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 #include "fib_lookup.h"
119 #define RT_FL_TOS(oldflp4) \
120 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
122 #define RT_GC_TIMEOUT (300*HZ)
124 static int ip_rt_max_size
;
125 static int ip_rt_redirect_number __read_mostly
= 9;
126 static int ip_rt_redirect_load __read_mostly
= HZ
/ 50;
127 static int ip_rt_redirect_silence __read_mostly
= ((HZ
/ 50) << (9 + 1));
128 static int ip_rt_error_cost __read_mostly
= HZ
;
129 static int ip_rt_error_burst __read_mostly
= 5 * HZ
;
130 static int ip_rt_mtu_expires __read_mostly
= 10 * 60 * HZ
;
131 static u32 ip_rt_min_pmtu __read_mostly
= 512 + 20 + 20;
132 static int ip_rt_min_advmss __read_mostly
= 256;
134 static int ip_rt_gc_timeout __read_mostly
= RT_GC_TIMEOUT
;
136 static int ip_min_valid_pmtu __read_mostly
= IPV4_MIN_MTU
;
139 * Interface to generic destination cache.
142 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
);
143 static unsigned int ipv4_default_advmss(const struct dst_entry
*dst
);
144 static unsigned int ipv4_mtu(const struct dst_entry
*dst
);
145 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
);
146 static void ipv4_link_failure(struct sk_buff
*skb
);
147 static void ip_rt_update_pmtu(struct dst_entry
*dst
, struct sock
*sk
,
148 struct sk_buff
*skb
, u32 mtu
);
149 static void ip_do_redirect(struct dst_entry
*dst
, struct sock
*sk
,
150 struct sk_buff
*skb
);
151 static void ipv4_dst_destroy(struct dst_entry
*dst
);
153 static u32
*ipv4_cow_metrics(struct dst_entry
*dst
, unsigned long old
)
159 static struct neighbour
*ipv4_neigh_lookup(const struct dst_entry
*dst
,
162 static void ipv4_confirm_neigh(const struct dst_entry
*dst
, const void *daddr
);
164 static struct dst_ops ipv4_dst_ops
= {
166 .check
= ipv4_dst_check
,
167 .default_advmss
= ipv4_default_advmss
,
169 .cow_metrics
= ipv4_cow_metrics
,
170 .destroy
= ipv4_dst_destroy
,
171 .negative_advice
= ipv4_negative_advice
,
172 .link_failure
= ipv4_link_failure
,
173 .update_pmtu
= ip_rt_update_pmtu
,
174 .redirect
= ip_do_redirect
,
175 .local_out
= __ip_local_out
,
176 .neigh_lookup
= ipv4_neigh_lookup
,
177 .confirm_neigh
= ipv4_confirm_neigh
,
180 #define ECN_OR_COST(class) TC_PRIO_##class
182 const __u8 ip_tos2prio
[16] = {
184 ECN_OR_COST(BESTEFFORT
),
186 ECN_OR_COST(BESTEFFORT
),
192 ECN_OR_COST(INTERACTIVE
),
194 ECN_OR_COST(INTERACTIVE
),
195 TC_PRIO_INTERACTIVE_BULK
,
196 ECN_OR_COST(INTERACTIVE_BULK
),
197 TC_PRIO_INTERACTIVE_BULK
,
198 ECN_OR_COST(INTERACTIVE_BULK
)
200 EXPORT_SYMBOL(ip_tos2prio
);
202 static DEFINE_PER_CPU(struct rt_cache_stat
, rt_cache_stat
);
203 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
205 #ifdef CONFIG_PROC_FS
206 static void *rt_cache_seq_start(struct seq_file
*seq
, loff_t
*pos
)
210 return SEQ_START_TOKEN
;
213 static void *rt_cache_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
219 static void rt_cache_seq_stop(struct seq_file
*seq
, void *v
)
223 static int rt_cache_seq_show(struct seq_file
*seq
, void *v
)
225 if (v
== SEQ_START_TOKEN
)
226 seq_printf(seq
, "%-127s\n",
227 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
228 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
233 static const struct seq_operations rt_cache_seq_ops
= {
234 .start
= rt_cache_seq_start
,
235 .next
= rt_cache_seq_next
,
236 .stop
= rt_cache_seq_stop
,
237 .show
= rt_cache_seq_show
,
240 static int rt_cache_seq_open(struct inode
*inode
, struct file
*file
)
242 return seq_open(file
, &rt_cache_seq_ops
);
245 static const struct file_operations rt_cache_seq_fops
= {
246 .owner
= THIS_MODULE
,
247 .open
= rt_cache_seq_open
,
250 .release
= seq_release
,
254 static void *rt_cpu_seq_start(struct seq_file
*seq
, loff_t
*pos
)
259 return SEQ_START_TOKEN
;
261 for (cpu
= *pos
-1; cpu
< nr_cpu_ids
; ++cpu
) {
262 if (!cpu_possible(cpu
))
265 return &per_cpu(rt_cache_stat
, cpu
);
270 static void *rt_cpu_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
274 for (cpu
= *pos
; cpu
< nr_cpu_ids
; ++cpu
) {
275 if (!cpu_possible(cpu
))
278 return &per_cpu(rt_cache_stat
, cpu
);
284 static void rt_cpu_seq_stop(struct seq_file
*seq
, void *v
)
289 static int rt_cpu_seq_show(struct seq_file
*seq
, void *v
)
291 struct rt_cache_stat
*st
= v
;
293 if (v
== SEQ_START_TOKEN
) {
294 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");
298 seq_printf(seq
,"%08x %08x %08x %08x %08x %08x %08x %08x "
299 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
300 dst_entries_get_slow(&ipv4_dst_ops
),
313 0, /* st->gc_total */
314 0, /* st->gc_ignored */
315 0, /* st->gc_goal_miss */
316 0, /* st->gc_dst_overflow */
317 0, /* st->in_hlist_search */
318 0 /* st->out_hlist_search */
323 static const struct seq_operations rt_cpu_seq_ops
= {
324 .start
= rt_cpu_seq_start
,
325 .next
= rt_cpu_seq_next
,
326 .stop
= rt_cpu_seq_stop
,
327 .show
= rt_cpu_seq_show
,
331 static int rt_cpu_seq_open(struct inode
*inode
, struct file
*file
)
333 return seq_open(file
, &rt_cpu_seq_ops
);
336 static const struct file_operations rt_cpu_seq_fops
= {
337 .owner
= THIS_MODULE
,
338 .open
= rt_cpu_seq_open
,
341 .release
= seq_release
,
344 #ifdef CONFIG_IP_ROUTE_CLASSID
345 static int rt_acct_proc_show(struct seq_file
*m
, void *v
)
347 struct ip_rt_acct
*dst
, *src
;
350 dst
= kcalloc(256, sizeof(struct ip_rt_acct
), GFP_KERNEL
);
354 for_each_possible_cpu(i
) {
355 src
= (struct ip_rt_acct
*)per_cpu_ptr(ip_rt_acct
, i
);
356 for (j
= 0; j
< 256; j
++) {
357 dst
[j
].o_bytes
+= src
[j
].o_bytes
;
358 dst
[j
].o_packets
+= src
[j
].o_packets
;
359 dst
[j
].i_bytes
+= src
[j
].i_bytes
;
360 dst
[j
].i_packets
+= src
[j
].i_packets
;
364 seq_write(m
, dst
, 256 * sizeof(struct ip_rt_acct
));
369 static int rt_acct_proc_open(struct inode
*inode
, struct file
*file
)
371 return single_open(file
, rt_acct_proc_show
, NULL
);
374 static const struct file_operations rt_acct_proc_fops
= {
375 .owner
= THIS_MODULE
,
376 .open
= rt_acct_proc_open
,
379 .release
= single_release
,
383 static int __net_init
ip_rt_do_proc_init(struct net
*net
)
385 struct proc_dir_entry
*pde
;
387 pde
= proc_create("rt_cache", S_IRUGO
, net
->proc_net
,
392 pde
= proc_create("rt_cache", S_IRUGO
,
393 net
->proc_net_stat
, &rt_cpu_seq_fops
);
397 #ifdef CONFIG_IP_ROUTE_CLASSID
398 pde
= proc_create("rt_acct", 0, net
->proc_net
, &rt_acct_proc_fops
);
404 #ifdef CONFIG_IP_ROUTE_CLASSID
406 remove_proc_entry("rt_cache", net
->proc_net_stat
);
409 remove_proc_entry("rt_cache", net
->proc_net
);
414 static void __net_exit
ip_rt_do_proc_exit(struct net
*net
)
416 remove_proc_entry("rt_cache", net
->proc_net_stat
);
417 remove_proc_entry("rt_cache", net
->proc_net
);
418 #ifdef CONFIG_IP_ROUTE_CLASSID
419 remove_proc_entry("rt_acct", net
->proc_net
);
423 static struct pernet_operations ip_rt_proc_ops __net_initdata
= {
424 .init
= ip_rt_do_proc_init
,
425 .exit
= ip_rt_do_proc_exit
,
428 static int __init
ip_rt_proc_init(void)
430 return register_pernet_subsys(&ip_rt_proc_ops
);
434 static inline int ip_rt_proc_init(void)
438 #endif /* CONFIG_PROC_FS */
440 static inline bool rt_is_expired(const struct rtable
*rth
)
442 return rth
->rt_genid
!= rt_genid_ipv4(dev_net(rth
->dst
.dev
));
445 void rt_cache_flush(struct net
*net
)
447 rt_genid_bump_ipv4(net
);
450 static struct neighbour
*ipv4_neigh_lookup(const struct dst_entry
*dst
,
454 struct net_device
*dev
= dst
->dev
;
455 const __be32
*pkey
= daddr
;
456 const struct rtable
*rt
;
459 rt
= (const struct rtable
*) dst
;
461 pkey
= (const __be32
*) &rt
->rt_gateway
;
463 pkey
= &ip_hdr(skb
)->daddr
;
465 n
= __ipv4_neigh_lookup(dev
, *(__force u32
*)pkey
);
468 return neigh_create(&arp_tbl
, pkey
, dev
);
471 static void ipv4_confirm_neigh(const struct dst_entry
*dst
, const void *daddr
)
473 struct net_device
*dev
= dst
->dev
;
474 const __be32
*pkey
= daddr
;
475 const struct rtable
*rt
;
477 rt
= (const struct rtable
*)dst
;
479 pkey
= (const __be32
*)&rt
->rt_gateway
;
482 (RTCF_MULTICAST
| RTCF_BROADCAST
| RTCF_LOCAL
)))
485 __ipv4_confirm_neigh(dev
, *(__force u32
*)pkey
);
488 #define IP_IDENTS_SZ 2048u
490 static atomic_t
*ip_idents __read_mostly
;
491 static u32
*ip_tstamps __read_mostly
;
493 /* In order to protect privacy, we add a perturbation to identifiers
494 * if one generator is seldom used. This makes hard for an attacker
495 * to infer how many packets were sent between two points in time.
497 u32
ip_idents_reserve(u32 hash
, int segs
)
499 u32
*p_tstamp
= ip_tstamps
+ hash
% IP_IDENTS_SZ
;
500 atomic_t
*p_id
= ip_idents
+ hash
% IP_IDENTS_SZ
;
501 u32 old
= ACCESS_ONCE(*p_tstamp
);
502 u32 now
= (u32
)jiffies
;
505 if (old
!= now
&& cmpxchg(p_tstamp
, old
, now
) == old
)
506 delta
= prandom_u32_max(now
- old
);
508 /* Do not use atomic_add_return() as it makes UBSAN unhappy */
510 old
= (u32
)atomic_read(p_id
);
511 new = old
+ delta
+ segs
;
512 } while (atomic_cmpxchg(p_id
, old
, new) != old
);
516 EXPORT_SYMBOL(ip_idents_reserve
);
518 void __ip_select_ident(struct net
*net
, struct iphdr
*iph
, int segs
)
522 /* Note the following code is not safe, but this is okay. */
523 if (unlikely(siphash_key_is_zero(&net
->ipv4
.ip_id_key
)))
524 get_random_bytes(&net
->ipv4
.ip_id_key
,
525 sizeof(net
->ipv4
.ip_id_key
));
527 hash
= siphash_3u32((__force u32
)iph
->daddr
,
528 (__force u32
)iph
->saddr
,
530 &net
->ipv4
.ip_id_key
);
531 id
= ip_idents_reserve(hash
, segs
);
534 EXPORT_SYMBOL(__ip_select_ident
);
536 static void __build_flow_key(const struct net
*net
, struct flowi4
*fl4
,
537 const struct sock
*sk
,
538 const struct iphdr
*iph
,
540 u8 prot
, u32 mark
, int flow_flags
)
543 const struct inet_sock
*inet
= inet_sk(sk
);
545 oif
= sk
->sk_bound_dev_if
;
547 tos
= RT_CONN_FLAGS(sk
);
548 prot
= inet
->hdrincl
? IPPROTO_RAW
: sk
->sk_protocol
;
550 flowi4_init_output(fl4
, oif
, mark
, tos
,
551 RT_SCOPE_UNIVERSE
, prot
,
553 iph
->daddr
, iph
->saddr
, 0, 0,
554 sock_net_uid(net
, sk
));
557 static void build_skb_flow_key(struct flowi4
*fl4
, const struct sk_buff
*skb
,
558 const struct sock
*sk
)
560 const struct net
*net
= dev_net(skb
->dev
);
561 const struct iphdr
*iph
= ip_hdr(skb
);
562 int oif
= skb
->dev
->ifindex
;
563 u8 tos
= RT_TOS(iph
->tos
);
564 u8 prot
= iph
->protocol
;
565 u32 mark
= skb
->mark
;
567 __build_flow_key(net
, fl4
, sk
, iph
, oif
, tos
, prot
, mark
, 0);
570 static void build_sk_flow_key(struct flowi4
*fl4
, const struct sock
*sk
)
572 const struct inet_sock
*inet
= inet_sk(sk
);
573 const struct ip_options_rcu
*inet_opt
;
574 __be32 daddr
= inet
->inet_daddr
;
577 inet_opt
= rcu_dereference(inet
->inet_opt
);
578 if (inet_opt
&& inet_opt
->opt
.srr
)
579 daddr
= inet_opt
->opt
.faddr
;
580 flowi4_init_output(fl4
, sk
->sk_bound_dev_if
, sk
->sk_mark
,
581 RT_CONN_FLAGS(sk
), RT_SCOPE_UNIVERSE
,
582 inet
->hdrincl
? IPPROTO_RAW
: sk
->sk_protocol
,
583 inet_sk_flowi_flags(sk
),
584 daddr
, inet
->inet_saddr
, 0, 0, sk
->sk_uid
);
588 static void ip_rt_build_flow_key(struct flowi4
*fl4
, const struct sock
*sk
,
589 const struct sk_buff
*skb
)
592 build_skb_flow_key(fl4
, skb
, sk
);
594 build_sk_flow_key(fl4
, sk
);
597 static DEFINE_SPINLOCK(fnhe_lock
);
599 static void fnhe_flush_routes(struct fib_nh_exception
*fnhe
)
603 rt
= rcu_dereference(fnhe
->fnhe_rth_input
);
605 RCU_INIT_POINTER(fnhe
->fnhe_rth_input
, NULL
);
606 dst_dev_put(&rt
->dst
);
607 dst_release(&rt
->dst
);
609 rt
= rcu_dereference(fnhe
->fnhe_rth_output
);
611 RCU_INIT_POINTER(fnhe
->fnhe_rth_output
, NULL
);
612 dst_dev_put(&rt
->dst
);
613 dst_release(&rt
->dst
);
617 static struct fib_nh_exception
*fnhe_oldest(struct fnhe_hash_bucket
*hash
)
619 struct fib_nh_exception
*fnhe
, *oldest
;
621 oldest
= rcu_dereference(hash
->chain
);
622 for (fnhe
= rcu_dereference(oldest
->fnhe_next
); fnhe
;
623 fnhe
= rcu_dereference(fnhe
->fnhe_next
)) {
624 if (time_before(fnhe
->fnhe_stamp
, oldest
->fnhe_stamp
))
627 fnhe_flush_routes(oldest
);
631 static inline u32
fnhe_hashfun(__be32 daddr
)
633 static u32 fnhe_hashrnd __read_mostly
;
636 net_get_random_once(&fnhe_hashrnd
, sizeof(fnhe_hashrnd
));
637 hval
= jhash_1word((__force u32
) daddr
, fnhe_hashrnd
);
638 return hash_32(hval
, FNHE_HASH_SHIFT
);
641 static void fill_route_from_fnhe(struct rtable
*rt
, struct fib_nh_exception
*fnhe
)
643 rt
->rt_pmtu
= fnhe
->fnhe_pmtu
;
644 rt
->rt_mtu_locked
= fnhe
->fnhe_mtu_locked
;
645 rt
->dst
.expires
= fnhe
->fnhe_expires
;
648 rt
->rt_flags
|= RTCF_REDIRECTED
;
649 rt
->rt_gateway
= fnhe
->fnhe_gw
;
650 rt
->rt_uses_gateway
= 1;
654 static void update_or_create_fnhe(struct fib_nh
*nh
, __be32 daddr
, __be32 gw
,
655 u32 pmtu
, bool lock
, unsigned long expires
)
657 struct fnhe_hash_bucket
*hash
;
658 struct fib_nh_exception
*fnhe
;
664 genid
= fnhe_genid(dev_net(nh
->nh_dev
));
665 hval
= fnhe_hashfun(daddr
);
667 spin_lock_bh(&fnhe_lock
);
669 hash
= rcu_dereference(nh
->nh_exceptions
);
671 hash
= kzalloc(FNHE_HASH_SIZE
* sizeof(*hash
), GFP_ATOMIC
);
674 rcu_assign_pointer(nh
->nh_exceptions
, hash
);
680 for (fnhe
= rcu_dereference(hash
->chain
); fnhe
;
681 fnhe
= rcu_dereference(fnhe
->fnhe_next
)) {
682 if (fnhe
->fnhe_daddr
== daddr
)
688 if (fnhe
->fnhe_genid
!= genid
)
689 fnhe
->fnhe_genid
= genid
;
693 fnhe
->fnhe_pmtu
= pmtu
;
694 fnhe
->fnhe_mtu_locked
= lock
;
696 fnhe
->fnhe_expires
= max(1UL, expires
);
697 /* Update all cached dsts too */
698 rt
= rcu_dereference(fnhe
->fnhe_rth_input
);
700 fill_route_from_fnhe(rt
, fnhe
);
701 rt
= rcu_dereference(fnhe
->fnhe_rth_output
);
703 fill_route_from_fnhe(rt
, fnhe
);
705 if (depth
> FNHE_RECLAIM_DEPTH
)
706 fnhe
= fnhe_oldest(hash
);
708 fnhe
= kzalloc(sizeof(*fnhe
), GFP_ATOMIC
);
712 fnhe
->fnhe_next
= hash
->chain
;
713 rcu_assign_pointer(hash
->chain
, fnhe
);
715 fnhe
->fnhe_genid
= genid
;
716 fnhe
->fnhe_daddr
= daddr
;
718 fnhe
->fnhe_pmtu
= pmtu
;
719 fnhe
->fnhe_mtu_locked
= lock
;
720 fnhe
->fnhe_expires
= max(1UL, expires
);
722 /* Exception created; mark the cached routes for the nexthop
723 * stale, so anyone caching it rechecks if this exception
726 rt
= rcu_dereference(nh
->nh_rth_input
);
728 rt
->dst
.obsolete
= DST_OBSOLETE_KILL
;
730 for_each_possible_cpu(i
) {
731 struct rtable __rcu
**prt
;
732 prt
= per_cpu_ptr(nh
->nh_pcpu_rth_output
, i
);
733 rt
= rcu_dereference(*prt
);
735 rt
->dst
.obsolete
= DST_OBSOLETE_KILL
;
739 fnhe
->fnhe_stamp
= jiffies
;
742 spin_unlock_bh(&fnhe_lock
);
745 static void __ip_do_redirect(struct rtable
*rt
, struct sk_buff
*skb
, struct flowi4
*fl4
,
748 __be32 new_gw
= icmp_hdr(skb
)->un
.gateway
;
749 __be32 old_gw
= ip_hdr(skb
)->saddr
;
750 struct net_device
*dev
= skb
->dev
;
751 struct in_device
*in_dev
;
752 struct fib_result res
;
756 switch (icmp_hdr(skb
)->code
& 7) {
758 case ICMP_REDIR_NETTOS
:
759 case ICMP_REDIR_HOST
:
760 case ICMP_REDIR_HOSTTOS
:
767 if (rt
->rt_gateway
!= old_gw
)
770 in_dev
= __in_dev_get_rcu(dev
);
775 if (new_gw
== old_gw
|| !IN_DEV_RX_REDIRECTS(in_dev
) ||
776 ipv4_is_multicast(new_gw
) || ipv4_is_lbcast(new_gw
) ||
777 ipv4_is_zeronet(new_gw
))
778 goto reject_redirect
;
780 if (!IN_DEV_SHARED_MEDIA(in_dev
)) {
781 if (!inet_addr_onlink(in_dev
, new_gw
, old_gw
))
782 goto reject_redirect
;
783 if (IN_DEV_SEC_REDIRECTS(in_dev
) && ip_fib_check_default(new_gw
, dev
))
784 goto reject_redirect
;
786 if (inet_addr_type(net
, new_gw
) != RTN_UNICAST
)
787 goto reject_redirect
;
790 n
= __ipv4_neigh_lookup(rt
->dst
.dev
, new_gw
);
792 n
= neigh_create(&arp_tbl
, &new_gw
, rt
->dst
.dev
);
794 if (!(n
->nud_state
& NUD_VALID
)) {
795 neigh_event_send(n
, NULL
);
797 if (fib_lookup(net
, fl4
, &res
, 0) == 0) {
798 struct fib_nh
*nh
= &FIB_RES_NH(res
);
800 update_or_create_fnhe(nh
, fl4
->daddr
, new_gw
,
802 jiffies
+ ip_rt_gc_timeout
);
805 rt
->dst
.obsolete
= DST_OBSOLETE_KILL
;
806 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE
, n
);
813 #ifdef CONFIG_IP_ROUTE_VERBOSE
814 if (IN_DEV_LOG_MARTIANS(in_dev
)) {
815 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
816 __be32 daddr
= iph
->daddr
;
817 __be32 saddr
= iph
->saddr
;
819 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
820 " Advised path = %pI4 -> %pI4\n",
821 &old_gw
, dev
->name
, &new_gw
,
828 static void ip_do_redirect(struct dst_entry
*dst
, struct sock
*sk
, struct sk_buff
*skb
)
832 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
833 struct net
*net
= dev_net(skb
->dev
);
834 int oif
= skb
->dev
->ifindex
;
835 u8 tos
= RT_TOS(iph
->tos
);
836 u8 prot
= iph
->protocol
;
837 u32 mark
= skb
->mark
;
839 rt
= (struct rtable
*) dst
;
841 __build_flow_key(net
, &fl4
, sk
, iph
, oif
, tos
, prot
, mark
, 0);
842 __ip_do_redirect(rt
, skb
, &fl4
, true);
845 static struct dst_entry
*ipv4_negative_advice(struct dst_entry
*dst
)
847 struct rtable
*rt
= (struct rtable
*)dst
;
848 struct dst_entry
*ret
= dst
;
851 if (dst
->obsolete
> 0) {
854 } else if ((rt
->rt_flags
& RTCF_REDIRECTED
) ||
865 * 1. The first ip_rt_redirect_number redirects are sent
866 * with exponential backoff, then we stop sending them at all,
867 * assuming that the host ignores our redirects.
868 * 2. If we did not see packets requiring redirects
869 * during ip_rt_redirect_silence, we assume that the host
870 * forgot redirected route and start to send redirects again.
872 * This algorithm is much cheaper and more intelligent than dumb load limiting
875 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
876 * and "frag. need" (breaks PMTU discovery) in icmp.c.
879 void ip_rt_send_redirect(struct sk_buff
*skb
)
881 struct rtable
*rt
= skb_rtable(skb
);
882 struct in_device
*in_dev
;
883 struct inet_peer
*peer
;
889 in_dev
= __in_dev_get_rcu(rt
->dst
.dev
);
890 if (!in_dev
|| !IN_DEV_TX_REDIRECTS(in_dev
)) {
894 log_martians
= IN_DEV_LOG_MARTIANS(in_dev
);
895 vif
= l3mdev_master_ifindex_rcu(rt
->dst
.dev
);
898 net
= dev_net(rt
->dst
.dev
);
899 peer
= inet_getpeer_v4(net
->ipv4
.peers
, ip_hdr(skb
)->saddr
, vif
, 1);
901 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
,
902 rt_nexthop(rt
, ip_hdr(skb
)->daddr
));
906 /* No redirected packets during ip_rt_redirect_silence;
907 * reset the algorithm.
909 if (time_after(jiffies
, peer
->rate_last
+ ip_rt_redirect_silence
)) {
910 peer
->rate_tokens
= 0;
911 peer
->n_redirects
= 0;
914 /* Too many ignored redirects; do not send anything
915 * set dst.rate_last to the last seen redirected packet.
917 if (peer
->n_redirects
>= ip_rt_redirect_number
) {
918 peer
->rate_last
= jiffies
;
922 /* Check for load limit; set rate_last to the latest sent
925 if (peer
->rate_tokens
== 0 ||
928 (ip_rt_redirect_load
<< peer
->rate_tokens
)))) {
929 __be32 gw
= rt_nexthop(rt
, ip_hdr(skb
)->daddr
);
931 icmp_send(skb
, ICMP_REDIRECT
, ICMP_REDIR_HOST
, gw
);
932 peer
->rate_last
= jiffies
;
935 #ifdef CONFIG_IP_ROUTE_VERBOSE
937 peer
->rate_tokens
== ip_rt_redirect_number
)
938 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
939 &ip_hdr(skb
)->saddr
, inet_iif(skb
),
940 &ip_hdr(skb
)->daddr
, &gw
);
947 static int ip_error(struct sk_buff
*skb
)
949 struct in_device
*in_dev
= __in_dev_get_rcu(skb
->dev
);
950 struct rtable
*rt
= skb_rtable(skb
);
951 struct inet_peer
*peer
;
957 /* IP on this device is disabled. */
961 net
= dev_net(rt
->dst
.dev
);
962 if (!IN_DEV_FORWARD(in_dev
)) {
963 switch (rt
->dst
.error
) {
965 __IP_INC_STATS(net
, IPSTATS_MIB_INADDRERRORS
);
969 __IP_INC_STATS(net
, IPSTATS_MIB_INNOROUTES
);
975 switch (rt
->dst
.error
) {
980 code
= ICMP_HOST_UNREACH
;
983 code
= ICMP_NET_UNREACH
;
984 __IP_INC_STATS(net
, IPSTATS_MIB_INNOROUTES
);
987 code
= ICMP_PKT_FILTERED
;
991 peer
= inet_getpeer_v4(net
->ipv4
.peers
, ip_hdr(skb
)->saddr
,
992 l3mdev_master_ifindex(skb
->dev
), 1);
997 peer
->rate_tokens
+= now
- peer
->rate_last
;
998 if (peer
->rate_tokens
> ip_rt_error_burst
)
999 peer
->rate_tokens
= ip_rt_error_burst
;
1000 peer
->rate_last
= now
;
1001 if (peer
->rate_tokens
>= ip_rt_error_cost
)
1002 peer
->rate_tokens
-= ip_rt_error_cost
;
1008 icmp_send(skb
, ICMP_DEST_UNREACH
, code
, 0);
1010 out
: kfree_skb(skb
);
1014 static void __ip_rt_update_pmtu(struct rtable
*rt
, struct flowi4
*fl4
, u32 mtu
)
1016 struct dst_entry
*dst
= &rt
->dst
;
1017 struct fib_result res
;
1020 if (ip_mtu_locked(dst
))
1023 if (ipv4_mtu(dst
) < mtu
)
1026 if (mtu
< ip_rt_min_pmtu
) {
1028 mtu
= ip_rt_min_pmtu
;
1031 if (rt
->rt_pmtu
== mtu
&&
1032 time_before(jiffies
, dst
->expires
- ip_rt_mtu_expires
/ 2))
1036 if (fib_lookup(dev_net(dst
->dev
), fl4
, &res
, 0) == 0) {
1037 struct fib_nh
*nh
= &FIB_RES_NH(res
);
1039 update_or_create_fnhe(nh
, fl4
->daddr
, 0, mtu
, lock
,
1040 jiffies
+ ip_rt_mtu_expires
);
1045 static void ip_rt_update_pmtu(struct dst_entry
*dst
, struct sock
*sk
,
1046 struct sk_buff
*skb
, u32 mtu
)
1048 struct rtable
*rt
= (struct rtable
*) dst
;
1051 ip_rt_build_flow_key(&fl4
, sk
, skb
);
1052 __ip_rt_update_pmtu(rt
, &fl4
, mtu
);
1055 void ipv4_update_pmtu(struct sk_buff
*skb
, struct net
*net
, u32 mtu
,
1056 int oif
, u32 mark
, u8 protocol
, int flow_flags
)
1058 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
1063 mark
= IP4_REPLY_MARK(net
, skb
->mark
);
1065 __build_flow_key(net
, &fl4
, NULL
, iph
, oif
,
1066 RT_TOS(iph
->tos
), protocol
, mark
, flow_flags
);
1067 rt
= __ip_route_output_key(net
, &fl4
);
1069 __ip_rt_update_pmtu(rt
, &fl4
, mtu
);
1073 EXPORT_SYMBOL_GPL(ipv4_update_pmtu
);
1075 static void __ipv4_sk_update_pmtu(struct sk_buff
*skb
, struct sock
*sk
, u32 mtu
)
1077 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
1081 __build_flow_key(sock_net(sk
), &fl4
, sk
, iph
, 0, 0, 0, 0, 0);
1083 if (!fl4
.flowi4_mark
)
1084 fl4
.flowi4_mark
= IP4_REPLY_MARK(sock_net(sk
), skb
->mark
);
1086 rt
= __ip_route_output_key(sock_net(sk
), &fl4
);
1088 __ip_rt_update_pmtu(rt
, &fl4
, mtu
);
1093 void ipv4_sk_update_pmtu(struct sk_buff
*skb
, struct sock
*sk
, u32 mtu
)
1095 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
1098 struct dst_entry
*odst
= NULL
;
1100 struct net
*net
= sock_net(sk
);
1104 if (!ip_sk_accept_pmtu(sk
))
1107 odst
= sk_dst_get(sk
);
1109 if (sock_owned_by_user(sk
) || !odst
) {
1110 __ipv4_sk_update_pmtu(skb
, sk
, mtu
);
1114 __build_flow_key(net
, &fl4
, sk
, iph
, 0, 0, 0, 0, 0);
1116 rt
= (struct rtable
*)odst
;
1117 if (odst
->obsolete
&& !odst
->ops
->check(odst
, 0)) {
1118 rt
= ip_route_output_flow(sock_net(sk
), &fl4
, sk
);
1125 __ip_rt_update_pmtu((struct rtable
*) rt
->dst
.path
, &fl4
, mtu
);
1127 if (!dst_check(&rt
->dst
, 0)) {
1129 dst_release(&rt
->dst
);
1131 rt
= ip_route_output_flow(sock_net(sk
), &fl4
, sk
);
1139 sk_dst_set(sk
, &rt
->dst
);
1145 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu
);
1147 void ipv4_redirect(struct sk_buff
*skb
, struct net
*net
,
1148 int oif
, u32 mark
, u8 protocol
, int flow_flags
)
1150 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
1154 __build_flow_key(net
, &fl4
, NULL
, iph
, oif
,
1155 RT_TOS(iph
->tos
), protocol
, mark
, flow_flags
);
1156 rt
= __ip_route_output_key(net
, &fl4
);
1158 __ip_do_redirect(rt
, skb
, &fl4
, false);
1162 EXPORT_SYMBOL_GPL(ipv4_redirect
);
1164 void ipv4_sk_redirect(struct sk_buff
*skb
, struct sock
*sk
)
1166 const struct iphdr
*iph
= (const struct iphdr
*) skb
->data
;
1169 struct net
*net
= sock_net(sk
);
1171 __build_flow_key(net
, &fl4
, sk
, iph
, 0, 0, 0, 0, 0);
1172 rt
= __ip_route_output_key(net
, &fl4
);
1174 __ip_do_redirect(rt
, skb
, &fl4
, false);
1178 EXPORT_SYMBOL_GPL(ipv4_sk_redirect
);
1180 static struct dst_entry
*ipv4_dst_check(struct dst_entry
*dst
, u32 cookie
)
1182 struct rtable
*rt
= (struct rtable
*) dst
;
1184 /* All IPV4 dsts are created with ->obsolete set to the value
1185 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1186 * into this function always.
1188 * When a PMTU/redirect information update invalidates a route,
1189 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1190 * DST_OBSOLETE_DEAD by dst_free().
1192 if (dst
->obsolete
!= DST_OBSOLETE_FORCE_CHK
|| rt_is_expired(rt
))
1197 static void ipv4_link_failure(struct sk_buff
*skb
)
1201 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_HOST_UNREACH
, 0);
1203 rt
= skb_rtable(skb
);
1205 dst_set_expires(&rt
->dst
, 0);
1208 static int ip_rt_bug(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
1210 pr_debug("%s: %pI4 -> %pI4, %s\n",
1211 __func__
, &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1212 skb
->dev
? skb
->dev
->name
: "?");
1219 We do not cache source address of outgoing interface,
1220 because it is used only by IP RR, TS and SRR options,
1221 so that it out of fast path.
1223 BTW remember: "addr" is allowed to be not aligned
1227 void ip_rt_get_source(u8
*addr
, struct sk_buff
*skb
, struct rtable
*rt
)
1231 if (rt_is_output_route(rt
))
1232 src
= ip_hdr(skb
)->saddr
;
1234 struct fib_result res
;
1240 memset(&fl4
, 0, sizeof(fl4
));
1241 fl4
.daddr
= iph
->daddr
;
1242 fl4
.saddr
= iph
->saddr
;
1243 fl4
.flowi4_tos
= RT_TOS(iph
->tos
);
1244 fl4
.flowi4_oif
= rt
->dst
.dev
->ifindex
;
1245 fl4
.flowi4_iif
= skb
->dev
->ifindex
;
1246 fl4
.flowi4_mark
= skb
->mark
;
1249 if (fib_lookup(dev_net(rt
->dst
.dev
), &fl4
, &res
, 0) == 0)
1250 src
= FIB_RES_PREFSRC(dev_net(rt
->dst
.dev
), res
);
1252 src
= inet_select_addr(rt
->dst
.dev
,
1253 rt_nexthop(rt
, iph
->daddr
),
1257 memcpy(addr
, &src
, 4);
1260 #ifdef CONFIG_IP_ROUTE_CLASSID
1261 static void set_class_tag(struct rtable
*rt
, u32 tag
)
1263 if (!(rt
->dst
.tclassid
& 0xFFFF))
1264 rt
->dst
.tclassid
|= tag
& 0xFFFF;
1265 if (!(rt
->dst
.tclassid
& 0xFFFF0000))
1266 rt
->dst
.tclassid
|= tag
& 0xFFFF0000;
1270 static unsigned int ipv4_default_advmss(const struct dst_entry
*dst
)
1272 unsigned int header_size
= sizeof(struct tcphdr
) + sizeof(struct iphdr
);
1273 unsigned int advmss
= max_t(unsigned int, ipv4_mtu(dst
) - header_size
,
1276 return min(advmss
, IPV4_MAX_PMTU
- header_size
);
1279 static unsigned int ipv4_mtu(const struct dst_entry
*dst
)
1281 const struct rtable
*rt
= (const struct rtable
*) dst
;
1282 unsigned int mtu
= rt
->rt_pmtu
;
1284 if (!mtu
|| time_after_eq(jiffies
, rt
->dst
.expires
))
1285 mtu
= dst_metric_raw(dst
, RTAX_MTU
);
1290 mtu
= READ_ONCE(dst
->dev
->mtu
);
1292 if (unlikely(ip_mtu_locked(dst
))) {
1293 if (rt
->rt_uses_gateway
&& mtu
> 576)
1297 mtu
= min_t(unsigned int, mtu
, IP_MAX_MTU
);
1299 return mtu
- lwtunnel_headroom(dst
->lwtstate
, mtu
);
1302 static void ip_del_fnhe(struct fib_nh
*nh
, __be32 daddr
)
1304 struct fnhe_hash_bucket
*hash
;
1305 struct fib_nh_exception
*fnhe
, __rcu
**fnhe_p
;
1306 u32 hval
= fnhe_hashfun(daddr
);
1308 spin_lock_bh(&fnhe_lock
);
1310 hash
= rcu_dereference_protected(nh
->nh_exceptions
,
1311 lockdep_is_held(&fnhe_lock
));
1314 fnhe_p
= &hash
->chain
;
1315 fnhe
= rcu_dereference_protected(*fnhe_p
, lockdep_is_held(&fnhe_lock
));
1317 if (fnhe
->fnhe_daddr
== daddr
) {
1318 rcu_assign_pointer(*fnhe_p
, rcu_dereference_protected(
1319 fnhe
->fnhe_next
, lockdep_is_held(&fnhe_lock
)));
1320 /* set fnhe_daddr to 0 to ensure it won't bind with
1321 * new dsts in rt_bind_exception().
1323 fnhe
->fnhe_daddr
= 0;
1324 fnhe_flush_routes(fnhe
);
1325 kfree_rcu(fnhe
, rcu
);
1328 fnhe_p
= &fnhe
->fnhe_next
;
1329 fnhe
= rcu_dereference_protected(fnhe
->fnhe_next
,
1330 lockdep_is_held(&fnhe_lock
));
1333 spin_unlock_bh(&fnhe_lock
);
1336 static struct fib_nh_exception
*find_exception(struct fib_nh
*nh
, __be32 daddr
)
1338 struct fnhe_hash_bucket
*hash
= rcu_dereference(nh
->nh_exceptions
);
1339 struct fib_nh_exception
*fnhe
;
1345 hval
= fnhe_hashfun(daddr
);
1347 for (fnhe
= rcu_dereference(hash
[hval
].chain
); fnhe
;
1348 fnhe
= rcu_dereference(fnhe
->fnhe_next
)) {
1349 if (fnhe
->fnhe_daddr
== daddr
) {
1350 if (fnhe
->fnhe_expires
&&
1351 time_after(jiffies
, fnhe
->fnhe_expires
)) {
1352 ip_del_fnhe(nh
, daddr
);
1361 static bool rt_bind_exception(struct rtable
*rt
, struct fib_nh_exception
*fnhe
,
1362 __be32 daddr
, const bool do_cache
)
1366 spin_lock_bh(&fnhe_lock
);
1368 if (daddr
== fnhe
->fnhe_daddr
) {
1369 struct rtable __rcu
**porig
;
1370 struct rtable
*orig
;
1371 int genid
= fnhe_genid(dev_net(rt
->dst
.dev
));
1373 if (rt_is_input_route(rt
))
1374 porig
= &fnhe
->fnhe_rth_input
;
1376 porig
= &fnhe
->fnhe_rth_output
;
1377 orig
= rcu_dereference(*porig
);
1379 if (fnhe
->fnhe_genid
!= genid
) {
1380 fnhe
->fnhe_genid
= genid
;
1382 fnhe
->fnhe_pmtu
= 0;
1383 fnhe
->fnhe_expires
= 0;
1384 fnhe_flush_routes(fnhe
);
1387 fill_route_from_fnhe(rt
, fnhe
);
1388 if (!rt
->rt_gateway
)
1389 rt
->rt_gateway
= daddr
;
1393 rcu_assign_pointer(*porig
, rt
);
1395 dst_dev_put(&orig
->dst
);
1396 dst_release(&orig
->dst
);
1401 fnhe
->fnhe_stamp
= jiffies
;
1403 spin_unlock_bh(&fnhe_lock
);
1408 static bool rt_cache_route(struct fib_nh
*nh
, struct rtable
*rt
)
1410 struct rtable
*orig
, *prev
, **p
;
1413 if (rt_is_input_route(rt
)) {
1414 p
= (struct rtable
**)&nh
->nh_rth_input
;
1416 p
= (struct rtable
**)raw_cpu_ptr(nh
->nh_pcpu_rth_output
);
1420 /* hold dst before doing cmpxchg() to avoid race condition
1424 prev
= cmpxchg(p
, orig
, rt
);
1427 dst_dev_put(&orig
->dst
);
1428 dst_release(&orig
->dst
);
1431 dst_release(&rt
->dst
);
1438 struct uncached_list
{
1440 struct list_head head
;
1443 static DEFINE_PER_CPU_ALIGNED(struct uncached_list
, rt_uncached_list
);
1445 static void rt_add_uncached_list(struct rtable
*rt
)
1447 struct uncached_list
*ul
= raw_cpu_ptr(&rt_uncached_list
);
1449 rt
->rt_uncached_list
= ul
;
1451 spin_lock_bh(&ul
->lock
);
1452 list_add_tail(&rt
->rt_uncached
, &ul
->head
);
1453 spin_unlock_bh(&ul
->lock
);
1456 static void ipv4_dst_destroy(struct dst_entry
*dst
)
1458 struct dst_metrics
*p
= (struct dst_metrics
*)DST_METRICS_PTR(dst
);
1459 struct rtable
*rt
= (struct rtable
*) dst
;
1461 if (p
!= &dst_default_metrics
&& refcount_dec_and_test(&p
->refcnt
))
1464 if (!list_empty(&rt
->rt_uncached
)) {
1465 struct uncached_list
*ul
= rt
->rt_uncached_list
;
1467 spin_lock_bh(&ul
->lock
);
1468 list_del(&rt
->rt_uncached
);
1469 spin_unlock_bh(&ul
->lock
);
1473 void rt_flush_dev(struct net_device
*dev
)
1475 struct net
*net
= dev_net(dev
);
1479 for_each_possible_cpu(cpu
) {
1480 struct uncached_list
*ul
= &per_cpu(rt_uncached_list
, cpu
);
1482 spin_lock_bh(&ul
->lock
);
1483 list_for_each_entry(rt
, &ul
->head
, rt_uncached
) {
1484 if (rt
->dst
.dev
!= dev
)
1486 rt
->dst
.dev
= net
->loopback_dev
;
1487 dev_hold(rt
->dst
.dev
);
1490 spin_unlock_bh(&ul
->lock
);
1494 static bool rt_cache_valid(const struct rtable
*rt
)
1497 rt
->dst
.obsolete
== DST_OBSOLETE_FORCE_CHK
&&
1501 static void rt_set_nexthop(struct rtable
*rt
, __be32 daddr
,
1502 const struct fib_result
*res
,
1503 struct fib_nh_exception
*fnhe
,
1504 struct fib_info
*fi
, u16 type
, u32 itag
,
1505 const bool do_cache
)
1507 bool cached
= false;
1510 struct fib_nh
*nh
= &FIB_RES_NH(*res
);
1512 if (nh
->nh_gw
&& nh
->nh_scope
== RT_SCOPE_LINK
) {
1513 rt
->rt_gateway
= nh
->nh_gw
;
1514 rt
->rt_uses_gateway
= 1;
1516 dst_init_metrics(&rt
->dst
, fi
->fib_metrics
->metrics
, true);
1517 if (fi
->fib_metrics
!= &dst_default_metrics
) {
1518 rt
->dst
._metrics
|= DST_METRICS_REFCOUNTED
;
1519 refcount_inc(&fi
->fib_metrics
->refcnt
);
1521 #ifdef CONFIG_IP_ROUTE_CLASSID
1522 rt
->dst
.tclassid
= nh
->nh_tclassid
;
1524 rt
->dst
.lwtstate
= lwtstate_get(nh
->nh_lwtstate
);
1526 cached
= rt_bind_exception(rt
, fnhe
, daddr
, do_cache
);
1528 cached
= rt_cache_route(nh
, rt
);
1529 if (unlikely(!cached
)) {
1530 /* Routes we intend to cache in nexthop exception or
1531 * FIB nexthop have the DST_NOCACHE bit clear.
1532 * However, if we are unsuccessful at storing this
1533 * route into the cache we really need to set it.
1535 if (!rt
->rt_gateway
)
1536 rt
->rt_gateway
= daddr
;
1537 rt_add_uncached_list(rt
);
1540 rt_add_uncached_list(rt
);
1542 #ifdef CONFIG_IP_ROUTE_CLASSID
1543 #ifdef CONFIG_IP_MULTIPLE_TABLES
1544 set_class_tag(rt
, res
->tclassid
);
1546 set_class_tag(rt
, itag
);
1550 struct rtable
*rt_dst_alloc(struct net_device
*dev
,
1551 unsigned int flags
, u16 type
,
1552 bool nopolicy
, bool noxfrm
, bool will_cache
)
1556 rt
= dst_alloc(&ipv4_dst_ops
, dev
, 1, DST_OBSOLETE_FORCE_CHK
,
1557 (will_cache
? 0 : DST_HOST
) |
1558 (nopolicy
? DST_NOPOLICY
: 0) |
1559 (noxfrm
? DST_NOXFRM
: 0));
1562 rt
->rt_genid
= rt_genid_ipv4(dev_net(dev
));
1563 rt
->rt_flags
= flags
;
1565 rt
->rt_is_input
= 0;
1568 rt
->rt_mtu_locked
= 0;
1570 rt
->rt_uses_gateway
= 0;
1571 rt
->rt_table_id
= 0;
1572 INIT_LIST_HEAD(&rt
->rt_uncached
);
1574 rt
->dst
.output
= ip_output
;
1575 if (flags
& RTCF_LOCAL
)
1576 rt
->dst
.input
= ip_local_deliver
;
1581 EXPORT_SYMBOL(rt_dst_alloc
);
1583 /* called in rcu_read_lock() section */
1584 int ip_mc_validate_source(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
1585 u8 tos
, struct net_device
*dev
,
1586 struct in_device
*in_dev
, u32
*itag
)
1590 /* Primary sanity checks. */
1594 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
) ||
1595 skb
->protocol
!= htons(ETH_P_IP
))
1598 if (ipv4_is_loopback(saddr
) && !IN_DEV_ROUTE_LOCALNET(in_dev
))
1601 if (ipv4_is_zeronet(saddr
)) {
1602 if (!ipv4_is_local_multicast(daddr
))
1605 err
= fib_validate_source(skb
, saddr
, 0, tos
, 0, dev
,
1613 /* called in rcu_read_lock() section */
1614 static int ip_route_input_mc(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
1615 u8 tos
, struct net_device
*dev
, int our
)
1617 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
1618 unsigned int flags
= RTCF_MULTICAST
;
1623 err
= ip_mc_validate_source(skb
, daddr
, saddr
, tos
, dev
, in_dev
, &itag
);
1628 flags
|= RTCF_LOCAL
;
1630 rth
= rt_dst_alloc(dev_net(dev
)->loopback_dev
, flags
, RTN_MULTICAST
,
1631 IN_DEV_CONF_GET(in_dev
, NOPOLICY
), false, false);
1635 #ifdef CONFIG_IP_ROUTE_CLASSID
1636 rth
->dst
.tclassid
= itag
;
1638 rth
->dst
.output
= ip_rt_bug
;
1639 rth
->rt_is_input
= 1;
1641 #ifdef CONFIG_IP_MROUTE
1642 if (!ipv4_is_local_multicast(daddr
) && IN_DEV_MFORWARD(in_dev
))
1643 rth
->dst
.input
= ip_mr_input
;
1645 RT_CACHE_STAT_INC(in_slow_mc
);
1647 skb_dst_set(skb
, &rth
->dst
);
1652 static void ip_handle_martian_source(struct net_device
*dev
,
1653 struct in_device
*in_dev
,
1654 struct sk_buff
*skb
,
1658 RT_CACHE_STAT_INC(in_martian_src
);
1659 #ifdef CONFIG_IP_ROUTE_VERBOSE
1660 if (IN_DEV_LOG_MARTIANS(in_dev
) && net_ratelimit()) {
1662 * RFC1812 recommendation, if source is martian,
1663 * the only hint is MAC header.
1665 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1666 &daddr
, &saddr
, dev
->name
);
1667 if (dev
->hard_header_len
&& skb_mac_header_was_set(skb
)) {
1668 print_hex_dump(KERN_WARNING
, "ll header: ",
1669 DUMP_PREFIX_OFFSET
, 16, 1,
1670 skb_mac_header(skb
),
1671 dev
->hard_header_len
, true);
1677 static void set_lwt_redirect(struct rtable
*rth
)
1679 if (lwtunnel_output_redirect(rth
->dst
.lwtstate
)) {
1680 rth
->dst
.lwtstate
->orig_output
= rth
->dst
.output
;
1681 rth
->dst
.output
= lwtunnel_output
;
1684 if (lwtunnel_input_redirect(rth
->dst
.lwtstate
)) {
1685 rth
->dst
.lwtstate
->orig_input
= rth
->dst
.input
;
1686 rth
->dst
.input
= lwtunnel_input
;
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;
1763 rth
->rt_table_id
= res
->table
->tb_id
;
1764 RT_CACHE_STAT_INC(in_slow_tot
);
1766 rth
->dst
.input
= ip_forward
;
1768 rt_set_nexthop(rth
, daddr
, res
, fnhe
, res
->fi
, res
->type
, itag
,
1770 set_lwt_redirect(rth
);
1771 skb_dst_set(skb
, &rth
->dst
);
1778 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1779 /* To make ICMP packets follow the right flow, the multipath hash is
1780 * calculated from the inner IP addresses.
1782 static void ip_multipath_l3_keys(const struct sk_buff
*skb
,
1783 struct flow_keys
*hash_keys
)
1785 const struct iphdr
*outer_iph
= ip_hdr(skb
);
1786 const struct iphdr
*inner_iph
;
1787 const struct icmphdr
*icmph
;
1788 struct iphdr _inner_iph
;
1789 struct icmphdr _icmph
;
1791 hash_keys
->addrs
.v4addrs
.src
= outer_iph
->saddr
;
1792 hash_keys
->addrs
.v4addrs
.dst
= outer_iph
->daddr
;
1793 if (likely(outer_iph
->protocol
!= IPPROTO_ICMP
))
1796 if (unlikely((outer_iph
->frag_off
& htons(IP_OFFSET
)) != 0))
1799 icmph
= skb_header_pointer(skb
, outer_iph
->ihl
* 4, sizeof(_icmph
),
1804 if (icmph
->type
!= ICMP_DEST_UNREACH
&&
1805 icmph
->type
!= ICMP_REDIRECT
&&
1806 icmph
->type
!= ICMP_TIME_EXCEEDED
&&
1807 icmph
->type
!= ICMP_PARAMETERPROB
)
1810 inner_iph
= skb_header_pointer(skb
,
1811 outer_iph
->ihl
* 4 + sizeof(_icmph
),
1812 sizeof(_inner_iph
), &_inner_iph
);
1815 hash_keys
->addrs
.v4addrs
.src
= inner_iph
->saddr
;
1816 hash_keys
->addrs
.v4addrs
.dst
= inner_iph
->daddr
;
1819 /* if skb is set it will be used and fl4 can be NULL */
1820 int fib_multipath_hash(const struct fib_info
*fi
, const struct flowi4
*fl4
,
1821 const struct sk_buff
*skb
)
1823 struct net
*net
= fi
->fib_net
;
1824 struct flow_keys hash_keys
;
1827 switch (net
->ipv4
.sysctl_fib_multipath_hash_policy
) {
1829 memset(&hash_keys
, 0, sizeof(hash_keys
));
1830 hash_keys
.control
.addr_type
= FLOW_DISSECTOR_KEY_IPV4_ADDRS
;
1832 ip_multipath_l3_keys(skb
, &hash_keys
);
1834 hash_keys
.addrs
.v4addrs
.src
= fl4
->saddr
;
1835 hash_keys
.addrs
.v4addrs
.dst
= fl4
->daddr
;
1839 /* skb is currently provided only when forwarding */
1841 unsigned int flag
= FLOW_DISSECTOR_F_STOP_AT_ENCAP
;
1842 struct flow_keys keys
;
1844 /* short-circuit if we already have L4 hash present */
1846 return skb_get_hash_raw(skb
) >> 1;
1847 memset(&hash_keys
, 0, sizeof(hash_keys
));
1848 skb_flow_dissect_flow_keys(skb
, &keys
, flag
);
1850 hash_keys
.control
.addr_type
= FLOW_DISSECTOR_KEY_IPV4_ADDRS
;
1851 hash_keys
.addrs
.v4addrs
.src
= keys
.addrs
.v4addrs
.src
;
1852 hash_keys
.addrs
.v4addrs
.dst
= keys
.addrs
.v4addrs
.dst
;
1853 hash_keys
.ports
.src
= keys
.ports
.src
;
1854 hash_keys
.ports
.dst
= keys
.ports
.dst
;
1855 hash_keys
.basic
.ip_proto
= keys
.basic
.ip_proto
;
1857 memset(&hash_keys
, 0, sizeof(hash_keys
));
1858 hash_keys
.control
.addr_type
= FLOW_DISSECTOR_KEY_IPV4_ADDRS
;
1859 hash_keys
.addrs
.v4addrs
.src
= fl4
->saddr
;
1860 hash_keys
.addrs
.v4addrs
.dst
= fl4
->daddr
;
1861 hash_keys
.ports
.src
= fl4
->fl4_sport
;
1862 hash_keys
.ports
.dst
= fl4
->fl4_dport
;
1863 hash_keys
.basic
.ip_proto
= fl4
->flowi4_proto
;
1867 mhash
= flow_hash_from_keys(&hash_keys
);
1871 EXPORT_SYMBOL_GPL(fib_multipath_hash
);
1872 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1874 static int ip_mkroute_input(struct sk_buff
*skb
,
1875 struct fib_result
*res
,
1876 struct in_device
*in_dev
,
1877 __be32 daddr
, __be32 saddr
, u32 tos
)
1879 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1880 if (res
->fi
&& res
->fi
->fib_nhs
> 1) {
1881 int h
= fib_multipath_hash(res
->fi
, NULL
, skb
);
1883 fib_select_multipath(res
, h
);
1887 /* create a routing cache entry */
1888 return __mkroute_input(skb
, res
, in_dev
, daddr
, saddr
, tos
);
1892 * NOTE. We drop all the packets that has local source
1893 * addresses, because every properly looped back packet
1894 * must have correct destination already attached by output routine.
1896 * Such approach solves two big problems:
1897 * 1. Not simplex devices are handled properly.
1898 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1899 * called with rcu_read_lock()
1902 static int ip_route_input_slow(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
1903 u8 tos
, struct net_device
*dev
,
1904 struct fib_result
*res
)
1906 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
1907 struct ip_tunnel_info
*tun_info
;
1909 unsigned int flags
= 0;
1913 struct net
*net
= dev_net(dev
);
1916 /* IP on this device is disabled. */
1921 /* Check for the most weird martians, which can be not detected
1925 tun_info
= skb_tunnel_info(skb
);
1926 if (tun_info
&& !(tun_info
->mode
& IP_TUNNEL_INFO_TX
))
1927 fl4
.flowi4_tun_key
.tun_id
= tun_info
->key
.tun_id
;
1929 fl4
.flowi4_tun_key
.tun_id
= 0;
1932 if (ipv4_is_multicast(saddr
) || ipv4_is_lbcast(saddr
))
1933 goto martian_source
;
1937 if (ipv4_is_lbcast(daddr
) || (saddr
== 0 && daddr
== 0))
1940 /* Accept zero addresses only to limited broadcast;
1941 * I even do not know to fix it or not. Waiting for complains :-)
1943 if (ipv4_is_zeronet(saddr
))
1944 goto martian_source
;
1946 if (ipv4_is_zeronet(daddr
))
1947 goto martian_destination
;
1949 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1950 * and call it once if daddr or/and saddr are loopback addresses
1952 if (ipv4_is_loopback(daddr
)) {
1953 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev
, net
))
1954 goto martian_destination
;
1955 } else if (ipv4_is_loopback(saddr
)) {
1956 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev
, net
))
1957 goto martian_source
;
1961 * Now we are ready to route packet.
1964 fl4
.flowi4_iif
= dev
->ifindex
;
1965 fl4
.flowi4_mark
= skb
->mark
;
1966 fl4
.flowi4_tos
= tos
;
1967 fl4
.flowi4_scope
= RT_SCOPE_UNIVERSE
;
1968 fl4
.flowi4_flags
= 0;
1971 fl4
.flowi4_uid
= sock_net_uid(net
, NULL
);
1972 err
= fib_lookup(net
, &fl4
, res
, 0);
1974 if (!IN_DEV_FORWARD(in_dev
))
1975 err
= -EHOSTUNREACH
;
1979 if (res
->type
== RTN_BROADCAST
)
1982 if (res
->type
== RTN_LOCAL
) {
1983 err
= fib_validate_source(skb
, saddr
, daddr
, tos
,
1984 0, dev
, in_dev
, &itag
);
1986 goto martian_source
;
1990 if (!IN_DEV_FORWARD(in_dev
)) {
1991 err
= -EHOSTUNREACH
;
1994 if (res
->type
!= RTN_UNICAST
)
1995 goto martian_destination
;
1997 err
= ip_mkroute_input(skb
, res
, in_dev
, daddr
, saddr
, tos
);
2001 if (skb
->protocol
!= htons(ETH_P_IP
))
2004 if (!ipv4_is_zeronet(saddr
)) {
2005 err
= fib_validate_source(skb
, saddr
, 0, tos
, 0, dev
,
2008 goto martian_source
;
2010 flags
|= RTCF_BROADCAST
;
2011 res
->type
= RTN_BROADCAST
;
2012 RT_CACHE_STAT_INC(in_brd
);
2018 rth
= rcu_dereference(FIB_RES_NH(*res
).nh_rth_input
);
2019 if (rt_cache_valid(rth
)) {
2020 skb_dst_set_noref(skb
, &rth
->dst
);
2028 rth
= rt_dst_alloc(l3mdev_master_dev_rcu(dev
) ? : net
->loopback_dev
,
2029 flags
| RTCF_LOCAL
, res
->type
,
2030 IN_DEV_CONF_GET(in_dev
, NOPOLICY
), false, do_cache
);
2034 rth
->dst
.output
= ip_rt_bug
;
2035 #ifdef CONFIG_IP_ROUTE_CLASSID
2036 rth
->dst
.tclassid
= itag
;
2038 rth
->rt_is_input
= 1;
2040 rth
->rt_table_id
= res
->table
->tb_id
;
2042 RT_CACHE_STAT_INC(in_slow_tot
);
2043 if (res
->type
== RTN_UNREACHABLE
) {
2044 rth
->dst
.input
= ip_error
;
2045 rth
->dst
.error
= -err
;
2046 rth
->rt_flags
&= ~RTCF_LOCAL
;
2050 struct fib_nh
*nh
= &FIB_RES_NH(*res
);
2052 rth
->dst
.lwtstate
= lwtstate_get(nh
->nh_lwtstate
);
2053 if (lwtunnel_input_redirect(rth
->dst
.lwtstate
)) {
2054 WARN_ON(rth
->dst
.input
== lwtunnel_input
);
2055 rth
->dst
.lwtstate
->orig_input
= rth
->dst
.input
;
2056 rth
->dst
.input
= lwtunnel_input
;
2059 if (unlikely(!rt_cache_route(nh
, rth
)))
2060 rt_add_uncached_list(rth
);
2062 skb_dst_set(skb
, &rth
->dst
);
2067 RT_CACHE_STAT_INC(in_no_route
);
2068 res
->type
= RTN_UNREACHABLE
;
2074 * Do not cache martian addresses: they should be logged (RFC1812)
2076 martian_destination
:
2077 RT_CACHE_STAT_INC(in_martian_dst
);
2078 #ifdef CONFIG_IP_ROUTE_VERBOSE
2079 if (IN_DEV_LOG_MARTIANS(in_dev
))
2080 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2081 &daddr
, &saddr
, dev
->name
);
2093 ip_handle_martian_source(dev
, in_dev
, skb
, daddr
, saddr
);
2097 int ip_route_input_noref(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2098 u8 tos
, struct net_device
*dev
)
2100 struct fib_result res
;
2103 tos
&= IPTOS_RT_MASK
;
2105 err
= ip_route_input_rcu(skb
, daddr
, saddr
, tos
, dev
, &res
);
2110 EXPORT_SYMBOL(ip_route_input_noref
);
2112 /* called with rcu_read_lock held */
2113 int ip_route_input_rcu(struct sk_buff
*skb
, __be32 daddr
, __be32 saddr
,
2114 u8 tos
, struct net_device
*dev
, struct fib_result
*res
)
2116 /* Multicast recognition logic is moved from route cache to here.
2117 The problem was that too many Ethernet cards have broken/missing
2118 hardware multicast filters :-( As result the host on multicasting
2119 network acquires a lot of useless route cache entries, sort of
2120 SDR messages from all the world. Now we try to get rid of them.
2121 Really, provided software IP multicast filter is organized
2122 reasonably (at least, hashed), it does not result in a slowdown
2123 comparing with route cache reject entries.
2124 Note, that multicast routers are not affected, because
2125 route cache entry is created eventually.
2127 if (ipv4_is_multicast(daddr
)) {
2128 struct in_device
*in_dev
= __in_dev_get_rcu(dev
);
2134 our
= ip_check_mc_rcu(in_dev
, daddr
, saddr
,
2135 ip_hdr(skb
)->protocol
);
2137 /* check l3 master if no match yet */
2138 if (!our
&& netif_is_l3_slave(dev
)) {
2139 struct in_device
*l3_in_dev
;
2141 l3_in_dev
= __in_dev_get_rcu(skb
->dev
);
2143 our
= ip_check_mc_rcu(l3_in_dev
, daddr
, saddr
,
2144 ip_hdr(skb
)->protocol
);
2148 #ifdef CONFIG_IP_MROUTE
2150 (!ipv4_is_local_multicast(daddr
) &&
2151 IN_DEV_MFORWARD(in_dev
))
2154 err
= ip_route_input_mc(skb
, daddr
, saddr
,
2160 return ip_route_input_slow(skb
, daddr
, saddr
, tos
, dev
, res
);
2163 /* called with rcu_read_lock() */
2164 static struct rtable
*__mkroute_output(const struct fib_result
*res
,
2165 const struct flowi4
*fl4
, int orig_oif
,
2166 struct net_device
*dev_out
,
2169 struct fib_info
*fi
= res
->fi
;
2170 struct fib_nh_exception
*fnhe
;
2171 struct in_device
*in_dev
;
2172 u16 type
= res
->type
;
2176 in_dev
= __in_dev_get_rcu(dev_out
);
2178 return ERR_PTR(-EINVAL
);
2180 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev
)))
2181 if (ipv4_is_loopback(fl4
->saddr
) &&
2182 !(dev_out
->flags
& IFF_LOOPBACK
) &&
2183 !netif_is_l3_master(dev_out
))
2184 return ERR_PTR(-EINVAL
);
2186 if (ipv4_is_lbcast(fl4
->daddr
))
2187 type
= RTN_BROADCAST
;
2188 else if (ipv4_is_multicast(fl4
->daddr
))
2189 type
= RTN_MULTICAST
;
2190 else if (ipv4_is_zeronet(fl4
->daddr
))
2191 return ERR_PTR(-EINVAL
);
2193 if (dev_out
->flags
& IFF_LOOPBACK
)
2194 flags
|= RTCF_LOCAL
;
2197 if (type
== RTN_BROADCAST
) {
2198 flags
|= RTCF_BROADCAST
| RTCF_LOCAL
;
2200 } else if (type
== RTN_MULTICAST
) {
2201 flags
|= RTCF_MULTICAST
| RTCF_LOCAL
;
2202 if (!ip_check_mc_rcu(in_dev
, fl4
->daddr
, fl4
->saddr
,
2204 flags
&= ~RTCF_LOCAL
;
2207 /* If multicast route do not exist use
2208 * default one, but do not gateway in this case.
2211 if (fi
&& res
->prefixlen
< 4)
2213 } else if ((type
== RTN_LOCAL
) && (orig_oif
!= 0) &&
2214 (orig_oif
!= dev_out
->ifindex
)) {
2215 /* For local routes that require a particular output interface
2216 * we do not want to cache the result. Caching the result
2217 * causes incorrect behaviour when there are multiple source
2218 * addresses on the interface, the end result being that if the
2219 * intended recipient is waiting on that interface for the
2220 * packet he won't receive it because it will be delivered on
2221 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2222 * be set to the loopback interface as well.
2228 do_cache
&= fi
!= NULL
;
2230 struct rtable __rcu
**prth
;
2231 struct fib_nh
*nh
= &FIB_RES_NH(*res
);
2233 fnhe
= find_exception(nh
, fl4
->daddr
);
2237 prth
= &fnhe
->fnhe_rth_output
;
2239 if (unlikely(fl4
->flowi4_flags
&
2240 FLOWI_FLAG_KNOWN_NH
&&
2242 nh
->nh_scope
== RT_SCOPE_LINK
))) {
2246 prth
= raw_cpu_ptr(nh
->nh_pcpu_rth_output
);
2248 rth
= rcu_dereference(*prth
);
2249 if (rt_cache_valid(rth
) && dst_hold_safe(&rth
->dst
))
2254 rth
= rt_dst_alloc(dev_out
, flags
, type
,
2255 IN_DEV_CONF_GET(in_dev
, NOPOLICY
),
2256 IN_DEV_CONF_GET(in_dev
, NOXFRM
),
2259 return ERR_PTR(-ENOBUFS
);
2261 rth
->rt_iif
= orig_oif
;
2263 rth
->rt_table_id
= res
->table
->tb_id
;
2265 RT_CACHE_STAT_INC(out_slow_tot
);
2267 if (flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) {
2268 if (flags
& RTCF_LOCAL
&&
2269 !(dev_out
->flags
& IFF_LOOPBACK
)) {
2270 rth
->dst
.output
= ip_mc_output
;
2271 RT_CACHE_STAT_INC(out_slow_mc
);
2273 #ifdef CONFIG_IP_MROUTE
2274 if (type
== RTN_MULTICAST
) {
2275 if (IN_DEV_MFORWARD(in_dev
) &&
2276 !ipv4_is_local_multicast(fl4
->daddr
)) {
2277 rth
->dst
.input
= ip_mr_input
;
2278 rth
->dst
.output
= ip_mc_output
;
2284 rt_set_nexthop(rth
, fl4
->daddr
, res
, fnhe
, fi
, type
, 0, do_cache
);
2285 set_lwt_redirect(rth
);
2291 * Major route resolver routine.
2294 struct rtable
*ip_route_output_key_hash(struct net
*net
, struct flowi4
*fl4
,
2295 const struct sk_buff
*skb
)
2297 __u8 tos
= RT_FL_TOS(fl4
);
2298 struct fib_result res
= {
2306 fl4
->flowi4_iif
= LOOPBACK_IFINDEX
;
2307 fl4
->flowi4_tos
= tos
& IPTOS_RT_MASK
;
2308 fl4
->flowi4_scope
= ((tos
& RTO_ONLINK
) ?
2309 RT_SCOPE_LINK
: RT_SCOPE_UNIVERSE
);
2312 rth
= ip_route_output_key_hash_rcu(net
, fl4
, &res
, skb
);
2317 EXPORT_SYMBOL_GPL(ip_route_output_key_hash
);
2319 struct rtable
*ip_route_output_key_hash_rcu(struct net
*net
, struct flowi4
*fl4
,
2320 struct fib_result
*res
,
2321 const struct sk_buff
*skb
)
2323 struct net_device
*dev_out
= NULL
;
2324 int orig_oif
= fl4
->flowi4_oif
;
2325 unsigned int flags
= 0;
2327 int err
= -ENETUNREACH
;
2330 rth
= ERR_PTR(-EINVAL
);
2331 if (ipv4_is_multicast(fl4
->saddr
) ||
2332 ipv4_is_lbcast(fl4
->saddr
) ||
2333 ipv4_is_zeronet(fl4
->saddr
))
2336 /* I removed check for oif == dev_out->oif here.
2337 It was wrong for two reasons:
2338 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2339 is assigned to multiple interfaces.
2340 2. Moreover, we are allowed to send packets with saddr
2341 of another iface. --ANK
2344 if (fl4
->flowi4_oif
== 0 &&
2345 (ipv4_is_multicast(fl4
->daddr
) ||
2346 ipv4_is_lbcast(fl4
->daddr
))) {
2347 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2348 dev_out
= __ip_dev_find(net
, fl4
->saddr
, false);
2352 /* Special hack: user can direct multicasts
2353 and limited broadcast via necessary interface
2354 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2355 This hack is not just for fun, it allows
2356 vic,vat and friends to work.
2357 They bind socket to loopback, set ttl to zero
2358 and expect that it will work.
2359 From the viewpoint of routing cache they are broken,
2360 because we are not allowed to build multicast path
2361 with loopback source addr (look, routing cache
2362 cannot know, that ttl is zero, so that packet
2363 will not leave this host and route is valid).
2364 Luckily, this hack is good workaround.
2367 fl4
->flowi4_oif
= dev_out
->ifindex
;
2371 if (!(fl4
->flowi4_flags
& FLOWI_FLAG_ANYSRC
)) {
2372 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2373 if (!__ip_dev_find(net
, fl4
->saddr
, false))
2379 if (fl4
->flowi4_oif
) {
2380 dev_out
= dev_get_by_index_rcu(net
, fl4
->flowi4_oif
);
2381 rth
= ERR_PTR(-ENODEV
);
2385 /* RACE: Check return value of inet_select_addr instead. */
2386 if (!(dev_out
->flags
& IFF_UP
) || !__in_dev_get_rcu(dev_out
)) {
2387 rth
= ERR_PTR(-ENETUNREACH
);
2390 if (ipv4_is_local_multicast(fl4
->daddr
) ||
2391 ipv4_is_lbcast(fl4
->daddr
) ||
2392 fl4
->flowi4_proto
== IPPROTO_IGMP
) {
2394 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2399 if (ipv4_is_multicast(fl4
->daddr
))
2400 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2402 else if (!fl4
->daddr
)
2403 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2409 fl4
->daddr
= fl4
->saddr
;
2411 fl4
->daddr
= fl4
->saddr
= htonl(INADDR_LOOPBACK
);
2412 dev_out
= net
->loopback_dev
;
2413 fl4
->flowi4_oif
= LOOPBACK_IFINDEX
;
2414 res
->type
= RTN_LOCAL
;
2415 flags
|= RTCF_LOCAL
;
2419 err
= fib_lookup(net
, fl4
, res
, 0);
2423 if (fl4
->flowi4_oif
&&
2424 (ipv4_is_multicast(fl4
->daddr
) ||
2425 !netif_index_is_l3_master(net
, fl4
->flowi4_oif
))) {
2426 /* Apparently, routing tables are wrong. Assume,
2427 that the destination is on link.
2430 Because we are allowed to send to iface
2431 even if it has NO routes and NO assigned
2432 addresses. When oif is specified, routing
2433 tables are looked up with only one purpose:
2434 to catch if destination is gatewayed, rather than
2435 direct. Moreover, if MSG_DONTROUTE is set,
2436 we send packet, ignoring both routing tables
2437 and ifaddr state. --ANK
2440 We could make it even if oif is unknown,
2441 likely IPv6, but we do not.
2444 if (fl4
->saddr
== 0)
2445 fl4
->saddr
= inet_select_addr(dev_out
, 0,
2447 res
->type
= RTN_UNICAST
;
2454 if (res
->type
== RTN_LOCAL
) {
2456 if (res
->fi
->fib_prefsrc
)
2457 fl4
->saddr
= res
->fi
->fib_prefsrc
;
2459 fl4
->saddr
= fl4
->daddr
;
2462 /* L3 master device is the loopback for that domain */
2463 dev_out
= l3mdev_master_dev_rcu(FIB_RES_DEV(*res
)) ? :
2466 /* make sure orig_oif points to fib result device even
2467 * though packet rx/tx happens over loopback or l3mdev
2469 orig_oif
= FIB_RES_OIF(*res
);
2471 fl4
->flowi4_oif
= dev_out
->ifindex
;
2472 flags
|= RTCF_LOCAL
;
2476 fib_select_path(net
, res
, fl4
, skb
);
2478 dev_out
= FIB_RES_DEV(*res
);
2479 fl4
->flowi4_oif
= dev_out
->ifindex
;
2483 rth
= __mkroute_output(res
, fl4
, orig_oif
, dev_out
, flags
);
2489 static struct dst_entry
*ipv4_blackhole_dst_check(struct dst_entry
*dst
, u32 cookie
)
2494 static unsigned int ipv4_blackhole_mtu(const struct dst_entry
*dst
)
2496 unsigned int mtu
= dst_metric_raw(dst
, RTAX_MTU
);
2498 return mtu
? : dst
->dev
->mtu
;
2501 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry
*dst
, struct sock
*sk
,
2502 struct sk_buff
*skb
, u32 mtu
)
2506 static void ipv4_rt_blackhole_redirect(struct dst_entry
*dst
, struct sock
*sk
,
2507 struct sk_buff
*skb
)
2511 static u32
*ipv4_rt_blackhole_cow_metrics(struct dst_entry
*dst
,
2517 static struct dst_ops ipv4_dst_blackhole_ops
= {
2519 .check
= ipv4_blackhole_dst_check
,
2520 .mtu
= ipv4_blackhole_mtu
,
2521 .default_advmss
= ipv4_default_advmss
,
2522 .update_pmtu
= ipv4_rt_blackhole_update_pmtu
,
2523 .redirect
= ipv4_rt_blackhole_redirect
,
2524 .cow_metrics
= ipv4_rt_blackhole_cow_metrics
,
2525 .neigh_lookup
= ipv4_neigh_lookup
,
2528 struct dst_entry
*ipv4_blackhole_route(struct net
*net
, struct dst_entry
*dst_orig
)
2530 struct rtable
*ort
= (struct rtable
*) dst_orig
;
2533 rt
= dst_alloc(&ipv4_dst_blackhole_ops
, NULL
, 1, DST_OBSOLETE_DEAD
, 0);
2535 struct dst_entry
*new = &rt
->dst
;
2538 new->input
= dst_discard
;
2539 new->output
= dst_discard_out
;
2541 new->dev
= net
->loopback_dev
;
2545 rt
->rt_is_input
= ort
->rt_is_input
;
2546 rt
->rt_iif
= ort
->rt_iif
;
2547 rt
->rt_pmtu
= ort
->rt_pmtu
;
2548 rt
->rt_mtu_locked
= ort
->rt_mtu_locked
;
2550 rt
->rt_genid
= rt_genid_ipv4(net
);
2551 rt
->rt_flags
= ort
->rt_flags
;
2552 rt
->rt_type
= ort
->rt_type
;
2553 rt
->rt_gateway
= ort
->rt_gateway
;
2554 rt
->rt_uses_gateway
= ort
->rt_uses_gateway
;
2556 INIT_LIST_HEAD(&rt
->rt_uncached
);
2559 dst_release(dst_orig
);
2561 return rt
? &rt
->dst
: ERR_PTR(-ENOMEM
);
2564 struct rtable
*ip_route_output_flow(struct net
*net
, struct flowi4
*flp4
,
2565 const struct sock
*sk
)
2567 struct rtable
*rt
= __ip_route_output_key(net
, flp4
);
2572 if (flp4
->flowi4_proto
)
2573 rt
= (struct rtable
*)xfrm_lookup_route(net
, &rt
->dst
,
2574 flowi4_to_flowi(flp4
),
2579 EXPORT_SYMBOL_GPL(ip_route_output_flow
);
2581 /* called with rcu_read_lock held */
2582 static int rt_fill_info(struct net
*net
, __be32 dst
, __be32 src
, u32 table_id
,
2583 struct flowi4
*fl4
, struct sk_buff
*skb
, u32 portid
,
2586 struct rtable
*rt
= skb_rtable(skb
);
2588 struct nlmsghdr
*nlh
;
2589 unsigned long expires
= 0;
2591 u32 metrics
[RTAX_MAX
];
2593 nlh
= nlmsg_put(skb
, portid
, seq
, RTM_NEWROUTE
, sizeof(*r
), 0);
2597 r
= nlmsg_data(nlh
);
2598 r
->rtm_family
= AF_INET
;
2599 r
->rtm_dst_len
= 32;
2601 r
->rtm_tos
= fl4
->flowi4_tos
;
2602 r
->rtm_table
= table_id
< 256 ? table_id
: RT_TABLE_COMPAT
;
2603 if (nla_put_u32(skb
, RTA_TABLE
, table_id
))
2604 goto nla_put_failure
;
2605 r
->rtm_type
= rt
->rt_type
;
2606 r
->rtm_scope
= RT_SCOPE_UNIVERSE
;
2607 r
->rtm_protocol
= RTPROT_UNSPEC
;
2608 r
->rtm_flags
= (rt
->rt_flags
& ~0xFFFF) | RTM_F_CLONED
;
2609 if (rt
->rt_flags
& RTCF_NOTIFY
)
2610 r
->rtm_flags
|= RTM_F_NOTIFY
;
2611 if (IPCB(skb
)->flags
& IPSKB_DOREDIRECT
)
2612 r
->rtm_flags
|= RTCF_DOREDIRECT
;
2614 if (nla_put_in_addr(skb
, RTA_DST
, dst
))
2615 goto nla_put_failure
;
2617 r
->rtm_src_len
= 32;
2618 if (nla_put_in_addr(skb
, RTA_SRC
, src
))
2619 goto nla_put_failure
;
2622 nla_put_u32(skb
, RTA_OIF
, rt
->dst
.dev
->ifindex
))
2623 goto nla_put_failure
;
2624 #ifdef CONFIG_IP_ROUTE_CLASSID
2625 if (rt
->dst
.tclassid
&&
2626 nla_put_u32(skb
, RTA_FLOW
, rt
->dst
.tclassid
))
2627 goto nla_put_failure
;
2629 if (!rt_is_input_route(rt
) &&
2630 fl4
->saddr
!= src
) {
2631 if (nla_put_in_addr(skb
, RTA_PREFSRC
, fl4
->saddr
))
2632 goto nla_put_failure
;
2634 if (rt
->rt_uses_gateway
&&
2635 nla_put_in_addr(skb
, RTA_GATEWAY
, rt
->rt_gateway
))
2636 goto nla_put_failure
;
2638 expires
= rt
->dst
.expires
;
2640 unsigned long now
= jiffies
;
2642 if (time_before(now
, expires
))
2648 memcpy(metrics
, dst_metrics_ptr(&rt
->dst
), sizeof(metrics
));
2649 if (rt
->rt_pmtu
&& expires
)
2650 metrics
[RTAX_MTU
- 1] = rt
->rt_pmtu
;
2651 if (rt
->rt_mtu_locked
&& expires
)
2652 metrics
[RTAX_LOCK
- 1] |= BIT(RTAX_MTU
);
2653 if (rtnetlink_put_metrics(skb
, metrics
) < 0)
2654 goto nla_put_failure
;
2656 if (fl4
->flowi4_mark
&&
2657 nla_put_u32(skb
, RTA_MARK
, fl4
->flowi4_mark
))
2658 goto nla_put_failure
;
2660 if (!uid_eq(fl4
->flowi4_uid
, INVALID_UID
) &&
2661 nla_put_u32(skb
, RTA_UID
,
2662 from_kuid_munged(current_user_ns(), fl4
->flowi4_uid
)))
2663 goto nla_put_failure
;
2665 error
= rt
->dst
.error
;
2667 if (rt_is_input_route(rt
)) {
2668 #ifdef CONFIG_IP_MROUTE
2669 if (ipv4_is_multicast(dst
) && !ipv4_is_local_multicast(dst
) &&
2670 IPV4_DEVCONF_ALL(net
, MC_FORWARDING
)) {
2671 int err
= ipmr_get_route(net
, skb
,
2672 fl4
->saddr
, fl4
->daddr
,
2678 goto nla_put_failure
;
2682 if (nla_put_u32(skb
, RTA_IIF
, skb
->dev
->ifindex
))
2683 goto nla_put_failure
;
2686 if (rtnl_put_cacheinfo(skb
, &rt
->dst
, 0, expires
, error
) < 0)
2687 goto nla_put_failure
;
2689 nlmsg_end(skb
, nlh
);
2693 nlmsg_cancel(skb
, nlh
);
2697 static int inet_rtm_getroute(struct sk_buff
*in_skb
, struct nlmsghdr
*nlh
,
2698 struct netlink_ext_ack
*extack
)
2700 struct net
*net
= sock_net(in_skb
->sk
);
2702 struct nlattr
*tb
[RTA_MAX
+1];
2703 struct fib_result res
= {};
2704 struct rtable
*rt
= NULL
;
2711 struct sk_buff
*skb
;
2712 u32 table_id
= RT_TABLE_MAIN
;
2715 err
= nlmsg_parse(nlh
, sizeof(*rtm
), tb
, RTA_MAX
, rtm_ipv4_policy
,
2720 rtm
= nlmsg_data(nlh
);
2722 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
2728 /* Reserve room for dummy headers, this skb can pass
2729 through good chunk of routing engine.
2731 skb_reset_mac_header(skb
);
2732 skb_reset_network_header(skb
);
2734 src
= tb
[RTA_SRC
] ? nla_get_in_addr(tb
[RTA_SRC
]) : 0;
2735 dst
= tb
[RTA_DST
] ? nla_get_in_addr(tb
[RTA_DST
]) : 0;
2736 iif
= tb
[RTA_IIF
] ? nla_get_u32(tb
[RTA_IIF
]) : 0;
2737 mark
= tb
[RTA_MARK
] ? nla_get_u32(tb
[RTA_MARK
]) : 0;
2739 uid
= make_kuid(current_user_ns(), nla_get_u32(tb
[RTA_UID
]));
2741 uid
= (iif
? INVALID_UID
: current_uid());
2743 /* Bugfix: need to give ip_route_input enough of an IP header to
2746 ip_hdr(skb
)->protocol
= IPPROTO_UDP
;
2747 ip_hdr(skb
)->saddr
= src
;
2748 ip_hdr(skb
)->daddr
= dst
;
2750 skb_reserve(skb
, MAX_HEADER
+ sizeof(struct iphdr
));
2752 memset(&fl4
, 0, sizeof(fl4
));
2755 fl4
.flowi4_tos
= rtm
->rtm_tos
;
2756 fl4
.flowi4_oif
= tb
[RTA_OIF
] ? nla_get_u32(tb
[RTA_OIF
]) : 0;
2757 fl4
.flowi4_mark
= mark
;
2758 fl4
.flowi4_uid
= uid
;
2763 struct net_device
*dev
;
2765 dev
= dev_get_by_index_rcu(net
, iif
);
2771 skb
->protocol
= htons(ETH_P_IP
);
2774 err
= ip_route_input_rcu(skb
, dst
, src
, rtm
->rtm_tos
,
2777 rt
= skb_rtable(skb
);
2778 if (err
== 0 && rt
->dst
.error
)
2779 err
= -rt
->dst
.error
;
2781 fl4
.flowi4_iif
= LOOPBACK_IFINDEX
;
2782 rt
= ip_route_output_key_hash_rcu(net
, &fl4
, &res
, skb
);
2787 skb_dst_set(skb
, &rt
->dst
);
2793 if (rtm
->rtm_flags
& RTM_F_NOTIFY
)
2794 rt
->rt_flags
|= RTCF_NOTIFY
;
2796 if (rtm
->rtm_flags
& RTM_F_LOOKUP_TABLE
)
2797 table_id
= rt
->rt_table_id
;
2799 if (rtm
->rtm_flags
& RTM_F_FIB_MATCH
) {
2801 err
= fib_props
[res
.type
].error
;
2803 err
= -EHOSTUNREACH
;
2806 err
= fib_dump_info(skb
, NETLINK_CB(in_skb
).portid
,
2807 nlh
->nlmsg_seq
, RTM_NEWROUTE
, table_id
,
2808 rt
->rt_type
, res
.prefix
, res
.prefixlen
,
2809 fl4
.flowi4_tos
, res
.fi
, 0);
2811 err
= rt_fill_info(net
, dst
, src
, table_id
, &fl4
, skb
,
2812 NETLINK_CB(in_skb
).portid
, nlh
->nlmsg_seq
);
2819 err
= rtnl_unicast(skb
, net
, NETLINK_CB(in_skb
).portid
);
2829 void ip_rt_multicast_event(struct in_device
*in_dev
)
2831 rt_cache_flush(dev_net(in_dev
->dev
));
2834 #ifdef CONFIG_SYSCTL
2835 static int ip_rt_gc_interval __read_mostly
= 60 * HZ
;
2836 static int ip_rt_gc_min_interval __read_mostly
= HZ
/ 2;
2837 static int ip_rt_gc_elasticity __read_mostly
= 8;
2839 static int ipv4_sysctl_rtcache_flush(struct ctl_table
*__ctl
, int write
,
2840 void __user
*buffer
,
2841 size_t *lenp
, loff_t
*ppos
)
2843 struct net
*net
= (struct net
*)__ctl
->extra1
;
2846 rt_cache_flush(net
);
2847 fnhe_genid_bump(net
);
2854 static struct ctl_table ipv4_route_table
[] = {
2856 .procname
= "gc_thresh",
2857 .data
= &ipv4_dst_ops
.gc_thresh
,
2858 .maxlen
= sizeof(int),
2860 .proc_handler
= proc_dointvec
,
2863 .procname
= "max_size",
2864 .data
= &ip_rt_max_size
,
2865 .maxlen
= sizeof(int),
2867 .proc_handler
= proc_dointvec
,
2870 /* Deprecated. Use gc_min_interval_ms */
2872 .procname
= "gc_min_interval",
2873 .data
= &ip_rt_gc_min_interval
,
2874 .maxlen
= sizeof(int),
2876 .proc_handler
= proc_dointvec_jiffies
,
2879 .procname
= "gc_min_interval_ms",
2880 .data
= &ip_rt_gc_min_interval
,
2881 .maxlen
= sizeof(int),
2883 .proc_handler
= proc_dointvec_ms_jiffies
,
2886 .procname
= "gc_timeout",
2887 .data
= &ip_rt_gc_timeout
,
2888 .maxlen
= sizeof(int),
2890 .proc_handler
= proc_dointvec_jiffies
,
2893 .procname
= "gc_interval",
2894 .data
= &ip_rt_gc_interval
,
2895 .maxlen
= sizeof(int),
2897 .proc_handler
= proc_dointvec_jiffies
,
2900 .procname
= "redirect_load",
2901 .data
= &ip_rt_redirect_load
,
2902 .maxlen
= sizeof(int),
2904 .proc_handler
= proc_dointvec
,
2907 .procname
= "redirect_number",
2908 .data
= &ip_rt_redirect_number
,
2909 .maxlen
= sizeof(int),
2911 .proc_handler
= proc_dointvec
,
2914 .procname
= "redirect_silence",
2915 .data
= &ip_rt_redirect_silence
,
2916 .maxlen
= sizeof(int),
2918 .proc_handler
= proc_dointvec
,
2921 .procname
= "error_cost",
2922 .data
= &ip_rt_error_cost
,
2923 .maxlen
= sizeof(int),
2925 .proc_handler
= proc_dointvec
,
2928 .procname
= "error_burst",
2929 .data
= &ip_rt_error_burst
,
2930 .maxlen
= sizeof(int),
2932 .proc_handler
= proc_dointvec
,
2935 .procname
= "gc_elasticity",
2936 .data
= &ip_rt_gc_elasticity
,
2937 .maxlen
= sizeof(int),
2939 .proc_handler
= proc_dointvec
,
2942 .procname
= "mtu_expires",
2943 .data
= &ip_rt_mtu_expires
,
2944 .maxlen
= sizeof(int),
2946 .proc_handler
= proc_dointvec_jiffies
,
2949 .procname
= "min_pmtu",
2950 .data
= &ip_rt_min_pmtu
,
2951 .maxlen
= sizeof(int),
2953 .proc_handler
= proc_dointvec_minmax
,
2954 .extra1
= &ip_min_valid_pmtu
,
2957 .procname
= "min_adv_mss",
2958 .data
= &ip_rt_min_advmss
,
2959 .maxlen
= sizeof(int),
2961 .proc_handler
= proc_dointvec
,
2966 static struct ctl_table ipv4_route_flush_table
[] = {
2968 .procname
= "flush",
2969 .maxlen
= sizeof(int),
2971 .proc_handler
= ipv4_sysctl_rtcache_flush
,
2976 static __net_init
int sysctl_route_net_init(struct net
*net
)
2978 struct ctl_table
*tbl
;
2980 tbl
= ipv4_route_flush_table
;
2981 if (!net_eq(net
, &init_net
)) {
2982 tbl
= kmemdup(tbl
, sizeof(ipv4_route_flush_table
), GFP_KERNEL
);
2986 /* Don't export sysctls to unprivileged users */
2987 if (net
->user_ns
!= &init_user_ns
)
2988 tbl
[0].procname
= NULL
;
2990 tbl
[0].extra1
= net
;
2992 net
->ipv4
.route_hdr
= register_net_sysctl(net
, "net/ipv4/route", tbl
);
2993 if (!net
->ipv4
.route_hdr
)
2998 if (tbl
!= ipv4_route_flush_table
)
3004 static __net_exit
void sysctl_route_net_exit(struct net
*net
)
3006 struct ctl_table
*tbl
;
3008 tbl
= net
->ipv4
.route_hdr
->ctl_table_arg
;
3009 unregister_net_sysctl_table(net
->ipv4
.route_hdr
);
3010 BUG_ON(tbl
== ipv4_route_flush_table
);
3014 static __net_initdata
struct pernet_operations sysctl_route_ops
= {
3015 .init
= sysctl_route_net_init
,
3016 .exit
= sysctl_route_net_exit
,
3020 static __net_init
int rt_genid_init(struct net
*net
)
3022 atomic_set(&net
->ipv4
.rt_genid
, 0);
3023 atomic_set(&net
->fnhe_genid
, 0);
3024 atomic_set(&net
->ipv4
.dev_addr_genid
, get_random_int());
3028 static __net_initdata
struct pernet_operations rt_genid_ops
= {
3029 .init
= rt_genid_init
,
3032 static int __net_init
ipv4_inetpeer_init(struct net
*net
)
3034 struct inet_peer_base
*bp
= kmalloc(sizeof(*bp
), GFP_KERNEL
);
3038 inet_peer_base_init(bp
);
3039 net
->ipv4
.peers
= bp
;
3043 static void __net_exit
ipv4_inetpeer_exit(struct net
*net
)
3045 struct inet_peer_base
*bp
= net
->ipv4
.peers
;
3047 net
->ipv4
.peers
= NULL
;
3048 inetpeer_invalidate_tree(bp
);
3052 static __net_initdata
struct pernet_operations ipv4_inetpeer_ops
= {
3053 .init
= ipv4_inetpeer_init
,
3054 .exit
= ipv4_inetpeer_exit
,
3057 #ifdef CONFIG_IP_ROUTE_CLASSID
3058 struct ip_rt_acct __percpu
*ip_rt_acct __read_mostly
;
3059 #endif /* CONFIG_IP_ROUTE_CLASSID */
3061 int __init
ip_rt_init(void)
3066 ip_idents
= kmalloc(IP_IDENTS_SZ
* sizeof(*ip_idents
), GFP_KERNEL
);
3068 panic("IP: failed to allocate ip_idents\n");
3070 prandom_bytes(ip_idents
, IP_IDENTS_SZ
* sizeof(*ip_idents
));
3072 ip_tstamps
= kcalloc(IP_IDENTS_SZ
, sizeof(*ip_tstamps
), GFP_KERNEL
);
3074 panic("IP: failed to allocate ip_tstamps\n");
3076 for_each_possible_cpu(cpu
) {
3077 struct uncached_list
*ul
= &per_cpu(rt_uncached_list
, cpu
);
3079 INIT_LIST_HEAD(&ul
->head
);
3080 spin_lock_init(&ul
->lock
);
3082 #ifdef CONFIG_IP_ROUTE_CLASSID
3083 ip_rt_acct
= __alloc_percpu(256 * sizeof(struct ip_rt_acct
), __alignof__(struct ip_rt_acct
));
3085 panic("IP: failed to allocate ip_rt_acct\n");
3088 ipv4_dst_ops
.kmem_cachep
=
3089 kmem_cache_create("ip_dst_cache", sizeof(struct rtable
), 0,
3090 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3092 ipv4_dst_blackhole_ops
.kmem_cachep
= ipv4_dst_ops
.kmem_cachep
;
3094 if (dst_entries_init(&ipv4_dst_ops
) < 0)
3095 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3097 if (dst_entries_init(&ipv4_dst_blackhole_ops
) < 0)
3098 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3100 ipv4_dst_ops
.gc_thresh
= ~0;
3101 ip_rt_max_size
= INT_MAX
;
3106 if (ip_rt_proc_init())
3107 pr_err("Unable to create route proc files\n");
3112 rtnl_register(PF_INET
, RTM_GETROUTE
, inet_rtm_getroute
, NULL
,
3113 RTNL_FLAG_DOIT_UNLOCKED
);
3115 #ifdef CONFIG_SYSCTL
3116 register_pernet_subsys(&sysctl_route_ops
);
3118 register_pernet_subsys(&rt_genid_ops
);
3119 register_pernet_subsys(&ipv4_inetpeer_ops
);
3123 #ifdef CONFIG_SYSCTL
3125 * We really need to sanitize the damn ipv4 init order, then all
3126 * this nonsense will go away.
3128 void __init
ip_static_sysctl_init(void)
3130 register_net_sysctl(&init_net
, "net/ipv4/route", ipv4_route_table
);