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"
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/condvar.h>
49 #include <sys/malloc.h>
51 #include <sys/kernel.h>
54 #include <sys/module.h>
56 #include <sys/rwlock.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/sysctl.h>
60 #include <sys/syslog.h>
61 #include <sys/ucred.h>
63 #include <net/radix.h>
64 #include <net/route.h>
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/in_var.h>
68 #include <netinet/in_pcb.h>
69 #include <netinet/ip.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/ip_icmp.h>
72 #include <netinet/ip_fw.h>
73 #include <netinet/ip_divert.h>
74 #include <netinet/ip_dummynet.h>
75 #include <netinet/pim.h>
76 #include <netinet/tcp.h>
77 #include <netinet/tcp_timer.h>
78 #include <netinet/tcp_var.h>
79 #include <netinet/tcpip.h>
80 #include <netinet/udp.h>
81 #include <netinet/udp_var.h>
83 #include <netgraph/ng_ipfw.h>
85 #include <altq/if_altq.h>
88 #include <netinet6/ipsec.h>
91 #include <netinet/ip6.h>
92 #include <netinet/icmp6.h>
94 #include <netinet6/scope6_var.h>
97 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
99 #include <machine/in_cksum.h> /* XXX for in_cksum */
102 * set_disable contains one bit per set value (0..31).
103 * If the bit is set, all rules with the corresponding set
104 * are disabled. Set RESVD_SET(31) is reserved for the default rule
105 * and rules that are not deleted by the flush command,
106 * and CANNOT be disabled.
107 * Rules in set RESVD_SET can only be deleted explicitly.
109 static u_int32_t set_disable;
111 static int fw_verbose;
112 static int verbose_limit;
114 static struct callout ipfw_timeout;
115 static uma_zone_t ipfw_dyn_rule_zone;
116 #define IPFW_DEFAULT_RULE 65535
119 * Data structure to cache our ucred related
120 * information. This structure only gets used if
121 * the user specified UID/GID based constraints in
125 gid_t fw_groups[NGROUPS];
131 #define IPFW_TABLES_MAX 128
133 struct ip_fw *rules; /* list of rules */
134 struct ip_fw *reap; /* list of rules to reap */
135 struct radix_node_head *tables[IPFW_TABLES_MAX];
138 #define IPFW_LOCK_INIT(_chain) \
139 rw_init(&(_chain)->rwmtx, "IPFW static rules")
140 #define IPFW_LOCK_DESTROY(_chain) rw_destroy(&(_chain)->rwmtx)
141 #define IPFW_WLOCK_ASSERT(_chain) do { \
142 rw_assert(&(_chain)->rwmtx, RA_WLOCKED); \
143 NET_ASSERT_GIANT(); \
146 #define IPFW_RLOCK(p) rw_rlock(&(p)->rwmtx)
147 #define IPFW_RUNLOCK(p) rw_runlock(&(p)->rwmtx)
148 #define IPFW_WLOCK(p) rw_wlock(&(p)->rwmtx)
149 #define IPFW_WUNLOCK(p) rw_wunlock(&(p)->rwmtx)
152 * list of rules for layer 3
154 static struct ip_fw_chain layer3_chain;
156 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
157 MALLOC_DEFINE(M_IPFW_TBL, "ipfw_tbl", "IpFw tables");
160 struct radix_node rn[2];
161 struct sockaddr_in addr, mask;
165 static int fw_debug = 1;
166 static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
168 extern int ipfw_chg_hook(SYSCTL_HANDLER_ARGS);
171 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
172 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, enable,
173 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE3, &fw_enable, 0,
174 ipfw_chg_hook, "I", "Enable ipfw");
175 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
176 &autoinc_step, 0, "Rule number autincrement step");
177 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, one_pass,
178 CTLFLAG_RW | CTLFLAG_SECURE3,
180 "Only do a single pass through ipfw when using dummynet(4)");
181 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
182 &fw_debug, 0, "Enable printing of debug ip_fw statements");
183 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose,
184 CTLFLAG_RW | CTLFLAG_SECURE3,
185 &fw_verbose, 0, "Log matches to ipfw rules");
186 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
187 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
190 * Description of dynamic rules.
192 * Dynamic rules are stored in lists accessed through a hash table
193 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
194 * be modified through the sysctl variable dyn_buckets which is
195 * updated when the table becomes empty.
197 * XXX currently there is only one list, ipfw_dyn.
199 * When a packet is received, its address fields are first masked
200 * with the mask defined for the rule, then hashed, then matched
201 * against the entries in the corresponding list.
202 * Dynamic rules can be used for different purposes:
204 * + enforcing limits on the number of sessions;
205 * + in-kernel NAT (not implemented yet)
207 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
208 * measured in seconds and depending on the flags.
210 * The total number of dynamic rules is stored in dyn_count.
211 * The max number of dynamic rules is dyn_max. When we reach
212 * the maximum number of rules we do not create anymore. This is
213 * done to avoid consuming too much memory, but also too much
214 * time when searching on each packet (ideally, we should try instead
215 * to put a limit on the length of the list on each bucket...).
217 * Each dynamic rule holds a pointer to the parent ipfw rule so
218 * we know what action to perform. Dynamic rules are removed when
219 * the parent rule is deleted. XXX we should make them survive.
221 * There are some limitations with dynamic rules -- we do not
222 * obey the 'randomized match', and we do not do multiple
223 * passes through the firewall. XXX check the latter!!!
225 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
226 static u_int32_t dyn_buckets = 256; /* must be power of 2 */
227 static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */
229 static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */
230 #define IPFW_DYN_LOCK_INIT() \
231 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF)
232 #define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx)
233 #define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx)
234 #define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx)
235 #define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED)
238 * Timeouts for various events in handing dynamic rules.
240 static u_int32_t dyn_ack_lifetime = 300;
241 static u_int32_t dyn_syn_lifetime = 20;
242 static u_int32_t dyn_fin_lifetime = 1;
243 static u_int32_t dyn_rst_lifetime = 1;
244 static u_int32_t dyn_udp_lifetime = 10;
245 static u_int32_t dyn_short_lifetime = 5;
248 * Keepalives are sent if dyn_keepalive is set. They are sent every
249 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
250 * seconds of lifetime of a rule.
251 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
252 * than dyn_keepalive_period.
255 static u_int32_t dyn_keepalive_interval = 20;
256 static u_int32_t dyn_keepalive_period = 5;
257 static u_int32_t dyn_keepalive = 1; /* do send keepalives */
259 static u_int32_t static_count; /* # of static rules */
260 static u_int32_t static_len; /* size in bytes of static rules */
261 static u_int32_t dyn_count; /* # of dynamic rules */
262 static u_int32_t dyn_max = 4096; /* max # of dynamic rules */
264 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
265 &dyn_buckets, 0, "Number of dyn. buckets");
266 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
267 &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
268 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
269 &dyn_count, 0, "Number of dyn. rules");
270 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
271 &dyn_max, 0, "Max number of dyn. rules");
272 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
273 &static_count, 0, "Number of static rules");
274 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
275 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
276 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
277 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
278 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
279 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
280 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
281 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
282 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
283 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
284 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
285 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
286 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
287 &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
291 * IPv6 specific variables
293 SYSCTL_DECL(_net_inet6_ip6);
295 static struct sysctl_ctx_list ip6_fw_sysctl_ctx;
296 static struct sysctl_oid *ip6_fw_sysctl_tree;
298 #endif /* SYSCTL_NODE */
300 static int fw_deny_unknown_exthdrs = 1;
304 * L3HDR maps an ipv4 pointer into a layer3 header pointer of type T
305 * Other macros just cast void * into the appropriate type
307 #define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
308 #define TCP(p) ((struct tcphdr *)(p))
309 #define UDP(p) ((struct udphdr *)(p))
310 #define ICMP(p) ((struct icmphdr *)(p))
311 #define ICMP6(p) ((struct icmp6_hdr *)(p))
314 icmptype_match(struct icmphdr *icmp, ipfw_insn_u32 *cmd)
316 int type = icmp->icmp_type;
318 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
321 #define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
322 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
325 is_icmp_query(struct icmphdr *icmp)
327 int type = icmp->icmp_type;
329 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
334 * The following checks use two arrays of 8 or 16 bits to store the
335 * bits that we want set or clear, respectively. They are in the
336 * low and high half of cmd->arg1 or cmd->d[0].
338 * We scan options and store the bits we find set. We succeed if
340 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
342 * The code is sometimes optimized not to store additional variables.
346 flags_match(ipfw_insn *cmd, u_int8_t bits)
351 if ( ((cmd->arg1 & 0xff) & bits) != 0)
352 return 0; /* some bits we want set were clear */
353 want_clear = (cmd->arg1 >> 8) & 0xff;
354 if ( (want_clear & bits) != want_clear)
355 return 0; /* some bits we want clear were set */
360 ipopts_match(struct ip *ip, ipfw_insn *cmd)
362 int optlen, bits = 0;
363 u_char *cp = (u_char *)(ip + 1);
364 int x = (ip->ip_hl << 2) - sizeof (struct ip);
366 for (; x > 0; x -= optlen, cp += optlen) {
367 int opt = cp[IPOPT_OPTVAL];
369 if (opt == IPOPT_EOL)
371 if (opt == IPOPT_NOP)
374 optlen = cp[IPOPT_OLEN];
375 if (optlen <= 0 || optlen > x)
376 return 0; /* invalid or truncated */
384 bits |= IP_FW_IPOPT_LSRR;
388 bits |= IP_FW_IPOPT_SSRR;
392 bits |= IP_FW_IPOPT_RR;
396 bits |= IP_FW_IPOPT_TS;
400 return (flags_match(cmd, bits));
404 tcpopts_match(struct tcphdr *tcp, ipfw_insn *cmd)
406 int optlen, bits = 0;
407 u_char *cp = (u_char *)(tcp + 1);
408 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
410 for (; x > 0; x -= optlen, cp += optlen) {
412 if (opt == TCPOPT_EOL)
414 if (opt == TCPOPT_NOP)
428 bits |= IP_FW_TCPOPT_MSS;
432 bits |= IP_FW_TCPOPT_WINDOW;
435 case TCPOPT_SACK_PERMITTED:
437 bits |= IP_FW_TCPOPT_SACK;
440 case TCPOPT_TIMESTAMP:
441 bits |= IP_FW_TCPOPT_TS;
446 return (flags_match(cmd, bits));
450 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
452 if (ifp == NULL) /* no iface with this packet, match fails */
454 /* Check by name or by IP address */
455 if (cmd->name[0] != '\0') { /* match by name */
458 if (fnmatch(cmd->name, ifp->if_xname, 0) == 0)
461 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
468 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
469 if (ia->ifa_addr->sa_family != AF_INET)
471 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
472 (ia->ifa_addr))->sin_addr.s_addr)
473 return(1); /* match */
476 return(0); /* no match, fail ... */
480 * The verify_path function checks if a route to the src exists and
481 * if it is reachable via ifp (when provided).
483 * The 'verrevpath' option checks that the interface that an IP packet
484 * arrives on is the same interface that traffic destined for the
485 * packet's source address would be routed out of. The 'versrcreach'
486 * option just checks that the source address is reachable via any route
487 * (except default) in the routing table. These two are a measure to block
488 * forged packets. This is also commonly known as "anti-spoofing" or Unicast
489 * Reverse Path Forwarding (Unicast RFP) in Cisco-ese. The name of the knobs
490 * is purposely reminiscent of the Cisco IOS command,
492 * ip verify unicast reverse-path
493 * ip verify unicast source reachable-via any
495 * which implements the same functionality. But note that syntax is
496 * misleading. The check may be performed on all IP packets whether unicast,
497 * multicast, or broadcast.
500 verify_path(struct in_addr src, struct ifnet *ifp)
503 struct sockaddr_in *dst;
505 bzero(&ro, sizeof(ro));
507 dst = (struct sockaddr_in *)&(ro.ro_dst);
508 dst->sin_family = AF_INET;
509 dst->sin_len = sizeof(*dst);
511 rtalloc_ign(&ro, RTF_CLONING);
513 if (ro.ro_rt == NULL)
517 * If ifp is provided, check for equality with rtentry.
518 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
519 * in order to pass packets injected back by if_simloop():
520 * if useloopback == 1 routing entry (via lo0) for our own address
521 * may exist, so we need to handle routing assymetry.
523 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
528 /* if no ifp provided, check if rtentry is not default route */
530 satosin(rt_key(ro.ro_rt))->sin_addr.s_addr == INADDR_ANY) {
535 /* or if this is a blackhole/reject route */
536 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
541 /* found valid route */
548 * ipv6 specific rules here...
551 icmp6type_match (int type, ipfw_insn_u32 *cmd)
553 return (type <= ICMP6_MAXTYPE && (cmd->d[type/32] & (1<<(type%32)) ) );
557 flow6id_match( int curr_flow, ipfw_insn_u32 *cmd )
560 for (i=0; i <= cmd->o.arg1; ++i )
561 if (curr_flow == cmd->d[i] )
566 /* support for IP6_*_ME opcodes */
568 search_ip6_addr_net (struct in6_addr * ip6_addr)
572 struct in6_ifaddr *fdm;
573 struct in6_addr copia;
575 TAILQ_FOREACH(mdc, &ifnet, if_link)
576 TAILQ_FOREACH(mdc2, &mdc->if_addrlist, ifa_list) {
577 if (mdc2->ifa_addr->sa_family == AF_INET6) {
578 fdm = (struct in6_ifaddr *)mdc2;
579 copia = fdm->ia_addr.sin6_addr;
580 /* need for leaving scope_id in the sock_addr */
581 in6_clearscope(&copia);
582 if (IN6_ARE_ADDR_EQUAL(ip6_addr, &copia))
590 verify_path6(struct in6_addr *src, struct ifnet *ifp)
593 struct sockaddr_in6 *dst;
595 bzero(&ro, sizeof(ro));
597 dst = (struct sockaddr_in6 * )&(ro.ro_dst);
598 dst->sin6_family = AF_INET6;
599 dst->sin6_len = sizeof(*dst);
600 dst->sin6_addr = *src;
601 rtalloc_ign((struct route *)&ro, RTF_CLONING);
603 if (ro.ro_rt == NULL)
607 * if ifp is provided, check for equality with rtentry
608 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
609 * to support the case of sending packets to an address of our own.
610 * (where the former interface is the first argument of if_simloop()
611 * (=ifp), the latter is lo0)
613 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
618 /* if no ifp provided, check if rtentry is not default route */
620 IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(ro.ro_rt))->sin6_addr)) {
625 /* or if this is a blackhole/reject route */
626 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
631 /* found valid route */
637 hash_packet6(struct ipfw_flow_id *id)
640 i = (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^
641 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^
642 (id->src_ip6.__u6_addr.__u6_addr32[2]) ^
643 (id->src_ip6.__u6_addr.__u6_addr32[3]) ^
644 (id->dst_port) ^ (id->src_port);
649 is_icmp6_query(int icmp6_type)
651 if ((icmp6_type <= ICMP6_MAXTYPE) &&
652 (icmp6_type == ICMP6_ECHO_REQUEST ||
653 icmp6_type == ICMP6_MEMBERSHIP_QUERY ||
654 icmp6_type == ICMP6_WRUREQUEST ||
655 icmp6_type == ICMP6_FQDN_QUERY ||
656 icmp6_type == ICMP6_NI_QUERY))
663 send_reject6(struct ip_fw_args *args, int code, u_int hlen)
665 if (code == ICMP6_UNREACH_RST && args->f_id.proto == IPPROTO_TCP) {
675 if (args->m->m_len < (hlen+sizeof(struct tcphdr))) {
676 args->m = m_pullup(args->m, hlen+sizeof(struct tcphdr));
681 ip6 = mtod(args->m, struct ip6_hdr *);
682 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
684 if ((tcp->th_flags & TH_RST) != 0) {
691 ti.th.th_seq = ntohl(ti.th.th_seq);
692 ti.th.th_ack = ntohl(ti.th.th_ack);
693 ti.ip6.ip6_nxt = IPPROTO_TCP;
695 if (ti.th.th_flags & TH_ACK) {
701 if (((args->m)->m_flags & M_PKTHDR) != 0) {
702 ack += (args->m)->m_pkthdr.len - hlen
703 - (ti.th.th_off << 2);
704 } else if (ip6->ip6_plen) {
705 ack += ntohs(ip6->ip6_plen) + sizeof(*ip6)
706 - hlen - (ti.th.th_off << 2);
711 if (tcp->th_flags & TH_SYN)
714 flags = TH_RST|TH_ACK;
716 bcopy(&ti, ip6, sizeof(ti));
717 tcp_respond(NULL, ip6, (struct tcphdr *)(ip6 + 1),
718 args->m, ack, seq, flags);
720 } else if (code != ICMP6_UNREACH_RST) { /* Send an ICMPv6 unreach. */
721 icmp6_error(args->m, ICMP6_DST_UNREACH, code, 0);
731 static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */
733 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
734 #define SNP(buf) buf, sizeof(buf)
737 * We enter here when we have a rule with O_LOG.
738 * XXX this function alone takes about 2Kbytes of code!
741 ipfw_log(struct ip_fw *f, u_int hlen, struct ip_fw_args *args,
742 struct mbuf *m, struct ifnet *oif, u_short offset, uint32_t tablearg)
744 struct ether_header *eh = args->eh;
746 int limit_reached = 0;
747 char action2[40], proto[128], fragment[32];
752 if (f == NULL) { /* bogus pkt */
753 if (verbose_limit != 0 && norule_counter >= verbose_limit)
756 if (norule_counter == verbose_limit)
757 limit_reached = verbose_limit;
759 } else { /* O_LOG is the first action, find the real one */
760 ipfw_insn *cmd = ACTION_PTR(f);
761 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
763 if (l->max_log != 0 && l->log_left == 0)
766 if (l->log_left == 0)
767 limit_reached = l->max_log;
768 cmd += F_LEN(cmd); /* point to first action */
769 if (cmd->opcode == O_ALTQ) {
770 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
772 snprintf(SNPARGS(action2, 0), "Altq %d",
776 if (cmd->opcode == O_PROB)
779 if (cmd->opcode == O_TAG)
783 switch (cmd->opcode) {
789 if (cmd->arg1==ICMP_REJECT_RST)
791 else if (cmd->arg1==ICMP_UNREACH_HOST)
794 snprintf(SNPARGS(action2, 0), "Unreach %d",
799 if (cmd->arg1==ICMP6_UNREACH_RST)
802 snprintf(SNPARGS(action2, 0), "Unreach %d",
813 snprintf(SNPARGS(action2, 0), "Divert %d",
817 snprintf(SNPARGS(action2, 0), "Tee %d",
821 snprintf(SNPARGS(action2, 0), "SkipTo %d",
825 snprintf(SNPARGS(action2, 0), "Pipe %d",
829 snprintf(SNPARGS(action2, 0), "Queue %d",
833 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
835 struct in_addr dummyaddr;
836 if (sa->sa.sin_addr.s_addr == INADDR_ANY)
837 dummyaddr.s_addr = htonl(tablearg);
839 dummyaddr.s_addr = sa->sa.sin_addr.s_addr;
841 len = snprintf(SNPARGS(action2, 0), "Forward to %s",
842 inet_ntoa(dummyaddr));
845 snprintf(SNPARGS(action2, len), ":%d",
850 snprintf(SNPARGS(action2, 0), "Netgraph %d",
854 snprintf(SNPARGS(action2, 0), "Ngtee %d",
863 if (hlen == 0) { /* non-ip */
864 snprintf(SNPARGS(proto, 0), "MAC");
868 char src[48], dst[48];
869 struct icmphdr *icmp;
872 /* Initialize to make compiler happy. */
873 struct ip *ip = NULL;
875 struct ip6_hdr *ip6 = NULL;
876 struct icmp6_hdr *icmp6;
881 if (args->f_id.addr_type == 6) {
882 snprintf(src, sizeof(src), "[%s]",
883 ip6_sprintf(&args->f_id.src_ip6));
884 snprintf(dst, sizeof(dst), "[%s]",
885 ip6_sprintf(&args->f_id.dst_ip6));
887 ip6 = (struct ip6_hdr *)mtod(m, struct ip6_hdr *);
888 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
889 udp = (struct udphdr *)(mtod(args->m, char *) + hlen);
893 ip = mtod(m, struct ip *);
894 tcp = L3HDR(struct tcphdr, ip);
895 udp = L3HDR(struct udphdr, ip);
897 inet_ntoa_r(ip->ip_src, src);
898 inet_ntoa_r(ip->ip_dst, dst);
901 switch (args->f_id.proto) {
903 len = snprintf(SNPARGS(proto, 0), "TCP %s", src);
905 snprintf(SNPARGS(proto, len), ":%d %s:%d",
906 ntohs(tcp->th_sport),
908 ntohs(tcp->th_dport));
910 snprintf(SNPARGS(proto, len), " %s", dst);
914 len = snprintf(SNPARGS(proto, 0), "UDP %s", src);
916 snprintf(SNPARGS(proto, len), ":%d %s:%d",
917 ntohs(udp->uh_sport),
919 ntohs(udp->uh_dport));
921 snprintf(SNPARGS(proto, len), " %s", dst);
925 icmp = L3HDR(struct icmphdr, ip);
927 len = snprintf(SNPARGS(proto, 0),
929 icmp->icmp_type, icmp->icmp_code);
931 len = snprintf(SNPARGS(proto, 0), "ICMP ");
932 len += snprintf(SNPARGS(proto, len), "%s", src);
933 snprintf(SNPARGS(proto, len), " %s", dst);
937 icmp6 = (struct icmp6_hdr *)(mtod(args->m, char *) + hlen);
939 len = snprintf(SNPARGS(proto, 0),
941 icmp6->icmp6_type, icmp6->icmp6_code);
943 len = snprintf(SNPARGS(proto, 0), "ICMPv6 ");
944 len += snprintf(SNPARGS(proto, len), "%s", src);
945 snprintf(SNPARGS(proto, len), " %s", dst);
949 len = snprintf(SNPARGS(proto, 0), "P:%d %s",
950 args->f_id.proto, src);
951 snprintf(SNPARGS(proto, len), " %s", dst);
956 if (args->f_id.addr_type == 6) {
957 if (offset & (IP6F_OFF_MASK | IP6F_MORE_FRAG))
958 snprintf(SNPARGS(fragment, 0),
959 " (frag %08x:%d@%d%s)",
961 ntohs(ip6->ip6_plen) - hlen,
962 ntohs(offset & IP6F_OFF_MASK) << 3,
963 (offset & IP6F_MORE_FRAG) ? "+" : "");
968 if (eh != NULL) { /* layer 2 packets are as on the wire */
969 ip_off = ntohs(ip->ip_off);
970 ip_len = ntohs(ip->ip_len);
975 if (ip_off & (IP_MF | IP_OFFMASK))
976 snprintf(SNPARGS(fragment, 0),
977 " (frag %d:%d@%d%s)",
978 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
980 (ip_off & IP_MF) ? "+" : "");
983 if (oif || m->m_pkthdr.rcvif)
984 log(LOG_SECURITY | LOG_INFO,
985 "ipfw: %d %s %s %s via %s%s\n",
987 action, proto, oif ? "out" : "in",
988 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
991 log(LOG_SECURITY | LOG_INFO,
992 "ipfw: %d %s %s [no if info]%s\n",
994 action, proto, fragment);
996 log(LOG_SECURITY | LOG_NOTICE,
997 "ipfw: limit %d reached on entry %d\n",
998 limit_reached, f ? f->rulenum : -1);
1002 * IMPORTANT: the hash function for dynamic rules must be commutative
1003 * in source and destination (ip,port), because rules are bidirectional
1004 * and we want to find both in the same bucket.
1007 hash_packet(struct ipfw_flow_id *id)
1012 if (IS_IP6_FLOW_ID(id))
1013 i = hash_packet6(id);
1016 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
1017 i &= (curr_dyn_buckets - 1);
1022 * unlink a dynamic rule from a chain. prev is a pointer to
1023 * the previous one, q is a pointer to the rule to delete,
1024 * head is a pointer to the head of the queue.
1025 * Modifies q and potentially also head.
1027 #define UNLINK_DYN_RULE(prev, head, q) { \
1028 ipfw_dyn_rule *old_q = q; \
1030 /* remove a refcount to the parent */ \
1031 if (q->dyn_type == O_LIMIT) \
1032 q->parent->count--; \
1033 DEB(printf("ipfw: unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",\
1034 (q->id.src_ip), (q->id.src_port), \
1035 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \
1037 prev->next = q = q->next; \
1039 head = q = q->next; \
1041 uma_zfree(ipfw_dyn_rule_zone, old_q); }
1043 #define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
1046 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
1048 * If keep_me == NULL, rules are deleted even if not expired,
1049 * otherwise only expired rules are removed.
1051 * The value of the second parameter is also used to point to identify
1052 * a rule we absolutely do not want to remove (e.g. because we are
1053 * holding a reference to it -- this is the case with O_LIMIT_PARENT
1054 * rules). The pointer is only used for comparison, so any non-null
1058 remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
1060 static u_int32_t last_remove = 0;
1062 #define FORCE (keep_me == NULL)
1064 ipfw_dyn_rule *prev, *q;
1065 int i, pass = 0, max_pass = 0;
1067 IPFW_DYN_LOCK_ASSERT();
1069 if (ipfw_dyn_v == NULL || dyn_count == 0)
1071 /* do not expire more than once per second, it is useless */
1072 if (!FORCE && last_remove == time_uptime)
1074 last_remove = time_uptime;
1077 * because O_LIMIT refer to parent rules, during the first pass only
1078 * remove child and mark any pending LIMIT_PARENT, and remove
1079 * them in a second pass.
1082 for (i = 0 ; i < curr_dyn_buckets ; i++) {
1083 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
1085 * Logic can become complex here, so we split tests.
1089 if (rule != NULL && rule != q->rule)
1090 goto next; /* not the one we are looking for */
1091 if (q->dyn_type == O_LIMIT_PARENT) {
1093 * handle parent in the second pass,
1094 * record we need one.
1099 if (FORCE && q->count != 0 ) {
1100 /* XXX should not happen! */
1101 printf("ipfw: OUCH! cannot remove rule,"
1102 " count %d\n", q->count);
1106 !TIME_LEQ( q->expire, time_uptime ))
1109 if (q->dyn_type != O_LIMIT_PARENT || !q->count) {
1110 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1118 if (pass++ < max_pass)
1124 * lookup a dynamic rule.
1126 static ipfw_dyn_rule *
1127 lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction,
1131 * stateful ipfw extensions.
1132 * Lookup into dynamic session queue
1134 #define MATCH_REVERSE 0
1135 #define MATCH_FORWARD 1
1136 #define MATCH_NONE 2
1137 #define MATCH_UNKNOWN 3
1138 int i, dir = MATCH_NONE;
1139 ipfw_dyn_rule *prev, *q=NULL;
1141 IPFW_DYN_LOCK_ASSERT();
1143 if (ipfw_dyn_v == NULL)
1144 goto done; /* not found */
1145 i = hash_packet( pkt );
1146 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
1147 if (q->dyn_type == O_LIMIT_PARENT && q->count)
1149 if (TIME_LEQ( q->expire, time_uptime)) { /* expire entry */
1150 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1153 if (pkt->proto == q->id.proto &&
1154 q->dyn_type != O_LIMIT_PARENT) {
1155 if (IS_IP6_FLOW_ID(pkt)) {
1156 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1157 &(q->id.src_ip6)) &&
1158 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1159 &(q->id.dst_ip6)) &&
1160 pkt->src_port == q->id.src_port &&
1161 pkt->dst_port == q->id.dst_port ) {
1162 dir = MATCH_FORWARD;
1165 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1166 &(q->id.dst_ip6)) &&
1167 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1168 &(q->id.src_ip6)) &&
1169 pkt->src_port == q->id.dst_port &&
1170 pkt->dst_port == q->id.src_port ) {
1171 dir = MATCH_REVERSE;
1175 if (pkt->src_ip == q->id.src_ip &&
1176 pkt->dst_ip == q->id.dst_ip &&
1177 pkt->src_port == q->id.src_port &&
1178 pkt->dst_port == q->id.dst_port ) {
1179 dir = MATCH_FORWARD;
1182 if (pkt->src_ip == q->id.dst_ip &&
1183 pkt->dst_ip == q->id.src_ip &&
1184 pkt->src_port == q->id.dst_port &&
1185 pkt->dst_port == q->id.src_port ) {
1186 dir = MATCH_REVERSE;
1196 goto done; /* q = NULL, not found */
1198 if ( prev != NULL) { /* found and not in front */
1199 prev->next = q->next;
1200 q->next = ipfw_dyn_v[i];
1203 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
1204 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
1206 #define BOTH_SYN (TH_SYN | (TH_SYN << 8))
1207 #define BOTH_FIN (TH_FIN | (TH_FIN << 8))
1208 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
1210 case TH_SYN: /* opening */
1211 q->expire = time_uptime + dyn_syn_lifetime;
1214 case BOTH_SYN: /* move to established */
1215 case BOTH_SYN | TH_FIN : /* one side tries to close */
1216 case BOTH_SYN | (TH_FIN << 8) :
1218 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
1219 u_int32_t ack = ntohl(tcp->th_ack);
1220 if (dir == MATCH_FORWARD) {
1221 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
1223 else { /* ignore out-of-sequence */
1227 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
1229 else { /* ignore out-of-sequence */
1234 q->expire = time_uptime + dyn_ack_lifetime;
1237 case BOTH_SYN | BOTH_FIN: /* both sides closed */
1238 if (dyn_fin_lifetime >= dyn_keepalive_period)
1239 dyn_fin_lifetime = dyn_keepalive_period - 1;
1240 q->expire = time_uptime + dyn_fin_lifetime;
1246 * reset or some invalid combination, but can also
1247 * occur if we use keep-state the wrong way.
1249 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
1250 printf("invalid state: 0x%x\n", q->state);
1252 if (dyn_rst_lifetime >= dyn_keepalive_period)
1253 dyn_rst_lifetime = dyn_keepalive_period - 1;
1254 q->expire = time_uptime + dyn_rst_lifetime;
1257 } else if (pkt->proto == IPPROTO_UDP) {
1258 q->expire = time_uptime + dyn_udp_lifetime;
1260 /* other protocols */
1261 q->expire = time_uptime + dyn_short_lifetime;
1264 if (match_direction)
1265 *match_direction = dir;
1269 static ipfw_dyn_rule *
1270 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
1276 q = lookup_dyn_rule_locked(pkt, match_direction, tcp);
1279 /* NB: return table locked when q is not NULL */
1284 realloc_dynamic_table(void)
1286 IPFW_DYN_LOCK_ASSERT();
1289 * Try reallocation, make sure we have a power of 2 and do
1290 * not allow more than 64k entries. In case of overflow,
1294 if (dyn_buckets > 65536)
1296 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
1297 dyn_buckets = curr_dyn_buckets; /* reset */
1300 curr_dyn_buckets = dyn_buckets;
1301 if (ipfw_dyn_v != NULL)
1302 free(ipfw_dyn_v, M_IPFW);
1304 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
1305 M_IPFW, M_NOWAIT | M_ZERO);
1306 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
1308 curr_dyn_buckets /= 2;
1313 * Install state of type 'type' for a dynamic session.
1314 * The hash table contains two type of rules:
1315 * - regular rules (O_KEEP_STATE)
1316 * - rules for sessions with limited number of sess per user
1317 * (O_LIMIT). When they are created, the parent is
1318 * increased by 1, and decreased on delete. In this case,
1319 * the third parameter is the parent rule and not the chain.
1320 * - "parent" rules for the above (O_LIMIT_PARENT).
1322 static ipfw_dyn_rule *
1323 add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
1328 IPFW_DYN_LOCK_ASSERT();
1330 if (ipfw_dyn_v == NULL ||
1331 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1332 realloc_dynamic_table();
1333 if (ipfw_dyn_v == NULL)
1334 return NULL; /* failed ! */
1336 i = hash_packet(id);
1338 r = uma_zalloc(ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO);
1340 printf ("ipfw: sorry cannot allocate state\n");
1344 /* increase refcount on parent, and set pointer */
1345 if (dyn_type == O_LIMIT) {
1346 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1347 if ( parent->dyn_type != O_LIMIT_PARENT)
1348 panic("invalid parent");
1351 rule = parent->rule;
1355 r->expire = time_uptime + dyn_syn_lifetime;
1357 r->dyn_type = dyn_type;
1358 r->pcnt = r->bcnt = 0;
1362 r->next = ipfw_dyn_v[i];
1365 DEB(printf("ipfw: add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1367 (r->id.src_ip), (r->id.src_port),
1368 (r->id.dst_ip), (r->id.dst_port),
1374 * lookup dynamic parent rule using pkt and rule as search keys.
1375 * If the lookup fails, then install one.
1377 static ipfw_dyn_rule *
1378 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1383 IPFW_DYN_LOCK_ASSERT();
1386 int is_v6 = IS_IP6_FLOW_ID(pkt);
1387 i = hash_packet( pkt );
1388 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
1389 if (q->dyn_type == O_LIMIT_PARENT &&
1391 pkt->proto == q->id.proto &&
1392 pkt->src_port == q->id.src_port &&
1393 pkt->dst_port == q->id.dst_port &&
1396 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1397 &(q->id.src_ip6)) &&
1398 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1399 &(q->id.dst_ip6))) ||
1401 pkt->src_ip == q->id.src_ip &&
1402 pkt->dst_ip == q->id.dst_ip)
1405 q->expire = time_uptime + dyn_short_lifetime;
1406 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);)
1410 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1414 * Install dynamic state for rule type cmd->o.opcode
1416 * Returns 1 (failure) if state is not installed because of errors or because
1417 * session limitations are enforced.
1420 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1421 struct ip_fw_args *args, uint32_t tablearg)
1423 static int last_log;
1428 printf("ipfw: %s: type %d 0x%08x %u -> 0x%08x %u\n",
1429 __func__, cmd->o.opcode,
1430 (args->f_id.src_ip), (args->f_id.src_port),
1431 (args->f_id.dst_ip), (args->f_id.dst_port));
1436 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
1438 if (q != NULL) { /* should never occur */
1439 if (last_log != time_uptime) {
1440 last_log = time_uptime;
1441 printf("ipfw: %s: entry already present, done\n",
1448 if (dyn_count >= dyn_max)
1449 /* Run out of slots, try to remove any expired rule. */
1450 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1452 if (dyn_count >= dyn_max) {
1453 if (last_log != time_uptime) {
1454 last_log = time_uptime;
1455 printf("ipfw: %s: Too many dynamic rules\n", __func__);
1458 return (1); /* cannot install, notify caller */
1461 switch (cmd->o.opcode) {
1462 case O_KEEP_STATE: /* bidir rule */
1463 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1466 case O_LIMIT: { /* limit number of sessions */
1467 struct ipfw_flow_id id;
1468 ipfw_dyn_rule *parent;
1469 uint32_t conn_limit;
1470 uint16_t limit_mask = cmd->limit_mask;
1472 conn_limit = (cmd->conn_limit == IP_FW_TABLEARG) ?
1473 tablearg : cmd->conn_limit;
1476 if (cmd->conn_limit == IP_FW_TABLEARG)
1477 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u "
1478 "(tablearg)\n", __func__, conn_limit);
1480 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u\n",
1481 __func__, conn_limit);
1484 id.dst_ip = id.src_ip = id.dst_port = id.src_port = 0;
1485 id.proto = args->f_id.proto;
1486 id.addr_type = args->f_id.addr_type;
1488 if (IS_IP6_FLOW_ID (&(args->f_id))) {
1489 if (limit_mask & DYN_SRC_ADDR)
1490 id.src_ip6 = args->f_id.src_ip6;
1491 if (limit_mask & DYN_DST_ADDR)
1492 id.dst_ip6 = args->f_id.dst_ip6;
1494 if (limit_mask & DYN_SRC_ADDR)
1495 id.src_ip = args->f_id.src_ip;
1496 if (limit_mask & DYN_DST_ADDR)
1497 id.dst_ip = args->f_id.dst_ip;
1499 if (limit_mask & DYN_SRC_PORT)
1500 id.src_port = args->f_id.src_port;
1501 if (limit_mask & DYN_DST_PORT)
1502 id.dst_port = args->f_id.dst_port;
1503 if ((parent = lookup_dyn_parent(&id, rule)) == NULL) {
1504 printf("ipfw: %s: add parent failed\n", __func__);
1509 if (parent->count >= conn_limit) {
1510 /* See if we can remove some expired rule. */
1511 remove_dyn_rule(rule, parent);
1512 if (parent->count >= conn_limit) {
1513 if (fw_verbose && last_log != time_uptime) {
1514 last_log = time_uptime;
1515 log(LOG_SECURITY | LOG_DEBUG,
1516 "drop session, too many entries\n");
1522 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1526 printf("ipfw: %s: unknown dynamic rule type %u\n",
1527 __func__, cmd->o.opcode);
1532 /* XXX just set lifetime */
1533 lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
1540 * Generate a TCP packet, containing either a RST or a keepalive.
1541 * When flags & TH_RST, we are sending a RST packet, because of a
1542 * "reset" action matched the packet.
1543 * Otherwise we are sending a keepalive, and flags & TH_
1545 static struct mbuf *
1546 send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags)
1552 MGETHDR(m, M_DONTWAIT, MT_DATA);
1555 m->m_pkthdr.rcvif = (struct ifnet *)0;
1556 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1557 m->m_data += max_linkhdr;
1559 ip = mtod(m, struct ip *);
1560 bzero(ip, m->m_len);
1561 tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1562 ip->ip_p = IPPROTO_TCP;
1565 * Assume we are sending a RST (or a keepalive in the reverse
1566 * direction), swap src and destination addresses and ports.
1568 ip->ip_src.s_addr = htonl(id->dst_ip);
1569 ip->ip_dst.s_addr = htonl(id->src_ip);
1570 tcp->th_sport = htons(id->dst_port);
1571 tcp->th_dport = htons(id->src_port);
1572 if (flags & TH_RST) { /* we are sending a RST */
1573 if (flags & TH_ACK) {
1574 tcp->th_seq = htonl(ack);
1575 tcp->th_ack = htonl(0);
1576 tcp->th_flags = TH_RST;
1580 tcp->th_seq = htonl(0);
1581 tcp->th_ack = htonl(seq);
1582 tcp->th_flags = TH_RST | TH_ACK;
1586 * We are sending a keepalive. flags & TH_SYN determines
1587 * the direction, forward if set, reverse if clear.
1588 * NOTE: seq and ack are always assumed to be correct
1589 * as set by the caller. This may be confusing...
1591 if (flags & TH_SYN) {
1593 * we have to rewrite the correct addresses!
1595 ip->ip_dst.s_addr = htonl(id->dst_ip);
1596 ip->ip_src.s_addr = htonl(id->src_ip);
1597 tcp->th_dport = htons(id->dst_port);
1598 tcp->th_sport = htons(id->src_port);
1600 tcp->th_seq = htonl(seq);
1601 tcp->th_ack = htonl(ack);
1602 tcp->th_flags = TH_ACK;
1605 * set ip_len to the payload size so we can compute
1606 * the tcp checksum on the pseudoheader
1607 * XXX check this, could save a couple of words ?
1609 ip->ip_len = htons(sizeof(struct tcphdr));
1610 tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1612 * now fill fields left out earlier
1614 ip->ip_ttl = ip_defttl;
1615 ip->ip_len = m->m_pkthdr.len;
1616 m->m_flags |= M_SKIP_FIREWALL;
1621 * sends a reject message, consuming the mbuf passed as an argument.
1624 send_reject(struct ip_fw_args *args, int code, int ip_len)
1627 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1628 /* We need the IP header in host order for icmp_error(). */
1629 if (args->eh != NULL) {
1630 struct ip *ip = mtod(args->m, struct ip *);
1631 ip->ip_len = ntohs(ip->ip_len);
1632 ip->ip_off = ntohs(ip->ip_off);
1634 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1635 } else if (args->f_id.proto == IPPROTO_TCP) {
1636 struct tcphdr *const tcp =
1637 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1638 if ( (tcp->th_flags & TH_RST) == 0) {
1640 m = send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1642 tcp->th_flags | TH_RST);
1644 ip_output(m, NULL, NULL, 0, NULL, NULL);
1654 * Given an ip_fw *, lookup_next_rule will return a pointer
1655 * to the next rule, which can be either the jump
1656 * target (for skipto instructions) or the next one in the list (in
1657 * all other cases including a missing jump target).
1658 * The result is also written in the "next_rule" field of the rule.
1659 * Backward jumps are not allowed, so start looking from the next
1662 * This never returns NULL -- in case we do not have an exact match,
1663 * the next rule is returned. When the ruleset is changed,
1664 * pointers are flushed so we are always correct.
1667 static struct ip_fw *
1668 lookup_next_rule(struct ip_fw *me)
1670 struct ip_fw *rule = NULL;
1673 /* look for action, in case it is a skipto */
1674 cmd = ACTION_PTR(me);
1675 if (cmd->opcode == O_LOG)
1677 if (cmd->opcode == O_ALTQ)
1679 if (cmd->opcode == O_TAG)
1681 if ( cmd->opcode == O_SKIPTO )
1682 for (rule = me->next; rule ; rule = rule->next)
1683 if (rule->rulenum >= cmd->arg1)
1685 if (rule == NULL) /* failure or not a skipto */
1687 me->next_rule = rule;
1692 add_table_entry(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1693 uint8_t mlen, uint32_t value)
1695 struct radix_node_head *rnh;
1696 struct table_entry *ent;
1698 if (tbl >= IPFW_TABLES_MAX)
1700 rnh = ch->tables[tbl];
1701 ent = malloc(sizeof(*ent), M_IPFW_TBL, M_NOWAIT | M_ZERO);
1705 ent->addr.sin_len = ent->mask.sin_len = 8;
1706 ent->mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1707 ent->addr.sin_addr.s_addr = addr & ent->mask.sin_addr.s_addr;
1708 IPFW_WLOCK(&layer3_chain);
1709 if (rnh->rnh_addaddr(&ent->addr, &ent->mask, rnh, (void *)ent) ==
1711 IPFW_WUNLOCK(&layer3_chain);
1712 free(ent, M_IPFW_TBL);
1715 IPFW_WUNLOCK(&layer3_chain);
1720 del_table_entry(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1723 struct radix_node_head *rnh;
1724 struct table_entry *ent;
1725 struct sockaddr_in sa, mask;
1727 if (tbl >= IPFW_TABLES_MAX)
1729 rnh = ch->tables[tbl];
1730 sa.sin_len = mask.sin_len = 8;
1731 mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1732 sa.sin_addr.s_addr = addr & mask.sin_addr.s_addr;
1734 ent = (struct table_entry *)rnh->rnh_deladdr(&sa, &mask, rnh);
1740 free(ent, M_IPFW_TBL);
1745 flush_table_entry(struct radix_node *rn, void *arg)
1747 struct radix_node_head * const rnh = arg;
1748 struct table_entry *ent;
1750 ent = (struct table_entry *)
1751 rnh->rnh_deladdr(rn->rn_key, rn->rn_mask, rnh);
1753 free(ent, M_IPFW_TBL);
1758 flush_table(struct ip_fw_chain *ch, uint16_t tbl)
1760 struct radix_node_head *rnh;
1762 IPFW_WLOCK_ASSERT(ch);
1764 if (tbl >= IPFW_TABLES_MAX)
1766 rnh = ch->tables[tbl];
1767 KASSERT(rnh != NULL, ("NULL IPFW table"));
1768 rnh->rnh_walktree(rnh, flush_table_entry, rnh);
1773 flush_tables(struct ip_fw_chain *ch)
1777 IPFW_WLOCK_ASSERT(ch);
1779 for (tbl = 0; tbl < IPFW_TABLES_MAX; tbl++)
1780 flush_table(ch, tbl);
1784 init_tables(struct ip_fw_chain *ch)
1789 for (i = 0; i < IPFW_TABLES_MAX; i++) {
1790 if (!rn_inithead((void **)&ch->tables[i], 32)) {
1791 for (j = 0; j < i; j++) {
1792 (void) flush_table(ch, j);
1801 lookup_table(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1804 struct radix_node_head *rnh;
1805 struct table_entry *ent;
1806 struct sockaddr_in sa;
1808 if (tbl >= IPFW_TABLES_MAX)
1810 rnh = ch->tables[tbl];
1812 sa.sin_addr.s_addr = addr;
1813 ent = (struct table_entry *)(rnh->rnh_lookup(&sa, NULL, rnh));
1822 count_table_entry(struct radix_node *rn, void *arg)
1824 u_int32_t * const cnt = arg;
1831 count_table(struct ip_fw_chain *ch, uint32_t tbl, uint32_t *cnt)
1833 struct radix_node_head *rnh;
1835 if (tbl >= IPFW_TABLES_MAX)
1837 rnh = ch->tables[tbl];
1839 rnh->rnh_walktree(rnh, count_table_entry, cnt);
1844 dump_table_entry(struct radix_node *rn, void *arg)
1846 struct table_entry * const n = (struct table_entry *)rn;
1847 ipfw_table * const tbl = arg;
1848 ipfw_table_entry *ent;
1850 if (tbl->cnt == tbl->size)
1852 ent = &tbl->ent[tbl->cnt];
1853 ent->tbl = tbl->tbl;
1854 if (in_nullhost(n->mask.sin_addr))
1857 ent->masklen = 33 - ffs(ntohl(n->mask.sin_addr.s_addr));
1858 ent->addr = n->addr.sin_addr.s_addr;
1859 ent->value = n->value;
1865 dump_table(struct ip_fw_chain *ch, ipfw_table *tbl)
1867 struct radix_node_head *rnh;
1869 if (tbl->tbl >= IPFW_TABLES_MAX)
1871 rnh = ch->tables[tbl->tbl];
1873 rnh->rnh_walktree(rnh, dump_table_entry, tbl);
1878 fill_ugid_cache(struct inpcb *inp, struct ip_fw_ugid *ugp)
1882 if (inp->inp_socket != NULL) {
1883 cr = inp->inp_socket->so_cred;
1884 ugp->fw_prid = jailed(cr) ?
1885 cr->cr_prison->pr_id : -1;
1886 ugp->fw_uid = cr->cr_uid;
1887 ugp->fw_ngroups = cr->cr_ngroups;
1888 bcopy(cr->cr_groups, ugp->fw_groups,
1889 sizeof(ugp->fw_groups));
1894 check_uidgid(ipfw_insn_u32 *insn,
1895 int proto, struct ifnet *oif,
1896 struct in_addr dst_ip, u_int16_t dst_port,
1897 struct in_addr src_ip, u_int16_t src_port,
1898 struct ip_fw_ugid *ugp, int *lookup, struct inpcb *inp)
1900 struct inpcbinfo *pi;
1907 * Check to see if the UDP or TCP stack supplied us with
1908 * the PCB. If so, rather then holding a lock and looking
1909 * up the PCB, we can use the one that was supplied.
1911 if (inp && *lookup == 0) {
1912 INP_LOCK_ASSERT(inp);
1913 if (inp->inp_socket != NULL) {
1914 fill_ugid_cache(inp, ugp);
1919 * If we have already been here and the packet has no
1920 * PCB entry associated with it, then we can safely
1921 * assume that this is a no match.
1925 if (proto == IPPROTO_TCP) {
1928 } else if (proto == IPPROTO_UDP) {
1929 wildcard = INPLOOKUP_WILDCARD;
1937 in_pcblookup_hash(pi,
1938 dst_ip, htons(dst_port),
1939 src_ip, htons(src_port),
1941 in_pcblookup_hash(pi,
1942 src_ip, htons(src_port),
1943 dst_ip, htons(dst_port),
1947 if (pcb->inp_socket != NULL) {
1948 fill_ugid_cache(pcb, ugp);
1953 INP_INFO_RUNLOCK(pi);
1956 * If the lookup did not yield any results, there
1957 * is no sense in coming back and trying again. So
1958 * we can set lookup to -1 and ensure that we wont
1959 * bother the pcb system again.
1965 if (insn->o.opcode == O_UID)
1966 match = (ugp->fw_uid == (uid_t)insn->d[0]);
1967 else if (insn->o.opcode == O_GID) {
1968 for (gp = ugp->fw_groups;
1969 gp < &ugp->fw_groups[ugp->fw_ngroups]; gp++)
1970 if (*gp == (gid_t)insn->d[0]) {
1974 } else if (insn->o.opcode == O_JAIL)
1975 match = (ugp->fw_prid == (int)insn->d[0]);
1980 * The main check routine for the firewall.
1982 * All arguments are in args so we can modify them and return them
1983 * back to the caller.
1987 * args->m (in/out) The packet; we set to NULL when/if we nuke it.
1988 * Starts with the IP header.
1989 * args->eh (in) Mac header if present, or NULL for layer3 packet.
1990 * args->oif Outgoing interface, or NULL if packet is incoming.
1991 * The incoming interface is in the mbuf. (in)
1992 * args->divert_rule (in/out)
1993 * Skip up to the first rule past this rule number;
1994 * upon return, non-zero port number for divert or tee.
1996 * args->rule Pointer to the last matching rule (in/out)
1997 * args->next_hop Socket we are forwarding to (out).
1998 * args->f_id Addresses grabbed from the packet (out)
1999 * args->cookie a cookie depending on rule action
2003 * IP_FW_PASS the packet must be accepted
2004 * IP_FW_DENY the packet must be dropped
2005 * IP_FW_DIVERT divert packet, port in m_tag
2006 * IP_FW_TEE tee packet, port in m_tag
2007 * IP_FW_DUMMYNET to dummynet, pipe in args->cookie
2008 * IP_FW_NETGRAPH into netgraph, cookie args->cookie
2013 ipfw_chk(struct ip_fw_args *args)
2016 * Local variables hold state during the processing of a packet.
2018 * IMPORTANT NOTE: to speed up the processing of rules, there
2019 * are some assumption on the values of the variables, which
2020 * are documented here. Should you change them, please check
2021 * the implementation of the various instructions to make sure
2022 * that they still work.
2024 * args->eh The MAC header. It is non-null for a layer2
2025 * packet, it is NULL for a layer-3 packet.
2027 * m | args->m Pointer to the mbuf, as received from the caller.
2028 * It may change if ipfw_chk() does an m_pullup, or if it
2029 * consumes the packet because it calls send_reject().
2030 * XXX This has to change, so that ipfw_chk() never modifies
2031 * or consumes the buffer.
2032 * ip is simply an alias of the value of m, and it is kept
2033 * in sync with it (the packet is supposed to start with
2036 struct mbuf *m = args->m;
2037 struct ip *ip = mtod(m, struct ip *);
2040 * For rules which contain uid/gid or jail constraints, cache
2041 * a copy of the users credentials after the pcb lookup has been
2042 * executed. This will speed up the processing of rules with
2043 * these types of constraints, as well as decrease contention
2044 * on pcb related locks.
2046 struct ip_fw_ugid fw_ugid_cache;
2047 int ugid_lookup = 0;
2050 * divinput_flags If non-zero, set to the IP_FW_DIVERT_*_FLAG
2051 * associated with a packet input on a divert socket. This
2052 * will allow to distinguish traffic and its direction when
2053 * it originates from a divert socket.
2055 u_int divinput_flags = 0;
2058 * oif | args->oif If NULL, ipfw_chk has been called on the
2059 * inbound path (ether_input, ip_input).
2060 * If non-NULL, ipfw_chk has been called on the outbound path
2061 * (ether_output, ip_output).
2063 struct ifnet *oif = args->oif;
2065 struct ip_fw *f = NULL; /* matching rule */
2069 * hlen The length of the IP header.
2071 u_int hlen = 0; /* hlen >0 means we have an IP pkt */
2074 * offset The offset of a fragment. offset != 0 means that
2075 * we have a fragment at this offset of an IPv4 packet.
2076 * offset == 0 means that (if this is an IPv4 packet)
2077 * this is the first or only fragment.
2078 * For IPv6 offset == 0 means there is no Fragment Header.
2079 * If offset != 0 for IPv6 always use correct mask to
2080 * get the correct offset because we add IP6F_MORE_FRAG
2081 * to be able to dectect the first fragment which would
2082 * otherwise have offset = 0.
2087 * Local copies of addresses. They are only valid if we have
2090 * proto The protocol. Set to 0 for non-ip packets,
2091 * or to the protocol read from the packet otherwise.
2092 * proto != 0 means that we have an IPv4 packet.
2094 * src_port, dst_port port numbers, in HOST format. Only
2095 * valid for TCP and UDP packets.
2097 * src_ip, dst_ip ip addresses, in NETWORK format.
2098 * Only valid for IPv4 packets.
2101 u_int16_t src_port = 0, dst_port = 0; /* NOTE: host format */
2102 struct in_addr src_ip, dst_ip; /* NOTE: network format */
2107 * dyn_dir = MATCH_UNKNOWN when rules unchecked,
2108 * MATCH_NONE when checked and not matched (q = NULL),
2109 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
2111 int dyn_dir = MATCH_UNKNOWN;
2112 ipfw_dyn_rule *q = NULL;
2113 struct ip_fw_chain *chain = &layer3_chain;
2117 * We store in ulp a pointer to the upper layer protocol header.
2118 * In the ipv4 case this is easy to determine from the header,
2119 * but for ipv6 we might have some additional headers in the middle.
2120 * ulp is NULL if not found.
2122 void *ulp = NULL; /* upper layer protocol pointer. */
2123 /* XXX ipv6 variables */
2125 u_int16_t ext_hd = 0; /* bits vector for extension header filtering */
2126 /* end of ipv6 variables */
2129 if (m->m_flags & M_SKIP_FIREWALL)
2130 return (IP_FW_PASS); /* accept */
2132 pktlen = m->m_pkthdr.len;
2133 proto = args->f_id.proto = 0; /* mark f_id invalid */
2134 /* XXX 0 is a valid proto: IP/IPv6 Hop-by-Hop Option */
2137 * PULLUP_TO(len, p, T) makes sure that len + sizeof(T) is contiguous,
2138 * then it sets p to point at the offset "len" in the mbuf. WARNING: the
2139 * pointer might become stale after other pullups (but we never use it
2142 #define PULLUP_TO(len, p, T) \
2144 int x = (len) + sizeof(T); \
2145 if ((m)->m_len < x) { \
2146 args->m = m = m_pullup(m, x); \
2148 goto pullup_failed; \
2150 p = (mtod(m, char *) + (len)); \
2153 /* Identify IP packets and fill up variables. */
2154 if (pktlen >= sizeof(struct ip6_hdr) &&
2155 (args->eh == NULL || ntohs(args->eh->ether_type)==ETHERTYPE_IPV6) &&
2156 mtod(m, struct ip *)->ip_v == 6) {
2158 args->f_id.addr_type = 6;
2159 hlen = sizeof(struct ip6_hdr);
2160 proto = mtod(m, struct ip6_hdr *)->ip6_nxt;
2162 /* Search extension headers to find upper layer protocols */
2163 while (ulp == NULL) {
2165 case IPPROTO_ICMPV6:
2166 PULLUP_TO(hlen, ulp, struct icmp6_hdr);
2167 args->f_id.flags = ICMP6(ulp)->icmp6_type;
2171 PULLUP_TO(hlen, ulp, struct tcphdr);
2172 dst_port = TCP(ulp)->th_dport;
2173 src_port = TCP(ulp)->th_sport;
2174 args->f_id.flags = TCP(ulp)->th_flags;
2178 PULLUP_TO(hlen, ulp, struct udphdr);
2179 dst_port = UDP(ulp)->uh_dport;
2180 src_port = UDP(ulp)->uh_sport;
2183 case IPPROTO_HOPOPTS: /* RFC 2460 */
2184 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2185 ext_hd |= EXT_HOPOPTS;
2186 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2187 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2191 case IPPROTO_ROUTING: /* RFC 2460 */
2192 PULLUP_TO(hlen, ulp, struct ip6_rthdr);
2193 switch (((struct ip6_rthdr *)ulp)->ip6r_type) {
2197 printf("IPFW2: IPV6 - Unknown Routing "
2198 "Header type(%d)\n",
2199 ((struct ip6_rthdr *)ulp)->ip6r_type);
2200 if (fw_deny_unknown_exthdrs)
2201 return (IP_FW_DENY);
2204 ext_hd |= EXT_ROUTING;
2205 hlen += (((struct ip6_rthdr *)ulp)->ip6r_len + 1) << 3;
2206 proto = ((struct ip6_rthdr *)ulp)->ip6r_nxt;
2210 case IPPROTO_FRAGMENT: /* RFC 2460 */
2211 PULLUP_TO(hlen, ulp, struct ip6_frag);
2212 ext_hd |= EXT_FRAGMENT;
2213 hlen += sizeof (struct ip6_frag);
2214 proto = ((struct ip6_frag *)ulp)->ip6f_nxt;
2215 offset = ((struct ip6_frag *)ulp)->ip6f_offlg &
2217 /* Add IP6F_MORE_FRAG for offset of first
2218 * fragment to be != 0. */
2219 offset |= ((struct ip6_frag *)ulp)->ip6f_offlg &
2222 printf("IPFW2: IPV6 - Invalid Fragment "
2224 if (fw_deny_unknown_exthdrs)
2225 return (IP_FW_DENY);
2228 args->f_id.frag_id6 =
2229 ntohl(((struct ip6_frag *)ulp)->ip6f_ident);
2233 case IPPROTO_DSTOPTS: /* RFC 2460 */
2234 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2235 ext_hd |= EXT_DSTOPTS;
2236 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2237 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2241 case IPPROTO_AH: /* RFC 2402 */
2242 PULLUP_TO(hlen, ulp, struct ip6_ext);
2244 hlen += (((struct ip6_ext *)ulp)->ip6e_len + 2) << 2;
2245 proto = ((struct ip6_ext *)ulp)->ip6e_nxt;
2249 case IPPROTO_ESP: /* RFC 2406 */
2250 PULLUP_TO(hlen, ulp, uint32_t); /* SPI, Seq# */
2251 /* Anything past Seq# is variable length and
2252 * data past this ext. header is encrypted. */
2256 case IPPROTO_NONE: /* RFC 2460 */
2257 PULLUP_TO(hlen, ulp, struct ip6_ext);
2258 /* Packet ends here. if ip6e_len!=0 octets
2259 * must be ignored. */
2262 case IPPROTO_OSPFIGP:
2263 /* XXX OSPF header check? */
2264 PULLUP_TO(hlen, ulp, struct ip6_ext);
2268 /* XXX PIM header check? */
2269 PULLUP_TO(hlen, ulp, struct pim);
2272 case IPPROTO_IPV6: /* RFC 2893 */
2273 PULLUP_TO(hlen, ulp, struct ip6_hdr);
2276 case IPPROTO_IPV4: /* RFC 2893 */
2277 PULLUP_TO(hlen, ulp, struct ip);
2281 printf("IPFW2: IPV6 - Unknown Extension "
2282 "Header(%d), ext_hd=%x\n", proto, ext_hd);
2283 if (fw_deny_unknown_exthdrs)
2284 return (IP_FW_DENY);
2285 PULLUP_TO(hlen, ulp, struct ip6_ext);
2289 args->f_id.src_ip6 = mtod(m,struct ip6_hdr *)->ip6_src;
2290 args->f_id.dst_ip6 = mtod(m,struct ip6_hdr *)->ip6_dst;
2291 args->f_id.src_ip = 0;
2292 args->f_id.dst_ip = 0;
2293 args->f_id.flow_id6 = ntohl(mtod(m, struct ip6_hdr *)->ip6_flow);
2294 } else if (pktlen >= sizeof(struct ip) &&
2295 (args->eh == NULL || ntohs(args->eh->ether_type) == ETHERTYPE_IP) &&
2296 mtod(m, struct ip *)->ip_v == 4) {
2298 ip = mtod(m, struct ip *);
2299 hlen = ip->ip_hl << 2;
2300 args->f_id.addr_type = 4;
2303 * Collect parameters into local variables for faster matching.
2306 src_ip = ip->ip_src;
2307 dst_ip = ip->ip_dst;
2308 if (args->eh != NULL) { /* layer 2 packets are as on the wire */
2309 offset = ntohs(ip->ip_off) & IP_OFFMASK;
2310 ip_len = ntohs(ip->ip_len);
2312 offset = ip->ip_off & IP_OFFMASK;
2313 ip_len = ip->ip_len;
2315 pktlen = ip_len < pktlen ? ip_len : pktlen;
2320 PULLUP_TO(hlen, ulp, struct tcphdr);
2321 dst_port = TCP(ulp)->th_dport;
2322 src_port = TCP(ulp)->th_sport;
2323 args->f_id.flags = TCP(ulp)->th_flags;
2327 PULLUP_TO(hlen, ulp, struct udphdr);
2328 dst_port = UDP(ulp)->uh_dport;
2329 src_port = UDP(ulp)->uh_sport;
2333 PULLUP_TO(hlen, ulp, struct icmphdr);
2334 args->f_id.flags = ICMP(ulp)->icmp_type;
2342 args->f_id.src_ip = ntohl(src_ip.s_addr);
2343 args->f_id.dst_ip = ntohl(dst_ip.s_addr);
2346 if (proto) { /* we may have port numbers, store them */
2347 args->f_id.proto = proto;
2348 args->f_id.src_port = src_port = ntohs(src_port);
2349 args->f_id.dst_port = dst_port = ntohs(dst_port);
2353 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
2356 * Packet has already been tagged. Look for the next rule
2357 * to restart processing.
2359 * If fw_one_pass != 0 then just accept it.
2360 * XXX should not happen here, but optimized out in
2364 IPFW_RUNLOCK(chain);
2365 return (IP_FW_PASS);
2368 f = args->rule->next_rule;
2370 f = lookup_next_rule(args->rule);
2373 * Find the starting rule. It can be either the first
2374 * one, or the one after divert_rule if asked so.
2376 int skipto = mtag ? divert_cookie(mtag) : 0;
2379 if (args->eh == NULL && skipto != 0) {
2380 if (skipto >= IPFW_DEFAULT_RULE) {
2381 IPFW_RUNLOCK(chain);
2382 return (IP_FW_DENY); /* invalid */
2384 while (f && f->rulenum <= skipto)
2386 if (f == NULL) { /* drop packet */
2387 IPFW_RUNLOCK(chain);
2388 return (IP_FW_DENY);
2392 /* reset divert rule to avoid confusion later */
2394 divinput_flags = divert_info(mtag) &
2395 (IP_FW_DIVERT_OUTPUT_FLAG | IP_FW_DIVERT_LOOPBACK_FLAG);
2396 m_tag_delete(m, mtag);
2400 * Now scan the rules, and parse microinstructions for each rule.
2402 for (; f; f = f->next) {
2404 uint32_t tablearg = 0;
2405 int l, cmdlen, skip_or; /* skip rest of OR block */
2408 if (set_disable & (1 << f->set) )
2412 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ;
2413 l -= cmdlen, cmd += cmdlen) {
2417 * check_body is a jump target used when we find a
2418 * CHECK_STATE, and need to jump to the body of
2423 cmdlen = F_LEN(cmd);
2425 * An OR block (insn_1 || .. || insn_n) has the
2426 * F_OR bit set in all but the last instruction.
2427 * The first match will set "skip_or", and cause
2428 * the following instructions to be skipped until
2429 * past the one with the F_OR bit clear.
2431 if (skip_or) { /* skip this instruction */
2432 if ((cmd->len & F_OR) == 0)
2433 skip_or = 0; /* next one is good */
2436 match = 0; /* set to 1 if we succeed */
2438 switch (cmd->opcode) {
2440 * The first set of opcodes compares the packet's
2441 * fields with some pattern, setting 'match' if a
2442 * match is found. At the end of the loop there is
2443 * logic to deal with F_NOT and F_OR flags associated
2451 printf("ipfw: opcode %d unimplemented\n",
2459 * We only check offset == 0 && proto != 0,
2460 * as this ensures that we have a
2461 * packet with the ports info.
2465 if (is_ipv6) /* XXX to be fixed later */
2467 if (proto == IPPROTO_TCP ||
2468 proto == IPPROTO_UDP)
2469 match = check_uidgid(
2470 (ipfw_insn_u32 *)cmd,
2473 src_ip, src_port, &fw_ugid_cache,
2474 &ugid_lookup, args->inp);
2478 match = iface_match(m->m_pkthdr.rcvif,
2479 (ipfw_insn_if *)cmd);
2483 match = iface_match(oif, (ipfw_insn_if *)cmd);
2487 match = iface_match(oif ? oif :
2488 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
2492 if (args->eh != NULL) { /* have MAC header */
2493 u_int32_t *want = (u_int32_t *)
2494 ((ipfw_insn_mac *)cmd)->addr;
2495 u_int32_t *mask = (u_int32_t *)
2496 ((ipfw_insn_mac *)cmd)->mask;
2497 u_int32_t *hdr = (u_int32_t *)args->eh;
2500 ( want[0] == (hdr[0] & mask[0]) &&
2501 want[1] == (hdr[1] & mask[1]) &&
2502 want[2] == (hdr[2] & mask[2]) );
2507 if (args->eh != NULL) {
2509 ntohs(args->eh->ether_type);
2511 ((ipfw_insn_u16 *)cmd)->ports;
2514 for (i = cmdlen - 1; !match && i>0;
2516 match = (t>=p[0] && t<=p[1]);
2521 match = (offset != 0);
2524 case O_IN: /* "out" is "not in" */
2525 match = (oif == NULL);
2529 match = (args->eh != NULL);
2533 match = (cmd->arg1 & 1 && divinput_flags &
2534 IP_FW_DIVERT_LOOPBACK_FLAG) ||
2535 (cmd->arg1 & 2 && divinput_flags &
2536 IP_FW_DIVERT_OUTPUT_FLAG);
2541 * We do not allow an arg of 0 so the
2542 * check of "proto" only suffices.
2544 match = (proto == cmd->arg1);
2549 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2553 case O_IP_SRC_LOOKUP:
2554 case O_IP_DST_LOOKUP:
2557 (cmd->opcode == O_IP_DST_LOOKUP) ?
2558 dst_ip.s_addr : src_ip.s_addr;
2561 match = lookup_table(chain, cmd->arg1, a,
2565 if (cmdlen == F_INSN_SIZE(ipfw_insn_u32))
2567 ((ipfw_insn_u32 *)cmd)->d[0] == v;
2577 (cmd->opcode == O_IP_DST_MASK) ?
2578 dst_ip.s_addr : src_ip.s_addr;
2579 uint32_t *p = ((ipfw_insn_u32 *)cmd)->d;
2582 for (; !match && i>0; i-= 2, p+= 2)
2583 match = (p[0] == (a & p[1]));
2591 INADDR_TO_IFP(src_ip, tif);
2592 match = (tif != NULL);
2599 u_int32_t *d = (u_int32_t *)(cmd+1);
2601 cmd->opcode == O_IP_DST_SET ?
2607 addr -= d[0]; /* subtract base */
2608 match = (addr < cmd->arg1) &&
2609 ( d[ 1 + (addr>>5)] &
2610 (1<<(addr & 0x1f)) );
2616 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2624 INADDR_TO_IFP(dst_ip, tif);
2625 match = (tif != NULL);
2632 * offset == 0 && proto != 0 is enough
2633 * to guarantee that we have a
2634 * packet with port info.
2636 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
2639 (cmd->opcode == O_IP_SRCPORT) ?
2640 src_port : dst_port ;
2642 ((ipfw_insn_u16 *)cmd)->ports;
2645 for (i = cmdlen - 1; !match && i>0;
2647 match = (x>=p[0] && x<=p[1]);
2652 match = (offset == 0 && proto==IPPROTO_ICMP &&
2653 icmptype_match(ICMP(ulp), (ipfw_insn_u32 *)cmd) );
2658 match = is_ipv6 && offset == 0 &&
2659 proto==IPPROTO_ICMPV6 &&
2661 ICMP6(ulp)->icmp6_type,
2662 (ipfw_insn_u32 *)cmd);
2668 ipopts_match(mtod(m, struct ip *), cmd) );
2673 cmd->arg1 == mtod(m, struct ip *)->ip_v);
2679 if (is_ipv4) { /* only for IP packets */
2684 if (cmd->opcode == O_IPLEN)
2686 else if (cmd->opcode == O_IPTTL)
2687 x = mtod(m, struct ip *)->ip_ttl;
2688 else /* must be IPID */
2689 x = ntohs(mtod(m, struct ip *)->ip_id);
2691 match = (cmd->arg1 == x);
2694 /* otherwise we have ranges */
2695 p = ((ipfw_insn_u16 *)cmd)->ports;
2697 for (; !match && i>0; i--, p += 2)
2698 match = (x >= p[0] && x <= p[1]);
2702 case O_IPPRECEDENCE:
2704 (cmd->arg1 == (mtod(m, struct ip *)->ip_tos & 0xe0)) );
2709 flags_match(cmd, mtod(m, struct ip *)->ip_tos));
2713 if (proto == IPPROTO_TCP && offset == 0) {
2721 ((ip->ip_hl + tcp->th_off) << 2);
2723 match = (cmd->arg1 == x);
2726 /* otherwise we have ranges */
2727 p = ((ipfw_insn_u16 *)cmd)->ports;
2729 for (; !match && i>0; i--, p += 2)
2730 match = (x >= p[0] && x <= p[1]);
2735 match = (proto == IPPROTO_TCP && offset == 0 &&
2736 flags_match(cmd, TCP(ulp)->th_flags));
2740 match = (proto == IPPROTO_TCP && offset == 0 &&
2741 tcpopts_match(TCP(ulp), cmd));
2745 match = (proto == IPPROTO_TCP && offset == 0 &&
2746 ((ipfw_insn_u32 *)cmd)->d[0] ==
2751 match = (proto == IPPROTO_TCP && offset == 0 &&
2752 ((ipfw_insn_u32 *)cmd)->d[0] ==
2757 match = (proto == IPPROTO_TCP && offset == 0 &&
2758 cmd->arg1 == TCP(ulp)->th_win);
2762 /* reject packets which have SYN only */
2763 /* XXX should i also check for TH_ACK ? */
2764 match = (proto == IPPROTO_TCP && offset == 0 &&
2765 (TCP(ulp)->th_flags &
2766 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2770 struct altq_tag *at;
2771 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
2774 mtag = m_tag_find(m, PACKET_TAG_PF_QID, NULL);
2777 mtag = m_tag_get(PACKET_TAG_PF_QID,
2778 sizeof(struct altq_tag),
2782 * Let the packet fall back to the
2787 at = (struct altq_tag *)(mtag+1);
2788 at->qid = altq->qid;
2794 m_tag_prepend(m, mtag);
2800 ipfw_log(f, hlen, args, m,
2801 oif, offset, tablearg);
2806 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]);
2810 /* Outgoing packets automatically pass/match */
2811 match = ((oif != NULL) ||
2812 (m->m_pkthdr.rcvif == NULL) ||
2816 verify_path6(&(args->f_id.src_ip6),
2817 m->m_pkthdr.rcvif) :
2819 verify_path(src_ip, m->m_pkthdr.rcvif)));
2823 /* Outgoing packets automatically pass/match */
2824 match = (hlen > 0 && ((oif != NULL) ||
2827 verify_path6(&(args->f_id.src_ip6),
2830 verify_path(src_ip, NULL)));
2834 /* Outgoing packets automatically pass/match */
2835 if (oif == NULL && hlen > 0 &&
2836 ( (is_ipv4 && in_localaddr(src_ip))
2839 in6_localaddr(&(args->f_id.src_ip6)))
2844 is_ipv6 ? verify_path6(
2845 &(args->f_id.src_ip6),
2846 m->m_pkthdr.rcvif) :
2856 match = (m_tag_find(m,
2857 PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL);
2860 match = (ipsec_getnhist(m) != 0);
2862 /* otherwise no match */
2868 IN6_ARE_ADDR_EQUAL(&args->f_id.src_ip6,
2869 &((ipfw_insn_ip6 *)cmd)->addr6);
2874 IN6_ARE_ADDR_EQUAL(&args->f_id.dst_ip6,
2875 &((ipfw_insn_ip6 *)cmd)->addr6);
2877 case O_IP6_SRC_MASK:
2879 ipfw_insn_ip6 *te = (ipfw_insn_ip6 *)cmd;
2880 struct in6_addr p = args->f_id.src_ip6;
2882 APPLY_MASK(&p, &te->mask6);
2883 match = IN6_ARE_ADDR_EQUAL(&te->addr6, &p);
2887 case O_IP6_DST_MASK:
2889 ipfw_insn_ip6 *te = (ipfw_insn_ip6 *)cmd;
2890 struct in6_addr p = args->f_id.dst_ip6;
2892 APPLY_MASK(&p, &te->mask6);
2893 match = IN6_ARE_ADDR_EQUAL(&te->addr6, &p);
2898 match= is_ipv6 && search_ip6_addr_net(&args->f_id.src_ip6);
2902 match= is_ipv6 && search_ip6_addr_net(&args->f_id.dst_ip6);
2907 flow6id_match(args->f_id.flow_id6,
2908 (ipfw_insn_u32 *) cmd);
2913 (ext_hd & ((ipfw_insn *) cmd)->arg1);
2926 uint32_t tag = (cmd->arg1 == IP_FW_TABLEARG) ?
2927 tablearg : cmd->arg1;
2929 /* Packet is already tagged with this tag? */
2930 mtag = m_tag_locate(m, MTAG_IPFW, tag, NULL);
2932 /* We have `untag' action when F_NOT flag is
2933 * present. And we must remove this mtag from
2934 * mbuf and reset `match' to zero (`match' will
2935 * be inversed later).
2936 * Otherwise we should allocate new mtag and
2937 * push it into mbuf.
2939 if (cmd->len & F_NOT) { /* `untag' action */
2941 m_tag_delete(m, mtag);
2942 } else if (mtag == NULL) {
2943 if ((mtag = m_tag_alloc(MTAG_IPFW,
2944 tag, 0, M_NOWAIT)) != NULL)
2945 m_tag_prepend(m, mtag);
2947 match = (cmd->len & F_NOT) ? 0: 1;
2952 uint32_t tag = (cmd->arg1 == IP_FW_TABLEARG) ?
2953 tablearg : cmd->arg1;
2956 match = m_tag_locate(m, MTAG_IPFW,
2961 /* we have ranges */
2962 for (mtag = m_tag_first(m);
2963 mtag != NULL && !match;
2964 mtag = m_tag_next(m, mtag)) {
2968 if (mtag->m_tag_cookie != MTAG_IPFW)
2971 p = ((ipfw_insn_u16 *)cmd)->ports;
2973 for(; !match && i > 0; i--, p += 2)
2975 mtag->m_tag_id >= p[0] &&
2976 mtag->m_tag_id <= p[1];
2982 * The second set of opcodes represents 'actions',
2983 * i.e. the terminal part of a rule once the packet
2984 * matches all previous patterns.
2985 * Typically there is only one action for each rule,
2986 * and the opcode is stored at the end of the rule
2987 * (but there are exceptions -- see below).
2989 * In general, here we set retval and terminate the
2990 * outer loop (would be a 'break 3' in some language,
2991 * but we need to do a 'goto done').
2994 * O_COUNT and O_SKIPTO actions:
2995 * instead of terminating, we jump to the next rule
2996 * ('goto next_rule', equivalent to a 'break 2'),
2997 * or to the SKIPTO target ('goto again' after
2998 * having set f, cmd and l), respectively.
3000 * O_TAG, O_LOG and O_ALTQ action parameters:
3001 * perform some action and set match = 1;
3003 * O_LIMIT and O_KEEP_STATE: these opcodes are
3004 * not real 'actions', and are stored right
3005 * before the 'action' part of the rule.
3006 * These opcodes try to install an entry in the
3007 * state tables; if successful, we continue with
3008 * the next opcode (match=1; break;), otherwise
3009 * the packet * must be dropped
3010 * ('goto done' after setting retval);
3012 * O_PROBE_STATE and O_CHECK_STATE: these opcodes
3013 * cause a lookup of the state table, and a jump
3014 * to the 'action' part of the parent rule
3015 * ('goto check_body') if an entry is found, or
3016 * (CHECK_STATE only) a jump to the next rule if
3017 * the entry is not found ('goto next_rule').
3018 * The result of the lookup is cached to make
3019 * further instances of these opcodes are
3024 if (install_state(f,
3025 (ipfw_insn_limit *)cmd, args, tablearg)) {
3026 retval = IP_FW_DENY;
3027 goto done; /* error/limit violation */
3035 * dynamic rules are checked at the first
3036 * keep-state or check-state occurrence,
3037 * with the result being stored in dyn_dir.
3038 * The compiler introduces a PROBE_STATE
3039 * instruction for us when we have a
3040 * KEEP_STATE (because PROBE_STATE needs
3043 if (dyn_dir == MATCH_UNKNOWN &&
3044 (q = lookup_dyn_rule(&args->f_id,
3045 &dyn_dir, proto == IPPROTO_TCP ?
3049 * Found dynamic entry, update stats
3050 * and jump to the 'action' part of
3056 cmd = ACTION_PTR(f);
3057 l = f->cmd_len - f->act_ofs;
3062 * Dynamic entry not found. If CHECK_STATE,
3063 * skip to next rule, if PROBE_STATE just
3064 * ignore and continue with next opcode.
3066 if (cmd->opcode == O_CHECK_STATE)
3072 retval = 0; /* accept */
3077 args->rule = f; /* report matching rule */
3078 if (cmd->arg1 == IP_FW_TABLEARG)
3079 args->cookie = tablearg;
3081 args->cookie = cmd->arg1;
3082 retval = IP_FW_DUMMYNET;
3087 struct divert_tag *dt;
3089 if (args->eh) /* not on layer 2 */
3091 mtag = m_tag_get(PACKET_TAG_DIVERT,
3092 sizeof(struct divert_tag),
3097 IPFW_RUNLOCK(chain);
3098 return (IP_FW_DENY);
3100 dt = (struct divert_tag *)(mtag+1);
3101 dt->cookie = f->rulenum;
3102 if (cmd->arg1 == IP_FW_TABLEARG)
3103 dt->info = tablearg;
3105 dt->info = cmd->arg1;
3106 m_tag_prepend(m, mtag);
3107 retval = (cmd->opcode == O_DIVERT) ?
3108 IP_FW_DIVERT : IP_FW_TEE;
3114 f->pcnt++; /* update stats */
3116 f->timestamp = time_uptime;
3117 if (cmd->opcode == O_COUNT)
3120 if (f->next_rule == NULL)
3121 lookup_next_rule(f);
3127 * Drop the packet and send a reject notice
3128 * if the packet is not ICMP (or is an ICMP
3129 * query), and it is not multicast/broadcast.
3131 if (hlen > 0 && is_ipv4 && offset == 0 &&
3132 (proto != IPPROTO_ICMP ||
3133 is_icmp_query(ICMP(ulp))) &&
3134 !(m->m_flags & (M_BCAST|M_MCAST)) &&
3135 !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
3136 send_reject(args, cmd->arg1, ip_len);
3142 if (hlen > 0 && is_ipv6 &&
3143 ((offset & IP6F_OFF_MASK) == 0) &&
3144 (proto != IPPROTO_ICMPV6 ||
3145 (is_icmp6_query(args->f_id.flags) == 1)) &&
3146 !(m->m_flags & (M_BCAST|M_MCAST)) &&
3147 !IN6_IS_ADDR_MULTICAST(&args->f_id.dst_ip6)) {
3148 send_reject6(args, cmd->arg1, hlen);
3154 retval = IP_FW_DENY;
3157 case O_FORWARD_IP: {
3158 struct sockaddr_in *sa;
3159 sa = &(((ipfw_insn_sa *)cmd)->sa);
3160 if (args->eh) /* not valid on layer2 pkts */
3162 if (!q || dyn_dir == MATCH_FORWARD) {
3163 if (sa->sin_addr.s_addr == INADDR_ANY) {
3164 bcopy(sa, &args->hopstore,
3166 args->hopstore.sin_addr.s_addr =
3171 args->next_hop = sa;
3174 retval = IP_FW_PASS;
3180 args->rule = f; /* report matching rule */
3181 if (cmd->arg1 == IP_FW_TABLEARG)
3182 args->cookie = tablearg;
3184 args->cookie = cmd->arg1;
3185 retval = (cmd->opcode == O_NETGRAPH) ?
3186 IP_FW_NETGRAPH : IP_FW_NGTEE;
3190 panic("-- unknown opcode %d\n", cmd->opcode);
3191 } /* end of switch() on opcodes */
3193 if (cmd->len & F_NOT)
3197 if (cmd->len & F_OR)
3200 if (!(cmd->len & F_OR)) /* not an OR block, */
3201 break; /* try next rule */
3204 } /* end of inner for, scan opcodes */
3206 next_rule:; /* try next rule */
3208 } /* end of outer for, scan rules */
3209 printf("ipfw: ouch!, skip past end of rules, denying packet\n");
3210 IPFW_RUNLOCK(chain);
3211 return (IP_FW_DENY);
3214 /* Update statistics */
3217 f->timestamp = time_uptime;
3218 IPFW_RUNLOCK(chain);
3223 printf("ipfw: pullup failed\n");
3224 return (IP_FW_DENY);
3228 * When a rule is added/deleted, clear the next_rule pointers in all rules.
3229 * These will be reconstructed on the fly as packets are matched.
3232 flush_rule_ptrs(struct ip_fw_chain *chain)
3236 IPFW_WLOCK_ASSERT(chain);
3238 for (rule = chain->rules; rule; rule = rule->next)
3239 rule->next_rule = NULL;
3243 * Add a new rule to the list. Copy the rule into a malloc'ed area, then
3244 * possibly create a rule number and add the rule to the list.
3245 * Update the rule_number in the input struct so the caller knows it as well.
3248 add_rule(struct ip_fw_chain *chain, struct ip_fw *input_rule)
3250 struct ip_fw *rule, *f, *prev;
3251 int l = RULESIZE(input_rule);
3253 if (chain->rules == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE)
3256 rule = malloc(l, M_IPFW, M_NOWAIT | M_ZERO);
3260 bcopy(input_rule, rule, l);
3263 rule->next_rule = NULL;
3267 rule->timestamp = 0;
3271 if (chain->rules == NULL) { /* default rule */
3272 chain->rules = rule;
3277 * If rulenum is 0, find highest numbered rule before the
3278 * default rule, and add autoinc_step
3280 if (autoinc_step < 1)
3282 else if (autoinc_step > 1000)
3283 autoinc_step = 1000;
3284 if (rule->rulenum == 0) {
3286 * locate the highest numbered rule before default
3288 for (f = chain->rules; f; f = f->next) {
3289 if (f->rulenum == IPFW_DEFAULT_RULE)
3291 rule->rulenum = f->rulenum;
3293 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
3294 rule->rulenum += autoinc_step;
3295 input_rule->rulenum = rule->rulenum;
3299 * Now insert the new rule in the right place in the sorted list.
3301 for (prev = NULL, f = chain->rules; f; prev = f, f = f->next) {
3302 if (f->rulenum > rule->rulenum) { /* found the location */
3306 } else { /* head insert */
3307 rule->next = chain->rules;
3308 chain->rules = rule;
3313 flush_rule_ptrs(chain);
3317 IPFW_WUNLOCK(chain);
3318 DEB(printf("ipfw: installed rule %d, static count now %d\n",
3319 rule->rulenum, static_count);)
3324 * Remove a static rule (including derived * dynamic rules)
3325 * and place it on the ``reap list'' for later reclamation.
3326 * The caller is in charge of clearing rule pointers to avoid
3327 * dangling pointers.
3328 * @return a pointer to the next entry.
3329 * Arguments are not checked, so they better be correct.
3331 static struct ip_fw *
3332 remove_rule(struct ip_fw_chain *chain, struct ip_fw *rule, struct ip_fw *prev)
3335 int l = RULESIZE(rule);
3337 IPFW_WLOCK_ASSERT(chain);
3341 remove_dyn_rule(rule, NULL /* force removal */);
3350 rule->next = chain->reap;
3357 * Reclaim storage associated with a list of rules. This is
3358 * typically the list created using remove_rule.
3361 reap_rules(struct ip_fw *head)
3365 while ((rule = head) != NULL) {
3367 if (DUMMYNET_LOADED)
3368 ip_dn_ruledel_ptr(rule);
3374 * Remove all rules from a chain (except rules in set RESVD_SET
3375 * unless kill_default = 1). The caller is responsible for
3376 * reclaiming storage for the rules left in chain->reap.
3379 free_chain(struct ip_fw_chain *chain, int kill_default)
3381 struct ip_fw *prev, *rule;
3383 IPFW_WLOCK_ASSERT(chain);
3385 flush_rule_ptrs(chain); /* more efficient to do outside the loop */
3386 for (prev = NULL, rule = chain->rules; rule ; )
3387 if (kill_default || rule->set != RESVD_SET)
3388 rule = remove_rule(chain, rule, prev);
3396 * Remove all rules with given number, and also do set manipulation.
3397 * Assumes chain != NULL && *chain != NULL.
3399 * The argument is an u_int32_t. The low 16 bit are the rule or set number,
3400 * the next 8 bits are the new set, the top 8 bits are the command:
3402 * 0 delete rules with given number
3403 * 1 delete rules with given set number
3404 * 2 move rules with given number to new set
3405 * 3 move rules with given set number to new set
3406 * 4 swap sets with given numbers
3409 del_entry(struct ip_fw_chain *chain, u_int32_t arg)
3411 struct ip_fw *prev = NULL, *rule;
3412 u_int16_t rulenum; /* rule or old_set */
3413 u_int8_t cmd, new_set;
3415 rulenum = arg & 0xffff;
3416 cmd = (arg >> 24) & 0xff;
3417 new_set = (arg >> 16) & 0xff;
3421 if (new_set > RESVD_SET)
3423 if (cmd == 0 || cmd == 2) {
3424 if (rulenum >= IPFW_DEFAULT_RULE)
3427 if (rulenum > RESVD_SET) /* old_set */
3432 rule = chain->rules;
3435 case 0: /* delete rules with given number */
3437 * locate first rule to delete
3439 for (; rule->rulenum < rulenum; prev = rule, rule = rule->next)
3441 if (rule->rulenum != rulenum) {
3442 IPFW_WUNLOCK(chain);
3447 * flush pointers outside the loop, then delete all matching
3448 * rules. prev remains the same throughout the cycle.
3450 flush_rule_ptrs(chain);
3451 while (rule->rulenum == rulenum)
3452 rule = remove_rule(chain, rule, prev);
3455 case 1: /* delete all rules with given set number */
3456 flush_rule_ptrs(chain);
3457 rule = chain->rules;
3458 while (rule->rulenum < IPFW_DEFAULT_RULE)
3459 if (rule->set == rulenum)
3460 rule = remove_rule(chain, rule, prev);
3467 case 2: /* move rules with given number to new set */
3468 rule = chain->rules;
3469 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3470 if (rule->rulenum == rulenum)
3471 rule->set = new_set;
3474 case 3: /* move rules with given set number to new set */
3475 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3476 if (rule->set == rulenum)
3477 rule->set = new_set;
3480 case 4: /* swap two sets */
3481 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3482 if (rule->set == rulenum)
3483 rule->set = new_set;
3484 else if (rule->set == new_set)
3485 rule->set = rulenum;
3489 * Look for rules to reclaim. We grab the list before
3490 * releasing the lock then reclaim them w/o the lock to
3491 * avoid a LOR with dummynet.
3495 IPFW_WUNLOCK(chain);
3502 * Clear counters for a specific rule.
3503 * The enclosing "table" is assumed locked.
3506 clear_counters(struct ip_fw *rule, int log_only)
3508 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
3510 if (log_only == 0) {
3511 rule->bcnt = rule->pcnt = 0;
3512 rule->timestamp = 0;
3514 if (l->o.opcode == O_LOG)
3515 l->log_left = l->max_log;
3519 * Reset some or all counters on firewall rules.
3520 * @arg frwl is null to clear all entries, or contains a specific
3522 * @arg log_only is 1 if we only want to reset logs, zero otherwise.
3525 zero_entry(struct ip_fw_chain *chain, int rulenum, int log_only)
3533 for (rule = chain->rules; rule; rule = rule->next)
3534 clear_counters(rule, log_only);
3535 msg = log_only ? "ipfw: All logging counts reset.\n" :
3536 "ipfw: Accounting cleared.\n";
3540 * We can have multiple rules with the same number, so we
3541 * need to clear them all.
3543 for (rule = chain->rules; rule; rule = rule->next)
3544 if (rule->rulenum == rulenum) {
3545 while (rule && rule->rulenum == rulenum) {
3546 clear_counters(rule, log_only);
3552 if (!cleared) { /* we did not find any matching rules */
3553 IPFW_WUNLOCK(chain);
3556 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
3557 "ipfw: Entry %d cleared.\n";
3559 IPFW_WUNLOCK(chain);
3562 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
3567 * Check validity of the structure before insert.
3568 * Fortunately rules are simple, so this mostly need to check rule sizes.
3571 check_ipfw_struct(struct ip_fw *rule, int size)
3577 if (size < sizeof(*rule)) {
3578 printf("ipfw: rule too short\n");
3581 /* first, check for valid size */
3584 printf("ipfw: size mismatch (have %d want %d)\n", size, l);
3587 if (rule->act_ofs >= rule->cmd_len) {
3588 printf("ipfw: bogus action offset (%u > %u)\n",
3589 rule->act_ofs, rule->cmd_len - 1);
3593 * Now go for the individual checks. Very simple ones, basically only
3594 * instruction sizes.
3596 for (l = rule->cmd_len, cmd = rule->cmd ;
3597 l > 0 ; l -= cmdlen, cmd += cmdlen) {
3598 cmdlen = F_LEN(cmd);
3600 printf("ipfw: opcode %d size truncated\n",
3604 DEB(printf("ipfw: opcode %d\n", cmd->opcode);)
3605 switch (cmd->opcode) {
3617 case O_IPPRECEDENCE:
3635 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3648 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3653 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
3658 if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
3661 ((ipfw_insn_log *)cmd)->log_left =
3662 ((ipfw_insn_log *)cmd)->max_log;
3668 /* only odd command lengths */
3669 if ( !(cmdlen & 1) || cmdlen > 31)
3675 if (cmd->arg1 == 0 || cmd->arg1 > 256) {
3676 printf("ipfw: invalid set size %d\n",
3680 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3685 case O_IP_SRC_LOOKUP:
3686 case O_IP_DST_LOOKUP:
3687 if (cmd->arg1 >= IPFW_TABLES_MAX) {
3688 printf("ipfw: invalid table number %d\n",
3692 if (cmdlen != F_INSN_SIZE(ipfw_insn) &&
3693 cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3698 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
3708 if (cmdlen < 1 || cmdlen > 31)
3714 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
3715 if (cmdlen < 2 || cmdlen > 31)
3722 if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
3727 if (cmdlen != F_INSN_SIZE(ipfw_insn_altq))
3733 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3738 #ifdef IPFIREWALL_FORWARD
3739 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
3748 if (ip_divert_ptr == NULL)
3754 if (!NG_IPFW_LOADED)
3758 case O_FORWARD_MAC: /* XXX not implemented yet */
3769 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3773 printf("ipfw: opcode %d, multiple actions"
3780 printf("ipfw: opcode %d, action must be"
3789 if (cmdlen != F_INSN_SIZE(struct in6_addr) +
3790 F_INSN_SIZE(ipfw_insn))
3795 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3796 ((ipfw_insn_u32 *)cmd)->o.arg1)
3800 case O_IP6_SRC_MASK:
3801 case O_IP6_DST_MASK:
3802 if ( !(cmdlen & 1) || cmdlen > 127)
3806 if( cmdlen != F_INSN_SIZE( ipfw_insn_icmp6 ) )
3812 switch (cmd->opcode) {
3822 case O_IP6_SRC_MASK:
3823 case O_IP6_DST_MASK:
3825 printf("ipfw: no IPv6 support in kernel\n");
3826 return EPROTONOSUPPORT;
3829 printf("ipfw: opcode %d, unknown opcode\n",
3835 if (have_action == 0) {
3836 printf("ipfw: missing action\n");
3842 printf("ipfw: opcode %d size %d wrong\n",
3843 cmd->opcode, cmdlen);
3848 * Copy the static and dynamic rules to the supplied buffer
3849 * and return the amount of space actually used.
3852 ipfw_getrules(struct ip_fw_chain *chain, void *buf, size_t space)
3855 char *ep = bp + space;
3859 /* XXX this can take a long time and locking will block packet flow */
3861 for (rule = chain->rules; rule ; rule = rule->next) {
3863 * Verify the entry fits in the buffer in case the
3864 * rules changed between calculating buffer space and
3865 * now. This would be better done using a generation
3866 * number but should suffice for now.
3871 bcopy(&set_disable, &(((struct ip_fw *)bp)->next_rule),
3872 sizeof(set_disable));
3876 IPFW_RUNLOCK(chain);
3878 ipfw_dyn_rule *p, *last = NULL;
3881 for (i = 0 ; i < curr_dyn_buckets; i++)
3882 for (p = ipfw_dyn_v[i] ; p != NULL; p = p->next) {
3883 if (bp + sizeof *p <= ep) {
3884 ipfw_dyn_rule *dst =
3885 (ipfw_dyn_rule *)bp;
3886 bcopy(p, dst, sizeof *p);
3887 bcopy(&(p->rule->rulenum), &(dst->rule),
3888 sizeof(p->rule->rulenum));
3890 * store a non-null value in "next".
3891 * The userland code will interpret a
3892 * NULL here as a marker
3893 * for the last dynamic rule.
3895 bcopy(&dst, &dst->next, sizeof(dst));
3898 TIME_LEQ(dst->expire, time_uptime) ?
3899 0 : dst->expire - time_uptime ;
3900 bp += sizeof(ipfw_dyn_rule);
3904 if (last != NULL) /* mark last dynamic rule */
3905 bzero(&last->next, sizeof(last));
3907 return (bp - (char *)buf);
3912 * {set|get}sockopt parser.
3915 ipfw_ctl(struct sockopt *sopt)
3917 #define RULE_MAXSIZE (256*sizeof(u_int32_t))
3918 int error, rule_num;
3920 struct ip_fw *buf, *rule;
3921 u_int32_t rulenum[2];
3923 error = suser(sopt->sopt_td);
3928 * Disallow modifications in really-really secure mode, but still allow
3929 * the logging counters to be reset.
3931 if (sopt->sopt_name == IP_FW_ADD ||
3932 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
3933 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
3940 switch (sopt->sopt_name) {
3943 * pass up a copy of the current rules. Static rules
3944 * come first (the last of which has number IPFW_DEFAULT_RULE),
3945 * followed by a possibly empty list of dynamic rule.
3946 * The last dynamic rule has NULL in the "next" field.
3948 * Note that the calculated size is used to bound the
3949 * amount of data returned to the user. The rule set may
3950 * change between calculating the size and returning the
3951 * data in which case we'll just return what fits.
3953 size = static_len; /* size of static rules */
3954 if (ipfw_dyn_v) /* add size of dyn.rules */
3955 size += (dyn_count * sizeof(ipfw_dyn_rule));
3958 * XXX todo: if the user passes a short length just to know
3959 * how much room is needed, do not bother filling up the
3960 * buffer, just jump to the sooptcopyout.
3962 buf = malloc(size, M_TEMP, M_WAITOK);
3963 error = sooptcopyout(sopt, buf,
3964 ipfw_getrules(&layer3_chain, buf, size));
3970 * Normally we cannot release the lock on each iteration.
3971 * We could do it here only because we start from the head all
3972 * the times so there is no risk of missing some entries.
3973 * On the other hand, the risk is that we end up with
3974 * a very inconsistent ruleset, so better keep the lock
3975 * around the whole cycle.
3977 * XXX this code can be improved by resetting the head of
3978 * the list to point to the default rule, and then freeing
3979 * the old list without the need for a lock.
3982 IPFW_WLOCK(&layer3_chain);
3983 layer3_chain.reap = NULL;
3984 free_chain(&layer3_chain, 0 /* keep default rule */);
3985 rule = layer3_chain.reap;
3986 layer3_chain.reap = NULL;
3987 IPFW_WUNLOCK(&layer3_chain);
3993 rule = malloc(RULE_MAXSIZE, M_TEMP, M_WAITOK);
3994 error = sooptcopyin(sopt, rule, RULE_MAXSIZE,
3995 sizeof(struct ip_fw) );
3997 error = check_ipfw_struct(rule, sopt->sopt_valsize);
3999 error = add_rule(&layer3_chain, rule);
4000 size = RULESIZE(rule);
4001 if (!error && sopt->sopt_dir == SOPT_GET)
4002 error = sooptcopyout(sopt, rule, size);
4009 * IP_FW_DEL is used for deleting single rules or sets,
4010 * and (ab)used to atomically manipulate sets. Argument size
4011 * is used to distinguish between the two:
4013 * delete single rule or set of rules,
4014 * or reassign rules (or sets) to a different set.
4015 * 2*sizeof(u_int32_t)
4016 * atomic disable/enable sets.
4017 * first u_int32_t contains sets to be disabled,
4018 * second u_int32_t contains sets to be enabled.
4020 error = sooptcopyin(sopt, rulenum,
4021 2*sizeof(u_int32_t), sizeof(u_int32_t));
4024 size = sopt->sopt_valsize;
4025 if (size == sizeof(u_int32_t)) /* delete or reassign */
4026 error = del_entry(&layer3_chain, rulenum[0]);
4027 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */
4029 (set_disable | rulenum[0]) & ~rulenum[1] &
4030 ~(1<<RESVD_SET); /* set RESVD_SET always enabled */
4036 case IP_FW_RESETLOG: /* argument is an int, the rule number */
4038 if (sopt->sopt_val != 0) {
4039 error = sooptcopyin(sopt, &rule_num,
4040 sizeof(int), sizeof(int));
4044 error = zero_entry(&layer3_chain, rule_num,
4045 sopt->sopt_name == IP_FW_RESETLOG);
4048 case IP_FW_TABLE_ADD:
4050 ipfw_table_entry ent;
4052 error = sooptcopyin(sopt, &ent,
4053 sizeof(ent), sizeof(ent));
4056 error = add_table_entry(&layer3_chain, ent.tbl,
4057 ent.addr, ent.masklen, ent.value);
4061 case IP_FW_TABLE_DEL:
4063 ipfw_table_entry ent;
4065 error = sooptcopyin(sopt, &ent,
4066 sizeof(ent), sizeof(ent));
4069 error = del_table_entry(&layer3_chain, ent.tbl,
4070 ent.addr, ent.masklen);
4074 case IP_FW_TABLE_FLUSH:
4078 error = sooptcopyin(sopt, &tbl,
4079 sizeof(tbl), sizeof(tbl));
4082 IPFW_WLOCK(&layer3_chain);
4083 error = flush_table(&layer3_chain, tbl);
4084 IPFW_WUNLOCK(&layer3_chain);
4088 case IP_FW_TABLE_GETSIZE:
4092 if ((error = sooptcopyin(sopt, &tbl, sizeof(tbl),
4095 IPFW_RLOCK(&layer3_chain);
4096 error = count_table(&layer3_chain, tbl, &cnt);
4097 IPFW_RUNLOCK(&layer3_chain);
4100 error = sooptcopyout(sopt, &cnt, sizeof(cnt));
4104 case IP_FW_TABLE_LIST:
4108 if (sopt->sopt_valsize < sizeof(*tbl)) {
4112 size = sopt->sopt_valsize;
4113 tbl = malloc(size, M_TEMP, M_WAITOK);
4114 error = sooptcopyin(sopt, tbl, size, sizeof(*tbl));
4119 tbl->size = (size - sizeof(*tbl)) /
4120 sizeof(ipfw_table_entry);
4121 IPFW_RLOCK(&layer3_chain);
4122 error = dump_table(&layer3_chain, tbl);
4123 IPFW_RUNLOCK(&layer3_chain);
4128 error = sooptcopyout(sopt, tbl, size);
4134 printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name);
4143 * dummynet needs a reference to the default rule, because rules can be
4144 * deleted while packets hold a reference to them. When this happens,
4145 * dummynet changes the reference to the default rule (it could well be a
4146 * NULL pointer, but this way we do not need to check for the special
4147 * case, plus here he have info on the default behaviour).
4149 struct ip_fw *ip_fw_default_rule;
4152 * This procedure is only used to handle keepalives. It is invoked
4153 * every dyn_keepalive_period
4156 ipfw_tick(void * __unused unused)
4158 struct mbuf *m0, *m, *mnext, **mtailp;
4162 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
4166 * We make a chain of packets to go out here -- not deferring
4167 * until after we drop the IPFW dynamic rule lock would result
4168 * in a lock order reversal with the normal packet input -> ipfw
4174 for (i = 0 ; i < curr_dyn_buckets ; i++) {
4175 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
4176 if (q->dyn_type == O_LIMIT_PARENT)
4178 if (q->id.proto != IPPROTO_TCP)
4180 if ( (q->state & BOTH_SYN) != BOTH_SYN)
4182 if (TIME_LEQ( time_uptime+dyn_keepalive_interval,
4184 continue; /* too early */
4185 if (TIME_LEQ(q->expire, time_uptime))
4186 continue; /* too late, rule expired */
4188 *mtailp = send_pkt(&(q->id), q->ack_rev - 1,
4189 q->ack_fwd, TH_SYN);
4190 if (*mtailp != NULL)
4191 mtailp = &(*mtailp)->m_nextpkt;
4192 *mtailp = send_pkt(&(q->id), q->ack_fwd - 1,
4194 if (*mtailp != NULL)
4195 mtailp = &(*mtailp)->m_nextpkt;
4199 for (m = mnext = m0; m != NULL; m = mnext) {
4200 mnext = m->m_nextpkt;
4201 m->m_nextpkt = NULL;
4202 ip_output(m, NULL, NULL, 0, NULL, NULL);
4205 callout_reset(&ipfw_timeout, dyn_keepalive_period*hz, ipfw_tick, NULL);
4211 struct ip_fw default_rule;
4215 /* Setup IPv6 fw sysctl tree. */
4216 sysctl_ctx_init(&ip6_fw_sysctl_ctx);
4217 ip6_fw_sysctl_tree = SYSCTL_ADD_NODE(&ip6_fw_sysctl_ctx,
4218 SYSCTL_STATIC_CHILDREN(_net_inet6_ip6), OID_AUTO, "fw",
4219 CTLFLAG_RW | CTLFLAG_SECURE, 0, "Firewall");
4220 SYSCTL_ADD_PROC(&ip6_fw_sysctl_ctx, SYSCTL_CHILDREN(ip6_fw_sysctl_tree),
4221 OID_AUTO, "enable", CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE3,
4222 &fw6_enable, 0, ipfw_chg_hook, "I", "Enable ipfw+6");
4223 SYSCTL_ADD_INT(&ip6_fw_sysctl_ctx, SYSCTL_CHILDREN(ip6_fw_sysctl_tree),
4224 OID_AUTO, "deny_unknown_exthdrs", CTLFLAG_RW | CTLFLAG_SECURE,
4225 &fw_deny_unknown_exthdrs, 0,
4226 "Deny packets with unknown IPv6 Extension Headers");
4229 layer3_chain.rules = NULL;
4230 IPFW_LOCK_INIT(&layer3_chain);
4231 ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule zone",
4232 sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL,
4234 IPFW_DYN_LOCK_INIT();
4235 callout_init(&ipfw_timeout, NET_CALLOUT_MPSAFE);
4237 bzero(&default_rule, sizeof default_rule);
4239 default_rule.act_ofs = 0;
4240 default_rule.rulenum = IPFW_DEFAULT_RULE;
4241 default_rule.cmd_len = 1;
4242 default_rule.set = RESVD_SET;
4244 default_rule.cmd[0].len = 1;
4245 default_rule.cmd[0].opcode =
4246 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
4251 error = add_rule(&layer3_chain, &default_rule);
4253 printf("ipfw2: error %u initializing default rule "
4254 "(support disabled)\n", error);
4255 IPFW_DYN_LOCK_DESTROY();
4256 IPFW_LOCK_DESTROY(&layer3_chain);
4257 uma_zdestroy(ipfw_dyn_rule_zone);
4261 ip_fw_default_rule = layer3_chain.rules;
4266 "initialized, divert %s, "
4267 "rule-based forwarding "
4268 #ifdef IPFIREWALL_FORWARD
4273 "default to %s, logging ",
4279 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny");
4281 #ifdef IPFIREWALL_VERBOSE
4284 #ifdef IPFIREWALL_VERBOSE_LIMIT
4285 verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
4287 if (fw_verbose == 0)
4288 printf("disabled\n");
4289 else if (verbose_limit == 0)
4290 printf("unlimited\n");
4292 printf("limited to %d packets/entry by default\n",
4295 error = init_tables(&layer3_chain);
4297 IPFW_DYN_LOCK_DESTROY();
4298 IPFW_LOCK_DESTROY(&layer3_chain);
4299 uma_zdestroy(ipfw_dyn_rule_zone);
4302 ip_fw_ctl_ptr = ipfw_ctl;
4303 ip_fw_chk_ptr = ipfw_chk;
4304 callout_reset(&ipfw_timeout, hz, ipfw_tick, NULL);
4314 ip_fw_chk_ptr = NULL;
4315 ip_fw_ctl_ptr = NULL;
4316 callout_drain(&ipfw_timeout);
4317 IPFW_WLOCK(&layer3_chain);
4318 flush_tables(&layer3_chain);
4319 layer3_chain.reap = NULL;
4320 free_chain(&layer3_chain, 1 /* kill default rule */);
4321 reap = layer3_chain.reap, layer3_chain.reap = NULL;
4322 IPFW_WUNLOCK(&layer3_chain);
4325 IPFW_DYN_LOCK_DESTROY();
4326 uma_zdestroy(ipfw_dyn_rule_zone);
4327 IPFW_LOCK_DESTROY(&layer3_chain);
4330 /* Free IPv6 fw sysctl tree. */
4331 sysctl_ctx_free(&ip6_fw_sysctl_ctx);
4334 printf("IP firewall unloaded\n");