blob: a3c1b9dfc9a19e239b5cf3176f8cb1b5dd6b8356 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 *
Jesper Juhl02c30a82005-05-05 16:16:16 -070010 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/types.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/ip.h>
27#include <linux/in.h>
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -030028#include <linux/skbuff.h>
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020029
30#include <net/inet_sock.h>
Francesco Fuscoaa661582013-09-24 15:43:09 +020031#include <net/route.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <net/snmp.h>
KOVACS Krisztian86b08d82008-10-01 07:44:42 -070033#include <net/flow.h>
Jiri Pirko1bd758e2015-05-12 14:56:07 +020034#include <net/flow_dissector.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Eric Dumazetc2f78bf2017-12-11 07:17:39 -080036#define IPV4_MIN_MTU 68 /* RFC 791 */
37
Linus Torvalds1da177e2005-04-16 15:20:36 -070038struct sock;
39
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +000040struct inet_skb_parm {
David Ahern0b922b72016-05-10 11:19:51 -070041 int iif;
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 struct ip_options opt; /* Compiled IP options */
David Aherna04a4802016-10-16 20:02:52 -070043 u16 flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Hannes Frederic Sowadf4d9252015-01-23 12:01:26 +010045#define IPSKB_FORWARDED BIT(0)
46#define IPSKB_XFRM_TUNNEL_SIZE BIT(1)
47#define IPSKB_XFRM_TRANSFORMED BIT(2)
48#define IPSKB_FRAG_COMPLETE BIT(3)
49#define IPSKB_REROUTED BIT(4)
50#define IPSKB_DOREDIRECT BIT(5)
Florian Westphald6b915e2015-05-22 16:32:51 +020051#define IPSKB_FRAG_PMTU BIT(6)
Lance Richardson9ee6c5d2016-11-02 16:36:17 -040052#define IPSKB_L3SLAVE BIT(7)
Patrick McHardy5f2d04f2012-08-26 19:13:55 +020053
54 u16 frag_max_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055};
56
David Aherna04a4802016-10-16 20:02:52 -070057static inline bool ipv4_l3mdev_skb(u16 flags)
58{
59 return !!(flags & IPSKB_L3SLAVE);
60}
61
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -030062static inline unsigned int ip_hdrlen(const struct sk_buff *skb)
63{
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -070064 return ip_hdr(skb)->ihl * 4;
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -030065}
66
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +000067struct ipcm_cookie {
Soheil Hassas Yeganeh24025c42016-04-02 23:08:10 -040068 struct sockcm_cookie sockc;
Al Viroc1d18f92006-09-27 18:28:28 -070069 __be32 addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 int oif;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +000071 struct ip_options_rcu *opt;
Oliver Hartkopp2244d072010-08-17 08:59:14 +000072 __u8 tx_flags;
Francesco Fuscof02db312013-09-24 15:43:08 +020073 __u8 ttl;
74 __s16 tos;
75 char priority;
Linus Torvalds1da177e2005-04-16 15:20:36 -070076};
77
78#define IPCB(skb) ((struct inet_skb_parm*)((skb)->cb))
Hannes Frederic Sowa4b261c72014-01-20 03:43:08 +010079#define PKTINFO_SKB_CB(skb) ((struct in_pktinfo *)((skb)->cb))
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +000081struct ip_ra_chain {
Eric Dumazet43a951e2010-10-25 03:32:44 +000082 struct ip_ra_chain __rcu *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 struct sock *sk;
Eric Dumazet592fcb92010-06-09 16:21:07 +000084 union {
85 void (*destructor)(struct sock *);
86 struct sock *saved_sk;
87 };
Eric Dumazet66018502010-06-07 03:12:08 +000088 struct rcu_head rcu;
Linus Torvalds1da177e2005-04-16 15:20:36 -070089};
90
Eric Dumazet43a951e2010-10-25 03:32:44 +000091extern struct ip_ra_chain __rcu *ip_ra_chain;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
93/* IP flags. */
94#define IP_CE 0x8000 /* Flag: "Congestion" */
95#define IP_DF 0x4000 /* Flag: "Don't Fragment" */
96#define IP_MF 0x2000 /* Flag: "More Fragments" */
97#define IP_OFFSET 0x1FFF /* "Fragment Offset" part */
98
99#define IP_FRAG_TIME (30 * HZ) /* fragment lifetime */
100
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -0200101struct msghdr;
102struct net_device;
103struct packet_type;
104struct rtable;
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -0200105struct sockaddr;
106
WANG Cong72c1d3b2014-01-10 16:09:45 -0800107int igmp_mc_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108
109/*
110 * Functions provided by ip.c
111 */
112
Eric Dumazetcfe673b2015-09-25 07:39:16 -0700113int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700114 __be32 saddr, __be32 daddr,
115 struct ip_options_rcu *opt);
116int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
117 struct net_device *orig_dev);
118int ip_local_deliver(struct sk_buff *skb);
119int ip_mr_input(struct sk_buff *skb);
Eric W. Biedermanede20592015-10-07 16:48:47 -0500120int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb);
121int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb);
Eric W. Biederman694869b2015-06-12 21:55:31 -0500122int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
123 int (*output)(struct net *, struct sock *, struct sk_buff *));
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700124void ip_send_check(struct iphdr *ip);
Eric W. Biedermancf91a992015-10-07 16:48:45 -0500125int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb);
Eric W. Biederman33224b12015-10-07 16:48:46 -0500126int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb);
Eric Dumazetaad88722014-04-15 13:47:15 -0400127
Eric Dumazetb0270e92014-04-15 12:58:34 -0400128int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl);
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700129void ip_init(void);
130int ip_append_data(struct sock *sk, struct flowi4 *fl4,
131 int getfrag(void *from, char *to, int offset, int len,
132 int odd, struct sk_buff *skb),
133 void *from, int len, int protolen,
134 struct ipcm_cookie *ipc,
135 struct rtable **rt,
136 unsigned int flags);
137int ip_generic_getfrag(void *from, char *to, int offset, int len, int odd,
138 struct sk_buff *skb);
139ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
140 int offset, size_t size, int flags);
141struct sk_buff *__ip_make_skb(struct sock *sk, struct flowi4 *fl4,
142 struct sk_buff_head *queue,
143 struct inet_cork *cork);
144int ip_send_skb(struct net *net, struct sk_buff *skb);
145int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4);
146void ip_flush_pending_frames(struct sock *sk);
147struct sk_buff *ip_make_skb(struct sock *sk, struct flowi4 *fl4,
148 int getfrag(void *from, char *to, int offset,
149 int len, int odd, struct sk_buff *skb),
150 void *from, int length, int transhdrlen,
151 struct ipcm_cookie *ipc, struct rtable **rtp,
152 unsigned int flags);
Herbert Xu1c32c5a2011-03-01 02:36:47 +0000153
David S. Miller77968b72011-05-08 17:12:19 -0700154static inline struct sk_buff *ip_finish_skb(struct sock *sk, struct flowi4 *fl4)
Herbert Xu1c32c5a2011-03-01 02:36:47 +0000155{
David S. Miller77968b72011-05-08 17:12:19 -0700156 return __ip_make_skb(sk, fl4, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
Herbert Xu1c32c5a2011-03-01 02:36:47 +0000157}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
Francesco Fuscoaa661582013-09-24 15:43:09 +0200159static inline __u8 get_rttos(struct ipcm_cookie* ipc, struct inet_sock *inet)
160{
161 return (ipc->tos != -1) ? RT_TOS(ipc->tos) : RT_TOS(inet->tos);
162}
163
164static inline __u8 get_rtconn_flags(struct ipcm_cookie* ipc, struct sock* sk)
165{
166 return (ipc->tos != -1) ? RT_CONN_FLAGS_TOS(sk, ipc->tos) : RT_CONN_FLAGS(sk);
167}
168
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169/* datagram.c */
Eric Dumazet03645a12015-07-14 08:10:22 +0200170int __ip4_datagram_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700171int ip4_datagram_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700173void ip4_datagram_release_cb(struct sock *sk);
Steffen Klassert8141ed92013-01-21 02:00:03 +0000174
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175struct ip_reply_arg {
176 struct kvec iov[1];
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700177 int flags;
Al Virod6f5493c2006-11-14 21:26:08 -0800178 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 int csumoffset; /* u16 offset of csum in iov[0].iov_base */
180 /* -1 if not needed */
Patrick McHardyf0e48db2007-06-04 21:32:46 -0700181 int bound_dev_if;
Eric Dumazet66b13d92011-10-24 03:06:21 -0400182 u8 tos;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183};
184
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700185#define IP_REPLY_ARG_NOSRCCHECK 1
186
KOVACS Krisztian86b08d82008-10-01 07:44:42 -0700187static inline __u8 ip_reply_arg_flowi_flags(const struct ip_reply_arg *arg)
188{
189 return (arg->flags & IP_REPLY_ARG_NOSRCCHECK) ? FLOWI_FLAG_ANYSRC : 0;
190}
191
Eric Dumazetbdbbb852015-01-29 21:35:05 -0800192void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
Eric Dumazet24a2d432014-09-27 09:50:55 -0700193 const struct ip_options *sopt,
194 __be32 daddr, __be32 saddr,
195 const struct ip_reply_arg *arg,
David S. Miller70e73412012-06-28 03:21:41 -0700196 unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
Eric Dumazet4ce3c182010-06-30 13:31:19 -0700198#define IP_INC_STATS(net, field) SNMP_INC_STATS64((net)->mib.ip_statistics, field)
Eric Dumazet13415e42016-04-27 16:44:43 -0700199#define __IP_INC_STATS(net, field) __SNMP_INC_STATS64((net)->mib.ip_statistics, field)
Eric Dumazet4ce3c182010-06-30 13:31:19 -0700200#define IP_ADD_STATS(net, field, val) SNMP_ADD_STATS64((net)->mib.ip_statistics, field, val)
Eric Dumazet13415e42016-04-27 16:44:43 -0700201#define __IP_ADD_STATS(net, field, val) __SNMP_ADD_STATS64((net)->mib.ip_statistics, field, val)
Eric Dumazet4ce3c182010-06-30 13:31:19 -0700202#define IP_UPD_PO_STATS(net, field, val) SNMP_UPD_PO_STATS64((net)->mib.ip_statistics, field, val)
Eric Dumazet13415e42016-04-27 16:44:43 -0700203#define __IP_UPD_PO_STATS(net, field, val) __SNMP_UPD_PO_STATS64((net)->mib.ip_statistics, field, val)
Pavel Emelyanov61a7e262008-07-18 04:03:08 -0700204#define NET_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.net_statistics, field)
Eric Dumazet13415e42016-04-27 16:44:43 -0700205#define __NET_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.net_statistics, field)
David S. Millerf7324ac2014-03-06 15:03:17 -0500206#define NET_ADD_STATS(net, field, adnd) SNMP_ADD_STATS((net)->mib.net_statistics, field, adnd)
Eric Dumazet13415e42016-04-27 16:44:43 -0700207#define __NET_ADD_STATS(net, field, adnd) __SNMP_ADD_STATS((net)->mib.net_statistics, field, adnd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
Raghavendra K Tc4c6bc32015-08-30 11:29:41 +0530209u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offct);
WANG Cong698365f2014-05-05 15:55:55 -0700210unsigned long snmp_fold_field(void __percpu *mib, int offt);
Eric Dumazet4ce3c182010-06-30 13:31:19 -0700211#if BITS_PER_LONG==32
Raghavendra K Tc4c6bc32015-08-30 11:29:41 +0530212u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offct,
213 size_t syncp_offset);
WANG Cong698365f2014-05-05 15:55:55 -0700214u64 snmp_fold_field64(void __percpu *mib, int offt, size_t sync_off);
Eric Dumazet4ce3c182010-06-30 13:31:19 -0700215#else
Raghavendra K Tc4c6bc32015-08-30 11:29:41 +0530216static inline u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offct,
217 size_t syncp_offset)
218{
219 return snmp_get_cpu_field(mib, cpu, offct);
220
221}
222
WANG Cong698365f2014-05-05 15:55:55 -0700223static inline u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_off)
Eric Dumazet4ce3c182010-06-30 13:31:19 -0700224{
225 return snmp_fold_field(mib, offt);
226}
227#endif
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -0700228
Jia He6348ef22016-09-30 11:28:58 +0800229#define snmp_get_cpu_field64_batch(buff64, stats_list, mib_statistic, offset) \
230{ \
231 int i, c; \
232 for_each_possible_cpu(c) { \
233 for (i = 0; stats_list[i].name; i++) \
234 buff64[i] += snmp_get_cpu_field64( \
235 mib_statistic, \
236 c, stats_list[i].entry, \
237 offset); \
238 } \
239}
240
241#define snmp_get_cpu_field_batch(buff, stats_list, mib_statistic) \
242{ \
243 int i, c; \
244 for_each_possible_cpu(c) { \
245 for (i = 0; stats_list[i].name; i++) \
246 buff[i] += snmp_get_cpu_field( \
247 mib_statistic, \
248 c, stats_list[i].entry); \
249 } \
250}
251
Eric W. Biederman0bbf87d2013-09-28 14:10:59 -0700252void inet_get_local_port_range(struct net *net, int *low, int *high);
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700253
David S. Millerfcd77db2014-05-15 13:43:14 -0400254#ifdef CONFIG_SYSCTL
WANG Cong122ff242014-05-12 16:04:53 -0700255static inline int inet_is_local_reserved_port(struct net *net, int port)
Amerigo Wange3826f12010-05-05 00:27:06 +0000256{
WANG Cong122ff242014-05-12 16:04:53 -0700257 if (!net->ipv4.sysctl_local_reserved_ports)
258 return 0;
259 return test_bit(port, net->ipv4.sysctl_local_reserved_ports);
Amerigo Wange3826f12010-05-05 00:27:06 +0000260}
WANG Cong20e61da2014-07-25 15:25:08 -0700261
262static inline bool sysctl_dev_name_is_allowed(const char *name)
263{
264 return strcmp(name, "default") != 0 && strcmp(name, "all") != 0;
265}
266
WANG Cong122ff242014-05-12 16:04:53 -0700267#else
268static inline int inet_is_local_reserved_port(struct net *net, int port)
269{
270 return 0;
271}
272#endif
Amerigo Wange3826f12010-05-05 00:27:06 +0000273
Deepa Dinamani822c8682016-02-27 00:32:15 -0800274__be32 inet_current_timestamp(void);
275
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300276/* From inetpeer.c */
277extern int inet_peer_threshold;
278extern int inet_peer_minttl;
279extern int inet_peer_maxttl;
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300280
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700281void ipfrag_init(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300282
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700283void ip_static_sysctl_init(void);
Al Virobd7b1532008-07-15 16:00:59 -0400284
Lorenzo Colittie1108612014-05-13 10:17:33 -0700285#define IP4_REPLY_MARK(net, mark) \
286 ((net)->ipv4.sysctl_fwmark_reflect ? (mark) : 0)
287
David S. Millerd18cd552011-06-23 21:28:52 -0700288static inline bool ip_is_fragment(const struct iphdr *iph)
289{
290 return (iph->frag_off & htons(IP_MF | IP_OFFSET)) != 0;
291}
292
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293#ifdef CONFIG_INET
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -0200294#include <net/dst.h>
295
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296/* The function in 2.2 was invalid, producing wrong result for
297 * check=0xFEFF. It was noticed by Arthur Skawina _year_ ago. --ANK(000625) */
298static inline
299int ip_decrease_ttl(struct iphdr *iph)
300{
Al Viro5c78f272006-11-14 21:42:26 -0800301 u32 check = (__force u32)iph->check;
302 check += (__force u32)htons(0x0100);
303 iph->check = (__force __sum16)(check + (check>=0xFFFF));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 return --iph->ttl;
305}
306
Sabrina Dubrocad5439072018-03-14 10:21:14 +0100307static inline int ip_mtu_locked(const struct dst_entry *dst)
308{
309 const struct rtable *rt = (const struct rtable *)dst;
310
311 return rt->rt_mtu_locked || dst_metric_locked(dst, RTAX_MTU);
312}
313
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314static inline
Eric Dumazet4e3f5d72015-09-25 07:39:14 -0700315int ip_dont_fragment(const struct sock *sk, const struct dst_entry *dst)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316{
Eric Dumazet4e3f5d72015-09-25 07:39:14 -0700317 u8 pmtudisc = READ_ONCE(inet_sk(sk)->pmtudisc);
318
319 return pmtudisc == IP_PMTUDISC_DO ||
320 (pmtudisc == IP_PMTUDISC_WANT &&
Sabrina Dubrocad5439072018-03-14 10:21:14 +0100321 !ip_mtu_locked(dst));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322}
323
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100324static inline bool ip_sk_accept_pmtu(const struct sock *sk)
325{
Hannes Frederic Sowa1b346572014-02-26 01:20:42 +0100326 return inet_sk(sk)->pmtudisc != IP_PMTUDISC_INTERFACE &&
327 inet_sk(sk)->pmtudisc != IP_PMTUDISC_OMIT;
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100328}
329
330static inline bool ip_sk_use_pmtu(const struct sock *sk)
331{
332 return inet_sk(sk)->pmtudisc < IP_PMTUDISC_PROBE;
333}
334
WANG Cong60ff7462014-05-04 16:39:18 -0700335static inline bool ip_sk_ignore_df(const struct sock *sk)
Hannes Frederic Sowa1b346572014-02-26 01:20:42 +0100336{
337 return inet_sk(sk)->pmtudisc < IP_PMTUDISC_DO ||
338 inet_sk(sk)->pmtudisc == IP_PMTUDISC_OMIT;
339}
340
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100341static inline unsigned int ip_dst_mtu_maybe_forward(const struct dst_entry *dst,
342 bool forwarding)
343{
344 struct net *net = dev_net(dst->dev);
345
346 if (net->ipv4.sysctl_ip_fwd_use_pmtu ||
Sabrina Dubrocad5439072018-03-14 10:21:14 +0100347 ip_mtu_locked(dst) ||
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100348 !forwarding)
349 return dst_mtu(dst);
350
Eric Dumazetf29c9f42017-08-16 11:09:12 -0700351 return min(READ_ONCE(dst->dev->mtu), IP_MAX_MTU);
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100352}
353
Shmulik Ladkanifedbb6b42016-06-29 21:47:03 +0300354static inline unsigned int ip_skb_dst_mtu(struct sock *sk,
355 const struct sk_buff *skb)
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100356{
Eric Dumazetcaf3f262015-10-04 21:08:08 -0700357 if (!sk || !sk_fullsock(sk) || ip_sk_use_pmtu(sk)) {
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100358 bool forwarding = IPCB(skb)->flags & IPSKB_FORWARDED;
Eric Dumazetcaf3f262015-10-04 21:08:08 -0700359
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100360 return ip_dst_mtu_maybe_forward(skb_dst(skb), forwarding);
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100361 }
Eric Dumazetcaf3f262015-10-04 21:08:08 -0700362
Eric Dumazetf29c9f42017-08-16 11:09:12 -0700363 return min(READ_ONCE(skb_dst(skb)->dev->mtu), IP_MAX_MTU);
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100364}
365
Eric Dumazet04ca6972014-07-26 08:58:10 +0200366u32 ip_idents_reserve(u32 hash, int segs);
Hannes Frederic Sowab6a77192015-03-25 17:07:44 +0100367void __ip_select_ident(struct net *net, struct iphdr *iph, int segs);
Eric Dumazet73f156a2014-06-02 05:26:03 -0700368
Hannes Frederic Sowab6a77192015-03-25 17:07:44 +0100369static inline void ip_select_ident_segs(struct net *net, struct sk_buff *skb,
370 struct sock *sk, int segs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371{
Ansis Atteka703133d2013-09-18 15:29:53 -0700372 struct iphdr *iph = ip_hdr(skb);
373
WANG Cong60ff7462014-05-04 16:39:18 -0700374 if ((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 /* This is only to work around buggy Windows95/2000
376 * VJ compression implementations. If the ID field
377 * does not change, they drop every other packet in
378 * a TCP stream using header compression.
379 */
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000380 if (sk && inet_sk(sk)->inet_daddr) {
381 iph->id = htons(inet_sk(sk)->inet_id);
Eric Dumazet73f156a2014-06-02 05:26:03 -0700382 inet_sk(sk)->inet_id += segs;
383 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 iph->id = 0;
Eric Dumazet73f156a2014-06-02 05:26:03 -0700385 }
386 } else {
Hannes Frederic Sowab6a77192015-03-25 17:07:44 +0100387 __ip_select_ident(net, iph, segs);
Eric Dumazet73f156a2014-06-02 05:26:03 -0700388 }
389}
390
Hannes Frederic Sowab6a77192015-03-25 17:07:44 +0100391static inline void ip_select_ident(struct net *net, struct sk_buff *skb,
392 struct sock *sk)
Eric Dumazet73f156a2014-06-02 05:26:03 -0700393{
Hannes Frederic Sowab6a77192015-03-25 17:07:44 +0100394 ip_select_ident_segs(net, skb, sk, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395}
396
Tom Herberted70fcf2014-05-02 16:29:38 -0700397static inline __wsum inet_compute_pseudo(struct sk_buff *skb, int proto)
398{
399 return csum_tcpudp_nofold(ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
400 skb->len, proto, 0);
401}
402
Tom Herbertc3f83242015-06-04 09:16:40 -0700403/* copy IPv4 saddr & daddr to flow_keys, possibly using 64bit load/store
404 * Equivalent to : flow->v4addrs.src = iph->saddr;
405 * flow->v4addrs.dst = iph->daddr;
406 */
407static inline void iph_to_flow_copy_v4addrs(struct flow_keys *flow,
408 const struct iphdr *iph)
409{
410 BUILD_BUG_ON(offsetof(typeof(flow->addrs), v4addrs.dst) !=
411 offsetof(typeof(flow->addrs), v4addrs.src) +
412 sizeof(flow->addrs.v4addrs.src));
413 memcpy(&flow->addrs.v4addrs, &iph->saddr, sizeof(flow->addrs.v4addrs));
414 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
415}
416
Tom Herbert1933a782014-08-22 13:34:04 -0700417static inline __wsum inet_gro_compute_pseudo(struct sk_buff *skb, int proto)
418{
419 const struct iphdr *iph = skb_gro_network_header(skb);
420
421 return csum_tcpudp_nofold(iph->saddr, iph->daddr,
422 skb_gro_len(skb), proto, 0);
423}
424
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425/*
426 * Map a multicast IP onto multicast MAC for type ethernet.
427 */
428
Al Viro714e85b2006-11-14 20:51:49 -0800429static inline void ip_eth_mc_map(__be32 naddr, char *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430{
Al Viro714e85b2006-11-14 20:51:49 -0800431 __u32 addr=ntohl(naddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 buf[0]=0x01;
433 buf[1]=0x00;
434 buf[2]=0x5e;
435 buf[5]=addr&0xFF;
436 addr>>=8;
437 buf[4]=addr&0xFF;
438 addr>>=8;
439 buf[3]=addr&0x7F;
440}
441
442/*
443 * Map a multicast IP onto multicast MAC for type IP-over-InfiniBand.
444 * Leave P_Key as 0 to be filled in by driver.
445 */
446
Rolf Manderscheida9e527e2007-12-10 13:38:41 -0700447static inline void ip_ib_mc_map(__be32 naddr, const unsigned char *broadcast, char *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448{
Al Viro714e85b2006-11-14 20:51:49 -0800449 __u32 addr;
Rolf Manderscheida9e527e2007-12-10 13:38:41 -0700450 unsigned char scope = broadcast[5] & 0xF;
451
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 buf[0] = 0; /* Reserved */
453 buf[1] = 0xff; /* Multicast QPN */
454 buf[2] = 0xff;
455 buf[3] = 0xff;
Al Viro714e85b2006-11-14 20:51:49 -0800456 addr = ntohl(naddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 buf[4] = 0xff;
Rolf Manderscheida9e527e2007-12-10 13:38:41 -0700458 buf[5] = 0x10 | scope; /* scope from broadcast address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 buf[6] = 0x40; /* IPv4 signature */
460 buf[7] = 0x1b;
Rolf Manderscheida9e527e2007-12-10 13:38:41 -0700461 buf[8] = broadcast[8]; /* P_Key */
462 buf[9] = broadcast[9];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 buf[10] = 0;
464 buf[11] = 0;
465 buf[12] = 0;
466 buf[13] = 0;
467 buf[14] = 0;
468 buf[15] = 0;
469 buf[19] = addr & 0xff;
470 addr >>= 8;
471 buf[18] = addr & 0xff;
472 addr >>= 8;
473 buf[17] = addr & 0xff;
474 addr >>= 8;
475 buf[16] = addr & 0x0f;
476}
477
Timo Teräs93ca3bb2011-03-28 22:40:53 +0000478static inline void ip_ipgre_mc_map(__be32 naddr, const unsigned char *broadcast, char *buf)
479{
480 if ((broadcast[0] | broadcast[1] | broadcast[2] | broadcast[3]) != 0)
481 memcpy(buf, broadcast, 4);
482 else
483 memcpy(buf, &naddr, sizeof(naddr));
484}
485
Eric Dumazetdfd56b82011-12-10 09:48:31 +0000486#if IS_ENABLED(CONFIG_IPV6)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487#include <linux/ipv6.h>
488#endif
489
490static __inline__ void inet_reset_saddr(struct sock *sk)
491{
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000492 inet_sk(sk)->inet_rcv_saddr = inet_sk(sk)->inet_saddr = 0;
Eric Dumazetdfd56b82011-12-10 09:48:31 +0000493#if IS_ENABLED(CONFIG_IPV6)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 if (sk->sk_family == PF_INET6) {
495 struct ipv6_pinfo *np = inet6_sk(sk);
496
497 memset(&np->saddr, 0, sizeof(np->saddr));
Eric Dumazetefe42082013-10-03 15:42:29 -0700498 memset(&sk->sk_v6_rcv_saddr, 0, sizeof(sk->sk_v6_rcv_saddr));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 }
500#endif
501}
502
503#endif
504
David Ahern72afa352015-08-27 16:06:59 -0700505static inline unsigned int ipv4_addr_hash(__be32 ip)
506{
507 return (__force unsigned int) ip;
508}
509
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700510bool ip_call_ra_chain(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
512/*
Rami Rosenb7982322008-01-01 21:13:09 -0800513 * Functions provided by ip_fragment.c
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 */
515
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000516enum ip_defrag_users {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 IP_DEFRAG_LOCAL_DELIVER,
518 IP_DEFRAG_CALL_RA_CHAIN,
519 IP_DEFRAG_CONNTRACK_IN,
Alexey Dobriyan4be929b2010-05-24 14:33:03 -0700520 __IP_DEFRAG_CONNTRACK_IN_END = IP_DEFRAG_CONNTRACK_IN + USHRT_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 IP_DEFRAG_CONNTRACK_OUT,
Alexey Dobriyan4be929b2010-05-24 14:33:03 -0700522 __IP_DEFRAG_CONNTRACK_OUT_END = IP_DEFRAG_CONNTRACK_OUT + USHRT_MAX,
Patrick McHardy8fa9ff62009-12-15 16:59:59 +0100523 IP_DEFRAG_CONNTRACK_BRIDGE_IN,
Alexey Dobriyan4be929b2010-05-24 14:33:03 -0700524 __IP_DEFRAG_CONNTRACK_BRIDGE_IN = IP_DEFRAG_CONNTRACK_BRIDGE_IN + USHRT_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 IP_DEFRAG_VS_IN,
526 IP_DEFRAG_VS_OUT,
David S. Miller595fc712011-07-05 01:05:48 -0700527 IP_DEFRAG_VS_FWD,
528 IP_DEFRAG_AF_PACKET,
Eric Dumazetbc416d92011-10-06 10:28:31 +0000529 IP_DEFRAG_MACVLAN,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530};
531
Andy Zhou5cf42282015-05-15 14:15:35 -0700532/* Return true if the value of 'user' is between 'lower_bond'
533 * and 'upper_bond' inclusively.
534 */
535static inline bool ip_defrag_user_in_between(u32 user,
536 enum ip_defrag_users lower_bond,
537 enum ip_defrag_users upper_bond)
538{
539 return user >= lower_bond && user <= upper_bond;
540}
541
Eric W. Biederman19bcf9f2015-10-09 13:44:54 -0500542int ip_defrag(struct net *net, struct sk_buff *skb, u32 user);
Eric Dumazetbc416d92011-10-06 10:28:31 +0000543#ifdef CONFIG_INET
Eric W. Biederman19bcf9f2015-10-09 13:44:54 -0500544struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user);
Eric Dumazetbc416d92011-10-06 10:28:31 +0000545#else
Eric W. Biederman19bcf9f2015-10-09 13:44:54 -0500546static inline struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
Eric Dumazetbc416d92011-10-06 10:28:31 +0000547{
548 return skb;
549}
550#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
552/*
553 * Functions provided by ip_forward.c
554 */
555
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700556int ip_forward(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
558/*
559 * Functions provided by ip_options.c
560 */
561
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700562void ip_options_build(struct sk_buff *skb, struct ip_options *opt,
563 __be32 daddr, struct rtable *rt, int is_frag);
Eric Dumazet24a2d432014-09-27 09:50:55 -0700564
565int __ip_options_echo(struct ip_options *dopt, struct sk_buff *skb,
566 const struct ip_options *sopt);
567static inline int ip_options_echo(struct ip_options *dopt, struct sk_buff *skb)
568{
569 return __ip_options_echo(dopt, skb, &IPCB(skb)->opt);
570}
571
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700572void ip_options_fragment(struct sk_buff *skb);
Nazarov Sergey3b448972019-02-25 19:27:15 +0300573int __ip_options_compile(struct net *net, struct ip_options *opt,
574 struct sk_buff *skb, __be32 *info);
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700575int ip_options_compile(struct net *net, struct ip_options *opt,
576 struct sk_buff *skb);
577int ip_options_get(struct net *net, struct ip_options_rcu **optp,
578 unsigned char *data, int optlen);
579int ip_options_get_from_user(struct net *net, struct ip_options_rcu **optp,
580 unsigned char __user *data, int optlen);
581void ip_options_undo(struct ip_options *opt);
582void ip_forward_options(struct sk_buff *skb);
Stephen Suryaputra5f5d6282019-04-01 09:17:32 -0400583int ip_options_rcv_srr(struct sk_buff *skb, struct net_device *dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
585/*
586 * Functions provided by ip_sockglue.c
587 */
588
Shawn Bohrerfbf88662013-10-07 11:01:40 -0500589void ipv4_pktinfo_prepare(const struct sock *sk, struct sk_buff *skb);
Eric Dumazet10df8e62016-10-23 18:03:06 -0700590void ip_cmsg_recv_offset(struct msghdr *msg, struct sk_buff *skb, int tlen, int offset);
Soheil Hassas Yeganeh24025c42016-04-02 23:08:10 -0400591int ip_cmsg_send(struct sock *sk, struct msghdr *msg,
Hannes Frederic Sowac8e6ad02014-02-18 21:38:08 +0100592 struct ipcm_cookie *ipc, bool allow_ipv6);
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700593int ip_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
594 unsigned int optlen);
595int ip_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
596 int __user *optlen);
597int compat_ip_setsockopt(struct sock *sk, int level, int optname,
598 char __user *optval, unsigned int optlen);
599int compat_ip_getsockopt(struct sock *sk, int level, int optname,
600 char __user *optval, int __user *optlen);
601int ip_ra_control(struct sock *sk, unsigned char on,
602 void (*destructor)(struct sock *));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
Hannes Frederic Sowa85fbaa72013-11-23 00:46:12 +0100604int ip_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len);
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700605void ip_icmp_error(struct sock *sk, struct sk_buff *skb, int err, __be16 port,
606 u32 info, u8 *payload);
607void ip_local_error(struct sock *sk, int err, __be32 daddr, __be16 dport,
608 u32 info);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
Tom Herbert5961de92015-01-05 13:56:16 -0800610static inline void ip_cmsg_recv(struct msghdr *msg, struct sk_buff *skb)
611{
Eric Dumazet10df8e62016-10-23 18:03:06 -0700612 ip_cmsg_recv_offset(msg, skb, 0, 0);
Tom Herbert5961de92015-01-05 13:56:16 -0800613}
614
Eric Dumazet4cdf5072014-09-19 07:38:40 -0700615bool icmp_global_allow(void);
616extern int sysctl_icmp_msgs_per_sec;
617extern int sysctl_icmp_msgs_burst;
618
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300619#ifdef CONFIG_PROC_FS
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700620int ip_misc_proc_init(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300621#endif
622
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623#endif /* _IP_H */