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: releng/5.2/sys/netinet/ip_fw2.c 123763 2003-12-23 12:25:56Z maxim $
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 #define IPFW2 1
47 #if IPFW2
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/kernel.h>
53 #include <sys/proc.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/sysctl.h>
57 #include <sys/syslog.h>
58 #include <sys/ucred.h>
59 #include <net/if.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 * This is used to avoid that a firewall-generated packet
87 * loops forever through the firewall. Note that it must
88 * be a flag that is unused by other protocols that might
89 * be called from ip_output (e.g. IPsec) and it must be
90 * listed in M_COPYFLAGS in mbuf.h so that if the mbuf chain
91 * is altered on the way through ip_output it is not lost.
92 * It might be better to add an m_tag since the this happens
93 * infrequently.
94 */
95 #define M_SKIP_FIREWALL M_PROTO6
96
97 /*
98 * set_disable contains one bit per set value (0..31).
99 * If the bit is set, all rules with the corresponding set
100 * are disabled. Set RESVD_SET(31) is reserved for the default rule
101 * and rules that are not deleted by the flush command,
102 * and CANNOT be disabled.
103 * Rules in set RESVD_SET can only be deleted explicitly.
104 */
105 static u_int32_t set_disable;
106
107 static int fw_verbose;
108 static int verbose_limit;
109
110 static struct callout ipfw_timeout;
111 #define IPFW_DEFAULT_RULE 65535
112
113 struct ip_fw_chain {
114 struct ip_fw *rules; /* list of rules */
115 struct ip_fw *reap; /* list of rules to reap */
116 struct mtx mtx; /* lock guarding rule list */
117 };
118 #define IPFW_LOCK_INIT(_chain) \
119 mtx_init(&(_chain)->mtx, "IPFW static rules", NULL, \
120 MTX_DEF | MTX_RECURSE)
121 #define IPFW_LOCK_DESTROY(_chain) mtx_destroy(&(_chain)->mtx)
122 #define IPFW_LOCK(_chain) mtx_lock(&(_chain)->mtx)
123 #define IPFW_UNLOCK(_chain) mtx_unlock(&(_chain)->mtx)
124 #define IPFW_LOCK_ASSERT(_chain) mtx_assert(&(_chain)->mtx, MA_OWNED)
125
126 /*
127 * list of rules for layer 3
128 */
129 static struct ip_fw_chain layer3_chain;
130
131 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
132
133 static int fw_debug = 1;
134 static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
135
136 #ifdef SYSCTL_NODE
137 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
138 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable,
139 CTLFLAG_RW | CTLFLAG_SECURE3,
140 &fw_enable, 0, "Enable ipfw");
141 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
142 &autoinc_step, 0, "Rule number autincrement step");
143 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, one_pass,
144 CTLFLAG_RW | CTLFLAG_SECURE3,
145 &fw_one_pass, 0,
146 "Only do a single pass through ipfw when using dummynet(4)");
147 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
148 &fw_debug, 0, "Enable printing of debug ip_fw statements");
149 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose,
150 CTLFLAG_RW | CTLFLAG_SECURE3,
151 &fw_verbose, 0, "Log matches to ipfw rules");
152 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
153 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
154
155 /*
156 * Description of dynamic rules.
157 *
158 * Dynamic rules are stored in lists accessed through a hash table
159 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
160 * be modified through the sysctl variable dyn_buckets which is
161 * updated when the table becomes empty.
162 *
163 * XXX currently there is only one list, ipfw_dyn.
164 *
165 * When a packet is received, its address fields are first masked
166 * with the mask defined for the rule, then hashed, then matched
167 * against the entries in the corresponding list.
168 * Dynamic rules can be used for different purposes:
169 * + stateful rules;
170 * + enforcing limits on the number of sessions;
171 * + in-kernel NAT (not implemented yet)
172 *
173 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
174 * measured in seconds and depending on the flags.
175 *
176 * The total number of dynamic rules is stored in dyn_count.
177 * The max number of dynamic rules is dyn_max. When we reach
178 * the maximum number of rules we do not create anymore. This is
179 * done to avoid consuming too much memory, but also too much
180 * time when searching on each packet (ideally, we should try instead
181 * to put a limit on the length of the list on each bucket...).
182 *
183 * Each dynamic rule holds a pointer to the parent ipfw rule so
184 * we know what action to perform. Dynamic rules are removed when
185 * the parent rule is deleted. XXX we should make them survive.
186 *
187 * There are some limitations with dynamic rules -- we do not
188 * obey the 'randomized match', and we do not do multiple
189 * passes through the firewall. XXX check the latter!!!
190 */
191 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
192 static u_int32_t dyn_buckets = 256; /* must be power of 2 */
193 static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */
194
195 static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */
196 #define IPFW_DYN_LOCK_INIT() \
197 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF)
198 #define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx)
199 #define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx)
200 #define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx)
201 #define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED)
202
203 /*
204 * Timeouts for various events in handing dynamic rules.
205 */
206 static u_int32_t dyn_ack_lifetime = 300;
207 static u_int32_t dyn_syn_lifetime = 20;
208 static u_int32_t dyn_fin_lifetime = 1;
209 static u_int32_t dyn_rst_lifetime = 1;
210 static u_int32_t dyn_udp_lifetime = 10;
211 static u_int32_t dyn_short_lifetime = 5;
212
213 /*
214 * Keepalives are sent if dyn_keepalive is set. They are sent every
215 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
216 * seconds of lifetime of a rule.
217 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
218 * than dyn_keepalive_period.
219 */
220
221 static u_int32_t dyn_keepalive_interval = 20;
222 static u_int32_t dyn_keepalive_period = 5;
223 static u_int32_t dyn_keepalive = 1; /* do send keepalives */
224
225 static u_int32_t static_count; /* # of static rules */
226 static u_int32_t static_len; /* size in bytes of static rules */
227 static u_int32_t dyn_count; /* # of dynamic rules */
228 static u_int32_t dyn_max = 4096; /* max # of dynamic rules */
229
230 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
231 &dyn_buckets, 0, "Number of dyn. buckets");
232 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
233 &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
234 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
235 &dyn_count, 0, "Number of dyn. rules");
236 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
237 &dyn_max, 0, "Max number of dyn. rules");
238 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
239 &static_count, 0, "Number of static rules");
240 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
241 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
242 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
243 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
244 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
245 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
246 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
247 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
248 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
249 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
250 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
251 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
252 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
253 &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
254
255 #endif /* SYSCTL_NODE */
256
257
258 static ip_fw_chk_t ipfw_chk;
259
260 ip_dn_ruledel_t *ip_dn_ruledel_ptr = NULL; /* hook into dummynet */
261
262 /*
263 * This macro maps an ip pointer into a layer3 header pointer of type T
264 */
265 #define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
266
267 static __inline int
268 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
269 {
270 int type = L3HDR(struct icmp,ip)->icmp_type;
271
272 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
273 }
274
275 #define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
276 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
277
278 static int
279 is_icmp_query(struct ip *ip)
280 {
281 int type = L3HDR(struct icmp, ip)->icmp_type;
282 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
283 }
284 #undef TT
285
286 /*
287 * The following checks use two arrays of 8 or 16 bits to store the
288 * bits that we want set or clear, respectively. They are in the
289 * low and high half of cmd->arg1 or cmd->d[0].
290 *
291 * We scan options and store the bits we find set. We succeed if
292 *
293 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
294 *
295 * The code is sometimes optimized not to store additional variables.
296 */
297
298 static int
299 flags_match(ipfw_insn *cmd, u_int8_t bits)
300 {
301 u_char want_clear;
302 bits = ~bits;
303
304 if ( ((cmd->arg1 & 0xff) & bits) != 0)
305 return 0; /* some bits we want set were clear */
306 want_clear = (cmd->arg1 >> 8) & 0xff;
307 if ( (want_clear & bits) != want_clear)
308 return 0; /* some bits we want clear were set */
309 return 1;
310 }
311
312 static int
313 ipopts_match(struct ip *ip, ipfw_insn *cmd)
314 {
315 int optlen, bits = 0;
316 u_char *cp = (u_char *)(ip + 1);
317 int x = (ip->ip_hl << 2) - sizeof (struct ip);
318
319 for (; x > 0; x -= optlen, cp += optlen) {
320 int opt = cp[IPOPT_OPTVAL];
321
322 if (opt == IPOPT_EOL)
323 break;
324 if (opt == IPOPT_NOP)
325 optlen = 1;
326 else {
327 optlen = cp[IPOPT_OLEN];
328 if (optlen <= 0 || optlen > x)
329 return 0; /* invalid or truncated */
330 }
331 switch (opt) {
332
333 default:
334 break;
335
336 case IPOPT_LSRR:
337 bits |= IP_FW_IPOPT_LSRR;
338 break;
339
340 case IPOPT_SSRR:
341 bits |= IP_FW_IPOPT_SSRR;
342 break;
343
344 case IPOPT_RR:
345 bits |= IP_FW_IPOPT_RR;
346 break;
347
348 case IPOPT_TS:
349 bits |= IP_FW_IPOPT_TS;
350 break;
351 }
352 }
353 return (flags_match(cmd, bits));
354 }
355
356 static int
357 tcpopts_match(struct ip *ip, ipfw_insn *cmd)
358 {
359 int optlen, bits = 0;
360 struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
361 u_char *cp = (u_char *)(tcp + 1);
362 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
363
364 for (; x > 0; x -= optlen, cp += optlen) {
365 int opt = cp[0];
366 if (opt == TCPOPT_EOL)
367 break;
368 if (opt == TCPOPT_NOP)
369 optlen = 1;
370 else {
371 optlen = cp[1];
372 if (optlen <= 0)
373 break;
374 }
375
376 switch (opt) {
377
378 default:
379 break;
380
381 case TCPOPT_MAXSEG:
382 bits |= IP_FW_TCPOPT_MSS;
383 break;
384
385 case TCPOPT_WINDOW:
386 bits |= IP_FW_TCPOPT_WINDOW;
387 break;
388
389 case TCPOPT_SACK_PERMITTED:
390 case TCPOPT_SACK:
391 bits |= IP_FW_TCPOPT_SACK;
392 break;
393
394 case TCPOPT_TIMESTAMP:
395 bits |= IP_FW_TCPOPT_TS;
396 break;
397
398 case TCPOPT_CC:
399 case TCPOPT_CCNEW:
400 case TCPOPT_CCECHO:
401 bits |= IP_FW_TCPOPT_CC;
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 /* Check by name or by IP address */
414 if (cmd->name[0] != '\0') { /* match by name */
415 /* Check name */
416 if (cmd->p.glob) {
417 if (fnmatch(cmd->name, ifp->if_xname, 0) == 0)
418 return(1);
419 } else {
420 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
421 return(1);
422 }
423 } else {
424 struct ifaddr *ia;
425
426 /* XXX lock? */
427 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
428 if (ia->ifa_addr == NULL)
429 continue;
430 if (ia->ifa_addr->sa_family != AF_INET)
431 continue;
432 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
433 (ia->ifa_addr))->sin_addr.s_addr)
434 return(1); /* match */
435 }
436 }
437 return(0); /* no match, fail ... */
438 }
439
440 /*
441 * The 'verrevpath' option checks that the interface that an IP packet
442 * arrives on is the same interface that traffic destined for the
443 * packet's source address would be routed out of. This is a measure
444 * to block forged packets. This is also commonly known as "anti-spoofing"
445 * or Unicast Reverse Path Forwarding (Unicast RFP) in Cisco-ese. The
446 * name of the knob is purposely reminisent of the Cisco IOS command,
447 *
448 * ip verify unicast reverse-path
449 *
450 * which implements the same functionality. But note that syntax is
451 * misleading. The check may be performed on all IP packets whether unicast,
452 * multicast, or broadcast.
453 */
454 static int
455 verify_rev_path(struct in_addr src, struct ifnet *ifp)
456 {
457 struct route ro;
458 struct sockaddr_in *dst;
459
460 bzero(&ro, sizeof(ro));
461
462 dst = (struct sockaddr_in *)&(ro.ro_dst);
463 dst->sin_family = AF_INET;
464 dst->sin_len = sizeof(*dst);
465 dst->sin_addr = src;
466 rtalloc_ign(&ro, RTF_CLONING);
467
468 if (ro.ro_rt == NULL)
469 return 0;
470 if ((ifp == NULL) || (ro.ro_rt->rt_ifp->if_index != ifp->if_index)) {
471 RTFREE(ro.ro_rt);
472 return 0;
473 }
474 RTFREE(ro.ro_rt);
475 return 1;
476 }
477
478
479 static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */
480
481 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
482 #define SNP(buf) buf, sizeof(buf)
483
484 /*
485 * We enter here when we have a rule with O_LOG.
486 * XXX this function alone takes about 2Kbytes of code!
487 */
488 static void
489 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
490 struct mbuf *m, struct ifnet *oif)
491 {
492 char *action;
493 int limit_reached = 0;
494 char action2[40], proto[48], fragment[28];
495
496 fragment[0] = '\0';
497 proto[0] = '\0';
498
499 if (f == NULL) { /* bogus pkt */
500 if (verbose_limit != 0 && norule_counter >= verbose_limit)
501 return;
502 norule_counter++;
503 if (norule_counter == verbose_limit)
504 limit_reached = verbose_limit;
505 action = "Refuse";
506 } else { /* O_LOG is the first action, find the real one */
507 ipfw_insn *cmd = ACTION_PTR(f);
508 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
509
510 if (l->max_log != 0 && l->log_left == 0)
511 return;
512 l->log_left--;
513 if (l->log_left == 0)
514 limit_reached = l->max_log;
515 cmd += F_LEN(cmd); /* point to first action */
516 if (cmd->opcode == O_PROB)
517 cmd += F_LEN(cmd);
518
519 action = action2;
520 switch (cmd->opcode) {
521 case O_DENY:
522 action = "Deny";
523 break;
524
525 case O_REJECT:
526 if (cmd->arg1==ICMP_REJECT_RST)
527 action = "Reset";
528 else if (cmd->arg1==ICMP_UNREACH_HOST)
529 action = "Reject";
530 else
531 snprintf(SNPARGS(action2, 0), "Unreach %d",
532 cmd->arg1);
533 break;
534
535 case O_ACCEPT:
536 action = "Accept";
537 break;
538 case O_COUNT:
539 action = "Count";
540 break;
541 case O_DIVERT:
542 snprintf(SNPARGS(action2, 0), "Divert %d",
543 cmd->arg1);
544 break;
545 case O_TEE:
546 snprintf(SNPARGS(action2, 0), "Tee %d",
547 cmd->arg1);
548 break;
549 case O_SKIPTO:
550 snprintf(SNPARGS(action2, 0), "SkipTo %d",
551 cmd->arg1);
552 break;
553 case O_PIPE:
554 snprintf(SNPARGS(action2, 0), "Pipe %d",
555 cmd->arg1);
556 break;
557 case O_QUEUE:
558 snprintf(SNPARGS(action2, 0), "Queue %d",
559 cmd->arg1);
560 break;
561 case O_FORWARD_IP: {
562 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
563 int len;
564
565 len = snprintf(SNPARGS(action2, 0), "Forward to %s",
566 inet_ntoa(sa->sa.sin_addr));
567 if (sa->sa.sin_port)
568 snprintf(SNPARGS(action2, len), ":%d",
569 sa->sa.sin_port);
570 }
571 break;
572 default:
573 action = "UNKNOWN";
574 break;
575 }
576 }
577
578 if (hlen == 0) { /* non-ip */
579 snprintf(SNPARGS(proto, 0), "MAC");
580 } else {
581 struct ip *ip = mtod(m, struct ip *);
582 /* these three are all aliases to the same thing */
583 struct icmp *const icmp = L3HDR(struct icmp, ip);
584 struct tcphdr *const tcp = (struct tcphdr *)icmp;
585 struct udphdr *const udp = (struct udphdr *)icmp;
586
587 int ip_off, offset, ip_len;
588
589 int len;
590
591 if (eh != NULL) { /* layer 2 packets are as on the wire */
592 ip_off = ntohs(ip->ip_off);
593 ip_len = ntohs(ip->ip_len);
594 } else {
595 ip_off = ip->ip_off;
596 ip_len = ip->ip_len;
597 }
598 offset = ip_off & IP_OFFMASK;
599 switch (ip->ip_p) {
600 case IPPROTO_TCP:
601 len = snprintf(SNPARGS(proto, 0), "TCP %s",
602 inet_ntoa(ip->ip_src));
603 if (offset == 0)
604 snprintf(SNPARGS(proto, len), ":%d %s:%d",
605 ntohs(tcp->th_sport),
606 inet_ntoa(ip->ip_dst),
607 ntohs(tcp->th_dport));
608 else
609 snprintf(SNPARGS(proto, len), " %s",
610 inet_ntoa(ip->ip_dst));
611 break;
612
613 case IPPROTO_UDP:
614 len = snprintf(SNPARGS(proto, 0), "UDP %s",
615 inet_ntoa(ip->ip_src));
616 if (offset == 0)
617 snprintf(SNPARGS(proto, len), ":%d %s:%d",
618 ntohs(udp->uh_sport),
619 inet_ntoa(ip->ip_dst),
620 ntohs(udp->uh_dport));
621 else
622 snprintf(SNPARGS(proto, len), " %s",
623 inet_ntoa(ip->ip_dst));
624 break;
625
626 case IPPROTO_ICMP:
627 if (offset == 0)
628 len = snprintf(SNPARGS(proto, 0),
629 "ICMP:%u.%u ",
630 icmp->icmp_type, icmp->icmp_code);
631 else
632 len = snprintf(SNPARGS(proto, 0), "ICMP ");
633 len += snprintf(SNPARGS(proto, len), "%s",
634 inet_ntoa(ip->ip_src));
635 snprintf(SNPARGS(proto, len), " %s",
636 inet_ntoa(ip->ip_dst));
637 break;
638
639 default:
640 len = snprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
641 inet_ntoa(ip->ip_src));
642 snprintf(SNPARGS(proto, len), " %s",
643 inet_ntoa(ip->ip_dst));
644 break;
645 }
646
647 if (ip_off & (IP_MF | IP_OFFMASK))
648 snprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
649 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
650 offset << 3,
651 (ip_off & IP_MF) ? "+" : "");
652 }
653 if (oif || m->m_pkthdr.rcvif)
654 log(LOG_SECURITY | LOG_INFO,
655 "ipfw: %d %s %s %s via %s%s\n",
656 f ? f->rulenum : -1,
657 action, proto, oif ? "out" : "in",
658 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
659 fragment);
660 else
661 log(LOG_SECURITY | LOG_INFO,
662 "ipfw: %d %s %s [no if info]%s\n",
663 f ? f->rulenum : -1,
664 action, proto, fragment);
665 if (limit_reached)
666 log(LOG_SECURITY | LOG_NOTICE,
667 "ipfw: limit %d reached on entry %d\n",
668 limit_reached, f ? f->rulenum : -1);
669 }
670
671 /*
672 * IMPORTANT: the hash function for dynamic rules must be commutative
673 * in source and destination (ip,port), because rules are bidirectional
674 * and we want to find both in the same bucket.
675 */
676 static __inline int
677 hash_packet(struct ipfw_flow_id *id)
678 {
679 u_int32_t i;
680
681 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
682 i &= (curr_dyn_buckets - 1);
683 return i;
684 }
685
686 /**
687 * unlink a dynamic rule from a chain. prev is a pointer to
688 * the previous one, q is a pointer to the rule to delete,
689 * head is a pointer to the head of the queue.
690 * Modifies q and potentially also head.
691 */
692 #define UNLINK_DYN_RULE(prev, head, q) { \
693 ipfw_dyn_rule *old_q = q; \
694 \
695 /* remove a refcount to the parent */ \
696 if (q->dyn_type == O_LIMIT) \
697 q->parent->count--; \
698 DEB(printf("ipfw: unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",\
699 (q->id.src_ip), (q->id.src_port), \
700 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \
701 if (prev != NULL) \
702 prev->next = q = q->next; \
703 else \
704 head = q = q->next; \
705 dyn_count--; \
706 free(old_q, M_IPFW); }
707
708 #define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
709
710 /**
711 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
712 *
713 * If keep_me == NULL, rules are deleted even if not expired,
714 * otherwise only expired rules are removed.
715 *
716 * The value of the second parameter is also used to point to identify
717 * a rule we absolutely do not want to remove (e.g. because we are
718 * holding a reference to it -- this is the case with O_LIMIT_PARENT
719 * rules). The pointer is only used for comparison, so any non-null
720 * value will do.
721 */
722 static void
723 remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
724 {
725 static u_int32_t last_remove = 0;
726
727 #define FORCE (keep_me == NULL)
728
729 ipfw_dyn_rule *prev, *q;
730 int i, pass = 0, max_pass = 0;
731
732 IPFW_DYN_LOCK_ASSERT();
733
734 if (ipfw_dyn_v == NULL || dyn_count == 0)
735 return;
736 /* do not expire more than once per second, it is useless */
737 if (!FORCE && last_remove == time_second)
738 return;
739 last_remove = time_second;
740
741 /*
742 * because O_LIMIT refer to parent rules, during the first pass only
743 * remove child and mark any pending LIMIT_PARENT, and remove
744 * them in a second pass.
745 */
746 next_pass:
747 for (i = 0 ; i < curr_dyn_buckets ; i++) {
748 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
749 /*
750 * Logic can become complex here, so we split tests.
751 */
752 if (q == keep_me)
753 goto next;
754 if (rule != NULL && rule != q->rule)
755 goto next; /* not the one we are looking for */
756 if (q->dyn_type == O_LIMIT_PARENT) {
757 /*
758 * handle parent in the second pass,
759 * record we need one.
760 */
761 max_pass = 1;
762 if (pass == 0)
763 goto next;
764 if (FORCE && q->count != 0 ) {
765 /* XXX should not happen! */
766 printf("ipfw: OUCH! cannot remove rule,"
767 " count %d\n", q->count);
768 }
769 } else {
770 if (!FORCE &&
771 !TIME_LEQ( q->expire, time_second ))
772 goto next;
773 }
774 if (q->dyn_type != O_LIMIT_PARENT || !q->count) {
775 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
776 continue;
777 }
778 next:
779 prev=q;
780 q=q->next;
781 }
782 }
783 if (pass++ < max_pass)
784 goto next_pass;
785 }
786
787
788 /**
789 * lookup a dynamic rule.
790 */
791 static ipfw_dyn_rule *
792 lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction,
793 struct tcphdr *tcp)
794 {
795 /*
796 * stateful ipfw extensions.
797 * Lookup into dynamic session queue
798 */
799 #define MATCH_REVERSE 0
800 #define MATCH_FORWARD 1
801 #define MATCH_NONE 2
802 #define MATCH_UNKNOWN 3
803 int i, dir = MATCH_NONE;
804 ipfw_dyn_rule *prev, *q=NULL;
805
806 IPFW_DYN_LOCK_ASSERT();
807
808 if (ipfw_dyn_v == NULL)
809 goto done; /* not found */
810 i = hash_packet( pkt );
811 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
812 if (q->dyn_type == O_LIMIT_PARENT && q->count)
813 goto next;
814 if (TIME_LEQ( q->expire, time_second)) { /* expire entry */
815 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
816 continue;
817 }
818 if (pkt->proto == q->id.proto &&
819 q->dyn_type != O_LIMIT_PARENT) {
820 if (pkt->src_ip == q->id.src_ip &&
821 pkt->dst_ip == q->id.dst_ip &&
822 pkt->src_port == q->id.src_port &&
823 pkt->dst_port == q->id.dst_port ) {
824 dir = MATCH_FORWARD;
825 break;
826 }
827 if (pkt->src_ip == q->id.dst_ip &&
828 pkt->dst_ip == q->id.src_ip &&
829 pkt->src_port == q->id.dst_port &&
830 pkt->dst_port == q->id.src_port ) {
831 dir = MATCH_REVERSE;
832 break;
833 }
834 }
835 next:
836 prev = q;
837 q = q->next;
838 }
839 if (q == NULL)
840 goto done; /* q = NULL, not found */
841
842 if ( prev != NULL) { /* found and not in front */
843 prev->next = q->next;
844 q->next = ipfw_dyn_v[i];
845 ipfw_dyn_v[i] = q;
846 }
847 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
848 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
849
850 #define BOTH_SYN (TH_SYN | (TH_SYN << 8))
851 #define BOTH_FIN (TH_FIN | (TH_FIN << 8))
852 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
853 switch (q->state) {
854 case TH_SYN: /* opening */
855 q->expire = time_second + dyn_syn_lifetime;
856 break;
857
858 case BOTH_SYN: /* move to established */
859 case BOTH_SYN | TH_FIN : /* one side tries to close */
860 case BOTH_SYN | (TH_FIN << 8) :
861 if (tcp) {
862 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
863 u_int32_t ack = ntohl(tcp->th_ack);
864 if (dir == MATCH_FORWARD) {
865 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
866 q->ack_fwd = ack;
867 else { /* ignore out-of-sequence */
868 break;
869 }
870 } else {
871 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
872 q->ack_rev = ack;
873 else { /* ignore out-of-sequence */
874 break;
875 }
876 }
877 }
878 q->expire = time_second + dyn_ack_lifetime;
879 break;
880
881 case BOTH_SYN | BOTH_FIN: /* both sides closed */
882 if (dyn_fin_lifetime >= dyn_keepalive_period)
883 dyn_fin_lifetime = dyn_keepalive_period - 1;
884 q->expire = time_second + dyn_fin_lifetime;
885 break;
886
887 default:
888 #if 0
889 /*
890 * reset or some invalid combination, but can also
891 * occur if we use keep-state the wrong way.
892 */
893 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
894 printf("invalid state: 0x%x\n", q->state);
895 #endif
896 if (dyn_rst_lifetime >= dyn_keepalive_period)
897 dyn_rst_lifetime = dyn_keepalive_period - 1;
898 q->expire = time_second + dyn_rst_lifetime;
899 break;
900 }
901 } else if (pkt->proto == IPPROTO_UDP) {
902 q->expire = time_second + dyn_udp_lifetime;
903 } else {
904 /* other protocols */
905 q->expire = time_second + dyn_short_lifetime;
906 }
907 done:
908 if (match_direction)
909 *match_direction = dir;
910 return q;
911 }
912
913 static ipfw_dyn_rule *
914 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
915 struct tcphdr *tcp)
916 {
917 ipfw_dyn_rule *q;
918
919 IPFW_DYN_LOCK();
920 q = lookup_dyn_rule_locked(pkt, match_direction, tcp);
921 if (q == NULL)
922 IPFW_DYN_UNLOCK();
923 /* NB: return table locked when q is not NULL */
924 return q;
925 }
926
927 static void
928 realloc_dynamic_table(void)
929 {
930 IPFW_DYN_LOCK_ASSERT();
931
932 /*
933 * Try reallocation, make sure we have a power of 2 and do
934 * not allow more than 64k entries. In case of overflow,
935 * default to 1024.
936 */
937
938 if (dyn_buckets > 65536)
939 dyn_buckets = 1024;
940 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
941 dyn_buckets = curr_dyn_buckets; /* reset */
942 return;
943 }
944 curr_dyn_buckets = dyn_buckets;
945 if (ipfw_dyn_v != NULL)
946 free(ipfw_dyn_v, M_IPFW);
947 for (;;) {
948 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
949 M_IPFW, M_NOWAIT | M_ZERO);
950 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
951 break;
952 curr_dyn_buckets /= 2;
953 }
954 }
955
956 /**
957 * Install state of type 'type' for a dynamic session.
958 * The hash table contains two type of rules:
959 * - regular rules (O_KEEP_STATE)
960 * - rules for sessions with limited number of sess per user
961 * (O_LIMIT). When they are created, the parent is
962 * increased by 1, and decreased on delete. In this case,
963 * the third parameter is the parent rule and not the chain.
964 * - "parent" rules for the above (O_LIMIT_PARENT).
965 */
966 static ipfw_dyn_rule *
967 add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
968 {
969 ipfw_dyn_rule *r;
970 int i;
971
972 IPFW_DYN_LOCK_ASSERT();
973
974 if (ipfw_dyn_v == NULL ||
975 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
976 realloc_dynamic_table();
977 if (ipfw_dyn_v == NULL)
978 return NULL; /* failed ! */
979 }
980 i = hash_packet(id);
981
982 r = malloc(sizeof *r, M_IPFW, M_NOWAIT | M_ZERO);
983 if (r == NULL) {
984 printf ("ipfw: sorry cannot allocate state\n");
985 return NULL;
986 }
987
988 /* increase refcount on parent, and set pointer */
989 if (dyn_type == O_LIMIT) {
990 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
991 if ( parent->dyn_type != O_LIMIT_PARENT)
992 panic("invalid parent");
993 parent->count++;
994 r->parent = parent;
995 rule = parent->rule;
996 }
997
998 r->id = *id;
999 r->expire = time_second + dyn_syn_lifetime;
1000 r->rule = rule;
1001 r->dyn_type = dyn_type;
1002 r->pcnt = r->bcnt = 0;
1003 r->count = 0;
1004
1005 r->bucket = i;
1006 r->next = ipfw_dyn_v[i];
1007 ipfw_dyn_v[i] = r;
1008 dyn_count++;
1009 DEB(printf("ipfw: add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1010 dyn_type,
1011 (r->id.src_ip), (r->id.src_port),
1012 (r->id.dst_ip), (r->id.dst_port),
1013 dyn_count ); )
1014 return r;
1015 }
1016
1017 /**
1018 * lookup dynamic parent rule using pkt and rule as search keys.
1019 * If the lookup fails, then install one.
1020 */
1021 static ipfw_dyn_rule *
1022 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1023 {
1024 ipfw_dyn_rule *q;
1025 int i;
1026
1027 IPFW_DYN_LOCK_ASSERT();
1028
1029 if (ipfw_dyn_v) {
1030 i = hash_packet( pkt );
1031 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
1032 if (q->dyn_type == O_LIMIT_PARENT &&
1033 rule== q->rule &&
1034 pkt->proto == q->id.proto &&
1035 pkt->src_ip == q->id.src_ip &&
1036 pkt->dst_ip == q->id.dst_ip &&
1037 pkt->src_port == q->id.src_port &&
1038 pkt->dst_port == q->id.dst_port) {
1039 q->expire = time_second + dyn_short_lifetime;
1040 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);)
1041 return q;
1042 }
1043 }
1044 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1045 }
1046
1047 /**
1048 * Install dynamic state for rule type cmd->o.opcode
1049 *
1050 * Returns 1 (failure) if state is not installed because of errors or because
1051 * session limitations are enforced.
1052 */
1053 static int
1054 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1055 struct ip_fw_args *args)
1056 {
1057 static int last_log;
1058
1059 ipfw_dyn_rule *q;
1060
1061 DEB(printf("ipfw: install state type %d 0x%08x %u -> 0x%08x %u\n",
1062 cmd->o.opcode,
1063 (args->f_id.src_ip), (args->f_id.src_port),
1064 (args->f_id.dst_ip), (args->f_id.dst_port) );)
1065
1066 IPFW_DYN_LOCK();
1067
1068 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
1069
1070 if (q != NULL) { /* should never occur */
1071 if (last_log != time_second) {
1072 last_log = time_second;
1073 printf("ipfw: install_state: entry already present, done\n");
1074 }
1075 IPFW_DYN_UNLOCK();
1076 return 0;
1077 }
1078
1079 if (dyn_count >= dyn_max)
1080 /*
1081 * Run out of slots, try to remove any expired rule.
1082 */
1083 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1084
1085 if (dyn_count >= dyn_max) {
1086 if (last_log != time_second) {
1087 last_log = time_second;
1088 printf("ipfw: install_state: Too many dynamic rules\n");
1089 }
1090 IPFW_DYN_UNLOCK();
1091 return 1; /* cannot install, notify caller */
1092 }
1093
1094 switch (cmd->o.opcode) {
1095 case O_KEEP_STATE: /* bidir rule */
1096 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1097 break;
1098
1099 case O_LIMIT: /* limit number of sessions */
1100 {
1101 u_int16_t limit_mask = cmd->limit_mask;
1102 struct ipfw_flow_id id;
1103 ipfw_dyn_rule *parent;
1104
1105 DEB(printf("ipfw: installing dyn-limit rule %d\n",
1106 cmd->conn_limit);)
1107
1108 id.dst_ip = id.src_ip = 0;
1109 id.dst_port = id.src_port = 0;
1110 id.proto = args->f_id.proto;
1111
1112 if (limit_mask & DYN_SRC_ADDR)
1113 id.src_ip = args->f_id.src_ip;
1114 if (limit_mask & DYN_DST_ADDR)
1115 id.dst_ip = args->f_id.dst_ip;
1116 if (limit_mask & DYN_SRC_PORT)
1117 id.src_port = args->f_id.src_port;
1118 if (limit_mask & DYN_DST_PORT)
1119 id.dst_port = args->f_id.dst_port;
1120 parent = lookup_dyn_parent(&id, rule);
1121 if (parent == NULL) {
1122 printf("ipfw: add parent failed\n");
1123 return 1;
1124 }
1125 if (parent->count >= cmd->conn_limit) {
1126 /*
1127 * See if we can remove some expired rule.
1128 */
1129 remove_dyn_rule(rule, parent);
1130 if (parent->count >= cmd->conn_limit) {
1131 if (fw_verbose && last_log != time_second) {
1132 last_log = time_second;
1133 log(LOG_SECURITY | LOG_DEBUG,
1134 "drop session, too many entries\n");
1135 }
1136 IPFW_DYN_UNLOCK();
1137 return 1;
1138 }
1139 }
1140 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1141 }
1142 break;
1143 default:
1144 printf("ipfw: unknown dynamic rule type %u\n", cmd->o.opcode);
1145 IPFW_DYN_UNLOCK();
1146 return 1;
1147 }
1148 lookup_dyn_rule_locked(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1149 IPFW_DYN_UNLOCK();
1150 return 0;
1151 }
1152
1153 /*
1154 * Transmit a TCP packet, containing either a RST or a keepalive.
1155 * When flags & TH_RST, we are sending a RST packet, because of a
1156 * "reset" action matched the packet.
1157 * Otherwise we are sending a keepalive, and flags & TH_
1158 */
1159 static void
1160 send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags)
1161 {
1162 struct mbuf *m;
1163 struct ip *ip;
1164 struct tcphdr *tcp;
1165
1166 MGETHDR(m, M_DONTWAIT, MT_HEADER);
1167 if (m == 0)
1168 return;
1169 m->m_pkthdr.rcvif = (struct ifnet *)0;
1170 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1171 m->m_data += max_linkhdr;
1172
1173 ip = mtod(m, struct ip *);
1174 bzero(ip, m->m_len);
1175 tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1176 ip->ip_p = IPPROTO_TCP;
1177 tcp->th_off = 5;
1178 /*
1179 * Assume we are sending a RST (or a keepalive in the reverse
1180 * direction), swap src and destination addresses and ports.
1181 */
1182 ip->ip_src.s_addr = htonl(id->dst_ip);
1183 ip->ip_dst.s_addr = htonl(id->src_ip);
1184 tcp->th_sport = htons(id->dst_port);
1185 tcp->th_dport = htons(id->src_port);
1186 if (flags & TH_RST) { /* we are sending a RST */
1187 if (flags & TH_ACK) {
1188 tcp->th_seq = htonl(ack);
1189 tcp->th_ack = htonl(0);
1190 tcp->th_flags = TH_RST;
1191 } else {
1192 if (flags & TH_SYN)
1193 seq++;
1194 tcp->th_seq = htonl(0);
1195 tcp->th_ack = htonl(seq);
1196 tcp->th_flags = TH_RST | TH_ACK;
1197 }
1198 } else {
1199 /*
1200 * We are sending a keepalive. flags & TH_SYN determines
1201 * the direction, forward if set, reverse if clear.
1202 * NOTE: seq and ack are always assumed to be correct
1203 * as set by the caller. This may be confusing...
1204 */
1205 if (flags & TH_SYN) {
1206 /*
1207 * we have to rewrite the correct addresses!
1208 */
1209 ip->ip_dst.s_addr = htonl(id->dst_ip);
1210 ip->ip_src.s_addr = htonl(id->src_ip);
1211 tcp->th_dport = htons(id->dst_port);
1212 tcp->th_sport = htons(id->src_port);
1213 }
1214 tcp->th_seq = htonl(seq);
1215 tcp->th_ack = htonl(ack);
1216 tcp->th_flags = TH_ACK;
1217 }
1218 /*
1219 * set ip_len to the payload size so we can compute
1220 * the tcp checksum on the pseudoheader
1221 * XXX check this, could save a couple of words ?
1222 */
1223 ip->ip_len = htons(sizeof(struct tcphdr));
1224 tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1225 /*
1226 * now fill fields left out earlier
1227 */
1228 ip->ip_ttl = ip_defttl;
1229 ip->ip_len = m->m_pkthdr.len;
1230 m->m_flags |= M_SKIP_FIREWALL;
1231 ip_output(m, NULL, NULL, 0, NULL, NULL);
1232 }
1233
1234 /*
1235 * sends a reject message, consuming the mbuf passed as an argument.
1236 */
1237 static void
1238 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1239 {
1240
1241 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1242 /* We need the IP header in host order for icmp_error(). */
1243 if (args->eh != NULL) {
1244 struct ip *ip = mtod(args->m, struct ip *);
1245 ip->ip_len = ntohs(ip->ip_len);
1246 ip->ip_off = ntohs(ip->ip_off);
1247 }
1248 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1249 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1250 struct tcphdr *const tcp =
1251 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1252 if ( (tcp->th_flags & TH_RST) == 0)
1253 send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1254 ntohl(tcp->th_ack),
1255 tcp->th_flags | TH_RST);
1256 m_freem(args->m);
1257 } else
1258 m_freem(args->m);
1259 args->m = NULL;
1260 }
1261
1262 /**
1263 *
1264 * Given an ip_fw *, lookup_next_rule will return a pointer
1265 * to the next rule, which can be either the jump
1266 * target (for skipto instructions) or the next one in the list (in
1267 * all other cases including a missing jump target).
1268 * The result is also written in the "next_rule" field of the rule.
1269 * Backward jumps are not allowed, so start looking from the next
1270 * rule...
1271 *
1272 * This never returns NULL -- in case we do not have an exact match,
1273 * the next rule is returned. When the ruleset is changed,
1274 * pointers are flushed so we are always correct.
1275 */
1276
1277 static struct ip_fw *
1278 lookup_next_rule(struct ip_fw *me)
1279 {
1280 struct ip_fw *rule = NULL;
1281 ipfw_insn *cmd;
1282
1283 /* look for action, in case it is a skipto */
1284 cmd = ACTION_PTR(me);
1285 if (cmd->opcode == O_LOG)
1286 cmd += F_LEN(cmd);
1287 if ( cmd->opcode == O_SKIPTO )
1288 for (rule = me->next; rule ; rule = rule->next)
1289 if (rule->rulenum >= cmd->arg1)
1290 break;
1291 if (rule == NULL) /* failure or not a skipto */
1292 rule = me->next;
1293 me->next_rule = rule;
1294 return rule;
1295 }
1296
1297 static int
1298 check_uidgid(ipfw_insn_u32 *insn,
1299 int proto, struct ifnet *oif,
1300 struct in_addr dst_ip, u_int16_t dst_port,
1301 struct in_addr src_ip, u_int16_t src_port)
1302 {
1303 struct inpcbinfo *pi;
1304 int wildcard;
1305 struct inpcb *pcb;
1306 int match;
1307
1308 if (proto == IPPROTO_TCP) {
1309 wildcard = 0;
1310 pi = &tcbinfo;
1311 } else if (proto == IPPROTO_UDP) {
1312 wildcard = 1;
1313 pi = &udbinfo;
1314 } else
1315 return 0;
1316
1317 match = 0;
1318
1319 INP_INFO_RLOCK(pi); /* XXX LOR with IPFW */
1320 pcb = (oif) ?
1321 in_pcblookup_hash(pi,
1322 dst_ip, htons(dst_port),
1323 src_ip, htons(src_port),
1324 wildcard, oif) :
1325 in_pcblookup_hash(pi,
1326 src_ip, htons(src_port),
1327 dst_ip, htons(dst_port),
1328 wildcard, NULL);
1329 if (pcb != NULL) {
1330 INP_LOCK(pcb);
1331 if (pcb->inp_socket != NULL) {
1332 #if __FreeBSD_version < 500034
1333 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1334 #endif
1335 if (insn->o.opcode == O_UID) {
1336 match = !socheckuid(pcb->inp_socket,
1337 (uid_t)insn->d[0]);
1338 } else {
1339 match = groupmember((uid_t)insn->d[0],
1340 pcb->inp_socket->so_cred);
1341 }
1342 }
1343 INP_UNLOCK(pcb);
1344 }
1345 INP_INFO_RUNLOCK(pi);
1346
1347 return match;
1348 }
1349
1350 /*
1351 * The main check routine for the firewall.
1352 *
1353 * All arguments are in args so we can modify them and return them
1354 * back to the caller.
1355 *
1356 * Parameters:
1357 *
1358 * args->m (in/out) The packet; we set to NULL when/if we nuke it.
1359 * Starts with the IP header.
1360 * args->eh (in) Mac header if present, or NULL for layer3 packet.
1361 * args->oif Outgoing interface, or NULL if packet is incoming.
1362 * The incoming interface is in the mbuf. (in)
1363 * args->divert_rule (in/out)
1364 * Skip up to the first rule past this rule number;
1365 * upon return, non-zero port number for divert or tee.
1366 *
1367 * args->rule Pointer to the last matching rule (in/out)
1368 * args->next_hop Socket we are forwarding to (out).
1369 * args->f_id Addresses grabbed from the packet (out)
1370 *
1371 * Return value:
1372 *
1373 * IP_FW_PORT_DENY_FLAG the packet must be dropped.
1374 * 0 The packet is to be accepted and routed normally OR
1375 * the packet was denied/rejected and has been dropped;
1376 * in the latter case, *m is equal to NULL upon return.
1377 * port Divert the packet to port, with these caveats:
1378 *
1379 * - If IP_FW_PORT_TEE_FLAG is set, tee the packet instead
1380 * of diverting it (ie, 'ipfw tee').
1381 *
1382 * - If IP_FW_PORT_DYNT_FLAG is set, interpret the lower
1383 * 16 bits as a dummynet pipe number instead of diverting
1384 */
1385
1386 static int
1387 ipfw_chk(struct ip_fw_args *args)
1388 {
1389 /*
1390 * Local variables hold state during the processing of a packet.
1391 *
1392 * IMPORTANT NOTE: to speed up the processing of rules, there
1393 * are some assumption on the values of the variables, which
1394 * are documented here. Should you change them, please check
1395 * the implementation of the various instructions to make sure
1396 * that they still work.
1397 *
1398 * args->eh The MAC header. It is non-null for a layer2
1399 * packet, it is NULL for a layer-3 packet.
1400 *
1401 * m | args->m Pointer to the mbuf, as received from the caller.
1402 * It may change if ipfw_chk() does an m_pullup, or if it
1403 * consumes the packet because it calls send_reject().
1404 * XXX This has to change, so that ipfw_chk() never modifies
1405 * or consumes the buffer.
1406 * ip is simply an alias of the value of m, and it is kept
1407 * in sync with it (the packet is supposed to start with
1408 * the ip header).
1409 */
1410 struct mbuf *m = args->m;
1411 struct ip *ip = mtod(m, struct ip *);
1412
1413 /*
1414 * oif | args->oif If NULL, ipfw_chk has been called on the
1415 * inbound path (ether_input, bdg_forward, ip_input).
1416 * If non-NULL, ipfw_chk has been called on the outbound path
1417 * (ether_output, ip_output).
1418 */
1419 struct ifnet *oif = args->oif;
1420
1421 struct ip_fw *f = NULL; /* matching rule */
1422 int retval = 0;
1423
1424 /*
1425 * hlen The length of the IPv4 header.
1426 * hlen >0 means we have an IPv4 packet.
1427 */
1428 u_int hlen = 0; /* hlen >0 means we have an IP pkt */
1429
1430 /*
1431 * offset The offset of a fragment. offset != 0 means that
1432 * we have a fragment at this offset of an IPv4 packet.
1433 * offset == 0 means that (if this is an IPv4 packet)
1434 * this is the first or only fragment.
1435 */
1436 u_short offset = 0;
1437
1438 /*
1439 * Local copies of addresses. They are only valid if we have
1440 * an IP packet.
1441 *
1442 * proto The protocol. Set to 0 for non-ip packets,
1443 * or to the protocol read from the packet otherwise.
1444 * proto != 0 means that we have an IPv4 packet.
1445 *
1446 * src_port, dst_port port numbers, in HOST format. Only
1447 * valid for TCP and UDP packets.
1448 *
1449 * src_ip, dst_ip ip addresses, in NETWORK format.
1450 * Only valid for IPv4 packets.
1451 */
1452 u_int8_t proto;
1453 u_int16_t src_port = 0, dst_port = 0; /* NOTE: host format */
1454 struct in_addr src_ip, dst_ip; /* NOTE: network format */
1455 u_int16_t ip_len=0;
1456 int pktlen;
1457 int dyn_dir = MATCH_UNKNOWN;
1458 ipfw_dyn_rule *q = NULL;
1459 struct ip_fw_chain *chain = &layer3_chain;
1460
1461 if (m->m_flags & M_SKIP_FIREWALL)
1462 return 0; /* accept */
1463 /*
1464 * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1465 * MATCH_NONE when checked and not matched (q = NULL),
1466 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
1467 */
1468
1469 pktlen = m->m_pkthdr.len;
1470 if (args->eh == NULL || /* layer 3 packet */
1471 ( m->m_pkthdr.len >= sizeof(struct ip) &&
1472 ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1473 hlen = ip->ip_hl << 2;
1474
1475 /*
1476 * Collect parameters into local variables for faster matching.
1477 */
1478 if (hlen == 0) { /* do not grab addresses for non-ip pkts */
1479 proto = args->f_id.proto = 0; /* mark f_id invalid */
1480 goto after_ip_checks;
1481 }
1482
1483 proto = args->f_id.proto = ip->ip_p;
1484 src_ip = ip->ip_src;
1485 dst_ip = ip->ip_dst;
1486 if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1487 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1488 ip_len = ntohs(ip->ip_len);
1489 } else {
1490 offset = ip->ip_off & IP_OFFMASK;
1491 ip_len = ip->ip_len;
1492 }
1493 pktlen = ip_len < pktlen ? ip_len : pktlen;
1494
1495 #define PULLUP_TO(len) \
1496 do { \
1497 if ((m)->m_len < (len)) { \
1498 args->m = m = m_pullup(m, (len)); \
1499 if (m == 0) \
1500 goto pullup_failed; \
1501 ip = mtod(m, struct ip *); \
1502 } \
1503 } while (0)
1504
1505 if (offset == 0) {
1506 switch (proto) {
1507 case IPPROTO_TCP:
1508 {
1509 struct tcphdr *tcp;
1510
1511 PULLUP_TO(hlen + sizeof(struct tcphdr));
1512 tcp = L3HDR(struct tcphdr, ip);
1513 dst_port = tcp->th_dport;
1514 src_port = tcp->th_sport;
1515 args->f_id.flags = tcp->th_flags;
1516 }
1517 break;
1518
1519 case IPPROTO_UDP:
1520 {
1521 struct udphdr *udp;
1522
1523 PULLUP_TO(hlen + sizeof(struct udphdr));
1524 udp = L3HDR(struct udphdr, ip);
1525 dst_port = udp->uh_dport;
1526 src_port = udp->uh_sport;
1527 }
1528 break;
1529
1530 case IPPROTO_ICMP:
1531 PULLUP_TO(hlen + 4); /* type, code and checksum. */
1532 args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1533 break;
1534
1535 default:
1536 break;
1537 }
1538 #undef PULLUP_TO
1539 }
1540
1541 args->f_id.src_ip = ntohl(src_ip.s_addr);
1542 args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1543 args->f_id.src_port = src_port = ntohs(src_port);
1544 args->f_id.dst_port = dst_port = ntohs(dst_port);
1545
1546 after_ip_checks:
1547 IPFW_LOCK(chain); /* XXX expensive? can we run lock free? */
1548 if (args->rule) {
1549 /*
1550 * Packet has already been tagged. Look for the next rule
1551 * to restart processing.
1552 *
1553 * If fw_one_pass != 0 then just accept it.
1554 * XXX should not happen here, but optimized out in
1555 * the caller.
1556 */
1557 if (fw_one_pass) {
1558 IPFW_UNLOCK(chain); /* XXX optimize */
1559 return 0;
1560 }
1561
1562 f = args->rule->next_rule;
1563 if (f == NULL)
1564 f = lookup_next_rule(args->rule);
1565 } else {
1566 /*
1567 * Find the starting rule. It can be either the first
1568 * one, or the one after divert_rule if asked so.
1569 */
1570 int skipto = args->divert_rule;
1571
1572 f = chain->rules;
1573 if (args->eh == NULL && skipto != 0) {
1574 if (skipto >= IPFW_DEFAULT_RULE) {
1575 IPFW_UNLOCK(chain);
1576 return(IP_FW_PORT_DENY_FLAG); /* invalid */
1577 }
1578 while (f && f->rulenum <= skipto)
1579 f = f->next;
1580 if (f == NULL) { /* drop packet */
1581 IPFW_UNLOCK(chain);
1582 return(IP_FW_PORT_DENY_FLAG);
1583 }
1584 }
1585 }
1586 args->divert_rule = 0; /* reset to avoid confusion later */
1587
1588 /*
1589 * Now scan the rules, and parse microinstructions for each rule.
1590 */
1591 for (; f; f = f->next) {
1592 int l, cmdlen;
1593 ipfw_insn *cmd;
1594 int skip_or; /* skip rest of OR block */
1595
1596 again:
1597 if (set_disable & (1 << f->set) )
1598 continue;
1599
1600 skip_or = 0;
1601 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ;
1602 l -= cmdlen, cmd += cmdlen) {
1603 int match;
1604
1605 /*
1606 * check_body is a jump target used when we find a
1607 * CHECK_STATE, and need to jump to the body of
1608 * the target rule.
1609 */
1610
1611 check_body:
1612 cmdlen = F_LEN(cmd);
1613 /*
1614 * An OR block (insn_1 || .. || insn_n) has the
1615 * F_OR bit set in all but the last instruction.
1616 * The first match will set "skip_or", and cause
1617 * the following instructions to be skipped until
1618 * past the one with the F_OR bit clear.
1619 */
1620 if (skip_or) { /* skip this instruction */
1621 if ((cmd->len & F_OR) == 0)
1622 skip_or = 0; /* next one is good */
1623 continue;
1624 }
1625 match = 0; /* set to 1 if we succeed */
1626
1627 switch (cmd->opcode) {
1628 /*
1629 * The first set of opcodes compares the packet's
1630 * fields with some pattern, setting 'match' if a
1631 * match is found. At the end of the loop there is
1632 * logic to deal with F_NOT and F_OR flags associated
1633 * with the opcode.
1634 */
1635 case O_NOP:
1636 match = 1;
1637 break;
1638
1639 case O_FORWARD_MAC:
1640 printf("ipfw: opcode %d unimplemented\n",
1641 cmd->opcode);
1642 break;
1643
1644 case O_GID:
1645 case O_UID:
1646 /*
1647 * We only check offset == 0 && proto != 0,
1648 * as this ensures that we have an IPv4
1649 * packet with the ports info.
1650 */
1651 if (offset!=0)
1652 break;
1653 if (proto == IPPROTO_TCP ||
1654 proto == IPPROTO_UDP)
1655 match = check_uidgid(
1656 (ipfw_insn_u32 *)cmd,
1657 proto, oif,
1658 dst_ip, dst_port,
1659 src_ip, src_port);
1660 break;
1661
1662 case O_RECV:
1663 match = iface_match(m->m_pkthdr.rcvif,
1664 (ipfw_insn_if *)cmd);
1665 break;
1666
1667 case O_XMIT:
1668 match = iface_match(oif, (ipfw_insn_if *)cmd);
1669 break;
1670
1671 case O_VIA:
1672 match = iface_match(oif ? oif :
1673 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1674 break;
1675
1676 case O_MACADDR2:
1677 if (args->eh != NULL) { /* have MAC header */
1678 u_int32_t *want = (u_int32_t *)
1679 ((ipfw_insn_mac *)cmd)->addr;
1680 u_int32_t *mask = (u_int32_t *)
1681 ((ipfw_insn_mac *)cmd)->mask;
1682 u_int32_t *hdr = (u_int32_t *)args->eh;
1683
1684 match =
1685 ( want[0] == (hdr[0] & mask[0]) &&
1686 want[1] == (hdr[1] & mask[1]) &&
1687 want[2] == (hdr[2] & mask[2]) );
1688 }
1689 break;
1690
1691 case O_MAC_TYPE:
1692 if (args->eh != NULL) {
1693 u_int16_t t =
1694 ntohs(args->eh->ether_type);
1695 u_int16_t *p =
1696 ((ipfw_insn_u16 *)cmd)->ports;
1697 int i;
1698
1699 for (i = cmdlen - 1; !match && i>0;
1700 i--, p += 2)
1701 match = (t>=p[0] && t<=p[1]);
1702 }
1703 break;
1704
1705 case O_FRAG:
1706 match = (hlen > 0 && offset != 0);
1707 break;
1708
1709 case O_IN: /* "out" is "not in" */
1710 match = (oif == NULL);
1711 break;
1712
1713 case O_LAYER2:
1714 match = (args->eh != NULL);
1715 break;
1716
1717 case O_PROTO:
1718 /*
1719 * We do not allow an arg of 0 so the
1720 * check of "proto" only suffices.
1721 */
1722 match = (proto == cmd->arg1);
1723 break;
1724
1725 case O_IP_SRC:
1726 match = (hlen > 0 &&
1727 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1728 src_ip.s_addr);
1729 break;
1730
1731 case O_IP_SRC_MASK:
1732 case O_IP_DST_MASK:
1733 if (hlen > 0) {
1734 uint32_t a =
1735 (cmd->opcode == O_IP_DST_MASK) ?
1736 dst_ip.s_addr : src_ip.s_addr;
1737 uint32_t *p = ((ipfw_insn_u32 *)cmd)->d;
1738 int i = cmdlen-1;
1739
1740 for (; !match && i>0; i-= 2, p+= 2)
1741 match = (p[0] == (a & p[1]));
1742 }
1743 break;
1744
1745 case O_IP_SRC_ME:
1746 if (hlen > 0) {
1747 struct ifnet *tif;
1748
1749 INADDR_TO_IFP(src_ip, tif);
1750 match = (tif != NULL);
1751 }
1752 break;
1753
1754 case O_IP_DST_SET:
1755 case O_IP_SRC_SET:
1756 if (hlen > 0) {
1757 u_int32_t *d = (u_int32_t *)(cmd+1);
1758 u_int32_t addr =
1759 cmd->opcode == O_IP_DST_SET ?
1760 args->f_id.dst_ip :
1761 args->f_id.src_ip;
1762
1763 if (addr < d[0])
1764 break;
1765 addr -= d[0]; /* subtract base */
1766 match = (addr < cmd->arg1) &&
1767 ( d[ 1 + (addr>>5)] &
1768 (1<<(addr & 0x1f)) );
1769 }
1770 break;
1771
1772 case O_IP_DST:
1773 match = (hlen > 0 &&
1774 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
1775 dst_ip.s_addr);
1776 break;
1777
1778 case O_IP_DST_ME:
1779 if (hlen > 0) {
1780 struct ifnet *tif;
1781
1782 INADDR_TO_IFP(dst_ip, tif);
1783 match = (tif != NULL);
1784 }
1785 break;
1786
1787 case O_IP_SRCPORT:
1788 case O_IP_DSTPORT:
1789 /*
1790 * offset == 0 && proto != 0 is enough
1791 * to guarantee that we have an IPv4
1792 * packet with port info.
1793 */
1794 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
1795 && offset == 0) {
1796 u_int16_t x =
1797 (cmd->opcode == O_IP_SRCPORT) ?
1798 src_port : dst_port ;
1799 u_int16_t *p =
1800 ((ipfw_insn_u16 *)cmd)->ports;
1801 int i;
1802
1803 for (i = cmdlen - 1; !match && i>0;
1804 i--, p += 2)
1805 match = (x>=p[0] && x<=p[1]);
1806 }
1807 break;
1808
1809 case O_ICMPTYPE:
1810 match = (offset == 0 && proto==IPPROTO_ICMP &&
1811 icmptype_match(ip, (ipfw_insn_u32 *)cmd) );
1812 break;
1813
1814 case O_IPOPT:
1815 match = (hlen > 0 && ipopts_match(ip, cmd) );
1816 break;
1817
1818 case O_IPVER:
1819 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
1820 break;
1821
1822 case O_IPID:
1823 case O_IPLEN:
1824 case O_IPTTL:
1825 if (hlen > 0) { /* only for IP packets */
1826 uint16_t x;
1827 uint16_t *p;
1828 int i;
1829
1830 if (cmd->opcode == O_IPLEN)
1831 x = ip_len;
1832 else if (cmd->opcode == O_IPTTL)
1833 x = ip->ip_ttl;
1834 else /* must be IPID */
1835 x = ntohs(ip->ip_id);
1836 if (cmdlen == 1) {
1837 match = (cmd->arg1 == x);
1838 break;
1839 }
1840 /* otherwise we have ranges */
1841 p = ((ipfw_insn_u16 *)cmd)->ports;
1842 i = cmdlen - 1;
1843 for (; !match && i>0; i--, p += 2)
1844 match = (x >= p[0] && x <= p[1]);
1845 }
1846 break;
1847
1848 case O_IPPRECEDENCE:
1849 match = (hlen > 0 &&
1850 (cmd->arg1 == (ip->ip_tos & 0xe0)) );
1851 break;
1852
1853 case O_IPTOS:
1854 match = (hlen > 0 &&
1855 flags_match(cmd, ip->ip_tos));
1856 break;
1857
1858 case O_TCPFLAGS:
1859 match = (proto == IPPROTO_TCP && offset == 0 &&
1860 flags_match(cmd,
1861 L3HDR(struct tcphdr,ip)->th_flags));
1862 break;
1863
1864 case O_TCPOPTS:
1865 match = (proto == IPPROTO_TCP && offset == 0 &&
1866 tcpopts_match(ip, cmd));
1867 break;
1868
1869 case O_TCPSEQ:
1870 match = (proto == IPPROTO_TCP && offset == 0 &&
1871 ((ipfw_insn_u32 *)cmd)->d[0] ==
1872 L3HDR(struct tcphdr,ip)->th_seq);
1873 break;
1874
1875 case O_TCPACK:
1876 match = (proto == IPPROTO_TCP && offset == 0 &&
1877 ((ipfw_insn_u32 *)cmd)->d[0] ==
1878 L3HDR(struct tcphdr,ip)->th_ack);
1879 break;
1880
1881 case O_TCPWIN:
1882 match = (proto == IPPROTO_TCP && offset == 0 &&
1883 cmd->arg1 ==
1884 L3HDR(struct tcphdr,ip)->th_win);
1885 break;
1886
1887 case O_ESTAB:
1888 /* reject packets which have SYN only */
1889 /* XXX should i also check for TH_ACK ? */
1890 match = (proto == IPPROTO_TCP && offset == 0 &&
1891 (L3HDR(struct tcphdr,ip)->th_flags &
1892 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
1893 break;
1894
1895 case O_LOG:
1896 if (fw_verbose)
1897 ipfw_log(f, hlen, args->eh, m, oif);
1898 match = 1;
1899 break;
1900
1901 case O_PROB:
1902 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]);
1903 break;
1904
1905 case O_VERREVPATH:
1906 /* Outgoing packets automatically pass/match */
1907 match = ((oif != NULL) ||
1908 (m->m_pkthdr.rcvif == NULL) ||
1909 verify_rev_path(src_ip, m->m_pkthdr.rcvif));
1910 break;
1911
1912 case O_IPSEC:
1913 #ifdef FAST_IPSEC
1914 match = (m_tag_find(m,
1915 PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL);
1916 #endif
1917 #ifdef IPSEC
1918 match = (ipsec_getnhist(m) != NULL);
1919 #endif
1920 /* otherwise no match */
1921 break;
1922
1923 /*
1924 * The second set of opcodes represents 'actions',
1925 * i.e. the terminal part of a rule once the packet
1926 * matches all previous patterns.
1927 * Typically there is only one action for each rule,
1928 * and the opcode is stored at the end of the rule
1929 * (but there are exceptions -- see below).
1930 *
1931 * In general, here we set retval and terminate the
1932 * outer loop (would be a 'break 3' in some language,
1933 * but we need to do a 'goto done').
1934 *
1935 * Exceptions:
1936 * O_COUNT and O_SKIPTO actions:
1937 * instead of terminating, we jump to the next rule
1938 * ('goto next_rule', equivalent to a 'break 2'),
1939 * or to the SKIPTO target ('goto again' after
1940 * having set f, cmd and l), respectively.
1941 *
1942 * O_LIMIT and O_KEEP_STATE: these opcodes are
1943 * not real 'actions', and are stored right
1944 * before the 'action' part of the rule.
1945 * These opcodes try to install an entry in the
1946 * state tables; if successful, we continue with
1947 * the next opcode (match=1; break;), otherwise
1948 * the packet * must be dropped
1949 * ('goto done' after setting retval);
1950 *
1951 * O_PROBE_STATE and O_CHECK_STATE: these opcodes
1952 * cause a lookup of the state table, and a jump
1953 * to the 'action' part of the parent rule
1954 * ('goto check_body') if an entry is found, or
1955 * (CHECK_STATE only) a jump to the next rule if
1956 * the entry is not found ('goto next_rule').
1957 * The result of the lookup is cached to make
1958 * further instances of these opcodes are
1959 * effectively NOPs.
1960 */
1961 case O_LIMIT:
1962 case O_KEEP_STATE:
1963 if (install_state(f,
1964 (ipfw_insn_limit *)cmd, args)) {
1965 retval = IP_FW_PORT_DENY_FLAG;
1966 goto done; /* error/limit violation */
1967 }
1968 match = 1;
1969 break;
1970
1971 case O_PROBE_STATE:
1972 case O_CHECK_STATE:
1973 /*
1974 * dynamic rules are checked at the first
1975 * keep-state or check-state occurrence,
1976 * with the result being stored in dyn_dir.
1977 * The compiler introduces a PROBE_STATE
1978 * instruction for us when we have a
1979 * KEEP_STATE (because PROBE_STATE needs
1980 * to be run first).
1981 */
1982 if (dyn_dir == MATCH_UNKNOWN &&
1983 (q = lookup_dyn_rule(&args->f_id,
1984 &dyn_dir, proto == IPPROTO_TCP ?
1985 L3HDR(struct tcphdr, ip) : NULL))
1986 != NULL) {
1987 /*
1988 * Found dynamic entry, update stats
1989 * and jump to the 'action' part of
1990 * the parent rule.
1991 */
1992 q->pcnt++;
1993 q->bcnt += pktlen;
1994 f = q->rule;
1995 cmd = ACTION_PTR(f);
1996 l = f->cmd_len - f->act_ofs;
1997 IPFW_DYN_UNLOCK();
1998 goto check_body;
1999 }
2000 /*
2001 * Dynamic entry not found. If CHECK_STATE,
2002 * skip to next rule, if PROBE_STATE just
2003 * ignore and continue with next opcode.
2004 */
2005 if (cmd->opcode == O_CHECK_STATE)
2006 goto next_rule;
2007 match = 1;
2008 break;
2009
2010 case O_ACCEPT:
2011 retval = 0; /* accept */
2012 goto done;
2013
2014 case O_PIPE:
2015 case O_QUEUE:
2016 args->rule = f; /* report matching rule */
2017 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG;
2018 goto done;
2019
2020 case O_DIVERT:
2021 case O_TEE:
2022 if (args->eh) /* not on layer 2 */
2023 break;
2024 args->divert_rule = f->rulenum;
2025 retval = (cmd->opcode == O_DIVERT) ?
2026 cmd->arg1 :
2027 cmd->arg1 | IP_FW_PORT_TEE_FLAG;
2028 goto done;
2029
2030 case O_COUNT:
2031 case O_SKIPTO:
2032 f->pcnt++; /* update stats */
2033 f->bcnt += pktlen;
2034 f->timestamp = time_second;
2035 if (cmd->opcode == O_COUNT)
2036 goto next_rule;
2037 /* handle skipto */
2038 if (f->next_rule == NULL)
2039 lookup_next_rule(f);
2040 f = f->next_rule;
2041 goto again;
2042
2043 case O_REJECT:
2044 /*
2045 * Drop the packet and send a reject notice
2046 * if the packet is not ICMP (or is an ICMP
2047 * query), and it is not multicast/broadcast.
2048 */
2049 if (hlen > 0 &&
2050 (proto != IPPROTO_ICMP ||
2051 is_icmp_query(ip)) &&
2052 !(m->m_flags & (M_BCAST|M_MCAST)) &&
2053 !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
2054 send_reject(args, cmd->arg1,
2055 offset,ip_len);
2056 m = args->m;
2057 }
2058 /* FALLTHROUGH */
2059 case O_DENY:
2060 retval = IP_FW_PORT_DENY_FLAG;
2061 goto done;
2062
2063 case O_FORWARD_IP:
2064 if (args->eh) /* not valid on layer2 pkts */
2065 break;
2066 if (!q || dyn_dir == MATCH_FORWARD)
2067 args->next_hop =
2068 &((ipfw_insn_sa *)cmd)->sa;
2069 retval = 0;
2070 goto done;
2071
2072 default:
2073 panic("-- unknown opcode %d\n", cmd->opcode);
2074 } /* end of switch() on opcodes */
2075
2076 if (cmd->len & F_NOT)
2077 match = !match;
2078
2079 if (match) {
2080 if (cmd->len & F_OR)
2081 skip_or = 1;
2082 } else {
2083 if (!(cmd->len & F_OR)) /* not an OR block, */
2084 break; /* try next rule */
2085 }
2086
2087 } /* end of inner for, scan opcodes */
2088
2089 next_rule:; /* try next rule */
2090
2091 } /* end of outer for, scan rules */
2092 printf("ipfw: ouch!, skip past end of rules, denying packet\n");
2093 IPFW_UNLOCK(chain);
2094 return(IP_FW_PORT_DENY_FLAG);
2095
2096 done:
2097 /* Update statistics */
2098 f->pcnt++;
2099 f->bcnt += pktlen;
2100 f->timestamp = time_second;
2101 IPFW_UNLOCK(chain);
2102 return retval;
2103
2104 pullup_failed:
2105 if (fw_verbose)
2106 printf("ipfw: pullup failed\n");
2107 return(IP_FW_PORT_DENY_FLAG);
2108 }
2109
2110 /*
2111 * When a rule is added/deleted, clear the next_rule pointers in all rules.
2112 * These will be reconstructed on the fly as packets are matched.
2113 */
2114 static void
2115 flush_rule_ptrs(struct ip_fw_chain *chain)
2116 {
2117 struct ip_fw *rule;
2118
2119 IPFW_LOCK_ASSERT(chain);
2120
2121 for (rule = chain->rules; rule; rule = rule->next)
2122 rule->next_rule = NULL;
2123 }
2124
2125 /*
2126 * When pipes/queues are deleted, clear the "pipe_ptr" pointer to a given
2127 * pipe/queue, or to all of them (match == NULL).
2128 */
2129 void
2130 flush_pipe_ptrs(struct dn_flow_set *match)
2131 {
2132 struct ip_fw *rule;
2133
2134 IPFW_LOCK(&layer3_chain);
2135 for (rule = layer3_chain.rules; rule; rule = rule->next) {
2136 ipfw_insn_pipe *cmd = (ipfw_insn_pipe *)ACTION_PTR(rule);
2137
2138 if (cmd->o.opcode != O_PIPE && cmd->o.opcode != O_QUEUE)
2139 continue;
2140 /*
2141 * XXX Use bcmp/bzero to handle pipe_ptr to overcome
2142 * possible alignment problems on 64-bit architectures.
2143 * This code is seldom used so we do not worry too
2144 * much about efficiency.
2145 */
2146 if (match == NULL ||
2147 !bcmp(&cmd->pipe_ptr, &match, sizeof(match)) )
2148 bzero(&cmd->pipe_ptr, sizeof(cmd->pipe_ptr));
2149 }
2150 IPFW_UNLOCK(&layer3_chain);
2151 }
2152
2153 /*
2154 * Add a new rule to the list. Copy the rule into a malloc'ed area, then
2155 * possibly create a rule number and add the rule to the list.
2156 * Update the rule_number in the input struct so the caller knows it as well.
2157 */
2158 static int
2159 add_rule(struct ip_fw_chain *chain, struct ip_fw *input_rule)
2160 {
2161 struct ip_fw *rule, *f, *prev;
2162 int l = RULESIZE(input_rule);
2163
2164 if (chain->rules == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE)
2165 return (EINVAL);
2166
2167 rule = malloc(l, M_IPFW, M_NOWAIT | M_ZERO);
2168 if (rule == NULL)
2169 return (ENOSPC);
2170
2171 bcopy(input_rule, rule, l);
2172
2173 rule->next = NULL;
2174 rule->next_rule = NULL;
2175
2176 rule->pcnt = 0;
2177 rule->bcnt = 0;
2178 rule->timestamp = 0;
2179
2180 IPFW_LOCK(chain);
2181
2182 if (chain->rules == NULL) { /* default rule */
2183 chain->rules = rule;
2184 goto done;
2185 }
2186
2187 /*
2188 * If rulenum is 0, find highest numbered rule before the
2189 * default rule, and add autoinc_step
2190 */
2191 if (autoinc_step < 1)
2192 autoinc_step = 1;
2193 else if (autoinc_step > 1000)
2194 autoinc_step = 1000;
2195 if (rule->rulenum == 0) {
2196 /*
2197 * locate the highest numbered rule before default
2198 */
2199 for (f = chain->rules; f; f = f->next) {
2200 if (f->rulenum == IPFW_DEFAULT_RULE)
2201 break;
2202 rule->rulenum = f->rulenum;
2203 }
2204 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
2205 rule->rulenum += autoinc_step;
2206 input_rule->rulenum = rule->rulenum;
2207 }
2208
2209 /*
2210 * Now insert the new rule in the right place in the sorted list.
2211 */
2212 for (prev = NULL, f = chain->rules; f; prev = f, f = f->next) {
2213 if (f->rulenum > rule->rulenum) { /* found the location */
2214 if (prev) {
2215 rule->next = f;
2216 prev->next = rule;
2217 } else { /* head insert */
2218 rule->next = chain->rules;
2219 chain->rules = rule;
2220 }
2221 break;
2222 }
2223 }
2224 flush_rule_ptrs(chain);
2225 done:
2226 static_count++;
2227 static_len += l;
2228 IPFW_UNLOCK(chain);
2229 DEB(printf("ipfw: installed rule %d, static count now %d\n",
2230 rule->rulenum, static_count);)
2231 return (0);
2232 }
2233
2234 /**
2235 * Remove a static rule (including derived * dynamic rules)
2236 * and place it on the ``reap list'' for later reclamation.
2237 * The caller is in charge of clearing rule pointers to avoid
2238 * dangling pointers.
2239 * @return a pointer to the next entry.
2240 * Arguments are not checked, so they better be correct.
2241 */
2242 static struct ip_fw *
2243 remove_rule(struct ip_fw_chain *chain, struct ip_fw *rule, struct ip_fw *prev)
2244 {
2245 struct ip_fw *n;
2246 int l = RULESIZE(rule);
2247
2248 IPFW_LOCK_ASSERT(chain);
2249
2250 n = rule->next;
2251 IPFW_DYN_LOCK();
2252 remove_dyn_rule(rule, NULL /* force removal */);
2253 IPFW_DYN_UNLOCK();
2254 if (prev == NULL)
2255 chain->rules = n;
2256 else
2257 prev->next = n;
2258 static_count--;
2259 static_len -= l;
2260
2261 rule->next = chain->reap;
2262 chain->reap = rule;
2263
2264 return n;
2265 }
2266
2267 /**
2268 * Reclaim storage associated with a list of rules. This is
2269 * typically the list created using remove_rule.
2270 */
2271 static void
2272 reap_rules(struct ip_fw *head)
2273 {
2274 struct ip_fw *rule;
2275
2276 while ((rule = head) != NULL) {
2277 head = head->next;
2278 if (DUMMYNET_LOADED)
2279 ip_dn_ruledel_ptr(rule);
2280 free(rule, M_IPFW);
2281 }
2282 }
2283
2284 /*
2285 * Remove all rules from a chain (except rules in set RESVD_SET
2286 * unless kill_default = 1). The caller is responsible for
2287 * reclaiming storage for the rules left in chain->reap.
2288 */
2289 static void
2290 free_chain(struct ip_fw_chain *chain, int kill_default)
2291 {
2292 struct ip_fw *prev, *rule;
2293
2294 IPFW_LOCK_ASSERT(chain);
2295
2296 flush_rule_ptrs(chain); /* more efficient to do outside the loop */
2297 for (prev = NULL, rule = chain->rules; rule ; )
2298 if (kill_default || rule->set != RESVD_SET)
2299 rule = remove_rule(chain, rule, prev);
2300 else {
2301 prev = rule;
2302 rule = rule->next;
2303 }
2304 }
2305
2306 /**
2307 * Remove all rules with given number, and also do set manipulation.
2308 * Assumes chain != NULL && *chain != NULL.
2309 *
2310 * The argument is an u_int32_t. The low 16 bit are the rule or set number,
2311 * the next 8 bits are the new set, the top 8 bits are the command:
2312 *
2313 * 0 delete rules with given number
2314 * 1 delete rules with given set number
2315 * 2 move rules with given number to new set
2316 * 3 move rules with given set number to new set
2317 * 4 swap sets with given numbers
2318 */
2319 static int
2320 del_entry(struct ip_fw_chain *chain, u_int32_t arg)
2321 {
2322 struct ip_fw *prev = NULL, *rule;
2323 u_int16_t rulenum; /* rule or old_set */
2324 u_int8_t cmd, new_set;
2325
2326 rulenum = arg & 0xffff;
2327 cmd = (arg >> 24) & 0xff;
2328 new_set = (arg >> 16) & 0xff;
2329
2330 if (cmd > 4)
2331 return EINVAL;
2332 if (new_set > RESVD_SET)
2333 return EINVAL;
2334 if (cmd == 0 || cmd == 2) {
2335 if (rulenum >= IPFW_DEFAULT_RULE)
2336 return EINVAL;
2337 } else {
2338 if (rulenum > RESVD_SET) /* old_set */
2339 return EINVAL;
2340 }
2341
2342 IPFW_LOCK(chain);
2343 rule = chain->rules;
2344 chain->reap = NULL;
2345 switch (cmd) {
2346 case 0: /* delete rules with given number */
2347 /*
2348 * locate first rule to delete
2349 */
2350 for (; rule->rulenum < rulenum; prev = rule, rule = rule->next)
2351 ;
2352 if (rule->rulenum != rulenum) {
2353 IPFW_UNLOCK(chain);
2354 return EINVAL;
2355 }
2356
2357 /*
2358 * flush pointers outside the loop, then delete all matching
2359 * rules. prev remains the same throughout the cycle.
2360 */
2361 flush_rule_ptrs(chain);
2362 while (rule->rulenum == rulenum)
2363 rule = remove_rule(chain, rule, prev);
2364 break;
2365
2366 case 1: /* delete all rules with given set number */
2367 flush_rule_ptrs(chain);
2368 rule = chain->rules;
2369 while (rule->rulenum < IPFW_DEFAULT_RULE)
2370 if (rule->set == rulenum)
2371 rule = remove_rule(chain, rule, prev);
2372 else {
2373 prev = rule;
2374 rule = rule->next;
2375 }
2376 break;
2377
2378 case 2: /* move rules with given number to new set */
2379 rule = chain->rules;
2380 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
2381 if (rule->rulenum == rulenum)
2382 rule->set = new_set;
2383 break;
2384
2385 case 3: /* move rules with given set number to new set */
2386 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
2387 if (rule->set == rulenum)
2388 rule->set = new_set;
2389 break;
2390
2391 case 4: /* swap two sets */
2392 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
2393 if (rule->set == rulenum)
2394 rule->set = new_set;
2395 else if (rule->set == new_set)
2396 rule->set = rulenum;
2397 break;
2398 }
2399 /*
2400 * Look for rules to reclaim. We grab the list before
2401 * releasing the lock then reclaim them w/o the lock to
2402 * avoid a LOR with dummynet.
2403 */
2404 rule = chain->reap;
2405 chain->reap = NULL;
2406 IPFW_UNLOCK(chain);
2407 if (rule)
2408 reap_rules(rule);
2409 return 0;
2410 }
2411
2412 /*
2413 * Clear counters for a specific rule.
2414 * The enclosing "table" is assumed locked.
2415 */
2416 static void
2417 clear_counters(struct ip_fw *rule, int log_only)
2418 {
2419 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
2420
2421 if (log_only == 0) {
2422 rule->bcnt = rule->pcnt = 0;
2423 rule->timestamp = 0;
2424 }
2425 if (l->o.opcode == O_LOG)
2426 l->log_left = l->max_log;
2427 }
2428
2429 /**
2430 * Reset some or all counters on firewall rules.
2431 * @arg frwl is null to clear all entries, or contains a specific
2432 * rule number.
2433 * @arg log_only is 1 if we only want to reset logs, zero otherwise.
2434 */
2435 static int
2436 zero_entry(struct ip_fw_chain *chain, int rulenum, int log_only)
2437 {
2438 struct ip_fw *rule;
2439 char *msg;
2440
2441 IPFW_LOCK(chain);
2442 if (rulenum == 0) {
2443 norule_counter = 0;
2444 for (rule = chain->rules; rule; rule = rule->next)
2445 clear_counters(rule, log_only);
2446 msg = log_only ? "ipfw: All logging counts reset.\n" :
2447 "ipfw: Accounting cleared.\n";
2448 } else {
2449 int cleared = 0;
2450 /*
2451 * We can have multiple rules with the same number, so we
2452 * need to clear them all.
2453 */
2454 for (rule = chain->rules; rule; rule = rule->next)
2455 if (rule->rulenum == rulenum) {
2456 while (rule && rule->rulenum == rulenum) {
2457 clear_counters(rule, log_only);
2458 rule = rule->next;
2459 }
2460 cleared = 1;
2461 break;
2462 }
2463 if (!cleared) { /* we did not find any matching rules */
2464 IPFW_UNLOCK(chain);
2465 return (EINVAL);
2466 }
2467 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
2468 "ipfw: Entry %d cleared.\n";
2469 }
2470 IPFW_UNLOCK(chain);
2471
2472 if (fw_verbose)
2473 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
2474 return (0);
2475 }
2476
2477 /*
2478 * Check validity of the structure before insert.
2479 * Fortunately rules are simple, so this mostly need to check rule sizes.
2480 */
2481 static int
2482 check_ipfw_struct(struct ip_fw *rule, int size)
2483 {
2484 int l, cmdlen = 0;
2485 int have_action=0;
2486 ipfw_insn *cmd;
2487
2488 if (size < sizeof(*rule)) {
2489 printf("ipfw: rule too short\n");
2490 return (EINVAL);
2491 }
2492 /* first, check for valid size */
2493 l = RULESIZE(rule);
2494 if (l != size) {
2495 printf("ipfw: size mismatch (have %d want %d)\n", size, l);
2496 return (EINVAL);
2497 }
2498 /*
2499 * Now go for the individual checks. Very simple ones, basically only
2500 * instruction sizes.
2501 */
2502 for (l = rule->cmd_len, cmd = rule->cmd ;
2503 l > 0 ; l -= cmdlen, cmd += cmdlen) {
2504 cmdlen = F_LEN(cmd);
2505 if (cmdlen > l) {
2506 printf("ipfw: opcode %d size truncated\n",
2507 cmd->opcode);
2508 return EINVAL;
2509 }
2510 DEB(printf("ipfw: opcode %d\n", cmd->opcode);)
2511 switch (cmd->opcode) {
2512 case O_PROBE_STATE:
2513 case O_KEEP_STATE:
2514 case O_PROTO:
2515 case O_IP_SRC_ME:
2516 case O_IP_DST_ME:
2517 case O_LAYER2:
2518 case O_IN:
2519 case O_FRAG:
2520 case O_IPOPT:
2521 case O_IPTOS:
2522 case O_IPPRECEDENCE:
2523 case O_IPVER:
2524 case O_TCPWIN:
2525 case O_TCPFLAGS:
2526 case O_TCPOPTS:
2527 case O_ESTAB:
2528 case O_VERREVPATH:
2529 case O_IPSEC:
2530 if (cmdlen != F_INSN_SIZE(ipfw_insn))
2531 goto bad_size;
2532 break;
2533
2534 case O_UID:
2535 case O_GID:
2536 case O_IP_SRC:
2537 case O_IP_DST:
2538 case O_TCPSEQ:
2539 case O_TCPACK:
2540 case O_PROB:
2541 case O_ICMPTYPE:
2542 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
2543 goto bad_size;
2544 break;
2545
2546 case O_LIMIT:
2547 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
2548 goto bad_size;
2549 break;
2550
2551 case O_LOG:
2552 if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
2553 goto bad_size;
2554
2555 ((ipfw_insn_log *)cmd)->log_left =
2556 ((ipfw_insn_log *)cmd)->max_log;
2557
2558 break;
2559
2560 case O_IP_SRC_MASK:
2561 case O_IP_DST_MASK:
2562 /* only odd command lengths */
2563 if ( !(cmdlen & 1) || cmdlen > 31)
2564 goto bad_size;
2565 break;
2566
2567 case O_IP_SRC_SET:
2568 case O_IP_DST_SET:
2569 if (cmd->arg1 == 0 || cmd->arg1 > 256) {
2570 printf("ipfw: invalid set size %d\n",
2571 cmd->arg1);
2572 return EINVAL;
2573 }
2574 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
2575 (cmd->arg1+31)/32 )
2576 goto bad_size;
2577 break;
2578
2579 case O_MACADDR2:
2580 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
2581 goto bad_size;
2582 break;
2583
2584 case O_NOP:
2585 case O_IPID:
2586 case O_IPTTL:
2587 case O_IPLEN:
2588 if (cmdlen < 1 || cmdlen > 31)
2589 goto bad_size;
2590 break;
2591
2592 case O_MAC_TYPE:
2593 case O_IP_SRCPORT:
2594 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
2595 if (cmdlen < 2 || cmdlen > 31)
2596 goto bad_size;
2597 break;
2598
2599 case O_RECV:
2600 case O_XMIT:
2601 case O_VIA:
2602 if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
2603 goto bad_size;
2604 break;
2605
2606 case O_PIPE:
2607 case O_QUEUE:
2608 if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
2609 goto bad_size;
2610 goto check_action;
2611
2612 case O_FORWARD_IP:
2613 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
2614 goto bad_size;
2615 goto check_action;
2616
2617 case O_FORWARD_MAC: /* XXX not implemented yet */
2618 case O_CHECK_STATE:
2619 case O_COUNT:
2620 case O_ACCEPT:
2621 case O_DENY:
2622 case O_REJECT:
2623 case O_SKIPTO:
2624 case O_DIVERT:
2625 case O_TEE:
2626 if (cmdlen != F_INSN_SIZE(ipfw_insn))
2627 goto bad_size;
2628 check_action:
2629 if (have_action) {
2630 printf("ipfw: opcode %d, multiple actions"
2631 " not allowed\n",
2632 cmd->opcode);
2633 return EINVAL;
2634 }
2635 have_action = 1;
2636 if (l != cmdlen) {
2637 printf("ipfw: opcode %d, action must be"
2638 " last opcode\n",
2639 cmd->opcode);
2640 return EINVAL;
2641 }
2642 break;
2643 default:
2644 printf("ipfw: opcode %d, unknown opcode\n",
2645 cmd->opcode);
2646 return EINVAL;
2647 }
2648 }
2649 if (have_action == 0) {
2650 printf("ipfw: missing action\n");
2651 return EINVAL;
2652 }
2653 return 0;
2654
2655 bad_size:
2656 printf("ipfw: opcode %d size %d wrong\n",
2657 cmd->opcode, cmdlen);
2658 return EINVAL;
2659 }
2660
2661 /*
2662 * Copy the static and dynamic rules to the supplied buffer
2663 * and return the amount of space actually used.
2664 */
2665 static size_t
2666 ipfw_getrules(struct ip_fw_chain *chain, void *buf, size_t space)
2667 {
2668 char *bp = buf;
2669 char *ep = bp + space;
2670 struct ip_fw *rule;
2671 int i;
2672
2673 /* XXX this can take a long time and locking will block packet flow */
2674 IPFW_LOCK(chain);
2675 for (rule = chain->rules; rule ; rule = rule->next) {
2676 /*
2677 * Verify the entry fits in the buffer in case the
2678 * rules changed between calculating buffer space and
2679 * now. This would be better done using a generation
2680 * number but should suffice for now.
2681 */
2682 i = RULESIZE(rule);
2683 if (bp + i <= ep) {
2684 bcopy(rule, bp, i);
2685 bcopy(&set_disable, &(((struct ip_fw *)bp)->next_rule),
2686 sizeof(set_disable));
2687 bp += i;
2688 }
2689 }
2690 IPFW_UNLOCK(chain);
2691 if (ipfw_dyn_v) {
2692 ipfw_dyn_rule *p, *last = NULL;
2693
2694 IPFW_DYN_LOCK();
2695 for (i = 0 ; i < curr_dyn_buckets; i++)
2696 for (p = ipfw_dyn_v[i] ; p != NULL; p = p->next) {
2697 if (bp + sizeof *p <= ep) {
2698 ipfw_dyn_rule *dst =
2699 (ipfw_dyn_rule *)bp;
2700 bcopy(p, dst, sizeof *p);
2701 bcopy(&(p->rule->rulenum), &(dst->rule),
2702 sizeof(p->rule->rulenum));
2703 /*
2704 * store a non-null value in "next".
2705 * The userland code will interpret a
2706 * NULL here as a marker
2707 * for the last dynamic rule.
2708 */
2709 bcopy(&dst, &dst->next, sizeof(dst));
2710 last = dst;
2711 dst->expire =
2712 TIME_LEQ(dst->expire, time_second) ?
2713 0 : dst->expire - time_second ;
2714 bp += sizeof(ipfw_dyn_rule);
2715 }
2716 }
2717 IPFW_DYN_UNLOCK();
2718 if (last != NULL) /* mark last dynamic rule */
2719 bzero(&last->next, sizeof(last));
2720 }
2721 return (bp - (char *)buf);
2722 }
2723
2724
2725 /**
2726 * {set|get}sockopt parser.
2727 */
2728 static int
2729 ipfw_ctl(struct sockopt *sopt)
2730 {
2731 #define RULE_MAXSIZE (256*sizeof(u_int32_t))
2732 int error, rule_num;
2733 size_t size;
2734 struct ip_fw *buf, *rule;
2735 u_int32_t rulenum[2];
2736
2737 /*
2738 * Disallow modifications in really-really secure mode, but still allow
2739 * the logging counters to be reset.
2740 */
2741 if (sopt->sopt_name == IP_FW_ADD ||
2742 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
2743 #if __FreeBSD_version >= 500034
2744 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
2745 if (error)
2746 return (error);
2747 #else /* FreeBSD 4.x */
2748 if (securelevel >= 3)
2749 return (EPERM);
2750 #endif
2751 }
2752
2753 error = 0;
2754
2755 switch (sopt->sopt_name) {
2756 case IP_FW_GET:
2757 /*
2758 * pass up a copy of the current rules. Static rules
2759 * come first (the last of which has number IPFW_DEFAULT_RULE),
2760 * followed by a possibly empty list of dynamic rule.
2761 * The last dynamic rule has NULL in the "next" field.
2762 *
2763 * Note that the calculated size is used to bound the
2764 * amount of data returned to the user. The rule set may
2765 * change between calculating the size and returning the
2766 * data in which case we'll just return what fits.
2767 */
2768 size = static_len; /* size of static rules */
2769 if (ipfw_dyn_v) /* add size of dyn.rules */
2770 size += (dyn_count * sizeof(ipfw_dyn_rule));
2771
2772 /*
2773 * XXX todo: if the user passes a short length just to know
2774 * how much room is needed, do not bother filling up the
2775 * buffer, just jump to the sooptcopyout.
2776 */
2777 buf = malloc(size, M_TEMP, M_WAITOK);
2778 error = sooptcopyout(sopt, buf,
2779 ipfw_getrules(&layer3_chain, buf, size));
2780 free(buf, M_TEMP);
2781 break;
2782
2783 case IP_FW_FLUSH:
2784 /*
2785 * Normally we cannot release the lock on each iteration.
2786 * We could do it here only because we start from the head all
2787 * the times so there is no risk of missing some entries.
2788 * On the other hand, the risk is that we end up with
2789 * a very inconsistent ruleset, so better keep the lock
2790 * around the whole cycle.
2791 *
2792 * XXX this code can be improved by resetting the head of
2793 * the list to point to the default rule, and then freeing
2794 * the old list without the need for a lock.
2795 */
2796
2797 IPFW_LOCK(&layer3_chain);
2798 layer3_chain.reap = NULL;
2799 free_chain(&layer3_chain, 0 /* keep default rule */);
2800 rule = layer3_chain.reap, layer3_chain.reap = NULL;
2801 IPFW_UNLOCK(&layer3_chain);
2802 if (layer3_chain.reap != NULL)
2803 reap_rules(rule);
2804 break;
2805
2806 case IP_FW_ADD:
2807 rule = malloc(RULE_MAXSIZE, M_TEMP, M_WAITOK);
2808 error = sooptcopyin(sopt, rule, RULE_MAXSIZE,
2809 sizeof(struct ip_fw) );
2810 if (error == 0)
2811 error = check_ipfw_struct(rule, sopt->sopt_valsize);
2812 if (error == 0) {
2813 error = add_rule(&layer3_chain, rule);
2814 size = RULESIZE(rule);
2815 if (!error && sopt->sopt_dir == SOPT_GET)
2816 error = sooptcopyout(sopt, rule, size);
2817 }
2818 free(rule, M_TEMP);
2819 break;
2820
2821 case IP_FW_DEL:
2822 /*
2823 * IP_FW_DEL is used for deleting single rules or sets,
2824 * and (ab)used to atomically manipulate sets. Argument size
2825 * is used to distinguish between the two:
2826 * sizeof(u_int32_t)
2827 * delete single rule or set of rules,
2828 * or reassign rules (or sets) to a different set.
2829 * 2*sizeof(u_int32_t)
2830 * atomic disable/enable sets.
2831 * first u_int32_t contains sets to be disabled,
2832 * second u_int32_t contains sets to be enabled.
2833 */
2834 error = sooptcopyin(sopt, rulenum,
2835 2*sizeof(u_int32_t), sizeof(u_int32_t));
2836 if (error)
2837 break;
2838 size = sopt->sopt_valsize;
2839 if (size == sizeof(u_int32_t)) /* delete or reassign */
2840 error = del_entry(&layer3_chain, rulenum[0]);
2841 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */
2842 set_disable =
2843 (set_disable | rulenum[0]) & ~rulenum[1] &
2844 ~(1<<RESVD_SET); /* set RESVD_SET always enabled */
2845 else
2846 error = EINVAL;
2847 break;
2848
2849 case IP_FW_ZERO:
2850 case IP_FW_RESETLOG: /* argument is an int, the rule number */
2851 rule_num = 0;
2852 if (sopt->sopt_val != 0) {
2853 error = sooptcopyin(sopt, &rule_num,
2854 sizeof(int), sizeof(int));
2855 if (error)
2856 break;
2857 }
2858 error = zero_entry(&layer3_chain, rule_num,
2859 sopt->sopt_name == IP_FW_RESETLOG);
2860 break;
2861
2862 default:
2863 printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name);
2864 error = EINVAL;
2865 }
2866
2867 return (error);
2868 #undef RULE_MAXSIZE
2869 }
2870
2871 /**
2872 * dummynet needs a reference to the default rule, because rules can be
2873 * deleted while packets hold a reference to them. When this happens,
2874 * dummynet changes the reference to the default rule (it could well be a
2875 * NULL pointer, but this way we do not need to check for the special
2876 * case, plus here he have info on the default behaviour).
2877 */
2878 struct ip_fw *ip_fw_default_rule;
2879
2880 /*
2881 * This procedure is only used to handle keepalives. It is invoked
2882 * every dyn_keepalive_period
2883 */
2884 static void
2885 ipfw_tick(void * __unused unused)
2886 {
2887 int i;
2888 ipfw_dyn_rule *q;
2889
2890 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
2891 goto done;
2892
2893 IPFW_DYN_LOCK();
2894 for (i = 0 ; i < curr_dyn_buckets ; i++) {
2895 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
2896 if (q->dyn_type == O_LIMIT_PARENT)
2897 continue;
2898 if (q->id.proto != IPPROTO_TCP)
2899 continue;
2900 if ( (q->state & BOTH_SYN) != BOTH_SYN)
2901 continue;
2902 if (TIME_LEQ( time_second+dyn_keepalive_interval,
2903 q->expire))
2904 continue; /* too early */
2905 if (TIME_LEQ(q->expire, time_second))
2906 continue; /* too late, rule expired */
2907
2908 send_pkt(&(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN);
2909 send_pkt(&(q->id), q->ack_fwd - 1, q->ack_rev, 0);
2910 }
2911 }
2912 IPFW_DYN_UNLOCK();
2913 done:
2914 callout_reset(&ipfw_timeout, dyn_keepalive_period*hz, ipfw_tick, NULL);
2915 }
2916
2917 static int
2918 ipfw_init(void)
2919 {
2920 struct ip_fw default_rule;
2921 int error;
2922
2923 layer3_chain.rules = NULL;
2924 IPFW_LOCK_INIT(&layer3_chain);
2925 IPFW_DYN_LOCK_INIT();
2926 callout_init(&ipfw_timeout, debug_mpsafenet ? CALLOUT_MPSAFE : 0);
2927
2928 bzero(&default_rule, sizeof default_rule);
2929
2930 default_rule.act_ofs = 0;
2931 default_rule.rulenum = IPFW_DEFAULT_RULE;
2932 default_rule.cmd_len = 1;
2933 default_rule.set = RESVD_SET;
2934
2935 default_rule.cmd[0].len = 1;
2936 default_rule.cmd[0].opcode =
2937 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
2938 1 ? O_ACCEPT :
2939 #endif
2940 O_DENY;
2941
2942 error = add_rule(&layer3_chain, &default_rule);
2943 if (error != 0) {
2944 printf("ipfw2: error %u initializing default rule "
2945 "(support disabled)\n", error);
2946 IPFW_DYN_LOCK_DESTROY();
2947 IPFW_LOCK_DESTROY(&layer3_chain);
2948 return (error);
2949 }
2950
2951 ip_fw_default_rule = layer3_chain.rules;
2952 printf("ipfw2 initialized, divert %s, "
2953 "rule-based forwarding enabled, default to %s, logging ",
2954 #ifdef IPDIVERT
2955 "enabled",
2956 #else
2957 "disabled",
2958 #endif
2959 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny");
2960
2961 #ifdef IPFIREWALL_VERBOSE
2962 fw_verbose = 1;
2963 #endif
2964 #ifdef IPFIREWALL_VERBOSE_LIMIT
2965 verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
2966 #endif
2967 if (fw_verbose == 0)
2968 printf("disabled\n");
2969 else if (verbose_limit == 0)
2970 printf("unlimited\n");
2971 else
2972 printf("limited to %d packets/entry by default\n",
2973 verbose_limit);
2974
2975 ip_fw_chk_ptr = ipfw_chk;
2976 ip_fw_ctl_ptr = ipfw_ctl;
2977 callout_reset(&ipfw_timeout, hz, ipfw_tick, NULL);
2978
2979 return (0);
2980 }
2981
2982 static void
2983 ipfw_destroy(void)
2984 {
2985 struct ip_fw *reap;
2986
2987 IPFW_LOCK(&layer3_chain);
2988 callout_stop(&ipfw_timeout);
2989 ip_fw_chk_ptr = NULL;
2990 ip_fw_ctl_ptr = NULL;
2991 layer3_chain.reap = NULL;
2992 free_chain(&layer3_chain, 1 /* kill default rule */);
2993 reap = layer3_chain.reap, layer3_chain.reap = NULL;
2994 IPFW_UNLOCK(&layer3_chain);
2995 if (reap != NULL)
2996 reap_rules(reap);
2997
2998 IPFW_DYN_LOCK_DESTROY();
2999 IPFW_LOCK_DESTROY(&layer3_chain);
3000 printf("IP firewall unloaded\n");
3001 }
3002
3003 static int
3004 ipfw_modevent(module_t mod, int type, void *unused)
3005 {
3006 int err = 0;
3007
3008 switch (type) {
3009 case MOD_LOAD:
3010 if (IPFW_LOADED) {
3011 printf("IP firewall already loaded\n");
3012 err = EEXIST;
3013 } else {
3014 err = ipfw_init();
3015 }
3016 break;
3017
3018 case MOD_UNLOAD:
3019 ipfw_destroy();
3020 err = 0;
3021 break;
3022 default:
3023 break;
3024 }
3025 return err;
3026 }
3027
3028 static moduledata_t ipfwmod = {
3029 "ipfw",
3030 ipfw_modevent,
3031 0
3032 };
3033 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PSEUDO, SI_ORDER_ANY);
3034 MODULE_VERSION(ipfw, 1);
3035 #endif /* IPFW2 */
Cache object: f2a4bc6062b55b3da3e9bcf4039a1051
|