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
25 * $FreeBSD: src/sys/netinet/ip_fw2.c,v 1.6.2.12 2003/04/08 10:42:32 maxim Exp $
26 * $DragonFly: src/sys/net/ipfw/ip_fw2.c,v 1.67 2008/08/09 06:09:18 sephe Exp $
33 * Implement IP packet firewall (new version)
39 #include "opt_ipdivert.h"
42 #error IPFIREWALL requires INET.
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/malloc.h>
50 #include <sys/kernel.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
54 #include <sys/sysctl.h>
55 #include <sys/syslog.h>
56 #include <sys/thread2.h>
57 #include <sys/ucred.h>
58 #include <sys/in_cksum.h>
62 #include <net/route.h>
63 #include <net/netmsg2.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>
73 #include <net/dummynet/ip_dummynet.h>
74 #include <netinet/tcp.h>
75 #include <netinet/tcp_timer.h>
76 #include <netinet/tcp_var.h>
77 #include <netinet/tcpip.h>
78 #include <netinet/udp.h>
79 #include <netinet/udp_var.h>
81 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
83 #define IPFW_AUTOINC_STEP_MIN 1
84 #define IPFW_AUTOINC_STEP_MAX 1000
85 #define IPFW_AUTOINC_STEP_DEF 100
88 * set_disable contains one bit per set value (0..31).
89 * If the bit is set, all rules with the corresponding set
90 * are disabled. Set 31 is reserved for the default rule
91 * and CANNOT be disabled.
93 static uint32_t set_disable;
95 static int fw_verbose;
96 static int verbose_limit;
99 static int ipfw_refcnt;
102 static struct callout ipfw_timeout_h;
103 #define IPFW_DEFAULT_RULE 65535
106 * list of rules for layer 3
108 static struct ip_fw *layer3_chain;
110 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
112 static int fw_debug = 1;
113 static int autoinc_step = IPFW_AUTOINC_STEP_DEF;
114 static uint32_t static_count; /* # of static rules */
115 static uint32_t static_ioc_len; /* bytes of static rules */
116 static uint32_t static_gen; /* generation of static rules */
118 static int ipfw_sysctl_autoinc_step(SYSCTL_HANDLER_ARGS);
119 static int ipfw_sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS);
120 static int ipfw_sysctl_dyn_fin(SYSCTL_HANDLER_ARGS);
121 static int ipfw_sysctl_dyn_rst(SYSCTL_HANDLER_ARGS);
124 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
125 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable, CTLFLAG_RW,
126 &fw_enable, 0, "Enable ipfw");
127 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLTYPE_INT | CTLFLAG_RW,
128 &autoinc_step, 0, ipfw_sysctl_autoinc_step, "I",
129 "Rule number autincrement step");
130 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW,
132 "Only do a single pass through ipfw when using dummynet(4)");
133 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
134 &fw_debug, 0, "Enable printing of debug ip_fw statements");
135 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW,
136 &fw_verbose, 0, "Log matches to ipfw rules");
137 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
138 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
141 * Description of dynamic rules.
143 * Dynamic rules are stored in lists accessed through a hash table
144 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
145 * be modified through the sysctl variable dyn_buckets which is
146 * updated when the table becomes empty.
148 * XXX currently there is only one list, ipfw_dyn.
150 * When a packet is received, its address fields are first masked
151 * with the mask defined for the rule, then hashed, then matched
152 * against the entries in the corresponding list.
153 * Dynamic rules can be used for different purposes:
155 * + enforcing limits on the number of sessions;
156 * + in-kernel NAT (not implemented yet)
158 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
159 * measured in seconds and depending on the flags.
161 * The total number of dynamic rules is stored in dyn_count.
162 * The max number of dynamic rules is dyn_max. When we reach
163 * the maximum number of rules we do not create anymore. This is
164 * done to avoid consuming too much memory, but also too much
165 * time when searching on each packet (ideally, we should try instead
166 * to put a limit on the length of the list on each bucket...).
168 * Each dynamic rule holds a pointer to the parent ipfw rule so
169 * we know what action to perform. Dynamic rules are removed when
170 * the parent rule is deleted. XXX we should make them survive.
172 * There are some limitations with dynamic rules -- we do not
173 * obey the 'randomized match', and we do not do multiple
174 * passes through the firewall. XXX check the latter!!!
176 * NOTE about the SHARED LOCKMGR LOCK during dynamic rule looking up:
177 * Only TCP state transition will change dynamic rule's state and ack
178 * sequences, while all packets of one TCP connection only goes through
179 * one TCP thread, so it is safe to use shared lockmgr lock during dynamic
180 * rule looking up. The keep alive callout uses exclusive lockmgr lock
181 * when it tries to find suitable dynamic rules to send keep alive, so
182 * it will not see half updated state and ack sequences. Though the expire
183 * field updating looks racy for other protocols, the resolution (second)
184 * of expire field makes this kind of race harmless.
185 * XXX statistics' updating is _not_ MPsafe!!!
186 * XXX once UDP output path is fixed, we could use lockless dynamic rule
189 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
190 static uint32_t dyn_buckets = 256; /* must be power of 2 */
191 static uint32_t curr_dyn_buckets = 256; /* must be power of 2 */
192 static uint32_t dyn_buckets_gen; /* generation of dyn buckets array */
193 static struct lock dyn_lock; /* dynamic rules' hash table lock */
196 * Timeouts for various events in handing dynamic rules.
198 static uint32_t dyn_ack_lifetime = 300;
199 static uint32_t dyn_syn_lifetime = 20;
200 static uint32_t dyn_fin_lifetime = 1;
201 static uint32_t dyn_rst_lifetime = 1;
202 static uint32_t dyn_udp_lifetime = 10;
203 static uint32_t dyn_short_lifetime = 5;
206 * Keepalives are sent if dyn_keepalive is set. They are sent every
207 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
208 * seconds of lifetime of a rule.
209 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
210 * than dyn_keepalive_period.
213 static uint32_t dyn_keepalive_interval = 20;
214 static uint32_t dyn_keepalive_period = 5;
215 static uint32_t dyn_keepalive = 1; /* do send keepalives */
217 static uint32_t dyn_count; /* # of dynamic rules */
218 static uint32_t dyn_max = 4096; /* max # of dynamic rules */
220 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLTYPE_INT | CTLFLAG_RW,
221 &dyn_buckets, 0, ipfw_sysctl_dyn_buckets, "I", "Number of dyn. buckets");
222 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
223 &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
224 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
225 &dyn_count, 0, "Number of dyn. rules");
226 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
227 &dyn_max, 0, "Max number of dyn. rules");
228 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
229 &static_count, 0, "Number of static rules");
230 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
231 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
232 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
233 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
234 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime,
235 CTLTYPE_INT | CTLFLAG_RW, &dyn_fin_lifetime, 0, ipfw_sysctl_dyn_fin, "I",
236 "Lifetime of dyn. rules for fin");
237 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime,
238 CTLTYPE_INT | CTLFLAG_RW, &dyn_rst_lifetime, 0, ipfw_sysctl_dyn_rst, "I",
239 "Lifetime of dyn. rules for rst");
240 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
241 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
242 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
243 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
244 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
245 &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
247 #endif /* SYSCTL_NODE */
249 static struct ip_fw *ip_fw_default_rule;
251 static ip_fw_chk_t ipfw_chk;
254 ipfw_free_rule(struct ip_fw *rule)
256 KASSERT(rule->refcnt > 0, ("invalid refcnt %u\n", rule->refcnt));
257 atomic_subtract_int(&rule->refcnt, 1);
258 if (atomic_cmpset_int(&rule->refcnt, 0, 1)) {
266 ipfw_unref_rule(void *priv)
268 ipfw_free_rule(priv);
270 atomic_subtract_int(&ipfw_refcnt, 1);
275 ipfw_ref_rule(struct ip_fw *rule)
278 atomic_add_int(&ipfw_refcnt, 1);
280 atomic_add_int(&rule->refcnt, 1);
284 * This macro maps an ip pointer into a layer3 header pointer of type T
286 #define L3HDR(T, ip) ((T *)((uint32_t *)(ip) + (ip)->ip_hl))
289 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
291 int type = L3HDR(struct icmp,ip)->icmp_type;
293 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1 << type)));
296 #define TT ((1 << ICMP_ECHO) | \
297 (1 << ICMP_ROUTERSOLICIT) | \
298 (1 << ICMP_TSTAMP) | \
303 is_icmp_query(struct ip *ip)
305 int type = L3HDR(struct icmp, ip)->icmp_type;
307 return (type <= ICMP_MAXTYPE && (TT & (1 << type)));
313 * The following checks use two arrays of 8 or 16 bits to store the
314 * bits that we want set or clear, respectively. They are in the
315 * low and high half of cmd->arg1 or cmd->d[0].
317 * We scan options and store the bits we find set. We succeed if
319 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
321 * The code is sometimes optimized not to store additional variables.
325 flags_match(ipfw_insn *cmd, uint8_t bits)
330 if (((cmd->arg1 & 0xff) & bits) != 0)
331 return 0; /* some bits we want set were clear */
333 want_clear = (cmd->arg1 >> 8) & 0xff;
334 if ((want_clear & bits) != want_clear)
335 return 0; /* some bits we want clear were set */
340 ipopts_match(struct ip *ip, ipfw_insn *cmd)
342 int optlen, bits = 0;
343 u_char *cp = (u_char *)(ip + 1);
344 int x = (ip->ip_hl << 2) - sizeof(struct ip);
346 for (; x > 0; x -= optlen, cp += optlen) {
347 int opt = cp[IPOPT_OPTVAL];
349 if (opt == IPOPT_EOL)
352 if (opt == IPOPT_NOP) {
355 optlen = cp[IPOPT_OLEN];
356 if (optlen <= 0 || optlen > x)
357 return 0; /* invalid or truncated */
362 bits |= IP_FW_IPOPT_LSRR;
366 bits |= IP_FW_IPOPT_SSRR;
370 bits |= IP_FW_IPOPT_RR;
374 bits |= IP_FW_IPOPT_TS;
381 return (flags_match(cmd, bits));
385 tcpopts_match(struct ip *ip, ipfw_insn *cmd)
387 int optlen, bits = 0;
388 struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
389 u_char *cp = (u_char *)(tcp + 1);
390 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
392 for (; x > 0; x -= optlen, cp += optlen) {
395 if (opt == TCPOPT_EOL)
398 if (opt == TCPOPT_NOP) {
408 bits |= IP_FW_TCPOPT_MSS;
412 bits |= IP_FW_TCPOPT_WINDOW;
415 case TCPOPT_SACK_PERMITTED:
417 bits |= IP_FW_TCPOPT_SACK;
420 case TCPOPT_TIMESTAMP:
421 bits |= IP_FW_TCPOPT_TS;
427 bits |= IP_FW_TCPOPT_CC;
434 return (flags_match(cmd, bits));
438 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
440 if (ifp == NULL) /* no iface with this packet, match fails */
443 /* Check by name or by IP address */
444 if (cmd->name[0] != '\0') { /* match by name */
447 if (kfnmatch(cmd->name, ifp->if_xname, 0) == 0)
450 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
454 struct ifaddr_container *ifac;
456 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
457 struct ifaddr *ia = ifac->ifa;
459 if (ia->ifa_addr == NULL)
461 if (ia->ifa_addr->sa_family != AF_INET)
463 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
464 (ia->ifa_addr))->sin_addr.s_addr)
465 return(1); /* match */
468 return(0); /* no match, fail ... */
471 static uint64_t norule_counter; /* counter for ipfw_log(NULL...) */
473 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
476 * We enter here when we have a rule with O_LOG.
477 * XXX this function alone takes about 2Kbytes of code!
480 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
481 struct mbuf *m, struct ifnet *oif)
484 int limit_reached = 0;
485 char action2[40], proto[48], fragment[28];
490 if (f == NULL) { /* bogus pkt */
491 if (verbose_limit != 0 && norule_counter >= verbose_limit)
494 if (norule_counter == verbose_limit)
495 limit_reached = verbose_limit;
497 } else { /* O_LOG is the first action, find the real one */
498 ipfw_insn *cmd = ACTION_PTR(f);
499 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
501 if (l->max_log != 0 && l->log_left == 0)
504 if (l->log_left == 0)
505 limit_reached = l->max_log;
506 cmd += F_LEN(cmd); /* point to first action */
507 if (cmd->opcode == O_PROB)
511 switch (cmd->opcode) {
517 if (cmd->arg1==ICMP_REJECT_RST) {
519 } else if (cmd->arg1==ICMP_UNREACH_HOST) {
522 ksnprintf(SNPARGS(action2, 0), "Unreach %d",
536 ksnprintf(SNPARGS(action2, 0), "Divert %d", cmd->arg1);
540 ksnprintf(SNPARGS(action2, 0), "Tee %d", cmd->arg1);
544 ksnprintf(SNPARGS(action2, 0), "SkipTo %d", cmd->arg1);
548 ksnprintf(SNPARGS(action2, 0), "Pipe %d", cmd->arg1);
552 ksnprintf(SNPARGS(action2, 0), "Queue %d", cmd->arg1);
557 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
560 len = ksnprintf(SNPARGS(action2, 0),
562 inet_ntoa(sa->sa.sin_addr));
563 if (sa->sa.sin_port) {
564 ksnprintf(SNPARGS(action2, len), ":%d",
576 if (hlen == 0) { /* non-ip */
577 ksnprintf(SNPARGS(proto, 0), "MAC");
579 struct ip *ip = mtod(m, struct ip *);
580 /* these three are all aliases to the same thing */
581 struct icmp *const icmp = L3HDR(struct icmp, ip);
582 struct tcphdr *const tcp = (struct tcphdr *)icmp;
583 struct udphdr *const udp = (struct udphdr *)icmp;
585 int ip_off, offset, ip_len;
588 if (eh != NULL) { /* layer 2 packets are as on the wire */
589 ip_off = ntohs(ip->ip_off);
590 ip_len = ntohs(ip->ip_len);
595 offset = ip_off & IP_OFFMASK;
598 len = ksnprintf(SNPARGS(proto, 0), "TCP %s",
599 inet_ntoa(ip->ip_src));
601 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
602 ntohs(tcp->th_sport),
603 inet_ntoa(ip->ip_dst),
604 ntohs(tcp->th_dport));
606 ksnprintf(SNPARGS(proto, len), " %s",
607 inet_ntoa(ip->ip_dst));
612 len = ksnprintf(SNPARGS(proto, 0), "UDP %s",
613 inet_ntoa(ip->ip_src));
615 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
616 ntohs(udp->uh_sport),
617 inet_ntoa(ip->ip_dst),
618 ntohs(udp->uh_dport));
620 ksnprintf(SNPARGS(proto, len), " %s",
621 inet_ntoa(ip->ip_dst));
627 len = ksnprintf(SNPARGS(proto, 0),
632 len = ksnprintf(SNPARGS(proto, 0), "ICMP ");
634 len += ksnprintf(SNPARGS(proto, len), "%s",
635 inet_ntoa(ip->ip_src));
636 ksnprintf(SNPARGS(proto, len), " %s",
637 inet_ntoa(ip->ip_dst));
641 len = ksnprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
642 inet_ntoa(ip->ip_src));
643 ksnprintf(SNPARGS(proto, len), " %s",
644 inet_ntoa(ip->ip_dst));
648 if (ip_off & (IP_MF | IP_OFFMASK)) {
649 ksnprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
650 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
651 offset << 3, (ip_off & IP_MF) ? "+" : "");
655 if (oif || m->m_pkthdr.rcvif) {
656 log(LOG_SECURITY | LOG_INFO,
657 "ipfw: %d %s %s %s via %s%s\n",
659 action, proto, oif ? "out" : "in",
660 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
663 log(LOG_SECURITY | LOG_INFO,
664 "ipfw: %d %s %s [no if info]%s\n",
666 action, proto, fragment);
670 log(LOG_SECURITY | LOG_NOTICE,
671 "ipfw: limit %d reached on entry %d\n",
672 limit_reached, f ? f->rulenum : -1);
679 * IMPORTANT: the hash function for dynamic rules must be commutative
680 * in source and destination (ip,port), because rules are bidirectional
681 * and we want to find both in the same bucket.
684 hash_packet(struct ipfw_flow_id *id)
688 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
689 i &= (curr_dyn_buckets - 1);
694 * unlink a dynamic rule from a chain. prev is a pointer to
695 * the previous one, q is a pointer to the rule to delete,
696 * head is a pointer to the head of the queue.
697 * Modifies q and potentially also head.
699 #define UNLINK_DYN_RULE(prev, head, q) \
701 ipfw_dyn_rule *old_q = q; \
703 /* remove a refcount to the parent */ \
704 if (q->dyn_type == O_LIMIT) \
705 q->parent->count--; \
706 DEB(kprintf("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n", \
707 (q->id.src_ip), (q->id.src_port), \
708 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \
710 prev->next = q = q->next; \
712 head = q = q->next; \
713 KASSERT(dyn_count > 0, ("invalid dyn count %u\n", dyn_count)); \
715 kfree(old_q, M_IPFW); \
718 #define TIME_LEQ(a, b) ((int)((a) - (b)) <= 0)
721 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
723 * If keep_me == NULL, rules are deleted even if not expired,
724 * otherwise only expired rules are removed.
726 * The value of the second parameter is also used to point to identify
727 * a rule we absolutely do not want to remove (e.g. because we are
728 * holding a reference to it -- this is the case with O_LIMIT_PARENT
729 * rules). The pointer is only used for comparison, so any non-null
733 remove_dyn_rule_locked(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
735 static uint32_t last_remove = 0; /* XXX */
737 #define FORCE (keep_me == NULL)
739 ipfw_dyn_rule *prev, *q;
740 int i, pass = 0, max_pass = 0, unlinked = 0;
742 if (ipfw_dyn_v == NULL || dyn_count == 0)
744 /* do not expire more than once per second, it is useless */
745 if (!FORCE && last_remove == time_second)
747 last_remove = time_second;
750 * because O_LIMIT refer to parent rules, during the first pass only
751 * remove child and mark any pending LIMIT_PARENT, and remove
752 * them in a second pass.
755 for (i = 0; i < curr_dyn_buckets; i++) {
756 for (prev = NULL, q = ipfw_dyn_v[i]; q;) {
758 * Logic can become complex here, so we split tests.
762 if (rule != NULL && rule != q->rule)
763 goto next; /* not the one we are looking for */
764 if (q->dyn_type == O_LIMIT_PARENT) {
766 * handle parent in the second pass,
767 * record we need one.
772 if (FORCE && q->count != 0) {
773 /* XXX should not happen! */
774 kprintf("OUCH! cannot remove rule, "
775 "count %d\n", q->count);
778 if (!FORCE && !TIME_LEQ(q->expire, time_second))
782 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
789 if (pass++ < max_pass)
799 remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
801 lockmgr(&dyn_lock, LK_EXCLUSIVE);
802 remove_dyn_rule_locked(rule, keep_me);
803 lockmgr(&dyn_lock, LK_RELEASE);
807 * lookup a dynamic rule.
809 static ipfw_dyn_rule *
810 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
814 * stateful ipfw extensions.
815 * Lookup into dynamic session queue
817 #define MATCH_REVERSE 0
818 #define MATCH_FORWARD 1
820 #define MATCH_UNKNOWN 3
821 int i, dir = MATCH_NONE;
822 ipfw_dyn_rule *prev, *q=NULL;
824 if (ipfw_dyn_v == NULL)
825 goto done; /* not found */
827 i = hash_packet(pkt);
828 for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) {
829 if (q->dyn_type == O_LIMIT_PARENT)
832 if (TIME_LEQ(q->expire, time_second)) {
834 * Entry expired; skip.
835 * Let ipfw_tick() take care of it
840 if (pkt->proto == q->id.proto) {
841 if (pkt->src_ip == q->id.src_ip &&
842 pkt->dst_ip == q->id.dst_ip &&
843 pkt->src_port == q->id.src_port &&
844 pkt->dst_port == q->id.dst_port) {
848 if (pkt->src_ip == q->id.dst_ip &&
849 pkt->dst_ip == q->id.src_ip &&
850 pkt->src_port == q->id.dst_port &&
851 pkt->dst_port == q->id.src_port) {
861 goto done; /* q = NULL, not found */
863 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
864 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
866 #define BOTH_SYN (TH_SYN | (TH_SYN << 8))
867 #define BOTH_FIN (TH_FIN | (TH_FIN << 8))
869 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
871 case TH_SYN: /* opening */
872 q->expire = time_second + dyn_syn_lifetime;
875 case BOTH_SYN: /* move to established */
876 case BOTH_SYN | TH_FIN : /* one side tries to close */
877 case BOTH_SYN | (TH_FIN << 8) :
879 uint32_t ack = ntohl(tcp->th_ack);
881 #define _SEQ_GE(a, b) ((int)(a) - (int)(b) >= 0)
883 if (dir == MATCH_FORWARD) {
884 if (q->ack_fwd == 0 ||
885 _SEQ_GE(ack, q->ack_fwd))
887 else /* ignore out-of-sequence */
890 if (q->ack_rev == 0 ||
891 _SEQ_GE(ack, q->ack_rev))
893 else /* ignore out-of-sequence */
898 q->expire = time_second + dyn_ack_lifetime;
901 case BOTH_SYN | BOTH_FIN: /* both sides closed */
902 KKASSERT(dyn_fin_lifetime < dyn_keepalive_period);
903 q->expire = time_second + dyn_fin_lifetime;
909 * reset or some invalid combination, but can also
910 * occur if we use keep-state the wrong way.
912 if ((q->state & ((TH_RST << 8) | TH_RST)) == 0)
913 kprintf("invalid state: 0x%x\n", q->state);
915 KKASSERT(dyn_rst_lifetime < dyn_keepalive_period);
916 q->expire = time_second + dyn_rst_lifetime;
919 } else if (pkt->proto == IPPROTO_UDP) {
920 q->expire = time_second + dyn_udp_lifetime;
922 /* other protocols */
923 q->expire = time_second + dyn_short_lifetime;
927 *match_direction = dir;
931 static struct ip_fw *
932 lookup_rule(struct ipfw_flow_id *pkt, int *match_direction, struct tcphdr *tcp,
933 uint16_t len, int *deny)
935 struct ip_fw *rule = NULL;
942 lockmgr(&dyn_lock, LK_SHARED);
944 if (static_gen != gen) {
946 * Static rules had been change when we were waiting
947 * for the dynamic hash table lock; deny this packet,
948 * since it is _not_ known whether it is safe to keep
949 * iterating the static rules.
955 q = lookup_dyn_rule(pkt, match_direction, tcp);
966 lockmgr(&dyn_lock, LK_RELEASE);
971 realloc_dynamic_table(void)
973 ipfw_dyn_rule **old_dyn_v;
974 uint32_t old_curr_dyn_buckets;
976 KASSERT(dyn_buckets <= 65536 && (dyn_buckets & (dyn_buckets - 1)) == 0,
977 ("invalid dyn_buckets %d\n", dyn_buckets));
979 /* Save the current buckets array for later error recovery */
980 old_dyn_v = ipfw_dyn_v;
981 old_curr_dyn_buckets = curr_dyn_buckets;
983 curr_dyn_buckets = dyn_buckets;
985 ipfw_dyn_v = kmalloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
986 M_IPFW, M_NOWAIT | M_ZERO);
987 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
990 curr_dyn_buckets /= 2;
991 if (curr_dyn_buckets <= old_curr_dyn_buckets &&
994 * Don't try allocating smaller buckets array, reuse
995 * the old one, which alreay contains enough buckets
1001 if (ipfw_dyn_v != NULL) {
1002 if (old_dyn_v != NULL)
1003 kfree(old_dyn_v, M_IPFW);
1005 /* Allocation failed, restore old buckets array */
1006 ipfw_dyn_v = old_dyn_v;
1007 curr_dyn_buckets = old_curr_dyn_buckets;
1010 if (ipfw_dyn_v != NULL)
1015 * Install state of type 'type' for a dynamic session.
1016 * The hash table contains two type of rules:
1017 * - regular rules (O_KEEP_STATE)
1018 * - rules for sessions with limited number of sess per user
1019 * (O_LIMIT). When they are created, the parent is
1020 * increased by 1, and decreased on delete. In this case,
1021 * the third parameter is the parent rule and not the chain.
1022 * - "parent" rules for the above (O_LIMIT_PARENT).
1024 static ipfw_dyn_rule *
1025 add_dyn_rule(struct ipfw_flow_id *id, uint8_t dyn_type, struct ip_fw *rule)
1030 if (ipfw_dyn_v == NULL ||
1031 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1032 realloc_dynamic_table();
1033 if (ipfw_dyn_v == NULL)
1034 return NULL; /* failed ! */
1036 i = hash_packet(id);
1038 r = kmalloc(sizeof(*r), M_IPFW, M_NOWAIT | M_ZERO);
1040 kprintf ("sorry cannot allocate state\n");
1044 /* increase refcount on parent, and set pointer */
1045 if (dyn_type == O_LIMIT) {
1046 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1048 if (parent->dyn_type != O_LIMIT_PARENT)
1049 panic("invalid parent");
1052 rule = parent->rule;
1056 r->expire = time_second + dyn_syn_lifetime;
1058 r->dyn_type = dyn_type;
1059 r->pcnt = r->bcnt = 0;
1063 r->next = ipfw_dyn_v[i];
1067 DEB(kprintf("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1069 (r->id.src_ip), (r->id.src_port),
1070 (r->id.dst_ip), (r->id.dst_port),
1076 * lookup dynamic parent rule using pkt and rule as search keys.
1077 * If the lookup fails, then install one.
1079 static ipfw_dyn_rule *
1080 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1086 i = hash_packet(pkt);
1087 for (q = ipfw_dyn_v[i]; q != NULL; q = q->next) {
1088 if (q->dyn_type == O_LIMIT_PARENT &&
1090 pkt->proto == q->id.proto &&
1091 pkt->src_ip == q->id.src_ip &&
1092 pkt->dst_ip == q->id.dst_ip &&
1093 pkt->src_port == q->id.src_port &&
1094 pkt->dst_port == q->id.dst_port) {
1095 q->expire = time_second + dyn_short_lifetime;
1096 DEB(kprintf("lookup_dyn_parent found 0x%p\n",q);)
1101 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1105 * Install dynamic state for rule type cmd->o.opcode
1107 * Returns 1 (failure) if state is not installed because of errors or because
1108 * session limitations are enforced.
1111 install_state_locked(struct ip_fw *rule, ipfw_insn_limit *cmd,
1112 struct ip_fw_args *args)
1114 static int last_log; /* XXX */
1118 DEB(kprintf("-- install state type %d 0x%08x %u -> 0x%08x %u\n",
1120 (args->f_id.src_ip), (args->f_id.src_port),
1121 (args->f_id.dst_ip), (args->f_id.dst_port) );)
1123 q = lookup_dyn_rule(&args->f_id, NULL, NULL);
1124 if (q != NULL) { /* should never occur */
1125 if (last_log != time_second) {
1126 last_log = time_second;
1127 kprintf(" install_state: entry already present, done\n");
1132 if (dyn_count >= dyn_max) {
1134 * Run out of slots, try to remove any expired rule.
1136 remove_dyn_rule_locked(NULL, (ipfw_dyn_rule *)1);
1137 if (dyn_count >= dyn_max) {
1138 if (last_log != time_second) {
1139 last_log = time_second;
1140 kprintf("install_state: "
1141 "Too many dynamic rules\n");
1143 return 1; /* cannot install, notify caller */
1147 switch (cmd->o.opcode) {
1148 case O_KEEP_STATE: /* bidir rule */
1149 if (add_dyn_rule(&args->f_id, O_KEEP_STATE, rule) == NULL)
1153 case O_LIMIT: /* limit number of sessions */
1155 uint16_t limit_mask = cmd->limit_mask;
1156 struct ipfw_flow_id id;
1157 ipfw_dyn_rule *parent;
1159 DEB(kprintf("installing dyn-limit rule %d\n",
1162 id.dst_ip = id.src_ip = 0;
1163 id.dst_port = id.src_port = 0;
1164 id.proto = args->f_id.proto;
1166 if (limit_mask & DYN_SRC_ADDR)
1167 id.src_ip = args->f_id.src_ip;
1168 if (limit_mask & DYN_DST_ADDR)
1169 id.dst_ip = args->f_id.dst_ip;
1170 if (limit_mask & DYN_SRC_PORT)
1171 id.src_port = args->f_id.src_port;
1172 if (limit_mask & DYN_DST_PORT)
1173 id.dst_port = args->f_id.dst_port;
1175 parent = lookup_dyn_parent(&id, rule);
1176 if (parent == NULL) {
1177 kprintf("add parent failed\n");
1181 if (parent->count >= cmd->conn_limit) {
1183 * See if we can remove some expired rule.
1185 remove_dyn_rule_locked(rule, parent);
1186 if (parent->count >= cmd->conn_limit) {
1188 last_log != time_second) {
1189 last_log = time_second;
1190 log(LOG_SECURITY | LOG_DEBUG,
1192 "too many entries\n");
1197 if (add_dyn_rule(&args->f_id, O_LIMIT,
1198 (struct ip_fw *)parent) == NULL)
1203 kprintf("unknown dynamic rule type %u\n", cmd->o.opcode);
1206 lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1211 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1212 struct ip_fw_args *args, int *deny)
1220 lockmgr(&dyn_lock, LK_EXCLUSIVE);
1221 if (static_gen != gen) {
1222 /* See the comment in lookup_rule() */
1225 ret = install_state_locked(rule, cmd, args);
1227 lockmgr(&dyn_lock, LK_RELEASE);
1233 * Transmit a TCP packet, containing either a RST or a keepalive.
1234 * When flags & TH_RST, we are sending a RST packet, because of a
1235 * "reset" action matched the packet.
1236 * Otherwise we are sending a keepalive, and flags & TH_
1239 send_pkt(struct ipfw_flow_id *id, uint32_t seq, uint32_t ack, int flags)
1244 struct route sro; /* fake route */
1246 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1249 m->m_pkthdr.rcvif = NULL;
1250 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1251 m->m_data += max_linkhdr;
1253 ip = mtod(m, struct ip *);
1254 bzero(ip, m->m_len);
1255 tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1256 ip->ip_p = IPPROTO_TCP;
1260 * Assume we are sending a RST (or a keepalive in the reverse
1261 * direction), swap src and destination addresses and ports.
1263 ip->ip_src.s_addr = htonl(id->dst_ip);
1264 ip->ip_dst.s_addr = htonl(id->src_ip);
1265 tcp->th_sport = htons(id->dst_port);
1266 tcp->th_dport = htons(id->src_port);
1267 if (flags & TH_RST) { /* we are sending a RST */
1268 if (flags & TH_ACK) {
1269 tcp->th_seq = htonl(ack);
1270 tcp->th_ack = htonl(0);
1271 tcp->th_flags = TH_RST;
1275 tcp->th_seq = htonl(0);
1276 tcp->th_ack = htonl(seq);
1277 tcp->th_flags = TH_RST | TH_ACK;
1281 * We are sending a keepalive. flags & TH_SYN determines
1282 * the direction, forward if set, reverse if clear.
1283 * NOTE: seq and ack are always assumed to be correct
1284 * as set by the caller. This may be confusing...
1286 if (flags & TH_SYN) {
1288 * we have to rewrite the correct addresses!
1290 ip->ip_dst.s_addr = htonl(id->dst_ip);
1291 ip->ip_src.s_addr = htonl(id->src_ip);
1292 tcp->th_dport = htons(id->dst_port);
1293 tcp->th_sport = htons(id->src_port);
1295 tcp->th_seq = htonl(seq);
1296 tcp->th_ack = htonl(ack);
1297 tcp->th_flags = TH_ACK;
1301 * set ip_len to the payload size so we can compute
1302 * the tcp checksum on the pseudoheader
1303 * XXX check this, could save a couple of words ?
1305 ip->ip_len = htons(sizeof(struct tcphdr));
1306 tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1309 * now fill fields left out earlier
1311 ip->ip_ttl = ip_defttl;
1312 ip->ip_len = m->m_pkthdr.len;
1314 bzero(&sro, sizeof(sro));
1315 ip_rtaddr(ip->ip_dst, &sro);
1317 m->m_pkthdr.fw_flags |= IPFW_MBUF_GENERATED;
1318 ip_output(m, NULL, &sro, 0, NULL, NULL);
1324 * sends a reject message, consuming the mbuf passed as an argument.
1327 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1329 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1330 /* We need the IP header in host order for icmp_error(). */
1331 if (args->eh != NULL) {
1332 struct ip *ip = mtod(args->m, struct ip *);
1334 ip->ip_len = ntohs(ip->ip_len);
1335 ip->ip_off = ntohs(ip->ip_off);
1337 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1338 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1339 struct tcphdr *const tcp =
1340 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1342 if ((tcp->th_flags & TH_RST) == 0) {
1343 send_pkt(&args->f_id, ntohl(tcp->th_seq),
1344 ntohl(tcp->th_ack), tcp->th_flags | TH_RST);
1355 * Given an ip_fw *, lookup_next_rule will return a pointer
1356 * to the next rule, which can be either the jump
1357 * target (for skipto instructions) or the next one in the list (in
1358 * all other cases including a missing jump target).
1359 * The result is also written in the "next_rule" field of the rule.
1360 * Backward jumps are not allowed, so start looking from the next
1363 * This never returns NULL -- in case we do not have an exact match,
1364 * the next rule is returned. When the ruleset is changed,
1365 * pointers are flushed so we are always correct.
1368 static struct ip_fw *
1369 lookup_next_rule(struct ip_fw *me)
1371 struct ip_fw *rule = NULL;
1374 /* look for action, in case it is a skipto */
1375 cmd = ACTION_PTR(me);
1376 if (cmd->opcode == O_LOG)
1378 if (cmd->opcode == O_SKIPTO) {
1379 for (rule = me->next; rule; rule = rule->next) {
1380 if (rule->rulenum >= cmd->arg1)
1384 if (rule == NULL) /* failure or not a skipto */
1386 me->next_rule = rule;
1391 * The main check routine for the firewall.
1393 * All arguments are in args so we can modify them and return them
1394 * back to the caller.
1398 * args->m (in/out) The packet; we set to NULL when/if we nuke it.
1399 * Starts with the IP header.
1400 * args->eh (in) Mac header if present, or NULL for layer3 packet.
1401 * args->oif Outgoing interface, or NULL if packet is incoming.
1402 * The incoming interface is in the mbuf. (in)
1404 * args->rule Pointer to the last matching rule (in/out)
1405 * args->next_hop Socket we are forwarding to (out).
1406 * args->f_id Addresses grabbed from the packet (out)
1410 * IP_FW_PORT_DENY_FLAG the packet must be dropped.
1411 * 0 The packet is to be accepted and routed normally OR
1412 * the packet was denied/rejected and has been dropped;
1413 * in the latter case, *m is equal to NULL upon return.
1414 * port Divert the packet to port, with these caveats:
1416 * - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead
1417 * of diverting it (ie, 'ipfw tee').
1419 * - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower
1420 * 16 bits as a dummynet pipe number instead of diverting
1424 ipfw_chk(struct ip_fw_args *args)
1427 * Local variables hold state during the processing of a packet.
1429 * IMPORTANT NOTE: to speed up the processing of rules, there
1430 * are some assumption on the values of the variables, which
1431 * are documented here. Should you change them, please check
1432 * the implementation of the various instructions to make sure
1433 * that they still work.
1435 * args->eh The MAC header. It is non-null for a layer2
1436 * packet, it is NULL for a layer-3 packet.
1438 * m | args->m Pointer to the mbuf, as received from the caller.
1439 * It may change if ipfw_chk() does an m_pullup, or if it
1440 * consumes the packet because it calls send_reject().
1441 * XXX This has to change, so that ipfw_chk() never modifies
1442 * or consumes the buffer.
1443 * ip is simply an alias of the value of m, and it is kept
1444 * in sync with it (the packet is supposed to start with
1447 struct mbuf *m = args->m;
1448 struct ip *ip = mtod(m, struct ip *);
1451 * oif | args->oif If NULL, ipfw_chk has been called on the
1452 * inbound path (ether_input, ip_input).
1453 * If non-NULL, ipfw_chk has been called on the outbound path
1454 * (ether_output, ip_output).
1456 struct ifnet *oif = args->oif;
1458 struct ip_fw *f = NULL; /* matching rule */
1463 * hlen The length of the IPv4 header.
1464 * hlen >0 means we have an IPv4 packet.
1466 u_int hlen = 0; /* hlen >0 means we have an IP pkt */
1469 * offset The offset of a fragment. offset != 0 means that
1470 * we have a fragment at this offset of an IPv4 packet.
1471 * offset == 0 means that (if this is an IPv4 packet)
1472 * this is the first or only fragment.
1477 * Local copies of addresses. They are only valid if we have
1480 * proto The protocol. Set to 0 for non-ip packets,
1481 * or to the protocol read from the packet otherwise.
1482 * proto != 0 means that we have an IPv4 packet.
1484 * src_port, dst_port port numbers, in HOST format. Only
1485 * valid for TCP and UDP packets.
1487 * src_ip, dst_ip ip addresses, in NETWORK format.
1488 * Only valid for IPv4 packets.
1491 uint16_t src_port = 0, dst_port = 0; /* NOTE: host format */
1492 struct in_addr src_ip, dst_ip; /* NOTE: network format */
1493 uint16_t ip_len = 0;
1494 int dyn_dir = MATCH_UNKNOWN;
1495 ipfw_dyn_rule *q = NULL;
1497 if (m->m_pkthdr.fw_flags & IPFW_MBUF_GENERATED)
1498 return 0; /* accept */
1500 * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1501 * MATCH_NONE when checked and not matched (q = NULL),
1502 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
1505 if (args->eh == NULL || /* layer 3 packet */
1506 (m->m_pkthdr.len >= sizeof(struct ip) &&
1507 ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1508 hlen = ip->ip_hl << 2;
1511 * Collect parameters into local variables for faster matching.
1513 if (hlen == 0) { /* do not grab addresses for non-ip pkts */
1514 proto = args->f_id.proto = 0; /* mark f_id invalid */
1515 goto after_ip_checks;
1518 proto = args->f_id.proto = ip->ip_p;
1519 src_ip = ip->ip_src;
1520 dst_ip = ip->ip_dst;
1521 if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1522 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1523 ip_len = ntohs(ip->ip_len);
1525 offset = ip->ip_off & IP_OFFMASK;
1526 ip_len = ip->ip_len;
1529 #define PULLUP_TO(len) \
1531 if (m->m_len < (len)) { \
1532 args->m = m = m_pullup(m, (len));\
1534 goto pullup_failed; \
1535 ip = mtod(m, struct ip *); \
1545 PULLUP_TO(hlen + sizeof(struct tcphdr));
1546 tcp = L3HDR(struct tcphdr, ip);
1547 dst_port = tcp->th_dport;
1548 src_port = tcp->th_sport;
1549 args->f_id.flags = tcp->th_flags;
1557 PULLUP_TO(hlen + sizeof(struct udphdr));
1558 udp = L3HDR(struct udphdr, ip);
1559 dst_port = udp->uh_dport;
1560 src_port = udp->uh_sport;
1565 PULLUP_TO(hlen + 4); /* type, code and checksum. */
1566 args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1576 args->f_id.src_ip = ntohl(src_ip.s_addr);
1577 args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1578 args->f_id.src_port = src_port = ntohs(src_port);
1579 args->f_id.dst_port = dst_port = ntohs(dst_port);
1584 * Packet has already been tagged. Look for the next rule
1585 * to restart processing.
1587 * If fw_one_pass != 0 then just accept it.
1588 * XXX should not happen here, but optimized out in
1594 /* This rule was deleted */
1595 if (args->rule->rule_flags & IPFW_RULE_F_INVALID)
1596 return IP_FW_PORT_DENY_FLAG;
1598 f = args->rule->next_rule;
1600 f = lookup_next_rule(args->rule);
1603 * Find the starting rule. It can be either the first
1604 * one, or the one after divert_rule if asked so.
1608 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1610 skipto = *(uint16_t *)m_tag_data(mtag);
1615 if (args->eh == NULL && skipto != 0) {
1616 if (skipto >= IPFW_DEFAULT_RULE)
1617 return(IP_FW_PORT_DENY_FLAG); /* invalid */
1618 while (f && f->rulenum <= skipto)
1620 if (f == NULL) /* drop packet */
1621 return(IP_FW_PORT_DENY_FLAG);
1624 if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
1625 m_tag_delete(m, mtag);
1628 * Now scan the rules, and parse microinstructions for each rule.
1630 for (; f; f = f->next) {
1633 int skip_or; /* skip rest of OR block */
1636 if (set_disable & (1 << f->set))
1640 for (l = f->cmd_len, cmd = f->cmd; l > 0;
1641 l -= cmdlen, cmd += cmdlen) {
1645 * check_body is a jump target used when we find a
1646 * CHECK_STATE, and need to jump to the body of
1651 cmdlen = F_LEN(cmd);
1653 * An OR block (insn_1 || .. || insn_n) has the
1654 * F_OR bit set in all but the last instruction.
1655 * The first match will set "skip_or", and cause
1656 * the following instructions to be skipped until
1657 * past the one with the F_OR bit clear.
1659 if (skip_or) { /* skip this instruction */
1660 if ((cmd->len & F_OR) == 0)
1661 skip_or = 0; /* next one is good */
1664 match = 0; /* set to 1 if we succeed */
1666 switch (cmd->opcode) {
1668 * The first set of opcodes compares the packet's
1669 * fields with some pattern, setting 'match' if a
1670 * match is found. At the end of the loop there is
1671 * logic to deal with F_NOT and F_OR flags associated
1679 kprintf("ipfw: opcode %d unimplemented\n",
1686 * We only check offset == 0 && proto != 0,
1687 * as this ensures that we have an IPv4
1688 * packet with the ports info.
1693 struct inpcbinfo *pi;
1697 if (proto == IPPROTO_TCP) {
1699 pi = &tcbinfo[mycpu->gd_cpuid];
1700 } else if (proto == IPPROTO_UDP) {
1707 in_pcblookup_hash(pi,
1708 dst_ip, htons(dst_port),
1709 src_ip, htons(src_port),
1711 in_pcblookup_hash(pi,
1712 src_ip, htons(src_port),
1713 dst_ip, htons(dst_port),
1716 if (pcb == NULL || pcb->inp_socket == NULL)
1719 if (cmd->opcode == O_UID) {
1720 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1722 !socheckuid(pcb->inp_socket,
1723 (uid_t)((ipfw_insn_u32 *)cmd)->d[0]);
1726 match = groupmember(
1727 (uid_t)((ipfw_insn_u32 *)cmd)->d[0],
1728 pcb->inp_socket->so_cred);
1734 match = iface_match(m->m_pkthdr.rcvif,
1735 (ipfw_insn_if *)cmd);
1739 match = iface_match(oif, (ipfw_insn_if *)cmd);
1743 match = iface_match(oif ? oif :
1744 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1748 if (args->eh != NULL) { /* have MAC header */
1749 uint32_t *want = (uint32_t *)
1750 ((ipfw_insn_mac *)cmd)->addr;
1751 uint32_t *mask = (uint32_t *)
1752 ((ipfw_insn_mac *)cmd)->mask;
1753 uint32_t *hdr = (uint32_t *)args->eh;
1756 (want[0] == (hdr[0] & mask[0]) &&
1757 want[1] == (hdr[1] & mask[1]) &&
1758 want[2] == (hdr[2] & mask[2]));
1763 if (args->eh != NULL) {
1765 ntohs(args->eh->ether_type);
1767 ((ipfw_insn_u16 *)cmd)->ports;
1770 /* Special vlan handling */
1771 if (m->m_flags & M_VLANTAG)
1774 for (i = cmdlen - 1; !match && i > 0;
1777 (t >= p[0] && t <= p[1]);
1783 match = (hlen > 0 && offset != 0);
1786 case O_IN: /* "out" is "not in" */
1787 match = (oif == NULL);
1791 match = (args->eh != NULL);
1796 * We do not allow an arg of 0 so the
1797 * check of "proto" only suffices.
1799 match = (proto == cmd->arg1);
1803 match = (hlen > 0 &&
1804 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1809 match = (hlen > 0 &&
1810 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1812 ((ipfw_insn_ip *)cmd)->mask.s_addr));
1819 tif = INADDR_TO_IFP(&src_ip);
1820 match = (tif != NULL);
1827 uint32_t *d = (uint32_t *)(cmd + 1);
1829 cmd->opcode == O_IP_DST_SET ?
1835 addr -= d[0]; /* subtract base */
1837 (addr < cmd->arg1) &&
1838 (d[1 + (addr >> 5)] &
1839 (1 << (addr & 0x1f)));
1844 match = (hlen > 0 &&
1845 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1850 match = (hlen > 0) &&
1851 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
1853 ((ipfw_insn_ip *)cmd)->mask.s_addr));
1860 tif = INADDR_TO_IFP(&dst_ip);
1861 match = (tif != NULL);
1868 * offset == 0 && proto != 0 is enough
1869 * to guarantee that we have an IPv4
1870 * packet with port info.
1872 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
1875 (cmd->opcode == O_IP_SRCPORT) ?
1876 src_port : dst_port ;
1878 ((ipfw_insn_u16 *)cmd)->ports;
1881 for (i = cmdlen - 1; !match && i > 0;
1884 (x >= p[0] && x <= p[1]);
1890 match = (offset == 0 && proto==IPPROTO_ICMP &&
1891 icmptype_match(ip, (ipfw_insn_u32 *)cmd));
1895 match = (hlen > 0 && ipopts_match(ip, cmd));
1899 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
1903 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl);
1907 match = (hlen > 0 &&
1908 cmd->arg1 == ntohs(ip->ip_id));
1912 match = (hlen > 0 && cmd->arg1 == ip_len);
1915 case O_IPPRECEDENCE:
1916 match = (hlen > 0 &&
1917 (cmd->arg1 == (ip->ip_tos & 0xe0)));
1921 match = (hlen > 0 &&
1922 flags_match(cmd, ip->ip_tos));
1926 match = (proto == IPPROTO_TCP && offset == 0 &&
1928 L3HDR(struct tcphdr,ip)->th_flags));
1932 match = (proto == IPPROTO_TCP && offset == 0 &&
1933 tcpopts_match(ip, cmd));
1937 match = (proto == IPPROTO_TCP && offset == 0 &&
1938 ((ipfw_insn_u32 *)cmd)->d[0] ==
1939 L3HDR(struct tcphdr,ip)->th_seq);
1943 match = (proto == IPPROTO_TCP && offset == 0 &&
1944 ((ipfw_insn_u32 *)cmd)->d[0] ==
1945 L3HDR(struct tcphdr,ip)->th_ack);
1949 match = (proto == IPPROTO_TCP && offset == 0 &&
1951 L3HDR(struct tcphdr,ip)->th_win);
1955 /* reject packets which have SYN only */
1956 /* XXX should i also check for TH_ACK ? */
1957 match = (proto == IPPROTO_TCP && offset == 0 &&
1958 (L3HDR(struct tcphdr,ip)->th_flags &
1959 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
1964 ipfw_log(f, hlen, args->eh, m, oif);
1969 match = (krandom() <
1970 ((ipfw_insn_u32 *)cmd)->d[0]);
1974 * The second set of opcodes represents 'actions',
1975 * i.e. the terminal part of a rule once the packet
1976 * matches all previous patterns.
1977 * Typically there is only one action for each rule,
1978 * and the opcode is stored at the end of the rule
1979 * (but there are exceptions -- see below).
1981 * In general, here we set retval and terminate the
1982 * outer loop (would be a 'break 3' in some language,
1983 * but we need to do a 'goto done').
1986 * O_COUNT and O_SKIPTO actions:
1987 * instead of terminating, we jump to the next rule
1988 * ('goto next_rule', equivalent to a 'break 2'),
1989 * or to the SKIPTO target ('goto again' after
1990 * having set f, cmd and l), respectively.
1992 * O_LIMIT and O_KEEP_STATE: these opcodes are
1993 * not real 'actions', and are stored right
1994 * before the 'action' part of the rule.
1995 * These opcodes try to install an entry in the
1996 * state tables; if successful, we continue with
1997 * the next opcode (match=1; break;), otherwise
1998 * the packet * must be dropped
1999 * ('goto done' after setting retval);
2001 * O_PROBE_STATE and O_CHECK_STATE: these opcodes
2002 * cause a lookup of the state table, and a jump
2003 * to the 'action' part of the parent rule
2004 * ('goto check_body') if an entry is found, or
2005 * (CHECK_STATE only) a jump to the next rule if
2006 * the entry is not found ('goto next_rule').
2007 * The result of the lookup is cached to make
2008 * further instances of these opcodes are
2013 if (install_state(f,
2014 (ipfw_insn_limit *)cmd, args, &deny)) {
2016 return IP_FW_PORT_DENY_FLAG;
2018 retval = IP_FW_PORT_DENY_FLAG;
2019 goto done; /* error/limit violation */
2022 return IP_FW_PORT_DENY_FLAG;
2029 * dynamic rules are checked at the first
2030 * keep-state or check-state occurrence,
2031 * with the result being stored in dyn_dir.
2032 * The compiler introduces a PROBE_STATE
2033 * instruction for us when we have a
2034 * KEEP_STATE (because PROBE_STATE needs
2037 if (dyn_dir == MATCH_UNKNOWN) {
2038 struct ip_fw *dyn_f;
2040 dyn_f = lookup_rule(&args->f_id,
2042 proto == IPPROTO_TCP ?
2043 L3HDR(struct tcphdr, ip) : NULL,
2046 return IP_FW_PORT_DENY_FLAG;
2047 if (dyn_f != NULL) {
2049 * Found a rule from a dynamic
2050 * entry; jump to the 'action'
2054 cmd = ACTION_PTR(f);
2055 l = f->cmd_len - f->act_ofs;
2060 * Dynamic entry not found. If CHECK_STATE,
2061 * skip to next rule, if PROBE_STATE just
2062 * ignore and continue with next opcode.
2064 if (cmd->opcode == O_CHECK_STATE)
2070 retval = 0; /* accept */
2075 args->rule = f; /* report matching rule */
2076 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG;
2081 if (args->eh) /* not on layer 2 */
2084 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT,
2085 sizeof(uint16_t), MB_DONTWAIT);
2087 retval = IP_FW_PORT_DENY_FLAG;
2090 *(uint16_t *)m_tag_data(mtag) = f->rulenum;
2091 m_tag_prepend(m, mtag);
2092 retval = (cmd->opcode == O_DIVERT) ?
2094 cmd->arg1 | IP_FW_PORT_TEE_FLAG;
2099 f->pcnt++; /* update stats */
2101 f->timestamp = time_second;
2102 if (cmd->opcode == O_COUNT)
2105 if (f->next_rule == NULL)
2106 lookup_next_rule(f);
2112 * Drop the packet and send a reject notice
2113 * if the packet is not ICMP (or is an ICMP
2114 * query), and it is not multicast/broadcast.
2117 (proto != IPPROTO_ICMP ||
2118 is_icmp_query(ip)) &&
2119 !(m->m_flags & (M_BCAST|M_MCAST)) &&
2120 !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
2121 send_reject(args, cmd->arg1,
2127 retval = IP_FW_PORT_DENY_FLAG;
2131 if (args->eh) /* not valid on layer2 pkts */
2133 if (!q || dyn_dir == MATCH_FORWARD) {
2135 &((ipfw_insn_sa *)cmd)->sa;
2141 panic("-- unknown opcode %d\n", cmd->opcode);
2142 } /* end of switch() on opcodes */
2144 if (cmd->len & F_NOT)
2148 if (cmd->len & F_OR)
2151 if (!(cmd->len & F_OR)) /* not an OR block, */
2152 break; /* try next rule */
2155 } /* end of inner for, scan opcodes */
2157 next_rule:; /* try next rule */
2159 } /* end of outer for, scan rules */
2160 kprintf("+++ ipfw: ouch!, skip past end of rules, denying packet\n");
2161 return(IP_FW_PORT_DENY_FLAG);
2164 /* Update statistics */
2167 f->timestamp = time_second;
2172 kprintf("pullup failed\n");
2173 return(IP_FW_PORT_DENY_FLAG);
2177 ipfw_dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa)
2182 const struct ipfw_flow_id *id;
2183 struct dn_flow_id *fid;
2187 mtag = m_tag_get(PACKET_TAG_DUMMYNET, sizeof(*pkt), MB_DONTWAIT);
2192 m_tag_prepend(m, mtag);
2194 pkt = m_tag_data(mtag);
2195 bzero(pkt, sizeof(*pkt));
2197 cmd = fwa->rule->cmd + fwa->rule->act_ofs;
2198 if (cmd->opcode == O_LOG)
2200 KASSERT(cmd->opcode == O_PIPE || cmd->opcode == O_QUEUE,
2201 ("Rule is not PIPE or QUEUE, opcode %d\n", cmd->opcode));
2204 pkt->dn_flags = (dir & DN_FLAGS_DIR_MASK);
2205 pkt->ifp = fwa->oif;
2206 pkt->cpuid = mycpu->gd_cpuid;
2207 pkt->pipe_nr = pipe_nr;
2211 fid->fid_dst_ip = id->dst_ip;
2212 fid->fid_src_ip = id->src_ip;
2213 fid->fid_dst_port = id->dst_port;
2214 fid->fid_src_port = id->src_port;
2215 fid->fid_proto = id->proto;
2216 fid->fid_flags = id->flags;
2218 ipfw_ref_rule(fwa->rule);
2219 pkt->dn_priv = fwa->rule;
2220 pkt->dn_unref_priv = ipfw_unref_rule;
2222 if (cmd->opcode == O_PIPE)
2223 pkt->dn_flags |= DN_FLAGS_IS_PIPE;
2225 if (dir == DN_TO_IP_OUT) {
2227 * We need to copy *ro because for ICMP pkts (and maybe
2228 * others) the caller passed a pointer into the stack;
2229 * dst might also be a pointer into *ro so it needs to
2232 pkt->ro = *(fwa->ro);
2234 fwa->ro->ro_rt->rt_refcnt++;
2235 if (fwa->dst == (struct sockaddr_in *)&fwa->ro->ro_dst) {
2236 /* 'dst' points into 'ro' */
2237 fwa->dst = (struct sockaddr_in *)&(pkt->ro.ro_dst);
2239 pkt->dn_dst = fwa->dst;
2240 pkt->flags = fwa->flags;
2243 m->m_pkthdr.fw_flags |= DUMMYNET_MBUF_TAGGED;
2248 * When a rule is added/deleted, clear the next_rule pointers in all rules.
2249 * These will be reconstructed on the fly as packets are matched.
2250 * Must be called at splimp().
2253 flush_rule_ptrs(void)
2257 for (rule = layer3_chain; rule; rule = rule->next)
2258 rule->next_rule = NULL;
2261 static __inline void
2262 ipfw_inc_static_count(struct ip_fw *rule)
2264 IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2267 static_ioc_len += IOC_RULESIZE(rule);
2270 static __inline void
2271 ipfw_dec_static_count(struct ip_fw *rule)
2273 int l = IOC_RULESIZE(rule);
2275 IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2277 KASSERT(static_count > 0, ("invalid static count %u\n", static_count));
2280 KASSERT(static_ioc_len >= l,
2281 ("invalid static len %u\n", static_ioc_len));
2282 static_ioc_len -= l;
2285 static struct ip_fw *
2286 ipfw_create_rule(const struct ipfw_ioc_rule *ioc_rule)
2290 rule = kmalloc(RULESIZE(ioc_rule), M_IPFW, M_WAITOK | M_ZERO);
2292 rule->act_ofs = ioc_rule->act_ofs;
2293 rule->cmd_len = ioc_rule->cmd_len;
2294 rule->rulenum = ioc_rule->rulenum;
2295 rule->set = ioc_rule->set;
2296 rule->usr_flags = ioc_rule->usr_flags;
2298 bcopy(ioc_rule->cmd, rule->cmd, rule->cmd_len * 4 /* XXX */);
2306 * Add a new rule to the list. Copy the rule into a malloc'ed area,
2307 * then possibly create a rule number and add the rule to the list.
2308 * Update the rule_number in the input struct so the caller knows
2312 ipfw_add_rule(struct ip_fw **head, struct ipfw_ioc_rule *ioc_rule)
2314 struct ip_fw *rule, *f, *prev;
2316 KKASSERT(*head != NULL);
2317 IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2321 rule = ipfw_create_rule(ioc_rule);
2326 * If rulenum is 0, find highest numbered rule before the
2327 * default rule, and add rule number incremental step
2329 if (rule->rulenum == 0) {
2330 int step = autoinc_step;
2332 KKASSERT(step >= IPFW_AUTOINC_STEP_MIN &&
2333 step <= IPFW_AUTOINC_STEP_MAX);
2336 * Locate the highest numbered rule before default
2338 for (f = *head; f; f = f->next) {
2339 if (f->rulenum == IPFW_DEFAULT_RULE)
2341 rule->rulenum = f->rulenum;
2343 if (rule->rulenum < IPFW_DEFAULT_RULE - step)
2344 rule->rulenum += step;
2346 /* Update the input structure */
2347 ioc_rule->rulenum = rule->rulenum;
2351 * Now insert the new rule in the right place in the sorted list.
2353 for (prev = NULL, f = *head; f; prev = f, f = f->next) {
2354 if (f->rulenum > rule->rulenum) {
2355 /* Found the location */
2368 ipfw_inc_static_count(rule);
2372 DEB(kprintf("++ installed rule %d, static count now %d\n",
2373 rule->rulenum, static_count);)
2377 * Free storage associated with a static rule (including derived
2379 * The caller is in charge of clearing rule pointers to avoid
2380 * dangling pointers.
2381 * @return a pointer to the next entry.
2382 * Arguments are not checked, so they better be correct.
2383 * Must be called at splimp().
2385 static struct ip_fw *
2386 delete_rule(struct ip_fw **head, struct ip_fw *prev, struct ip_fw *rule)
2393 remove_dyn_rule(rule, NULL /* force removal */);
2398 ipfw_dec_static_count(rule);
2400 /* Mark the rule as invalid */
2401 rule->rule_flags |= IPFW_RULE_F_INVALID;
2402 rule->next_rule = NULL;
2404 /* Try to free this rule */
2405 ipfw_free_rule(rule);
2411 * Deletes all rules from a chain (including the default rule
2412 * if the second argument is set).
2413 * Must be called at splimp().
2416 free_chain(struct ip_fw **chain, int kill_default)
2420 flush_rule_ptrs(); /* more efficient to do outside the loop */
2422 while ((rule = *chain) != NULL &&
2423 (kill_default || rule->rulenum != IPFW_DEFAULT_RULE))
2424 delete_rule(chain, NULL, rule);
2426 KASSERT(dyn_count == 0, ("%u dyn rule remains\n", dyn_count));
2429 ip_fw_default_rule = NULL; /* Reset default rule */
2431 if (ipfw_dyn_v != NULL) {
2433 * Free dynamic rules(state) hash table
2435 kfree(ipfw_dyn_v, M_IPFW);
2439 KASSERT(static_count == 0,
2440 ("%u static rules remains\n", static_count));
2441 KASSERT(static_ioc_len == 0,
2442 ("%u bytes of static rules remains\n", static_ioc_len));
2444 KASSERT(static_count == 1,
2445 ("%u static rules remains\n", static_count));
2446 KASSERT(static_ioc_len == IOC_RULESIZE(ip_fw_default_rule),
2447 ("%u bytes of static rules remains, should be %u\n",
2448 static_ioc_len, IOC_RULESIZE(ip_fw_default_rule)));
2453 * Remove all rules with given number, and also do set manipulation.
2455 * The argument is an uint32_t. The low 16 bit are the rule or set number,
2456 * the next 8 bits are the new set, the top 8 bits are the command:
2458 * 0 delete rules with given number
2459 * 1 delete rules with given set number
2460 * 2 move rules with given number to new set
2461 * 3 move rules with given set number to new set
2462 * 4 swap sets with given numbers
2465 del_entry(struct ip_fw **chain, uint32_t arg)
2467 struct ip_fw *prev, *rule;
2469 uint8_t cmd, new_set;
2471 rulenum = arg & 0xffff;
2472 cmd = (arg >> 24) & 0xff;
2473 new_set = (arg >> 16) & 0xff;
2479 if (cmd == 0 || cmd == 2) {
2480 if (rulenum == IPFW_DEFAULT_RULE)
2488 case 0: /* delete rules with given number */
2490 * locate first rule to delete
2492 for (prev = NULL, rule = *chain;
2493 rule && rule->rulenum < rulenum;
2494 prev = rule, rule = rule->next)
2496 if (rule->rulenum != rulenum)
2499 crit_enter(); /* no access to rules while removing */
2501 * flush pointers outside the loop, then delete all matching
2502 * rules. prev remains the same throughout the cycle.
2505 while (rule && rule->rulenum == rulenum)
2506 rule = delete_rule(chain, prev, rule);
2510 case 1: /* delete all rules with given set number */
2513 for (prev = NULL, rule = *chain; rule;) {
2514 if (rule->set == rulenum) {
2515 rule = delete_rule(chain, prev, rule);
2524 case 2: /* move rules with given number to new set */
2526 for (rule = *chain; rule; rule = rule->next) {
2527 if (rule->rulenum == rulenum)
2528 rule->set = new_set;
2533 case 3: /* move rules with given set number to new set */
2535 for (rule = *chain; rule; rule = rule->next) {
2536 if (rule->set == rulenum)
2537 rule->set = new_set;
2542 case 4: /* swap two sets */
2544 for (rule = *chain; rule; rule = rule->next) {
2545 if (rule->set == rulenum)
2546 rule->set = new_set;
2547 else if (rule->set == new_set)
2548 rule->set = rulenum;
2557 * Clear counters for a specific rule.
2560 clear_counters(struct ip_fw *rule, int log_only)
2562 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
2564 if (log_only == 0) {
2565 rule->bcnt = rule->pcnt = 0;
2566 rule->timestamp = 0;
2568 if (l->o.opcode == O_LOG)
2569 l->log_left = l->max_log;
2573 * Reset some or all counters on firewall rules.
2574 * @arg frwl is null to clear all entries, or contains a specific
2576 * @arg log_only is 1 if we only want to reset logs, zero otherwise.
2579 zero_entry(int rulenum, int log_only)
2587 for (rule = layer3_chain; rule; rule = rule->next)
2588 clear_counters(rule, log_only);
2590 msg = log_only ? "ipfw: All logging counts reset.\n"
2591 : "ipfw: Accounting cleared.\n";
2596 * We can have multiple rules with the same number, so we
2597 * need to clear them all.
2599 for (rule = layer3_chain; rule; rule = rule->next) {
2600 if (rule->rulenum == rulenum) {
2602 while (rule && rule->rulenum == rulenum) {
2603 clear_counters(rule, log_only);
2611 if (!cleared) /* we did not find any matching rules */
2613 msg = log_only ? "ipfw: Entry %d logging count reset.\n"
2614 : "ipfw: Entry %d cleared.\n";
2617 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
2622 * Check validity of the structure before insert.
2623 * Fortunately rules are simple, so this mostly need to check rule sizes.
2626 ipfw_ctl_check_rule(struct ipfw_ioc_rule *rule, int size)
2629 int have_action = 0;
2632 /* Check for valid size */
2633 if (size < sizeof(*rule)) {
2634 kprintf("ipfw: rule too short\n");
2637 l = IOC_RULESIZE(rule);
2639 kprintf("ipfw: size mismatch (have %d want %d)\n", size, l);
2644 * Now go for the individual checks. Very simple ones, basically only
2645 * instruction sizes.
2647 for (l = rule->cmd_len, cmd = rule->cmd; l > 0;
2648 l -= cmdlen, cmd += cmdlen) {
2649 cmdlen = F_LEN(cmd);
2651 kprintf("ipfw: opcode %d size truncated\n",
2655 DEB(kprintf("ipfw: opcode %d\n", cmd->opcode);)
2656 switch (cmd->opcode) {
2670 case O_IPPRECEDENCE:
2677 if (cmdlen != F_INSN_SIZE(ipfw_insn))
2689 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
2694 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
2699 if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
2702 ((ipfw_insn_log *)cmd)->log_left =
2703 ((ipfw_insn_log *)cmd)->max_log;
2709 if (cmdlen != F_INSN_SIZE(ipfw_insn_ip))
2711 if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) {
2712 kprintf("ipfw: opcode %d, useless rule\n",
2720 if (cmd->arg1 == 0 || cmd->arg1 > 256) {
2721 kprintf("ipfw: invalid set size %d\n",
2725 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
2731 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
2737 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
2738 if (cmdlen < 2 || cmdlen > 31)
2745 if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
2751 if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
2756 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
2760 case O_FORWARD_MAC: /* XXX not implemented yet */
2769 if (cmdlen != F_INSN_SIZE(ipfw_insn))
2773 kprintf("ipfw: opcode %d, multiple actions"
2780 kprintf("ipfw: opcode %d, action must be"
2787 kprintf("ipfw: opcode %d, unknown opcode\n",
2792 if (have_action == 0) {
2793 kprintf("ipfw: missing action\n");
2799 kprintf("ipfw: opcode %d size %d wrong\n",
2800 cmd->opcode, cmdlen);
2805 ipfw_ctl_add_rule(struct sockopt *sopt)
2807 struct ipfw_ioc_rule *ioc_rule;
2811 size = sopt->sopt_valsize;
2812 if (size > (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX) ||
2813 size < sizeof(*ioc_rule)) {
2816 if (size != (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX)) {
2817 sopt->sopt_val = krealloc(sopt->sopt_val, sizeof(uint32_t) *
2818 IPFW_RULE_SIZE_MAX, M_TEMP, M_WAITOK);
2820 ioc_rule = sopt->sopt_val;
2822 error = ipfw_ctl_check_rule(ioc_rule, size);
2826 ipfw_add_rule(&layer3_chain, ioc_rule);
2828 if (sopt->sopt_dir == SOPT_GET)
2829 sopt->sopt_valsize = IOC_RULESIZE(ioc_rule);
2834 ipfw_copy_rule(const struct ip_fw *rule, struct ipfw_ioc_rule *ioc_rule)
2836 ioc_rule->act_ofs = rule->act_ofs;
2837 ioc_rule->cmd_len = rule->cmd_len;
2838 ioc_rule->rulenum = rule->rulenum;
2839 ioc_rule->set = rule->set;
2840 ioc_rule->usr_flags = rule->usr_flags;
2842 ioc_rule->set_disable = set_disable;
2843 ioc_rule->static_count = static_count;
2844 ioc_rule->static_len = static_ioc_len;
2846 ioc_rule->pcnt = rule->pcnt;
2847 ioc_rule->bcnt = rule->bcnt;
2848 ioc_rule->timestamp = rule->timestamp;
2850 bcopy(rule->cmd, ioc_rule->cmd, ioc_rule->cmd_len * 4 /* XXX */);
2852 return ((uint8_t *)ioc_rule + IOC_RULESIZE(ioc_rule));
2856 ipfw_copy_state(const ipfw_dyn_rule *dyn_rule,
2857 struct ipfw_ioc_state *ioc_state)
2859 const struct ipfw_flow_id *id;
2860 struct ipfw_ioc_flowid *ioc_id;
2862 ioc_state->expire = TIME_LEQ(dyn_rule->expire, time_second) ?
2863 0 : dyn_rule->expire - time_second;
2864 ioc_state->pcnt = dyn_rule->pcnt;
2865 ioc_state->bcnt = dyn_rule->bcnt;
2867 ioc_state->dyn_type = dyn_rule->dyn_type;
2868 ioc_state->count = dyn_rule->count;
2870 ioc_state->rulenum = dyn_rule->rule->rulenum;
2873 ioc_id = &ioc_state->id;
2875 ioc_id->type = ETHERTYPE_IP;
2876 ioc_id->u.ip.dst_ip = id->dst_ip;
2877 ioc_id->u.ip.src_ip = id->src_ip;
2878 ioc_id->u.ip.dst_port = id->dst_port;
2879 ioc_id->u.ip.src_port = id->src_port;
2880 ioc_id->u.ip.proto = id->proto;
2884 ipfw_ctl_get_rules(struct sockopt *sopt)
2889 uint32_t dcount = 0;
2892 * pass up a copy of the current rules. Static rules
2893 * come first (the last of which has number IPFW_DEFAULT_RULE),
2894 * followed by a possibly empty list of dynamic rule.
2898 size = static_ioc_len; /* size of static rules */
2899 if (ipfw_dyn_v) { /* add size of dyn.rules */
2901 size += dcount * sizeof(struct ipfw_ioc_state);
2904 if (sopt->sopt_valsize < size) {
2905 /* short length, no need to return incomplete rules */
2906 /* XXX: if superuser, no need to zero buffer */
2907 bzero(sopt->sopt_val, sopt->sopt_valsize);
2910 bp = sopt->sopt_val;
2912 for (rule = layer3_chain; rule; rule = rule->next)
2913 bp = ipfw_copy_rule(rule, bp);
2915 if (ipfw_dyn_v && dcount != 0) {
2916 struct ipfw_ioc_state *ioc_state = bp;
2917 uint32_t dcount2 = 0;
2919 size_t old_size = size;
2923 lockmgr(&dyn_lock, LK_SHARED);
2924 for (i = 0; i < curr_dyn_buckets; i++) {
2928 * The # of dynamic rules may have grown after the
2929 * snapshot of 'dyn_count' was taken, so we will have
2930 * to check 'dcount' (snapshot of dyn_count) here to
2931 * make sure that we don't overflow the pre-allocated
2934 for (p = ipfw_dyn_v[i]; p != NULL && dcount != 0;
2935 p = p->next, ioc_state++, dcount--, dcount2++)
2936 ipfw_copy_state(p, ioc_state);
2938 lockmgr(&dyn_lock, LK_RELEASE);
2941 * The # of dynamic rules may be shrinked after the
2942 * snapshot of 'dyn_count' was taken. To give user a
2943 * correct dynamic rule count, we use the 'dcount2'
2944 * calculated above (with shared lockmgr lock held).
2946 size = static_ioc_len +
2947 (dcount2 * sizeof(struct ipfw_ioc_state));
2948 KKASSERT(size <= old_size);
2953 sopt->sopt_valsize = size;
2958 * {set|get}sockopt parser.
2961 ipfw_ctl(struct sockopt *sopt)
2969 switch (sopt->sopt_name) {
2971 error = ipfw_ctl_get_rules(sopt);
2976 * Normally we cannot release the lock on each iteration.
2977 * We could do it here only because we start from the head all
2978 * the times so there is no risk of missing some entries.
2979 * On the other hand, the risk is that we end up with
2980 * a very inconsistent ruleset, so better keep the lock
2981 * around the whole cycle.
2983 * XXX this code can be improved by resetting the head of
2984 * the list to point to the default rule, and then freeing
2985 * the old list without the need for a lock.
2989 free_chain(&layer3_chain, 0 /* keep default rule */);
2994 error = ipfw_ctl_add_rule(sopt);
2999 * IP_FW_DEL is used for deleting single rules or sets,
3000 * and (ab)used to atomically manipulate sets. Argument size
3001 * is used to distinguish between the two:
3003 * delete single rule or set of rules,
3004 * or reassign rules (or sets) to a different set.
3005 * 2*sizeof(uint32_t)
3006 * atomic disable/enable sets.
3007 * first uint32_t contains sets to be disabled,
3008 * second uint32_t contains sets to be enabled.
3010 masks = sopt->sopt_val;
3011 size = sopt->sopt_valsize;
3012 if (size == sizeof(*masks)) {
3014 * Delete or reassign static rule
3016 error = del_entry(&layer3_chain, masks[0]);
3017 } else if (size == (2 * sizeof(*masks))) {
3019 * Set enable/disable
3024 (set_disable | masks[0]) & ~masks[1] &
3025 ~(1 << 31); /* set 31 always enabled */
3034 case IP_FW_RESETLOG: /* argument is an int, the rule number */
3037 if (sopt->sopt_val != 0) {
3038 error = soopt_to_kbuf(sopt, &rulenum,
3039 sizeof(int), sizeof(int));
3043 error = zero_entry(rulenum, sopt->sopt_name == IP_FW_RESETLOG);
3047 kprintf("ipfw_ctl invalid option %d\n", sopt->sopt_name);
3054 * This procedure is only used to handle keepalives. It is invoked
3055 * every dyn_keepalive_period
3058 ipfw_tick(void *dummy __unused)
3064 if (ipfw_dyn_v == NULL || dyn_count == 0)
3067 keep_alive = time_second;
3069 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3071 gen = dyn_buckets_gen;
3072 for (i = 0; i < curr_dyn_buckets; i++) {
3073 ipfw_dyn_rule *q, *prev;
3075 for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) {
3076 uint32_t ack_rev, ack_fwd;
3077 struct ipfw_flow_id id;
3079 if (q->dyn_type == O_LIMIT_PARENT)
3082 if (TIME_LEQ(q->expire, time_second)) {
3084 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
3089 * Keep alive processing
3094 if (q->id.proto != IPPROTO_TCP)
3096 if ((q->state & BOTH_SYN) != BOTH_SYN)
3098 if (TIME_LEQ(time_second + dyn_keepalive_interval,
3100 goto next; /* too early */
3101 if (q->keep_alive == keep_alive)
3102 goto next; /* alreay done */
3105 * Save necessary information, so that they could
3106 * survive after possible blocking in send_pkt()
3109 ack_rev = q->ack_rev;
3110 ack_fwd = q->ack_fwd;
3112 /* Sending has been started */
3113 q->keep_alive = keep_alive;
3115 /* Release lock to avoid possible dead lock */
3116 lockmgr(&dyn_lock, LK_RELEASE);
3117 send_pkt(&id, ack_rev - 1, ack_fwd, TH_SYN);
3118 send_pkt(&id, ack_fwd - 1, ack_rev, 0);
3119 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3121 if (gen != dyn_buckets_gen) {
3123 * Dyn bucket array has been changed during
3124 * the above two sending; reiterate.
3133 lockmgr(&dyn_lock, LK_RELEASE);
3135 callout_reset(&ipfw_timeout_h, dyn_keepalive_period * hz,
3140 ipfw_init_default_rule(struct ip_fw **head)
3142 struct ip_fw *def_rule;
3144 KKASSERT(*head == NULL);
3146 def_rule = kmalloc(sizeof(*def_rule), M_IPFW, M_WAITOK | M_ZERO);
3148 def_rule->act_ofs = 0;
3149 def_rule->rulenum = IPFW_DEFAULT_RULE;
3150 def_rule->cmd_len = 1;
3153 def_rule->cmd[0].len = 1;
3154 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
3155 def_rule->cmd[0].opcode = O_ACCEPT;
3157 def_rule->cmd[0].opcode = O_DENY;
3160 def_rule->refcnt = 1;
3163 ipfw_inc_static_count(def_rule);
3165 /* Install the default rule */
3166 ip_fw_default_rule = def_rule;
3170 ipfw_sysctl_autoinc_step(SYSCTL_HANDLER_ARGS)
3172 return sysctl_int_range(oidp, arg1, arg2, req,
3173 IPFW_AUTOINC_STEP_MIN, IPFW_AUTOINC_STEP_MAX);
3177 ipfw_sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS)
3181 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3183 value = dyn_buckets;
3184 error = sysctl_handle_int(oidp, &value, 0, req);
3185 if (error || !req->newptr)
3189 * Make sure we have a power of 2 and
3190 * do not allow more than 64k entries.
3193 if (value < 0 || value > 65536)
3195 if ((value & (value - 1)) != 0)
3199 dyn_buckets = value;
3201 lockmgr(&dyn_lock, LK_RELEASE);
3206 ipfw_sysctl_dyn_fin(SYSCTL_HANDLER_ARGS)
3208 return sysctl_int_range(oidp, arg1, arg2, req,
3209 1, dyn_keepalive_period - 1);
3213 ipfw_sysctl_dyn_rst(SYSCTL_HANDLER_ARGS)
3215 return sysctl_int_range(oidp, arg1, arg2, req,
3216 1, dyn_keepalive_period - 1);
3220 ipfw_init_dispatch(struct netmsg *nmsg)
3227 kprintf("IP firewall already loaded\n");
3232 ip_fw_chk_ptr = ipfw_chk;
3233 ip_fw_ctl_ptr = ipfw_ctl;
3234 ip_fw_dn_io_ptr = ipfw_dummynet_io;
3236 layer3_chain = NULL;
3237 ipfw_init_default_rule(&layer3_chain);
3239 kprintf("ipfw2 initialized, divert %s, "
3240 "rule-based forwarding enabled, default to %s, logging ",
3246 ip_fw_default_rule->cmd[0].opcode == O_ACCEPT ?
3249 #ifdef IPFIREWALL_VERBOSE
3252 #ifdef IPFIREWALL_VERBOSE_LIMIT
3253 verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
3255 if (fw_verbose == 0) {
3256 kprintf("disabled\n");
3257 } else if (verbose_limit == 0) {
3258 kprintf("unlimited\n");
3260 kprintf("limited to %d packets/entry by default\n",
3264 callout_init(&ipfw_timeout_h);
3265 lockinit(&dyn_lock, "ipfw_dyn", 0, 0);
3268 callout_reset(&ipfw_timeout_h, hz, ipfw_tick, NULL);
3271 lwkt_replymsg(&nmsg->nm_lmsg, error);
3279 netmsg_init(&smsg, &curthread->td_msgport, 0, ipfw_init_dispatch);
3280 return lwkt_domsg(IPFW_CFGPORT, &smsg.nm_lmsg, 0);
3286 ipfw_fini_dispatch(struct netmsg *nmsg)
3292 if (ipfw_refcnt != 0) {
3297 callout_stop(&ipfw_timeout_h);
3300 netmsg_service_sync();
3302 ip_fw_chk_ptr = NULL;
3303 ip_fw_ctl_ptr = NULL;
3304 ip_fw_dn_io_ptr = NULL;
3305 free_chain(&layer3_chain, 1 /* kill default rule */);
3307 kprintf("IP firewall unloaded\n");
3310 lwkt_replymsg(&nmsg->nm_lmsg, error);
3318 netmsg_init(&smsg, &curthread->td_msgport, 0, ipfw_fini_dispatch);
3319 return lwkt_domsg(IPFW_CFGPORT, &smsg.nm_lmsg, 0);
3322 #endif /* KLD_MODULE */
3325 ipfw_modevent(module_t mod, int type, void *unused)
3336 kprintf("ipfw statically compiled, cannot unload\n");
3348 static moduledata_t ipfwmod = {
3353 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PROTO_END, SI_ORDER_ANY);
3354 MODULE_VERSION(ipfw, 1);