ip_state.c revision 102520
1/*
2 * Copyright (C) 1995-2002 by Darren Reed.
3 *
4 * See the IPFILTER.LICENCE file for details on licencing.
5 */
6
7#ifdef __sgi
8# include <sys/ptimers.h>
9#endif
10#include <sys/errno.h>
11#include <sys/types.h>
12#include <sys/param.h>
13#include <sys/file.h>
14#if defined(__NetBSD__) && (NetBSD >= 199905) && !defined(IPFILTER_LKM) && \
15    defined(_KERNEL)
16# include "opt_ipfilter_log.h"
17#endif
18#if defined(_KERNEL) && defined(__FreeBSD_version) && \
19    (__FreeBSD_version >= 400000) && !defined(KLD_MODULE)
20#include "opt_inet6.h"
21#endif
22#if !defined(_KERNEL) && !defined(KERNEL) && !defined(__KERNEL__)
23# include <stdio.h>
24# include <stdlib.h>
25# include <string.h>
26#else
27# ifdef linux
28#  include <linux/kernel.h>
29#  include <linux/module.h>
30# endif
31#endif
32#if (defined(KERNEL) || defined(_KERNEL)) && (__FreeBSD_version >= 220000)
33# include <sys/filio.h>
34# include <sys/fcntl.h>
35# if (__FreeBSD_version >= 300000) && !defined(IPFILTER_LKM)
36#  include "opt_ipfilter.h"
37# endif
38#else
39# include <sys/ioctl.h>
40#endif
41#include <sys/time.h>
42#ifndef linux
43# include <sys/protosw.h>
44#endif
45#include <sys/socket.h>
46#if (defined(_KERNEL) || defined(KERNEL)) && !defined(linux)
47# include <sys/systm.h>
48#endif
49#if !defined(__SVR4) && !defined(__svr4__)
50# ifndef linux
51#  include <sys/mbuf.h>
52# endif
53#else
54# include <sys/filio.h>
55# include <sys/byteorder.h>
56# ifdef _KERNEL
57#  include <sys/dditypes.h>
58# endif
59# include <sys/stream.h>
60# include <sys/kmem.h>
61#endif
62
63#include <net/if.h>
64#ifdef sun
65# include <net/af.h>
66#endif
67#include <net/route.h>
68#include <netinet/in.h>
69#include <netinet/in_systm.h>
70#include <netinet/ip.h>
71#include <netinet/tcp.h>
72#ifndef linux
73# include <netinet/ip_var.h>
74# include <netinet/tcp_fsm.h>
75#endif
76#include <netinet/udp.h>
77#include <netinet/ip_icmp.h>
78#include "netinet/ip_compat.h"
79#include <netinet/tcpip.h>
80#include "netinet/ip_fil.h"
81#include "netinet/ip_nat.h"
82#include "netinet/ip_frag.h"
83#include "netinet/ip_state.h"
84#ifdef	USE_INET6
85#include <netinet/icmp6.h>
86#endif
87#if (__FreeBSD_version >= 300000)
88# include <sys/malloc.h>
89# if (defined(_KERNEL) || defined(KERNEL)) && !defined(IPFILTER_LKM)
90#  include <sys/libkern.h>
91#  include <sys/systm.h>
92# endif
93#endif
94
95#if !defined(lint)
96static const char sccsid[] = "@(#)ip_state.c	1.8 6/5/96 (C) 1993-2000 Darren Reed";
97/* static const char rcsid[] = "@(#)$Id: ip_state.c,v 2.30.2.38 2001/07/23 13:49:46 darrenr Exp $"; */
98static const char rcsid[] = "@(#)$FreeBSD: head/sys/contrib/ipfilter/netinet/ip_state.c 102520 2002-08-28 13:41:36Z darrenr $";
99#endif
100
101#ifndef	MIN
102# define	MIN(a,b)	(((a)<(b))?(a):(b))
103#endif
104
105#define	TCP_CLOSE	(TH_FIN|TH_RST)
106
107static ipstate_t **ips_table = NULL;
108static int	ips_num = 0;
109static int	ips_wild = 0;
110static ips_stat_t ips_stats;
111#if	(SOLARIS || defined(__sgi)) && defined(_KERNEL)
112extern	KRWLOCK_T	ipf_state, ipf_mutex;
113extern	kmutex_t	ipf_rw;
114#endif
115
116#ifdef	USE_INET6
117static frentry_t *fr_checkicmp6matchingstate __P((ip6_t *, fr_info_t *));
118#endif
119static int fr_matchsrcdst __P((ipstate_t *, union i6addr, union i6addr,
120			       fr_info_t *, tcphdr_t *));
121static frentry_t *fr_checkicmpmatchingstate __P((ip_t *, fr_info_t *));
122static int fr_matchicmpqueryreply __P((int, ipstate_t *, icmphdr_t *));
123static int fr_state_flush __P((int));
124static ips_stat_t *fr_statetstats __P((void));
125static void fr_delstate __P((ipstate_t *));
126static int fr_state_remove __P((caddr_t));
127static void fr_ipsmove __P((ipstate_t **, ipstate_t *, u_int));
128static int fr_tcpoptions __P((tcphdr_t *));
129int fr_stputent __P((caddr_t));
130int fr_stgetent __P((caddr_t));
131void fr_stinsert __P((ipstate_t *));
132
133
134#define	FIVE_DAYS	(2 * 5 * 86400)	/* 5 days: half closed session */
135
136#define	TCP_MSL	240			/* 2 minutes */
137u_long	fr_tcpidletimeout = FIVE_DAYS,
138	fr_tcpclosewait = 2 * TCP_MSL,
139	fr_tcplastack = 2 * TCP_MSL,
140	fr_tcptimeout = 2 * TCP_MSL,
141	fr_tcpclosed = 120,
142	fr_tcphalfclosed = 2 * 2 * 3600,    /* 2 hours */
143	fr_udptimeout = 240,
144	fr_udpacktimeout = 24,
145	fr_icmptimeout = 120,
146	fr_icmpacktimeout = 12;
147int	fr_statemax = IPSTATE_MAX,
148	fr_statesize = IPSTATE_SIZE;
149int	fr_state_doflush = 0,
150	fr_state_lock = 0;
151ipstate_t *ips_list = NULL;
152
153static 	int icmpreplytype4[ICMP_MAXTYPE + 1];
154#ifdef	USE_INET6
155static 	int icmpreplytype6[ICMP6_MAXTYPE + 1];
156#endif
157
158int fr_stateinit()
159{
160	int i;
161
162	KMALLOCS(ips_table, ipstate_t **, fr_statesize * sizeof(ipstate_t *));
163	if (ips_table != NULL)
164		bzero((char *)ips_table, fr_statesize * sizeof(ipstate_t *));
165	else
166		return -1;
167
168	/* fill icmp reply type table */
169	for (i = 0; i <= ICMP_MAXTYPE; i++)
170		icmpreplytype4[i] = -1;
171	icmpreplytype4[ICMP_ECHO] = ICMP_ECHOREPLY;
172	icmpreplytype4[ICMP_TSTAMP] = ICMP_TSTAMPREPLY;
173	icmpreplytype4[ICMP_IREQ] = ICMP_IREQREPLY;
174	icmpreplytype4[ICMP_MASKREQ] = ICMP_MASKREPLY;
175#ifdef	USE_INET6
176	/* fill icmp reply type table */
177	for (i = 0; i <= ICMP6_MAXTYPE; i++)
178		icmpreplytype6[i] = -1;
179	icmpreplytype6[ICMP6_ECHO_REQUEST] = ICMP6_ECHO_REPLY;
180	icmpreplytype6[ICMP6_MEMBERSHIP_QUERY] = ICMP6_MEMBERSHIP_REPORT;
181	icmpreplytype6[ICMP6_NI_QUERY] = ICMP6_NI_REPLY;
182	icmpreplytype6[ND_ROUTER_SOLICIT] = ND_ROUTER_ADVERT;
183	icmpreplytype6[ND_NEIGHBOR_SOLICIT] = ND_NEIGHBOR_ADVERT;
184#endif
185
186	return 0;
187}
188
189
190static ips_stat_t *fr_statetstats()
191{
192	ips_stats.iss_active = ips_num;
193	ips_stats.iss_table = ips_table;
194	ips_stats.iss_list = ips_list;
195	return &ips_stats;
196}
197
198
199/*
200 * flush state tables.  two actions currently defined:
201 * which == 0 : flush all state table entries
202 * which == 1 : flush TCP connections which have started to close but are
203 *	        stuck for some reason.
204 * which == 2 : flush TCP connections which have been idle for a long time,
205 *              starting at > 4 days idle and working back in successive half-
206 *              days to at most 12 hours old.
207 */
208static int fr_state_flush(which)
209int which;
210{
211	ipstate_t *is, **isp;
212#if defined(_KERNEL) && !SOLARIS
213	int s;
214#endif
215	int delete, removed = 0, try;
216
217	SPL_NET(s);
218	for (isp = &ips_list; (is = *isp); ) {
219		delete = 0;
220
221		switch (which)
222		{
223		case 0 :
224			delete = 1;
225			break;
226		case 1 :
227		case 2 :
228			if (is->is_p != IPPROTO_TCP)
229				break;
230			if ((is->is_state[0] != TCPS_ESTABLISHED) ||
231			    (is->is_state[1] != TCPS_ESTABLISHED))
232				delete = 1;
233			break;
234		}
235
236		if (delete) {
237			if (is->is_p == IPPROTO_TCP)
238				ips_stats.iss_fin++;
239			else
240				ips_stats.iss_expire++;
241#ifdef	IPFILTER_LOG
242			ipstate_log(is, ISL_FLUSH);
243#endif
244			fr_delstate(is);
245			removed++;
246		} else
247			isp = &is->is_next;
248	}
249
250	/*
251	 * Asked to remove inactive entries, try again if first attempt
252	 * failed.  In this case, 86400 is half a day because the counter is
253	 * activated every half second.
254	 */
255	if ((which == 2) && (removed == 0)) {
256		try = 86400;	/* half a day */
257		for (; (try < FIVE_DAYS) && (removed == 0); try += 86400) {
258			for (isp = &ips_list; (is = *isp); ) {
259				delete = 0;
260				if ((is->is_p == IPPROTO_TCP) &&
261				    ((is->is_state[0] == TCPS_ESTABLISHED) ||
262				     (is->is_state[1] == TCPS_ESTABLISHED)) &&
263				    (is->is_age < try)) {
264					ips_stats.iss_fin++;
265					delete = 1;
266				} else if ((is->is_p != IPPROTO_TCP) &&
267					   (is->is_pkts > 1)) {
268					ips_stats.iss_expire++;
269					delete = 1;
270				}
271				if (delete) {
272#ifdef	IPFILTER_LOG
273					ipstate_log(is, ISL_FLUSH);
274#endif
275					fr_delstate(is);
276					removed++;
277				} else
278					isp = &is->is_next;
279			}
280		}
281	}
282
283	SPL_X(s);
284	return removed;
285}
286
287
288static int fr_state_remove(data)
289caddr_t data;
290{
291	ipstate_t *sp, st;
292	int error;
293
294	sp = &st;
295	error = IRCOPYPTR(data, (caddr_t)&st, sizeof(st));
296	if (error)
297		return EFAULT;
298
299	WRITE_ENTER(&ipf_state);
300	for (sp = ips_list; sp; sp = sp->is_next)
301		if ((sp->is_p == st.is_p) && (sp->is_v == st.is_v) &&
302		    !bcmp((char *)&sp->is_src, (char *)&st.is_src,
303			  sizeof(st.is_src)) &&
304		    !bcmp((char *)&sp->is_dst, (char *)&st.is_dst,
305			  sizeof(st.is_dst)) &&
306		    !bcmp((char *)&sp->is_ps, (char *)&st.is_ps,
307			  sizeof(st.is_ps))) {
308#ifdef	IPFILTER_LOG
309			ipstate_log(sp, ISL_REMOVE);
310#endif
311			fr_delstate(sp);
312			RWLOCK_EXIT(&ipf_state);
313			return 0;
314		}
315	RWLOCK_EXIT(&ipf_state);
316	return ESRCH;
317}
318
319
320int fr_state_ioctl(data, cmd, mode)
321caddr_t data;
322#if defined(__NetBSD__) || defined(__OpenBSD__)
323u_long cmd;
324#else
325int cmd;
326#endif
327int mode;
328{
329	int arg, ret, error = 0;
330
331	switch (cmd)
332	{
333	case SIOCDELST :
334		error = fr_state_remove(data);
335		break;
336	case SIOCIPFFL :
337		error = IRCOPY(data, (caddr_t)&arg, sizeof(arg));
338		if (error)
339			break;
340		if (arg == 0 || arg == 1) {
341			WRITE_ENTER(&ipf_state);
342			ret = fr_state_flush(arg);
343			RWLOCK_EXIT(&ipf_state);
344			error = IWCOPY((caddr_t)&ret, data, sizeof(ret));
345		} else
346			error = EINVAL;
347		break;
348#ifdef	IPFILTER_LOG
349	case SIOCIPFFB :
350		if (!(mode & FWRITE))
351			error = EPERM;
352		else {
353			int tmp;
354
355			tmp = ipflog_clear(IPL_LOGSTATE);
356			IWCOPY((char *)&tmp, data, sizeof(tmp));
357		}
358		break;
359#endif
360	case SIOCGETFS :
361		error = IWCOPYPTR((caddr_t)fr_statetstats(), data,
362				  sizeof(ips_stat_t));
363		break;
364	case FIONREAD :
365#ifdef	IPFILTER_LOG
366		arg = (int)iplused[IPL_LOGSTATE];
367		error = IWCOPY((caddr_t)&arg, (caddr_t)data, sizeof(arg));
368#endif
369		break;
370	case SIOCSTLCK :
371		error = fr_lock(data, &fr_state_lock);
372		break;
373	case SIOCSTPUT :
374		if (!fr_state_lock) {
375			error = EACCES;
376			break;
377		}
378		error = fr_stputent(data);
379		break;
380	case SIOCSTGET :
381		if (!fr_state_lock) {
382			error = EACCES;
383			break;
384		}
385		error = fr_stgetent(data);
386		break;
387	default :
388		error = EINVAL;
389		break;
390	}
391	return error;
392}
393
394
395/*
396 * Copy out state information from the kernel to a user space process.
397 */
398int fr_stgetent(data)
399caddr_t data;
400{
401	register ipstate_t *is, *isn;
402	ipstate_save_t ips;
403	int error;
404
405	error = IRCOPYPTR(data, (caddr_t)&ips, sizeof(ips));
406	if (error)
407		return error;
408
409	isn = ips.ips_next;
410	if (!isn) {
411		isn = ips_list;
412		if (isn == NULL) {
413			if (ips.ips_next == NULL)
414				return ENOENT;
415			return 0;
416		}
417	} else {
418		/*
419		 * Make sure the pointer we're copying from exists in the
420		 * current list of entries.  Security precaution to prevent
421		 * copying of random kernel data.
422		 */
423		for (is = ips_list; is; is = is->is_next)
424			if (is == isn)
425				break;
426		if (!is)
427			return ESRCH;
428	}
429	ips.ips_next = isn->is_next;
430	bcopy((char *)isn, (char *)&ips.ips_is, sizeof(ips.ips_is));
431	if (isn->is_rule)
432		bcopy((char *)isn->is_rule, (char *)&ips.ips_fr,
433		      sizeof(ips.ips_fr));
434	error = IWCOPYPTR((caddr_t)&ips, data, sizeof(ips));
435	if (error)
436		error = EFAULT;
437	return error;
438}
439
440
441int fr_stputent(data)
442caddr_t data;
443{
444	register ipstate_t *is, *isn;
445	ipstate_save_t ips;
446	int error, out, i;
447	frentry_t *fr;
448	char *name;
449
450	error = IRCOPYPTR(data, (caddr_t)&ips, sizeof(ips));
451	if (error)
452		return error;
453
454	KMALLOC(isn, ipstate_t *);
455	if (isn == NULL)
456		return ENOMEM;
457
458	bcopy((char *)&ips.ips_is, (char *)isn, sizeof(*isn));
459	fr = isn->is_rule;
460	if (fr != NULL) {
461		if (isn->is_flags & FI_NEWFR) {
462			KMALLOC(fr, frentry_t *);
463			if (fr == NULL) {
464				KFREE(isn);
465				return ENOMEM;
466			}
467			bcopy((char *)&ips.ips_fr, (char *)fr, sizeof(*fr));
468			out = fr->fr_flags & FR_OUTQUE ? 1 : 0;
469			isn->is_rule = fr;
470			ips.ips_is.is_rule = fr;
471
472			/*
473			 * Look up all the interface names in the rule.
474			 */
475			for (i = 0; i < 4; i++) {
476				name = fr->fr_ifnames[i];
477				if ((name[1] == '\0') &&
478				    ((name[0] == '-') || (name[0] == '*'))) {
479					fr->fr_ifas[i] = NULL;
480				} else if (*name != '\0') {
481					fr->fr_ifas[i] = GETUNIT(name,
482								 fr->fr_v);
483					if (fr->fr_ifas[i] == NULL)
484						fr->fr_ifas[i] = (void *)-1;
485					else {
486						strncpy(isn->is_ifname[i],
487							IFNAME(fr->fr_ifas[i]),
488							IFNAMSIZ);
489					}
490				}
491				isn->is_ifp[out] = fr->fr_ifas[i];
492			}
493
494			/*
495			 * send a copy back to userland of what we ended up
496			 * to allow for verification.
497			 */
498			error = IWCOPYPTR((caddr_t)&ips, data, sizeof(ips));
499			if (error) {
500				KFREE(isn);
501				KFREE(fr);
502				return EFAULT;
503			}
504		} else {
505			for (is = ips_list; is; is = is->is_next)
506				if (is->is_rule == fr)
507					break;
508			if (!is) {
509				KFREE(isn);
510				return ESRCH;
511			}
512		}
513	}
514	fr_stinsert(isn);
515	return 0;
516}
517
518
519/*
520 * Insert a state table entry manually.
521 */
522void fr_stinsert(is)
523register ipstate_t *is;
524{
525	register u_int hv = is->is_hv;
526	char *name;
527	int i;
528
529	MUTEX_INIT(&is->is_lock, "ipf state entry", NULL);
530
531	/*
532	 * Look up all the interface names in the state entry.
533	 */
534	for (i = 0; i < 4; i++) {
535		name = is->is_ifname[i];
536		if ((name[1] == '\0') &&
537		    ((name[0] == '-') || (name[0] == '*'))) {
538			is->is_ifp[0] = NULL;
539		} else if (*name != '\0') {
540			is->is_ifp[i] = GETUNIT(name, is->is_v);
541			if (is->is_ifp[i] == NULL)
542				is->is_ifp[i] = (void *)-1;
543		}
544	}
545
546
547	/*
548	 * add into list table.
549	 */
550	if (ips_list)
551		ips_list->is_pnext = &is->is_next;
552	is->is_pnext = &ips_list;
553	is->is_next = ips_list;
554	ips_list = is;
555	if (ips_table[hv])
556		ips_table[hv]->is_phnext = &is->is_hnext;
557	else
558		ips_stats.iss_inuse++;
559	is->is_phnext = ips_table + hv;
560	is->is_hnext = ips_table[hv];
561	ips_table[hv] = is;
562	ips_num++;
563}
564
565
566/*
567 * Create a new ipstate structure and hang it off the hash table.
568 */
569ipstate_t *fr_addstate(ip, fin, stsave, flags)
570ip_t *ip;
571fr_info_t *fin;
572ipstate_t **stsave;
573u_int flags;
574{
575	register tcphdr_t *tcp = NULL;
576	register ipstate_t *is;
577	register u_int hv;
578	struct icmp *ic;
579	ipstate_t ips;
580	int out, ws;
581	u_int pass;
582	void *ifp;
583
584	if (fr_state_lock || (fin->fin_off != 0) || (fin->fin_fl & FI_SHORT) ||
585	    (fin->fin_misc & FM_BADSTATE))
586		return NULL;
587	if (ips_num == fr_statemax) {
588		ips_stats.iss_max++;
589		fr_state_doflush = 1;
590		return NULL;
591	}
592	out = fin->fin_out;
593	is = &ips;
594	bzero((char *)is, sizeof(*is));
595	ips.is_age = 1;
596	/*
597	 * Copy and calculate...
598	 */
599	hv = (is->is_p = fin->fin_fi.fi_p);
600	is->is_src = fin->fin_fi.fi_src;
601	hv += is->is_saddr;
602	is->is_dst = fin->fin_fi.fi_dst;
603	hv += is->is_daddr;
604#ifdef	USE_INET6
605	if (fin->fin_v == 6) {
606		if ((is->is_p == IPPROTO_ICMPV6) &&
607		    IN6_IS_ADDR_MULTICAST(&is->is_dst.in6)) {
608			/*
609			 * So you can do keep state with neighbour discovery.
610			 */
611			flags |= FI_W_DADDR;
612			hv -= is->is_daddr;
613		} else {
614			hv += is->is_dst.i6[1];
615			hv += is->is_dst.i6[2];
616			hv += is->is_dst.i6[3];
617		}
618		hv += is->is_src.i6[1];
619		hv += is->is_src.i6[2];
620		hv += is->is_src.i6[3];
621	}
622#endif
623
624	switch (is->is_p)
625	{
626		int off;
627
628#ifdef	USE_INET6
629	case IPPROTO_ICMPV6 :
630		ic = (struct icmp *)fin->fin_dp;
631		if ((ic->icmp_type & ICMP6_INFOMSG_MASK) == 0)
632			return NULL;
633
634		switch (ic->icmp_type)
635		{
636		case ICMP6_ECHO_REQUEST :
637			is->is_icmp.ics_type = ic->icmp_type;
638			hv += (is->is_icmp.ics_id = ic->icmp_id);
639			hv += (is->is_icmp.ics_seq = ic->icmp_seq);
640			break;
641		case ICMP6_MEMBERSHIP_QUERY :
642		case ND_ROUTER_SOLICIT :
643		case ND_NEIGHBOR_SOLICIT :
644		case ICMP6_NI_QUERY :
645			is->is_icmp.ics_type = ic->icmp_type;
646			break;
647		default :
648			return NULL;
649		}
650		ATOMIC_INCL(ips_stats.iss_icmp);
651		is->is_age = fr_icmptimeout;
652		break;
653#endif
654	case IPPROTO_ICMP :
655		ic = (struct icmp *)fin->fin_dp;
656
657		switch (ic->icmp_type)
658		{
659		case ICMP_ECHO :
660		case ICMP_TSTAMP :
661		case ICMP_IREQ :
662		case ICMP_MASKREQ :
663			is->is_icmp.ics_type = ic->icmp_type;
664			hv += (is->is_icmp.ics_id = ic->icmp_id);
665			hv += (is->is_icmp.ics_seq = ic->icmp_seq);
666			break;
667		default :
668			return NULL;
669		}
670		ATOMIC_INCL(ips_stats.iss_icmp);
671		is->is_age = fr_icmptimeout;
672		break;
673	case IPPROTO_TCP :
674		tcp = (tcphdr_t *)fin->fin_dp;
675
676		if (tcp->th_flags & TH_RST)
677			return NULL;
678		/*
679		 * The endian of the ports doesn't matter, but the ack and
680		 * sequence numbers do as we do mathematics on them later.
681		 */
682		is->is_sport = htons(fin->fin_data[0]);
683		is->is_dport = htons(fin->fin_data[1]);
684		if ((flags & (FI_W_DPORT|FI_W_SPORT)) == 0) {
685			hv += is->is_sport;
686			hv += is->is_dport;
687		}
688		if ((flags & FI_IGNOREPKT) == 0) {
689			is->is_send = ntohl(tcp->th_seq) + fin->fin_dlen -
690				      (off = (tcp->th_off << 2)) +
691				      ((tcp->th_flags & TH_SYN) ? 1 : 0) +
692				      ((tcp->th_flags & TH_FIN) ? 1 : 0);
693			is->is_maxsend = is->is_send;
694
695			if ((tcp->th_flags & TH_SYN) &&
696			    ((tcp->th_off << 2) >= (sizeof(*tcp) + 4))) {
697				ws = fr_tcpoptions(tcp);
698				if (ws >= 0)
699					is->is_swscale = ws;
700			}
701		}
702
703		is->is_maxdwin = 1;
704		is->is_maxswin = ntohs(tcp->th_win);
705		if (is->is_maxswin == 0)
706			is->is_maxswin = 1;
707
708		if ((tcp->th_flags & TH_OPENING) == TH_SYN)
709			is->is_fsm = 1;
710
711		/*
712		 * If we're creating state for a starting connection, start the
713		 * timer on it as we'll never see an error if it fails to
714		 * connect.
715		 */
716		ATOMIC_INCL(ips_stats.iss_tcp);
717		break;
718
719	case IPPROTO_UDP :
720		tcp = (tcphdr_t *)fin->fin_dp;
721
722		is->is_sport = htons(fin->fin_data[0]);
723		is->is_dport = htons(fin->fin_data[1]);
724		if ((flags & (FI_W_DPORT|FI_W_SPORT)) == 0) {
725			hv += is->is_sport;
726			hv += is->is_dport;
727		}
728		ATOMIC_INCL(ips_stats.iss_udp);
729		is->is_age = fr_udptimeout;
730		break;
731	default :
732		is->is_age = fr_udptimeout;
733		break;
734	}
735
736	KMALLOC(is, ipstate_t *);
737	if (is == NULL) {
738		ATOMIC_INCL(ips_stats.iss_nomem);
739		return NULL;
740	}
741	bcopy((char *)&ips, (char *)is, sizeof(*is));
742	hv %= fr_statesize;
743	is->is_hv = hv;
744	is->is_rule = fin->fin_fr;
745	if (is->is_rule != NULL) {
746		is->is_group = is->is_rule->fr_group;
747		ATOMIC_INC32(is->is_rule->fr_ref);
748		pass = is->is_rule->fr_flags;
749		is->is_frage[0] = is->is_rule->fr_age[0];
750		is->is_frage[1] = is->is_rule->fr_age[1];
751		if (is->is_frage[0] != 0)
752			is->is_age = is->is_frage[0];
753
754		is->is_ifp[(out << 1) + 1] = is->is_rule->fr_ifas[1];
755		is->is_ifp[(1 - out) << 1] = is->is_rule->fr_ifas[2];
756		is->is_ifp[((1 - out) << 1) + 1] = is->is_rule->fr_ifas[3];
757
758		if (((ifp = is->is_rule->fr_ifas[1]) != NULL) &&
759		    (ifp != (void *)-1))
760			strncpy(is->is_ifname[(out << 1) + 1],
761				IFNAME(ifp), IFNAMSIZ);
762		if (((ifp = is->is_rule->fr_ifas[2]) != NULL) &&
763		    (ifp != (void *)-1))
764			strncpy(is->is_ifname[(1 - out) << 1],
765				IFNAME(ifp), IFNAMSIZ);
766		if (((ifp = is->is_rule->fr_ifas[3]) != NULL) &&
767		    (ifp != (void *)-1))
768			strncpy(is->is_ifname[((1 - out) << 1) + 1],
769				IFNAME(ifp), IFNAMSIZ);
770	} else
771		pass = fr_flags;
772
773	is->is_ifp[out << 1] = fin->fin_ifp;
774	strncpy(is->is_ifname[out << 1], IFNAME(fin->fin_ifp), IFNAMSIZ);
775
776	WRITE_ENTER(&ipf_state);
777
778	is->is_pass = pass;
779	if ((flags & FI_IGNOREPKT) == 0) {
780		is->is_pkts = 1;
781		is->is_bytes = fin->fin_dlen + fin->fin_hlen;
782	}
783	/*
784	 * We want to check everything that is a property of this packet,
785	 * but we don't (automatically) care about it's fragment status as
786	 * this may change.
787	 */
788	is->is_v = fin->fin_v;
789	is->is_rulen = fin->fin_rule;
790	is->is_opt = fin->fin_fi.fi_optmsk;
791	is->is_optmsk = 0xffffffff;
792	is->is_sec = fin->fin_fi.fi_secmsk;
793	is->is_secmsk = 0xffff;
794	is->is_auth = fin->fin_fi.fi_auth;
795	is->is_authmsk = 0xffff;
796	is->is_flags = fin->fin_fl & FI_CMP;
797	is->is_flags |= FI_CMP << 4;
798	is->is_flags |= flags & (FI_WILDP|FI_WILDA);
799	if (flags & (FI_WILDP|FI_WILDA))
800		ips_wild++;
801
802	if (pass & FR_LOGFIRST)
803		is->is_pass &= ~(FR_LOGFIRST|FR_LOG);
804	fr_stinsert(is);
805	is->is_me = stsave;
806	if (is->is_p == IPPROTO_TCP) {
807		fr_tcp_age(&is->is_age, is->is_state, fin,
808			   0, is->is_fsm); /* 0 = packet from the source */
809	}
810#ifdef	IPFILTER_LOG
811	ipstate_log(is, ISL_NEW);
812#endif
813	RWLOCK_EXIT(&ipf_state);
814	fin->fin_rev = IP6NEQ(is->is_dst, fin->fin_fi.fi_dst);
815	if ((fin->fin_fl & FI_FRAG) && (pass & FR_KEEPFRAG))
816		ipfr_newfrag(ip, fin);
817	return is;
818}
819
820
821static int fr_tcpoptions(tcp)
822tcphdr_t *tcp;
823{
824	u_char *opt, *last;
825	int wscale;
826
827	opt = (u_char *) (tcp + 1);
828	last = ((u_char *)tcp) + (tcp->th_off << 2);
829
830	/* If we don't find wscale here, we need to clear it */
831	wscale = -2;
832
833	/* Termination condition picked such that opt[0 .. 2] exist */
834	while ((opt < last - 2)  && (*opt != TCPOPT_EOL)) {
835		switch (*opt) {
836		case TCPOPT_NOP:
837			opt++;
838			continue;
839		case TCPOPT_WSCALE:
840			/* Proper length ? */
841			if (opt[1] == 3) {
842				if (opt[2] > 14)
843					wscale = 14;
844				else
845					wscale = opt[2];
846			}
847			break;
848		default:
849			/* Unknown options must be two bytes+ */
850			if (opt[1] < 2)
851				break;
852			opt += opt[1];
853			continue;
854		}
855		break;
856	}
857	return wscale;
858}
859
860
861
862/*
863 * check to see if a packet with TCP headers fits within the TCP window.
864 * change timeout depending on whether new packet is a SYN-ACK returning for a
865 * SYN or a RST or FIN which indicate time to close up shop.
866 */
867int fr_tcpstate(is, fin, ip, tcp)
868register ipstate_t *is;
869fr_info_t *fin;
870ip_t *ip;
871tcphdr_t *tcp;
872{
873	register tcp_seq seq, ack, end;
874	register int ackskew;
875	tcpdata_t  *fdata, *tdata;
876	u_32_t	win, maxwin;
877	int ret = 0, off;
878	int source;
879	int wscale;
880
881	/*
882	 * Find difference between last checked packet and this packet.
883	 */
884	source = IP6EQ(fin->fin_fi.fi_src, is->is_src);
885	if (source && (ntohs(is->is_sport) != fin->fin_data[0]))
886		source = 0;
887	fdata = &is->is_tcp.ts_data[!source];
888	tdata = &is->is_tcp.ts_data[source];
889	off = tcp->th_off << 2;
890	seq = ntohl(tcp->th_seq);
891	ack = ntohl(tcp->th_ack);
892	win = ntohs(tcp->th_win);
893	end = seq + fin->fin_dlen - off +
894	       ((tcp->th_flags & TH_SYN) ? 1 : 0) +
895	       ((tcp->th_flags & TH_FIN) ? 1 : 0);
896
897
898	if ((tcp->th_flags & TH_SYN) && (off >= sizeof(*tcp) + 4))
899		wscale = fr_tcpoptions(tcp);
900	else
901		wscale = -1;
902
903	MUTEX_ENTER(&is->is_lock);
904
905	if (wscale >= 0)
906		fdata->td_wscale = wscale;
907	else if (wscale == -2)
908		fdata->td_wscale = tdata->td_wscale = 0;
909	win <<= fdata->td_wscale;
910
911	if ((fdata->td_end == 0) &&
912	    (!is->is_fsm || ((tcp->th_flags & TH_OPENING) == TH_OPENING))) {
913		/*
914		 * Must be a (outgoing) SYN-ACK in reply to a SYN.
915		 */
916		fdata->td_end = end;
917		fdata->td_maxwin = 1;
918		fdata->td_maxend = end + win;
919		if (win == 0)
920			fdata->td_maxend++;
921	}
922
923	if (!(tcp->th_flags & TH_ACK)) {  /* Pretend an ack was sent */
924		ack = tdata->td_end;
925	} else if (((tcp->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) &&
926		   (ack == 0)) {
927		/* gross hack to get around certain broken tcp stacks */
928		ack = tdata->td_end;
929	}
930
931	if (seq == end)
932		seq = end = fdata->td_end;
933
934	maxwin = tdata->td_maxwin;
935	ackskew = tdata->td_end - ack;
936
937#define	SEQ_GE(a,b)	((int)((a) - (b)) >= 0)
938#define	SEQ_GT(a,b)	((int)((a) - (b)) > 0)
939	if ((SEQ_GE(fdata->td_maxend, end)) &&
940	    (SEQ_GE(seq, fdata->td_end - maxwin)) &&
941/* XXX what about big packets */
942#define MAXACKWINDOW 66000
943	    (ackskew >= -MAXACKWINDOW) &&
944	    (ackskew <= MAXACKWINDOW)) {
945		/* if ackskew < 0 then this should be due to fragented
946		 * packets. There is no way to know the length of the
947		 * total packet in advance.
948		 * We do know the total length from the fragment cache though.
949		 * Note however that there might be more sessions with
950		 * exactly the same source and destination paramters in the
951		 * state cache (and source and destination is the only stuff
952		 * that is saved in the fragment cache). Note further that
953		 * some TCP connections in the state cache are hashed with
954		 * sport and dport as well which makes it not worthwhile to
955		 * look for them.
956		 * Thus, when ackskew is negative but still seems to belong
957		 * to this session, we bump up the destinations end value.
958		 */
959		/*
960		 * Nearing end of connection, start timeout.
961		 */
962		/* source ? 0 : 1 -> !source */
963		if (fr_tcp_age(&is->is_age, is->is_state, fin, !source,
964			       (int)is->is_fsm) == 0) {
965			if (ackskew < 0)
966				tdata->td_end = ack;
967
968			/* update max window seen */
969			if (fdata->td_maxwin < win)
970				fdata->td_maxwin = win;
971			if (SEQ_GT(end, fdata->td_end))
972				fdata->td_end = end;
973			if (SEQ_GE(ack + win, tdata->td_maxend)) {
974				tdata->td_maxend = ack + win;
975				if (win == 0)
976					tdata->td_maxend++;
977			}
978
979			ATOMIC_INCL(ips_stats.iss_hits);
980			ret = 1;
981		}
982	}
983	MUTEX_EXIT(&is->is_lock);
984	if ((ret == 0) && ((tcp->th_flags & TH_OPENING) != TH_SYN))
985		fin->fin_misc |= FM_BADSTATE;
986	return ret;
987}
988
989
990/*
991 * Match a state table entry against an IP packet.
992 */
993static int fr_matchsrcdst(is, src, dst, fin, tcp)
994ipstate_t *is;
995union i6addr src, dst;
996fr_info_t *fin;
997tcphdr_t *tcp;
998{
999	int ret = 0, rev, out, flags, idx;
1000	u_short sp, dp;
1001	void *ifp;
1002
1003	rev = IP6NEQ(is->is_dst, dst);
1004	ifp = fin->fin_ifp;
1005	out = fin->fin_out;
1006	flags = is->is_flags & (FI_WILDA|FI_WILDP);
1007	sp = 0;
1008	dp = 0;
1009
1010	if (tcp != NULL) {
1011		flags = is->is_flags;
1012		sp = tcp->th_sport;
1013		dp = tcp->th_dport;
1014		if (!rev) {
1015			if (!(flags & FI_W_SPORT) && (sp != is->is_sport))
1016				rev = 1;
1017			else if (!(flags & FI_W_DPORT) && (dp != is->is_dport))
1018				rev = 1;
1019		}
1020	}
1021
1022	idx = (out << 1) + rev;
1023
1024	if ((is->is_ifp[idx] == NULL &&
1025	     (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) ||
1026	    is->is_ifp[idx] == ifp)
1027		ret = 1;
1028
1029	if (ret == 0)
1030		return 0;
1031	ret = 0;
1032
1033	if (rev == 0) {
1034		if ((IP6EQ(is->is_dst, dst) || (flags & FI_W_DADDR)) &&
1035		    (IP6EQ(is->is_src, src) || (flags & FI_W_SADDR)) &&
1036		    (!tcp || ((sp == is->is_sport || flags & FI_W_SPORT) &&
1037		     (dp == is->is_dport || flags & FI_W_DPORT)))) {
1038			ret = 1;
1039		}
1040	} else {
1041		if ((IP6EQ(is->is_dst, src) || (flags & FI_W_DADDR)) &&
1042		    (IP6EQ(is->is_src, dst) || (flags & FI_W_SADDR)) &&
1043		    (!tcp || ((sp == is->is_dport || flags & FI_W_DPORT) &&
1044		     (dp == is->is_sport || flags & FI_W_SPORT)))) {
1045			ret = 1;
1046		}
1047	}
1048	if (ret == 0)
1049		return 0;
1050
1051	/*
1052	 * Whether or not this should be here, is questionable, but the aim
1053	 * is to get this out of the main line.
1054	 */
1055	if (tcp == NULL)
1056		flags = is->is_flags & (FI_CMP|(FI_CMP<<4));
1057
1058	if (((fin->fin_fl & (flags >> 4)) != (flags & FI_CMP)) ||
1059	    (fin->fin_fi.fi_optmsk != is->is_opt) ||
1060	    (fin->fin_fi.fi_secmsk != is->is_sec) ||
1061	    (fin->fin_fi.fi_auth != is->is_auth))
1062		return 0;
1063
1064	flags = is->is_flags & (FI_WILDA|FI_WILDP);
1065	if ((flags & (FI_W_SADDR|FI_W_DADDR))) {
1066		if ((flags & FI_W_SADDR) != 0) {
1067			if (rev == 0) {
1068				is->is_src = fin->fin_fi.fi_src;
1069			} else {
1070				is->is_src = fin->fin_fi.fi_dst;
1071			}
1072		} else if ((flags & FI_W_DADDR) != 0) {
1073			if (rev == 0) {
1074				is->is_dst = fin->fin_fi.fi_dst;
1075			} else {
1076				is->is_dst = fin->fin_fi.fi_src;
1077			}
1078		}
1079		is->is_flags &= ~(FI_W_SADDR|FI_W_DADDR);
1080		if ((is->is_flags & (FI_WILDA|FI_WILDP)) == 0)
1081			ips_wild--;
1082	}
1083
1084	if ((flags & (FI_W_SPORT|FI_W_DPORT))) {
1085		if ((flags & FI_W_SPORT) != 0) {
1086			if (rev == 0) {
1087				is->is_sport = sp;
1088				is->is_send = htonl(tcp->th_seq);
1089			} else {
1090				is->is_sport = dp;
1091				is->is_send = htonl(tcp->th_ack);
1092			}
1093			is->is_maxsend = is->is_send + 1;
1094		} else if ((flags & FI_W_DPORT) != 0) {
1095			if (rev == 0) {
1096				is->is_dport = dp;
1097				is->is_dend = htonl(tcp->th_ack);
1098			} else {
1099				is->is_dport = sp;
1100				is->is_dend = htonl(tcp->th_seq);
1101			}
1102			is->is_maxdend = is->is_dend + 1;
1103		}
1104		is->is_flags &= ~(FI_W_SPORT|FI_W_DPORT);
1105		ips_wild--;
1106	}
1107
1108	ret = -1;
1109
1110	if (is->is_ifp[idx] == NULL &&
1111	    (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*'))
1112		ret = idx;
1113
1114	if (ret >= 0) {
1115		is->is_ifp[ret] = ifp;
1116		strncpy(is->is_ifname[ret], IFNAME(ifp),
1117			sizeof(is->is_ifname[ret]));
1118	}
1119	fin->fin_rev = rev;
1120	return 1;
1121}
1122
1123static int fr_matchicmpqueryreply(v, is, icmp)
1124int v;
1125ipstate_t *is;
1126icmphdr_t *icmp;
1127{
1128	if (v == 4) {
1129		/*
1130		 * If we matched its type on the way in, then when going out
1131		 * it will still be the same type.
1132		 */
1133		if (((icmp->icmp_type == is->is_type) ||
1134		     (icmpreplytype4[is->is_type] == icmp->icmp_type))) {
1135			if (icmp->icmp_type != ICMP_ECHOREPLY)
1136				return 1;
1137			if ((icmp->icmp_id == is->is_icmp.ics_id) &&
1138			    (icmp->icmp_seq == is->is_icmp.ics_seq))
1139				return 1;
1140		}
1141	}
1142#ifdef	USE_INET6
1143	else if (is->is_v == 6) {
1144		if (((icmp->icmp_type == is->is_type) ||
1145		     (icmpreplytype6[is->is_type] == icmp->icmp_type))) {
1146			if (icmp->icmp_type != ICMP6_ECHO_REPLY)
1147				return 1;
1148			if ((icmp->icmp_id == is->is_icmp.ics_id) &&
1149			    (icmp->icmp_seq == is->is_icmp.ics_seq))
1150				return 1;
1151		}
1152	}
1153#endif
1154	return 0;
1155}
1156
1157static frentry_t *fr_checkicmpmatchingstate(ip, fin)
1158ip_t *ip;
1159fr_info_t *fin;
1160{
1161	register ipstate_t *is, **isp;
1162	register u_short sport, dport;
1163	register u_char	pr;
1164	u_short savelen, ohlen;
1165	union i6addr dst, src;
1166	struct icmp *ic;
1167	icmphdr_t *icmp;
1168	fr_info_t ofin;
1169	int type, len;
1170	tcphdr_t *tcp;
1171	frentry_t *fr;
1172	ip_t *oip;
1173	u_int hv;
1174
1175	/*
1176	 * Does it at least have the return (basic) IP header ?
1177	 * Only a basic IP header (no options) should be with
1178	 * an ICMP error header.
1179	 */
1180	if (((ip->ip_v != 4) || (ip->ip_hl != 5)) ||
1181	    (fin->fin_plen < ICMPERR_MINPKTLEN))
1182		return NULL;
1183
1184	ic = (struct icmp *)fin->fin_dp;
1185	type = ic->icmp_type;
1186	/*
1187	 * If it's not an error type, then return
1188	 */
1189	if ((type != ICMP_UNREACH) && (type != ICMP_SOURCEQUENCH) &&
1190    	    (type != ICMP_REDIRECT) && (type != ICMP_TIMXCEED) &&
1191    	    (type != ICMP_PARAMPROB))
1192		return NULL;
1193
1194	oip = (ip_t *)((char *)ic + ICMPERR_ICMPHLEN);
1195	ohlen = oip->ip_hl << 2;
1196	if (fin->fin_plen < ICMPERR_MAXPKTLEN + ohlen - sizeof(*oip))
1197		return NULL;
1198
1199	/*
1200	 * Sanity checks.
1201	 */
1202	len = fin->fin_dlen - ICMPERR_ICMPHLEN;
1203	if ((len <= 0) || (ohlen > len))
1204		return NULL;
1205
1206	/*
1207	 * Is the buffer big enough for all of it ?  It's the size of the IP
1208	 * header claimed in the encapsulated part which is of concern.  It
1209	 * may be too big to be in this buffer but not so big that it's
1210	 * outside the ICMP packet, leading to TCP deref's causing problems.
1211	 * This is possible because we don't know how big oip_hl is when we
1212	 * do the pullup early in fr_check() and thus can't gaurantee it is
1213	 * all here now.
1214	 */
1215#ifdef  _KERNEL
1216	{
1217	mb_t *m;
1218
1219# if SOLARIS
1220	m = fin->fin_qfm;
1221	if ((char *)oip + len > (char *)m->b_wptr)
1222		return NULL;
1223# else
1224	m = *(mb_t **)fin->fin_mp;
1225	if ((char *)oip + len > (char *)ip + m->m_len)
1226		return NULL;
1227# endif
1228	}
1229#endif
1230
1231	/*
1232	 * in the IPv4 case we must zero the i6addr union otherwise
1233	 * the IP6EQ and IP6NEQ macros produce the wrong results because
1234	 * of the 'junk' in the unused part of the union
1235	 */
1236	bzero((char *)&src, sizeof(src));
1237	bzero((char *)&dst, sizeof(dst));
1238	bzero((char *)&ofin, sizeof(ofin));
1239	ofin.fin_ifp = fin->fin_ifp;
1240	ofin.fin_out = !fin->fin_out;
1241	ofin.fin_v = 4;
1242	fr = NULL;
1243
1244	switch (oip->ip_p)
1245	{
1246	case IPPROTO_ICMP :
1247		icmp = (icmphdr_t *)((char *)oip + ohlen);
1248
1249		/*
1250		 * a ICMP error can only be generated as a result of an
1251		 * ICMP query, not as the response on an ICMP error
1252		 *
1253		 * XXX theoretically ICMP_ECHOREP and the other reply's are
1254		 * ICMP query's as well, but adding them here seems strange XXX
1255		 */
1256		 if ((icmp->icmp_type != ICMP_ECHO) &&
1257		     (icmp->icmp_type != ICMP_TSTAMP) &&
1258		     (icmp->icmp_type != ICMP_IREQ) &&
1259		     (icmp->icmp_type != ICMP_MASKREQ))
1260		    	return NULL;
1261
1262		/*
1263		 * perform a lookup of the ICMP packet in the state table
1264		 */
1265		hv = (pr = oip->ip_p);
1266		src.in4 = oip->ip_src;
1267		hv += src.in4.s_addr;
1268		dst.in4 = oip->ip_dst;
1269		hv += dst.in4.s_addr;
1270		hv += icmp->icmp_id;
1271		hv += icmp->icmp_seq;
1272		hv %= fr_statesize;
1273
1274		savelen = oip->ip_len;
1275		oip->ip_len = len;
1276		fr_makefrip(ohlen, oip, &ofin);
1277		oip->ip_len = savelen;
1278
1279		READ_ENTER(&ipf_state);
1280		for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext)
1281			if ((is->is_p == pr) && (is->is_v == 4) &&
1282			    fr_matchsrcdst(is, src, dst, &ofin, NULL) &&
1283			    fr_matchicmpqueryreply(is->is_v, is, icmp)) {
1284				ips_stats.iss_hits++;
1285				is->is_pkts++;
1286				is->is_bytes += ip->ip_len;
1287				fr = is->is_rule;
1288				break;
1289			}
1290		RWLOCK_EXIT(&ipf_state);
1291		return fr;
1292
1293	case IPPROTO_TCP :
1294	case IPPROTO_UDP :
1295		if (fin->fin_plen < ICMPERR_MAXPKTLEN)
1296			return NULL;
1297		break;
1298	default :
1299		return NULL;
1300	}
1301
1302	tcp = (tcphdr_t *)((char *)oip + ohlen);
1303	dport = tcp->th_dport;
1304	sport = tcp->th_sport;
1305
1306	hv = (pr = oip->ip_p);
1307	src.in4 = oip->ip_src;
1308	hv += src.in4.s_addr;
1309	dst.in4 = oip->ip_dst;
1310	hv += dst.in4.s_addr;
1311	hv += dport;
1312	hv += sport;
1313	hv %= fr_statesize;
1314	/*
1315	 * we make an fin entry to be able to feed it to
1316	 * matchsrcdst note that not all fields are encessary
1317	 * but this is the cleanest way. Note further we fill
1318	 * in fin_mp such that if someone uses it we'll get
1319	 * a kernel panic. fr_matchsrcdst does not use this.
1320	 *
1321	 * watch out here, as ip is in host order and oip in network
1322	 * order. Any change we make must be undone afterwards.
1323	 */
1324	savelen = oip->ip_len;
1325	oip->ip_len = len;
1326	fr_makefrip(ohlen, oip, &ofin);
1327	oip->ip_len = savelen;
1328	READ_ENTER(&ipf_state);
1329	for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) {
1330		/*
1331		 * Only allow this icmp though if the
1332		 * encapsulated packet was allowed through the
1333		 * other way around. Note that the minimal amount
1334		 * of info present does not allow for checking against
1335		 * tcp internals such as seq and ack numbers.
1336		 */
1337		if ((is->is_p == pr) && (is->is_v == 4) &&
1338		    fr_matchsrcdst(is, src, dst, &ofin, tcp)) {
1339			fr = is->is_rule;
1340			ips_stats.iss_hits++;
1341			is->is_pkts++;
1342			is->is_bytes += fin->fin_plen;
1343			/*
1344			 * we deliberately do not touch the timeouts
1345			 * for the accompanying state table entry.
1346			 * It remains to be seen if that is correct. XXX
1347			 */
1348			break;
1349		}
1350	}
1351	RWLOCK_EXIT(&ipf_state);
1352	return fr;
1353}
1354
1355
1356/*
1357 * Move a state hash table entry from its old location at is->is_hv to
1358 * its new location, indexed by hv % fr_statesize.
1359 */
1360static void fr_ipsmove(isp, is, hv)
1361ipstate_t **isp, *is;
1362u_int hv;
1363{
1364	u_int hvm;
1365
1366	hvm = is->is_hv;
1367	/*
1368	 * Remove the hash from the old location...
1369	 */
1370	if (is->is_hnext)
1371		is->is_hnext->is_phnext = isp;
1372	*isp = is->is_hnext;
1373	if (ips_table[hvm] == NULL)
1374		ips_stats.iss_inuse--;
1375
1376	/*
1377	 * ...and put the hash in the new one.
1378	 */
1379	hvm = hv % fr_statesize;
1380	is->is_hv = hvm;
1381	isp = &ips_table[hvm];
1382	if (*isp)
1383		(*isp)->is_phnext = &is->is_hnext;
1384	else
1385		ips_stats.iss_inuse++;
1386	is->is_phnext = isp;
1387	is->is_hnext = *isp;
1388	*isp = is;
1389}
1390
1391
1392/*
1393 * Check if a packet has a registered state.
1394 */
1395frentry_t *fr_checkstate(ip, fin)
1396ip_t *ip;
1397fr_info_t *fin;
1398{
1399	union i6addr dst, src;
1400	register ipstate_t *is, **isp;
1401	register u_char pr;
1402	u_int hv, hvm, hlen, tryagain, pass, v;
1403	struct icmp *ic;
1404	frentry_t *fr;
1405	tcphdr_t *tcp;
1406	int rev;
1407
1408	if ((ips_list == NULL) || (fin->fin_off != 0) || fr_state_lock ||
1409	    (fin->fin_fl & FI_SHORT))
1410		return NULL;
1411
1412	is = NULL;
1413	hlen = fin->fin_hlen;
1414	tcp = (tcphdr_t *)((char *)ip + hlen);
1415	ic = (struct icmp *)tcp;
1416	hv = (pr = fin->fin_fi.fi_p);
1417	src = fin->fin_fi.fi_src;
1418	dst = fin->fin_fi.fi_dst;
1419	hv += src.in4.s_addr;
1420	hv += dst.in4.s_addr;
1421
1422	/*
1423	 * Search the hash table for matching packet header info.
1424	 * At the bottom of this switch statement, the following is expected:
1425	 * is == NULL, no lock on ipf_state is held.
1426	 * is != NULL, a lock on ipf_state is held.
1427	 */
1428	v = fin->fin_fi.fi_v;
1429#ifdef	USE_INET6
1430	if (v == 6) {
1431		hv += fin->fin_fi.fi_src.i6[1];
1432		hv += fin->fin_fi.fi_src.i6[2];
1433		hv += fin->fin_fi.fi_src.i6[3];
1434
1435		if ((fin->fin_p == IPPROTO_ICMPV6) &&
1436		    IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_dst.in6)) {
1437			hv -= dst.in4.s_addr;
1438		} else {
1439			hv += fin->fin_fi.fi_dst.i6[1];
1440			hv += fin->fin_fi.fi_dst.i6[2];
1441			hv += fin->fin_fi.fi_dst.i6[3];
1442		}
1443	}
1444#endif
1445
1446	switch (fin->fin_p)
1447	{
1448#ifdef	USE_INET6
1449	case IPPROTO_ICMPV6 :
1450		tcp = NULL;
1451		tryagain = 0;
1452		if (v == 6) {
1453			if ((ic->icmp_type == ICMP6_ECHO_REQUEST) ||
1454			    (ic->icmp_type == ICMP6_ECHO_REPLY)) {
1455				hv += ic->icmp_id;
1456				hv += ic->icmp_seq;
1457			}
1458		}
1459		READ_ENTER(&ipf_state);
1460icmp6again:
1461		hvm = hv % fr_statesize;
1462		for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext)
1463			if ((is->is_p == pr) && (is->is_v == v) &&
1464			    fr_matchsrcdst(is, src, dst, fin, NULL) &&
1465			    fr_matchicmpqueryreply(v, is, ic)) {
1466				rev = fin->fin_rev;
1467				if (is->is_frage[rev] != 0)
1468					is->is_age = is->is_frage[rev];
1469				else if (rev != 0)
1470					is->is_age = fr_icmpacktimeout;
1471				else
1472					is->is_age = fr_icmptimeout;
1473				break;
1474			}
1475
1476		if (is != NULL) {
1477			if (tryagain && !(is->is_flags & FI_W_DADDR)) {
1478				hv += fin->fin_fi.fi_src.i6[0];
1479				hv += fin->fin_fi.fi_src.i6[1];
1480				hv += fin->fin_fi.fi_src.i6[2];
1481				hv += fin->fin_fi.fi_src.i6[3];
1482				fr_ipsmove(isp, is, hv);
1483				MUTEX_DOWNGRADE(&ipf_state);
1484			}
1485			break;
1486		}
1487		RWLOCK_EXIT(&ipf_state);
1488
1489		/*
1490		 * No matching icmp state entry. Perhaps this is a
1491		 * response to another state entry.
1492		 */
1493		if ((ips_wild != 0) && (v == 6) && (tryagain == 0) &&
1494		    !IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_src.in6)) {
1495			hv -= fin->fin_fi.fi_src.i6[0];
1496			hv -= fin->fin_fi.fi_src.i6[1];
1497			hv -= fin->fin_fi.fi_src.i6[2];
1498			hv -= fin->fin_fi.fi_src.i6[3];
1499			tryagain = 1;
1500			WRITE_ENTER(&ipf_state);
1501			goto icmp6again;
1502		}
1503
1504		fr = fr_checkicmp6matchingstate((ip6_t *)ip, fin);
1505		if (fr)
1506			return fr;
1507		break;
1508#endif
1509	case IPPROTO_ICMP :
1510		tcp = NULL;
1511		if (v == 4) {
1512			hv += ic->icmp_id;
1513			hv += ic->icmp_seq;
1514		}
1515		hvm = hv % fr_statesize;
1516		READ_ENTER(&ipf_state);
1517		for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext)
1518			if ((is->is_p == pr) && (is->is_v == v) &&
1519			    fr_matchsrcdst(is, src, dst, fin, NULL) &&
1520			    fr_matchicmpqueryreply(v, is, ic)) {
1521				rev = fin->fin_rev;
1522				if (is->is_frage[rev] != 0)
1523					is->is_age = is->is_frage[rev];
1524				else if (fin->fin_rev)
1525					is->is_age = fr_icmpacktimeout;
1526				else
1527					is->is_age = fr_icmptimeout;
1528				break;
1529			}
1530
1531		if (is != NULL)
1532			break;
1533		RWLOCK_EXIT(&ipf_state);
1534		/*
1535		 * No matching icmp state entry. Perhaps this is a
1536		 * response to another state entry.
1537		 */
1538		fr = fr_checkicmpmatchingstate(ip, fin);
1539		if (fr)
1540			return fr;
1541		break;
1542	case IPPROTO_TCP :
1543		/*
1544		 * Just plain ignore RST flag set with either FIN or SYN.
1545		 */
1546		if ((tcp->th_flags & TH_RST) &&
1547		    ((tcp->th_flags & (TH_FIN|TH_SYN|TH_RST)) != TH_RST))
1548			break;
1549	case IPPROTO_UDP :
1550	    {
1551		register u_short dport, sport;
1552
1553		dport = tcp->th_dport;
1554		sport = tcp->th_sport;
1555		tryagain = 0;
1556		hv += dport;
1557		hv += sport;
1558		READ_ENTER(&ipf_state);
1559retry_tcpudp:
1560		hvm = hv % fr_statesize;
1561		for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext)
1562			if ((is->is_p == pr) && (is->is_v == v) &&
1563			    fr_matchsrcdst(is, src, dst, fin, tcp)) {
1564				rev = fin->fin_rev;
1565				if ((pr == IPPROTO_TCP)) {
1566					if (!fr_tcpstate(is, fin, ip, tcp))
1567						is = NULL;
1568				} else if ((pr == IPPROTO_UDP)) {
1569					if (is->is_frage[rev] != 0)
1570						is->is_age = is->is_frage[rev];
1571					else if (fin->fin_rev)
1572						is->is_age = fr_udpacktimeout;
1573					else
1574						is->is_age = fr_udptimeout;
1575				}
1576				break;
1577			}
1578		if (is != NULL) {
1579			if (tryagain &&
1580			    !(is->is_flags & (FI_WILDP|FI_WILDA))) {
1581				hv += dport;
1582				hv += sport;
1583				fr_ipsmove(isp, is, hv);
1584				MUTEX_DOWNGRADE(&ipf_state);
1585			}
1586			break;
1587		}
1588
1589		RWLOCK_EXIT(&ipf_state);
1590		if (!tryagain && ips_wild) {
1591			hv -= dport;
1592			hv -= sport;
1593			tryagain = 1;
1594			WRITE_ENTER(&ipf_state);
1595			goto retry_tcpudp;
1596		}
1597		break;
1598	    }
1599	default :
1600		tcp = NULL;
1601		hv %= fr_statesize;
1602		READ_ENTER(&ipf_state);
1603		for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) {
1604			if ((is->is_p == pr) && (is->is_v == v) &&
1605			    fr_matchsrcdst(is, src, dst, fin, NULL)) {
1606				rev = fin->fin_rev;
1607				if (is->is_frage[rev] != 0)
1608					is->is_age = is->is_frage[rev];
1609				else
1610					is->is_age = fr_udptimeout;
1611				break;
1612			}
1613		}
1614		if (is == NULL) {
1615			RWLOCK_EXIT(&ipf_state);
1616		}
1617		break;
1618	}
1619
1620	if (is == NULL) {
1621		ATOMIC_INCL(ips_stats.iss_miss);
1622		return NULL;
1623	}
1624
1625	MUTEX_ENTER(&is->is_lock);
1626	is->is_bytes += fin->fin_plen;
1627	ips_stats.iss_hits++;
1628	is->is_pkts++;
1629	MUTEX_EXIT(&is->is_lock);
1630	fr = is->is_rule;
1631	fin->fin_rule = is->is_rulen;
1632	if (fr != NULL) {
1633		fin->fin_group = fr->fr_group;
1634		fin->fin_icode = fr->fr_icode;
1635	}
1636	fin->fin_fr = fr;
1637	pass = is->is_pass;
1638	RWLOCK_EXIT(&ipf_state);
1639	if ((fin->fin_fl & FI_FRAG) && (pass & FR_KEEPFRAG))
1640		ipfr_newfrag(ip, fin);
1641#ifndef	_KERNEL
1642	if ((tcp != NULL) && (tcp->th_flags & TCP_CLOSE))
1643		fr_delstate(is);
1644#endif
1645	return fr;
1646}
1647
1648
1649/*
1650 * Sync. state entries.  If interfaces come or go or just change position,
1651 * this is needed.
1652 */
1653void ip_statesync(ifp)
1654void *ifp;
1655{
1656	register ipstate_t *is;
1657	int i;
1658
1659	WRITE_ENTER(&ipf_state);
1660	for (is = ips_list; is; is = is->is_next) {
1661		for (i = 0; i < 4; i++) {
1662			if (is->is_ifp[i] == ifp) {
1663				is->is_ifpin = GETUNIT(is->is_ifname[i],
1664						       is->is_v);
1665				if (!is->is_ifp[i])
1666					is->is_ifp[i] = (void *)-1;
1667			}
1668		}
1669	}
1670	RWLOCK_EXIT(&ipf_state);
1671}
1672
1673
1674/*
1675 * Must always be called with fr_ipfstate held as a write lock.
1676 */
1677static void fr_delstate(is)
1678ipstate_t *is;
1679{
1680	frentry_t *fr;
1681
1682	if (is->is_flags & (FI_WILDP|FI_WILDA))
1683		ips_wild--;
1684	if (is->is_next)
1685		is->is_next->is_pnext = is->is_pnext;
1686	*is->is_pnext = is->is_next;
1687	if (is->is_hnext)
1688		is->is_hnext->is_phnext = is->is_phnext;
1689	*is->is_phnext = is->is_hnext;
1690	if (ips_table[is->is_hv] == NULL)
1691		ips_stats.iss_inuse--;
1692	if (is->is_me)
1693		*is->is_me = NULL;
1694
1695	fr = is->is_rule;
1696	if (fr != NULL) {
1697		fr->fr_ref--;
1698		if (fr->fr_ref == 0) {
1699			KFREE(fr);
1700		}
1701	}
1702#ifdef	_KERNEL
1703	MUTEX_DESTROY(&is->is_lock);
1704#endif
1705	KFREE(is);
1706	ips_num--;
1707}
1708
1709
1710/*
1711 * Free memory in use by all state info. kept.
1712 */
1713void fr_stateunload()
1714{
1715	register ipstate_t *is;
1716
1717	WRITE_ENTER(&ipf_state);
1718	while ((is = ips_list))
1719		fr_delstate(is);
1720	ips_stats.iss_inuse = 0;
1721	ips_num = 0;
1722	RWLOCK_EXIT(&ipf_state);
1723	if (ips_table)
1724		KFREES(ips_table, fr_statesize * sizeof(ipstate_t *));
1725	ips_table = NULL;
1726}
1727
1728
1729/*
1730 * Slowly expire held state for thingslike UDP and ICMP.  Timeouts are set
1731 * in expectation of this being called twice per second.
1732 */
1733void fr_timeoutstate()
1734{
1735	register ipstate_t *is, **isp;
1736#if defined(_KERNEL) && !SOLARIS
1737	int s;
1738#endif
1739
1740	SPL_NET(s);
1741	WRITE_ENTER(&ipf_state);
1742	for (isp = &ips_list; (is = *isp); )
1743		if (is->is_age && !--is->is_age) {
1744			if (is->is_p == IPPROTO_TCP)
1745				ips_stats.iss_fin++;
1746			else
1747				ips_stats.iss_expire++;
1748#ifdef	IPFILTER_LOG
1749			ipstate_log(is, ISL_EXPIRE);
1750#endif
1751			fr_delstate(is);
1752		} else
1753			isp = &is->is_next;
1754	if (fr_state_doflush) {
1755		(void) fr_state_flush(2);
1756		fr_state_doflush = 0;
1757	}
1758	RWLOCK_EXIT(&ipf_state);
1759	SPL_X(s);
1760}
1761
1762
1763/*
1764 * Original idea freom Pradeep Krishnan for use primarily with NAT code.
1765 * (pkrishna@netcom.com)
1766 *
1767 * Rewritten by Arjan de Vet <Arjan.deVet@adv.iae.nl>, 2000-07-29:
1768 *
1769 * - (try to) base state transitions on real evidence only,
1770 *   i.e. packets that are sent and have been received by ipfilter;
1771 *   diagram 18.12 of TCP/IP volume 1 by W. Richard Stevens was used.
1772 *
1773 * - deal with half-closed connections correctly;
1774 *
1775 * - store the state of the source in state[0] such that ipfstat
1776 *   displays the state as source/dest instead of dest/source; the calls
1777 *   to fr_tcp_age have been changed accordingly.
1778 *
1779 * Parameters:
1780 *
1781 *    state[0] = state of source (host that initiated connection)
1782 *    state[1] = state of dest   (host that accepted the connection)
1783 *
1784 *    dir == 0 : a packet from source to dest
1785 *    dir == 1 : a packet from dest to source
1786 *
1787 */
1788int fr_tcp_age(age, state, fin, dir, fsm)
1789u_long *age;
1790u_char *state;
1791fr_info_t *fin;
1792int dir, fsm;
1793{
1794	tcphdr_t *tcp = (tcphdr_t *)fin->fin_dp;
1795	u_char flags = tcp->th_flags;
1796	int dlen, ostate;
1797	u_long newage;
1798
1799	ostate = state[1 - dir];
1800
1801	dlen = fin->fin_plen - fin->fin_hlen - (tcp->th_off << 2);
1802
1803	if (flags & TH_RST) {
1804		if (!(tcp->th_flags & TH_PUSH) && !dlen) {
1805			*age = fr_tcpclosed;
1806			state[dir] = TCPS_CLOSED;
1807		} else {
1808			*age = fr_tcpclosewait;
1809			state[dir] = TCPS_CLOSE_WAIT;
1810		}
1811		return 0;
1812	}
1813
1814	newage = 0;
1815
1816	switch(state[dir])
1817	{
1818	case TCPS_CLOSED: /* 0 */
1819		if ((flags & TH_OPENING) == TH_OPENING) {
1820			/*
1821			 * 'dir' received an S and sends SA in response,
1822			 * CLOSED -> SYN_RECEIVED
1823			 */
1824			state[dir] = TCPS_SYN_RECEIVED;
1825			newage = fr_tcptimeout;
1826		} else if ((flags & TH_OPENING) == TH_SYN) {
1827			/* 'dir' sent S, CLOSED -> SYN_SENT */
1828			state[dir] = TCPS_SYN_SENT;
1829			newage = fr_tcptimeout;
1830		}
1831		/*
1832		 * The next piece of code makes it possible to get
1833		 * already established connections into the state table
1834		 * after a restart or reload of the filter rules; this
1835		 * does not work when a strict 'flags S keep state' is
1836		 * used for tcp connections of course
1837		 */
1838		if (!fsm && (flags & (TH_FIN|TH_SYN|TH_RST|TH_ACK)) == TH_ACK) {
1839			/* we saw an A, guess 'dir' is in ESTABLISHED mode */
1840			if (state[1 - dir] == TCPS_CLOSED ||
1841			    state[1 - dir] == TCPS_ESTABLISHED) {
1842				state[dir] = TCPS_ESTABLISHED;
1843				newage = fr_tcpidletimeout;
1844			}
1845		}
1846		/*
1847		 * TODO: besides regular ACK packets we can have other
1848		 * packets as well; it is yet to be determined how we
1849		 * should initialize the states in those cases
1850		 */
1851		break;
1852
1853	case TCPS_LISTEN: /* 1 */
1854		/* NOT USED */
1855		break;
1856
1857	case TCPS_SYN_SENT: /* 2 */
1858		if ((flags & ~(TH_ECN|TH_CWR)) == TH_SYN) {
1859			/*
1860			 * A retransmitted SYN packet.  We do not reset the
1861			 * timeout here to fr_tcptimeout because a connection
1862			 * connect timeout does not renew after every packet
1863			 * that is sent.  We need to set newage to something
1864			 * to indicate the packet has passed the check for its
1865			 * flags being valid in the TCP FSM.
1866			 */
1867			newage = *age;
1868		} else if ((flags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) {
1869			/*
1870			 * We see an A from 'dir' which is in SYN_SENT
1871			 * state: 'dir' sent an A in response to an SA
1872			 * which it received, SYN_SENT -> ESTABLISHED
1873			 */
1874			state[dir] = TCPS_ESTABLISHED;
1875			newage = fr_tcpidletimeout;
1876		} else if (flags & TH_FIN) {
1877			/*
1878			 * We see an F from 'dir' which is in SYN_SENT
1879			 * state and wants to close its side of the
1880			 * connection; SYN_SENT -> FIN_WAIT_1
1881			 */
1882			state[dir] = TCPS_FIN_WAIT_1;
1883			newage = fr_tcpidletimeout; /* or fr_tcptimeout? */
1884		} else if ((flags & TH_OPENING) == TH_OPENING) {
1885			/*
1886			 * We see an SA from 'dir' which is already in
1887			 * SYN_SENT state, this means we have a
1888			 * simultaneous open; SYN_SENT -> SYN_RECEIVED
1889			 */
1890			state[dir] = TCPS_SYN_RECEIVED;
1891			newage = fr_tcptimeout;
1892		}
1893		break;
1894
1895	case TCPS_SYN_RECEIVED: /* 3 */
1896		if ((flags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) {
1897			/*
1898			 * We see an A from 'dir' which was in SYN_RECEIVED
1899			 * state so it must now be in established state,
1900			 * SYN_RECEIVED -> ESTABLISHED
1901			 */
1902			state[dir] = TCPS_ESTABLISHED;
1903			newage = fr_tcpidletimeout;
1904		} else if ((flags & ~(TH_ECN|TH_CWR)) == TH_OPENING) {
1905			/*
1906			 * We see an SA from 'dir' which is already in
1907			 * SYN_RECEIVED state.
1908			 */
1909			newage = fr_tcptimeout;
1910		} else if (flags & TH_FIN) {
1911			/*
1912			 * We see an F from 'dir' which is in SYN_RECEIVED
1913			 * state and wants to close its side of the connection;
1914			 * SYN_RECEIVED -> FIN_WAIT_1
1915			 */
1916			state[dir] = TCPS_FIN_WAIT_1;
1917			newage = fr_tcpidletimeout;
1918		}
1919		break;
1920
1921	case TCPS_ESTABLISHED: /* 4 */
1922		if (flags & TH_FIN) {
1923			/*
1924			 * 'dir' closed its side of the connection; this
1925			 * gives us a half-closed connection;
1926			 * ESTABLISHED -> FIN_WAIT_1
1927			 */
1928			state[dir] = TCPS_FIN_WAIT_1;
1929			newage = fr_tcphalfclosed;
1930		} else if (flags & TH_ACK) {
1931			/* an ACK, should we exclude other flags here? */
1932			if (ostate == TCPS_FIN_WAIT_1) {
1933				/*
1934				 * We know the other side did an active close,
1935				 * so we are ACKing the recvd FIN packet (does
1936				 * the window matching code guarantee this?)
1937				 * and go into CLOSE_WAIT state; this gives us
1938				 * a half-closed connection
1939				 */
1940				state[dir] = TCPS_CLOSE_WAIT;
1941				newage = fr_tcphalfclosed;
1942			} else if (ostate < TCPS_CLOSE_WAIT)
1943				/*
1944				 * Still a fully established connection,
1945				 * reset timeout
1946				 */
1947				newage = fr_tcpidletimeout;
1948		}
1949		break;
1950
1951	case TCPS_CLOSE_WAIT: /* 5 */
1952		if (flags & TH_FIN) {
1953			/*
1954			 * Application closed and 'dir' sent a FIN, we're now
1955			 * going into LAST_ACK state
1956			 */
1957			newage  = fr_tcplastack;
1958			state[dir] = TCPS_LAST_ACK;
1959		} else {
1960			/*
1961			 * We remain in CLOSE_WAIT because the other side has
1962			 * closed already and we did not close our side yet;
1963			 * reset timeout
1964			 */
1965			newage  = fr_tcphalfclosed;
1966		}
1967		break;
1968
1969	case TCPS_FIN_WAIT_1: /* 6 */
1970		if ((flags & TH_ACK) && ostate > TCPS_CLOSE_WAIT) {
1971			/*
1972			 * If the other side is not active anymore it has sent
1973			 * us a FIN packet that we are ack'ing now with an ACK;
1974			 * this means both sides have now closed the connection
1975			 * and we go into TIME_WAIT
1976			 */
1977			/*
1978			 * XXX: how do we know we really are ACKing the FIN
1979			 * packet here? does the window code guarantee that?
1980			 */
1981			state[dir] = TCPS_TIME_WAIT;
1982			newage = fr_tcptimeout;
1983		} else
1984			/*
1985			 * We closed our side of the connection already but the
1986			 * other side is still active (ESTABLISHED/CLOSE_WAIT);
1987			 * continue with this half-closed connection
1988			 */
1989			newage = fr_tcphalfclosed;
1990		break;
1991
1992	case TCPS_CLOSING: /* 7 */
1993		/* NOT USED */
1994		break;
1995
1996	case TCPS_LAST_ACK: /* 8 */
1997		if (flags & TH_ACK) {
1998			if ((flags & TH_PUSH) || dlen)
1999				/*
2000				 * There is still data to be delivered, reset
2001				 * timeout
2002				 */
2003				newage  = fr_tcplastack;
2004			else
2005				newage = *age;
2006		}
2007		/*
2008		 * We cannot detect when we go out of LAST_ACK state to CLOSED
2009		 * because that is based on the reception of ACK packets;
2010		 * ipfilter can only detect that a packet has been sent by a
2011		 * host
2012		 */
2013		break;
2014
2015	case TCPS_FIN_WAIT_2: /* 9 */
2016		/* NOT USED */
2017		break;
2018
2019	case TCPS_TIME_WAIT: /* 10 */
2020		newage = fr_tcptimeout; /* default 4 mins */
2021		/* we're in 2MSL timeout now */
2022		break;
2023	}
2024
2025	if (newage != 0) {
2026		*age = newage;
2027		return 0;
2028	}
2029	return -1;
2030}
2031
2032
2033#ifdef	IPFILTER_LOG
2034void ipstate_log(is, type)
2035struct ipstate *is;
2036u_int type;
2037{
2038	struct	ipslog	ipsl;
2039	void *items[1];
2040	size_t sizes[1];
2041	int types[1];
2042
2043	ipsl.isl_type = type;
2044	ipsl.isl_pkts = is->is_pkts;
2045	ipsl.isl_bytes = is->is_bytes;
2046	ipsl.isl_src = is->is_src;
2047	ipsl.isl_dst = is->is_dst;
2048	ipsl.isl_p = is->is_p;
2049	ipsl.isl_v = is->is_v;
2050	ipsl.isl_flags = is->is_flags;
2051	ipsl.isl_rulen = is->is_rulen;
2052	ipsl.isl_group = is->is_group;
2053	if (ipsl.isl_p == IPPROTO_TCP || ipsl.isl_p == IPPROTO_UDP) {
2054		ipsl.isl_sport = is->is_sport;
2055		ipsl.isl_dport = is->is_dport;
2056		if (ipsl.isl_p == IPPROTO_TCP) {
2057			ipsl.isl_state[0] = is->is_state[0];
2058			ipsl.isl_state[1] = is->is_state[1];
2059		}
2060	} else if (ipsl.isl_p == IPPROTO_ICMP) {
2061		ipsl.isl_itype = is->is_icmp.ics_type;
2062	} else if (ipsl.isl_p == IPPROTO_ICMPV6) {
2063		ipsl.isl_itype = is->is_icmp.ics_type;
2064	} else {
2065		ipsl.isl_ps.isl_filler[0] = 0;
2066		ipsl.isl_ps.isl_filler[1] = 0;
2067	}
2068	items[0] = &ipsl;
2069	sizes[0] = sizeof(ipsl);
2070	types[0] = 0;
2071
2072	(void) ipllog(IPL_LOGSTATE, NULL, items, sizes, types, 1);
2073}
2074#endif
2075
2076
2077#ifdef	USE_INET6
2078frentry_t *fr_checkicmp6matchingstate(ip, fin)
2079ip6_t *ip;
2080fr_info_t *fin;
2081{
2082	register ipstate_t *is, **isp;
2083	register u_short sport, dport;
2084	register u_char	pr;
2085	struct icmp6_hdr *ic, *oic;
2086	union i6addr dst, src;
2087	u_short savelen;
2088	fr_info_t ofin;
2089	tcphdr_t *tcp;
2090	frentry_t *fr;
2091	ip6_t *oip;
2092	int type;
2093	u_int hv;
2094
2095	/*
2096	 * Does it at least have the return (basic) IP header ?
2097	 * Only a basic IP header (no options) should be with
2098	 * an ICMP error header.
2099	 */
2100	if ((fin->fin_v != 6) || (fin->fin_plen < ICMP6ERR_MINPKTLEN))
2101		return NULL;
2102	ic = (struct icmp6_hdr *)fin->fin_dp;
2103	type = ic->icmp6_type;
2104	/*
2105	 * If it's not an error type, then return
2106	 */
2107	if ((type != ICMP6_DST_UNREACH) && (type != ICMP6_PACKET_TOO_BIG) &&
2108	    (type != ICMP6_TIME_EXCEEDED) && (type != ICMP6_PARAM_PROB))
2109		return NULL;
2110
2111	oip = (ip6_t *)((char *)ic + ICMPERR_ICMPHLEN);
2112	if (fin->fin_plen < sizeof(*oip))
2113		return NULL;
2114
2115	if ((oip->ip6_nxt != IPPROTO_TCP) && (oip->ip6_nxt != IPPROTO_UDP) &&
2116	    (oip->ip6_nxt != IPPROTO_ICMPV6))
2117		return NULL;
2118
2119	bzero((char *)&ofin, sizeof(ofin));
2120	ofin.fin_out = !fin->fin_out;
2121	ofin.fin_ifp = fin->fin_ifp;
2122	ofin.fin_v = 6;
2123
2124	if (oip->ip6_nxt == IPPROTO_ICMPV6) {
2125		oic = (struct icmp6_hdr *)(oip + 1);
2126		/*
2127		 * a ICMP error can only be generated as a result of an
2128		 * ICMP query, not as the response on an ICMP error
2129		 *
2130		 * XXX theoretically ICMP_ECHOREP and the other reply's are
2131		 * ICMP query's as well, but adding them here seems strange XXX
2132		 */
2133		 if (!(oic->icmp6_type & ICMP6_INFOMSG_MASK))
2134		    	return NULL;
2135
2136		/*
2137		 * perform a lookup of the ICMP packet in the state table
2138		 */
2139		hv = (pr = oip->ip6_nxt);
2140		src.in6 = oip->ip6_src;
2141		hv += src.in4.s_addr;
2142		dst.in6 = oip->ip6_dst;
2143		hv += dst.in4.s_addr;
2144		hv += oic->icmp6_id;
2145		hv += oic->icmp6_seq;
2146		hv %= fr_statesize;
2147
2148		oip->ip6_plen = ntohs(oip->ip6_plen);
2149		fr_makefrip(sizeof(*oip), (ip_t *)oip, &ofin);
2150		oip->ip6_plen = htons(oip->ip6_plen);
2151
2152		READ_ENTER(&ipf_state);
2153		for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext)
2154			if ((is->is_p == pr) &&
2155			    (oic->icmp6_id == is->is_icmp.ics_id) &&
2156			    (oic->icmp6_seq == is->is_icmp.ics_seq) &&
2157			    fr_matchsrcdst(is, src, dst, &ofin, NULL)) {
2158			    	/*
2159			    	 * in the state table ICMP query's are stored
2160			    	 * with the type of the corresponding ICMP
2161			    	 * response. Correct here
2162			    	 */
2163				if (((is->is_type == ICMP6_ECHO_REPLY) &&
2164				     (oic->icmp6_type == ICMP6_ECHO_REQUEST)) ||
2165				     (is->is_type - 1 == oic->icmp6_type )) {
2166				    	ips_stats.iss_hits++;
2167    					is->is_pkts++;
2168					is->is_bytes += fin->fin_plen;
2169					return is->is_rule;
2170				}
2171			}
2172		RWLOCK_EXIT(&ipf_state);
2173
2174		return NULL;
2175	}
2176
2177	tcp = (tcphdr_t *)(oip + 1);
2178	dport = tcp->th_dport;
2179	sport = tcp->th_sport;
2180
2181	hv = (pr = oip->ip6_nxt);
2182	src.in6 = oip->ip6_src;
2183	hv += src.in4.s_addr;
2184	hv += src.i6[1];
2185	hv += src.i6[2];
2186	hv += src.i6[3];
2187	dst.in6 = oip->ip6_dst;
2188	hv += dst.in4.s_addr;
2189	hv += dst.i6[1];
2190	hv += dst.i6[2];
2191	hv += dst.i6[3];
2192	hv += dport;
2193	hv += sport;
2194	hv %= fr_statesize;
2195	/*
2196	 * we make an fin entry to be able to feed it to
2197	 * matchsrcdst note that not all fields are encessary
2198	 * but this is the cleanest way. Note further we fill
2199	 * in fin_mp such that if someone uses it we'll get
2200	 * a kernel panic. fr_matchsrcdst does not use this.
2201	 *
2202	 * watch out here, as ip is in host order and oip in network
2203	 * order. Any change we make must be undone afterwards.
2204	 */
2205	savelen = oip->ip6_plen;
2206	oip->ip6_plen = ip->ip6_plen - sizeof(*ip) - ICMPERR_ICMPHLEN;
2207	fr_makefrip(sizeof(*oip), (ip_t *)oip, &ofin);
2208	oip->ip6_plen = savelen;
2209	READ_ENTER(&ipf_state);
2210	for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) {
2211		/*
2212		 * Only allow this icmp though if the
2213		 * encapsulated packet was allowed through the
2214		 * other way around. Note that the minimal amount
2215		 * of info present does not allow for checking against
2216		 * tcp internals such as seq and ack numbers.
2217		 */
2218		if ((is->is_p == pr) && (is->is_v == 6) &&
2219		    fr_matchsrcdst(is, src, dst, &ofin, tcp)) {
2220			fr = is->is_rule;
2221			ips_stats.iss_hits++;
2222			is->is_pkts++;
2223			is->is_bytes += fin->fin_plen;
2224			/*
2225			 * we deliberately do not touch the timeouts
2226			 * for the accompanying state table entry.
2227			 * It remains to be seen if that is correct. XXX
2228			 */
2229			RWLOCK_EXIT(&ipf_state);
2230			return fr;
2231		}
2232	}
2233	RWLOCK_EXIT(&ipf_state);
2234	return NULL;
2235}
2236#endif
2237