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 <linux/uaccess.h>
69 #include <linux/bitops.h>
70 #include <linux/types.h>
71 #include <linux/kernel.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/skbuff.h>
83 #include <linux/inetdevice.h>
84 #include <linux/igmp.h>
85 #include <linux/pkt_sched.h>
86 #include <linux/mroute.h>
87 #include <linux/netfilter_ipv4.h>
88 #include <linux/random.h>
89 #include <linux/rcupdate.h>
90 #include <linux/times.h>
91 #include <linux/slab.h>
92 #include <linux/jhash.h>
94 #include <net/dst_metadata.h>
95 #include <net/net_namespace.h>
96 #include <net/protocol.h>
98 #include <net/route.h>
99 #include <net/inetpeer.h>
100 #include <net/sock.h>
101 #include <net/ip_fib.h>
104 #include <net/icmp.h>
105 #include <net/xfrm.h>
106 #include <net/lwtunnel.h>
107 #include <net/netevent.h>
108 #include <net/rtnetlink.h>
110 #include <linux/sysctl.h>
111 #include <linux/kmemleak.h>
113 #include <net/secure_seq.h>
114 #include <net/ip_tunnels.h>
115 #include <net/l3mdev.h>
117 #define RT_FL_TOS(oldflp4) \
118 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
120 #define RT_GC_TIMEOUT (300*HZ)
122 static int ip_rt_max_size
;
123 static int ip_rt_redirect_number __read_mostly
= 9;
124 static int ip_rt_redirect_load __read_mostly
= HZ
/ 50;
125 static int ip_rt_redirect_silence __read_mostly
= ((HZ
/ 50) << (9 + 1));
126 static int ip_rt_error_cost __read_mostly
= HZ
;
127 static int ip_rt_error_burst __read_mostly
= 5 * HZ
;
128 static int ip_rt_mtu_expires __read_mostly
= 10 * 60 * HZ
;
129 static int ip_rt_min_pmtu __read_mostly
= 512 + 20 + 20;
130 static int ip_rt_min_advmss __read_mostly
= 256;
132 static int ip_rt_gc_timeout __read_mostly
= RT_GC_TIMEOUT
;
134 * Interface to generic destination cache.
137 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
);
138 static unsigned int ipv4_default_advmss(const struct dst_entry
*dst
);
139 static unsigned int ipv4_mtu(const struct dst_entry
*dst
);
140 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
);
141 static void ipv4_link_failure(struct sk_buff
*skb
);
142 static void ip_rt_update_pmtu(struct dst_entry
*dst
, struct sock
*sk
,
143 struct sk_buff
*skb
, u32 mtu
);
144 static void ip_do_redirect(struct dst_entry
*dst
, struct sock
*sk
,
145 struct sk_buff
*skb
);
146 static void ipv4_dst_destroy(struct dst_entry
*dst
);
148 static u32
*ipv4_cow_metrics(struct dst_entry
*dst
, unsigned long old
)
154 static struct neighbour
*ipv4_neigh_lookup(const struct dst_entry
*dst
,
158 static struct dst_ops ipv4_dst_ops
= {
160 .check
= ipv4_dst_check
,
161 .default_advmss
= ipv4_default_advmss
,
163 .cow_metrics
= ipv4_cow_metrics
,
164 .destroy
= ipv4_dst_destroy
,
165 .negative_advice
= ipv4_negative_advice
,
166 .link_failure
= ipv4_link_failure
,
167 .update_pmtu
= ip_rt_update_pmtu
,
168 .redirect
= ip_do_redirect
,
169 .local_out
= __ip_local_out
,
170 .neigh_lookup
= ipv4_neigh_lookup
,
173 #define ECN_OR_COST(class) TC_PRIO_##class
175 const __u8 ip_tos2prio
[16] = {
177 ECN_OR_COST(BESTEFFORT
),
179 ECN_OR_COST(BESTEFFORT
),
185 ECN_OR_COST(INTERACTIVE
),
187 ECN_OR_COST(INTERACTIVE
),
188 TC_PRIO_INTERACTIVE_BULK
,
189 ECN_OR_COST(INTERACTIVE_BULK
),
190 TC_PRIO_INTERACTIVE_BULK
,
191 ECN_OR_COST(INTERACTIVE_BULK
)
193 EXPORT_SYMBOL(ip_tos2prio
);
195 static DEFINE_PER_CPU(struct rt_cache_stat
, rt_cache_stat
);
196 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
198 #ifdef CONFIG_PROC_FS
199 static void *rt_cache_seq_start(struct seq_file
*seq
, loff_t
*pos
)
203 return SEQ_START_TOKEN
;
206 static void *rt_cache_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
212 static void rt_cache_seq_stop(struct seq_file
*seq
, void *v
)
216 static int rt_cache_seq_show(struct seq_file
*seq
, void *v
)
218 if (v
== SEQ_START_TOKEN
)
219 seq_printf(seq
, "%-127s\n",
220 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
221 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
226 static const struct seq_operations rt_cache_seq_ops
= {
227 .start
= rt_cache_seq_start
,
228 .next
= rt_cache_seq_next
,
229 .stop
= rt_cache_seq_stop
,
230 .show
= rt_cache_seq_show
,
233 static int rt_cache_seq_open(struct inode
*inode
, struct file
*file
)
235 return seq_open(file
, &rt_cache_seq_ops
);
238 static const struct file_operations rt_cache_seq_fops
= {
239 .owner
= THIS_MODULE
,
240 .open
= rt_cache_seq_open
,
243 .release
= seq_release
,
247 static void *rt_cpu_seq_start(struct seq_file
*seq
, loff_t
*pos
)
252 return SEQ_START_TOKEN
;
254 for (cpu
= *pos
-1; cpu
< nr_cpu_ids
; ++cpu
) {
255 if (!cpu_possible(cpu
))
258 return &per_cpu(rt_cache_stat
, cpu
);
263 static void *rt_cpu_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
267 for (cpu
= *pos
; cpu
< nr_cpu_ids
; ++cpu
) {
268 if (!cpu_possible(cpu
))
271 return &per_cpu(rt_cache_stat
, cpu
);
277 static void rt_cpu_seq_stop(struct seq_file
*seq
, void *v
)
282 static int rt_cpu_seq_show(struct seq_file
*seq
, void *v
)
284 struct rt_cache_stat
*st
= v
;
286 if (v
== SEQ_START_TOKEN
) {
287 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");
291 seq_printf(seq
,"%08x %08x %08x %08x %08x %08x %08x %08x "
292 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
293 dst_entries_get_slow(&ipv4_dst_ops
),
306 0, /* st->gc_total */
307 0, /* st->gc_ignored */
308 0, /* st->gc_goal_miss */
309 0, /* st->gc_dst_overflow */
310 0, /* st->in_hlist_search */
311 0 /* st->out_hlist_search */
316 static const struct seq_operations rt_cpu_seq_ops
= {
317 .start
= rt_cpu_seq_start
,
318 .next
= rt_cpu_seq_next
,
319 .stop
= rt_cpu_seq_stop
,
320 .show
= rt_cpu_seq_show
,
324 static int rt_cpu_seq_open(struct inode
*inode
, struct file
*file
)
326 return seq_open(file
, &rt_cpu_seq_ops
);
329 static const struct file_operations rt_cpu_seq_fops
= {
330 .owner
= THIS_MODULE
,
331 .open
= rt_cpu_seq_open
,
334 .release
= seq_release
,
337 #ifdef CONFIG_IP_ROUTE_CLASSID
338 static int rt_acct_proc_show(struct seq_file
*m
, void *v
)
340 struct ip_rt_acct
*dst
, *src
;
343 dst
= kcalloc(256, sizeof(struct ip_rt_acct
), GFP_KERNEL
);
347 for_each_possible_cpu(i
) {
348 src
= (struct ip_rt_acct
*)per_cpu_ptr(ip_rt_acct
, i
);
349 for (j
= 0; j
< 256; j
++) {
350 dst
[j
].o_bytes
+= src
[j
].o_bytes
;
351 dst
[j
].o_packets
+= src
[j
].o_packets
;
352 dst
[j
].i_bytes
+= src
[j
].i_bytes
;
353 dst
[j
].i_packets
+= src
[j
].i_packets
;
357 seq_write(m
, dst
, 256 * sizeof(struct ip_rt_acct
));
362 static int rt_acct_proc_open(struct inode
*inode
, struct file
*file
)
364 return single_open(file
, rt_acct_proc_show
, NULL
);
367 static const struct file_operations rt_acct_proc_fops
= {
368 .owner
= THIS_MODULE
,
369 .open
= rt_acct_proc_open
,
372 .release
= single_release
,
376 static int __net_init
ip_rt_do_proc_init(struct net
*net
)
378 struct proc_dir_entry
*pde
;
380 pde
= proc_create("rt_cache", S_IRUGO
, net
->proc_net
,
385 pde
= proc_create("rt_cache", S_IRUGO
,
386 net
->proc_net_stat
, &rt_cpu_seq_fops
);
390 #ifdef CONFIG_IP_ROUTE_CLASSID
391 pde
= proc_create("rt_acct", 0, net
->proc_net
, &rt_acct_proc_fops
);
397 #ifdef CONFIG_IP_ROUTE_CLASSID
399 remove_proc_entry("rt_cache", net
->proc_net_stat
);
402 remove_proc_entry("rt_cache", net
->proc_net
);
407 static void __net_exit
ip_rt_do_proc_exit(struct net
*net
)
409 remove_proc_entry("rt_cache", net
->proc_net_stat
);
410 remove_proc_entry("rt_cache", net
->proc_net
);
411 #ifdef CONFIG_IP_ROUTE_CLASSID
412 remove_proc_entry("rt_acct", net
->proc_net
);
416 static struct pernet_operations ip_rt_proc_ops __net_initdata
= {
417 .init
= ip_rt_do_proc_init
,
418 .exit
= ip_rt_do_proc_exit
,
421 static int __init
ip_rt_proc_init(void)
423 return register_pernet_subsys(&ip_rt_proc_ops
);
427 static inline int ip_rt_proc_init(void)
431 #endif /* CONFIG_PROC_FS */
433 static inline bool rt_is_expired(const struct rtable
*rth
)
435 return rth
->rt_genid
!= rt_genid_ipv4(dev_net(rth
->dst
.dev
));
438 void rt_cache_flush(struct net
*net
)
440 rt_genid_bump_ipv4(net
);
443 static struct neighbour
*ipv4_neigh_lookup(const struct dst_entry
*dst
,
447 struct net_device
*dev
= dst
->dev
;
448 const __be32
*pkey
= daddr
;
449 const struct rtable
*rt
;
452 rt
= (const struct rtable
*) dst
;
454 pkey
= (const __be32
*) &rt
->rt_gateway
;
456 pkey
= &ip_hdr(skb
)->daddr
;
458 n
= __ipv4_neigh_lookup(dev
, *(__force u32
*)pkey
);
461 return neigh_create(&arp_tbl
, pkey
, dev
);
464 #define IP_IDENTS_SZ 2048u
466 static atomic_t
*ip_idents __read_mostly
;
467 static u32
*ip_tstamps __read_mostly
;
469 /* In order to protect privacy, we add a perturbation to identifiers
470 * if one generator is seldom used. This makes hard for an attacker
471 * to infer how many packets were sent between two points in time.
473 u32
ip_idents_reserve(u32 hash
, int segs
)
475 u32
*p_tstamp
= ip_tstamps
+ hash
% IP_IDENTS_SZ
;
476 atomic_t
*p_id
= ip_idents
+ hash
% IP_IDENTS_SZ
;
477 u32 old
= ACCESS_ONCE(*p_tstamp
);
478 u32 now
= (u32
)jiffies
;
481 if (old
!= now
&& cmpxchg(p_tstamp
, old
, now
) == old
)
482 delta
= prandom_u32_max(now
- old
);
484 /* Do not use atomic_add_return() as it makes UBSAN unhappy */
486 old
= (u32
)atomic_read(p_id
);
487 new = old
+ delta
+ segs
;
488 } while (atomic_cmpxchg(p_id
, old
, new) != old
);
492 EXPORT_SYMBOL(ip_idents_reserve
);
494 void __ip_select_ident(struct net
*net
, struct iphdr
*iph
, int segs
)
496 static u32 ip_idents_hashrnd __read_mostly
;
499 net_get_random_once(&ip_idents_hashrnd
, sizeof(ip_idents_hashrnd
));
501 hash
= jhash_3words((__force u32
)iph
->daddr
,
502 (__force u32
)iph
->saddr
,
503 iph
->protocol
^ net_hash_mix(net
),
505 id
= ip_idents_reserve(hash
, segs
);
508 EXPORT_SYMBOL(__ip_select_ident
);
510 static void __build_flow_key(const struct net
*net
, struct flowi4
*fl4
,
511 const struct sock
*sk
,
512 const struct iphdr
*iph
,
514 u8 prot
, u32 mark
, int flow_flags
)
517 const struct inet_sock
*inet
= inet_sk(sk
);
519 oif
= sk
->sk_bound_dev_if
;
521 tos
= RT_CONN_FLAGS(sk
);
522 prot
= inet
->hdrincl
? IPPROTO_RAW
: sk
->sk_protocol
;
524 flowi4_init_output(fl4
, oif
, mark
, tos
,
525 RT_SCOPE_UNIVERSE
, prot
,
527 iph
->daddr
, iph
->saddr
, 0, 0,
528 sock_net_uid(net
, sk
));
531 static void build_skb_flow_key(struct flowi4
*fl4
, const struct sk_buff
*skb
,
532 const struct sock
*sk
)
534 const struct net
*net
= dev_net(skb
->dev
);
535 const struct iphdr
*iph
= ip_hdr(skb
);
536 int oif
= skb
->dev
->ifindex
;
537 u8 tos
= RT_TOS(iph
->tos
);
538 u8 prot
= iph
->protocol
;
539 u32 mark
= skb
->mark
;
541 __build_flow_key(net
, fl4
, sk
, iph
, oif
, tos
, prot
, mark
, 0);
544 static void build_sk_flow_key(struct flowi4
*fl4
, const struct sock
*sk
)
546 const struct inet_sock
*inet
= inet_sk(sk
);
547 const struct ip_options_rcu
*inet_opt
;
548 __be32 daddr
= inet
->inet_daddr
;
551 inet_opt
= rcu_dereference(inet
->inet_opt
);
552 if (inet_opt
&& inet_opt
->opt
.srr
)
553 daddr
= inet_opt
->opt
.faddr
;
554 flowi4_init_output(fl4
, sk
->sk_bound_dev_if
, sk
->sk_mark
,
555 RT_CONN_FLAGS(sk
), RT_SCOPE_UNIVERSE
,
556 inet
->hdrincl
? IPPROTO_RAW
: sk
->sk_protocol
,
557 inet_sk_flowi_flags(sk
),
558 daddr
, inet
->inet_saddr
, 0, 0, sk
->sk_uid
);
562 static void ip_rt_build_flow_key(struct flowi4
*fl4
, const struct sock
*sk
,
563 const struct sk_buff
*skb
)
566 build_skb_flow_key(fl4
, skb
, sk
);
568 build_sk_flow_key(fl4
, sk
);
571 static inline void rt_free(struct rtable
*rt
)
573 call_rcu(&rt
->dst
.rcu_head
, dst_rcu_free
);
576 static DEFINE_SPINLOCK(fnhe_lock
);
578 static void fnhe_flush_routes(struct fib_nh_exception
*fnhe
)
582 rt
= rcu_dereference(fnhe
->fnhe_rth_input
);
584 RCU_INIT_POINTER(fnhe
->fnhe_rth_input
, NULL
);
587 rt
= rcu_dereference(fnhe
->fnhe_rth_output
);
589 RCU_INIT_POINTER(fnhe
->fnhe_rth_output
, NULL
);
594 static struct fib_nh_exception
*fnhe_oldest(struct fnhe_hash_bucket
*hash
)
596 struct fib_nh_exception
*fnhe
, *oldest
;
598 oldest
= rcu_dereference(hash
->chain
);
599 for (fnhe
= rcu_dereference(oldest
->fnhe_next
); fnhe
;
600 fnhe
= rcu_dereference(fnhe
->fnhe_next
)) {
601 if (time_before(fnhe
->fnhe_stamp
, oldest
->fnhe_stamp
))
604 fnhe_flush_routes(oldest
);
608 static inline u32
fnhe_hashfun(__be32 daddr
)
610 static u32 fnhe_hashrnd __read_mostly
;
613 net_get_random_once(&fnhe_hashrnd
, sizeof(fnhe_hashrnd
));
614 hval
= jhash_1word((__force u32
) daddr
, fnhe_hashrnd
);
615 return hash_32(hval
, FNHE_HASH_SHIFT
);
618 static void fill_route_from_fnhe(struct rtable
*rt
, struct fib_nh_exception
*fnhe
)
620 rt
->rt_pmtu
= fnhe
->fnhe_pmtu
;
621 rt
->dst
.expires
= fnhe
->fnhe_expires
;
624 rt
->rt_flags
|= RTCF_REDIRECTED
;
625 rt
->rt_gateway
= fnhe
->fnhe_gw
;
626 rt
->rt_uses_gateway
= 1;
630 static void update_or_create_fnhe(struct fib_nh
*nh
, __be32 daddr
, __be32 gw
,
631 u32 pmtu
, unsigned long expires
)
633 struct fnhe_hash_bucket
*hash
;
634 struct fib_nh_exception
*fnhe
;
638 u32 hval
= fnhe_hashfun(daddr
);
640 spin_lock_bh(&fnhe_lock
);
642 hash
= rcu_dereference(nh
->nh_exceptions
);
644 hash
= kzalloc(FNHE_HASH_SIZE
* sizeof(*hash
), GFP_ATOMIC
);
647 rcu_assign_pointer(nh
->nh_exceptions
, hash
);
653 for (fnhe
= rcu_dereference(hash
->chain
); fnhe
;
654 fnhe
= rcu_dereference(fnhe
->fnhe_next
)) {
655 if (fnhe
->fnhe_daddr
== daddr
)
664 fnhe
->fnhe_pmtu
= pmtu
;
665 fnhe
->fnhe_expires
= max(1UL, expires
);
667 /* Update all cached dsts too */
668 rt
= rcu_dereference(fnhe
->fnhe_rth_input
);
670 fill_route_from_fnhe(rt
, fnhe
);
671 rt
= rcu_dereference(fnhe
->fnhe_rth_output
);
673 fill_route_from_fnhe(rt
, fnhe
);
675 if (depth
> FNHE_RECLAIM_DEPTH
)
676 fnhe
= fnhe_oldest(hash
);
678 fnhe
= kzalloc(sizeof(*fnhe
), GFP_ATOMIC
);
682 fnhe
->fnhe_next
= hash
->chain
;
683 rcu_assign_pointer(hash
->chain
, fnhe
);
685 fnhe
->fnhe_genid
= fnhe_genid(dev_net(nh
->nh_dev
));
686 fnhe
->fnhe_daddr
= daddr
;
688 fnhe
->fnhe_pmtu
= pmtu
;
689 fnhe
->fnhe_expires
= expires
;
691 /* Exception created; mark the cached routes for the nexthop
692 * stale, so anyone caching it rechecks if this exception
695 rt
= rcu_dereference(nh
->nh_rth_input
);
697 rt
->dst
.obsolete
= DST_OBSOLETE_KILL
;
699 for_each_possible_cpu(i
) {
700 struct rtable __rcu
**prt
;
701 prt
= per_cpu_ptr(nh
->nh_pcpu_rth_output
, i
);
702 rt
= rcu_dereference(*prt
);
704 rt
->dst
.obsolete
= DST_OBSOLETE_KILL
;
708 fnhe
->fnhe_stamp
= jiffies
;
711 spin_unlock_bh(&fnhe_lock
);
714 static void __ip_do_redirect(struct rtable
*rt
, struct sk_buff
*skb
, struct flowi4
*fl4
,
717 __be32 new_gw
= icmp_hdr(skb
)->un
.gateway
;
718 __be32 old_gw
= ip_hdr(skb
)->saddr
;
719 struct net_device
*dev
= skb
->dev
;
720 struct in_device
*in_dev
;
721 struct fib_result res
;
725 switch (icmp_hdr(skb
)->code
& 7) {
727 case ICMP_REDIR_NETTOS
:
728 case ICMP_REDIR_HOST
:
729 case ICMP_REDIR_HOSTTOS
:
736 if (rt
->rt_gateway
!= old_gw
)
739 in_dev
= __in_dev_get_rcu(dev
);
744 if (new_gw
== old_gw
|| !IN_DEV_RX_REDIRECTS(in_dev
) ||
745 ipv4_is_multicast(new_gw
) || ipv4_is_lbcast(new_gw
) ||
746 ipv4_is_zeronet(new_gw
))
747 goto reject_redirect
;
749 if (!IN_DEV_SHARED_MEDIA(in_dev
)) {
750 if (!inet_addr_onlink(in_dev
, new_gw
, old_gw
))
751 goto reject_redirect
;
752 if (IN_DEV_SEC_REDIRECTS(in_dev
) && ip_fib_check_default(new_gw
, dev
))
753 goto reject_redirect
;
755 if (inet_addr_type(net
, new_gw
) != RTN_UNICAST
)
756 goto reject_redirect
;
759 n
= __ipv4_neigh_lookup(rt
->dst
.dev
, new_gw
);
761 n
= neigh_create(&arp_tbl
, &new_gw
, rt
->dst
.dev
);
763 if (!(n
->nud_state
& NUD_VALID
)) {
764 neigh_event_send(n
, NULL
);
766 if (fib_lookup(net
, fl4
, &res
, 0) == 0) {
767 struct fib_nh
*nh
= &FIB_RES_NH(res
);
769 update_or_create_fnhe(nh
, fl4
->daddr
, new_gw
,
770 0, jiffies
+ ip_rt_gc_timeout
);
773 rt
->dst
.obsolete
= DST_OBSOLETE_KILL
;
774 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE
, n
);
781 #ifdef CONFIG_IP_ROUTE_VERBOSE
782 if (IN_DEV_LOG_MARTIANS(in_dev
)) {
783 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
784 __be32 daddr
= iph
->daddr
;
785 __be32 saddr
= iph
->saddr
;
787 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
788 " Advised path = %pI4 -> %pI4\n",
789 &old_gw
, dev
->name
, &new_gw
,
796 static void ip_do_redirect(struct dst_entry
*dst
, struct sock
*sk
, struct sk_buff
*skb
)
800 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
801 struct net
*net
= dev_net(skb
->dev
);
802 int oif
= skb
->dev
->ifindex
;
803 u8 tos
= RT_TOS(iph
->tos
);
804 u8 prot
= iph
->protocol
;
805 u32 mark
= skb
->mark
;
807 rt
= (struct rtable
*) dst
;
809 __build_flow_key(net
, &fl4
, sk
, iph
, oif
, tos
, prot
, mark
, 0);
810 __ip_do_redirect(rt
, skb
, &fl4
, true);
813 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
)
815 struct rtable
*rt
= (struct rtable
*)dst
;
816 struct dst_entry
*ret
= dst
;
819 if (dst
->obsolete
> 0) {
822 } else if ((rt
->rt_flags
& RTCF_REDIRECTED
) ||
833 * 1. The first ip_rt_redirect_number redirects are sent
834 * with exponential backoff, then we stop sending them at all,
835 * assuming that the host ignores our redirects.
836 * 2. If we did not see packets requiring redirects
837 * during ip_rt_redirect_silence, we assume that the host
838 * forgot redirected route and start to send redirects again.
840 * This algorithm is much cheaper and more intelligent than dumb load limiting
843 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
844 * and "frag. need" (breaks PMTU discovery) in icmp.c.
847 void ip_rt_send_redirect(struct sk_buff
*skb
)
849 struct rtable
*rt
= skb_rtable(skb
);
850 struct in_device
*in_dev
;
851 struct inet_peer
*peer
;
857 in_dev
= __in_dev_get_rcu(rt
->dst
.dev
);
858 if (!in_dev
|| !IN_DEV_TX_REDIRECTS(in_dev
)) {
862 log_martians
= IN_DEV_LOG_MARTIANS(in_dev
);
863 vif
= l3mdev_master_ifindex_rcu(rt
->dst
.dev
);
866 net
= dev_net(rt
->dst
.dev
);
867 peer
= inet_getpeer_v4(net
->ipv4
.peers
, ip_hdr(skb
)->saddr
, vif
, 1);
869 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
,
870 rt_nexthop(rt
, ip_hdr(skb
)->daddr
));
874 /* No redirected packets during ip_rt_redirect_silence;
875 * reset the algorithm.
877 if (time_after(jiffies
, peer
->rate_last
+ ip_rt_redirect_silence
))
878 peer
->rate_tokens
= 0;
880 /* Too many ignored redirects; do not send anything
881 * set dst.rate_last to the last seen redirected packet.
883 if (peer
->rate_tokens
>= ip_rt_redirect_number
) {
884 peer
->rate_last
= jiffies
;
888 /* Check for load limit; set rate_last to the latest sent
891 if (peer
->rate_tokens
== 0 ||
894 (ip_rt_redirect_load
<< peer
->rate_tokens
)))) {
895 __be32 gw
= rt_nexthop(rt
, ip_hdr(skb
)->daddr
);
897 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
, gw
);
898 peer
->rate_last
= jiffies
;
900 #ifdef CONFIG_IP_ROUTE_VERBOSE
902 peer
->rate_tokens
== ip_rt_redirect_number
)
903 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
904 &ip_hdr(skb
)->saddr
, inet_iif(skb
),
905 &ip_hdr(skb
)->daddr
, &gw
);
912 static int ip_error(struct sk_buff
*skb
)
914 struct in_device
*in_dev
= __in_dev_get_rcu(skb
->dev
);
915 struct rtable
*rt
= skb_rtable(skb
);
916 struct inet_peer
*peer
;
922 /* IP on this device is disabled. */
926 net
= dev_net(rt
->dst
.dev
);
927 if (!IN_DEV_FORWARD(in_dev
)) {
928 switch (rt
->dst
.error
) {
930 __IP_INC_STATS(net
, IPSTATS_MIB_INADDRERRORS
);
934 __IP_INC_STATS(net
, IPSTATS_MIB_INNOROUTES
);
940 switch (rt
->dst
.error
) {
945 code
= ICMP_HOST_UNREACH
;
948 code
= ICMP_NET_UNREACH
;
949 __IP_INC_STATS(net
, IPSTATS_MIB_INNOROUTES
);
952 code
= ICMP_PKT_FILTERED
;
956 peer
= inet_getpeer_v4(net
->ipv4
.peers
, ip_hdr(skb
)->saddr
,
957 l3mdev_master_ifindex(skb
->dev
), 1);
962 peer
->rate_tokens
+= now
- peer
->rate_last
;
963 if (peer
->rate_tokens
> ip_rt_error_burst
)
964 peer
->rate_tokens
= ip_rt_error_burst
;
965 peer
->rate_last
= now
;
966 if (peer
->rate_tokens
>= ip_rt_error_cost
)
967 peer
->rate_tokens
-= ip_rt_error_cost
;
973 icmp_send(skb
, ICMP_DEST_UNREACH
, code
, 0);
979 static void __ip_rt_update_pmtu(struct rtable
*rt
, struct flowi4
*fl4
, u32 mtu
)
981 struct dst_entry
*dst
= &rt
->dst
;
982 struct fib_result res
;
984 if (dst_metric_locked(dst
, RTAX_MTU
))
987 if (ipv4_mtu(dst
) < mtu
)
990 if (mtu
< ip_rt_min_pmtu
)
991 mtu
= ip_rt_min_pmtu
;
993 if (rt
->rt_pmtu
== mtu
&&
994 time_before(jiffies
, dst
->expires
- ip_rt_mtu_expires
/ 2))
998 if (fib_lookup(dev_net(dst
->dev
), fl4
, &res
, 0) == 0) {
999 struct fib_nh
*nh
= &FIB_RES_NH(res
);
1001 update_or_create_fnhe(nh
, fl4
->daddr
, 0, mtu
,
1002 jiffies
+ ip_rt_mtu_expires
);
1007 static void ip_rt_update_pmtu(struct dst_entry
*dst
, struct sock
*sk
,
1008 struct sk_buff
*skb
, u32 mtu
)
1010 struct rtable
*rt
= (struct rtable
*) dst
;
1013 ip_rt_build_flow_key(&fl4
, sk
, skb
);
1014 __ip_rt_update_pmtu(rt
, &fl4
, mtu
);
1017 void ipv4_update_pmtu(struct sk_buff
*skb
, struct net
*net
, u32 mtu
,
1018 int oif
, u32 mark
, u8 protocol
, int flow_flags
)
1020 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
1025 mark
= IP4_REPLY_MARK(net
, skb
->mark
);
1027 __build_flow_key(net
, &fl4
, NULL
, iph
, oif
,
1028 RT_TOS(iph
->tos
), protocol
, mark
, flow_flags
);
1029 rt
= __ip_route_output_key(net
, &fl4
);
1031 __ip_rt_update_pmtu(rt
, &fl4
, mtu
);
1035 EXPORT_SYMBOL_GPL(ipv4_update_pmtu
);
1037 static void __ipv4_sk_update_pmtu(struct sk_buff
*skb
, struct sock
*sk
, u32 mtu
)
1039 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
1043 __build_flow_key(sock_net(sk
), &fl4
, sk
, iph
, 0, 0, 0, 0, 0);
1045 if (!fl4
.flowi4_mark
)
1046 fl4
.flowi4_mark
= IP4_REPLY_MARK(sock_net(sk
), skb
->mark
);
1048 rt
= __ip_route_output_key(sock_net(sk
), &fl4
);
1050 __ip_rt_update_pmtu(rt
, &fl4
, mtu
);
1055 void ipv4_sk_update_pmtu(struct sk_buff
*skb
, struct sock
*sk
, u32 mtu
)
1057 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
1060 struct dst_entry
*odst
= NULL
;
1062 struct net
*net
= sock_net(sk
);
1066 if (!ip_sk_accept_pmtu(sk
))
1069 odst
= sk_dst_get(sk
);
1071 if (sock_owned_by_user(sk
) || !odst
) {
1072 __ipv4_sk_update_pmtu(skb
, sk
, mtu
);
1076 __build_flow_key(net
, &fl4
, sk
, iph
, 0, 0, 0, 0, 0);
1078 rt
= (struct rtable
*)odst
;
1079 if (odst
->obsolete
&& !odst
->ops
->check(odst
, 0)) {
1080 rt
= ip_route_output_flow(sock_net(sk
), &fl4
, sk
);
1087 __ip_rt_update_pmtu((struct rtable
*) rt
->dst
.path
, &fl4
, mtu
);
1089 if (!dst_check(&rt
->dst
, 0)) {
1091 dst_release(&rt
->dst
);
1093 rt
= ip_route_output_flow(sock_net(sk
), &fl4
, sk
);
1101 sk_dst_set(sk
, &rt
->dst
);
1107 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu
);
1109 void ipv4_redirect(struct sk_buff
*skb
, struct net
*net
,
1110 int oif
, u32 mark
, u8 protocol
, int flow_flags
)
1112 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
1116 __build_flow_key(net
, &fl4
, NULL
, iph
, oif
,
1117 RT_TOS(iph
->tos
), protocol
, mark
, flow_flags
);
1118 rt
= __ip_route_output_key(net
, &fl4
);
1120 __ip_do_redirect(rt
, skb
, &fl4
, false);
1124 EXPORT_SYMBOL_GPL(ipv4_redirect
);
1126 void ipv4_sk_redirect(struct sk_buff
*skb
, struct sock
*sk
)
1128 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
1131 struct net
*net
= sock_net(sk
);
1133 __build_flow_key(net
, &fl4
, sk
, iph
, 0, 0, 0, 0, 0);
1134 rt
= __ip_route_output_key(net
, &fl4
);
1136 __ip_do_redirect(rt
, skb
, &fl4
, false);
1140 EXPORT_SYMBOL_GPL(ipv4_sk_redirect
);
1142 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
)
1144 struct rtable
*rt
= (struct rtable
*) dst
;
1146 /* All IPV4 dsts are created with ->obsolete set to the value
1147 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1148 * into this function always.
1150 * When a PMTU/redirect information update invalidates a route,
1151 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1152 * DST_OBSOLETE_DEAD by dst_free().
1154 if (dst
->obsolete
!= DST_OBSOLETE_FORCE_CHK
|| rt_is_expired(rt
))
1159 static void ipv4_link_failure(struct sk_buff
*skb
)
1163 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_HOST_UNREACH
, 0);
1165 rt
= skb_rtable(skb
);
1167 dst_set_expires(&rt
->dst
, 0);
1170 static int ip_rt_bug(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
1172 pr_debug("%s: %pI4 -> %pI4, %s\n",
1173 __func__
, &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1174 skb
->dev
? skb
->dev
->name
: "?");
1181 We do not cache source address of outgoing interface,
1182 because it is used only by IP RR, TS and SRR options,
1183 so that it out of fast path.
1185 BTW remember: "addr" is allowed to be not aligned
1189 void ip_rt_get_source(u8
*addr
, struct sk_buff
*skb
, struct rtable
*rt
)
1193 if (rt_is_output_route(rt
))
1194 src
= ip_hdr(skb
)->saddr
;
1196 struct fib_result res
;
1202 memset(&fl4
, 0, sizeof(fl4
));
1203 fl4
.daddr
= iph
->daddr
;
1204 fl4
.saddr
= iph
->saddr
;
1205 fl4
.flowi4_tos
= RT_TOS(iph
->tos
);
1206 fl4
.flowi4_oif
= rt
->dst
.dev
->ifindex
;
1207 fl4
.flowi4_iif
= skb
->dev
->ifindex
;
1208 fl4
.flowi4_mark
= skb
->mark
;
1211 if (fib_lookup(dev_net(rt
->dst
.dev
), &fl4
, &res
, 0) == 0)
1212 src
= FIB_RES_PREFSRC(dev_net(rt
->dst
.dev
), res
);
1214 src
= inet_select_addr(rt
->dst
.dev
,
1215 rt_nexthop(rt
, iph
->daddr
),
1219 memcpy(addr
, &src
, 4);
1222 #ifdef CONFIG_IP_ROUTE_CLASSID
1223 static void set_class_tag(struct rtable
*rt
, u32 tag
)
1225 if (!(rt
->dst
.tclassid
& 0xFFFF))
1226 rt
->dst
.tclassid
|= tag
& 0xFFFF;
1227 if (!(rt
->dst
.tclassid
& 0xFFFF0000))
1228 rt
->dst
.tclassid
|= tag
& 0xFFFF0000;
1232 static unsigned int ipv4_default_advmss(const struct dst_entry
*dst
)
1234 unsigned int advmss
= dst_metric_raw(dst
, RTAX_ADVMSS
);
1237 advmss
= max_t(unsigned int, dst
->dev
->mtu
- 40,
1239 if (advmss
> 65535 - 40)
1240 advmss
= 65535 - 40;
1245 static unsigned int ipv4_mtu(const struct dst_entry
*dst
)
1247 const struct rtable
*rt
= (const struct rtable
*) dst
;
1248 unsigned int mtu
= rt
->rt_pmtu
;
1250 if (!mtu
|| time_after_eq(jiffies
, rt
->dst
.expires
))
1251 mtu
= dst_metric_raw(dst
, RTAX_MTU
);
1256 mtu
= dst
->dev
->mtu
;
1258 if (unlikely(dst_metric_locked(dst
, RTAX_MTU
))) {
1259 if (rt
->rt_uses_gateway
&& mtu
> 576)
1263 mtu
= min_t(unsigned int, mtu
, IP_MAX_MTU
);
1265 return mtu
- lwtunnel_headroom(dst
->lwtstate
, mtu
);
1268 static struct fib_nh_exception
*find_exception(struct fib_nh
*nh
, __be32 daddr
)
1270 struct fnhe_hash_bucket
*hash
= rcu_dereference(nh
->nh_exceptions
);
1271 struct fib_nh_exception
*fnhe
;
1277 hval
= fnhe_hashfun(daddr
);
1279 for (fnhe
= rcu_dereference(hash
[hval
].chain
); fnhe
;
1280 fnhe
= rcu_dereference(fnhe
->fnhe_next
)) {
1281 if (fnhe
->fnhe_daddr
== daddr
)
1287 static bool rt_bind_exception(struct rtable
*rt
, struct fib_nh_exception
*fnhe
,
1292 spin_lock_bh(&fnhe_lock
);
1294 if (daddr
== fnhe
->fnhe_daddr
) {
1295 struct rtable __rcu
**porig
;
1296 struct rtable
*orig
;
1297 int genid
= fnhe_genid(dev_net(rt
->dst
.dev
));
1299 if (rt_is_input_route(rt
))
1300 porig
= &fnhe
->fnhe_rth_input
;
1302 porig
= &fnhe
->fnhe_rth_output
;
1303 orig
= rcu_dereference(*porig
);
1305 if (fnhe
->fnhe_genid
!= genid
) {
1306 fnhe
->fnhe_genid
= genid
;
1308 fnhe
->fnhe_pmtu
= 0;
1309 fnhe
->fnhe_expires
= 0;
1310 fnhe_flush_routes(fnhe
);
1313 fill_route_from_fnhe(rt
, fnhe
);
1314 if (!rt
->rt_gateway
)
1315 rt
->rt_gateway
= daddr
;
1317 if (!(rt
->dst
.flags
& DST_NOCACHE
)) {
1318 rcu_assign_pointer(*porig
, rt
);
1324 fnhe
->fnhe_stamp
= jiffies
;
1326 spin_unlock_bh(&fnhe_lock
);
1331 static bool rt_cache_route(struct fib_nh
*nh
, struct rtable
*rt
)
1333 struct rtable
*orig
, *prev
, **p
;
1336 if (rt_is_input_route(rt
)) {
1337 p
= (struct rtable
**)&nh
->nh_rth_input
;
1339 p
= (struct rtable
**)raw_cpu_ptr(nh
->nh_pcpu_rth_output
);
1343 prev
= cmpxchg(p
, orig
, rt
);
1353 struct uncached_list
{
1355 struct list_head head
;
1358 static DEFINE_PER_CPU_ALIGNED(struct uncached_list
, rt_uncached_list
);
1360 static void rt_add_uncached_list(struct rtable
*rt
)
1362 struct uncached_list
*ul
= raw_cpu_ptr(&rt_uncached_list
);
1364 rt
->rt_uncached_list
= ul
;
1366 spin_lock_bh(&ul
->lock
);
1367 list_add_tail(&rt
->rt_uncached
, &ul
->head
);
1368 spin_unlock_bh(&ul
->lock
);
1371 static void ipv4_dst_destroy(struct dst_entry
*dst
)
1373 struct rtable
*rt
= (struct rtable
*) dst
;
1375 if (!list_empty(&rt
->rt_uncached
)) {
1376 struct uncached_list
*ul
= rt
->rt_uncached_list
;
1378 spin_lock_bh(&ul
->lock
);
1379 list_del(&rt
->rt_uncached
);
1380 spin_unlock_bh(&ul
->lock
);
1384 void rt_flush_dev(struct net_device
*dev
)
1386 struct net
*net
= dev_net(dev
);
1390 for_each_possible_cpu(cpu
) {
1391 struct uncached_list
*ul
= &per_cpu(rt_uncached_list
, cpu
);
1393 spin_lock_bh(&ul
->lock
);
1394 list_for_each_entry(rt
, &ul
->head
, rt_uncached
) {
1395 if (rt
->dst
.dev
!= dev
)
1397 rt
->dst
.dev
= net
->loopback_dev
;
1398 dev_hold(rt
->dst
.dev
);
1401 spin_unlock_bh(&ul
->lock
);
1405 static bool rt_cache_valid(const struct rtable
*rt
)
1408 rt
->dst
.obsolete
== DST_OBSOLETE_FORCE_CHK
&&
1412 static void rt_set_nexthop(struct rtable
*rt
, __be32 daddr
,
1413 const struct fib_result
*res
,
1414 struct fib_nh_exception
*fnhe
,
1415 struct fib_info
*fi
, u16 type
, u32 itag
)
1417 bool cached
= false;
1420 struct fib_nh
*nh
= &FIB_RES_NH(*res
);
1422 if (nh
->nh_gw
&& nh
->nh_scope
== RT_SCOPE_LINK
) {
1423 rt
->rt_gateway
= nh
->nh_gw
;
1424 rt
->rt_uses_gateway
= 1;
1426 dst_init_metrics(&rt
->dst
, fi
->fib_metrics
, true);
1427 #ifdef CONFIG_IP_ROUTE_CLASSID
1428 rt
->dst
.tclassid
= nh
->nh_tclassid
;
1430 rt
->dst
.lwtstate
= lwtstate_get(nh
->nh_lwtstate
);
1432 cached
= rt_bind_exception(rt
, fnhe
, daddr
);
1433 else if (!(rt
->dst
.flags
& DST_NOCACHE
))
1434 cached
= rt_cache_route(nh
, rt
);
1435 if (unlikely(!cached
)) {
1436 /* Routes we intend to cache in nexthop exception or
1437 * FIB nexthop have the DST_NOCACHE bit clear.
1438 * However, if we are unsuccessful at storing this
1439 * route into the cache we really need to set it.
1441 rt
->dst
.flags
|= DST_NOCACHE
;
1442 if (!rt
->rt_gateway
)
1443 rt
->rt_gateway
= daddr
;
1444 rt_add_uncached_list(rt
);
1447 rt_add_uncached_list(rt
);
1449 #ifdef CONFIG_IP_ROUTE_CLASSID
1450 #ifdef CONFIG_IP_MULTIPLE_TABLES
1451 set_class_tag(rt
, res
->tclassid
);
1453 set_class_tag(rt
, itag
);
1457 struct rtable
*rt_dst_alloc(struct net_device
*dev
,
1458 unsigned int flags
, u16 type
,
1459 bool nopolicy
, bool noxfrm
, bool will_cache
)
1463 rt
= dst_alloc(&ipv4_dst_ops
, dev
, 1, DST_OBSOLETE_FORCE_CHK
,
1464 (will_cache
? 0 : (DST_HOST
| DST_NOCACHE
)) |
1465 (nopolicy
? DST_NOPOLICY
: 0) |
1466 (noxfrm
? DST_NOXFRM
: 0));
1469 rt
->rt_genid
= rt_genid_ipv4(dev_net(dev
));
1470 rt
->rt_flags
= flags
;
1472 rt
->rt_is_input
= 0;
1476 rt
->rt_uses_gateway
= 0;
1477 rt
->rt_table_id
= 0;
1478 INIT_LIST_HEAD(&rt
->rt_uncached
);
1480 rt
->dst
.output
= ip_output
;
1481 if (flags
& RTCF_LOCAL
)
1482 rt
->dst
.input
= ip_local_deliver
;
1487 EXPORT_SYMBOL(rt_dst_alloc
);
1489 /* called in rcu_read_lock() section */
1490 static int ip_route_input_mc(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
1491 u8 tos
, struct net_device
*dev
, int our
)
1494 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
1495 unsigned int flags
= RTCF_MULTICAST
;
1499 /* Primary sanity checks. */
1504 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
1505 skb
->protocol
!= htons(ETH_P_IP
))
1508 if (ipv4_is_loopback(saddr
) && !IN_DEV_ROUTE_LOCALNET(in_dev
))
1511 if (ipv4_is_zeronet(saddr
)) {
1512 if (!ipv4_is_local_multicast(daddr
))
1515 err
= fib_validate_source(skb
, saddr
, 0, tos
, 0, dev
,
1521 flags
|= RTCF_LOCAL
;
1523 rth
= rt_dst_alloc(dev_net(dev
)->loopback_dev
, flags
, RTN_MULTICAST
,
1524 IN_DEV_CONF_GET(in_dev
, NOPOLICY
), false, false);
1528 #ifdef CONFIG_IP_ROUTE_CLASSID
1529 rth
->dst
.tclassid
= itag
;
1531 rth
->dst
.output
= ip_rt_bug
;
1532 rth
->rt_is_input
= 1;
1534 #ifdef CONFIG_IP_MROUTE
1535 if (!ipv4_is_local_multicast(daddr
) && IN_DEV_MFORWARD(in_dev
))
1536 rth
->dst
.input
= ip_mr_input
;
1538 RT_CACHE_STAT_INC(in_slow_mc
);
1540 skb_dst_set(skb
, &rth
->dst
);
1552 static void ip_handle_martian_source(struct net_device
*dev
,
1553 struct in_device
*in_dev
,
1554 struct sk_buff
*skb
,
1558 RT_CACHE_STAT_INC(in_martian_src
);
1559 #ifdef CONFIG_IP_ROUTE_VERBOSE
1560 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit()) {
1562 * RFC1812 recommendation, if source is martian,
1563 * the only hint is MAC header.
1565 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1566 &daddr
, &saddr
, dev
->name
);
1567 if (dev
->hard_header_len
&& skb_mac_header_was_set(skb
)) {
1568 print_hex_dump(KERN_WARNING
, "ll header: ",
1569 DUMP_PREFIX_OFFSET
, 16, 1,
1570 skb_mac_header(skb
),
1571 dev
->hard_header_len
, true);
1577 static void ip_del_fnhe(struct fib_nh
*nh
, __be32 daddr
)
1579 struct fnhe_hash_bucket
*hash
;
1580 struct fib_nh_exception
*fnhe
, __rcu
**fnhe_p
;
1581 u32 hval
= fnhe_hashfun(daddr
);
1583 spin_lock_bh(&fnhe_lock
);
1585 hash
= rcu_dereference_protected(nh
->nh_exceptions
,
1586 lockdep_is_held(&fnhe_lock
));
1589 fnhe_p
= &hash
->chain
;
1590 fnhe
= rcu_dereference_protected(*fnhe_p
, lockdep_is_held(&fnhe_lock
));
1592 if (fnhe
->fnhe_daddr
== daddr
) {
1593 rcu_assign_pointer(*fnhe_p
, rcu_dereference_protected(
1594 fnhe
->fnhe_next
, lockdep_is_held(&fnhe_lock
)));
1595 fnhe_flush_routes(fnhe
);
1596 kfree_rcu(fnhe
, rcu
);
1599 fnhe_p
= &fnhe
->fnhe_next
;
1600 fnhe
= rcu_dereference_protected(fnhe
->fnhe_next
,
1601 lockdep_is_held(&fnhe_lock
));
1604 spin_unlock_bh(&fnhe_lock
);
1607 static void set_lwt_redirect(struct rtable
*rth
)
1609 if (lwtunnel_output_redirect(rth
->dst
.lwtstate
)) {
1610 rth
->dst
.lwtstate
->orig_output
= rth
->dst
.output
;
1611 rth
->dst
.output
= lwtunnel_output
;
1614 if (lwtunnel_input_redirect(rth
->dst
.lwtstate
)) {
1615 rth
->dst
.lwtstate
->orig_input
= rth
->dst
.input
;
1616 rth
->dst
.input
= lwtunnel_input
;
1620 /* called in rcu_read_lock() section */
1621 static int __mkroute_input(struct sk_buff
*skb
,
1622 const struct fib_result
*res
,
1623 struct in_device
*in_dev
,
1624 __be32 daddr
, __be32 saddr
, u32 tos
)
1626 struct fib_nh_exception
*fnhe
;
1629 struct in_device
*out_dev
;
1633 /* get a working reference to the output device */
1634 out_dev
= __in_dev_get_rcu(FIB_RES_DEV(*res
));
1636 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1640 err
= fib_validate_source(skb
, saddr
, daddr
, tos
, FIB_RES_OIF(*res
),
1641 in_dev
->dev
, in_dev
, &itag
);
1643 ip_handle_martian_source(in_dev
->dev
, in_dev
, skb
, daddr
,
1649 do_cache
= res
->fi
&& !itag
;
1650 if (out_dev
== in_dev
&& err
&& IN_DEV_TX_REDIRECTS(out_dev
) &&
1651 skb
->protocol
== htons(ETH_P_IP
) &&
1652 (IN_DEV_SHARED_MEDIA(out_dev
) ||
1653 inet_addr_onlink(out_dev
, saddr
, FIB_RES_GW(*res
))))
1654 IPCB(skb
)->flags
|= IPSKB_DOREDIRECT
;
1656 if (skb
->protocol
!= htons(ETH_P_IP
)) {
1657 /* Not IP (i.e. ARP). Do not create route, if it is
1658 * invalid for proxy arp. DNAT routes are always valid.
1660 * Proxy arp feature have been extended to allow, ARP
1661 * replies back to the same interface, to support
1662 * Private VLAN switch technologies. See arp.c.
1664 if (out_dev
== in_dev
&&
1665 IN_DEV_PROXY_ARP_PVLAN(in_dev
) == 0) {
1671 fnhe
= find_exception(&FIB_RES_NH(*res
), daddr
);
1674 rth
= rcu_dereference(fnhe
->fnhe_rth_input
);
1675 if (rth
&& rth
->dst
.expires
&&
1676 time_after(jiffies
, rth
->dst
.expires
)) {
1677 ip_del_fnhe(&FIB_RES_NH(*res
), daddr
);
1684 rth
= rcu_dereference(FIB_RES_NH(*res
).nh_rth_input
);
1687 if (rt_cache_valid(rth
)) {
1688 skb_dst_set_noref(skb
, &rth
->dst
);
1693 rth
= rt_dst_alloc(out_dev
->dev
, 0, res
->type
,
1694 IN_DEV_CONF_GET(in_dev
, NOPOLICY
),
1695 IN_DEV_CONF_GET(out_dev
, NOXFRM
), do_cache
);
1701 rth
->rt_is_input
= 1;
1703 rth
->rt_table_id
= res
->table
->tb_id
;
1704 RT_CACHE_STAT_INC(in_slow_tot
);
1706 rth
->dst
.input
= ip_forward
;
1708 rt_set_nexthop(rth
, daddr
, res
, fnhe
, res
->fi
, res
->type
, itag
);
1709 set_lwt_redirect(rth
);
1710 skb_dst_set(skb
, &rth
->dst
);
1717 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1719 /* To make ICMP packets follow the right flow, the multipath hash is
1720 * calculated from the inner IP addresses in reverse order.
1722 static int ip_multipath_icmp_hash(struct sk_buff
*skb
)
1724 const struct iphdr
*outer_iph
= ip_hdr(skb
);
1725 struct icmphdr _icmph
;
1726 const struct icmphdr
*icmph
;
1727 struct iphdr _inner_iph
;
1728 const struct iphdr
*inner_iph
;
1730 if (unlikely((outer_iph
->frag_off
& htons(IP_OFFSET
)) != 0))
1733 icmph
= skb_header_pointer(skb
, outer_iph
->ihl
* 4, sizeof(_icmph
),
1738 if (icmph
->type
!= ICMP_DEST_UNREACH
&&
1739 icmph
->type
!= ICMP_REDIRECT
&&
1740 icmph
->type
!= ICMP_TIME_EXCEEDED
&&
1741 icmph
->type
!= ICMP_PARAMETERPROB
) {
1745 inner_iph
= skb_header_pointer(skb
,
1746 outer_iph
->ihl
* 4 + sizeof(_icmph
),
1747 sizeof(_inner_iph
), &_inner_iph
);
1751 return fib_multipath_hash(inner_iph
->daddr
, inner_iph
->saddr
);
1754 return fib_multipath_hash(outer_iph
->saddr
, outer_iph
->daddr
);
1757 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1759 static int ip_mkroute_input(struct sk_buff
*skb
,
1760 struct fib_result
*res
,
1761 struct in_device
*in_dev
,
1762 __be32 daddr
, __be32 saddr
, u32 tos
)
1764 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1765 if (res
->fi
&& res
->fi
->fib_nhs
> 1) {
1768 if (unlikely(ip_hdr(skb
)->protocol
== IPPROTO_ICMP
))
1769 h
= ip_multipath_icmp_hash(skb
);
1771 h
= fib_multipath_hash(saddr
, daddr
);
1772 fib_select_multipath(res
, h
);
1776 /* create a routing cache entry */
1777 return __mkroute_input(skb
, res
, in_dev
, daddr
, saddr
, tos
);
1781 * NOTE. We drop all the packets that has local source
1782 * addresses, because every properly looped back packet
1783 * must have correct destination already attached by output routine.
1785 * Such approach solves two big problems:
1786 * 1. Not simplex devices are handled properly.
1787 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1788 * called with rcu_read_lock()
1791 static int ip_route_input_slow(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
1792 u8 tos
, struct net_device
*dev
)
1794 struct fib_result res
;
1795 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
1796 struct ip_tunnel_info
*tun_info
;
1798 unsigned int flags
= 0;
1802 struct net
*net
= dev_net(dev
);
1805 /* IP on this device is disabled. */
1810 /* Check for the most weird martians, which can be not detected
1814 tun_info
= skb_tunnel_info(skb
);
1815 if (tun_info
&& !(tun_info
->mode
& IP_TUNNEL_INFO_TX
))
1816 fl4
.flowi4_tun_key
.tun_id
= tun_info
->key
.tun_id
;
1818 fl4
.flowi4_tun_key
.tun_id
= 0;
1821 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
))
1822 goto martian_source
;
1826 if (ipv4_is_lbcast(daddr
) || (saddr
== 0 && daddr
== 0))
1829 /* Accept zero addresses only to limited broadcast;
1830 * I even do not know to fix it or not. Waiting for complains :-)
1832 if (ipv4_is_zeronet(saddr
))
1833 goto martian_source
;
1835 if (ipv4_is_zeronet(daddr
))
1836 goto martian_destination
;
1838 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1839 * and call it once if daddr or/and saddr are loopback addresses
1841 if (ipv4_is_loopback(daddr
)) {
1842 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev
, net
))
1843 goto martian_destination
;
1844 } else if (ipv4_is_loopback(saddr
)) {
1845 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev
, net
))
1846 goto martian_source
;
1850 * Now we are ready to route packet.
1853 fl4
.flowi4_iif
= dev
->ifindex
;
1854 fl4
.flowi4_mark
= skb
->mark
;
1855 fl4
.flowi4_tos
= tos
;
1856 fl4
.flowi4_scope
= RT_SCOPE_UNIVERSE
;
1857 fl4
.flowi4_flags
= 0;
1860 err
= fib_lookup(net
, &fl4
, &res
, 0);
1862 if (!IN_DEV_FORWARD(in_dev
))
1863 err
= -EHOSTUNREACH
;
1867 if (res
.type
== RTN_BROADCAST
)
1870 if (res
.type
== RTN_LOCAL
) {
1871 err
= fib_validate_source(skb
, saddr
, daddr
, tos
,
1872 0, dev
, in_dev
, &itag
);
1874 goto martian_source
;
1878 if (!IN_DEV_FORWARD(in_dev
)) {
1879 err
= -EHOSTUNREACH
;
1882 if (res
.type
!= RTN_UNICAST
)
1883 goto martian_destination
;
1885 err
= ip_mkroute_input(skb
, &res
, in_dev
, daddr
, saddr
, tos
);
1889 if (skb
->protocol
!= htons(ETH_P_IP
))
1892 if (!ipv4_is_zeronet(saddr
)) {
1893 err
= fib_validate_source(skb
, saddr
, 0, tos
, 0, dev
,
1896 goto martian_source
;
1898 flags
|= RTCF_BROADCAST
;
1899 res
.type
= RTN_BROADCAST
;
1900 RT_CACHE_STAT_INC(in_brd
);
1906 rth
= rcu_dereference(FIB_RES_NH(res
).nh_rth_input
);
1907 if (rt_cache_valid(rth
)) {
1908 skb_dst_set_noref(skb
, &rth
->dst
);
1916 rth
= rt_dst_alloc(l3mdev_master_dev_rcu(dev
) ? : net
->loopback_dev
,
1917 flags
| RTCF_LOCAL
, res
.type
,
1918 IN_DEV_CONF_GET(in_dev
, NOPOLICY
), false, do_cache
);
1922 rth
->dst
.output
= ip_rt_bug
;
1923 #ifdef CONFIG_IP_ROUTE_CLASSID
1924 rth
->dst
.tclassid
= itag
;
1926 rth
->rt_is_input
= 1;
1928 rth
->rt_table_id
= res
.table
->tb_id
;
1930 RT_CACHE_STAT_INC(in_slow_tot
);
1931 if (res
.type
== RTN_UNREACHABLE
) {
1932 rth
->dst
.input
= ip_error
;
1933 rth
->dst
.error
= -err
;
1934 rth
->rt_flags
&= ~RTCF_LOCAL
;
1938 struct fib_nh
*nh
= &FIB_RES_NH(res
);
1940 rth
->dst
.lwtstate
= lwtstate_get(nh
->nh_lwtstate
);
1941 if (lwtunnel_input_redirect(rth
->dst
.lwtstate
)) {
1942 WARN_ON(rth
->dst
.input
== lwtunnel_input
);
1943 rth
->dst
.lwtstate
->orig_input
= rth
->dst
.input
;
1944 rth
->dst
.input
= lwtunnel_input
;
1947 if (unlikely(!rt_cache_route(nh
, rth
))) {
1948 rth
->dst
.flags
|= DST_NOCACHE
;
1949 rt_add_uncached_list(rth
);
1952 skb_dst_set(skb
, &rth
->dst
);
1957 RT_CACHE_STAT_INC(in_no_route
);
1958 res
.type
= RTN_UNREACHABLE
;
1964 * Do not cache martian addresses: they should be logged (RFC1812)
1966 martian_destination
:
1967 RT_CACHE_STAT_INC(in_martian_dst
);
1968 #ifdef CONFIG_IP_ROUTE_VERBOSE
1969 if (IN_DEV_LOG_MARTIANS(in_dev
))
1970 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1971 &daddr
, &saddr
, dev
->name
);
1983 ip_handle_martian_source(dev
, in_dev
, skb
, daddr
, saddr
);
1987 int ip_route_input_noref(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
1988 u8 tos
, struct net_device
*dev
)
1994 /* Multicast recognition logic is moved from route cache to here.
1995 The problem was that too many Ethernet cards have broken/missing
1996 hardware multicast filters :-( As result the host on multicasting
1997 network acquires a lot of useless route cache entries, sort of
1998 SDR messages from all the world. Now we try to get rid of them.
1999 Really, provided software IP multicast filter is organized
2000 reasonably (at least, hashed), it does not result in a slowdown
2001 comparing with route cache reject entries.
2002 Note, that multicast routers are not affected, because
2003 route cache entry is created eventually.
2005 if (ipv4_is_multicast(daddr
)) {
2006 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
2010 our
= ip_check_mc_rcu(in_dev
, daddr
, saddr
,
2011 ip_hdr(skb
)->protocol
);
2013 /* check l3 master if no match yet */
2014 if ((!in_dev
|| !our
) && netif_is_l3_slave(dev
)) {
2015 struct in_device
*l3_in_dev
;
2017 l3_in_dev
= __in_dev_get_rcu(skb
->dev
);
2019 our
= ip_check_mc_rcu(l3_in_dev
, daddr
, saddr
,
2020 ip_hdr(skb
)->protocol
);
2025 #ifdef CONFIG_IP_MROUTE
2027 (!ipv4_is_local_multicast(daddr
) &&
2028 IN_DEV_MFORWARD(in_dev
))
2031 res
= ip_route_input_mc(skb
, daddr
, saddr
,
2037 res
= ip_route_input_slow(skb
, daddr
, saddr
, tos
, dev
);
2041 EXPORT_SYMBOL(ip_route_input_noref
);
2043 /* called with rcu_read_lock() */
2044 static struct rtable
*__mkroute_output(const struct fib_result
*res
,
2045 const struct flowi4
*fl4
, int orig_oif
,
2046 struct net_device
*dev_out
,
2049 struct fib_info
*fi
= res
->fi
;
2050 struct fib_nh_exception
*fnhe
;
2051 struct in_device
*in_dev
;
2052 u16 type
= res
->type
;
2056 in_dev
= __in_dev_get_rcu(dev_out
);
2058 return ERR_PTR(-EINVAL
);
2060 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev
)))
2061 if (ipv4_is_loopback(fl4
->saddr
) &&
2062 !(dev_out
->flags
& IFF_LOOPBACK
) &&
2063 !netif_is_l3_master(dev_out
))
2064 return ERR_PTR(-EINVAL
);
2066 if (ipv4_is_lbcast(fl4
->daddr
))
2067 type
= RTN_BROADCAST
;
2068 else if (ipv4_is_multicast(fl4
->daddr
))
2069 type
= RTN_MULTICAST
;
2070 else if (ipv4_is_zeronet(fl4
->daddr
))
2071 return ERR_PTR(-EINVAL
);
2073 if (dev_out
->flags
& IFF_LOOPBACK
)
2074 flags
|= RTCF_LOCAL
;
2077 if (type
== RTN_BROADCAST
) {
2078 flags
|= RTCF_BROADCAST
| RTCF_LOCAL
;
2080 } else if (type
== RTN_MULTICAST
) {
2081 flags
|= RTCF_MULTICAST
| RTCF_LOCAL
;
2082 if (!ip_check_mc_rcu(in_dev
, fl4
->daddr
, fl4
->saddr
,
2084 flags
&= ~RTCF_LOCAL
;
2087 /* If multicast route do not exist use
2088 * default one, but do not gateway in this case.
2091 if (fi
&& res
->prefixlen
< 4)
2093 } else if ((type
== RTN_LOCAL
) && (orig_oif
!= 0) &&
2094 (orig_oif
!= dev_out
->ifindex
)) {
2095 /* For local routes that require a particular output interface
2096 * we do not want to cache the result. Caching the result
2097 * causes incorrect behaviour when there are multiple source
2098 * addresses on the interface, the end result being that if the
2099 * intended recipient is waiting on that interface for the
2100 * packet he won't receive it because it will be delivered on
2101 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2102 * be set to the loopback interface as well.
2108 do_cache
&= fi
!= NULL
;
2110 struct rtable __rcu
**prth
;
2111 struct fib_nh
*nh
= &FIB_RES_NH(*res
);
2113 fnhe
= find_exception(nh
, fl4
->daddr
);
2115 prth
= &fnhe
->fnhe_rth_output
;
2116 rth
= rcu_dereference(*prth
);
2117 if (rth
&& rth
->dst
.expires
&&
2118 time_after(jiffies
, rth
->dst
.expires
)) {
2119 ip_del_fnhe(nh
, fl4
->daddr
);
2126 if (unlikely(fl4
->flowi4_flags
&
2127 FLOWI_FLAG_KNOWN_NH
&&
2129 nh
->nh_scope
== RT_SCOPE_LINK
))) {
2133 prth
= raw_cpu_ptr(nh
->nh_pcpu_rth_output
);
2134 rth
= rcu_dereference(*prth
);
2137 if (rt_cache_valid(rth
)) {
2138 dst_hold(&rth
->dst
);
2144 rth
= rt_dst_alloc(dev_out
, flags
, type
,
2145 IN_DEV_CONF_GET(in_dev
, NOPOLICY
),
2146 IN_DEV_CONF_GET(in_dev
, NOXFRM
),
2149 return ERR_PTR(-ENOBUFS
);
2151 rth
->rt_iif
= orig_oif
? : 0;
2153 rth
->rt_table_id
= res
->table
->tb_id
;
2155 RT_CACHE_STAT_INC(out_slow_tot
);
2157 if (flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) {
2158 if (flags
& RTCF_LOCAL
&&
2159 !(dev_out
->flags
& IFF_LOOPBACK
)) {
2160 rth
->dst
.output
= ip_mc_output
;
2161 RT_CACHE_STAT_INC(out_slow_mc
);
2163 #ifdef CONFIG_IP_MROUTE
2164 if (type
== RTN_MULTICAST
) {
2165 if (IN_DEV_MFORWARD(in_dev
) &&
2166 !ipv4_is_local_multicast(fl4
->daddr
)) {
2167 rth
->dst
.input
= ip_mr_input
;
2168 rth
->dst
.output
= ip_mc_output
;
2174 rt_set_nexthop(rth
, fl4
->daddr
, res
, fnhe
, fi
, type
, 0);
2175 set_lwt_redirect(rth
);
2181 * Major route resolver routine.
2184 struct rtable
*__ip_route_output_key_hash(struct net
*net
, struct flowi4
*fl4
,
2187 struct net_device
*dev_out
= NULL
;
2188 __u8 tos
= RT_FL_TOS(fl4
);
2189 unsigned int flags
= 0;
2190 struct fib_result res
;
2193 int err
= -ENETUNREACH
;
2199 orig_oif
= fl4
->flowi4_oif
;
2201 fl4
->flowi4_iif
= LOOPBACK_IFINDEX
;
2202 fl4
->flowi4_tos
= tos
& IPTOS_RT_MASK
;
2203 fl4
->flowi4_scope
= ((tos
& RTO_ONLINK
) ?
2204 RT_SCOPE_LINK
: RT_SCOPE_UNIVERSE
);
2208 rth
= ERR_PTR(-EINVAL
);
2209 if (ipv4_is_multicast(fl4
->saddr
) ||
2210 ipv4_is_lbcast(fl4
->saddr
) ||
2211 ipv4_is_zeronet(fl4
->saddr
))
2214 /* I removed check for oif == dev_out->oif here.
2215 It was wrong for two reasons:
2216 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2217 is assigned to multiple interfaces.
2218 2. Moreover, we are allowed to send packets with saddr
2219 of another iface. --ANK
2222 if (fl4
->flowi4_oif
== 0 &&
2223 (ipv4_is_multicast(fl4
->daddr
) ||
2224 ipv4_is_lbcast(fl4
->daddr
))) {
2225 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2226 dev_out
= __ip_dev_find(net
, fl4
->saddr
, false);
2230 /* Special hack: user can direct multicasts
2231 and limited broadcast via necessary interface
2232 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2233 This hack is not just for fun, it allows
2234 vic,vat and friends to work.
2235 They bind socket to loopback, set ttl to zero
2236 and expect that it will work.
2237 From the viewpoint of routing cache they are broken,
2238 because we are not allowed to build multicast path
2239 with loopback source addr (look, routing cache
2240 cannot know, that ttl is zero, so that packet
2241 will not leave this host and route is valid).
2242 Luckily, this hack is good workaround.
2245 fl4
->flowi4_oif
= dev_out
->ifindex
;
2249 if (!(fl4
->flowi4_flags
& FLOWI_FLAG_ANYSRC
)) {
2250 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2251 if (!__ip_dev_find(net
, fl4
->saddr
, false))
2257 if (fl4
->flowi4_oif
) {
2258 dev_out
= dev_get_by_index_rcu(net
, fl4
->flowi4_oif
);
2259 rth
= ERR_PTR(-ENODEV
);
2263 /* RACE: Check return value of inet_select_addr instead. */
2264 if (!(dev_out
->flags
& IFF_UP
) || !__in_dev_get_rcu(dev_out
)) {
2265 rth
= ERR_PTR(-ENETUNREACH
);
2268 if (ipv4_is_local_multicast(fl4
->daddr
) ||
2269 ipv4_is_lbcast(fl4
->daddr
) ||
2270 fl4
->flowi4_proto
== IPPROTO_IGMP
) {
2272 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2277 if (ipv4_is_multicast(fl4
->daddr
))
2278 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2280 else if (!fl4
->daddr
)
2281 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2287 fl4
->daddr
= fl4
->saddr
;
2289 fl4
->daddr
= fl4
->saddr
= htonl(INADDR_LOOPBACK
);
2290 dev_out
= net
->loopback_dev
;
2291 fl4
->flowi4_oif
= LOOPBACK_IFINDEX
;
2292 res
.type
= RTN_LOCAL
;
2293 flags
|= RTCF_LOCAL
;
2297 err
= fib_lookup(net
, fl4
, &res
, 0);
2301 if (fl4
->flowi4_oif
&&
2302 (ipv4_is_multicast(fl4
->daddr
) ||
2303 !netif_index_is_l3_master(net
, fl4
->flowi4_oif
))) {
2304 /* Apparently, routing tables are wrong. Assume,
2305 that the destination is on link.
2308 Because we are allowed to send to iface
2309 even if it has NO routes and NO assigned
2310 addresses. When oif is specified, routing
2311 tables are looked up with only one purpose:
2312 to catch if destination is gatewayed, rather than
2313 direct. Moreover, if MSG_DONTROUTE is set,
2314 we send packet, ignoring both routing tables
2315 and ifaddr state. --ANK
2318 We could make it even if oif is unknown,
2319 likely IPv6, but we do not.
2322 if (fl4
->saddr
== 0)
2323 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2325 res
.type
= RTN_UNICAST
;
2332 if (res
.type
== RTN_LOCAL
) {
2334 if (res
.fi
->fib_prefsrc
)
2335 fl4
->saddr
= res
.fi
->fib_prefsrc
;
2337 fl4
->saddr
= fl4
->daddr
;
2340 /* L3 master device is the loopback for that domain */
2341 dev_out
= l3mdev_master_dev_rcu(dev_out
) ? : net
->loopback_dev
;
2342 fl4
->flowi4_oif
= dev_out
->ifindex
;
2343 flags
|= RTCF_LOCAL
;
2347 fib_select_path(net
, &res
, fl4
, mp_hash
);
2349 dev_out
= FIB_RES_DEV(res
);
2350 fl4
->flowi4_oif
= dev_out
->ifindex
;
2354 rth
= __mkroute_output(&res
, fl4
, orig_oif
, dev_out
, flags
);
2360 EXPORT_SYMBOL_GPL(__ip_route_output_key_hash
);
2362 static struct dst_entry
*ipv4_blackhole_dst_check(struct dst_entry
*dst
, u32 cookie
)
2367 static unsigned int ipv4_blackhole_mtu(const struct dst_entry
*dst
)
2369 unsigned int mtu
= dst_metric_raw(dst
, RTAX_MTU
);
2371 return mtu
? : dst
->dev
->mtu
;
2374 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry
*dst
, struct sock
*sk
,
2375 struct sk_buff
*skb
, u32 mtu
)
2379 static void ipv4_rt_blackhole_redirect(struct dst_entry
*dst
, struct sock
*sk
,
2380 struct sk_buff
*skb
)
2384 static u32
*ipv4_rt_blackhole_cow_metrics(struct dst_entry
*dst
,
2390 static struct dst_ops ipv4_dst_blackhole_ops
= {
2392 .check
= ipv4_blackhole_dst_check
,
2393 .mtu
= ipv4_blackhole_mtu
,
2394 .default_advmss
= ipv4_default_advmss
,
2395 .update_pmtu
= ipv4_rt_blackhole_update_pmtu
,
2396 .redirect
= ipv4_rt_blackhole_redirect
,
2397 .cow_metrics
= ipv4_rt_blackhole_cow_metrics
,
2398 .neigh_lookup
= ipv4_neigh_lookup
,
2401 struct dst_entry
*ipv4_blackhole_route(struct net
*net
, struct dst_entry
*dst_orig
)
2403 struct rtable
*ort
= (struct rtable
*) dst_orig
;
2406 rt
= dst_alloc(&ipv4_dst_blackhole_ops
, NULL
, 1, DST_OBSOLETE_NONE
, 0);
2408 struct dst_entry
*new = &rt
->dst
;
2411 new->input
= dst_discard
;
2412 new->output
= dst_discard_out
;
2414 new->dev
= ort
->dst
.dev
;
2418 rt
->rt_is_input
= ort
->rt_is_input
;
2419 rt
->rt_iif
= ort
->rt_iif
;
2420 rt
->rt_pmtu
= ort
->rt_pmtu
;
2422 rt
->rt_genid
= rt_genid_ipv4(net
);
2423 rt
->rt_flags
= ort
->rt_flags
;
2424 rt
->rt_type
= ort
->rt_type
;
2425 rt
->rt_gateway
= ort
->rt_gateway
;
2426 rt
->rt_uses_gateway
= ort
->rt_uses_gateway
;
2428 INIT_LIST_HEAD(&rt
->rt_uncached
);
2432 dst_release(dst_orig
);
2434 return rt
? &rt
->dst
: ERR_PTR(-ENOMEM
);
2437 struct rtable
*ip_route_output_flow(struct net
*net
, struct flowi4
*flp4
,
2438 const struct sock
*sk
)
2440 struct rtable
*rt
= __ip_route_output_key(net
, flp4
);
2445 if (flp4
->flowi4_proto
)
2446 rt
= (struct rtable
*)xfrm_lookup_route(net
, &rt
->dst
,
2447 flowi4_to_flowi(flp4
),
2452 EXPORT_SYMBOL_GPL(ip_route_output_flow
);
2454 static int rt_fill_info(struct net
*net
, __be32 dst
, __be32 src
, u32 table_id
,
2455 struct flowi4
*fl4
, struct sk_buff
*skb
, u32 portid
,
2458 struct rtable
*rt
= skb_rtable(skb
);
2460 struct nlmsghdr
*nlh
;
2461 unsigned long expires
= 0;
2463 u32 metrics
[RTAX_MAX
];
2465 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*r
), 0);
2469 r
= nlmsg_data(nlh
);
2470 r
->rtm_family
= AF_INET
;
2471 r
->rtm_dst_len
= 32;
2473 r
->rtm_tos
= fl4
->flowi4_tos
;
2474 r
->rtm_table
= table_id
< 256 ? table_id
: RT_TABLE_COMPAT
;
2475 if (nla_put_u32(skb
, RTA_TABLE
, table_id
))
2476 goto nla_put_failure
;
2477 r
->rtm_type
= rt
->rt_type
;
2478 r
->rtm_scope
= RT_SCOPE_UNIVERSE
;
2479 r
->rtm_protocol
= RTPROT_UNSPEC
;
2480 r
->rtm_flags
= (rt
->rt_flags
& ~0xFFFF) | RTM_F_CLONED
;
2481 if (rt
->rt_flags
& RTCF_NOTIFY
)
2482 r
->rtm_flags
|= RTM_F_NOTIFY
;
2483 if (IPCB(skb
)->flags
& IPSKB_DOREDIRECT
)
2484 r
->rtm_flags
|= RTCF_DOREDIRECT
;
2486 if (nla_put_in_addr(skb
, RTA_DST
, dst
))
2487 goto nla_put_failure
;
2489 r
->rtm_src_len
= 32;
2490 if (nla_put_in_addr(skb
, RTA_SRC
, src
))
2491 goto nla_put_failure
;
2494 nla_put_u32(skb
, RTA_OIF
, rt
->dst
.dev
->ifindex
))
2495 goto nla_put_failure
;
2496 #ifdef CONFIG_IP_ROUTE_CLASSID
2497 if (rt
->dst
.tclassid
&&
2498 nla_put_u32(skb
, RTA_FLOW
, rt
->dst
.tclassid
))
2499 goto nla_put_failure
;
2501 if (!rt_is_input_route(rt
) &&
2502 fl4
->saddr
!= src
) {
2503 if (nla_put_in_addr(skb
, RTA_PREFSRC
, fl4
->saddr
))
2504 goto nla_put_failure
;
2506 if (rt
->rt_uses_gateway
&&
2507 nla_put_in_addr(skb
, RTA_GATEWAY
, rt
->rt_gateway
))
2508 goto nla_put_failure
;
2510 expires
= rt
->dst
.expires
;
2512 unsigned long now
= jiffies
;
2514 if (time_before(now
, expires
))
2520 memcpy(metrics
, dst_metrics_ptr(&rt
->dst
), sizeof(metrics
));
2521 if (rt
->rt_pmtu
&& expires
)
2522 metrics
[RTAX_MTU
- 1] = rt
->rt_pmtu
;
2523 if (rtnetlink_put_metrics(skb
, metrics
) < 0)
2524 goto nla_put_failure
;
2526 if (fl4
->flowi4_mark
&&
2527 nla_put_u32(skb
, RTA_MARK
, fl4
->flowi4_mark
))
2528 goto nla_put_failure
;
2530 if (!uid_eq(fl4
->flowi4_uid
, INVALID_UID
) &&
2531 nla_put_u32(skb
, RTA_UID
,
2532 from_kuid_munged(current_user_ns(), fl4
->flowi4_uid
)))
2533 goto nla_put_failure
;
2535 error
= rt
->dst
.error
;
2537 if (rt_is_input_route(rt
)) {
2538 #ifdef CONFIG_IP_MROUTE
2539 if (ipv4_is_multicast(dst
) && !ipv4_is_local_multicast(dst
) &&
2540 IPV4_DEVCONF_ALL(net
, MC_FORWARDING
)) {
2541 int err
= ipmr_get_route(net
, skb
,
2542 fl4
->saddr
, fl4
->daddr
,
2548 goto nla_put_failure
;
2552 if (nla_put_u32(skb
, RTA_IIF
, skb
->dev
->ifindex
))
2553 goto nla_put_failure
;
2556 if (rtnl_put_cacheinfo(skb
, &rt
->dst
, 0, expires
, error
) < 0)
2557 goto nla_put_failure
;
2559 nlmsg_end(skb
, nlh
);
2563 nlmsg_cancel(skb
, nlh
);
2567 static int inet_rtm_getroute(struct sk_buff
*in_skb
, struct nlmsghdr
*nlh
)
2569 struct net
*net
= sock_net(in_skb
->sk
);
2571 struct nlattr
*tb
[RTA_MAX
+1];
2572 struct rtable
*rt
= NULL
;
2579 struct sk_buff
*skb
;
2580 u32 table_id
= RT_TABLE_MAIN
;
2583 err
= nlmsg_parse(nlh
, sizeof(*rtm
), tb
, RTA_MAX
, rtm_ipv4_policy
);
2587 rtm
= nlmsg_data(nlh
);
2589 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
2595 /* Reserve room for dummy headers, this skb can pass
2596 through good chunk of routing engine.
2598 skb_reset_mac_header(skb
);
2599 skb_reset_network_header(skb
);
2601 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2602 ip_hdr(skb
)->protocol
= IPPROTO_ICMP
;
2603 skb_reserve(skb
, MAX_HEADER
+ sizeof(struct iphdr
));
2605 src
= tb
[RTA_SRC
] ? nla_get_in_addr(tb
[RTA_SRC
]) : 0;
2606 dst
= tb
[RTA_DST
] ? nla_get_in_addr(tb
[RTA_DST
]) : 0;
2607 iif
= tb
[RTA_IIF
] ? nla_get_u32(tb
[RTA_IIF
]) : 0;
2608 mark
= tb
[RTA_MARK
] ? nla_get_u32(tb
[RTA_MARK
]) : 0;
2610 uid
= make_kuid(current_user_ns(), nla_get_u32(tb
[RTA_UID
]));
2612 uid
= (iif
? INVALID_UID
: current_uid());
2614 memset(&fl4
, 0, sizeof(fl4
));
2617 fl4
.flowi4_tos
= rtm
->rtm_tos
;
2618 fl4
.flowi4_oif
= tb
[RTA_OIF
] ? nla_get_u32(tb
[RTA_OIF
]) : 0;
2619 fl4
.flowi4_mark
= mark
;
2620 fl4
.flowi4_uid
= uid
;
2623 struct net_device
*dev
;
2625 dev
= __dev_get_by_index(net
, iif
);
2631 skb
->protocol
= htons(ETH_P_IP
);
2634 err
= ip_route_input(skb
, dst
, src
, rtm
->rtm_tos
, dev
);
2636 rt
= skb_rtable(skb
);
2637 if (err
== 0 && rt
->dst
.error
)
2638 err
= -rt
->dst
.error
;
2640 rt
= ip_route_output_key(net
, &fl4
);
2650 skb_dst_set(skb
, &rt
->dst
);
2651 if (rtm
->rtm_flags
& RTM_F_NOTIFY
)
2652 rt
->rt_flags
|= RTCF_NOTIFY
;
2654 if (rtm
->rtm_flags
& RTM_F_LOOKUP_TABLE
)
2655 table_id
= rt
->rt_table_id
;
2657 err
= rt_fill_info(net
, dst
, src
, table_id
, &fl4
, skb
,
2658 NETLINK_CB(in_skb
).portid
, nlh
->nlmsg_seq
,
2663 err
= rtnl_unicast(skb
, net
, NETLINK_CB(in_skb
).portid
);
2672 void ip_rt_multicast_event(struct in_device
*in_dev
)
2674 rt_cache_flush(dev_net(in_dev
->dev
));
2677 #ifdef CONFIG_SYSCTL
2678 static int ip_rt_gc_interval __read_mostly
= 60 * HZ
;
2679 static int ip_rt_gc_min_interval __read_mostly
= HZ
/ 2;
2680 static int ip_rt_gc_elasticity __read_mostly
= 8;
2682 static int ipv4_sysctl_rtcache_flush(struct ctl_table
*__ctl
, int write
,
2683 void __user
*buffer
,
2684 size_t *lenp
, loff_t
*ppos
)
2686 struct net
*net
= (struct net
*)__ctl
->extra1
;
2689 rt_cache_flush(net
);
2690 fnhe_genid_bump(net
);
2697 static struct ctl_table ipv4_route_table
[] = {
2699 .procname
= "gc_thresh",
2700 .data
= &ipv4_dst_ops
.gc_thresh
,
2701 .maxlen
= sizeof(int),
2703 .proc_handler
= proc_dointvec
,
2706 .procname
= "max_size",
2707 .data
= &ip_rt_max_size
,
2708 .maxlen
= sizeof(int),
2710 .proc_handler
= proc_dointvec
,
2713 /* Deprecated. Use gc_min_interval_ms */
2715 .procname
= "gc_min_interval",
2716 .data
= &ip_rt_gc_min_interval
,
2717 .maxlen
= sizeof(int),
2719 .proc_handler
= proc_dointvec_jiffies
,
2722 .procname
= "gc_min_interval_ms",
2723 .data
= &ip_rt_gc_min_interval
,
2724 .maxlen
= sizeof(int),
2726 .proc_handler
= proc_dointvec_ms_jiffies
,
2729 .procname
= "gc_timeout",
2730 .data
= &ip_rt_gc_timeout
,
2731 .maxlen
= sizeof(int),
2733 .proc_handler
= proc_dointvec_jiffies
,
2736 .procname
= "gc_interval",
2737 .data
= &ip_rt_gc_interval
,
2738 .maxlen
= sizeof(int),
2740 .proc_handler
= proc_dointvec_jiffies
,
2743 .procname
= "redirect_load",
2744 .data
= &ip_rt_redirect_load
,
2745 .maxlen
= sizeof(int),
2747 .proc_handler
= proc_dointvec
,
2750 .procname
= "redirect_number",
2751 .data
= &ip_rt_redirect_number
,
2752 .maxlen
= sizeof(int),
2754 .proc_handler
= proc_dointvec
,
2757 .procname
= "redirect_silence",
2758 .data
= &ip_rt_redirect_silence
,
2759 .maxlen
= sizeof(int),
2761 .proc_handler
= proc_dointvec
,
2764 .procname
= "error_cost",
2765 .data
= &ip_rt_error_cost
,
2766 .maxlen
= sizeof(int),
2768 .proc_handler
= proc_dointvec
,
2771 .procname
= "error_burst",
2772 .data
= &ip_rt_error_burst
,
2773 .maxlen
= sizeof(int),
2775 .proc_handler
= proc_dointvec
,
2778 .procname
= "gc_elasticity",
2779 .data
= &ip_rt_gc_elasticity
,
2780 .maxlen
= sizeof(int),
2782 .proc_handler
= proc_dointvec
,
2785 .procname
= "mtu_expires",
2786 .data
= &ip_rt_mtu_expires
,
2787 .maxlen
= sizeof(int),
2789 .proc_handler
= proc_dointvec_jiffies
,
2792 .procname
= "min_pmtu",
2793 .data
= &ip_rt_min_pmtu
,
2794 .maxlen
= sizeof(int),
2796 .proc_handler
= proc_dointvec
,
2799 .procname
= "min_adv_mss",
2800 .data
= &ip_rt_min_advmss
,
2801 .maxlen
= sizeof(int),
2803 .proc_handler
= proc_dointvec
,
2808 static struct ctl_table ipv4_route_flush_table
[] = {
2810 .procname
= "flush",
2811 .maxlen
= sizeof(int),
2813 .proc_handler
= ipv4_sysctl_rtcache_flush
,
2818 static __net_init
int sysctl_route_net_init(struct net
*net
)
2820 struct ctl_table
*tbl
;
2822 tbl
= ipv4_route_flush_table
;
2823 if (!net_eq(net
, &init_net
)) {
2824 tbl
= kmemdup(tbl
, sizeof(ipv4_route_flush_table
), GFP_KERNEL
);
2828 /* Don't export sysctls to unprivileged users */
2829 if (net
->user_ns
!= &init_user_ns
)
2830 tbl
[0].procname
= NULL
;
2832 tbl
[0].extra1
= net
;
2834 net
->ipv4
.route_hdr
= register_net_sysctl(net
, "net/ipv4/route", tbl
);
2835 if (!net
->ipv4
.route_hdr
)
2840 if (tbl
!= ipv4_route_flush_table
)
2846 static __net_exit
void sysctl_route_net_exit(struct net
*net
)
2848 struct ctl_table
*tbl
;
2850 tbl
= net
->ipv4
.route_hdr
->ctl_table_arg
;
2851 unregister_net_sysctl_table(net
->ipv4
.route_hdr
);
2852 BUG_ON(tbl
== ipv4_route_flush_table
);
2856 static __net_initdata
struct pernet_operations sysctl_route_ops
= {
2857 .init
= sysctl_route_net_init
,
2858 .exit
= sysctl_route_net_exit
,
2862 static __net_init
int rt_genid_init(struct net
*net
)
2864 atomic_set(&net
->ipv4
.rt_genid
, 0);
2865 atomic_set(&net
->fnhe_genid
, 0);
2866 get_random_bytes(&net
->ipv4
.dev_addr_genid
,
2867 sizeof(net
->ipv4
.dev_addr_genid
));
2871 static __net_initdata
struct pernet_operations rt_genid_ops
= {
2872 .init
= rt_genid_init
,
2875 static int __net_init
ipv4_inetpeer_init(struct net
*net
)
2877 struct inet_peer_base
*bp
= kmalloc(sizeof(*bp
), GFP_KERNEL
);
2881 inet_peer_base_init(bp
);
2882 net
->ipv4
.peers
= bp
;
2886 static void __net_exit
ipv4_inetpeer_exit(struct net
*net
)
2888 struct inet_peer_base
*bp
= net
->ipv4
.peers
;
2890 net
->ipv4
.peers
= NULL
;
2891 inetpeer_invalidate_tree(bp
);
2895 static __net_initdata
struct pernet_operations ipv4_inetpeer_ops
= {
2896 .init
= ipv4_inetpeer_init
,
2897 .exit
= ipv4_inetpeer_exit
,
2900 #ifdef CONFIG_IP_ROUTE_CLASSID
2901 struct ip_rt_acct __percpu
*ip_rt_acct __read_mostly
;
2902 #endif /* CONFIG_IP_ROUTE_CLASSID */
2904 int __init
ip_rt_init(void)
2909 ip_idents
= kmalloc(IP_IDENTS_SZ
* sizeof(*ip_idents
), GFP_KERNEL
);
2911 panic("IP: failed to allocate ip_idents\n");
2913 prandom_bytes(ip_idents
, IP_IDENTS_SZ
* sizeof(*ip_idents
));
2915 ip_tstamps
= kcalloc(IP_IDENTS_SZ
, sizeof(*ip_tstamps
), GFP_KERNEL
);
2917 panic("IP: failed to allocate ip_tstamps\n");
2919 for_each_possible_cpu(cpu
) {
2920 struct uncached_list
*ul
= &per_cpu(rt_uncached_list
, cpu
);
2922 INIT_LIST_HEAD(&ul
->head
);
2923 spin_lock_init(&ul
->lock
);
2925 #ifdef CONFIG_IP_ROUTE_CLASSID
2926 ip_rt_acct
= __alloc_percpu(256 * sizeof(struct ip_rt_acct
), __alignof__(struct ip_rt_acct
));
2928 panic("IP: failed to allocate ip_rt_acct\n");
2931 ipv4_dst_ops
.kmem_cachep
=
2932 kmem_cache_create("ip_dst_cache", sizeof(struct rtable
), 0,
2933 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2935 ipv4_dst_blackhole_ops
.kmem_cachep
= ipv4_dst_ops
.kmem_cachep
;
2937 if (dst_entries_init(&ipv4_dst_ops
) < 0)
2938 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2940 if (dst_entries_init(&ipv4_dst_blackhole_ops
) < 0)
2941 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2943 ipv4_dst_ops
.gc_thresh
= ~0;
2944 ip_rt_max_size
= INT_MAX
;
2949 if (ip_rt_proc_init())
2950 pr_err("Unable to create route proc files\n");
2955 rtnl_register(PF_INET
, RTM_GETROUTE
, inet_rtm_getroute
, NULL
, NULL
);
2957 #ifdef CONFIG_SYSCTL
2958 register_pernet_subsys(&sysctl_route_ops
);
2960 register_pernet_subsys(&rt_genid_ops
);
2961 register_pernet_subsys(&ipv4_inetpeer_ops
);
2965 #ifdef CONFIG_SYSCTL
2967 * We really need to sanitize the damn ipv4 init order, then all
2968 * this nonsense will go away.
2970 void __init
ip_static_sysctl_init(void)
2972 register_net_sysctl(&init_net
, "net/ipv4/route", ipv4_route_table
);