White space and style changes
[dragonfly.git] / sys / net / ipfw / ip_fw2.c
1 /*
2  * Copyright (c) 2002 Luigi Rizzo, Universita` di Pisa
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
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.
12  *
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
23  * SUCH DAMAGE.
24  *
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 1.55 2008/07/31 11:58:42 sephe Exp $
27  */
28
29 #define        DEB(x)
30 #define        DDB(x) x
31
32 /*
33  * Implement IP packet firewall (new version)
34  */
35
36 #ifndef KLD_MODULE
37 #include "opt_ipfw.h"
38 #include "opt_ipdn.h"
39 #include "opt_ipdivert.h"
40 #include "opt_inet.h"
41 #ifndef INET
42 #error IPFIREWALL requires INET.
43 #endif /* INET */
44 #endif
45
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/kernel.h>
51 #include <sys/proc.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>
59
60 #include <net/if.h>
61 #include <net/route.h>
62 #include <net/netmsg2.h>
63
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/in_var.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/ip.h>
69 #include <netinet/ip_var.h>
70 #include <netinet/ip_icmp.h>
71 #include "ip_fw.h"
72 #include <net/dummynet/ip_dummynet.h>
73 #include <netinet/tcp.h>
74 #include <netinet/tcp_timer.h>
75 #include <netinet/tcp_var.h>
76 #include <netinet/tcpip.h>
77 #include <netinet/udp.h>
78 #include <netinet/udp_var.h>
79
80 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
81
82 /*
83  * set_disable contains one bit per set value (0..31).
84  * If the bit is set, all rules with the corresponding set
85  * are disabled. Set 31 is reserved for the default rule
86  * and CANNOT be disabled.
87  */
88 static uint32_t set_disable;
89
90 static int fw_verbose;
91 static int verbose_limit;
92
93 #ifdef KLD_MODULE
94 static int ipfw_refcnt;
95 #endif
96
97 static struct callout ipfw_timeout_h;
98 #define IPFW_DEFAULT_RULE       65535
99
100 /*
101  * list of rules for layer 3
102  */
103 static struct ip_fw *layer3_chain;
104
105 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
106
107 static int fw_debug = 1;
108 static int autoinc_step = 100; /* bounded to 1..1000 in ipfw_add_rule() */
109
110 #ifdef SYSCTL_NODE
111 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
112 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable, CTLFLAG_RW,
113     &fw_enable, 0, "Enable ipfw");
114 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
115     &autoinc_step, 0, "Rule number autincrement step");
116 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW,
117     &fw_one_pass, 0,
118     "Only do a single pass through ipfw when using dummynet(4)");
119 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
120     &fw_debug, 0, "Enable printing of debug ip_fw statements");
121 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW,
122     &fw_verbose, 0, "Log matches to ipfw rules");
123 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
124     &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
125
126 /*
127  * Description of dynamic rules.
128  *
129  * Dynamic rules are stored in lists accessed through a hash table
130  * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
131  * be modified through the sysctl variable dyn_buckets which is
132  * updated when the table becomes empty.
133  *
134  * XXX currently there is only one list, ipfw_dyn.
135  *
136  * When a packet is received, its address fields are first masked
137  * with the mask defined for the rule, then hashed, then matched
138  * against the entries in the corresponding list.
139  * Dynamic rules can be used for different purposes:
140  *  + stateful rules;
141  *  + enforcing limits on the number of sessions;
142  *  + in-kernel NAT (not implemented yet)
143  *
144  * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
145  * measured in seconds and depending on the flags.
146  *
147  * The total number of dynamic rules is stored in dyn_count.
148  * The max number of dynamic rules is dyn_max. When we reach
149  * the maximum number of rules we do not create anymore. This is
150  * done to avoid consuming too much memory, but also too much
151  * time when searching on each packet (ideally, we should try instead
152  * to put a limit on the length of the list on each bucket...).
153  *
154  * Each dynamic rule holds a pointer to the parent ipfw rule so
155  * we know what action to perform. Dynamic rules are removed when
156  * the parent rule is deleted. XXX we should make them survive.
157  *
158  * There are some limitations with dynamic rules -- we do not
159  * obey the 'randomized match', and we do not do multiple
160  * passes through the firewall. XXX check the latter!!!
161  */
162 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
163 static uint32_t dyn_buckets = 256; /* must be power of 2 */
164 static uint32_t curr_dyn_buckets = 256; /* must be power of 2 */
165
166 /*
167  * Timeouts for various events in handing dynamic rules.
168  */
169 static uint32_t dyn_ack_lifetime = 300;
170 static uint32_t dyn_syn_lifetime = 20;
171 static uint32_t dyn_fin_lifetime = 1;
172 static uint32_t dyn_rst_lifetime = 1;
173 static uint32_t dyn_udp_lifetime = 10;
174 static uint32_t dyn_short_lifetime = 5;
175
176 /*
177  * Keepalives are sent if dyn_keepalive is set. They are sent every
178  * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
179  * seconds of lifetime of a rule.
180  * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
181  * than dyn_keepalive_period.
182  */
183
184 static uint32_t dyn_keepalive_interval = 20;
185 static uint32_t dyn_keepalive_period = 5;
186 static uint32_t dyn_keepalive = 1;      /* do send keepalives */
187
188 static uint32_t static_count;           /* # of static rules */
189 static uint32_t static_ioc_len; /* bytes of static rules */
190 static uint32_t dyn_count;              /* # of dynamic rules */
191 static uint32_t dyn_max = 4096; /* max # of dynamic rules */
192
193 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
194     &dyn_buckets, 0, "Number of dyn. buckets");
195 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
196     &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
197 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
198     &dyn_count, 0, "Number of dyn. rules");
199 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
200     &dyn_max, 0, "Max number of dyn. rules");
201 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
202     &static_count, 0, "Number of static rules");
203 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
204     &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
205 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
206     &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
207 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
208     &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
209 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
210     &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
211 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
212     &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
213 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
214     &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
215 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
216     &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
217
218 #endif /* SYSCTL_NODE */
219
220 static struct ip_fw *ip_fw_default_rule;
221
222 static ip_fw_chk_t      ipfw_chk;
223
224 static __inline int
225 ipfw_free_rule(struct ip_fw *rule)
226 {
227         KASSERT(rule->refcnt > 0, ("invalid refcnt %u\n", rule->refcnt));
228         atomic_subtract_int(&rule->refcnt, 1);
229         if (atomic_cmpset_int(&rule->refcnt, 0, 1)) {
230                 kfree(rule, M_IPFW);
231                 return 1;
232         }
233         return 0;
234 }
235
236 static void
237 ipfw_unref_rule(void *priv)
238 {
239         ipfw_free_rule(priv);
240 #ifdef KLD_MODULE
241         atomic_subtract_int(&ipfw_refcnt, 1);
242 #endif
243 }
244
245 static __inline void
246 ipfw_ref_rule(struct ip_fw *rule)
247 {
248 #ifdef KLD_MODULE
249         atomic_add_int(&ipfw_refcnt, 1);
250 #endif
251         atomic_add_int(&rule->refcnt, 1);
252 }
253
254 /*
255  * This macro maps an ip pointer into a layer3 header pointer of type T
256  */
257 #define L3HDR(T, ip) ((T *)((uint32_t *)(ip) + (ip)->ip_hl))
258
259 static __inline int
260 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
261 {
262         int type = L3HDR(struct icmp,ip)->icmp_type;
263
264         return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1 << type)));
265 }
266
267 #define TT      ((1 << ICMP_ECHO) | \
268                  (1 << ICMP_ROUTERSOLICIT) | \
269                  (1 << ICMP_TSTAMP) | \
270                  (1 << ICMP_IREQ) | \
271                  (1 << ICMP_MASKREQ))
272
273 static int
274 is_icmp_query(struct ip *ip)
275 {
276         int type = L3HDR(struct icmp, ip)->icmp_type;
277
278         return (type <= ICMP_MAXTYPE && (TT & (1 << type)));
279 }
280
281 #undef TT
282
283 /*
284  * The following checks use two arrays of 8 or 16 bits to store the
285  * bits that we want set or clear, respectively. They are in the
286  * low and high half of cmd->arg1 or cmd->d[0].
287  *
288  * We scan options and store the bits we find set. We succeed if
289  *
290  *      (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
291  *
292  * The code is sometimes optimized not to store additional variables.
293  */
294
295 static int
296 flags_match(ipfw_insn *cmd, uint8_t bits)
297 {
298         u_char want_clear;
299         bits = ~bits;
300
301         if (((cmd->arg1 & 0xff) & bits) != 0)
302                 return 0; /* some bits we want set were clear */
303
304         want_clear = (cmd->arg1 >> 8) & 0xff;
305         if ((want_clear & bits) != want_clear)
306                 return 0; /* some bits we want clear were set */
307         return 1;
308 }
309
310 static int
311 ipopts_match(struct ip *ip, ipfw_insn *cmd)
312 {
313         int optlen, bits = 0;
314         u_char *cp = (u_char *)(ip + 1);
315         int x = (ip->ip_hl << 2) - sizeof(struct ip);
316
317         for (; x > 0; x -= optlen, cp += optlen) {
318                 int opt = cp[IPOPT_OPTVAL];
319
320                 if (opt == IPOPT_EOL)
321                         break;
322
323                 if (opt == IPOPT_NOP) {
324                         optlen = 1;
325                 } else {
326                         optlen = cp[IPOPT_OLEN];
327                         if (optlen <= 0 || optlen > x)
328                                 return 0; /* invalid or truncated */
329                 }
330
331                 switch (opt) {
332                 case IPOPT_LSRR:
333                         bits |= IP_FW_IPOPT_LSRR;
334                         break;
335
336                 case IPOPT_SSRR:
337                         bits |= IP_FW_IPOPT_SSRR;
338                         break;
339
340                 case IPOPT_RR:
341                         bits |= IP_FW_IPOPT_RR;
342                         break;
343
344                 case IPOPT_TS:
345                         bits |= IP_FW_IPOPT_TS;
346                         break;
347
348                 default:
349                         break;
350                 }
351         }
352         return (flags_match(cmd, bits));
353 }
354
355 static int
356 tcpopts_match(struct ip *ip, ipfw_insn *cmd)
357 {
358         int optlen, bits = 0;
359         struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
360         u_char *cp = (u_char *)(tcp + 1);
361         int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
362
363         for (; x > 0; x -= optlen, cp += optlen) {
364                 int opt = cp[0];
365
366                 if (opt == TCPOPT_EOL)
367                         break;
368
369                 if (opt == TCPOPT_NOP) {
370                         optlen = 1;
371                 } else {
372                         optlen = cp[1];
373                         if (optlen <= 0)
374                                 break;
375                 }
376
377                 switch (opt) {
378                 case TCPOPT_MAXSEG:
379                         bits |= IP_FW_TCPOPT_MSS;
380                         break;
381
382                 case TCPOPT_WINDOW:
383                         bits |= IP_FW_TCPOPT_WINDOW;
384                         break;
385
386                 case TCPOPT_SACK_PERMITTED:
387                 case TCPOPT_SACK:
388                         bits |= IP_FW_TCPOPT_SACK;
389                         break;
390
391                 case TCPOPT_TIMESTAMP:
392                         bits |= IP_FW_TCPOPT_TS;
393                         break;
394
395                 case TCPOPT_CC:
396                 case TCPOPT_CCNEW:
397                 case TCPOPT_CCECHO:
398                         bits |= IP_FW_TCPOPT_CC;
399                         break;
400
401                 default:
402                         break;
403                 }
404         }
405         return (flags_match(cmd, bits));
406 }
407
408 static int
409 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
410 {
411         if (ifp == NULL)        /* no iface with this packet, match fails */
412                 return 0;
413
414         /* Check by name or by IP address */
415         if (cmd->name[0] != '\0') { /* match by name */
416                 /* Check name */
417                 if (cmd->p.glob) {
418                         if (kfnmatch(cmd->name, ifp->if_xname, 0) == 0)
419                                 return(1);
420                 } else {
421                         if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
422                                 return(1);
423                 }
424         } else {
425                 struct ifaddr_container *ifac;
426
427                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
428                         struct ifaddr *ia = ifac->ifa;
429
430                         if (ia->ifa_addr == NULL)
431                                 continue;
432                         if (ia->ifa_addr->sa_family != AF_INET)
433                                 continue;
434                         if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
435                             (ia->ifa_addr))->sin_addr.s_addr)
436                                 return(1);      /* match */
437                 }
438         }
439         return(0);      /* no match, fail ... */
440 }
441
442 static uint64_t norule_counter; /* counter for ipfw_log(NULL...) */
443
444 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
445
446 /*
447  * We enter here when we have a rule with O_LOG.
448  * XXX this function alone takes about 2Kbytes of code!
449  */
450 static void
451 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
452          struct mbuf *m, struct ifnet *oif)
453 {
454         char *action;
455         int limit_reached = 0;
456         char action2[40], proto[48], fragment[28];
457
458         fragment[0] = '\0';
459         proto[0] = '\0';
460
461         if (f == NULL) {        /* bogus pkt */
462                 if (verbose_limit != 0 && norule_counter >= verbose_limit)
463                         return;
464                 norule_counter++;
465                 if (norule_counter == verbose_limit)
466                         limit_reached = verbose_limit;
467                 action = "Refuse";
468         } else {        /* O_LOG is the first action, find the real one */
469                 ipfw_insn *cmd = ACTION_PTR(f);
470                 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
471
472                 if (l->max_log != 0 && l->log_left == 0)
473                         return;
474                 l->log_left--;
475                 if (l->log_left == 0)
476                         limit_reached = l->max_log;
477                 cmd += F_LEN(cmd);      /* point to first action */
478                 if (cmd->opcode == O_PROB)
479                         cmd += F_LEN(cmd);
480
481                 action = action2;
482                 switch (cmd->opcode) {
483                 case O_DENY:
484                         action = "Deny";
485                         break;
486
487                 case O_REJECT:
488                         if (cmd->arg1==ICMP_REJECT_RST) {
489                                 action = "Reset";
490                         } else if (cmd->arg1==ICMP_UNREACH_HOST) {
491                                 action = "Reject";
492                         } else {
493                                 ksnprintf(SNPARGS(action2, 0), "Unreach %d",
494                                           cmd->arg1);
495                         }
496                         break;
497
498                 case O_ACCEPT:
499                         action = "Accept";
500                         break;
501
502                 case O_COUNT:
503                         action = "Count";
504                         break;
505
506                 case O_DIVERT:
507                         ksnprintf(SNPARGS(action2, 0), "Divert %d", cmd->arg1);
508                         break;
509
510                 case O_TEE:
511                         ksnprintf(SNPARGS(action2, 0), "Tee %d", cmd->arg1);
512                         break;
513
514                 case O_SKIPTO:
515                         ksnprintf(SNPARGS(action2, 0), "SkipTo %d", cmd->arg1);
516                         break;
517
518                 case O_PIPE:
519                         ksnprintf(SNPARGS(action2, 0), "Pipe %d", cmd->arg1);
520                         break;
521
522                 case O_QUEUE:
523                         ksnprintf(SNPARGS(action2, 0), "Queue %d", cmd->arg1);
524                         break;
525
526                 case O_FORWARD_IP:
527                         {
528                                 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
529                                 int len;
530
531                                 len = ksnprintf(SNPARGS(action2, 0),
532                                                 "Forward to %s",
533                                                 inet_ntoa(sa->sa.sin_addr));
534                                 if (sa->sa.sin_port) {
535                                         ksnprintf(SNPARGS(action2, len), ":%d",
536                                                   sa->sa.sin_port);
537                                 }
538                         }
539                         break;
540
541                 default:
542                         action = "UNKNOWN";
543                         break;
544                 }
545         }
546
547         if (hlen == 0) {        /* non-ip */
548                 ksnprintf(SNPARGS(proto, 0), "MAC");
549         } else {
550                 struct ip *ip = mtod(m, struct ip *);
551                 /* these three are all aliases to the same thing */
552                 struct icmp *const icmp = L3HDR(struct icmp, ip);
553                 struct tcphdr *const tcp = (struct tcphdr *)icmp;
554                 struct udphdr *const udp = (struct udphdr *)icmp;
555
556                 int ip_off, offset, ip_len;
557                 int len;
558
559                 if (eh != NULL) { /* layer 2 packets are as on the wire */
560                         ip_off = ntohs(ip->ip_off);
561                         ip_len = ntohs(ip->ip_len);
562                 } else {
563                         ip_off = ip->ip_off;
564                         ip_len = ip->ip_len;
565                 }
566                 offset = ip_off & IP_OFFMASK;
567                 switch (ip->ip_p) {
568                 case IPPROTO_TCP:
569                         len = ksnprintf(SNPARGS(proto, 0), "TCP %s",
570                                         inet_ntoa(ip->ip_src));
571                         if (offset == 0) {
572                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
573                                           ntohs(tcp->th_sport),
574                                           inet_ntoa(ip->ip_dst),
575                                           ntohs(tcp->th_dport));
576                         } else {
577                                 ksnprintf(SNPARGS(proto, len), " %s",
578                                           inet_ntoa(ip->ip_dst));
579                         }
580                         break;
581
582                 case IPPROTO_UDP:
583                         len = ksnprintf(SNPARGS(proto, 0), "UDP %s",
584                                         inet_ntoa(ip->ip_src));
585                         if (offset == 0) {
586                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
587                                           ntohs(udp->uh_sport),
588                                           inet_ntoa(ip->ip_dst),
589                                           ntohs(udp->uh_dport));
590                         } else {
591                                 ksnprintf(SNPARGS(proto, len), " %s",
592                                           inet_ntoa(ip->ip_dst));
593                         }
594                         break;
595
596                 case IPPROTO_ICMP:
597                         if (offset == 0) {
598                                 len = ksnprintf(SNPARGS(proto, 0),
599                                                 "ICMP:%u.%u ",
600                                                 icmp->icmp_type,
601                                                 icmp->icmp_code);
602                         } else {
603                                 len = ksnprintf(SNPARGS(proto, 0), "ICMP ");
604                         }
605                         len += ksnprintf(SNPARGS(proto, len), "%s",
606                                          inet_ntoa(ip->ip_src));
607                         ksnprintf(SNPARGS(proto, len), " %s",
608                                   inet_ntoa(ip->ip_dst));
609                         break;
610
611                 default:
612                         len = ksnprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
613                                         inet_ntoa(ip->ip_src));
614                         ksnprintf(SNPARGS(proto, len), " %s",
615                                   inet_ntoa(ip->ip_dst));
616                         break;
617                 }
618
619                 if (ip_off & (IP_MF | IP_OFFMASK)) {
620                         ksnprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
621                                   ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
622                                   offset << 3, (ip_off & IP_MF) ? "+" : "");
623                 }
624         }
625
626         if (oif || m->m_pkthdr.rcvif) {
627                 log(LOG_SECURITY | LOG_INFO,
628                     "ipfw: %d %s %s %s via %s%s\n",
629                     f ? f->rulenum : -1,
630                     action, proto, oif ? "out" : "in",
631                     oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
632                     fragment);
633         } else {
634                 log(LOG_SECURITY | LOG_INFO,
635                     "ipfw: %d %s %s [no if info]%s\n",
636                     f ? f->rulenum : -1,
637                     action, proto, fragment);
638         }
639
640         if (limit_reached) {
641                 log(LOG_SECURITY | LOG_NOTICE,
642                     "ipfw: limit %d reached on entry %d\n",
643                     limit_reached, f ? f->rulenum : -1);
644         }
645 }
646
647 #undef SNPARGS
648
649 /*
650  * IMPORTANT: the hash function for dynamic rules must be commutative
651  * in source and destination (ip,port), because rules are bidirectional
652  * and we want to find both in the same bucket.
653  */
654 static __inline int
655 hash_packet(struct ipfw_flow_id *id)
656 {
657         uint32_t i;
658
659         i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
660         i &= (curr_dyn_buckets - 1);
661         return i;
662 }
663
664 /**
665  * unlink a dynamic rule from a chain. prev is a pointer to
666  * the previous one, q is a pointer to the rule to delete,
667  * head is a pointer to the head of the queue.
668  * Modifies q and potentially also head.
669  */
670 #define UNLINK_DYN_RULE(prev, head, q) {                                \
671         ipfw_dyn_rule *old_q = q;                                       \
672                                                                         \
673         /* remove a refcount to the parent */                           \
674         if (q->dyn_type == O_LIMIT)                                     \
675                 q->parent->count--;                                     \
676         DEB(kprintf("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",        \
677                 (q->id.src_ip), (q->id.src_port),                       \
678                 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); )      \
679         if (prev != NULL)                                               \
680                 prev->next = q = q->next;                               \
681         else                                                            \
682                 head = q = q->next;                                     \
683         KASSERT(dyn_count > 0, ("invalid dyn count %u\n", dyn_count));  \
684         dyn_count--;                                                    \
685         kfree(old_q, M_IPFW); }
686
687 #define TIME_LEQ(a,b)       ((int)((a)-(b)) <= 0)
688
689 /**
690  * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
691  *
692  * If keep_me == NULL, rules are deleted even if not expired,
693  * otherwise only expired rules are removed.
694  *
695  * The value of the second parameter is also used to point to identify
696  * a rule we absolutely do not want to remove (e.g. because we are
697  * holding a reference to it -- this is the case with O_LIMIT_PARENT
698  * rules). The pointer is only used for comparison, so any non-null
699  * value will do.
700  */
701 static void
702 remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
703 {
704         static uint32_t last_remove = 0;
705
706 #define FORCE (keep_me == NULL)
707
708         ipfw_dyn_rule *prev, *q;
709         int i, pass = 0, max_pass = 0;
710
711         if (ipfw_dyn_v == NULL || dyn_count == 0)
712                 return;
713         /* do not expire more than once per second, it is useless */
714         if (!FORCE && last_remove == time_second)
715                 return;
716         last_remove = time_second;
717
718         /*
719          * because O_LIMIT refer to parent rules, during the first pass only
720          * remove child and mark any pending LIMIT_PARENT, and remove
721          * them in a second pass.
722          */
723 next_pass:
724         for (i = 0 ; i < curr_dyn_buckets ; i++) {
725                 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
726                         /*
727                          * Logic can become complex here, so we split tests.
728                          */
729                         if (q == keep_me)
730                                 goto next;
731                         if (rule != NULL && rule != q->rule)
732                                 goto next; /* not the one we are looking for */
733                         if (q->dyn_type == O_LIMIT_PARENT) {
734                                 /*
735                                  * handle parent in the second pass,
736                                  * record we need one.
737                                  */
738                                 max_pass = 1;
739                                 if (pass == 0)
740                                         goto next;
741                                 if (FORCE && q->count != 0 ) {
742                                         /* XXX should not happen! */
743                                         kprintf( "OUCH! cannot remove rule,"
744                                              " count %d\n", q->count);
745                                 }
746                         } else {
747                                 if (!FORCE &&
748                                     !TIME_LEQ( q->expire, time_second ))
749                                         goto next;
750                         }
751                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
752                         continue;
753 next:
754                         prev=q;
755                         q=q->next;
756                 }
757         }
758         if (pass++ < max_pass)
759                 goto next_pass;
760 }
761
762
763 /**
764  * lookup a dynamic rule.
765  */
766 static ipfw_dyn_rule *
767 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
768         struct tcphdr *tcp)
769 {
770         /*
771          * stateful ipfw extensions.
772          * Lookup into dynamic session queue
773          */
774 #define MATCH_REVERSE   0
775 #define MATCH_FORWARD   1
776 #define MATCH_NONE      2
777 #define MATCH_UNKNOWN   3
778         int i, dir = MATCH_NONE;
779         ipfw_dyn_rule *prev, *q=NULL;
780
781         if (ipfw_dyn_v == NULL)
782                 goto done;      /* not found */
783         i = hash_packet( pkt );
784         for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
785                 if (q->dyn_type == O_LIMIT_PARENT)
786                         goto next;
787                 if (TIME_LEQ( q->expire, time_second)) { /* expire entry */
788                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
789                         continue;
790                 }
791                 if ( pkt->proto == q->id.proto) {
792                         if (pkt->src_ip == q->id.src_ip &&
793                             pkt->dst_ip == q->id.dst_ip &&
794                             pkt->src_port == q->id.src_port &&
795                             pkt->dst_port == q->id.dst_port ) {
796                                 dir = MATCH_FORWARD;
797                                 break;
798                         }
799                         if (pkt->src_ip == q->id.dst_ip &&
800                             pkt->dst_ip == q->id.src_ip &&
801                             pkt->src_port == q->id.dst_port &&
802                             pkt->dst_port == q->id.src_port ) {
803                                 dir = MATCH_REVERSE;
804                                 break;
805                         }
806                 }
807 next:
808                 prev = q;
809                 q = q->next;
810         }
811         if (q == NULL)
812                 goto done; /* q = NULL, not found */
813
814         if ( prev != NULL) { /* found and not in front */
815                 prev->next = q->next;
816                 q->next = ipfw_dyn_v[i];
817                 ipfw_dyn_v[i] = q;
818         }
819         if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
820                 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
821
822 #define BOTH_SYN        (TH_SYN | (TH_SYN << 8))
823 #define BOTH_FIN        (TH_FIN | (TH_FIN << 8))
824                 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
825                 switch (q->state) {
826                 case TH_SYN:                            /* opening */
827                         q->expire = time_second + dyn_syn_lifetime;
828                         break;
829
830                 case BOTH_SYN:                  /* move to established */
831                 case BOTH_SYN | TH_FIN :        /* one side tries to close */
832                 case BOTH_SYN | (TH_FIN << 8) :
833                         if (tcp) {
834 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
835                             uint32_t ack = ntohl(tcp->th_ack);
836                             if (dir == MATCH_FORWARD) {
837                                 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
838                                     q->ack_fwd = ack;
839                                 else { /* ignore out-of-sequence */
840                                     break;
841                                 }
842                             } else {
843                                 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
844                                     q->ack_rev = ack;
845                                 else { /* ignore out-of-sequence */
846                                     break;
847                                 }
848                             }
849                         }
850                         q->expire = time_second + dyn_ack_lifetime;
851                         break;
852
853                 case BOTH_SYN | BOTH_FIN:       /* both sides closed */
854                         if (dyn_fin_lifetime >= dyn_keepalive_period)
855                                 dyn_fin_lifetime = dyn_keepalive_period - 1;
856                         q->expire = time_second + dyn_fin_lifetime;
857                         break;
858
859                 default:
860 #if 0
861                         /*
862                          * reset or some invalid combination, but can also
863                          * occur if we use keep-state the wrong way.
864                          */
865                         if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
866                                 kprintf("invalid state: 0x%x\n", q->state);
867 #endif
868                         if (dyn_rst_lifetime >= dyn_keepalive_period)
869                                 dyn_rst_lifetime = dyn_keepalive_period - 1;
870                         q->expire = time_second + dyn_rst_lifetime;
871                         break;
872                 }
873         } else if (pkt->proto == IPPROTO_UDP) {
874                 q->expire = time_second + dyn_udp_lifetime;
875         } else {
876                 /* other protocols */
877                 q->expire = time_second + dyn_short_lifetime;
878         }
879 done:
880         if (match_direction)
881                 *match_direction = dir;
882         return q;
883 }
884
885 static void
886 realloc_dynamic_table(void)
887 {
888         /*
889          * Try reallocation, make sure we have a power of 2 and do
890          * not allow more than 64k entries. In case of overflow,
891          * default to 1024.
892          */
893
894         if (dyn_buckets > 65536)
895                 dyn_buckets = 1024;
896         if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
897                 dyn_buckets = curr_dyn_buckets; /* reset */
898                 return;
899         }
900         curr_dyn_buckets = dyn_buckets;
901         if (ipfw_dyn_v != NULL)
902                 kfree(ipfw_dyn_v, M_IPFW);
903         for (;;) {
904                 ipfw_dyn_v = kmalloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
905                        M_IPFW, M_INTWAIT | M_NULLOK | M_ZERO);
906                 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
907                         break;
908                 curr_dyn_buckets /= 2;
909         }
910 }
911
912 /**
913  * Install state of type 'type' for a dynamic session.
914  * The hash table contains two type of rules:
915  * - regular rules (O_KEEP_STATE)
916  * - rules for sessions with limited number of sess per user
917  *   (O_LIMIT). When they are created, the parent is
918  *   increased by 1, and decreased on delete. In this case,
919  *   the third parameter is the parent rule and not the chain.
920  * - "parent" rules for the above (O_LIMIT_PARENT).
921  */
922 static ipfw_dyn_rule *
923 add_dyn_rule(struct ipfw_flow_id *id, uint8_t dyn_type, struct ip_fw *rule)
924 {
925         ipfw_dyn_rule *r;
926         int i;
927
928         if (ipfw_dyn_v == NULL ||
929             (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
930                 realloc_dynamic_table();
931                 if (ipfw_dyn_v == NULL)
932                         return NULL; /* failed ! */
933         }
934         i = hash_packet(id);
935
936         r = kmalloc(sizeof *r, M_IPFW, M_INTWAIT | M_NULLOK | M_ZERO);
937         if (r == NULL) {
938                 kprintf ("sorry cannot allocate state\n");
939                 return NULL;
940         }
941
942         /* increase refcount on parent, and set pointer */
943         if (dyn_type == O_LIMIT) {
944                 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
945                 if ( parent->dyn_type != O_LIMIT_PARENT)
946                         panic("invalid parent");
947                 parent->count++;
948                 r->parent = parent;
949                 rule = parent->rule;
950         }
951
952         r->id = *id;
953         r->expire = time_second + dyn_syn_lifetime;
954         r->rule = rule;
955         r->dyn_type = dyn_type;
956         r->pcnt = r->bcnt = 0;
957         r->count = 0;
958
959         r->bucket = i;
960         r->next = ipfw_dyn_v[i];
961         ipfw_dyn_v[i] = r;
962         dyn_count++;
963         DEB(kprintf("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
964            dyn_type,
965            (r->id.src_ip), (r->id.src_port),
966            (r->id.dst_ip), (r->id.dst_port),
967            dyn_count ); )
968         return r;
969 }
970
971 /**
972  * lookup dynamic parent rule using pkt and rule as search keys.
973  * If the lookup fails, then install one.
974  */
975 static ipfw_dyn_rule *
976 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
977 {
978         ipfw_dyn_rule *q;
979         int i;
980
981         if (ipfw_dyn_v) {
982                 i = hash_packet( pkt );
983                 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
984                         if (q->dyn_type == O_LIMIT_PARENT &&
985                             rule== q->rule &&
986                             pkt->proto == q->id.proto &&
987                             pkt->src_ip == q->id.src_ip &&
988                             pkt->dst_ip == q->id.dst_ip &&
989                             pkt->src_port == q->id.src_port &&
990                             pkt->dst_port == q->id.dst_port) {
991                                 q->expire = time_second + dyn_short_lifetime;
992                                 DEB(kprintf("lookup_dyn_parent found 0x%p\n",q);)
993                                 return q;
994                         }
995         }
996         return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
997 }
998
999 /**
1000  * Install dynamic state for rule type cmd->o.opcode
1001  *
1002  * Returns 1 (failure) if state is not installed because of errors or because
1003  * session limitations are enforced.
1004  */
1005 static int
1006 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1007         struct ip_fw_args *args)
1008 {
1009         static int last_log;
1010
1011         ipfw_dyn_rule *q;
1012
1013         DEB(kprintf("-- install state type %d 0x%08x %u -> 0x%08x %u\n",
1014             cmd->o.opcode,
1015             (args->f_id.src_ip), (args->f_id.src_port),
1016             (args->f_id.dst_ip), (args->f_id.dst_port) );)
1017
1018         q = lookup_dyn_rule(&args->f_id, NULL, NULL);
1019
1020         if (q != NULL) { /* should never occur */
1021                 if (last_log != time_second) {
1022                         last_log = time_second;
1023                         kprintf(" install_state: entry already present, done\n");
1024                 }
1025                 return 0;
1026         }
1027
1028         if (dyn_count >= dyn_max)
1029                 /*
1030                  * Run out of slots, try to remove any expired rule.
1031                  */
1032                 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1033
1034         if (dyn_count >= dyn_max) {
1035                 if (last_log != time_second) {
1036                         last_log = time_second;
1037                         kprintf("install_state: Too many dynamic rules\n");
1038                 }
1039                 return 1; /* cannot install, notify caller */
1040         }
1041
1042         switch (cmd->o.opcode) {
1043         case O_KEEP_STATE: /* bidir rule */
1044                 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1045                 break;
1046
1047         case O_LIMIT: /* limit number of sessions */
1048             {
1049                 uint16_t limit_mask = cmd->limit_mask;
1050                 struct ipfw_flow_id id;
1051                 ipfw_dyn_rule *parent;
1052
1053                 DEB(kprintf("installing dyn-limit rule %d\n", cmd->conn_limit);)
1054
1055                 id.dst_ip = id.src_ip = 0;
1056                 id.dst_port = id.src_port = 0;
1057                 id.proto = args->f_id.proto;
1058
1059                 if (limit_mask & DYN_SRC_ADDR)
1060                         id.src_ip = args->f_id.src_ip;
1061                 if (limit_mask & DYN_DST_ADDR)
1062                         id.dst_ip = args->f_id.dst_ip;
1063                 if (limit_mask & DYN_SRC_PORT)
1064                         id.src_port = args->f_id.src_port;
1065                 if (limit_mask & DYN_DST_PORT)
1066                         id.dst_port = args->f_id.dst_port;
1067                 parent = lookup_dyn_parent(&id, rule);
1068                 if (parent == NULL) {
1069                         kprintf("add parent failed\n");
1070                         return 1;
1071                 }
1072                 if (parent->count >= cmd->conn_limit) {
1073                         /*
1074                          * See if we can remove some expired rule.
1075                          */
1076                         remove_dyn_rule(rule, parent);
1077                         if (parent->count >= cmd->conn_limit) {
1078                                 if (fw_verbose && last_log != time_second) {
1079                                         last_log = time_second;
1080                                         log(LOG_SECURITY | LOG_DEBUG,
1081                                             "drop session, too many entries\n");
1082                                 }
1083                                 return 1;
1084                         }
1085                 }
1086                 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1087             }
1088                 break;
1089         default:
1090                 kprintf("unknown dynamic rule type %u\n", cmd->o.opcode);
1091                 return 1;
1092         }
1093         lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1094         return 0;
1095 }
1096
1097 /*
1098  * Transmit a TCP packet, containing either a RST or a keepalive.
1099  * When flags & TH_RST, we are sending a RST packet, because of a
1100  * "reset" action matched the packet.
1101  * Otherwise we are sending a keepalive, and flags & TH_
1102  */
1103 static void
1104 send_pkt(struct ipfw_flow_id *id, uint32_t seq, uint32_t ack, int flags)
1105 {
1106         struct mbuf *m;
1107         struct ip *ip;
1108         struct tcphdr *tcp;
1109         struct route sro;       /* fake route */
1110
1111         MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1112         if (m == 0)
1113                 return;
1114         m->m_pkthdr.rcvif = NULL;
1115         m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1116         m->m_data += max_linkhdr;
1117
1118         ip = mtod(m, struct ip *);
1119         bzero(ip, m->m_len);
1120         tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1121         ip->ip_p = IPPROTO_TCP;
1122         tcp->th_off = 5;
1123
1124         /*
1125          * Assume we are sending a RST (or a keepalive in the reverse
1126          * direction), swap src and destination addresses and ports.
1127          */
1128         ip->ip_src.s_addr = htonl(id->dst_ip);
1129         ip->ip_dst.s_addr = htonl(id->src_ip);
1130         tcp->th_sport = htons(id->dst_port);
1131         tcp->th_dport = htons(id->src_port);
1132         if (flags & TH_RST) {   /* we are sending a RST */
1133                 if (flags & TH_ACK) {
1134                         tcp->th_seq = htonl(ack);
1135                         tcp->th_ack = htonl(0);
1136                         tcp->th_flags = TH_RST;
1137                 } else {
1138                         if (flags & TH_SYN)
1139                                 seq++;
1140                         tcp->th_seq = htonl(0);
1141                         tcp->th_ack = htonl(seq);
1142                         tcp->th_flags = TH_RST | TH_ACK;
1143                 }
1144         } else {
1145                 /*
1146                  * We are sending a keepalive. flags & TH_SYN determines
1147                  * the direction, forward if set, reverse if clear.
1148                  * NOTE: seq and ack are always assumed to be correct
1149                  * as set by the caller. This may be confusing...
1150                  */
1151                 if (flags & TH_SYN) {
1152                         /*
1153                          * we have to rewrite the correct addresses!
1154                          */
1155                         ip->ip_dst.s_addr = htonl(id->dst_ip);
1156                         ip->ip_src.s_addr = htonl(id->src_ip);
1157                         tcp->th_dport = htons(id->dst_port);
1158                         tcp->th_sport = htons(id->src_port);
1159                 }
1160                 tcp->th_seq = htonl(seq);
1161                 tcp->th_ack = htonl(ack);
1162                 tcp->th_flags = TH_ACK;
1163         }
1164
1165         /*
1166          * set ip_len to the payload size so we can compute
1167          * the tcp checksum on the pseudoheader
1168          * XXX check this, could save a couple of words ?
1169          */
1170         ip->ip_len = htons(sizeof(struct tcphdr));
1171         tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1172
1173         /*
1174          * now fill fields left out earlier
1175          */
1176         ip->ip_ttl = ip_defttl;
1177         ip->ip_len = m->m_pkthdr.len;
1178
1179         bzero(&sro, sizeof(sro));
1180         ip_rtaddr(ip->ip_dst, &sro);
1181
1182         m->m_pkthdr.fw_flags |= IPFW_MBUF_GENERATED;
1183         ip_output(m, NULL, &sro, 0, NULL, NULL);
1184         if (sro.ro_rt)
1185                 RTFREE(sro.ro_rt);
1186 }
1187
1188 /*
1189  * sends a reject message, consuming the mbuf passed as an argument.
1190  */
1191 static void
1192 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1193 {
1194
1195         if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1196                 /* We need the IP header in host order for icmp_error(). */
1197                 if (args->eh != NULL) {
1198                         struct ip *ip = mtod(args->m, struct ip *);
1199                         ip->ip_len = ntohs(ip->ip_len);
1200                         ip->ip_off = ntohs(ip->ip_off);
1201                 }
1202                 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1203         } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1204                 struct tcphdr *const tcp =
1205                     L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1206                 if ( (tcp->th_flags & TH_RST) == 0)
1207                         send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1208                                 ntohl(tcp->th_ack),
1209                                 tcp->th_flags | TH_RST);
1210                 m_freem(args->m);
1211         } else
1212                 m_freem(args->m);
1213         args->m = NULL;
1214 }
1215
1216 /**
1217  *
1218  * Given an ip_fw *, lookup_next_rule will return a pointer
1219  * to the next rule, which can be either the jump
1220  * target (for skipto instructions) or the next one in the list (in
1221  * all other cases including a missing jump target).
1222  * The result is also written in the "next_rule" field of the rule.
1223  * Backward jumps are not allowed, so start looking from the next
1224  * rule...
1225  *
1226  * This never returns NULL -- in case we do not have an exact match,
1227  * the next rule is returned. When the ruleset is changed,
1228  * pointers are flushed so we are always correct.
1229  */
1230
1231 static struct ip_fw *
1232 lookup_next_rule(struct ip_fw *me)
1233 {
1234         struct ip_fw *rule = NULL;
1235         ipfw_insn *cmd;
1236
1237         /* look for action, in case it is a skipto */
1238         cmd = ACTION_PTR(me);
1239         if (cmd->opcode == O_LOG)
1240                 cmd += F_LEN(cmd);
1241         if ( cmd->opcode == O_SKIPTO )
1242                 for (rule = me->next; rule ; rule = rule->next)
1243                         if (rule->rulenum >= cmd->arg1)
1244                                 break;
1245         if (rule == NULL)                       /* failure or not a skipto */
1246                 rule = me->next;
1247         me->next_rule = rule;
1248         return rule;
1249 }
1250
1251 /*
1252  * The main check routine for the firewall.
1253  *
1254  * All arguments are in args so we can modify them and return them
1255  * back to the caller.
1256  *
1257  * Parameters:
1258  *
1259  *      args->m (in/out) The packet; we set to NULL when/if we nuke it.
1260  *              Starts with the IP header.
1261  *      args->eh (in)   Mac header if present, or NULL for layer3 packet.
1262  *      args->oif       Outgoing interface, or NULL if packet is incoming.
1263  *              The incoming interface is in the mbuf. (in)
1264  *
1265  *      args->rule      Pointer to the last matching rule (in/out)
1266  *      args->next_hop  Socket we are forwarding to (out).
1267  *      args->f_id      Addresses grabbed from the packet (out)
1268  *
1269  * Return value:
1270  *
1271  *      IP_FW_PORT_DENY_FLAG    the packet must be dropped.
1272  *      0       The packet is to be accepted and routed normally OR
1273  *              the packet was denied/rejected and has been dropped;
1274  *              in the latter case, *m is equal to NULL upon return.
1275  *      port    Divert the packet to port, with these caveats:
1276  *
1277  *              - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead
1278  *                of diverting it (ie, 'ipfw tee').
1279  *
1280  *              - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower
1281  *                16 bits as a dummynet pipe number instead of diverting
1282  */
1283
1284 static int
1285 ipfw_chk(struct ip_fw_args *args)
1286 {
1287         /*
1288          * Local variables hold state during the processing of a packet.
1289          *
1290          * IMPORTANT NOTE: to speed up the processing of rules, there
1291          * are some assumption on the values of the variables, which
1292          * are documented here. Should you change them, please check
1293          * the implementation of the various instructions to make sure
1294          * that they still work.
1295          *
1296          * args->eh     The MAC header. It is non-null for a layer2
1297          *      packet, it is NULL for a layer-3 packet.
1298          *
1299          * m | args->m  Pointer to the mbuf, as received from the caller.
1300          *      It may change if ipfw_chk() does an m_pullup, or if it
1301          *      consumes the packet because it calls send_reject().
1302          *      XXX This has to change, so that ipfw_chk() never modifies
1303          *      or consumes the buffer.
1304          * ip   is simply an alias of the value of m, and it is kept
1305          *      in sync with it (the packet is  supposed to start with
1306          *      the ip header).
1307          */
1308         struct mbuf *m = args->m;
1309         struct ip *ip = mtod(m, struct ip *);
1310
1311         /*
1312          * oif | args->oif      If NULL, ipfw_chk has been called on the
1313          *      inbound path (ether_input, ip_input).
1314          *      If non-NULL, ipfw_chk has been called on the outbound path
1315          *      (ether_output, ip_output).
1316          */
1317         struct ifnet *oif = args->oif;
1318
1319         struct ip_fw *f = NULL;         /* matching rule */
1320         int retval = 0;
1321         struct m_tag *mtag;
1322
1323         /*
1324          * hlen The length of the IPv4 header.
1325          *      hlen >0 means we have an IPv4 packet.
1326          */
1327         u_int hlen = 0;         /* hlen >0 means we have an IP pkt */
1328
1329         /*
1330          * offset       The offset of a fragment. offset != 0 means that
1331          *      we have a fragment at this offset of an IPv4 packet.
1332          *      offset == 0 means that (if this is an IPv4 packet)
1333          *      this is the first or only fragment.
1334          */
1335         u_short offset = 0;
1336
1337         /*
1338          * Local copies of addresses. They are only valid if we have
1339          * an IP packet.
1340          *
1341          * proto        The protocol. Set to 0 for non-ip packets,
1342          *      or to the protocol read from the packet otherwise.
1343          *      proto != 0 means that we have an IPv4 packet.
1344          *
1345          * src_port, dst_port   port numbers, in HOST format. Only
1346          *      valid for TCP and UDP packets.
1347          *
1348          * src_ip, dst_ip       ip addresses, in NETWORK format.
1349          *      Only valid for IPv4 packets.
1350          */
1351         uint8_t proto;
1352         uint16_t src_port = 0, dst_port = 0;    /* NOTE: host format    */
1353         struct in_addr src_ip, dst_ip;          /* NOTE: network format */
1354         uint16_t ip_len = 0;
1355         int dyn_dir = MATCH_UNKNOWN;
1356         ipfw_dyn_rule *q = NULL;
1357
1358         if (m->m_pkthdr.fw_flags & IPFW_MBUF_GENERATED)
1359                 return 0;       /* accept */
1360         /*
1361          * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1362          *      MATCH_NONE when checked and not matched (q = NULL),
1363          *      MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
1364          */
1365
1366         if (args->eh == NULL ||         /* layer 3 packet */
1367             (m->m_pkthdr.len >= sizeof(struct ip) &&
1368              ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1369                 hlen = ip->ip_hl << 2;
1370
1371         /*
1372          * Collect parameters into local variables for faster matching.
1373          */
1374         if (hlen == 0) {        /* do not grab addresses for non-ip pkts */
1375                 proto = args->f_id.proto = 0;   /* mark f_id invalid */
1376                 goto after_ip_checks;
1377         }
1378
1379         proto = args->f_id.proto = ip->ip_p;
1380         src_ip = ip->ip_src;
1381         dst_ip = ip->ip_dst;
1382         if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1383                 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1384                 ip_len = ntohs(ip->ip_len);
1385         } else {
1386                 offset = ip->ip_off & IP_OFFMASK;
1387                 ip_len = ip->ip_len;
1388         }
1389
1390 #define PULLUP_TO(len)                          \
1391 do {                                            \
1392         if (m->m_len < (len)) {                 \
1393                 args->m = m = m_pullup(m, (len));\
1394                 if (m == NULL)                  \
1395                         goto pullup_failed;     \
1396                 ip = mtod(m, struct ip *);      \
1397         }                                       \
1398 } while (0)
1399
1400         if (offset == 0) {
1401                 switch (proto) {
1402                 case IPPROTO_TCP:
1403                         {
1404                                 struct tcphdr *tcp;
1405
1406                                 PULLUP_TO(hlen + sizeof(struct tcphdr));
1407                                 tcp = L3HDR(struct tcphdr, ip);
1408                                 dst_port = tcp->th_dport;
1409                                 src_port = tcp->th_sport;
1410                                 args->f_id.flags = tcp->th_flags;
1411                         }
1412                         break;
1413
1414                 case IPPROTO_UDP:
1415                         {
1416                                 struct udphdr *udp;
1417
1418                                 PULLUP_TO(hlen + sizeof(struct udphdr));
1419                                 udp = L3HDR(struct udphdr, ip);
1420                                 dst_port = udp->uh_dport;
1421                                 src_port = udp->uh_sport;
1422                         }
1423                         break;
1424
1425                 case IPPROTO_ICMP:
1426                         PULLUP_TO(hlen + 4);    /* type, code and checksum. */
1427                         args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1428                         break;
1429
1430                 default:
1431                         break;
1432                 }
1433         }
1434
1435 #undef PULLUP_TO
1436
1437         args->f_id.src_ip = ntohl(src_ip.s_addr);
1438         args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1439         args->f_id.src_port = src_port = ntohs(src_port);
1440         args->f_id.dst_port = dst_port = ntohs(dst_port);
1441
1442 after_ip_checks:
1443         if (args->rule) {
1444                 /*
1445                  * Packet has already been tagged. Look for the next rule
1446                  * to restart processing.
1447                  *
1448                  * If fw_one_pass != 0 then just accept it.
1449                  * XXX should not happen here, but optimized out in
1450                  * the caller.
1451                  */
1452                 if (fw_one_pass)
1453                         return 0;
1454
1455                 /* This rule was deleted */
1456                 if (args->rule->rule_flags & IPFW_RULE_F_INVALID)
1457                         return IP_FW_PORT_DENY_FLAG;
1458
1459                 f = args->rule->next_rule;
1460                 if (f == NULL)
1461                         f = lookup_next_rule(args->rule);
1462         } else {
1463                 /*
1464                  * Find the starting rule. It can be either the first
1465                  * one, or the one after divert_rule if asked so.
1466                  */
1467                 int skipto;
1468
1469                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1470                 if (mtag != NULL)
1471                         skipto = *(uint16_t *)m_tag_data(mtag);
1472                 else
1473                         skipto = 0;
1474
1475                 f = layer3_chain;
1476                 if (args->eh == NULL && skipto != 0) {
1477                         if (skipto >= IPFW_DEFAULT_RULE)
1478                                 return(IP_FW_PORT_DENY_FLAG); /* invalid */
1479                         while (f && f->rulenum <= skipto)
1480                                 f = f->next;
1481                         if (f == NULL)  /* drop packet */
1482                                 return(IP_FW_PORT_DENY_FLAG);
1483                 }
1484         }
1485         if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
1486                 m_tag_delete(m, mtag);
1487
1488         /*
1489          * Now scan the rules, and parse microinstructions for each rule.
1490          */
1491         for (; f; f = f->next) {
1492                 int l, cmdlen;
1493                 ipfw_insn *cmd;
1494                 int skip_or; /* skip rest of OR block */
1495
1496 again:
1497                 if (set_disable & (1 << f->set))
1498                         continue;
1499
1500                 skip_or = 0;
1501                 for (l = f->cmd_len, cmd = f->cmd; l > 0;
1502                      l -= cmdlen, cmd += cmdlen) {
1503                         int match;
1504
1505                         /*
1506                          * check_body is a jump target used when we find a
1507                          * CHECK_STATE, and need to jump to the body of
1508                          * the target rule.
1509                          */
1510
1511 check_body:
1512                         cmdlen = F_LEN(cmd);
1513                         /*
1514                          * An OR block (insn_1 || .. || insn_n) has the
1515                          * F_OR bit set in all but the last instruction.
1516                          * The first match will set "skip_or", and cause
1517                          * the following instructions to be skipped until
1518                          * past the one with the F_OR bit clear.
1519                          */
1520                         if (skip_or) {          /* skip this instruction */
1521                                 if ((cmd->len & F_OR) == 0)
1522                                         skip_or = 0;    /* next one is good */
1523                                 continue;
1524                         }
1525                         match = 0; /* set to 1 if we succeed */
1526
1527                         switch (cmd->opcode) {
1528                         /*
1529                          * The first set of opcodes compares the packet's
1530                          * fields with some pattern, setting 'match' if a
1531                          * match is found. At the end of the loop there is
1532                          * logic to deal with F_NOT and F_OR flags associated
1533                          * with the opcode.
1534                          */
1535                         case O_NOP:
1536                                 match = 1;
1537                                 break;
1538
1539                         case O_FORWARD_MAC:
1540                                 kprintf("ipfw: opcode %d unimplemented\n",
1541                                         cmd->opcode);
1542                                 break;
1543
1544                         case O_GID:
1545                         case O_UID:
1546                                 /*
1547                                  * We only check offset == 0 && proto != 0,
1548                                  * as this ensures that we have an IPv4
1549                                  * packet with the ports info.
1550                                  */
1551                                 if (offset!=0)
1552                                         break;
1553                             {
1554                                 struct inpcbinfo *pi;
1555                                 int wildcard;
1556                                 struct inpcb *pcb;
1557
1558                                 if (proto == IPPROTO_TCP) {
1559                                         wildcard = 0;
1560                                         pi = &tcbinfo[mycpu->gd_cpuid];
1561                                 } else if (proto == IPPROTO_UDP) {
1562                                         wildcard = 1;
1563                                         pi = &udbinfo;
1564                                 } else
1565                                         break;
1566
1567                                 pcb =  (oif) ?
1568                                         in_pcblookup_hash(pi,
1569                                             dst_ip, htons(dst_port),
1570                                             src_ip, htons(src_port),
1571                                             wildcard, oif) :
1572                                         in_pcblookup_hash(pi,
1573                                             src_ip, htons(src_port),
1574                                             dst_ip, htons(dst_port),
1575                                             wildcard, NULL);
1576
1577                                 if (pcb == NULL || pcb->inp_socket == NULL)
1578                                         break;
1579
1580                                 if (cmd->opcode == O_UID) {
1581 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1582                                         match =
1583                                           !socheckuid(pcb->inp_socket,
1584                                            (uid_t)((ipfw_insn_u32 *)cmd)->d[0]);
1585 #undef socheckuid
1586                                 } else  {
1587                                         match = groupmember(
1588                                             (uid_t)((ipfw_insn_u32 *)cmd)->d[0],
1589                                             pcb->inp_socket->so_cred);
1590                                 }
1591                             }
1592                                 break;
1593
1594                         case O_RECV:
1595                                 match = iface_match(m->m_pkthdr.rcvif,
1596                                     (ipfw_insn_if *)cmd);
1597                                 break;
1598
1599                         case O_XMIT:
1600                                 match = iface_match(oif, (ipfw_insn_if *)cmd);
1601                                 break;
1602
1603                         case O_VIA:
1604                                 match = iface_match(oif ? oif :
1605                                     m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1606                                 break;
1607
1608                         case O_MACADDR2:
1609                                 if (args->eh != NULL) { /* have MAC header */
1610                                         uint32_t *want = (uint32_t *)
1611                                                 ((ipfw_insn_mac *)cmd)->addr;
1612                                         uint32_t *mask = (uint32_t *)
1613                                                 ((ipfw_insn_mac *)cmd)->mask;
1614                                         uint32_t *hdr = (uint32_t *)args->eh;
1615
1616                                         match =
1617                                         (want[0] == (hdr[0] & mask[0]) &&
1618                                          want[1] == (hdr[1] & mask[1]) &&
1619                                          want[2] == (hdr[2] & mask[2]));
1620                                 }
1621                                 break;
1622
1623                         case O_MAC_TYPE:
1624                                 if (args->eh != NULL) {
1625                                         uint16_t t =
1626                                             ntohs(args->eh->ether_type);
1627                                         uint16_t *p =
1628                                             ((ipfw_insn_u16 *)cmd)->ports;
1629                                         int i;
1630
1631                                         /* Special vlan handling */
1632                                         if (m->m_flags & M_VLANTAG)
1633                                                 t = ETHERTYPE_VLAN;
1634
1635                                         for (i = cmdlen - 1; !match && i > 0;
1636                                              i--, p += 2) {
1637                                                 match =
1638                                                 (t >= p[0] && t <= p[1]);
1639                                         }
1640                                 }
1641                                 break;
1642
1643                         case O_FRAG:
1644                                 match = (hlen > 0 && offset != 0);
1645                                 break;
1646
1647                         case O_IN:      /* "out" is "not in" */
1648                                 match = (oif == NULL);
1649                                 break;
1650
1651                         case O_LAYER2:
1652                                 match = (args->eh != NULL);
1653                                 break;
1654
1655                         case O_PROTO:
1656                                 /*
1657                                  * We do not allow an arg of 0 so the
1658                                  * check of "proto" only suffices.
1659                                  */
1660                                 match = (proto == cmd->arg1);
1661                                 break;
1662
1663                         case O_IP_SRC:
1664                                 match = (hlen > 0 &&
1665                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1666                                     src_ip.s_addr);
1667                                 break;
1668
1669                         case O_IP_SRC_MASK:
1670                                 match = (hlen > 0 &&
1671                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1672                                      (src_ip.s_addr &
1673                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
1674                                 break;
1675
1676                         case O_IP_SRC_ME:
1677                                 if (hlen > 0) {
1678                                         struct ifnet *tif;
1679
1680                                         tif = INADDR_TO_IFP(&src_ip);
1681                                         match = (tif != NULL);
1682                                 }
1683                                 break;
1684
1685                         case O_IP_DST_SET:
1686                         case O_IP_SRC_SET:
1687                                 if (hlen > 0) {
1688                                         uint32_t *d = (uint32_t *)(cmd+1);
1689                                         uint32_t addr =
1690                                             cmd->opcode == O_IP_DST_SET ?
1691                                                 args->f_id.dst_ip :
1692                                                 args->f_id.src_ip;
1693
1694                                         if (addr < d[0])
1695                                                 break;
1696                                         addr -= d[0]; /* subtract base */
1697                                         match =
1698                                         (addr < cmd->arg1) &&
1699                                          (d[1 + (addr >> 5)] &
1700                                           (1 << (addr & 0x1f)));
1701                                 }
1702                                 break;
1703
1704                         case O_IP_DST:
1705                                 match = (hlen > 0 &&
1706                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1707                                     dst_ip.s_addr);
1708                                 break;
1709
1710                         case O_IP_DST_MASK:
1711                                 match = (hlen > 0) &&
1712                                     (((ipfw_insn_ip *)cmd)->addr.s_addr ==
1713                                      (dst_ip.s_addr &
1714                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
1715                                 break;
1716
1717                         case O_IP_DST_ME:
1718                                 if (hlen > 0) {
1719                                         struct ifnet *tif;
1720
1721                                         tif = INADDR_TO_IFP(&dst_ip);
1722                                         match = (tif != NULL);
1723                                 }
1724                                 break;
1725
1726                         case O_IP_SRCPORT:
1727                         case O_IP_DSTPORT:
1728                                 /*
1729                                  * offset == 0 && proto != 0 is enough
1730                                  * to guarantee that we have an IPv4
1731                                  * packet with port info.
1732                                  */
1733                                 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
1734                                     && offset == 0) {
1735                                         uint16_t x =
1736                                             (cmd->opcode == O_IP_SRCPORT) ?
1737                                                 src_port : dst_port ;
1738                                         uint16_t *p =
1739                                             ((ipfw_insn_u16 *)cmd)->ports;
1740                                         int i;
1741
1742                                         for (i = cmdlen - 1; !match && i > 0;
1743                                              i--, p += 2) {
1744                                                 match =
1745                                                 (x >= p[0] && x <= p[1]);
1746                                         }
1747                                 }
1748                                 break;
1749
1750                         case O_ICMPTYPE:
1751                                 match = (offset == 0 && proto==IPPROTO_ICMP &&
1752                                     icmptype_match(ip, (ipfw_insn_u32 *)cmd));
1753                                 break;
1754
1755                         case O_IPOPT:
1756                                 match = (hlen > 0 && ipopts_match(ip, cmd));
1757                                 break;
1758
1759                         case O_IPVER:
1760                                 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
1761                                 break;
1762
1763                         case O_IPTTL:
1764                                 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl);
1765                                 break;
1766
1767                         case O_IPID:
1768                                 match = (hlen > 0 &&
1769                                     cmd->arg1 == ntohs(ip->ip_id));
1770                                 break;
1771
1772                         case O_IPLEN:
1773                                 match = (hlen > 0 && cmd->arg1 == ip_len);
1774                                 break;
1775
1776                         case O_IPPRECEDENCE:
1777                                 match = (hlen > 0 &&
1778                                     (cmd->arg1 == (ip->ip_tos & 0xe0)));
1779                                 break;
1780
1781                         case O_IPTOS:
1782                                 match = (hlen > 0 &&
1783                                     flags_match(cmd, ip->ip_tos));
1784                                 break;
1785
1786                         case O_TCPFLAGS:
1787                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1788                                     flags_match(cmd,
1789                                         L3HDR(struct tcphdr,ip)->th_flags));
1790                                 break;
1791
1792                         case O_TCPOPTS:
1793                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1794                                     tcpopts_match(ip, cmd));
1795                                 break;
1796
1797                         case O_TCPSEQ:
1798                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1799                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
1800                                         L3HDR(struct tcphdr,ip)->th_seq);
1801                                 break;
1802
1803                         case O_TCPACK:
1804                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1805                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
1806                                         L3HDR(struct tcphdr,ip)->th_ack);
1807                                 break;
1808
1809                         case O_TCPWIN:
1810                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1811                                     cmd->arg1 ==
1812                                         L3HDR(struct tcphdr,ip)->th_win);
1813                                 break;
1814
1815                         case O_ESTAB:
1816                                 /* reject packets which have SYN only */
1817                                 /* XXX should i also check for TH_ACK ? */
1818                                 match = (proto == IPPROTO_TCP && offset == 0 &&
1819                                     (L3HDR(struct tcphdr,ip)->th_flags &
1820                                      (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
1821                                 break;
1822
1823                         case O_LOG:
1824                                 if (fw_verbose)
1825                                         ipfw_log(f, hlen, args->eh, m, oif);
1826                                 match = 1;
1827                                 break;
1828
1829                         case O_PROB:
1830                                 match = (krandom() <
1831                                         ((ipfw_insn_u32 *)cmd)->d[0]);
1832                                 break;
1833
1834                         /*
1835                          * The second set of opcodes represents 'actions',
1836                          * i.e. the terminal part of a rule once the packet
1837                          * matches all previous patterns.
1838                          * Typically there is only one action for each rule,
1839                          * and the opcode is stored at the end of the rule
1840                          * (but there are exceptions -- see below).
1841                          *
1842                          * In general, here we set retval and terminate the
1843                          * outer loop (would be a 'break 3' in some language,
1844                          * but we need to do a 'goto done').
1845                          *
1846                          * Exceptions:
1847                          * O_COUNT and O_SKIPTO actions:
1848                          *   instead of terminating, we jump to the next rule
1849                          *   ('goto next_rule', equivalent to a 'break 2'),
1850                          *   or to the SKIPTO target ('goto again' after
1851                          *   having set f, cmd and l), respectively.
1852                          *
1853                          * O_LIMIT and O_KEEP_STATE: these opcodes are
1854                          *   not real 'actions', and are stored right
1855                          *   before the 'action' part of the rule.
1856                          *   These opcodes try to install an entry in the
1857                          *   state tables; if successful, we continue with
1858                          *   the next opcode (match=1; break;), otherwise
1859                          *   the packet *   must be dropped
1860                          *   ('goto done' after setting retval);
1861                          *
1862                          * O_PROBE_STATE and O_CHECK_STATE: these opcodes
1863                          *   cause a lookup of the state table, and a jump
1864                          *   to the 'action' part of the parent rule
1865                          *   ('goto check_body') if an entry is found, or
1866                          *   (CHECK_STATE only) a jump to the next rule if
1867                          *   the entry is not found ('goto next_rule').
1868                          *   The result of the lookup is cached to make
1869                          *   further instances of these opcodes are
1870                          *   effectively NOPs.
1871                          */
1872                         case O_LIMIT:
1873                         case O_KEEP_STATE:
1874                                 if (install_state(f,
1875                                     (ipfw_insn_limit *)cmd, args)) {
1876                                         retval = IP_FW_PORT_DENY_FLAG;
1877                                         goto done; /* error/limit violation */
1878                                 }
1879                                 match = 1;
1880                                 break;
1881
1882                         case O_PROBE_STATE:
1883                         case O_CHECK_STATE:
1884                                 /*
1885                                  * dynamic rules are checked at the first
1886                                  * keep-state or check-state occurrence,
1887                                  * with the result being stored in dyn_dir.
1888                                  * The compiler introduces a PROBE_STATE
1889                                  * instruction for us when we have a
1890                                  * KEEP_STATE (because PROBE_STATE needs
1891                                  * to be run first).
1892                                  */
1893                                 if (dyn_dir == MATCH_UNKNOWN &&
1894                                     (q = lookup_dyn_rule(&args->f_id,
1895                                      &dyn_dir, proto == IPPROTO_TCP ?
1896                                         L3HDR(struct tcphdr, ip) : NULL))
1897                                         != NULL) {
1898                                         /*
1899                                          * Found dynamic entry, update stats
1900                                          * and jump to the 'action' part of
1901                                          * the parent rule.
1902                                          */
1903                                         q->pcnt++;
1904                                         q->bcnt += ip_len;
1905                                         f = q->rule;
1906                                         cmd = ACTION_PTR(f);
1907                                         l = f->cmd_len - f->act_ofs;
1908                                         goto check_body;
1909                                 }
1910                                 /*
1911                                  * Dynamic entry not found. If CHECK_STATE,
1912                                  * skip to next rule, if PROBE_STATE just
1913                                  * ignore and continue with next opcode.
1914                                  */
1915                                 if (cmd->opcode == O_CHECK_STATE)
1916                                         goto next_rule;
1917                                 match = 1;
1918                                 break;
1919
1920                         case O_ACCEPT:
1921                                 retval = 0;     /* accept */
1922                                 goto done;
1923
1924                         case O_PIPE:
1925                         case O_QUEUE:
1926                                 args->rule = f; /* report matching rule */
1927                                 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG;
1928                                 goto done;
1929
1930                         case O_DIVERT:
1931                         case O_TEE:
1932                                 if (args->eh) /* not on layer 2 */
1933                                         break;
1934
1935                                 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT,
1936                                                  sizeof(uint16_t), MB_DONTWAIT);
1937                                 if (mtag == NULL) {
1938                                         retval = IP_FW_PORT_DENY_FLAG;
1939                                         goto done;
1940                                 }
1941                                 *(uint16_t *)m_tag_data(mtag) = f->rulenum;
1942                                 m_tag_prepend(m, mtag);
1943                                 retval = (cmd->opcode == O_DIVERT) ?
1944                                     cmd->arg1 :
1945                                     cmd->arg1 | IP_FW_PORT_TEE_FLAG;
1946                                 goto done;
1947
1948                         case O_COUNT:
1949                         case O_SKIPTO:
1950                                 f->pcnt++;      /* update stats */
1951                                 f->bcnt += ip_len;
1952                                 f->timestamp = time_second;
1953                                 if (cmd->opcode == O_COUNT)
1954                                         goto next_rule;
1955                                 /* handle skipto */
1956                                 if (f->next_rule == NULL)
1957                                         lookup_next_rule(f);
1958                                 f = f->next_rule;
1959                                 goto again;
1960
1961                         case O_REJECT:
1962                                 /*
1963                                  * Drop the packet and send a reject notice
1964                                  * if the packet is not ICMP (or is an ICMP
1965                                  * query), and it is not multicast/broadcast.
1966                                  */
1967                                 if (hlen > 0 &&
1968                                     (proto != IPPROTO_ICMP ||
1969                                      is_icmp_query(ip)) &&
1970                                     !(m->m_flags & (M_BCAST|M_MCAST)) &&
1971                                     !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
1972                                         send_reject(args, cmd->arg1,
1973                                             offset,ip_len);
1974                                         m = args->m;
1975                                 }
1976                                 /* FALLTHROUGH */
1977                         case O_DENY:
1978                                 retval = IP_FW_PORT_DENY_FLAG;
1979                                 goto done;
1980
1981                         case O_FORWARD_IP:
1982                                 if (args->eh)   /* not valid on layer2 pkts */
1983                                         break;
1984                                 if (!q || dyn_dir == MATCH_FORWARD) {
1985                                         args->next_hop =
1986                                             &((ipfw_insn_sa *)cmd)->sa;
1987                                 }
1988                                 retval = 0;
1989                                 goto done;
1990
1991                         default:
1992                                 panic("-- unknown opcode %d\n", cmd->opcode);
1993                         } /* end of switch() on opcodes */
1994
1995                         if (cmd->len & F_NOT)
1996                                 match = !match;
1997
1998                         if (match) {
1999                                 if (cmd->len & F_OR)
2000                                         skip_or = 1;
2001                         } else {
2002                                 if (!(cmd->len & F_OR)) /* not an OR block, */
2003                                         break;          /* try next rule    */
2004                         }
2005
2006                 }       /* end of inner for, scan opcodes */
2007
2008 next_rule:;             /* try next rule                */
2009
2010         }               /* end of outer for, scan rules */
2011         kprintf("+++ ipfw: ouch!, skip past end of rules, denying packet\n");
2012         return(IP_FW_PORT_DENY_FLAG);
2013
2014 done:
2015         /* Update statistics */
2016         f->pcnt++;
2017         f->bcnt += ip_len;
2018         f->timestamp = time_second;
2019         return retval;
2020
2021 pullup_failed:
2022         if (fw_verbose)
2023                 kprintf("pullup failed\n");
2024         return(IP_FW_PORT_DENY_FLAG);
2025 }
2026
2027 static void
2028 ipfw_dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa)
2029 {
2030         struct m_tag *mtag;
2031         struct dn_pkt *pkt;
2032         ipfw_insn *cmd;
2033         const struct ipfw_flow_id *id;
2034         struct dn_flow_id *fid;
2035
2036         M_ASSERTPKTHDR(m);
2037
2038         mtag = m_tag_get(PACKET_TAG_DUMMYNET, sizeof(*pkt), MB_DONTWAIT);
2039         if (mtag == NULL) {
2040                 m_freem(m);
2041                 return;
2042         }
2043         m_tag_prepend(m, mtag);
2044
2045         pkt = m_tag_data(mtag);
2046         bzero(pkt, sizeof(*pkt));
2047
2048         cmd = fwa->rule->cmd + fwa->rule->act_ofs;
2049         if (cmd->opcode == O_LOG)
2050                 cmd += F_LEN(cmd);
2051         KASSERT(cmd->opcode == O_PIPE || cmd->opcode == O_QUEUE,
2052                 ("Rule is not PIPE or QUEUE, opcode %d\n", cmd->opcode));
2053
2054         pkt->dn_m = m;
2055         pkt->dn_flags = (dir & DN_FLAGS_DIR_MASK);
2056         pkt->ifp = fwa->oif;
2057         pkt->cpuid = mycpu->gd_cpuid;
2058         pkt->pipe_nr = pipe_nr;
2059
2060         id = &fwa->f_id;
2061         fid = &pkt->id;
2062         fid->fid_dst_ip = id->dst_ip;
2063         fid->fid_src_ip = id->src_ip;
2064         fid->fid_dst_port = id->dst_port;
2065         fid->fid_src_port = id->src_port;
2066         fid->fid_proto = id->proto;
2067         fid->fid_flags = id->flags;
2068
2069         ipfw_ref_rule(fwa->rule);
2070         pkt->dn_priv = fwa->rule;
2071         pkt->dn_unref_priv = ipfw_unref_rule;
2072
2073         if (cmd->opcode == O_PIPE)
2074                 pkt->dn_flags |= DN_FLAGS_IS_PIPE;
2075
2076         if (dir == DN_TO_IP_OUT) {
2077                 /*
2078                  * We need to copy *ro because for ICMP pkts (and maybe
2079                  * others) the caller passed a pointer into the stack;
2080                  * dst might also be a pointer into *ro so it needs to
2081                  * be updated.
2082                  */
2083                 pkt->ro = *(fwa->ro);
2084                 if (fwa->ro->ro_rt)
2085                         fwa->ro->ro_rt->rt_refcnt++;
2086                 if (fwa->dst == (struct sockaddr_in *)&fwa->ro->ro_dst) {
2087                         /* 'dst' points into 'ro' */
2088                         fwa->dst = (struct sockaddr_in *)&(pkt->ro.ro_dst);
2089                 }
2090                 pkt->dn_dst = fwa->dst;
2091                 pkt->flags = fwa->flags;
2092         }
2093
2094         m->m_pkthdr.fw_flags |= DUMMYNET_MBUF_TAGGED;
2095         ip_dn_queue(m);
2096 }
2097
2098 /*
2099  * When a rule is added/deleted, clear the next_rule pointers in all rules.
2100  * These will be reconstructed on the fly as packets are matched.
2101  * Must be called at splimp().
2102  */
2103 static void
2104 flush_rule_ptrs(void)
2105 {
2106         struct ip_fw *rule;
2107
2108         for (rule = layer3_chain; rule; rule = rule->next)
2109                 rule->next_rule = NULL;
2110 }
2111
2112 static __inline void
2113 ipfw_inc_static_count(struct ip_fw *rule)
2114 {
2115         static_count++;
2116         static_ioc_len += IOC_RULESIZE(rule);
2117 }
2118
2119 static __inline void
2120 ipfw_dec_static_count(struct ip_fw *rule)
2121 {
2122         int l = IOC_RULESIZE(rule);
2123
2124         KASSERT(static_count > 0, ("invalid static count %u\n", static_count));
2125         static_count--;
2126
2127         KASSERT(static_ioc_len >= l,
2128                 ("invalid static len %u\n", static_ioc_len));
2129         static_ioc_len -= l;
2130 }
2131
2132 static struct ip_fw *
2133 ipfw_create_rule(const struct ipfw_ioc_rule *ioc_rule)
2134 {
2135         struct ip_fw *rule;
2136
2137         rule = kmalloc(RULESIZE(ioc_rule), M_IPFW, M_WAITOK | M_ZERO);
2138
2139         rule->act_ofs = ioc_rule->act_ofs;
2140         rule->cmd_len = ioc_rule->cmd_len;
2141         rule->rulenum = ioc_rule->rulenum;
2142         rule->set = ioc_rule->set;
2143         rule->usr_flags = ioc_rule->usr_flags;
2144
2145         bcopy(ioc_rule->cmd, rule->cmd, rule->cmd_len * 4 /* XXX */);
2146
2147         rule->refcnt = 1;
2148
2149         return rule;
2150 }
2151
2152 /*
2153  * Add a new rule to the list. Copy the rule into a malloc'ed area, then
2154  * possibly create a rule number and add the rule to the list.
2155  * Update the rule_number in the input struct so the caller knows it as well.
2156  */
2157 static int
2158 ipfw_add_rule(struct ip_fw **head, struct ipfw_ioc_rule *ioc_rule)
2159 {
2160         struct ip_fw *rule, *f, *prev;
2161
2162         KKASSERT(*head != NULL);
2163
2164         rule = ipfw_create_rule(ioc_rule);
2165
2166         crit_enter();
2167
2168         /*
2169          * If rulenum is 0, find highest numbered rule before the
2170          * default rule, and add autoinc_step
2171          */
2172         if (autoinc_step < 1)
2173                 autoinc_step = 1;
2174         else if (autoinc_step > 1000)
2175                 autoinc_step = 1000;
2176         if (rule->rulenum == 0) {
2177                 /*
2178                  * locate the highest numbered rule before default
2179                  */
2180                 for (f = *head; f; f = f->next) {
2181                         if (f->rulenum == IPFW_DEFAULT_RULE)
2182                                 break;
2183                         rule->rulenum = f->rulenum;
2184                 }
2185                 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
2186                         rule->rulenum += autoinc_step;
2187                 ioc_rule->rulenum = rule->rulenum;
2188         }
2189
2190         /*
2191          * Now insert the new rule in the right place in the sorted list.
2192          */
2193         for (prev = NULL, f = *head; f; prev = f, f = f->next) {
2194                 if (f->rulenum > rule->rulenum) { /* found the location */
2195                         if (prev) {
2196                                 rule->next = f;
2197                                 prev->next = rule;
2198                         } else { /* head insert */
2199                                 rule->next = *head;
2200                                 *head = rule;
2201                         }
2202                         break;
2203                 }
2204         }
2205
2206         flush_rule_ptrs();
2207         ipfw_inc_static_count(rule);
2208
2209         crit_exit();
2210
2211         DEB(kprintf("++ installed rule %d, static count now %d\n",
2212                 rule->rulenum, static_count);)
2213         return (0);
2214 }
2215
2216 /**
2217  * Free storage associated with a static rule (including derived
2218  * dynamic rules).
2219  * The caller is in charge of clearing rule pointers to avoid
2220  * dangling pointers.
2221  * @return a pointer to the next entry.
2222  * Arguments are not checked, so they better be correct.
2223  * Must be called at splimp().
2224  */
2225 static struct ip_fw *
2226 delete_rule(struct ip_fw **head, struct ip_fw *prev, struct ip_fw *rule)
2227 {
2228         struct ip_fw *n;
2229
2230         n = rule->next;
2231         remove_dyn_rule(rule, NULL /* force removal */);
2232         if (prev == NULL)
2233                 *head = n;
2234         else
2235                 prev->next = n;
2236         ipfw_dec_static_count(rule);
2237
2238         /* Mark the rule as invalid */
2239         rule->rule_flags |= IPFW_RULE_F_INVALID;
2240         rule->next_rule = NULL;
2241
2242         /* Try to free this rule */
2243         ipfw_free_rule(rule);
2244
2245         return n;
2246 }
2247
2248 /*
2249  * Deletes all rules from a chain (including the default rule
2250  * if the second argument is set).
2251  * Must be called at splimp().
2252  */
2253 static void
2254 free_chain(struct ip_fw **chain, int kill_default)
2255 {
2256         struct ip_fw *rule;
2257
2258         flush_rule_ptrs(); /* more efficient to do outside the loop */
2259
2260         while ((rule = *chain) != NULL &&
2261                (kill_default || rule->rulenum != IPFW_DEFAULT_RULE))
2262                 delete_rule(chain, NULL, rule);
2263
2264         KASSERT(dyn_count == 0, ("%u dyn rule remains\n", dyn_count));
2265
2266         if (kill_default) {
2267                 ip_fw_default_rule = NULL;      /* Reset default rule */
2268
2269                 if (ipfw_dyn_v != NULL) {
2270                         /*
2271                          * Free dynamic rules(state) hash table
2272                          */
2273                         kfree(ipfw_dyn_v, M_IPFW);
2274                         ipfw_dyn_v = NULL;
2275                 }
2276
2277                 KASSERT(static_count == 0,
2278                         ("%u static rules remains\n", static_count));
2279                 KASSERT(static_ioc_len == 0,
2280                         ("%u bytes of static rules remains\n", static_ioc_len));
2281         } else {
2282                 KASSERT(static_count == 1,
2283                         ("%u static rules remains\n", static_count));
2284                 KASSERT(static_ioc_len == IOC_RULESIZE(ip_fw_default_rule),
2285                         ("%u bytes of static rules remains, should be %u\n",
2286                          static_ioc_len, IOC_RULESIZE(ip_fw_default_rule)));
2287         }
2288 }
2289
2290 /**
2291  * Remove all rules with given number, and also do set manipulation.
2292  *
2293  * The argument is an uint32_t. The low 16 bit are the rule or set number,
2294  * the next 8 bits are the new set, the top 8 bits are the command:
2295  *
2296  *      0       delete rules with given number
2297  *      1       delete rules with given set number
2298  *      2       move rules with given number to new set
2299  *      3       move rules with given set number to new set
2300  *      4       swap sets with given numbers
2301  */
2302 static int
2303 del_entry(struct ip_fw **chain, uint32_t arg)
2304 {
2305         struct ip_fw *prev, *rule;
2306         uint16_t rulenum;
2307         uint8_t cmd, new_set;
2308
2309         rulenum = arg & 0xffff;
2310         cmd = (arg >> 24) & 0xff;
2311         new_set = (arg >> 16) & 0xff;
2312
2313         if (cmd > 4)
2314                 return EINVAL;
2315         if (new_set > 30)
2316                 return EINVAL;
2317         if (cmd == 0 || cmd == 2) {
2318                 if (rulenum == IPFW_DEFAULT_RULE)
2319                         return EINVAL;
2320         } else {
2321                 if (rulenum > 30)
2322                         return EINVAL;
2323         }
2324
2325         switch (cmd) {
2326         case 0: /* delete rules with given number */
2327                 /*
2328                  * locate first rule to delete
2329                  */
2330                 for (prev = NULL, rule = *chain;
2331                      rule && rule->rulenum < rulenum;
2332                      prev = rule, rule = rule->next)
2333                         ;
2334                 if (rule->rulenum != rulenum)
2335                         return EINVAL;
2336
2337                 crit_enter(); /* no access to rules while removing */
2338                 /*
2339                  * flush pointers outside the loop, then delete all matching
2340                  * rules. prev remains the same throughout the cycle.
2341                  */
2342                 flush_rule_ptrs();
2343                 while (rule && rule->rulenum == rulenum)
2344                         rule = delete_rule(chain, prev, rule);
2345                 crit_exit();
2346                 break;
2347
2348         case 1: /* delete all rules with given set number */
2349                 crit_enter();
2350                 flush_rule_ptrs();
2351                 for (prev = NULL, rule = *chain; rule;) {
2352                         if (rule->set == rulenum) {
2353                                 rule = delete_rule(chain, prev, rule);
2354                         } else {
2355                                 prev = rule;
2356                                 rule = rule->next;
2357                         }
2358                 }
2359                 crit_exit();
2360                 break;
2361
2362         case 2: /* move rules with given number to new set */
2363                 crit_enter();
2364                 for (rule = *chain; rule; rule = rule->next) {
2365                         if (rule->rulenum == rulenum)
2366                                 rule->set = new_set;
2367                 }
2368                 crit_exit();
2369                 break;
2370
2371         case 3: /* move rules with given set number to new set */
2372                 crit_enter();
2373                 for (rule = *chain; rule; rule = rule->next) {
2374                         if (rule->set == rulenum)
2375                                 rule->set = new_set;
2376                 }
2377                 crit_exit();
2378                 break;
2379
2380         case 4: /* swap two sets */
2381                 crit_enter();
2382                 for (rule = *chain; rule; rule = rule->next) {
2383                         if (rule->set == rulenum)
2384                                 rule->set = new_set;
2385                         else if (rule->set == new_set)
2386                                 rule->set = rulenum;
2387                 }
2388                 crit_exit();
2389                 break;
2390         }
2391         return 0;
2392 }
2393
2394 /*
2395  * Clear counters for a specific rule.
2396  */
2397 static void
2398 clear_counters(struct ip_fw *rule, int log_only)
2399 {
2400         ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
2401
2402         if (log_only == 0) {
2403                 rule->bcnt = rule->pcnt = 0;
2404                 rule->timestamp = 0;
2405         }
2406         if (l->o.opcode == O_LOG)
2407                 l->log_left = l->max_log;
2408 }
2409
2410 /**
2411  * Reset some or all counters on firewall rules.
2412  * @arg frwl is null to clear all entries, or contains a specific
2413  * rule number.
2414  * @arg log_only is 1 if we only want to reset logs, zero otherwise.
2415  */
2416 static int
2417 zero_entry(int rulenum, int log_only)
2418 {
2419         struct ip_fw *rule;
2420         char *msg;
2421
2422         if (rulenum == 0) {
2423                 crit_enter();
2424                 norule_counter = 0;
2425                 for (rule = layer3_chain; rule; rule = rule->next)
2426                         clear_counters(rule, log_only);
2427                 crit_exit();
2428                 msg = log_only ? "ipfw: All logging counts reset.\n"
2429                                : "ipfw: Accounting cleared.\n";
2430         } else {
2431                 int cleared = 0;
2432
2433                 /*
2434                  * We can have multiple rules with the same number, so we
2435                  * need to clear them all.
2436                  */
2437                 for (rule = layer3_chain; rule; rule = rule->next) {
2438                         if (rule->rulenum == rulenum) {
2439                                 crit_enter();
2440                                 while (rule && rule->rulenum == rulenum) {
2441                                         clear_counters(rule, log_only);
2442                                         rule = rule->next;
2443                                 }
2444                                 crit_exit();
2445                                 cleared = 1;
2446                                 break;
2447                         }
2448                 }
2449                 if (!cleared)   /* we did not find any matching rules */
2450                         return (EINVAL);
2451                 msg = log_only ? "ipfw: Entry %d logging count reset.\n"
2452                                : "ipfw: Entry %d cleared.\n";
2453         }
2454         if (fw_verbose)
2455                 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
2456         return (0);
2457 }
2458
2459 /*
2460  * Check validity of the structure before insert.
2461  * Fortunately rules are simple, so this mostly need to check rule sizes.
2462  */
2463 static int
2464 ipfw_ctl_check_rule(struct ipfw_ioc_rule *rule, int size)
2465 {
2466         int l, cmdlen = 0;
2467         int have_action = 0;
2468         ipfw_insn *cmd;
2469
2470         /* Check for valid size */
2471         if (size < sizeof(*rule)) {
2472                 kprintf("ipfw: rule too short\n");
2473                 return EINVAL;
2474         }
2475         l = IOC_RULESIZE(rule);
2476         if (l != size) {
2477                 kprintf("ipfw: size mismatch (have %d want %d)\n", size, l);
2478                 return EINVAL;
2479         }
2480
2481         /*
2482          * Now go for the individual checks. Very simple ones, basically only
2483          * instruction sizes.
2484          */
2485         for (l = rule->cmd_len, cmd = rule->cmd; l > 0;
2486              l -= cmdlen, cmd += cmdlen) {
2487                 cmdlen = F_LEN(cmd);
2488                 if (cmdlen > l) {
2489                         kprintf("ipfw: opcode %d size truncated\n",
2490                                 cmd->opcode);
2491                         return EINVAL;
2492                 }
2493                 DEB(kprintf("ipfw: opcode %d\n", cmd->opcode);)
2494                 switch (cmd->opcode) {
2495                 case O_NOP:
2496                 case O_PROBE_STATE:
2497                 case O_KEEP_STATE:
2498                 case O_PROTO:
2499                 case O_IP_SRC_ME:
2500                 case O_IP_DST_ME:
2501                 case O_LAYER2:
2502                 case O_IN:
2503                 case O_FRAG:
2504                 case O_IPOPT:
2505                 case O_IPLEN:
2506                 case O_IPID:
2507                 case O_IPTOS:
2508                 case O_IPPRECEDENCE:
2509                 case O_IPTTL:
2510                 case O_IPVER:
2511                 case O_TCPWIN:
2512                 case O_TCPFLAGS:
2513                 case O_TCPOPTS:
2514                 case O_ESTAB:
2515                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
2516                                 goto bad_size;
2517                         break;
2518
2519                 case O_UID:
2520                 case O_GID:
2521                 case O_IP_SRC:
2522                 case O_IP_DST:
2523                 case O_TCPSEQ:
2524                 case O_TCPACK:
2525                 case O_PROB:
2526                 case O_ICMPTYPE:
2527                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
2528                                 goto bad_size;
2529                         break;
2530
2531                 case O_LIMIT:
2532                         if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
2533                                 goto bad_size;
2534                         break;
2535
2536                 case O_LOG:
2537                         if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
2538                                 goto bad_size;
2539
2540                         ((ipfw_insn_log *)cmd)->log_left =
2541                             ((ipfw_insn_log *)cmd)->max_log;
2542
2543                         break;
2544
2545                 case O_IP_SRC_MASK:
2546                 case O_IP_DST_MASK:
2547                         if (cmdlen != F_INSN_SIZE(ipfw_insn_ip))
2548                                 goto bad_size;
2549                         if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) {
2550                                 kprintf("ipfw: opcode %d, useless rule\n",
2551                                         cmd->opcode);
2552                                 return EINVAL;
2553                         }
2554                         break;
2555
2556                 case O_IP_SRC_SET:
2557                 case O_IP_DST_SET:
2558                         if (cmd->arg1 == 0 || cmd->arg1 > 256) {
2559                                 kprintf("ipfw: invalid set size %d\n",
2560                                         cmd->arg1);
2561                                 return EINVAL;
2562                         }
2563                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
2564                             (cmd->arg1+31)/32 )
2565                                 goto bad_size;
2566                         break;
2567
2568                 case O_MACADDR2:
2569                         if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
2570                                 goto bad_size;
2571                         break;
2572
2573                 case O_MAC_TYPE:
2574                 case O_IP_SRCPORT:
2575                 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
2576                         if (cmdlen < 2 || cmdlen > 31)
2577                                 goto bad_size;
2578                         break;
2579
2580                 case O_RECV:
2581                 case O_XMIT:
2582                 case O_VIA:
2583                         if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
2584                                 goto bad_size;
2585                         break;
2586
2587                 case O_PIPE:
2588                 case O_QUEUE:
2589                         if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
2590                                 goto bad_size;
2591                         goto check_action;
2592
2593                 case O_FORWARD_IP:
2594                         if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
2595                                 goto bad_size;
2596                         goto check_action;
2597
2598                 case O_FORWARD_MAC: /* XXX not implemented yet */
2599                 case O_CHECK_STATE:
2600                 case O_COUNT:
2601                 case O_ACCEPT:
2602                 case O_DENY:
2603                 case O_REJECT:
2604                 case O_SKIPTO:
2605                 case O_DIVERT:
2606                 case O_TEE:
2607                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
2608                                 goto bad_size;
2609 check_action:
2610                         if (have_action) {
2611                                 kprintf("ipfw: opcode %d, multiple actions"
2612                                         " not allowed\n",
2613                                         cmd->opcode);
2614                                 return EINVAL;
2615                         }
2616                         have_action = 1;
2617                         if (l != cmdlen) {
2618                                 kprintf("ipfw: opcode %d, action must be"
2619                                         " last opcode\n",
2620                                         cmd->opcode);
2621                                 return EINVAL;
2622                         }
2623                         break;
2624                 default:
2625                         kprintf("ipfw: opcode %d, unknown opcode\n",
2626                                 cmd->opcode);
2627                         return EINVAL;
2628                 }
2629         }
2630         if (have_action == 0) {
2631                 kprintf("ipfw: missing action\n");
2632                 return EINVAL;
2633         }
2634         return 0;
2635
2636 bad_size:
2637         kprintf("ipfw: opcode %d size %d wrong\n",
2638                 cmd->opcode, cmdlen);
2639         return EINVAL;
2640 }
2641
2642 static int
2643 ipfw_ctl_add_rule(struct sockopt *sopt)
2644 {
2645         struct ipfw_ioc_rule *ioc_rule;
2646         size_t size;
2647         int error;
2648         
2649         size = sopt->sopt_valsize;
2650         if (size > (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX) ||
2651             size < sizeof(*ioc_rule)) {
2652                 return EINVAL;
2653         }
2654         if (size != (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX)) {
2655                 sopt->sopt_val = krealloc(sopt->sopt_val, sizeof(uint32_t) *
2656                                           IPFW_RULE_SIZE_MAX, M_TEMP, M_WAITOK);
2657         }
2658         ioc_rule = sopt->sopt_val;
2659
2660         error = ipfw_ctl_check_rule(ioc_rule, size);
2661         if (error)
2662                 return error;
2663
2664         error = ipfw_add_rule(&layer3_chain, ioc_rule);
2665         if (error)
2666                 return error;
2667
2668         if (sopt->sopt_dir == SOPT_GET)
2669                 sopt->sopt_valsize = IOC_RULESIZE(ioc_rule);
2670         return 0;
2671 }
2672
2673 static void *
2674 ipfw_copy_rule(const struct ip_fw *rule, struct ipfw_ioc_rule *ioc_rule)
2675 {
2676         ioc_rule->act_ofs = rule->act_ofs;
2677         ioc_rule->cmd_len = rule->cmd_len;
2678         ioc_rule->rulenum = rule->rulenum;
2679         ioc_rule->set = rule->set;
2680         ioc_rule->usr_flags = rule->usr_flags;
2681
2682         ioc_rule->set_disable = set_disable;
2683         ioc_rule->static_count = static_count;
2684         ioc_rule->static_len = static_ioc_len;
2685
2686         ioc_rule->pcnt = rule->pcnt;
2687         ioc_rule->bcnt = rule->bcnt;
2688         ioc_rule->timestamp = rule->timestamp;
2689
2690         bcopy(rule->cmd, ioc_rule->cmd, ioc_rule->cmd_len * 4 /* XXX */);
2691
2692         return ((uint8_t *)ioc_rule + IOC_RULESIZE(ioc_rule));
2693 }
2694
2695 static void
2696 ipfw_copy_state(const ipfw_dyn_rule *dyn_rule,
2697                 struct ipfw_ioc_state *ioc_state)
2698 {
2699         const struct ipfw_flow_id *id;
2700         struct ipfw_ioc_flowid *ioc_id;
2701
2702         ioc_state->expire = TIME_LEQ(dyn_rule->expire, time_second) ?
2703                             0 : dyn_rule->expire - time_second;
2704         ioc_state->pcnt = dyn_rule->pcnt;
2705         ioc_state->bcnt = dyn_rule->bcnt;
2706
2707         ioc_state->dyn_type = dyn_rule->dyn_type;
2708         ioc_state->count = dyn_rule->count;
2709
2710         ioc_state->rulenum = dyn_rule->rule->rulenum;
2711
2712         id = &dyn_rule->id;
2713         ioc_id = &ioc_state->id;
2714
2715         ioc_id->type = ETHERTYPE_IP;
2716         ioc_id->u.ip.dst_ip = id->dst_ip;
2717         ioc_id->u.ip.src_ip = id->src_ip;
2718         ioc_id->u.ip.dst_port = id->dst_port;
2719         ioc_id->u.ip.src_port = id->src_port;
2720         ioc_id->u.ip.proto = id->proto;
2721 }
2722
2723 static int
2724 ipfw_ctl_get_rules(struct sockopt *sopt)
2725 {
2726         struct ip_fw *rule;
2727         void *bp;
2728         size_t size;
2729
2730         /*
2731          * pass up a copy of the current rules. Static rules
2732          * come first (the last of which has number IPFW_DEFAULT_RULE),
2733          * followed by a possibly empty list of dynamic rule.
2734          */
2735         crit_enter();
2736
2737         size = static_ioc_len;  /* size of static rules */
2738         if (ipfw_dyn_v)         /* add size of dyn.rules */
2739                 size += (dyn_count * sizeof(struct ipfw_ioc_state));
2740
2741         if (sopt->sopt_valsize < size) {
2742                 /* short length, no need to return incomplete rules */
2743                 /* XXX: if superuser, no need to zero buffer */
2744                 bzero(sopt->sopt_val, sopt->sopt_valsize); 
2745                 return 0;
2746         }
2747         bp = sopt->sopt_val;
2748
2749         for (rule = layer3_chain; rule; rule = rule->next)
2750                 bp = ipfw_copy_rule(rule, bp);
2751
2752         if (ipfw_dyn_v) {
2753                 struct ipfw_ioc_state *ioc_state;
2754                 int i;
2755
2756                 ioc_state = bp;
2757                 for (i = 0; i < curr_dyn_buckets; i++) {
2758                         ipfw_dyn_rule *p;
2759
2760                         for (p = ipfw_dyn_v[i]; p != NULL;
2761                              p = p->next, ioc_state++)
2762                                 ipfw_copy_state(p, ioc_state);
2763                 }
2764         }
2765
2766         crit_exit();
2767
2768         sopt->sopt_valsize = size;
2769         return 0;
2770 }
2771
2772 /**
2773  * {set|get}sockopt parser.
2774  */
2775 static int
2776 ipfw_ctl(struct sockopt *sopt)
2777 {
2778         int error, rulenum;
2779         uint32_t *masks;
2780         size_t size;
2781
2782         error = 0;
2783
2784         switch (sopt->sopt_name) {
2785         case IP_FW_GET:
2786                 error = ipfw_ctl_get_rules(sopt);
2787                 break;
2788
2789         case IP_FW_FLUSH:
2790                 /*
2791                  * Normally we cannot release the lock on each iteration.
2792                  * We could do it here only because we start from the head all
2793                  * the times so there is no risk of missing some entries.
2794                  * On the other hand, the risk is that we end up with
2795                  * a very inconsistent ruleset, so better keep the lock
2796                  * around the whole cycle.
2797                  *
2798                  * XXX this code can be improved by resetting the head of
2799                  * the list to point to the default rule, and then freeing
2800                  * the old list without the need for a lock.
2801                  */
2802
2803                 crit_enter();
2804                 free_chain(&layer3_chain, 0 /* keep default rule */);
2805                 crit_exit();
2806                 break;
2807
2808         case IP_FW_ADD:
2809                 error = ipfw_ctl_add_rule(sopt);
2810                 break;
2811
2812         case IP_FW_DEL:
2813                 /*
2814                  * IP_FW_DEL is used for deleting single rules or sets,
2815                  * and (ab)used to atomically manipulate sets. Argument size
2816                  * is used to distinguish between the two:
2817                  *    sizeof(uint32_t)
2818                  *      delete single rule or set of rules,
2819                  *      or reassign rules (or sets) to a different set.
2820                  *    2*sizeof(uint32_t)
2821                  *      atomic disable/enable sets.
2822                  *      first uint32_t contains sets to be disabled,
2823                  *      second uint32_t contains sets to be enabled.
2824                  */
2825                 masks = sopt->sopt_val;
2826                 size = sopt->sopt_valsize;
2827                 if (size == sizeof(*masks)) {
2828                         /*
2829                          * Delete or reassign static rule
2830                          */
2831                         error = del_entry(&layer3_chain, masks[0]);
2832                 } else if (size == (2 * sizeof(*masks))) {
2833                         /*
2834                          * Set enable/disable
2835                          */
2836                         crit_enter();
2837
2838                         set_disable =
2839                             (set_disable | masks[0]) & ~masks[1] &
2840                             ~(1 << 31); /* set 31 always enabled */
2841
2842                         crit_exit();
2843                 } else {
2844                         error = EINVAL;
2845                 }
2846                 break;
2847
2848         case IP_FW_ZERO:
2849         case IP_FW_RESETLOG: /* argument is an int, the rule number */
2850                 rulenum=0;
2851
2852                 if (sopt->sopt_val != 0) {
2853                     error = soopt_to_kbuf(sopt, &rulenum,
2854                             sizeof(int), sizeof(int));
2855                     if (error)
2856                         break;
2857                 }
2858                 error = zero_entry(rulenum, sopt->sopt_name == IP_FW_RESETLOG);
2859                 break;
2860
2861         default:
2862                 kprintf("ipfw_ctl invalid option %d\n", sopt->sopt_name);
2863                 error = EINVAL;
2864         }
2865         return error;
2866 }
2867
2868 /*
2869  * This procedure is only used to handle keepalives. It is invoked
2870  * every dyn_keepalive_period
2871  */
2872 static void
2873 ipfw_tick(void *unused __unused)
2874 {
2875         int i;
2876         ipfw_dyn_rule *q;
2877
2878         if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
2879                 goto done;
2880
2881         crit_enter();
2882         for (i = 0; i < curr_dyn_buckets; i++) {
2883                 for (q = ipfw_dyn_v[i]; q; q = q->next) {
2884                         if (q->dyn_type == O_LIMIT_PARENT)
2885                                 continue;
2886                         if (q->id.proto != IPPROTO_TCP)
2887                                 continue;
2888                         if ((q->state & BOTH_SYN) != BOTH_SYN)
2889                                 continue;
2890                         if (TIME_LEQ(time_second + dyn_keepalive_interval,
2891                             q->expire))
2892                                 continue;       /* too early */
2893                         if (TIME_LEQ(q->expire, time_second))
2894                                 continue;       /* too late, rule expired */
2895
2896                         send_pkt(&q->id, q->ack_rev - 1, q->ack_fwd, TH_SYN);
2897                         send_pkt(&q->id, q->ack_fwd - 1, q->ack_rev, 0);
2898                 }
2899         }
2900         crit_exit();
2901 done:
2902         callout_reset(&ipfw_timeout_h, dyn_keepalive_period * hz,
2903                       ipfw_tick, NULL);
2904 }
2905
2906 static void
2907 ipfw_init_default_rule(struct ip_fw **head)
2908 {
2909         struct ip_fw *def_rule;
2910
2911         KKASSERT(*head == NULL);
2912
2913         def_rule = kmalloc(sizeof(*def_rule), M_IPFW, M_WAITOK | M_ZERO);
2914
2915         def_rule->act_ofs = 0;
2916         def_rule->rulenum = IPFW_DEFAULT_RULE;
2917         def_rule->cmd_len = 1;
2918         def_rule->set = 31;
2919
2920         def_rule->cmd[0].len = 1;
2921 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
2922         def_rule->cmd[0].opcode = O_ACCEPT;
2923 #else
2924         def_rule->cmd[0].opcode = O_DENY;
2925 #endif
2926
2927         def_rule->refcnt = 1;
2928
2929         *head = def_rule;
2930         ipfw_inc_static_count(def_rule);
2931
2932         /* Install the default rule */
2933         ip_fw_default_rule = def_rule;
2934 }
2935
2936 static void
2937 ipfw_init_dispatch(struct netmsg *nmsg)
2938 {
2939         int error = 0;
2940
2941         crit_enter();
2942
2943         if (IPFW_LOADED) {
2944                 kprintf("IP firewall already loaded\n");
2945                 error = EEXIST;
2946                 goto reply;
2947         }
2948
2949         ip_fw_chk_ptr = ipfw_chk;
2950         ip_fw_ctl_ptr = ipfw_ctl;
2951         ip_fw_dn_io_ptr = ipfw_dummynet_io;
2952
2953         layer3_chain = NULL;
2954         ipfw_init_default_rule(&layer3_chain);
2955
2956         kprintf("ipfw2 initialized, divert %s, "
2957                 "rule-based forwarding enabled, default to %s, logging ",
2958 #ifdef IPDIVERT
2959                 "enabled",
2960 #else
2961                 "disabled",
2962 #endif
2963                 ip_fw_default_rule->cmd[0].opcode == O_ACCEPT ?
2964                 "accept" : "deny");
2965
2966 #ifdef IPFIREWALL_VERBOSE
2967         fw_verbose = 1;
2968 #endif
2969 #ifdef IPFIREWALL_VERBOSE_LIMIT
2970         verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
2971 #endif
2972         if (fw_verbose == 0) {
2973                 kprintf("disabled\n");
2974         } else if (verbose_limit == 0) {
2975                 kprintf("unlimited\n");
2976         } else {
2977                 kprintf("limited to %d packets/entry by default\n",
2978                         verbose_limit);
2979         }
2980         callout_init(&ipfw_timeout_h);
2981
2982         ip_fw_loaded = 1;
2983         callout_reset(&ipfw_timeout_h, hz, ipfw_tick, NULL);
2984 reply:
2985         crit_exit();
2986         lwkt_replymsg(&nmsg->nm_lmsg, error);
2987 }
2988
2989 static int
2990 ipfw_init(void)
2991 {
2992         struct netmsg smsg;
2993
2994         netmsg_init(&smsg, &curthread->td_msgport, 0, ipfw_init_dispatch);
2995         return lwkt_domsg(cpu_portfn(0), &smsg.nm_lmsg, 0);
2996 }
2997
2998 #ifdef KLD_MODULE
2999
3000 static void
3001 ipfw_fini_dispatch(struct netmsg *nmsg)
3002 {
3003         int error = 0;
3004
3005         crit_enter();
3006
3007         if (ipfw_refcnt != 0) {
3008                 error = EBUSY;
3009                 goto reply;
3010         }
3011
3012         callout_stop(&ipfw_timeout_h);
3013
3014         ip_fw_loaded = 0;
3015         netmsg_service_sync();
3016
3017         ip_fw_chk_ptr = NULL;
3018         ip_fw_ctl_ptr = NULL;
3019         ip_fw_dn_io_ptr = NULL;
3020         free_chain(&layer3_chain, 1 /* kill default rule */);
3021
3022         kprintf("IP firewall unloaded\n");
3023 reply:
3024         crit_exit();
3025         lwkt_replymsg(&nmsg->nm_lmsg, error);
3026 }
3027
3028 static int
3029 ipfw_fini(void)
3030 {
3031         struct netmsg smsg;
3032
3033         netmsg_init(&smsg, &curthread->td_msgport, 0, ipfw_fini_dispatch);
3034         return lwkt_domsg(cpu_portfn(0), &smsg.nm_lmsg, 0);
3035 }
3036
3037 #endif  /* KLD_MODULE */
3038
3039 static int
3040 ipfw_modevent(module_t mod, int type, void *unused)
3041 {
3042         int err = 0;
3043
3044         switch (type) {
3045         case MOD_LOAD:
3046                 err = ipfw_init();
3047                 break;
3048
3049         case MOD_UNLOAD:
3050 #ifndef KLD_MODULE
3051                 kprintf("ipfw statically compiled, cannot unload\n");
3052                 err = EBUSY;
3053 #else
3054                 err = ipfw_fini();
3055 #endif
3056                 break;
3057         default:
3058                 break;
3059         }
3060         return err;
3061 }
3062
3063 static moduledata_t ipfwmod = {
3064         "ipfw",
3065         ipfw_modevent,
3066         0
3067 };
3068 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PROTO_END, SI_ORDER_ANY);
3069 MODULE_VERSION(ipfw, 1);