
->flowi4_oif
) {
2707 dev_out
= dev_get_by_index_rcu(net
, fl4
->flowi4_oif
);
2708 rth
= ERR_PTR(-ENODEV
);
2709 if (dev_out
== NULL
)
2712 /* RACE: Check return value of inet_select_addr instead. */
2713 if (!(dev_out
->flags
& IFF_UP
) || !__in_dev_get_rcu(dev_out
)) {
2714 rth
= ERR_PTR(-ENETUNREACH
);
2717 if (ipv4_is_local_multicast(fl4
->daddr
) ||
2718 ipv4_is_lbcast(fl4
->daddr
)) {
2720 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2725 if (ipv4_is_multicast(fl4
->daddr
))
2726 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2728 else if (!fl4
->daddr
)
2729 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2735 fl4
->daddr
= fl4
->saddr
;
2737 fl4
->daddr
= fl4
->saddr
= htonl(INADDR_LOOPBACK
);
2738 dev_out
= net
->loopback_dev
;
2739 fl4
->flowi4_oif
= net
->loopback_dev
->ifindex
;
2740 res
.type
= RTN_LOCAL
;
2741 flags
|= RTCF_LOCAL
;
2745 if (fib_lookup(net
, fl4
, &res
)) {
2747 if (fl4
->flowi4_oif
) {
2748 /* Apparently, routing tables are wrong. Assume,
2749 that the destination is on link.
2752 Because we are allowed to send to iface
2753 even if it has NO routes and NO assigned
2754 addresses. When oif is specified, routing
2755 tables are looked up with only one purpose:
2756 to catch if destination is gatewayed, rather than
2757 direct. Moreover, if MSG_DONTROUTE is set,
2758 we send packet, ignoring both routing tables
2759 and ifaddr state. --ANK
2762 We could make it even if oif is unknown,
2763 likely IPv6, but we do not.
2766 if (fl4
->saddr
== 0)
2767 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2769 res
.type
= RTN_UNICAST
;
2772 rth
= ERR_PTR(-ENETUNREACH
);
2776 if (res
.type
== RTN_LOCAL
) {
2778 if (res
.fi
->fib_prefsrc
)
2779 fl4
->saddr
= res
.fi
->fib_prefsrc
;
2781 fl4
->saddr
= fl4
->daddr
;
2783 dev_out
= net
->loopback_dev
;
2784 fl4
->flowi4_oif
= dev_out
->ifindex
;
2786 flags
|= RTCF_LOCAL
;
2790 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2791 if (res
.fi
->fib_nhs
> 1 && fl4
->flowi4_oif
== 0)
2792 fib_select_multipath(&res
);
2795 if (!res
.prefixlen
&&
2796 res
.table
->tb_num_default
> 1 &&
2797 res
.type
== RTN_UNICAST
&& !fl4
->flowi4_oif
)
2798 fib_select_default(&res
);
2801 fl4
->saddr
= FIB_RES_PREFSRC(net
, res
);
2803 dev_out
= FIB_RES_DEV(res
);
2804 fl4
->flowi4_oif
= dev_out
->ifindex
;
2808 rth
= __mkroute_output(&res
, fl4
, orig_daddr
, orig_saddr
, orig_oif
,
2809 tos
, dev_out
, flags
);
2813 hash
= rt_hash(orig_daddr
, orig_saddr
, orig_oif
,
2814 rt_genid(dev_net(dev_out
)));
2815 rth
= rt_intern_hash(hash
, rth
, NULL
, orig_oif
);
2823 struct rtable
*__ip_route_output_key(struct net
*net
, struct flowi4
*flp4
)
2828 if (!rt_caching(net
))
2831 hash
= rt_hash(flp4
->daddr
, flp4
->saddr
, flp4
->flowi4_oif
, rt_genid(net
));
2834 for (rth
= rcu_dereference_bh(rt_hash_table
[hash
].chain
); rth
;
2835 rth
= rcu_dereference_bh(rth
->dst
.rt_next
)) {
2836 if (rth
->rt_key_dst
== flp4
->daddr
&&
2837 rth
->rt_key_src
== flp4
->saddr
&&
2838 rt_is_output_route(rth
) &&
2839 rth
->rt_oif
== flp4
->flowi4_oif
&&
2840 rth
->rt_mark
== flp4
->flowi4_mark
&&
2841 !((rth
->rt_key_tos
^ flp4
->flowi4_tos
) &
2842 (IPTOS_RT_MASK
| RTO_ONLINK
)) &&
2843 net_eq(dev_net(rth
->dst
.dev
), net
) &&
2844 !rt_is_expired(rth
)) {
2845 ipv4_validate_peer(rth
);
2846 dst_use(&rth
->dst
, jiffies
);
2847 RT_CACHE_STAT_INC(out_hit
);
2848 rcu_read_unlock_bh();
2850 flp4
->saddr
= rth
->rt_src
;
2852 flp4
->daddr
= rth
->rt_dst
;
2855 RT_CACHE_STAT_INC(out_hlist_search
);
2857 rcu_read_unlock_bh();
2860 return ip_route_output_slow(net
, flp4
);
2862 EXPORT_SYMBOL_GPL(__ip_route_output_key
);
2864 static struct dst_entry
*ipv4_blackhole_dst_check(struct dst_entry
*dst
, u32 cookie
)
2869 static unsigned int ipv4_blackhole_default_mtu(const struct dst_entry
*dst
)
2874 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
2878 static u32
*ipv4_rt_blackhole_cow_metrics(struct dst_entry
*dst
,
2884 static struct dst_ops ipv4_dst_blackhole_ops
= {
2886 .protocol
= cpu_to_be16(ETH_P_IP
),
2887 .destroy
= ipv4_dst_destroy
,
2888 .check
= ipv4_blackhole_dst_check
,
2889 .default_mtu
= ipv4_blackhole_default_mtu
,
2890 .default_advmss
= ipv4_default_advmss
,
2891 .update_pmtu
= ipv4_rt_blackhole_update_pmtu
,
2892 .cow_metrics
= ipv4_rt_blackhole_cow_metrics
,
2893 .neigh_lookup
= ipv4_neigh_lookup
,
2896 struct dst_entry
*ipv4_blackhole_route(struct net
*net
, struct dst_entry
*dst_orig
)
2898 struct rtable
*rt
= dst_alloc(&ipv4_dst_blackhole_ops
, NULL
, 1, 0, 0);
2899 struct rtable
*ort
= (struct rtable
*) dst_orig
;
2902 struct dst_entry
*new = &rt
->dst
;
2905 new->input
= dst_discard
;
2906 new->output
= dst_discard
;
2907 dst_copy_metrics(new, &ort
->dst
);
2909 new->dev
= ort
->dst
.dev
;
2913 rt
->rt_key_dst
= ort
->rt_key_dst
;
2914 rt
->rt_key_src
= ort
->rt_key_src
;
2915 rt
->rt_key_tos
= ort
->rt_key_tos
;
2916 rt
->rt_route_iif
= ort
->rt_route_iif
;
2917 rt
->rt_iif
= ort
->rt_iif
;
2918 rt
->rt_oif
= ort
->rt_oif
;
2919 rt
->rt_mark
= ort
->rt_mark
;
2921 rt
->rt_genid
= rt_genid(net
);
2922 rt
->rt_flags
= ort
->rt_flags
;
2923 rt
->rt_type
= ort
->rt_type
;
2924 rt
->rt_dst
= ort
->rt_dst
;
2925 rt
->rt_src
= ort
->rt_src
;
2926 rt
->rt_gateway
= ort
->rt_gateway
;
2927 rt
->rt_spec_dst
= ort
->rt_spec_dst
;
2928 rt
->peer
= ort
->peer
;
2930 atomic_inc(&rt
->peer
->refcnt
);
2933 atomic_inc(&rt
->fi
->fib_clntref
);
2938 dst_release(dst_orig
);
2940 return rt
? &rt
->dst
: ERR_PTR(-ENOMEM
);
2943 struct rtable
*ip_route_output_flow(struct net
*net
, struct flowi4
*flp4
,
2946 struct rtable
*rt
= __ip_route_output_key(net
, flp4
);
2951 if (flp4
->flowi4_proto
)
2952 rt
= (struct rtable
*) xfrm_lookup(net
, &rt
->dst
,
2953 flowi4_to_flowi(flp4
),
2958 EXPORT_SYMBOL_GPL(ip_route_output_flow
);
2960 static int rt_fill_info(struct net
*net
,
2961 struct sk_buff
*skb
, u32 pid
, u32 seq
, int event
,
2962 int nowait
, unsigned int flags
)
2964 struct rtable
*rt
= skb_rtable(skb
);
2966 struct nlmsghdr
*nlh
;
2968 const struct inet_peer
*peer
= rt
->peer
;
2969 u32 id
= 0, ts
= 0, tsage
= 0, error
;
2971 nlh
= nlmsg_put(skb
, pid
, seq
, event
, sizeof(*r
), flags
);
2975 r
= nlmsg_data(nlh
);
2976 r
->rtm_family
= AF_INET
;
2977 r
->rtm_dst_len
= 32;
2979 r
->rtm_tos
= rt
->rt_key_tos
;
2980 r
->rtm_table
= RT_TABLE_MAIN
;
2981 NLA_PUT_U32(skb
, RTA_TABLE
, RT_TABLE_MAIN
);
2982 r
->rtm_type
= rt
->rt_type
;
2983 r
->rtm_scope
= RT_SCOPE_UNIVERSE
;
2984 r
->rtm_protocol
= RTPROT_UNSPEC
;
2985 r
->rtm_flags
= (rt
->rt_flags
& ~0xFFFF) | RTM_F_CLONED
;
2986 if (rt
->rt_flags
& RTCF_NOTIFY
)
2987 r
->rtm_flags
|= RTM_F_NOTIFY
;
2989 NLA_PUT_BE32(skb
, RTA_DST
, rt
->rt_dst
);
2991 if (rt
->rt_key_src
) {
2992 r
->rtm_src_len
= 32;
2993 NLA_PUT_BE32(skb
, RTA_SRC
, rt
->rt_key_src
);
2996 NLA_PUT_U32(skb
, RTA_OIF
, rt
->dst
.dev
->ifindex
);
2997 #ifdef CONFIG_IP_ROUTE_CLASSID
2998 if (rt
->dst
.tclassid
)
2999 NLA_PUT_U32(skb
, RTA_FLOW
, rt
->dst
.tclassid
);
3001 if (rt_is_input_route(rt
))
3002 NLA_PUT_BE32(skb
, RTA_PREFSRC
, rt
->rt_spec_dst
);
3003 else if (rt
->rt_src
!= rt
->rt_key_src
)
3004 NLA_PUT_BE32(skb
, RTA_PREFSRC
, rt
->rt_src
);
3006 if (rt
->rt_dst
!= rt
->rt_gateway
)
3007 NLA_PUT_BE32(skb
, RTA_GATEWAY
, rt
->rt_gateway
);
3009 if (rtnetlink_put_metrics(skb
, dst_metrics_ptr(&rt
->dst
)) < 0)
3010 goto nla_put_failure
;
3013 NLA_PUT_BE32(skb
, RTA_MARK
, rt
->rt_mark
);
3015 error
= rt
->dst
.error
;
3017 inet_peer_refcheck(rt
->peer
);
3018 id
= atomic_read(&peer
->ip_id_count
) & 0xffff;
3019 if (peer
->tcp_ts_stamp
) {
3021 tsage
= get_seconds() - peer
->tcp_ts_stamp
;
3023 expires
= ACCESS_ONCE(peer
->pmtu_expires
);
3028 if (rt_is_input_route(rt
)) {
3029 #ifdef CONFIG_IP_MROUTE
3030 __be32 dst
= rt
->rt_dst
;
3032 if (ipv4_is_multicast(dst
) && !ipv4_is_local_multicast(dst
) &&
3033 IPV4_DEVCONF_ALL(net
, MC_FORWARDING
)) {
3034 int err
= ipmr_get_route(net
, skb
,
3035 rt
->rt_src
, rt
->rt_dst
,
3041 goto nla_put_failure
;
3043 if (err
== -EMSGSIZE
)
3044 goto nla_put_failure
;
3050 NLA_PUT_U32(skb
, RTA_IIF
, rt
->rt_iif
);
3053 if (rtnl_put_cacheinfo(skb
, &rt
->dst
, id
, ts
, tsage
,
3054 expires
, error
) < 0)
3055 goto nla_put_failure
;
3057 return nlmsg_end(skb
, nlh
);
3060 nlmsg_cancel(skb
, nlh
);
3064 static int inet_rtm_getroute(struct sk_buff
*in_skb
, struct nlmsghdr
* nlh
, void *arg
)
3066 struct net
*net
= sock_net(in_skb
->sk
);
3068 struct nlattr
*tb
[RTA_MAX
+1];
3069 struct rtable
*rt
= NULL
;
3075 struct sk_buff
*skb
;
3077 err
= nlmsg_parse(nlh
, sizeof(*rtm
), tb
, RTA_MAX
, rtm_ipv4_policy
);
3081 rtm
= nlmsg_data(nlh
);
3083 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
3089 /* Reserve room for dummy headers, this skb can pass
3090 through good chunk of routing engine.
3092 skb_reset_mac_header(skb
);
3093 skb_reset_network_header(skb
);
3095 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
3096 ip_hdr(skb
)->protocol
= IPPROTO_ICMP
;
3097 skb_reserve(skb
, MAX_HEADER
+ sizeof(struct iphdr
));
3099 src
= tb
[RTA_SRC
] ? nla_get_be32(tb
[RTA_SRC
]) : 0;
3100 dst
= tb
[RTA_DST
] ? nla_get_be32(tb
[RTA_DST
]) : 0;
3101 iif
= tb
[RTA_IIF
] ? nla_get_u32(tb
[RTA_IIF
]) : 0;
3102 mark
= tb
[RTA_MARK
] ? nla_get_u32(tb
[RTA_MARK
]) : 0;
3105 struct net_device
*dev
;
3107 dev
= __dev_get_by_index(net
, iif
);
3113 skb
->protocol
= htons(ETH_P_IP
);
3117 err
= ip_route_input(skb
, dst
, src
, rtm
->rtm_tos
, dev
);
3120 rt
= skb_rtable(skb
);
3121 if (err
== 0 && rt
->dst
.error
)
3122 err
= -rt
->dst
.error
;
3124 struct flowi4 fl4
= {
3127 .flowi4_tos
= rtm
->rtm_tos
,
3128 .flowi4_oif
= tb
[RTA_OIF
] ? nla_get_u32(tb
[RTA_OIF
]) : 0,
3129 .flowi4_mark
= mark
,
3131 rt
= ip_route_output_key(net
, &fl4
);
3141 skb_dst_set(skb
, &rt
->dst
);
3142 if (rtm
->rtm_flags
& RTM_F_NOTIFY
)
3143 rt
->rt_flags
|= RTCF_NOTIFY
;
3145 err
= rt_fill_info(net
, skb
, NETLINK_CB(in_skb
).pid
, nlh
->nlmsg_seq
,
3146 RTM_NEWROUTE
, 0, 0);
3150 err
= rtnl_unicast(skb
, net
, NETLINK_CB(in_skb
).pid
);
3159 int ip_rt_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
3166 net
= sock_net(skb
->sk
);
3171 s_idx
= idx
= cb
->args
[1];
3172 for (h
= s_h
; h
<= rt_hash_mask
; h
++, s_idx
= 0) {
3173 if (!rt_hash_table
[h
].chain
)
3176 for (rt
= rcu_dereference_bh(rt_hash_table
[h
].chain
), idx
= 0; rt
;
3177 rt
= rcu_dereference_bh(rt
->dst
.rt_next
), idx
++) {
3178 if (!net_eq(dev_net(rt
->dst
.dev
), net
) || idx
< s_idx
)
3180 if (rt_is_expired(rt
))
3182 skb_dst_set_noref(skb
, &rt
->dst
);
3183 if (rt_fill_info(net
, skb
, NETLINK_CB(cb
->skb
).pid
,
3184 cb
->nlh
->nlmsg_seq
, RTM_NEWROUTE
,
3185 1, NLM_F_MULTI
) <= 0) {
3187 rcu_read_unlock_bh();
3192 rcu_read_unlock_bh();
3201 void ip_rt_multicast_event(struct in_device
*in_dev
)
3203 rt_cache_flush(dev_net(in_dev
->dev
), 0);
3206 #ifdef CONFIG_SYSCTL
3207 static int ipv4_sysctl_rtcache_flush(ctl_table
*__ctl
, int write
,
3208 void __user
*buffer
,
3209 size_t *lenp
, loff_t
*ppos
)
3216 memcpy(&ctl
, __ctl
, sizeof(ctl
));
3217 ctl
.data
= &flush_delay
;
3218 proc_dointvec(&ctl
, write
, buffer
, lenp
, ppos
);
3220 net
= (struct net
*)__ctl
->extra1
;
3221 rt_cache_flush(net
, flush_delay
);
3228 static ctl_table ipv4_route_table
[] = {
3230 .procname
= "gc_thresh",
3231 .data
= &ipv4_dst_ops
.gc_thresh
,
3232 .maxlen
= sizeof(int),
3234 .proc_handler
= proc_dointvec
,
3237 .procname
= "max_size",
3238 .data
= &ip_rt_max_size
,
3239 .maxlen
= sizeof(int),
3241 .proc_handler
= proc_dointvec
,
3244 /* Deprecated. Use gc_min_interval_ms */
3246 .procname
= "gc_min_interval",
3247 .data
= &ip_rt_gc_min_interval
,
3248 .maxlen
= sizeof(int),
3250 .proc_handler
= proc_dointvec_jiffies
,
3253 .procname
= "gc_min_interval_ms",
3254 .data
= &ip_rt_gc_min_interval
,
3255 .maxlen
= sizeof(int),
3257 .proc_handler
= proc_dointvec_ms_jiffies
,
3260 .procname
= "gc_timeout",
3261 .data
= &ip_rt_gc_timeout
,
3262 .maxlen
= sizeof(int),
3264 .proc_handler
= proc_dointvec_jiffies
,
3267 .procname
= "gc_interval",
3268 .data
= &ip_rt_gc_interval
,
3269 .maxlen
= sizeof(int),
3271 .proc_handler
= proc_dointvec_jiffies
,
3274 .procname
= "gc_interval",
3275 .data
= &ip_rt_gc_interval
,
3276 .maxlen
= sizeof(int),
3278 .proc_handler
= proc_dointvec_jiffies
,
3281 .procname
= "redirect_load",
3282 .data
= &ip_rt_redirect_load
,
3283 .maxlen
= sizeof(int),
3285 .proc_handler
= proc_dointvec
,
3288 .procname
= "redirect_number",
3289 .data
= &ip_rt_redirect_number
,
3290 .maxlen
= sizeof(int),
3292 .proc_handler
= proc_dointvec
,
3295 .procname
= "redirect_silence",
3296 .data
= &ip_rt_redirect_silence
,
3297 .maxlen
= sizeof(int),
3299 .proc_handler
= proc_dointvec
,
3302 .procname
= "error_cost",
3303 .data
= &ip_rt_error_cost
,
3304 .maxlen
= sizeof(int),
3306 .proc_handler
= proc_dointvec
,
3309 .procname
= "error_burst",
3310 .data
= &ip_rt_error_burst
,
3311 .maxlen
= sizeof(int),
3313 .proc_handler
= proc_dointvec
,
3316 .procname
= "gc_elasticity",
3317 .data
= &ip_rt_gc_elasticity
,
3318 .maxlen
= sizeof(int),
3320 .proc_handler
= proc_dointvec
,
3323 .procname
= "mtu_expires",
3324 .data
= &ip_rt_mtu_expires
,
3325 .maxlen
= sizeof(int),
3327 .proc_handler
= proc_dointvec_jiffies
,
3330 .procname
= "min_pmtu",
3331 .data
= &ip_rt_min_pmtu
,
3332 .maxlen
= sizeof(int),
3334 .proc_handler
= proc_dointvec
,
3337 .procname
= "min_adv_mss",
3338 .data
= &ip_rt_min_advmss
,
3339 .maxlen
= sizeof(int),
3341 .proc_handler
= proc_dointvec
,
3346 static struct ctl_table empty
[1];
3348 static struct ctl_table ipv4_skeleton
[] =
3350 { .procname
= "route",
3351 .mode
= 0555, .child
= ipv4_route_table
},
3352 { .procname
= "neigh",
3353 .mode
= 0555, .child
= empty
},
3357 static __net_initdata
struct ctl_path ipv4_path
[] = {
3358 { .procname
= "net", },
3359 { .procname
= "ipv4", },
3363 static struct ctl_table ipv4_route_flush_table
[] = {
3365 .procname
= "flush",
3366 .maxlen
= sizeof(int),
3368 .proc_handler
= ipv4_sysctl_rtcache_flush
,
3373 static __net_initdata
struct ctl_path ipv4_route_path
[] = {
3374 { .procname
= "net", },
3375 { .procname
= "ipv4", },
3376 { .procname
= "route", },
3380 static __net_init
int sysctl_route_net_init(struct net
*net
)
3382 struct ctl_table
*tbl
;
3384 tbl
= ipv4_route_flush_table
;
3385 if (!net_eq(net
, &init_net
)) {
3386 tbl
= kmemdup(tbl
, sizeof(ipv4_route_flush_table
), GFP_KERNEL
);
3390 tbl
[0].extra1
= net
;
3392 net
->ipv4
.route_hdr
=
3393 register_net_sysctl_table(net
, ipv4_route_path
, tbl
);
3394 if (net
->ipv4
.route_hdr
== NULL
)
3399 if (tbl
!= ipv4_route_flush_table
)
3405 static __net_exit
void sysctl_route_net_exit(struct net
*net
)
3407 struct ctl_table
*tbl
;
3409 tbl
= net
->ipv4
.route_hdr
->ctl_table_arg
;
3410 unregister_net_sysctl_table(net
->ipv4
.route_hdr
);
3411 BUG_ON(tbl
== ipv4_route_flush_table
);
3415 static __net_initdata
struct pernet_operations sysctl_route_ops
= {
3416 .init
= sysctl_route_net_init
,
3417 .exit
= sysctl_route_net_exit
,
3421 static __net_init
int rt_genid_init(struct net
*net
)
3423 get_random_bytes(&net
->ipv4
.rt_genid
,
3424 sizeof(net
->ipv4
.rt_genid
));
3425 get_random_bytes(&net
->ipv4
.dev_addr_genid
,
3426 sizeof(net
->ipv4
.dev_addr_genid
));
3430 static __net_initdata
struct pernet_operations rt_genid_ops
= {
3431 .init
= rt_genid_init
,
3435 #ifdef CONFIG_IP_ROUTE_CLASSID
3436 struct ip_rt_acct __percpu
*ip_rt_acct __read_mostly
;
3437 #endif /* CONFIG_IP_ROUTE_CLASSID */
3439 static __initdata
unsigned long rhash_entries
;
3440 static int __init
set_rhash_entries(char *str
)
3444 rhash_entries
= simple_strtoul(str
, &str
, 0);
3447 __setup("rhash_entries=", set_rhash_entries
);
3449 int __init
ip_rt_init(void)
3453 #ifdef CONFIG_IP_ROUTE_CLASSID
3454 ip_rt_acct
= __alloc_percpu(256 * sizeof(struct ip_rt_acct
), __alignof__(struct ip_rt_acct
));
3456 panic("IP: failed to allocate ip_rt_acct\n");
3459 ipv4_dst_ops
.kmem_cachep
=
3460 kmem_cache_create("ip_dst_cache", sizeof(struct rtable
), 0,
3461 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3463 ipv4_dst_blackhole_ops
.kmem_cachep
= ipv4_dst_ops
.kmem_cachep
;
3465 if (dst_entries_init(&ipv4_dst_ops
) < 0)
3466 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3468 if (dst_entries_init(&ipv4_dst_blackhole_ops
) < 0)
3469 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3471 rt_hash_table
= (struct rt_hash_bucket
*)
3472 alloc_large_system_hash("IP route cache",
3473 sizeof(struct rt_hash_bucket
),
3475 (totalram_pages
>= 128 * 1024) ?
3480 rhash_entries
? 0 : 512 * 1024);
3481 memset(rt_hash_table
, 0, (rt_hash_mask
+ 1) * sizeof(struct rt_hash_bucket
));
3482 rt_hash_lock_init();
3484 ipv4_dst_ops
.gc_thresh
= (rt_hash_mask
+ 1);
3485 ip_rt_max_size
= (rt_hash_mask
+ 1) * 16;
3490 INIT_DELAYED_WORK_DEFERRABLE(&expires_work
, rt_worker_func
);
3491 expires_ljiffies
= jiffies
;
3492 schedule_delayed_work(&expires_work
,
3493 net_random() % ip_rt_gc_interval
+ ip_rt_gc_interval
);
3495 if (ip_rt_proc_init())
3496 printk(KERN_ERR
"Unable to create route proc files\n");
3499 xfrm4_init(ip_rt_max_size
);
3501 rtnl_register(PF_INET
, RTM_GETROUTE
, inet_rtm_getroute
, NULL
, NULL
);
3503 #ifdef CONFIG_SYSCTL
3504 register_pernet_subsys(&sysctl_route_ops
);
3506 register_pernet_subsys(&rt_genid_ops
);
3510 #ifdef CONFIG_SYSCTL
3512 * We really need to sanitize the damn ipv4 init order, then all
3513 * this nonsense will go away.
3515 void __init
ip_static_sysctl_init(void)
3517 register_sysctl_paths(ipv4_path
, ipv4_skeleton
);