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 1.7 2004/07/28 00:22:37 hmp 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));
616 * Copy and calculate...
618 hv = (is->is_p = fin->fin_fi.fi_p);
619 is->is_src = fin->fin_fi.fi_src;
621 is->is_dst = fin->fin_fi.fi_dst;
624 if (fin->fin_v == 6) {
625 if ((is->is_p == IPPROTO_ICMPV6) &&
626 IN6_IS_ADDR_MULTICAST(&is->is_dst.in6)) {
628 * So you can do keep state with neighbour discovery.
633 hv += is->is_dst.i6[1];
634 hv += is->is_dst.i6[2];
635 hv += is->is_dst.i6[3];
637 hv += is->is_src.i6[1];
638 hv += is->is_src.i6[2];
639 hv += is->is_src.i6[3];
648 case IPPROTO_ICMPV6 :
649 ic = (struct icmp *)fin->fin_dp;
650 if ((ic->icmp_type & ICMP6_INFOMSG_MASK) == 0)
653 switch (ic->icmp_type)
655 case ICMP6_ECHO_REQUEST :
656 is->is_icmp.ics_type = ic->icmp_type;
657 hv += (is->is_icmp.ics_id = ic->icmp_id);
658 hv += (is->is_icmp.ics_seq = ic->icmp_seq);
660 case ICMP6_MEMBERSHIP_QUERY :
661 case ND_ROUTER_SOLICIT :
662 case ND_NEIGHBOR_SOLICIT :
663 case ICMP6_NI_QUERY :
664 is->is_icmp.ics_type = ic->icmp_type;
669 ATOMIC_INCL(ips_stats.iss_icmp);
670 is->is_age = fr_icmptimeout;
674 ic = (struct icmp *)fin->fin_dp;
676 switch (ic->icmp_type)
682 is->is_icmp.ics_type = ic->icmp_type;
683 hv += (is->is_icmp.ics_id = ic->icmp_id);
684 hv += (is->is_icmp.ics_seq = ic->icmp_seq);
689 ATOMIC_INCL(ips_stats.iss_icmp);
690 is->is_age = fr_icmptimeout;
693 tcp = (tcphdr_t *)fin->fin_dp;
695 if (tcp->th_flags & TH_RST)
698 * The endian of the ports doesn't matter, but the ack and
699 * sequence numbers do as we do mathematics on them later.
701 is->is_sport = htons(fin->fin_data[0]);
702 is->is_dport = htons(fin->fin_data[1]);
703 if ((flags & (FI_W_DPORT|FI_W_SPORT)) == 0) {
707 if ((flags & FI_IGNOREPKT) == 0) {
708 is->is_send = ntohl(tcp->th_seq) + fin->fin_dlen -
709 (off = (tcp->th_off << 2)) +
710 ((tcp->th_flags & TH_SYN) ? 1 : 0) +
711 ((tcp->th_flags & TH_FIN) ? 1 : 0);
712 is->is_maxsend = is->is_send;
714 if ((tcp->th_flags & TH_SYN) &&
715 ((tcp->th_off << 2) >= (sizeof(*tcp) + 4))) {
716 ws = fr_tcpoptions(tcp);
723 is->is_maxswin = ntohs(tcp->th_win);
724 if (is->is_maxswin == 0)
727 if ((tcp->th_flags & TH_OPENING) == TH_SYN)
731 * If we're creating state for a starting connection, start the
732 * timer on it as we'll never see an error if it fails to
735 ATOMIC_INCL(ips_stats.iss_tcp);
739 tcp = (tcphdr_t *)fin->fin_dp;
741 is->is_sport = htons(fin->fin_data[0]);
742 is->is_dport = htons(fin->fin_data[1]);
743 if ((flags & (FI_W_DPORT|FI_W_SPORT)) == 0) {
747 ATOMIC_INCL(ips_stats.iss_udp);
748 is->is_age = fr_udptimeout;
751 is->is_age = fr_udptimeout;
755 KMALLOC(is, ipstate_t *);
757 ATOMIC_INCL(ips_stats.iss_nomem);
760 bcopy((char *)&ips, (char *)is, sizeof(*is));
763 is->is_rule = fin->fin_fr;
764 if (is->is_rule != NULL) {
765 is->is_group = is->is_rule->fr_group;
766 ATOMIC_INC32(is->is_rule->fr_ref);
767 pass = is->is_rule->fr_flags;
768 is->is_frage[0] = is->is_rule->fr_age[0];
769 is->is_frage[1] = is->is_rule->fr_age[1];
770 if (is->is_frage[0] != 0)
771 is->is_age = is->is_frage[0];
773 is->is_ifp[(out << 1) + 1] = is->is_rule->fr_ifas[1];
774 is->is_ifp[(1 - out) << 1] = is->is_rule->fr_ifas[2];
775 is->is_ifp[((1 - out) << 1) + 1] = is->is_rule->fr_ifas[3];
777 if (((ifp = is->is_rule->fr_ifas[1]) != NULL) &&
779 strncpy(is->is_ifname[(out << 1) + 1],
780 IFNAME(ifp), IFNAMSIZ);
781 if (((ifp = is->is_rule->fr_ifas[2]) != NULL) &&
783 strncpy(is->is_ifname[(1 - out) << 1],
784 IFNAME(ifp), IFNAMSIZ);
785 if (((ifp = is->is_rule->fr_ifas[3]) != NULL) &&
787 strncpy(is->is_ifname[((1 - out) << 1) + 1],
788 IFNAME(ifp), IFNAMSIZ);
792 is->is_ifp[out << 1] = fin->fin_ifp;
793 strncpy(is->is_ifname[out << 1], IFNAME(fin->fin_ifp), IFNAMSIZ);
795 WRITE_ENTER(&ipf_state);
798 if ((flags & FI_IGNOREPKT) == 0) {
800 is->is_bytes = fin->fin_dlen + fin->fin_hlen;
803 * We want to check everything that is a property of this packet,
804 * but we don't (automatically) care about it's fragment status as
807 is->is_v = fin->fin_v;
808 is->is_rulen = fin->fin_rule;
809 is->is_opt = fin->fin_fi.fi_optmsk;
810 is->is_optmsk = 0xffffffff;
811 is->is_sec = fin->fin_fi.fi_secmsk;
812 is->is_secmsk = 0xffff;
813 is->is_auth = fin->fin_fi.fi_auth;
814 is->is_authmsk = 0xffff;
815 is->is_flags = fin->fin_fl & FI_CMP;
816 is->is_flags |= FI_CMP << 4;
817 is->is_flags |= flags & (FI_WILDP|FI_WILDA);
818 if (flags & (FI_WILDP|FI_WILDA))
821 if (pass & FR_LOGFIRST)
822 is->is_pass &= ~(FR_LOGFIRST|FR_LOG);
825 if (is->is_p == IPPROTO_TCP) {
826 fr_tcp_age(&is->is_age, is->is_state, fin,
827 0, is->is_fsm); /* 0 = packet from the source */
830 ipstate_log(is, ISL_NEW);
832 RWLOCK_EXIT(&ipf_state);
833 fin->fin_rev = IP6NEQ(is->is_dst, fin->fin_fi.fi_dst);
834 if ((fin->fin_fl & FI_FRAG) && (pass & FR_KEEPFRAG))
835 ipfr_newfrag(ip, fin);
840 static int fr_tcpoptions(tcp)
846 opt = (u_char *) (tcp + 1);
847 last = ((u_char *)tcp) + (tcp->th_off << 2);
849 /* If we don't find wscale here, we need to clear it */
852 /* Termination condition picked such that opt[0 .. 2] exist */
853 while ((opt < last - 2) && (*opt != TCPOPT_EOL)) {
859 /* Proper length ? */
868 /* Unknown options must be two bytes+ */
882 * check to see if a packet with TCP headers fits within the TCP window.
883 * change timeout depending on whether new packet is a SYN-ACK returning for a
884 * SYN or a RST or FIN which indicate time to close up shop.
886 int fr_tcpstate(is, fin, ip, tcp)
892 tcp_seq seq, ack, end;
894 tcpdata_t *fdata, *tdata;
901 * Find difference between last checked packet and this packet.
903 source = IP6EQ(fin->fin_fi.fi_src, is->is_src);
904 if (source && (ntohs(is->is_sport) != fin->fin_data[0]))
906 fdata = &is->is_tcp.ts_data[!source];
907 tdata = &is->is_tcp.ts_data[source];
908 off = tcp->th_off << 2;
909 seq = ntohl(tcp->th_seq);
910 ack = ntohl(tcp->th_ack);
911 win = ntohs(tcp->th_win);
912 end = seq + fin->fin_dlen - off +
913 ((tcp->th_flags & TH_SYN) ? 1 : 0) +
914 ((tcp->th_flags & TH_FIN) ? 1 : 0);
917 if ((tcp->th_flags & TH_SYN) && (off >= sizeof(*tcp) + 4))
918 wscale = fr_tcpoptions(tcp);
922 MUTEX_ENTER(&is->is_lock);
925 fdata->td_wscale = wscale;
926 else if (wscale == -2)
927 fdata->td_wscale = tdata->td_wscale = 0;
928 if (!(tcp->th_flags & TH_SYN))
929 win <<= fdata->td_wscale;
931 if ((fdata->td_end == 0) &&
932 (!is->is_fsm || ((tcp->th_flags & TH_OPENING) == TH_OPENING))) {
934 * Must be a (outgoing) SYN-ACK in reply to a SYN.
937 fdata->td_maxwin = 1;
938 fdata->td_maxend = end + win;
943 if (!(tcp->th_flags & TH_ACK)) { /* Pretend an ack was sent */
945 } else if (((tcp->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) &&
947 /* gross hack to get around certain broken tcp stacks */
952 seq = end = fdata->td_end;
954 maxwin = tdata->td_maxwin;
955 ackskew = tdata->td_end - ack;
957 #define SEQ_GE(a,b) ((int)((a) - (b)) >= 0)
958 #define SEQ_GT(a,b) ((int)((a) - (b)) > 0)
959 if ((SEQ_GE(fdata->td_maxend, end)) &&
960 (SEQ_GE(seq, fdata->td_end - maxwin)) &&
961 /* XXX what about big packets */
962 #define MAXACKWINDOW 66000
963 (-ackskew <= (MAXACKWINDOW << tdata->td_wscale)) &&
964 ( ackskew <= (MAXACKWINDOW << tdata->td_wscale))) {
966 /* if ackskew < 0 then this should be due to fragmented
967 * packets. There is no way to know the length of the
968 * total packet in advance.
969 * We do know the total length from the fragment cache though.
970 * Note however that there might be more sessions with
971 * exactly the same source and destination parameters in the
972 * state cache (and source and destination is the only stuff
973 * that is saved in the fragment cache). Note further that
974 * some TCP connections in the state cache are hashed with
975 * sport and dport as well which makes it not worthwhile to
977 * Thus, when ackskew is negative but still seems to belong
978 * to this session, we bump up the destinations end value.
981 * Nearing end of connection, start timeout.
983 /* source ? 0 : 1 -> !source */
984 if (fr_tcp_age(&is->is_age, is->is_state, fin, !source,
985 (int)is->is_fsm) == 0) {
989 /* update max window seen */
990 if (fdata->td_maxwin < win)
991 fdata->td_maxwin = win;
992 if (SEQ_GT(end, fdata->td_end))
994 if (SEQ_GE(ack + win, tdata->td_maxend)) {
995 tdata->td_maxend = ack + win;
1000 ATOMIC_INCL(ips_stats.iss_hits);
1004 MUTEX_EXIT(&is->is_lock);
1005 if ((ret == 0) && ((tcp->th_flags & TH_OPENING) != TH_SYN))
1006 fin->fin_misc |= FM_BADSTATE;
1012 * Match a state table entry against an IP packet.
1014 static int fr_matchsrcdst(is, src, dst, fin, tcp)
1016 union i6addr src, dst;
1020 int ret = 0, rev, out, flags, idx;
1024 rev = IP6NEQ(is->is_dst, dst);
1027 flags = is->is_flags & (FI_WILDA|FI_WILDP);
1032 flags = is->is_flags;
1036 if (!(flags & FI_W_SPORT) && (sp != is->is_sport))
1038 else if (!(flags & FI_W_DPORT) && (dp != is->is_dport))
1043 idx = (out << 1) + rev;
1045 if ((is->is_ifp[idx] == NULL &&
1046 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) ||
1047 is->is_ifp[idx] == ifp)
1055 if ((IP6EQ(is->is_dst, dst) || (flags & FI_W_DADDR)) &&
1056 (IP6EQ(is->is_src, src) || (flags & FI_W_SADDR)) &&
1057 (!tcp || ((sp == is->is_sport || flags & FI_W_SPORT) &&
1058 (dp == is->is_dport || flags & FI_W_DPORT)))) {
1062 if ((IP6EQ(is->is_dst, src) || (flags & FI_W_DADDR)) &&
1063 (IP6EQ(is->is_src, dst) || (flags & FI_W_SADDR)) &&
1064 (!tcp || ((sp == is->is_dport || flags & FI_W_DPORT) &&
1065 (dp == is->is_sport || flags & FI_W_SPORT)))) {
1073 * Whether or not this should be here, is questionable, but the aim
1074 * is to get this out of the main line.
1077 flags = is->is_flags & (FI_CMP|(FI_CMP<<4));
1079 if (((fin->fin_fl & (flags >> 4)) != (flags & FI_CMP)) ||
1080 (fin->fin_fi.fi_optmsk != is->is_opt) ||
1081 (fin->fin_fi.fi_secmsk != is->is_sec) ||
1082 (fin->fin_fi.fi_auth != is->is_auth))
1085 flags = is->is_flags & (FI_WILDA|FI_WILDP);
1086 if ((flags & (FI_W_SADDR|FI_W_DADDR))) {
1087 if ((flags & FI_W_SADDR) != 0) {
1089 is->is_src = fin->fin_fi.fi_src;
1091 is->is_src = fin->fin_fi.fi_dst;
1093 } else if ((flags & FI_W_DADDR) != 0) {
1095 is->is_dst = fin->fin_fi.fi_dst;
1097 is->is_dst = fin->fin_fi.fi_src;
1100 is->is_flags &= ~(FI_W_SADDR|FI_W_DADDR);
1101 if ((is->is_flags & (FI_WILDA|FI_WILDP)) == 0)
1105 if ((flags & (FI_W_SPORT|FI_W_DPORT))) {
1106 if ((flags & FI_W_SPORT) != 0) {
1109 is->is_send = htonl(tcp->th_seq);
1112 is->is_send = htonl(tcp->th_ack);
1114 is->is_maxsend = is->is_send + 1;
1115 } else if ((flags & FI_W_DPORT) != 0) {
1118 is->is_dend = htonl(tcp->th_ack);
1121 is->is_dend = htonl(tcp->th_seq);
1123 is->is_maxdend = is->is_dend + 1;
1125 is->is_flags &= ~(FI_W_SPORT|FI_W_DPORT);
1131 if (is->is_ifp[idx] == NULL &&
1132 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*'))
1136 is->is_ifp[ret] = ifp;
1137 strncpy(is->is_ifname[ret], IFNAME(ifp),
1138 sizeof(is->is_ifname[ret]));
1144 static int fr_matchicmpqueryreply(v, is, icmp, rev)
1151 * If we matched its type on the way in, then when going out
1152 * it will still be the same type.
1154 if ((!rev && (icmp->icmp_type == is->is_type)) ||
1155 (rev && (icmpreplytype4[is->is_type] == icmp->icmp_type))) {
1156 if (icmp->icmp_type != ICMP_ECHOREPLY)
1158 if ((icmp->icmp_id == is->is_icmp.ics_id) &&
1159 (icmp->icmp_seq == is->is_icmp.ics_seq))
1164 else if (is->is_v == 6) {
1165 if ((!rev && (icmp->icmp_type == is->is_type)) ||
1166 (rev && (icmpreplytype6[is->is_type] == icmp->icmp_type))) {
1167 if (icmp->icmp_type != ICMP6_ECHO_REPLY)
1169 if ((icmp->icmp_id == is->is_icmp.ics_id) &&
1170 (icmp->icmp_seq == is->is_icmp.ics_seq))
1178 static frentry_t *fr_checkicmpmatchingstate(ip, fin)
1182 ipstate_t *is, **isp;
1183 u_short sport, dport;
1185 u_short savelen, ohlen;
1186 union i6addr dst, src;
1197 * Does it at least have the return (basic) IP header ?
1198 * Only a basic IP header (no options) should be with
1199 * an ICMP error header.
1201 if (((ip->ip_v != 4) || (ip->ip_hl != 5)) ||
1202 (fin->fin_plen < ICMPERR_MINPKTLEN))
1205 ic = (struct icmp *)fin->fin_dp;
1206 type = ic->icmp_type;
1208 * If it's not an error type, then return
1210 if ((type != ICMP_UNREACH) && (type != ICMP_SOURCEQUENCH) &&
1211 (type != ICMP_REDIRECT) && (type != ICMP_TIMXCEED) &&
1212 (type != ICMP_PARAMPROB))
1215 oip = (ip_t *)((char *)ic + ICMPERR_ICMPHLEN);
1216 ohlen = oip->ip_hl << 2;
1218 * Check if the at least the old IP header (with options) and
1219 * 8 bytes of payload is present.
1221 if (fin->fin_plen < ICMPERR_MAXPKTLEN + ohlen - sizeof(*oip))
1227 len = fin->fin_dlen - ICMPERR_ICMPHLEN;
1228 if ((len <= 0) || (ohlen > len))
1232 * Is the buffer big enough for all of it ? It's the size of the IP
1233 * header claimed in the encapsulated part which is of concern. It
1234 * may be too big to be in this buffer but not so big that it's
1235 * outside the ICMP packet, leading to TCP deref's causing problems.
1236 * This is possible because we don't know how big oip_hl is when we
1237 * do the pullup early in fr_check() and thus can't guarantee it is
1246 if ((char *)oip + len > (char *)m->b_wptr)
1249 m = *(mb_t **)fin->fin_mp;
1250 if ((char *)oip + len > (char *)ip + m->m_len)
1257 * in the IPv4 case we must zero the i6addr union otherwise
1258 * the IP6EQ and IP6NEQ macros produce the wrong results because
1259 * of the 'junk' in the unused part of the union
1261 bzero((char *)&src, sizeof(src));
1262 bzero((char *)&dst, sizeof(dst));
1263 bzero((char *)&ofin, sizeof(ofin));
1265 * We make an fin entry to be able to feed it to
1266 * matchsrcdst. Note that not all fields are encessary
1267 * but this is the cleanest way. Note further that we
1268 * fill in fin_mp such that if someone uses it we'll get
1269 * a kernel panic. fr_matchsrcdst does not use this.
1271 ofin.fin_ifp = fin->fin_ifp;
1272 ofin.fin_out = !fin->fin_out;
1276 * watch out here, as ip is in host order and oip in network
1277 * order. Any change we make must be undone afterwards, like
1278 * oip->ip_off - it is still in network byte order so fix it.
1280 savelen = oip->ip_len;
1282 oip->ip_off = ntohs(oip->ip_off);
1283 (void) fr_makefrip(ohlen, oip, &ofin);
1285 * Reset the short flag here because in fr_matchsrcdst() the flags
1286 * for the current packet (fin_fl) are compared against * those for
1287 * the existing session.
1289 ofin.fin_fl &= ~FI_SHORT;
1292 * Put old values of ip_len and ip_off back as we don't know
1293 * if we have to forward the packet (or process it again.
1295 oip->ip_len = savelen;
1296 oip->ip_off = htons(oip->ip_off);
1299 ofin.fin_qfm = NULL;
1306 icmp = (icmphdr_t *)((char *)oip + ohlen);
1309 * an ICMP error can only be generated as a result of an
1310 * ICMP query, not as the response on an ICMP error
1312 * XXX theoretically ICMP_ECHOREP and the other reply's are
1313 * ICMP query's as well, but adding them here seems strange XXX
1315 if ((icmp->icmp_type != ICMP_ECHO) &&
1316 (icmp->icmp_type != ICMP_TSTAMP) &&
1317 (icmp->icmp_type != ICMP_IREQ) &&
1318 (icmp->icmp_type != ICMP_MASKREQ))
1322 * perform a lookup of the ICMP packet in the state table
1324 hv = (pr = oip->ip_p);
1325 src.in4 = oip->ip_src;
1326 hv += src.in4.s_addr;
1327 dst.in4 = oip->ip_dst;
1328 hv += dst.in4.s_addr;
1329 hv += icmp->icmp_id;
1330 hv += icmp->icmp_seq;
1333 READ_ENTER(&ipf_state);
1334 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext)
1335 if ((is->is_p == pr) && (is->is_v == 4) &&
1336 (is->is_icmppkts < is->is_pkts) &&
1337 fr_matchsrcdst(is, src, dst, &ofin, NULL) &&
1338 fr_matchicmpqueryreply(is->is_v, is, icmp,
1340 ips_stats.iss_hits++;
1342 is->is_bytes += ip->ip_len;
1346 RWLOCK_EXIT(&ipf_state);
1351 if (fin->fin_plen < ICMPERR_MAXPKTLEN)
1358 tcp = (tcphdr_t *)((char *)oip + ohlen);
1359 dport = tcp->th_dport;
1360 sport = tcp->th_sport;
1362 hv = (pr = oip->ip_p);
1363 src.in4 = oip->ip_src;
1364 hv += src.in4.s_addr;
1365 dst.in4 = oip->ip_dst;
1366 hv += dst.in4.s_addr;
1371 READ_ENTER(&ipf_state);
1372 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) {
1374 * Only allow this icmp though if the
1375 * encapsulated packet was allowed through the
1376 * other way around. Note that the minimal amount
1377 * of info present does not allow for checking against
1378 * tcp internals such as seq and ack numbers. Only the
1379 * ports are known to be present and can be even if the
1380 * short flag is set.
1382 if ((is->is_p == pr) && (is->is_v == 4) &&
1383 (is->is_icmppkts < is->is_pkts) &&
1384 fr_matchsrcdst(is, src, dst, &ofin, tcp)) {
1386 ips_stats.iss_hits++;
1388 is->is_bytes += fin->fin_plen;
1390 * we deliberately do not touch the timeouts
1391 * for the accompanying state table entry.
1392 * It remains to be seen if that is correct. XXX
1397 RWLOCK_EXIT(&ipf_state);
1403 * Move a state hash table entry from its old location at is->is_hv to
1404 * its new location, indexed by hv % fr_statesize.
1406 static void fr_ipsmove(isp, is, hv)
1407 ipstate_t **isp, *is;
1414 * Remove the hash from the old location...
1417 is->is_hnext->is_phnext = isp;
1418 *isp = is->is_hnext;
1419 if (ips_table[hvm] == NULL)
1420 ips_stats.iss_inuse--;
1423 * ...and put the hash in the new one.
1425 hvm = hv % fr_statesize;
1427 isp = &ips_table[hvm];
1429 (*isp)->is_phnext = &is->is_hnext;
1431 ips_stats.iss_inuse++;
1432 is->is_phnext = isp;
1433 is->is_hnext = *isp;
1439 * Check if a packet has a registered state.
1441 frentry_t *fr_checkstate(ip, fin)
1445 union i6addr dst, src;
1446 ipstate_t *is, **isp;
1448 u_int hv, hvm, hlen, tryagain, pass, v;
1454 if ((ips_list == NULL) || (fin->fin_off != 0) || fr_state_lock ||
1455 (fin->fin_fl & FI_SHORT))
1459 hlen = fin->fin_hlen;
1460 tcp = (tcphdr_t *)((char *)ip + hlen);
1461 ic = (struct icmp *)tcp;
1462 hv = (pr = fin->fin_fi.fi_p);
1463 src = fin->fin_fi.fi_src;
1464 dst = fin->fin_fi.fi_dst;
1465 hv += src.in4.s_addr;
1466 hv += dst.in4.s_addr;
1469 * Search the hash table for matching packet header info.
1470 * At the bottom of this switch statement, the following is expected:
1471 * is == NULL, no lock on ipf_state is held.
1472 * is != NULL, a lock on ipf_state is held.
1474 v = fin->fin_fi.fi_v;
1477 hv += fin->fin_fi.fi_src.i6[1];
1478 hv += fin->fin_fi.fi_src.i6[2];
1479 hv += fin->fin_fi.fi_src.i6[3];
1481 if ((fin->fin_p == IPPROTO_ICMPV6) &&
1482 IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_dst.in6)) {
1483 hv -= dst.in4.s_addr;
1485 hv += fin->fin_fi.fi_dst.i6[1];
1486 hv += fin->fin_fi.fi_dst.i6[2];
1487 hv += fin->fin_fi.fi_dst.i6[3];
1495 case IPPROTO_ICMPV6 :
1499 if ((ic->icmp_type == ICMP6_ECHO_REQUEST) ||
1500 (ic->icmp_type == ICMP6_ECHO_REPLY)) {
1505 READ_ENTER(&ipf_state);
1507 hvm = hv % fr_statesize;
1508 for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext)
1509 if ((is->is_p == pr) && (is->is_v == v) &&
1510 fr_matchsrcdst(is, src, dst, fin, NULL) &&
1511 fr_matchicmpqueryreply(v, is, ic, fin->fin_rev)) {
1513 if (is->is_frage[rev] != 0)
1514 is->is_age = is->is_frage[rev];
1516 is->is_age = fr_icmpacktimeout;
1518 is->is_age = fr_icmptimeout;
1523 if (tryagain && !(is->is_flags & FI_W_DADDR)) {
1524 hv += fin->fin_fi.fi_src.i6[0];
1525 hv += fin->fin_fi.fi_src.i6[1];
1526 hv += fin->fin_fi.fi_src.i6[2];
1527 hv += fin->fin_fi.fi_src.i6[3];
1528 fr_ipsmove(isp, is, hv);
1529 MUTEX_DOWNGRADE(&ipf_state);
1533 RWLOCK_EXIT(&ipf_state);
1536 * No matching icmp state entry. Perhaps this is a
1537 * response to another state entry.
1539 if ((ips_wild != 0) && (v == 6) && (tryagain == 0) &&
1540 !IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_src.in6)) {
1541 hv -= fin->fin_fi.fi_src.i6[0];
1542 hv -= fin->fin_fi.fi_src.i6[1];
1543 hv -= fin->fin_fi.fi_src.i6[2];
1544 hv -= fin->fin_fi.fi_src.i6[3];
1546 WRITE_ENTER(&ipf_state);
1550 fr = fr_checkicmp6matchingstate((ip6_t *)ip, fin);
1561 hvm = hv % fr_statesize;
1562 READ_ENTER(&ipf_state);
1563 for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext)
1564 if ((is->is_p == pr) && (is->is_v == v) &&
1565 fr_matchsrcdst(is, src, dst, fin, NULL) &&
1566 fr_matchicmpqueryreply(v, is, ic, fin->fin_rev)) {
1568 if (is->is_frage[rev] != 0)
1569 is->is_age = is->is_frage[rev];
1570 else if (fin->fin_rev)
1571 is->is_age = fr_icmpacktimeout;
1573 is->is_age = fr_icmptimeout;
1579 RWLOCK_EXIT(&ipf_state);
1581 * No matching icmp state entry. Perhaps this is a
1582 * response to another state entry.
1584 fr = fr_checkicmpmatchingstate(ip, fin);
1590 * Just plain ignore RST flag set with either FIN or SYN.
1592 if ((tcp->th_flags & TH_RST) &&
1593 ((tcp->th_flags & (TH_FIN|TH_SYN|TH_RST)) != TH_RST))
1597 u_short dport, sport;
1599 dport = tcp->th_dport;
1600 sport = tcp->th_sport;
1604 READ_ENTER(&ipf_state);
1606 hvm = hv % fr_statesize;
1607 for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext)
1608 if ((is->is_p == pr) && (is->is_v == v) &&
1609 fr_matchsrcdst(is, src, dst, fin, tcp)) {
1611 if ((pr == IPPROTO_TCP)) {
1612 if (!fr_tcpstate(is, fin, ip, tcp))
1614 } else if ((pr == IPPROTO_UDP)) {
1615 if (is->is_frage[rev] != 0)
1616 is->is_age = is->is_frage[rev];
1617 else if (fin->fin_rev)
1618 is->is_age = fr_udpacktimeout;
1620 is->is_age = fr_udptimeout;
1626 !(is->is_flags & (FI_WILDP|FI_WILDA))) {
1629 fr_ipsmove(isp, is, hv);
1630 MUTEX_DOWNGRADE(&ipf_state);
1635 RWLOCK_EXIT(&ipf_state);
1636 if (!tryagain && ips_wild) {
1640 WRITE_ENTER(&ipf_state);
1648 READ_ENTER(&ipf_state);
1649 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) {
1650 if ((is->is_p == pr) && (is->is_v == v) &&
1651 fr_matchsrcdst(is, src, dst, fin, NULL)) {
1653 if (is->is_frage[rev] != 0)
1654 is->is_age = is->is_frage[rev];
1656 is->is_age = fr_udptimeout;
1661 RWLOCK_EXIT(&ipf_state);
1667 ATOMIC_INCL(ips_stats.iss_miss);
1671 MUTEX_ENTER(&is->is_lock);
1672 is->is_bytes += fin->fin_plen;
1673 ips_stats.iss_hits++;
1675 MUTEX_EXIT(&is->is_lock);
1677 fin->fin_rule = is->is_rulen;
1679 fin->fin_group = fr->fr_group;
1680 fin->fin_icode = fr->fr_icode;
1684 RWLOCK_EXIT(&ipf_state);
1685 if ((fin->fin_fl & FI_FRAG) && (pass & FR_KEEPFRAG))
1686 ipfr_newfrag(ip, fin);
1688 if ((tcp != NULL) && (tcp->th_flags & TCP_CLOSE))
1696 * Sync. state entries. If interfaces come or go or just change position,
1699 void ip_statesync(ifp)
1705 WRITE_ENTER(&ipf_state);
1706 for (is = ips_list; is; is = is->is_next) {
1707 for (i = 0; i < 4; i++) {
1708 if (is->is_ifp[i] == ifp) {
1709 is->is_ifp[i] = GETUNIT(is->is_ifname[i],
1712 is->is_ifp[i] = (void *)-1;
1716 RWLOCK_EXIT(&ipf_state);
1721 * Must always be called with fr_ipfstate held as a write lock.
1723 static void fr_delstate(is)
1728 if (is->is_flags & (FI_WILDP|FI_WILDA))
1731 is->is_next->is_pnext = is->is_pnext;
1732 *is->is_pnext = is->is_next;
1734 is->is_hnext->is_phnext = is->is_phnext;
1735 *is->is_phnext = is->is_hnext;
1736 if (ips_table[is->is_hv] == NULL)
1737 ips_stats.iss_inuse--;
1744 if (fr->fr_ref == 0) {
1749 MUTEX_DESTROY(&is->is_lock);
1757 * Free memory in use by all state info. kept.
1759 void fr_stateunload()
1763 WRITE_ENTER(&ipf_state);
1764 while ((is = ips_list))
1766 ips_stats.iss_inuse = 0;
1768 RWLOCK_EXIT(&ipf_state);
1770 KFREES(ips_table, fr_statesize * sizeof(ipstate_t *));
1776 * Slowly expire held state for thingslike UDP and ICMP. Timeouts are set
1777 * in expectation of this being called twice per second.
1779 void fr_timeoutstate()
1781 ipstate_t *is, **isp;
1782 #if defined(_KERNEL) && !SOLARIS
1787 WRITE_ENTER(&ipf_state);
1788 for (isp = &ips_list; (is = *isp); )
1789 if (is->is_age && !--is->is_age) {
1790 if (is->is_p == IPPROTO_TCP)
1791 ips_stats.iss_fin++;
1793 ips_stats.iss_expire++;
1795 ipstate_log(is, ISL_EXPIRE);
1800 if (fr_state_doflush) {
1801 (void) fr_state_flush(2, 0);
1802 fr_state_doflush = 0;
1804 RWLOCK_EXIT(&ipf_state);
1810 * Original idea freom Pradeep Krishnan for use primarily with NAT code.
1811 * (pkrishna@netcom.com)
1813 * Rewritten by Arjan de Vet <Arjan.deVet@adv.iae.nl>, 2000-07-29:
1815 * - (try to) base state transitions on real evidence only,
1816 * i.e. packets that are sent and have been received by ipfilter;
1817 * diagram 18.12 of TCP/IP volume 1 by W. Richard Stevens was used.
1819 * - deal with half-closed connections correctly;
1821 * - store the state of the source in state[0] such that ipfstat
1822 * displays the state as source/dest instead of dest/source; the calls
1823 * to fr_tcp_age have been changed accordingly.
1827 * state[0] = state of source (host that initiated connection)
1828 * state[1] = state of dest (host that accepted the connection)
1830 * dir == 0 : a packet from source to dest
1831 * dir == 1 : a packet from dest to source
1834 int fr_tcp_age(age, state, fin, dir, fsm)
1840 tcphdr_t *tcp = (tcphdr_t *)fin->fin_dp;
1841 u_char flags = tcp->th_flags;
1845 ostate = state[1 - dir];
1847 dlen = fin->fin_plen - fin->fin_hlen - (tcp->th_off << 2);
1849 if (flags & TH_RST) {
1850 if (!(tcp->th_flags & TH_PUSH) && !dlen) {
1851 *age = fr_tcpclosed;
1852 state[dir] = TCPS_CLOSED;
1854 *age = fr_tcpclosewait;
1855 state[dir] = TCPS_CLOSE_WAIT;
1864 case TCPS_CLOSED: /* 0 */
1865 if ((flags & TH_OPENING) == TH_OPENING) {
1867 * 'dir' received an S and sends SA in response,
1868 * CLOSED -> SYN_RECEIVED
1870 state[dir] = TCPS_SYN_RECEIVED;
1871 newage = fr_tcptimeout;
1872 } else if ((flags & TH_OPENING) == TH_SYN) {
1873 /* 'dir' sent S, CLOSED -> SYN_SENT */
1874 state[dir] = TCPS_SYN_SENT;
1875 newage = fr_tcptimeout;
1879 * It is apparently possible that a hosts sends two syncs
1880 * before the remote party is able to respond with a SA. In
1881 * such a case the remote server sometimes ACK's the second
1882 * sync, and then responds with a SA. The following code
1883 * is used to prevent this ack from being blocked.
1885 * We do not reset the timeout here to fr_tcptimeout because
1886 * a connection connect timeout does not renew after every
1887 * packet that is sent. We need to set newage to something
1888 * to indicate the packet has passed the check for its flags
1889 * being valid in the TCP FSM.
1891 else if ((ostate == TCPS_SYN_SENT) &&
1892 ((flags & (TH_FIN|TH_SYN|TH_RST|TH_ACK)) == TH_ACK)) {
1897 * The next piece of code makes it possible to get
1898 * already established connections into the state table
1899 * after a restart or reload of the filter rules; this
1900 * does not work when a strict 'flags S keep state' is
1901 * used for tcp connections of course, however, use a
1902 * lower time-out so the state disappears quickly if
1903 * the other side does not pick it up.
1906 (flags & (TH_FIN|TH_SYN|TH_RST|TH_ACK)) == TH_ACK) {
1907 /* we saw an A, guess 'dir' is in ESTABLISHED mode */
1908 if (ostate == TCPS_CLOSED) {
1909 state[dir] = TCPS_ESTABLISHED;
1910 newage = fr_tcptimeout;
1911 } else if (ostate == TCPS_ESTABLISHED) {
1912 state[dir] = TCPS_ESTABLISHED;
1913 newage = fr_tcpidletimeout;
1917 * TODO: besides regular ACK packets we can have other
1918 * packets as well; it is yet to be determined how we
1919 * should initialize the states in those cases
1923 case TCPS_LISTEN: /* 1 */
1927 case TCPS_SYN_SENT: /* 2 */
1928 if ((flags & ~(TH_ECN|TH_CWR)) == TH_SYN) {
1930 * A retransmitted SYN packet. We do not reset the
1931 * timeout here to fr_tcptimeout because a connection
1932 * connect timeout does not renew after every packet
1933 * that is sent. We need to set newage to something
1934 * to indicate the packet has passed the check for its
1935 * flags being valid in the TCP FSM.
1938 } else if ((flags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) {
1940 * We see an A from 'dir' which is in SYN_SENT
1941 * state: 'dir' sent an A in response to an SA
1942 * which it received, SYN_SENT -> ESTABLISHED
1944 state[dir] = TCPS_ESTABLISHED;
1945 newage = fr_tcpidletimeout;
1946 } else if (flags & TH_FIN) {
1948 * We see an F from 'dir' which is in SYN_SENT
1949 * state and wants to close its side of the
1950 * connection; SYN_SENT -> FIN_WAIT_1
1952 state[dir] = TCPS_FIN_WAIT_1;
1953 newage = fr_tcpidletimeout; /* or fr_tcptimeout? */
1954 } else if ((flags & TH_OPENING) == TH_OPENING) {
1956 * We see an SA from 'dir' which is already in
1957 * SYN_SENT state, this means we have a
1958 * simultaneous open; SYN_SENT -> SYN_RECEIVED
1960 state[dir] = TCPS_SYN_RECEIVED;
1961 newage = fr_tcptimeout;
1965 case TCPS_SYN_RECEIVED: /* 3 */
1966 if ((flags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) {
1968 * We see an A from 'dir' which was in SYN_RECEIVED
1969 * state so it must now be in established state,
1970 * SYN_RECEIVED -> ESTABLISHED
1972 state[dir] = TCPS_ESTABLISHED;
1973 newage = fr_tcpidletimeout;
1974 } else if ((flags & ~(TH_ECN|TH_CWR)) == TH_OPENING) {
1976 * We see an SA from 'dir' which is already in
1977 * SYN_RECEIVED state.
1979 newage = fr_tcptimeout;
1980 } else if (flags & TH_FIN) {
1982 * We see an F from 'dir' which is in SYN_RECEIVED
1983 * state and wants to close its side of the connection;
1984 * SYN_RECEIVED -> FIN_WAIT_1
1986 state[dir] = TCPS_FIN_WAIT_1;
1987 newage = fr_tcpidletimeout;
1991 case TCPS_ESTABLISHED: /* 4 */
1992 if (flags & TH_FIN) {
1994 * 'dir' closed its side of the connection; this
1995 * gives us a half-closed connection;
1996 * ESTABLISHED -> FIN_WAIT_1
1998 state[dir] = TCPS_FIN_WAIT_1;
1999 newage = fr_tcphalfclosed;
2000 } else if (flags & TH_ACK) {
2001 /* an ACK, should we exclude other flags here? */
2002 if (ostate == TCPS_FIN_WAIT_1) {
2004 * We know the other side did an active close,
2005 * so we are ACKing the recvd FIN packet (does
2006 * the window matching code guarantee this?)
2007 * and go into CLOSE_WAIT state; this gives us
2008 * a half-closed connection
2010 state[dir] = TCPS_CLOSE_WAIT;
2011 newage = fr_tcphalfclosed;
2012 } else if (ostate < TCPS_CLOSE_WAIT)
2014 * Still a fully established connection,
2017 newage = fr_tcpidletimeout;
2021 case TCPS_CLOSE_WAIT: /* 5 */
2022 if (flags & TH_FIN) {
2024 * Application closed and 'dir' sent a FIN, we're now
2025 * going into LAST_ACK state
2027 newage = fr_tcplastack;
2028 state[dir] = TCPS_LAST_ACK;
2031 * We remain in CLOSE_WAIT because the other side has
2032 * closed already and we did not close our side yet;
2035 newage = fr_tcphalfclosed;
2039 case TCPS_FIN_WAIT_1: /* 6 */
2040 if ((flags & TH_ACK) && ostate > TCPS_CLOSE_WAIT) {
2042 * If the other side is not active anymore it has sent
2043 * us a FIN packet that we are ack'ing now with an ACK;
2044 * this means both sides have now closed the connection
2045 * and we go into TIME_WAIT
2048 * XXX: how do we know we really are ACKing the FIN
2049 * packet here? does the window code guarantee that?
2051 state[dir] = TCPS_TIME_WAIT;
2052 newage = fr_tcptimeout;
2055 * We closed our side of the connection already but the
2056 * other side is still active (ESTABLISHED/CLOSE_WAIT);
2057 * continue with this half-closed connection
2059 newage = fr_tcphalfclosed;
2062 case TCPS_CLOSING: /* 7 */
2066 case TCPS_LAST_ACK: /* 8 */
2067 if (flags & TH_ACK) {
2068 if ((flags & TH_PUSH) || dlen)
2070 * There is still data to be delivered, reset
2073 newage = fr_tcplastack;
2078 * We cannot detect when we go out of LAST_ACK state to CLOSED
2079 * because that is based on the reception of ACK packets;
2080 * ipfilter can only detect that a packet has been sent by a
2085 case TCPS_FIN_WAIT_2: /* 9 */
2089 case TCPS_TIME_WAIT: /* 10 */
2090 newage = fr_tcptimeout; /* default 4 mins */
2091 /* we're in 2MSL timeout now */
2104 void ipstate_log(is, type)
2113 ipsl.isl_type = type;
2114 ipsl.isl_pkts = is->is_pkts + is->is_icmppkts;
2115 ipsl.isl_bytes = is->is_bytes;
2116 ipsl.isl_src = is->is_src;
2117 ipsl.isl_dst = is->is_dst;
2118 ipsl.isl_p = is->is_p;
2119 ipsl.isl_v = is->is_v;
2120 ipsl.isl_flags = is->is_flags;
2121 ipsl.isl_rulen = is->is_rulen;
2122 ipsl.isl_group = is->is_group;
2123 if (ipsl.isl_p == IPPROTO_TCP || ipsl.isl_p == IPPROTO_UDP) {
2124 ipsl.isl_sport = is->is_sport;
2125 ipsl.isl_dport = is->is_dport;
2126 if (ipsl.isl_p == IPPROTO_TCP) {
2127 ipsl.isl_state[0] = is->is_state[0];
2128 ipsl.isl_state[1] = is->is_state[1];
2130 } else if (ipsl.isl_p == IPPROTO_ICMP) {
2131 ipsl.isl_itype = is->is_icmp.ics_type;
2132 } else if (ipsl.isl_p == IPPROTO_ICMPV6) {
2133 ipsl.isl_itype = is->is_icmp.ics_type;
2135 ipsl.isl_ps.isl_filler[0] = 0;
2136 ipsl.isl_ps.isl_filler[1] = 0;
2139 sizes[0] = sizeof(ipsl);
2142 if (ipllog(IPL_LOGSTATE, NULL, items, sizes, types, 1)) {
2143 ATOMIC_INCL(ips_stats.iss_logged);
2145 ATOMIC_INCL(ips_stats.iss_logfail);
2152 frentry_t *fr_checkicmp6matchingstate(ip, fin)
2156 ipstate_t *is, **isp;
2157 u_short sport, dport;
2159 struct icmp6_hdr *ic, *oic;
2160 union i6addr dst, src;
2170 * Does it at least have the return (basic) IP header ?
2171 * Only a basic IP header (no options) should be with
2172 * an ICMP error header.
2174 if ((fin->fin_v != 6) || (fin->fin_plen < ICMP6ERR_MINPKTLEN))
2176 ic = (struct icmp6_hdr *)fin->fin_dp;
2177 type = ic->icmp6_type;
2179 * If it's not an error type, then return
2181 if ((type != ICMP6_DST_UNREACH) && (type != ICMP6_PACKET_TOO_BIG) &&
2182 (type != ICMP6_TIME_EXCEEDED) && (type != ICMP6_PARAM_PROB))
2185 oip = (ip6_t *)((char *)ic + ICMPERR_ICMPHLEN);
2186 if (fin->fin_plen < sizeof(*oip))
2189 if ((oip->ip6_nxt != IPPROTO_TCP) && (oip->ip6_nxt != IPPROTO_UDP) &&
2190 (oip->ip6_nxt != IPPROTO_ICMPV6))
2193 bzero((char *)&ofin, sizeof(ofin));
2194 ofin.fin_out = !fin->fin_out;
2195 ofin.fin_ifp = fin->fin_ifp;
2199 ofin.fin_qfm = NULL;
2202 * We make a fin entry to be able to feed it to
2203 * matchsrcdst. Note that not all fields are necessary
2204 * but this is the cleanest way. Note further we fill
2205 * in fin_mp such that if someone uses it we'll get
2206 * a kernel panic. fr_matchsrcdst does not use this.
2208 * watch out here, as ip is in host order and oip in network
2209 * order. Any change we make must be undone afterwards.
2211 savelen = oip->ip6_plen;
2212 oip->ip6_plen = ip->ip6_plen - sizeof(*ip) - ICMPERR_ICMPHLEN;
2213 fr_makefrip(sizeof(*oip), (ip_t *)oip, &ofin);
2214 oip->ip6_plen = savelen;
2216 if (oip->ip6_nxt == IPPROTO_ICMPV6) {
2217 oic = (struct icmp6_hdr *)(oip + 1);
2219 * an ICMP error can only be generated as a result of an
2220 * ICMP query, not as the response on an ICMP error
2222 * XXX theoretically ICMP_ECHOREP and the other reply's are
2223 * ICMP query's as well, but adding them here seems strange XXX
2225 if (!(oic->icmp6_type & ICMP6_INFOMSG_MASK))
2229 * perform a lookup of the ICMP packet in the state table
2231 hv = (pr = oip->ip6_nxt);
2232 src.in6 = oip->ip6_src;
2233 hv += src.in4.s_addr;
2234 dst.in6 = oip->ip6_dst;
2235 hv += dst.in4.s_addr;
2236 hv += oic->icmp6_id;
2237 hv += oic->icmp6_seq;
2240 READ_ENTER(&ipf_state);
2241 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext)
2242 if ((is->is_p == pr) &&
2243 (oic->icmp6_id == is->is_icmp.ics_id) &&
2244 (oic->icmp6_seq == is->is_icmp.ics_seq) &&
2245 fr_matchsrcdst(is, src, dst, &ofin, NULL)) {
2247 * in the state table ICMP query's are stored
2248 * with the type of the corresponding ICMP
2249 * response. Correct here
2251 if (((is->is_type == ICMP6_ECHO_REPLY) &&
2252 (oic->icmp6_type == ICMP6_ECHO_REQUEST)) ||
2253 (is->is_type - 1 == oic->icmp6_type )) {
2254 ips_stats.iss_hits++;
2256 is->is_bytes += fin->fin_plen;
2260 RWLOCK_EXIT(&ipf_state);
2265 tcp = (tcphdr_t *)(oip + 1);
2266 dport = tcp->th_dport;
2267 sport = tcp->th_sport;
2269 hv = (pr = oip->ip6_nxt);
2270 src.in6 = oip->ip6_src;
2271 hv += src.in4.s_addr;
2275 dst.in6 = oip->ip6_dst;
2276 hv += dst.in4.s_addr;
2284 READ_ENTER(&ipf_state);
2285 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) {
2287 * Only allow this icmp though if the
2288 * encapsulated packet was allowed through the
2289 * other way around. Note that the minimal amount
2290 * of info present does not allow for checking against
2291 * tcp internals such as seq and ack numbers.
2293 if ((is->is_p == pr) && (is->is_v == 6) &&
2294 fr_matchsrcdst(is, src, dst, &ofin, tcp)) {
2296 ips_stats.iss_hits++;
2298 is->is_bytes += fin->fin_plen;
2300 * we deliberately do not touch the timeouts
2301 * for the accompanying state table entry.
2302 * It remains to be seen if that is correct. XXX
2304 RWLOCK_EXIT(&ipf_state);
2308 RWLOCK_EXIT(&ipf_state);