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_ip6fw.h"
38#include "opt_ipdn.h"
39#include "opt_inet.h"
40#ifndef INET
41#error IPFIREWALL requires INET.
42#endif /* INET */
43#endif
44#include "opt_inet6.h"
45#include "opt_ipsec.h"
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/condvar.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#include <netgraph/ng_ipfw.h>
82
83#include <altq/if_altq.h>
84
85#ifdef IPSEC
86#include <netinet6/ipsec.h>
87#endif
88
89#include <netinet/ip6.h>
90#include <netinet/icmp6.h>
91#ifdef INET6
92#include <netinet6/scope6_var.h>
93#endif
94
95#include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
96
97#include <machine/in_cksum.h> /* XXX for in_cksum */
98
99/*
100 * set_disable contains one bit per set value (0..31).
101 * If the bit is set, all rules with the corresponding set
102 * are disabled. Set RESVD_SET(31) is reserved for the default rule
103 * and rules that are not deleted by the flush command,
104 * and CANNOT be disabled.
105 * Rules in set RESVD_SET can only be deleted explicitly.
106 */
107static u_int32_t set_disable;
108
109static int fw_verbose;
110static int verbose_limit;
111
112static struct callout ipfw_timeout;
113static uma_zone_t ipfw_dyn_rule_zone;
114#define IPFW_DEFAULT_RULE 65535
115
116/*
117 * Data structure to cache our ucred related
118 * information. This structure only gets used if
119 * the user specified UID/GID based constraints in
120 * a firewall rule.
121 */
122struct ip_fw_ugid {
123 gid_t fw_groups[NGROUPS];
124 int fw_ngroups;
125 uid_t fw_uid;
126 int fw_prid;
127};
128
129#define IPFW_TABLES_MAX 128
130struct ip_fw_chain {
131 struct ip_fw *rules; /* list of rules */
132 struct ip_fw *reap; /* list of rules to reap */
133 struct radix_node_head *tables[IPFW_TABLES_MAX];
134 struct mtx mtx; /* lock guarding rule list */
135 int busy_count; /* busy count for rw locks */
136 int want_write;
137 struct cv cv;
138};
139#define IPFW_LOCK_INIT(_chain) \
140 mtx_init(&(_chain)->mtx, "IPFW static rules", NULL, \
141 MTX_DEF | MTX_RECURSE)
142#define IPFW_LOCK_DESTROY(_chain) mtx_destroy(&(_chain)->mtx)
143#define IPFW_WLOCK_ASSERT(_chain) do { \
144 mtx_assert(&(_chain)->mtx, MA_OWNED); \
145 NET_ASSERT_GIANT(); \
146} while (0)
147
148static __inline void
149IPFW_RLOCK(struct ip_fw_chain *chain)
150{
151 mtx_lock(&chain->mtx);
152 chain->busy_count++;
153 mtx_unlock(&chain->mtx);
154}
155
156static __inline void
157IPFW_RUNLOCK(struct ip_fw_chain *chain)
158{
159 mtx_lock(&chain->mtx);
160 chain->busy_count--;
161 if (chain->busy_count == 0 && chain->want_write)
162 cv_signal(&chain->cv);
163 mtx_unlock(&chain->mtx);
164}
165
166static __inline void
167IPFW_WLOCK(struct ip_fw_chain *chain)
168{
169 mtx_lock(&chain->mtx);
170 chain->want_write++;
171 while (chain->busy_count > 0)
172 cv_wait(&chain->cv, &chain->mtx);
173}
174
175static __inline void
176IPFW_WUNLOCK(struct ip_fw_chain *chain)
177{
178 chain->want_write--;
179 cv_signal(&chain->cv);
180 mtx_unlock(&chain->mtx);
181}
182
183/*
184 * list of rules for layer 3
185 */
186static struct ip_fw_chain layer3_chain;
187
188MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
189MALLOC_DEFINE(M_IPFW_TBL, "ipfw_tbl", "IpFw tables");
190
191struct table_entry {
192 struct radix_node rn[2];
193 struct sockaddr_in addr, mask;
194 u_int32_t value;
195};
196
197static int fw_debug = 1;
198static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
199
200#ifdef SYSCTL_NODE
201SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
202SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable,
203 CTLFLAG_RW | CTLFLAG_SECURE3,
204 &fw_enable, 0, "Enable ipfw");
205SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
206 &autoinc_step, 0, "Rule number autincrement step");
207SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, one_pass,
208 CTLFLAG_RW | CTLFLAG_SECURE3,
209 &fw_one_pass, 0,
210 "Only do a single pass through ipfw when using dummynet(4)");
211SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
212 &fw_debug, 0, "Enable printing of debug ip_fw statements");
213SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose,
214 CTLFLAG_RW | CTLFLAG_SECURE3,
215 &fw_verbose, 0, "Log matches to ipfw rules");
216SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
217 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
218
219/*
220 * Description of dynamic rules.
221 *
222 * Dynamic rules are stored in lists accessed through a hash table
223 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
224 * be modified through the sysctl variable dyn_buckets which is
225 * updated when the table becomes empty.
226 *
227 * XXX currently there is only one list, ipfw_dyn.
228 *
229 * When a packet is received, its address fields are first masked
230 * with the mask defined for the rule, then hashed, then matched
231 * against the entries in the corresponding list.
232 * Dynamic rules can be used for different purposes:
233 * + stateful rules;
234 * + enforcing limits on the number of sessions;
235 * + in-kernel NAT (not implemented yet)
236 *
237 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
238 * measured in seconds and depending on the flags.
239 *
240 * The total number of dynamic rules is stored in dyn_count.
241 * The max number of dynamic rules is dyn_max. When we reach
242 * the maximum number of rules we do not create anymore. This is
243 * done to avoid consuming too much memory, but also too much
244 * time when searching on each packet (ideally, we should try instead
245 * to put a limit on the length of the list on each bucket...).
246 *
247 * Each dynamic rule holds a pointer to the parent ipfw rule so
248 * we know what action to perform. Dynamic rules are removed when
249 * the parent rule is deleted. XXX we should make them survive.
250 *
251 * There are some limitations with dynamic rules -- we do not
252 * obey the 'randomized match', and we do not do multiple
253 * passes through the firewall. XXX check the latter!!!
254 */
255static ipfw_dyn_rule **ipfw_dyn_v = NULL;
256static u_int32_t dyn_buckets = 256; /* must be power of 2 */
257static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */
258
259static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */
260#define IPFW_DYN_LOCK_INIT() \
261 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF)
262#define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx)
263#define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx)
264#define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx)
265#define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED)
266
267/*
268 * Timeouts for various events in handing dynamic rules.
269 */
270static u_int32_t dyn_ack_lifetime = 300;
271static u_int32_t dyn_syn_lifetime = 20;
272static u_int32_t dyn_fin_lifetime = 1;
273static u_int32_t dyn_rst_lifetime = 1;
274static u_int32_t dyn_udp_lifetime = 10;
275static u_int32_t dyn_short_lifetime = 5;
276
277/*
278 * Keepalives are sent if dyn_keepalive is set. They are sent every
279 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
280 * seconds of lifetime of a rule.
281 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
282 * than dyn_keepalive_period.
283 */
284
285static u_int32_t dyn_keepalive_interval = 20;
286static u_int32_t dyn_keepalive_period = 5;
287static u_int32_t dyn_keepalive = 1; /* do send keepalives */
288
289static u_int32_t static_count; /* # of static rules */
290static u_int32_t static_len; /* size in bytes of static rules */
291static u_int32_t dyn_count; /* # of dynamic rules */
292static u_int32_t dyn_max = 4096; /* max # of dynamic rules */
293
294SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
295 &dyn_buckets, 0, "Number of dyn. buckets");
296SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
297 &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
298SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
299 &dyn_count, 0, "Number of dyn. rules");
300SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
301 &dyn_max, 0, "Max number of dyn. rules");
302SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
303 &static_count, 0, "Number of static rules");
304SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
305 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
306SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
307 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
308SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
309 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
310SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
311 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
312SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
313 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
314SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
315 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
316SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
317 &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
318
319#ifdef INET6
320/*
321 * IPv6 specific variables
322 */
323SYSCTL_DECL(_net_inet6_ip6);
324
325static struct sysctl_ctx_list ip6_fw_sysctl_ctx;
326static struct sysctl_oid *ip6_fw_sysctl_tree;
327#endif /* INET6 */
328#endif /* SYSCTL_NODE */
329
330static int fw_deny_unknown_exthdrs = 1;
331
332
333/*
334 * L3HDR maps an ipv4 pointer into a layer3 header pointer of type T
335 * Other macros just cast void * into the appropriate type
336 */
337#define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
338#define TCP(p) ((struct tcphdr *)(p))
339#define UDP(p) ((struct udphdr *)(p))
340#define ICMP(p) ((struct icmphdr *)(p))
341#define ICMP6(p) ((struct icmp6_hdr *)(p))
342
343static __inline int
344icmptype_match(struct icmphdr *icmp, ipfw_insn_u32 *cmd)
345{
346 int type = icmp->icmp_type;
347
348 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
349}
350
351#define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
352 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
353
354static int
355is_icmp_query(struct icmphdr *icmp)
356{
357 int type = icmp->icmp_type;
358
359 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
360}
361#undef TT
362
363/*
364 * The following checks use two arrays of 8 or 16 bits to store the
365 * bits that we want set or clear, respectively. They are in the
366 * low and high half of cmd->arg1 or cmd->d[0].
367 *
368 * We scan options and store the bits we find set. We succeed if
369 *
370 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
371 *
372 * The code is sometimes optimized not to store additional variables.
373 */
374
375static int
376flags_match(ipfw_insn *cmd, u_int8_t bits)
377{
378 u_char want_clear;
379 bits = ~bits;
380
381 if ( ((cmd->arg1 & 0xff) & bits) != 0)
382 return 0; /* some bits we want set were clear */
383 want_clear = (cmd->arg1 >> 8) & 0xff;
384 if ( (want_clear & bits) != want_clear)
385 return 0; /* some bits we want clear were set */
386 return 1;
387}
388
389static int
390ipopts_match(struct ip *ip, ipfw_insn *cmd)
391{
392 int optlen, bits = 0;
393 u_char *cp = (u_char *)(ip + 1);
394 int x = (ip->ip_hl << 2) - sizeof (struct ip);
395
396 for (; x > 0; x -= optlen, cp += optlen) {
397 int opt = cp[IPOPT_OPTVAL];
398
399 if (opt == IPOPT_EOL)
400 break;
401 if (opt == IPOPT_NOP)
402 optlen = 1;
403 else {
404 optlen = cp[IPOPT_OLEN];
405 if (optlen <= 0 || optlen > x)
406 return 0; /* invalid or truncated */
407 }
408 switch (opt) {
409
410 default:
411 break;
412
413 case IPOPT_LSRR:
414 bits |= IP_FW_IPOPT_LSRR;
415 break;
416
417 case IPOPT_SSRR:
418 bits |= IP_FW_IPOPT_SSRR;
419 break;
420
421 case IPOPT_RR:
422 bits |= IP_FW_IPOPT_RR;
423 break;
424
425 case IPOPT_TS:
426 bits |= IP_FW_IPOPT_TS;
427 break;
428 }
429 }
430 return (flags_match(cmd, bits));
431}
432
433static int
434tcpopts_match(struct tcphdr *tcp, ipfw_insn *cmd)
435{
436 int optlen, bits = 0;
437 u_char *cp = (u_char *)(tcp + 1);
438 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
439
440 for (; x > 0; x -= optlen, cp += optlen) {
441 int opt = cp[0];
442 if (opt == TCPOPT_EOL)
443 break;
444 if (opt == TCPOPT_NOP)
445 optlen = 1;
446 else {
447 optlen = cp[1];
448 if (optlen <= 0)
449 break;
450 }
451
452 switch (opt) {
453
454 default:
455 break;
456
457 case TCPOPT_MAXSEG:
458 bits |= IP_FW_TCPOPT_MSS;
459 break;
460
461 case TCPOPT_WINDOW:
462 bits |= IP_FW_TCPOPT_WINDOW;
463 break;
464
465 case TCPOPT_SACK_PERMITTED:
466 case TCPOPT_SACK:
467 bits |= IP_FW_TCPOPT_SACK;
468 break;
469
470 case TCPOPT_TIMESTAMP:
471 bits |= IP_FW_TCPOPT_TS;
472 break;
473
474 }
475 }
476 return (flags_match(cmd, bits));
477}
478
479static int
480iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
481{
482 if (ifp == NULL) /* no iface with this packet, match fails */
483 return 0;
484 /* Check by name or by IP address */
485 if (cmd->name[0] != '\0') { /* match by name */
486 /* Check name */
487 if (cmd->p.glob) {
488 if (fnmatch(cmd->name, ifp->if_xname, 0) == 0)
489 return(1);
490 } else {
491 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
492 return(1);
493 }
494 } else {
495 struct ifaddr *ia;
496
497 /* XXX lock? */
498 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
499 if (ia->ifa_addr == NULL)
500 continue;
501 if (ia->ifa_addr->sa_family != AF_INET)
502 continue;
503 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
504 (ia->ifa_addr))->sin_addr.s_addr)
505 return(1); /* match */
506 }
507 }
508 return(0); /* no match, fail ... */
509}
510
511/*
512 * The verify_path function checks if a route to the src exists and
513 * if it is reachable via ifp (when provided).
514 *
515 * The 'verrevpath' option checks that the interface that an IP packet
516 * arrives on is the same interface that traffic destined for the
517 * packet's source address would be routed out of. The 'versrcreach'
518 * option just checks that the source address is reachable via any route
519 * (except default) in the routing table. These two are a measure to block
520 * forged packets. This is also commonly known as "anti-spoofing" or Unicast
521 * Reverse Path Forwarding (Unicast RFP) in Cisco-ese. The name of the knobs
522 * is purposely reminiscent of the Cisco IOS command,
523 *
524 * ip verify unicast reverse-path
525 * ip verify unicast source reachable-via any
526 *
527 * which implements the same functionality. But note that syntax is
528 * misleading. The check may be performed on all IP packets whether unicast,
529 * multicast, or broadcast.
530 */
531static int
532verify_path(struct in_addr src, struct ifnet *ifp)
533{
534 struct route ro;
535 struct sockaddr_in *dst;
536
537 bzero(&ro, sizeof(ro));
538
539 dst = (struct sockaddr_in *)&(ro.ro_dst);
540 dst->sin_family = AF_INET;
541 dst->sin_len = sizeof(*dst);
542 dst->sin_addr = src;
543 rtalloc_ign(&ro, RTF_CLONING);
544
545 if (ro.ro_rt == NULL)
546 return 0;
547
548 /* if ifp is provided, check for equality with rtentry */
549 if (ifp != NULL && ro.ro_rt->rt_ifp != ifp) {
550 RTFREE(ro.ro_rt);
551 return 0;
552 }
553
554 /* if no ifp provided, check if rtentry is not default route */
555 if (ifp == NULL &&
556 satosin(rt_key(ro.ro_rt))->sin_addr.s_addr == INADDR_ANY) {
557 RTFREE(ro.ro_rt);
558 return 0;
559 }
560
561 /* or if this is a blackhole/reject route */
562 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
563 RTFREE(ro.ro_rt);
564 return 0;
565 }
566
567 /* found valid route */
568 RTFREE(ro.ro_rt);
569 return 1;
570}
571
572#ifdef INET6
573/*
574 * ipv6 specific rules here...
575 */
576static __inline int
577icmp6type_match (int type, ipfw_insn_u32 *cmd)
578{
579 return (type <= ICMP6_MAXTYPE && (cmd->d[type/32] & (1<<(type%32)) ) );
580}
581
582static int
583flow6id_match( int curr_flow, ipfw_insn_u32 *cmd )
584{
585 int i;
586 for (i=0; i <= cmd->o.arg1; ++i )
587 if (curr_flow == cmd->d[i] )
588 return 1;
589 return 0;
590}
591
592/* support for IP6_*_ME opcodes */
593static int
594search_ip6_addr_net (struct in6_addr * ip6_addr)
595{
596 struct ifnet *mdc;
597 struct ifaddr *mdc2;
598 struct in6_ifaddr *fdm;
599 struct in6_addr copia;
600
601 TAILQ_FOREACH(mdc, &ifnet, if_link)
602 for (mdc2 = mdc->if_addrlist.tqh_first; mdc2;
603 mdc2 = mdc2->ifa_list.tqe_next) {
604 if (!mdc2->ifa_addr)
605 continue;
606 if (mdc2->ifa_addr->sa_family == AF_INET6) {
607 fdm = (struct in6_ifaddr *)mdc2;
608 copia = fdm->ia_addr.sin6_addr;
609 /* need for leaving scope_id in the sock_addr */
610 in6_clearscope(&copia);
611 if (IN6_ARE_ADDR_EQUAL(ip6_addr, &copia))
612 return 1;
613 }
614 }
615 return 0;
616}
617
618static int
619verify_path6(struct in6_addr *src, struct ifnet *ifp)
620{
621 struct route_in6 ro;
622 struct sockaddr_in6 *dst;
623
624 bzero(&ro, sizeof(ro));
625
626 dst = (struct sockaddr_in6 * )&(ro.ro_dst);
627 dst->sin6_family = AF_INET6;
628 dst->sin6_len = sizeof(*dst);
629 dst->sin6_addr = *src;
630 rtalloc_ign((struct route *)&ro, RTF_CLONING);
631
632 if (ro.ro_rt == NULL)
633 return 0;
634
635 /*
636 * if ifp is provided, check for equality with rtentry
637 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
638 * to support the case of sending packets to an address of our own.
639 * (where the former interface is the first argument of if_simloop()
640 * (=ifp), the latter is lo0)
641 */
642 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
643 RTFREE(ro.ro_rt);
644 return 0;
645 }
646
647 /* if no ifp provided, check if rtentry is not default route */
648 if (ifp == NULL &&
649 IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(ro.ro_rt))->sin6_addr)) {
650 RTFREE(ro.ro_rt);
651 return 0;
652 }
653
654 /* or if this is a blackhole/reject route */
655 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
656 RTFREE(ro.ro_rt);
657 return 0;
658 }
659
660 /* found valid route */
661 RTFREE(ro.ro_rt);
662 return 1;
663
664}
665static __inline int
666hash_packet6(struct ipfw_flow_id *id)
667{
668 u_int32_t i;
669 i = (id->dst_ip6.__u6_addr.__u6_addr32[0]) ^
670 (id->dst_ip6.__u6_addr.__u6_addr32[1]) ^
671 (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^
672 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^
673 (id->dst_port) ^ (id->src_port) ^ (id->flow_id6);
674 return i;
675}
676
677static int
678is_icmp6_query(int icmp6_type)
679{
680 if ((icmp6_type <= ICMP6_MAXTYPE) &&
681 (icmp6_type == ICMP6_ECHO_REQUEST ||
682 icmp6_type == ICMP6_MEMBERSHIP_QUERY ||
683 icmp6_type == ICMP6_WRUREQUEST ||
684 icmp6_type == ICMP6_FQDN_QUERY ||
685 icmp6_type == ICMP6_NI_QUERY))
686 return (1);
687
688 return (0);
689}
690
691static void
692send_reject6(struct ip_fw_args *args, int code, u_short offset, u_int hlen)
693{
694 if (code == ICMP6_UNREACH_RST && offset == 0 &&
695 args->f_id.proto == IPPROTO_TCP) {
696 struct ip6_hdr *ip6;
697 struct tcphdr *tcp;
698 tcp_seq ack, seq;
699 int flags;
700 struct {
701 struct ip6_hdr ip6;
702 struct tcphdr th;
703 } ti;
704
705 if (args->m->m_len < (hlen+sizeof(struct tcphdr))) {
706 args->m = m_pullup(args->m, hlen+sizeof(struct tcphdr));
707 if (args->m == NULL)
708 return;
709 }
710
711 ip6 = mtod(args->m, struct ip6_hdr *);
712 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
713
714 if ((tcp->th_flags & TH_RST) != 0) {
715 m_freem(args->m);
716 return;
717 }
718
719 ti.ip6 = *ip6;
720 ti.th = *tcp;
721 ti.th.th_seq = ntohl(ti.th.th_seq);
722 ti.th.th_ack = ntohl(ti.th.th_ack);
723 ti.ip6.ip6_nxt = IPPROTO_TCP;
724
725 if (ti.th.th_flags & TH_ACK) {
726 ack = 0;
727 seq = ti.th.th_ack;
728 flags = TH_RST;
729 } else {
730 ack = ti.th.th_seq;
731 if (((args->m)->m_flags & M_PKTHDR) != 0) {
732 ack += (args->m)->m_pkthdr.len - hlen
733 - (ti.th.th_off << 2);
734 } else if (ip6->ip6_plen) {
735 ack += ntohs(ip6->ip6_plen) + sizeof(*ip6)
736 - hlen - (ti.th.th_off << 2);
737 } else {
738 m_freem(args->m);
739 return;
740 }
741 if (tcp->th_flags & TH_SYN)
742 ack++;
743 seq = 0;
744 flags = TH_RST|TH_ACK;
745 }
746 bcopy(&ti, ip6, sizeof(ti));
747 tcp_respond(NULL, ip6, (struct tcphdr *)(ip6 + 1),
748 args->m, ack, seq, flags);
749
750 } else if (code != ICMP6_UNREACH_RST) { /* Send an ICMPv6 unreach. */
751 icmp6_error(args->m, ICMP6_DST_UNREACH, code, 0);
752
753 } else
754 m_freem(args->m);
755
756 args->m = NULL;
757}
758
759#endif /* INET6 */
760
761static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */
762
763#define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
764#define SNP(buf) buf, sizeof(buf)
765
766/*
767 * We enter here when we have a rule with O_LOG.
768 * XXX this function alone takes about 2Kbytes of code!
769 */
770static void
771ipfw_log(struct ip_fw *f, u_int hlen, struct ip_fw_args *args,
772 struct mbuf *m, struct ifnet *oif, u_short offset)
773{
774 struct ether_header *eh = args->eh;
775 char *action;
776 int limit_reached = 0;
777 char action2[40], proto[128], fragment[32];
778
779 fragment[0] = '\0';
780 proto[0] = '\0';
781
782 if (f == NULL) { /* bogus pkt */
783 if (verbose_limit != 0 && norule_counter >= verbose_limit)
784 return;
785 norule_counter++;
786 if (norule_counter == verbose_limit)
787 limit_reached = verbose_limit;
788 action = "Refuse";
789 } else { /* O_LOG is the first action, find the real one */
790 ipfw_insn *cmd = ACTION_PTR(f);
791 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
792
793 if (l->max_log != 0 && l->log_left == 0)
794 return;
795 l->log_left--;
796 if (l->log_left == 0)
797 limit_reached = l->max_log;
798 cmd += F_LEN(cmd); /* point to first action */
799 if (cmd->opcode == O_ALTQ) {
800 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
801
802 snprintf(SNPARGS(action2, 0), "Altq %d",
803 altq->qid);
804 cmd += F_LEN(cmd);
805 }
806 if (cmd->opcode == O_PROB)
807 cmd += F_LEN(cmd);
808
809 action = action2;
810 switch (cmd->opcode) {
811 case O_DENY:
812 action = "Deny";
813 break;
814
815 case O_REJECT:
816 if (cmd->arg1==ICMP_REJECT_RST)
817 action = "Reset";
818 else if (cmd->arg1==ICMP_UNREACH_HOST)
819 action = "Reject";
820 else
821 snprintf(SNPARGS(action2, 0), "Unreach %d",
822 cmd->arg1);
823 break;
824
825 case O_UNREACH6:
826 if (cmd->arg1==ICMP6_UNREACH_RST)
827 action = "Reset";
828 else
829 snprintf(SNPARGS(action2, 0), "Unreach %d",
830 cmd->arg1);
831 break;
832
833 case O_ACCEPT:
834 action = "Accept";
835 break;
836 case O_COUNT:
837 action = "Count";
838 break;
839 case O_DIVERT:
840 snprintf(SNPARGS(action2, 0), "Divert %d",
841 cmd->arg1);
842 break;
843 case O_TEE:
844 snprintf(SNPARGS(action2, 0), "Tee %d",
845 cmd->arg1);
846 break;
847 case O_SKIPTO:
848 snprintf(SNPARGS(action2, 0), "SkipTo %d",
849 cmd->arg1);
850 break;
851 case O_PIPE:
852 snprintf(SNPARGS(action2, 0), "Pipe %d",
853 cmd->arg1);
854 break;
855 case O_QUEUE:
856 snprintf(SNPARGS(action2, 0), "Queue %d",
857 cmd->arg1);
858 break;
859 case O_FORWARD_IP: {
860 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
861 int len;
862
863 len = snprintf(SNPARGS(action2, 0), "Forward to %s",
864 inet_ntoa(sa->sa.sin_addr));
865 if (sa->sa.sin_port)
866 snprintf(SNPARGS(action2, len), ":%d",
867 sa->sa.sin_port);
868 }
869 break;
870 case O_NETGRAPH:
871 snprintf(SNPARGS(action2, 0), "Netgraph %d",
872 cmd->arg1);
873 break;
874 case O_NGTEE:
875 snprintf(SNPARGS(action2, 0), "Ngtee %d",
876 cmd->arg1);
877 break;
878 default:
879 action = "UNKNOWN";
880 break;
881 }
882 }
883
884 if (hlen == 0) { /* non-ip */
885 snprintf(SNPARGS(proto, 0), "MAC");
886
887 } else {
888 int len;
889 char src[48], dst[48];
890 struct icmphdr *icmp;
891 struct tcphdr *tcp;
892 struct udphdr *udp;
893 /* Initialize to make compiler happy. */
894 struct ip *ip = NULL;
895#ifdef INET6
896 struct ip6_hdr *ip6 = NULL;
897 struct icmp6_hdr *icmp6;
898#endif
899 src[0] = '\0';
900 dst[0] = '\0';
901#ifdef INET6
902 if (args->f_id.addr_type == 6) {
903 snprintf(src, sizeof(src), "[%s]",
904 ip6_sprintf(&args->f_id.src_ip6));
905 snprintf(dst, sizeof(dst), "[%s]",
906 ip6_sprintf(&args->f_id.dst_ip6));
907
908 ip6 = (struct ip6_hdr *)mtod(m, struct ip6_hdr *);
909 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
910 udp = (struct udphdr *)(mtod(args->m, char *) + hlen);
911 } else
912#endif
913 {
914 ip = mtod(m, struct ip *);
915 tcp = L3HDR(struct tcphdr, ip);
916 udp = L3HDR(struct udphdr, ip);
917
918 inet_ntoa_r(ip->ip_src, src);
919 inet_ntoa_r(ip->ip_dst, dst);
920 }
921
922 switch (args->f_id.proto) {
923 case IPPROTO_TCP:
924 len = snprintf(SNPARGS(proto, 0), "TCP %s", src);
925 if (offset == 0)
926 snprintf(SNPARGS(proto, len), ":%d %s:%d",
927 ntohs(tcp->th_sport),
928 dst,
929 ntohs(tcp->th_dport));
930 else
931 snprintf(SNPARGS(proto, len), " %s", dst);
932 break;
933
934 case IPPROTO_UDP:
935 len = snprintf(SNPARGS(proto, 0), "UDP %s", src);
936 if (offset == 0)
937 snprintf(SNPARGS(proto, len), ":%d %s:%d",
938 ntohs(udp->uh_sport),
939 dst,
940 ntohs(udp->uh_dport));
941 else
942 snprintf(SNPARGS(proto, len), " %s", dst);
943 break;
944
945 case IPPROTO_ICMP:
946 icmp = L3HDR(struct icmphdr, ip);
947 if (offset == 0)
948 len = snprintf(SNPARGS(proto, 0),
949 "ICMP:%u.%u ",
950 icmp->icmp_type, icmp->icmp_code);
951 else
952 len = snprintf(SNPARGS(proto, 0), "ICMP ");
953 len += snprintf(SNPARGS(proto, len), "%s", src);
954 snprintf(SNPARGS(proto, len), " %s", dst);
955 break;
956#ifdef INET6
957 case IPPROTO_ICMPV6:
958 icmp6 = (struct icmp6_hdr *)(mtod(args->m, char *) + hlen);
959 if (offset == 0)
960 len = snprintf(SNPARGS(proto, 0),
961 "ICMPv6:%u.%u ",
962 icmp6->icmp6_type, icmp6->icmp6_code);
963 else
964 len = snprintf(SNPARGS(proto, 0), "ICMPv6 ");
965 len += snprintf(SNPARGS(proto, len), "%s", src);
966 snprintf(SNPARGS(proto, len), " %s", dst);
967 break;
968#endif
969 default:
970 len = snprintf(SNPARGS(proto, 0), "P:%d %s",
971 args->f_id.proto, src);
972 snprintf(SNPARGS(proto, len), " %s", dst);
973 break;
974 }
975
976#ifdef INET6
977 if (args->f_id.addr_type == 6) {
978 if (offset & (IP6F_OFF_MASK | IP6F_MORE_FRAG))
979 snprintf(SNPARGS(fragment, 0),
980 " (frag %08x:%d@%d%s)",
981 args->f_id.frag_id6,
982 ntohs(ip6->ip6_plen) - hlen,
983 ntohs(offset & IP6F_OFF_MASK) << 3,
984 (offset & IP6F_MORE_FRAG) ? "+" : "");
985 } else
986#endif
987 {
988 int ip_off, ip_len;
989 if (eh != NULL) { /* layer 2 packets are as on the wire */
990 ip_off = ntohs(ip->ip_off);
991 ip_len = ntohs(ip->ip_len);
992 } else {
993 ip_off = ip->ip_off;
994 ip_len = ip->ip_len;
995 }
996 if (ip_off & (IP_MF | IP_OFFMASK))
997 snprintf(SNPARGS(fragment, 0),
998 " (frag %d:%d@%d%s)",
999 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
1000 offset << 3,
1001 (ip_off & IP_MF) ? "+" : "");
1002 }
1003 }
1004 if (oif || m->m_pkthdr.rcvif)
1005 log(LOG_SECURITY | LOG_INFO,
1006 "ipfw: %d %s %s %s via %s%s\n",
1007 f ? f->rulenum : -1,
1008 action, proto, oif ? "out" : "in",
1009 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
1010 fragment);
1011 else
1012 log(LOG_SECURITY | LOG_INFO,
1013 "ipfw: %d %s %s [no if info]%s\n",
1014 f ? f->rulenum : -1,
1015 action, proto, fragment);
1016 if (limit_reached)
1017 log(LOG_SECURITY | LOG_NOTICE,
1018 "ipfw: limit %d reached on entry %d\n",
1019 limit_reached, f ? f->rulenum : -1);
1020}
1021
1022/*
1023 * IMPORTANT: the hash function for dynamic rules must be commutative
1024 * in source and destination (ip,port), because rules are bidirectional
1025 * and we want to find both in the same bucket.
1026 */
1027static __inline int
1028hash_packet(struct ipfw_flow_id *id)
1029{
1030 u_int32_t i;
1031
1032#ifdef INET6
1033 if (IS_IP6_FLOW_ID(id))
1034 i = hash_packet6(id);
1035 else
1036#endif /* INET6 */
1037 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
1038 i &= (curr_dyn_buckets - 1);
1039 return i;
1040}
1041
1042/**
1043 * unlink a dynamic rule from a chain. prev is a pointer to
1044 * the previous one, q is a pointer to the rule to delete,
1045 * head is a pointer to the head of the queue.
1046 * Modifies q and potentially also head.
1047 */
1048#define UNLINK_DYN_RULE(prev, head, q) { \
1049 ipfw_dyn_rule *old_q = q; \
1050 \
1051 /* remove a refcount to the parent */ \
1052 if (q->dyn_type == O_LIMIT) \
1053 q->parent->count--; \
1054 DEB(printf("ipfw: unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",\
1055 (q->id.src_ip), (q->id.src_port), \
1056 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \
1057 if (prev != NULL) \
1058 prev->next = q = q->next; \
1059 else \
1060 head = q = q->next; \
1061 dyn_count--; \
1062 uma_zfree(ipfw_dyn_rule_zone, old_q); }
1063
1064#define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
1065
1066/**
1067 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
1068 *
1069 * If keep_me == NULL, rules are deleted even if not expired,
1070 * otherwise only expired rules are removed.
1071 *
1072 * The value of the second parameter is also used to point to identify
1073 * a rule we absolutely do not want to remove (e.g. because we are
1074 * holding a reference to it -- this is the case with O_LIMIT_PARENT
1075 * rules). The pointer is only used for comparison, so any non-null
1076 * value will do.
1077 */
1078static void
1079remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
1080{
1081 static u_int32_t last_remove = 0;
1082
1083#define FORCE (keep_me == NULL)
1084
1085 ipfw_dyn_rule *prev, *q;
1086 int i, pass = 0, max_pass = 0;
1087
1088 IPFW_DYN_LOCK_ASSERT();
1089
1090 if (ipfw_dyn_v == NULL || dyn_count == 0)
1091 return;
1092 /* do not expire more than once per second, it is useless */
1093 if (!FORCE && last_remove == time_uptime)
1094 return;
1095 last_remove = time_uptime;
1096
1097 /*
1098 * because O_LIMIT refer to parent rules, during the first pass only
1099 * remove child and mark any pending LIMIT_PARENT, and remove
1100 * them in a second pass.
1101 */
1102next_pass:
1103 for (i = 0 ; i < curr_dyn_buckets ; i++) {
1104 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
1105 /*
1106 * Logic can become complex here, so we split tests.
1107 */
1108 if (q == keep_me)
1109 goto next;
1110 if (rule != NULL && rule != q->rule)
1111 goto next; /* not the one we are looking for */
1112 if (q->dyn_type == O_LIMIT_PARENT) {
1113 /*
1114 * handle parent in the second pass,
1115 * record we need one.
1116 */
1117 max_pass = 1;
1118 if (pass == 0)
1119 goto next;
1120 if (FORCE && q->count != 0 ) {
1121 /* XXX should not happen! */
1122 printf("ipfw: OUCH! cannot remove rule,"
1123 " count %d\n", q->count);
1124 }
1125 } else {
1126 if (!FORCE &&
1127 !TIME_LEQ( q->expire, time_uptime ))
1128 goto next;
1129 }
1130 if (q->dyn_type != O_LIMIT_PARENT || !q->count) {
1131 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1132 continue;
1133 }
1134next:
1135 prev=q;
1136 q=q->next;
1137 }
1138 }
1139 if (pass++ < max_pass)
1140 goto next_pass;
1141}
1142
1143
1144/**
1145 * lookup a dynamic rule.
1146 */
1147static ipfw_dyn_rule *
1148lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction,
1149 struct tcphdr *tcp)
1150{
1151 /*
1152 * stateful ipfw extensions.
1153 * Lookup into dynamic session queue
1154 */
1155#define MATCH_REVERSE 0
1156#define MATCH_FORWARD 1
1157#define MATCH_NONE 2
1158#define MATCH_UNKNOWN 3
1159 int i, dir = MATCH_NONE;
1160 ipfw_dyn_rule *prev, *q=NULL;
1161
1162 IPFW_DYN_LOCK_ASSERT();
1163
1164 if (ipfw_dyn_v == NULL)
1165 goto done; /* not found */
1166 i = hash_packet( pkt );
1167 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
1168 if (q->dyn_type == O_LIMIT_PARENT && q->count)
1169 goto next;
1170 if (TIME_LEQ( q->expire, time_uptime)) { /* expire entry */
1171 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1172 continue;
1173 }
1174 if (pkt->proto == q->id.proto &&
1175 q->dyn_type != O_LIMIT_PARENT) {
1176 if (IS_IP6_FLOW_ID(pkt)) {
1177 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1178 &(q->id.src_ip6)) &&
1179 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1180 &(q->id.dst_ip6)) &&
1181 pkt->src_port == q->id.src_port &&
1182 pkt->dst_port == q->id.dst_port ) {
1183 dir = MATCH_FORWARD;
1184 break;
1185 }
1186 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1187 &(q->id.dst_ip6)) &&
1188 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1189 &(q->id.src_ip6)) &&
1190 pkt->src_port == q->id.dst_port &&
1191 pkt->dst_port == q->id.src_port ) {
1192 dir = MATCH_REVERSE;
1193 break;
1194 }
1195 } else {
1196 if (pkt->src_ip == q->id.src_ip &&
1197 pkt->dst_ip == q->id.dst_ip &&
1198 pkt->src_port == q->id.src_port &&
1199 pkt->dst_port == q->id.dst_port ) {
1200 dir = MATCH_FORWARD;
1201 break;
1202 }
1203 if (pkt->src_ip == q->id.dst_ip &&
1204 pkt->dst_ip == q->id.src_ip &&
1205 pkt->src_port == q->id.dst_port &&
1206 pkt->dst_port == q->id.src_port ) {
1207 dir = MATCH_REVERSE;
1208 break;
1209 }
1210 }
1211 }
1212next:
1213 prev = q;
1214 q = q->next;
1215 }
1216 if (q == NULL)
1217 goto done; /* q = NULL, not found */
1218
1219 if ( prev != NULL) { /* found and not in front */
1220 prev->next = q->next;
1221 q->next = ipfw_dyn_v[i];
1222 ipfw_dyn_v[i] = q;
1223 }
1224 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
1225 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
1226
1227#define BOTH_SYN (TH_SYN | (TH_SYN << 8))
1228#define BOTH_FIN (TH_FIN | (TH_FIN << 8))
1229 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
1230 switch (q->state) {
1231 case TH_SYN: /* opening */
1232 q->expire = time_uptime + dyn_syn_lifetime;
1233 break;
1234
1235 case BOTH_SYN: /* move to established */
1236 case BOTH_SYN | TH_FIN : /* one side tries to close */
1237 case BOTH_SYN | (TH_FIN << 8) :
1238 if (tcp) {
1239#define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
1240 u_int32_t ack = ntohl(tcp->th_ack);
1241 if (dir == MATCH_FORWARD) {
1242 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
1243 q->ack_fwd = ack;
1244 else { /* ignore out-of-sequence */
1245 break;
1246 }
1247 } else {
1248 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
1249 q->ack_rev = ack;
1250 else { /* ignore out-of-sequence */
1251 break;
1252 }
1253 }
1254 }
1255 q->expire = time_uptime + dyn_ack_lifetime;
1256 break;
1257
1258 case BOTH_SYN | BOTH_FIN: /* both sides closed */
1259 if (dyn_fin_lifetime >= dyn_keepalive_period)
1260 dyn_fin_lifetime = dyn_keepalive_period - 1;
1261 q->expire = time_uptime + dyn_fin_lifetime;
1262 break;
1263
1264 default:
1265#if 0
1266 /*
1267 * reset or some invalid combination, but can also
1268 * occur if we use keep-state the wrong way.
1269 */
1270 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
1271 printf("invalid state: 0x%x\n", q->state);
1272#endif
1273 if (dyn_rst_lifetime >= dyn_keepalive_period)
1274 dyn_rst_lifetime = dyn_keepalive_period - 1;
1275 q->expire = time_uptime + dyn_rst_lifetime;
1276 break;
1277 }
1278 } else if (pkt->proto == IPPROTO_UDP) {
1279 q->expire = time_uptime + dyn_udp_lifetime;
1280 } else {
1281 /* other protocols */
1282 q->expire = time_uptime + dyn_short_lifetime;
1283 }
1284done:
1285 if (match_direction)
1286 *match_direction = dir;
1287 return q;
1288}
1289
1290static ipfw_dyn_rule *
1291lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
1292 struct tcphdr *tcp)
1293{
1294 ipfw_dyn_rule *q;
1295
1296 IPFW_DYN_LOCK();
1297 q = lookup_dyn_rule_locked(pkt, match_direction, tcp);
1298 if (q == NULL)
1299 IPFW_DYN_UNLOCK();
1300 /* NB: return table locked when q is not NULL */
1301 return q;
1302}
1303
1304static void
1305realloc_dynamic_table(void)
1306{
1307 IPFW_DYN_LOCK_ASSERT();
1308
1309 /*
1310 * Try reallocation, make sure we have a power of 2 and do
1311 * not allow more than 64k entries. In case of overflow,
1312 * default to 1024.
1313 */
1314
1315 if (dyn_buckets > 65536)
1316 dyn_buckets = 1024;
1317 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
1318 dyn_buckets = curr_dyn_buckets; /* reset */
1319 return;
1320 }
1321 curr_dyn_buckets = dyn_buckets;
1322 if (ipfw_dyn_v != NULL)
1323 free(ipfw_dyn_v, M_IPFW);
1324 for (;;) {
1325 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
1326 M_IPFW, M_NOWAIT | M_ZERO);
1327 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
1328 break;
1329 curr_dyn_buckets /= 2;
1330 }
1331}
1332
1333/**
1334 * Install state of type 'type' for a dynamic session.
1335 * The hash table contains two type of rules:
1336 * - regular rules (O_KEEP_STATE)
1337 * - rules for sessions with limited number of sess per user
1338 * (O_LIMIT). When they are created, the parent is
1339 * increased by 1, and decreased on delete. In this case,
1340 * the third parameter is the parent rule and not the chain.
1341 * - "parent" rules for the above (O_LIMIT_PARENT).
1342 */
1343static ipfw_dyn_rule *
1344add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
1345{
1346 ipfw_dyn_rule *r;
1347 int i;
1348
1349 IPFW_DYN_LOCK_ASSERT();
1350
1351 if (ipfw_dyn_v == NULL ||
1352 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1353 realloc_dynamic_table();
1354 if (ipfw_dyn_v == NULL)
1355 return NULL; /* failed ! */
1356 }
1357 i = hash_packet(id);
1358
1359 r = uma_zalloc(ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO);
1360 if (r == NULL) {
1361 printf ("ipfw: sorry cannot allocate state\n");
1362 return NULL;
1363 }
1364
1365 /* increase refcount on parent, and set pointer */
1366 if (dyn_type == O_LIMIT) {
1367 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1368 if ( parent->dyn_type != O_LIMIT_PARENT)
1369 panic("invalid parent");
1370 parent->count++;
1371 r->parent = parent;
1372 rule = parent->rule;
1373 }
1374
1375 r->id = *id;
1376 r->expire = time_uptime + dyn_syn_lifetime;
1377 r->rule = rule;
1378 r->dyn_type = dyn_type;
1379 r->pcnt = r->bcnt = 0;
1380 r->count = 0;
1381
1382 r->bucket = i;
1383 r->next = ipfw_dyn_v[i];
1384 ipfw_dyn_v[i] = r;
1385 dyn_count++;
1386 DEB(printf("ipfw: add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1387 dyn_type,
1388 (r->id.src_ip), (r->id.src_port),
1389 (r->id.dst_ip), (r->id.dst_port),
1390 dyn_count ); )
1391 return r;
1392}
1393
1394/**
1395 * lookup dynamic parent rule using pkt and rule as search keys.
1396 * If the lookup fails, then install one.
1397 */
1398static ipfw_dyn_rule *
1399lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1400{
1401 ipfw_dyn_rule *q;
1402 int i;
1403
1404 IPFW_DYN_LOCK_ASSERT();
1405
1406 if (ipfw_dyn_v) {
1407 int is_v6 = IS_IP6_FLOW_ID(pkt);
1408 i = hash_packet( pkt );
1409 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
1410 if (q->dyn_type == O_LIMIT_PARENT &&
1411 rule== q->rule &&
1412 pkt->proto == q->id.proto &&
1413 pkt->src_port == q->id.src_port &&
1414 pkt->dst_port == q->id.dst_port &&
1415 (
1416 (is_v6 &&
1417 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1418 &(q->id.src_ip6)) &&
1419 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1420 &(q->id.dst_ip6))) ||
1421 (!is_v6 &&
1422 pkt->src_ip == q->id.src_ip &&
1423 pkt->dst_ip == q->id.dst_ip)
1424 )
1425 ) {
1426 q->expire = time_uptime + dyn_short_lifetime;
1427 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);)
1428 return q;
1429 }
1430 }
1431 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1432}
1433
1434/**
1435 * Install dynamic state for rule type cmd->o.opcode
1436 *
1437 * Returns 1 (failure) if state is not installed because of errors or because
1438 * session limitations are enforced.
1439 */
1440static int
1441install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1442 struct ip_fw_args *args)
1443{
1444 static int last_log;
1445
1446 ipfw_dyn_rule *q;
1447
1448 DEB(printf("ipfw: install state type %d 0x%08x %u -> 0x%08x %u\n",
1449 cmd->o.opcode,
1450 (args->f_id.src_ip), (args->f_id.src_port),
1451 (args->f_id.dst_ip), (args->f_id.dst_port) );)
1452
1453 IPFW_DYN_LOCK();
1454
1455 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
1456
1457 if (q != NULL) { /* should never occur */
1458 if (last_log != time_uptime) {
1459 last_log = time_uptime;
1460 printf("ipfw: install_state: entry already present, done\n");
1461 }
1462 IPFW_DYN_UNLOCK();
1463 return 0;
1464 }
1465
1466 if (dyn_count >= dyn_max)
1467 /*
1468 * Run out of slots, try to remove any expired rule.
1469 */
1470 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1471
1472 if (dyn_count >= dyn_max) {
1473 if (last_log != time_uptime) {
1474 last_log = time_uptime;
1475 printf("ipfw: install_state: Too many dynamic rules\n");
1476 }
1477 IPFW_DYN_UNLOCK();
1478 return 1; /* cannot install, notify caller */
1479 }
1480
1481 switch (cmd->o.opcode) {
1482 case O_KEEP_STATE: /* bidir rule */
1483 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1484 break;
1485
1486 case O_LIMIT: /* limit number of sessions */
1487 {
1488 u_int16_t limit_mask = cmd->limit_mask;
1489 struct ipfw_flow_id id;
1490 ipfw_dyn_rule *parent;
1491
1492 DEB(printf("ipfw: installing dyn-limit rule %d\n",
1493 cmd->conn_limit);)
1494
1495 id.dst_ip = id.src_ip = 0;
1496 id.dst_port = id.src_port = 0;
1497 id.proto = args->f_id.proto;
1498
1499 if (IS_IP6_FLOW_ID (&(args->f_id))) {
1500 if (limit_mask & DYN_SRC_ADDR)
1501 id.src_ip6 = args->f_id.src_ip6;
1502 if (limit_mask & DYN_DST_ADDR)
1503 id.dst_ip6 = args->f_id.dst_ip6;
1504 } else {
1505 if (limit_mask & DYN_SRC_ADDR)
1506 id.src_ip = args->f_id.src_ip;
1507 if (limit_mask & DYN_DST_ADDR)
1508 id.dst_ip = args->f_id.dst_ip;
1509 }
1510 if (limit_mask & DYN_SRC_PORT)
1511 id.src_port = args->f_id.src_port;
1512 if (limit_mask & DYN_DST_PORT)
1513 id.dst_port = args->f_id.dst_port;
1514 parent = lookup_dyn_parent(&id, rule);
1515 if (parent == NULL) {
1516 printf("ipfw: add parent failed\n");
1517 IPFW_DYN_UNLOCK();
1518 return 1;
1519 }
1520 if (parent->count >= cmd->conn_limit) {
1521 /*
1522 * See if we can remove some expired rule.
1523 */
1524 remove_dyn_rule(rule, parent);
1525 if (parent->count >= cmd->conn_limit) {
1526 if (fw_verbose && last_log != time_uptime) {
1527 last_log = time_uptime;
1528 log(LOG_SECURITY | LOG_DEBUG,
1529 "drop session, too many entries\n");
1530 }
1531 IPFW_DYN_UNLOCK();
1532 return 1;
1533 }
1534 }
1535 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1536 }
1537 break;
1538 default:
1539 printf("ipfw: unknown dynamic rule type %u\n", cmd->o.opcode);
1540 IPFW_DYN_UNLOCK();
1541 return 1;
1542 }
1543 lookup_dyn_rule_locked(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1544 IPFW_DYN_UNLOCK();
1545 return 0;
1546}
1547
1548/*
1549 * Generate a TCP packet, containing either a RST or a keepalive.
1550 * When flags & TH_RST, we are sending a RST packet, because of a
1551 * "reset" action matched the packet.
1552 * Otherwise we are sending a keepalive, and flags & TH_
1553 */
1554static struct mbuf *
1555send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags)
1556{
1557 struct mbuf *m;
1558 struct ip *ip;
1559 struct tcphdr *tcp;
1560
1561 MGETHDR(m, M_DONTWAIT, MT_DATA);
1562 if (m == 0)
1563 return (NULL);
1564 m->m_pkthdr.rcvif = (struct ifnet *)0;
1565 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1566 m->m_data += max_linkhdr;
1567
1568 ip = mtod(m, struct ip *);
1569 bzero(ip, m->m_len);
1570 tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1571 ip->ip_p = IPPROTO_TCP;
1572 tcp->th_off = 5;
1573 /*
1574 * Assume we are sending a RST (or a keepalive in the reverse
1575 * direction), swap src and destination addresses and ports.
1576 */
1577 ip->ip_src.s_addr = htonl(id->dst_ip);
1578 ip->ip_dst.s_addr = htonl(id->src_ip);
1579 tcp->th_sport = htons(id->dst_port);
1580 tcp->th_dport = htons(id->src_port);
1581 if (flags & TH_RST) { /* we are sending a RST */
1582 if (flags & TH_ACK) {
1583 tcp->th_seq = htonl(ack);
1584 tcp->th_ack = htonl(0);
1585 tcp->th_flags = TH_RST;
1586 } else {
1587 if (flags & TH_SYN)
1588 seq++;
1589 tcp->th_seq = htonl(0);
1590 tcp->th_ack = htonl(seq);
1591 tcp->th_flags = TH_RST | TH_ACK;
1592 }
1593 } else {
1594 /*
1595 * We are sending a keepalive. flags & TH_SYN determines
1596 * the direction, forward if set, reverse if clear.
1597 * NOTE: seq and ack are always assumed to be correct
1598 * as set by the caller. This may be confusing...
1599 */
1600 if (flags & TH_SYN) {
1601 /*
1602 * we have to rewrite the correct addresses!
1603 */
1604 ip->ip_dst.s_addr = htonl(id->dst_ip);
1605 ip->ip_src.s_addr = htonl(id->src_ip);
1606 tcp->th_dport = htons(id->dst_port);
1607 tcp->th_sport = htons(id->src_port);
1608 }
1609 tcp->th_seq = htonl(seq);
1610 tcp->th_ack = htonl(ack);
1611 tcp->th_flags = TH_ACK;
1612 }
1613 /*
1614 * set ip_len to the payload size so we can compute
1615 * the tcp checksum on the pseudoheader
1616 * XXX check this, could save a couple of words ?
1617 */
1618 ip->ip_len = htons(sizeof(struct tcphdr));
1619 tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1620 /*
1621 * now fill fields left out earlier
1622 */
1623 ip->ip_ttl = ip_defttl;
1624 ip->ip_len = m->m_pkthdr.len;
1625 m->m_flags |= M_SKIP_FIREWALL;
1626 return (m);
1627}
1628
1629/*
1630 * sends a reject message, consuming the mbuf passed as an argument.
1631 */
1632static void
1633send_reject(struct ip_fw_args *args, int code, u_short offset, int ip_len)
1634{
1635
1636 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1637 /* We need the IP header in host order for icmp_error(). */
1638 if (args->eh != NULL) {
1639 struct ip *ip = mtod(args->m, struct ip *);
1640 ip->ip_len = ntohs(ip->ip_len);
1641 ip->ip_off = ntohs(ip->ip_off);
1642 }
1643 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1644 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1645 struct tcphdr *const tcp =
1646 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1647 if ( (tcp->th_flags & TH_RST) == 0) {
1648 struct mbuf *m;
1649 m = send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1650 ntohl(tcp->th_ack),
1651 tcp->th_flags | TH_RST);
1652 if (m != NULL)
1653 ip_output(m, NULL, NULL, 0, NULL, NULL);
1654 }
1655 m_freem(args->m);
1656 } else
1657 m_freem(args->m);
1658 args->m = NULL;
1659}
1660
1661/**
1662 *
1663 * Given an ip_fw *, lookup_next_rule will return a pointer
1664 * to the next rule, which can be either the jump
1665 * target (for skipto instructions) or the next one in the list (in
1666 * all other cases including a missing jump target).
1667 * The result is also written in the "next_rule" field of the rule.
1668 * Backward jumps are not allowed, so start looking from the next
1669 * rule...
1670 *
1671 * This never returns NULL -- in case we do not have an exact match,
1672 * the next rule is returned. When the ruleset is changed,
1673 * pointers are flushed so we are always correct.
1674 */
1675
1676static struct ip_fw *
1677lookup_next_rule(struct ip_fw *me)
1678{
1679 struct ip_fw *rule = NULL;
1680 ipfw_insn *cmd;
1681
1682 /* look for action, in case it is a skipto */
1683 cmd = ACTION_PTR(me);
1684 if (cmd->opcode == O_LOG)
1685 cmd += F_LEN(cmd);
1686 if (cmd->opcode == O_ALTQ)
1687 cmd += F_LEN(cmd);
1688 if ( cmd->opcode == O_SKIPTO )
1689 for (rule = me->next; rule ; rule = rule->next)
1690 if (rule->rulenum >= cmd->arg1)
1691 break;
1692 if (rule == NULL) /* failure or not a skipto */
1693 rule = me->next;
1694 me->next_rule = rule;
1695 return rule;
1696}
1697
1698static void
1699init_tables(struct ip_fw_chain *ch)
1700{
1701 int i;
1702
1703 for (i = 0; i < IPFW_TABLES_MAX; i++)
1704 rn_inithead((void **)&ch->tables[i], 32);
1705}
1706
1707static int
1708add_table_entry(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1709 uint8_t mlen, uint32_t value)
1710{
1711 struct radix_node_head *rnh;
1712 struct table_entry *ent;
1713
1714 if (tbl >= IPFW_TABLES_MAX)
1715 return (EINVAL);
1716 rnh = ch->tables[tbl];
1717 ent = malloc(sizeof(*ent), M_IPFW_TBL, M_NOWAIT | M_ZERO);
1718 if (ent == NULL)
1719 return (ENOMEM);
1720 ent->value = value;
1721 ent->addr.sin_len = ent->mask.sin_len = 8;
1722 ent->mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1723 ent->addr.sin_addr.s_addr = addr & ent->mask.sin_addr.s_addr;
1724 IPFW_WLOCK(&layer3_chain);
1725 if (rnh->rnh_addaddr(&ent->addr, &ent->mask, rnh, (void *)ent) ==
1726 NULL) {
1727 IPFW_WUNLOCK(&layer3_chain);
1728 free(ent, M_IPFW_TBL);
1729 return (EEXIST);
1730 }
1731 IPFW_WUNLOCK(&layer3_chain);
1732 return (0);
1733}
1734
1735static int
1736del_table_entry(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1737 uint8_t mlen)
1738{
1739 struct radix_node_head *rnh;
1740 struct table_entry *ent;
1741 struct sockaddr_in sa, mask;
1742
1743 if (tbl >= IPFW_TABLES_MAX)
1744 return (EINVAL);
1745 rnh = ch->tables[tbl];
1746 sa.sin_len = mask.sin_len = 8;
1747 mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1748 sa.sin_addr.s_addr = addr & mask.sin_addr.s_addr;
1749 IPFW_WLOCK(ch);
1750 ent = (struct table_entry *)rnh->rnh_deladdr(&sa, &mask, rnh);
1751 if (ent == NULL) {
1752 IPFW_WUNLOCK(ch);
1753 return (ESRCH);
1754 }
1755 IPFW_WUNLOCK(ch);
1756 free(ent, M_IPFW_TBL);
1757 return (0);
1758}
1759
1760static int
1761flush_table_entry(struct radix_node *rn, void *arg)
1762{
1763 struct radix_node_head * const rnh = arg;
1764 struct table_entry *ent;
1765
1766 ent = (struct table_entry *)
1767 rnh->rnh_deladdr(rn->rn_key, rn->rn_mask, rnh);
1768 if (ent != NULL)
1769 free(ent, M_IPFW_TBL);
1770 return (0);
1771}
1772
1773static int
1774flush_table(struct ip_fw_chain *ch, uint16_t tbl)
1775{
1776 struct radix_node_head *rnh;
1777
1778 IPFW_WLOCK_ASSERT(ch);
1779
1780 if (tbl >= IPFW_TABLES_MAX)
1781 return (EINVAL);
1782 rnh = ch->tables[tbl];
1783 rnh->rnh_walktree(rnh, flush_table_entry, rnh);
1784 return (0);
1785}
1786
1787static void
1788flush_tables(struct ip_fw_chain *ch)
1789{
1790 uint16_t tbl;
1791
1792 IPFW_WLOCK_ASSERT(ch);
1793
1794 for (tbl = 0; tbl < IPFW_TABLES_MAX; tbl++)
1795 flush_table(ch, tbl);
1796}
1797
1798static int
1799lookup_table(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1800 uint32_t *val)
1801{
1802 struct radix_node_head *rnh;
1803 struct table_entry *ent;
1804 struct sockaddr_in sa;
1805
1806 if (tbl >= IPFW_TABLES_MAX)
1807 return (0);
1808 rnh = ch->tables[tbl];
1809 sa.sin_len = 8;
1810 sa.sin_addr.s_addr = addr;
1811 ent = (struct table_entry *)(rnh->rnh_lookup(&sa, NULL, rnh));
1812 if (ent != NULL) {
1813 *val = ent->value;
1814 return (1);
1815 }
1816 return (0);
1817}
1818
1819static int
1820count_table_entry(struct radix_node *rn, void *arg)
1821{
1822 u_int32_t * const cnt = arg;
1823
1824 (*cnt)++;
1825 return (0);
1826}
1827
1828static int
1829count_table(struct ip_fw_chain *ch, uint32_t tbl, uint32_t *cnt)
1830{
1831 struct radix_node_head *rnh;
1832
1833 if (tbl >= IPFW_TABLES_MAX)
1834 return (EINVAL);
1835 rnh = ch->tables[tbl];
1836 *cnt = 0;
1837 rnh->rnh_walktree(rnh, count_table_entry, cnt);
1838 return (0);
1839}
1840
1841static int
1842dump_table_entry(struct radix_node *rn, void *arg)
1843{
1844 struct table_entry * const n = (struct table_entry *)rn;
1845 ipfw_table * const tbl = arg;
1846 ipfw_table_entry *ent;
1847
1848 if (tbl->cnt == tbl->size)
1849 return (1);
1850 ent = &tbl->ent[tbl->cnt];
1851 ent->tbl = tbl->tbl;
1852 if (in_nullhost(n->mask.sin_addr))
1853 ent->masklen = 0;
1854 else
1855 ent->masklen = 33 - ffs(ntohl(n->mask.sin_addr.s_addr));
1856 ent->addr = n->addr.sin_addr.s_addr;
1857 ent->value = n->value;
1858 tbl->cnt++;
1859 return (0);
1860}
1861
1862static int
1863dump_table(struct ip_fw_chain *ch, ipfw_table *tbl)
1864{
1865 struct radix_node_head *rnh;
1866
1867 IPFW_WLOCK_ASSERT(ch);
1868
1869 if (tbl->tbl >= IPFW_TABLES_MAX)
1870 return (EINVAL);
1871 rnh = ch->tables[tbl->tbl];
1872 tbl->cnt = 0;
1873 rnh->rnh_walktree(rnh, dump_table_entry, tbl);
1874 return (0);
1875}
1876
1877static void
1878fill_ugid_cache(struct inpcb *inp, struct ip_fw_ugid *ugp)
1879{
1880 struct ucred *cr;
1881
1882 if (inp->inp_socket != NULL) {
1883 cr = inp->inp_socket->so_cred;
1884 ugp->fw_prid = jailed(cr) ?
1885 cr->cr_prison->pr_id : -1;
1886 ugp->fw_uid = cr->cr_uid;
1887 ugp->fw_ngroups = cr->cr_ngroups;
1888 bcopy(cr->cr_groups, ugp->fw_groups,
1889 sizeof(ugp->fw_groups));
1890 }
1891}
1892
1893static int
1894check_uidgid(ipfw_insn_u32 *insn,
1895 int proto, struct ifnet *oif,
1896 struct in_addr dst_ip, u_int16_t dst_port,
1897 struct in_addr src_ip, u_int16_t src_port,
1898 struct ip_fw_ugid *ugp, int *lookup, struct inpcb *inp)
1899{
1900 struct inpcbinfo *pi;
1901 int wildcard;
1902 struct inpcb *pcb;
1903 int match;
1904 gid_t *gp;
1905
1906 /*
1907 * Check to see if the UDP or TCP stack supplied us with
1908 * the PCB. If so, rather then holding a lock and looking
1909 * up the PCB, we can use the one that was supplied.
1910 */
1911 if (inp && *lookup == 0) {
1912 INP_LOCK_ASSERT(inp);
1913 if (inp->inp_socket != NULL) {
1914 fill_ugid_cache(inp, ugp);
1915 *lookup = 1;
1916 }
1917 }
1918 /*
1919 * If we have already been here and the packet has no
1920 * PCB entry associated with it, then we can safely
1921 * assume that this is a no match.
1922 */
1923 if (*lookup == -1)
1924 return (0);
1925 if (proto == IPPROTO_TCP) {
1926 wildcard = 0;
1927 pi = &tcbinfo;
1928 } else if (proto == IPPROTO_UDP) {
1929 wildcard = 1;
1930 pi = &udbinfo;
1931 } else
1932 return 0;
1933 match = 0;
1934 if (*lookup == 0) {
1935 INP_INFO_RLOCK(pi);
1936 pcb = (oif) ?
1937 in_pcblookup_hash(pi,
1938 dst_ip, htons(dst_port),
1939 src_ip, htons(src_port),
1940 wildcard, oif) :
1941 in_pcblookup_hash(pi,
1942 src_ip, htons(src_port),
1943 dst_ip, htons(dst_port),
1944 wildcard, NULL);
1945 if (pcb != NULL) {
1946 INP_LOCK(pcb);
1947 if (pcb->inp_socket != NULL) {
1948 fill_ugid_cache(pcb, ugp);
1949 *lookup = 1;
1950 }
1951 INP_UNLOCK(pcb);
1952 }
1953 INP_INFO_RUNLOCK(pi);
1954 if (*lookup == 0) {
1955 /*
1956 * If the lookup did not yield any results, there
1957 * is no sense in coming back and trying again. So
1958 * we can set lookup to -1 and ensure that we wont
1959 * bother the pcb system again.
1960 */
1961 *lookup = -1;
1962 return (0);
1963 }
1964 }
1965 if (insn->o.opcode == O_UID)
1966 match = (ugp->fw_uid == (uid_t)insn->d[0]);
1967 else if (insn->o.opcode == O_GID) {
1968 for (gp = ugp->fw_groups;
1969 gp < &ugp->fw_groups[ugp->fw_ngroups]; gp++)
1970 if (*gp == (gid_t)insn->d[0]) {
1971 match = 1;
1972 break;
1973 }
1974 } else if (insn->o.opcode == O_JAIL)
1975 match = (ugp->fw_prid == (int)insn->d[0]);
1976 return match;
1977}
1978
1979/*
1980 * The main check routine for the firewall.
1981 *
1982 * All arguments are in args so we can modify them and return them
1983 * back to the caller.
1984 *
1985 * Parameters:
1986 *
1987 * args->m (in/out) The packet; we set to NULL when/if we nuke it.
1988 * Starts with the IP header.
1989 * args->eh (in) Mac header if present, or NULL for layer3 packet.
1990 * args->oif Outgoing interface, or NULL if packet is incoming.
1991 * The incoming interface is in the mbuf. (in)
1992 * args->divert_rule (in/out)
1993 * Skip up to the first rule past this rule number;
1994 * upon return, non-zero port number for divert or tee.
1995 *
1996 * args->rule Pointer to the last matching rule (in/out)
1997 * args->next_hop Socket we are forwarding to (out).
1998 * args->f_id Addresses grabbed from the packet (out)
1999 * args->cookie a cookie depending on rule action
2000 *
2001 * Return value:
2002 *
2003 * IP_FW_PASS the packet must be accepted
2004 * IP_FW_DENY the packet must be dropped
2005 * IP_FW_DIVERT divert packet, port in m_tag
2006 * IP_FW_TEE tee packet, port in m_tag
2007 * IP_FW_DUMMYNET to dummynet, pipe in args->cookie
2008 * IP_FW_NETGRAPH into netgraph, cookie args->cookie
2009 *
2010 */
2011
2012int
2013ipfw_chk(struct ip_fw_args *args)
2014{
2015 /*
2016 * Local variables hold state during the processing of a packet.
2017 *
2018 * IMPORTANT NOTE: to speed up the processing of rules, there
2019 * are some assumption on the values of the variables, which
2020 * are documented here. Should you change them, please check
2021 * the implementation of the various instructions to make sure
2022 * that they still work.
2023 *
2024 * args->eh The MAC header. It is non-null for a layer2
2025 * packet, it is NULL for a layer-3 packet.
2026 *
2027 * m | args->m Pointer to the mbuf, as received from the caller.
2028 * It may change if ipfw_chk() does an m_pullup, or if it
2029 * consumes the packet because it calls send_reject().
2030 * XXX This has to change, so that ipfw_chk() never modifies
2031 * or consumes the buffer.
2032 * ip is simply an alias of the value of m, and it is kept
2033 * in sync with it (the packet is supposed to start with
2034 * the ip header).
2035 */
2036 struct mbuf *m = args->m;
2037 struct ip *ip = mtod(m, struct ip *);
2038
2039 /*
2040 * For rules which contain uid/gid or jail constraints, cache
2041 * a copy of the users credentials after the pcb lookup has been
2042 * executed. This will speed up the processing of rules with
2043 * these types of constraints, as well as decrease contention
2044 * on pcb related locks.
2045 */
2046 struct ip_fw_ugid fw_ugid_cache;
2047 int ugid_lookup = 0;
2048
2049 /*
2050 * divinput_flags If non-zero, set to the IP_FW_DIVERT_*_FLAG
2051 * associated with a packet input on a divert socket. This
2052 * will allow to distinguish traffic and its direction when
2053 * it originates from a divert socket.
2054 */
2055 u_int divinput_flags = 0;
2056
2057 /*
2058 * oif | args->oif If NULL, ipfw_chk has been called on the
2059 * inbound path (ether_input, ip_input).
2060 * If non-NULL, ipfw_chk has been called on the outbound path
2061 * (ether_output, ip_output).
2062 */
2063 struct ifnet *oif = args->oif;
2064
2065 struct ip_fw *f = NULL; /* matching rule */
2066 int retval = 0;
2067
2068 /*
2069 * hlen The length of the IP header.
2070 */
2071 u_int hlen = 0; /* hlen >0 means we have an IP pkt */
2072
2073 /*
2074 * offset The offset of a fragment. offset != 0 means that
2075 * we have a fragment at this offset of an IPv4 packet.
2076 * offset == 0 means that (if this is an IPv4 packet)
2077 * this is the first or only fragment.
2078 * For IPv6 offset == 0 means there is no Fragment Header.
2079 * If offset != 0 for IPv6 always use correct mask to
2080 * get the correct offset because we add IP6F_MORE_FRAG
2081 * to be able to dectect the first fragment which would
2082 * otherwise have offset = 0.
2083 */
2084 u_short offset = 0;
2085
2086 /*
2087 * Local copies of addresses. They are only valid if we have
2088 * an IP packet.
2089 *
2090 * proto The protocol. Set to 0 for non-ip packets,
2091 * or to the protocol read from the packet otherwise.
2092 * proto != 0 means that we have an IPv4 packet.
2093 *
2094 * src_port, dst_port port numbers, in HOST format. Only
2095 * valid for TCP and UDP packets.
2096 *
2097 * src_ip, dst_ip ip addresses, in NETWORK format.
2098 * Only valid for IPv4 packets.
2099 */
2100 u_int8_t proto;
2101 u_int16_t src_port = 0, dst_port = 0; /* NOTE: host format */
2102 struct in_addr src_ip, dst_ip; /* NOTE: network format */
2103 u_int16_t ip_len=0;
2104 int pktlen;
2105
2106 /*
2107 * dyn_dir = MATCH_UNKNOWN when rules unchecked,
2108 * MATCH_NONE when checked and not matched (q = NULL),
2109 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
2110 */
2111 int dyn_dir = MATCH_UNKNOWN;
2112 ipfw_dyn_rule *q = NULL;
2113 struct ip_fw_chain *chain = &layer3_chain;
2114 struct m_tag *mtag;
2115
2116 /*
2117 * We store in ulp a pointer to the upper layer protocol header.
2118 * In the ipv4 case this is easy to determine from the header,
2119 * but for ipv6 we might have some additional headers in the middle.
2120 * ulp is NULL if not found.
2121 */
2122 void *ulp = NULL; /* upper layer protocol pointer. */
2123 /* XXX ipv6 variables */
2124 int is_ipv6 = 0;
2125 u_int16_t ext_hd = 0; /* bits vector for extension header filtering */
2126 /* end of ipv6 variables */
2127 int is_ipv4 = 0;
2128
2129 if (m->m_flags & M_SKIP_FIREWALL)
2130 return (IP_FW_PASS); /* accept */
2131
2132 pktlen = m->m_pkthdr.len;
2133 proto = args->f_id.proto = 0; /* mark f_id invalid */
2134 /* XXX 0 is a valid proto: IP/IPv6 Hop-by-Hop Option */
2135
2136/*
2137 * PULLUP_TO(len, p, T) makes sure that len + sizeof(T) is contiguous,
2138 * then it sets p to point at the offset "len" in the mbuf. WARNING: the
2139 * pointer might become stale after other pullups (but we never use it
2140 * this way).
2141 */
2142#define PULLUP_TO(len, p, T) \
2143do { \
2144 int x = (len) + sizeof(T); \
2145 if ((m)->m_len < x) { \
2146 args->m = m = m_pullup(m, x); \
2147 if (m == NULL) \
2148 goto pullup_failed; \
2149 } \
2150 p = (mtod(m, char *) + (len)); \
2151} while (0)
2152
2153 /* Identify IP packets and fill up variables. */
2154 if (pktlen >= sizeof(struct ip6_hdr) &&
2155 (args->eh == NULL || ntohs(args->eh->ether_type)==ETHERTYPE_IPV6) &&
2156 mtod(m, struct ip *)->ip_v == 6) {
2157 is_ipv6 = 1;
2158 args->f_id.addr_type = 6;
2159 hlen = sizeof(struct ip6_hdr);
2160 proto = mtod(m, struct ip6_hdr *)->ip6_nxt;
2161
2162 /* Search extension headers to find upper layer protocols */
2163 while (ulp == NULL) {
2164 switch (proto) {
2165 case IPPROTO_ICMPV6:
2166 PULLUP_TO(hlen, ulp, struct icmp6_hdr);
2167 args->f_id.flags = ICMP6(ulp)->icmp6_type;
2168 break;
2169
2170 case IPPROTO_TCP:
2171 PULLUP_TO(hlen, ulp, struct tcphdr);
2172 dst_port = TCP(ulp)->th_dport;
2173 src_port = TCP(ulp)->th_sport;
2174 args->f_id.flags = TCP(ulp)->th_flags;
2175 break;
2176
2177 case IPPROTO_UDP:
2178 PULLUP_TO(hlen, ulp, struct udphdr);
2179 dst_port = UDP(ulp)->uh_dport;
2180 src_port = UDP(ulp)->uh_sport;
2181 break;
2182
2183 case IPPROTO_HOPOPTS: /* RFC 2460 */
2184 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2185 ext_hd |= EXT_HOPOPTS;
2186 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2187 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2188 ulp = NULL;
2189 break;
2190
2191 case IPPROTO_ROUTING: /* RFC 2460 */
2192 PULLUP_TO(hlen, ulp, struct ip6_rthdr);
2193 if (((struct ip6_rthdr *)ulp)->ip6r_type != 0) {
2194 printf("IPFW2: IPV6 - Unknown Routing "
2195 "Header type(%d)\n",
2196 ((struct ip6_rthdr *)ulp)->ip6r_type);
2197 if (fw_deny_unknown_exthdrs)
2198 return (IP_FW_DENY);
2199 break;
2200 }
2201 ext_hd |= EXT_ROUTING;
2202 hlen += (((struct ip6_rthdr *)ulp)->ip6r_len + 1) << 3;
2203 proto = ((struct ip6_rthdr *)ulp)->ip6r_nxt;
2204 ulp = NULL;
2205 break;
2206
2207 case IPPROTO_FRAGMENT: /* RFC 2460 */
2208 PULLUP_TO(hlen, ulp, struct ip6_frag);
2209 ext_hd |= EXT_FRAGMENT;
2210 hlen += sizeof (struct ip6_frag);
2211 proto = ((struct ip6_frag *)ulp)->ip6f_nxt;
2212 offset = ((struct ip6_frag *)ulp)->ip6f_offlg &
2213 IP6F_OFF_MASK;
2214 /* Add IP6F_MORE_FRAG for offset of first
2215 * fragment to be != 0. */
2216 offset |= ((struct ip6_frag *)ulp)->ip6f_offlg &
2217 IP6F_MORE_FRAG;
2218 if (offset == 0) {
2219 printf("IPFW2: IPV6 - Invalid Fragment "
2220 "Header\n");
2221 if (fw_deny_unknown_exthdrs)
2222 return (IP_FW_DENY);
2223 break;
2224 }
2225 args->f_id.frag_id6 =
2226 ntohl(((struct ip6_frag *)ulp)->ip6f_ident);
2227 ulp = NULL;
2228 break;
2229
2230 case IPPROTO_DSTOPTS: /* RFC 2460 */
2231 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2232 ext_hd |= EXT_DSTOPTS;
2233 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2234 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2235 ulp = NULL;
2236 break;
2237
2238 case IPPROTO_AH: /* RFC 2402 */
2239 PULLUP_TO(hlen, ulp, struct ip6_ext);
2240 ext_hd |= EXT_AH;
2241 hlen += (((struct ip6_ext *)ulp)->ip6e_len + 2) << 2;
2242 proto = ((struct ip6_ext *)ulp)->ip6e_nxt;
2243 ulp = NULL;
2244 break;
2245
2246 case IPPROTO_ESP: /* RFC 2406 */
2247 PULLUP_TO(hlen, ulp, uint32_t); /* SPI, Seq# */
2248 /* Anything past Seq# is variable length and
2249 * data past this ext. header is encrypted. */
2250 ext_hd |= EXT_ESP;
2251 break;
2252
2253 case IPPROTO_NONE: /* RFC 2460 */
2254 PULLUP_TO(hlen, ulp, struct ip6_ext);
2255 /* Packet ends here. if ip6e_len!=0 octets
2256 * must be ignored. */
2257 break;
2258
2259 case IPPROTO_OSPFIGP:
2260 /* XXX OSPF header check? */
2261 PULLUP_TO(hlen, ulp, struct ip6_ext);
2262 break;
2263
2264 default:
2265 printf("IPFW2: IPV6 - Unknown Extension "
2266 "Header(%d), ext_hd=%x\n", proto, ext_hd);
2267 if (fw_deny_unknown_exthdrs)
2268 return (IP_FW_DENY);
2269 break;
2270 } /*switch */
2271 }
2272 args->f_id.src_ip6 = mtod(m,struct ip6_hdr *)->ip6_src;
2273 args->f_id.dst_ip6 = mtod(m,struct ip6_hdr *)->ip6_dst;
2274 args->f_id.src_ip = 0;
2275 args->f_id.dst_ip = 0;
2276 args->f_id.flow_id6 = ntohl(mtod(m, struct ip6_hdr *)->ip6_flow);
2277 } else if (pktlen >= sizeof(struct ip) &&
2278 (args->eh == NULL || ntohs(args->eh->ether_type) == ETHERTYPE_IP) &&
2279 mtod(m, struct ip *)->ip_v == 4) {
2280 is_ipv4 = 1;
2281 ip = mtod(m, struct ip *);
2282 hlen = ip->ip_hl << 2;
2283 args->f_id.addr_type = 4;
2284
2285 /*
2286 * Collect parameters into local variables for faster matching.
2287 */
2288 proto = ip->ip_p;
2289 src_ip = ip->ip_src;
2290 dst_ip = ip->ip_dst;
2291 if (args->eh != NULL) { /* layer 2 packets are as on the wire */
2292 offset = ntohs(ip->ip_off) & IP_OFFMASK;
2293 ip_len = ntohs(ip->ip_len);
2294 } else {
2295 offset = ip->ip_off & IP_OFFMASK;
2296 ip_len = ip->ip_len;
2297 }
2298 pktlen = ip_len < pktlen ? ip_len : pktlen;
2299
2300 if (offset == 0) {
2301 switch (proto) {
2302 case IPPROTO_TCP:
2303 PULLUP_TO(hlen, ulp, struct tcphdr);
2304 dst_port = TCP(ulp)->th_dport;
2305 src_port = TCP(ulp)->th_sport;
2306 args->f_id.flags = TCP(ulp)->th_flags;
2307 break;
2308
2309 case IPPROTO_UDP:
2310 PULLUP_TO(hlen, ulp, struct udphdr);
2311 dst_port = UDP(ulp)->uh_dport;
2312 src_port = UDP(ulp)->uh_sport;
2313 break;
2314
2315 case IPPROTO_ICMP:
2316 PULLUP_TO(hlen, ulp, struct icmphdr);
2317 args->f_id.flags = ICMP(ulp)->icmp_type;
2318 break;
2319
2320 default:
2321 break;
2322 }
2323 }
2324
2325 args->f_id.src_ip = ntohl(src_ip.s_addr);
2326 args->f_id.dst_ip = ntohl(dst_ip.s_addr);
2327 }
2328#undef PULLUP_TO
2329 if (proto) { /* we may have port numbers, store them */
2330 args->f_id.proto = proto;
2331 args->f_id.src_port = src_port = ntohs(src_port);
2332 args->f_id.dst_port = dst_port = ntohs(dst_port);
2333 }
2334
2335 IPFW_RLOCK(chain);
2336 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
2337 if (args->rule) {
2338 /*
2339 * Packet has already been tagged. Look for the next rule
2340 * to restart processing.
2341 *
2342 * If fw_one_pass != 0 then just accept it.
2343 * XXX should not happen here, but optimized out in
2344 * the caller.
2345 */
2346 if (fw_one_pass) {
2347 IPFW_RUNLOCK(chain);
2348 return (IP_FW_PASS);
2349 }
2350
2351 f = args->rule->next_rule;
2352 if (f == NULL)
2353 f = lookup_next_rule(args->rule);
2354 } else {
2355 /*
2356 * Find the starting rule. It can be either the first
2357 * one, or the one after divert_rule if asked so.
2358 */
2359 int skipto = mtag ? divert_cookie(mtag) : 0;
2360
2361 f = chain->rules;
2362 if (args->eh == NULL && skipto != 0) {
2363 if (skipto >= IPFW_DEFAULT_RULE) {
2364 IPFW_RUNLOCK(chain);
2365 return (IP_FW_DENY); /* invalid */
2366 }
2367 while (f && f->rulenum <= skipto)
2368 f = f->next;
2369 if (f == NULL) { /* drop packet */
2370 IPFW_RUNLOCK(chain);
2371 return (IP_FW_DENY);
2372 }
2373 }
2374 }
2375 /* reset divert rule to avoid confusion later */
2376 if (mtag) {
2377 divinput_flags = divert_info(mtag) &
2378 (IP_FW_DIVERT_OUTPUT_FLAG | IP_FW_DIVERT_LOOPBACK_FLAG);
2379 m_tag_delete(m, mtag);
2380 }
2381
2382 /*
2383 * Now scan the rules, and parse microinstructions for each rule.
2384 */
2385 for (; f; f = f->next) {
| 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_ip6fw.h"
38#include "opt_ipdn.h"
39#include "opt_inet.h"
40#ifndef INET
41#error IPFIREWALL requires INET.
42#endif /* INET */
43#endif
44#include "opt_inet6.h"
45#include "opt_ipsec.h"
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/condvar.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#include <netgraph/ng_ipfw.h>
82
83#include <altq/if_altq.h>
84
85#ifdef IPSEC
86#include <netinet6/ipsec.h>
87#endif
88
89#include <netinet/ip6.h>
90#include <netinet/icmp6.h>
91#ifdef INET6
92#include <netinet6/scope6_var.h>
93#endif
94
95#include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
96
97#include <machine/in_cksum.h> /* XXX for in_cksum */
98
99/*
100 * set_disable contains one bit per set value (0..31).
101 * If the bit is set, all rules with the corresponding set
102 * are disabled. Set RESVD_SET(31) is reserved for the default rule
103 * and rules that are not deleted by the flush command,
104 * and CANNOT be disabled.
105 * Rules in set RESVD_SET can only be deleted explicitly.
106 */
107static u_int32_t set_disable;
108
109static int fw_verbose;
110static int verbose_limit;
111
112static struct callout ipfw_timeout;
113static uma_zone_t ipfw_dyn_rule_zone;
114#define IPFW_DEFAULT_RULE 65535
115
116/*
117 * Data structure to cache our ucred related
118 * information. This structure only gets used if
119 * the user specified UID/GID based constraints in
120 * a firewall rule.
121 */
122struct ip_fw_ugid {
123 gid_t fw_groups[NGROUPS];
124 int fw_ngroups;
125 uid_t fw_uid;
126 int fw_prid;
127};
128
129#define IPFW_TABLES_MAX 128
130struct ip_fw_chain {
131 struct ip_fw *rules; /* list of rules */
132 struct ip_fw *reap; /* list of rules to reap */
133 struct radix_node_head *tables[IPFW_TABLES_MAX];
134 struct mtx mtx; /* lock guarding rule list */
135 int busy_count; /* busy count for rw locks */
136 int want_write;
137 struct cv cv;
138};
139#define IPFW_LOCK_INIT(_chain) \
140 mtx_init(&(_chain)->mtx, "IPFW static rules", NULL, \
141 MTX_DEF | MTX_RECURSE)
142#define IPFW_LOCK_DESTROY(_chain) mtx_destroy(&(_chain)->mtx)
143#define IPFW_WLOCK_ASSERT(_chain) do { \
144 mtx_assert(&(_chain)->mtx, MA_OWNED); \
145 NET_ASSERT_GIANT(); \
146} while (0)
147
148static __inline void
149IPFW_RLOCK(struct ip_fw_chain *chain)
150{
151 mtx_lock(&chain->mtx);
152 chain->busy_count++;
153 mtx_unlock(&chain->mtx);
154}
155
156static __inline void
157IPFW_RUNLOCK(struct ip_fw_chain *chain)
158{
159 mtx_lock(&chain->mtx);
160 chain->busy_count--;
161 if (chain->busy_count == 0 && chain->want_write)
162 cv_signal(&chain->cv);
163 mtx_unlock(&chain->mtx);
164}
165
166static __inline void
167IPFW_WLOCK(struct ip_fw_chain *chain)
168{
169 mtx_lock(&chain->mtx);
170 chain->want_write++;
171 while (chain->busy_count > 0)
172 cv_wait(&chain->cv, &chain->mtx);
173}
174
175static __inline void
176IPFW_WUNLOCK(struct ip_fw_chain *chain)
177{
178 chain->want_write--;
179 cv_signal(&chain->cv);
180 mtx_unlock(&chain->mtx);
181}
182
183/*
184 * list of rules for layer 3
185 */
186static struct ip_fw_chain layer3_chain;
187
188MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
189MALLOC_DEFINE(M_IPFW_TBL, "ipfw_tbl", "IpFw tables");
190
191struct table_entry {
192 struct radix_node rn[2];
193 struct sockaddr_in addr, mask;
194 u_int32_t value;
195};
196
197static int fw_debug = 1;
198static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
199
200#ifdef SYSCTL_NODE
201SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
202SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable,
203 CTLFLAG_RW | CTLFLAG_SECURE3,
204 &fw_enable, 0, "Enable ipfw");
205SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
206 &autoinc_step, 0, "Rule number autincrement step");
207SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, one_pass,
208 CTLFLAG_RW | CTLFLAG_SECURE3,
209 &fw_one_pass, 0,
210 "Only do a single pass through ipfw when using dummynet(4)");
211SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
212 &fw_debug, 0, "Enable printing of debug ip_fw statements");
213SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose,
214 CTLFLAG_RW | CTLFLAG_SECURE3,
215 &fw_verbose, 0, "Log matches to ipfw rules");
216SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
217 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
218
219/*
220 * Description of dynamic rules.
221 *
222 * Dynamic rules are stored in lists accessed through a hash table
223 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
224 * be modified through the sysctl variable dyn_buckets which is
225 * updated when the table becomes empty.
226 *
227 * XXX currently there is only one list, ipfw_dyn.
228 *
229 * When a packet is received, its address fields are first masked
230 * with the mask defined for the rule, then hashed, then matched
231 * against the entries in the corresponding list.
232 * Dynamic rules can be used for different purposes:
233 * + stateful rules;
234 * + enforcing limits on the number of sessions;
235 * + in-kernel NAT (not implemented yet)
236 *
237 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
238 * measured in seconds and depending on the flags.
239 *
240 * The total number of dynamic rules is stored in dyn_count.
241 * The max number of dynamic rules is dyn_max. When we reach
242 * the maximum number of rules we do not create anymore. This is
243 * done to avoid consuming too much memory, but also too much
244 * time when searching on each packet (ideally, we should try instead
245 * to put a limit on the length of the list on each bucket...).
246 *
247 * Each dynamic rule holds a pointer to the parent ipfw rule so
248 * we know what action to perform. Dynamic rules are removed when
249 * the parent rule is deleted. XXX we should make them survive.
250 *
251 * There are some limitations with dynamic rules -- we do not
252 * obey the 'randomized match', and we do not do multiple
253 * passes through the firewall. XXX check the latter!!!
254 */
255static ipfw_dyn_rule **ipfw_dyn_v = NULL;
256static u_int32_t dyn_buckets = 256; /* must be power of 2 */
257static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */
258
259static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */
260#define IPFW_DYN_LOCK_INIT() \
261 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF)
262#define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx)
263#define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx)
264#define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx)
265#define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED)
266
267/*
268 * Timeouts for various events in handing dynamic rules.
269 */
270static u_int32_t dyn_ack_lifetime = 300;
271static u_int32_t dyn_syn_lifetime = 20;
272static u_int32_t dyn_fin_lifetime = 1;
273static u_int32_t dyn_rst_lifetime = 1;
274static u_int32_t dyn_udp_lifetime = 10;
275static u_int32_t dyn_short_lifetime = 5;
276
277/*
278 * Keepalives are sent if dyn_keepalive is set. They are sent every
279 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
280 * seconds of lifetime of a rule.
281 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
282 * than dyn_keepalive_period.
283 */
284
285static u_int32_t dyn_keepalive_interval = 20;
286static u_int32_t dyn_keepalive_period = 5;
287static u_int32_t dyn_keepalive = 1; /* do send keepalives */
288
289static u_int32_t static_count; /* # of static rules */
290static u_int32_t static_len; /* size in bytes of static rules */
291static u_int32_t dyn_count; /* # of dynamic rules */
292static u_int32_t dyn_max = 4096; /* max # of dynamic rules */
293
294SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
295 &dyn_buckets, 0, "Number of dyn. buckets");
296SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
297 &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
298SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
299 &dyn_count, 0, "Number of dyn. rules");
300SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
301 &dyn_max, 0, "Max number of dyn. rules");
302SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
303 &static_count, 0, "Number of static rules");
304SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
305 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
306SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
307 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
308SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
309 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
310SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
311 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
312SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
313 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
314SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
315 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
316SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
317 &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
318
319#ifdef INET6
320/*
321 * IPv6 specific variables
322 */
323SYSCTL_DECL(_net_inet6_ip6);
324
325static struct sysctl_ctx_list ip6_fw_sysctl_ctx;
326static struct sysctl_oid *ip6_fw_sysctl_tree;
327#endif /* INET6 */
328#endif /* SYSCTL_NODE */
329
330static int fw_deny_unknown_exthdrs = 1;
331
332
333/*
334 * L3HDR maps an ipv4 pointer into a layer3 header pointer of type T
335 * Other macros just cast void * into the appropriate type
336 */
337#define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
338#define TCP(p) ((struct tcphdr *)(p))
339#define UDP(p) ((struct udphdr *)(p))
340#define ICMP(p) ((struct icmphdr *)(p))
341#define ICMP6(p) ((struct icmp6_hdr *)(p))
342
343static __inline int
344icmptype_match(struct icmphdr *icmp, ipfw_insn_u32 *cmd)
345{
346 int type = icmp->icmp_type;
347
348 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
349}
350
351#define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
352 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
353
354static int
355is_icmp_query(struct icmphdr *icmp)
356{
357 int type = icmp->icmp_type;
358
359 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
360}
361#undef TT
362
363/*
364 * The following checks use two arrays of 8 or 16 bits to store the
365 * bits that we want set or clear, respectively. They are in the
366 * low and high half of cmd->arg1 or cmd->d[0].
367 *
368 * We scan options and store the bits we find set. We succeed if
369 *
370 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
371 *
372 * The code is sometimes optimized not to store additional variables.
373 */
374
375static int
376flags_match(ipfw_insn *cmd, u_int8_t bits)
377{
378 u_char want_clear;
379 bits = ~bits;
380
381 if ( ((cmd->arg1 & 0xff) & bits) != 0)
382 return 0; /* some bits we want set were clear */
383 want_clear = (cmd->arg1 >> 8) & 0xff;
384 if ( (want_clear & bits) != want_clear)
385 return 0; /* some bits we want clear were set */
386 return 1;
387}
388
389static int
390ipopts_match(struct ip *ip, ipfw_insn *cmd)
391{
392 int optlen, bits = 0;
393 u_char *cp = (u_char *)(ip + 1);
394 int x = (ip->ip_hl << 2) - sizeof (struct ip);
395
396 for (; x > 0; x -= optlen, cp += optlen) {
397 int opt = cp[IPOPT_OPTVAL];
398
399 if (opt == IPOPT_EOL)
400 break;
401 if (opt == IPOPT_NOP)
402 optlen = 1;
403 else {
404 optlen = cp[IPOPT_OLEN];
405 if (optlen <= 0 || optlen > x)
406 return 0; /* invalid or truncated */
407 }
408 switch (opt) {
409
410 default:
411 break;
412
413 case IPOPT_LSRR:
414 bits |= IP_FW_IPOPT_LSRR;
415 break;
416
417 case IPOPT_SSRR:
418 bits |= IP_FW_IPOPT_SSRR;
419 break;
420
421 case IPOPT_RR:
422 bits |= IP_FW_IPOPT_RR;
423 break;
424
425 case IPOPT_TS:
426 bits |= IP_FW_IPOPT_TS;
427 break;
428 }
429 }
430 return (flags_match(cmd, bits));
431}
432
433static int
434tcpopts_match(struct tcphdr *tcp, ipfw_insn *cmd)
435{
436 int optlen, bits = 0;
437 u_char *cp = (u_char *)(tcp + 1);
438 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
439
440 for (; x > 0; x -= optlen, cp += optlen) {
441 int opt = cp[0];
442 if (opt == TCPOPT_EOL)
443 break;
444 if (opt == TCPOPT_NOP)
445 optlen = 1;
446 else {
447 optlen = cp[1];
448 if (optlen <= 0)
449 break;
450 }
451
452 switch (opt) {
453
454 default:
455 break;
456
457 case TCPOPT_MAXSEG:
458 bits |= IP_FW_TCPOPT_MSS;
459 break;
460
461 case TCPOPT_WINDOW:
462 bits |= IP_FW_TCPOPT_WINDOW;
463 break;
464
465 case TCPOPT_SACK_PERMITTED:
466 case TCPOPT_SACK:
467 bits |= IP_FW_TCPOPT_SACK;
468 break;
469
470 case TCPOPT_TIMESTAMP:
471 bits |= IP_FW_TCPOPT_TS;
472 break;
473
474 }
475 }
476 return (flags_match(cmd, bits));
477}
478
479static int
480iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
481{
482 if (ifp == NULL) /* no iface with this packet, match fails */
483 return 0;
484 /* Check by name or by IP address */
485 if (cmd->name[0] != '\0') { /* match by name */
486 /* Check name */
487 if (cmd->p.glob) {
488 if (fnmatch(cmd->name, ifp->if_xname, 0) == 0)
489 return(1);
490 } else {
491 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
492 return(1);
493 }
494 } else {
495 struct ifaddr *ia;
496
497 /* XXX lock? */
498 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
499 if (ia->ifa_addr == NULL)
500 continue;
501 if (ia->ifa_addr->sa_family != AF_INET)
502 continue;
503 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
504 (ia->ifa_addr))->sin_addr.s_addr)
505 return(1); /* match */
506 }
507 }
508 return(0); /* no match, fail ... */
509}
510
511/*
512 * The verify_path function checks if a route to the src exists and
513 * if it is reachable via ifp (when provided).
514 *
515 * The 'verrevpath' option checks that the interface that an IP packet
516 * arrives on is the same interface that traffic destined for the
517 * packet's source address would be routed out of. The 'versrcreach'
518 * option just checks that the source address is reachable via any route
519 * (except default) in the routing table. These two are a measure to block
520 * forged packets. This is also commonly known as "anti-spoofing" or Unicast
521 * Reverse Path Forwarding (Unicast RFP) in Cisco-ese. The name of the knobs
522 * is purposely reminiscent of the Cisco IOS command,
523 *
524 * ip verify unicast reverse-path
525 * ip verify unicast source reachable-via any
526 *
527 * which implements the same functionality. But note that syntax is
528 * misleading. The check may be performed on all IP packets whether unicast,
529 * multicast, or broadcast.
530 */
531static int
532verify_path(struct in_addr src, struct ifnet *ifp)
533{
534 struct route ro;
535 struct sockaddr_in *dst;
536
537 bzero(&ro, sizeof(ro));
538
539 dst = (struct sockaddr_in *)&(ro.ro_dst);
540 dst->sin_family = AF_INET;
541 dst->sin_len = sizeof(*dst);
542 dst->sin_addr = src;
543 rtalloc_ign(&ro, RTF_CLONING);
544
545 if (ro.ro_rt == NULL)
546 return 0;
547
548 /* if ifp is provided, check for equality with rtentry */
549 if (ifp != NULL && ro.ro_rt->rt_ifp != ifp) {
550 RTFREE(ro.ro_rt);
551 return 0;
552 }
553
554 /* if no ifp provided, check if rtentry is not default route */
555 if (ifp == NULL &&
556 satosin(rt_key(ro.ro_rt))->sin_addr.s_addr == INADDR_ANY) {
557 RTFREE(ro.ro_rt);
558 return 0;
559 }
560
561 /* or if this is a blackhole/reject route */
562 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
563 RTFREE(ro.ro_rt);
564 return 0;
565 }
566
567 /* found valid route */
568 RTFREE(ro.ro_rt);
569 return 1;
570}
571
572#ifdef INET6
573/*
574 * ipv6 specific rules here...
575 */
576static __inline int
577icmp6type_match (int type, ipfw_insn_u32 *cmd)
578{
579 return (type <= ICMP6_MAXTYPE && (cmd->d[type/32] & (1<<(type%32)) ) );
580}
581
582static int
583flow6id_match( int curr_flow, ipfw_insn_u32 *cmd )
584{
585 int i;
586 for (i=0; i <= cmd->o.arg1; ++i )
587 if (curr_flow == cmd->d[i] )
588 return 1;
589 return 0;
590}
591
592/* support for IP6_*_ME opcodes */
593static int
594search_ip6_addr_net (struct in6_addr * ip6_addr)
595{
596 struct ifnet *mdc;
597 struct ifaddr *mdc2;
598 struct in6_ifaddr *fdm;
599 struct in6_addr copia;
600
601 TAILQ_FOREACH(mdc, &ifnet, if_link)
602 for (mdc2 = mdc->if_addrlist.tqh_first; mdc2;
603 mdc2 = mdc2->ifa_list.tqe_next) {
604 if (!mdc2->ifa_addr)
605 continue;
606 if (mdc2->ifa_addr->sa_family == AF_INET6) {
607 fdm = (struct in6_ifaddr *)mdc2;
608 copia = fdm->ia_addr.sin6_addr;
609 /* need for leaving scope_id in the sock_addr */
610 in6_clearscope(&copia);
611 if (IN6_ARE_ADDR_EQUAL(ip6_addr, &copia))
612 return 1;
613 }
614 }
615 return 0;
616}
617
618static int
619verify_path6(struct in6_addr *src, struct ifnet *ifp)
620{
621 struct route_in6 ro;
622 struct sockaddr_in6 *dst;
623
624 bzero(&ro, sizeof(ro));
625
626 dst = (struct sockaddr_in6 * )&(ro.ro_dst);
627 dst->sin6_family = AF_INET6;
628 dst->sin6_len = sizeof(*dst);
629 dst->sin6_addr = *src;
630 rtalloc_ign((struct route *)&ro, RTF_CLONING);
631
632 if (ro.ro_rt == NULL)
633 return 0;
634
635 /*
636 * if ifp is provided, check for equality with rtentry
637 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
638 * to support the case of sending packets to an address of our own.
639 * (where the former interface is the first argument of if_simloop()
640 * (=ifp), the latter is lo0)
641 */
642 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
643 RTFREE(ro.ro_rt);
644 return 0;
645 }
646
647 /* if no ifp provided, check if rtentry is not default route */
648 if (ifp == NULL &&
649 IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(ro.ro_rt))->sin6_addr)) {
650 RTFREE(ro.ro_rt);
651 return 0;
652 }
653
654 /* or if this is a blackhole/reject route */
655 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
656 RTFREE(ro.ro_rt);
657 return 0;
658 }
659
660 /* found valid route */
661 RTFREE(ro.ro_rt);
662 return 1;
663
664}
665static __inline int
666hash_packet6(struct ipfw_flow_id *id)
667{
668 u_int32_t i;
669 i = (id->dst_ip6.__u6_addr.__u6_addr32[0]) ^
670 (id->dst_ip6.__u6_addr.__u6_addr32[1]) ^
671 (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^
672 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^
673 (id->dst_port) ^ (id->src_port) ^ (id->flow_id6);
674 return i;
675}
676
677static int
678is_icmp6_query(int icmp6_type)
679{
680 if ((icmp6_type <= ICMP6_MAXTYPE) &&
681 (icmp6_type == ICMP6_ECHO_REQUEST ||
682 icmp6_type == ICMP6_MEMBERSHIP_QUERY ||
683 icmp6_type == ICMP6_WRUREQUEST ||
684 icmp6_type == ICMP6_FQDN_QUERY ||
685 icmp6_type == ICMP6_NI_QUERY))
686 return (1);
687
688 return (0);
689}
690
691static void
692send_reject6(struct ip_fw_args *args, int code, u_short offset, u_int hlen)
693{
694 if (code == ICMP6_UNREACH_RST && offset == 0 &&
695 args->f_id.proto == IPPROTO_TCP) {
696 struct ip6_hdr *ip6;
697 struct tcphdr *tcp;
698 tcp_seq ack, seq;
699 int flags;
700 struct {
701 struct ip6_hdr ip6;
702 struct tcphdr th;
703 } ti;
704
705 if (args->m->m_len < (hlen+sizeof(struct tcphdr))) {
706 args->m = m_pullup(args->m, hlen+sizeof(struct tcphdr));
707 if (args->m == NULL)
708 return;
709 }
710
711 ip6 = mtod(args->m, struct ip6_hdr *);
712 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
713
714 if ((tcp->th_flags & TH_RST) != 0) {
715 m_freem(args->m);
716 return;
717 }
718
719 ti.ip6 = *ip6;
720 ti.th = *tcp;
721 ti.th.th_seq = ntohl(ti.th.th_seq);
722 ti.th.th_ack = ntohl(ti.th.th_ack);
723 ti.ip6.ip6_nxt = IPPROTO_TCP;
724
725 if (ti.th.th_flags & TH_ACK) {
726 ack = 0;
727 seq = ti.th.th_ack;
728 flags = TH_RST;
729 } else {
730 ack = ti.th.th_seq;
731 if (((args->m)->m_flags & M_PKTHDR) != 0) {
732 ack += (args->m)->m_pkthdr.len - hlen
733 - (ti.th.th_off << 2);
734 } else if (ip6->ip6_plen) {
735 ack += ntohs(ip6->ip6_plen) + sizeof(*ip6)
736 - hlen - (ti.th.th_off << 2);
737 } else {
738 m_freem(args->m);
739 return;
740 }
741 if (tcp->th_flags & TH_SYN)
742 ack++;
743 seq = 0;
744 flags = TH_RST|TH_ACK;
745 }
746 bcopy(&ti, ip6, sizeof(ti));
747 tcp_respond(NULL, ip6, (struct tcphdr *)(ip6 + 1),
748 args->m, ack, seq, flags);
749
750 } else if (code != ICMP6_UNREACH_RST) { /* Send an ICMPv6 unreach. */
751 icmp6_error(args->m, ICMP6_DST_UNREACH, code, 0);
752
753 } else
754 m_freem(args->m);
755
756 args->m = NULL;
757}
758
759#endif /* INET6 */
760
761static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */
762
763#define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
764#define SNP(buf) buf, sizeof(buf)
765
766/*
767 * We enter here when we have a rule with O_LOG.
768 * XXX this function alone takes about 2Kbytes of code!
769 */
770static void
771ipfw_log(struct ip_fw *f, u_int hlen, struct ip_fw_args *args,
772 struct mbuf *m, struct ifnet *oif, u_short offset)
773{
774 struct ether_header *eh = args->eh;
775 char *action;
776 int limit_reached = 0;
777 char action2[40], proto[128], fragment[32];
778
779 fragment[0] = '\0';
780 proto[0] = '\0';
781
782 if (f == NULL) { /* bogus pkt */
783 if (verbose_limit != 0 && norule_counter >= verbose_limit)
784 return;
785 norule_counter++;
786 if (norule_counter == verbose_limit)
787 limit_reached = verbose_limit;
788 action = "Refuse";
789 } else { /* O_LOG is the first action, find the real one */
790 ipfw_insn *cmd = ACTION_PTR(f);
791 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
792
793 if (l->max_log != 0 && l->log_left == 0)
794 return;
795 l->log_left--;
796 if (l->log_left == 0)
797 limit_reached = l->max_log;
798 cmd += F_LEN(cmd); /* point to first action */
799 if (cmd->opcode == O_ALTQ) {
800 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
801
802 snprintf(SNPARGS(action2, 0), "Altq %d",
803 altq->qid);
804 cmd += F_LEN(cmd);
805 }
806 if (cmd->opcode == O_PROB)
807 cmd += F_LEN(cmd);
808
809 action = action2;
810 switch (cmd->opcode) {
811 case O_DENY:
812 action = "Deny";
813 break;
814
815 case O_REJECT:
816 if (cmd->arg1==ICMP_REJECT_RST)
817 action = "Reset";
818 else if (cmd->arg1==ICMP_UNREACH_HOST)
819 action = "Reject";
820 else
821 snprintf(SNPARGS(action2, 0), "Unreach %d",
822 cmd->arg1);
823 break;
824
825 case O_UNREACH6:
826 if (cmd->arg1==ICMP6_UNREACH_RST)
827 action = "Reset";
828 else
829 snprintf(SNPARGS(action2, 0), "Unreach %d",
830 cmd->arg1);
831 break;
832
833 case O_ACCEPT:
834 action = "Accept";
835 break;
836 case O_COUNT:
837 action = "Count";
838 break;
839 case O_DIVERT:
840 snprintf(SNPARGS(action2, 0), "Divert %d",
841 cmd->arg1);
842 break;
843 case O_TEE:
844 snprintf(SNPARGS(action2, 0), "Tee %d",
845 cmd->arg1);
846 break;
847 case O_SKIPTO:
848 snprintf(SNPARGS(action2, 0), "SkipTo %d",
849 cmd->arg1);
850 break;
851 case O_PIPE:
852 snprintf(SNPARGS(action2, 0), "Pipe %d",
853 cmd->arg1);
854 break;
855 case O_QUEUE:
856 snprintf(SNPARGS(action2, 0), "Queue %d",
857 cmd->arg1);
858 break;
859 case O_FORWARD_IP: {
860 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
861 int len;
862
863 len = snprintf(SNPARGS(action2, 0), "Forward to %s",
864 inet_ntoa(sa->sa.sin_addr));
865 if (sa->sa.sin_port)
866 snprintf(SNPARGS(action2, len), ":%d",
867 sa->sa.sin_port);
868 }
869 break;
870 case O_NETGRAPH:
871 snprintf(SNPARGS(action2, 0), "Netgraph %d",
872 cmd->arg1);
873 break;
874 case O_NGTEE:
875 snprintf(SNPARGS(action2, 0), "Ngtee %d",
876 cmd->arg1);
877 break;
878 default:
879 action = "UNKNOWN";
880 break;
881 }
882 }
883
884 if (hlen == 0) { /* non-ip */
885 snprintf(SNPARGS(proto, 0), "MAC");
886
887 } else {
888 int len;
889 char src[48], dst[48];
890 struct icmphdr *icmp;
891 struct tcphdr *tcp;
892 struct udphdr *udp;
893 /* Initialize to make compiler happy. */
894 struct ip *ip = NULL;
895#ifdef INET6
896 struct ip6_hdr *ip6 = NULL;
897 struct icmp6_hdr *icmp6;
898#endif
899 src[0] = '\0';
900 dst[0] = '\0';
901#ifdef INET6
902 if (args->f_id.addr_type == 6) {
903 snprintf(src, sizeof(src), "[%s]",
904 ip6_sprintf(&args->f_id.src_ip6));
905 snprintf(dst, sizeof(dst), "[%s]",
906 ip6_sprintf(&args->f_id.dst_ip6));
907
908 ip6 = (struct ip6_hdr *)mtod(m, struct ip6_hdr *);
909 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
910 udp = (struct udphdr *)(mtod(args->m, char *) + hlen);
911 } else
912#endif
913 {
914 ip = mtod(m, struct ip *);
915 tcp = L3HDR(struct tcphdr, ip);
916 udp = L3HDR(struct udphdr, ip);
917
918 inet_ntoa_r(ip->ip_src, src);
919 inet_ntoa_r(ip->ip_dst, dst);
920 }
921
922 switch (args->f_id.proto) {
923 case IPPROTO_TCP:
924 len = snprintf(SNPARGS(proto, 0), "TCP %s", src);
925 if (offset == 0)
926 snprintf(SNPARGS(proto, len), ":%d %s:%d",
927 ntohs(tcp->th_sport),
928 dst,
929 ntohs(tcp->th_dport));
930 else
931 snprintf(SNPARGS(proto, len), " %s", dst);
932 break;
933
934 case IPPROTO_UDP:
935 len = snprintf(SNPARGS(proto, 0), "UDP %s", src);
936 if (offset == 0)
937 snprintf(SNPARGS(proto, len), ":%d %s:%d",
938 ntohs(udp->uh_sport),
939 dst,
940 ntohs(udp->uh_dport));
941 else
942 snprintf(SNPARGS(proto, len), " %s", dst);
943 break;
944
945 case IPPROTO_ICMP:
946 icmp = L3HDR(struct icmphdr, ip);
947 if (offset == 0)
948 len = snprintf(SNPARGS(proto, 0),
949 "ICMP:%u.%u ",
950 icmp->icmp_type, icmp->icmp_code);
951 else
952 len = snprintf(SNPARGS(proto, 0), "ICMP ");
953 len += snprintf(SNPARGS(proto, len), "%s", src);
954 snprintf(SNPARGS(proto, len), " %s", dst);
955 break;
956#ifdef INET6
957 case IPPROTO_ICMPV6:
958 icmp6 = (struct icmp6_hdr *)(mtod(args->m, char *) + hlen);
959 if (offset == 0)
960 len = snprintf(SNPARGS(proto, 0),
961 "ICMPv6:%u.%u ",
962 icmp6->icmp6_type, icmp6->icmp6_code);
963 else
964 len = snprintf(SNPARGS(proto, 0), "ICMPv6 ");
965 len += snprintf(SNPARGS(proto, len), "%s", src);
966 snprintf(SNPARGS(proto, len), " %s", dst);
967 break;
968#endif
969 default:
970 len = snprintf(SNPARGS(proto, 0), "P:%d %s",
971 args->f_id.proto, src);
972 snprintf(SNPARGS(proto, len), " %s", dst);
973 break;
974 }
975
976#ifdef INET6
977 if (args->f_id.addr_type == 6) {
978 if (offset & (IP6F_OFF_MASK | IP6F_MORE_FRAG))
979 snprintf(SNPARGS(fragment, 0),
980 " (frag %08x:%d@%d%s)",
981 args->f_id.frag_id6,
982 ntohs(ip6->ip6_plen) - hlen,
983 ntohs(offset & IP6F_OFF_MASK) << 3,
984 (offset & IP6F_MORE_FRAG) ? "+" : "");
985 } else
986#endif
987 {
988 int ip_off, ip_len;
989 if (eh != NULL) { /* layer 2 packets are as on the wire */
990 ip_off = ntohs(ip->ip_off);
991 ip_len = ntohs(ip->ip_len);
992 } else {
993 ip_off = ip->ip_off;
994 ip_len = ip->ip_len;
995 }
996 if (ip_off & (IP_MF | IP_OFFMASK))
997 snprintf(SNPARGS(fragment, 0),
998 " (frag %d:%d@%d%s)",
999 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
1000 offset << 3,
1001 (ip_off & IP_MF) ? "+" : "");
1002 }
1003 }
1004 if (oif || m->m_pkthdr.rcvif)
1005 log(LOG_SECURITY | LOG_INFO,
1006 "ipfw: %d %s %s %s via %s%s\n",
1007 f ? f->rulenum : -1,
1008 action, proto, oif ? "out" : "in",
1009 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
1010 fragment);
1011 else
1012 log(LOG_SECURITY | LOG_INFO,
1013 "ipfw: %d %s %s [no if info]%s\n",
1014 f ? f->rulenum : -1,
1015 action, proto, fragment);
1016 if (limit_reached)
1017 log(LOG_SECURITY | LOG_NOTICE,
1018 "ipfw: limit %d reached on entry %d\n",
1019 limit_reached, f ? f->rulenum : -1);
1020}
1021
1022/*
1023 * IMPORTANT: the hash function for dynamic rules must be commutative
1024 * in source and destination (ip,port), because rules are bidirectional
1025 * and we want to find both in the same bucket.
1026 */
1027static __inline int
1028hash_packet(struct ipfw_flow_id *id)
1029{
1030 u_int32_t i;
1031
1032#ifdef INET6
1033 if (IS_IP6_FLOW_ID(id))
1034 i = hash_packet6(id);
1035 else
1036#endif /* INET6 */
1037 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
1038 i &= (curr_dyn_buckets - 1);
1039 return i;
1040}
1041
1042/**
1043 * unlink a dynamic rule from a chain. prev is a pointer to
1044 * the previous one, q is a pointer to the rule to delete,
1045 * head is a pointer to the head of the queue.
1046 * Modifies q and potentially also head.
1047 */
1048#define UNLINK_DYN_RULE(prev, head, q) { \
1049 ipfw_dyn_rule *old_q = q; \
1050 \
1051 /* remove a refcount to the parent */ \
1052 if (q->dyn_type == O_LIMIT) \
1053 q->parent->count--; \
1054 DEB(printf("ipfw: unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",\
1055 (q->id.src_ip), (q->id.src_port), \
1056 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \
1057 if (prev != NULL) \
1058 prev->next = q = q->next; \
1059 else \
1060 head = q = q->next; \
1061 dyn_count--; \
1062 uma_zfree(ipfw_dyn_rule_zone, old_q); }
1063
1064#define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
1065
1066/**
1067 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
1068 *
1069 * If keep_me == NULL, rules are deleted even if not expired,
1070 * otherwise only expired rules are removed.
1071 *
1072 * The value of the second parameter is also used to point to identify
1073 * a rule we absolutely do not want to remove (e.g. because we are
1074 * holding a reference to it -- this is the case with O_LIMIT_PARENT
1075 * rules). The pointer is only used for comparison, so any non-null
1076 * value will do.
1077 */
1078static void
1079remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
1080{
1081 static u_int32_t last_remove = 0;
1082
1083#define FORCE (keep_me == NULL)
1084
1085 ipfw_dyn_rule *prev, *q;
1086 int i, pass = 0, max_pass = 0;
1087
1088 IPFW_DYN_LOCK_ASSERT();
1089
1090 if (ipfw_dyn_v == NULL || dyn_count == 0)
1091 return;
1092 /* do not expire more than once per second, it is useless */
1093 if (!FORCE && last_remove == time_uptime)
1094 return;
1095 last_remove = time_uptime;
1096
1097 /*
1098 * because O_LIMIT refer to parent rules, during the first pass only
1099 * remove child and mark any pending LIMIT_PARENT, and remove
1100 * them in a second pass.
1101 */
1102next_pass:
1103 for (i = 0 ; i < curr_dyn_buckets ; i++) {
1104 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
1105 /*
1106 * Logic can become complex here, so we split tests.
1107 */
1108 if (q == keep_me)
1109 goto next;
1110 if (rule != NULL && rule != q->rule)
1111 goto next; /* not the one we are looking for */
1112 if (q->dyn_type == O_LIMIT_PARENT) {
1113 /*
1114 * handle parent in the second pass,
1115 * record we need one.
1116 */
1117 max_pass = 1;
1118 if (pass == 0)
1119 goto next;
1120 if (FORCE && q->count != 0 ) {
1121 /* XXX should not happen! */
1122 printf("ipfw: OUCH! cannot remove rule,"
1123 " count %d\n", q->count);
1124 }
1125 } else {
1126 if (!FORCE &&
1127 !TIME_LEQ( q->expire, time_uptime ))
1128 goto next;
1129 }
1130 if (q->dyn_type != O_LIMIT_PARENT || !q->count) {
1131 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1132 continue;
1133 }
1134next:
1135 prev=q;
1136 q=q->next;
1137 }
1138 }
1139 if (pass++ < max_pass)
1140 goto next_pass;
1141}
1142
1143
1144/**
1145 * lookup a dynamic rule.
1146 */
1147static ipfw_dyn_rule *
1148lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction,
1149 struct tcphdr *tcp)
1150{
1151 /*
1152 * stateful ipfw extensions.
1153 * Lookup into dynamic session queue
1154 */
1155#define MATCH_REVERSE 0
1156#define MATCH_FORWARD 1
1157#define MATCH_NONE 2
1158#define MATCH_UNKNOWN 3
1159 int i, dir = MATCH_NONE;
1160 ipfw_dyn_rule *prev, *q=NULL;
1161
1162 IPFW_DYN_LOCK_ASSERT();
1163
1164 if (ipfw_dyn_v == NULL)
1165 goto done; /* not found */
1166 i = hash_packet( pkt );
1167 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
1168 if (q->dyn_type == O_LIMIT_PARENT && q->count)
1169 goto next;
1170 if (TIME_LEQ( q->expire, time_uptime)) { /* expire entry */
1171 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1172 continue;
1173 }
1174 if (pkt->proto == q->id.proto &&
1175 q->dyn_type != O_LIMIT_PARENT) {
1176 if (IS_IP6_FLOW_ID(pkt)) {
1177 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1178 &(q->id.src_ip6)) &&
1179 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1180 &(q->id.dst_ip6)) &&
1181 pkt->src_port == q->id.src_port &&
1182 pkt->dst_port == q->id.dst_port ) {
1183 dir = MATCH_FORWARD;
1184 break;
1185 }
1186 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1187 &(q->id.dst_ip6)) &&
1188 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1189 &(q->id.src_ip6)) &&
1190 pkt->src_port == q->id.dst_port &&
1191 pkt->dst_port == q->id.src_port ) {
1192 dir = MATCH_REVERSE;
1193 break;
1194 }
1195 } else {
1196 if (pkt->src_ip == q->id.src_ip &&
1197 pkt->dst_ip == q->id.dst_ip &&
1198 pkt->src_port == q->id.src_port &&
1199 pkt->dst_port == q->id.dst_port ) {
1200 dir = MATCH_FORWARD;
1201 break;
1202 }
1203 if (pkt->src_ip == q->id.dst_ip &&
1204 pkt->dst_ip == q->id.src_ip &&
1205 pkt->src_port == q->id.dst_port &&
1206 pkt->dst_port == q->id.src_port ) {
1207 dir = MATCH_REVERSE;
1208 break;
1209 }
1210 }
1211 }
1212next:
1213 prev = q;
1214 q = q->next;
1215 }
1216 if (q == NULL)
1217 goto done; /* q = NULL, not found */
1218
1219 if ( prev != NULL) { /* found and not in front */
1220 prev->next = q->next;
1221 q->next = ipfw_dyn_v[i];
1222 ipfw_dyn_v[i] = q;
1223 }
1224 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
1225 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
1226
1227#define BOTH_SYN (TH_SYN | (TH_SYN << 8))
1228#define BOTH_FIN (TH_FIN | (TH_FIN << 8))
1229 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
1230 switch (q->state) {
1231 case TH_SYN: /* opening */
1232 q->expire = time_uptime + dyn_syn_lifetime;
1233 break;
1234
1235 case BOTH_SYN: /* move to established */
1236 case BOTH_SYN | TH_FIN : /* one side tries to close */
1237 case BOTH_SYN | (TH_FIN << 8) :
1238 if (tcp) {
1239#define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
1240 u_int32_t ack = ntohl(tcp->th_ack);
1241 if (dir == MATCH_FORWARD) {
1242 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
1243 q->ack_fwd = ack;
1244 else { /* ignore out-of-sequence */
1245 break;
1246 }
1247 } else {
1248 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
1249 q->ack_rev = ack;
1250 else { /* ignore out-of-sequence */
1251 break;
1252 }
1253 }
1254 }
1255 q->expire = time_uptime + dyn_ack_lifetime;
1256 break;
1257
1258 case BOTH_SYN | BOTH_FIN: /* both sides closed */
1259 if (dyn_fin_lifetime >= dyn_keepalive_period)
1260 dyn_fin_lifetime = dyn_keepalive_period - 1;
1261 q->expire = time_uptime + dyn_fin_lifetime;
1262 break;
1263
1264 default:
1265#if 0
1266 /*
1267 * reset or some invalid combination, but can also
1268 * occur if we use keep-state the wrong way.
1269 */
1270 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
1271 printf("invalid state: 0x%x\n", q->state);
1272#endif
1273 if (dyn_rst_lifetime >= dyn_keepalive_period)
1274 dyn_rst_lifetime = dyn_keepalive_period - 1;
1275 q->expire = time_uptime + dyn_rst_lifetime;
1276 break;
1277 }
1278 } else if (pkt->proto == IPPROTO_UDP) {
1279 q->expire = time_uptime + dyn_udp_lifetime;
1280 } else {
1281 /* other protocols */
1282 q->expire = time_uptime + dyn_short_lifetime;
1283 }
1284done:
1285 if (match_direction)
1286 *match_direction = dir;
1287 return q;
1288}
1289
1290static ipfw_dyn_rule *
1291lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
1292 struct tcphdr *tcp)
1293{
1294 ipfw_dyn_rule *q;
1295
1296 IPFW_DYN_LOCK();
1297 q = lookup_dyn_rule_locked(pkt, match_direction, tcp);
1298 if (q == NULL)
1299 IPFW_DYN_UNLOCK();
1300 /* NB: return table locked when q is not NULL */
1301 return q;
1302}
1303
1304static void
1305realloc_dynamic_table(void)
1306{
1307 IPFW_DYN_LOCK_ASSERT();
1308
1309 /*
1310 * Try reallocation, make sure we have a power of 2 and do
1311 * not allow more than 64k entries. In case of overflow,
1312 * default to 1024.
1313 */
1314
1315 if (dyn_buckets > 65536)
1316 dyn_buckets = 1024;
1317 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
1318 dyn_buckets = curr_dyn_buckets; /* reset */
1319 return;
1320 }
1321 curr_dyn_buckets = dyn_buckets;
1322 if (ipfw_dyn_v != NULL)
1323 free(ipfw_dyn_v, M_IPFW);
1324 for (;;) {
1325 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
1326 M_IPFW, M_NOWAIT | M_ZERO);
1327 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
1328 break;
1329 curr_dyn_buckets /= 2;
1330 }
1331}
1332
1333/**
1334 * Install state of type 'type' for a dynamic session.
1335 * The hash table contains two type of rules:
1336 * - regular rules (O_KEEP_STATE)
1337 * - rules for sessions with limited number of sess per user
1338 * (O_LIMIT). When they are created, the parent is
1339 * increased by 1, and decreased on delete. In this case,
1340 * the third parameter is the parent rule and not the chain.
1341 * - "parent" rules for the above (O_LIMIT_PARENT).
1342 */
1343static ipfw_dyn_rule *
1344add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
1345{
1346 ipfw_dyn_rule *r;
1347 int i;
1348
1349 IPFW_DYN_LOCK_ASSERT();
1350
1351 if (ipfw_dyn_v == NULL ||
1352 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1353 realloc_dynamic_table();
1354 if (ipfw_dyn_v == NULL)
1355 return NULL; /* failed ! */
1356 }
1357 i = hash_packet(id);
1358
1359 r = uma_zalloc(ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO);
1360 if (r == NULL) {
1361 printf ("ipfw: sorry cannot allocate state\n");
1362 return NULL;
1363 }
1364
1365 /* increase refcount on parent, and set pointer */
1366 if (dyn_type == O_LIMIT) {
1367 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1368 if ( parent->dyn_type != O_LIMIT_PARENT)
1369 panic("invalid parent");
1370 parent->count++;
1371 r->parent = parent;
1372 rule = parent->rule;
1373 }
1374
1375 r->id = *id;
1376 r->expire = time_uptime + dyn_syn_lifetime;
1377 r->rule = rule;
1378 r->dyn_type = dyn_type;
1379 r->pcnt = r->bcnt = 0;
1380 r->count = 0;
1381
1382 r->bucket = i;
1383 r->next = ipfw_dyn_v[i];
1384 ipfw_dyn_v[i] = r;
1385 dyn_count++;
1386 DEB(printf("ipfw: add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1387 dyn_type,
1388 (r->id.src_ip), (r->id.src_port),
1389 (r->id.dst_ip), (r->id.dst_port),
1390 dyn_count ); )
1391 return r;
1392}
1393
1394/**
1395 * lookup dynamic parent rule using pkt and rule as search keys.
1396 * If the lookup fails, then install one.
1397 */
1398static ipfw_dyn_rule *
1399lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1400{
1401 ipfw_dyn_rule *q;
1402 int i;
1403
1404 IPFW_DYN_LOCK_ASSERT();
1405
1406 if (ipfw_dyn_v) {
1407 int is_v6 = IS_IP6_FLOW_ID(pkt);
1408 i = hash_packet( pkt );
1409 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
1410 if (q->dyn_type == O_LIMIT_PARENT &&
1411 rule== q->rule &&
1412 pkt->proto == q->id.proto &&
1413 pkt->src_port == q->id.src_port &&
1414 pkt->dst_port == q->id.dst_port &&
1415 (
1416 (is_v6 &&
1417 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1418 &(q->id.src_ip6)) &&
1419 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1420 &(q->id.dst_ip6))) ||
1421 (!is_v6 &&
1422 pkt->src_ip == q->id.src_ip &&
1423 pkt->dst_ip == q->id.dst_ip)
1424 )
1425 ) {
1426 q->expire = time_uptime + dyn_short_lifetime;
1427 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);)
1428 return q;
1429 }
1430 }
1431 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1432}
1433
1434/**
1435 * Install dynamic state for rule type cmd->o.opcode
1436 *
1437 * Returns 1 (failure) if state is not installed because of errors or because
1438 * session limitations are enforced.
1439 */
1440static int
1441install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1442 struct ip_fw_args *args)
1443{
1444 static int last_log;
1445
1446 ipfw_dyn_rule *q;
1447
1448 DEB(printf("ipfw: install state type %d 0x%08x %u -> 0x%08x %u\n",
1449 cmd->o.opcode,
1450 (args->f_id.src_ip), (args->f_id.src_port),
1451 (args->f_id.dst_ip), (args->f_id.dst_port) );)
1452
1453 IPFW_DYN_LOCK();
1454
1455 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
1456
1457 if (q != NULL) { /* should never occur */
1458 if (last_log != time_uptime) {
1459 last_log = time_uptime;
1460 printf("ipfw: install_state: entry already present, done\n");
1461 }
1462 IPFW_DYN_UNLOCK();
1463 return 0;
1464 }
1465
1466 if (dyn_count >= dyn_max)
1467 /*
1468 * Run out of slots, try to remove any expired rule.
1469 */
1470 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1471
1472 if (dyn_count >= dyn_max) {
1473 if (last_log != time_uptime) {
1474 last_log = time_uptime;
1475 printf("ipfw: install_state: Too many dynamic rules\n");
1476 }
1477 IPFW_DYN_UNLOCK();
1478 return 1; /* cannot install, notify caller */
1479 }
1480
1481 switch (cmd->o.opcode) {
1482 case O_KEEP_STATE: /* bidir rule */
1483 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1484 break;
1485
1486 case O_LIMIT: /* limit number of sessions */
1487 {
1488 u_int16_t limit_mask = cmd->limit_mask;
1489 struct ipfw_flow_id id;
1490 ipfw_dyn_rule *parent;
1491
1492 DEB(printf("ipfw: installing dyn-limit rule %d\n",
1493 cmd->conn_limit);)
1494
1495 id.dst_ip = id.src_ip = 0;
1496 id.dst_port = id.src_port = 0;
1497 id.proto = args->f_id.proto;
1498
1499 if (IS_IP6_FLOW_ID (&(args->f_id))) {
1500 if (limit_mask & DYN_SRC_ADDR)
1501 id.src_ip6 = args->f_id.src_ip6;
1502 if (limit_mask & DYN_DST_ADDR)
1503 id.dst_ip6 = args->f_id.dst_ip6;
1504 } else {
1505 if (limit_mask & DYN_SRC_ADDR)
1506 id.src_ip = args->f_id.src_ip;
1507 if (limit_mask & DYN_DST_ADDR)
1508 id.dst_ip = args->f_id.dst_ip;
1509 }
1510 if (limit_mask & DYN_SRC_PORT)
1511 id.src_port = args->f_id.src_port;
1512 if (limit_mask & DYN_DST_PORT)
1513 id.dst_port = args->f_id.dst_port;
1514 parent = lookup_dyn_parent(&id, rule);
1515 if (parent == NULL) {
1516 printf("ipfw: add parent failed\n");
1517 IPFW_DYN_UNLOCK();
1518 return 1;
1519 }
1520 if (parent->count >= cmd->conn_limit) {
1521 /*
1522 * See if we can remove some expired rule.
1523 */
1524 remove_dyn_rule(rule, parent);
1525 if (parent->count >= cmd->conn_limit) {
1526 if (fw_verbose && last_log != time_uptime) {
1527 last_log = time_uptime;
1528 log(LOG_SECURITY | LOG_DEBUG,
1529 "drop session, too many entries\n");
1530 }
1531 IPFW_DYN_UNLOCK();
1532 return 1;
1533 }
1534 }
1535 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1536 }
1537 break;
1538 default:
1539 printf("ipfw: unknown dynamic rule type %u\n", cmd->o.opcode);
1540 IPFW_DYN_UNLOCK();
1541 return 1;
1542 }
1543 lookup_dyn_rule_locked(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1544 IPFW_DYN_UNLOCK();
1545 return 0;
1546}
1547
1548/*
1549 * Generate a TCP packet, containing either a RST or a keepalive.
1550 * When flags & TH_RST, we are sending a RST packet, because of a
1551 * "reset" action matched the packet.
1552 * Otherwise we are sending a keepalive, and flags & TH_
1553 */
1554static struct mbuf *
1555send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags)
1556{
1557 struct mbuf *m;
1558 struct ip *ip;
1559 struct tcphdr *tcp;
1560
1561 MGETHDR(m, M_DONTWAIT, MT_DATA);
1562 if (m == 0)
1563 return (NULL);
1564 m->m_pkthdr.rcvif = (struct ifnet *)0;
1565 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1566 m->m_data += max_linkhdr;
1567
1568 ip = mtod(m, struct ip *);
1569 bzero(ip, m->m_len);
1570 tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1571 ip->ip_p = IPPROTO_TCP;
1572 tcp->th_off = 5;
1573 /*
1574 * Assume we are sending a RST (or a keepalive in the reverse
1575 * direction), swap src and destination addresses and ports.
1576 */
1577 ip->ip_src.s_addr = htonl(id->dst_ip);
1578 ip->ip_dst.s_addr = htonl(id->src_ip);
1579 tcp->th_sport = htons(id->dst_port);
1580 tcp->th_dport = htons(id->src_port);
1581 if (flags & TH_RST) { /* we are sending a RST */
1582 if (flags & TH_ACK) {
1583 tcp->th_seq = htonl(ack);
1584 tcp->th_ack = htonl(0);
1585 tcp->th_flags = TH_RST;
1586 } else {
1587 if (flags & TH_SYN)
1588 seq++;
1589 tcp->th_seq = htonl(0);
1590 tcp->th_ack = htonl(seq);
1591 tcp->th_flags = TH_RST | TH_ACK;
1592 }
1593 } else {
1594 /*
1595 * We are sending a keepalive. flags & TH_SYN determines
1596 * the direction, forward if set, reverse if clear.
1597 * NOTE: seq and ack are always assumed to be correct
1598 * as set by the caller. This may be confusing...
1599 */
1600 if (flags & TH_SYN) {
1601 /*
1602 * we have to rewrite the correct addresses!
1603 */
1604 ip->ip_dst.s_addr = htonl(id->dst_ip);
1605 ip->ip_src.s_addr = htonl(id->src_ip);
1606 tcp->th_dport = htons(id->dst_port);
1607 tcp->th_sport = htons(id->src_port);
1608 }
1609 tcp->th_seq = htonl(seq);
1610 tcp->th_ack = htonl(ack);
1611 tcp->th_flags = TH_ACK;
1612 }
1613 /*
1614 * set ip_len to the payload size so we can compute
1615 * the tcp checksum on the pseudoheader
1616 * XXX check this, could save a couple of words ?
1617 */
1618 ip->ip_len = htons(sizeof(struct tcphdr));
1619 tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1620 /*
1621 * now fill fields left out earlier
1622 */
1623 ip->ip_ttl = ip_defttl;
1624 ip->ip_len = m->m_pkthdr.len;
1625 m->m_flags |= M_SKIP_FIREWALL;
1626 return (m);
1627}
1628
1629/*
1630 * sends a reject message, consuming the mbuf passed as an argument.
1631 */
1632static void
1633send_reject(struct ip_fw_args *args, int code, u_short offset, int ip_len)
1634{
1635
1636 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1637 /* We need the IP header in host order for icmp_error(). */
1638 if (args->eh != NULL) {
1639 struct ip *ip = mtod(args->m, struct ip *);
1640 ip->ip_len = ntohs(ip->ip_len);
1641 ip->ip_off = ntohs(ip->ip_off);
1642 }
1643 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1644 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1645 struct tcphdr *const tcp =
1646 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1647 if ( (tcp->th_flags & TH_RST) == 0) {
1648 struct mbuf *m;
1649 m = send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1650 ntohl(tcp->th_ack),
1651 tcp->th_flags | TH_RST);
1652 if (m != NULL)
1653 ip_output(m, NULL, NULL, 0, NULL, NULL);
1654 }
1655 m_freem(args->m);
1656 } else
1657 m_freem(args->m);
1658 args->m = NULL;
1659}
1660
1661/**
1662 *
1663 * Given an ip_fw *, lookup_next_rule will return a pointer
1664 * to the next rule, which can be either the jump
1665 * target (for skipto instructions) or the next one in the list (in
1666 * all other cases including a missing jump target).
1667 * The result is also written in the "next_rule" field of the rule.
1668 * Backward jumps are not allowed, so start looking from the next
1669 * rule...
1670 *
1671 * This never returns NULL -- in case we do not have an exact match,
1672 * the next rule is returned. When the ruleset is changed,
1673 * pointers are flushed so we are always correct.
1674 */
1675
1676static struct ip_fw *
1677lookup_next_rule(struct ip_fw *me)
1678{
1679 struct ip_fw *rule = NULL;
1680 ipfw_insn *cmd;
1681
1682 /* look for action, in case it is a skipto */
1683 cmd = ACTION_PTR(me);
1684 if (cmd->opcode == O_LOG)
1685 cmd += F_LEN(cmd);
1686 if (cmd->opcode == O_ALTQ)
1687 cmd += F_LEN(cmd);
1688 if ( cmd->opcode == O_SKIPTO )
1689 for (rule = me->next; rule ; rule = rule->next)
1690 if (rule->rulenum >= cmd->arg1)
1691 break;
1692 if (rule == NULL) /* failure or not a skipto */
1693 rule = me->next;
1694 me->next_rule = rule;
1695 return rule;
1696}
1697
1698static void
1699init_tables(struct ip_fw_chain *ch)
1700{
1701 int i;
1702
1703 for (i = 0; i < IPFW_TABLES_MAX; i++)
1704 rn_inithead((void **)&ch->tables[i], 32);
1705}
1706
1707static int
1708add_table_entry(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1709 uint8_t mlen, uint32_t value)
1710{
1711 struct radix_node_head *rnh;
1712 struct table_entry *ent;
1713
1714 if (tbl >= IPFW_TABLES_MAX)
1715 return (EINVAL);
1716 rnh = ch->tables[tbl];
1717 ent = malloc(sizeof(*ent), M_IPFW_TBL, M_NOWAIT | M_ZERO);
1718 if (ent == NULL)
1719 return (ENOMEM);
1720 ent->value = value;
1721 ent->addr.sin_len = ent->mask.sin_len = 8;
1722 ent->mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1723 ent->addr.sin_addr.s_addr = addr & ent->mask.sin_addr.s_addr;
1724 IPFW_WLOCK(&layer3_chain);
1725 if (rnh->rnh_addaddr(&ent->addr, &ent->mask, rnh, (void *)ent) ==
1726 NULL) {
1727 IPFW_WUNLOCK(&layer3_chain);
1728 free(ent, M_IPFW_TBL);
1729 return (EEXIST);
1730 }
1731 IPFW_WUNLOCK(&layer3_chain);
1732 return (0);
1733}
1734
1735static int
1736del_table_entry(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1737 uint8_t mlen)
1738{
1739 struct radix_node_head *rnh;
1740 struct table_entry *ent;
1741 struct sockaddr_in sa, mask;
1742
1743 if (tbl >= IPFW_TABLES_MAX)
1744 return (EINVAL);
1745 rnh = ch->tables[tbl];
1746 sa.sin_len = mask.sin_len = 8;
1747 mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1748 sa.sin_addr.s_addr = addr & mask.sin_addr.s_addr;
1749 IPFW_WLOCK(ch);
1750 ent = (struct table_entry *)rnh->rnh_deladdr(&sa, &mask, rnh);
1751 if (ent == NULL) {
1752 IPFW_WUNLOCK(ch);
1753 return (ESRCH);
1754 }
1755 IPFW_WUNLOCK(ch);
1756 free(ent, M_IPFW_TBL);
1757 return (0);
1758}
1759
1760static int
1761flush_table_entry(struct radix_node *rn, void *arg)
1762{
1763 struct radix_node_head * const rnh = arg;
1764 struct table_entry *ent;
1765
1766 ent = (struct table_entry *)
1767 rnh->rnh_deladdr(rn->rn_key, rn->rn_mask, rnh);
1768 if (ent != NULL)
1769 free(ent, M_IPFW_TBL);
1770 return (0);
1771}
1772
1773static int
1774flush_table(struct ip_fw_chain *ch, uint16_t tbl)
1775{
1776 struct radix_node_head *rnh;
1777
1778 IPFW_WLOCK_ASSERT(ch);
1779
1780 if (tbl >= IPFW_TABLES_MAX)
1781 return (EINVAL);
1782 rnh = ch->tables[tbl];
1783 rnh->rnh_walktree(rnh, flush_table_entry, rnh);
1784 return (0);
1785}
1786
1787static void
1788flush_tables(struct ip_fw_chain *ch)
1789{
1790 uint16_t tbl;
1791
1792 IPFW_WLOCK_ASSERT(ch);
1793
1794 for (tbl = 0; tbl < IPFW_TABLES_MAX; tbl++)
1795 flush_table(ch, tbl);
1796}
1797
1798static int
1799lookup_table(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1800 uint32_t *val)
1801{
1802 struct radix_node_head *rnh;
1803 struct table_entry *ent;
1804 struct sockaddr_in sa;
1805
1806 if (tbl >= IPFW_TABLES_MAX)
1807 return (0);
1808 rnh = ch->tables[tbl];
1809 sa.sin_len = 8;
1810 sa.sin_addr.s_addr = addr;
1811 ent = (struct table_entry *)(rnh->rnh_lookup(&sa, NULL, rnh));
1812 if (ent != NULL) {
1813 *val = ent->value;
1814 return (1);
1815 }
1816 return (0);
1817}
1818
1819static int
1820count_table_entry(struct radix_node *rn, void *arg)
1821{
1822 u_int32_t * const cnt = arg;
1823
1824 (*cnt)++;
1825 return (0);
1826}
1827
1828static int
1829count_table(struct ip_fw_chain *ch, uint32_t tbl, uint32_t *cnt)
1830{
1831 struct radix_node_head *rnh;
1832
1833 if (tbl >= IPFW_TABLES_MAX)
1834 return (EINVAL);
1835 rnh = ch->tables[tbl];
1836 *cnt = 0;
1837 rnh->rnh_walktree(rnh, count_table_entry, cnt);
1838 return (0);
1839}
1840
1841static int
1842dump_table_entry(struct radix_node *rn, void *arg)
1843{
1844 struct table_entry * const n = (struct table_entry *)rn;
1845 ipfw_table * const tbl = arg;
1846 ipfw_table_entry *ent;
1847
1848 if (tbl->cnt == tbl->size)
1849 return (1);
1850 ent = &tbl->ent[tbl->cnt];
1851 ent->tbl = tbl->tbl;
1852 if (in_nullhost(n->mask.sin_addr))
1853 ent->masklen = 0;
1854 else
1855 ent->masklen = 33 - ffs(ntohl(n->mask.sin_addr.s_addr));
1856 ent->addr = n->addr.sin_addr.s_addr;
1857 ent->value = n->value;
1858 tbl->cnt++;
1859 return (0);
1860}
1861
1862static int
1863dump_table(struct ip_fw_chain *ch, ipfw_table *tbl)
1864{
1865 struct radix_node_head *rnh;
1866
1867 IPFW_WLOCK_ASSERT(ch);
1868
1869 if (tbl->tbl >= IPFW_TABLES_MAX)
1870 return (EINVAL);
1871 rnh = ch->tables[tbl->tbl];
1872 tbl->cnt = 0;
1873 rnh->rnh_walktree(rnh, dump_table_entry, tbl);
1874 return (0);
1875}
1876
1877static void
1878fill_ugid_cache(struct inpcb *inp, struct ip_fw_ugid *ugp)
1879{
1880 struct ucred *cr;
1881
1882 if (inp->inp_socket != NULL) {
1883 cr = inp->inp_socket->so_cred;
1884 ugp->fw_prid = jailed(cr) ?
1885 cr->cr_prison->pr_id : -1;
1886 ugp->fw_uid = cr->cr_uid;
1887 ugp->fw_ngroups = cr->cr_ngroups;
1888 bcopy(cr->cr_groups, ugp->fw_groups,
1889 sizeof(ugp->fw_groups));
1890 }
1891}
1892
1893static int
1894check_uidgid(ipfw_insn_u32 *insn,
1895 int proto, struct ifnet *oif,
1896 struct in_addr dst_ip, u_int16_t dst_port,
1897 struct in_addr src_ip, u_int16_t src_port,
1898 struct ip_fw_ugid *ugp, int *lookup, struct inpcb *inp)
1899{
1900 struct inpcbinfo *pi;
1901 int wildcard;
1902 struct inpcb *pcb;
1903 int match;
1904 gid_t *gp;
1905
1906 /*
1907 * Check to see if the UDP or TCP stack supplied us with
1908 * the PCB. If so, rather then holding a lock and looking
1909 * up the PCB, we can use the one that was supplied.
1910 */
1911 if (inp && *lookup == 0) {
1912 INP_LOCK_ASSERT(inp);
1913 if (inp->inp_socket != NULL) {
1914 fill_ugid_cache(inp, ugp);
1915 *lookup = 1;
1916 }
1917 }
1918 /*
1919 * If we have already been here and the packet has no
1920 * PCB entry associated with it, then we can safely
1921 * assume that this is a no match.
1922 */
1923 if (*lookup == -1)
1924 return (0);
1925 if (proto == IPPROTO_TCP) {
1926 wildcard = 0;
1927 pi = &tcbinfo;
1928 } else if (proto == IPPROTO_UDP) {
1929 wildcard = 1;
1930 pi = &udbinfo;
1931 } else
1932 return 0;
1933 match = 0;
1934 if (*lookup == 0) {
1935 INP_INFO_RLOCK(pi);
1936 pcb = (oif) ?
1937 in_pcblookup_hash(pi,
1938 dst_ip, htons(dst_port),
1939 src_ip, htons(src_port),
1940 wildcard, oif) :
1941 in_pcblookup_hash(pi,
1942 src_ip, htons(src_port),
1943 dst_ip, htons(dst_port),
1944 wildcard, NULL);
1945 if (pcb != NULL) {
1946 INP_LOCK(pcb);
1947 if (pcb->inp_socket != NULL) {
1948 fill_ugid_cache(pcb, ugp);
1949 *lookup = 1;
1950 }
1951 INP_UNLOCK(pcb);
1952 }
1953 INP_INFO_RUNLOCK(pi);
1954 if (*lookup == 0) {
1955 /*
1956 * If the lookup did not yield any results, there
1957 * is no sense in coming back and trying again. So
1958 * we can set lookup to -1 and ensure that we wont
1959 * bother the pcb system again.
1960 */
1961 *lookup = -1;
1962 return (0);
1963 }
1964 }
1965 if (insn->o.opcode == O_UID)
1966 match = (ugp->fw_uid == (uid_t)insn->d[0]);
1967 else if (insn->o.opcode == O_GID) {
1968 for (gp = ugp->fw_groups;
1969 gp < &ugp->fw_groups[ugp->fw_ngroups]; gp++)
1970 if (*gp == (gid_t)insn->d[0]) {
1971 match = 1;
1972 break;
1973 }
1974 } else if (insn->o.opcode == O_JAIL)
1975 match = (ugp->fw_prid == (int)insn->d[0]);
1976 return match;
1977}
1978
1979/*
1980 * The main check routine for the firewall.
1981 *
1982 * All arguments are in args so we can modify them and return them
1983 * back to the caller.
1984 *
1985 * Parameters:
1986 *
1987 * args->m (in/out) The packet; we set to NULL when/if we nuke it.
1988 * Starts with the IP header.
1989 * args->eh (in) Mac header if present, or NULL for layer3 packet.
1990 * args->oif Outgoing interface, or NULL if packet is incoming.
1991 * The incoming interface is in the mbuf. (in)
1992 * args->divert_rule (in/out)
1993 * Skip up to the first rule past this rule number;
1994 * upon return, non-zero port number for divert or tee.
1995 *
1996 * args->rule Pointer to the last matching rule (in/out)
1997 * args->next_hop Socket we are forwarding to (out).
1998 * args->f_id Addresses grabbed from the packet (out)
1999 * args->cookie a cookie depending on rule action
2000 *
2001 * Return value:
2002 *
2003 * IP_FW_PASS the packet must be accepted
2004 * IP_FW_DENY the packet must be dropped
2005 * IP_FW_DIVERT divert packet, port in m_tag
2006 * IP_FW_TEE tee packet, port in m_tag
2007 * IP_FW_DUMMYNET to dummynet, pipe in args->cookie
2008 * IP_FW_NETGRAPH into netgraph, cookie args->cookie
2009 *
2010 */
2011
2012int
2013ipfw_chk(struct ip_fw_args *args)
2014{
2015 /*
2016 * Local variables hold state during the processing of a packet.
2017 *
2018 * IMPORTANT NOTE: to speed up the processing of rules, there
2019 * are some assumption on the values of the variables, which
2020 * are documented here. Should you change them, please check
2021 * the implementation of the various instructions to make sure
2022 * that they still work.
2023 *
2024 * args->eh The MAC header. It is non-null for a layer2
2025 * packet, it is NULL for a layer-3 packet.
2026 *
2027 * m | args->m Pointer to the mbuf, as received from the caller.
2028 * It may change if ipfw_chk() does an m_pullup, or if it
2029 * consumes the packet because it calls send_reject().
2030 * XXX This has to change, so that ipfw_chk() never modifies
2031 * or consumes the buffer.
2032 * ip is simply an alias of the value of m, and it is kept
2033 * in sync with it (the packet is supposed to start with
2034 * the ip header).
2035 */
2036 struct mbuf *m = args->m;
2037 struct ip *ip = mtod(m, struct ip *);
2038
2039 /*
2040 * For rules which contain uid/gid or jail constraints, cache
2041 * a copy of the users credentials after the pcb lookup has been
2042 * executed. This will speed up the processing of rules with
2043 * these types of constraints, as well as decrease contention
2044 * on pcb related locks.
2045 */
2046 struct ip_fw_ugid fw_ugid_cache;
2047 int ugid_lookup = 0;
2048
2049 /*
2050 * divinput_flags If non-zero, set to the IP_FW_DIVERT_*_FLAG
2051 * associated with a packet input on a divert socket. This
2052 * will allow to distinguish traffic and its direction when
2053 * it originates from a divert socket.
2054 */
2055 u_int divinput_flags = 0;
2056
2057 /*
2058 * oif | args->oif If NULL, ipfw_chk has been called on the
2059 * inbound path (ether_input, ip_input).
2060 * If non-NULL, ipfw_chk has been called on the outbound path
2061 * (ether_output, ip_output).
2062 */
2063 struct ifnet *oif = args->oif;
2064
2065 struct ip_fw *f = NULL; /* matching rule */
2066 int retval = 0;
2067
2068 /*
2069 * hlen The length of the IP header.
2070 */
2071 u_int hlen = 0; /* hlen >0 means we have an IP pkt */
2072
2073 /*
2074 * offset The offset of a fragment. offset != 0 means that
2075 * we have a fragment at this offset of an IPv4 packet.
2076 * offset == 0 means that (if this is an IPv4 packet)
2077 * this is the first or only fragment.
2078 * For IPv6 offset == 0 means there is no Fragment Header.
2079 * If offset != 0 for IPv6 always use correct mask to
2080 * get the correct offset because we add IP6F_MORE_FRAG
2081 * to be able to dectect the first fragment which would
2082 * otherwise have offset = 0.
2083 */
2084 u_short offset = 0;
2085
2086 /*
2087 * Local copies of addresses. They are only valid if we have
2088 * an IP packet.
2089 *
2090 * proto The protocol. Set to 0 for non-ip packets,
2091 * or to the protocol read from the packet otherwise.
2092 * proto != 0 means that we have an IPv4 packet.
2093 *
2094 * src_port, dst_port port numbers, in HOST format. Only
2095 * valid for TCP and UDP packets.
2096 *
2097 * src_ip, dst_ip ip addresses, in NETWORK format.
2098 * Only valid for IPv4 packets.
2099 */
2100 u_int8_t proto;
2101 u_int16_t src_port = 0, dst_port = 0; /* NOTE: host format */
2102 struct in_addr src_ip, dst_ip; /* NOTE: network format */
2103 u_int16_t ip_len=0;
2104 int pktlen;
2105
2106 /*
2107 * dyn_dir = MATCH_UNKNOWN when rules unchecked,
2108 * MATCH_NONE when checked and not matched (q = NULL),
2109 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
2110 */
2111 int dyn_dir = MATCH_UNKNOWN;
2112 ipfw_dyn_rule *q = NULL;
2113 struct ip_fw_chain *chain = &layer3_chain;
2114 struct m_tag *mtag;
2115
2116 /*
2117 * We store in ulp a pointer to the upper layer protocol header.
2118 * In the ipv4 case this is easy to determine from the header,
2119 * but for ipv6 we might have some additional headers in the middle.
2120 * ulp is NULL if not found.
2121 */
2122 void *ulp = NULL; /* upper layer protocol pointer. */
2123 /* XXX ipv6 variables */
2124 int is_ipv6 = 0;
2125 u_int16_t ext_hd = 0; /* bits vector for extension header filtering */
2126 /* end of ipv6 variables */
2127 int is_ipv4 = 0;
2128
2129 if (m->m_flags & M_SKIP_FIREWALL)
2130 return (IP_FW_PASS); /* accept */
2131
2132 pktlen = m->m_pkthdr.len;
2133 proto = args->f_id.proto = 0; /* mark f_id invalid */
2134 /* XXX 0 is a valid proto: IP/IPv6 Hop-by-Hop Option */
2135
2136/*
2137 * PULLUP_TO(len, p, T) makes sure that len + sizeof(T) is contiguous,
2138 * then it sets p to point at the offset "len" in the mbuf. WARNING: the
2139 * pointer might become stale after other pullups (but we never use it
2140 * this way).
2141 */
2142#define PULLUP_TO(len, p, T) \
2143do { \
2144 int x = (len) + sizeof(T); \
2145 if ((m)->m_len < x) { \
2146 args->m = m = m_pullup(m, x); \
2147 if (m == NULL) \
2148 goto pullup_failed; \
2149 } \
2150 p = (mtod(m, char *) + (len)); \
2151} while (0)
2152
2153 /* Identify IP packets and fill up variables. */
2154 if (pktlen >= sizeof(struct ip6_hdr) &&
2155 (args->eh == NULL || ntohs(args->eh->ether_type)==ETHERTYPE_IPV6) &&
2156 mtod(m, struct ip *)->ip_v == 6) {
2157 is_ipv6 = 1;
2158 args->f_id.addr_type = 6;
2159 hlen = sizeof(struct ip6_hdr);
2160 proto = mtod(m, struct ip6_hdr *)->ip6_nxt;
2161
2162 /* Search extension headers to find upper layer protocols */
2163 while (ulp == NULL) {
2164 switch (proto) {
2165 case IPPROTO_ICMPV6:
2166 PULLUP_TO(hlen, ulp, struct icmp6_hdr);
2167 args->f_id.flags = ICMP6(ulp)->icmp6_type;
2168 break;
2169
2170 case IPPROTO_TCP:
2171 PULLUP_TO(hlen, ulp, struct tcphdr);
2172 dst_port = TCP(ulp)->th_dport;
2173 src_port = TCP(ulp)->th_sport;
2174 args->f_id.flags = TCP(ulp)->th_flags;
2175 break;
2176
2177 case IPPROTO_UDP:
2178 PULLUP_TO(hlen, ulp, struct udphdr);
2179 dst_port = UDP(ulp)->uh_dport;
2180 src_port = UDP(ulp)->uh_sport;
2181 break;
2182
2183 case IPPROTO_HOPOPTS: /* RFC 2460 */
2184 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2185 ext_hd |= EXT_HOPOPTS;
2186 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2187 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2188 ulp = NULL;
2189 break;
2190
2191 case IPPROTO_ROUTING: /* RFC 2460 */
2192 PULLUP_TO(hlen, ulp, struct ip6_rthdr);
2193 if (((struct ip6_rthdr *)ulp)->ip6r_type != 0) {
2194 printf("IPFW2: IPV6 - Unknown Routing "
2195 "Header type(%d)\n",
2196 ((struct ip6_rthdr *)ulp)->ip6r_type);
2197 if (fw_deny_unknown_exthdrs)
2198 return (IP_FW_DENY);
2199 break;
2200 }
2201 ext_hd |= EXT_ROUTING;
2202 hlen += (((struct ip6_rthdr *)ulp)->ip6r_len + 1) << 3;
2203 proto = ((struct ip6_rthdr *)ulp)->ip6r_nxt;
2204 ulp = NULL;
2205 break;
2206
2207 case IPPROTO_FRAGMENT: /* RFC 2460 */
2208 PULLUP_TO(hlen, ulp, struct ip6_frag);
2209 ext_hd |= EXT_FRAGMENT;
2210 hlen += sizeof (struct ip6_frag);
2211 proto = ((struct ip6_frag *)ulp)->ip6f_nxt;
2212 offset = ((struct ip6_frag *)ulp)->ip6f_offlg &
2213 IP6F_OFF_MASK;
2214 /* Add IP6F_MORE_FRAG for offset of first
2215 * fragment to be != 0. */
2216 offset |= ((struct ip6_frag *)ulp)->ip6f_offlg &
2217 IP6F_MORE_FRAG;
2218 if (offset == 0) {
2219 printf("IPFW2: IPV6 - Invalid Fragment "
2220 "Header\n");
2221 if (fw_deny_unknown_exthdrs)
2222 return (IP_FW_DENY);
2223 break;
2224 }
2225 args->f_id.frag_id6 =
2226 ntohl(((struct ip6_frag *)ulp)->ip6f_ident);
2227 ulp = NULL;
2228 break;
2229
2230 case IPPROTO_DSTOPTS: /* RFC 2460 */
2231 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2232 ext_hd |= EXT_DSTOPTS;
2233 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2234 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2235 ulp = NULL;
2236 break;
2237
2238 case IPPROTO_AH: /* RFC 2402 */
2239 PULLUP_TO(hlen, ulp, struct ip6_ext);
2240 ext_hd |= EXT_AH;
2241 hlen += (((struct ip6_ext *)ulp)->ip6e_len + 2) << 2;
2242 proto = ((struct ip6_ext *)ulp)->ip6e_nxt;
2243 ulp = NULL;
2244 break;
2245
2246 case IPPROTO_ESP: /* RFC 2406 */
2247 PULLUP_TO(hlen, ulp, uint32_t); /* SPI, Seq# */
2248 /* Anything past Seq# is variable length and
2249 * data past this ext. header is encrypted. */
2250 ext_hd |= EXT_ESP;
2251 break;
2252
2253 case IPPROTO_NONE: /* RFC 2460 */
2254 PULLUP_TO(hlen, ulp, struct ip6_ext);
2255 /* Packet ends here. if ip6e_len!=0 octets
2256 * must be ignored. */
2257 break;
2258
2259 case IPPROTO_OSPFIGP:
2260 /* XXX OSPF header check? */
2261 PULLUP_TO(hlen, ulp, struct ip6_ext);
2262 break;
2263
2264 default:
2265 printf("IPFW2: IPV6 - Unknown Extension "
2266 "Header(%d), ext_hd=%x\n", proto, ext_hd);
2267 if (fw_deny_unknown_exthdrs)
2268 return (IP_FW_DENY);
2269 break;
2270 } /*switch */
2271 }
2272 args->f_id.src_ip6 = mtod(m,struct ip6_hdr *)->ip6_src;
2273 args->f_id.dst_ip6 = mtod(m,struct ip6_hdr *)->ip6_dst;
2274 args->f_id.src_ip = 0;
2275 args->f_id.dst_ip = 0;
2276 args->f_id.flow_id6 = ntohl(mtod(m, struct ip6_hdr *)->ip6_flow);
2277 } else if (pktlen >= sizeof(struct ip) &&
2278 (args->eh == NULL || ntohs(args->eh->ether_type) == ETHERTYPE_IP) &&
2279 mtod(m, struct ip *)->ip_v == 4) {
2280 is_ipv4 = 1;
2281 ip = mtod(m, struct ip *);
2282 hlen = ip->ip_hl << 2;
2283 args->f_id.addr_type = 4;
2284
2285 /*
2286 * Collect parameters into local variables for faster matching.
2287 */
2288 proto = ip->ip_p;
2289 src_ip = ip->ip_src;
2290 dst_ip = ip->ip_dst;
2291 if (args->eh != NULL) { /* layer 2 packets are as on the wire */
2292 offset = ntohs(ip->ip_off) & IP_OFFMASK;
2293 ip_len = ntohs(ip->ip_len);
2294 } else {
2295 offset = ip->ip_off & IP_OFFMASK;
2296 ip_len = ip->ip_len;
2297 }
2298 pktlen = ip_len < pktlen ? ip_len : pktlen;
2299
2300 if (offset == 0) {
2301 switch (proto) {
2302 case IPPROTO_TCP:
2303 PULLUP_TO(hlen, ulp, struct tcphdr);
2304 dst_port = TCP(ulp)->th_dport;
2305 src_port = TCP(ulp)->th_sport;
2306 args->f_id.flags = TCP(ulp)->th_flags;
2307 break;
2308
2309 case IPPROTO_UDP:
2310 PULLUP_TO(hlen, ulp, struct udphdr);
2311 dst_port = UDP(ulp)->uh_dport;
2312 src_port = UDP(ulp)->uh_sport;
2313 break;
2314
2315 case IPPROTO_ICMP:
2316 PULLUP_TO(hlen, ulp, struct icmphdr);
2317 args->f_id.flags = ICMP(ulp)->icmp_type;
2318 break;
2319
2320 default:
2321 break;
2322 }
2323 }
2324
2325 args->f_id.src_ip = ntohl(src_ip.s_addr);
2326 args->f_id.dst_ip = ntohl(dst_ip.s_addr);
2327 }
2328#undef PULLUP_TO
2329 if (proto) { /* we may have port numbers, store them */
2330 args->f_id.proto = proto;
2331 args->f_id.src_port = src_port = ntohs(src_port);
2332 args->f_id.dst_port = dst_port = ntohs(dst_port);
2333 }
2334
2335 IPFW_RLOCK(chain);
2336 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
2337 if (args->rule) {
2338 /*
2339 * Packet has already been tagged. Look for the next rule
2340 * to restart processing.
2341 *
2342 * If fw_one_pass != 0 then just accept it.
2343 * XXX should not happen here, but optimized out in
2344 * the caller.
2345 */
2346 if (fw_one_pass) {
2347 IPFW_RUNLOCK(chain);
2348 return (IP_FW_PASS);
2349 }
2350
2351 f = args->rule->next_rule;
2352 if (f == NULL)
2353 f = lookup_next_rule(args->rule);
2354 } else {
2355 /*
2356 * Find the starting rule. It can be either the first
2357 * one, or the one after divert_rule if asked so.
2358 */
2359 int skipto = mtag ? divert_cookie(mtag) : 0;
2360
2361 f = chain->rules;
2362 if (args->eh == NULL && skipto != 0) {
2363 if (skipto >= IPFW_DEFAULT_RULE) {
2364 IPFW_RUNLOCK(chain);
2365 return (IP_FW_DENY); /* invalid */
2366 }
2367 while (f && f->rulenum <= skipto)
2368 f = f->next;
2369 if (f == NULL) { /* drop packet */
2370 IPFW_RUNLOCK(chain);
2371 return (IP_FW_DENY);
2372 }
2373 }
2374 }
2375 /* reset divert rule to avoid confusion later */
2376 if (mtag) {
2377 divinput_flags = divert_info(mtag) &
2378 (IP_FW_DIVERT_OUTPUT_FLAG | IP_FW_DIVERT_LOOPBACK_FLAG);
2379 m_tag_delete(m, mtag);
2380 }
2381
2382 /*
2383 * Now scan the rules, and parse microinstructions for each rule.
2384 */
2385 for (; f; f = f->next) {
|
3089 retval = (cmd->opcode == O_NETGRAPH) ?
3090 IP_FW_NETGRAPH : IP_FW_NGTEE;
3091 goto done;
3092
3093 default:
3094 panic("-- unknown opcode %d\n", cmd->opcode);
3095 } /* end of switch() on opcodes */
3096
3097 if (cmd->len & F_NOT)
3098 match = !match;
3099
3100 if (match) {
3101 if (cmd->len & F_OR)
3102 skip_or = 1;
3103 } else {
3104 if (!(cmd->len & F_OR)) /* not an OR block, */
3105 break; /* try next rule */
3106 }
3107
3108 } /* end of inner for, scan opcodes */
3109
3110next_rule:; /* try next rule */
3111
3112 } /* end of outer for, scan rules */
3113 printf("ipfw: ouch!, skip past end of rules, denying packet\n");
3114 IPFW_RUNLOCK(chain);
3115 return (IP_FW_DENY);
3116
3117done:
3118 /* Update statistics */
3119 f->pcnt++;
3120 f->bcnt += pktlen;
3121 f->timestamp = time_uptime;
3122 IPFW_RUNLOCK(chain);
3123 return (retval);
3124
3125pullup_failed:
3126 if (fw_verbose)
3127 printf("ipfw: pullup failed\n");
3128 return (IP_FW_DENY);
3129}
3130
3131/*
3132 * When a rule is added/deleted, clear the next_rule pointers in all rules.
3133 * These will be reconstructed on the fly as packets are matched.
3134 */
3135static void
3136flush_rule_ptrs(struct ip_fw_chain *chain)
3137{
3138 struct ip_fw *rule;
3139
3140 IPFW_WLOCK_ASSERT(chain);
3141
3142 for (rule = chain->rules; rule; rule = rule->next)
3143 rule->next_rule = NULL;
3144}
3145
3146/*
3147 * Add a new rule to the list. Copy the rule into a malloc'ed area, then
3148 * possibly create a rule number and add the rule to the list.
3149 * Update the rule_number in the input struct so the caller knows it as well.
3150 */
3151static int
3152add_rule(struct ip_fw_chain *chain, struct ip_fw *input_rule)
3153{
3154 struct ip_fw *rule, *f, *prev;
3155 int l = RULESIZE(input_rule);
3156
3157 if (chain->rules == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE)
3158 return (EINVAL);
3159
3160 rule = malloc(l, M_IPFW, M_NOWAIT | M_ZERO);
3161 if (rule == NULL)
3162 return (ENOSPC);
3163
3164 bcopy(input_rule, rule, l);
3165
3166 rule->next = NULL;
3167 rule->next_rule = NULL;
3168
3169 rule->pcnt = 0;
3170 rule->bcnt = 0;
3171 rule->timestamp = 0;
3172
3173 IPFW_WLOCK(chain);
3174
3175 if (chain->rules == NULL) { /* default rule */
3176 chain->rules = rule;
3177 goto done;
3178 }
3179
3180 /*
3181 * If rulenum is 0, find highest numbered rule before the
3182 * default rule, and add autoinc_step
3183 */
3184 if (autoinc_step < 1)
3185 autoinc_step = 1;
3186 else if (autoinc_step > 1000)
3187 autoinc_step = 1000;
3188 if (rule->rulenum == 0) {
3189 /*
3190 * locate the highest numbered rule before default
3191 */
3192 for (f = chain->rules; f; f = f->next) {
3193 if (f->rulenum == IPFW_DEFAULT_RULE)
3194 break;
3195 rule->rulenum = f->rulenum;
3196 }
3197 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
3198 rule->rulenum += autoinc_step;
3199 input_rule->rulenum = rule->rulenum;
3200 }
3201
3202 /*
3203 * Now insert the new rule in the right place in the sorted list.
3204 */
3205 for (prev = NULL, f = chain->rules; f; prev = f, f = f->next) {
3206 if (f->rulenum > rule->rulenum) { /* found the location */
3207 if (prev) {
3208 rule->next = f;
3209 prev->next = rule;
3210 } else { /* head insert */
3211 rule->next = chain->rules;
3212 chain->rules = rule;
3213 }
3214 break;
3215 }
3216 }
3217 flush_rule_ptrs(chain);
3218done:
3219 static_count++;
3220 static_len += l;
3221 IPFW_WUNLOCK(chain);
3222 DEB(printf("ipfw: installed rule %d, static count now %d\n",
3223 rule->rulenum, static_count);)
3224 return (0);
3225}
3226
3227/**
3228 * Remove a static rule (including derived * dynamic rules)
3229 * and place it on the ``reap list'' for later reclamation.
3230 * The caller is in charge of clearing rule pointers to avoid
3231 * dangling pointers.
3232 * @return a pointer to the next entry.
3233 * Arguments are not checked, so they better be correct.
3234 */
3235static struct ip_fw *
3236remove_rule(struct ip_fw_chain *chain, struct ip_fw *rule, struct ip_fw *prev)
3237{
3238 struct ip_fw *n;
3239 int l = RULESIZE(rule);
3240
3241 IPFW_WLOCK_ASSERT(chain);
3242
3243 n = rule->next;
3244 IPFW_DYN_LOCK();
3245 remove_dyn_rule(rule, NULL /* force removal */);
3246 IPFW_DYN_UNLOCK();
3247 if (prev == NULL)
3248 chain->rules = n;
3249 else
3250 prev->next = n;
3251 static_count--;
3252 static_len -= l;
3253
3254 rule->next = chain->reap;
3255 chain->reap = rule;
3256
3257 return n;
3258}
3259
3260/**
3261 * Reclaim storage associated with a list of rules. This is
3262 * typically the list created using remove_rule.
3263 */
3264static void
3265reap_rules(struct ip_fw *head)
3266{
3267 struct ip_fw *rule;
3268
3269 while ((rule = head) != NULL) {
3270 head = head->next;
3271 if (DUMMYNET_LOADED)
3272 ip_dn_ruledel_ptr(rule);
3273 free(rule, M_IPFW);
3274 }
3275}
3276
3277/*
3278 * Remove all rules from a chain (except rules in set RESVD_SET
3279 * unless kill_default = 1). The caller is responsible for
3280 * reclaiming storage for the rules left in chain->reap.
3281 */
3282static void
3283free_chain(struct ip_fw_chain *chain, int kill_default)
3284{
3285 struct ip_fw *prev, *rule;
3286
3287 IPFW_WLOCK_ASSERT(chain);
3288
3289 flush_rule_ptrs(chain); /* more efficient to do outside the loop */
3290 for (prev = NULL, rule = chain->rules; rule ; )
3291 if (kill_default || rule->set != RESVD_SET)
3292 rule = remove_rule(chain, rule, prev);
3293 else {
3294 prev = rule;
3295 rule = rule->next;
3296 }
3297}
3298
3299/**
3300 * Remove all rules with given number, and also do set manipulation.
3301 * Assumes chain != NULL && *chain != NULL.
3302 *
3303 * The argument is an u_int32_t. The low 16 bit are the rule or set number,
3304 * the next 8 bits are the new set, the top 8 bits are the command:
3305 *
3306 * 0 delete rules with given number
3307 * 1 delete rules with given set number
3308 * 2 move rules with given number to new set
3309 * 3 move rules with given set number to new set
3310 * 4 swap sets with given numbers
3311 */
3312static int
3313del_entry(struct ip_fw_chain *chain, u_int32_t arg)
3314{
3315 struct ip_fw *prev = NULL, *rule;
3316 u_int16_t rulenum; /* rule or old_set */
3317 u_int8_t cmd, new_set;
3318
3319 rulenum = arg & 0xffff;
3320 cmd = (arg >> 24) & 0xff;
3321 new_set = (arg >> 16) & 0xff;
3322
3323 if (cmd > 4)
3324 return EINVAL;
3325 if (new_set > RESVD_SET)
3326 return EINVAL;
3327 if (cmd == 0 || cmd == 2) {
3328 if (rulenum >= IPFW_DEFAULT_RULE)
3329 return EINVAL;
3330 } else {
3331 if (rulenum > RESVD_SET) /* old_set */
3332 return EINVAL;
3333 }
3334
3335 IPFW_WLOCK(chain);
3336 rule = chain->rules;
3337 chain->reap = NULL;
3338 switch (cmd) {
3339 case 0: /* delete rules with given number */
3340 /*
3341 * locate first rule to delete
3342 */
3343 for (; rule->rulenum < rulenum; prev = rule, rule = rule->next)
3344 ;
3345 if (rule->rulenum != rulenum) {
3346 IPFW_WUNLOCK(chain);
3347 return EINVAL;
3348 }
3349
3350 /*
3351 * flush pointers outside the loop, then delete all matching
3352 * rules. prev remains the same throughout the cycle.
3353 */
3354 flush_rule_ptrs(chain);
3355 while (rule->rulenum == rulenum)
3356 rule = remove_rule(chain, rule, prev);
3357 break;
3358
3359 case 1: /* delete all rules with given set number */
3360 flush_rule_ptrs(chain);
3361 rule = chain->rules;
3362 while (rule->rulenum < IPFW_DEFAULT_RULE)
3363 if (rule->set == rulenum)
3364 rule = remove_rule(chain, rule, prev);
3365 else {
3366 prev = rule;
3367 rule = rule->next;
3368 }
3369 break;
3370
3371 case 2: /* move rules with given number to new set */
3372 rule = chain->rules;
3373 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3374 if (rule->rulenum == rulenum)
3375 rule->set = new_set;
3376 break;
3377
3378 case 3: /* move rules with given set number to new set */
3379 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3380 if (rule->set == rulenum)
3381 rule->set = new_set;
3382 break;
3383
3384 case 4: /* swap two sets */
3385 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3386 if (rule->set == rulenum)
3387 rule->set = new_set;
3388 else if (rule->set == new_set)
3389 rule->set = rulenum;
3390 break;
3391 }
3392 /*
3393 * Look for rules to reclaim. We grab the list before
3394 * releasing the lock then reclaim them w/o the lock to
3395 * avoid a LOR with dummynet.
3396 */
3397 rule = chain->reap;
3398 chain->reap = NULL;
3399 IPFW_WUNLOCK(chain);
3400 if (rule)
3401 reap_rules(rule);
3402 return 0;
3403}
3404
3405/*
3406 * Clear counters for a specific rule.
3407 * The enclosing "table" is assumed locked.
3408 */
3409static void
3410clear_counters(struct ip_fw *rule, int log_only)
3411{
3412 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
3413
3414 if (log_only == 0) {
3415 rule->bcnt = rule->pcnt = 0;
3416 rule->timestamp = 0;
3417 }
3418 if (l->o.opcode == O_LOG)
3419 l->log_left = l->max_log;
3420}
3421
3422/**
3423 * Reset some or all counters on firewall rules.
3424 * @arg frwl is null to clear all entries, or contains a specific
3425 * rule number.
3426 * @arg log_only is 1 if we only want to reset logs, zero otherwise.
3427 */
3428static int
3429zero_entry(struct ip_fw_chain *chain, int rulenum, int log_only)
3430{
3431 struct ip_fw *rule;
3432 char *msg;
3433
3434 IPFW_WLOCK(chain);
3435 if (rulenum == 0) {
3436 norule_counter = 0;
3437 for (rule = chain->rules; rule; rule = rule->next)
3438 clear_counters(rule, log_only);
3439 msg = log_only ? "ipfw: All logging counts reset.\n" :
3440 "ipfw: Accounting cleared.\n";
3441 } else {
3442 int cleared = 0;
3443 /*
3444 * We can have multiple rules with the same number, so we
3445 * need to clear them all.
3446 */
3447 for (rule = chain->rules; rule; rule = rule->next)
3448 if (rule->rulenum == rulenum) {
3449 while (rule && rule->rulenum == rulenum) {
3450 clear_counters(rule, log_only);
3451 rule = rule->next;
3452 }
3453 cleared = 1;
3454 break;
3455 }
3456 if (!cleared) { /* we did not find any matching rules */
3457 IPFW_WUNLOCK(chain);
3458 return (EINVAL);
3459 }
3460 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
3461 "ipfw: Entry %d cleared.\n";
3462 }
3463 IPFW_WUNLOCK(chain);
3464
3465 if (fw_verbose)
3466 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
3467 return (0);
3468}
3469
3470/*
3471 * Check validity of the structure before insert.
3472 * Fortunately rules are simple, so this mostly need to check rule sizes.
3473 */
3474static int
3475check_ipfw_struct(struct ip_fw *rule, int size)
3476{
3477 int l, cmdlen = 0;
3478 int have_action=0;
3479 ipfw_insn *cmd;
3480
3481 if (size < sizeof(*rule)) {
3482 printf("ipfw: rule too short\n");
3483 return (EINVAL);
3484 }
3485 /* first, check for valid size */
3486 l = RULESIZE(rule);
3487 if (l != size) {
3488 printf("ipfw: size mismatch (have %d want %d)\n", size, l);
3489 return (EINVAL);
3490 }
3491 if (rule->act_ofs >= rule->cmd_len) {
3492 printf("ipfw: bogus action offset (%u > %u)\n",
3493 rule->act_ofs, rule->cmd_len - 1);
3494 return (EINVAL);
3495 }
3496 /*
3497 * Now go for the individual checks. Very simple ones, basically only
3498 * instruction sizes.
3499 */
3500 for (l = rule->cmd_len, cmd = rule->cmd ;
3501 l > 0 ; l -= cmdlen, cmd += cmdlen) {
3502 cmdlen = F_LEN(cmd);
3503 if (cmdlen > l) {
3504 printf("ipfw: opcode %d size truncated\n",
3505 cmd->opcode);
3506 return EINVAL;
3507 }
3508 DEB(printf("ipfw: opcode %d\n", cmd->opcode);)
3509 switch (cmd->opcode) {
3510 case O_PROBE_STATE:
3511 case O_KEEP_STATE:
3512 case O_PROTO:
3513 case O_IP_SRC_ME:
3514 case O_IP_DST_ME:
3515 case O_LAYER2:
3516 case O_IN:
3517 case O_FRAG:
3518 case O_DIVERTED:
3519 case O_IPOPT:
3520 case O_IPTOS:
3521 case O_IPPRECEDENCE:
3522 case O_IPVER:
3523 case O_TCPWIN:
3524 case O_TCPFLAGS:
3525 case O_TCPOPTS:
3526 case O_ESTAB:
3527 case O_VERREVPATH:
3528 case O_VERSRCREACH:
3529 case O_ANTISPOOF:
3530 case O_IPSEC:
3531#ifdef INET6
3532 case O_IP6_SRC_ME:
3533 case O_IP6_DST_ME:
3534 case O_EXT_HDR:
3535 case O_IP6:
3536#endif
3537 case O_IP4:
3538 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3539 goto bad_size;
3540 break;
3541
3542 case O_UID:
3543 case O_GID:
3544 case O_JAIL:
3545 case O_IP_SRC:
3546 case O_IP_DST:
3547 case O_TCPSEQ:
3548 case O_TCPACK:
3549 case O_PROB:
3550 case O_ICMPTYPE:
3551 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3552 goto bad_size;
3553 break;
3554
3555 case O_LIMIT:
3556 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
3557 goto bad_size;
3558 break;
3559
3560 case O_LOG:
3561 if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
3562 goto bad_size;
3563
3564 ((ipfw_insn_log *)cmd)->log_left =
3565 ((ipfw_insn_log *)cmd)->max_log;
3566
3567 break;
3568
3569 case O_IP_SRC_MASK:
3570 case O_IP_DST_MASK:
3571 /* only odd command lengths */
3572 if ( !(cmdlen & 1) || cmdlen > 31)
3573 goto bad_size;
3574 break;
3575
3576 case O_IP_SRC_SET:
3577 case O_IP_DST_SET:
3578 if (cmd->arg1 == 0 || cmd->arg1 > 256) {
3579 printf("ipfw: invalid set size %d\n",
3580 cmd->arg1);
3581 return EINVAL;
3582 }
3583 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3584 (cmd->arg1+31)/32 )
3585 goto bad_size;
3586 break;
3587
3588 case O_IP_SRC_LOOKUP:
3589 case O_IP_DST_LOOKUP:
3590 if (cmd->arg1 >= IPFW_TABLES_MAX) {
3591 printf("ipfw: invalid table number %d\n",
3592 cmd->arg1);
3593 return (EINVAL);
3594 }
3595 if (cmdlen != F_INSN_SIZE(ipfw_insn) &&
3596 cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3597 goto bad_size;
3598 break;
3599
3600 case O_MACADDR2:
3601 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
3602 goto bad_size;
3603 break;
3604
3605 case O_NOP:
3606 case O_IPID:
3607 case O_IPTTL:
3608 case O_IPLEN:
3609 case O_TCPDATALEN:
3610 if (cmdlen < 1 || cmdlen > 31)
3611 goto bad_size;
3612 break;
3613
3614 case O_MAC_TYPE:
3615 case O_IP_SRCPORT:
3616 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
3617 if (cmdlen < 2 || cmdlen > 31)
3618 goto bad_size;
3619 break;
3620
3621 case O_RECV:
3622 case O_XMIT:
3623 case O_VIA:
3624 if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
3625 goto bad_size;
3626 break;
3627
3628 case O_ALTQ:
3629 if (cmdlen != F_INSN_SIZE(ipfw_insn_altq))
3630 goto bad_size;
3631 break;
3632
3633 case O_PIPE:
3634 case O_QUEUE:
3635 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3636 goto bad_size;
3637 goto check_action;
3638
3639 case O_FORWARD_IP:
3640#ifdef IPFIREWALL_FORWARD
3641 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
3642 goto bad_size;
3643 goto check_action;
3644#else
3645 return EINVAL;
3646#endif
3647
3648 case O_DIVERT:
3649 case O_TEE:
3650 if (ip_divert_ptr == NULL)
3651 return EINVAL;
3652 else
3653 goto check_size;
3654 case O_NETGRAPH:
3655 case O_NGTEE:
3656 if (!NG_IPFW_LOADED)
3657 return EINVAL;
3658 else
3659 goto check_size;
3660 case O_FORWARD_MAC: /* XXX not implemented yet */
3661 case O_CHECK_STATE:
3662 case O_COUNT:
3663 case O_ACCEPT:
3664 case O_DENY:
3665 case O_REJECT:
3666#ifdef INET6
3667 case O_UNREACH6:
3668#endif
3669 case O_SKIPTO:
3670check_size:
3671 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3672 goto bad_size;
3673check_action:
3674 if (have_action) {
3675 printf("ipfw: opcode %d, multiple actions"
3676 " not allowed\n",
3677 cmd->opcode);
3678 return EINVAL;
3679 }
3680 have_action = 1;
3681 if (l != cmdlen) {
3682 printf("ipfw: opcode %d, action must be"
3683 " last opcode\n",
3684 cmd->opcode);
3685 return EINVAL;
3686 }
3687 break;
3688#ifdef INET6
3689 case O_IP6_SRC:
3690 case O_IP6_DST:
3691 if (cmdlen != F_INSN_SIZE(struct in6_addr) +
3692 F_INSN_SIZE(ipfw_insn))
3693 goto bad_size;
3694 break;
3695
3696 case O_FLOW6ID:
3697 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3698 ((ipfw_insn_u32 *)cmd)->o.arg1)
3699 goto bad_size;
3700 break;
3701
3702 case O_IP6_SRC_MASK:
3703 case O_IP6_DST_MASK:
3704 if ( !(cmdlen & 1) || cmdlen > 127)
3705 goto bad_size;
3706 break;
3707 case O_ICMP6TYPE:
3708 if( cmdlen != F_INSN_SIZE( ipfw_insn_icmp6 ) )
3709 goto bad_size;
3710 break;
3711#endif
3712
3713 default:
3714 switch (cmd->opcode) {
3715#ifndef INET6
3716 case O_IP6_SRC_ME:
3717 case O_IP6_DST_ME:
3718 case O_EXT_HDR:
3719 case O_IP6:
3720 case O_UNREACH6:
3721 case O_IP6_SRC:
3722 case O_IP6_DST:
3723 case O_FLOW6ID:
3724 case O_IP6_SRC_MASK:
3725 case O_IP6_DST_MASK:
3726 case O_ICMP6TYPE:
3727 printf("ipfw: no IPv6 support in kernel\n");
3728 return EPROTONOSUPPORT;
3729#endif
3730 default:
3731 printf("ipfw: opcode %d, unknown opcode\n",
3732 cmd->opcode);
3733 return EINVAL;
3734 }
3735 }
3736 }
3737 if (have_action == 0) {
3738 printf("ipfw: missing action\n");
3739 return EINVAL;
3740 }
3741 return 0;
3742
3743bad_size:
3744 printf("ipfw: opcode %d size %d wrong\n",
3745 cmd->opcode, cmdlen);
3746 return EINVAL;
3747}
3748
3749/*
3750 * Copy the static and dynamic rules to the supplied buffer
3751 * and return the amount of space actually used.
3752 */
3753static size_t
3754ipfw_getrules(struct ip_fw_chain *chain, void *buf, size_t space)
3755{
3756 char *bp = buf;
3757 char *ep = bp + space;
3758 struct ip_fw *rule;
3759 int i;
3760
3761 /* XXX this can take a long time and locking will block packet flow */
3762 IPFW_RLOCK(chain);
3763 for (rule = chain->rules; rule ; rule = rule->next) {
3764 /*
3765 * Verify the entry fits in the buffer in case the
3766 * rules changed between calculating buffer space and
3767 * now. This would be better done using a generation
3768 * number but should suffice for now.
3769 */
3770 i = RULESIZE(rule);
3771 if (bp + i <= ep) {
3772 bcopy(rule, bp, i);
3773 bcopy(&set_disable, &(((struct ip_fw *)bp)->next_rule),
3774 sizeof(set_disable));
3775 bp += i;
3776 }
3777 }
3778 IPFW_RUNLOCK(chain);
3779 if (ipfw_dyn_v) {
3780 ipfw_dyn_rule *p, *last = NULL;
3781
3782 IPFW_DYN_LOCK();
3783 for (i = 0 ; i < curr_dyn_buckets; i++)
3784 for (p = ipfw_dyn_v[i] ; p != NULL; p = p->next) {
3785 if (bp + sizeof *p <= ep) {
3786 ipfw_dyn_rule *dst =
3787 (ipfw_dyn_rule *)bp;
3788 bcopy(p, dst, sizeof *p);
3789 bcopy(&(p->rule->rulenum), &(dst->rule),
3790 sizeof(p->rule->rulenum));
3791 /*
3792 * store a non-null value in "next".
3793 * The userland code will interpret a
3794 * NULL here as a marker
3795 * for the last dynamic rule.
3796 */
3797 bcopy(&dst, &dst->next, sizeof(dst));
3798 last = dst;
3799 dst->expire =
3800 TIME_LEQ(dst->expire, time_uptime) ?
3801 0 : dst->expire - time_uptime ;
3802 bp += sizeof(ipfw_dyn_rule);
3803 }
3804 }
3805 IPFW_DYN_UNLOCK();
3806 if (last != NULL) /* mark last dynamic rule */
3807 bzero(&last->next, sizeof(last));
3808 }
3809 return (bp - (char *)buf);
3810}
3811
3812
3813/**
3814 * {set|get}sockopt parser.
3815 */
3816static int
3817ipfw_ctl(struct sockopt *sopt)
3818{
3819#define RULE_MAXSIZE (256*sizeof(u_int32_t))
3820 int error, rule_num;
3821 size_t size;
3822 struct ip_fw *buf, *rule;
3823 u_int32_t rulenum[2];
3824
3825 error = suser(sopt->sopt_td);
3826 if (error)
3827 return (error);
3828
3829 /*
3830 * Disallow modifications in really-really secure mode, but still allow
3831 * the logging counters to be reset.
3832 */
3833 if (sopt->sopt_name == IP_FW_ADD ||
3834 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
3835#if __FreeBSD_version >= 500034
3836 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
3837 if (error)
3838 return (error);
3839#else /* FreeBSD 4.x */
3840 if (securelevel >= 3)
3841 return (EPERM);
3842#endif
3843 }
3844
3845 error = 0;
3846
3847 switch (sopt->sopt_name) {
3848 case IP_FW_GET:
3849 /*
3850 * pass up a copy of the current rules. Static rules
3851 * come first (the last of which has number IPFW_DEFAULT_RULE),
3852 * followed by a possibly empty list of dynamic rule.
3853 * The last dynamic rule has NULL in the "next" field.
3854 *
3855 * Note that the calculated size is used to bound the
3856 * amount of data returned to the user. The rule set may
3857 * change between calculating the size and returning the
3858 * data in which case we'll just return what fits.
3859 */
3860 size = static_len; /* size of static rules */
3861 if (ipfw_dyn_v) /* add size of dyn.rules */
3862 size += (dyn_count * sizeof(ipfw_dyn_rule));
3863
3864 /*
3865 * XXX todo: if the user passes a short length just to know
3866 * how much room is needed, do not bother filling up the
3867 * buffer, just jump to the sooptcopyout.
3868 */
3869 buf = malloc(size, M_TEMP, M_WAITOK);
3870 error = sooptcopyout(sopt, buf,
3871 ipfw_getrules(&layer3_chain, buf, size));
3872 free(buf, M_TEMP);
3873 break;
3874
3875 case IP_FW_FLUSH:
3876 /*
3877 * Normally we cannot release the lock on each iteration.
3878 * We could do it here only because we start from the head all
3879 * the times so there is no risk of missing some entries.
3880 * On the other hand, the risk is that we end up with
3881 * a very inconsistent ruleset, so better keep the lock
3882 * around the whole cycle.
3883 *
3884 * XXX this code can be improved by resetting the head of
3885 * the list to point to the default rule, and then freeing
3886 * the old list without the need for a lock.
3887 */
3888
3889 IPFW_WLOCK(&layer3_chain);
3890 layer3_chain.reap = NULL;
3891 free_chain(&layer3_chain, 0 /* keep default rule */);
3892 rule = layer3_chain.reap, layer3_chain.reap = NULL;
3893 IPFW_WUNLOCK(&layer3_chain);
3894 if (layer3_chain.reap != NULL)
3895 reap_rules(rule);
3896 break;
3897
3898 case IP_FW_ADD:
3899 rule = malloc(RULE_MAXSIZE, M_TEMP, M_WAITOK);
3900 error = sooptcopyin(sopt, rule, RULE_MAXSIZE,
3901 sizeof(struct ip_fw) );
3902 if (error == 0)
3903 error = check_ipfw_struct(rule, sopt->sopt_valsize);
3904 if (error == 0) {
3905 error = add_rule(&layer3_chain, rule);
3906 size = RULESIZE(rule);
3907 if (!error && sopt->sopt_dir == SOPT_GET)
3908 error = sooptcopyout(sopt, rule, size);
3909 }
3910 free(rule, M_TEMP);
3911 break;
3912
3913 case IP_FW_DEL:
3914 /*
3915 * IP_FW_DEL is used for deleting single rules or sets,
3916 * and (ab)used to atomically manipulate sets. Argument size
3917 * is used to distinguish between the two:
3918 * sizeof(u_int32_t)
3919 * delete single rule or set of rules,
3920 * or reassign rules (or sets) to a different set.
3921 * 2*sizeof(u_int32_t)
3922 * atomic disable/enable sets.
3923 * first u_int32_t contains sets to be disabled,
3924 * second u_int32_t contains sets to be enabled.
3925 */
3926 error = sooptcopyin(sopt, rulenum,
3927 2*sizeof(u_int32_t), sizeof(u_int32_t));
3928 if (error)
3929 break;
3930 size = sopt->sopt_valsize;
3931 if (size == sizeof(u_int32_t)) /* delete or reassign */
3932 error = del_entry(&layer3_chain, rulenum[0]);
3933 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */
3934 set_disable =
3935 (set_disable | rulenum[0]) & ~rulenum[1] &
3936 ~(1<<RESVD_SET); /* set RESVD_SET always enabled */
3937 else
3938 error = EINVAL;
3939 break;
3940
3941 case IP_FW_ZERO:
3942 case IP_FW_RESETLOG: /* argument is an int, the rule number */
3943 rule_num = 0;
3944 if (sopt->sopt_val != 0) {
3945 error = sooptcopyin(sopt, &rule_num,
3946 sizeof(int), sizeof(int));
3947 if (error)
3948 break;
3949 }
3950 error = zero_entry(&layer3_chain, rule_num,
3951 sopt->sopt_name == IP_FW_RESETLOG);
3952 break;
3953
3954 case IP_FW_TABLE_ADD:
3955 {
3956 ipfw_table_entry ent;
3957
3958 error = sooptcopyin(sopt, &ent,
3959 sizeof(ent), sizeof(ent));
3960 if (error)
3961 break;
3962 error = add_table_entry(&layer3_chain, ent.tbl,
3963 ent.addr, ent.masklen, ent.value);
3964 }
3965 break;
3966
3967 case IP_FW_TABLE_DEL:
3968 {
3969 ipfw_table_entry ent;
3970
3971 error = sooptcopyin(sopt, &ent,
3972 sizeof(ent), sizeof(ent));
3973 if (error)
3974 break;
3975 error = del_table_entry(&layer3_chain, ent.tbl,
3976 ent.addr, ent.masklen);
3977 }
3978 break;
3979
3980 case IP_FW_TABLE_FLUSH:
3981 {
3982 u_int16_t tbl;
3983
3984 error = sooptcopyin(sopt, &tbl,
3985 sizeof(tbl), sizeof(tbl));
3986 if (error)
3987 break;
3988 IPFW_WLOCK(&layer3_chain);
3989 error = flush_table(&layer3_chain, tbl);
3990 IPFW_WUNLOCK(&layer3_chain);
3991 }
3992 break;
3993
3994 case IP_FW_TABLE_GETSIZE:
3995 {
3996 u_int32_t tbl, cnt;
3997
3998 if ((error = sooptcopyin(sopt, &tbl, sizeof(tbl),
3999 sizeof(tbl))))
4000 break;
4001 IPFW_RLOCK(&layer3_chain);
4002 if ((error = count_table(&layer3_chain, tbl, &cnt)))
4003 break;
4004 IPFW_RUNLOCK(&layer3_chain);
4005 error = sooptcopyout(sopt, &cnt, sizeof(cnt));
4006 }
4007 break;
4008
4009 case IP_FW_TABLE_LIST:
4010 {
4011 ipfw_table *tbl;
4012
4013 if (sopt->sopt_valsize < sizeof(*tbl)) {
4014 error = EINVAL;
4015 break;
4016 }
4017 size = sopt->sopt_valsize;
4018 tbl = malloc(size, M_TEMP, M_WAITOK);
4019 if (tbl == NULL) {
4020 error = ENOMEM;
4021 break;
4022 }
4023 error = sooptcopyin(sopt, tbl, size, sizeof(*tbl));
4024 if (error) {
4025 free(tbl, M_TEMP);
4026 break;
4027 }
4028 tbl->size = (size - sizeof(*tbl)) /
4029 sizeof(ipfw_table_entry);
4030 IPFW_WLOCK(&layer3_chain);
4031 error = dump_table(&layer3_chain, tbl);
4032 if (error) {
4033 IPFW_WUNLOCK(&layer3_chain);
4034 free(tbl, M_TEMP);
4035 break;
4036 }
4037 IPFW_WUNLOCK(&layer3_chain);
4038 error = sooptcopyout(sopt, tbl, size);
4039 free(tbl, M_TEMP);
4040 }
4041 break;
4042
4043 default:
4044 printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name);
4045 error = EINVAL;
4046 }
4047
4048 return (error);
4049#undef RULE_MAXSIZE
4050}
4051
4052/**
4053 * dummynet needs a reference to the default rule, because rules can be
4054 * deleted while packets hold a reference to them. When this happens,
4055 * dummynet changes the reference to the default rule (it could well be a
4056 * NULL pointer, but this way we do not need to check for the special
4057 * case, plus here he have info on the default behaviour).
4058 */
4059struct ip_fw *ip_fw_default_rule;
4060
4061/*
4062 * This procedure is only used to handle keepalives. It is invoked
4063 * every dyn_keepalive_period
4064 */
4065static void
4066ipfw_tick(void * __unused unused)
4067{
4068 struct mbuf *m0, *m, *mnext, **mtailp;
4069 int i;
4070 ipfw_dyn_rule *q;
4071
4072 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
4073 goto done;
4074
4075 /*
4076 * We make a chain of packets to go out here -- not deferring
4077 * until after we drop the IPFW dynamic rule lock would result
4078 * in a lock order reversal with the normal packet input -> ipfw
4079 * call stack.
4080 */
4081 m0 = NULL;
4082 mtailp = &m0;
4083 IPFW_DYN_LOCK();
4084 for (i = 0 ; i < curr_dyn_buckets ; i++) {
4085 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
4086 if (q->dyn_type == O_LIMIT_PARENT)
4087 continue;
4088 if (q->id.proto != IPPROTO_TCP)
4089 continue;
4090 if ( (q->state & BOTH_SYN) != BOTH_SYN)
4091 continue;
4092 if (TIME_LEQ( time_uptime+dyn_keepalive_interval,
4093 q->expire))
4094 continue; /* too early */
4095 if (TIME_LEQ(q->expire, time_uptime))
4096 continue; /* too late, rule expired */
4097
4098 *mtailp = send_pkt(&(q->id), q->ack_rev - 1,
4099 q->ack_fwd, TH_SYN);
4100 if (*mtailp != NULL)
4101 mtailp = &(*mtailp)->m_nextpkt;
4102 *mtailp = send_pkt(&(q->id), q->ack_fwd - 1,
4103 q->ack_rev, 0);
4104 if (*mtailp != NULL)
4105 mtailp = &(*mtailp)->m_nextpkt;
4106 }
4107 }
4108 IPFW_DYN_UNLOCK();
4109 for (m = mnext = m0; m != NULL; m = mnext) {
4110 mnext = m->m_nextpkt;
4111 m->m_nextpkt = NULL;
4112 ip_output(m, NULL, NULL, 0, NULL, NULL);
4113 }
4114done:
4115 callout_reset(&ipfw_timeout, dyn_keepalive_period*hz, ipfw_tick, NULL);
4116}
4117
4118int
4119ipfw_init(void)
4120{
4121 struct ip_fw default_rule;
4122 int error;
4123
4124#ifdef INET6
4125 /* Setup IPv6 fw sysctl tree. */
4126 sysctl_ctx_init(&ip6_fw_sysctl_ctx);
4127 ip6_fw_sysctl_tree = SYSCTL_ADD_NODE(&ip6_fw_sysctl_ctx,
4128 SYSCTL_STATIC_CHILDREN(_net_inet6_ip6), OID_AUTO, "fw",
4129 CTLFLAG_RW | CTLFLAG_SECURE, 0, "Firewall");
4130 SYSCTL_ADD_INT(&ip6_fw_sysctl_ctx, SYSCTL_CHILDREN(ip6_fw_sysctl_tree),
4131 OID_AUTO, "deny_unknown_exthdrs", CTLFLAG_RW | CTLFLAG_SECURE,
4132 &fw_deny_unknown_exthdrs, 0,
4133 "Deny packets with unknown IPv6 Extension Headers");
4134#endif
4135
4136 layer3_chain.rules = NULL;
4137 layer3_chain.want_write = 0;
4138 layer3_chain.busy_count = 0;
4139 cv_init(&layer3_chain.cv, "Condition variable for IPFW rw locks");
4140 IPFW_LOCK_INIT(&layer3_chain);
4141 ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule zone",
4142 sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL,
4143 UMA_ALIGN_PTR, 0);
4144 IPFW_DYN_LOCK_INIT();
4145 callout_init(&ipfw_timeout, NET_CALLOUT_MPSAFE);
4146
4147 bzero(&default_rule, sizeof default_rule);
4148
4149 default_rule.act_ofs = 0;
4150 default_rule.rulenum = IPFW_DEFAULT_RULE;
4151 default_rule.cmd_len = 1;
4152 default_rule.set = RESVD_SET;
4153
4154 default_rule.cmd[0].len = 1;
4155 default_rule.cmd[0].opcode =
4156#ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
4157 1 ? O_ACCEPT :
4158#endif
4159 O_DENY;
4160
4161 error = add_rule(&layer3_chain, &default_rule);
4162 if (error != 0) {
4163 printf("ipfw2: error %u initializing default rule "
4164 "(support disabled)\n", error);
4165 IPFW_DYN_LOCK_DESTROY();
4166 IPFW_LOCK_DESTROY(&layer3_chain);
4167 return (error);
4168 }
4169
4170 ip_fw_default_rule = layer3_chain.rules;
4171 printf("ipfw2 (+ipv6) initialized, divert %s, "
4172 "rule-based forwarding "
4173#ifdef IPFIREWALL_FORWARD
4174 "enabled, "
4175#else
4176 "disabled, "
4177#endif
4178 "default to %s, logging ",
4179#ifdef IPDIVERT
4180 "enabled",
4181#else
4182 "loadable",
4183#endif
4184 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny");
4185
4186#ifdef IPFIREWALL_VERBOSE
4187 fw_verbose = 1;
4188#endif
4189#ifdef IPFIREWALL_VERBOSE_LIMIT
4190 verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
4191#endif
4192 if (fw_verbose == 0)
4193 printf("disabled\n");
4194 else if (verbose_limit == 0)
4195 printf("unlimited\n");
4196 else
4197 printf("limited to %d packets/entry by default\n",
4198 verbose_limit);
4199
4200 init_tables(&layer3_chain);
4201 ip_fw_ctl_ptr = ipfw_ctl;
4202 ip_fw_chk_ptr = ipfw_chk;
4203 callout_reset(&ipfw_timeout, hz, ipfw_tick, NULL);
4204
4205 return (0);
4206}
4207
4208void
4209ipfw_destroy(void)
4210{
4211 struct ip_fw *reap;
4212
4213 ip_fw_chk_ptr = NULL;
4214 ip_fw_ctl_ptr = NULL;
4215 callout_drain(&ipfw_timeout);
4216 IPFW_WLOCK(&layer3_chain);
4217 flush_tables(&layer3_chain);
4218 layer3_chain.reap = NULL;
4219 free_chain(&layer3_chain, 1 /* kill default rule */);
4220 reap = layer3_chain.reap, layer3_chain.reap = NULL;
4221 IPFW_WUNLOCK(&layer3_chain);
4222 if (reap != NULL)
4223 reap_rules(reap);
4224 IPFW_DYN_LOCK_DESTROY();
4225 uma_zdestroy(ipfw_dyn_rule_zone);
4226 IPFW_LOCK_DESTROY(&layer3_chain);
4227
4228#ifdef INET6
4229 /* Free IPv6 fw sysctl tree. */
4230 sysctl_ctx_free(&ip6_fw_sysctl_ctx);
4231#endif
4232
4233 printf("IP firewall unloaded\n");
4234}
| 3100 retval = (cmd->opcode == O_NETGRAPH) ?
3101 IP_FW_NETGRAPH : IP_FW_NGTEE;
3102 goto done;
3103
3104 default:
3105 panic("-- unknown opcode %d\n", cmd->opcode);
3106 } /* end of switch() on opcodes */
3107
3108 if (cmd->len & F_NOT)
3109 match = !match;
3110
3111 if (match) {
3112 if (cmd->len & F_OR)
3113 skip_or = 1;
3114 } else {
3115 if (!(cmd->len & F_OR)) /* not an OR block, */
3116 break; /* try next rule */
3117 }
3118
3119 } /* end of inner for, scan opcodes */
3120
3121next_rule:; /* try next rule */
3122
3123 } /* end of outer for, scan rules */
3124 printf("ipfw: ouch!, skip past end of rules, denying packet\n");
3125 IPFW_RUNLOCK(chain);
3126 return (IP_FW_DENY);
3127
3128done:
3129 /* Update statistics */
3130 f->pcnt++;
3131 f->bcnt += pktlen;
3132 f->timestamp = time_uptime;
3133 IPFW_RUNLOCK(chain);
3134 return (retval);
3135
3136pullup_failed:
3137 if (fw_verbose)
3138 printf("ipfw: pullup failed\n");
3139 return (IP_FW_DENY);
3140}
3141
3142/*
3143 * When a rule is added/deleted, clear the next_rule pointers in all rules.
3144 * These will be reconstructed on the fly as packets are matched.
3145 */
3146static void
3147flush_rule_ptrs(struct ip_fw_chain *chain)
3148{
3149 struct ip_fw *rule;
3150
3151 IPFW_WLOCK_ASSERT(chain);
3152
3153 for (rule = chain->rules; rule; rule = rule->next)
3154 rule->next_rule = NULL;
3155}
3156
3157/*
3158 * Add a new rule to the list. Copy the rule into a malloc'ed area, then
3159 * possibly create a rule number and add the rule to the list.
3160 * Update the rule_number in the input struct so the caller knows it as well.
3161 */
3162static int
3163add_rule(struct ip_fw_chain *chain, struct ip_fw *input_rule)
3164{
3165 struct ip_fw *rule, *f, *prev;
3166 int l = RULESIZE(input_rule);
3167
3168 if (chain->rules == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE)
3169 return (EINVAL);
3170
3171 rule = malloc(l, M_IPFW, M_NOWAIT | M_ZERO);
3172 if (rule == NULL)
3173 return (ENOSPC);
3174
3175 bcopy(input_rule, rule, l);
3176
3177 rule->next = NULL;
3178 rule->next_rule = NULL;
3179
3180 rule->pcnt = 0;
3181 rule->bcnt = 0;
3182 rule->timestamp = 0;
3183
3184 IPFW_WLOCK(chain);
3185
3186 if (chain->rules == NULL) { /* default rule */
3187 chain->rules = rule;
3188 goto done;
3189 }
3190
3191 /*
3192 * If rulenum is 0, find highest numbered rule before the
3193 * default rule, and add autoinc_step
3194 */
3195 if (autoinc_step < 1)
3196 autoinc_step = 1;
3197 else if (autoinc_step > 1000)
3198 autoinc_step = 1000;
3199 if (rule->rulenum == 0) {
3200 /*
3201 * locate the highest numbered rule before default
3202 */
3203 for (f = chain->rules; f; f = f->next) {
3204 if (f->rulenum == IPFW_DEFAULT_RULE)
3205 break;
3206 rule->rulenum = f->rulenum;
3207 }
3208 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
3209 rule->rulenum += autoinc_step;
3210 input_rule->rulenum = rule->rulenum;
3211 }
3212
3213 /*
3214 * Now insert the new rule in the right place in the sorted list.
3215 */
3216 for (prev = NULL, f = chain->rules; f; prev = f, f = f->next) {
3217 if (f->rulenum > rule->rulenum) { /* found the location */
3218 if (prev) {
3219 rule->next = f;
3220 prev->next = rule;
3221 } else { /* head insert */
3222 rule->next = chain->rules;
3223 chain->rules = rule;
3224 }
3225 break;
3226 }
3227 }
3228 flush_rule_ptrs(chain);
3229done:
3230 static_count++;
3231 static_len += l;
3232 IPFW_WUNLOCK(chain);
3233 DEB(printf("ipfw: installed rule %d, static count now %d\n",
3234 rule->rulenum, static_count);)
3235 return (0);
3236}
3237
3238/**
3239 * Remove a static rule (including derived * dynamic rules)
3240 * and place it on the ``reap list'' for later reclamation.
3241 * The caller is in charge of clearing rule pointers to avoid
3242 * dangling pointers.
3243 * @return a pointer to the next entry.
3244 * Arguments are not checked, so they better be correct.
3245 */
3246static struct ip_fw *
3247remove_rule(struct ip_fw_chain *chain, struct ip_fw *rule, struct ip_fw *prev)
3248{
3249 struct ip_fw *n;
3250 int l = RULESIZE(rule);
3251
3252 IPFW_WLOCK_ASSERT(chain);
3253
3254 n = rule->next;
3255 IPFW_DYN_LOCK();
3256 remove_dyn_rule(rule, NULL /* force removal */);
3257 IPFW_DYN_UNLOCK();
3258 if (prev == NULL)
3259 chain->rules = n;
3260 else
3261 prev->next = n;
3262 static_count--;
3263 static_len -= l;
3264
3265 rule->next = chain->reap;
3266 chain->reap = rule;
3267
3268 return n;
3269}
3270
3271/**
3272 * Reclaim storage associated with a list of rules. This is
3273 * typically the list created using remove_rule.
3274 */
3275static void
3276reap_rules(struct ip_fw *head)
3277{
3278 struct ip_fw *rule;
3279
3280 while ((rule = head) != NULL) {
3281 head = head->next;
3282 if (DUMMYNET_LOADED)
3283 ip_dn_ruledel_ptr(rule);
3284 free(rule, M_IPFW);
3285 }
3286}
3287
3288/*
3289 * Remove all rules from a chain (except rules in set RESVD_SET
3290 * unless kill_default = 1). The caller is responsible for
3291 * reclaiming storage for the rules left in chain->reap.
3292 */
3293static void
3294free_chain(struct ip_fw_chain *chain, int kill_default)
3295{
3296 struct ip_fw *prev, *rule;
3297
3298 IPFW_WLOCK_ASSERT(chain);
3299
3300 flush_rule_ptrs(chain); /* more efficient to do outside the loop */
3301 for (prev = NULL, rule = chain->rules; rule ; )
3302 if (kill_default || rule->set != RESVD_SET)
3303 rule = remove_rule(chain, rule, prev);
3304 else {
3305 prev = rule;
3306 rule = rule->next;
3307 }
3308}
3309
3310/**
3311 * Remove all rules with given number, and also do set manipulation.
3312 * Assumes chain != NULL && *chain != NULL.
3313 *
3314 * The argument is an u_int32_t. The low 16 bit are the rule or set number,
3315 * the next 8 bits are the new set, the top 8 bits are the command:
3316 *
3317 * 0 delete rules with given number
3318 * 1 delete rules with given set number
3319 * 2 move rules with given number to new set
3320 * 3 move rules with given set number to new set
3321 * 4 swap sets with given numbers
3322 */
3323static int
3324del_entry(struct ip_fw_chain *chain, u_int32_t arg)
3325{
3326 struct ip_fw *prev = NULL, *rule;
3327 u_int16_t rulenum; /* rule or old_set */
3328 u_int8_t cmd, new_set;
3329
3330 rulenum = arg & 0xffff;
3331 cmd = (arg >> 24) & 0xff;
3332 new_set = (arg >> 16) & 0xff;
3333
3334 if (cmd > 4)
3335 return EINVAL;
3336 if (new_set > RESVD_SET)
3337 return EINVAL;
3338 if (cmd == 0 || cmd == 2) {
3339 if (rulenum >= IPFW_DEFAULT_RULE)
3340 return EINVAL;
3341 } else {
3342 if (rulenum > RESVD_SET) /* old_set */
3343 return EINVAL;
3344 }
3345
3346 IPFW_WLOCK(chain);
3347 rule = chain->rules;
3348 chain->reap = NULL;
3349 switch (cmd) {
3350 case 0: /* delete rules with given number */
3351 /*
3352 * locate first rule to delete
3353 */
3354 for (; rule->rulenum < rulenum; prev = rule, rule = rule->next)
3355 ;
3356 if (rule->rulenum != rulenum) {
3357 IPFW_WUNLOCK(chain);
3358 return EINVAL;
3359 }
3360
3361 /*
3362 * flush pointers outside the loop, then delete all matching
3363 * rules. prev remains the same throughout the cycle.
3364 */
3365 flush_rule_ptrs(chain);
3366 while (rule->rulenum == rulenum)
3367 rule = remove_rule(chain, rule, prev);
3368 break;
3369
3370 case 1: /* delete all rules with given set number */
3371 flush_rule_ptrs(chain);
3372 rule = chain->rules;
3373 while (rule->rulenum < IPFW_DEFAULT_RULE)
3374 if (rule->set == rulenum)
3375 rule = remove_rule(chain, rule, prev);
3376 else {
3377 prev = rule;
3378 rule = rule->next;
3379 }
3380 break;
3381
3382 case 2: /* move rules with given number to new set */
3383 rule = chain->rules;
3384 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3385 if (rule->rulenum == rulenum)
3386 rule->set = new_set;
3387 break;
3388
3389 case 3: /* move rules with given set number to new set */
3390 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3391 if (rule->set == rulenum)
3392 rule->set = new_set;
3393 break;
3394
3395 case 4: /* swap two sets */
3396 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3397 if (rule->set == rulenum)
3398 rule->set = new_set;
3399 else if (rule->set == new_set)
3400 rule->set = rulenum;
3401 break;
3402 }
3403 /*
3404 * Look for rules to reclaim. We grab the list before
3405 * releasing the lock then reclaim them w/o the lock to
3406 * avoid a LOR with dummynet.
3407 */
3408 rule = chain->reap;
3409 chain->reap = NULL;
3410 IPFW_WUNLOCK(chain);
3411 if (rule)
3412 reap_rules(rule);
3413 return 0;
3414}
3415
3416/*
3417 * Clear counters for a specific rule.
3418 * The enclosing "table" is assumed locked.
3419 */
3420static void
3421clear_counters(struct ip_fw *rule, int log_only)
3422{
3423 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
3424
3425 if (log_only == 0) {
3426 rule->bcnt = rule->pcnt = 0;
3427 rule->timestamp = 0;
3428 }
3429 if (l->o.opcode == O_LOG)
3430 l->log_left = l->max_log;
3431}
3432
3433/**
3434 * Reset some or all counters on firewall rules.
3435 * @arg frwl is null to clear all entries, or contains a specific
3436 * rule number.
3437 * @arg log_only is 1 if we only want to reset logs, zero otherwise.
3438 */
3439static int
3440zero_entry(struct ip_fw_chain *chain, int rulenum, int log_only)
3441{
3442 struct ip_fw *rule;
3443 char *msg;
3444
3445 IPFW_WLOCK(chain);
3446 if (rulenum == 0) {
3447 norule_counter = 0;
3448 for (rule = chain->rules; rule; rule = rule->next)
3449 clear_counters(rule, log_only);
3450 msg = log_only ? "ipfw: All logging counts reset.\n" :
3451 "ipfw: Accounting cleared.\n";
3452 } else {
3453 int cleared = 0;
3454 /*
3455 * We can have multiple rules with the same number, so we
3456 * need to clear them all.
3457 */
3458 for (rule = chain->rules; rule; rule = rule->next)
3459 if (rule->rulenum == rulenum) {
3460 while (rule && rule->rulenum == rulenum) {
3461 clear_counters(rule, log_only);
3462 rule = rule->next;
3463 }
3464 cleared = 1;
3465 break;
3466 }
3467 if (!cleared) { /* we did not find any matching rules */
3468 IPFW_WUNLOCK(chain);
3469 return (EINVAL);
3470 }
3471 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
3472 "ipfw: Entry %d cleared.\n";
3473 }
3474 IPFW_WUNLOCK(chain);
3475
3476 if (fw_verbose)
3477 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
3478 return (0);
3479}
3480
3481/*
3482 * Check validity of the structure before insert.
3483 * Fortunately rules are simple, so this mostly need to check rule sizes.
3484 */
3485static int
3486check_ipfw_struct(struct ip_fw *rule, int size)
3487{
3488 int l, cmdlen = 0;
3489 int have_action=0;
3490 ipfw_insn *cmd;
3491
3492 if (size < sizeof(*rule)) {
3493 printf("ipfw: rule too short\n");
3494 return (EINVAL);
3495 }
3496 /* first, check for valid size */
3497 l = RULESIZE(rule);
3498 if (l != size) {
3499 printf("ipfw: size mismatch (have %d want %d)\n", size, l);
3500 return (EINVAL);
3501 }
3502 if (rule->act_ofs >= rule->cmd_len) {
3503 printf("ipfw: bogus action offset (%u > %u)\n",
3504 rule->act_ofs, rule->cmd_len - 1);
3505 return (EINVAL);
3506 }
3507 /*
3508 * Now go for the individual checks. Very simple ones, basically only
3509 * instruction sizes.
3510 */
3511 for (l = rule->cmd_len, cmd = rule->cmd ;
3512 l > 0 ; l -= cmdlen, cmd += cmdlen) {
3513 cmdlen = F_LEN(cmd);
3514 if (cmdlen > l) {
3515 printf("ipfw: opcode %d size truncated\n",
3516 cmd->opcode);
3517 return EINVAL;
3518 }
3519 DEB(printf("ipfw: opcode %d\n", cmd->opcode);)
3520 switch (cmd->opcode) {
3521 case O_PROBE_STATE:
3522 case O_KEEP_STATE:
3523 case O_PROTO:
3524 case O_IP_SRC_ME:
3525 case O_IP_DST_ME:
3526 case O_LAYER2:
3527 case O_IN:
3528 case O_FRAG:
3529 case O_DIVERTED:
3530 case O_IPOPT:
3531 case O_IPTOS:
3532 case O_IPPRECEDENCE:
3533 case O_IPVER:
3534 case O_TCPWIN:
3535 case O_TCPFLAGS:
3536 case O_TCPOPTS:
3537 case O_ESTAB:
3538 case O_VERREVPATH:
3539 case O_VERSRCREACH:
3540 case O_ANTISPOOF:
3541 case O_IPSEC:
3542#ifdef INET6
3543 case O_IP6_SRC_ME:
3544 case O_IP6_DST_ME:
3545 case O_EXT_HDR:
3546 case O_IP6:
3547#endif
3548 case O_IP4:
3549 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3550 goto bad_size;
3551 break;
3552
3553 case O_UID:
3554 case O_GID:
3555 case O_JAIL:
3556 case O_IP_SRC:
3557 case O_IP_DST:
3558 case O_TCPSEQ:
3559 case O_TCPACK:
3560 case O_PROB:
3561 case O_ICMPTYPE:
3562 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3563 goto bad_size;
3564 break;
3565
3566 case O_LIMIT:
3567 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
3568 goto bad_size;
3569 break;
3570
3571 case O_LOG:
3572 if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
3573 goto bad_size;
3574
3575 ((ipfw_insn_log *)cmd)->log_left =
3576 ((ipfw_insn_log *)cmd)->max_log;
3577
3578 break;
3579
3580 case O_IP_SRC_MASK:
3581 case O_IP_DST_MASK:
3582 /* only odd command lengths */
3583 if ( !(cmdlen & 1) || cmdlen > 31)
3584 goto bad_size;
3585 break;
3586
3587 case O_IP_SRC_SET:
3588 case O_IP_DST_SET:
3589 if (cmd->arg1 == 0 || cmd->arg1 > 256) {
3590 printf("ipfw: invalid set size %d\n",
3591 cmd->arg1);
3592 return EINVAL;
3593 }
3594 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3595 (cmd->arg1+31)/32 )
3596 goto bad_size;
3597 break;
3598
3599 case O_IP_SRC_LOOKUP:
3600 case O_IP_DST_LOOKUP:
3601 if (cmd->arg1 >= IPFW_TABLES_MAX) {
3602 printf("ipfw: invalid table number %d\n",
3603 cmd->arg1);
3604 return (EINVAL);
3605 }
3606 if (cmdlen != F_INSN_SIZE(ipfw_insn) &&
3607 cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3608 goto bad_size;
3609 break;
3610
3611 case O_MACADDR2:
3612 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
3613 goto bad_size;
3614 break;
3615
3616 case O_NOP:
3617 case O_IPID:
3618 case O_IPTTL:
3619 case O_IPLEN:
3620 case O_TCPDATALEN:
3621 if (cmdlen < 1 || cmdlen > 31)
3622 goto bad_size;
3623 break;
3624
3625 case O_MAC_TYPE:
3626 case O_IP_SRCPORT:
3627 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
3628 if (cmdlen < 2 || cmdlen > 31)
3629 goto bad_size;
3630 break;
3631
3632 case O_RECV:
3633 case O_XMIT:
3634 case O_VIA:
3635 if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
3636 goto bad_size;
3637 break;
3638
3639 case O_ALTQ:
3640 if (cmdlen != F_INSN_SIZE(ipfw_insn_altq))
3641 goto bad_size;
3642 break;
3643
3644 case O_PIPE:
3645 case O_QUEUE:
3646 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3647 goto bad_size;
3648 goto check_action;
3649
3650 case O_FORWARD_IP:
3651#ifdef IPFIREWALL_FORWARD
3652 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
3653 goto bad_size;
3654 goto check_action;
3655#else
3656 return EINVAL;
3657#endif
3658
3659 case O_DIVERT:
3660 case O_TEE:
3661 if (ip_divert_ptr == NULL)
3662 return EINVAL;
3663 else
3664 goto check_size;
3665 case O_NETGRAPH:
3666 case O_NGTEE:
3667 if (!NG_IPFW_LOADED)
3668 return EINVAL;
3669 else
3670 goto check_size;
3671 case O_FORWARD_MAC: /* XXX not implemented yet */
3672 case O_CHECK_STATE:
3673 case O_COUNT:
3674 case O_ACCEPT:
3675 case O_DENY:
3676 case O_REJECT:
3677#ifdef INET6
3678 case O_UNREACH6:
3679#endif
3680 case O_SKIPTO:
3681check_size:
3682 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3683 goto bad_size;
3684check_action:
3685 if (have_action) {
3686 printf("ipfw: opcode %d, multiple actions"
3687 " not allowed\n",
3688 cmd->opcode);
3689 return EINVAL;
3690 }
3691 have_action = 1;
3692 if (l != cmdlen) {
3693 printf("ipfw: opcode %d, action must be"
3694 " last opcode\n",
3695 cmd->opcode);
3696 return EINVAL;
3697 }
3698 break;
3699#ifdef INET6
3700 case O_IP6_SRC:
3701 case O_IP6_DST:
3702 if (cmdlen != F_INSN_SIZE(struct in6_addr) +
3703 F_INSN_SIZE(ipfw_insn))
3704 goto bad_size;
3705 break;
3706
3707 case O_FLOW6ID:
3708 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3709 ((ipfw_insn_u32 *)cmd)->o.arg1)
3710 goto bad_size;
3711 break;
3712
3713 case O_IP6_SRC_MASK:
3714 case O_IP6_DST_MASK:
3715 if ( !(cmdlen & 1) || cmdlen > 127)
3716 goto bad_size;
3717 break;
3718 case O_ICMP6TYPE:
3719 if( cmdlen != F_INSN_SIZE( ipfw_insn_icmp6 ) )
3720 goto bad_size;
3721 break;
3722#endif
3723
3724 default:
3725 switch (cmd->opcode) {
3726#ifndef INET6
3727 case O_IP6_SRC_ME:
3728 case O_IP6_DST_ME:
3729 case O_EXT_HDR:
3730 case O_IP6:
3731 case O_UNREACH6:
3732 case O_IP6_SRC:
3733 case O_IP6_DST:
3734 case O_FLOW6ID:
3735 case O_IP6_SRC_MASK:
3736 case O_IP6_DST_MASK:
3737 case O_ICMP6TYPE:
3738 printf("ipfw: no IPv6 support in kernel\n");
3739 return EPROTONOSUPPORT;
3740#endif
3741 default:
3742 printf("ipfw: opcode %d, unknown opcode\n",
3743 cmd->opcode);
3744 return EINVAL;
3745 }
3746 }
3747 }
3748 if (have_action == 0) {
3749 printf("ipfw: missing action\n");
3750 return EINVAL;
3751 }
3752 return 0;
3753
3754bad_size:
3755 printf("ipfw: opcode %d size %d wrong\n",
3756 cmd->opcode, cmdlen);
3757 return EINVAL;
3758}
3759
3760/*
3761 * Copy the static and dynamic rules to the supplied buffer
3762 * and return the amount of space actually used.
3763 */
3764static size_t
3765ipfw_getrules(struct ip_fw_chain *chain, void *buf, size_t space)
3766{
3767 char *bp = buf;
3768 char *ep = bp + space;
3769 struct ip_fw *rule;
3770 int i;
3771
3772 /* XXX this can take a long time and locking will block packet flow */
3773 IPFW_RLOCK(chain);
3774 for (rule = chain->rules; rule ; rule = rule->next) {
3775 /*
3776 * Verify the entry fits in the buffer in case the
3777 * rules changed between calculating buffer space and
3778 * now. This would be better done using a generation
3779 * number but should suffice for now.
3780 */
3781 i = RULESIZE(rule);
3782 if (bp + i <= ep) {
3783 bcopy(rule, bp, i);
3784 bcopy(&set_disable, &(((struct ip_fw *)bp)->next_rule),
3785 sizeof(set_disable));
3786 bp += i;
3787 }
3788 }
3789 IPFW_RUNLOCK(chain);
3790 if (ipfw_dyn_v) {
3791 ipfw_dyn_rule *p, *last = NULL;
3792
3793 IPFW_DYN_LOCK();
3794 for (i = 0 ; i < curr_dyn_buckets; i++)
3795 for (p = ipfw_dyn_v[i] ; p != NULL; p = p->next) {
3796 if (bp + sizeof *p <= ep) {
3797 ipfw_dyn_rule *dst =
3798 (ipfw_dyn_rule *)bp;
3799 bcopy(p, dst, sizeof *p);
3800 bcopy(&(p->rule->rulenum), &(dst->rule),
3801 sizeof(p->rule->rulenum));
3802 /*
3803 * store a non-null value in "next".
3804 * The userland code will interpret a
3805 * NULL here as a marker
3806 * for the last dynamic rule.
3807 */
3808 bcopy(&dst, &dst->next, sizeof(dst));
3809 last = dst;
3810 dst->expire =
3811 TIME_LEQ(dst->expire, time_uptime) ?
3812 0 : dst->expire - time_uptime ;
3813 bp += sizeof(ipfw_dyn_rule);
3814 }
3815 }
3816 IPFW_DYN_UNLOCK();
3817 if (last != NULL) /* mark last dynamic rule */
3818 bzero(&last->next, sizeof(last));
3819 }
3820 return (bp - (char *)buf);
3821}
3822
3823
3824/**
3825 * {set|get}sockopt parser.
3826 */
3827static int
3828ipfw_ctl(struct sockopt *sopt)
3829{
3830#define RULE_MAXSIZE (256*sizeof(u_int32_t))
3831 int error, rule_num;
3832 size_t size;
3833 struct ip_fw *buf, *rule;
3834 u_int32_t rulenum[2];
3835
3836 error = suser(sopt->sopt_td);
3837 if (error)
3838 return (error);
3839
3840 /*
3841 * Disallow modifications in really-really secure mode, but still allow
3842 * the logging counters to be reset.
3843 */
3844 if (sopt->sopt_name == IP_FW_ADD ||
3845 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
3846#if __FreeBSD_version >= 500034
3847 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
3848 if (error)
3849 return (error);
3850#else /* FreeBSD 4.x */
3851 if (securelevel >= 3)
3852 return (EPERM);
3853#endif
3854 }
3855
3856 error = 0;
3857
3858 switch (sopt->sopt_name) {
3859 case IP_FW_GET:
3860 /*
3861 * pass up a copy of the current rules. Static rules
3862 * come first (the last of which has number IPFW_DEFAULT_RULE),
3863 * followed by a possibly empty list of dynamic rule.
3864 * The last dynamic rule has NULL in the "next" field.
3865 *
3866 * Note that the calculated size is used to bound the
3867 * amount of data returned to the user. The rule set may
3868 * change between calculating the size and returning the
3869 * data in which case we'll just return what fits.
3870 */
3871 size = static_len; /* size of static rules */
3872 if (ipfw_dyn_v) /* add size of dyn.rules */
3873 size += (dyn_count * sizeof(ipfw_dyn_rule));
3874
3875 /*
3876 * XXX todo: if the user passes a short length just to know
3877 * how much room is needed, do not bother filling up the
3878 * buffer, just jump to the sooptcopyout.
3879 */
3880 buf = malloc(size, M_TEMP, M_WAITOK);
3881 error = sooptcopyout(sopt, buf,
3882 ipfw_getrules(&layer3_chain, buf, size));
3883 free(buf, M_TEMP);
3884 break;
3885
3886 case IP_FW_FLUSH:
3887 /*
3888 * Normally we cannot release the lock on each iteration.
3889 * We could do it here only because we start from the head all
3890 * the times so there is no risk of missing some entries.
3891 * On the other hand, the risk is that we end up with
3892 * a very inconsistent ruleset, so better keep the lock
3893 * around the whole cycle.
3894 *
3895 * XXX this code can be improved by resetting the head of
3896 * the list to point to the default rule, and then freeing
3897 * the old list without the need for a lock.
3898 */
3899
3900 IPFW_WLOCK(&layer3_chain);
3901 layer3_chain.reap = NULL;
3902 free_chain(&layer3_chain, 0 /* keep default rule */);
3903 rule = layer3_chain.reap, layer3_chain.reap = NULL;
3904 IPFW_WUNLOCK(&layer3_chain);
3905 if (layer3_chain.reap != NULL)
3906 reap_rules(rule);
3907 break;
3908
3909 case IP_FW_ADD:
3910 rule = malloc(RULE_MAXSIZE, M_TEMP, M_WAITOK);
3911 error = sooptcopyin(sopt, rule, RULE_MAXSIZE,
3912 sizeof(struct ip_fw) );
3913 if (error == 0)
3914 error = check_ipfw_struct(rule, sopt->sopt_valsize);
3915 if (error == 0) {
3916 error = add_rule(&layer3_chain, rule);
3917 size = RULESIZE(rule);
3918 if (!error && sopt->sopt_dir == SOPT_GET)
3919 error = sooptcopyout(sopt, rule, size);
3920 }
3921 free(rule, M_TEMP);
3922 break;
3923
3924 case IP_FW_DEL:
3925 /*
3926 * IP_FW_DEL is used for deleting single rules or sets,
3927 * and (ab)used to atomically manipulate sets. Argument size
3928 * is used to distinguish between the two:
3929 * sizeof(u_int32_t)
3930 * delete single rule or set of rules,
3931 * or reassign rules (or sets) to a different set.
3932 * 2*sizeof(u_int32_t)
3933 * atomic disable/enable sets.
3934 * first u_int32_t contains sets to be disabled,
3935 * second u_int32_t contains sets to be enabled.
3936 */
3937 error = sooptcopyin(sopt, rulenum,
3938 2*sizeof(u_int32_t), sizeof(u_int32_t));
3939 if (error)
3940 break;
3941 size = sopt->sopt_valsize;
3942 if (size == sizeof(u_int32_t)) /* delete or reassign */
3943 error = del_entry(&layer3_chain, rulenum[0]);
3944 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */
3945 set_disable =
3946 (set_disable | rulenum[0]) & ~rulenum[1] &
3947 ~(1<<RESVD_SET); /* set RESVD_SET always enabled */
3948 else
3949 error = EINVAL;
3950 break;
3951
3952 case IP_FW_ZERO:
3953 case IP_FW_RESETLOG: /* argument is an int, the rule number */
3954 rule_num = 0;
3955 if (sopt->sopt_val != 0) {
3956 error = sooptcopyin(sopt, &rule_num,
3957 sizeof(int), sizeof(int));
3958 if (error)
3959 break;
3960 }
3961 error = zero_entry(&layer3_chain, rule_num,
3962 sopt->sopt_name == IP_FW_RESETLOG);
3963 break;
3964
3965 case IP_FW_TABLE_ADD:
3966 {
3967 ipfw_table_entry ent;
3968
3969 error = sooptcopyin(sopt, &ent,
3970 sizeof(ent), sizeof(ent));
3971 if (error)
3972 break;
3973 error = add_table_entry(&layer3_chain, ent.tbl,
3974 ent.addr, ent.masklen, ent.value);
3975 }
3976 break;
3977
3978 case IP_FW_TABLE_DEL:
3979 {
3980 ipfw_table_entry ent;
3981
3982 error = sooptcopyin(sopt, &ent,
3983 sizeof(ent), sizeof(ent));
3984 if (error)
3985 break;
3986 error = del_table_entry(&layer3_chain, ent.tbl,
3987 ent.addr, ent.masklen);
3988 }
3989 break;
3990
3991 case IP_FW_TABLE_FLUSH:
3992 {
3993 u_int16_t tbl;
3994
3995 error = sooptcopyin(sopt, &tbl,
3996 sizeof(tbl), sizeof(tbl));
3997 if (error)
3998 break;
3999 IPFW_WLOCK(&layer3_chain);
4000 error = flush_table(&layer3_chain, tbl);
4001 IPFW_WUNLOCK(&layer3_chain);
4002 }
4003 break;
4004
4005 case IP_FW_TABLE_GETSIZE:
4006 {
4007 u_int32_t tbl, cnt;
4008
4009 if ((error = sooptcopyin(sopt, &tbl, sizeof(tbl),
4010 sizeof(tbl))))
4011 break;
4012 IPFW_RLOCK(&layer3_chain);
4013 if ((error = count_table(&layer3_chain, tbl, &cnt)))
4014 break;
4015 IPFW_RUNLOCK(&layer3_chain);
4016 error = sooptcopyout(sopt, &cnt, sizeof(cnt));
4017 }
4018 break;
4019
4020 case IP_FW_TABLE_LIST:
4021 {
4022 ipfw_table *tbl;
4023
4024 if (sopt->sopt_valsize < sizeof(*tbl)) {
4025 error = EINVAL;
4026 break;
4027 }
4028 size = sopt->sopt_valsize;
4029 tbl = malloc(size, M_TEMP, M_WAITOK);
4030 if (tbl == NULL) {
4031 error = ENOMEM;
4032 break;
4033 }
4034 error = sooptcopyin(sopt, tbl, size, sizeof(*tbl));
4035 if (error) {
4036 free(tbl, M_TEMP);
4037 break;
4038 }
4039 tbl->size = (size - sizeof(*tbl)) /
4040 sizeof(ipfw_table_entry);
4041 IPFW_WLOCK(&layer3_chain);
4042 error = dump_table(&layer3_chain, tbl);
4043 if (error) {
4044 IPFW_WUNLOCK(&layer3_chain);
4045 free(tbl, M_TEMP);
4046 break;
4047 }
4048 IPFW_WUNLOCK(&layer3_chain);
4049 error = sooptcopyout(sopt, tbl, size);
4050 free(tbl, M_TEMP);
4051 }
4052 break;
4053
4054 default:
4055 printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name);
4056 error = EINVAL;
4057 }
4058
4059 return (error);
4060#undef RULE_MAXSIZE
4061}
4062
4063/**
4064 * dummynet needs a reference to the default rule, because rules can be
4065 * deleted while packets hold a reference to them. When this happens,
4066 * dummynet changes the reference to the default rule (it could well be a
4067 * NULL pointer, but this way we do not need to check for the special
4068 * case, plus here he have info on the default behaviour).
4069 */
4070struct ip_fw *ip_fw_default_rule;
4071
4072/*
4073 * This procedure is only used to handle keepalives. It is invoked
4074 * every dyn_keepalive_period
4075 */
4076static void
4077ipfw_tick(void * __unused unused)
4078{
4079 struct mbuf *m0, *m, *mnext, **mtailp;
4080 int i;
4081 ipfw_dyn_rule *q;
4082
4083 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
4084 goto done;
4085
4086 /*
4087 * We make a chain of packets to go out here -- not deferring
4088 * until after we drop the IPFW dynamic rule lock would result
4089 * in a lock order reversal with the normal packet input -> ipfw
4090 * call stack.
4091 */
4092 m0 = NULL;
4093 mtailp = &m0;
4094 IPFW_DYN_LOCK();
4095 for (i = 0 ; i < curr_dyn_buckets ; i++) {
4096 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
4097 if (q->dyn_type == O_LIMIT_PARENT)
4098 continue;
4099 if (q->id.proto != IPPROTO_TCP)
4100 continue;
4101 if ( (q->state & BOTH_SYN) != BOTH_SYN)
4102 continue;
4103 if (TIME_LEQ( time_uptime+dyn_keepalive_interval,
4104 q->expire))
4105 continue; /* too early */
4106 if (TIME_LEQ(q->expire, time_uptime))
4107 continue; /* too late, rule expired */
4108
4109 *mtailp = send_pkt(&(q->id), q->ack_rev - 1,
4110 q->ack_fwd, TH_SYN);
4111 if (*mtailp != NULL)
4112 mtailp = &(*mtailp)->m_nextpkt;
4113 *mtailp = send_pkt(&(q->id), q->ack_fwd - 1,
4114 q->ack_rev, 0);
4115 if (*mtailp != NULL)
4116 mtailp = &(*mtailp)->m_nextpkt;
4117 }
4118 }
4119 IPFW_DYN_UNLOCK();
4120 for (m = mnext = m0; m != NULL; m = mnext) {
4121 mnext = m->m_nextpkt;
4122 m->m_nextpkt = NULL;
4123 ip_output(m, NULL, NULL, 0, NULL, NULL);
4124 }
4125done:
4126 callout_reset(&ipfw_timeout, dyn_keepalive_period*hz, ipfw_tick, NULL);
4127}
4128
4129int
4130ipfw_init(void)
4131{
4132 struct ip_fw default_rule;
4133 int error;
4134
4135#ifdef INET6
4136 /* Setup IPv6 fw sysctl tree. */
4137 sysctl_ctx_init(&ip6_fw_sysctl_ctx);
4138 ip6_fw_sysctl_tree = SYSCTL_ADD_NODE(&ip6_fw_sysctl_ctx,
4139 SYSCTL_STATIC_CHILDREN(_net_inet6_ip6), OID_AUTO, "fw",
4140 CTLFLAG_RW | CTLFLAG_SECURE, 0, "Firewall");
4141 SYSCTL_ADD_INT(&ip6_fw_sysctl_ctx, SYSCTL_CHILDREN(ip6_fw_sysctl_tree),
4142 OID_AUTO, "deny_unknown_exthdrs", CTLFLAG_RW | CTLFLAG_SECURE,
4143 &fw_deny_unknown_exthdrs, 0,
4144 "Deny packets with unknown IPv6 Extension Headers");
4145#endif
4146
4147 layer3_chain.rules = NULL;
4148 layer3_chain.want_write = 0;
4149 layer3_chain.busy_count = 0;
4150 cv_init(&layer3_chain.cv, "Condition variable for IPFW rw locks");
4151 IPFW_LOCK_INIT(&layer3_chain);
4152 ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule zone",
4153 sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL,
4154 UMA_ALIGN_PTR, 0);
4155 IPFW_DYN_LOCK_INIT();
4156 callout_init(&ipfw_timeout, NET_CALLOUT_MPSAFE);
4157
4158 bzero(&default_rule, sizeof default_rule);
4159
4160 default_rule.act_ofs = 0;
4161 default_rule.rulenum = IPFW_DEFAULT_RULE;
4162 default_rule.cmd_len = 1;
4163 default_rule.set = RESVD_SET;
4164
4165 default_rule.cmd[0].len = 1;
4166 default_rule.cmd[0].opcode =
4167#ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
4168 1 ? O_ACCEPT :
4169#endif
4170 O_DENY;
4171
4172 error = add_rule(&layer3_chain, &default_rule);
4173 if (error != 0) {
4174 printf("ipfw2: error %u initializing default rule "
4175 "(support disabled)\n", error);
4176 IPFW_DYN_LOCK_DESTROY();
4177 IPFW_LOCK_DESTROY(&layer3_chain);
4178 return (error);
4179 }
4180
4181 ip_fw_default_rule = layer3_chain.rules;
4182 printf("ipfw2 (+ipv6) initialized, divert %s, "
4183 "rule-based forwarding "
4184#ifdef IPFIREWALL_FORWARD
4185 "enabled, "
4186#else
4187 "disabled, "
4188#endif
4189 "default to %s, logging ",
4190#ifdef IPDIVERT
4191 "enabled",
4192#else
4193 "loadable",
4194#endif
4195 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny");
4196
4197#ifdef IPFIREWALL_VERBOSE
4198 fw_verbose = 1;
4199#endif
4200#ifdef IPFIREWALL_VERBOSE_LIMIT
4201 verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
4202#endif
4203 if (fw_verbose == 0)
4204 printf("disabled\n");
4205 else if (verbose_limit == 0)
4206 printf("unlimited\n");
4207 else
4208 printf("limited to %d packets/entry by default\n",
4209 verbose_limit);
4210
4211 init_tables(&layer3_chain);
4212 ip_fw_ctl_ptr = ipfw_ctl;
4213 ip_fw_chk_ptr = ipfw_chk;
4214 callout_reset(&ipfw_timeout, hz, ipfw_tick, NULL);
4215
4216 return (0);
4217}
4218
4219void
4220ipfw_destroy(void)
4221{
4222 struct ip_fw *reap;
4223
4224 ip_fw_chk_ptr = NULL;
4225 ip_fw_ctl_ptr = NULL;
4226 callout_drain(&ipfw_timeout);
4227 IPFW_WLOCK(&layer3_chain);
4228 flush_tables(&layer3_chain);
4229 layer3_chain.reap = NULL;
4230 free_chain(&layer3_chain, 1 /* kill default rule */);
4231 reap = layer3_chain.reap, layer3_chain.reap = NULL;
4232 IPFW_WUNLOCK(&layer3_chain);
4233 if (reap != NULL)
4234 reap_rules(reap);
4235 IPFW_DYN_LOCK_DESTROY();
4236 uma_zdestroy(ipfw_dyn_rule_zone);
4237 IPFW_LOCK_DESTROY(&layer3_chain);
4238
4239#ifdef INET6
4240 /* Free IPv6 fw sysctl tree. */
4241 sysctl_ctx_free(&ip6_fw_sysctl_ctx);
4242#endif
4243
4244 printf("IP firewall unloaded\n");
4245}
|