xref: /freebsd/sys/netinet/tcp_syncache.c (revision 09e8dea79366f1e5b3a73e8a271b26e4b6bf2e6a)
1 /*-
2  * Copyright (c) 2001 Networks Associates Technology, Inc.
3  * All rights reserved.
4  *
5  * This software was developed for the FreeBSD Project by Jonathan Lemon
6  * and NAI Labs, the Security Research Division of Network Associates, Inc.
7  * under 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  * 3. The name of the author may not be used to endorse or promote
19  *    products derived from this software without specific prior written
20  *    permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * $FreeBSD$
35  */
36 
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/sysctl.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/md5.h>
47 #include <sys/proc.h>		/* for proc0 declaration */
48 #include <sys/random.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 
52 #include <net/if.h>
53 #include <net/route.h>
54 
55 #include <netinet/in.h>
56 #include <netinet/in_systm.h>
57 #include <netinet/ip.h>
58 #include <netinet/in_var.h>
59 #include <netinet/in_pcb.h>
60 #include <netinet/ip_var.h>
61 #ifdef INET6
62 #include <netinet/ip6.h>
63 #include <netinet/icmp6.h>
64 #include <netinet6/nd6.h>
65 #include <netinet6/ip6_var.h>
66 #include <netinet6/in6_pcb.h>
67 #endif
68 #include <netinet/tcp.h>
69 #include <netinet/tcp_fsm.h>
70 #include <netinet/tcp_seq.h>
71 #include <netinet/tcp_timer.h>
72 #include <netinet/tcp_var.h>
73 #ifdef INET6
74 #include <netinet6/tcp6_var.h>
75 #endif
76 
77 #ifdef IPSEC
78 #include <netinet6/ipsec.h>
79 #ifdef INET6
80 #include <netinet6/ipsec6.h>
81 #endif
82 #include <netkey/key.h>
83 #endif /*IPSEC*/
84 
85 #include <machine/in_cksum.h>
86 #include <vm/uma.h>
87 
88 static int tcp_syncookies = 1;
89 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
90     &tcp_syncookies, 0,
91     "Use TCP SYN cookies if the syncache overflows");
92 
93 static void	 syncache_drop(struct syncache *, struct syncache_head *);
94 static void	 syncache_free(struct syncache *);
95 static void	 syncache_insert(struct syncache *, struct syncache_head *);
96 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
97 static int	 syncache_respond(struct syncache *, struct mbuf *);
98 static struct 	 socket *syncache_socket(struct syncache *, struct socket *,
99 		    struct mbuf *m);
100 static void	 syncache_timer(void *);
101 static u_int32_t syncookie_generate(struct syncache *);
102 static struct syncache *syncookie_lookup(struct in_conninfo *,
103 		    struct tcphdr *, struct socket *);
104 
105 /*
106  * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
107  * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds,
108  * the odds are that the user has given up attempting to connect by then.
109  */
110 #define SYNCACHE_MAXREXMTS		3
111 
112 /* Arbitrary values */
113 #define TCP_SYNCACHE_HASHSIZE		512
114 #define TCP_SYNCACHE_BUCKETLIMIT	30
115 
116 struct tcp_syncache {
117 	struct	syncache_head *hashbase;
118 	uma_zone_t zone;
119 	u_int	hashsize;
120 	u_int	hashmask;
121 	u_int	bucket_limit;
122 	u_int	cache_count;
123 	u_int	cache_limit;
124 	u_int	rexmt_limit;
125 	u_int	hash_secret;
126 	u_int	next_reseed;
127 	TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1];
128 	struct	callout tt_timerq[SYNCACHE_MAXREXMTS + 1];
129 };
130 static struct tcp_syncache tcp_syncache;
131 
132 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
133 
134 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD,
135      &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
136 
137 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD,
138      &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
139 
140 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
141      &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
142 
143 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD,
144      &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
145 
146 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
147      &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
148 
149 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
150 
151 #define SYNCACHE_HASH(inc, mask) 					\
152 	((tcp_syncache.hash_secret ^					\
153 	  (inc)->inc_faddr.s_addr ^					\
154 	  ((inc)->inc_faddr.s_addr >> 16) ^ 				\
155 	  (inc)->inc_fport ^ (inc)->inc_lport) & mask)
156 
157 #define SYNCACHE_HASH6(inc, mask) 					\
158 	((tcp_syncache.hash_secret ^					\
159 	  (inc)->inc6_faddr.s6_addr32[0] ^ 				\
160 	  (inc)->inc6_faddr.s6_addr32[3] ^ 				\
161 	  (inc)->inc_fport ^ (inc)->inc_lport) & mask)
162 
163 #define ENDPTS_EQ(a, b) (						\
164 	(a)->ie_fport == (b)->ie_fport &&				\
165 	(a)->ie_lport == (b)->ie_lport &&				\
166 	(a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr &&			\
167 	(a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr			\
168 )
169 
170 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
171 
172 #define SYNCACHE_TIMEOUT(sc, slot) do {					\
173 	sc->sc_rxtslot = slot;						\
174 	sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[slot];	\
175 	TAILQ_INSERT_TAIL(&tcp_syncache.timerq[slot], sc, sc_timerq);	\
176 	if (!callout_active(&tcp_syncache.tt_timerq[slot]))		\
177 		callout_reset(&tcp_syncache.tt_timerq[slot],		\
178 		    TCPTV_RTOBASE * tcp_backoff[slot],			\
179 		    syncache_timer, (void *)((intptr_t)slot));		\
180 } while (0)
181 
182 static void
183 syncache_free(struct syncache *sc)
184 {
185 	struct rtentry *rt;
186 
187 	if (sc->sc_ipopts)
188 		(void) m_free(sc->sc_ipopts);
189 #ifdef INET6
190 	if (sc->sc_inc.inc_isipv6)
191 		rt = sc->sc_route6.ro_rt;
192 	else
193 #endif
194 		rt = sc->sc_route.ro_rt;
195 	if (rt != NULL) {
196 		/*
197 		 * If this is the only reference to a protocol cloned
198 		 * route, remove it immediately.
199 		 */
200 		if (rt->rt_flags & RTF_WASCLONED &&
201 		    (sc->sc_flags & SCF_KEEPROUTE) == 0 &&
202 		    rt->rt_refcnt == 1)
203 			rtrequest(RTM_DELETE, rt_key(rt),
204 			    rt->rt_gateway, rt_mask(rt),
205 			    rt->rt_flags, NULL);
206 		RTFREE(rt);
207 	}
208 	uma_zfree(tcp_syncache.zone, sc);
209 }
210 
211 void
212 syncache_init(void)
213 {
214 	int i;
215 
216 	tcp_syncache.cache_count = 0;
217 	tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
218 	tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
219 	tcp_syncache.cache_limit =
220 	    tcp_syncache.hashsize * tcp_syncache.bucket_limit;
221 	tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
222 	tcp_syncache.next_reseed = 0;
223 	tcp_syncache.hash_secret = arc4random();
224 
225         TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
226 	    &tcp_syncache.hashsize);
227         TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
228 	    &tcp_syncache.cache_limit);
229         TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
230 	    &tcp_syncache.bucket_limit);
231 	if (!powerof2(tcp_syncache.hashsize)) {
232                 printf("WARNING: syncache hash size is not a power of 2.\n");
233 		tcp_syncache.hashsize = 512;	/* safe default */
234         }
235 	tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
236 
237 	/* Allocate the hash table. */
238 	MALLOC(tcp_syncache.hashbase, struct syncache_head *,
239 	    tcp_syncache.hashsize * sizeof(struct syncache_head),
240 	    M_SYNCACHE, M_WAITOK);
241 
242 	/* Initialize the hash buckets. */
243 	for (i = 0; i < tcp_syncache.hashsize; i++) {
244 		TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket);
245 		tcp_syncache.hashbase[i].sch_length = 0;
246 	}
247 
248 	/* Initialize the timer queues. */
249 	for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) {
250 		TAILQ_INIT(&tcp_syncache.timerq[i]);
251 		callout_init(&tcp_syncache.tt_timerq[i], 0);
252 	}
253 
254 	/*
255 	 * Allocate the syncache entries.  Allow the zone to allocate one
256 	 * more entry than cache limit, so a new entry can bump out an
257 	 * older one.
258 	 */
259 	tcp_syncache.cache_limit -= 1;
260 	tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
261 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
262 	uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit);
263 }
264 
265 static void
266 syncache_insert(sc, sch)
267 	struct syncache *sc;
268 	struct syncache_head *sch;
269 {
270 	struct syncache *sc2;
271 	int s, i;
272 
273 	/*
274 	 * Make sure that we don't overflow the per-bucket
275 	 * limit or the total cache size limit.
276 	 */
277 	s = splnet();
278 	if (sch->sch_length >= tcp_syncache.bucket_limit) {
279 		/*
280 		 * The bucket is full, toss the oldest element.
281 		 */
282 		sc2 = TAILQ_FIRST(&sch->sch_bucket);
283 		sc2->sc_tp->ts_recent = ticks;
284 		syncache_drop(sc2, sch);
285 		tcpstat.tcps_sc_bucketoverflow++;
286 	} else if (tcp_syncache.cache_count >= tcp_syncache.cache_limit) {
287 		/*
288 		 * The cache is full.  Toss the oldest entry in the
289 		 * entire cache.  This is the front entry in the
290 		 * first non-empty timer queue with the largest
291 		 * timeout value.
292 		 */
293 		for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
294 			sc2 = TAILQ_FIRST(&tcp_syncache.timerq[i]);
295 			if (sc2 != NULL)
296 				break;
297 		}
298 		sc2->sc_tp->ts_recent = ticks;
299 		syncache_drop(sc2, NULL);
300 		tcpstat.tcps_sc_cacheoverflow++;
301 	}
302 
303 	/* Initialize the entry's timer. */
304 	SYNCACHE_TIMEOUT(sc, 0);
305 
306 	/* Put it into the bucket. */
307 	TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash);
308 	sch->sch_length++;
309 	tcp_syncache.cache_count++;
310 	tcpstat.tcps_sc_added++;
311 	splx(s);
312 }
313 
314 static void
315 syncache_drop(sc, sch)
316 	struct syncache *sc;
317 	struct syncache_head *sch;
318 {
319 	int s;
320 
321 	if (sch == NULL) {
322 #ifdef INET6
323 		if (sc->sc_inc.inc_isipv6) {
324 			sch = &tcp_syncache.hashbase[
325 			    SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)];
326 		} else
327 #endif
328 		{
329 			sch = &tcp_syncache.hashbase[
330 			    SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)];
331 		}
332 	}
333 
334 	s = splnet();
335 
336 	TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
337 	sch->sch_length--;
338 	tcp_syncache.cache_count--;
339 
340 	TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], sc, sc_timerq);
341 	if (TAILQ_EMPTY(&tcp_syncache.timerq[sc->sc_rxtslot]))
342 		callout_stop(&tcp_syncache.tt_timerq[sc->sc_rxtslot]);
343 	splx(s);
344 
345 	syncache_free(sc);
346 }
347 
348 /*
349  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
350  * If we have retransmitted an entry the maximum number of times, expire it.
351  */
352 static void
353 syncache_timer(xslot)
354 	void *xslot;
355 {
356 	intptr_t slot = (intptr_t)xslot;
357 	struct syncache *sc, *nsc;
358 	struct inpcb *inp;
359 	int s;
360 
361 	s = splnet();
362         if (callout_pending(&tcp_syncache.tt_timerq[slot]) ||
363             !callout_active(&tcp_syncache.tt_timerq[slot])) {
364                 splx(s);
365                 return;
366         }
367         callout_deactivate(&tcp_syncache.tt_timerq[slot]);
368 
369         nsc = TAILQ_FIRST(&tcp_syncache.timerq[slot]);
370 	INP_INFO_RLOCK(&tcbinfo);
371 	while (nsc != NULL) {
372 		if (ticks < nsc->sc_rxttime)
373 			break;
374 		sc = nsc;
375 		nsc = TAILQ_NEXT(sc, sc_timerq);
376 		inp = sc->sc_tp->t_inpcb;
377 		INP_LOCK(inp);
378 		if (slot == SYNCACHE_MAXREXMTS ||
379 		    slot >= tcp_syncache.rexmt_limit ||
380 		    inp->inp_gencnt != sc->sc_inp_gencnt) {
381 			syncache_drop(sc, NULL);
382 			tcpstat.tcps_sc_stale++;
383 			INP_UNLOCK(inp);
384 			continue;
385 		}
386 		(void) syncache_respond(sc, NULL);
387 		INP_UNLOCK(inp);
388 		tcpstat.tcps_sc_retransmitted++;
389 		TAILQ_REMOVE(&tcp_syncache.timerq[slot], sc, sc_timerq);
390 		SYNCACHE_TIMEOUT(sc, slot + 1);
391 	}
392 	INP_INFO_RUNLOCK(&tcbinfo);
393 	if (nsc != NULL)
394 		callout_reset(&tcp_syncache.tt_timerq[slot],
395 		    nsc->sc_rxttime - ticks, syncache_timer, (void *)(slot));
396 	splx(s);
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 	int s;
410 
411 #ifdef INET6
412 	if (inc->inc_isipv6) {
413 		sch = &tcp_syncache.hashbase[
414 		    SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
415 		*schp = sch;
416 		s = splnet();
417 		TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
418 			if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) {
419 				splx(s);
420 				return (sc);
421 			}
422 		}
423 		splx(s);
424 	} else
425 #endif
426 	{
427 		sch = &tcp_syncache.hashbase[
428 		    SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
429 		*schp = sch;
430 		s = splnet();
431 		TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
432 #ifdef INET6
433 			if (sc->sc_inc.inc_isipv6)
434 				continue;
435 #endif
436 			if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) {
437 				splx(s);
438 				return (sc);
439 			}
440 		}
441 		splx(s);
442 	}
443 	return (NULL);
444 }
445 
446 /*
447  * This function is called when we get a RST for a
448  * non-existent connection, so that we can see if the
449  * connection is in the syn cache.  If it is, zap it.
450  */
451 void
452 syncache_chkrst(inc, th)
453 	struct in_conninfo *inc;
454 	struct tcphdr *th;
455 {
456 	struct syncache *sc;
457 	struct syncache_head *sch;
458 
459 	sc = syncache_lookup(inc, &sch);
460 	if (sc == NULL)
461 		return;
462 	/*
463 	 * If the RST bit is set, check the sequence number to see
464 	 * if this is a valid reset segment.
465 	 * RFC 793 page 37:
466 	 *   In all states except SYN-SENT, all reset (RST) segments
467 	 *   are validated by checking their SEQ-fields.  A reset is
468 	 *   valid if its sequence number is in the window.
469 	 *
470 	 *   The sequence number in the reset segment is normally an
471 	 *   echo of our outgoing acknowlegement numbers, but some hosts
472 	 *   send a reset with the sequence number at the rightmost edge
473 	 *   of our receive window, and we have to handle this case.
474 	 */
475 	if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
476 	    SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
477 		syncache_drop(sc, sch);
478 		tcpstat.tcps_sc_reset++;
479 	}
480 }
481 
482 void
483 syncache_badack(inc)
484 	struct in_conninfo *inc;
485 {
486 	struct syncache *sc;
487 	struct syncache_head *sch;
488 
489 	sc = syncache_lookup(inc, &sch);
490 	if (sc != NULL) {
491 		syncache_drop(sc, sch);
492 		tcpstat.tcps_sc_badack++;
493 	}
494 }
495 
496 void
497 syncache_unreach(inc, th)
498 	struct in_conninfo *inc;
499 	struct tcphdr *th;
500 {
501 	struct syncache *sc;
502 	struct syncache_head *sch;
503 
504 	/* we are called at splnet() here */
505 	sc = syncache_lookup(inc, &sch);
506 	if (sc == NULL)
507 		return;
508 
509 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
510 	if (ntohl(th->th_seq) != sc->sc_iss)
511 		return;
512 
513 	/*
514 	 * If we've rertransmitted 3 times and this is our second error,
515 	 * we remove the entry.  Otherwise, we allow it to continue on.
516 	 * This prevents us from incorrectly nuking an entry during a
517 	 * spurious network outage.
518 	 *
519 	 * See tcp_notify().
520 	 */
521 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) {
522 		sc->sc_flags |= SCF_UNREACH;
523 		return;
524 	}
525 	syncache_drop(sc, sch);
526 	tcpstat.tcps_sc_unreach++;
527 }
528 
529 /*
530  * Build a new TCP socket structure from a syncache entry.
531  */
532 static struct socket *
533 syncache_socket(sc, lso, m)
534 	struct syncache *sc;
535 	struct socket *lso;
536 	struct mbuf *m;
537 {
538 	struct inpcb *inp = NULL;
539 	struct socket *so;
540 	struct tcpcb *tp;
541 
542 	/*
543 	 * Ok, create the full blown connection, and set things up
544 	 * as they would have been set up if we had created the
545 	 * connection when the SYN arrived.  If we can't create
546 	 * the connection, abort it.
547 	 */
548 	so = sonewconn(lso, SS_ISCONNECTED);
549 	if (so == NULL) {
550 		/*
551 		 * Drop the connection; we will send a RST if the peer
552 		 * retransmits the ACK,
553 		 */
554 		tcpstat.tcps_listendrop++;
555 		goto abort;
556 	}
557 
558 	inp = sotoinpcb(so);
559 
560 	/*
561 	 * Insert new socket into hash list.
562 	 */
563 	inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
564 #ifdef INET6
565 	if (sc->sc_inc.inc_isipv6) {
566 		inp->in6p_laddr = sc->sc_inc.inc6_laddr;
567 	} else {
568 		inp->inp_vflag &= ~INP_IPV6;
569 		inp->inp_vflag |= INP_IPV4;
570 #endif
571 		inp->inp_laddr = sc->sc_inc.inc_laddr;
572 #ifdef INET6
573 	}
574 #endif
575 	inp->inp_lport = sc->sc_inc.inc_lport;
576 	if (in_pcbinshash(inp) != 0) {
577 		/*
578 		 * Undo the assignments above if we failed to
579 		 * put the PCB on the hash lists.
580 		 */
581 #ifdef INET6
582 		if (sc->sc_inc.inc_isipv6)
583 			inp->in6p_laddr = in6addr_any;
584        		else
585 #endif
586 			inp->inp_laddr.s_addr = INADDR_ANY;
587 		inp->inp_lport = 0;
588 		goto abort;
589 	}
590 #ifdef IPSEC
591 	/* copy old policy into new socket's */
592 	if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
593 		printf("syncache_expand: could not copy policy\n");
594 #endif
595 #ifdef INET6
596 	if (sc->sc_inc.inc_isipv6) {
597 		struct inpcb *oinp = sotoinpcb(lso);
598 		struct in6_addr laddr6;
599 		struct sockaddr_in6 *sin6;
600 		/*
601 		 * Inherit socket options from the listening socket.
602 		 * Note that in6p_inputopts are not (and should not be)
603 		 * copied, since it stores previously received options and is
604 		 * used to detect if each new option is different than the
605 		 * previous one and hence should be passed to a user.
606                  * If we copied in6p_inputopts, a user would not be able to
607 		 * receive options just after calling the accept system call.
608 		 */
609 		inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
610 		if (oinp->in6p_outputopts)
611 			inp->in6p_outputopts =
612 			    ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
613 		inp->in6p_route = sc->sc_route6;
614 		sc->sc_route6.ro_rt = NULL;
615 
616 		MALLOC(sin6, struct sockaddr_in6 *, sizeof *sin6,
617 		    M_SONAME, M_NOWAIT | M_ZERO);
618 		if (sin6 == NULL)
619 			goto abort;
620 		sin6->sin6_family = AF_INET6;
621 		sin6->sin6_len = sizeof(*sin6);
622 		sin6->sin6_addr = sc->sc_inc.inc6_faddr;
623 		sin6->sin6_port = sc->sc_inc.inc_fport;
624 		laddr6 = inp->in6p_laddr;
625 		if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
626 			inp->in6p_laddr = sc->sc_inc.inc6_laddr;
627 		if (in6_pcbconnect(inp, (struct sockaddr *)sin6, &thread0)) {
628 			inp->in6p_laddr = laddr6;
629 			FREE(sin6, M_SONAME);
630 			goto abort;
631 		}
632 		FREE(sin6, M_SONAME);
633 	} else
634 #endif
635 	{
636 		struct in_addr laddr;
637 		struct sockaddr_in *sin;
638 
639 		inp->inp_options = ip_srcroute();
640 		if (inp->inp_options == NULL) {
641 			inp->inp_options = sc->sc_ipopts;
642 			sc->sc_ipopts = NULL;
643 		}
644 		inp->inp_route = sc->sc_route;
645 		sc->sc_route.ro_rt = NULL;
646 
647 		MALLOC(sin, struct sockaddr_in *, sizeof *sin,
648 		    M_SONAME, M_NOWAIT | M_ZERO);
649 		if (sin == NULL)
650 			goto abort;
651 		sin->sin_family = AF_INET;
652 		sin->sin_len = sizeof(*sin);
653 		sin->sin_addr = sc->sc_inc.inc_faddr;
654 		sin->sin_port = sc->sc_inc.inc_fport;
655 		bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero));
656 		laddr = inp->inp_laddr;
657 		if (inp->inp_laddr.s_addr == INADDR_ANY)
658 			inp->inp_laddr = sc->sc_inc.inc_laddr;
659 		if (in_pcbconnect(inp, (struct sockaddr *)sin, &thread0)) {
660 			inp->inp_laddr = laddr;
661 			FREE(sin, M_SONAME);
662 			goto abort;
663 		}
664 		FREE(sin, M_SONAME);
665 	}
666 
667 	tp = intotcpcb(inp);
668 	tp->t_state = TCPS_SYN_RECEIVED;
669 	tp->iss = sc->sc_iss;
670 	tp->irs = sc->sc_irs;
671 	tcp_rcvseqinit(tp);
672 	tcp_sendseqinit(tp);
673 	tp->snd_wl1 = sc->sc_irs;
674 	tp->rcv_up = sc->sc_irs + 1;
675 	tp->rcv_wnd = sc->sc_wnd;
676 	tp->rcv_adv += tp->rcv_wnd;
677 
678 	tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
679 	if (sc->sc_flags & SCF_NOOPT)
680 		tp->t_flags |= TF_NOOPT;
681 	if (sc->sc_flags & SCF_WINSCALE) {
682 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
683 		tp->requested_s_scale = sc->sc_requested_s_scale;
684 		tp->request_r_scale = sc->sc_request_r_scale;
685 	}
686 	if (sc->sc_flags & SCF_TIMESTAMP) {
687 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
688 		tp->ts_recent = sc->sc_tsrecent;
689 		tp->ts_recent_age = ticks;
690 	}
691 	if (sc->sc_flags & SCF_CC) {
692 		/*
693 		 * Initialization of the tcpcb for transaction;
694 		 *   set SND.WND = SEG.WND,
695 		 *   initialize CCsend and CCrecv.
696 		 */
697 		tp->t_flags |= TF_REQ_CC|TF_RCVD_CC;
698 		tp->cc_send = sc->sc_cc_send;
699 		tp->cc_recv = sc->sc_cc_recv;
700 	}
701 
702 	tcp_mss(tp, sc->sc_peer_mss);
703 
704 	/*
705 	 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
706 	 */
707 	if (sc->sc_rxtslot != 0)
708                 tp->snd_cwnd = tp->t_maxseg;
709 	callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
710 
711 	tcpstat.tcps_accepts++;
712 	return (so);
713 
714 abort:
715 	if (so != NULL)
716 		(void) soabort(so);
717 	return (NULL);
718 }
719 
720 /*
721  * This function gets called when we receive an ACK for a
722  * socket in the LISTEN state.  We look up the connection
723  * in the syncache, and if its there, we pull it out of
724  * the cache and turn it into a full-blown connection in
725  * the SYN-RECEIVED state.
726  */
727 int
728 syncache_expand(inc, th, sop, m)
729 	struct in_conninfo *inc;
730 	struct tcphdr *th;
731 	struct socket **sop;
732 	struct mbuf *m;
733 {
734 	struct syncache *sc;
735 	struct syncache_head *sch;
736 	struct socket *so;
737 
738 	sc = syncache_lookup(inc, &sch);
739 	if (sc == NULL) {
740 		/*
741 		 * There is no syncache entry, so see if this ACK is
742 		 * a returning syncookie.  To do this, first:
743 		 *  A. See if this socket has had a syncache entry dropped in
744 		 *     the past.  We don't want to accept a bogus syncookie
745  		 *     if we've never received a SYN.
746 		 *  B. check that the syncookie is valid.  If it is, then
747 		 *     cobble up a fake syncache entry, and return.
748 		 */
749 		if (!tcp_syncookies)
750 			return (0);
751 		sc = syncookie_lookup(inc, th, *sop);
752 		if (sc == NULL)
753 			return (0);
754 		sch = NULL;
755 		tcpstat.tcps_sc_recvcookie++;
756 	}
757 
758 	/*
759 	 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
760 	 */
761 	if (th->th_ack != sc->sc_iss + 1)
762 		return (0);
763 
764 	so = syncache_socket(sc, *sop, m);
765 	if (so == NULL) {
766 #if 0
767 resetandabort:
768 		/* XXXjlemon check this - is this correct? */
769 		(void) tcp_respond(NULL, m, m, th,
770 		    th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK);
771 #endif
772 		m_freem(m);			/* XXX only needed for above */
773 		tcpstat.tcps_sc_aborted++;
774 	} else {
775 		sc->sc_flags |= SCF_KEEPROUTE;
776 		tcpstat.tcps_sc_completed++;
777 	}
778 	if (sch == NULL)
779 		syncache_free(sc);
780 	else
781 		syncache_drop(sc, sch);
782 	*sop = so;
783 	return (1);
784 }
785 
786 /*
787  * Given a LISTEN socket and an inbound SYN request, add
788  * this to the syn cache, and send back a segment:
789  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
790  * to the source.
791  *
792  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
793  * Doing so would require that we hold onto the data and deliver it
794  * to the application.  However, if we are the target of a SYN-flood
795  * DoS attack, an attacker could send data which would eventually
796  * consume all available buffer space if it were ACKed.  By not ACKing
797  * the data, we avoid this DoS scenario.
798  */
799 int
800 syncache_add(inc, to, th, sop, m)
801 	struct in_conninfo *inc;
802 	struct tcpopt *to;
803 	struct tcphdr *th;
804 	struct socket **sop;
805 	struct mbuf *m;
806 {
807 	struct tcpcb *tp;
808 	struct socket *so;
809 	struct syncache *sc = NULL;
810 	struct syncache_head *sch;
811 	struct mbuf *ipopts = NULL;
812 	struct rmxp_tao *taop;
813 	int i, s, win;
814 
815 	so = *sop;
816 	tp = sototcpcb(so);
817 
818 	/*
819 	 * Remember the IP options, if any.
820 	 */
821 #ifdef INET6
822 	if (!inc->inc_isipv6)
823 #endif
824 		ipopts = ip_srcroute();
825 
826 	/*
827 	 * See if we already have an entry for this connection.
828 	 * If we do, resend the SYN,ACK, and reset the retransmit timer.
829 	 *
830 	 * XXX
831 	 * should the syncache be re-initialized with the contents
832 	 * of the new SYN here (which may have different options?)
833 	 */
834 	sc = syncache_lookup(inc, &sch);
835 	if (sc != NULL) {
836 		tcpstat.tcps_sc_dupsyn++;
837 		if (ipopts) {
838 			/*
839 			 * If we were remembering a previous source route,
840 			 * forget it and use the new one we've been given.
841 			 */
842 			if (sc->sc_ipopts)
843 				(void) m_free(sc->sc_ipopts);
844 			sc->sc_ipopts = ipopts;
845 		}
846 		/*
847 		 * Update timestamp if present.
848 		 */
849 		if (sc->sc_flags & SCF_TIMESTAMP)
850 			sc->sc_tsrecent = to->to_tsval;
851 		/*
852 		 * PCB may have changed, pick up new values.
853 		 */
854 		sc->sc_tp = tp;
855 		sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
856 		if (syncache_respond(sc, m) == 0) {
857 		        s = splnet();
858 			TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot],
859 			    sc, sc_timerq);
860 			SYNCACHE_TIMEOUT(sc, sc->sc_rxtslot);
861 		        splx(s);
862 		 	tcpstat.tcps_sndacks++;
863 			tcpstat.tcps_sndtotal++;
864 		}
865 		*sop = NULL;
866 		return (1);
867 	}
868 
869 	sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT);
870 	if (sc == NULL) {
871 		/*
872 		 * The zone allocator couldn't provide more entries.
873 		 * Treat this as if the cache was full; drop the oldest
874 		 * entry and insert the new one.
875 		 */
876 		s = splnet();
877 		for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
878 			sc = TAILQ_FIRST(&tcp_syncache.timerq[i]);
879 			if (sc != NULL)
880 				break;
881 		}
882 		sc->sc_tp->ts_recent = ticks;
883 		syncache_drop(sc, NULL);
884 		splx(s);
885 		tcpstat.tcps_sc_zonefail++;
886 		sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT);
887 		if (sc == NULL) {
888 			if (ipopts)
889 				(void) m_free(ipopts);
890 			return (0);
891 		}
892 	}
893 
894 	/*
895 	 * Fill in the syncache values.
896 	 */
897 	bzero(sc, sizeof(*sc));
898 	sc->sc_tp = tp;
899 	sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
900 	sc->sc_ipopts = ipopts;
901 	sc->sc_inc.inc_fport = inc->inc_fport;
902 	sc->sc_inc.inc_lport = inc->inc_lport;
903 #ifdef INET6
904 	sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
905 	if (inc->inc_isipv6) {
906 		sc->sc_inc.inc6_faddr = inc->inc6_faddr;
907 		sc->sc_inc.inc6_laddr = inc->inc6_laddr;
908 		sc->sc_route6.ro_rt = NULL;
909 	} else
910 #endif
911 	{
912 		sc->sc_inc.inc_faddr = inc->inc_faddr;
913 		sc->sc_inc.inc_laddr = inc->inc_laddr;
914 		sc->sc_route.ro_rt = NULL;
915 	}
916 	sc->sc_irs = th->th_seq;
917 	if (tcp_syncookies)
918 		sc->sc_iss = syncookie_generate(sc);
919 	else
920 		sc->sc_iss = arc4random();
921 
922 	/* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */
923 	win = sbspace(&so->so_rcv);
924 	win = imax(win, 0);
925 	win = imin(win, TCP_MAXWIN);
926 	sc->sc_wnd = win;
927 
928 	sc->sc_flags = 0;
929 	sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0;
930 	if (tcp_do_rfc1323) {
931 		/*
932 		 * A timestamp received in a SYN makes
933 		 * it ok to send timestamp requests and replies.
934 		 */
935 		if (to->to_flags & TOF_TS) {
936 			sc->sc_tsrecent = to->to_tsval;
937 			sc->sc_flags |= SCF_TIMESTAMP;
938 		}
939 		if (to->to_flags & TOF_SCALE) {
940 			int wscale = 0;
941 
942 			/* Compute proper scaling value from buffer space */
943 			while (wscale < TCP_MAX_WINSHIFT &&
944 			    (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat)
945 				wscale++;
946 			sc->sc_request_r_scale = wscale;
947 			sc->sc_requested_s_scale = to->to_requested_s_scale;
948 			sc->sc_flags |= SCF_WINSCALE;
949 		}
950 	}
951 	if (tcp_do_rfc1644) {
952 		/*
953 		 * A CC or CC.new option received in a SYN makes
954 		 * it ok to send CC in subsequent segments.
955 		 */
956 		if (to->to_flags & (TOF_CC|TOF_CCNEW)) {
957 			sc->sc_cc_recv = to->to_cc;
958 			sc->sc_cc_send = CC_INC(tcp_ccgen);
959 			sc->sc_flags |= SCF_CC;
960 		}
961 	}
962 	if (tp->t_flags & TF_NOOPT)
963 		sc->sc_flags = SCF_NOOPT;
964 
965 	/*
966 	 * XXX
967 	 * We have the option here of not doing TAO (even if the segment
968 	 * qualifies) and instead fall back to a normal 3WHS via the syncache.
969 	 * This allows us to apply synflood protection to TAO-qualifying SYNs
970 	 * also. However, there should be a hueristic to determine when to
971 	 * do this, and is not present at the moment.
972 	 */
973 
974 	/*
975 	 * Perform TAO test on incoming CC (SEG.CC) option, if any.
976 	 * - compare SEG.CC against cached CC from the same host, if any.
977 	 * - if SEG.CC > chached value, SYN must be new and is accepted
978 	 *	immediately: save new CC in the cache, mark the socket
979 	 *	connected, enter ESTABLISHED state, turn on flag to
980 	 *	send a SYN in the next segment.
981 	 *	A virtual advertised window is set in rcv_adv to
982 	 *	initialize SWS prevention.  Then enter normal segment
983 	 *	processing: drop SYN, process data and FIN.
984 	 * - otherwise do a normal 3-way handshake.
985 	 */
986 	taop = tcp_gettaocache(&sc->sc_inc);
987 	if ((to->to_flags & TOF_CC) != 0) {
988 		if (((tp->t_flags & TF_NOPUSH) != 0) &&
989 		    sc->sc_flags & SCF_CC &&
990 		    taop != NULL && taop->tao_cc != 0 &&
991 		    CC_GT(to->to_cc, taop->tao_cc)) {
992 			sc->sc_rxtslot = 0;
993 			so = syncache_socket(sc, *sop, m);
994 			if (so != NULL) {
995 				sc->sc_flags |= SCF_KEEPROUTE;
996 				taop->tao_cc = to->to_cc;
997 				*sop = so;
998 			}
999 			syncache_free(sc);
1000 			return (so != NULL);
1001 		}
1002 	} else {
1003 		/*
1004 		 * No CC option, but maybe CC.NEW: invalidate cached value.
1005 		 */
1006 		if (taop != NULL)
1007 			taop->tao_cc = 0;
1008 	}
1009 	/*
1010 	 * TAO test failed or there was no CC option,
1011 	 *    do a standard 3-way handshake.
1012 	 */
1013 	if (syncache_respond(sc, m) == 0) {
1014 		syncache_insert(sc, sch);
1015 		tcpstat.tcps_sndacks++;
1016 		tcpstat.tcps_sndtotal++;
1017 	} else {
1018 		syncache_free(sc);
1019 		tcpstat.tcps_sc_dropped++;
1020 	}
1021 	*sop = NULL;
1022 	return (1);
1023 }
1024 
1025 static int
1026 syncache_respond(sc, m)
1027 	struct syncache *sc;
1028 	struct mbuf *m;
1029 {
1030 	u_int8_t *optp;
1031 	int optlen, error;
1032 	u_int16_t tlen, hlen, mssopt;
1033 	struct ip *ip = NULL;
1034 	struct rtentry *rt;
1035 	struct tcphdr *th;
1036 #ifdef INET6
1037 	struct ip6_hdr *ip6 = NULL;
1038 #endif
1039 
1040 #ifdef INET6
1041 	if (sc->sc_inc.inc_isipv6) {
1042 		rt = tcp_rtlookup6(&sc->sc_inc);
1043 		if (rt != NULL)
1044 			mssopt = rt->rt_ifp->if_mtu -
1045 			     (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1046 		else
1047 			mssopt = tcp_v6mssdflt;
1048 		hlen = sizeof(struct ip6_hdr);
1049 	} else
1050 #endif
1051 	{
1052 		rt = tcp_rtlookup(&sc->sc_inc);
1053 		if (rt != NULL)
1054 			mssopt = rt->rt_ifp->if_mtu -
1055 			     (sizeof(struct ip) + sizeof(struct tcphdr));
1056 		else
1057 			mssopt = tcp_mssdflt;
1058 		hlen = sizeof(struct ip);
1059 	}
1060 
1061 	/* Compute the size of the TCP options. */
1062 	if (sc->sc_flags & SCF_NOOPT) {
1063 		optlen = 0;
1064 	} else {
1065 		optlen = TCPOLEN_MAXSEG +
1066 		    ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) +
1067 		    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) +
1068 		    ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0);
1069 	}
1070 	tlen = hlen + sizeof(struct tcphdr) + optlen;
1071 
1072 	/*
1073 	 * XXX
1074 	 * assume that the entire packet will fit in a header mbuf
1075 	 */
1076 	KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small"));
1077 
1078 	/*
1079 	 * XXX shouldn't this reuse the mbuf if possible ?
1080 	 * Create the IP+TCP header from scratch.
1081 	 */
1082 	if (m)
1083 		m_freem(m);
1084 
1085 	m = m_gethdr(M_DONTWAIT, MT_HEADER);
1086 	if (m == NULL)
1087 		return (ENOBUFS);
1088 	m->m_data += max_linkhdr;
1089 	m->m_len = tlen;
1090 	m->m_pkthdr.len = tlen;
1091 	m->m_pkthdr.rcvif = NULL;
1092 
1093 #ifdef IPSEC
1094 	/* use IPsec policy on listening socket to send SYN,ACK */
1095 	if (ipsec_setsocket(m, sc->sc_tp->t_inpcb->inp_socket) != 0) {
1096 		m_freem(m);
1097 		return (ENOBUFS);
1098 	}
1099 #endif
1100 
1101 #ifdef INET6
1102 	if (sc->sc_inc.inc_isipv6) {
1103 		ip6 = mtod(m, struct ip6_hdr *);
1104 		ip6->ip6_vfc = IPV6_VERSION;
1105 		ip6->ip6_nxt = IPPROTO_TCP;
1106 		ip6->ip6_src = sc->sc_inc.inc6_laddr;
1107 		ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1108 		ip6->ip6_plen = htons(tlen - hlen);
1109 		/* ip6_hlim is set after checksum */
1110 		/* ip6_flow = ??? */
1111 
1112 		th = (struct tcphdr *)(ip6 + 1);
1113 	} else
1114 #endif
1115 	{
1116 		ip = mtod(m, struct ip *);
1117 		ip->ip_v = IPVERSION;
1118 		ip->ip_hl = sizeof(struct ip) >> 2;
1119 		ip->ip_len = tlen;
1120 		ip->ip_id = 0;
1121 		ip->ip_off = 0;
1122 		ip->ip_sum = 0;
1123 		ip->ip_p = IPPROTO_TCP;
1124 		ip->ip_src = sc->sc_inc.inc_laddr;
1125 		ip->ip_dst = sc->sc_inc.inc_faddr;
1126 		ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl;   /* XXX */
1127 		ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos;   /* XXX */
1128 
1129 		/*
1130 		 * See if we should do MTU discovery.  We do it only if the following
1131 		 * are true:
1132 		 *      1) we have a valid route to the destination
1133 		 *      2) the MTU is not locked (if it is, then discovery has been
1134 		 *         disabled)
1135 		 */
1136 		if (path_mtu_discovery
1137 		    && (rt != NULL)
1138 		    && rt->rt_flags & RTF_UP
1139 		    && !(rt->rt_rmx.rmx_locks & RTV_MTU)) {
1140 		       ip->ip_off |= IP_DF;
1141 		}
1142 
1143 		th = (struct tcphdr *)(ip + 1);
1144 	}
1145 	th->th_sport = sc->sc_inc.inc_lport;
1146 	th->th_dport = sc->sc_inc.inc_fport;
1147 
1148 	th->th_seq = htonl(sc->sc_iss);
1149 	th->th_ack = htonl(sc->sc_irs + 1);
1150 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1151 	th->th_x2 = 0;
1152 	th->th_flags = TH_SYN|TH_ACK;
1153 	th->th_win = htons(sc->sc_wnd);
1154 	th->th_urp = 0;
1155 
1156 	/* Tack on the TCP options. */
1157 	if (optlen == 0)
1158 		goto no_options;
1159 	optp = (u_int8_t *)(th + 1);
1160 	*optp++ = TCPOPT_MAXSEG;
1161 	*optp++ = TCPOLEN_MAXSEG;
1162 	*optp++ = (mssopt >> 8) & 0xff;
1163 	*optp++ = mssopt & 0xff;
1164 
1165 	if (sc->sc_flags & SCF_WINSCALE) {
1166 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
1167 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
1168 		    sc->sc_request_r_scale);
1169 		optp += 4;
1170 	}
1171 
1172 	if (sc->sc_flags & SCF_TIMESTAMP) {
1173 		u_int32_t *lp = (u_int32_t *)(optp);
1174 
1175 		/* Form timestamp option as shown in appendix A of RFC 1323. */
1176 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
1177 		*lp++ = htonl(ticks);
1178 		*lp   = htonl(sc->sc_tsrecent);
1179 		optp += TCPOLEN_TSTAMP_APPA;
1180 	}
1181 
1182 	/*
1183          * Send CC and CC.echo if we received CC from our peer.
1184          */
1185         if (sc->sc_flags & SCF_CC) {
1186 		u_int32_t *lp = (u_int32_t *)(optp);
1187 
1188 		*lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC));
1189 		*lp++ = htonl(sc->sc_cc_send);
1190 		*lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO));
1191 		*lp   = htonl(sc->sc_cc_recv);
1192 		optp += TCPOLEN_CC_APPA * 2;
1193 	}
1194 no_options:
1195 
1196 #ifdef INET6
1197 	if (sc->sc_inc.inc_isipv6) {
1198 		struct route_in6 *ro6 = &sc->sc_route6;
1199 
1200 		th->th_sum = 0;
1201 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
1202 		ip6->ip6_hlim = in6_selecthlim(NULL,
1203 		    ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL);
1204 		error = ip6_output(m, NULL, ro6, 0, NULL, NULL);
1205 	} else
1206 #endif
1207 	{
1208         	th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1209 		    htons(tlen - hlen + IPPROTO_TCP));
1210 		m->m_pkthdr.csum_flags = CSUM_TCP;
1211 		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1212 		error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 0, NULL);
1213 	}
1214 	return (error);
1215 }
1216 
1217 /*
1218  * cookie layers:
1219  *
1220  *	|. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .|
1221  *	| peer iss                                                      |
1222  *	| MD5(laddr,faddr,lport,fport,secret)             |. . . . . . .|
1223  *	|                     0                       |(A)|             |
1224  * (A): peer mss index
1225  */
1226 
1227 /*
1228  * The values below are chosen to minimize the size of the tcp_secret
1229  * table, as well as providing roughly a 4 second lifetime for the cookie.
1230  */
1231 
1232 #define SYNCOOKIE_HASHSHIFT	2	/* log2(# of 32bit words from hash) */
1233 #define SYNCOOKIE_WNDBITS	7	/* exposed bits for window indexing */
1234 #define SYNCOOKIE_TIMESHIFT	5	/* scale ticks to window time units */
1235 
1236 #define SYNCOOKIE_HASHMASK	((1 << SYNCOOKIE_HASHSHIFT) - 1)
1237 #define SYNCOOKIE_WNDMASK	((1 << SYNCOOKIE_WNDBITS) - 1)
1238 #define SYNCOOKIE_NSECRETS	(1 << (SYNCOOKIE_WNDBITS - SYNCOOKIE_HASHSHIFT))
1239 #define SYNCOOKIE_TIMEOUT \
1240     (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT))
1241 #define SYNCOOKIE_DATAMASK 	((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK)
1242 
1243 static struct {
1244 	u_int32_t	ts_secbits;
1245 	u_int		ts_expire;
1246 } tcp_secret[SYNCOOKIE_NSECRETS];
1247 
1248 static int tcp_msstab[] = { 0, 536, 1460, 8960 };
1249 
1250 static MD5_CTX syn_ctx;
1251 
1252 #define MD5Add(v)	MD5Update(&syn_ctx, (u_char *)&v, sizeof(v))
1253 
1254 /*
1255  * Consider the problem of a recreated (and retransmitted) cookie.  If the
1256  * original SYN was accepted, the connection is established.  The second
1257  * SYN is inflight, and if it arrives with an ISN that falls within the
1258  * receive window, the connection is killed.
1259  *
1260  * However, since cookies have other problems, this may not be worth
1261  * worrying about.
1262  */
1263 
1264 static u_int32_t
1265 syncookie_generate(struct syncache *sc)
1266 {
1267 	u_int32_t md5_buffer[4];
1268 	u_int32_t data;
1269 	int wnd, idx;
1270 
1271 	wnd = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK;
1272 	idx = wnd >> SYNCOOKIE_HASHSHIFT;
1273 	if (tcp_secret[idx].ts_expire < ticks) {
1274 		tcp_secret[idx].ts_secbits = arc4random();
1275 		tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT;
1276 	}
1277 	for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--)
1278 		if (tcp_msstab[data] <= sc->sc_peer_mss)
1279 			break;
1280 	data = (data << SYNCOOKIE_WNDBITS) | wnd;
1281 	data ^= sc->sc_irs;				/* peer's iss */
1282 	MD5Init(&syn_ctx);
1283 #ifdef INET6
1284 	if (sc->sc_inc.inc_isipv6) {
1285 		MD5Add(sc->sc_inc.inc6_laddr);
1286 		MD5Add(sc->sc_inc.inc6_faddr);
1287 	} else
1288 #endif
1289 	{
1290 		MD5Add(sc->sc_inc.inc_laddr);
1291 		MD5Add(sc->sc_inc.inc_faddr);
1292 	}
1293 	MD5Add(sc->sc_inc.inc_lport);
1294 	MD5Add(sc->sc_inc.inc_fport);
1295 	MD5Add(tcp_secret[idx].ts_secbits);
1296 	MD5Final((u_char *)&md5_buffer, &syn_ctx);
1297 	data ^= (md5_buffer[wnd & SYNCOOKIE_HASHMASK] & ~SYNCOOKIE_WNDMASK);
1298 	return (data);
1299 }
1300 
1301 static struct syncache *
1302 syncookie_lookup(inc, th, so)
1303 	struct in_conninfo *inc;
1304 	struct tcphdr *th;
1305 	struct socket *so;
1306 {
1307 	u_int32_t md5_buffer[4];
1308 	struct syncache *sc;
1309 	u_int32_t data;
1310 	int wnd, idx;
1311 
1312 	data = (th->th_ack - 1) ^ (th->th_seq - 1);	/* remove ISS */
1313 	wnd = data & SYNCOOKIE_WNDMASK;
1314 	idx = wnd >> SYNCOOKIE_HASHSHIFT;
1315 	if (tcp_secret[idx].ts_expire < ticks ||
1316 	    sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks)
1317 		return (NULL);
1318 	MD5Init(&syn_ctx);
1319 #ifdef INET6
1320 	if (inc->inc_isipv6) {
1321 		MD5Add(inc->inc6_laddr);
1322 		MD5Add(inc->inc6_faddr);
1323 	} else
1324 #endif
1325 	{
1326 		MD5Add(inc->inc_laddr);
1327 		MD5Add(inc->inc_faddr);
1328 	}
1329 	MD5Add(inc->inc_lport);
1330 	MD5Add(inc->inc_fport);
1331 	MD5Add(tcp_secret[idx].ts_secbits);
1332 	MD5Final((u_char *)&md5_buffer, &syn_ctx);
1333 	data ^= md5_buffer[wnd & SYNCOOKIE_HASHMASK];
1334 	if ((data & ~SYNCOOKIE_DATAMASK) != 0)
1335 		return (NULL);
1336 	data = data >> SYNCOOKIE_WNDBITS;
1337 
1338 	sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT);
1339 	if (sc == NULL)
1340 		return (NULL);
1341 	/*
1342 	 * Fill in the syncache values.
1343 	 * XXX duplicate code from syncache_add
1344 	 */
1345 	sc->sc_ipopts = NULL;
1346 	sc->sc_inc.inc_fport = inc->inc_fport;
1347 	sc->sc_inc.inc_lport = inc->inc_lport;
1348 #ifdef INET6
1349 	sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1350 	if (inc->inc_isipv6) {
1351 		sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1352 		sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1353 		sc->sc_route6.ro_rt = NULL;
1354 	} else
1355 #endif
1356 	{
1357 		sc->sc_inc.inc_faddr = inc->inc_faddr;
1358 		sc->sc_inc.inc_laddr = inc->inc_laddr;
1359 		sc->sc_route.ro_rt = NULL;
1360 	}
1361 	sc->sc_irs = th->th_seq - 1;
1362 	sc->sc_iss = th->th_ack - 1;
1363 	wnd = sbspace(&so->so_rcv);
1364 	wnd = imax(wnd, 0);
1365 	wnd = imin(wnd, TCP_MAXWIN);
1366 	sc->sc_wnd = wnd;
1367 	sc->sc_flags = 0;
1368 	sc->sc_rxtslot = 0;
1369 	sc->sc_peer_mss = tcp_msstab[data];
1370 	return (sc);
1371 }
1372