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>
112 #include <net/secure_seq.h>
113 #include <net/ip_tunnels.h>
114 #include <net/l3mdev.h>
116 #include "fib_lookup.h"
118 #define RT_FL_TOS(oldflp4) \
119 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
121 #define RT_GC_TIMEOUT (300*HZ)
123 static int ip_rt_max_size
;
124 static int ip_rt_redirect_number __read_mostly
= 9;
125 static int ip_rt_redirect_load __read_mostly
= HZ
/ 50;
126 static int ip_rt_redirect_silence __read_mostly
= ((HZ
/ 50) << (9 + 1));
127 static int ip_rt_error_cost __read_mostly
= HZ
;
128 static int ip_rt_error_burst __read_mostly
= 5 * HZ
;
129 static int ip_rt_mtu_expires __read_mostly
= 10 * 60 * HZ
;
130 static u32 ip_rt_min_pmtu __read_mostly
= 512 + 20 + 20;
131 static int ip_rt_min_advmss __read_mostly
= 256;
133 static int ip_rt_gc_timeout __read_mostly
= RT_GC_TIMEOUT
;
136 * Interface to generic destination cache.
139 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
);
140 static unsigned int ipv4_default_advmss(const struct dst_entry
*dst
);
141 static unsigned int ipv4_mtu(const struct dst_entry
*dst
);
142 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
);
143 static void ipv4_link_failure(struct sk_buff
*skb
);
144 static void ip_rt_update_pmtu(struct dst_entry
*dst
, struct sock
*sk
,
145 struct sk_buff
*skb
, u32 mtu
);
146 static void ip_do_redirect(struct dst_entry
*dst
, struct sock
*sk
,
147 struct sk_buff
*skb
);
148 static void ipv4_dst_destroy(struct dst_entry
*dst
);
150 static u32
*ipv4_cow_metrics(struct dst_entry
*dst
, unsigned long old
)
156 static struct neighbour
*ipv4_neigh_lookup(const struct dst_entry
*dst
,
159 static void ipv4_confirm_neigh(const struct dst_entry
*dst
, const void *daddr
);
161 static struct dst_ops ipv4_dst_ops
= {
163 .check
= ipv4_dst_check
,
164 .default_advmss
= ipv4_default_advmss
,
166 .cow_metrics
= ipv4_cow_metrics
,
167 .destroy
= ipv4_dst_destroy
,
168 .negative_advice
= ipv4_negative_advice
,
169 .link_failure
= ipv4_link_failure
,
170 .update_pmtu
= ip_rt_update_pmtu
,
171 .redirect
= ip_do_redirect
,
172 .local_out
= __ip_local_out
,
173 .neigh_lookup
= ipv4_neigh_lookup
,
174 .confirm_neigh
= ipv4_confirm_neigh
,
177 #define ECN_OR_COST(class) TC_PRIO_##class
179 const __u8 ip_tos2prio
[16] = {
181 ECN_OR_COST(BESTEFFORT
),
183 ECN_OR_COST(BESTEFFORT
),
189 ECN_OR_COST(INTERACTIVE
),
191 ECN_OR_COST(INTERACTIVE
),
192 TC_PRIO_INTERACTIVE_BULK
,
193 ECN_OR_COST(INTERACTIVE_BULK
),
194 TC_PRIO_INTERACTIVE_BULK
,
195 ECN_OR_COST(INTERACTIVE_BULK
)
197 EXPORT_SYMBOL(ip_tos2prio
);
199 static DEFINE_PER_CPU(struct rt_cache_stat
, rt_cache_stat
);
200 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
202 #ifdef CONFIG_PROC_FS
203 static void *rt_cache_seq_start(struct seq_file
*seq
, loff_t
*pos
)
207 return SEQ_START_TOKEN
;
210 static void *rt_cache_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
216 static void rt_cache_seq_stop(struct seq_file
*seq
, void *v
)
220 static int rt_cache_seq_show(struct seq_file
*seq
, void *v
)
222 if (v
== SEQ_START_TOKEN
)
223 seq_printf(seq
, "%-127s\n",
224 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
225 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
230 static const struct seq_operations rt_cache_seq_ops
= {
231 .start
= rt_cache_seq_start
,
232 .next
= rt_cache_seq_next
,
233 .stop
= rt_cache_seq_stop
,
234 .show
= rt_cache_seq_show
,
237 static int rt_cache_seq_open(struct inode
*inode
, struct file
*file
)
239 return seq_open(file
, &rt_cache_seq_ops
);
242 static const struct file_operations rt_cache_seq_fops
= {
243 .open
= rt_cache_seq_open
,
246 .release
= seq_release
,
250 static void *rt_cpu_seq_start(struct seq_file
*seq
, loff_t
*pos
)
255 return SEQ_START_TOKEN
;
257 for (cpu
= *pos
-1; cpu
< nr_cpu_ids
; ++cpu
) {
258 if (!cpu_possible(cpu
))
261 return &per_cpu(rt_cache_stat
, cpu
);
266 static void *rt_cpu_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
270 for (cpu
= *pos
; cpu
< nr_cpu_ids
; ++cpu
) {
271 if (!cpu_possible(cpu
))
274 return &per_cpu(rt_cache_stat
, cpu
);
280 static void rt_cpu_seq_stop(struct seq_file
*seq
, void *v
)
285 static int rt_cpu_seq_show(struct seq_file
*seq
, void *v
)
287 struct rt_cache_stat
*st
= v
;
289 if (v
== SEQ_START_TOKEN
) {
290 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");
294 seq_printf(seq
,"%08x %08x %08x %08x %08x %08x %08x %08x "
295 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
296 dst_entries_get_slow(&ipv4_dst_ops
),
309 0, /* st->gc_total */
310 0, /* st->gc_ignored */
311 0, /* st->gc_goal_miss */
312 0, /* st->gc_dst_overflow */
313 0, /* st->in_hlist_search */
314 0 /* st->out_hlist_search */
319 static const struct seq_operations rt_cpu_seq_ops
= {
320 .start
= rt_cpu_seq_start
,
321 .next
= rt_cpu_seq_next
,
322 .stop
= rt_cpu_seq_stop
,
323 .show
= rt_cpu_seq_show
,
327 static int rt_cpu_seq_open(struct inode
*inode
, struct file
*file
)
329 return seq_open(file
, &rt_cpu_seq_ops
);
332 static const struct file_operations rt_cpu_seq_fops
= {
333 .open
= rt_cpu_seq_open
,
336 .release
= seq_release
,
339 #ifdef CONFIG_IP_ROUTE_CLASSID
340 static int rt_acct_proc_show(struct seq_file
*m
, void *v
)
342 struct ip_rt_acct
*dst
, *src
;
345 dst
= kcalloc(256, sizeof(struct ip_rt_acct
), GFP_KERNEL
);
349 for_each_possible_cpu(i
) {
350 src
= (struct ip_rt_acct
*)per_cpu_ptr(ip_rt_acct
, i
);
351 for (j
= 0; j
< 256; j
++) {
352 dst
[j
].o_bytes
+= src
[j
].o_bytes
;
353 dst
[j
].o_packets
+= src
[j
].o_packets
;
354 dst
[j
].i_bytes
+= src
[j
].i_bytes
;
355 dst
[j
].i_packets
+= src
[j
].i_packets
;
359 seq_write(m
, dst
, 256 * sizeof(struct ip_rt_acct
));
365 static int __net_init
ip_rt_do_proc_init(struct net
*net
)
367 struct proc_dir_entry
*pde
;
369 pde
= proc_create("rt_cache", 0444, net
->proc_net
,
374 pde
= proc_create("rt_cache", 0444,
375 net
->proc_net_stat
, &rt_cpu_seq_fops
);
379 #ifdef CONFIG_IP_ROUTE_CLASSID
380 pde
= proc_create_single("rt_acct", 0, net
->proc_net
,
387 #ifdef CONFIG_IP_ROUTE_CLASSID
389 remove_proc_entry("rt_cache", net
->proc_net_stat
);
392 remove_proc_entry("rt_cache", net
->proc_net
);
397 static void __net_exit
ip_rt_do_proc_exit(struct net
*net
)
399 remove_proc_entry("rt_cache", net
->proc_net_stat
);
400 remove_proc_entry("rt_cache", net
->proc_net
);
401 #ifdef CONFIG_IP_ROUTE_CLASSID
402 remove_proc_entry("rt_acct", net
->proc_net
);
406 static struct pernet_operations ip_rt_proc_ops __net_initdata
= {
407 .init
= ip_rt_do_proc_init
,
408 .exit
= ip_rt_do_proc_exit
,
411 static int __init
ip_rt_proc_init(void)
413 return register_pernet_subsys(&ip_rt_proc_ops
);
417 static inline int ip_rt_proc_init(void)
421 #endif /* CONFIG_PROC_FS */
423 static inline bool rt_is_expired(const struct rtable
*rth
)
425 return rth
->rt_genid
!= rt_genid_ipv4(dev_net(rth
->dst
.dev
));
428 void rt_cache_flush(struct net
*net
)
430 rt_genid_bump_ipv4(net
);
433 static struct neighbour
*ipv4_neigh_lookup(const struct dst_entry
*dst
,
437 struct net_device
*dev
= dst
->dev
;
438 const __be32
*pkey
= daddr
;
439 const struct rtable
*rt
;
442 rt
= (const struct rtable
*) dst
;
444 pkey
= (const __be32
*) &rt
->rt_gateway
;
446 pkey
= &ip_hdr(skb
)->daddr
;
448 n
= __ipv4_neigh_lookup(dev
, *(__force u32
*)pkey
);
451 return neigh_create(&arp_tbl
, pkey
, dev
);
454 static void ipv4_confirm_neigh(const struct dst_entry
*dst
, const void *daddr
)
456 struct net_device
*dev
= dst
->dev
;
457 const __be32
*pkey
= daddr
;
458 const struct rtable
*rt
;
460 rt
= (const struct rtable
*)dst
;
462 pkey
= (const __be32
*)&rt
->rt_gateway
;
465 (RTCF_MULTICAST
| RTCF_BROADCAST
| RTCF_LOCAL
)))
468 __ipv4_confirm_neigh(dev
, *(__force u32
*)pkey
);
471 #define IP_IDENTS_SZ 2048u
473 static atomic_t
*ip_idents __read_mostly
;
474 static u32
*ip_tstamps __read_mostly
;
476 /* In order to protect privacy, we add a perturbation to identifiers
477 * if one generator is seldom used. This makes hard for an attacker
478 * to infer how many packets were sent between two points in time.
480 u32
ip_idents_reserve(u32 hash
, int segs
)
482 u32
*p_tstamp
= ip_tstamps
+ hash
% IP_IDENTS_SZ
;
483 atomic_t
*p_id
= ip_idents
+ hash
% IP_IDENTS_SZ
;
484 u32 old
= READ_ONCE(*p_tstamp
);
485 u32 now
= (u32
)jiffies
;
488 if (old
!= now
&& cmpxchg(p_tstamp
, old
, now
) == old
)
489 delta
= prandom_u32_max(now
- old
);
491 /* Do not use atomic_add_return() as it makes UBSAN unhappy */
493 old
= (u32
)atomic_read(p_id
);
494 new = old
+ delta
+ segs
;
495 } while (atomic_cmpxchg(p_id
, old
, new) != old
);
499 EXPORT_SYMBOL(ip_idents_reserve
);
501 void __ip_select_ident(struct net
*net
, struct iphdr
*iph
, int segs
)
503 static u32 ip_idents_hashrnd __read_mostly
;
506 net_get_random_once(&ip_idents_hashrnd
, sizeof(ip_idents_hashrnd
));
508 hash
= jhash_3words((__force u32
)iph
->daddr
,
509 (__force u32
)iph
->saddr
,
510 iph
->protocol
^ net_hash_mix(net
),
512 id
= ip_idents_reserve(hash
, segs
);
515 EXPORT_SYMBOL(__ip_select_ident
);
517 static void __build_flow_key(const struct net
*net
, struct flowi4
*fl4
,
518 const struct sock
*sk
,
519 const struct iphdr
*iph
,
521 u8 prot
, u32 mark
, int flow_flags
)
524 const struct inet_sock
*inet
= inet_sk(sk
);
526 oif
= sk
->sk_bound_dev_if
;
528 tos
= RT_CONN_FLAGS(sk
);
529 prot
= inet
->hdrincl
? IPPROTO_RAW
: sk
->sk_protocol
;
531 flowi4_init_output(fl4
, oif
, mark
, tos
,
532 RT_SCOPE_UNIVERSE
, prot
,
534 iph
->daddr
, iph
->saddr
, 0, 0,
535 sock_net_uid(net
, sk
));
538 static void build_skb_flow_key(struct flowi4
*fl4
, const struct sk_buff
*skb
,
539 const struct sock
*sk
)
541 const struct net
*net
= dev_net(skb
->dev
);
542 const struct iphdr
*iph
= ip_hdr(skb
);
543 int oif
= skb
->dev
->ifindex
;
544 u8 tos
= RT_TOS(iph
->tos
);
545 u8 prot
= iph
->protocol
;
546 u32 mark
= skb
->mark
;
548 __build_flow_key(net
, fl4
, sk
, iph
, oif
, tos
, prot
, mark
, 0);
551 static void build_sk_flow_key(struct flowi4
*fl4
, const struct sock
*sk
)
553 const struct inet_sock
*inet
= inet_sk(sk
);
554 const struct ip_options_rcu
*inet_opt
;
555 __be32 daddr
= inet
->inet_daddr
;
558 inet_opt
= rcu_dereference(inet
->inet_opt
);
559 if (inet_opt
&& inet_opt
->opt
.srr
)
560 daddr
= inet_opt
->opt
.faddr
;
561 flowi4_init_output(fl4
, sk
->sk_bound_dev_if
, sk
->sk_mark
,
562 RT_CONN_FLAGS(sk
), RT_SCOPE_UNIVERSE
,
563 inet
->hdrincl
? IPPROTO_RAW
: sk
->sk_protocol
,
564 inet_sk_flowi_flags(sk
),
565 daddr
, inet
->inet_saddr
, 0, 0, sk
->sk_uid
);
569 static void ip_rt_build_flow_key(struct flowi4
*fl4
, const struct sock
*sk
,
570 const struct sk_buff
*skb
)
573 build_skb_flow_key(fl4
, skb
, sk
);
575 build_sk_flow_key(fl4
, sk
);
578 static DEFINE_SPINLOCK(fnhe_lock
);
580 static void fnhe_flush_routes(struct fib_nh_exception
*fnhe
)
584 rt
= rcu_dereference(fnhe
->fnhe_rth_input
);
586 RCU_INIT_POINTER(fnhe
->fnhe_rth_input
, NULL
);
587 dst_dev_put(&rt
->dst
);
588 dst_release(&rt
->dst
);
590 rt
= rcu_dereference(fnhe
->fnhe_rth_output
);
592 RCU_INIT_POINTER(fnhe
->fnhe_rth_output
, NULL
);
593 dst_dev_put(&rt
->dst
);
594 dst_release(&rt
->dst
);
598 static struct fib_nh_exception
*fnhe_oldest(struct fnhe_hash_bucket
*hash
)
600 struct fib_nh_exception
*fnhe
, *oldest
;
602 oldest
= rcu_dereference(hash
->chain
);
603 for (fnhe
= rcu_dereference(oldest
->fnhe_next
); fnhe
;
604 fnhe
= rcu_dereference(fnhe
->fnhe_next
)) {
605 if (time_before(fnhe
->fnhe_stamp
, oldest
->fnhe_stamp
))
608 fnhe_flush_routes(oldest
);
612 static inline u32
fnhe_hashfun(__be32 daddr
)
614 static u32 fnhe_hashrnd __read_mostly
;
617 net_get_random_once(&fnhe_hashrnd
, sizeof(fnhe_hashrnd
));
618 hval
= jhash_1word((__force u32
) daddr
, fnhe_hashrnd
);
619 return hash_32(hval
, FNHE_HASH_SHIFT
);
622 static void fill_route_from_fnhe(struct rtable
*rt
, struct fib_nh_exception
*fnhe
)
624 rt
->rt_pmtu
= fnhe
->fnhe_pmtu
;
625 rt
->rt_mtu_locked
= fnhe
->fnhe_mtu_locked
;
626 rt
->dst
.expires
= fnhe
->fnhe_expires
;
629 rt
->rt_flags
|= RTCF_REDIRECTED
;
630 rt
->rt_gateway
= fnhe
->fnhe_gw
;
631 rt
->rt_uses_gateway
= 1;
635 static void update_or_create_fnhe(struct fib_nh
*nh
, __be32 daddr
, __be32 gw
,
636 u32 pmtu
, bool lock
, unsigned long expires
)
638 struct fnhe_hash_bucket
*hash
;
639 struct fib_nh_exception
*fnhe
;
645 genid
= fnhe_genid(dev_net(nh
->nh_dev
));
646 hval
= fnhe_hashfun(daddr
);
648 spin_lock_bh(&fnhe_lock
);
650 hash
= rcu_dereference(nh
->nh_exceptions
);
652 hash
= kcalloc(FNHE_HASH_SIZE
, sizeof(*hash
), GFP_ATOMIC
);
655 rcu_assign_pointer(nh
->nh_exceptions
, hash
);
661 for (fnhe
= rcu_dereference(hash
->chain
); fnhe
;
662 fnhe
= rcu_dereference(fnhe
->fnhe_next
)) {
663 if (fnhe
->fnhe_daddr
== daddr
)
669 if (fnhe
->fnhe_genid
!= genid
)
670 fnhe
->fnhe_genid
= genid
;
674 fnhe
->fnhe_pmtu
= pmtu
;
675 fnhe
->fnhe_mtu_locked
= lock
;
677 fnhe
->fnhe_expires
= max(1UL, expires
);
678 /* Update all cached dsts too */
679 rt
= rcu_dereference(fnhe
->fnhe_rth_input
);
681 fill_route_from_fnhe(rt
, fnhe
);
682 rt
= rcu_dereference(fnhe
->fnhe_rth_output
);
684 fill_route_from_fnhe(rt
, fnhe
);
686 if (depth
> FNHE_RECLAIM_DEPTH
)
687 fnhe
= fnhe_oldest(hash
);
689 fnhe
= kzalloc(sizeof(*fnhe
), GFP_ATOMIC
);
693 fnhe
->fnhe_next
= hash
->chain
;
694 rcu_assign_pointer(hash
->chain
, fnhe
);
696 fnhe
->fnhe_genid
= genid
;
697 fnhe
->fnhe_daddr
= daddr
;
699 fnhe
->fnhe_pmtu
= pmtu
;
700 fnhe
->fnhe_mtu_locked
= lock
;
701 fnhe
->fnhe_expires
= max(1UL, expires
);
703 /* Exception created; mark the cached routes for the nexthop
704 * stale, so anyone caching it rechecks if this exception
707 rt
= rcu_dereference(nh
->nh_rth_input
);
709 rt
->dst
.obsolete
= DST_OBSOLETE_KILL
;
711 for_each_possible_cpu(i
) {
712 struct rtable __rcu
**prt
;
713 prt
= per_cpu_ptr(nh
->nh_pcpu_rth_output
, i
);
714 rt
= rcu_dereference(*prt
);
716 rt
->dst
.obsolete
= DST_OBSOLETE_KILL
;
720 fnhe
->fnhe_stamp
= jiffies
;
723 spin_unlock_bh(&fnhe_lock
);
726 static void __ip_do_redirect(struct rtable
*rt
, struct sk_buff
*skb
, struct flowi4
*fl4
,
729 __be32 new_gw
= icmp_hdr(skb
)->un
.gateway
;
730 __be32 old_gw
= ip_hdr(skb
)->saddr
;
731 struct net_device
*dev
= skb
->dev
;
732 struct in_device
*in_dev
;
733 struct fib_result res
;
737 switch (icmp_hdr(skb
)->code
& 7) {
739 case ICMP_REDIR_NETTOS
:
740 case ICMP_REDIR_HOST
:
741 case ICMP_REDIR_HOSTTOS
:
748 if (rt
->rt_gateway
!= old_gw
)
751 in_dev
= __in_dev_get_rcu(dev
);
756 if (new_gw
== old_gw
|| !IN_DEV_RX_REDIRECTS(in_dev
) ||
757 ipv4_is_multicast(new_gw
) || ipv4_is_lbcast(new_gw
) ||
758 ipv4_is_zeronet(new_gw
))
759 goto reject_redirect
;
761 if (!IN_DEV_SHARED_MEDIA(in_dev
)) {
762 if (!inet_addr_onlink(in_dev
, new_gw
, old_gw
))
763 goto reject_redirect
;
764 if (IN_DEV_SEC_REDIRECTS(in_dev
) && ip_fib_check_default(new_gw
, dev
))
765 goto reject_redirect
;
767 if (inet_addr_type(net
, new_gw
) != RTN_UNICAST
)
768 goto reject_redirect
;
771 n
= __ipv4_neigh_lookup(rt
->dst
.dev
, new_gw
);
773 n
= neigh_create(&arp_tbl
, &new_gw
, rt
->dst
.dev
);
775 if (!(n
->nud_state
& NUD_VALID
)) {
776 neigh_event_send(n
, NULL
);
778 if (fib_lookup(net
, fl4
, &res
, 0) == 0) {
779 struct fib_nh
*nh
= &FIB_RES_NH(res
);
781 update_or_create_fnhe(nh
, fl4
->daddr
, new_gw
,
783 jiffies
+ ip_rt_gc_timeout
);
786 rt
->dst
.obsolete
= DST_OBSOLETE_KILL
;
787 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE
, n
);
794 #ifdef CONFIG_IP_ROUTE_VERBOSE
795 if (IN_DEV_LOG_MARTIANS(in_dev
)) {
796 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
797 __be32 daddr
= iph
->daddr
;
798 __be32 saddr
= iph
->saddr
;
800 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
801 " Advised path = %pI4 -> %pI4\n",
802 &old_gw
, dev
->name
, &new_gw
,
809 static void ip_do_redirect(struct dst_entry
*dst
, struct sock
*sk
, struct sk_buff
*skb
)
813 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
814 struct net
*net
= dev_net(skb
->dev
);
815 int oif
= skb
->dev
->ifindex
;
816 u8 tos
= RT_TOS(iph
->tos
);
817 u8 prot
= iph
->protocol
;
818 u32 mark
= skb
->mark
;
820 rt
= (struct rtable
*) dst
;
822 __build_flow_key(net
, &fl4
, sk
, iph
, oif
, tos
, prot
, mark
, 0);
823 __ip_do_redirect(rt
, skb
, &fl4
, true);
826 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
)
828 struct rtable
*rt
= (struct rtable
*)dst
;
829 struct dst_entry
*ret
= dst
;
832 if (dst
->obsolete
> 0) {
835 } else if ((rt
->rt_flags
& RTCF_REDIRECTED
) ||
846 * 1. The first ip_rt_redirect_number redirects are sent
847 * with exponential backoff, then we stop sending them at all,
848 * assuming that the host ignores our redirects.
849 * 2. If we did not see packets requiring redirects
850 * during ip_rt_redirect_silence, we assume that the host
851 * forgot redirected route and start to send redirects again.
853 * This algorithm is much cheaper and more intelligent than dumb load limiting
856 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
857 * and "frag. need" (breaks PMTU discovery) in icmp.c.
860 void ip_rt_send_redirect(struct sk_buff
*skb
)
862 struct rtable
*rt
= skb_rtable(skb
);
863 struct in_device
*in_dev
;
864 struct inet_peer
*peer
;
870 in_dev
= __in_dev_get_rcu(rt
->dst
.dev
);
871 if (!in_dev
|| !IN_DEV_TX_REDIRECTS(in_dev
)) {
875 log_martians
= IN_DEV_LOG_MARTIANS(in_dev
);
876 vif
= l3mdev_master_ifindex_rcu(rt
->dst
.dev
);
879 net
= dev_net(rt
->dst
.dev
);
880 peer
= inet_getpeer_v4(net
->ipv4
.peers
, ip_hdr(skb
)->saddr
, vif
, 1);
882 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
,
883 rt_nexthop(rt
, ip_hdr(skb
)->daddr
));
887 /* No redirected packets during ip_rt_redirect_silence;
888 * reset the algorithm.
890 if (time_after(jiffies
, peer
->rate_last
+ ip_rt_redirect_silence
)) {
891 peer
->rate_tokens
= 0;
892 peer
->n_redirects
= 0;
895 /* Too many ignored redirects; do not send anything
896 * set dst.rate_last to the last seen redirected packet.
898 if (peer
->n_redirects
>= ip_rt_redirect_number
) {
899 peer
->rate_last
= jiffies
;
903 /* Check for load limit; set rate_last to the latest sent
906 if (peer
->rate_tokens
== 0 ||
909 (ip_rt_redirect_load
<< peer
->rate_tokens
)))) {
910 __be32 gw
= rt_nexthop(rt
, ip_hdr(skb
)->daddr
);
912 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
, gw
);
913 peer
->rate_last
= jiffies
;
916 #ifdef CONFIG_IP_ROUTE_VERBOSE
918 peer
->rate_tokens
== ip_rt_redirect_number
)
919 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
920 &ip_hdr(skb
)->saddr
, inet_iif(skb
),
921 &ip_hdr(skb
)->daddr
, &gw
);
928 static int ip_error(struct sk_buff
*skb
)
930 struct rtable
*rt
= skb_rtable(skb
);
931 struct net_device
*dev
= skb
->dev
;
932 struct in_device
*in_dev
;
933 struct inet_peer
*peer
;
939 if (netif_is_l3_master(skb
->dev
)) {
940 dev
= __dev_get_by_index(dev_net(skb
->dev
), IPCB(skb
)->iif
);
945 in_dev
= __in_dev_get_rcu(dev
);
947 /* IP on this device is disabled. */
951 net
= dev_net(rt
->dst
.dev
);
952 if (!IN_DEV_FORWARD(in_dev
)) {
953 switch (rt
->dst
.error
) {
955 __IP_INC_STATS(net
, IPSTATS_MIB_INADDRERRORS
);
959 __IP_INC_STATS(net
, IPSTATS_MIB_INNOROUTES
);
965 switch (rt
->dst
.error
) {
970 code
= ICMP_HOST_UNREACH
;
973 code
= ICMP_NET_UNREACH
;
974 __IP_INC_STATS(net
, IPSTATS_MIB_INNOROUTES
);
977 code
= ICMP_PKT_FILTERED
;
981 peer
= inet_getpeer_v4(net
->ipv4
.peers
, ip_hdr(skb
)->saddr
,
982 l3mdev_master_ifindex(skb
->dev
), 1);
987 peer
->rate_tokens
+= now
- peer
->rate_last
;
988 if (peer
->rate_tokens
> ip_rt_error_burst
)
989 peer
->rate_tokens
= ip_rt_error_burst
;
990 peer
->rate_last
= now
;
991 if (peer
->rate_tokens
>= ip_rt_error_cost
)
992 peer
->rate_tokens
-= ip_rt_error_cost
;
998 icmp_send(skb
, ICMP_DEST_UNREACH
, code
, 0);
1000 out
: kfree_skb(skb
);
1004 static void __ip_rt_update_pmtu(struct rtable
*rt
, struct flowi4
*fl4
, u32 mtu
)
1006 struct dst_entry
*dst
= &rt
->dst
;
1007 u32 old_mtu
= ipv4_mtu(dst
);
1008 struct fib_result res
;
1011 if (ip_mtu_locked(dst
))
1017 if (mtu
< ip_rt_min_pmtu
) {
1019 mtu
= min(old_mtu
, ip_rt_min_pmtu
);
1022 if (rt
->rt_pmtu
== mtu
&& !lock
&&
1023 time_before(jiffies
, dst
->expires
- ip_rt_mtu_expires
/ 2))
1027 if (fib_lookup(dev_net(dst
->dev
), fl4
, &res
, 0) == 0) {
1028 struct fib_nh
*nh
= &FIB_RES_NH(res
);
1030 update_or_create_fnhe(nh
, fl4
->daddr
, 0, mtu
, lock
,
1031 jiffies
+ ip_rt_mtu_expires
);
1036 static void ip_rt_update_pmtu(struct dst_entry
*dst
, struct sock
*sk
,
1037 struct sk_buff
*skb
, u32 mtu
)
1039 struct rtable
*rt
= (struct rtable
*) dst
;
1042 ip_rt_build_flow_key(&fl4
, sk
, skb
);
1043 __ip_rt_update_pmtu(rt
, &fl4
, mtu
);
1046 void ipv4_update_pmtu(struct sk_buff
*skb
, struct net
*net
, u32 mtu
,
1047 int oif
, u8 protocol
)
1049 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
1052 u32 mark
= IP4_REPLY_MARK(net
, skb
->mark
);
1054 __build_flow_key(net
, &fl4
, NULL
, iph
, oif
,
1055 RT_TOS(iph
->tos
), protocol
, mark
, 0);
1056 rt
= __ip_route_output_key(net
, &fl4
);
1058 __ip_rt_update_pmtu(rt
, &fl4
, mtu
);
1062 EXPORT_SYMBOL_GPL(ipv4_update_pmtu
);
1064 static void __ipv4_sk_update_pmtu(struct sk_buff
*skb
, struct sock
*sk
, u32 mtu
)
1066 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
1070 __build_flow_key(sock_net(sk
), &fl4
, sk
, iph
, 0, 0, 0, 0, 0);
1072 if (!fl4
.flowi4_mark
)
1073 fl4
.flowi4_mark
= IP4_REPLY_MARK(sock_net(sk
), skb
->mark
);
1075 rt
= __ip_route_output_key(sock_net(sk
), &fl4
);
1077 __ip_rt_update_pmtu(rt
, &fl4
, mtu
);
1082 void ipv4_sk_update_pmtu(struct sk_buff
*skb
, struct sock
*sk
, u32 mtu
)
1084 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
1087 struct dst_entry
*odst
= NULL
;
1089 struct net
*net
= sock_net(sk
);
1093 if (!ip_sk_accept_pmtu(sk
))
1096 odst
= sk_dst_get(sk
);
1098 if (sock_owned_by_user(sk
) || !odst
) {
1099 __ipv4_sk_update_pmtu(skb
, sk
, mtu
);
1103 __build_flow_key(net
, &fl4
, sk
, iph
, 0, 0, 0, 0, 0);
1105 rt
= (struct rtable
*)odst
;
1106 if (odst
->obsolete
&& !odst
->ops
->check(odst
, 0)) {
1107 rt
= ip_route_output_flow(sock_net(sk
), &fl4
, sk
);
1114 __ip_rt_update_pmtu((struct rtable
*) xfrm_dst_path(&rt
->dst
), &fl4
, mtu
);
1116 if (!dst_check(&rt
->dst
, 0)) {
1118 dst_release(&rt
->dst
);
1120 rt
= ip_route_output_flow(sock_net(sk
), &fl4
, sk
);
1128 sk_dst_set(sk
, &rt
->dst
);
1134 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu
);
1136 void ipv4_redirect(struct sk_buff
*skb
, struct net
*net
,
1137 int oif
, u8 protocol
)
1139 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
1143 __build_flow_key(net
, &fl4
, NULL
, iph
, oif
,
1144 RT_TOS(iph
->tos
), protocol
, 0, 0);
1145 rt
= __ip_route_output_key(net
, &fl4
);
1147 __ip_do_redirect(rt
, skb
, &fl4
, false);
1151 EXPORT_SYMBOL_GPL(ipv4_redirect
);
1153 void ipv4_sk_redirect(struct sk_buff
*skb
, struct sock
*sk
)
1155 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
1158 struct net
*net
= sock_net(sk
);
1160 __build_flow_key(net
, &fl4
, sk
, iph
, 0, 0, 0, 0, 0);
1161 rt
= __ip_route_output_key(net
, &fl4
);
1163 __ip_do_redirect(rt
, skb
, &fl4
, false);
1167 EXPORT_SYMBOL_GPL(ipv4_sk_redirect
);
1169 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
)
1171 struct rtable
*rt
= (struct rtable
*) dst
;
1173 /* All IPV4 dsts are created with ->obsolete set to the value
1174 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1175 * into this function always.
1177 * When a PMTU/redirect information update invalidates a route,
1178 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1179 * DST_OBSOLETE_DEAD by dst_free().
1181 if (dst
->obsolete
!= DST_OBSOLETE_FORCE_CHK
|| rt_is_expired(rt
))
1186 static void ipv4_link_failure(struct sk_buff
*skb
)
1190 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_HOST_UNREACH
, 0);
1192 rt
= skb_rtable(skb
);
1194 dst_set_expires(&rt
->dst
, 0);
1197 static int ip_rt_bug(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
1199 pr_debug("%s: %pI4 -> %pI4, %s\n",
1200 __func__
, &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1201 skb
->dev
? skb
->dev
->name
: "?");
1208 We do not cache source address of outgoing interface,
1209 because it is used only by IP RR, TS and SRR options,
1210 so that it out of fast path.
1212 BTW remember: "addr" is allowed to be not aligned
1216 void ip_rt_get_source(u8
*addr
, struct sk_buff
*skb
, struct rtable
*rt
)
1220 if (rt_is_output_route(rt
))
1221 src
= ip_hdr(skb
)->saddr
;
1223 struct fib_result res
;
1224 struct iphdr
*iph
= ip_hdr(skb
);
1225 struct flowi4 fl4
= {
1226 .daddr
= iph
->daddr
,
1227 .saddr
= iph
->saddr
,
1228 .flowi4_tos
= RT_TOS(iph
->tos
),
1229 .flowi4_oif
= rt
->dst
.dev
->ifindex
,
1230 .flowi4_iif
= skb
->dev
->ifindex
,
1231 .flowi4_mark
= skb
->mark
,
1235 if (fib_lookup(dev_net(rt
->dst
.dev
), &fl4
, &res
, 0) == 0)
1236 src
= FIB_RES_PREFSRC(dev_net(rt
->dst
.dev
), res
);
1238 src
= inet_select_addr(rt
->dst
.dev
,
1239 rt_nexthop(rt
, iph
->daddr
),
1243 memcpy(addr
, &src
, 4);
1246 #ifdef CONFIG_IP_ROUTE_CLASSID
1247 static void set_class_tag(struct rtable
*rt
, u32 tag
)
1249 if (!(rt
->dst
.tclassid
& 0xFFFF))
1250 rt
->dst
.tclassid
|= tag
& 0xFFFF;
1251 if (!(rt
->dst
.tclassid
& 0xFFFF0000))
1252 rt
->dst
.tclassid
|= tag
& 0xFFFF0000;
1256 static unsigned int ipv4_default_advmss(const struct dst_entry
*dst
)
1258 unsigned int header_size
= sizeof(struct tcphdr
) + sizeof(struct iphdr
);
1259 unsigned int advmss
= max_t(unsigned int, ipv4_mtu(dst
) - header_size
,
1262 return min(advmss
, IPV4_MAX_PMTU
- header_size
);
1265 static unsigned int ipv4_mtu(const struct dst_entry
*dst
)
1267 const struct rtable
*rt
= (const struct rtable
*) dst
;
1268 unsigned int mtu
= rt
->rt_pmtu
;
1270 if (!mtu
|| time_after_eq(jiffies
, rt
->dst
.expires
))
1271 mtu
= dst_metric_raw(dst
, RTAX_MTU
);
1276 mtu
= READ_ONCE(dst
->dev
->mtu
);
1278 if (unlikely(ip_mtu_locked(dst
))) {
1279 if (rt
->rt_uses_gateway
&& mtu
> 576)
1283 mtu
= min_t(unsigned int, mtu
, IP_MAX_MTU
);
1285 return mtu
- lwtunnel_headroom(dst
->lwtstate
, mtu
);
1288 static void ip_del_fnhe(struct fib_nh
*nh
, __be32 daddr
)
1290 struct fnhe_hash_bucket
*hash
;
1291 struct fib_nh_exception
*fnhe
, __rcu
**fnhe_p
;
1292 u32 hval
= fnhe_hashfun(daddr
);
1294 spin_lock_bh(&fnhe_lock
);
1296 hash
= rcu_dereference_protected(nh
->nh_exceptions
,
1297 lockdep_is_held(&fnhe_lock
));
1300 fnhe_p
= &hash
->chain
;
1301 fnhe
= rcu_dereference_protected(*fnhe_p
, lockdep_is_held(&fnhe_lock
));
1303 if (fnhe
->fnhe_daddr
== daddr
) {
1304 rcu_assign_pointer(*fnhe_p
, rcu_dereference_protected(
1305 fnhe
->fnhe_next
, lockdep_is_held(&fnhe_lock
)));
1306 fnhe_flush_routes(fnhe
);
1307 kfree_rcu(fnhe
, rcu
);
1310 fnhe_p
= &fnhe
->fnhe_next
;
1311 fnhe
= rcu_dereference_protected(fnhe
->fnhe_next
,
1312 lockdep_is_held(&fnhe_lock
));
1315 spin_unlock_bh(&fnhe_lock
);
1318 static struct fib_nh_exception
*find_exception(struct fib_nh
*nh
, __be32 daddr
)
1320 struct fnhe_hash_bucket
*hash
= rcu_dereference(nh
->nh_exceptions
);
1321 struct fib_nh_exception
*fnhe
;
1327 hval
= fnhe_hashfun(daddr
);
1329 for (fnhe
= rcu_dereference(hash
[hval
].chain
); fnhe
;
1330 fnhe
= rcu_dereference(fnhe
->fnhe_next
)) {
1331 if (fnhe
->fnhe_daddr
== daddr
) {
1332 if (fnhe
->fnhe_expires
&&
1333 time_after(jiffies
, fnhe
->fnhe_expires
)) {
1334 ip_del_fnhe(nh
, daddr
);
1344 * 1. mtu on route is locked - use it
1345 * 2. mtu from nexthop exception
1346 * 3. mtu from egress device
1349 u32
ip_mtu_from_fib_result(struct fib_result
*res
, __be32 daddr
)
1351 struct fib_info
*fi
= res
->fi
;
1352 struct fib_nh
*nh
= &fi
->fib_nh
[res
->nh_sel
];
1353 struct net_device
*dev
= nh
->nh_dev
;
1356 if (dev_net(dev
)->ipv4
.sysctl_ip_fwd_use_pmtu
||
1357 fi
->fib_metrics
->metrics
[RTAX_LOCK
- 1] & (1 << RTAX_MTU
))
1361 struct fib_nh_exception
*fnhe
;
1363 fnhe
= find_exception(nh
, daddr
);
1364 if (fnhe
&& !time_after_eq(jiffies
, fnhe
->fnhe_expires
))
1365 mtu
= fnhe
->fnhe_pmtu
;
1369 mtu
= min(READ_ONCE(dev
->mtu
), IP_MAX_MTU
);
1371 return mtu
- lwtunnel_headroom(nh
->nh_lwtstate
, mtu
);
1374 static bool rt_bind_exception(struct rtable
*rt
, struct fib_nh_exception
*fnhe
,
1375 __be32 daddr
, const bool do_cache
)
1379 spin_lock_bh(&fnhe_lock
);
1381 if (daddr
== fnhe
->fnhe_daddr
) {
1382 struct rtable __rcu
**porig
;
1383 struct rtable
*orig
;
1384 int genid
= fnhe_genid(dev_net(rt
->dst
.dev
));
1386 if (rt_is_input_route(rt
))
1387 porig
= &fnhe
->fnhe_rth_input
;
1389 porig
= &fnhe
->fnhe_rth_output
;
1390 orig
= rcu_dereference(*porig
);
1392 if (fnhe
->fnhe_genid
!= genid
) {
1393 fnhe
->fnhe_genid
= genid
;
1395 fnhe
->fnhe_pmtu
= 0;
1396 fnhe
->fnhe_expires
= 0;
1397 fnhe
->fnhe_mtu_locked
= false;
1398 fnhe_flush_routes(fnhe
);
1401 fill_route_from_fnhe(rt
, fnhe
);
1402 if (!rt
->rt_gateway
)
1403 rt
->rt_gateway
= daddr
;
1407 rcu_assign_pointer(*porig
, rt
);
1409 dst_dev_put(&orig
->dst
);
1410 dst_release(&orig
->dst
);
1415 fnhe
->fnhe_stamp
= jiffies
;
1417 spin_unlock_bh(&fnhe_lock
);
1422 static bool rt_cache_route(struct fib_nh
*nh
, struct rtable
*rt
)
1424 struct rtable
*orig
, *prev
, **p
;
1427 if (rt_is_input_route(rt
)) {
1428 p
= (struct rtable
**)&nh
->nh_rth_input
;
1430 p
= (struct rtable
**)raw_cpu_ptr(nh
->nh_pcpu_rth_output
);
1434 /* hold dst before doing cmpxchg() to avoid race condition
1438 prev
= cmpxchg(p
, orig
, rt
);
1441 dst_dev_put(&orig
->dst
);
1442 dst_release(&orig
->dst
);
1445 dst_release(&rt
->dst
);
1452 struct uncached_list
{
1454 struct list_head head
;
1457 static DEFINE_PER_CPU_ALIGNED(struct uncached_list
, rt_uncached_list
);
1459 void rt_add_uncached_list(struct rtable
*rt
)
1461 struct uncached_list
*ul
= raw_cpu_ptr(&rt_uncached_list
);
1463 rt
->rt_uncached_list
= ul
;
1465 spin_lock_bh(&ul
->lock
);
1466 list_add_tail(&rt
->rt_uncached
, &ul
->head
);
1467 spin_unlock_bh(&ul
->lock
);
1470 void rt_del_uncached_list(struct rtable
*rt
)
1472 if (!list_empty(&rt
->rt_uncached
)) {
1473 struct uncached_list
*ul
= rt
->rt_uncached_list
;
1475 spin_lock_bh(&ul
->lock
);
1476 list_del(&rt
->rt_uncached
);
1477 spin_unlock_bh(&ul
->lock
);
1481 static void ipv4_dst_destroy(struct dst_entry
*dst
)
1483 struct rtable
*rt
= (struct rtable
*)dst
;
1485 ip_dst_metrics_put(dst
);
1486 rt_del_uncached_list(rt
);
1489 void rt_flush_dev(struct net_device
*dev
)
1491 struct net
*net
= dev_net(dev
);
1495 for_each_possible_cpu(cpu
) {
1496 struct uncached_list
*ul
= &per_cpu(rt_uncached_list
, cpu
);
1498 spin_lock_bh(&ul
->lock
);
1499 list_for_each_entry(rt
, &ul
->head
, rt_uncached
) {
1500 if (rt
->dst
.dev
!= dev
)
1502 rt
->dst
.dev
= net
->loopback_dev
;
1503 dev_hold(rt
->dst
.dev
);
1506 spin_unlock_bh(&ul
->lock
);
1510 static bool rt_cache_valid(const struct rtable
*rt
)
1513 rt
->dst
.obsolete
== DST_OBSOLETE_FORCE_CHK
&&
1517 static void rt_set_nexthop(struct rtable
*rt
, __be32 daddr
,
1518 const struct fib_result
*res
,
1519 struct fib_nh_exception
*fnhe
,
1520 struct fib_info
*fi
, u16 type
, u32 itag
,
1521 const bool do_cache
)
1523 bool cached
= false;
1526 struct fib_nh
*nh
= &FIB_RES_NH(*res
);
1528 if (nh
->nh_gw
&& nh
->nh_scope
== RT_SCOPE_LINK
) {
1529 rt
->rt_gateway
= nh
->nh_gw
;
1530 rt
->rt_uses_gateway
= 1;
1532 ip_dst_init_metrics(&rt
->dst
, fi
->fib_metrics
);
1534 #ifdef CONFIG_IP_ROUTE_CLASSID
1535 rt
->dst
.tclassid
= nh
->nh_tclassid
;
1537 rt
->dst
.lwtstate
= lwtstate_get(nh
->nh_lwtstate
);
1539 cached
= rt_bind_exception(rt
, fnhe
, daddr
, do_cache
);
1541 cached
= rt_cache_route(nh
, rt
);
1542 if (unlikely(!cached
)) {
1543 /* Routes we intend to cache in nexthop exception or
1544 * FIB nexthop have the DST_NOCACHE bit clear.
1545 * However, if we are unsuccessful at storing this
1546 * route into the cache we really need to set it.
1548 if (!rt
->rt_gateway
)
1549 rt
->rt_gateway
= daddr
;
1550 rt_add_uncached_list(rt
);
1553 rt_add_uncached_list(rt
);
1555 #ifdef CONFIG_IP_ROUTE_CLASSID
1556 #ifdef CONFIG_IP_MULTIPLE_TABLES
1557 set_class_tag(rt
, res
->tclassid
);
1559 set_class_tag(rt
, itag
);
1563 struct rtable
*rt_dst_alloc(struct net_device
*dev
,
1564 unsigned int flags
, u16 type
,
1565 bool nopolicy
, bool noxfrm
, bool will_cache
)
1569 rt
= dst_alloc(&ipv4_dst_ops
, dev
, 1, DST_OBSOLETE_FORCE_CHK
,
1570 (will_cache
? 0 : DST_HOST
) |
1571 (nopolicy
? DST_NOPOLICY
: 0) |
1572 (noxfrm
? DST_NOXFRM
: 0));
1575 rt
->rt_genid
= rt_genid_ipv4(dev_net(dev
));
1576 rt
->rt_flags
= flags
;
1578 rt
->rt_is_input
= 0;
1581 rt
->rt_mtu_locked
= 0;
1583 rt
->rt_uses_gateway
= 0;
1584 INIT_LIST_HEAD(&rt
->rt_uncached
);
1586 rt
->dst
.output
= ip_output
;
1587 if (flags
& RTCF_LOCAL
)
1588 rt
->dst
.input
= ip_local_deliver
;
1593 EXPORT_SYMBOL(rt_dst_alloc
);
1595 /* called in rcu_read_lock() section */
1596 int ip_mc_validate_source(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
1597 u8 tos
, struct net_device
*dev
,
1598 struct in_device
*in_dev
, u32
*itag
)
1602 /* Primary sanity checks. */
1606 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
1607 skb
->protocol
!= htons(ETH_P_IP
))
1610 if (ipv4_is_loopback(saddr
) && !IN_DEV_ROUTE_LOCALNET(in_dev
))
1613 if (ipv4_is_zeronet(saddr
)) {
1614 if (!ipv4_is_local_multicast(daddr
) &&
1615 ip_hdr(skb
)->protocol
!= IPPROTO_IGMP
)
1618 err
= fib_validate_source(skb
, saddr
, 0, tos
, 0, dev
,
1626 /* called in rcu_read_lock() section */
1627 static int ip_route_input_mc(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
1628 u8 tos
, struct net_device
*dev
, int our
)
1630 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
1631 unsigned int flags
= RTCF_MULTICAST
;
1636 err
= ip_mc_validate_source(skb
, daddr
, saddr
, tos
, dev
, in_dev
, &itag
);
1641 flags
|= RTCF_LOCAL
;
1643 rth
= rt_dst_alloc(dev_net(dev
)->loopback_dev
, flags
, RTN_MULTICAST
,
1644 IN_DEV_CONF_GET(in_dev
, NOPOLICY
), false, false);
1648 #ifdef CONFIG_IP_ROUTE_CLASSID
1649 rth
->dst
.tclassid
= itag
;
1651 rth
->dst
.output
= ip_rt_bug
;
1652 rth
->rt_is_input
= 1;
1654 #ifdef CONFIG_IP_MROUTE
1655 if (!ipv4_is_local_multicast(daddr
) && IN_DEV_MFORWARD(in_dev
))
1656 rth
->dst
.input
= ip_mr_input
;
1658 RT_CACHE_STAT_INC(in_slow_mc
);
1660 skb_dst_set(skb
, &rth
->dst
);
1665 static void ip_handle_martian_source(struct net_device
*dev
,
1666 struct in_device
*in_dev
,
1667 struct sk_buff
*skb
,
1671 RT_CACHE_STAT_INC(in_martian_src
);
1672 #ifdef CONFIG_IP_ROUTE_VERBOSE
1673 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit()) {
1675 * RFC1812 recommendation, if source is martian,
1676 * the only hint is MAC header.
1678 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1679 &daddr
, &saddr
, dev
->name
);
1680 if (dev
->hard_header_len
&& skb_mac_header_was_set(skb
)) {
1681 print_hex_dump(KERN_WARNING
, "ll header: ",
1682 DUMP_PREFIX_OFFSET
, 16, 1,
1683 skb_mac_header(skb
),
1684 dev
->hard_header_len
, false);
1690 /* called in rcu_read_lock() section */
1691 static int __mkroute_input(struct sk_buff
*skb
,
1692 const struct fib_result
*res
,
1693 struct in_device
*in_dev
,
1694 __be32 daddr
, __be32 saddr
, u32 tos
)
1696 struct fib_nh_exception
*fnhe
;
1699 struct in_device
*out_dev
;
1703 /* get a working reference to the output device */
1704 out_dev
= __in_dev_get_rcu(FIB_RES_DEV(*res
));
1706 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1710 err
= fib_validate_source(skb
, saddr
, daddr
, tos
, FIB_RES_OIF(*res
),
1711 in_dev
->dev
, in_dev
, &itag
);
1713 ip_handle_martian_source(in_dev
->dev
, in_dev
, skb
, daddr
,
1719 do_cache
= res
->fi
&& !itag
;
1720 if (out_dev
== in_dev
&& err
&& IN_DEV_TX_REDIRECTS(out_dev
) &&
1721 skb
->protocol
== htons(ETH_P_IP
) &&
1722 (IN_DEV_SHARED_MEDIA(out_dev
) ||
1723 inet_addr_onlink(out_dev
, saddr
, FIB_RES_GW(*res
))))
1724 IPCB(skb
)->flags
|= IPSKB_DOREDIRECT
;
1726 if (skb
->protocol
!= htons(ETH_P_IP
)) {
1727 /* Not IP (i.e. ARP). Do not create route, if it is
1728 * invalid for proxy arp. DNAT routes are always valid.
1730 * Proxy arp feature have been extended to allow, ARP
1731 * replies back to the same interface, to support
1732 * Private VLAN switch technologies. See arp.c.
1734 if (out_dev
== in_dev
&&
1735 IN_DEV_PROXY_ARP_PVLAN(in_dev
) == 0) {
1741 fnhe
= find_exception(&FIB_RES_NH(*res
), daddr
);
1744 rth
= rcu_dereference(fnhe
->fnhe_rth_input
);
1746 rth
= rcu_dereference(FIB_RES_NH(*res
).nh_rth_input
);
1747 if (rt_cache_valid(rth
)) {
1748 skb_dst_set_noref(skb
, &rth
->dst
);
1753 rth
= rt_dst_alloc(out_dev
->dev
, 0, res
->type
,
1754 IN_DEV_CONF_GET(in_dev
, NOPOLICY
),
1755 IN_DEV_CONF_GET(out_dev
, NOXFRM
), do_cache
);
1761 rth
->rt_is_input
= 1;
1762 RT_CACHE_STAT_INC(in_slow_tot
);
1764 rth
->dst
.input
= ip_forward
;
1766 rt_set_nexthop(rth
, daddr
, res
, fnhe
, res
->fi
, res
->type
, itag
,
1768 lwtunnel_set_redirect(&rth
->dst
);
1769 skb_dst_set(skb
, &rth
->dst
);
1776 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1777 /* To make ICMP packets follow the right flow, the multipath hash is
1778 * calculated from the inner IP addresses.
1780 static void ip_multipath_l3_keys(const struct sk_buff
*skb
,
1781 struct flow_keys
*hash_keys
)
1783 const struct iphdr
*outer_iph
= ip_hdr(skb
);
1784 const struct iphdr
*key_iph
= outer_iph
;
1785 const struct iphdr
*inner_iph
;
1786 const struct icmphdr
*icmph
;
1787 struct iphdr _inner_iph
;
1788 struct icmphdr _icmph
;
1790 if (likely(outer_iph
->protocol
!= IPPROTO_ICMP
))
1793 if (unlikely((outer_iph
->frag_off
& htons(IP_OFFSET
)) != 0))
1796 icmph
= skb_header_pointer(skb
, outer_iph
->ihl
* 4, sizeof(_icmph
),
1801 if (icmph
->type
!= ICMP_DEST_UNREACH
&&
1802 icmph
->type
!= ICMP_REDIRECT
&&
1803 icmph
->type
!= ICMP_TIME_EXCEEDED
&&
1804 icmph
->type
!= ICMP_PARAMETERPROB
)
1807 inner_iph
= skb_header_pointer(skb
,
1808 outer_iph
->ihl
* 4 + sizeof(_icmph
),
1809 sizeof(_inner_iph
), &_inner_iph
);
1813 key_iph
= inner_iph
;
1815 hash_keys
->addrs
.v4addrs
.src
= key_iph
->saddr
;
1816 hash_keys
->addrs
.v4addrs
.dst
= key_iph
->daddr
;
1819 /* if skb is set it will be used and fl4 can be NULL */
1820 int fib_multipath_hash(const struct net
*net
, const struct flowi4
*fl4
,
1821 const struct sk_buff
*skb
, struct flow_keys
*flkeys
)
1823 struct flow_keys hash_keys
;
1826 switch (net
->ipv4
.sysctl_fib_multipath_hash_policy
) {
1828 memset(&hash_keys
, 0, sizeof(hash_keys
));
1829 hash_keys
.control
.addr_type
= FLOW_DISSECTOR_KEY_IPV4_ADDRS
;
1831 ip_multipath_l3_keys(skb
, &hash_keys
);
1833 hash_keys
.addrs
.v4addrs
.src
= fl4
->saddr
;
1834 hash_keys
.addrs
.v4addrs
.dst
= fl4
->daddr
;
1838 /* skb is currently provided only when forwarding */
1840 unsigned int flag
= FLOW_DISSECTOR_F_STOP_AT_ENCAP
;
1841 struct flow_keys keys
;
1843 /* short-circuit if we already have L4 hash present */
1845 return skb_get_hash_raw(skb
) >> 1;
1847 memset(&hash_keys
, 0, sizeof(hash_keys
));
1850 skb_flow_dissect_flow_keys(skb
, &keys
, flag
);
1854 hash_keys
.control
.addr_type
= FLOW_DISSECTOR_KEY_IPV4_ADDRS
;
1855 hash_keys
.addrs
.v4addrs
.src
= flkeys
->addrs
.v4addrs
.src
;
1856 hash_keys
.addrs
.v4addrs
.dst
= flkeys
->addrs
.v4addrs
.dst
;
1857 hash_keys
.ports
.src
= flkeys
->ports
.src
;
1858 hash_keys
.ports
.dst
= flkeys
->ports
.dst
;
1859 hash_keys
.basic
.ip_proto
= flkeys
->basic
.ip_proto
;
1861 memset(&hash_keys
, 0, sizeof(hash_keys
));
1862 hash_keys
.control
.addr_type
= FLOW_DISSECTOR_KEY_IPV4_ADDRS
;
1863 hash_keys
.addrs
.v4addrs
.src
= fl4
->saddr
;
1864 hash_keys
.addrs
.v4addrs
.dst
= fl4
->daddr
;
1865 hash_keys
.ports
.src
= fl4
->fl4_sport
;
1866 hash_keys
.ports
.dst
= fl4
->fl4_dport
;
1867 hash_keys
.basic
.ip_proto
= fl4
->flowi4_proto
;
1871 mhash
= flow_hash_from_keys(&hash_keys
);
1875 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1877 static int ip_mkroute_input(struct sk_buff
*skb
,
1878 struct fib_result
*res
,
1879 struct in_device
*in_dev
,
1880 __be32 daddr
, __be32 saddr
, u32 tos
,
1881 struct flow_keys
*hkeys
)
1883 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1884 if (res
->fi
&& res
->fi
->fib_nhs
> 1) {
1885 int h
= fib_multipath_hash(res
->fi
->fib_net
, NULL
, skb
, hkeys
);
1887 fib_select_multipath(res
, h
);
1891 /* create a routing cache entry */
1892 return __mkroute_input(skb
, res
, in_dev
, daddr
, saddr
, tos
);
1896 * NOTE. We drop all the packets that has local source
1897 * addresses, because every properly looped back packet
1898 * must have correct destination already attached by output routine.
1900 * Such approach solves two big problems:
1901 * 1. Not simplex devices are handled properly.
1902 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1903 * called with rcu_read_lock()
1906 static int ip_route_input_slow(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
1907 u8 tos
, struct net_device
*dev
,
1908 struct fib_result
*res
)
1910 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
1911 struct flow_keys
*flkeys
= NULL
, _flkeys
;
1912 struct net
*net
= dev_net(dev
);
1913 struct ip_tunnel_info
*tun_info
;
1915 unsigned int flags
= 0;
1921 /* IP on this device is disabled. */
1926 /* Check for the most weird martians, which can be not detected
1930 tun_info
= skb_tunnel_info(skb
);
1931 if (tun_info
&& !(tun_info
->mode
& IP_TUNNEL_INFO_TX
))
1932 fl4
.flowi4_tun_key
.tun_id
= tun_info
->key
.tun_id
;
1934 fl4
.flowi4_tun_key
.tun_id
= 0;
1937 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
))
1938 goto martian_source
;
1942 if (ipv4_is_lbcast(daddr
) || (saddr
== 0 && daddr
== 0))
1945 /* Accept zero addresses only to limited broadcast;
1946 * I even do not know to fix it or not. Waiting for complains :-)
1948 if (ipv4_is_zeronet(saddr
))
1949 goto martian_source
;
1951 if (ipv4_is_zeronet(daddr
))
1952 goto martian_destination
;
1954 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1955 * and call it once if daddr or/and saddr are loopback addresses
1957 if (ipv4_is_loopback(daddr
)) {
1958 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev
, net
))
1959 goto martian_destination
;
1960 } else if (ipv4_is_loopback(saddr
)) {
1961 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev
, net
))
1962 goto martian_source
;
1966 * Now we are ready to route packet.
1969 fl4
.flowi4_iif
= dev
->ifindex
;
1970 fl4
.flowi4_mark
= skb
->mark
;
1971 fl4
.flowi4_tos
= tos
;
1972 fl4
.flowi4_scope
= RT_SCOPE_UNIVERSE
;
1973 fl4
.flowi4_flags
= 0;
1976 fl4
.flowi4_uid
= sock_net_uid(net
, NULL
);
1978 if (fib4_rules_early_flow_dissect(net
, skb
, &fl4
, &_flkeys
)) {
1981 fl4
.flowi4_proto
= 0;
1986 err
= fib_lookup(net
, &fl4
, res
, 0);
1988 if (!IN_DEV_FORWARD(in_dev
))
1989 err
= -EHOSTUNREACH
;
1993 if (res
->type
== RTN_BROADCAST
) {
1994 if (IN_DEV_BFORWARD(in_dev
))
1999 if (res
->type
== RTN_LOCAL
) {
2000 err
= fib_validate_source(skb
, saddr
, daddr
, tos
,
2001 0, dev
, in_dev
, &itag
);
2003 goto martian_source
;
2007 if (!IN_DEV_FORWARD(in_dev
)) {
2008 err
= -EHOSTUNREACH
;
2011 if (res
->type
!= RTN_UNICAST
)
2012 goto martian_destination
;
2015 err
= ip_mkroute_input(skb
, res
, in_dev
, daddr
, saddr
, tos
, flkeys
);
2019 if (skb
->protocol
!= htons(ETH_P_IP
))
2022 if (!ipv4_is_zeronet(saddr
)) {
2023 err
= fib_validate_source(skb
, saddr
, 0, tos
, 0, dev
,
2026 goto martian_source
;
2028 flags
|= RTCF_BROADCAST
;
2029 res
->type
= RTN_BROADCAST
;
2030 RT_CACHE_STAT_INC(in_brd
);
2036 rth
= rcu_dereference(FIB_RES_NH(*res
).nh_rth_input
);
2037 if (rt_cache_valid(rth
)) {
2038 skb_dst_set_noref(skb
, &rth
->dst
);
2046 rth
= rt_dst_alloc(l3mdev_master_dev_rcu(dev
) ? : net
->loopback_dev
,
2047 flags
| RTCF_LOCAL
, res
->type
,
2048 IN_DEV_CONF_GET(in_dev
, NOPOLICY
), false, do_cache
);
2052 rth
->dst
.output
= ip_rt_bug
;
2053 #ifdef CONFIG_IP_ROUTE_CLASSID
2054 rth
->dst
.tclassid
= itag
;
2056 rth
->rt_is_input
= 1;
2058 RT_CACHE_STAT_INC(in_slow_tot
);
2059 if (res
->type
== RTN_UNREACHABLE
) {
2060 rth
->dst
.input
= ip_error
;
2061 rth
->dst
.error
= -err
;
2062 rth
->rt_flags
&= ~RTCF_LOCAL
;
2066 struct fib_nh
*nh
= &FIB_RES_NH(*res
);
2068 rth
->dst
.lwtstate
= lwtstate_get(nh
->nh_lwtstate
);
2069 if (lwtunnel_input_redirect(rth
->dst
.lwtstate
)) {
2070 WARN_ON(rth
->dst
.input
== lwtunnel_input
);
2071 rth
->dst
.lwtstate
->orig_input
= rth
->dst
.input
;
2072 rth
->dst
.input
= lwtunnel_input
;
2075 if (unlikely(!rt_cache_route(nh
, rth
)))
2076 rt_add_uncached_list(rth
);
2078 skb_dst_set(skb
, &rth
->dst
);
2083 RT_CACHE_STAT_INC(in_no_route
);
2084 res
->type
= RTN_UNREACHABLE
;
2090 * Do not cache martian addresses: they should be logged (RFC1812)
2092 martian_destination
:
2093 RT_CACHE_STAT_INC(in_martian_dst
);
2094 #ifdef CONFIG_IP_ROUTE_VERBOSE
2095 if (IN_DEV_LOG_MARTIANS(in_dev
))
2096 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2097 &daddr
, &saddr
, dev
->name
);
2109 ip_handle_martian_source(dev
, in_dev
, skb
, daddr
, saddr
);
2113 int ip_route_input_noref(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2114 u8 tos
, struct net_device
*dev
)
2116 struct fib_result res
;
2119 tos
&= IPTOS_RT_MASK
;
2121 err
= ip_route_input_rcu(skb
, daddr
, saddr
, tos
, dev
, &res
);
2126 EXPORT_SYMBOL(ip_route_input_noref
);
2128 /* called with rcu_read_lock held */
2129 int ip_route_input_rcu(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2130 u8 tos
, struct net_device
*dev
, struct fib_result
*res
)
2132 /* Multicast recognition logic is moved from route cache to here.
2133 The problem was that too many Ethernet cards have broken/missing
2134 hardware multicast filters :-( As result the host on multicasting
2135 network acquires a lot of useless route cache entries, sort of
2136 SDR messages from all the world. Now we try to get rid of them.
2137 Really, provided software IP multicast filter is organized
2138 reasonably (at least, hashed), it does not result in a slowdown
2139 comparing with route cache reject entries.
2140 Note, that multicast routers are not affected, because
2141 route cache entry is created eventually.
2143 if (ipv4_is_multicast(daddr
)) {
2144 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
2149 our
= ip_check_mc_rcu(in_dev
, daddr
, saddr
,
2150 ip_hdr(skb
)->protocol
);
2152 /* check l3 master if no match yet */
2153 if ((!in_dev
|| !our
) && netif_is_l3_slave(dev
)) {
2154 struct in_device
*l3_in_dev
;
2156 l3_in_dev
= __in_dev_get_rcu(skb
->dev
);
2158 our
= ip_check_mc_rcu(l3_in_dev
, daddr
, saddr
,
2159 ip_hdr(skb
)->protocol
);
2163 #ifdef CONFIG_IP_MROUTE
2165 (!ipv4_is_local_multicast(daddr
) &&
2166 IN_DEV_MFORWARD(in_dev
))
2169 err
= ip_route_input_mc(skb
, daddr
, saddr
,
2175 return ip_route_input_slow(skb
, daddr
, saddr
, tos
, dev
, res
);
2178 /* called with rcu_read_lock() */
2179 static struct rtable
*__mkroute_output(const struct fib_result
*res
,
2180 const struct flowi4
*fl4
, int orig_oif
,
2181 struct net_device
*dev_out
,
2184 struct fib_info
*fi
= res
->fi
;
2185 struct fib_nh_exception
*fnhe
;
2186 struct in_device
*in_dev
;
2187 u16 type
= res
->type
;
2191 in_dev
= __in_dev_get_rcu(dev_out
);
2193 return ERR_PTR(-EINVAL
);
2195 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev
)))
2196 if (ipv4_is_loopback(fl4
->saddr
) &&
2197 !(dev_out
->flags
& IFF_LOOPBACK
) &&
2198 !netif_is_l3_master(dev_out
))
2199 return ERR_PTR(-EINVAL
);
2201 if (ipv4_is_lbcast(fl4
->daddr
))
2202 type
= RTN_BROADCAST
;
2203 else if (ipv4_is_multicast(fl4
->daddr
))
2204 type
= RTN_MULTICAST
;
2205 else if (ipv4_is_zeronet(fl4
->daddr
))
2206 return ERR_PTR(-EINVAL
);
2208 if (dev_out
->flags
& IFF_LOOPBACK
)
2209 flags
|= RTCF_LOCAL
;
2212 if (type
== RTN_BROADCAST
) {
2213 flags
|= RTCF_BROADCAST
| RTCF_LOCAL
;
2215 } else if (type
== RTN_MULTICAST
) {
2216 flags
|= RTCF_MULTICAST
| RTCF_LOCAL
;
2217 if (!ip_check_mc_rcu(in_dev
, fl4
->daddr
, fl4
->saddr
,
2219 flags
&= ~RTCF_LOCAL
;
2222 /* If multicast route do not exist use
2223 * default one, but do not gateway in this case.
2226 if (fi
&& res
->prefixlen
< 4)
2228 } else if ((type
== RTN_LOCAL
) && (orig_oif
!= 0) &&
2229 (orig_oif
!= dev_out
->ifindex
)) {
2230 /* For local routes that require a particular output interface
2231 * we do not want to cache the result. Caching the result
2232 * causes incorrect behaviour when there are multiple source
2233 * addresses on the interface, the end result being that if the
2234 * intended recipient is waiting on that interface for the
2235 * packet he won't receive it because it will be delivered on
2236 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2237 * be set to the loopback interface as well.
2243 do_cache
&= fi
!= NULL
;
2245 struct rtable __rcu
**prth
;
2246 struct fib_nh
*nh
= &FIB_RES_NH(*res
);
2248 fnhe
= find_exception(nh
, fl4
->daddr
);
2252 prth
= &fnhe
->fnhe_rth_output
;
2254 if (unlikely(fl4
->flowi4_flags
&
2255 FLOWI_FLAG_KNOWN_NH
&&
2257 nh
->nh_scope
== RT_SCOPE_LINK
))) {
2261 prth
= raw_cpu_ptr(nh
->nh_pcpu_rth_output
);
2263 rth
= rcu_dereference(*prth
);
2264 if (rt_cache_valid(rth
) && dst_hold_safe(&rth
->dst
))
2269 rth
= rt_dst_alloc(dev_out
, flags
, type
,
2270 IN_DEV_CONF_GET(in_dev
, NOPOLICY
),
2271 IN_DEV_CONF_GET(in_dev
, NOXFRM
),
2274 return ERR_PTR(-ENOBUFS
);
2276 rth
->rt_iif
= orig_oif
;
2278 RT_CACHE_STAT_INC(out_slow_tot
);
2280 if (flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) {
2281 if (flags
& RTCF_LOCAL
&&
2282 !(dev_out
->flags
& IFF_LOOPBACK
)) {
2283 rth
->dst
.output
= ip_mc_output
;
2284 RT_CACHE_STAT_INC(out_slow_mc
);
2286 #ifdef CONFIG_IP_MROUTE
2287 if (type
== RTN_MULTICAST
) {
2288 if (IN_DEV_MFORWARD(in_dev
) &&
2289 !ipv4_is_local_multicast(fl4
->daddr
)) {
2290 rth
->dst
.input
= ip_mr_input
;
2291 rth
->dst
.output
= ip_mc_output
;
2297 rt_set_nexthop(rth
, fl4
->daddr
, res
, fnhe
, fi
, type
, 0, do_cache
);
2298 lwtunnel_set_redirect(&rth
->dst
);
2304 * Major route resolver routine.
2307 struct rtable
*ip_route_output_key_hash(struct net
*net
, struct flowi4
*fl4
,
2308 const struct sk_buff
*skb
)
2310 __u8 tos
= RT_FL_TOS(fl4
);
2311 struct fib_result res
= {
2319 fl4
->flowi4_iif
= LOOPBACK_IFINDEX
;
2320 fl4
->flowi4_tos
= tos
& IPTOS_RT_MASK
;
2321 fl4
->flowi4_scope
= ((tos
& RTO_ONLINK
) ?
2322 RT_SCOPE_LINK
: RT_SCOPE_UNIVERSE
);
2325 rth
= ip_route_output_key_hash_rcu(net
, fl4
, &res
, skb
);
2330 EXPORT_SYMBOL_GPL(ip_route_output_key_hash
);
2332 struct rtable
*ip_route_output_key_hash_rcu(struct net
*net
, struct flowi4
*fl4
,
2333 struct fib_result
*res
,
2334 const struct sk_buff
*skb
)
2336 struct net_device
*dev_out
= NULL
;
2337 int orig_oif
= fl4
->flowi4_oif
;
2338 unsigned int flags
= 0;
2340 int err
= -ENETUNREACH
;
2343 rth
= ERR_PTR(-EINVAL
);
2344 if (ipv4_is_multicast(fl4
->saddr
) ||
2345 ipv4_is_lbcast(fl4
->saddr
) ||
2346 ipv4_is_zeronet(fl4
->saddr
))
2349 /* I removed check for oif == dev_out->oif here.
2350 It was wrong for two reasons:
2351 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2352 is assigned to multiple interfaces.
2353 2. Moreover, we are allowed to send packets with saddr
2354 of another iface. --ANK
2357 if (fl4
->flowi4_oif
== 0 &&
2358 (ipv4_is_multicast(fl4
->daddr
) ||
2359 ipv4_is_lbcast(fl4
->daddr
))) {
2360 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2361 dev_out
= __ip_dev_find(net
, fl4
->saddr
, false);
2365 /* Special hack: user can direct multicasts
2366 and limited broadcast via necessary interface
2367 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2368 This hack is not just for fun, it allows
2369 vic,vat and friends to work.
2370 They bind socket to loopback, set ttl to zero
2371 and expect that it will work.
2372 From the viewpoint of routing cache they are broken,
2373 because we are not allowed to build multicast path
2374 with loopback source addr (look, routing cache
2375 cannot know, that ttl is zero, so that packet
2376 will not leave this host and route is valid).
2377 Luckily, this hack is good workaround.
2380 fl4
->flowi4_oif
= dev_out
->ifindex
;
2384 if (!(fl4
->flowi4_flags
& FLOWI_FLAG_ANYSRC
)) {
2385 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2386 if (!__ip_dev_find(net
, fl4
->saddr
, false))
2392 if (fl4
->flowi4_oif
) {
2393 dev_out
= dev_get_by_index_rcu(net
, fl4
->flowi4_oif
);
2394 rth
= ERR_PTR(-ENODEV
);
2398 /* RACE: Check return value of inet_select_addr instead. */
2399 if (!(dev_out
->flags
& IFF_UP
) || !__in_dev_get_rcu(dev_out
)) {
2400 rth
= ERR_PTR(-ENETUNREACH
);
2403 if (ipv4_is_local_multicast(fl4
->daddr
) ||
2404 ipv4_is_lbcast(fl4
->daddr
) ||
2405 fl4
->flowi4_proto
== IPPROTO_IGMP
) {
2407 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2412 if (ipv4_is_multicast(fl4
->daddr
))
2413 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2415 else if (!fl4
->daddr
)
2416 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2422 fl4
->daddr
= fl4
->saddr
;
2424 fl4
->daddr
= fl4
->saddr
= htonl(INADDR_LOOPBACK
);
2425 dev_out
= net
->loopback_dev
;
2426 fl4
->flowi4_oif
= LOOPBACK_IFINDEX
;
2427 res
->type
= RTN_LOCAL
;
2428 flags
|= RTCF_LOCAL
;
2432 err
= fib_lookup(net
, fl4
, res
, 0);
2436 if (fl4
->flowi4_oif
&&
2437 (ipv4_is_multicast(fl4
->daddr
) ||
2438 !netif_index_is_l3_master(net
, fl4
->flowi4_oif
))) {
2439 /* Apparently, routing tables are wrong. Assume,
2440 that the destination is on link.
2443 Because we are allowed to send to iface
2444 even if it has NO routes and NO assigned
2445 addresses. When oif is specified, routing
2446 tables are looked up with only one purpose:
2447 to catch if destination is gatewayed, rather than
2448 direct. Moreover, if MSG_DONTROUTE is set,
2449 we send packet, ignoring both routing tables
2450 and ifaddr state. --ANK
2453 We could make it even if oif is unknown,
2454 likely IPv6, but we do not.
2457 if (fl4
->saddr
== 0)
2458 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2460 res
->type
= RTN_UNICAST
;
2467 if (res
->type
== RTN_LOCAL
) {
2469 if (res
->fi
->fib_prefsrc
)
2470 fl4
->saddr
= res
->fi
->fib_prefsrc
;
2472 fl4
->saddr
= fl4
->daddr
;
2475 /* L3 master device is the loopback for that domain */
2476 dev_out
= l3mdev_master_dev_rcu(FIB_RES_DEV(*res
)) ? :
2479 /* make sure orig_oif points to fib result device even
2480 * though packet rx/tx happens over loopback or l3mdev
2482 orig_oif
= FIB_RES_OIF(*res
);
2484 fl4
->flowi4_oif
= dev_out
->ifindex
;
2485 flags
|= RTCF_LOCAL
;
2489 fib_select_path(net
, res
, fl4
, skb
);
2491 dev_out
= FIB_RES_DEV(*res
);
2492 fl4
->flowi4_oif
= dev_out
->ifindex
;
2496 rth
= __mkroute_output(res
, fl4
, orig_oif
, dev_out
, flags
);
2502 static struct dst_entry
*ipv4_blackhole_dst_check(struct dst_entry
*dst
, u32 cookie
)
2507 static unsigned int ipv4_blackhole_mtu(const struct dst_entry
*dst
)
2509 unsigned int mtu
= dst_metric_raw(dst
, RTAX_MTU
);
2511 return mtu
? : dst
->dev
->mtu
;
2514 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry
*dst
, struct sock
*sk
,
2515 struct sk_buff
*skb
, u32 mtu
)
2519 static void ipv4_rt_blackhole_redirect(struct dst_entry
*dst
, struct sock
*sk
,
2520 struct sk_buff
*skb
)
2524 static u32
*ipv4_rt_blackhole_cow_metrics(struct dst_entry
*dst
,
2530 static struct dst_ops ipv4_dst_blackhole_ops
= {
2532 .check
= ipv4_blackhole_dst_check
,
2533 .mtu
= ipv4_blackhole_mtu
,
2534 .default_advmss
= ipv4_default_advmss
,
2535 .update_pmtu
= ipv4_rt_blackhole_update_pmtu
,
2536 .redirect
= ipv4_rt_blackhole_redirect
,
2537 .cow_metrics
= ipv4_rt_blackhole_cow_metrics
,
2538 .neigh_lookup
= ipv4_neigh_lookup
,
2541 struct dst_entry
*ipv4_blackhole_route(struct net
*net
, struct dst_entry
*dst_orig
)
2543 struct rtable
*ort
= (struct rtable
*) dst_orig
;
2546 rt
= dst_alloc(&ipv4_dst_blackhole_ops
, NULL
, 1, DST_OBSOLETE_DEAD
, 0);
2548 struct dst_entry
*new = &rt
->dst
;
2551 new->input
= dst_discard
;
2552 new->output
= dst_discard_out
;
2554 new->dev
= net
->loopback_dev
;
2558 rt
->rt_is_input
= ort
->rt_is_input
;
2559 rt
->rt_iif
= ort
->rt_iif
;
2560 rt
->rt_pmtu
= ort
->rt_pmtu
;
2561 rt
->rt_mtu_locked
= ort
->rt_mtu_locked
;
2563 rt
->rt_genid
= rt_genid_ipv4(net
);
2564 rt
->rt_flags
= ort
->rt_flags
;
2565 rt
->rt_type
= ort
->rt_type
;
2566 rt
->rt_gateway
= ort
->rt_gateway
;
2567 rt
->rt_uses_gateway
= ort
->rt_uses_gateway
;
2569 INIT_LIST_HEAD(&rt
->rt_uncached
);
2572 dst_release(dst_orig
);
2574 return rt
? &rt
->dst
: ERR_PTR(-ENOMEM
);
2577 struct rtable
*ip_route_output_flow(struct net
*net
, struct flowi4
*flp4
,
2578 const struct sock
*sk
)
2580 struct rtable
*rt
= __ip_route_output_key(net
, flp4
);
2585 if (flp4
->flowi4_proto
)
2586 rt
= (struct rtable
*)xfrm_lookup_route(net
, &rt
->dst
,
2587 flowi4_to_flowi(flp4
),
2592 EXPORT_SYMBOL_GPL(ip_route_output_flow
);
2594 /* called with rcu_read_lock held */
2595 static int rt_fill_info(struct net
*net
, __be32 dst
, __be32 src
,
2596 struct rtable
*rt
, u32 table_id
, struct flowi4
*fl4
,
2597 struct sk_buff
*skb
, u32 portid
, u32 seq
)
2600 struct nlmsghdr
*nlh
;
2601 unsigned long expires
= 0;
2603 u32 metrics
[RTAX_MAX
];
2605 nlh
= nlmsg_put(skb
, portid
, seq
, RTM_NEWROUTE
, sizeof(*r
), 0);
2609 r
= nlmsg_data(nlh
);
2610 r
->rtm_family
= AF_INET
;
2611 r
->rtm_dst_len
= 32;
2613 r
->rtm_tos
= fl4
->flowi4_tos
;
2614 r
->rtm_table
= table_id
< 256 ? table_id
: RT_TABLE_COMPAT
;
2615 if (nla_put_u32(skb
, RTA_TABLE
, table_id
))
2616 goto nla_put_failure
;
2617 r
->rtm_type
= rt
->rt_type
;
2618 r
->rtm_scope
= RT_SCOPE_UNIVERSE
;
2619 r
->rtm_protocol
= RTPROT_UNSPEC
;
2620 r
->rtm_flags
= (rt
->rt_flags
& ~0xFFFF) | RTM_F_CLONED
;
2621 if (rt
->rt_flags
& RTCF_NOTIFY
)
2622 r
->rtm_flags
|= RTM_F_NOTIFY
;
2623 if (IPCB(skb
)->flags
& IPSKB_DOREDIRECT
)
2624 r
->rtm_flags
|= RTCF_DOREDIRECT
;
2626 if (nla_put_in_addr(skb
, RTA_DST
, dst
))
2627 goto nla_put_failure
;
2629 r
->rtm_src_len
= 32;
2630 if (nla_put_in_addr(skb
, RTA_SRC
, src
))
2631 goto nla_put_failure
;
2634 nla_put_u32(skb
, RTA_OIF
, rt
->dst
.dev
->ifindex
))
2635 goto nla_put_failure
;
2636 #ifdef CONFIG_IP_ROUTE_CLASSID
2637 if (rt
->dst
.tclassid
&&
2638 nla_put_u32(skb
, RTA_FLOW
, rt
->dst
.tclassid
))
2639 goto nla_put_failure
;
2641 if (!rt_is_input_route(rt
) &&
2642 fl4
->saddr
!= src
) {
2643 if (nla_put_in_addr(skb
, RTA_PREFSRC
, fl4
->saddr
))
2644 goto nla_put_failure
;
2646 if (rt
->rt_uses_gateway
&&
2647 nla_put_in_addr(skb
, RTA_GATEWAY
, rt
->rt_gateway
))
2648 goto nla_put_failure
;
2650 expires
= rt
->dst
.expires
;
2652 unsigned long now
= jiffies
;
2654 if (time_before(now
, expires
))
2660 memcpy(metrics
, dst_metrics_ptr(&rt
->dst
), sizeof(metrics
));
2661 if (rt
->rt_pmtu
&& expires
)
2662 metrics
[RTAX_MTU
- 1] = rt
->rt_pmtu
;
2663 if (rt
->rt_mtu_locked
&& expires
)
2664 metrics
[RTAX_LOCK
- 1] |= BIT(RTAX_MTU
);
2665 if (rtnetlink_put_metrics(skb
, metrics
) < 0)
2666 goto nla_put_failure
;
2668 if (fl4
->flowi4_mark
&&
2669 nla_put_u32(skb
, RTA_MARK
, fl4
->flowi4_mark
))
2670 goto nla_put_failure
;
2672 if (!uid_eq(fl4
->flowi4_uid
, INVALID_UID
) &&
2673 nla_put_u32(skb
, RTA_UID
,
2674 from_kuid_munged(current_user_ns(), fl4
->flowi4_uid
)))
2675 goto nla_put_failure
;
2677 error
= rt
->dst
.error
;
2679 if (rt_is_input_route(rt
)) {
2680 #ifdef CONFIG_IP_MROUTE
2681 if (ipv4_is_multicast(dst
) && !ipv4_is_local_multicast(dst
) &&
2682 IPV4_DEVCONF_ALL(net
, MC_FORWARDING
)) {
2683 int err
= ipmr_get_route(net
, skb
,
2684 fl4
->saddr
, fl4
->daddr
,
2690 goto nla_put_failure
;
2694 if (nla_put_u32(skb
, RTA_IIF
, fl4
->flowi4_iif
))
2695 goto nla_put_failure
;
2698 if (rtnl_put_cacheinfo(skb
, &rt
->dst
, 0, expires
, error
) < 0)
2699 goto nla_put_failure
;
2701 nlmsg_end(skb
, nlh
);
2705 nlmsg_cancel(skb
, nlh
);
2709 static struct sk_buff
*inet_rtm_getroute_build_skb(__be32 src
, __be32 dst
,
2710 u8 ip_proto
, __be16 sport
,
2713 struct sk_buff
*skb
;
2716 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
2720 /* Reserve room for dummy headers, this skb can pass
2721 * through good chunk of routing engine.
2723 skb_reset_mac_header(skb
);
2724 skb_reset_network_header(skb
);
2725 skb
->protocol
= htons(ETH_P_IP
);
2726 iph
= skb_put(skb
, sizeof(struct iphdr
));
2727 iph
->protocol
= ip_proto
;
2733 skb_set_transport_header(skb
, skb
->len
);
2735 switch (iph
->protocol
) {
2737 struct udphdr
*udph
;
2739 udph
= skb_put_zero(skb
, sizeof(struct udphdr
));
2740 udph
->source
= sport
;
2742 udph
->len
= sizeof(struct udphdr
);
2747 struct tcphdr
*tcph
;
2749 tcph
= skb_put_zero(skb
, sizeof(struct tcphdr
));
2750 tcph
->source
= sport
;
2752 tcph
->doff
= sizeof(struct tcphdr
) / 4;
2754 tcph
->check
= ~tcp_v4_check(sizeof(struct tcphdr
),
2758 case IPPROTO_ICMP
: {
2759 struct icmphdr
*icmph
;
2761 icmph
= skb_put_zero(skb
, sizeof(struct icmphdr
));
2762 icmph
->type
= ICMP_ECHO
;
2770 static int inet_rtm_valid_getroute_req(struct sk_buff
*skb
,
2771 const struct nlmsghdr
*nlh
,
2773 struct netlink_ext_ack
*extack
)
2778 if (nlh
->nlmsg_len
< nlmsg_msg_size(sizeof(*rtm
))) {
2779 NL_SET_ERR_MSG(extack
,
2780 "ipv4: Invalid header for route get request");
2784 if (!netlink_strict_get_check(skb
))
2785 return nlmsg_parse(nlh
, sizeof(*rtm
), tb
, RTA_MAX
,
2786 rtm_ipv4_policy
, extack
);
2788 rtm
= nlmsg_data(nlh
);
2789 if ((rtm
->rtm_src_len
&& rtm
->rtm_src_len
!= 32) ||
2790 (rtm
->rtm_dst_len
&& rtm
->rtm_dst_len
!= 32) ||
2791 rtm
->rtm_table
|| rtm
->rtm_protocol
||
2792 rtm
->rtm_scope
|| rtm
->rtm_type
) {
2793 NL_SET_ERR_MSG(extack
, "ipv4: Invalid values in header for route get request");
2797 if (rtm
->rtm_flags
& ~(RTM_F_NOTIFY
|
2798 RTM_F_LOOKUP_TABLE
|
2800 NL_SET_ERR_MSG(extack
, "ipv4: Unsupported rtm_flags for route get request");
2804 err
= nlmsg_parse_strict(nlh
, sizeof(*rtm
), tb
, RTA_MAX
,
2805 rtm_ipv4_policy
, extack
);
2809 if ((tb
[RTA_SRC
] && !rtm
->rtm_src_len
) ||
2810 (tb
[RTA_DST
] && !rtm
->rtm_dst_len
)) {
2811 NL_SET_ERR_MSG(extack
, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
2815 for (i
= 0; i
<= RTA_MAX
; i
++) {
2831 NL_SET_ERR_MSG(extack
, "ipv4: Unsupported attribute in route get request");
2839 static int inet_rtm_getroute(struct sk_buff
*in_skb
, struct nlmsghdr
*nlh
,
2840 struct netlink_ext_ack
*extack
)
2842 struct net
*net
= sock_net(in_skb
->sk
);
2843 struct nlattr
*tb
[RTA_MAX
+1];
2844 u32 table_id
= RT_TABLE_MAIN
;
2845 __be16 sport
= 0, dport
= 0;
2846 struct fib_result res
= {};
2847 u8 ip_proto
= IPPROTO_UDP
;
2848 struct rtable
*rt
= NULL
;
2849 struct sk_buff
*skb
;
2851 struct flowi4 fl4
= {};
2859 err
= inet_rtm_valid_getroute_req(in_skb
, nlh
, tb
, extack
);
2863 rtm
= nlmsg_data(nlh
);
2864 src
= tb
[RTA_SRC
] ? nla_get_in_addr(tb
[RTA_SRC
]) : 0;
2865 dst
= tb
[RTA_DST
] ? nla_get_in_addr(tb
[RTA_DST
]) : 0;
2866 iif
= tb
[RTA_IIF
] ? nla_get_u32(tb
[RTA_IIF
]) : 0;
2867 mark
= tb
[RTA_MARK
] ? nla_get_u32(tb
[RTA_MARK
]) : 0;
2869 uid
= make_kuid(current_user_ns(), nla_get_u32(tb
[RTA_UID
]));
2871 uid
= (iif
? INVALID_UID
: current_uid());
2873 if (tb
[RTA_IP_PROTO
]) {
2874 err
= rtm_getroute_parse_ip_proto(tb
[RTA_IP_PROTO
],
2881 sport
= nla_get_be16(tb
[RTA_SPORT
]);
2884 dport
= nla_get_be16(tb
[RTA_DPORT
]);
2886 skb
= inet_rtm_getroute_build_skb(src
, dst
, ip_proto
, sport
, dport
);
2892 fl4
.flowi4_tos
= rtm
->rtm_tos
;
2893 fl4
.flowi4_oif
= tb
[RTA_OIF
] ? nla_get_u32(tb
[RTA_OIF
]) : 0;
2894 fl4
.flowi4_mark
= mark
;
2895 fl4
.flowi4_uid
= uid
;
2897 fl4
.fl4_sport
= sport
;
2899 fl4
.fl4_dport
= dport
;
2900 fl4
.flowi4_proto
= ip_proto
;
2905 struct net_device
*dev
;
2907 dev
= dev_get_by_index_rcu(net
, iif
);
2913 fl4
.flowi4_iif
= iif
; /* for rt_fill_info */
2916 err
= ip_route_input_rcu(skb
, dst
, src
, rtm
->rtm_tos
,
2919 rt
= skb_rtable(skb
);
2920 if (err
== 0 && rt
->dst
.error
)
2921 err
= -rt
->dst
.error
;
2923 fl4
.flowi4_iif
= LOOPBACK_IFINDEX
;
2924 skb
->dev
= net
->loopback_dev
;
2925 rt
= ip_route_output_key_hash_rcu(net
, &fl4
, &res
, skb
);
2930 skb_dst_set(skb
, &rt
->dst
);
2936 if (rtm
->rtm_flags
& RTM_F_NOTIFY
)
2937 rt
->rt_flags
|= RTCF_NOTIFY
;
2939 if (rtm
->rtm_flags
& RTM_F_LOOKUP_TABLE
)
2940 table_id
= res
.table
? res
.table
->tb_id
: 0;
2942 /* reset skb for netlink reply msg */
2944 skb_reset_network_header(skb
);
2945 skb_reset_transport_header(skb
);
2946 skb_reset_mac_header(skb
);
2948 if (rtm
->rtm_flags
& RTM_F_FIB_MATCH
) {
2950 err
= fib_props
[res
.type
].error
;
2952 err
= -EHOSTUNREACH
;
2955 err
= fib_dump_info(skb
, NETLINK_CB(in_skb
).portid
,
2956 nlh
->nlmsg_seq
, RTM_NEWROUTE
, table_id
,
2957 rt
->rt_type
, res
.prefix
, res
.prefixlen
,
2958 fl4
.flowi4_tos
, res
.fi
, 0);
2960 err
= rt_fill_info(net
, dst
, src
, rt
, table_id
, &fl4
, skb
,
2961 NETLINK_CB(in_skb
).portid
, nlh
->nlmsg_seq
);
2968 err
= rtnl_unicast(skb
, net
, NETLINK_CB(in_skb
).portid
);
2978 void ip_rt_multicast_event(struct in_device
*in_dev
)
2980 rt_cache_flush(dev_net(in_dev
->dev
));
2983 #ifdef CONFIG_SYSCTL
2984 static int ip_rt_gc_interval __read_mostly
= 60 * HZ
;
2985 static int ip_rt_gc_min_interval __read_mostly
= HZ
/ 2;
2986 static int ip_rt_gc_elasticity __read_mostly
= 8;
2987 static int ip_min_valid_pmtu __read_mostly
= IPV4_MIN_MTU
;
2989 static int ipv4_sysctl_rtcache_flush(struct ctl_table
*__ctl
, int write
,
2990 void __user
*buffer
,
2991 size_t *lenp
, loff_t
*ppos
)
2993 struct net
*net
= (struct net
*)__ctl
->extra1
;
2996 rt_cache_flush(net
);
2997 fnhe_genid_bump(net
);
3004 static struct ctl_table ipv4_route_table
[] = {
3006 .procname
= "gc_thresh",
3007 .data
= &ipv4_dst_ops
.gc_thresh
,
3008 .maxlen
= sizeof(int),
3010 .proc_handler
= proc_dointvec
,
3013 .procname
= "max_size",
3014 .data
= &ip_rt_max_size
,
3015 .maxlen
= sizeof(int),
3017 .proc_handler
= proc_dointvec
,
3020 /* Deprecated. Use gc_min_interval_ms */
3022 .procname
= "gc_min_interval",
3023 .data
= &ip_rt_gc_min_interval
,
3024 .maxlen
= sizeof(int),
3026 .proc_handler
= proc_dointvec_jiffies
,
3029 .procname
= "gc_min_interval_ms",
3030 .data
= &ip_rt_gc_min_interval
,
3031 .maxlen
= sizeof(int),
3033 .proc_handler
= proc_dointvec_ms_jiffies
,
3036 .procname
= "gc_timeout",
3037 .data
= &ip_rt_gc_timeout
,
3038 .maxlen
= sizeof(int),
3040 .proc_handler
= proc_dointvec_jiffies
,
3043 .procname
= "gc_interval",
3044 .data
= &ip_rt_gc_interval
,
3045 .maxlen
= sizeof(int),
3047 .proc_handler
= proc_dointvec_jiffies
,
3050 .procname
= "redirect_load",
3051 .data
= &ip_rt_redirect_load
,
3052 .maxlen
= sizeof(int),
3054 .proc_handler
= proc_dointvec
,
3057 .procname
= "redirect_number",
3058 .data
= &ip_rt_redirect_number
,
3059 .maxlen
= sizeof(int),
3061 .proc_handler
= proc_dointvec
,
3064 .procname
= "redirect_silence",
3065 .data
= &ip_rt_redirect_silence
,
3066 .maxlen
= sizeof(int),
3068 .proc_handler
= proc_dointvec
,
3071 .procname
= "error_cost",
3072 .data
= &ip_rt_error_cost
,
3073 .maxlen
= sizeof(int),
3075 .proc_handler
= proc_dointvec
,
3078 .procname
= "error_burst",
3079 .data
= &ip_rt_error_burst
,
3080 .maxlen
= sizeof(int),
3082 .proc_handler
= proc_dointvec
,
3085 .procname
= "gc_elasticity",
3086 .data
= &ip_rt_gc_elasticity
,
3087 .maxlen
= sizeof(int),
3089 .proc_handler
= proc_dointvec
,
3092 .procname
= "mtu_expires",
3093 .data
= &ip_rt_mtu_expires
,
3094 .maxlen
= sizeof(int),
3096 .proc_handler
= proc_dointvec_jiffies
,
3099 .procname
= "min_pmtu",
3100 .data
= &ip_rt_min_pmtu
,
3101 .maxlen
= sizeof(int),
3103 .proc_handler
= proc_dointvec_minmax
,
3104 .extra1
= &ip_min_valid_pmtu
,
3107 .procname
= "min_adv_mss",
3108 .data
= &ip_rt_min_advmss
,
3109 .maxlen
= sizeof(int),
3111 .proc_handler
= proc_dointvec
,
3116 static struct ctl_table ipv4_route_flush_table
[] = {
3118 .procname
= "flush",
3119 .maxlen
= sizeof(int),
3121 .proc_handler
= ipv4_sysctl_rtcache_flush
,
3126 static __net_init
int sysctl_route_net_init(struct net
*net
)
3128 struct ctl_table
*tbl
;
3130 tbl
= ipv4_route_flush_table
;
3131 if (!net_eq(net
, &init_net
)) {
3132 tbl
= kmemdup(tbl
, sizeof(ipv4_route_flush_table
), GFP_KERNEL
);
3136 /* Don't export sysctls to unprivileged users */
3137 if (net
->user_ns
!= &init_user_ns
)
3138 tbl
[0].procname
= NULL
;
3140 tbl
[0].extra1
= net
;
3142 net
->ipv4
.route_hdr
= register_net_sysctl(net
, "net/ipv4/route", tbl
);
3143 if (!net
->ipv4
.route_hdr
)
3148 if (tbl
!= ipv4_route_flush_table
)
3154 static __net_exit
void sysctl_route_net_exit(struct net
*net
)
3156 struct ctl_table
*tbl
;
3158 tbl
= net
->ipv4
.route_hdr
->ctl_table_arg
;
3159 unregister_net_sysctl_table(net
->ipv4
.route_hdr
);
3160 BUG_ON(tbl
== ipv4_route_flush_table
);
3164 static __net_initdata
struct pernet_operations sysctl_route_ops
= {
3165 .init
= sysctl_route_net_init
,
3166 .exit
= sysctl_route_net_exit
,
3170 static __net_init
int rt_genid_init(struct net
*net
)
3172 atomic_set(&net
->ipv4
.rt_genid
, 0);
3173 atomic_set(&net
->fnhe_genid
, 0);
3174 atomic_set(&net
->ipv4
.dev_addr_genid
, get_random_int());
3178 static __net_initdata
struct pernet_operations rt_genid_ops
= {
3179 .init
= rt_genid_init
,
3182 static int __net_init
ipv4_inetpeer_init(struct net
*net
)
3184 struct inet_peer_base
*bp
= kmalloc(sizeof(*bp
), GFP_KERNEL
);
3188 inet_peer_base_init(bp
);
3189 net
->ipv4
.peers
= bp
;
3193 static void __net_exit
ipv4_inetpeer_exit(struct net
*net
)
3195 struct inet_peer_base
*bp
= net
->ipv4
.peers
;
3197 net
->ipv4
.peers
= NULL
;
3198 inetpeer_invalidate_tree(bp
);
3202 static __net_initdata
struct pernet_operations ipv4_inetpeer_ops
= {
3203 .init
= ipv4_inetpeer_init
,
3204 .exit
= ipv4_inetpeer_exit
,
3207 #ifdef CONFIG_IP_ROUTE_CLASSID
3208 struct ip_rt_acct __percpu
*ip_rt_acct __read_mostly
;
3209 #endif /* CONFIG_IP_ROUTE_CLASSID */
3211 int __init
ip_rt_init(void)
3215 ip_idents
= kmalloc_array(IP_IDENTS_SZ
, sizeof(*ip_idents
),
3218 panic("IP: failed to allocate ip_idents\n");
3220 prandom_bytes(ip_idents
, IP_IDENTS_SZ
* sizeof(*ip_idents
));
3222 ip_tstamps
= kcalloc(IP_IDENTS_SZ
, sizeof(*ip_tstamps
), GFP_KERNEL
);
3224 panic("IP: failed to allocate ip_tstamps\n");
3226 for_each_possible_cpu(cpu
) {
3227 struct uncached_list
*ul
= &per_cpu(rt_uncached_list
, cpu
);
3229 INIT_LIST_HEAD(&ul
->head
);
3230 spin_lock_init(&ul
->lock
);
3232 #ifdef CONFIG_IP_ROUTE_CLASSID
3233 ip_rt_acct
= __alloc_percpu(256 * sizeof(struct ip_rt_acct
), __alignof__(struct ip_rt_acct
));
3235 panic("IP: failed to allocate ip_rt_acct\n");
3238 ipv4_dst_ops
.kmem_cachep
=
3239 kmem_cache_create("ip_dst_cache", sizeof(struct rtable
), 0,
3240 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3242 ipv4_dst_blackhole_ops
.kmem_cachep
= ipv4_dst_ops
.kmem_cachep
;
3244 if (dst_entries_init(&ipv4_dst_ops
) < 0)
3245 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3247 if (dst_entries_init(&ipv4_dst_blackhole_ops
) < 0)
3248 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3250 ipv4_dst_ops
.gc_thresh
= ~0;
3251 ip_rt_max_size
= INT_MAX
;
3256 if (ip_rt_proc_init())
3257 pr_err("Unable to create route proc files\n");
3262 rtnl_register(PF_INET
, RTM_GETROUTE
, inet_rtm_getroute
, NULL
,
3263 RTNL_FLAG_DOIT_UNLOCKED
);
3265 #ifdef CONFIG_SYSCTL
3266 register_pernet_subsys(&sysctl_route_ops
);
3268 register_pernet_subsys(&rt_genid_ops
);
3269 register_pernet_subsys(&ipv4_inetpeer_ops
);
3273 #ifdef CONFIG_SYSCTL
3275 * We really need to sanitize the damn ipv4 init order, then all
3276 * this nonsense will go away.
3278 void __init
ip_static_sysctl_init(void)
3280 register_net_sysctl(&init_net
, "net/ipv4/route", ipv4_route_table
);