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 #define pr_fmt(fmt) "IPv4: " fmt
67 #include <linux/module.h>
68 #include <asm/uaccess.h>
69 #include <linux/bitops.h>
70 #include <linux/types.h>
71 #include <linux/kernel.h>
73 #include <linux/bootmem.h>
74 #include <linux/string.h>
75 #include <linux/socket.h>
76 #include <linux/sockios.h>
77 #include <linux/errno.h>
79 #include <linux/inet.h>
80 #include <linux/netdevice.h>
81 #include <linux/proc_fs.h>
82 #include <linux/init.h>
83 #include <linux/workqueue.h>
84 #include <linux/skbuff.h>
85 #include <linux/inetdevice.h>
86 #include <linux/igmp.h>
87 #include <linux/pkt_sched.h>
88 #include <linux/mroute.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/jhash.h>
92 #include <linux/rcupdate.h>
93 #include <linux/times.h>
94 #include <linux/slab.h>
95 #include <linux/prefetch.h>
97 #include <net/net_namespace.h>
98 #include <net/protocol.h>
100 #include <net/route.h>
101 #include <net/inetpeer.h>
102 #include <net/sock.h>
103 #include <net/ip_fib.h>
106 #include <net/icmp.h>
107 #include <net/xfrm.h>
108 #include <net/netevent.h>
109 #include <net/rtnetlink.h>
111 #include <linux/sysctl.h>
112 #include <linux/kmemleak.h>
114 #include <net/secure_seq.h>
116 #define RT_FL_TOS(oldflp4) \
117 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
119 #define IP_MAX_MTU 0xFFF0
121 #define RT_GC_TIMEOUT (300*HZ)
123 static int ip_rt_max_size
;
124 static int ip_rt_gc_timeout __read_mostly
= RT_GC_TIMEOUT
;
125 static int ip_rt_gc_interval __read_mostly
= 60 * HZ
;
126 static int ip_rt_gc_min_interval __read_mostly
= HZ
/ 2;
127 static int ip_rt_redirect_number __read_mostly
= 9;
128 static int ip_rt_redirect_load __read_mostly
= HZ
/ 50;
129 static int ip_rt_redirect_silence __read_mostly
= ((HZ
/ 50) << (9 + 1));
130 static int ip_rt_error_cost __read_mostly
= HZ
;
131 static int ip_rt_error_burst __read_mostly
= 5 * HZ
;
132 static int ip_rt_gc_elasticity __read_mostly
= 8;
133 static int ip_rt_mtu_expires __read_mostly
= 10 * 60 * HZ
;
134 static int ip_rt_min_pmtu __read_mostly
= 512 + 20 + 20;
135 static int ip_rt_min_advmss __read_mostly
= 256;
136 static int rt_chain_length_max __read_mostly
= 20;
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_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
,
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
)
233 EXPORT_SYMBOL(ip_tos2prio
);
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 int 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_access_pointer(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_access_pointer(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_noref(&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_access_pointer(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
);
941 inetpeer_invalidate_tree(AF_INET
);
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
)
963 net_warn_ratelimited("Route hash chain too long!\n");
964 rt_cache_invalidate(net
);
968 Short description of GC goals.
970 We want to build algorithm, which will keep routing cache
971 at some equilibrium point, when number of aged off entries
972 is kept approximately equal to newly generated ones.
974 Current expiration strength is variable "expire".
975 We try to adjust it dynamically, so that if networking
976 is idle expires is large enough to keep enough of warm entries,
977 and when load increases it reduces to limit cache size.
980 static int rt_garbage_collect(struct dst_ops
*ops
)
982 static unsigned long expire
= RT_GC_TIMEOUT
;
983 static unsigned long last_gc
;
985 static int equilibrium
;
987 struct rtable __rcu
**rthp
;
988 unsigned long now
= jiffies
;
990 int entries
= dst_entries_get_fast(&ipv4_dst_ops
);
993 * Garbage collection is pretty expensive,
994 * do not make it too frequently.
997 RT_CACHE_STAT_INC(gc_total
);
999 if (now
- last_gc
< ip_rt_gc_min_interval
&&
1000 entries
< ip_rt_max_size
) {
1001 RT_CACHE_STAT_INC(gc_ignored
);
1005 entries
= dst_entries_get_slow(&ipv4_dst_ops
);
1006 /* Calculate number of entries, which we want to expire now. */
1007 goal
= entries
- (ip_rt_gc_elasticity
<< rt_hash_log
);
1009 if (equilibrium
< ipv4_dst_ops
.gc_thresh
)
1010 equilibrium
= ipv4_dst_ops
.gc_thresh
;
1011 goal
= entries
- equilibrium
;
1013 equilibrium
+= min_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
1014 goal
= entries
- equilibrium
;
1017 /* We are in dangerous area. Try to reduce cache really
1020 goal
= max_t(unsigned int, goal
>> 1, rt_hash_mask
+ 1);
1021 equilibrium
= entries
- goal
;
1024 if (now
- last_gc
>= ip_rt_gc_min_interval
)
1028 equilibrium
+= goal
;
1035 for (i
= rt_hash_mask
, k
= rover
; i
>= 0; i
--) {
1036 unsigned long tmo
= expire
;
1038 k
= (k
+ 1) & rt_hash_mask
;
1039 rthp
= &rt_hash_table
[k
].chain
;
1040 spin_lock_bh(rt_hash_lock_addr(k
));
1041 while ((rth
= rcu_dereference_protected(*rthp
,
1042 lockdep_is_held(rt_hash_lock_addr(k
)))) != NULL
) {
1043 if (!rt_is_expired(rth
) &&
1044 !rt_may_expire(rth
, tmo
, expire
)) {
1046 rthp
= &rth
->dst
.rt_next
;
1049 *rthp
= rth
->dst
.rt_next
;
1053 spin_unlock_bh(rt_hash_lock_addr(k
));
1062 /* Goal is not achieved. We stop process if:
1064 - if expire reduced to zero. Otherwise, expire is halfed.
1065 - if table is not full.
1066 - if we are called from interrupt.
1067 - jiffies check is just fallback/debug loop breaker.
1068 We will not spin here for long time in any case.
1071 RT_CACHE_STAT_INC(gc_goal_miss
);
1078 if (dst_entries_get_fast(&ipv4_dst_ops
) < ip_rt_max_size
)
1080 } while (!in_softirq() && time_before_eq(jiffies
, now
));
1082 if (dst_entries_get_fast(&ipv4_dst_ops
) < ip_rt_max_size
)
1084 if (dst_entries_get_slow(&ipv4_dst_ops
) < ip_rt_max_size
)
1086 net_warn_ratelimited("dst cache overflow\n");
1087 RT_CACHE_STAT_INC(gc_dst_overflow
);
1091 expire
+= ip_rt_gc_min_interval
;
1092 if (expire
> ip_rt_gc_timeout
||
1093 dst_entries_get_fast(&ipv4_dst_ops
) < ipv4_dst_ops
.gc_thresh
||
1094 dst_entries_get_slow(&ipv4_dst_ops
) < ipv4_dst_ops
.gc_thresh
)
1095 expire
= ip_rt_gc_timeout
;
1100 * Returns number of entries in a hash chain that have different hash_inputs
1102 static int slow_chain_length(const struct rtable
*head
)
1105 const struct rtable
*rth
= head
;
1108 length
+= has_noalias(head
, rth
);
1109 rth
= rcu_dereference_protected(rth
->dst
.rt_next
, 1);
1111 return length
>> FRACT_BITS
;
1114 static struct neighbour
*ipv4_neigh_lookup(const struct dst_entry
*dst
, const void *daddr
)
1116 static const __be32 inaddr_any
= 0;
1117 struct net_device
*dev
= dst
->dev
;
1118 const __be32
*pkey
= daddr
;
1119 const struct rtable
*rt
;
1120 struct neighbour
*n
;
1122 rt
= (const struct rtable
*) dst
;
1124 if (dev
->flags
& (IFF_LOOPBACK
| IFF_POINTOPOINT
))
1126 else if (rt
->rt_gateway
)
1127 pkey
= (const __be32
*) &rt
->rt_gateway
;
1129 n
= __ipv4_neigh_lookup(dev
, *(__force u32
*)pkey
);
1132 return neigh_create(&arp_tbl
, pkey
, dev
);
1135 static int rt_bind_neighbour(struct rtable
*rt
)
1137 struct neighbour
*n
= ipv4_neigh_lookup(&rt
->dst
, &rt
->rt_gateway
);
1140 dst_set_neighbour(&rt
->dst
, n
);
1145 static struct rtable
*rt_intern_hash(unsigned int hash
, struct rtable
*rt
,
1146 struct sk_buff
*skb
, int ifindex
)
1148 struct rtable
*rth
, *cand
;
1149 struct rtable __rcu
**rthp
, **candp
;
1153 int attempts
= !in_softirq();
1157 min_score
= ~(u32
)0;
1162 if (!rt_caching(dev_net(rt
->dst
.dev
))) {
1164 * If we're not caching, just tell the caller we
1165 * were successful and don't touch the route. The
1166 * caller hold the sole reference to the cache entry, and
1167 * it will be released when the caller is done with it.
1168 * If we drop it here, the callers have no way to resolve routes
1169 * when we're not caching. Instead, just point *rp at rt, so
1170 * the caller gets a single use out of the route
1171 * Note that we do rt_free on this new route entry, so that
1172 * once its refcount hits zero, we are still able to reap it
1174 * Note: To avoid expensive rcu stuff for this uncached dst,
1175 * we set DST_NOCACHE so that dst_release() can free dst without
1176 * waiting a grace period.
1179 rt
->dst
.flags
|= DST_NOCACHE
;
1180 if (rt
->rt_type
== RTN_UNICAST
|| rt_is_output_route(rt
)) {
1181 int err
= rt_bind_neighbour(rt
);
1183 net_warn_ratelimited("Neighbour table failure & not caching routes\n");
1185 return ERR_PTR(err
);
1192 rthp
= &rt_hash_table
[hash
].chain
;
1194 spin_lock_bh(rt_hash_lock_addr(hash
));
1195 while ((rth
= rcu_dereference_protected(*rthp
,
1196 lockdep_is_held(rt_hash_lock_addr(hash
)))) != NULL
) {
1197 if (rt_is_expired(rth
)) {
1198 *rthp
= rth
->dst
.rt_next
;
1202 if (compare_keys(rth
, rt
) && compare_netns(rth
, rt
)) {
1204 *rthp
= rth
->dst
.rt_next
;
1206 * Since lookup is lockfree, the deletion
1207 * must be visible to another weakly ordered CPU before
1208 * the insertion at the start of the hash chain.
1210 rcu_assign_pointer(rth
->dst
.rt_next
,
1211 rt_hash_table
[hash
].chain
);
1213 * Since lookup is lockfree, the update writes
1214 * must be ordered for consistency on SMP.
1216 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rth
);
1218 dst_use(&rth
->dst
, now
);
1219 spin_unlock_bh(rt_hash_lock_addr(hash
));
1223 skb_dst_set(skb
, &rth
->dst
);
1227 if (!atomic_read(&rth
->dst
.__refcnt
)) {
1228 u32 score
= rt_score(rth
);
1230 if (score
<= min_score
) {
1239 rthp
= &rth
->dst
.rt_next
;
1243 /* ip_rt_gc_elasticity used to be average length of chain
1244 * length, when exceeded gc becomes really aggressive.
1246 * The second limit is less certain. At the moment it allows
1247 * only 2 entries per bucket. We will see.
1249 if (chain_length
> ip_rt_gc_elasticity
) {
1250 *candp
= cand
->dst
.rt_next
;
1254 if (chain_length
> rt_chain_length_max
&&
1255 slow_chain_length(rt_hash_table
[hash
].chain
) > rt_chain_length_max
) {
1256 struct net
*net
= dev_net(rt
->dst
.dev
);
1257 int num
= ++net
->ipv4
.current_rt_cache_rebuild_count
;
1258 if (!rt_caching(net
)) {
1259 pr_warn("%s: %d rebuilds is over limit, route caching disabled\n",
1260 rt
->dst
.dev
->name
, num
);
1262 rt_emergency_hash_rebuild(net
);
1263 spin_unlock_bh(rt_hash_lock_addr(hash
));
1265 hash
= rt_hash(rt
->rt_key_dst
, rt
->rt_key_src
,
1266 ifindex
, rt_genid(net
));
1271 /* Try to bind route to arp only if it is output
1272 route or unicast forwarding path.
1274 if (rt
->rt_type
== RTN_UNICAST
|| rt_is_output_route(rt
)) {
1275 int err
= rt_bind_neighbour(rt
);
1277 spin_unlock_bh(rt_hash_lock_addr(hash
));
1279 if (err
!= -ENOBUFS
) {
1281 return ERR_PTR(err
);
1284 /* Neighbour tables are full and nothing
1285 can be released. Try to shrink route cache,
1286 it is most likely it holds some neighbour records.
1288 if (attempts
-- > 0) {
1289 int saved_elasticity
= ip_rt_gc_elasticity
;
1290 int saved_int
= ip_rt_gc_min_interval
;
1291 ip_rt_gc_elasticity
= 1;
1292 ip_rt_gc_min_interval
= 0;
1293 rt_garbage_collect(&ipv4_dst_ops
);
1294 ip_rt_gc_min_interval
= saved_int
;
1295 ip_rt_gc_elasticity
= saved_elasticity
;
1299 net_warn_ratelimited("Neighbour table overflow\n");
1301 return ERR_PTR(-ENOBUFS
);
1305 rt
->dst
.rt_next
= rt_hash_table
[hash
].chain
;
1308 * Since lookup is lockfree, we must make sure
1309 * previous writes to rt are committed to memory
1310 * before making rt visible to other CPUS.
1312 rcu_assign_pointer(rt_hash_table
[hash
].chain
, rt
);
1314 spin_unlock_bh(rt_hash_lock_addr(hash
));
1318 skb_dst_set(skb
, &rt
->dst
);
1322 static atomic_t __rt_peer_genid
= ATOMIC_INIT(0);
1324 static u32
rt_peer_genid(void)
1326 return atomic_read(&__rt_peer_genid
);
1329 void rt_bind_peer(struct rtable
*rt
, __be32 daddr
, int create
)
1331 struct inet_peer
*peer
;
1333 peer
= inet_getpeer_v4(daddr
, create
);
1335 if (peer
&& cmpxchg(&rt
->peer
, NULL
, peer
) != NULL
)
1338 rt
->rt_peer_genid
= rt_peer_genid();
1342 * Peer allocation may fail only in serious out-of-memory conditions. However
1343 * we still can generate some output.
1344 * Random ID selection looks a bit dangerous because we have no chances to
1345 * select ID being unique in a reasonable period of time.
1346 * But broken packet identifier may be better than no packet at all.
1348 static void ip_select_fb_ident(struct iphdr
*iph
)
1350 static DEFINE_SPINLOCK(ip_fb_id_lock
);
1351 static u32 ip_fallback_id
;
1354 spin_lock_bh(&ip_fb_id_lock
);
1355 salt
= secure_ip_id((__force __be32
)ip_fallback_id
^ iph
->daddr
);
1356 iph
->id
= htons(salt
& 0xFFFF);
1357 ip_fallback_id
= salt
;
1358 spin_unlock_bh(&ip_fb_id_lock
);
1361 void __ip_select_ident(struct iphdr
*iph
, struct dst_entry
*dst
, int more
)
1363 struct rtable
*rt
= (struct rtable
*) dst
;
1365 if (rt
&& !(rt
->dst
.flags
& DST_NOPEER
)) {
1366 if (rt
->peer
== NULL
)
1367 rt_bind_peer(rt
, rt
->rt_dst
, 1);
1369 /* If peer is attached to destination, it is never detached,
1370 so that we need not to grab a lock to dereference it.
1373 iph
->id
= htons(inet_getid(rt
->peer
, more
));
1377 pr_debug("rt_bind_peer(0) @%p\n", __builtin_return_address(0));
1379 ip_select_fb_ident(iph
);
1381 EXPORT_SYMBOL(__ip_select_ident
);
1383 static void rt_del(unsigned int hash
, struct rtable
*rt
)
1385 struct rtable __rcu
**rthp
;
1388 rthp
= &rt_hash_table
[hash
].chain
;
1389 spin_lock_bh(rt_hash_lock_addr(hash
));
1391 while ((aux
= rcu_dereference_protected(*rthp
,
1392 lockdep_is_held(rt_hash_lock_addr(hash
)))) != NULL
) {
1393 if (aux
== rt
|| rt_is_expired(aux
)) {
1394 *rthp
= aux
->dst
.rt_next
;
1398 rthp
= &aux
->dst
.rt_next
;
1400 spin_unlock_bh(rt_hash_lock_addr(hash
));
1403 static void check_peer_redir(struct dst_entry
*dst
, struct inet_peer
*peer
)
1405 struct rtable
*rt
= (struct rtable
*) dst
;
1406 __be32 orig_gw
= rt
->rt_gateway
;
1407 struct neighbour
*n
, *old_n
;
1409 dst_confirm(&rt
->dst
);
1411 rt
->rt_gateway
= peer
->redirect_learned
.a4
;
1413 n
= ipv4_neigh_lookup(&rt
->dst
, &rt
->rt_gateway
);
1415 rt
->rt_gateway
= orig_gw
;
1418 old_n
= xchg(&rt
->dst
._neighbour
, n
);
1420 neigh_release(old_n
);
1421 if (!(n
->nud_state
& NUD_VALID
)) {
1422 neigh_event_send(n
, NULL
);
1424 rt
->rt_flags
|= RTCF_REDIRECTED
;
1425 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE
, n
);
1429 /* called in rcu_read_lock() section */
1430 void ip_rt_redirect(__be32 old_gw
, __be32 daddr
, __be32 new_gw
,
1431 __be32 saddr
, struct net_device
*dev
)
1434 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
1435 __be32 skeys
[2] = { saddr
, 0 };
1436 int ikeys
[2] = { dev
->ifindex
, 0 };
1437 struct inet_peer
*peer
;
1444 if (new_gw
== old_gw
|| !IN_DEV_RX_REDIRECTS(in_dev
) ||
1445 ipv4_is_multicast(new_gw
) || ipv4_is_lbcast(new_gw
) ||
1446 ipv4_is_zeronet(new_gw
))
1447 goto reject_redirect
;
1449 if (!IN_DEV_SHARED_MEDIA(in_dev
)) {
1450 if (!inet_addr_onlink(in_dev
, new_gw
, old_gw
))
1451 goto reject_redirect
;
1452 if (IN_DEV_SEC_REDIRECTS(in_dev
) && ip_fib_check_default(new_gw
, dev
))
1453 goto reject_redirect
;
1455 if (inet_addr_type(net
, new_gw
) != RTN_UNICAST
)
1456 goto reject_redirect
;
1459 for (s
= 0; s
< 2; s
++) {
1460 for (i
= 0; i
< 2; i
++) {
1462 struct rtable __rcu
**rthp
;
1465 hash
= rt_hash(daddr
, skeys
[s
], ikeys
[i
], rt_genid(net
));
1467 rthp
= &rt_hash_table
[hash
].chain
;
1469 while ((rt
= rcu_dereference(*rthp
)) != NULL
) {
1470 rthp
= &rt
->dst
.rt_next
;
1472 if (rt
->rt_key_dst
!= daddr
||
1473 rt
->rt_key_src
!= skeys
[s
] ||
1474 rt
->rt_oif
!= ikeys
[i
] ||
1475 rt_is_input_route(rt
) ||
1476 rt_is_expired(rt
) ||
1477 !net_eq(dev_net(rt
->dst
.dev
), net
) ||
1479 rt
->dst
.dev
!= dev
||
1480 rt
->rt_gateway
!= old_gw
)
1484 rt_bind_peer(rt
, rt
->rt_dst
, 1);
1488 if (peer
->redirect_learned
.a4
!= new_gw
) {
1489 peer
->redirect_learned
.a4
= new_gw
;
1490 atomic_inc(&__rt_peer_genid
);
1492 check_peer_redir(&rt
->dst
, peer
);
1500 #ifdef CONFIG_IP_ROUTE_VERBOSE
1501 if (IN_DEV_LOG_MARTIANS(in_dev
))
1502 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
1503 " Advised path = %pI4 -> %pI4\n",
1504 &old_gw
, dev
->name
, &new_gw
,
1510 static bool peer_pmtu_expired(struct inet_peer
*peer
)
1512 unsigned long orig
= ACCESS_ONCE(peer
->pmtu_expires
);
1515 time_after_eq(jiffies
, orig
) &&
1516 cmpxchg(&peer
->pmtu_expires
, orig
, 0) == orig
;
1519 static bool peer_pmtu_cleaned(struct inet_peer
*peer
)
1521 unsigned long orig
= ACCESS_ONCE(peer
->pmtu_expires
);
1524 cmpxchg(&peer
->pmtu_expires
, orig
, 0) == orig
;
1527 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
)
1529 struct rtable
*rt
= (struct rtable
*)dst
;
1530 struct dst_entry
*ret
= dst
;
1533 if (dst
->obsolete
> 0) {
1536 } else if (rt
->rt_flags
& RTCF_REDIRECTED
) {
1537 unsigned int hash
= rt_hash(rt
->rt_key_dst
, rt
->rt_key_src
,
1539 rt_genid(dev_net(dst
->dev
)));
1542 } else if (rt
->peer
&& peer_pmtu_expired(rt
->peer
)) {
1543 dst_metric_set(dst
, RTAX_MTU
, rt
->peer
->pmtu_orig
);
1551 * 1. The first ip_rt_redirect_number redirects are sent
1552 * with exponential backoff, then we stop sending them at all,
1553 * assuming that the host ignores our redirects.
1554 * 2. If we did not see packets requiring redirects
1555 * during ip_rt_redirect_silence, we assume that the host
1556 * forgot redirected route and start to send redirects again.
1558 * This algorithm is much cheaper and more intelligent than dumb load limiting
1561 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1562 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1565 void ip_rt_send_redirect(struct sk_buff
*skb
)
1567 struct rtable
*rt
= skb_rtable(skb
);
1568 struct in_device
*in_dev
;
1569 struct inet_peer
*peer
;
1573 in_dev
= __in_dev_get_rcu(rt
->dst
.dev
);
1574 if (!in_dev
|| !IN_DEV_TX_REDIRECTS(in_dev
)) {
1578 log_martians
= IN_DEV_LOG_MARTIANS(in_dev
);
1582 rt_bind_peer(rt
, rt
->rt_dst
, 1);
1585 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
, rt
->rt_gateway
);
1589 /* No redirected packets during ip_rt_redirect_silence;
1590 * reset the algorithm.
1592 if (time_after(jiffies
, peer
->rate_last
+ ip_rt_redirect_silence
))
1593 peer
->rate_tokens
= 0;
1595 /* Too many ignored redirects; do not send anything
1596 * set dst.rate_last to the last seen redirected packet.
1598 if (peer
->rate_tokens
>= ip_rt_redirect_number
) {
1599 peer
->rate_last
= jiffies
;
1603 /* Check for load limit; set rate_last to the latest sent
1606 if (peer
->rate_tokens
== 0 ||
1609 (ip_rt_redirect_load
<< peer
->rate_tokens
)))) {
1610 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
, rt
->rt_gateway
);
1611 peer
->rate_last
= jiffies
;
1612 ++peer
->rate_tokens
;
1613 #ifdef CONFIG_IP_ROUTE_VERBOSE
1615 peer
->rate_tokens
== ip_rt_redirect_number
)
1616 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
1617 &ip_hdr(skb
)->saddr
, rt
->rt_iif
,
1618 &rt
->rt_dst
, &rt
->rt_gateway
);
1623 static int ip_error(struct sk_buff
*skb
)
1625 struct rtable
*rt
= skb_rtable(skb
);
1626 struct inet_peer
*peer
;
1631 switch (rt
->dst
.error
) {
1636 code
= ICMP_HOST_UNREACH
;
1639 code
= ICMP_NET_UNREACH
;
1640 IP_INC_STATS_BH(dev_net(rt
->dst
.dev
),
1641 IPSTATS_MIB_INNOROUTES
);
1644 code
= ICMP_PKT_FILTERED
;
1649 rt_bind_peer(rt
, rt
->rt_dst
, 1);
1655 peer
->rate_tokens
+= now
- peer
->rate_last
;
1656 if (peer
->rate_tokens
> ip_rt_error_burst
)
1657 peer
->rate_tokens
= ip_rt_error_burst
;
1658 peer
->rate_last
= now
;
1659 if (peer
->rate_tokens
>= ip_rt_error_cost
)
1660 peer
->rate_tokens
-= ip_rt_error_cost
;
1665 icmp_send(skb
, ICMP_DEST_UNREACH
, code
, 0);
1667 out
: kfree_skb(skb
);
1672 * The last two values are not from the RFC but
1673 * are needed for AMPRnet AX.25 paths.
1676 static const unsigned short mtu_plateau
[] =
1677 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1679 static inline unsigned short guess_mtu(unsigned short old_mtu
)
1683 for (i
= 0; i
< ARRAY_SIZE(mtu_plateau
); i
++)
1684 if (old_mtu
> mtu_plateau
[i
])
1685 return mtu_plateau
[i
];
1689 unsigned short ip_rt_frag_needed(struct net
*net
, const struct iphdr
*iph
,
1690 unsigned short new_mtu
,
1691 struct net_device
*dev
)
1693 unsigned short old_mtu
= ntohs(iph
->tot_len
);
1694 unsigned short est_mtu
= 0;
1695 struct inet_peer
*peer
;
1697 peer
= inet_getpeer_v4(iph
->daddr
, 1);
1699 unsigned short mtu
= new_mtu
;
1701 if (new_mtu
< 68 || new_mtu
>= old_mtu
) {
1702 /* BSD 4.2 derived systems incorrectly adjust
1703 * tot_len by the IP header length, and report
1704 * a zero MTU in the ICMP message.
1707 old_mtu
>= 68 + (iph
->ihl
<< 2))
1708 old_mtu
-= iph
->ihl
<< 2;
1709 mtu
= guess_mtu(old_mtu
);
1712 if (mtu
< ip_rt_min_pmtu
)
1713 mtu
= ip_rt_min_pmtu
;
1714 if (!peer
->pmtu_expires
|| mtu
< peer
->pmtu_learned
) {
1715 unsigned long pmtu_expires
;
1717 pmtu_expires
= jiffies
+ ip_rt_mtu_expires
;
1722 peer
->pmtu_learned
= mtu
;
1723 peer
->pmtu_expires
= pmtu_expires
;
1724 atomic_inc(&__rt_peer_genid
);
1729 return est_mtu
? : new_mtu
;
1732 static void check_peer_pmtu(struct dst_entry
*dst
, struct inet_peer
*peer
)
1734 unsigned long expires
= ACCESS_ONCE(peer
->pmtu_expires
);
1738 if (time_before(jiffies
, expires
)) {
1739 u32 orig_dst_mtu
= dst_mtu(dst
);
1740 if (peer
->pmtu_learned
< orig_dst_mtu
) {
1741 if (!peer
->pmtu_orig
)
1742 peer
->pmtu_orig
= dst_metric_raw(dst
, RTAX_MTU
);
1743 dst_metric_set(dst
, RTAX_MTU
, peer
->pmtu_learned
);
1745 } else if (cmpxchg(&peer
->pmtu_expires
, expires
, 0) == expires
)
1746 dst_metric_set(dst
, RTAX_MTU
, peer
->pmtu_orig
);
1749 static void ip_rt_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
1751 struct rtable
*rt
= (struct rtable
*) dst
;
1752 struct inet_peer
*peer
;
1757 rt_bind_peer(rt
, rt
->rt_dst
, 1);
1760 unsigned long pmtu_expires
= ACCESS_ONCE(peer
->pmtu_expires
);
1762 if (mtu
< ip_rt_min_pmtu
)
1763 mtu
= ip_rt_min_pmtu
;
1764 if (!pmtu_expires
|| mtu
< peer
->pmtu_learned
) {
1766 pmtu_expires
= jiffies
+ ip_rt_mtu_expires
;
1770 peer
->pmtu_learned
= mtu
;
1771 peer
->pmtu_expires
= pmtu_expires
;
1773 atomic_inc(&__rt_peer_genid
);
1774 rt
->rt_peer_genid
= rt_peer_genid();
1776 check_peer_pmtu(dst
, peer
);
1781 static void ipv4_validate_peer(struct rtable
*rt
)
1783 if (rt
->rt_peer_genid
!= rt_peer_genid()) {
1784 struct inet_peer
*peer
;
1787 rt_bind_peer(rt
, rt
->rt_dst
, 0);
1791 check_peer_pmtu(&rt
->dst
, peer
);
1793 if (peer
->redirect_learned
.a4
&&
1794 peer
->redirect_learned
.a4
!= rt
->rt_gateway
)
1795 check_peer_redir(&rt
->dst
, peer
);
1798 rt
->rt_peer_genid
= rt_peer_genid();
1802 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
)
1804 struct rtable
*rt
= (struct rtable
*) dst
;
1806 if (rt_is_expired(rt
))
1808 ipv4_validate_peer(rt
);
1812 static void ipv4_dst_destroy(struct dst_entry
*dst
)
1814 struct rtable
*rt
= (struct rtable
*) dst
;
1815 struct inet_peer
*peer
= rt
->peer
;
1818 fib_info_put(rt
->fi
);
1828 static void ipv4_link_failure(struct sk_buff
*skb
)
1832 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_HOST_UNREACH
, 0);
1834 rt
= skb_rtable(skb
);
1835 if (rt
&& rt
->peer
&& peer_pmtu_cleaned(rt
->peer
))
1836 dst_metric_set(&rt
->dst
, RTAX_MTU
, rt
->peer
->pmtu_orig
);
1839 static int ip_rt_bug(struct sk_buff
*skb
)
1841 pr_debug("%s: %pI4 -> %pI4, %s\n",
1842 __func__
, &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1843 skb
->dev
? skb
->dev
->name
: "?");
1850 We do not cache source address of outgoing interface,
1851 because it is used only by IP RR, TS and SRR options,
1852 so that it out of fast path.
1854 BTW remember: "addr" is allowed to be not aligned
1858 void ip_rt_get_source(u8
*addr
, struct sk_buff
*skb
, struct rtable
*rt
)
1862 if (rt_is_output_route(rt
))
1863 src
= ip_hdr(skb
)->saddr
;
1865 struct fib_result res
;
1871 memset(&fl4
, 0, sizeof(fl4
));
1872 fl4
.daddr
= iph
->daddr
;
1873 fl4
.saddr
= iph
->saddr
;
1874 fl4
.flowi4_tos
= RT_TOS(iph
->tos
);
1875 fl4
.flowi4_oif
= rt
->dst
.dev
->ifindex
;
1876 fl4
.flowi4_iif
= skb
->dev
->ifindex
;
1877 fl4
.flowi4_mark
= skb
->mark
;
1880 if (fib_lookup(dev_net(rt
->dst
.dev
), &fl4
, &res
) == 0)
1881 src
= FIB_RES_PREFSRC(dev_net(rt
->dst
.dev
), res
);
1883 src
= inet_select_addr(rt
->dst
.dev
, rt
->rt_gateway
,
1887 memcpy(addr
, &src
, 4);
1890 #ifdef CONFIG_IP_ROUTE_CLASSID
1891 static void set_class_tag(struct rtable
*rt
, u32 tag
)
1893 if (!(rt
->dst
.tclassid
& 0xFFFF))
1894 rt
->dst
.tclassid
|= tag
& 0xFFFF;
1895 if (!(rt
->dst
.tclassid
& 0xFFFF0000))
1896 rt
->dst
.tclassid
|= tag
& 0xFFFF0000;
1900 static unsigned int ipv4_default_advmss(const struct dst_entry
*dst
)
1902 unsigned int advmss
= dst_metric_raw(dst
, RTAX_ADVMSS
);
1905 advmss
= max_t(unsigned int, dst
->dev
->mtu
- 40,
1907 if (advmss
> 65535 - 40)
1908 advmss
= 65535 - 40;
1913 static unsigned int ipv4_mtu(const struct dst_entry
*dst
)
1915 const struct rtable
*rt
= (const struct rtable
*) dst
;
1916 unsigned int mtu
= dst_metric_raw(dst
, RTAX_MTU
);
1918 if (mtu
&& rt_is_output_route(rt
))
1921 mtu
= dst
->dev
->mtu
;
1923 if (unlikely(dst_metric_locked(dst
, RTAX_MTU
))) {
1925 if (rt
->rt_gateway
!= rt
->rt_dst
&& mtu
> 576)
1929 if (mtu
> IP_MAX_MTU
)
1935 static void rt_init_metrics(struct rtable
*rt
, const struct flowi4
*fl4
,
1936 struct fib_info
*fi
)
1938 struct inet_peer
*peer
;
1941 /* If a peer entry exists for this destination, we must hook
1942 * it up in order to get at cached metrics.
1944 if (fl4
&& (fl4
->flowi4_flags
& FLOWI_FLAG_PRECOW_METRICS
))
1947 rt
->peer
= peer
= inet_getpeer_v4(rt
->rt_dst
, create
);
1949 rt
->rt_peer_genid
= rt_peer_genid();
1950 if (inet_metrics_new(peer
))
1951 memcpy(peer
->metrics
, fi
->fib_metrics
,
1952 sizeof(u32
) * RTAX_MAX
);
1953 dst_init_metrics(&rt
->dst
, peer
->metrics
, false);
1955 check_peer_pmtu(&rt
->dst
, peer
);
1957 if (peer
->redirect_learned
.a4
&&
1958 peer
->redirect_learned
.a4
!= rt
->rt_gateway
) {
1959 rt
->rt_gateway
= peer
->redirect_learned
.a4
;
1960 rt
->rt_flags
|= RTCF_REDIRECTED
;
1963 if (fi
->fib_metrics
!= (u32
*) dst_default_metrics
) {
1965 atomic_inc(&fi
->fib_clntref
);
1967 dst_init_metrics(&rt
->dst
, fi
->fib_metrics
, true);
1971 static void rt_set_nexthop(struct rtable
*rt
, const struct flowi4
*fl4
,
1972 const struct fib_result
*res
,
1973 struct fib_info
*fi
, u16 type
, u32 itag
)
1975 struct dst_entry
*dst
= &rt
->dst
;
1978 if (FIB_RES_GW(*res
) &&
1979 FIB_RES_NH(*res
).nh_scope
== RT_SCOPE_LINK
)
1980 rt
->rt_gateway
= FIB_RES_GW(*res
);
1981 rt_init_metrics(rt
, fl4
, fi
);
1982 #ifdef CONFIG_IP_ROUTE_CLASSID
1983 dst
->tclassid
= FIB_RES_NH(*res
).nh_tclassid
;
1987 if (dst_mtu(dst
) > IP_MAX_MTU
)
1988 dst_metric_set(dst
, RTAX_MTU
, IP_MAX_MTU
);
1989 if (dst_metric_raw(dst
, RTAX_ADVMSS
) > 65535 - 40)
1990 dst_metric_set(dst
, RTAX_ADVMSS
, 65535 - 40);
1992 #ifdef CONFIG_IP_ROUTE_CLASSID
1993 #ifdef CONFIG_IP_MULTIPLE_TABLES
1994 set_class_tag(rt
, fib_rules_tclass(res
));
1996 set_class_tag(rt
, itag
);
2000 static struct rtable
*rt_dst_alloc(struct net_device
*dev
,
2001 bool nopolicy
, bool noxfrm
)
2003 return dst_alloc(&ipv4_dst_ops
, dev
, 1, -1,
2005 (nopolicy
? DST_NOPOLICY
: 0) |
2006 (noxfrm
? DST_NOXFRM
: 0));
2009 /* called in rcu_read_lock() section */
2010 static int ip_route_input_mc(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2011 u8 tos
, struct net_device
*dev
, int our
)
2016 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
2020 /* Primary sanity checks. */
2025 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
2026 ipv4_is_loopback(saddr
) || skb
->protocol
!= htons(ETH_P_IP
))
2029 if (ipv4_is_zeronet(saddr
)) {
2030 if (!ipv4_is_local_multicast(daddr
))
2032 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
2034 err
= fib_validate_source(skb
, saddr
, 0, tos
, 0, dev
, &spec_dst
,
2039 rth
= rt_dst_alloc(dev_net(dev
)->loopback_dev
,
2040 IN_DEV_CONF_GET(in_dev
, NOPOLICY
), false);
2044 #ifdef CONFIG_IP_ROUTE_CLASSID
2045 rth
->dst
.tclassid
= itag
;
2047 rth
->dst
.output
= ip_rt_bug
;
2049 rth
->rt_key_dst
= daddr
;
2050 rth
->rt_key_src
= saddr
;
2051 rth
->rt_genid
= rt_genid(dev_net(dev
));
2052 rth
->rt_flags
= RTCF_MULTICAST
;
2053 rth
->rt_type
= RTN_MULTICAST
;
2054 rth
->rt_key_tos
= tos
;
2055 rth
->rt_dst
= daddr
;
2056 rth
->rt_src
= saddr
;
2057 rth
->rt_route_iif
= dev
->ifindex
;
2058 rth
->rt_iif
= dev
->ifindex
;
2060 rth
->rt_mark
= skb
->mark
;
2061 rth
->rt_gateway
= daddr
;
2062 rth
->rt_spec_dst
= spec_dst
;
2063 rth
->rt_peer_genid
= 0;
2067 rth
->dst
.input
= ip_local_deliver
;
2068 rth
->rt_flags
|= RTCF_LOCAL
;
2071 #ifdef CONFIG_IP_MROUTE
2072 if (!ipv4_is_local_multicast(daddr
) && IN_DEV_MFORWARD(in_dev
))
2073 rth
->dst
.input
= ip_mr_input
;
2075 RT_CACHE_STAT_INC(in_slow_mc
);
2077 hash
= rt_hash(daddr
, saddr
, dev
->ifindex
, rt_genid(dev_net(dev
)));
2078 rth
= rt_intern_hash(hash
, rth
, skb
, dev
->ifindex
);
2079 return IS_ERR(rth
) ? PTR_ERR(rth
) : 0;
2090 static void ip_handle_martian_source(struct net_device
*dev
,
2091 struct in_device
*in_dev
,
2092 struct sk_buff
*skb
,
2096 RT_CACHE_STAT_INC(in_martian_src
);
2097 #ifdef CONFIG_IP_ROUTE_VERBOSE
2098 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit()) {
2100 * RFC1812 recommendation, if source is martian,
2101 * the only hint is MAC header.
2103 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
2104 &daddr
, &saddr
, dev
->name
);
2105 if (dev
->hard_header_len
&& skb_mac_header_was_set(skb
)) {
2106 print_hex_dump(KERN_WARNING
, "ll header: ",
2107 DUMP_PREFIX_OFFSET
, 16, 1,
2108 skb_mac_header(skb
),
2109 dev
->hard_header_len
, true);
2115 /* called in rcu_read_lock() section */
2116 static int __mkroute_input(struct sk_buff
*skb
,
2117 const struct fib_result
*res
,
2118 struct in_device
*in_dev
,
2119 __be32 daddr
, __be32 saddr
, u32 tos
,
2120 struct rtable
**result
)
2124 struct in_device
*out_dev
;
2125 unsigned int flags
= 0;
2129 /* get a working reference to the output device */
2130 out_dev
= __in_dev_get_rcu(FIB_RES_DEV(*res
));
2131 if (out_dev
== NULL
) {
2132 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
2137 err
= fib_validate_source(skb
, saddr
, daddr
, tos
, FIB_RES_OIF(*res
),
2138 in_dev
->dev
, &spec_dst
, &itag
);
2140 ip_handle_martian_source(in_dev
->dev
, in_dev
, skb
, daddr
,
2147 flags
|= RTCF_DIRECTSRC
;
2149 if (out_dev
== in_dev
&& err
&&
2150 (IN_DEV_SHARED_MEDIA(out_dev
) ||
2151 inet_addr_onlink(out_dev
, saddr
, FIB_RES_GW(*res
))))
2152 flags
|= RTCF_DOREDIRECT
;
2154 if (skb
->protocol
!= htons(ETH_P_IP
)) {
2155 /* Not IP (i.e. ARP). Do not create route, if it is
2156 * invalid for proxy arp. DNAT routes are always valid.
2158 * Proxy arp feature have been extended to allow, ARP
2159 * replies back to the same interface, to support
2160 * Private VLAN switch technologies. See arp.c.
2162 if (out_dev
== in_dev
&&
2163 IN_DEV_PROXY_ARP_PVLAN(in_dev
) == 0) {
2169 rth
= rt_dst_alloc(out_dev
->dev
,
2170 IN_DEV_CONF_GET(in_dev
, NOPOLICY
),
2171 IN_DEV_CONF_GET(out_dev
, NOXFRM
));
2177 rth
->rt_key_dst
= daddr
;
2178 rth
->rt_key_src
= saddr
;
2179 rth
->rt_genid
= rt_genid(dev_net(rth
->dst
.dev
));
2180 rth
->rt_flags
= flags
;
2181 rth
->rt_type
= res
->type
;
2182 rth
->rt_key_tos
= tos
;
2183 rth
->rt_dst
= daddr
;
2184 rth
->rt_src
= saddr
;
2185 rth
->rt_route_iif
= in_dev
->dev
->ifindex
;
2186 rth
->rt_iif
= in_dev
->dev
->ifindex
;
2188 rth
->rt_mark
= skb
->mark
;
2189 rth
->rt_gateway
= daddr
;
2190 rth
->rt_spec_dst
= spec_dst
;
2191 rth
->rt_peer_genid
= 0;
2195 rth
->dst
.input
= ip_forward
;
2196 rth
->dst
.output
= ip_output
;
2198 rt_set_nexthop(rth
, NULL
, res
, res
->fi
, res
->type
, itag
);
2206 static int ip_mkroute_input(struct sk_buff
*skb
,
2207 struct fib_result
*res
,
2208 const struct flowi4
*fl4
,
2209 struct in_device
*in_dev
,
2210 __be32 daddr
, __be32 saddr
, u32 tos
)
2212 struct rtable
*rth
= NULL
;
2216 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2217 if (res
->fi
&& res
->fi
->fib_nhs
> 1)
2218 fib_select_multipath(res
);
2221 /* create a routing cache entry */
2222 err
= __mkroute_input(skb
, res
, in_dev
, daddr
, saddr
, tos
, &rth
);
2226 /* put it into the cache */
2227 hash
= rt_hash(daddr
, saddr
, fl4
->flowi4_iif
,
2228 rt_genid(dev_net(rth
->dst
.dev
)));
2229 rth
= rt_intern_hash(hash
, rth
, skb
, fl4
->flowi4_iif
);
2231 return PTR_ERR(rth
);
2236 * NOTE. We drop all the packets that has local source
2237 * addresses, because every properly looped back packet
2238 * must have correct destination already attached by output routine.
2240 * Such approach solves two big problems:
2241 * 1. Not simplex devices are handled properly.
2242 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2243 * called with rcu_read_lock()
2246 static int ip_route_input_slow(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2247 u8 tos
, struct net_device
*dev
)
2249 struct fib_result res
;
2250 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
2252 unsigned int flags
= 0;
2258 struct net
*net
= dev_net(dev
);
2260 /* IP on this device is disabled. */
2265 /* Check for the most weird martians, which can be not detected
2269 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
2270 ipv4_is_loopback(saddr
))
2271 goto martian_source
;
2273 if (ipv4_is_lbcast(daddr
) || (saddr
== 0 && daddr
== 0))
2276 /* Accept zero addresses only to limited broadcast;
2277 * I even do not know to fix it or not. Waiting for complains :-)
2279 if (ipv4_is_zeronet(saddr
))
2280 goto martian_source
;
2282 if (ipv4_is_zeronet(daddr
) || ipv4_is_loopback(daddr
))
2283 goto martian_destination
;
2286 * Now we are ready to route packet.
2289 fl4
.flowi4_iif
= dev
->ifindex
;
2290 fl4
.flowi4_mark
= skb
->mark
;
2291 fl4
.flowi4_tos
= tos
;
2292 fl4
.flowi4_scope
= RT_SCOPE_UNIVERSE
;
2295 err
= fib_lookup(net
, &fl4
, &res
);
2297 if (!IN_DEV_FORWARD(in_dev
))
2302 RT_CACHE_STAT_INC(in_slow_tot
);
2304 if (res
.type
== RTN_BROADCAST
)
2307 if (res
.type
== RTN_LOCAL
) {
2308 err
= fib_validate_source(skb
, saddr
, daddr
, tos
,
2309 net
->loopback_dev
->ifindex
,
2310 dev
, &spec_dst
, &itag
);
2312 goto martian_source_keep_err
;
2314 flags
|= RTCF_DIRECTSRC
;
2319 if (!IN_DEV_FORWARD(in_dev
))
2321 if (res
.type
!= RTN_UNICAST
)
2322 goto martian_destination
;
2324 err
= ip_mkroute_input(skb
, &res
, &fl4
, in_dev
, daddr
, saddr
, tos
);
2328 if (skb
->protocol
!= htons(ETH_P_IP
))
2331 if (ipv4_is_zeronet(saddr
))
2332 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
2334 err
= fib_validate_source(skb
, saddr
, 0, tos
, 0, dev
, &spec_dst
,
2337 goto martian_source_keep_err
;
2339 flags
|= RTCF_DIRECTSRC
;
2341 flags
|= RTCF_BROADCAST
;
2342 res
.type
= RTN_BROADCAST
;
2343 RT_CACHE_STAT_INC(in_brd
);
2346 rth
= rt_dst_alloc(net
->loopback_dev
,
2347 IN_DEV_CONF_GET(in_dev
, NOPOLICY
), false);
2351 rth
->dst
.input
= ip_local_deliver
;
2352 rth
->dst
.output
= ip_rt_bug
;
2353 #ifdef CONFIG_IP_ROUTE_CLASSID
2354 rth
->dst
.tclassid
= itag
;
2357 rth
->rt_key_dst
= daddr
;
2358 rth
->rt_key_src
= saddr
;
2359 rth
->rt_genid
= rt_genid(net
);
2360 rth
->rt_flags
= flags
|RTCF_LOCAL
;
2361 rth
->rt_type
= res
.type
;
2362 rth
->rt_key_tos
= tos
;
2363 rth
->rt_dst
= daddr
;
2364 rth
->rt_src
= saddr
;
2365 #ifdef CONFIG_IP_ROUTE_CLASSID
2366 rth
->dst
.tclassid
= itag
;
2368 rth
->rt_route_iif
= dev
->ifindex
;
2369 rth
->rt_iif
= dev
->ifindex
;
2371 rth
->rt_mark
= skb
->mark
;
2372 rth
->rt_gateway
= daddr
;
2373 rth
->rt_spec_dst
= spec_dst
;
2374 rth
->rt_peer_genid
= 0;
2377 if (res
.type
== RTN_UNREACHABLE
) {
2378 rth
->dst
.input
= ip_error
;
2379 rth
->dst
.error
= -err
;
2380 rth
->rt_flags
&= ~RTCF_LOCAL
;
2382 hash
= rt_hash(daddr
, saddr
, fl4
.flowi4_iif
, rt_genid(net
));
2383 rth
= rt_intern_hash(hash
, rth
, skb
, fl4
.flowi4_iif
);
2390 RT_CACHE_STAT_INC(in_no_route
);
2391 spec_dst
= inet_select_addr(dev
, 0, RT_SCOPE_UNIVERSE
);
2392 res
.type
= RTN_UNREACHABLE
;
2398 * Do not cache martian addresses: they should be logged (RFC1812)
2400 martian_destination
:
2401 RT_CACHE_STAT_INC(in_martian_dst
);
2402 #ifdef CONFIG_IP_ROUTE_VERBOSE
2403 if (IN_DEV_LOG_MARTIANS(in_dev
))
2404 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2405 &daddr
, &saddr
, dev
->name
);
2409 err
= -EHOSTUNREACH
;
2422 martian_source_keep_err
:
2423 ip_handle_martian_source(dev
, in_dev
, skb
, daddr
, saddr
);
2427 int ip_route_input_common(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2428 u8 tos
, struct net_device
*dev
, bool noref
)
2432 int iif
= dev
->ifindex
;
2440 if (!rt_caching(net
))
2443 tos
&= IPTOS_RT_MASK
;
2444 hash
= rt_hash(daddr
, saddr
, iif
, rt_genid(net
));
2446 for (rth
= rcu_dereference(rt_hash_table
[hash
].chain
); rth
;
2447 rth
= rcu_dereference(rth
->dst
.rt_next
)) {
2448 if ((((__force u32
)rth
->rt_key_dst
^ (__force u32
)daddr
) |
2449 ((__force u32
)rth
->rt_key_src
^ (__force u32
)saddr
) |
2450 (rth
->rt_route_iif
^ iif
) |
2451 (rth
->rt_key_tos
^ tos
)) == 0 &&
2452 rth
->rt_mark
== skb
->mark
&&
2453 net_eq(dev_net(rth
->dst
.dev
), net
) &&
2454 !rt_is_expired(rth
)) {
2455 ipv4_validate_peer(rth
);
2457 dst_use_noref(&rth
->dst
, jiffies
);
2458 skb_dst_set_noref(skb
, &rth
->dst
);
2460 dst_use(&rth
->dst
, jiffies
);
2461 skb_dst_set(skb
, &rth
->dst
);
2463 RT_CACHE_STAT_INC(in_hit
);
2467 RT_CACHE_STAT_INC(in_hlist_search
);
2471 /* Multicast recognition logic is moved from route cache to here.
2472 The problem was that too many Ethernet cards have broken/missing
2473 hardware multicast filters :-( As result the host on multicasting
2474 network acquires a lot of useless route cache entries, sort of
2475 SDR messages from all the world. Now we try to get rid of them.
2476 Really, provided software IP multicast filter is organized
2477 reasonably (at least, hashed), it does not result in a slowdown
2478 comparing with route cache reject entries.
2479 Note, that multicast routers are not affected, because
2480 route cache entry is created eventually.
2482 if (ipv4_is_multicast(daddr
)) {
2483 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
2486 int our
= ip_check_mc_rcu(in_dev
, daddr
, saddr
,
2487 ip_hdr(skb
)->protocol
);
2489 #ifdef CONFIG_IP_MROUTE
2491 (!ipv4_is_local_multicast(daddr
) &&
2492 IN_DEV_MFORWARD(in_dev
))
2495 int res
= ip_route_input_mc(skb
, daddr
, saddr
,
2504 res
= ip_route_input_slow(skb
, daddr
, saddr
, tos
, dev
);
2508 EXPORT_SYMBOL(ip_route_input_common
);
2510 /* called with rcu_read_lock() */
2511 static struct rtable
*__mkroute_output(const struct fib_result
*res
,
2512 const struct flowi4
*fl4
,
2513 __be32 orig_daddr
, __be32 orig_saddr
,
2514 int orig_oif
, __u8 orig_rtos
,
2515 struct net_device
*dev_out
,
2518 struct fib_info
*fi
= res
->fi
;
2519 struct in_device
*in_dev
;
2520 u16 type
= res
->type
;
2523 if (ipv4_is_loopback(fl4
->saddr
) && !(dev_out
->flags
& IFF_LOOPBACK
))
2524 return ERR_PTR(-EINVAL
);
2526 if (ipv4_is_lbcast(fl4
->daddr
))
2527 type
= RTN_BROADCAST
;
2528 else if (ipv4_is_multicast(fl4
->daddr
))
2529 type
= RTN_MULTICAST
;
2530 else if (ipv4_is_zeronet(fl4
->daddr
))
2531 return ERR_PTR(-EINVAL
);
2533 if (dev_out
->flags
& IFF_LOOPBACK
)
2534 flags
|= RTCF_LOCAL
;
2536 in_dev
= __in_dev_get_rcu(dev_out
);
2538 return ERR_PTR(-EINVAL
);
2540 if (type
== RTN_BROADCAST
) {
2541 flags
|= RTCF_BROADCAST
| RTCF_LOCAL
;
2543 } else if (type
== RTN_MULTICAST
) {
2544 flags
|= RTCF_MULTICAST
| RTCF_LOCAL
;
2545 if (!ip_check_mc_rcu(in_dev
, fl4
->daddr
, fl4
->saddr
,
2547 flags
&= ~RTCF_LOCAL
;
2548 /* If multicast route do not exist use
2549 * default one, but do not gateway in this case.
2552 if (fi
&& res
->prefixlen
< 4)
2556 rth
= rt_dst_alloc(dev_out
,
2557 IN_DEV_CONF_GET(in_dev
, NOPOLICY
),
2558 IN_DEV_CONF_GET(in_dev
, NOXFRM
));
2560 return ERR_PTR(-ENOBUFS
);
2562 rth
->dst
.output
= ip_output
;
2564 rth
->rt_key_dst
= orig_daddr
;
2565 rth
->rt_key_src
= orig_saddr
;
2566 rth
->rt_genid
= rt_genid(dev_net(dev_out
));
2567 rth
->rt_flags
= flags
;
2568 rth
->rt_type
= type
;
2569 rth
->rt_key_tos
= orig_rtos
;
2570 rth
->rt_dst
= fl4
->daddr
;
2571 rth
->rt_src
= fl4
->saddr
;
2572 rth
->rt_route_iif
= 0;
2573 rth
->rt_iif
= orig_oif
? : dev_out
->ifindex
;
2574 rth
->rt_oif
= orig_oif
;
2575 rth
->rt_mark
= fl4
->flowi4_mark
;
2576 rth
->rt_gateway
= fl4
->daddr
;
2577 rth
->rt_spec_dst
= fl4
->saddr
;
2578 rth
->rt_peer_genid
= 0;
2582 RT_CACHE_STAT_INC(out_slow_tot
);
2584 if (flags
& RTCF_LOCAL
) {
2585 rth
->dst
.input
= ip_local_deliver
;
2586 rth
->rt_spec_dst
= fl4
->daddr
;
2588 if (flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) {
2589 rth
->rt_spec_dst
= fl4
->saddr
;
2590 if (flags
& RTCF_LOCAL
&&
2591 !(dev_out
->flags
& IFF_LOOPBACK
)) {
2592 rth
->dst
.output
= ip_mc_output
;
2593 RT_CACHE_STAT_INC(out_slow_mc
);
2595 #ifdef CONFIG_IP_MROUTE
2596 if (type
== RTN_MULTICAST
) {
2597 if (IN_DEV_MFORWARD(in_dev
) &&
2598 !ipv4_is_local_multicast(fl4
->daddr
)) {
2599 rth
->dst
.input
= ip_mr_input
;
2600 rth
->dst
.output
= ip_mc_output
;
2606 rt_set_nexthop(rth
, fl4
, res
, fi
, type
, 0);
2612 * Major route resolver routine.
2613 * called with rcu_read_lock();
2616 static struct rtable
*ip_route_output_slow(struct net
*net
, struct flowi4
*fl4
)
2618 struct net_device
*dev_out
= NULL
;
2619 __u8 tos
= RT_FL_TOS(fl4
);
2620 unsigned int flags
= 0;
2621 struct fib_result res
;
2628 #ifdef CONFIG_IP_MULTIPLE_TABLES
2632 orig_daddr
= fl4
->daddr
;
2633 orig_saddr
= fl4
->saddr
;
2634 orig_oif
= fl4
->flowi4_oif
;
2636 fl4
->flowi4_iif
= net
->loopback_dev
->ifindex
;
2637 fl4
->flowi4_tos
= tos
& IPTOS_RT_MASK
;
2638 fl4
->flowi4_scope
= ((tos
& RTO_ONLINK
) ?
2639 RT_SCOPE_LINK
: RT_SCOPE_UNIVERSE
);
2643 rth
= ERR_PTR(-EINVAL
);
2644 if (ipv4_is_multicast(fl4
->saddr
) ||
2645 ipv4_is_lbcast(fl4
->saddr
) ||
2646 ipv4_is_zeronet(fl4
->saddr
))
2649 /* I removed check for oif == dev_out->oif here.
2650 It was wrong for two reasons:
2651 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2652 is assigned to multiple interfaces.
2653 2. Moreover, we are allowed to send packets with saddr
2654 of another iface. --ANK
2657 if (fl4
->flowi4_oif
== 0 &&
2658 (ipv4_is_multicast(fl4
->daddr
) ||
2659 ipv4_is_lbcast(fl4
->daddr
))) {
2660 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2661 dev_out
= __ip_dev_find(net
, fl4
->saddr
, false);
2662 if (dev_out
== NULL
)
2665 /* Special hack: user can direct multicasts
2666 and limited broadcast via necessary interface
2667 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2668 This hack is not just for fun, it allows
2669 vic,vat and friends to work.
2670 They bind socket to loopback, set ttl to zero
2671 and expect that it will work.
2672 From the viewpoint of routing cache they are broken,
2673 because we are not allowed to build multicast path
2674 with loopback source addr (look, routing cache
2675 cannot know, that ttl is zero, so that packet
2676 will not leave this host and route is valid).
2677 Luckily, this hack is good workaround.
2680 fl4
->flowi4_oif
= dev_out
->ifindex
;
2684 if (!(fl4
->flowi4_flags
& FLOWI_FLAG_ANYSRC
)) {
2685 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2686 if (!__ip_dev_find(net
, fl4
->saddr
, false))
2692 if (fl4
->flowi4_oif
) {
2693 dev_out
= dev_get_by_index_rcu(net
, fl4
->flowi4_oif
);
2694 rth
= ERR_PTR(-ENODEV
);
2695 if (dev_out
== NULL
)
2698 /* RACE: Check return value of inet_select_addr instead. */
2699 if (!(dev_out
->flags
& IFF_UP
) || !__in_dev_get_rcu(dev_out
)) {
2700 rth
= ERR_PTR(-ENETUNREACH
);
2703 if (ipv4_is_local_multicast(fl4
->daddr
) ||
2704 ipv4_is_lbcast(fl4
->daddr
)) {
2706 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2711 if (ipv4_is_multicast(fl4
->daddr
))
2712 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2714 else if (!fl4
->daddr
)
2715 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2721 fl4
->daddr
= fl4
->saddr
;
2723 fl4
->daddr
= fl4
->saddr
= htonl(INADDR_LOOPBACK
);
2724 dev_out
= net
->loopback_dev
;
2725 fl4
->flowi4_oif
= net
->loopback_dev
->ifindex
;
2726 res
.type
= RTN_LOCAL
;
2727 flags
|= RTCF_LOCAL
;
2731 if (fib_lookup(net
, fl4
, &res
)) {
2733 if (fl4
->flowi4_oif
) {
2734 /* Apparently, routing tables are wrong. Assume,
2735 that the destination is on link.
2738 Because we are allowed to send to iface
2739 even if it has NO routes and NO assigned
2740 addresses. When oif is specified, routing
2741 tables are looked up with only one purpose:
2742 to catch if destination is gatewayed, rather than
2743 direct. Moreover, if MSG_DONTROUTE is set,
2744 we send packet, ignoring both routing tables
2745 and ifaddr state. --ANK
2748 We could make it even if oif is unknown,
2749 likely IPv6, but we do not.
2752 if (fl4
->saddr
== 0)
2753 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2755 res
.type
= RTN_UNICAST
;
2758 rth
= ERR_PTR(-ENETUNREACH
);
2762 if (res
.type
== RTN_LOCAL
) {
2764 if (res
.fi
->fib_prefsrc
)
2765 fl4
->saddr
= res
.fi
->fib_prefsrc
;
2767 fl4
->saddr
= fl4
->daddr
;
2769 dev_out
= net
->loopback_dev
;
2770 fl4
->flowi4_oif
= dev_out
->ifindex
;
2772 flags
|= RTCF_LOCAL
;
2776 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2777 if (res
.fi
->fib_nhs
> 1 && fl4
->flowi4_oif
== 0)
2778 fib_select_multipath(&res
);
2781 if (!res
.prefixlen
&&
2782 res
.table
->tb_num_default
> 1 &&
2783 res
.type
== RTN_UNICAST
&& !fl4
->flowi4_oif
)
2784 fib_select_default(&res
);
2787 fl4
->saddr
= FIB_RES_PREFSRC(net
, res
);
2789 dev_out
= FIB_RES_DEV(res
);
2790 fl4
->flowi4_oif
= dev_out
->ifindex
;
2794 rth
= __mkroute_output(&res
, fl4
, orig_daddr
, orig_saddr
, orig_oif
,
2795 tos
, dev_out
, flags
);
2799 hash
= rt_hash(orig_daddr
, orig_saddr
, orig_oif
,
2800 rt_genid(dev_net(dev_out
)));
2801 rth
= rt_intern_hash(hash
, rth
, NULL
, orig_oif
);
2809 struct rtable
*__ip_route_output_key(struct net
*net
, struct flowi4
*flp4
)
2814 if (!rt_caching(net
))
2817 hash
= rt_hash(flp4
->daddr
, flp4
->saddr
, flp4
->flowi4_oif
, rt_genid(net
));
2820 for (rth
= rcu_dereference_bh(rt_hash_table
[hash
].chain
); rth
;
2821 rth
= rcu_dereference_bh(rth
->dst
.rt_next
)) {
2822 if (rth
->rt_key_dst
== flp4
->daddr
&&
2823 rth
->rt_key_src
== flp4
->saddr
&&
2824 rt_is_output_route(rth
) &&
2825 rth
->rt_oif
== flp4
->flowi4_oif
&&
2826 rth
->rt_mark
== flp4
->flowi4_mark
&&
2827 !((rth
->rt_key_tos
^ flp4
->flowi4_tos
) &
2828 (IPTOS_RT_MASK
| RTO_ONLINK
)) &&
2829 net_eq(dev_net(rth
->dst
.dev
), net
) &&
2830 !rt_is_expired(rth
)) {
2831 ipv4_validate_peer(rth
);
2832 dst_use(&rth
->dst
, jiffies
);
2833 RT_CACHE_STAT_INC(out_hit
);
2834 rcu_read_unlock_bh();
2836 flp4
->saddr
= rth
->rt_src
;
2838 flp4
->daddr
= rth
->rt_dst
;
2841 RT_CACHE_STAT_INC(out_hlist_search
);
2843 rcu_read_unlock_bh();
2846 return ip_route_output_slow(net
, flp4
);
2848 EXPORT_SYMBOL_GPL(__ip_route_output_key
);
2850 static struct dst_entry
*ipv4_blackhole_dst_check(struct dst_entry
*dst
, u32 cookie
)
2855 static unsigned int ipv4_blackhole_mtu(const struct dst_entry
*dst
)
2857 unsigned int mtu
= dst_metric_raw(dst
, RTAX_MTU
);
2859 return mtu
? : dst
->dev
->mtu
;
2862 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry
*dst
, u32 mtu
)
2866 static u32
*ipv4_rt_blackhole_cow_metrics(struct dst_entry
*dst
,
2872 static struct dst_ops ipv4_dst_blackhole_ops
= {
2874 .protocol
= cpu_to_be16(ETH_P_IP
),
2875 .destroy
= ipv4_dst_destroy
,
2876 .check
= ipv4_blackhole_dst_check
,
2877 .mtu
= ipv4_blackhole_mtu
,
2878 .default_advmss
= ipv4_default_advmss
,
2879 .update_pmtu
= ipv4_rt_blackhole_update_pmtu
,
2880 .cow_metrics
= ipv4_rt_blackhole_cow_metrics
,
2881 .neigh_lookup
= ipv4_neigh_lookup
,
2884 struct dst_entry
*ipv4_blackhole_route(struct net
*net
, struct dst_entry
*dst_orig
)
2886 struct rtable
*rt
= dst_alloc(&ipv4_dst_blackhole_ops
, NULL
, 1, 0, 0);
2887 struct rtable
*ort
= (struct rtable
*) dst_orig
;
2890 struct dst_entry
*new = &rt
->dst
;
2893 new->input
= dst_discard
;
2894 new->output
= dst_discard
;
2895 dst_copy_metrics(new, &ort
->dst
);
2897 new->dev
= ort
->dst
.dev
;
2901 rt
->rt_key_dst
= ort
->rt_key_dst
;
2902 rt
->rt_key_src
= ort
->rt_key_src
;
2903 rt
->rt_key_tos
= ort
->rt_key_tos
;
2904 rt
->rt_route_iif
= ort
->rt_route_iif
;
2905 rt
->rt_iif
= ort
->rt_iif
;
2906 rt
->rt_oif
= ort
->rt_oif
;
2907 rt
->rt_mark
= ort
->rt_mark
;
2909 rt
->rt_genid
= rt_genid(net
);
2910 rt
->rt_flags
= ort
->rt_flags
;
2911 rt
->rt_type
= ort
->rt_type
;
2912 rt
->rt_dst
= ort
->rt_dst
;
2913 rt
->rt_src
= ort
->rt_src
;
2914 rt
->rt_gateway
= ort
->rt_gateway
;
2915 rt
->rt_spec_dst
= ort
->rt_spec_dst
;
2916 rt
->peer
= ort
->peer
;
2918 atomic_inc(&rt
->peer
->refcnt
);
2921 atomic_inc(&rt
->fi
->fib_clntref
);
2926 dst_release(dst_orig
);
2928 return rt
? &rt
->dst
: ERR_PTR(-ENOMEM
);
2931 struct rtable
*ip_route_output_flow(struct net
*net
, struct flowi4
*flp4
,
2934 struct rtable
*rt
= __ip_route_output_key(net
, flp4
);
2939 if (flp4
->flowi4_proto
)
2940 rt
= (struct rtable
*) xfrm_lookup(net
, &rt
->dst
,
2941 flowi4_to_flowi(flp4
),
2946 EXPORT_SYMBOL_GPL(ip_route_output_flow
);
2948 static int rt_fill_info(struct net
*net
,
2949 struct sk_buff
*skb
, u32 pid
, u32 seq
, int event
,
2950 int nowait
, unsigned int flags
)
2952 struct rtable
*rt
= skb_rtable(skb
);
2954 struct nlmsghdr
*nlh
;
2955 unsigned long expires
= 0;
2956 const struct inet_peer
*peer
= rt
->peer
;
2957 u32 id
= 0, ts
= 0, tsage
= 0, error
;
2959 nlh
= nlmsg_put(skb
, pid
, seq
, event
, sizeof(*r
), flags
);
2963 r
= nlmsg_data(nlh
);
2964 r
->rtm_family
= AF_INET
;
2965 r
->rtm_dst_len
= 32;
2967 r
->rtm_tos
= rt
->rt_key_tos
;
2968 r
->rtm_table
= RT_TABLE_MAIN
;
2969 if (nla_put_u32(skb
, RTA_TABLE
, RT_TABLE_MAIN
))
2970 goto nla_put_failure
;
2971 r
->rtm_type
= rt
->rt_type
;
2972 r
->rtm_scope
= RT_SCOPE_UNIVERSE
;
2973 r
->rtm_protocol
= RTPROT_UNSPEC
;
2974 r
->rtm_flags
= (rt
->rt_flags
& ~0xFFFF) | RTM_F_CLONED
;
2975 if (rt
->rt_flags
& RTCF_NOTIFY
)
2976 r
->rtm_flags
|= RTM_F_NOTIFY
;
2978 if (nla_put_be32(skb
, RTA_DST
, rt
->rt_dst
))
2979 goto nla_put_failure
;
2980 if (rt
->rt_key_src
) {
2981 r
->rtm_src_len
= 32;
2982 if (nla_put_be32(skb
, RTA_SRC
, rt
->rt_key_src
))
2983 goto nla_put_failure
;
2986 nla_put_u32(skb
, RTA_OIF
, rt
->dst
.dev
->ifindex
))
2987 goto nla_put_failure
;
2988 #ifdef CONFIG_IP_ROUTE_CLASSID
2989 if (rt
->dst
.tclassid
&&
2990 nla_put_u32(skb
, RTA_FLOW
, rt
->dst
.tclassid
))
2991 goto nla_put_failure
;
2993 if (rt_is_input_route(rt
)) {
2994 if (nla_put_be32(skb
, RTA_PREFSRC
, rt
->rt_spec_dst
))
2995 goto nla_put_failure
;
2996 } else if (rt
->rt_src
!= rt
->rt_key_src
) {
2997 if (nla_put_be32(skb
, RTA_PREFSRC
, rt
->rt_src
))
2998 goto nla_put_failure
;
3000 if (rt
->rt_dst
!= rt
->rt_gateway
&&
3001 nla_put_be32(skb
, RTA_GATEWAY
, rt
->rt_gateway
))
3002 goto nla_put_failure
;
3004 if (rtnetlink_put_metrics(skb
, dst_metrics_ptr(&rt
->dst
)) < 0)
3005 goto nla_put_failure
;
3008 nla_put_be32(skb
, RTA_MARK
, rt
->rt_mark
))
3009 goto nla_put_failure
;
3011 error
= rt
->dst
.error
;
3013 inet_peer_refcheck(rt
->peer
);
3014 id
= atomic_read(&peer
->ip_id_count
) & 0xffff;
3015 if (peer
->tcp_ts_stamp
) {
3017 tsage
= get_seconds() - peer
->tcp_ts_stamp
;
3019 expires
= ACCESS_ONCE(peer
->pmtu_expires
);
3021 if (time_before(jiffies
, 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 if (nla_put_u32(skb
, RTA_IIF
, rt
->rt_iif
))
3051 goto nla_put_failure
;
3054 if (rtnl_put_cacheinfo(skb
, &rt
->dst
, id
, ts
, tsage
,
3055 expires
, error
) < 0)
3056 goto nla_put_failure
;
3058 return nlmsg_end(skb
, nlh
);
3061 nlmsg_cancel(skb
, nlh
);
3065 static int inet_rtm_getroute(struct sk_buff
*in_skb
, struct nlmsghdr
*nlh
, void *arg
)
3067 struct net
*net
= sock_net(in_skb
->sk
);
3069 struct nlattr
*tb
[RTA_MAX
+1];
3070 struct rtable
*rt
= NULL
;
3076 struct sk_buff
*skb
;
3078 err
= nlmsg_parse(nlh
, sizeof(*rtm
), tb
, RTA_MAX
, rtm_ipv4_policy
);
3082 rtm
= nlmsg_data(nlh
);
3084 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
3090 /* Reserve room for dummy headers, this skb can pass
3091 through good chunk of routing engine.
3093 skb_reset_mac_header(skb
);
3094 skb_reset_network_header(skb
);
3096 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
3097 ip_hdr(skb
)->protocol
= IPPROTO_ICMP
;
3098 skb_reserve(skb
, MAX_HEADER
+ sizeof(struct iphdr
));
3100 src
= tb
[RTA_SRC
] ? nla_get_be32(tb
[RTA_SRC
]) : 0;
3101 dst
= tb
[RTA_DST
] ? nla_get_be32(tb
[RTA_DST
]) : 0;
3102 iif
= tb
[RTA_IIF
] ? nla_get_u32(tb
[RTA_IIF
]) : 0;
3103 mark
= tb
[RTA_MARK
] ? nla_get_u32(tb
[RTA_MARK
]) : 0;
3106 struct net_device
*dev
;
3108 dev
= __dev_get_by_index(net
, iif
);
3114 skb
->protocol
= htons(ETH_P_IP
);
3118 err
= ip_route_input(skb
, dst
, src
, rtm
->rtm_tos
, dev
);
3121 rt
= skb_rtable(skb
);
3122 if (err
== 0 && rt
->dst
.error
)
3123 err
= -rt
->dst
.error
;
3125 struct flowi4 fl4
= {
3128 .flowi4_tos
= rtm
->rtm_tos
,
3129 .flowi4_oif
= tb
[RTA_OIF
] ? nla_get_u32(tb
[RTA_OIF
]) : 0,
3130 .flowi4_mark
= mark
,
3132 rt
= ip_route_output_key(net
, &fl4
);
3142 skb_dst_set(skb
, &rt
->dst
);
3143 if (rtm
->rtm_flags
& RTM_F_NOTIFY
)
3144 rt
->rt_flags
|= RTCF_NOTIFY
;
3146 err
= rt_fill_info(net
, skb
, NETLINK_CB(in_skb
).pid
, nlh
->nlmsg_seq
,
3147 RTM_NEWROUTE
, 0, 0);
3151 err
= rtnl_unicast(skb
, net
, NETLINK_CB(in_skb
).pid
);
3160 int ip_rt_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
3167 net
= sock_net(skb
->sk
);
3172 s_idx
= idx
= cb
->args
[1];
3173 for (h
= s_h
; h
<= rt_hash_mask
; h
++, s_idx
= 0) {
3174 if (!rt_hash_table
[h
].chain
)
3177 for (rt
= rcu_dereference_bh(rt_hash_table
[h
].chain
), idx
= 0; rt
;
3178 rt
= rcu_dereference_bh(rt
->dst
.rt_next
), idx
++) {
3179 if (!net_eq(dev_net(rt
->dst
.dev
), net
) || idx
< s_idx
)
3181 if (rt_is_expired(rt
))
3183 skb_dst_set_noref(skb
, &rt
->dst
);
3184 if (rt_fill_info(net
, skb
, NETLINK_CB(cb
->skb
).pid
,
3185 cb
->nlh
->nlmsg_seq
, RTM_NEWROUTE
,
3186 1, NLM_F_MULTI
) <= 0) {
3188 rcu_read_unlock_bh();
3193 rcu_read_unlock_bh();
3202 void ip_rt_multicast_event(struct in_device
*in_dev
)
3204 rt_cache_flush(dev_net(in_dev
->dev
), 0);
3207 #ifdef CONFIG_SYSCTL
3208 static int ipv4_sysctl_rtcache_flush(ctl_table
*__ctl
, int write
,
3209 void __user
*buffer
,
3210 size_t *lenp
, loff_t
*ppos
)
3217 memcpy(&ctl
, __ctl
, sizeof(ctl
));
3218 ctl
.data
= &flush_delay
;
3219 proc_dointvec(&ctl
, write
, buffer
, lenp
, ppos
);
3221 net
= (struct net
*)__ctl
->extra1
;
3222 rt_cache_flush(net
, flush_delay
);
3229 static ctl_table ipv4_route_table
[] = {
3231 .procname
= "gc_thresh",
3232 .data
= &ipv4_dst_ops
.gc_thresh
,
3233 .maxlen
= sizeof(int),
3235 .proc_handler
= proc_dointvec
,
3238 .procname
= "max_size",
3239 .data
= &ip_rt_max_size
,
3240 .maxlen
= sizeof(int),
3242 .proc_handler
= proc_dointvec
,
3245 /* Deprecated. Use gc_min_interval_ms */
3247 .procname
= "gc_min_interval",
3248 .data
= &ip_rt_gc_min_interval
,
3249 .maxlen
= sizeof(int),
3251 .proc_handler
= proc_dointvec_jiffies
,
3254 .procname
= "gc_min_interval_ms",
3255 .data
= &ip_rt_gc_min_interval
,
3256 .maxlen
= sizeof(int),
3258 .proc_handler
= proc_dointvec_ms_jiffies
,
3261 .procname
= "gc_timeout",
3262 .data
= &ip_rt_gc_timeout
,
3263 .maxlen
= sizeof(int),
3265 .proc_handler
= proc_dointvec_jiffies
,
3268 .procname
= "gc_interval",
3269 .data
= &ip_rt_gc_interval
,
3270 .maxlen
= sizeof(int),
3272 .proc_handler
= proc_dointvec_jiffies
,
3275 .procname
= "redirect_load",
3276 .data
= &ip_rt_redirect_load
,
3277 .maxlen
= sizeof(int),
3279 .proc_handler
= proc_dointvec
,
3282 .procname
= "redirect_number",
3283 .data
= &ip_rt_redirect_number
,
3284 .maxlen
= sizeof(int),
3286 .proc_handler
= proc_dointvec
,
3289 .procname
= "redirect_silence",
3290 .data
= &ip_rt_redirect_silence
,
3291 .maxlen
= sizeof(int),
3293 .proc_handler
= proc_dointvec
,
3296 .procname
= "error_cost",
3297 .data
= &ip_rt_error_cost
,
3298 .maxlen
= sizeof(int),
3300 .proc_handler
= proc_dointvec
,
3303 .procname
= "error_burst",
3304 .data
= &ip_rt_error_burst
,
3305 .maxlen
= sizeof(int),
3307 .proc_handler
= proc_dointvec
,
3310 .procname
= "gc_elasticity",
3311 .data
= &ip_rt_gc_elasticity
,
3312 .maxlen
= sizeof(int),
3314 .proc_handler
= proc_dointvec
,
3317 .procname
= "mtu_expires",
3318 .data
= &ip_rt_mtu_expires
,
3319 .maxlen
= sizeof(int),
3321 .proc_handler
= proc_dointvec_jiffies
,
3324 .procname
= "min_pmtu",
3325 .data
= &ip_rt_min_pmtu
,
3326 .maxlen
= sizeof(int),
3328 .proc_handler
= proc_dointvec
,
3331 .procname
= "min_adv_mss",
3332 .data
= &ip_rt_min_advmss
,
3333 .maxlen
= sizeof(int),
3335 .proc_handler
= proc_dointvec
,
3340 static struct ctl_table ipv4_route_flush_table
[] = {
3342 .procname
= "flush",
3343 .maxlen
= sizeof(int),
3345 .proc_handler
= ipv4_sysctl_rtcache_flush
,
3350 static __net_init
int sysctl_route_net_init(struct net
*net
)
3352 struct ctl_table
*tbl
;
3354 tbl
= ipv4_route_flush_table
;
3355 if (!net_eq(net
, &init_net
)) {
3356 tbl
= kmemdup(tbl
, sizeof(ipv4_route_flush_table
), GFP_KERNEL
);
3360 tbl
[0].extra1
= net
;
3362 net
->ipv4
.route_hdr
= register_net_sysctl(net
, "net/ipv4/route", tbl
);
3363 if (net
->ipv4
.route_hdr
== NULL
)
3368 if (tbl
!= ipv4_route_flush_table
)
3374 static __net_exit
void sysctl_route_net_exit(struct net
*net
)
3376 struct ctl_table
*tbl
;
3378 tbl
= net
->ipv4
.route_hdr
->ctl_table_arg
;
3379 unregister_net_sysctl_table(net
->ipv4
.route_hdr
);
3380 BUG_ON(tbl
== ipv4_route_flush_table
);
3384 static __net_initdata
struct pernet_operations sysctl_route_ops
= {
3385 .init
= sysctl_route_net_init
,
3386 .exit
= sysctl_route_net_exit
,
3390 static __net_init
int rt_genid_init(struct net
*net
)
3392 get_random_bytes(&net
->ipv4
.rt_genid
,
3393 sizeof(net
->ipv4
.rt_genid
));
3394 get_random_bytes(&net
->ipv4
.dev_addr_genid
,
3395 sizeof(net
->ipv4
.dev_addr_genid
));
3399 static __net_initdata
struct pernet_operations rt_genid_ops
= {
3400 .init
= rt_genid_init
,
3404 #ifdef CONFIG_IP_ROUTE_CLASSID
3405 struct ip_rt_acct __percpu
*ip_rt_acct __read_mostly
;
3406 #endif /* CONFIG_IP_ROUTE_CLASSID */
3408 static __initdata
unsigned long rhash_entries
;
3409 static int __init
set_rhash_entries(char *str
)
3416 ret
= kstrtoul(str
, 0, &rhash_entries
);
3422 __setup("rhash_entries=", set_rhash_entries
);
3424 int __init
ip_rt_init(void)
3428 #ifdef CONFIG_IP_ROUTE_CLASSID
3429 ip_rt_acct
= __alloc_percpu(256 * sizeof(struct ip_rt_acct
), __alignof__(struct ip_rt_acct
));
3431 panic("IP: failed to allocate ip_rt_acct\n");
3434 ipv4_dst_ops
.kmem_cachep
=
3435 kmem_cache_create("ip_dst_cache", sizeof(struct rtable
), 0,
3436 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3438 ipv4_dst_blackhole_ops
.kmem_cachep
= ipv4_dst_ops
.kmem_cachep
;
3440 if (dst_entries_init(&ipv4_dst_ops
) < 0)
3441 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3443 if (dst_entries_init(&ipv4_dst_blackhole_ops
) < 0)
3444 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3446 rt_hash_table
= (struct rt_hash_bucket
*)
3447 alloc_large_system_hash("IP route cache",
3448 sizeof(struct rt_hash_bucket
),
3450 (totalram_pages
>= 128 * 1024) ?
3456 rhash_entries
? 0 : 512 * 1024);
3457 memset(rt_hash_table
, 0, (rt_hash_mask
+ 1) * sizeof(struct rt_hash_bucket
));
3458 rt_hash_lock_init();
3460 ipv4_dst_ops
.gc_thresh
= (rt_hash_mask
+ 1);
3461 ip_rt_max_size
= (rt_hash_mask
+ 1) * 16;
3466 INIT_DELAYED_WORK_DEFERRABLE(&expires_work
, rt_worker_func
);
3467 expires_ljiffies
= jiffies
;
3468 schedule_delayed_work(&expires_work
,
3469 net_random() % ip_rt_gc_interval
+ ip_rt_gc_interval
);
3471 if (ip_rt_proc_init())
3472 pr_err("Unable to create route proc files\n");
3475 xfrm4_init(ip_rt_max_size
);
3477 rtnl_register(PF_INET
, RTM_GETROUTE
, inet_rtm_getroute
, NULL
, NULL
);
3479 #ifdef CONFIG_SYSCTL
3480 register_pernet_subsys(&sysctl_route_ops
);
3482 register_pernet_subsys(&rt_genid_ops
);
3486 #ifdef CONFIG_SYSCTL
3488 * We really need to sanitize the damn ipv4 init order, then all
3489 * this nonsense will go away.
3491 void __init
ip_static_sysctl_init(void)
3493 register_net_sysctl(&init_net
, "net/ipv4/route", ipv4_route_table
);