4 * Copyright (C) 2012 by Darren Reed.
6 * See the IPFILTER.LICENCE file for details on licencing.
8 * Copyright 2008 Sun Microsystems.
12 #if defined(KERNEL) || defined(_KERNEL)
18 #include <sys/errno.h>
19 #include <sys/types.h>
20 #include <sys/param.h>
22 #if defined(_KERNEL) && defined(__FreeBSD_version) && \
23 (__FreeBSD_version >= 400000) && !defined(KLD_MODULE)
24 #include "opt_inet6.h"
26 #if !defined(_KERNEL) && !defined(__KERNEL__)
37 #if defined(_KERNEL) && (__FreeBSD_version >= 220000)
38 # include <sys/filio.h>
39 # include <sys/fcntl.h>
41 # include <sys/ioctl.h>
45 # include <sys/protosw.h>
47 #include <sys/socket.h>
49 # include <sys/systm.h>
50 # if !defined(__SVR4) && !defined(__svr4__)
51 # include <sys/mbuf.h>
54 #if defined(__SVR4) || defined(__svr4__)
55 # include <sys/filio.h>
56 # include <sys/byteorder.h>
58 # include <sys/dditypes.h>
60 # include <sys/stream.h>
61 # include <sys/kmem.h>
68 #include <netinet/in.h>
69 #include <netinet/in_systm.h>
70 #include <netinet/ip.h>
71 #include <netinet/tcp.h>
72 #if !defined(__hpux) && !defined(linux)
73 # include <netinet/tcp_fsm.h>
75 #include <netinet/udp.h>
76 #include <netinet/ip_icmp.h>
80 #include "netinet/ip_compat.h"
81 #include "netinet/ip_fil.h"
82 #include "netinet/ip_nat.h"
83 #include "netinet/ip_frag.h"
84 #include "netinet/ip_state.h"
85 #include "netinet/ip_proxy.h"
86 #include "netinet/ip_lookup.h"
87 #include "netinet/ip_dstlist.h"
88 #include "netinet/ip_sync.h"
90 #include <netinet/icmp6.h>
92 #if FREEBSD_GE_REV(300000)
93 # include <sys/malloc.h>
94 # if defined(_KERNEL) && !defined(IPFILTER_LKM)
95 # include <sys/libkern.h>
96 # include <sys/systm.h>
103 static const char sccsid[] = "@(#)ip_state.c 1.8 6/5/96 (C) 1993-2000 Darren Reed";
104 static const char rcsid[] = "@(#)$Id$";
108 static ipftuneable_t ipf_state_tuneables[] = {
109 { { (void *)offsetof(ipf_state_softc_t, ipf_state_max) },
110 "state_max", 1, 0x7fffffff,
111 stsizeof(ipf_state_softc_t, ipf_state_max),
113 { { (void *)offsetof(ipf_state_softc_t, ipf_state_size) },
114 "state_size", 1, 0x7fffffff,
115 stsizeof(ipf_state_softc_t, ipf_state_size),
116 0, NULL, ipf_state_rehash },
117 { { (void *)offsetof(ipf_state_softc_t, ipf_state_lock) },
119 stsizeof(ipf_state_softc_t, ipf_state_lock),
120 IPFT_RDONLY, NULL, NULL },
121 { { (void *)offsetof(ipf_state_softc_t, ipf_state_maxbucket) },
122 "state_maxbucket", 1, 0x7fffffff,
123 stsizeof(ipf_state_softc_t, ipf_state_maxbucket),
125 { { (void *)offsetof(ipf_state_softc_t, ipf_state_logging) },
126 "state_logging",0, 1,
127 stsizeof(ipf_state_softc_t, ipf_state_logging),
129 { { (void *)offsetof(ipf_state_softc_t, ipf_state_wm_high) },
130 "state_wm_high",2, 100,
131 stsizeof(ipf_state_softc_t, ipf_state_wm_high),
133 { { (void *)offsetof(ipf_state_softc_t, ipf_state_wm_low) },
134 "state_wm_low", 1, 99,
135 stsizeof(ipf_state_softc_t, ipf_state_wm_low),
137 { { (void *)offsetof(ipf_state_softc_t, ipf_state_wm_freq) },
138 "state_wm_freq",2, 999999,
139 stsizeof(ipf_state_softc_t, ipf_state_wm_freq),
147 #define SINCL(x) ATOMIC_INCL(softs->x)
148 #define SBUMP(x) (softs->x)++
149 #define SBUMPD(x, y) do { (softs->x.y)++; DT(y); } while (0)
150 #define SBUMPDX(x, y, z)do { (softs->x.y)++; DT(z); } while (0)
153 static ipstate_t *ipf_checkicmp6matchingstate __P((fr_info_t *));
155 static int ipf_allowstateicmp __P((fr_info_t *, ipstate_t *, i6addr_t *));
156 static ipstate_t *ipf_matchsrcdst __P((fr_info_t *, ipstate_t *, i6addr_t *,
157 i6addr_t *, tcphdr_t *, u_32_t));
158 static ipstate_t *ipf_checkicmpmatchingstate __P((fr_info_t *));
159 static int ipf_state_flush_entry __P((ipf_main_softc_t *, void *));
160 static ips_stat_t *ipf_state_stats __P((ipf_main_softc_t *));
161 static int ipf_state_del __P((ipf_main_softc_t *, ipstate_t *, int));
162 static int ipf_state_remove __P((ipf_main_softc_t *, caddr_t));
163 static int ipf_state_match __P((ipstate_t *is1, ipstate_t *is2));
164 static int ipf_state_matchaddresses __P((ipstate_t *is1, ipstate_t *is2));
165 static int ipf_state_matchipv4addrs __P((ipstate_t *is1, ipstate_t *is2));
166 static int ipf_state_matchipv6addrs __P((ipstate_t *is1, ipstate_t *is2));
167 static int ipf_state_matchisps __P((ipstate_t *is1, ipstate_t *is2));
168 static int ipf_state_matchports __P((udpinfo_t *is1, udpinfo_t *is2));
169 static int ipf_state_matcharray __P((ipstate_t *, int *, u_long));
170 static void ipf_ipsmove __P((ipf_state_softc_t *, ipstate_t *, u_int));
171 static int ipf_state_tcp __P((ipf_main_softc_t *, ipf_state_softc_t *,
172 fr_info_t *, tcphdr_t *, ipstate_t *));
173 static int ipf_tcpoptions __P((ipf_state_softc_t *, fr_info_t *,
174 tcphdr_t *, tcpdata_t *));
175 static ipstate_t *ipf_state_clone __P((fr_info_t *, tcphdr_t *, ipstate_t *));
176 static void ipf_fixinisn __P((fr_info_t *, ipstate_t *));
177 static void ipf_fixoutisn __P((fr_info_t *, ipstate_t *));
178 static void ipf_checknewisn __P((fr_info_t *, ipstate_t *));
179 static int ipf_state_iter __P((ipf_main_softc_t *, ipftoken_t *,
180 ipfgeniter_t *, ipfobj_t *));
181 static int ipf_state_gettable __P((ipf_main_softc_t *, ipf_state_softc_t *,
183 static int ipf_state_tcpinwindow __P((struct fr_info *, struct tcpdata *,
184 struct tcpdata *, tcphdr_t *, int));
186 static int ipf_state_getent __P((ipf_main_softc_t *, ipf_state_softc_t *,
188 static int ipf_state_putent __P((ipf_main_softc_t *, ipf_state_softc_t *,
191 #define ONE_DAY IPF_TTLVAL(1 * 86400) /* 1 day */
192 #define FIVE_DAYS (5 * ONE_DAY)
193 #define DOUBLE_HASH(x) (((x) + softs->ipf_state_seed[(x) % \
194 softs->ipf_state_size]) % softs->ipf_state_size)
197 /* ------------------------------------------------------------------------ */
198 /* Function: ipf_state_main_load */
199 /* Returns: int - 0 == success, -1 == failure */
200 /* Parameters: Nil */
202 /* A null-op function that exists as a placeholder so that the flow in */
203 /* other functions is obvious. */
204 /* ------------------------------------------------------------------------ */
206 ipf_state_main_load()
212 /* ------------------------------------------------------------------------ */
213 /* Function: ipf_state_main_unload */
214 /* Returns: int - 0 == success, -1 == failure */
215 /* Parameters: Nil */
217 /* A null-op function that exists as a placeholder so that the flow in */
218 /* other functions is obvious. */
219 /* ------------------------------------------------------------------------ */
221 ipf_state_main_unload()
227 /* ------------------------------------------------------------------------ */
228 /* Function: ipf_state_soft_create */
229 /* Returns: void * - NULL = failure, else pointer to soft context */
230 /* Parameters: softc(I) - pointer to soft context main structure */
232 /* Create a new state soft context structure and populate it with the list */
233 /* of tunables and other default settings. */
234 /* ------------------------------------------------------------------------ */
236 ipf_state_soft_create(softc)
237 ipf_main_softc_t *softc;
239 ipf_state_softc_t *softs;
241 KMALLOC(softs, ipf_state_softc_t *);
245 bzero((char *)softs, sizeof(*softs));
247 softs->ipf_state_tune = ipf_tune_array_copy(softs,
248 sizeof(ipf_state_tuneables),
249 ipf_state_tuneables);
250 if (softs->ipf_state_tune == NULL) {
251 ipf_state_soft_destroy(softc, softs);
254 if (ipf_tune_array_link(softc, softs->ipf_state_tune) == -1) {
255 ipf_state_soft_destroy(softc, softs);
260 softs->ipf_state_logging = 1;
262 softs->ipf_state_logging = 0;
264 softs->ipf_state_size = IPSTATE_SIZE,
265 softs->ipf_state_maxbucket = 0;
266 softs->ipf_state_wm_freq = IPF_TTLVAL(10);
267 softs->ipf_state_max = IPSTATE_MAX;
268 softs->ipf_state_wm_last = 0;
269 softs->ipf_state_wm_high = 99;
270 softs->ipf_state_wm_low = 90;
271 softs->ipf_state_inited = 0;
272 softs->ipf_state_lock = 0;
273 softs->ipf_state_doflush = 0;
279 /* ------------------------------------------------------------------------ */
280 /* Function: ipf_state_soft_destroy */
282 /* Parameters: softc(I) - pointer to soft context main structure */
283 /* arg(I) - pointer to local context to use */
285 /* Undo only what we did in soft create: unlink and free the tunables and */
286 /* free the soft context structure itself. */
287 /* ------------------------------------------------------------------------ */
289 ipf_state_soft_destroy(softc, arg)
290 ipf_main_softc_t *softc;
293 ipf_state_softc_t *softs = arg;
295 if (softs->ipf_state_tune != NULL) {
296 ipf_tune_array_unlink(softc, softs->ipf_state_tune);
297 KFREES(softs->ipf_state_tune, sizeof(ipf_state_tuneables));
298 softs->ipf_state_tune = NULL;
305 ipf_state_seed_alloc(u_int state_size, u_int state_max)
309 KMALLOCS(state_seed, u_long *, state_size * sizeof(*state_seed));
310 if (state_seed == NULL)
313 for (i = 0; i < state_size; i++) {
315 * XXX - ipf_state_seed[X] should be a random number of sorts.
317 #if FREEBSD_GE_REV(400000)
318 state_seed[i] = arc4random();
320 state_seed[i] = ((u_long)state_seed + i) * state_size;
321 state_seed[i] ^= 0xa5a55a5a;
322 state_seed[i] *= (u_long)state_seed;
323 state_seed[i] ^= 0x5a5aa5a5;
324 state_seed[i] *= state_max;
331 /* ------------------------------------------------------------------------ */
332 /* Function: ipf_state_soft_init */
333 /* Returns: int - 0 == success, -1 == failure */
334 /* Parameters: softc(I) - pointer to soft context main structure */
335 /* arg(I) - pointer to local context to use */
337 /* Initialise the state soft context structure so it is ready for use. */
339 /* - allocating a hash table and zero'ing it out */
340 /* - building a secondary table of seeds for double hashing to make it more */
341 /* difficult to attempt to attack the hash table itself (for DoS) */
342 /* - initialise all of the timeout queues, including a table for TCP, some */
343 /* pairs of query/response for UDP and other IP protocols (typically the */
344 /* reply queue has a shorter timeout than the query) */
345 /* ------------------------------------------------------------------------ */
347 ipf_state_soft_init(softc, arg)
348 ipf_main_softc_t *softc;
351 ipf_state_softc_t *softs = arg;
354 KMALLOCS(softs->ipf_state_table,
355 ipstate_t **, softs->ipf_state_size * sizeof(ipstate_t *));
356 if (softs->ipf_state_table == NULL)
359 bzero((char *)softs->ipf_state_table,
360 softs->ipf_state_size * sizeof(ipstate_t *));
362 softs->ipf_state_seed = ipf_state_seed_alloc(softs->ipf_state_size,
363 softs->ipf_state_max);
364 if (softs->ipf_state_seed == NULL)
367 KMALLOCS(softs->ipf_state_stats.iss_bucketlen, u_int *,
368 softs->ipf_state_size * sizeof(u_int));
369 if (softs->ipf_state_stats.iss_bucketlen == NULL)
372 bzero((char *)softs->ipf_state_stats.iss_bucketlen,
373 softs->ipf_state_size * sizeof(u_int));
375 if (softs->ipf_state_maxbucket == 0) {
376 for (i = softs->ipf_state_size; i > 0; i >>= 1)
377 softs->ipf_state_maxbucket++;
378 softs->ipf_state_maxbucket *= 2;
381 ipf_sttab_init(softc, softs->ipf_state_tcptq);
382 softs->ipf_state_stats.iss_tcptab = softs->ipf_state_tcptq;
383 softs->ipf_state_tcptq[IPF_TCP_NSTATES - 1].ifq_next =
384 &softs->ipf_state_udptq;
386 IPFTQ_INIT(&softs->ipf_state_udptq, softc->ipf_udptimeout,
388 softs->ipf_state_udptq.ifq_next = &softs->ipf_state_udpacktq;
390 IPFTQ_INIT(&softs->ipf_state_udpacktq, softc->ipf_udpacktimeout,
392 softs->ipf_state_udpacktq.ifq_next = &softs->ipf_state_icmptq;
394 IPFTQ_INIT(&softs->ipf_state_icmptq, softc->ipf_icmptimeout,
396 softs->ipf_state_icmptq.ifq_next = &softs->ipf_state_icmpacktq;
398 IPFTQ_INIT(&softs->ipf_state_icmpacktq, softc->ipf_icmpacktimeout,
399 "ipftq icmpack tab");
400 softs->ipf_state_icmpacktq.ifq_next = &softs->ipf_state_iptq;
402 IPFTQ_INIT(&softs->ipf_state_iptq, softc->ipf_iptimeout,
403 "ipftq iptimeout tab");
404 softs->ipf_state_iptq.ifq_next = &softs->ipf_state_pending;
406 IPFTQ_INIT(&softs->ipf_state_pending, IPF_HZ_DIVIDE, "ipftq pending");
407 softs->ipf_state_pending.ifq_next = &softs->ipf_state_deletetq;
409 IPFTQ_INIT(&softs->ipf_state_deletetq, 1, "ipftq delete");
410 softs->ipf_state_deletetq.ifq_next = NULL;
412 MUTEX_INIT(&softs->ipf_stinsert, "ipf state insert mutex");
415 softs->ipf_state_wm_last = softc->ipf_ticks;
416 softs->ipf_state_inited = 1;
422 /* ------------------------------------------------------------------------ */
423 /* Function: ipf_state_soft_fini */
424 /* Returns: int - 0 = success, -1 = failure */
425 /* Parameters: softc(I) - pointer to soft context main structure */
426 /* arg(I) - pointer to local context to use */
428 /* Release and destroy any resources acquired or initialised so that */
429 /* IPFilter can be unloaded or re-initialised. */
430 /* ------------------------------------------------------------------------ */
432 ipf_state_soft_fini(softc, arg)
433 ipf_main_softc_t *softc;
436 ipf_state_softc_t *softs = arg;
437 ipftq_t *ifq, *ifqnext;
440 while ((is = softs->ipf_state_list) != NULL)
441 ipf_state_del(softc, is, ISL_UNLOAD);
444 * Proxy timeout queues are not cleaned here because although they
445 * exist on the state list, appr_unload is called after
446 * ipf_state_unload and the proxies actually are responsible for them
447 * being created. Should the proxy timeouts have their own list?
448 * There's no real justification as this is the only complication.
450 for (ifq = softs->ipf_state_usertq; ifq != NULL; ifq = ifqnext) {
451 ifqnext = ifq->ifq_next;
453 if (ipf_deletetimeoutqueue(ifq) == 0)
454 ipf_freetimeoutqueue(softc, ifq);
457 softs->ipf_state_stats.iss_inuse = 0;
458 softs->ipf_state_stats.iss_active = 0;
460 if (softs->ipf_state_inited == 1) {
461 softs->ipf_state_inited = 0;
462 ipf_sttab_destroy(softs->ipf_state_tcptq);
463 MUTEX_DESTROY(&softs->ipf_state_udptq.ifq_lock);
464 MUTEX_DESTROY(&softs->ipf_state_icmptq.ifq_lock);
465 MUTEX_DESTROY(&softs->ipf_state_udpacktq.ifq_lock);
466 MUTEX_DESTROY(&softs->ipf_state_icmpacktq.ifq_lock);
467 MUTEX_DESTROY(&softs->ipf_state_iptq.ifq_lock);
468 MUTEX_DESTROY(&softs->ipf_state_deletetq.ifq_lock);
469 MUTEX_DESTROY(&softs->ipf_state_pending.ifq_lock);
470 MUTEX_DESTROY(&softs->ipf_stinsert);
473 if (softs->ipf_state_table != NULL) {
474 KFREES(softs->ipf_state_table,
475 softs->ipf_state_size * sizeof(*softs->ipf_state_table));
476 softs->ipf_state_table = NULL;
479 if (softs->ipf_state_seed != NULL) {
480 KFREES(softs->ipf_state_seed,
481 softs->ipf_state_size * sizeof(*softs->ipf_state_seed));
482 softs->ipf_state_seed = NULL;
485 if (softs->ipf_state_stats.iss_bucketlen != NULL) {
486 KFREES(softs->ipf_state_stats.iss_bucketlen,
487 softs->ipf_state_size * sizeof(u_int));
488 softs->ipf_state_stats.iss_bucketlen = NULL;
495 /* ------------------------------------------------------------------------ */
496 /* Function: ipf_state_setlock */
498 /* Parameters: arg(I) - pointer to local context to use */
499 /* tmp(I) - new value for lock */
501 /* Stub function that allows for external manipulation of ipf_state_lock */
502 /* ------------------------------------------------------------------------ */
504 ipf_state_setlock(arg, tmp)
508 ipf_state_softc_t *softs = arg;
510 softs->ipf_state_lock = tmp;
514 /* ------------------------------------------------------------------------ */
515 /* Function: ipf_state_stats */
516 /* Returns: ips_state_t* - pointer to state stats structure */
517 /* Parameters: softc(I) - pointer to soft context main structure */
519 /* Put all the current numbers and pointers into a single struct and return */
520 /* a pointer to it. */
521 /* ------------------------------------------------------------------------ */
523 ipf_state_stats(softc)
524 ipf_main_softc_t *softc;
526 ipf_state_softc_t *softs = softc->ipf_state_soft;
527 ips_stat_t *issp = &softs->ipf_state_stats;
529 issp->iss_state_size = softs->ipf_state_size;
530 issp->iss_state_max = softs->ipf_state_max;
531 issp->iss_table = softs->ipf_state_table;
532 issp->iss_list = softs->ipf_state_list;
533 issp->iss_ticks = softc->ipf_ticks;
535 #ifdef IPFILTER_LOGGING
536 issp->iss_log_ok = ipf_log_logok(softc, IPF_LOGSTATE);
537 issp->iss_log_fail = ipf_log_failures(softc, IPF_LOGSTATE);
539 issp->iss_log_ok = 0;
540 issp->iss_log_fail = 0;
545 /* ------------------------------------------------------------------------ */
546 /* Function: ipf_state_remove */
547 /* Returns: int - 0 == success, != 0 == failure */
548 /* Parameters: softc(I) - pointer to soft context main structure */
549 /* data(I) - pointer to state structure to delete from table */
551 /* Search for a state structure that matches the one passed, according to */
552 /* the IP addresses and other protocol specific information. */
553 /* ------------------------------------------------------------------------ */
555 ipf_state_remove(softc, data)
556 ipf_main_softc_t *softc;
559 ipf_state_softc_t *softs = softc->ipf_state_soft;
564 error = ipf_inobj(softc, data, NULL, &st, IPFOBJ_IPSTATE);
568 WRITE_ENTER(&softc->ipf_state);
569 for (sp = softs->ipf_state_list; sp; sp = sp->is_next)
570 if ((sp->is_p == st.is_p) && (sp->is_v == st.is_v) &&
571 !bcmp((caddr_t)&sp->is_src, (caddr_t)&st.is_src,
572 sizeof(st.is_src)) &&
573 !bcmp((caddr_t)&sp->is_dst, (caddr_t)&st.is_dst,
574 sizeof(st.is_dst)) &&
575 !bcmp((caddr_t)&sp->is_ps, (caddr_t)&st.is_ps,
577 ipf_state_del(softc, sp, ISL_REMOVE);
578 RWLOCK_EXIT(&softc->ipf_state);
581 RWLOCK_EXIT(&softc->ipf_state);
588 /* ------------------------------------------------------------------------ */
589 /* Function: ipf_state_ioctl */
590 /* Returns: int - 0 == success, != 0 == failure */
591 /* Parameters: softc(I) - pointer to soft context main structure */
592 /* data(I) - pointer to ioctl data */
593 /* cmd(I) - ioctl command integer */
594 /* mode(I) - file mode bits used with open */
595 /* uid(I) - uid of process making the ioctl call */
596 /* ctx(I) - pointer specific to context of the call */
598 /* Processes an ioctl call made to operate on the IP Filter state device. */
599 /* ------------------------------------------------------------------------ */
601 ipf_state_ioctl(softc, data, cmd, mode, uid, ctx)
602 ipf_main_softc_t *softc;
608 ipf_state_softc_t *softs = softc->ipf_state_soft;
609 int arg, ret, error = 0;
615 * Delete an entry from the state table.
618 error = ipf_state_remove(softc, data);
622 * Flush the state table
625 error = BCOPYIN(data, &arg, sizeof(arg));
631 WRITE_ENTER(&softc->ipf_state);
632 ret = ipf_state_flush(softc, arg, 4);
633 RWLOCK_EXIT(&softc->ipf_state);
635 error = BCOPYOUT(&ret, data, sizeof(ret));
645 error = BCOPYIN(data, &arg, sizeof(arg));
651 WRITE_ENTER(&softc->ipf_state);
652 ret = ipf_state_flush(softc, arg, 6);
653 RWLOCK_EXIT(&softc->ipf_state);
655 error = BCOPYOUT(&ret, data, sizeof(ret));
664 case SIOCMATCHFLUSH :
665 WRITE_ENTER(&softc->ipf_state);
666 error = ipf_state_matchflush(softc, data);
667 RWLOCK_EXIT(&softc->ipf_state);
672 * Flush the state log.
675 if (!(mode & FWRITE)) {
681 tmp = ipf_log_clear(softc, IPL_LOGSTATE);
682 error = BCOPYOUT(&tmp, data, sizeof(tmp));
691 * Turn logging of state information on/off.
694 if (!(mode & FWRITE)) {
698 error = BCOPYIN(data, &softs->ipf_state_logging,
699 sizeof(softs->ipf_state_logging));
708 * Return the current state of logging.
711 error = BCOPYOUT(&softs->ipf_state_logging, data,
712 sizeof(softs->ipf_state_logging));
720 * Return the number of bytes currently waiting to be read.
723 arg = ipf_log_bytesused(softc, IPL_LOGSTATE);
724 error = BCOPYOUT(&arg, data, sizeof(arg));
733 * Get the current state statistics.
736 error = ipf_outobj(softc, data, ipf_state_stats(softc),
741 * Lock/Unlock the state table. (Locking prevents any changes, which
742 * means no packets match).
745 if (!(mode & FWRITE)) {
749 error = ipf_lock(data, &softs->ipf_state_lock);
754 * Add an entry to the current state table.
757 if (!softs->ipf_state_lock || !(mode &FWRITE)) {
762 error = ipf_state_putent(softc, softs, data);
766 * Get a state table entry.
769 if (!softs->ipf_state_lock) {
774 error = ipf_state_getent(softc, softs, data);
778 * Return a copy of the hash table bucket lengths
781 error = BCOPYOUT(softs->ipf_state_stats.iss_bucketlen, data,
782 softs->ipf_state_size * sizeof(u_int));
795 error = ipf_inobj(softc, data, &obj, &iter, IPFOBJ_GENITER);
800 token = ipf_token_find(softc, IPFGENITER_STATE, uid, ctx);
802 error = ipf_state_iter(softc, token, &iter, &obj);
803 WRITE_ENTER(&softc->ipf_tokens);
804 ipf_token_deref(softc, token);
805 RWLOCK_EXIT(&softc->ipf_tokens);
815 error = ipf_state_gettable(softc, softs, data);
819 error = BCOPYIN(data, &arg, sizeof(arg));
825 error = ipf_token_del(softc, arg, uid, ctx);
831 error = ipf_outobj(softc, data, softs->ipf_state_tcptq,
844 /* ------------------------------------------------------------------------ */
845 /* Function: ipf_state_getent */
846 /* Returns: int - 0 == success, != 0 == failure */
847 /* Parameters: softc(I) - pointer to soft context main structure */
848 /* softs(I) - pointer to state context structure */
849 /* data(I) - pointer to state structure to retrieve from table*/
851 /* Copy out state information from the kernel to a user space process. If */
852 /* there is a filter rule associated with the state entry, copy that out */
853 /* as well. The entry to copy out is taken from the value of "ips_next" in */
854 /* the struct passed in and if not null and not found in the list of current*/
855 /* state entries, the retrieval fails. */
856 /* ------------------------------------------------------------------------ */
858 ipf_state_getent(softc, softs, data)
859 ipf_main_softc_t *softc;
860 ipf_state_softc_t *softs;
867 error = ipf_inobj(softc, data, NULL, &ips, IPFOBJ_STATESAVE);
871 READ_ENTER(&softc->ipf_state);
874 isn = softs->ipf_state_list;
876 if (ips.ips_next == NULL) {
877 RWLOCK_EXIT(&softc->ipf_state);
885 * Make sure the pointer we're copying from exists in the
886 * current list of entries. Security precaution to prevent
887 * copying of random kernel data.
889 for (is = softs->ipf_state_list; is; is = is->is_next)
893 RWLOCK_EXIT(&softc->ipf_state);
898 ips.ips_next = isn->is_next;
899 bcopy((char *)isn, (char *)&ips.ips_is, sizeof(ips.ips_is));
900 ips.ips_rule = isn->is_rule;
901 if (isn->is_rule != NULL)
902 bcopy((char *)isn->is_rule, (char *)&ips.ips_fr,
904 RWLOCK_EXIT(&softc->ipf_state);
905 error = ipf_outobj(softc, data, &ips, IPFOBJ_STATESAVE);
910 /* ------------------------------------------------------------------------ */
911 /* Function: ipf_state_putent */
912 /* Returns: int - 0 == success, != 0 == failure */
913 /* Parameters: softc(I) - pointer to soft context main structure */
914 /* softs(I) - pointer to state context structure */
915 /* data(I) - pointer to state information struct */
917 /* This function implements the SIOCSTPUT ioctl: insert a state entry into */
918 /* the state table. If the state info. includes a pointer to a filter rule */
919 /* then also add in an orphaned rule (will not show up in any "ipfstat -io" */
921 /* ------------------------------------------------------------------------ */
923 ipf_state_putent(softc, softs, data)
924 ipf_main_softc_t *softc;
925 ipf_state_softc_t *softs;
934 error = ipf_inobj(softc, data, NULL, &ips, IPFOBJ_STATESAVE);
938 KMALLOC(isn, ipstate_t *);
944 bcopy((char *)&ips.ips_is, (char *)isn, sizeof(*isn));
945 bzero((char *)isn, offsetof(struct ipstate, is_pkts));
946 isn->is_sti.tqe_pnext = NULL;
947 isn->is_sti.tqe_next = NULL;
948 isn->is_sti.tqe_ifq = NULL;
949 isn->is_sti.tqe_parent = isn;
950 isn->is_ifp[0] = NULL;
951 isn->is_ifp[1] = NULL;
952 isn->is_ifp[2] = NULL;
953 isn->is_ifp[3] = NULL;
960 READ_ENTER(&softc->ipf_state);
961 inserr = ipf_state_insert(softc, isn, 0);
962 MUTEX_EXIT(&isn->is_lock);
963 RWLOCK_EXIT(&softc->ipf_state);
968 if (isn->is_flags & SI_NEWFR) {
969 KMALLOC(fr, frentry_t *);
975 bcopy((char *)&ips.ips_fr, (char *)fr, sizeof(*fr));
976 out = fr->fr_flags & FR_OUTQUE ? 1 : 0;
978 ips.ips_is.is_rule = fr;
979 MUTEX_NUKE(&fr->fr_lock);
980 MUTEX_INIT(&fr->fr_lock, "state filter rule lock");
983 * Look up all the interface names in the rule.
985 for (i = 0; i < 4; i++) {
986 if (fr->fr_ifnames[i] == -1) {
987 fr->fr_ifas[i] = NULL;
990 name = fr->fr_names + fr->fr_ifnames[i];
991 fr->fr_ifas[i] = ipf_resolvenic(softc, name,
995 for (i = 0; i < 4; i++) {
996 name = isn->is_ifname[i];
997 isn->is_ifp[i] = ipf_resolvenic(softc, name,
1004 fr->fr_type = FR_T_NONE;
1006 (void) ipf_resolvedest(softc, fr->fr_names, &fr->fr_tifs[0],
1008 (void) ipf_resolvedest(softc, fr->fr_names, &fr->fr_tifs[1],
1010 (void) ipf_resolvedest(softc, fr->fr_names, &fr->fr_dif,
1014 * send a copy back to userland of what we ended up
1015 * to allow for verification.
1017 error = ipf_outobj(softc, data, &ips, IPFOBJ_STATESAVE);
1020 MUTEX_DESTROY(&fr->fr_lock);
1025 READ_ENTER(&softc->ipf_state);
1026 error = ipf_state_insert(softc, isn, 0);
1027 MUTEX_EXIT(&isn->is_lock);
1028 RWLOCK_EXIT(&softc->ipf_state);
1031 READ_ENTER(&softc->ipf_state);
1032 for (is = softs->ipf_state_list; is; is = is->is_next)
1033 if (is->is_rule == fr) {
1034 error = ipf_state_insert(softc, isn, 0);
1035 MUTEX_EXIT(&isn->is_lock);
1043 RWLOCK_EXIT(&softc->ipf_state);
1055 /* ------------------------------------------------------------------------ */
1056 /* Function: ipf_state_insert */
1057 /* Returns: int - 0 == success, -1 == failure */
1058 /* Parameters: softc(I) - pointer to soft context main structure */
1059 /* Parameters: is(I) - pointer to state structure */
1060 /* rev(I) - flag indicating direction of packet */
1062 /* Inserts a state structure into the hash table (for lookups) and the list */
1063 /* of state entries (for enumeration). Resolves all of the interface names */
1064 /* to pointers and adjusts running stats for the hash table as appropriate. */
1066 /* This function can fail if the filter rule has had a population policy of */
1067 /* IP addresses used with stateful filtering assigned to it. */
1069 /* Locking: it is assumed that some kind of lock on ipf_state is held. */
1070 /* Exits with is_lock initialised and held - *EVEN IF ERROR*. */
1071 /* ------------------------------------------------------------------------ */
1073 ipf_state_insert(softc, is, rev)
1074 ipf_main_softc_t *softc;
1078 ipf_state_softc_t *softs = softc->ipf_state_soft;
1084 * Look up all the interface names in the state entry.
1086 for (i = 0; i < 4; i++) {
1087 if (is->is_ifp[i] != NULL)
1089 is->is_ifp[i] = ipf_resolvenic(softc, is->is_ifname[i],
1094 * If we could trust is_hv, then the modulus would not be needed,
1095 * but when running with IPFILTER_SYNC, this stops bad values.
1097 hv = is->is_hv % softs->ipf_state_size;
1102 * We need to get both of these locks...the first because it is
1103 * possible that once the insert is complete another packet might
1104 * come along, match the entry and want to update it.
1106 MUTEX_INIT(&is->is_lock, "ipf state entry");
1107 MUTEX_ENTER(&is->is_lock);
1108 MUTEX_ENTER(&softs->ipf_stinsert);
1112 if ((fr->fr_srctrack.ht_max_nodes != 0) &&
1113 (ipf_ht_node_add(softc, &fr->fr_srctrack,
1114 is->is_family, &is->is_src) == -1)) {
1115 SBUMPD(ipf_state_stats, iss_max_track);
1116 MUTEX_EXIT(&softs->ipf_stinsert);
1120 MUTEX_ENTER(&fr->fr_lock);
1122 MUTEX_EXIT(&fr->fr_lock);
1126 if (is->is_flags & (SI_WILDP|SI_WILDA)) {
1127 DT(iss_wild_plus_one);
1128 SINCL(ipf_state_stats.iss_wild);
1131 SBUMP(ipf_state_stats.iss_proto[is->is_p]);
1132 SBUMP(ipf_state_stats.iss_active_proto[is->is_p]);
1135 * add into list table.
1137 if (softs->ipf_state_list != NULL)
1138 softs->ipf_state_list->is_pnext = &is->is_next;
1139 is->is_pnext = &softs->ipf_state_list;
1140 is->is_next = softs->ipf_state_list;
1141 softs->ipf_state_list = is;
1143 if (softs->ipf_state_table[hv] != NULL)
1144 softs->ipf_state_table[hv]->is_phnext = &is->is_hnext;
1146 softs->ipf_state_stats.iss_inuse++;
1147 is->is_phnext = softs->ipf_state_table + hv;
1148 is->is_hnext = softs->ipf_state_table[hv];
1149 softs->ipf_state_table[hv] = is;
1150 softs->ipf_state_stats.iss_bucketlen[hv]++;
1151 softs->ipf_state_stats.iss_active++;
1152 MUTEX_EXIT(&softs->ipf_stinsert);
1154 ipf_state_setqueue(softc, is, rev);
1160 /* ------------------------------------------------------------------------ */
1161 /* Function: ipf_state_matchipv4addrs */
1162 /* Returns: int - 2 addresses match (strong match), 1 reverse match, */
1164 /* Parameters: is1, is2 pointers to states we are checking */
1166 /* Function matches IPv4 addresses it returns strong match for ICMP proto */
1167 /* even there is only reverse match */
1168 /* ------------------------------------------------------------------------ */
1170 ipf_state_matchipv4addrs(is1, is2)
1171 ipstate_t *is1, *is2;
1175 if (is1->is_saddr == is2->is_saddr && is1->is_daddr == is2->is_daddr)
1177 else if (is1->is_saddr == is2->is_daddr &&
1178 is1->is_daddr == is2->is_saddr) {
1179 /* force strong match for ICMP protocol */
1180 rv = (is1->is_p == IPPROTO_ICMP) ? 2 : 1;
1189 /* ------------------------------------------------------------------------ */
1190 /* Function: ipf_state_matchipv6addrs */
1191 /* Returns: int - 2 addresses match (strong match), 1 reverse match, */
1193 /* Parameters: is1, is2 pointers to states we are checking */
1195 /* Function matches IPv6 addresses it returns strong match for ICMP proto */
1196 /* even there is only reverse match */
1197 /* ------------------------------------------------------------------------ */
1199 ipf_state_matchipv6addrs(is1, is2)
1200 ipstate_t *is1, *is2;
1204 if (IP6_EQ(&is1->is_src, &is2->is_src) &&
1205 IP6_EQ(&is1->is_dst, &is2->is_dst))
1207 else if (IP6_EQ(&is1->is_src, &is2->is_dst) &&
1208 IP6_EQ(&is1->is_dst, &is2->is_src)) {
1209 /* force strong match for ICMPv6 protocol */
1210 rv = (is1->is_p == IPPROTO_ICMPV6) ? 2 : 1;
1219 /* ------------------------------------------------------------------------ */
1220 /* Function: ipf_state_matchaddresses */
1221 /* Returns: int - 2 addresses match, 1 reverse match, zero no match */
1222 /* Parameters: is1, is2 pointers to states we are checking */
1224 /* function retruns true if two pairs of addresses belong to single */
1225 /* connection. suppose there are two endpoints: */
1226 /* endpoint1 1.1.1.1 */
1227 /* endpoint2 1.1.1.2 */
1229 /* the state is established by packet flying from .1 to .2 so we see: */
1230 /* is1->src = 1.1.1.1 */
1231 /* is1->dst = 1.1.1.2 */
1232 /* now endpoint 1.1.1.2 sends answer */
1233 /* retreives is1 record created by first packat and compares it with is2 */
1234 /* temporal record, is2 is initialized as follows: */
1235 /* is2->src = 1.1.1.2 */
1236 /* is2->dst = 1.1.1.1 */
1237 /* in this case 1 will be returned */
1239 /* the ipf_matchaddresses() assumes those two records to be same. of course */
1240 /* the ipf_matchaddresses() also assume records are same in case you pass */
1241 /* identical arguments (i.e. ipf_matchaddress(is1, is1) would return 2 */
1242 /* ------------------------------------------------------------------------ */
1244 ipf_state_matchaddresses(is1, is2)
1245 ipstate_t *is1, *is2;
1249 if (is1->is_v == 4) {
1250 rv = ipf_state_matchipv4addrs(is1, is2);
1253 rv = ipf_state_matchipv6addrs(is1, is2);
1260 /* ------------------------------------------------------------------------ */
1261 /* Function: ipf_matchports */
1262 /* Returns: int - 2 match, 1 rverse match, 0 no match */
1263 /* Parameters: ppairs1, ppairs - src, dst ports we want to match */
1265 /* performs the same match for isps members as for addresses */
1266 /* ------------------------------------------------------------------------ */
1268 ipf_state_matchports(ppairs1, ppairs2)
1269 udpinfo_t *ppairs1, *ppairs2;
1273 if (ppairs1->us_sport == ppairs2->us_sport &&
1274 ppairs1->us_dport == ppairs2->us_dport)
1276 else if (ppairs1->us_sport == ppairs2->us_dport &&
1277 ppairs1->us_dport == ppairs2->us_sport)
1286 /* ------------------------------------------------------------------------ */
1287 /* Function: ipf_matchisps */
1288 /* Returns: int - nonzero if isps members match, 0 nomatch */
1289 /* Parameters: is1, is2 - states we want to match */
1291 /* performs the same match for isps members as for addresses */
1292 /* ------------------------------------------------------------------------ */
1294 ipf_state_matchisps(is1, is2)
1295 ipstate_t *is1, *is2;
1299 if (is1->is_p == is2->is_p) {
1305 /* greinfo_t can be also interprted as port pair */
1306 rv = ipf_state_matchports(&is1->is_ps.is_us,
1311 case IPPROTO_ICMPV6 :
1312 /* force strong match for ICMP datagram. */
1313 if (bcmp(&is1->is_ps, &is2->is_ps,
1314 sizeof(icmpinfo_t)) == 0) {
1332 /* ------------------------------------------------------------------------ */
1333 /* Function: ipf_state_match */
1334 /* Returns: int - nonzero match, zero no match */
1335 /* Parameters: is1, is2 - states we want to match */
1337 /* ------------------------------------------------------------------------ */
1339 ipf_state_match(is1, is2)
1340 ipstate_t *is1, *is2;
1346 if (bcmp(&is1->is_pass, &is2->is_pass,
1347 offsetof(struct ipstate, is_authmsk) -
1348 offsetof(struct ipstate, is_pass)) == 0) {
1350 pomatch = ipf_state_matchisps(is1, is2);
1351 amatch = ipf_state_matchaddresses(is1, is2);
1352 rv = (amatch != 0) && (amatch == pomatch);
1360 /* ------------------------------------------------------------------------ */
1361 /* Function: ipf_state_add */
1362 /* Returns: ipstate_t - 0 = success */
1363 /* Parameters: softc(I) - pointer to soft context main structure */
1364 /* fin(I) - pointer to packet information */
1365 /* stsave(O) - pointer to place to save pointer to created */
1366 /* state structure. */
1367 /* flags(I) - flags to use when creating the structure */
1369 /* Creates a new IP state structure from the packet information collected. */
1370 /* Inserts it into the state table and appends to the bottom of the active */
1371 /* list. If the capacity of the table has reached the maximum allowed then */
1372 /* the call will fail and a flush is scheduled for the next timeout call. */
1374 /* NOTE: The use of stsave to point to nat_state will result in memory */
1375 /* corruption. It should only be used to point to objects that will */
1376 /* either outlive this (not expired) or will deref the ip_state_t */
1377 /* when they are deleted. */
1378 /* ------------------------------------------------------------------------ */
1380 ipf_state_add(softc, fin, stsave, flags)
1381 ipf_main_softc_t *softc;
1386 ipf_state_softc_t *softs = softc->ipf_state_soft;
1396 * If a locally created packet is trying to egress but it
1397 * does not match because of this lock, it is likely that
1398 * the policy will block it and return network unreachable further
1399 * up the stack. To mitigate this error, EAGAIN is returned instead,
1400 * telling the IP stack to try sending this packet again later.
1402 if (softs->ipf_state_lock) {
1403 SBUMPD(ipf_state_stats, iss_add_locked);
1404 fin->fin_error = EAGAIN;
1408 if (fin->fin_flx & (FI_SHORT|FI_STATE|FI_FRAGBODY|FI_BAD)) {
1409 SBUMPD(ipf_state_stats, iss_add_bad);
1413 if ((fin->fin_flx & FI_OOW) && !(fin->fin_tcpf & TH_SYN)) {
1414 SBUMPD(ipf_state_stats, iss_add_oow);
1418 if ((softs->ipf_state_stats.iss_active * 100 / softs->ipf_state_max) >
1419 softs->ipf_state_wm_high) {
1420 softs->ipf_state_doflush = 1;
1424 * If a "keep state" rule has reached the maximum number of references
1425 * to it, then schedule an automatic flush in case we can clear out
1426 * some "dead old wood". Note that because the lock isn't held on
1427 * fr it is possible that we could overflow. The cost of overflowing
1428 * is being ignored here as the number by which it can overflow is
1429 * a product of the number of simultaneous threads that could be
1430 * executing in here, so a limit of 100 won't result in 200, but could
1431 * result in 101 or 102.
1435 if ((softs->ipf_state_stats.iss_active >=
1436 softs->ipf_state_max) && (fr->fr_statemax == 0)) {
1437 SBUMPD(ipf_state_stats, iss_max);
1440 if ((fr->fr_statemax != 0) &&
1441 (fr->fr_statecnt >= fr->fr_statemax)) {
1442 SBUMPD(ipf_state_stats, iss_max_ref);
1449 pass = softc->ipf_flags;
1450 is->is_tag = FR_NOLOGTAG;
1452 pass = fr->fr_flags;
1458 bzero((char *)is, sizeof(*is));
1459 is->is_die = 1 + softc->ipf_ticks;
1461 * We want to check everything that is a property of this packet,
1462 * but we don't (automatically) care about its fragment status as
1466 is->is_v = fin->fin_v;
1467 is->is_sec = fin->fin_secmsk;
1468 is->is_secmsk = 0xffff;
1469 is->is_auth = fin->fin_auth;
1470 is->is_authmsk = 0xffff;
1471 is->is_family = fin->fin_family;
1472 is->is_opt[0] = fin->fin_optmsk;
1473 is->is_optmsk[0] = 0xffffffff;
1474 if (is->is_v == 6) {
1475 is->is_opt[0] &= ~0x8;
1476 is->is_optmsk[0] &= ~0x8;
1480 * Copy and calculate...
1482 hv = (is->is_p = fin->fin_fi.fi_p);
1483 is->is_src = fin->fin_fi.fi_src;
1485 is->is_dst = fin->fin_fi.fi_dst;
1488 if (fin->fin_v == 6) {
1490 * For ICMPv6, we check to see if the destination address is
1491 * a multicast address. If it is, do not include it in the
1492 * calculation of the hash because the correct reply will come
1493 * back from a real address, not a multicast address.
1495 if ((is->is_p == IPPROTO_ICMPV6) &&
1496 IN6_IS_ADDR_MULTICAST(&is->is_dst.in6)) {
1498 * So you can do keep state with neighbour discovery.
1500 * Here we could use the address from the neighbour
1501 * solicit message to put in the state structure and
1502 * we could use that without a wildcard flag too...
1504 flags |= SI_W_DADDR;
1507 hv += is->is_dst.i6[1];
1508 hv += is->is_dst.i6[2];
1509 hv += is->is_dst.i6[3];
1511 hv += is->is_src.i6[1];
1512 hv += is->is_src.i6[2];
1513 hv += is->is_src.i6[3];
1516 if ((fin->fin_v == 4) &&
1517 (fin->fin_flx & (FI_MULTICAST|FI_BROADCAST|FI_MBCAST))) {
1518 flags |= SI_W_DADDR;
1525 case IPPROTO_ICMPV6 :
1528 switch (ic->icmp_type)
1530 case ICMP6_ECHO_REQUEST :
1531 hv += (is->is_icmp.ici_id = ic->icmp_id);
1533 case ICMP6_MEMBERSHIP_QUERY :
1534 case ND_ROUTER_SOLICIT :
1535 case ND_NEIGHBOR_SOLICIT :
1536 case ICMP6_NI_QUERY :
1537 is->is_icmp.ici_type = ic->icmp_type;
1540 SBUMPD(ipf_state_stats, iss_icmp6_notquery);
1548 switch (ic->icmp_type)
1554 is->is_icmp.ici_type = ic->icmp_type;
1555 hv += (is->is_icmp.ici_id = ic->icmp_id);
1558 SBUMPD(ipf_state_stats, iss_icmp_notquery);
1567 is->is_gre.gs_flags = gre->gr_flags;
1568 is->is_gre.gs_ptype = gre->gr_ptype;
1569 if (GRE_REV(is->is_gre.gs_flags) == 1) {
1570 is->is_call[0] = fin->fin_data[0];
1571 is->is_call[1] = fin->fin_data[1];
1579 if (tcp->th_flags & TH_RST) {
1580 SBUMPD(ipf_state_stats, iss_tcp_rstadd);
1584 /* TRACE is, flags, hv */
1587 * The endian of the ports doesn't matter, but the ack and
1588 * sequence numbers do as we do mathematics on them later.
1590 is->is_sport = htons(fin->fin_data[0]);
1591 is->is_dport = htons(fin->fin_data[1]);
1592 if ((flags & (SI_W_DPORT|SI_W_SPORT)) == 0) {
1597 /* TRACE is, flags, hv */
1600 * If this is a real packet then initialise fields in the
1601 * state information structure from the TCP header information.
1605 is->is_maxswin = ntohs(tcp->th_win);
1606 if (is->is_maxswin == 0)
1609 if ((fin->fin_flx & FI_IGNORE) == 0) {
1610 is->is_send = ntohl(tcp->th_seq) + fin->fin_dlen -
1611 (TCP_OFF(tcp) << 2) +
1612 ((tcp->th_flags & TH_SYN) ? 1 : 0) +
1613 ((tcp->th_flags & TH_FIN) ? 1 : 0);
1614 is->is_maxsend = is->is_send;
1617 * Window scale option is only present in
1618 * SYN/SYN-ACK packet.
1620 if ((tcp->th_flags & ~(TH_FIN|TH_ACK|TH_ECNALL)) ==
1622 (TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) {
1623 if (ipf_tcpoptions(softs, fin, tcp,
1624 &is->is_tcp.ts_data[0]) == -1)
1625 fin->fin_flx |= FI_BAD;
1628 if ((fin->fin_out != 0) && (pass & FR_NEWISN) != 0) {
1629 ipf_checknewisn(fin, is);
1630 ipf_fixoutisn(fin, is);
1633 if ((tcp->th_flags & TH_OPENING) == TH_SYN)
1636 is->is_maxdwin = is->is_maxswin * 2;
1637 is->is_dend = ntohl(tcp->th_ack);
1638 is->is_maxdend = ntohl(tcp->th_ack);
1639 is->is_maxdwin *= 2;
1644 * If we're creating state for a starting connection, start
1645 * the timer on it as we'll never see an error if it fails
1653 is->is_sport = htons(fin->fin_data[0]);
1654 is->is_dport = htons(fin->fin_data[1]);
1655 if ((flags & (SI_W_DPORT|SI_W_SPORT)) == 0) {
1656 hv += tcp->th_dport;
1657 hv += tcp->th_sport;
1664 hv = DOUBLE_HASH(hv);
1668 * Look for identical state.
1670 for (is = softs->ipf_state_table[hv % softs->ipf_state_size];
1671 is != NULL; is = is->is_hnext) {
1672 if (ipf_state_match(&ips, is) == 1)
1676 SBUMPD(ipf_state_stats, iss_add_dup);
1680 if (softs->ipf_state_stats.iss_bucketlen[hv] >=
1681 softs->ipf_state_maxbucket) {
1682 SBUMPD(ipf_state_stats, iss_bucket_full);
1687 * No existing state; create new
1689 KMALLOC(is, ipstate_t *);
1691 SBUMPD(ipf_state_stats, iss_nomem);
1694 bcopy((char *)&ips, (char *)is, sizeof(*is));
1695 is->is_flags = flags & IS_INHERITED;
1696 is->is_rulen = fin->fin_rule;
1700 * Do not do the modulus here, it is done in ipf_state_insert().
1705 (void) strncpy(is->is_group, FR_NAME(fr, fr_group),
1707 if (fr->fr_age[0] != 0) {
1708 tq = ipf_addtimeoutqueue(softc,
1709 &softs->ipf_state_usertq,
1711 is->is_tqehead[0] = tq;
1712 is->is_sti.tqe_flags |= TQE_RULEBASED;
1714 if (fr->fr_age[1] != 0) {
1715 tq = ipf_addtimeoutqueue(softc,
1716 &softs->ipf_state_usertq,
1718 is->is_tqehead[1] = tq;
1719 is->is_sti.tqe_flags |= TQE_RULEBASED;
1722 is->is_tag = fr->fr_logtag;
1726 * It may seem strange to set is_ref to 2, but if stsave is not NULL
1727 * then a copy of the pointer is being stored somewhere else and in
1728 * the end, it will expect to be able to do something with it.
1731 if (stsave != NULL) {
1737 is->is_pkts[0] = 0, is->is_bytes[0] = 0;
1738 is->is_pkts[1] = 0, is->is_bytes[1] = 0;
1739 is->is_pkts[2] = 0, is->is_bytes[2] = 0;
1740 is->is_pkts[3] = 0, is->is_bytes[3] = 0;
1741 if ((fin->fin_flx & FI_IGNORE) == 0) {
1742 is->is_pkts[out] = 1;
1743 fin->fin_pktnum = 1;
1744 is->is_bytes[out] = fin->fin_plen;
1745 is->is_flx[out][0] = fin->fin_flx & FI_CMP;
1746 is->is_flx[out][0] &= ~FI_OOW;
1749 if (pass & FR_STLOOSE)
1750 is->is_flags |= IS_LOOSE;
1752 if (pass & FR_STSTRICT)
1753 is->is_flags |= IS_STRICT;
1755 if (pass & FR_STATESYNC)
1756 is->is_flags |= IS_STATESYNC;
1758 if (pass & FR_LOGFIRST)
1759 is->is_pass &= ~(FR_LOGFIRST|FR_LOG);
1761 READ_ENTER(&softc->ipf_state);
1763 if (ipf_state_insert(softc, is, fin->fin_rev) == -1) {
1764 RWLOCK_EXIT(&softc->ipf_state);
1766 * This is a bit more manual than it should be but
1767 * ipf_state_del cannot be called.
1769 MUTEX_EXIT(&is->is_lock);
1770 MUTEX_DESTROY(&is->is_lock);
1771 if (is->is_tqehead[0] != NULL) {
1772 if (ipf_deletetimeoutqueue(is->is_tqehead[0]) == 0)
1773 ipf_freetimeoutqueue(softc, is->is_tqehead[0]);
1774 is->is_tqehead[0] = NULL;
1776 if (is->is_tqehead[1] != NULL) {
1777 if (ipf_deletetimeoutqueue(is->is_tqehead[1]) == 0)
1778 ipf_freetimeoutqueue(softc, is->is_tqehead[1]);
1779 is->is_tqehead[1] = NULL;
1786 * Filling in the interface name is after the insert so that an
1787 * event (such as add/delete) of an interface that is referenced
1788 * by this rule will see this state entry.
1792 * The name '-' is special for network interfaces and causes
1793 * a NULL name to be present, always, allowing packets to
1794 * match it, regardless of their interface.
1796 if ((fin->fin_ifp == NULL) ||
1797 (fr->fr_ifnames[out << 1] != -1 &&
1798 fr->fr_names[fr->fr_ifnames[out << 1] + 0] == '-' &&
1799 fr->fr_names[fr->fr_ifnames[out << 1] + 1] == '\0')) {
1800 is->is_ifp[out << 1] = fr->fr_ifas[0];
1801 strncpy(is->is_ifname[out << 1],
1802 fr->fr_names + fr->fr_ifnames[0],
1803 sizeof(fr->fr_ifnames[0]));
1805 is->is_ifp[out << 1] = fin->fin_ifp;
1806 COPYIFNAME(fin->fin_v, fin->fin_ifp,
1807 is->is_ifname[out << 1]);
1810 is->is_ifp[(out << 1) + 1] = fr->fr_ifas[1];
1811 if (fr->fr_ifnames[1] != -1) {
1812 strncpy(is->is_ifname[(out << 1) + 1],
1813 fr->fr_names + fr->fr_ifnames[1],
1814 sizeof(fr->fr_ifnames[1]));
1817 is->is_ifp[(1 - out) << 1] = fr->fr_ifas[2];
1818 if (fr->fr_ifnames[2] != -1) {
1819 strncpy(is->is_ifname[((1 - out) << 1)],
1820 fr->fr_names + fr->fr_ifnames[2],
1821 sizeof(fr->fr_ifnames[2]));
1824 is->is_ifp[((1 - out) << 1) + 1] = fr->fr_ifas[3];
1825 if (fr->fr_ifnames[3] != -1) {
1826 strncpy(is->is_ifname[((1 - out) << 1) + 1],
1827 fr->fr_names + fr->fr_ifnames[3],
1828 sizeof(fr->fr_ifnames[3]));
1831 if (fin->fin_ifp != NULL) {
1832 is->is_ifp[out << 1] = fin->fin_ifp;
1833 COPYIFNAME(fin->fin_v, fin->fin_ifp,
1834 is->is_ifname[out << 1]);
1838 if (fin->fin_p == IPPROTO_TCP) {
1840 * If we're creating state for a starting connection, start the
1841 * timer on it as we'll never see an error if it fails to
1844 (void) ipf_tcp_age(&is->is_sti, fin, softs->ipf_state_tcptq,
1847 MUTEX_EXIT(&is->is_lock);
1848 if ((is->is_flags & IS_STATESYNC) && ((is->is_flags & SI_CLONE) == 0))
1849 is->is_sync = ipf_sync_new(softc, SMC_STATE, fin, is);
1850 if (softs->ipf_state_logging)
1851 ipf_state_log(softc, is, ISL_NEW);
1853 RWLOCK_EXIT(&softc->ipf_state);
1855 fin->fin_flx |= FI_STATE;
1856 if (fin->fin_flx & FI_FRAG)
1857 (void) ipf_frag_new(softc, fin, pass);
1859 fdp = &fr->fr_tifs[0];
1860 if (fdp->fd_type == FRD_DSTLIST) {
1861 ipf_dstlist_select_node(fin, fdp->fd_ptr, NULL,
1864 bcopy(fdp, &is->is_tifs[0], sizeof(*fdp));
1867 fdp = &fr->fr_tifs[1];
1868 if (fdp->fd_type == FRD_DSTLIST) {
1869 ipf_dstlist_select_node(fin, fdp->fd_ptr, NULL,
1872 bcopy(fdp, &is->is_tifs[1], sizeof(*fdp));
1874 fin->fin_tif = &is->is_tifs[fin->fin_rev];
1877 if (fdp->fd_type == FRD_DSTLIST) {
1878 ipf_dstlist_select_node(fin, fdp->fd_ptr, NULL,
1881 bcopy(fdp, &is->is_dif, sizeof(*fdp));
1883 fin->fin_dif = &is->is_dif;
1889 /* ------------------------------------------------------------------------ */
1890 /* Function: ipf_tcpoptions */
1891 /* Returns: int - 1 == packet matches state entry, 0 == it does not, */
1892 /* -1 == packet has bad TCP options data */
1893 /* Parameters: softs(I) - pointer to state context structure */
1894 /* fin(I) - pointer to packet information */
1895 /* tcp(I) - pointer to TCP packet header */
1896 /* td(I) - pointer to TCP data held as part of the state */
1898 /* Look after the TCP header for any options and deal with those that are */
1899 /* present. Record details about those that we recogise. */
1900 /* ------------------------------------------------------------------------ */
1902 ipf_tcpoptions(softs, fin, tcp, td)
1903 ipf_state_softc_t *softs;
1908 int off, mlen, ol, i, len, retval;
1909 char buf[64], *s, opt;
1912 len = (TCP_OFF(tcp) << 2);
1913 if (fin->fin_dlen < len) {
1914 SBUMPD(ipf_state_stats, iss_tcp_toosmall);
1917 len -= sizeof(*tcp);
1919 off = fin->fin_plen - fin->fin_dlen + sizeof(*tcp) + fin->fin_ipoff;
1922 mlen = MSGDSIZE(m) - off;
1930 COPYDATA(m, off, len, buf);
1932 for (s = buf; len > 0; ) {
1934 if (opt == TCPOPT_EOL)
1936 else if (opt == TCPOPT_NOP)
1942 if (ol < 2 || ol > len)
1946 * Extract the TCP options we are interested in out of
1947 * the header and store them in the the tcpdata struct.
1951 case TCPOPT_WINDOW :
1952 if (ol == TCPOLEN_WINDOW) {
1954 if (i > TCP_WSCALE_MAX)
1958 td->td_winscale = i;
1959 td->td_winflags |= TCP_WSCALE_SEEN|
1964 case TCPOPT_MAXSEG :
1966 * So, if we wanted to set the TCP MAXSEG,
1967 * it should be done here...
1969 if (ol == TCPOLEN_MAXSEG) {
1977 case TCPOPT_SACK_PERMITTED :
1978 if (ol == TCPOLEN_SACK_PERMITTED)
1979 td->td_winflags |= TCP_SACK_PERMIT;
1989 SBUMPD(ipf_state_stats, iss_tcp_badopt);
1995 /* ------------------------------------------------------------------------ */
1996 /* Function: ipf_state_tcp */
1997 /* Returns: int - 1 == packet matches state entry, 0 == it does not */
1998 /* Parameters: softc(I) - pointer to soft context main structure */
1999 /* softs(I) - pointer to state context structure */
2000 /* fin(I) - pointer to packet information */
2001 /* tcp(I) - pointer to TCP packet header */
2002 /* is(I) - pointer to master state structure */
2004 /* Check to see if a packet with TCP headers fits within the TCP window. */
2005 /* Change timeout depending on whether new packet is a SYN-ACK returning */
2006 /* for a SYN or a RST or FIN which indicate time to close up shop. */
2007 /* ------------------------------------------------------------------------ */
2009 ipf_state_tcp(softc, softs, fin, tcp, is)
2010 ipf_main_softc_t *softc;
2011 ipf_state_softc_t *softs;
2016 tcpdata_t *fdata, *tdata;
2017 int source, ret, flags;
2019 source = !fin->fin_rev;
2020 if (((is->is_flags & IS_TCPFSM) != 0) && (source == 1) &&
2021 (ntohs(is->is_sport) != fin->fin_data[0]))
2023 fdata = &is->is_tcp.ts_data[!source];
2024 tdata = &is->is_tcp.ts_data[source];
2026 MUTEX_ENTER(&is->is_lock);
2029 * If a SYN packet is received for a connection that is on the way out
2030 * but hasn't yet departed then advance this session along the way.
2032 if ((tcp->th_flags & TH_OPENING) == TH_SYN) {
2033 if ((is->is_state[0] > IPF_TCPS_ESTABLISHED) &&
2034 (is->is_state[1] > IPF_TCPS_ESTABLISHED)) {
2035 is->is_state[!source] = IPF_TCPS_CLOSED;
2036 ipf_movequeue(softc->ipf_ticks, &is->is_sti,
2038 &softs->ipf_state_deletetq);
2039 MUTEX_EXIT(&is->is_lock);
2040 DT1(iss_tcp_closing, ipstate_t *, is);
2041 SBUMP(ipf_state_stats.iss_tcp_closing);
2046 if (is->is_flags & IS_LOOSE)
2049 ret = ipf_state_tcpinwindow(fin, fdata, tdata, tcp,
2053 * Nearing end of connection, start timeout.
2055 ret = ipf_tcp_age(&is->is_sti, fin, softs->ipf_state_tcptq,
2058 MUTEX_EXIT(&is->is_lock);
2059 DT2(iss_tcp_fsm, fr_info_t *, fin, ipstate_t *, is);
2060 SBUMP(ipf_state_stats.iss_tcp_fsm);
2064 if (softs->ipf_state_logging > 4)
2065 ipf_state_log(softc, is, ISL_STATECHANGE);
2068 * set s0's as appropriate. Use syn-ack packet as it
2069 * contains both pieces of required information.
2072 * Window scale option is only present in SYN/SYN-ACK packet.
2073 * Compare with ~TH_FIN to mask out T/TCP setups.
2075 flags = tcp->th_flags & ~(TH_FIN|TH_ECNALL);
2076 if (flags == (TH_SYN|TH_ACK)) {
2077 is->is_s0[source] = ntohl(tcp->th_ack);
2078 is->is_s0[!source] = ntohl(tcp->th_seq) + 1;
2079 if ((TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) {
2080 if (ipf_tcpoptions(softs, fin, tcp,
2082 fin->fin_flx |= FI_BAD;
2084 if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN))
2085 ipf_checknewisn(fin, is);
2086 } else if (flags == TH_SYN) {
2087 is->is_s0[source] = ntohl(tcp->th_seq) + 1;
2088 if ((TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) {
2089 if (ipf_tcpoptions(softs, fin, tcp,
2091 fin->fin_flx |= FI_BAD;
2094 if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN))
2095 ipf_checknewisn(fin, is);
2100 DT2(iss_tcp_oow, fr_info_t *, fin, ipstate_t *, is);
2101 SBUMP(ipf_state_stats.iss_tcp_oow);
2104 MUTEX_EXIT(&is->is_lock);
2109 /* ------------------------------------------------------------------------ */
2110 /* Function: ipf_checknewisn */
2112 /* Parameters: fin(I) - pointer to packet information */
2113 /* is(I) - pointer to master state structure */
2115 /* Check to see if this TCP connection is expecting and needs a new */
2116 /* sequence number for a particular direction of the connection. */
2118 /* NOTE: This does not actually change the sequence numbers, only gets new */
2120 /* ------------------------------------------------------------------------ */
2122 ipf_checknewisn(fin, is)
2126 u_32_t sumd, old, new;
2133 if (((i == 0) && !(is->is_flags & IS_ISNSYN)) ||
2134 ((i == 1) && !(is->is_flags & IS_ISNACK))) {
2135 old = ntohl(tcp->th_seq);
2136 new = ipf_newisn(fin);
2137 is->is_isninc[i] = new - old;
2138 CALC_SUMD(old, new, sumd);
2139 is->is_sumd[i] = (sumd & 0xffff) + (sumd >> 16);
2141 is->is_flags |= ((i == 0) ? IS_ISNSYN : IS_ISNACK);
2146 /* ------------------------------------------------------------------------ */
2147 /* Function: ipf_state_tcpinwindow */
2148 /* Returns: int - 1 == packet inside TCP "window", 0 == not inside. */
2149 /* Parameters: fin(I) - pointer to packet information */
2150 /* fdata(I) - pointer to tcp state informatio (forward) */
2151 /* tdata(I) - pointer to tcp state informatio (reverse) */
2152 /* tcp(I) - pointer to TCP packet header */
2154 /* Given a packet has matched addresses and ports, check to see if it is */
2155 /* within the TCP data window. In a show of generosity, allow packets that */
2156 /* are within the window space behind the current sequence # as well. */
2157 /* ------------------------------------------------------------------------ */
2159 ipf_state_tcpinwindow(fin, fdata, tdata, tcp, flags)
2161 tcpdata_t *fdata, *tdata;
2165 ipf_main_softc_t *softc = fin->fin_main_soft;
2166 ipf_state_softc_t *softs = softc->ipf_state_soft;
2167 tcp_seq seq, ack, end;
2168 int ackskew, tcpflags;
2173 * Find difference between last checked packet and this packet.
2175 tcpflags = tcp->th_flags;
2176 seq = ntohl(tcp->th_seq);
2177 ack = ntohl(tcp->th_ack);
2178 if (tcpflags & TH_SYN)
2179 win = ntohs(tcp->th_win);
2181 win = ntohs(tcp->th_win) << fdata->td_winscale;
2184 * A window of 0 produces undesirable behaviour from this function.
2189 dsize = fin->fin_dlen - (TCP_OFF(tcp) << 2) +
2190 ((tcpflags & TH_SYN) ? 1 : 0) + ((tcpflags & TH_FIN) ? 1 : 0);
2193 * if window scaling is present, the scaling is only allowed
2194 * for windows not in the first SYN packet. In that packet the
2195 * window is 65535 to specify the largest window possible
2196 * for receivers not implementing the window scale option.
2197 * Currently, we do not assume TTCP here. That means that
2198 * if we see a second packet from a host (after the initial
2199 * SYN), we can assume that the receiver of the SYN did
2200 * already send back the SYN/ACK (and thus that we know if
2201 * the receiver also does window scaling)
2203 if (!(tcpflags & TH_SYN) && (fdata->td_winflags & TCP_WSCALE_FIRST)) {
2204 fdata->td_winflags &= ~TCP_WSCALE_FIRST;
2205 fdata->td_maxwin = win;
2210 if ((fdata->td_end == 0) &&
2211 (!(flags & IS_TCPFSM) ||
2212 ((tcpflags & TH_OPENING) == TH_OPENING))) {
2214 * Must be a (outgoing) SYN-ACK in reply to a SYN.
2216 fdata->td_end = end - 1;
2217 fdata->td_maxwin = 1;
2218 fdata->td_maxend = end + win;
2221 if (!(tcpflags & TH_ACK)) { /* Pretend an ack was sent */
2222 ack = tdata->td_end;
2223 } else if (((tcpflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) &&
2225 /* gross hack to get around certain broken tcp stacks */
2226 ack = tdata->td_end;
2229 maxwin = tdata->td_maxwin;
2230 ackskew = tdata->td_end - ack;
2233 * Strict sequencing only allows in-order delivery.
2235 if ((flags & IS_STRICT) != 0) {
2236 if (seq != fdata->td_end) {
2237 DT2(iss_tcp_struct, tcpdata_t *, fdata, int, seq);
2238 SBUMP(ipf_state_stats.iss_tcp_strict);
2239 fin->fin_flx |= FI_OOW;
2244 #define SEQ_GE(a,b) ((int)((a) - (b)) >= 0)
2245 #define SEQ_GT(a,b) ((int)((a) - (b)) > 0)
2247 if ((SEQ_GE(fdata->td_maxend, end)) &&
2248 (SEQ_GE(seq, fdata->td_end - maxwin)) &&
2249 /* XXX what about big packets */
2250 #define MAXACKWINDOW 66000
2251 (-ackskew <= (MAXACKWINDOW)) &&
2252 ( ackskew <= (MAXACKWINDOW << fdata->td_winscale))) {
2255 * Microsoft Windows will send the next packet to the right of the
2256 * window if SACK is in use.
2258 } else if ((seq == fdata->td_maxend) && (ackskew == 0) &&
2259 (fdata->td_winflags & TCP_SACK_PERMIT) &&
2260 (tdata->td_winflags & TCP_SACK_PERMIT)) {
2261 DT2(iss_sinsack, tcpdata_t *, fdata, int, seq);
2262 SBUMP(ipf_state_stats.iss_winsack);
2265 * Sometimes a TCP RST will be generated with only the ACK field
2268 } else if ((seq == 0) && (tcpflags == (TH_RST|TH_ACK)) &&
2269 (ackskew >= -1) && (ackskew <= 1)) {
2271 } else if (!(flags & IS_TCPFSM)) {
2274 i = (fin->fin_rev << 1) + fin->fin_out;
2277 if (is_pkts[i]0 == 0) {
2279 * Picking up a connection in the middle, the "next"
2280 * packet seen from a direction that is new should be
2281 * accepted, even if it appears out of sequence.
2286 if (!(fdata->td_winflags &
2287 (TCP_WSCALE_SEEN|TCP_WSCALE_FIRST))) {
2289 * No TCPFSM and no window scaling, so make some
2292 if ((seq == fdata->td_maxend) && (ackskew == 0))
2294 else if (SEQ_GE(seq + maxwin, fdata->td_end - maxwin))
2299 /* TRACE(inseq, fdata, tdata, seq, end, ack, ackskew, win, maxwin) */
2302 /* if ackskew < 0 then this should be due to fragmented
2303 * packets. There is no way to know the length of the
2304 * total packet in advance.
2305 * We do know the total length from the fragment cache though.
2306 * Note however that there might be more sessions with
2307 * exactly the same source and destination parameters in the
2308 * state cache (and source and destination is the only stuff
2309 * that is saved in the fragment cache). Note further that
2310 * some TCP connections in the state cache are hashed with
2311 * sport and dport as well which makes it not worthwhile to
2313 * Thus, when ackskew is negative but still seems to belong
2314 * to this session, we bump up the destinations end value.
2317 tdata->td_end = ack;
2319 /* update max window seen */
2320 if (fdata->td_maxwin < win)
2321 fdata->td_maxwin = win;
2322 if (SEQ_GT(end, fdata->td_end))
2323 fdata->td_end = end;
2324 if (SEQ_GE(ack + win, tdata->td_maxend))
2325 tdata->td_maxend = ack + win;
2328 SBUMP(ipf_state_stats.iss_oow);
2329 fin->fin_flx |= FI_OOW;
2334 /* ------------------------------------------------------------------------ */
2335 /* Function: ipf_state_clone */
2336 /* Returns: ipstate_t* - NULL == cloning failed, */
2337 /* else pointer to new state structure */
2338 /* Parameters: fin(I) - pointer to packet information */
2339 /* tcp(I) - pointer to TCP/UDP header */
2340 /* is(I) - pointer to master state structure */
2342 /* Create a "duplcate" state table entry from the master. */
2343 /* ------------------------------------------------------------------------ */
2345 ipf_state_clone(fin, tcp, is)
2350 ipf_main_softc_t *softc = fin->fin_main_soft;
2351 ipf_state_softc_t *softs = softc->ipf_state_soft;
2355 if (softs->ipf_state_stats.iss_active == softs->ipf_state_max) {
2356 SBUMPD(ipf_state_stats, iss_max);
2357 softs->ipf_state_doflush = 1;
2360 KMALLOC(clone, ipstate_t *);
2361 if (clone == NULL) {
2362 SBUMPD(ipf_state_stats, iss_clone_nomem);
2365 bcopy((char *)is, (char *)clone, sizeof(*clone));
2367 MUTEX_NUKE(&clone->is_lock);
2369 * It has not yet been placed on any timeout queue, so make sure
2370 * all of that data is zero'd out.
2372 clone->is_sti.tqe_pnext = NULL;
2373 clone->is_sti.tqe_next = NULL;
2374 clone->is_sti.tqe_ifq = NULL;
2375 clone->is_sti.tqe_parent = clone;
2377 clone->is_die = ONE_DAY + softc->ipf_ticks;
2378 clone->is_state[0] = 0;
2379 clone->is_state[1] = 0;
2380 send = ntohl(tcp->th_seq) + fin->fin_dlen - (TCP_OFF(tcp) << 2) +
2381 ((tcp->th_flags & TH_SYN) ? 1 : 0) +
2382 ((tcp->th_flags & TH_FIN) ? 1 : 0);
2384 if (fin->fin_rev == 1) {
2385 clone->is_dend = send;
2386 clone->is_maxdend = send;
2388 clone->is_maxswin = 1;
2389 clone->is_maxdwin = ntohs(tcp->th_win);
2390 if (clone->is_maxdwin == 0)
2391 clone->is_maxdwin = 1;
2393 clone->is_send = send;
2394 clone->is_maxsend = send;
2396 clone->is_maxdwin = 1;
2397 clone->is_maxswin = ntohs(tcp->th_win);
2398 if (clone->is_maxswin == 0)
2399 clone->is_maxswin = 1;
2402 clone->is_flags &= ~SI_CLONE;
2403 clone->is_flags |= SI_CLONED;
2404 if (ipf_state_insert(softc, clone, fin->fin_rev) == -1) {
2410 if (clone->is_p == IPPROTO_TCP) {
2411 (void) ipf_tcp_age(&clone->is_sti, fin, softs->ipf_state_tcptq,
2412 clone->is_flags, 2);
2414 MUTEX_EXIT(&clone->is_lock);
2415 if (is->is_flags & IS_STATESYNC)
2416 clone->is_sync = ipf_sync_new(softc, SMC_STATE, fin, clone);
2417 DT2(iss_clone, ipstate_t *, is, ipstate_t *, clone);
2418 SBUMP(ipf_state_stats.iss_cloned);
2423 /* ------------------------------------------------------------------------ */
2424 /* Function: ipf_matchsrcdst */
2426 /* Parameters: fin(I) - pointer to packet information */
2427 /* is(I) - pointer to state structure */
2428 /* src(I) - pointer to source address */
2429 /* dst(I) - pointer to destination address */
2430 /* tcp(I) - pointer to TCP/UDP header */
2431 /* cmask(I) - mask of FI_* bits to check */
2433 /* Match a state table entry against an IP packet. The logic below is that */
2434 /* ret gets set to one if the match succeeds, else remains 0. If it is */
2435 /* still 0 after the test. no match. */
2436 /* ------------------------------------------------------------------------ */
2438 ipf_matchsrcdst(fin, is, src, dst, tcp, cmask)
2441 i6addr_t *src, *dst;
2445 ipf_main_softc_t *softc = fin->fin_main_soft;
2446 ipf_state_softc_t *softs = softc->ipf_state_soft;
2447 int ret = 0, rev, out, flags, flx = 0, idx;
2453 * If a connection is about to be deleted, no packets
2454 * are allowed to match it.
2456 if (is->is_sti.tqe_ifq == &softs->ipf_state_deletetq)
2459 rev = IP6_NEQ(&is->is_dst, dst);
2462 flags = is->is_flags;
2467 sp = htons(fin->fin_sport);
2468 dp = ntohs(fin->fin_dport);
2472 if (!(flags & SI_W_SPORT) && (sp != is->is_sport))
2474 else if (!(flags & SI_W_DPORT) && (dp != is->is_dport))
2479 idx = (out << 1) + rev;
2482 * If the interface for this 'direction' is set, make sure it matches.
2483 * An interface name that is not set matches any, as does a name of *.
2485 if ((is->is_ifp[idx] == ifp) || (is->is_ifp[idx] == NULL &&
2486 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '-' ||
2487 *is->is_ifname[idx] == '*')))
2491 DT2(iss_lookup_badifp, fr_info_t *, fin, ipstate_t *, is);
2492 SBUMP(ipf_state_stats.iss_lookup_badifp);
2493 /* TRACE is, out, rev, idx */
2499 * Match addresses and ports.
2502 if ((IP6_EQ(&is->is_dst, dst) || (flags & SI_W_DADDR)) &&
2503 (IP6_EQ(&is->is_src, src) || (flags & SI_W_SADDR))) {
2505 if ((sp == is->is_sport || flags & SI_W_SPORT)
2507 (dp == is->is_dport || flags & SI_W_DPORT))
2514 if ((IP6_EQ(&is->is_dst, src) || (flags & SI_W_DADDR)) &&
2515 (IP6_EQ(&is->is_src, dst) || (flags & SI_W_SADDR))) {
2517 if ((dp == is->is_sport || flags & SI_W_SPORT)
2519 (sp == is->is_dport || flags & SI_W_DPORT))
2528 SBUMP(ipf_state_stats.iss_lookup_badport);
2529 DT2(iss_lookup_badport, fr_info_t *, fin, ipstate_t *, is);
2530 /* TRACE rev, is, sp, dp, src, dst */
2535 * Whether or not this should be here, is questionable, but the aim
2536 * is to get this out of the main line.
2539 flags = is->is_flags & ~(SI_WILDP|SI_NEWFR|SI_CLONE|SI_CLONED);
2542 * Only one of the source or destination address can be flaged as a
2543 * wildcard. Fill in the missing address, if set.
2544 * For IPv6, if the address being copied in is multicast, then
2545 * don't reset the wild flag - multicast causes it to be set in the
2548 if ((flags & (SI_W_SADDR|SI_W_DADDR))) {
2549 fr_ip_t *fi = &fin->fin_fi;
2551 if ((flags & SI_W_SADDR) != 0) {
2553 is->is_src = fi->fi_src;
2554 is->is_flags &= ~SI_W_SADDR;
2556 if (!(fin->fin_flx & (FI_MULTICAST|FI_MBCAST))){
2557 is->is_src = fi->fi_dst;
2558 is->is_flags &= ~SI_W_SADDR;
2561 } else if ((flags & SI_W_DADDR) != 0) {
2563 if (!(fin->fin_flx & (FI_MULTICAST|FI_MBCAST))){
2564 is->is_dst = fi->fi_dst;
2565 is->is_flags &= ~SI_W_DADDR;
2568 is->is_dst = fi->fi_src;
2569 is->is_flags &= ~SI_W_DADDR;
2572 if ((is->is_flags & (SI_WILDA|SI_WILDP)) == 0) {
2573 ATOMIC_DECL(softs->ipf_state_stats.iss_wild);
2577 flx = fin->fin_flx & cmask;
2578 cflx = is->is_flx[out][rev];
2581 * Match up any flags set from IP options.
2583 if ((cflx && (flx != (cflx & cmask))) ||
2584 ((fin->fin_optmsk & is->is_optmsk[rev]) != is->is_opt[rev]) ||
2585 ((fin->fin_secmsk & is->is_secmsk) != is->is_sec) ||
2586 ((fin->fin_auth & is->is_authmsk) != is->is_auth)) {
2587 SBUMPD(ipf_state_stats, iss_miss_mask);
2591 if ((fin->fin_flx & FI_IGNORE) != 0) {
2597 * Only one of the source or destination port can be flagged as a
2598 * wildcard. When filling it in, fill in a copy of the matched entry
2599 * if it has the cloning flag set.
2601 if ((flags & (SI_W_SPORT|SI_W_DPORT))) {
2602 if ((flags & SI_CLONE) != 0) {
2605 clone = ipf_state_clone(fin, tcp, is);
2610 ATOMIC_DECL(softs->ipf_state_stats.iss_wild);
2613 if ((flags & SI_W_SPORT) != 0) {
2616 is->is_send = ntohl(tcp->th_seq);
2619 is->is_send = ntohl(tcp->th_ack);
2621 is->is_maxsend = is->is_send + 1;
2622 } else if ((flags & SI_W_DPORT) != 0) {
2625 is->is_dend = ntohl(tcp->th_ack);
2628 is->is_dend = ntohl(tcp->th_seq);
2630 is->is_maxdend = is->is_dend + 1;
2632 is->is_flags &= ~(SI_W_SPORT|SI_W_DPORT);
2633 if ((flags & SI_CLONED) && softs->ipf_state_logging)
2634 ipf_state_log(softc, is, ISL_CLONE);
2639 if (is->is_flx[out][rev] == 0) {
2640 is->is_flx[out][rev] = flx;
2641 if (rev == 1 && is->is_optmsk[1] == 0) {
2642 is->is_opt[1] = fin->fin_optmsk;
2643 is->is_optmsk[1] = 0xffffffff;
2644 if (is->is_v == 6) {
2645 is->is_opt[1] &= ~0x8;
2646 is->is_optmsk[1] &= ~0x8;
2652 * Check if the interface name for this "direction" is set and if not,
2655 if (is->is_ifp[idx] == NULL &&
2656 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) {
2657 is->is_ifp[idx] = ifp;
2658 COPYIFNAME(fin->fin_v, ifp, is->is_ifname[idx]);
2665 /* ------------------------------------------------------------------------ */
2666 /* Function: ipf_checkicmpmatchingstate */
2668 /* Parameters: fin(I) - pointer to packet information */
2670 /* If we've got an ICMP error message, using the information stored in the */
2671 /* ICMP packet, look for a matching state table entry. */
2673 /* If we return NULL then no lock on ipf_state is held. */
2674 /* If we return non-null then a read-lock on ipf_state is held. */
2675 /* ------------------------------------------------------------------------ */
2677 ipf_checkicmpmatchingstate(fin)
2680 ipf_main_softc_t *softc = fin->fin_main_soft;
2681 ipf_state_softc_t *softs = softc->ipf_state_soft;
2682 ipstate_t *is, **isp;
2695 * Does it at least have the return (basic) IP header ?
2696 * Is it an actual recognised ICMP error type?
2697 * Only a basic IP header (no options) should be with
2698 * an ICMP error header.
2700 if ((fin->fin_v != 4) || (fin->fin_hlen != sizeof(ip_t)) ||
2701 (fin->fin_plen < ICMPERR_MINPKTLEN) ||
2702 !(fin->fin_flx & FI_ICMPERR)) {
2703 SBUMPD(ipf_state_stats, iss_icmp_bad);
2707 type = ic->icmp_type;
2709 oip = (ip_t *)((char *)ic + ICMPERR_ICMPHLEN);
2711 * Check if the at least the old IP header (with options) and
2712 * 8 bytes of payload is present.
2714 if (fin->fin_plen < ICMPERR_MAXPKTLEN + ((IP_HL(oip) - 5) << 2)) {
2715 SBUMPDX(ipf_state_stats, iss_icmp_short, iss_icmp_short_1);
2722 len = fin->fin_dlen - ICMPERR_ICMPHLEN;
2723 if ((len <= 0) || ((IP_HL(oip) << 2) > len)) {
2724 DT2(iss_icmp_len, fr_info_t *, fin, struct ip*, oip);
2725 SBUMPDX(ipf_state_stats, iss_icmp_short, iss_icmp_short_1);
2730 * Is the buffer big enough for all of it ? It's the size of the IP
2731 * header claimed in the encapsulated part which is of concern. It
2732 * may be too big to be in this buffer but not so big that it's
2733 * outside the ICMP packet, leading to TCP deref's causing problems.
2734 * This is possible because we don't know how big oip_hl is when we
2735 * do the pullup early in ipf_check() and thus can't guarantee it is
2743 # if defined(MENTAT)
2744 if ((char *)oip + len > (char *)m->b_wptr) {
2745 SBUMPDX(ipf_state_stats, iss_icmp_short, iss_icmp_short_2);
2749 if ((char *)oip + len > (char *)fin->fin_ip + m->m_len) {
2750 SBUMPDX(ipf_state_stats, iss_icmp_short, iss_icmp_short_3);
2757 bcopy((char *)fin, (char *)&ofin, sizeof(*fin));
2760 * in the IPv4 case we must zero the i6addr union otherwise
2761 * the IP6_EQ and IP6_NEQ macros produce the wrong results because
2762 * of the 'junk' in the unused part of the union
2764 bzero((char *)&src, sizeof(src));
2765 bzero((char *)&dst, sizeof(dst));
2768 * we make an fin entry to be able to feed it to
2769 * matchsrcdst note that not all fields are encessary
2770 * but this is the cleanest way. Note further we fill
2771 * in fin_mp such that if someone uses it we'll get
2772 * a kernel panic. ipf_matchsrcdst does not use this.
2774 * watch out here, as ip is in host order and oip in network
2775 * order. Any change we make must be undone afterwards, like
2778 savelen = oip->ip_len;
2779 oip->ip_len = htons(len);
2781 ofin.fin_flx = FI_NOCKSUM;
2784 ofin.fin_m = NULL; /* if dereferenced, panic XXX */
2785 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */
2786 (void) ipf_makefrip(IP_HL(oip) << 2, oip, &ofin);
2787 ofin.fin_ifp = fin->fin_ifp;
2788 ofin.fin_out = !fin->fin_out;
2790 hv = (pr = oip->ip_p);
2791 src.in4 = oip->ip_src;
2792 hv += src.in4.s_addr;
2793 dst.in4 = oip->ip_dst;
2794 hv += dst.in4.s_addr;
2797 * Reset the short and bad flag here because in ipf_matchsrcdst()
2798 * the flags for the current packet (fin_flx) are compared against
2799 * those for the existing session.
2801 ofin.fin_flx &= ~(FI_BAD|FI_SHORT);
2804 * Put old values of ip_len back as we don't know
2805 * if we have to forward the packet or process it again.
2807 oip->ip_len = savelen;
2813 * an ICMP error can only be generated as a result of an
2814 * ICMP query, not as the response on an ICMP error
2816 * XXX theoretically ICMP_ECHOREP and the other reply's are
2817 * ICMP query's as well, but adding them here seems strange XXX
2819 if ((ofin.fin_flx & FI_ICMPERR) != 0) {
2820 DT1(iss_icmp_icmperr, fr_info_t *, &ofin);
2821 SBUMP(ipf_state_stats.iss_icmp_icmperr);
2826 * perform a lookup of the ICMP packet in the state table
2828 icmp = (icmphdr_t *)((char *)oip + (IP_HL(oip) << 2));
2829 hv += icmp->icmp_id;
2830 hv = DOUBLE_HASH(hv);
2832 READ_ENTER(&softc->ipf_state);
2833 for (isp = &softs->ipf_state_table[hv];
2834 ((is = *isp) != NULL); ) {
2835 isp = &is->is_hnext;
2836 if ((is->is_p != pr) || (is->is_v != 4))
2838 if (is->is_pass & FR_NOICMPERR)
2841 is = ipf_matchsrcdst(&ofin, is, &src, &dst,
2843 if ((is != NULL) && !ipf_allowstateicmp(fin, is, &src))
2846 RWLOCK_EXIT(&softc->ipf_state);
2847 SBUMPDX(ipf_state_stats, iss_icmp_miss, iss_icmp_miss_1);
2853 SBUMPDX(ipf_state_stats, iss_icmp_miss, iss_icmp_miss_2);
2857 tcp = (tcphdr_t *)((char *)oip + (IP_HL(oip) << 2));
2859 hv += tcp->th_dport;;
2860 hv += tcp->th_sport;;
2861 hv = DOUBLE_HASH(hv);
2863 READ_ENTER(&softc->ipf_state);
2864 for (isp = &softs->ipf_state_table[hv]; ((is = *isp) != NULL); ) {
2865 isp = &is->is_hnext;
2867 * Only allow this icmp though if the
2868 * encapsulated packet was allowed through the
2869 * other way around. Note that the minimal amount
2870 * of info present does not allow for checking against
2871 * tcp internals such as seq and ack numbers. Only the
2872 * ports are known to be present and can be even if the
2873 * short flag is set.
2875 if ((is->is_p == pr) && (is->is_v == 4) &&
2876 (is = ipf_matchsrcdst(&ofin, is, &src, &dst,
2877 tcp, FI_ICMPCMP))) {
2878 if (ipf_allowstateicmp(fin, is, &src) == 0)
2882 RWLOCK_EXIT(&softc->ipf_state);
2883 SBUMPDX(ipf_state_stats, iss_icmp_miss, iss_icmp_miss_3);
2888 /* ------------------------------------------------------------------------ */
2889 /* Function: ipf_allowstateicmp */
2890 /* Returns: int - 1 = packet denied, 0 = packet allowed */
2891 /* Parameters: fin(I) - pointer to packet information */
2892 /* is(I) - pointer to state table entry */
2893 /* src(I) - source address to check permission for */
2895 /* For an ICMP packet that has so far matched a state table entry, check if */
2896 /* there are any further refinements that might mean we want to block this */
2897 /* packet. This code isn't specific to either IPv4 or IPv6. */
2898 /* ------------------------------------------------------------------------ */
2900 ipf_allowstateicmp(fin, is, src)
2905 ipf_main_softc_t *softc = fin->fin_main_soft;
2906 ipf_state_softc_t *softs = softc->ipf_state_soft;
2915 if (fr != NULL && fr->fr_icmpgrp != NULL) {
2916 savefr = fin->fin_fr;
2917 fin->fin_fr = fr->fr_icmpgrp->fg_start;
2919 ipass = ipf_scanlist(fin, softc->ipf_pass);
2920 fin->fin_fr = savefr;
2921 if (FR_ISBLOCK(ipass)) {
2922 SBUMPD(ipf_state_stats, iss_icmp_headblock);
2928 * i : the index of this packet (the icmp unreachable)
2929 * oi : the index of the original packet found in the
2930 * icmp header (i.e. the packet causing this icmp)
2931 * backward : original packet was backward compared to
2934 backward = IP6_NEQ(&is->is_src, src);
2935 fin->fin_rev = !backward;
2936 i = (!backward << 1) + fin->fin_out;
2937 oi = (backward << 1) + !fin->fin_out;
2939 if (is->is_pass & FR_NOICMPERR) {
2940 SBUMPD(ipf_state_stats, iss_icmp_banned);
2943 if (is->is_icmppkts[i] > is->is_pkts[oi]) {
2944 SBUMPD(ipf_state_stats, iss_icmp_toomany);
2948 DT2(iss_icmp_hits, fr_info_t *, fin, ipstate_t *, is);
2949 SBUMP(ipf_state_stats.iss_icmp_hits);
2950 is->is_icmppkts[i]++;
2953 * we deliberately do not touch the timeouts
2954 * for the accompanying state table entry.
2955 * It remains to be seen if that is correct. XXX
2961 /* ------------------------------------------------------------------------ */
2962 /* Function: ipf_ipsmove */
2964 /* Parameters: is(I) - pointer to state table entry */
2965 /* hv(I) - new hash value for state table entry */
2966 /* Write Locks: ipf_state */
2968 /* Move a state entry from one position in the hash table to another. */
2969 /* ------------------------------------------------------------------------ */
2971 ipf_ipsmove(softs, is, hv)
2972 ipf_state_softc_t *softs;
2981 /* TRACE is, is_hv, hvm */
2984 * Remove the hash from the old location...
2986 isp = is->is_phnext;
2988 is->is_hnext->is_phnext = isp;
2989 *isp = is->is_hnext;
2990 if (softs->ipf_state_table[hvm] == NULL)
2991 softs->ipf_state_stats.iss_inuse--;
2992 softs->ipf_state_stats.iss_bucketlen[hvm]--;
2995 * ...and put the hash in the new one.
2997 hvm = DOUBLE_HASH(hv);
3000 /* TRACE is, hv, is_hv, hvm */
3002 isp = &softs->ipf_state_table[hvm];
3004 (*isp)->is_phnext = &is->is_hnext;
3006 softs->ipf_state_stats.iss_inuse++;
3007 softs->ipf_state_stats.iss_bucketlen[hvm]++;
3008 is->is_phnext = isp;
3009 is->is_hnext = *isp;
3014 /* ------------------------------------------------------------------------ */
3015 /* Function: ipf_state_lookup */
3016 /* Returns: ipstate_t* - NULL == no matching state found, */
3017 /* else pointer to state information is returned */
3018 /* Parameters: fin(I) - pointer to packet information */
3019 /* tcp(I) - pointer to TCP/UDP header. */
3020 /* ifqp(O) - pointer for storing tailq timeout */
3022 /* Search the state table for a matching entry to the packet described by */
3023 /* the contents of *fin. For certain protocols, when a match is found the */
3024 /* timeout queue is also selected and stored in ifpq if it is non-NULL. */
3026 /* If we return NULL then no lock on ipf_state is held. */
3027 /* If we return non-null then a read-lock on ipf_state is held. */
3028 /* ------------------------------------------------------------------------ */
3030 ipf_state_lookup(fin, tcp, ifqp)
3035 ipf_main_softc_t *softc = fin->fin_main_soft;
3036 ipf_state_softc_t *softs = softc->ipf_state_soft;
3037 u_int hv, hvm, pr, v, tryagain;
3038 ipstate_t *is, **isp;
3039 u_short dport, sport;
3048 ic = (struct icmp *)tcp;
3049 hv = (pr = fin->fin_fi.fi_p);
3050 src = fin->fin_fi.fi_src;
3051 dst = fin->fin_fi.fi_dst;
3052 hv += src.in4.s_addr;
3053 hv += dst.in4.s_addr;
3055 v = fin->fin_fi.fi_v;
3058 hv += fin->fin_fi.fi_src.i6[1];
3059 hv += fin->fin_fi.fi_src.i6[2];
3060 hv += fin->fin_fi.fi_src.i6[3];
3062 if ((fin->fin_p == IPPROTO_ICMPV6) &&
3063 IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_dst.in6)) {
3064 hv -= dst.in4.s_addr;
3066 hv += fin->fin_fi.fi_dst.i6[1];
3067 hv += fin->fin_fi.fi_dst.i6[2];
3068 hv += fin->fin_fi.fi_dst.i6[3];
3073 (fin->fin_flx & (FI_MULTICAST|FI_BROADCAST|FI_MBCAST))) {
3074 if (fin->fin_out == 0) {
3075 hv -= src.in4.s_addr;
3077 hv -= dst.in4.s_addr;
3081 /* TRACE fin_saddr, fin_daddr, hv */
3084 * Search the hash table for matching packet header info.
3089 case IPPROTO_ICMPV6 :
3092 if ((ic->icmp_type == ICMP6_ECHO_REQUEST) ||
3093 (ic->icmp_type == ICMP6_ECHO_REPLY)) {
3097 READ_ENTER(&softc->ipf_state);
3099 hvm = DOUBLE_HASH(hv);
3100 for (isp = &softs->ipf_state_table[hvm];
3101 ((is = *isp) != NULL); ) {
3102 isp = &is->is_hnext;
3103 if ((is->is_p != pr) || (is->is_v != v))
3105 is = ipf_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP);
3107 ipf_matchicmpqueryreply(v, &is->is_icmp,
3108 ic, fin->fin_rev)) {
3110 ifq = &softs->ipf_state_icmpacktq;
3112 ifq = &softs->ipf_state_icmptq;
3118 if ((tryagain != 0) && !(is->is_flags & SI_W_DADDR)) {
3119 hv += fin->fin_fi.fi_src.i6[0];
3120 hv += fin->fin_fi.fi_src.i6[1];
3121 hv += fin->fin_fi.fi_src.i6[2];
3122 hv += fin->fin_fi.fi_src.i6[3];
3123 ipf_ipsmove(softs, is, hv);
3124 MUTEX_DOWNGRADE(&softc->ipf_state);
3128 RWLOCK_EXIT(&softc->ipf_state);
3131 * No matching icmp state entry. Perhaps this is a
3132 * response to another state entry.
3134 * XXX With some ICMP6 packets, the "other" address is already
3135 * in the packet, after the ICMP6 header, and this could be
3136 * used in place of the multicast address. However, taking
3137 * advantage of this requires some significant code changes
3138 * to handle the specific types where that is the case.
3140 if ((softs->ipf_state_stats.iss_wild != 0) &&
3141 ((fin->fin_flx & FI_NOWILD) == 0) &&
3142 (v == 6) && (tryagain == 0)) {
3143 hv -= fin->fin_fi.fi_src.i6[0];
3144 hv -= fin->fin_fi.fi_src.i6[1];
3145 hv -= fin->fin_fi.fi_src.i6[2];
3146 hv -= fin->fin_fi.fi_src.i6[3];
3148 WRITE_ENTER(&softc->ipf_state);
3152 is = ipf_checkicmp6matchingstate(fin);
3162 hv = DOUBLE_HASH(hv);
3163 READ_ENTER(&softc->ipf_state);
3164 for (isp = &softs->ipf_state_table[hv];
3165 ((is = *isp) != NULL); ) {
3166 isp = &is->is_hnext;
3167 if ((is->is_p != pr) || (is->is_v != v))
3169 is = ipf_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP);
3171 (ic->icmp_id == is->is_icmp.ici_id) &&
3172 ipf_matchicmpqueryreply(v, &is->is_icmp,
3173 ic, fin->fin_rev)) {
3175 ifq = &softs->ipf_state_icmpacktq;
3177 ifq = &softs->ipf_state_icmptq;
3182 RWLOCK_EXIT(&softc->ipf_state);
3189 sport = htons(fin->fin_data[0]);
3191 dport = htons(fin->fin_data[1]);
3195 READ_ENTER(&softc->ipf_state);
3197 hvm = DOUBLE_HASH(hv);
3201 for (isp = &softs->ipf_state_table[hvm];
3202 ((is = *isp) != NULL); ) {
3203 isp = &is->is_hnext;
3204 if ((is->is_p != pr) || (is->is_v != v))
3206 fin->fin_flx &= ~FI_OOW;
3207 is = ipf_matchsrcdst(fin, is, &src, &dst, tcp, FI_CMP);
3209 if (pr == IPPROTO_TCP) {
3210 if (!ipf_state_tcp(softc, softs, fin,
3212 oow |= fin->fin_flx & FI_OOW;
3221 !(is->is_flags & (SI_CLONE|SI_WILDP|SI_WILDA))) {
3224 ipf_ipsmove(softs, is, hv);
3225 MUTEX_DOWNGRADE(&softc->ipf_state);
3229 RWLOCK_EXIT(&softc->ipf_state);
3231 if ((softs->ipf_state_stats.iss_wild != 0) &&
3232 ((fin->fin_flx & FI_NOWILD) == 0)) {
3233 if (tryagain == 0) {
3236 } else if (tryagain == 1) {
3237 hv = fin->fin_fi.fi_p;
3239 * If we try to pretend this is a reply to a
3240 * multicast/broadcast packet then we need to
3241 * exclude part of the address from the hash
3244 if (fin->fin_out == 0) {
3245 hv += src.in4.s_addr;
3247 hv += dst.in4.s_addr;
3253 if (tryagain <= 2) {
3254 WRITE_ENTER(&softc->ipf_state);
3258 fin->fin_flx |= oow;
3264 if (GRE_REV(gre->gr_flags) == 1) {
3271 hvm = DOUBLE_HASH(hv);
3272 READ_ENTER(&softc->ipf_state);
3273 for (isp = &softs->ipf_state_table[hvm];
3274 ((is = *isp) != NULL); ) {
3275 isp = &is->is_hnext;
3276 if ((is->is_p != pr) || (is->is_v != v))
3278 is = ipf_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP);
3280 ifq = &softs->ipf_state_iptq;
3285 RWLOCK_EXIT(&softc->ipf_state);
3291 if (((is->is_sti.tqe_flags & TQE_RULEBASED) != 0) &&
3292 (is->is_tqehead[fin->fin_rev] != NULL))
3293 ifq = is->is_tqehead[fin->fin_rev];
3294 if (ifq != NULL && ifqp != NULL)
3297 SBUMP(ipf_state_stats.iss_lookup_miss);
3303 /* ------------------------------------------------------------------------ */
3304 /* Function: ipf_state_check */
3305 /* Returns: frentry_t* - NULL == search failed, */
3306 /* else pointer to rule for matching state */
3307 /* Parameters: fin(I) - pointer to packet information */
3308 /* passp(I) - pointer to filtering result flags */
3310 /* Check if a packet is associated with an entry in the state table. */
3311 /* ------------------------------------------------------------------------ */
3313 ipf_state_check(fin, passp)
3317 ipf_main_softc_t *softc = fin->fin_main_soft;
3318 ipf_state_softc_t *softs = softc->ipf_state_soft;
3327 if (softs->ipf_state_lock || (softs->ipf_state_list == NULL))
3330 if (fin->fin_flx & (FI_SHORT|FI_FRAGBODY|FI_BAD)) {
3331 SBUMPD(ipf_state_stats, iss_check_bad);
3335 if ((fin->fin_flx & FI_TCPUDP) ||
3336 (fin->fin_fi.fi_p == IPPROTO_ICMP)
3338 || (fin->fin_fi.fi_p == IPPROTO_ICMPV6)
3347 * Search the hash table for matching packet header info.
3349 is = ipf_state_lookup(fin, tcp, &ifq);
3354 case IPPROTO_ICMPV6 :
3357 if (fin->fin_v == 6) {
3358 is = ipf_checkicmp6matchingstate(fin);
3366 * No matching icmp state entry. Perhaps this is a
3367 * response to another state entry.
3369 is = ipf_checkicmpmatchingstate(fin);
3376 if (is->is_pass & FR_NEWISN) {
3377 if (fin->fin_out == 0)
3378 ipf_fixinisn(fin, is);
3379 else if (fin->fin_out == 1)
3380 ipf_fixoutisn(fin, is);
3385 ifq = &softs->ipf_state_udpacktq;
3387 ifq = &softs->ipf_state_udptq;
3391 SBUMP(ipf_state_stats.iss_check_miss);
3397 if ((fin->fin_out == 0) && (fr->fr_nattag.ipt_num[0] != 0)) {
3398 if (fin->fin_nattag == NULL) {
3399 RWLOCK_EXIT(&softc->ipf_state);
3400 SBUMPD(ipf_state_stats, iss_check_notag);
3403 if (ipf_matchtag(&fr->fr_nattag, fin->fin_nattag)!=0) {
3404 RWLOCK_EXIT(&softc->ipf_state);
3405 SBUMPD(ipf_state_stats, iss_check_nattag);
3409 (void) strncpy(fin->fin_group, FR_NAME(fr, fr_group),
3411 fin->fin_icode = fr->fr_icode;
3414 fin->fin_rule = is->is_rulen;
3418 * If this packet is a fragment and the rule says to track fragments,
3419 * then create a new fragment cache entry.
3421 if (fin->fin_flx & FI_FRAG && FR_ISPASS(is->is_pass) &&
3422 is->is_pass & FR_KEEPFRAG)
3423 (void) ipf_frag_new(softc, fin, is->is_pass);
3426 * For TCP packets, ifq == NULL. For all others, check if this new
3427 * queue is different to the last one it was on and move it if so.
3430 if ((tqe->tqe_flags & TQE_RULEBASED) != 0)
3431 ifq = is->is_tqehead[fin->fin_rev];
3433 MUTEX_ENTER(&is->is_lock);
3436 ipf_movequeue(softc->ipf_ticks, tqe, tqe->tqe_ifq, ifq);
3438 inout = (fin->fin_rev << 1) + fin->fin_out;
3439 is->is_pkts[inout]++;
3440 is->is_bytes[inout] += fin->fin_plen;
3441 fin->fin_pktnum = is->is_pkts[inout] + is->is_icmppkts[inout];
3443 MUTEX_EXIT(&is->is_lock);
3447 if (is->is_flags & IS_STATESYNC)
3448 ipf_sync_update(softc, SMC_STATE, fin, is->is_sync);
3450 RWLOCK_EXIT(&softc->ipf_state);
3452 SBUMP(ipf_state_stats.iss_hits);
3454 fin->fin_dif = &is->is_dif;
3455 fin->fin_tif = &is->is_tifs[fin->fin_rev];
3456 fin->fin_flx |= FI_STATE;
3457 if ((pass & FR_LOGFIRST) != 0)
3458 pass &= ~(FR_LOGFIRST|FR_LOG);
3464 /* ------------------------------------------------------------------------ */
3465 /* Function: ipf_fixoutisn */
3467 /* Parameters: fin(I) - pointer to packet information */
3468 /* is(I) - pointer to master state structure */
3470 /* Called only for outbound packets, adjusts the sequence number and the */
3471 /* TCP checksum to match that change. */
3472 /* ------------------------------------------------------------------------ */
3474 ipf_fixoutisn(fin, is)
3484 if ((is->is_flags & IS_ISNSYN) != 0) {
3485 if ((rev == 0) && (fin->fin_cksum < FI_CK_L4PART)) {
3486 seq = ntohl(tcp->th_seq);
3487 seq += is->is_isninc[0];
3488 tcp->th_seq = htonl(seq);
3489 ipf_fix_outcksum(0, &tcp->th_sum, is->is_sumd[0], 0);
3492 if ((is->is_flags & IS_ISNACK) != 0) {
3493 if ((rev == 1) && (fin->fin_cksum < FI_CK_L4PART)) {
3494 seq = ntohl(tcp->th_seq);
3495 seq += is->is_isninc[1];
3496 tcp->th_seq = htonl(seq);
3497 ipf_fix_outcksum(0, &tcp->th_sum, is->is_sumd[1], 0);
3503 /* ------------------------------------------------------------------------ */
3504 /* Function: ipf_fixinisn */
3506 /* Parameters: fin(I) - pointer to packet information */
3507 /* is(I) - pointer to master state structure */
3509 /* Called only for inbound packets, adjusts the acknowledge number and the */
3510 /* TCP checksum to match that change. */
3511 /* ------------------------------------------------------------------------ */
3513 ipf_fixinisn(fin, is)
3523 if ((is->is_flags & IS_ISNSYN) != 0) {
3524 if ((rev == 1) && (fin->fin_cksum < FI_CK_L4PART)) {
3525 ack = ntohl(tcp->th_ack);
3526 ack -= is->is_isninc[0];
3527 tcp->th_ack = htonl(ack);
3528 ipf_fix_incksum(0, &tcp->th_sum, is->is_sumd[0], 0);
3531 if ((is->is_flags & IS_ISNACK) != 0) {
3532 if ((rev == 0) && (fin->fin_cksum < FI_CK_L4PART)) {
3533 ack = ntohl(tcp->th_ack);
3534 ack -= is->is_isninc[1];
3535 tcp->th_ack = htonl(ack);
3536 ipf_fix_incksum(0, &tcp->th_sum, is->is_sumd[1], 0);
3542 /* ------------------------------------------------------------------------ */
3543 /* Function: ipf_state_sync */
3545 /* Parameters: softc(I) - pointer to soft context main structure */
3546 /* ifp(I) - pointer to interface */
3548 /* Walk through all state entries and if an interface pointer match is */
3549 /* found then look it up again, based on its name in case the pointer has */
3550 /* changed since last time. */
3552 /* If ifp is passed in as being non-null then we are only doing updates for */
3553 /* existing, matching, uses of it. */
3554 /* ------------------------------------------------------------------------ */
3556 ipf_state_sync(softc, ifp)
3557 ipf_main_softc_t *softc;
3560 ipf_state_softc_t *softs = softc->ipf_state_soft;
3564 if (softc->ipf_running <= 0)
3567 WRITE_ENTER(&softc->ipf_state);
3569 if (softc->ipf_running <= 0) {
3570 RWLOCK_EXIT(&softc->ipf_state);
3574 for (is = softs->ipf_state_list; is; is = is->is_next) {
3576 * Look up all the interface names in the state entry.
3578 for (i = 0; i < 4; i++) {
3579 if (ifp == NULL || ifp == is->is_ifp[i])
3580 is->is_ifp[i] = ipf_resolvenic(softc,
3585 RWLOCK_EXIT(&softc->ipf_state);
3589 /* ------------------------------------------------------------------------ */
3590 /* Function: ipf_state_del */
3591 /* Returns: int - 0 = deleted, else refernce count on active struct */
3592 /* Parameters: softc(I) - pointer to soft context main structure */
3593 /* is(I) - pointer to state structure to delete */
3594 /* why(I) - if not 0, log reason why it was deleted */
3595 /* Write Locks: ipf_state */
3597 /* Deletes a state entry from the enumerated list as well as the hash table */
3598 /* and timeout queue lists. Make adjustments to hash table statistics and */
3599 /* global counters as required. */
3600 /* ------------------------------------------------------------------------ */
3602 ipf_state_del(softc, is, why)
3603 ipf_main_softc_t *softc;
3607 ipf_state_softc_t *softs = softc->ipf_state_soft;
3612 * Since we want to delete this, remove it from the state table,
3613 * where it can be found & used, first.
3615 if (is->is_phnext != NULL) {
3616 *is->is_phnext = is->is_hnext;
3617 if (is->is_hnext != NULL)
3618 is->is_hnext->is_phnext = is->is_phnext;
3619 if (softs->ipf_state_table[is->is_hv] == NULL)
3620 softs->ipf_state_stats.iss_inuse--;
3621 softs->ipf_state_stats.iss_bucketlen[is->is_hv]--;
3623 is->is_phnext = NULL;
3624 is->is_hnext = NULL;
3629 * Because ipf_state_stats.iss_wild is a count of entries in the state
3630 * table that have wildcard flags set, only decerement it once
3633 if (is->is_flags & (SI_WILDP|SI_WILDA)) {
3634 if (!(is->is_flags & SI_CLONED)) {
3635 ATOMIC_DECL(softs->ipf_state_stats.iss_wild);
3637 is->is_flags &= ~(SI_WILDP|SI_WILDA);
3641 * Next, remove it from the timeout queue it is in.
3643 if (is->is_sti.tqe_ifq != NULL)
3644 ipf_deletequeueentry(&is->is_sti);
3647 * If it is still in use by something else, do not go any further,
3648 * but note that at this point it is now an orphan. How can this
3649 * be? ipf_state_flush() calls ipf_delete() directly because it wants
3650 * to empty the table out and if something has a hold on a state
3651 * entry (such as ipfstat), it'll do the deref path that'll bring
3652 * us back here to do the real delete & free.
3654 MUTEX_ENTER(&is->is_lock);
3655 if (is->is_me != NULL) {
3661 if (is->is_ref > 0) {
3665 MUTEX_EXIT(&is->is_lock);
3667 softs->ipf_state_stats.iss_orphan++;
3674 if (fr->fr_srctrack.ht_max_nodes != 0) {
3675 (void) ipf_ht_node_del(&fr->fr_srctrack,
3676 is->is_family, &is->is_src);
3680 ASSERT(is->is_ref == 0);
3681 MUTEX_EXIT(&is->is_lock);
3683 if (is->is_tqehead[0] != NULL) {
3684 if (ipf_deletetimeoutqueue(is->is_tqehead[0]) == 0)
3685 ipf_freetimeoutqueue(softc, is->is_tqehead[0]);
3687 if (is->is_tqehead[1] != NULL) {
3688 if (ipf_deletetimeoutqueue(is->is_tqehead[1]) == 0)
3689 ipf_freetimeoutqueue(softc, is->is_tqehead[1]);
3693 ipf_sync_del_state(softc->ipf_sync_soft, is->is_sync);
3696 * Now remove it from the linked list of known states
3698 if (is->is_pnext != NULL) {
3699 *is->is_pnext = is->is_next;
3701 if (is->is_next != NULL)
3702 is->is_next->is_pnext = is->is_pnext;
3704 is->is_pnext = NULL;
3708 if (softs->ipf_state_logging != 0 && why != 0)
3709 ipf_state_log(softc, is, why);
3711 if (is->is_p == IPPROTO_TCP)
3712 softs->ipf_state_stats.iss_fin++;
3714 softs->ipf_state_stats.iss_expire++;
3716 softs->ipf_state_stats.iss_orphan--;
3720 (void) ipf_derefrule(softc, &fr);
3723 softs->ipf_state_stats.iss_active_proto[is->is_p]--;
3725 MUTEX_DESTROY(&is->is_lock);
3727 softs->ipf_state_stats.iss_active--;
3733 /* ------------------------------------------------------------------------ */
3734 /* Function: ipf_state_expire */
3736 /* Parameters: softc(I) - pointer to soft context main structure */
3738 /* Slowly expire held state for thingslike UDP and ICMP. The algorithm */
3739 /* used here is to keep the queue sorted with the oldest things at the top */
3740 /* and the youngest at the bottom. So if the top one doesn't need to be */
3741 /* expired then neither will any under it. */
3742 /* ------------------------------------------------------------------------ */
3744 ipf_state_expire(softc)
3745 ipf_main_softc_t *softc;
3747 ipf_state_softc_t *softs = softc->ipf_state_soft;
3748 ipftq_t *ifq, *ifqnext;
3749 ipftqent_t *tqe, *tqn;
3754 WRITE_ENTER(&softc->ipf_state);
3755 for (ifq = softs->ipf_state_tcptq; ifq != NULL; ifq = ifq->ifq_next)
3756 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3757 if (tqe->tqe_die > softc->ipf_ticks)
3759 tqn = tqe->tqe_next;
3760 is = tqe->tqe_parent;
3761 ipf_state_del(softc, is, ISL_EXPIRE);
3764 for (ifq = softs->ipf_state_usertq; ifq != NULL; ifq = ifqnext) {
3765 ifqnext = ifq->ifq_next;
3767 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3768 if (tqe->tqe_die > softc->ipf_ticks)
3770 tqn = tqe->tqe_next;
3771 is = tqe->tqe_parent;
3772 ipf_state_del(softc, is, ISL_EXPIRE);
3776 for (ifq = softs->ipf_state_usertq; ifq != NULL; ifq = ifqnext) {
3777 ifqnext = ifq->ifq_next;
3779 if (((ifq->ifq_flags & IFQF_DELETE) != 0) &&
3780 (ifq->ifq_ref == 0)) {
3781 ipf_freetimeoutqueue(softc, ifq);
3785 if (softs->ipf_state_doflush) {
3786 (void) ipf_state_flush(softc, 2, 0);
3787 softs->ipf_state_doflush = 0;
3788 softs->ipf_state_wm_last = softc->ipf_ticks;
3791 RWLOCK_EXIT(&softc->ipf_state);
3796 /* ------------------------------------------------------------------------ */
3797 /* Function: ipf_state_flush */
3798 /* Returns: int - 0 == success, -1 == failure */
3799 /* Parameters: softc(I) - pointer to soft context main structure */
3800 /* which(I) - which flush action to perform */
3801 /* proto(I) - which protocol to flush (0 == ALL) */
3802 /* Write Locks: ipf_state */
3804 /* Flush state tables. Three actions currently defined: */
3805 /* which == 0 : flush all state table entries */
3806 /* which == 1 : flush TCP connections which have started to close but are */
3807 /* stuck for some reason. */
3808 /* which == 2 : flush TCP connections which have been idle for a long time, */
3809 /* starting at > 4 days idle and working back in successive half-*/
3810 /* days to at most 12 hours old. If this fails to free enough */
3811 /* slots then work backwards in half hour slots to 30 minutes. */
3812 /* If that too fails, then work backwards in 30 second intervals */
3813 /* for the last 30 minutes to at worst 30 seconds idle. */
3814 /* ------------------------------------------------------------------------ */
3816 ipf_state_flush(softc, which, proto)
3817 ipf_main_softc_t *softc;
3820 ipf_state_softc_t *softs = softc->ipf_state_soft;
3821 ipftqent_t *tqe, *tqn;
3822 ipstate_t *is, **isp;
3834 SBUMP(ipf_state_stats.iss_flush_all);
3836 * Style 0 flush removes everything...
3838 for (isp = &softs->ipf_state_list; ((is = *isp) != NULL); ) {
3839 if ((proto != 0) && (is->is_v != proto)) {
3843 if (ipf_state_del(softc, is, ISL_FLUSH) == 0)
3851 SBUMP(ipf_state_stats.iss_flush_closing);
3853 * Since we're only interested in things that are closing,
3854 * we can start with the appropriate timeout queue.
3856 for (ifq = softs->ipf_state_tcptq + IPF_TCPS_CLOSE_WAIT;
3857 ifq != NULL; ifq = ifq->ifq_next) {
3859 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3860 tqn = tqe->tqe_next;
3861 is = tqe->tqe_parent;
3862 if (is->is_p != IPPROTO_TCP)
3864 if (ipf_state_del(softc, is, ISL_FLUSH) == 0)
3870 * Also need to look through the user defined queues.
3872 for (ifq = softs->ipf_state_usertq; ifq != NULL;
3873 ifq = ifq->ifq_next) {
3874 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3875 tqn = tqe->tqe_next;
3876 is = tqe->tqe_parent;
3877 if (is->is_p != IPPROTO_TCP)
3880 if ((is->is_state[0] > IPF_TCPS_ESTABLISHED) &&
3881 (is->is_state[1] > IPF_TCPS_ESTABLISHED)) {
3882 if (ipf_state_del(softc, is,
3894 * Args 5-11 correspond to flushing those particular states
3895 * for TCP connections.
3897 case IPF_TCPS_CLOSE_WAIT :
3898 case IPF_TCPS_FIN_WAIT_1 :
3899 case IPF_TCPS_CLOSING :
3900 case IPF_TCPS_LAST_ACK :
3901 case IPF_TCPS_FIN_WAIT_2 :
3902 case IPF_TCPS_TIME_WAIT :
3903 case IPF_TCPS_CLOSED :
3904 SBUMP(ipf_state_stats.iss_flush_queue);
3905 tqn = softs->ipf_state_tcptq[which].ifq_head;
3906 while (tqn != NULL) {
3908 tqn = tqe->tqe_next;
3909 is = tqe->tqe_parent;
3910 if (ipf_state_del(softc, is, ISL_FLUSH) == 0)
3919 SBUMP(ipf_state_stats.iss_flush_state);
3921 * Take a large arbitrary number to mean the number of seconds
3922 * for which which consider to be the maximum value we'll allow
3923 * the expiration to be.
3925 which = IPF_TTLVAL(which);
3926 for (isp = &softs->ipf_state_list; ((is = *isp) != NULL); ) {
3927 if ((proto == 0) || (is->is_v == proto)) {
3928 if (softc->ipf_ticks - is->is_touched > which) {
3929 if (ipf_state_del(softc, is,
3946 SBUMP(ipf_state_stats.iss_flush_timeout);
3948 * Asked to remove inactive entries because the table is full, try
3949 * again, 3 times, if first attempt failed with a different criteria
3950 * each time. The order tried in must be in decreasing age.
3951 * Another alternative is to implement random drop and drop N entries
3952 * at random until N have been freed up.
3954 if (softc->ipf_ticks - softs->ipf_state_wm_last >
3955 softs->ipf_state_wm_freq) {
3956 removed = ipf_queueflush(softc, ipf_state_flush_entry,
3957 softs->ipf_state_tcptq,
3958 softs->ipf_state_usertq,
3959 &softs->ipf_state_stats.iss_active,
3960 softs->ipf_state_size,
3961 softs->ipf_state_wm_low);
3962 softs->ipf_state_wm_last = softc->ipf_ticks;
3970 /* ------------------------------------------------------------------------ */
3971 /* Function: ipf_state_flush_entry */
3972 /* Returns: int - 0 = entry deleted, else not deleted */
3973 /* Parameters: softc(I) - pointer to soft context main structure */
3974 /* entry(I) - pointer to state structure to delete */
3975 /* Write Locks: ipf_state */
3977 /* This function is a stepping stone between ipf_queueflush() and */
3978 /* ipf_state_del(). It is used so we can provide a uniform interface via */
3979 /* the ipf_queueflush() function. */
3980 /* ------------------------------------------------------------------------ */
3982 ipf_state_flush_entry(softc, entry)
3983 ipf_main_softc_t *softc;
3986 return ipf_state_del(softc, entry, ISL_FLUSH);
3990 /* ------------------------------------------------------------------------ */
3991 /* Function: ipf_tcp_age */
3992 /* Returns: int - 1 == state transition made, 0 == no change (rejected) */
3993 /* Parameters: tqe(I) - pointer to timeout queue information */
3994 /* fin(I) - pointer to packet information */
3995 /* tqtab(I) - TCP timeout queue table this is in */
3996 /* flags(I) - flags from state/NAT entry */
3997 /* ok(I) - can we advance state */
3999 /* Rewritten by Arjan de Vet <Arjan.deVet@adv.iae.nl>, 2000-07-29: */
4001 /* - (try to) base state transitions on real evidence only, */
4002 /* i.e. packets that are sent and have been received by ipfilter; */
4003 /* diagram 18.12 of TCP/IP volume 1 by W. Richard Stevens was used. */
4005 /* - deal with half-closed connections correctly; */
4007 /* - store the state of the source in state[0] such that ipfstat */
4008 /* displays the state as source/dest instead of dest/source; the calls */
4009 /* to ipf_tcp_age have been changed accordingly. */
4011 /* Internal Parameters: */
4013 /* state[0] = state of source (host that initiated connection) */
4014 /* state[1] = state of dest (host that accepted the connection) */
4016 /* dir == 0 : a packet from source to dest */
4017 /* dir == 1 : a packet from dest to source */
4019 /* A typical procession for a connection is as follows: */
4021 /* +--------------+-------------------+ */
4022 /* | Side '0' | Side '1' | */
4023 /* +--------------+-------------------+ */
4024 /* | 0 -> 1 (SYN) | | */
4025 /* | | 0 -> 2 (SYN-ACK) | */
4026 /* | 1 -> 3 (ACK) | | */
4027 /* | | 2 -> 4 (ACK-PUSH) | */
4028 /* | 3 -> 4 (ACK) | | */
4030 /* | | 4 -> 6 (FIN-ACK) | */
4031 /* | 4 -> 5 (ACK) | | */
4032 /* | | 6 -> 6 (ACK-PUSH) | */
4033 /* | 5 -> 5 (ACK) | | */
4034 /* | 5 -> 8 (FIN) | | */
4035 /* | | 6 -> 10 (ACK) | */
4036 /* +--------------+-------------------+ */
4038 /* Locking: it is assumed that the parent of the tqe structure is locked. */
4039 /* ------------------------------------------------------------------------ */
4041 ipf_tcp_age(tqe, fin, tqtab, flags, ok)
4047 ipf_main_softc_t *softc = fin->fin_main_soft;
4048 int dlen, ostate, nstate, rval, dir;
4056 tcpflags = tcp->th_flags;
4057 dlen = fin->fin_dlen - (TCP_OFF(tcp) << 2);
4058 ostate = tqe->tqe_state[1 - dir];
4059 nstate = tqe->tqe_state[dir];
4061 if (tcpflags & TH_RST) {
4062 if (!(tcpflags & TH_PUSH) && !dlen)
4063 nstate = IPF_TCPS_CLOSED;
4065 nstate = IPF_TCPS_CLOSE_WAIT;
4067 if (ostate <= IPF_TCPS_ESTABLISHED) {
4068 tqe->tqe_state[1 - dir] = IPF_TCPS_CLOSE_WAIT;
4074 case IPF_TCPS_LISTEN: /* 0 */
4075 if ((tcpflags & TH_OPENING) == TH_OPENING) {
4077 * 'dir' received an S and sends SA in
4078 * response, LISTEN -> SYN_RECEIVED
4080 nstate = IPF_TCPS_SYN_RECEIVED;
4082 } else if ((tcpflags & TH_OPENING) == TH_SYN) {
4083 /* 'dir' sent S, LISTEN -> SYN_SENT */
4084 nstate = IPF_TCPS_SYN_SENT;
4088 * the next piece of code makes it possible to get
4089 * already established connections into the state table
4090 * after a restart or reload of the filter rules; this
4091 * does not work when a strict 'flags S keep state' is
4092 * used for tcp connections of course
4094 if (((flags & IS_TCPFSM) == 0) &&
4095 ((tcpflags & TH_ACKMASK) == TH_ACK)) {
4097 * we saw an A, guess 'dir' is in ESTABLISHED
4102 case IPF_TCPS_LISTEN :
4103 case IPF_TCPS_SYN_RECEIVED :
4104 nstate = IPF_TCPS_HALF_ESTAB;
4107 case IPF_TCPS_HALF_ESTAB :
4108 case IPF_TCPS_ESTABLISHED :
4109 nstate = IPF_TCPS_ESTABLISHED;
4117 * TODO: besides regular ACK packets we can have other
4118 * packets as well; it is yet to be determined how we
4119 * should initialize the states in those cases
4123 case IPF_TCPS_SYN_SENT: /* 1 */
4124 if ((tcpflags & ~(TH_ECN|TH_CWR)) == TH_SYN) {
4126 * A retransmitted SYN packet. We do not reset
4127 * the timeout here to ipf_tcptimeout because a
4128 * connection connect timeout does not renew
4129 * after every packet that is sent. We need to
4130 * set rval so as to indicate the packet has
4131 * passed the check for its flags being valid
4132 * in the TCP FSM. Setting rval to 2 has the
4133 * result of not resetting the timeout.
4136 } else if ((tcpflags & (TH_SYN|TH_FIN|TH_ACK)) ==
4139 * we see an A from 'dir' which is in SYN_SENT
4140 * state: 'dir' sent an A in response to an SA
4141 * which it received, SYN_SENT -> ESTABLISHED
4143 nstate = IPF_TCPS_ESTABLISHED;
4145 } else if (tcpflags & TH_FIN) {
4147 * we see an F from 'dir' which is in SYN_SENT
4148 * state and wants to close its side of the
4149 * connection; SYN_SENT -> FIN_WAIT_1
4151 nstate = IPF_TCPS_FIN_WAIT_1;
4153 } else if ((tcpflags & TH_OPENING) == TH_OPENING) {
4155 * we see an SA from 'dir' which is already in
4156 * SYN_SENT state, this means we have a
4157 * simultaneous open; SYN_SENT -> SYN_RECEIVED
4159 nstate = IPF_TCPS_SYN_RECEIVED;
4164 case IPF_TCPS_SYN_RECEIVED: /* 2 */
4165 if ((tcpflags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) {
4167 * we see an A from 'dir' which was in
4168 * SYN_RECEIVED state so it must now be in
4169 * established state, SYN_RECEIVED ->
4172 nstate = IPF_TCPS_ESTABLISHED;
4174 } else if ((tcpflags & ~(TH_ECN|TH_CWR)) ==
4177 * We see an SA from 'dir' which is already in
4178 * SYN_RECEIVED state.
4181 } else if (tcpflags & TH_FIN) {
4183 * we see an F from 'dir' which is in
4184 * SYN_RECEIVED state and wants to close its
4185 * side of the connection; SYN_RECEIVED ->
4188 nstate = IPF_TCPS_FIN_WAIT_1;
4193 case IPF_TCPS_HALF_ESTAB: /* 3 */
4194 if (tcpflags & TH_FIN) {
4195 nstate = IPF_TCPS_FIN_WAIT_1;
4197 } else if ((tcpflags & TH_ACKMASK) == TH_ACK) {
4199 * If we've picked up a connection in mid
4200 * flight, we could be looking at a follow on
4201 * packet from the same direction as the one
4202 * that created this state. Recognise it but
4203 * do not advance the entire connection's
4208 case IPF_TCPS_LISTEN :
4209 case IPF_TCPS_SYN_SENT :
4210 case IPF_TCPS_SYN_RECEIVED :
4213 case IPF_TCPS_HALF_ESTAB :
4214 case IPF_TCPS_ESTABLISHED :
4215 nstate = IPF_TCPS_ESTABLISHED;
4224 case IPF_TCPS_ESTABLISHED: /* 4 */
4226 if (tcpflags & TH_FIN) {
4228 * 'dir' closed its side of the connection;
4229 * this gives us a half-closed connection;
4230 * ESTABLISHED -> FIN_WAIT_1
4232 if (ostate == IPF_TCPS_FIN_WAIT_1) {
4233 nstate = IPF_TCPS_CLOSING;
4235 nstate = IPF_TCPS_FIN_WAIT_1;
4237 } else if (tcpflags & TH_ACK) {
4239 * an ACK, should we exclude other flags here?
4241 if (ostate == IPF_TCPS_FIN_WAIT_1) {
4243 * We know the other side did an active
4244 * close, so we are ACKing the recvd
4245 * FIN packet (does the window matching
4246 * code guarantee this?) and go into
4247 * CLOSE_WAIT state; this gives us a
4248 * half-closed connection
4250 nstate = IPF_TCPS_CLOSE_WAIT;
4251 } else if (ostate < IPF_TCPS_CLOSE_WAIT) {
4253 * still a fully established
4254 * connection reset timeout
4256 nstate = IPF_TCPS_ESTABLISHED;
4261 case IPF_TCPS_CLOSE_WAIT: /* 5 */
4263 if (tcpflags & TH_FIN) {
4265 * application closed and 'dir' sent a FIN,
4266 * we're now going into LAST_ACK state
4268 nstate = IPF_TCPS_LAST_ACK;
4271 * we remain in CLOSE_WAIT because the other
4272 * side has closed already and we did not
4273 * close our side yet; reset timeout
4275 nstate = IPF_TCPS_CLOSE_WAIT;
4279 case IPF_TCPS_FIN_WAIT_1: /* 6 */
4281 if ((tcpflags & TH_ACK) &&
4282 ostate > IPF_TCPS_CLOSE_WAIT) {
4284 * if the other side is not active anymore
4285 * it has sent us a FIN packet that we are
4286 * ack'ing now with an ACK; this means both
4287 * sides have now closed the connection and
4288 * we go into TIME_WAIT
4291 * XXX: how do we know we really are ACKing
4292 * the FIN packet here? does the window code
4295 nstate = IPF_TCPS_LAST_ACK;
4298 * we closed our side of the connection
4299 * already but the other side is still active
4300 * (ESTABLISHED/CLOSE_WAIT); continue with
4301 * this half-closed connection
4303 nstate = IPF_TCPS_FIN_WAIT_1;
4307 case IPF_TCPS_CLOSING: /* 7 */
4308 if ((tcpflags & (TH_FIN|TH_ACK)) == TH_ACK) {
4309 nstate = IPF_TCPS_TIME_WAIT;
4314 case IPF_TCPS_LAST_ACK: /* 8 */
4315 if (tcpflags & TH_ACK) {
4319 * we cannot detect when we go out of LAST_ACK state
4320 * to CLOSED because that is based on the reception
4321 * of ACK packets; ipfilter can only detect that a
4322 * packet has been sent by a host
4326 case IPF_TCPS_FIN_WAIT_2: /* 9 */
4330 case IPF_TCPS_TIME_WAIT: /* 10 */
4331 /* we're in 2MSL timeout now */
4332 if (ostate == IPF_TCPS_LAST_ACK) {
4333 nstate = IPF_TCPS_CLOSED;
4340 case IPF_TCPS_CLOSED: /* 11 */
4345 #if !defined(_KERNEL)
4353 * If rval == 2 then do not update the queue position, but treat the
4354 * packet as being ok.
4358 else if (rval == 1) {
4360 tqe->tqe_state[dir] = nstate;
4361 if ((tqe->tqe_flags & TQE_RULEBASED) == 0)
4362 ipf_movequeue(softc->ipf_ticks, tqe, tqe->tqe_ifq,
4370 /* ------------------------------------------------------------------------ */
4371 /* Function: ipf_state_log */
4373 /* Parameters: softc(I) - pointer to soft context main structure */
4374 /* is(I) - pointer to state structure */
4375 /* type(I) - type of log entry to create */
4377 /* Creates a state table log entry using the state structure and type info. */
4378 /* passed in. Log packet/byte counts, source/destination address and other */
4379 /* protocol specific information. */
4380 /* ------------------------------------------------------------------------ */
4382 ipf_state_log(softc, is, type)
4383 ipf_main_softc_t *softc;
4394 * Copy information out of the ipstate_t structure and into the
4395 * structure used for logging.
4397 ipsl.isl_type = type;
4398 ipsl.isl_pkts[0] = is->is_pkts[0] + is->is_icmppkts[0];
4399 ipsl.isl_bytes[0] = is->is_bytes[0];
4400 ipsl.isl_pkts[1] = is->is_pkts[1] + is->is_icmppkts[1];
4401 ipsl.isl_bytes[1] = is->is_bytes[1];
4402 ipsl.isl_pkts[2] = is->is_pkts[2] + is->is_icmppkts[2];
4403 ipsl.isl_bytes[2] = is->is_bytes[2];
4404 ipsl.isl_pkts[3] = is->is_pkts[3] + is->is_icmppkts[3];
4405 ipsl.isl_bytes[3] = is->is_bytes[3];
4406 ipsl.isl_src = is->is_src;
4407 ipsl.isl_dst = is->is_dst;
4408 ipsl.isl_p = is->is_p;
4409 ipsl.isl_v = is->is_v;
4410 ipsl.isl_flags = is->is_flags;
4411 ipsl.isl_tag = is->is_tag;
4412 ipsl.isl_rulen = is->is_rulen;
4413 (void) strncpy(ipsl.isl_group, is->is_group, FR_GROUPLEN);
4415 if (ipsl.isl_p == IPPROTO_TCP || ipsl.isl_p == IPPROTO_UDP) {
4416 ipsl.isl_sport = is->is_sport;
4417 ipsl.isl_dport = is->is_dport;
4418 if (ipsl.isl_p == IPPROTO_TCP) {
4419 ipsl.isl_state[0] = is->is_state[0];
4420 ipsl.isl_state[1] = is->is_state[1];
4422 } else if (ipsl.isl_p == IPPROTO_ICMP) {
4423 ipsl.isl_itype = is->is_icmp.ici_type;
4424 } else if (ipsl.isl_p == IPPROTO_ICMPV6) {
4425 ipsl.isl_itype = is->is_icmp.ici_type;
4427 ipsl.isl_ps.isl_filler[0] = 0;
4428 ipsl.isl_ps.isl_filler[1] = 0;
4432 sizes[0] = sizeof(ipsl);
4435 (void) ipf_log_items(softc, IPL_LOGSTATE, NULL, items, sizes, types, 1);
4441 /* ------------------------------------------------------------------------ */
4442 /* Function: ipf_checkicmp6matchingstate */
4443 /* Returns: ipstate_t* - NULL == no match found, */
4444 /* else pointer to matching state entry */
4445 /* Parameters: fin(I) - pointer to packet information */
4446 /* Locks: NULL == no locks, else Read Lock on ipf_state */
4448 /* If we've got an ICMPv6 error message, using the information stored in */
4449 /* the ICMPv6 packet, look for a matching state table entry. */
4450 /* ------------------------------------------------------------------------ */
4452 ipf_checkicmp6matchingstate(fin)
4455 ipf_main_softc_t *softc = fin->fin_main_soft;
4456 ipf_state_softc_t *softs = softc->ipf_state_soft;
4457 struct icmp6_hdr *ic6, *oic;
4458 ipstate_t *is, **isp;
4459 u_short sport, dport;
4471 * Does it at least have the return (basic) IP header ?
4472 * Is it an actual recognised ICMP error type?
4473 * Only a basic IP header (no options) should be with
4474 * an ICMP error header.
4476 if ((fin->fin_v != 6) || (fin->fin_plen < ICMP6ERR_MINPKTLEN) ||
4477 !(fin->fin_flx & FI_ICMPERR)) {
4478 SBUMPD(ipf_state_stats, iss_icmp_bad);
4483 type = ic6->icmp6_type;
4485 oip6 = (ip6_t *)((char *)ic6 + ICMPERR_ICMPHLEN);
4486 if (fin->fin_plen < sizeof(*oip6)) {
4487 SBUMPD(ipf_state_stats, iss_icmp_short);
4491 bcopy((char *)fin, (char *)&ofin, sizeof(*fin));
4493 ofin.fin_ifp = fin->fin_ifp;
4494 ofin.fin_out = !fin->fin_out;
4495 ofin.fin_m = NULL; /* if dereferenced, panic XXX */
4496 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */
4499 * We make a fin entry to be able to feed it to
4500 * matchsrcdst. Note that not all fields are necessary
4501 * but this is the cleanest way. Note further we fill
4502 * in fin_mp such that if someone uses it we'll get
4503 * a kernel panic. ipf_matchsrcdst does not use this.
4505 * watch out here, as ip is in host order and oip6 in network
4506 * order. Any change we make must be undone afterwards.
4508 savelen = oip6->ip6_plen;
4509 oip6->ip6_plen = htons(fin->fin_dlen - ICMPERR_ICMPHLEN);
4510 ofin.fin_flx = FI_NOCKSUM;
4511 ofin.fin_ip = (ip_t *)oip6;
4512 (void) ipf_makefrip(sizeof(*oip6), (ip_t *)oip6, &ofin);
4513 ofin.fin_flx &= ~(FI_BAD|FI_SHORT);
4514 oip6->ip6_plen = savelen;
4518 * an ICMP error can never generate an ICMP error in response.
4520 if (ofin.fin_flx & FI_ICMPERR) {
4521 DT1(iss_icmp6_icmperr, fr_info_t *, &ofin);
4522 SBUMP(ipf_state_stats.iss_icmp6_icmperr);
4526 if (oip6->ip6_nxt == IPPROTO_ICMPV6) {
4529 * an ICMP error can only be generated as a result of an
4530 * ICMP query, not as the response on an ICMP error
4532 * XXX theoretically ICMP_ECHOREP and the other reply's are
4533 * ICMP query's as well, but adding them here seems strange XXX
4535 if (!(oic->icmp6_type & ICMP6_INFOMSG_MASK)) {
4536 DT1(iss_icmp6_notinfo, fr_info_t *, &ofin);
4537 SBUMP(ipf_state_stats.iss_icmp6_notinfo);
4542 * perform a lookup of the ICMP packet in the state table
4544 hv = (pr = oip6->ip6_nxt);
4545 src.in6 = oip6->ip6_src;
4546 hv += src.in4.s_addr;
4547 dst.in6 = oip6->ip6_dst;
4548 hv += dst.in4.s_addr;
4549 hv += oic->icmp6_id;
4550 hv += oic->icmp6_seq;
4551 hv = DOUBLE_HASH(hv);
4553 READ_ENTER(&softc->ipf_state);
4554 for (isp = &softs->ipf_state_table[hv];
4555 ((is = *isp) != NULL); ) {
4557 isp = &is->is_hnext;
4558 if ((is->is_p == pr) &&
4559 !(is->is_pass & FR_NOICMPERR) &&
4560 (oic->icmp6_id == ic->ici_id) &&
4561 (oic->icmp6_seq == ic->ici_seq) &&
4562 (is = ipf_matchsrcdst(&ofin, is, &src,
4563 &dst, NULL, FI_ICMPCMP))) {
4565 * in the state table ICMP query's are stored
4566 * with the type of the corresponding ICMP
4567 * response. Correct here
4569 if (((ic->ici_type == ICMP6_ECHO_REPLY) &&
4570 (oic->icmp6_type == ICMP6_ECHO_REQUEST)) ||
4571 (ic->ici_type - 1 == oic->icmp6_type )) {
4572 if (!ipf_allowstateicmp(fin, is, &src))
4577 RWLOCK_EXIT(&softc->ipf_state);
4578 SBUMPD(ipf_state_stats, iss_icmp6_miss);
4582 hv = (pr = oip6->ip6_nxt);
4583 src.in6 = oip6->ip6_src;
4588 dst.in6 = oip6->ip6_dst;
4596 switch (oip6->ip6_nxt)
4600 tcp = (tcphdr_t *)(oip6 + 1);
4601 dport = tcp->th_dport;
4602 sport = tcp->th_sport;
4607 case IPPROTO_ICMPV6 :
4608 oic = (struct icmp6_hdr *)(oip6 + 1);
4609 hv += oic->icmp6_id;
4610 hv += oic->icmp6_seq;
4617 hv = DOUBLE_HASH(hv);
4619 READ_ENTER(&softc->ipf_state);
4620 for (isp = &softs->ipf_state_table[hv]; ((is = *isp) != NULL); ) {
4621 isp = &is->is_hnext;
4623 * Only allow this icmp though if the
4624 * encapsulated packet was allowed through the
4625 * other way around. Note that the minimal amount
4626 * of info present does not allow for checking against
4627 * tcp internals such as seq and ack numbers.
4629 if ((is->is_p != pr) || (is->is_v != 6) ||
4630 (is->is_pass & FR_NOICMPERR))
4632 is = ipf_matchsrcdst(&ofin, is, &src, &dst, tcp, FI_ICMPCMP);
4633 if ((is != NULL) && (ipf_allowstateicmp(fin, is, &src) == 0))
4636 RWLOCK_EXIT(&softc->ipf_state);
4637 SBUMPD(ipf_state_stats, iss_icmp_miss);
4643 /* ------------------------------------------------------------------------ */
4644 /* Function: ipf_sttab_init */
4646 /* Parameters: softc(I) - pointer to soft context main structure */
4647 /* tqp(I) - pointer to an array of timeout queues for TCP */
4649 /* Initialise the array of timeout queues for TCP. */
4650 /* ------------------------------------------------------------------------ */
4652 ipf_sttab_init(softc, tqp)
4653 ipf_main_softc_t *softc;
4658 for (i = IPF_TCP_NSTATES - 1; i >= 0; i--) {
4659 IPFTQ_INIT(&tqp[i], 0, "ipftq tcp tab");
4660 tqp[i].ifq_next = tqp + i + 1;
4662 tqp[IPF_TCP_NSTATES - 1].ifq_next = NULL;
4663 tqp[IPF_TCPS_CLOSED].ifq_ttl = softc->ipf_tcpclosed;
4664 tqp[IPF_TCPS_LISTEN].ifq_ttl = softc->ipf_tcptimeout;
4665 tqp[IPF_TCPS_SYN_SENT].ifq_ttl = softc->ipf_tcpsynsent;
4666 tqp[IPF_TCPS_SYN_RECEIVED].ifq_ttl = softc->ipf_tcpsynrecv;
4667 tqp[IPF_TCPS_ESTABLISHED].ifq_ttl = softc->ipf_tcpidletimeout;
4668 tqp[IPF_TCPS_CLOSE_WAIT].ifq_ttl = softc->ipf_tcphalfclosed;
4669 tqp[IPF_TCPS_FIN_WAIT_1].ifq_ttl = softc->ipf_tcphalfclosed;
4670 tqp[IPF_TCPS_CLOSING].ifq_ttl = softc->ipf_tcptimeout;
4671 tqp[IPF_TCPS_LAST_ACK].ifq_ttl = softc->ipf_tcplastack;
4672 tqp[IPF_TCPS_FIN_WAIT_2].ifq_ttl = softc->ipf_tcpclosewait;
4673 tqp[IPF_TCPS_TIME_WAIT].ifq_ttl = softc->ipf_tcptimewait;
4674 tqp[IPF_TCPS_HALF_ESTAB].ifq_ttl = softc->ipf_tcptimeout;
4678 /* ------------------------------------------------------------------------ */
4679 /* Function: ipf_sttab_destroy */
4681 /* Parameters: tqp(I) - pointer to an array of timeout queues for TCP */
4683 /* Do whatever is necessary to "destroy" each of the entries in the array */
4684 /* of timeout queues for TCP. */
4685 /* ------------------------------------------------------------------------ */
4687 ipf_sttab_destroy(tqp)
4692 for (i = IPF_TCP_NSTATES - 1; i >= 0; i--)
4693 MUTEX_DESTROY(&tqp[i].ifq_lock);
4697 /* ------------------------------------------------------------------------ */
4698 /* Function: ipf_state_deref */
4700 /* Parameters: softc(I) - pointer to soft context main structure */
4701 /* isp(I) - pointer to pointer to state table entry */
4703 /* Decrement the reference counter for this state table entry and free it */
4704 /* if there are no more things using it. */
4706 /* This function is only called when cleaning up after increasing is_ref by */
4707 /* one earlier in the 'code path' so if is_ref is 1 when entering, we do */
4708 /* have an orphan, otherwise not. However there is a possible race between */
4709 /* the entry being deleted via flushing with an ioctl call (that calls the */
4710 /* delete function directly) and the tail end of packet processing so we */
4711 /* need to grab is_lock before doing the check to synchronise the two code */
4714 /* When operating in userland (ipftest), we have no timers to clear a state */
4715 /* entry. Therefore, we make a few simple tests before deleting an entry */
4716 /* outright. We compare states on each side looking for a combination of */
4717 /* TIME_WAIT (should really be FIN_WAIT_2?) and LAST_ACK. Then we factor */
4718 /* in packet direction with the interface list to make sure we don't */
4719 /* prematurely delete an entry on a final inbound packet that's we're also */
4720 /* supposed to route elsewhere. */
4722 /* Internal parameters: */
4723 /* state[0] = state of source (host that initiated connection) */
4724 /* state[1] = state of dest (host that accepted the connection) */
4726 /* dir == 0 : a packet from source to dest */
4727 /* dir == 1 : a packet from dest to source */
4728 /* ------------------------------------------------------------------------ */
4730 ipf_state_deref(softc, isp)
4731 ipf_main_softc_t *softc;
4734 ipstate_t *is = *isp;
4739 MUTEX_ENTER(&is->is_lock);
4740 if (is->is_ref > 1) {
4742 MUTEX_EXIT(&is->is_lock);
4744 if ((is->is_sti.tqe_state[0] > IPF_TCPS_ESTABLISHED) ||
4745 (is->is_sti.tqe_state[1] > IPF_TCPS_ESTABLISHED)) {
4746 ipf_state_del(softc, is, ISL_EXPIRE);
4751 MUTEX_EXIT(&is->is_lock);
4753 WRITE_ENTER(&softc->ipf_state);
4754 ipf_state_del(softc, is, ISL_ORPHAN);
4755 RWLOCK_EXIT(&softc->ipf_state);
4759 /* ------------------------------------------------------------------------ */
4760 /* Function: ipf_state_setqueue */
4762 /* Parameters: softc(I) - pointer to soft context main structure */
4763 /* is(I) - pointer to state structure */
4764 /* rev(I) - forward(0) or reverse(1) direction */
4765 /* Locks: ipf_state (read or write) */
4767 /* Put the state entry on its default queue entry, using rev as a helped in */
4768 /* determining which queue it should be placed on. */
4769 /* ------------------------------------------------------------------------ */
4771 ipf_state_setqueue(softc, is, rev)
4772 ipf_main_softc_t *softc;
4776 ipf_state_softc_t *softs = softc->ipf_state_soft;
4777 ipftq_t *oifq, *nifq;
4779 if ((is->is_sti.tqe_flags & TQE_RULEBASED) != 0)
4780 nifq = is->is_tqehead[rev];
4788 case IPPROTO_ICMPV6 :
4790 nifq = &softs->ipf_state_icmpacktq;
4792 nifq = &softs->ipf_state_icmptq;
4797 nifq = &softs->ipf_state_icmpacktq;
4799 nifq = &softs->ipf_state_icmptq;
4802 nifq = softs->ipf_state_tcptq + is->is_state[rev];
4807 nifq = &softs->ipf_state_udpacktq;
4809 nifq = &softs->ipf_state_udptq;
4813 nifq = &softs->ipf_state_iptq;
4818 oifq = is->is_sti.tqe_ifq;
4820 * If it's currently on a timeout queue, move it from one queue to
4821 * another, else put it on the end of the newly determined queue.
4824 ipf_movequeue(softc->ipf_ticks, &is->is_sti, oifq, nifq);
4826 ipf_queueappend(softc->ipf_ticks, &is->is_sti, nifq, is);
4831 /* ------------------------------------------------------------------------ */
4832 /* Function: ipf_state_iter */
4833 /* Returns: int - 0 == success, else error */
4834 /* Parameters: softc(I) - pointer to main soft context */
4835 /* token(I) - pointer to ipftoken structure */
4836 /* itp(I) - pointer to ipfgeniter structure */
4837 /* obj(I) - pointer to data description structure */
4839 /* This function handles the SIOCGENITER ioctl for the state tables and */
4840 /* walks through the list of entries in the state table list (softs->ipf_state_list.) */
4841 /* ------------------------------------------------------------------------ */
4843 ipf_state_iter(softc, token, itp, obj)
4844 ipf_main_softc_t *softc;
4849 ipf_state_softc_t *softs = softc->ipf_state_soft;
4850 ipstate_t *is, *next, zero;
4853 if (itp->igi_data == NULL) {
4858 if (itp->igi_nitems < 1) {
4863 if (itp->igi_type != IPFGENITER_STATE) {
4868 is = token->ipt_data;
4869 if (is == (void *)-1) {
4875 obj->ipfo_type = IPFOBJ_IPSTATE;
4876 obj->ipfo_size = sizeof(ipstate_t);
4878 READ_ENTER(&softc->ipf_state);
4880 is = token->ipt_data;
4882 next = softs->ipf_state_list;
4888 * If we find a state entry to use, bump its reference count so that
4889 * it can be used for is_next when we come back.
4892 MUTEX_ENTER(&next->is_lock);
4894 MUTEX_EXIT(&next->is_lock);
4895 token->ipt_data = next;
4897 bzero(&zero, sizeof(zero));
4899 token->ipt_data = NULL;
4901 if (next->is_next == NULL)
4902 ipf_token_mark_complete(token);
4904 RWLOCK_EXIT(&softc->ipf_state);
4906 obj->ipfo_ptr = itp->igi_data;
4907 error = ipf_outobjk(softc, obj, next);
4909 ipf_state_deref(softc, &is);
4915 /* ------------------------------------------------------------------------ */
4916 /* Function: ipf_state_gettable */
4917 /* Returns: int - 0 = success, else error */
4918 /* Parameters: softc(I) - pointer to main soft context */
4919 /* softs(I) - pointer to state context structure */
4920 /* data(I) - pointer to ioctl data */
4922 /* This function handles ioctl requests for tables of state information. */
4923 /* At present the only table it deals with is the hash bucket statistics. */
4924 /* ------------------------------------------------------------------------ */
4926 ipf_state_gettable(softc, softs, data)
4927 ipf_main_softc_t *softc;
4928 ipf_state_softc_t *softs;
4934 error = ipf_inobj(softc, data, NULL, &table, IPFOBJ_GTABLE);
4938 if (table.ita_type != IPFTABLE_BUCKETS) {
4943 error = COPYOUT(softs->ipf_state_stats.iss_bucketlen, table.ita_table,
4944 softs->ipf_state_size * sizeof(u_int));
4953 /* ------------------------------------------------------------------------ */
4954 /* Function: ipf_state_setpending */
4956 /* Parameters: softc(I) - pointer to main soft context */
4957 /* is(I) - pointer to state structure */
4958 /* Locks: ipf_state (read or write) */
4960 /* Put the state entry on to the pending queue - this queue has a very */
4961 /* short lifetime where items are put that can't be deleted straight away */
4962 /* because of locking issues but we want to delete them ASAP, anyway. */
4963 /* ------------------------------------------------------------------------ */
4965 ipf_state_setpending(softc, is)
4966 ipf_main_softc_t *softc;
4969 ipf_state_softc_t *softs = softc->ipf_state_soft;
4972 oifq = is->is_sti.tqe_ifq;
4974 ipf_movequeue(softc->ipf_ticks, &is->is_sti, oifq,
4975 &softs->ipf_state_pending);
4977 ipf_queueappend(softc->ipf_ticks, &is->is_sti,
4978 &softs->ipf_state_pending, is);
4980 MUTEX_ENTER(&is->is_lock);
4981 if (is->is_me != NULL) {
4986 MUTEX_EXIT(&is->is_lock);
4990 /* ------------------------------------------------------------------------ */
4991 /* Function: ipf_state_matchflush */
4993 /* Parameters: softc(I) - pointer to main soft context */
4994 /* data(I) - pointer to state structure */
4995 /* Locks: ipf_state (read or write) */
4997 /* Flush all entries from the list of state entries that match the */
4998 /* properties in the array loaded. */
4999 /* ------------------------------------------------------------------------ */
5001 ipf_state_matchflush(softc, data)
5002 ipf_main_softc_t *softc;
5005 ipf_state_softc_t *softs = softc->ipf_state_soft;
5006 int *array, flushed, error;
5007 ipstate_t *state, *statenext;
5010 error = ipf_matcharray_load(softc, data, &obj, &array);
5016 for (state = softs->ipf_state_list; state != NULL; state = statenext) {
5017 statenext = state->is_next;
5018 if (ipf_state_matcharray(state, array, softc->ipf_ticks) == 0) {
5019 ipf_state_del(softc, state, ISL_FLUSH);
5024 obj.ipfo_retval = flushed;
5025 error = BCOPYOUT(&obj, data, sizeof(obj));
5027 KFREES(array, array[0] * sizeof(*array));
5033 /* ------------------------------------------------------------------------ */
5034 /* Function: ipf_state_matcharray */
5035 /* Returns: int - 0 = no match, 1 = match */
5036 /* Parameters: state(I) - pointer to state structure */
5037 /* array(I) - pointer to ipf matching expression */
5038 /* ticks(I) - current value of ipfilter tick timer */
5039 /* Locks: ipf_state (read or write) */
5041 /* Compare a state entry with the match array passed in and return a value */
5042 /* to indicate whether or not the matching was successful. */
5043 /* ------------------------------------------------------------------------ */
5045 ipf_state_matcharray(state, array, ticks)
5050 int i, n, *x, rv, p;
5057 for (; n > 0; x += 3 + x[3], rv = 0) {
5060 if (x[0] == IPF_EXP_END)
5064 * If we need to match the protocol and that doesn't match,
5065 * don't even both with the instruction array.
5067 p = e->ipfe_cmd >> 16;
5068 if ((p != 0) && (p != state->is_p))
5071 switch (e->ipfe_cmd)
5073 case IPF_EXP_IP_PR :
5074 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5075 rv |= (state->is_p == e->ipfe_arg0[i]);
5079 case IPF_EXP_IP_SRCADDR :
5080 if (state->is_v != 4)
5082 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5083 rv |= ((state->is_saddr &
5084 e->ipfe_arg0[i * 2 + 1]) ==
5085 e->ipfe_arg0[i * 2]);
5089 case IPF_EXP_IP_DSTADDR :
5090 if (state->is_v != 4)
5092 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5093 rv |= ((state->is_daddr &
5094 e->ipfe_arg0[i * 2 + 1]) ==
5095 e->ipfe_arg0[i * 2]);
5099 case IPF_EXP_IP_ADDR :
5100 if (state->is_v != 4)
5102 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5103 rv |= ((state->is_saddr &
5104 e->ipfe_arg0[i * 2 + 1]) ==
5105 e->ipfe_arg0[i * 2]) ||
5107 e->ipfe_arg0[i * 2 + 1]) ==
5108 e->ipfe_arg0[i * 2]);
5113 case IPF_EXP_IP6_SRCADDR :
5114 if (state->is_v != 6)
5116 for (i = 0; !rv && i < x[3]; i++) {
5117 rv |= IP6_MASKEQ(&state->is_src.in6,
5118 &e->ipfe_arg0[i * 8 + 4],
5119 &e->ipfe_arg0[i * 8]);
5123 case IPF_EXP_IP6_DSTADDR :
5124 if (state->is_v != 6)
5126 for (i = 0; !rv && i < x[3]; i++) {
5127 rv |= IP6_MASKEQ(&state->is_dst.in6,
5128 &e->ipfe_arg0[i * 8 + 4],
5129 &e->ipfe_arg0[i * 8]);
5133 case IPF_EXP_IP6_ADDR :
5134 if (state->is_v != 6)
5136 for (i = 0; !rv && i < x[3]; i++) {
5137 rv |= IP6_MASKEQ(&state->is_src.in6,
5138 &e->ipfe_arg0[i * 8 + 4],
5139 &e->ipfe_arg0[i * 8]) ||
5140 IP6_MASKEQ(&state->is_dst.in6,
5141 &e->ipfe_arg0[i * 8 + 4],
5142 &e->ipfe_arg0[i * 8]);
5147 case IPF_EXP_UDP_PORT :
5148 case IPF_EXP_TCP_PORT :
5149 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5150 rv |= (state->is_sport == e->ipfe_arg0[i]) ||
5151 (state->is_dport == e->ipfe_arg0[i]);
5155 case IPF_EXP_UDP_SPORT :
5156 case IPF_EXP_TCP_SPORT :
5157 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5158 rv |= (state->is_sport == e->ipfe_arg0[i]);
5162 case IPF_EXP_UDP_DPORT :
5163 case IPF_EXP_TCP_DPORT :
5164 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5165 rv |= (state->is_dport == e->ipfe_arg0[i]);
5169 case IPF_EXP_TCP_STATE :
5170 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5171 rv |= (state->is_state[0] == e->ipfe_arg0[i]) ||
5172 (state->is_state[1] == e->ipfe_arg0[i]);
5176 case IPF_EXP_IDLE_GT :
5177 rv |= (ticks - state->is_touched > e->ipfe_arg0[0]);
5182 * Factor in doing a negative match.
5194 /* ------------------------------------------------------------------------ */
5195 /* Function: ipf_state_settimeout */
5196 /* Returns: int 0 = success, else failure */
5197 /* Parameters: softc(I) - pointer to main soft context */
5198 /* t(I) - pointer to tuneable being changed */
5199 /* p(I) - pointer to the new value */
5201 /* Sets a timeout value for one of the many timeout queues. We find the */
5202 /* correct queue using a somewhat manual process of comparing the timeout */
5203 /* names for each specific value available and calling ipf_apply_timeout on */
5204 /* that queue so that all of the items on it are updated accordingly. */
5205 /* ------------------------------------------------------------------------ */
5207 ipf_state_settimeout(softc, t, p)
5208 struct ipf_main_softc_s *softc;
5212 ipf_state_softc_t *softs = softc->ipf_state_soft;
5215 * In case there is nothing to do...
5217 if (*t->ipft_pint == p->ipftu_int)
5220 if (!strncmp(t->ipft_name, "tcp_", 4))
5221 return ipf_settimeout_tcp(t, p, softs->ipf_state_tcptq);
5223 if (!strcmp(t->ipft_name, "udp_timeout")) {
5224 ipf_apply_timeout(&softs->ipf_state_udptq, p->ipftu_int);
5225 } else if (!strcmp(t->ipft_name, "udp_ack_timeout")) {
5226 ipf_apply_timeout(&softs->ipf_state_udpacktq, p->ipftu_int);
5227 } else if (!strcmp(t->ipft_name, "icmp_timeout")) {
5228 ipf_apply_timeout(&softs->ipf_state_icmptq, p->ipftu_int);
5229 } else if (!strcmp(t->ipft_name, "icmp_ack_timeout")) {
5230 ipf_apply_timeout(&softs->ipf_state_icmpacktq, p->ipftu_int);
5231 } else if (!strcmp(t->ipft_name, "ip_timeout")) {
5232 ipf_apply_timeout(&softs->ipf_state_iptq, p->ipftu_int);
5239 * Update the tuneable being set.
5241 *t->ipft_pint = p->ipftu_int;
5247 /* ------------------------------------------------------------------------ */
5248 /* Function: ipf_state_rehash */
5249 /* Returns: int 0 = success, else failure */
5250 /* Parameters: softc(I) - pointer to main soft context */
5251 /* t(I) - pointer to tuneable being changed */
5252 /* p(I) - pointer to the new value */
5254 /* To change the size of the state hash table at runtime, a new table has */
5255 /* to be allocated and then all of the existing entries put in it, bumping */
5256 /* up the bucketlength for it as we go along. */
5257 /* ------------------------------------------------------------------------ */
5259 ipf_state_rehash(softc, t, p)
5260 ipf_main_softc_t *softc;
5264 ipf_state_softc_t *softs = softc->ipf_state_soft;
5265 ipstate_t **newtab, *is;
5273 newsize = p->ipftu_int;
5275 * In case there is nothing to do...
5277 if (newsize == softs->ipf_state_size)
5280 KMALLOCS(newtab, ipstate_t **, newsize * sizeof(ipstate_t *));
5281 if (newtab == NULL) {
5286 KMALLOCS(bucketlens, u_int *, newsize * sizeof(u_int));
5287 if (bucketlens == NULL) {
5288 KFREES(newtab, newsize * sizeof(*softs->ipf_state_table));
5293 newseed = ipf_state_seed_alloc(newsize, softs->ipf_state_max);
5294 if (newseed == NULL) {
5295 KFREES(bucketlens, newsize * sizeof(*bucketlens));
5296 KFREES(newtab, newsize * sizeof(*newtab));
5301 for (maxbucket = 0, i = newsize; i > 0; i >>= 1)
5305 bzero((char *)newtab, newsize * sizeof(ipstate_t *));
5306 bzero((char *)bucketlens, newsize * sizeof(u_int));
5308 WRITE_ENTER(&softc->ipf_state);
5310 if (softs->ipf_state_table != NULL) {
5311 KFREES(softs->ipf_state_table,
5312 softs->ipf_state_size * sizeof(*softs->ipf_state_table));
5314 softs->ipf_state_table = newtab;
5316 if (softs->ipf_state_seed != NULL) {
5317 KFREES(softs->ipf_state_seed,
5318 softs->ipf_state_size * sizeof(*softs->ipf_state_seed));
5320 softs->ipf_state_seed = newseed;
5322 if (softs->ipf_state_stats.iss_bucketlen != NULL) {
5323 KFREES(softs->ipf_state_stats.iss_bucketlen,
5324 softs->ipf_state_size * sizeof(u_int));
5326 softs->ipf_state_stats.iss_bucketlen = bucketlens;
5327 softs->ipf_state_maxbucket = maxbucket;
5328 softs->ipf_state_size = newsize;
5331 * Walk through the entire list of state table entries and put them
5332 * in the new state table, somewhere. Because we have a new table,
5333 * we need to restart the counter of how many chains are in use.
5335 softs->ipf_state_stats.iss_inuse = 0;
5336 for (is = softs->ipf_state_list; is != NULL; is = is->is_next) {
5337 is->is_hnext = NULL;
5338 is->is_phnext = NULL;
5339 hv = is->is_hv % softs->ipf_state_size;
5341 if (softs->ipf_state_table[hv] != NULL)
5342 softs->ipf_state_table[hv]->is_phnext = &is->is_hnext;
5344 softs->ipf_state_stats.iss_inuse++;
5345 is->is_phnext = softs->ipf_state_table + hv;
5346 is->is_hnext = softs->ipf_state_table[hv];
5347 softs->ipf_state_table[hv] = is;
5348 softs->ipf_state_stats.iss_bucketlen[hv]++;
5350 RWLOCK_EXIT(&softc->ipf_state);
5356 /* ------------------------------------------------------------------------ */
5357 /* Function: ipf_state_add_tq */
5358 /* Returns: ipftq_t * - NULL = failure, else pointer to new timeout */
5360 /* Parameters: softc(I) - pointer to main soft context */
5361 /* ttl(I) - pointer to the ttl for the new queue */
5363 /* Request a pointer to a timeout queue that has a ttl as given by the */
5364 /* value being passed in. The timeout queue is added tot the list of those */
5365 /* used internally for stateful filtering. */
5366 /* ------------------------------------------------------------------------ */
5368 ipf_state_add_tq(softc, ttl)
5369 ipf_main_softc_t *softc;
5372 ipf_state_softc_t *softs = softc->ipf_state_soft;
5374 return ipf_addtimeoutqueue(softc, &softs->ipf_state_usertq, ttl);
5380 * Display the built up state table rules and mapping entries.
5383 ipf_state_dump(softc, arg)
5384 ipf_main_softc_t *softc;
5387 ipf_state_softc_t *softs = arg;
5390 printf("List of active state sessions:\n");
5391 for (ips = softs->ipf_state_list; ips != NULL; )
5392 ips = printstate(ips, opts & (OPT_DEBUG|OPT_VERBOSE),