xref: /freebsd/sys/netinet/tcp_syncache.c (revision 25dd52cdb10d223b9258836e23cc6ae4ea333b86)
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 #include "opt_mac.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
45 #include <sys/limits.h>
46 #include <sys/lock.h>
47 #include <sys/mutex.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/md5.h>
51 #include <sys/proc.h>		/* for proc0 declaration */
52 #include <sys/random.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/syslog.h>
56 #include <sys/ucred.h>
57 #include <sys/vimage.h>
58 
59 #include <vm/uma.h>
60 
61 #include <net/if.h>
62 #include <net/route.h>
63 
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/ip.h>
67 #include <netinet/in_var.h>
68 #include <netinet/in_pcb.h>
69 #include <netinet/ip_var.h>
70 #include <netinet/ip_options.h>
71 #ifdef INET6
72 #include <netinet/ip6.h>
73 #include <netinet/icmp6.h>
74 #include <netinet6/nd6.h>
75 #include <netinet6/ip6_var.h>
76 #include <netinet6/in6_pcb.h>
77 #endif
78 #include <netinet/tcp.h>
79 #include <netinet/tcp_fsm.h>
80 #include <netinet/tcp_seq.h>
81 #include <netinet/tcp_timer.h>
82 #include <netinet/tcp_var.h>
83 #include <netinet/tcp_syncache.h>
84 #include <netinet/tcp_offload.h>
85 #ifdef INET6
86 #include <netinet6/tcp6_var.h>
87 #endif
88 #include <netinet/vinet.h>
89 
90 #ifdef IPSEC
91 #include <netipsec/ipsec.h>
92 #ifdef INET6
93 #include <netipsec/ipsec6.h>
94 #endif
95 #include <netipsec/key.h>
96 #endif /*IPSEC*/
97 
98 #include <machine/in_cksum.h>
99 
100 #include <security/mac/mac_framework.h>
101 
102 #ifdef VIMAGE_GLOBALS
103 static struct tcp_syncache tcp_syncache;
104 static int tcp_syncookies;
105 static int tcp_syncookiesonly;
106 int tcp_sc_rst_sock_fail;
107 #endif
108 
109 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, syncookies,
110     CTLFLAG_RW, tcp_syncookies, 0,
111     "Use TCP SYN cookies if the syncache overflows");
112 
113 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, syncookies_only,
114     CTLFLAG_RW, 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_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO,
154     bucketlimit, CTLFLAG_RDTUN,
155     tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
156 
157 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO,
158     cachelimit, CTLFLAG_RDTUN,
159     tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
160 
161 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO,
162     count, CTLFLAG_RD,
163     tcp_syncache.cache_count, 0, "Current number of entries in syncache");
164 
165 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO,
166     hashsize, CTLFLAG_RDTUN,
167     tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
168 
169 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO,
170     rexmtlimit, CTLFLAG_RW,
171     tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
172 
173 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO,
174      rst_on_sock_fail, CTLFLAG_RW,
175      tcp_sc_rst_sock_fail, 0, "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 	INIT_VNET_INET(curvnet);
211 
212 	if (sc->sc_ipopts)
213 		(void) m_free(sc->sc_ipopts);
214 	if (sc->sc_cred)
215 		crfree(sc->sc_cred);
216 #ifdef MAC
217 	mac_syncache_destroy(&sc->sc_label);
218 #endif
219 
220 	uma_zfree(V_tcp_syncache.zone, sc);
221 }
222 
223 void
224 syncache_init(void)
225 {
226 	INIT_VNET_INET(curvnet);
227 	int i;
228 
229 	V_tcp_syncookies = 1;
230 	V_tcp_syncookiesonly = 0;
231 	V_tcp_sc_rst_sock_fail = 1;
232 
233 	V_tcp_syncache.cache_count = 0;
234 	V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
235 	V_tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
236 	V_tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
237 	V_tcp_syncache.hash_secret = arc4random();
238 
239 	TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
240 	    &V_tcp_syncache.hashsize);
241 	TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
242 	    &V_tcp_syncache.bucket_limit);
243 	if (!powerof2(V_tcp_syncache.hashsize) ||
244 	    V_tcp_syncache.hashsize == 0) {
245 		printf("WARNING: syncache hash size is not a power of 2.\n");
246 		V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
247 	}
248 	V_tcp_syncache.hashmask = V_tcp_syncache.hashsize - 1;
249 
250 	/* Set limits. */
251 	V_tcp_syncache.cache_limit =
252 	    V_tcp_syncache.hashsize * V_tcp_syncache.bucket_limit;
253 	TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
254 	    &V_tcp_syncache.cache_limit);
255 
256 	/* Allocate the hash table. */
257 	V_tcp_syncache.hashbase = malloc(V_tcp_syncache.hashsize *
258 	    sizeof(struct syncache_head), M_SYNCACHE, M_WAITOK | M_ZERO);
259 
260 	/* Initialize the hash buckets. */
261 	for (i = 0; i < V_tcp_syncache.hashsize; i++) {
262 #ifdef VIMAGE
263 		V_tcp_syncache.hashbase[i].sch_vnet = curvnet;
264 #endif
265 		TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket);
266 		mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
267 			 NULL, MTX_DEF);
268 		callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer,
269 			 &V_tcp_syncache.hashbase[i].sch_mtx, 0);
270 		V_tcp_syncache.hashbase[i].sch_length = 0;
271 	}
272 
273 	/* Create the syncache entry zone. */
274 	V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
275 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
276 	uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit);
277 }
278 
279 /*
280  * Inserts a syncache entry into the specified bucket row.
281  * Locks and unlocks the syncache_head autonomously.
282  */
283 static void
284 syncache_insert(struct syncache *sc, struct syncache_head *sch)
285 {
286 	INIT_VNET_INET(sch->sch_vnet);
287 	struct syncache *sc2;
288 
289 	SCH_LOCK(sch);
290 
291 	/*
292 	 * Make sure that we don't overflow the per-bucket limit.
293 	 * If the bucket is full, toss the oldest element.
294 	 */
295 	if (sch->sch_length >= V_tcp_syncache.bucket_limit) {
296 		KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
297 			("sch->sch_length incorrect"));
298 		sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
299 		syncache_drop(sc2, sch);
300 		TCPSTAT_INC(tcps_sc_bucketoverflow);
301 	}
302 
303 	/* Put it into the bucket. */
304 	TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
305 	sch->sch_length++;
306 
307 	/* Reinitialize the bucket row's timer. */
308 	if (sch->sch_length == 1)
309 		sch->sch_nextc = ticks + INT_MAX;
310 	syncache_timeout(sc, sch, 1);
311 
312 	SCH_UNLOCK(sch);
313 
314 	V_tcp_syncache.cache_count++;
315 	TCPSTAT_INC(tcps_sc_added);
316 }
317 
318 /*
319  * Remove and free entry from syncache bucket row.
320  * Expects locked syncache head.
321  */
322 static void
323 syncache_drop(struct syncache *sc, struct syncache_head *sch)
324 {
325 	INIT_VNET_INET(sch->sch_vnet);
326 
327 	SCH_LOCK_ASSERT(sch);
328 
329 	TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
330 	sch->sch_length--;
331 
332 #ifndef TCP_OFFLOAD_DISABLE
333 	if (sc->sc_tu)
334 		sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb);
335 #endif
336 	syncache_free(sc);
337 	V_tcp_syncache.cache_count--;
338 }
339 
340 /*
341  * Engage/reengage time on bucket row.
342  */
343 static void
344 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
345 {
346 	sc->sc_rxttime = ticks +
347 		TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
348 	sc->sc_rxmits++;
349 	if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
350 		sch->sch_nextc = sc->sc_rxttime;
351 		if (docallout)
352 			callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
353 			    syncache_timer, (void *)sch);
354 	}
355 }
356 
357 /*
358  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
359  * If we have retransmitted an entry the maximum number of times, expire it.
360  * One separate timer for each bucket row.
361  */
362 static void
363 syncache_timer(void *xsch)
364 {
365 	struct syncache_head *sch = (struct syncache_head *)xsch;
366 	struct syncache *sc, *nsc;
367 	int tick = ticks;
368 	char *s;
369 
370 	CURVNET_SET(sch->sch_vnet);
371 	INIT_VNET_INET(sch->sch_vnet);
372 
373 	/* NB: syncache_head has already been locked by the callout. */
374 	SCH_LOCK_ASSERT(sch);
375 
376 	/*
377 	 * In the following cycle we may remove some entries and/or
378 	 * advance some timeouts, so re-initialize the bucket timer.
379 	 */
380 	sch->sch_nextc = tick + INT_MAX;
381 
382 	TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
383 		/*
384 		 * We do not check if the listen socket still exists
385 		 * and accept the case where the listen socket may be
386 		 * gone by the time we resend the SYN/ACK.  We do
387 		 * not expect this to happens often. If it does,
388 		 * then the RST will be sent by the time the remote
389 		 * host does the SYN/ACK->ACK.
390 		 */
391 		if (TSTMP_GT(sc->sc_rxttime, tick)) {
392 			if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
393 				sch->sch_nextc = sc->sc_rxttime;
394 			continue;
395 		}
396 		if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) {
397 			if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
398 				log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
399 				    "giving up and removing syncache entry\n",
400 				    s, __func__);
401 				free(s, M_TCPLOG);
402 			}
403 			syncache_drop(sc, sch);
404 			TCPSTAT_INC(tcps_sc_stale);
405 			continue;
406 		}
407 		if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
408 			log(LOG_DEBUG, "%s; %s: Response timeout, "
409 			    "retransmitting (%u) SYN|ACK\n",
410 			    s, __func__, sc->sc_rxmits);
411 			free(s, M_TCPLOG);
412 		}
413 
414 		(void) syncache_respond(sc);
415 		TCPSTAT_INC(tcps_sc_retransmitted);
416 		syncache_timeout(sc, sch, 0);
417 	}
418 	if (!TAILQ_EMPTY(&(sch)->sch_bucket))
419 		callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
420 			syncache_timer, (void *)(sch));
421 	CURVNET_RESTORE();
422 }
423 
424 /*
425  * Find an entry in the syncache.
426  * Returns always with locked syncache_head plus a matching entry or NULL.
427  */
428 struct syncache *
429 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
430 {
431 	INIT_VNET_INET(curvnet);
432 	struct syncache *sc;
433 	struct syncache_head *sch;
434 
435 #ifdef INET6
436 	if (inc->inc_flags & INC_ISIPV6) {
437 		sch = &V_tcp_syncache.hashbase[
438 		    SYNCACHE_HASH6(inc, V_tcp_syncache.hashmask)];
439 		*schp = sch;
440 
441 		SCH_LOCK(sch);
442 
443 		/* Circle through bucket row to find matching entry. */
444 		TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
445 			if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
446 				return (sc);
447 		}
448 	} else
449 #endif
450 	{
451 		sch = &V_tcp_syncache.hashbase[
452 		    SYNCACHE_HASH(inc, V_tcp_syncache.hashmask)];
453 		*schp = sch;
454 
455 		SCH_LOCK(sch);
456 
457 		/* Circle through bucket row to find matching entry. */
458 		TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
459 #ifdef INET6
460 			if (sc->sc_inc.inc_flags & INC_ISIPV6)
461 				continue;
462 #endif
463 			if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
464 				return (sc);
465 		}
466 	}
467 	SCH_LOCK_ASSERT(*schp);
468 	return (NULL);			/* always returns with locked sch */
469 }
470 
471 /*
472  * This function is called when we get a RST for a
473  * non-existent connection, so that we can see if the
474  * connection is in the syn cache.  If it is, zap it.
475  */
476 void
477 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
478 {
479 	INIT_VNET_INET(curvnet);
480 	struct syncache *sc;
481 	struct syncache_head *sch;
482 	char *s = NULL;
483 
484 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
485 	SCH_LOCK_ASSERT(sch);
486 
487 	/*
488 	 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
489 	 * See RFC 793 page 65, section SEGMENT ARRIVES.
490 	 */
491 	if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
492 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
493 			log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
494 			    "FIN flag set, segment ignored\n", s, __func__);
495 		TCPSTAT_INC(tcps_badrst);
496 		goto done;
497 	}
498 
499 	/*
500 	 * No corresponding connection was found in syncache.
501 	 * If syncookies are enabled and possibly exclusively
502 	 * used, or we are under memory pressure, a valid RST
503 	 * may not find a syncache entry.  In that case we're
504 	 * done and no SYN|ACK retransmissions will happen.
505 	 * Otherwise the the RST was misdirected or spoofed.
506 	 */
507 	if (sc == NULL) {
508 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
509 			log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
510 			    "syncache entry (possibly syncookie only), "
511 			    "segment ignored\n", s, __func__);
512 		TCPSTAT_INC(tcps_badrst);
513 		goto done;
514 	}
515 
516 	/*
517 	 * If the RST bit is set, check the sequence number to see
518 	 * if this is a valid reset segment.
519 	 * RFC 793 page 37:
520 	 *   In all states except SYN-SENT, all reset (RST) segments
521 	 *   are validated by checking their SEQ-fields.  A reset is
522 	 *   valid if its sequence number is in the window.
523 	 *
524 	 *   The sequence number in the reset segment is normally an
525 	 *   echo of our outgoing acknowlegement numbers, but some hosts
526 	 *   send a reset with the sequence number at the rightmost edge
527 	 *   of our receive window, and we have to handle this case.
528 	 */
529 	if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
530 	    SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
531 		syncache_drop(sc, sch);
532 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
533 			log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
534 			    "connection attempt aborted by remote endpoint\n",
535 			    s, __func__);
536 		TCPSTAT_INC(tcps_sc_reset);
537 	} else {
538 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
539 			log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != "
540 			    "IRS %u (+WND %u), segment ignored\n",
541 			    s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
542 		TCPSTAT_INC(tcps_badrst);
543 	}
544 
545 done:
546 	if (s != NULL)
547 		free(s, M_TCPLOG);
548 	SCH_UNLOCK(sch);
549 }
550 
551 void
552 syncache_badack(struct in_conninfo *inc)
553 {
554 	INIT_VNET_INET(curvnet);
555 	struct syncache *sc;
556 	struct syncache_head *sch;
557 
558 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
559 	SCH_LOCK_ASSERT(sch);
560 	if (sc != NULL) {
561 		syncache_drop(sc, sch);
562 		TCPSTAT_INC(tcps_sc_badack);
563 	}
564 	SCH_UNLOCK(sch);
565 }
566 
567 void
568 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
569 {
570 	INIT_VNET_INET(curvnet);
571 	struct syncache *sc;
572 	struct syncache_head *sch;
573 
574 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
575 	SCH_LOCK_ASSERT(sch);
576 	if (sc == NULL)
577 		goto done;
578 
579 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
580 	if (ntohl(th->th_seq) != sc->sc_iss)
581 		goto done;
582 
583 	/*
584 	 * If we've rertransmitted 3 times and this is our second error,
585 	 * we remove the entry.  Otherwise, we allow it to continue on.
586 	 * This prevents us from incorrectly nuking an entry during a
587 	 * spurious network outage.
588 	 *
589 	 * See tcp_notify().
590 	 */
591 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
592 		sc->sc_flags |= SCF_UNREACH;
593 		goto done;
594 	}
595 	syncache_drop(sc, sch);
596 	TCPSTAT_INC(tcps_sc_unreach);
597 done:
598 	SCH_UNLOCK(sch);
599 }
600 
601 /*
602  * Build a new TCP socket structure from a syncache entry.
603  */
604 static struct socket *
605 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
606 {
607 	INIT_VNET_INET(lso->so_vnet);
608 	struct inpcb *inp = NULL;
609 	struct socket *so;
610 	struct tcpcb *tp;
611 	char *s;
612 
613 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
614 
615 	/*
616 	 * Ok, create the full blown connection, and set things up
617 	 * as they would have been set up if we had created the
618 	 * connection when the SYN arrived.  If we can't create
619 	 * the connection, abort it.
620 	 */
621 	so = sonewconn(lso, SS_ISCONNECTED);
622 	if (so == NULL) {
623 		/*
624 		 * Drop the connection; we will either send a RST or
625 		 * have the peer retransmit its SYN again after its
626 		 * RTO and try again.
627 		 */
628 		TCPSTAT_INC(tcps_listendrop);
629 		if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
630 			log(LOG_DEBUG, "%s; %s: Socket create failed "
631 			    "due to limits or memory shortage\n",
632 			    s, __func__);
633 			free(s, M_TCPLOG);
634 		}
635 		goto abort2;
636 	}
637 #ifdef MAC
638 	SOCK_LOCK(so);
639 	mac_socketpeer_set_from_mbuf(m, so);
640 	SOCK_UNLOCK(so);
641 #endif
642 
643 	inp = sotoinpcb(so);
644 	inp->inp_inc.inc_fibnum = sc->sc_inc.inc_fibnum;
645 	so->so_fibnum = sc->sc_inc.inc_fibnum;
646 	INP_WLOCK(inp);
647 
648 	/* Insert new socket into PCB hash list. */
649 	inp->inp_inc.inc_flags = sc->sc_inc.inc_flags;
650 #ifdef INET6
651 	if (sc->sc_inc.inc_flags & INC_ISIPV6) {
652 		inp->in6p_laddr = sc->sc_inc.inc6_laddr;
653 	} else {
654 		inp->inp_vflag &= ~INP_IPV6;
655 		inp->inp_vflag |= INP_IPV4;
656 #endif
657 		inp->inp_laddr = sc->sc_inc.inc_laddr;
658 #ifdef INET6
659 	}
660 #endif
661 	inp->inp_lport = sc->sc_inc.inc_lport;
662 	if (in_pcbinshash(inp) != 0) {
663 		/*
664 		 * Undo the assignments above if we failed to
665 		 * put the PCB on the hash lists.
666 		 */
667 #ifdef INET6
668 		if (sc->sc_inc.inc_flags & INC_ISIPV6)
669 			inp->in6p_laddr = in6addr_any;
670 		else
671 #endif
672 			inp->inp_laddr.s_addr = INADDR_ANY;
673 		inp->inp_lport = 0;
674 		goto abort;
675 	}
676 #ifdef IPSEC
677 	/* Copy old policy into new socket's. */
678 	if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
679 		printf("syncache_socket: could not copy policy\n");
680 #endif
681 #ifdef INET6
682 	if (sc->sc_inc.inc_flags & INC_ISIPV6) {
683 		struct inpcb *oinp = sotoinpcb(lso);
684 		struct in6_addr laddr6;
685 		struct sockaddr_in6 sin6;
686 		/*
687 		 * Inherit socket options from the listening socket.
688 		 * Note that in6p_inputopts are not (and should not be)
689 		 * copied, since it stores previously received options and is
690 		 * used to detect if each new option is different than the
691 		 * previous one and hence should be passed to a user.
692 		 * If we copied in6p_inputopts, a user would not be able to
693 		 * receive options just after calling the accept system call.
694 		 */
695 		inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
696 		if (oinp->in6p_outputopts)
697 			inp->in6p_outputopts =
698 			    ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
699 
700 		sin6.sin6_family = AF_INET6;
701 		sin6.sin6_len = sizeof(sin6);
702 		sin6.sin6_addr = sc->sc_inc.inc6_faddr;
703 		sin6.sin6_port = sc->sc_inc.inc_fport;
704 		sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
705 		laddr6 = inp->in6p_laddr;
706 		if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
707 			inp->in6p_laddr = sc->sc_inc.inc6_laddr;
708 		if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
709 		    thread0.td_ucred)) {
710 			inp->in6p_laddr = laddr6;
711 			goto abort;
712 		}
713 		/* Override flowlabel from in6_pcbconnect. */
714 		inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
715 		inp->inp_flow |= sc->sc_flowlabel;
716 	} else
717 #endif
718 	{
719 		struct in_addr laddr;
720 		struct sockaddr_in sin;
721 
722 		inp->inp_options = (m) ? ip_srcroute(m) : NULL;
723 
724 		if (inp->inp_options == NULL) {
725 			inp->inp_options = sc->sc_ipopts;
726 			sc->sc_ipopts = NULL;
727 		}
728 
729 		sin.sin_family = AF_INET;
730 		sin.sin_len = sizeof(sin);
731 		sin.sin_addr = sc->sc_inc.inc_faddr;
732 		sin.sin_port = sc->sc_inc.inc_fport;
733 		bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
734 		laddr = inp->inp_laddr;
735 		if (inp->inp_laddr.s_addr == INADDR_ANY)
736 			inp->inp_laddr = sc->sc_inc.inc_laddr;
737 		if (in_pcbconnect(inp, (struct sockaddr *)&sin,
738 		    thread0.td_ucred)) {
739 			inp->inp_laddr = laddr;
740 			goto abort;
741 		}
742 	}
743 	tp = intotcpcb(inp);
744 	tp->t_state = TCPS_SYN_RECEIVED;
745 	tp->iss = sc->sc_iss;
746 	tp->irs = sc->sc_irs;
747 	tcp_rcvseqinit(tp);
748 	tcp_sendseqinit(tp);
749 	tp->snd_wl1 = sc->sc_irs;
750 	tp->snd_max = tp->iss + 1;
751 	tp->snd_nxt = tp->iss + 1;
752 	tp->rcv_up = sc->sc_irs + 1;
753 	tp->rcv_wnd = sc->sc_wnd;
754 	tp->rcv_adv += tp->rcv_wnd;
755 	tp->last_ack_sent = tp->rcv_nxt;
756 
757 	tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
758 	if (sc->sc_flags & SCF_NOOPT)
759 		tp->t_flags |= TF_NOOPT;
760 	else {
761 		if (sc->sc_flags & SCF_WINSCALE) {
762 			tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
763 			tp->snd_scale = sc->sc_requested_s_scale;
764 			tp->request_r_scale = sc->sc_requested_r_scale;
765 		}
766 		if (sc->sc_flags & SCF_TIMESTAMP) {
767 			tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
768 			tp->ts_recent = sc->sc_tsreflect;
769 			tp->ts_recent_age = ticks;
770 			tp->ts_offset = sc->sc_tsoff;
771 		}
772 #ifdef TCP_SIGNATURE
773 		if (sc->sc_flags & SCF_SIGNATURE)
774 			tp->t_flags |= TF_SIGNATURE;
775 #endif
776 		if (sc->sc_flags & SCF_SACK)
777 			tp->t_flags |= TF_SACK_PERMIT;
778 	}
779 
780 	if (sc->sc_flags & SCF_ECN)
781 		tp->t_flags |= TF_ECN_PERMIT;
782 
783 	/*
784 	 * Set up MSS and get cached values from tcp_hostcache.
785 	 * This might overwrite some of the defaults we just set.
786 	 */
787 	tcp_mss(tp, sc->sc_peer_mss);
788 
789 	/*
790 	 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
791 	 */
792 	if (sc->sc_rxmits)
793 		tp->snd_cwnd = tp->t_maxseg;
794 	tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
795 
796 	INP_WUNLOCK(inp);
797 
798 	TCPSTAT_INC(tcps_accepts);
799 	return (so);
800 
801 abort:
802 	INP_WUNLOCK(inp);
803 abort2:
804 	if (so != NULL)
805 		soabort(so);
806 	return (NULL);
807 }
808 
809 /*
810  * This function gets called when we receive an ACK for a
811  * socket in the LISTEN state.  We look up the connection
812  * in the syncache, and if its there, we pull it out of
813  * the cache and turn it into a full-blown connection in
814  * the SYN-RECEIVED state.
815  */
816 int
817 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
818     struct socket **lsop, struct mbuf *m)
819 {
820 	INIT_VNET_INET(curvnet);
821 	struct syncache *sc;
822 	struct syncache_head *sch;
823 	struct syncache scs;
824 	char *s;
825 
826 	/*
827 	 * Global TCP locks are held because we manipulate the PCB lists
828 	 * and create a new socket.
829 	 */
830 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
831 	KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
832 	    ("%s: can handle only ACK", __func__));
833 
834 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
835 	SCH_LOCK_ASSERT(sch);
836 	if (sc == NULL) {
837 		/*
838 		 * There is no syncache entry, so see if this ACK is
839 		 * a returning syncookie.  To do this, first:
840 		 *  A. See if this socket has had a syncache entry dropped in
841 		 *     the past.  We don't want to accept a bogus syncookie
842 		 *     if we've never received a SYN.
843 		 *  B. check that the syncookie is valid.  If it is, then
844 		 *     cobble up a fake syncache entry, and return.
845 		 */
846 		if (!V_tcp_syncookies) {
847 			SCH_UNLOCK(sch);
848 			if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
849 				log(LOG_DEBUG, "%s; %s: Spurious ACK, "
850 				    "segment rejected (syncookies disabled)\n",
851 				    s, __func__);
852 			goto failed;
853 		}
854 		bzero(&scs, sizeof(scs));
855 		sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
856 		SCH_UNLOCK(sch);
857 		if (sc == NULL) {
858 			if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
859 				log(LOG_DEBUG, "%s; %s: Segment failed "
860 				    "SYNCOOKIE authentication, segment rejected "
861 				    "(probably spoofed)\n", s, __func__);
862 			goto failed;
863 		}
864 	} else {
865 		/* Pull out the entry to unlock the bucket row. */
866 		TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
867 		sch->sch_length--;
868 		V_tcp_syncache.cache_count--;
869 		SCH_UNLOCK(sch);
870 	}
871 
872 	/*
873 	 * Segment validation:
874 	 * ACK must match our initial sequence number + 1 (the SYN|ACK).
875 	 */
876 	if (th->th_ack != sc->sc_iss + 1 && !TOEPCB_ISSET(sc)) {
877 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
878 			log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment "
879 			    "rejected\n", s, __func__, th->th_ack, sc->sc_iss);
880 		goto failed;
881 	}
882 
883 	/*
884 	 * The SEQ must fall in the window starting at the received
885 	 * initial receive sequence number + 1 (the SYN).
886 	 */
887 	if ((SEQ_LEQ(th->th_seq, sc->sc_irs) ||
888 	    SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) &&
889 	    !TOEPCB_ISSET(sc)) {
890 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
891 			log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment "
892 			    "rejected\n", s, __func__, th->th_seq, sc->sc_irs);
893 		goto failed;
894 	}
895 
896 	if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
897 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
898 			log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
899 			    "segment rejected\n", s, __func__);
900 		goto failed;
901 	}
902 	/*
903 	 * If timestamps were negotiated the reflected timestamp
904 	 * must be equal to what we actually sent in the SYN|ACK.
905 	 */
906 	if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts &&
907 	    !TOEPCB_ISSET(sc)) {
908 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
909 			log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
910 			    "segment rejected\n",
911 			    s, __func__, to->to_tsecr, sc->sc_ts);
912 		goto failed;
913 	}
914 
915 	*lsop = syncache_socket(sc, *lsop, m);
916 
917 	if (*lsop == NULL)
918 		TCPSTAT_INC(tcps_sc_aborted);
919 	else
920 		TCPSTAT_INC(tcps_sc_completed);
921 
922 /* how do we find the inp for the new socket? */
923 	if (sc != &scs)
924 		syncache_free(sc);
925 	return (1);
926 failed:
927 	if (sc != NULL && sc != &scs)
928 		syncache_free(sc);
929 	if (s != NULL)
930 		free(s, M_TCPLOG);
931 	*lsop = NULL;
932 	return (0);
933 }
934 
935 int
936 tcp_offload_syncache_expand(struct in_conninfo *inc, struct tcpopt *to,
937     struct tcphdr *th, struct socket **lsop, struct mbuf *m)
938 {
939 	INIT_VNET_INET(curvnet);
940 	int rc;
941 
942 	INP_INFO_WLOCK(&V_tcbinfo);
943 	rc = syncache_expand(inc, to, th, lsop, m);
944 	INP_INFO_WUNLOCK(&V_tcbinfo);
945 
946 	return (rc);
947 }
948 
949 /*
950  * Given a LISTEN socket and an inbound SYN request, add
951  * this to the syn cache, and send back a segment:
952  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
953  * to the source.
954  *
955  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
956  * Doing so would require that we hold onto the data and deliver it
957  * to the application.  However, if we are the target of a SYN-flood
958  * DoS attack, an attacker could send data which would eventually
959  * consume all available buffer space if it were ACKed.  By not ACKing
960  * the data, we avoid this DoS scenario.
961  */
962 static void
963 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
964     struct inpcb *inp, struct socket **lsop, struct mbuf *m,
965     struct toe_usrreqs *tu, void *toepcb)
966 {
967 	INIT_VNET_INET(inp->inp_vnet);
968 	struct tcpcb *tp;
969 	struct socket *so;
970 	struct syncache *sc = NULL;
971 	struct syncache_head *sch;
972 	struct mbuf *ipopts = NULL;
973 	u_int32_t flowtmp;
974 	int win, sb_hiwat, ip_ttl, ip_tos, noopt;
975 	char *s;
976 #ifdef INET6
977 	int autoflowlabel = 0;
978 #endif
979 #ifdef MAC
980 	struct label *maclabel;
981 #endif
982 	struct syncache scs;
983 	struct ucred *cred;
984 
985 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
986 	INP_WLOCK_ASSERT(inp);			/* listen socket */
987 	KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
988 	    ("%s: unexpected tcp flags", __func__));
989 
990 	/*
991 	 * Combine all so/tp operations very early to drop the INP lock as
992 	 * soon as possible.
993 	 */
994 	so = *lsop;
995 	tp = sototcpcb(so);
996 	cred = crhold(so->so_cred);
997 
998 #ifdef INET6
999 	if ((inc->inc_flags & INC_ISIPV6) &&
1000 	    (inp->inp_flags & IN6P_AUTOFLOWLABEL))
1001 		autoflowlabel = 1;
1002 #endif
1003 	ip_ttl = inp->inp_ip_ttl;
1004 	ip_tos = inp->inp_ip_tos;
1005 	win = sbspace(&so->so_rcv);
1006 	sb_hiwat = so->so_rcv.sb_hiwat;
1007 	noopt = (tp->t_flags & TF_NOOPT);
1008 
1009 	/* By the time we drop the lock these should no longer be used. */
1010 	so = NULL;
1011 	tp = NULL;
1012 
1013 #ifdef MAC
1014 	if (mac_syncache_init(&maclabel) != 0) {
1015 		INP_WUNLOCK(inp);
1016 		INP_INFO_WUNLOCK(&V_tcbinfo);
1017 		goto done;
1018 	} else
1019 		mac_syncache_create(maclabel, inp);
1020 #endif
1021 	INP_WUNLOCK(inp);
1022 	INP_INFO_WUNLOCK(&V_tcbinfo);
1023 
1024 	/*
1025 	 * Remember the IP options, if any.
1026 	 */
1027 #ifdef INET6
1028 	if (!(inc->inc_flags & INC_ISIPV6))
1029 #endif
1030 		ipopts = (m) ? ip_srcroute(m) : NULL;
1031 
1032 	/*
1033 	 * See if we already have an entry for this connection.
1034 	 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1035 	 *
1036 	 * XXX: should the syncache be re-initialized with the contents
1037 	 * of the new SYN here (which may have different options?)
1038 	 *
1039 	 * XXX: We do not check the sequence number to see if this is a
1040 	 * real retransmit or a new connection attempt.  The question is
1041 	 * how to handle such a case; either ignore it as spoofed, or
1042 	 * drop the current entry and create a new one?
1043 	 */
1044 	sc = syncache_lookup(inc, &sch);	/* returns locked entry */
1045 	SCH_LOCK_ASSERT(sch);
1046 	if (sc != NULL) {
1047 #ifndef TCP_OFFLOAD_DISABLE
1048 		if (sc->sc_tu)
1049 			sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT,
1050 			    sc->sc_toepcb);
1051 #endif
1052 		TCPSTAT_INC(tcps_sc_dupsyn);
1053 		if (ipopts) {
1054 			/*
1055 			 * If we were remembering a previous source route,
1056 			 * forget it and use the new one we've been given.
1057 			 */
1058 			if (sc->sc_ipopts)
1059 				(void) m_free(sc->sc_ipopts);
1060 			sc->sc_ipopts = ipopts;
1061 		}
1062 		/*
1063 		 * Update timestamp if present.
1064 		 */
1065 		if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
1066 			sc->sc_tsreflect = to->to_tsval;
1067 		else
1068 			sc->sc_flags &= ~SCF_TIMESTAMP;
1069 #ifdef MAC
1070 		/*
1071 		 * Since we have already unconditionally allocated label
1072 		 * storage, free it up.  The syncache entry will already
1073 		 * have an initialized label we can use.
1074 		 */
1075 		mac_syncache_destroy(&maclabel);
1076 #endif
1077 		/* Retransmit SYN|ACK and reset retransmit count. */
1078 		if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
1079 			log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
1080 			    "resetting timer and retransmitting SYN|ACK\n",
1081 			    s, __func__);
1082 			free(s, M_TCPLOG);
1083 		}
1084 		if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) {
1085 			sc->sc_rxmits = 0;
1086 			syncache_timeout(sc, sch, 1);
1087 			TCPSTAT_INC(tcps_sndacks);
1088 			TCPSTAT_INC(tcps_sndtotal);
1089 		}
1090 		SCH_UNLOCK(sch);
1091 		goto done;
1092 	}
1093 
1094 	sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1095 	if (sc == NULL) {
1096 		/*
1097 		 * The zone allocator couldn't provide more entries.
1098 		 * Treat this as if the cache was full; drop the oldest
1099 		 * entry and insert the new one.
1100 		 */
1101 		TCPSTAT_INC(tcps_sc_zonefail);
1102 		if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1103 			syncache_drop(sc, sch);
1104 		sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1105 		if (sc == NULL) {
1106 			if (V_tcp_syncookies) {
1107 				bzero(&scs, sizeof(scs));
1108 				sc = &scs;
1109 			} else {
1110 				SCH_UNLOCK(sch);
1111 				if (ipopts)
1112 					(void) m_free(ipopts);
1113 				goto done;
1114 			}
1115 		}
1116 	}
1117 
1118 	/*
1119 	 * Fill in the syncache values.
1120 	 */
1121 #ifdef MAC
1122 	sc->sc_label = maclabel;
1123 #endif
1124 	sc->sc_cred = cred;
1125 	cred = NULL;
1126 	sc->sc_ipopts = ipopts;
1127 	/* XXX-BZ this fib assignment is just useless. */
1128 	sc->sc_inc.inc_fibnum = inp->inp_inc.inc_fibnum;
1129 	bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1130 #ifdef INET6
1131 	if (!(inc->inc_flags & INC_ISIPV6))
1132 #endif
1133 	{
1134 		sc->sc_ip_tos = ip_tos;
1135 		sc->sc_ip_ttl = ip_ttl;
1136 	}
1137 #ifndef TCP_OFFLOAD_DISABLE
1138 	sc->sc_tu = tu;
1139 	sc->sc_toepcb = toepcb;
1140 #endif
1141 	sc->sc_irs = th->th_seq;
1142 	sc->sc_iss = arc4random();
1143 	sc->sc_flags = 0;
1144 	sc->sc_flowlabel = 0;
1145 
1146 	/*
1147 	 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
1148 	 * win was derived from socket earlier in the function.
1149 	 */
1150 	win = imax(win, 0);
1151 	win = imin(win, TCP_MAXWIN);
1152 	sc->sc_wnd = win;
1153 
1154 	if (V_tcp_do_rfc1323) {
1155 		/*
1156 		 * A timestamp received in a SYN makes
1157 		 * it ok to send timestamp requests and replies.
1158 		 */
1159 		if (to->to_flags & TOF_TS) {
1160 			sc->sc_tsreflect = to->to_tsval;
1161 			sc->sc_ts = ticks;
1162 			sc->sc_flags |= SCF_TIMESTAMP;
1163 		}
1164 		if (to->to_flags & TOF_SCALE) {
1165 			int wscale = 0;
1166 
1167 			/*
1168 			 * Pick the smallest possible scaling factor that
1169 			 * will still allow us to scale up to sb_max, aka
1170 			 * kern.ipc.maxsockbuf.
1171 			 *
1172 			 * We do this because there are broken firewalls that
1173 			 * will corrupt the window scale option, leading to
1174 			 * the other endpoint believing that our advertised
1175 			 * window is unscaled.  At scale factors larger than
1176 			 * 5 the unscaled window will drop below 1500 bytes,
1177 			 * leading to serious problems when traversing these
1178 			 * broken firewalls.
1179 			 *
1180 			 * With the default maxsockbuf of 256K, a scale factor
1181 			 * of 3 will be chosen by this algorithm.  Those who
1182 			 * choose a larger maxsockbuf should watch out
1183 			 * for the compatiblity problems mentioned above.
1184 			 *
1185 			 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1186 			 * or <SYN,ACK>) segment itself is never scaled.
1187 			 */
1188 			while (wscale < TCP_MAX_WINSHIFT &&
1189 			    (TCP_MAXWIN << wscale) < sb_max)
1190 				wscale++;
1191 			sc->sc_requested_r_scale = wscale;
1192 			sc->sc_requested_s_scale = to->to_wscale;
1193 			sc->sc_flags |= SCF_WINSCALE;
1194 		}
1195 	}
1196 #ifdef TCP_SIGNATURE
1197 	/*
1198 	 * If listening socket requested TCP digests, and received SYN
1199 	 * contains the option, flag this in the syncache so that
1200 	 * syncache_respond() will do the right thing with the SYN+ACK.
1201 	 * XXX: Currently we always record the option by default and will
1202 	 * attempt to use it in syncache_respond().
1203 	 */
1204 	if (to->to_flags & TOF_SIGNATURE)
1205 		sc->sc_flags |= SCF_SIGNATURE;
1206 #endif
1207 	if (to->to_flags & TOF_SACKPERM)
1208 		sc->sc_flags |= SCF_SACK;
1209 	if (to->to_flags & TOF_MSS)
1210 		sc->sc_peer_mss = to->to_mss;	/* peer mss may be zero */
1211 	if (noopt)
1212 		sc->sc_flags |= SCF_NOOPT;
1213 	if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn)
1214 		sc->sc_flags |= SCF_ECN;
1215 
1216 	if (V_tcp_syncookies) {
1217 		syncookie_generate(sch, sc, &flowtmp);
1218 #ifdef INET6
1219 		if (autoflowlabel)
1220 			sc->sc_flowlabel = flowtmp;
1221 #endif
1222 	} else {
1223 #ifdef INET6
1224 		if (autoflowlabel)
1225 			sc->sc_flowlabel =
1226 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1227 #endif
1228 	}
1229 	SCH_UNLOCK(sch);
1230 
1231 	/*
1232 	 * Do a standard 3-way handshake.
1233 	 */
1234 	if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) {
1235 		if (V_tcp_syncookies && V_tcp_syncookiesonly && sc != &scs)
1236 			syncache_free(sc);
1237 		else if (sc != &scs)
1238 			syncache_insert(sc, sch);   /* locks and unlocks sch */
1239 		TCPSTAT_INC(tcps_sndacks);
1240 		TCPSTAT_INC(tcps_sndtotal);
1241 	} else {
1242 		if (sc != &scs)
1243 			syncache_free(sc);
1244 		TCPSTAT_INC(tcps_sc_dropped);
1245 	}
1246 
1247 done:
1248 	if (cred != NULL)
1249 		crfree(cred);
1250 #ifdef MAC
1251 	if (sc == &scs)
1252 		mac_syncache_destroy(&maclabel);
1253 #endif
1254 	if (m) {
1255 
1256 		*lsop = NULL;
1257 		m_freem(m);
1258 	}
1259 }
1260 
1261 static int
1262 syncache_respond(struct syncache *sc)
1263 {
1264 	INIT_VNET_INET(curvnet);
1265 	struct ip *ip = NULL;
1266 	struct mbuf *m;
1267 	struct tcphdr *th;
1268 	int optlen, error;
1269 	u_int16_t hlen, tlen, mssopt;
1270 	struct tcpopt to;
1271 #ifdef INET6
1272 	struct ip6_hdr *ip6 = NULL;
1273 #endif
1274 
1275 	hlen =
1276 #ifdef INET6
1277 	       (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) :
1278 #endif
1279 		sizeof(struct ip);
1280 	tlen = hlen + sizeof(struct tcphdr);
1281 
1282 	/* Determine MSS we advertize to other end of connection. */
1283 	mssopt = tcp_mssopt(&sc->sc_inc);
1284 	if (sc->sc_peer_mss)
1285 		mssopt = max( min(sc->sc_peer_mss, mssopt), V_tcp_minmss);
1286 
1287 	/* XXX: Assume that the entire packet will fit in a header mbuf. */
1288 	KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1289 	    ("syncache: mbuf too small"));
1290 
1291 	/* Create the IP+TCP header from scratch. */
1292 	m = m_gethdr(M_DONTWAIT, MT_DATA);
1293 	if (m == NULL)
1294 		return (ENOBUFS);
1295 #ifdef MAC
1296 	mac_syncache_create_mbuf(sc->sc_label, m);
1297 #endif
1298 	m->m_data += max_linkhdr;
1299 	m->m_len = tlen;
1300 	m->m_pkthdr.len = tlen;
1301 	m->m_pkthdr.rcvif = NULL;
1302 
1303 #ifdef INET6
1304 	if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1305 		ip6 = mtod(m, struct ip6_hdr *);
1306 		ip6->ip6_vfc = IPV6_VERSION;
1307 		ip6->ip6_nxt = IPPROTO_TCP;
1308 		ip6->ip6_src = sc->sc_inc.inc6_laddr;
1309 		ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1310 		ip6->ip6_plen = htons(tlen - hlen);
1311 		/* ip6_hlim is set after checksum */
1312 		ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1313 		ip6->ip6_flow |= sc->sc_flowlabel;
1314 
1315 		th = (struct tcphdr *)(ip6 + 1);
1316 	} else
1317 #endif
1318 	{
1319 		ip = mtod(m, struct ip *);
1320 		ip->ip_v = IPVERSION;
1321 		ip->ip_hl = sizeof(struct ip) >> 2;
1322 		ip->ip_len = tlen;
1323 		ip->ip_id = 0;
1324 		ip->ip_off = 0;
1325 		ip->ip_sum = 0;
1326 		ip->ip_p = IPPROTO_TCP;
1327 		ip->ip_src = sc->sc_inc.inc_laddr;
1328 		ip->ip_dst = sc->sc_inc.inc_faddr;
1329 		ip->ip_ttl = sc->sc_ip_ttl;
1330 		ip->ip_tos = sc->sc_ip_tos;
1331 
1332 		/*
1333 		 * See if we should do MTU discovery.  Route lookups are
1334 		 * expensive, so we will only unset the DF bit if:
1335 		 *
1336 		 *	1) path_mtu_discovery is disabled
1337 		 *	2) the SCF_UNREACH flag has been set
1338 		 */
1339 		if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1340 		       ip->ip_off |= IP_DF;
1341 
1342 		th = (struct tcphdr *)(ip + 1);
1343 	}
1344 	th->th_sport = sc->sc_inc.inc_lport;
1345 	th->th_dport = sc->sc_inc.inc_fport;
1346 
1347 	th->th_seq = htonl(sc->sc_iss);
1348 	th->th_ack = htonl(sc->sc_irs + 1);
1349 	th->th_off = sizeof(struct tcphdr) >> 2;
1350 	th->th_x2 = 0;
1351 	th->th_flags = TH_SYN|TH_ACK;
1352 	th->th_win = htons(sc->sc_wnd);
1353 	th->th_urp = 0;
1354 
1355 	if (sc->sc_flags & SCF_ECN) {
1356 		th->th_flags |= TH_ECE;
1357 		TCPSTAT_INC(tcps_ecn_shs);
1358 	}
1359 
1360 	/* Tack on the TCP options. */
1361 	if ((sc->sc_flags & SCF_NOOPT) == 0) {
1362 		to.to_flags = 0;
1363 
1364 		to.to_mss = mssopt;
1365 		to.to_flags = TOF_MSS;
1366 		if (sc->sc_flags & SCF_WINSCALE) {
1367 			to.to_wscale = sc->sc_requested_r_scale;
1368 			to.to_flags |= TOF_SCALE;
1369 		}
1370 		if (sc->sc_flags & SCF_TIMESTAMP) {
1371 			/* Virgin timestamp or TCP cookie enhanced one. */
1372 			to.to_tsval = sc->sc_ts;
1373 			to.to_tsecr = sc->sc_tsreflect;
1374 			to.to_flags |= TOF_TS;
1375 		}
1376 		if (sc->sc_flags & SCF_SACK)
1377 			to.to_flags |= TOF_SACKPERM;
1378 #ifdef TCP_SIGNATURE
1379 		if (sc->sc_flags & SCF_SIGNATURE)
1380 			to.to_flags |= TOF_SIGNATURE;
1381 #endif
1382 		optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1383 
1384 		/* Adjust headers by option size. */
1385 		th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1386 		m->m_len += optlen;
1387 		m->m_pkthdr.len += optlen;
1388 
1389 #ifdef TCP_SIGNATURE
1390 		if (sc->sc_flags & SCF_SIGNATURE)
1391 			tcp_signature_compute(m, 0, 0, optlen,
1392 			    to.to_signature, IPSEC_DIR_OUTBOUND);
1393 #endif
1394 #ifdef INET6
1395 		if (sc->sc_inc.inc_flags & INC_ISIPV6)
1396 			ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1397 		else
1398 #endif
1399 			ip->ip_len += optlen;
1400 	} else
1401 		optlen = 0;
1402 
1403 #ifdef INET6
1404 	if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1405 		th->th_sum = 0;
1406 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
1407 				       tlen + optlen - hlen);
1408 		ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1409 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1410 	} else
1411 #endif
1412 	{
1413 		th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1414 		    htons(tlen + optlen - hlen + IPPROTO_TCP));
1415 		m->m_pkthdr.csum_flags = CSUM_TCP;
1416 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1417 		error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1418 	}
1419 	return (error);
1420 }
1421 
1422 void
1423 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1424     struct inpcb *inp, struct socket **lsop, struct mbuf *m)
1425 {
1426 	_syncache_add(inc, to, th, inp, lsop, m, NULL, NULL);
1427 }
1428 
1429 void
1430 tcp_offload_syncache_add(struct in_conninfo *inc, struct tcpopt *to,
1431     struct tcphdr *th, struct inpcb *inp, struct socket **lsop,
1432     struct toe_usrreqs *tu, void *toepcb)
1433 {
1434 	INIT_VNET_INET(curvnet);
1435 
1436 	INP_INFO_WLOCK(&V_tcbinfo);
1437 	INP_WLOCK(inp);
1438 	_syncache_add(inc, to, th, inp, lsop, NULL, tu, toepcb);
1439 }
1440 
1441 /*
1442  * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1443  * receive and to be able to handle SYN floods from bogus source addresses
1444  * (where we will never receive any reply).  SYN floods try to exhaust all
1445  * our memory and available slots in the SYN cache table to cause a denial
1446  * of service to legitimate users of the local host.
1447  *
1448  * The idea of SYN cookies is to encode and include all necessary information
1449  * about the connection setup state within the SYN-ACK we send back and thus
1450  * to get along without keeping any local state until the ACK to the SYN-ACK
1451  * arrives (if ever).  Everything we need to know should be available from
1452  * the information we encoded in the SYN-ACK.
1453  *
1454  * More information about the theory behind SYN cookies and its first
1455  * discussion and specification can be found at:
1456  *  http://cr.yp.to/syncookies.html    (overview)
1457  *  http://cr.yp.to/syncookies/archive (gory details)
1458  *
1459  * This implementation extends the orginal idea and first implementation
1460  * of FreeBSD by using not only the initial sequence number field to store
1461  * information but also the timestamp field if present.  This way we can
1462  * keep track of the entire state we need to know to recreate the session in
1463  * its original form.  Almost all TCP speakers implement RFC1323 timestamps
1464  * these days.  For those that do not we still have to live with the known
1465  * shortcomings of the ISN only SYN cookies.
1466  *
1467  * Cookie layers:
1468  *
1469  * Initial sequence number we send:
1470  * 31|................................|0
1471  *    DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1472  *    D = MD5 Digest (first dword)
1473  *    M = MSS index
1474  *    R = Rotation of secret
1475  *    P = Odd or Even secret
1476  *
1477  * The MD5 Digest is computed with over following parameters:
1478  *  a) randomly rotated secret
1479  *  b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1480  *  c) the received initial sequence number from remote host
1481  *  d) the rotation offset and odd/even bit
1482  *
1483  * Timestamp we send:
1484  * 31|................................|0
1485  *    DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1486  *    D = MD5 Digest (third dword) (only as filler)
1487  *    S = Requested send window scale
1488  *    R = Requested receive window scale
1489  *    A = SACK allowed
1490  *    5 = TCP-MD5 enabled (not implemented yet)
1491  *    XORed with MD5 Digest (forth dword)
1492  *
1493  * The timestamp isn't cryptographically secure and doesn't need to be.
1494  * The double use of the MD5 digest dwords ties it to a specific remote/
1495  * local host/port, remote initial sequence number and our local time
1496  * limited secret.  A received timestamp is reverted (XORed) and then
1497  * the contained MD5 dword is compared to the computed one to ensure the
1498  * timestamp belongs to the SYN-ACK we sent.  The other parameters may
1499  * have been tampered with but this isn't different from supplying bogus
1500  * values in the SYN in the first place.
1501  *
1502  * Some problems with SYN cookies remain however:
1503  * Consider the problem of a recreated (and retransmitted) cookie.  If the
1504  * original SYN was accepted, the connection is established.  The second
1505  * SYN is inflight, and if it arrives with an ISN that falls within the
1506  * receive window, the connection is killed.
1507  *
1508  * Notes:
1509  * A heuristic to determine when to accept syn cookies is not necessary.
1510  * An ACK flood would cause the syncookie verification to be attempted,
1511  * but a SYN flood causes syncookies to be generated.  Both are of equal
1512  * cost, so there's no point in trying to optimize the ACK flood case.
1513  * Also, if you don't process certain ACKs for some reason, then all someone
1514  * would have to do is launch a SYN and ACK flood at the same time, which
1515  * would stop cookie verification and defeat the entire purpose of syncookies.
1516  */
1517 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1518 
1519 static void
1520 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1521     u_int32_t *flowlabel)
1522 {
1523 	INIT_VNET_INET(curvnet);
1524 	MD5_CTX ctx;
1525 	u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1526 	u_int32_t data;
1527 	u_int32_t *secbits;
1528 	u_int off, pmss, mss;
1529 	int i;
1530 
1531 	SCH_LOCK_ASSERT(sch);
1532 
1533 	/* Which of the two secrets to use. */
1534 	secbits = sch->sch_oddeven ?
1535 			sch->sch_secbits_odd : sch->sch_secbits_even;
1536 
1537 	/* Reseed secret if too old. */
1538 	if (sch->sch_reseed < time_uptime) {
1539 		sch->sch_oddeven = sch->sch_oddeven ? 0 : 1;	/* toggle */
1540 		secbits = sch->sch_oddeven ?
1541 				sch->sch_secbits_odd : sch->sch_secbits_even;
1542 		for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1543 			secbits[i] = arc4random();
1544 		sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1545 	}
1546 
1547 	/* Secret rotation offset. */
1548 	off = sc->sc_iss & 0x7;			/* iss was randomized before */
1549 
1550 	/* Maximum segment size calculation. */
1551 	pmss =
1552 	    max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)),	V_tcp_minmss);
1553 	for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1554 		if (tcp_sc_msstab[mss] <= pmss)
1555 			break;
1556 
1557 	/* Fold parameters and MD5 digest into the ISN we will send. */
1558 	data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1559 	data |= off << 1;	/* secret offset, derived from iss, 3 bits */
1560 	data |= mss << 4;	/* mss, 3 bits */
1561 
1562 	MD5Init(&ctx);
1563 	MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1564 	    SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1565 	MD5Update(&ctx, secbits, off);
1566 	MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1567 	MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1568 	MD5Update(&ctx, &data, sizeof(data));
1569 	MD5Final((u_int8_t *)&md5_buffer, &ctx);
1570 
1571 	data |= (md5_buffer[0] << 7);
1572 	sc->sc_iss = data;
1573 
1574 #ifdef INET6
1575 	*flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1576 #endif
1577 
1578 	/* Additional parameters are stored in the timestamp if present. */
1579 	if (sc->sc_flags & SCF_TIMESTAMP) {
1580 		data =  ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1581 		data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1582 		data |= sc->sc_requested_s_scale << 2;  /* SWIN scale, 4 bits */
1583 		data |= sc->sc_requested_r_scale << 6;  /* RWIN scale, 4 bits */
1584 		data |= md5_buffer[2] << 10;		/* more digest bits */
1585 		data ^= md5_buffer[3];
1586 		sc->sc_ts = data;
1587 		sc->sc_tsoff = data - ticks;		/* after XOR */
1588 	}
1589 
1590 	TCPSTAT_INC(tcps_sc_sendcookie);
1591 }
1592 
1593 static struct syncache *
1594 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1595     struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1596     struct socket *so)
1597 {
1598 	INIT_VNET_INET(curvnet);
1599 	MD5_CTX ctx;
1600 	u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1601 	u_int32_t data = 0;
1602 	u_int32_t *secbits;
1603 	tcp_seq ack, seq;
1604 	int off, mss, wnd, flags;
1605 
1606 	SCH_LOCK_ASSERT(sch);
1607 
1608 	/*
1609 	 * Pull information out of SYN-ACK/ACK and
1610 	 * revert sequence number advances.
1611 	 */
1612 	ack = th->th_ack - 1;
1613 	seq = th->th_seq - 1;
1614 	off = (ack >> 1) & 0x7;
1615 	mss = (ack >> 4) & 0x7;
1616 	flags = ack & 0x7f;
1617 
1618 	/* Which of the two secrets to use. */
1619 	secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1620 
1621 	/*
1622 	 * The secret wasn't updated for the lifetime of a syncookie,
1623 	 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1624 	 */
1625 	if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) {
1626 		return (NULL);
1627 	}
1628 
1629 	/* Recompute the digest so we can compare it. */
1630 	MD5Init(&ctx);
1631 	MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1632 	    SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1633 	MD5Update(&ctx, secbits, off);
1634 	MD5Update(&ctx, inc, sizeof(*inc));
1635 	MD5Update(&ctx, &seq, sizeof(seq));
1636 	MD5Update(&ctx, &flags, sizeof(flags));
1637 	MD5Final((u_int8_t *)&md5_buffer, &ctx);
1638 
1639 	/* Does the digest part of or ACK'ed ISS match? */
1640 	if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1641 		return (NULL);
1642 
1643 	/* Does the digest part of our reflected timestamp match? */
1644 	if (to->to_flags & TOF_TS) {
1645 		data = md5_buffer[3] ^ to->to_tsecr;
1646 		if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1647 			return (NULL);
1648 	}
1649 
1650 	/* Fill in the syncache values. */
1651 	bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1652 	sc->sc_ipopts = NULL;
1653 
1654 	sc->sc_irs = seq;
1655 	sc->sc_iss = ack;
1656 
1657 #ifdef INET6
1658 	if (inc->inc_flags & INC_ISIPV6) {
1659 		if (sotoinpcb(so)->inp_flags & IN6P_AUTOFLOWLABEL)
1660 			sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1661 	} else
1662 #endif
1663 	{
1664 		sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1665 		sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1666 	}
1667 
1668 	/* Additional parameters that were encoded in the timestamp. */
1669 	if (data) {
1670 		sc->sc_flags |= SCF_TIMESTAMP;
1671 		sc->sc_tsreflect = to->to_tsval;
1672 		sc->sc_ts = to->to_tsecr;
1673 		sc->sc_tsoff = to->to_tsecr - ticks;
1674 		sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1675 		sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1676 		sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1677 		    TCP_MAX_WINSHIFT);
1678 		sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1679 		    TCP_MAX_WINSHIFT);
1680 		if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1681 			sc->sc_flags |= SCF_WINSCALE;
1682 	} else
1683 		sc->sc_flags |= SCF_NOOPT;
1684 
1685 	wnd = sbspace(&so->so_rcv);
1686 	wnd = imax(wnd, 0);
1687 	wnd = imin(wnd, TCP_MAXWIN);
1688 	sc->sc_wnd = wnd;
1689 
1690 	sc->sc_rxmits = 0;
1691 	sc->sc_peer_mss = tcp_sc_msstab[mss];
1692 
1693 	TCPSTAT_INC(tcps_sc_recvcookie);
1694 	return (sc);
1695 }
1696 
1697 /*
1698  * Returns the current number of syncache entries.  This number
1699  * will probably change before you get around to calling
1700  * syncache_pcblist.
1701  */
1702 
1703 int
1704 syncache_pcbcount(void)
1705 {
1706 	INIT_VNET_INET(curvnet);
1707 	struct syncache_head *sch;
1708 	int count, i;
1709 
1710 	for (count = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1711 		/* No need to lock for a read. */
1712 		sch = &V_tcp_syncache.hashbase[i];
1713 		count += sch->sch_length;
1714 	}
1715 	return count;
1716 }
1717 
1718 /*
1719  * Exports the syncache entries to userland so that netstat can display
1720  * them alongside the other sockets.  This function is intended to be
1721  * called only from tcp_pcblist.
1722  *
1723  * Due to concurrency on an active system, the number of pcbs exported
1724  * may have no relation to max_pcbs.  max_pcbs merely indicates the
1725  * amount of space the caller allocated for this function to use.
1726  */
1727 int
1728 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
1729 {
1730 	INIT_VNET_INET(curvnet);
1731 	struct xtcpcb xt;
1732 	struct syncache *sc;
1733 	struct syncache_head *sch;
1734 	int count, error, i;
1735 
1736 	for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1737 		sch = &V_tcp_syncache.hashbase[i];
1738 		SCH_LOCK(sch);
1739 		TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
1740 			if (count >= max_pcbs) {
1741 				SCH_UNLOCK(sch);
1742 				goto exit;
1743 			}
1744 			if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0)
1745 				continue;
1746 			bzero(&xt, sizeof(xt));
1747 			xt.xt_len = sizeof(xt);
1748 			if (sc->sc_inc.inc_flags & INC_ISIPV6)
1749 				xt.xt_inp.inp_vflag = INP_IPV6;
1750 			else
1751 				xt.xt_inp.inp_vflag = INP_IPV4;
1752 			bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
1753 			xt.xt_tp.t_inpcb = &xt.xt_inp;
1754 			xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
1755 			xt.xt_socket.xso_protocol = IPPROTO_TCP;
1756 			xt.xt_socket.xso_len = sizeof (struct xsocket);
1757 			xt.xt_socket.so_type = SOCK_STREAM;
1758 			xt.xt_socket.so_state = SS_ISCONNECTING;
1759 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1760 			if (error) {
1761 				SCH_UNLOCK(sch);
1762 				goto exit;
1763 			}
1764 			count++;
1765 		}
1766 		SCH_UNLOCK(sch);
1767 	}
1768 exit:
1769 	*pcbs_exported = count;
1770 	return error;
1771 }
1772