xref: /freebsd/sys/netinet/tcp_syncache.c (revision 0371d265dcb4b7a41cb788636eb80c76e7857a9d)
1 /*-
2  * Copyright (c) 2001 McAfee, Inc.
3  * Copyright (c) 2006 Andre Oppermann, Internet Business Solutions AG
4  * All rights reserved.
5  *
6  * This software was developed for the FreeBSD Project by Jonathan Lemon
7  * and McAfee Research, the Security Research Division of McAfee, Inc. under
8  * DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
9  * DARPA CHATS research program.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/sysctl.h>
44 #include <sys/limits.h>
45 #include <sys/lock.h>
46 #include <sys/mutex.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/md5.h>
50 #include <sys/proc.h>		/* for proc0 declaration */
51 #include <sys/random.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
54 #include <sys/syslog.h>
55 #include <sys/ucred.h>
56 
57 #include <vm/uma.h>
58 
59 #include <net/if.h>
60 #include <net/route.h>
61 #include <net/vnet.h>
62 
63 #include <netinet/in.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/ip.h>
66 #include <netinet/in_var.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/ip_var.h>
69 #include <netinet/ip_options.h>
70 #ifdef INET6
71 #include <netinet/ip6.h>
72 #include <netinet/icmp6.h>
73 #include <netinet6/nd6.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/in6_pcb.h>
76 #endif
77 #include <netinet/tcp.h>
78 #include <netinet/tcp_fsm.h>
79 #include <netinet/tcp_seq.h>
80 #include <netinet/tcp_timer.h>
81 #include <netinet/tcp_var.h>
82 #include <netinet/tcp_syncache.h>
83 #include <netinet/tcp_offload.h>
84 #ifdef INET6
85 #include <netinet6/tcp6_var.h>
86 #endif
87 
88 #ifdef IPSEC
89 #include <netipsec/ipsec.h>
90 #ifdef INET6
91 #include <netipsec/ipsec6.h>
92 #endif
93 #include <netipsec/key.h>
94 #endif /*IPSEC*/
95 
96 #include <machine/in_cksum.h>
97 
98 #include <security/mac/mac_framework.h>
99 
100 static VNET_DEFINE(struct tcp_syncache, tcp_syncache);
101 static VNET_DEFINE(int, tcp_syncookies);
102 static VNET_DEFINE(int, tcp_syncookiesonly);
103 VNET_DEFINE(int, tcp_sc_rst_sock_fail);
104 
105 #define	V_tcp_syncache			VNET(tcp_syncache)
106 #define	V_tcp_syncookies		VNET(tcp_syncookies)
107 #define	V_tcp_syncookiesonly		VNET(tcp_syncookiesonly)
108 
109 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
110     &VNET_NAME(tcp_syncookies), 0,
111     "Use TCP SYN cookies if the syncache overflows");
112 
113 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW,
114     &VNET_NAME(tcp_syncookiesonly), 0,
115     "Use only TCP SYN cookies");
116 
117 #ifdef TCP_OFFLOAD_DISABLE
118 #define TOEPCB_ISSET(sc) (0)
119 #else
120 #define TOEPCB_ISSET(sc) ((sc)->sc_toepcb != NULL)
121 #endif
122 
123 static void	 syncache_drop(struct syncache *, struct syncache_head *);
124 static void	 syncache_free(struct syncache *);
125 static void	 syncache_insert(struct syncache *, struct syncache_head *);
126 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
127 static int	 syncache_respond(struct syncache *);
128 static struct	 socket *syncache_socket(struct syncache *, struct socket *,
129 		    struct mbuf *m);
130 static void	 syncache_timeout(struct syncache *sc, struct syncache_head *sch,
131 		    int docallout);
132 static void	 syncache_timer(void *);
133 static void	 syncookie_generate(struct syncache_head *, struct syncache *,
134 		    u_int32_t *);
135 static struct syncache
136 		*syncookie_lookup(struct in_conninfo *, struct syncache_head *,
137 		    struct syncache *, struct tcpopt *, struct tcphdr *,
138 		    struct socket *);
139 
140 /*
141  * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
142  * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds,
143  * the odds are that the user has given up attempting to connect by then.
144  */
145 #define SYNCACHE_MAXREXMTS		3
146 
147 /* Arbitrary values */
148 #define TCP_SYNCACHE_HASHSIZE		512
149 #define TCP_SYNCACHE_BUCKETLIMIT	30
150 
151 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
152 
153 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
154     &VNET_NAME(tcp_syncache.bucket_limit), 0,
155     "Per-bucket hash limit for syncache");
156 
157 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
158     &VNET_NAME(tcp_syncache.cache_limit), 0,
159     "Overall entry limit for syncache");
160 
161 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
162     &VNET_NAME(tcp_syncache.cache_count), 0,
163     "Current number of entries in syncache");
164 
165 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
166     &VNET_NAME(tcp_syncache.hashsize), 0,
167     "Size of TCP syncache hashtable");
168 
169 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
170     &VNET_NAME(tcp_syncache.rexmt_limit), 0,
171     "Limit on SYN/ACK retransmissions");
172 
173 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail,
174     CTLFLAG_RW, &VNET_NAME(tcp_sc_rst_sock_fail), 0,
175     "Send reset on socket allocation failure");
176 
177 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
178 
179 #define SYNCACHE_HASH(inc, mask)					\
180 	((V_tcp_syncache.hash_secret ^					\
181 	  (inc)->inc_faddr.s_addr ^					\
182 	  ((inc)->inc_faddr.s_addr >> 16) ^				\
183 	  (inc)->inc_fport ^ (inc)->inc_lport) & mask)
184 
185 #define SYNCACHE_HASH6(inc, mask)					\
186 	((V_tcp_syncache.hash_secret ^					\
187 	  (inc)->inc6_faddr.s6_addr32[0] ^				\
188 	  (inc)->inc6_faddr.s6_addr32[3] ^				\
189 	  (inc)->inc_fport ^ (inc)->inc_lport) & mask)
190 
191 #define ENDPTS_EQ(a, b) (						\
192 	(a)->ie_fport == (b)->ie_fport &&				\
193 	(a)->ie_lport == (b)->ie_lport &&				\
194 	(a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr &&			\
195 	(a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr			\
196 )
197 
198 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
199 
200 #define	SCH_LOCK(sch)		mtx_lock(&(sch)->sch_mtx)
201 #define	SCH_UNLOCK(sch)		mtx_unlock(&(sch)->sch_mtx)
202 #define	SCH_LOCK_ASSERT(sch)	mtx_assert(&(sch)->sch_mtx, MA_OWNED)
203 
204 /*
205  * Requires the syncache entry to be already removed from the bucket list.
206  */
207 static void
208 syncache_free(struct syncache *sc)
209 {
210 
211 	if (sc->sc_ipopts)
212 		(void) m_free(sc->sc_ipopts);
213 	if (sc->sc_cred)
214 		crfree(sc->sc_cred);
215 #ifdef MAC
216 	mac_syncache_destroy(&sc->sc_label);
217 #endif
218 
219 	uma_zfree(V_tcp_syncache.zone, sc);
220 }
221 
222 void
223 syncache_init(void)
224 {
225 	int i;
226 
227 	V_tcp_syncookies = 1;
228 	V_tcp_syncookiesonly = 0;
229 	V_tcp_sc_rst_sock_fail = 1;
230 
231 	V_tcp_syncache.cache_count = 0;
232 	V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
233 	V_tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
234 	V_tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
235 	V_tcp_syncache.hash_secret = arc4random();
236 
237 	TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
238 	    &V_tcp_syncache.hashsize);
239 	TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
240 	    &V_tcp_syncache.bucket_limit);
241 	if (!powerof2(V_tcp_syncache.hashsize) ||
242 	    V_tcp_syncache.hashsize == 0) {
243 		printf("WARNING: syncache hash size is not a power of 2.\n");
244 		V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
245 	}
246 	V_tcp_syncache.hashmask = V_tcp_syncache.hashsize - 1;
247 
248 	/* Set limits. */
249 	V_tcp_syncache.cache_limit =
250 	    V_tcp_syncache.hashsize * V_tcp_syncache.bucket_limit;
251 	TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
252 	    &V_tcp_syncache.cache_limit);
253 
254 	/* Allocate the hash table. */
255 	V_tcp_syncache.hashbase = malloc(V_tcp_syncache.hashsize *
256 	    sizeof(struct syncache_head), M_SYNCACHE, M_WAITOK | M_ZERO);
257 
258 	/* Initialize the hash buckets. */
259 	for (i = 0; i < V_tcp_syncache.hashsize; i++) {
260 #ifdef VIMAGE
261 		V_tcp_syncache.hashbase[i].sch_vnet = curvnet;
262 #endif
263 		TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket);
264 		mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
265 			 NULL, MTX_DEF);
266 		callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer,
267 			 &V_tcp_syncache.hashbase[i].sch_mtx, 0);
268 		V_tcp_syncache.hashbase[i].sch_length = 0;
269 	}
270 
271 	/* Create the syncache entry zone. */
272 	V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
273 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
274 	uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit);
275 }
276 
277 #ifdef VIMAGE
278 void
279 syncache_destroy(void)
280 {
281 	struct syncache_head *sch;
282 	struct syncache *sc, *nsc;
283 	int i;
284 
285 	/* Cleanup hash buckets: stop timers, free entries, destroy locks. */
286 	for (i = 0; i < V_tcp_syncache.hashsize; i++) {
287 
288 		sch = &V_tcp_syncache.hashbase[i];
289 		callout_drain(&sch->sch_timer);
290 
291 		SCH_LOCK(sch);
292 		TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc)
293 			syncache_drop(sc, sch);
294 		SCH_UNLOCK(sch);
295 		KASSERT(TAILQ_EMPTY(&sch->sch_bucket),
296 		    ("%s: sch->sch_bucket not empty", __func__));
297 		KASSERT(sch->sch_length == 0, ("%s: sch->sch_length %d not 0",
298 		    __func__, sch->sch_length));
299 		mtx_destroy(&sch->sch_mtx);
300 	}
301 
302 	KASSERT(V_tcp_syncache.cache_count == 0, ("%s: cache_count %d not 0",
303 	    __func__, V_tcp_syncache.cache_count));
304 
305 	/* Free the allocated global resources. */
306 	uma_zdestroy(V_tcp_syncache.zone);
307 	free(V_tcp_syncache.hashbase, M_SYNCACHE);
308 }
309 #endif
310 
311 /*
312  * Inserts a syncache entry into the specified bucket row.
313  * Locks and unlocks the syncache_head autonomously.
314  */
315 static void
316 syncache_insert(struct syncache *sc, struct syncache_head *sch)
317 {
318 	struct syncache *sc2;
319 
320 	SCH_LOCK(sch);
321 
322 	/*
323 	 * Make sure that we don't overflow the per-bucket limit.
324 	 * If the bucket is full, toss the oldest element.
325 	 */
326 	if (sch->sch_length >= V_tcp_syncache.bucket_limit) {
327 		KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
328 			("sch->sch_length incorrect"));
329 		sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
330 		syncache_drop(sc2, sch);
331 		TCPSTAT_INC(tcps_sc_bucketoverflow);
332 	}
333 
334 	/* Put it into the bucket. */
335 	TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
336 	sch->sch_length++;
337 
338 	/* Reinitialize the bucket row's timer. */
339 	if (sch->sch_length == 1)
340 		sch->sch_nextc = ticks + INT_MAX;
341 	syncache_timeout(sc, sch, 1);
342 
343 	SCH_UNLOCK(sch);
344 
345 	V_tcp_syncache.cache_count++;
346 	TCPSTAT_INC(tcps_sc_added);
347 }
348 
349 /*
350  * Remove and free entry from syncache bucket row.
351  * Expects locked syncache head.
352  */
353 static void
354 syncache_drop(struct syncache *sc, struct syncache_head *sch)
355 {
356 
357 	SCH_LOCK_ASSERT(sch);
358 
359 	TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
360 	sch->sch_length--;
361 
362 #ifndef TCP_OFFLOAD_DISABLE
363 	if (sc->sc_tu)
364 		sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb);
365 #endif
366 	syncache_free(sc);
367 	V_tcp_syncache.cache_count--;
368 }
369 
370 /*
371  * Engage/reengage time on bucket row.
372  */
373 static void
374 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
375 {
376 	sc->sc_rxttime = ticks +
377 		TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
378 	sc->sc_rxmits++;
379 	if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
380 		sch->sch_nextc = sc->sc_rxttime;
381 		if (docallout)
382 			callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
383 			    syncache_timer, (void *)sch);
384 	}
385 }
386 
387 /*
388  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
389  * If we have retransmitted an entry the maximum number of times, expire it.
390  * One separate timer for each bucket row.
391  */
392 static void
393 syncache_timer(void *xsch)
394 {
395 	struct syncache_head *sch = (struct syncache_head *)xsch;
396 	struct syncache *sc, *nsc;
397 	int tick = ticks;
398 	char *s;
399 
400 	CURVNET_SET(sch->sch_vnet);
401 
402 	/* NB: syncache_head has already been locked by the callout. */
403 	SCH_LOCK_ASSERT(sch);
404 
405 	/*
406 	 * In the following cycle we may remove some entries and/or
407 	 * advance some timeouts, so re-initialize the bucket timer.
408 	 */
409 	sch->sch_nextc = tick + INT_MAX;
410 
411 	TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
412 		/*
413 		 * We do not check if the listen socket still exists
414 		 * and accept the case where the listen socket may be
415 		 * gone by the time we resend the SYN/ACK.  We do
416 		 * not expect this to happens often. If it does,
417 		 * then the RST will be sent by the time the remote
418 		 * host does the SYN/ACK->ACK.
419 		 */
420 		if (TSTMP_GT(sc->sc_rxttime, tick)) {
421 			if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
422 				sch->sch_nextc = sc->sc_rxttime;
423 			continue;
424 		}
425 		if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) {
426 			if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
427 				log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
428 				    "giving up and removing syncache entry\n",
429 				    s, __func__);
430 				free(s, M_TCPLOG);
431 			}
432 			syncache_drop(sc, sch);
433 			TCPSTAT_INC(tcps_sc_stale);
434 			continue;
435 		}
436 		if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
437 			log(LOG_DEBUG, "%s; %s: Response timeout, "
438 			    "retransmitting (%u) SYN|ACK\n",
439 			    s, __func__, sc->sc_rxmits);
440 			free(s, M_TCPLOG);
441 		}
442 
443 		(void) syncache_respond(sc);
444 		TCPSTAT_INC(tcps_sc_retransmitted);
445 		syncache_timeout(sc, sch, 0);
446 	}
447 	if (!TAILQ_EMPTY(&(sch)->sch_bucket))
448 		callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
449 			syncache_timer, (void *)(sch));
450 	CURVNET_RESTORE();
451 }
452 
453 /*
454  * Find an entry in the syncache.
455  * Returns always with locked syncache_head plus a matching entry or NULL.
456  */
457 struct syncache *
458 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
459 {
460 	struct syncache *sc;
461 	struct syncache_head *sch;
462 
463 #ifdef INET6
464 	if (inc->inc_flags & INC_ISIPV6) {
465 		sch = &V_tcp_syncache.hashbase[
466 		    SYNCACHE_HASH6(inc, V_tcp_syncache.hashmask)];
467 		*schp = sch;
468 
469 		SCH_LOCK(sch);
470 
471 		/* Circle through bucket row to find matching entry. */
472 		TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
473 			if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
474 				return (sc);
475 		}
476 	} else
477 #endif
478 	{
479 		sch = &V_tcp_syncache.hashbase[
480 		    SYNCACHE_HASH(inc, V_tcp_syncache.hashmask)];
481 		*schp = sch;
482 
483 		SCH_LOCK(sch);
484 
485 		/* Circle through bucket row to find matching entry. */
486 		TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
487 #ifdef INET6
488 			if (sc->sc_inc.inc_flags & INC_ISIPV6)
489 				continue;
490 #endif
491 			if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
492 				return (sc);
493 		}
494 	}
495 	SCH_LOCK_ASSERT(*schp);
496 	return (NULL);			/* always returns with locked sch */
497 }
498 
499 /*
500  * This function is called when we get a RST for a
501  * non-existent connection, so that we can see if the
502  * connection is in the syn cache.  If it is, zap it.
503  */
504 void
505 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
506 {
507 	struct syncache *sc;
508 	struct syncache_head *sch;
509 	char *s = NULL;
510 
511 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
512 	SCH_LOCK_ASSERT(sch);
513 
514 	/*
515 	 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
516 	 * See RFC 793 page 65, section SEGMENT ARRIVES.
517 	 */
518 	if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
519 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
520 			log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
521 			    "FIN flag set, segment ignored\n", s, __func__);
522 		TCPSTAT_INC(tcps_badrst);
523 		goto done;
524 	}
525 
526 	/*
527 	 * No corresponding connection was found in syncache.
528 	 * If syncookies are enabled and possibly exclusively
529 	 * used, or we are under memory pressure, a valid RST
530 	 * may not find a syncache entry.  In that case we're
531 	 * done and no SYN|ACK retransmissions will happen.
532 	 * Otherwise the the RST was misdirected or spoofed.
533 	 */
534 	if (sc == NULL) {
535 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
536 			log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
537 			    "syncache entry (possibly syncookie only), "
538 			    "segment ignored\n", s, __func__);
539 		TCPSTAT_INC(tcps_badrst);
540 		goto done;
541 	}
542 
543 	/*
544 	 * If the RST bit is set, check the sequence number to see
545 	 * if this is a valid reset segment.
546 	 * RFC 793 page 37:
547 	 *   In all states except SYN-SENT, all reset (RST) segments
548 	 *   are validated by checking their SEQ-fields.  A reset is
549 	 *   valid if its sequence number is in the window.
550 	 *
551 	 *   The sequence number in the reset segment is normally an
552 	 *   echo of our outgoing acknowlegement numbers, but some hosts
553 	 *   send a reset with the sequence number at the rightmost edge
554 	 *   of our receive window, and we have to handle this case.
555 	 */
556 	if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
557 	    SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
558 		syncache_drop(sc, sch);
559 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
560 			log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
561 			    "connection attempt aborted by remote endpoint\n",
562 			    s, __func__);
563 		TCPSTAT_INC(tcps_sc_reset);
564 	} else {
565 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
566 			log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != "
567 			    "IRS %u (+WND %u), segment ignored\n",
568 			    s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
569 		TCPSTAT_INC(tcps_badrst);
570 	}
571 
572 done:
573 	if (s != NULL)
574 		free(s, M_TCPLOG);
575 	SCH_UNLOCK(sch);
576 }
577 
578 void
579 syncache_badack(struct in_conninfo *inc)
580 {
581 	struct syncache *sc;
582 	struct syncache_head *sch;
583 
584 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
585 	SCH_LOCK_ASSERT(sch);
586 	if (sc != NULL) {
587 		syncache_drop(sc, sch);
588 		TCPSTAT_INC(tcps_sc_badack);
589 	}
590 	SCH_UNLOCK(sch);
591 }
592 
593 void
594 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
595 {
596 	struct syncache *sc;
597 	struct syncache_head *sch;
598 
599 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
600 	SCH_LOCK_ASSERT(sch);
601 	if (sc == NULL)
602 		goto done;
603 
604 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
605 	if (ntohl(th->th_seq) != sc->sc_iss)
606 		goto done;
607 
608 	/*
609 	 * If we've rertransmitted 3 times and this is our second error,
610 	 * we remove the entry.  Otherwise, we allow it to continue on.
611 	 * This prevents us from incorrectly nuking an entry during a
612 	 * spurious network outage.
613 	 *
614 	 * See tcp_notify().
615 	 */
616 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
617 		sc->sc_flags |= SCF_UNREACH;
618 		goto done;
619 	}
620 	syncache_drop(sc, sch);
621 	TCPSTAT_INC(tcps_sc_unreach);
622 done:
623 	SCH_UNLOCK(sch);
624 }
625 
626 /*
627  * Build a new TCP socket structure from a syncache entry.
628  */
629 static struct socket *
630 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
631 {
632 	struct inpcb *inp = NULL;
633 	struct socket *so;
634 	struct tcpcb *tp;
635 	char *s;
636 
637 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
638 
639 	/*
640 	 * Ok, create the full blown connection, and set things up
641 	 * as they would have been set up if we had created the
642 	 * connection when the SYN arrived.  If we can't create
643 	 * the connection, abort it.
644 	 */
645 	so = sonewconn(lso, SS_ISCONNECTED);
646 	if (so == NULL) {
647 		/*
648 		 * Drop the connection; we will either send a RST or
649 		 * have the peer retransmit its SYN again after its
650 		 * RTO and try again.
651 		 */
652 		TCPSTAT_INC(tcps_listendrop);
653 		if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
654 			log(LOG_DEBUG, "%s; %s: Socket create failed "
655 			    "due to limits or memory shortage\n",
656 			    s, __func__);
657 			free(s, M_TCPLOG);
658 		}
659 		goto abort2;
660 	}
661 #ifdef MAC
662 	mac_socketpeer_set_from_mbuf(m, so);
663 #endif
664 
665 	inp = sotoinpcb(so);
666 	inp->inp_inc.inc_fibnum = so->so_fibnum;
667 	INP_WLOCK(inp);
668 
669 	/* Insert new socket into PCB hash list. */
670 	inp->inp_inc.inc_flags = sc->sc_inc.inc_flags;
671 #ifdef INET6
672 	if (sc->sc_inc.inc_flags & INC_ISIPV6) {
673 		inp->in6p_laddr = sc->sc_inc.inc6_laddr;
674 	} else {
675 		inp->inp_vflag &= ~INP_IPV6;
676 		inp->inp_vflag |= INP_IPV4;
677 #endif
678 		inp->inp_laddr = sc->sc_inc.inc_laddr;
679 #ifdef INET6
680 	}
681 #endif
682 	inp->inp_lport = sc->sc_inc.inc_lport;
683 	if (in_pcbinshash(inp) != 0) {
684 		/*
685 		 * Undo the assignments above if we failed to
686 		 * put the PCB on the hash lists.
687 		 */
688 #ifdef INET6
689 		if (sc->sc_inc.inc_flags & INC_ISIPV6)
690 			inp->in6p_laddr = in6addr_any;
691 		else
692 #endif
693 			inp->inp_laddr.s_addr = INADDR_ANY;
694 		inp->inp_lport = 0;
695 		goto abort;
696 	}
697 #ifdef IPSEC
698 	/* Copy old policy into new socket's. */
699 	if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
700 		printf("syncache_socket: could not copy policy\n");
701 #endif
702 #ifdef INET6
703 	if (sc->sc_inc.inc_flags & INC_ISIPV6) {
704 		struct inpcb *oinp = sotoinpcb(lso);
705 		struct in6_addr laddr6;
706 		struct sockaddr_in6 sin6;
707 		/*
708 		 * Inherit socket options from the listening socket.
709 		 * Note that in6p_inputopts are not (and should not be)
710 		 * copied, since it stores previously received options and is
711 		 * used to detect if each new option is different than the
712 		 * previous one and hence should be passed to a user.
713 		 * If we copied in6p_inputopts, a user would not be able to
714 		 * receive options just after calling the accept system call.
715 		 */
716 		inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
717 		if (oinp->in6p_outputopts)
718 			inp->in6p_outputopts =
719 			    ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
720 
721 		sin6.sin6_family = AF_INET6;
722 		sin6.sin6_len = sizeof(sin6);
723 		sin6.sin6_addr = sc->sc_inc.inc6_faddr;
724 		sin6.sin6_port = sc->sc_inc.inc_fport;
725 		sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
726 		laddr6 = inp->in6p_laddr;
727 		if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
728 			inp->in6p_laddr = sc->sc_inc.inc6_laddr;
729 		if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
730 		    thread0.td_ucred)) {
731 			inp->in6p_laddr = laddr6;
732 			goto abort;
733 		}
734 		/* Override flowlabel from in6_pcbconnect. */
735 		inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
736 		inp->inp_flow |= sc->sc_flowlabel;
737 	} else
738 #endif
739 	{
740 		struct in_addr laddr;
741 		struct sockaddr_in sin;
742 
743 		inp->inp_options = (m) ? ip_srcroute(m) : NULL;
744 
745 		if (inp->inp_options == NULL) {
746 			inp->inp_options = sc->sc_ipopts;
747 			sc->sc_ipopts = NULL;
748 		}
749 
750 		sin.sin_family = AF_INET;
751 		sin.sin_len = sizeof(sin);
752 		sin.sin_addr = sc->sc_inc.inc_faddr;
753 		sin.sin_port = sc->sc_inc.inc_fport;
754 		bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
755 		laddr = inp->inp_laddr;
756 		if (inp->inp_laddr.s_addr == INADDR_ANY)
757 			inp->inp_laddr = sc->sc_inc.inc_laddr;
758 		if (in_pcbconnect(inp, (struct sockaddr *)&sin,
759 		    thread0.td_ucred)) {
760 			inp->inp_laddr = laddr;
761 			goto abort;
762 		}
763 	}
764 	tp = intotcpcb(inp);
765 	tp->t_state = TCPS_SYN_RECEIVED;
766 	tp->iss = sc->sc_iss;
767 	tp->irs = sc->sc_irs;
768 	tcp_rcvseqinit(tp);
769 	tcp_sendseqinit(tp);
770 	tp->snd_wl1 = sc->sc_irs;
771 	tp->snd_max = tp->iss + 1;
772 	tp->snd_nxt = tp->iss + 1;
773 	tp->rcv_up = sc->sc_irs + 1;
774 	tp->rcv_wnd = sc->sc_wnd;
775 	tp->rcv_adv += tp->rcv_wnd;
776 	tp->last_ack_sent = tp->rcv_nxt;
777 
778 	tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
779 	if (sc->sc_flags & SCF_NOOPT)
780 		tp->t_flags |= TF_NOOPT;
781 	else {
782 		if (sc->sc_flags & SCF_WINSCALE) {
783 			tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
784 			tp->snd_scale = sc->sc_requested_s_scale;
785 			tp->request_r_scale = sc->sc_requested_r_scale;
786 		}
787 		if (sc->sc_flags & SCF_TIMESTAMP) {
788 			tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
789 			tp->ts_recent = sc->sc_tsreflect;
790 			tp->ts_recent_age = ticks;
791 			tp->ts_offset = sc->sc_tsoff;
792 		}
793 #ifdef TCP_SIGNATURE
794 		if (sc->sc_flags & SCF_SIGNATURE)
795 			tp->t_flags |= TF_SIGNATURE;
796 #endif
797 		if (sc->sc_flags & SCF_SACK)
798 			tp->t_flags |= TF_SACK_PERMIT;
799 	}
800 
801 	if (sc->sc_flags & SCF_ECN)
802 		tp->t_flags |= TF_ECN_PERMIT;
803 
804 	/*
805 	 * Set up MSS and get cached values from tcp_hostcache.
806 	 * This might overwrite some of the defaults we just set.
807 	 */
808 	tcp_mss(tp, sc->sc_peer_mss);
809 
810 	/*
811 	 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
812 	 */
813 	if (sc->sc_rxmits)
814 		tp->snd_cwnd = tp->t_maxseg;
815 	tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
816 
817 	INP_WUNLOCK(inp);
818 
819 	TCPSTAT_INC(tcps_accepts);
820 	return (so);
821 
822 abort:
823 	INP_WUNLOCK(inp);
824 abort2:
825 	if (so != NULL)
826 		soabort(so);
827 	return (NULL);
828 }
829 
830 /*
831  * This function gets called when we receive an ACK for a
832  * socket in the LISTEN state.  We look up the connection
833  * in the syncache, and if its there, we pull it out of
834  * the cache and turn it into a full-blown connection in
835  * the SYN-RECEIVED state.
836  */
837 int
838 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
839     struct socket **lsop, struct mbuf *m)
840 {
841 	struct syncache *sc;
842 	struct syncache_head *sch;
843 	struct syncache scs;
844 	char *s;
845 
846 	/*
847 	 * Global TCP locks are held because we manipulate the PCB lists
848 	 * and create a new socket.
849 	 */
850 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
851 	KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
852 	    ("%s: can handle only ACK", __func__));
853 
854 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
855 	SCH_LOCK_ASSERT(sch);
856 	if (sc == NULL) {
857 		/*
858 		 * There is no syncache entry, so see if this ACK is
859 		 * a returning syncookie.  To do this, first:
860 		 *  A. See if this socket has had a syncache entry dropped in
861 		 *     the past.  We don't want to accept a bogus syncookie
862 		 *     if we've never received a SYN.
863 		 *  B. check that the syncookie is valid.  If it is, then
864 		 *     cobble up a fake syncache entry, and return.
865 		 */
866 		if (!V_tcp_syncookies) {
867 			SCH_UNLOCK(sch);
868 			if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
869 				log(LOG_DEBUG, "%s; %s: Spurious ACK, "
870 				    "segment rejected (syncookies disabled)\n",
871 				    s, __func__);
872 			goto failed;
873 		}
874 		bzero(&scs, sizeof(scs));
875 		sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
876 		SCH_UNLOCK(sch);
877 		if (sc == NULL) {
878 			if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
879 				log(LOG_DEBUG, "%s; %s: Segment failed "
880 				    "SYNCOOKIE authentication, segment rejected "
881 				    "(probably spoofed)\n", s, __func__);
882 			goto failed;
883 		}
884 	} else {
885 		/* Pull out the entry to unlock the bucket row. */
886 		TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
887 		sch->sch_length--;
888 		V_tcp_syncache.cache_count--;
889 		SCH_UNLOCK(sch);
890 	}
891 
892 	/*
893 	 * Segment validation:
894 	 * ACK must match our initial sequence number + 1 (the SYN|ACK).
895 	 */
896 	if (th->th_ack != sc->sc_iss + 1 && !TOEPCB_ISSET(sc)) {
897 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
898 			log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment "
899 			    "rejected\n", s, __func__, th->th_ack, sc->sc_iss);
900 		goto failed;
901 	}
902 
903 	/*
904 	 * The SEQ must fall in the window starting at the received
905 	 * initial receive sequence number + 1 (the SYN).
906 	 */
907 	if ((SEQ_LEQ(th->th_seq, sc->sc_irs) ||
908 	    SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) &&
909 	    !TOEPCB_ISSET(sc)) {
910 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
911 			log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment "
912 			    "rejected\n", s, __func__, th->th_seq, sc->sc_irs);
913 		goto failed;
914 	}
915 
916 	if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
917 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
918 			log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
919 			    "segment rejected\n", s, __func__);
920 		goto failed;
921 	}
922 	/*
923 	 * If timestamps were negotiated the reflected timestamp
924 	 * must be equal to what we actually sent in the SYN|ACK.
925 	 */
926 	if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts &&
927 	    !TOEPCB_ISSET(sc)) {
928 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
929 			log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
930 			    "segment rejected\n",
931 			    s, __func__, to->to_tsecr, sc->sc_ts);
932 		goto failed;
933 	}
934 
935 	*lsop = syncache_socket(sc, *lsop, m);
936 
937 	if (*lsop == NULL)
938 		TCPSTAT_INC(tcps_sc_aborted);
939 	else
940 		TCPSTAT_INC(tcps_sc_completed);
941 
942 /* how do we find the inp for the new socket? */
943 	if (sc != &scs)
944 		syncache_free(sc);
945 	return (1);
946 failed:
947 	if (sc != NULL && sc != &scs)
948 		syncache_free(sc);
949 	if (s != NULL)
950 		free(s, M_TCPLOG);
951 	*lsop = NULL;
952 	return (0);
953 }
954 
955 int
956 tcp_offload_syncache_expand(struct in_conninfo *inc, struct toeopt *toeo,
957     struct tcphdr *th, struct socket **lsop, struct mbuf *m)
958 {
959 	struct tcpopt to;
960 	int rc;
961 
962 	bzero(&to, sizeof(struct tcpopt));
963 	to.to_mss = toeo->to_mss;
964 	to.to_wscale = toeo->to_wscale;
965 	to.to_flags = toeo->to_flags;
966 
967 	INP_INFO_WLOCK(&V_tcbinfo);
968 	rc = syncache_expand(inc, &to, th, lsop, m);
969 	INP_INFO_WUNLOCK(&V_tcbinfo);
970 
971 	return (rc);
972 }
973 
974 /*
975  * Given a LISTEN socket and an inbound SYN request, add
976  * this to the syn cache, and send back a segment:
977  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
978  * to the source.
979  *
980  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
981  * Doing so would require that we hold onto the data and deliver it
982  * to the application.  However, if we are the target of a SYN-flood
983  * DoS attack, an attacker could send data which would eventually
984  * consume all available buffer space if it were ACKed.  By not ACKing
985  * the data, we avoid this DoS scenario.
986  */
987 static void
988 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
989     struct inpcb *inp, struct socket **lsop, struct mbuf *m,
990     struct toe_usrreqs *tu, void *toepcb)
991 {
992 	struct tcpcb *tp;
993 	struct socket *so;
994 	struct syncache *sc = NULL;
995 	struct syncache_head *sch;
996 	struct mbuf *ipopts = NULL;
997 	u_int32_t flowtmp;
998 	int win, sb_hiwat, ip_ttl, ip_tos, noopt;
999 	char *s;
1000 #ifdef INET6
1001 	int autoflowlabel = 0;
1002 #endif
1003 #ifdef MAC
1004 	struct label *maclabel;
1005 #endif
1006 	struct syncache scs;
1007 	struct ucred *cred;
1008 
1009 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1010 	INP_WLOCK_ASSERT(inp);			/* listen socket */
1011 	KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
1012 	    ("%s: unexpected tcp flags", __func__));
1013 
1014 	/*
1015 	 * Combine all so/tp operations very early to drop the INP lock as
1016 	 * soon as possible.
1017 	 */
1018 	so = *lsop;
1019 	tp = sototcpcb(so);
1020 	cred = crhold(so->so_cred);
1021 
1022 #ifdef INET6
1023 	if ((inc->inc_flags & INC_ISIPV6) &&
1024 	    (inp->inp_flags & IN6P_AUTOFLOWLABEL))
1025 		autoflowlabel = 1;
1026 #endif
1027 	ip_ttl = inp->inp_ip_ttl;
1028 	ip_tos = inp->inp_ip_tos;
1029 	win = sbspace(&so->so_rcv);
1030 	sb_hiwat = so->so_rcv.sb_hiwat;
1031 	noopt = (tp->t_flags & TF_NOOPT);
1032 
1033 	/* By the time we drop the lock these should no longer be used. */
1034 	so = NULL;
1035 	tp = NULL;
1036 
1037 #ifdef MAC
1038 	if (mac_syncache_init(&maclabel) != 0) {
1039 		INP_WUNLOCK(inp);
1040 		INP_INFO_WUNLOCK(&V_tcbinfo);
1041 		goto done;
1042 	} else
1043 		mac_syncache_create(maclabel, inp);
1044 #endif
1045 	INP_WUNLOCK(inp);
1046 	INP_INFO_WUNLOCK(&V_tcbinfo);
1047 
1048 	/*
1049 	 * Remember the IP options, if any.
1050 	 */
1051 #ifdef INET6
1052 	if (!(inc->inc_flags & INC_ISIPV6))
1053 #endif
1054 		ipopts = (m) ? ip_srcroute(m) : NULL;
1055 
1056 	/*
1057 	 * See if we already have an entry for this connection.
1058 	 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1059 	 *
1060 	 * XXX: should the syncache be re-initialized with the contents
1061 	 * of the new SYN here (which may have different options?)
1062 	 *
1063 	 * XXX: We do not check the sequence number to see if this is a
1064 	 * real retransmit or a new connection attempt.  The question is
1065 	 * how to handle such a case; either ignore it as spoofed, or
1066 	 * drop the current entry and create a new one?
1067 	 */
1068 	sc = syncache_lookup(inc, &sch);	/* returns locked entry */
1069 	SCH_LOCK_ASSERT(sch);
1070 	if (sc != NULL) {
1071 #ifndef TCP_OFFLOAD_DISABLE
1072 		if (sc->sc_tu)
1073 			sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT,
1074 			    sc->sc_toepcb);
1075 #endif
1076 		TCPSTAT_INC(tcps_sc_dupsyn);
1077 		if (ipopts) {
1078 			/*
1079 			 * If we were remembering a previous source route,
1080 			 * forget it and use the new one we've been given.
1081 			 */
1082 			if (sc->sc_ipopts)
1083 				(void) m_free(sc->sc_ipopts);
1084 			sc->sc_ipopts = ipopts;
1085 		}
1086 		/*
1087 		 * Update timestamp if present.
1088 		 */
1089 		if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
1090 			sc->sc_tsreflect = to->to_tsval;
1091 		else
1092 			sc->sc_flags &= ~SCF_TIMESTAMP;
1093 #ifdef MAC
1094 		/*
1095 		 * Since we have already unconditionally allocated label
1096 		 * storage, free it up.  The syncache entry will already
1097 		 * have an initialized label we can use.
1098 		 */
1099 		mac_syncache_destroy(&maclabel);
1100 #endif
1101 		/* Retransmit SYN|ACK and reset retransmit count. */
1102 		if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
1103 			log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
1104 			    "resetting timer and retransmitting SYN|ACK\n",
1105 			    s, __func__);
1106 			free(s, M_TCPLOG);
1107 		}
1108 		if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) {
1109 			sc->sc_rxmits = 0;
1110 			syncache_timeout(sc, sch, 1);
1111 			TCPSTAT_INC(tcps_sndacks);
1112 			TCPSTAT_INC(tcps_sndtotal);
1113 		}
1114 		SCH_UNLOCK(sch);
1115 		goto done;
1116 	}
1117 
1118 	sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1119 	if (sc == NULL) {
1120 		/*
1121 		 * The zone allocator couldn't provide more entries.
1122 		 * Treat this as if the cache was full; drop the oldest
1123 		 * entry and insert the new one.
1124 		 */
1125 		TCPSTAT_INC(tcps_sc_zonefail);
1126 		if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1127 			syncache_drop(sc, sch);
1128 		sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1129 		if (sc == NULL) {
1130 			if (V_tcp_syncookies) {
1131 				bzero(&scs, sizeof(scs));
1132 				sc = &scs;
1133 			} else {
1134 				SCH_UNLOCK(sch);
1135 				if (ipopts)
1136 					(void) m_free(ipopts);
1137 				goto done;
1138 			}
1139 		}
1140 	}
1141 
1142 	/*
1143 	 * Fill in the syncache values.
1144 	 */
1145 #ifdef MAC
1146 	sc->sc_label = maclabel;
1147 #endif
1148 	sc->sc_cred = cred;
1149 	cred = NULL;
1150 	sc->sc_ipopts = ipopts;
1151 	bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1152 #ifdef INET6
1153 	if (!(inc->inc_flags & INC_ISIPV6))
1154 #endif
1155 	{
1156 		sc->sc_ip_tos = ip_tos;
1157 		sc->sc_ip_ttl = ip_ttl;
1158 	}
1159 #ifndef TCP_OFFLOAD_DISABLE
1160 	sc->sc_tu = tu;
1161 	sc->sc_toepcb = toepcb;
1162 #endif
1163 	sc->sc_irs = th->th_seq;
1164 	sc->sc_iss = arc4random();
1165 	sc->sc_flags = 0;
1166 	sc->sc_flowlabel = 0;
1167 
1168 	/*
1169 	 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
1170 	 * win was derived from socket earlier in the function.
1171 	 */
1172 	win = imax(win, 0);
1173 	win = imin(win, TCP_MAXWIN);
1174 	sc->sc_wnd = win;
1175 
1176 	if (V_tcp_do_rfc1323) {
1177 		/*
1178 		 * A timestamp received in a SYN makes
1179 		 * it ok to send timestamp requests and replies.
1180 		 */
1181 		if (to->to_flags & TOF_TS) {
1182 			sc->sc_tsreflect = to->to_tsval;
1183 			sc->sc_ts = ticks;
1184 			sc->sc_flags |= SCF_TIMESTAMP;
1185 		}
1186 		if (to->to_flags & TOF_SCALE) {
1187 			int wscale = 0;
1188 
1189 			/*
1190 			 * Pick the smallest possible scaling factor that
1191 			 * will still allow us to scale up to sb_max, aka
1192 			 * kern.ipc.maxsockbuf.
1193 			 *
1194 			 * We do this because there are broken firewalls that
1195 			 * will corrupt the window scale option, leading to
1196 			 * the other endpoint believing that our advertised
1197 			 * window is unscaled.  At scale factors larger than
1198 			 * 5 the unscaled window will drop below 1500 bytes,
1199 			 * leading to serious problems when traversing these
1200 			 * broken firewalls.
1201 			 *
1202 			 * With the default maxsockbuf of 256K, a scale factor
1203 			 * of 3 will be chosen by this algorithm.  Those who
1204 			 * choose a larger maxsockbuf should watch out
1205 			 * for the compatiblity problems mentioned above.
1206 			 *
1207 			 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1208 			 * or <SYN,ACK>) segment itself is never scaled.
1209 			 */
1210 			while (wscale < TCP_MAX_WINSHIFT &&
1211 			    (TCP_MAXWIN << wscale) < sb_max)
1212 				wscale++;
1213 			sc->sc_requested_r_scale = wscale;
1214 			sc->sc_requested_s_scale = to->to_wscale;
1215 			sc->sc_flags |= SCF_WINSCALE;
1216 		}
1217 	}
1218 #ifdef TCP_SIGNATURE
1219 	/*
1220 	 * If listening socket requested TCP digests, and received SYN
1221 	 * contains the option, flag this in the syncache so that
1222 	 * syncache_respond() will do the right thing with the SYN+ACK.
1223 	 * XXX: Currently we always record the option by default and will
1224 	 * attempt to use it in syncache_respond().
1225 	 */
1226 	if (to->to_flags & TOF_SIGNATURE)
1227 		sc->sc_flags |= SCF_SIGNATURE;
1228 #endif
1229 	if (to->to_flags & TOF_SACKPERM)
1230 		sc->sc_flags |= SCF_SACK;
1231 	if (to->to_flags & TOF_MSS)
1232 		sc->sc_peer_mss = to->to_mss;	/* peer mss may be zero */
1233 	if (noopt)
1234 		sc->sc_flags |= SCF_NOOPT;
1235 	if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn)
1236 		sc->sc_flags |= SCF_ECN;
1237 
1238 	if (V_tcp_syncookies) {
1239 		syncookie_generate(sch, sc, &flowtmp);
1240 #ifdef INET6
1241 		if (autoflowlabel)
1242 			sc->sc_flowlabel = flowtmp;
1243 #endif
1244 	} else {
1245 #ifdef INET6
1246 		if (autoflowlabel)
1247 			sc->sc_flowlabel =
1248 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1249 #endif
1250 	}
1251 	SCH_UNLOCK(sch);
1252 
1253 	/*
1254 	 * Do a standard 3-way handshake.
1255 	 */
1256 	if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) {
1257 		if (V_tcp_syncookies && V_tcp_syncookiesonly && sc != &scs)
1258 			syncache_free(sc);
1259 		else if (sc != &scs)
1260 			syncache_insert(sc, sch);   /* locks and unlocks sch */
1261 		TCPSTAT_INC(tcps_sndacks);
1262 		TCPSTAT_INC(tcps_sndtotal);
1263 	} else {
1264 		if (sc != &scs)
1265 			syncache_free(sc);
1266 		TCPSTAT_INC(tcps_sc_dropped);
1267 	}
1268 
1269 done:
1270 	if (cred != NULL)
1271 		crfree(cred);
1272 #ifdef MAC
1273 	if (sc == &scs)
1274 		mac_syncache_destroy(&maclabel);
1275 #endif
1276 	if (m) {
1277 
1278 		*lsop = NULL;
1279 		m_freem(m);
1280 	}
1281 }
1282 
1283 static int
1284 syncache_respond(struct syncache *sc)
1285 {
1286 	struct ip *ip = NULL;
1287 	struct mbuf *m;
1288 	struct tcphdr *th;
1289 	int optlen, error;
1290 	u_int16_t hlen, tlen, mssopt;
1291 	struct tcpopt to;
1292 #ifdef INET6
1293 	struct ip6_hdr *ip6 = NULL;
1294 #endif
1295 
1296 	hlen =
1297 #ifdef INET6
1298 	       (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) :
1299 #endif
1300 		sizeof(struct ip);
1301 	tlen = hlen + sizeof(struct tcphdr);
1302 
1303 	/* Determine MSS we advertize to other end of connection. */
1304 	mssopt = tcp_mssopt(&sc->sc_inc);
1305 	if (sc->sc_peer_mss)
1306 		mssopt = max( min(sc->sc_peer_mss, mssopt), V_tcp_minmss);
1307 
1308 	/* XXX: Assume that the entire packet will fit in a header mbuf. */
1309 	KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1310 	    ("syncache: mbuf too small"));
1311 
1312 	/* Create the IP+TCP header from scratch. */
1313 	m = m_gethdr(M_DONTWAIT, MT_DATA);
1314 	if (m == NULL)
1315 		return (ENOBUFS);
1316 #ifdef MAC
1317 	mac_syncache_create_mbuf(sc->sc_label, m);
1318 #endif
1319 	m->m_data += max_linkhdr;
1320 	m->m_len = tlen;
1321 	m->m_pkthdr.len = tlen;
1322 	m->m_pkthdr.rcvif = NULL;
1323 
1324 #ifdef INET6
1325 	if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1326 		ip6 = mtod(m, struct ip6_hdr *);
1327 		ip6->ip6_vfc = IPV6_VERSION;
1328 		ip6->ip6_nxt = IPPROTO_TCP;
1329 		ip6->ip6_src = sc->sc_inc.inc6_laddr;
1330 		ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1331 		ip6->ip6_plen = htons(tlen - hlen);
1332 		/* ip6_hlim is set after checksum */
1333 		ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1334 		ip6->ip6_flow |= sc->sc_flowlabel;
1335 
1336 		th = (struct tcphdr *)(ip6 + 1);
1337 	} else
1338 #endif
1339 	{
1340 		ip = mtod(m, struct ip *);
1341 		ip->ip_v = IPVERSION;
1342 		ip->ip_hl = sizeof(struct ip) >> 2;
1343 		ip->ip_len = tlen;
1344 		ip->ip_id = 0;
1345 		ip->ip_off = 0;
1346 		ip->ip_sum = 0;
1347 		ip->ip_p = IPPROTO_TCP;
1348 		ip->ip_src = sc->sc_inc.inc_laddr;
1349 		ip->ip_dst = sc->sc_inc.inc_faddr;
1350 		ip->ip_ttl = sc->sc_ip_ttl;
1351 		ip->ip_tos = sc->sc_ip_tos;
1352 
1353 		/*
1354 		 * See if we should do MTU discovery.  Route lookups are
1355 		 * expensive, so we will only unset the DF bit if:
1356 		 *
1357 		 *	1) path_mtu_discovery is disabled
1358 		 *	2) the SCF_UNREACH flag has been set
1359 		 */
1360 		if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1361 		       ip->ip_off |= IP_DF;
1362 
1363 		th = (struct tcphdr *)(ip + 1);
1364 	}
1365 	th->th_sport = sc->sc_inc.inc_lport;
1366 	th->th_dport = sc->sc_inc.inc_fport;
1367 
1368 	th->th_seq = htonl(sc->sc_iss);
1369 	th->th_ack = htonl(sc->sc_irs + 1);
1370 	th->th_off = sizeof(struct tcphdr) >> 2;
1371 	th->th_x2 = 0;
1372 	th->th_flags = TH_SYN|TH_ACK;
1373 	th->th_win = htons(sc->sc_wnd);
1374 	th->th_urp = 0;
1375 
1376 	if (sc->sc_flags & SCF_ECN) {
1377 		th->th_flags |= TH_ECE;
1378 		TCPSTAT_INC(tcps_ecn_shs);
1379 	}
1380 
1381 	/* Tack on the TCP options. */
1382 	if ((sc->sc_flags & SCF_NOOPT) == 0) {
1383 		to.to_flags = 0;
1384 
1385 		to.to_mss = mssopt;
1386 		to.to_flags = TOF_MSS;
1387 		if (sc->sc_flags & SCF_WINSCALE) {
1388 			to.to_wscale = sc->sc_requested_r_scale;
1389 			to.to_flags |= TOF_SCALE;
1390 		}
1391 		if (sc->sc_flags & SCF_TIMESTAMP) {
1392 			/* Virgin timestamp or TCP cookie enhanced one. */
1393 			to.to_tsval = sc->sc_ts;
1394 			to.to_tsecr = sc->sc_tsreflect;
1395 			to.to_flags |= TOF_TS;
1396 		}
1397 		if (sc->sc_flags & SCF_SACK)
1398 			to.to_flags |= TOF_SACKPERM;
1399 #ifdef TCP_SIGNATURE
1400 		if (sc->sc_flags & SCF_SIGNATURE)
1401 			to.to_flags |= TOF_SIGNATURE;
1402 #endif
1403 		optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1404 
1405 		/* Adjust headers by option size. */
1406 		th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1407 		m->m_len += optlen;
1408 		m->m_pkthdr.len += optlen;
1409 
1410 #ifdef TCP_SIGNATURE
1411 		if (sc->sc_flags & SCF_SIGNATURE)
1412 			tcp_signature_compute(m, 0, 0, optlen,
1413 			    to.to_signature, IPSEC_DIR_OUTBOUND);
1414 #endif
1415 #ifdef INET6
1416 		if (sc->sc_inc.inc_flags & INC_ISIPV6)
1417 			ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1418 		else
1419 #endif
1420 			ip->ip_len += optlen;
1421 	} else
1422 		optlen = 0;
1423 
1424 	M_SETFIB(m, sc->sc_inc.inc_fibnum);
1425 #ifdef INET6
1426 	if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1427 		th->th_sum = 0;
1428 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
1429 				       tlen + optlen - hlen);
1430 		ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1431 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1432 	} else
1433 #endif
1434 	{
1435 		th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1436 		    htons(tlen + optlen - hlen + IPPROTO_TCP));
1437 		m->m_pkthdr.csum_flags = CSUM_TCP;
1438 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1439 		error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1440 	}
1441 	return (error);
1442 }
1443 
1444 void
1445 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1446     struct inpcb *inp, struct socket **lsop, struct mbuf *m)
1447 {
1448 	_syncache_add(inc, to, th, inp, lsop, m, NULL, NULL);
1449 }
1450 
1451 void
1452 tcp_offload_syncache_add(struct in_conninfo *inc, struct toeopt *toeo,
1453     struct tcphdr *th, struct inpcb *inp, struct socket **lsop,
1454     struct toe_usrreqs *tu, void *toepcb)
1455 {
1456 	struct tcpopt to;
1457 
1458 	bzero(&to, sizeof(struct tcpopt));
1459 	to.to_mss = toeo->to_mss;
1460 	to.to_wscale = toeo->to_wscale;
1461 	to.to_flags = toeo->to_flags;
1462 
1463 	INP_INFO_WLOCK(&V_tcbinfo);
1464 	INP_WLOCK(inp);
1465 
1466 	_syncache_add(inc, &to, th, inp, lsop, NULL, tu, toepcb);
1467 }
1468 
1469 /*
1470  * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1471  * receive and to be able to handle SYN floods from bogus source addresses
1472  * (where we will never receive any reply).  SYN floods try to exhaust all
1473  * our memory and available slots in the SYN cache table to cause a denial
1474  * of service to legitimate users of the local host.
1475  *
1476  * The idea of SYN cookies is to encode and include all necessary information
1477  * about the connection setup state within the SYN-ACK we send back and thus
1478  * to get along without keeping any local state until the ACK to the SYN-ACK
1479  * arrives (if ever).  Everything we need to know should be available from
1480  * the information we encoded in the SYN-ACK.
1481  *
1482  * More information about the theory behind SYN cookies and its first
1483  * discussion and specification can be found at:
1484  *  http://cr.yp.to/syncookies.html    (overview)
1485  *  http://cr.yp.to/syncookies/archive (gory details)
1486  *
1487  * This implementation extends the orginal idea and first implementation
1488  * of FreeBSD by using not only the initial sequence number field to store
1489  * information but also the timestamp field if present.  This way we can
1490  * keep track of the entire state we need to know to recreate the session in
1491  * its original form.  Almost all TCP speakers implement RFC1323 timestamps
1492  * these days.  For those that do not we still have to live with the known
1493  * shortcomings of the ISN only SYN cookies.
1494  *
1495  * Cookie layers:
1496  *
1497  * Initial sequence number we send:
1498  * 31|................................|0
1499  *    DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1500  *    D = MD5 Digest (first dword)
1501  *    M = MSS index
1502  *    R = Rotation of secret
1503  *    P = Odd or Even secret
1504  *
1505  * The MD5 Digest is computed with over following parameters:
1506  *  a) randomly rotated secret
1507  *  b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1508  *  c) the received initial sequence number from remote host
1509  *  d) the rotation offset and odd/even bit
1510  *
1511  * Timestamp we send:
1512  * 31|................................|0
1513  *    DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1514  *    D = MD5 Digest (third dword) (only as filler)
1515  *    S = Requested send window scale
1516  *    R = Requested receive window scale
1517  *    A = SACK allowed
1518  *    5 = TCP-MD5 enabled (not implemented yet)
1519  *    XORed with MD5 Digest (forth dword)
1520  *
1521  * The timestamp isn't cryptographically secure and doesn't need to be.
1522  * The double use of the MD5 digest dwords ties it to a specific remote/
1523  * local host/port, remote initial sequence number and our local time
1524  * limited secret.  A received timestamp is reverted (XORed) and then
1525  * the contained MD5 dword is compared to the computed one to ensure the
1526  * timestamp belongs to the SYN-ACK we sent.  The other parameters may
1527  * have been tampered with but this isn't different from supplying bogus
1528  * values in the SYN in the first place.
1529  *
1530  * Some problems with SYN cookies remain however:
1531  * Consider the problem of a recreated (and retransmitted) cookie.  If the
1532  * original SYN was accepted, the connection is established.  The second
1533  * SYN is inflight, and if it arrives with an ISN that falls within the
1534  * receive window, the connection is killed.
1535  *
1536  * Notes:
1537  * A heuristic to determine when to accept syn cookies is not necessary.
1538  * An ACK flood would cause the syncookie verification to be attempted,
1539  * but a SYN flood causes syncookies to be generated.  Both are of equal
1540  * cost, so there's no point in trying to optimize the ACK flood case.
1541  * Also, if you don't process certain ACKs for some reason, then all someone
1542  * would have to do is launch a SYN and ACK flood at the same time, which
1543  * would stop cookie verification and defeat the entire purpose of syncookies.
1544  */
1545 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1546 
1547 static void
1548 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1549     u_int32_t *flowlabel)
1550 {
1551 	MD5_CTX ctx;
1552 	u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1553 	u_int32_t data;
1554 	u_int32_t *secbits;
1555 	u_int off, pmss, mss;
1556 	int i;
1557 
1558 	SCH_LOCK_ASSERT(sch);
1559 
1560 	/* Which of the two secrets to use. */
1561 	secbits = sch->sch_oddeven ?
1562 			sch->sch_secbits_odd : sch->sch_secbits_even;
1563 
1564 	/* Reseed secret if too old. */
1565 	if (sch->sch_reseed < time_uptime) {
1566 		sch->sch_oddeven = sch->sch_oddeven ? 0 : 1;	/* toggle */
1567 		secbits = sch->sch_oddeven ?
1568 				sch->sch_secbits_odd : sch->sch_secbits_even;
1569 		for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1570 			secbits[i] = arc4random();
1571 		sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1572 	}
1573 
1574 	/* Secret rotation offset. */
1575 	off = sc->sc_iss & 0x7;			/* iss was randomized before */
1576 
1577 	/* Maximum segment size calculation. */
1578 	pmss =
1579 	    max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)),	V_tcp_minmss);
1580 	for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1581 		if (tcp_sc_msstab[mss] <= pmss)
1582 			break;
1583 
1584 	/* Fold parameters and MD5 digest into the ISN we will send. */
1585 	data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1586 	data |= off << 1;	/* secret offset, derived from iss, 3 bits */
1587 	data |= mss << 4;	/* mss, 3 bits */
1588 
1589 	MD5Init(&ctx);
1590 	MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1591 	    SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1592 	MD5Update(&ctx, secbits, off);
1593 	MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1594 	MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1595 	MD5Update(&ctx, &data, sizeof(data));
1596 	MD5Final((u_int8_t *)&md5_buffer, &ctx);
1597 
1598 	data |= (md5_buffer[0] << 7);
1599 	sc->sc_iss = data;
1600 
1601 #ifdef INET6
1602 	*flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1603 #endif
1604 
1605 	/* Additional parameters are stored in the timestamp if present. */
1606 	if (sc->sc_flags & SCF_TIMESTAMP) {
1607 		data =  ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1608 		data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1609 		data |= sc->sc_requested_s_scale << 2;  /* SWIN scale, 4 bits */
1610 		data |= sc->sc_requested_r_scale << 6;  /* RWIN scale, 4 bits */
1611 		data |= md5_buffer[2] << 10;		/* more digest bits */
1612 		data ^= md5_buffer[3];
1613 		sc->sc_ts = data;
1614 		sc->sc_tsoff = data - ticks;		/* after XOR */
1615 	}
1616 
1617 	TCPSTAT_INC(tcps_sc_sendcookie);
1618 }
1619 
1620 static struct syncache *
1621 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1622     struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1623     struct socket *so)
1624 {
1625 	MD5_CTX ctx;
1626 	u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1627 	u_int32_t data = 0;
1628 	u_int32_t *secbits;
1629 	tcp_seq ack, seq;
1630 	int off, mss, wnd, flags;
1631 
1632 	SCH_LOCK_ASSERT(sch);
1633 
1634 	/*
1635 	 * Pull information out of SYN-ACK/ACK and
1636 	 * revert sequence number advances.
1637 	 */
1638 	ack = th->th_ack - 1;
1639 	seq = th->th_seq - 1;
1640 	off = (ack >> 1) & 0x7;
1641 	mss = (ack >> 4) & 0x7;
1642 	flags = ack & 0x7f;
1643 
1644 	/* Which of the two secrets to use. */
1645 	secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1646 
1647 	/*
1648 	 * The secret wasn't updated for the lifetime of a syncookie,
1649 	 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1650 	 */
1651 	if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) {
1652 		return (NULL);
1653 	}
1654 
1655 	/* Recompute the digest so we can compare it. */
1656 	MD5Init(&ctx);
1657 	MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1658 	    SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1659 	MD5Update(&ctx, secbits, off);
1660 	MD5Update(&ctx, inc, sizeof(*inc));
1661 	MD5Update(&ctx, &seq, sizeof(seq));
1662 	MD5Update(&ctx, &flags, sizeof(flags));
1663 	MD5Final((u_int8_t *)&md5_buffer, &ctx);
1664 
1665 	/* Does the digest part of or ACK'ed ISS match? */
1666 	if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1667 		return (NULL);
1668 
1669 	/* Does the digest part of our reflected timestamp match? */
1670 	if (to->to_flags & TOF_TS) {
1671 		data = md5_buffer[3] ^ to->to_tsecr;
1672 		if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1673 			return (NULL);
1674 	}
1675 
1676 	/* Fill in the syncache values. */
1677 	bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1678 	sc->sc_ipopts = NULL;
1679 
1680 	sc->sc_irs = seq;
1681 	sc->sc_iss = ack;
1682 
1683 #ifdef INET6
1684 	if (inc->inc_flags & INC_ISIPV6) {
1685 		if (sotoinpcb(so)->inp_flags & IN6P_AUTOFLOWLABEL)
1686 			sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1687 	} else
1688 #endif
1689 	{
1690 		sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1691 		sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1692 	}
1693 
1694 	/* Additional parameters that were encoded in the timestamp. */
1695 	if (data) {
1696 		sc->sc_flags |= SCF_TIMESTAMP;
1697 		sc->sc_tsreflect = to->to_tsval;
1698 		sc->sc_ts = to->to_tsecr;
1699 		sc->sc_tsoff = to->to_tsecr - ticks;
1700 		sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1701 		sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1702 		sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1703 		    TCP_MAX_WINSHIFT);
1704 		sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1705 		    TCP_MAX_WINSHIFT);
1706 		if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1707 			sc->sc_flags |= SCF_WINSCALE;
1708 	} else
1709 		sc->sc_flags |= SCF_NOOPT;
1710 
1711 	wnd = sbspace(&so->so_rcv);
1712 	wnd = imax(wnd, 0);
1713 	wnd = imin(wnd, TCP_MAXWIN);
1714 	sc->sc_wnd = wnd;
1715 
1716 	sc->sc_rxmits = 0;
1717 	sc->sc_peer_mss = tcp_sc_msstab[mss];
1718 
1719 	TCPSTAT_INC(tcps_sc_recvcookie);
1720 	return (sc);
1721 }
1722 
1723 /*
1724  * Returns the current number of syncache entries.  This number
1725  * will probably change before you get around to calling
1726  * syncache_pcblist.
1727  */
1728 
1729 int
1730 syncache_pcbcount(void)
1731 {
1732 	struct syncache_head *sch;
1733 	int count, i;
1734 
1735 	for (count = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1736 		/* No need to lock for a read. */
1737 		sch = &V_tcp_syncache.hashbase[i];
1738 		count += sch->sch_length;
1739 	}
1740 	return count;
1741 }
1742 
1743 /*
1744  * Exports the syncache entries to userland so that netstat can display
1745  * them alongside the other sockets.  This function is intended to be
1746  * called only from tcp_pcblist.
1747  *
1748  * Due to concurrency on an active system, the number of pcbs exported
1749  * may have no relation to max_pcbs.  max_pcbs merely indicates the
1750  * amount of space the caller allocated for this function to use.
1751  */
1752 int
1753 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
1754 {
1755 	struct xtcpcb xt;
1756 	struct syncache *sc;
1757 	struct syncache_head *sch;
1758 	int count, error, i;
1759 
1760 	for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1761 		sch = &V_tcp_syncache.hashbase[i];
1762 		SCH_LOCK(sch);
1763 		TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
1764 			if (count >= max_pcbs) {
1765 				SCH_UNLOCK(sch);
1766 				goto exit;
1767 			}
1768 			if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0)
1769 				continue;
1770 			bzero(&xt, sizeof(xt));
1771 			xt.xt_len = sizeof(xt);
1772 			if (sc->sc_inc.inc_flags & INC_ISIPV6)
1773 				xt.xt_inp.inp_vflag = INP_IPV6;
1774 			else
1775 				xt.xt_inp.inp_vflag = INP_IPV4;
1776 			bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
1777 			xt.xt_tp.t_inpcb = &xt.xt_inp;
1778 			xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
1779 			xt.xt_socket.xso_protocol = IPPROTO_TCP;
1780 			xt.xt_socket.xso_len = sizeof (struct xsocket);
1781 			xt.xt_socket.so_type = SOCK_STREAM;
1782 			xt.xt_socket.so_state = SS_ISCONNECTING;
1783 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1784 			if (error) {
1785 				SCH_UNLOCK(sch);
1786 				goto exit;
1787 			}
1788 			count++;
1789 		}
1790 		SCH_UNLOCK(sch);
1791 	}
1792 exit:
1793 	*pcbs_exported = count;
1794 	return error;
1795 }
1796