2 * Copyright (C) 1995-2002 by Darren Reed.
4 * See the IPFILTER.LICENCE file for details on licencing.
6 * @(#)ip_state.c 1.8 6/5/96 (C) 1993-2000 Darren Reed
7 * @(#)$Id: ip_state.c,v 2.30.2.74 2002/07/27 15:58:10 darrenr Exp $
8 * $FreeBSD: src/sys/contrib/ipfilter/netinet/ip_state.c,v 1.21.2.6 2004/07/04 09:24:39 darrenr Exp $
9 * $DragonFly: src/sys/contrib/ipfilter/netinet/ip_state.c,v 1.8 2004/08/28 07:27:02 dillon Exp $
12 #if defined(__sgi) && (IRIX > 602)
13 # include <sys/ptimers.h>
15 #include <sys/errno.h>
16 #include <sys/types.h>
17 #include <sys/param.h>
19 #if defined(__NetBSD__) && (NetBSD >= 199905) && !defined(IPFILTER_LKM) && \
21 # include "opt_ipfilter_log.h"
23 #if defined(_KERNEL) && (defined(__DragonFly__) || \
24 (defined(__FreeBSD_version) && \
25 __FreeBSD_version >= 400000)) && !defined(KLD_MODULE)
26 #include "opt_inet6.h"
28 #if !defined(_KERNEL) && !defined(KERNEL) && !defined(__KERNEL__)
34 # include <linux/kernel.h>
35 # include <linux/module.h>
38 #if (defined(KERNEL) || defined(_KERNEL)) && (defined(__DragonFly__) || __FreeBSD_version >= 220000)
39 # include <sys/filio.h>
40 # include <sys/fcntl.h>
41 # if (defined(__DragonFly__) || __FreeBSD_version >= 300000) && !defined(IPFILTER_LKM)
42 # include "opt_ipfilter.h"
45 # include <sys/ioctl.h>
49 # include <sys/protosw.h>
51 #include <sys/socket.h>
52 #if (defined(_KERNEL) || defined(KERNEL)) && !defined(linux)
53 # include <sys/systm.h>
55 #if !defined(__SVR4) && !defined(__svr4__)
57 # include <sys/mbuf.h>
60 # include <sys/filio.h>
61 # include <sys/byteorder.h>
63 # include <sys/dditypes.h>
65 # include <sys/stream.h>
66 # include <sys/kmem.h>
73 #include <net/route.h>
74 #include <netinet/in.h>
75 #include <netinet/in_systm.h>
76 #include <netinet/ip.h>
77 #include <netinet/tcp.h>
79 # include <netinet/ip_var.h>
80 # include <netinet/tcp_fsm.h>
82 #include <netinet/udp.h>
83 #include <netinet/ip_icmp.h>
84 #include "ip_compat.h"
85 #include <netinet/tcpip.h>
91 #include <netinet/icmp6.h>
93 #if defined(__DragonFly__) || (__FreeBSD_version >= 300000)
94 # include <sys/malloc.h>
95 # if (defined(_KERNEL) || defined(KERNEL)) && !defined(IPFILTER_LKM)
96 # include <sys/libkern.h>
97 # include <sys/systm.h>
101 static const char sccsid[] = "@(#)ip_state.c 1.8 6/5/96 (C) 1993-2000 Darren Reed";
104 # define MIN(a,b) (((a)<(b))?(a):(b))
107 #define TCP_CLOSE (TH_FIN|TH_RST)
109 static ipstate_t **ips_table = NULL;
110 static int ips_num = 0;
111 static int ips_wild = 0;
112 static ips_stat_t ips_stats;
113 #if (SOLARIS || defined(__sgi)) && defined(_KERNEL)
114 extern KRWLOCK_T ipf_state, ipf_mutex;
115 extern kmutex_t ipf_rw;
119 static frentry_t *fr_checkicmp6matchingstate (ip6_t *, fr_info_t *);
121 static int fr_matchsrcdst (ipstate_t *, union i6addr, union i6addr,
122 fr_info_t *, tcphdr_t *);
123 static frentry_t *fr_checkicmpmatchingstate (ip_t *, fr_info_t *);
124 static int fr_matchicmpqueryreply (int, ipstate_t *, icmphdr_t *, int);
125 static int fr_state_flush (int, int);
126 static ips_stat_t *fr_statetstats (void);
127 static void fr_delstate (ipstate_t *);
128 static int fr_state_remove (caddr_t);
129 static void fr_ipsmove (ipstate_t **, ipstate_t *, u_int);
130 static int fr_tcpoptions (tcphdr_t *);
131 int fr_stputent (caddr_t);
132 int fr_stgetent (caddr_t);
133 void fr_stinsert (ipstate_t *);
136 #define FIVE_DAYS (2 * 5 * 86400) /* 5 days: half closed session */
138 #define TCP_MSL 240 /* 2 minutes */
139 u_long fr_tcpidletimeout = FIVE_DAYS,
140 fr_tcpclosewait = 2 * TCP_MSL,
141 fr_tcplastack = 2 * TCP_MSL,
142 fr_tcptimeout = 2 * TCP_MSL,
144 fr_tcphalfclosed = 2 * 2 * 3600, /* 2 hours */
146 fr_udpacktimeout = 24,
147 fr_icmptimeout = 120,
148 fr_icmpacktimeout = 12;
149 int fr_statemax = IPSTATE_MAX,
150 fr_statesize = IPSTATE_SIZE;
151 int fr_state_doflush = 0,
153 ipstate_t *ips_list = NULL;
155 static int icmpreplytype4[ICMP_MAXTYPE + 1];
157 static int icmpreplytype6[ICMP6_MAXTYPE + 1];
164 KMALLOCS(ips_table, ipstate_t **, fr_statesize * sizeof(ipstate_t *));
165 if (ips_table != NULL)
166 bzero((char *)ips_table, fr_statesize * sizeof(ipstate_t *));
170 /* fill icmp reply type table */
171 for (i = 0; i <= ICMP_MAXTYPE; i++)
172 icmpreplytype4[i] = -1;
173 icmpreplytype4[ICMP_ECHO] = ICMP_ECHOREPLY;
174 icmpreplytype4[ICMP_TSTAMP] = ICMP_TSTAMPREPLY;
175 icmpreplytype4[ICMP_IREQ] = ICMP_IREQREPLY;
176 icmpreplytype4[ICMP_MASKREQ] = ICMP_MASKREPLY;
178 /* fill icmp reply type table */
179 for (i = 0; i <= ICMP6_MAXTYPE; i++)
180 icmpreplytype6[i] = -1;
181 icmpreplytype6[ICMP6_ECHO_REQUEST] = ICMP6_ECHO_REPLY;
182 icmpreplytype6[ICMP6_MEMBERSHIP_QUERY] = ICMP6_MEMBERSHIP_REPORT;
183 icmpreplytype6[ICMP6_NI_QUERY] = ICMP6_NI_REPLY;
184 icmpreplytype6[ND_ROUTER_SOLICIT] = ND_ROUTER_ADVERT;
185 icmpreplytype6[ND_NEIGHBOR_SOLICIT] = ND_NEIGHBOR_ADVERT;
192 static ips_stat_t *fr_statetstats()
194 ips_stats.iss_active = ips_num;
195 ips_stats.iss_table = ips_table;
196 ips_stats.iss_list = ips_list;
202 * flush state tables. two actions currently defined:
203 * which == 0 : flush all state table entries
204 * which == 1 : flush TCP connections which have started to close but are
205 * stuck for some reason.
206 * which == 2 : flush TCP connections which have been idle for a long time,
207 * starting at > 4 days idle and working back in successive half-
208 * days to at most 12 hours old.
210 static int fr_state_flush(which, proto)
213 ipstate_t *is, **isp;
214 #if defined(_KERNEL) && !SOLARIS
217 int delete, removed = 0, try;
220 for (isp = &ips_list; (is = *isp); ) {
223 if ((proto != 0) && (is->is_v != proto))
233 if (is->is_p != IPPROTO_TCP)
235 if ((is->is_state[0] != TCPS_ESTABLISHED) ||
236 (is->is_state[1] != TCPS_ESTABLISHED))
242 if (is->is_p == IPPROTO_TCP)
245 ips_stats.iss_expire++;
247 ipstate_log(is, ISL_FLUSH);
256 * Asked to remove inactive entries, try again if first attempt
257 * failed. In this case, 86400 is half a day because the counter is
258 * activated every half second.
260 if ((which == 2) && (removed == 0)) {
261 try = 86400; /* half a day */
262 for (; (try < FIVE_DAYS) && (removed == 0); try += 86400) {
263 for (isp = &ips_list; (is = *isp); ) {
265 if ((is->is_p == IPPROTO_TCP) &&
266 ((is->is_state[0] == TCPS_ESTABLISHED) ||
267 (is->is_state[1] == TCPS_ESTABLISHED)) &&
268 (is->is_age < try)) {
271 } else if ((is->is_p != IPPROTO_TCP) &&
273 ips_stats.iss_expire++;
278 ipstate_log(is, ISL_FLUSH);
293 static int fr_state_remove(data)
300 error = IRCOPYPTR(data, (caddr_t)&st, sizeof(st));
304 WRITE_ENTER(&ipf_state);
305 for (sp = ips_list; sp; sp = sp->is_next)
306 if ((sp->is_p == st.is_p) && (sp->is_v == st.is_v) &&
307 !bcmp((char *)&sp->is_src, (char *)&st.is_src,
308 sizeof(st.is_src)) &&
309 !bcmp((char *)&sp->is_dst, (char *)&st.is_dst,
310 sizeof(st.is_dst)) &&
311 !bcmp((char *)&sp->is_ps, (char *)&st.is_ps,
314 ipstate_log(sp, ISL_REMOVE);
317 RWLOCK_EXIT(&ipf_state);
320 RWLOCK_EXIT(&ipf_state);
325 int fr_state_ioctl(data, cmd, mode)
327 #if defined(__NetBSD__) || defined(__OpenBSD__)
334 int arg, ret, error = 0;
339 error = fr_state_remove(data);
342 error = IRCOPY(data, (caddr_t)&arg, sizeof(arg));
345 if (arg == 0 || arg == 1) {
346 WRITE_ENTER(&ipf_state);
347 ret = fr_state_flush(arg, 4);
348 RWLOCK_EXIT(&ipf_state);
349 error = IWCOPY((caddr_t)&ret, data, sizeof(ret));
355 error = IRCOPY(data, (caddr_t)&arg, sizeof(arg));
358 if (arg == 0 || arg == 1) {
359 WRITE_ENTER(&ipf_state);
360 ret = fr_state_flush(arg, 6);
361 RWLOCK_EXIT(&ipf_state);
362 error = IWCOPY((caddr_t)&ret, data, sizeof(ret));
369 if (!(mode & FWRITE))
374 tmp = ipflog_clear(IPL_LOGSTATE);
375 IWCOPY((char *)&tmp, data, sizeof(tmp));
380 error = IWCOPYPTR((caddr_t)fr_statetstats(), data,
385 arg = (int)iplused[IPL_LOGSTATE];
386 error = IWCOPY((caddr_t)&arg, (caddr_t)data, sizeof(arg));
390 error = fr_lock(data, &fr_state_lock);
393 if (!fr_state_lock) {
397 error = fr_stputent(data);
400 if (!fr_state_lock) {
404 error = fr_stgetent(data);
415 * Copy out state information from the kernel to a user space process.
417 int fr_stgetent(data)
424 error = IRCOPYPTR(data, (caddr_t)&ips, sizeof(ips));
432 if (ips.ips_next == NULL)
438 * Make sure the pointer we're copying from exists in the
439 * current list of entries. Security precaution to prevent
440 * copying of random kernel data.
442 for (is = ips_list; is; is = is->is_next)
448 ips.ips_next = isn->is_next;
449 bcopy((char *)isn, (char *)&ips.ips_is, sizeof(ips.ips_is));
451 bcopy((char *)isn->is_rule, (char *)&ips.ips_fr,
453 error = IWCOPYPTR((caddr_t)&ips, data, sizeof(ips));
460 int fr_stputent(data)
469 error = IRCOPYPTR(data, (caddr_t)&ips, sizeof(ips));
473 KMALLOC(isn, ipstate_t *);
477 bcopy((char *)&ips.ips_is, (char *)isn, sizeof(*isn));
480 if (isn->is_flags & FI_NEWFR) {
481 KMALLOC(fr, frentry_t *);
486 bcopy((char *)&ips.ips_fr, (char *)fr, sizeof(*fr));
487 out = fr->fr_flags & FR_OUTQUE ? 1 : 0;
489 ips.ips_is.is_rule = fr;
492 * Look up all the interface names in the rule.
494 for (i = 0; i < 4; i++) {
495 name = fr->fr_ifnames[i];
496 if ((name[1] == '\0') &&
497 ((name[0] == '-') || (name[0] == '*'))) {
498 fr->fr_ifas[i] = NULL;
499 } else if (*name != '\0') {
500 fr->fr_ifas[i] = GETUNIT(name,
502 if (fr->fr_ifas[i] == NULL)
503 fr->fr_ifas[i] = (void *)-1;
505 strncpy(isn->is_ifname[i],
506 IFNAME(fr->fr_ifas[i]),
510 isn->is_ifp[out] = fr->fr_ifas[i];
514 * send a copy back to userland of what we ended up
515 * to allow for verification.
517 error = IWCOPYPTR((caddr_t)&ips, data, sizeof(ips));
524 for (is = ips_list; is; is = is->is_next)
525 if (is->is_rule == fr)
539 * Insert a state table entry manually.
544 u_int hv = is->is_hv;
548 MUTEX_INIT(&is->is_lock, "ipf state entry", NULL);
551 * Look up all the interface names in the state entry.
553 for (i = 0; i < 4; i++) {
554 name = is->is_ifname[i];
555 if ((name[1] == '\0') &&
556 ((name[0] == '-') || (name[0] == '*'))) {
557 is->is_ifp[0] = NULL;
558 } else if (*name != '\0') {
559 is->is_ifp[i] = GETUNIT(name, is->is_v);
560 if (is->is_ifp[i] == NULL)
561 is->is_ifp[i] = (void *)-1;
567 * add into list table.
570 ips_list->is_pnext = &is->is_next;
571 is->is_pnext = &ips_list;
572 is->is_next = ips_list;
575 ips_table[hv]->is_phnext = &is->is_hnext;
577 ips_stats.iss_inuse++;
578 is->is_phnext = ips_table + hv;
579 is->is_hnext = ips_table[hv];
586 * Create a new ipstate structure and hang it off the hash table.
588 ipstate_t *fr_addstate(ip, fin, stsave, flags)
594 tcphdr_t *tcp = NULL;
603 if (fr_state_lock || (fin->fin_off != 0) || (fin->fin_fl & FI_SHORT) ||
604 (fin->fin_misc & FM_BADSTATE))
606 if (ips_num == fr_statemax) {
608 fr_state_doflush = 1;
613 bzero((char *)is, sizeof(*is));
615 ips.is_state[0] = TCPS_CLOSED;
616 ips.is_state[1] = TCPS_CLOSED;
619 * Copy and calculate...
621 hv = (is->is_p = fin->fin_fi.fi_p);
622 is->is_src = fin->fin_fi.fi_src;
624 is->is_dst = fin->fin_fi.fi_dst;
627 if (fin->fin_v == 6) {
628 if ((is->is_p == IPPROTO_ICMPV6) &&
629 IN6_IS_ADDR_MULTICAST(&is->is_dst.in6)) {
631 * So you can do keep state with neighbour discovery.
636 hv += is->is_dst.i6[1];
637 hv += is->is_dst.i6[2];
638 hv += is->is_dst.i6[3];
640 hv += is->is_src.i6[1];
641 hv += is->is_src.i6[2];
642 hv += is->is_src.i6[3];
651 case IPPROTO_ICMPV6 :
652 ic = (struct icmp *)fin->fin_dp;
653 if ((ic->icmp_type & ICMP6_INFOMSG_MASK) == 0)
656 switch (ic->icmp_type)
658 case ICMP6_ECHO_REQUEST :
659 is->is_icmp.ics_type = ic->icmp_type;
660 hv += (is->is_icmp.ics_id = ic->icmp_id);
661 hv += (is->is_icmp.ics_seq = ic->icmp_seq);
663 case ICMP6_MEMBERSHIP_QUERY :
664 case ND_ROUTER_SOLICIT :
665 case ND_NEIGHBOR_SOLICIT :
666 case ICMP6_NI_QUERY :
667 is->is_icmp.ics_type = ic->icmp_type;
672 ATOMIC_INCL(ips_stats.iss_icmp);
673 is->is_age = fr_icmptimeout;
677 ic = (struct icmp *)fin->fin_dp;
679 switch (ic->icmp_type)
685 is->is_icmp.ics_type = ic->icmp_type;
686 hv += (is->is_icmp.ics_id = ic->icmp_id);
687 hv += (is->is_icmp.ics_seq = ic->icmp_seq);
692 ATOMIC_INCL(ips_stats.iss_icmp);
693 is->is_age = fr_icmptimeout;
696 tcp = (tcphdr_t *)fin->fin_dp;
698 if (tcp->th_flags & TH_RST)
701 * The endian of the ports doesn't matter, but the ack and
702 * sequence numbers do as we do mathematics on them later.
704 is->is_sport = htons(fin->fin_data[0]);
705 is->is_dport = htons(fin->fin_data[1]);
706 if ((flags & (FI_W_DPORT|FI_W_SPORT)) == 0) {
710 if ((flags & FI_IGNOREPKT) == 0) {
711 is->is_send = ntohl(tcp->th_seq) + fin->fin_dlen -
712 (off = (tcp->th_off << 2)) +
713 ((tcp->th_flags & TH_SYN) ? 1 : 0) +
714 ((tcp->th_flags & TH_FIN) ? 1 : 0);
715 is->is_maxsend = is->is_send;
717 if ((tcp->th_flags & TH_SYN) &&
718 ((tcp->th_off << 2) >= (sizeof(*tcp) + 4))) {
719 ws = fr_tcpoptions(tcp);
726 is->is_maxswin = ntohs(tcp->th_win);
727 if (is->is_maxswin == 0)
730 if ((tcp->th_flags & TH_OPENING) == TH_SYN)
734 * If we're creating state for a starting connection, start the
735 * timer on it as we'll never see an error if it fails to
738 ATOMIC_INCL(ips_stats.iss_tcp);
742 tcp = (tcphdr_t *)fin->fin_dp;
744 is->is_sport = htons(fin->fin_data[0]);
745 is->is_dport = htons(fin->fin_data[1]);
746 if ((flags & (FI_W_DPORT|FI_W_SPORT)) == 0) {
750 ATOMIC_INCL(ips_stats.iss_udp);
751 is->is_age = fr_udptimeout;
754 is->is_age = fr_udptimeout;
758 KMALLOC(is, ipstate_t *);
760 ATOMIC_INCL(ips_stats.iss_nomem);
763 bcopy((char *)&ips, (char *)is, sizeof(*is));
766 is->is_rule = fin->fin_fr;
767 if (is->is_rule != NULL) {
768 is->is_group = is->is_rule->fr_group;
769 ATOMIC_INC32(is->is_rule->fr_ref);
770 pass = is->is_rule->fr_flags;
771 is->is_frage[0] = is->is_rule->fr_age[0];
772 is->is_frage[1] = is->is_rule->fr_age[1];
773 if (is->is_frage[0] != 0)
774 is->is_age = is->is_frage[0];
776 is->is_ifp[(out << 1) + 1] = is->is_rule->fr_ifas[1];
777 is->is_ifp[(1 - out) << 1] = is->is_rule->fr_ifas[2];
778 is->is_ifp[((1 - out) << 1) + 1] = is->is_rule->fr_ifas[3];
780 if (((ifp = is->is_rule->fr_ifas[1]) != NULL) &&
782 strncpy(is->is_ifname[(out << 1) + 1],
783 IFNAME(ifp), IFNAMSIZ);
784 if (((ifp = is->is_rule->fr_ifas[2]) != NULL) &&
786 strncpy(is->is_ifname[(1 - out) << 1],
787 IFNAME(ifp), IFNAMSIZ);
788 if (((ifp = is->is_rule->fr_ifas[3]) != NULL) &&
790 strncpy(is->is_ifname[((1 - out) << 1) + 1],
791 IFNAME(ifp), IFNAMSIZ);
795 is->is_ifp[out << 1] = fin->fin_ifp;
796 strncpy(is->is_ifname[out << 1], IFNAME(fin->fin_ifp), IFNAMSIZ);
798 WRITE_ENTER(&ipf_state);
801 if ((flags & FI_IGNOREPKT) == 0) {
803 is->is_bytes = fin->fin_dlen + fin->fin_hlen;
806 * We want to check everything that is a property of this packet,
807 * but we don't (automatically) care about it's fragment status as
810 is->is_v = fin->fin_v;
811 is->is_rulen = fin->fin_rule;
812 is->is_opt = fin->fin_fi.fi_optmsk;
813 is->is_optmsk = 0xffffffff;
814 is->is_sec = fin->fin_fi.fi_secmsk;
815 is->is_secmsk = 0xffff;
816 is->is_auth = fin->fin_fi.fi_auth;
817 is->is_authmsk = 0xffff;
818 is->is_flags = fin->fin_fl & FI_CMP;
819 is->is_flags |= FI_CMP << 4;
820 is->is_flags |= flags & (FI_WILDP|FI_WILDA);
821 if (flags & (FI_WILDP|FI_WILDA))
824 if (pass & FR_LOGFIRST)
825 is->is_pass &= ~(FR_LOGFIRST|FR_LOG);
828 if (is->is_p == IPPROTO_TCP) {
829 fr_tcp_age(&is->is_age, is->is_state, fin,
830 0, is->is_fsm); /* 0 = packet from the source */
833 ipstate_log(is, ISL_NEW);
835 RWLOCK_EXIT(&ipf_state);
836 fin->fin_rev = IP6NEQ(is->is_dst, fin->fin_fi.fi_dst);
837 if ((fin->fin_fl & FI_FRAG) && (pass & FR_KEEPFRAG))
838 ipfr_newfrag(ip, fin);
843 static int fr_tcpoptions(tcp)
849 opt = (u_char *) (tcp + 1);
850 last = ((u_char *)tcp) + (tcp->th_off << 2);
852 /* If we don't find wscale here, we need to clear it */
855 /* Termination condition picked such that opt[0 .. 2] exist */
856 while ((opt < last - 2) && (*opt != TCPOPT_EOL)) {
862 /* Proper length ? */
871 /* Unknown options must be two bytes+ */
885 * check to see if a packet with TCP headers fits within the TCP window.
886 * change timeout depending on whether new packet is a SYN-ACK returning for a
887 * SYN or a RST or FIN which indicate time to close up shop.
889 int fr_tcpstate(is, fin, ip, tcp)
895 tcp_seq seq, ack, end;
897 tcpdata_t *fdata, *tdata;
904 * Find difference between last checked packet and this packet.
906 source = IP6EQ(fin->fin_fi.fi_src, is->is_src);
907 if (source && (ntohs(is->is_sport) != fin->fin_data[0]))
909 fdata = &is->is_tcp.ts_data[!source];
910 tdata = &is->is_tcp.ts_data[source];
911 off = tcp->th_off << 2;
912 seq = ntohl(tcp->th_seq);
913 ack = ntohl(tcp->th_ack);
914 win = ntohs(tcp->th_win);
915 end = seq + fin->fin_dlen - off +
916 ((tcp->th_flags & TH_SYN) ? 1 : 0) +
917 ((tcp->th_flags & TH_FIN) ? 1 : 0);
920 if ((tcp->th_flags & TH_SYN) && (off >= sizeof(*tcp) + 4))
921 wscale = fr_tcpoptions(tcp);
925 MUTEX_ENTER(&is->is_lock);
928 fdata->td_wscale = wscale;
929 else if (wscale == -2)
930 fdata->td_wscale = tdata->td_wscale = 0;
931 if (!(tcp->th_flags & TH_SYN))
932 win <<= fdata->td_wscale;
934 if ((fdata->td_end == 0) &&
935 (!is->is_fsm || ((tcp->th_flags & TH_OPENING) == TH_OPENING))) {
937 * Must be a (outgoing) SYN-ACK in reply to a SYN.
940 fdata->td_maxwin = 1;
941 fdata->td_maxend = end + win;
946 if (!(tcp->th_flags & TH_ACK)) { /* Pretend an ack was sent */
948 } else if (((tcp->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) &&
950 /* gross hack to get around certain broken tcp stacks */
955 seq = end = fdata->td_end;
957 maxwin = tdata->td_maxwin;
958 ackskew = tdata->td_end - ack;
960 #define SEQ_GE(a,b) ((int)((a) - (b)) >= 0)
961 #define SEQ_GT(a,b) ((int)((a) - (b)) > 0)
962 if ((SEQ_GE(fdata->td_maxend, end)) &&
963 (SEQ_GE(seq, fdata->td_end - maxwin)) &&
964 /* XXX what about big packets */
965 #define MAXACKWINDOW 66000
966 (-ackskew <= (MAXACKWINDOW << tdata->td_wscale)) &&
967 ( ackskew <= (MAXACKWINDOW << tdata->td_wscale))) {
969 /* if ackskew < 0 then this should be due to fragmented
970 * packets. There is no way to know the length of the
971 * total packet in advance.
972 * We do know the total length from the fragment cache though.
973 * Note however that there might be more sessions with
974 * exactly the same source and destination parameters in the
975 * state cache (and source and destination is the only stuff
976 * that is saved in the fragment cache). Note further that
977 * some TCP connections in the state cache are hashed with
978 * sport and dport as well which makes it not worthwhile to
980 * Thus, when ackskew is negative but still seems to belong
981 * to this session, we bump up the destinations end value.
984 * Nearing end of connection, start timeout.
986 /* source ? 0 : 1 -> !source */
987 if (fr_tcp_age(&is->is_age, is->is_state, fin, !source,
988 (int)is->is_fsm) == 0) {
992 /* update max window seen */
993 if (fdata->td_maxwin < win)
994 fdata->td_maxwin = win;
995 if (SEQ_GT(end, fdata->td_end))
997 if (SEQ_GE(ack + win, tdata->td_maxend)) {
998 tdata->td_maxend = ack + win;
1003 ATOMIC_INCL(ips_stats.iss_hits);
1007 MUTEX_EXIT(&is->is_lock);
1008 if ((ret == 0) && ((tcp->th_flags & TH_OPENING) != TH_SYN))
1009 fin->fin_misc |= FM_BADSTATE;
1015 * Match a state table entry against an IP packet.
1017 static int fr_matchsrcdst(is, src, dst, fin, tcp)
1019 union i6addr src, dst;
1023 int ret = 0, rev, out, flags, idx;
1027 rev = IP6NEQ(is->is_dst, dst);
1030 flags = is->is_flags & (FI_WILDA|FI_WILDP);
1035 flags = is->is_flags;
1039 if (!(flags & FI_W_SPORT) && (sp != is->is_sport))
1041 else if (!(flags & FI_W_DPORT) && (dp != is->is_dport))
1046 idx = (out << 1) + rev;
1048 if ((is->is_ifp[idx] == NULL &&
1049 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) ||
1050 is->is_ifp[idx] == ifp)
1058 if ((IP6EQ(is->is_dst, dst) || (flags & FI_W_DADDR)) &&
1059 (IP6EQ(is->is_src, src) || (flags & FI_W_SADDR)) &&
1060 (!tcp || ((sp == is->is_sport || flags & FI_W_SPORT) &&
1061 (dp == is->is_dport || flags & FI_W_DPORT)))) {
1065 if ((IP6EQ(is->is_dst, src) || (flags & FI_W_DADDR)) &&
1066 (IP6EQ(is->is_src, dst) || (flags & FI_W_SADDR)) &&
1067 (!tcp || ((sp == is->is_dport || flags & FI_W_DPORT) &&
1068 (dp == is->is_sport || flags & FI_W_SPORT)))) {
1076 * Whether or not this should be here, is questionable, but the aim
1077 * is to get this out of the main line.
1080 flags = is->is_flags & (FI_CMP|(FI_CMP<<4));
1082 if (((fin->fin_fl & (flags >> 4)) != (flags & FI_CMP)) ||
1083 (fin->fin_fi.fi_optmsk != is->is_opt) ||
1084 (fin->fin_fi.fi_secmsk != is->is_sec) ||
1085 (fin->fin_fi.fi_auth != is->is_auth))
1088 flags = is->is_flags & (FI_WILDA|FI_WILDP);
1089 if ((flags & (FI_W_SADDR|FI_W_DADDR))) {
1090 if ((flags & FI_W_SADDR) != 0) {
1092 is->is_src = fin->fin_fi.fi_src;
1094 is->is_src = fin->fin_fi.fi_dst;
1096 } else if ((flags & FI_W_DADDR) != 0) {
1098 is->is_dst = fin->fin_fi.fi_dst;
1100 is->is_dst = fin->fin_fi.fi_src;
1103 is->is_flags &= ~(FI_W_SADDR|FI_W_DADDR);
1104 if ((is->is_flags & (FI_WILDA|FI_WILDP)) == 0)
1108 if ((flags & (FI_W_SPORT|FI_W_DPORT))) {
1109 if ((flags & FI_W_SPORT) != 0) {
1112 is->is_send = htonl(tcp->th_seq);
1115 is->is_send = htonl(tcp->th_ack);
1117 is->is_maxsend = is->is_send + 1;
1118 } else if ((flags & FI_W_DPORT) != 0) {
1121 is->is_dend = htonl(tcp->th_ack);
1124 is->is_dend = htonl(tcp->th_seq);
1126 is->is_maxdend = is->is_dend + 1;
1128 is->is_flags &= ~(FI_W_SPORT|FI_W_DPORT);
1134 if (is->is_ifp[idx] == NULL &&
1135 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*'))
1139 is->is_ifp[ret] = ifp;
1140 strncpy(is->is_ifname[ret], IFNAME(ifp),
1141 sizeof(is->is_ifname[ret]));
1147 static int fr_matchicmpqueryreply(v, is, icmp, rev)
1154 * If we matched its type on the way in, then when going out
1155 * it will still be the same type.
1157 if ((!rev && (icmp->icmp_type == is->is_type)) ||
1158 (rev && (icmpreplytype4[is->is_type] == icmp->icmp_type))) {
1159 if (icmp->icmp_type != ICMP_ECHOREPLY)
1161 if ((icmp->icmp_id == is->is_icmp.ics_id) &&
1162 (icmp->icmp_seq == is->is_icmp.ics_seq))
1167 else if (is->is_v == 6) {
1168 if ((!rev && (icmp->icmp_type == is->is_type)) ||
1169 (rev && (icmpreplytype6[is->is_type] == icmp->icmp_type))) {
1170 if (icmp->icmp_type != ICMP6_ECHO_REPLY)
1172 if ((icmp->icmp_id == is->is_icmp.ics_id) &&
1173 (icmp->icmp_seq == is->is_icmp.ics_seq))
1181 static frentry_t *fr_checkicmpmatchingstate(ip, fin)
1185 ipstate_t *is, **isp;
1186 u_short sport, dport;
1188 u_short savelen, ohlen;
1189 union i6addr dst, src;
1200 * Does it at least have the return (basic) IP header ?
1201 * Only a basic IP header (no options) should be with
1202 * an ICMP error header.
1204 if (((ip->ip_v != 4) || (ip->ip_hl != 5)) ||
1205 (fin->fin_plen < ICMPERR_MINPKTLEN))
1208 ic = (struct icmp *)fin->fin_dp;
1209 type = ic->icmp_type;
1211 * If it's not an error type, then return
1213 if ((type != ICMP_UNREACH) && (type != ICMP_SOURCEQUENCH) &&
1214 (type != ICMP_REDIRECT) && (type != ICMP_TIMXCEED) &&
1215 (type != ICMP_PARAMPROB))
1218 oip = (ip_t *)((char *)ic + ICMPERR_ICMPHLEN);
1219 ohlen = oip->ip_hl << 2;
1221 * Check if the at least the old IP header (with options) and
1222 * 8 bytes of payload is present.
1224 if (fin->fin_plen < ICMPERR_MAXPKTLEN + ohlen - sizeof(*oip))
1230 len = fin->fin_dlen - ICMPERR_ICMPHLEN;
1231 if ((len <= 0) || (ohlen > len))
1235 * Is the buffer big enough for all of it ? It's the size of the IP
1236 * header claimed in the encapsulated part which is of concern. It
1237 * may be too big to be in this buffer but not so big that it's
1238 * outside the ICMP packet, leading to TCP deref's causing problems.
1239 * This is possible because we don't know how big oip_hl is when we
1240 * do the pullup early in fr_check() and thus can't guarantee it is
1249 if ((char *)oip + len > (char *)m->b_wptr)
1252 m = *(mb_t **)fin->fin_mp;
1253 if ((char *)oip + len > (char *)ip + m->m_len)
1260 * in the IPv4 case we must zero the i6addr union otherwise
1261 * the IP6EQ and IP6NEQ macros produce the wrong results because
1262 * of the 'junk' in the unused part of the union
1264 bzero((char *)&src, sizeof(src));
1265 bzero((char *)&dst, sizeof(dst));
1266 bzero((char *)&ofin, sizeof(ofin));
1268 * We make an fin entry to be able to feed it to
1269 * matchsrcdst. Note that not all fields are encessary
1270 * but this is the cleanest way. Note further that we
1271 * fill in fin_mp such that if someone uses it we'll get
1272 * a kernel panic. fr_matchsrcdst does not use this.
1274 ofin.fin_ifp = fin->fin_ifp;
1275 ofin.fin_out = !fin->fin_out;
1279 * watch out here, as ip is in host order and oip in network
1280 * order. Any change we make must be undone afterwards, like
1281 * oip->ip_off - it is still in network byte order so fix it.
1283 savelen = oip->ip_len;
1285 oip->ip_off = ntohs(oip->ip_off);
1286 (void) fr_makefrip(ohlen, oip, &ofin);
1288 * Reset the short flag here because in fr_matchsrcdst() the flags
1289 * for the current packet (fin_fl) are compared against * those for
1290 * the existing session.
1292 ofin.fin_fl &= ~FI_SHORT;
1295 * Put old values of ip_len and ip_off back as we don't know
1296 * if we have to forward the packet (or process it again.
1298 oip->ip_len = savelen;
1299 oip->ip_off = htons(oip->ip_off);
1302 ofin.fin_qfm = NULL;
1309 icmp = (icmphdr_t *)((char *)oip + ohlen);
1312 * an ICMP error can only be generated as a result of an
1313 * ICMP query, not as the response on an ICMP error
1315 * XXX theoretically ICMP_ECHOREP and the other reply's are
1316 * ICMP query's as well, but adding them here seems strange XXX
1318 if ((icmp->icmp_type != ICMP_ECHO) &&
1319 (icmp->icmp_type != ICMP_TSTAMP) &&
1320 (icmp->icmp_type != ICMP_IREQ) &&
1321 (icmp->icmp_type != ICMP_MASKREQ))
1325 * perform a lookup of the ICMP packet in the state table
1327 hv = (pr = oip->ip_p);
1328 src.in4 = oip->ip_src;
1329 hv += src.in4.s_addr;
1330 dst.in4 = oip->ip_dst;
1331 hv += dst.in4.s_addr;
1332 hv += icmp->icmp_id;
1333 hv += icmp->icmp_seq;
1336 READ_ENTER(&ipf_state);
1337 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext)
1338 if ((is->is_p == pr) && (is->is_v == 4) &&
1339 (is->is_icmppkts < is->is_pkts) &&
1340 fr_matchsrcdst(is, src, dst, &ofin, NULL) &&
1341 fr_matchicmpqueryreply(is->is_v, is, icmp,
1343 ips_stats.iss_hits++;
1345 is->is_bytes += ip->ip_len;
1349 RWLOCK_EXIT(&ipf_state);
1354 if (fin->fin_plen < ICMPERR_MAXPKTLEN)
1361 tcp = (tcphdr_t *)((char *)oip + ohlen);
1362 dport = tcp->th_dport;
1363 sport = tcp->th_sport;
1365 hv = (pr = oip->ip_p);
1366 src.in4 = oip->ip_src;
1367 hv += src.in4.s_addr;
1368 dst.in4 = oip->ip_dst;
1369 hv += dst.in4.s_addr;
1374 READ_ENTER(&ipf_state);
1375 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) {
1377 * Only allow this icmp though if the
1378 * encapsulated packet was allowed through the
1379 * other way around. Note that the minimal amount
1380 * of info present does not allow for checking against
1381 * tcp internals such as seq and ack numbers. Only the
1382 * ports are known to be present and can be even if the
1383 * short flag is set.
1385 if ((is->is_p == pr) && (is->is_v == 4) &&
1386 (is->is_icmppkts < is->is_pkts) &&
1387 fr_matchsrcdst(is, src, dst, &ofin, tcp)) {
1389 ips_stats.iss_hits++;
1391 is->is_bytes += fin->fin_plen;
1393 * we deliberately do not touch the timeouts
1394 * for the accompanying state table entry.
1395 * It remains to be seen if that is correct. XXX
1400 RWLOCK_EXIT(&ipf_state);
1406 * Move a state hash table entry from its old location at is->is_hv to
1407 * its new location, indexed by hv % fr_statesize.
1409 static void fr_ipsmove(isp, is, hv)
1410 ipstate_t **isp, *is;
1417 * Remove the hash from the old location...
1420 is->is_hnext->is_phnext = isp;
1421 *isp = is->is_hnext;
1422 if (ips_table[hvm] == NULL)
1423 ips_stats.iss_inuse--;
1426 * ...and put the hash in the new one.
1428 hvm = hv % fr_statesize;
1430 isp = &ips_table[hvm];
1432 (*isp)->is_phnext = &is->is_hnext;
1434 ips_stats.iss_inuse++;
1435 is->is_phnext = isp;
1436 is->is_hnext = *isp;
1442 * Check if a packet has a registered state.
1444 frentry_t *fr_checkstate(ip, fin)
1448 union i6addr dst, src;
1449 ipstate_t *is, **isp;
1451 u_int hv, hvm, hlen, tryagain, pass, v;
1457 if ((ips_list == NULL) || (fin->fin_off != 0) || fr_state_lock ||
1458 (fin->fin_fl & FI_SHORT))
1462 hlen = fin->fin_hlen;
1463 tcp = (tcphdr_t *)((char *)ip + hlen);
1464 ic = (struct icmp *)tcp;
1465 hv = (pr = fin->fin_fi.fi_p);
1466 src = fin->fin_fi.fi_src;
1467 dst = fin->fin_fi.fi_dst;
1468 hv += src.in4.s_addr;
1469 hv += dst.in4.s_addr;
1472 * Search the hash table for matching packet header info.
1473 * At the bottom of this switch statement, the following is expected:
1474 * is == NULL, no lock on ipf_state is held.
1475 * is != NULL, a lock on ipf_state is held.
1477 v = fin->fin_fi.fi_v;
1480 hv += fin->fin_fi.fi_src.i6[1];
1481 hv += fin->fin_fi.fi_src.i6[2];
1482 hv += fin->fin_fi.fi_src.i6[3];
1484 if ((fin->fin_p == IPPROTO_ICMPV6) &&
1485 IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_dst.in6)) {
1486 hv -= dst.in4.s_addr;
1488 hv += fin->fin_fi.fi_dst.i6[1];
1489 hv += fin->fin_fi.fi_dst.i6[2];
1490 hv += fin->fin_fi.fi_dst.i6[3];
1498 case IPPROTO_ICMPV6 :
1502 if ((ic->icmp_type == ICMP6_ECHO_REQUEST) ||
1503 (ic->icmp_type == ICMP6_ECHO_REPLY)) {
1508 READ_ENTER(&ipf_state);
1510 hvm = hv % fr_statesize;
1511 for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext)
1512 if ((is->is_p == pr) && (is->is_v == v) &&
1513 fr_matchsrcdst(is, src, dst, fin, NULL) &&
1514 fr_matchicmpqueryreply(v, is, ic, fin->fin_rev)) {
1516 if (is->is_frage[rev] != 0)
1517 is->is_age = is->is_frage[rev];
1519 is->is_age = fr_icmpacktimeout;
1521 is->is_age = fr_icmptimeout;
1526 if (tryagain && !(is->is_flags & FI_W_DADDR)) {
1527 hv += fin->fin_fi.fi_src.i6[0];
1528 hv += fin->fin_fi.fi_src.i6[1];
1529 hv += fin->fin_fi.fi_src.i6[2];
1530 hv += fin->fin_fi.fi_src.i6[3];
1531 fr_ipsmove(isp, is, hv);
1532 MUTEX_DOWNGRADE(&ipf_state);
1536 RWLOCK_EXIT(&ipf_state);
1539 * No matching icmp state entry. Perhaps this is a
1540 * response to another state entry.
1542 if ((ips_wild != 0) && (v == 6) && (tryagain == 0) &&
1543 !IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_src.in6)) {
1544 hv -= fin->fin_fi.fi_src.i6[0];
1545 hv -= fin->fin_fi.fi_src.i6[1];
1546 hv -= fin->fin_fi.fi_src.i6[2];
1547 hv -= fin->fin_fi.fi_src.i6[3];
1549 WRITE_ENTER(&ipf_state);
1553 fr = fr_checkicmp6matchingstate((ip6_t *)ip, fin);
1564 hvm = hv % fr_statesize;
1565 READ_ENTER(&ipf_state);
1566 for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext)
1567 if ((is->is_p == pr) && (is->is_v == v) &&
1568 fr_matchsrcdst(is, src, dst, fin, NULL) &&
1569 fr_matchicmpqueryreply(v, is, ic, fin->fin_rev)) {
1571 if (is->is_frage[rev] != 0)
1572 is->is_age = is->is_frage[rev];
1573 else if (fin->fin_rev)
1574 is->is_age = fr_icmpacktimeout;
1576 is->is_age = fr_icmptimeout;
1582 RWLOCK_EXIT(&ipf_state);
1584 * No matching icmp state entry. Perhaps this is a
1585 * response to another state entry.
1587 fr = fr_checkicmpmatchingstate(ip, fin);
1593 * Just plain ignore RST flag set with either FIN or SYN.
1595 if ((tcp->th_flags & TH_RST) &&
1596 ((tcp->th_flags & (TH_FIN|TH_SYN|TH_RST)) != TH_RST))
1600 u_short dport, sport;
1602 dport = tcp->th_dport;
1603 sport = tcp->th_sport;
1607 READ_ENTER(&ipf_state);
1609 hvm = hv % fr_statesize;
1610 for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext)
1611 if ((is->is_p == pr) && (is->is_v == v) &&
1612 fr_matchsrcdst(is, src, dst, fin, tcp)) {
1614 if ((pr == IPPROTO_TCP)) {
1615 if (!fr_tcpstate(is, fin, ip, tcp))
1617 } else if ((pr == IPPROTO_UDP)) {
1618 if (is->is_frage[rev] != 0)
1619 is->is_age = is->is_frage[rev];
1620 else if (fin->fin_rev)
1621 is->is_age = fr_udpacktimeout;
1623 is->is_age = fr_udptimeout;
1629 !(is->is_flags & (FI_WILDP|FI_WILDA))) {
1632 fr_ipsmove(isp, is, hv);
1633 MUTEX_DOWNGRADE(&ipf_state);
1638 RWLOCK_EXIT(&ipf_state);
1639 if (!tryagain && ips_wild) {
1643 WRITE_ENTER(&ipf_state);
1651 READ_ENTER(&ipf_state);
1652 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) {
1653 if ((is->is_p == pr) && (is->is_v == v) &&
1654 fr_matchsrcdst(is, src, dst, fin, NULL)) {
1656 if (is->is_frage[rev] != 0)
1657 is->is_age = is->is_frage[rev];
1659 is->is_age = fr_udptimeout;
1664 RWLOCK_EXIT(&ipf_state);
1670 ATOMIC_INCL(ips_stats.iss_miss);
1674 MUTEX_ENTER(&is->is_lock);
1675 is->is_bytes += fin->fin_plen;
1676 ips_stats.iss_hits++;
1678 MUTEX_EXIT(&is->is_lock);
1680 fin->fin_rule = is->is_rulen;
1682 fin->fin_group = fr->fr_group;
1683 fin->fin_icode = fr->fr_icode;
1687 RWLOCK_EXIT(&ipf_state);
1688 if ((fin->fin_fl & FI_FRAG) && (pass & FR_KEEPFRAG))
1689 ipfr_newfrag(ip, fin);
1691 if ((tcp != NULL) && (tcp->th_flags & TCP_CLOSE))
1699 * Sync. state entries. If interfaces come or go or just change position,
1702 void ip_statesync(ifp)
1708 WRITE_ENTER(&ipf_state);
1709 for (is = ips_list; is; is = is->is_next) {
1710 for (i = 0; i < 4; i++) {
1711 if (is->is_ifp[i] == ifp) {
1712 is->is_ifp[i] = GETUNIT(is->is_ifname[i],
1715 is->is_ifp[i] = (void *)-1;
1719 RWLOCK_EXIT(&ipf_state);
1724 * Must always be called with fr_ipfstate held as a write lock.
1726 static void fr_delstate(is)
1731 if (is->is_flags & (FI_WILDP|FI_WILDA))
1734 is->is_next->is_pnext = is->is_pnext;
1735 *is->is_pnext = is->is_next;
1737 is->is_hnext->is_phnext = is->is_phnext;
1738 *is->is_phnext = is->is_hnext;
1739 if (ips_table[is->is_hv] == NULL)
1740 ips_stats.iss_inuse--;
1747 if (fr->fr_ref == 0) {
1752 MUTEX_DESTROY(&is->is_lock);
1760 * Free memory in use by all state info. kept.
1762 void fr_stateunload()
1766 WRITE_ENTER(&ipf_state);
1767 while ((is = ips_list))
1769 ips_stats.iss_inuse = 0;
1771 RWLOCK_EXIT(&ipf_state);
1773 KFREES(ips_table, fr_statesize * sizeof(ipstate_t *));
1779 * Slowly expire held state for thingslike UDP and ICMP. Timeouts are set
1780 * in expectation of this being called twice per second.
1782 void fr_timeoutstate()
1784 ipstate_t *is, **isp;
1785 #if defined(_KERNEL) && !SOLARIS
1790 WRITE_ENTER(&ipf_state);
1791 for (isp = &ips_list; (is = *isp); )
1792 if (is->is_age && !--is->is_age) {
1793 if (is->is_p == IPPROTO_TCP)
1794 ips_stats.iss_fin++;
1796 ips_stats.iss_expire++;
1798 ipstate_log(is, ISL_EXPIRE);
1803 if (fr_state_doflush) {
1804 (void) fr_state_flush(2, 0);
1805 fr_state_doflush = 0;
1807 RWLOCK_EXIT(&ipf_state);
1813 * Original idea freom Pradeep Krishnan for use primarily with NAT code.
1814 * (pkrishna@netcom.com)
1816 * Rewritten by Arjan de Vet <Arjan.deVet@adv.iae.nl>, 2000-07-29:
1818 * - (try to) base state transitions on real evidence only,
1819 * i.e. packets that are sent and have been received by ipfilter;
1820 * diagram 18.12 of TCP/IP volume 1 by W. Richard Stevens was used.
1822 * - deal with half-closed connections correctly;
1824 * - store the state of the source in state[0] such that ipfstat
1825 * displays the state as source/dest instead of dest/source; the calls
1826 * to fr_tcp_age have been changed accordingly.
1830 * state[0] = state of source (host that initiated connection)
1831 * state[1] = state of dest (host that accepted the connection)
1833 * dir == 0 : a packet from source to dest
1834 * dir == 1 : a packet from dest to source
1837 int fr_tcp_age(age, state, fin, dir, fsm)
1843 tcphdr_t *tcp = (tcphdr_t *)fin->fin_dp;
1844 u_char flags = tcp->th_flags;
1848 ostate = state[1 - dir];
1850 dlen = fin->fin_plen - fin->fin_hlen - (tcp->th_off << 2);
1852 if (flags & TH_RST) {
1853 if (!(tcp->th_flags & TH_PUSH) && !dlen) {
1854 *age = fr_tcpclosed;
1855 state[dir] = TCPS_CLOSED;
1857 *age = fr_tcpclosewait;
1858 state[dir] = TCPS_CLOSE_WAIT;
1867 case TCPS_CLOSED: /* 0 */
1868 if ((flags & TH_OPENING) == TH_OPENING) {
1870 * 'dir' received an S and sends SA in response,
1871 * CLOSED -> SYN_RECEIVED
1873 state[dir] = TCPS_SYN_RECEIVED;
1874 newage = fr_tcptimeout;
1875 } else if ((flags & TH_OPENING) == TH_SYN) {
1876 /* 'dir' sent S, CLOSED -> SYN_SENT */
1877 state[dir] = TCPS_SYN_SENT;
1878 newage = fr_tcptimeout;
1882 * It is apparently possible that a hosts sends two syncs
1883 * before the remote party is able to respond with a SA. In
1884 * such a case the remote server sometimes ACK's the second
1885 * sync, and then responds with a SA. The following code
1886 * is used to prevent this ack from being blocked.
1888 * We do not reset the timeout here to fr_tcptimeout because
1889 * a connection connect timeout does not renew after every
1890 * packet that is sent. We need to set newage to something
1891 * to indicate the packet has passed the check for its flags
1892 * being valid in the TCP FSM.
1894 else if ((ostate == TCPS_SYN_SENT) &&
1895 ((flags & (TH_FIN|TH_SYN|TH_RST|TH_ACK)) == TH_ACK)) {
1900 * The next piece of code makes it possible to get
1901 * already established connections into the state table
1902 * after a restart or reload of the filter rules; this
1903 * does not work when a strict 'flags S keep state' is
1904 * used for tcp connections of course, however, use a
1905 * lower time-out so the state disappears quickly if
1906 * the other side does not pick it up.
1909 (flags & (TH_FIN|TH_SYN|TH_RST|TH_ACK)) == TH_ACK) {
1910 /* we saw an A, guess 'dir' is in ESTABLISHED mode */
1911 if (ostate == TCPS_CLOSED) {
1912 state[dir] = TCPS_ESTABLISHED;
1913 newage = fr_tcptimeout;
1914 } else if (ostate == TCPS_ESTABLISHED) {
1915 state[dir] = TCPS_ESTABLISHED;
1916 newage = fr_tcpidletimeout;
1920 * TODO: besides regular ACK packets we can have other
1921 * packets as well; it is yet to be determined how we
1922 * should initialize the states in those cases
1926 case TCPS_LISTEN: /* 1 */
1930 case TCPS_SYN_SENT: /* 2 */
1931 if ((flags & ~(TH_ECN|TH_CWR)) == TH_SYN) {
1933 * A retransmitted SYN packet. We do not reset the
1934 * timeout here to fr_tcptimeout because a connection
1935 * connect timeout does not renew after every packet
1936 * that is sent. We need to set newage to something
1937 * to indicate the packet has passed the check for its
1938 * flags being valid in the TCP FSM.
1941 } else if ((flags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) {
1943 * We see an A from 'dir' which is in SYN_SENT
1944 * state: 'dir' sent an A in response to an SA
1945 * which it received, SYN_SENT -> ESTABLISHED
1947 state[dir] = TCPS_ESTABLISHED;
1948 newage = fr_tcpidletimeout;
1949 } else if (flags & TH_FIN) {
1951 * We see an F from 'dir' which is in SYN_SENT
1952 * state and wants to close its side of the
1953 * connection; SYN_SENT -> FIN_WAIT_1
1955 state[dir] = TCPS_FIN_WAIT_1;
1956 newage = fr_tcpidletimeout; /* or fr_tcptimeout? */
1957 } else if ((flags & TH_OPENING) == TH_OPENING) {
1959 * We see an SA from 'dir' which is already in
1960 * SYN_SENT state, this means we have a
1961 * simultaneous open; SYN_SENT -> SYN_RECEIVED
1963 state[dir] = TCPS_SYN_RECEIVED;
1964 newage = fr_tcptimeout;
1968 case TCPS_SYN_RECEIVED: /* 3 */
1969 if ((flags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) {
1971 * We see an A from 'dir' which was in SYN_RECEIVED
1972 * state so it must now be in established state,
1973 * SYN_RECEIVED -> ESTABLISHED
1975 state[dir] = TCPS_ESTABLISHED;
1976 newage = fr_tcpidletimeout;
1977 } else if ((flags & ~(TH_ECN|TH_CWR)) == TH_OPENING) {
1979 * We see an SA from 'dir' which is already in
1980 * SYN_RECEIVED state.
1982 newage = fr_tcptimeout;
1983 } else if (flags & TH_FIN) {
1985 * We see an F from 'dir' which is in SYN_RECEIVED
1986 * state and wants to close its side of the connection;
1987 * SYN_RECEIVED -> FIN_WAIT_1
1989 state[dir] = TCPS_FIN_WAIT_1;
1990 newage = fr_tcpidletimeout;
1994 case TCPS_ESTABLISHED: /* 4 */
1995 if (flags & TH_FIN) {
1997 * 'dir' closed its side of the connection; this
1998 * gives us a half-closed connection;
1999 * ESTABLISHED -> FIN_WAIT_1
2001 state[dir] = TCPS_FIN_WAIT_1;
2002 newage = fr_tcphalfclosed;
2003 } else if (flags & TH_ACK) {
2004 /* an ACK, should we exclude other flags here? */
2005 if (ostate == TCPS_FIN_WAIT_1) {
2007 * We know the other side did an active close,
2008 * so we are ACKing the recvd FIN packet (does
2009 * the window matching code guarantee this?)
2010 * and go into CLOSE_WAIT state; this gives us
2011 * a half-closed connection
2013 state[dir] = TCPS_CLOSE_WAIT;
2014 newage = fr_tcphalfclosed;
2015 } else if (ostate < TCPS_CLOSE_WAIT)
2017 * Still a fully established connection,
2020 newage = fr_tcpidletimeout;
2024 case TCPS_CLOSE_WAIT: /* 5 */
2025 if (flags & TH_FIN) {
2027 * Application closed and 'dir' sent a FIN, we're now
2028 * going into LAST_ACK state
2030 newage = fr_tcplastack;
2031 state[dir] = TCPS_LAST_ACK;
2034 * We remain in CLOSE_WAIT because the other side has
2035 * closed already and we did not close our side yet;
2038 newage = fr_tcphalfclosed;
2042 case TCPS_FIN_WAIT_1: /* 6 */
2043 if ((flags & TH_ACK) && ostate > TCPS_CLOSE_WAIT) {
2045 * If the other side is not active anymore it has sent
2046 * us a FIN packet that we are ack'ing now with an ACK;
2047 * this means both sides have now closed the connection
2048 * and we go into TIME_WAIT
2051 * XXX: how do we know we really are ACKing the FIN
2052 * packet here? does the window code guarantee that?
2054 state[dir] = TCPS_TIME_WAIT;
2055 newage = fr_tcptimeout;
2058 * We closed our side of the connection already but the
2059 * other side is still active (ESTABLISHED/CLOSE_WAIT);
2060 * continue with this half-closed connection
2062 newage = fr_tcphalfclosed;
2065 case TCPS_CLOSING: /* 7 */
2069 case TCPS_LAST_ACK: /* 8 */
2070 if (flags & TH_ACK) {
2071 if ((flags & TH_PUSH) || dlen)
2073 * There is still data to be delivered, reset
2076 newage = fr_tcplastack;
2081 * We cannot detect when we go out of LAST_ACK state to CLOSED
2082 * because that is based on the reception of ACK packets;
2083 * ipfilter can only detect that a packet has been sent by a
2088 case TCPS_FIN_WAIT_2: /* 9 */
2092 case TCPS_TIME_WAIT: /* 10 */
2093 newage = fr_tcptimeout; /* default 4 mins */
2094 /* we're in 2MSL timeout now */
2107 void ipstate_log(is, type)
2116 ipsl.isl_type = type;
2117 ipsl.isl_pkts = is->is_pkts + is->is_icmppkts;
2118 ipsl.isl_bytes = is->is_bytes;
2119 ipsl.isl_src = is->is_src;
2120 ipsl.isl_dst = is->is_dst;
2121 ipsl.isl_p = is->is_p;
2122 ipsl.isl_v = is->is_v;
2123 ipsl.isl_flags = is->is_flags;
2124 ipsl.isl_rulen = is->is_rulen;
2125 ipsl.isl_group = is->is_group;
2126 if (ipsl.isl_p == IPPROTO_TCP || ipsl.isl_p == IPPROTO_UDP) {
2127 ipsl.isl_sport = is->is_sport;
2128 ipsl.isl_dport = is->is_dport;
2129 if (ipsl.isl_p == IPPROTO_TCP) {
2130 ipsl.isl_state[0] = is->is_state[0];
2131 ipsl.isl_state[1] = is->is_state[1];
2133 } else if (ipsl.isl_p == IPPROTO_ICMP) {
2134 ipsl.isl_itype = is->is_icmp.ics_type;
2135 } else if (ipsl.isl_p == IPPROTO_ICMPV6) {
2136 ipsl.isl_itype = is->is_icmp.ics_type;
2138 ipsl.isl_ps.isl_filler[0] = 0;
2139 ipsl.isl_ps.isl_filler[1] = 0;
2142 sizes[0] = sizeof(ipsl);
2145 if (ipllog(IPL_LOGSTATE, NULL, items, sizes, types, 1)) {
2146 ATOMIC_INCL(ips_stats.iss_logged);
2148 ATOMIC_INCL(ips_stats.iss_logfail);
2155 frentry_t *fr_checkicmp6matchingstate(ip, fin)
2159 ipstate_t *is, **isp;
2160 u_short sport, dport;
2162 struct icmp6_hdr *ic, *oic;
2163 union i6addr dst, src;
2173 * Does it at least have the return (basic) IP header ?
2174 * Only a basic IP header (no options) should be with
2175 * an ICMP error header.
2177 if ((fin->fin_v != 6) || (fin->fin_plen < ICMP6ERR_MINPKTLEN))
2179 ic = (struct icmp6_hdr *)fin->fin_dp;
2180 type = ic->icmp6_type;
2182 * If it's not an error type, then return
2184 if ((type != ICMP6_DST_UNREACH) && (type != ICMP6_PACKET_TOO_BIG) &&
2185 (type != ICMP6_TIME_EXCEEDED) && (type != ICMP6_PARAM_PROB))
2188 oip = (ip6_t *)((char *)ic + ICMPERR_ICMPHLEN);
2189 if (fin->fin_plen < sizeof(*oip))
2192 if ((oip->ip6_nxt != IPPROTO_TCP) && (oip->ip6_nxt != IPPROTO_UDP) &&
2193 (oip->ip6_nxt != IPPROTO_ICMPV6))
2196 bzero((char *)&ofin, sizeof(ofin));
2197 ofin.fin_out = !fin->fin_out;
2198 ofin.fin_ifp = fin->fin_ifp;
2202 ofin.fin_qfm = NULL;
2205 * We make a fin entry to be able to feed it to
2206 * matchsrcdst. Note that not all fields are necessary
2207 * but this is the cleanest way. Note further we fill
2208 * in fin_mp such that if someone uses it we'll get
2209 * a kernel panic. fr_matchsrcdst does not use this.
2211 * watch out here, as ip is in host order and oip in network
2212 * order. Any change we make must be undone afterwards.
2214 savelen = oip->ip6_plen;
2215 oip->ip6_plen = ip->ip6_plen - sizeof(*ip) - ICMPERR_ICMPHLEN;
2216 fr_makefrip(sizeof(*oip), (ip_t *)oip, &ofin);
2217 oip->ip6_plen = savelen;
2219 if (oip->ip6_nxt == IPPROTO_ICMPV6) {
2220 oic = (struct icmp6_hdr *)(oip + 1);
2222 * an ICMP error can only be generated as a result of an
2223 * ICMP query, not as the response on an ICMP error
2225 * XXX theoretically ICMP_ECHOREP and the other reply's are
2226 * ICMP query's as well, but adding them here seems strange XXX
2228 if (!(oic->icmp6_type & ICMP6_INFOMSG_MASK))
2232 * perform a lookup of the ICMP packet in the state table
2234 hv = (pr = oip->ip6_nxt);
2235 src.in6 = oip->ip6_src;
2236 hv += src.in4.s_addr;
2237 dst.in6 = oip->ip6_dst;
2238 hv += dst.in4.s_addr;
2239 hv += oic->icmp6_id;
2240 hv += oic->icmp6_seq;
2243 READ_ENTER(&ipf_state);
2244 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext)
2245 if ((is->is_p == pr) &&
2246 (oic->icmp6_id == is->is_icmp.ics_id) &&
2247 (oic->icmp6_seq == is->is_icmp.ics_seq) &&
2248 fr_matchsrcdst(is, src, dst, &ofin, NULL)) {
2250 * in the state table ICMP query's are stored
2251 * with the type of the corresponding ICMP
2252 * response. Correct here
2254 if (((is->is_type == ICMP6_ECHO_REPLY) &&
2255 (oic->icmp6_type == ICMP6_ECHO_REQUEST)) ||
2256 (is->is_type - 1 == oic->icmp6_type )) {
2257 ips_stats.iss_hits++;
2259 is->is_bytes += fin->fin_plen;
2263 RWLOCK_EXIT(&ipf_state);
2268 tcp = (tcphdr_t *)(oip + 1);
2269 dport = tcp->th_dport;
2270 sport = tcp->th_sport;
2272 hv = (pr = oip->ip6_nxt);
2273 src.in6 = oip->ip6_src;
2274 hv += src.in4.s_addr;
2278 dst.in6 = oip->ip6_dst;
2279 hv += dst.in4.s_addr;
2287 READ_ENTER(&ipf_state);
2288 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) {
2290 * Only allow this icmp though if the
2291 * encapsulated packet was allowed through the
2292 * other way around. Note that the minimal amount
2293 * of info present does not allow for checking against
2294 * tcp internals such as seq and ack numbers.
2296 if ((is->is_p == pr) && (is->is_v == 6) &&
2297 fr_matchsrcdst(is, src, dst, &ofin, tcp)) {
2299 ips_stats.iss_hits++;
2301 is->is_bytes += fin->fin_plen;
2303 * we deliberately do not touch the timeouts
2304 * for the accompanying state table entry.
2305 * It remains to be seen if that is correct. XXX
2307 RWLOCK_EXIT(&ipf_state);
2311 RWLOCK_EXIT(&ipf_state);