xref: /freebsd/sys/netinet/tcp_syncache.c (revision 9336e0699bda8a301cd2bfa37106b6ec5e32012e)
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 
57 #include <vm/uma.h>
58 
59 #include <net/if.h>
60 #include <net/route.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/ip.h>
65 #include <netinet/in_var.h>
66 #include <netinet/in_pcb.h>
67 #include <netinet/ip_var.h>
68 #include <netinet/ip_options.h>
69 #ifdef INET6
70 #include <netinet/ip6.h>
71 #include <netinet/icmp6.h>
72 #include <netinet6/nd6.h>
73 #include <netinet6/ip6_var.h>
74 #include <netinet6/in6_pcb.h>
75 #endif
76 #include <netinet/tcp.h>
77 #include <netinet/tcp_fsm.h>
78 #include <netinet/tcp_seq.h>
79 #include <netinet/tcp_timer.h>
80 #include <netinet/tcp_var.h>
81 #include <netinet/tcp_syncache.h>
82 #include <netinet/tcp_offload.h>
83 #ifdef INET6
84 #include <netinet6/tcp6_var.h>
85 #endif
86 
87 #ifdef IPSEC
88 #include <netipsec/ipsec.h>
89 #ifdef INET6
90 #include <netipsec/ipsec6.h>
91 #endif
92 #include <netipsec/key.h>
93 #endif /*IPSEC*/
94 
95 #include <machine/in_cksum.h>
96 
97 #include <security/mac/mac_framework.h>
98 
99 static int tcp_syncookies = 1;
100 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
101     &tcp_syncookies, 0,
102     "Use TCP SYN cookies if the syncache overflows");
103 
104 static int tcp_syncookiesonly = 0;
105 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW,
106     &tcp_syncookiesonly, 0,
107     "Use only TCP SYN cookies");
108 
109 #define	SYNCOOKIE_SECRET_SIZE	8	/* dwords */
110 #define	SYNCOOKIE_LIFETIME	16	/* seconds */
111 
112 struct syncache {
113 	TAILQ_ENTRY(syncache)	sc_hash;
114 	struct		in_conninfo sc_inc;	/* addresses */
115 	int		sc_rxttime;		/* retransmit time */
116 	u_int16_t	sc_rxmits;		/* retransmit counter */
117 
118 	u_int32_t	sc_tsreflect;		/* timestamp to reflect */
119 	u_int32_t	sc_ts;			/* our timestamp to send */
120 	u_int32_t	sc_tsoff;		/* ts offset w/ syncookies */
121 	u_int32_t	sc_flowlabel;		/* IPv6 flowlabel */
122 	tcp_seq		sc_irs;			/* seq from peer */
123 	tcp_seq		sc_iss;			/* our ISS */
124 	struct		mbuf *sc_ipopts;	/* source route */
125 
126 	u_int16_t	sc_peer_mss;		/* peer's MSS */
127 	u_int16_t	sc_wnd;			/* advertised window */
128 	u_int8_t	sc_ip_ttl;		/* IPv4 TTL */
129 	u_int8_t	sc_ip_tos;		/* IPv4 TOS */
130 	u_int8_t	sc_requested_s_scale:4,
131 			sc_requested_r_scale:4;
132 	u_int8_t	sc_flags;
133 #define SCF_NOOPT	0x01			/* no TCP options */
134 #define SCF_WINSCALE	0x02			/* negotiated window scaling */
135 #define SCF_TIMESTAMP	0x04			/* negotiated timestamps */
136 						/* MSS is implicit */
137 #define SCF_UNREACH	0x10			/* icmp unreachable received */
138 #define SCF_SIGNATURE	0x20			/* send MD5 digests */
139 #define SCF_SACK	0x80			/* send SACK option */
140 #ifndef TCP_OFFLOAD_DISABLE
141 	struct toe_usrreqs *sc_tu;		/* TOE operations */
142 	void 		*sc_toepcb;		/* TOE protocol block */
143 #endif
144 #ifdef MAC
145 	struct label	*sc_label;		/* MAC label reference */
146 #endif
147 };
148 
149 #ifdef TCP_OFFLOAD_DISABLE
150 #define TOEPCB_ISSET(sc) (0)
151 #else
152 #define TOEPCB_ISSET(sc) ((sc)->sc_toepcb != NULL)
153 #endif
154 
155 
156 struct syncache_head {
157 	struct mtx	sch_mtx;
158 	TAILQ_HEAD(sch_head, syncache)	sch_bucket;
159 	struct callout	sch_timer;
160 	int		sch_nextc;
161 	u_int		sch_length;
162 	u_int		sch_oddeven;
163 	u_int32_t	sch_secbits_odd[SYNCOOKIE_SECRET_SIZE];
164 	u_int32_t	sch_secbits_even[SYNCOOKIE_SECRET_SIZE];
165 	u_int		sch_reseed;		/* time_uptime, seconds */
166 };
167 
168 static void	 syncache_drop(struct syncache *, struct syncache_head *);
169 static void	 syncache_free(struct syncache *);
170 static void	 syncache_insert(struct syncache *, struct syncache_head *);
171 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
172 static int	 syncache_respond(struct syncache *);
173 static struct	 socket *syncache_socket(struct syncache *, struct socket *,
174 		    struct mbuf *m);
175 static void	 syncache_timeout(struct syncache *sc, struct syncache_head *sch,
176 		    int docallout);
177 static void	 syncache_timer(void *);
178 static void	 syncookie_generate(struct syncache_head *, struct syncache *,
179 		    u_int32_t *);
180 static struct syncache
181 		*syncookie_lookup(struct in_conninfo *, struct syncache_head *,
182 		    struct syncache *, struct tcpopt *, struct tcphdr *,
183 		    struct socket *);
184 
185 /*
186  * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
187  * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds,
188  * the odds are that the user has given up attempting to connect by then.
189  */
190 #define SYNCACHE_MAXREXMTS		3
191 
192 /* Arbitrary values */
193 #define TCP_SYNCACHE_HASHSIZE		512
194 #define TCP_SYNCACHE_BUCKETLIMIT	30
195 
196 struct tcp_syncache {
197 	struct	syncache_head *hashbase;
198 	uma_zone_t zone;
199 	u_int	hashsize;
200 	u_int	hashmask;
201 	u_int	bucket_limit;
202 	u_int	cache_count;		/* XXX: unprotected */
203 	u_int	cache_limit;
204 	u_int	rexmt_limit;
205 	u_int	hash_secret;
206 };
207 static struct tcp_syncache tcp_syncache;
208 
209 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
210 
211 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
212      &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
213 
214 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
215      &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
216 
217 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
218      &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
219 
220 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
221      &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
222 
223 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
224      &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
225 
226 int	tcp_sc_rst_sock_fail = 1;
227 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail, CTLFLAG_RW,
228      &tcp_sc_rst_sock_fail, 0, "Send reset on socket allocation failure");
229 
230 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
231 
232 #define SYNCACHE_HASH(inc, mask)					\
233 	((tcp_syncache.hash_secret ^					\
234 	  (inc)->inc_faddr.s_addr ^					\
235 	  ((inc)->inc_faddr.s_addr >> 16) ^				\
236 	  (inc)->inc_fport ^ (inc)->inc_lport) & mask)
237 
238 #define SYNCACHE_HASH6(inc, mask)					\
239 	((tcp_syncache.hash_secret ^					\
240 	  (inc)->inc6_faddr.s6_addr32[0] ^				\
241 	  (inc)->inc6_faddr.s6_addr32[3] ^				\
242 	  (inc)->inc_fport ^ (inc)->inc_lport) & mask)
243 
244 #define ENDPTS_EQ(a, b) (						\
245 	(a)->ie_fport == (b)->ie_fport &&				\
246 	(a)->ie_lport == (b)->ie_lport &&				\
247 	(a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr &&			\
248 	(a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr			\
249 )
250 
251 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
252 
253 #define	SCH_LOCK(sch)		mtx_lock(&(sch)->sch_mtx)
254 #define	SCH_UNLOCK(sch)		mtx_unlock(&(sch)->sch_mtx)
255 #define	SCH_LOCK_ASSERT(sch)	mtx_assert(&(sch)->sch_mtx, MA_OWNED)
256 
257 /*
258  * Requires the syncache entry to be already removed from the bucket list.
259  */
260 static void
261 syncache_free(struct syncache *sc)
262 {
263 	if (sc->sc_ipopts)
264 		(void) m_free(sc->sc_ipopts);
265 #ifdef MAC
266 	mac_syncache_destroy(&sc->sc_label);
267 #endif
268 
269 	uma_zfree(tcp_syncache.zone, sc);
270 }
271 
272 void
273 syncache_init(void)
274 {
275 	int i;
276 
277 	tcp_syncache.cache_count = 0;
278 	tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
279 	tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
280 	tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
281 	tcp_syncache.hash_secret = arc4random();
282 
283 	TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
284 	    &tcp_syncache.hashsize);
285 	TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
286 	    &tcp_syncache.bucket_limit);
287 	if (!powerof2(tcp_syncache.hashsize) || tcp_syncache.hashsize == 0) {
288 		printf("WARNING: syncache hash size is not a power of 2.\n");
289 		tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
290 	}
291 	tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
292 
293 	/* Set limits. */
294 	tcp_syncache.cache_limit =
295 	    tcp_syncache.hashsize * tcp_syncache.bucket_limit;
296 	TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
297 	    &tcp_syncache.cache_limit);
298 
299 	/* Allocate the hash table. */
300 	MALLOC(tcp_syncache.hashbase, struct syncache_head *,
301 	    tcp_syncache.hashsize * sizeof(struct syncache_head),
302 	    M_SYNCACHE, M_WAITOK | M_ZERO);
303 
304 	/* Initialize the hash buckets. */
305 	for (i = 0; i < tcp_syncache.hashsize; i++) {
306 		TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket);
307 		mtx_init(&tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
308 			 NULL, MTX_DEF);
309 		callout_init_mtx(&tcp_syncache.hashbase[i].sch_timer,
310 			 &tcp_syncache.hashbase[i].sch_mtx, 0);
311 		tcp_syncache.hashbase[i].sch_length = 0;
312 	}
313 
314 	/* Create the syncache entry zone. */
315 	tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
316 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
317 	uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit);
318 }
319 
320 /*
321  * Inserts a syncache entry into the specified bucket row.
322  * Locks and unlocks the syncache_head autonomously.
323  */
324 static void
325 syncache_insert(struct syncache *sc, struct syncache_head *sch)
326 {
327 	struct syncache *sc2;
328 
329 	SCH_LOCK(sch);
330 
331 	/*
332 	 * Make sure that we don't overflow the per-bucket limit.
333 	 * If the bucket is full, toss the oldest element.
334 	 */
335 	if (sch->sch_length >= tcp_syncache.bucket_limit) {
336 		KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
337 			("sch->sch_length incorrect"));
338 		sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
339 		syncache_drop(sc2, sch);
340 		tcpstat.tcps_sc_bucketoverflow++;
341 	}
342 
343 	/* Put it into the bucket. */
344 	TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
345 	sch->sch_length++;
346 
347 	/* Reinitialize the bucket row's timer. */
348 	if (sch->sch_length == 1)
349 		sch->sch_nextc = ticks + INT_MAX;
350 	syncache_timeout(sc, sch, 1);
351 
352 	SCH_UNLOCK(sch);
353 
354 	tcp_syncache.cache_count++;
355 	tcpstat.tcps_sc_added++;
356 }
357 
358 /*
359  * Remove and free entry from syncache bucket row.
360  * Expects locked syncache head.
361  */
362 static void
363 syncache_drop(struct syncache *sc, struct syncache_head *sch)
364 {
365 
366 	SCH_LOCK_ASSERT(sch);
367 
368 	TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
369 	sch->sch_length--;
370 
371 #ifndef TCP_OFFLOAD_DISABLE
372 	if (sc->sc_tu)
373 		sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb);
374 #endif
375 	syncache_free(sc);
376 	tcp_syncache.cache_count--;
377 }
378 
379 /*
380  * Engage/reengage time on bucket row.
381  */
382 static void
383 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
384 {
385 	sc->sc_rxttime = ticks +
386 		TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
387 	sc->sc_rxmits++;
388 	if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
389 		sch->sch_nextc = sc->sc_rxttime;
390 		if (docallout)
391 			callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
392 			    syncache_timer, (void *)sch);
393 	}
394 }
395 
396 /*
397  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
398  * If we have retransmitted an entry the maximum number of times, expire it.
399  * One separate timer for each bucket row.
400  */
401 static void
402 syncache_timer(void *xsch)
403 {
404 	struct syncache_head *sch = (struct syncache_head *)xsch;
405 	struct syncache *sc, *nsc;
406 	int tick = ticks;
407 	char *s;
408 
409 	/* NB: syncache_head has already been locked by the callout. */
410 	SCH_LOCK_ASSERT(sch);
411 
412 	/*
413 	 * In the following cycle we may remove some entries and/or
414 	 * advance some timeouts, so re-initialize the bucket timer.
415 	 */
416 	sch->sch_nextc = tick + INT_MAX;
417 
418 	TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
419 		/*
420 		 * We do not check if the listen socket still exists
421 		 * and accept the case where the listen socket may be
422 		 * gone by the time we resend the SYN/ACK.  We do
423 		 * not expect this to happens often. If it does,
424 		 * then the RST will be sent by the time the remote
425 		 * host does the SYN/ACK->ACK.
426 		 */
427 		if (TSTMP_GT(sc->sc_rxttime, tick)) {
428 			if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
429 				sch->sch_nextc = sc->sc_rxttime;
430 			continue;
431 		}
432 		if (sc->sc_rxmits > tcp_syncache.rexmt_limit) {
433 			if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
434 				log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
435 				    "giving up and removing syncache entry\n",
436 				    s, __func__);
437 				free(s, M_TCPLOG);
438 			}
439 			syncache_drop(sc, sch);
440 			tcpstat.tcps_sc_stale++;
441 			continue;
442 		}
443 		if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
444 			log(LOG_DEBUG, "%s; %s: Response timeout, "
445 			    "retransmitting (%u) SYN|ACK\n",
446 			    s, __func__, sc->sc_rxmits);
447 			free(s, M_TCPLOG);
448 		}
449 
450 		(void) syncache_respond(sc);
451 		tcpstat.tcps_sc_retransmitted++;
452 		syncache_timeout(sc, sch, 0);
453 	}
454 	if (!TAILQ_EMPTY(&(sch)->sch_bucket))
455 		callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
456 			syncache_timer, (void *)(sch));
457 }
458 
459 /*
460  * Find an entry in the syncache.
461  * Returns always with locked syncache_head plus a matching entry or NULL.
462  */
463 struct syncache *
464 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
465 {
466 	struct syncache *sc;
467 	struct syncache_head *sch;
468 
469 #ifdef INET6
470 	if (inc->inc_isipv6) {
471 		sch = &tcp_syncache.hashbase[
472 		    SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
473 		*schp = sch;
474 
475 		SCH_LOCK(sch);
476 
477 		/* Circle through bucket row to find matching entry. */
478 		TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
479 			if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
480 				return (sc);
481 		}
482 	} else
483 #endif
484 	{
485 		sch = &tcp_syncache.hashbase[
486 		    SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
487 		*schp = sch;
488 
489 		SCH_LOCK(sch);
490 
491 		/* Circle through bucket row to find matching entry. */
492 		TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
493 #ifdef INET6
494 			if (sc->sc_inc.inc_isipv6)
495 				continue;
496 #endif
497 			if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
498 				return (sc);
499 		}
500 	}
501 	SCH_LOCK_ASSERT(*schp);
502 	return (NULL);			/* always returns with locked sch */
503 }
504 
505 /*
506  * This function is called when we get a RST for a
507  * non-existent connection, so that we can see if the
508  * connection is in the syn cache.  If it is, zap it.
509  */
510 void
511 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
512 {
513 	struct syncache *sc;
514 	struct syncache_head *sch;
515 	char *s = NULL;
516 
517 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
518 	SCH_LOCK_ASSERT(sch);
519 
520 	/*
521 	 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
522 	 * See RFC 793 page 65, section SEGMENT ARRIVES.
523 	 */
524 	if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
525 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
526 			log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
527 			    "FIN flag set, segment ignored\n", s, __func__);
528 		tcpstat.tcps_badrst++;
529 		goto done;
530 	}
531 
532 	/*
533 	 * No corresponding connection was found in syncache.
534 	 * If syncookies are enabled and possibly exclusively
535 	 * used, or we are under memory pressure, a valid RST
536 	 * may not find a syncache entry.  In that case we're
537 	 * done and no SYN|ACK retransmissions will happen.
538 	 * Otherwise the the RST was misdirected or spoofed.
539 	 */
540 	if (sc == NULL) {
541 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
542 			log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
543 			    "syncache entry (possibly syncookie only), "
544 			    "segment ignored\n", s, __func__);
545 		tcpstat.tcps_badrst++;
546 		goto done;
547 	}
548 
549 	/*
550 	 * If the RST bit is set, check the sequence number to see
551 	 * if this is a valid reset segment.
552 	 * RFC 793 page 37:
553 	 *   In all states except SYN-SENT, all reset (RST) segments
554 	 *   are validated by checking their SEQ-fields.  A reset is
555 	 *   valid if its sequence number is in the window.
556 	 *
557 	 *   The sequence number in the reset segment is normally an
558 	 *   echo of our outgoing acknowlegement numbers, but some hosts
559 	 *   send a reset with the sequence number at the rightmost edge
560 	 *   of our receive window, and we have to handle this case.
561 	 */
562 	if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
563 	    SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
564 		syncache_drop(sc, sch);
565 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
566 			log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
567 			    "connection attempt aborted by remote endpoint\n",
568 			    s, __func__);
569 		tcpstat.tcps_sc_reset++;
570 	} else if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
571 		log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != IRS %u "
572 		    "(+WND %u), segment ignored\n",
573 		    s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
574 		tcpstat.tcps_badrst++;
575 	}
576 
577 done:
578 	if (s != NULL)
579 		free(s, M_TCPLOG);
580 	SCH_UNLOCK(sch);
581 }
582 
583 void
584 syncache_badack(struct in_conninfo *inc)
585 {
586 	struct syncache *sc;
587 	struct syncache_head *sch;
588 
589 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
590 	SCH_LOCK_ASSERT(sch);
591 	if (sc != NULL) {
592 		syncache_drop(sc, sch);
593 		tcpstat.tcps_sc_badack++;
594 	}
595 	SCH_UNLOCK(sch);
596 }
597 
598 void
599 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
600 {
601 	struct syncache *sc;
602 	struct syncache_head *sch;
603 
604 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
605 	SCH_LOCK_ASSERT(sch);
606 	if (sc == NULL)
607 		goto done;
608 
609 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
610 	if (ntohl(th->th_seq) != sc->sc_iss)
611 		goto done;
612 
613 	/*
614 	 * If we've rertransmitted 3 times and this is our second error,
615 	 * we remove the entry.  Otherwise, we allow it to continue on.
616 	 * This prevents us from incorrectly nuking an entry during a
617 	 * spurious network outage.
618 	 *
619 	 * See tcp_notify().
620 	 */
621 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
622 		sc->sc_flags |= SCF_UNREACH;
623 		goto done;
624 	}
625 	syncache_drop(sc, sch);
626 	tcpstat.tcps_sc_unreach++;
627 done:
628 	SCH_UNLOCK(sch);
629 }
630 
631 /*
632  * Build a new TCP socket structure from a syncache entry.
633  */
634 static struct socket *
635 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
636 {
637 	struct inpcb *inp = NULL;
638 	struct socket *so;
639 	struct tcpcb *tp;
640 	char *s;
641 
642 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
643 
644 	/*
645 	 * Ok, create the full blown connection, and set things up
646 	 * as they would have been set up if we had created the
647 	 * connection when the SYN arrived.  If we can't create
648 	 * the connection, abort it.
649 	 */
650 	so = sonewconn(lso, SS_ISCONNECTED);
651 	if (so == NULL) {
652 		/*
653 		 * Drop the connection; we will either send a RST or
654 		 * have the peer retransmit its SYN again after its
655 		 * RTO and try again.
656 		 */
657 		tcpstat.tcps_listendrop++;
658 		if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
659 			log(LOG_DEBUG, "%s; %s: Socket create failed "
660 			    "due to limits or memory shortage\n",
661 			    s, __func__);
662 			free(s, M_TCPLOG);
663 		}
664 		goto abort2;
665 	}
666 #ifdef MAC
667 	SOCK_LOCK(so);
668 	mac_socketpeer_set_from_mbuf(m, so);
669 	SOCK_UNLOCK(so);
670 #endif
671 
672 	inp = sotoinpcb(so);
673 	INP_LOCK(inp);
674 
675 	/* Insert new socket into PCB hash list. */
676 	inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
677 #ifdef INET6
678 	if (sc->sc_inc.inc_isipv6) {
679 		inp->in6p_laddr = sc->sc_inc.inc6_laddr;
680 	} else {
681 		inp->inp_vflag &= ~INP_IPV6;
682 		inp->inp_vflag |= INP_IPV4;
683 #endif
684 		inp->inp_laddr = sc->sc_inc.inc_laddr;
685 #ifdef INET6
686 	}
687 #endif
688 	inp->inp_lport = sc->sc_inc.inc_lport;
689 	if (in_pcbinshash(inp) != 0) {
690 		/*
691 		 * Undo the assignments above if we failed to
692 		 * put the PCB on the hash lists.
693 		 */
694 #ifdef INET6
695 		if (sc->sc_inc.inc_isipv6)
696 			inp->in6p_laddr = in6addr_any;
697 		else
698 #endif
699 			inp->inp_laddr.s_addr = INADDR_ANY;
700 		inp->inp_lport = 0;
701 		goto abort;
702 	}
703 #ifdef IPSEC
704 	/* Copy old policy into new socket's. */
705 	if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
706 		printf("syncache_socket: could not copy policy\n");
707 #endif
708 #ifdef INET6
709 	if (sc->sc_inc.inc_isipv6) {
710 		struct inpcb *oinp = sotoinpcb(lso);
711 		struct in6_addr laddr6;
712 		struct sockaddr_in6 sin6;
713 		/*
714 		 * Inherit socket options from the listening socket.
715 		 * Note that in6p_inputopts are not (and should not be)
716 		 * copied, since it stores previously received options and is
717 		 * used to detect if each new option is different than the
718 		 * previous one and hence should be passed to a user.
719 		 * If we copied in6p_inputopts, a user would not be able to
720 		 * receive options just after calling the accept system call.
721 		 */
722 		inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
723 		if (oinp->in6p_outputopts)
724 			inp->in6p_outputopts =
725 			    ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
726 
727 		sin6.sin6_family = AF_INET6;
728 		sin6.sin6_len = sizeof(sin6);
729 		sin6.sin6_addr = sc->sc_inc.inc6_faddr;
730 		sin6.sin6_port = sc->sc_inc.inc_fport;
731 		sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
732 		laddr6 = inp->in6p_laddr;
733 		if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
734 			inp->in6p_laddr = sc->sc_inc.inc6_laddr;
735 		if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
736 		    thread0.td_ucred)) {
737 			inp->in6p_laddr = laddr6;
738 			goto abort;
739 		}
740 		/* Override flowlabel from in6_pcbconnect. */
741 		inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK;
742 		inp->in6p_flowinfo |= sc->sc_flowlabel;
743 	} else
744 #endif
745 	{
746 		struct in_addr laddr;
747 		struct sockaddr_in sin;
748 
749 		inp->inp_options = (m) ? ip_srcroute(m) : NULL;
750 
751 		if (inp->inp_options == NULL) {
752 			inp->inp_options = sc->sc_ipopts;
753 			sc->sc_ipopts = NULL;
754 		}
755 
756 		sin.sin_family = AF_INET;
757 		sin.sin_len = sizeof(sin);
758 		sin.sin_addr = sc->sc_inc.inc_faddr;
759 		sin.sin_port = sc->sc_inc.inc_fport;
760 		bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
761 		laddr = inp->inp_laddr;
762 		if (inp->inp_laddr.s_addr == INADDR_ANY)
763 			inp->inp_laddr = sc->sc_inc.inc_laddr;
764 		if (in_pcbconnect(inp, (struct sockaddr *)&sin,
765 		    thread0.td_ucred)) {
766 			inp->inp_laddr = laddr;
767 			goto abort;
768 		}
769 	}
770 	tp = intotcpcb(inp);
771 	tp->t_state = TCPS_SYN_RECEIVED;
772 	tp->iss = sc->sc_iss;
773 	tp->irs = sc->sc_irs;
774 	tcp_rcvseqinit(tp);
775 	tcp_sendseqinit(tp);
776 	tp->snd_wl1 = sc->sc_irs;
777 	tp->snd_max = tp->iss + 1;
778 	tp->snd_nxt = tp->iss + 1;
779 	tp->rcv_up = sc->sc_irs + 1;
780 	tp->rcv_wnd = sc->sc_wnd;
781 	tp->rcv_adv += tp->rcv_wnd;
782 	tp->last_ack_sent = tp->rcv_nxt;
783 
784 	tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
785 	if (sc->sc_flags & SCF_NOOPT)
786 		tp->t_flags |= TF_NOOPT;
787 	else {
788 		if (sc->sc_flags & SCF_WINSCALE) {
789 			tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
790 			tp->snd_scale = sc->sc_requested_s_scale;
791 			tp->request_r_scale = sc->sc_requested_r_scale;
792 		}
793 		if (sc->sc_flags & SCF_TIMESTAMP) {
794 			tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
795 			tp->ts_recent = sc->sc_tsreflect;
796 			tp->ts_recent_age = ticks;
797 			tp->ts_offset = sc->sc_tsoff;
798 		}
799 #ifdef TCP_SIGNATURE
800 		if (sc->sc_flags & SCF_SIGNATURE)
801 			tp->t_flags |= TF_SIGNATURE;
802 #endif
803 		if (sc->sc_flags & SCF_SACK)
804 			tp->t_flags |= TF_SACK_PERMIT;
805 	}
806 
807 	/*
808 	 * Set up MSS and get cached values from tcp_hostcache.
809 	 * This might overwrite some of the defaults we just set.
810 	 */
811 	tcp_mss(tp, sc->sc_peer_mss);
812 
813 	/*
814 	 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
815 	 */
816 	if (sc->sc_rxmits)
817 		tp->snd_cwnd = tp->t_maxseg;
818 	tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
819 
820 	INP_UNLOCK(inp);
821 
822 	tcpstat.tcps_accepts++;
823 	return (so);
824 
825 abort:
826 	INP_UNLOCK(inp);
827 abort2:
828 	if (so != NULL)
829 		soabort(so);
830 	return (NULL);
831 }
832 
833 /*
834  * This function gets called when we receive an ACK for a
835  * socket in the LISTEN state.  We look up the connection
836  * in the syncache, and if its there, we pull it out of
837  * the cache and turn it into a full-blown connection in
838  * the SYN-RECEIVED state.
839  */
840 int
841 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
842     struct socket **lsop, struct mbuf *m)
843 {
844 	struct syncache *sc;
845 	struct syncache_head *sch;
846 	struct syncache scs;
847 	char *s;
848 
849 	/*
850 	 * Global TCP locks are held because we manipulate the PCB lists
851 	 * and create a new socket.
852 	 */
853 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
854 	KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
855 	    ("%s: can handle only ACK", __func__));
856 
857 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
858 	SCH_LOCK_ASSERT(sch);
859 	if (sc == NULL) {
860 		/*
861 		 * There is no syncache entry, so see if this ACK is
862 		 * a returning syncookie.  To do this, first:
863 		 *  A. See if this socket has had a syncache entry dropped in
864 		 *     the past.  We don't want to accept a bogus syncookie
865 		 *     if we've never received a SYN.
866 		 *  B. check that the syncookie is valid.  If it is, then
867 		 *     cobble up a fake syncache entry, and return.
868 		 */
869 		if (!tcp_syncookies) {
870 			SCH_UNLOCK(sch);
871 			if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
872 				log(LOG_DEBUG, "%s; %s: Spurious ACK, "
873 				    "segment rejected (syncookies disabled)\n",
874 				    s, __func__);
875 			goto failed;
876 		}
877 		bzero(&scs, sizeof(scs));
878 		sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
879 		SCH_UNLOCK(sch);
880 		if (sc == NULL) {
881 			if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
882 				log(LOG_DEBUG, "%s; %s: Segment failed "
883 				    "SYNCOOKIE authentication, segment rejected "
884 				    "(probably spoofed)\n", s, __func__);
885 			goto failed;
886 		}
887 	} else {
888 		/* Pull out the entry to unlock the bucket row. */
889 		TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
890 		sch->sch_length--;
891 		tcp_syncache.cache_count--;
892 		SCH_UNLOCK(sch);
893 	}
894 
895 	/*
896 	 * Segment validation:
897 	 * ACK must match our initial sequence number + 1 (the SYN|ACK).
898 	 */
899 	if (th->th_ack != sc->sc_iss + 1 && !TOEPCB_ISSET(sc)) {
900 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
901 			log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment "
902 			    "rejected\n", s, __func__, th->th_ack, sc->sc_iss);
903 		goto failed;
904 	}
905 	/*
906 	 * The SEQ must match the received initial receive sequence
907 	 * number + 1 (the SYN) because we didn't ACK any data that
908 	 * may have come with the SYN.
909 	 */
910 	if (th->th_seq != sc->sc_irs + 1 && !TOEPCB_ISSET(sc)) {
911 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
912 			log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment "
913 			    "rejected\n", s, __func__, th->th_seq, sc->sc_irs);
914 		goto failed;
915 	}
916 
917 	if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
918 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
919 			log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
920 			    "segment rejected\n", s, __func__);
921 		goto failed;
922 	}
923 	/*
924 	 * If timestamps were negotiated the reflected timestamp
925 	 * must be equal to what we actually sent in the SYN|ACK.
926 	 */
927 	if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts &&
928 	    !TOEPCB_ISSET(sc)) {
929 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
930 			log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
931 			    "segment rejected\n",
932 			    s, __func__, to->to_tsecr, sc->sc_ts);
933 		goto failed;
934 	}
935 
936 	*lsop = syncache_socket(sc, *lsop, m);
937 
938 	if (*lsop == NULL)
939 		tcpstat.tcps_sc_aborted++;
940 	else
941 		tcpstat.tcps_sc_completed++;
942 
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 /*
956  * Given a LISTEN socket and an inbound SYN request, add
957  * this to the syn cache, and send back a segment:
958  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
959  * to the source.
960  *
961  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
962  * Doing so would require that we hold onto the data and deliver it
963  * to the application.  However, if we are the target of a SYN-flood
964  * DoS attack, an attacker could send data which would eventually
965  * consume all available buffer space if it were ACKed.  By not ACKing
966  * the data, we avoid this DoS scenario.
967  */
968 static void
969 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
970     struct inpcb *inp, struct socket **lsop, struct mbuf *m,
971     struct toe_usrreqs *tu, void *toepcb)
972 {
973 	struct tcpcb *tp;
974 	struct socket *so;
975 	struct syncache *sc = NULL;
976 	struct syncache_head *sch;
977 	struct mbuf *ipopts = NULL;
978 	u_int32_t flowtmp;
979 	int win, sb_hiwat, ip_ttl, ip_tos, noopt;
980 	char *s;
981 #ifdef INET6
982 	int autoflowlabel = 0;
983 #endif
984 #ifdef MAC
985 	struct label *maclabel;
986 #endif
987 	struct syncache scs;
988 
989 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
990 	INP_LOCK_ASSERT(inp);			/* listen socket */
991 	KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
992 	    ("%s: unexpected tcp flags", __func__));
993 
994 	/*
995 	 * Combine all so/tp operations very early to drop the INP lock as
996 	 * soon as possible.
997 	 */
998 	so = *lsop;
999 	tp = sototcpcb(so);
1000 
1001 #ifdef INET6
1002 	if (inc->inc_isipv6 &&
1003 	    (inp->in6p_flags & IN6P_AUTOFLOWLABEL))
1004 		autoflowlabel = 1;
1005 #endif
1006 	ip_ttl = inp->inp_ip_ttl;
1007 	ip_tos = inp->inp_ip_tos;
1008 	win = sbspace(&so->so_rcv);
1009 	sb_hiwat = so->so_rcv.sb_hiwat;
1010 	noopt = (tp->t_flags & TF_NOOPT);
1011 
1012 	so = NULL;
1013 	tp = NULL;
1014 
1015 #ifdef MAC
1016 	if (mac_syncache_init(&maclabel) != 0) {
1017 		INP_UNLOCK(inp);
1018 		INP_INFO_WUNLOCK(&tcbinfo);
1019 		goto done;
1020 	} else
1021 		mac_syncache_create(maclabel, inp);
1022 #endif
1023 	INP_UNLOCK(inp);
1024 	INP_INFO_WUNLOCK(&tcbinfo);
1025 
1026 	/*
1027 	 * Remember the IP options, if any.
1028 	 */
1029 #ifdef INET6
1030 	if (!inc->inc_isipv6)
1031 #endif
1032 		ipopts = (m) ? ip_srcroute(m) : NULL;
1033 
1034 	/*
1035 	 * See if we already have an entry for this connection.
1036 	 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1037 	 *
1038 	 * XXX: should the syncache be re-initialized with the contents
1039 	 * of the new SYN here (which may have different options?)
1040 	 *
1041 	 * XXX: We do not check the sequence number to see if this is a
1042 	 * real retransmit or a new connection attempt.  The question is
1043 	 * how to handle such a case; either ignore it as spoofed, or
1044 	 * drop the current entry and create a new one?
1045 	 */
1046 	sc = syncache_lookup(inc, &sch);	/* returns locked entry */
1047 	SCH_LOCK_ASSERT(sch);
1048 	if (sc != NULL) {
1049 #ifndef TCP_OFFLOAD_DISABLE
1050 		if (sc->sc_tu)
1051 			sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT,
1052 			    sc->sc_toepcb);
1053 #endif
1054 		tcpstat.tcps_sc_dupsyn++;
1055 		if (ipopts) {
1056 			/*
1057 			 * If we were remembering a previous source route,
1058 			 * forget it and use the new one we've been given.
1059 			 */
1060 			if (sc->sc_ipopts)
1061 				(void) m_free(sc->sc_ipopts);
1062 			sc->sc_ipopts = ipopts;
1063 		}
1064 		/*
1065 		 * Update timestamp if present.
1066 		 */
1067 		if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
1068 			sc->sc_tsreflect = to->to_tsval;
1069 		else
1070 			sc->sc_flags &= ~SCF_TIMESTAMP;
1071 #ifdef MAC
1072 		/*
1073 		 * Since we have already unconditionally allocated label
1074 		 * storage, free it up.  The syncache entry will already
1075 		 * have an initialized label we can use.
1076 		 */
1077 		mac_syncache_destroy(&maclabel);
1078 		KASSERT(sc->sc_label != NULL,
1079 		    ("%s: label not initialized", __func__));
1080 #endif
1081 		/* Retransmit SYN|ACK and reset retransmit count. */
1082 		if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
1083 			log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
1084 			    "resetting timer and retransmitting SYN|ACK\n",
1085 			    s, __func__);
1086 			free(s, M_TCPLOG);
1087 		}
1088 		if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) {
1089 			sc->sc_rxmits = 0;
1090 			syncache_timeout(sc, sch, 1);
1091 			tcpstat.tcps_sndacks++;
1092 			tcpstat.tcps_sndtotal++;
1093 		}
1094 		SCH_UNLOCK(sch);
1095 		goto done;
1096 	}
1097 
1098 	sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
1099 	if (sc == NULL) {
1100 		/*
1101 		 * The zone allocator couldn't provide more entries.
1102 		 * Treat this as if the cache was full; drop the oldest
1103 		 * entry and insert the new one.
1104 		 */
1105 		tcpstat.tcps_sc_zonefail++;
1106 		if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1107 			syncache_drop(sc, sch);
1108 		sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
1109 		if (sc == NULL) {
1110 			if (tcp_syncookies) {
1111 				bzero(&scs, sizeof(scs));
1112 				sc = &scs;
1113 			} else {
1114 				SCH_UNLOCK(sch);
1115 				if (ipopts)
1116 					(void) m_free(ipopts);
1117 				goto done;
1118 			}
1119 		}
1120 	}
1121 
1122 	/*
1123 	 * Fill in the syncache values.
1124 	 */
1125 #ifdef MAC
1126 	sc->sc_label = maclabel;
1127 #endif
1128 	sc->sc_ipopts = ipopts;
1129 	bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1130 #ifdef INET6
1131 	if (!inc->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 (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 
1214 	if (tcp_syncookies) {
1215 		syncookie_generate(sch, sc, &flowtmp);
1216 #ifdef INET6
1217 		if (autoflowlabel)
1218 			sc->sc_flowlabel = flowtmp;
1219 #endif
1220 	} else {
1221 #ifdef INET6
1222 		if (autoflowlabel)
1223 			sc->sc_flowlabel =
1224 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1225 #endif
1226 	}
1227 	SCH_UNLOCK(sch);
1228 
1229 	/*
1230 	 * Do a standard 3-way handshake.
1231 	 */
1232 	if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) {
1233 		if (tcp_syncookies && tcp_syncookiesonly && sc != &scs)
1234 			syncache_free(sc);
1235 		else if (sc != &scs)
1236 			syncache_insert(sc, sch);   /* locks and unlocks sch */
1237 		tcpstat.tcps_sndacks++;
1238 		tcpstat.tcps_sndtotal++;
1239 	} else {
1240 		if (sc != &scs)
1241 			syncache_free(sc);
1242 		tcpstat.tcps_sc_dropped++;
1243 	}
1244 
1245 done:
1246 #ifdef MAC
1247 	if (sc == &scs)
1248 		mac_syncache_destroy(&maclabel);
1249 #endif
1250 	if (m) {
1251 
1252 		*lsop = NULL;
1253 		m_freem(m);
1254 	}
1255 	return;
1256 }
1257 
1258 static int
1259 syncache_respond(struct syncache *sc)
1260 {
1261 	struct ip *ip = NULL;
1262 	struct mbuf *m;
1263 	struct tcphdr *th;
1264 	int optlen, error;
1265 	u_int16_t hlen, tlen, mssopt;
1266 	struct tcpopt to;
1267 #ifdef INET6
1268 	struct ip6_hdr *ip6 = NULL;
1269 #endif
1270 
1271 	hlen =
1272 #ifdef INET6
1273 	       (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) :
1274 #endif
1275 		sizeof(struct ip);
1276 	tlen = hlen + sizeof(struct tcphdr);
1277 
1278 	/* Determine MSS we advertize to other end of connection. */
1279 	mssopt = tcp_mssopt(&sc->sc_inc);
1280 	if (sc->sc_peer_mss)
1281 		mssopt = max( min(sc->sc_peer_mss, mssopt), tcp_minmss);
1282 
1283 	/* XXX: Assume that the entire packet will fit in a header mbuf. */
1284 	KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1285 	    ("syncache: mbuf too small"));
1286 
1287 	/* Create the IP+TCP header from scratch. */
1288 	m = m_gethdr(M_DONTWAIT, MT_DATA);
1289 	if (m == NULL)
1290 		return (ENOBUFS);
1291 #ifdef MAC
1292 	mac_syncache_create_mbuf(sc->sc_label, m);
1293 #endif
1294 	m->m_data += max_linkhdr;
1295 	m->m_len = tlen;
1296 	m->m_pkthdr.len = tlen;
1297 	m->m_pkthdr.rcvif = NULL;
1298 
1299 #ifdef INET6
1300 	if (sc->sc_inc.inc_isipv6) {
1301 		ip6 = mtod(m, struct ip6_hdr *);
1302 		ip6->ip6_vfc = IPV6_VERSION;
1303 		ip6->ip6_nxt = IPPROTO_TCP;
1304 		ip6->ip6_src = sc->sc_inc.inc6_laddr;
1305 		ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1306 		ip6->ip6_plen = htons(tlen - hlen);
1307 		/* ip6_hlim is set after checksum */
1308 		ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1309 		ip6->ip6_flow |= sc->sc_flowlabel;
1310 
1311 		th = (struct tcphdr *)(ip6 + 1);
1312 	} else
1313 #endif
1314 	{
1315 		ip = mtod(m, struct ip *);
1316 		ip->ip_v = IPVERSION;
1317 		ip->ip_hl = sizeof(struct ip) >> 2;
1318 		ip->ip_len = tlen;
1319 		ip->ip_id = 0;
1320 		ip->ip_off = 0;
1321 		ip->ip_sum = 0;
1322 		ip->ip_p = IPPROTO_TCP;
1323 		ip->ip_src = sc->sc_inc.inc_laddr;
1324 		ip->ip_dst = sc->sc_inc.inc_faddr;
1325 		ip->ip_ttl = sc->sc_ip_ttl;
1326 		ip->ip_tos = sc->sc_ip_tos;
1327 
1328 		/*
1329 		 * See if we should do MTU discovery.  Route lookups are
1330 		 * expensive, so we will only unset the DF bit if:
1331 		 *
1332 		 *	1) path_mtu_discovery is disabled
1333 		 *	2) the SCF_UNREACH flag has been set
1334 		 */
1335 		if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1336 		       ip->ip_off |= IP_DF;
1337 
1338 		th = (struct tcphdr *)(ip + 1);
1339 	}
1340 	th->th_sport = sc->sc_inc.inc_lport;
1341 	th->th_dport = sc->sc_inc.inc_fport;
1342 
1343 	th->th_seq = htonl(sc->sc_iss);
1344 	th->th_ack = htonl(sc->sc_irs + 1);
1345 	th->th_off = sizeof(struct tcphdr) >> 2;
1346 	th->th_x2 = 0;
1347 	th->th_flags = TH_SYN|TH_ACK;
1348 	th->th_win = htons(sc->sc_wnd);
1349 	th->th_urp = 0;
1350 
1351 	/* Tack on the TCP options. */
1352 	if ((sc->sc_flags & SCF_NOOPT) == 0) {
1353 		to.to_flags = 0;
1354 
1355 		to.to_mss = mssopt;
1356 		to.to_flags = TOF_MSS;
1357 		if (sc->sc_flags & SCF_WINSCALE) {
1358 			to.to_wscale = sc->sc_requested_r_scale;
1359 			to.to_flags |= TOF_SCALE;
1360 		}
1361 		if (sc->sc_flags & SCF_TIMESTAMP) {
1362 			/* Virgin timestamp or TCP cookie enhanced one. */
1363 			to.to_tsval = sc->sc_ts;
1364 			to.to_tsecr = sc->sc_tsreflect;
1365 			to.to_flags |= TOF_TS;
1366 		}
1367 		if (sc->sc_flags & SCF_SACK)
1368 			to.to_flags |= TOF_SACKPERM;
1369 #ifdef TCP_SIGNATURE
1370 		if (sc->sc_flags & SCF_SIGNATURE)
1371 			to.to_flags |= TOF_SIGNATURE;
1372 #endif
1373 		optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1374 
1375 		/* Adjust headers by option size. */
1376 		th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1377 		m->m_len += optlen;
1378 		m->m_pkthdr.len += optlen;
1379 
1380 #ifdef TCP_SIGNATURE
1381 		if (sc->sc_flags & SCF_SIGNATURE)
1382 			tcp_signature_compute(m, sizeof(struct ip), 0, optlen,
1383 			    to.to_signature, IPSEC_DIR_OUTBOUND);
1384 #endif
1385 #ifdef INET6
1386 		if (sc->sc_inc.inc_isipv6)
1387 			ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1388 		else
1389 #endif
1390 			ip->ip_len += optlen;
1391 	} else
1392 		optlen = 0;
1393 
1394 #ifdef INET6
1395 	if (sc->sc_inc.inc_isipv6) {
1396 		th->th_sum = 0;
1397 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
1398 				       tlen + optlen - hlen);
1399 		ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1400 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1401 	} else
1402 #endif
1403 	{
1404 		th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1405 		    htons(tlen + optlen - hlen + IPPROTO_TCP));
1406 		m->m_pkthdr.csum_flags = CSUM_TCP;
1407 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1408 		error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1409 	}
1410 	return (error);
1411 }
1412 
1413 void
1414 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1415     struct inpcb *inp, struct socket **lsop, struct mbuf *m)
1416 {
1417 	_syncache_add(inc, to, th, inp, lsop, m, NULL, NULL);
1418 }
1419 
1420 void
1421 syncache_offload_add(struct in_conninfo *inc, struct tcpopt *to,
1422     struct tcphdr *th, struct inpcb *inp, struct socket **lsop,
1423     struct toe_usrreqs *tu, void *toepcb)
1424 {
1425 	_syncache_add(inc, to, th, inp, lsop, NULL, tu, toepcb);
1426 }
1427 
1428 /*
1429  * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1430  * receive and to be able to handle SYN floods from bogus source addresses
1431  * (where we will never receive any reply).  SYN floods try to exhaust all
1432  * our memory and available slots in the SYN cache table to cause a denial
1433  * of service to legitimate users of the local host.
1434  *
1435  * The idea of SYN cookies is to encode and include all necessary information
1436  * about the connection setup state within the SYN-ACK we send back and thus
1437  * to get along without keeping any local state until the ACK to the SYN-ACK
1438  * arrives (if ever).  Everything we need to know should be available from
1439  * the information we encoded in the SYN-ACK.
1440  *
1441  * More information about the theory behind SYN cookies and its first
1442  * discussion and specification can be found at:
1443  *  http://cr.yp.to/syncookies.html    (overview)
1444  *  http://cr.yp.to/syncookies/archive (gory details)
1445  *
1446  * This implementation extends the orginal idea and first implementation
1447  * of FreeBSD by using not only the initial sequence number field to store
1448  * information but also the timestamp field if present.  This way we can
1449  * keep track of the entire state we need to know to recreate the session in
1450  * its original form.  Almost all TCP speakers implement RFC1323 timestamps
1451  * these days.  For those that do not we still have to live with the known
1452  * shortcomings of the ISN only SYN cookies.
1453  *
1454  * Cookie layers:
1455  *
1456  * Initial sequence number we send:
1457  * 31|................................|0
1458  *    DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1459  *    D = MD5 Digest (first dword)
1460  *    M = MSS index
1461  *    R = Rotation of secret
1462  *    P = Odd or Even secret
1463  *
1464  * The MD5 Digest is computed with over following parameters:
1465  *  a) randomly rotated secret
1466  *  b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1467  *  c) the received initial sequence number from remote host
1468  *  d) the rotation offset and odd/even bit
1469  *
1470  * Timestamp we send:
1471  * 31|................................|0
1472  *    DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1473  *    D = MD5 Digest (third dword) (only as filler)
1474  *    S = Requested send window scale
1475  *    R = Requested receive window scale
1476  *    A = SACK allowed
1477  *    5 = TCP-MD5 enabled (not implemented yet)
1478  *    XORed with MD5 Digest (forth dword)
1479  *
1480  * The timestamp isn't cryptographically secure and doesn't need to be.
1481  * The double use of the MD5 digest dwords ties it to a specific remote/
1482  * local host/port, remote initial sequence number and our local time
1483  * limited secret.  A received timestamp is reverted (XORed) and then
1484  * the contained MD5 dword is compared to the computed one to ensure the
1485  * timestamp belongs to the SYN-ACK we sent.  The other parameters may
1486  * have been tampered with but this isn't different from supplying bogus
1487  * values in the SYN in the first place.
1488  *
1489  * Some problems with SYN cookies remain however:
1490  * Consider the problem of a recreated (and retransmitted) cookie.  If the
1491  * original SYN was accepted, the connection is established.  The second
1492  * SYN is inflight, and if it arrives with an ISN that falls within the
1493  * receive window, the connection is killed.
1494  *
1495  * Notes:
1496  * A heuristic to determine when to accept syn cookies is not necessary.
1497  * An ACK flood would cause the syncookie verification to be attempted,
1498  * but a SYN flood causes syncookies to be generated.  Both are of equal
1499  * cost, so there's no point in trying to optimize the ACK flood case.
1500  * Also, if you don't process certain ACKs for some reason, then all someone
1501  * would have to do is launch a SYN and ACK flood at the same time, which
1502  * would stop cookie verification and defeat the entire purpose of syncookies.
1503  */
1504 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1505 
1506 static void
1507 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1508     u_int32_t *flowlabel)
1509 {
1510 	MD5_CTX ctx;
1511 	u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1512 	u_int32_t data;
1513 	u_int32_t *secbits;
1514 	u_int off, pmss, mss;
1515 	int i;
1516 
1517 	SCH_LOCK_ASSERT(sch);
1518 
1519 	/* Which of the two secrets to use. */
1520 	secbits = sch->sch_oddeven ?
1521 			sch->sch_secbits_odd : sch->sch_secbits_even;
1522 
1523 	/* Reseed secret if too old. */
1524 	if (sch->sch_reseed < time_uptime) {
1525 		sch->sch_oddeven = sch->sch_oddeven ? 0 : 1;	/* toggle */
1526 		secbits = sch->sch_oddeven ?
1527 				sch->sch_secbits_odd : sch->sch_secbits_even;
1528 		for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1529 			secbits[i] = arc4random();
1530 		sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1531 	}
1532 
1533 	/* Secret rotation offset. */
1534 	off = sc->sc_iss & 0x7;			/* iss was randomized before */
1535 
1536 	/* Maximum segment size calculation. */
1537 	pmss = max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), tcp_minmss);
1538 	for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1539 		if (tcp_sc_msstab[mss] <= pmss)
1540 			break;
1541 
1542 	/* Fold parameters and MD5 digest into the ISN we will send. */
1543 	data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1544 	data |= off << 1;	/* secret offset, derived from iss, 3 bits */
1545 	data |= mss << 4;	/* mss, 3 bits */
1546 
1547 	MD5Init(&ctx);
1548 	MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1549 	    SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1550 	MD5Update(&ctx, secbits, off);
1551 	MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1552 	MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1553 	MD5Update(&ctx, &data, sizeof(data));
1554 	MD5Final((u_int8_t *)&md5_buffer, &ctx);
1555 
1556 	data |= (md5_buffer[0] << 7);
1557 	sc->sc_iss = data;
1558 
1559 #ifdef INET6
1560 	*flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1561 #endif
1562 
1563 	/* Additional parameters are stored in the timestamp if present. */
1564 	if (sc->sc_flags & SCF_TIMESTAMP) {
1565 		data =  ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1566 		data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1567 		data |= sc->sc_requested_s_scale << 2;  /* SWIN scale, 4 bits */
1568 		data |= sc->sc_requested_r_scale << 6;  /* RWIN scale, 4 bits */
1569 		data |= md5_buffer[2] << 10;		/* more digest bits */
1570 		data ^= md5_buffer[3];
1571 		sc->sc_ts = data;
1572 		sc->sc_tsoff = data - ticks;		/* after XOR */
1573 	}
1574 
1575 	tcpstat.tcps_sc_sendcookie++;
1576 	return;
1577 }
1578 
1579 static struct syncache *
1580 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1581     struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1582     struct socket *so)
1583 {
1584 	MD5_CTX ctx;
1585 	u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1586 	u_int32_t data = 0;
1587 	u_int32_t *secbits;
1588 	tcp_seq ack, seq;
1589 	int off, mss, wnd, flags;
1590 
1591 	SCH_LOCK_ASSERT(sch);
1592 
1593 	/*
1594 	 * Pull information out of SYN-ACK/ACK and
1595 	 * revert sequence number advances.
1596 	 */
1597 	ack = th->th_ack - 1;
1598 	seq = th->th_seq - 1;
1599 	off = (ack >> 1) & 0x7;
1600 	mss = (ack >> 4) & 0x7;
1601 	flags = ack & 0x7f;
1602 
1603 	/* Which of the two secrets to use. */
1604 	secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1605 
1606 	/*
1607 	 * The secret wasn't updated for the lifetime of a syncookie,
1608 	 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1609 	 */
1610 	if (sch->sch_reseed < time_uptime) {
1611 		return (NULL);
1612 	}
1613 
1614 	/* Recompute the digest so we can compare it. */
1615 	MD5Init(&ctx);
1616 	MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1617 	    SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1618 	MD5Update(&ctx, secbits, off);
1619 	MD5Update(&ctx, inc, sizeof(*inc));
1620 	MD5Update(&ctx, &seq, sizeof(seq));
1621 	MD5Update(&ctx, &flags, sizeof(flags));
1622 	MD5Final((u_int8_t *)&md5_buffer, &ctx);
1623 
1624 	/* Does the digest part of or ACK'ed ISS match? */
1625 	if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1626 		return (NULL);
1627 
1628 	/* Does the digest part of our reflected timestamp match? */
1629 	if (to->to_flags & TOF_TS) {
1630 		data = md5_buffer[3] ^ to->to_tsecr;
1631 		if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1632 			return (NULL);
1633 	}
1634 
1635 	/* Fill in the syncache values. */
1636 	bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1637 	sc->sc_ipopts = NULL;
1638 
1639 	sc->sc_irs = seq;
1640 	sc->sc_iss = ack;
1641 
1642 #ifdef INET6
1643 	if (inc->inc_isipv6) {
1644 		if (sotoinpcb(so)->in6p_flags & IN6P_AUTOFLOWLABEL)
1645 			sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1646 	} else
1647 #endif
1648 	{
1649 		sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1650 		sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1651 	}
1652 
1653 	/* Additional parameters that were encoded in the timestamp. */
1654 	if (data) {
1655 		sc->sc_flags |= SCF_TIMESTAMP;
1656 		sc->sc_tsreflect = to->to_tsval;
1657 		sc->sc_ts = to->to_tsecr;
1658 		sc->sc_tsoff = to->to_tsecr - ticks;
1659 		sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1660 		sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1661 		sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1662 		    TCP_MAX_WINSHIFT);
1663 		sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1664 		    TCP_MAX_WINSHIFT);
1665 		if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1666 			sc->sc_flags |= SCF_WINSCALE;
1667 	} else
1668 		sc->sc_flags |= SCF_NOOPT;
1669 
1670 	wnd = sbspace(&so->so_rcv);
1671 	wnd = imax(wnd, 0);
1672 	wnd = imin(wnd, TCP_MAXWIN);
1673 	sc->sc_wnd = wnd;
1674 
1675 	sc->sc_rxmits = 0;
1676 	sc->sc_peer_mss = tcp_sc_msstab[mss];
1677 
1678 	tcpstat.tcps_sc_recvcookie++;
1679 	return (sc);
1680 }
1681 
1682 /*
1683  * Returns the current number of syncache entries.  This number
1684  * will probably change before you get around to calling
1685  * syncache_pcblist.
1686  */
1687 
1688 int
1689 syncache_pcbcount(void)
1690 {
1691 	struct syncache_head *sch;
1692 	int count, i;
1693 
1694 	for (count = 0, i = 0; i < tcp_syncache.hashsize; i++) {
1695 		/* No need to lock for a read. */
1696 		sch = &tcp_syncache.hashbase[i];
1697 		count += sch->sch_length;
1698 	}
1699 	return count;
1700 }
1701 
1702 /*
1703  * Exports the syncache entries to userland so that netstat can display
1704  * them alongside the other sockets.  This function is intended to be
1705  * called only from tcp_pcblist.
1706  *
1707  * Due to concurrency on an active system, the number of pcbs exported
1708  * may have no relation to max_pcbs.  max_pcbs merely indicates the
1709  * amount of space the caller allocated for this function to use.
1710  */
1711 int
1712 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
1713 {
1714 	struct xtcpcb xt;
1715 	struct syncache *sc;
1716 	struct syncache_head *sch;
1717 	int count, error, i;
1718 
1719 	for (count = 0, error = 0, i = 0; i < tcp_syncache.hashsize; i++) {
1720 		sch = &tcp_syncache.hashbase[i];
1721 		SCH_LOCK(sch);
1722 		TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
1723 			if (count >= max_pcbs) {
1724 				SCH_UNLOCK(sch);
1725 				goto exit;
1726 			}
1727 			bzero(&xt, sizeof(xt));
1728 			xt.xt_len = sizeof(xt);
1729 			if (sc->sc_inc.inc_isipv6)
1730 				xt.xt_inp.inp_vflag = INP_IPV6;
1731 			else
1732 				xt.xt_inp.inp_vflag = INP_IPV4;
1733 			bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
1734 			xt.xt_tp.t_inpcb = &xt.xt_inp;
1735 			xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
1736 			xt.xt_socket.xso_protocol = IPPROTO_TCP;
1737 			xt.xt_socket.xso_len = sizeof (struct xsocket);
1738 			xt.xt_socket.so_type = SOCK_STREAM;
1739 			xt.xt_socket.so_state = SS_ISCONNECTING;
1740 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1741 			if (error) {
1742 				SCH_UNLOCK(sch);
1743 				goto exit;
1744 			}
1745 			count++;
1746 		}
1747 		SCH_UNLOCK(sch);
1748 	}
1749 exit:
1750 	*pcbs_exported = count;
1751 	return error;
1752 }
1753