xref: /freebsd/sys/netinet/tcp_syncache.c (revision 419ecd5dee28e23b759fc90ff551d2fd188cb86f)
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  * $FreeBSD$
33  */
34 
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 #include "opt_ipsec.h"
38 #include "opt_mac.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/sysctl.h>
44 #include <sys/lock.h>
45 #include <sys/mutex.h>
46 #include <sys/malloc.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 #include <sys/syslog.h>
54 
55 #include <vm/uma.h>
56 
57 #include <net/if.h>
58 #include <net/route.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/ip.h>
63 #include <netinet/in_var.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/ip_var.h>
66 #include <netinet/ip_options.h>
67 #ifdef INET6
68 #include <netinet/ip6.h>
69 #include <netinet/icmp6.h>
70 #include <netinet6/nd6.h>
71 #include <netinet6/ip6_var.h>
72 #include <netinet6/in6_pcb.h>
73 #endif
74 #include <netinet/tcp.h>
75 #include <netinet/tcp_fsm.h>
76 #include <netinet/tcp_seq.h>
77 #include <netinet/tcp_timer.h>
78 #include <netinet/tcp_var.h>
79 #ifdef INET6
80 #include <netinet6/tcp6_var.h>
81 #endif
82 
83 #ifdef IPSEC
84 #include <netinet6/ipsec.h>
85 #ifdef INET6
86 #include <netinet6/ipsec6.h>
87 #endif
88 #endif /*IPSEC*/
89 
90 #ifdef FAST_IPSEC
91 #include <netipsec/ipsec.h>
92 #ifdef INET6
93 #include <netipsec/ipsec6.h>
94 #endif
95 #include <netipsec/key.h>
96 #endif /*FAST_IPSEC*/
97 
98 #include <machine/in_cksum.h>
99 
100 #include <security/mac/mac_framework.h>
101 
102 static int tcp_syncookies = 1;
103 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
104     &tcp_syncookies, 0,
105     "Use TCP SYN cookies if the syncache overflows");
106 
107 static int tcp_syncookiesonly = 0;
108 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW,
109     &tcp_syncookiesonly, 0,
110     "Use only TCP SYN cookies");
111 
112 #define	SYNCOOKIE_SECRET_SIZE	8	/* dwords */
113 #define	SYNCOOKIE_LIFETIME	16	/* seconds */
114 
115 struct syncache {
116 	TAILQ_ENTRY(syncache)	sc_hash;
117 	struct		in_conninfo sc_inc;	/* addresses */
118 	u_long		sc_rxttime;		/* retransmit time */
119 	u_int16_t	sc_rxmits;		/* retransmit counter */
120 
121 	u_int32_t	sc_tsreflect;		/* timestamp to reflect */
122 	u_int32_t	sc_ts;			/* our timestamp to send */
123 	u_int32_t	sc_tsoff;		/* ts offset w/ syncookies */
124 	u_int32_t	sc_flowlabel;		/* IPv6 flowlabel */
125 	tcp_seq		sc_irs;			/* seq from peer */
126 	tcp_seq		sc_iss;			/* our ISS */
127 	struct		mbuf *sc_ipopts;	/* source route */
128 
129 	u_int16_t	sc_peer_mss;		/* peer's MSS */
130 	u_int16_t	sc_wnd;			/* advertised window */
131 	u_int8_t	sc_ip_ttl;		/* IPv4 TTL */
132 	u_int8_t	sc_ip_tos;		/* IPv4 TOS */
133 	u_int8_t	sc_requested_s_scale:4,
134 			sc_requested_r_scale:4;
135 	u_int8_t	sc_flags;
136 #define SCF_NOOPT	0x01			/* no TCP options */
137 #define SCF_WINSCALE	0x02			/* negotiated window scaling */
138 #define SCF_TIMESTAMP	0x04			/* negotiated timestamps */
139 						/* MSS is implicit */
140 #define SCF_UNREACH	0x10			/* icmp unreachable received */
141 #define SCF_SIGNATURE	0x20			/* send MD5 digests */
142 #define SCF_SACK	0x80			/* send SACK option */
143 #ifdef MAC
144 	struct label	*sc_label;		/* MAC label reference */
145 #endif
146 };
147 
148 struct syncache_head {
149 	struct mtx	sch_mtx;
150 	TAILQ_HEAD(sch_head, syncache)	sch_bucket;
151 	struct callout	sch_timer;
152 	int		sch_nextc;
153 	u_int		sch_length;
154 	u_int		sch_oddeven;
155 	u_int32_t	sch_secbits_odd[SYNCOOKIE_SECRET_SIZE];
156 	u_int32_t	sch_secbits_even[SYNCOOKIE_SECRET_SIZE];
157 	u_int		sch_reseed;		/* time_uptime, seconds */
158 };
159 
160 static void	 syncache_drop(struct syncache *, struct syncache_head *);
161 static void	 syncache_free(struct syncache *);
162 static void	 syncache_insert(struct syncache *, struct syncache_head *);
163 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
164 static int	 syncache_respond(struct syncache *);
165 static struct	 socket *syncache_socket(struct syncache *, struct socket *,
166 		    struct mbuf *m);
167 static void	 syncache_timer(void *);
168 static void	 syncookie_generate(struct syncache_head *, struct syncache *,
169 		    u_int32_t *);
170 static struct syncache
171 		*syncookie_lookup(struct in_conninfo *, struct syncache_head *,
172 		    struct syncache *, struct tcpopt *, struct tcphdr *,
173 		    struct socket *);
174 
175 /*
176  * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
177  * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds,
178  * the odds are that the user has given up attempting to connect by then.
179  */
180 #define SYNCACHE_MAXREXMTS		3
181 
182 /* Arbitrary values */
183 #define TCP_SYNCACHE_HASHSIZE		512
184 #define TCP_SYNCACHE_BUCKETLIMIT	30
185 
186 struct tcp_syncache {
187 	struct	syncache_head *hashbase;
188 	uma_zone_t zone;
189 	u_int	hashsize;
190 	u_int	hashmask;
191 	u_int	bucket_limit;
192 	u_int	cache_count;		/* XXX: unprotected */
193 	u_int	cache_limit;
194 	u_int	rexmt_limit;
195 	u_int	hash_secret;
196 };
197 static struct tcp_syncache tcp_syncache;
198 
199 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
200 
201 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
202      &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
203 
204 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
205      &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
206 
207 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
208      &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
209 
210 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
211      &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
212 
213 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
214      &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
215 
216 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
217 
218 #define SYNCACHE_HASH(inc, mask)					\
219 	((tcp_syncache.hash_secret ^					\
220 	  (inc)->inc_faddr.s_addr ^					\
221 	  ((inc)->inc_faddr.s_addr >> 16) ^				\
222 	  (inc)->inc_fport ^ (inc)->inc_lport) & mask)
223 
224 #define SYNCACHE_HASH6(inc, mask)					\
225 	((tcp_syncache.hash_secret ^					\
226 	  (inc)->inc6_faddr.s6_addr32[0] ^				\
227 	  (inc)->inc6_faddr.s6_addr32[3] ^				\
228 	  (inc)->inc_fport ^ (inc)->inc_lport) & mask)
229 
230 #define ENDPTS_EQ(a, b) (						\
231 	(a)->ie_fport == (b)->ie_fport &&				\
232 	(a)->ie_lport == (b)->ie_lport &&				\
233 	(a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr &&			\
234 	(a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr			\
235 )
236 
237 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
238 
239 #define SYNCACHE_TIMEOUT(sc, sch, co) do {				\
240 	(sc)->sc_rxmits++;						\
241 	(sc)->sc_rxttime = ticks +					\
242 		TCPTV_RTOBASE * tcp_backoff[(sc)->sc_rxmits - 1];	\
243 	if ((sch)->sch_nextc > (sc)->sc_rxttime)			\
244 		(sch)->sch_nextc = (sc)->sc_rxttime;			\
245 	if (!TAILQ_EMPTY(&(sch)->sch_bucket) && !(co))			\
246 		callout_reset(&(sch)->sch_timer,			\
247 			(sch)->sch_nextc - ticks,			\
248 			syncache_timer, (void *)(sch));			\
249 } while (0)
250 
251 #define	SCH_LOCK(sch)		mtx_lock(&(sch)->sch_mtx)
252 #define	SCH_UNLOCK(sch)		mtx_unlock(&(sch)->sch_mtx)
253 #define	SCH_LOCK_ASSERT(sch)	mtx_assert(&(sch)->sch_mtx, MA_OWNED)
254 
255 /*
256  * Requires the syncache entry to be already removed from the bucket list.
257  */
258 static void
259 syncache_free(struct syncache *sc)
260 {
261 	if (sc->sc_ipopts)
262 		(void) m_free(sc->sc_ipopts);
263 #ifdef MAC
264 	mac_destroy_syncache(&sc->sc_label);
265 #endif
266 
267 	uma_zfree(tcp_syncache.zone, sc);
268 }
269 
270 void
271 syncache_init(void)
272 {
273 	int i;
274 
275 	tcp_syncache.cache_count = 0;
276 	tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
277 	tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
278 	tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
279 	tcp_syncache.hash_secret = arc4random();
280 
281 	TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
282 	    &tcp_syncache.hashsize);
283 	TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
284 	    &tcp_syncache.bucket_limit);
285 	if (!powerof2(tcp_syncache.hashsize) || tcp_syncache.hashsize == 0) {
286 		printf("WARNING: syncache hash size is not a power of 2.\n");
287 		tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
288 	}
289 	tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
290 
291 	/* Set limits. */
292 	tcp_syncache.cache_limit =
293 	    tcp_syncache.hashsize * tcp_syncache.bucket_limit;
294 	TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
295 	    &tcp_syncache.cache_limit);
296 
297 	/* Allocate the hash table. */
298 	MALLOC(tcp_syncache.hashbase, struct syncache_head *,
299 	    tcp_syncache.hashsize * sizeof(struct syncache_head),
300 	    M_SYNCACHE, M_WAITOK | M_ZERO);
301 
302 	/* Initialize the hash buckets. */
303 	for (i = 0; i < tcp_syncache.hashsize; i++) {
304 		TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket);
305 		mtx_init(&tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
306 			 NULL, MTX_DEF);
307 		callout_init_mtx(&tcp_syncache.hashbase[i].sch_timer,
308 			 &tcp_syncache.hashbase[i].sch_mtx, 0);
309 		tcp_syncache.hashbase[i].sch_length = 0;
310 	}
311 
312 	/* Create the syncache entry zone. */
313 	tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
314 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
315 	uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit);
316 }
317 
318 /*
319  * Inserts a syncache entry into the specified bucket row.
320  * Locks and unlocks the syncache_head autonomously.
321  */
322 static void
323 syncache_insert(struct syncache *sc, struct syncache_head *sch)
324 {
325 	struct syncache *sc2;
326 
327 	SCH_LOCK(sch);
328 
329 	/*
330 	 * Make sure that we don't overflow the per-bucket limit.
331 	 * If the bucket is full, toss the oldest element.
332 	 */
333 	if (sch->sch_length >= tcp_syncache.bucket_limit) {
334 		KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
335 			("sch->sch_length incorrect"));
336 		sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
337 		syncache_drop(sc2, sch);
338 		tcpstat.tcps_sc_bucketoverflow++;
339 	}
340 
341 	/* Put it into the bucket. */
342 	TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
343 	sch->sch_length++;
344 
345 	/* Reinitialize the bucket row's timer. */
346 	SYNCACHE_TIMEOUT(sc, sch, 1);
347 
348 	SCH_UNLOCK(sch);
349 
350 	tcp_syncache.cache_count++;
351 	tcpstat.tcps_sc_added++;
352 }
353 
354 /*
355  * Remove and free entry from syncache bucket row.
356  * Expects locked syncache head.
357  */
358 static void
359 syncache_drop(struct syncache *sc, struct syncache_head *sch)
360 {
361 
362 	SCH_LOCK_ASSERT(sch);
363 
364 	TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
365 	sch->sch_length--;
366 
367 	syncache_free(sc);
368 	tcp_syncache.cache_count--;
369 }
370 
371 /*
372  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
373  * If we have retransmitted an entry the maximum number of times, expire it.
374  * One separate timer for each bucket row.
375  */
376 static void
377 syncache_timer(void *xsch)
378 {
379 	struct syncache_head *sch = (struct syncache_head *)xsch;
380 	struct syncache *sc, *nsc;
381 	int tick = ticks;
382 	char *s;
383 
384 	/* NB: syncache_head has already been locked by the callout. */
385 	SCH_LOCK_ASSERT(sch);
386 
387 	TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
388 		/*
389 		 * We do not check if the listen socket still exists
390 		 * and accept the case where the listen socket may be
391 		 * gone by the time we resend the SYN/ACK.  We do
392 		 * not expect this to happens often. If it does,
393 		 * then the RST will be sent by the time the remote
394 		 * host does the SYN/ACK->ACK.
395 		 */
396 		if (sc->sc_rxttime >= tick) {
397 			if (sc->sc_rxttime < sch->sch_nextc)
398 				sch->sch_nextc = sc->sc_rxttime;
399 			continue;
400 		}
401 
402 		if (sc->sc_rxmits > tcp_syncache.rexmt_limit) {
403 			if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
404 				log(LOG_DEBUG, "%s; %s: Response timeout\n",
405 				    s, __func__);
406 				free(s, M_TCPLOG);
407 			}
408 			syncache_drop(sc, sch);
409 			tcpstat.tcps_sc_stale++;
410 			continue;
411 		}
412 
413 		(void) syncache_respond(sc);
414 		tcpstat.tcps_sc_retransmitted++;
415 		SYNCACHE_TIMEOUT(sc, sch, 0);
416 	}
417 	if (!TAILQ_EMPTY(&(sch)->sch_bucket))
418 		callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
419 			syncache_timer, (void *)(sch));
420 }
421 
422 /*
423  * Find an entry in the syncache.
424  * Returns always with locked syncache_head plus a matching entry or NULL.
425  */
426 struct syncache *
427 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
428 {
429 	struct syncache *sc;
430 	struct syncache_head *sch;
431 
432 #ifdef INET6
433 	if (inc->inc_isipv6) {
434 		sch = &tcp_syncache.hashbase[
435 		    SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
436 		*schp = sch;
437 
438 		SCH_LOCK(sch);
439 
440 		/* Circle through bucket row to find matching entry. */
441 		TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
442 			if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
443 				return (sc);
444 		}
445 	} else
446 #endif
447 	{
448 		sch = &tcp_syncache.hashbase[
449 		    SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
450 		*schp = sch;
451 
452 		SCH_LOCK(sch);
453 
454 		/* Circle through bucket row to find matching entry. */
455 		TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
456 #ifdef INET6
457 			if (sc->sc_inc.inc_isipv6)
458 				continue;
459 #endif
460 			if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
461 				return (sc);
462 		}
463 	}
464 	SCH_LOCK_ASSERT(*schp);
465 	return (NULL);			/* always returns with locked sch */
466 }
467 
468 /*
469  * This function is called when we get a RST for a
470  * non-existent connection, so that we can see if the
471  * connection is in the syn cache.  If it is, zap it.
472  */
473 void
474 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
475 {
476 	struct syncache *sc;
477 	struct syncache_head *sch;
478 
479 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
480 	SCH_LOCK_ASSERT(sch);
481 	if (sc == NULL)
482 		goto done;
483 
484 	/*
485 	 * If the RST bit is set, check the sequence number to see
486 	 * if this is a valid reset segment.
487 	 * RFC 793 page 37:
488 	 *   In all states except SYN-SENT, all reset (RST) segments
489 	 *   are validated by checking their SEQ-fields.  A reset is
490 	 *   valid if its sequence number is in the window.
491 	 *
492 	 *   The sequence number in the reset segment is normally an
493 	 *   echo of our outgoing acknowlegement numbers, but some hosts
494 	 *   send a reset with the sequence number at the rightmost edge
495 	 *   of our receive window, and we have to handle this case.
496 	 */
497 	if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
498 	    SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
499 		syncache_drop(sc, sch);
500 		tcpstat.tcps_sc_reset++;
501 	}
502 done:
503 	SCH_UNLOCK(sch);
504 }
505 
506 void
507 syncache_badack(struct in_conninfo *inc)
508 {
509 	struct syncache *sc;
510 	struct syncache_head *sch;
511 
512 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
513 	SCH_LOCK_ASSERT(sch);
514 	if (sc != NULL) {
515 		syncache_drop(sc, sch);
516 		tcpstat.tcps_sc_badack++;
517 	}
518 	SCH_UNLOCK(sch);
519 }
520 
521 void
522 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
523 {
524 	struct syncache *sc;
525 	struct syncache_head *sch;
526 
527 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
528 	SCH_LOCK_ASSERT(sch);
529 	if (sc == NULL)
530 		goto done;
531 
532 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
533 	if (ntohl(th->th_seq) != sc->sc_iss)
534 		goto done;
535 
536 	/*
537 	 * If we've rertransmitted 3 times and this is our second error,
538 	 * we remove the entry.  Otherwise, we allow it to continue on.
539 	 * This prevents us from incorrectly nuking an entry during a
540 	 * spurious network outage.
541 	 *
542 	 * See tcp_notify().
543 	 */
544 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
545 		sc->sc_flags |= SCF_UNREACH;
546 		goto done;
547 	}
548 	syncache_drop(sc, sch);
549 	tcpstat.tcps_sc_unreach++;
550 done:
551 	SCH_UNLOCK(sch);
552 }
553 
554 /*
555  * Build a new TCP socket structure from a syncache entry.
556  */
557 static struct socket *
558 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
559 {
560 	struct inpcb *inp = NULL;
561 	struct socket *so;
562 	struct tcpcb *tp;
563 	char *s;
564 
565 	NET_ASSERT_GIANT();
566 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
567 
568 	/*
569 	 * Ok, create the full blown connection, and set things up
570 	 * as they would have been set up if we had created the
571 	 * connection when the SYN arrived.  If we can't create
572 	 * the connection, abort it.
573 	 */
574 	so = sonewconn(lso, SS_ISCONNECTED);
575 	if (so == NULL) {
576 		/*
577 		 * Drop the connection; we will either send a RST or
578 		 * have the peer retransmit its SYN again after its
579 		 * RTO and try again.
580 		 */
581 		tcpstat.tcps_listendrop++;
582 		if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
583 			log(LOG_DEBUG, "%s; %s: Socket create failed "
584 			    "due to limits or memory shortage\n",
585 			    s, __func__);
586 			free(s, M_TCPLOG);
587 		}
588 		goto abort2;
589 	}
590 #ifdef MAC
591 	SOCK_LOCK(so);
592 	mac_set_socket_peer_from_mbuf(m, so);
593 	SOCK_UNLOCK(so);
594 #endif
595 
596 	inp = sotoinpcb(so);
597 	INP_LOCK(inp);
598 
599 	/* Insert new socket into PCB hash list. */
600 	inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
601 #ifdef INET6
602 	if (sc->sc_inc.inc_isipv6) {
603 		inp->in6p_laddr = sc->sc_inc.inc6_laddr;
604 	} else {
605 		inp->inp_vflag &= ~INP_IPV6;
606 		inp->inp_vflag |= INP_IPV4;
607 #endif
608 		inp->inp_laddr = sc->sc_inc.inc_laddr;
609 #ifdef INET6
610 	}
611 #endif
612 	inp->inp_lport = sc->sc_inc.inc_lport;
613 	if (in_pcbinshash(inp) != 0) {
614 		/*
615 		 * Undo the assignments above if we failed to
616 		 * put the PCB on the hash lists.
617 		 */
618 #ifdef INET6
619 		if (sc->sc_inc.inc_isipv6)
620 			inp->in6p_laddr = in6addr_any;
621 		else
622 #endif
623 			inp->inp_laddr.s_addr = INADDR_ANY;
624 		inp->inp_lport = 0;
625 		goto abort;
626 	}
627 #ifdef IPSEC
628 	/* Copy old policy into new socket's. */
629 	if (ipsec_copy_pcbpolicy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
630 		printf("syncache_socket: could not copy policy\n");
631 #endif
632 #ifdef FAST_IPSEC
633 	/* Copy old policy into new socket's. */
634 	if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
635 		printf("syncache_socket: could not copy policy\n");
636 #endif
637 #ifdef INET6
638 	if (sc->sc_inc.inc_isipv6) {
639 		struct inpcb *oinp = sotoinpcb(lso);
640 		struct in6_addr laddr6;
641 		struct sockaddr_in6 sin6;
642 		/*
643 		 * Inherit socket options from the listening socket.
644 		 * Note that in6p_inputopts are not (and should not be)
645 		 * copied, since it stores previously received options and is
646 		 * used to detect if each new option is different than the
647 		 * previous one and hence should be passed to a user.
648 		 * If we copied in6p_inputopts, a user would not be able to
649 		 * receive options just after calling the accept system call.
650 		 */
651 		inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
652 		if (oinp->in6p_outputopts)
653 			inp->in6p_outputopts =
654 			    ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
655 
656 		sin6.sin6_family = AF_INET6;
657 		sin6.sin6_len = sizeof(sin6);
658 		sin6.sin6_addr = sc->sc_inc.inc6_faddr;
659 		sin6.sin6_port = sc->sc_inc.inc_fport;
660 		sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
661 		laddr6 = inp->in6p_laddr;
662 		if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
663 			inp->in6p_laddr = sc->sc_inc.inc6_laddr;
664 		if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
665 		    thread0.td_ucred)) {
666 			inp->in6p_laddr = laddr6;
667 			goto abort;
668 		}
669 		/* Override flowlabel from in6_pcbconnect. */
670 		inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK;
671 		inp->in6p_flowinfo |= sc->sc_flowlabel;
672 	} else
673 #endif
674 	{
675 		struct in_addr laddr;
676 		struct sockaddr_in sin;
677 
678 		inp->inp_options = ip_srcroute(m);
679 		if (inp->inp_options == NULL) {
680 			inp->inp_options = sc->sc_ipopts;
681 			sc->sc_ipopts = NULL;
682 		}
683 
684 		sin.sin_family = AF_INET;
685 		sin.sin_len = sizeof(sin);
686 		sin.sin_addr = sc->sc_inc.inc_faddr;
687 		sin.sin_port = sc->sc_inc.inc_fport;
688 		bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
689 		laddr = inp->inp_laddr;
690 		if (inp->inp_laddr.s_addr == INADDR_ANY)
691 			inp->inp_laddr = sc->sc_inc.inc_laddr;
692 		if (in_pcbconnect(inp, (struct sockaddr *)&sin,
693 		    thread0.td_ucred)) {
694 			inp->inp_laddr = laddr;
695 			goto abort;
696 		}
697 	}
698 	tp = intotcpcb(inp);
699 	tp->t_state = TCPS_SYN_RECEIVED;
700 	tp->iss = sc->sc_iss;
701 	tp->irs = sc->sc_irs;
702 	tcp_rcvseqinit(tp);
703 	tcp_sendseqinit(tp);
704 	tp->snd_wl1 = sc->sc_irs;
705 	tp->snd_max = tp->iss + 1;
706 	tp->snd_nxt = tp->iss + 1;
707 	tp->rcv_up = sc->sc_irs + 1;
708 	tp->rcv_wnd = sc->sc_wnd;
709 	tp->rcv_adv += tp->rcv_wnd;
710 	tp->last_ack_sent = tp->rcv_nxt;
711 
712 	tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
713 	if (sc->sc_flags & SCF_NOOPT)
714 		tp->t_flags |= TF_NOOPT;
715 	else {
716 		if (sc->sc_flags & SCF_WINSCALE) {
717 			tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
718 			tp->snd_scale = sc->sc_requested_s_scale;
719 			tp->request_r_scale = sc->sc_requested_r_scale;
720 		}
721 		if (sc->sc_flags & SCF_TIMESTAMP) {
722 			tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
723 			tp->ts_recent = sc->sc_tsreflect;
724 			tp->ts_recent_age = ticks;
725 			tp->ts_offset = sc->sc_tsoff;
726 		}
727 #ifdef TCP_SIGNATURE
728 		if (sc->sc_flags & SCF_SIGNATURE)
729 			tp->t_flags |= TF_SIGNATURE;
730 #endif
731 		if (sc->sc_flags & SCF_SACK)
732 			tp->t_flags |= TF_SACK_PERMIT;
733 	}
734 
735 	/*
736 	 * Set up MSS and get cached values from tcp_hostcache.
737 	 * This might overwrite some of the defaults we just set.
738 	 */
739 	tcp_mss(tp, sc->sc_peer_mss);
740 
741 	/*
742 	 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
743 	 */
744 	if (sc->sc_rxmits > 1)
745 		tp->snd_cwnd = tp->t_maxseg;
746 	tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
747 
748 	INP_UNLOCK(inp);
749 
750 	tcpstat.tcps_accepts++;
751 	return (so);
752 
753 abort:
754 	INP_UNLOCK(inp);
755 abort2:
756 	if (so != NULL)
757 		soabort(so);
758 	return (NULL);
759 }
760 
761 /*
762  * This function gets called when we receive an ACK for a
763  * socket in the LISTEN state.  We look up the connection
764  * in the syncache, and if its there, we pull it out of
765  * the cache and turn it into a full-blown connection in
766  * the SYN-RECEIVED state.
767  */
768 int
769 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
770     struct socket **lsop, struct mbuf *m)
771 {
772 	struct syncache *sc;
773 	struct syncache_head *sch;
774 	struct syncache scs;
775 	char *s;
776 
777 	/*
778 	 * Global TCP locks are held because we manipulate the PCB lists
779 	 * and create a new socket.
780 	 */
781 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
782 	KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
783 	    ("%s: can handle only ACK", __func__));
784 
785 	sc = syncache_lookup(inc, &sch);	/* returns locked sch */
786 	SCH_LOCK_ASSERT(sch);
787 	if (sc == NULL) {
788 		/*
789 		 * There is no syncache entry, so see if this ACK is
790 		 * a returning syncookie.  To do this, first:
791 		 *  A. See if this socket has had a syncache entry dropped in
792 		 *     the past.  We don't want to accept a bogus syncookie
793 		 *     if we've never received a SYN.
794 		 *  B. check that the syncookie is valid.  If it is, then
795 		 *     cobble up a fake syncache entry, and return.
796 		 */
797 		if (!tcp_syncookies) {
798 			SCH_UNLOCK(sch);
799 			if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
800 				log(LOG_DEBUG, "%s; %s: Spurious ACK\n",
801 				    s, __func__);
802 			goto failed;
803 		}
804 		bzero(&scs, sizeof(scs));
805 		sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
806 		SCH_UNLOCK(sch);
807 		if (sc == NULL) {
808 			if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
809 				log(LOG_DEBUG, "%s; %s: Segment failed "
810 				    "SYNCOOKIE authentication\n",
811 				    s, __func__);
812 			goto failed;
813 		}
814 		tcpstat.tcps_sc_recvcookie++;
815 	} else {
816 		/* Pull out the entry to unlock the bucket row. */
817 		TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
818 		sch->sch_length--;
819 		tcp_syncache.cache_count--;
820 		SCH_UNLOCK(sch);
821 	}
822 
823 	/*
824 	 * Segment validation:
825 	 * ACK must match our initial sequence number + 1 (the SYN|ACK).
826 	 */
827 	if (th->th_ack != sc->sc_iss + 1) {
828 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
829 			log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u\n",
830 			    s, __func__, th->th_ack, sc->sc_iss);
831 		goto failed;
832 	}
833 	/*
834 	 * The SEQ must match the received initial receive sequence
835 	 * number + 1 (the SYN) because we didn't ACK any data that
836 	 * may have come with the SYN.
837 	 */
838 	if (th->th_seq != sc->sc_irs + 1) {
839 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
840 			log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u\n",
841 			    s, __func__, th->th_ack, sc->sc_iss);
842 		goto failed;
843 	}
844 	/*
845 	 * If timestamps were present in the SYN and we accepted
846 	 * them in our SYN|ACK we require them to be present from
847 	 * now on.  And vice versa.
848 	 */
849 	if ((sc->sc_flags & SCF_TIMESTAMP) && !(to->to_flags & TOF_TS)) {
850 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
851 			log(LOG_DEBUG, "%s; %s: Timestamp missing\n",
852 			    s, __func__);
853 		goto failed;
854 	}
855 	if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
856 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
857 			log(LOG_DEBUG, "%s; %s: Timestamp not expected\n",
858 			    s, __func__);
859 		goto failed;
860 	}
861 	/*
862 	 * If timestamps were negotiated the reflected timestamp
863 	 * must be equal to what we actually sent in the SYN|ACK.
864 	 */
865 	if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts) {
866 		if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
867 			log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u\n",
868 			    s, __func__, to->to_tsecr, sc->sc_ts);
869 		goto failed;
870 	}
871 
872 	*lsop = syncache_socket(sc, *lsop, m);
873 
874 	if (*lsop == NULL)
875 		tcpstat.tcps_sc_aborted++;
876 	else
877 		tcpstat.tcps_sc_completed++;
878 
879 	if (sc != &scs)
880 		syncache_free(sc);
881 	return (1);
882 failed:
883 	if (sc != NULL && sc != &scs)
884 		syncache_free(sc);
885 	if (s != NULL)
886 		free(s, M_TCPLOG);
887 	*lsop = NULL;
888 	return (0);
889 }
890 
891 /*
892  * Given a LISTEN socket and an inbound SYN request, add
893  * this to the syn cache, and send back a segment:
894  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
895  * to the source.
896  *
897  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
898  * Doing so would require that we hold onto the data and deliver it
899  * to the application.  However, if we are the target of a SYN-flood
900  * DoS attack, an attacker could send data which would eventually
901  * consume all available buffer space if it were ACKed.  By not ACKing
902  * the data, we avoid this DoS scenario.
903  */
904 void
905 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
906     struct inpcb *inp, struct socket **lsop, struct mbuf *m)
907 {
908 	struct tcpcb *tp;
909 	struct socket *so;
910 	struct syncache *sc = NULL;
911 	struct syncache_head *sch;
912 	struct mbuf *ipopts = NULL;
913 	u_int32_t flowtmp;
914 	int win, sb_hiwat, ip_ttl, ip_tos, noopt;
915 #ifdef INET6
916 	int autoflowlabel = 0;
917 #endif
918 #ifdef MAC
919 	struct label *maclabel;
920 #endif
921 	struct syncache scs;
922 
923 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
924 	INP_LOCK_ASSERT(inp);			/* listen socket */
925 
926 	/*
927 	 * Combine all so/tp operations very early to drop the INP lock as
928 	 * soon as possible.
929 	 */
930 	so = *lsop;
931 	tp = sototcpcb(so);
932 
933 #ifdef INET6
934 	if (inc->inc_isipv6 &&
935 	    (inp->in6p_flags & IN6P_AUTOFLOWLABEL))
936 		autoflowlabel = 1;
937 #endif
938 	ip_ttl = inp->inp_ip_ttl;
939 	ip_tos = inp->inp_ip_tos;
940 	win = sbspace(&so->so_rcv);
941 	sb_hiwat = so->so_rcv.sb_hiwat;
942 	noopt = (tp->t_flags & TF_NOOPT);
943 
944 	so = NULL;
945 	tp = NULL;
946 
947 #ifdef MAC
948 	if (mac_init_syncache(&maclabel) != 0) {
949 		INP_UNLOCK(inp);
950 		INP_INFO_WUNLOCK(&tcbinfo);
951 		goto done;
952 	} else
953 		mac_init_syncache_from_inpcb(maclabel, inp);
954 #endif
955 	INP_UNLOCK(inp);
956 	INP_INFO_WUNLOCK(&tcbinfo);
957 
958 	/*
959 	 * Remember the IP options, if any.
960 	 */
961 #ifdef INET6
962 	if (!inc->inc_isipv6)
963 #endif
964 		ipopts = ip_srcroute(m);
965 
966 	/*
967 	 * See if we already have an entry for this connection.
968 	 * If we do, resend the SYN,ACK, and reset the retransmit timer.
969 	 *
970 	 * XXX: should the syncache be re-initialized with the contents
971 	 * of the new SYN here (which may have different options?)
972 	 */
973 	sc = syncache_lookup(inc, &sch);	/* returns locked entry */
974 	SCH_LOCK_ASSERT(sch);
975 	if (sc != NULL) {
976 		tcpstat.tcps_sc_dupsyn++;
977 		if (ipopts) {
978 			/*
979 			 * If we were remembering a previous source route,
980 			 * forget it and use the new one we've been given.
981 			 */
982 			if (sc->sc_ipopts)
983 				(void) m_free(sc->sc_ipopts);
984 			sc->sc_ipopts = ipopts;
985 		}
986 		/*
987 		 * Update timestamp if present.
988 		 */
989 		if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
990 			sc->sc_tsreflect = to->to_tsval;
991 		else
992 			sc->sc_flags &= ~SCF_TIMESTAMP;
993 #ifdef MAC
994 		/*
995 		 * Since we have already unconditionally allocated label
996 		 * storage, free it up.  The syncache entry will already
997 		 * have an initialized label we can use.
998 		 */
999 		mac_destroy_syncache(&maclabel);
1000 		KASSERT(sc->sc_label != NULL,
1001 		    ("%s: label not initialized", __func__));
1002 #endif
1003 		if (syncache_respond(sc) == 0) {
1004 			SYNCACHE_TIMEOUT(sc, sch, 1);
1005 			tcpstat.tcps_sndacks++;
1006 			tcpstat.tcps_sndtotal++;
1007 		}
1008 		SCH_UNLOCK(sch);
1009 		goto done;
1010 	}
1011 
1012 	sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
1013 	if (sc == NULL) {
1014 		/*
1015 		 * The zone allocator couldn't provide more entries.
1016 		 * Treat this as if the cache was full; drop the oldest
1017 		 * entry and insert the new one.
1018 		 */
1019 		tcpstat.tcps_sc_zonefail++;
1020 		if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1021 			syncache_drop(sc, sch);
1022 		sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
1023 		if (sc == NULL) {
1024 			if (tcp_syncookies) {
1025 				bzero(&scs, sizeof(scs));
1026 				sc = &scs;
1027 			} else {
1028 				SCH_UNLOCK(sch);
1029 				if (ipopts)
1030 					(void) m_free(ipopts);
1031 				goto done;
1032 			}
1033 		}
1034 	}
1035 
1036 	/*
1037 	 * Fill in the syncache values.
1038 	 */
1039 #ifdef MAC
1040 	sc->sc_label = maclabel;
1041 #endif
1042 	sc->sc_ipopts = ipopts;
1043 	bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1044 #ifdef INET6
1045 	if (!inc->inc_isipv6)
1046 #endif
1047 	{
1048 		sc->sc_ip_tos = ip_tos;
1049 		sc->sc_ip_ttl = ip_ttl;
1050 	}
1051 
1052 	sc->sc_irs = th->th_seq;
1053 	sc->sc_iss = arc4random();
1054 	sc->sc_flags = 0;
1055 	sc->sc_flowlabel = 0;
1056 
1057 	/*
1058 	 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
1059 	 * win was derived from socket earlier in the function.
1060 	 */
1061 	win = imax(win, 0);
1062 	win = imin(win, TCP_MAXWIN);
1063 	sc->sc_wnd = win;
1064 
1065 	if (tcp_do_rfc1323) {
1066 		/*
1067 		 * A timestamp received in a SYN makes
1068 		 * it ok to send timestamp requests and replies.
1069 		 */
1070 		if (to->to_flags & TOF_TS) {
1071 			sc->sc_tsreflect = to->to_tsval;
1072 			sc->sc_ts = ticks;
1073 			sc->sc_flags |= SCF_TIMESTAMP;
1074 		}
1075 		if (to->to_flags & TOF_SCALE) {
1076 			int wscale = 0;
1077 
1078 			/*
1079 			 * Compute proper scaling value from buffer space.
1080 			 * Leave enough room for the socket buffer to grow
1081 			 * with auto sizing.  This allows us to scale the
1082 			 * receive buffer over a wide range while not losing
1083 			 * any efficiency or fine granularity.
1084 			 *
1085 			 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1086 			 * or <SYN,ACK>) segment itself is never scaled.
1087 			 */
1088 			while (wscale < TCP_MAX_WINSHIFT &&
1089 			    (0x1 << wscale) < tcp_minmss)
1090 				wscale++;
1091 			sc->sc_requested_r_scale = wscale;
1092 			sc->sc_requested_s_scale = to->to_wscale;
1093 			sc->sc_flags |= SCF_WINSCALE;
1094 		}
1095 	}
1096 #ifdef TCP_SIGNATURE
1097 	/*
1098 	 * If listening socket requested TCP digests, and received SYN
1099 	 * contains the option, flag this in the syncache so that
1100 	 * syncache_respond() will do the right thing with the SYN+ACK.
1101 	 * XXX: Currently we always record the option by default and will
1102 	 * attempt to use it in syncache_respond().
1103 	 */
1104 	if (to->to_flags & TOF_SIGNATURE)
1105 		sc->sc_flags |= SCF_SIGNATURE;
1106 #endif
1107 	if (to->to_flags & TOF_SACK)
1108 		sc->sc_flags |= SCF_SACK;
1109 	if (to->to_flags & TOF_MSS)
1110 		sc->sc_peer_mss = to->to_mss;	/* peer mss may be zero */
1111 	if (noopt)
1112 		sc->sc_flags |= SCF_NOOPT;
1113 
1114 	if (tcp_syncookies) {
1115 		syncookie_generate(sch, sc, &flowtmp);
1116 #ifdef INET6
1117 		if (autoflowlabel)
1118 			sc->sc_flowlabel = flowtmp;
1119 #endif
1120 	} else {
1121 #ifdef INET6
1122 		if (autoflowlabel)
1123 			sc->sc_flowlabel =
1124 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1125 #endif
1126 	}
1127 	SCH_UNLOCK(sch);
1128 
1129 	/*
1130 	 * Do a standard 3-way handshake.
1131 	 */
1132 	if (syncache_respond(sc) == 0) {
1133 		if (tcp_syncookies && tcp_syncookiesonly && sc != &scs)
1134 			syncache_free(sc);
1135 		else if (sc != &scs)
1136 			syncache_insert(sc, sch);   /* locks and unlocks sch */
1137 		tcpstat.tcps_sndacks++;
1138 		tcpstat.tcps_sndtotal++;
1139 	} else {
1140 		if (sc != &scs)
1141 			syncache_free(sc);
1142 		tcpstat.tcps_sc_dropped++;
1143 	}
1144 
1145 done:
1146 #ifdef MAC
1147 	if (sc == &scs)
1148 		mac_destroy_syncache(&maclabel);
1149 #endif
1150 	*lsop = NULL;
1151 	m_freem(m);
1152 	return;
1153 }
1154 
1155 static int
1156 syncache_respond(struct syncache *sc)
1157 {
1158 	struct ip *ip = NULL;
1159 	struct mbuf *m;
1160 	struct tcphdr *th;
1161 	int optlen, error;
1162 	u_int16_t hlen, tlen, mssopt;
1163 	struct tcpopt to;
1164 #ifdef INET6
1165 	struct ip6_hdr *ip6 = NULL;
1166 #endif
1167 
1168 	hlen =
1169 #ifdef INET6
1170 	       (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) :
1171 #endif
1172 		sizeof(struct ip);
1173 	tlen = hlen + sizeof(struct tcphdr);
1174 
1175 	/* Determine MSS we advertize to other end of connection. */
1176 	mssopt = tcp_mssopt(&sc->sc_inc);
1177 	if (sc->sc_peer_mss)
1178 		mssopt = max( min(sc->sc_peer_mss, mssopt), tcp_minmss);
1179 
1180 	/* XXX: Assume that the entire packet will fit in a header mbuf. */
1181 	KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1182 	    ("syncache: mbuf too small"));
1183 
1184 	/* Create the IP+TCP header from scratch. */
1185 	m = m_gethdr(M_DONTWAIT, MT_DATA);
1186 	if (m == NULL)
1187 		return (ENOBUFS);
1188 #ifdef MAC
1189 	mac_create_mbuf_from_syncache(sc->sc_label, m);
1190 #endif
1191 	m->m_data += max_linkhdr;
1192 	m->m_len = tlen;
1193 	m->m_pkthdr.len = tlen;
1194 	m->m_pkthdr.rcvif = NULL;
1195 
1196 #ifdef INET6
1197 	if (sc->sc_inc.inc_isipv6) {
1198 		ip6 = mtod(m, struct ip6_hdr *);
1199 		ip6->ip6_vfc = IPV6_VERSION;
1200 		ip6->ip6_nxt = IPPROTO_TCP;
1201 		ip6->ip6_src = sc->sc_inc.inc6_laddr;
1202 		ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1203 		ip6->ip6_plen = htons(tlen - hlen);
1204 		/* ip6_hlim is set after checksum */
1205 		ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1206 		ip6->ip6_flow |= sc->sc_flowlabel;
1207 
1208 		th = (struct tcphdr *)(ip6 + 1);
1209 	} else
1210 #endif
1211 	{
1212 		ip = mtod(m, struct ip *);
1213 		ip->ip_v = IPVERSION;
1214 		ip->ip_hl = sizeof(struct ip) >> 2;
1215 		ip->ip_len = tlen;
1216 		ip->ip_id = 0;
1217 		ip->ip_off = 0;
1218 		ip->ip_sum = 0;
1219 		ip->ip_p = IPPROTO_TCP;
1220 		ip->ip_src = sc->sc_inc.inc_laddr;
1221 		ip->ip_dst = sc->sc_inc.inc_faddr;
1222 		ip->ip_ttl = sc->sc_ip_ttl;
1223 		ip->ip_tos = sc->sc_ip_tos;
1224 
1225 		/*
1226 		 * See if we should do MTU discovery.  Route lookups are
1227 		 * expensive, so we will only unset the DF bit if:
1228 		 *
1229 		 *	1) path_mtu_discovery is disabled
1230 		 *	2) the SCF_UNREACH flag has been set
1231 		 */
1232 		if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1233 		       ip->ip_off |= IP_DF;
1234 
1235 		th = (struct tcphdr *)(ip + 1);
1236 	}
1237 	th->th_sport = sc->sc_inc.inc_lport;
1238 	th->th_dport = sc->sc_inc.inc_fport;
1239 
1240 	th->th_seq = htonl(sc->sc_iss);
1241 	th->th_ack = htonl(sc->sc_irs + 1);
1242 	th->th_off = sizeof(struct tcphdr) >> 2;
1243 	th->th_x2 = 0;
1244 	th->th_flags = TH_SYN|TH_ACK;
1245 	th->th_win = htons(sc->sc_wnd);
1246 	th->th_urp = 0;
1247 
1248 	/* Tack on the TCP options. */
1249 	if ((sc->sc_flags & SCF_NOOPT) == 0) {
1250 		to.to_flags = 0;
1251 
1252 		to.to_mss = mssopt;
1253 		to.to_flags = TOF_MSS;
1254 		if (sc->sc_flags & SCF_WINSCALE) {
1255 			to.to_wscale = sc->sc_requested_r_scale;
1256 			to.to_flags |= TOF_SCALE;
1257 		}
1258 		if (sc->sc_flags & SCF_TIMESTAMP) {
1259 			/* Virgin timestamp or TCP cookie enhanced one. */
1260 			to.to_tsval = sc->sc_ts;
1261 			to.to_tsecr = sc->sc_tsreflect;
1262 			to.to_flags |= TOF_TS;
1263 		}
1264 		if (sc->sc_flags & SCF_SACK)
1265 			to.to_flags |= TOF_SACKPERM;
1266 #ifdef TCP_SIGNATURE
1267 		if (sc->sc_flags & SCF_SIGNATURE)
1268 			to.to_flags |= TOF_SIGNATURE;
1269 #endif
1270 		optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1271 
1272 #ifdef TCP_SIGNATURE
1273 		tcp_signature_compute(m, sizeof(struct ip), 0, optlen,
1274 		    to.to_signature, IPSEC_DIR_OUTBOUND);
1275 #endif
1276 
1277 		/* Adjust headers by option size. */
1278 		th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1279 		m->m_len += optlen;
1280 		m->m_pkthdr.len += optlen;
1281 #ifdef INET6
1282 		if (sc->sc_inc.inc_isipv6)
1283 			ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1284 		else
1285 #endif
1286 			ip->ip_len += optlen;
1287 	} else
1288 		optlen = 0;
1289 
1290 #ifdef INET6
1291 	if (sc->sc_inc.inc_isipv6) {
1292 		th->th_sum = 0;
1293 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
1294 				       tlen + optlen - hlen);
1295 		ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1296 		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1297 	} else
1298 #endif
1299 	{
1300 		th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1301 		    htons(tlen + optlen - hlen + IPPROTO_TCP));
1302 		m->m_pkthdr.csum_flags = CSUM_TCP;
1303 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1304 		error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1305 	}
1306 	return (error);
1307 }
1308 
1309 /*
1310  * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1311  * receive and to be able to handle SYN floods from bogus source addresses
1312  * (where we will never receive any reply).  SYN floods try to exhaust all
1313  * our memory and available slots in the SYN cache table to cause a denial
1314  * of service to legitimate users of the local host.
1315  *
1316  * The idea of SYN cookies is to encode and include all necessary information
1317  * about the connection setup state within the SYN-ACK we send back and thus
1318  * to get along without keeping any local state until the ACK to the SYN-ACK
1319  * arrives (if ever).  Everything we need to know should be available from
1320  * the information we encoded in the SYN-ACK.
1321  *
1322  * More information about the theory behind SYN cookies and its first
1323  * discussion and specification can be found at:
1324  *  http://cr.yp.to/syncookies.html    (overview)
1325  *  http://cr.yp.to/syncookies/archive (gory details)
1326  *
1327  * This implementation extends the orginal idea and first implementation
1328  * of FreeBSD by using not only the initial sequence number field to store
1329  * information but also the timestamp field if present.  This way we can
1330  * keep track of the entire state we need to know to recreate the session in
1331  * its original form.  Almost all TCP speakers implement RFC1323 timestamps
1332  * these days.  For those that do not we still have to live with the known
1333  * shortcomings of the ISN only SYN cookies.
1334  *
1335  * Cookie layers:
1336  *
1337  * Initial sequence number we send:
1338  * 31|................................|0
1339  *    DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1340  *    D = MD5 Digest (first dword)
1341  *    M = MSS index
1342  *    R = Rotation of secret
1343  *    P = Odd or Even secret
1344  *
1345  * The MD5 Digest is computed with over following parameters:
1346  *  a) randomly rotated secret
1347  *  b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1348  *  c) the received initial sequence number from remote host
1349  *  d) the rotation offset and odd/even bit
1350  *
1351  * Timestamp we send:
1352  * 31|................................|0
1353  *    DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1354  *    D = MD5 Digest (third dword) (only as filler)
1355  *    S = Requested send window scale
1356  *    R = Requested receive window scale
1357  *    A = SACK allowed
1358  *    5 = TCP-MD5 enabled (not implemented yet)
1359  *    XORed with MD5 Digest (forth dword)
1360  *
1361  * The timestamp isn't cryptographically secure and doesn't need to be.
1362  * The double use of the MD5 digest dwords ties it to a specific remote/
1363  * local host/port, remote initial sequence number and our local time
1364  * limited secret.  A received timestamp is reverted (XORed) and then
1365  * the contained MD5 dword is compared to the computed one to ensure the
1366  * timestamp belongs to the SYN-ACK we sent.  The other parameters may
1367  * have been tampered with but this isn't different from supplying bogus
1368  * values in the SYN in the first place.
1369  *
1370  * Some problems with SYN cookies remain however:
1371  * Consider the problem of a recreated (and retransmitted) cookie.  If the
1372  * original SYN was accepted, the connection is established.  The second
1373  * SYN is inflight, and if it arrives with an ISN that falls within the
1374  * receive window, the connection is killed.
1375  *
1376  * Notes:
1377  * A heuristic to determine when to accept syn cookies is not necessary.
1378  * An ACK flood would cause the syncookie verification to be attempted,
1379  * but a SYN flood causes syncookies to be generated.  Both are of equal
1380  * cost, so there's no point in trying to optimize the ACK flood case.
1381  * Also, if you don't process certain ACKs for some reason, then all someone
1382  * would have to do is launch a SYN and ACK flood at the same time, which
1383  * would stop cookie verification and defeat the entire purpose of syncookies.
1384  */
1385 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1386 
1387 static void
1388 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1389     u_int32_t *flowlabel)
1390 {
1391 	MD5_CTX ctx;
1392 	u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1393 	u_int32_t data;
1394 	u_int32_t *secbits;
1395 	u_int off, pmss, mss;
1396 	int i;
1397 
1398 	SCH_LOCK_ASSERT(sch);
1399 
1400 	/* Which of the two secrets to use. */
1401 	secbits = sch->sch_oddeven ?
1402 			sch->sch_secbits_odd : sch->sch_secbits_even;
1403 
1404 	/* Reseed secret if too old. */
1405 	if (sch->sch_reseed < time_uptime) {
1406 		sch->sch_oddeven = sch->sch_oddeven ? 0 : 1;	/* toggle */
1407 		secbits = sch->sch_oddeven ?
1408 				sch->sch_secbits_odd : sch->sch_secbits_even;
1409 		for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1410 			secbits[i] = arc4random();
1411 		sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1412 	}
1413 
1414 	/* Secret rotation offset. */
1415 	off = sc->sc_iss & 0x7;			/* iss was randomized before */
1416 
1417 	/* Maximum segment size calculation. */
1418 	pmss = max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), tcp_minmss);
1419 	for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1420 		if (tcp_sc_msstab[mss] <= pmss)
1421 			break;
1422 
1423 	/* Fold parameters and MD5 digest into the ISN we will send. */
1424 	data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1425 	data |= off << 1;	/* secret offset, derived from iss, 3 bits */
1426 	data |= mss << 4;	/* mss, 3 bits */
1427 
1428 	MD5Init(&ctx);
1429 	MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1430 	    SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1431 	MD5Update(&ctx, secbits, off);
1432 	MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1433 	MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1434 	MD5Update(&ctx, &data, sizeof(data));
1435 	MD5Final((u_int8_t *)&md5_buffer, &ctx);
1436 
1437 	data |= (md5_buffer[0] << 7);
1438 	sc->sc_iss = data;
1439 
1440 #ifdef INET6
1441 	*flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1442 #endif
1443 
1444 	/* Additional parameters are stored in the timestamp if present. */
1445 	if (sc->sc_flags & SCF_TIMESTAMP) {
1446 		data =  ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1447 		data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1448 		data |= sc->sc_requested_s_scale << 2;  /* SWIN scale, 4 bits */
1449 		data |= sc->sc_requested_r_scale << 6;  /* RWIN scale, 4 bits */
1450 		data |= md5_buffer[2] << 10;		/* more digest bits */
1451 		data ^= md5_buffer[3];
1452 		sc->sc_ts = data;
1453 		sc->sc_tsoff = data - ticks;		/* after XOR */
1454 	}
1455 
1456 	return;
1457 }
1458 
1459 static struct syncache *
1460 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1461     struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1462     struct socket *so)
1463 {
1464 	MD5_CTX ctx;
1465 	u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1466 	u_int32_t data = 0;
1467 	u_int32_t *secbits;
1468 	tcp_seq ack, seq;
1469 	int off, mss, wnd, flags;
1470 
1471 	SCH_LOCK_ASSERT(sch);
1472 
1473 	/*
1474 	 * Pull information out of SYN-ACK/ACK and
1475 	 * revert sequence number advances.
1476 	 */
1477 	ack = th->th_ack - 1;
1478 	seq = th->th_seq - 1;
1479 	off = (ack >> 1) & 0x7;
1480 	mss = (ack >> 4) & 0x7;
1481 	flags = ack & 0x7f;
1482 
1483 	/* Which of the two secrets to use. */
1484 	secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1485 
1486 	/*
1487 	 * The secret wasn't updated for the lifetime of a syncookie,
1488 	 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1489 	 */
1490 	if (sch->sch_reseed < time_uptime) {
1491 		return (NULL);
1492 	}
1493 
1494 	/* Recompute the digest so we can compare it. */
1495 	MD5Init(&ctx);
1496 	MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1497 	    SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1498 	MD5Update(&ctx, secbits, off);
1499 	MD5Update(&ctx, inc, sizeof(*inc));
1500 	MD5Update(&ctx, &seq, sizeof(seq));
1501 	MD5Update(&ctx, &flags, sizeof(flags));
1502 	MD5Final((u_int8_t *)&md5_buffer, &ctx);
1503 
1504 	/* Does the digest part of or ACK'ed ISS match? */
1505 	if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1506 		return (NULL);
1507 
1508 	/* Does the digest part of our reflected timestamp match? */
1509 	if (to->to_flags & TOF_TS) {
1510 		data = md5_buffer[3] ^ to->to_tsecr;
1511 		if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1512 			return (NULL);
1513 	}
1514 
1515 	/* Fill in the syncache values. */
1516 	bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1517 	sc->sc_ipopts = NULL;
1518 
1519 	sc->sc_irs = seq;
1520 	sc->sc_iss = ack;
1521 
1522 #ifdef INET6
1523 	if (inc->inc_isipv6) {
1524 		if (sotoinpcb(so)->in6p_flags & IN6P_AUTOFLOWLABEL)
1525 			sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1526 	} else
1527 #endif
1528 	{
1529 		sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1530 		sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1531 	}
1532 
1533 	/* Additional parameters that were encoded in the timestamp. */
1534 	if (data) {
1535 		sc->sc_flags |= SCF_TIMESTAMP;
1536 		sc->sc_tsreflect = to->to_tsval;
1537 		sc->sc_ts = to->to_tsecr;
1538 		sc->sc_tsoff = to->to_tsecr - ticks;
1539 		sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1540 		sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1541 		sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1542 		    TCP_MAX_WINSHIFT);
1543 		sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1544 		    TCP_MAX_WINSHIFT);
1545 		if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1546 			sc->sc_flags |= SCF_WINSCALE;
1547 	} else
1548 		sc->sc_flags |= SCF_NOOPT;
1549 
1550 	wnd = sbspace(&so->so_rcv);
1551 	wnd = imax(wnd, 0);
1552 	wnd = imin(wnd, TCP_MAXWIN);
1553 	sc->sc_wnd = wnd;
1554 
1555 	sc->sc_rxmits = 0;
1556 	sc->sc_peer_mss = tcp_sc_msstab[mss];
1557 
1558 	return (sc);
1559 }
1560