ipv4: Invalidate the socket cached route on pmtu events if possible
[deliverable/linux.git] / include / net / ip.h
CommitLineData
1da177e4
LT
1/*
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.
5 *
6 * Definitions for the IP module.
7 *
8 * Version: @(#)ip.h 1.0.2 05/07/93
9 *
02c30a84 10 * Authors: Ross Biro
1da177e4
LT
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 *
14 * Changes:
15 * Mike McLagan : Routing by source
16 *
17 * This program is free software; you can redistribute it and/or
18 * modify it under the terms of the GNU General Public License
19 * as published by the Free Software Foundation; either version
20 * 2 of the License, or (at your option) any later version.
21 */
22#ifndef _IP_H
23#define _IP_H
24
1da177e4 25#include <linux/types.h>
1da177e4
LT
26#include <linux/ip.h>
27#include <linux/in.h>
c9bdd4b5 28#include <linux/skbuff.h>
14c85021
ACM
29
30#include <net/inet_sock.h>
1da177e4 31#include <net/snmp.h>
86b08d86 32#include <net/flow.h>
1da177e4
LT
33
34struct sock;
35
fd2c3ef7 36struct inet_skb_parm {
1da177e4
LT
37 struct ip_options opt; /* Compiled IP options */
38 unsigned char flags;
39
8cdfab8a
PM
40#define IPSKB_FORWARDED 1
41#define IPSKB_XFRM_TUNNEL_SIZE 2
3e3850e9
PM
42#define IPSKB_XFRM_TRANSFORMED 4
43#define IPSKB_FRAG_COMPLETE 8
48d5cad8 44#define IPSKB_REROUTED 16
5f2d04f1
PM
45
46 u16 frag_max_size;
1da177e4
LT
47};
48
c9bdd4b5
ACM
49static inline unsigned int ip_hdrlen(const struct sk_buff *skb)
50{
eddc9ec5 51 return ip_hdr(skb)->ihl * 4;
c9bdd4b5
ACM
52}
53
fd2c3ef7 54struct ipcm_cookie {
c1d18f9f 55 __be32 addr;
1da177e4 56 int oif;
f6d8bd05 57 struct ip_options_rcu *opt;
2244d07b 58 __u8 tx_flags;
1da177e4
LT
59};
60
61#define IPCB(skb) ((struct inet_skb_parm*)((skb)->cb))
62
fd2c3ef7 63struct ip_ra_chain {
43a951e9 64 struct ip_ra_chain __rcu *next;
1da177e4 65 struct sock *sk;
592fcb9d
ED
66 union {
67 void (*destructor)(struct sock *);
68 struct sock *saved_sk;
69 };
66018506 70 struct rcu_head rcu;
1da177e4
LT
71};
72
43a951e9 73extern struct ip_ra_chain __rcu *ip_ra_chain;
1da177e4
LT
74
75/* IP flags. */
76#define IP_CE 0x8000 /* Flag: "Congestion" */
77#define IP_DF 0x4000 /* Flag: "Don't Fragment" */
78#define IP_MF 0x2000 /* Flag: "More Fragments" */
79#define IP_OFFSET 0x1FFF /* "Fragment Offset" part */
80
81#define IP_FRAG_TIME (30 * HZ) /* fragment lifetime */
82
14c85021
ACM
83struct msghdr;
84struct net_device;
85struct packet_type;
86struct rtable;
14c85021
ACM
87struct sockaddr;
88
1da177e4
LT
89extern int igmp_mc_proc_init(void);
90
91/*
92 * Functions provided by ip.c
93 */
94
95extern int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
13d8eaa0 96 __be32 saddr, __be32 daddr,
f6d8bd05 97 struct ip_options_rcu *opt);
1da177e4 98extern int ip_rcv(struct sk_buff *skb, struct net_device *dev,
f2ccd8fa 99 struct packet_type *pt, struct net_device *orig_dev);
1da177e4
LT
100extern int ip_local_deliver(struct sk_buff *skb);
101extern int ip_mr_input(struct sk_buff *skb);
102extern int ip_output(struct sk_buff *skb);
103extern int ip_mc_output(struct sk_buff *skb);
2e2f7aef 104extern int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
1da177e4
LT
105extern int ip_do_nat(struct sk_buff *skb);
106extern void ip_send_check(struct iphdr *ip);
c439cb2e
HX
107extern int __ip_local_out(struct sk_buff *skb);
108extern int ip_local_out(struct sk_buff *skb);
d9d8da80 109extern int ip_queue_xmit(struct sk_buff *skb, struct flowi *fl);
1da177e4 110extern void ip_init(void);
f5fca608 111extern int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1da177e4
LT
112 int getfrag(void *from, char *to, int offset, int len,
113 int odd, struct sk_buff *skb),
114 void *from, int len, int protolen,
115 struct ipcm_cookie *ipc,
2e77d89b 116 struct rtable **rt,
1da177e4
LT
117 unsigned int flags);
118extern int ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb);
f5fca608 119extern ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
1da177e4 120 int offset, size_t size, int flags);
1c32c5ad 121extern struct sk_buff *__ip_make_skb(struct sock *sk,
77968b78 122 struct flowi4 *fl4,
1c32c5ad
HX
123 struct sk_buff_head *queue,
124 struct inet_cork *cork);
b5ec8eea 125extern int ip_send_skb(struct net *net, struct sk_buff *skb);
77968b78 126extern int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4);
1da177e4 127extern void ip_flush_pending_frames(struct sock *sk);
1c32c5ad 128extern struct sk_buff *ip_make_skb(struct sock *sk,
77968b78 129 struct flowi4 *fl4,
1c32c5ad
HX
130 int getfrag(void *from, char *to, int offset, int len,
131 int odd, struct sk_buff *skb),
132 void *from, int length, int transhdrlen,
133 struct ipcm_cookie *ipc,
134 struct rtable **rtp,
135 unsigned int flags);
136
77968b78 137static inline struct sk_buff *ip_finish_skb(struct sock *sk, struct flowi4 *fl4)
1c32c5ad 138{
77968b78 139 return __ip_make_skb(sk, fl4, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1c32c5ad 140}
1da177e4
LT
141
142/* datagram.c */
143extern int ip4_datagram_connect(struct sock *sk,
144 struct sockaddr *uaddr, int addr_len);
145
1da177e4
LT
146struct ip_reply_arg {
147 struct kvec iov[1];
88ef4a5a 148 int flags;
d6f5493c 149 __wsum csum;
1da177e4
LT
150 int csumoffset; /* u16 offset of csum in iov[0].iov_base */
151 /* -1 if not needed */
f0e48dbf 152 int bound_dev_if;
66b13d99 153 u8 tos;
1da177e4
LT
154};
155
88ef4a5a
KK
156#define IP_REPLY_ARG_NOSRCCHECK 1
157
86b08d86
KK
158static inline __u8 ip_reply_arg_flowi_flags(const struct ip_reply_arg *arg)
159{
160 return (arg->flags & IP_REPLY_ARG_NOSRCCHECK) ? FLOWI_FLAG_ANYSRC : 0;
161}
162
be9f4a44 163void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
70e73416
DM
164 __be32 saddr, const struct ip_reply_arg *arg,
165 unsigned int len);
1da177e4 166
fd2c3ef7 167struct ipv4_config {
1da177e4 168 int log_martians;
1da177e4
LT
169 int no_pmtu_disc;
170};
171
172extern struct ipv4_config ipv4_config;
4ce3c183
ED
173#define IP_INC_STATS(net, field) SNMP_INC_STATS64((net)->mib.ip_statistics, field)
174#define IP_INC_STATS_BH(net, field) SNMP_INC_STATS64_BH((net)->mib.ip_statistics, field)
175#define IP_ADD_STATS(net, field, val) SNMP_ADD_STATS64((net)->mib.ip_statistics, field, val)
176#define IP_ADD_STATS_BH(net, field, val) SNMP_ADD_STATS64_BH((net)->mib.ip_statistics, field, val)
177#define IP_UPD_PO_STATS(net, field, val) SNMP_UPD_PO_STATS64((net)->mib.ip_statistics, field, val)
178#define IP_UPD_PO_STATS_BH(net, field, val) SNMP_UPD_PO_STATS64_BH((net)->mib.ip_statistics, field, val)
61a7e260
PE
179#define NET_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.net_statistics, field)
180#define NET_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.net_statistics, field)
181#define NET_INC_STATS_USER(net, field) SNMP_INC_STATS_USER((net)->mib.net_statistics, field)
182#define NET_ADD_STATS_BH(net, field, adnd) SNMP_ADD_STATS_BH((net)->mib.net_statistics, field, adnd)
183#define NET_ADD_STATS_USER(net, field, adnd) SNMP_ADD_STATS_USER((net)->mib.net_statistics, field, adnd)
1da177e4 184
7d720c3e 185extern unsigned long snmp_fold_field(void __percpu *mib[], int offt);
4ce3c183
ED
186#if BITS_PER_LONG==32
187extern u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t sync_off);
188#else
189static inline u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_off)
190{
191 return snmp_fold_field(mib, offt);
192}
193#endif
1823e4c8 194extern int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align);
7d720c3e 195extern void snmp_mib_free(void __percpu *ptr[2]);
33490170 196
3c689b73
ED
197extern struct local_ports {
198 seqlock_t lock;
199 int range[2];
200} sysctl_local_ports;
227b60f5
SH
201extern void inet_get_local_port_range(int *low, int *high);
202
e3826f1e
AW
203extern unsigned long *sysctl_local_reserved_ports;
204static inline int inet_is_reserved_local_port(int port)
205{
206 return test_bit(port, sysctl_local_reserved_ports);
207}
208
cdac4e07 209extern int sysctl_ip_nonlocal_bind;
1da177e4 210
20380731
ACM
211/* From inetpeer.c */
212extern int inet_peer_threshold;
213extern int inet_peer_minttl;
214extern int inet_peer_maxttl;
20380731 215
6648bd7e
AD
216/* From ip_input.c */
217extern int sysctl_ip_early_demux;
218
20380731
ACM
219/* From ip_output.c */
220extern int sysctl_ip_dynaddr;
221
222extern void ipfrag_init(void);
223
bd7b1533
AV
224extern void ip_static_sysctl_init(void);
225
d18cd551
DM
226static inline bool ip_is_fragment(const struct iphdr *iph)
227{
228 return (iph->frag_off & htons(IP_MF | IP_OFFSET)) != 0;
229}
230
1da177e4 231#ifdef CONFIG_INET
14c85021
ACM
232#include <net/dst.h>
233
1da177e4
LT
234/* The function in 2.2 was invalid, producing wrong result for
235 * check=0xFEFF. It was noticed by Arthur Skawina _year_ ago. --ANK(000625) */
236static inline
237int ip_decrease_ttl(struct iphdr *iph)
238{
5c78f275
AV
239 u32 check = (__force u32)iph->check;
240 check += (__force u32)htons(0x0100);
241 iph->check = (__force __sum16)(check + (check>=0xFFFF));
1da177e4
LT
242 return --iph->ttl;
243}
244
245static inline
246int ip_dont_fragment(struct sock *sk, struct dst_entry *dst)
247{
a02cec21 248 return inet_sk(sk)->pmtudisc == IP_PMTUDISC_DO ||
1da177e4 249 (inet_sk(sk)->pmtudisc == IP_PMTUDISC_WANT &&
a02cec21 250 !(dst_metric_locked(dst, RTAX_MTU)));
1da177e4
LT
251}
252
253extern void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more);
254
255static inline void ip_select_ident(struct iphdr *iph, struct dst_entry *dst, struct sock *sk)
256{
257 if (iph->frag_off & htons(IP_DF)) {
258 /* This is only to work around buggy Windows95/2000
259 * VJ compression implementations. If the ID field
260 * does not change, they drop every other packet in
261 * a TCP stream using header compression.
262 */
c720c7e8
ED
263 iph->id = (sk && inet_sk(sk)->inet_daddr) ?
264 htons(inet_sk(sk)->inet_id++) : 0;
1da177e4
LT
265 } else
266 __ip_select_ident(iph, dst, 0);
267}
268
269static inline void ip_select_ident_more(struct iphdr *iph, struct dst_entry *dst, struct sock *sk, int more)
270{
271 if (iph->frag_off & htons(IP_DF)) {
c720c7e8
ED
272 if (sk && inet_sk(sk)->inet_daddr) {
273 iph->id = htons(inet_sk(sk)->inet_id);
274 inet_sk(sk)->inet_id += 1 + more;
1da177e4
LT
275 } else
276 iph->id = 0;
277 } else
278 __ip_select_ident(iph, dst, more);
279}
280
281/*
282 * Map a multicast IP onto multicast MAC for type ethernet.
283 */
284
714e85be 285static inline void ip_eth_mc_map(__be32 naddr, char *buf)
1da177e4 286{
714e85be 287 __u32 addr=ntohl(naddr);
1da177e4
LT
288 buf[0]=0x01;
289 buf[1]=0x00;
290 buf[2]=0x5e;
291 buf[5]=addr&0xFF;
292 addr>>=8;
293 buf[4]=addr&0xFF;
294 addr>>=8;
295 buf[3]=addr&0x7F;
296}
297
298/*
299 * Map a multicast IP onto multicast MAC for type IP-over-InfiniBand.
300 * Leave P_Key as 0 to be filled in by driver.
301 */
302
a9e527e3 303static inline void ip_ib_mc_map(__be32 naddr, const unsigned char *broadcast, char *buf)
1da177e4 304{
714e85be 305 __u32 addr;
a9e527e3
RM
306 unsigned char scope = broadcast[5] & 0xF;
307
1da177e4
LT
308 buf[0] = 0; /* Reserved */
309 buf[1] = 0xff; /* Multicast QPN */
310 buf[2] = 0xff;
311 buf[3] = 0xff;
714e85be 312 addr = ntohl(naddr);
1da177e4 313 buf[4] = 0xff;
a9e527e3 314 buf[5] = 0x10 | scope; /* scope from broadcast address */
1da177e4
LT
315 buf[6] = 0x40; /* IPv4 signature */
316 buf[7] = 0x1b;
a9e527e3
RM
317 buf[8] = broadcast[8]; /* P_Key */
318 buf[9] = broadcast[9];
1da177e4
LT
319 buf[10] = 0;
320 buf[11] = 0;
321 buf[12] = 0;
322 buf[13] = 0;
323 buf[14] = 0;
324 buf[15] = 0;
325 buf[19] = addr & 0xff;
326 addr >>= 8;
327 buf[18] = addr & 0xff;
328 addr >>= 8;
329 buf[17] = addr & 0xff;
330 addr >>= 8;
331 buf[16] = addr & 0x0f;
332}
333
93ca3bb5
TT
334static inline void ip_ipgre_mc_map(__be32 naddr, const unsigned char *broadcast, char *buf)
335{
336 if ((broadcast[0] | broadcast[1] | broadcast[2] | broadcast[3]) != 0)
337 memcpy(buf, broadcast, 4);
338 else
339 memcpy(buf, &naddr, sizeof(naddr));
340}
341
dfd56b8b 342#if IS_ENABLED(CONFIG_IPV6)
1da177e4
LT
343#include <linux/ipv6.h>
344#endif
345
346static __inline__ void inet_reset_saddr(struct sock *sk)
347{
c720c7e8 348 inet_sk(sk)->inet_rcv_saddr = inet_sk(sk)->inet_saddr = 0;
dfd56b8b 349#if IS_ENABLED(CONFIG_IPV6)
1da177e4
LT
350 if (sk->sk_family == PF_INET6) {
351 struct ipv6_pinfo *np = inet6_sk(sk);
352
353 memset(&np->saddr, 0, sizeof(np->saddr));
354 memset(&np->rcv_saddr, 0, sizeof(np->rcv_saddr));
355 }
356#endif
357}
358
359#endif
360
7ad6848c
OP
361static inline int sk_mc_loop(struct sock *sk)
362{
363 if (!sk)
364 return 1;
365 switch (sk->sk_family) {
366 case AF_INET:
367 return inet_sk(sk)->mc_loop;
dfd56b8b 368#if IS_ENABLED(CONFIG_IPV6)
7ad6848c
OP
369 case AF_INET6:
370 return inet6_sk(sk)->mc_loop;
371#endif
372 }
cd65c3c7 373 WARN_ON(1);
7ad6848c
OP
374 return 1;
375}
376
ba57b4db 377extern bool ip_call_ra_chain(struct sk_buff *skb);
1da177e4
LT
378
379/*
b798232f 380 * Functions provided by ip_fragment.c
1da177e4
LT
381 */
382
fd2c3ef7 383enum ip_defrag_users {
1da177e4
LT
384 IP_DEFRAG_LOCAL_DELIVER,
385 IP_DEFRAG_CALL_RA_CHAIN,
386 IP_DEFRAG_CONNTRACK_IN,
4be929be 387 __IP_DEFRAG_CONNTRACK_IN_END = IP_DEFRAG_CONNTRACK_IN + USHRT_MAX,
1da177e4 388 IP_DEFRAG_CONNTRACK_OUT,
4be929be 389 __IP_DEFRAG_CONNTRACK_OUT_END = IP_DEFRAG_CONNTRACK_OUT + USHRT_MAX,
8fa9ff68 390 IP_DEFRAG_CONNTRACK_BRIDGE_IN,
4be929be 391 __IP_DEFRAG_CONNTRACK_BRIDGE_IN = IP_DEFRAG_CONNTRACK_BRIDGE_IN + USHRT_MAX,
1da177e4
LT
392 IP_DEFRAG_VS_IN,
393 IP_DEFRAG_VS_OUT,
595fc71b
DM
394 IP_DEFRAG_VS_FWD,
395 IP_DEFRAG_AF_PACKET,
bc416d97 396 IP_DEFRAG_MACVLAN,
1da177e4
LT
397};
398
776c729e 399int ip_defrag(struct sk_buff *skb, u32 user);
bc416d97
ED
400#ifdef CONFIG_INET
401struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user);
402#else
403static inline struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
404{
405 return skb;
406}
407#endif
6ddc0822 408int ip_frag_mem(struct net *net);
e5a2bb84 409int ip_frag_nqueues(struct net *net);
1da177e4
LT
410
411/*
412 * Functions provided by ip_forward.c
413 */
414
415extern int ip_forward(struct sk_buff *skb);
1da177e4
LT
416
417/*
418 * Functions provided by ip_options.c
419 */
420
f6d8bd05
ED
421extern void ip_options_build(struct sk_buff *skb, struct ip_options *opt,
422 __be32 daddr, struct rtable *rt, int is_frag);
1da177e4
LT
423extern int ip_options_echo(struct ip_options *dopt, struct sk_buff *skb);
424extern void ip_options_fragment(struct sk_buff *skb);
0e6bd4a1
DL
425extern int ip_options_compile(struct net *net,
426 struct ip_options *opt, struct sk_buff *skb);
f6d8bd05 427extern int ip_options_get(struct net *net, struct ip_options_rcu **optp,
4c6ea29d 428 unsigned char *data, int optlen);
f6d8bd05 429extern int ip_options_get_from_user(struct net *net, struct ip_options_rcu **optp,
4c6ea29d 430 unsigned char __user *data, int optlen);
1da177e4
LT
431extern void ip_options_undo(struct ip_options * opt);
432extern void ip_forward_options(struct sk_buff *skb);
433extern int ip_options_rcv_srr(struct sk_buff *skb);
434
435/*
436 * Functions provided by ip_sockglue.c
437 */
438
d826eb14 439extern void ipv4_pktinfo_prepare(struct sk_buff *skb);
1da177e4 440extern void ip_cmsg_recv(struct msghdr *msg, struct sk_buff *skb);
7a6adb92
DL
441extern int ip_cmsg_send(struct net *net,
442 struct msghdr *msg, struct ipcm_cookie *ipc);
b7058842 443extern int ip_setsockopt(struct sock *sk, int level, int optname, char __user *optval, unsigned int optlen);
1da177e4 444extern int ip_getsockopt(struct sock *sk, int level, int optname, char __user *optval, int __user *optlen);
3fdadf7d 445extern int compat_ip_setsockopt(struct sock *sk, int level,
b7058842 446 int optname, char __user *optval, unsigned int optlen);
3fdadf7d
DM
447extern int compat_ip_getsockopt(struct sock *sk, int level,
448 int optname, char __user *optval, int __user *optlen);
1da177e4
LT
449extern int ip_ra_control(struct sock *sk, unsigned char on, void (*destructor)(struct sock *));
450
451extern int ip_recv_error(struct sock *sk, struct msghdr *msg, int len);
452extern void ip_icmp_error(struct sock *sk, struct sk_buff *skb, int err,
35986b32 453 __be16 port, u32 info, u8 *payload);
0579016e 454extern void ip_local_error(struct sock *sk, int err, __be32 daddr, __be16 dport,
1da177e4
LT
455 u32 info);
456
20380731
ACM
457#ifdef CONFIG_PROC_FS
458extern int ip_misc_proc_init(void);
459#endif
460
1da177e4 461#endif /* _IP_H */
This page took 0.740115 seconds and 5 git commands to generate.