xref: /freebsd/sys/netinet/tcp_syncache.c (revision 6af83ee0d2941d18880b6aaa2b4facd1d30c6106)
1 /*-
2  * Copyright (c) 2001 McAfee, Inc.
3  * All rights reserved.
4  *
5  * This software was developed for the FreeBSD Project by Jonathan Lemon
6  * and McAfee Research, the Security Research Division of McAfee, Inc. under
7  * DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
8  * DARPA CHATS research program.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  * $FreeBSD$
32  */
33 
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_ipsec.h"
37 #include "opt_mac.h"
38 #include "opt_tcpdebug.h"
39 #include "opt_tcp_sack.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/malloc.h>
46 #include <sys/mac.h>
47 #include <sys/mbuf.h>
48 #include <sys/md5.h>
49 #include <sys/proc.h>		/* for proc0 declaration */
50 #include <sys/random.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 
54 #include <net/if.h>
55 #include <net/route.h>
56 
57 #include <netinet/in.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/ip.h>
60 #include <netinet/in_var.h>
61 #include <netinet/in_pcb.h>
62 #include <netinet/ip_var.h>
63 #ifdef INET6
64 #include <netinet/ip6.h>
65 #include <netinet/icmp6.h>
66 #include <netinet6/nd6.h>
67 #include <netinet6/ip6_var.h>
68 #include <netinet6/in6_pcb.h>
69 #endif
70 #include <netinet/tcp.h>
71 #ifdef TCPDEBUG
72 #include <netinet/tcpip.h>
73 #endif
74 #include <netinet/tcp_fsm.h>
75 #include <netinet/tcp_seq.h>
76 #include <netinet/tcp_timer.h>
77 #include <netinet/tcp_var.h>
78 #ifdef TCPDEBUG
79 #include <netinet/tcp_debug.h>
80 #endif
81 #ifdef INET6
82 #include <netinet6/tcp6_var.h>
83 #endif
84 
85 #ifdef IPSEC
86 #include <netinet6/ipsec.h>
87 #ifdef INET6
88 #include <netinet6/ipsec6.h>
89 #endif
90 #endif /*IPSEC*/
91 
92 #ifdef FAST_IPSEC
93 #include <netipsec/ipsec.h>
94 #ifdef INET6
95 #include <netipsec/ipsec6.h>
96 #endif
97 #include <netipsec/key.h>
98 #endif /*FAST_IPSEC*/
99 
100 #include <machine/in_cksum.h>
101 #include <vm/uma.h>
102 
103 static int tcp_syncookies = 1;
104 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
105     &tcp_syncookies, 0,
106     "Use TCP SYN cookies if the syncache overflows");
107 
108 static void	 syncache_drop(struct syncache *, struct syncache_head *);
109 static void	 syncache_free(struct syncache *);
110 static void	 syncache_insert(struct syncache *, struct syncache_head *);
111 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
112 #ifdef TCPDEBUG
113 static int	 syncache_respond(struct syncache *, struct mbuf *, struct socket *);
114 #else
115 static int	 syncache_respond(struct syncache *, struct mbuf *);
116 #endif
117 static struct	 socket *syncache_socket(struct syncache *, struct socket *,
118 		    struct mbuf *m);
119 static void	 syncache_timer(void *);
120 static u_int32_t syncookie_generate(struct syncache *, u_int32_t *);
121 static struct syncache *syncookie_lookup(struct in_conninfo *,
122 		    struct tcphdr *, struct socket *);
123 
124 /*
125  * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
126  * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds,
127  * the odds are that the user has given up attempting to connect by then.
128  */
129 #define SYNCACHE_MAXREXMTS		3
130 
131 /* Arbitrary values */
132 #define TCP_SYNCACHE_HASHSIZE		512
133 #define TCP_SYNCACHE_BUCKETLIMIT	30
134 
135 struct tcp_syncache {
136 	struct	syncache_head *hashbase;
137 	uma_zone_t zone;
138 	u_int	hashsize;
139 	u_int	hashmask;
140 	u_int	bucket_limit;
141 	u_int	cache_count;
142 	u_int	cache_limit;
143 	u_int	rexmt_limit;
144 	u_int	hash_secret;
145 	TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1];
146 	struct	callout tt_timerq[SYNCACHE_MAXREXMTS + 1];
147 };
148 static struct tcp_syncache tcp_syncache;
149 
150 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
151 
152 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
153      &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
154 
155 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
156      &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
157 
158 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
159      &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
160 
161 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
162      &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
163 
164 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
165      &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
166 
167 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
168 
169 #define SYNCACHE_HASH(inc, mask)					\
170 	((tcp_syncache.hash_secret ^					\
171 	  (inc)->inc_faddr.s_addr ^					\
172 	  ((inc)->inc_faddr.s_addr >> 16) ^				\
173 	  (inc)->inc_fport ^ (inc)->inc_lport) & mask)
174 
175 #define SYNCACHE_HASH6(inc, mask)					\
176 	((tcp_syncache.hash_secret ^					\
177 	  (inc)->inc6_faddr.s6_addr32[0] ^				\
178 	  (inc)->inc6_faddr.s6_addr32[3] ^				\
179 	  (inc)->inc_fport ^ (inc)->inc_lport) & mask)
180 
181 #define ENDPTS_EQ(a, b) (						\
182 	(a)->ie_fport == (b)->ie_fport &&				\
183 	(a)->ie_lport == (b)->ie_lport &&				\
184 	(a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr &&			\
185 	(a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr			\
186 )
187 
188 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
189 
190 #define SYNCACHE_TIMEOUT(sc, slot) do {				\
191 	sc->sc_rxtslot = (slot);					\
192 	sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[(slot)];	\
193 	TAILQ_INSERT_TAIL(&tcp_syncache.timerq[(slot)], sc, sc_timerq);	\
194 	if (!callout_active(&tcp_syncache.tt_timerq[(slot)]))		\
195 		callout_reset(&tcp_syncache.tt_timerq[(slot)],		\
196 		    TCPTV_RTOBASE * tcp_backoff[(slot)],		\
197 		    syncache_timer, (void *)((intptr_t)(slot)));	\
198 } while (0)
199 
200 static void
201 syncache_free(struct syncache *sc)
202 {
203 	if (sc->sc_ipopts)
204 		(void) m_free(sc->sc_ipopts);
205 
206 	uma_zfree(tcp_syncache.zone, sc);
207 }
208 
209 void
210 syncache_init(void)
211 {
212 	int i;
213 
214 	tcp_syncache.cache_count = 0;
215 	tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
216 	tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
217 	tcp_syncache.cache_limit =
218 	    tcp_syncache.hashsize * tcp_syncache.bucket_limit;
219 	tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
220 	tcp_syncache.hash_secret = arc4random();
221 
222 	TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
223 	    &tcp_syncache.hashsize);
224 	TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
225 	    &tcp_syncache.cache_limit);
226 	TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
227 	    &tcp_syncache.bucket_limit);
228 	if (!powerof2(tcp_syncache.hashsize)) {
229 		printf("WARNING: syncache hash size is not a power of 2.\n");
230 		tcp_syncache.hashsize = 512;	/* safe default */
231 	}
232 	tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
233 
234 	/* Allocate the hash table. */
235 	MALLOC(tcp_syncache.hashbase, struct syncache_head *,
236 	    tcp_syncache.hashsize * sizeof(struct syncache_head),
237 	    M_SYNCACHE, M_WAITOK);
238 
239 	/* Initialize the hash buckets. */
240 	for (i = 0; i < tcp_syncache.hashsize; i++) {
241 		TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket);
242 		tcp_syncache.hashbase[i].sch_length = 0;
243 	}
244 
245 	/* Initialize the timer queues. */
246 	for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) {
247 		TAILQ_INIT(&tcp_syncache.timerq[i]);
248 		callout_init(&tcp_syncache.tt_timerq[i],
249 			debug_mpsafenet ? CALLOUT_MPSAFE : 0);
250 	}
251 
252 	/*
253 	 * Allocate the syncache entries.  Allow the zone to allocate one
254 	 * more entry than cache limit, so a new entry can bump out an
255 	 * older one.
256 	 */
257 	tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
258 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
259 	uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit);
260 	tcp_syncache.cache_limit -= 1;
261 }
262 
263 static void
264 syncache_insert(sc, sch)
265 	struct syncache *sc;
266 	struct syncache_head *sch;
267 {
268 	struct syncache *sc2;
269 	int i;
270 
271 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
272 
273 	/*
274 	 * Make sure that we don't overflow the per-bucket
275 	 * limit or the total cache size limit.
276 	 */
277 	if (sch->sch_length >= tcp_syncache.bucket_limit) {
278 		/*
279 		 * The bucket is full, toss the oldest element.
280 		 */
281 		sc2 = TAILQ_FIRST(&sch->sch_bucket);
282 		sc2->sc_tp->ts_recent = ticks;
283 		syncache_drop(sc2, sch);
284 		tcpstat.tcps_sc_bucketoverflow++;
285 	} else if (tcp_syncache.cache_count >= tcp_syncache.cache_limit) {
286 		/*
287 		 * The cache is full.  Toss the oldest entry in the
288 		 * entire cache.  This is the front entry in the
289 		 * first non-empty timer queue with the largest
290 		 * timeout value.
291 		 */
292 		for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
293 			sc2 = TAILQ_FIRST(&tcp_syncache.timerq[i]);
294 			if (sc2 != NULL)
295 				break;
296 		}
297 		sc2->sc_tp->ts_recent = ticks;
298 		syncache_drop(sc2, NULL);
299 		tcpstat.tcps_sc_cacheoverflow++;
300 	}
301 
302 	/* Initialize the entry's timer. */
303 	SYNCACHE_TIMEOUT(sc, 0);
304 
305 	/* Put it into the bucket. */
306 	TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash);
307 	sch->sch_length++;
308 	tcp_syncache.cache_count++;
309 	tcpstat.tcps_sc_added++;
310 }
311 
312 static void
313 syncache_drop(sc, sch)
314 	struct syncache *sc;
315 	struct syncache_head *sch;
316 {
317 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
318 
319 	if (sch == NULL) {
320 #ifdef INET6
321 		if (sc->sc_inc.inc_isipv6) {
322 			sch = &tcp_syncache.hashbase[
323 			    SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)];
324 		} else
325 #endif
326 		{
327 			sch = &tcp_syncache.hashbase[
328 			    SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)];
329 		}
330 	}
331 
332 	TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
333 	sch->sch_length--;
334 	tcp_syncache.cache_count--;
335 
336 	TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], sc, sc_timerq);
337 	if (TAILQ_EMPTY(&tcp_syncache.timerq[sc->sc_rxtslot]))
338 		callout_stop(&tcp_syncache.tt_timerq[sc->sc_rxtslot]);
339 
340 	syncache_free(sc);
341 }
342 
343 /*
344  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
345  * If we have retransmitted an entry the maximum number of times, expire it.
346  */
347 static void
348 syncache_timer(xslot)
349 	void *xslot;
350 {
351 	intptr_t slot = (intptr_t)xslot;
352 	struct syncache *sc, *nsc;
353 	struct inpcb *inp;
354 
355 	INP_INFO_WLOCK(&tcbinfo);
356 	if (callout_pending(&tcp_syncache.tt_timerq[slot]) ||
357 	    !callout_active(&tcp_syncache.tt_timerq[slot])) {
358 		/* XXX can this happen? */
359 		INP_INFO_WUNLOCK(&tcbinfo);
360 		return;
361 	}
362 	callout_deactivate(&tcp_syncache.tt_timerq[slot]);
363 
364 	nsc = TAILQ_FIRST(&tcp_syncache.timerq[slot]);
365 	while (nsc != NULL) {
366 		if (ticks < nsc->sc_rxttime)
367 			break;
368 		sc = nsc;
369 		inp = sc->sc_tp->t_inpcb;
370 		if (slot == SYNCACHE_MAXREXMTS ||
371 		    slot >= tcp_syncache.rexmt_limit ||
372 		    inp == NULL || inp->inp_gencnt != sc->sc_inp_gencnt) {
373 			nsc = TAILQ_NEXT(sc, sc_timerq);
374 			syncache_drop(sc, NULL);
375 			tcpstat.tcps_sc_stale++;
376 			continue;
377 		}
378 		/*
379 		 * syncache_respond() may call back into the syncache to
380 		 * to modify another entry, so do not obtain the next
381 		 * entry on the timer chain until it has completed.
382 		 */
383 #ifdef TCPDEBUG
384 		(void) syncache_respond(sc, NULL, NULL);
385 #else
386 		(void) syncache_respond(sc, NULL);
387 #endif
388 		nsc = TAILQ_NEXT(sc, sc_timerq);
389 		tcpstat.tcps_sc_retransmitted++;
390 		TAILQ_REMOVE(&tcp_syncache.timerq[slot], sc, sc_timerq);
391 		SYNCACHE_TIMEOUT(sc, slot + 1);
392 	}
393 	if (nsc != NULL)
394 		callout_reset(&tcp_syncache.tt_timerq[slot],
395 		    nsc->sc_rxttime - ticks, syncache_timer, (void *)(slot));
396 	INP_INFO_WUNLOCK(&tcbinfo);
397 }
398 
399 /*
400  * Find an entry in the syncache.
401  */
402 struct syncache *
403 syncache_lookup(inc, schp)
404 	struct in_conninfo *inc;
405 	struct syncache_head **schp;
406 {
407 	struct syncache *sc;
408 	struct syncache_head *sch;
409 
410 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
411 
412 #ifdef INET6
413 	if (inc->inc_isipv6) {
414 		sch = &tcp_syncache.hashbase[
415 		    SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
416 		*schp = sch;
417 		TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
418 			if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
419 				return (sc);
420 		}
421 	} else
422 #endif
423 	{
424 		sch = &tcp_syncache.hashbase[
425 		    SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
426 		*schp = sch;
427 		TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
428 #ifdef INET6
429 			if (sc->sc_inc.inc_isipv6)
430 				continue;
431 #endif
432 			if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
433 				return (sc);
434 		}
435 	}
436 	return (NULL);
437 }
438 
439 /*
440  * This function is called when we get a RST for a
441  * non-existent connection, so that we can see if the
442  * connection is in the syn cache.  If it is, zap it.
443  */
444 void
445 syncache_chkrst(inc, th)
446 	struct in_conninfo *inc;
447 	struct tcphdr *th;
448 {
449 	struct syncache *sc;
450 	struct syncache_head *sch;
451 
452 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
453 
454 	sc = syncache_lookup(inc, &sch);
455 	if (sc == NULL)
456 		return;
457 	/*
458 	 * If the RST bit is set, check the sequence number to see
459 	 * if this is a valid reset segment.
460 	 * RFC 793 page 37:
461 	 *   In all states except SYN-SENT, all reset (RST) segments
462 	 *   are validated by checking their SEQ-fields.  A reset is
463 	 *   valid if its sequence number is in the window.
464 	 *
465 	 *   The sequence number in the reset segment is normally an
466 	 *   echo of our outgoing acknowlegement numbers, but some hosts
467 	 *   send a reset with the sequence number at the rightmost edge
468 	 *   of our receive window, and we have to handle this case.
469 	 */
470 	if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
471 	    SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
472 		syncache_drop(sc, sch);
473 		tcpstat.tcps_sc_reset++;
474 	}
475 }
476 
477 void
478 syncache_badack(inc)
479 	struct in_conninfo *inc;
480 {
481 	struct syncache *sc;
482 	struct syncache_head *sch;
483 
484 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
485 
486 	sc = syncache_lookup(inc, &sch);
487 	if (sc != NULL) {
488 		syncache_drop(sc, sch);
489 		tcpstat.tcps_sc_badack++;
490 	}
491 }
492 
493 void
494 syncache_unreach(inc, th)
495 	struct in_conninfo *inc;
496 	struct tcphdr *th;
497 {
498 	struct syncache *sc;
499 	struct syncache_head *sch;
500 
501 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
502 
503 	/* we are called at splnet() here */
504 	sc = syncache_lookup(inc, &sch);
505 	if (sc == NULL)
506 		return;
507 
508 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
509 	if (ntohl(th->th_seq) != sc->sc_iss)
510 		return;
511 
512 	/*
513 	 * If we've rertransmitted 3 times and this is our second error,
514 	 * we remove the entry.  Otherwise, we allow it to continue on.
515 	 * This prevents us from incorrectly nuking an entry during a
516 	 * spurious network outage.
517 	 *
518 	 * See tcp_notify().
519 	 */
520 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) {
521 		sc->sc_flags |= SCF_UNREACH;
522 		return;
523 	}
524 	syncache_drop(sc, sch);
525 	tcpstat.tcps_sc_unreach++;
526 }
527 
528 /*
529  * Build a new TCP socket structure from a syncache entry.
530  */
531 static struct socket *
532 syncache_socket(sc, lso, m)
533 	struct syncache *sc;
534 	struct socket *lso;
535 	struct mbuf *m;
536 {
537 	struct inpcb *inp = NULL;
538 	struct socket *so;
539 	struct tcpcb *tp;
540 
541 	NET_ASSERT_GIANT();
542 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
543 
544 	/*
545 	 * Ok, create the full blown connection, and set things up
546 	 * as they would have been set up if we had created the
547 	 * connection when the SYN arrived.  If we can't create
548 	 * the connection, abort it.
549 	 */
550 	so = sonewconn(lso, SS_ISCONNECTED);
551 	if (so == NULL) {
552 		/*
553 		 * Drop the connection; we will send a RST if the peer
554 		 * retransmits the ACK,
555 		 */
556 		tcpstat.tcps_listendrop++;
557 		goto abort2;
558 	}
559 #ifdef MAC
560 	SOCK_LOCK(so);
561 	mac_set_socket_peer_from_mbuf(m, so);
562 	SOCK_UNLOCK(so);
563 #endif
564 
565 	inp = sotoinpcb(so);
566 	INP_LOCK(inp);
567 
568 	/*
569 	 * Insert new socket into hash list.
570 	 */
571 	inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
572 #ifdef INET6
573 	if (sc->sc_inc.inc_isipv6) {
574 		inp->in6p_laddr = sc->sc_inc.inc6_laddr;
575 	} else {
576 		inp->inp_vflag &= ~INP_IPV6;
577 		inp->inp_vflag |= INP_IPV4;
578 #endif
579 		inp->inp_laddr = sc->sc_inc.inc_laddr;
580 #ifdef INET6
581 	}
582 #endif
583 	inp->inp_lport = sc->sc_inc.inc_lport;
584 	if (in_pcbinshash(inp) != 0) {
585 		/*
586 		 * Undo the assignments above if we failed to
587 		 * put the PCB on the hash lists.
588 		 */
589 #ifdef INET6
590 		if (sc->sc_inc.inc_isipv6)
591 			inp->in6p_laddr = in6addr_any;
592 		else
593 #endif
594 			inp->inp_laddr.s_addr = INADDR_ANY;
595 		inp->inp_lport = 0;
596 		goto abort;
597 	}
598 #ifdef IPSEC
599 	/* copy old policy into new socket's */
600 	if (ipsec_copy_pcbpolicy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
601 		printf("syncache_expand: could not copy policy\n");
602 #endif
603 #ifdef FAST_IPSEC
604 	/* copy old policy into new socket's */
605 	if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
606 		printf("syncache_expand: could not copy policy\n");
607 #endif
608 #ifdef INET6
609 	if (sc->sc_inc.inc_isipv6) {
610 		struct inpcb *oinp = sotoinpcb(lso);
611 		struct in6_addr laddr6;
612 		struct sockaddr_in6 sin6;
613 		/*
614 		 * Inherit socket options from the listening socket.
615 		 * Note that in6p_inputopts are not (and should not be)
616 		 * copied, since it stores previously received options and is
617 		 * used to detect if each new option is different than the
618 		 * previous one and hence should be passed to a user.
619 		 * If we copied in6p_inputopts, a user would not be able to
620 		 * receive options just after calling the accept system call.
621 		 */
622 		inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
623 		if (oinp->in6p_outputopts)
624 			inp->in6p_outputopts =
625 			    ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
626 
627 		sin6.sin6_family = AF_INET6;
628 		sin6.sin6_len = sizeof(sin6);
629 		sin6.sin6_addr = sc->sc_inc.inc6_faddr;
630 		sin6.sin6_port = sc->sc_inc.inc_fport;
631 		sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
632 		laddr6 = inp->in6p_laddr;
633 		if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
634 			inp->in6p_laddr = sc->sc_inc.inc6_laddr;
635 		if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
636 		    thread0.td_ucred)) {
637 			inp->in6p_laddr = laddr6;
638 			goto abort;
639 		}
640 		/* Override flowlabel from in6_pcbconnect. */
641 		inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK;
642 		inp->in6p_flowinfo |= sc->sc_flowlabel;
643 	} else
644 #endif
645 	{
646 		struct in_addr laddr;
647 		struct sockaddr_in sin;
648 
649 		inp->inp_options = ip_srcroute(m);
650 		if (inp->inp_options == NULL) {
651 			inp->inp_options = sc->sc_ipopts;
652 			sc->sc_ipopts = NULL;
653 		}
654 
655 		sin.sin_family = AF_INET;
656 		sin.sin_len = sizeof(sin);
657 		sin.sin_addr = sc->sc_inc.inc_faddr;
658 		sin.sin_port = sc->sc_inc.inc_fport;
659 		bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
660 		laddr = inp->inp_laddr;
661 		if (inp->inp_laddr.s_addr == INADDR_ANY)
662 			inp->inp_laddr = sc->sc_inc.inc_laddr;
663 		if (in_pcbconnect(inp, (struct sockaddr *)&sin,
664 		    thread0.td_ucred)) {
665 			inp->inp_laddr = laddr;
666 			goto abort;
667 		}
668 	}
669 
670 	tp = intotcpcb(inp);
671 	tp->t_state = TCPS_SYN_RECEIVED;
672 	tp->iss = sc->sc_iss;
673 	tp->irs = sc->sc_irs;
674 	tcp_rcvseqinit(tp);
675 	tcp_sendseqinit(tp);
676 	tp->snd_wl1 = sc->sc_irs;
677 	tp->rcv_up = sc->sc_irs + 1;
678 	tp->rcv_wnd = sc->sc_wnd;
679 	tp->rcv_adv += tp->rcv_wnd;
680 
681 	tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
682 	if (sc->sc_flags & SCF_NOOPT)
683 		tp->t_flags |= TF_NOOPT;
684 	if (sc->sc_flags & SCF_WINSCALE) {
685 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
686 		tp->requested_s_scale = sc->sc_requested_s_scale;
687 		tp->request_r_scale = sc->sc_request_r_scale;
688 	}
689 	if (sc->sc_flags & SCF_TIMESTAMP) {
690 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
691 		tp->ts_recent = sc->sc_tsrecent;
692 		tp->ts_recent_age = ticks;
693 	}
694 #ifdef TCP_SIGNATURE
695 	if (sc->sc_flags & SCF_SIGNATURE)
696 		tp->t_flags |= TF_SIGNATURE;
697 #endif
698 	if (sc->sc_flags & SCF_SACK) {
699 		tp->sack_enable = 1;
700 		tp->t_flags |= TF_SACK_PERMIT;
701 	}
702 	/*
703 	 * Set up MSS and get cached values from tcp_hostcache.
704 	 * This might overwrite some of the defaults we just set.
705 	 */
706 	tcp_mss(tp, sc->sc_peer_mss);
707 
708 	/*
709 	 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
710 	 */
711 	if (sc->sc_rxtslot != 0)
712 		tp->snd_cwnd = tp->t_maxseg;
713 	callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
714 
715 	INP_UNLOCK(inp);
716 
717 	tcpstat.tcps_accepts++;
718 	return (so);
719 
720 abort:
721 	INP_UNLOCK(inp);
722 abort2:
723 	if (so != NULL)
724 		(void) soabort(so);
725 	return (NULL);
726 }
727 
728 /*
729  * This function gets called when we receive an ACK for a
730  * socket in the LISTEN state.  We look up the connection
731  * in the syncache, and if its there, we pull it out of
732  * the cache and turn it into a full-blown connection in
733  * the SYN-RECEIVED state.
734  */
735 int
736 syncache_expand(inc, th, sop, m)
737 	struct in_conninfo *inc;
738 	struct tcphdr *th;
739 	struct socket **sop;
740 	struct mbuf *m;
741 {
742 	struct syncache *sc;
743 	struct syncache_head *sch;
744 	struct socket *so;
745 
746 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
747 
748 	sc = syncache_lookup(inc, &sch);
749 	if (sc == NULL) {
750 		/*
751 		 * There is no syncache entry, so see if this ACK is
752 		 * a returning syncookie.  To do this, first:
753 		 *  A. See if this socket has had a syncache entry dropped in
754 		 *     the past.  We don't want to accept a bogus syncookie
755 		 *     if we've never received a SYN.
756 		 *  B. check that the syncookie is valid.  If it is, then
757 		 *     cobble up a fake syncache entry, and return.
758 		 */
759 		if (!tcp_syncookies)
760 			return (0);
761 		sc = syncookie_lookup(inc, th, *sop);
762 		if (sc == NULL)
763 			return (0);
764 		sch = NULL;
765 		tcpstat.tcps_sc_recvcookie++;
766 	}
767 
768 	/*
769 	 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
770 	 */
771 	if (th->th_ack != sc->sc_iss + 1)
772 		return (0);
773 
774 	so = syncache_socket(sc, *sop, m);
775 	if (so == NULL) {
776 #if 0
777 resetandabort:
778 		/* XXXjlemon check this - is this correct? */
779 		(void) tcp_respond(NULL, m, m, th,
780 		    th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK);
781 #endif
782 		m_freem(m);			/* XXX only needed for above */
783 		tcpstat.tcps_sc_aborted++;
784 	} else
785 		tcpstat.tcps_sc_completed++;
786 
787 	if (sch == NULL)
788 		syncache_free(sc);
789 	else
790 		syncache_drop(sc, sch);
791 	*sop = so;
792 	return (1);
793 }
794 
795 /*
796  * Given a LISTEN socket and an inbound SYN request, add
797  * this to the syn cache, and send back a segment:
798  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
799  * to the source.
800  *
801  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
802  * Doing so would require that we hold onto the data and deliver it
803  * to the application.  However, if we are the target of a SYN-flood
804  * DoS attack, an attacker could send data which would eventually
805  * consume all available buffer space if it were ACKed.  By not ACKing
806  * the data, we avoid this DoS scenario.
807  */
808 int
809 syncache_add(inc, to, th, sop, m)
810 	struct in_conninfo *inc;
811 	struct tcpopt *to;
812 	struct tcphdr *th;
813 	struct socket **sop;
814 	struct mbuf *m;
815 {
816 	struct tcpcb *tp;
817 	struct socket *so;
818 	struct syncache *sc = NULL;
819 	struct syncache_head *sch;
820 	struct mbuf *ipopts = NULL;
821 	u_int32_t flowtmp;
822 	int i, win;
823 
824 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
825 
826 	so = *sop;
827 	tp = sototcpcb(so);
828 
829 	/*
830 	 * Remember the IP options, if any.
831 	 */
832 #ifdef INET6
833 	if (!inc->inc_isipv6)
834 #endif
835 		ipopts = ip_srcroute(m);
836 
837 	/*
838 	 * See if we already have an entry for this connection.
839 	 * If we do, resend the SYN,ACK, and reset the retransmit timer.
840 	 *
841 	 * XXX
842 	 * should the syncache be re-initialized with the contents
843 	 * of the new SYN here (which may have different options?)
844 	 */
845 	sc = syncache_lookup(inc, &sch);
846 	if (sc != NULL) {
847 		tcpstat.tcps_sc_dupsyn++;
848 		if (ipopts) {
849 			/*
850 			 * If we were remembering a previous source route,
851 			 * forget it and use the new one we've been given.
852 			 */
853 			if (sc->sc_ipopts)
854 				(void) m_free(sc->sc_ipopts);
855 			sc->sc_ipopts = ipopts;
856 		}
857 		/*
858 		 * Update timestamp if present.
859 		 */
860 		if (sc->sc_flags & SCF_TIMESTAMP)
861 			sc->sc_tsrecent = to->to_tsval;
862 		/*
863 		 * PCB may have changed, pick up new values.
864 		 */
865 		sc->sc_tp = tp;
866 		sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
867 #ifdef TCPDEBUG
868 		if (syncache_respond(sc, m, so) == 0) {
869 #else
870 		if (syncache_respond(sc, m) == 0) {
871 #endif
872 			/* NB: guarded by INP_INFO_WLOCK(&tcbinfo) */
873 			TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot],
874 			    sc, sc_timerq);
875 			SYNCACHE_TIMEOUT(sc, sc->sc_rxtslot);
876 			tcpstat.tcps_sndacks++;
877 			tcpstat.tcps_sndtotal++;
878 		}
879 		*sop = NULL;
880 		return (1);
881 	}
882 
883 	sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT);
884 	if (sc == NULL) {
885 		/*
886 		 * The zone allocator couldn't provide more entries.
887 		 * Treat this as if the cache was full; drop the oldest
888 		 * entry and insert the new one.
889 		 */
890 		/* NB: guarded by INP_INFO_WLOCK(&tcbinfo) */
891 		for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
892 			sc = TAILQ_FIRST(&tcp_syncache.timerq[i]);
893 			if (sc != NULL)
894 				break;
895 		}
896 		sc->sc_tp->ts_recent = ticks;
897 		syncache_drop(sc, NULL);
898 		tcpstat.tcps_sc_zonefail++;
899 		sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT);
900 		if (sc == NULL) {
901 			if (ipopts)
902 				(void) m_free(ipopts);
903 			return (0);
904 		}
905 	}
906 
907 	/*
908 	 * Fill in the syncache values.
909 	 */
910 	bzero(sc, sizeof(*sc));
911 	sc->sc_tp = tp;
912 	sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
913 	sc->sc_ipopts = ipopts;
914 	sc->sc_inc.inc_fport = inc->inc_fport;
915 	sc->sc_inc.inc_lport = inc->inc_lport;
916 #ifdef INET6
917 	sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
918 	if (inc->inc_isipv6) {
919 		sc->sc_inc.inc6_faddr = inc->inc6_faddr;
920 		sc->sc_inc.inc6_laddr = inc->inc6_laddr;
921 	} else
922 #endif
923 	{
924 		sc->sc_inc.inc_faddr = inc->inc_faddr;
925 		sc->sc_inc.inc_laddr = inc->inc_laddr;
926 	}
927 	sc->sc_irs = th->th_seq;
928 	sc->sc_flags = 0;
929 	sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0;
930 	sc->sc_flowlabel = 0;
931 	if (tcp_syncookies) {
932 		sc->sc_iss = syncookie_generate(sc, &flowtmp);
933 #ifdef INET6
934 		if (inc->inc_isipv6 &&
935 		    (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL)) {
936 			sc->sc_flowlabel = flowtmp & IPV6_FLOWLABEL_MASK;
937 		}
938 #endif
939 	} else {
940 		sc->sc_iss = arc4random();
941 #ifdef INET6
942 		if (inc->inc_isipv6 &&
943 		    (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL)) {
944 			sc->sc_flowlabel =
945 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
946 		}
947 #endif
948 	}
949 
950 	/* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */
951 	win = sbspace(&so->so_rcv);
952 	win = imax(win, 0);
953 	win = imin(win, TCP_MAXWIN);
954 	sc->sc_wnd = win;
955 
956 	if (tcp_do_rfc1323) {
957 		/*
958 		 * A timestamp received in a SYN makes
959 		 * it ok to send timestamp requests and replies.
960 		 */
961 		if (to->to_flags & TOF_TS) {
962 			sc->sc_tsrecent = to->to_tsval;
963 			sc->sc_flags |= SCF_TIMESTAMP;
964 		}
965 		if (to->to_flags & TOF_SCALE) {
966 			int wscale = 0;
967 
968 			/* Compute proper scaling value from buffer space */
969 			while (wscale < TCP_MAX_WINSHIFT &&
970 			    (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat)
971 				wscale++;
972 			sc->sc_request_r_scale = wscale;
973 			sc->sc_requested_s_scale = to->to_requested_s_scale;
974 			sc->sc_flags |= SCF_WINSCALE;
975 		}
976 	}
977 	if (tp->t_flags & TF_NOOPT)
978 		sc->sc_flags = SCF_NOOPT;
979 #ifdef TCP_SIGNATURE
980 	/*
981 	 * If listening socket requested TCP digests, and received SYN
982 	 * contains the option, flag this in the syncache so that
983 	 * syncache_respond() will do the right thing with the SYN+ACK.
984 	 * XXX Currently we always record the option by default and will
985 	 * attempt to use it in syncache_respond().
986 	 */
987 	if (to->to_flags & TOF_SIGNATURE)
988 		sc->sc_flags = SCF_SIGNATURE;
989 #endif
990 
991 	if (to->to_flags & TOF_SACK)
992 		sc->sc_flags |= SCF_SACK;
993 
994 	/*
995 	 * Do a standard 3-way handshake.
996 	 */
997 #ifdef TCPDEBUG
998 	if (syncache_respond(sc, m, so) == 0) {
999 #else
1000 	if (syncache_respond(sc, m) == 0) {
1001 #endif
1002 		syncache_insert(sc, sch);
1003 		tcpstat.tcps_sndacks++;
1004 		tcpstat.tcps_sndtotal++;
1005 	} else {
1006 		syncache_free(sc);
1007 		tcpstat.tcps_sc_dropped++;
1008 	}
1009 	*sop = NULL;
1010 	return (1);
1011 }
1012 
1013 #ifdef TCPDEBUG
1014 static int
1015 syncache_respond(sc, m, so)
1016 	struct syncache *sc;
1017 	struct mbuf *m;
1018 	struct socket *so;
1019 #else
1020 static int
1021 syncache_respond(sc, m)
1022 	struct syncache *sc;
1023 	struct mbuf *m;
1024 #endif
1025 {
1026 	u_int8_t *optp;
1027 	int optlen, error;
1028 	u_int16_t tlen, hlen, mssopt;
1029 	struct ip *ip = NULL;
1030 	struct tcphdr *th;
1031 	struct inpcb *inp;
1032 #ifdef INET6
1033 	struct ip6_hdr *ip6 = NULL;
1034 #endif
1035 
1036 	hlen =
1037 #ifdef INET6
1038 	       (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) :
1039 #endif
1040 		sizeof(struct ip);
1041 
1042 	KASSERT((&sc->sc_inc) != NULL, ("syncache_respond with NULL in_conninfo pointer"));
1043 
1044 	/* Determine MSS we advertize to other end of connection */
1045 	mssopt = tcp_mssopt(&sc->sc_inc);
1046 
1047 	/* Compute the size of the TCP options. */
1048 	if (sc->sc_flags & SCF_NOOPT) {
1049 		optlen = 0;
1050 	} else {
1051 		optlen = TCPOLEN_MAXSEG +
1052 		    ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) +
1053 		    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
1054 #ifdef TCP_SIGNATURE
1055 		optlen += (sc->sc_flags & SCF_SIGNATURE) ?
1056 		    TCPOLEN_SIGNATURE + 2 : 0;
1057 #endif
1058 		optlen += ((sc->sc_flags & SCF_SACK) ? 4 : 0);
1059 	}
1060 	tlen = hlen + sizeof(struct tcphdr) + optlen;
1061 
1062 	/*
1063 	 * XXX
1064 	 * assume that the entire packet will fit in a header mbuf
1065 	 */
1066 	KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small"));
1067 
1068 	/*
1069 	 * XXX shouldn't this reuse the mbuf if possible ?
1070 	 * Create the IP+TCP header from scratch.
1071 	 */
1072 	if (m)
1073 		m_freem(m);
1074 
1075 	m = m_gethdr(M_DONTWAIT, MT_HEADER);
1076 	if (m == NULL)
1077 		return (ENOBUFS);
1078 	m->m_data += max_linkhdr;
1079 	m->m_len = tlen;
1080 	m->m_pkthdr.len = tlen;
1081 	m->m_pkthdr.rcvif = NULL;
1082 	inp = sc->sc_tp->t_inpcb;
1083 	INP_LOCK(inp);
1084 #ifdef MAC
1085 	mac_create_mbuf_from_inpcb(inp, m);
1086 #endif
1087 
1088 #ifdef INET6
1089 	if (sc->sc_inc.inc_isipv6) {
1090 		ip6 = mtod(m, struct ip6_hdr *);
1091 		ip6->ip6_vfc = IPV6_VERSION;
1092 		ip6->ip6_nxt = IPPROTO_TCP;
1093 		ip6->ip6_src = sc->sc_inc.inc6_laddr;
1094 		ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1095 		ip6->ip6_plen = htons(tlen - hlen);
1096 		/* ip6_hlim is set after checksum */
1097 		ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1098 		ip6->ip6_flow |= sc->sc_flowlabel;
1099 
1100 		th = (struct tcphdr *)(ip6 + 1);
1101 	} else
1102 #endif
1103 	{
1104 		ip = mtod(m, struct ip *);
1105 		ip->ip_v = IPVERSION;
1106 		ip->ip_hl = sizeof(struct ip) >> 2;
1107 		ip->ip_len = tlen;
1108 		ip->ip_id = 0;
1109 		ip->ip_off = 0;
1110 		ip->ip_sum = 0;
1111 		ip->ip_p = IPPROTO_TCP;
1112 		ip->ip_src = sc->sc_inc.inc_laddr;
1113 		ip->ip_dst = sc->sc_inc.inc_faddr;
1114 		ip->ip_ttl = inp->inp_ip_ttl;   /* XXX */
1115 		ip->ip_tos = inp->inp_ip_tos;   /* XXX */
1116 
1117 		/*
1118 		 * See if we should do MTU discovery.  Route lookups are
1119 		 * expensive, so we will only unset the DF bit if:
1120 		 *
1121 		 *	1) path_mtu_discovery is disabled
1122 		 *	2) the SCF_UNREACH flag has been set
1123 		 */
1124 		if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1125 		       ip->ip_off |= IP_DF;
1126 
1127 		th = (struct tcphdr *)(ip + 1);
1128 	}
1129 	th->th_sport = sc->sc_inc.inc_lport;
1130 	th->th_dport = sc->sc_inc.inc_fport;
1131 
1132 	th->th_seq = htonl(sc->sc_iss);
1133 	th->th_ack = htonl(sc->sc_irs + 1);
1134 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1135 	th->th_x2 = 0;
1136 	th->th_flags = TH_SYN|TH_ACK;
1137 	th->th_win = htons(sc->sc_wnd);
1138 	th->th_urp = 0;
1139 
1140 	/* Tack on the TCP options. */
1141 	if (optlen != 0) {
1142 		optp = (u_int8_t *)(th + 1);
1143 		*optp++ = TCPOPT_MAXSEG;
1144 		*optp++ = TCPOLEN_MAXSEG;
1145 		*optp++ = (mssopt >> 8) & 0xff;
1146 		*optp++ = mssopt & 0xff;
1147 
1148 		if (sc->sc_flags & SCF_WINSCALE) {
1149 			*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
1150 			    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
1151 			    sc->sc_request_r_scale);
1152 			optp += 4;
1153 		}
1154 
1155 		if (sc->sc_flags & SCF_TIMESTAMP) {
1156 			u_int32_t *lp = (u_int32_t *)(optp);
1157 
1158 			/* Form timestamp option per appendix A of RFC 1323. */
1159 			*lp++ = htonl(TCPOPT_TSTAMP_HDR);
1160 			*lp++ = htonl(ticks);
1161 			*lp   = htonl(sc->sc_tsrecent);
1162 			optp += TCPOLEN_TSTAMP_APPA;
1163 		}
1164 
1165 #ifdef TCP_SIGNATURE
1166 		/*
1167 		 * Handle TCP-MD5 passive opener response.
1168 		 */
1169 		if (sc->sc_flags & SCF_SIGNATURE) {
1170 			u_int8_t *bp = optp;
1171 			int i;
1172 
1173 			*bp++ = TCPOPT_SIGNATURE;
1174 			*bp++ = TCPOLEN_SIGNATURE;
1175 			for (i = 0; i < TCP_SIGLEN; i++)
1176 				*bp++ = 0;
1177 			tcp_signature_compute(m, sizeof(struct ip), 0, optlen,
1178 			    optp + 2, IPSEC_DIR_OUTBOUND);
1179 			*bp++ = TCPOPT_NOP;
1180 			*bp++ = TCPOPT_EOL;
1181 			optp += TCPOLEN_SIGNATURE + 2;
1182 		}
1183 #endif /* TCP_SIGNATURE */
1184 
1185 	if (sc->sc_flags & SCF_SACK) {
1186 		*(u_int32_t *)optp = htonl(TCPOPT_SACK_PERMIT_HDR);
1187 		optp += 4;
1188 	}
1189 	}
1190 
1191 #ifdef INET6
1192 	if (sc->sc_inc.inc_isipv6) {
1193 		th->th_sum = 0;
1194 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
1195 		ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1196 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1197 	} else
1198 #endif
1199 	{
1200 		th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1201 		    htons(tlen - hlen + IPPROTO_TCP));
1202 		m->m_pkthdr.csum_flags = CSUM_TCP;
1203 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1204 #ifdef TCPDEBUG
1205 		/*
1206 		 * Trace.
1207 		 */
1208 		if (so != NULL && so->so_options & SO_DEBUG) {
1209 			struct tcpcb *tp = sototcpcb(so);
1210 			tcp_trace(TA_OUTPUT, tp->t_state, tp,
1211 			    mtod(m, void *), th, 0);
1212 		}
1213 #endif
1214 		error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, inp);
1215 	}
1216 	INP_UNLOCK(inp);
1217 	return (error);
1218 }
1219 
1220 /*
1221  * cookie layers:
1222  *
1223  *	|. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .|
1224  *	| peer iss                                                      |
1225  *	| MD5(laddr,faddr,secret,lport,fport)             |. . . . . . .|
1226  *	|                     0                       |(A)|             |
1227  * (A): peer mss index
1228  */
1229 
1230 /*
1231  * The values below are chosen to minimize the size of the tcp_secret
1232  * table, as well as providing roughly a 16 second lifetime for the cookie.
1233  */
1234 
1235 #define SYNCOOKIE_WNDBITS	5	/* exposed bits for window indexing */
1236 #define SYNCOOKIE_TIMESHIFT	1	/* scale ticks to window time units */
1237 
1238 #define SYNCOOKIE_WNDMASK	((1 << SYNCOOKIE_WNDBITS) - 1)
1239 #define SYNCOOKIE_NSECRETS	(1 << SYNCOOKIE_WNDBITS)
1240 #define SYNCOOKIE_TIMEOUT \
1241     (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT))
1242 #define SYNCOOKIE_DATAMASK	((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK)
1243 
1244 static struct {
1245 	u_int32_t	ts_secbits[4];
1246 	u_int		ts_expire;
1247 } tcp_secret[SYNCOOKIE_NSECRETS];
1248 
1249 static int tcp_msstab[] = { 0, 536, 1460, 8960 };
1250 
1251 static MD5_CTX syn_ctx;
1252 
1253 #define MD5Add(v)	MD5Update(&syn_ctx, (u_char *)&v, sizeof(v))
1254 
1255 struct md5_add {
1256 	u_int32_t laddr, faddr;
1257 	u_int32_t secbits[4];
1258 	u_int16_t lport, fport;
1259 };
1260 
1261 #ifdef CTASSERT
1262 CTASSERT(sizeof(struct md5_add) == 28);
1263 #endif
1264 
1265 /*
1266  * Consider the problem of a recreated (and retransmitted) cookie.  If the
1267  * original SYN was accepted, the connection is established.  The second
1268  * SYN is inflight, and if it arrives with an ISN that falls within the
1269  * receive window, the connection is killed.
1270  *
1271  * However, since cookies have other problems, this may not be worth
1272  * worrying about.
1273  */
1274 
1275 static u_int32_t
1276 syncookie_generate(struct syncache *sc, u_int32_t *flowid)
1277 {
1278 	u_int32_t md5_buffer[4];
1279 	u_int32_t data;
1280 	int idx, i;
1281 	struct md5_add add;
1282 
1283 	/* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */
1284 
1285 	idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK;
1286 	if (tcp_secret[idx].ts_expire < ticks) {
1287 		for (i = 0; i < 4; i++)
1288 			tcp_secret[idx].ts_secbits[i] = arc4random();
1289 		tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT;
1290 	}
1291 	for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--)
1292 		if (tcp_msstab[data] <= sc->sc_peer_mss)
1293 			break;
1294 	data = (data << SYNCOOKIE_WNDBITS) | idx;
1295 	data ^= sc->sc_irs;				/* peer's iss */
1296 	MD5Init(&syn_ctx);
1297 #ifdef INET6
1298 	if (sc->sc_inc.inc_isipv6) {
1299 		MD5Add(sc->sc_inc.inc6_laddr);
1300 		MD5Add(sc->sc_inc.inc6_faddr);
1301 		add.laddr = 0;
1302 		add.faddr = 0;
1303 	} else
1304 #endif
1305 	{
1306 		add.laddr = sc->sc_inc.inc_laddr.s_addr;
1307 		add.faddr = sc->sc_inc.inc_faddr.s_addr;
1308 	}
1309 	add.lport = sc->sc_inc.inc_lport;
1310 	add.fport = sc->sc_inc.inc_fport;
1311 	add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1312 	add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1313 	add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1314 	add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1315 	MD5Add(add);
1316 	MD5Final((u_char *)&md5_buffer, &syn_ctx);
1317 	data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK);
1318 	*flowid = md5_buffer[1];
1319 	return (data);
1320 }
1321 
1322 static struct syncache *
1323 syncookie_lookup(inc, th, so)
1324 	struct in_conninfo *inc;
1325 	struct tcphdr *th;
1326 	struct socket *so;
1327 {
1328 	u_int32_t md5_buffer[4];
1329 	struct syncache *sc;
1330 	u_int32_t data;
1331 	int wnd, idx;
1332 	struct md5_add add;
1333 
1334 	/* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */
1335 
1336 	data = (th->th_ack - 1) ^ (th->th_seq - 1);	/* remove ISS */
1337 	idx = data & SYNCOOKIE_WNDMASK;
1338 	if (tcp_secret[idx].ts_expire < ticks ||
1339 	    sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks)
1340 		return (NULL);
1341 	MD5Init(&syn_ctx);
1342 #ifdef INET6
1343 	if (inc->inc_isipv6) {
1344 		MD5Add(inc->inc6_laddr);
1345 		MD5Add(inc->inc6_faddr);
1346 		add.laddr = 0;
1347 		add.faddr = 0;
1348 	} else
1349 #endif
1350 	{
1351 		add.laddr = inc->inc_laddr.s_addr;
1352 		add.faddr = inc->inc_faddr.s_addr;
1353 	}
1354 	add.lport = inc->inc_lport;
1355 	add.fport = inc->inc_fport;
1356 	add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1357 	add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1358 	add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1359 	add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1360 	MD5Add(add);
1361 	MD5Final((u_char *)&md5_buffer, &syn_ctx);
1362 	data ^= md5_buffer[0];
1363 	if ((data & ~SYNCOOKIE_DATAMASK) != 0)
1364 		return (NULL);
1365 	data = data >> SYNCOOKIE_WNDBITS;
1366 
1367 	sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT);
1368 	if (sc == NULL)
1369 		return (NULL);
1370 	/*
1371 	 * Fill in the syncache values.
1372 	 * XXX duplicate code from syncache_add
1373 	 */
1374 	sc->sc_ipopts = NULL;
1375 	sc->sc_inc.inc_fport = inc->inc_fport;
1376 	sc->sc_inc.inc_lport = inc->inc_lport;
1377 	sc->sc_tp = sototcpcb(so);
1378 #ifdef INET6
1379 	sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1380 	if (inc->inc_isipv6) {
1381 		sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1382 		sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1383 		if (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL)
1384 			sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1385 	} else
1386 #endif
1387 	{
1388 		sc->sc_inc.inc_faddr = inc->inc_faddr;
1389 		sc->sc_inc.inc_laddr = inc->inc_laddr;
1390 	}
1391 	sc->sc_irs = th->th_seq - 1;
1392 	sc->sc_iss = th->th_ack - 1;
1393 	wnd = sbspace(&so->so_rcv);
1394 	wnd = imax(wnd, 0);
1395 	wnd = imin(wnd, TCP_MAXWIN);
1396 	sc->sc_wnd = wnd;
1397 	sc->sc_flags = 0;
1398 	sc->sc_rxtslot = 0;
1399 	sc->sc_peer_mss = tcp_msstab[data];
1400 	return (sc);
1401 }
1402