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__) && \
24 #include "opt_inet6.h"
26 #if !defined(_KERNEL) && !defined(__KERNEL__)
34 #if defined(_KERNEL) && defined(__FreeBSD__)
35 # include <sys/filio.h>
36 # include <sys/fcntl.h>
38 # include <sys/ioctl.h>
41 # include <sys/protosw.h>
42 #include <sys/socket.h>
44 # include <sys/systm.h>
46 # include <sys/mbuf.h>
50 # include <sys/filio.h>
51 # include <sys/byteorder.h>
53 # include <sys/dditypes.h>
55 # include <sys/stream.h>
56 # include <sys/kmem.h>
63 #include <netinet/in.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/ip.h>
66 #include <netinet/tcp.h>
67 # include <netinet/tcp_fsm.h>
68 #include <netinet/udp.h>
69 #include <netinet/ip_icmp.h>
73 #include "netinet/ip_compat.h"
74 #include "netinet/ip_fil.h"
75 #include "netinet/ip_nat.h"
76 #include "netinet/ip_frag.h"
77 #include "netinet/ip_state.h"
78 #include "netinet/ip_proxy.h"
79 #include "netinet/ip_lookup.h"
80 #include "netinet/ip_dstlist.h"
81 #include "netinet/ip_sync.h"
83 #include <netinet/icmp6.h>
86 # include <sys/malloc.h>
87 # if defined(_KERNEL) && !defined(IPFILTER_LKM)
88 # include <sys/libkern.h>
89 # include <sys/systm.h>
96 static const char sccsid[] = "@(#)ip_state.c 1.8 6/5/96 (C) 1993-2000 Darren Reed";
97 static const char rcsid[] = "@(#)$Id$";
101 static ipftuneable_t ipf_state_tuneables[] = {
102 { { (void *)offsetof(ipf_state_softc_t, ipf_state_max) },
103 "state_max", 1, 0x7fffffff,
104 stsizeof(ipf_state_softc_t, ipf_state_max),
106 { { (void *)offsetof(ipf_state_softc_t, ipf_state_size) },
107 "state_size", 1, 0x7fffffff,
108 stsizeof(ipf_state_softc_t, ipf_state_size),
109 0, NULL, ipf_state_rehash },
110 { { (void *)offsetof(ipf_state_softc_t, ipf_state_lock) },
112 stsizeof(ipf_state_softc_t, ipf_state_lock),
113 IPFT_RDONLY, NULL, NULL },
114 { { (void *)offsetof(ipf_state_softc_t, ipf_state_maxbucket) },
115 "state_maxbucket", 1, 0x7fffffff,
116 stsizeof(ipf_state_softc_t, ipf_state_maxbucket),
118 { { (void *)offsetof(ipf_state_softc_t, ipf_state_logging) },
119 "state_logging",0, 1,
120 stsizeof(ipf_state_softc_t, ipf_state_logging),
122 { { (void *)offsetof(ipf_state_softc_t, ipf_state_wm_high) },
123 "state_wm_high",2, 100,
124 stsizeof(ipf_state_softc_t, ipf_state_wm_high),
126 { { (void *)offsetof(ipf_state_softc_t, ipf_state_wm_low) },
127 "state_wm_low", 1, 99,
128 stsizeof(ipf_state_softc_t, ipf_state_wm_low),
130 { { (void *)offsetof(ipf_state_softc_t, ipf_state_wm_freq) },
131 "state_wm_freq",2, 999999,
132 stsizeof(ipf_state_softc_t, ipf_state_wm_freq),
140 #define SINCL(x) ATOMIC_INCL(softs->x)
141 #define SBUMP(x) (softs->x)++
142 #define SBUMPD(x, y) do { (softs->x.y)++; DT(y); } while (0)
143 #define SBUMPDX(x, y, z)do { (softs->x.y)++; DT(z); } while (0)
146 static ipstate_t *ipf_checkicmp6matchingstate(fr_info_t *);
148 static int ipf_allowstateicmp(fr_info_t *, ipstate_t *, i6addr_t *);
149 static ipstate_t *ipf_matchsrcdst(fr_info_t *, ipstate_t *, i6addr_t *,
150 i6addr_t *, tcphdr_t *, u_32_t);
151 static ipstate_t *ipf_checkicmpmatchingstate(fr_info_t *);
152 static int ipf_state_flush_entry(ipf_main_softc_t *, void *);
153 static ips_stat_t *ipf_state_stats(ipf_main_softc_t *);
154 static int ipf_state_del(ipf_main_softc_t *, ipstate_t *, int);
155 static int ipf_state_remove(ipf_main_softc_t *, caddr_t);
156 static int ipf_state_match(ipstate_t *is1, ipstate_t *is2);
157 static int ipf_state_matchaddresses(ipstate_t *is1, ipstate_t *is2);
158 static int ipf_state_matchipv4addrs(ipstate_t *is1, ipstate_t *is2);
159 static int ipf_state_matchipv6addrs(ipstate_t *is1, ipstate_t *is2);
160 static int ipf_state_matchisps(ipstate_t *is1, ipstate_t *is2);
161 static int ipf_state_matchports(udpinfo_t *is1, udpinfo_t *is2);
162 static int ipf_state_matcharray(ipstate_t *, int *, u_long);
163 static void ipf_ipsmove(ipf_state_softc_t *, ipstate_t *, u_int);
164 static int ipf_state_tcp(ipf_main_softc_t *, ipf_state_softc_t *,
165 fr_info_t *, tcphdr_t *, ipstate_t *);
166 static int ipf_tcpoptions(ipf_state_softc_t *, fr_info_t *,
167 tcphdr_t *, tcpdata_t *);
168 static ipstate_t *ipf_state_clone(fr_info_t *, tcphdr_t *, ipstate_t *);
169 static void ipf_fixinisn(fr_info_t *, ipstate_t *);
170 static void ipf_fixoutisn(fr_info_t *, ipstate_t *);
171 static void ipf_checknewisn(fr_info_t *, ipstate_t *);
172 static int ipf_state_iter(ipf_main_softc_t *, ipftoken_t *,
173 ipfgeniter_t *, ipfobj_t *);
174 static int ipf_state_gettable(ipf_main_softc_t *, ipf_state_softc_t *,
176 static int ipf_state_tcpinwindow(struct fr_info *, struct tcpdata *,
177 struct tcpdata *, tcphdr_t *, int);
179 static int ipf_state_getent(ipf_main_softc_t *, ipf_state_softc_t *,
181 static int ipf_state_putent(ipf_main_softc_t *, ipf_state_softc_t *,
184 #define ONE_DAY IPF_TTLVAL(1 * 86400) /* 1 day */
185 #define FIVE_DAYS (5 * ONE_DAY)
186 #define DOUBLE_HASH(x) (((x) + softs->ipf_state_seed[(x) % \
187 softs->ipf_state_size]) % softs->ipf_state_size)
190 /* ------------------------------------------------------------------------ */
191 /* Function: ipf_state_main_load */
192 /* Returns: int - 0 == success, -1 == failure */
193 /* Parameters: Nil */
195 /* A null-op function that exists as a placeholder so that the flow in */
196 /* other functions is obvious. */
197 /* ------------------------------------------------------------------------ */
199 ipf_state_main_load()
205 /* ------------------------------------------------------------------------ */
206 /* Function: ipf_state_main_unload */
207 /* Returns: int - 0 == success, -1 == failure */
208 /* Parameters: Nil */
210 /* A null-op function that exists as a placeholder so that the flow in */
211 /* other functions is obvious. */
212 /* ------------------------------------------------------------------------ */
214 ipf_state_main_unload()
220 /* ------------------------------------------------------------------------ */
221 /* Function: ipf_state_soft_create */
222 /* Returns: void * - NULL = failure, else pointer to soft context */
223 /* Parameters: softc(I) - pointer to soft context main structure */
225 /* Create a new state soft context structure and populate it with the list */
226 /* of tunables and other default settings. */
227 /* ------------------------------------------------------------------------ */
229 ipf_state_soft_create(softc)
230 ipf_main_softc_t *softc;
232 ipf_state_softc_t *softs;
234 KMALLOC(softs, ipf_state_softc_t *);
238 bzero((char *)softs, sizeof(*softs));
240 softs->ipf_state_tune = ipf_tune_array_copy(softs,
241 sizeof(ipf_state_tuneables),
242 ipf_state_tuneables);
243 if (softs->ipf_state_tune == NULL) {
244 ipf_state_soft_destroy(softc, softs);
247 if (ipf_tune_array_link(softc, softs->ipf_state_tune) == -1) {
248 ipf_state_soft_destroy(softc, softs);
253 softs->ipf_state_logging = 1;
255 softs->ipf_state_logging = 0;
257 softs->ipf_state_size = IPSTATE_SIZE,
258 softs->ipf_state_maxbucket = 0;
259 softs->ipf_state_wm_freq = IPF_TTLVAL(10);
260 softs->ipf_state_max = IPSTATE_MAX;
261 softs->ipf_state_wm_last = 0;
262 softs->ipf_state_wm_high = 99;
263 softs->ipf_state_wm_low = 90;
264 softs->ipf_state_inited = 0;
265 softs->ipf_state_lock = 0;
266 softs->ipf_state_doflush = 0;
272 /* ------------------------------------------------------------------------ */
273 /* Function: ipf_state_soft_destroy */
275 /* Parameters: softc(I) - pointer to soft context main structure */
276 /* arg(I) - pointer to local context to use */
278 /* Undo only what we did in soft create: unlink and free the tunables and */
279 /* free the soft context structure itself. */
280 /* ------------------------------------------------------------------------ */
282 ipf_state_soft_destroy(softc, arg)
283 ipf_main_softc_t *softc;
286 ipf_state_softc_t *softs = arg;
288 if (softs->ipf_state_tune != NULL) {
289 ipf_tune_array_unlink(softc, softs->ipf_state_tune);
290 KFREES(softs->ipf_state_tune, sizeof(ipf_state_tuneables));
291 softs->ipf_state_tune = NULL;
298 ipf_state_seed_alloc(u_int state_size, u_int state_max)
302 KMALLOCS(state_seed, u_long *, state_size * sizeof(*state_seed));
303 if (state_seed == NULL)
306 for (i = 0; i < state_size; i++) {
308 * XXX - ipf_state_seed[X] should be a random number of sorts.
311 state_seed[i] = arc4random();
313 state_seed[i] = ((u_long)state_seed + i) * state_size;
314 state_seed[i] ^= 0xa5a55a5a;
315 state_seed[i] *= (u_long)state_seed;
316 state_seed[i] ^= 0x5a5aa5a5;
317 state_seed[i] *= state_max;
324 /* ------------------------------------------------------------------------ */
325 /* Function: ipf_state_soft_init */
326 /* Returns: int - 0 == success, -1 == failure */
327 /* Parameters: softc(I) - pointer to soft context main structure */
328 /* arg(I) - pointer to local context to use */
330 /* Initialise the state soft context structure so it is ready for use. */
332 /* - allocating a hash table and zero'ing it out */
333 /* - building a secondary table of seeds for double hashing to make it more */
334 /* difficult to attempt to attack the hash table itself (for DoS) */
335 /* - initialise all of the timeout queues, including a table for TCP, some */
336 /* pairs of query/response for UDP and other IP protocols (typically the */
337 /* reply queue has a shorter timeout than the query) */
338 /* ------------------------------------------------------------------------ */
340 ipf_state_soft_init(softc, arg)
341 ipf_main_softc_t *softc;
344 ipf_state_softc_t *softs = arg;
347 KMALLOCS(softs->ipf_state_table,
348 ipstate_t **, softs->ipf_state_size * sizeof(ipstate_t *));
349 if (softs->ipf_state_table == NULL)
352 bzero((char *)softs->ipf_state_table,
353 softs->ipf_state_size * sizeof(ipstate_t *));
355 softs->ipf_state_seed = ipf_state_seed_alloc(softs->ipf_state_size,
356 softs->ipf_state_max);
357 if (softs->ipf_state_seed == NULL)
360 KMALLOCS(softs->ipf_state_stats.iss_bucketlen, u_int *,
361 softs->ipf_state_size * sizeof(u_int));
362 if (softs->ipf_state_stats.iss_bucketlen == NULL)
365 bzero((char *)softs->ipf_state_stats.iss_bucketlen,
366 softs->ipf_state_size * sizeof(u_int));
368 if (softs->ipf_state_maxbucket == 0) {
369 for (i = softs->ipf_state_size; i > 0; i >>= 1)
370 softs->ipf_state_maxbucket++;
371 softs->ipf_state_maxbucket *= 2;
374 ipf_sttab_init(softc, softs->ipf_state_tcptq);
375 softs->ipf_state_stats.iss_tcptab = softs->ipf_state_tcptq;
376 softs->ipf_state_tcptq[IPF_TCP_NSTATES - 1].ifq_next =
377 &softs->ipf_state_udptq;
379 IPFTQ_INIT(&softs->ipf_state_udptq, softc->ipf_udptimeout,
381 softs->ipf_state_udptq.ifq_next = &softs->ipf_state_udpacktq;
383 IPFTQ_INIT(&softs->ipf_state_udpacktq, softc->ipf_udpacktimeout,
385 softs->ipf_state_udpacktq.ifq_next = &softs->ipf_state_icmptq;
387 IPFTQ_INIT(&softs->ipf_state_icmptq, softc->ipf_icmptimeout,
389 softs->ipf_state_icmptq.ifq_next = &softs->ipf_state_icmpacktq;
391 IPFTQ_INIT(&softs->ipf_state_icmpacktq, softc->ipf_icmpacktimeout,
392 "ipftq icmpack tab");
393 softs->ipf_state_icmpacktq.ifq_next = &softs->ipf_state_iptq;
395 IPFTQ_INIT(&softs->ipf_state_iptq, softc->ipf_iptimeout,
396 "ipftq iptimeout tab");
397 softs->ipf_state_iptq.ifq_next = &softs->ipf_state_pending;
399 IPFTQ_INIT(&softs->ipf_state_pending, IPF_HZ_DIVIDE, "ipftq pending");
400 softs->ipf_state_pending.ifq_next = &softs->ipf_state_deletetq;
402 IPFTQ_INIT(&softs->ipf_state_deletetq, 1, "ipftq delete");
403 softs->ipf_state_deletetq.ifq_next = NULL;
405 MUTEX_INIT(&softs->ipf_stinsert, "ipf state insert mutex");
408 softs->ipf_state_wm_last = softc->ipf_ticks;
409 softs->ipf_state_inited = 1;
415 /* ------------------------------------------------------------------------ */
416 /* Function: ipf_state_soft_fini */
417 /* Returns: int - 0 = success, -1 = failure */
418 /* Parameters: softc(I) - pointer to soft context main structure */
419 /* arg(I) - pointer to local context to use */
421 /* Release and destroy any resources acquired or initialised so that */
422 /* IPFilter can be unloaded or re-initialised. */
423 /* ------------------------------------------------------------------------ */
425 ipf_state_soft_fini(softc, arg)
426 ipf_main_softc_t *softc;
429 ipf_state_softc_t *softs = arg;
430 ipftq_t *ifq, *ifqnext;
433 while ((is = softs->ipf_state_list) != NULL)
434 ipf_state_del(softc, is, ISL_UNLOAD);
437 * Proxy timeout queues are not cleaned here because although they
438 * exist on the state list, appr_unload is called after
439 * ipf_state_unload and the proxies actually are responsible for them
440 * being created. Should the proxy timeouts have their own list?
441 * There's no real justification as this is the only complication.
443 for (ifq = softs->ipf_state_usertq; ifq != NULL; ifq = ifqnext) {
444 ifqnext = ifq->ifq_next;
446 if (ipf_deletetimeoutqueue(ifq) == 0)
447 ipf_freetimeoutqueue(softc, ifq);
450 softs->ipf_state_stats.iss_inuse = 0;
451 softs->ipf_state_stats.iss_active = 0;
453 if (softs->ipf_state_inited == 1) {
454 softs->ipf_state_inited = 0;
455 ipf_sttab_destroy(softs->ipf_state_tcptq);
456 MUTEX_DESTROY(&softs->ipf_state_udptq.ifq_lock);
457 MUTEX_DESTROY(&softs->ipf_state_icmptq.ifq_lock);
458 MUTEX_DESTROY(&softs->ipf_state_udpacktq.ifq_lock);
459 MUTEX_DESTROY(&softs->ipf_state_icmpacktq.ifq_lock);
460 MUTEX_DESTROY(&softs->ipf_state_iptq.ifq_lock);
461 MUTEX_DESTROY(&softs->ipf_state_deletetq.ifq_lock);
462 MUTEX_DESTROY(&softs->ipf_state_pending.ifq_lock);
463 MUTEX_DESTROY(&softs->ipf_stinsert);
466 if (softs->ipf_state_table != NULL) {
467 KFREES(softs->ipf_state_table,
468 softs->ipf_state_size * sizeof(*softs->ipf_state_table));
469 softs->ipf_state_table = NULL;
472 if (softs->ipf_state_seed != NULL) {
473 KFREES(softs->ipf_state_seed,
474 softs->ipf_state_size * sizeof(*softs->ipf_state_seed));
475 softs->ipf_state_seed = NULL;
478 if (softs->ipf_state_stats.iss_bucketlen != NULL) {
479 KFREES(softs->ipf_state_stats.iss_bucketlen,
480 softs->ipf_state_size * sizeof(u_int));
481 softs->ipf_state_stats.iss_bucketlen = NULL;
488 /* ------------------------------------------------------------------------ */
489 /* Function: ipf_state_setlock */
491 /* Parameters: arg(I) - pointer to local context to use */
492 /* tmp(I) - new value for lock */
494 /* Stub function that allows for external manipulation of ipf_state_lock */
495 /* ------------------------------------------------------------------------ */
497 ipf_state_setlock(arg, tmp)
501 ipf_state_softc_t *softs = arg;
503 softs->ipf_state_lock = tmp;
507 /* ------------------------------------------------------------------------ */
508 /* Function: ipf_state_stats */
509 /* Returns: ips_state_t* - pointer to state stats structure */
510 /* Parameters: softc(I) - pointer to soft context main structure */
512 /* Put all the current numbers and pointers into a single struct and return */
513 /* a pointer to it. */
514 /* ------------------------------------------------------------------------ */
516 ipf_state_stats(softc)
517 ipf_main_softc_t *softc;
519 ipf_state_softc_t *softs = softc->ipf_state_soft;
520 ips_stat_t *issp = &softs->ipf_state_stats;
522 issp->iss_state_size = softs->ipf_state_size;
523 issp->iss_state_max = softs->ipf_state_max;
524 issp->iss_table = softs->ipf_state_table;
525 issp->iss_list = softs->ipf_state_list;
526 issp->iss_ticks = softc->ipf_ticks;
528 #ifdef IPFILTER_LOGGING
529 issp->iss_log_ok = ipf_log_logok(softc, IPF_LOGSTATE);
530 issp->iss_log_fail = ipf_log_failures(softc, IPF_LOGSTATE);
532 issp->iss_log_ok = 0;
533 issp->iss_log_fail = 0;
538 /* ------------------------------------------------------------------------ */
539 /* Function: ipf_state_remove */
540 /* Returns: int - 0 == success, != 0 == failure */
541 /* Parameters: softc(I) - pointer to soft context main structure */
542 /* data(I) - pointer to state structure to delete from table */
544 /* Search for a state structure that matches the one passed, according to */
545 /* the IP addresses and other protocol specific information. */
546 /* ------------------------------------------------------------------------ */
548 ipf_state_remove(softc, data)
549 ipf_main_softc_t *softc;
552 ipf_state_softc_t *softs = softc->ipf_state_soft;
557 error = ipf_inobj(softc, data, NULL, &st, IPFOBJ_IPSTATE);
561 WRITE_ENTER(&softc->ipf_state);
562 for (sp = softs->ipf_state_list; sp; sp = sp->is_next)
563 if ((sp->is_p == st.is_p) && (sp->is_v == st.is_v) &&
564 !bcmp((caddr_t)&sp->is_src, (caddr_t)&st.is_src,
565 sizeof(st.is_src)) &&
566 !bcmp((caddr_t)&sp->is_dst, (caddr_t)&st.is_dst,
567 sizeof(st.is_dst)) &&
568 !bcmp((caddr_t)&sp->is_ps, (caddr_t)&st.is_ps,
570 ipf_state_del(softc, sp, ISL_REMOVE);
571 RWLOCK_EXIT(&softc->ipf_state);
574 RWLOCK_EXIT(&softc->ipf_state);
581 /* ------------------------------------------------------------------------ */
582 /* Function: ipf_state_ioctl */
583 /* Returns: int - 0 == success, != 0 == failure */
584 /* Parameters: softc(I) - pointer to soft context main structure */
585 /* data(I) - pointer to ioctl data */
586 /* cmd(I) - ioctl command integer */
587 /* mode(I) - file mode bits used with open */
588 /* uid(I) - uid of process making the ioctl call */
589 /* ctx(I) - pointer specific to context of the call */
591 /* Processes an ioctl call made to operate on the IP Filter state device. */
592 /* ------------------------------------------------------------------------ */
594 ipf_state_ioctl(softc, data, cmd, mode, uid, ctx)
595 ipf_main_softc_t *softc;
601 ipf_state_softc_t *softs = softc->ipf_state_soft;
602 int arg, ret, error = 0;
608 * Delete an entry from the state table.
611 error = ipf_state_remove(softc, data);
615 * Flush the state table
618 error = BCOPYIN(data, &arg, sizeof(arg));
624 WRITE_ENTER(&softc->ipf_state);
625 ret = ipf_state_flush(softc, arg, 4);
626 RWLOCK_EXIT(&softc->ipf_state);
628 error = BCOPYOUT(&ret, data, sizeof(ret));
638 error = BCOPYIN(data, &arg, sizeof(arg));
644 WRITE_ENTER(&softc->ipf_state);
645 ret = ipf_state_flush(softc, arg, 6);
646 RWLOCK_EXIT(&softc->ipf_state);
648 error = BCOPYOUT(&ret, data, sizeof(ret));
657 case SIOCMATCHFLUSH :
658 WRITE_ENTER(&softc->ipf_state);
659 error = ipf_state_matchflush(softc, data);
660 RWLOCK_EXIT(&softc->ipf_state);
665 * Flush the state log.
668 if (!(mode & FWRITE)) {
674 tmp = ipf_log_clear(softc, IPL_LOGSTATE);
675 error = BCOPYOUT(&tmp, data, sizeof(tmp));
684 * Turn logging of state information on/off.
687 if (!(mode & FWRITE)) {
691 error = BCOPYIN(data, &softs->ipf_state_logging,
692 sizeof(softs->ipf_state_logging));
701 * Return the current state of logging.
704 error = BCOPYOUT(&softs->ipf_state_logging, data,
705 sizeof(softs->ipf_state_logging));
713 * Return the number of bytes currently waiting to be read.
716 arg = ipf_log_bytesused(softc, IPL_LOGSTATE);
717 error = BCOPYOUT(&arg, data, sizeof(arg));
726 * Get the current state statistics.
729 error = ipf_outobj(softc, data, ipf_state_stats(softc),
734 * Lock/Unlock the state table. (Locking prevents any changes, which
735 * means no packets match).
738 if (!(mode & FWRITE)) {
742 error = ipf_lock(data, &softs->ipf_state_lock);
747 * Add an entry to the current state table.
750 if (!softs->ipf_state_lock || !(mode &FWRITE)) {
755 error = ipf_state_putent(softc, softs, data);
759 * Get a state table entry.
762 if (!softs->ipf_state_lock) {
767 error = ipf_state_getent(softc, softs, data);
771 * Return a copy of the hash table bucket lengths
774 error = BCOPYOUT(softs->ipf_state_stats.iss_bucketlen, data,
775 softs->ipf_state_size * sizeof(u_int));
788 error = ipf_inobj(softc, data, &obj, &iter, IPFOBJ_GENITER);
793 token = ipf_token_find(softc, IPFGENITER_STATE, uid, ctx);
795 error = ipf_state_iter(softc, token, &iter, &obj);
796 WRITE_ENTER(&softc->ipf_tokens);
797 ipf_token_deref(softc, token);
798 RWLOCK_EXIT(&softc->ipf_tokens);
808 error = ipf_state_gettable(softc, softs, data);
812 error = BCOPYIN(data, &arg, sizeof(arg));
818 error = ipf_token_del(softc, arg, uid, ctx);
824 error = ipf_outobj(softc, data, softs->ipf_state_tcptq,
837 /* ------------------------------------------------------------------------ */
838 /* Function: ipf_state_getent */
839 /* Returns: int - 0 == success, != 0 == failure */
840 /* Parameters: softc(I) - pointer to soft context main structure */
841 /* softs(I) - pointer to state context structure */
842 /* data(I) - pointer to state structure to retrieve from table*/
844 /* Copy out state information from the kernel to a user space process. If */
845 /* there is a filter rule associated with the state entry, copy that out */
846 /* as well. The entry to copy out is taken from the value of "ips_next" in */
847 /* the struct passed in and if not null and not found in the list of current*/
848 /* state entries, the retrieval fails. */
849 /* ------------------------------------------------------------------------ */
851 ipf_state_getent(softc, softs, data)
852 ipf_main_softc_t *softc;
853 ipf_state_softc_t *softs;
860 error = ipf_inobj(softc, data, NULL, &ips, IPFOBJ_STATESAVE);
864 READ_ENTER(&softc->ipf_state);
867 isn = softs->ipf_state_list;
869 if (ips.ips_next == NULL) {
870 RWLOCK_EXIT(&softc->ipf_state);
878 * Make sure the pointer we're copying from exists in the
879 * current list of entries. Security precaution to prevent
880 * copying of random kernel data.
882 for (is = softs->ipf_state_list; is; is = is->is_next)
886 RWLOCK_EXIT(&softc->ipf_state);
891 ips.ips_next = isn->is_next;
892 bcopy((char *)isn, (char *)&ips.ips_is, sizeof(ips.ips_is));
893 ips.ips_rule = isn->is_rule;
894 if (isn->is_rule != NULL)
895 bcopy((char *)isn->is_rule, (char *)&ips.ips_fr,
897 RWLOCK_EXIT(&softc->ipf_state);
898 error = ipf_outobj(softc, data, &ips, IPFOBJ_STATESAVE);
903 /* ------------------------------------------------------------------------ */
904 /* Function: ipf_state_putent */
905 /* Returns: int - 0 == success, != 0 == failure */
906 /* Parameters: softc(I) - pointer to soft context main structure */
907 /* softs(I) - pointer to state context structure */
908 /* data(I) - pointer to state information struct */
910 /* This function implements the SIOCSTPUT ioctl: insert a state entry into */
911 /* the state table. If the state info. includes a pointer to a filter rule */
912 /* then also add in an orphaned rule (will not show up in any "ipfstat -io" */
914 /* ------------------------------------------------------------------------ */
916 ipf_state_putent(softc, softs, data)
917 ipf_main_softc_t *softc;
918 ipf_state_softc_t *softs;
927 error = ipf_inobj(softc, data, NULL, &ips, IPFOBJ_STATESAVE);
931 KMALLOC(isn, ipstate_t *);
937 bcopy((char *)&ips.ips_is, (char *)isn, sizeof(*isn));
938 bzero((char *)isn, offsetof(struct ipstate, is_pkts));
939 isn->is_sti.tqe_pnext = NULL;
940 isn->is_sti.tqe_next = NULL;
941 isn->is_sti.tqe_ifq = NULL;
942 isn->is_sti.tqe_parent = isn;
943 isn->is_ifp[0] = NULL;
944 isn->is_ifp[1] = NULL;
945 isn->is_ifp[2] = NULL;
946 isn->is_ifp[3] = NULL;
953 READ_ENTER(&softc->ipf_state);
954 inserr = ipf_state_insert(softc, isn, 0);
955 MUTEX_EXIT(&isn->is_lock);
956 RWLOCK_EXIT(&softc->ipf_state);
961 if (isn->is_flags & SI_NEWFR) {
962 KMALLOC(fr, frentry_t *);
968 bcopy((char *)&ips.ips_fr, (char *)fr, sizeof(*fr));
969 out = fr->fr_flags & FR_OUTQUE ? 1 : 0;
971 ips.ips_is.is_rule = fr;
972 MUTEX_NUKE(&fr->fr_lock);
973 MUTEX_INIT(&fr->fr_lock, "state filter rule lock");
976 * Look up all the interface names in the rule.
978 for (i = 0; i < FR_NUM(fr->fr_ifnames); i++) {
979 if (fr->fr_ifnames[i] == -1) {
980 fr->fr_ifas[i] = NULL;
983 name = FR_NAME(fr, fr_ifnames[i]);
984 fr->fr_ifas[i] = ipf_resolvenic(softc, name,
988 for (i = 0; i < FR_NUM(isn->is_ifname); i++) {
989 name = isn->is_ifname[i];
990 isn->is_ifp[i] = ipf_resolvenic(softc, name,
997 fr->fr_type = FR_T_NONE;
999 (void) ipf_resolvedest(softc, fr->fr_names, &fr->fr_tifs[0],
1001 (void) ipf_resolvedest(softc, fr->fr_names, &fr->fr_tifs[1],
1003 (void) ipf_resolvedest(softc, fr->fr_names, &fr->fr_dif,
1007 * send a copy back to userland of what we ended up
1008 * to allow for verification.
1010 error = ipf_outobj(softc, data, &ips, IPFOBJ_STATESAVE);
1013 MUTEX_DESTROY(&fr->fr_lock);
1018 READ_ENTER(&softc->ipf_state);
1019 error = ipf_state_insert(softc, isn, 0);
1020 MUTEX_EXIT(&isn->is_lock);
1021 RWLOCK_EXIT(&softc->ipf_state);
1024 READ_ENTER(&softc->ipf_state);
1025 for (is = softs->ipf_state_list; is; is = is->is_next)
1026 if (is->is_rule == fr) {
1027 error = ipf_state_insert(softc, isn, 0);
1028 MUTEX_EXIT(&isn->is_lock);
1036 RWLOCK_EXIT(&softc->ipf_state);
1048 /* ------------------------------------------------------------------------ */
1049 /* Function: ipf_state_insert */
1050 /* Returns: int - 0 == success, -1 == failure */
1051 /* Parameters: softc(I) - pointer to soft context main structure */
1052 /* Parameters: is(I) - pointer to state structure */
1053 /* rev(I) - flag indicating direction of packet */
1055 /* Inserts a state structure into the hash table (for lookups) and the list */
1056 /* of state entries (for enumeration). Resolves all of the interface names */
1057 /* to pointers and adjusts running stats for the hash table as appropriate. */
1059 /* This function can fail if the filter rule has had a population policy of */
1060 /* IP addresses used with stateful filtering assigned to it. */
1062 /* Locking: it is assumed that some kind of lock on ipf_state is held. */
1063 /* Exits with is_lock initialised and held - *EVEN IF ERROR*. */
1064 /* ------------------------------------------------------------------------ */
1066 ipf_state_insert(softc, is, rev)
1067 ipf_main_softc_t *softc;
1071 ipf_state_softc_t *softs = softc->ipf_state_soft;
1077 * Look up all the interface names in the state entry.
1079 for (i = 0; i < FR_NUM(is->is_ifp); i++) {
1080 if (is->is_ifp[i] != NULL)
1082 is->is_ifp[i] = ipf_resolvenic(softc, is->is_ifname[i],
1087 * If we could trust is_hv, then the modulus would not be needed,
1088 * but when running with IPFILTER_SYNC, this stops bad values.
1090 hv = is->is_hv % softs->ipf_state_size;
1095 * We need to get both of these locks...the first because it is
1096 * possible that once the insert is complete another packet might
1097 * come along, match the entry and want to update it.
1099 MUTEX_INIT(&is->is_lock, "ipf state entry");
1100 MUTEX_ENTER(&is->is_lock);
1101 MUTEX_ENTER(&softs->ipf_stinsert);
1105 if ((fr->fr_srctrack.ht_max_nodes != 0) &&
1106 (ipf_ht_node_add(softc, &fr->fr_srctrack,
1107 is->is_family, &is->is_src) == -1)) {
1108 SBUMPD(ipf_state_stats, iss_max_track);
1109 MUTEX_EXIT(&softs->ipf_stinsert);
1113 MUTEX_ENTER(&fr->fr_lock);
1115 MUTEX_EXIT(&fr->fr_lock);
1119 if (is->is_flags & (SI_WILDP|SI_WILDA)) {
1120 DT(iss_wild_plus_one);
1121 SINCL(ipf_state_stats.iss_wild);
1124 SBUMP(ipf_state_stats.iss_proto[is->is_p]);
1125 SBUMP(ipf_state_stats.iss_active_proto[is->is_p]);
1128 * add into list table.
1130 if (softs->ipf_state_list != NULL)
1131 softs->ipf_state_list->is_pnext = &is->is_next;
1132 is->is_pnext = &softs->ipf_state_list;
1133 is->is_next = softs->ipf_state_list;
1134 softs->ipf_state_list = is;
1136 if (softs->ipf_state_table[hv] != NULL)
1137 softs->ipf_state_table[hv]->is_phnext = &is->is_hnext;
1139 softs->ipf_state_stats.iss_inuse++;
1140 is->is_phnext = softs->ipf_state_table + hv;
1141 is->is_hnext = softs->ipf_state_table[hv];
1142 softs->ipf_state_table[hv] = is;
1143 softs->ipf_state_stats.iss_bucketlen[hv]++;
1144 softs->ipf_state_stats.iss_active++;
1145 MUTEX_EXIT(&softs->ipf_stinsert);
1147 ipf_state_setqueue(softc, is, rev);
1153 /* ------------------------------------------------------------------------ */
1154 /* Function: ipf_state_matchipv4addrs */
1155 /* Returns: int - 2 addresses match (strong match), 1 reverse match, */
1157 /* Parameters: is1, is2 pointers to states we are checking */
1159 /* Function matches IPv4 addresses it returns strong match for ICMP proto */
1160 /* even there is only reverse match */
1161 /* ------------------------------------------------------------------------ */
1163 ipf_state_matchipv4addrs(is1, is2)
1164 ipstate_t *is1, *is2;
1168 if (is1->is_saddr == is2->is_saddr && is1->is_daddr == is2->is_daddr)
1170 else if (is1->is_saddr == is2->is_daddr &&
1171 is1->is_daddr == is2->is_saddr) {
1172 /* force strong match for ICMP protocol */
1173 rv = (is1->is_p == IPPROTO_ICMP) ? 2 : 1;
1182 /* ------------------------------------------------------------------------ */
1183 /* Function: ipf_state_matchipv6addrs */
1184 /* Returns: int - 2 addresses match (strong match), 1 reverse match, */
1186 /* Parameters: is1, is2 pointers to states we are checking */
1188 /* Function matches IPv6 addresses it returns strong match for ICMP proto */
1189 /* even there is only reverse match */
1190 /* ------------------------------------------------------------------------ */
1192 ipf_state_matchipv6addrs(is1, is2)
1193 ipstate_t *is1, *is2;
1197 if (IP6_EQ(&is1->is_src, &is2->is_src) &&
1198 IP6_EQ(&is1->is_dst, &is2->is_dst))
1200 else if (IP6_EQ(&is1->is_src, &is2->is_dst) &&
1201 IP6_EQ(&is1->is_dst, &is2->is_src)) {
1202 /* force strong match for ICMPv6 protocol */
1203 rv = (is1->is_p == IPPROTO_ICMPV6) ? 2 : 1;
1212 /* ------------------------------------------------------------------------ */
1213 /* Function: ipf_state_matchaddresses */
1214 /* Returns: int - 2 addresses match, 1 reverse match, zero no match */
1215 /* Parameters: is1, is2 pointers to states we are checking */
1217 /* function retruns true if two pairs of addresses belong to single */
1218 /* connection. suppose there are two endpoints: */
1219 /* endpoint1 1.1.1.1 */
1220 /* endpoint2 1.1.1.2 */
1222 /* the state is established by packet flying from .1 to .2 so we see: */
1223 /* is1->src = 1.1.1.1 */
1224 /* is1->dst = 1.1.1.2 */
1225 /* now endpoint 1.1.1.2 sends answer */
1226 /* retreives is1 record created by first packat and compares it with is2 */
1227 /* temporal record, is2 is initialized as follows: */
1228 /* is2->src = 1.1.1.2 */
1229 /* is2->dst = 1.1.1.1 */
1230 /* in this case 1 will be returned */
1232 /* the ipf_matchaddresses() assumes those two records to be same. of course */
1233 /* the ipf_matchaddresses() also assume records are same in case you pass */
1234 /* identical arguments (i.e. ipf_matchaddress(is1, is1) would return 2 */
1235 /* ------------------------------------------------------------------------ */
1237 ipf_state_matchaddresses(is1, is2)
1238 ipstate_t *is1, *is2;
1242 if (is1->is_v == 4) {
1243 rv = ipf_state_matchipv4addrs(is1, is2);
1246 rv = ipf_state_matchipv6addrs(is1, is2);
1253 /* ------------------------------------------------------------------------ */
1254 /* Function: ipf_matchports */
1255 /* Returns: int - 2 match, 1 rverse match, 0 no match */
1256 /* Parameters: ppairs1, ppairs - src, dst ports we want to match */
1258 /* performs the same match for isps members as for addresses */
1259 /* ------------------------------------------------------------------------ */
1261 ipf_state_matchports(ppairs1, ppairs2)
1262 udpinfo_t *ppairs1, *ppairs2;
1266 if (ppairs1->us_sport == ppairs2->us_sport &&
1267 ppairs1->us_dport == ppairs2->us_dport)
1269 else if (ppairs1->us_sport == ppairs2->us_dport &&
1270 ppairs1->us_dport == ppairs2->us_sport)
1279 /* ------------------------------------------------------------------------ */
1280 /* Function: ipf_matchisps */
1281 /* Returns: int - nonzero if isps members match, 0 nomatch */
1282 /* Parameters: is1, is2 - states we want to match */
1284 /* performs the same match for isps members as for addresses */
1285 /* ------------------------------------------------------------------------ */
1287 ipf_state_matchisps(is1, is2)
1288 ipstate_t *is1, *is2;
1292 if (is1->is_p == is2->is_p) {
1298 /* greinfo_t can be also interprted as port pair */
1299 rv = ipf_state_matchports(&is1->is_ps.is_us,
1304 case IPPROTO_ICMPV6 :
1305 /* force strong match for ICMP datagram. */
1306 if (bcmp(&is1->is_ps, &is2->is_ps,
1307 sizeof(icmpinfo_t)) == 0) {
1325 /* ------------------------------------------------------------------------ */
1326 /* Function: ipf_state_match */
1327 /* Returns: int - nonzero match, zero no match */
1328 /* Parameters: is1, is2 - states we want to match */
1330 /* ------------------------------------------------------------------------ */
1332 ipf_state_match(is1, is2)
1333 ipstate_t *is1, *is2;
1339 if (bcmp(&is1->is_pass, &is2->is_pass,
1340 offsetof(struct ipstate, is_authmsk) -
1341 offsetof(struct ipstate, is_pass)) == 0) {
1343 pomatch = ipf_state_matchisps(is1, is2);
1344 amatch = ipf_state_matchaddresses(is1, is2);
1345 rv = (amatch != 0) && (amatch == pomatch);
1353 /* ------------------------------------------------------------------------ */
1354 /* Function: ipf_state_add */
1355 /* Returns: ipstate_t - 0 = success */
1356 /* Parameters: softc(I) - pointer to soft context main structure */
1357 /* fin(I) - pointer to packet information */
1358 /* stsave(O) - pointer to place to save pointer to created */
1359 /* state structure. */
1360 /* flags(I) - flags to use when creating the structure */
1362 /* Creates a new IP state structure from the packet information collected. */
1363 /* Inserts it into the state table and appends to the bottom of the active */
1364 /* list. If the capacity of the table has reached the maximum allowed then */
1365 /* the call will fail and a flush is scheduled for the next timeout call. */
1367 /* NOTE: The use of stsave to point to nat_state will result in memory */
1368 /* corruption. It should only be used to point to objects that will */
1369 /* either outlive this (not expired) or will deref the ip_state_t */
1370 /* when they are deleted. */
1371 /* ------------------------------------------------------------------------ */
1373 ipf_state_add(softc, fin, stsave, flags)
1374 ipf_main_softc_t *softc;
1379 ipf_state_softc_t *softs = softc->ipf_state_soft;
1389 * If a locally created packet is trying to egress but it
1390 * does not match because of this lock, it is likely that
1391 * the policy will block it and return network unreachable further
1392 * up the stack. To mitigate this error, EAGAIN is returned instead,
1393 * telling the IP stack to try sending this packet again later.
1395 if (softs->ipf_state_lock) {
1396 SBUMPD(ipf_state_stats, iss_add_locked);
1397 fin->fin_error = EAGAIN;
1401 if (fin->fin_flx & (FI_SHORT|FI_STATE|FI_FRAGBODY|FI_BAD)) {
1402 SBUMPD(ipf_state_stats, iss_add_bad);
1406 if ((fin->fin_flx & FI_OOW) && !(fin->fin_tcpf & TH_SYN)) {
1407 SBUMPD(ipf_state_stats, iss_add_oow);
1411 if ((softs->ipf_state_stats.iss_active * 100 / softs->ipf_state_max) >
1412 softs->ipf_state_wm_high) {
1413 softs->ipf_state_doflush = 1;
1417 * If a "keep state" rule has reached the maximum number of references
1418 * to it, then schedule an automatic flush in case we can clear out
1419 * some "dead old wood". Note that because the lock isn't held on
1420 * fr it is possible that we could overflow. The cost of overflowing
1421 * is being ignored here as the number by which it can overflow is
1422 * a product of the number of simultaneous threads that could be
1423 * executing in here, so a limit of 100 won't result in 200, but could
1424 * result in 101 or 102.
1428 if ((softs->ipf_state_stats.iss_active >=
1429 softs->ipf_state_max) && (fr->fr_statemax == 0)) {
1430 SBUMPD(ipf_state_stats, iss_max);
1433 if ((fr->fr_statemax != 0) &&
1434 (fr->fr_statecnt >= fr->fr_statemax)) {
1435 SBUMPD(ipf_state_stats, iss_max_ref);
1442 pass = softc->ipf_flags;
1443 is->is_tag = FR_NOLOGTAG;
1445 pass = fr->fr_flags;
1451 bzero((char *)is, sizeof(*is));
1452 is->is_die = 1 + softc->ipf_ticks;
1454 * We want to check everything that is a property of this packet,
1455 * but we don't (automatically) care about its fragment status as
1459 is->is_v = fin->fin_v;
1460 is->is_sec = fin->fin_secmsk;
1461 is->is_secmsk = 0xffff;
1462 is->is_auth = fin->fin_auth;
1463 is->is_authmsk = 0xffff;
1464 is->is_family = fin->fin_family;
1465 is->is_opt[0] = fin->fin_optmsk;
1466 is->is_optmsk[0] = 0xffffffff;
1467 if (is->is_v == 6) {
1468 is->is_opt[0] &= ~0x8;
1469 is->is_optmsk[0] &= ~0x8;
1473 * Copy and calculate...
1475 hv = (is->is_p = fin->fin_fi.fi_p);
1476 is->is_src = fin->fin_fi.fi_src;
1478 is->is_dst = fin->fin_fi.fi_dst;
1481 if (fin->fin_v == 6) {
1483 * For ICMPv6, we check to see if the destination address is
1484 * a multicast address. If it is, do not include it in the
1485 * calculation of the hash because the correct reply will come
1486 * back from a real address, not a multicast address.
1488 if ((is->is_p == IPPROTO_ICMPV6) &&
1489 IN6_IS_ADDR_MULTICAST(&is->is_dst.in6)) {
1491 * So you can do keep state with neighbour discovery.
1493 * Here we could use the address from the neighbour
1494 * solicit message to put in the state structure and
1495 * we could use that without a wildcard flag too...
1497 flags |= SI_W_DADDR;
1500 hv += is->is_dst.i6[1];
1501 hv += is->is_dst.i6[2];
1502 hv += is->is_dst.i6[3];
1504 hv += is->is_src.i6[1];
1505 hv += is->is_src.i6[2];
1506 hv += is->is_src.i6[3];
1509 if ((fin->fin_v == 4) &&
1510 (fin->fin_flx & (FI_MULTICAST|FI_BROADCAST|FI_MBCAST))) {
1511 flags |= SI_W_DADDR;
1518 case IPPROTO_ICMPV6 :
1521 switch (ic->icmp_type)
1523 case ICMP6_ECHO_REQUEST :
1524 hv += (is->is_icmp.ici_id = ic->icmp_id);
1526 case ICMP6_MEMBERSHIP_QUERY :
1527 case ND_ROUTER_SOLICIT :
1528 case ND_NEIGHBOR_SOLICIT :
1529 case ICMP6_NI_QUERY :
1530 is->is_icmp.ici_type = ic->icmp_type;
1533 SBUMPD(ipf_state_stats, iss_icmp6_notquery);
1541 switch (ic->icmp_type)
1547 is->is_icmp.ici_type = ic->icmp_type;
1548 hv += (is->is_icmp.ici_id = ic->icmp_id);
1551 SBUMPD(ipf_state_stats, iss_icmp_notquery);
1560 is->is_gre.gs_flags = gre->gr_flags;
1561 is->is_gre.gs_ptype = gre->gr_ptype;
1562 if (GRE_REV(is->is_gre.gs_flags) == 1) {
1563 is->is_call[0] = fin->fin_data[0];
1564 is->is_call[1] = fin->fin_data[1];
1572 if (tcp->th_flags & TH_RST) {
1573 SBUMPD(ipf_state_stats, iss_tcp_rstadd);
1577 /* TRACE is, flags, hv */
1580 * The endian of the ports doesn't matter, but the ack and
1581 * sequence numbers do as we do mathematics on them later.
1583 is->is_sport = htons(fin->fin_data[0]);
1584 is->is_dport = htons(fin->fin_data[1]);
1585 if ((flags & (SI_W_DPORT|SI_W_SPORT)) == 0) {
1590 /* TRACE is, flags, hv */
1593 * If this is a real packet then initialise fields in the
1594 * state information structure from the TCP header information.
1598 is->is_maxswin = ntohs(tcp->th_win);
1599 if (is->is_maxswin == 0)
1602 if ((fin->fin_flx & FI_IGNORE) == 0) {
1603 is->is_send = ntohl(tcp->th_seq) + fin->fin_dlen -
1604 (TCP_OFF(tcp) << 2) +
1605 ((tcp->th_flags & TH_SYN) ? 1 : 0) +
1606 ((tcp->th_flags & TH_FIN) ? 1 : 0);
1607 is->is_maxsend = is->is_send;
1610 * Window scale option is only present in
1611 * SYN/SYN-ACK packet.
1613 if ((tcp->th_flags & ~(TH_FIN|TH_ACK|TH_ECNALL)) ==
1615 (TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) {
1616 if (ipf_tcpoptions(softs, fin, tcp,
1617 &is->is_tcp.ts_data[0]) == -1) {
1618 fin->fin_flx |= FI_BAD;
1619 DT1(ipf_fi_bad_tcpoptions_th_fin_ack_ecnall, fr_info_t *, fin);
1623 if ((fin->fin_out != 0) && (pass & FR_NEWISN) != 0) {
1624 ipf_checknewisn(fin, is);
1625 ipf_fixoutisn(fin, is);
1628 if ((tcp->th_flags & TH_OPENING) == TH_SYN)
1631 is->is_maxdwin = is->is_maxswin * 2;
1632 is->is_dend = ntohl(tcp->th_ack);
1633 is->is_maxdend = ntohl(tcp->th_ack);
1634 is->is_maxdwin *= 2;
1639 * If we're creating state for a starting connection, start
1640 * the timer on it as we'll never see an error if it fails
1648 is->is_sport = htons(fin->fin_data[0]);
1649 is->is_dport = htons(fin->fin_data[1]);
1650 if ((flags & (SI_W_DPORT|SI_W_SPORT)) == 0) {
1651 hv += tcp->th_dport;
1652 hv += tcp->th_sport;
1659 hv = DOUBLE_HASH(hv);
1663 * Look for identical state.
1665 for (is = softs->ipf_state_table[hv % softs->ipf_state_size];
1666 is != NULL; is = is->is_hnext) {
1667 if (ipf_state_match(&ips, is) == 1)
1671 SBUMPD(ipf_state_stats, iss_add_dup);
1675 if (softs->ipf_state_stats.iss_bucketlen[hv] >=
1676 softs->ipf_state_maxbucket) {
1677 SBUMPD(ipf_state_stats, iss_bucket_full);
1682 * No existing state; create new
1684 KMALLOC(is, ipstate_t *);
1686 SBUMPD(ipf_state_stats, iss_nomem);
1689 bcopy((char *)&ips, (char *)is, sizeof(*is));
1690 is->is_flags = flags & IS_INHERITED;
1691 is->is_rulen = fin->fin_rule;
1695 * Do not do the modulus here, it is done in ipf_state_insert().
1700 (void) strncpy(is->is_group, FR_NAME(fr, fr_group),
1702 if (fr->fr_age[0] != 0) {
1703 tq = ipf_addtimeoutqueue(softc,
1704 &softs->ipf_state_usertq,
1706 is->is_tqehead[0] = tq;
1707 is->is_sti.tqe_flags |= TQE_RULEBASED;
1709 if (fr->fr_age[1] != 0) {
1710 tq = ipf_addtimeoutqueue(softc,
1711 &softs->ipf_state_usertq,
1713 is->is_tqehead[1] = tq;
1714 is->is_sti.tqe_flags |= TQE_RULEBASED;
1717 is->is_tag = fr->fr_logtag;
1721 * It may seem strange to set is_ref to 2, but if stsave is not NULL
1722 * then a copy of the pointer is being stored somewhere else and in
1723 * the end, it will expect to be able to do something with it.
1726 if (stsave != NULL) {
1732 is->is_pkts[0] = 0, is->is_bytes[0] = 0;
1733 is->is_pkts[1] = 0, is->is_bytes[1] = 0;
1734 is->is_pkts[2] = 0, is->is_bytes[2] = 0;
1735 is->is_pkts[3] = 0, is->is_bytes[3] = 0;
1736 if ((fin->fin_flx & FI_IGNORE) == 0) {
1737 is->is_pkts[out] = 1;
1738 fin->fin_pktnum = 1;
1739 is->is_bytes[out] = fin->fin_plen;
1740 is->is_flx[out][0] = fin->fin_flx & FI_CMP;
1741 is->is_flx[out][0] &= ~FI_OOW;
1744 if (pass & FR_STLOOSE)
1745 is->is_flags |= IS_LOOSE;
1747 if (pass & FR_STSTRICT)
1748 is->is_flags |= IS_STRICT;
1750 if (pass & FR_STATESYNC)
1751 is->is_flags |= IS_STATESYNC;
1753 if (pass & FR_LOGFIRST)
1754 is->is_pass &= ~(FR_LOGFIRST|FR_LOG);
1756 READ_ENTER(&softc->ipf_state);
1758 if (ipf_state_insert(softc, is, fin->fin_rev) == -1) {
1759 RWLOCK_EXIT(&softc->ipf_state);
1761 * This is a bit more manual than it should be but
1762 * ipf_state_del cannot be called.
1764 MUTEX_EXIT(&is->is_lock);
1765 MUTEX_DESTROY(&is->is_lock);
1766 if (is->is_tqehead[0] != NULL) {
1767 if (ipf_deletetimeoutqueue(is->is_tqehead[0]) == 0)
1768 ipf_freetimeoutqueue(softc, is->is_tqehead[0]);
1769 is->is_tqehead[0] = NULL;
1771 if (is->is_tqehead[1] != NULL) {
1772 if (ipf_deletetimeoutqueue(is->is_tqehead[1]) == 0)
1773 ipf_freetimeoutqueue(softc, is->is_tqehead[1]);
1774 is->is_tqehead[1] = NULL;
1781 * Filling in the interface name is after the insert so that an
1782 * event (such as add/delete) of an interface that is referenced
1783 * by this rule will see this state entry.
1787 * The name '-' is special for network interfaces and causes
1788 * a NULL name to be present, always, allowing packets to
1789 * match it, regardless of their interface.
1791 if ((fin->fin_ifp == NULL) ||
1792 (fr->fr_ifnames[out << 1] != -1 &&
1793 fr->fr_names[fr->fr_ifnames[out << 1] + 0] == '-' &&
1794 fr->fr_names[fr->fr_ifnames[out << 1] + 1] == '\0')) {
1795 is->is_ifp[out << 1] = fr->fr_ifas[0];
1796 strncpy(is->is_ifname[out << 1],
1797 FR_NAME(fr, fr_ifnames[0]),
1798 sizeof(fr->fr_ifnames[0]));
1800 is->is_ifp[out << 1] = fin->fin_ifp;
1801 COPYIFNAME(fin->fin_v, fin->fin_ifp,
1802 is->is_ifname[out << 1]);
1805 is->is_ifp[(out << 1) + 1] = fr->fr_ifas[1];
1806 if (fr->fr_ifnames[1] != -1) {
1807 strncpy(is->is_ifname[(out << 1) + 1],
1808 FR_NAME(fr, fr_ifnames[1]),
1809 sizeof(fr->fr_ifnames[1]));
1812 is->is_ifp[(1 - out) << 1] = fr->fr_ifas[2];
1813 if (fr->fr_ifnames[2] != -1) {
1814 strncpy(is->is_ifname[((1 - out) << 1)],
1815 FR_NAME(fr, fr_ifnames[2]),
1816 sizeof(fr->fr_ifnames[2]));
1819 is->is_ifp[((1 - out) << 1) + 1] = fr->fr_ifas[3];
1820 if (fr->fr_ifnames[3] != -1) {
1821 strncpy(is->is_ifname[((1 - out) << 1) + 1],
1822 FR_NAME(fr, fr_ifnames[3]),
1823 sizeof(fr->fr_ifnames[3]));
1826 if (fin->fin_ifp != NULL) {
1827 is->is_ifp[out << 1] = fin->fin_ifp;
1828 COPYIFNAME(fin->fin_v, fin->fin_ifp,
1829 is->is_ifname[out << 1]);
1833 if (fin->fin_p == IPPROTO_TCP) {
1835 * If we're creating state for a starting connection, start the
1836 * timer on it as we'll never see an error if it fails to
1839 (void) ipf_tcp_age(&is->is_sti, fin, softs->ipf_state_tcptq,
1842 MUTEX_EXIT(&is->is_lock);
1843 if ((is->is_flags & IS_STATESYNC) && ((is->is_flags & SI_CLONE) == 0))
1844 is->is_sync = ipf_sync_new(softc, SMC_STATE, fin, is);
1845 if (softs->ipf_state_logging)
1846 ipf_state_log(softc, is, ISL_NEW);
1848 RWLOCK_EXIT(&softc->ipf_state);
1850 fin->fin_flx |= FI_STATE;
1851 if (fin->fin_flx & FI_FRAG)
1852 (void) ipf_frag_new(softc, fin, pass);
1854 fdp = &fr->fr_tifs[0];
1855 if (fdp->fd_type == FRD_DSTLIST) {
1856 ipf_dstlist_select_node(fin, fdp->fd_ptr, NULL,
1859 bcopy(fdp, &is->is_tifs[0], sizeof(*fdp));
1862 fdp = &fr->fr_tifs[1];
1863 if (fdp->fd_type == FRD_DSTLIST) {
1864 ipf_dstlist_select_node(fin, fdp->fd_ptr, NULL,
1867 bcopy(fdp, &is->is_tifs[1], sizeof(*fdp));
1869 fin->fin_tif = &is->is_tifs[fin->fin_rev];
1872 if (fdp->fd_type == FRD_DSTLIST) {
1873 ipf_dstlist_select_node(fin, fdp->fd_ptr, NULL,
1876 bcopy(fdp, &is->is_dif, sizeof(*fdp));
1878 fin->fin_dif = &is->is_dif;
1884 /* ------------------------------------------------------------------------ */
1885 /* Function: ipf_tcpoptions */
1886 /* Returns: int - 1 == packet matches state entry, 0 == it does not, */
1887 /* -1 == packet has bad TCP options data */
1888 /* Parameters: softs(I) - pointer to state context structure */
1889 /* fin(I) - pointer to packet information */
1890 /* tcp(I) - pointer to TCP packet header */
1891 /* td(I) - pointer to TCP data held as part of the state */
1893 /* Look after the TCP header for any options and deal with those that are */
1894 /* present. Record details about those that we recogise. */
1895 /* ------------------------------------------------------------------------ */
1897 ipf_tcpoptions(softs, fin, tcp, td)
1898 ipf_state_softc_t *softs;
1903 int off, mlen, ol, i, len, retval;
1904 char buf[64], *s, opt;
1907 len = (TCP_OFF(tcp) << 2);
1908 if (fin->fin_dlen < len) {
1909 SBUMPD(ipf_state_stats, iss_tcp_toosmall);
1912 len -= sizeof(*tcp);
1914 off = fin->fin_plen - fin->fin_dlen + sizeof(*tcp) + fin->fin_ipoff;
1917 mlen = MSGDSIZE(m) - off;
1925 COPYDATA(m, off, len, buf);
1927 for (s = buf; len > 0; ) {
1929 if (opt == TCPOPT_EOL)
1931 else if (opt == TCPOPT_NOP)
1937 if (ol < 2 || ol > len)
1941 * Extract the TCP options we are interested in out of
1942 * the header and store them in the the tcpdata struct.
1946 case TCPOPT_WINDOW :
1947 if (ol == TCPOLEN_WINDOW) {
1949 if (i > TCP_WSCALE_MAX)
1953 td->td_winscale = i;
1954 td->td_winflags |= TCP_WSCALE_SEEN|
1959 case TCPOPT_MAXSEG :
1961 * So, if we wanted to set the TCP MAXSEG,
1962 * it should be done here...
1964 if (ol == TCPOLEN_MAXSEG) {
1972 case TCPOPT_SACK_PERMITTED :
1973 if (ol == TCPOLEN_SACK_PERMITTED)
1974 td->td_winflags |= TCP_SACK_PERMIT;
1984 SBUMPD(ipf_state_stats, iss_tcp_badopt);
1990 /* ------------------------------------------------------------------------ */
1991 /* Function: ipf_state_tcp */
1992 /* Returns: int - 1 == packet matches state entry, 0 == it does not */
1993 /* Parameters: softc(I) - pointer to soft context main structure */
1994 /* softs(I) - pointer to state context structure */
1995 /* fin(I) - pointer to packet information */
1996 /* tcp(I) - pointer to TCP packet header */
1997 /* is(I) - pointer to master state structure */
1999 /* Check to see if a packet with TCP headers fits within the TCP window. */
2000 /* Change timeout depending on whether new packet is a SYN-ACK returning */
2001 /* for a SYN or a RST or FIN which indicate time to close up shop. */
2002 /* ------------------------------------------------------------------------ */
2004 ipf_state_tcp(softc, softs, fin, tcp, is)
2005 ipf_main_softc_t *softc;
2006 ipf_state_softc_t *softs;
2011 tcpdata_t *fdata, *tdata;
2012 int source, ret, flags;
2014 source = !fin->fin_rev;
2015 if (((is->is_flags & IS_TCPFSM) != 0) && (source == 1) &&
2016 (ntohs(is->is_sport) != fin->fin_data[0]))
2018 fdata = &is->is_tcp.ts_data[!source];
2019 tdata = &is->is_tcp.ts_data[source];
2021 MUTEX_ENTER(&is->is_lock);
2024 * If a SYN packet is received for a connection that is on the way out
2025 * but hasn't yet departed then advance this session along the way.
2027 if ((tcp->th_flags & TH_OPENING) == TH_SYN) {
2028 if ((is->is_state[0] > IPF_TCPS_ESTABLISHED) &&
2029 (is->is_state[1] > IPF_TCPS_ESTABLISHED)) {
2030 is->is_state[!source] = IPF_TCPS_CLOSED;
2031 ipf_movequeue(softc->ipf_ticks, &is->is_sti,
2033 &softs->ipf_state_deletetq);
2034 MUTEX_EXIT(&is->is_lock);
2035 DT1(iss_tcp_closing, ipstate_t *, is);
2036 SBUMP(ipf_state_stats.iss_tcp_closing);
2041 if (is->is_flags & IS_LOOSE)
2044 ret = ipf_state_tcpinwindow(fin, fdata, tdata, tcp,
2048 * Nearing end of connection, start timeout.
2050 ret = ipf_tcp_age(&is->is_sti, fin, softs->ipf_state_tcptq,
2053 MUTEX_EXIT(&is->is_lock);
2054 DT2(iss_tcp_fsm, fr_info_t *, fin, ipstate_t *, is);
2055 SBUMP(ipf_state_stats.iss_tcp_fsm);
2059 if (softs->ipf_state_logging > 4)
2060 ipf_state_log(softc, is, ISL_STATECHANGE);
2063 * set s0's as appropriate. Use syn-ack packet as it
2064 * contains both pieces of required information.
2067 * Window scale option is only present in SYN/SYN-ACK packet.
2068 * Compare with ~TH_FIN to mask out T/TCP setups.
2070 flags = tcp->th_flags & ~(TH_FIN|TH_ECNALL);
2071 if (flags == (TH_SYN|TH_ACK)) {
2072 is->is_s0[source] = ntohl(tcp->th_ack);
2073 is->is_s0[!source] = ntohl(tcp->th_seq) + 1;
2074 if ((TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) {
2075 if (ipf_tcpoptions(softs, fin, tcp,
2077 fin->fin_flx |= FI_BAD;
2078 DT1(ipf_fi_bad_winscale_syn_ack, fr_info_t *, fin);
2081 if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN))
2082 ipf_checknewisn(fin, is);
2083 } else if (flags == TH_SYN) {
2084 is->is_s0[source] = ntohl(tcp->th_seq) + 1;
2085 if ((TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) {
2086 if (ipf_tcpoptions(softs, fin, tcp,
2088 fin->fin_flx |= FI_BAD;
2089 DT1(ipf_fi_bad_winscale_syn, fr_info_t *, fin);
2093 if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN))
2094 ipf_checknewisn(fin, is);
2099 DT2(iss_tcp_oow, fr_info_t *, fin, ipstate_t *, is);
2100 SBUMP(ipf_state_stats.iss_tcp_oow);
2103 MUTEX_EXIT(&is->is_lock);
2108 /* ------------------------------------------------------------------------ */
2109 /* Function: ipf_checknewisn */
2111 /* Parameters: fin(I) - pointer to packet information */
2112 /* is(I) - pointer to master state structure */
2114 /* Check to see if this TCP connection is expecting and needs a new */
2115 /* sequence number for a particular direction of the connection. */
2117 /* NOTE: This does not actually change the sequence numbers, only gets new */
2119 /* ------------------------------------------------------------------------ */
2121 ipf_checknewisn(fin, is)
2125 u_32_t sumd, old, new;
2132 if (((i == 0) && !(is->is_flags & IS_ISNSYN)) ||
2133 ((i == 1) && !(is->is_flags & IS_ISNACK))) {
2134 old = ntohl(tcp->th_seq);
2135 new = ipf_newisn(fin);
2136 is->is_isninc[i] = new - old;
2137 CALC_SUMD(old, new, sumd);
2138 is->is_sumd[i] = (sumd & 0xffff) + (sumd >> 16);
2140 is->is_flags |= ((i == 0) ? IS_ISNSYN : IS_ISNACK);
2145 /* ------------------------------------------------------------------------ */
2146 /* Function: ipf_state_tcpinwindow */
2147 /* Returns: int - 1 == packet inside TCP "window", 0 == not inside. */
2148 /* Parameters: fin(I) - pointer to packet information */
2149 /* fdata(I) - pointer to tcp state informatio (forward) */
2150 /* tdata(I) - pointer to tcp state informatio (reverse) */
2151 /* tcp(I) - pointer to TCP packet header */
2153 /* Given a packet has matched addresses and ports, check to see if it is */
2154 /* within the TCP data window. In a show of generosity, allow packets that */
2155 /* are within the window space behind the current sequence # as well. */
2156 /* ------------------------------------------------------------------------ */
2158 ipf_state_tcpinwindow(fin, fdata, tdata, tcp, flags)
2160 tcpdata_t *fdata, *tdata;
2164 ipf_main_softc_t *softc = fin->fin_main_soft;
2165 ipf_state_softc_t *softs = softc->ipf_state_soft;
2166 tcp_seq seq, ack, end;
2167 int ackskew, tcpflags;
2172 * Find difference between last checked packet and this packet.
2174 tcpflags = tcp->th_flags;
2175 seq = ntohl(tcp->th_seq);
2176 ack = ntohl(tcp->th_ack);
2177 if (tcpflags & TH_SYN)
2178 win = ntohs(tcp->th_win);
2180 win = ntohs(tcp->th_win) << fdata->td_winscale;
2183 * A window of 0 produces undesirable behaviour from this function.
2188 dsize = fin->fin_dlen - (TCP_OFF(tcp) << 2) +
2189 ((tcpflags & TH_SYN) ? 1 : 0) + ((tcpflags & TH_FIN) ? 1 : 0);
2192 * if window scaling is present, the scaling is only allowed
2193 * for windows not in the first SYN packet. In that packet the
2194 * window is 65535 to specify the largest window possible
2195 * for receivers not implementing the window scale option.
2196 * Currently, we do not assume TTCP here. That means that
2197 * if we see a second packet from a host (after the initial
2198 * SYN), we can assume that the receiver of the SYN did
2199 * already send back the SYN/ACK (and thus that we know if
2200 * the receiver also does window scaling)
2202 if (!(tcpflags & TH_SYN) && (fdata->td_winflags & TCP_WSCALE_FIRST)) {
2203 fdata->td_winflags &= ~TCP_WSCALE_FIRST;
2204 fdata->td_maxwin = win;
2209 if ((fdata->td_end == 0) &&
2210 (!(flags & IS_TCPFSM) ||
2211 ((tcpflags & TH_OPENING) == TH_OPENING))) {
2213 * Must be a (outgoing) SYN-ACK in reply to a SYN.
2215 fdata->td_end = end - 1;
2216 fdata->td_maxwin = 1;
2217 fdata->td_maxend = end + win;
2220 if (!(tcpflags & TH_ACK)) { /* Pretend an ack was sent */
2221 ack = tdata->td_end;
2222 } else if (((tcpflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) &&
2224 /* gross hack to get around certain broken tcp stacks */
2225 ack = tdata->td_end;
2228 maxwin = tdata->td_maxwin;
2229 ackskew = tdata->td_end - ack;
2232 * Strict sequencing only allows in-order delivery.
2234 if ((flags & IS_STRICT) != 0) {
2235 if (seq != fdata->td_end) {
2236 DT2(iss_tcp_struct, tcpdata_t *, fdata, int, seq);
2237 SBUMP(ipf_state_stats.iss_tcp_strict);
2238 fin->fin_flx |= FI_OOW;
2243 #define SEQ_GE(a,b) ((int)((a) - (b)) >= 0)
2244 #define SEQ_GT(a,b) ((int)((a) - (b)) > 0)
2246 if ((SEQ_GE(fdata->td_maxend, end)) &&
2247 (SEQ_GE(seq, fdata->td_end - maxwin)) &&
2248 /* XXX what about big packets */
2249 #define MAXACKWINDOW 66000
2250 (-ackskew <= (MAXACKWINDOW)) &&
2251 ( ackskew <= (MAXACKWINDOW << fdata->td_winscale))) {
2254 * Microsoft Windows will send the next packet to the right of the
2255 * window if SACK is in use.
2257 } else if ((seq == fdata->td_maxend) && (ackskew == 0) &&
2258 (fdata->td_winflags & TCP_SACK_PERMIT) &&
2259 (tdata->td_winflags & TCP_SACK_PERMIT)) {
2260 DT2(iss_sinsack, tcpdata_t *, fdata, int, seq);
2261 SBUMP(ipf_state_stats.iss_winsack);
2264 * Sometimes a TCP RST will be generated with only the ACK field
2267 } else if ((seq == 0) && (tcpflags == (TH_RST|TH_ACK)) &&
2268 (ackskew >= -1) && (ackskew <= 1)) {
2270 } else if (!(flags & IS_TCPFSM)) {
2273 i = (fin->fin_rev << 1) + fin->fin_out;
2276 if (is_pkts[i]0 == 0) {
2278 * Picking up a connection in the middle, the "next"
2279 * packet seen from a direction that is new should be
2280 * accepted, even if it appears out of sequence.
2285 if (!(fdata->td_winflags &
2286 (TCP_WSCALE_SEEN|TCP_WSCALE_FIRST))) {
2288 * No TCPFSM and no window scaling, so make some
2291 if ((seq == fdata->td_maxend) && (ackskew == 0))
2293 else if (SEQ_GE(seq + maxwin, fdata->td_end - maxwin))
2298 /* TRACE(inseq, fdata, tdata, seq, end, ack, ackskew, win, maxwin) */
2301 /* if ackskew < 0 then this should be due to fragmented
2302 * packets. There is no way to know the length of the
2303 * total packet in advance.
2304 * We do know the total length from the fragment cache though.
2305 * Note however that there might be more sessions with
2306 * exactly the same source and destination parameters in the
2307 * state cache (and source and destination is the only stuff
2308 * that is saved in the fragment cache). Note further that
2309 * some TCP connections in the state cache are hashed with
2310 * sport and dport as well which makes it not worthwhile to
2312 * Thus, when ackskew is negative but still seems to belong
2313 * to this session, we bump up the destinations end value.
2316 tdata->td_end = ack;
2318 /* update max window seen */
2319 if (fdata->td_maxwin < win)
2320 fdata->td_maxwin = win;
2321 if (SEQ_GT(end, fdata->td_end))
2322 fdata->td_end = end;
2323 if (SEQ_GE(ack + win, tdata->td_maxend))
2324 tdata->td_maxend = ack + win;
2327 SBUMP(ipf_state_stats.iss_oow);
2328 fin->fin_flx |= FI_OOW;
2333 /* ------------------------------------------------------------------------ */
2334 /* Function: ipf_state_clone */
2335 /* Returns: ipstate_t* - NULL == cloning failed, */
2336 /* else pointer to new state structure */
2337 /* Parameters: fin(I) - pointer to packet information */
2338 /* tcp(I) - pointer to TCP/UDP header */
2339 /* is(I) - pointer to master state structure */
2341 /* Create a "duplcate" state table entry from the master. */
2342 /* ------------------------------------------------------------------------ */
2344 ipf_state_clone(fin, tcp, is)
2349 ipf_main_softc_t *softc = fin->fin_main_soft;
2350 ipf_state_softc_t *softs = softc->ipf_state_soft;
2354 if (softs->ipf_state_stats.iss_active == softs->ipf_state_max) {
2355 SBUMPD(ipf_state_stats, iss_max);
2356 softs->ipf_state_doflush = 1;
2359 KMALLOC(clone, ipstate_t *);
2360 if (clone == NULL) {
2361 SBUMPD(ipf_state_stats, iss_clone_nomem);
2364 bcopy((char *)is, (char *)clone, sizeof(*clone));
2366 MUTEX_NUKE(&clone->is_lock);
2368 * It has not yet been placed on any timeout queue, so make sure
2369 * all of that data is zero'd out.
2371 clone->is_sti.tqe_pnext = NULL;
2372 clone->is_sti.tqe_next = NULL;
2373 clone->is_sti.tqe_ifq = NULL;
2374 clone->is_sti.tqe_parent = clone;
2376 clone->is_die = ONE_DAY + softc->ipf_ticks;
2377 clone->is_state[0] = 0;
2378 clone->is_state[1] = 0;
2379 send = ntohl(tcp->th_seq) + fin->fin_dlen - (TCP_OFF(tcp) << 2) +
2380 ((tcp->th_flags & TH_SYN) ? 1 : 0) +
2381 ((tcp->th_flags & TH_FIN) ? 1 : 0);
2383 if (fin->fin_rev == 1) {
2384 clone->is_dend = send;
2385 clone->is_maxdend = send;
2387 clone->is_maxswin = 1;
2388 clone->is_maxdwin = ntohs(tcp->th_win);
2389 if (clone->is_maxdwin == 0)
2390 clone->is_maxdwin = 1;
2392 clone->is_send = send;
2393 clone->is_maxsend = send;
2395 clone->is_maxdwin = 1;
2396 clone->is_maxswin = ntohs(tcp->th_win);
2397 if (clone->is_maxswin == 0)
2398 clone->is_maxswin = 1;
2401 clone->is_flags &= ~SI_CLONE;
2402 clone->is_flags |= SI_CLONED;
2403 if (ipf_state_insert(softc, clone, fin->fin_rev) == -1) {
2409 if (clone->is_p == IPPROTO_TCP) {
2410 (void) ipf_tcp_age(&clone->is_sti, fin, softs->ipf_state_tcptq,
2411 clone->is_flags, 2);
2413 MUTEX_EXIT(&clone->is_lock);
2414 if (is->is_flags & IS_STATESYNC)
2415 clone->is_sync = ipf_sync_new(softc, SMC_STATE, fin, clone);
2416 DT2(iss_clone, ipstate_t *, is, ipstate_t *, clone);
2417 SBUMP(ipf_state_stats.iss_cloned);
2422 /* ------------------------------------------------------------------------ */
2423 /* Function: ipf_matchsrcdst */
2425 /* Parameters: fin(I) - pointer to packet information */
2426 /* is(I) - pointer to state structure */
2427 /* src(I) - pointer to source address */
2428 /* dst(I) - pointer to destination address */
2429 /* tcp(I) - pointer to TCP/UDP header */
2430 /* cmask(I) - mask of FI_* bits to check */
2432 /* Match a state table entry against an IP packet. The logic below is that */
2433 /* ret gets set to one if the match succeeds, else remains 0. If it is */
2434 /* still 0 after the test. no match. */
2435 /* ------------------------------------------------------------------------ */
2437 ipf_matchsrcdst(fin, is, src, dst, tcp, cmask)
2440 i6addr_t *src, *dst;
2444 ipf_main_softc_t *softc = fin->fin_main_soft;
2445 ipf_state_softc_t *softs = softc->ipf_state_soft;
2446 int ret = 0, rev, out, flags, flx = 0, idx;
2452 * If a connection is about to be deleted, no packets
2453 * are allowed to match it.
2455 if (is->is_sti.tqe_ifq == &softs->ipf_state_deletetq)
2458 rev = IP6_NEQ(&is->is_dst, dst);
2461 flags = is->is_flags;
2466 sp = htons(fin->fin_sport);
2467 dp = ntohs(fin->fin_dport);
2471 if (!(flags & SI_W_SPORT) && (sp != is->is_sport))
2473 else if (!(flags & SI_W_DPORT) && (dp != is->is_dport))
2478 idx = (out << 1) + rev;
2481 * If the interface for this 'direction' is set, make sure it matches.
2482 * An interface name that is not set matches any, as does a name of *.
2484 if ((is->is_ifp[idx] == ifp) || (is->is_ifp[idx] == NULL &&
2485 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '-' ||
2486 *is->is_ifname[idx] == '*')))
2490 DT2(iss_lookup_badifp, fr_info_t *, fin, ipstate_t *, is);
2491 SBUMP(ipf_state_stats.iss_lookup_badifp);
2492 /* TRACE is, out, rev, idx */
2498 * Match addresses and ports.
2501 if ((IP6_EQ(&is->is_dst, dst) || (flags & SI_W_DADDR)) &&
2502 (IP6_EQ(&is->is_src, src) || (flags & SI_W_SADDR))) {
2504 if ((sp == is->is_sport || flags & SI_W_SPORT)
2506 (dp == is->is_dport || flags & SI_W_DPORT))
2513 if ((IP6_EQ(&is->is_dst, src) || (flags & SI_W_DADDR)) &&
2514 (IP6_EQ(&is->is_src, dst) || (flags & SI_W_SADDR))) {
2516 if ((dp == is->is_sport || flags & SI_W_SPORT)
2518 (sp == is->is_dport || flags & SI_W_DPORT))
2527 SBUMP(ipf_state_stats.iss_lookup_badport);
2528 DT2(iss_lookup_badport, fr_info_t *, fin, ipstate_t *, is);
2529 /* TRACE rev, is, sp, dp, src, dst */
2534 * Whether or not this should be here, is questionable, but the aim
2535 * is to get this out of the main line.
2538 flags = is->is_flags & ~(SI_WILDP|SI_NEWFR|SI_CLONE|SI_CLONED);
2541 * Only one of the source or destination address can be flaged as a
2542 * wildcard. Fill in the missing address, if set.
2543 * For IPv6, if the address being copied in is multicast, then
2544 * don't reset the wild flag - multicast causes it to be set in the
2547 if ((flags & (SI_W_SADDR|SI_W_DADDR))) {
2548 fr_ip_t *fi = &fin->fin_fi;
2550 if ((flags & SI_W_SADDR) != 0) {
2552 is->is_src = fi->fi_src;
2553 is->is_flags &= ~SI_W_SADDR;
2555 if (!(fin->fin_flx & (FI_MULTICAST|FI_MBCAST))){
2556 is->is_src = fi->fi_dst;
2557 is->is_flags &= ~SI_W_SADDR;
2560 } else if ((flags & SI_W_DADDR) != 0) {
2562 if (!(fin->fin_flx & (FI_MULTICAST|FI_MBCAST))){
2563 is->is_dst = fi->fi_dst;
2564 is->is_flags &= ~SI_W_DADDR;
2567 is->is_dst = fi->fi_src;
2568 is->is_flags &= ~SI_W_DADDR;
2571 if ((is->is_flags & (SI_WILDA|SI_WILDP)) == 0) {
2572 ATOMIC_DECL(softs->ipf_state_stats.iss_wild);
2576 flx = fin->fin_flx & cmask;
2577 cflx = is->is_flx[out][rev];
2580 * Match up any flags set from IP options.
2582 if ((cflx && (flx != (cflx & cmask))) ||
2583 ((fin->fin_optmsk & is->is_optmsk[rev]) != is->is_opt[rev]) ||
2584 ((fin->fin_secmsk & is->is_secmsk) != is->is_sec) ||
2585 ((fin->fin_auth & is->is_authmsk) != is->is_auth)) {
2586 SBUMPD(ipf_state_stats, iss_miss_mask);
2590 if ((fin->fin_flx & FI_IGNORE) != 0) {
2596 * Only one of the source or destination port can be flagged as a
2597 * wildcard. When filling it in, fill in a copy of the matched entry
2598 * if it has the cloning flag set.
2600 if ((flags & (SI_W_SPORT|SI_W_DPORT))) {
2601 if ((flags & SI_CLONE) != 0) {
2604 clone = ipf_state_clone(fin, tcp, is);
2609 ATOMIC_DECL(softs->ipf_state_stats.iss_wild);
2612 if ((flags & SI_W_SPORT) != 0) {
2615 is->is_send = ntohl(tcp->th_seq);
2618 is->is_send = ntohl(tcp->th_ack);
2620 is->is_maxsend = is->is_send + 1;
2621 } else if ((flags & SI_W_DPORT) != 0) {
2624 is->is_dend = ntohl(tcp->th_ack);
2627 is->is_dend = ntohl(tcp->th_seq);
2629 is->is_maxdend = is->is_dend + 1;
2631 is->is_flags &= ~(SI_W_SPORT|SI_W_DPORT);
2632 if ((flags & SI_CLONED) && softs->ipf_state_logging)
2633 ipf_state_log(softc, is, ISL_CLONE);
2638 if (is->is_flx[out][rev] == 0) {
2639 is->is_flx[out][rev] = flx;
2640 if (rev == 1 && is->is_optmsk[1] == 0) {
2641 is->is_opt[1] = fin->fin_optmsk;
2642 is->is_optmsk[1] = 0xffffffff;
2643 if (is->is_v == 6) {
2644 is->is_opt[1] &= ~0x8;
2645 is->is_optmsk[1] &= ~0x8;
2651 * Check if the interface name for this "direction" is set and if not,
2654 if (is->is_ifp[idx] == NULL &&
2655 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) {
2656 is->is_ifp[idx] = ifp;
2657 COPYIFNAME(fin->fin_v, ifp, is->is_ifname[idx]);
2664 /* ------------------------------------------------------------------------ */
2665 /* Function: ipf_checkicmpmatchingstate */
2667 /* Parameters: fin(I) - pointer to packet information */
2669 /* If we've got an ICMP error message, using the information stored in the */
2670 /* ICMP packet, look for a matching state table entry. */
2672 /* If we return NULL then no lock on ipf_state is held. */
2673 /* If we return non-null then a read-lock on ipf_state is held. */
2674 /* ------------------------------------------------------------------------ */
2676 ipf_checkicmpmatchingstate(fin)
2679 ipf_main_softc_t *softc = fin->fin_main_soft;
2680 ipf_state_softc_t *softs = softc->ipf_state_soft;
2681 ipstate_t *is, **isp;
2694 * Does it at least have the return (basic) IP header ?
2695 * Is it an actual recognised ICMP error type?
2696 * Only a basic IP header (no options) should be with
2697 * an ICMP error header.
2699 if ((fin->fin_v != 4) || (fin->fin_hlen != sizeof(ip_t)) ||
2700 (fin->fin_plen < ICMPERR_MINPKTLEN) ||
2701 !(fin->fin_flx & FI_ICMPERR)) {
2702 SBUMPD(ipf_state_stats, iss_icmp_bad);
2706 type = ic->icmp_type;
2708 oip = (ip_t *)((char *)ic + ICMPERR_ICMPHLEN);
2710 * Check if the at least the old IP header (with options) and
2711 * 8 bytes of payload is present.
2713 if (fin->fin_plen < ICMPERR_MAXPKTLEN + ((IP_HL(oip) - 5) << 2)) {
2714 SBUMPDX(ipf_state_stats, iss_icmp_short, iss_icmp_short_1);
2721 len = fin->fin_dlen - ICMPERR_ICMPHLEN;
2722 if ((len <= 0) || ((IP_HL(oip) << 2) > len)) {
2723 DT2(iss_icmp_len, fr_info_t *, fin, struct ip*, oip);
2724 SBUMPDX(ipf_state_stats, iss_icmp_short, iss_icmp_short_1);
2729 * Is the buffer big enough for all of it ? It's the size of the IP
2730 * header claimed in the encapsulated part which is of concern. It
2731 * may be too big to be in this buffer but not so big that it's
2732 * outside the ICMP packet, leading to TCP deref's causing problems.
2733 * This is possible because we don't know how big oip_hl is when we
2734 * do the pullup early in ipf_check() and thus can't guarantee it is
2743 if ((char *)oip + len > (char *)m->b_wptr) {
2744 SBUMPDX(ipf_state_stats, iss_icmp_short, iss_icmp_short_2);
2748 if ((char *)oip + len > (char *)fin->fin_ip + m->m_len) {
2749 SBUMPDX(ipf_state_stats, iss_icmp_short, iss_icmp_short_3);
2756 bcopy((char *)fin, (char *)&ofin, sizeof(*fin));
2759 * in the IPv4 case we must zero the i6addr union otherwise
2760 * the IP6_EQ and IP6_NEQ macros produce the wrong results because
2761 * of the 'junk' in the unused part of the union
2763 bzero((char *)&src, sizeof(src));
2764 bzero((char *)&dst, sizeof(dst));
2767 * we make an fin entry to be able to feed it to
2768 * matchsrcdst note that not all fields are encessary
2769 * but this is the cleanest way. Note further we fill
2770 * in fin_mp such that if someone uses it we'll get
2771 * a kernel panic. ipf_matchsrcdst does not use this.
2773 * watch out here, as ip is in host order and oip in network
2774 * order. Any change we make must be undone afterwards, like
2777 savelen = oip->ip_len;
2778 oip->ip_len = htons(len);
2780 ofin.fin_flx = FI_NOCKSUM;
2783 ofin.fin_m = NULL; /* if dereferenced, panic XXX */
2784 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */
2785 (void) ipf_makefrip(IP_HL(oip) << 2, oip, &ofin);
2786 ofin.fin_ifp = fin->fin_ifp;
2787 ofin.fin_out = !fin->fin_out;
2789 hv = (pr = oip->ip_p);
2790 src.in4 = oip->ip_src;
2791 hv += src.in4.s_addr;
2792 dst.in4 = oip->ip_dst;
2793 hv += dst.in4.s_addr;
2796 * Reset the short and bad flag here because in ipf_matchsrcdst()
2797 * the flags for the current packet (fin_flx) are compared against
2798 * those for the existing session.
2800 ofin.fin_flx &= ~(FI_BAD|FI_SHORT);
2803 * Put old values of ip_len back as we don't know
2804 * if we have to forward the packet or process it again.
2806 oip->ip_len = savelen;
2812 * an ICMP error can only be generated as a result of an
2813 * ICMP query, not as the response on an ICMP error
2815 * XXX theoretically ICMP_ECHOREP and the other reply's are
2816 * ICMP query's as well, but adding them here seems strange XXX
2818 if ((ofin.fin_flx & FI_ICMPERR) != 0) {
2819 DT1(iss_icmp_icmperr, fr_info_t *, &ofin);
2820 SBUMP(ipf_state_stats.iss_icmp_icmperr);
2825 * perform a lookup of the ICMP packet in the state table
2827 icmp = (icmphdr_t *)((char *)oip + (IP_HL(oip) << 2));
2828 hv += icmp->icmp_id;
2829 hv = DOUBLE_HASH(hv);
2831 READ_ENTER(&softc->ipf_state);
2832 for (isp = &softs->ipf_state_table[hv];
2833 ((is = *isp) != NULL); ) {
2834 isp = &is->is_hnext;
2835 if ((is->is_p != pr) || (is->is_v != 4))
2837 if (is->is_pass & FR_NOICMPERR)
2840 is = ipf_matchsrcdst(&ofin, is, &src, &dst,
2842 if ((is != NULL) && !ipf_allowstateicmp(fin, is, &src))
2845 RWLOCK_EXIT(&softc->ipf_state);
2846 SBUMPDX(ipf_state_stats, iss_icmp_miss, iss_icmp_miss_1);
2852 SBUMPDX(ipf_state_stats, iss_icmp_miss, iss_icmp_miss_2);
2856 tcp = (tcphdr_t *)((char *)oip + (IP_HL(oip) << 2));
2858 hv += tcp->th_dport;;
2859 hv += tcp->th_sport;;
2860 hv = DOUBLE_HASH(hv);
2862 READ_ENTER(&softc->ipf_state);
2863 for (isp = &softs->ipf_state_table[hv]; ((is = *isp) != NULL); ) {
2864 isp = &is->is_hnext;
2866 * Only allow this icmp though if the
2867 * encapsulated packet was allowed through the
2868 * other way around. Note that the minimal amount
2869 * of info present does not allow for checking against
2870 * tcp internals such as seq and ack numbers. Only the
2871 * ports are known to be present and can be even if the
2872 * short flag is set.
2874 if ((is->is_p == pr) && (is->is_v == 4) &&
2875 (is = ipf_matchsrcdst(&ofin, is, &src, &dst,
2876 tcp, FI_ICMPCMP))) {
2877 if (ipf_allowstateicmp(fin, is, &src) == 0)
2881 RWLOCK_EXIT(&softc->ipf_state);
2882 SBUMPDX(ipf_state_stats, iss_icmp_miss, iss_icmp_miss_3);
2887 /* ------------------------------------------------------------------------ */
2888 /* Function: ipf_allowstateicmp */
2889 /* Returns: int - 1 = packet denied, 0 = packet allowed */
2890 /* Parameters: fin(I) - pointer to packet information */
2891 /* is(I) - pointer to state table entry */
2892 /* src(I) - source address to check permission for */
2894 /* For an ICMP packet that has so far matched a state table entry, check if */
2895 /* there are any further refinements that might mean we want to block this */
2896 /* packet. This code isn't specific to either IPv4 or IPv6. */
2897 /* ------------------------------------------------------------------------ */
2899 ipf_allowstateicmp(fin, is, src)
2904 ipf_main_softc_t *softc = fin->fin_main_soft;
2905 ipf_state_softc_t *softs = softc->ipf_state_soft;
2914 if (fr != NULL && fr->fr_icmpgrp != NULL) {
2915 savefr = fin->fin_fr;
2916 fin->fin_fr = fr->fr_icmpgrp->fg_start;
2918 ipass = ipf_scanlist(fin, softc->ipf_pass);
2919 fin->fin_fr = savefr;
2920 if (FR_ISBLOCK(ipass)) {
2921 SBUMPD(ipf_state_stats, iss_icmp_headblock);
2927 * i : the index of this packet (the icmp unreachable)
2928 * oi : the index of the original packet found in the
2929 * icmp header (i.e. the packet causing this icmp)
2930 * backward : original packet was backward compared to
2933 backward = IP6_NEQ(&is->is_src, src);
2934 fin->fin_rev = !backward;
2935 i = (!backward << 1) + fin->fin_out;
2936 oi = (backward << 1) + !fin->fin_out;
2938 if (is->is_pass & FR_NOICMPERR) {
2939 SBUMPD(ipf_state_stats, iss_icmp_banned);
2942 if (is->is_icmppkts[i] > is->is_pkts[oi]) {
2943 SBUMPD(ipf_state_stats, iss_icmp_toomany);
2947 DT2(iss_icmp_hits, fr_info_t *, fin, ipstate_t *, is);
2948 SBUMP(ipf_state_stats.iss_icmp_hits);
2949 is->is_icmppkts[i]++;
2952 * we deliberately do not touch the timeouts
2953 * for the accompanying state table entry.
2954 * It remains to be seen if that is correct. XXX
2960 /* ------------------------------------------------------------------------ */
2961 /* Function: ipf_ipsmove */
2963 /* Parameters: is(I) - pointer to state table entry */
2964 /* hv(I) - new hash value for state table entry */
2965 /* Write Locks: ipf_state */
2967 /* Move a state entry from one position in the hash table to another. */
2968 /* ------------------------------------------------------------------------ */
2970 ipf_ipsmove(softs, is, hv)
2971 ipf_state_softc_t *softs;
2980 /* TRACE is, is_hv, hvm */
2983 * Remove the hash from the old location...
2985 isp = is->is_phnext;
2987 is->is_hnext->is_phnext = isp;
2988 *isp = is->is_hnext;
2989 if (softs->ipf_state_table[hvm] == NULL)
2990 softs->ipf_state_stats.iss_inuse--;
2991 softs->ipf_state_stats.iss_bucketlen[hvm]--;
2994 * ...and put the hash in the new one.
2996 hvm = DOUBLE_HASH(hv);
2999 /* TRACE is, hv, is_hv, hvm */
3001 isp = &softs->ipf_state_table[hvm];
3003 (*isp)->is_phnext = &is->is_hnext;
3005 softs->ipf_state_stats.iss_inuse++;
3006 softs->ipf_state_stats.iss_bucketlen[hvm]++;
3007 is->is_phnext = isp;
3008 is->is_hnext = *isp;
3013 /* ------------------------------------------------------------------------ */
3014 /* Function: ipf_state_lookup */
3015 /* Returns: ipstate_t* - NULL == no matching state found, */
3016 /* else pointer to state information is returned */
3017 /* Parameters: fin(I) - pointer to packet information */
3018 /* tcp(I) - pointer to TCP/UDP header. */
3019 /* ifqp(O) - pointer for storing tailq timeout */
3021 /* Search the state table for a matching entry to the packet described by */
3022 /* the contents of *fin. For certain protocols, when a match is found the */
3023 /* timeout queue is also selected and stored in ifpq if it is non-NULL. */
3025 /* If we return NULL then no lock on ipf_state is held. */
3026 /* If we return non-null then a read-lock on ipf_state is held. */
3027 /* ------------------------------------------------------------------------ */
3029 ipf_state_lookup(fin, tcp, ifqp)
3034 ipf_main_softc_t *softc = fin->fin_main_soft;
3035 ipf_state_softc_t *softs = softc->ipf_state_soft;
3036 u_int hv, hvm, pr, v, tryagain;
3037 ipstate_t *is, **isp;
3038 u_short dport, sport;
3047 ic = (struct icmp *)tcp;
3048 hv = (pr = fin->fin_fi.fi_p);
3049 src = fin->fin_fi.fi_src;
3050 dst = fin->fin_fi.fi_dst;
3051 hv += src.in4.s_addr;
3052 hv += dst.in4.s_addr;
3054 v = fin->fin_fi.fi_v;
3057 hv += fin->fin_fi.fi_src.i6[1];
3058 hv += fin->fin_fi.fi_src.i6[2];
3059 hv += fin->fin_fi.fi_src.i6[3];
3061 if ((fin->fin_p == IPPROTO_ICMPV6) &&
3062 IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_dst.in6)) {
3063 hv -= dst.in4.s_addr;
3065 hv += fin->fin_fi.fi_dst.i6[1];
3066 hv += fin->fin_fi.fi_dst.i6[2];
3067 hv += fin->fin_fi.fi_dst.i6[3];
3072 (fin->fin_flx & (FI_MULTICAST|FI_BROADCAST|FI_MBCAST))) {
3073 if (fin->fin_out == 0) {
3074 hv -= src.in4.s_addr;
3076 hv -= dst.in4.s_addr;
3080 /* TRACE fin_saddr, fin_daddr, hv */
3083 * Search the hash table for matching packet header info.
3088 case IPPROTO_ICMPV6 :
3091 if ((ic->icmp_type == ICMP6_ECHO_REQUEST) ||
3092 (ic->icmp_type == ICMP6_ECHO_REPLY)) {
3096 READ_ENTER(&softc->ipf_state);
3098 hvm = DOUBLE_HASH(hv);
3099 for (isp = &softs->ipf_state_table[hvm];
3100 ((is = *isp) != NULL); ) {
3101 isp = &is->is_hnext;
3102 if ((is->is_p != pr) || (is->is_v != v))
3104 is = ipf_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP);
3106 ipf_matchicmpqueryreply(v, &is->is_icmp,
3107 ic, fin->fin_rev)) {
3109 ifq = &softs->ipf_state_icmpacktq;
3111 ifq = &softs->ipf_state_icmptq;
3117 if ((tryagain != 0) && !(is->is_flags & SI_W_DADDR)) {
3118 hv += fin->fin_fi.fi_src.i6[0];
3119 hv += fin->fin_fi.fi_src.i6[1];
3120 hv += fin->fin_fi.fi_src.i6[2];
3121 hv += fin->fin_fi.fi_src.i6[3];
3122 ipf_ipsmove(softs, is, hv);
3123 MUTEX_DOWNGRADE(&softc->ipf_state);
3127 RWLOCK_EXIT(&softc->ipf_state);
3130 * No matching icmp state entry. Perhaps this is a
3131 * response to another state entry.
3133 * XXX With some ICMP6 packets, the "other" address is already
3134 * in the packet, after the ICMP6 header, and this could be
3135 * used in place of the multicast address. However, taking
3136 * advantage of this requires some significant code changes
3137 * to handle the specific types where that is the case.
3139 if ((softs->ipf_state_stats.iss_wild != 0) &&
3140 ((fin->fin_flx & FI_NOWILD) == 0) &&
3141 (v == 6) && (tryagain == 0)) {
3142 hv -= fin->fin_fi.fi_src.i6[0];
3143 hv -= fin->fin_fi.fi_src.i6[1];
3144 hv -= fin->fin_fi.fi_src.i6[2];
3145 hv -= fin->fin_fi.fi_src.i6[3];
3147 WRITE_ENTER(&softc->ipf_state);
3151 is = ipf_checkicmp6matchingstate(fin);
3161 hv = DOUBLE_HASH(hv);
3162 READ_ENTER(&softc->ipf_state);
3163 for (isp = &softs->ipf_state_table[hv];
3164 ((is = *isp) != NULL); ) {
3165 isp = &is->is_hnext;
3166 if ((is->is_p != pr) || (is->is_v != v))
3168 is = ipf_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP);
3170 (ic->icmp_id == is->is_icmp.ici_id) &&
3171 ipf_matchicmpqueryreply(v, &is->is_icmp,
3172 ic, fin->fin_rev)) {
3174 ifq = &softs->ipf_state_icmpacktq;
3176 ifq = &softs->ipf_state_icmptq;
3181 RWLOCK_EXIT(&softc->ipf_state);
3188 sport = htons(fin->fin_data[0]);
3190 dport = htons(fin->fin_data[1]);
3194 READ_ENTER(&softc->ipf_state);
3196 hvm = DOUBLE_HASH(hv);
3200 for (isp = &softs->ipf_state_table[hvm];
3201 ((is = *isp) != NULL); ) {
3202 isp = &is->is_hnext;
3203 if ((is->is_p != pr) || (is->is_v != v))
3205 fin->fin_flx &= ~FI_OOW;
3206 is = ipf_matchsrcdst(fin, is, &src, &dst, tcp, FI_CMP);
3208 if (pr == IPPROTO_TCP) {
3209 if (!ipf_state_tcp(softc, softs, fin,
3211 oow |= fin->fin_flx & FI_OOW;
3220 !(is->is_flags & (SI_CLONE|SI_WILDP|SI_WILDA))) {
3223 ipf_ipsmove(softs, is, hv);
3224 MUTEX_DOWNGRADE(&softc->ipf_state);
3228 RWLOCK_EXIT(&softc->ipf_state);
3230 if ((softs->ipf_state_stats.iss_wild != 0) &&
3231 ((fin->fin_flx & FI_NOWILD) == 0)) {
3232 if (tryagain == 0) {
3235 } else if (tryagain == 1) {
3236 hv = fin->fin_fi.fi_p;
3238 * If we try to pretend this is a reply to a
3239 * multicast/broadcast packet then we need to
3240 * exclude part of the address from the hash
3243 if (fin->fin_out == 0) {
3244 hv += src.in4.s_addr;
3246 hv += dst.in4.s_addr;
3252 if (tryagain <= 2) {
3253 WRITE_ENTER(&softc->ipf_state);
3257 fin->fin_flx |= oow;
3263 if (GRE_REV(gre->gr_flags) == 1) {
3270 hvm = DOUBLE_HASH(hv);
3271 READ_ENTER(&softc->ipf_state);
3272 for (isp = &softs->ipf_state_table[hvm];
3273 ((is = *isp) != NULL); ) {
3274 isp = &is->is_hnext;
3275 if ((is->is_p != pr) || (is->is_v != v))
3277 is = ipf_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP);
3279 ifq = &softs->ipf_state_iptq;
3284 RWLOCK_EXIT(&softc->ipf_state);
3290 if (((is->is_sti.tqe_flags & TQE_RULEBASED) != 0) &&
3291 (is->is_tqehead[fin->fin_rev] != NULL))
3292 ifq = is->is_tqehead[fin->fin_rev];
3293 if (ifq != NULL && ifqp != NULL)
3296 SBUMP(ipf_state_stats.iss_lookup_miss);
3302 /* ------------------------------------------------------------------------ */
3303 /* Function: ipf_state_check */
3304 /* Returns: frentry_t* - NULL == search failed, */
3305 /* else pointer to rule for matching state */
3306 /* Parameters: fin(I) - pointer to packet information */
3307 /* passp(I) - pointer to filtering result flags */
3309 /* Check if a packet is associated with an entry in the state table. */
3310 /* ------------------------------------------------------------------------ */
3312 ipf_state_check(fin, passp)
3316 ipf_main_softc_t *softc = fin->fin_main_soft;
3317 ipf_state_softc_t *softs = softc->ipf_state_soft;
3326 if (softs->ipf_state_lock || (softs->ipf_state_list == NULL))
3329 if (fin->fin_flx & (FI_SHORT|FI_FRAGBODY|FI_BAD)) {
3330 SBUMPD(ipf_state_stats, iss_check_bad);
3334 if ((fin->fin_flx & FI_TCPUDP) ||
3335 (fin->fin_fi.fi_p == IPPROTO_ICMP)
3337 || (fin->fin_fi.fi_p == IPPROTO_ICMPV6)
3346 * Search the hash table for matching packet header info.
3348 is = ipf_state_lookup(fin, tcp, &ifq);
3353 case IPPROTO_ICMPV6 :
3356 if (fin->fin_v == 6) {
3357 is = ipf_checkicmp6matchingstate(fin);
3365 * No matching icmp state entry. Perhaps this is a
3366 * response to another state entry.
3368 is = ipf_checkicmpmatchingstate(fin);
3375 if (is->is_pass & FR_NEWISN) {
3376 if (fin->fin_out == 0)
3377 ipf_fixinisn(fin, is);
3378 else if (fin->fin_out == 1)
3379 ipf_fixoutisn(fin, is);
3384 ifq = &softs->ipf_state_udpacktq;
3386 ifq = &softs->ipf_state_udptq;
3390 SBUMP(ipf_state_stats.iss_check_miss);
3396 if ((fin->fin_out == 0) && (fr->fr_nattag.ipt_num[0] != 0)) {
3397 if (fin->fin_nattag == NULL) {
3398 RWLOCK_EXIT(&softc->ipf_state);
3399 SBUMPD(ipf_state_stats, iss_check_notag);
3402 if (ipf_matchtag(&fr->fr_nattag, fin->fin_nattag)!=0) {
3403 RWLOCK_EXIT(&softc->ipf_state);
3404 SBUMPD(ipf_state_stats, iss_check_nattag);
3408 (void) strncpy(fin->fin_group, FR_NAME(fr, fr_group),
3410 fin->fin_icode = fr->fr_icode;
3413 fin->fin_rule = is->is_rulen;
3417 * If this packet is a fragment and the rule says to track fragments,
3418 * then create a new fragment cache entry.
3420 if (fin->fin_flx & FI_FRAG && FR_ISPASS(is->is_pass) &&
3421 is->is_pass & FR_KEEPFRAG)
3422 (void) ipf_frag_new(softc, fin, is->is_pass);
3425 * For TCP packets, ifq == NULL. For all others, check if this new
3426 * queue is different to the last one it was on and move it if so.
3429 if ((tqe->tqe_flags & TQE_RULEBASED) != 0)
3430 ifq = is->is_tqehead[fin->fin_rev];
3432 MUTEX_ENTER(&is->is_lock);
3435 ipf_movequeue(softc->ipf_ticks, tqe, tqe->tqe_ifq, ifq);
3437 inout = (fin->fin_rev << 1) + fin->fin_out;
3438 is->is_pkts[inout]++;
3439 is->is_bytes[inout] += fin->fin_plen;
3440 fin->fin_pktnum = is->is_pkts[inout] + is->is_icmppkts[inout];
3442 MUTEX_EXIT(&is->is_lock);
3446 if (is->is_flags & IS_STATESYNC)
3447 ipf_sync_update(softc, SMC_STATE, fin, is->is_sync);
3449 RWLOCK_EXIT(&softc->ipf_state);
3451 SBUMP(ipf_state_stats.iss_hits);
3453 fin->fin_dif = &is->is_dif;
3454 fin->fin_tif = &is->is_tifs[fin->fin_rev];
3455 fin->fin_flx |= FI_STATE;
3456 if ((pass & FR_LOGFIRST) != 0)
3457 pass &= ~(FR_LOGFIRST|FR_LOG);
3463 /* ------------------------------------------------------------------------ */
3464 /* Function: ipf_fixoutisn */
3466 /* Parameters: fin(I) - pointer to packet information */
3467 /* is(I) - pointer to master state structure */
3469 /* Called only for outbound packets, adjusts the sequence number and the */
3470 /* TCP checksum to match that change. */
3471 /* ------------------------------------------------------------------------ */
3473 ipf_fixoutisn(fin, is)
3483 if ((is->is_flags & IS_ISNSYN) != 0) {
3484 if ((rev == 0) && (fin->fin_cksum < FI_CK_L4PART)) {
3485 seq = ntohl(tcp->th_seq);
3486 seq += is->is_isninc[0];
3487 tcp->th_seq = htonl(seq);
3488 ipf_fix_outcksum(0, &tcp->th_sum, is->is_sumd[0], 0);
3491 if ((is->is_flags & IS_ISNACK) != 0) {
3492 if ((rev == 1) && (fin->fin_cksum < FI_CK_L4PART)) {
3493 seq = ntohl(tcp->th_seq);
3494 seq += is->is_isninc[1];
3495 tcp->th_seq = htonl(seq);
3496 ipf_fix_outcksum(0, &tcp->th_sum, is->is_sumd[1], 0);
3502 /* ------------------------------------------------------------------------ */
3503 /* Function: ipf_fixinisn */
3505 /* Parameters: fin(I) - pointer to packet information */
3506 /* is(I) - pointer to master state structure */
3508 /* Called only for inbound packets, adjusts the acknowledge number and the */
3509 /* TCP checksum to match that change. */
3510 /* ------------------------------------------------------------------------ */
3512 ipf_fixinisn(fin, is)
3522 if ((is->is_flags & IS_ISNSYN) != 0) {
3523 if ((rev == 1) && (fin->fin_cksum < FI_CK_L4PART)) {
3524 ack = ntohl(tcp->th_ack);
3525 ack -= is->is_isninc[0];
3526 tcp->th_ack = htonl(ack);
3527 ipf_fix_incksum(0, &tcp->th_sum, is->is_sumd[0], 0);
3530 if ((is->is_flags & IS_ISNACK) != 0) {
3531 if ((rev == 0) && (fin->fin_cksum < FI_CK_L4PART)) {
3532 ack = ntohl(tcp->th_ack);
3533 ack -= is->is_isninc[1];
3534 tcp->th_ack = htonl(ack);
3535 ipf_fix_incksum(0, &tcp->th_sum, is->is_sumd[1], 0);
3541 /* ------------------------------------------------------------------------ */
3542 /* Function: ipf_state_sync */
3544 /* Parameters: softc(I) - pointer to soft context main structure */
3545 /* ifp(I) - pointer to interface */
3547 /* Walk through all state entries and if an interface pointer match is */
3548 /* found then look it up again, based on its name in case the pointer has */
3549 /* changed since last time. */
3551 /* If ifp is passed in as being non-null then we are only doing updates for */
3552 /* existing, matching, uses of it. */
3553 /* ------------------------------------------------------------------------ */
3555 ipf_state_sync(softc, ifp)
3556 ipf_main_softc_t *softc;
3559 ipf_state_softc_t *softs = softc->ipf_state_soft;
3563 if (softc->ipf_running <= 0)
3566 WRITE_ENTER(&softc->ipf_state);
3568 if (softc->ipf_running <= 0) {
3569 RWLOCK_EXIT(&softc->ipf_state);
3573 for (is = softs->ipf_state_list; is; is = is->is_next) {
3575 * Look up all the interface names in the state entry.
3577 for (i = 0; i < FR_NUM(is->is_ifp); i++) {
3578 if (ifp == NULL || ifp == is->is_ifp[i])
3579 is->is_ifp[i] = ipf_resolvenic(softc,
3584 RWLOCK_EXIT(&softc->ipf_state);
3588 /* ------------------------------------------------------------------------ */
3589 /* Function: ipf_state_del */
3590 /* Returns: int - 0 = deleted, else refernce count on active struct */
3591 /* Parameters: softc(I) - pointer to soft context main structure */
3592 /* is(I) - pointer to state structure to delete */
3593 /* why(I) - if not 0, log reason why it was deleted */
3594 /* Write Locks: ipf_state */
3596 /* Deletes a state entry from the enumerated list as well as the hash table */
3597 /* and timeout queue lists. Make adjustments to hash table statistics and */
3598 /* global counters as required. */
3599 /* ------------------------------------------------------------------------ */
3601 ipf_state_del(softc, is, why)
3602 ipf_main_softc_t *softc;
3606 ipf_state_softc_t *softs = softc->ipf_state_soft;
3611 * Since we want to delete this, remove it from the state table,
3612 * where it can be found & used, first.
3614 if (is->is_phnext != NULL) {
3615 *is->is_phnext = is->is_hnext;
3616 if (is->is_hnext != NULL)
3617 is->is_hnext->is_phnext = is->is_phnext;
3618 if (softs->ipf_state_table[is->is_hv] == NULL)
3619 softs->ipf_state_stats.iss_inuse--;
3620 softs->ipf_state_stats.iss_bucketlen[is->is_hv]--;
3622 is->is_phnext = NULL;
3623 is->is_hnext = NULL;
3628 * Because ipf_state_stats.iss_wild is a count of entries in the state
3629 * table that have wildcard flags set, only decerement it once
3632 if (is->is_flags & (SI_WILDP|SI_WILDA)) {
3633 if (!(is->is_flags & SI_CLONED)) {
3634 ATOMIC_DECL(softs->ipf_state_stats.iss_wild);
3636 is->is_flags &= ~(SI_WILDP|SI_WILDA);
3640 * Next, remove it from the timeout queue it is in.
3642 if (is->is_sti.tqe_ifq != NULL)
3643 ipf_deletequeueentry(&is->is_sti);
3646 * If it is still in use by something else, do not go any further,
3647 * but note that at this point it is now an orphan. How can this
3648 * be? ipf_state_flush() calls ipf_delete() directly because it wants
3649 * to empty the table out and if something has a hold on a state
3650 * entry (such as ipfstat), it'll do the deref path that'll bring
3651 * us back here to do the real delete & free.
3653 MUTEX_ENTER(&is->is_lock);
3654 if (is->is_me != NULL) {
3660 if (is->is_ref > 0) {
3664 MUTEX_EXIT(&is->is_lock);
3666 softs->ipf_state_stats.iss_orphan++;
3673 if (fr->fr_srctrack.ht_max_nodes != 0) {
3674 (void) ipf_ht_node_del(&fr->fr_srctrack,
3675 is->is_family, &is->is_src);
3679 ASSERT(is->is_ref == 0);
3680 MUTEX_EXIT(&is->is_lock);
3682 if (is->is_tqehead[0] != NULL) {
3683 if (ipf_deletetimeoutqueue(is->is_tqehead[0]) == 0)
3684 ipf_freetimeoutqueue(softc, is->is_tqehead[0]);
3686 if (is->is_tqehead[1] != NULL) {
3687 if (ipf_deletetimeoutqueue(is->is_tqehead[1]) == 0)
3688 ipf_freetimeoutqueue(softc, is->is_tqehead[1]);
3692 ipf_sync_del_state(softc->ipf_sync_soft, is->is_sync);
3695 * Now remove it from the linked list of known states
3697 if (is->is_pnext != NULL) {
3698 *is->is_pnext = is->is_next;
3700 if (is->is_next != NULL)
3701 is->is_next->is_pnext = is->is_pnext;
3703 is->is_pnext = NULL;
3707 if (softs->ipf_state_logging != 0 && why != 0)
3708 ipf_state_log(softc, is, why);
3710 if (is->is_p == IPPROTO_TCP)
3711 softs->ipf_state_stats.iss_fin++;
3713 softs->ipf_state_stats.iss_expire++;
3715 softs->ipf_state_stats.iss_orphan--;
3719 (void) ipf_derefrule(softc, &fr);
3722 softs->ipf_state_stats.iss_active_proto[is->is_p]--;
3724 MUTEX_DESTROY(&is->is_lock);
3726 softs->ipf_state_stats.iss_active--;
3732 /* ------------------------------------------------------------------------ */
3733 /* Function: ipf_state_expire */
3735 /* Parameters: softc(I) - pointer to soft context main structure */
3737 /* Slowly expire held state for thingslike UDP and ICMP. The algorithm */
3738 /* used here is to keep the queue sorted with the oldest things at the top */
3739 /* and the youngest at the bottom. So if the top one doesn't need to be */
3740 /* expired then neither will any under it. */
3741 /* ------------------------------------------------------------------------ */
3743 ipf_state_expire(softc)
3744 ipf_main_softc_t *softc;
3746 ipf_state_softc_t *softs = softc->ipf_state_soft;
3747 ipftq_t *ifq, *ifqnext;
3748 ipftqent_t *tqe, *tqn;
3753 WRITE_ENTER(&softc->ipf_state);
3754 for (ifq = softs->ipf_state_tcptq; ifq != NULL; ifq = ifq->ifq_next)
3755 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3756 if (tqe->tqe_die > softc->ipf_ticks)
3758 tqn = tqe->tqe_next;
3759 is = tqe->tqe_parent;
3760 ipf_state_del(softc, is, ISL_EXPIRE);
3763 for (ifq = softs->ipf_state_usertq; ifq != NULL; ifq = ifqnext) {
3764 ifqnext = ifq->ifq_next;
3766 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3767 if (tqe->tqe_die > softc->ipf_ticks)
3769 tqn = tqe->tqe_next;
3770 is = tqe->tqe_parent;
3771 ipf_state_del(softc, is, ISL_EXPIRE);
3775 for (ifq = softs->ipf_state_usertq; ifq != NULL; ifq = ifqnext) {
3776 ifqnext = ifq->ifq_next;
3778 if (((ifq->ifq_flags & IFQF_DELETE) != 0) &&
3779 (ifq->ifq_ref == 0)) {
3780 ipf_freetimeoutqueue(softc, ifq);
3784 if (softs->ipf_state_doflush) {
3785 (void) ipf_state_flush(softc, 2, 0);
3786 softs->ipf_state_doflush = 0;
3787 softs->ipf_state_wm_last = softc->ipf_ticks;
3790 RWLOCK_EXIT(&softc->ipf_state);
3795 /* ------------------------------------------------------------------------ */
3796 /* Function: ipf_state_flush */
3797 /* Returns: int - 0 == success, -1 == failure */
3798 /* Parameters: softc(I) - pointer to soft context main structure */
3799 /* which(I) - which flush action to perform */
3800 /* proto(I) - which protocol to flush (0 == ALL) */
3801 /* Write Locks: ipf_state */
3803 /* Flush state tables. Three actions currently defined: */
3804 /* which == 0 : flush all state table entries */
3805 /* which == 1 : flush TCP connections which have started to close but are */
3806 /* stuck for some reason. */
3807 /* which == 2 : flush TCP connections which have been idle for a long time, */
3808 /* starting at > 4 days idle and working back in successive half-*/
3809 /* days to at most 12 hours old. If this fails to free enough */
3810 /* slots then work backwards in half hour slots to 30 minutes. */
3811 /* If that too fails, then work backwards in 30 second intervals */
3812 /* for the last 30 minutes to at worst 30 seconds idle. */
3813 /* ------------------------------------------------------------------------ */
3815 ipf_state_flush(softc, which, proto)
3816 ipf_main_softc_t *softc;
3819 ipf_state_softc_t *softs = softc->ipf_state_soft;
3820 ipftqent_t *tqe, *tqn;
3821 ipstate_t *is, **isp;
3833 SBUMP(ipf_state_stats.iss_flush_all);
3835 * Style 0 flush removes everything...
3837 for (isp = &softs->ipf_state_list; ((is = *isp) != NULL); ) {
3838 if ((proto != 0) && (is->is_v != proto)) {
3842 if (ipf_state_del(softc, is, ISL_FLUSH) == 0)
3850 SBUMP(ipf_state_stats.iss_flush_closing);
3852 * Since we're only interested in things that are closing,
3853 * we can start with the appropriate timeout queue.
3855 for (ifq = softs->ipf_state_tcptq + IPF_TCPS_CLOSE_WAIT;
3856 ifq != NULL; ifq = ifq->ifq_next) {
3858 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3859 tqn = tqe->tqe_next;
3860 is = tqe->tqe_parent;
3861 if (is->is_p != IPPROTO_TCP)
3863 if (ipf_state_del(softc, is, ISL_FLUSH) == 0)
3869 * Also need to look through the user defined queues.
3871 for (ifq = softs->ipf_state_usertq; ifq != NULL;
3872 ifq = ifq->ifq_next) {
3873 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3874 tqn = tqe->tqe_next;
3875 is = tqe->tqe_parent;
3876 if (is->is_p != IPPROTO_TCP)
3879 if ((is->is_state[0] > IPF_TCPS_ESTABLISHED) &&
3880 (is->is_state[1] > IPF_TCPS_ESTABLISHED)) {
3881 if (ipf_state_del(softc, is,
3893 * Args 5-11 correspond to flushing those particular states
3894 * for TCP connections.
3896 case IPF_TCPS_CLOSE_WAIT :
3897 case IPF_TCPS_FIN_WAIT_1 :
3898 case IPF_TCPS_CLOSING :
3899 case IPF_TCPS_LAST_ACK :
3900 case IPF_TCPS_FIN_WAIT_2 :
3901 case IPF_TCPS_TIME_WAIT :
3902 case IPF_TCPS_CLOSED :
3903 SBUMP(ipf_state_stats.iss_flush_queue);
3904 tqn = softs->ipf_state_tcptq[which].ifq_head;
3905 while (tqn != NULL) {
3907 tqn = tqe->tqe_next;
3908 is = tqe->tqe_parent;
3909 if (ipf_state_del(softc, is, ISL_FLUSH) == 0)
3918 SBUMP(ipf_state_stats.iss_flush_state);
3920 * Take a large arbitrary number to mean the number of seconds
3921 * for which which consider to be the maximum value we'll allow
3922 * the expiration to be.
3924 which = IPF_TTLVAL(which);
3925 for (isp = &softs->ipf_state_list; ((is = *isp) != NULL); ) {
3926 if ((proto == 0) || (is->is_v == proto)) {
3927 if (softc->ipf_ticks - is->is_touched > which) {
3928 if (ipf_state_del(softc, is,
3945 SBUMP(ipf_state_stats.iss_flush_timeout);
3947 * Asked to remove inactive entries because the table is full, try
3948 * again, 3 times, if first attempt failed with a different criteria
3949 * each time. The order tried in must be in decreasing age.
3950 * Another alternative is to implement random drop and drop N entries
3951 * at random until N have been freed up.
3953 if (softc->ipf_ticks - softs->ipf_state_wm_last >
3954 softs->ipf_state_wm_freq) {
3955 removed = ipf_queueflush(softc, ipf_state_flush_entry,
3956 softs->ipf_state_tcptq,
3957 softs->ipf_state_usertq,
3958 &softs->ipf_state_stats.iss_active,
3959 softs->ipf_state_size,
3960 softs->ipf_state_wm_low);
3961 softs->ipf_state_wm_last = softc->ipf_ticks;
3969 /* ------------------------------------------------------------------------ */
3970 /* Function: ipf_state_flush_entry */
3971 /* Returns: int - 0 = entry deleted, else not deleted */
3972 /* Parameters: softc(I) - pointer to soft context main structure */
3973 /* entry(I) - pointer to state structure to delete */
3974 /* Write Locks: ipf_state */
3976 /* This function is a stepping stone between ipf_queueflush() and */
3977 /* ipf_state_del(). It is used so we can provide a uniform interface via */
3978 /* the ipf_queueflush() function. */
3979 /* ------------------------------------------------------------------------ */
3981 ipf_state_flush_entry(softc, entry)
3982 ipf_main_softc_t *softc;
3985 return ipf_state_del(softc, entry, ISL_FLUSH);
3989 /* ------------------------------------------------------------------------ */
3990 /* Function: ipf_tcp_age */
3991 /* Returns: int - 1 == state transition made, 0 == no change (rejected) */
3992 /* Parameters: tqe(I) - pointer to timeout queue information */
3993 /* fin(I) - pointer to packet information */
3994 /* tqtab(I) - TCP timeout queue table this is in */
3995 /* flags(I) - flags from state/NAT entry */
3996 /* ok(I) - can we advance state */
3998 /* Rewritten by Arjan de Vet <Arjan.deVet@adv.iae.nl>, 2000-07-29: */
4000 /* - (try to) base state transitions on real evidence only, */
4001 /* i.e. packets that are sent and have been received by ipfilter; */
4002 /* diagram 18.12 of TCP/IP volume 1 by W. Richard Stevens was used. */
4004 /* - deal with half-closed connections correctly; */
4006 /* - store the state of the source in state[0] such that ipfstat */
4007 /* displays the state as source/dest instead of dest/source; the calls */
4008 /* to ipf_tcp_age have been changed accordingly. */
4010 /* Internal Parameters: */
4012 /* state[0] = state of source (host that initiated connection) */
4013 /* state[1] = state of dest (host that accepted the connection) */
4015 /* dir == 0 : a packet from source to dest */
4016 /* dir == 1 : a packet from dest to source */
4018 /* A typical procession for a connection is as follows: */
4020 /* +--------------+-------------------+ */
4021 /* | Side '0' | Side '1' | */
4022 /* +--------------+-------------------+ */
4023 /* | 0 -> 1 (SYN) | | */
4024 /* | | 0 -> 2 (SYN-ACK) | */
4025 /* | 1 -> 3 (ACK) | | */
4026 /* | | 2 -> 4 (ACK-PUSH) | */
4027 /* | 3 -> 4 (ACK) | | */
4029 /* | | 4 -> 6 (FIN-ACK) | */
4030 /* | 4 -> 5 (ACK) | | */
4031 /* | | 6 -> 6 (ACK-PUSH) | */
4032 /* | 5 -> 5 (ACK) | | */
4033 /* | 5 -> 8 (FIN) | | */
4034 /* | | 6 -> 10 (ACK) | */
4035 /* +--------------+-------------------+ */
4037 /* Locking: it is assumed that the parent of the tqe structure is locked. */
4038 /* ------------------------------------------------------------------------ */
4040 ipf_tcp_age(tqe, fin, tqtab, flags, ok)
4046 ipf_main_softc_t *softc = fin->fin_main_soft;
4047 int dlen, ostate, nstate, rval, dir;
4055 tcpflags = tcp->th_flags;
4056 dlen = fin->fin_dlen - (TCP_OFF(tcp) << 2);
4057 ostate = tqe->tqe_state[1 - dir];
4058 nstate = tqe->tqe_state[dir];
4060 if (tcpflags & TH_RST) {
4061 if (!(tcpflags & TH_PUSH) && !dlen)
4062 nstate = IPF_TCPS_CLOSED;
4064 nstate = IPF_TCPS_CLOSE_WAIT;
4066 if (ostate <= IPF_TCPS_ESTABLISHED) {
4067 tqe->tqe_state[1 - dir] = IPF_TCPS_CLOSE_WAIT;
4073 case IPF_TCPS_LISTEN: /* 0 */
4074 if ((tcpflags & TH_OPENING) == TH_OPENING) {
4076 * 'dir' received an S and sends SA in
4077 * response, LISTEN -> SYN_RECEIVED
4079 nstate = IPF_TCPS_SYN_RECEIVED;
4081 } else if ((tcpflags & TH_OPENING) == TH_SYN) {
4082 /* 'dir' sent S, LISTEN -> SYN_SENT */
4083 nstate = IPF_TCPS_SYN_SENT;
4087 * the next piece of code makes it possible to get
4088 * already established connections into the state table
4089 * after a restart or reload of the filter rules; this
4090 * does not work when a strict 'flags S keep state' is
4091 * used for tcp connections of course
4093 if (((flags & IS_TCPFSM) == 0) &&
4094 ((tcpflags & TH_ACKMASK) == TH_ACK)) {
4096 * we saw an A, guess 'dir' is in ESTABLISHED
4101 case IPF_TCPS_LISTEN :
4102 case IPF_TCPS_SYN_RECEIVED :
4103 nstate = IPF_TCPS_HALF_ESTAB;
4106 case IPF_TCPS_HALF_ESTAB :
4107 case IPF_TCPS_ESTABLISHED :
4108 nstate = IPF_TCPS_ESTABLISHED;
4116 * TODO: besides regular ACK packets we can have other
4117 * packets as well; it is yet to be determined how we
4118 * should initialize the states in those cases
4122 case IPF_TCPS_SYN_SENT: /* 1 */
4123 if ((tcpflags & ~(TH_ECN|TH_CWR)) == TH_SYN) {
4125 * A retransmitted SYN packet. We do not reset
4126 * the timeout here to ipf_tcptimeout because a
4127 * connection connect timeout does not renew
4128 * after every packet that is sent. We need to
4129 * set rval so as to indicate the packet has
4130 * passed the check for its flags being valid
4131 * in the TCP FSM. Setting rval to 2 has the
4132 * result of not resetting the timeout.
4135 } else if ((tcpflags & (TH_SYN|TH_FIN|TH_ACK)) ==
4138 * we see an A from 'dir' which is in SYN_SENT
4139 * state: 'dir' sent an A in response to an SA
4140 * which it received, SYN_SENT -> ESTABLISHED
4142 nstate = IPF_TCPS_ESTABLISHED;
4144 } else if (tcpflags & TH_FIN) {
4146 * we see an F from 'dir' which is in SYN_SENT
4147 * state and wants to close its side of the
4148 * connection; SYN_SENT -> FIN_WAIT_1
4150 nstate = IPF_TCPS_FIN_WAIT_1;
4152 } else if ((tcpflags & TH_OPENING) == TH_OPENING) {
4154 * we see an SA from 'dir' which is already in
4155 * SYN_SENT state, this means we have a
4156 * simultaneous open; SYN_SENT -> SYN_RECEIVED
4158 nstate = IPF_TCPS_SYN_RECEIVED;
4163 case IPF_TCPS_SYN_RECEIVED: /* 2 */
4164 if ((tcpflags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) {
4166 * we see an A from 'dir' which was in
4167 * SYN_RECEIVED state so it must now be in
4168 * established state, SYN_RECEIVED ->
4171 nstate = IPF_TCPS_ESTABLISHED;
4173 } else if ((tcpflags & ~(TH_ECN|TH_CWR)) ==
4176 * We see an SA from 'dir' which is already in
4177 * SYN_RECEIVED state.
4180 } else if (tcpflags & TH_FIN) {
4182 * we see an F from 'dir' which is in
4183 * SYN_RECEIVED state and wants to close its
4184 * side of the connection; SYN_RECEIVED ->
4187 nstate = IPF_TCPS_FIN_WAIT_1;
4192 case IPF_TCPS_HALF_ESTAB: /* 3 */
4193 if (tcpflags & TH_FIN) {
4194 nstate = IPF_TCPS_FIN_WAIT_1;
4196 } else if ((tcpflags & TH_ACKMASK) == TH_ACK) {
4198 * If we've picked up a connection in mid
4199 * flight, we could be looking at a follow on
4200 * packet from the same direction as the one
4201 * that created this state. Recognise it but
4202 * do not advance the entire connection's
4207 case IPF_TCPS_LISTEN :
4208 case IPF_TCPS_SYN_SENT :
4209 case IPF_TCPS_SYN_RECEIVED :
4212 case IPF_TCPS_HALF_ESTAB :
4213 case IPF_TCPS_ESTABLISHED :
4214 nstate = IPF_TCPS_ESTABLISHED;
4223 case IPF_TCPS_ESTABLISHED: /* 4 */
4225 if (tcpflags & TH_FIN) {
4227 * 'dir' closed its side of the connection;
4228 * this gives us a half-closed connection;
4229 * ESTABLISHED -> FIN_WAIT_1
4231 if (ostate == IPF_TCPS_FIN_WAIT_1) {
4232 nstate = IPF_TCPS_CLOSING;
4234 nstate = IPF_TCPS_FIN_WAIT_1;
4236 } else if (tcpflags & TH_ACK) {
4238 * an ACK, should we exclude other flags here?
4240 if (ostate == IPF_TCPS_FIN_WAIT_1) {
4242 * We know the other side did an active
4243 * close, so we are ACKing the recvd
4244 * FIN packet (does the window matching
4245 * code guarantee this?) and go into
4246 * CLOSE_WAIT state; this gives us a
4247 * half-closed connection
4249 nstate = IPF_TCPS_CLOSE_WAIT;
4250 } else if (ostate < IPF_TCPS_CLOSE_WAIT) {
4252 * still a fully established
4253 * connection reset timeout
4255 nstate = IPF_TCPS_ESTABLISHED;
4260 case IPF_TCPS_CLOSE_WAIT: /* 5 */
4262 if (tcpflags & TH_FIN) {
4264 * application closed and 'dir' sent a FIN,
4265 * we're now going into LAST_ACK state
4267 nstate = IPF_TCPS_LAST_ACK;
4270 * we remain in CLOSE_WAIT because the other
4271 * side has closed already and we did not
4272 * close our side yet; reset timeout
4274 nstate = IPF_TCPS_CLOSE_WAIT;
4278 case IPF_TCPS_FIN_WAIT_1: /* 6 */
4280 if ((tcpflags & TH_ACK) &&
4281 ostate > IPF_TCPS_CLOSE_WAIT) {
4283 * if the other side is not active anymore
4284 * it has sent us a FIN packet that we are
4285 * ack'ing now with an ACK; this means both
4286 * sides have now closed the connection and
4287 * we go into TIME_WAIT
4290 * XXX: how do we know we really are ACKing
4291 * the FIN packet here? does the window code
4294 nstate = IPF_TCPS_LAST_ACK;
4297 * we closed our side of the connection
4298 * already but the other side is still active
4299 * (ESTABLISHED/CLOSE_WAIT); continue with
4300 * this half-closed connection
4302 nstate = IPF_TCPS_FIN_WAIT_1;
4306 case IPF_TCPS_CLOSING: /* 7 */
4307 if ((tcpflags & (TH_FIN|TH_ACK)) == TH_ACK) {
4308 nstate = IPF_TCPS_TIME_WAIT;
4313 case IPF_TCPS_LAST_ACK: /* 8 */
4314 if (tcpflags & TH_ACK) {
4318 * we cannot detect when we go out of LAST_ACK state
4319 * to CLOSED because that is based on the reception
4320 * of ACK packets; ipfilter can only detect that a
4321 * packet has been sent by a host
4325 case IPF_TCPS_FIN_WAIT_2: /* 9 */
4329 case IPF_TCPS_TIME_WAIT: /* 10 */
4330 /* we're in 2MSL timeout now */
4331 if (ostate == IPF_TCPS_LAST_ACK) {
4332 nstate = IPF_TCPS_CLOSED;
4339 case IPF_TCPS_CLOSED: /* 11 */
4344 #if !defined(_KERNEL)
4352 * If rval == 2 then do not update the queue position, but treat the
4353 * packet as being ok.
4357 else if (rval == 1) {
4359 tqe->tqe_state[dir] = nstate;
4360 if ((tqe->tqe_flags & TQE_RULEBASED) == 0)
4361 ipf_movequeue(softc->ipf_ticks, tqe, tqe->tqe_ifq,
4369 /* ------------------------------------------------------------------------ */
4370 /* Function: ipf_state_log */
4372 /* Parameters: softc(I) - pointer to soft context main structure */
4373 /* is(I) - pointer to state structure */
4374 /* type(I) - type of log entry to create */
4376 /* Creates a state table log entry using the state structure and type info. */
4377 /* passed in. Log packet/byte counts, source/destination address and other */
4378 /* protocol specific information. */
4379 /* ------------------------------------------------------------------------ */
4381 ipf_state_log(softc, is, type)
4382 ipf_main_softc_t *softc;
4393 * Copy information out of the ipstate_t structure and into the
4394 * structure used for logging.
4396 ipsl.isl_type = type;
4397 ipsl.isl_pkts[0] = is->is_pkts[0] + is->is_icmppkts[0];
4398 ipsl.isl_bytes[0] = is->is_bytes[0];
4399 ipsl.isl_pkts[1] = is->is_pkts[1] + is->is_icmppkts[1];
4400 ipsl.isl_bytes[1] = is->is_bytes[1];
4401 ipsl.isl_pkts[2] = is->is_pkts[2] + is->is_icmppkts[2];
4402 ipsl.isl_bytes[2] = is->is_bytes[2];
4403 ipsl.isl_pkts[3] = is->is_pkts[3] + is->is_icmppkts[3];
4404 ipsl.isl_bytes[3] = is->is_bytes[3];
4405 ipsl.isl_src = is->is_src;
4406 ipsl.isl_dst = is->is_dst;
4407 ipsl.isl_p = is->is_p;
4408 ipsl.isl_v = is->is_v;
4409 ipsl.isl_flags = is->is_flags;
4410 ipsl.isl_tag = is->is_tag;
4411 ipsl.isl_rulen = is->is_rulen;
4412 (void) strncpy(ipsl.isl_group, is->is_group, FR_GROUPLEN);
4414 if (ipsl.isl_p == IPPROTO_TCP || ipsl.isl_p == IPPROTO_UDP) {
4415 ipsl.isl_sport = is->is_sport;
4416 ipsl.isl_dport = is->is_dport;
4417 if (ipsl.isl_p == IPPROTO_TCP) {
4418 ipsl.isl_state[0] = is->is_state[0];
4419 ipsl.isl_state[1] = is->is_state[1];
4421 } else if (ipsl.isl_p == IPPROTO_ICMP) {
4422 ipsl.isl_itype = is->is_icmp.ici_type;
4423 } else if (ipsl.isl_p == IPPROTO_ICMPV6) {
4424 ipsl.isl_itype = is->is_icmp.ici_type;
4426 ipsl.isl_ps.isl_filler[0] = 0;
4427 ipsl.isl_ps.isl_filler[1] = 0;
4431 sizes[0] = sizeof(ipsl);
4434 (void) ipf_log_items(softc, IPL_LOGSTATE, NULL, items, sizes, types, 1);
4440 /* ------------------------------------------------------------------------ */
4441 /* Function: ipf_checkicmp6matchingstate */
4442 /* Returns: ipstate_t* - NULL == no match found, */
4443 /* else pointer to matching state entry */
4444 /* Parameters: fin(I) - pointer to packet information */
4445 /* Locks: NULL == no locks, else Read Lock on ipf_state */
4447 /* If we've got an ICMPv6 error message, using the information stored in */
4448 /* the ICMPv6 packet, look for a matching state table entry. */
4449 /* ------------------------------------------------------------------------ */
4451 ipf_checkicmp6matchingstate(fin)
4454 ipf_main_softc_t *softc = fin->fin_main_soft;
4455 ipf_state_softc_t *softs = softc->ipf_state_soft;
4456 struct icmp6_hdr *ic6, *oic;
4457 ipstate_t *is, **isp;
4458 u_short sport, dport;
4470 * Does it at least have the return (basic) IP header ?
4471 * Is it an actual recognised ICMP error type?
4472 * Only a basic IP header (no options) should be with
4473 * an ICMP error header.
4475 if ((fin->fin_v != 6) || (fin->fin_plen < ICMP6ERR_MINPKTLEN) ||
4476 !(fin->fin_flx & FI_ICMPERR)) {
4477 SBUMPD(ipf_state_stats, iss_icmp_bad);
4482 type = ic6->icmp6_type;
4484 oip6 = (ip6_t *)((char *)ic6 + ICMPERR_ICMPHLEN);
4485 if (fin->fin_plen < sizeof(*oip6)) {
4486 SBUMPD(ipf_state_stats, iss_icmp_short);
4490 bcopy((char *)fin, (char *)&ofin, sizeof(*fin));
4492 ofin.fin_ifp = fin->fin_ifp;
4493 ofin.fin_out = !fin->fin_out;
4494 ofin.fin_m = NULL; /* if dereferenced, panic XXX */
4495 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */
4498 * We make a fin entry to be able to feed it to
4499 * matchsrcdst. Note that not all fields are necessary
4500 * but this is the cleanest way. Note further we fill
4501 * in fin_mp such that if someone uses it we'll get
4502 * a kernel panic. ipf_matchsrcdst does not use this.
4504 * watch out here, as ip is in host order and oip6 in network
4505 * order. Any change we make must be undone afterwards.
4507 savelen = oip6->ip6_plen;
4508 oip6->ip6_plen = htons(fin->fin_dlen - ICMPERR_ICMPHLEN);
4509 ofin.fin_flx = FI_NOCKSUM;
4510 ofin.fin_ip = (ip_t *)oip6;
4511 (void) ipf_makefrip(sizeof(*oip6), (ip_t *)oip6, &ofin);
4512 ofin.fin_flx &= ~(FI_BAD|FI_SHORT);
4513 oip6->ip6_plen = savelen;
4517 * an ICMP error can never generate an ICMP error in response.
4519 if (ofin.fin_flx & FI_ICMPERR) {
4520 DT1(iss_icmp6_icmperr, fr_info_t *, &ofin);
4521 SBUMP(ipf_state_stats.iss_icmp6_icmperr);
4525 if (oip6->ip6_nxt == IPPROTO_ICMPV6) {
4528 * an ICMP error can only be generated as a result of an
4529 * ICMP query, not as the response on an ICMP error
4531 * XXX theoretically ICMP_ECHOREP and the other reply's are
4532 * ICMP query's as well, but adding them here seems strange XXX
4534 if (!(oic->icmp6_type & ICMP6_INFOMSG_MASK)) {
4535 DT1(iss_icmp6_notinfo, fr_info_t *, &ofin);
4536 SBUMP(ipf_state_stats.iss_icmp6_notinfo);
4541 * perform a lookup of the ICMP packet in the state table
4543 hv = (pr = oip6->ip6_nxt);
4544 src.in6 = oip6->ip6_src;
4545 hv += src.in4.s_addr;
4546 dst.in6 = oip6->ip6_dst;
4547 hv += dst.in4.s_addr;
4548 hv += oic->icmp6_id;
4549 hv += oic->icmp6_seq;
4550 hv = DOUBLE_HASH(hv);
4552 READ_ENTER(&softc->ipf_state);
4553 for (isp = &softs->ipf_state_table[hv];
4554 ((is = *isp) != NULL); ) {
4556 isp = &is->is_hnext;
4557 if ((is->is_p == pr) &&
4558 !(is->is_pass & FR_NOICMPERR) &&
4559 (oic->icmp6_id == ic->ici_id) &&
4560 (oic->icmp6_seq == ic->ici_seq) &&
4561 (is = ipf_matchsrcdst(&ofin, is, &src,
4562 &dst, NULL, FI_ICMPCMP))) {
4564 * in the state table ICMP query's are stored
4565 * with the type of the corresponding ICMP
4566 * response. Correct here
4568 if (((ic->ici_type == ICMP6_ECHO_REPLY) &&
4569 (oic->icmp6_type == ICMP6_ECHO_REQUEST)) ||
4570 (ic->ici_type - 1 == oic->icmp6_type )) {
4571 if (!ipf_allowstateicmp(fin, is, &src))
4576 RWLOCK_EXIT(&softc->ipf_state);
4577 SBUMPD(ipf_state_stats, iss_icmp6_miss);
4581 hv = (pr = oip6->ip6_nxt);
4582 src.in6 = oip6->ip6_src;
4587 dst.in6 = oip6->ip6_dst;
4595 switch (oip6->ip6_nxt)
4599 tcp = (tcphdr_t *)(oip6 + 1);
4600 dport = tcp->th_dport;
4601 sport = tcp->th_sport;
4606 case IPPROTO_ICMPV6 :
4607 oic = (struct icmp6_hdr *)(oip6 + 1);
4608 hv += oic->icmp6_id;
4609 hv += oic->icmp6_seq;
4616 hv = DOUBLE_HASH(hv);
4618 READ_ENTER(&softc->ipf_state);
4619 for (isp = &softs->ipf_state_table[hv]; ((is = *isp) != NULL); ) {
4620 isp = &is->is_hnext;
4622 * Only allow this icmp though if the
4623 * encapsulated packet was allowed through the
4624 * other way around. Note that the minimal amount
4625 * of info present does not allow for checking against
4626 * tcp internals such as seq and ack numbers.
4628 if ((is->is_p != pr) || (is->is_v != 6) ||
4629 (is->is_pass & FR_NOICMPERR))
4631 is = ipf_matchsrcdst(&ofin, is, &src, &dst, tcp, FI_ICMPCMP);
4632 if ((is != NULL) && (ipf_allowstateicmp(fin, is, &src) == 0))
4635 RWLOCK_EXIT(&softc->ipf_state);
4636 SBUMPD(ipf_state_stats, iss_icmp_miss);
4642 /* ------------------------------------------------------------------------ */
4643 /* Function: ipf_sttab_init */
4645 /* Parameters: softc(I) - pointer to soft context main structure */
4646 /* tqp(I) - pointer to an array of timeout queues for TCP */
4648 /* Initialise the array of timeout queues for TCP. */
4649 /* ------------------------------------------------------------------------ */
4651 ipf_sttab_init(softc, tqp)
4652 ipf_main_softc_t *softc;
4657 for (i = IPF_TCP_NSTATES - 1; i >= 0; i--) {
4658 IPFTQ_INIT(&tqp[i], 0, "ipftq tcp tab");
4659 tqp[i].ifq_next = tqp + i + 1;
4661 tqp[IPF_TCP_NSTATES - 1].ifq_next = NULL;
4662 tqp[IPF_TCPS_CLOSED].ifq_ttl = softc->ipf_tcpclosed;
4663 tqp[IPF_TCPS_LISTEN].ifq_ttl = softc->ipf_tcptimeout;
4664 tqp[IPF_TCPS_SYN_SENT].ifq_ttl = softc->ipf_tcpsynsent;
4665 tqp[IPF_TCPS_SYN_RECEIVED].ifq_ttl = softc->ipf_tcpsynrecv;
4666 tqp[IPF_TCPS_ESTABLISHED].ifq_ttl = softc->ipf_tcpidletimeout;
4667 tqp[IPF_TCPS_CLOSE_WAIT].ifq_ttl = softc->ipf_tcphalfclosed;
4668 tqp[IPF_TCPS_FIN_WAIT_1].ifq_ttl = softc->ipf_tcphalfclosed;
4669 tqp[IPF_TCPS_CLOSING].ifq_ttl = softc->ipf_tcptimeout;
4670 tqp[IPF_TCPS_LAST_ACK].ifq_ttl = softc->ipf_tcplastack;
4671 tqp[IPF_TCPS_FIN_WAIT_2].ifq_ttl = softc->ipf_tcpclosewait;
4672 tqp[IPF_TCPS_TIME_WAIT].ifq_ttl = softc->ipf_tcptimewait;
4673 tqp[IPF_TCPS_HALF_ESTAB].ifq_ttl = softc->ipf_tcptimeout;
4677 /* ------------------------------------------------------------------------ */
4678 /* Function: ipf_sttab_destroy */
4680 /* Parameters: tqp(I) - pointer to an array of timeout queues for TCP */
4682 /* Do whatever is necessary to "destroy" each of the entries in the array */
4683 /* of timeout queues for TCP. */
4684 /* ------------------------------------------------------------------------ */
4686 ipf_sttab_destroy(tqp)
4691 for (i = IPF_TCP_NSTATES - 1; i >= 0; i--)
4692 MUTEX_DESTROY(&tqp[i].ifq_lock);
4696 /* ------------------------------------------------------------------------ */
4697 /* Function: ipf_state_deref */
4699 /* Parameters: softc(I) - pointer to soft context main structure */
4700 /* isp(I) - pointer to pointer to state table entry */
4702 /* Decrement the reference counter for this state table entry and free it */
4703 /* if there are no more things using it. */
4705 /* This function is only called when cleaning up after increasing is_ref by */
4706 /* one earlier in the 'code path' so if is_ref is 1 when entering, we do */
4707 /* have an orphan, otherwise not. However there is a possible race between */
4708 /* the entry being deleted via flushing with an ioctl call (that calls the */
4709 /* delete function directly) and the tail end of packet processing so we */
4710 /* need to grab is_lock before doing the check to synchronise the two code */
4713 /* When operating in userland (ipftest), we have no timers to clear a state */
4714 /* entry. Therefore, we make a few simple tests before deleting an entry */
4715 /* outright. We compare states on each side looking for a combination of */
4716 /* TIME_WAIT (should really be FIN_WAIT_2?) and LAST_ACK. Then we factor */
4717 /* in packet direction with the interface list to make sure we don't */
4718 /* prematurely delete an entry on a final inbound packet that's we're also */
4719 /* supposed to route elsewhere. */
4721 /* Internal parameters: */
4722 /* state[0] = state of source (host that initiated connection) */
4723 /* state[1] = state of dest (host that accepted the connection) */
4725 /* dir == 0 : a packet from source to dest */
4726 /* dir == 1 : a packet from dest to source */
4727 /* ------------------------------------------------------------------------ */
4729 ipf_state_deref(softc, isp)
4730 ipf_main_softc_t *softc;
4733 ipstate_t *is = *isp;
4738 MUTEX_ENTER(&is->is_lock);
4739 if (is->is_ref > 1) {
4741 MUTEX_EXIT(&is->is_lock);
4743 if ((is->is_sti.tqe_state[0] > IPF_TCPS_ESTABLISHED) ||
4744 (is->is_sti.tqe_state[1] > IPF_TCPS_ESTABLISHED)) {
4745 ipf_state_del(softc, is, ISL_EXPIRE);
4750 MUTEX_EXIT(&is->is_lock);
4752 WRITE_ENTER(&softc->ipf_state);
4753 ipf_state_del(softc, is, ISL_ORPHAN);
4754 RWLOCK_EXIT(&softc->ipf_state);
4758 /* ------------------------------------------------------------------------ */
4759 /* Function: ipf_state_setqueue */
4761 /* Parameters: softc(I) - pointer to soft context main structure */
4762 /* is(I) - pointer to state structure */
4763 /* rev(I) - forward(0) or reverse(1) direction */
4764 /* Locks: ipf_state (read or write) */
4766 /* Put the state entry on its default queue entry, using rev as a helped in */
4767 /* determining which queue it should be placed on. */
4768 /* ------------------------------------------------------------------------ */
4770 ipf_state_setqueue(softc, is, rev)
4771 ipf_main_softc_t *softc;
4775 ipf_state_softc_t *softs = softc->ipf_state_soft;
4776 ipftq_t *oifq, *nifq;
4778 if ((is->is_sti.tqe_flags & TQE_RULEBASED) != 0)
4779 nifq = is->is_tqehead[rev];
4787 case IPPROTO_ICMPV6 :
4789 nifq = &softs->ipf_state_icmpacktq;
4791 nifq = &softs->ipf_state_icmptq;
4796 nifq = &softs->ipf_state_icmpacktq;
4798 nifq = &softs->ipf_state_icmptq;
4801 nifq = softs->ipf_state_tcptq + is->is_state[rev];
4806 nifq = &softs->ipf_state_udpacktq;
4808 nifq = &softs->ipf_state_udptq;
4812 nifq = &softs->ipf_state_iptq;
4817 oifq = is->is_sti.tqe_ifq;
4819 * If it's currently on a timeout queue, move it from one queue to
4820 * another, else put it on the end of the newly determined queue.
4823 ipf_movequeue(softc->ipf_ticks, &is->is_sti, oifq, nifq);
4825 ipf_queueappend(softc->ipf_ticks, &is->is_sti, nifq, is);
4830 /* ------------------------------------------------------------------------ */
4831 /* Function: ipf_state_iter */
4832 /* Returns: int - 0 == success, else error */
4833 /* Parameters: softc(I) - pointer to main soft context */
4834 /* token(I) - pointer to ipftoken structure */
4835 /* itp(I) - pointer to ipfgeniter structure */
4836 /* obj(I) - pointer to data description structure */
4838 /* This function handles the SIOCGENITER ioctl for the state tables and */
4839 /* walks through the list of entries in the state table list (softs->ipf_state_list.) */
4840 /* ------------------------------------------------------------------------ */
4842 ipf_state_iter(softc, token, itp, obj)
4843 ipf_main_softc_t *softc;
4848 ipf_state_softc_t *softs = softc->ipf_state_soft;
4849 ipstate_t *is, *next, zero;
4852 if (itp->igi_data == NULL) {
4857 if (itp->igi_nitems < 1) {
4862 if (itp->igi_type != IPFGENITER_STATE) {
4867 is = token->ipt_data;
4868 if (is == (void *)-1) {
4874 obj->ipfo_type = IPFOBJ_IPSTATE;
4875 obj->ipfo_size = sizeof(ipstate_t);
4877 READ_ENTER(&softc->ipf_state);
4879 is = token->ipt_data;
4881 next = softs->ipf_state_list;
4887 * If we find a state entry to use, bump its reference count so that
4888 * it can be used for is_next when we come back.
4891 MUTEX_ENTER(&next->is_lock);
4893 MUTEX_EXIT(&next->is_lock);
4894 token->ipt_data = next;
4896 bzero(&zero, sizeof(zero));
4898 token->ipt_data = NULL;
4900 if (next->is_next == NULL)
4901 ipf_token_mark_complete(token);
4903 RWLOCK_EXIT(&softc->ipf_state);
4905 obj->ipfo_ptr = itp->igi_data;
4906 error = ipf_outobjk(softc, obj, next);
4908 ipf_state_deref(softc, &is);
4914 /* ------------------------------------------------------------------------ */
4915 /* Function: ipf_state_gettable */
4916 /* Returns: int - 0 = success, else error */
4917 /* Parameters: softc(I) - pointer to main soft context */
4918 /* softs(I) - pointer to state context structure */
4919 /* data(I) - pointer to ioctl data */
4921 /* This function handles ioctl requests for tables of state information. */
4922 /* At present the only table it deals with is the hash bucket statistics. */
4923 /* ------------------------------------------------------------------------ */
4925 ipf_state_gettable(softc, softs, data)
4926 ipf_main_softc_t *softc;
4927 ipf_state_softc_t *softs;
4933 error = ipf_inobj(softc, data, NULL, &table, IPFOBJ_GTABLE);
4937 if (table.ita_type != IPFTABLE_BUCKETS) {
4942 error = COPYOUT(softs->ipf_state_stats.iss_bucketlen, table.ita_table,
4943 softs->ipf_state_size * sizeof(u_int));
4952 /* ------------------------------------------------------------------------ */
4953 /* Function: ipf_state_setpending */
4955 /* Parameters: softc(I) - pointer to main soft context */
4956 /* is(I) - pointer to state structure */
4957 /* Locks: ipf_state (read or write) */
4959 /* Put the state entry on to the pending queue - this queue has a very */
4960 /* short lifetime where items are put that can't be deleted straight away */
4961 /* because of locking issues but we want to delete them ASAP, anyway. */
4962 /* ------------------------------------------------------------------------ */
4964 ipf_state_setpending(softc, is)
4965 ipf_main_softc_t *softc;
4968 ipf_state_softc_t *softs = softc->ipf_state_soft;
4971 oifq = is->is_sti.tqe_ifq;
4973 ipf_movequeue(softc->ipf_ticks, &is->is_sti, oifq,
4974 &softs->ipf_state_pending);
4976 ipf_queueappend(softc->ipf_ticks, &is->is_sti,
4977 &softs->ipf_state_pending, is);
4979 MUTEX_ENTER(&is->is_lock);
4980 if (is->is_me != NULL) {
4985 MUTEX_EXIT(&is->is_lock);
4989 /* ------------------------------------------------------------------------ */
4990 /* Function: ipf_state_matchflush */
4992 /* Parameters: softc(I) - pointer to main soft context */
4993 /* data(I) - pointer to state structure */
4994 /* Locks: ipf_state (read or write) */
4996 /* Flush all entries from the list of state entries that match the */
4997 /* properties in the array loaded. */
4998 /* ------------------------------------------------------------------------ */
5000 ipf_state_matchflush(softc, data)
5001 ipf_main_softc_t *softc;
5004 ipf_state_softc_t *softs = softc->ipf_state_soft;
5005 int *array, flushed, error;
5006 ipstate_t *state, *statenext;
5009 error = ipf_matcharray_load(softc, data, &obj, &array);
5015 for (state = softs->ipf_state_list; state != NULL; state = statenext) {
5016 statenext = state->is_next;
5017 if (ipf_state_matcharray(state, array, softc->ipf_ticks) == 0) {
5018 ipf_state_del(softc, state, ISL_FLUSH);
5023 obj.ipfo_retval = flushed;
5024 error = BCOPYOUT(&obj, data, sizeof(obj));
5026 KFREES(array, array[0] * sizeof(*array));
5032 /* ------------------------------------------------------------------------ */
5033 /* Function: ipf_state_matcharray */
5034 /* Returns: int - 0 = no match, 1 = match */
5035 /* Parameters: state(I) - pointer to state structure */
5036 /* array(I) - pointer to ipf matching expression */
5037 /* ticks(I) - current value of ipfilter tick timer */
5038 /* Locks: ipf_state (read or write) */
5040 /* Compare a state entry with the match array passed in and return a value */
5041 /* to indicate whether or not the matching was successful. */
5042 /* ------------------------------------------------------------------------ */
5044 ipf_state_matcharray(state, array, ticks)
5049 int i, n, *x, rv, p;
5056 for (; n > 0; x += 3 + x[3], rv = 0) {
5059 if (x[0] == IPF_EXP_END)
5063 * If we need to match the protocol and that doesn't match,
5064 * don't even both with the instruction array.
5066 p = e->ipfe_cmd >> 16;
5067 if ((p != 0) && (p != state->is_p))
5070 switch (e->ipfe_cmd)
5072 case IPF_EXP_IP_PR :
5073 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5074 rv |= (state->is_p == e->ipfe_arg0[i]);
5078 case IPF_EXP_IP_SRCADDR :
5079 if (state->is_v != 4)
5081 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5082 rv |= ((state->is_saddr &
5083 e->ipfe_arg0[i * 2 + 1]) ==
5084 e->ipfe_arg0[i * 2]);
5088 case IPF_EXP_IP_DSTADDR :
5089 if (state->is_v != 4)
5091 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5092 rv |= ((state->is_daddr &
5093 e->ipfe_arg0[i * 2 + 1]) ==
5094 e->ipfe_arg0[i * 2]);
5098 case IPF_EXP_IP_ADDR :
5099 if (state->is_v != 4)
5101 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5102 rv |= ((state->is_saddr &
5103 e->ipfe_arg0[i * 2 + 1]) ==
5104 e->ipfe_arg0[i * 2]) ||
5106 e->ipfe_arg0[i * 2 + 1]) ==
5107 e->ipfe_arg0[i * 2]);
5112 case IPF_EXP_IP6_SRCADDR :
5113 if (state->is_v != 6)
5115 for (i = 0; !rv && i < x[3]; i++) {
5116 rv |= IP6_MASKEQ(&state->is_src.in6,
5117 &e->ipfe_arg0[i * 8 + 4],
5118 &e->ipfe_arg0[i * 8]);
5122 case IPF_EXP_IP6_DSTADDR :
5123 if (state->is_v != 6)
5125 for (i = 0; !rv && i < x[3]; i++) {
5126 rv |= IP6_MASKEQ(&state->is_dst.in6,
5127 &e->ipfe_arg0[i * 8 + 4],
5128 &e->ipfe_arg0[i * 8]);
5132 case IPF_EXP_IP6_ADDR :
5133 if (state->is_v != 6)
5135 for (i = 0; !rv && i < x[3]; i++) {
5136 rv |= IP6_MASKEQ(&state->is_src.in6,
5137 &e->ipfe_arg0[i * 8 + 4],
5138 &e->ipfe_arg0[i * 8]) ||
5139 IP6_MASKEQ(&state->is_dst.in6,
5140 &e->ipfe_arg0[i * 8 + 4],
5141 &e->ipfe_arg0[i * 8]);
5146 case IPF_EXP_UDP_PORT :
5147 case IPF_EXP_TCP_PORT :
5148 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5149 rv |= (state->is_sport == e->ipfe_arg0[i]) ||
5150 (state->is_dport == e->ipfe_arg0[i]);
5154 case IPF_EXP_UDP_SPORT :
5155 case IPF_EXP_TCP_SPORT :
5156 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5157 rv |= (state->is_sport == e->ipfe_arg0[i]);
5161 case IPF_EXP_UDP_DPORT :
5162 case IPF_EXP_TCP_DPORT :
5163 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5164 rv |= (state->is_dport == e->ipfe_arg0[i]);
5168 case IPF_EXP_TCP_STATE :
5169 for (i = 0; !rv && i < e->ipfe_narg; i++) {
5170 rv |= (state->is_state[0] == e->ipfe_arg0[i]) ||
5171 (state->is_state[1] == e->ipfe_arg0[i]);
5175 case IPF_EXP_IDLE_GT :
5176 rv |= (ticks - state->is_touched > e->ipfe_arg0[0]);
5181 * Factor in doing a negative match.
5193 /* ------------------------------------------------------------------------ */
5194 /* Function: ipf_state_settimeout */
5195 /* Returns: int 0 = success, else failure */
5196 /* Parameters: softc(I) - pointer to main soft context */
5197 /* t(I) - pointer to tuneable being changed */
5198 /* p(I) - pointer to the new value */
5200 /* Sets a timeout value for one of the many timeout queues. We find the */
5201 /* correct queue using a somewhat manual process of comparing the timeout */
5202 /* names for each specific value available and calling ipf_apply_timeout on */
5203 /* that queue so that all of the items on it are updated accordingly. */
5204 /* ------------------------------------------------------------------------ */
5206 ipf_state_settimeout(softc, t, p)
5207 struct ipf_main_softc_s *softc;
5211 ipf_state_softc_t *softs = softc->ipf_state_soft;
5214 * In case there is nothing to do...
5216 if (*t->ipft_pint == p->ipftu_int)
5219 if (!strncmp(t->ipft_name, "tcp_", 4))
5220 return ipf_settimeout_tcp(t, p, softs->ipf_state_tcptq);
5222 if (!strcmp(t->ipft_name, "udp_timeout")) {
5223 ipf_apply_timeout(&softs->ipf_state_udptq, p->ipftu_int);
5224 } else if (!strcmp(t->ipft_name, "udp_ack_timeout")) {
5225 ipf_apply_timeout(&softs->ipf_state_udpacktq, p->ipftu_int);
5226 } else if (!strcmp(t->ipft_name, "icmp_timeout")) {
5227 ipf_apply_timeout(&softs->ipf_state_icmptq, p->ipftu_int);
5228 } else if (!strcmp(t->ipft_name, "icmp_ack_timeout")) {
5229 ipf_apply_timeout(&softs->ipf_state_icmpacktq, p->ipftu_int);
5230 } else if (!strcmp(t->ipft_name, "ip_timeout")) {
5231 ipf_apply_timeout(&softs->ipf_state_iptq, p->ipftu_int);
5238 * Update the tuneable being set.
5240 *t->ipft_pint = p->ipftu_int;
5246 /* ------------------------------------------------------------------------ */
5247 /* Function: ipf_state_rehash */
5248 /* Returns: int 0 = success, else failure */
5249 /* Parameters: softc(I) - pointer to main soft context */
5250 /* t(I) - pointer to tuneable being changed */
5251 /* p(I) - pointer to the new value */
5253 /* To change the size of the state hash table at runtime, a new table has */
5254 /* to be allocated and then all of the existing entries put in it, bumping */
5255 /* up the bucketlength for it as we go along. */
5256 /* ------------------------------------------------------------------------ */
5258 ipf_state_rehash(softc, t, p)
5259 ipf_main_softc_t *softc;
5263 ipf_state_softc_t *softs = softc->ipf_state_soft;
5264 ipstate_t **newtab, *is;
5272 newsize = p->ipftu_int;
5274 * In case there is nothing to do...
5276 if (newsize == softs->ipf_state_size)
5279 KMALLOCS(newtab, ipstate_t **, newsize * sizeof(ipstate_t *));
5280 if (newtab == NULL) {
5285 KMALLOCS(bucketlens, u_int *, newsize * sizeof(u_int));
5286 if (bucketlens == NULL) {
5287 KFREES(newtab, newsize * sizeof(*softs->ipf_state_table));
5292 newseed = ipf_state_seed_alloc(newsize, softs->ipf_state_max);
5293 if (newseed == NULL) {
5294 KFREES(bucketlens, newsize * sizeof(*bucketlens));
5295 KFREES(newtab, newsize * sizeof(*newtab));
5300 for (maxbucket = 0, i = newsize; i > 0; i >>= 1)
5304 bzero((char *)newtab, newsize * sizeof(ipstate_t *));
5305 bzero((char *)bucketlens, newsize * sizeof(u_int));
5307 WRITE_ENTER(&softc->ipf_state);
5309 if (softs->ipf_state_table != NULL) {
5310 KFREES(softs->ipf_state_table,
5311 softs->ipf_state_size * sizeof(*softs->ipf_state_table));
5313 softs->ipf_state_table = newtab;
5315 if (softs->ipf_state_seed != NULL) {
5316 KFREES(softs->ipf_state_seed,
5317 softs->ipf_state_size * sizeof(*softs->ipf_state_seed));
5319 softs->ipf_state_seed = newseed;
5321 if (softs->ipf_state_stats.iss_bucketlen != NULL) {
5322 KFREES(softs->ipf_state_stats.iss_bucketlen,
5323 softs->ipf_state_size * sizeof(u_int));
5325 softs->ipf_state_stats.iss_bucketlen = bucketlens;
5326 softs->ipf_state_maxbucket = maxbucket;
5327 softs->ipf_state_size = newsize;
5330 * Walk through the entire list of state table entries and put them
5331 * in the new state table, somewhere. Because we have a new table,
5332 * we need to restart the counter of how many chains are in use.
5334 softs->ipf_state_stats.iss_inuse = 0;
5335 for (is = softs->ipf_state_list; is != NULL; is = is->is_next) {
5336 is->is_hnext = NULL;
5337 is->is_phnext = NULL;
5338 hv = is->is_hv % softs->ipf_state_size;
5340 if (softs->ipf_state_table[hv] != NULL)
5341 softs->ipf_state_table[hv]->is_phnext = &is->is_hnext;
5343 softs->ipf_state_stats.iss_inuse++;
5344 is->is_phnext = softs->ipf_state_table + hv;
5345 is->is_hnext = softs->ipf_state_table[hv];
5346 softs->ipf_state_table[hv] = is;
5347 softs->ipf_state_stats.iss_bucketlen[hv]++;
5349 RWLOCK_EXIT(&softc->ipf_state);
5355 /* ------------------------------------------------------------------------ */
5356 /* Function: ipf_state_add_tq */
5357 /* Returns: ipftq_t * - NULL = failure, else pointer to new timeout */
5359 /* Parameters: softc(I) - pointer to main soft context */
5360 /* ttl(I) - pointer to the ttl for the new queue */
5362 /* Request a pointer to a timeout queue that has a ttl as given by the */
5363 /* value being passed in. The timeout queue is added tot the list of those */
5364 /* used internally for stateful filtering. */
5365 /* ------------------------------------------------------------------------ */
5367 ipf_state_add_tq(softc, ttl)
5368 ipf_main_softc_t *softc;
5371 ipf_state_softc_t *softs = softc->ipf_state_soft;
5373 return ipf_addtimeoutqueue(softc, &softs->ipf_state_usertq, ttl);
5379 * Display the built up state table rules and mapping entries.
5382 ipf_state_dump(softc, arg)
5383 ipf_main_softc_t *softc;
5386 ipf_state_softc_t *softs = arg;
5389 printf("List of active state sessions:\n");
5390 for (ips = softs->ipf_state_list; ips != NULL; )
5391 ips = printstate(ips, opts & (OPT_DEBUG|OPT_VERBOSE),