blob: ecffd843e7b896a83416847fdaa452be6223f3dc [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>
Martin KaFai Lauf0b1e642017-12-01 12:52:30 -080029#include <linux/jhash.h>
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020030
31#include <net/inet_sock.h>
Francesco Fuscoaa661582013-09-24 15:43:09 +020032#include <net/route.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <net/snmp.h>
KOVACS Krisztian86b08d82008-10-01 07:44:42 -070034#include <net/flow.h>
Jiri Pirko1bd758e2015-05-12 14:56:07 +020035#include <net/flow_dissector.h>
Martin KaFai Lauf0b1e642017-12-01 12:52:30 -080036#include <net/netns/hash.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Gao Feng7ed14d92017-04-12 12:34:03 +080038#define IPV4_MAX_PMTU 65535U /* RFC 2675, Section 5.1 */
Eric Dumazetb5476022017-12-11 07:17:39 -080039#define IPV4_MIN_MTU 68 /* RFC 791 */
Gao Feng7ed14d92017-04-12 12:34:03 +080040
Linus Torvalds1da177e2005-04-16 15:20:36 -070041struct sock;
42
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +000043struct inet_skb_parm {
David Ahern0b922b72016-05-10 11:19:51 -070044 int iif;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 struct ip_options opt; /* Compiled IP options */
David Aherna04a4802016-10-16 20:02:52 -070046 u16 flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
Hannes Frederic Sowadf4d9252015-01-23 12:01:26 +010048#define IPSKB_FORWARDED BIT(0)
49#define IPSKB_XFRM_TUNNEL_SIZE BIT(1)
50#define IPSKB_XFRM_TRANSFORMED BIT(2)
51#define IPSKB_FRAG_COMPLETE BIT(3)
52#define IPSKB_REROUTED BIT(4)
53#define IPSKB_DOREDIRECT BIT(5)
Florian Westphald6b915e2015-05-22 16:32:51 +020054#define IPSKB_FRAG_PMTU BIT(6)
Lance Richardson9ee6c5d2016-11-02 16:36:17 -040055#define IPSKB_L3SLAVE BIT(7)
Patrick McHardy5f2d04f2012-08-26 19:13:55 +020056
57 u16 frag_max_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058};
59
David Aherna04a4802016-10-16 20:02:52 -070060static inline bool ipv4_l3mdev_skb(u16 flags)
61{
62 return !!(flags & IPSKB_L3SLAVE);
63}
64
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -030065static inline unsigned int ip_hdrlen(const struct sk_buff *skb)
66{
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -070067 return ip_hdr(skb)->ihl * 4;
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -030068}
69
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +000070struct ipcm_cookie {
Soheil Hassas Yeganeh24025c42016-04-02 23:08:10 -040071 struct sockcm_cookie sockc;
Al Viroc1d18f92006-09-27 18:28:28 -070072 __be32 addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 int oif;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +000074 struct ip_options_rcu *opt;
Oliver Hartkopp2244d072010-08-17 08:59:14 +000075 __u8 tx_flags;
Francesco Fuscof02db312013-09-24 15:43:08 +020076 __u8 ttl;
77 __s16 tos;
78 char priority;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079};
80
81#define IPCB(skb) ((struct inet_skb_parm*)((skb)->cb))
Hannes Frederic Sowa4b261c72014-01-20 03:43:08 +010082#define PKTINFO_SKB_CB(skb) ((struct in_pktinfo *)((skb)->cb))
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
David Ahernfb74c272017-08-07 08:44:16 -070084/* return enslaved device index if relevant */
85static inline int inet_sdif(struct sk_buff *skb)
86{
87#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
88 if (skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
89 return IPCB(skb)->iif;
90#endif
91 return 0;
92}
93
Kirill Tkhai5796ef72018-03-22 12:45:32 +030094/* Special input handler for packets caught by router alert option.
95 They are selected only by protocol field, and then processed likely
96 local ones; but only if someone wants them! Otherwise, router
97 not running rsvpd will kill RSVP.
98
99 It is user level problem, what it will make with them.
100 I have no idea, how it will masquearde or NAT them (it is joke, joke :-)),
101 but receiver should be enough clever f.e. to forward mtrace requests,
102 sent to multicast group to reach destination designated router.
103 */
104
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000105struct ip_ra_chain {
Eric Dumazet43a951e2010-10-25 03:32:44 +0000106 struct ip_ra_chain __rcu *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 struct sock *sk;
Eric Dumazet592fcb92010-06-09 16:21:07 +0000108 union {
109 void (*destructor)(struct sock *);
110 struct sock *saved_sk;
111 };
Eric Dumazet66018502010-06-07 03:12:08 +0000112 struct rcu_head rcu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113};
114
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115/* IP flags. */
116#define IP_CE 0x8000 /* Flag: "Congestion" */
117#define IP_DF 0x4000 /* Flag: "Don't Fragment" */
118#define IP_MF 0x2000 /* Flag: "More Fragments" */
119#define IP_OFFSET 0x1FFF /* "Fragment Offset" part */
120
121#define IP_FRAG_TIME (30 * HZ) /* fragment lifetime */
122
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -0200123struct msghdr;
124struct net_device;
125struct packet_type;
126struct rtable;
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -0200127struct sockaddr;
128
WANG Cong72c1d3b2014-01-10 16:09:45 -0800129int igmp_mc_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
131/*
132 * Functions provided by ip.c
133 */
134
Eric Dumazetcfe673b2015-09-25 07:39:16 -0700135int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700136 __be32 saddr, __be32 daddr,
137 struct ip_options_rcu *opt);
138int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
139 struct net_device *orig_dev);
140int ip_local_deliver(struct sk_buff *skb);
141int ip_mr_input(struct sk_buff *skb);
Eric W. Biedermanede20592015-10-07 16:48:47 -0500142int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb);
143int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb);
Eric W. Biederman694869b2015-06-12 21:55:31 -0500144int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
145 int (*output)(struct net *, struct sock *, struct sk_buff *));
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700146void ip_send_check(struct iphdr *ip);
Eric W. Biedermancf91a992015-10-07 16:48:45 -0500147int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb);
Eric W. Biederman33224b12015-10-07 16:48:46 -0500148int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb);
Eric Dumazetaad88722014-04-15 13:47:15 -0400149
Eric Dumazetb0270e92014-04-15 12:58:34 -0400150int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl);
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700151void ip_init(void);
152int ip_append_data(struct sock *sk, struct flowi4 *fl4,
153 int getfrag(void *from, char *to, int offset, int len,
154 int odd, struct sk_buff *skb),
155 void *from, int len, int protolen,
156 struct ipcm_cookie *ipc,
157 struct rtable **rt,
158 unsigned int flags);
159int ip_generic_getfrag(void *from, char *to, int offset, int len, int odd,
160 struct sk_buff *skb);
161ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
162 int offset, size_t size, int flags);
163struct sk_buff *__ip_make_skb(struct sock *sk, struct flowi4 *fl4,
164 struct sk_buff_head *queue,
165 struct inet_cork *cork);
166int ip_send_skb(struct net *net, struct sk_buff *skb);
167int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4);
168void ip_flush_pending_frames(struct sock *sk);
169struct sk_buff *ip_make_skb(struct sock *sk, struct flowi4 *fl4,
170 int getfrag(void *from, char *to, int offset,
171 int len, int odd, struct sk_buff *skb),
172 void *from, int length, int transhdrlen,
173 struct ipcm_cookie *ipc, struct rtable **rtp,
174 unsigned int flags);
Herbert Xu1c32c5a2011-03-01 02:36:47 +0000175
David S. Miller77968b72011-05-08 17:12:19 -0700176static inline struct sk_buff *ip_finish_skb(struct sock *sk, struct flowi4 *fl4)
Herbert Xu1c32c5a2011-03-01 02:36:47 +0000177{
David S. Miller77968b72011-05-08 17:12:19 -0700178 return __ip_make_skb(sk, fl4, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
Herbert Xu1c32c5a2011-03-01 02:36:47 +0000179}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Francesco Fuscoaa661582013-09-24 15:43:09 +0200181static inline __u8 get_rttos(struct ipcm_cookie* ipc, struct inet_sock *inet)
182{
183 return (ipc->tos != -1) ? RT_TOS(ipc->tos) : RT_TOS(inet->tos);
184}
185
186static inline __u8 get_rtconn_flags(struct ipcm_cookie* ipc, struct sock* sk)
187{
188 return (ipc->tos != -1) ? RT_CONN_FLAGS_TOS(sk, ipc->tos) : RT_CONN_FLAGS(sk);
189}
190
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191/* datagram.c */
Eric Dumazet03645a12015-07-14 08:10:22 +0200192int __ip4_datagram_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700193int ip4_datagram_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700195void ip4_datagram_release_cb(struct sock *sk);
Steffen Klassert8141ed92013-01-21 02:00:03 +0000196
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197struct ip_reply_arg {
Stephen Hemminger82695b32018-02-27 15:48:21 -0800198 struct kvec iov[1];
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700199 int flags;
Al Virod6f5493c2006-11-14 21:26:08 -0800200 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 int csumoffset; /* u16 offset of csum in iov[0].iov_base */
Stephen Hemminger82695b32018-02-27 15:48:21 -0800202 /* -1 if not needed */
Patrick McHardyf0e48db2007-06-04 21:32:46 -0700203 int bound_dev_if;
Eric Dumazet66b13d92011-10-24 03:06:21 -0400204 u8 tos;
Lorenzo Colittie2d118a2016-11-04 02:23:43 +0900205 kuid_t uid;
Stephen Hemminger82695b32018-02-27 15:48:21 -0800206};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700208#define IP_REPLY_ARG_NOSRCCHECK 1
209
KOVACS Krisztian86b08d82008-10-01 07:44:42 -0700210static inline __u8 ip_reply_arg_flowi_flags(const struct ip_reply_arg *arg)
211{
212 return (arg->flags & IP_REPLY_ARG_NOSRCCHECK) ? FLOWI_FLAG_ANYSRC : 0;
213}
214
Eric Dumazetbdbbb852015-01-29 21:35:05 -0800215void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
Eric Dumazet24a2d432014-09-27 09:50:55 -0700216 const struct ip_options *sopt,
217 __be32 daddr, __be32 saddr,
218 const struct ip_reply_arg *arg,
David S. Miller70e73412012-06-28 03:21:41 -0700219 unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
Eric Dumazet4ce3c182010-06-30 13:31:19 -0700221#define IP_INC_STATS(net, field) SNMP_INC_STATS64((net)->mib.ip_statistics, field)
Eric Dumazet13415e42016-04-27 16:44:43 -0700222#define __IP_INC_STATS(net, field) __SNMP_INC_STATS64((net)->mib.ip_statistics, field)
Eric Dumazet4ce3c182010-06-30 13:31:19 -0700223#define IP_ADD_STATS(net, field, val) SNMP_ADD_STATS64((net)->mib.ip_statistics, field, val)
Eric Dumazet13415e42016-04-27 16:44:43 -0700224#define __IP_ADD_STATS(net, field, val) __SNMP_ADD_STATS64((net)->mib.ip_statistics, field, val)
Eric Dumazet4ce3c182010-06-30 13:31:19 -0700225#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 -0700226#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 -0700227#define NET_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.net_statistics, field)
Eric Dumazet13415e42016-04-27 16:44:43 -0700228#define __NET_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.net_statistics, field)
David S. Millerf7324ac2014-03-06 15:03:17 -0500229#define NET_ADD_STATS(net, field, adnd) SNMP_ADD_STATS((net)->mib.net_statistics, field, adnd)
Eric Dumazet13415e42016-04-27 16:44:43 -0700230#define __NET_ADD_STATS(net, field, adnd) __SNMP_ADD_STATS((net)->mib.net_statistics, field, adnd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
Raghavendra K Tc4c6bc32015-08-30 11:29:41 +0530232u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offct);
WANG Cong698365f2014-05-05 15:55:55 -0700233unsigned long snmp_fold_field(void __percpu *mib, int offt);
Eric Dumazet4ce3c182010-06-30 13:31:19 -0700234#if BITS_PER_LONG==32
Raghavendra K Tc4c6bc32015-08-30 11:29:41 +0530235u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offct,
236 size_t syncp_offset);
WANG Cong698365f2014-05-05 15:55:55 -0700237u64 snmp_fold_field64(void __percpu *mib, int offt, size_t sync_off);
Eric Dumazet4ce3c182010-06-30 13:31:19 -0700238#else
Raghavendra K Tc4c6bc32015-08-30 11:29:41 +0530239static inline u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offct,
240 size_t syncp_offset)
241{
242 return snmp_get_cpu_field(mib, cpu, offct);
243
244}
245
WANG Cong698365f2014-05-05 15:55:55 -0700246static inline u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_off)
Eric Dumazet4ce3c182010-06-30 13:31:19 -0700247{
248 return snmp_fold_field(mib, offt);
249}
250#endif
YOSHIFUJI Hideaki33490172007-04-20 15:57:15 -0700251
Jia He6348ef22016-09-30 11:28:58 +0800252#define snmp_get_cpu_field64_batch(buff64, stats_list, mib_statistic, offset) \
253{ \
254 int i, c; \
255 for_each_possible_cpu(c) { \
256 for (i = 0; stats_list[i].name; i++) \
257 buff64[i] += snmp_get_cpu_field64( \
258 mib_statistic, \
259 c, stats_list[i].entry, \
260 offset); \
261 } \
262}
263
264#define snmp_get_cpu_field_batch(buff, stats_list, mib_statistic) \
265{ \
266 int i, c; \
267 for_each_possible_cpu(c) { \
268 for (i = 0; stats_list[i].name; i++) \
269 buff[i] += snmp_get_cpu_field( \
270 mib_statistic, \
271 c, stats_list[i].entry); \
272 } \
273}
274
Eric W. Biederman0bbf87d2013-09-28 14:10:59 -0700275void inet_get_local_port_range(struct net *net, int *low, int *high);
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700276
David S. Millerfcd77db2014-05-15 13:43:14 -0400277#ifdef CONFIG_SYSCTL
WANG Cong122ff242014-05-12 16:04:53 -0700278static inline int inet_is_local_reserved_port(struct net *net, int port)
Amerigo Wange3826f12010-05-05 00:27:06 +0000279{
WANG Cong122ff242014-05-12 16:04:53 -0700280 if (!net->ipv4.sysctl_local_reserved_ports)
281 return 0;
282 return test_bit(port, net->ipv4.sysctl_local_reserved_ports);
Amerigo Wange3826f12010-05-05 00:27:06 +0000283}
WANG Cong20e61da2014-07-25 15:25:08 -0700284
285static inline bool sysctl_dev_name_is_allowed(const char *name)
286{
287 return strcmp(name, "default") != 0 && strcmp(name, "all") != 0;
288}
289
Krister Johansen4548b682017-01-20 17:49:11 -0800290static inline int inet_prot_sock(struct net *net)
291{
292 return net->ipv4.sysctl_ip_prot_sock;
293}
294
WANG Cong122ff242014-05-12 16:04:53 -0700295#else
296static inline int inet_is_local_reserved_port(struct net *net, int port)
297{
298 return 0;
299}
Krister Johansen4548b682017-01-20 17:49:11 -0800300
301static inline int inet_prot_sock(struct net *net)
302{
303 return PROT_SOCK;
304}
WANG Cong122ff242014-05-12 16:04:53 -0700305#endif
Amerigo Wange3826f12010-05-05 00:27:06 +0000306
Deepa Dinamani822c8682016-02-27 00:32:15 -0800307__be32 inet_current_timestamp(void);
308
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300309/* From inetpeer.c */
310extern int inet_peer_threshold;
311extern int inet_peer_minttl;
312extern int inet_peer_maxttl;
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300313
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700314void ipfrag_init(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300315
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700316void ip_static_sysctl_init(void);
Al Virobd7b1532008-07-15 16:00:59 -0400317
Lorenzo Colittie1108612014-05-13 10:17:33 -0700318#define IP4_REPLY_MARK(net, mark) \
319 ((net)->ipv4.sysctl_fwmark_reflect ? (mark) : 0)
320
David S. Millerd18cd552011-06-23 21:28:52 -0700321static inline bool ip_is_fragment(const struct iphdr *iph)
322{
323 return (iph->frag_off & htons(IP_MF | IP_OFFSET)) != 0;
324}
325
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326#ifdef CONFIG_INET
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -0200327#include <net/dst.h>
328
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329/* The function in 2.2 was invalid, producing wrong result for
330 * check=0xFEFF. It was noticed by Arthur Skawina _year_ ago. --ANK(000625) */
331static inline
332int ip_decrease_ttl(struct iphdr *iph)
333{
Al Viro5c78f272006-11-14 21:42:26 -0800334 u32 check = (__force u32)iph->check;
335 check += (__force u32)htons(0x0100);
336 iph->check = (__force __sum16)(check + (check>=0xFFFF));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 return --iph->ttl;
338}
339
Sabrina Dubrocad52e5a72018-03-14 10:21:14 +0100340static inline int ip_mtu_locked(const struct dst_entry *dst)
341{
342 const struct rtable *rt = (const struct rtable *)dst;
343
344 return rt->rt_mtu_locked || dst_metric_locked(dst, RTAX_MTU);
345}
346
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347static inline
Eric Dumazet4e3f5d72015-09-25 07:39:14 -0700348int ip_dont_fragment(const struct sock *sk, const struct dst_entry *dst)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349{
Eric Dumazet4e3f5d72015-09-25 07:39:14 -0700350 u8 pmtudisc = READ_ONCE(inet_sk(sk)->pmtudisc);
351
352 return pmtudisc == IP_PMTUDISC_DO ||
353 (pmtudisc == IP_PMTUDISC_WANT &&
Sabrina Dubrocad52e5a72018-03-14 10:21:14 +0100354 !ip_mtu_locked(dst));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355}
356
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100357static inline bool ip_sk_accept_pmtu(const struct sock *sk)
358{
Hannes Frederic Sowa1b346572014-02-26 01:20:42 +0100359 return inet_sk(sk)->pmtudisc != IP_PMTUDISC_INTERFACE &&
360 inet_sk(sk)->pmtudisc != IP_PMTUDISC_OMIT;
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100361}
362
363static inline bool ip_sk_use_pmtu(const struct sock *sk)
364{
365 return inet_sk(sk)->pmtudisc < IP_PMTUDISC_PROBE;
366}
367
WANG Cong60ff7462014-05-04 16:39:18 -0700368static inline bool ip_sk_ignore_df(const struct sock *sk)
Hannes Frederic Sowa1b346572014-02-26 01:20:42 +0100369{
370 return inet_sk(sk)->pmtudisc < IP_PMTUDISC_DO ||
371 inet_sk(sk)->pmtudisc == IP_PMTUDISC_OMIT;
372}
373
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100374static inline unsigned int ip_dst_mtu_maybe_forward(const struct dst_entry *dst,
375 bool forwarding)
376{
377 struct net *net = dev_net(dst->dev);
378
379 if (net->ipv4.sysctl_ip_fwd_use_pmtu ||
Sabrina Dubrocad52e5a72018-03-14 10:21:14 +0100380 ip_mtu_locked(dst) ||
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100381 !forwarding)
382 return dst_mtu(dst);
383
Eric Dumazetc780a042017-08-16 11:09:12 -0700384 return min(READ_ONCE(dst->dev->mtu), IP_MAX_MTU);
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100385}
386
Shmulik Ladkanifedbb6b42016-06-29 21:47:03 +0300387static inline unsigned int ip_skb_dst_mtu(struct sock *sk,
388 const struct sk_buff *skb)
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100389{
Eric Dumazetcaf3f262015-10-04 21:08:08 -0700390 if (!sk || !sk_fullsock(sk) || ip_sk_use_pmtu(sk)) {
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100391 bool forwarding = IPCB(skb)->flags & IPSKB_FORWARDED;
Eric Dumazetcaf3f262015-10-04 21:08:08 -0700392
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100393 return ip_dst_mtu_maybe_forward(skb_dst(skb), forwarding);
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100394 }
Eric Dumazetcaf3f262015-10-04 21:08:08 -0700395
Eric Dumazetc780a042017-08-16 11:09:12 -0700396 return min(READ_ONCE(skb_dst(skb)->dev->mtu), IP_MAX_MTU);
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100397}
398
Eric Dumazet04ca6972014-07-26 08:58:10 +0200399u32 ip_idents_reserve(u32 hash, int segs);
Hannes Frederic Sowab6a77192015-03-25 17:07:44 +0100400void __ip_select_ident(struct net *net, struct iphdr *iph, int segs);
Eric Dumazet73f156a2014-06-02 05:26:03 -0700401
Hannes Frederic Sowab6a77192015-03-25 17:07:44 +0100402static inline void ip_select_ident_segs(struct net *net, struct sk_buff *skb,
403 struct sock *sk, int segs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404{
Ansis Atteka703133d2013-09-18 15:29:53 -0700405 struct iphdr *iph = ip_hdr(skb);
406
WANG Cong60ff7462014-05-04 16:39:18 -0700407 if ((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 /* This is only to work around buggy Windows95/2000
409 * VJ compression implementations. If the ID field
410 * does not change, they drop every other packet in
411 * a TCP stream using header compression.
412 */
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000413 if (sk && inet_sk(sk)->inet_daddr) {
414 iph->id = htons(inet_sk(sk)->inet_id);
Eric Dumazet73f156a2014-06-02 05:26:03 -0700415 inet_sk(sk)->inet_id += segs;
416 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 iph->id = 0;
Eric Dumazet73f156a2014-06-02 05:26:03 -0700418 }
419 } else {
Hannes Frederic Sowab6a77192015-03-25 17:07:44 +0100420 __ip_select_ident(net, iph, segs);
Eric Dumazet73f156a2014-06-02 05:26:03 -0700421 }
422}
423
Hannes Frederic Sowab6a77192015-03-25 17:07:44 +0100424static inline void ip_select_ident(struct net *net, struct sk_buff *skb,
425 struct sock *sk)
Eric Dumazet73f156a2014-06-02 05:26:03 -0700426{
Hannes Frederic Sowab6a77192015-03-25 17:07:44 +0100427 ip_select_ident_segs(net, skb, sk, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428}
429
Tom Herberted70fcf2014-05-02 16:29:38 -0700430static inline __wsum inet_compute_pseudo(struct sk_buff *skb, int proto)
431{
432 return csum_tcpudp_nofold(ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
433 skb->len, proto, 0);
434}
435
Tom Herbertc3f83242015-06-04 09:16:40 -0700436/* copy IPv4 saddr & daddr to flow_keys, possibly using 64bit load/store
437 * Equivalent to : flow->v4addrs.src = iph->saddr;
438 * flow->v4addrs.dst = iph->daddr;
439 */
440static inline void iph_to_flow_copy_v4addrs(struct flow_keys *flow,
441 const struct iphdr *iph)
442{
443 BUILD_BUG_ON(offsetof(typeof(flow->addrs), v4addrs.dst) !=
444 offsetof(typeof(flow->addrs), v4addrs.src) +
445 sizeof(flow->addrs.v4addrs.src));
446 memcpy(&flow->addrs.v4addrs, &iph->saddr, sizeof(flow->addrs.v4addrs));
447 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
448}
449
Tom Herbert1933a782014-08-22 13:34:04 -0700450static inline __wsum inet_gro_compute_pseudo(struct sk_buff *skb, int proto)
451{
452 const struct iphdr *iph = skb_gro_network_header(skb);
453
454 return csum_tcpudp_nofold(iph->saddr, iph->daddr,
455 skb_gro_len(skb), proto, 0);
456}
457
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458/*
459 * Map a multicast IP onto multicast MAC for type ethernet.
460 */
461
Al Viro714e85b2006-11-14 20:51:49 -0800462static inline void ip_eth_mc_map(__be32 naddr, char *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463{
Al Viro714e85b2006-11-14 20:51:49 -0800464 __u32 addr=ntohl(naddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 buf[0]=0x01;
466 buf[1]=0x00;
467 buf[2]=0x5e;
468 buf[5]=addr&0xFF;
469 addr>>=8;
470 buf[4]=addr&0xFF;
471 addr>>=8;
472 buf[3]=addr&0x7F;
473}
474
475/*
476 * Map a multicast IP onto multicast MAC for type IP-over-InfiniBand.
477 * Leave P_Key as 0 to be filled in by driver.
478 */
479
Rolf Manderscheida9e527e2007-12-10 13:38:41 -0700480static inline void ip_ib_mc_map(__be32 naddr, const unsigned char *broadcast, char *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481{
Al Viro714e85b2006-11-14 20:51:49 -0800482 __u32 addr;
Rolf Manderscheida9e527e2007-12-10 13:38:41 -0700483 unsigned char scope = broadcast[5] & 0xF;
484
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 buf[0] = 0; /* Reserved */
486 buf[1] = 0xff; /* Multicast QPN */
487 buf[2] = 0xff;
488 buf[3] = 0xff;
Al Viro714e85b2006-11-14 20:51:49 -0800489 addr = ntohl(naddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 buf[4] = 0xff;
Rolf Manderscheida9e527e2007-12-10 13:38:41 -0700491 buf[5] = 0x10 | scope; /* scope from broadcast address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 buf[6] = 0x40; /* IPv4 signature */
493 buf[7] = 0x1b;
Rolf Manderscheida9e527e2007-12-10 13:38:41 -0700494 buf[8] = broadcast[8]; /* P_Key */
495 buf[9] = broadcast[9];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 buf[10] = 0;
497 buf[11] = 0;
498 buf[12] = 0;
499 buf[13] = 0;
500 buf[14] = 0;
501 buf[15] = 0;
502 buf[19] = addr & 0xff;
503 addr >>= 8;
504 buf[18] = addr & 0xff;
505 addr >>= 8;
506 buf[17] = addr & 0xff;
507 addr >>= 8;
508 buf[16] = addr & 0x0f;
509}
510
Timo Teräs93ca3bb2011-03-28 22:40:53 +0000511static inline void ip_ipgre_mc_map(__be32 naddr, const unsigned char *broadcast, char *buf)
512{
513 if ((broadcast[0] | broadcast[1] | broadcast[2] | broadcast[3]) != 0)
514 memcpy(buf, broadcast, 4);
515 else
516 memcpy(buf, &naddr, sizeof(naddr));
517}
518
Eric Dumazetdfd56b82011-12-10 09:48:31 +0000519#if IS_ENABLED(CONFIG_IPV6)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520#include <linux/ipv6.h>
521#endif
522
523static __inline__ void inet_reset_saddr(struct sock *sk)
524{
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000525 inet_sk(sk)->inet_rcv_saddr = inet_sk(sk)->inet_saddr = 0;
Eric Dumazetdfd56b82011-12-10 09:48:31 +0000526#if IS_ENABLED(CONFIG_IPV6)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 if (sk->sk_family == PF_INET6) {
528 struct ipv6_pinfo *np = inet6_sk(sk);
529
530 memset(&np->saddr, 0, sizeof(np->saddr));
Eric Dumazetefe42082013-10-03 15:42:29 -0700531 memset(&sk->sk_v6_rcv_saddr, 0, sizeof(sk->sk_v6_rcv_saddr));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 }
533#endif
534}
535
536#endif
537
David Ahern72afa352015-08-27 16:06:59 -0700538static inline unsigned int ipv4_addr_hash(__be32 ip)
539{
540 return (__force unsigned int) ip;
541}
542
Martin KaFai Lauf0b1e642017-12-01 12:52:30 -0800543static inline u32 ipv4_portaddr_hash(const struct net *net,
544 __be32 saddr,
545 unsigned int port)
546{
547 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
548}
549
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700550bool ip_call_ra_chain(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
552/*
Rami Rosenb7982322008-01-01 21:13:09 -0800553 * Functions provided by ip_fragment.c
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 */
555
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000556enum ip_defrag_users {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 IP_DEFRAG_LOCAL_DELIVER,
558 IP_DEFRAG_CALL_RA_CHAIN,
559 IP_DEFRAG_CONNTRACK_IN,
Alexey Dobriyan4be929b2010-05-24 14:33:03 -0700560 __IP_DEFRAG_CONNTRACK_IN_END = IP_DEFRAG_CONNTRACK_IN + USHRT_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 IP_DEFRAG_CONNTRACK_OUT,
Alexey Dobriyan4be929b2010-05-24 14:33:03 -0700562 __IP_DEFRAG_CONNTRACK_OUT_END = IP_DEFRAG_CONNTRACK_OUT + USHRT_MAX,
Patrick McHardy8fa9ff62009-12-15 16:59:59 +0100563 IP_DEFRAG_CONNTRACK_BRIDGE_IN,
Alexey Dobriyan4be929b2010-05-24 14:33:03 -0700564 __IP_DEFRAG_CONNTRACK_BRIDGE_IN = IP_DEFRAG_CONNTRACK_BRIDGE_IN + USHRT_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 IP_DEFRAG_VS_IN,
566 IP_DEFRAG_VS_OUT,
David S. Miller595fc712011-07-05 01:05:48 -0700567 IP_DEFRAG_VS_FWD,
568 IP_DEFRAG_AF_PACKET,
Eric Dumazetbc416d92011-10-06 10:28:31 +0000569 IP_DEFRAG_MACVLAN,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570};
571
Andy Zhou5cf42282015-05-15 14:15:35 -0700572/* Return true if the value of 'user' is between 'lower_bond'
573 * and 'upper_bond' inclusively.
574 */
575static inline bool ip_defrag_user_in_between(u32 user,
576 enum ip_defrag_users lower_bond,
577 enum ip_defrag_users upper_bond)
578{
579 return user >= lower_bond && user <= upper_bond;
580}
581
Eric W. Biederman19bcf9f2015-10-09 13:44:54 -0500582int ip_defrag(struct net *net, struct sk_buff *skb, u32 user);
Eric Dumazetbc416d92011-10-06 10:28:31 +0000583#ifdef CONFIG_INET
Eric W. Biederman19bcf9f2015-10-09 13:44:54 -0500584struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user);
Eric Dumazetbc416d92011-10-06 10:28:31 +0000585#else
Eric W. Biederman19bcf9f2015-10-09 13:44:54 -0500586static inline struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
Eric Dumazetbc416d92011-10-06 10:28:31 +0000587{
588 return skb;
589}
590#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
592/*
593 * Functions provided by ip_forward.c
594 */
Stephen Hemminger82695b32018-02-27 15:48:21 -0800595
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700596int ip_forward(struct sk_buff *skb);
Stephen Hemminger82695b32018-02-27 15:48:21 -0800597
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598/*
599 * Functions provided by ip_options.c
600 */
Stephen Hemminger82695b32018-02-27 15:48:21 -0800601
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700602void ip_options_build(struct sk_buff *skb, struct ip_options *opt,
603 __be32 daddr, struct rtable *rt, int is_frag);
Eric Dumazet24a2d432014-09-27 09:50:55 -0700604
Paolo Abeni91ed1e62017-08-03 18:07:06 +0200605int __ip_options_echo(struct net *net, struct ip_options *dopt,
606 struct sk_buff *skb, const struct ip_options *sopt);
607static inline int ip_options_echo(struct net *net, struct ip_options *dopt,
608 struct sk_buff *skb)
Eric Dumazet24a2d432014-09-27 09:50:55 -0700609{
Paolo Abeni91ed1e62017-08-03 18:07:06 +0200610 return __ip_options_echo(net, dopt, skb, &IPCB(skb)->opt);
Eric Dumazet24a2d432014-09-27 09:50:55 -0700611}
612
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700613void ip_options_fragment(struct sk_buff *skb);
614int ip_options_compile(struct net *net, struct ip_options *opt,
615 struct sk_buff *skb);
616int ip_options_get(struct net *net, struct ip_options_rcu **optp,
617 unsigned char *data, int optlen);
618int ip_options_get_from_user(struct net *net, struct ip_options_rcu **optp,
619 unsigned char __user *data, int optlen);
620void ip_options_undo(struct ip_options *opt);
621void ip_forward_options(struct sk_buff *skb);
622int ip_options_rcv_srr(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
624/*
625 * Functions provided by ip_sockglue.c
626 */
627
Shawn Bohrerfbf88662013-10-07 11:01:40 -0500628void ipv4_pktinfo_prepare(const struct sock *sk, struct sk_buff *skb);
Paolo Abeniad959032016-11-04 11:28:58 +0100629void ip_cmsg_recv_offset(struct msghdr *msg, struct sock *sk,
630 struct sk_buff *skb, int tlen, int offset);
Soheil Hassas Yeganeh24025c42016-04-02 23:08:10 -0400631int ip_cmsg_send(struct sock *sk, struct msghdr *msg,
Hannes Frederic Sowac8e6ad02014-02-18 21:38:08 +0100632 struct ipcm_cookie *ipc, bool allow_ipv6);
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700633int ip_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
634 unsigned int optlen);
635int ip_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
636 int __user *optlen);
637int compat_ip_setsockopt(struct sock *sk, int level, int optname,
638 char __user *optval, unsigned int optlen);
639int compat_ip_getsockopt(struct sock *sk, int level, int optname,
640 char __user *optval, int __user *optlen);
641int ip_ra_control(struct sock *sk, unsigned char on,
642 void (*destructor)(struct sock *));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
Hannes Frederic Sowa85fbaa72013-11-23 00:46:12 +0100644int ip_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len);
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700645void ip_icmp_error(struct sock *sk, struct sk_buff *skb, int err, __be16 port,
646 u32 info, u8 *payload);
647void ip_local_error(struct sock *sk, int err, __be32 daddr, __be16 dport,
648 u32 info);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649
Tom Herbert5961de92015-01-05 13:56:16 -0800650static inline void ip_cmsg_recv(struct msghdr *msg, struct sk_buff *skb)
651{
Paolo Abeniad959032016-11-04 11:28:58 +0100652 ip_cmsg_recv_offset(msg, skb->sk, skb, 0, 0);
Tom Herbert5961de92015-01-05 13:56:16 -0800653}
654
Eric Dumazet4cdf5072014-09-19 07:38:40 -0700655bool icmp_global_allow(void);
656extern int sysctl_icmp_msgs_per_sec;
657extern int sysctl_icmp_msgs_burst;
658
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300659#ifdef CONFIG_PROC_FS
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700660int ip_misc_proc_init(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -0300661#endif
662
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663#endif /* _IP_H */