2 * Copyright (c) 2002 Luigi Rizzo, Universita` di Pisa
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
7 * 1. Redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution.
13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * Implement IP packet firewall (new version)
35 #if !defined(KLD_MODULE)
40 #error IPFIREWALL requires INET.
43 #include "opt_inet6.h"
44 #include "opt_ipsec.h"
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/condvar.h>
50 #include <sys/malloc.h>
52 #include <sys/kernel.h>
56 #include <sys/module.h>
58 #include <sys/rwlock.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/sysctl.h>
62 #include <sys/syslog.h>
63 #include <sys/ucred.h>
65 #include <net/radix.h>
66 #include <net/route.h>
67 #include <netinet/in.h>
68 #include <netinet/in_systm.h>
69 #include <netinet/in_var.h>
70 #include <netinet/in_pcb.h>
71 #include <netinet/ip.h>
72 #include <netinet/ip_var.h>
73 #include <netinet/ip_icmp.h>
74 #include <netinet/ip_fw.h>
75 #include <netinet/ip_divert.h>
76 #include <netinet/ip_dummynet.h>
77 #include <netinet/pim.h>
78 #include <netinet/tcp.h>
79 #include <netinet/tcp_timer.h>
80 #include <netinet/tcp_var.h>
81 #include <netinet/tcpip.h>
82 #include <netinet/udp.h>
83 #include <netinet/udp_var.h>
85 #include <netgraph/ng_ipfw.h>
87 #include <altq/if_altq.h>
90 #include <netinet6/ipsec.h>
93 #include <netinet/ip6.h>
94 #include <netinet/icmp6.h>
96 #include <netinet6/scope6_var.h>
99 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
101 #include <machine/in_cksum.h> /* XXX for in_cksum */
104 * set_disable contains one bit per set value (0..31).
105 * If the bit is set, all rules with the corresponding set
106 * are disabled. Set RESVD_SET(31) is reserved for the default rule
107 * and rules that are not deleted by the flush command,
108 * and CANNOT be disabled.
109 * Rules in set RESVD_SET can only be deleted explicitly.
111 static u_int32_t set_disable;
113 static int fw_verbose;
114 static int verbose_limit;
116 static struct callout ipfw_timeout;
117 static uma_zone_t ipfw_dyn_rule_zone;
118 #define IPFW_DEFAULT_RULE 65535
121 * Data structure to cache our ucred related
122 * information. This structure only gets used if
123 * the user specified UID/GID based constraints in
127 gid_t fw_groups[NGROUPS];
133 #define IPFW_TABLES_MAX 128
135 struct ip_fw *rules; /* list of rules */
136 struct ip_fw *reap; /* list of rules to reap */
137 struct radix_node_head *tables[IPFW_TABLES_MAX];
140 #define IPFW_LOCK_INIT(_chain) \
141 rw_init(&(_chain)->rwmtx, "IPFW static rules")
142 #define IPFW_LOCK_DESTROY(_chain) rw_destroy(&(_chain)->rwmtx)
143 #define IPFW_WLOCK_ASSERT(_chain) do { \
144 rw_assert(&(_chain)->rwmtx, RA_WLOCKED); \
145 NET_ASSERT_GIANT(); \
148 #define IPFW_RLOCK(p) rw_rlock(&(p)->rwmtx)
149 #define IPFW_RUNLOCK(p) rw_runlock(&(p)->rwmtx)
150 #define IPFW_WLOCK(p) rw_wlock(&(p)->rwmtx)
151 #define IPFW_WUNLOCK(p) rw_wunlock(&(p)->rwmtx)
154 * list of rules for layer 3
156 static struct ip_fw_chain layer3_chain;
158 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
159 MALLOC_DEFINE(M_IPFW_TBL, "ipfw_tbl", "IpFw tables");
162 struct radix_node rn[2];
163 struct sockaddr_in addr, mask;
167 static int fw_debug = 1;
168 static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
170 extern int ipfw_chg_hook(SYSCTL_HANDLER_ARGS);
173 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
174 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, enable,
175 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE3, &fw_enable, 0,
176 ipfw_chg_hook, "I", "Enable ipfw");
177 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
178 &autoinc_step, 0, "Rule number autincrement step");
179 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, one_pass,
180 CTLFLAG_RW | CTLFLAG_SECURE3,
182 "Only do a single pass through ipfw when using dummynet(4)");
183 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
184 &fw_debug, 0, "Enable printing of debug ip_fw statements");
185 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose,
186 CTLFLAG_RW | CTLFLAG_SECURE3,
187 &fw_verbose, 0, "Log matches to ipfw rules");
188 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
189 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
192 * Description of dynamic rules.
194 * Dynamic rules are stored in lists accessed through a hash table
195 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
196 * be modified through the sysctl variable dyn_buckets which is
197 * updated when the table becomes empty.
199 * XXX currently there is only one list, ipfw_dyn.
201 * When a packet is received, its address fields are first masked
202 * with the mask defined for the rule, then hashed, then matched
203 * against the entries in the corresponding list.
204 * Dynamic rules can be used for different purposes:
206 * + enforcing limits on the number of sessions;
207 * + in-kernel NAT (not implemented yet)
209 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
210 * measured in seconds and depending on the flags.
212 * The total number of dynamic rules is stored in dyn_count.
213 * The max number of dynamic rules is dyn_max. When we reach
214 * the maximum number of rules we do not create anymore. This is
215 * done to avoid consuming too much memory, but also too much
216 * time when searching on each packet (ideally, we should try instead
217 * to put a limit on the length of the list on each bucket...).
219 * Each dynamic rule holds a pointer to the parent ipfw rule so
220 * we know what action to perform. Dynamic rules are removed when
221 * the parent rule is deleted. XXX we should make them survive.
223 * There are some limitations with dynamic rules -- we do not
224 * obey the 'randomized match', and we do not do multiple
225 * passes through the firewall. XXX check the latter!!!
227 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
228 static u_int32_t dyn_buckets = 256; /* must be power of 2 */
229 static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */
231 static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */
232 #define IPFW_DYN_LOCK_INIT() \
233 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF)
234 #define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx)
235 #define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx)
236 #define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx)
237 #define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED)
240 * Timeouts for various events in handing dynamic rules.
242 static u_int32_t dyn_ack_lifetime = 300;
243 static u_int32_t dyn_syn_lifetime = 20;
244 static u_int32_t dyn_fin_lifetime = 1;
245 static u_int32_t dyn_rst_lifetime = 1;
246 static u_int32_t dyn_udp_lifetime = 10;
247 static u_int32_t dyn_short_lifetime = 5;
250 * Keepalives are sent if dyn_keepalive is set. They are sent every
251 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
252 * seconds of lifetime of a rule.
253 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
254 * than dyn_keepalive_period.
257 static u_int32_t dyn_keepalive_interval = 20;
258 static u_int32_t dyn_keepalive_period = 5;
259 static u_int32_t dyn_keepalive = 1; /* do send keepalives */
261 static u_int32_t static_count; /* # of static rules */
262 static u_int32_t static_len; /* size in bytes of static rules */
263 static u_int32_t dyn_count; /* # of dynamic rules */
264 static u_int32_t dyn_max = 4096; /* max # of dynamic rules */
266 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
267 &dyn_buckets, 0, "Number of dyn. buckets");
268 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
269 &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
270 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
271 &dyn_count, 0, "Number of dyn. rules");
272 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
273 &dyn_max, 0, "Max number of dyn. rules");
274 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
275 &static_count, 0, "Number of static rules");
276 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
277 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
278 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
279 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
280 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
281 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
282 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
283 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
284 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
285 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
286 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
287 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
288 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
289 &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
293 * IPv6 specific variables
295 SYSCTL_DECL(_net_inet6_ip6);
297 static struct sysctl_ctx_list ip6_fw_sysctl_ctx;
298 static struct sysctl_oid *ip6_fw_sysctl_tree;
300 #endif /* SYSCTL_NODE */
302 static int fw_deny_unknown_exthdrs = 1;
306 * L3HDR maps an ipv4 pointer into a layer3 header pointer of type T
307 * Other macros just cast void * into the appropriate type
309 #define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
310 #define TCP(p) ((struct tcphdr *)(p))
311 #define UDP(p) ((struct udphdr *)(p))
312 #define ICMP(p) ((struct icmphdr *)(p))
313 #define ICMP6(p) ((struct icmp6_hdr *)(p))
316 icmptype_match(struct icmphdr *icmp, ipfw_insn_u32 *cmd)
318 int type = icmp->icmp_type;
320 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
323 #define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
324 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
327 is_icmp_query(struct icmphdr *icmp)
329 int type = icmp->icmp_type;
331 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
336 * The following checks use two arrays of 8 or 16 bits to store the
337 * bits that we want set or clear, respectively. They are in the
338 * low and high half of cmd->arg1 or cmd->d[0].
340 * We scan options and store the bits we find set. We succeed if
342 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
344 * The code is sometimes optimized not to store additional variables.
348 flags_match(ipfw_insn *cmd, u_int8_t bits)
353 if ( ((cmd->arg1 & 0xff) & bits) != 0)
354 return 0; /* some bits we want set were clear */
355 want_clear = (cmd->arg1 >> 8) & 0xff;
356 if ( (want_clear & bits) != want_clear)
357 return 0; /* some bits we want clear were set */
362 ipopts_match(struct ip *ip, ipfw_insn *cmd)
364 int optlen, bits = 0;
365 u_char *cp = (u_char *)(ip + 1);
366 int x = (ip->ip_hl << 2) - sizeof (struct ip);
368 for (; x > 0; x -= optlen, cp += optlen) {
369 int opt = cp[IPOPT_OPTVAL];
371 if (opt == IPOPT_EOL)
373 if (opt == IPOPT_NOP)
376 optlen = cp[IPOPT_OLEN];
377 if (optlen <= 0 || optlen > x)
378 return 0; /* invalid or truncated */
386 bits |= IP_FW_IPOPT_LSRR;
390 bits |= IP_FW_IPOPT_SSRR;
394 bits |= IP_FW_IPOPT_RR;
398 bits |= IP_FW_IPOPT_TS;
402 return (flags_match(cmd, bits));
406 tcpopts_match(struct tcphdr *tcp, ipfw_insn *cmd)
408 int optlen, bits = 0;
409 u_char *cp = (u_char *)(tcp + 1);
410 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
412 for (; x > 0; x -= optlen, cp += optlen) {
414 if (opt == TCPOPT_EOL)
416 if (opt == TCPOPT_NOP)
430 bits |= IP_FW_TCPOPT_MSS;
434 bits |= IP_FW_TCPOPT_WINDOW;
437 case TCPOPT_SACK_PERMITTED:
439 bits |= IP_FW_TCPOPT_SACK;
442 case TCPOPT_TIMESTAMP:
443 bits |= IP_FW_TCPOPT_TS;
448 return (flags_match(cmd, bits));
452 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
454 if (ifp == NULL) /* no iface with this packet, match fails */
456 /* Check by name or by IP address */
457 if (cmd->name[0] != '\0') { /* match by name */
460 if (fnmatch(cmd->name, ifp->if_xname, 0) == 0)
463 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
470 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
471 if (ia->ifa_addr->sa_family != AF_INET)
473 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
474 (ia->ifa_addr))->sin_addr.s_addr)
475 return(1); /* match */
478 return(0); /* no match, fail ... */
482 * The verify_path function checks if a route to the src exists and
483 * if it is reachable via ifp (when provided).
485 * The 'verrevpath' option checks that the interface that an IP packet
486 * arrives on is the same interface that traffic destined for the
487 * packet's source address would be routed out of. The 'versrcreach'
488 * option just checks that the source address is reachable via any route
489 * (except default) in the routing table. These two are a measure to block
490 * forged packets. This is also commonly known as "anti-spoofing" or Unicast
491 * Reverse Path Forwarding (Unicast RFP) in Cisco-ese. The name of the knobs
492 * is purposely reminiscent of the Cisco IOS command,
494 * ip verify unicast reverse-path
495 * ip verify unicast source reachable-via any
497 * which implements the same functionality. But note that syntax is
498 * misleading. The check may be performed on all IP packets whether unicast,
499 * multicast, or broadcast.
502 verify_path(struct in_addr src, struct ifnet *ifp)
505 struct sockaddr_in *dst;
507 bzero(&ro, sizeof(ro));
509 dst = (struct sockaddr_in *)&(ro.ro_dst);
510 dst->sin_family = AF_INET;
511 dst->sin_len = sizeof(*dst);
513 rtalloc_ign(&ro, RTF_CLONING);
515 if (ro.ro_rt == NULL)
519 * If ifp is provided, check for equality with rtentry.
520 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
521 * in order to pass packets injected back by if_simloop():
522 * if useloopback == 1 routing entry (via lo0) for our own address
523 * may exist, so we need to handle routing assymetry.
525 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
530 /* if no ifp provided, check if rtentry is not default route */
532 satosin(rt_key(ro.ro_rt))->sin_addr.s_addr == INADDR_ANY) {
537 /* or if this is a blackhole/reject route */
538 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
543 /* found valid route */
550 * ipv6 specific rules here...
553 icmp6type_match (int type, ipfw_insn_u32 *cmd)
555 return (type <= ICMP6_MAXTYPE && (cmd->d[type/32] & (1<<(type%32)) ) );
559 flow6id_match( int curr_flow, ipfw_insn_u32 *cmd )
562 for (i=0; i <= cmd->o.arg1; ++i )
563 if (curr_flow == cmd->d[i] )
568 /* support for IP6_*_ME opcodes */
570 search_ip6_addr_net (struct in6_addr * ip6_addr)
574 struct in6_ifaddr *fdm;
575 struct in6_addr copia;
577 TAILQ_FOREACH(mdc, &ifnet, if_link)
578 TAILQ_FOREACH(mdc2, &mdc->if_addrlist, ifa_list) {
579 if (mdc2->ifa_addr->sa_family == AF_INET6) {
580 fdm = (struct in6_ifaddr *)mdc2;
581 copia = fdm->ia_addr.sin6_addr;
582 /* need for leaving scope_id in the sock_addr */
583 in6_clearscope(&copia);
584 if (IN6_ARE_ADDR_EQUAL(ip6_addr, &copia))
592 verify_path6(struct in6_addr *src, struct ifnet *ifp)
595 struct sockaddr_in6 *dst;
597 bzero(&ro, sizeof(ro));
599 dst = (struct sockaddr_in6 * )&(ro.ro_dst);
600 dst->sin6_family = AF_INET6;
601 dst->sin6_len = sizeof(*dst);
602 dst->sin6_addr = *src;
603 rtalloc_ign((struct route *)&ro, RTF_CLONING);
605 if (ro.ro_rt == NULL)
609 * if ifp is provided, check for equality with rtentry
610 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
611 * to support the case of sending packets to an address of our own.
612 * (where the former interface is the first argument of if_simloop()
613 * (=ifp), the latter is lo0)
615 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
620 /* if no ifp provided, check if rtentry is not default route */
622 IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(ro.ro_rt))->sin6_addr)) {
627 /* or if this is a blackhole/reject route */
628 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
633 /* found valid route */
639 hash_packet6(struct ipfw_flow_id *id)
642 i = (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^
643 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^
644 (id->src_ip6.__u6_addr.__u6_addr32[2]) ^
645 (id->src_ip6.__u6_addr.__u6_addr32[3]) ^
646 (id->dst_port) ^ (id->src_port);
651 is_icmp6_query(int icmp6_type)
653 if ((icmp6_type <= ICMP6_MAXTYPE) &&
654 (icmp6_type == ICMP6_ECHO_REQUEST ||
655 icmp6_type == ICMP6_MEMBERSHIP_QUERY ||
656 icmp6_type == ICMP6_WRUREQUEST ||
657 icmp6_type == ICMP6_FQDN_QUERY ||
658 icmp6_type == ICMP6_NI_QUERY))
665 send_reject6(struct ip_fw_args *args, int code, u_int hlen)
667 if (code == ICMP6_UNREACH_RST && args->f_id.proto == IPPROTO_TCP) {
677 if (args->m->m_len < (hlen+sizeof(struct tcphdr))) {
678 args->m = m_pullup(args->m, hlen+sizeof(struct tcphdr));
683 ip6 = mtod(args->m, struct ip6_hdr *);
684 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
686 if ((tcp->th_flags & TH_RST) != 0) {
693 ti.th.th_seq = ntohl(ti.th.th_seq);
694 ti.th.th_ack = ntohl(ti.th.th_ack);
695 ti.ip6.ip6_nxt = IPPROTO_TCP;
697 if (ti.th.th_flags & TH_ACK) {
703 if (((args->m)->m_flags & M_PKTHDR) != 0) {
704 ack += (args->m)->m_pkthdr.len - hlen
705 - (ti.th.th_off << 2);
706 } else if (ip6->ip6_plen) {
707 ack += ntohs(ip6->ip6_plen) + sizeof(*ip6)
708 - hlen - (ti.th.th_off << 2);
713 if (tcp->th_flags & TH_SYN)
716 flags = TH_RST|TH_ACK;
718 bcopy(&ti, ip6, sizeof(ti));
719 tcp_respond(NULL, ip6, (struct tcphdr *)(ip6 + 1),
720 args->m, ack, seq, flags);
722 } else if (code != ICMP6_UNREACH_RST) { /* Send an ICMPv6 unreach. */
723 icmp6_error(args->m, ICMP6_DST_UNREACH, code, 0);
733 static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */
735 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
736 #define SNP(buf) buf, sizeof(buf)
739 * We enter here when we have a rule with O_LOG.
740 * XXX this function alone takes about 2Kbytes of code!
743 ipfw_log(struct ip_fw *f, u_int hlen, struct ip_fw_args *args,
744 struct mbuf *m, struct ifnet *oif, u_short offset, uint32_t tablearg)
746 struct ether_header *eh = args->eh;
748 int limit_reached = 0;
749 char action2[40], proto[128], fragment[32];
754 if (f == NULL) { /* bogus pkt */
755 if (verbose_limit != 0 && norule_counter >= verbose_limit)
758 if (norule_counter == verbose_limit)
759 limit_reached = verbose_limit;
761 } else { /* O_LOG is the first action, find the real one */
762 ipfw_insn *cmd = ACTION_PTR(f);
763 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
765 if (l->max_log != 0 && l->log_left == 0)
768 if (l->log_left == 0)
769 limit_reached = l->max_log;
770 cmd += F_LEN(cmd); /* point to first action */
771 if (cmd->opcode == O_ALTQ) {
772 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
774 snprintf(SNPARGS(action2, 0), "Altq %d",
778 if (cmd->opcode == O_PROB)
781 if (cmd->opcode == O_TAG)
785 switch (cmd->opcode) {
791 if (cmd->arg1==ICMP_REJECT_RST)
793 else if (cmd->arg1==ICMP_UNREACH_HOST)
796 snprintf(SNPARGS(action2, 0), "Unreach %d",
801 if (cmd->arg1==ICMP6_UNREACH_RST)
804 snprintf(SNPARGS(action2, 0), "Unreach %d",
815 snprintf(SNPARGS(action2, 0), "Divert %d",
819 snprintf(SNPARGS(action2, 0), "Tee %d",
823 snprintf(SNPARGS(action2, 0), "SkipTo %d",
827 snprintf(SNPARGS(action2, 0), "Pipe %d",
831 snprintf(SNPARGS(action2, 0), "Queue %d",
835 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
837 struct in_addr dummyaddr;
838 if (sa->sa.sin_addr.s_addr == INADDR_ANY)
839 dummyaddr.s_addr = htonl(tablearg);
841 dummyaddr.s_addr = sa->sa.sin_addr.s_addr;
843 len = snprintf(SNPARGS(action2, 0), "Forward to %s",
844 inet_ntoa(dummyaddr));
847 snprintf(SNPARGS(action2, len), ":%d",
852 snprintf(SNPARGS(action2, 0), "Netgraph %d",
856 snprintf(SNPARGS(action2, 0), "Ngtee %d",
865 if (hlen == 0) { /* non-ip */
866 snprintf(SNPARGS(proto, 0), "MAC");
870 char src[48], dst[48];
871 struct icmphdr *icmp;
874 /* Initialize to make compiler happy. */
875 struct ip *ip = NULL;
877 struct ip6_hdr *ip6 = NULL;
878 struct icmp6_hdr *icmp6;
883 if (args->f_id.addr_type == 6) {
884 snprintf(src, sizeof(src), "[%s]",
885 ip6_sprintf(&args->f_id.src_ip6));
886 snprintf(dst, sizeof(dst), "[%s]",
887 ip6_sprintf(&args->f_id.dst_ip6));
889 ip6 = (struct ip6_hdr *)mtod(m, struct ip6_hdr *);
890 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
891 udp = (struct udphdr *)(mtod(args->m, char *) + hlen);
895 ip = mtod(m, struct ip *);
896 tcp = L3HDR(struct tcphdr, ip);
897 udp = L3HDR(struct udphdr, ip);
899 inet_ntoa_r(ip->ip_src, src);
900 inet_ntoa_r(ip->ip_dst, dst);
903 switch (args->f_id.proto) {
905 len = snprintf(SNPARGS(proto, 0), "TCP %s", src);
907 snprintf(SNPARGS(proto, len), ":%d %s:%d",
908 ntohs(tcp->th_sport),
910 ntohs(tcp->th_dport));
912 snprintf(SNPARGS(proto, len), " %s", dst);
916 len = snprintf(SNPARGS(proto, 0), "UDP %s", src);
918 snprintf(SNPARGS(proto, len), ":%d %s:%d",
919 ntohs(udp->uh_sport),
921 ntohs(udp->uh_dport));
923 snprintf(SNPARGS(proto, len), " %s", dst);
927 icmp = L3HDR(struct icmphdr, ip);
929 len = snprintf(SNPARGS(proto, 0),
931 icmp->icmp_type, icmp->icmp_code);
933 len = snprintf(SNPARGS(proto, 0), "ICMP ");
934 len += snprintf(SNPARGS(proto, len), "%s", src);
935 snprintf(SNPARGS(proto, len), " %s", dst);
939 icmp6 = (struct icmp6_hdr *)(mtod(args->m, char *) + hlen);
941 len = snprintf(SNPARGS(proto, 0),
943 icmp6->icmp6_type, icmp6->icmp6_code);
945 len = snprintf(SNPARGS(proto, 0), "ICMPv6 ");
946 len += snprintf(SNPARGS(proto, len), "%s", src);
947 snprintf(SNPARGS(proto, len), " %s", dst);
951 len = snprintf(SNPARGS(proto, 0), "P:%d %s",
952 args->f_id.proto, src);
953 snprintf(SNPARGS(proto, len), " %s", dst);
958 if (args->f_id.addr_type == 6) {
959 if (offset & (IP6F_OFF_MASK | IP6F_MORE_FRAG))
960 snprintf(SNPARGS(fragment, 0),
961 " (frag %08x:%d@%d%s)",
963 ntohs(ip6->ip6_plen) - hlen,
964 ntohs(offset & IP6F_OFF_MASK) << 3,
965 (offset & IP6F_MORE_FRAG) ? "+" : "");
970 if (eh != NULL) { /* layer 2 packets are as on the wire */
971 ip_off = ntohs(ip->ip_off);
972 ip_len = ntohs(ip->ip_len);
977 if (ip_off & (IP_MF | IP_OFFMASK))
978 snprintf(SNPARGS(fragment, 0),
979 " (frag %d:%d@%d%s)",
980 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
982 (ip_off & IP_MF) ? "+" : "");
985 if (oif || m->m_pkthdr.rcvif)
986 log(LOG_SECURITY | LOG_INFO,
987 "ipfw: %d %s %s %s via %s%s\n",
989 action, proto, oif ? "out" : "in",
990 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
993 log(LOG_SECURITY | LOG_INFO,
994 "ipfw: %d %s %s [no if info]%s\n",
996 action, proto, fragment);
998 log(LOG_SECURITY | LOG_NOTICE,
999 "ipfw: limit %d reached on entry %d\n",
1000 limit_reached, f ? f->rulenum : -1);
1004 * IMPORTANT: the hash function for dynamic rules must be commutative
1005 * in source and destination (ip,port), because rules are bidirectional
1006 * and we want to find both in the same bucket.
1009 hash_packet(struct ipfw_flow_id *id)
1014 if (IS_IP6_FLOW_ID(id))
1015 i = hash_packet6(id);
1018 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
1019 i &= (curr_dyn_buckets - 1);
1024 * unlink a dynamic rule from a chain. prev is a pointer to
1025 * the previous one, q is a pointer to the rule to delete,
1026 * head is a pointer to the head of the queue.
1027 * Modifies q and potentially also head.
1029 #define UNLINK_DYN_RULE(prev, head, q) { \
1030 ipfw_dyn_rule *old_q = q; \
1032 /* remove a refcount to the parent */ \
1033 if (q->dyn_type == O_LIMIT) \
1034 q->parent->count--; \
1035 DEB(printf("ipfw: unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",\
1036 (q->id.src_ip), (q->id.src_port), \
1037 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \
1039 prev->next = q = q->next; \
1041 head = q = q->next; \
1043 uma_zfree(ipfw_dyn_rule_zone, old_q); }
1045 #define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
1048 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
1050 * If keep_me == NULL, rules are deleted even if not expired,
1051 * otherwise only expired rules are removed.
1053 * The value of the second parameter is also used to point to identify
1054 * a rule we absolutely do not want to remove (e.g. because we are
1055 * holding a reference to it -- this is the case with O_LIMIT_PARENT
1056 * rules). The pointer is only used for comparison, so any non-null
1060 remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
1062 static u_int32_t last_remove = 0;
1064 #define FORCE (keep_me == NULL)
1066 ipfw_dyn_rule *prev, *q;
1067 int i, pass = 0, max_pass = 0;
1069 IPFW_DYN_LOCK_ASSERT();
1071 if (ipfw_dyn_v == NULL || dyn_count == 0)
1073 /* do not expire more than once per second, it is useless */
1074 if (!FORCE && last_remove == time_uptime)
1076 last_remove = time_uptime;
1079 * because O_LIMIT refer to parent rules, during the first pass only
1080 * remove child and mark any pending LIMIT_PARENT, and remove
1081 * them in a second pass.
1084 for (i = 0 ; i < curr_dyn_buckets ; i++) {
1085 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
1087 * Logic can become complex here, so we split tests.
1091 if (rule != NULL && rule != q->rule)
1092 goto next; /* not the one we are looking for */
1093 if (q->dyn_type == O_LIMIT_PARENT) {
1095 * handle parent in the second pass,
1096 * record we need one.
1101 if (FORCE && q->count != 0 ) {
1102 /* XXX should not happen! */
1103 printf("ipfw: OUCH! cannot remove rule,"
1104 " count %d\n", q->count);
1108 !TIME_LEQ( q->expire, time_uptime ))
1111 if (q->dyn_type != O_LIMIT_PARENT || !q->count) {
1112 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1120 if (pass++ < max_pass)
1126 * lookup a dynamic rule.
1128 static ipfw_dyn_rule *
1129 lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction,
1133 * stateful ipfw extensions.
1134 * Lookup into dynamic session queue
1136 #define MATCH_REVERSE 0
1137 #define MATCH_FORWARD 1
1138 #define MATCH_NONE 2
1139 #define MATCH_UNKNOWN 3
1140 int i, dir = MATCH_NONE;
1141 ipfw_dyn_rule *prev, *q=NULL;
1143 IPFW_DYN_LOCK_ASSERT();
1145 if (ipfw_dyn_v == NULL)
1146 goto done; /* not found */
1147 i = hash_packet( pkt );
1148 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
1149 if (q->dyn_type == O_LIMIT_PARENT && q->count)
1151 if (TIME_LEQ( q->expire, time_uptime)) { /* expire entry */
1152 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1155 if (pkt->proto == q->id.proto &&
1156 q->dyn_type != O_LIMIT_PARENT) {
1157 if (IS_IP6_FLOW_ID(pkt)) {
1158 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1159 &(q->id.src_ip6)) &&
1160 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1161 &(q->id.dst_ip6)) &&
1162 pkt->src_port == q->id.src_port &&
1163 pkt->dst_port == q->id.dst_port ) {
1164 dir = MATCH_FORWARD;
1167 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1168 &(q->id.dst_ip6)) &&
1169 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1170 &(q->id.src_ip6)) &&
1171 pkt->src_port == q->id.dst_port &&
1172 pkt->dst_port == q->id.src_port ) {
1173 dir = MATCH_REVERSE;
1177 if (pkt->src_ip == q->id.src_ip &&
1178 pkt->dst_ip == q->id.dst_ip &&
1179 pkt->src_port == q->id.src_port &&
1180 pkt->dst_port == q->id.dst_port ) {
1181 dir = MATCH_FORWARD;
1184 if (pkt->src_ip == q->id.dst_ip &&
1185 pkt->dst_ip == q->id.src_ip &&
1186 pkt->src_port == q->id.dst_port &&
1187 pkt->dst_port == q->id.src_port ) {
1188 dir = MATCH_REVERSE;
1198 goto done; /* q = NULL, not found */
1200 if ( prev != NULL) { /* found and not in front */
1201 prev->next = q->next;
1202 q->next = ipfw_dyn_v[i];
1205 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
1206 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
1208 #define BOTH_SYN (TH_SYN | (TH_SYN << 8))
1209 #define BOTH_FIN (TH_FIN | (TH_FIN << 8))
1210 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
1212 case TH_SYN: /* opening */
1213 q->expire = time_uptime + dyn_syn_lifetime;
1216 case BOTH_SYN: /* move to established */
1217 case BOTH_SYN | TH_FIN : /* one side tries to close */
1218 case BOTH_SYN | (TH_FIN << 8) :
1220 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
1221 u_int32_t ack = ntohl(tcp->th_ack);
1222 if (dir == MATCH_FORWARD) {
1223 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
1225 else { /* ignore out-of-sequence */
1229 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
1231 else { /* ignore out-of-sequence */
1236 q->expire = time_uptime + dyn_ack_lifetime;
1239 case BOTH_SYN | BOTH_FIN: /* both sides closed */
1240 if (dyn_fin_lifetime >= dyn_keepalive_period)
1241 dyn_fin_lifetime = dyn_keepalive_period - 1;
1242 q->expire = time_uptime + dyn_fin_lifetime;
1248 * reset or some invalid combination, but can also
1249 * occur if we use keep-state the wrong way.
1251 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
1252 printf("invalid state: 0x%x\n", q->state);
1254 if (dyn_rst_lifetime >= dyn_keepalive_period)
1255 dyn_rst_lifetime = dyn_keepalive_period - 1;
1256 q->expire = time_uptime + dyn_rst_lifetime;
1259 } else if (pkt->proto == IPPROTO_UDP) {
1260 q->expire = time_uptime + dyn_udp_lifetime;
1262 /* other protocols */
1263 q->expire = time_uptime + dyn_short_lifetime;
1266 if (match_direction)
1267 *match_direction = dir;
1271 static ipfw_dyn_rule *
1272 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
1278 q = lookup_dyn_rule_locked(pkt, match_direction, tcp);
1281 /* NB: return table locked when q is not NULL */
1286 realloc_dynamic_table(void)
1288 IPFW_DYN_LOCK_ASSERT();
1291 * Try reallocation, make sure we have a power of 2 and do
1292 * not allow more than 64k entries. In case of overflow,
1296 if (dyn_buckets > 65536)
1298 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
1299 dyn_buckets = curr_dyn_buckets; /* reset */
1302 curr_dyn_buckets = dyn_buckets;
1303 if (ipfw_dyn_v != NULL)
1304 free(ipfw_dyn_v, M_IPFW);
1306 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
1307 M_IPFW, M_NOWAIT | M_ZERO);
1308 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
1310 curr_dyn_buckets /= 2;
1315 * Install state of type 'type' for a dynamic session.
1316 * The hash table contains two type of rules:
1317 * - regular rules (O_KEEP_STATE)
1318 * - rules for sessions with limited number of sess per user
1319 * (O_LIMIT). When they are created, the parent is
1320 * increased by 1, and decreased on delete. In this case,
1321 * the third parameter is the parent rule and not the chain.
1322 * - "parent" rules for the above (O_LIMIT_PARENT).
1324 static ipfw_dyn_rule *
1325 add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
1330 IPFW_DYN_LOCK_ASSERT();
1332 if (ipfw_dyn_v == NULL ||
1333 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1334 realloc_dynamic_table();
1335 if (ipfw_dyn_v == NULL)
1336 return NULL; /* failed ! */
1338 i = hash_packet(id);
1340 r = uma_zalloc(ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO);
1342 printf ("ipfw: sorry cannot allocate state\n");
1346 /* increase refcount on parent, and set pointer */
1347 if (dyn_type == O_LIMIT) {
1348 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1349 if ( parent->dyn_type != O_LIMIT_PARENT)
1350 panic("invalid parent");
1353 rule = parent->rule;
1357 r->expire = time_uptime + dyn_syn_lifetime;
1359 r->dyn_type = dyn_type;
1360 r->pcnt = r->bcnt = 0;
1364 r->next = ipfw_dyn_v[i];
1367 DEB(printf("ipfw: add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1369 (r->id.src_ip), (r->id.src_port),
1370 (r->id.dst_ip), (r->id.dst_port),
1376 * lookup dynamic parent rule using pkt and rule as search keys.
1377 * If the lookup fails, then install one.
1379 static ipfw_dyn_rule *
1380 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1385 IPFW_DYN_LOCK_ASSERT();
1388 int is_v6 = IS_IP6_FLOW_ID(pkt);
1389 i = hash_packet( pkt );
1390 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
1391 if (q->dyn_type == O_LIMIT_PARENT &&
1393 pkt->proto == q->id.proto &&
1394 pkt->src_port == q->id.src_port &&
1395 pkt->dst_port == q->id.dst_port &&
1398 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1399 &(q->id.src_ip6)) &&
1400 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1401 &(q->id.dst_ip6))) ||
1403 pkt->src_ip == q->id.src_ip &&
1404 pkt->dst_ip == q->id.dst_ip)
1407 q->expire = time_uptime + dyn_short_lifetime;
1408 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);)
1412 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1416 * Install dynamic state for rule type cmd->o.opcode
1418 * Returns 1 (failure) if state is not installed because of errors or because
1419 * session limitations are enforced.
1422 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1423 struct ip_fw_args *args, uint32_t tablearg)
1425 static int last_log;
1430 printf("ipfw: %s: type %d 0x%08x %u -> 0x%08x %u\n",
1431 __func__, cmd->o.opcode,
1432 (args->f_id.src_ip), (args->f_id.src_port),
1433 (args->f_id.dst_ip), (args->f_id.dst_port));
1438 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
1440 if (q != NULL) { /* should never occur */
1441 if (last_log != time_uptime) {
1442 last_log = time_uptime;
1443 printf("ipfw: %s: entry already present, done\n",
1450 if (dyn_count >= dyn_max)
1451 /* Run out of slots, try to remove any expired rule. */
1452 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1454 if (dyn_count >= dyn_max) {
1455 if (last_log != time_uptime) {
1456 last_log = time_uptime;
1457 printf("ipfw: %s: Too many dynamic rules\n", __func__);
1460 return (1); /* cannot install, notify caller */
1463 switch (cmd->o.opcode) {
1464 case O_KEEP_STATE: /* bidir rule */
1465 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1468 case O_LIMIT: { /* limit number of sessions */
1469 struct ipfw_flow_id id;
1470 ipfw_dyn_rule *parent;
1471 uint32_t conn_limit;
1472 uint16_t limit_mask = cmd->limit_mask;
1474 conn_limit = (cmd->conn_limit == IP_FW_TABLEARG) ?
1475 tablearg : cmd->conn_limit;
1478 if (cmd->conn_limit == IP_FW_TABLEARG)
1479 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u "
1480 "(tablearg)\n", __func__, conn_limit);
1482 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u\n",
1483 __func__, conn_limit);
1486 id.dst_ip = id.src_ip = id.dst_port = id.src_port = 0;
1487 id.proto = args->f_id.proto;
1488 id.addr_type = args->f_id.addr_type;
1490 if (IS_IP6_FLOW_ID (&(args->f_id))) {
1491 if (limit_mask & DYN_SRC_ADDR)
1492 id.src_ip6 = args->f_id.src_ip6;
1493 if (limit_mask & DYN_DST_ADDR)
1494 id.dst_ip6 = args->f_id.dst_ip6;
1496 if (limit_mask & DYN_SRC_ADDR)
1497 id.src_ip = args->f_id.src_ip;
1498 if (limit_mask & DYN_DST_ADDR)
1499 id.dst_ip = args->f_id.dst_ip;
1501 if (limit_mask & DYN_SRC_PORT)
1502 id.src_port = args->f_id.src_port;
1503 if (limit_mask & DYN_DST_PORT)
1504 id.dst_port = args->f_id.dst_port;
1505 if ((parent = lookup_dyn_parent(&id, rule)) == NULL) {
1506 printf("ipfw: %s: add parent failed\n", __func__);
1511 if (parent->count >= conn_limit) {
1512 /* See if we can remove some expired rule. */
1513 remove_dyn_rule(rule, parent);
1514 if (parent->count >= conn_limit) {
1515 if (fw_verbose && last_log != time_uptime) {
1516 last_log = time_uptime;
1517 log(LOG_SECURITY | LOG_DEBUG,
1518 "drop session, too many entries\n");
1524 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1528 printf("ipfw: %s: unknown dynamic rule type %u\n",
1529 __func__, cmd->o.opcode);
1534 /* XXX just set lifetime */
1535 lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
1542 * Generate a TCP packet, containing either a RST or a keepalive.
1543 * When flags & TH_RST, we are sending a RST packet, because of a
1544 * "reset" action matched the packet.
1545 * Otherwise we are sending a keepalive, and flags & TH_
1546 * The 'replyto' mbuf is the mbuf being replied to, if any, and is required
1547 * so that MAC can label the reply appropriately.
1549 static struct mbuf *
1550 send_pkt(struct mbuf *replyto, struct ipfw_flow_id *id, u_int32_t seq,
1551 u_int32_t ack, int flags)
1557 MGETHDR(m, M_DONTWAIT, MT_DATA);
1560 m->m_pkthdr.rcvif = (struct ifnet *)0;
1563 if (replyto != NULL)
1564 mac_create_mbuf_netlayer(replyto, m);
1566 mac_create_mbuf_from_firewall(m);
1568 (void)replyto; /* don't warn about unused arg */
1571 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1572 m->m_data += max_linkhdr;
1574 ip = mtod(m, struct ip *);
1575 bzero(ip, m->m_len);
1576 tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1577 ip->ip_p = IPPROTO_TCP;
1580 * Assume we are sending a RST (or a keepalive in the reverse
1581 * direction), swap src and destination addresses and ports.
1583 ip->ip_src.s_addr = htonl(id->dst_ip);
1584 ip->ip_dst.s_addr = htonl(id->src_ip);
1585 tcp->th_sport = htons(id->dst_port);
1586 tcp->th_dport = htons(id->src_port);
1587 if (flags & TH_RST) { /* we are sending a RST */
1588 if (flags & TH_ACK) {
1589 tcp->th_seq = htonl(ack);
1590 tcp->th_ack = htonl(0);
1591 tcp->th_flags = TH_RST;
1595 tcp->th_seq = htonl(0);
1596 tcp->th_ack = htonl(seq);
1597 tcp->th_flags = TH_RST | TH_ACK;
1601 * We are sending a keepalive. flags & TH_SYN determines
1602 * the direction, forward if set, reverse if clear.
1603 * NOTE: seq and ack are always assumed to be correct
1604 * as set by the caller. This may be confusing...
1606 if (flags & TH_SYN) {
1608 * we have to rewrite the correct addresses!
1610 ip->ip_dst.s_addr = htonl(id->dst_ip);
1611 ip->ip_src.s_addr = htonl(id->src_ip);
1612 tcp->th_dport = htons(id->dst_port);
1613 tcp->th_sport = htons(id->src_port);
1615 tcp->th_seq = htonl(seq);
1616 tcp->th_ack = htonl(ack);
1617 tcp->th_flags = TH_ACK;
1620 * set ip_len to the payload size so we can compute
1621 * the tcp checksum on the pseudoheader
1622 * XXX check this, could save a couple of words ?
1624 ip->ip_len = htons(sizeof(struct tcphdr));
1625 tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1627 * now fill fields left out earlier
1629 ip->ip_ttl = ip_defttl;
1630 ip->ip_len = m->m_pkthdr.len;
1631 m->m_flags |= M_SKIP_FIREWALL;
1636 * sends a reject message, consuming the mbuf passed as an argument.
1639 send_reject(struct ip_fw_args *args, int code, int ip_len)
1642 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1643 /* We need the IP header in host order for icmp_error(). */
1644 if (args->eh != NULL) {
1645 struct ip *ip = mtod(args->m, struct ip *);
1646 ip->ip_len = ntohs(ip->ip_len);
1647 ip->ip_off = ntohs(ip->ip_off);
1649 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1650 } else if (args->f_id.proto == IPPROTO_TCP) {
1651 struct tcphdr *const tcp =
1652 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1653 if ( (tcp->th_flags & TH_RST) == 0) {
1655 m = send_pkt(args->m, &(args->f_id),
1656 ntohl(tcp->th_seq), ntohl(tcp->th_ack),
1657 tcp->th_flags | TH_RST);
1659 ip_output(m, NULL, NULL, 0, NULL, NULL);
1669 * Given an ip_fw *, lookup_next_rule will return a pointer
1670 * to the next rule, which can be either the jump
1671 * target (for skipto instructions) or the next one in the list (in
1672 * all other cases including a missing jump target).
1673 * The result is also written in the "next_rule" field of the rule.
1674 * Backward jumps are not allowed, so start looking from the next
1677 * This never returns NULL -- in case we do not have an exact match,
1678 * the next rule is returned. When the ruleset is changed,
1679 * pointers are flushed so we are always correct.
1682 static struct ip_fw *
1683 lookup_next_rule(struct ip_fw *me)
1685 struct ip_fw *rule = NULL;
1688 /* look for action, in case it is a skipto */
1689 cmd = ACTION_PTR(me);
1690 if (cmd->opcode == O_LOG)
1692 if (cmd->opcode == O_ALTQ)
1694 if (cmd->opcode == O_TAG)
1696 if ( cmd->opcode == O_SKIPTO )
1697 for (rule = me->next; rule ; rule = rule->next)
1698 if (rule->rulenum >= cmd->arg1)
1700 if (rule == NULL) /* failure or not a skipto */
1702 me->next_rule = rule;
1707 add_table_entry(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1708 uint8_t mlen, uint32_t value)
1710 struct radix_node_head *rnh;
1711 struct table_entry *ent;
1713 if (tbl >= IPFW_TABLES_MAX)
1715 rnh = ch->tables[tbl];
1716 ent = malloc(sizeof(*ent), M_IPFW_TBL, M_NOWAIT | M_ZERO);
1720 ent->addr.sin_len = ent->mask.sin_len = 8;
1721 ent->mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1722 ent->addr.sin_addr.s_addr = addr & ent->mask.sin_addr.s_addr;
1723 IPFW_WLOCK(&layer3_chain);
1724 if (rnh->rnh_addaddr(&ent->addr, &ent->mask, rnh, (void *)ent) ==
1726 IPFW_WUNLOCK(&layer3_chain);
1727 free(ent, M_IPFW_TBL);
1730 IPFW_WUNLOCK(&layer3_chain);
1735 del_table_entry(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1738 struct radix_node_head *rnh;
1739 struct table_entry *ent;
1740 struct sockaddr_in sa, mask;
1742 if (tbl >= IPFW_TABLES_MAX)
1744 rnh = ch->tables[tbl];
1745 sa.sin_len = mask.sin_len = 8;
1746 mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1747 sa.sin_addr.s_addr = addr & mask.sin_addr.s_addr;
1749 ent = (struct table_entry *)rnh->rnh_deladdr(&sa, &mask, rnh);
1755 free(ent, M_IPFW_TBL);
1760 flush_table_entry(struct radix_node *rn, void *arg)
1762 struct radix_node_head * const rnh = arg;
1763 struct table_entry *ent;
1765 ent = (struct table_entry *)
1766 rnh->rnh_deladdr(rn->rn_key, rn->rn_mask, rnh);
1768 free(ent, M_IPFW_TBL);
1773 flush_table(struct ip_fw_chain *ch, uint16_t tbl)
1775 struct radix_node_head *rnh;
1777 IPFW_WLOCK_ASSERT(ch);
1779 if (tbl >= IPFW_TABLES_MAX)
1781 rnh = ch->tables[tbl];
1782 KASSERT(rnh != NULL, ("NULL IPFW table"));
1783 rnh->rnh_walktree(rnh, flush_table_entry, rnh);
1788 flush_tables(struct ip_fw_chain *ch)
1792 IPFW_WLOCK_ASSERT(ch);
1794 for (tbl = 0; tbl < IPFW_TABLES_MAX; tbl++)
1795 flush_table(ch, tbl);
1799 init_tables(struct ip_fw_chain *ch)
1804 for (i = 0; i < IPFW_TABLES_MAX; i++) {
1805 if (!rn_inithead((void **)&ch->tables[i], 32)) {
1806 for (j = 0; j < i; j++) {
1807 (void) flush_table(ch, j);
1816 lookup_table(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1819 struct radix_node_head *rnh;
1820 struct table_entry *ent;
1821 struct sockaddr_in sa;
1823 if (tbl >= IPFW_TABLES_MAX)
1825 rnh = ch->tables[tbl];
1827 sa.sin_addr.s_addr = addr;
1828 ent = (struct table_entry *)(rnh->rnh_lookup(&sa, NULL, rnh));
1837 count_table_entry(struct radix_node *rn, void *arg)
1839 u_int32_t * const cnt = arg;
1846 count_table(struct ip_fw_chain *ch, uint32_t tbl, uint32_t *cnt)
1848 struct radix_node_head *rnh;
1850 if (tbl >= IPFW_TABLES_MAX)
1852 rnh = ch->tables[tbl];
1854 rnh->rnh_walktree(rnh, count_table_entry, cnt);
1859 dump_table_entry(struct radix_node *rn, void *arg)
1861 struct table_entry * const n = (struct table_entry *)rn;
1862 ipfw_table * const tbl = arg;
1863 ipfw_table_entry *ent;
1865 if (tbl->cnt == tbl->size)
1867 ent = &tbl->ent[tbl->cnt];
1868 ent->tbl = tbl->tbl;
1869 if (in_nullhost(n->mask.sin_addr))
1872 ent->masklen = 33 - ffs(ntohl(n->mask.sin_addr.s_addr));
1873 ent->addr = n->addr.sin_addr.s_addr;
1874 ent->value = n->value;
1880 dump_table(struct ip_fw_chain *ch, ipfw_table *tbl)
1882 struct radix_node_head *rnh;
1884 if (tbl->tbl >= IPFW_TABLES_MAX)
1886 rnh = ch->tables[tbl->tbl];
1888 rnh->rnh_walktree(rnh, dump_table_entry, tbl);
1893 fill_ugid_cache(struct inpcb *inp, struct ip_fw_ugid *ugp)
1897 if (inp->inp_socket != NULL) {
1898 cr = inp->inp_socket->so_cred;
1899 ugp->fw_prid = jailed(cr) ?
1900 cr->cr_prison->pr_id : -1;
1901 ugp->fw_uid = cr->cr_uid;
1902 ugp->fw_ngroups = cr->cr_ngroups;
1903 bcopy(cr->cr_groups, ugp->fw_groups,
1904 sizeof(ugp->fw_groups));
1909 check_uidgid(ipfw_insn_u32 *insn,
1910 int proto, struct ifnet *oif,
1911 struct in_addr dst_ip, u_int16_t dst_port,
1912 struct in_addr src_ip, u_int16_t src_port,
1913 struct ip_fw_ugid *ugp, int *lookup, struct inpcb *inp)
1915 struct inpcbinfo *pi;
1922 * Check to see if the UDP or TCP stack supplied us with
1923 * the PCB. If so, rather then holding a lock and looking
1924 * up the PCB, we can use the one that was supplied.
1926 if (inp && *lookup == 0) {
1927 INP_LOCK_ASSERT(inp);
1928 if (inp->inp_socket != NULL) {
1929 fill_ugid_cache(inp, ugp);
1934 * If we have already been here and the packet has no
1935 * PCB entry associated with it, then we can safely
1936 * assume that this is a no match.
1940 if (proto == IPPROTO_TCP) {
1943 } else if (proto == IPPROTO_UDP) {
1944 wildcard = INPLOOKUP_WILDCARD;
1952 in_pcblookup_hash(pi,
1953 dst_ip, htons(dst_port),
1954 src_ip, htons(src_port),
1956 in_pcblookup_hash(pi,
1957 src_ip, htons(src_port),
1958 dst_ip, htons(dst_port),
1962 if (pcb->inp_socket != NULL) {
1963 fill_ugid_cache(pcb, ugp);
1968 INP_INFO_RUNLOCK(pi);
1971 * If the lookup did not yield any results, there
1972 * is no sense in coming back and trying again. So
1973 * we can set lookup to -1 and ensure that we wont
1974 * bother the pcb system again.
1980 if (insn->o.opcode == O_UID)
1981 match = (ugp->fw_uid == (uid_t)insn->d[0]);
1982 else if (insn->o.opcode == O_GID) {
1983 for (gp = ugp->fw_groups;
1984 gp < &ugp->fw_groups[ugp->fw_ngroups]; gp++)
1985 if (*gp == (gid_t)insn->d[0]) {
1989 } else if (insn->o.opcode == O_JAIL)
1990 match = (ugp->fw_prid == (int)insn->d[0]);
1995 * The main check routine for the firewall.
1997 * All arguments are in args so we can modify them and return them
1998 * back to the caller.
2002 * args->m (in/out) The packet; we set to NULL when/if we nuke it.
2003 * Starts with the IP header.
2004 * args->eh (in) Mac header if present, or NULL for layer3 packet.
2005 * args->oif Outgoing interface, or NULL if packet is incoming.
2006 * The incoming interface is in the mbuf. (in)
2007 * args->divert_rule (in/out)
2008 * Skip up to the first rule past this rule number;
2009 * upon return, non-zero port number for divert or tee.
2011 * args->rule Pointer to the last matching rule (in/out)
2012 * args->next_hop Socket we are forwarding to (out).
2013 * args->f_id Addresses grabbed from the packet (out)
2014 * args->cookie a cookie depending on rule action
2018 * IP_FW_PASS the packet must be accepted
2019 * IP_FW_DENY the packet must be dropped
2020 * IP_FW_DIVERT divert packet, port in m_tag
2021 * IP_FW_TEE tee packet, port in m_tag
2022 * IP_FW_DUMMYNET to dummynet, pipe in args->cookie
2023 * IP_FW_NETGRAPH into netgraph, cookie args->cookie
2028 ipfw_chk(struct ip_fw_args *args)
2031 * Local variables hold state during the processing of a packet.
2033 * IMPORTANT NOTE: to speed up the processing of rules, there
2034 * are some assumption on the values of the variables, which
2035 * are documented here. Should you change them, please check
2036 * the implementation of the various instructions to make sure
2037 * that they still work.
2039 * args->eh The MAC header. It is non-null for a layer2
2040 * packet, it is NULL for a layer-3 packet.
2042 * m | args->m Pointer to the mbuf, as received from the caller.
2043 * It may change if ipfw_chk() does an m_pullup, or if it
2044 * consumes the packet because it calls send_reject().
2045 * XXX This has to change, so that ipfw_chk() never modifies
2046 * or consumes the buffer.
2047 * ip is simply an alias of the value of m, and it is kept
2048 * in sync with it (the packet is supposed to start with
2051 struct mbuf *m = args->m;
2052 struct ip *ip = mtod(m, struct ip *);
2055 * For rules which contain uid/gid or jail constraints, cache
2056 * a copy of the users credentials after the pcb lookup has been
2057 * executed. This will speed up the processing of rules with
2058 * these types of constraints, as well as decrease contention
2059 * on pcb related locks.
2061 struct ip_fw_ugid fw_ugid_cache;
2062 int ugid_lookup = 0;
2065 * divinput_flags If non-zero, set to the IP_FW_DIVERT_*_FLAG
2066 * associated with a packet input on a divert socket. This
2067 * will allow to distinguish traffic and its direction when
2068 * it originates from a divert socket.
2070 u_int divinput_flags = 0;
2073 * oif | args->oif If NULL, ipfw_chk has been called on the
2074 * inbound path (ether_input, ip_input).
2075 * If non-NULL, ipfw_chk has been called on the outbound path
2076 * (ether_output, ip_output).
2078 struct ifnet *oif = args->oif;
2080 struct ip_fw *f = NULL; /* matching rule */
2084 * hlen The length of the IP header.
2086 u_int hlen = 0; /* hlen >0 means we have an IP pkt */
2089 * offset The offset of a fragment. offset != 0 means that
2090 * we have a fragment at this offset of an IPv4 packet.
2091 * offset == 0 means that (if this is an IPv4 packet)
2092 * this is the first or only fragment.
2093 * For IPv6 offset == 0 means there is no Fragment Header.
2094 * If offset != 0 for IPv6 always use correct mask to
2095 * get the correct offset because we add IP6F_MORE_FRAG
2096 * to be able to dectect the first fragment which would
2097 * otherwise have offset = 0.
2102 * Local copies of addresses. They are only valid if we have
2105 * proto The protocol. Set to 0 for non-ip packets,
2106 * or to the protocol read from the packet otherwise.
2107 * proto != 0 means that we have an IPv4 packet.
2109 * src_port, dst_port port numbers, in HOST format. Only
2110 * valid for TCP and UDP packets.
2112 * src_ip, dst_ip ip addresses, in NETWORK format.
2113 * Only valid for IPv4 packets.
2116 u_int16_t src_port = 0, dst_port = 0; /* NOTE: host format */
2117 struct in_addr src_ip, dst_ip; /* NOTE: network format */
2122 * dyn_dir = MATCH_UNKNOWN when rules unchecked,
2123 * MATCH_NONE when checked and not matched (q = NULL),
2124 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
2126 int dyn_dir = MATCH_UNKNOWN;
2127 ipfw_dyn_rule *q = NULL;
2128 struct ip_fw_chain *chain = &layer3_chain;
2132 * We store in ulp a pointer to the upper layer protocol header.
2133 * In the ipv4 case this is easy to determine from the header,
2134 * but for ipv6 we might have some additional headers in the middle.
2135 * ulp is NULL if not found.
2137 void *ulp = NULL; /* upper layer protocol pointer. */
2138 /* XXX ipv6 variables */
2140 u_int16_t ext_hd = 0; /* bits vector for extension header filtering */
2141 /* end of ipv6 variables */
2144 if (m->m_flags & M_SKIP_FIREWALL)
2145 return (IP_FW_PASS); /* accept */
2147 pktlen = m->m_pkthdr.len;
2148 proto = args->f_id.proto = 0; /* mark f_id invalid */
2149 /* XXX 0 is a valid proto: IP/IPv6 Hop-by-Hop Option */
2152 * PULLUP_TO(len, p, T) makes sure that len + sizeof(T) is contiguous,
2153 * then it sets p to point at the offset "len" in the mbuf. WARNING: the
2154 * pointer might become stale after other pullups (but we never use it
2157 #define PULLUP_TO(len, p, T) \
2159 int x = (len) + sizeof(T); \
2160 if ((m)->m_len < x) { \
2161 args->m = m = m_pullup(m, x); \
2163 goto pullup_failed; \
2165 p = (mtod(m, char *) + (len)); \
2168 /* Identify IP packets and fill up variables. */
2169 if (pktlen >= sizeof(struct ip6_hdr) &&
2170 (args->eh == NULL || ntohs(args->eh->ether_type)==ETHERTYPE_IPV6) &&
2171 mtod(m, struct ip *)->ip_v == 6) {
2173 args->f_id.addr_type = 6;
2174 hlen = sizeof(struct ip6_hdr);
2175 proto = mtod(m, struct ip6_hdr *)->ip6_nxt;
2177 /* Search extension headers to find upper layer protocols */
2178 while (ulp == NULL) {
2180 case IPPROTO_ICMPV6:
2181 PULLUP_TO(hlen, ulp, struct icmp6_hdr);
2182 args->f_id.flags = ICMP6(ulp)->icmp6_type;
2186 PULLUP_TO(hlen, ulp, struct tcphdr);
2187 dst_port = TCP(ulp)->th_dport;
2188 src_port = TCP(ulp)->th_sport;
2189 args->f_id.flags = TCP(ulp)->th_flags;
2193 PULLUP_TO(hlen, ulp, struct udphdr);
2194 dst_port = UDP(ulp)->uh_dport;
2195 src_port = UDP(ulp)->uh_sport;
2198 case IPPROTO_HOPOPTS: /* RFC 2460 */
2199 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2200 ext_hd |= EXT_HOPOPTS;
2201 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2202 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2206 case IPPROTO_ROUTING: /* RFC 2460 */
2207 PULLUP_TO(hlen, ulp, struct ip6_rthdr);
2208 switch (((struct ip6_rthdr *)ulp)->ip6r_type) {
2212 printf("IPFW2: IPV6 - Unknown Routing "
2213 "Header type(%d)\n",
2214 ((struct ip6_rthdr *)ulp)->ip6r_type);
2215 if (fw_deny_unknown_exthdrs)
2216 return (IP_FW_DENY);
2219 ext_hd |= EXT_ROUTING;
2220 hlen += (((struct ip6_rthdr *)ulp)->ip6r_len + 1) << 3;
2221 proto = ((struct ip6_rthdr *)ulp)->ip6r_nxt;
2225 case IPPROTO_FRAGMENT: /* RFC 2460 */
2226 PULLUP_TO(hlen, ulp, struct ip6_frag);
2227 ext_hd |= EXT_FRAGMENT;
2228 hlen += sizeof (struct ip6_frag);
2229 proto = ((struct ip6_frag *)ulp)->ip6f_nxt;
2230 offset = ((struct ip6_frag *)ulp)->ip6f_offlg &
2232 /* Add IP6F_MORE_FRAG for offset of first
2233 * fragment to be != 0. */
2234 offset |= ((struct ip6_frag *)ulp)->ip6f_offlg &
2237 printf("IPFW2: IPV6 - Invalid Fragment "
2239 if (fw_deny_unknown_exthdrs)
2240 return (IP_FW_DENY);
2243 args->f_id.frag_id6 =
2244 ntohl(((struct ip6_frag *)ulp)->ip6f_ident);
2248 case IPPROTO_DSTOPTS: /* RFC 2460 */
2249 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2250 ext_hd |= EXT_DSTOPTS;
2251 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2252 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2256 case IPPROTO_AH: /* RFC 2402 */
2257 PULLUP_TO(hlen, ulp, struct ip6_ext);
2259 hlen += (((struct ip6_ext *)ulp)->ip6e_len + 2) << 2;
2260 proto = ((struct ip6_ext *)ulp)->ip6e_nxt;
2264 case IPPROTO_ESP: /* RFC 2406 */
2265 PULLUP_TO(hlen, ulp, uint32_t); /* SPI, Seq# */
2266 /* Anything past Seq# is variable length and
2267 * data past this ext. header is encrypted. */
2271 case IPPROTO_NONE: /* RFC 2460 */
2272 PULLUP_TO(hlen, ulp, struct ip6_ext);
2273 /* Packet ends here. if ip6e_len!=0 octets
2274 * must be ignored. */
2277 case IPPROTO_OSPFIGP:
2278 /* XXX OSPF header check? */
2279 PULLUP_TO(hlen, ulp, struct ip6_ext);
2283 /* XXX PIM header check? */
2284 PULLUP_TO(hlen, ulp, struct pim);
2287 case IPPROTO_IPV6: /* RFC 2893 */
2288 PULLUP_TO(hlen, ulp, struct ip6_hdr);
2291 case IPPROTO_IPV4: /* RFC 2893 */
2292 PULLUP_TO(hlen, ulp, struct ip);
2296 printf("IPFW2: IPV6 - Unknown Extension "
2297 "Header(%d), ext_hd=%x\n", proto, ext_hd);
2298 if (fw_deny_unknown_exthdrs)
2299 return (IP_FW_DENY);
2300 PULLUP_TO(hlen, ulp, struct ip6_ext);
2304 args->f_id.src_ip6 = mtod(m,struct ip6_hdr *)->ip6_src;
2305 args->f_id.dst_ip6 = mtod(m,struct ip6_hdr *)->ip6_dst;
2306 args->f_id.src_ip = 0;
2307 args->f_id.dst_ip = 0;
2308 args->f_id.flow_id6 = ntohl(mtod(m, struct ip6_hdr *)->ip6_flow);
2309 } else if (pktlen >= sizeof(struct ip) &&
2310 (args->eh == NULL || ntohs(args->eh->ether_type) == ETHERTYPE_IP) &&
2311 mtod(m, struct ip *)->ip_v == 4) {
2313 ip = mtod(m, struct ip *);
2314 hlen = ip->ip_hl << 2;
2315 args->f_id.addr_type = 4;
2318 * Collect parameters into local variables for faster matching.
2321 src_ip = ip->ip_src;
2322 dst_ip = ip->ip_dst;
2323 if (args->eh != NULL) { /* layer 2 packets are as on the wire */
2324 offset = ntohs(ip->ip_off) & IP_OFFMASK;
2325 ip_len = ntohs(ip->ip_len);
2327 offset = ip->ip_off & IP_OFFMASK;
2328 ip_len = ip->ip_len;
2330 pktlen = ip_len < pktlen ? ip_len : pktlen;
2335 PULLUP_TO(hlen, ulp, struct tcphdr);
2336 dst_port = TCP(ulp)->th_dport;
2337 src_port = TCP(ulp)->th_sport;
2338 args->f_id.flags = TCP(ulp)->th_flags;
2342 PULLUP_TO(hlen, ulp, struct udphdr);
2343 dst_port = UDP(ulp)->uh_dport;
2344 src_port = UDP(ulp)->uh_sport;
2348 PULLUP_TO(hlen, ulp, struct icmphdr);
2349 args->f_id.flags = ICMP(ulp)->icmp_type;
2357 args->f_id.src_ip = ntohl(src_ip.s_addr);
2358 args->f_id.dst_ip = ntohl(dst_ip.s_addr);
2361 if (proto) { /* we may have port numbers, store them */
2362 args->f_id.proto = proto;
2363 args->f_id.src_port = src_port = ntohs(src_port);
2364 args->f_id.dst_port = dst_port = ntohs(dst_port);
2368 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
2371 * Packet has already been tagged. Look for the next rule
2372 * to restart processing.
2374 * If fw_one_pass != 0 then just accept it.
2375 * XXX should not happen here, but optimized out in
2379 IPFW_RUNLOCK(chain);
2380 return (IP_FW_PASS);
2383 f = args->rule->next_rule;
2385 f = lookup_next_rule(args->rule);
2388 * Find the starting rule. It can be either the first
2389 * one, or the one after divert_rule if asked so.
2391 int skipto = mtag ? divert_cookie(mtag) : 0;
2394 if (args->eh == NULL && skipto != 0) {
2395 if (skipto >= IPFW_DEFAULT_RULE) {
2396 IPFW_RUNLOCK(chain);
2397 return (IP_FW_DENY); /* invalid */
2399 while (f && f->rulenum <= skipto)
2401 if (f == NULL) { /* drop packet */
2402 IPFW_RUNLOCK(chain);
2403 return (IP_FW_DENY);
2407 /* reset divert rule to avoid confusion later */
2409 divinput_flags = divert_info(mtag) &
2410 (IP_FW_DIVERT_OUTPUT_FLAG | IP_FW_DIVERT_LOOPBACK_FLAG);
2411 m_tag_delete(m, mtag);
2415 * Now scan the rules, and parse microinstructions for each rule.
2417 for (; f; f = f->next) {
2419 uint32_t tablearg = 0;
2420 int l, cmdlen, skip_or; /* skip rest of OR block */
2423 if (set_disable & (1 << f->set) )
2427 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ;
2428 l -= cmdlen, cmd += cmdlen) {
2432 * check_body is a jump target used when we find a
2433 * CHECK_STATE, and need to jump to the body of
2438 cmdlen = F_LEN(cmd);
2440 * An OR block (insn_1 || .. || insn_n) has the
2441 * F_OR bit set in all but the last instruction.
2442 * The first match will set "skip_or", and cause
2443 * the following instructions to be skipped until
2444 * past the one with the F_OR bit clear.
2446 if (skip_or) { /* skip this instruction */
2447 if ((cmd->len & F_OR) == 0)
2448 skip_or = 0; /* next one is good */
2451 match = 0; /* set to 1 if we succeed */
2453 switch (cmd->opcode) {
2455 * The first set of opcodes compares the packet's
2456 * fields with some pattern, setting 'match' if a
2457 * match is found. At the end of the loop there is
2458 * logic to deal with F_NOT and F_OR flags associated
2466 printf("ipfw: opcode %d unimplemented\n",
2474 * We only check offset == 0 && proto != 0,
2475 * as this ensures that we have a
2476 * packet with the ports info.
2480 if (is_ipv6) /* XXX to be fixed later */
2482 if (proto == IPPROTO_TCP ||
2483 proto == IPPROTO_UDP)
2484 match = check_uidgid(
2485 (ipfw_insn_u32 *)cmd,
2488 src_ip, src_port, &fw_ugid_cache,
2489 &ugid_lookup, args->inp);
2493 match = iface_match(m->m_pkthdr.rcvif,
2494 (ipfw_insn_if *)cmd);
2498 match = iface_match(oif, (ipfw_insn_if *)cmd);
2502 match = iface_match(oif ? oif :
2503 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
2507 if (args->eh != NULL) { /* have MAC header */
2508 u_int32_t *want = (u_int32_t *)
2509 ((ipfw_insn_mac *)cmd)->addr;
2510 u_int32_t *mask = (u_int32_t *)
2511 ((ipfw_insn_mac *)cmd)->mask;
2512 u_int32_t *hdr = (u_int32_t *)args->eh;
2515 ( want[0] == (hdr[0] & mask[0]) &&
2516 want[1] == (hdr[1] & mask[1]) &&
2517 want[2] == (hdr[2] & mask[2]) );
2522 if (args->eh != NULL) {
2524 ntohs(args->eh->ether_type);
2526 ((ipfw_insn_u16 *)cmd)->ports;
2529 for (i = cmdlen - 1; !match && i>0;
2531 match = (t>=p[0] && t<=p[1]);
2536 match = (offset != 0);
2539 case O_IN: /* "out" is "not in" */
2540 match = (oif == NULL);
2544 match = (args->eh != NULL);
2548 match = (cmd->arg1 & 1 && divinput_flags &
2549 IP_FW_DIVERT_LOOPBACK_FLAG) ||
2550 (cmd->arg1 & 2 && divinput_flags &
2551 IP_FW_DIVERT_OUTPUT_FLAG);
2556 * We do not allow an arg of 0 so the
2557 * check of "proto" only suffices.
2559 match = (proto == cmd->arg1);
2564 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2568 case O_IP_SRC_LOOKUP:
2569 case O_IP_DST_LOOKUP:
2572 (cmd->opcode == O_IP_DST_LOOKUP) ?
2573 dst_ip.s_addr : src_ip.s_addr;
2576 match = lookup_table(chain, cmd->arg1, a,
2580 if (cmdlen == F_INSN_SIZE(ipfw_insn_u32))
2582 ((ipfw_insn_u32 *)cmd)->d[0] == v;
2592 (cmd->opcode == O_IP_DST_MASK) ?
2593 dst_ip.s_addr : src_ip.s_addr;
2594 uint32_t *p = ((ipfw_insn_u32 *)cmd)->d;
2597 for (; !match && i>0; i-= 2, p+= 2)
2598 match = (p[0] == (a & p[1]));
2606 INADDR_TO_IFP(src_ip, tif);
2607 match = (tif != NULL);
2614 u_int32_t *d = (u_int32_t *)(cmd+1);
2616 cmd->opcode == O_IP_DST_SET ?
2622 addr -= d[0]; /* subtract base */
2623 match = (addr < cmd->arg1) &&
2624 ( d[ 1 + (addr>>5)] &
2625 (1<<(addr & 0x1f)) );
2631 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2639 INADDR_TO_IFP(dst_ip, tif);
2640 match = (tif != NULL);
2647 * offset == 0 && proto != 0 is enough
2648 * to guarantee that we have a
2649 * packet with port info.
2651 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
2654 (cmd->opcode == O_IP_SRCPORT) ?
2655 src_port : dst_port ;
2657 ((ipfw_insn_u16 *)cmd)->ports;
2660 for (i = cmdlen - 1; !match && i>0;
2662 match = (x>=p[0] && x<=p[1]);
2667 match = (offset == 0 && proto==IPPROTO_ICMP &&
2668 icmptype_match(ICMP(ulp), (ipfw_insn_u32 *)cmd) );
2673 match = is_ipv6 && offset == 0 &&
2674 proto==IPPROTO_ICMPV6 &&
2676 ICMP6(ulp)->icmp6_type,
2677 (ipfw_insn_u32 *)cmd);
2683 ipopts_match(mtod(m, struct ip *), cmd) );
2688 cmd->arg1 == mtod(m, struct ip *)->ip_v);
2694 if (is_ipv4) { /* only for IP packets */
2699 if (cmd->opcode == O_IPLEN)
2701 else if (cmd->opcode == O_IPTTL)
2702 x = mtod(m, struct ip *)->ip_ttl;
2703 else /* must be IPID */
2704 x = ntohs(mtod(m, struct ip *)->ip_id);
2706 match = (cmd->arg1 == x);
2709 /* otherwise we have ranges */
2710 p = ((ipfw_insn_u16 *)cmd)->ports;
2712 for (; !match && i>0; i--, p += 2)
2713 match = (x >= p[0] && x <= p[1]);
2717 case O_IPPRECEDENCE:
2719 (cmd->arg1 == (mtod(m, struct ip *)->ip_tos & 0xe0)) );
2724 flags_match(cmd, mtod(m, struct ip *)->ip_tos));
2728 if (proto == IPPROTO_TCP && offset == 0) {
2736 ((ip->ip_hl + tcp->th_off) << 2);
2738 match = (cmd->arg1 == x);
2741 /* otherwise we have ranges */
2742 p = ((ipfw_insn_u16 *)cmd)->ports;
2744 for (; !match && i>0; i--, p += 2)
2745 match = (x >= p[0] && x <= p[1]);
2750 match = (proto == IPPROTO_TCP && offset == 0 &&
2751 flags_match(cmd, TCP(ulp)->th_flags));
2755 match = (proto == IPPROTO_TCP && offset == 0 &&
2756 tcpopts_match(TCP(ulp), cmd));
2760 match = (proto == IPPROTO_TCP && offset == 0 &&
2761 ((ipfw_insn_u32 *)cmd)->d[0] ==
2766 match = (proto == IPPROTO_TCP && offset == 0 &&
2767 ((ipfw_insn_u32 *)cmd)->d[0] ==
2772 match = (proto == IPPROTO_TCP && offset == 0 &&
2773 cmd->arg1 == TCP(ulp)->th_win);
2777 /* reject packets which have SYN only */
2778 /* XXX should i also check for TH_ACK ? */
2779 match = (proto == IPPROTO_TCP && offset == 0 &&
2780 (TCP(ulp)->th_flags &
2781 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2785 struct altq_tag *at;
2786 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
2789 mtag = m_tag_find(m, PACKET_TAG_PF_QID, NULL);
2792 mtag = m_tag_get(PACKET_TAG_PF_QID,
2793 sizeof(struct altq_tag),
2797 * Let the packet fall back to the
2802 at = (struct altq_tag *)(mtag+1);
2803 at->qid = altq->qid;
2809 m_tag_prepend(m, mtag);
2815 ipfw_log(f, hlen, args, m,
2816 oif, offset, tablearg);
2821 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]);
2825 /* Outgoing packets automatically pass/match */
2826 match = ((oif != NULL) ||
2827 (m->m_pkthdr.rcvif == NULL) ||
2831 verify_path6(&(args->f_id.src_ip6),
2832 m->m_pkthdr.rcvif) :
2834 verify_path(src_ip, m->m_pkthdr.rcvif)));
2838 /* Outgoing packets automatically pass/match */
2839 match = (hlen > 0 && ((oif != NULL) ||
2842 verify_path6(&(args->f_id.src_ip6),
2845 verify_path(src_ip, NULL)));
2849 /* Outgoing packets automatically pass/match */
2850 if (oif == NULL && hlen > 0 &&
2851 ( (is_ipv4 && in_localaddr(src_ip))
2854 in6_localaddr(&(args->f_id.src_ip6)))
2859 is_ipv6 ? verify_path6(
2860 &(args->f_id.src_ip6),
2861 m->m_pkthdr.rcvif) :
2871 match = (m_tag_find(m,
2872 PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL);
2875 match = (ipsec_getnhist(m) != 0);
2877 /* otherwise no match */
2883 IN6_ARE_ADDR_EQUAL(&args->f_id.src_ip6,
2884 &((ipfw_insn_ip6 *)cmd)->addr6);
2889 IN6_ARE_ADDR_EQUAL(&args->f_id.dst_ip6,
2890 &((ipfw_insn_ip6 *)cmd)->addr6);
2892 case O_IP6_SRC_MASK:
2893 case O_IP6_DST_MASK:
2897 struct in6_addr *d =
2898 &((ipfw_insn_ip6 *)cmd)->addr6;
2900 for (; !match && i > 0; d += 2,
2901 i -= F_INSN_SIZE(struct in6_addr)
2907 APPLY_MASK(&p, &d[1]);
2909 IN6_ARE_ADDR_EQUAL(&d[0],
2916 match= is_ipv6 && search_ip6_addr_net(&args->f_id.src_ip6);
2920 match= is_ipv6 && search_ip6_addr_net(&args->f_id.dst_ip6);
2925 flow6id_match(args->f_id.flow_id6,
2926 (ipfw_insn_u32 *) cmd);
2931 (ext_hd & ((ipfw_insn *) cmd)->arg1);
2944 uint32_t tag = (cmd->arg1 == IP_FW_TABLEARG) ?
2945 tablearg : cmd->arg1;
2947 /* Packet is already tagged with this tag? */
2948 mtag = m_tag_locate(m, MTAG_IPFW, tag, NULL);
2950 /* We have `untag' action when F_NOT flag is
2951 * present. And we must remove this mtag from
2952 * mbuf and reset `match' to zero (`match' will
2953 * be inversed later).
2954 * Otherwise we should allocate new mtag and
2955 * push it into mbuf.
2957 if (cmd->len & F_NOT) { /* `untag' action */
2959 m_tag_delete(m, mtag);
2960 } else if (mtag == NULL) {
2961 if ((mtag = m_tag_alloc(MTAG_IPFW,
2962 tag, 0, M_NOWAIT)) != NULL)
2963 m_tag_prepend(m, mtag);
2965 match = (cmd->len & F_NOT) ? 0: 1;
2970 uint32_t tag = (cmd->arg1 == IP_FW_TABLEARG) ?
2971 tablearg : cmd->arg1;
2974 match = m_tag_locate(m, MTAG_IPFW,
2979 /* we have ranges */
2980 for (mtag = m_tag_first(m);
2981 mtag != NULL && !match;
2982 mtag = m_tag_next(m, mtag)) {
2986 if (mtag->m_tag_cookie != MTAG_IPFW)
2989 p = ((ipfw_insn_u16 *)cmd)->ports;
2991 for(; !match && i > 0; i--, p += 2)
2993 mtag->m_tag_id >= p[0] &&
2994 mtag->m_tag_id <= p[1];
3000 * The second set of opcodes represents 'actions',
3001 * i.e. the terminal part of a rule once the packet
3002 * matches all previous patterns.
3003 * Typically there is only one action for each rule,
3004 * and the opcode is stored at the end of the rule
3005 * (but there are exceptions -- see below).
3007 * In general, here we set retval and terminate the
3008 * outer loop (would be a 'break 3' in some language,
3009 * but we need to do a 'goto done').
3012 * O_COUNT and O_SKIPTO actions:
3013 * instead of terminating, we jump to the next rule
3014 * ('goto next_rule', equivalent to a 'break 2'),
3015 * or to the SKIPTO target ('goto again' after
3016 * having set f, cmd and l), respectively.
3018 * O_TAG, O_LOG and O_ALTQ action parameters:
3019 * perform some action and set match = 1;
3021 * O_LIMIT and O_KEEP_STATE: these opcodes are
3022 * not real 'actions', and are stored right
3023 * before the 'action' part of the rule.
3024 * These opcodes try to install an entry in the
3025 * state tables; if successful, we continue with
3026 * the next opcode (match=1; break;), otherwise
3027 * the packet * must be dropped
3028 * ('goto done' after setting retval);
3030 * O_PROBE_STATE and O_CHECK_STATE: these opcodes
3031 * cause a lookup of the state table, and a jump
3032 * to the 'action' part of the parent rule
3033 * ('goto check_body') if an entry is found, or
3034 * (CHECK_STATE only) a jump to the next rule if
3035 * the entry is not found ('goto next_rule').
3036 * The result of the lookup is cached to make
3037 * further instances of these opcodes are
3042 if (install_state(f,
3043 (ipfw_insn_limit *)cmd, args, tablearg)) {
3044 retval = IP_FW_DENY;
3045 goto done; /* error/limit violation */
3053 * dynamic rules are checked at the first
3054 * keep-state or check-state occurrence,
3055 * with the result being stored in dyn_dir.
3056 * The compiler introduces a PROBE_STATE
3057 * instruction for us when we have a
3058 * KEEP_STATE (because PROBE_STATE needs
3061 if (dyn_dir == MATCH_UNKNOWN &&
3062 (q = lookup_dyn_rule(&args->f_id,
3063 &dyn_dir, proto == IPPROTO_TCP ?
3067 * Found dynamic entry, update stats
3068 * and jump to the 'action' part of
3074 cmd = ACTION_PTR(f);
3075 l = f->cmd_len - f->act_ofs;
3080 * Dynamic entry not found. If CHECK_STATE,
3081 * skip to next rule, if PROBE_STATE just
3082 * ignore and continue with next opcode.
3084 if (cmd->opcode == O_CHECK_STATE)
3090 retval = 0; /* accept */
3095 args->rule = f; /* report matching rule */
3096 if (cmd->arg1 == IP_FW_TABLEARG)
3097 args->cookie = tablearg;
3099 args->cookie = cmd->arg1;
3100 retval = IP_FW_DUMMYNET;
3105 struct divert_tag *dt;
3107 if (args->eh) /* not on layer 2 */
3109 mtag = m_tag_get(PACKET_TAG_DIVERT,
3110 sizeof(struct divert_tag),
3115 IPFW_RUNLOCK(chain);
3116 return (IP_FW_DENY);
3118 dt = (struct divert_tag *)(mtag+1);
3119 dt->cookie = f->rulenum;
3120 if (cmd->arg1 == IP_FW_TABLEARG)
3121 dt->info = tablearg;
3123 dt->info = cmd->arg1;
3124 m_tag_prepend(m, mtag);
3125 retval = (cmd->opcode == O_DIVERT) ?
3126 IP_FW_DIVERT : IP_FW_TEE;
3132 f->pcnt++; /* update stats */
3134 f->timestamp = time_uptime;
3135 if (cmd->opcode == O_COUNT)
3138 if (f->next_rule == NULL)
3139 lookup_next_rule(f);
3145 * Drop the packet and send a reject notice
3146 * if the packet is not ICMP (or is an ICMP
3147 * query), and it is not multicast/broadcast.
3149 if (hlen > 0 && is_ipv4 && offset == 0 &&
3150 (proto != IPPROTO_ICMP ||
3151 is_icmp_query(ICMP(ulp))) &&
3152 !(m->m_flags & (M_BCAST|M_MCAST)) &&
3153 !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
3154 send_reject(args, cmd->arg1, ip_len);
3160 if (hlen > 0 && is_ipv6 &&
3161 ((offset & IP6F_OFF_MASK) == 0) &&
3162 (proto != IPPROTO_ICMPV6 ||
3163 (is_icmp6_query(args->f_id.flags) == 1)) &&
3164 !(m->m_flags & (M_BCAST|M_MCAST)) &&
3165 !IN6_IS_ADDR_MULTICAST(&args->f_id.dst_ip6)) {
3166 send_reject6(args, cmd->arg1, hlen);
3172 retval = IP_FW_DENY;
3175 case O_FORWARD_IP: {
3176 struct sockaddr_in *sa;
3177 sa = &(((ipfw_insn_sa *)cmd)->sa);
3178 if (args->eh) /* not valid on layer2 pkts */
3180 if (!q || dyn_dir == MATCH_FORWARD) {
3181 if (sa->sin_addr.s_addr == INADDR_ANY) {
3182 bcopy(sa, &args->hopstore,
3184 args->hopstore.sin_addr.s_addr =
3189 args->next_hop = sa;
3192 retval = IP_FW_PASS;
3198 args->rule = f; /* report matching rule */
3199 if (cmd->arg1 == IP_FW_TABLEARG)
3200 args->cookie = tablearg;
3202 args->cookie = cmd->arg1;
3203 retval = (cmd->opcode == O_NETGRAPH) ?
3204 IP_FW_NETGRAPH : IP_FW_NGTEE;
3208 panic("-- unknown opcode %d\n", cmd->opcode);
3209 } /* end of switch() on opcodes */
3211 if (cmd->len & F_NOT)
3215 if (cmd->len & F_OR)
3218 if (!(cmd->len & F_OR)) /* not an OR block, */
3219 break; /* try next rule */
3222 } /* end of inner for, scan opcodes */
3224 next_rule:; /* try next rule */
3226 } /* end of outer for, scan rules */
3227 printf("ipfw: ouch!, skip past end of rules, denying packet\n");
3228 IPFW_RUNLOCK(chain);
3229 return (IP_FW_DENY);
3232 /* Update statistics */
3235 f->timestamp = time_uptime;
3236 IPFW_RUNLOCK(chain);
3241 printf("ipfw: pullup failed\n");
3242 return (IP_FW_DENY);
3246 * When a rule is added/deleted, clear the next_rule pointers in all rules.
3247 * These will be reconstructed on the fly as packets are matched.
3250 flush_rule_ptrs(struct ip_fw_chain *chain)
3254 IPFW_WLOCK_ASSERT(chain);
3256 for (rule = chain->rules; rule; rule = rule->next)
3257 rule->next_rule = NULL;
3261 * Add a new rule to the list. Copy the rule into a malloc'ed area, then
3262 * possibly create a rule number and add the rule to the list.
3263 * Update the rule_number in the input struct so the caller knows it as well.
3266 add_rule(struct ip_fw_chain *chain, struct ip_fw *input_rule)
3268 struct ip_fw *rule, *f, *prev;
3269 int l = RULESIZE(input_rule);
3271 if (chain->rules == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE)
3274 rule = malloc(l, M_IPFW, M_NOWAIT | M_ZERO);
3278 bcopy(input_rule, rule, l);
3281 rule->next_rule = NULL;
3285 rule->timestamp = 0;
3289 if (chain->rules == NULL) { /* default rule */
3290 chain->rules = rule;
3295 * If rulenum is 0, find highest numbered rule before the
3296 * default rule, and add autoinc_step
3298 if (autoinc_step < 1)
3300 else if (autoinc_step > 1000)
3301 autoinc_step = 1000;
3302 if (rule->rulenum == 0) {
3304 * locate the highest numbered rule before default
3306 for (f = chain->rules; f; f = f->next) {
3307 if (f->rulenum == IPFW_DEFAULT_RULE)
3309 rule->rulenum = f->rulenum;
3311 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
3312 rule->rulenum += autoinc_step;
3313 input_rule->rulenum = rule->rulenum;
3317 * Now insert the new rule in the right place in the sorted list.
3319 for (prev = NULL, f = chain->rules; f; prev = f, f = f->next) {
3320 if (f->rulenum > rule->rulenum) { /* found the location */
3324 } else { /* head insert */
3325 rule->next = chain->rules;
3326 chain->rules = rule;
3331 flush_rule_ptrs(chain);
3335 IPFW_WUNLOCK(chain);
3336 DEB(printf("ipfw: installed rule %d, static count now %d\n",
3337 rule->rulenum, static_count);)
3342 * Remove a static rule (including derived * dynamic rules)
3343 * and place it on the ``reap list'' for later reclamation.
3344 * The caller is in charge of clearing rule pointers to avoid
3345 * dangling pointers.
3346 * @return a pointer to the next entry.
3347 * Arguments are not checked, so they better be correct.
3349 static struct ip_fw *
3350 remove_rule(struct ip_fw_chain *chain, struct ip_fw *rule, struct ip_fw *prev)
3353 int l = RULESIZE(rule);
3355 IPFW_WLOCK_ASSERT(chain);
3359 remove_dyn_rule(rule, NULL /* force removal */);
3368 rule->next = chain->reap;
3375 * Reclaim storage associated with a list of rules. This is
3376 * typically the list created using remove_rule.
3379 reap_rules(struct ip_fw *head)
3383 while ((rule = head) != NULL) {
3385 if (DUMMYNET_LOADED)
3386 ip_dn_ruledel_ptr(rule);
3392 * Remove all rules from a chain (except rules in set RESVD_SET
3393 * unless kill_default = 1). The caller is responsible for
3394 * reclaiming storage for the rules left in chain->reap.
3397 free_chain(struct ip_fw_chain *chain, int kill_default)
3399 struct ip_fw *prev, *rule;
3401 IPFW_WLOCK_ASSERT(chain);
3403 flush_rule_ptrs(chain); /* more efficient to do outside the loop */
3404 for (prev = NULL, rule = chain->rules; rule ; )
3405 if (kill_default || rule->set != RESVD_SET)
3406 rule = remove_rule(chain, rule, prev);
3414 * Remove all rules with given number, and also do set manipulation.
3415 * Assumes chain != NULL && *chain != NULL.
3417 * The argument is an u_int32_t. The low 16 bit are the rule or set number,
3418 * the next 8 bits are the new set, the top 8 bits are the command:
3420 * 0 delete rules with given number
3421 * 1 delete rules with given set number
3422 * 2 move rules with given number to new set
3423 * 3 move rules with given set number to new set
3424 * 4 swap sets with given numbers
3427 del_entry(struct ip_fw_chain *chain, u_int32_t arg)
3429 struct ip_fw *prev = NULL, *rule;
3430 u_int16_t rulenum; /* rule or old_set */
3431 u_int8_t cmd, new_set;
3433 rulenum = arg & 0xffff;
3434 cmd = (arg >> 24) & 0xff;
3435 new_set = (arg >> 16) & 0xff;
3439 if (new_set > RESVD_SET)
3441 if (cmd == 0 || cmd == 2) {
3442 if (rulenum >= IPFW_DEFAULT_RULE)
3445 if (rulenum > RESVD_SET) /* old_set */
3450 rule = chain->rules;
3453 case 0: /* delete rules with given number */
3455 * locate first rule to delete
3457 for (; rule->rulenum < rulenum; prev = rule, rule = rule->next)
3459 if (rule->rulenum != rulenum) {
3460 IPFW_WUNLOCK(chain);
3465 * flush pointers outside the loop, then delete all matching
3466 * rules. prev remains the same throughout the cycle.
3468 flush_rule_ptrs(chain);
3469 while (rule->rulenum == rulenum)
3470 rule = remove_rule(chain, rule, prev);
3473 case 1: /* delete all rules with given set number */
3474 flush_rule_ptrs(chain);
3475 rule = chain->rules;
3476 while (rule->rulenum < IPFW_DEFAULT_RULE)
3477 if (rule->set == rulenum)
3478 rule = remove_rule(chain, rule, prev);
3485 case 2: /* move rules with given number to new set */
3486 rule = chain->rules;
3487 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3488 if (rule->rulenum == rulenum)
3489 rule->set = new_set;
3492 case 3: /* move rules with given set number to new set */
3493 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3494 if (rule->set == rulenum)
3495 rule->set = new_set;
3498 case 4: /* swap two sets */
3499 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3500 if (rule->set == rulenum)
3501 rule->set = new_set;
3502 else if (rule->set == new_set)
3503 rule->set = rulenum;
3507 * Look for rules to reclaim. We grab the list before
3508 * releasing the lock then reclaim them w/o the lock to
3509 * avoid a LOR with dummynet.
3513 IPFW_WUNLOCK(chain);
3520 * Clear counters for a specific rule.
3521 * The enclosing "table" is assumed locked.
3524 clear_counters(struct ip_fw *rule, int log_only)
3526 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
3528 if (log_only == 0) {
3529 rule->bcnt = rule->pcnt = 0;
3530 rule->timestamp = 0;
3532 if (l->o.opcode == O_LOG)
3533 l->log_left = l->max_log;
3537 * Reset some or all counters on firewall rules.
3538 * @arg frwl is null to clear all entries, or contains a specific
3540 * @arg log_only is 1 if we only want to reset logs, zero otherwise.
3543 zero_entry(struct ip_fw_chain *chain, int rulenum, int log_only)
3551 for (rule = chain->rules; rule; rule = rule->next)
3552 clear_counters(rule, log_only);
3553 msg = log_only ? "ipfw: All logging counts reset.\n" :
3554 "ipfw: Accounting cleared.\n";
3558 * We can have multiple rules with the same number, so we
3559 * need to clear them all.
3561 for (rule = chain->rules; rule; rule = rule->next)
3562 if (rule->rulenum == rulenum) {
3563 while (rule && rule->rulenum == rulenum) {
3564 clear_counters(rule, log_only);
3570 if (!cleared) { /* we did not find any matching rules */
3571 IPFW_WUNLOCK(chain);
3574 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
3575 "ipfw: Entry %d cleared.\n";
3577 IPFW_WUNLOCK(chain);
3580 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
3585 * Check validity of the structure before insert.
3586 * Fortunately rules are simple, so this mostly need to check rule sizes.
3589 check_ipfw_struct(struct ip_fw *rule, int size)
3595 if (size < sizeof(*rule)) {
3596 printf("ipfw: rule too short\n");
3599 /* first, check for valid size */
3602 printf("ipfw: size mismatch (have %d want %d)\n", size, l);
3605 if (rule->act_ofs >= rule->cmd_len) {
3606 printf("ipfw: bogus action offset (%u > %u)\n",
3607 rule->act_ofs, rule->cmd_len - 1);
3611 * Now go for the individual checks. Very simple ones, basically only
3612 * instruction sizes.
3614 for (l = rule->cmd_len, cmd = rule->cmd ;
3615 l > 0 ; l -= cmdlen, cmd += cmdlen) {
3616 cmdlen = F_LEN(cmd);
3618 printf("ipfw: opcode %d size truncated\n",
3622 DEB(printf("ipfw: opcode %d\n", cmd->opcode);)
3623 switch (cmd->opcode) {
3635 case O_IPPRECEDENCE:
3653 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3666 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3671 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
3676 if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
3679 ((ipfw_insn_log *)cmd)->log_left =
3680 ((ipfw_insn_log *)cmd)->max_log;
3686 /* only odd command lengths */
3687 if ( !(cmdlen & 1) || cmdlen > 31)
3693 if (cmd->arg1 == 0 || cmd->arg1 > 256) {
3694 printf("ipfw: invalid set size %d\n",
3698 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3703 case O_IP_SRC_LOOKUP:
3704 case O_IP_DST_LOOKUP:
3705 if (cmd->arg1 >= IPFW_TABLES_MAX) {
3706 printf("ipfw: invalid table number %d\n",
3710 if (cmdlen != F_INSN_SIZE(ipfw_insn) &&
3711 cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3716 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
3726 if (cmdlen < 1 || cmdlen > 31)
3732 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
3733 if (cmdlen < 2 || cmdlen > 31)
3740 if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
3745 if (cmdlen != F_INSN_SIZE(ipfw_insn_altq))
3751 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3756 #ifdef IPFIREWALL_FORWARD
3757 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
3766 if (ip_divert_ptr == NULL)
3772 if (!NG_IPFW_LOADED)
3776 case O_FORWARD_MAC: /* XXX not implemented yet */
3787 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3791 printf("ipfw: opcode %d, multiple actions"
3798 printf("ipfw: opcode %d, action must be"
3807 if (cmdlen != F_INSN_SIZE(struct in6_addr) +
3808 F_INSN_SIZE(ipfw_insn))
3813 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3814 ((ipfw_insn_u32 *)cmd)->o.arg1)
3818 case O_IP6_SRC_MASK:
3819 case O_IP6_DST_MASK:
3820 if ( !(cmdlen & 1) || cmdlen > 127)
3824 if( cmdlen != F_INSN_SIZE( ipfw_insn_icmp6 ) )
3830 switch (cmd->opcode) {
3840 case O_IP6_SRC_MASK:
3841 case O_IP6_DST_MASK:
3843 printf("ipfw: no IPv6 support in kernel\n");
3844 return EPROTONOSUPPORT;
3847 printf("ipfw: opcode %d, unknown opcode\n",
3853 if (have_action == 0) {
3854 printf("ipfw: missing action\n");
3860 printf("ipfw: opcode %d size %d wrong\n",
3861 cmd->opcode, cmdlen);
3866 * Copy the static and dynamic rules to the supplied buffer
3867 * and return the amount of space actually used.
3870 ipfw_getrules(struct ip_fw_chain *chain, void *buf, size_t space)
3873 char *ep = bp + space;
3877 /* XXX this can take a long time and locking will block packet flow */
3879 for (rule = chain->rules; rule ; rule = rule->next) {
3881 * Verify the entry fits in the buffer in case the
3882 * rules changed between calculating buffer space and
3883 * now. This would be better done using a generation
3884 * number but should suffice for now.
3889 bcopy(&set_disable, &(((struct ip_fw *)bp)->next_rule),
3890 sizeof(set_disable));
3894 IPFW_RUNLOCK(chain);
3896 ipfw_dyn_rule *p, *last = NULL;
3899 for (i = 0 ; i < curr_dyn_buckets; i++)
3900 for (p = ipfw_dyn_v[i] ; p != NULL; p = p->next) {
3901 if (bp + sizeof *p <= ep) {
3902 ipfw_dyn_rule *dst =
3903 (ipfw_dyn_rule *)bp;
3904 bcopy(p, dst, sizeof *p);
3905 bcopy(&(p->rule->rulenum), &(dst->rule),
3906 sizeof(p->rule->rulenum));
3908 * store a non-null value in "next".
3909 * The userland code will interpret a
3910 * NULL here as a marker
3911 * for the last dynamic rule.
3913 bcopy(&dst, &dst->next, sizeof(dst));
3916 TIME_LEQ(dst->expire, time_uptime) ?
3917 0 : dst->expire - time_uptime ;
3918 bp += sizeof(ipfw_dyn_rule);
3922 if (last != NULL) /* mark last dynamic rule */
3923 bzero(&last->next, sizeof(last));
3925 return (bp - (char *)buf);
3930 * {set|get}sockopt parser.
3933 ipfw_ctl(struct sockopt *sopt)
3935 #define RULE_MAXSIZE (256*sizeof(u_int32_t))
3936 int error, rule_num;
3938 struct ip_fw *buf, *rule;
3939 u_int32_t rulenum[2];
3941 error = suser(sopt->sopt_td);
3946 * Disallow modifications in really-really secure mode, but still allow
3947 * the logging counters to be reset.
3949 if (sopt->sopt_name == IP_FW_ADD ||
3950 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
3951 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
3958 switch (sopt->sopt_name) {
3961 * pass up a copy of the current rules. Static rules
3962 * come first (the last of which has number IPFW_DEFAULT_RULE),
3963 * followed by a possibly empty list of dynamic rule.
3964 * The last dynamic rule has NULL in the "next" field.
3966 * Note that the calculated size is used to bound the
3967 * amount of data returned to the user. The rule set may
3968 * change between calculating the size and returning the
3969 * data in which case we'll just return what fits.
3971 size = static_len; /* size of static rules */
3972 if (ipfw_dyn_v) /* add size of dyn.rules */
3973 size += (dyn_count * sizeof(ipfw_dyn_rule));
3976 * XXX todo: if the user passes a short length just to know
3977 * how much room is needed, do not bother filling up the
3978 * buffer, just jump to the sooptcopyout.
3980 buf = malloc(size, M_TEMP, M_WAITOK);
3981 error = sooptcopyout(sopt, buf,
3982 ipfw_getrules(&layer3_chain, buf, size));
3988 * Normally we cannot release the lock on each iteration.
3989 * We could do it here only because we start from the head all
3990 * the times so there is no risk of missing some entries.
3991 * On the other hand, the risk is that we end up with
3992 * a very inconsistent ruleset, so better keep the lock
3993 * around the whole cycle.
3995 * XXX this code can be improved by resetting the head of
3996 * the list to point to the default rule, and then freeing
3997 * the old list without the need for a lock.
4000 IPFW_WLOCK(&layer3_chain);
4001 layer3_chain.reap = NULL;
4002 free_chain(&layer3_chain, 0 /* keep default rule */);
4003 rule = layer3_chain.reap;
4004 layer3_chain.reap = NULL;
4005 IPFW_WUNLOCK(&layer3_chain);
4011 rule = malloc(RULE_MAXSIZE, M_TEMP, M_WAITOK);
4012 error = sooptcopyin(sopt, rule, RULE_MAXSIZE,
4013 sizeof(struct ip_fw) );
4015 error = check_ipfw_struct(rule, sopt->sopt_valsize);
4017 error = add_rule(&layer3_chain, rule);
4018 size = RULESIZE(rule);
4019 if (!error && sopt->sopt_dir == SOPT_GET)
4020 error = sooptcopyout(sopt, rule, size);
4027 * IP_FW_DEL is used for deleting single rules or sets,
4028 * and (ab)used to atomically manipulate sets. Argument size
4029 * is used to distinguish between the two:
4031 * delete single rule or set of rules,
4032 * or reassign rules (or sets) to a different set.
4033 * 2*sizeof(u_int32_t)
4034 * atomic disable/enable sets.
4035 * first u_int32_t contains sets to be disabled,
4036 * second u_int32_t contains sets to be enabled.
4038 error = sooptcopyin(sopt, rulenum,
4039 2*sizeof(u_int32_t), sizeof(u_int32_t));
4042 size = sopt->sopt_valsize;
4043 if (size == sizeof(u_int32_t)) /* delete or reassign */
4044 error = del_entry(&layer3_chain, rulenum[0]);
4045 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */
4047 (set_disable | rulenum[0]) & ~rulenum[1] &
4048 ~(1<<RESVD_SET); /* set RESVD_SET always enabled */
4054 case IP_FW_RESETLOG: /* argument is an int, the rule number */
4056 if (sopt->sopt_val != 0) {
4057 error = sooptcopyin(sopt, &rule_num,
4058 sizeof(int), sizeof(int));
4062 error = zero_entry(&layer3_chain, rule_num,
4063 sopt->sopt_name == IP_FW_RESETLOG);
4066 case IP_FW_TABLE_ADD:
4068 ipfw_table_entry ent;
4070 error = sooptcopyin(sopt, &ent,
4071 sizeof(ent), sizeof(ent));
4074 error = add_table_entry(&layer3_chain, ent.tbl,
4075 ent.addr, ent.masklen, ent.value);
4079 case IP_FW_TABLE_DEL:
4081 ipfw_table_entry ent;
4083 error = sooptcopyin(sopt, &ent,
4084 sizeof(ent), sizeof(ent));
4087 error = del_table_entry(&layer3_chain, ent.tbl,
4088 ent.addr, ent.masklen);
4092 case IP_FW_TABLE_FLUSH:
4096 error = sooptcopyin(sopt, &tbl,
4097 sizeof(tbl), sizeof(tbl));
4100 IPFW_WLOCK(&layer3_chain);
4101 error = flush_table(&layer3_chain, tbl);
4102 IPFW_WUNLOCK(&layer3_chain);
4106 case IP_FW_TABLE_GETSIZE:
4110 if ((error = sooptcopyin(sopt, &tbl, sizeof(tbl),
4113 IPFW_RLOCK(&layer3_chain);
4114 error = count_table(&layer3_chain, tbl, &cnt);
4115 IPFW_RUNLOCK(&layer3_chain);
4118 error = sooptcopyout(sopt, &cnt, sizeof(cnt));
4122 case IP_FW_TABLE_LIST:
4126 if (sopt->sopt_valsize < sizeof(*tbl)) {
4130 size = sopt->sopt_valsize;
4131 tbl = malloc(size, M_TEMP, M_WAITOK);
4132 error = sooptcopyin(sopt, tbl, size, sizeof(*tbl));
4137 tbl->size = (size - sizeof(*tbl)) /
4138 sizeof(ipfw_table_entry);
4139 IPFW_RLOCK(&layer3_chain);
4140 error = dump_table(&layer3_chain, tbl);
4141 IPFW_RUNLOCK(&layer3_chain);
4146 error = sooptcopyout(sopt, tbl, size);
4152 printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name);
4161 * dummynet needs a reference to the default rule, because rules can be
4162 * deleted while packets hold a reference to them. When this happens,
4163 * dummynet changes the reference to the default rule (it could well be a
4164 * NULL pointer, but this way we do not need to check for the special
4165 * case, plus here he have info on the default behaviour).
4167 struct ip_fw *ip_fw_default_rule;
4170 * This procedure is only used to handle keepalives. It is invoked
4171 * every dyn_keepalive_period
4174 ipfw_tick(void * __unused unused)
4176 struct mbuf *m0, *m, *mnext, **mtailp;
4180 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
4184 * We make a chain of packets to go out here -- not deferring
4185 * until after we drop the IPFW dynamic rule lock would result
4186 * in a lock order reversal with the normal packet input -> ipfw
4192 for (i = 0 ; i < curr_dyn_buckets ; i++) {
4193 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
4194 if (q->dyn_type == O_LIMIT_PARENT)
4196 if (q->id.proto != IPPROTO_TCP)
4198 if ( (q->state & BOTH_SYN) != BOTH_SYN)
4200 if (TIME_LEQ( time_uptime+dyn_keepalive_interval,
4202 continue; /* too early */
4203 if (TIME_LEQ(q->expire, time_uptime))
4204 continue; /* too late, rule expired */
4206 *mtailp = send_pkt(NULL, &(q->id), q->ack_rev - 1,
4207 q->ack_fwd, TH_SYN);
4208 if (*mtailp != NULL)
4209 mtailp = &(*mtailp)->m_nextpkt;
4210 *mtailp = send_pkt(NULL, &(q->id), q->ack_fwd - 1,
4212 if (*mtailp != NULL)
4213 mtailp = &(*mtailp)->m_nextpkt;
4217 for (m = mnext = m0; m != NULL; m = mnext) {
4218 mnext = m->m_nextpkt;
4219 m->m_nextpkt = NULL;
4220 ip_output(m, NULL, NULL, 0, NULL, NULL);
4223 callout_reset(&ipfw_timeout, dyn_keepalive_period*hz, ipfw_tick, NULL);
4229 struct ip_fw default_rule;
4233 /* Setup IPv6 fw sysctl tree. */
4234 sysctl_ctx_init(&ip6_fw_sysctl_ctx);
4235 ip6_fw_sysctl_tree = SYSCTL_ADD_NODE(&ip6_fw_sysctl_ctx,
4236 SYSCTL_STATIC_CHILDREN(_net_inet6_ip6), OID_AUTO, "fw",
4237 CTLFLAG_RW | CTLFLAG_SECURE, 0, "Firewall");
4238 SYSCTL_ADD_PROC(&ip6_fw_sysctl_ctx, SYSCTL_CHILDREN(ip6_fw_sysctl_tree),
4239 OID_AUTO, "enable", CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE3,
4240 &fw6_enable, 0, ipfw_chg_hook, "I", "Enable ipfw+6");
4241 SYSCTL_ADD_INT(&ip6_fw_sysctl_ctx, SYSCTL_CHILDREN(ip6_fw_sysctl_tree),
4242 OID_AUTO, "deny_unknown_exthdrs", CTLFLAG_RW | CTLFLAG_SECURE,
4243 &fw_deny_unknown_exthdrs, 0,
4244 "Deny packets with unknown IPv6 Extension Headers");
4247 layer3_chain.rules = NULL;
4248 IPFW_LOCK_INIT(&layer3_chain);
4249 ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule zone",
4250 sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL,
4252 IPFW_DYN_LOCK_INIT();
4253 callout_init(&ipfw_timeout, NET_CALLOUT_MPSAFE);
4255 bzero(&default_rule, sizeof default_rule);
4257 default_rule.act_ofs = 0;
4258 default_rule.rulenum = IPFW_DEFAULT_RULE;
4259 default_rule.cmd_len = 1;
4260 default_rule.set = RESVD_SET;
4262 default_rule.cmd[0].len = 1;
4263 default_rule.cmd[0].opcode =
4264 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
4269 error = add_rule(&layer3_chain, &default_rule);
4271 printf("ipfw2: error %u initializing default rule "
4272 "(support disabled)\n", error);
4273 IPFW_DYN_LOCK_DESTROY();
4274 IPFW_LOCK_DESTROY(&layer3_chain);
4275 uma_zdestroy(ipfw_dyn_rule_zone);
4279 ip_fw_default_rule = layer3_chain.rules;
4284 "initialized, divert %s, "
4285 "rule-based forwarding "
4286 #ifdef IPFIREWALL_FORWARD
4291 "default to %s, logging ",
4297 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny");
4299 #ifdef IPFIREWALL_VERBOSE
4302 #ifdef IPFIREWALL_VERBOSE_LIMIT
4303 verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
4305 if (fw_verbose == 0)
4306 printf("disabled\n");
4307 else if (verbose_limit == 0)
4308 printf("unlimited\n");
4310 printf("limited to %d packets/entry by default\n",
4313 error = init_tables(&layer3_chain);
4315 IPFW_DYN_LOCK_DESTROY();
4316 IPFW_LOCK_DESTROY(&layer3_chain);
4317 uma_zdestroy(ipfw_dyn_rule_zone);
4320 ip_fw_ctl_ptr = ipfw_ctl;
4321 ip_fw_chk_ptr = ipfw_chk;
4322 callout_reset(&ipfw_timeout, hz, ipfw_tick, NULL);
4332 ip_fw_chk_ptr = NULL;
4333 ip_fw_ctl_ptr = NULL;
4334 callout_drain(&ipfw_timeout);
4335 IPFW_WLOCK(&layer3_chain);
4336 flush_tables(&layer3_chain);
4337 layer3_chain.reap = NULL;
4338 free_chain(&layer3_chain, 1 /* kill default rule */);
4339 reap = layer3_chain.reap, layer3_chain.reap = NULL;
4340 IPFW_WUNLOCK(&layer3_chain);
4343 IPFW_DYN_LOCK_DESTROY();
4344 uma_zdestroy(ipfw_dyn_rule_zone);
4345 IPFW_LOCK_DESTROY(&layer3_chain);
4348 /* Free IPv6 fw sysctl tree. */
4349 sysctl_ctx_free(&ip6_fw_sysctl_ctx);
4352 printf("IP firewall unloaded\n");