1
0
mirror of https://git.FreeBSD.org/src.git synced 2024-12-22 11:17:19 +00:00
freebsd/sys/netinet/ip_fil.h

526 lines
16 KiB
C

/*
* Copyright (C) 1993-1997 by Darren Reed.
*
* Redistribution and use in source and binary forms are permitted
* provided that this notice is preserved and due credit is given
* to the original author and the contributors.
*
* @(#)ip_fil.h 1.35 6/5/96
* $Id: ip_fil.h,v 1.5 1998/06/08 09:47:41 bde Exp $
*/
#ifndef __IP_FIL_H__
#define __IP_FIL_H__
/*
* Pathnames for various IP Filter control devices. Used by LKM
* and userland, so defined here.
*/
#define IPNAT_NAME "/dev/ipnat"
#define IPSTATE_NAME "/dev/ipstate"
#define IPAUTH_NAME "/dev/ipauth"
#ifndef SOLARIS
#define SOLARIS (defined(sun) && (defined(__svr4__) || defined(__SVR4)))
#endif
#if defined(KERNEL) && !defined(_KERNEL)
#define _KERNEL
#endif
#ifndef __P
# ifdef __STDC__
# define __P(x) x
# else
# define __P(x) ()
# endif
#endif
#if defined(__STDC__) || defined(__GNUC__)
#define SIOCADAFR _IOW('r', 60, struct frentry)
#define SIOCRMAFR _IOW('r', 61, struct frentry)
#define SIOCSETFF _IOW('r', 62, u_int)
#define SIOCGETFF _IOR('r', 63, u_int)
#define SIOCGETFS _IOR('r', 64, struct friostat)
#define SIOCIPFFL _IOWR('r', 65, int)
#define SIOCIPFFB _IOR('r', 66, int)
#define SIOCADIFR _IOW('r', 67, struct frentry)
#define SIOCRMIFR _IOW('r', 68, struct frentry)
#define SIOCSWAPA _IOR('r', 69, u_int)
#define SIOCINAFR _IOW('r', 70, struct frentry)
#define SIOCINIFR _IOW('r', 71, struct frentry)
#define SIOCFRENB _IOW('r', 72, u_int)
#define SIOCFRSYN _IOW('r', 73, u_int)
#define SIOCFRZST _IOWR('r', 74, struct friostat)
#define SIOCZRLST _IOWR('r', 75, struct frentry)
#define SIOCAUTHW _IOWR('r', 76, struct fr_info)
#define SIOCAUTHR _IOWR('r', 77, struct fr_info)
#define SIOCATHST _IOWR('r', 78, struct fr_authstat)
#else
#define SIOCADAFR _IOW(r, 60, struct frentry)
#define SIOCRMAFR _IOW(r, 61, struct frentry)
#define SIOCSETFF _IOW(r, 62, u_int)
#define SIOCGETFF _IOR(r, 63, u_int)
#define SIOCGETFS _IOR(r, 64, struct friostat)
#define SIOCIPFFL _IOWR(r, 65, int)
#define SIOCIPFFB _IOR(r, 66, int)
#define SIOCADIFR _IOW(r, 67, struct frentry)
#define SIOCRMIFR _IOW(r, 68, struct frentry)
#define SIOCSWAPA _IOR(r, 69, u_int)
#define SIOCINAFR _IOW(r, 70, struct frentry)
#define SIOCINIFR _IOW(r, 71, struct frentry)
#define SIOCFRENB _IOW(r, 72, u_int)
#define SIOCFRSYN _IOW(r, 73, u_int)
#define SIOCFRZST _IOWR(r, 74, struct friostat)
#define SIOCZRLST _IOWR(r, 75, struct frentry)
#define SIOCAUTHW _IOWR(r, 76, struct fr_info)
#define SIOCAUTHR _IOWR(r, 77, struct fr_info)
#define SIOCATHST _IOWR(r, 78, struct fr_authstat)
#endif
#define SIOCADDFR SIOCADAFR
#define SIOCDELFR SIOCRMAFR
#define SIOCINSFR SIOCINAFR
typedef struct fr_ip {
u_int fi_v:4; /* IP version */
u_int fi_fl:4; /* packet flags */
u_char fi_tos;
u_char fi_ttl;
u_char fi_p;
struct in_addr fi_src;
struct in_addr fi_dst;
u_32_t fi_optmsk; /* bitmask composed from IP options */
u_short fi_secmsk; /* bitmask composed from IP security options */
u_short fi_auth;
} fr_ip_t;
#define FI_OPTIONS (FF_OPTIONS >> 24)
#define FI_TCPUDP (FF_TCPUDP >> 24) /* TCP/UCP implied comparison*/
#define FI_FRAG (FF_FRAG >> 24)
#define FI_SHORT (FF_SHORT >> 24)
typedef struct fr_info {
struct fr_ip fin_fi;
u_short fin_data[2];
u_short fin_out;
u_short fin_hlen;
u_char fin_tcpf;
u_char fin_icode; /* From here on is packet specific */
u_short fin_rule;
u_short fin_group;
u_short fin_dlen;
u_short fin_id;
void *fin_ifp;
struct frentry *fin_fr;
char *fin_dp; /* start of data past IP header */
void *fin_mp;
} fr_info_t;
/*
* Size for compares on fr_info structures
*/
#define FI_CSIZE (sizeof(struct fr_ip) + sizeof(u_short) * 4 + \
sizeof(u_char))
/*
* Size for copying cache fr_info structure
*/
#define FI_COPYSIZE (sizeof(fr_info_t) - sizeof(void *) * 2)
typedef struct frdest {
void *fd_ifp;
struct in_addr fd_ip;
char fd_ifname[IFNAMSIZ];
} frdest_t;
typedef struct frentry {
struct frentry *fr_next;
u_short fr_group; /* group to which this rule belongs */
u_short fr_grhead; /* group # which this rule starts */
struct frentry *fr_grp;
int fr_ref; /* reference count - for grouping */
void *fr_ifa;
/*
* These are only incremented when a packet matches this rule and
* it is the last match
*/
U_QUAD_T fr_hits;
U_QUAD_T fr_bytes;
/*
* Fields after this may not change whilst in the kernel.
*/
struct fr_ip fr_ip;
struct fr_ip fr_mip; /* mask structure */
u_char fr_tcpfm; /* tcp flags mask */
u_char fr_tcpf; /* tcp flags */
u_short fr_icmpm; /* data for ICMP packets (mask) */
u_short fr_icmp;
u_char fr_scmp; /* data for port comparisons */
u_char fr_dcmp;
u_short fr_dport;
u_short fr_sport;
u_short fr_stop; /* top port for <> and >< */
u_short fr_dtop; /* top port for <> and >< */
u_32_t fr_flags; /* per-rule flags && options (see below) */
int fr_skip; /* # of rules to skip */
int (*fr_func) __P((int, ip_t *, fr_info_t *)); /* call this function */
char fr_icode; /* return ICMP code */
char fr_ifname[IFNAMSIZ];
struct frdest fr_tif; /* "to" interface */
struct frdest fr_dif; /* duplicate packet interfaces */
} frentry_t;
#define fr_proto fr_ip.fi_p
#define fr_ttl fr_ip.fi_ttl
#define fr_tos fr_ip.fi_tos
#define fr_dst fr_ip.fi_dst
#define fr_src fr_ip.fi_src
#define fr_dmsk fr_mip.fi_dst
#define fr_smsk fr_mip.fi_src
#ifndef offsetof
#define offsetof(t,m) (int)((&((t *)0L)->m))
#endif
#define FR_CMPSIZ (sizeof(struct frentry) - offsetof(frentry_t, fr_ip))
/*
* fr_flags
*/
#define FR_BLOCK 0x00001 /* do not allow packet to pass */
#define FR_PASS 0x00002 /* allow packet to pass */
#define FR_OUTQUE 0x00004 /* outgoing packets */
#define FR_INQUE 0x00008 /* ingoing packets */
#define FR_LOG 0x00010 /* Log */
#define FR_LOGB 0x00011 /* Log-fail */
#define FR_LOGP 0x00012 /* Log-pass */
#define FR_LOGBODY 0x00020 /* Log the body */
#define FR_LOGFIRST 0x00040 /* Log the first byte if state held */
#define FR_RETRST 0x00080 /* Return TCP RST packet - reset connection */
#define FR_RETICMP 0x00100 /* Return ICMP unreachable packet */
#define FR_NOMATCH 0x00200 /* no match occured */
#define FR_ACCOUNT 0x00400 /* count packet bytes */
#define FR_KEEPFRAG 0x00800 /* keep fragment information */
#define FR_KEEPSTATE 0x01000 /* keep `connection' state information */
#define FR_INACTIVE 0x02000
#define FR_QUICK 0x04000 /* match & stop processing list */
#define FR_FASTROUTE 0x08000 /* bypass normal routing */
#define FR_CALLNOW 0x10000 /* call another function (fr_func) if matches */
#define FR_DUP 0x20000 /* duplicate packet */
#define FR_LOGORBLOCK 0x40000 /* block the packet if it can't be logged */
#define FR_NOTSRCIP 0x80000 /* not the src IP# */
#define FR_NOTDSTIP 0x100000 /* not the dst IP# */
#define FR_AUTH 0x200000 /* use authentication */
#define FR_PREAUTH 0x400000 /* require preauthentication */
#define FR_LOGMASK (FR_LOG|FR_LOGP|FR_LOGB)
/*
* These correspond to #define's for FI_* and are stored in fr_flags
*/
#define FF_OPTIONS 0x01000000
#define FF_TCPUDP 0x02000000
#define FF_FRAG 0x04000000
#define FF_SHORT 0x08000000
/*
* recognized flags for SIOCGETFF and SIOCSETFF, and get put in fr_flags
*/
#define FF_LOGPASS 0x10000000
#define FF_LOGBLOCK 0x20000000
#define FF_LOGNOMATCH 0x40000000
#define FF_LOGGING (FF_LOGPASS|FF_LOGBLOCK|FF_LOGNOMATCH)
#define FF_BLOCKNONIP 0x80000000 /* Solaris2 Only */
#define FR_NONE 0
#define FR_EQUAL 1
#define FR_NEQUAL 2
#define FR_LESST 3
#define FR_GREATERT 4
#define FR_LESSTE 5
#define FR_GREATERTE 6
#define FR_OUTRANGE 7
#define FR_INRANGE 8
typedef struct filterstats {
u_long fr_pass; /* packets allowed */
u_long fr_block; /* packets denied */
u_long fr_nom; /* packets which don't match any rule */
u_long fr_ppkl; /* packets allowed and logged */
u_long fr_bpkl; /* packets denied and logged */
u_long fr_npkl; /* packets unmatched and logged */
u_long fr_pkl; /* packets logged */
u_long fr_skip; /* packets to be logged but buffer full */
u_long fr_ret; /* packets for which a return is sent */
u_long fr_acct; /* packets for which counting was performed */
u_long fr_bnfr; /* bad attempts to allocate fragment state */
u_long fr_nfr; /* new fragment state kept */
u_long fr_cfr; /* add new fragment state but complete pkt */
u_long fr_bads; /* bad attempts to allocate packet state */
u_long fr_ads; /* new packet state kept */
u_long fr_chit; /* cached hit */
u_long fr_tcpbad; /* TCP checksum check failures */
u_long fr_pull[2]; /* good and bad pullup attempts */
#if SOLARIS
u_long fr_bad; /* bad IP packets to the filter */
u_long fr_notip; /* packets passed through no on ip queue */
u_long fr_drop; /* packets dropped - no info for them! */
#endif
} filterstats_t;
/*
* For SIOCGETFS
*/
typedef struct friostat {
struct filterstats f_st[2];
struct frentry *f_fin[2];
struct frentry *f_fout[2];
struct frentry *f_acctin[2];
struct frentry *f_acctout[2];
struct frentry *f_auth;
u_long f_froute[2];
int f_active;
} friostat_t;
typedef struct optlist {
u_short ol_val;
int ol_bit;
} optlist_t;
/*
* Group list structure.
*/
typedef struct frgroup {
u_short fg_num;
struct frgroup *fg_next;
struct frentry *fg_head;
struct frentry **fg_start;
} frgroup_t;
/*
* Log structure. Each packet header logged is prepended by one of these.
* Following this in the log records read from the device will be an ipflog
* structure which is then followed by any packet data.
*/
typedef struct iplog {
u_long ipl_magic;
u_long ipl_sec;
u_long ipl_usec;
u_int ipl_len;
u_int ipl_count;
size_t ipl_dsize;
struct iplog *ipl_next;
} iplog_t;
#define IPL_MAGIC 0x49504c4d /* 'IPLM' */
typedef struct ipflog {
#if (defined(NetBSD) && (NetBSD <= 1991011) && (NetBSD >= 199603)) || \
(defined(OpenBSD) && (OpenBSD >= 199603))
u_char fl_ifname[IFNAMSIZ];
#else
u_int fl_unit;
u_char fl_ifname[4];
#endif
u_char fl_plen; /* extra data after hlen */
u_char fl_hlen; /* length of IP headers saved */
u_short fl_rule; /* assume never more than 64k rules, total */
u_short fl_group;
u_32_t fl_flags;
} ipflog_t;
#ifndef ICMP_UNREACH_FILTER
#define ICMP_UNREACH_FILTER 13
#endif
#ifndef IPF_LOGGING
#define IPF_LOGGING 0
#endif
#ifndef IPF_DEFAULT_PASS
#define IPF_DEFAULT_PASS FR_PASS
#endif
#define IPMINLEN(i, h) ((i)->ip_len >= ((i)->ip_hl * 4 + sizeof(struct h)))
#define IPLLOGSIZE 8192
/*
* Device filenames for reading log information. Use ipf on Solaris2 because
* ipl is already a name used by something else.
*/
#ifndef IPL_NAME
# if SOLARIS
# define IPL_NAME "/dev/ipf"
# else
# define IPL_NAME "/dev/ipl"
# endif
#endif
#define IPL_NAT IPNAT_NAME
#define IPL_STATE IPSTATE_NAME
#define IPL_AUTH IPAUTH_NAME
#define IPL_LOGIPF 0 /* Minor device #'s for accessing logs */
#define IPL_LOGNAT 1
#define IPL_LOGSTATE 2
#define IPL_LOGAUTH 3
#define IPL_LOGMAX 3
#if !defined(CDEV_MAJOR) && defined (__FreeBSD_version) && \
(__FreeBSD_version >= 220000)
# define CDEV_MAJOR 79
#endif
#ifndef _KERNEL
struct ifnet;
extern int fr_check __P((ip_t *, int, void *, int, mb_t **));
extern int (*fr_checkp) __P((ip_t *, int, void *, int, mb_t **));
extern int send_reset __P((ip_t *, struct ifnet *));
extern int icmp_error __P((ip_t *, struct ifnet *));
extern int ipf_log __P((void));
extern void ipfr_fastroute __P((ip_t *, fr_info_t *, frdest_t *));
extern struct ifnet *get_unit __P((char *));
# define FR_SCANLIST(p, ip, fi, m) fr_scanlist(p, ip, fi, m)
# if defined(__NetBSD__) || defined(__OpenBSD__) || \
(_BSDI_VERSION >= 199701) || (__FreeBSD_version >= 300003)
extern int iplioctl __P((dev_t, u_long, caddr_t, int));
# else
extern int iplioctl __P((dev_t, int, caddr_t, int));
# endif
extern int iplopen __P((dev_t, int));
extern int iplclose __P((dev_t, int));
#else /* #ifndef _KERNEL */
# if defined(__NetBSD__) && defined(PFIL_HOOKS)
extern int ipfilterattach __P((int));
# endif
extern int iplattach __P((void));
extern int ipl_enable __P((void));
extern int ipl_disable __P((void));
extern void ipflog_init __P((void));
extern int ipflog_clear __P((int));
extern int ipflog_read __P((int, struct uio *));
extern int ipflog __P((u_int, ip_t *, fr_info_t *, mb_t *));
extern int ipllog __P((int, u_long, void **, size_t *, int *, int));
# if SOLARIS
extern int fr_check __P((ip_t *, int, void *, int, qif_t *, mb_t **));
extern int (*fr_checkp) __P((ip_t *, int, void *,
int, qif_t *, mb_t **));
extern int icmp_error __P((ip_t *, int, int, qif_t *,
struct in_addr));
extern int iplioctl __P((dev_t, int, int, int, cred_t *, int *));
extern int iplopen __P((dev_t *, int, int, cred_t *));
extern int iplclose __P((dev_t, int, int, cred_t *));
extern int ipfsync __P((void));
extern int send_reset __P((ip_t *, qif_t *));
extern int ipfr_fastroute __P((qif_t *, ip_t *, mblk_t *, mblk_t **,
fr_info_t *, frdest_t *));
extern void copyin_mblk __P((mblk_t *, int, int, char *));
extern void copyout_mblk __P((mblk_t *, int, int, char *));
extern int fr_qin __P((queue_t *, mblk_t *));
extern int fr_qout __P((queue_t *, mblk_t *));
# ifdef IPFILTER_LOG
extern int iplread __P((dev_t, struct uio *, cred_t *));
# endif
# else /* SOLARIS */
extern int fr_check __P((ip_t *, int, void *, int, mb_t **));
extern int (*fr_checkp) __P((ip_t *, int, void *, int, mb_t **));
# ifdef linux
extern int send_reset __P((tcpiphdr_t *, struct ifnet *));
# else
extern int send_reset __P((tcpiphdr_t *));
# endif
extern void ipfr_fastroute __P((mb_t *, fr_info_t *, frdest_t *));
extern size_t mbufchainlen __P((mb_t *));
# ifdef __sgi
# include <sys/cred.h>
extern int iplioctl __P((dev_t, int, caddr_t, int, cred_t *, int *));
extern int iplopen __P((dev_t *, int, int, cred_t *));
extern int iplclose __P((dev_t, int, int, cred_t *));
extern int iplread __P((dev_t, struct uio *, cred_t *));
extern int ipfsync __P((void));
extern int ipfilter_sgi_attach __P((void));
extern void ipfilter_sgi_detach __P((void));
extern void ipfilter_sgi_intfsync __P((void));
# else
# ifdef IPFILTER_LKM
extern int iplidentify __P((char *));
# endif
# if (_BSDI_VERSION >= 199510) || (__FreeBSD_version >= 220000) || \
(NetBSD >= 199511)
# if defined(__NetBSD__) || (_BSDI_VERSION >= 199701) || \
(__FreeBSD_version >= 300003)
extern int iplioctl __P((dev_t, u_long, caddr_t, int, struct proc *));
# else
extern int iplioctl __P((dev_t, int, caddr_t, int, struct proc *));
# endif
extern int iplopen __P((dev_t, int, int, struct proc *));
extern int iplclose __P((dev_t, int, int, struct proc *));
# else
# if defined(__OpenBSD__)
extern int iplioctl __P((dev_t, u_long, caddr_t, int));
# else /* __OpenBSD__ */
# ifndef linux
extern int iplioctl __P((dev_t, int, caddr_t, int));
# else
extern int iplioctl(struct inode *, struct file *, u_int, u_long);
# endif
# endif /* __OpenBSD__ */
# ifndef linux
extern int iplopen __P((dev_t, int));
extern int iplclose __P((dev_t, int));
# else
extern int iplopen __P((struct inode *, struct file *));
extern void iplclose __P((struct inode *, struct file *));
# endif /* !linux */
# endif /* (_BSDI_VERSION >= 199510) */
# if BSD >= 199306
extern int iplread __P((dev_t, struct uio *, int));
# else
# ifndef linux
extern int iplread __P((dev_t, struct uio *));
# else
extern int iplread(struct inode *, struct file *, char *, int);
# endif /* !linux */
# endif /* BSD >= 199306 */
# endif /* __ sgi */
# endif /* SOLARIS */
#endif /* #ifndef _KERNEL */
/*
* Post NetBSD 1.2 has the PFIL interface for packet filters. This turns
* on those hooks. We don't need any special mods in non-IP Filter code
* with this!
*/
#if (defined(NetBSD) && (NetBSD > 199609) && (NetBSD <= 1991011)) || \
(defined(NetBSD1_2) && NetBSD1_2 > 1)
# define NETBSD_PF
#endif
extern int ipldetach __P((void));
extern u_short fr_tcpsum __P((mb_t *, ip_t *, tcphdr_t *, int));
#define FR_SCANLIST(p, ip, fi, m) fr_scanlist(p, ip, fi, m)
extern int fr_scanlist __P((int, ip_t *, fr_info_t *, void *));
extern u_short ipf_cksum __P((u_short *, int));
extern int fr_copytolog __P((int, char *, int));
extern void frflush __P((int, int *));
extern frgroup_t *fr_addgroup __P((u_short, frentry_t *, int, int));
extern frgroup_t *fr_findgroup __P((u_short, u_32_t, int, int, frgroup_t ***));
extern void fr_delgroup __P((u_short, u_32_t, int, int));
extern int ipl_unreach;
extern int ipl_inited;
extern u_long ipl_frouteok[2];
extern int fr_pass;
extern int fr_flags;
extern int fr_active;
extern fr_info_t frcache[2];
#ifdef IPFILTER_LOG
extern iplog_t **iplh[IPL_LOGMAX+1], *iplt[IPL_LOGMAX+1];
extern int iplused[IPL_LOGMAX + 1];
#endif
extern struct frentry *ipfilter[2][2], *ipacct[2][2];
extern struct frgroup *ipfgroups[3][2];
extern struct filterstats frstats[];
#endif /* __IP_FIL_H__ */