Gitiles
Code Review Sign In
code.illumos.org / illumos-gate / f4b3ec61df05330d25f55a36b975b4d7519fdeb1 / . / usr / src / uts / common / inet / ipf / ip_state.c
blob: 15e3b7ac8a2ab69b6c66181897c01d626085c084 [file] [log] [blame]
/*
* Copyright (C) 1995-2003 by Darren Reed.
*
* See the IPFILTER.LICENCE file for details on licencing.
*
* Copyright 2007 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
*/
#pragma ident "%Z%%M% %I% %E% SMI"
#if defined(KERNEL) || defined(_KERNEL)
# undef KERNEL
# undef _KERNEL
# define KERNEL 1
# define _KERNEL 1
#endif
#include <sys/errno.h>
#include <sys/types.h>
#include <sys/param.h>
#include <sys/file.h>
#if defined(__NetBSD__) && (NetBSD >= 199905) && !defined(IPFILTER_LKM) && \
defined(_KERNEL)
# include "opt_ipfilter_log.h"
#endif
#if defined(_KERNEL) && defined(__FreeBSD_version) && \
(__FreeBSD_version >= 400000) && !defined(KLD_MODULE)
#include "opt_inet6.h"
#endif
#if !defined(_KERNEL) && !defined(__KERNEL__)
# include <stdio.h>
# include <stdlib.h>
# include <string.h>
# define _KERNEL
# ifdef __OpenBSD__
struct file;
# endif
# include <sys/uio.h>
# undef _KERNEL
#endif
#if defined(_KERNEL) && (__FreeBSD_version >= 220000)
# include <sys/filio.h>
# include <sys/fcntl.h>
# if (__FreeBSD_version >= 300000) && !defined(IPFILTER_LKM)
# include "opt_ipfilter.h"
# endif
#else
# include <sys/ioctl.h>
#endif
#include <sys/time.h>
#if !defined(linux)
# include <sys/protosw.h>
#endif
#include <sys/socket.h>
#if defined(_KERNEL)
# include <sys/systm.h>
# if !defined(__SVR4) && !defined(__svr4__)
# include <sys/mbuf.h>
# endif
#endif
#if defined(__SVR4) || defined(__svr4__)
# include <sys/filio.h>
# include <sys/byteorder.h>
# ifdef _KERNEL
# include <sys/dditypes.h>
# endif
# include <sys/stream.h>
# include <sys/kmem.h>
#endif
#include <net/if.h>
#ifdef sun
# include <net/af.h>
#endif
#include <net/route.h>
#include <netinet/in.h>
#include <netinet/in_systm.h>
#include <netinet/ip.h>
#include <netinet/tcp.h>
#if !defined(linux)
# include <netinet/ip_var.h>
#endif
#if !defined(__hpux) && !defined(linux)
# include <netinet/tcp_fsm.h>
#endif
#include <netinet/udp.h>
#include <netinet/ip_icmp.h>
#include "netinet/ip_compat.h"
#include <netinet/tcpip.h>
#include "netinet/ip_fil.h"
#include "netinet/ip_nat.h"
#include "netinet/ip_frag.h"
#include "netinet/ip_state.h"
#include "netinet/ip_proxy.h"
#include "netinet/ipf_stack.h"
#ifdef IPFILTER_SYNC
#include "netinet/ip_sync.h"
#endif
#ifdef IPFILTER_SCAN
#include "netinet/ip_scan.h"
#endif
#ifdef USE_INET6
#include <netinet/icmp6.h>
#endif
#if (__FreeBSD_version >= 300000)
# include <sys/malloc.h>
# if defined(_KERNEL) && !defined(IPFILTER_LKM)
# include <sys/libkern.h>
# include <sys/systm.h>
# endif
#endif
/* END OF INCLUDES */
#if !defined(lint)
static const char sccsid[] = "@(#)ip_state.c 1.8 6/5/96 (C) 1993-2000 Darren Reed";
static const char rcsid[] = "@(#)$Id: ip_state.c,v 2.186.2.36 2005/08/11 19:58:03 darrenr Exp $";
#endif
#ifdef USE_INET6
static ipstate_t *fr_checkicmp6matchingstate __P((fr_info_t *));
#endif
static ipstate_t *fr_matchsrcdst __P((fr_info_t *, ipstate_t *, i6addr_t *,
i6addr_t *, tcphdr_t *, u_32_t));
static ipstate_t *fr_checkicmpmatchingstate __P((fr_info_t *));
static int fr_state_flush __P((int, int, ipf_stack_t *));
static ips_stat_t *fr_statetstats __P((ipf_stack_t *));
static void fr_delstate __P((ipstate_t *, int, ipf_stack_t *));
static int fr_state_remove __P((caddr_t, ipf_stack_t *));
static void fr_ipsmove __P((ipstate_t *, u_int, ipf_stack_t *));
static int fr_tcpstate __P((fr_info_t *, tcphdr_t *, ipstate_t *));
static int fr_tcpoptions __P((fr_info_t *, tcphdr_t *, tcpdata_t *));
static ipstate_t *fr_stclone __P((fr_info_t *, tcphdr_t *, ipstate_t *));
static void fr_fixinisn __P((fr_info_t *, ipstate_t *));
static void fr_fixoutisn __P((fr_info_t *, ipstate_t *));
static void fr_checknewisn __P((fr_info_t *, ipstate_t *));
static int fr_stateiter __P((ipftoken_t *, ipfgeniter_t *, ipf_stack_t *));
int fr_stputent __P((caddr_t, ipf_stack_t *));
int fr_stgetent __P((caddr_t, ipf_stack_t *));
#define ONE_DAY IPF_TTLVAL(1 * 86400) /* 1 day */
#define FIVE_DAYS (5 * ONE_DAY)
#define DOUBLE_HASH(x, ifs) \
(((x) + ifs->ifs_ips_seed[(x) % ifs->ifs_fr_statesize]) % ifs->ifs_fr_statesize)
/* ------------------------------------------------------------------------ */
/* Function: fr_stateinit */
/* Returns: int - 0 == success, -1 == failure */
/* Parameters: Nil */
/* */
/* Initialise all the global variables used within the state code. */
/* This action also includes initiailising locks. */
/* ------------------------------------------------------------------------ */
int fr_stateinit(ifs)
ipf_stack_t *ifs;
{
int i;
ifs->ifs_fr_tcpidletimeout = FIVE_DAYS;
ifs->ifs_fr_tcpclosewait = IPF_TTLVAL(TCP_MSL);
ifs->ifs_fr_tcplastack = IPF_TTLVAL(TCP_MSL);
ifs->ifs_fr_tcptimeout = IPF_TTLVAL(TCP_MSL);
ifs->ifs_fr_tcpclosed = IPF_TTLVAL(60);
ifs->ifs_fr_tcphalfclosed = IPF_TTLVAL(2 * 3600); /* 2 hours */
ifs->ifs_fr_udptimeout = IPF_TTLVAL(120);
ifs->ifs_fr_udpacktimeout = IPF_TTLVAL(12);
ifs->ifs_fr_icmptimeout = IPF_TTLVAL(60);
ifs->ifs_fr_icmpacktimeout = IPF_TTLVAL(6);
ifs->ifs_fr_iptimeout = IPF_TTLVAL(60);
ifs->ifs_fr_statemax = IPSTATE_MAX;
ifs->ifs_fr_statesize = IPSTATE_SIZE;
ifs->ifs_fr_state_maxbucket_reset = 1;
#ifdef IPFILTER_LOG
ifs->ifs_ipstate_logging = 1;
#else
ifs->ifs_ipstate_logging = 0;
#endif
KMALLOCS(ifs->ifs_ips_table, ipstate_t **,
ifs->ifs_fr_statesize * sizeof(ipstate_t *));
if (ifs->ifs_ips_table == NULL)
return -1;
bzero((char *)ifs->ifs_ips_table,
ifs->ifs_fr_statesize * sizeof(ipstate_t *));
KMALLOCS(ifs->ifs_ips_seed, u_long *,
ifs->ifs_fr_statesize * sizeof(*ifs->ifs_ips_seed));
if (ifs->ifs_ips_seed == NULL)
return -2;
for (i = 0; i < ifs->ifs_fr_statesize; i++) {
/*
* XXX - ips_seed[X] should be a random number of sorts.
*/
#if (__FreeBSD_version >= 400000)
ifs->ifs_ips_seed[i] = arc4random();
#else
ifs->ifs_ips_seed[i] = ((u_long)ifs->ifs_ips_seed + i) *
ifs->ifs_fr_statesize;
ifs->ifs_ips_seed[i] ^= 0xa5a55a5a;
ifs->ifs_ips_seed[i] *= (u_long)ifs->ifs_ips_seed;
ifs->ifs_ips_seed[i] ^= 0x5a5aa5a5;
ifs->ifs_ips_seed[i] *= ifs->ifs_fr_statemax;
#endif
}
/* fill icmp reply type table */
for (i = 0; i <= ICMP_MAXTYPE; i++)
icmpreplytype4[i] = -1;
icmpreplytype4[ICMP_ECHO] = ICMP_ECHOREPLY;
icmpreplytype4[ICMP_TSTAMP] = ICMP_TSTAMPREPLY;
icmpreplytype4[ICMP_IREQ] = ICMP_IREQREPLY;
icmpreplytype4[ICMP_MASKREQ] = ICMP_MASKREPLY;
#ifdef USE_INET6
/* fill icmp reply type table */
for (i = 0; i <= ICMP6_MAXTYPE; i++)
icmpreplytype6[i] = -1;
icmpreplytype6[ICMP6_ECHO_REQUEST] = ICMP6_ECHO_REPLY;
icmpreplytype6[ICMP6_MEMBERSHIP_QUERY] = ICMP6_MEMBERSHIP_REPORT;
icmpreplytype6[ICMP6_NI_QUERY] = ICMP6_NI_REPLY;
icmpreplytype6[ND_ROUTER_SOLICIT] = ND_ROUTER_ADVERT;
icmpreplytype6[ND_NEIGHBOR_SOLICIT] = ND_NEIGHBOR_ADVERT;
#endif
KMALLOCS(ifs->ifs_ips_stats.iss_bucketlen, u_long *,
ifs->ifs_fr_statesize * sizeof(u_long));
if (ifs->ifs_ips_stats.iss_bucketlen == NULL)
return -1;
bzero((char *)ifs->ifs_ips_stats.iss_bucketlen,
ifs->ifs_fr_statesize * sizeof(u_long));
if (ifs->ifs_fr_state_maxbucket == 0) {
for (i = ifs->ifs_fr_statesize; i > 0; i >>= 1)
ifs->ifs_fr_state_maxbucket++;
ifs->ifs_fr_state_maxbucket *= 2;
}
fr_sttab_init(ifs->ifs_ips_tqtqb, ifs);
ifs->ifs_ips_tqtqb[IPF_TCP_NSTATES - 1].ifq_next = &ifs->ifs_ips_udptq;
ifs->ifs_ips_udptq.ifq_ttl = (u_long)ifs->ifs_fr_udptimeout;
ifs->ifs_ips_udptq.ifq_ref = 1;
ifs->ifs_ips_udptq.ifq_head = NULL;
ifs->ifs_ips_udptq.ifq_tail = &ifs->ifs_ips_udptq.ifq_head;
MUTEX_INIT(&ifs->ifs_ips_udptq.ifq_lock, "ipftq udp tab");
ifs->ifs_ips_udptq.ifq_next = &ifs->ifs_ips_udpacktq;
ifs->ifs_ips_udpacktq.ifq_ttl = (u_long)ifs->ifs_fr_udpacktimeout;
ifs->ifs_ips_udpacktq.ifq_ref = 1;
ifs->ifs_ips_udpacktq.ifq_head = NULL;
ifs->ifs_ips_udpacktq.ifq_tail = &ifs->ifs_ips_udpacktq.ifq_head;
MUTEX_INIT(&ifs->ifs_ips_udpacktq.ifq_lock, "ipftq udpack tab");
ifs->ifs_ips_udpacktq.ifq_next = &ifs->ifs_ips_icmptq;
ifs->ifs_ips_icmptq.ifq_ttl = (u_long)ifs->ifs_fr_icmptimeout;
ifs->ifs_ips_icmptq.ifq_ref = 1;
ifs->ifs_ips_icmptq.ifq_head = NULL;
ifs->ifs_ips_icmptq.ifq_tail = &ifs->ifs_ips_icmptq.ifq_head;
MUTEX_INIT(&ifs->ifs_ips_icmptq.ifq_lock, "ipftq icmp tab");
ifs->ifs_ips_icmptq.ifq_next = &ifs->ifs_ips_icmpacktq;
ifs->ifs_ips_icmpacktq.ifq_ttl = (u_long)ifs->ifs_fr_icmpacktimeout;
ifs->ifs_ips_icmpacktq.ifq_ref = 1;
ifs->ifs_ips_icmpacktq.ifq_head = NULL;
ifs->ifs_ips_icmpacktq.ifq_tail = &ifs->ifs_ips_icmpacktq.ifq_head;
MUTEX_INIT(&ifs->ifs_ips_icmpacktq.ifq_lock, "ipftq icmpack tab");
ifs->ifs_ips_icmpacktq.ifq_next = &ifs->ifs_ips_iptq;
ifs->ifs_ips_iptq.ifq_ttl = (u_long)ifs->ifs_fr_iptimeout;
ifs->ifs_ips_iptq.ifq_ref = 1;
ifs->ifs_ips_iptq.ifq_head = NULL;
ifs->ifs_ips_iptq.ifq_tail = &ifs->ifs_ips_iptq.ifq_head;
MUTEX_INIT(&ifs->ifs_ips_iptq.ifq_lock, "ipftq ip tab");
ifs->ifs_ips_iptq.ifq_next = NULL;
RWLOCK_INIT(&ifs->ifs_ipf_state, "ipf IP state rwlock");
MUTEX_INIT(&ifs->ifs_ipf_stinsert, "ipf state insert mutex");
ifs->ifs_fr_state_init = 1;
ifs->ifs_ips_last_force_flush = ifs->ifs_fr_ticks;
return 0;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_stateunload */
/* Returns: Nil */
/* Parameters: Nil */
/* */
/* Release and destroy any resources acquired or initialised so that */
/* IPFilter can be unloaded or re-initialised. */
/* ------------------------------------------------------------------------ */
void fr_stateunload(ifs)
ipf_stack_t *ifs;
{
ipftq_t *ifq, *ifqnext;
ipstate_t *is;
while ((is = ifs->ifs_ips_list) != NULL)
fr_delstate(is, 0, ifs);
/*
* Proxy timeout queues are not cleaned here because although they
* exist on the state list, appr_unload is called after fr_stateunload
* and the proxies actually are responsible for them being created.
* Should the proxy timeouts have their own list? There's no real
* justification as this is the only complicationA
*/
for (ifq = ifs->ifs_ips_utqe; ifq != NULL; ifq = ifqnext) {
ifqnext = ifq->ifq_next;
if (((ifq->ifq_flags & IFQF_PROXY) == 0) &&
(fr_deletetimeoutqueue(ifq) == 0))
fr_freetimeoutqueue(ifq, ifs);
}
ifs->ifs_ips_stats.iss_inuse = 0;
ifs->ifs_ips_num = 0;
if (ifs->ifs_fr_state_init == 1) {
fr_sttab_destroy(ifs->ifs_ips_tqtqb);
MUTEX_DESTROY(&ifs->ifs_ips_udptq.ifq_lock);
MUTEX_DESTROY(&ifs->ifs_ips_icmptq.ifq_lock);
MUTEX_DESTROY(&ifs->ifs_ips_udpacktq.ifq_lock);
MUTEX_DESTROY(&ifs->ifs_ips_icmpacktq.ifq_lock);
MUTEX_DESTROY(&ifs->ifs_ips_iptq.ifq_lock);
}
if (ifs->ifs_ips_table != NULL) {
KFREES(ifs->ifs_ips_table,
ifs->ifs_fr_statesize * sizeof(*ifs->ifs_ips_table));
ifs->ifs_ips_table = NULL;
}
if (ifs->ifs_ips_seed != NULL) {
KFREES(ifs->ifs_ips_seed,
ifs->ifs_fr_statesize * sizeof(*ifs->ifs_ips_seed));
ifs->ifs_ips_seed = NULL;
}
if (ifs->ifs_ips_stats.iss_bucketlen != NULL) {
KFREES(ifs->ifs_ips_stats.iss_bucketlen,
ifs->ifs_fr_statesize * sizeof(u_long));
ifs->ifs_ips_stats.iss_bucketlen = NULL;
}
if (ifs->ifs_fr_state_maxbucket_reset == 1)
ifs->ifs_fr_state_maxbucket = 0;
if (ifs->ifs_fr_state_init == 1) {
ifs->ifs_fr_state_init = 0;
RW_DESTROY(&ifs->ifs_ipf_state);
MUTEX_DESTROY(&ifs->ifs_ipf_stinsert);
}
}
/* ------------------------------------------------------------------------ */
/* Function: fr_statetstats */
/* Returns: ips_state_t* - pointer to state stats structure */
/* Parameters: Nil */
/* */
/* Put all the current numbers and pointers into a single struct and return */
/* a pointer to it. */
/* ------------------------------------------------------------------------ */
static ips_stat_t *fr_statetstats(ifs)
ipf_stack_t *ifs;
{
ifs->ifs_ips_stats.iss_active = ifs->ifs_ips_num;
ifs->ifs_ips_stats.iss_statesize = ifs->ifs_fr_statesize;
ifs->ifs_ips_stats.iss_statemax = ifs->ifs_fr_statemax;
ifs->ifs_ips_stats.iss_table = ifs->ifs_ips_table;
ifs->ifs_ips_stats.iss_list = ifs->ifs_ips_list;
ifs->ifs_ips_stats.iss_ticks = ifs->ifs_fr_ticks;
return &ifs->ifs_ips_stats;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_state_remove */
/* Returns: int - 0 == success, != 0 == failure */
/* Parameters: data(I) - pointer to state structure to delete from table */
/* */
/* Search for a state structure that matches the one passed, according to */
/* the IP addresses and other protocol specific information. */
/* ------------------------------------------------------------------------ */
static int fr_state_remove(data, ifs)
caddr_t data;
ipf_stack_t *ifs;
{
ipstate_t *sp, st;
int error;
sp = &st;
error = fr_inobj(data, &st, IPFOBJ_IPSTATE);
if (error)
return EFAULT;
WRITE_ENTER(&ifs->ifs_ipf_state);
for (sp = ifs->ifs_ips_list; sp; sp = sp->is_next)
if ((sp->is_p == st.is_p) && (sp->is_v == st.is_v) &&
!bcmp((caddr_t)&sp->is_src, (caddr_t)&st.is_src,
sizeof(st.is_src)) &&
!bcmp((caddr_t)&sp->is_dst, (caddr_t)&st.is_src,
sizeof(st.is_dst)) &&
!bcmp((caddr_t)&sp->is_ps, (caddr_t)&st.is_ps,
sizeof(st.is_ps))) {
fr_delstate(sp, ISL_REMOVE, ifs);
RWLOCK_EXIT(&ifs->ifs_ipf_state);
return 0;
}
RWLOCK_EXIT(&ifs->ifs_ipf_state);
return ESRCH;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_state_ioctl */
/* Returns: int - 0 == success, != 0 == failure */
/* Parameters: data(I) - pointer to ioctl data */
/* cmd(I) - ioctl command integer */
/* mode(I) - file mode bits used with open */
/* */
/* Processes an ioctl call made to operate on the IP Filter state device. */
/* ------------------------------------------------------------------------ */
int fr_state_ioctl(data, cmd, mode, uid, ctx, ifs)
caddr_t data;
ioctlcmd_t cmd;
int mode, uid;
void *ctx;
ipf_stack_t *ifs;
{
int arg, ret, error = 0;
switch (cmd)
{
/*
* Delete an entry from the state table.
*/
case SIOCDELST :
error = fr_state_remove(data, ifs);
break;
/*
* Flush the state table
*/
case SIOCIPFFL :
BCOPYIN(data, (char *)&arg, sizeof(arg));
if (arg == 0 || arg == 1) {
WRITE_ENTER(&ifs->ifs_ipf_state);
ret = fr_state_flush(arg, 4, ifs);
RWLOCK_EXIT(&ifs->ifs_ipf_state);
BCOPYOUT((char *)&ret, data, sizeof(ret));
} else
error = EINVAL;
break;
#ifdef USE_INET6
case SIOCIPFL6 :
BCOPYIN(data, (char *)&arg, sizeof(arg));
if (arg == 0 || arg == 1) {
WRITE_ENTER(&ifs->ifs_ipf_state);
ret = fr_state_flush(arg, 6, ifs);
RWLOCK_EXIT(&ifs->ifs_ipf_state);
BCOPYOUT((char *)&ret, data, sizeof(ret));
} else
error = EINVAL;
break;
#endif
#ifdef IPFILTER_LOG
/*
* Flush the state log.
*/
case SIOCIPFFB :
if (!(mode & FWRITE))
error = EPERM;
else {
int tmp;
tmp = ipflog_clear(IPL_LOGSTATE, ifs);
BCOPYOUT((char *)&tmp, data, sizeof(tmp));
}
break;
/*
* Turn logging of state information on/off.
*/
case SIOCSETLG :
if (!(mode & FWRITE))
error = EPERM;
else {
BCOPYIN((char *)data,
(char *)&ifs->ifs_ipstate_logging,
sizeof(ifs->ifs_ipstate_logging));
}
break;
/*
* Return the current state of logging.
*/
case SIOCGETLG :
BCOPYOUT((char *)&ifs->ifs_ipstate_logging, (char *)data,
sizeof(ifs->ifs_ipstate_logging));
break;
/*
* Return the number of bytes currently waiting to be read.
*/
case FIONREAD :
arg = ifs->ifs_iplused[IPL_LOGSTATE]; /* returned in an int */
BCOPYOUT((char *)&arg, data, sizeof(arg));
break;
#endif
/*
* Get the current state statistics.
*/
case SIOCGETFS :
error = fr_outobj(data, fr_statetstats(ifs), IPFOBJ_STATESTAT);
break;
/*
* Lock/Unlock the state table. (Locking prevents any changes, which
* means no packets match).
*/
case SIOCSTLCK :
if (!(mode & FWRITE)) {
error = EPERM;
} else {
fr_lock(data, &ifs->ifs_fr_state_lock);
}
break;
/*
* Add an entry to the current state table.
*/
case SIOCSTPUT :
if (!ifs->ifs_fr_state_lock || !(mode &FWRITE)) {
error = EACCES;
break;
}
error = fr_stputent(data, ifs);
break;
/*
* Get a state table entry.
*/
case SIOCSTGET :
if (!ifs->ifs_fr_state_lock) {
error = EACCES;
break;
}
error = fr_stgetent(data, ifs);
break;
case SIOCGENITER :
{
ipftoken_t *token;
ipfgeniter_t iter;
error = fr_inobj(data, &iter, IPFOBJ_GENITER);
if (error != 0)
break;
token = ipf_findtoken(IPFGENITER_STATE, uid, ctx, ifs);
if (token != NULL)
error = fr_stateiter(token, &iter, ifs);
else
error = ESRCH;
RWLOCK_EXIT(&ifs->ifs_ipf_tokens);
break;
}
case SIOCIPFDELTOK :
(void) BCOPYIN(data, (char *)&arg, sizeof(arg));
error = ipf_deltoken(arg, uid, ctx, ifs);
break;
default :
error = EINVAL;
break;
}
return error;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_stgetent */
/* Returns: int - 0 == success, != 0 == failure */
/* Parameters: data(I) - pointer to state structure to retrieve from table */
/* */
/* Copy out state information from the kernel to a user space process. If */
/* there is a filter rule associated with the state entry, copy that out */
/* as well. The entry to copy out is taken from the value of "ips_next" in */
/* the struct passed in and if not null and not found in the list of current*/
/* state entries, the retrieval fails. */
/* ------------------------------------------------------------------------ */
int fr_stgetent(data, ifs)
caddr_t data;
ipf_stack_t *ifs;
{
ipstate_t *is, *isn;
ipstate_save_t ips;
int error;
error = fr_inobj(data, &ips, IPFOBJ_STATESAVE);
if (error)
return EFAULT;
isn = ips.ips_next;
if (isn == NULL) {
isn = ifs->ifs_ips_list;
if (isn == NULL) {
if (ips.ips_next == NULL)
return ENOENT;
return 0;
}
} else {
/*
* Make sure the pointer we're copying from exists in the
* current list of entries. Security precaution to prevent
* copying of random kernel data.
*/
for (is = ifs->ifs_ips_list; is; is = is->is_next)
if (is == isn)
break;
if (!is)
return ESRCH;
}
ips.ips_next = isn->is_next;
bcopy((char *)isn, (char *)&ips.ips_is, sizeof(ips.ips_is));
ips.ips_rule = isn->is_rule;
if (isn->is_rule != NULL)
bcopy((char *)isn->is_rule, (char *)&ips.ips_fr,
sizeof(ips.ips_fr));
error = fr_outobj(data, &ips, IPFOBJ_STATESAVE);
if (error)
return EFAULT;
return 0;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_stputent */
/* Returns: int - 0 == success, != 0 == failure */
/* Parameters: data(I) - pointer to state information struct */
/* */
/* This function implements the SIOCSTPUT ioctl: insert a state entry into */
/* the state table. If the state info. includes a pointer to a filter rule */
/* then also add in an orphaned rule (will not show up in any "ipfstat -io" */
/* output. */
/* ------------------------------------------------------------------------ */
int fr_stputent(data, ifs)
caddr_t data;
ipf_stack_t *ifs;
{
ipstate_t *is, *isn;
ipstate_save_t ips;
int error, i;
frentry_t *fr;
char *name;
error = fr_inobj(data, &ips, IPFOBJ_STATESAVE);
if (error)
return EFAULT;
KMALLOC(isn, ipstate_t *);
if (isn == NULL)
return ENOMEM;
bcopy((char *)&ips.ips_is, (char *)isn, sizeof(*isn));
bzero((char *)isn, offsetof(struct ipstate, is_pkts));
isn->is_sti.tqe_pnext = NULL;
isn->is_sti.tqe_next = NULL;
isn->is_sti.tqe_ifq = NULL;
isn->is_sti.tqe_parent = isn;
isn->is_ifp[0] = NULL;
isn->is_ifp[1] = NULL;
isn->is_ifp[2] = NULL;
isn->is_ifp[3] = NULL;
isn->is_sync = NULL;
fr = ips.ips_rule;
if (fr == NULL) {
READ_ENTER(&ifs->ifs_ipf_state);
fr_stinsert(isn, 0, ifs);
MUTEX_EXIT(&isn->is_lock);
RWLOCK_EXIT(&ifs->ifs_ipf_state);
return 0;
}
if (isn->is_flags & SI_NEWFR) {
KMALLOC(fr, frentry_t *);
if (fr == NULL) {
KFREE(isn);
return ENOMEM;
}
bcopy((char *)&ips.ips_fr, (char *)fr, sizeof(*fr));
isn->is_rule = fr;
ips.ips_is.is_rule = fr;
MUTEX_NUKE(&fr->fr_lock);
MUTEX_INIT(&fr->fr_lock, "state filter rule lock");
/*
* Look up all the interface names in the rule.
*/
for (i = 0; i < 4; i++) {
name = fr->fr_ifnames[i];
fr->fr_ifas[i] = fr_resolvenic(name, fr->fr_v, ifs);
name = isn->is_ifname[i];
isn->is_ifp[i] = fr_resolvenic(name, isn->is_v, ifs);
}
fr->fr_ref = 0;
fr->fr_dsize = 0;
fr->fr_data = NULL;
fr_resolvedest(&fr->fr_tif, fr->fr_v, ifs);
fr_resolvedest(&fr->fr_dif, fr->fr_v, ifs);
/*
* send a copy back to userland of what we ended up
* to allow for verification.
*/
error = fr_outobj(data, &ips, IPFOBJ_STATESAVE);
if (error) {
KFREE(isn);
MUTEX_DESTROY(&fr->fr_lock);
KFREE(fr);
return EFAULT;
}
READ_ENTER(&ifs->ifs_ipf_state);
fr_stinsert(isn, 0, ifs);
MUTEX_EXIT(&isn->is_lock);
RWLOCK_EXIT(&ifs->ifs_ipf_state);
} else {
READ_ENTER(&ifs->ifs_ipf_state);
for (is = ifs->ifs_ips_list; is; is = is->is_next)
if (is->is_rule == fr) {
fr_stinsert(isn, 0, ifs);
MUTEX_EXIT(&isn->is_lock);
break;
}
if (is == NULL) {
KFREE(isn);
isn = NULL;
}
RWLOCK_EXIT(&ifs->ifs_ipf_state);
return (isn == NULL) ? ESRCH : 0;
}
return 0;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_stinsert */
/* Returns: Nil */
/* Parameters: is(I) - pointer to state structure */
/* rev(I) - flag indicating forward/reverse direction of packet */
/* */
/* Inserts a state structure into the hash table (for lookups) and the list */
/* of state entries (for enumeration). Resolves all of the interface names */
/* to pointers and adjusts running stats for the hash table as appropriate. */
/* */
/* Locking: it is assumed that some kind of lock on ipf_state is held. */
/* Exits with is_lock initialised and held. */
/* ------------------------------------------------------------------------ */
void fr_stinsert(is, rev, ifs)
ipstate_t *is;
int rev;
ipf_stack_t *ifs;
{
frentry_t *fr;
u_int hv;
int i;
MUTEX_INIT(&is->is_lock, "ipf state entry");
fr = is->is_rule;
if (fr != NULL) {
MUTEX_ENTER(&fr->fr_lock);
fr->fr_ref++;
fr->fr_statecnt++;
MUTEX_EXIT(&fr->fr_lock);
}
/*
* Look up all the interface names in the state entry.
*/
for (i = 0; i < 4; i++) {
if (is->is_ifp[i] != NULL)
continue;
is->is_ifp[i] = fr_resolvenic(is->is_ifname[i], is->is_v, ifs);
}
/*
* If we could trust is_hv, then the modulous would not be needed, but
* when running with IPFILTER_SYNC, this stops bad values.
*/
hv = is->is_hv % ifs->ifs_fr_statesize;
is->is_hv = hv;
/*
* We need to get both of these locks...the first because it is
* possible that once the insert is complete another packet might
* come along, match the entry and want to update it.
*/
MUTEX_ENTER(&is->is_lock);
MUTEX_ENTER(&ifs->ifs_ipf_stinsert);
/*
* add into list table.
*/
if (ifs->ifs_ips_list != NULL)
ifs->ifs_ips_list->is_pnext = &is->is_next;
is->is_pnext = &ifs->ifs_ips_list;
is->is_next = ifs->ifs_ips_list;
ifs->ifs_ips_list = is;
if (ifs->ifs_ips_table[hv] != NULL)
ifs->ifs_ips_table[hv]->is_phnext = &is->is_hnext;
else
ifs->ifs_ips_stats.iss_inuse++;
is->is_phnext = ifs->ifs_ips_table + hv;
is->is_hnext = ifs->ifs_ips_table[hv];
ifs->ifs_ips_table[hv] = is;
ifs->ifs_ips_stats.iss_bucketlen[hv]++;
ifs->ifs_ips_num++;
MUTEX_EXIT(&ifs->ifs_ipf_stinsert);
fr_setstatequeue(is, rev, ifs);
}
/* ------------------------------------------------------------------------ */
/* Function: fr_addstate */
/* Returns: ipstate_t* - NULL == failure, else pointer to new state */
/* Parameters: fin(I) - pointer to packet information */
/* stsave(O) - pointer to place to save pointer to created */
/* state structure. */
/* flags(I) - flags to use when creating the structure */
/* */
/* Creates a new IP state structure from the packet information collected. */
/* Inserts it into the state table and appends to the bottom of the active */
/* list. If the capacity of the table has reached the maximum allowed then */
/* the call will fail and a flush is scheduled for the next timeout call. */
/* ------------------------------------------------------------------------ */
ipstate_t *fr_addstate(fin, stsave, flags)
fr_info_t *fin;
ipstate_t **stsave;
u_int flags;
{
ipstate_t *is, ips;
struct icmp *ic;
u_int pass, hv;
frentry_t *fr;
tcphdr_t *tcp;
grehdr_t *gre;
void *ifp;
int out;
ipf_stack_t *ifs = fin->fin_ifs;
if (ifs->ifs_fr_state_lock ||
(fin->fin_flx & (FI_SHORT|FI_STATE|FI_FRAGBODY|FI_BAD)))
return NULL;
if ((fin->fin_flx & FI_OOW) && !(fin->fin_tcpf & TH_SYN))
return NULL;
/*
* If a "keep state" rule has reached the maximum number of references
* to it, then schedule an automatic flush in case we can clear out
* some "dead old wood". Note that because the lock isn't held on
* fr it is possible that we could overflow. The cost of overflowing
* is being ignored here as the number by which it can overflow is
* a product of the number of simultaneous threads that could be
* executing in here, so a limit of 100 won't result in 200, but could
* result in 101 or 102.
*/
fr = fin->fin_fr;
if (fr != NULL) {
if ((ifs->ifs_ips_num == ifs->ifs_fr_statemax) && (fr->fr_statemax == 0)) {
ATOMIC_INCL(ifs->ifs_ips_stats.iss_max);
ifs->ifs_fr_state_doflush = 1;
return NULL;
}
if ((fr->fr_statemax != 0) &&
(fr->fr_statecnt >= fr->fr_statemax)) {
ATOMIC_INCL(ifs->ifs_ips_stats.iss_maxref);
ifs->ifs_fr_state_doflush = 1;
return NULL;
}
}
pass = (fr == NULL) ? 0 : fr->fr_flags;
ic = NULL;
tcp = NULL;
out = fin->fin_out;
is = &ips;
bzero((char *)is, sizeof(*is));
is->is_die = 1 + ifs->ifs_fr_ticks;
/*
* Copy and calculate...
*/
hv = (is->is_p = fin->fin_fi.fi_p);
is->is_src = fin->fin_fi.fi_src;
hv += is->is_saddr;
is->is_dst = fin->fin_fi.fi_dst;
hv += is->is_daddr;
#ifdef USE_INET6
if (fin->fin_v == 6) {
/*
* For ICMPv6, we check to see if the destination address is
* a multicast address. If it is, do not include it in the
* calculation of the hash because the correct reply will come
* back from a real address, not a multicast address.
*/
if ((is->is_p == IPPROTO_ICMPV6) &&
IN6_IS_ADDR_MULTICAST(&is->is_dst.in6)) {
/*
* So you can do keep state with neighbour discovery.
*
* Here we could use the address from the neighbour
* solicit message to put in the state structure and
* we could use that without a wildcard flag too...
*/
is->is_flags |= SI_W_DADDR;
hv -= is->is_daddr;
} else {
hv += is->is_dst.i6[1];
hv += is->is_dst.i6[2];
hv += is->is_dst.i6[3];
}
hv += is->is_src.i6[1];
hv += is->is_src.i6[2];
hv += is->is_src.i6[3];
}
#endif
switch (is->is_p)
{
#ifdef USE_INET6
case IPPROTO_ICMPV6 :
ic = fin->fin_dp;
switch (ic->icmp_type)
{
case ICMP6_ECHO_REQUEST :
is->is_icmp.ici_type = ic->icmp_type;
hv += (is->is_icmp.ici_id = ic->icmp_id);
break;
case ICMP6_MEMBERSHIP_QUERY :
case ND_ROUTER_SOLICIT :
case ND_NEIGHBOR_SOLICIT :
case ICMP6_NI_QUERY :
is->is_icmp.ici_type = ic->icmp_type;
break;
default :
return NULL;
}
ATOMIC_INCL(ifs->ifs_ips_stats.iss_icmp);
break;
#endif
case IPPROTO_ICMP :
ic = fin->fin_dp;
switch (ic->icmp_type)
{
case ICMP_ECHO :
case ICMP_TSTAMP :
case ICMP_IREQ :
case ICMP_MASKREQ :
is->is_icmp.ici_type = ic->icmp_type;
hv += (is->is_icmp.ici_id = ic->icmp_id);
break;
default :
return NULL;
}
ATOMIC_INCL(ifs->ifs_ips_stats.iss_icmp);
break;
case IPPROTO_GRE :
gre = fin->fin_dp;
is->is_gre.gs_flags = gre->gr_flags;
is->is_gre.gs_ptype = gre->gr_ptype;
if (GRE_REV(is->is_gre.gs_flags) == 1) {
is->is_call[0] = fin->fin_data[0];
is->is_call[1] = fin->fin_data[1];
}
break;
case IPPROTO_TCP :
tcp = fin->fin_dp;
if (tcp->th_flags & TH_RST)
return NULL;
/*
* The endian of the ports doesn't matter, but the ack and
* sequence numbers do as we do mathematics on them later.
*/
is->is_sport = htons(fin->fin_data[0]);
is->is_dport = htons(fin->fin_data[1]);
if ((flags & (SI_W_DPORT|SI_W_SPORT)) == 0) {
hv += is->is_sport;
hv += is->is_dport;
}
/*
* If this is a real packet then initialise fields in the
* state information structure from the TCP header information.
*/
is->is_maxdwin = 1;
is->is_maxswin = ntohs(tcp->th_win);
if (is->is_maxswin == 0)
is->is_maxswin = 1;
if ((fin->fin_flx & FI_IGNORE) == 0) {
is->is_send = ntohl(tcp->th_seq) + fin->fin_dlen -
(TCP_OFF(tcp) << 2) +
((tcp->th_flags & TH_SYN) ? 1 : 0) +
((tcp->th_flags & TH_FIN) ? 1 : 0);
is->is_maxsend = is->is_send;
/*
* Window scale option is only present in
* SYN/SYN-ACK packet.
*/
if ((tcp->th_flags & ~(TH_FIN|TH_ACK|TH_ECNALL)) ==
TH_SYN &&
(TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) {
if (fr_tcpoptions(fin, tcp,
&is->is_tcp.ts_data[0]))
is->is_swinflags = TCP_WSCALE_SEEN|
TCP_WSCALE_FIRST;
}
if ((fin->fin_out != 0) && (pass & FR_NEWISN) != 0) {
fr_checknewisn(fin, is);
fr_fixoutisn(fin, is);
}
if ((tcp->th_flags & TH_OPENING) == TH_SYN)
flags |= IS_TCPFSM;
else {
is->is_maxdwin = is->is_maxswin * 2;
is->is_dend = ntohl(tcp->th_ack);
is->is_maxdend = ntohl(tcp->th_ack);
is->is_maxdwin *= 2;
}
}
/*
* If we're creating state for a starting connection, start the
* timer on it as we'll never see an error if it fails to
* connect.
*/
ATOMIC_INCL(ifs->ifs_ips_stats.iss_tcp);
break;
case IPPROTO_UDP :
tcp = fin->fin_dp;
is->is_sport = htons(fin->fin_data[0]);
is->is_dport = htons(fin->fin_data[1]);
if ((flags & (SI_W_DPORT|SI_W_SPORT)) == 0) {
hv += tcp->th_dport;
hv += tcp->th_sport;
}
ATOMIC_INCL(ifs->ifs_ips_stats.iss_udp);
break;
default :
break;
}
hv = DOUBLE_HASH(hv, ifs);
is->is_hv = hv;
is->is_rule = fr;
is->is_flags = flags & IS_INHERITED;
/*
* Look for identical state.
*/
for (is = ifs->ifs_ips_table[is->is_hv % ifs->ifs_fr_statesize];
is != NULL;
is = is->is_hnext) {
if (bcmp(&ips.is_src, &is->is_src,
offsetof(struct ipstate, is_ps) -
offsetof(struct ipstate, is_src)) == 0)
break;
}
if (is != NULL)
return NULL;
if (ifs->ifs_ips_stats.iss_bucketlen[hv] >= ifs->ifs_fr_state_maxbucket) {
ATOMIC_INCL(ifs->ifs_ips_stats.iss_bucketfull);
return NULL;
}
KMALLOC(is, ipstate_t *);
if (is == NULL) {
ATOMIC_INCL(ifs->ifs_ips_stats.iss_nomem);
return NULL;
}
bcopy((char *)&ips, (char *)is, sizeof(*is));
/*
* Do not do the modulous here, it is done in fr_stinsert().
*/
if (fr != NULL) {
(void) strncpy(is->is_group, fr->fr_group, FR_GROUPLEN);
if (fr->fr_age[0] != 0) {
is->is_tqehead[0] =
fr_addtimeoutqueue(&ifs->ifs_ips_utqe,
fr->fr_age[0], ifs);
is->is_sti.tqe_flags |= TQE_RULEBASED;
}
if (fr->fr_age[1] != 0) {
is->is_tqehead[1] =
fr_addtimeoutqueue(&ifs->ifs_ips_utqe,
fr->fr_age[1], ifs);
is->is_sti.tqe_flags |= TQE_RULEBASED;
}
is->is_tag = fr->fr_logtag;
is->is_ifp[(out << 1) + 1] = fr->fr_ifas[1];
is->is_ifp[(1 - out) << 1] = fr->fr_ifas[2];
is->is_ifp[((1 - out) << 1) + 1] = fr->fr_ifas[3];
if (((ifp = fr->fr_ifas[1]) != NULL) &&
(ifp != (void *)-1)) {
COPYIFNAME(ifp, is->is_ifname[(out << 1) + 1], fr->fr_v);
}
if (((ifp = fr->fr_ifas[2]) != NULL) &&
(ifp != (void *)-1)) {
COPYIFNAME(ifp, is->is_ifname[(1 - out) << 1], fr->fr_v);
}
if (((ifp = fr->fr_ifas[3]) != NULL) &&
(ifp != (void *)-1)) {
COPYIFNAME(ifp, is->is_ifname[((1 - out) << 1) + 1], fr->fr_v);
}
} else {
pass = ifs->ifs_fr_flags;
is->is_tag = FR_NOLOGTAG;
}
is->is_ifp[out << 1] = fin->fin_ifp;
if (fin->fin_ifp != NULL) {
COPYIFNAME(fin->fin_ifp, is->is_ifname[out << 1], fr->fr_v);
}
/*
* It may seem strange to set is_ref to 2, but fr_check() will call
* fr_statederef() after calling fr_addstate() and the idea is to
* have it exist at the end of fr_check() with is_ref == 1.
*/
is->is_ref = 2;
is->is_pass = pass;
is->is_pkts[0] = 0, is->is_bytes[0] = 0;
is->is_pkts[1] = 0, is->is_bytes[1] = 0;
is->is_pkts[2] = 0, is->is_bytes[2] = 0;
is->is_pkts[3] = 0, is->is_bytes[3] = 0;
if ((fin->fin_flx & FI_IGNORE) == 0) {
is->is_pkts[out] = 1;
is->is_bytes[out] = fin->fin_plen;
is->is_flx[out][0] = fin->fin_flx & FI_CMP;
is->is_flx[out][0] &= ~FI_OOW;
}
if (pass & FR_STSTRICT)
is->is_flags |= IS_STRICT;
if (pass & FR_STATESYNC)
is->is_flags |= IS_STATESYNC;
/*
* We want to check everything that is a property of this packet,
* but we don't (automatically) care about it's fragment status as
* this may change.
*/
is->is_v = fin->fin_v;
is->is_opt[0] = fin->fin_optmsk;
is->is_optmsk[0] = 0xffffffff;
is->is_optmsk[1] = 0xffffffff;
if (is->is_v == 6) {
is->is_opt[0] &= ~0x8;
is->is_optmsk[0] &= ~0x8;
is->is_optmsk[1] &= ~0x8;
}
is->is_sec = fin->fin_secmsk;
is->is_secmsk = 0xffff;
is->is_auth = fin->fin_auth;
is->is_authmsk = 0xffff;
if (flags & (SI_WILDP|SI_WILDA)) {
ATOMIC_INCL(ifs->ifs_ips_stats.iss_wild);
}
is->is_rulen = fin->fin_rule;
if (pass & FR_LOGFIRST)
is->is_pass &= ~(FR_LOGFIRST|FR_LOG);
READ_ENTER(&ifs->ifs_ipf_state);
is->is_me = stsave;
fr_stinsert(is, fin->fin_rev, ifs);
if (fin->fin_p == IPPROTO_TCP) {
/*
* If we're creating state for a starting connection, start the
* timer on it as we'll never see an error if it fails to
* connect.
*/
(void) fr_tcp_age(&is->is_sti, fin, ifs->ifs_ips_tqtqb,
is->is_flags);
MUTEX_EXIT(&is->is_lock);
#ifdef IPFILTER_SCAN
if ((is->is_flags & SI_CLONE) == 0)
(void) ipsc_attachis(is);
#endif
} else {
MUTEX_EXIT(&is->is_lock);
}
#ifdef IPFILTER_SYNC
if ((is->is_flags & IS_STATESYNC) && ((is->is_flags & SI_CLONE) == 0))
is->is_sync = ipfsync_new(SMC_STATE, fin, is);
#endif
if (ifs->ifs_ipstate_logging)
ipstate_log(is, ISL_NEW, ifs);
RWLOCK_EXIT(&ifs->ifs_ipf_state);
fin->fin_state = is;
fin->fin_rev = IP6_NEQ(&is->is_dst, &fin->fin_daddr);
fin->fin_flx |= FI_STATE;
if (fin->fin_flx & FI_FRAG)
(void) fr_newfrag(fin, pass ^ FR_KEEPSTATE);
return is;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_tcpoptions */
/* Returns: int - 1 == packet matches state entry, 0 == it does not */
/* Parameters: fin(I) - pointer to packet information */
/* tcp(I) - pointer to TCP packet header */
/* td(I) - pointer to TCP data held as part of the state */
/* */
/* Look after the TCP header for any options and deal with those that are */
/* present. Record details about those that we recogise. */
/* ------------------------------------------------------------------------ */
static int fr_tcpoptions(fin, tcp, td)
fr_info_t *fin;
tcphdr_t *tcp;
tcpdata_t *td;
{
int off, mlen, ol, i, len, retval;
char buf[64], *s, opt;
mb_t *m = NULL;
len = (TCP_OFF(tcp) << 2);
if (fin->fin_dlen < len)
return 0;
len -= sizeof(*tcp);
off = fin->fin_plen - fin->fin_dlen + sizeof(*tcp) + fin->fin_ipoff;
m = fin->fin_m;
mlen = MSGDSIZE(m) - off;
if (len > mlen) {
len = mlen;
retval = 0;
} else {
retval = 1;
}
COPYDATA(m, off, len, buf);
for (s = buf; len > 0; ) {
opt = *s;
if (opt == TCPOPT_EOL)
break;
else if (opt == TCPOPT_NOP)
ol = 1;
else {
if (len < 2)
break;
ol = (int)*(s + 1);
if (ol < 2 || ol > len)
break;
/*
* Extract the TCP options we are interested in out of
* the header and store them in the the tcpdata struct.
*/
switch (opt)
{
case TCPOPT_WINDOW :
if (ol == TCPOLEN_WINDOW) {
i = (int)*(s + 2);
if (i > TCP_WSCALE_MAX)
i = TCP_WSCALE_MAX;
else if (i < 0)
i = 0;
td->td_winscale = i;
}
break;
case TCPOPT_MAXSEG :
/*
* So, if we wanted to set the TCP MAXSEG,
* it should be done here...
*/
if (ol == TCPOLEN_MAXSEG) {
i = (int)*(s + 2);
i <<= 8;
i += (int)*(s + 3);
td->td_maxseg = i;
}
break;
}
}
len -= ol;
s += ol;
}
return retval;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_tcpstate */
/* Returns: int - 1 == packet matches state entry, 0 == it does not */
/* Parameters: fin(I) - pointer to packet information */
/* tcp(I) - pointer to TCP packet header */
/* is(I) - pointer to master state structure */
/* */
/* Check to see if a packet with TCP headers fits within the TCP window. */
/* Change timeout depending on whether new packet is a SYN-ACK returning */
/* for a SYN or a RST or FIN which indicate time to close up shop. */
/* ------------------------------------------------------------------------ */
static int fr_tcpstate(fin, tcp, is)
fr_info_t *fin;
tcphdr_t *tcp;
ipstate_t *is;
{
int source, ret = 0, flags;
tcpdata_t *fdata, *tdata;
ipf_stack_t *ifs = fin->fin_ifs;
source = !fin->fin_rev;
if (((is->is_flags & IS_TCPFSM) != 0) && (source == 1) &&
(ntohs(is->is_sport) != fin->fin_data[0]))
source = 0;
fdata = &is->is_tcp.ts_data[!source];
tdata = &is->is_tcp.ts_data[source];
MUTEX_ENTER(&is->is_lock);
if (fr_tcpinwindow(fin, fdata, tdata, tcp, is->is_flags)) {
#ifdef IPFILTER_SCAN
if (is->is_flags & (IS_SC_CLIENT|IS_SC_SERVER)) {
ipsc_packet(fin, is);
if (FR_ISBLOCK(is->is_pass)) {
MUTEX_EXIT(&is->is_lock);
return 1;
}
}
#endif
/*
* Nearing end of connection, start timeout.
*/
ret = fr_tcp_age(&is->is_sti, fin, ifs->ifs_ips_tqtqb,
is->is_flags);
if (ret == 0) {
MUTEX_EXIT(&is->is_lock);
return 0;
}
/*
* set s0's as appropriate. Use syn-ack packet as it
* contains both pieces of required information.
*/
/*
* Window scale option is only present in SYN/SYN-ACK packet.
* Compare with ~TH_FIN to mask out T/TCP setups.
*/
flags = tcp->th_flags & ~(TH_FIN|TH_ECNALL);
if (flags == (TH_SYN|TH_ACK)) {
is->is_s0[source] = ntohl(tcp->th_ack);
is->is_s0[!source] = ntohl(tcp->th_seq) + 1;
if ((TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2)) &&
tdata->td_winscale) {
if (fr_tcpoptions(fin, tcp, fdata)) {
fdata->td_winflags = TCP_WSCALE_SEEN|
TCP_WSCALE_FIRST;
} else {
if (!fdata->td_winscale)
tdata->td_winscale = 0;
}
}
if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN))
fr_checknewisn(fin, is);
} else if (flags == TH_SYN) {
is->is_s0[source] = ntohl(tcp->th_seq) + 1;
if ((TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2)))
if (fr_tcpoptions(fin, tcp, tdata)) {
tdata->td_winflags = TCP_WSCALE_SEEN|
TCP_WSCALE_FIRST;
}
if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN))
fr_checknewisn(fin, is);
}
ret = 1;
} else
fin->fin_flx |= FI_OOW;
MUTEX_EXIT(&is->is_lock);
return ret;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_checknewisn */
/* Returns: Nil */
/* Parameters: fin(I) - pointer to packet information */
/* is(I) - pointer to master state structure */
/* */
/* Check to see if this TCP connection is expecting and needs a new */
/* sequence number for a particular direction of the connection. */
/* */
/* NOTE: This does not actually change the sequence numbers, only gets new */
/* one ready. */
/* ------------------------------------------------------------------------ */
static void fr_checknewisn(fin, is)
fr_info_t *fin;
ipstate_t *is;
{
u_32_t sumd, old, new;
tcphdr_t *tcp;
int i;
i = fin->fin_rev;
tcp = fin->fin_dp;
if (((i == 0) && !(is->is_flags & IS_ISNSYN)) ||
((i == 1) && !(is->is_flags & IS_ISNACK))) {
old = ntohl(tcp->th_seq);
new = fr_newisn(fin);
is->is_isninc[i] = new - old;
CALC_SUMD(old, new, sumd);
is->is_sumd[i] = (sumd & 0xffff) + (sumd >> 16);
is->is_flags |= ((i == 0) ? IS_ISNSYN : IS_ISNACK);
}
}
/* ------------------------------------------------------------------------ */
/* Function: fr_tcpinwindow */
/* Returns: int - 1 == packet inside TCP "window", 0 == not inside. */
/* Parameters: fin(I) - pointer to packet information */
/* fdata(I) - pointer to tcp state informatio (forward) */
/* tdata(I) - pointer to tcp state informatio (reverse) */
/* tcp(I) - pointer to TCP packet header */
/* */
/* Given a packet has matched addresses and ports, check to see if it is */
/* within the TCP data window. In a show of generosity, allow packets that */
/* are within the window space behind the current sequence # as well. */
/* ------------------------------------------------------------------------ */
int fr_tcpinwindow(fin, fdata, tdata, tcp, flags)
fr_info_t *fin;
tcpdata_t *fdata, *tdata;
tcphdr_t *tcp;
int flags;
{
tcp_seq seq, ack, end;
int ackskew, tcpflags;
u_32_t win, maxwin;
/*
* Find difference between last checked packet and this packet.
*/
tcpflags = tcp->th_flags;
seq = ntohl(tcp->th_seq);
ack = ntohl(tcp->th_ack);
if (tcpflags & TH_SYN)
win = ntohs(tcp->th_win);
else
win = ntohs(tcp->th_win) << fdata->td_winscale;
if (win == 0)
win = 1;
/*
* if window scaling is present, the scaling is only allowed
* for windows not in the first SYN packet. In that packet the
* window is 65535 to specify the largest window possible
* for receivers not implementing the window scale option.
* Currently, we do not assume TTCP here. That means that
* if we see a second packet from a host (after the initial
* SYN), we can assume that the receiver of the SYN did
* already send back the SYN/ACK (and thus that we know if
* the receiver also does window scaling)
*/
if (!(tcpflags & TH_SYN) && (fdata->td_winflags & TCP_WSCALE_FIRST)) {
if (tdata->td_winflags & TCP_WSCALE_SEEN) {
fdata->td_winflags &= ~TCP_WSCALE_FIRST;
fdata->td_maxwin = win;
} else {
fdata->td_winscale = 0;
fdata->td_winflags = 0;
tdata->td_winscale = 0;
tdata->td_winflags = 0;
}
}
end = seq + fin->fin_dlen - (TCP_OFF(tcp) << 2) +
((tcpflags & TH_SYN) ? 1 : 0) + ((tcpflags & TH_FIN) ? 1 : 0);
if ((fdata->td_end == 0) &&
(!(flags & IS_TCPFSM) ||
((tcpflags & TH_OPENING) == TH_OPENING))) {
/*
* Must be a (outgoing) SYN-ACK in reply to a SYN.
*/
fdata->td_end = end;
fdata->td_maxwin = 1;
fdata->td_maxend = end + win;
}
if (!(tcpflags & TH_ACK)) { /* Pretend an ack was sent */
ack = tdata->td_end;
} else if (((tcpflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) &&
(ack == 0)) {
/* gross hack to get around certain broken tcp stacks */
ack = tdata->td_end;
}
if (seq == end)
seq = end = fdata->td_end;
maxwin = tdata->td_maxwin;
ackskew = tdata->td_end - ack;
/*
* Strict sequencing only allows in-order delivery.
*/
if ((flags & IS_STRICT) != 0) {
if (seq != fdata->td_end) {
return 0;
}
}
#define SEQ_GE(a,b) ((int)((a) - (b)) >= 0)
#define SEQ_GT(a,b) ((int)((a) - (b)) > 0)
if (
#if defined(_KERNEL)
(SEQ_GE(fdata->td_maxend, end)) &&
(SEQ_GE(seq, fdata->td_end - maxwin)) &&
#endif
/* XXX what about big packets */
#define MAXACKWINDOW 66000
(-ackskew <= (MAXACKWINDOW << fdata->td_winscale)) &&
( ackskew <= (MAXACKWINDOW << fdata->td_winscale))) {
/* if ackskew < 0 then this should be due to fragmented
* packets. There is no way to know the length of the
* total packet in advance.
* We do know the total length from the fragment cache though.
* Note however that there might be more sessions with
* exactly the same source and destination parameters in the
* state cache (and source and destination is the only stuff
* that is saved in the fragment cache). Note further that
* some TCP connections in the state cache are hashed with
* sport and dport as well which makes it not worthwhile to
* look for them.
* Thus, when ackskew is negative but still seems to belong
* to this session, we bump up the destinations end value.
*/
if (ackskew < 0)
tdata->td_end = ack;
/* update max window seen */
if (fdata->td_maxwin < win)
fdata->td_maxwin = win;
if (SEQ_GT(end, fdata->td_end))
fdata->td_end = end;
if (SEQ_GE(ack + win, tdata->td_maxend))
tdata->td_maxend = ack + win;
return 1;
}
return 0;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_stclone */
/* Returns: ipstate_t* - NULL == cloning failed, */
/* else pointer to new state structure */
/* Parameters: fin(I) - pointer to packet information */
/* tcp(I) - pointer to TCP/UDP header */
/* is(I) - pointer to master state structure */
/* */
/* Create a "duplcate" state table entry from the master. */
/* ------------------------------------------------------------------------ */
static ipstate_t *fr_stclone(fin, tcp, is)
fr_info_t *fin;
tcphdr_t *tcp;
ipstate_t *is;
{
ipstate_t *clone;
u_32_t send;
ipf_stack_t *ifs = fin->fin_ifs;
if (ifs->ifs_ips_num == ifs->ifs_fr_statemax) {
ATOMIC_INCL(ifs->ifs_ips_stats.iss_max);
ifs->ifs_fr_state_doflush = 1;
return NULL;
}
KMALLOC(clone, ipstate_t *);
if (clone == NULL)
return NULL;
bcopy((char *)is, (char *)clone, sizeof(*clone));
MUTEX_NUKE(&clone->is_lock);
clone->is_die = ONE_DAY + ifs->ifs_fr_ticks;
clone->is_state[0] = 0;
clone->is_state[1] = 0;
send = ntohl(tcp->th_seq) + fin->fin_dlen - (TCP_OFF(tcp) << 2) +
((tcp->th_flags & TH_SYN) ? 1 : 0) +
((tcp->th_flags & TH_FIN) ? 1 : 0);
if (fin->fin_rev == 1) {
clone->is_dend = send;
clone->is_maxdend = send;
clone->is_send = 0;
clone->is_maxswin = 1;
clone->is_maxdwin = ntohs(tcp->th_win);
if (clone->is_maxdwin == 0)
clone->is_maxdwin = 1;
} else {
clone->is_send = send;
clone->is_maxsend = send;
clone->is_dend = 0;
clone->is_maxdwin = 1;
clone->is_maxswin = ntohs(tcp->th_win);
if (clone->is_maxswin == 0)
clone->is_maxswin = 1;
}
clone->is_flags &= ~SI_CLONE;
clone->is_flags |= SI_CLONED;
fr_stinsert(clone, fin->fin_rev, ifs);
clone->is_ref = 2;
if (clone->is_p == IPPROTO_TCP) {
(void) fr_tcp_age(&clone->is_sti, fin, ifs->ifs_ips_tqtqb,
clone->is_flags);
}
MUTEX_EXIT(&clone->is_lock);
#ifdef IPFILTER_SCAN
(void) ipsc_attachis(is);
#endif
#ifdef IPFILTER_SYNC
if (is->is_flags & IS_STATESYNC)
clone->is_sync = ipfsync_new(SMC_STATE, fin, clone);
#endif
return clone;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_matchsrcdst */
/* Returns: Nil */
/* Parameters: fin(I) - pointer to packet information */
/* is(I) - pointer to state structure */
/* src(I) - pointer to source address */
/* dst(I) - pointer to destination address */
/* tcp(I) - pointer to TCP/UDP header */
/* */
/* Match a state table entry against an IP packet. The logic below is that */
/* ret gets set to one if the match succeeds, else remains 0. If it is */
/* still 0 after the test. no match. */
/* ------------------------------------------------------------------------ */
static ipstate_t *fr_matchsrcdst(fin, is, src, dst, tcp, cmask)
fr_info_t *fin;
ipstate_t *is;
i6addr_t *src, *dst;
tcphdr_t *tcp;
u_32_t cmask;
{
int ret = 0, rev, out, flags, flx = 0, idx;
u_short sp, dp;
u_32_t cflx;
void *ifp;
ipf_stack_t *ifs = fin->fin_ifs;
rev = IP6_NEQ(&is->is_dst, dst);
ifp = fin->fin_ifp;
out = fin->fin_out;
flags = is->is_flags;
sp = 0;
dp = 0;
if (tcp != NULL) {
sp = htons(fin->fin_sport);
dp = ntohs(fin->fin_dport);
}
if (!rev) {
if (tcp != NULL) {
if (!(flags & SI_W_SPORT) && (sp != is->is_sport))
rev = 1;
else if (!(flags & SI_W_DPORT) && (dp != is->is_dport))
rev = 1;
}
}
idx = (out << 1) + rev;
/*
* If the interface for this 'direction' is set, make sure it matches.
* An interface name that is not set matches any, as does a name of *.
*/
if ((is->is_ifp[idx] == NULL &&
(*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) ||
is->is_ifp[idx] == ifp)
ret = 1;
if (ret == 0)
return NULL;
ret = 0;
/*
* Match addresses and ports.
*/
if (rev == 0) {
if ((IP6_EQ(&is->is_dst, dst) || (flags & SI_W_DADDR)) &&
(IP6_EQ(&is->is_src, src) || (flags & SI_W_SADDR))) {
if (tcp) {
if ((sp == is->is_sport || flags & SI_W_SPORT)&&
(dp == is->is_dport || flags & SI_W_DPORT))
ret = 1;
} else {
ret = 1;
}
}
} else {
if ((IP6_EQ(&is->is_dst, src) || (flags & SI_W_DADDR)) &&
(IP6_EQ(&is->is_src, dst) || (flags & SI_W_SADDR))) {
if (tcp) {
if ((dp == is->is_sport || flags & SI_W_SPORT)&&
(sp == is->is_dport || flags & SI_W_DPORT))
ret = 1;
} else {
ret = 1;
}
}
}
if (ret == 0)
return NULL;
/*
* Whether or not this should be here, is questionable, but the aim
* is to get this out of the main line.
*/
if (tcp == NULL)
flags = is->is_flags & ~(SI_WILDP|SI_NEWFR|SI_CLONE|SI_CLONED);
/*
* Only one of the source or destination address can be flaged as a
* wildcard. Fill in the missing address, if set.
* For IPv6, if the address being copied in is multicast, then
* don't reset the wild flag - multicast causes it to be set in the
* first place!
*/
if ((flags & (SI_W_SADDR|SI_W_DADDR))) {
fr_ip_t *fi = &fin->fin_fi;
if ((flags & SI_W_SADDR) != 0) {
if (rev == 0) {
#ifdef USE_INET6
if (is->is_v == 6 &&
IN6_IS_ADDR_MULTICAST(&fi->fi_src.in6))
/*EMPTY*/;
else
#endif
{
is->is_src = fi->fi_src;
is->is_flags &= ~SI_W_SADDR;
}
} else {
#ifdef USE_INET6
if (is->is_v == 6 &&
IN6_IS_ADDR_MULTICAST(&fi->fi_dst.in6))
/*EMPTY*/;
else
#endif
{
is->is_src = fi->fi_dst;
is->is_flags &= ~SI_W_SADDR;
}
}
} else if ((flags & SI_W_DADDR) != 0) {
if (rev == 0) {
#ifdef USE_INET6
if (is->is_v == 6 &&
IN6_IS_ADDR_MULTICAST(&fi->fi_dst.in6))
/*EMPTY*/;
else
#endif
{
is->is_dst = fi->fi_dst;
is->is_flags &= ~SI_W_DADDR;
}
} else {
#ifdef USE_INET6
if (is->is_v == 6 &&
IN6_IS_ADDR_MULTICAST(&fi->fi_src.in6))
/*EMPTY*/;
else
#endif
{
is->is_dst = fi->fi_src;
is->is_flags &= ~SI_W_DADDR;
}
}
}
if ((is->is_flags & (SI_WILDA|SI_WILDP)) == 0) {
ATOMIC_DECL(ifs->ifs_ips_stats.iss_wild);
}
}
flx = fin->fin_flx & cmask;
cflx = is->is_flx[out][rev];
/*
* Match up any flags set from IP options.
*/
if ((cflx && (flx != (cflx & cmask))) ||
((fin->fin_optmsk & is->is_optmsk[rev]) != is->is_opt[rev]) ||
((fin->fin_secmsk & is->is_secmsk) != is->is_sec) ||
((fin->fin_auth & is->is_authmsk) != is->is_auth))
return NULL;
/*
* Only one of the source or destination port can be flagged as a
* wildcard. When filling it in, fill in a copy of the matched entry
* if it has the cloning flag set.
*/
if ((fin->fin_flx & FI_IGNORE) != 0) {
fin->fin_rev = rev;
return is;
}
if ((flags & (SI_W_SPORT|SI_W_DPORT))) {
if ((flags & SI_CLONE) != 0) {
ipstate_t *clone;
clone = fr_stclone(fin, tcp, is);
if (clone == NULL)
return NULL;
is = clone;
} else {
ATOMIC_DECL(ifs->ifs_ips_stats.iss_wild);
}
if ((flags & SI_W_SPORT) != 0) {
if (rev == 0) {
is->is_sport = sp;
is->is_send = ntohl(tcp->th_seq);
} else {
is->is_sport = dp;
is->is_send = ntohl(tcp->th_ack);
}
is->is_maxsend = is->is_send + 1;
} else if ((flags & SI_W_DPORT) != 0) {
if (rev == 0) {
is->is_dport = dp;
is->is_dend = ntohl(tcp->th_ack);
} else {
is->is_dport = sp;
is->is_dend = ntohl(tcp->th_seq);
}
is->is_maxdend = is->is_dend + 1;
}
is->is_flags &= ~(SI_W_SPORT|SI_W_DPORT);
if ((flags & SI_CLONED) && ifs->ifs_ipstate_logging)
ipstate_log(is, ISL_CLONE, ifs);
}
ret = -1;
if (is->is_flx[out][rev] == 0) {
is->is_flx[out][rev] = flx;
is->is_opt[rev] = fin->fin_optmsk;
if (is->is_v == 6) {
is->is_opt[rev] &= ~0x8;
is->is_optmsk[rev] &= ~0x8;
}
}
/*
* Check if the interface name for this "direction" is set and if not,
* fill it in.
*/
if (is->is_ifp[idx] == NULL &&
(*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) {
is->is_ifp[idx] = ifp;
COPYIFNAME(ifp, is->is_ifname[idx], fin->fin_v);
}
fin->fin_rev = rev;
return is;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_checkicmpmatchingstate */
/* Returns: Nil */
/* Parameters: fin(I) - pointer to packet information */
/* */
/* If we've got an ICMP error message, using the information stored in the */
/* ICMP packet, look for a matching state table entry. */
/* */
/* If we return NULL then no lock on ipf_state is held. */
/* If we return non-null then a read-lock on ipf_state is held. */
/* ------------------------------------------------------------------------ */
static ipstate_t *fr_checkicmpmatchingstate(fin)
fr_info_t *fin;
{
ipstate_t *is, **isp;
u_short sport, dport;
u_char pr;
int backward, i, oi;
i6addr_t dst, src;
struct icmp *ic;
u_short savelen;
icmphdr_t *icmp;
fr_info_t ofin;
tcphdr_t *tcp;
int len;
ip_t *oip;
u_int hv;
ipf_stack_t *ifs = fin->fin_ifs;
/*
* Does it at least have the return (basic) IP header ?
* Is it an actual recognised ICMP error type?
* Only a basic IP header (no options) should be with
* an ICMP error header.
*/
if ((fin->fin_v != 4) || (fin->fin_hlen != sizeof(ip_t)) ||
(fin->fin_plen < ICMPERR_MINPKTLEN) ||
!(fin->fin_flx & FI_ICMPERR))
return NULL;
ic = fin->fin_dp;
oip = (ip_t *)((char *)ic + ICMPERR_ICMPHLEN);
/*
* Check if the at least the old IP header (with options) and
* 8 bytes of payload is present.
*/
if (fin->fin_plen < ICMPERR_MAXPKTLEN + ((IP_HL(oip) - 5) << 2))
return NULL;
/*
* Sanity Checks.
*/
len = fin->fin_dlen - ICMPERR_ICMPHLEN;
if ((len <= 0) || ((IP_HL(oip) << 2) > len))
return NULL;
/*
* Is the buffer big enough for all of it ? It's the size of the IP
* header claimed in the encapsulated part which is of concern. It
* may be too big to be in this buffer but not so big that it's
* outside the ICMP packet, leading to TCP deref's causing problems.
* This is possible because we don't know how big oip_hl is when we
* do the pullup early in fr_check() and thus can't guarantee it is
* all here now.
*/
#ifdef _KERNEL
{
mb_t *m;
m = fin->fin_m;
# if defined(MENTAT)
if ((char *)oip + len > (char *)m->b_wptr)
return NULL;
# else
if ((char *)oip + len > (char *)fin->fin_ip + m->m_len)
return NULL;
# endif
}
#endif
bcopy((char *)fin, (char *)&ofin, sizeof(fin));
/*
* in the IPv4 case we must zero the i6addr union otherwise
* the IP6_EQ and IP6_NEQ macros produce the wrong results because
* of the 'junk' in the unused part of the union
*/
bzero((char *)&src, sizeof(src));
bzero((char *)&dst, sizeof(dst));
/*
* we make an fin entry to be able to feed it to
* matchsrcdst note that not all fields are encessary
* but this is the cleanest way. Note further we fill
* in fin_mp such that if someone uses it we'll get
* a kernel panic. fr_matchsrcdst does not use this.
*
* watch out here, as ip is in host order and oip in network
* order. Any change we make must be undone afterwards, like
* oip->ip_off - it is still in network byte order so fix it.
*/
savelen = oip->ip_len;
oip->ip_len = len;
oip->ip_off = ntohs(oip->ip_off);
ofin.fin_flx = FI_NOCKSUM;
ofin.fin_v = 4;
ofin.fin_ip = oip;
ofin.fin_m = NULL; /* if dereferenced, panic XXX */
ofin.fin_mp = NULL; /* if dereferenced, panic XXX */
ofin.fin_plen = fin->fin_dlen - ICMPERR_ICMPHLEN;
(void) fr_makefrip(IP_HL(oip) << 2, oip, &ofin);
ofin.fin_ifp = fin->fin_ifp;
ofin.fin_out = !fin->fin_out;
/*
* Reset the short and bad flag here because in fr_matchsrcdst()
* the flags for the current packet (fin_flx) are compared against
* those for the existing session.
*/
ofin.fin_flx &= ~(FI_BAD|FI_SHORT);
/*
* Put old values of ip_len and ip_off back as we don't know
* if we have to forward the packet (or process it again.
*/
oip->ip_len = savelen;
oip->ip_off = htons(oip->ip_off);
switch (oip->ip_p)
{
case IPPROTO_ICMP :
/*
* an ICMP error can only be generated as a result of an
* ICMP query, not as the response on an ICMP error
*
* XXX theoretically ICMP_ECHOREP and the other reply's are
* ICMP query's as well, but adding them here seems strange XXX
*/
if ((ofin.fin_flx & FI_ICMPERR) != 0)
return NULL;
/*
* perform a lookup of the ICMP packet in the state table
*/
icmp = (icmphdr_t *)((char *)oip + (IP_HL(oip) << 2));
hv = (pr = oip->ip_p);
src.in4 = oip->ip_src;
hv += src.in4.s_addr;
dst.in4 = oip->ip_dst;
hv += dst.in4.s_addr;
hv += icmp->icmp_id;
hv = DOUBLE_HASH(hv, ifs);
READ_ENTER(&ifs->ifs_ipf_state);
for (isp = &ifs->ifs_ips_table[hv]; ((is = *isp) != NULL); ) {
isp = &is->is_hnext;
if ((is->is_p != pr) || (is->is_v != 4))
continue;
if (is->is_pass & FR_NOICMPERR)
continue;
is = fr_matchsrcdst(&ofin, is, &src, &dst,
NULL, FI_ICMPCMP);
if (is != NULL) {
if ((is->is_pass & FR_NOICMPERR) != 0) {
RWLOCK_EXIT(&ifs->ifs_ipf_state);
return NULL;
}
/*
* i : the index of this packet (the icmp
* unreachable)
* oi : the index of the original packet found
* in the icmp header (i.e. the packet
* causing this icmp)
* backward : original packet was backward
* compared to the state
*/
backward = IP6_NEQ(&is->is_src, &src);
fin->fin_rev = !backward;
i = (!backward << 1) + fin->fin_out;
oi = (backward << 1) + ofin.fin_out;
if (is->is_icmppkts[i] > is->is_pkts[oi])
continue;
ifs->ifs_ips_stats.iss_hits++;
is->is_icmppkts[i]++;
return is;
}
}
RWLOCK_EXIT(&ifs->ifs_ipf_state);
return NULL;
case IPPROTO_TCP :
case IPPROTO_UDP :
break;
default :
return NULL;
}
tcp = (tcphdr_t *)((char *)oip + (IP_HL(oip) << 2));
dport = tcp->th_dport;
sport = tcp->th_sport;
hv = (pr = oip->ip_p);
src.in4 = oip->ip_src;
hv += src.in4.s_addr;
dst.in4 = oip->ip_dst;
hv += dst.in4.s_addr;
hv += dport;
hv += sport;
hv = DOUBLE_HASH(hv, ifs);
READ_ENTER(&ifs->ifs_ipf_state);
for (isp = &ifs->ifs_ips_table[hv]; ((is = *isp) != NULL); ) {
isp = &is->is_hnext;
/*
* Only allow this icmp though if the
* encapsulated packet was allowed through the
* other way around. Note that the minimal amount
* of info present does not allow for checking against
* tcp internals such as seq and ack numbers. Only the
* ports are known to be present and can be even if the
* short flag is set.
*/
if ((is->is_p == pr) && (is->is_v == 4) &&
(is = fr_matchsrcdst(&ofin, is, &src, &dst,
tcp, FI_ICMPCMP))) {
/*
* i : the index of this packet (the icmp unreachable)
* oi : the index of the original packet found in the
* icmp header (i.e. the packet causing this icmp)
* backward : original packet was backward compared to
* the state
*/
backward = IP6_NEQ(&is->is_src, &src);
fin->fin_rev = !backward;
i = (!backward << 1) + fin->fin_out;
oi = (backward << 1) + ofin.fin_out;
if (((is->is_pass & FR_NOICMPERR) != 0) ||
(is->is_icmppkts[i] > is->is_pkts[oi]))
break;
ifs->ifs_ips_stats.iss_hits++;
is->is_icmppkts[i]++;
/*
* we deliberately do not touch the timeouts
* for the accompanying state table entry.
* It remains to be seen if that is correct. XXX
*/
return is;
}
}
RWLOCK_EXIT(&ifs->ifs_ipf_state);
return NULL;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_ipsmove */
/* Returns: Nil */
/* Parameters: is(I) - pointer to state table entry */
/* hv(I) - new hash value for state table entry */
/* Write Locks: ipf_state */
/* */
/* Move a state entry from one position in the hash table to another. */
/* ------------------------------------------------------------------------ */
static void fr_ipsmove(is, hv, ifs)
ipstate_t *is;
u_int hv;
ipf_stack_t *ifs;
{
ipstate_t **isp;
u_int hvm;
ASSERT(rw_read_locked(&ifs->ifs_ipf_state.ipf_lk) == 0);
hvm = is->is_hv;
/*
* Remove the hash from the old location...
*/
isp = is->is_phnext;
if (is->is_hnext)
is->is_hnext->is_phnext = isp;
*isp = is->is_hnext;
if (ifs->ifs_ips_table[hvm] == NULL)
ifs->ifs_ips_stats.iss_inuse--;
ifs->ifs_ips_stats.iss_bucketlen[hvm]--;
/*
* ...and put the hash in the new one.
*/
hvm = DOUBLE_HASH(hv, ifs);
is->is_hv = hvm;
isp = &ifs->ifs_ips_table[hvm];
if (*isp)
(*isp)->is_phnext = &is->is_hnext;
else
ifs->ifs_ips_stats.iss_inuse++;
ifs->ifs_ips_stats.iss_bucketlen[hvm]++;
is->is_phnext = isp;
is->is_hnext = *isp;
*isp = is;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_stlookup */
/* Returns: ipstate_t* - NULL == no matching state found, */
/* else pointer to state information is returned */
/* Parameters: fin(I) - pointer to packet information */
/* tcp(I) - pointer to TCP/UDP header. */
/* */
/* Search the state table for a matching entry to the packet described by */
/* the contents of *fin. */
/* */
/* If we return NULL then no lock on ipf_state is held. */
/* If we return non-null then a read-lock on ipf_state is held. */
/* ------------------------------------------------------------------------ */
ipstate_t *fr_stlookup(fin, tcp, ifqp)
fr_info_t *fin;
tcphdr_t *tcp;
ipftq_t **ifqp;
{
u_int hv, hvm, pr, v, tryagain;
ipstate_t *is, **isp;
u_short dport, sport;
i6addr_t src, dst;
struct icmp *ic;
ipftq_t *ifq;
int oow;
ipf_stack_t *ifs = fin->fin_ifs;
is = NULL;
ifq = NULL;
tcp = fin->fin_dp;
ic = (struct icmp *)tcp;
hv = (pr = fin->fin_fi.fi_p);
src = fin->fin_fi.fi_src;
dst = fin->fin_fi.fi_dst;
hv += src.in4.s_addr;
hv += dst.in4.s_addr;
v = fin->fin_fi.fi_v;
#ifdef USE_INET6
if (v == 6) {
hv += fin->fin_fi.fi_src.i6[1];
hv += fin->fin_fi.fi_src.i6[2];
hv += fin->fin_fi.fi_src.i6[3];
if ((fin->fin_p == IPPROTO_ICMPV6) &&
IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_dst.in6)) {
hv -= dst.in4.s_addr;
} else {
hv += fin->fin_fi.fi_dst.i6[1];
hv += fin->fin_fi.fi_dst.i6[2];
hv += fin->fin_fi.fi_dst.i6[3];
}
}
#endif
/*
* Search the hash table for matching packet header info.
*/
switch (pr)
{
#ifdef USE_INET6
case IPPROTO_ICMPV6 :
tryagain = 0;
if (v == 6) {
if ((ic->icmp_type == ICMP6_ECHO_REQUEST) ||
(ic->icmp_type == ICMP6_ECHO_REPLY)) {
hv += ic->icmp_id;
}
}
READ_ENTER(&ifs->ifs_ipf_state);
icmp6again:
hvm = DOUBLE_HASH(hv, ifs);
for (isp = &ifs->ifs_ips_table[hvm]; ((is = *isp) != NULL); ) {
isp = &is->is_hnext;
if ((is->is_p != pr) || (is->is_v != v))
continue;
is = fr_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP);
if (is != NULL &&
fr_matchicmpqueryreply(v, &is->is_icmp,
ic, fin->fin_rev)) {
if (fin->fin_rev)
ifq = &ifs->ifs_ips_icmpacktq;
else
ifq = &ifs->ifs_ips_icmptq;
break;
}
}
if (is != NULL) {
if ((tryagain != 0) && !(is->is_flags & SI_W_DADDR)) {
hv += fin->fin_fi.fi_src.i6[0];
hv += fin->fin_fi.fi_src.i6[1];
hv += fin->fin_fi.fi_src.i6[2];
hv += fin->fin_fi.fi_src.i6[3];
fr_ipsmove(is, hv, ifs);
MUTEX_DOWNGRADE(&ifs->ifs_ipf_state);
}
break;
}
RWLOCK_EXIT(&ifs->ifs_ipf_state);
/*
* No matching icmp state entry. Perhaps this is a
* response to another state entry.
*
* XXX With some ICMP6 packets, the "other" address is already
* in the packet, after the ICMP6 header, and this could be
* used in place of the multicast address. However, taking
* advantage of this requires some significant code changes
* to handle the specific types where that is the case.
*/
if ((ifs->ifs_ips_stats.iss_wild != 0) && (v == 6) && (tryagain == 0) &&
!IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_src.in6)) {
hv -= fin->fin_fi.fi_src.i6[0];
hv -= fin->fin_fi.fi_src.i6[1];
hv -= fin->fin_fi.fi_src.i6[2];
hv -= fin->fin_fi.fi_src.i6[3];
tryagain = 1;
WRITE_ENTER(&ifs->ifs_ipf_state);
goto icmp6again;
}
is = fr_checkicmp6matchingstate(fin);
if (is != NULL)
return is;
break;
#endif
case IPPROTO_ICMP :
if (v == 4) {
hv += ic->icmp_id;
}
hv = DOUBLE_HASH(hv, ifs);
READ_ENTER(&ifs->ifs_ipf_state);
for (isp = &ifs->ifs_ips_table[hv]; ((is = *isp) != NULL); ) {
isp = &is->is_hnext;
if ((is->is_p != pr) || (is->is_v != v))
continue;
is = fr_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP);
if (is != NULL &&
fr_matchicmpqueryreply(v, &is->is_icmp,
ic, fin->fin_rev)) {
if (fin->fin_rev)
ifq = &ifs->ifs_ips_icmpacktq;
else
ifq = &ifs->ifs_ips_icmptq;
break;
}
}
if (is == NULL) {
RWLOCK_EXIT(&ifs->ifs_ipf_state);
}
break;
case IPPROTO_TCP :
case IPPROTO_UDP :
ifqp = NULL;
sport = htons(fin->fin_data[0]);
hv += sport;
dport = htons(fin->fin_data[1]);
hv += dport;
oow = 0;
tryagain = 0;
READ_ENTER(&ifs->ifs_ipf_state);
retry_tcpudp:
hvm = DOUBLE_HASH(hv, ifs);
for (isp = &ifs->ifs_ips_table[hvm]; ((is = *isp) != NULL); ) {
isp = &is->is_hnext;
if ((is->is_p != pr) || (is->is_v != v))
continue;
fin->fin_flx &= ~FI_OOW;
is = fr_matchsrcdst(fin, is, &src, &dst, tcp, FI_CMP);
if (is != NULL) {
if (pr == IPPROTO_TCP) {
if (!fr_tcpstate(fin, tcp, is)) {
oow |= fin->fin_flx & FI_OOW;
continue;
}
}
break;
}
}
if (is != NULL) {
if (tryagain &&
!(is->is_flags & (SI_CLONE|SI_WILDP|SI_WILDA))) {
hv += dport;
hv += sport;
fr_ipsmove(is, hv, ifs);
MUTEX_DOWNGRADE(&ifs->ifs_ipf_state);
}
break;
}
RWLOCK_EXIT(&ifs->ifs_ipf_state);
if (!tryagain && ifs->ifs_ips_stats.iss_wild) {
hv -= dport;
hv -= sport;
tryagain = 1;
WRITE_ENTER(&ifs->ifs_ipf_state);
goto retry_tcpudp;
}
fin->fin_flx |= oow;
break;
#if 0
case IPPROTO_GRE :
gre = fin->fin_dp;
if (GRE_REV(gre->gr_flags) == 1) {
hv += gre->gr_call;
}
/* FALLTHROUGH */
#endif
default :
ifqp = NULL;
hvm = DOUBLE_HASH(hv, ifs);
READ_ENTER(&ifs->ifs_ipf_state);
for (isp = &ifs->ifs_ips_table[hvm]; ((is = *isp) != NULL); ) {
isp = &is->is_hnext;
if ((is->is_p != pr) || (is->is_v != v))
continue;
is = fr_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP);
if (is != NULL) {
ifq = &ifs->ifs_ips_iptq;
break;
}
}
if (is == NULL) {
RWLOCK_EXIT(&ifs->ifs_ipf_state);
}
break;
}
if ((is != NULL) && ((is->is_sti.tqe_flags & TQE_RULEBASED) != 0) &&
(is->is_tqehead[fin->fin_rev] != NULL))
ifq = is->is_tqehead[fin->fin_rev];
if (ifq != NULL && ifqp != NULL)
*ifqp = ifq;
return is;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_updatestate */
/* Returns: Nil */
/* Parameters: fin(I) - pointer to packet information */
/* is(I) - pointer to state table entry */
/* Read Locks: ipf_state */
/* */
/* Updates packet and byte counters for a newly received packet. Seeds the */
/* fragment cache with a new entry as required. */
/* ------------------------------------------------------------------------ */
void fr_updatestate(fin, is, ifq)
fr_info_t *fin;
ipstate_t *is;
ipftq_t *ifq;
{
ipftqent_t *tqe;
int i, pass;
ipf_stack_t *ifs = fin->fin_ifs;
i = (fin->fin_rev << 1) + fin->fin_out;
/*
* For TCP packets, ifq == NULL. For all others, check if this new
* queue is different to the last one it was on and move it if so.
*/
tqe = &is->is_sti;
MUTEX_ENTER(&is->is_lock);
if ((tqe->tqe_flags & TQE_RULEBASED) != 0)
ifq = is->is_tqehead[fin->fin_rev];
if (ifq != NULL)
fr_movequeue(tqe, tqe->tqe_ifq, ifq, ifs);
is->is_pkts[i]++;
is->is_bytes[i] += fin->fin_plen;
MUTEX_EXIT(&is->is_lock);
#ifdef IPFILTER_SYNC
if (is->is_flags & IS_STATESYNC)
ipfsync_update(SMC_STATE, fin, is->is_sync);
#endif
ATOMIC_INCL(ifs->ifs_ips_stats.iss_hits);
fin->fin_fr = is->is_rule;
/*
* If this packet is a fragment and the rule says to track fragments,
* then create a new fragment cache entry.
*/
pass = is->is_pass;
if ((fin->fin_flx & FI_FRAG) && FR_ISPASS(pass))
(void) fr_newfrag(fin, pass ^ FR_KEEPSTATE);
}
/* ------------------------------------------------------------------------ */
/* Function: fr_checkstate */
/* Returns: frentry_t* - NULL == search failed, */
/* else pointer to rule for matching state */
/* Parameters: ifp(I) - pointer to interface */
/* passp(I) - pointer to filtering result flags */
/* */
/* Check if a packet is associated with an entry in the state table. */
/* ------------------------------------------------------------------------ */
frentry_t *fr_checkstate(fin, passp)
fr_info_t *fin;
u_32_t *passp;
{
ipstate_t *is;
frentry_t *fr;
tcphdr_t *tcp;
ipftq_t *ifq;
u_int pass;
ipf_stack_t *ifs = fin->fin_ifs;
if (ifs->ifs_fr_state_lock || (ifs->ifs_ips_list == NULL) ||
(fin->fin_flx & (FI_SHORT|FI_STATE|FI_FRAGBODY|FI_BAD)))
return NULL;
is = NULL;
if ((fin->fin_flx & FI_TCPUDP) ||
(fin->fin_fi.fi_p == IPPROTO_ICMP)
#ifdef USE_INET6
|| (fin->fin_fi.fi_p == IPPROTO_ICMPV6)
#endif
)
tcp = fin->fin_dp;
else
tcp = NULL;
/*
* Search the hash table for matching packet header info.
*/
ifq = NULL;
is = fin->fin_state;
if (is == NULL)
is = fr_stlookup(fin, tcp, &ifq);
switch (fin->fin_p)
{
#ifdef USE_INET6
case IPPROTO_ICMPV6 :
if (is != NULL)
break;
if (fin->fin_v == 6) {
is = fr_checkicmp6matchingstate(fin);
if (is != NULL)
goto matched;
}
break;
#endif
case IPPROTO_ICMP :
if (is != NULL)
break;
/*
* No matching icmp state entry. Perhaps this is a
* response to another state entry.
*/
is = fr_checkicmpmatchingstate(fin);
if (is != NULL)
goto matched;
break;
case IPPROTO_TCP :
if (is == NULL)
break;
if (is->is_pass & FR_NEWISN) {
if (fin->fin_out == 0)
fr_fixinisn(fin, is);
else if (fin->fin_out == 1)
fr_fixoutisn(fin, is);
}
break;
default :
if (fin->fin_rev)
ifq = &ifs->ifs_ips_udpacktq;
else
ifq = &ifs->ifs_ips_udptq;
break;
}
if (is == NULL) {
ATOMIC_INCL(ifs->ifs_ips_stats.iss_miss);
return NULL;
}
matched:
fr = is->is_rule;
if (fr != NULL) {
if ((fin->fin_out == 0) && (fr->fr_nattag.ipt_num[0] != 0)) {
if (fin->fin_nattag == NULL)
return NULL;
if (fr_matchtag(&fr->fr_nattag, fin->fin_nattag) != 0)
return NULL;
}
(void) strncpy(fin->fin_group, fr->fr_group, FR_GROUPLEN);
fin->fin_icode = fr->fr_icode;
}
fin->fin_rule = is->is_rulen;
pass = is->is_pass;
fr_updatestate(fin, is, ifq);
if (fin->fin_out == 1)
fin->fin_nat = is->is_nat[fin->fin_rev];
fin->fin_state = is;
is->is_touched = ifs->ifs_fr_ticks;
MUTEX_ENTER(&is->is_lock);
is->is_ref++;
MUTEX_EXIT(&is->is_lock);
RWLOCK_EXIT(&ifs->ifs_ipf_state);
fin->fin_flx |= FI_STATE;
if ((pass & FR_LOGFIRST) != 0)
pass &= ~(FR_LOGFIRST|FR_LOG);
*passp = pass;
return fr;
}
/* ------------------------------------------------------------------------ */
/* Function: fr_fixoutisn */
/* Returns: Nil */
/* Parameters: fin(I) - pointer to packet information */
/* is(I) - pointer to master state structure */
/* */
/* Called only for outbound packets, adjusts the sequence number and the */
/* TCP checksum to match that change. */
/* ------------------------------------------------------------------------ */
static void fr_fixoutisn(fin, is)
fr_info_t *fin;
ipstate_t *is;
{
tcphdr_t *tcp;
int rev;
u_32_t seq;
tcp = fin->fin_dp;
rev = fin->fin_rev;
if ((is->is_flags & IS_ISNSYN