2 * Copyright (c) 2001 McAfee, Inc.
3 * Copyright (c) 2006 Andre Oppermann, Internet Business Solutions AG
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.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
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.
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
36 #include "opt_inet6.h"
37 #include "opt_ipsec.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/sysctl.h>
45 #include <sys/mutex.h>
46 #include <sys/malloc.h>
49 #include <sys/proc.h> /* for proc0 declaration */
50 #include <sys/random.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
57 #include <net/route.h>
59 #include <netinet/in.h>
60 #include <netinet/in_systm.h>
61 #include <netinet/ip.h>
62 #include <netinet/in_var.h>
63 #include <netinet/in_pcb.h>
64 #include <netinet/ip_var.h>
65 #include <netinet/ip_options.h>
67 #include <netinet/ip6.h>
68 #include <netinet/icmp6.h>
69 #include <netinet6/nd6.h>
70 #include <netinet6/ip6_var.h>
71 #include <netinet6/in6_pcb.h>
73 #include <netinet/tcp.h>
74 #include <netinet/tcp_fsm.h>
75 #include <netinet/tcp_seq.h>
76 #include <netinet/tcp_timer.h>
77 #include <netinet/tcp_var.h>
79 #include <netinet6/tcp6_var.h>
83 #include <netinet6/ipsec.h>
85 #include <netinet6/ipsec6.h>
90 #include <netipsec/ipsec.h>
92 #include <netipsec/ipsec6.h>
94 #include <netipsec/key.h>
97 #include <machine/in_cksum.h>
99 #include <security/mac/mac_framework.h>
101 static int tcp_syncookies = 1;
102 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
104 "Use TCP SYN cookies if the syncache overflows");
106 static int tcp_syncookiesonly = 0;
107 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW,
108 &tcp_syncookiesonly, 0,
109 "Use only TCP SYN cookies");
111 #define SYNCOOKIE_SECRET_SIZE 8 /* dwords */
112 #define SYNCOOKIE_LIFETIME 16 /* seconds */
115 TAILQ_ENTRY(syncache) sc_hash;
116 struct in_conninfo sc_inc; /* addresses */
117 u_long sc_rxttime; /* retransmit time */
118 u_int16_t sc_rxmits; /* retransmit counter */
120 u_int32_t sc_tsreflect; /* timestamp to reflect */
121 u_int32_t sc_ts; /* our timestamp to send */
122 u_int32_t sc_tsoff; /* ts offset w/ syncookies */
123 u_int32_t sc_flowlabel; /* IPv6 flowlabel */
124 tcp_seq sc_irs; /* seq from peer */
125 tcp_seq sc_iss; /* our ISS */
126 struct mbuf *sc_ipopts; /* source route */
128 u_int16_t sc_peer_mss; /* peer's MSS */
129 u_int16_t sc_wnd; /* advertised window */
130 u_int8_t sc_ip_ttl; /* IPv4 TTL */
131 u_int8_t sc_ip_tos; /* IPv4 TOS */
132 u_int8_t sc_requested_s_scale:4,
133 sc_requested_r_scale:4;
135 #define SCF_NOOPT 0x01 /* no TCP options */
136 #define SCF_WINSCALE 0x02 /* negotiated window scaling */
137 #define SCF_TIMESTAMP 0x04 /* negotiated timestamps */
138 /* MSS is implicit */
139 #define SCF_UNREACH 0x10 /* icmp unreachable received */
140 #define SCF_SIGNATURE 0x20 /* send MD5 digests */
141 #define SCF_SACK 0x80 /* send SACK option */
143 struct label *sc_label; /* MAC label reference */
147 struct syncache_head {
149 TAILQ_HEAD(sch_head, syncache) sch_bucket;
150 struct callout sch_timer;
154 u_int32_t sch_secbits_odd[SYNCOOKIE_SECRET_SIZE];
155 u_int32_t sch_secbits_even[SYNCOOKIE_SECRET_SIZE];
156 u_int sch_reseed; /* time_uptime, seconds */
159 static void syncache_drop(struct syncache *, struct syncache_head *);
160 static void syncache_free(struct syncache *);
161 static void syncache_insert(struct syncache *, struct syncache_head *);
162 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
163 static int syncache_respond(struct syncache *, struct mbuf *);
164 static struct socket *syncache_socket(struct syncache *, struct socket *,
166 static void syncache_timer(void *);
167 static void syncookie_generate(struct syncache_head *, struct syncache *,
169 static struct syncache
170 *syncookie_lookup(struct in_conninfo *, struct syncache_head *,
171 struct syncache *, struct tcpopt *, struct tcphdr *,
175 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
176 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds,
177 * the odds are that the user has given up attempting to connect by then.
179 #define SYNCACHE_MAXREXMTS 3
181 /* Arbitrary values */
182 #define TCP_SYNCACHE_HASHSIZE 512
183 #define TCP_SYNCACHE_BUCKETLIMIT 30
185 struct tcp_syncache {
186 struct syncache_head *hashbase;
191 u_int cache_count; /* XXX: unprotected */
196 static struct tcp_syncache tcp_syncache;
198 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
200 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
201 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
203 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
204 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
206 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
207 &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
209 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
210 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
212 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
213 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
215 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
217 #define SYNCACHE_HASH(inc, mask) \
218 ((tcp_syncache.hash_secret ^ \
219 (inc)->inc_faddr.s_addr ^ \
220 ((inc)->inc_faddr.s_addr >> 16) ^ \
221 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
223 #define SYNCACHE_HASH6(inc, mask) \
224 ((tcp_syncache.hash_secret ^ \
225 (inc)->inc6_faddr.s6_addr32[0] ^ \
226 (inc)->inc6_faddr.s6_addr32[3] ^ \
227 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
229 #define ENDPTS_EQ(a, b) ( \
230 (a)->ie_fport == (b)->ie_fport && \
231 (a)->ie_lport == (b)->ie_lport && \
232 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
233 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
236 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
238 #define SYNCACHE_TIMEOUT(sc, sch, co) do { \
240 (sc)->sc_rxttime = ticks + \
241 TCPTV_RTOBASE * tcp_backoff[(sc)->sc_rxmits - 1]; \
242 if ((sch)->sch_nextc > (sc)->sc_rxttime) \
243 (sch)->sch_nextc = (sc)->sc_rxttime; \
244 if (!TAILQ_EMPTY(&(sch)->sch_bucket) && !(co)) \
245 callout_reset(&(sch)->sch_timer, \
246 (sch)->sch_nextc - ticks, \
247 syncache_timer, (void *)(sch)); \
250 #define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx)
251 #define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx)
252 #define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED)
255 * Requires the syncache entry to be already removed from the bucket list.
258 syncache_free(struct syncache *sc)
261 (void) m_free(sc->sc_ipopts);
263 mac_destroy_syncache(&sc->sc_label);
266 uma_zfree(tcp_syncache.zone, sc);
274 tcp_syncache.cache_count = 0;
275 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
276 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
277 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
278 tcp_syncache.hash_secret = arc4random();
280 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
281 &tcp_syncache.hashsize);
282 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
283 &tcp_syncache.bucket_limit);
284 if (!powerof2(tcp_syncache.hashsize) || tcp_syncache.hashsize == 0) {
285 printf("WARNING: syncache hash size is not a power of 2.\n");
286 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
288 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
291 tcp_syncache.cache_limit =
292 tcp_syncache.hashsize * tcp_syncache.bucket_limit;
293 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
294 &tcp_syncache.cache_limit);
296 /* Allocate the hash table. */
297 MALLOC(tcp_syncache.hashbase, struct syncache_head *,
298 tcp_syncache.hashsize * sizeof(struct syncache_head),
299 M_SYNCACHE, M_WAITOK | M_ZERO);
301 /* Initialize the hash buckets. */
302 for (i = 0; i < tcp_syncache.hashsize; i++) {
303 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket);
304 mtx_init(&tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
306 callout_init_mtx(&tcp_syncache.hashbase[i].sch_timer,
307 &tcp_syncache.hashbase[i].sch_mtx, 0);
308 tcp_syncache.hashbase[i].sch_length = 0;
311 /* Create the syncache entry zone. */
312 tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
313 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
314 uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit);
318 * Inserts a syncache entry into the specified bucket row.
319 * Locks and unlocks the syncache_head autonomously.
322 syncache_insert(struct syncache *sc, struct syncache_head *sch)
324 struct syncache *sc2;
329 * Make sure that we don't overflow the per-bucket limit.
330 * If the bucket is full, toss the oldest element.
332 if (sch->sch_length >= tcp_syncache.bucket_limit) {
333 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
334 ("sch->sch_length incorrect"));
335 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
336 syncache_drop(sc2, sch);
337 tcpstat.tcps_sc_bucketoverflow++;
340 /* Put it into the bucket. */
341 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
344 /* Reinitialize the bucket row's timer. */
345 SYNCACHE_TIMEOUT(sc, sch, 1);
349 tcp_syncache.cache_count++;
350 tcpstat.tcps_sc_added++;
354 * Remove and free entry from syncache bucket row.
355 * Expects locked syncache head.
358 syncache_drop(struct syncache *sc, struct syncache_head *sch)
361 SCH_LOCK_ASSERT(sch);
363 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
367 tcp_syncache.cache_count--;
371 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
372 * If we have retransmitted an entry the maximum number of times, expire it.
373 * One separate timer for each bucket row.
376 syncache_timer(void *xsch)
378 struct syncache_head *sch = (struct syncache_head *)xsch;
379 struct syncache *sc, *nsc;
382 /* NB: syncache_head has already been locked by the callout. */
383 SCH_LOCK_ASSERT(sch);
385 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
387 * We do not check if the listen socket still exists
388 * and accept the case where the listen socket may be
389 * gone by the time we resend the SYN/ACK. We do
390 * not expect this to happens often. If it does,
391 * then the RST will be sent by the time the remote
392 * host does the SYN/ACK->ACK.
394 if (sc->sc_rxttime >= tick) {
395 if (sc->sc_rxttime < sch->sch_nextc)
396 sch->sch_nextc = sc->sc_rxttime;
400 if (sc->sc_rxmits > tcp_syncache.rexmt_limit) {
401 syncache_drop(sc, sch);
402 tcpstat.tcps_sc_stale++;
406 (void) syncache_respond(sc, NULL);
407 tcpstat.tcps_sc_retransmitted++;
408 SYNCACHE_TIMEOUT(sc, sch, 0);
410 if (!TAILQ_EMPTY(&(sch)->sch_bucket))
411 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
412 syncache_timer, (void *)(sch));
416 * Find an entry in the syncache.
417 * Returns always with locked syncache_head plus a matching entry or NULL.
420 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
423 struct syncache_head *sch;
426 if (inc->inc_isipv6) {
427 sch = &tcp_syncache.hashbase[
428 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
433 /* Circle through bucket row to find matching entry. */
434 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
435 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
441 sch = &tcp_syncache.hashbase[
442 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
447 /* Circle through bucket row to find matching entry. */
448 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
450 if (sc->sc_inc.inc_isipv6)
453 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
457 SCH_LOCK_ASSERT(*schp);
458 return (NULL); /* always returns with locked sch */
462 * This function is called when we get a RST for a
463 * non-existent connection, so that we can see if the
464 * connection is in the syn cache. If it is, zap it.
467 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
470 struct syncache_head *sch;
472 sc = syncache_lookup(inc, &sch); /* returns locked sch */
473 SCH_LOCK_ASSERT(sch);
478 * If the RST bit is set, check the sequence number to see
479 * if this is a valid reset segment.
481 * In all states except SYN-SENT, all reset (RST) segments
482 * are validated by checking their SEQ-fields. A reset is
483 * valid if its sequence number is in the window.
485 * The sequence number in the reset segment is normally an
486 * echo of our outgoing acknowlegement numbers, but some hosts
487 * send a reset with the sequence number at the rightmost edge
488 * of our receive window, and we have to handle this case.
490 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
491 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
492 syncache_drop(sc, sch);
493 tcpstat.tcps_sc_reset++;
500 syncache_badack(struct in_conninfo *inc)
503 struct syncache_head *sch;
505 sc = syncache_lookup(inc, &sch); /* returns locked sch */
506 SCH_LOCK_ASSERT(sch);
508 syncache_drop(sc, sch);
509 tcpstat.tcps_sc_badack++;
515 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
518 struct syncache_head *sch;
520 sc = syncache_lookup(inc, &sch); /* returns locked sch */
521 SCH_LOCK_ASSERT(sch);
525 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
526 if (ntohl(th->th_seq) != sc->sc_iss)
530 * If we've rertransmitted 3 times and this is our second error,
531 * we remove the entry. Otherwise, we allow it to continue on.
532 * This prevents us from incorrectly nuking an entry during a
533 * spurious network outage.
537 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
538 sc->sc_flags |= SCF_UNREACH;
541 syncache_drop(sc, sch);
542 tcpstat.tcps_sc_unreach++;
548 * Build a new TCP socket structure from a syncache entry.
550 static struct socket *
551 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
553 struct inpcb *inp = NULL;
558 INP_INFO_WLOCK_ASSERT(&tcbinfo);
561 * Ok, create the full blown connection, and set things up
562 * as they would have been set up if we had created the
563 * connection when the SYN arrived. If we can't create
564 * the connection, abort it.
566 so = sonewconn(lso, SS_ISCONNECTED);
569 * Drop the connection; we will send a RST if the peer
570 * retransmits the ACK,
572 tcpstat.tcps_listendrop++;
577 mac_set_socket_peer_from_mbuf(m, so);
584 /* Insert new socket into PCB hash list. */
585 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
587 if (sc->sc_inc.inc_isipv6) {
588 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
590 inp->inp_vflag &= ~INP_IPV6;
591 inp->inp_vflag |= INP_IPV4;
593 inp->inp_laddr = sc->sc_inc.inc_laddr;
597 inp->inp_lport = sc->sc_inc.inc_lport;
598 if (in_pcbinshash(inp) != 0) {
600 * Undo the assignments above if we failed to
601 * put the PCB on the hash lists.
604 if (sc->sc_inc.inc_isipv6)
605 inp->in6p_laddr = in6addr_any;
608 inp->inp_laddr.s_addr = INADDR_ANY;
613 /* Copy old policy into new socket's. */
614 if (ipsec_copy_pcbpolicy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
615 printf("syncache_socket: could not copy policy\n");
618 /* Copy old policy into new socket's. */
619 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
620 printf("syncache_socket: could not copy policy\n");
623 if (sc->sc_inc.inc_isipv6) {
624 struct inpcb *oinp = sotoinpcb(lso);
625 struct in6_addr laddr6;
626 struct sockaddr_in6 sin6;
628 * Inherit socket options from the listening socket.
629 * Note that in6p_inputopts are not (and should not be)
630 * copied, since it stores previously received options and is
631 * used to detect if each new option is different than the
632 * previous one and hence should be passed to a user.
633 * If we copied in6p_inputopts, a user would not be able to
634 * receive options just after calling the accept system call.
636 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
637 if (oinp->in6p_outputopts)
638 inp->in6p_outputopts =
639 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
641 sin6.sin6_family = AF_INET6;
642 sin6.sin6_len = sizeof(sin6);
643 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
644 sin6.sin6_port = sc->sc_inc.inc_fport;
645 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
646 laddr6 = inp->in6p_laddr;
647 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
648 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
649 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
651 inp->in6p_laddr = laddr6;
654 /* Override flowlabel from in6_pcbconnect. */
655 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK;
656 inp->in6p_flowinfo |= sc->sc_flowlabel;
660 struct in_addr laddr;
661 struct sockaddr_in sin;
663 inp->inp_options = ip_srcroute(m);
664 if (inp->inp_options == NULL) {
665 inp->inp_options = sc->sc_ipopts;
666 sc->sc_ipopts = NULL;
669 sin.sin_family = AF_INET;
670 sin.sin_len = sizeof(sin);
671 sin.sin_addr = sc->sc_inc.inc_faddr;
672 sin.sin_port = sc->sc_inc.inc_fport;
673 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
674 laddr = inp->inp_laddr;
675 if (inp->inp_laddr.s_addr == INADDR_ANY)
676 inp->inp_laddr = sc->sc_inc.inc_laddr;
677 if (in_pcbconnect(inp, (struct sockaddr *)&sin,
679 inp->inp_laddr = laddr;
684 tp->t_state = TCPS_SYN_RECEIVED;
685 tp->iss = sc->sc_iss;
686 tp->irs = sc->sc_irs;
689 tp->snd_wl1 = sc->sc_irs;
690 tp->rcv_up = sc->sc_irs + 1;
691 tp->rcv_wnd = sc->sc_wnd;
692 tp->rcv_adv += tp->rcv_wnd;
694 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
695 if (sc->sc_flags & SCF_NOOPT)
696 tp->t_flags |= TF_NOOPT;
698 if (sc->sc_flags & SCF_WINSCALE) {
699 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
700 tp->snd_scale = sc->sc_requested_s_scale;
701 tp->request_r_scale = sc->sc_requested_r_scale;
703 if (sc->sc_flags & SCF_TIMESTAMP) {
704 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
705 tp->ts_recent = sc->sc_tsreflect;
706 tp->ts_recent_age = ticks;
707 tp->ts_offset = sc->sc_tsoff;
710 if (sc->sc_flags & SCF_SIGNATURE)
711 tp->t_flags |= TF_SIGNATURE;
713 if (sc->sc_flags & SCF_SACK) {
715 tp->t_flags |= TF_SACK_PERMIT;
720 * Set up MSS and get cached values from tcp_hostcache.
721 * This might overwrite some of the defaults we just set.
723 tcp_mss(tp, sc->sc_peer_mss);
726 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
728 if (sc->sc_rxmits > 1)
729 tp->snd_cwnd = tp->t_maxseg;
730 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
734 tcpstat.tcps_accepts++;
746 * This function gets called when we receive an ACK for a
747 * socket in the LISTEN state. We look up the connection
748 * in the syncache, and if its there, we pull it out of
749 * the cache and turn it into a full-blown connection in
750 * the SYN-RECEIVED state.
753 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
754 struct socket **lsop, struct mbuf *m)
757 struct syncache_head *sch;
762 * Global TCP locks are held because we manipulate the PCB lists
763 * and create a new socket.
765 INP_INFO_WLOCK_ASSERT(&tcbinfo);
767 sc = syncache_lookup(inc, &sch); /* returns locked sch */
768 SCH_LOCK_ASSERT(sch);
771 * There is no syncache entry, so see if this ACK is
772 * a returning syncookie. To do this, first:
773 * A. See if this socket has had a syncache entry dropped in
774 * the past. We don't want to accept a bogus syncookie
775 * if we've never received a SYN.
776 * B. check that the syncookie is valid. If it is, then
777 * cobble up a fake syncache entry, and return.
779 if (!tcp_syncookies) {
783 bzero(&scs, sizeof(scs));
784 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
788 tcpstat.tcps_sc_recvcookie++;
790 /* Pull out the entry to unlock the bucket row. */
791 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
793 tcp_syncache.cache_count--;
798 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
800 if (th->th_ack != sc->sc_iss + 1)
803 so = syncache_socket(sc, *lsop, m);
808 /* XXXjlemon check this - is this correct? */
809 (void) tcp_respond(NULL, m, m, th,
810 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK);
812 m_freem(m); /* XXX: only needed for above */
813 tcpstat.tcps_sc_aborted++;
815 syncache_insert(sc, sch); /* try again later */
820 tcpstat.tcps_sc_completed++;
827 if (sc != NULL && sc != &scs)
833 * Given a LISTEN socket and an inbound SYN request, add
834 * this to the syn cache, and send back a segment:
835 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
838 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
839 * Doing so would require that we hold onto the data and deliver it
840 * to the application. However, if we are the target of a SYN-flood
841 * DoS attack, an attacker could send data which would eventually
842 * consume all available buffer space if it were ACKed. By not ACKing
843 * the data, we avoid this DoS scenario.
846 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
847 struct inpcb *inp, struct socket **lsop, struct mbuf *m)
851 struct syncache *sc = NULL;
852 struct syncache_head *sch;
853 struct mbuf *ipopts = NULL;
855 int win, sb_hiwat, ip_ttl, ip_tos, noopt;
857 int autoflowlabel = 0;
860 struct label *maclabel;
864 INP_INFO_WLOCK_ASSERT(&tcbinfo);
865 INP_LOCK_ASSERT(inp); /* listen socket */
868 * Combine all so/tp operations very early to drop the INP lock as
875 if (inc->inc_isipv6 &&
876 (inp->in6p_flags & IN6P_AUTOFLOWLABEL))
879 ip_ttl = inp->inp_ip_ttl;
880 ip_tos = inp->inp_ip_tos;
881 win = sbspace(&so->so_rcv);
882 sb_hiwat = so->so_rcv.sb_hiwat;
883 noopt = (tp->t_flags & TF_NOOPT);
889 if (mac_init_syncache(&maclabel) != 0) {
892 INP_INFO_WUNLOCK(&tcbinfo);
895 mac_init_syncache_from_inpcb(maclabel, inp);
898 INP_INFO_WUNLOCK(&tcbinfo);
901 * Remember the IP options, if any.
904 if (!inc->inc_isipv6)
906 ipopts = ip_srcroute(m);
909 * See if we already have an entry for this connection.
910 * If we do, resend the SYN,ACK, and reset the retransmit timer.
912 * XXX: should the syncache be re-initialized with the contents
913 * of the new SYN here (which may have different options?)
915 sc = syncache_lookup(inc, &sch); /* returns locked entry */
916 SCH_LOCK_ASSERT(sch);
918 tcpstat.tcps_sc_dupsyn++;
921 * If we were remembering a previous source route,
922 * forget it and use the new one we've been given.
925 (void) m_free(sc->sc_ipopts);
926 sc->sc_ipopts = ipopts;
929 * Update timestamp if present.
931 if (sc->sc_flags & SCF_TIMESTAMP)
932 sc->sc_tsreflect = to->to_tsval;
935 * Since we have already unconditionally allocated label
936 * storage, free it up. The syncache entry will already
937 * have an initialized label we can use.
939 mac_destroy_syncache(&maclabel);
940 KASSERT(sc->sc_label != NULL,
941 ("%s: label not initialized", __func__));
943 if (syncache_respond(sc, m) == 0) {
944 SYNCACHE_TIMEOUT(sc, sch, 1);
945 tcpstat.tcps_sndacks++;
946 tcpstat.tcps_sndtotal++;
952 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
955 * The zone allocator couldn't provide more entries.
956 * Treat this as if the cache was full; drop the oldest
957 * entry and insert the new one.
959 tcpstat.tcps_sc_zonefail++;
960 sc = TAILQ_LAST(&sch->sch_bucket, sch_head);
961 syncache_drop(sc, sch);
962 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
964 if (tcp_syncookies) {
965 bzero(&scs, sizeof(scs));
970 (void) m_free(ipopts);
977 * Fill in the syncache values.
980 sc->sc_label = maclabel;
982 sc->sc_ipopts = ipopts;
983 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
985 if (!inc->inc_isipv6)
988 sc->sc_ip_tos = ip_tos;
989 sc->sc_ip_ttl = ip_ttl;
992 sc->sc_irs = th->th_seq;
993 sc->sc_iss = arc4random();
995 sc->sc_flowlabel = 0;
998 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
999 * win was derived from socket earlier in the function.
1002 win = imin(win, TCP_MAXWIN);
1005 if (tcp_do_rfc1323) {
1007 * A timestamp received in a SYN makes
1008 * it ok to send timestamp requests and replies.
1010 if (to->to_flags & TOF_TS) {
1011 sc->sc_tsreflect = to->to_tsval;
1012 sc->sc_flags |= SCF_TIMESTAMP;
1014 if (to->to_flags & TOF_SCALE) {
1018 * Compute proper scaling value from buffer space.
1019 * Leave enough room for the socket buffer to grow
1020 * with auto sizing. This allows us to scale the
1021 * receive buffer over a wide range while not losing
1022 * any efficiency or fine granularity.
1024 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1025 * or <SYN,ACK>) segment itself is never scaled.
1027 while (wscale < TCP_MAX_WINSHIFT &&
1028 (0x1 << wscale) < tcp_minmss)
1030 sc->sc_requested_r_scale = wscale;
1031 sc->sc_requested_s_scale = to->to_wscale;
1032 sc->sc_flags |= SCF_WINSCALE;
1035 #ifdef TCP_SIGNATURE
1037 * If listening socket requested TCP digests, and received SYN
1038 * contains the option, flag this in the syncache so that
1039 * syncache_respond() will do the right thing with the SYN+ACK.
1040 * XXX: Currently we always record the option by default and will
1041 * attempt to use it in syncache_respond().
1043 if (to->to_flags & TOF_SIGNATURE)
1044 sc->sc_flags |= SCF_SIGNATURE;
1046 if (to->to_flags & TOF_SACK)
1047 sc->sc_flags |= SCF_SACK;
1048 if (to->to_flags & TOF_MSS)
1049 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */
1051 sc->sc_flags |= SCF_NOOPT;
1053 if (tcp_syncookies) {
1054 syncookie_generate(sch, sc, &flowtmp);
1057 sc->sc_flowlabel = flowtmp;
1063 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1069 * Do a standard 3-way handshake.
1071 if (syncache_respond(sc, m) == 0) {
1072 if (tcp_syncookies && tcp_syncookiesonly && sc != &scs)
1074 else if (sc != &scs)
1075 syncache_insert(sc, sch); /* locks and unlocks sch */
1078 mac_destroy_syncache(&sc->sc_label);
1080 tcpstat.tcps_sndacks++;
1081 tcpstat.tcps_sndtotal++;
1087 mac_destroy_syncache(&sc->sc_label);
1089 tcpstat.tcps_sc_dropped++;
1098 syncache_respond(struct syncache *sc, struct mbuf *m)
1100 struct ip *ip = NULL;
1103 u_int16_t hlen, tlen, mssopt;
1106 struct ip6_hdr *ip6 = NULL;
1111 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) :
1114 tlen = hlen + sizeof(struct tcphdr);
1116 /* Determine MSS we advertize to other end of connection. */
1117 mssopt = tcp_mssopt(&sc->sc_inc);
1118 if (sc->sc_peer_mss)
1119 mssopt = max( min(sc->sc_peer_mss, mssopt), tcp_minmss);
1121 /* XXX: Assume that the entire packet will fit in a header mbuf. */
1122 KASSERT(max_linkhdr + tlen + MAX_TCPOPTLEN <= MHLEN,
1123 ("syncache: mbuf too small"));
1125 /* Create the IP+TCP header from scratch. */
1129 m = m_gethdr(M_DONTWAIT, MT_DATA);
1133 mac_create_mbuf_from_syncache(sc->sc_label, m);
1135 m->m_data += max_linkhdr;
1137 m->m_pkthdr.len = tlen;
1138 m->m_pkthdr.rcvif = NULL;
1141 if (sc->sc_inc.inc_isipv6) {
1142 ip6 = mtod(m, struct ip6_hdr *);
1143 ip6->ip6_vfc = IPV6_VERSION;
1144 ip6->ip6_nxt = IPPROTO_TCP;
1145 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1146 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1147 ip6->ip6_plen = htons(tlen - hlen);
1148 /* ip6_hlim is set after checksum */
1149 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1150 ip6->ip6_flow |= sc->sc_flowlabel;
1152 th = (struct tcphdr *)(ip6 + 1);
1156 ip = mtod(m, struct ip *);
1157 ip->ip_v = IPVERSION;
1158 ip->ip_hl = sizeof(struct ip) >> 2;
1163 ip->ip_p = IPPROTO_TCP;
1164 ip->ip_src = sc->sc_inc.inc_laddr;
1165 ip->ip_dst = sc->sc_inc.inc_faddr;
1166 ip->ip_ttl = sc->sc_ip_ttl;
1167 ip->ip_tos = sc->sc_ip_tos;
1170 * See if we should do MTU discovery. Route lookups are
1171 * expensive, so we will only unset the DF bit if:
1173 * 1) path_mtu_discovery is disabled
1174 * 2) the SCF_UNREACH flag has been set
1176 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1177 ip->ip_off |= IP_DF;
1179 th = (struct tcphdr *)(ip + 1);
1181 th->th_sport = sc->sc_inc.inc_lport;
1182 th->th_dport = sc->sc_inc.inc_fport;
1184 th->th_seq = htonl(sc->sc_iss);
1185 th->th_ack = htonl(sc->sc_irs + 1);
1186 th->th_off = sizeof(struct tcphdr) >> 2;
1188 th->th_flags = TH_SYN|TH_ACK;
1189 th->th_win = htons(sc->sc_wnd);
1192 /* Tack on the TCP options. */
1193 if ((sc->sc_flags & SCF_NOOPT) == 0) {
1197 to.to_flags = TOF_MSS;
1198 if (sc->sc_flags & SCF_WINSCALE) {
1199 to.to_wscale = sc->sc_requested_r_scale;
1200 to.to_flags |= TOF_SCALE;
1202 if (sc->sc_flags & SCF_TIMESTAMP) {
1203 /* Virgin timestamp or TCP cookie enhanced one. */
1204 to.to_tsval = sc->sc_ts ? sc->sc_ts : ticks;
1205 to.to_tsecr = sc->sc_tsreflect;
1206 to.to_flags |= TOF_TS;
1208 if (sc->sc_flags & SCF_SACK)
1209 to.to_flags |= TOF_SACKPERM;
1210 #ifdef TCP_SIGNATURE
1211 if (sc->sc_flags & SCF_SIGNATURE)
1212 to.to_flags |= TOF_SIGNATURE;
1214 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1216 #ifdef TCP_SIGNATURE
1217 tcp_signature_compute(m, sizeof(struct ip), 0, optlen,
1218 to.to_signature, IPSEC_DIR_OUTBOUND);
1221 /* Adjust headers by option size. */
1222 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1224 m->m_pkthdr.len += optlen;
1226 if (sc->sc_inc.inc_isipv6)
1227 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1230 ip->ip_len += optlen;
1235 if (sc->sc_inc.inc_isipv6) {
1237 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
1238 tlen + optlen - hlen);
1239 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1240 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1244 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1245 htons(tlen + optlen - hlen + IPPROTO_TCP));
1246 m->m_pkthdr.csum_flags = CSUM_TCP;
1247 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1248 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1254 * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1255 * receive and to be able to handle SYN floods from bogus source addresses
1256 * (where we will never receive any reply). SYN floods try to exhaust all
1257 * our memory and available slots in the SYN cache table to cause a denial
1258 * of service to legitimate users of the local host.
1260 * The idea of SYN cookies is to encode and include all necessary information
1261 * about the connection setup state within the SYN-ACK we send back and thus
1262 * to get along without keeping any local state until the ACK to the SYN-ACK
1263 * arrives (if ever). Everything we need to know should be available from
1264 * the information we encoded in the SYN-ACK.
1266 * More information about the theory behind SYN cookies and its first
1267 * discussion and specification can be found at:
1268 * http://cr.yp.to/syncookies.html (overview)
1269 * http://cr.yp.to/syncookies/archive (gory details)
1271 * This implementation extends the orginal idea and first implementation
1272 * of FreeBSD by using not only the initial sequence number field to store
1273 * information but also the timestamp field if present. This way we can
1274 * keep track of the entire state we need to know to recreate the session in
1275 * its original form. Almost all TCP speakers implement RFC1323 timestamps
1276 * these days. For those that do not we still have to live with the known
1277 * shortcomings of the ISN only SYN cookies.
1281 * Initial sequence number we send:
1282 * 31|................................|0
1283 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1284 * D = MD5 Digest (first dword)
1286 * R = Rotation of secret
1287 * P = Odd or Even secret
1289 * The MD5 Digest is computed with over following parameters:
1290 * a) randomly rotated secret
1291 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1292 * c) the received initial sequence number from remote host
1293 * d) the rotation offset and odd/even bit
1295 * Timestamp we send:
1296 * 31|................................|0
1297 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1298 * D = MD5 Digest (third dword) (only as filler)
1299 * S = Requested send window scale
1300 * R = Requested receive window scale
1302 * 5 = TCP-MD5 enabled (not implemented yet)
1303 * XORed with MD5 Digest (forth dword)
1305 * The timestamp isn't cryptographically secure and doesn't need to be.
1306 * The double use of the MD5 digest dwords ties it to a specific remote/
1307 * local host/port, remote initial sequence number and our local time
1308 * limited secret. A received timestamp is reverted (XORed) and then
1309 * the contained MD5 dword is compared to the computed one to ensure the
1310 * timestamp belongs to the SYN-ACK we sent. The other parameters may
1311 * have been tampered with but this isn't different from supplying bogus
1312 * values in the SYN in the first place.
1314 * Some problems with SYN cookies remain however:
1315 * Consider the problem of a recreated (and retransmitted) cookie. If the
1316 * original SYN was accepted, the connection is established. The second
1317 * SYN is inflight, and if it arrives with an ISN that falls within the
1318 * receive window, the connection is killed.
1321 * A heuristic to determine when to accept syn cookies is not necessary.
1322 * An ACK flood would cause the syncookie verification to be attempted,
1323 * but a SYN flood causes syncookies to be generated. Both are of equal
1324 * cost, so there's no point in trying to optimize the ACK flood case.
1325 * Also, if you don't process certain ACKs for some reason, then all someone
1326 * would have to do is launch a SYN and ACK flood at the same time, which
1327 * would stop cookie verification and defeat the entire purpose of syncookies.
1329 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1332 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1333 u_int32_t *flowlabel)
1336 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1339 u_int off, pmss, mss;
1342 SCH_LOCK_ASSERT(sch);
1344 /* Which of the two secrets to use. */
1345 secbits = sch->sch_oddeven ?
1346 sch->sch_secbits_odd : sch->sch_secbits_even;
1348 /* Reseed secret if too old. */
1349 if (sch->sch_reseed < time_uptime) {
1350 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */
1351 secbits = sch->sch_oddeven ?
1352 sch->sch_secbits_odd : sch->sch_secbits_even;
1353 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1354 secbits[i] = arc4random();
1355 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1358 /* Secret rotation offset. */
1359 off = sc->sc_iss & 0x7; /* iss was randomized before */
1361 /* Maximum segment size calculation. */
1362 pmss = max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), tcp_minmss);
1363 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1364 if (tcp_sc_msstab[mss] <= pmss)
1367 /* Fold parameters and MD5 digest into the ISN we will send. */
1368 data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1369 data |= off << 1; /* secret offset, derived from iss, 3 bits */
1370 data |= mss << 4; /* mss, 3 bits */
1373 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1374 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1375 MD5Update(&ctx, secbits, off);
1376 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1377 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1378 MD5Update(&ctx, &data, sizeof(data));
1379 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1381 data |= (md5_buffer[0] << 7);
1385 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1388 /* Additional parameters are stored in the timestamp if present. */
1389 if (sc->sc_flags & SCF_TIMESTAMP) {
1390 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1391 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1392 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */
1393 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */
1394 data |= md5_buffer[2] << 10; /* more digest bits */
1395 data ^= md5_buffer[3];
1397 sc->sc_tsoff = data - ticks; /* after XOR */
1404 static struct syncache *
1405 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1406 struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1410 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1414 int off, mss, wnd, flags;
1416 SCH_LOCK_ASSERT(sch);
1419 * Pull information out of SYN-ACK/ACK and
1420 * revert sequence number advances.
1422 ack = th->th_ack - 1;
1423 seq = th->th_seq - 1;
1424 off = (ack >> 1) & 0x7;
1425 mss = (ack >> 4) & 0x7;
1428 /* Which of the two secrets to use. */
1429 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1432 * The secret wasn't updated for the lifetime of a syncookie,
1433 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1435 if (sch->sch_reseed < time_uptime) {
1439 /* Recompute the digest so we can compare it. */
1441 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1442 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1443 MD5Update(&ctx, secbits, off);
1444 MD5Update(&ctx, inc, sizeof(*inc));
1445 MD5Update(&ctx, &seq, sizeof(seq));
1446 MD5Update(&ctx, &flags, sizeof(flags));
1447 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1449 /* Does the digest part of or ACK'ed ISS match? */
1450 if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1453 /* Does the digest part of our reflected timestamp match? */
1454 if (to->to_flags & TOF_TS) {
1455 data = md5_buffer[3] ^ to->to_tsecr;
1456 if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1460 /* Fill in the syncache values. */
1461 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1462 sc->sc_ipopts = NULL;
1468 if (inc->inc_isipv6) {
1469 if (sotoinpcb(so)->in6p_flags & IN6P_AUTOFLOWLABEL)
1470 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1474 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1475 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1478 /* Additional parameters that were encoded in the timestamp. */
1480 sc->sc_flags |= SCF_TIMESTAMP;
1481 sc->sc_tsreflect = to->to_tsval;
1482 sc->sc_tsoff = to->to_tsecr - ticks;
1483 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1484 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1485 sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1487 sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1489 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1490 sc->sc_flags |= SCF_WINSCALE;
1492 sc->sc_flags |= SCF_NOOPT;
1494 wnd = sbspace(&so->so_rcv);
1496 wnd = imin(wnd, TCP_MAXWIN);
1500 sc->sc_peer_mss = tcp_sc_msstab[mss];