NetBSD/sys/netinet/ip_fil.c

1330 lines
28 KiB
C
Raw Normal View History

/* $NetBSD: ip_fil.c,v 1.33 1999/01/19 23:39:57 mycroft Exp $ */
/*
1998-11-22 18:17:18 +03:00
* Copyright (C) 1993-1998 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.
*/
#if !defined(lint)
static const char sccsid[] = "@(#)ip_fil.c 2.41 6/5/96 (C) 1993-1995 Darren Reed";
1998-11-22 18:17:18 +03:00
static const char rcsid[] = "@(#)Id: ip_fil.c,v 2.0.2.44.2.10 1998/11/22 01:50:22 darrenr Exp ";
#endif
#ifndef SOLARIS
#define SOLARIS (defined(sun) && (defined(__svr4__) || defined(__SVR4)))
#endif
1997-11-14 15:40:06 +03:00
#if defined(KERNEL) && !defined(_KERNEL)
# define _KERNEL
#endif
1997-05-25 16:40:11 +04:00
#ifdef __FreeBSD__
# if defined(_KERNEL) && !defined(IPFILTER_LKM)
# include <sys/osreldate.h>
# else
# include <osreldate.h>
# endif
#endif
#ifndef _KERNEL
1997-07-05 09:38:14 +04:00
# include <stdio.h>
# include <string.h>
# include <stdlib.h>
# include <ctype.h>
#endif
#include <sys/errno.h>
#include <sys/types.h>
#include <sys/param.h>
#include <sys/file.h>
1997-05-25 16:40:11 +04:00
#if __FreeBSD_version >= 220000 && defined(_KERNEL)
# include <sys/fcntl.h>
# include <sys/filio.h>
#else
# include <sys/ioctl.h>
#endif
#include <sys/time.h>
#ifdef _KERNEL
1997-07-05 09:38:14 +04:00
# include <sys/systm.h>
#endif
#include <sys/uio.h>
#if !SOLARIS
# if (NetBSD > 199609) || (OpenBSD > 199603)
# include <sys/dirent.h>
# else
# include <sys/dir.h>
# endif
1997-07-05 09:38:14 +04:00
# include <sys/mbuf.h>
#else
1997-07-05 09:38:14 +04:00
# include <sys/filio.h>
#endif
#include <sys/protosw.h>
#include <sys/socket.h>
#include <net/if.h>
#ifdef sun
1997-07-05 09:38:14 +04:00
# include <net/af.h>
1997-05-25 16:40:11 +04:00
#endif
#if __FreeBSD_version >= 300000
# include <net/if_var.h>
#endif
#ifdef __sgi
#include <sys/debug.h>
# ifdef IFF_DRVRLOCK /* IRIX6 */
#include <sys/hashing.h>
# endif
#endif
#include <net/route.h>
#include <netinet/in.h>
#if !(defined(__sgi) && !defined(IFF_DRVRLOCK)) /* IRIX < 6 */
#include <netinet/in_var.h>
#endif
#include <netinet/in_systm.h>
#include <netinet/ip.h>
#include <netinet/ip_var.h>
#include <netinet/tcp.h>
#include <netinet/udp.h>
#include <netinet/tcpip.h>
#include <netinet/ip_icmp.h>
1997-05-25 16:40:11 +04:00
#ifndef _KERNEL
# include <syslog.h>
#endif
1997-05-25 16:40:11 +04:00
#include "netinet/ip_compat.h"
#include "netinet/ip_fil.h"
#include "netinet/ip_proxy.h"
#include "netinet/ip_nat.h"
#include "netinet/ip_frag.h"
#include "netinet/ip_state.h"
1997-07-05 09:38:14 +04:00
#include "netinet/ip_auth.h"
#ifndef MIN
1997-05-25 16:40:11 +04:00
#define MIN(a,b) (((a)<(b))?(a):(b))
#endif
#if !SOLARIS && defined(_KERNEL)
extern int ip_optcopy __P((struct ip *, struct ip *));
#endif
1997-05-25 16:40:11 +04:00
extern struct protosw inetsw[];
1997-07-05 09:38:14 +04:00
#ifndef _KERNEL
# include "ipt.h"
static struct ifnet **ifneta = NULL;
static int nifs = 0;
#else
# if (BSD < 199306) && !defined(__sgi)
static int (*fr_saveslowtimo) __P((void));
1997-07-05 09:38:14 +04:00
# else
static void (*fr_saveslowtimo) __P((void));
1997-07-05 09:38:14 +04:00
# endif
# if (BSD < 199306) || defined(__sgi)
extern int tcp_ttl;
# endif
#endif
int ipl_inited = 0;
int ipl_unreach = ICMP_UNREACH_FILTER;
u_long ipl_frouteok[2] = {0, 0};
static void fixskip __P((frentry_t **, frentry_t *, int));
static void frzerostats __P((caddr_t));
static void frsync __P((void));
#if defined(__NetBSD__) || defined(__OpenBSD__)
1997-07-05 09:38:14 +04:00
static int frrequest __P((int, u_long, caddr_t, int));
#else
1997-07-05 09:38:14 +04:00
static int frrequest __P((int, int, caddr_t, int));
#endif
#ifdef _KERNEL
1997-11-14 15:40:06 +03:00
static int (*fr_savep) __P((ip_t *, int, void *, int, struct mbuf **));
1998-11-22 18:17:18 +03:00
# ifdef __sgi
extern kmutex_t ipf_rw;
# endif
#else
int ipllog __P((void));
void init_ifp __P((void));
# ifdef __sgi
static int no_output __P((struct ifnet *, struct mbuf *,
struct sockaddr *));
static int write_output __P((struct ifnet *, struct mbuf *,
struct sockaddr *));
# else
static int no_output __P((struct ifnet *, struct mbuf *,
struct sockaddr *, struct rtentry *));
static int write_output __P((struct ifnet *, struct mbuf *,
struct sockaddr *, struct rtentry *));
# endif
#endif
#if (_BSDI_VERSION >= 199510) && defined(_KERNEL)
# include <sys/device.h>
# include <sys/conf.h>
struct cfdriver iplcd = {
NULL, "ipl", NULL, NULL, DV_DULL, 0
};
struct devsw iplsw = {
&iplcd,
iplopen, iplclose, iplread, nowrite, iplioctl, noselect, nommap,
nostrat, nodump, nopsize, 0,
nostop
};
#endif /* _BSDI_VERSION >= 199510 && _KERNEL */
1998-05-17 20:50:15 +04:00
#if defined(__NetBSD__) || defined(__OpenBSD__) || (_BSDI_VERSION >= 199701)
# include <sys/conf.h>
# if defined(NETBSD_PF)
# include <net/pfil.h>
/*
* We provide the fr_checkp name just to minimize changes later.
*/
int (*fr_checkp) __P((ip_t *ip, int hlen, void *ifp, int out, mb_t **mp));
# endif /* NETBSD_PF */
#endif /* __NetBSD__ */
#ifdef _KERNEL
# if defined(IPFILTER_LKM) && !defined(__sgi)
int iplidentify(s)
char *s;
{
if (strcmp(s, "ipl") == 0)
return 1;
return 0;
}
# endif /* IPFILTER_LKM */
Fix an ... interesting bug that resulted from namespace collision. Description: - A BSD pseudo-device initialization routine is declared as void <pseudo-device name>attach __P((int count)); in ioconf.c by config(8). main() calls these functions from a table. - IP Filter has functions iplattach() and ipldetach() (or, in the NetBSD case, were erroneously renamed ipfilterattach() and ipfilterdetach()). These functions are used to establish and disestablish the IP Filter "filter rule check" hook in the IP input/output stream. They are declared: int iplattach __P((void)); int ipldetach __P((void)); ..and are expected to return a value by iplioctl(). - When main() calls (by sheer coincidence!) iplattach(), the filter hook is established, and the IP Filter machinery labeled as "initialized". This causes all packets, whether or not the user intents to use filter rules, to be passed to the filter rule checker if "ipfilter" is configured into the kernel. - As a result of the above, a kludge existed to default to passing all packets (I can only assume that when this was originally committed, the symptom of the bug was noticed by the integrator, but the bug not actually found/fixed). - In iplioctl(), if the SIOCFRENB ioctl is issued with an argument of "enable" (i.e. user executed "ipf -E"), iplattach() will notice that the machinery is already initialized and return EBUSY. Fix: - Rename iplattach()/ipldetach() to ipl_enable() and ipl_disable(). - Create a pseudo-device entry stub named ipfilterattach() (NetBSD case) or iplattach() (all other). This is a noop; none of the machinery should be initialized until the caller expicitly enables the filter with ipf -E. Add a comment to note that.
1997-03-29 04:57:55 +03:00
/*
* BSD pseudo-device attach routine; this is a no-op.
Fix an ... interesting bug that resulted from namespace collision. Description: - A BSD pseudo-device initialization routine is declared as void <pseudo-device name>attach __P((int count)); in ioconf.c by config(8). main() calls these functions from a table. - IP Filter has functions iplattach() and ipldetach() (or, in the NetBSD case, were erroneously renamed ipfilterattach() and ipfilterdetach()). These functions are used to establish and disestablish the IP Filter "filter rule check" hook in the IP input/output stream. They are declared: int iplattach __P((void)); int ipldetach __P((void)); ..and are expected to return a value by iplioctl(). - When main() calls (by sheer coincidence!) iplattach(), the filter hook is established, and the IP Filter machinery labeled as "initialized". This causes all packets, whether or not the user intents to use filter rules, to be passed to the filter rule checker if "ipfilter" is configured into the kernel. - As a result of the above, a kludge existed to default to passing all packets (I can only assume that when this was originally committed, the symptom of the bug was noticed by the integrator, but the bug not actually found/fixed). - In iplioctl(), if the SIOCFRENB ioctl is issued with an argument of "enable" (i.e. user executed "ipf -E"), iplattach() will notice that the machinery is already initialized and return EBUSY. Fix: - Rename iplattach()/ipldetach() to ipl_enable() and ipl_disable(). - Create a pseudo-device entry stub named ipfilterattach() (NetBSD case) or iplattach() (all other). This is a noop; none of the machinery should be initialized until the caller expicitly enables the filter with ipf -E. Add a comment to note that.
1997-03-29 04:57:55 +03:00
*/
# if defined(__NetBSD__)
Fix an ... interesting bug that resulted from namespace collision. Description: - A BSD pseudo-device initialization routine is declared as void <pseudo-device name>attach __P((int count)); in ioconf.c by config(8). main() calls these functions from a table. - IP Filter has functions iplattach() and ipldetach() (or, in the NetBSD case, were erroneously renamed ipfilterattach() and ipfilterdetach()). These functions are used to establish and disestablish the IP Filter "filter rule check" hook in the IP input/output stream. They are declared: int iplattach __P((void)); int ipldetach __P((void)); ..and are expected to return a value by iplioctl(). - When main() calls (by sheer coincidence!) iplattach(), the filter hook is established, and the IP Filter machinery labeled as "initialized". This causes all packets, whether or not the user intents to use filter rules, to be passed to the filter rule checker if "ipfilter" is configured into the kernel. - As a result of the above, a kludge existed to default to passing all packets (I can only assume that when this was originally committed, the symptom of the bug was noticed by the integrator, but the bug not actually found/fixed). - In iplioctl(), if the SIOCFRENB ioctl is issued with an argument of "enable" (i.e. user executed "ipf -E"), iplattach() will notice that the machinery is already initialized and return EBUSY. Fix: - Rename iplattach()/ipldetach() to ipl_enable() and ipl_disable(). - Create a pseudo-device entry stub named ipfilterattach() (NetBSD case) or iplattach() (all other). This is a noop; none of the machinery should be initialized until the caller expicitly enables the filter with ipf -E. Add a comment to note that.
1997-03-29 04:57:55 +03:00
void
ipfilterattach(count)
int count;
{
/*
* Do nothing here, really. The filter will be enabled
* by the SIOCFRENB ioctl.
*/
}
# endif
# if defined(__NetBSD__)
int ipl_enable()
# else
int iplattach()
# endif
Fix an ... interesting bug that resulted from namespace collision. Description: - A BSD pseudo-device initialization routine is declared as void <pseudo-device name>attach __P((int count)); in ioconf.c by config(8). main() calls these functions from a table. - IP Filter has functions iplattach() and ipldetach() (or, in the NetBSD case, were erroneously renamed ipfilterattach() and ipfilterdetach()). These functions are used to establish and disestablish the IP Filter "filter rule check" hook in the IP input/output stream. They are declared: int iplattach __P((void)); int ipldetach __P((void)); ..and are expected to return a value by iplioctl(). - When main() calls (by sheer coincidence!) iplattach(), the filter hook is established, and the IP Filter machinery labeled as "initialized". This causes all packets, whether or not the user intents to use filter rules, to be passed to the filter rule checker if "ipfilter" is configured into the kernel. - As a result of the above, a kludge existed to default to passing all packets (I can only assume that when this was originally committed, the symptom of the bug was noticed by the integrator, but the bug not actually found/fixed). - In iplioctl(), if the SIOCFRENB ioctl is issued with an argument of "enable" (i.e. user executed "ipf -E"), iplattach() will notice that the machinery is already initialized and return EBUSY. Fix: - Rename iplattach()/ipldetach() to ipl_enable() and ipl_disable(). - Create a pseudo-device entry stub named ipfilterattach() (NetBSD case) or iplattach() (all other). This is a noop; none of the machinery should be initialized until the caller expicitly enables the filter with ipf -E. Add a comment to note that.
1997-03-29 04:57:55 +03:00
{
1997-05-25 16:40:11 +04:00
char *defpass;
int s;
# ifdef __sgi
int error;
# endif
SPL_NET(s);
if (ipl_inited || (fr_checkp == fr_check)) {
printf("IP Filter: already initialized\n");
SPL_X(s);
return EBUSY;
}
Fix an ... interesting bug that resulted from namespace collision. Description: - A BSD pseudo-device initialization routine is declared as void <pseudo-device name>attach __P((int count)); in ioconf.c by config(8). main() calls these functions from a table. - IP Filter has functions iplattach() and ipldetach() (or, in the NetBSD case, were erroneously renamed ipfilterattach() and ipfilterdetach()). These functions are used to establish and disestablish the IP Filter "filter rule check" hook in the IP input/output stream. They are declared: int iplattach __P((void)); int ipldetach __P((void)); ..and are expected to return a value by iplioctl(). - When main() calls (by sheer coincidence!) iplattach(), the filter hook is established, and the IP Filter machinery labeled as "initialized". This causes all packets, whether or not the user intents to use filter rules, to be passed to the filter rule checker if "ipfilter" is configured into the kernel. - As a result of the above, a kludge existed to default to passing all packets (I can only assume that when this was originally committed, the symptom of the bug was noticed by the integrator, but the bug not actually found/fixed). - In iplioctl(), if the SIOCFRENB ioctl is issued with an argument of "enable" (i.e. user executed "ipf -E"), iplattach() will notice that the machinery is already initialized and return EBUSY. Fix: - Rename iplattach()/ipldetach() to ipl_enable() and ipl_disable(). - Create a pseudo-device entry stub named ipfilterattach() (NetBSD case) or iplattach() (all other). This is a noop; none of the machinery should be initialized until the caller expicitly enables the filter with ipf -E. Add a comment to note that.
1997-03-29 04:57:55 +03:00
# ifdef NETBSD_PF
pfil_add_hook((void *)fr_check, PFIL_IN|PFIL_OUT);
# endif
# ifdef __sgi
error = ipfilter_sgi_attach();
if (error) {
SPL_X(s);
return error;
}
# endif
ipl_inited = 1;
bzero((char *)frcache, sizeof(frcache));
bzero((char *)nat_table, sizeof(nat_table));
fr_savep = fr_checkp;
fr_checkp = fr_check;
fr_saveslowtimo = inetsw[0].pr_slowtimo;
inetsw[0].pr_slowtimo = ipfr_slowtimer;
# ifdef IPFILTER_LOG
ipflog_init();
# endif
SPL_X(s);
1997-05-25 16:40:11 +04:00
if (fr_pass & FR_PASS)
defpass = "pass";
else if (fr_pass & FR_BLOCK)
defpass = "block";
else
defpass = "no-match -> block";
1997-11-14 15:40:06 +03:00
printf("IP Filter: initialized. Default = %s all, Logging = %s\n",
defpass,
# ifdef IPFILTER_LOG
"enabled");
# else
"disabled");
# endif
return 0;
}
Fix an ... interesting bug that resulted from namespace collision. Description: - A BSD pseudo-device initialization routine is declared as void <pseudo-device name>attach __P((int count)); in ioconf.c by config(8). main() calls these functions from a table. - IP Filter has functions iplattach() and ipldetach() (or, in the NetBSD case, were erroneously renamed ipfilterattach() and ipfilterdetach()). These functions are used to establish and disestablish the IP Filter "filter rule check" hook in the IP input/output stream. They are declared: int iplattach __P((void)); int ipldetach __P((void)); ..and are expected to return a value by iplioctl(). - When main() calls (by sheer coincidence!) iplattach(), the filter hook is established, and the IP Filter machinery labeled as "initialized". This causes all packets, whether or not the user intents to use filter rules, to be passed to the filter rule checker if "ipfilter" is configured into the kernel. - As a result of the above, a kludge existed to default to passing all packets (I can only assume that when this was originally committed, the symptom of the bug was noticed by the integrator, but the bug not actually found/fixed). - In iplioctl(), if the SIOCFRENB ioctl is issued with an argument of "enable" (i.e. user executed "ipf -E"), iplattach() will notice that the machinery is already initialized and return EBUSY. Fix: - Rename iplattach()/ipldetach() to ipl_enable() and ipl_disable(). - Create a pseudo-device entry stub named ipfilterattach() (NetBSD case) or iplattach() (all other). This is a noop; none of the machinery should be initialized until the caller expicitly enables the filter with ipf -E. Add a comment to note that.
1997-03-29 04:57:55 +03:00
/*
* Disable the filter by removing the hooks from the IP input/output
* stream.
*/
# if defined(__NetBSD__)
int ipl_disable()
# else
int ipldetach()
# endif
{
int s, i = FR_INQUE|FR_OUTQUE;
SPL_NET(s);
if (!ipl_inited)
{
printf("IP Filter: not initialized\n");
SPL_X(s);
return 0;
}
fr_checkp = fr_savep;
inetsw[0].pr_slowtimo = fr_saveslowtimo;
1998-05-17 20:50:15 +04:00
frflush(IPL_LOGIPF, &i);
ipl_inited = 0;
# ifdef NETBSD_PF
pfil_remove_hook((void *)fr_check, PFIL_IN|PFIL_OUT);
# endif
# ifdef __sgi
ipfilter_sgi_detach();
# endif
ipfr_unload();
ip_natunload();
fr_stateunload();
1997-07-05 09:38:14 +04:00
fr_authunload();
SPL_X(s);
return 0;
}
#endif /* _KERNEL */
static void frzerostats(data)
caddr_t data;
{
struct friostat fio;
bcopy((char *)frstats, (char *)fio.f_st,
sizeof(struct filterstats) * 2);
fio.f_fin[0] = ipfilter[0][0];
fio.f_fin[1] = ipfilter[0][1];
fio.f_fout[0] = ipfilter[1][0];
fio.f_fout[1] = ipfilter[1][1];
fio.f_acctin[0] = ipacct[0][0];
fio.f_acctin[1] = ipacct[0][1];
fio.f_acctout[0] = ipacct[1][0];
fio.f_acctout[1] = ipacct[1][1];
fio.f_active = fr_active;
fio.f_froute[0] = ipl_frouteok[0];
fio.f_froute[1] = ipl_frouteok[1];
IWCOPY((caddr_t)&fio, data, sizeof(fio));
bzero((char *)frstats, sizeof(*frstats) * 2);
}
/*
* Filter ioctl interface.
*/
#ifdef __sgi
int IPL_EXTERN(ioctl)(dev_t dev, int cmd, caddr_t data, int mode
# ifdef _KERNEL
, cred_t *cp, int *rp
# endif
)
#else
int IPL_EXTERN(ioctl)(dev, cmd, data, mode
1997-05-25 16:40:11 +04:00
#if ((_BSDI_VERSION >= 199510) || (BSD >= 199506) || (NetBSD >= 199511) || \
(__FreeBSD_version >= 220000)) && defined(_KERNEL)
, p)
struct proc *p;
#else
)
#endif
dev_t dev;
1998-05-17 20:50:15 +04:00
#if defined(__NetBSD__) || defined(__OpenBSD__) || (_BSDI_VERSION >= 199701)
u_long cmd;
#else
int cmd;
#endif
caddr_t data;
int mode;
#endif /* __sgi */
{
1997-07-05 09:38:14 +04:00
#if defined(_KERNEL) && !SOLARIS
int s;
#endif
1998-05-17 20:50:15 +04:00
int error = 0, unit = 0, tmp;
#ifdef _KERNEL
unit = GET_MINOR(dev);
1997-07-05 09:38:14 +04:00
if ((IPL_LOGMAX < unit) || (unit < 0))
return ENXIO;
#endif
SPL_NET(s);
1997-05-25 16:40:11 +04:00
if (unit == IPL_LOGNAT) {
error = nat_ioctl(data, cmd, mode);
SPL_X(s);
1997-05-25 16:40:11 +04:00
return error;
}
if (unit == IPL_LOGSTATE) {
error = fr_state_ioctl(data, cmd, mode);
SPL_X(s);
1997-05-25 16:40:11 +04:00
return error;
}
switch (cmd) {
case FIONREAD :
#ifdef IPFILTER_LOG
IWCOPY((caddr_t)&iplused[IPL_LOGIPF], (caddr_t)data,
sizeof(iplused[IPL_LOGIPF]));
#endif
break;
#if (!defined(IPFILTER_LKM) || defined(__NetBSD__)) && defined(_KERNEL)
case SIOCFRENB :
{
u_int enable;
# if defined(__NetBSD__) && defined(_KERNEL)
if ((securelevel >= 2) || !(mode & FWRITE))
#else
if (!(mode & FWRITE))
#endif
error = EPERM;
else {
IRCOPY(data, (caddr_t)&enable, sizeof(enable));
# if defined(__NetBSD__)
if (enable)
Fix an ... interesting bug that resulted from namespace collision. Description: - A BSD pseudo-device initialization routine is declared as void <pseudo-device name>attach __P((int count)); in ioconf.c by config(8). main() calls these functions from a table. - IP Filter has functions iplattach() and ipldetach() (or, in the NetBSD case, were erroneously renamed ipfilterattach() and ipfilterdetach()). These functions are used to establish and disestablish the IP Filter "filter rule check" hook in the IP input/output stream. They are declared: int iplattach __P((void)); int ipldetach __P((void)); ..and are expected to return a value by iplioctl(). - When main() calls (by sheer coincidence!) iplattach(), the filter hook is established, and the IP Filter machinery labeled as "initialized". This causes all packets, whether or not the user intents to use filter rules, to be passed to the filter rule checker if "ipfilter" is configured into the kernel. - As a result of the above, a kludge existed to default to passing all packets (I can only assume that when this was originally committed, the symptom of the bug was noticed by the integrator, but the bug not actually found/fixed). - In iplioctl(), if the SIOCFRENB ioctl is issued with an argument of "enable" (i.e. user executed "ipf -E"), iplattach() will notice that the machinery is already initialized and return EBUSY. Fix: - Rename iplattach()/ipldetach() to ipl_enable() and ipl_disable(). - Create a pseudo-device entry stub named ipfilterattach() (NetBSD case) or iplattach() (all other). This is a noop; none of the machinery should be initialized until the caller expicitly enables the filter with ipf -E. Add a comment to note that.
1997-03-29 04:57:55 +03:00
error = ipl_enable();
else
Fix an ... interesting bug that resulted from namespace collision. Description: - A BSD pseudo-device initialization routine is declared as void <pseudo-device name>attach __P((int count)); in ioconf.c by config(8). main() calls these functions from a table. - IP Filter has functions iplattach() and ipldetach() (or, in the NetBSD case, were erroneously renamed ipfilterattach() and ipfilterdetach()). These functions are used to establish and disestablish the IP Filter "filter rule check" hook in the IP input/output stream. They are declared: int iplattach __P((void)); int ipldetach __P((void)); ..and are expected to return a value by iplioctl(). - When main() calls (by sheer coincidence!) iplattach(), the filter hook is established, and the IP Filter machinery labeled as "initialized". This causes all packets, whether or not the user intents to use filter rules, to be passed to the filter rule checker if "ipfilter" is configured into the kernel. - As a result of the above, a kludge existed to default to passing all packets (I can only assume that when this was originally committed, the symptom of the bug was noticed by the integrator, but the bug not actually found/fixed). - In iplioctl(), if the SIOCFRENB ioctl is issued with an argument of "enable" (i.e. user executed "ipf -E"), iplattach() will notice that the machinery is already initialized and return EBUSY. Fix: - Rename iplattach()/ipldetach() to ipl_enable() and ipl_disable(). - Create a pseudo-device entry stub named ipfilterattach() (NetBSD case) or iplattach() (all other). This is a noop; none of the machinery should be initialized until the caller expicitly enables the filter with ipf -E. Add a comment to note that.
1997-03-29 04:57:55 +03:00
error = ipl_disable();
# else
if (enable)
error = iplattach();
else
error = ipldetach();
# endif
}
break;
}
#endif
case SIOCSETFF :
#if defined(__NetBSD__) && defined(_KERNEL)
if ((securelevel >= 2) || !(mode & FWRITE))
#else
if (!(mode & FWRITE))
#endif
error = EPERM;
else
IRCOPY(data, (caddr_t)&fr_flags, sizeof(fr_flags));
break;
case SIOCGETFF :
IWCOPY((caddr_t)&fr_flags, data, sizeof(fr_flags));
break;
case SIOCINAFR :
case SIOCRMAFR :
case SIOCADAFR :
case SIOCZRLST :
#if defined(__NetBSD__) && defined(_KERNEL)
if ((securelevel >= 2) || !(mode & FWRITE))
#else
if (!(mode & FWRITE))
#endif
error = EPERM;
else
1997-07-05 09:38:14 +04:00
error = frrequest(unit, cmd, data, fr_active);
break;
case SIOCINIFR :
case SIOCRMIFR :
case SIOCADIFR :
#if defined(__NetBSD__) && defined(_KERNEL)
if ((securelevel >= 2) || !(mode & FWRITE))
#else
if (!(mode & FWRITE))
#endif
error = EPERM;
else
1997-07-05 09:38:14 +04:00
error = frrequest(unit, cmd, data, 1 - fr_active);
break;
case SIOCSWAPA :
#if defined(__NetBSD__) && defined(_KERNEL)
if((securelevel >= 2) || !(mode & FWRITE))
#else
if (!(mode & FWRITE))
#endif
error = EPERM;
else {
bzero((char *)frcache, sizeof(frcache[0]) * 2);
*(u_int *)data = fr_active;
fr_active = 1 - fr_active;
}
break;
case SIOCGETFS :
{
struct friostat fio;
bcopy((char *)frstats, (char *)fio.f_st,
sizeof(struct filterstats) * 2);
fio.f_fin[0] = ipfilter[0][0];
fio.f_fin[1] = ipfilter[0][1];
fio.f_fout[0] = ipfilter[1][0];
fio.f_fout[1] = ipfilter[1][1];
fio.f_acctin[0] = ipacct[0][0];
fio.f_acctin[1] = ipacct[0][1];
fio.f_acctout[0] = ipacct[1][0];
fio.f_acctout[1] = ipacct[1][1];
1997-07-05 09:38:14 +04:00
fio.f_auth = ipauth;
fio.f_active = fr_active;
fio.f_froute[0] = ipl_frouteok[0];
fio.f_froute[1] = ipl_frouteok[1];
IWCOPY((caddr_t)&fio, data, sizeof(fio));
break;
}
case SIOCFRZST :
#if defined(__NetBSD__) && defined(_KERNEL)
if ((securelevel >= 2) || !(mode & FWRITE))
#else
if (!(mode & FWRITE))
#endif
error = EPERM;
else
frzerostats(data);
break;
case SIOCIPFFL :
#if defined(__NetBSD__) && defined(_KERNEL)
if ((securelevel >= 2) || !(mode & FWRITE))
#else
if (!(mode & FWRITE))
#endif
error = EPERM;
1998-05-17 20:50:15 +04:00
else {
IRCOPY(data, (caddr_t)&tmp, sizeof(tmp));
frflush(unit, &tmp);
IWCOPY((caddr_t)&tmp, data, sizeof(tmp));
}
break;
#ifdef IPFILTER_LOG
case SIOCIPFFB :
#if defined(__NetBSD__) && defined(_KERNEL)
if ((securelevel >= 2) || !(mode & FWRITE))
#else
if (!(mode & FWRITE))
#endif
error = EPERM;
else
*(int *)data = ipflog_clear(unit);
break;
#endif /* IPFILTER_LOG */
case SIOCGFRST :
IWCOPY((caddr_t)ipfr_fragstats(), data, sizeof(ipfrstat_t));
break;
case SIOCAUTHW :
case SIOCAUTHR :
#if defined(__NetBSD__) && defined(_KERNEL)
if ((securelevel >= 2) || !(mode & FWRITE)) {
#else
1997-07-05 09:38:14 +04:00
if (!(mode & FWRITE)) {
#endif
1997-07-05 09:38:14 +04:00
error = EPERM;
break;
}
case SIOCATHST :
1997-07-05 09:38:14 +04:00
error = fr_auth_ioctl(data, cmd, NULL, NULL);
break;
case SIOCFRSYN :
#if defined(__NetBSD__) && defined(_KERNEL)
if ((securelevel >= 2) || !(mode & FWRITE))
#else
if (!(mode & FWRITE))
#endif
error = EPERM;
else {
#if defined(_KERNEL) && defined(__sgi)
ipfsync();
#endif
frsync();
}
break;
default :
error = EINVAL;
break;
}
SPL_X(s);
return error;
}
static void frsync()
{
#ifdef _KERNEL
struct ifnet *ifp;
# if defined(__OpenBSD__) || (NetBSD >= 199511)
for (ifp = ifnet.tqh_first; ifp; ifp = ifp->if_list.tqe_next)
# else
for (ifp = ifnet; ifp; ifp = ifp->if_next)
# endif
ip_natsync(ifp);
#endif
}
1997-07-05 09:38:14 +04:00
static void fixskip(listp, rp, addremove)
frentry_t **listp, *rp;
int addremove;
{
frentry_t *fp;
int rules = 0, rn = 0;
for (fp = *listp; fp && (fp != rp); fp = fp->fr_next, rules++)
;
if (!fp)
return;
for (fp = *listp; fp && (fp != rp); fp = fp->fr_next, rn++)
if (fp->fr_skip && (rn + fp->fr_skip >= rules))
fp->fr_skip += addremove;
}
static int frrequest(unit, req, data, set)
int unit;
#if defined(__NetBSD__) || defined(__OpenBSD__)
1997-03-18 10:14:45 +03:00
u_long req;
#else
int req;
#endif
1997-03-18 10:14:45 +03:00
int set;
caddr_t data;
{
register frentry_t *fp, *f, **fprev;
register frentry_t **ftail;
frentry_t frd;
frdest_t *fdp;
frgroup_t *fg = NULL;
int error = 0, in, group;
fp = &frd;
IRCOPY(data, (caddr_t)fp, sizeof(*fp));
/*
* Check that the group number does exist and that if a head group
* has been specified, doesn't exist.
*/
if (fp->fr_grhead &&
fr_findgroup(fp->fr_grhead, fp->fr_flags, unit, set, NULL))
return EEXIST;
if (fp->fr_group &&
!fr_findgroup(fp->fr_group, fp->fr_flags, unit, set, NULL))
return ESRCH;
in = (fp->fr_flags & FR_INQUE) ? 0 : 1;
1997-07-05 09:38:14 +04:00
if (unit == IPL_LOGAUTH)
ftail = fprev = &ipauth;
else if (fp->fr_flags & FR_ACCOUNT)
ftail = fprev = &ipacct[in][set];
1997-07-05 09:38:14 +04:00
else if (fp->fr_flags & (FR_OUTQUE|FR_INQUE))
ftail = fprev = &ipfilter[in][set];
else
return ESRCH;
if ((group = fp->fr_group)) {
if (!(fg = fr_findgroup(group, fp->fr_flags, unit, set, NULL)))
return ESRCH;
ftail = fprev = fg->fg_start;
}
bzero((char *)frcache, sizeof(frcache[0]) * 2);
if (*fp->fr_ifname) {
fp->fr_ifa = GETUNIT(fp->fr_ifname);
if (!fp->fr_ifa)
fp->fr_ifa = (void *)-1;
}
fdp = &fp->fr_dif;
fp->fr_flags &= ~FR_DUP;
if (*fdp->fd_ifname) {
fdp->fd_ifp = GETUNIT(fdp->fd_ifname);
if (!fdp->fd_ifp)
fdp->fd_ifp = (struct ifnet *)-1;
else
fp->fr_flags |= FR_DUP;
}
fdp = &fp->fr_tif;
if (*fdp->fd_ifname) {
fdp->fd_ifp = GETUNIT(fdp->fd_ifname);
if (!fdp->fd_ifp)
fdp->fd_ifp = (struct ifnet *)-1;
}
/*
* Look for a matching filter rule, but don't include the next or
* interface pointer in the comparison (fr_next, fr_ifa).
*/
for (; (f = *ftail); ftail = &f->fr_next)
if (bcmp((char *)&f->fr_ip, (char *)&fp->fr_ip,
FR_CMPSIZ) == 0)
break;
/*
* If zero'ing statistics, copy current to caller and zero.
*/
if (req == SIOCZRLST) {
if (!f)
return ESRCH;
IWCOPY((caddr_t)f, data, sizeof(*f));
f->fr_hits = 0;
f->fr_bytes = 0;
return 0;
}
if (!f) {
ftail = fprev;
if (req != SIOCINAFR && req != SIOCINIFR)
while ((f = *ftail))
ftail = &f->fr_next;
else if (fp->fr_hits)
while (--fp->fr_hits && (f = *ftail))
ftail = &f->fr_next;
f = NULL;
}
if (req == SIOCDELFR || req == SIOCRMIFR) {
if (!f)
error = ESRCH;
else {
if (f->fr_ref > 1)
return EBUSY;
if (fg && fg->fg_head)
fg->fg_head->fr_ref--;
1997-07-05 09:38:14 +04:00
if (unit == IPL_LOGAUTH)
return fr_auth_ioctl(data, req, f, ftail);
if (f->fr_grhead)
fr_delgroup(f->fr_grhead, fp->fr_flags, unit,
set);
1997-07-05 09:38:14 +04:00
fixskip(fprev, f, -1);
*ftail = f->fr_next;
KFREE(f);
}
} else {
if (f)
error = EEXIST;
else {
1997-07-05 09:38:14 +04:00
if (unit == IPL_LOGAUTH)
return fr_auth_ioctl(data, req, f, ftail);
KMALLOC(f, frentry_t *, sizeof(*f));
if (f != NULL) {
if (fg && fg->fg_head)
fg->fg_head->fr_ref++;
bcopy((char *)fp, (char *)f, sizeof(*f));
f->fr_ref = 1;
f->fr_hits = 0;
f->fr_next = *ftail;
*ftail = f;
1997-07-05 09:38:14 +04:00
if (req == SIOCINIFR || req == SIOCINAFR)
fixskip(fprev, f, 1);
f->fr_grp = NULL;
if ((group = f->fr_grhead))
fg = fr_addgroup(group, f, unit, set);
} else
error = ENOMEM;
}
}
return (error);
}
#ifdef _KERNEL
/*
* routines below for saving IP headers to buffer
*/
#ifdef __sgi
# ifdef _KERNEL
int IPL_EXTERN(open)(dev_t *pdev, int flags, int devtype, cred_t *cp)
# else
int IPL_EXTERN(open)(dev_t dev, int flags)
# endif
#else
int IPL_EXTERN(open)(dev, flags
1997-07-05 09:38:14 +04:00
# if ((_BSDI_VERSION >= 199510) || (BSD >= 199506) || (NetBSD >= 199511) || \
1997-05-25 16:40:11 +04:00
(__FreeBSD_version >= 220000)) && defined(_KERNEL)
, devtype, p)
int devtype;
struct proc *p;
# else
)
# endif
dev_t dev;
int flags;
#endif /* __sgi */
{
#if defined(__sgi) && defined(_KERNEL)
u_int min = geteminor(*pdev);
#else
u_int min = GET_MINOR(dev);
#endif
1997-05-25 16:40:11 +04:00
if (2 < min)
min = ENXIO;
else
min = 0;
return min;
}
#ifdef __sgi
int IPL_EXTERN(close)(dev_t dev, int flags, int devtype, cred_t *cp)
#else
int IPL_EXTERN(close)(dev, flags
1997-07-05 09:38:14 +04:00
# if ((_BSDI_VERSION >= 199510) || (BSD >= 199506) || (NetBSD >= 199511) || \
1997-05-25 16:40:11 +04:00
(__FreeBSD_version >= 220000)) && defined(_KERNEL)
, devtype, p)
int devtype;
struct proc *p;
# else
)
# endif
dev_t dev;
int flags;
#endif /* __sgi */
{
u_int min = GET_MINOR(dev);
1997-05-25 16:40:11 +04:00
if (2 < min)
min = ENXIO;
else
min = 0;
return min;
}
/*
* iplread/ipllog
* both of these must operate with at least splnet() lest they be
* called during packet processing and cause an inconsistancy to appear in
* the filter lists.
*/
#ifdef __sgi
int IPL_EXTERN(read)(dev_t dev, uio_t *uio, cred_t *crp)
#else
# if BSD >= 199306
int IPL_EXTERN(read)(dev, uio, ioflag)
int ioflag;
# else
int IPL_EXTERN(read)(dev, uio)
# endif
dev_t dev;
register struct uio *uio;
#endif /* __sgi */
{
1997-07-05 09:38:14 +04:00
# ifdef IPFILTER_LOG
return ipflog_read(GET_MINOR(dev), uio);
1997-07-05 09:38:14 +04:00
# else
return ENXIO;
1997-07-05 09:38:14 +04:00
# endif
}
/*
* send_reset - this could conceivably be a call to tcp_respond(), but that
* requires a large amount of setting up and isn't any more efficient.
*/
int send_reset(ti)
struct tcpiphdr *ti;
{
struct tcpiphdr *tp;
struct tcphdr *tcp;
struct mbuf *m;
1998-05-17 20:50:15 +04:00
int tlen = 0, err;
1997-11-14 15:40:06 +03:00
ip_t *ip;
1997-07-05 09:38:14 +04:00
# if defined(__FreeBSD_version) && (__FreeBSD_version >= 220000)
1997-05-25 16:40:11 +04:00
struct route ro;
1997-07-05 09:38:14 +04:00
# endif
if (ti->ti_flags & TH_RST)
return -1; /* feedback loop */
# if (BSD < 199306) || defined(__sgi)
m = m_get(M_DONTWAIT, MT_HEADER);
# else
m = m_gethdr(M_DONTWAIT, MT_HEADER);
m->m_data += max_linkhdr;
# endif
if (m == NULL)
return -1;
if (ti->ti_flags & TH_SYN)
tlen = 1;
m->m_len = sizeof (struct tcpiphdr);
# if BSD >= 199306
m->m_pkthdr.len = sizeof (struct tcpiphdr);
m->m_pkthdr.rcvif = (struct ifnet *)0;
# endif
bzero(mtod(m, char *), sizeof(struct tcpiphdr));
ip = mtod(m, struct ip *);
tp = mtod(m, struct tcpiphdr *);
tcp = (struct tcphdr *)((char *)ip + sizeof(struct ip));
ip->ip_src.s_addr = ti->ti_dst.s_addr;
ip->ip_dst.s_addr = ti->ti_src.s_addr;
tcp->th_dport = ti->ti_sport;
tcp->th_sport = ti->ti_dport;
tcp->th_ack = htonl(ntohl(ti->ti_seq) + tlen);
tcp->th_off = sizeof(struct tcphdr) >> 2;
tcp->th_flags = TH_RST|TH_ACK;
tp->ti_pr = ((struct ip *)ti)->ip_p;
tp->ti_len = htons(sizeof(struct tcphdr));
tcp->th_sum = in_cksum(m, sizeof(struct tcpiphdr));
ip->ip_tos = ((struct ip *)ti)->ip_tos;
ip->ip_p = ((struct ip *)ti)->ip_p;
ip->ip_len = sizeof (struct tcpiphdr);
# if (BSD < 199306) || defined(__sgi)
ip->ip_ttl = tcp_ttl;
# else
ip->ip_ttl = ip_defttl;
# endif
1997-07-05 09:38:14 +04:00
# if defined(__FreeBSD_version) && (__FreeBSD_version >= 220000)
1997-05-25 16:40:11 +04:00
bzero((char *)&ro, sizeof(ro));
1998-05-17 20:50:15 +04:00
err = ip_output(m, (struct mbuf *)0, &ro, 0, 0);
1997-05-25 16:40:11 +04:00
if (ro.ro_rt)
RTFREE(ro.ro_rt);
1997-07-05 09:38:14 +04:00
# else
/*
* extra 0 in case of multicast
*/
1998-05-17 20:50:15 +04:00
err = ip_output(m, (struct mbuf *)0, 0, 0, 0);
1997-07-05 09:38:14 +04:00
# endif
1998-05-17 20:50:15 +04:00
return err;
}
# if !defined(IPFILTER_LKM) && (__FreeBSD_version < 300000) && !defined(__sgi)
# if (BSD < 199306)
1997-05-25 16:40:11 +04:00
int iplinit __P((void));
int
1997-05-25 16:40:11 +04:00
# else
void iplinit __P((void));
void
1997-05-25 16:40:11 +04:00
# endif
iplinit()
{
# if defined(__NetBSD__)
(void) ipl_enable();
# else
(void) iplattach();
# endif
ip_init();
}
# endif /* !IPFILTER_LKM && __FreeBSD_version < 300000 */
size_t mbufchainlen(m0)
register struct mbuf *m0;
{
register size_t len = 0;
for (; m0; m0 = m0->m_next)
len += m0->m_len;
return len;
}
void ipfr_fastroute(m0, fin, fdp)
struct mbuf *m0;
fr_info_t *fin;
frdest_t *fdp;
{
register struct ip *ip, *mhip;
register struct mbuf *m = m0;
register struct route *ro;
struct ifnet *ifp = fdp->fd_ifp;
int len, off, error = 0;
int hlen = fin->fin_hlen;
struct route iproute;
struct sockaddr_in *dst;
ip = mtod(m0, struct ip *);
/*
* Route packet.
*/
ro = &iproute;
bzero((caddr_t)ro, sizeof (*ro));
dst = (struct sockaddr_in *)&ro->ro_dst;
dst->sin_family = AF_INET;
dst->sin_addr = fdp->fd_ip.s_addr ? fdp->fd_ip : ip->ip_dst;
1997-07-05 09:38:14 +04:00
# ifdef __bsdi__
dst->sin_len = sizeof(*dst);
1997-07-05 09:38:14 +04:00
# endif
# if (BSD >= 199306) && !defined(__NetBSD__) && !defined(__bsdi__) && \
!defined(__OpenBSD__)
# ifdef RTF_CLONING
rtalloc_ign(ro, RTF_CLONING);
# else
rtalloc_ign(ro, RTF_PRCLONING);
# endif
# else
rtalloc(ro);
# endif
if (!ifp) {
if (!(fin->fin_fr->fr_flags & FR_FASTROUTE)) {
error = -2;
goto bad;
}
if (ro->ro_rt == 0 || (ifp = ro->ro_rt->rt_ifp) == 0) {
if (in_localaddr(ip->ip_dst))
error = EHOSTUNREACH;
else
error = ENETUNREACH;
goto bad;
}
if (ro->ro_rt->rt_flags & RTF_GATEWAY)
dst = (struct sockaddr_in *)&ro->ro_rt->rt_gateway;
}
1998-05-17 20:50:15 +04:00
if (ro->ro_rt)
ro->ro_rt->rt_use++;
/*
* For input packets which are being "fastrouted", they won't
* go back through output filtering and miss their chance to get
1998-11-22 18:17:18 +03:00
* NAT'd and counted.
*/
1998-11-22 18:17:18 +03:00
if (fin->fin_out == 0) {
if ((fin->fin_fr = ipacct[1][fr_active]) &&
(FR_SCANLIST(FR_NOMATCH, ip, fin, m) & FR_ACCOUNT)) {
ATOMIC_INC(frstats[1].fr_acct);
}
fin->fin_fr = NULL;
(void) ip_natout(ip, hlen, fin);
}
if (fin->fin_out)
ip->ip_sum = 0;
/*
* If small enough for interface, can just send directly.
*/
if (ip->ip_len <= ifp->if_mtu) {
# ifndef sparc
# ifndef __NetBSD__
ip->ip_id = htons(ip->ip_id);
# endif
ip->ip_len = htons(ip->ip_len);
ip->ip_off = htons(ip->ip_off);
# endif
if (!ip->ip_sum)
ip->ip_sum = in_cksum(m, hlen);
# if BSD >= 199306
error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst,
ro->ro_rt);
# else
error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst);
# endif
goto done;
}
/*
* Too large for interface; fragment if possible.
* Must be able to put at least 8 bytes per fragment.
*/
if (ip->ip_off & IP_DF) {
/* Send ICMP error here */
struct mbuf *mcopy;
mcopy = m_copy(m, 0, imin((int)ip->ip_len, 68));
if (mcopy)
icmp_error(mcopy, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
ip->ip_dst.s_addr, ifp);
error = EMSGSIZE;
goto bad;
}
len = (ifp->if_mtu - hlen) &~ 7;
if (len < 8) {
error = EMSGSIZE;
goto bad;
}
{
int mhlen, firstlen = len;
struct mbuf **mnext = &m->m_act;
/*
* Loop through length of segment after first fragment,
* make new header and copy data of each part and link onto chain.
*/
m0 = m;
mhlen = sizeof (struct ip);
for (off = hlen + len; off < ip->ip_len; off += len) {
MGETHDR(m, M_DONTWAIT, MT_HEADER);
if (m == 0) {
error = ENOBUFS;
goto bad;
}
# if BSD >= 199306
m->m_data += max_linkhdr;
# else
m->m_off = MMAXOFF - hlen;
# endif
mhip = mtod(m, struct ip *);
bcopy((char *)ip, (char *)mhip, sizeof(*ip));
if (hlen > sizeof (struct ip)) {
mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
mhip->ip_hl = mhlen >> 2;
}
m->m_len = mhlen;
mhip->ip_off = ((off - hlen) >> 3) + (ip->ip_off & ~IP_MF);
if (ip->ip_off & IP_MF)
mhip->ip_off |= IP_MF;
if (off + len >= ip->ip_len)
len = ip->ip_len - off;
else
mhip->ip_off |= IP_MF;
mhip->ip_len = htons((u_short)(len + mhlen));
m->m_next = m_copy(m0, off, len);
if (m->m_next == 0) {
error = ENOBUFS; /* ??? */
goto sendorfree;
}
# ifndef sparc
mhip->ip_off = htons((u_short)mhip->ip_off);
# endif
mhip->ip_sum = 0;
mhip->ip_sum = in_cksum(m, mhlen);
*mnext = m;
mnext = &m->m_act;
}
/*
* Update first fragment by trimming what's been copied out
* and updating header, then send each fragment (in order).
*/
m_adj(m0, hlen + firstlen - ip->ip_len);
ip->ip_len = htons((u_short)(hlen + firstlen));
ip->ip_off = htons((u_short)(ip->ip_off | IP_MF));
ip->ip_sum = 0;
ip->ip_sum = in_cksum(m0, hlen);
sendorfree:
for (m = m0; m; m = m0) {
m0 = m->m_act;
m->m_act = 0;
if (error == 0)
# if BSD >= 199306
error = (*ifp->if_output)(ifp, m,
(struct sockaddr *)dst, ro->ro_rt);
# else
error = (*ifp->if_output)(ifp, m,
(struct sockaddr *)dst);
# endif
else
m_freem(m);
}
}
done:
if (!error)
ipl_frouteok[0]++;
else
ipl_frouteok[1]++;
1998-11-22 18:17:18 +03:00
if (ro->ro_rt)
RTFREE(ro->ro_rt);
return;
bad:
m_freem(m);
goto done;
}
#else /* #ifdef _KERNEL */
#ifdef __sgi
static int no_output __P((struct ifnet *ifp, struct mbuf *m,
struct sockaddr *s))
#else
static int no_output __P((struct ifnet *ifp, struct mbuf *m,
struct sockaddr *s, struct rtentry *rt))
#endif
{
return 0;
}
# ifdef __STDC__
#ifdef __sgi
static int write_output __P((struct ifnet *ifp, struct mbuf *m,
struct sockaddr *s))
#else
static int write_output __P((struct ifnet *ifp, struct mbuf *m,
struct sockaddr *s, struct rtentry *rt))
#endif
{
# if !(defined(NetBSD) && (NetBSD <= 1991011) && (NetBSD >= 199606)) || \
(defined(OpenBSD) && (OpenBSD >= 199603))
ip_t *ip = (ip_t *)m;
1997-07-05 09:38:14 +04:00
# endif
# else
static int write_output(ifp, ip)
struct ifnet *ifp;
ip_t *ip;
{
# endif
FILE *fp;
char fname[32];
# if (defined(NetBSD) && (NetBSD <= 1991011) && (NetBSD >= 199606)) || \
(defined(OpenBSD) && (OpenBSD >= 199603))
sprintf(fname, "/tmp/%s", ifp->if_xname);
if ((fp = fopen(fname, "a"))) {
fclose(fp);
}
1997-07-05 09:38:14 +04:00
# else
sprintf(fname, "/tmp/%s%d", ifp->if_name, ifp->if_unit);
if ((fp = fopen(fname, "a"))) {
fwrite((char *)ip, ntohs(ip->ip_len), 1, fp);
fclose(fp);
}
1997-07-05 09:38:14 +04:00
# endif
return 0;
}
struct ifnet *get_unit(name)
char *name;
{
struct ifnet *ifp, **ifa;
# if (defined(NetBSD) && (NetBSD <= 1991011) && (NetBSD >= 199606)) || \
(defined(OpenBSD) && (OpenBSD >= 199603))
for (ifa = ifneta; ifa && (ifp = *ifa); ifa++) {
if (!strcmp(name, ifp->if_xname))
return ifp;
}
1997-07-05 09:38:14 +04:00
# else
char ifname[32], *s;
for (ifa = ifneta; ifa && (ifp = *ifa); ifa++) {
(void) sprintf(ifname, "%s%d", ifp->if_name, ifp->if_unit);
if (!strcmp(name, ifname))
return ifp;
}
1997-07-05 09:38:14 +04:00
# endif
if (!ifneta) {
ifneta = (struct ifnet **)malloc(sizeof(ifp) * 2);
ifneta[1] = NULL;
ifneta[0] = (struct ifnet *)calloc(1, sizeof(*ifp));
nifs = 1;
} else {
nifs++;
ifneta = (struct ifnet **)realloc(ifneta,
(nifs + 1) * sizeof(*ifa));
ifneta[nifs] = NULL;
ifneta[nifs - 1] = (struct ifnet *)malloc(sizeof(*ifp));
}
ifp = ifneta[nifs - 1];
# if (defined(NetBSD) && (NetBSD <= 1991011) && (NetBSD >= 199606)) || \
(defined(OpenBSD) && (OpenBSD >= 199603))
strncpy(ifp->if_xname, name, sizeof(ifp->if_xname));
1997-07-05 09:38:14 +04:00
# else
for (s = name; *s && !isdigit(*s); s++)
;
if (*s && isdigit(*s)) {
ifp->if_unit = atoi(s);
ifp->if_name = (char *)malloc(s - name + 1);
strncpy(ifp->if_name, name, s - name);
ifp->if_name[s - name] = '\0';
} else {
ifp->if_name = strdup(name);
ifp->if_unit = -1;
}
1997-07-05 09:38:14 +04:00
# endif
ifp->if_output = no_output;
return ifp;
}
1997-07-05 09:38:14 +04:00
void init_ifp()
{
FILE *fp;
struct ifnet *ifp, **ifa;
char fname[32];
# if (defined(NetBSD) && (NetBSD <= 1991011) && (NetBSD >= 199606)) || \
(defined(OpenBSD) && (OpenBSD >= 199603))
for (ifa = ifneta; ifa && (ifp = *ifa); ifa++) {
ifp->if_output = write_output;
sprintf(fname, "/tmp/%s", ifp->if_xname);
if ((fp = fopen(fname, "w")))
fclose(fp);
}
1997-07-05 09:38:14 +04:00
# else
for (ifa = ifneta; ifa && (ifp = *ifa); ifa++) {
ifp->if_output = write_output;
sprintf(fname, "/tmp/%s%d", ifp->if_name, ifp->if_unit);
if ((fp = fopen(fname, "w")))
fclose(fp);
}
1997-07-05 09:38:14 +04:00
# endif
}
void ipfr_fastroute(ip, fin, fdp)
ip_t *ip;
fr_info_t *fin;
frdest_t *fdp;
{
struct ifnet *ifp = fdp->fd_ifp;
if (!ifp)
return; /* no routing table out here */
ip->ip_len = htons((u_short)ip->ip_len);
ip->ip_off = htons((u_short)(ip->ip_off | IP_MF));
ip->ip_sum = 0;
#ifdef __sgi
(*ifp->if_output)(ifp, (void *)ip, NULL);
#else
(*ifp->if_output)(ifp, (void *)ip, NULL, 0);
#endif
}
int ipllog __P((void))
{
verbose("l");
return 0;
}
int send_reset(ip, ifp)
ip_t *ip;
struct ifnet *ifp;
{
verbose("- TCP RST sent\n");
return 0;
}
int icmp_error(ip, ifp)
ip_t *ip;
struct ifnet *ifp;
{
verbose("- TCP RST sent\n");
return 0;
}
#endif /* _KERNEL */