xref: /openbmc/linux/include/net/ip.h (revision c0e297dc)
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  *
10  * Authors:	Ross Biro
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 
25 #include <linux/types.h>
26 #include <linux/ip.h>
27 #include <linux/in.h>
28 #include <linux/skbuff.h>
29 
30 #include <net/inet_sock.h>
31 #include <net/route.h>
32 #include <net/snmp.h>
33 #include <net/flow.h>
34 #include <net/flow_dissector.h>
35 
36 struct sock;
37 
38 struct inet_skb_parm {
39 	struct ip_options	opt;		/* Compiled IP options		*/
40 	unsigned char		flags;
41 
42 #define IPSKB_FORWARDED		BIT(0)
43 #define IPSKB_XFRM_TUNNEL_SIZE	BIT(1)
44 #define IPSKB_XFRM_TRANSFORMED	BIT(2)
45 #define IPSKB_FRAG_COMPLETE	BIT(3)
46 #define IPSKB_REROUTED		BIT(4)
47 #define IPSKB_DOREDIRECT	BIT(5)
48 #define IPSKB_FRAG_PMTU		BIT(6)
49 
50 	u16			frag_max_size;
51 };
52 
53 static inline unsigned int ip_hdrlen(const struct sk_buff *skb)
54 {
55 	return ip_hdr(skb)->ihl * 4;
56 }
57 
58 struct ipcm_cookie {
59 	__be32			addr;
60 	int			oif;
61 	struct ip_options_rcu	*opt;
62 	__u8			tx_flags;
63 	__u8			ttl;
64 	__s16			tos;
65 	char			priority;
66 };
67 
68 #define IPCB(skb) ((struct inet_skb_parm*)((skb)->cb))
69 #define PKTINFO_SKB_CB(skb) ((struct in_pktinfo *)((skb)->cb))
70 
71 struct ip_ra_chain {
72 	struct ip_ra_chain __rcu *next;
73 	struct sock		*sk;
74 	union {
75 		void			(*destructor)(struct sock *);
76 		struct sock		*saved_sk;
77 	};
78 	struct rcu_head		rcu;
79 };
80 
81 extern struct ip_ra_chain __rcu *ip_ra_chain;
82 
83 /* IP flags. */
84 #define IP_CE		0x8000		/* Flag: "Congestion"		*/
85 #define IP_DF		0x4000		/* Flag: "Don't Fragment"	*/
86 #define IP_MF		0x2000		/* Flag: "More Fragments"	*/
87 #define IP_OFFSET	0x1FFF		/* "Fragment Offset" part	*/
88 
89 #define IP_FRAG_TIME	(30 * HZ)		/* fragment lifetime	*/
90 
91 struct msghdr;
92 struct net_device;
93 struct packet_type;
94 struct rtable;
95 struct sockaddr;
96 
97 int igmp_mc_init(void);
98 
99 /*
100  *	Functions provided by ip.c
101  */
102 
103 int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
104 			  __be32 saddr, __be32 daddr,
105 			  struct ip_options_rcu *opt);
106 int ip_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt,
107 	   struct net_device *orig_dev);
108 int ip_local_deliver(struct sk_buff *skb);
109 int ip_mr_input(struct sk_buff *skb);
110 int ip_output(struct sock *sk, struct sk_buff *skb);
111 int ip_mc_output(struct sock *sk, struct sk_buff *skb);
112 int ip_do_fragment(struct sock *sk, struct sk_buff *skb,
113 		   int (*output)(struct sock *, struct sk_buff *));
114 void ip_send_check(struct iphdr *ip);
115 int __ip_local_out(struct sk_buff *skb);
116 int ip_local_out_sk(struct sock *sk, struct sk_buff *skb);
117 static inline int ip_local_out(struct sk_buff *skb)
118 {
119 	return ip_local_out_sk(skb->sk, skb);
120 }
121 
122 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl);
123 void ip_init(void);
124 int ip_append_data(struct sock *sk, struct flowi4 *fl4,
125 		   int getfrag(void *from, char *to, int offset, int len,
126 			       int odd, struct sk_buff *skb),
127 		   void *from, int len, int protolen,
128 		   struct ipcm_cookie *ipc,
129 		   struct rtable **rt,
130 		   unsigned int flags);
131 int ip_generic_getfrag(void *from, char *to, int offset, int len, int odd,
132 		       struct sk_buff *skb);
133 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
134 		       int offset, size_t size, int flags);
135 struct sk_buff *__ip_make_skb(struct sock *sk, struct flowi4 *fl4,
136 			      struct sk_buff_head *queue,
137 			      struct inet_cork *cork);
138 int ip_send_skb(struct net *net, struct sk_buff *skb);
139 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4);
140 void ip_flush_pending_frames(struct sock *sk);
141 struct sk_buff *ip_make_skb(struct sock *sk, struct flowi4 *fl4,
142 			    int getfrag(void *from, char *to, int offset,
143 					int len, int odd, struct sk_buff *skb),
144 			    void *from, int length, int transhdrlen,
145 			    struct ipcm_cookie *ipc, struct rtable **rtp,
146 			    unsigned int flags);
147 
148 static inline struct sk_buff *ip_finish_skb(struct sock *sk, struct flowi4 *fl4)
149 {
150 	return __ip_make_skb(sk, fl4, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
151 }
152 
153 static inline __u8 get_rttos(struct ipcm_cookie* ipc, struct inet_sock *inet)
154 {
155 	return (ipc->tos != -1) ? RT_TOS(ipc->tos) : RT_TOS(inet->tos);
156 }
157 
158 static inline __u8 get_rtconn_flags(struct ipcm_cookie* ipc, struct sock* sk)
159 {
160 	return (ipc->tos != -1) ? RT_CONN_FLAGS_TOS(sk, ipc->tos) : RT_CONN_FLAGS(sk);
161 }
162 
163 /* datagram.c */
164 int __ip4_datagram_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
165 int ip4_datagram_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
166 
167 void ip4_datagram_release_cb(struct sock *sk);
168 
169 struct ip_reply_arg {
170 	struct kvec iov[1];
171 	int	    flags;
172 	__wsum 	    csum;
173 	int	    csumoffset; /* u16 offset of csum in iov[0].iov_base */
174 				/* -1 if not needed */
175 	int	    bound_dev_if;
176 	u8  	    tos;
177 };
178 
179 #define IP_REPLY_ARG_NOSRCCHECK 1
180 
181 static inline __u8 ip_reply_arg_flowi_flags(const struct ip_reply_arg *arg)
182 {
183 	return (arg->flags & IP_REPLY_ARG_NOSRCCHECK) ? FLOWI_FLAG_ANYSRC : 0;
184 }
185 
186 void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
187 			   const struct ip_options *sopt,
188 			   __be32 daddr, __be32 saddr,
189 			   const struct ip_reply_arg *arg,
190 			   unsigned int len);
191 
192 #define IP_INC_STATS(net, field)	SNMP_INC_STATS64((net)->mib.ip_statistics, field)
193 #define IP_INC_STATS_BH(net, field)	SNMP_INC_STATS64_BH((net)->mib.ip_statistics, field)
194 #define IP_ADD_STATS(net, field, val)	SNMP_ADD_STATS64((net)->mib.ip_statistics, field, val)
195 #define IP_ADD_STATS_BH(net, field, val) SNMP_ADD_STATS64_BH((net)->mib.ip_statistics, field, val)
196 #define IP_UPD_PO_STATS(net, field, val) SNMP_UPD_PO_STATS64((net)->mib.ip_statistics, field, val)
197 #define IP_UPD_PO_STATS_BH(net, field, val) SNMP_UPD_PO_STATS64_BH((net)->mib.ip_statistics, field, val)
198 #define NET_INC_STATS(net, field)	SNMP_INC_STATS((net)->mib.net_statistics, field)
199 #define NET_INC_STATS_BH(net, field)	SNMP_INC_STATS_BH((net)->mib.net_statistics, field)
200 #define NET_INC_STATS_USER(net, field) 	SNMP_INC_STATS_USER((net)->mib.net_statistics, field)
201 #define NET_ADD_STATS(net, field, adnd)	SNMP_ADD_STATS((net)->mib.net_statistics, field, adnd)
202 #define NET_ADD_STATS_BH(net, field, adnd) SNMP_ADD_STATS_BH((net)->mib.net_statistics, field, adnd)
203 #define NET_ADD_STATS_USER(net, field, adnd) SNMP_ADD_STATS_USER((net)->mib.net_statistics, field, adnd)
204 
205 unsigned long snmp_fold_field(void __percpu *mib, int offt);
206 #if BITS_PER_LONG==32
207 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t sync_off);
208 #else
209 static inline u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_off)
210 {
211 	return snmp_fold_field(mib, offt);
212 }
213 #endif
214 
215 void inet_get_local_port_range(struct net *net, int *low, int *high);
216 
217 #ifdef CONFIG_SYSCTL
218 static inline int inet_is_local_reserved_port(struct net *net, int port)
219 {
220 	if (!net->ipv4.sysctl_local_reserved_ports)
221 		return 0;
222 	return test_bit(port, net->ipv4.sysctl_local_reserved_ports);
223 }
224 
225 static inline bool sysctl_dev_name_is_allowed(const char *name)
226 {
227 	return strcmp(name, "default") != 0  && strcmp(name, "all") != 0;
228 }
229 
230 #else
231 static inline int inet_is_local_reserved_port(struct net *net, int port)
232 {
233 	return 0;
234 }
235 #endif
236 
237 /* From inetpeer.c */
238 extern int inet_peer_threshold;
239 extern int inet_peer_minttl;
240 extern int inet_peer_maxttl;
241 
242 /* From ip_input.c */
243 extern int sysctl_ip_early_demux;
244 
245 /* From ip_output.c */
246 extern int sysctl_ip_dynaddr;
247 
248 void ipfrag_init(void);
249 
250 void ip_static_sysctl_init(void);
251 
252 #define IP4_REPLY_MARK(net, mark) \
253 	((net)->ipv4.sysctl_fwmark_reflect ? (mark) : 0)
254 
255 static inline bool ip_is_fragment(const struct iphdr *iph)
256 {
257 	return (iph->frag_off & htons(IP_MF | IP_OFFSET)) != 0;
258 }
259 
260 #ifdef CONFIG_INET
261 #include <net/dst.h>
262 
263 /* The function in 2.2 was invalid, producing wrong result for
264  * check=0xFEFF. It was noticed by Arthur Skawina _year_ ago. --ANK(000625) */
265 static inline
266 int ip_decrease_ttl(struct iphdr *iph)
267 {
268 	u32 check = (__force u32)iph->check;
269 	check += (__force u32)htons(0x0100);
270 	iph->check = (__force __sum16)(check + (check>=0xFFFF));
271 	return --iph->ttl;
272 }
273 
274 static inline
275 int ip_dont_fragment(struct sock *sk, struct dst_entry *dst)
276 {
277 	return  inet_sk(sk)->pmtudisc == IP_PMTUDISC_DO ||
278 		(inet_sk(sk)->pmtudisc == IP_PMTUDISC_WANT &&
279 		 !(dst_metric_locked(dst, RTAX_MTU)));
280 }
281 
282 static inline bool ip_sk_accept_pmtu(const struct sock *sk)
283 {
284 	return inet_sk(sk)->pmtudisc != IP_PMTUDISC_INTERFACE &&
285 	       inet_sk(sk)->pmtudisc != IP_PMTUDISC_OMIT;
286 }
287 
288 static inline bool ip_sk_use_pmtu(const struct sock *sk)
289 {
290 	return inet_sk(sk)->pmtudisc < IP_PMTUDISC_PROBE;
291 }
292 
293 static inline bool ip_sk_ignore_df(const struct sock *sk)
294 {
295 	return inet_sk(sk)->pmtudisc < IP_PMTUDISC_DO ||
296 	       inet_sk(sk)->pmtudisc == IP_PMTUDISC_OMIT;
297 }
298 
299 static inline unsigned int ip_dst_mtu_maybe_forward(const struct dst_entry *dst,
300 						    bool forwarding)
301 {
302 	struct net *net = dev_net(dst->dev);
303 
304 	if (net->ipv4.sysctl_ip_fwd_use_pmtu ||
305 	    dst_metric_locked(dst, RTAX_MTU) ||
306 	    !forwarding)
307 		return dst_mtu(dst);
308 
309 	return min(dst->dev->mtu, IP_MAX_MTU);
310 }
311 
312 static inline unsigned int ip_skb_dst_mtu(const struct sk_buff *skb)
313 {
314 	if (!skb->sk || ip_sk_use_pmtu(skb->sk)) {
315 		bool forwarding = IPCB(skb)->flags & IPSKB_FORWARDED;
316 		return ip_dst_mtu_maybe_forward(skb_dst(skb), forwarding);
317 	} else {
318 		return min(skb_dst(skb)->dev->mtu, IP_MAX_MTU);
319 	}
320 }
321 
322 u32 ip_idents_reserve(u32 hash, int segs);
323 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs);
324 
325 static inline void ip_select_ident_segs(struct net *net, struct sk_buff *skb,
326 					struct sock *sk, int segs)
327 {
328 	struct iphdr *iph = ip_hdr(skb);
329 
330 	if ((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) {
331 		/* This is only to work around buggy Windows95/2000
332 		 * VJ compression implementations.  If the ID field
333 		 * does not change, they drop every other packet in
334 		 * a TCP stream using header compression.
335 		 */
336 		if (sk && inet_sk(sk)->inet_daddr) {
337 			iph->id = htons(inet_sk(sk)->inet_id);
338 			inet_sk(sk)->inet_id += segs;
339 		} else {
340 			iph->id = 0;
341 		}
342 	} else {
343 		__ip_select_ident(net, iph, segs);
344 	}
345 }
346 
347 static inline void ip_select_ident(struct net *net, struct sk_buff *skb,
348 				   struct sock *sk)
349 {
350 	ip_select_ident_segs(net, skb, sk, 1);
351 }
352 
353 static inline __wsum inet_compute_pseudo(struct sk_buff *skb, int proto)
354 {
355 	return csum_tcpudp_nofold(ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
356 				  skb->len, proto, 0);
357 }
358 
359 /* copy IPv4 saddr & daddr to flow_keys, possibly using 64bit load/store
360  * Equivalent to :	flow->v4addrs.src = iph->saddr;
361  *			flow->v4addrs.dst = iph->daddr;
362  */
363 static inline void iph_to_flow_copy_v4addrs(struct flow_keys *flow,
364 					    const struct iphdr *iph)
365 {
366 	BUILD_BUG_ON(offsetof(typeof(flow->addrs), v4addrs.dst) !=
367 		     offsetof(typeof(flow->addrs), v4addrs.src) +
368 			      sizeof(flow->addrs.v4addrs.src));
369 	memcpy(&flow->addrs.v4addrs, &iph->saddr, sizeof(flow->addrs.v4addrs));
370 	flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
371 }
372 
373 static inline void inet_set_txhash(struct sock *sk)
374 {
375 	struct inet_sock *inet = inet_sk(sk);
376 	struct flow_keys keys;
377 
378 	memset(&keys, 0, sizeof(keys));
379 
380 	keys.addrs.v4addrs.src = inet->inet_saddr;
381 	keys.addrs.v4addrs.dst = inet->inet_daddr;
382 	keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
383 	keys.ports.src = inet->inet_sport;
384 	keys.ports.dst = inet->inet_dport;
385 
386 	sk->sk_txhash = flow_hash_from_keys(&keys);
387 }
388 
389 static inline __wsum inet_gro_compute_pseudo(struct sk_buff *skb, int proto)
390 {
391 	const struct iphdr *iph = skb_gro_network_header(skb);
392 
393 	return csum_tcpudp_nofold(iph->saddr, iph->daddr,
394 				  skb_gro_len(skb), proto, 0);
395 }
396 
397 /*
398  *	Map a multicast IP onto multicast MAC for type ethernet.
399  */
400 
401 static inline void ip_eth_mc_map(__be32 naddr, char *buf)
402 {
403 	__u32 addr=ntohl(naddr);
404 	buf[0]=0x01;
405 	buf[1]=0x00;
406 	buf[2]=0x5e;
407 	buf[5]=addr&0xFF;
408 	addr>>=8;
409 	buf[4]=addr&0xFF;
410 	addr>>=8;
411 	buf[3]=addr&0x7F;
412 }
413 
414 /*
415  *	Map a multicast IP onto multicast MAC for type IP-over-InfiniBand.
416  *	Leave P_Key as 0 to be filled in by driver.
417  */
418 
419 static inline void ip_ib_mc_map(__be32 naddr, const unsigned char *broadcast, char *buf)
420 {
421 	__u32 addr;
422 	unsigned char scope = broadcast[5] & 0xF;
423 
424 	buf[0]  = 0;		/* Reserved */
425 	buf[1]  = 0xff;		/* Multicast QPN */
426 	buf[2]  = 0xff;
427 	buf[3]  = 0xff;
428 	addr    = ntohl(naddr);
429 	buf[4]  = 0xff;
430 	buf[5]  = 0x10 | scope;	/* scope from broadcast address */
431 	buf[6]  = 0x40;		/* IPv4 signature */
432 	buf[7]  = 0x1b;
433 	buf[8]  = broadcast[8];		/* P_Key */
434 	buf[9]  = broadcast[9];
435 	buf[10] = 0;
436 	buf[11] = 0;
437 	buf[12] = 0;
438 	buf[13] = 0;
439 	buf[14] = 0;
440 	buf[15] = 0;
441 	buf[19] = addr & 0xff;
442 	addr  >>= 8;
443 	buf[18] = addr & 0xff;
444 	addr  >>= 8;
445 	buf[17] = addr & 0xff;
446 	addr  >>= 8;
447 	buf[16] = addr & 0x0f;
448 }
449 
450 static inline void ip_ipgre_mc_map(__be32 naddr, const unsigned char *broadcast, char *buf)
451 {
452 	if ((broadcast[0] | broadcast[1] | broadcast[2] | broadcast[3]) != 0)
453 		memcpy(buf, broadcast, 4);
454 	else
455 		memcpy(buf, &naddr, sizeof(naddr));
456 }
457 
458 #if IS_ENABLED(CONFIG_IPV6)
459 #include <linux/ipv6.h>
460 #endif
461 
462 static __inline__ void inet_reset_saddr(struct sock *sk)
463 {
464 	inet_sk(sk)->inet_rcv_saddr = inet_sk(sk)->inet_saddr = 0;
465 #if IS_ENABLED(CONFIG_IPV6)
466 	if (sk->sk_family == PF_INET6) {
467 		struct ipv6_pinfo *np = inet6_sk(sk);
468 
469 		memset(&np->saddr, 0, sizeof(np->saddr));
470 		memset(&sk->sk_v6_rcv_saddr, 0, sizeof(sk->sk_v6_rcv_saddr));
471 	}
472 #endif
473 }
474 
475 #endif
476 
477 bool ip_call_ra_chain(struct sk_buff *skb);
478 
479 /*
480  *	Functions provided by ip_fragment.c
481  */
482 
483 enum ip_defrag_users {
484 	IP_DEFRAG_LOCAL_DELIVER,
485 	IP_DEFRAG_CALL_RA_CHAIN,
486 	IP_DEFRAG_CONNTRACK_IN,
487 	__IP_DEFRAG_CONNTRACK_IN_END	= IP_DEFRAG_CONNTRACK_IN + USHRT_MAX,
488 	IP_DEFRAG_CONNTRACK_OUT,
489 	__IP_DEFRAG_CONNTRACK_OUT_END	= IP_DEFRAG_CONNTRACK_OUT + USHRT_MAX,
490 	IP_DEFRAG_CONNTRACK_BRIDGE_IN,
491 	__IP_DEFRAG_CONNTRACK_BRIDGE_IN = IP_DEFRAG_CONNTRACK_BRIDGE_IN + USHRT_MAX,
492 	IP_DEFRAG_VS_IN,
493 	IP_DEFRAG_VS_OUT,
494 	IP_DEFRAG_VS_FWD,
495 	IP_DEFRAG_AF_PACKET,
496 	IP_DEFRAG_MACVLAN,
497 };
498 
499 /* Return true if the value of 'user' is between 'lower_bond'
500  * and 'upper_bond' inclusively.
501  */
502 static inline bool ip_defrag_user_in_between(u32 user,
503 					     enum ip_defrag_users lower_bond,
504 					     enum ip_defrag_users upper_bond)
505 {
506 	return user >= lower_bond && user <= upper_bond;
507 }
508 
509 int ip_defrag(struct sk_buff *skb, u32 user);
510 #ifdef CONFIG_INET
511 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user);
512 #else
513 static inline struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
514 {
515 	return skb;
516 }
517 #endif
518 int ip_frag_mem(struct net *net);
519 
520 /*
521  *	Functions provided by ip_forward.c
522  */
523 
524 int ip_forward(struct sk_buff *skb);
525 
526 /*
527  *	Functions provided by ip_options.c
528  */
529 
530 void ip_options_build(struct sk_buff *skb, struct ip_options *opt,
531 		      __be32 daddr, struct rtable *rt, int is_frag);
532 
533 int __ip_options_echo(struct ip_options *dopt, struct sk_buff *skb,
534 		      const struct ip_options *sopt);
535 static inline int ip_options_echo(struct ip_options *dopt, struct sk_buff *skb)
536 {
537 	return __ip_options_echo(dopt, skb, &IPCB(skb)->opt);
538 }
539 
540 void ip_options_fragment(struct sk_buff *skb);
541 int ip_options_compile(struct net *net, struct ip_options *opt,
542 		       struct sk_buff *skb);
543 int ip_options_get(struct net *net, struct ip_options_rcu **optp,
544 		   unsigned char *data, int optlen);
545 int ip_options_get_from_user(struct net *net, struct ip_options_rcu **optp,
546 			     unsigned char __user *data, int optlen);
547 void ip_options_undo(struct ip_options *opt);
548 void ip_forward_options(struct sk_buff *skb);
549 int ip_options_rcv_srr(struct sk_buff *skb);
550 
551 /*
552  *	Functions provided by ip_sockglue.c
553  */
554 
555 void ipv4_pktinfo_prepare(const struct sock *sk, struct sk_buff *skb);
556 void ip_cmsg_recv_offset(struct msghdr *msg, struct sk_buff *skb, int offset);
557 int ip_cmsg_send(struct net *net, struct msghdr *msg,
558 		 struct ipcm_cookie *ipc, bool allow_ipv6);
559 int ip_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
560 		  unsigned int optlen);
561 int ip_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
562 		  int __user *optlen);
563 int compat_ip_setsockopt(struct sock *sk, int level, int optname,
564 			 char __user *optval, unsigned int optlen);
565 int compat_ip_getsockopt(struct sock *sk, int level, int optname,
566 			 char __user *optval, int __user *optlen);
567 int ip_ra_control(struct sock *sk, unsigned char on,
568 		  void (*destructor)(struct sock *));
569 
570 int ip_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len);
571 void ip_icmp_error(struct sock *sk, struct sk_buff *skb, int err, __be16 port,
572 		   u32 info, u8 *payload);
573 void ip_local_error(struct sock *sk, int err, __be32 daddr, __be16 dport,
574 		    u32 info);
575 
576 static inline void ip_cmsg_recv(struct msghdr *msg, struct sk_buff *skb)
577 {
578 	ip_cmsg_recv_offset(msg, skb, 0);
579 }
580 
581 bool icmp_global_allow(void);
582 extern int sysctl_icmp_msgs_per_sec;
583 extern int sysctl_icmp_msgs_burst;
584 
585 #ifdef CONFIG_PROC_FS
586 int ip_misc_proc_init(void);
587 #endif
588 
589 #endif	/* _IP_H */
590