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;
1626 DT1(ipf_fi_bad_tcpoptions_th_fin_ack_ecnall, fr_info_t *, fin);
1630 if ((fin->fin_out != 0) && (pass & FR_NEWISN) != 0) {
1631 ipf_checknewisn(fin, is);
1632 ipf_fixoutisn(fin, is);
1635 if ((tcp->th_flags & TH_OPENING) == TH_SYN)
1638 is->is_maxdwin = is->is_maxswin * 2;
1639 is->is_dend = ntohl(tcp->th_ack);
1640 is->is_maxdend = ntohl(tcp->th_ack);
1641 is->is_maxdwin *= 2;
1646 * If we're creating state for a starting connection, start
1647 * the timer on it as we'll never see an error if it fails
1655 is->is_sport = htons(fin->fin_data[0]);
1656 is->is_dport = htons(fin->fin_data[1]);
1657 if ((flags & (SI_W_DPORT|SI_W_SPORT)) == 0) {
1658 hv += tcp->th_dport;
1659 hv += tcp->th_sport;
1666 hv = DOUBLE_HASH(hv);
1670 * Look for identical state.
1672 for (is = softs->ipf_state_table[hv % softs->ipf_state_size];
1673 is != NULL; is = is->is_hnext) {
1674 if (ipf_state_match(&ips, is) == 1)
1678 SBUMPD(ipf_state_stats, iss_add_dup);
1682 if (softs->ipf_state_stats.iss_bucketlen[hv] >=
1683 softs->ipf_state_maxbucket) {
1684 SBUMPD(ipf_state_stats, iss_bucket_full);
1689 * No existing state; create new
1691 KMALLOC(is, ipstate_t *);
1693 SBUMPD(ipf_state_stats, iss_nomem);
1696 bcopy((char *)&ips, (char *)is, sizeof(*is));
1697 is->is_flags = flags & IS_INHERITED;
1698 is->is_rulen = fin->fin_rule;
1702 * Do not do the modulus here, it is done in ipf_state_insert().
1707 (void) strncpy(is->is_group, FR_NAME(fr, fr_group),
1709 if (fr->fr_age[0] != 0) {
1710 tq = ipf_addtimeoutqueue(softc,
1711 &softs->ipf_state_usertq,
1713 is->is_tqehead[0] = tq;
1714 is->is_sti.tqe_flags |= TQE_RULEBASED;
1716 if (fr->fr_age[1] != 0) {
1717 tq = ipf_addtimeoutqueue(softc,
1718 &softs->ipf_state_usertq,
1720 is->is_tqehead[1] = tq;
1721 is->is_sti.tqe_flags |= TQE_RULEBASED;
1724 is->is_tag = fr->fr_logtag;
1728 * It may seem strange to set is_ref to 2, but if stsave is not NULL
1729 * then a copy of the pointer is being stored somewhere else and in
1730 * the end, it will expect to be able to do something with it.
1733 if (stsave != NULL) {
1739 is->is_pkts[0] = 0, is->is_bytes[0] = 0;
1740 is->is_pkts[1] = 0, is->is_bytes[1] = 0;
1741 is->is_pkts[2] = 0, is->is_bytes[2] = 0;
1742 is->is_pkts[3] = 0, is->is_bytes[3] = 0;
1743 if ((fin->fin_flx & FI_IGNORE) == 0) {
1744 is->is_pkts[out] = 1;
1745 fin->fin_pktnum = 1;
1746 is->is_bytes[out] = fin->fin_plen;
1747 is->is_flx[out][0] = fin->fin_flx & FI_CMP;
1748 is->is_flx[out][0] &= ~FI_OOW;
1751 if (pass & FR_STLOOSE)
1752 is->is_flags |= IS_LOOSE;
1754 if (pass & FR_STSTRICT)
1755 is->is_flags |= IS_STRICT;
1757 if (pass & FR_STATESYNC)
1758 is->is_flags |= IS_STATESYNC;
1760 if (pass & FR_LOGFIRST)
1761 is->is_pass &= ~(FR_LOGFIRST|FR_LOG);
1763 READ_ENTER(&softc->ipf_state);
1765 if (ipf_state_insert(softc, is, fin->fin_rev) == -1) {
1766 RWLOCK_EXIT(&softc->ipf_state);
1768 * This is a bit more manual than it should be but
1769 * ipf_state_del cannot be called.
1771 MUTEX_EXIT(&is->is_lock);
1772 MUTEX_DESTROY(&is->is_lock);
1773 if (is->is_tqehead[0] != NULL) {
1774 if (ipf_deletetimeoutqueue(is->is_tqehead[0]) == 0)
1775 ipf_freetimeoutqueue(softc, is->is_tqehead[0]);
1776 is->is_tqehead[0] = NULL;
1778 if (is->is_tqehead[1] != NULL) {
1779 if (ipf_deletetimeoutqueue(is->is_tqehead[1]) == 0)
1780 ipf_freetimeoutqueue(softc, is->is_tqehead[1]);
1781 is->is_tqehead[1] = NULL;
1788 * Filling in the interface name is after the insert so that an
1789 * event (such as add/delete) of an interface that is referenced
1790 * by this rule will see this state entry.
1794 * The name '-' is special for network interfaces and causes
1795 * a NULL name to be present, always, allowing packets to
1796 * match it, regardless of their interface.
1798 if ((fin->fin_ifp == NULL) ||
1799 (fr->fr_ifnames[out << 1] != -1 &&
1800 fr->fr_names[fr->fr_ifnames[out << 1] + 0] == '-' &&
1801 fr->fr_names[fr->fr_ifnames[out << 1] + 1] == '\0')) {
1802 is->is_ifp[out << 1] = fr->fr_ifas[0];
1803 strncpy(is->is_ifname[out << 1],
1804 fr->fr_names + fr->fr_ifnames[0],
1805 sizeof(fr->fr_ifnames[0]));
1807 is->is_ifp[out << 1] = fin->fin_ifp;
1808 COPYIFNAME(fin->fin_v, fin->fin_ifp,
1809 is->is_ifname[out << 1]);
1812 is->is_ifp[(out << 1) + 1] = fr->fr_ifas[1];
1813 if (fr->fr_ifnames[1] != -1) {
1814 strncpy(is->is_ifname[(out << 1) + 1],
1815 fr->fr_names + fr->fr_ifnames[1],
1816 sizeof(fr->fr_ifnames[1]));
1819 is->is_ifp[(1 - out) << 1] = fr->fr_ifas[2];
1820 if (fr->fr_ifnames[2] != -1) {
1821 strncpy(is->is_ifname[((1 - out) << 1)],
1822 fr->fr_names + fr->fr_ifnames[2],
1823 sizeof(fr->fr_ifnames[2]));
1826 is->is_ifp[((1 - out) << 1) + 1] = fr->fr_ifas[3];
1827 if (fr->fr_ifnames[3] != -1) {
1828 strncpy(is->is_ifname[((1 - out) << 1) + 1],
1829 fr->fr_names + fr->fr_ifnames[3],
1830 sizeof(fr->fr_ifnames[3]));
1833 if (fin->fin_ifp != NULL) {
1834 is->is_ifp[out << 1] = fin->fin_ifp;
1835 COPYIFNAME(fin->fin_v, fin->fin_ifp,
1836 is->is_ifname[out << 1]);
1840 if (fin->fin_p == IPPROTO_TCP) {
1842 * If we're creating state for a starting connection, start the
1843 * timer on it as we'll never see an error if it fails to
1846 (void) ipf_tcp_age(&is->is_sti, fin, softs->ipf_state_tcptq,
1849 MUTEX_EXIT(&is->is_lock);
1850 if ((is->is_flags & IS_STATESYNC) && ((is->is_flags & SI_CLONE) == 0))
1851 is->is_sync = ipf_sync_new(softc, SMC_STATE, fin, is);
1852 if (softs->ipf_state_logging)
1853 ipf_state_log(softc, is, ISL_NEW);
1855 RWLOCK_EXIT(&softc->ipf_state);
1857 fin->fin_flx |= FI_STATE;
1858 if (fin->fin_flx & FI_FRAG)
1859 (void) ipf_frag_new(softc, fin, pass);
1861 fdp = &fr->fr_tifs[0];
1862 if (fdp->fd_type == FRD_DSTLIST) {
1863 ipf_dstlist_select_node(fin, fdp->fd_ptr, NULL,
1866 bcopy(fdp, &is->is_tifs[0], sizeof(*fdp));
1869 fdp = &fr->fr_tifs[1];
1870 if (fdp->fd_type == FRD_DSTLIST) {
1871 ipf_dstlist_select_node(fin, fdp->fd_ptr, NULL,
1874 bcopy(fdp, &is->is_tifs[1], sizeof(*fdp));
1876 fin->fin_tif = &is->is_tifs[fin->fin_rev];
1879 if (fdp->fd_type == FRD_DSTLIST) {
1880 ipf_dstlist_select_node(fin, fdp->fd_ptr, NULL,
1883 bcopy(fdp, &is->is_dif, sizeof(*fdp));
1885 fin->fin_dif = &is->is_dif;
1891 /* ------------------------------------------------------------------------ */
1892 /* Function: ipf_tcpoptions */
1893 /* Returns: int - 1 == packet matches state entry, 0 == it does not, */
1894 /* -1 == packet has bad TCP options data */
1895 /* Parameters: softs(I) - pointer to state context structure */
1896 /* fin(I) - pointer to packet information */
1897 /* tcp(I) - pointer to TCP packet header */
1898 /* td(I) - pointer to TCP data held as part of the state */
1900 /* Look after the TCP header for any options and deal with those that are */
1901 /* present. Record details about those that we recogise. */
1902 /* ------------------------------------------------------------------------ */
1904 ipf_tcpoptions(softs, fin, tcp, td)
1905 ipf_state_softc_t *softs;
1910 int off, mlen, ol, i, len, retval;
1911 char buf[64], *s, opt;
1914 len = (TCP_OFF(tcp) << 2);
1915 if (fin->fin_dlen < len) {
1916 SBUMPD(ipf_state_stats, iss_tcp_toosmall);
1919 len -= sizeof(*tcp);
1921 off = fin->fin_plen - fin->fin_dlen + sizeof(*tcp) + fin->fin_ipoff;
1924 mlen = MSGDSIZE(m) - off;
1932 COPYDATA(m, off, len, buf);
1934 for (s = buf; len > 0; ) {
1936 if (opt == TCPOPT_EOL)
1938 else if (opt == TCPOPT_NOP)
1944 if (ol < 2 || ol > len)
1948 * Extract the TCP options we are interested in out of
1949 * the header and store them in the the tcpdata struct.
1953 case TCPOPT_WINDOW :
1954 if (ol == TCPOLEN_WINDOW) {
1956 if (i > TCP_WSCALE_MAX)
1960 td->td_winscale = i;
1961 td->td_winflags |= TCP_WSCALE_SEEN|
1966 case TCPOPT_MAXSEG :
1968 * So, if we wanted to set the TCP MAXSEG,
1969 * it should be done here...
1971 if (ol == TCPOLEN_MAXSEG) {
1979 case TCPOPT_SACK_PERMITTED :
1980 if (ol == TCPOLEN_SACK_PERMITTED)
1981 td->td_winflags |= TCP_SACK_PERMIT;
1991 SBUMPD(ipf_state_stats, iss_tcp_badopt);
1997 /* ------------------------------------------------------------------------ */
1998 /* Function: ipf_state_tcp */
1999 /* Returns: int - 1 == packet matches state entry, 0 == it does not */
2000 /* Parameters: softc(I) - pointer to soft context main structure */
2001 /* softs(I) - pointer to state context structure */
2002 /* fin(I) - pointer to packet information */
2003 /* tcp(I) - pointer to TCP packet header */
2004 /* is(I) - pointer to master state structure */
2006 /* Check to see if a packet with TCP headers fits within the TCP window. */
2007 /* Change timeout depending on whether new packet is a SYN-ACK returning */
2008 /* for a SYN or a RST or FIN which indicate time to close up shop. */
2009 /* ------------------------------------------------------------------------ */
2011 ipf_state_tcp(softc, softs, fin, tcp, is)
2012 ipf_main_softc_t *softc;
2013 ipf_state_softc_t *softs;
2018 tcpdata_t *fdata, *tdata;
2019 int source, ret, flags;
2021 source = !fin->fin_rev;
2022 if (((is->is_flags & IS_TCPFSM) != 0) && (source == 1) &&
2023 (ntohs(is->is_sport) != fin->fin_data[0]))
2025 fdata = &is->is_tcp.ts_data[!source];
2026 tdata = &is->is_tcp.ts_data[source];
2028 MUTEX_ENTER(&is->is_lock);
2031 * If a SYN packet is received for a connection that is on the way out
2032 * but hasn't yet departed then advance this session along the way.
2034 if ((tcp->th_flags & TH_OPENING) == TH_SYN) {
2035 if ((is->is_state[0] > IPF_TCPS_ESTABLISHED) &&
2036 (is->is_state[1] > IPF_TCPS_ESTABLISHED)) {
2037 is->is_state[!source] = IPF_TCPS_CLOSED;
2038 ipf_movequeue(softc->ipf_ticks, &is->is_sti,
2040 &softs->ipf_state_deletetq);
2041 MUTEX_EXIT(&is->is_lock);
2042 DT1(iss_tcp_closing, ipstate_t *, is);
2043 SBUMP(ipf_state_stats.iss_tcp_closing);
2048 if (is->is_flags & IS_LOOSE)
2051 ret = ipf_state_tcpinwindow(fin, fdata, tdata, tcp,
2055 * Nearing end of connection, start timeout.
2057 ret = ipf_tcp_age(&is->is_sti, fin, softs->ipf_state_tcptq,
2060 MUTEX_EXIT(&is->is_lock);
2061 DT2(iss_tcp_fsm, fr_info_t *, fin, ipstate_t *, is);
2062 SBUMP(ipf_state_stats.iss_tcp_fsm);
2066 if (softs->ipf_state_logging > 4)
2067 ipf_state_log(softc, is, ISL_STATECHANGE);
2070 * set s0's as appropriate. Use syn-ack packet as it
2071 * contains both pieces of required information.
2074 * Window scale option is only present in SYN/SYN-ACK packet.
2075 * Compare with ~TH_FIN to mask out T/TCP setups.
2077 flags = tcp->th_flags & ~(TH_FIN|TH_ECNALL);
2078 if (flags == (TH_SYN|TH_ACK)) {
2079 is->is_s0[source] = ntohl(tcp->th_ack);
2080 is->is_s0[!source] = ntohl(tcp->th_seq) + 1;
2081 if ((TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) {
2082 if (ipf_tcpoptions(softs, fin, tcp,
2084 fin->fin_flx |= FI_BAD;
2085 DT1(ipf_fi_bad_winscale_syn_ack, fr_info_t *, fin);
2088 if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN))
2089 ipf_checknewisn(fin, is);
2090 } else if (flags == TH_SYN) {
2091 is->is_s0[source] = ntohl(tcp->th_seq) + 1;
2092 if ((TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) {
2093 if (ipf_tcpoptions(softs, fin, tcp,
2095 fin->fin_flx |= FI_BAD;
2096 DT1(ipf_fi_bad_winscale_syn, fr_info_t *, fin);
2100 if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN))
2101 ipf_checknewisn(fin, is);
2106 DT2(iss_tcp_oow, fr_info_t *, fin, ipstate_t *, is);
2107 SBUMP(ipf_state_stats.iss_tcp_oow);
2110 MUTEX_EXIT(&is->is_lock);
2115 /* ------------------------------------------------------------------------ */
2116 /* Function: ipf_checknewisn */
2118 /* Parameters: fin(I) - pointer to packet information */
2119 /* is(I) - pointer to master state structure */
2121 /* Check to see if this TCP connection is expecting and needs a new */
2122 /* sequence number for a particular direction of the connection. */
2124 /* NOTE: This does not actually change the sequence numbers, only gets new */
2126 /* ------------------------------------------------------------------------ */
2128 ipf_checknewisn(fin, is)
2132 u_32_t sumd, old, new;
2139 if (((i == 0) && !(is->is_flags & IS_ISNSYN)) ||
2140 ((i == 1) && !(is->is_flags & IS_ISNACK))) {
2141 old = ntohl(tcp->th_seq);
2142 new = ipf_newisn(fin);
2143 is->is_isninc[i] = new - old;
2144 CALC_SUMD(old, new, sumd);
2145 is->is_sumd[i] = (sumd & 0xffff) + (sumd >> 16);
2147 is->is_flags |= ((i == 0) ? IS_ISNSYN : IS_ISNACK);
2152 /* ------------------------------------------------------------------------ */
2153 /* Function: ipf_state_tcpinwindow */
2154 /* Returns: int - 1 == packet inside TCP "window", 0 == not inside. */
2155 /* Parameters: fin(I) - pointer to packet information */
2156 /* fdata(I) - pointer to tcp state informatio (forward) */
2157 /* tdata(I) - pointer to tcp state informatio (reverse) */
2158 /* tcp(I) - pointer to TCP packet header */
2160 /* Given a packet has matched addresses and ports, check to see if it is */
2161 /* within the TCP data window. In a show of generosity, allow packets that */
2162 /* are within the window space behind the current sequence # as well. */
2163 /* ------------------------------------------------------------------------ */
2165 ipf_state_tcpinwindow(fin, fdata, tdata, tcp, flags)
2167 tcpdata_t *fdata, *tdata;
2171 ipf_main_softc_t *softc = fin->fin_main_soft;
2172 ipf_state_softc_t *softs = softc->ipf_state_soft;
2173 tcp_seq seq, ack, end;
2174 int ackskew, tcpflags;
2179 * Find difference between last checked packet and this packet.
2181 tcpflags = tcp->th_flags;
2182 seq = ntohl(tcp->th_seq);
2183 ack = ntohl(tcp->th_ack);
2184 if (tcpflags & TH_SYN)
2185 win = ntohs(tcp->th_win);
2187 win = ntohs(tcp->th_win) << fdata->td_winscale;
2190 * A window of 0 produces undesirable behaviour from this function.
2195 dsize = fin->fin_dlen - (TCP_OFF(tcp) << 2) +
2196 ((tcpflags & TH_SYN) ? 1 : 0) + ((tcpflags & TH_FIN) ? 1 : 0);
2199 * if window scaling is present, the scaling is only allowed
2200 * for windows not in the first SYN packet. In that packet the
2201 * window is 65535 to specify the largest window possible
2202 * for receivers not implementing the window scale option.
2203 * Currently, we do not assume TTCP here. That means that
2204 * if we see a second packet from a host (after the initial
2205 * SYN), we can assume that the receiver of the SYN did
2206 * already send back the SYN/ACK (and thus that we know if
2207 * the receiver also does window scaling)
2209 if (!(tcpflags & TH_SYN) && (fdata->td_winflags & TCP_WSCALE_FIRST)) {
2210 fdata->td_winflags &= ~TCP_WSCALE_FIRST;
2211 fdata->td_maxwin = win;
2216 if ((fdata->td_end == 0) &&
2217 (!(flags & IS_TCPFSM) ||
2218 ((tcpflags & TH_OPENING) == TH_OPENING))) {
2220 * Must be a (outgoing) SYN-ACK in reply to a SYN.
2222 fdata->td_end = end - 1;
2223 fdata->td_maxwin = 1;
2224 fdata->td_maxend = end + win;
2227 if (!(tcpflags & TH_ACK)) { /* Pretend an ack was sent */
2228 ack = tdata->td_end;
2229 } else if (((tcpflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) &&
2231 /* gross hack to get around certain broken tcp stacks */
2232 ack = tdata->td_end;
2235 maxwin = tdata->td_maxwin;
2236 ackskew = tdata->td_end - ack;
2239 * Strict sequencing only allows in-order delivery.
2241 if ((flags & IS_STRICT) != 0) {
2242 if (seq != fdata->td_end) {
2243 DT2(iss_tcp_struct, tcpdata_t *, fdata, int, seq);
2244 SBUMP(ipf_state_stats.iss_tcp_strict);
2245 fin->fin_flx |= FI_OOW;
2250 #define SEQ_GE(a,b) ((int)((a) - (b)) >= 0)
2251 #define SEQ_GT(a,b) ((int)((a) - (b)) > 0)
2253 if ((SEQ_GE(fdata->td_maxend, end)) &&
2254 (SEQ_GE(seq, fdata->td_end - maxwin)) &&
2255 /* XXX what about big packets */
2256 #define MAXACKWINDOW 66000
2257 (-ackskew <= (MAXACKWINDOW)) &&
2258 ( ackskew <= (MAXACKWINDOW << fdata->td_winscale))) {
2261 * Microsoft Windows will send the next packet to the right of the
2262 * window if SACK is in use.
2264 } else if ((seq == fdata->td_maxend) && (ackskew == 0) &&
2265 (fdata->td_winflags & TCP_SACK_PERMIT) &&
2266 (tdata->td_winflags & TCP_SACK_PERMIT)) {
2267 DT2(iss_sinsack, tcpdata_t *, fdata, int, seq);
2268 SBUMP(ipf_state_stats.iss_winsack);
2271 * Sometimes a TCP RST will be generated with only the ACK field
2274 } else if ((seq == 0) && (tcpflags == (TH_RST|TH_ACK)) &&
2275 (ackskew >= -1) && (ackskew <= 1)) {
2277 } else if (!(flags & IS_TCPFSM)) {
2280 i = (fin->fin_rev << 1) + fin->fin_out;
2283 if (is_pkts[i]0 == 0) {
2285 * Picking up a connection in the middle, the "next"
2286 * packet seen from a direction that is new should be
2287 * accepted, even if it appears out of sequence.
2292 if (!(fdata->td_winflags &
2293 (TCP_WSCALE_SEEN|TCP_WSCALE_FIRST))) {
2295 * No TCPFSM and no window scaling, so make some
2298 if ((seq == fdata->td_maxend) && (ackskew == 0))
2300 else if (SEQ_GE(seq + maxwin, fdata->td_end - maxwin))
2305 /* TRACE(inseq, fdata, tdata, seq, end, ack, ackskew, win, maxwin) */
2308 /* if ackskew < 0 then this should be due to fragmented
2309 * packets. There is no way to know the length of the
2310 * total packet in advance.
2311 * We do know the total length from the fragment cache though.
2312 * Note however that there might be more sessions with
2313 * exactly the same source and destination parameters in the
2314 * state cache (and source and destination is the only stuff
2315 * that is saved in the fragment cache). Note further that
2316 * some TCP connections in the state cache are hashed with
2317 * sport and dport as well which makes it not worthwhile to
2319 * Thus, when ackskew is negative but still seems to belong
2320 * to this session, we bump up the destinations end value.
2323 tdata->td_end = ack;
2325 /* update max window seen */
2326 if (fdata->td_maxwin < win)
2327 fdata->td_maxwin = win;
2328 if (SEQ_GT(end, fdata->td_end))
2329 fdata->td_end = end;
2330 if (SEQ_GE(ack + win, tdata->td_maxend))
2331 tdata->td_maxend = ack + win;
2334 SBUMP(ipf_state_stats.iss_oow);
2335 fin->fin_flx |= FI_OOW;
2340 /* ------------------------------------------------------------------------ */
2341 /* Function: ipf_state_clone */
2342 /* Returns: ipstate_t* - NULL == cloning failed, */
2343 /* else pointer to new state structure */
2344 /* Parameters: fin(I) - pointer to packet information */
2345 /* tcp(I) - pointer to TCP/UDP header */
2346 /* is(I) - pointer to master state structure */
2348 /* Create a "duplcate" state table entry from the master. */
2349 /* ------------------------------------------------------------------------ */
2351 ipf_state_clone(fin, tcp, is)
2356 ipf_main_softc_t *softc = fin->fin_main_soft;
2357 ipf_state_softc_t *softs = softc->ipf_state_soft;
2361 if (softs->ipf_state_stats.iss_active == softs->ipf_state_max) {
2362 SBUMPD(ipf_state_stats, iss_max);
2363 softs->ipf_state_doflush = 1;
2366 KMALLOC(clone, ipstate_t *);
2367 if (clone == NULL) {
2368 SBUMPD(ipf_state_stats, iss_clone_nomem);
2371 bcopy((char *)is, (char *)clone, sizeof(*clone));
2373 MUTEX_NUKE(&clone->is_lock);
2375 * It has not yet been placed on any timeout queue, so make sure
2376 * all of that data is zero'd out.
2378 clone->is_sti.tqe_pnext = NULL;
2379 clone->is_sti.tqe_next = NULL;
2380 clone->is_sti.tqe_ifq = NULL;
2381 clone->is_sti.tqe_parent = clone;
2383 clone->is_die = ONE_DAY + softc->ipf_ticks;
2384 clone->is_state[0] = 0;
2385 clone->is_state[1] = 0;
2386 send = ntohl(tcp->th_seq) + fin->fin_dlen - (TCP_OFF(tcp) << 2) +
2387 ((tcp->th_flags & TH_SYN) ? 1 : 0) +
2388 ((tcp->th_flags & TH_FIN) ? 1 : 0);
2390 if (fin->fin_rev == 1) {
2391 clone->is_dend = send;
2392 clone->is_maxdend = send;
2394 clone->is_maxswin = 1;
2395 clone->is_maxdwin = ntohs(tcp->th_win);
2396 if (clone->is_maxdwin == 0)
2397 clone->is_maxdwin = 1;
2399 clone->is_send = send;
2400 clone->is_maxsend = send;
2402 clone->is_maxdwin = 1;
2403 clone->is_maxswin = ntohs(tcp->th_win);
2404 if (clone->is_maxswin == 0)
2405 clone->is_maxswin = 1;
2408 clone->is_flags &= ~SI_CLONE;
2409 clone->is_flags |= SI_CLONED;
2410 if (ipf_state_insert(softc, clone, fin->fin_rev) == -1) {
2416 if (clone->is_p == IPPROTO_TCP) {
2417 (void) ipf_tcp_age(&clone->is_sti, fin, softs->ipf_state_tcptq,
2418 clone->is_flags, 2);
2420 MUTEX_EXIT(&clone->is_lock);
2421 if (is->is_flags & IS_STATESYNC)
2422 clone->is_sync = ipf_sync_new(softc, SMC_STATE, fin, clone);
2423 DT2(iss_clone, ipstate_t *, is, ipstate_t *, clone);
2424 SBUMP(ipf_state_stats.iss_cloned);
2429 /* ------------------------------------------------------------------------ */
2430 /* Function: ipf_matchsrcdst */
2432 /* Parameters: fin(I) - pointer to packet information */
2433 /* is(I) - pointer to state structure */
2434 /* src(I) - pointer to source address */
2435 /* dst(I) - pointer to destination address */
2436 /* tcp(I) - pointer to TCP/UDP header */
2437 /* cmask(I) - mask of FI_* bits to check */
2439 /* Match a state table entry against an IP packet. The logic below is that */
2440 /* ret gets set to one if the match succeeds, else remains 0. If it is */
2441 /* still 0 after the test. no match. */
2442 /* ------------------------------------------------------------------------ */
2444 ipf_matchsrcdst(fin, is, src, dst, tcp, cmask)
2447 i6addr_t *src, *dst;
2451 ipf_main_softc_t *softc = fin->fin_main_soft;
2452 ipf_state_softc_t *softs = softc->ipf_state_soft;
2453 int ret = 0, rev, out, flags, flx = 0, idx;
2459 * If a connection is about to be deleted, no packets
2460 * are allowed to match it.
2462 if (is->is_sti.tqe_ifq == &softs->ipf_state_deletetq)
2465 rev = IP6_NEQ(&is->is_dst, dst);
2468 flags = is->is_flags;
2473 sp = htons(fin->fin_sport);
2474 dp = ntohs(fin->fin_dport);
2478 if (!(flags & SI_W_SPORT) && (sp != is->is_sport))
2480 else if (!(flags & SI_W_DPORT) && (dp != is->is_dport))
2485 idx = (out << 1) + rev;
2488 * If the interface for this 'direction' is set, make sure it matches.
2489 * An interface name that is not set matches any, as does a name of *.
2491 if ((is->is_ifp[idx] == ifp) || (is->is_ifp[idx] == NULL &&
2492 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '-' ||
2493 *is->is_ifname[idx] == '*')))
2497 DT2(iss_lookup_badifp, fr_info_t *, fin, ipstate_t *, is);
2498 SBUMP(ipf_state_stats.iss_lookup_badifp);
2499 /* TRACE is, out, rev, idx */
2505 * Match addresses and ports.
2508 if ((IP6_EQ(&is->is_dst, dst) || (flags & SI_W_DADDR)) &&
2509 (IP6_EQ(&is->is_src, src) || (flags & SI_W_SADDR))) {
2511 if ((sp == is->is_sport || flags & SI_W_SPORT)
2513 (dp == is->is_dport || flags & SI_W_DPORT))
2520 if ((IP6_EQ(&is->is_dst, src) || (flags & SI_W_DADDR)) &&
2521 (IP6_EQ(&is->is_src, dst) || (flags & SI_W_SADDR))) {
2523 if ((dp == is->is_sport || flags & SI_W_SPORT)
2525 (sp == is->is_dport || flags & SI_W_DPORT))
2534 SBUMP(ipf_state_stats.iss_lookup_badport);
2535 DT2(iss_lookup_badport, fr_info_t *, fin, ipstate_t *, is);
2536 /* TRACE rev, is, sp, dp, src, dst */
2541 * Whether or not this should be here, is questionable, but the aim
2542 * is to get this out of the main line.
2545 flags = is->is_flags & ~(SI_WILDP|SI_NEWFR|SI_CLONE|SI_CLONED);
2548 * Only one of the source or destination address can be flaged as a
2549 * wildcard. Fill in the missing address, if set.
2550 * For IPv6, if the address being copied in is multicast, then
2551 * don't reset the wild flag - multicast causes it to be set in the
2554 if ((flags & (SI_W_SADDR|SI_W_DADDR))) {
2555 fr_ip_t *fi = &fin->fin_fi;
2557 if ((flags & SI_W_SADDR) != 0) {
2559 is->is_src = fi->fi_src;
2560 is->is_flags &= ~SI_W_SADDR;
2562 if (!(fin->fin_flx & (FI_MULTICAST|FI_MBCAST))){
2563 is->is_src = fi->fi_dst;
2564 is->is_flags &= ~SI_W_SADDR;
2567 } else if ((flags & SI_W_DADDR) != 0) {
2569 if (!(fin->fin_flx & (FI_MULTICAST|FI_MBCAST))){
2570 is->is_dst = fi->fi_dst;
2571 is->is_flags &= ~SI_W_DADDR;
2574 is->is_dst = fi->fi_src;
2575 is->is_flags &= ~SI_W_DADDR;
2578 if ((is->is_flags & (SI_WILDA|SI_WILDP)) == 0) {
2579 ATOMIC_DECL(softs->ipf_state_stats.iss_wild);
2583 flx = fin->fin_flx & cmask;
2584 cflx = is->is_flx[out][rev];
2587 * Match up any flags set from IP options.
2589 if ((cflx && (flx != (cflx & cmask))) ||
2590 ((fin->fin_optmsk & is->is_optmsk[rev]) != is->is_opt[rev]) ||
2591 ((fin->fin_secmsk & is->is_secmsk) != is->is_sec) ||
2592 ((fin->fin_auth & is->is_authmsk) != is->is_auth)) {
2593 SBUMPD(ipf_state_stats, iss_miss_mask);
2597 if ((fin->fin_flx & FI_IGNORE) != 0) {
2603 * Only one of the source or destination port can be flagged as a
2604 * wildcard. When filling it in, fill in a copy of the matched entry
2605 * if it has the cloning flag set.
2607 if ((flags & (SI_W_SPORT|SI_W_DPORT))) {
2608 if ((flags & SI_CLONE) != 0) {
2611 clone = ipf_state_clone(fin, tcp, is);
2616 ATOMIC_DECL(softs->ipf_state_stats.iss_wild);
2619 if ((flags & SI_W_SPORT) != 0) {
2622 is->is_send = ntohl(tcp->th_seq);
2625 is->is_send = ntohl(tcp->th_ack);
2627 is->is_maxsend = is->is_send + 1;
2628 } else if ((flags & SI_W_DPORT) != 0) {
2631 is->is_dend = ntohl(tcp->th_ack);
2634 is->is_dend = ntohl(tcp->th_seq);
2636 is->is_maxdend = is->is_dend + 1;
2638 is->is_flags &= ~(SI_W_SPORT|SI_W_DPORT);
2639 if ((flags & SI_CLONED) && softs->ipf_state_logging)
2640 ipf_state_log(softc, is, ISL_CLONE);
2645 if (is->is_flx[out][rev] == 0) {
2646 is->is_flx[out][rev] = flx;
2647 if (rev == 1 && is->is_optmsk[1] == 0) {
2648 is->is_opt[1] = fin->fin_optmsk;
2649 is->is_optmsk[1] = 0xffffffff;
2650 if (is->is_v == 6) {
2651 is->is_opt[1] &= ~0x8;
2652 is->is_optmsk[1] &= ~0x8;
2658 * Check if the interface name for this "direction" is set and if not,
2661 if (is->is_ifp[idx] == NULL &&
2662 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) {
2663 is->is_ifp[idx] = ifp;
2664 COPYIFNAME(fin->fin_v, ifp, is->is_ifname[idx]);
2671 /* ------------------------------------------------------------------------ */
2672 /* Function: ipf_checkicmpmatchingstate */
2674 /* Parameters: fin(I) - pointer to packet information */
2676 /* If we've got an ICMP error message, using the information stored in the */
2677 /* ICMP packet, look for a matching state table entry. */
2679 /* If we return NULL then no lock on ipf_state is held. */
2680 /* If we return non-null then a read-lock on ipf_state is held. */
2681 /* ------------------------------------------------------------------------ */
2683 ipf_checkicmpmatchingstate(fin)
2686 ipf_main_softc_t *softc = fin->fin_main_soft;
2687 ipf_state_softc_t *softs = softc->ipf_state_soft;
2688 ipstate_t *is, **isp;
2701 * Does it at least have the return (basic) IP header ?
2702 * Is it an actual recognised ICMP error type?
2703 * Only a basic IP header (no options) should be with
2704 * an ICMP error header.
2706 if ((fin->fin_v != 4) || (fin->fin_hlen != sizeof(ip_t)) ||
2707 (fin->fin_plen < ICMPERR_MINPKTLEN) ||
2708 !(fin->fin_flx & FI_ICMPERR)) {
2709 SBUMPD(ipf_state_stats, iss_icmp_bad);
2713 type = ic->icmp_type;
2715 oip = (ip_t *)((char *)ic + ICMPERR_ICMPHLEN);
2717 * Check if the at least the old IP header (with options) and
2718 * 8 bytes of payload is present.
2720 if (fin->fin_plen < ICMPERR_MAXPKTLEN + ((IP_HL(oip) - 5) << 2)) {
2721 SBUMPDX(ipf_state_stats, iss_icmp_short, iss_icmp_short_1);
2728 len = fin->fin_dlen - ICMPERR_ICMPHLEN;
2729 if ((len <= 0) || ((IP_HL(oip) << 2) > len)) {
2730 DT2(iss_icmp_len, fr_info_t *, fin, struct ip*, oip);
2731 SBUMPDX(ipf_state_stats, iss_icmp_short, iss_icmp_short_1);
2736 * Is the buffer big enough for all of it ? It's the size of the IP
2737 * header claimed in the encapsulated part which is of concern. It
2738 * may be too big to be in this buffer but not so big that it's
2739 * outside the ICMP packet, leading to TCP deref's causing problems.
2740 * This is possible because we don't know how big oip_hl is when we
2741 * do the pullup early in ipf_check() and thus can't guarantee it is
2749 # if defined(MENTAT)
2750 if ((char *)oip + len > (char *)m->b_wptr) {
2751 SBUMPDX(ipf_state_stats, iss_icmp_short, iss_icmp_short_2);
2755 if ((char *)oip + len > (char *)fin->fin_ip + m->m_len) {
2756 SBUMPDX(ipf_state_stats, iss_icmp_short, iss_icmp_short_3);
2763 bcopy((char *)fin, (char *)&ofin, sizeof(*fin));
2766 * in the IPv4 case we must zero the i6addr union otherwise
2767 * the IP6_EQ and IP6_NEQ macros produce the wrong results because
2768 * of the 'junk' in the unused part of the union
2770 bzero((char *)&src, sizeof(src));
2771 bzero((char *)&dst, sizeof(dst));
2774 * we make an fin entry to be able to feed it to
2775 * matchsrcdst note that not all fields are encessary
2776 * but this is the cleanest way. Note further we fill
2777 * in fin_mp such that if someone uses it we'll get
2778 * a kernel panic. ipf_matchsrcdst does not use this.
2780 * watch out here, as ip is in host order and oip in network
2781 * order. Any change we make must be undone afterwards, like
2784 savelen = oip->ip_len;
2785 oip->ip_len = htons(len);
2787 ofin.fin_flx = FI_NOCKSUM;
2790 ofin.fin_m = NULL; /* if dereferenced, panic XXX */
2791 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */
2792 (void) ipf_makefrip(IP_HL(oip) << 2, oip, &ofin);
2793 ofin.fin_ifp = fin->fin_ifp;
2794 ofin.fin_out = !fin->fin_out;
2796 hv = (pr = oip->ip_p);
2797 src.in4 = oip->ip_src;
2798 hv += src.in4.s_addr;
2799 dst.in4 = oip->ip_dst;
2800 hv += dst.in4.s_addr;
2803 * Reset the short and bad flag here because in ipf_matchsrcdst()
2804 * the flags for the current packet (fin_flx) are compared against
2805 * those for the existing session.
2807 ofin.fin_flx &= ~(FI_BAD|FI_SHORT);
2810 * Put old values of ip_len back as we don't know
2811 * if we have to forward the packet or process it again.
2813 oip->ip_len = savelen;
2819 * an ICMP error can only be generated as a result of an
2820 * ICMP query, not as the response on an ICMP error
2822 * XXX theoretically ICMP_ECHOREP and the other reply's are
2823 * ICMP query's as well, but adding them here seems strange XXX
2825 if ((ofin.fin_flx & FI_ICMPERR) != 0) {
2826 DT1(iss_icmp_icmperr, fr_info_t *, &ofin);
2827 SBUMP(ipf_state_stats.iss_icmp_icmperr);
2832 * perform a lookup of the ICMP packet in the state table
2834 icmp = (icmphdr_t *)((char *)oip + (IP_HL(oip) << 2));
2835 hv += icmp->icmp_id;
2836 hv = DOUBLE_HASH(hv);
2838 READ_ENTER(&softc->ipf_state);
2839 for (isp = &softs->ipf_state_table[hv];
2840 ((is = *isp) != NULL); ) {
2841 isp = &is->is_hnext;
2842 if ((is->is_p != pr) || (is->is_v != 4))
2844 if (is->is_pass & FR_NOICMPERR)
2847 is = ipf_matchsrcdst(&ofin, is, &src, &dst,
2849 if ((is != NULL) && !ipf_allowstateicmp(fin, is, &src))
2852 RWLOCK_EXIT(&softc->ipf_state);
2853 SBUMPDX(ipf_state_stats, iss_icmp_miss, iss_icmp_miss_1);
2859 SBUMPDX(ipf_state_stats, iss_icmp_miss, iss_icmp_miss_2);
2863 tcp = (tcphdr_t *)((char *)oip + (IP_HL(oip) << 2));
2865 hv += tcp->th_dport;;
2866 hv += tcp->th_sport;;
2867 hv = DOUBLE_HASH(hv);
2869 READ_ENTER(&softc->ipf_state);
2870 for (isp = &softs->ipf_state_table[hv]; ((is = *isp) != NULL); ) {
2871 isp = &is->is_hnext;
2873 * Only allow this icmp though if the
2874 * encapsulated packet was allowed through the
2875 * other way around. Note that the minimal amount
2876 * of info present does not allow for checking against
2877 * tcp internals such as seq and ack numbers. Only the
2878 * ports are known to be present and can be even if the
2879 * short flag is set.
2881 if ((is->is_p == pr) && (is->is_v == 4) &&
2882 (is = ipf_matchsrcdst(&ofin, is, &src, &dst,
2883 tcp, FI_ICMPCMP))) {
2884 if (ipf_allowstateicmp(fin, is, &src) == 0)
2888 RWLOCK_EXIT(&softc->ipf_state);
2889 SBUMPDX(ipf_state_stats, iss_icmp_miss, iss_icmp_miss_3);
2894 /* ------------------------------------------------------------------------ */
2895 /* Function: ipf_allowstateicmp */
2896 /* Returns: int - 1 = packet denied, 0 = packet allowed */
2897 /* Parameters: fin(I) - pointer to packet information */
2898 /* is(I) - pointer to state table entry */
2899 /* src(I) - source address to check permission for */
2901 /* For an ICMP packet that has so far matched a state table entry, check if */
2902 /* there are any further refinements that might mean we want to block this */
2903 /* packet. This code isn't specific to either IPv4 or IPv6. */
2904 /* ------------------------------------------------------------------------ */
2906 ipf_allowstateicmp(fin, is, src)
2911 ipf_main_softc_t *softc = fin->fin_main_soft;
2912 ipf_state_softc_t *softs = softc->ipf_state_soft;
2921 if (fr != NULL && fr->fr_icmpgrp != NULL) {
2922 savefr = fin->fin_fr;
2923 fin->fin_fr = fr->fr_icmpgrp->fg_start;
2925 ipass = ipf_scanlist(fin, softc->ipf_pass);
2926 fin->fin_fr = savefr;
2927 if (FR_ISBLOCK(ipass)) {
2928 SBUMPD(ipf_state_stats, iss_icmp_headblock);
2934 * i : the index of this packet (the icmp unreachable)
2935 * oi : the index of the original packet found in the
2936 * icmp header (i.e. the packet causing this icmp)
2937 * backward : original packet was backward compared to
2940 backward = IP6_NEQ(&is->is_src, src);
2941 fin->fin_rev = !backward;
2942 i = (!backward << 1) + fin->fin_out;
2943 oi = (backward << 1) + !fin->fin_out;
2945 if (is->is_pass & FR_NOICMPERR) {
2946 SBUMPD(ipf_state_stats, iss_icmp_banned);
2949 if (is->is_icmppkts[i] > is->is_pkts[oi]) {
2950 SBUMPD(ipf_state_stats, iss_icmp_toomany);
2954 DT2(iss_icmp_hits, fr_info_t *, fin, ipstate_t *, is);
2955 SBUMP(ipf_state_stats.iss_icmp_hits);
2956 is->is_icmppkts[i]++;
2959 * we deliberately do not touch the timeouts
2960 * for the accompanying state table entry.
2961 * It remains to be seen if that is correct. XXX
2967 /* ------------------------------------------------------------------------ */
2968 /* Function: ipf_ipsmove */
2970 /* Parameters: is(I) - pointer to state table entry */
2971 /* hv(I) - new hash value for state table entry */
2972 /* Write Locks: ipf_state */
2974 /* Move a state entry from one position in the hash table to another. */
2975 /* ------------------------------------------------------------------------ */
2977 ipf_ipsmove(softs, is, hv)
2978 ipf_state_softc_t *softs;
2987 /* TRACE is, is_hv, hvm */
2990 * Remove the hash from the old location...
2992 isp = is->is_phnext;
2994 is->is_hnext->is_phnext = isp;
2995 *isp = is->is_hnext;
2996 if (softs->ipf_state_table[hvm] == NULL)
2997 softs->ipf_state_stats.iss_inuse--;
2998 softs->ipf_state_stats.iss_bucketlen[hvm]--;
3001 * ...and put the hash in the new one.
3003 hvm = DOUBLE_HASH(hv);
3006 /* TRACE is, hv, is_hv, hvm */
3008 isp = &softs->ipf_state_table[hvm];
3010 (*isp)->is_phnext = &is->is_hnext;
3012 softs->ipf_state_stats.iss_inuse++;
3013 softs->ipf_state_stats.iss_bucketlen[hvm]++;
3014 is->is_phnext = isp;
3015 is->is_hnext = *isp;
3020 /* ------------------------------------------------------------------------ */
3021 /* Function: ipf_state_lookup */
3022 /* Returns: ipstate_t* - NULL == no matching state found, */
3023 /* else pointer to state information is returned */
3024 /* Parameters: fin(I) - pointer to packet information */
3025 /* tcp(I) - pointer to TCP/UDP header. */
3026 /* ifqp(O) - pointer for storing tailq timeout */
3028 /* Search the state table for a matching entry to the packet described by */
3029 /* the contents of *fin. For certain protocols, when a match is found the */
3030 /* timeout queue is also selected and stored in ifpq if it is non-NULL. */
3032 /* If we return NULL then no lock on ipf_state is held. */
3033 /* If we return non-null then a read-lock on ipf_state is held. */
3034 /* ------------------------------------------------------------------------ */
3036 ipf_state_lookup(fin, tcp, ifqp)
3041 ipf_main_softc_t *softc = fin->fin_main_soft;
3042 ipf_state_softc_t *softs = softc->ipf_state_soft;
3043 u_int hv, hvm, pr, v, tryagain;
3044 ipstate_t *is, **isp;
3045 u_short dport, sport;
3054 ic = (struct icmp *)tcp;
3055 hv = (pr = fin->fin_fi.fi_p);
3056 src = fin->fin_fi.fi_src;
3057 dst = fin->fin_fi.fi_dst;
3058 hv += src.in4.s_addr;
3059 hv += dst.in4.s_addr;
3061 v = fin->fin_fi.fi_v;
3064 hv += fin->fin_fi.fi_src.i6[1];
3065 hv += fin->fin_fi.fi_src.i6[2];
3066 hv += fin->fin_fi.fi_src.i6[3];
3068 if ((fin->fin_p == IPPROTO_ICMPV6) &&
3069 IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_dst.in6)) {
3070 hv -= dst.in4.s_addr;
3072 hv += fin->fin_fi.fi_dst.i6[1];
3073 hv += fin->fin_fi.fi_dst.i6[2];
3074 hv += fin->fin_fi.fi_dst.i6[3];
3079 (fin->fin_flx & (FI_MULTICAST|FI_BROADCAST|FI_MBCAST))) {
3080 if (fin->fin_out == 0) {
3081 hv -= src.in4.s_addr;
3083 hv -= dst.in4.s_addr;
3087 /* TRACE fin_saddr, fin_daddr, hv */
3090 * Search the hash table for matching packet header info.
3095 case IPPROTO_ICMPV6 :
3098 if ((ic->icmp_type == ICMP6_ECHO_REQUEST) ||
3099 (ic->icmp_type == ICMP6_ECHO_REPLY)) {
3103 READ_ENTER(&softc->ipf_state);
3105 hvm = DOUBLE_HASH(hv);
3106 for (isp = &softs->ipf_state_table[hvm];
3107 ((is = *isp) != NULL); ) {
3108 isp = &is->is_hnext;
3109 if ((is->is_p != pr) || (is->is_v != v))
3111 is = ipf_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP);
3113 ipf_matchicmpqueryreply(v, &is->is_icmp,
3114 ic, fin->fin_rev)) {
3116 ifq = &softs->ipf_state_icmpacktq;
3118 ifq = &softs->ipf_state_icmptq;
3124 if ((tryagain != 0) && !(is->is_flags & SI_W_DADDR)) {
3125 hv += fin->fin_fi.fi_src.i6[0];
3126 hv += fin->fin_fi.fi_src.i6[1];
3127 hv += fin->fin_fi.fi_src.i6[2];
3128 hv += fin->fin_fi.fi_src.i6[3];
3129 ipf_ipsmove(softs, is, hv);
3130 MUTEX_DOWNGRADE(&softc->ipf_state);
3134 RWLOCK_EXIT(&softc->ipf_state);
3137 * No matching icmp state entry. Perhaps this is a
3138 * response to another state entry.
3140 * XXX With some ICMP6 packets, the "other" address is already
3141 * in the packet, after the ICMP6 header, and this could be
3142 * used in place of the multicast address. However, taking
3143 * advantage of this requires some significant code changes
3144 * to handle the specific types where that is the case.
3146 if ((softs->ipf_state_stats.iss_wild != 0) &&
3147 ((fin->fin_flx & FI_NOWILD) == 0) &&
3148 (v == 6) && (tryagain == 0)) {
3149 hv -= fin->fin_fi.fi_src.i6[0];
3150 hv -= fin->fin_fi.fi_src.i6[1];
3151 hv -= fin->fin_fi.fi_src.i6[2];
3152 hv -= fin->fin_fi.fi_src.i6[3];
3154 WRITE_ENTER(&softc->ipf_state);
3158 is = ipf_checkicmp6matchingstate(fin);
3168 hv = DOUBLE_HASH(hv);
3169 READ_ENTER(&softc->ipf_state);
3170 for (isp = &softs->ipf_state_table[hv];
3171 ((is = *isp) != NULL); ) {
3172 isp = &is->is_hnext;
3173 if ((is->is_p != pr) || (is->is_v != v))
3175 is = ipf_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP);
3177 (ic->icmp_id == is->is_icmp.ici_id) &&
3178 ipf_matchicmpqueryreply(v, &is->is_icmp,
3179 ic, fin->fin_rev)) {
3181 ifq = &softs->ipf_state_icmpacktq;
3183 ifq = &softs->ipf_state_icmptq;
3188 RWLOCK_EXIT(&softc->ipf_state);
3195 sport = htons(fin->fin_data[0]);
3197 dport = htons(fin->fin_data[1]);
3201 READ_ENTER(&softc->ipf_state);
3203 hvm = DOUBLE_HASH(hv);
3207 for (isp = &softs->ipf_state_table[hvm];
3208 ((is = *isp) != NULL); ) {
3209 isp = &is->is_hnext;
3210 if ((is->is_p != pr) || (is->is_v != v))
3212 fin->fin_flx &= ~FI_OOW;
3213 is = ipf_matchsrcdst(fin, is, &src, &dst, tcp, FI_CMP);
3215 if (pr == IPPROTO_TCP) {
3216 if (!ipf_state_tcp(softc, softs, fin,
3218 oow |= fin->fin_flx & FI_OOW;
3227 !(is->is_flags & (SI_CLONE|SI_WILDP|SI_WILDA))) {
3230 ipf_ipsmove(softs, is, hv);
3231 MUTEX_DOWNGRADE(&softc->ipf_state);
3235 RWLOCK_EXIT(&softc->ipf_state);
3237 if ((softs->ipf_state_stats.iss_wild != 0) &&
3238 ((fin->fin_flx & FI_NOWILD) == 0)) {
3239 if (tryagain == 0) {
3242 } else if (tryagain == 1) {
3243 hv = fin->fin_fi.fi_p;
3245 * If we try to pretend this is a reply to a
3246 * multicast/broadcast packet then we need to
3247 * exclude part of the address from the hash
3250 if (fin->fin_out == 0) {
3251 hv += src.in4.s_addr;
3253 hv += dst.in4.s_addr;
3259 if (tryagain <= 2) {
3260 WRITE_ENTER(&softc->ipf_state);
3264 fin->fin_flx |= oow;
3270 if (GRE_REV(gre->gr_flags) == 1) {
3277 hvm = DOUBLE_HASH(hv);
3278 READ_ENTER(&softc->ipf_state);
3279 for (isp = &softs->ipf_state_table[hvm];
3280 ((is = *isp) != NULL); ) {
3281 isp = &is->is_hnext;
3282 if ((is->is_p != pr) || (is->is_v != v))
3284 is = ipf_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP);
3286 ifq = &softs->ipf_state_iptq;
3291 RWLOCK_EXIT(&softc->ipf_state);
3297 if (((is->is_sti.tqe_flags & TQE_RULEBASED) != 0) &&
3298 (is->is_tqehead[fin->fin_rev] != NULL))
3299 ifq = is->is_tqehead[fin->fin_rev];
3300 if (ifq != NULL && ifqp != NULL)
3303 SBUMP(ipf_state_stats.iss_lookup_miss);
3309 /* ------------------------------------------------------------------------ */
3310 /* Function: ipf_state_check */
3311 /* Returns: frentry_t* - NULL == search failed, */
3312 /* else pointer to rule for matching state */
3313 /* Parameters: fin(I) - pointer to packet information */
3314 /* passp(I) - pointer to filtering result flags */
3316 /* Check if a packet is associated with an entry in the state table. */
3317 /* ------------------------------------------------------------------------ */
3319 ipf_state_check(fin, passp)
3323 ipf_main_softc_t *softc = fin->fin_main_soft;
3324 ipf_state_softc_t *softs = softc->ipf_state_soft;
3333 if (softs->ipf_state_lock || (softs->ipf_state_list == NULL))
3336 if (fin->fin_flx & (FI_SHORT|FI_FRAGBODY|FI_BAD)) {
3337 SBUMPD(ipf_state_stats, iss_check_bad);
3341 if ((fin->fin_flx & FI_TCPUDP) ||
3342 (fin->fin_fi.fi_p == IPPROTO_ICMP)
3344 || (fin->fin_fi.fi_p == IPPROTO_ICMPV6)
3353 * Search the hash table for matching packet header info.
3355 is = ipf_state_lookup(fin, tcp, &ifq);
3360 case IPPROTO_ICMPV6 :
3363 if (fin->fin_v == 6) {
3364 is = ipf_checkicmp6matchingstate(fin);
3372 * No matching icmp state entry. Perhaps this is a
3373 * response to another state entry.
3375 is = ipf_checkicmpmatchingstate(fin);
3382 if (is->is_pass & FR_NEWISN) {
3383 if (fin->fin_out == 0)
3384 ipf_fixinisn(fin, is);
3385 else if (fin->fin_out == 1)
3386 ipf_fixoutisn(fin, is);
3391 ifq = &softs->ipf_state_udpacktq;
3393 ifq = &softs->ipf_state_udptq;
3397 SBUMP(ipf_state_stats.iss_check_miss);
3403 if ((fin->fin_out == 0) && (fr->fr_nattag.ipt_num[0] != 0)) {
3404 if (fin->fin_nattag == NULL) {
3405 RWLOCK_EXIT(&softc->ipf_state);
3406 SBUMPD(ipf_state_stats, iss_check_notag);
3409 if (ipf_matchtag(&fr->fr_nattag, fin->fin_nattag)!=0) {
3410 RWLOCK_EXIT(&softc->ipf_state);
3411 SBUMPD(ipf_state_stats, iss_check_nattag);
3415 (void) strncpy(fin->fin_group, FR_NAME(fr, fr_group),
3417 fin->fin_icode = fr->fr_icode;
3420 fin->fin_rule = is->is_rulen;
3424 * If this packet is a fragment and the rule says to track fragments,
3425 * then create a new fragment cache entry.
3427 if (fin->fin_flx & FI_FRAG && FR_ISPASS(is->is_pass) &&
3428 is->is_pass & FR_KEEPFRAG)
3429 (void) ipf_frag_new(softc, fin, is->is_pass);
3432 * For TCP packets, ifq == NULL. For all others, check if this new
3433 * queue is different to the last one it was on and move it if so.
3436 if ((tqe->tqe_flags & TQE_RULEBASED) != 0)
3437 ifq = is->is_tqehead[fin->fin_rev];
3439 MUTEX_ENTER(&is->is_lock);
3442 ipf_movequeue(softc->ipf_ticks, tqe, tqe->tqe_ifq, ifq);
3444 inout = (fin->fin_rev << 1) + fin->fin_out;
3445 is->is_pkts[inout]++;
3446 is->is_bytes[inout] += fin->fin_plen;
3447 fin->fin_pktnum = is->is_pkts[inout] + is->is_icmppkts[inout];
3449 MUTEX_EXIT(&is->is_lock);
3453 if (is->is_flags & IS_STATESYNC)
3454 ipf_sync_update(softc, SMC_STATE, fin, is->is_sync);
3456 RWLOCK_EXIT(&softc->ipf_state);
3458 SBUMP(ipf_state_stats.iss_hits);
3460 fin->fin_dif = &is->is_dif;
3461 fin->fin_tif = &is->is_tifs[fin->fin_rev];
3462 fin->fin_flx |= FI_STATE;
3463 if ((pass & FR_LOGFIRST) != 0)
3464 pass &= ~(FR_LOGFIRST|FR_LOG);
3470 /* ------------------------------------------------------------------------ */
3471 /* Function: ipf_fixoutisn */
3473 /* Parameters: fin(I) - pointer to packet information */
3474 /* is(I) - pointer to master state structure */
3476 /* Called only for outbound packets, adjusts the sequence number and the */
3477 /* TCP checksum to match that change. */
3478 /* ------------------------------------------------------------------------ */
3480 ipf_fixoutisn(fin, is)
3490 if ((is->is_flags & IS_ISNSYN) != 0) {
3491 if ((rev == 0) && (fin->fin_cksum < FI_CK_L4PART)) {
3492 seq = ntohl(tcp->th_seq);
3493 seq += is->is_isninc[0];
3494 tcp->th_seq = htonl(seq);
3495 ipf_fix_outcksum(0, &tcp->th_sum, is->is_sumd[0], 0);
3498 if ((is->is_flags & IS_ISNACK) != 0) {
3499 if ((rev == 1) && (fin->fin_cksum < FI_CK_L4PART)) {
3500 seq = ntohl(tcp->th_seq);
3501 seq += is->is_isninc[1];
3502 tcp->th_seq = htonl(seq);
3503 ipf_fix_outcksum(0, &tcp->th_sum, is->is_sumd[1], 0);
3509 /* ------------------------------------------------------------------------ */
3510 /* Function: ipf_fixinisn */
3512 /* Parameters: fin(I) - pointer to packet information */
3513 /* is(I) - pointer to master state structure */
3515 /* Called only for inbound packets, adjusts the acknowledge number and the */
3516 /* TCP checksum to match that change. */
3517 /* ------------------------------------------------------------------------ */
3519 ipf_fixinisn(fin, is)
3529 if ((is->is_flags & IS_ISNSYN) != 0) {
3530 if ((rev == 1) && (fin->fin_cksum < FI_CK_L4PART)) {
3531 ack = ntohl(tcp->th_ack);
3532 ack -= is->is_isninc[0];
3533 tcp->th_ack = htonl(ack);
3534 ipf_fix_incksum(0, &tcp->th_sum, is->is_sumd[0], 0);
3537 if ((is->is_flags & IS_ISNACK) != 0) {
3538 if ((rev == 0) && (fin->fin_cksum < FI_CK_L4PART)) {
3539 ack = ntohl(tcp->th_ack);
3540 ack -= is->is_isninc[1];
3541 tcp->th_ack = htonl(ack);
3542 ipf_fix_incksum(0, &tcp->th_sum, is->is_sumd[1], 0);
3548 /* ------------------------------------------------------------------------ */
3549 /* Function: ipf_state_sync */
3551 /* Parameters: softc(I) - pointer to soft context main structure */
3552 /* ifp(I) - pointer to interface */
3554 /* Walk through all state entries and if an interface pointer match is */
3555 /* found then look it up again, based on its name in case the pointer has */
3556 /* changed since last time. */
3558 /* If ifp is passed in as being non-null then we are only doing updates for */
3559 /* existing, matching, uses of it. */
3560 /* ------------------------------------------------------------------------ */
3562 ipf_state_sync(softc, ifp)
3563 ipf_main_softc_t *softc;
3566 ipf_state_softc_t *softs = softc->ipf_state_soft;
3570 if (softc->ipf_running <= 0)
3573 WRITE_ENTER(&softc->ipf_state);
3575 if (softc->ipf_running <= 0) {
3576 RWLOCK_EXIT(&softc->ipf_state);
3580 for (is = softs->ipf_state_list; is; is = is->is_next) {
3582 * Look up all the interface names in the state entry.
3584 for (i = 0; i < 4; i++) {
3585 if (ifp == NULL || ifp == is->is_ifp[i])
3586 is->is_ifp[i] = ipf_resolvenic(softc,
3591 RWLOCK_EXIT(&softc->ipf_state);
3595 /* ------------------------------------------------------------------------ */
3596 /* Function: ipf_state_del */
3597 /* Returns: int - 0 = deleted, else refernce count on active struct */
3598 /* Parameters: softc(I) - pointer to soft context main structure */
3599 /* is(I) - pointer to state structure to delete */
3600 /* why(I) - if not 0, log reason why it was deleted */
3601 /* Write Locks: ipf_state */
3603 /* Deletes a state entry from the enumerated list as well as the hash table */
3604 /* and timeout queue lists. Make adjustments to hash table statistics and */
3605 /* global counters as required. */
3606 /* ------------------------------------------------------------------------ */
3608 ipf_state_del(softc, is, why)
3609 ipf_main_softc_t *softc;
3613 ipf_state_softc_t *softs = softc->ipf_state_soft;
3618 * Since we want to delete this, remove it from the state table,
3619 * where it can be found & used, first.
3621 if (is->is_phnext != NULL) {
3622 *is->is_phnext = is->is_hnext;
3623 if (is->is_hnext != NULL)
3624 is->is_hnext->is_phnext = is->is_phnext;
3625 if (softs->ipf_state_table[is->is_hv] == NULL)
3626 softs->ipf_state_stats.iss_inuse--;
3627 softs->ipf_state_stats.iss_bucketlen[is->is_hv]--;
3629 is->is_phnext = NULL;
3630 is->is_hnext = NULL;
3635 * Because ipf_state_stats.iss_wild is a count of entries in the state
3636 * table that have wildcard flags set, only decerement it once
3639 if (is->is_flags & (SI_WILDP|SI_WILDA)) {
3640 if (!(is->is_flags & SI_CLONED)) {
3641 ATOMIC_DECL(softs->ipf_state_stats.iss_wild);
3643 is->is_flags &= ~(SI_WILDP|SI_WILDA);
3647 * Next, remove it from the timeout queue it is in.
3649 if (is->is_sti.tqe_ifq != NULL)
3650 ipf_deletequeueentry(&is->is_sti);
3653 * If it is still in use by something else, do not go any further,
3654 * but note that at this point it is now an orphan. How can this
3655 * be? ipf_state_flush() calls ipf_delete() directly because it wants
3656 * to empty the table out and if something has a hold on a state
3657 * entry (such as ipfstat), it'll do the deref path that'll bring
3658 * us back here to do the real delete & free.
3660 MUTEX_ENTER(&is->is_lock);
3661 if (is->is_me != NULL) {
3667 if (is->is_ref > 0) {
3671 MUTEX_EXIT(&is->is_lock);
3673 softs->ipf_state_stats.iss_orphan++;
3680 if (fr->fr_srctrack.ht_max_nodes != 0) {
3681 (void) ipf_ht_node_del(&fr->fr_srctrack,
3682 is->is_family, &is->is_src);
3686 ASSERT(is->is_ref == 0);
3687 MUTEX_EXIT(&is->is_lock);
3689 if (is->is_tqehead[0] != NULL) {
3690 if (ipf_deletetimeoutqueue(is->is_tqehead[0]) == 0)
3691 ipf_freetimeoutqueue(softc, is->is_tqehead[0]);
3693 if (is->is_tqehead[1] != NULL) {
3694 if (ipf_deletetimeoutqueue(is->is_tqehead[1]) == 0)
3695 ipf_freetimeoutqueue(softc, is->is_tqehead[1]);
3699 ipf_sync_del_state(softc->ipf_sync_soft, is->is_sync);
3702 * Now remove it from the linked list of known states
3704 if (is->is_pnext != NULL) {
3705 *is->is_pnext = is->is_next;
3707 if (is->is_next != NULL)
3708 is->is_next->is_pnext = is->is_pnext;
3710 is->is_pnext = NULL;
3714 if (softs->ipf_state_logging != 0 && why != 0)
3715 ipf_state_log(softc, is, why);
3717 if (is->is_p == IPPROTO_TCP)
3718 softs->ipf_state_stats.iss_fin++;
3720 softs->ipf_state_stats.iss_expire++;
3722 softs->ipf_state_stats.iss_orphan--;
3726 (void) ipf_derefrule(softc, &fr);
3729 softs->ipf_state_stats.iss_active_proto[is->is_p]--;
3731 MUTEX_DESTROY(&is->is_lock);
3733 softs->ipf_state_stats.iss_active--;
3739 /* ------------------------------------------------------------------------ */
3740 /* Function: ipf_state_expire */
3742 /* Parameters: softc(I) - pointer to soft context main structure */
3744 /* Slowly expire held state for thingslike UDP and ICMP. The algorithm */
3745 /* used here is to keep the queue sorted with the oldest things at the top */
3746 /* and the youngest at the bottom. So if the top one doesn't need to be */
3747 /* expired then neither will any under it. */
3748 /* ------------------------------------------------------------------------ */
3750 ipf_state_expire(softc)
3751 ipf_main_softc_t *softc;
3753 ipf_state_softc_t *softs = softc->ipf_state_soft;
3754 ipftq_t *ifq, *ifqnext;
3755 ipftqent_t *tqe, *tqn;
3760 WRITE_ENTER(&softc->ipf_state);
3761 for (ifq = softs->ipf_state_tcptq; ifq != NULL; ifq = ifq->ifq_next)
3762 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3763 if (tqe->tqe_die > softc->ipf_ticks)
3765 tqn = tqe->tqe_next;
3766 is = tqe->tqe_parent;
3767 ipf_state_del(softc, is, ISL_EXPIRE);
3770 for (ifq = softs->ipf_state_usertq; ifq != NULL; ifq = ifqnext) {
3771 ifqnext = ifq->ifq_next;
3773 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3774 if (tqe->tqe_die > softc->ipf_ticks)
3776 tqn = tqe->tqe_next;
3777 is = tqe->tqe_parent;
3778 ipf_state_del(softc, is, ISL_EXPIRE);
3782 for (ifq = softs->ipf_state_usertq; ifq != NULL; ifq = ifqnext) {
3783 ifqnext = ifq->ifq_next;
3785 if (((ifq->ifq_flags & IFQF_DELETE) != 0) &&
3786 (ifq->ifq_ref == 0)) {
3787 ipf_freetimeoutqueue(softc, ifq);
3791 if (softs->ipf_state_doflush) {
3792 (void) ipf_state_flush(softc, 2, 0);
3793 softs->ipf_state_doflush = 0;
3794 softs->ipf_state_wm_last = softc->ipf_ticks;
3797 RWLOCK_EXIT(&softc->ipf_state);
3802 /* ------------------------------------------------------------------------ */
3803 /* Function: ipf_state_flush */
3804 /* Returns: int - 0 == success, -1 == failure */
3805 /* Parameters: softc(I) - pointer to soft context main structure */
3806 /* which(I) - which flush action to perform */
3807 /* proto(I) - which protocol to flush (0 == ALL) */
3808 /* Write Locks: ipf_state */
3810 /* Flush state tables. Three actions currently defined: */
3811 /* which == 0 : flush all state table entries */
3812 /* which == 1 : flush TCP connections which have started to close but are */
3813 /* stuck for some reason. */
3814 /* which == 2 : flush TCP connections which have been idle for a long time, */
3815 /* starting at > 4 days idle and working back in successive half-*/
3816 /* days to at most 12 hours old. If this fails to free enough */
3817 /* slots then work backwards in half hour slots to 30 minutes. */
3818 /* If that too fails, then work backwards in 30 second intervals */
3819 /* for the last 30 minutes to at worst 30 seconds idle. */
3820 /* ------------------------------------------------------------------------ */
3822 ipf_state_flush(softc, which, proto)
3823 ipf_main_softc_t *softc;
3826 ipf_state_softc_t *softs = softc->ipf_state_soft;
3827 ipftqent_t *tqe, *tqn;
3828 ipstate_t *is, **isp;
3840 SBUMP(ipf_state_stats.iss_flush_all);
3842 * Style 0 flush removes everything...
3844 for (isp = &softs->ipf_state_list; ((is = *isp) != NULL); ) {
3845 if ((proto != 0) && (is->is_v != proto)) {
3849 if (ipf_state_del(softc, is, ISL_FLUSH) == 0)
3857 SBUMP(ipf_state_stats.iss_flush_closing);
3859 * Since we're only interested in things that are closing,
3860 * we can start with the appropriate timeout queue.
3862 for (ifq = softs->ipf_state_tcptq + IPF_TCPS_CLOSE_WAIT;
3863 ifq != NULL; ifq = ifq->ifq_next) {
3865 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3866 tqn = tqe->tqe_next;
3867 is = tqe->tqe_parent;
3868 if (is->is_p != IPPROTO_TCP)
3870 if (ipf_state_del(softc, is, ISL_FLUSH) == 0)
3876 * Also need to look through the user defined queues.
3878 for (ifq = softs->ipf_state_usertq; ifq != NULL;
3879 ifq = ifq->ifq_next) {
3880 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3881 tqn = tqe->tqe_next;
3882 is = tqe->tqe_parent;
3883 if (is->is_p != IPPROTO_TCP)
3886 if ((is->is_state[0] > IPF_TCPS_ESTABLISHED) &&
3887 (is->is_state[1] > IPF_TCPS_ESTABLISHED)) {
3888 if (ipf_state_del(softc, is,
3900 * Args 5-11 correspond to flushing those particular states
3901 * for TCP connections.
3903 case IPF_TCPS_CLOSE_WAIT :
3904 case IPF_TCPS_FIN_WAIT_1 :
3905 case IPF_TCPS_CLOSING :
3906 case IPF_TCPS_LAST_ACK :
3907 case IPF_TCPS_FIN_WAIT_2 :
3908 case IPF_TCPS_TIME_WAIT :
3909 case IPF_TCPS_CLOSED :
3910 SBUMP(ipf_state_stats.iss_flush_queue);
3911 tqn = softs->ipf_state_tcptq[which].ifq_head;
3912 while (tqn != NULL) {
3914 tqn = tqe->tqe_next;
3915 is = tqe->tqe_parent;
3916 if (ipf_state_del(softc, is, ISL_FLUSH) == 0)
3925 SBUMP(ipf_state_stats.iss_flush_state);
3927 * Take a large arbitrary number to mean the number of seconds
3928 * for which which consider to be the maximum value we'll allow
3929 * the expiration to be.
3931 which = IPF_TTLVAL(which);
3932 for (isp = &softs->ipf_state_list; ((is = *isp) != NULL); ) {
3933 if ((proto == 0) || (is->is_v == proto)) {
3934 if (softc->ipf_ticks - is->is_touched > which) {
3935 if (ipf_state_del(softc, is,
3952 SBUMP(ipf_state_stats.iss_flush_timeout);
3954 * Asked to remove inactive entries because the table is full, try
3955 * again, 3 times, if first attempt failed with a different criteria
3956 * each time. The order tried in must be in decreasing age.
3957 * Another alternative is to implement random drop and drop N entries
3958 * at random until N have been freed up.
3960 if (softc->ipf_ticks - softs->ipf_state_wm_last >
3961 softs->ipf_state_wm_freq) {
3962 removed = ipf_queueflush(softc, ipf_state_flush_entry,
3963 softs->ipf_state_tcptq,
3964 softs->ipf_state_usertq,
3965 &softs->ipf_state_stats.iss_active,
3966 softs->ipf_state_size,
3967 softs->ipf_state_wm_low);
3968 softs->ipf_state_wm_last = softc->ipf_ticks;
3976 /* ------------------------------------------------------------------------ */
3977 /* Function: ipf_state_flush_entry */
3978 /* Returns: int - 0 = entry deleted, else not deleted */
3979 /* Parameters: softc(I) - pointer to soft context main structure */
3980 /* entry(I) - pointer to state structure to delete */
3981 /* Write Locks: ipf_state */
3983 /* This function is a stepping stone between ipf_queueflush() and */
3984 /* ipf_state_del(). It is used so we can provide a uniform interface via */
3985 /* the ipf_queueflush() function. */
3986 /* ------------------------------------------------------------------------ */
3988 ipf_state_flush_entry(softc, entry)
3989 ipf_main_softc_t *softc;
3992 return ipf_state_del(softc, entry, ISL_FLUSH);
3996 /* ------------------------------------------------------------------------ */
3997 /* Function: ipf_tcp_age */
3998 /* Returns: int - 1 == state transition made, 0 == no change (rejected) */
3999 /* Parameters: tqe(I) - pointer to timeout queue information */
4000 /* fin(I) - pointer to packet information */
4001 /* tqtab(I) - TCP timeout queue table this is in */
4002 /* flags(I) - flags from state/NAT entry */
4003 /* ok(I) - can we advance state */
4005 /* Rewritten by Arjan de Vet <Arjan.deVet@adv.iae.nl>, 2000-07-29: */
4007 /* - (try to) base state transitions on real evidence only, */
4008 /* i.e. packets that are sent and have been received by ipfilter; */
4009 /* diagram 18.12 of TCP/IP volume 1 by W. Richard Stevens was used. */
4011 /* - deal with half-closed connections correctly; */
4013 /* - store the state of the source in state[0] such that ipfstat */
4014 /* displays the state as source/dest instead of dest/source; the calls */
4015 /* to ipf_tcp_age have been changed accordingly. */
4017 /* Internal Parameters: */
4019 /* state[0] = state of source (host that initiated connection) */
4020 /* state[1] = state of dest (host that accepted the connection) */
4022 /* dir == 0 : a packet from source to dest */
4023 /* dir == 1 : a packet from dest to source */
4025 /* A typical procession for a connection is as follows: */
4027 /* +--------------+-------------------+ */
4028 /* | Side '0' | Side '1' | */
4029 /* +--------------+-------------------+ */
4030 /* | 0 -> 1 (SYN) | | */
4031 /* | | 0 -> 2 (SYN-ACK) | */
4032 /* | 1 -> 3 (ACK) | | */
4033 /* | | 2 -> 4 (ACK-PUSH) | */
4034 /* | 3 -> 4 (ACK) | | */
4036 /* | | 4 -> 6 (FIN-ACK) | */
4037 /* | 4 -> 5 (ACK) | | */
4038 /* | | 6 -> 6 (ACK-PUSH) | */
4039 /* | 5 -> 5 (ACK) | | */
4040 /* | 5 -> 8 (FIN) | | */
4041 /* | | 6 -> 10 (ACK) | */
4042 /* +--------------+-------------------+ */
4044 /* Locking: it is assumed that the parent of the tqe structure is locked. */
4045 /* ------------------------------------------------------------------------ */
4047 ipf_tcp_age(tqe, fin, tqtab, flags, ok)
4053 ipf_main_softc_t *softc = fin->fin_main_soft;
4054 int dlen, ostate, nstate, rval, dir;
4062 tcpflags = tcp->th_flags;
4063 dlen = fin->fin_dlen - (TCP_OFF(tcp) << 2);
4064 ostate = tqe->tqe_state[1 - dir];
4065 nstate = tqe->tqe_state[dir];
4067 if (tcpflags & TH_RST) {
4068 if (!(tcpflags & TH_PUSH) && !dlen)
4069 nstate = IPF_TCPS_CLOSED;
4071 nstate = IPF_TCPS_CLOSE_WAIT;
4073 if (ostate <= IPF_TCPS_ESTABLISHED) {
4074 tqe->tqe_state[1 - dir] = IPF_TCPS_CLOSE_WAIT;
4080 case IPF_TCPS_LISTEN: /* 0 */
4081 if ((tcpflags & TH_OPENING) == TH_OPENING) {
4083 * 'dir' received an S and sends SA in
4084 * response, LISTEN -> SYN_RECEIVED
4086 nstate = IPF_TCPS_SYN_RECEIVED;
4088 } else if ((tcpflags & TH_OPENING) == TH_SYN) {
4089 /* 'dir' sent S, LISTEN -> SYN_SENT */
4090 nstate = IPF_TCPS_SYN_SENT;
4094 * the next piece of code makes it possible to get
4095 * already established connections into the state table
4096 * after a restart or reload of the filter rules; this
4097 * does not work when a strict 'flags S keep state' is
4098 * used for tcp connections of course
4100 if (((flags & IS_TCPFSM) == 0) &&
4101 ((tcpflags & TH_ACKMASK) == TH_ACK)) {
4103 * we saw an A, guess 'dir' is in ESTABLISHED
4108 case IPF_TCPS_LISTEN :
4109 case IPF_TCPS_SYN_RECEIVED :
4110 nstate = IPF_TCPS_HALF_ESTAB;
4113 case IPF_TCPS_HALF_ESTAB :
4114 case IPF_TCPS_ESTABLISHED :
4115 nstate = IPF_TCPS_ESTABLISHED;
4123 * TODO: besides regular ACK packets we can have other
4124 * packets as well; it is yet to be determined how we
4125 * should initialize the states in those cases
4129 case IPF_TCPS_SYN_SENT: /* 1 */
4130 if ((tcpflags & ~(TH_ECN|TH_CWR)) == TH_SYN) {
4132 * A retransmitted SYN packet. We do not reset
4133 * the timeout here to ipf_tcptimeout because a
4134 * connection connect timeout does not renew
4135 * after every packet that is sent. We need to
4136 * set rval so as to indicate the packet has
4137 * passed the check for its flags being valid
4138 * in the TCP FSM. Setting rval to 2 has the
4139 * result of not resetting the timeout.
4142 } else if ((tcpflags & (TH_SYN|TH_FIN|TH_ACK)) ==
4145 * we see an A from 'dir' which is in SYN_SENT
4146 * state: 'dir' sent an A in response to an SA
4147 * which it received, SYN_SENT -> ESTABLISHED
4149 nstate = IPF_TCPS_ESTABLISHED;
4151 } else if (tcpflags & TH_FIN) {
4153 * we see an F from 'dir' which is in SYN_SENT
4154 * state and wants to close its side of the
4155 * connection; SYN_SENT -> FIN_WAIT_1
4157 nstate = IPF_TCPS_FIN_WAIT_1;
4159 } else if ((tcpflags & TH_OPENING) == TH_OPENING) {
4161 * we see an SA from 'dir' which is already in
4162 * SYN_SENT state, this means we have a
4163 * simultaneous open; SYN_SENT -> SYN_RECEIVED
4165 nstate = IPF_TCPS_SYN_RECEIVED;
4170 case IPF_TCPS_SYN_RECEIVED: /* 2 */
4171 if ((tcpflags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) {
4173 * we see an A from 'dir' which was in
4174 * SYN_RECEIVED state so it must now be in
4175 * established state, SYN_RECEIVED ->
4178 nstate = IPF_TCPS_ESTABLISHED;
4180 } else if ((tcpflags & ~(TH_ECN|TH_CWR)) ==
4183 * We see an SA from 'dir' which is already in
4184 * SYN_RECEIVED state.
4187 } else if (tcpflags & TH_FIN) {
4189 * we see an F from 'dir' which is in
4190 * SYN_RECEIVED state and wants to close its
4191 * side of the connection; SYN_RECEIVED ->
4194 nstate = IPF_TCPS_FIN_WAIT_1;
4199 case IPF_TCPS_HALF_ESTAB: /* 3 */
4200 if (tcpflags & TH_FIN) {
4201 nstate = IPF_TCPS_FIN_WAIT_1;
4203 } else if ((tcpflags & TH_ACKMASK) == TH_ACK) {
4205 * If we've picked up a connection in mid
4206 * flight, we could be looking at a follow on
4207 * packet from the same direction as the one
4208 * that created this state. Recognise it but
4209 * do not advance the entire connection's
4214 case IPF_TCPS_LISTEN :
4215 case IPF_TCPS_SYN_SENT :
4216 case IPF_TCPS_SYN_RECEIVED :
4219 case IPF_TCPS_HALF_ESTAB :
4220 case IPF_TCPS_ESTABLISHED :
4221 nstate = IPF_TCPS_ESTABLISHED;
4230 case IPF_TCPS_ESTABLISHED: /* 4 */
4232 if (tcpflags & TH_FIN) {
4234 * 'dir' closed its side of the connection;
4235 * this gives us a half-closed connection;
4236 * ESTABLISHED -> FIN_WAIT_1
4238 if (ostate == IPF_TCPS_FIN_WAIT_1) {
4239 nstate = IPF_TCPS_CLOSING;
4241 nstate = IPF_TCPS_FIN_WAIT_1;
4243 } else if (tcpflags & TH_ACK) {
4245 * an ACK, should we exclude other flags here?
4247 if (ostate == IPF_TCPS_FIN_WAIT_1) {
4249 * We know the other side did an active
4250 * close, so we are ACKing the recvd
4251 * FIN packet (does the window matching
4252 * code guarantee this?) and go into
4253 * CLOSE_WAIT state; this gives us a
4254 * half-closed connection
4256 nstate = IPF_TCPS_CLOSE_WAIT;
4257 } else if (ostate < IPF_TCPS_CLOSE_WAIT) {
4259 * still a fully established
4260 * connection reset timeout
4262 nstate = IPF_TCPS_ESTABLISHED;
4267 case IPF_TCPS_CLOSE_WAIT: /* 5 */
4269 if (tcpflags & TH_FIN) {
4271 * application closed and 'dir' sent a FIN,
4272 * we're now going into LAST_ACK state
4274 nstate = IPF_TCPS_LAST_ACK;
4277 * we remain in CLOSE_WAIT because the other
4278 * side has closed already and we did not
4279 * close our side yet; reset timeout
4281 nstate = IPF_TCPS_CLOSE_WAIT;
4285 case IPF_TCPS_FIN_WAIT_1: /* 6 */
4287 if ((tcpflags & TH_ACK) &&
4288 ostate > IPF_TCPS_CLOSE_WAIT) {
4290 * if the other side is not active anymore
4291 * it has sent us a FIN packet that we are
4292 * ack'ing now with an ACK; this means both
4293 * sides have now closed the connection and
4294 * we go into TIME_WAIT
4297 * XXX: how do we know we really are ACKing
4298 * the FIN packet here? does the window code
4301 nstate = IPF_TCPS_LAST_ACK;
4304 * we closed our side of the connection
4305 * already but the other side is still active
4306 * (ESTABLISHED/CLOSE_WAIT); continue with
4307 * this half-closed connection
4309 nstate = IPF_TCPS_FIN_WAIT_1;
4313 case IPF_TCPS_CLOSING: /* 7 */
4314 if ((tcpflags & (TH_FIN|TH_ACK)) == TH_ACK) {
4315 nstate = IPF_TCPS_TIME_WAIT;
4320 case IPF_TCPS_LAST_ACK: /* 8 */
4321 if (tcpflags & TH_ACK) {
4325 * we cannot detect when we go out of LAST_ACK state
4326 * to CLOSED because that is based on the reception
4327 * of ACK packets; ipfilter can only detect that a
4328 * packet has been sent by a host
4332 case IPF_TCPS_FIN_WAIT_2: /* 9 */
4336 case IPF_TCPS_TIME_WAIT: /* 10 */
4337 /* we're in 2MSL timeout now */
4338 if (ostate == IPF_TCPS_LAST_ACK) {
4339 nstate = IPF_TCPS_CLOSED;
4346 case IPF_TCPS_CLOSED: /* 11 */
4351 #if !defined(_KERNEL)
4359 * If rval == 2 then do not update the queue position, but treat the
4360 * packet as being ok.
4364 else if (rval == 1) {
4366 tqe->tqe_state[dir] = nstate;
4367 if ((tqe->tqe_flags & TQE_RULEBASED) == 0)
4368 ipf_movequeue(softc->ipf_ticks, tqe, tqe->tqe_ifq,
4376 /* ------------------------------------------------------------------------ */
4377 /* Function: ipf_state_log */
4379 /* Parameters: softc(I) - pointer to soft context main structure */
4380 /* is(I) - pointer to state structure */
4381 /* type(I) - type of log entry to create */
4383 /* Creates a state table log entry using the state structure and type info. */
4384 /* passed in. Log packet/byte counts, source/destination address and other */
4385 /* protocol specific information. */
4386 /* ------------------------------------------------------------------------ */
4388 ipf_state_log(softc, is, type)
4389 ipf_main_softc_t *softc;
4400 * Copy information out of the ipstate_t structure and into the
4401 * structure used for logging.
4403 ipsl.isl_type = type;
4404 ipsl.isl_pkts[0] = is->is_pkts[0] + is->is_icmppkts[0];
4405 ipsl.isl_bytes[0] = is->is_bytes[0];
4406 ipsl.isl_pkts[1] = is->is_pkts[1] + is->is_icmppkts[1];
4407 ipsl.isl_bytes[1] = is->is_bytes[1];
4408 ipsl.isl_pkts[2] = is->is_pkts[2] + is->is_icmppkts[2];
4409 ipsl.isl_bytes[2] = is->is_bytes[2];
4410 ipsl.isl_pkts[3] = is->is_pkts[3] + is->is_icmppkts[3];
4411 ipsl.isl_bytes[3] = is->is_bytes[3];
4412 ipsl.isl_src = is->is_src;
4413 ipsl.isl_dst = is->is_dst;
4414 ipsl.isl_p = is->is_p;
4415 ipsl.isl_v = is->is_v;
4416 ipsl.isl_flags = is->is_flags;
4417 ipsl.isl_tag = is->is_tag;
4418 ipsl.isl_rulen = is->is_rulen;
4419 (void) strncpy(ipsl.isl_group, is->is_group, FR_GROUPLEN);
4421 if (ipsl.isl_p == IPPROTO_TCP || ipsl.isl_p == IPPROTO_UDP) {
4422 ipsl.isl_sport = is->is_sport;
4423 ipsl.isl_dport = is->is_dport;
4424 if (ipsl.isl_p == IPPROTO_TCP) {
4425 ipsl.isl_state[0] = is->is_state[0];
4426 ipsl.isl_state[1] = is->is_state[1];
4428 } else if (ipsl.isl_p == IPPROTO_ICMP) {
4429 ipsl.isl_itype = is->is_icmp.ici_type;
4430 } else if (ipsl.isl_p == IPPROTO_ICMPV6) {
4431 ipsl.isl_itype = is->is_icmp.ici_type;
4433 ipsl.isl_ps.isl_filler[0] = 0;
4434 ipsl.isl_ps.isl_filler[1] = 0;
4438 sizes[0] = sizeof(ipsl);
4441 (void) ipf_log_items(softc, IPL_LOGSTATE, NULL, items, sizes, types, 1);
4447 /* ------------------------------------------------------------------------ */
4448 /* Function: ipf_checkicmp6matchingstate */
4449 /* Returns: ipstate_t* - NULL == no match found, */
4450 /* else pointer to matching state entry */
4451 /* Parameters: fin(I) - pointer to packet information */
4452 /* Locks: NULL == no locks, else Read Lock on ipf_state */
4454 /* If we've got an ICMPv6 error message, using the information stored in */
4455 /* the ICMPv6 packet, look for a matching state table entry. */
4456 /* ------------------------------------------------------------------------ */
4458 ipf_checkicmp6matchingstate(fin)
4461 ipf_main_softc_t *softc = fin->fin_main_soft;
4462 ipf_state_softc_t *softs = softc->ipf_state_soft;
4463 struct icmp6_hdr *ic6, *oic;
4464 ipstate_t *is, **isp;
4465 u_short sport, dport;
4477 * Does it at least have the return (basic) IP header ?
4478 * Is it an actual recognised ICMP error type?
4479 * Only a basic IP header (no options) should be with
4480 * an ICMP error header.
4482 if ((fin->fin_v != 6) || (fin->fin_plen < ICMP6ERR_MINPKTLEN) ||
4483 !(fin->fin_flx & FI_ICMPERR)) {
4484 SBUMPD(ipf_state_stats, iss_icmp_bad);
4489 type = ic6->icmp6_type;
4491 oip6 = (ip6_t *)((char *)ic6 + ICMPERR_ICMPHLEN);
4492 if (fin->fin_plen < sizeof(*oip6)) {
4493 SBUMPD(ipf_state_stats, iss_icmp_short);
4497 bcopy((char *)fin, (char *)&ofin, sizeof(*fin));
4499 ofin.fin_ifp = fin->fin_ifp;
4500 ofin.fin_out = !fin->fin_out;
4501 ofin.fin_m = NULL; /* if dereferenced, panic XXX */
4502 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */
4505 * We make a fin entry to be able to feed it to
4506 * matchsrcdst. Note that not all fields are necessary
4507 * but this is the cleanest way. Note further we fill
4508 * in fin_mp such that if someone uses it we'll get
4509 * a kernel panic. ipf_matchsrcdst does not use this.
4511 * watch out here, as ip is in host order and oip6 in network
4512 * order. Any change we make must be undone afterwards.
4514 savelen = oip6->ip6_plen;
4515 oip6->ip6_plen = htons(fin->fin_dlen - ICMPERR_ICMPHLEN);
4516 ofin.fin_flx = FI_NOCKSUM;
4517 ofin.fin_ip = (ip_t *)oip6;
4518 (void) ipf_makefrip(sizeof(*oip6), (ip_t *)oip6, &ofin);
4519 ofin.fin_flx &= ~(FI_BAD|FI_SHORT);
4520 oip6->ip6_plen = savelen;
4524 * an ICMP error can never generate an ICMP error in response.
4526 if (ofin.fin_flx & FI_ICMPERR) {
4527 DT1(iss_icmp6_icmperr, fr_info_t *, &ofin);
4528 SBUMP(ipf_state_stats.iss_icmp6_icmperr);
4532 if (oip6->ip6_nxt == IPPROTO_ICMPV6) {
4535 * an ICMP error can only be generated as a result of an
4536 * ICMP query, not as the response on an ICMP error
4538 * XXX theoretically ICMP_ECHOREP and the other reply's are
4539 * ICMP query's as well, but adding them here seems strange XXX
4541 if (!(oic->icmp6_type & ICMP6_INFOMSG_MASK)) {
4542 DT1(iss_icmp6_notinfo, fr_info_t *, &ofin);
4543 SBUMP(ipf_state_stats.iss_icmp6_notinfo);
4548 * perform a lookup of the ICMP packet in the state table
4550 hv = (pr = oip6->ip6_nxt);
4551 src.in6 = oip6->ip6_src;
4552 hv += src.in4.s_addr;
4553 dst.in6 = oip6->ip6_dst;
4554 hv += dst.in4.s_addr;
4555 hv += oic->icmp6_id;
4556 hv += oic->icmp6_seq;
4557 hv = DOUBLE_HASH(hv);
4559 READ_ENTER(&softc->ipf_state);
4560 for (isp = &softs->ipf_state_table[hv];
4561 ((is = *isp) != NULL); ) {
4563 isp = &is->is_hnext;
4564 if ((is->is_p == pr) &&
4565 !(is->is_pass & FR_NOICMPERR) &&
4566 (oic->icmp6_id == ic->ici_id) &&
4567 (oic->icmp6_seq == ic->ici_seq) &&
4568 (is = ipf_matchsrcdst(&ofin, is, &src,
4569 &dst, NULL, FI_ICMPCMP))) {
4571 * in the state table ICMP query's are stored
4572 * with the type of the corresponding ICMP
4573 * response. Correct here
4575 if (((ic->ici_type == ICMP6_ECHO_REPLY) &&
4576 (oic->icmp6_type == ICMP6_ECHO_REQUEST)) ||
4577 (ic->ici_type - 1 == oic->icmp6_type )) {
4578 if (!ipf_allowstateicmp(fin, is, &src))
4583 RWLOCK_EXIT(&softc->ipf_state);
4584 SBUMPD(ipf_state_stats, iss_icmp6_miss);
4588 hv = (pr = oip6->ip6_nxt);
4589 src.in6 = oip6->ip6_src;
4594 dst.in6 = oip6->ip6_dst;
4602 switch (oip6->ip6_nxt)
4606 tcp = (tcphdr_t *)(oip6 + 1);
4607 dport = tcp->th_dport;
4608 sport = tcp->th_sport;
4613 case IPPROTO_ICMPV6 :
4614 oic = (struct icmp6_hdr *)(oip6 + 1);
4615 hv += oic->icmp6_id;
4616 hv += oic->icmp6_seq;
4623 hv = DOUBLE_HASH(hv);
4625 READ_ENTER(&softc->ipf_state);
4626 for (isp = &softs->ipf_state_table[hv]; ((is = *isp) != NULL); ) {
4627 isp = &is->is_hnext;
4629 * Only allow this icmp though if the
4630 * encapsulated packet was allowed through the
4631 * other way around. Note that the minimal amount
4632 * of info present does not allow for checking against
4633 * tcp internals such as seq and ack numbers.
4635 if ((is->is_p != pr) || (is->is_v != 6) ||
4636 (is->is_pass & FR_NOICMPERR))
4638 is = ipf_matchsrcdst(&ofin, is, &src, &dst, tcp, FI_ICMPCMP);
4639 if ((is != NULL) && (ipf_allowstateicmp(fin, is, &src) == 0))
4642 RWLOCK_EXIT(&softc->ipf_state);
4643 SBUMPD(ipf_state_stats, iss_icmp_miss);
4649 /* ------------------------------------------------------------------------ */
4650 /* Function: ipf_sttab_init */
4652 /* Parameters: softc(I) - pointer to soft context main structure */
4653 /* tqp(I) - pointer to an array of timeout queues for TCP */
4655 /* Initialise the array of timeout queues for TCP. */
4656 /* ------------------------------------------------------------------------ */
4658 ipf_sttab_init(softc, tqp)
4659 ipf_main_softc_t *softc;
4664 for (i = IPF_TCP_NSTATES - 1; i >= 0; i--) {
4665 IPFTQ_INIT(&tqp[i], 0, "ipftq tcp tab");
4666 tqp[i].ifq_next = tqp + i + 1;
4668 tqp[IPF_TCP_NSTATES - 1].ifq_next = NULL;
4669 tqp[IPF_TCPS_CLOSED].ifq_ttl = softc->ipf_tcpclosed;
4670 tqp[IPF_TCPS_LISTEN].ifq_ttl = softc->ipf_tcptimeout;
4671 tqp[IPF_TCPS_SYN_SENT].ifq_ttl = softc->ipf_tcpsynsent;
4672 tqp[IPF_TCPS_SYN_RECEIVED].ifq_ttl = softc->ipf_tcpsynrecv;
4673 tqp[IPF_TCPS_ESTABLISHED].ifq_ttl = softc->ipf_tcpidletimeout;
4674 tqp[IPF_TCPS_CLOSE_WAIT].ifq_ttl = softc->ipf_tcphalfclosed;
4675 tqp[IPF_TCPS_FIN_WAIT_1].ifq_ttl = softc->ipf_tcphalfclosed;
4676 tqp[IPF_TCPS_CLOSING].ifq_ttl = softc->ipf_tcptimeout;
4677 tqp[IPF_TCPS_LAST_ACK].ifq_ttl = softc->ipf_tcplastack;
4678 tqp[IPF_TCPS_FIN_WAIT_2].ifq_ttl = softc->ipf_tcpclosewait;
4679 tqp[IPF_TCPS_TIME_WAIT].ifq_ttl = softc->ipf_tcptimewait;
4680 tqp[IPF_TCPS_HALF_ESTAB].ifq_ttl = softc->ipf_tcptimeout;
4684 /* ------------------------------------------------------------------------ */
4685 /* Function: ipf_sttab_destroy */
4687 /* Parameters: tqp(I) - pointer to an array of timeout queues for TCP */
4689 /* Do whatever is necessary to "destroy" each of the entries in the array */
4690 /* of timeout queues for TCP. */
4691 /* ------------------------------------------------------------------------ */
4693 ipf_sttab_destroy(tqp)
4698 for (i = IPF_TCP_NSTATES - 1; i >= 0; i--)
4699 MUTEX_DESTROY(&tqp[i].ifq_lock);
4703 /* ------------------------------------------------------------------------ */
4704 /* Function: ipf_state_deref */
4706 /* Parameters: softc(I) - pointer to soft context main structure */
4707 /* isp(I) - pointer to pointer to state table entry */
4709 /* Decrement the reference counter for this state table entry and free it */
4710 /* if there are no more things using it. */
4712 /* This function is only called when cleaning up after increasing is_ref by */
4713 /* one earlier in the 'code path' so if is_ref is 1 when entering, we do */
4714 /* have an orphan, otherwise not. However there is a possible race between */
4715 /* the entry being deleted via flushing with an ioctl call (that calls the */
4716 /* delete function directly) and the tail end of packet processing so we */
4717 /* need to grab is_lock before doing the check to synchronise the two code */
4720 /* When operating in userland (ipftest), we have no timers to clear a state */
4721 /* entry. Therefore, we make a few simple tests before deleting an entry */
4722 /* outright. We compare states on each side looking for a combination of */
4723 /* TIME_WAIT (should really be FIN_WAIT_2?) and LAST_ACK. Then we factor */
4724 /* in packet direction with the interface list to make sure we don't */
4725 /* prematurely delete an entry on a final inbound packet that's we're also */
4726 /* supposed to route elsewhere. */
4728 /* Internal parameters: */
4729 /* state[0] = state of source (host that initiated connection) */
4730 /* state[1] = state of dest (host that accepted the connection) */
4732 /* dir == 0 : a packet from source to dest */
4733 /* dir == 1 : a packet from dest to source */
4734 /* ------------------------------------------------------------------------ */
4736 ipf_state_deref(softc, isp)
4737 ipf_main_softc_t *softc;
4740 ipstate_t *is = *isp;
4745 MUTEX_ENTER(&is->is_lock);
4746 if (is->is_ref > 1) {
4748 MUTEX_EXIT(&is->is_lock);
4750 if ((is->is_sti.tqe_state[0] > IPF_TCPS_ESTABLISHED) ||
4751 (is->is_sti.tqe_state[1] > IPF_TCPS_ESTABLISHED)) {
4752 ipf_state_del(softc, is, ISL_EXPIRE);
4757 MUTEX_EXIT(&is->is_lock);
4759 WRITE_ENTER(&softc->ipf_state);
4760 ipf_state_del(softc, is, ISL_ORPHAN);
4761 RWLOCK_EXIT(&softc->ipf_state);
4765 /* ------------------------------------------------------------------------ */
4766 /* Function: ipf_state_setqueue */
4768 /* Parameters: softc(I) - pointer to soft context main structure */
4769 /* is(I) - pointer to state structure */
4770 /* rev(I) - forward(0) or reverse(1) direction */
4771 /* Locks: ipf_state (read or write) */
4773 /* Put the state entry on its default queue entry, using rev as a helped in */
4774 /* determining which queue it should be placed on. */
4775 /* ------------------------------------------------------------------------ */
4777 ipf_state_setqueue(softc, is, rev)
4778 ipf_main_softc_t *softc;
4782 ipf_state_softc_t *softs = softc->ipf_state_soft;
4783 ipftq_t *oifq, *nifq;
4785 if ((is->is_sti.tqe_flags & TQE_RULEBASED) != 0)
4786 nifq = is->is_tqehead[rev];
4794 case IPPROTO_ICMPV6 :
4796 nifq = &softs->ipf_state_icmpacktq;
4798 nifq = &softs->ipf_state_icmptq;
4803 nifq = &softs->ipf_state_icmpacktq;
4805 nifq = &softs->ipf_state_icmptq;
4808 nifq = softs->ipf_state_tcptq + is->is_state[rev];
4813 nifq = &softs->ipf_state_udpacktq;
4815 nifq = &softs->ipf_state_udptq;
4819 nifq = &softs->ipf_state_iptq;
4824 oifq = is->is_sti.tqe_ifq;
4826 * If it's currently on a timeout queue, move it from one queue to
4827 * another, else put it on the end of the newly determined queue.
4830 ipf_movequeue(softc->ipf_ticks, &is->is_sti, oifq, nifq);
4832 ipf_queueappend(softc->ipf_ticks, &is->is_sti, nifq, is);
4837 /* ------------------------------------------------------------------------ */
4838 /* Function: ipf_state_iter */
4839 /* Returns: int - 0 == success, else error */
4840 /* Parameters: softc(I) - pointer to main soft context */
4841 /* token(I) - pointer to ipftoken structure */
4842 /* itp(I) - pointer to ipfgeniter structure */
4843 /* obj(I) - pointer to data description structure */
4845 /* This function handles the SIOCGENITER ioctl for the state tables and */
4846 /* walks through the list of entries in the state table list (softs->ipf_state_list.) */
4847 /* ------------------------------------------------------------------------ */
4849 ipf_state_iter(softc, token, itp, obj)
4850 ipf_main_softc_t *softc;
4855 ipf_state_softc_t *softs = softc->ipf_state_soft;
4856 ipstate_t *is, *next, zero;
4859 if (itp->igi_data == NULL) {
4864 if (itp->igi_nitems < 1) {
4869 if (itp->igi_type != IPFGENITER_STATE) {
4874 is = token->ipt_data;
4875 if (is == (void *)-1) {
4881 obj->ipfo_type = IPFOBJ_IPSTATE;
4882 obj->ipfo_size = sizeof(ipstate_t);
4884 READ_ENTER(&softc->ipf_state);
4886 is = token->ipt_data;
4888 next = softs->ipf_state_list;
4894 * If we find a state entry to use, bump its reference count so that
4895 * it can be used for is_next when we come back.
4898 MUTEX_ENTER(&next->is_lock);
4900 MUTEX_EXIT(&next->is_lock);
4901 token->ipt_data = next;
4903 bzero(&zero, sizeof(zero));
4905 token->ipt_data = NULL;
4907 if (next->is_next == NULL)
4908 ipf_token_mark_complete(token);
4910 RWLOCK_EXIT(&softc->ipf_state);
4912 obj->ipfo_ptr = itp->igi_data;
4913 error = ipf_outobjk(softc, obj, next);
4915 ipf_state_deref(softc, &is);
4921 /* ------------------------------------------------------------------------ */
4922 /* Function: ipf_state_gettable */
4923 /* Returns: int - 0 = success, else error */
4924 /* Parameters: softc(I) - pointer to main soft context */
4925 /* softs(I) - pointer to state context structure */
4926 /* data(I) - pointer to ioctl data */
4928 /* This function handles ioctl requests for tables of state information. */
4929 /* At present the only table it deals with is the hash bucket statistics. */
4930 /* ------------------------------------------------------------------------ */
4932 ipf_state_gettable(softc, softs, data)
4933 ipf_main_softc_t *softc;
4934 ipf_state_softc_t *softs;
4940 error = ipf_inobj(softc, data, NULL, &table, IPFOBJ_GTABLE);
4944 if (table.ita_type != IPFTABLE_BUCKETS) {
4949 error = COPYOUT(softs->ipf_state_stats.iss_bucketlen, table.ita_table,
4950 softs->ipf_state_size * sizeof(u_int));
4959 /* ------------------------------------------------------------------------ */
4960 /* Function: ipf_state_setpending */
4962 /* Parameters: softc(I) - pointer to main soft context */
4963 /* is(I) - pointer to state structure */
4964 /* Locks: ipf_state (read or write) */
4966 /* Put the state entry on to the pending queue - this queue has a very */
4967 /* short lifetime where items are put that can't be deleted straight away */
4968 /* because of locking issues but we want to delete them ASAP, anyway. */
4969 /* ------------------------------------------------------------------------ */
4971 ipf_state_setpending(softc, is)
4972 ipf_main_softc_t *softc;
4975 ipf_state_softc_t *softs = softc->ipf_state_soft;
4978 oifq = is->is_sti.tqe_ifq;
4980 ipf_movequeue(softc->ipf_ticks, &is->is_sti, oifq,
4981 &softs->ipf_state_pending);
4983 ipf_queueappend(softc->ipf_ticks, &is->is_sti,
4984 &softs->ipf_state_pending, is);
4986 MUTEX_ENTER(&is->is_lock);
4987 if (is->is_me != NULL) {
4992 MUTEX_EXIT(&is->is_lock);
4996 /* ------------------------------------------------------------------------ */
4997 /* Function: ipf_state_matchflush */
4999 /* Parameters: softc(I) - pointer to main soft context */
5000 /* data(I) - pointer to state structure */
5001 /* Locks: ipf_state (read or write) */
5003 /* Flush all entries from the list of state entries that match the */
5004 /* properties in the array loaded. */
5005 /* ------------------------------------------------------------------------ */
5007 ipf_state_matchflush(softc, data)
5008 ipf_main_softc_t *softc;
5011 ipf_state_softc_t *softs = softc->ipf_state_soft;
5012 int *array, flushed, error;
5013 ipstate_t *state, *statenext;
5016 error = ipf_matcharray_load(softc, data, &obj, &array);
5022 for (state = softs->ipf_state_list; state != NULL; state = statenext) {
5023 statenext = state->is_next;
5024 if (ipf_state_matcharray(state, array, softc->ipf_ticks) == 0) {
5025 ipf_state_del(softc, state, ISL_FLUSH);
5030 obj.ipfo_retval = flushed;
5031 error = BCOPYOUT(&obj, data, sizeof(obj));
5033 KFREES(array, array[0] * sizeof(*array));
5039 /* ------------------------------------------------------------------------ */
5040 /* Function: ipf_state_matcharray */
5041 /* Returns: int - 0 = no match, 1 = match */
5042 /* Parameters: state(I) - pointer to state structure */
5043 /* array(I) - pointer to ipf matching expression */
5044 /* ticks(I) - current value of ipfilter tick timer */
5045 /* Locks: ipf_state (read or write) */
5047 /* Compare a state entry with the match array passed in and return a value */
5048 /* to indicate whether or not the matching was successful. */
5049 /* ------------------------------------------------------------------------ */
5051 ipf_state_matcharray(state, array, ticks)
5056 int i, n, *x, rv, p;
5063 for (; n > 0; x += 3 + x[3], rv = 0) {
5066 if (x[0] == IPF_EXP_END)
5070 * If we need to match the protocol and that doesn't match,
5071 * don't even both with the instruction array.
5073 p = e->ipfe_cmd >> 16;
5074 if ((p != 0) && (p != state->is_p))
5077 switch (e->ipfe_cmd)
5079 case IPF_EXP_IP_PR :
5080 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5081 rv |= (state->is_p == e->ipfe_arg0[i]);
5085 case IPF_EXP_IP_SRCADDR :
5086 if (state->is_v != 4)
5088 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5089 rv |= ((state->is_saddr &
5090 e->ipfe_arg0[i * 2 + 1]) ==
5091 e->ipfe_arg0[i * 2]);
5095 case IPF_EXP_IP_DSTADDR :
5096 if (state->is_v != 4)
5098 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5099 rv |= ((state->is_daddr &
5100 e->ipfe_arg0[i * 2 + 1]) ==
5101 e->ipfe_arg0[i * 2]);
5105 case IPF_EXP_IP_ADDR :
5106 if (state->is_v != 4)
5108 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5109 rv |= ((state->is_saddr &
5110 e->ipfe_arg0[i * 2 + 1]) ==
5111 e->ipfe_arg0[i * 2]) ||
5113 e->ipfe_arg0[i * 2 + 1]) ==
5114 e->ipfe_arg0[i * 2]);
5119 case IPF_EXP_IP6_SRCADDR :
5120 if (state->is_v != 6)
5122 for (i = 0; !rv && i < x[3]; i++) {
5123 rv |= IP6_MASKEQ(&state->is_src.in6,
5124 &e->ipfe_arg0[i * 8 + 4],
5125 &e->ipfe_arg0[i * 8]);
5129 case IPF_EXP_IP6_DSTADDR :
5130 if (state->is_v != 6)
5132 for (i = 0; !rv && i < x[3]; i++) {
5133 rv |= IP6_MASKEQ(&state->is_dst.in6,
5134 &e->ipfe_arg0[i * 8 + 4],
5135 &e->ipfe_arg0[i * 8]);
5139 case IPF_EXP_IP6_ADDR :
5140 if (state->is_v != 6)
5142 for (i = 0; !rv && i < x[3]; i++) {
5143 rv |= IP6_MASKEQ(&state->is_src.in6,
5144 &e->ipfe_arg0[i * 8 + 4],
5145 &e->ipfe_arg0[i * 8]) ||
5146 IP6_MASKEQ(&state->is_dst.in6,
5147 &e->ipfe_arg0[i * 8 + 4],
5148 &e->ipfe_arg0[i * 8]);
5153 case IPF_EXP_UDP_PORT :
5154 case IPF_EXP_TCP_PORT :
5155 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5156 rv |= (state->is_sport == e->ipfe_arg0[i]) ||
5157 (state->is_dport == e->ipfe_arg0[i]);
5161 case IPF_EXP_UDP_SPORT :
5162 case IPF_EXP_TCP_SPORT :
5163 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5164 rv |= (state->is_sport == e->ipfe_arg0[i]);
5168 case IPF_EXP_UDP_DPORT :
5169 case IPF_EXP_TCP_DPORT :
5170 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5171 rv |= (state->is_dport == e->ipfe_arg0[i]);
5175 case IPF_EXP_TCP_STATE :
5176 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5177 rv |= (state->is_state[0] == e->ipfe_arg0[i]) ||
5178 (state->is_state[1] == e->ipfe_arg0[i]);
5182 case IPF_EXP_IDLE_GT :
5183 rv |= (ticks - state->is_touched > e->ipfe_arg0[0]);
5188 * Factor in doing a negative match.
5200 /* ------------------------------------------------------------------------ */
5201 /* Function: ipf_state_settimeout */
5202 /* Returns: int 0 = success, else failure */
5203 /* Parameters: softc(I) - pointer to main soft context */
5204 /* t(I) - pointer to tuneable being changed */
5205 /* p(I) - pointer to the new value */
5207 /* Sets a timeout value for one of the many timeout queues. We find the */
5208 /* correct queue using a somewhat manual process of comparing the timeout */
5209 /* names for each specific value available and calling ipf_apply_timeout on */
5210 /* that queue so that all of the items on it are updated accordingly. */
5211 /* ------------------------------------------------------------------------ */
5213 ipf_state_settimeout(softc, t, p)
5214 struct ipf_main_softc_s *softc;
5218 ipf_state_softc_t *softs = softc->ipf_state_soft;
5221 * In case there is nothing to do...
5223 if (*t->ipft_pint == p->ipftu_int)
5226 if (!strncmp(t->ipft_name, "tcp_", 4))
5227 return ipf_settimeout_tcp(t, p, softs->ipf_state_tcptq);
5229 if (!strcmp(t->ipft_name, "udp_timeout")) {
5230 ipf_apply_timeout(&softs->ipf_state_udptq, p->ipftu_int);
5231 } else if (!strcmp(t->ipft_name, "udp_ack_timeout")) {
5232 ipf_apply_timeout(&softs->ipf_state_udpacktq, p->ipftu_int);
5233 } else if (!strcmp(t->ipft_name, "icmp_timeout")) {
5234 ipf_apply_timeout(&softs->ipf_state_icmptq, p->ipftu_int);
5235 } else if (!strcmp(t->ipft_name, "icmp_ack_timeout")) {
5236 ipf_apply_timeout(&softs->ipf_state_icmpacktq, p->ipftu_int);
5237 } else if (!strcmp(t->ipft_name, "ip_timeout")) {
5238 ipf_apply_timeout(&softs->ipf_state_iptq, p->ipftu_int);
5245 * Update the tuneable being set.
5247 *t->ipft_pint = p->ipftu_int;
5253 /* ------------------------------------------------------------------------ */
5254 /* Function: ipf_state_rehash */
5255 /* Returns: int 0 = success, else failure */
5256 /* Parameters: softc(I) - pointer to main soft context */
5257 /* t(I) - pointer to tuneable being changed */
5258 /* p(I) - pointer to the new value */
5260 /* To change the size of the state hash table at runtime, a new table has */
5261 /* to be allocated and then all of the existing entries put in it, bumping */
5262 /* up the bucketlength for it as we go along. */
5263 /* ------------------------------------------------------------------------ */
5265 ipf_state_rehash(softc, t, p)
5266 ipf_main_softc_t *softc;
5270 ipf_state_softc_t *softs = softc->ipf_state_soft;
5271 ipstate_t **newtab, *is;
5279 newsize = p->ipftu_int;
5281 * In case there is nothing to do...
5283 if (newsize == softs->ipf_state_size)
5286 KMALLOCS(newtab, ipstate_t **, newsize * sizeof(ipstate_t *));
5287 if (newtab == NULL) {
5292 KMALLOCS(bucketlens, u_int *, newsize * sizeof(u_int));
5293 if (bucketlens == NULL) {
5294 KFREES(newtab, newsize * sizeof(*softs->ipf_state_table));
5299 newseed = ipf_state_seed_alloc(newsize, softs->ipf_state_max);
5300 if (newseed == NULL) {
5301 KFREES(bucketlens, newsize * sizeof(*bucketlens));
5302 KFREES(newtab, newsize * sizeof(*newtab));
5307 for (maxbucket = 0, i = newsize; i > 0; i >>= 1)
5311 bzero((char *)newtab, newsize * sizeof(ipstate_t *));
5312 bzero((char *)bucketlens, newsize * sizeof(u_int));
5314 WRITE_ENTER(&softc->ipf_state);
5316 if (softs->ipf_state_table != NULL) {
5317 KFREES(softs->ipf_state_table,
5318 softs->ipf_state_size * sizeof(*softs->ipf_state_table));
5320 softs->ipf_state_table = newtab;
5322 if (softs->ipf_state_seed != NULL) {
5323 KFREES(softs->ipf_state_seed,
5324 softs->ipf_state_size * sizeof(*softs->ipf_state_seed));
5326 softs->ipf_state_seed = newseed;
5328 if (softs->ipf_state_stats.iss_bucketlen != NULL) {
5329 KFREES(softs->ipf_state_stats.iss_bucketlen,
5330 softs->ipf_state_size * sizeof(u_int));
5332 softs->ipf_state_stats.iss_bucketlen = bucketlens;
5333 softs->ipf_state_maxbucket = maxbucket;
5334 softs->ipf_state_size = newsize;
5337 * Walk through the entire list of state table entries and put them
5338 * in the new state table, somewhere. Because we have a new table,
5339 * we need to restart the counter of how many chains are in use.
5341 softs->ipf_state_stats.iss_inuse = 0;
5342 for (is = softs->ipf_state_list; is != NULL; is = is->is_next) {
5343 is->is_hnext = NULL;
5344 is->is_phnext = NULL;
5345 hv = is->is_hv % softs->ipf_state_size;
5347 if (softs->ipf_state_table[hv] != NULL)
5348 softs->ipf_state_table[hv]->is_phnext = &is->is_hnext;
5350 softs->ipf_state_stats.iss_inuse++;
5351 is->is_phnext = softs->ipf_state_table + hv;
5352 is->is_hnext = softs->ipf_state_table[hv];
5353 softs->ipf_state_table[hv] = is;
5354 softs->ipf_state_stats.iss_bucketlen[hv]++;
5356 RWLOCK_EXIT(&softc->ipf_state);
5362 /* ------------------------------------------------------------------------ */
5363 /* Function: ipf_state_add_tq */
5364 /* Returns: ipftq_t * - NULL = failure, else pointer to new timeout */
5366 /* Parameters: softc(I) - pointer to main soft context */
5367 /* ttl(I) - pointer to the ttl for the new queue */
5369 /* Request a pointer to a timeout queue that has a ttl as given by the */
5370 /* value being passed in. The timeout queue is added tot the list of those */
5371 /* used internally for stateful filtering. */
5372 /* ------------------------------------------------------------------------ */
5374 ipf_state_add_tq(softc, ttl)
5375 ipf_main_softc_t *softc;
5378 ipf_state_softc_t *softs = softc->ipf_state_soft;
5380 return ipf_addtimeoutqueue(softc, &softs->ipf_state_usertq, ttl);
5386 * Display the built up state table rules and mapping entries.
5389 ipf_state_dump(softc, arg)
5390 ipf_main_softc_t *softc;
5393 ipf_state_softc_t *softs = arg;
5396 printf("List of active state sessions:\n");
5397 for (ips = softs->ipf_state_list; ips != NULL; )
5398 ips = printstate(ips, opts & (OPT_DEBUG|OPT_VERBOSE),