FreeBSD/Linux Kernel Cross Reference
sys/netinet/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 * $FreeBSD$
26 */
27
28 #define DEB(x)
29 #define DDB(x) x
30
31 /*
32 * Implement IP packet firewall (new version)
33 */
34
35 #if !defined(KLD_MODULE)
36 #include "opt_ipfw.h"
37 #include "opt_ipdn.h"
38 #include "opt_ipdivert.h"
39 #include "opt_inet.h"
40 #include "opt_ipsec.h"
41 #ifndef INET
42 #error IPFIREWALL requires INET.
43 #endif /* INET */
44 #endif
45
46 #if IPFW2
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #include <sys/kernel.h>
52 #include <sys/proc.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/sysctl.h>
56 #include <sys/syslog.h>
57 #include <sys/ucred.h>
58 #include <net/if.h>
59 #include <net/radix.h>
60 #include <net/route.h>
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/in_var.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/ip.h>
66 #include <netinet/ip_var.h>
67 #include <netinet/ip_icmp.h>
68 #include <netinet/ip_fw.h>
69 #include <netinet/ip_dummynet.h>
70 #include <netinet/tcp.h>
71 #include <netinet/tcp_timer.h>
72 #include <netinet/tcp_var.h>
73 #include <netinet/tcpip.h>
74 #include <netinet/udp.h>
75 #include <netinet/udp_var.h>
76
77 #ifdef IPSEC
78 #include <netinet6/ipsec.h>
79 #endif
80
81 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
82
83 #include <machine/in_cksum.h> /* XXX for in_cksum */
84
85 /*
86 * XXX This one should go in sys/mbuf.h. It is used to avoid that
87 * a firewall-generated packet loops forever through the firewall.
88 */
89 #ifndef M_SKIP_FIREWALL
90 #define M_SKIP_FIREWALL 0x4000
91 #endif
92
93 /*
94 * set_disable contains one bit per set value (0..31).
95 * If the bit is set, all rules with the corresponding set
96 * are disabled. Set RESVD_SET(31) is reserved for the default rule
97 * and rules that are not deleted by the flush command,
98 * and CANNOT be disabled.
99 * Rules in set RESVD_SET can only be deleted explicitly.
100 */
101 static u_int32_t set_disable;
102
103 static int fw_verbose;
104 static int verbose_limit;
105
106 static struct callout_handle ipfw_timeout_h;
107 #define IPFW_DEFAULT_RULE 65535
108
109 /*
110 * list of rules for layer 3
111 */
112 static struct ip_fw *layer3_chain;
113
114 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
115 MALLOC_DEFINE(M_IPFW_TBL, "ipfw_tbl", "IpFw tables");
116
117 struct table_entry {
118 struct radix_node rn[2];
119 struct sockaddr_in addr, mask;
120 u_int32_t value;
121 };
122
123 #define IPFW_TABLES_MAX 128
124 static struct {
125 struct radix_node_head *rnh;
126 int modified;
127 } ipfw_tables[IPFW_TABLES_MAX];
128
129 static int fw_debug = 1;
130 static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
131
132 #ifdef SYSCTL_NODE
133 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
134 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable,
135 CTLFLAG_RW|CTLFLAG_SECURE,
136 &fw_enable, 0, "Enable ipfw");
137 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
138 &autoinc_step, 0, "Rule number autincrement step");
139 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, one_pass,
140 CTLFLAG_RW|CTLFLAG_SECURE,
141 &fw_one_pass, 0,
142 "Only do a single pass through ipfw when using dummynet(4)");
143 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug,
144 CTLFLAG_RW,
145 &fw_debug, 0, "Enable printing of debug ip_fw statements");
146 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose,
147 CTLFLAG_RW|CTLFLAG_SECURE,
148 &fw_verbose, 0, "Log matches to ipfw rules");
149 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
150 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
151
152 /*
153 * Description of dynamic rules.
154 *
155 * Dynamic rules are stored in lists accessed through a hash table
156 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
157 * be modified through the sysctl variable dyn_buckets which is
158 * updated when the table becomes empty.
159 *
160 * XXX currently there is only one list, ipfw_dyn.
161 *
162 * When a packet is received, its address fields are first masked
163 * with the mask defined for the rule, then hashed, then matched
164 * against the entries in the corresponding list.
165 * Dynamic rules can be used for different purposes:
166 * + stateful rules;
167 * + enforcing limits on the number of sessions;
168 * + in-kernel NAT (not implemented yet)
169 *
170 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
171 * measured in seconds and depending on the flags.
172 *
173 * The total number of dynamic rules is stored in dyn_count.
174 * The max number of dynamic rules is dyn_max. When we reach
175 * the maximum number of rules we do not create anymore. This is
176 * done to avoid consuming too much memory, but also too much
177 * time when searching on each packet (ideally, we should try instead
178 * to put a limit on the length of the list on each bucket...).
179 *
180 * Each dynamic rule holds a pointer to the parent ipfw rule so
181 * we know what action to perform. Dynamic rules are removed when
182 * the parent rule is deleted. XXX we should make them survive.
183 *
184 * There are some limitations with dynamic rules -- we do not
185 * obey the 'randomized match', and we do not do multiple
186 * passes through the firewall. XXX check the latter!!!
187 */
188 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
189 static u_int32_t dyn_buckets = 256; /* must be power of 2 */
190 static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */
191
192 /*
193 * Timeouts for various events in handing dynamic rules.
194 */
195 static u_int32_t dyn_ack_lifetime = 300;
196 static u_int32_t dyn_syn_lifetime = 20;
197 static u_int32_t dyn_fin_lifetime = 1;
198 static u_int32_t dyn_rst_lifetime = 1;
199 static u_int32_t dyn_udp_lifetime = 10;
200 static u_int32_t dyn_short_lifetime = 5;
201
202 /*
203 * Keepalives are sent if dyn_keepalive is set. They are sent every
204 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
205 * seconds of lifetime of a rule.
206 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
207 * than dyn_keepalive_period.
208 */
209
210 static u_int32_t dyn_keepalive_interval = 20;
211 static u_int32_t dyn_keepalive_period = 5;
212 static u_int32_t dyn_keepalive = 1; /* do send keepalives */
213
214 static u_int32_t static_count; /* # of static rules */
215 static u_int32_t static_len; /* size in bytes of static rules */
216 static u_int32_t dyn_count; /* # of dynamic rules */
217 static u_int32_t dyn_max = 4096; /* max # of dynamic rules */
218
219 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
220 &dyn_buckets, 0, "Number of dyn. buckets");
221 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
222 &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
223 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
224 &dyn_count, 0, "Number of dyn. rules");
225 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
226 &dyn_max, 0, "Max number of dyn. rules");
227 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
228 &static_count, 0, "Number of static rules");
229 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
230 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
231 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
232 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
233 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
234 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
235 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
236 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
237 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
238 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
239 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
240 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
241 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
242 &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
243
244 #endif /* SYSCTL_NODE */
245
246
247 static ip_fw_chk_t ipfw_chk;
248
249 ip_dn_ruledel_t *ip_dn_ruledel_ptr = NULL; /* hook into dummynet */
250
251 /*
252 * This macro maps an ip pointer into a layer3 header pointer of type T
253 */
254 #define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
255
256 static __inline int
257 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
258 {
259 int type = L3HDR(struct icmp,ip)->icmp_type;
260
261 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
262 }
263
264 #define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
265 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
266
267 static int
268 is_icmp_query(struct ip *ip)
269 {
270 int type = L3HDR(struct icmp, ip)->icmp_type;
271 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
272 }
273 #undef TT
274
275 /*
276 * The following checks use two arrays of 8 or 16 bits to store the
277 * bits that we want set or clear, respectively. They are in the
278 * low and high half of cmd->arg1 or cmd->d[0].
279 *
280 * We scan options and store the bits we find set. We succeed if
281 *
282 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
283 *
284 * The code is sometimes optimized not to store additional variables.
285 */
286
287 static int
288 flags_match(ipfw_insn *cmd, u_int8_t bits)
289 {
290 u_char want_clear;
291 bits = ~bits;
292
293 if ( ((cmd->arg1 & 0xff) & bits) != 0)
294 return 0; /* some bits we want set were clear */
295 want_clear = (cmd->arg1 >> 8) & 0xff;
296 if ( (want_clear & bits) != want_clear)
297 return 0; /* some bits we want clear were set */
298 return 1;
299 }
300
301 static int
302 ipopts_match(struct ip *ip, ipfw_insn *cmd)
303 {
304 int optlen, bits = 0;
305 u_char *cp = (u_char *)(ip + 1);
306 int x = (ip->ip_hl << 2) - sizeof (struct ip);
307
308 for (; x > 0; x -= optlen, cp += optlen) {
309 int opt = cp[IPOPT_OPTVAL];
310
311 if (opt == IPOPT_EOL)
312 break;
313 if (opt == IPOPT_NOP)
314 optlen = 1;
315 else {
316 optlen = cp[IPOPT_OLEN];
317 if (optlen <= 0 || optlen > x)
318 return 0; /* invalid or truncated */
319 }
320 switch (opt) {
321
322 default:
323 break;
324
325 case IPOPT_LSRR:
326 bits |= IP_FW_IPOPT_LSRR;
327 break;
328
329 case IPOPT_SSRR:
330 bits |= IP_FW_IPOPT_SSRR;
331 break;
332
333 case IPOPT_RR:
334 bits |= IP_FW_IPOPT_RR;
335 break;
336
337 case IPOPT_TS:
338 bits |= IP_FW_IPOPT_TS;
339 break;
340 }
341 }
342 return (flags_match(cmd, bits));
343 }
344
345 static int
346 tcpopts_match(struct ip *ip, ipfw_insn *cmd)
347 {
348 int optlen, bits = 0;
349 struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
350 u_char *cp = (u_char *)(tcp + 1);
351 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
352
353 for (; x > 0; x -= optlen, cp += optlen) {
354 int opt = cp[0];
355 if (opt == TCPOPT_EOL)
356 break;
357 if (opt == TCPOPT_NOP)
358 optlen = 1;
359 else {
360 optlen = cp[1];
361 if (optlen <= 0)
362 break;
363 }
364
365 switch (opt) {
366
367 default:
368 break;
369
370 case TCPOPT_MAXSEG:
371 bits |= IP_FW_TCPOPT_MSS;
372 break;
373
374 case TCPOPT_WINDOW:
375 bits |= IP_FW_TCPOPT_WINDOW;
376 break;
377
378 case TCPOPT_SACK_PERMITTED:
379 case TCPOPT_SACK:
380 bits |= IP_FW_TCPOPT_SACK;
381 break;
382
383 case TCPOPT_TIMESTAMP:
384 bits |= IP_FW_TCPOPT_TS;
385 break;
386
387 case TCPOPT_CC:
388 case TCPOPT_CCNEW:
389 case TCPOPT_CCECHO:
390 bits |= IP_FW_TCPOPT_CC;
391 break;
392 }
393 }
394 return (flags_match(cmd, bits));
395 }
396
397 static int
398 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
399 {
400 if (ifp == NULL) /* no iface with this packet, match fails */
401 return 0;
402 /* Check by name or by IP address */
403 if (cmd->name[0] != '\0') { /* match by name */
404 /* Check unit number (-1 is wildcard) */
405 if (cmd->p.unit != -1 && cmd->p.unit != ifp->if_unit)
406 return(0);
407 /* Check name */
408 if (!strncmp(ifp->if_name, cmd->name, IFNAMSIZ))
409 return(1);
410 } else {
411 struct ifaddr *ia;
412
413 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
414 if (ia->ifa_addr == NULL)
415 continue;
416 if (ia->ifa_addr->sa_family != AF_INET)
417 continue;
418 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
419 (ia->ifa_addr))->sin_addr.s_addr)
420 return(1); /* match */
421 }
422 }
423 return(0); /* no match, fail ... */
424 }
425
426 /*
427 * The 'verrevpath' option checks that the interface that an IP packet
428 * arrives on is the same interface that traffic destined for the
429 * packet's source address would be routed out of. This is a measure
430 * to block forged packets. This is also commonly known as "anti-spoofing"
431 * or Unicast Reverse Path Forwarding (Unicast RFP) in Cisco-ese. The
432 * name of the knob is purposely reminisent of the Cisco IOS command,
433 *
434 * ip verify unicast reverse-path
435 *
436 * which implements the same functionality. But note that syntax is
437 * misleading. The check may be performed on all IP packets whether unicast,
438 * multicast, or broadcast.
439 */
440 static int
441 verify_rev_path(struct in_addr src, struct ifnet *ifp)
442 {
443 struct route ro;
444 struct sockaddr_in *dst;
445
446 bzero(&ro, sizeof(ro));
447
448 dst = (struct sockaddr_in *)&(ro.ro_dst);
449 dst->sin_family = AF_INET;
450 dst->sin_len = sizeof(*dst);
451 dst->sin_addr = src;
452 rtalloc_ign(&ro, RTF_CLONING|RTF_PRCLONING);
453
454 if (ro.ro_rt == NULL)
455 return 0;
456 if ((ifp == NULL) || (ro.ro_rt->rt_ifp->if_index != ifp->if_index)) {
457 RTFREE(ro.ro_rt);
458 return 0;
459 }
460 RTFREE(ro.ro_rt);
461 return 1;
462 }
463
464
465 static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */
466
467 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
468 #define SNP(buf) buf, sizeof(buf)
469
470 /*
471 * We enter here when we have a rule with O_LOG.
472 * XXX this function alone takes about 2Kbytes of code!
473 */
474 static void
475 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
476 struct mbuf *m, struct ifnet *oif)
477 {
478 char *action;
479 int limit_reached = 0;
480 char action2[40], proto[48], fragment[28];
481
482 fragment[0] = '\0';
483 proto[0] = '\0';
484
485 if (f == NULL) { /* bogus pkt */
486 if (verbose_limit != 0 && norule_counter >= verbose_limit)
487 return;
488 norule_counter++;
489 if (norule_counter == verbose_limit)
490 limit_reached = verbose_limit;
491 action = "Refuse";
492 } else { /* O_LOG is the first action, find the real one */
493 ipfw_insn *cmd = ACTION_PTR(f);
494 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
495
496 if (l->max_log != 0 && l->log_left == 0)
497 return;
498 l->log_left--;
499 if (l->log_left == 0)
500 limit_reached = l->max_log;
501 cmd += F_LEN(cmd); /* point to first action */
502 if (cmd->opcode == O_PROB)
503 cmd += F_LEN(cmd);
504
505 action = action2;
506 switch (cmd->opcode) {
507 case O_DENY:
508 action = "Deny";
509 break;
510
511 case O_REJECT:
512 if (cmd->arg1==ICMP_REJECT_RST)
513 action = "Reset";
514 else if (cmd->arg1==ICMP_UNREACH_HOST)
515 action = "Reject";
516 else
517 snprintf(SNPARGS(action2, 0), "Unreach %d",
518 cmd->arg1);
519 break;
520
521 case O_ACCEPT:
522 action = "Accept";
523 break;
524 case O_COUNT:
525 action = "Count";
526 break;
527 case O_DIVERT:
528 snprintf(SNPARGS(action2, 0), "Divert %d",
529 cmd->arg1);
530 break;
531 case O_TEE:
532 snprintf(SNPARGS(action2, 0), "Tee %d",
533 cmd->arg1);
534 break;
535 case O_SKIPTO:
536 snprintf(SNPARGS(action2, 0), "SkipTo %d",
537 cmd->arg1);
538 break;
539 case O_PIPE:
540 snprintf(SNPARGS(action2, 0), "Pipe %d",
541 cmd->arg1);
542 break;
543 case O_QUEUE:
544 snprintf(SNPARGS(action2, 0), "Queue %d",
545 cmd->arg1);
546 break;
547 case O_FORWARD_IP: {
548 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
549 int len;
550
551 len = snprintf(SNPARGS(action2, 0), "Forward to %s",
552 inet_ntoa(sa->sa.sin_addr));
553 if (sa->sa.sin_port)
554 snprintf(SNPARGS(action2, len), ":%d",
555 sa->sa.sin_port);
556 }
557 break;
558 default:
559 action = "UNKNOWN";
560 break;
561 }
562 }
563
564 if (hlen == 0) { /* non-ip */
565 snprintf(SNPARGS(proto, 0), "MAC");
566 } else {
567 struct ip *ip = mtod(m, struct ip *);
568 /* these three are all aliases to the same thing */
569 struct icmp *const icmp = L3HDR(struct icmp, ip);
570 struct tcphdr *const tcp = (struct tcphdr *)icmp;
571 struct udphdr *const udp = (struct udphdr *)icmp;
572
573 int ip_off, offset, ip_len;
574
575 int len;
576
577 if (eh != NULL) { /* layer 2 packets are as on the wire */
578 ip_off = ntohs(ip->ip_off);
579 ip_len = ntohs(ip->ip_len);
580 } else {
581 ip_off = ip->ip_off;
582 ip_len = ip->ip_len;
583 }
584 offset = ip_off & IP_OFFMASK;
585 switch (ip->ip_p) {
586 case IPPROTO_TCP:
587 len = snprintf(SNPARGS(proto, 0), "TCP %s",
588 inet_ntoa(ip->ip_src));
589 if (offset == 0)
590 snprintf(SNPARGS(proto, len), ":%d %s:%d",
591 ntohs(tcp->th_sport),
592 inet_ntoa(ip->ip_dst),
593 ntohs(tcp->th_dport));
594 else
595 snprintf(SNPARGS(proto, len), " %s",
596 inet_ntoa(ip->ip_dst));
597 break;
598
599 case IPPROTO_UDP:
600 len = snprintf(SNPARGS(proto, 0), "UDP %s",
601 inet_ntoa(ip->ip_src));
602 if (offset == 0)
603 snprintf(SNPARGS(proto, len), ":%d %s:%d",
604 ntohs(udp->uh_sport),
605 inet_ntoa(ip->ip_dst),
606 ntohs(udp->uh_dport));
607 else
608 snprintf(SNPARGS(proto, len), " %s",
609 inet_ntoa(ip->ip_dst));
610 break;
611
612 case IPPROTO_ICMP:
613 if (offset == 0)
614 len = snprintf(SNPARGS(proto, 0),
615 "ICMP:%u.%u ",
616 icmp->icmp_type, icmp->icmp_code);
617 else
618 len = snprintf(SNPARGS(proto, 0), "ICMP ");
619 len += snprintf(SNPARGS(proto, len), "%s",
620 inet_ntoa(ip->ip_src));
621 snprintf(SNPARGS(proto, len), " %s",
622 inet_ntoa(ip->ip_dst));
623 break;
624
625 default:
626 len = snprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
627 inet_ntoa(ip->ip_src));
628 snprintf(SNPARGS(proto, len), " %s",
629 inet_ntoa(ip->ip_dst));
630 break;
631 }
632
633 if (ip_off & (IP_MF | IP_OFFMASK))
634 snprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
635 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
636 offset << 3,
637 (ip_off & IP_MF) ? "+" : "");
638 }
639 if (oif || m->m_pkthdr.rcvif)
640 log(LOG_SECURITY | LOG_INFO,
641 "ipfw: %d %s %s %s via %s%d%s\n",
642 f ? f->rulenum : -1,
643 action, proto, oif ? "out" : "in",
644 oif ? oif->if_name : m->m_pkthdr.rcvif->if_name,
645 oif ? oif->if_unit : m->m_pkthdr.rcvif->if_unit,
646 fragment);
647 else
648 log(LOG_SECURITY | LOG_INFO,
649 "ipfw: %d %s %s [no if info]%s\n",
650 f ? f->rulenum : -1,
651 action, proto, fragment);
652 if (limit_reached)
653 log(LOG_SECURITY | LOG_NOTICE,
654 "ipfw: limit %d reached on entry %d\n",
655 limit_reached, f ? f->rulenum : -1);
656 }
657
658 /*
659 * IMPORTANT: the hash function for dynamic rules must be commutative
660 * in source and destination (ip,port), because rules are bidirectional
661 * and we want to find both in the same bucket.
662 */
663 static __inline int
664 hash_packet(struct ipfw_flow_id *id)
665 {
666 u_int32_t i;
667
668 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
669 i &= (curr_dyn_buckets - 1);
670 return i;
671 }
672
673 /**
674 * unlink a dynamic rule from a chain. prev is a pointer to
675 * the previous one, q is a pointer to the rule to delete,
676 * head is a pointer to the head of the queue.
677 * Modifies q and potentially also head.
678 */
679 #define UNLINK_DYN_RULE(prev, head, q) { \
680 ipfw_dyn_rule *old_q = q; \
681 \
682 /* remove a refcount to the parent */ \
683 if (q->dyn_type == O_LIMIT) \
684 q->parent->count--; \
685 DEB(printf("ipfw: unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",\
686 (q->id.src_ip), (q->id.src_port), \
687 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \
688 if (prev != NULL) \
689 prev->next = q = q->next; \
690 else \
691 head = q = q->next; \
692 dyn_count--; \
693 free(old_q, M_IPFW); }
694
695 #define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
696
697 /**
698 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
699 *
700 * If keep_me == NULL, rules are deleted even if not expired,
701 * otherwise only expired rules are removed.
702 *
703 * The value of the second parameter is also used to point to identify
704 * a rule we absolutely do not want to remove (e.g. because we are
705 * holding a reference to it -- this is the case with O_LIMIT_PARENT
706 * rules). The pointer is only used for comparison, so any non-null
707 * value will do.
708 */
709 static void
710 remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
711 {
712 static u_int32_t last_remove = 0;
713
714 #define FORCE (keep_me == NULL)
715
716 ipfw_dyn_rule *prev, *q;
717 int i, pass = 0, max_pass = 0;
718
719 if (ipfw_dyn_v == NULL || dyn_count == 0)
720 return;
721 /* do not expire more than once per second, it is useless */
722 if (!FORCE && last_remove == time_second)
723 return;
724 last_remove = time_second;
725
726 /*
727 * because O_LIMIT refer to parent rules, during the first pass only
728 * remove child and mark any pending LIMIT_PARENT, and remove
729 * them in a second pass.
730 */
731 next_pass:
732 for (i = 0 ; i < curr_dyn_buckets ; i++) {
733 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
734 /*
735 * Logic can become complex here, so we split tests.
736 */
737 if (q == keep_me)
738 goto next;
739 if (rule != NULL && rule != q->rule)
740 goto next; /* not the one we are looking for */
741 if (q->dyn_type == O_LIMIT_PARENT) {
742 /*
743 * handle parent in the second pass,
744 * record we need one.
745 */
746 max_pass = 1;
747 if (pass == 0)
748 goto next;
749 if (FORCE && q->count != 0 ) {
750 /* XXX should not happen! */
751 printf("ipfw: OUCH! cannot remove rule,"
752 " count %d\n", q->count);
753 }
754 } else {
755 if (!FORCE &&
756 !TIME_LEQ( q->expire, time_second ))
757 goto next;
758 }
759 if (q->dyn_type != O_LIMIT_PARENT || !q->count) {
760 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
761 continue;
762 }
763 next:
764 prev=q;
765 q=q->next;
766 }
767 }
768 if (pass++ < max_pass)
769 goto next_pass;
770 }
771
772
773 /**
774 * lookup a dynamic rule.
775 */
776 static ipfw_dyn_rule *
777 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
778 struct tcphdr *tcp)
779 {
780 /*
781 * stateful ipfw extensions.
782 * Lookup into dynamic session queue
783 */
784 #define MATCH_REVERSE 0
785 #define MATCH_FORWARD 1
786 #define MATCH_NONE 2
787 #define MATCH_UNKNOWN 3
788 int i, dir = MATCH_NONE;
789 ipfw_dyn_rule *prev, *q=NULL;
790
791 if (ipfw_dyn_v == NULL)
792 goto done; /* not found */
793 i = hash_packet( pkt );
794 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
795 if (q->dyn_type == O_LIMIT_PARENT && q->count)
796 goto next;
797 if (TIME_LEQ( q->expire, time_second)) { /* expire entry */
798 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
799 continue;
800 }
801 if (pkt->proto == q->id.proto &&
802 q->dyn_type != O_LIMIT_PARENT) {
803 if (pkt->src_ip == q->id.src_ip &&
804 pkt->dst_ip == q->id.dst_ip &&
805 pkt->src_port == q->id.src_port &&
806 pkt->dst_port == q->id.dst_port ) {
807 dir = MATCH_FORWARD;
808 break;
809 }
810 if (pkt->src_ip == q->id.dst_ip &&
811 pkt->dst_ip == q->id.src_ip &&
812 pkt->src_port == q->id.dst_port &&
813 pkt->dst_port == q->id.src_port ) {
814 dir = MATCH_REVERSE;
815 break;
816 }
817 }
818 next:
819 prev = q;
820 q = q->next;
821 }
822 if (q == NULL)
823 goto done; /* q = NULL, not found */
824
825 if ( prev != NULL) { /* found and not in front */
826 prev->next = q->next;
827 q->next = ipfw_dyn_v[i];
828 ipfw_dyn_v[i] = q;
829 }
830 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
831 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
832
833 #define BOTH_SYN (TH_SYN | (TH_SYN << 8))
834 #define BOTH_FIN (TH_FIN | (TH_FIN << 8))
835 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
836 switch (q->state) {
837 case TH_SYN: /* opening */
838 q->expire = time_second + dyn_syn_lifetime;
839 break;
840
841 case BOTH_SYN: /* move to established */
842 case BOTH_SYN | TH_FIN : /* one side tries to close */
843 case BOTH_SYN | (TH_FIN << 8) :
844 if (tcp) {
845 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
846 u_int32_t ack = ntohl(tcp->th_ack);
847 if (dir == MATCH_FORWARD) {
848 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
849 q->ack_fwd = ack;
850 else { /* ignore out-of-sequence */
851 break;
852 }
853 } else {
854 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
855 q->ack_rev = ack;
856 else { /* ignore out-of-sequence */
857 break;
858 }
859 }
860 }
861 q->expire = time_second + dyn_ack_lifetime;
862 break;
863
864 case BOTH_SYN | BOTH_FIN: /* both sides closed */
865 if (dyn_fin_lifetime >= dyn_keepalive_period)
866 dyn_fin_lifetime = dyn_keepalive_period - 1;
867 q->expire = time_second + dyn_fin_lifetime;
868 break;
869
870 default:
871 #if 0
872 /*
873 * reset or some invalid combination, but can also
874 * occur if we use keep-state the wrong way.
875 */
876 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
877 printf("invalid state: 0x%x\n", q->state);
878 #endif
879 if (dyn_rst_lifetime >= dyn_keepalive_period)
880 dyn_rst_lifetime = dyn_keepalive_period - 1;
881 q->expire = time_second + dyn_rst_lifetime;
882 break;
883 }
884 } else if (pkt->proto == IPPROTO_UDP) {
885 q->expire = time_second + dyn_udp_lifetime;
886 } else {
887 /* other protocols */
888 q->expire = time_second + dyn_short_lifetime;
889 }
890 done:
891 if (match_direction)
892 *match_direction = dir;
893 return q;
894 }
895
896 static void
897 realloc_dynamic_table(void)
898 {
899 /*
900 * Try reallocation, make sure we have a power of 2 and do
901 * not allow more than 64k entries. In case of overflow,
902 * default to 1024.
903 */
904
905 if (dyn_buckets > 65536)
906 dyn_buckets = 1024;
907 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
908 dyn_buckets = curr_dyn_buckets; /* reset */
909 return;
910 }
911 curr_dyn_buckets = dyn_buckets;
912 if (ipfw_dyn_v != NULL)
913 free(ipfw_dyn_v, M_IPFW);
914 for (;;) {
915 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
916 M_IPFW, M_NOWAIT | M_ZERO);
917 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
918 break;
919 curr_dyn_buckets /= 2;
920 }
921 }
922
923 /**
924 * Install state of type 'type' for a dynamic session.
925 * The hash table contains two type of rules:
926 * - regular rules (O_KEEP_STATE)
927 * - rules for sessions with limited number of sess per user
928 * (O_LIMIT). When they are created, the parent is
929 * increased by 1, and decreased on delete. In this case,
930 * the third parameter is the parent rule and not the chain.
931 * - "parent" rules for the above (O_LIMIT_PARENT).
932 */
933 static ipfw_dyn_rule *
934 add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
935 {
936 ipfw_dyn_rule *r;
937 int i;
938
939 if (ipfw_dyn_v == NULL ||
940 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
941 realloc_dynamic_table();
942 if (ipfw_dyn_v == NULL)
943 return NULL; /* failed ! */
944 }
945 i = hash_packet(id);
946
947 r = malloc(sizeof *r, M_IPFW, M_NOWAIT | M_ZERO);
948 if (r == NULL) {
949 printf ("ipfw: sorry cannot allocate state\n");
950 return NULL;
951 }
952
953 /* increase refcount on parent, and set pointer */
954 if (dyn_type == O_LIMIT) {
955 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
956 if ( parent->dyn_type != O_LIMIT_PARENT)
957 panic("invalid parent");
958 parent->count++;
959 r->parent = parent;
960 rule = parent->rule;
961 }
962
963 r->id = *id;
964 r->expire = time_second + dyn_syn_lifetime;
965 r->rule = rule;
966 r->dyn_type = dyn_type;
967 r->pcnt = r->bcnt = 0;
968 r->count = 0;
969
970 r->bucket = i;
971 r->next = ipfw_dyn_v[i];
972 ipfw_dyn_v[i] = r;
973 dyn_count++;
974 DEB(printf("ipfw: add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
975 dyn_type,
976 (r->id.src_ip), (r->id.src_port),
977 (r->id.dst_ip), (r->id.dst_port),
978 dyn_count ); )
979 return r;
980 }
981
982 /**
983 * lookup dynamic parent rule using pkt and rule as search keys.
984 * If the lookup fails, then install one.
985 */
986 static ipfw_dyn_rule *
987 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
988 {
989 ipfw_dyn_rule *q;
990 int i;
991
992 if (ipfw_dyn_v) {
993 i = hash_packet( pkt );
994 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
995 if (q->dyn_type == O_LIMIT_PARENT &&
996 rule== q->rule &&
997 pkt->proto == q->id.proto &&
998 pkt->src_ip == q->id.src_ip &&
999 pkt->dst_ip == q->id.dst_ip &&
1000 pkt->src_port == q->id.src_port &&
1001 pkt->dst_port == q->id.dst_port) {
1002 q->expire = time_second + dyn_short_lifetime;
1003 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);)
1004 return q;
1005 }
1006 }
1007 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1008 }
1009
1010 /**
1011 * Install dynamic state for rule type cmd->o.opcode
1012 *
1013 * Returns 1 (failure) if state is not installed because of errors or because
1014 * session limitations are enforced.
1015 */
1016 static int
1017 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1018 struct ip_fw_args *args)
1019 {
1020 static int last_log;
1021
1022 ipfw_dyn_rule *q;
1023
1024 DEB(printf("ipfw: install state type %d 0x%08x %u -> 0x%08x %u\n",
1025 cmd->o.opcode,
1026 (args->f_id.src_ip), (args->f_id.src_port),
1027 (args->f_id.dst_ip), (args->f_id.dst_port) );)
1028
1029 q = lookup_dyn_rule(&args->f_id, NULL, NULL);
1030
1031 if (q != NULL) { /* should never occur */
1032 if (last_log != time_second) {
1033 last_log = time_second;
1034 printf("ipfw: install_state: entry already present, done\n");
1035 }
1036 return 0;
1037 }
1038
1039 if (dyn_count >= dyn_max)
1040 /*
1041 * Run out of slots, try to remove any expired rule.
1042 */
1043 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1044
1045 if (dyn_count >= dyn_max) {
1046 if (last_log != time_second) {
1047 last_log = time_second;
1048 printf("ipfw: install_state: Too many dynamic rules\n");
1049 }
1050 return 1; /* cannot install, notify caller */
1051 }
1052
1053 switch (cmd->o.opcode) {
1054 case O_KEEP_STATE: /* bidir rule */
1055 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1056 break;
1057
1058 case O_LIMIT: /* limit number of sessions */
1059 {
1060 u_int16_t limit_mask = cmd->limit_mask;
1061 struct ipfw_flow_id id;
1062 ipfw_dyn_rule *parent;
1063
1064 DEB(printf("ipfw: installing dyn-limit rule %d\n",
1065 cmd->conn_limit);)
1066
1067 id.dst_ip = id.src_ip = 0;
1068 id.dst_port = id.src_port = 0;
1069 id.proto = args->f_id.proto;
1070
1071 if (limit_mask & DYN_SRC_ADDR)
1072 id.src_ip = args->f_id.src_ip;
1073 if (limit_mask & DYN_DST_ADDR)
1074 id.dst_ip = args->f_id.dst_ip;
1075 if (limit_mask & DYN_SRC_PORT)
1076 id.src_port = args->f_id.src_port;
1077 if (limit_mask & DYN_DST_PORT)
1078 id.dst_port = args->f_id.dst_port;
1079 parent = lookup_dyn_parent(&id, rule);
1080 if (parent == NULL) {
1081 printf("ipfw: add parent failed\n");
1082 return 1;
1083 }
1084 if (parent->count >= cmd->conn_limit) {
1085 /*
1086 * See if we can remove some expired rule.
1087 */
1088 remove_dyn_rule(rule, parent);
1089 if (parent->count >= cmd->conn_limit) {
1090 if (fw_verbose && last_log != time_second) {
1091 last_log = time_second;
1092 log(LOG_SECURITY | LOG_DEBUG,
1093 "drop session, too many entries\n");
1094 }
1095 return 1;
1096 }
1097 }
1098 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1099 }
1100 break;
1101 default:
1102 printf("ipfw: unknown dynamic rule type %u\n", cmd->o.opcode);
1103 return 1;
1104 }
1105 lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1106 return 0;
1107 }
1108
1109 /*
1110 * Transmit a TCP packet, containing either a RST or a keepalive.
1111 * When flags & TH_RST, we are sending a RST packet, because of a
1112 * "reset" action matched the packet.
1113 * Otherwise we are sending a keepalive, and flags & TH_
1114 */
1115 static void
1116 send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags)
1117 {
1118 struct mbuf *m;
1119 struct ip *ip;
1120 struct tcphdr *tcp;
1121 struct route sro; /* fake route */
1122
1123 MGETHDR(m, M_DONTWAIT, MT_HEADER);
1124 if (m == 0)
1125 return;
1126 m->m_pkthdr.rcvif = (struct ifnet *)0;
1127 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1128 m->m_data += max_linkhdr;
1129
1130 ip = mtod(m, struct ip *);
1131 bzero(ip, m->m_len);
1132 tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1133 ip->ip_p = IPPROTO_TCP;
1134 tcp->th_off = 5;
1135 /*
1136 * Assume we are sending a RST (or a keepalive in the reverse
1137 * direction), swap src and destination addresses and ports.
1138 */
1139 ip->ip_src.s_addr = htonl(id->dst_ip);
1140 ip->ip_dst.s_addr = htonl(id->src_ip);
1141 tcp->th_sport = htons(id->dst_port);
1142 tcp->th_dport = htons(id->src_port);
1143 if (flags & TH_RST) { /* we are sending a RST */
1144 if (flags & TH_ACK) {
1145 tcp->th_seq = htonl(ack);
1146 tcp->th_ack = htonl(0);
1147 tcp->th_flags = TH_RST;
1148 } else {
1149 if (flags & TH_SYN)
1150 seq++;
1151 tcp->th_seq = htonl(0);
1152 tcp->th_ack = htonl(seq);
1153 tcp->th_flags = TH_RST | TH_ACK;
1154 }
1155 } else {
1156 /*
1157 * We are sending a keepalive. flags & TH_SYN determines
1158 * the direction, forward if set, reverse if clear.
1159 * NOTE: seq and ack are always assumed to be correct
1160 * as set by the caller. This may be confusing...
1161 */
1162 if (flags & TH_SYN) {
1163 /*
1164 * we have to rewrite the correct addresses!
1165 */
1166 ip->ip_dst.s_addr = htonl(id->dst_ip);
1167 ip->ip_src.s_addr = htonl(id->src_ip);
1168 tcp->th_dport = htons(id->dst_port);
1169 tcp->th_sport = htons(id->src_port);
1170 }
1171 tcp->th_seq = htonl(seq);
1172 tcp->th_ack = htonl(ack);
1173 tcp->th_flags = TH_ACK;
1174 }
1175 /*
1176 * set ip_len to the payload size so we can compute
1177 * the tcp checksum on the pseudoheader
1178 * XXX check this, could save a couple of words ?
1179 */
1180 ip->ip_len = htons(sizeof(struct tcphdr));
1181 tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1182 /*
1183 * now fill fields left out earlier
1184 */
1185 ip->ip_ttl = ip_defttl;
1186 ip->ip_len = m->m_pkthdr.len;
1187 bzero (&sro, sizeof (sro));
1188 ip_rtaddr(ip->ip_dst, &sro);
1189 m->m_flags |= M_SKIP_FIREWALL;
1190 ip_output(m, NULL, &sro, 0, NULL, NULL);
1191 if (sro.ro_rt)
1192 RTFREE(sro.ro_rt);
1193 }
1194
1195 /*
1196 * sends a reject message, consuming the mbuf passed as an argument.
1197 */
1198 static void
1199 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1200 {
1201
1202 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1203 /* We need the IP header in host order for icmp_error(). */
1204 if (args->eh != NULL) {
1205 struct ip *ip = mtod(args->m, struct ip *);
1206 ip->ip_len = ntohs(ip->ip_len);
1207 ip->ip_off = ntohs(ip->ip_off);
1208 }
1209 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1210 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1211 struct tcphdr *const tcp =
1212 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1213 if ( (tcp->th_flags & TH_RST) == 0)
1214 send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1215 ntohl(tcp->th_ack),
1216 tcp->th_flags | TH_RST);
1217 m_freem(args->m);
1218 } else
1219 m_freem(args->m);
1220 args->m = NULL;
1221 }
1222
1223 /**
1224 *
1225 * Given an ip_fw *, lookup_next_rule will return a pointer
1226 * to the next rule, which can be either the jump
1227 * target (for skipto instructions) or the next one in the list (in
1228 * all other cases including a missing jump target).
1229 * The result is also written in the "next_rule" field of the rule.
1230 * Backward jumps are not allowed, so start looking from the next
1231 * rule...
1232 *
1233 * This never returns NULL -- in case we do not have an exact match,
1234 * the next rule is returned. When the ruleset is changed,
1235 * pointers are flushed so we are always correct.
1236 */
1237
1238 static struct ip_fw *
1239 lookup_next_rule(struct ip_fw *me)
1240 {
1241 struct ip_fw *rule = NULL;
1242 ipfw_insn *cmd;
1243
1244 /* look for action, in case it is a skipto */
1245 cmd = ACTION_PTR(me);
1246 if (cmd->opcode == O_LOG)
1247 cmd += F_LEN(cmd);
1248 if ( cmd->opcode == O_SKIPTO )
1249 for (rule = me->next; rule ; rule = rule->next)
1250 if (rule->rulenum >= cmd->arg1)
1251 break;
1252 if (rule == NULL) /* failure or not a skipto */
1253 rule = me->next;
1254 me->next_rule = rule;
1255 return rule;
1256 }
1257
1258 /*
1259 * The main check routine for the firewall.
1260 *
1261 * All arguments are in args so we can modify them and return them
1262 * back to the caller.
1263 *
1264 * Parameters:
1265 *
1266 * args->m (in/out) The packet; we set to NULL when/if we nuke it.
1267 * Starts with the IP header.
1268 * args->eh (in) Mac header if present, or NULL for layer3 packet.
1269 * args->oif Outgoing interface, or NULL if packet is incoming.
1270 * The incoming interface is in the mbuf. (in)
1271 * args->divert_rule (in/out)
1272 * Skip up to the first rule past this rule number;
1273 * upon return, non-zero port number for divert or tee.
1274 *
1275 * args->rule Pointer to the last matching rule (in/out)
1276 * args->next_hop Socket we are forwarding to (out).
1277 * args->f_id Addresses grabbed from the packet (out)
1278 *
1279 * Return value:
1280 *
1281 * IP_FW_PORT_DENY_FLAG the packet must be dropped.
1282 * 0 The packet is to be accepted and routed normally OR
1283 * the packet was denied/rejected and has been dropped;
1284 * in the latter case, *m is equal to NULL upon return.
1285 * port Divert the packet to port, with these caveats:
1286 *
1287 * - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead
1288 * of diverting it (ie, 'ipfw tee').
1289 *
1290 * - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower
1291 * 16 bits as a dummynet pipe number instead of diverting
1292 */
1293
1294 static void
1295 init_tables(void)
1296 {
1297 int i;
1298
1299 for (i = 0; i < IPFW_TABLES_MAX; i++) {
1300 rn_inithead((void **)&ipfw_tables[i].rnh, 32);
1301 ipfw_tables[i].modified = 1;
1302 }
1303 }
1304
1305 static int
1306 add_table_entry(u_int16_t tbl, in_addr_t addr, u_int8_t mlen, u_int32_t value)
1307 {
1308 struct radix_node_head *rnh;
1309 struct table_entry *ent;
1310 int s;
1311
1312 if (tbl >= IPFW_TABLES_MAX)
1313 return (EINVAL);
1314 rnh = ipfw_tables[tbl].rnh;
1315 ent = malloc(sizeof(*ent), M_IPFW_TBL, M_NOWAIT | M_ZERO);
1316 if (ent == NULL)
1317 return (ENOMEM);
1318 ent->value = value;
1319 ent->addr.sin_len = ent->mask.sin_len = 8;
1320 ent->mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1321 ent->addr.sin_addr.s_addr = addr & ent->mask.sin_addr.s_addr;
1322 s = splimp();
1323 if (rnh->rnh_addaddr(&ent->addr, &ent->mask, rnh, (void *)ent) ==
1324 NULL) {
1325 splx(s);
1326 free(ent, M_IPFW_TBL);
1327 return (EEXIST);
1328 }
1329 ipfw_tables[tbl].modified = 1;
1330 splx(s);
1331 return (0);
1332 }
1333
1334 static int
1335 del_table_entry(u_int16_t tbl, in_addr_t addr, u_int8_t mlen)
1336 {
1337 struct radix_node_head *rnh;
1338 struct table_entry *ent;
1339 struct sockaddr_in sa, mask;
1340 int s;
1341
1342 if (tbl >= IPFW_TABLES_MAX)
1343 return (EINVAL);
1344 rnh = ipfw_tables[tbl].rnh;
1345 sa.sin_len = mask.sin_len = 8;
1346 mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1347 sa.sin_addr.s_addr = addr & mask.sin_addr.s_addr;
1348 s = splimp();
1349 ent = (struct table_entry *)rnh->rnh_deladdr(&sa, &mask, rnh);
1350 if (ent == NULL) {
1351 splx(s);
1352 return (ESRCH);
1353 }
1354 ipfw_tables[tbl].modified = 1;
1355 splx(s);
1356 free(ent, M_IPFW_TBL);
1357 return (0);
1358 }
1359
1360 static int
1361 flush_table_entry(struct radix_node *rn, void *arg)
1362 {
1363 struct radix_node_head * const rnh = arg;
1364 struct table_entry *ent;
1365
1366 ent = (struct table_entry *)
1367 rnh->rnh_deladdr(rn->rn_key, rn->rn_mask, rnh);
1368 if (ent != NULL)
1369 free(ent, M_IPFW_TBL);
1370 return (0);
1371 }
1372
1373 static int
1374 flush_table(u_int16_t tbl)
1375 {
1376 struct radix_node_head *rnh;
1377 int s;
1378
1379 if (tbl >= IPFW_TABLES_MAX)
1380 return (EINVAL);
1381 rnh = ipfw_tables[tbl].rnh;
1382 s = splimp();
1383 rnh->rnh_walktree(rnh, flush_table_entry, rnh);
1384 ipfw_tables[tbl].modified = 1;
1385 splx(s);
1386 return (0);
1387 }
1388
1389 #if defined(KLD_MODULE)
1390 static void
1391 flush_tables(void)
1392 {
1393 u_int16_t tbl;
1394
1395 for (tbl = 0; tbl < IPFW_TABLES_MAX; tbl++)
1396 flush_table(tbl);
1397 }
1398 #endif
1399
1400 static int
1401 lookup_table(u_int16_t tbl, in_addr_t addr, u_int32_t *val)
1402 {
1403 struct radix_node_head *rnh;
1404 struct table_entry *ent;
1405 struct sockaddr_in sa;
1406 static in_addr_t last_addr;
1407 static int last_tbl;
1408 static int last_match;
1409 static u_int32_t last_value;
1410 int s;
1411
1412 if (tbl >= IPFW_TABLES_MAX)
1413 return (0);
1414 if (tbl == last_tbl && addr == last_addr &&
1415 !ipfw_tables[tbl].modified) {
1416 if (last_match)
1417 *val = last_value;
1418 return (last_match);
1419 }
1420 rnh = ipfw_tables[tbl].rnh;
1421 sa.sin_len = 8;
1422 sa.sin_addr.s_addr = addr;
1423 s = splimp();
1424 ipfw_tables[tbl].modified = 0;
1425 ent = (struct table_entry *)(rnh->rnh_lookup(&sa, NULL, rnh));
1426 splx(s);
1427 last_addr = addr;
1428 last_tbl = tbl;
1429 if (ent != NULL) {
1430 last_value = *val = ent->value;
1431 last_match = 1;
1432 return (1);
1433 }
1434 last_match = 0;
1435 return (0);
1436 }
1437
1438 static int
1439 count_table_entry(struct radix_node *rn, void *arg)
1440 {
1441 u_int32_t * const cnt = arg;
1442
1443 (*cnt)++;
1444 return (0);
1445 }
1446
1447 static int
1448 count_table(u_int32_t tbl, u_int32_t *cnt)
1449 {
1450 struct radix_node_head *rnh;
1451 int s;
1452
1453 if (tbl >= IPFW_TABLES_MAX)
1454 return (EINVAL);
1455 rnh = ipfw_tables[tbl].rnh;
1456 *cnt = 0;
1457 s = splimp();
1458 rnh->rnh_walktree(rnh, count_table_entry, cnt);
1459 splx(s);
1460 return (0);
1461 }
1462
1463 static int
1464 dump_table_entry(struct radix_node *rn, void *arg)
1465 {
1466 struct table_entry * const n = (struct table_entry *)rn;
1467 ipfw_table * const tbl = arg;
1468 ipfw_table_entry *ent;
1469
1470 if (tbl->cnt == tbl->size)
1471 return (1);
1472 ent = &tbl->ent[tbl->cnt];
1473 ent->tbl = tbl->tbl;
1474 if (in_nullhost(n->mask.sin_addr))
1475 ent->masklen = 0;
1476 else
1477 ent->masklen = 33 - ffs(ntohl(n->mask.sin_addr.s_addr));
1478 ent->addr = n->addr.sin_addr.s_addr;
1479 ent->value = n->value;
1480 tbl->cnt++;
1481 return (0);
1482 }
1483
1484 static int
1485 dump_table(ipfw_table *tbl)
1486 {
1487 struct radix_node_head *rnh;
1488 int s;
1489
1490 if (tbl->tbl >= IPFW_TABLES_MAX)
1491 return (EINVAL);
1492 rnh = ipfw_tables[tbl->tbl].rnh;
1493 tbl->cnt = 0;
1494 s = splimp();
1495 rnh->rnh_walktree(rnh, dump_table_entry, tbl);
1496 splx(s);
1497 return (0);
1498 }
1499
1500 static int
1501 ipfw_chk(struct ip_fw_args *args)
1502 {
1503 /*
1504 * Local variables hold state during the processing of a packet.
1505 *
1506 * IMPORTANT NOTE: to speed up the processing of rules, there
1507 * are some assumption on the values of the variables, which
1508 * are documented here. Should you change them, please check
1509 * the implementation of the various instructions to make sure
1510 * that they still work.
1511 *
1512 * args->eh The MAC header. It is non-null for a layer2
1513 * packet, it is NULL for a layer-3 packet.
1514 *
1515 * m | args->m Pointer to the mbuf, as received from the caller.
1516 * It may change if ipfw_chk() does an m_pullup, or if it
1517 * consumes the packet because it calls send_reject().
1518 * XXX This has to change, so that ipfw_chk() never modifies
1519 * or consumes the buffer.
1520 * ip is simply an alias of the value of m, and it is kept
1521 * in sync with it (the packet is supposed to start with
1522 * the ip header).
1523 */
1524 struct mbuf *m = args->m;
1525 struct ip *ip = mtod(m, struct ip *);
1526
1527 /*
1528 * oif | args->oif If NULL, ipfw_chk has been called on the
1529 * inbound path (ether_input, bdg_forward, ip_input).
1530 * If non-NULL, ipfw_chk has been called on the outbound path
1531 * (ether_output, ip_output).
1532 */
1533 struct ifnet *oif = args->oif;
1534
1535 struct ip_fw *f = NULL; /* matching rule */
1536 int retval = 0;
1537
1538 /*
1539 * hlen The length of the IPv4 header.
1540 * hlen >0 means we have an IPv4 packet.
1541 */
1542 u_int hlen = 0; /* hlen >0 means we have an IP pkt */
1543
1544 /*
1545 * offset The offset of a fragment. offset != 0 means that
1546 * we have a fragment at this offset of an IPv4 packet.
1547 * offset == 0 means that (if this is an IPv4 packet)
1548 * this is the first or only fragment.
1549 */
1550 u_short offset = 0;
1551
1552 /*
1553 * Local copies of addresses. They are only valid if we have
1554 * an IP packet.
1555 *
1556 * proto The protocol. Set to 0 for non-ip packets,
1557 * or to the protocol read from the packet otherwise.
1558 * proto != 0 means that we have an IPv4 packet.
1559 *
1560 * src_port, dst_port port numbers, in HOST format. Only
1561 * valid for TCP and UDP packets.
1562 *
1563 * src_ip, dst_ip ip addresses, in NETWORK format.
1564 * Only valid for IPv4 packets.
1565 */
1566 u_int8_t proto;
1567 u_int16_t src_port = 0, dst_port = 0; /* NOTE: host format */
1568 struct in_addr src_ip, dst_ip; /* NOTE: network format */
1569 u_int16_t ip_len=0;
1570 int pktlen;
1571 int dyn_dir = MATCH_UNKNOWN;
1572 ipfw_dyn_rule *q = NULL;
1573
1574 if (m->m_flags & M_SKIP_FIREWALL)
1575 return 0; /* accept */
1576 /*
1577 * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1578 * MATCH_NONE when checked and not matched (q = NULL),
1579 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
1580 */
1581
1582 pktlen = m->m_pkthdr.len;
1583 if (args->eh == NULL || /* layer 3 packet */
1584 ( m->m_pkthdr.len >= sizeof(struct ip) &&
1585 ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1586 hlen = ip->ip_hl << 2;
1587
1588 /*
1589 * Collect parameters into local variables for faster matching.
1590 */
1591 if (hlen == 0) { /* do not grab addresses for non-ip pkts */
1592 proto = args->f_id.proto = 0; /* mark f_id invalid */
1593 goto after_ip_checks;
1594 }
1595
1596 proto = args->f_id.proto = ip->ip_p;
1597 src_ip = ip->ip_src;
1598 dst_ip = ip->ip_dst;
1599 if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1600 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1601 ip_len = ntohs(ip->ip_len);
1602 } else {
1603 offset = ip->ip_off & IP_OFFMASK;
1604 ip_len = ip->ip_len;
1605 }
1606 pktlen = ip_len < pktlen ? ip_len : pktlen;
1607
1608 #define PULLUP_TO(len) \
1609 do { \
1610 if ((m)->m_len < (len)) { \
1611 args->m = m = m_pullup(m, (len)); \
1612 if (m == 0) \
1613 goto pullup_failed; \
1614 ip = mtod(m, struct ip *); \
1615 } \
1616 } while (0)
1617
1618 if (offset == 0) {
1619 switch (proto) {
1620 case IPPROTO_TCP:
1621 {
1622 struct tcphdr *tcp;
1623
1624 PULLUP_TO(hlen + sizeof(struct tcphdr));
1625 tcp = L3HDR(struct tcphdr, ip);
1626 dst_port = tcp->th_dport;
1627 src_port = tcp->th_sport;
1628 args->f_id.flags = tcp->th_flags;
1629 }
1630 break;
1631
1632 case IPPROTO_UDP:
1633 {
1634 struct udphdr *udp;
1635
1636 PULLUP_TO(hlen + sizeof(struct udphdr));
1637 udp = L3HDR(struct udphdr, ip);
1638 dst_port = udp->uh_dport;
1639 src_port = udp->uh_sport;
1640 }
1641 break;
1642
1643 case IPPROTO_ICMP:
1644 PULLUP_TO(hlen + 4); /* type, code and checksum. */
1645 args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1646 break;
1647
1648 default:
1649 break;
1650 }
1651 #undef PULLUP_TO
1652 }
1653
1654 args->f_id.src_ip = ntohl(src_ip.s_addr);
1655 args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1656 args->f_id.src_port = src_port = ntohs(src_port);
1657 args->f_id.dst_port = dst_port = ntohs(dst_port);
1658
1659 after_ip_checks:
1660 if (args->rule) {
1661 /*
1662 * Packet has already been tagged. Look for the next rule
1663 * to restart processing.
1664 *
1665 * If fw_one_pass != 0 then just accept it.
1666 * XXX should not happen here, but optimized out in
1667 * the caller.
1668 */
1669 if (fw_one_pass)
1670 return 0;
1671
1672 f = args->rule->next_rule;
1673 if (f == NULL)
1674 f = lookup_next_rule(args->rule);
1675 } else {
1676 /*
1677 * Find the starting rule. It can be either the first
1678 * one, or the one after divert_rule if asked so.
1679 */
1680 int skipto = args->divert_rule;
1681
1682 f = layer3_chain;
1683 if (args->eh == NULL && skipto != 0) {
1684 if (skipto >= IPFW_DEFAULT_RULE)
1685 return(IP_FW_PORT_DENY_FLAG); /* invalid */
1686 while (f && f->rulenum <= skipto)
1687 f = f->next;
1688 if (f == NULL) /* drop packet */
1689 return(IP_FW_PORT_DENY_FLAG);
1690 }
1691 }
1692 args->divert_rule = 0; /* reset to avoid confusion later */
1693
1694 /*
1695 * Now scan the rules, and parse microinstructions for each rule.
1696 */
1697 for (; f; f = f->next) {
1698 int l, cmdlen;
1699 ipfw_insn *cmd;
1700 int skip_or; /* skip rest of OR block */
1701
1702 again:
1703 if (set_disable & (1 << f->set) )
1704 continue;
1705
1706 skip_or = 0;
1707 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ;
1708 l -= cmdlen, cmd += cmdlen) {
1709 int match;
1710
1711 /*
1712 * check_body is a jump target used when we find a
1713 * CHECK_STATE, and need to jump to the body of
1714 * the target rule.
1715 */
1716
1717 check_body:
1718 cmdlen = F_LEN(cmd);
1719 /*
1720 * An OR block (insn_1 || .. || insn_n) has the
1721 * F_OR bit set in all but the last instruction.
1722 * The first match will set "skip_or", and cause
1723 * the following instructions to be skipped until
1724 * past the one with the F_OR bit clear.
1725 */
1726 if (skip_or) { /* skip this instruction */
1727 if ((cmd->len & F_OR) == 0)
1728 skip_or = 0; /* next one is good */
1729 continue;
1730 }
1731 match = 0; /* set to 1 if we succeed */
1732
1733 switch (cmd->opcode) {
1734 /*
1735 * The first set of opcodes compares the packet's
1736 * fields with some pattern, setting 'match' if a
1737 * match is found. At the end of the loop there is
1738 * logic to deal with F_NOT and F_OR flags associated
1739 * with the opcode.
1740 */
1741 case O_NOP:
1742 match = 1;
1743 break;
1744
1745 case O_FORWARD_MAC:
1746 printf("ipfw: opcode %d unimplemented\n",
1747 cmd->opcode);
1748 break;
1749
1750 case O_GID:
1751 case O_UID:
1752 /*
1753 * We only check offset == 0 && proto != 0,
1754 * as this ensures that we have an IPv4
1755 * packet with the ports info.
1756 */
1757 if (offset!=0)
1758 break;
1759 {
1760 struct inpcbinfo *pi;
1761 int wildcard;
1762 struct inpcb *pcb;
1763
1764 if (proto == IPPROTO_TCP) {
1765 wildcard = 0;
1766 pi = &tcbinfo;
1767 } else if (proto == IPPROTO_UDP) {
1768 wildcard = 1;
1769 pi = &udbinfo;
1770 } else
1771 break;
1772
1773 pcb = (oif) ?
1774 in_pcblookup_hash(pi,
1775 dst_ip, htons(dst_port),
1776 src_ip, htons(src_port),
1777 wildcard, oif) :
1778 in_pcblookup_hash(pi,
1779 src_ip, htons(src_port),
1780 dst_ip, htons(dst_port),
1781 wildcard, NULL);
1782
1783 if (pcb == NULL || pcb->inp_socket == NULL)
1784 break;
1785 #if __FreeBSD_version < 500034
1786 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1787 #endif
1788 if (cmd->opcode == O_UID) {
1789 match =
1790 !socheckuid(pcb->inp_socket,
1791 (uid_t)((ipfw_insn_u32 *)cmd)->d[0]);
1792 } else {
1793 match = groupmember(
1794 (uid_t)((ipfw_insn_u32 *)cmd)->d[0],
1795 pcb->inp_socket->so_cred);
1796 }
1797 }
1798 break;
1799
1800 case O_RECV:
1801 match = iface_match(m->m_pkthdr.rcvif,
1802 (ipfw_insn_if *)cmd);
1803 break;
1804
1805 case O_XMIT:
1806 match = iface_match(oif, (ipfw_insn_if *)cmd);
1807 break;
1808
1809 case O_VIA:
1810 match = iface_match(oif ? oif :
1811 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1812 break;
1813
1814 case O_MACADDR2:
1815 if (args->eh != NULL) { /* have MAC header */
1816 u_int32_t *want = (u_int32_t *)
1817 ((ipfw_insn_mac *)cmd)->addr;
1818 u_int32_t *mask = (u_int32_t *)
1819 ((ipfw_insn_mac *)cmd)->mask;
1820 u_int32_t *hdr = (u_int32_t *)args->eh;
1821
1822 match =
1823 ( want[0] == (hdr[0] & mask[0]) &&
1824 want[1] == (hdr[1] & mask[1]) &&
1825 want[2] == (hdr[2] & mask[2]) );
1826 }
1827 break;
1828
1829 case O_MAC_TYPE:
1830 if (args->eh != NULL) {
1831 u_int16_t t =
1832 ntohs(args->eh->ether_type);
1833 u_int16_t *p =
1834 ((ipfw_insn_u16 *)cmd)->ports;
1835 int i;
1836
1837 for (i = cmdlen - 1; !match && i>0;
1838 i--, p += 2)
1839 match = (t>=p[0] && t<=p[1]);
1840 }
1841 break;
1842
1843 case O_FRAG:
1844 match = (hlen > 0 && offset != 0);
1845 break;
1846
1847 case O_IN: /* "out" is "not in" */
1848 match = (oif == NULL);
1849 break;
1850
1851 case O_LAYER2:
1852 match = (args->eh != NULL);
1853 break;
1854
1855 case O_PROTO:
1856 /*
1857 * We do not allow an arg of 0 so the
1858 * check of "proto" only suffices.
1859 */
1860 match = (proto == cmd->arg1);
1861 break;
1862
1863 case O_IP_SRC:
1864 match = (hlen > 0 &&
1865 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1866 src_ip.s_addr);
1867 break;
1868
1869 case O_IP_SRC_LOOKUP:
1870 case O_IP_DST_LOOKUP:
1871 if (hlen > 0) {
1872 uint32_t a =
1873 (cmd->opcode == O_IP_DST_LOOKUP) ?
1874 dst_ip.s_addr : src_ip.s_addr;
1875 uint32_t v;
1876
1877 match = lookup_table(cmd->arg1, a, &v);
1878 if (!match)
1879 break;
1880 if (cmdlen == F_INSN_SIZE(ipfw_insn_u32))
1881 match =
1882 ((ipfw_insn_u32 *)cmd)->d[0] == v;
1883 }
1884 break;
1885
1886 case O_IP_SRC_MASK:
1887 case O_IP_DST_MASK:
1888 if (hlen > 0) {
1889 uint32_t a =
1890 (cmd->opcode == O_IP_DST_MASK) ?
1891 dst_ip.s_addr : src_ip.s_addr;
1892 uint32_t *p = ((ipfw_insn_u32 *)cmd)->d;
1893 int i = cmdlen-1;
1894
1895 for (; !match && i>0; i-= 2, p+= 2)
1896 match = (p[0] == (a & p[1]));
1897 }
1898 break;
1899
1900 case O_IP_SRC_ME:
1901 if (hlen > 0) {
1902 struct ifnet *tif;
1903
1904 INADDR_TO_IFP(src_ip, tif);
1905 match = (tif != NULL);
1906 }
1907 break;
1908
1909 case O_IP_DST_SET:
1910 case O_IP_SRC_SET:
1911 if (hlen > 0) {
1912 u_int32_t *d = (u_int32_t *)(cmd+1);
1913 u_int32_t addr =
1914 cmd->opcode == O_IP_DST_SET ?
1915 args->f_id.dst_ip :
1916 args->f_id.src_ip;
1917
1918 if (addr < d[0])
1919 break;
1920 addr -= d[0]; /* subtract base */
1921 match = (addr < cmd->arg1) &&
1922 ( d[ 1 + (addr>>5)] &
1923 (1<<(addr & 0x1f)) );
1924 }
1925 break;
1926
1927 case O_IP_DST:
1928 match = (hlen > 0 &&
1929 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1930 dst_ip.s_addr);
1931 break;
1932
1933 case O_IP_DST_ME:
1934 if (hlen > 0) {
1935 struct ifnet *tif;
1936
1937 INADDR_TO_IFP(dst_ip, tif);
1938 match = (tif != NULL);
1939 }
1940 break;
1941
1942 case O_IP_SRCPORT:
1943 case O_IP_DSTPORT:
1944 /*
1945 * offset == 0 && proto != 0 is enough
1946 * to guarantee that we have an IPv4
1947 * packet with port info.
1948 */
1949 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
1950 && offset == 0) {
1951 u_int16_t x =
1952 (cmd->opcode == O_IP_SRCPORT) ?
1953 src_port : dst_port ;
1954 u_int16_t *p =
1955 ((ipfw_insn_u16 *)cmd)->ports;
1956 int i;
1957
1958 for (i = cmdlen - 1; !match && i>0;
1959 i--, p += 2)
1960 match = (x>=p[0] && x<=p[1]);
1961 }
1962 break;
1963
1964 case O_ICMPTYPE:
1965 match = (offset == 0 && proto==IPPROTO_ICMP &&
1966 icmptype_match(ip, (ipfw_insn_u32 *)cmd) );
1967 break;
1968
1969 case O_IPOPT:
1970 match = (hlen > 0 && ipopts_match(ip, cmd) );
1971 break;
1972
1973 case O_IPVER:
1974 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
1975 break;
1976
1977 case O_IPID:
1978 case O_IPLEN:
1979 case O_IPTTL:
1980 if (hlen > 0) { /* only for IP packets */
1981 uint16_t x;
1982 uint16_t *p;
1983 int i;
1984
1985 if (cmd->opcode == O_IPLEN)
1986 x = ip_len;
1987 else if (cmd->opcode == O_IPTTL)
1988 x = ip->ip_ttl;
1989 else /* must be IPID */
1990 x = ntohs(ip->ip_id);
1991 if (cmdlen == 1) {
1992 match = (cmd->arg1 == x);
1993 break;
1994 }
1995 /* otherwise we have ranges */
1996 p = ((ipfw_insn_u16 *)cmd)->ports;
1997 i = cmdlen - 1;
1998 for (; !match && i>0; i--, p += 2)
1999 match = (x >= p[0] && x <= p[1]);
2000 }
2001 break;
2002
2003 case O_IPPRECEDENCE:
2004 match = (hlen > 0 &&
2005 (cmd->arg1 == (ip->ip_tos & 0xe0)) );
2006 break;
2007
2008 case O_IPTOS:
2009 match = (hlen > 0 &&
2010 flags_match(cmd, ip->ip_tos));
2011 break;
2012
2013 case O_TCPFLAGS:
2014 match = (proto == IPPROTO_TCP && offset == 0 &&
2015 flags_match(cmd,
2016 L3HDR(struct tcphdr,ip)->th_flags));
2017 break;
2018
2019 case O_TCPOPTS:
2020 match = (proto == IPPROTO_TCP && offset == 0 &&
2021 tcpopts_match(ip, cmd));
2022 break;
2023
2024 case O_TCPSEQ:
2025 match = (proto == IPPROTO_TCP && offset == 0 &&
2026 ((ipfw_insn_u32 *)cmd)->d[0] ==
2027 L3HDR(struct tcphdr,ip)->th_seq);
2028 break;
2029
2030 case O_TCPACK:
2031 match = (proto == IPPROTO_TCP && offset == 0 &&
2032 ((ipfw_insn_u32 *)cmd)->d[0] ==
2033 L3HDR(struct tcphdr,ip)->th_ack);
2034 break;
2035
2036 case O_TCPWIN:
2037 match = (proto == IPPROTO_TCP && offset == 0 &&
2038 cmd->arg1 ==
2039 L3HDR(struct tcphdr,ip)->th_win);
2040 break;
2041
2042 case O_ESTAB:
2043 /* reject packets which have SYN only */
2044 /* XXX should i also check for TH_ACK ? */
2045 match = (proto == IPPROTO_TCP && offset == 0 &&
2046 (L3HDR(struct tcphdr,ip)->th_flags &
2047 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2048 break;
2049
2050 case O_LOG:
2051 if (fw_verbose)
2052 ipfw_log(f, hlen, args->eh, m, oif);
2053 match = 1;
2054 break;
2055
2056 case O_PROB:
2057 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]);
2058 break;
2059
2060 case O_VERREVPATH:
2061 /* Outgoing packets automatically pass/match */
2062 match = ((oif != NULL) ||
2063 (m->m_pkthdr.rcvif == NULL) ||
2064 verify_rev_path(src_ip, m->m_pkthdr.rcvif));
2065 break;
2066
2067 case O_IPSEC:
2068 #ifdef FAST_IPSEC
2069 match = (m_tag_find(m,
2070 PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL);
2071 #endif
2072 #ifdef IPSEC
2073 match = (ipsec_gethist(m, NULL) != NULL);
2074 #endif
2075 /* otherwise no match */
2076 break;
2077
2078 /*
2079 * The second set of opcodes represents 'actions',
2080 * i.e. the terminal part of a rule once the packet
2081 * matches all previous patterns.
2082 * Typically there is only one action for each rule,
2083 * and the opcode is stored at the end of the rule
2084 * (but there are exceptions -- see below).
2085 *
2086 * In general, here we set retval and terminate the
2087 * outer loop (would be a 'break 3' in some language,
2088 * but we need to do a 'goto done').
2089 *
2090 * Exceptions:
2091 * O_COUNT and O_SKIPTO actions:
2092 * instead of terminating, we jump to the next rule
2093 * ('goto next_rule', equivalent to a 'break 2'),
2094 * or to the SKIPTO target ('goto again' after
2095 * having set f, cmd and l), respectively.
2096 *
2097 * O_LIMIT and O_KEEP_STATE: these opcodes are
2098 * not real 'actions', and are stored right
2099 * before the 'action' part of the rule.
2100 * These opcodes try to install an entry in the
2101 * state tables; if successful, we continue with
2102 * the next opcode (match=1; break;), otherwise
2103 * the packet * must be dropped
2104 * ('goto done' after setting retval);
2105 *
2106 * O_PROBE_STATE and O_CHECK_STATE: these opcodes
2107 * cause a lookup of the state table, and a jump
2108 * to the 'action' part of the parent rule
2109 * ('goto check_body') if an entry is found, or
2110 * (CHECK_STATE only) a jump to the next rule if
2111 * the entry is not found ('goto next_rule').
2112 * The result of the lookup is cached to make
2113 * further instances of these opcodes are
2114 * effectively NOPs.
2115 */
2116 case O_LIMIT:
2117 case O_KEEP_STATE:
2118 if (install_state(f,
2119 (ipfw_insn_limit *)cmd, args)) {
2120 retval = IP_FW_PORT_DENY_FLAG;
2121 goto done; /* error/limit violation */
2122 }
2123 match = 1;
2124 break;
2125
2126 case O_PROBE_STATE:
2127 case O_CHECK_STATE:
2128 /*
2129 * dynamic rules are checked at the first
2130 * keep-state or check-state occurrence,
2131 * with the result being stored in dyn_dir.
2132 * The compiler introduces a PROBE_STATE
2133 * instruction for us when we have a
2134 * KEEP_STATE (because PROBE_STATE needs
2135 * to be run first).
2136 */
2137 if (dyn_dir == MATCH_UNKNOWN &&
2138 (q = lookup_dyn_rule(&args->f_id,
2139 &dyn_dir, proto == IPPROTO_TCP ?
2140 L3HDR(struct tcphdr, ip) : NULL))
2141 != NULL) {
2142 /*
2143 * Found dynamic entry, update stats
2144 * and jump to the 'action' part of
2145 * the parent rule.
2146 */
2147 q->pcnt++;
2148 q->bcnt += pktlen;
2149 f = q->rule;
2150 cmd = ACTION_PTR(f);
2151 l = f->cmd_len - f->act_ofs;
2152 goto check_body;
2153 }
2154 /*
2155 * Dynamic entry not found. If CHECK_STATE,
2156 * skip to next rule, if PROBE_STATE just
2157 * ignore and continue with next opcode.
2158 */
2159 if (cmd->opcode == O_CHECK_STATE)
2160 goto next_rule;
2161 match = 1;
2162 break;
2163
2164 case O_ACCEPT:
2165 retval = 0; /* accept */
2166 goto done;
2167
2168 case O_PIPE:
2169 case O_QUEUE:
2170 args->rule = f; /* report matching rule */
2171 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG;
2172 goto done;
2173
2174 case O_DIVERT:
2175 case O_TEE:
2176 if (args->eh) /* not on layer 2 */
2177 break;
2178 args->divert_rule = f->rulenum;
2179 retval = (cmd->opcode == O_DIVERT) ?
2180 cmd->arg1 :
2181 cmd->arg1 | IP_FW_PORT_TEE_FLAG;
2182 goto done;
2183
2184 case O_COUNT:
2185 case O_SKIPTO:
2186 f->pcnt++; /* update stats */
2187 f->bcnt += pktlen;
2188 f->timestamp = time_second;
2189 if (cmd->opcode == O_COUNT)
2190 goto next_rule;
2191 /* handle skipto */
2192 if (f->next_rule == NULL)
2193 lookup_next_rule(f);
2194 f = f->next_rule;
2195 goto again;
2196
2197 case O_REJECT:
2198 /*
2199 * Drop the packet and send a reject notice
2200 * if the packet is not ICMP (or is an ICMP
2201 * query), and it is not multicast/broadcast.
2202 */
2203 if (hlen > 0 &&
2204 (proto != IPPROTO_ICMP ||
2205 is_icmp_query(ip)) &&
2206 !(m->m_flags & (M_BCAST|M_MCAST)) &&
2207 !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
2208 send_reject(args, cmd->arg1,
2209 offset,ip_len);
2210 m = args->m;
2211 }
2212 /* FALLTHROUGH */
2213 case O_DENY:
2214 retval = IP_FW_PORT_DENY_FLAG;
2215 goto done;
2216
2217 case O_FORWARD_IP:
2218 if (args->eh) /* not valid on layer2 pkts */
2219 break;
2220 if (!q || dyn_dir == MATCH_FORWARD)
2221 args->next_hop =
2222 &((ipfw_insn_sa *)cmd)->sa;
2223 retval = 0;
2224 goto done;
2225
2226 default:
2227 panic("-- unknown opcode %d\n", cmd->opcode);
2228 } /* end of switch() on opcodes */
2229
2230 if (cmd->len & F_NOT)
2231 match = !match;
2232
2233 if (match) {
2234 if (cmd->len & F_OR)
2235 skip_or = 1;
2236 } else {
2237 if (!(cmd->len & F_OR)) /* not an OR block, */
2238 break; /* try next rule */
2239 }
2240
2241 } /* end of inner for, scan opcodes */
2242
2243 next_rule:; /* try next rule */
2244
2245 } /* end of outer for, scan rules */
2246 printf("ipfw: ouch!, skip past end of rules, denying packet\n");
2247 return(IP_FW_PORT_DENY_FLAG);
2248
2249 done:
2250 /* Update statistics */
2251 f->pcnt++;
2252 f->bcnt += pktlen;
2253 f->timestamp = time_second;
2254 return retval;
2255
2256 pullup_failed:
2257 if (fw_verbose)
2258 printf("ipfw: pullup failed\n");
2259 return(IP_FW_PORT_DENY_FLAG);
2260 }
2261
2262 /*
2263 * When a rule is added/deleted, clear the next_rule pointers in all rules.
2264 * These will be reconstructed on the fly as packets are matched.
2265 * Must be called at splimp().
2266 */
2267 static void
2268 flush_rule_ptrs(void)
2269 {
2270 struct ip_fw *rule;
2271
2272 for (rule = layer3_chain; rule; rule = rule->next)
2273 rule->next_rule = NULL;
2274 }
2275
2276 /*
2277 * When pipes/queues are deleted, clear the "pipe_ptr" pointer to a given
2278 * pipe/queue, or to all of them (match == NULL).
2279 * Must be called at splimp().
2280 */
2281 void
2282 flush_pipe_ptrs(struct dn_flow_set *match)
2283 {
2284 struct ip_fw *rule;
2285
2286 for (rule = layer3_chain; rule; rule = rule->next) {
2287 ipfw_insn_pipe *cmd = (ipfw_insn_pipe *)ACTION_PTR(rule);
2288
2289 if (cmd->o.opcode != O_PIPE && cmd->o.opcode != O_QUEUE)
2290 continue;
2291 /*
2292 * XXX Use bcmp/bzero to handle pipe_ptr to overcome
2293 * possible alignment problems on 64-bit architectures.
2294 * This code is seldom used so we do not worry too
2295 * much about efficiency.
2296 */
2297 if (match == NULL ||
2298 !bcmp(&cmd->pipe_ptr, &match, sizeof(match)) )
2299 bzero(&cmd->pipe_ptr, sizeof(cmd->pipe_ptr));
2300 }
2301 }
2302
2303 /*
2304 * Add a new rule to the list. Copy the rule into a malloc'ed area, then
2305 * possibly create a rule number and add the rule to the list.
2306 * Update the rule_number in the input struct so the caller knows it as well.
2307 */
2308 static int
2309 add_rule(struct ip_fw **head, struct ip_fw *input_rule)
2310 {
2311 struct ip_fw *rule, *f, *prev;
2312 int s;
2313 int l = RULESIZE(input_rule);
2314
2315 if (*head == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE)
2316 return (EINVAL);
2317
2318 rule = malloc(l, M_IPFW, M_NOWAIT | M_ZERO);
2319 if (rule == NULL)
2320 return (ENOSPC);
2321
2322 bcopy(input_rule, rule, l);
2323
2324 rule->next = NULL;
2325 rule->next_rule = NULL;
2326
2327 rule->pcnt = 0;
2328 rule->bcnt = 0;
2329 rule->timestamp = 0;
2330
2331 s = splimp();
2332
2333 if (*head == NULL) { /* default rule */
2334 *head = rule;
2335 goto done;
2336 }
2337
2338 /*
2339 * If rulenum is 0, find highest numbered rule before the
2340 * default rule, and add autoinc_step
2341 */
2342 if (autoinc_step < 1)
2343 autoinc_step = 1;
2344 else if (autoinc_step > 1000)
2345 autoinc_step = 1000;
2346 if (rule->rulenum == 0) {
2347 /*
2348 * locate the highest numbered rule before default
2349 */
2350 for (f = *head; f; f = f->next) {
2351 if (f->rulenum == IPFW_DEFAULT_RULE)
2352 break;
2353 rule->rulenum = f->rulenum;
2354 }
2355 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
2356 rule->rulenum += autoinc_step;
2357 input_rule->rulenum = rule->rulenum;
2358 }
2359
2360 /*
2361 * Now insert the new rule in the right place in the sorted list.
2362 */
2363 for (prev = NULL, f = *head; f; prev = f, f = f->next) {
2364 if (f->rulenum > rule->rulenum) { /* found the location */
2365 if (prev) {
2366 rule->next = f;
2367 prev->next = rule;
2368 } else { /* head insert */
2369 rule->next = *head;
2370 *head = rule;
2371 }
2372 break;
2373 }
2374 }
2375 flush_rule_ptrs();
2376 done:
2377 static_count++;
2378 static_len += l;
2379 splx(s);
2380 DEB(printf("ipfw: installed rule %d, static count now %d\n",
2381 rule->rulenum, static_count);)
2382 return (0);
2383 }
2384
2385 /**
2386 * Free storage associated with a static rule (including derived
2387 * dynamic rules).
2388 * The caller is in charge of clearing rule pointers to avoid
2389 * dangling pointers.
2390 * @return a pointer to the next entry.
2391 * Arguments are not checked, so they better be correct.
2392 * Must be called at splimp().
2393 */
2394 static struct ip_fw *
2395 delete_rule(struct ip_fw **head, struct ip_fw *prev, struct ip_fw *rule)
2396 {
2397 struct ip_fw *n;
2398 int l = RULESIZE(rule);
2399
2400 n = rule->next;
2401 remove_dyn_rule(rule, NULL /* force removal */);
2402 if (prev == NULL)
2403 *head = n;
2404 else
2405 prev->next = n;
2406 static_count--;
2407 static_len -= l;
2408
2409 if (DUMMYNET_LOADED)
2410 ip_dn_ruledel_ptr(rule);
2411 free(rule, M_IPFW);
2412 return n;
2413 }
2414
2415 /*
2416 * Deletes all rules from a chain (except rules in set RESVD_SET
2417 * unless kill_default = 1).
2418 * Must be called at splimp().
2419 */
2420 static void
2421 free_chain(struct ip_fw **chain, int kill_default)
2422 {
2423 struct ip_fw *prev, *rule;
2424
2425 flush_rule_ptrs(); /* more efficient to do outside the loop */
2426 for (prev = NULL, rule = *chain; rule ; )
2427 if (kill_default || rule->set != RESVD_SET)
2428 rule = delete_rule(chain, prev, rule);
2429 else {
2430 prev = rule;
2431 rule = rule->next;
2432 }
2433 }
2434
2435 /**
2436 * Remove all rules with given number, and also do set manipulation.
2437 * Assumes chain != NULL && *chain != NULL.
2438 *
2439 * The argument is an u_int32_t. The low 16 bit are the rule or set number,
2440 * the next 8 bits are the new set, the top 8 bits are the command:
2441 *
2442 * 0 delete rules with given number
2443 * 1 delete rules with given set number
2444 * 2 move rules with given number to new set
2445 * 3 move rules with given set number to new set
2446 * 4 swap sets with given numbers
2447 */
2448 static int
2449 del_entry(struct ip_fw **chain, u_int32_t arg)
2450 {
2451 struct ip_fw *prev = NULL, *rule = *chain;
2452 int s;
2453 u_int16_t rulenum; /* rule or old_set */
2454 u_int8_t cmd, new_set;
2455
2456 rulenum = arg & 0xffff;
2457 cmd = (arg >> 24) & 0xff;
2458 new_set = (arg >> 16) & 0xff;
2459
2460 if (cmd > 4)
2461 return EINVAL;
2462 if (new_set > RESVD_SET)
2463 return EINVAL;
2464 if (cmd == 0 || cmd == 2) {
2465 if (rulenum >= IPFW_DEFAULT_RULE)
2466 return EINVAL;
2467 } else {
2468 if (rulenum > RESVD_SET) /* old_set */
2469 return EINVAL;
2470 }
2471
2472 switch (cmd) {
2473 case 0: /* delete rules with given number */
2474 /*
2475 * locate first rule to delete
2476 */
2477 for (; rule->rulenum < rulenum; prev = rule, rule = rule->next)
2478 ;
2479 if (rule->rulenum != rulenum)
2480 return EINVAL;
2481
2482 s = splimp(); /* no access to rules while removing */
2483 /*
2484 * flush pointers outside the loop, then delete all matching
2485 * rules. prev remains the same throughout the cycle.
2486 */
2487 flush_rule_ptrs();
2488 while (rule->rulenum == rulenum)
2489 rule = delete_rule(chain, prev, rule);
2490 splx(s);
2491 break;
2492
2493 case 1: /* delete all rules with given set number */
2494 s = splimp();
2495 flush_rule_ptrs();
2496 while (rule->rulenum < IPFW_DEFAULT_RULE)
2497 if (rule->set == rulenum)
2498 rule = delete_rule(chain, prev, rule);
2499 else {
2500 prev = rule;
2501 rule = rule->next;
2502 }
2503 splx(s);
2504 break;
2505
2506 case 2: /* move rules with given number to new set */
2507 s = splimp();
2508 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
2509 if (rule->rulenum == rulenum)
2510 rule->set = new_set;
2511 splx(s);
2512 break;
2513
2514 case 3: /* move rules with given set number to new set */
2515 s = splimp();
2516 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
2517 if (rule->set == rulenum)
2518 rule->set = new_set;
2519 splx(s);
2520 break;
2521
2522 case 4: /* swap two sets */
2523 s = splimp();
2524 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
2525 if (rule->set == rulenum)
2526 rule->set = new_set;
2527 else if (rule->set == new_set)
2528 rule->set = rulenum;
2529 splx(s);
2530 break;
2531 }
2532 return 0;
2533 }
2534
2535 /*
2536 * Clear counters for a specific rule.
2537 */
2538 static void
2539 clear_counters(struct ip_fw *rule, int log_only)
2540 {
2541 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
2542
2543 if (log_only == 0) {
2544 rule->bcnt = rule->pcnt = 0;
2545 rule->timestamp = 0;
2546 }
2547 if (l->o.opcode == O_LOG)
2548 l->log_left = l->max_log;
2549 }
2550
2551 /**
2552 * Reset some or all counters on firewall rules.
2553 * @arg frwl is null to clear all entries, or contains a specific
2554 * rule number.
2555 * @arg log_only is 1 if we only want to reset logs, zero otherwise.
2556 */
2557 static int
2558 zero_entry(int rulenum, int log_only)
2559 {
2560 struct ip_fw *rule;
2561 int s;
2562 char *msg;
2563
2564 if (rulenum == 0) {
2565 s = splimp();
2566 norule_counter = 0;
2567 for (rule = layer3_chain; rule; rule = rule->next)
2568 clear_counters(rule, log_only);
2569 splx(s);
2570 msg = log_only ? "ipfw: All logging counts reset.\n" :
2571 "ipfw: Accounting cleared.\n";
2572 } else {
2573 int cleared = 0;
2574 /*
2575 * We can have multiple rules with the same number, so we
2576 * need to clear them all.
2577 */
2578 for (rule = layer3_chain; rule; rule = rule->next)
2579 if (rule->rulenum == rulenum) {
2580 s = splimp();
2581 while (rule && rule->rulenum == rulenum) {
2582 clear_counters(rule, log_only);
2583 rule = rule->next;
2584 }
2585 splx(s);
2586 cleared = 1;
2587 break;
2588 }
2589 if (!cleared) /* we did not find any matching rules */
2590 return (EINVAL);
2591 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
2592 "ipfw: Entry %d cleared.\n";
2593 }
2594 if (fw_verbose)
2595 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
2596 return (0);
2597 }
2598
2599 /*
2600 * Check validity of the structure before insert.
2601 * Fortunately rules are simple, so this mostly need to check rule sizes.
2602 */
2603 static int
2604 check_ipfw_struct(struct ip_fw *rule, int size)
2605 {
2606 int l, cmdlen = 0;
2607 int have_action=0;
2608 ipfw_insn *cmd;
2609
2610 if (size < sizeof(*rule)) {
2611 printf("ipfw: rule too short\n");
2612 return (EINVAL);
2613 }
2614 /* first, check for valid size */
2615 l = RULESIZE(rule);
2616 if (l != size) {
2617 printf("ipfw: size mismatch (have %d want %d)\n", size, l);
2618 return (EINVAL);
2619 }
2620 /*
2621 * Now go for the individual checks. Very simple ones, basically only
2622 * instruction sizes.
2623 */
2624 for (l = rule->cmd_len, cmd = rule->cmd ;
2625 l > 0 ; l -= cmdlen, cmd += cmdlen) {
2626 cmdlen = F_LEN(cmd);
2627 if (cmdlen > l) {
2628 printf("ipfw: opcode %d size truncated\n",
2629 cmd->opcode);
2630 return EINVAL;
2631 }
2632 DEB(printf("ipfw: opcode %d\n", cmd->opcode);)
2633 switch (cmd->opcode) {
2634 case O_PROBE_STATE:
2635 case O_KEEP_STATE:
2636 case O_PROTO:
2637 case O_IP_SRC_ME:
2638 case O_IP_DST_ME:
2639 case O_LAYER2:
2640 case O_IN:
2641 case O_FRAG:
2642 case O_IPOPT:
2643 case O_IPTOS:
2644 case O_IPPRECEDENCE:
2645 case O_IPVER:
2646 case O_TCPWIN:
2647 case O_TCPFLAGS:
2648 case O_TCPOPTS:
2649 case O_ESTAB:
2650 case O_VERREVPATH:
2651 case O_IPSEC:
2652 if (cmdlen != F_INSN_SIZE(ipfw_insn))
2653 goto bad_size;
2654 break;
2655
2656 case O_UID:
2657 case O_GID:
2658 case O_IP_SRC:
2659 case O_IP_DST:
2660 case O_TCPSEQ:
2661 case O_TCPACK:
2662 case O_PROB:
2663 case O_ICMPTYPE:
2664 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
2665 goto bad_size;
2666 break;
2667
2668 case O_LIMIT:
2669 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
2670 goto bad_size;
2671 break;
2672
2673 case O_LOG:
2674 if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
2675 goto bad_size;
2676
2677 ((ipfw_insn_log *)cmd)->log_left =
2678 ((ipfw_insn_log *)cmd)->max_log;
2679
2680 break;
2681
2682 case O_IP_SRC_MASK:
2683 case O_IP_DST_MASK:
2684 /* only odd command lengths */
2685 if ( !(cmdlen & 1) || cmdlen > 31)
2686 goto bad_size;
2687 break;
2688
2689 case O_IP_SRC_SET:
2690 case O_IP_DST_SET:
2691 if (cmd->arg1 == 0 || cmd->arg1 > 256) {
2692 printf("ipfw: invalid set size %d\n",
2693 cmd->arg1);
2694 return EINVAL;
2695 }
2696 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
2697 (cmd->arg1+31)/32 )
2698 goto bad_size;
2699 break;
2700
2701 case O_IP_SRC_LOOKUP:
2702 case O_IP_DST_LOOKUP:
2703 if (cmd->arg1 >= IPFW_TABLES_MAX) {
2704 printf("ipfw: invalid table number %d\n",
2705 cmd->arg1);
2706 return (EINVAL);
2707 }
2708 if (cmdlen != F_INSN_SIZE(ipfw_insn) &&
2709 cmdlen != F_INSN_SIZE(ipfw_insn_u32))
2710 goto bad_size;
2711 break;
2712
2713 case O_MACADDR2:
2714 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
2715 goto bad_size;
2716 break;
2717
2718 case O_NOP:
2719 case O_IPID:
2720 case O_IPTTL:
2721 case O_IPLEN:
2722 if (cmdlen < 1 || cmdlen > 31)
2723 goto bad_size;
2724 break;
2725
2726 case O_MAC_TYPE:
2727 case O_IP_SRCPORT:
2728 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
2729 if (cmdlen < 2 || cmdlen > 31)
2730 goto bad_size;
2731 break;
2732
2733 case O_RECV:
2734 case O_XMIT:
2735 case O_VIA:
2736 if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
2737 goto bad_size;
2738 break;
2739
2740 case O_PIPE:
2741 case O_QUEUE:
2742 if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
2743 goto bad_size;
2744 goto check_action;
2745
2746 case O_FORWARD_IP:
2747 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
2748 goto bad_size;
2749 goto check_action;
2750
2751 case O_DIVERT:
2752 case O_TEE:
2753 #ifndef IPDIVERT
2754 return EINVAL;
2755 #endif
2756 case O_FORWARD_MAC: /* XXX not implemented yet */
2757 case O_CHECK_STATE:
2758 case O_COUNT:
2759 case O_ACCEPT:
2760 case O_DENY:
2761 case O_REJECT:
2762 case O_SKIPTO:
2763 if (cmdlen != F_INSN_SIZE(ipfw_insn))
2764 goto bad_size;
2765 check_action:
2766 if (have_action) {
2767 printf("ipfw: opcode %d, multiple actions"
2768 " not allowed\n",
2769 cmd->opcode);
2770 return EINVAL;
2771 }
2772 have_action = 1;
2773 if (l != cmdlen) {
2774 printf("ipfw: opcode %d, action must be"
2775 " last opcode\n",
2776 cmd->opcode);
2777 return EINVAL;
2778 }
2779 break;
2780 default:
2781 printf("ipfw: opcode %d, unknown opcode\n",
2782 cmd->opcode);
2783 return EINVAL;
2784 }
2785 }
2786 if (have_action == 0) {
2787 printf("ipfw: missing action\n");
2788 return EINVAL;
2789 }
2790 return 0;
2791
2792 bad_size:
2793 printf("ipfw: opcode %d size %d wrong\n",
2794 cmd->opcode, cmdlen);
2795 return EINVAL;
2796 }
2797
2798
2799 /**
2800 * {set|get}sockopt parser.
2801 */
2802 static int
2803 ipfw_ctl(struct sockopt *sopt)
2804 {
2805 int error, s, rulenum;
2806 size_t size;
2807 struct ip_fw *bp , *buf, *rule;
2808
2809 static u_int32_t rule_buf[255]; /* we copy the data here */
2810
2811 /*
2812 * Disallow modifications in really-really secure mode, but still allow
2813 * the logging counters to be reset.
2814 */
2815 if (sopt->sopt_name == IP_FW_ADD ||
2816 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
2817 #if __FreeBSD_version >= 500034
2818 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
2819 if (error)
2820 return (error);
2821 #else /* FreeBSD 4.x */
2822 if (securelevel >= 3)
2823 return (EPERM);
2824 #endif
2825 }
2826
2827 error = 0;
2828
2829 switch (sopt->sopt_name) {
2830 case IP_FW_GET:
2831 /*
2832 * pass up a copy of the current rules. Static rules
2833 * come first (the last of which has number IPFW_DEFAULT_RULE),
2834 * followed by a possibly empty list of dynamic rule.
2835 * The last dynamic rule has NULL in the "next" field.
2836 */
2837 s = splimp();
2838 size = static_len; /* size of static rules */
2839 if (ipfw_dyn_v) /* add size of dyn.rules */
2840 size += (dyn_count * sizeof(ipfw_dyn_rule));
2841
2842 /*
2843 * XXX todo: if the user passes a short length just to know
2844 * how much room is needed, do not bother filling up the
2845 * buffer, just jump to the sooptcopyout.
2846 */
2847 buf = malloc(size, M_TEMP, M_WAITOK);
2848 if (buf == 0) {
2849 splx(s);
2850 error = ENOBUFS;
2851 break;
2852 }
2853
2854 bp = buf;
2855 for (rule = layer3_chain; rule ; rule = rule->next) {
2856 int i = RULESIZE(rule);
2857 bcopy(rule, bp, i);
2858 bcopy(&set_disable, &(bp->next_rule),
2859 sizeof(set_disable));
2860 bp = (struct ip_fw *)((char *)bp + i);
2861 }
2862 if (ipfw_dyn_v) {
2863 int i;
2864 ipfw_dyn_rule *p, *dst, *last = NULL;
2865
2866 dst = (ipfw_dyn_rule *)bp;
2867 for (i = 0 ; i < curr_dyn_buckets ; i++ )
2868 for ( p = ipfw_dyn_v[i] ; p != NULL ;
2869 p = p->next, dst++ ) {
2870 bcopy(p, dst, sizeof *p);
2871 bcopy(&(p->rule->rulenum), &(dst->rule),
2872 sizeof(p->rule->rulenum));
2873 /*
2874 * store a non-null value in "next".
2875 * The userland code will interpret a
2876 * NULL here as a marker
2877 * for the last dynamic rule.
2878 */
2879 bcopy(&dst, &dst->next, sizeof(dst));
2880 last = dst ;
2881 dst->expire =
2882 TIME_LEQ(dst->expire, time_second) ?
2883 0 : dst->expire - time_second ;
2884 }
2885 if (last != NULL) /* mark last dynamic rule */
2886 bzero(&last->next, sizeof(last));
2887 }
2888 splx(s);
2889
2890 error = sooptcopyout(sopt, buf, size);
2891 free(buf, M_TEMP);
2892 break;
2893
2894 case IP_FW_FLUSH:
2895 /*
2896 * Normally we cannot release the lock on each iteration.
2897 * We could do it here only because we start from the head all
2898 * the times so there is no risk of missing some entries.
2899 * On the other hand, the risk is that we end up with
2900 * a very inconsistent ruleset, so better keep the lock
2901 * around the whole cycle.
2902 *
2903 * XXX this code can be improved by resetting the head of
2904 * the list to point to the default rule, and then freeing
2905 * the old list without the need for a lock.
2906 */
2907
2908 s = splimp();
2909 free_chain(&layer3_chain, 0 /* keep default rule */);
2910 splx(s);
2911 break;
2912
2913 case IP_FW_ADD:
2914 rule = (struct ip_fw *)rule_buf; /* XXX do a malloc */
2915 error = sooptcopyin(sopt, rule, sizeof(rule_buf),
2916 sizeof(struct ip_fw) );
2917 size = sopt->sopt_valsize;
2918 if (error || (error = check_ipfw_struct(rule, size)))
2919 break;
2920
2921 error = add_rule(&layer3_chain, rule);
2922 size = RULESIZE(rule);
2923 if (!error && sopt->sopt_dir == SOPT_GET)
2924 error = sooptcopyout(sopt, rule, size);
2925 break;
2926
2927 case IP_FW_DEL:
2928 /*
2929 * IP_FW_DEL is used for deleting single rules or sets,
2930 * and (ab)used to atomically manipulate sets. Argument size
2931 * is used to distinguish between the two:
2932 * sizeof(u_int32_t)
2933 * delete single rule or set of rules,
2934 * or reassign rules (or sets) to a different set.
2935 * 2*sizeof(u_int32_t)
2936 * atomic disable/enable sets.
2937 * first u_int32_t contains sets to be disabled,
2938 * second u_int32_t contains sets to be enabled.
2939 */
2940 error = sooptcopyin(sopt, rule_buf,
2941 2*sizeof(u_int32_t), sizeof(u_int32_t));
2942 if (error)
2943 break;
2944 size = sopt->sopt_valsize;
2945 if (size == sizeof(u_int32_t)) /* delete or reassign */
2946 error = del_entry(&layer3_chain, rule_buf[0]);
2947 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */
2948 set_disable =
2949 (set_disable | rule_buf[0]) & ~rule_buf[1] &
2950 ~(1<<RESVD_SET); /* set RESVD_SET always enabled */
2951 else
2952 error = EINVAL;
2953 break;
2954
2955 case IP_FW_ZERO:
2956 case IP_FW_RESETLOG: /* argument is an int, the rule number */
2957 rulenum=0;
2958
2959 if (sopt->sopt_val != 0) {
2960 error = sooptcopyin(sopt, &rulenum,
2961 sizeof(int), sizeof(int));
2962 if (error)
2963 break;
2964 }
2965 error = zero_entry(rulenum, sopt->sopt_name == IP_FW_RESETLOG);
2966 break;
2967
2968 case IP_FW_TABLE_ADD:
2969 {
2970 ipfw_table_entry ent;
2971
2972 error = sooptcopyin(sopt, &ent,
2973 sizeof(ent), sizeof(ent));
2974 if (error)
2975 break;
2976 error = add_table_entry(ent.tbl, ent.addr,
2977 ent.masklen, ent.value);
2978 }
2979 break;
2980
2981 case IP_FW_TABLE_DEL:
2982 {
2983 ipfw_table_entry ent;
2984
2985 error = sooptcopyin(sopt, &ent,
2986 sizeof(ent), sizeof(ent));
2987 if (error)
2988 break;
2989 error = del_table_entry(ent.tbl, ent.addr, ent.masklen);
2990 }
2991 break;
2992
2993 case IP_FW_TABLE_FLUSH:
2994 {
2995 u_int16_t tbl;
2996
2997 error = sooptcopyin(sopt, &tbl,
2998 sizeof(tbl), sizeof(tbl));
2999 if (error)
3000 break;
3001 error = flush_table(tbl);
3002 }
3003 break;
3004
3005 case IP_FW_TABLE_GETSIZE:
3006 {
3007 u_int32_t tbl, cnt;
3008
3009 if ((error = sooptcopyin(sopt, &tbl, sizeof(tbl),
3010 sizeof(tbl))))
3011 break;
3012 if ((error = count_table(tbl, &cnt)))
3013 break;
3014 error = sooptcopyout(sopt, &cnt, sizeof(cnt));
3015 }
3016 break;
3017
3018 case IP_FW_TABLE_LIST:
3019 {
3020 ipfw_table *tbl;
3021
3022 if (sopt->sopt_valsize < sizeof(*tbl)) {
3023 error = EINVAL;
3024 break;
3025 }
3026 size = sopt->sopt_valsize;
3027 tbl = malloc(size, M_TEMP, M_WAITOK);
3028 if (tbl == NULL) {
3029 error = ENOMEM;
3030 break;
3031 }
3032 error = sooptcopyin(sopt, tbl, size, sizeof(*tbl));
3033 if (error) {
3034 free(tbl, M_TEMP);
3035 break;
3036 }
3037 tbl->size = (size - sizeof(*tbl)) /
3038 sizeof(ipfw_table_entry);
3039 error = dump_table(tbl);
3040 if (error) {
3041 free(tbl, M_TEMP);
3042 break;
3043 }
3044 error = sooptcopyout(sopt, tbl, size);
3045 free(tbl, M_TEMP);
3046 }
3047 break;
3048
3049 default:
3050 printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name);
3051 error = EINVAL;
3052 }
3053
3054 return (error);
3055 }
3056
3057 /**
3058 * dummynet needs a reference to the default rule, because rules can be
3059 * deleted while packets hold a reference to them. When this happens,
3060 * dummynet changes the reference to the default rule (it could well be a
3061 * NULL pointer, but this way we do not need to check for the special
3062 * case, plus here he have info on the default behaviour).
3063 */
3064 struct ip_fw *ip_fw_default_rule;
3065
3066 /*
3067 * This procedure is only used to handle keepalives. It is invoked
3068 * every dyn_keepalive_period
3069 */
3070 static void
3071 ipfw_tick(void * __unused unused)
3072 {
3073 int i;
3074 int s;
3075 ipfw_dyn_rule *q;
3076
3077 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
3078 goto done;
3079
3080 s = splimp();
3081 for (i = 0 ; i < curr_dyn_buckets ; i++) {
3082 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
3083 if (q->dyn_type == O_LIMIT_PARENT)
3084 continue;
3085 if (q->id.proto != IPPROTO_TCP)
3086 continue;
3087 if ( (q->state & BOTH_SYN) != BOTH_SYN)
3088 continue;
3089 if (TIME_LEQ( time_second+dyn_keepalive_interval,
3090 q->expire))
3091 continue; /* too early */
3092 if (TIME_LEQ(q->expire, time_second))
3093 continue; /* too late, rule expired */
3094
3095 send_pkt(&(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN);
3096 send_pkt(&(q->id), q->ack_fwd - 1, q->ack_rev, 0);
3097 }
3098 }
3099 splx(s);
3100 done:
3101 ipfw_timeout_h = timeout(ipfw_tick, NULL, dyn_keepalive_period*hz);
3102 }
3103
3104 static void
3105 ipfw_init(void)
3106 {
3107 struct ip_fw default_rule;
3108
3109 ip_fw_chk_ptr = ipfw_chk;
3110 ip_fw_ctl_ptr = ipfw_ctl;
3111 layer3_chain = NULL;
3112
3113 bzero(&default_rule, sizeof default_rule);
3114
3115 default_rule.act_ofs = 0;
3116 default_rule.rulenum = IPFW_DEFAULT_RULE;
3117 default_rule.cmd_len = 1;
3118 default_rule.set = RESVD_SET;
3119
3120 default_rule.cmd[0].len = 1;
3121 default_rule.cmd[0].opcode =
3122 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
3123 1 ? O_ACCEPT :
3124 #endif
3125 O_DENY;
3126
3127 add_rule(&layer3_chain, &default_rule);
3128
3129 ip_fw_default_rule = layer3_chain;
3130 printf("ipfw2 initialized, divert %s, "
3131 "rule-based forwarding enabled, default to %s, logging ",
3132 #ifdef IPDIVERT
3133 "enabled",
3134 #else
3135 "disabled",
3136 #endif
3137 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny");
3138
3139 #ifdef IPFIREWALL_VERBOSE
3140 fw_verbose = 1;
3141 #endif
3142 #ifdef IPFIREWALL_VERBOSE_LIMIT
3143 verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
3144 #endif
3145 if (fw_verbose == 0)
3146 printf("disabled\n");
3147 else if (verbose_limit == 0)
3148 printf("unlimited\n");
3149 else
3150 printf("limited to %d packets/entry by default\n",
3151 verbose_limit);
3152 bzero(&ipfw_timeout_h, sizeof(struct callout_handle));
3153 ipfw_timeout_h = timeout(ipfw_tick, NULL, hz);
3154 }
3155
3156 static int
3157 ipfw_modevent(module_t mod, int type, void *unused)
3158 {
3159 int s;
3160 int err = 0;
3161
3162 switch (type) {
3163 case MOD_LOAD:
3164 s = splimp();
3165 if (IPFW_LOADED) {
3166 splx(s);
3167 printf("IP firewall already loaded\n");
3168 err = EEXIST;
3169 } else {
3170 ipfw_init();
3171 splx(s);
3172 }
3173 break;
3174
3175 case MOD_UNLOAD:
3176 #if !defined(KLD_MODULE)
3177 printf("ipfw statically compiled, cannot unload\n");
3178 err = EBUSY;
3179 #else
3180 s = splimp();
3181 untimeout(ipfw_tick, NULL, ipfw_timeout_h);
3182 ip_fw_chk_ptr = NULL;
3183 ip_fw_ctl_ptr = NULL;
3184 free_chain(&layer3_chain, 1 /* kill default rule */);
3185 flush_tables();
3186 splx(s);
3187 printf("IP firewall unloaded\n");
3188 #endif
3189 break;
3190 default:
3191 break;
3192 }
3193 return err;
3194 }
3195
3196 static moduledata_t ipfwmod = {
3197 "ipfw",
3198 ipfw_modevent,
3199 0
3200 };
3201 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PSEUDO, SI_ORDER_ANY);
3202 MODULE_VERSION(ipfw, 1);
3203
3204 /* Must be run after route_init(). */
3205 SYSINIT(ipfw, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, init_tables, 0)
3206
3207 #endif /* IPFW2 */
Cache object: 8ec212ce3e2374318d79bdf15fb6f58f
|