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/jail.h>
54#include <sys/module.h>
55#include <sys/proc.h>
56#include <sys/socket.h>
57#include <sys/socketvar.h>
58#include <sys/sysctl.h>
59#include <sys/syslog.h>
60#include <sys/ucred.h>
61#include <net/if.h>
62#include <net/radix.h>
63#include <net/route.h>
64#include <netinet/in.h>
65#include <netinet/in_systm.h>
66#include <netinet/in_var.h>
67#include <netinet/in_pcb.h>
68#include <netinet/ip.h>
69#include <netinet/ip_var.h>
70#include <netinet/ip_icmp.h>
71#include <netinet/ip_fw.h>
72#include <netinet/ip_divert.h>
73#include <netinet/ip_dummynet.h>
74#include <netinet/tcp.h>
75#include <netinet/tcp_timer.h>
76#include <netinet/tcp_var.h>
77#include <netinet/tcpip.h>
78#include <netinet/udp.h>
79#include <netinet/udp_var.h>
80
81#ifdef IPSEC
82#include <netinet6/ipsec.h>
83#endif
84
85#include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
86
87#include <machine/in_cksum.h> /* XXX for in_cksum */
88
89/*
90 * set_disable contains one bit per set value (0..31).
91 * If the bit is set, all rules with the corresponding set
92 * are disabled. Set RESVD_SET(31) is reserved for the default rule
93 * and rules that are not deleted by the flush command,
94 * and CANNOT be disabled.
95 * Rules in set RESVD_SET can only be deleted explicitly.
96 */
97static u_int32_t set_disable;
98
99static int fw_verbose;
100static int verbose_limit;
101
102static struct callout ipfw_timeout;
103#define IPFW_DEFAULT_RULE 65535
104
105/*
106 * Data structure to cache our ucred related
107 * information. This structure only gets used if
108 * the user specified UID/GID based constraints in
109 * a firewall rule.
110 */
111struct ip_fw_ugid {
112 gid_t fw_groups[NGROUPS];
113 int fw_ngroups;
114 uid_t fw_uid;
115 int fw_prid;
116};
117
118struct ip_fw_chain {
119 struct ip_fw *rules; /* list of rules */
120 struct ip_fw *reap; /* list of rules to reap */
121 struct mtx mtx; /* lock guarding rule list */
122};
123#define IPFW_LOCK_INIT(_chain) \
124 mtx_init(&(_chain)->mtx, "IPFW static rules", NULL, \
125 MTX_DEF | MTX_RECURSE)
126#define IPFW_LOCK_DESTROY(_chain) mtx_destroy(&(_chain)->mtx)
127#define IPFW_LOCK(_chain) mtx_lock(&(_chain)->mtx)
128#define IPFW_UNLOCK(_chain) mtx_unlock(&(_chain)->mtx)
129#define IPFW_LOCK_ASSERT(_chain) do { \
130 mtx_assert(&(_chain)->mtx, MA_OWNED); \
131 NET_ASSERT_GIANT(); \
132} while (0)
133
134/*
135 * list of rules for layer 3
136 */
137static struct ip_fw_chain layer3_chain;
138
139MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
140MALLOC_DEFINE(M_IPFW_TBL, "ipfw_tbl", "IpFw tables");
141
142struct table_entry {
143 struct radix_node rn[2];
144 struct sockaddr_in addr, mask;
145 u_int32_t value;
146};
147
148#define IPFW_TABLES_MAX 128
149static struct {
150 struct radix_node_head *rnh;
151 int modified;
152} ipfw_tables[IPFW_TABLES_MAX];
153
154static int fw_debug = 1;
155static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
156
157#ifdef SYSCTL_NODE
158SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
159SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable,
160 CTLFLAG_RW | CTLFLAG_SECURE3,
161 &fw_enable, 0, "Enable ipfw");
162SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
163 &autoinc_step, 0, "Rule number autincrement step");
164SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, one_pass,
165 CTLFLAG_RW | CTLFLAG_SECURE3,
166 &fw_one_pass, 0,
167 "Only do a single pass through ipfw when using dummynet(4)");
168SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
169 &fw_debug, 0, "Enable printing of debug ip_fw statements");
170SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose,
171 CTLFLAG_RW | CTLFLAG_SECURE3,
172 &fw_verbose, 0, "Log matches to ipfw rules");
173SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
174 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
175
176/*
177 * Description of dynamic rules.
178 *
179 * Dynamic rules are stored in lists accessed through a hash table
180 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
181 * be modified through the sysctl variable dyn_buckets which is
182 * updated when the table becomes empty.
183 *
184 * XXX currently there is only one list, ipfw_dyn.
185 *
186 * When a packet is received, its address fields are first masked
187 * with the mask defined for the rule, then hashed, then matched
188 * against the entries in the corresponding list.
189 * Dynamic rules can be used for different purposes:
190 * + stateful rules;
191 * + enforcing limits on the number of sessions;
192 * + in-kernel NAT (not implemented yet)
193 *
194 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
195 * measured in seconds and depending on the flags.
196 *
197 * The total number of dynamic rules is stored in dyn_count.
198 * The max number of dynamic rules is dyn_max. When we reach
199 * the maximum number of rules we do not create anymore. This is
200 * done to avoid consuming too much memory, but also too much
201 * time when searching on each packet (ideally, we should try instead
202 * to put a limit on the length of the list on each bucket...).
203 *
204 * Each dynamic rule holds a pointer to the parent ipfw rule so
205 * we know what action to perform. Dynamic rules are removed when
206 * the parent rule is deleted. XXX we should make them survive.
207 *
208 * There are some limitations with dynamic rules -- we do not
209 * obey the 'randomized match', and we do not do multiple
210 * passes through the firewall. XXX check the latter!!!
211 */
212static ipfw_dyn_rule **ipfw_dyn_v = NULL;
213static u_int32_t dyn_buckets = 256; /* must be power of 2 */
214static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */
215
216static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */
217#define IPFW_DYN_LOCK_INIT() \
218 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF)
219#define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx)
220#define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx)
221#define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx)
222#define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED)
223
224/*
225 * Timeouts for various events in handing dynamic rules.
226 */
227static u_int32_t dyn_ack_lifetime = 300;
228static u_int32_t dyn_syn_lifetime = 20;
229static u_int32_t dyn_fin_lifetime = 1;
230static u_int32_t dyn_rst_lifetime = 1;
231static u_int32_t dyn_udp_lifetime = 10;
232static u_int32_t dyn_short_lifetime = 5;
233
234/*
235 * Keepalives are sent if dyn_keepalive is set. They are sent every
236 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
237 * seconds of lifetime of a rule.
238 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
239 * than dyn_keepalive_period.
240 */
241
242static u_int32_t dyn_keepalive_interval = 20;
243static u_int32_t dyn_keepalive_period = 5;
244static u_int32_t dyn_keepalive = 1; /* do send keepalives */
245
246static u_int32_t static_count; /* # of static rules */
247static u_int32_t static_len; /* size in bytes of static rules */
248static u_int32_t dyn_count; /* # of dynamic rules */
249static u_int32_t dyn_max = 4096; /* max # of dynamic rules */
250
251SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
252 &dyn_buckets, 0, "Number of dyn. buckets");
253SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
254 &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
255SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
256 &dyn_count, 0, "Number of dyn. rules");
257SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
258 &dyn_max, 0, "Max number of dyn. rules");
259SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
260 &static_count, 0, "Number of static rules");
261SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
262 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
263SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
264 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
265SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
266 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
267SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
268 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
269SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
270 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
271SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
272 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
273SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
274 &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
275
276#endif /* SYSCTL_NODE */
277
278
279/*
280 * This macro maps an ip pointer into a layer3 header pointer of type T
281 */
282#define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
283
284static __inline int
285icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
286{
287 int type = L3HDR(struct icmp,ip)->icmp_type;
288
289 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
290}
291
292#define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
293 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
294
295static int
296is_icmp_query(struct ip *ip)
297{
298 int type = L3HDR(struct icmp, ip)->icmp_type;
299 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
300}
301#undef TT
302
303/*
304 * The following checks use two arrays of 8 or 16 bits to store the
305 * bits that we want set or clear, respectively. They are in the
306 * low and high half of cmd->arg1 or cmd->d[0].
307 *
308 * We scan options and store the bits we find set. We succeed if
309 *
310 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
311 *
312 * The code is sometimes optimized not to store additional variables.
313 */
314
315static int
316flags_match(ipfw_insn *cmd, u_int8_t bits)
317{
318 u_char want_clear;
319 bits = ~bits;
320
321 if ( ((cmd->arg1 & 0xff) & bits) != 0)
322 return 0; /* some bits we want set were clear */
323 want_clear = (cmd->arg1 >> 8) & 0xff;
324 if ( (want_clear & bits) != want_clear)
325 return 0; /* some bits we want clear were set */
326 return 1;
327}
328
329static int
330ipopts_match(struct ip *ip, ipfw_insn *cmd)
331{
332 int optlen, bits = 0;
333 u_char *cp = (u_char *)(ip + 1);
334 int x = (ip->ip_hl << 2) - sizeof (struct ip);
335
336 for (; x > 0; x -= optlen, cp += optlen) {
337 int opt = cp[IPOPT_OPTVAL];
338
339 if (opt == IPOPT_EOL)
340 break;
341 if (opt == IPOPT_NOP)
342 optlen = 1;
343 else {
344 optlen = cp[IPOPT_OLEN];
345 if (optlen <= 0 || optlen > x)
346 return 0; /* invalid or truncated */
347 }
348 switch (opt) {
349
350 default:
351 break;
352
353 case IPOPT_LSRR:
354 bits |= IP_FW_IPOPT_LSRR;
355 break;
356
357 case IPOPT_SSRR:
358 bits |= IP_FW_IPOPT_SSRR;
359 break;
360
361 case IPOPT_RR:
362 bits |= IP_FW_IPOPT_RR;
363 break;
364
365 case IPOPT_TS:
366 bits |= IP_FW_IPOPT_TS;
367 break;
368 }
369 }
370 return (flags_match(cmd, bits));
371}
372
373static int
374tcpopts_match(struct ip *ip, ipfw_insn *cmd)
375{
376 int optlen, bits = 0;
377 struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
378 u_char *cp = (u_char *)(tcp + 1);
379 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
380
381 for (; x > 0; x -= optlen, cp += optlen) {
382 int opt = cp[0];
383 if (opt == TCPOPT_EOL)
384 break;
385 if (opt == TCPOPT_NOP)
386 optlen = 1;
387 else {
388 optlen = cp[1];
389 if (optlen <= 0)
390 break;
391 }
392
393 switch (opt) {
394
395 default:
396 break;
397
398 case TCPOPT_MAXSEG:
399 bits |= IP_FW_TCPOPT_MSS;
400 break;
401
402 case TCPOPT_WINDOW:
403 bits |= IP_FW_TCPOPT_WINDOW;
404 break;
405
406 case TCPOPT_SACK_PERMITTED:
407 case TCPOPT_SACK:
408 bits |= IP_FW_TCPOPT_SACK;
409 break;
410
411 case TCPOPT_TIMESTAMP:
412 bits |= IP_FW_TCPOPT_TS;
413 break;
414
415 case TCPOPT_CC:
416 case TCPOPT_CCNEW:
417 case TCPOPT_CCECHO:
418 bits |= IP_FW_TCPOPT_CC;
419 break;
420 }
421 }
422 return (flags_match(cmd, bits));
423}
424
425static int
426iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
427{
428 if (ifp == NULL) /* no iface with this packet, match fails */
429 return 0;
430 /* Check by name or by IP address */
431 if (cmd->name[0] != '\0') { /* match by name */
432 /* Check name */
433 if (cmd->p.glob) {
434 if (fnmatch(cmd->name, ifp->if_xname, 0) == 0)
435 return(1);
436 } else {
437 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
438 return(1);
439 }
440 } else {
441 struct ifaddr *ia;
442
443 /* XXX lock? */
444 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
445 if (ia->ifa_addr == NULL)
446 continue;
447 if (ia->ifa_addr->sa_family != AF_INET)
448 continue;
449 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
450 (ia->ifa_addr))->sin_addr.s_addr)
451 return(1); /* match */
452 }
453 }
454 return(0); /* no match, fail ... */
455}
456
457/*
458 * The verify_path function checks if a route to the src exists and
459 * if it is reachable via ifp (when provided).
460 *
461 * The 'verrevpath' option checks that the interface that an IP packet
462 * arrives on is the same interface that traffic destined for the
463 * packet's source address would be routed out of. The 'versrcreach'
464 * option just checks that the source address is reachable via any route
465 * (except default) in the routing table. These two are a measure to block
466 * forged packets. This is also commonly known as "anti-spoofing" or Unicast
467 * Reverse Path Forwarding (Unicast RFP) in Cisco-ese. The name of the knobs
468 * is purposely reminiscent of the Cisco IOS command,
469 *
470 * ip verify unicast reverse-path
471 * ip verify unicast source reachable-via any
472 *
473 * which implements the same functionality. But note that syntax is
474 * misleading. The check may be performed on all IP packets whether unicast,
475 * multicast, or broadcast.
476 */
477static int
478verify_path(struct in_addr src, struct ifnet *ifp)
479{
480 struct route ro;
481 struct sockaddr_in *dst;
482
483 bzero(&ro, sizeof(ro));
484
485 dst = (struct sockaddr_in *)&(ro.ro_dst);
486 dst->sin_family = AF_INET;
487 dst->sin_len = sizeof(*dst);
488 dst->sin_addr = src;
489 rtalloc_ign(&ro, RTF_CLONING);
490
491 if (ro.ro_rt == NULL)
492 return 0;
493
494 /* if ifp is provided, check for equality with rtentry */
495 if (ifp != NULL && ro.ro_rt->rt_ifp != ifp) {
496 RTFREE(ro.ro_rt);
497 return 0;
498 }
499
500 /* if no ifp provided, check if rtentry is not default route */
501 if (ifp == NULL &&
502 satosin(rt_key(ro.ro_rt))->sin_addr.s_addr == INADDR_ANY) {
503 RTFREE(ro.ro_rt);
504 return 0;
505 }
506
507 /* or if this is a blackhole/reject route */
508 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
509 RTFREE(ro.ro_rt);
510 return 0;
511 }
512
513 /* found valid route */
514 RTFREE(ro.ro_rt);
515 return 1;
516}
517
518
519static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */
520
521#define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
522#define SNP(buf) buf, sizeof(buf)
523
524/*
525 * We enter here when we have a rule with O_LOG.
526 * XXX this function alone takes about 2Kbytes of code!
527 */
528static void
529ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
530 struct mbuf *m, struct ifnet *oif)
531{
532 char *action;
533 int limit_reached = 0;
534 char action2[40], proto[48], fragment[28];
535
536 fragment[0] = '\0';
537 proto[0] = '\0';
538
539 if (f == NULL) { /* bogus pkt */
540 if (verbose_limit != 0 && norule_counter >= verbose_limit)
541 return;
542 norule_counter++;
543 if (norule_counter == verbose_limit)
544 limit_reached = verbose_limit;
545 action = "Refuse";
546 } else { /* O_LOG is the first action, find the real one */
547 ipfw_insn *cmd = ACTION_PTR(f);
548 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
549
550 if (l->max_log != 0 && l->log_left == 0)
551 return;
552 l->log_left--;
553 if (l->log_left == 0)
554 limit_reached = l->max_log;
555 cmd += F_LEN(cmd); /* point to first action */
556 if (cmd->opcode == O_PROB)
557 cmd += F_LEN(cmd);
558
559 action = action2;
560 switch (cmd->opcode) {
561 case O_DENY:
562 action = "Deny";
563 break;
564
565 case O_REJECT:
566 if (cmd->arg1==ICMP_REJECT_RST)
567 action = "Reset";
568 else if (cmd->arg1==ICMP_UNREACH_HOST)
569 action = "Reject";
570 else
571 snprintf(SNPARGS(action2, 0), "Unreach %d",
572 cmd->arg1);
573 break;
574
575 case O_ACCEPT:
576 action = "Accept";
577 break;
578 case O_COUNT:
579 action = "Count";
580 break;
581 case O_DIVERT:
582 snprintf(SNPARGS(action2, 0), "Divert %d",
583 cmd->arg1);
584 break;
585 case O_TEE:
586 snprintf(SNPARGS(action2, 0), "Tee %d",
587 cmd->arg1);
588 break;
589 case O_SKIPTO:
590 snprintf(SNPARGS(action2, 0), "SkipTo %d",
591 cmd->arg1);
592 break;
593 case O_PIPE:
594 snprintf(SNPARGS(action2, 0), "Pipe %d",
595 cmd->arg1);
596 break;
597 case O_QUEUE:
598 snprintf(SNPARGS(action2, 0), "Queue %d",
599 cmd->arg1);
600 break;
601 case O_FORWARD_IP: {
602 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
603 int len;
604
605 len = snprintf(SNPARGS(action2, 0), "Forward to %s",
606 inet_ntoa(sa->sa.sin_addr));
607 if (sa->sa.sin_port)
608 snprintf(SNPARGS(action2, len), ":%d",
609 sa->sa.sin_port);
610 }
611 break;
612 default:
613 action = "UNKNOWN";
614 break;
615 }
616 }
617
618 if (hlen == 0) { /* non-ip */
619 snprintf(SNPARGS(proto, 0), "MAC");
620 } else {
621 struct ip *ip = mtod(m, struct ip *);
622 /* these three are all aliases to the same thing */
623 struct icmp *const icmp = L3HDR(struct icmp, ip);
624 struct tcphdr *const tcp = (struct tcphdr *)icmp;
625 struct udphdr *const udp = (struct udphdr *)icmp;
626
627 int ip_off, offset, ip_len;
628
629 int len;
630
631 if (eh != NULL) { /* layer 2 packets are as on the wire */
632 ip_off = ntohs(ip->ip_off);
633 ip_len = ntohs(ip->ip_len);
634 } else {
635 ip_off = ip->ip_off;
636 ip_len = ip->ip_len;
637 }
638 offset = ip_off & IP_OFFMASK;
639 switch (ip->ip_p) {
640 case IPPROTO_TCP:
641 len = snprintf(SNPARGS(proto, 0), "TCP %s",
642 inet_ntoa(ip->ip_src));
643 if (offset == 0)
644 snprintf(SNPARGS(proto, len), ":%d %s:%d",
645 ntohs(tcp->th_sport),
646 inet_ntoa(ip->ip_dst),
647 ntohs(tcp->th_dport));
648 else
649 snprintf(SNPARGS(proto, len), " %s",
650 inet_ntoa(ip->ip_dst));
651 break;
652
653 case IPPROTO_UDP:
654 len = snprintf(SNPARGS(proto, 0), "UDP %s",
655 inet_ntoa(ip->ip_src));
656 if (offset == 0)
657 snprintf(SNPARGS(proto, len), ":%d %s:%d",
658 ntohs(udp->uh_sport),
659 inet_ntoa(ip->ip_dst),
660 ntohs(udp->uh_dport));
661 else
662 snprintf(SNPARGS(proto, len), " %s",
663 inet_ntoa(ip->ip_dst));
664 break;
665
666 case IPPROTO_ICMP:
667 if (offset == 0)
668 len = snprintf(SNPARGS(proto, 0),
669 "ICMP:%u.%u ",
670 icmp->icmp_type, icmp->icmp_code);
671 else
672 len = snprintf(SNPARGS(proto, 0), "ICMP ");
673 len += snprintf(SNPARGS(proto, len), "%s",
674 inet_ntoa(ip->ip_src));
675 snprintf(SNPARGS(proto, len), " %s",
676 inet_ntoa(ip->ip_dst));
677 break;
678
679 default:
680 len = snprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
681 inet_ntoa(ip->ip_src));
682 snprintf(SNPARGS(proto, len), " %s",
683 inet_ntoa(ip->ip_dst));
684 break;
685 }
686
687 if (ip_off & (IP_MF | IP_OFFMASK))
688 snprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
689 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
690 offset << 3,
691 (ip_off & IP_MF) ? "+" : "");
692 }
693 if (oif || m->m_pkthdr.rcvif)
694 log(LOG_SECURITY | LOG_INFO,
695 "ipfw: %d %s %s %s via %s%s\n",
696 f ? f->rulenum : -1,
697 action, proto, oif ? "out" : "in",
698 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
699 fragment);
700 else
701 log(LOG_SECURITY | LOG_INFO,
702 "ipfw: %d %s %s [no if info]%s\n",
703 f ? f->rulenum : -1,
704 action, proto, fragment);
705 if (limit_reached)
706 log(LOG_SECURITY | LOG_NOTICE,
707 "ipfw: limit %d reached on entry %d\n",
708 limit_reached, f ? f->rulenum : -1);
709}
710
711/*
712 * IMPORTANT: the hash function for dynamic rules must be commutative
713 * in source and destination (ip,port), because rules are bidirectional
714 * and we want to find both in the same bucket.
715 */
716static __inline int
717hash_packet(struct ipfw_flow_id *id)
718{
719 u_int32_t i;
720
721 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
722 i &= (curr_dyn_buckets - 1);
723 return i;
724}
725
726/**
727 * unlink a dynamic rule from a chain. prev is a pointer to
728 * the previous one, q is a pointer to the rule to delete,
729 * head is a pointer to the head of the queue.
730 * Modifies q and potentially also head.
731 */
732#define UNLINK_DYN_RULE(prev, head, q) { \
733 ipfw_dyn_rule *old_q = q; \
734 \
735 /* remove a refcount to the parent */ \
736 if (q->dyn_type == O_LIMIT) \
737 q->parent->count--; \
738 DEB(printf("ipfw: unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",\
739 (q->id.src_ip), (q->id.src_port), \
740 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \
741 if (prev != NULL) \
742 prev->next = q = q->next; \
743 else \
744 head = q = q->next; \
745 dyn_count--; \
746 free(old_q, M_IPFW); }
747
748#define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
749
750/**
751 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
752 *
753 * If keep_me == NULL, rules are deleted even if not expired,
754 * otherwise only expired rules are removed.
755 *
756 * The value of the second parameter is also used to point to identify
757 * a rule we absolutely do not want to remove (e.g. because we are
758 * holding a reference to it -- this is the case with O_LIMIT_PARENT
759 * rules). The pointer is only used for comparison, so any non-null
760 * value will do.
761 */
762static void
763remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
764{
765 static u_int32_t last_remove = 0;
766
767#define FORCE (keep_me == NULL)
768
769 ipfw_dyn_rule *prev, *q;
770 int i, pass = 0, max_pass = 0;
771
772 IPFW_DYN_LOCK_ASSERT();
773
774 if (ipfw_dyn_v == NULL || dyn_count == 0)
775 return;
776 /* do not expire more than once per second, it is useless */
777 if (!FORCE && last_remove == time_second)
778 return;
779 last_remove = time_second;
780
781 /*
782 * because O_LIMIT refer to parent rules, during the first pass only
783 * remove child and mark any pending LIMIT_PARENT, and remove
784 * them in a second pass.
785 */
786next_pass:
787 for (i = 0 ; i < curr_dyn_buckets ; i++) {
788 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
789 /*
790 * Logic can become complex here, so we split tests.
791 */
792 if (q == keep_me)
793 goto next;
794 if (rule != NULL && rule != q->rule)
795 goto next; /* not the one we are looking for */
796 if (q->dyn_type == O_LIMIT_PARENT) {
797 /*
798 * handle parent in the second pass,
799 * record we need one.
800 */
801 max_pass = 1;
802 if (pass == 0)
803 goto next;
804 if (FORCE && q->count != 0 ) {
805 /* XXX should not happen! */
806 printf("ipfw: OUCH! cannot remove rule,"
807 " count %d\n", q->count);
808 }
809 } else {
810 if (!FORCE &&
811 !TIME_LEQ( q->expire, time_second ))
812 goto next;
813 }
814 if (q->dyn_type != O_LIMIT_PARENT || !q->count) {
815 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
816 continue;
817 }
818next:
819 prev=q;
820 q=q->next;
821 }
822 }
823 if (pass++ < max_pass)
824 goto next_pass;
825}
826
827
828/**
829 * lookup a dynamic rule.
830 */
831static ipfw_dyn_rule *
832lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction,
833 struct tcphdr *tcp)
834{
835 /*
836 * stateful ipfw extensions.
837 * Lookup into dynamic session queue
838 */
839#define MATCH_REVERSE 0
840#define MATCH_FORWARD 1
841#define MATCH_NONE 2
842#define MATCH_UNKNOWN 3
843 int i, dir = MATCH_NONE;
844 ipfw_dyn_rule *prev, *q=NULL;
845
846 IPFW_DYN_LOCK_ASSERT();
847
848 if (ipfw_dyn_v == NULL)
849 goto done; /* not found */
850 i = hash_packet( pkt );
851 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
852 if (q->dyn_type == O_LIMIT_PARENT && q->count)
853 goto next;
854 if (TIME_LEQ( q->expire, time_second)) { /* expire entry */
855 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
856 continue;
857 }
858 if (pkt->proto == q->id.proto &&
859 q->dyn_type != O_LIMIT_PARENT) {
860 if (pkt->src_ip == q->id.src_ip &&
861 pkt->dst_ip == q->id.dst_ip &&
862 pkt->src_port == q->id.src_port &&
863 pkt->dst_port == q->id.dst_port ) {
864 dir = MATCH_FORWARD;
865 break;
866 }
867 if (pkt->src_ip == q->id.dst_ip &&
868 pkt->dst_ip == q->id.src_ip &&
869 pkt->src_port == q->id.dst_port &&
870 pkt->dst_port == q->id.src_port ) {
871 dir = MATCH_REVERSE;
872 break;
873 }
874 }
875next:
876 prev = q;
877 q = q->next;
878 }
879 if (q == NULL)
880 goto done; /* q = NULL, not found */
881
882 if ( prev != NULL) { /* found and not in front */
883 prev->next = q->next;
884 q->next = ipfw_dyn_v[i];
885 ipfw_dyn_v[i] = q;
886 }
887 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
888 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
889
890#define BOTH_SYN (TH_SYN | (TH_SYN << 8))
891#define BOTH_FIN (TH_FIN | (TH_FIN << 8))
892 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
893 switch (q->state) {
894 case TH_SYN: /* opening */
895 q->expire = time_second + dyn_syn_lifetime;
896 break;
897
898 case BOTH_SYN: /* move to established */
899 case BOTH_SYN | TH_FIN : /* one side tries to close */
900 case BOTH_SYN | (TH_FIN << 8) :
901 if (tcp) {
902#define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
903 u_int32_t ack = ntohl(tcp->th_ack);
904 if (dir == MATCH_FORWARD) {
905 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
906 q->ack_fwd = ack;
907 else { /* ignore out-of-sequence */
908 break;
909 }
910 } else {
911 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
912 q->ack_rev = ack;
913 else { /* ignore out-of-sequence */
914 break;
915 }
916 }
917 }
918 q->expire = time_second + dyn_ack_lifetime;
919 break;
920
921 case BOTH_SYN | BOTH_FIN: /* both sides closed */
922 if (dyn_fin_lifetime >= dyn_keepalive_period)
923 dyn_fin_lifetime = dyn_keepalive_period - 1;
924 q->expire = time_second + dyn_fin_lifetime;
925 break;
926
927 default:
928#if 0
929 /*
930 * reset or some invalid combination, but can also
931 * occur if we use keep-state the wrong way.
932 */
933 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
934 printf("invalid state: 0x%x\n", q->state);
935#endif
936 if (dyn_rst_lifetime >= dyn_keepalive_period)
937 dyn_rst_lifetime = dyn_keepalive_period - 1;
938 q->expire = time_second + dyn_rst_lifetime;
939 break;
940 }
941 } else if (pkt->proto == IPPROTO_UDP) {
942 q->expire = time_second + dyn_udp_lifetime;
943 } else {
944 /* other protocols */
945 q->expire = time_second + dyn_short_lifetime;
946 }
947done:
948 if (match_direction)
949 *match_direction = dir;
950 return q;
951}
952
953static ipfw_dyn_rule *
954lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
955 struct tcphdr *tcp)
956{
957 ipfw_dyn_rule *q;
958
959 IPFW_DYN_LOCK();
960 q = lookup_dyn_rule_locked(pkt, match_direction, tcp);
961 if (q == NULL)
962 IPFW_DYN_UNLOCK();
963 /* NB: return table locked when q is not NULL */
964 return q;
965}
966
967static void
968realloc_dynamic_table(void)
969{
970 IPFW_DYN_LOCK_ASSERT();
971
972 /*
973 * Try reallocation, make sure we have a power of 2 and do
974 * not allow more than 64k entries. In case of overflow,
975 * default to 1024.
976 */
977
978 if (dyn_buckets > 65536)
979 dyn_buckets = 1024;
980 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
981 dyn_buckets = curr_dyn_buckets; /* reset */
982 return;
983 }
984 curr_dyn_buckets = dyn_buckets;
985 if (ipfw_dyn_v != NULL)
986 free(ipfw_dyn_v, M_IPFW);
987 for (;;) {
988 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
989 M_IPFW, M_NOWAIT | M_ZERO);
990 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
991 break;
992 curr_dyn_buckets /= 2;
993 }
994}
995
996/**
997 * Install state of type 'type' for a dynamic session.
998 * The hash table contains two type of rules:
999 * - regular rules (O_KEEP_STATE)
1000 * - rules for sessions with limited number of sess per user
1001 * (O_LIMIT). When they are created, the parent is
1002 * increased by 1, and decreased on delete. In this case,
1003 * the third parameter is the parent rule and not the chain.
1004 * - "parent" rules for the above (O_LIMIT_PARENT).
1005 */
1006static ipfw_dyn_rule *
1007add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
1008{
1009 ipfw_dyn_rule *r;
1010 int i;
1011
1012 IPFW_DYN_LOCK_ASSERT();
1013
1014 if (ipfw_dyn_v == NULL ||
1015 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1016 realloc_dynamic_table();
1017 if (ipfw_dyn_v == NULL)
1018 return NULL; /* failed ! */
1019 }
1020 i = hash_packet(id);
1021
1022 r = malloc(sizeof *r, M_IPFW, M_NOWAIT | M_ZERO);
1023 if (r == NULL) {
1024 printf ("ipfw: sorry cannot allocate state\n");
1025 return NULL;
1026 }
1027
1028 /* increase refcount on parent, and set pointer */
1029 if (dyn_type == O_LIMIT) {
1030 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1031 if ( parent->dyn_type != O_LIMIT_PARENT)
1032 panic("invalid parent");
1033 parent->count++;
1034 r->parent = parent;
1035 rule = parent->rule;
1036 }
1037
1038 r->id = *id;
1039 r->expire = time_second + dyn_syn_lifetime;
1040 r->rule = rule;
1041 r->dyn_type = dyn_type;
1042 r->pcnt = r->bcnt = 0;
1043 r->count = 0;
1044
1045 r->bucket = i;
1046 r->next = ipfw_dyn_v[i];
1047 ipfw_dyn_v[i] = r;
1048 dyn_count++;
1049 DEB(printf("ipfw: add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1050 dyn_type,
1051 (r->id.src_ip), (r->id.src_port),
1052 (r->id.dst_ip), (r->id.dst_port),
1053 dyn_count ); )
1054 return r;
1055}
1056
1057/**
1058 * lookup dynamic parent rule using pkt and rule as search keys.
1059 * If the lookup fails, then install one.
1060 */
1061static ipfw_dyn_rule *
1062lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1063{
1064 ipfw_dyn_rule *q;
1065 int i;
1066
1067 IPFW_DYN_LOCK_ASSERT();
1068
1069 if (ipfw_dyn_v) {
1070 i = hash_packet( pkt );
1071 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
1072 if (q->dyn_type == O_LIMIT_PARENT &&
1073 rule== q->rule &&
1074 pkt->proto == q->id.proto &&
1075 pkt->src_ip == q->id.src_ip &&
1076 pkt->dst_ip == q->id.dst_ip &&
1077 pkt->src_port == q->id.src_port &&
1078 pkt->dst_port == q->id.dst_port) {
1079 q->expire = time_second + dyn_short_lifetime;
1080 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);)
1081 return q;
1082 }
1083 }
1084 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1085}
1086
1087/**
1088 * Install dynamic state for rule type cmd->o.opcode
1089 *
1090 * Returns 1 (failure) if state is not installed because of errors or because
1091 * session limitations are enforced.
1092 */
1093static int
1094install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1095 struct ip_fw_args *args)
1096{
1097 static int last_log;
1098
1099 ipfw_dyn_rule *q;
1100
1101 DEB(printf("ipfw: install state type %d 0x%08x %u -> 0x%08x %u\n",
1102 cmd->o.opcode,
1103 (args->f_id.src_ip), (args->f_id.src_port),
1104 (args->f_id.dst_ip), (args->f_id.dst_port) );)
1105
1106 IPFW_DYN_LOCK();
1107
1108 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
1109
1110 if (q != NULL) { /* should never occur */
1111 if (last_log != time_second) {
1112 last_log = time_second;
1113 printf("ipfw: install_state: entry already present, done\n");
1114 }
1115 IPFW_DYN_UNLOCK();
1116 return 0;
1117 }
1118
1119 if (dyn_count >= dyn_max)
1120 /*
1121 * Run out of slots, try to remove any expired rule.
1122 */
1123 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1124
1125 if (dyn_count >= dyn_max) {
1126 if (last_log != time_second) {
1127 last_log = time_second;
1128 printf("ipfw: install_state: Too many dynamic rules\n");
1129 }
1130 IPFW_DYN_UNLOCK();
1131 return 1; /* cannot install, notify caller */
1132 }
1133
1134 switch (cmd->o.opcode) {
1135 case O_KEEP_STATE: /* bidir rule */
1136 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1137 break;
1138
1139 case O_LIMIT: /* limit number of sessions */
1140 {
1141 u_int16_t limit_mask = cmd->limit_mask;
1142 struct ipfw_flow_id id;
1143 ipfw_dyn_rule *parent;
1144
1145 DEB(printf("ipfw: installing dyn-limit rule %d\n",
1146 cmd->conn_limit);)
1147
1148 id.dst_ip = id.src_ip = 0;
1149 id.dst_port = id.src_port = 0;
1150 id.proto = args->f_id.proto;
1151
1152 if (limit_mask & DYN_SRC_ADDR)
1153 id.src_ip = args->f_id.src_ip;
1154 if (limit_mask & DYN_DST_ADDR)
1155 id.dst_ip = args->f_id.dst_ip;
1156 if (limit_mask & DYN_SRC_PORT)
1157 id.src_port = args->f_id.src_port;
1158 if (limit_mask & DYN_DST_PORT)
1159 id.dst_port = args->f_id.dst_port;
1160 parent = lookup_dyn_parent(&id, rule);
1161 if (parent == NULL) {
1162 printf("ipfw: add parent failed\n");
1163 return 1;
1164 }
1165 if (parent->count >= cmd->conn_limit) {
1166 /*
1167 * See if we can remove some expired rule.
1168 */
1169 remove_dyn_rule(rule, parent);
1170 if (parent->count >= cmd->conn_limit) {
1171 if (fw_verbose && last_log != time_second) {
1172 last_log = time_second;
1173 log(LOG_SECURITY | LOG_DEBUG,
1174 "drop session, too many entries\n");
1175 }
1176 IPFW_DYN_UNLOCK();
1177 return 1;
1178 }
1179 }
1180 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1181 }
1182 break;
1183 default:
1184 printf("ipfw: unknown dynamic rule type %u\n", cmd->o.opcode);
1185 IPFW_DYN_UNLOCK();
1186 return 1;
1187 }
1188 lookup_dyn_rule_locked(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1189 IPFW_DYN_UNLOCK();
1190 return 0;
1191}
1192
1193/*
1194 * Transmit a TCP packet, containing either a RST or a keepalive.
1195 * When flags & TH_RST, we are sending a RST packet, because of a
1196 * "reset" action matched the packet.
1197 * Otherwise we are sending a keepalive, and flags & TH_
1198 */
1199static void
1200send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags)
1201{
1202 struct mbuf *m;
1203 struct ip *ip;
1204 struct tcphdr *tcp;
1205
1206 MGETHDR(m, M_DONTWAIT, MT_HEADER);
1207 if (m == 0)
1208 return;
1209 m->m_pkthdr.rcvif = (struct ifnet *)0;
1210 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1211 m->m_data += max_linkhdr;
1212
1213 ip = mtod(m, struct ip *);
1214 bzero(ip, m->m_len);
1215 tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1216 ip->ip_p = IPPROTO_TCP;
1217 tcp->th_off = 5;
1218 /*
1219 * Assume we are sending a RST (or a keepalive in the reverse
1220 * direction), swap src and destination addresses and ports.
1221 */
1222 ip->ip_src.s_addr = htonl(id->dst_ip);
1223 ip->ip_dst.s_addr = htonl(id->src_ip);
1224 tcp->th_sport = htons(id->dst_port);
1225 tcp->th_dport = htons(id->src_port);
1226 if (flags & TH_RST) { /* we are sending a RST */
1227 if (flags & TH_ACK) {
1228 tcp->th_seq = htonl(ack);
1229 tcp->th_ack = htonl(0);
1230 tcp->th_flags = TH_RST;
1231 } else {
1232 if (flags & TH_SYN)
1233 seq++;
1234 tcp->th_seq = htonl(0);
1235 tcp->th_ack = htonl(seq);
1236 tcp->th_flags = TH_RST | TH_ACK;
1237 }
1238 } else {
1239 /*
1240 * We are sending a keepalive. flags & TH_SYN determines
1241 * the direction, forward if set, reverse if clear.
1242 * NOTE: seq and ack are always assumed to be correct
1243 * as set by the caller. This may be confusing...
1244 */
1245 if (flags & TH_SYN) {
1246 /*
1247 * we have to rewrite the correct addresses!
1248 */
1249 ip->ip_dst.s_addr = htonl(id->dst_ip);
1250 ip->ip_src.s_addr = htonl(id->src_ip);
1251 tcp->th_dport = htons(id->dst_port);
1252 tcp->th_sport = htons(id->src_port);
1253 }
1254 tcp->th_seq = htonl(seq);
1255 tcp->th_ack = htonl(ack);
1256 tcp->th_flags = TH_ACK;
1257 }
1258 /*
1259 * set ip_len to the payload size so we can compute
1260 * the tcp checksum on the pseudoheader
1261 * XXX check this, could save a couple of words ?
1262 */
1263 ip->ip_len = htons(sizeof(struct tcphdr));
1264 tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1265 /*
1266 * now fill fields left out earlier
1267 */
1268 ip->ip_ttl = ip_defttl;
1269 ip->ip_len = m->m_pkthdr.len;
1270 m->m_flags |= M_SKIP_FIREWALL;
1271 ip_output(m, NULL, NULL, 0, NULL, NULL);
1272}
1273
1274/*
1275 * sends a reject message, consuming the mbuf passed as an argument.
1276 */
1277static void
1278send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1279{
1280
1281 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1282 /* We need the IP header in host order for icmp_error(). */
1283 if (args->eh != NULL) {
1284 struct ip *ip = mtod(args->m, struct ip *);
1285 ip->ip_len = ntohs(ip->ip_len);
1286 ip->ip_off = ntohs(ip->ip_off);
1287 }
1288 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1289 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1290 struct tcphdr *const tcp =
1291 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1292 if ( (tcp->th_flags & TH_RST) == 0)
1293 send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1294 ntohl(tcp->th_ack),
1295 tcp->th_flags | TH_RST);
1296 m_freem(args->m);
1297 } else
1298 m_freem(args->m);
1299 args->m = NULL;
1300}
1301
1302/**
1303 *
1304 * Given an ip_fw *, lookup_next_rule will return a pointer
1305 * to the next rule, which can be either the jump
1306 * target (for skipto instructions) or the next one in the list (in
1307 * all other cases including a missing jump target).
1308 * The result is also written in the "next_rule" field of the rule.
1309 * Backward jumps are not allowed, so start looking from the next
1310 * rule...
1311 *
1312 * This never returns NULL -- in case we do not have an exact match,
1313 * the next rule is returned. When the ruleset is changed,
1314 * pointers are flushed so we are always correct.
1315 */
1316
1317static struct ip_fw *
1318lookup_next_rule(struct ip_fw *me)
1319{
1320 struct ip_fw *rule = NULL;
1321 ipfw_insn *cmd;
1322
1323 /* look for action, in case it is a skipto */
1324 cmd = ACTION_PTR(me);
1325 if (cmd->opcode == O_LOG)
1326 cmd += F_LEN(cmd);
1327 if ( cmd->opcode == O_SKIPTO )
1328 for (rule = me->next; rule ; rule = rule->next)
1329 if (rule->rulenum >= cmd->arg1)
1330 break;
1331 if (rule == NULL) /* failure or not a skipto */
1332 rule = me->next;
1333 me->next_rule = rule;
1334 return rule;
1335}
1336
1337static void
1338init_tables(void)
1339{
1340 int i;
1341
1342 for (i = 0; i < IPFW_TABLES_MAX; i++) {
1343 rn_inithead((void **)&ipfw_tables[i].rnh, 32);
1344 ipfw_tables[i].modified = 1;
1345 }
1346}
1347
1348static int
1349add_table_entry(u_int16_t tbl, in_addr_t addr, u_int8_t mlen, u_int32_t value)
1350{
1351 struct radix_node_head *rnh;
1352 struct table_entry *ent;
1353
1354 if (tbl >= IPFW_TABLES_MAX)
1355 return (EINVAL);
1356 rnh = ipfw_tables[tbl].rnh;
1357 ent = malloc(sizeof(*ent), M_IPFW_TBL, M_NOWAIT | M_ZERO);
1358 if (ent == NULL)
1359 return (ENOMEM);
1360 ent->value = value;
1361 ent->addr.sin_len = ent->mask.sin_len = 8;
1362 ent->mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1363 ent->addr.sin_addr.s_addr = addr & ent->mask.sin_addr.s_addr;
1364 RADIX_NODE_HEAD_LOCK(rnh);
1365 if (rnh->rnh_addaddr(&ent->addr, &ent->mask, rnh, (void *)ent) ==
1366 NULL) {
1367 RADIX_NODE_HEAD_UNLOCK(rnh);
1368 free(ent, M_IPFW_TBL);
1369 return (EEXIST);
1370 }
1371 ipfw_tables[tbl].modified = 1;
1372 RADIX_NODE_HEAD_UNLOCK(rnh);
1373 return (0);
1374}
1375
1376static int
1377del_table_entry(u_int16_t tbl, in_addr_t addr, u_int8_t mlen)
1378{
1379 struct radix_node_head *rnh;
1380 struct table_entry *ent;
1381 struct sockaddr_in sa, mask;
1382
1383 if (tbl >= IPFW_TABLES_MAX)
1384 return (EINVAL);
1385 rnh = ipfw_tables[tbl].rnh;
1386 sa.sin_len = mask.sin_len = 8;
1387 mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1388 sa.sin_addr.s_addr = addr & mask.sin_addr.s_addr;
1389 RADIX_NODE_HEAD_LOCK(rnh);
1390 ent = (struct table_entry *)rnh->rnh_deladdr(&sa, &mask, rnh);
1391 if (ent == NULL) {
1392 RADIX_NODE_HEAD_UNLOCK(rnh);
1393 return (ESRCH);
1394 }
1395 ipfw_tables[tbl].modified = 1;
1396 RADIX_NODE_HEAD_UNLOCK(rnh);
1397 free(ent, M_IPFW_TBL);
1398 return (0);
1399}
1400
1401static int
1402flush_table_entry(struct radix_node *rn, void *arg)
1403{
1404 struct radix_node_head * const rnh = arg;
1405 struct table_entry *ent;
1406
1407 ent = (struct table_entry *)
1408 rnh->rnh_deladdr(rn->rn_key, rn->rn_mask, rnh);
1409 if (ent != NULL)
1410 free(ent, M_IPFW_TBL);
1411 return (0);
1412}
1413
1414static int
1415flush_table(u_int16_t tbl)
1416{
1417 struct radix_node_head *rnh;
1418
1419 if (tbl >= IPFW_TABLES_MAX)
1420 return (EINVAL);
1421 rnh = ipfw_tables[tbl].rnh;
1422 RADIX_NODE_HEAD_LOCK(rnh);
1423 rnh->rnh_walktree(rnh, flush_table_entry, rnh);
1424 ipfw_tables[tbl].modified = 1;
1425 RADIX_NODE_HEAD_UNLOCK(rnh);
1426 return (0);
1427}
1428
1429static void
1430flush_tables(void)
1431{
1432 u_int16_t tbl;
1433
1434 for (tbl = 0; tbl < IPFW_TABLES_MAX; tbl++)
1435 flush_table(tbl);
1436}
1437
1438static int
1439lookup_table(u_int16_t tbl, in_addr_t addr, u_int32_t *val)
1440{
1441 struct radix_node_head *rnh;
1442 struct table_entry *ent;
1443 struct sockaddr_in sa;
1444 static in_addr_t last_addr;
1445 static int last_tbl;
1446 static int last_match;
1447 static u_int32_t last_value;
1448
1449 if (tbl >= IPFW_TABLES_MAX)
1450 return (0);
1451 if (tbl == last_tbl && addr == last_addr &&
1452 !ipfw_tables[tbl].modified) {
1453 if (last_match)
1454 *val = last_value;
1455 return (last_match);
1456 }
1457 rnh = ipfw_tables[tbl].rnh;
1458 sa.sin_len = 8;
1459 sa.sin_addr.s_addr = addr;
1460 RADIX_NODE_HEAD_LOCK(rnh);
1461 ipfw_tables[tbl].modified = 0;
1462 ent = (struct table_entry *)(rnh->rnh_lookup(&sa, NULL, rnh));
1463 RADIX_NODE_HEAD_UNLOCK(rnh);
1464 last_addr = addr;
1465 last_tbl = tbl;
1466 if (ent != NULL) {
1467 last_value = *val = ent->value;
1468 last_match = 1;
1469 return (1);
1470 }
1471 last_match = 0;
1472 return (0);
1473}
1474
1475static int
1476count_table_entry(struct radix_node *rn, void *arg)
1477{
1478 u_int32_t * const cnt = arg;
1479
1480 (*cnt)++;
1481 return (0);
1482}
1483
1484static int
1485count_table(u_int32_t tbl, u_int32_t *cnt)
1486{
1487 struct radix_node_head *rnh;
1488
1489 if (tbl >= IPFW_TABLES_MAX)
1490 return (EINVAL);
1491 rnh = ipfw_tables[tbl].rnh;
1492 *cnt = 0;
1493 RADIX_NODE_HEAD_LOCK(rnh);
1494 rnh->rnh_walktree(rnh, count_table_entry, cnt);
1495 RADIX_NODE_HEAD_UNLOCK(rnh);
1496 return (0);
1497}
1498
1499static int
1500dump_table_entry(struct radix_node *rn, void *arg)
1501{
1502 struct table_entry * const n = (struct table_entry *)rn;
1503 ipfw_table * const tbl = arg;
1504 ipfw_table_entry *ent;
1505
1506 if (tbl->cnt == tbl->size)
1507 return (1);
1508 ent = &tbl->ent[tbl->cnt];
1509 ent->tbl = tbl->tbl;
1510 if (in_nullhost(n->mask.sin_addr))
1511 ent->masklen = 0;
1512 else
1513 ent->masklen = 33 - ffs(ntohl(n->mask.sin_addr.s_addr));
1514 ent->addr = n->addr.sin_addr.s_addr;
1515 ent->value = n->value;
1516 tbl->cnt++;
1517 return (0);
1518}
1519
1520static int
1521dump_table(ipfw_table *tbl)
1522{
1523 struct radix_node_head *rnh;
1524
1525 if (tbl->tbl >= IPFW_TABLES_MAX)
1526 return (EINVAL);
1527 rnh = ipfw_tables[tbl->tbl].rnh;
1528 tbl->cnt = 0;
1529 RADIX_NODE_HEAD_LOCK(rnh);
1530 rnh->rnh_walktree(rnh, dump_table_entry, tbl);
1531 RADIX_NODE_HEAD_UNLOCK(rnh);
1532 return (0);
1533}
1534
| 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/jail.h>
54#include <sys/module.h>
55#include <sys/proc.h>
56#include <sys/socket.h>
57#include <sys/socketvar.h>
58#include <sys/sysctl.h>
59#include <sys/syslog.h>
60#include <sys/ucred.h>
61#include <net/if.h>
62#include <net/radix.h>
63#include <net/route.h>
64#include <netinet/in.h>
65#include <netinet/in_systm.h>
66#include <netinet/in_var.h>
67#include <netinet/in_pcb.h>
68#include <netinet/ip.h>
69#include <netinet/ip_var.h>
70#include <netinet/ip_icmp.h>
71#include <netinet/ip_fw.h>
72#include <netinet/ip_divert.h>
73#include <netinet/ip_dummynet.h>
74#include <netinet/tcp.h>
75#include <netinet/tcp_timer.h>
76#include <netinet/tcp_var.h>
77#include <netinet/tcpip.h>
78#include <netinet/udp.h>
79#include <netinet/udp_var.h>
80
81#ifdef IPSEC
82#include <netinet6/ipsec.h>
83#endif
84
85#include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
86
87#include <machine/in_cksum.h> /* XXX for in_cksum */
88
89/*
90 * set_disable contains one bit per set value (0..31).
91 * If the bit is set, all rules with the corresponding set
92 * are disabled. Set RESVD_SET(31) is reserved for the default rule
93 * and rules that are not deleted by the flush command,
94 * and CANNOT be disabled.
95 * Rules in set RESVD_SET can only be deleted explicitly.
96 */
97static u_int32_t set_disable;
98
99static int fw_verbose;
100static int verbose_limit;
101
102static struct callout ipfw_timeout;
103#define IPFW_DEFAULT_RULE 65535
104
105/*
106 * Data structure to cache our ucred related
107 * information. This structure only gets used if
108 * the user specified UID/GID based constraints in
109 * a firewall rule.
110 */
111struct ip_fw_ugid {
112 gid_t fw_groups[NGROUPS];
113 int fw_ngroups;
114 uid_t fw_uid;
115 int fw_prid;
116};
117
118struct ip_fw_chain {
119 struct ip_fw *rules; /* list of rules */
120 struct ip_fw *reap; /* list of rules to reap */
121 struct mtx mtx; /* lock guarding rule list */
122};
123#define IPFW_LOCK_INIT(_chain) \
124 mtx_init(&(_chain)->mtx, "IPFW static rules", NULL, \
125 MTX_DEF | MTX_RECURSE)
126#define IPFW_LOCK_DESTROY(_chain) mtx_destroy(&(_chain)->mtx)
127#define IPFW_LOCK(_chain) mtx_lock(&(_chain)->mtx)
128#define IPFW_UNLOCK(_chain) mtx_unlock(&(_chain)->mtx)
129#define IPFW_LOCK_ASSERT(_chain) do { \
130 mtx_assert(&(_chain)->mtx, MA_OWNED); \
131 NET_ASSERT_GIANT(); \
132} while (0)
133
134/*
135 * list of rules for layer 3
136 */
137static struct ip_fw_chain layer3_chain;
138
139MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
140MALLOC_DEFINE(M_IPFW_TBL, "ipfw_tbl", "IpFw tables");
141
142struct table_entry {
143 struct radix_node rn[2];
144 struct sockaddr_in addr, mask;
145 u_int32_t value;
146};
147
148#define IPFW_TABLES_MAX 128
149static struct {
150 struct radix_node_head *rnh;
151 int modified;
152} ipfw_tables[IPFW_TABLES_MAX];
153
154static int fw_debug = 1;
155static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
156
157#ifdef SYSCTL_NODE
158SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
159SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable,
160 CTLFLAG_RW | CTLFLAG_SECURE3,
161 &fw_enable, 0, "Enable ipfw");
162SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
163 &autoinc_step, 0, "Rule number autincrement step");
164SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, one_pass,
165 CTLFLAG_RW | CTLFLAG_SECURE3,
166 &fw_one_pass, 0,
167 "Only do a single pass through ipfw when using dummynet(4)");
168SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
169 &fw_debug, 0, "Enable printing of debug ip_fw statements");
170SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose,
171 CTLFLAG_RW | CTLFLAG_SECURE3,
172 &fw_verbose, 0, "Log matches to ipfw rules");
173SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
174 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
175
176/*
177 * Description of dynamic rules.
178 *
179 * Dynamic rules are stored in lists accessed through a hash table
180 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
181 * be modified through the sysctl variable dyn_buckets which is
182 * updated when the table becomes empty.
183 *
184 * XXX currently there is only one list, ipfw_dyn.
185 *
186 * When a packet is received, its address fields are first masked
187 * with the mask defined for the rule, then hashed, then matched
188 * against the entries in the corresponding list.
189 * Dynamic rules can be used for different purposes:
190 * + stateful rules;
191 * + enforcing limits on the number of sessions;
192 * + in-kernel NAT (not implemented yet)
193 *
194 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
195 * measured in seconds and depending on the flags.
196 *
197 * The total number of dynamic rules is stored in dyn_count.
198 * The max number of dynamic rules is dyn_max. When we reach
199 * the maximum number of rules we do not create anymore. This is
200 * done to avoid consuming too much memory, but also too much
201 * time when searching on each packet (ideally, we should try instead
202 * to put a limit on the length of the list on each bucket...).
203 *
204 * Each dynamic rule holds a pointer to the parent ipfw rule so
205 * we know what action to perform. Dynamic rules are removed when
206 * the parent rule is deleted. XXX we should make them survive.
207 *
208 * There are some limitations with dynamic rules -- we do not
209 * obey the 'randomized match', and we do not do multiple
210 * passes through the firewall. XXX check the latter!!!
211 */
212static ipfw_dyn_rule **ipfw_dyn_v = NULL;
213static u_int32_t dyn_buckets = 256; /* must be power of 2 */
214static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */
215
216static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */
217#define IPFW_DYN_LOCK_INIT() \
218 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF)
219#define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx)
220#define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx)
221#define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx)
222#define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED)
223
224/*
225 * Timeouts for various events in handing dynamic rules.
226 */
227static u_int32_t dyn_ack_lifetime = 300;
228static u_int32_t dyn_syn_lifetime = 20;
229static u_int32_t dyn_fin_lifetime = 1;
230static u_int32_t dyn_rst_lifetime = 1;
231static u_int32_t dyn_udp_lifetime = 10;
232static u_int32_t dyn_short_lifetime = 5;
233
234/*
235 * Keepalives are sent if dyn_keepalive is set. They are sent every
236 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
237 * seconds of lifetime of a rule.
238 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
239 * than dyn_keepalive_period.
240 */
241
242static u_int32_t dyn_keepalive_interval = 20;
243static u_int32_t dyn_keepalive_period = 5;
244static u_int32_t dyn_keepalive = 1; /* do send keepalives */
245
246static u_int32_t static_count; /* # of static rules */
247static u_int32_t static_len; /* size in bytes of static rules */
248static u_int32_t dyn_count; /* # of dynamic rules */
249static u_int32_t dyn_max = 4096; /* max # of dynamic rules */
250
251SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
252 &dyn_buckets, 0, "Number of dyn. buckets");
253SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
254 &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
255SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
256 &dyn_count, 0, "Number of dyn. rules");
257SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
258 &dyn_max, 0, "Max number of dyn. rules");
259SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
260 &static_count, 0, "Number of static rules");
261SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
262 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
263SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
264 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
265SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
266 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
267SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
268 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
269SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
270 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
271SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
272 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
273SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
274 &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
275
276#endif /* SYSCTL_NODE */
277
278
279/*
280 * This macro maps an ip pointer into a layer3 header pointer of type T
281 */
282#define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
283
284static __inline int
285icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
286{
287 int type = L3HDR(struct icmp,ip)->icmp_type;
288
289 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
290}
291
292#define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
293 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
294
295static int
296is_icmp_query(struct ip *ip)
297{
298 int type = L3HDR(struct icmp, ip)->icmp_type;
299 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
300}
301#undef TT
302
303/*
304 * The following checks use two arrays of 8 or 16 bits to store the
305 * bits that we want set or clear, respectively. They are in the
306 * low and high half of cmd->arg1 or cmd->d[0].
307 *
308 * We scan options and store the bits we find set. We succeed if
309 *
310 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
311 *
312 * The code is sometimes optimized not to store additional variables.
313 */
314
315static int
316flags_match(ipfw_insn *cmd, u_int8_t bits)
317{
318 u_char want_clear;
319 bits = ~bits;
320
321 if ( ((cmd->arg1 & 0xff) & bits) != 0)
322 return 0; /* some bits we want set were clear */
323 want_clear = (cmd->arg1 >> 8) & 0xff;
324 if ( (want_clear & bits) != want_clear)
325 return 0; /* some bits we want clear were set */
326 return 1;
327}
328
329static int
330ipopts_match(struct ip *ip, ipfw_insn *cmd)
331{
332 int optlen, bits = 0;
333 u_char *cp = (u_char *)(ip + 1);
334 int x = (ip->ip_hl << 2) - sizeof (struct ip);
335
336 for (; x > 0; x -= optlen, cp += optlen) {
337 int opt = cp[IPOPT_OPTVAL];
338
339 if (opt == IPOPT_EOL)
340 break;
341 if (opt == IPOPT_NOP)
342 optlen = 1;
343 else {
344 optlen = cp[IPOPT_OLEN];
345 if (optlen <= 0 || optlen > x)
346 return 0; /* invalid or truncated */
347 }
348 switch (opt) {
349
350 default:
351 break;
352
353 case IPOPT_LSRR:
354 bits |= IP_FW_IPOPT_LSRR;
355 break;
356
357 case IPOPT_SSRR:
358 bits |= IP_FW_IPOPT_SSRR;
359 break;
360
361 case IPOPT_RR:
362 bits |= IP_FW_IPOPT_RR;
363 break;
364
365 case IPOPT_TS:
366 bits |= IP_FW_IPOPT_TS;
367 break;
368 }
369 }
370 return (flags_match(cmd, bits));
371}
372
373static int
374tcpopts_match(struct ip *ip, ipfw_insn *cmd)
375{
376 int optlen, bits = 0;
377 struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
378 u_char *cp = (u_char *)(tcp + 1);
379 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
380
381 for (; x > 0; x -= optlen, cp += optlen) {
382 int opt = cp[0];
383 if (opt == TCPOPT_EOL)
384 break;
385 if (opt == TCPOPT_NOP)
386 optlen = 1;
387 else {
388 optlen = cp[1];
389 if (optlen <= 0)
390 break;
391 }
392
393 switch (opt) {
394
395 default:
396 break;
397
398 case TCPOPT_MAXSEG:
399 bits |= IP_FW_TCPOPT_MSS;
400 break;
401
402 case TCPOPT_WINDOW:
403 bits |= IP_FW_TCPOPT_WINDOW;
404 break;
405
406 case TCPOPT_SACK_PERMITTED:
407 case TCPOPT_SACK:
408 bits |= IP_FW_TCPOPT_SACK;
409 break;
410
411 case TCPOPT_TIMESTAMP:
412 bits |= IP_FW_TCPOPT_TS;
413 break;
414
415 case TCPOPT_CC:
416 case TCPOPT_CCNEW:
417 case TCPOPT_CCECHO:
418 bits |= IP_FW_TCPOPT_CC;
419 break;
420 }
421 }
422 return (flags_match(cmd, bits));
423}
424
425static int
426iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
427{
428 if (ifp == NULL) /* no iface with this packet, match fails */
429 return 0;
430 /* Check by name or by IP address */
431 if (cmd->name[0] != '\0') { /* match by name */
432 /* Check name */
433 if (cmd->p.glob) {
434 if (fnmatch(cmd->name, ifp->if_xname, 0) == 0)
435 return(1);
436 } else {
437 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
438 return(1);
439 }
440 } else {
441 struct ifaddr *ia;
442
443 /* XXX lock? */
444 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
445 if (ia->ifa_addr == NULL)
446 continue;
447 if (ia->ifa_addr->sa_family != AF_INET)
448 continue;
449 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
450 (ia->ifa_addr))->sin_addr.s_addr)
451 return(1); /* match */
452 }
453 }
454 return(0); /* no match, fail ... */
455}
456
457/*
458 * The verify_path function checks if a route to the src exists and
459 * if it is reachable via ifp (when provided).
460 *
461 * The 'verrevpath' option checks that the interface that an IP packet
462 * arrives on is the same interface that traffic destined for the
463 * packet's source address would be routed out of. The 'versrcreach'
464 * option just checks that the source address is reachable via any route
465 * (except default) in the routing table. These two are a measure to block
466 * forged packets. This is also commonly known as "anti-spoofing" or Unicast
467 * Reverse Path Forwarding (Unicast RFP) in Cisco-ese. The name of the knobs
468 * is purposely reminiscent of the Cisco IOS command,
469 *
470 * ip verify unicast reverse-path
471 * ip verify unicast source reachable-via any
472 *
473 * which implements the same functionality. But note that syntax is
474 * misleading. The check may be performed on all IP packets whether unicast,
475 * multicast, or broadcast.
476 */
477static int
478verify_path(struct in_addr src, struct ifnet *ifp)
479{
480 struct route ro;
481 struct sockaddr_in *dst;
482
483 bzero(&ro, sizeof(ro));
484
485 dst = (struct sockaddr_in *)&(ro.ro_dst);
486 dst->sin_family = AF_INET;
487 dst->sin_len = sizeof(*dst);
488 dst->sin_addr = src;
489 rtalloc_ign(&ro, RTF_CLONING);
490
491 if (ro.ro_rt == NULL)
492 return 0;
493
494 /* if ifp is provided, check for equality with rtentry */
495 if (ifp != NULL && ro.ro_rt->rt_ifp != ifp) {
496 RTFREE(ro.ro_rt);
497 return 0;
498 }
499
500 /* if no ifp provided, check if rtentry is not default route */
501 if (ifp == NULL &&
502 satosin(rt_key(ro.ro_rt))->sin_addr.s_addr == INADDR_ANY) {
503 RTFREE(ro.ro_rt);
504 return 0;
505 }
506
507 /* or if this is a blackhole/reject route */
508 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
509 RTFREE(ro.ro_rt);
510 return 0;
511 }
512
513 /* found valid route */
514 RTFREE(ro.ro_rt);
515 return 1;
516}
517
518
519static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */
520
521#define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
522#define SNP(buf) buf, sizeof(buf)
523
524/*
525 * We enter here when we have a rule with O_LOG.
526 * XXX this function alone takes about 2Kbytes of code!
527 */
528static void
529ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
530 struct mbuf *m, struct ifnet *oif)
531{
532 char *action;
533 int limit_reached = 0;
534 char action2[40], proto[48], fragment[28];
535
536 fragment[0] = '\0';
537 proto[0] = '\0';
538
539 if (f == NULL) { /* bogus pkt */
540 if (verbose_limit != 0 && norule_counter >= verbose_limit)
541 return;
542 norule_counter++;
543 if (norule_counter == verbose_limit)
544 limit_reached = verbose_limit;
545 action = "Refuse";
546 } else { /* O_LOG is the first action, find the real one */
547 ipfw_insn *cmd = ACTION_PTR(f);
548 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
549
550 if (l->max_log != 0 && l->log_left == 0)
551 return;
552 l->log_left--;
553 if (l->log_left == 0)
554 limit_reached = l->max_log;
555 cmd += F_LEN(cmd); /* point to first action */
556 if (cmd->opcode == O_PROB)
557 cmd += F_LEN(cmd);
558
559 action = action2;
560 switch (cmd->opcode) {
561 case O_DENY:
562 action = "Deny";
563 break;
564
565 case O_REJECT:
566 if (cmd->arg1==ICMP_REJECT_RST)
567 action = "Reset";
568 else if (cmd->arg1==ICMP_UNREACH_HOST)
569 action = "Reject";
570 else
571 snprintf(SNPARGS(action2, 0), "Unreach %d",
572 cmd->arg1);
573 break;
574
575 case O_ACCEPT:
576 action = "Accept";
577 break;
578 case O_COUNT:
579 action = "Count";
580 break;
581 case O_DIVERT:
582 snprintf(SNPARGS(action2, 0), "Divert %d",
583 cmd->arg1);
584 break;
585 case O_TEE:
586 snprintf(SNPARGS(action2, 0), "Tee %d",
587 cmd->arg1);
588 break;
589 case O_SKIPTO:
590 snprintf(SNPARGS(action2, 0), "SkipTo %d",
591 cmd->arg1);
592 break;
593 case O_PIPE:
594 snprintf(SNPARGS(action2, 0), "Pipe %d",
595 cmd->arg1);
596 break;
597 case O_QUEUE:
598 snprintf(SNPARGS(action2, 0), "Queue %d",
599 cmd->arg1);
600 break;
601 case O_FORWARD_IP: {
602 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
603 int len;
604
605 len = snprintf(SNPARGS(action2, 0), "Forward to %s",
606 inet_ntoa(sa->sa.sin_addr));
607 if (sa->sa.sin_port)
608 snprintf(SNPARGS(action2, len), ":%d",
609 sa->sa.sin_port);
610 }
611 break;
612 default:
613 action = "UNKNOWN";
614 break;
615 }
616 }
617
618 if (hlen == 0) { /* non-ip */
619 snprintf(SNPARGS(proto, 0), "MAC");
620 } else {
621 struct ip *ip = mtod(m, struct ip *);
622 /* these three are all aliases to the same thing */
623 struct icmp *const icmp = L3HDR(struct icmp, ip);
624 struct tcphdr *const tcp = (struct tcphdr *)icmp;
625 struct udphdr *const udp = (struct udphdr *)icmp;
626
627 int ip_off, offset, ip_len;
628
629 int len;
630
631 if (eh != NULL) { /* layer 2 packets are as on the wire */
632 ip_off = ntohs(ip->ip_off);
633 ip_len = ntohs(ip->ip_len);
634 } else {
635 ip_off = ip->ip_off;
636 ip_len = ip->ip_len;
637 }
638 offset = ip_off & IP_OFFMASK;
639 switch (ip->ip_p) {
640 case IPPROTO_TCP:
641 len = snprintf(SNPARGS(proto, 0), "TCP %s",
642 inet_ntoa(ip->ip_src));
643 if (offset == 0)
644 snprintf(SNPARGS(proto, len), ":%d %s:%d",
645 ntohs(tcp->th_sport),
646 inet_ntoa(ip->ip_dst),
647 ntohs(tcp->th_dport));
648 else
649 snprintf(SNPARGS(proto, len), " %s",
650 inet_ntoa(ip->ip_dst));
651 break;
652
653 case IPPROTO_UDP:
654 len = snprintf(SNPARGS(proto, 0), "UDP %s",
655 inet_ntoa(ip->ip_src));
656 if (offset == 0)
657 snprintf(SNPARGS(proto, len), ":%d %s:%d",
658 ntohs(udp->uh_sport),
659 inet_ntoa(ip->ip_dst),
660 ntohs(udp->uh_dport));
661 else
662 snprintf(SNPARGS(proto, len), " %s",
663 inet_ntoa(ip->ip_dst));
664 break;
665
666 case IPPROTO_ICMP:
667 if (offset == 0)
668 len = snprintf(SNPARGS(proto, 0),
669 "ICMP:%u.%u ",
670 icmp->icmp_type, icmp->icmp_code);
671 else
672 len = snprintf(SNPARGS(proto, 0), "ICMP ");
673 len += snprintf(SNPARGS(proto, len), "%s",
674 inet_ntoa(ip->ip_src));
675 snprintf(SNPARGS(proto, len), " %s",
676 inet_ntoa(ip->ip_dst));
677 break;
678
679 default:
680 len = snprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
681 inet_ntoa(ip->ip_src));
682 snprintf(SNPARGS(proto, len), " %s",
683 inet_ntoa(ip->ip_dst));
684 break;
685 }
686
687 if (ip_off & (IP_MF | IP_OFFMASK))
688 snprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
689 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
690 offset << 3,
691 (ip_off & IP_MF) ? "+" : "");
692 }
693 if (oif || m->m_pkthdr.rcvif)
694 log(LOG_SECURITY | LOG_INFO,
695 "ipfw: %d %s %s %s via %s%s\n",
696 f ? f->rulenum : -1,
697 action, proto, oif ? "out" : "in",
698 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
699 fragment);
700 else
701 log(LOG_SECURITY | LOG_INFO,
702 "ipfw: %d %s %s [no if info]%s\n",
703 f ? f->rulenum : -1,
704 action, proto, fragment);
705 if (limit_reached)
706 log(LOG_SECURITY | LOG_NOTICE,
707 "ipfw: limit %d reached on entry %d\n",
708 limit_reached, f ? f->rulenum : -1);
709}
710
711/*
712 * IMPORTANT: the hash function for dynamic rules must be commutative
713 * in source and destination (ip,port), because rules are bidirectional
714 * and we want to find both in the same bucket.
715 */
716static __inline int
717hash_packet(struct ipfw_flow_id *id)
718{
719 u_int32_t i;
720
721 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
722 i &= (curr_dyn_buckets - 1);
723 return i;
724}
725
726/**
727 * unlink a dynamic rule from a chain. prev is a pointer to
728 * the previous one, q is a pointer to the rule to delete,
729 * head is a pointer to the head of the queue.
730 * Modifies q and potentially also head.
731 */
732#define UNLINK_DYN_RULE(prev, head, q) { \
733 ipfw_dyn_rule *old_q = q; \
734 \
735 /* remove a refcount to the parent */ \
736 if (q->dyn_type == O_LIMIT) \
737 q->parent->count--; \
738 DEB(printf("ipfw: unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",\
739 (q->id.src_ip), (q->id.src_port), \
740 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \
741 if (prev != NULL) \
742 prev->next = q = q->next; \
743 else \
744 head = q = q->next; \
745 dyn_count--; \
746 free(old_q, M_IPFW); }
747
748#define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
749
750/**
751 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
752 *
753 * If keep_me == NULL, rules are deleted even if not expired,
754 * otherwise only expired rules are removed.
755 *
756 * The value of the second parameter is also used to point to identify
757 * a rule we absolutely do not want to remove (e.g. because we are
758 * holding a reference to it -- this is the case with O_LIMIT_PARENT
759 * rules). The pointer is only used for comparison, so any non-null
760 * value will do.
761 */
762static void
763remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
764{
765 static u_int32_t last_remove = 0;
766
767#define FORCE (keep_me == NULL)
768
769 ipfw_dyn_rule *prev, *q;
770 int i, pass = 0, max_pass = 0;
771
772 IPFW_DYN_LOCK_ASSERT();
773
774 if (ipfw_dyn_v == NULL || dyn_count == 0)
775 return;
776 /* do not expire more than once per second, it is useless */
777 if (!FORCE && last_remove == time_second)
778 return;
779 last_remove = time_second;
780
781 /*
782 * because O_LIMIT refer to parent rules, during the first pass only
783 * remove child and mark any pending LIMIT_PARENT, and remove
784 * them in a second pass.
785 */
786next_pass:
787 for (i = 0 ; i < curr_dyn_buckets ; i++) {
788 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
789 /*
790 * Logic can become complex here, so we split tests.
791 */
792 if (q == keep_me)
793 goto next;
794 if (rule != NULL && rule != q->rule)
795 goto next; /* not the one we are looking for */
796 if (q->dyn_type == O_LIMIT_PARENT) {
797 /*
798 * handle parent in the second pass,
799 * record we need one.
800 */
801 max_pass = 1;
802 if (pass == 0)
803 goto next;
804 if (FORCE && q->count != 0 ) {
805 /* XXX should not happen! */
806 printf("ipfw: OUCH! cannot remove rule,"
807 " count %d\n", q->count);
808 }
809 } else {
810 if (!FORCE &&
811 !TIME_LEQ( q->expire, time_second ))
812 goto next;
813 }
814 if (q->dyn_type != O_LIMIT_PARENT || !q->count) {
815 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
816 continue;
817 }
818next:
819 prev=q;
820 q=q->next;
821 }
822 }
823 if (pass++ < max_pass)
824 goto next_pass;
825}
826
827
828/**
829 * lookup a dynamic rule.
830 */
831static ipfw_dyn_rule *
832lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction,
833 struct tcphdr *tcp)
834{
835 /*
836 * stateful ipfw extensions.
837 * Lookup into dynamic session queue
838 */
839#define MATCH_REVERSE 0
840#define MATCH_FORWARD 1
841#define MATCH_NONE 2
842#define MATCH_UNKNOWN 3
843 int i, dir = MATCH_NONE;
844 ipfw_dyn_rule *prev, *q=NULL;
845
846 IPFW_DYN_LOCK_ASSERT();
847
848 if (ipfw_dyn_v == NULL)
849 goto done; /* not found */
850 i = hash_packet( pkt );
851 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
852 if (q->dyn_type == O_LIMIT_PARENT && q->count)
853 goto next;
854 if (TIME_LEQ( q->expire, time_second)) { /* expire entry */
855 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
856 continue;
857 }
858 if (pkt->proto == q->id.proto &&
859 q->dyn_type != O_LIMIT_PARENT) {
860 if (pkt->src_ip == q->id.src_ip &&
861 pkt->dst_ip == q->id.dst_ip &&
862 pkt->src_port == q->id.src_port &&
863 pkt->dst_port == q->id.dst_port ) {
864 dir = MATCH_FORWARD;
865 break;
866 }
867 if (pkt->src_ip == q->id.dst_ip &&
868 pkt->dst_ip == q->id.src_ip &&
869 pkt->src_port == q->id.dst_port &&
870 pkt->dst_port == q->id.src_port ) {
871 dir = MATCH_REVERSE;
872 break;
873 }
874 }
875next:
876 prev = q;
877 q = q->next;
878 }
879 if (q == NULL)
880 goto done; /* q = NULL, not found */
881
882 if ( prev != NULL) { /* found and not in front */
883 prev->next = q->next;
884 q->next = ipfw_dyn_v[i];
885 ipfw_dyn_v[i] = q;
886 }
887 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
888 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
889
890#define BOTH_SYN (TH_SYN | (TH_SYN << 8))
891#define BOTH_FIN (TH_FIN | (TH_FIN << 8))
892 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
893 switch (q->state) {
894 case TH_SYN: /* opening */
895 q->expire = time_second + dyn_syn_lifetime;
896 break;
897
898 case BOTH_SYN: /* move to established */
899 case BOTH_SYN | TH_FIN : /* one side tries to close */
900 case BOTH_SYN | (TH_FIN << 8) :
901 if (tcp) {
902#define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
903 u_int32_t ack = ntohl(tcp->th_ack);
904 if (dir == MATCH_FORWARD) {
905 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
906 q->ack_fwd = ack;
907 else { /* ignore out-of-sequence */
908 break;
909 }
910 } else {
911 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
912 q->ack_rev = ack;
913 else { /* ignore out-of-sequence */
914 break;
915 }
916 }
917 }
918 q->expire = time_second + dyn_ack_lifetime;
919 break;
920
921 case BOTH_SYN | BOTH_FIN: /* both sides closed */
922 if (dyn_fin_lifetime >= dyn_keepalive_period)
923 dyn_fin_lifetime = dyn_keepalive_period - 1;
924 q->expire = time_second + dyn_fin_lifetime;
925 break;
926
927 default:
928#if 0
929 /*
930 * reset or some invalid combination, but can also
931 * occur if we use keep-state the wrong way.
932 */
933 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
934 printf("invalid state: 0x%x\n", q->state);
935#endif
936 if (dyn_rst_lifetime >= dyn_keepalive_period)
937 dyn_rst_lifetime = dyn_keepalive_period - 1;
938 q->expire = time_second + dyn_rst_lifetime;
939 break;
940 }
941 } else if (pkt->proto == IPPROTO_UDP) {
942 q->expire = time_second + dyn_udp_lifetime;
943 } else {
944 /* other protocols */
945 q->expire = time_second + dyn_short_lifetime;
946 }
947done:
948 if (match_direction)
949 *match_direction = dir;
950 return q;
951}
952
953static ipfw_dyn_rule *
954lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
955 struct tcphdr *tcp)
956{
957 ipfw_dyn_rule *q;
958
959 IPFW_DYN_LOCK();
960 q = lookup_dyn_rule_locked(pkt, match_direction, tcp);
961 if (q == NULL)
962 IPFW_DYN_UNLOCK();
963 /* NB: return table locked when q is not NULL */
964 return q;
965}
966
967static void
968realloc_dynamic_table(void)
969{
970 IPFW_DYN_LOCK_ASSERT();
971
972 /*
973 * Try reallocation, make sure we have a power of 2 and do
974 * not allow more than 64k entries. In case of overflow,
975 * default to 1024.
976 */
977
978 if (dyn_buckets > 65536)
979 dyn_buckets = 1024;
980 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
981 dyn_buckets = curr_dyn_buckets; /* reset */
982 return;
983 }
984 curr_dyn_buckets = dyn_buckets;
985 if (ipfw_dyn_v != NULL)
986 free(ipfw_dyn_v, M_IPFW);
987 for (;;) {
988 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
989 M_IPFW, M_NOWAIT | M_ZERO);
990 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
991 break;
992 curr_dyn_buckets /= 2;
993 }
994}
995
996/**
997 * Install state of type 'type' for a dynamic session.
998 * The hash table contains two type of rules:
999 * - regular rules (O_KEEP_STATE)
1000 * - rules for sessions with limited number of sess per user
1001 * (O_LIMIT). When they are created, the parent is
1002 * increased by 1, and decreased on delete. In this case,
1003 * the third parameter is the parent rule and not the chain.
1004 * - "parent" rules for the above (O_LIMIT_PARENT).
1005 */
1006static ipfw_dyn_rule *
1007add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
1008{
1009 ipfw_dyn_rule *r;
1010 int i;
1011
1012 IPFW_DYN_LOCK_ASSERT();
1013
1014 if (ipfw_dyn_v == NULL ||
1015 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1016 realloc_dynamic_table();
1017 if (ipfw_dyn_v == NULL)
1018 return NULL; /* failed ! */
1019 }
1020 i = hash_packet(id);
1021
1022 r = malloc(sizeof *r, M_IPFW, M_NOWAIT | M_ZERO);
1023 if (r == NULL) {
1024 printf ("ipfw: sorry cannot allocate state\n");
1025 return NULL;
1026 }
1027
1028 /* increase refcount on parent, and set pointer */
1029 if (dyn_type == O_LIMIT) {
1030 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1031 if ( parent->dyn_type != O_LIMIT_PARENT)
1032 panic("invalid parent");
1033 parent->count++;
1034 r->parent = parent;
1035 rule = parent->rule;
1036 }
1037
1038 r->id = *id;
1039 r->expire = time_second + dyn_syn_lifetime;
1040 r->rule = rule;
1041 r->dyn_type = dyn_type;
1042 r->pcnt = r->bcnt = 0;
1043 r->count = 0;
1044
1045 r->bucket = i;
1046 r->next = ipfw_dyn_v[i];
1047 ipfw_dyn_v[i] = r;
1048 dyn_count++;
1049 DEB(printf("ipfw: add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1050 dyn_type,
1051 (r->id.src_ip), (r->id.src_port),
1052 (r->id.dst_ip), (r->id.dst_port),
1053 dyn_count ); )
1054 return r;
1055}
1056
1057/**
1058 * lookup dynamic parent rule using pkt and rule as search keys.
1059 * If the lookup fails, then install one.
1060 */
1061static ipfw_dyn_rule *
1062lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1063{
1064 ipfw_dyn_rule *q;
1065 int i;
1066
1067 IPFW_DYN_LOCK_ASSERT();
1068
1069 if (ipfw_dyn_v) {
1070 i = hash_packet( pkt );
1071 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
1072 if (q->dyn_type == O_LIMIT_PARENT &&
1073 rule== q->rule &&
1074 pkt->proto == q->id.proto &&
1075 pkt->src_ip == q->id.src_ip &&
1076 pkt->dst_ip == q->id.dst_ip &&
1077 pkt->src_port == q->id.src_port &&
1078 pkt->dst_port == q->id.dst_port) {
1079 q->expire = time_second + dyn_short_lifetime;
1080 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);)
1081 return q;
1082 }
1083 }
1084 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1085}
1086
1087/**
1088 * Install dynamic state for rule type cmd->o.opcode
1089 *
1090 * Returns 1 (failure) if state is not installed because of errors or because
1091 * session limitations are enforced.
1092 */
1093static int
1094install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1095 struct ip_fw_args *args)
1096{
1097 static int last_log;
1098
1099 ipfw_dyn_rule *q;
1100
1101 DEB(printf("ipfw: install state type %d 0x%08x %u -> 0x%08x %u\n",
1102 cmd->o.opcode,
1103 (args->f_id.src_ip), (args->f_id.src_port),
1104 (args->f_id.dst_ip), (args->f_id.dst_port) );)
1105
1106 IPFW_DYN_LOCK();
1107
1108 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
1109
1110 if (q != NULL) { /* should never occur */
1111 if (last_log != time_second) {
1112 last_log = time_second;
1113 printf("ipfw: install_state: entry already present, done\n");
1114 }
1115 IPFW_DYN_UNLOCK();
1116 return 0;
1117 }
1118
1119 if (dyn_count >= dyn_max)
1120 /*
1121 * Run out of slots, try to remove any expired rule.
1122 */
1123 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1124
1125 if (dyn_count >= dyn_max) {
1126 if (last_log != time_second) {
1127 last_log = time_second;
1128 printf("ipfw: install_state: Too many dynamic rules\n");
1129 }
1130 IPFW_DYN_UNLOCK();
1131 return 1; /* cannot install, notify caller */
1132 }
1133
1134 switch (cmd->o.opcode) {
1135 case O_KEEP_STATE: /* bidir rule */
1136 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1137 break;
1138
1139 case O_LIMIT: /* limit number of sessions */
1140 {
1141 u_int16_t limit_mask = cmd->limit_mask;
1142 struct ipfw_flow_id id;
1143 ipfw_dyn_rule *parent;
1144
1145 DEB(printf("ipfw: installing dyn-limit rule %d\n",
1146 cmd->conn_limit);)
1147
1148 id.dst_ip = id.src_ip = 0;
1149 id.dst_port = id.src_port = 0;
1150 id.proto = args->f_id.proto;
1151
1152 if (limit_mask & DYN_SRC_ADDR)
1153 id.src_ip = args->f_id.src_ip;
1154 if (limit_mask & DYN_DST_ADDR)
1155 id.dst_ip = args->f_id.dst_ip;
1156 if (limit_mask & DYN_SRC_PORT)
1157 id.src_port = args->f_id.src_port;
1158 if (limit_mask & DYN_DST_PORT)
1159 id.dst_port = args->f_id.dst_port;
1160 parent = lookup_dyn_parent(&id, rule);
1161 if (parent == NULL) {
1162 printf("ipfw: add parent failed\n");
1163 return 1;
1164 }
1165 if (parent->count >= cmd->conn_limit) {
1166 /*
1167 * See if we can remove some expired rule.
1168 */
1169 remove_dyn_rule(rule, parent);
1170 if (parent->count >= cmd->conn_limit) {
1171 if (fw_verbose && last_log != time_second) {
1172 last_log = time_second;
1173 log(LOG_SECURITY | LOG_DEBUG,
1174 "drop session, too many entries\n");
1175 }
1176 IPFW_DYN_UNLOCK();
1177 return 1;
1178 }
1179 }
1180 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1181 }
1182 break;
1183 default:
1184 printf("ipfw: unknown dynamic rule type %u\n", cmd->o.opcode);
1185 IPFW_DYN_UNLOCK();
1186 return 1;
1187 }
1188 lookup_dyn_rule_locked(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1189 IPFW_DYN_UNLOCK();
1190 return 0;
1191}
1192
1193/*
1194 * Transmit a TCP packet, containing either a RST or a keepalive.
1195 * When flags & TH_RST, we are sending a RST packet, because of a
1196 * "reset" action matched the packet.
1197 * Otherwise we are sending a keepalive, and flags & TH_
1198 */
1199static void
1200send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags)
1201{
1202 struct mbuf *m;
1203 struct ip *ip;
1204 struct tcphdr *tcp;
1205
1206 MGETHDR(m, M_DONTWAIT, MT_HEADER);
1207 if (m == 0)
1208 return;
1209 m->m_pkthdr.rcvif = (struct ifnet *)0;
1210 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1211 m->m_data += max_linkhdr;
1212
1213 ip = mtod(m, struct ip *);
1214 bzero(ip, m->m_len);
1215 tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1216 ip->ip_p = IPPROTO_TCP;
1217 tcp->th_off = 5;
1218 /*
1219 * Assume we are sending a RST (or a keepalive in the reverse
1220 * direction), swap src and destination addresses and ports.
1221 */
1222 ip->ip_src.s_addr = htonl(id->dst_ip);
1223 ip->ip_dst.s_addr = htonl(id->src_ip);
1224 tcp->th_sport = htons(id->dst_port);
1225 tcp->th_dport = htons(id->src_port);
1226 if (flags & TH_RST) { /* we are sending a RST */
1227 if (flags & TH_ACK) {
1228 tcp->th_seq = htonl(ack);
1229 tcp->th_ack = htonl(0);
1230 tcp->th_flags = TH_RST;
1231 } else {
1232 if (flags & TH_SYN)
1233 seq++;
1234 tcp->th_seq = htonl(0);
1235 tcp->th_ack = htonl(seq);
1236 tcp->th_flags = TH_RST | TH_ACK;
1237 }
1238 } else {
1239 /*
1240 * We are sending a keepalive. flags & TH_SYN determines
1241 * the direction, forward if set, reverse if clear.
1242 * NOTE: seq and ack are always assumed to be correct
1243 * as set by the caller. This may be confusing...
1244 */
1245 if (flags & TH_SYN) {
1246 /*
1247 * we have to rewrite the correct addresses!
1248 */
1249 ip->ip_dst.s_addr = htonl(id->dst_ip);
1250 ip->ip_src.s_addr = htonl(id->src_ip);
1251 tcp->th_dport = htons(id->dst_port);
1252 tcp->th_sport = htons(id->src_port);
1253 }
1254 tcp->th_seq = htonl(seq);
1255 tcp->th_ack = htonl(ack);
1256 tcp->th_flags = TH_ACK;
1257 }
1258 /*
1259 * set ip_len to the payload size so we can compute
1260 * the tcp checksum on the pseudoheader
1261 * XXX check this, could save a couple of words ?
1262 */
1263 ip->ip_len = htons(sizeof(struct tcphdr));
1264 tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1265 /*
1266 * now fill fields left out earlier
1267 */
1268 ip->ip_ttl = ip_defttl;
1269 ip->ip_len = m->m_pkthdr.len;
1270 m->m_flags |= M_SKIP_FIREWALL;
1271 ip_output(m, NULL, NULL, 0, NULL, NULL);
1272}
1273
1274/*
1275 * sends a reject message, consuming the mbuf passed as an argument.
1276 */
1277static void
1278send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1279{
1280
1281 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1282 /* We need the IP header in host order for icmp_error(). */
1283 if (args->eh != NULL) {
1284 struct ip *ip = mtod(args->m, struct ip *);
1285 ip->ip_len = ntohs(ip->ip_len);
1286 ip->ip_off = ntohs(ip->ip_off);
1287 }
1288 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1289 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1290 struct tcphdr *const tcp =
1291 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1292 if ( (tcp->th_flags & TH_RST) == 0)
1293 send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1294 ntohl(tcp->th_ack),
1295 tcp->th_flags | TH_RST);
1296 m_freem(args->m);
1297 } else
1298 m_freem(args->m);
1299 args->m = NULL;
1300}
1301
1302/**
1303 *
1304 * Given an ip_fw *, lookup_next_rule will return a pointer
1305 * to the next rule, which can be either the jump
1306 * target (for skipto instructions) or the next one in the list (in
1307 * all other cases including a missing jump target).
1308 * The result is also written in the "next_rule" field of the rule.
1309 * Backward jumps are not allowed, so start looking from the next
1310 * rule...
1311 *
1312 * This never returns NULL -- in case we do not have an exact match,
1313 * the next rule is returned. When the ruleset is changed,
1314 * pointers are flushed so we are always correct.
1315 */
1316
1317static struct ip_fw *
1318lookup_next_rule(struct ip_fw *me)
1319{
1320 struct ip_fw *rule = NULL;
1321 ipfw_insn *cmd;
1322
1323 /* look for action, in case it is a skipto */
1324 cmd = ACTION_PTR(me);
1325 if (cmd->opcode == O_LOG)
1326 cmd += F_LEN(cmd);
1327 if ( cmd->opcode == O_SKIPTO )
1328 for (rule = me->next; rule ; rule = rule->next)
1329 if (rule->rulenum >= cmd->arg1)
1330 break;
1331 if (rule == NULL) /* failure or not a skipto */
1332 rule = me->next;
1333 me->next_rule = rule;
1334 return rule;
1335}
1336
1337static void
1338init_tables(void)
1339{
1340 int i;
1341
1342 for (i = 0; i < IPFW_TABLES_MAX; i++) {
1343 rn_inithead((void **)&ipfw_tables[i].rnh, 32);
1344 ipfw_tables[i].modified = 1;
1345 }
1346}
1347
1348static int
1349add_table_entry(u_int16_t tbl, in_addr_t addr, u_int8_t mlen, u_int32_t value)
1350{
1351 struct radix_node_head *rnh;
1352 struct table_entry *ent;
1353
1354 if (tbl >= IPFW_TABLES_MAX)
1355 return (EINVAL);
1356 rnh = ipfw_tables[tbl].rnh;
1357 ent = malloc(sizeof(*ent), M_IPFW_TBL, M_NOWAIT | M_ZERO);
1358 if (ent == NULL)
1359 return (ENOMEM);
1360 ent->value = value;
1361 ent->addr.sin_len = ent->mask.sin_len = 8;
1362 ent->mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1363 ent->addr.sin_addr.s_addr = addr & ent->mask.sin_addr.s_addr;
1364 RADIX_NODE_HEAD_LOCK(rnh);
1365 if (rnh->rnh_addaddr(&ent->addr, &ent->mask, rnh, (void *)ent) ==
1366 NULL) {
1367 RADIX_NODE_HEAD_UNLOCK(rnh);
1368 free(ent, M_IPFW_TBL);
1369 return (EEXIST);
1370 }
1371 ipfw_tables[tbl].modified = 1;
1372 RADIX_NODE_HEAD_UNLOCK(rnh);
1373 return (0);
1374}
1375
1376static int
1377del_table_entry(u_int16_t tbl, in_addr_t addr, u_int8_t mlen)
1378{
1379 struct radix_node_head *rnh;
1380 struct table_entry *ent;
1381 struct sockaddr_in sa, mask;
1382
1383 if (tbl >= IPFW_TABLES_MAX)
1384 return (EINVAL);
1385 rnh = ipfw_tables[tbl].rnh;
1386 sa.sin_len = mask.sin_len = 8;
1387 mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1388 sa.sin_addr.s_addr = addr & mask.sin_addr.s_addr;
1389 RADIX_NODE_HEAD_LOCK(rnh);
1390 ent = (struct table_entry *)rnh->rnh_deladdr(&sa, &mask, rnh);
1391 if (ent == NULL) {
1392 RADIX_NODE_HEAD_UNLOCK(rnh);
1393 return (ESRCH);
1394 }
1395 ipfw_tables[tbl].modified = 1;
1396 RADIX_NODE_HEAD_UNLOCK(rnh);
1397 free(ent, M_IPFW_TBL);
1398 return (0);
1399}
1400
1401static int
1402flush_table_entry(struct radix_node *rn, void *arg)
1403{
1404 struct radix_node_head * const rnh = arg;
1405 struct table_entry *ent;
1406
1407 ent = (struct table_entry *)
1408 rnh->rnh_deladdr(rn->rn_key, rn->rn_mask, rnh);
1409 if (ent != NULL)
1410 free(ent, M_IPFW_TBL);
1411 return (0);
1412}
1413
1414static int
1415flush_table(u_int16_t tbl)
1416{
1417 struct radix_node_head *rnh;
1418
1419 if (tbl >= IPFW_TABLES_MAX)
1420 return (EINVAL);
1421 rnh = ipfw_tables[tbl].rnh;
1422 RADIX_NODE_HEAD_LOCK(rnh);
1423 rnh->rnh_walktree(rnh, flush_table_entry, rnh);
1424 ipfw_tables[tbl].modified = 1;
1425 RADIX_NODE_HEAD_UNLOCK(rnh);
1426 return (0);
1427}
1428
1429static void
1430flush_tables(void)
1431{
1432 u_int16_t tbl;
1433
1434 for (tbl = 0; tbl < IPFW_TABLES_MAX; tbl++)
1435 flush_table(tbl);
1436}
1437
1438static int
1439lookup_table(u_int16_t tbl, in_addr_t addr, u_int32_t *val)
1440{
1441 struct radix_node_head *rnh;
1442 struct table_entry *ent;
1443 struct sockaddr_in sa;
1444 static in_addr_t last_addr;
1445 static int last_tbl;
1446 static int last_match;
1447 static u_int32_t last_value;
1448
1449 if (tbl >= IPFW_TABLES_MAX)
1450 return (0);
1451 if (tbl == last_tbl && addr == last_addr &&
1452 !ipfw_tables[tbl].modified) {
1453 if (last_match)
1454 *val = last_value;
1455 return (last_match);
1456 }
1457 rnh = ipfw_tables[tbl].rnh;
1458 sa.sin_len = 8;
1459 sa.sin_addr.s_addr = addr;
1460 RADIX_NODE_HEAD_LOCK(rnh);
1461 ipfw_tables[tbl].modified = 0;
1462 ent = (struct table_entry *)(rnh->rnh_lookup(&sa, NULL, rnh));
1463 RADIX_NODE_HEAD_UNLOCK(rnh);
1464 last_addr = addr;
1465 last_tbl = tbl;
1466 if (ent != NULL) {
1467 last_value = *val = ent->value;
1468 last_match = 1;
1469 return (1);
1470 }
1471 last_match = 0;
1472 return (0);
1473}
1474
1475static int
1476count_table_entry(struct radix_node *rn, void *arg)
1477{
1478 u_int32_t * const cnt = arg;
1479
1480 (*cnt)++;
1481 return (0);
1482}
1483
1484static int
1485count_table(u_int32_t tbl, u_int32_t *cnt)
1486{
1487 struct radix_node_head *rnh;
1488
1489 if (tbl >= IPFW_TABLES_MAX)
1490 return (EINVAL);
1491 rnh = ipfw_tables[tbl].rnh;
1492 *cnt = 0;
1493 RADIX_NODE_HEAD_LOCK(rnh);
1494 rnh->rnh_walktree(rnh, count_table_entry, cnt);
1495 RADIX_NODE_HEAD_UNLOCK(rnh);
1496 return (0);
1497}
1498
1499static int
1500dump_table_entry(struct radix_node *rn, void *arg)
1501{
1502 struct table_entry * const n = (struct table_entry *)rn;
1503 ipfw_table * const tbl = arg;
1504 ipfw_table_entry *ent;
1505
1506 if (tbl->cnt == tbl->size)
1507 return (1);
1508 ent = &tbl->ent[tbl->cnt];
1509 ent->tbl = tbl->tbl;
1510 if (in_nullhost(n->mask.sin_addr))
1511 ent->masklen = 0;
1512 else
1513 ent->masklen = 33 - ffs(ntohl(n->mask.sin_addr.s_addr));
1514 ent->addr = n->addr.sin_addr.s_addr;
1515 ent->value = n->value;
1516 tbl->cnt++;
1517 return (0);
1518}
1519
1520static int
1521dump_table(ipfw_table *tbl)
1522{
1523 struct radix_node_head *rnh;
1524
1525 if (tbl->tbl >= IPFW_TABLES_MAX)
1526 return (EINVAL);
1527 rnh = ipfw_tables[tbl->tbl].rnh;
1528 tbl->cnt = 0;
1529 RADIX_NODE_HEAD_LOCK(rnh);
1530 rnh->rnh_walktree(rnh, dump_table_entry, tbl);
1531 RADIX_NODE_HEAD_UNLOCK(rnh);
1532 return (0);
1533}
1534
|
1942 break;
1943
1944 case O_RECV:
1945 match = iface_match(m->m_pkthdr.rcvif,
1946 (ipfw_insn_if *)cmd);
1947 break;
1948
1949 case O_XMIT:
1950 match = iface_match(oif, (ipfw_insn_if *)cmd);
1951 break;
1952
1953 case O_VIA:
1954 match = iface_match(oif ? oif :
1955 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1956 break;
1957
1958 case O_MACADDR2:
1959 if (args->eh != NULL) { /* have MAC header */
1960 u_int32_t *want = (u_int32_t *)
1961 ((ipfw_insn_mac *)cmd)->addr;
1962 u_int32_t *mask = (u_int32_t *)
1963 ((ipfw_insn_mac *)cmd)->mask;
1964 u_int32_t *hdr = (u_int32_t *)args->eh;
1965
1966 match =
1967 ( want[0] == (hdr[0] & mask[0]) &&
1968 want[1] == (hdr[1] & mask[1]) &&
1969 want[2] == (hdr[2] & mask[2]) );
1970 }
1971 break;
1972
1973 case O_MAC_TYPE:
1974 if (args->eh != NULL) {
1975 u_int16_t t =
1976 ntohs(args->eh->ether_type);
1977 u_int16_t *p =
1978 ((ipfw_insn_u16 *)cmd)->ports;
1979 int i;
1980
1981 for (i = cmdlen - 1; !match && i>0;
1982 i--, p += 2)
1983 match = (t>=p[0] && t<=p[1]);
1984 }
1985 break;
1986
1987 case O_FRAG:
1988 match = (hlen > 0 && offset != 0);
1989 break;
1990
1991 case O_IN: /* "out" is "not in" */
1992 match = (oif == NULL);
1993 break;
1994
1995 case O_LAYER2:
1996 match = (args->eh != NULL);
1997 break;
1998
1999 case O_PROTO:
2000 /*
2001 * We do not allow an arg of 0 so the
2002 * check of "proto" only suffices.
2003 */
2004 match = (proto == cmd->arg1);
2005 break;
2006
2007 case O_IP_SRC:
2008 match = (hlen > 0 &&
2009 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2010 src_ip.s_addr);
2011 break;
2012
2013 case O_IP_SRC_LOOKUP:
2014 case O_IP_DST_LOOKUP:
2015 if (hlen > 0) {
2016 uint32_t a =
2017 (cmd->opcode == O_IP_DST_LOOKUP) ?
2018 dst_ip.s_addr : src_ip.s_addr;
2019 uint32_t v;
2020
2021 match = lookup_table(cmd->arg1, a, &v);
2022 if (!match)
2023 break;
2024 if (cmdlen == F_INSN_SIZE(ipfw_insn_u32))
2025 match =
2026 ((ipfw_insn_u32 *)cmd)->d[0] == v;
2027 }
2028 break;
2029
2030 case O_IP_SRC_MASK:
2031 case O_IP_DST_MASK:
2032 if (hlen > 0) {
2033 uint32_t a =
2034 (cmd->opcode == O_IP_DST_MASK) ?
2035 dst_ip.s_addr : src_ip.s_addr;
2036 uint32_t *p = ((ipfw_insn_u32 *)cmd)->d;
2037 int i = cmdlen-1;
2038
2039 for (; !match && i>0; i-= 2, p+= 2)
2040 match = (p[0] == (a & p[1]));
2041 }
2042 break;
2043
2044 case O_IP_SRC_ME:
2045 if (hlen > 0) {
2046 struct ifnet *tif;
2047
2048 INADDR_TO_IFP(src_ip, tif);
2049 match = (tif != NULL);
2050 }
2051 break;
2052
2053 case O_IP_DST_SET:
2054 case O_IP_SRC_SET:
2055 if (hlen > 0) {
2056 u_int32_t *d = (u_int32_t *)(cmd+1);
2057 u_int32_t addr =
2058 cmd->opcode == O_IP_DST_SET ?
2059 args->f_id.dst_ip :
2060 args->f_id.src_ip;
2061
2062 if (addr < d[0])
2063 break;
2064 addr -= d[0]; /* subtract base */
2065 match = (addr < cmd->arg1) &&
2066 ( d[ 1 + (addr>>5)] &
2067 (1<<(addr & 0x1f)) );
2068 }
2069 break;
2070
2071 case O_IP_DST:
2072 match = (hlen > 0 &&
2073 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2074 dst_ip.s_addr);
2075 break;
2076
2077 case O_IP_DST_ME:
2078 if (hlen > 0) {
2079 struct ifnet *tif;
2080
2081 INADDR_TO_IFP(dst_ip, tif);
2082 match = (tif != NULL);
2083 }
2084 break;
2085
2086 case O_IP_SRCPORT:
2087 case O_IP_DSTPORT:
2088 /*
2089 * offset == 0 && proto != 0 is enough
2090 * to guarantee that we have an IPv4
2091 * packet with port info.
2092 */
2093 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
2094 && offset == 0) {
2095 u_int16_t x =
2096 (cmd->opcode == O_IP_SRCPORT) ?
2097 src_port : dst_port ;
2098 u_int16_t *p =
2099 ((ipfw_insn_u16 *)cmd)->ports;
2100 int i;
2101
2102 for (i = cmdlen - 1; !match && i>0;
2103 i--, p += 2)
2104 match = (x>=p[0] && x<=p[1]);
2105 }
2106 break;
2107
2108 case O_ICMPTYPE:
2109 match = (offset == 0 && proto==IPPROTO_ICMP &&
2110 icmptype_match(ip, (ipfw_insn_u32 *)cmd) );
2111 break;
2112
2113 case O_IPOPT:
2114 match = (hlen > 0 && ipopts_match(ip, cmd) );
2115 break;
2116
2117 case O_IPVER:
2118 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
2119 break;
2120
2121 case O_IPID:
2122 case O_IPLEN:
2123 case O_IPTTL:
2124 if (hlen > 0) { /* only for IP packets */
2125 uint16_t x;
2126 uint16_t *p;
2127 int i;
2128
2129 if (cmd->opcode == O_IPLEN)
2130 x = ip_len;
2131 else if (cmd->opcode == O_IPTTL)
2132 x = ip->ip_ttl;
2133 else /* must be IPID */
2134 x = ntohs(ip->ip_id);
2135 if (cmdlen == 1) {
2136 match = (cmd->arg1 == x);
2137 break;
2138 }
2139 /* otherwise we have ranges */
2140 p = ((ipfw_insn_u16 *)cmd)->ports;
2141 i = cmdlen - 1;
2142 for (; !match && i>0; i--, p += 2)
2143 match = (x >= p[0] && x <= p[1]);
2144 }
2145 break;
2146
2147 case O_IPPRECEDENCE:
2148 match = (hlen > 0 &&
2149 (cmd->arg1 == (ip->ip_tos & 0xe0)) );
2150 break;
2151
2152 case O_IPTOS:
2153 match = (hlen > 0 &&
2154 flags_match(cmd, ip->ip_tos));
2155 break;
2156
2157 case O_TCPFLAGS:
2158 match = (proto == IPPROTO_TCP && offset == 0 &&
2159 flags_match(cmd,
2160 L3HDR(struct tcphdr,ip)->th_flags));
2161 break;
2162
2163 case O_TCPOPTS:
2164 match = (proto == IPPROTO_TCP && offset == 0 &&
2165 tcpopts_match(ip, cmd));
2166 break;
2167
2168 case O_TCPSEQ:
2169 match = (proto == IPPROTO_TCP && offset == 0 &&
2170 ((ipfw_insn_u32 *)cmd)->d[0] ==
2171 L3HDR(struct tcphdr,ip)->th_seq);
2172 break;
2173
2174 case O_TCPACK:
2175 match = (proto == IPPROTO_TCP && offset == 0 &&
2176 ((ipfw_insn_u32 *)cmd)->d[0] ==
2177 L3HDR(struct tcphdr,ip)->th_ack);
2178 break;
2179
2180 case O_TCPWIN:
2181 match = (proto == IPPROTO_TCP && offset == 0 &&
2182 cmd->arg1 ==
2183 L3HDR(struct tcphdr,ip)->th_win);
2184 break;
2185
2186 case O_ESTAB:
2187 /* reject packets which have SYN only */
2188 /* XXX should i also check for TH_ACK ? */
2189 match = (proto == IPPROTO_TCP && offset == 0 &&
2190 (L3HDR(struct tcphdr,ip)->th_flags &
2191 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2192 break;
2193
2194 case O_LOG:
2195 if (fw_verbose)
2196 ipfw_log(f, hlen, args->eh, m, oif);
2197 match = 1;
2198 break;
2199
2200 case O_PROB:
2201 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]);
2202 break;
2203
2204 case O_VERREVPATH:
2205 /* Outgoing packets automatically pass/match */
2206 match = (hlen > 0 && ((oif != NULL) ||
2207 (m->m_pkthdr.rcvif == NULL) ||
2208 verify_path(src_ip, m->m_pkthdr.rcvif)));
2209 break;
2210
2211 case O_VERSRCREACH:
2212 /* Outgoing packets automatically pass/match */
2213 match = (hlen > 0 && ((oif != NULL) ||
2214 verify_path(src_ip, NULL)));
2215 break;
2216
2217 case O_ANTISPOOF:
2218 /* Outgoing packets automatically pass/match */
2219 if (oif == NULL && hlen > 0 &&
2220 in_localaddr(src_ip))
2221 match = verify_path(src_ip,
2222 m->m_pkthdr.rcvif);
2223 else
2224 match = 1;
2225 break;
2226
2227 case O_IPSEC:
2228#ifdef FAST_IPSEC
2229 match = (m_tag_find(m,
2230 PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL);
2231#endif
2232#ifdef IPSEC
2233 match = (ipsec_getnhist(m) != 0);
2234#endif
2235 /* otherwise no match */
2236 break;
2237
2238 /*
2239 * The second set of opcodes represents 'actions',
2240 * i.e. the terminal part of a rule once the packet
2241 * matches all previous patterns.
2242 * Typically there is only one action for each rule,
2243 * and the opcode is stored at the end of the rule
2244 * (but there are exceptions -- see below).
2245 *
2246 * In general, here we set retval and terminate the
2247 * outer loop (would be a 'break 3' in some language,
2248 * but we need to do a 'goto done').
2249 *
2250 * Exceptions:
2251 * O_COUNT and O_SKIPTO actions:
2252 * instead of terminating, we jump to the next rule
2253 * ('goto next_rule', equivalent to a 'break 2'),
2254 * or to the SKIPTO target ('goto again' after
2255 * having set f, cmd and l), respectively.
2256 *
2257 * O_LIMIT and O_KEEP_STATE: these opcodes are
2258 * not real 'actions', and are stored right
2259 * before the 'action' part of the rule.
2260 * These opcodes try to install an entry in the
2261 * state tables; if successful, we continue with
2262 * the next opcode (match=1; break;), otherwise
2263 * the packet * must be dropped
2264 * ('goto done' after setting retval);
2265 *
2266 * O_PROBE_STATE and O_CHECK_STATE: these opcodes
2267 * cause a lookup of the state table, and a jump
2268 * to the 'action' part of the parent rule
2269 * ('goto check_body') if an entry is found, or
2270 * (CHECK_STATE only) a jump to the next rule if
2271 * the entry is not found ('goto next_rule').
2272 * The result of the lookup is cached to make
2273 * further instances of these opcodes are
2274 * effectively NOPs.
2275 */
2276 case O_LIMIT:
2277 case O_KEEP_STATE:
2278 if (install_state(f,
2279 (ipfw_insn_limit *)cmd, args)) {
2280 retval = IP_FW_PORT_DENY_FLAG;
2281 goto done; /* error/limit violation */
2282 }
2283 match = 1;
2284 break;
2285
2286 case O_PROBE_STATE:
2287 case O_CHECK_STATE:
2288 /*
2289 * dynamic rules are checked at the first
2290 * keep-state or check-state occurrence,
2291 * with the result being stored in dyn_dir.
2292 * The compiler introduces a PROBE_STATE
2293 * instruction for us when we have a
2294 * KEEP_STATE (because PROBE_STATE needs
2295 * to be run first).
2296 */
2297 if (dyn_dir == MATCH_UNKNOWN &&
2298 (q = lookup_dyn_rule(&args->f_id,
2299 &dyn_dir, proto == IPPROTO_TCP ?
2300 L3HDR(struct tcphdr, ip) : NULL))
2301 != NULL) {
2302 /*
2303 * Found dynamic entry, update stats
2304 * and jump to the 'action' part of
2305 * the parent rule.
2306 */
2307 q->pcnt++;
2308 q->bcnt += pktlen;
2309 f = q->rule;
2310 cmd = ACTION_PTR(f);
2311 l = f->cmd_len - f->act_ofs;
2312 IPFW_DYN_UNLOCK();
2313 goto check_body;
2314 }
2315 /*
2316 * Dynamic entry not found. If CHECK_STATE,
2317 * skip to next rule, if PROBE_STATE just
2318 * ignore and continue with next opcode.
2319 */
2320 if (cmd->opcode == O_CHECK_STATE)
2321 goto next_rule;
2322 match = 1;
2323 break;
2324
2325 case O_ACCEPT:
2326 retval = 0; /* accept */
2327 goto done;
2328
2329 case O_PIPE:
2330 case O_QUEUE:
2331 args->rule = f; /* report matching rule */
2332 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG;
2333 goto done;
2334
2335 case O_DIVERT:
2336 case O_TEE: {
2337 struct divert_tag *dt;
2338
2339 if (args->eh) /* not on layer 2 */
2340 break;
2341 mtag = m_tag_get(PACKET_TAG_DIVERT,
2342 sizeof(struct divert_tag),
2343 M_NOWAIT);
2344 if (mtag == NULL) {
2345 /* XXX statistic */
2346 /* drop packet */
2347 IPFW_UNLOCK(chain);
2348 return IP_FW_PORT_DENY_FLAG;
2349 }
2350 dt = (struct divert_tag *)(mtag+1);
2351 dt->cookie = f->rulenum;
2352 dt->info = (cmd->opcode == O_DIVERT) ?
2353 cmd->arg1 :
2354 cmd->arg1 | IP_FW_PORT_TEE_FLAG;
2355 m_tag_prepend(m, mtag);
2356 retval = dt->info;
2357 goto done;
2358 }
2359
2360 case O_COUNT:
2361 case O_SKIPTO:
2362 f->pcnt++; /* update stats */
2363 f->bcnt += pktlen;
2364 f->timestamp = time_second;
2365 if (cmd->opcode == O_COUNT)
2366 goto next_rule;
2367 /* handle skipto */
2368 if (f->next_rule == NULL)
2369 lookup_next_rule(f);
2370 f = f->next_rule;
2371 goto again;
2372
2373 case O_REJECT:
2374 /*
2375 * Drop the packet and send a reject notice
2376 * if the packet is not ICMP (or is an ICMP
2377 * query), and it is not multicast/broadcast.
2378 */
2379 if (hlen > 0 &&
2380 (proto != IPPROTO_ICMP ||
2381 is_icmp_query(ip)) &&
2382 !(m->m_flags & (M_BCAST|M_MCAST)) &&
2383 !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
2384 send_reject(args, cmd->arg1,
2385 offset,ip_len);
2386 m = args->m;
2387 }
2388 /* FALLTHROUGH */
2389 case O_DENY:
2390 retval = IP_FW_PORT_DENY_FLAG;
2391 goto done;
2392
2393 case O_FORWARD_IP:
2394 if (args->eh) /* not valid on layer2 pkts */
2395 break;
2396 if (!q || dyn_dir == MATCH_FORWARD)
2397 args->next_hop =
2398 &((ipfw_insn_sa *)cmd)->sa;
2399 retval = 0;
2400 goto done;
2401
2402 default:
2403 panic("-- unknown opcode %d\n", cmd->opcode);
2404 } /* end of switch() on opcodes */
2405
2406 if (cmd->len & F_NOT)
2407 match = !match;
2408
2409 if (match) {
2410 if (cmd->len & F_OR)
2411 skip_or = 1;
2412 } else {
2413 if (!(cmd->len & F_OR)) /* not an OR block, */
2414 break; /* try next rule */
2415 }
2416
2417 } /* end of inner for, scan opcodes */
2418
2419next_rule:; /* try next rule */
2420
2421 } /* end of outer for, scan rules */
2422 printf("ipfw: ouch!, skip past end of rules, denying packet\n");
2423 IPFW_UNLOCK(chain);
2424 return(IP_FW_PORT_DENY_FLAG);
2425
2426done:
2427 /* Update statistics */
2428 f->pcnt++;
2429 f->bcnt += pktlen;
2430 f->timestamp = time_second;
2431 IPFW_UNLOCK(chain);
2432 return retval;
2433
2434pullup_failed:
2435 if (fw_verbose)
2436 printf("ipfw: pullup failed\n");
2437 return(IP_FW_PORT_DENY_FLAG);
2438}
2439
2440/*
2441 * When a rule is added/deleted, clear the next_rule pointers in all rules.
2442 * These will be reconstructed on the fly as packets are matched.
2443 */
2444static void
2445flush_rule_ptrs(struct ip_fw_chain *chain)
2446{
2447 struct ip_fw *rule;
2448
2449 IPFW_LOCK_ASSERT(chain);
2450
2451 for (rule = chain->rules; rule; rule = rule->next)
2452 rule->next_rule = NULL;
2453}
2454
2455/*
2456 * When pipes/queues are deleted, clear the "pipe_ptr" pointer to a given
2457 * pipe/queue, or to all of them (match == NULL).
2458 */
2459void
2460flush_pipe_ptrs(struct dn_flow_set *match)
2461{
2462 struct ip_fw *rule;
2463
2464 IPFW_LOCK(&layer3_chain);
2465 for (rule = layer3_chain.rules; rule; rule = rule->next) {
2466 ipfw_insn_pipe *cmd = (ipfw_insn_pipe *)ACTION_PTR(rule);
2467
2468 if (cmd->o.opcode != O_PIPE && cmd->o.opcode != O_QUEUE)
2469 continue;
2470 /*
2471 * XXX Use bcmp/bzero to handle pipe_ptr to overcome
2472 * possible alignment problems on 64-bit architectures.
2473 * This code is seldom used so we do not worry too
2474 * much about efficiency.
2475 */
2476 if (match == NULL ||
2477 !bcmp(&cmd->pipe_ptr, &match, sizeof(match)) )
2478 bzero(&cmd->pipe_ptr, sizeof(cmd->pipe_ptr));
2479 }
2480 IPFW_UNLOCK(&layer3_chain);
2481}
2482
2483/*
2484 * Add a new rule to the list. Copy the rule into a malloc'ed area, then
2485 * possibly create a rule number and add the rule to the list.
2486 * Update the rule_number in the input struct so the caller knows it as well.
2487 */
2488static int
2489add_rule(struct ip_fw_chain *chain, struct ip_fw *input_rule)
2490{
2491 struct ip_fw *rule, *f, *prev;
2492 int l = RULESIZE(input_rule);
2493
2494 if (chain->rules == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE)
2495 return (EINVAL);
2496
2497 rule = malloc(l, M_IPFW, M_NOWAIT | M_ZERO);
2498 if (rule == NULL)
2499 return (ENOSPC);
2500
2501 bcopy(input_rule, rule, l);
2502
2503 rule->next = NULL;
2504 rule->next_rule = NULL;
2505
2506 rule->pcnt = 0;
2507 rule->bcnt = 0;
2508 rule->timestamp = 0;
2509
2510 IPFW_LOCK(chain);
2511
2512 if (chain->rules == NULL) { /* default rule */
2513 chain->rules = rule;
2514 goto done;
2515 }
2516
2517 /*
2518 * If rulenum is 0, find highest numbered rule before the
2519 * default rule, and add autoinc_step
2520 */
2521 if (autoinc_step < 1)
2522 autoinc_step = 1;
2523 else if (autoinc_step > 1000)
2524 autoinc_step = 1000;
2525 if (rule->rulenum == 0) {
2526 /*
2527 * locate the highest numbered rule before default
2528 */
2529 for (f = chain->rules; f; f = f->next) {
2530 if (f->rulenum == IPFW_DEFAULT_RULE)
2531 break;
2532 rule->rulenum = f->rulenum;
2533 }
2534 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
2535 rule->rulenum += autoinc_step;
2536 input_rule->rulenum = rule->rulenum;
2537 }
2538
2539 /*
2540 * Now insert the new rule in the right place in the sorted list.
2541 */
2542 for (prev = NULL, f = chain->rules; f; prev = f, f = f->next) {
2543 if (f->rulenum > rule->rulenum) { /* found the location */
2544 if (prev) {
2545 rule->next = f;
2546 prev->next = rule;
2547 } else { /* head insert */
2548 rule->next = chain->rules;
2549 chain->rules = rule;
2550 }
2551 break;
2552 }
2553 }
2554 flush_rule_ptrs(chain);
2555done:
2556 static_count++;
2557 static_len += l;
2558 IPFW_UNLOCK(chain);
2559 DEB(printf("ipfw: installed rule %d, static count now %d\n",
2560 rule->rulenum, static_count);)
2561 return (0);
2562}
2563
2564/**
2565 * Remove a static rule (including derived * dynamic rules)
2566 * and place it on the ``reap list'' for later reclamation.
2567 * The caller is in charge of clearing rule pointers to avoid
2568 * dangling pointers.
2569 * @return a pointer to the next entry.
2570 * Arguments are not checked, so they better be correct.
2571 */
2572static struct ip_fw *
2573remove_rule(struct ip_fw_chain *chain, struct ip_fw *rule, struct ip_fw *prev)
2574{
2575 struct ip_fw *n;
2576 int l = RULESIZE(rule);
2577
2578 IPFW_LOCK_ASSERT(chain);
2579
2580 n = rule->next;
2581 IPFW_DYN_LOCK();
2582 remove_dyn_rule(rule, NULL /* force removal */);
2583 IPFW_DYN_UNLOCK();
2584 if (prev == NULL)
2585 chain->rules = n;
2586 else
2587 prev->next = n;
2588 static_count--;
2589 static_len -= l;
2590
2591 rule->next = chain->reap;
2592 chain->reap = rule;
2593
2594 return n;
2595}
2596
2597/**
2598 * Reclaim storage associated with a list of rules. This is
2599 * typically the list created using remove_rule.
2600 */
2601static void
2602reap_rules(struct ip_fw *head)
2603{
2604 struct ip_fw *rule;
2605
2606 while ((rule = head) != NULL) {
2607 head = head->next;
2608 if (DUMMYNET_LOADED)
2609 ip_dn_ruledel_ptr(rule);
2610 free(rule, M_IPFW);
2611 }
2612}
2613
2614/*
2615 * Remove all rules from a chain (except rules in set RESVD_SET
2616 * unless kill_default = 1). The caller is responsible for
2617 * reclaiming storage for the rules left in chain->reap.
2618 */
2619static void
2620free_chain(struct ip_fw_chain *chain, int kill_default)
2621{
2622 struct ip_fw *prev, *rule;
2623
2624 IPFW_LOCK_ASSERT(chain);
2625
2626 flush_rule_ptrs(chain); /* more efficient to do outside the loop */
2627 for (prev = NULL, rule = chain->rules; rule ; )
2628 if (kill_default || rule->set != RESVD_SET)
2629 rule = remove_rule(chain, rule, prev);
2630 else {
2631 prev = rule;
2632 rule = rule->next;
2633 }
2634}
2635
2636/**
2637 * Remove all rules with given number, and also do set manipulation.
2638 * Assumes chain != NULL && *chain != NULL.
2639 *
2640 * The argument is an u_int32_t. The low 16 bit are the rule or set number,
2641 * the next 8 bits are the new set, the top 8 bits are the command:
2642 *
2643 * 0 delete rules with given number
2644 * 1 delete rules with given set number
2645 * 2 move rules with given number to new set
2646 * 3 move rules with given set number to new set
2647 * 4 swap sets with given numbers
2648 */
2649static int
2650del_entry(struct ip_fw_chain *chain, u_int32_t arg)
2651{
2652 struct ip_fw *prev = NULL, *rule;
2653 u_int16_t rulenum; /* rule or old_set */
2654 u_int8_t cmd, new_set;
2655
2656 rulenum = arg & 0xffff;
2657 cmd = (arg >> 24) & 0xff;
2658 new_set = (arg >> 16) & 0xff;
2659
2660 if (cmd > 4)
2661 return EINVAL;
2662 if (new_set > RESVD_SET)
2663 return EINVAL;
2664 if (cmd == 0 || cmd == 2) {
2665 if (rulenum >= IPFW_DEFAULT_RULE)
2666 return EINVAL;
2667 } else {
2668 if (rulenum > RESVD_SET) /* old_set */
2669 return EINVAL;
2670 }
2671
2672 IPFW_LOCK(chain);
2673 rule = chain->rules;
2674 chain->reap = NULL;
2675 switch (cmd) {
2676 case 0: /* delete rules with given number */
2677 /*
2678 * locate first rule to delete
2679 */
2680 for (; rule->rulenum < rulenum; prev = rule, rule = rule->next)
2681 ;
2682 if (rule->rulenum != rulenum) {
2683 IPFW_UNLOCK(chain);
2684 return EINVAL;
2685 }
2686
2687 /*
2688 * flush pointers outside the loop, then delete all matching
2689 * rules. prev remains the same throughout the cycle.
2690 */
2691 flush_rule_ptrs(chain);
2692 while (rule->rulenum == rulenum)
2693 rule = remove_rule(chain, rule, prev);
2694 break;
2695
2696 case 1: /* delete all rules with given set number */
2697 flush_rule_ptrs(chain);
2698 rule = chain->rules;
2699 while (rule->rulenum < IPFW_DEFAULT_RULE)
2700 if (rule->set == rulenum)
2701 rule = remove_rule(chain, rule, prev);
2702 else {
2703 prev = rule;
2704 rule = rule->next;
2705 }
2706 break;
2707
2708 case 2: /* move rules with given number to new set */
2709 rule = chain->rules;
2710 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
2711 if (rule->rulenum == rulenum)
2712 rule->set = new_set;
2713 break;
2714
2715 case 3: /* move rules with given set number to new set */
2716 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
2717 if (rule->set == rulenum)
2718 rule->set = new_set;
2719 break;
2720
2721 case 4: /* swap two sets */
2722 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
2723 if (rule->set == rulenum)
2724 rule->set = new_set;
2725 else if (rule->set == new_set)
2726 rule->set = rulenum;
2727 break;
2728 }
2729 /*
2730 * Look for rules to reclaim. We grab the list before
2731 * releasing the lock then reclaim them w/o the lock to
2732 * avoid a LOR with dummynet.
2733 */
2734 rule = chain->reap;
2735 chain->reap = NULL;
2736 IPFW_UNLOCK(chain);
2737 if (rule)
2738 reap_rules(rule);
2739 return 0;
2740}
2741
2742/*
2743 * Clear counters for a specific rule.
2744 * The enclosing "table" is assumed locked.
2745 */
2746static void
2747clear_counters(struct ip_fw *rule, int log_only)
2748{
2749 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
2750
2751 if (log_only == 0) {
2752 rule->bcnt = rule->pcnt = 0;
2753 rule->timestamp = 0;
2754 }
2755 if (l->o.opcode == O_LOG)
2756 l->log_left = l->max_log;
2757}
2758
2759/**
2760 * Reset some or all counters on firewall rules.
2761 * @arg frwl is null to clear all entries, or contains a specific
2762 * rule number.
2763 * @arg log_only is 1 if we only want to reset logs, zero otherwise.
2764 */
2765static int
2766zero_entry(struct ip_fw_chain *chain, int rulenum, int log_only)
2767{
2768 struct ip_fw *rule;
2769 char *msg;
2770
2771 IPFW_LOCK(chain);
2772 if (rulenum == 0) {
2773 norule_counter = 0;
2774 for (rule = chain->rules; rule; rule = rule->next)
2775 clear_counters(rule, log_only);
2776 msg = log_only ? "ipfw: All logging counts reset.\n" :
2777 "ipfw: Accounting cleared.\n";
2778 } else {
2779 int cleared = 0;
2780 /*
2781 * We can have multiple rules with the same number, so we
2782 * need to clear them all.
2783 */
2784 for (rule = chain->rules; rule; rule = rule->next)
2785 if (rule->rulenum == rulenum) {
2786 while (rule && rule->rulenum == rulenum) {
2787 clear_counters(rule, log_only);
2788 rule = rule->next;
2789 }
2790 cleared = 1;
2791 break;
2792 }
2793 if (!cleared) { /* we did not find any matching rules */
2794 IPFW_UNLOCK(chain);
2795 return (EINVAL);
2796 }
2797 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
2798 "ipfw: Entry %d cleared.\n";
2799 }
2800 IPFW_UNLOCK(chain);
2801
2802 if (fw_verbose)
2803 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
2804 return (0);
2805}
2806
2807/*
2808 * Check validity of the structure before insert.
2809 * Fortunately rules are simple, so this mostly need to check rule sizes.
2810 */
2811static int
2812check_ipfw_struct(struct ip_fw *rule, int size)
2813{
2814 int l, cmdlen = 0;
2815 int have_action=0;
2816 ipfw_insn *cmd;
2817
2818 if (size < sizeof(*rule)) {
2819 printf("ipfw: rule too short\n");
2820 return (EINVAL);
2821 }
2822 /* first, check for valid size */
2823 l = RULESIZE(rule);
2824 if (l != size) {
2825 printf("ipfw: size mismatch (have %d want %d)\n", size, l);
2826 return (EINVAL);
2827 }
2828 /*
2829 * Now go for the individual checks. Very simple ones, basically only
2830 * instruction sizes.
2831 */
2832 for (l = rule->cmd_len, cmd = rule->cmd ;
2833 l > 0 ; l -= cmdlen, cmd += cmdlen) {
2834 cmdlen = F_LEN(cmd);
2835 if (cmdlen > l) {
2836 printf("ipfw: opcode %d size truncated\n",
2837 cmd->opcode);
2838 return EINVAL;
2839 }
2840 DEB(printf("ipfw: opcode %d\n", cmd->opcode);)
2841 switch (cmd->opcode) {
2842 case O_PROBE_STATE:
2843 case O_KEEP_STATE:
2844 case O_PROTO:
2845 case O_IP_SRC_ME:
2846 case O_IP_DST_ME:
2847 case O_LAYER2:
2848 case O_IN:
2849 case O_FRAG:
2850 case O_IPOPT:
2851 case O_IPTOS:
2852 case O_IPPRECEDENCE:
2853 case O_IPVER:
2854 case O_TCPWIN:
2855 case O_TCPFLAGS:
2856 case O_TCPOPTS:
2857 case O_ESTAB:
2858 case O_VERREVPATH:
2859 case O_VERSRCREACH:
2860 case O_ANTISPOOF:
2861 case O_IPSEC:
2862 if (cmdlen != F_INSN_SIZE(ipfw_insn))
2863 goto bad_size;
2864 break;
2865
2866 case O_UID:
2867 case O_GID:
2868 case O_JAIL:
2869 case O_IP_SRC:
2870 case O_IP_DST:
2871 case O_TCPSEQ:
2872 case O_TCPACK:
2873 case O_PROB:
2874 case O_ICMPTYPE:
2875 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
2876 goto bad_size;
2877 break;
2878
2879 case O_LIMIT:
2880 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
2881 goto bad_size;
2882 break;
2883
2884 case O_LOG:
2885 if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
2886 goto bad_size;
2887
2888 ((ipfw_insn_log *)cmd)->log_left =
2889 ((ipfw_insn_log *)cmd)->max_log;
2890
2891 break;
2892
2893 case O_IP_SRC_MASK:
2894 case O_IP_DST_MASK:
2895 /* only odd command lengths */
2896 if ( !(cmdlen & 1) || cmdlen > 31)
2897 goto bad_size;
2898 break;
2899
2900 case O_IP_SRC_SET:
2901 case O_IP_DST_SET:
2902 if (cmd->arg1 == 0 || cmd->arg1 > 256) {
2903 printf("ipfw: invalid set size %d\n",
2904 cmd->arg1);
2905 return EINVAL;
2906 }
2907 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
2908 (cmd->arg1+31)/32 )
2909 goto bad_size;
2910 break;
2911
2912 case O_IP_SRC_LOOKUP:
2913 case O_IP_DST_LOOKUP:
2914 if (cmd->arg1 >= IPFW_TABLES_MAX) {
2915 printf("ipfw: invalid table number %d\n",
2916 cmd->arg1);
2917 return (EINVAL);
2918 }
2919 if (cmdlen != F_INSN_SIZE(ipfw_insn) &&
2920 cmdlen != F_INSN_SIZE(ipfw_insn_u32))
2921 goto bad_size;
2922 break;
2923
2924 case O_MACADDR2:
2925 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
2926 goto bad_size;
2927 break;
2928
2929 case O_NOP:
2930 case O_IPID:
2931 case O_IPTTL:
2932 case O_IPLEN:
2933 if (cmdlen < 1 || cmdlen > 31)
2934 goto bad_size;
2935 break;
2936
2937 case O_MAC_TYPE:
2938 case O_IP_SRCPORT:
2939 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
2940 if (cmdlen < 2 || cmdlen > 31)
2941 goto bad_size;
2942 break;
2943
2944 case O_RECV:
2945 case O_XMIT:
2946 case O_VIA:
2947 if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
2948 goto bad_size;
2949 break;
2950
2951 case O_PIPE:
2952 case O_QUEUE:
2953 if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
2954 goto bad_size;
2955 goto check_action;
2956
2957 case O_FORWARD_IP:
2958#ifdef IPFIREWALL_FORWARD
2959 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
2960 goto bad_size;
2961 goto check_action;
2962#else
2963 return EINVAL;
2964#endif
2965
2966 case O_DIVERT:
2967 case O_TEE:
2968#ifndef IPDIVERT
2969 return EINVAL;
2970#endif
2971 case O_FORWARD_MAC: /* XXX not implemented yet */
2972 case O_CHECK_STATE:
2973 case O_COUNT:
2974 case O_ACCEPT:
2975 case O_DENY:
2976 case O_REJECT:
2977 case O_SKIPTO:
2978 if (cmdlen != F_INSN_SIZE(ipfw_insn))
2979 goto bad_size;
2980check_action:
2981 if (have_action) {
2982 printf("ipfw: opcode %d, multiple actions"
2983 " not allowed\n",
2984 cmd->opcode);
2985 return EINVAL;
2986 }
2987 have_action = 1;
2988 if (l != cmdlen) {
2989 printf("ipfw: opcode %d, action must be"
2990 " last opcode\n",
2991 cmd->opcode);
2992 return EINVAL;
2993 }
2994 break;
2995 default:
2996 printf("ipfw: opcode %d, unknown opcode\n",
2997 cmd->opcode);
2998 return EINVAL;
2999 }
3000 }
3001 if (have_action == 0) {
3002 printf("ipfw: missing action\n");
3003 return EINVAL;
3004 }
3005 return 0;
3006
3007bad_size:
3008 printf("ipfw: opcode %d size %d wrong\n",
3009 cmd->opcode, cmdlen);
3010 return EINVAL;
3011}
3012
3013/*
3014 * Copy the static and dynamic rules to the supplied buffer
3015 * and return the amount of space actually used.
3016 */
3017static size_t
3018ipfw_getrules(struct ip_fw_chain *chain, void *buf, size_t space)
3019{
3020 char *bp = buf;
3021 char *ep = bp + space;
3022 struct ip_fw *rule;
3023 int i;
3024
3025 /* XXX this can take a long time and locking will block packet flow */
3026 IPFW_LOCK(chain);
3027 for (rule = chain->rules; rule ; rule = rule->next) {
3028 /*
3029 * Verify the entry fits in the buffer in case the
3030 * rules changed between calculating buffer space and
3031 * now. This would be better done using a generation
3032 * number but should suffice for now.
3033 */
3034 i = RULESIZE(rule);
3035 if (bp + i <= ep) {
3036 bcopy(rule, bp, i);
3037 bcopy(&set_disable, &(((struct ip_fw *)bp)->next_rule),
3038 sizeof(set_disable));
3039 bp += i;
3040 }
3041 }
3042 IPFW_UNLOCK(chain);
3043 if (ipfw_dyn_v) {
3044 ipfw_dyn_rule *p, *last = NULL;
3045
3046 IPFW_DYN_LOCK();
3047 for (i = 0 ; i < curr_dyn_buckets; i++)
3048 for (p = ipfw_dyn_v[i] ; p != NULL; p = p->next) {
3049 if (bp + sizeof *p <= ep) {
3050 ipfw_dyn_rule *dst =
3051 (ipfw_dyn_rule *)bp;
3052 bcopy(p, dst, sizeof *p);
3053 bcopy(&(p->rule->rulenum), &(dst->rule),
3054 sizeof(p->rule->rulenum));
3055 /*
3056 * store a non-null value in "next".
3057 * The userland code will interpret a
3058 * NULL here as a marker
3059 * for the last dynamic rule.
3060 */
3061 bcopy(&dst, &dst->next, sizeof(dst));
3062 last = dst;
3063 dst->expire =
3064 TIME_LEQ(dst->expire, time_second) ?
3065 0 : dst->expire - time_second ;
3066 bp += sizeof(ipfw_dyn_rule);
3067 }
3068 }
3069 IPFW_DYN_UNLOCK();
3070 if (last != NULL) /* mark last dynamic rule */
3071 bzero(&last->next, sizeof(last));
3072 }
3073 return (bp - (char *)buf);
3074}
3075
3076
3077/**
3078 * {set|get}sockopt parser.
3079 */
3080static int
3081ipfw_ctl(struct sockopt *sopt)
3082{
3083#define RULE_MAXSIZE (256*sizeof(u_int32_t))
3084 int error, rule_num;
3085 size_t size;
3086 struct ip_fw *buf, *rule;
3087 u_int32_t rulenum[2];
3088
3089 error = suser(sopt->sopt_td);
3090 if (error)
3091 return (error);
3092
3093 /*
3094 * Disallow modifications in really-really secure mode, but still allow
3095 * the logging counters to be reset.
3096 */
3097 if (sopt->sopt_name == IP_FW_ADD ||
3098 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
3099#if __FreeBSD_version >= 500034
3100 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
3101 if (error)
3102 return (error);
3103#else /* FreeBSD 4.x */
3104 if (securelevel >= 3)
3105 return (EPERM);
3106#endif
3107 }
3108
3109 error = 0;
3110
3111 switch (sopt->sopt_name) {
3112 case IP_FW_GET:
3113 /*
3114 * pass up a copy of the current rules. Static rules
3115 * come first (the last of which has number IPFW_DEFAULT_RULE),
3116 * followed by a possibly empty list of dynamic rule.
3117 * The last dynamic rule has NULL in the "next" field.
3118 *
3119 * Note that the calculated size is used to bound the
3120 * amount of data returned to the user. The rule set may
3121 * change between calculating the size and returning the
3122 * data in which case we'll just return what fits.
3123 */
3124 size = static_len; /* size of static rules */
3125 if (ipfw_dyn_v) /* add size of dyn.rules */
3126 size += (dyn_count * sizeof(ipfw_dyn_rule));
3127
3128 /*
3129 * XXX todo: if the user passes a short length just to know
3130 * how much room is needed, do not bother filling up the
3131 * buffer, just jump to the sooptcopyout.
3132 */
3133 buf = malloc(size, M_TEMP, M_WAITOK);
3134 error = sooptcopyout(sopt, buf,
3135 ipfw_getrules(&layer3_chain, buf, size));
3136 free(buf, M_TEMP);
3137 break;
3138
3139 case IP_FW_FLUSH:
3140 /*
3141 * Normally we cannot release the lock on each iteration.
3142 * We could do it here only because we start from the head all
3143 * the times so there is no risk of missing some entries.
3144 * On the other hand, the risk is that we end up with
3145 * a very inconsistent ruleset, so better keep the lock
3146 * around the whole cycle.
3147 *
3148 * XXX this code can be improved by resetting the head of
3149 * the list to point to the default rule, and then freeing
3150 * the old list without the need for a lock.
3151 */
3152
3153 IPFW_LOCK(&layer3_chain);
3154 layer3_chain.reap = NULL;
3155 free_chain(&layer3_chain, 0 /* keep default rule */);
3156 rule = layer3_chain.reap, layer3_chain.reap = NULL;
3157 IPFW_UNLOCK(&layer3_chain);
3158 if (layer3_chain.reap != NULL)
3159 reap_rules(rule);
3160 break;
3161
3162 case IP_FW_ADD:
3163 rule = malloc(RULE_MAXSIZE, M_TEMP, M_WAITOK);
3164 error = sooptcopyin(sopt, rule, RULE_MAXSIZE,
3165 sizeof(struct ip_fw) );
3166 if (error == 0)
3167 error = check_ipfw_struct(rule, sopt->sopt_valsize);
3168 if (error == 0) {
3169 error = add_rule(&layer3_chain, rule);
3170 size = RULESIZE(rule);
3171 if (!error && sopt->sopt_dir == SOPT_GET)
3172 error = sooptcopyout(sopt, rule, size);
3173 }
3174 free(rule, M_TEMP);
3175 break;
3176
3177 case IP_FW_DEL:
3178 /*
3179 * IP_FW_DEL is used for deleting single rules or sets,
3180 * and (ab)used to atomically manipulate sets. Argument size
3181 * is used to distinguish between the two:
3182 * sizeof(u_int32_t)
3183 * delete single rule or set of rules,
3184 * or reassign rules (or sets) to a different set.
3185 * 2*sizeof(u_int32_t)
3186 * atomic disable/enable sets.
3187 * first u_int32_t contains sets to be disabled,
3188 * second u_int32_t contains sets to be enabled.
3189 */
3190 error = sooptcopyin(sopt, rulenum,
3191 2*sizeof(u_int32_t), sizeof(u_int32_t));
3192 if (error)
3193 break;
3194 size = sopt->sopt_valsize;
3195 if (size == sizeof(u_int32_t)) /* delete or reassign */
3196 error = del_entry(&layer3_chain, rulenum[0]);
3197 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */
3198 set_disable =
3199 (set_disable | rulenum[0]) & ~rulenum[1] &
3200 ~(1<<RESVD_SET); /* set RESVD_SET always enabled */
3201 else
3202 error = EINVAL;
3203 break;
3204
3205 case IP_FW_ZERO:
3206 case IP_FW_RESETLOG: /* argument is an int, the rule number */
3207 rule_num = 0;
3208 if (sopt->sopt_val != 0) {
3209 error = sooptcopyin(sopt, &rule_num,
3210 sizeof(int), sizeof(int));
3211 if (error)
3212 break;
3213 }
3214 error = zero_entry(&layer3_chain, rule_num,
3215 sopt->sopt_name == IP_FW_RESETLOG);
3216 break;
3217
3218 case IP_FW_TABLE_ADD:
3219 {
3220 ipfw_table_entry ent;
3221
3222 error = sooptcopyin(sopt, &ent,
3223 sizeof(ent), sizeof(ent));
3224 if (error)
3225 break;
3226 error = add_table_entry(ent.tbl, ent.addr,
3227 ent.masklen, ent.value);
3228 }
3229 break;
3230
3231 case IP_FW_TABLE_DEL:
3232 {
3233 ipfw_table_entry ent;
3234
3235 error = sooptcopyin(sopt, &ent,
3236 sizeof(ent), sizeof(ent));
3237 if (error)
3238 break;
3239 error = del_table_entry(ent.tbl, ent.addr, ent.masklen);
3240 }
3241 break;
3242
3243 case IP_FW_TABLE_FLUSH:
3244 {
3245 u_int16_t tbl;
3246
3247 error = sooptcopyin(sopt, &tbl,
3248 sizeof(tbl), sizeof(tbl));
3249 if (error)
3250 break;
3251 error = flush_table(tbl);
3252 }
3253 break;
3254
3255 case IP_FW_TABLE_GETSIZE:
3256 {
3257 u_int32_t tbl, cnt;
3258
3259 if ((error = sooptcopyin(sopt, &tbl, sizeof(tbl),
3260 sizeof(tbl))))
3261 break;
3262 if ((error = count_table(tbl, &cnt)))
3263 break;
3264 error = sooptcopyout(sopt, &cnt, sizeof(cnt));
3265 }
3266 break;
3267
3268 case IP_FW_TABLE_LIST:
3269 {
3270 ipfw_table *tbl;
3271
3272 if (sopt->sopt_valsize < sizeof(*tbl)) {
3273 error = EINVAL;
3274 break;
3275 }
3276 size = sopt->sopt_valsize;
3277 tbl = malloc(size, M_TEMP, M_WAITOK);
3278 if (tbl == NULL) {
3279 error = ENOMEM;
3280 break;
3281 }
3282 error = sooptcopyin(sopt, tbl, size, sizeof(*tbl));
3283 if (error) {
3284 free(tbl, M_TEMP);
3285 break;
3286 }
3287 tbl->size = (size - sizeof(*tbl)) /
3288 sizeof(ipfw_table_entry);
3289 error = dump_table(tbl);
3290 if (error) {
3291 free(tbl, M_TEMP);
3292 break;
3293 }
3294 error = sooptcopyout(sopt, tbl, size);
3295 free(tbl, M_TEMP);
3296 }
3297 break;
3298
3299 default:
3300 printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name);
3301 error = EINVAL;
3302 }
3303
3304 return (error);
3305#undef RULE_MAXSIZE
3306}
3307
3308/**
3309 * dummynet needs a reference to the default rule, because rules can be
3310 * deleted while packets hold a reference to them. When this happens,
3311 * dummynet changes the reference to the default rule (it could well be a
3312 * NULL pointer, but this way we do not need to check for the special
3313 * case, plus here he have info on the default behaviour).
3314 */
3315struct ip_fw *ip_fw_default_rule;
3316
3317/*
3318 * This procedure is only used to handle keepalives. It is invoked
3319 * every dyn_keepalive_period
3320 */
3321static void
3322ipfw_tick(void * __unused unused)
3323{
3324 int i;
3325 ipfw_dyn_rule *q;
3326
3327 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
3328 goto done;
3329
3330 IPFW_DYN_LOCK();
3331 for (i = 0 ; i < curr_dyn_buckets ; i++) {
3332 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
3333 if (q->dyn_type == O_LIMIT_PARENT)
3334 continue;
3335 if (q->id.proto != IPPROTO_TCP)
3336 continue;
3337 if ( (q->state & BOTH_SYN) != BOTH_SYN)
3338 continue;
3339 if (TIME_LEQ( time_second+dyn_keepalive_interval,
3340 q->expire))
3341 continue; /* too early */
3342 if (TIME_LEQ(q->expire, time_second))
3343 continue; /* too late, rule expired */
3344
3345 send_pkt(&(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN);
3346 send_pkt(&(q->id), q->ack_fwd - 1, q->ack_rev, 0);
3347 }
3348 }
3349 IPFW_DYN_UNLOCK();
3350done:
3351 callout_reset(&ipfw_timeout, dyn_keepalive_period*hz, ipfw_tick, NULL);
3352}
3353
3354int
3355ipfw_init(void)
3356{
3357 struct ip_fw default_rule;
3358 int error;
3359
3360 layer3_chain.rules = NULL;
3361 IPFW_LOCK_INIT(&layer3_chain);
3362 IPFW_DYN_LOCK_INIT();
3363 callout_init(&ipfw_timeout, debug_mpsafenet ? CALLOUT_MPSAFE : 0);
3364
3365 bzero(&default_rule, sizeof default_rule);
3366
3367 default_rule.act_ofs = 0;
3368 default_rule.rulenum = IPFW_DEFAULT_RULE;
3369 default_rule.cmd_len = 1;
3370 default_rule.set = RESVD_SET;
3371
3372 default_rule.cmd[0].len = 1;
3373 default_rule.cmd[0].opcode =
3374#ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
3375 1 ? O_ACCEPT :
3376#endif
3377 O_DENY;
3378
3379 error = add_rule(&layer3_chain, &default_rule);
3380 if (error != 0) {
3381 printf("ipfw2: error %u initializing default rule "
3382 "(support disabled)\n", error);
3383 IPFW_DYN_LOCK_DESTROY();
3384 IPFW_LOCK_DESTROY(&layer3_chain);
3385 return (error);
3386 }
3387
3388 ip_fw_default_rule = layer3_chain.rules;
3389 printf("ipfw2 initialized, divert %s, "
3390 "rule-based forwarding "
3391#ifdef IPFIREWALL_FORWARD
3392 "enabled, "
3393#else
3394 "disabled, "
3395#endif
3396 "default to %s, logging ",
3397#ifdef IPDIVERT
3398 "enabled",
3399#else
3400 "disabled",
3401#endif
3402 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny");
3403
3404#ifdef IPFIREWALL_VERBOSE
3405 fw_verbose = 1;
3406#endif
3407#ifdef IPFIREWALL_VERBOSE_LIMIT
3408 verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
3409#endif
3410 if (fw_verbose == 0)
3411 printf("disabled\n");
3412 else if (verbose_limit == 0)
3413 printf("unlimited\n");
3414 else
3415 printf("limited to %d packets/entry by default\n",
3416 verbose_limit);
3417
3418 init_tables();
3419 ip_fw_ctl_ptr = ipfw_ctl;
3420 ip_fw_chk_ptr = ipfw_chk;
3421 callout_reset(&ipfw_timeout, hz, ipfw_tick, NULL);
3422
3423 return (0);
3424}
3425
3426void
3427ipfw_destroy(void)
3428{
3429 struct ip_fw *reap;
3430
3431 ip_fw_chk_ptr = NULL;
3432 ip_fw_ctl_ptr = NULL;
3433 callout_drain(&ipfw_timeout);
3434 IPFW_LOCK(&layer3_chain);
3435 layer3_chain.reap = NULL;
3436 free_chain(&layer3_chain, 1 /* kill default rule */);
3437 reap = layer3_chain.reap, layer3_chain.reap = NULL;
3438 IPFW_UNLOCK(&layer3_chain);
3439 if (reap != NULL)
3440 reap_rules(reap);
3441 flush_tables();
3442 IPFW_DYN_LOCK_DESTROY();
3443 IPFW_LOCK_DESTROY(&layer3_chain);
3444 printf("IP firewall unloaded\n");
3445}
3446
3447#endif /* IPFW2 */
| 1963 break;
1964
1965 case O_RECV:
1966 match = iface_match(m->m_pkthdr.rcvif,
1967 (ipfw_insn_if *)cmd);
1968 break;
1969
1970 case O_XMIT:
1971 match = iface_match(oif, (ipfw_insn_if *)cmd);
1972 break;
1973
1974 case O_VIA:
1975 match = iface_match(oif ? oif :
1976 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1977 break;
1978
1979 case O_MACADDR2:
1980 if (args->eh != NULL) { /* have MAC header */
1981 u_int32_t *want = (u_int32_t *)
1982 ((ipfw_insn_mac *)cmd)->addr;
1983 u_int32_t *mask = (u_int32_t *)
1984 ((ipfw_insn_mac *)cmd)->mask;
1985 u_int32_t *hdr = (u_int32_t *)args->eh;
1986
1987 match =
1988 ( want[0] == (hdr[0] & mask[0]) &&
1989 want[1] == (hdr[1] & mask[1]) &&
1990 want[2] == (hdr[2] & mask[2]) );
1991 }
1992 break;
1993
1994 case O_MAC_TYPE:
1995 if (args->eh != NULL) {
1996 u_int16_t t =
1997 ntohs(args->eh->ether_type);
1998 u_int16_t *p =
1999 ((ipfw_insn_u16 *)cmd)->ports;
2000 int i;
2001
2002 for (i = cmdlen - 1; !match && i>0;
2003 i--, p += 2)
2004 match = (t>=p[0] && t<=p[1]);
2005 }
2006 break;
2007
2008 case O_FRAG:
2009 match = (hlen > 0 && offset != 0);
2010 break;
2011
2012 case O_IN: /* "out" is "not in" */
2013 match = (oif == NULL);
2014 break;
2015
2016 case O_LAYER2:
2017 match = (args->eh != NULL);
2018 break;
2019
2020 case O_PROTO:
2021 /*
2022 * We do not allow an arg of 0 so the
2023 * check of "proto" only suffices.
2024 */
2025 match = (proto == cmd->arg1);
2026 break;
2027
2028 case O_IP_SRC:
2029 match = (hlen > 0 &&
2030 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2031 src_ip.s_addr);
2032 break;
2033
2034 case O_IP_SRC_LOOKUP:
2035 case O_IP_DST_LOOKUP:
2036 if (hlen > 0) {
2037 uint32_t a =
2038 (cmd->opcode == O_IP_DST_LOOKUP) ?
2039 dst_ip.s_addr : src_ip.s_addr;
2040 uint32_t v;
2041
2042 match = lookup_table(cmd->arg1, a, &v);
2043 if (!match)
2044 break;
2045 if (cmdlen == F_INSN_SIZE(ipfw_insn_u32))
2046 match =
2047 ((ipfw_insn_u32 *)cmd)->d[0] == v;
2048 }
2049 break;
2050
2051 case O_IP_SRC_MASK:
2052 case O_IP_DST_MASK:
2053 if (hlen > 0) {
2054 uint32_t a =
2055 (cmd->opcode == O_IP_DST_MASK) ?
2056 dst_ip.s_addr : src_ip.s_addr;
2057 uint32_t *p = ((ipfw_insn_u32 *)cmd)->d;
2058 int i = cmdlen-1;
2059
2060 for (; !match && i>0; i-= 2, p+= 2)
2061 match = (p[0] == (a & p[1]));
2062 }
2063 break;
2064
2065 case O_IP_SRC_ME:
2066 if (hlen > 0) {
2067 struct ifnet *tif;
2068
2069 INADDR_TO_IFP(src_ip, tif);
2070 match = (tif != NULL);
2071 }
2072 break;
2073
2074 case O_IP_DST_SET:
2075 case O_IP_SRC_SET:
2076 if (hlen > 0) {
2077 u_int32_t *d = (u_int32_t *)(cmd+1);
2078 u_int32_t addr =
2079 cmd->opcode == O_IP_DST_SET ?
2080 args->f_id.dst_ip :
2081 args->f_id.src_ip;
2082
2083 if (addr < d[0])
2084 break;
2085 addr -= d[0]; /* subtract base */
2086 match = (addr < cmd->arg1) &&
2087 ( d[ 1 + (addr>>5)] &
2088 (1<<(addr & 0x1f)) );
2089 }
2090 break;
2091
2092 case O_IP_DST:
2093 match = (hlen > 0 &&
2094 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2095 dst_ip.s_addr);
2096 break;
2097
2098 case O_IP_DST_ME:
2099 if (hlen > 0) {
2100 struct ifnet *tif;
2101
2102 INADDR_TO_IFP(dst_ip, tif);
2103 match = (tif != NULL);
2104 }
2105 break;
2106
2107 case O_IP_SRCPORT:
2108 case O_IP_DSTPORT:
2109 /*
2110 * offset == 0 && proto != 0 is enough
2111 * to guarantee that we have an IPv4
2112 * packet with port info.
2113 */
2114 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
2115 && offset == 0) {
2116 u_int16_t x =
2117 (cmd->opcode == O_IP_SRCPORT) ?
2118 src_port : dst_port ;
2119 u_int16_t *p =
2120 ((ipfw_insn_u16 *)cmd)->ports;
2121 int i;
2122
2123 for (i = cmdlen - 1; !match && i>0;
2124 i--, p += 2)
2125 match = (x>=p[0] && x<=p[1]);
2126 }
2127 break;
2128
2129 case O_ICMPTYPE:
2130 match = (offset == 0 && proto==IPPROTO_ICMP &&
2131 icmptype_match(ip, (ipfw_insn_u32 *)cmd) );
2132 break;
2133
2134 case O_IPOPT:
2135 match = (hlen > 0 && ipopts_match(ip, cmd) );
2136 break;
2137
2138 case O_IPVER:
2139 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
2140 break;
2141
2142 case O_IPID:
2143 case O_IPLEN:
2144 case O_IPTTL:
2145 if (hlen > 0) { /* only for IP packets */
2146 uint16_t x;
2147 uint16_t *p;
2148 int i;
2149
2150 if (cmd->opcode == O_IPLEN)
2151 x = ip_len;
2152 else if (cmd->opcode == O_IPTTL)
2153 x = ip->ip_ttl;
2154 else /* must be IPID */
2155 x = ntohs(ip->ip_id);
2156 if (cmdlen == 1) {
2157 match = (cmd->arg1 == x);
2158 break;
2159 }
2160 /* otherwise we have ranges */
2161 p = ((ipfw_insn_u16 *)cmd)->ports;
2162 i = cmdlen - 1;
2163 for (; !match && i>0; i--, p += 2)
2164 match = (x >= p[0] && x <= p[1]);
2165 }
2166 break;
2167
2168 case O_IPPRECEDENCE:
2169 match = (hlen > 0 &&
2170 (cmd->arg1 == (ip->ip_tos & 0xe0)) );
2171 break;
2172
2173 case O_IPTOS:
2174 match = (hlen > 0 &&
2175 flags_match(cmd, ip->ip_tos));
2176 break;
2177
2178 case O_TCPFLAGS:
2179 match = (proto == IPPROTO_TCP && offset == 0 &&
2180 flags_match(cmd,
2181 L3HDR(struct tcphdr,ip)->th_flags));
2182 break;
2183
2184 case O_TCPOPTS:
2185 match = (proto == IPPROTO_TCP && offset == 0 &&
2186 tcpopts_match(ip, cmd));
2187 break;
2188
2189 case O_TCPSEQ:
2190 match = (proto == IPPROTO_TCP && offset == 0 &&
2191 ((ipfw_insn_u32 *)cmd)->d[0] ==
2192 L3HDR(struct tcphdr,ip)->th_seq);
2193 break;
2194
2195 case O_TCPACK:
2196 match = (proto == IPPROTO_TCP && offset == 0 &&
2197 ((ipfw_insn_u32 *)cmd)->d[0] ==
2198 L3HDR(struct tcphdr,ip)->th_ack);
2199 break;
2200
2201 case O_TCPWIN:
2202 match = (proto == IPPROTO_TCP && offset == 0 &&
2203 cmd->arg1 ==
2204 L3HDR(struct tcphdr,ip)->th_win);
2205 break;
2206
2207 case O_ESTAB:
2208 /* reject packets which have SYN only */
2209 /* XXX should i also check for TH_ACK ? */
2210 match = (proto == IPPROTO_TCP && offset == 0 &&
2211 (L3HDR(struct tcphdr,ip)->th_flags &
2212 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2213 break;
2214
2215 case O_LOG:
2216 if (fw_verbose)
2217 ipfw_log(f, hlen, args->eh, m, oif);
2218 match = 1;
2219 break;
2220
2221 case O_PROB:
2222 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]);
2223 break;
2224
2225 case O_VERREVPATH:
2226 /* Outgoing packets automatically pass/match */
2227 match = (hlen > 0 && ((oif != NULL) ||
2228 (m->m_pkthdr.rcvif == NULL) ||
2229 verify_path(src_ip, m->m_pkthdr.rcvif)));
2230 break;
2231
2232 case O_VERSRCREACH:
2233 /* Outgoing packets automatically pass/match */
2234 match = (hlen > 0 && ((oif != NULL) ||
2235 verify_path(src_ip, NULL)));
2236 break;
2237
2238 case O_ANTISPOOF:
2239 /* Outgoing packets automatically pass/match */
2240 if (oif == NULL && hlen > 0 &&
2241 in_localaddr(src_ip))
2242 match = verify_path(src_ip,
2243 m->m_pkthdr.rcvif);
2244 else
2245 match = 1;
2246 break;
2247
2248 case O_IPSEC:
2249#ifdef FAST_IPSEC
2250 match = (m_tag_find(m,
2251 PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL);
2252#endif
2253#ifdef IPSEC
2254 match = (ipsec_getnhist(m) != 0);
2255#endif
2256 /* otherwise no match */
2257 break;
2258
2259 /*
2260 * The second set of opcodes represents 'actions',
2261 * i.e. the terminal part of a rule once the packet
2262 * matches all previous patterns.
2263 * Typically there is only one action for each rule,
2264 * and the opcode is stored at the end of the rule
2265 * (but there are exceptions -- see below).
2266 *
2267 * In general, here we set retval and terminate the
2268 * outer loop (would be a 'break 3' in some language,
2269 * but we need to do a 'goto done').
2270 *
2271 * Exceptions:
2272 * O_COUNT and O_SKIPTO actions:
2273 * instead of terminating, we jump to the next rule
2274 * ('goto next_rule', equivalent to a 'break 2'),
2275 * or to the SKIPTO target ('goto again' after
2276 * having set f, cmd and l), respectively.
2277 *
2278 * O_LIMIT and O_KEEP_STATE: these opcodes are
2279 * not real 'actions', and are stored right
2280 * before the 'action' part of the rule.
2281 * These opcodes try to install an entry in the
2282 * state tables; if successful, we continue with
2283 * the next opcode (match=1; break;), otherwise
2284 * the packet * must be dropped
2285 * ('goto done' after setting retval);
2286 *
2287 * O_PROBE_STATE and O_CHECK_STATE: these opcodes
2288 * cause a lookup of the state table, and a jump
2289 * to the 'action' part of the parent rule
2290 * ('goto check_body') if an entry is found, or
2291 * (CHECK_STATE only) a jump to the next rule if
2292 * the entry is not found ('goto next_rule').
2293 * The result of the lookup is cached to make
2294 * further instances of these opcodes are
2295 * effectively NOPs.
2296 */
2297 case O_LIMIT:
2298 case O_KEEP_STATE:
2299 if (install_state(f,
2300 (ipfw_insn_limit *)cmd, args)) {
2301 retval = IP_FW_PORT_DENY_FLAG;
2302 goto done; /* error/limit violation */
2303 }
2304 match = 1;
2305 break;
2306
2307 case O_PROBE_STATE:
2308 case O_CHECK_STATE:
2309 /*
2310 * dynamic rules are checked at the first
2311 * keep-state or check-state occurrence,
2312 * with the result being stored in dyn_dir.
2313 * The compiler introduces a PROBE_STATE
2314 * instruction for us when we have a
2315 * KEEP_STATE (because PROBE_STATE needs
2316 * to be run first).
2317 */
2318 if (dyn_dir == MATCH_UNKNOWN &&
2319 (q = lookup_dyn_rule(&args->f_id,
2320 &dyn_dir, proto == IPPROTO_TCP ?
2321 L3HDR(struct tcphdr, ip) : NULL))
2322 != NULL) {
2323 /*
2324 * Found dynamic entry, update stats
2325 * and jump to the 'action' part of
2326 * the parent rule.
2327 */
2328 q->pcnt++;
2329 q->bcnt += pktlen;
2330 f = q->rule;
2331 cmd = ACTION_PTR(f);
2332 l = f->cmd_len - f->act_ofs;
2333 IPFW_DYN_UNLOCK();
2334 goto check_body;
2335 }
2336 /*
2337 * Dynamic entry not found. If CHECK_STATE,
2338 * skip to next rule, if PROBE_STATE just
2339 * ignore and continue with next opcode.
2340 */
2341 if (cmd->opcode == O_CHECK_STATE)
2342 goto next_rule;
2343 match = 1;
2344 break;
2345
2346 case O_ACCEPT:
2347 retval = 0; /* accept */
2348 goto done;
2349
2350 case O_PIPE:
2351 case O_QUEUE:
2352 args->rule = f; /* report matching rule */
2353 retval = cmd->arg1 | IP_FW_PORT_DYNT_FLAG;
2354 goto done;
2355
2356 case O_DIVERT:
2357 case O_TEE: {
2358 struct divert_tag *dt;
2359
2360 if (args->eh) /* not on layer 2 */
2361 break;
2362 mtag = m_tag_get(PACKET_TAG_DIVERT,
2363 sizeof(struct divert_tag),
2364 M_NOWAIT);
2365 if (mtag == NULL) {
2366 /* XXX statistic */
2367 /* drop packet */
2368 IPFW_UNLOCK(chain);
2369 return IP_FW_PORT_DENY_FLAG;
2370 }
2371 dt = (struct divert_tag *)(mtag+1);
2372 dt->cookie = f->rulenum;
2373 dt->info = (cmd->opcode == O_DIVERT) ?
2374 cmd->arg1 :
2375 cmd->arg1 | IP_FW_PORT_TEE_FLAG;
2376 m_tag_prepend(m, mtag);
2377 retval = dt->info;
2378 goto done;
2379 }
2380
2381 case O_COUNT:
2382 case O_SKIPTO:
2383 f->pcnt++; /* update stats */
2384 f->bcnt += pktlen;
2385 f->timestamp = time_second;
2386 if (cmd->opcode == O_COUNT)
2387 goto next_rule;
2388 /* handle skipto */
2389 if (f->next_rule == NULL)
2390 lookup_next_rule(f);
2391 f = f->next_rule;
2392 goto again;
2393
2394 case O_REJECT:
2395 /*
2396 * Drop the packet and send a reject notice
2397 * if the packet is not ICMP (or is an ICMP
2398 * query), and it is not multicast/broadcast.
2399 */
2400 if (hlen > 0 &&
2401 (proto != IPPROTO_ICMP ||
2402 is_icmp_query(ip)) &&
2403 !(m->m_flags & (M_BCAST|M_MCAST)) &&
2404 !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
2405 send_reject(args, cmd->arg1,
2406 offset,ip_len);
2407 m = args->m;
2408 }
2409 /* FALLTHROUGH */
2410 case O_DENY:
2411 retval = IP_FW_PORT_DENY_FLAG;
2412 goto done;
2413
2414 case O_FORWARD_IP:
2415 if (args->eh) /* not valid on layer2 pkts */
2416 break;
2417 if (!q || dyn_dir == MATCH_FORWARD)
2418 args->next_hop =
2419 &((ipfw_insn_sa *)cmd)->sa;
2420 retval = 0;
2421 goto done;
2422
2423 default:
2424 panic("-- unknown opcode %d\n", cmd->opcode);
2425 } /* end of switch() on opcodes */
2426
2427 if (cmd->len & F_NOT)
2428 match = !match;
2429
2430 if (match) {
2431 if (cmd->len & F_OR)
2432 skip_or = 1;
2433 } else {
2434 if (!(cmd->len & F_OR)) /* not an OR block, */
2435 break; /* try next rule */
2436 }
2437
2438 } /* end of inner for, scan opcodes */
2439
2440next_rule:; /* try next rule */
2441
2442 } /* end of outer for, scan rules */
2443 printf("ipfw: ouch!, skip past end of rules, denying packet\n");
2444 IPFW_UNLOCK(chain);
2445 return(IP_FW_PORT_DENY_FLAG);
2446
2447done:
2448 /* Update statistics */
2449 f->pcnt++;
2450 f->bcnt += pktlen;
2451 f->timestamp = time_second;
2452 IPFW_UNLOCK(chain);
2453 return retval;
2454
2455pullup_failed:
2456 if (fw_verbose)
2457 printf("ipfw: pullup failed\n");
2458 return(IP_FW_PORT_DENY_FLAG);
2459}
2460
2461/*
2462 * When a rule is added/deleted, clear the next_rule pointers in all rules.
2463 * These will be reconstructed on the fly as packets are matched.
2464 */
2465static void
2466flush_rule_ptrs(struct ip_fw_chain *chain)
2467{
2468 struct ip_fw *rule;
2469
2470 IPFW_LOCK_ASSERT(chain);
2471
2472 for (rule = chain->rules; rule; rule = rule->next)
2473 rule->next_rule = NULL;
2474}
2475
2476/*
2477 * When pipes/queues are deleted, clear the "pipe_ptr" pointer to a given
2478 * pipe/queue, or to all of them (match == NULL).
2479 */
2480void
2481flush_pipe_ptrs(struct dn_flow_set *match)
2482{
2483 struct ip_fw *rule;
2484
2485 IPFW_LOCK(&layer3_chain);
2486 for (rule = layer3_chain.rules; rule; rule = rule->next) {
2487 ipfw_insn_pipe *cmd = (ipfw_insn_pipe *)ACTION_PTR(rule);
2488
2489 if (cmd->o.opcode != O_PIPE && cmd->o.opcode != O_QUEUE)
2490 continue;
2491 /*
2492 * XXX Use bcmp/bzero to handle pipe_ptr to overcome
2493 * possible alignment problems on 64-bit architectures.
2494 * This code is seldom used so we do not worry too
2495 * much about efficiency.
2496 */
2497 if (match == NULL ||
2498 !bcmp(&cmd->pipe_ptr, &match, sizeof(match)) )
2499 bzero(&cmd->pipe_ptr, sizeof(cmd->pipe_ptr));
2500 }
2501 IPFW_UNLOCK(&layer3_chain);
2502}
2503
2504/*
2505 * Add a new rule to the list. Copy the rule into a malloc'ed area, then
2506 * possibly create a rule number and add the rule to the list.
2507 * Update the rule_number in the input struct so the caller knows it as well.
2508 */
2509static int
2510add_rule(struct ip_fw_chain *chain, struct ip_fw *input_rule)
2511{
2512 struct ip_fw *rule, *f, *prev;
2513 int l = RULESIZE(input_rule);
2514
2515 if (chain->rules == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE)
2516 return (EINVAL);
2517
2518 rule = malloc(l, M_IPFW, M_NOWAIT | M_ZERO);
2519 if (rule == NULL)
2520 return (ENOSPC);
2521
2522 bcopy(input_rule, rule, l);
2523
2524 rule->next = NULL;
2525 rule->next_rule = NULL;
2526
2527 rule->pcnt = 0;
2528 rule->bcnt = 0;
2529 rule->timestamp = 0;
2530
2531 IPFW_LOCK(chain);
2532
2533 if (chain->rules == NULL) { /* default rule */
2534 chain->rules = rule;
2535 goto done;
2536 }
2537
2538 /*
2539 * If rulenum is 0, find highest numbered rule before the
2540 * default rule, and add autoinc_step
2541 */
2542 if (autoinc_step < 1)
2543 autoinc_step = 1;
2544 else if (autoinc_step > 1000)
2545 autoinc_step = 1000;
2546 if (rule->rulenum == 0) {
2547 /*
2548 * locate the highest numbered rule before default
2549 */
2550 for (f = chain->rules; f; f = f->next) {
2551 if (f->rulenum == IPFW_DEFAULT_RULE)
2552 break;
2553 rule->rulenum = f->rulenum;
2554 }
2555 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
2556 rule->rulenum += autoinc_step;
2557 input_rule->rulenum = rule->rulenum;
2558 }
2559
2560 /*
2561 * Now insert the new rule in the right place in the sorted list.
2562 */
2563 for (prev = NULL, f = chain->rules; f; prev = f, f = f->next) {
2564 if (f->rulenum > rule->rulenum) { /* found the location */
2565 if (prev) {
2566 rule->next = f;
2567 prev->next = rule;
2568 } else { /* head insert */
2569 rule->next = chain->rules;
2570 chain->rules = rule;
2571 }
2572 break;
2573 }
2574 }
2575 flush_rule_ptrs(chain);
2576done:
2577 static_count++;
2578 static_len += l;
2579 IPFW_UNLOCK(chain);
2580 DEB(printf("ipfw: installed rule %d, static count now %d\n",
2581 rule->rulenum, static_count);)
2582 return (0);
2583}
2584
2585/**
2586 * Remove a static rule (including derived * dynamic rules)
2587 * and place it on the ``reap list'' for later reclamation.
2588 * The caller is in charge of clearing rule pointers to avoid
2589 * dangling pointers.
2590 * @return a pointer to the next entry.
2591 * Arguments are not checked, so they better be correct.
2592 */
2593static struct ip_fw *
2594remove_rule(struct ip_fw_chain *chain, struct ip_fw *rule, struct ip_fw *prev)
2595{
2596 struct ip_fw *n;
2597 int l = RULESIZE(rule);
2598
2599 IPFW_LOCK_ASSERT(chain);
2600
2601 n = rule->next;
2602 IPFW_DYN_LOCK();
2603 remove_dyn_rule(rule, NULL /* force removal */);
2604 IPFW_DYN_UNLOCK();
2605 if (prev == NULL)
2606 chain->rules = n;
2607 else
2608 prev->next = n;
2609 static_count--;
2610 static_len -= l;
2611
2612 rule->next = chain->reap;
2613 chain->reap = rule;
2614
2615 return n;
2616}
2617
2618/**
2619 * Reclaim storage associated with a list of rules. This is
2620 * typically the list created using remove_rule.
2621 */
2622static void
2623reap_rules(struct ip_fw *head)
2624{
2625 struct ip_fw *rule;
2626
2627 while ((rule = head) != NULL) {
2628 head = head->next;
2629 if (DUMMYNET_LOADED)
2630 ip_dn_ruledel_ptr(rule);
2631 free(rule, M_IPFW);
2632 }
2633}
2634
2635/*
2636 * Remove all rules from a chain (except rules in set RESVD_SET
2637 * unless kill_default = 1). The caller is responsible for
2638 * reclaiming storage for the rules left in chain->reap.
2639 */
2640static void
2641free_chain(struct ip_fw_chain *chain, int kill_default)
2642{
2643 struct ip_fw *prev, *rule;
2644
2645 IPFW_LOCK_ASSERT(chain);
2646
2647 flush_rule_ptrs(chain); /* more efficient to do outside the loop */
2648 for (prev = NULL, rule = chain->rules; rule ; )
2649 if (kill_default || rule->set != RESVD_SET)
2650 rule = remove_rule(chain, rule, prev);
2651 else {
2652 prev = rule;
2653 rule = rule->next;
2654 }
2655}
2656
2657/**
2658 * Remove all rules with given number, and also do set manipulation.
2659 * Assumes chain != NULL && *chain != NULL.
2660 *
2661 * The argument is an u_int32_t. The low 16 bit are the rule or set number,
2662 * the next 8 bits are the new set, the top 8 bits are the command:
2663 *
2664 * 0 delete rules with given number
2665 * 1 delete rules with given set number
2666 * 2 move rules with given number to new set
2667 * 3 move rules with given set number to new set
2668 * 4 swap sets with given numbers
2669 */
2670static int
2671del_entry(struct ip_fw_chain *chain, u_int32_t arg)
2672{
2673 struct ip_fw *prev = NULL, *rule;
2674 u_int16_t rulenum; /* rule or old_set */
2675 u_int8_t cmd, new_set;
2676
2677 rulenum = arg & 0xffff;
2678 cmd = (arg >> 24) & 0xff;
2679 new_set = (arg >> 16) & 0xff;
2680
2681 if (cmd > 4)
2682 return EINVAL;
2683 if (new_set > RESVD_SET)
2684 return EINVAL;
2685 if (cmd == 0 || cmd == 2) {
2686 if (rulenum >= IPFW_DEFAULT_RULE)
2687 return EINVAL;
2688 } else {
2689 if (rulenum > RESVD_SET) /* old_set */
2690 return EINVAL;
2691 }
2692
2693 IPFW_LOCK(chain);
2694 rule = chain->rules;
2695 chain->reap = NULL;
2696 switch (cmd) {
2697 case 0: /* delete rules with given number */
2698 /*
2699 * locate first rule to delete
2700 */
2701 for (; rule->rulenum < rulenum; prev = rule, rule = rule->next)
2702 ;
2703 if (rule->rulenum != rulenum) {
2704 IPFW_UNLOCK(chain);
2705 return EINVAL;
2706 }
2707
2708 /*
2709 * flush pointers outside the loop, then delete all matching
2710 * rules. prev remains the same throughout the cycle.
2711 */
2712 flush_rule_ptrs(chain);
2713 while (rule->rulenum == rulenum)
2714 rule = remove_rule(chain, rule, prev);
2715 break;
2716
2717 case 1: /* delete all rules with given set number */
2718 flush_rule_ptrs(chain);
2719 rule = chain->rules;
2720 while (rule->rulenum < IPFW_DEFAULT_RULE)
2721 if (rule->set == rulenum)
2722 rule = remove_rule(chain, rule, prev);
2723 else {
2724 prev = rule;
2725 rule = rule->next;
2726 }
2727 break;
2728
2729 case 2: /* move rules with given number to new set */
2730 rule = chain->rules;
2731 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
2732 if (rule->rulenum == rulenum)
2733 rule->set = new_set;
2734 break;
2735
2736 case 3: /* move rules with given set number to new set */
2737 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
2738 if (rule->set == rulenum)
2739 rule->set = new_set;
2740 break;
2741
2742 case 4: /* swap two sets */
2743 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
2744 if (rule->set == rulenum)
2745 rule->set = new_set;
2746 else if (rule->set == new_set)
2747 rule->set = rulenum;
2748 break;
2749 }
2750 /*
2751 * Look for rules to reclaim. We grab the list before
2752 * releasing the lock then reclaim them w/o the lock to
2753 * avoid a LOR with dummynet.
2754 */
2755 rule = chain->reap;
2756 chain->reap = NULL;
2757 IPFW_UNLOCK(chain);
2758 if (rule)
2759 reap_rules(rule);
2760 return 0;
2761}
2762
2763/*
2764 * Clear counters for a specific rule.
2765 * The enclosing "table" is assumed locked.
2766 */
2767static void
2768clear_counters(struct ip_fw *rule, int log_only)
2769{
2770 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
2771
2772 if (log_only == 0) {
2773 rule->bcnt = rule->pcnt = 0;
2774 rule->timestamp = 0;
2775 }
2776 if (l->o.opcode == O_LOG)
2777 l->log_left = l->max_log;
2778}
2779
2780/**
2781 * Reset some or all counters on firewall rules.
2782 * @arg frwl is null to clear all entries, or contains a specific
2783 * rule number.
2784 * @arg log_only is 1 if we only want to reset logs, zero otherwise.
2785 */
2786static int
2787zero_entry(struct ip_fw_chain *chain, int rulenum, int log_only)
2788{
2789 struct ip_fw *rule;
2790 char *msg;
2791
2792 IPFW_LOCK(chain);
2793 if (rulenum == 0) {
2794 norule_counter = 0;
2795 for (rule = chain->rules; rule; rule = rule->next)
2796 clear_counters(rule, log_only);
2797 msg = log_only ? "ipfw: All logging counts reset.\n" :
2798 "ipfw: Accounting cleared.\n";
2799 } else {
2800 int cleared = 0;
2801 /*
2802 * We can have multiple rules with the same number, so we
2803 * need to clear them all.
2804 */
2805 for (rule = chain->rules; rule; rule = rule->next)
2806 if (rule->rulenum == rulenum) {
2807 while (rule && rule->rulenum == rulenum) {
2808 clear_counters(rule, log_only);
2809 rule = rule->next;
2810 }
2811 cleared = 1;
2812 break;
2813 }
2814 if (!cleared) { /* we did not find any matching rules */
2815 IPFW_UNLOCK(chain);
2816 return (EINVAL);
2817 }
2818 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
2819 "ipfw: Entry %d cleared.\n";
2820 }
2821 IPFW_UNLOCK(chain);
2822
2823 if (fw_verbose)
2824 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
2825 return (0);
2826}
2827
2828/*
2829 * Check validity of the structure before insert.
2830 * Fortunately rules are simple, so this mostly need to check rule sizes.
2831 */
2832static int
2833check_ipfw_struct(struct ip_fw *rule, int size)
2834{
2835 int l, cmdlen = 0;
2836 int have_action=0;
2837 ipfw_insn *cmd;
2838
2839 if (size < sizeof(*rule)) {
2840 printf("ipfw: rule too short\n");
2841 return (EINVAL);
2842 }
2843 /* first, check for valid size */
2844 l = RULESIZE(rule);
2845 if (l != size) {
2846 printf("ipfw: size mismatch (have %d want %d)\n", size, l);
2847 return (EINVAL);
2848 }
2849 /*
2850 * Now go for the individual checks. Very simple ones, basically only
2851 * instruction sizes.
2852 */
2853 for (l = rule->cmd_len, cmd = rule->cmd ;
2854 l > 0 ; l -= cmdlen, cmd += cmdlen) {
2855 cmdlen = F_LEN(cmd);
2856 if (cmdlen > l) {
2857 printf("ipfw: opcode %d size truncated\n",
2858 cmd->opcode);
2859 return EINVAL;
2860 }
2861 DEB(printf("ipfw: opcode %d\n", cmd->opcode);)
2862 switch (cmd->opcode) {
2863 case O_PROBE_STATE:
2864 case O_KEEP_STATE:
2865 case O_PROTO:
2866 case O_IP_SRC_ME:
2867 case O_IP_DST_ME:
2868 case O_LAYER2:
2869 case O_IN:
2870 case O_FRAG:
2871 case O_IPOPT:
2872 case O_IPTOS:
2873 case O_IPPRECEDENCE:
2874 case O_IPVER:
2875 case O_TCPWIN:
2876 case O_TCPFLAGS:
2877 case O_TCPOPTS:
2878 case O_ESTAB:
2879 case O_VERREVPATH:
2880 case O_VERSRCREACH:
2881 case O_ANTISPOOF:
2882 case O_IPSEC:
2883 if (cmdlen != F_INSN_SIZE(ipfw_insn))
2884 goto bad_size;
2885 break;
2886
2887 case O_UID:
2888 case O_GID:
2889 case O_JAIL:
2890 case O_IP_SRC:
2891 case O_IP_DST:
2892 case O_TCPSEQ:
2893 case O_TCPACK:
2894 case O_PROB:
2895 case O_ICMPTYPE:
2896 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
2897 goto bad_size;
2898 break;
2899
2900 case O_LIMIT:
2901 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
2902 goto bad_size;
2903 break;
2904
2905 case O_LOG:
2906 if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
2907 goto bad_size;
2908
2909 ((ipfw_insn_log *)cmd)->log_left =
2910 ((ipfw_insn_log *)cmd)->max_log;
2911
2912 break;
2913
2914 case O_IP_SRC_MASK:
2915 case O_IP_DST_MASK:
2916 /* only odd command lengths */
2917 if ( !(cmdlen & 1) || cmdlen > 31)
2918 goto bad_size;
2919 break;
2920
2921 case O_IP_SRC_SET:
2922 case O_IP_DST_SET:
2923 if (cmd->arg1 == 0 || cmd->arg1 > 256) {
2924 printf("ipfw: invalid set size %d\n",
2925 cmd->arg1);
2926 return EINVAL;
2927 }
2928 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
2929 (cmd->arg1+31)/32 )
2930 goto bad_size;
2931 break;
2932
2933 case O_IP_SRC_LOOKUP:
2934 case O_IP_DST_LOOKUP:
2935 if (cmd->arg1 >= IPFW_TABLES_MAX) {
2936 printf("ipfw: invalid table number %d\n",
2937 cmd->arg1);
2938 return (EINVAL);
2939 }
2940 if (cmdlen != F_INSN_SIZE(ipfw_insn) &&
2941 cmdlen != F_INSN_SIZE(ipfw_insn_u32))
2942 goto bad_size;
2943 break;
2944
2945 case O_MACADDR2:
2946 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
2947 goto bad_size;
2948 break;
2949
2950 case O_NOP:
2951 case O_IPID:
2952 case O_IPTTL:
2953 case O_IPLEN:
2954 if (cmdlen < 1 || cmdlen > 31)
2955 goto bad_size;
2956 break;
2957
2958 case O_MAC_TYPE:
2959 case O_IP_SRCPORT:
2960 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
2961 if (cmdlen < 2 || cmdlen > 31)
2962 goto bad_size;
2963 break;
2964
2965 case O_RECV:
2966 case O_XMIT:
2967 case O_VIA:
2968 if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
2969 goto bad_size;
2970 break;
2971
2972 case O_PIPE:
2973 case O_QUEUE:
2974 if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
2975 goto bad_size;
2976 goto check_action;
2977
2978 case O_FORWARD_IP:
2979#ifdef IPFIREWALL_FORWARD
2980 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
2981 goto bad_size;
2982 goto check_action;
2983#else
2984 return EINVAL;
2985#endif
2986
2987 case O_DIVERT:
2988 case O_TEE:
2989#ifndef IPDIVERT
2990 return EINVAL;
2991#endif
2992 case O_FORWARD_MAC: /* XXX not implemented yet */
2993 case O_CHECK_STATE:
2994 case O_COUNT:
2995 case O_ACCEPT:
2996 case O_DENY:
2997 case O_REJECT:
2998 case O_SKIPTO:
2999 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3000 goto bad_size;
3001check_action:
3002 if (have_action) {
3003 printf("ipfw: opcode %d, multiple actions"
3004 " not allowed\n",
3005 cmd->opcode);
3006 return EINVAL;
3007 }
3008 have_action = 1;
3009 if (l != cmdlen) {
3010 printf("ipfw: opcode %d, action must be"
3011 " last opcode\n",
3012 cmd->opcode);
3013 return EINVAL;
3014 }
3015 break;
3016 default:
3017 printf("ipfw: opcode %d, unknown opcode\n",
3018 cmd->opcode);
3019 return EINVAL;
3020 }
3021 }
3022 if (have_action == 0) {
3023 printf("ipfw: missing action\n");
3024 return EINVAL;
3025 }
3026 return 0;
3027
3028bad_size:
3029 printf("ipfw: opcode %d size %d wrong\n",
3030 cmd->opcode, cmdlen);
3031 return EINVAL;
3032}
3033
3034/*
3035 * Copy the static and dynamic rules to the supplied buffer
3036 * and return the amount of space actually used.
3037 */
3038static size_t
3039ipfw_getrules(struct ip_fw_chain *chain, void *buf, size_t space)
3040{
3041 char *bp = buf;
3042 char *ep = bp + space;
3043 struct ip_fw *rule;
3044 int i;
3045
3046 /* XXX this can take a long time and locking will block packet flow */
3047 IPFW_LOCK(chain);
3048 for (rule = chain->rules; rule ; rule = rule->next) {
3049 /*
3050 * Verify the entry fits in the buffer in case the
3051 * rules changed between calculating buffer space and
3052 * now. This would be better done using a generation
3053 * number but should suffice for now.
3054 */
3055 i = RULESIZE(rule);
3056 if (bp + i <= ep) {
3057 bcopy(rule, bp, i);
3058 bcopy(&set_disable, &(((struct ip_fw *)bp)->next_rule),
3059 sizeof(set_disable));
3060 bp += i;
3061 }
3062 }
3063 IPFW_UNLOCK(chain);
3064 if (ipfw_dyn_v) {
3065 ipfw_dyn_rule *p, *last = NULL;
3066
3067 IPFW_DYN_LOCK();
3068 for (i = 0 ; i < curr_dyn_buckets; i++)
3069 for (p = ipfw_dyn_v[i] ; p != NULL; p = p->next) {
3070 if (bp + sizeof *p <= ep) {
3071 ipfw_dyn_rule *dst =
3072 (ipfw_dyn_rule *)bp;
3073 bcopy(p, dst, sizeof *p);
3074 bcopy(&(p->rule->rulenum), &(dst->rule),
3075 sizeof(p->rule->rulenum));
3076 /*
3077 * store a non-null value in "next".
3078 * The userland code will interpret a
3079 * NULL here as a marker
3080 * for the last dynamic rule.
3081 */
3082 bcopy(&dst, &dst->next, sizeof(dst));
3083 last = dst;
3084 dst->expire =
3085 TIME_LEQ(dst->expire, time_second) ?
3086 0 : dst->expire - time_second ;
3087 bp += sizeof(ipfw_dyn_rule);
3088 }
3089 }
3090 IPFW_DYN_UNLOCK();
3091 if (last != NULL) /* mark last dynamic rule */
3092 bzero(&last->next, sizeof(last));
3093 }
3094 return (bp - (char *)buf);
3095}
3096
3097
3098/**
3099 * {set|get}sockopt parser.
3100 */
3101static int
3102ipfw_ctl(struct sockopt *sopt)
3103{
3104#define RULE_MAXSIZE (256*sizeof(u_int32_t))
3105 int error, rule_num;
3106 size_t size;
3107 struct ip_fw *buf, *rule;
3108 u_int32_t rulenum[2];
3109
3110 error = suser(sopt->sopt_td);
3111 if (error)
3112 return (error);
3113
3114 /*
3115 * Disallow modifications in really-really secure mode, but still allow
3116 * the logging counters to be reset.
3117 */
3118 if (sopt->sopt_name == IP_FW_ADD ||
3119 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
3120#if __FreeBSD_version >= 500034
3121 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
3122 if (error)
3123 return (error);
3124#else /* FreeBSD 4.x */
3125 if (securelevel >= 3)
3126 return (EPERM);
3127#endif
3128 }
3129
3130 error = 0;
3131
3132 switch (sopt->sopt_name) {
3133 case IP_FW_GET:
3134 /*
3135 * pass up a copy of the current rules. Static rules
3136 * come first (the last of which has number IPFW_DEFAULT_RULE),
3137 * followed by a possibly empty list of dynamic rule.
3138 * The last dynamic rule has NULL in the "next" field.
3139 *
3140 * Note that the calculated size is used to bound the
3141 * amount of data returned to the user. The rule set may
3142 * change between calculating the size and returning the
3143 * data in which case we'll just return what fits.
3144 */
3145 size = static_len; /* size of static rules */
3146 if (ipfw_dyn_v) /* add size of dyn.rules */
3147 size += (dyn_count * sizeof(ipfw_dyn_rule));
3148
3149 /*
3150 * XXX todo: if the user passes a short length just to know
3151 * how much room is needed, do not bother filling up the
3152 * buffer, just jump to the sooptcopyout.
3153 */
3154 buf = malloc(size, M_TEMP, M_WAITOK);
3155 error = sooptcopyout(sopt, buf,
3156 ipfw_getrules(&layer3_chain, buf, size));
3157 free(buf, M_TEMP);
3158 break;
3159
3160 case IP_FW_FLUSH:
3161 /*
3162 * Normally we cannot release the lock on each iteration.
3163 * We could do it here only because we start from the head all
3164 * the times so there is no risk of missing some entries.
3165 * On the other hand, the risk is that we end up with
3166 * a very inconsistent ruleset, so better keep the lock
3167 * around the whole cycle.
3168 *
3169 * XXX this code can be improved by resetting the head of
3170 * the list to point to the default rule, and then freeing
3171 * the old list without the need for a lock.
3172 */
3173
3174 IPFW_LOCK(&layer3_chain);
3175 layer3_chain.reap = NULL;
3176 free_chain(&layer3_chain, 0 /* keep default rule */);
3177 rule = layer3_chain.reap, layer3_chain.reap = NULL;
3178 IPFW_UNLOCK(&layer3_chain);
3179 if (layer3_chain.reap != NULL)
3180 reap_rules(rule);
3181 break;
3182
3183 case IP_FW_ADD:
3184 rule = malloc(RULE_MAXSIZE, M_TEMP, M_WAITOK);
3185 error = sooptcopyin(sopt, rule, RULE_MAXSIZE,
3186 sizeof(struct ip_fw) );
3187 if (error == 0)
3188 error = check_ipfw_struct(rule, sopt->sopt_valsize);
3189 if (error == 0) {
3190 error = add_rule(&layer3_chain, rule);
3191 size = RULESIZE(rule);
3192 if (!error && sopt->sopt_dir == SOPT_GET)
3193 error = sooptcopyout(sopt, rule, size);
3194 }
3195 free(rule, M_TEMP);
3196 break;
3197
3198 case IP_FW_DEL:
3199 /*
3200 * IP_FW_DEL is used for deleting single rules or sets,
3201 * and (ab)used to atomically manipulate sets. Argument size
3202 * is used to distinguish between the two:
3203 * sizeof(u_int32_t)
3204 * delete single rule or set of rules,
3205 * or reassign rules (or sets) to a different set.
3206 * 2*sizeof(u_int32_t)
3207 * atomic disable/enable sets.
3208 * first u_int32_t contains sets to be disabled,
3209 * second u_int32_t contains sets to be enabled.
3210 */
3211 error = sooptcopyin(sopt, rulenum,
3212 2*sizeof(u_int32_t), sizeof(u_int32_t));
3213 if (error)
3214 break;
3215 size = sopt->sopt_valsize;
3216 if (size == sizeof(u_int32_t)) /* delete or reassign */
3217 error = del_entry(&layer3_chain, rulenum[0]);
3218 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */
3219 set_disable =
3220 (set_disable | rulenum[0]) & ~rulenum[1] &
3221 ~(1<<RESVD_SET); /* set RESVD_SET always enabled */
3222 else
3223 error = EINVAL;
3224 break;
3225
3226 case IP_FW_ZERO:
3227 case IP_FW_RESETLOG: /* argument is an int, the rule number */
3228 rule_num = 0;
3229 if (sopt->sopt_val != 0) {
3230 error = sooptcopyin(sopt, &rule_num,
3231 sizeof(int), sizeof(int));
3232 if (error)
3233 break;
3234 }
3235 error = zero_entry(&layer3_chain, rule_num,
3236 sopt->sopt_name == IP_FW_RESETLOG);
3237 break;
3238
3239 case IP_FW_TABLE_ADD:
3240 {
3241 ipfw_table_entry ent;
3242
3243 error = sooptcopyin(sopt, &ent,
3244 sizeof(ent), sizeof(ent));
3245 if (error)
3246 break;
3247 error = add_table_entry(ent.tbl, ent.addr,
3248 ent.masklen, ent.value);
3249 }
3250 break;
3251
3252 case IP_FW_TABLE_DEL:
3253 {
3254 ipfw_table_entry ent;
3255
3256 error = sooptcopyin(sopt, &ent,
3257 sizeof(ent), sizeof(ent));
3258 if (error)
3259 break;
3260 error = del_table_entry(ent.tbl, ent.addr, ent.masklen);
3261 }
3262 break;
3263
3264 case IP_FW_TABLE_FLUSH:
3265 {
3266 u_int16_t tbl;
3267
3268 error = sooptcopyin(sopt, &tbl,
3269 sizeof(tbl), sizeof(tbl));
3270 if (error)
3271 break;
3272 error = flush_table(tbl);
3273 }
3274 break;
3275
3276 case IP_FW_TABLE_GETSIZE:
3277 {
3278 u_int32_t tbl, cnt;
3279
3280 if ((error = sooptcopyin(sopt, &tbl, sizeof(tbl),
3281 sizeof(tbl))))
3282 break;
3283 if ((error = count_table(tbl, &cnt)))
3284 break;
3285 error = sooptcopyout(sopt, &cnt, sizeof(cnt));
3286 }
3287 break;
3288
3289 case IP_FW_TABLE_LIST:
3290 {
3291 ipfw_table *tbl;
3292
3293 if (sopt->sopt_valsize < sizeof(*tbl)) {
3294 error = EINVAL;
3295 break;
3296 }
3297 size = sopt->sopt_valsize;
3298 tbl = malloc(size, M_TEMP, M_WAITOK);
3299 if (tbl == NULL) {
3300 error = ENOMEM;
3301 break;
3302 }
3303 error = sooptcopyin(sopt, tbl, size, sizeof(*tbl));
3304 if (error) {
3305 free(tbl, M_TEMP);
3306 break;
3307 }
3308 tbl->size = (size - sizeof(*tbl)) /
3309 sizeof(ipfw_table_entry);
3310 error = dump_table(tbl);
3311 if (error) {
3312 free(tbl, M_TEMP);
3313 break;
3314 }
3315 error = sooptcopyout(sopt, tbl, size);
3316 free(tbl, M_TEMP);
3317 }
3318 break;
3319
3320 default:
3321 printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name);
3322 error = EINVAL;
3323 }
3324
3325 return (error);
3326#undef RULE_MAXSIZE
3327}
3328
3329/**
3330 * dummynet needs a reference to the default rule, because rules can be
3331 * deleted while packets hold a reference to them. When this happens,
3332 * dummynet changes the reference to the default rule (it could well be a
3333 * NULL pointer, but this way we do not need to check for the special
3334 * case, plus here he have info on the default behaviour).
3335 */
3336struct ip_fw *ip_fw_default_rule;
3337
3338/*
3339 * This procedure is only used to handle keepalives. It is invoked
3340 * every dyn_keepalive_period
3341 */
3342static void
3343ipfw_tick(void * __unused unused)
3344{
3345 int i;
3346 ipfw_dyn_rule *q;
3347
3348 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
3349 goto done;
3350
3351 IPFW_DYN_LOCK();
3352 for (i = 0 ; i < curr_dyn_buckets ; i++) {
3353 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
3354 if (q->dyn_type == O_LIMIT_PARENT)
3355 continue;
3356 if (q->id.proto != IPPROTO_TCP)
3357 continue;
3358 if ( (q->state & BOTH_SYN) != BOTH_SYN)
3359 continue;
3360 if (TIME_LEQ( time_second+dyn_keepalive_interval,
3361 q->expire))
3362 continue; /* too early */
3363 if (TIME_LEQ(q->expire, time_second))
3364 continue; /* too late, rule expired */
3365
3366 send_pkt(&(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN);
3367 send_pkt(&(q->id), q->ack_fwd - 1, q->ack_rev, 0);
3368 }
3369 }
3370 IPFW_DYN_UNLOCK();
3371done:
3372 callout_reset(&ipfw_timeout, dyn_keepalive_period*hz, ipfw_tick, NULL);
3373}
3374
3375int
3376ipfw_init(void)
3377{
3378 struct ip_fw default_rule;
3379 int error;
3380
3381 layer3_chain.rules = NULL;
3382 IPFW_LOCK_INIT(&layer3_chain);
3383 IPFW_DYN_LOCK_INIT();
3384 callout_init(&ipfw_timeout, debug_mpsafenet ? CALLOUT_MPSAFE : 0);
3385
3386 bzero(&default_rule, sizeof default_rule);
3387
3388 default_rule.act_ofs = 0;
3389 default_rule.rulenum = IPFW_DEFAULT_RULE;
3390 default_rule.cmd_len = 1;
3391 default_rule.set = RESVD_SET;
3392
3393 default_rule.cmd[0].len = 1;
3394 default_rule.cmd[0].opcode =
3395#ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
3396 1 ? O_ACCEPT :
3397#endif
3398 O_DENY;
3399
3400 error = add_rule(&layer3_chain, &default_rule);
3401 if (error != 0) {
3402 printf("ipfw2: error %u initializing default rule "
3403 "(support disabled)\n", error);
3404 IPFW_DYN_LOCK_DESTROY();
3405 IPFW_LOCK_DESTROY(&layer3_chain);
3406 return (error);
3407 }
3408
3409 ip_fw_default_rule = layer3_chain.rules;
3410 printf("ipfw2 initialized, divert %s, "
3411 "rule-based forwarding "
3412#ifdef IPFIREWALL_FORWARD
3413 "enabled, "
3414#else
3415 "disabled, "
3416#endif
3417 "default to %s, logging ",
3418#ifdef IPDIVERT
3419 "enabled",
3420#else
3421 "disabled",
3422#endif
3423 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny");
3424
3425#ifdef IPFIREWALL_VERBOSE
3426 fw_verbose = 1;
3427#endif
3428#ifdef IPFIREWALL_VERBOSE_LIMIT
3429 verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
3430#endif
3431 if (fw_verbose == 0)
3432 printf("disabled\n");
3433 else if (verbose_limit == 0)
3434 printf("unlimited\n");
3435 else
3436 printf("limited to %d packets/entry by default\n",
3437 verbose_limit);
3438
3439 init_tables();
3440 ip_fw_ctl_ptr = ipfw_ctl;
3441 ip_fw_chk_ptr = ipfw_chk;
3442 callout_reset(&ipfw_timeout, hz, ipfw_tick, NULL);
3443
3444 return (0);
3445}
3446
3447void
3448ipfw_destroy(void)
3449{
3450 struct ip_fw *reap;
3451
3452 ip_fw_chk_ptr = NULL;
3453 ip_fw_ctl_ptr = NULL;
3454 callout_drain(&ipfw_timeout);
3455 IPFW_LOCK(&layer3_chain);
3456 layer3_chain.reap = NULL;
3457 free_chain(&layer3_chain, 1 /* kill default rule */);
3458 reap = layer3_chain.reap, layer3_chain.reap = NULL;
3459 IPFW_UNLOCK(&layer3_chain);
3460 if (reap != NULL)
3461 reap_rules(reap);
3462 flush_tables();
3463 IPFW_DYN_LOCK_DESTROY();
3464 IPFW_LOCK_DESTROY(&layer3_chain);
3465 printf("IP firewall unloaded\n");
3466}
3467
3468#endif /* IPFW2 */
|