michael@0: /*-
michael@0:  * Copyright (c) 1982, 1986, 1990, 1993
michael@0:  *	The Regents of the University of California.  All rights reserved.
michael@0:  *
michael@0:  * Redistribution and use in source and binary forms, with or without
michael@0:  * modification, are permitted provided that the following conditions
michael@0:  * are met:
michael@0:  * 1. Redistributions of source code must retain the above copyright
michael@0:  *    notice, this list of conditions and the following disclaimer.
michael@0:  * 2. Redistributions in binary form must reproduce the above copyright
michael@0:  *    notice, this list of conditions and the following disclaimer in the
michael@0:  *    documentation and/or other materials provided with the distribution.
michael@0:  * 4. Neither the name of the University nor the names of its contributors
michael@0:  *    may be used to endorse or promote products derived from this software
michael@0:  *    without specific prior written permission.
michael@0:  *
michael@0:  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
michael@0:  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
michael@0:  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
michael@0:  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
michael@0:  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
michael@0:  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
michael@0:  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
michael@0:  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
michael@0:  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
michael@0:  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
michael@0:  * SUCH DAMAGE.
michael@0:  *
michael@0:  *	@(#)in_pcb.h	8.1 (Berkeley) 6/10/93
michael@0:  * $FreeBSD: src/sys/netinet/in_pcb.h,v 1.100.2.1 2007/12/07 05:46:08 kmacy Exp $
michael@0:  */
michael@0: 
michael@0: #ifndef _USER_INPCB_H_
michael@0: #define _USER_INPCB_H_
michael@0: 
michael@0: #include <user_route.h> /* was <net/route.h> */
michael@0: 
michael@0: #define	in6pcb		inpcb	/* for KAME src sync over BSD*'s */
michael@0: #define	in6p_sp		inp_sp	/* for KAME src sync over BSD*'s */
michael@0: struct inpcbpolicy;
michael@0: 
michael@0: /*
michael@0:  * Struct inpcb is the ommon structure pcb for the Internet Protocol
michael@0:  * implementation.
michael@0:  *
michael@0:  * Pointers to local and foreign host table entries, local and foreign socket
michael@0:  * numbers, and pointers up (to a socket structure) and down (to a
michael@0:  * protocol-specific control block) are stored here.
michael@0:  */
michael@0: LIST_HEAD(inpcbhead, inpcb);
michael@0: LIST_HEAD(inpcbporthead, inpcbport);
michael@0: 
michael@0: /*
michael@0:  * PCB with AF_INET6 null bind'ed laddr can receive AF_INET input packet.
michael@0:  * So, AF_INET6 null laddr is also used as AF_INET null laddr, by utilizing
michael@0:  * the following structure.
michael@0:  */
michael@0: struct in_addr_4in6 {
michael@0: 	u_int32_t	ia46_pad32[3];
michael@0: 	struct	in_addr	ia46_addr4;
michael@0: };
michael@0: 
michael@0: /*
michael@0:  * NOTE: ipv6 addrs should be 64-bit aligned, per RFC 2553.  in_conninfo has
michael@0:  * some extra padding to accomplish this.
michael@0:  */
michael@0: struct in_endpoints {
michael@0: 	u_int16_t	ie_fport;		/* foreign port */
michael@0: 	u_int16_t	ie_lport;		/* local port */
michael@0: 	/* protocol dependent part, local and foreign addr */
michael@0: 	union {
michael@0: 		/* foreign host table entry */
michael@0: 		struct	in_addr_4in6 ie46_foreign;
michael@0: 		struct	in6_addr ie6_foreign;
michael@0: 	} ie_dependfaddr;
michael@0: 	union {
michael@0: 		/* local host table entry */
michael@0: 		struct	in_addr_4in6 ie46_local;
michael@0: 		struct	in6_addr ie6_local;
michael@0: 	} ie_dependladdr;
michael@0: #define	ie_faddr	ie_dependfaddr.ie46_foreign.ia46_addr4
michael@0: #define	ie_laddr	ie_dependladdr.ie46_local.ia46_addr4
michael@0: #define	ie6_faddr	ie_dependfaddr.ie6_foreign
michael@0: #define	ie6_laddr	ie_dependladdr.ie6_local
michael@0: };
michael@0: 
michael@0: /*
michael@0:  * XXX The defines for inc_* are hacks and should be changed to direct
michael@0:  * references.
michael@0:  */
michael@0: struct in_conninfo {
michael@0: 	u_int8_t	inc_flags;
michael@0: 	u_int8_t	inc_len;
michael@0: 	u_int16_t	inc_pad;	/* XXX alignment for in_endpoints */
michael@0: 	/* protocol dependent part */
michael@0: 	struct	in_endpoints inc_ie;
michael@0: };
michael@0: #define inc_isipv6	inc_flags	/* temp compatability */
michael@0: #define	inc_fport	inc_ie.ie_fport
michael@0: #define	inc_lport	inc_ie.ie_lport
michael@0: #define	inc_faddr	inc_ie.ie_faddr
michael@0: #define	inc_laddr	inc_ie.ie_laddr
michael@0: #define	inc6_faddr	inc_ie.ie6_faddr
michael@0: #define	inc6_laddr	inc_ie.ie6_laddr
michael@0: 
michael@0: struct	icmp6_filter;
michael@0: 
michael@0: struct inpcb {
michael@0: 	LIST_ENTRY(inpcb) inp_hash;	/* hash list */
michael@0: 	LIST_ENTRY(inpcb) inp_list;	/* list for all PCBs of this proto */
michael@0: 	void	*inp_ppcb;		/* pointer to per-protocol pcb */
michael@0: 	struct	inpcbinfo *inp_pcbinfo;	/* PCB list info */
michael@0: 	struct	socket *inp_socket;	/* back pointer to socket */
michael@0: 
michael@0: 	u_int32_t	inp_flow;
michael@0: 	int	inp_flags;		/* generic IP/datagram flags */
michael@0: 
michael@0: 	u_char	inp_vflag;		/* IP version flag (v4/v6) */
michael@0: #define	INP_IPV4	0x1
michael@0: #define	INP_IPV6	0x2
michael@0: #define	INP_IPV6PROTO	0x4		/* opened under IPv6 protocol */
michael@0: #define	INP_TIMEWAIT	0x8		/* .. probably doesn't go here */
michael@0: #define	INP_ONESBCAST	0x10		/* send all-ones broadcast */
michael@0: #define	INP_DROPPED	0x20		/* protocol drop flag */
michael@0: #define	INP_SOCKREF	0x40		/* strong socket reference */
michael@0: #define INP_CONN        0x80
michael@0: 	u_char	inp_ip_ttl;		/* time to live proto */
michael@0: 	u_char	inp_ip_p;		/* protocol proto */
michael@0: 	u_char	inp_ip_minttl;		/* minimum TTL or drop */
michael@0: 	uint32_t inp_ispare1;		/* connection id / queue id */
michael@0: 	void	*inp_pspare[2];		/* rtentry / general use */
michael@0: 
michael@0: 	/* Local and foreign ports, local and foreign addr. */
michael@0: 	struct	in_conninfo inp_inc;
michael@0: 
michael@0: 					/* list for this PCB's local port */
michael@0: 	struct	label *inp_label;	/* MAC label */
michael@0: 	struct	inpcbpolicy *inp_sp;    /* for IPSEC */
michael@0: 
michael@0: 	/* Protocol-dependent part; options. */
michael@0: 	struct {
michael@0: 		u_char	inp4_ip_tos;		/* type of service proto */
michael@0: 		struct	mbuf *inp4_options;	/* IP options */
michael@0: 		struct	ip_moptions *inp4_moptions; /* IP multicast options */
michael@0: 	} inp_depend4;
michael@0: #define	inp_fport	inp_inc.inc_fport
michael@0: #define	inp_lport	inp_inc.inc_lport
michael@0: #define	inp_faddr	inp_inc.inc_faddr
michael@0: #define	inp_laddr	inp_inc.inc_laddr
michael@0: #define	inp_ip_tos	inp_depend4.inp4_ip_tos
michael@0: #define	inp_options	inp_depend4.inp4_options
michael@0: #define	inp_moptions	inp_depend4.inp4_moptions
michael@0: 	struct {
michael@0: 		/* IP options */
michael@0: 		struct	mbuf *inp6_options;
michael@0: 		/* IP6 options for outgoing packets */
michael@0: 		struct	ip6_pktopts *inp6_outputopts;
michael@0: 		/* IP multicast options */
michael@0: #if 0
michael@0: 		struct	ip6_moptions *inp6_moptions;
michael@0: #endif
michael@0: 		/* ICMPv6 code type filter */
michael@0: 		struct	icmp6_filter *inp6_icmp6filt;
michael@0: 		/* IPV6_CHECKSUM setsockopt */
michael@0: 		int	inp6_cksum;
michael@0: 		short	inp6_hops;
michael@0: 	} inp_depend6;
michael@0: 	LIST_ENTRY(inpcb) inp_portlist;
michael@0: 	struct	inpcbport *inp_phd;	/* head of this list */
michael@0: #define inp_zero_size offsetof(struct inpcb, inp_gencnt)
michael@0: 	struct mtx	inp_mtx;
michael@0: 
michael@0: #define	in6p_faddr	inp_inc.inc6_faddr
michael@0: #define	in6p_laddr	inp_inc.inc6_laddr
michael@0: #define	in6p_hops	inp_depend6.inp6_hops	/* default hop limit */
michael@0: #define	in6p_ip6_nxt	inp_ip_p
michael@0: #define	in6p_flowinfo	inp_flow
michael@0: #define	in6p_vflag	inp_vflag
michael@0: #define	in6p_options	inp_depend6.inp6_options
michael@0: #define	in6p_outputopts	inp_depend6.inp6_outputopts
michael@0: #if 0
michael@0: #define	in6p_moptions	inp_depend6.inp6_moptions
michael@0: #endif
michael@0: #define	in6p_icmp6filt	inp_depend6.inp6_icmp6filt
michael@0: #define	in6p_cksum	inp_depend6.inp6_cksum
michael@0: #define	in6p_flags	inp_flags  /* for KAME src sync over BSD*'s */
michael@0: #define	in6p_socket	inp_socket  /* for KAME src sync over BSD*'s */
michael@0: #define	in6p_lport	inp_lport  /* for KAME src sync over BSD*'s */
michael@0: #define	in6p_fport	inp_fport  /* for KAME src sync over BSD*'s */
michael@0: #define	in6p_ppcb	inp_ppcb  /* for KAME src sync over BSD*'s */
michael@0: };
michael@0: /*
michael@0:  * The range of the generation count, as used in this implementation, is 9e19.
michael@0:  * We would have to create 300 billion connections per second for this number
michael@0:  * to roll over in a year.  This seems sufficiently unlikely that we simply
michael@0:  * don't concern ourselves with that possibility.
michael@0:  */
michael@0: 
michael@0: struct inpcbport {
michael@0: 	LIST_ENTRY(inpcbport) phd_hash;
michael@0: 	struct inpcbhead phd_pcblist;
michael@0: 	u_short phd_port;
michael@0: };
michael@0: 
michael@0: /*
michael@0:  * Global data structure for each high-level protocol (UDP, TCP, ...) in both
michael@0:  * IPv4 and IPv6.  Holds inpcb lists and information for managing them.
michael@0:  */
michael@0: struct inpcbinfo {
michael@0: 	/*
michael@0: 	 * Global list of inpcbs on the protocol.
michael@0: 	 */
michael@0: 	struct inpcbhead	*ipi_listhead;
michael@0: 	u_int			 ipi_count;
michael@0: 
michael@0: 	/*
michael@0: 	 * Global hash of inpcbs, hashed by local and foreign addresses and
michael@0: 	 * port numbers.
michael@0: 	 */
michael@0: 	struct inpcbhead	*ipi_hashbase;
michael@0: 	u_long			 ipi_hashmask;
michael@0: 
michael@0: 	/*
michael@0: 	 * Global hash of inpcbs, hashed by only local port number.
michael@0: 	 */
michael@0: 	struct inpcbporthead	*ipi_porthashbase;
michael@0: 	u_long			 ipi_porthashmask;
michael@0: 
michael@0: 	/*
michael@0: 	 * Fields associated with port lookup and allocation.
michael@0: 	 */
michael@0: 	u_short			 ipi_lastport;
michael@0: 	u_short			 ipi_lastlow;
michael@0: 	u_short			 ipi_lasthi;
michael@0: 
michael@0: 	/*
michael@0: 	 * UMA zone from which inpcbs are allocated for this protocol.
michael@0: 	 */
michael@0: 	struct	uma_zone	*ipi_zone;
michael@0: 
michael@0: 	/*
michael@0: 	 * Generation count--incremented each time a connection is allocated
michael@0: 	 * or freed.
michael@0: 	 */
michael@0: 	struct mtx		 ipi_mtx;
michael@0: 
michael@0: 	/*
michael@0: 	 * vimage 1
michael@0: 	 * general use 1
michael@0: 	 */
michael@0: 	void 			*ipi_pspare[2];
michael@0: };
michael@0: 
michael@0: #define INP_LOCK_INIT(inp, d, t) \
michael@0: 	mtx_init(&(inp)->inp_mtx, (d), (t), MTX_DEF | MTX_RECURSE | MTX_DUPOK)
michael@0: #define INP_LOCK_DESTROY(inp)	mtx_destroy(&(inp)->inp_mtx)
michael@0: #define INP_LOCK(inp)		mtx_lock(&(inp)->inp_mtx)
michael@0: #define INP_UNLOCK(inp)		mtx_unlock(&(inp)->inp_mtx)
michael@0: #define INP_LOCK_ASSERT(inp)	mtx_assert(&(inp)->inp_mtx, MA_OWNED)
michael@0: #define	INP_UNLOCK_ASSERT(inp)	mtx_assert(&(inp)->inp_mtx, MA_NOTOWNED)
michael@0: 
michael@0: #define INP_INFO_LOCK_INIT(ipi, d) \
michael@0: 	mtx_init(&(ipi)->ipi_mtx, (d), NULL, MTX_DEF | MTX_RECURSE)
michael@0: #define INP_INFO_LOCK_DESTROY(ipi)  mtx_destroy(&(ipi)->ipi_mtx)
michael@0: #define INP_INFO_RLOCK(ipi)	mtx_lock(&(ipi)->ipi_mtx)
michael@0: #define INP_INFO_WLOCK(ipi)	mtx_lock(&(ipi)->ipi_mtx)
michael@0: #define INP_INFO_RUNLOCK(ipi)	mtx_unlock(&(ipi)->ipi_mtx)
michael@0: #define INP_INFO_WUNLOCK(ipi)	mtx_unlock(&(ipi)->ipi_mtx)
michael@0: #define INP_INFO_RLOCK_ASSERT(ipi)	mtx_assert(&(ipi)->ipi_mtx, MA_OWNED)
michael@0: #define INP_INFO_WLOCK_ASSERT(ipi)	mtx_assert(&(ipi)->ipi_mtx, MA_OWNED)
michael@0: #define INP_INFO_UNLOCK_ASSERT(ipi)	mtx_assert(&(ipi)->ipi_mtx, MA_NOTOWNED)
michael@0: 
michael@0: #define INP_PCBHASH(faddr, lport, fport, mask) \
michael@0: 	(((faddr) ^ ((faddr) >> 16) ^ ntohs((lport) ^ (fport))) & (mask))
michael@0: #define INP_PCBPORTHASH(lport, mask) \
michael@0: 	(ntohs((lport)) & (mask))
michael@0: 
michael@0: /* flags in inp_flags: */
michael@0: #define	INP_RECVOPTS		0x01	/* receive incoming IP options */
michael@0: #define	INP_RECVRETOPTS		0x02	/* receive IP options for reply */
michael@0: #define	INP_RECVDSTADDR		0x04	/* receive IP dst address */
michael@0: #define	INP_HDRINCL		0x08	/* user supplies entire IP header */
michael@0: #define	INP_HIGHPORT		0x10	/* user wants "high" port binding */
michael@0: #define	INP_LOWPORT		0x20	/* user wants "low" port binding */
michael@0: #define	INP_ANONPORT		0x40	/* port chosen for user */
michael@0: #define	INP_RECVIF		0x80	/* receive incoming interface */
michael@0: #define	INP_MTUDISC		0x100	/* user can do MTU discovery */
michael@0: #define	INP_FAITH		0x200	/* accept FAITH'ed connections */
michael@0: #define	INP_RECVTTL		0x400	/* receive incoming IP TTL */
michael@0: #define	INP_DONTFRAG		0x800	/* don't fragment packet */
michael@0: 
michael@0: #define IN6P_IPV6_V6ONLY	0x008000 /* restrict AF_INET6 socket for v6 */
michael@0: 
michael@0: #define	IN6P_PKTINFO		0x010000 /* receive IP6 dst and I/F */
michael@0: #define	IN6P_HOPLIMIT		0x020000 /* receive hoplimit */
michael@0: #define	IN6P_HOPOPTS		0x040000 /* receive hop-by-hop options */
michael@0: #define	IN6P_DSTOPTS		0x080000 /* receive dst options after rthdr */
michael@0: #define	IN6P_RTHDR		0x100000 /* receive routing header */
michael@0: #define	IN6P_RTHDRDSTOPTS	0x200000 /* receive dstoptions before rthdr */
michael@0: #define	IN6P_TCLASS		0x400000 /* receive traffic class value */
michael@0: #define	IN6P_AUTOFLOWLABEL	0x800000 /* attach flowlabel automatically */
michael@0: #define	IN6P_RFC2292		0x40000000 /* used RFC2292 API on the socket */
michael@0: #define	IN6P_MTU		0x80000000 /* receive path MTU */
michael@0: 
michael@0: #define	INP_CONTROLOPTS		(INP_RECVOPTS|INP_RECVRETOPTS|INP_RECVDSTADDR|\
michael@0: 				 INP_RECVIF|INP_RECVTTL|\
michael@0: 				 IN6P_PKTINFO|IN6P_HOPLIMIT|IN6P_HOPOPTS|\
michael@0: 				 IN6P_DSTOPTS|IN6P_RTHDR|IN6P_RTHDRDSTOPTS|\
michael@0: 				 IN6P_TCLASS|IN6P_AUTOFLOWLABEL|IN6P_RFC2292|\
michael@0: 				 IN6P_MTU)
michael@0: #define	INP_UNMAPPABLEOPTS	(IN6P_HOPOPTS|IN6P_DSTOPTS|IN6P_RTHDR|\
michael@0: 				 IN6P_TCLASS|IN6P_AUTOFLOWLABEL)
michael@0: 
michael@0:  /* for KAME src sync over BSD*'s */
michael@0: #define	IN6P_HIGHPORT		INP_HIGHPORT
michael@0: #define	IN6P_LOWPORT		INP_LOWPORT
michael@0: #define	IN6P_ANONPORT		INP_ANONPORT
michael@0: #define	IN6P_RECVIF		INP_RECVIF
michael@0: #define	IN6P_MTUDISC		INP_MTUDISC
michael@0: #define	IN6P_FAITH		INP_FAITH
michael@0: #define	IN6P_CONTROLOPTS INP_CONTROLOPTS
michael@0: 	/*
michael@0: 	 * socket AF version is {newer than,or include}
michael@0: 	 * actual datagram AF version
michael@0: 	 */
michael@0: 
michael@0: #define	INPLOOKUP_WILDCARD	1
michael@0: #define	sotoinpcb(so)	((struct inpcb *)(so)->so_pcb)
michael@0: #define	sotoin6pcb(so)	sotoinpcb(so) /* for KAME src sync over BSD*'s */
michael@0: 
michael@0: #define	INP_SOCKAF(so) so->so_proto->pr_domain->dom_family
michael@0: 
michael@0: #define	INP_CHECK_SOCKAF(so, af)	(INP_SOCKAF(so) == af)
michael@0: 
michael@0: /* #ifdef _KERNEL */
michael@0: extern int	ipport_reservedhigh;
michael@0: extern int	ipport_reservedlow;
michael@0: extern int	ipport_lowfirstauto;
michael@0: extern int	ipport_lowlastauto;
michael@0: extern int	ipport_firstauto;
michael@0: extern int	ipport_lastauto;
michael@0: extern int	ipport_hifirstauto;
michael@0: extern int	ipport_hilastauto;
michael@0: extern struct callout ipport_tick_callout;
michael@0: 
michael@0: void	in_pcbpurgeif0(struct inpcbinfo *, struct ifnet *);
michael@0: int	in_pcballoc(struct socket *, struct inpcbinfo *);
michael@0: int	in_pcbbind(struct inpcb *, struct sockaddr *, struct ucred *);
michael@0: int	in_pcbconnect(struct inpcb *, struct sockaddr *, struct ucred *);
michael@0: void	in_pcbdetach(struct inpcb *);
michael@0: void	in_pcbdisconnect(struct inpcb *);
michael@0: void	in_pcbdrop(struct inpcb *);
michael@0: void	in_pcbfree(struct inpcb *);
michael@0: int	in_pcbinshash(struct inpcb *);
michael@0: struct inpcb *
michael@0: 	in_pcblookup_local(struct inpcbinfo *,
michael@0: 	    struct in_addr, u_int, int);
michael@0: struct inpcb *
michael@0: 	in_pcblookup_hash(struct inpcbinfo *, struct in_addr, u_int,
michael@0: 	    struct in_addr, u_int, int, struct ifnet *);
michael@0: void	in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr,
michael@0: 	    int, struct inpcb *(*)(struct inpcb *, int));
michael@0: void	in_pcbrehash(struct inpcb *);
michael@0: void	in_pcbsetsolabel(struct socket *so);
michael@0: int	in_getpeeraddr(struct socket *so, struct sockaddr **nam);
michael@0: int	in_getsockaddr(struct socket *so, struct sockaddr **nam);
michael@0: void	in_pcbsosetlabel(struct socket *so);
michael@0: void	in_pcbremlists(struct inpcb *inp);
michael@0: void	ipport_tick(void *xtp);
michael@0: 
michael@0: /*
michael@0:  * Debugging routines compiled in when DDB is present.
michael@0:  */
michael@0: void	db_print_inpcb(struct inpcb *inp, const char *name, int indent);
michael@0: 
michael@0: /* #endif  _KERNEL */
michael@0: 
michael@0: #endif /* !_NETINET_IN_PCB_H_ */