2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * All advertising materials mentioning features or use of this software
36 * must display the following acknowledgement:
37 * This product includes software developed by Jeffrey M. Hsu.
39 * Copyright (c) 2001 Networks Associates Technologies, Inc.
40 * All rights reserved.
42 * This software was developed for the FreeBSD Project by Jonathan Lemon
43 * and NAI Labs, the Security Research Division of Network Associates, Inc.
44 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
45 * DARPA CHATS research program.
47 * Redistribution and use in source and binary forms, with or without
48 * modification, are permitted provided that the following conditions
50 * 1. Redistributions of source code must retain the above copyright
51 * notice, this list of conditions and the following disclaimer.
52 * 2. Redistributions in binary form must reproduce the above copyright
53 * notice, this list of conditions and the following disclaimer in the
54 * documentation and/or other materials provided with the distribution.
55 * 3. The name of the author may not be used to endorse or promote
56 * products derived from this software without specific prior written
59 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
71 * $FreeBSD: src/sys/netinet/tcp_syncache.c,v 1.5.2.14 2003/02/24 04:02:27 silby Exp $
72 * $DragonFly: src/sys/netinet/tcp_syncache.c,v 1.35 2008/11/22 11:03:35 sephe Exp $
75 #include "opt_inet6.h"
76 #include "opt_ipsec.h"
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/kernel.h>
81 #include <sys/sysctl.h>
82 #include <sys/malloc.h>
85 #include <sys/proc.h> /* for proc0 declaration */
86 #include <sys/random.h>
87 #include <sys/socket.h>
88 #include <sys/socketvar.h>
89 #include <sys/in_cksum.h>
91 #include <sys/msgport2.h>
92 #include <net/netmsg2.h>
95 #include <net/route.h>
97 #include <netinet/in.h>
98 #include <netinet/in_systm.h>
99 #include <netinet/ip.h>
100 #include <netinet/in_var.h>
101 #include <netinet/in_pcb.h>
102 #include <netinet/ip_var.h>
103 #include <netinet/ip6.h>
105 #include <netinet/icmp6.h>
106 #include <netinet6/nd6.h>
108 #include <netinet6/ip6_var.h>
109 #include <netinet6/in6_pcb.h>
110 #include <netinet/tcp.h>
111 #include <netinet/tcp_fsm.h>
112 #include <netinet/tcp_seq.h>
113 #include <netinet/tcp_timer.h>
114 #include <netinet/tcp_var.h>
115 #include <netinet6/tcp6_var.h>
118 #include <netinet6/ipsec.h>
120 #include <netinet6/ipsec6.h>
122 #include <netproto/key/key.h>
126 #include <netproto/ipsec/ipsec.h>
128 #include <netproto/ipsec/ipsec6.h>
130 #include <netproto/ipsec/key.h>
132 #endif /*FAST_IPSEC*/
134 #include <vm/vm_zone.h>
136 static int tcp_syncookies = 1;
137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
139 "Use TCP SYN cookies if the syncache overflows");
141 static void syncache_drop(struct syncache *, struct syncache_head *);
142 static void syncache_free(struct syncache *);
143 static void syncache_insert(struct syncache *, struct syncache_head *);
144 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
145 static int syncache_respond(struct syncache *, struct mbuf *);
146 static struct socket *syncache_socket(struct syncache *, struct socket *,
148 static void syncache_timer(void *);
149 static u_int32_t syncookie_generate(struct syncache *);
150 static struct syncache *syncookie_lookup(struct in_conninfo *,
151 struct tcphdr *, struct socket *);
154 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
155 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds,
156 * the odds are that the user has given up attempting to connect by then.
158 #define SYNCACHE_MAXREXMTS 3
160 /* Arbitrary values */
161 #define TCP_SYNCACHE_HASHSIZE 512
162 #define TCP_SYNCACHE_BUCKETLIMIT 30
164 struct netmsg_sc_timer {
165 struct netmsg nm_netmsg;
166 struct msgrec *nm_mrec; /* back pointer to containing msgrec */
170 struct netmsg_sc_timer msg;
171 lwkt_port_t port; /* constant after init */
172 int slot; /* constant after init */
175 static void syncache_timer_handler(netmsg_t);
177 struct tcp_syncache {
178 struct vm_zone *zone;
186 static struct tcp_syncache tcp_syncache;
188 struct tcp_syncache_percpu {
189 struct syncache_head *hashbase;
191 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1];
192 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1];
193 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1];
195 static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU];
197 static struct lwkt_port syncache_null_rport;
199 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
201 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD,
202 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
204 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD,
205 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
209 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
210 &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
213 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD,
214 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
216 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
217 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
219 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
221 #define SYNCACHE_HASH(inc, mask) \
222 ((tcp_syncache.hash_secret ^ \
223 (inc)->inc_faddr.s_addr ^ \
224 ((inc)->inc_faddr.s_addr >> 16) ^ \
225 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
227 #define SYNCACHE_HASH6(inc, mask) \
228 ((tcp_syncache.hash_secret ^ \
229 (inc)->inc6_faddr.s6_addr32[0] ^ \
230 (inc)->inc6_faddr.s6_addr32[3] ^ \
231 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
233 #define ENDPTS_EQ(a, b) ( \
234 (a)->ie_fport == (b)->ie_fport && \
235 (a)->ie_lport == (b)->ie_lport && \
236 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
237 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
240 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
243 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu,
244 struct syncache *sc, int slot)
246 sc->sc_rxtslot = slot;
247 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[slot];
248 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq);
249 if (!callout_active(&syncache_percpu->tt_timerq[slot])) {
250 callout_reset(&syncache_percpu->tt_timerq[slot],
251 TCPTV_RTOBASE * tcp_backoff[slot],
253 &syncache_percpu->mrec[slot]);
258 syncache_free(struct syncache *sc)
262 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
264 const boolean_t isipv6 = FALSE;
268 m_free(sc->sc_ipopts);
270 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt;
273 * If this is the only reference to a protocol-cloned
274 * route, remove it immediately.
276 if ((rt->rt_flags & RTF_WASCLONED) && rt->rt_refcnt == 1)
277 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
278 rt_mask(rt), rt->rt_flags, NULL);
282 zfree(tcp_syncache.zone, sc);
290 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
291 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
292 tcp_syncache.cache_limit =
293 tcp_syncache.hashsize * tcp_syncache.bucket_limit;
294 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
295 tcp_syncache.hash_secret = karc4random();
297 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
298 &tcp_syncache.hashsize);
299 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
300 &tcp_syncache.cache_limit);
301 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
302 &tcp_syncache.bucket_limit);
303 if (!powerof2(tcp_syncache.hashsize)) {
304 kprintf("WARNING: syncache hash size is not a power of 2.\n");
305 tcp_syncache.hashsize = 512; /* safe default */
307 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
309 lwkt_initport_replyonly_null(&syncache_null_rport);
311 for (cpu = 0; cpu < ncpus2; cpu++) {
312 struct tcp_syncache_percpu *syncache_percpu;
314 syncache_percpu = &tcp_syncache_percpu[cpu];
315 /* Allocate the hash table. */
316 MALLOC(syncache_percpu->hashbase, struct syncache_head *,
317 tcp_syncache.hashsize * sizeof(struct syncache_head),
318 M_SYNCACHE, M_WAITOK);
320 /* Initialize the hash buckets. */
321 for (i = 0; i < tcp_syncache.hashsize; i++) {
322 struct syncache_head *bucket;
324 bucket = &syncache_percpu->hashbase[i];
325 TAILQ_INIT(&bucket->sch_bucket);
326 bucket->sch_length = 0;
329 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) {
330 /* Initialize the timer queues. */
331 TAILQ_INIT(&syncache_percpu->timerq[i]);
332 callout_init(&syncache_percpu->tt_timerq[i]);
334 syncache_percpu->mrec[i].slot = i;
335 syncache_percpu->mrec[i].port = tcp_cport(cpu);
336 syncache_percpu->mrec[i].msg.nm_mrec =
337 &syncache_percpu->mrec[i];
338 netmsg_init(&syncache_percpu->mrec[i].msg.nm_netmsg,
339 &syncache_null_rport, 0,
340 syncache_timer_handler);
345 * Allocate the syncache entries. Allow the zone to allocate one
346 * more entry than cache limit, so a new entry can bump out an
349 tcp_syncache.zone = zinit("syncache", sizeof(struct syncache),
350 tcp_syncache.cache_limit * ncpus2, ZONE_INTERRUPT, 0);
351 tcp_syncache.cache_limit -= 1;
355 syncache_insert(struct syncache *sc, struct syncache_head *sch)
357 struct tcp_syncache_percpu *syncache_percpu;
358 struct syncache *sc2;
361 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
364 * Make sure that we don't overflow the per-bucket
365 * limit or the total cache size limit.
367 if (sch->sch_length >= tcp_syncache.bucket_limit) {
369 * The bucket is full, toss the oldest element.
371 sc2 = TAILQ_FIRST(&sch->sch_bucket);
372 sc2->sc_tp->ts_recent = ticks;
373 syncache_drop(sc2, sch);
374 tcpstat.tcps_sc_bucketoverflow++;
375 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) {
377 * The cache is full. Toss the oldest entry in the
378 * entire cache. This is the front entry in the
379 * first non-empty timer queue with the largest
382 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
383 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]);
387 sc2->sc_tp->ts_recent = ticks;
388 syncache_drop(sc2, NULL);
389 tcpstat.tcps_sc_cacheoverflow++;
392 /* Initialize the entry's timer. */
393 syncache_timeout(syncache_percpu, sc, 0);
395 /* Put it into the bucket. */
396 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash);
398 syncache_percpu->cache_count++;
399 tcpstat.tcps_sc_added++;
403 syncache_drop(struct syncache *sc, struct syncache_head *sch)
405 struct tcp_syncache_percpu *syncache_percpu;
407 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
409 const boolean_t isipv6 = FALSE;
412 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
416 sch = &syncache_percpu->hashbase[
417 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)];
419 sch = &syncache_percpu->hashbase[
420 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)];
424 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
426 syncache_percpu->cache_count--;
429 * Remove the entry from the syncache timer/timeout queue. Note
430 * that we do not try to stop any running timer since we do not know
431 * whether the timer's message is in-transit or not. Since timeouts
432 * are fairly long, taking an unneeded callout does not detrimentally
433 * effect performance.
435 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq);
441 * Place a timeout message on the TCP thread's message queue.
442 * This routine runs in soft interrupt context.
444 * An invariant is for this routine to be called, the callout must
445 * have been active. Note that the callout is not deactivated until
446 * after the message has been processed in syncache_timer_handler() below.
449 syncache_timer(void *p)
451 struct netmsg_sc_timer *msg = p;
453 lwkt_sendmsg(msg->nm_mrec->port, &msg->nm_netmsg.nm_lmsg);
457 * Service a timer message queued by timer expiration.
458 * This routine runs in the TCP protocol thread.
460 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
461 * If we have retransmitted an entry the maximum number of times, expire it.
463 * When we finish processing timed-out entries, we restart the timer if there
464 * are any entries still on the queue and deactivate it otherwise. Only after
465 * a timer has been deactivated here can it be restarted by syncache_timeout().
468 syncache_timer_handler(netmsg_t netmsg)
470 struct tcp_syncache_percpu *syncache_percpu;
471 struct syncache *sc, *nsc;
475 slot = ((struct netmsg_sc_timer *)netmsg)->nm_mrec->slot;
476 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
478 nsc = TAILQ_FIRST(&syncache_percpu->timerq[slot]);
479 while (nsc != NULL) {
480 if (ticks < nsc->sc_rxttime)
481 break; /* finished because timerq sorted by time */
483 inp = sc->sc_tp->t_inpcb;
484 if (slot == SYNCACHE_MAXREXMTS ||
485 slot >= tcp_syncache.rexmt_limit ||
486 inp->inp_gencnt != sc->sc_inp_gencnt) {
487 nsc = TAILQ_NEXT(sc, sc_timerq);
488 syncache_drop(sc, NULL);
489 tcpstat.tcps_sc_stale++;
493 * syncache_respond() may call back into the syncache to
494 * to modify another entry, so do not obtain the next
495 * entry on the timer chain until it has completed.
497 syncache_respond(sc, NULL);
498 nsc = TAILQ_NEXT(sc, sc_timerq);
499 tcpstat.tcps_sc_retransmitted++;
500 TAILQ_REMOVE(&syncache_percpu->timerq[slot], sc, sc_timerq);
501 syncache_timeout(syncache_percpu, sc, slot + 1);
504 callout_reset(&syncache_percpu->tt_timerq[slot],
505 nsc->sc_rxttime - ticks, syncache_timer,
506 &syncache_percpu->mrec[slot]);
508 callout_deactivate(&syncache_percpu->tt_timerq[slot]);
510 lwkt_replymsg(&netmsg->nm_lmsg, 0);
514 * Find an entry in the syncache.
517 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
519 struct tcp_syncache_percpu *syncache_percpu;
521 struct syncache_head *sch;
523 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
525 if (inc->inc_isipv6) {
526 sch = &syncache_percpu->hashbase[
527 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
529 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash)
530 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
535 sch = &syncache_percpu->hashbase[
536 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
538 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
540 if (sc->sc_inc.inc_isipv6)
543 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
551 * This function is called when we get a RST for a
552 * non-existent connection, so that we can see if the
553 * connection is in the syn cache. If it is, zap it.
556 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
559 struct syncache_head *sch;
561 sc = syncache_lookup(inc, &sch);
565 * If the RST bit is set, check the sequence number to see
566 * if this is a valid reset segment.
568 * In all states except SYN-SENT, all reset (RST) segments
569 * are validated by checking their SEQ-fields. A reset is
570 * valid if its sequence number is in the window.
572 * The sequence number in the reset segment is normally an
573 * echo of our outgoing acknowlegement numbers, but some hosts
574 * send a reset with the sequence number at the rightmost edge
575 * of our receive window, and we have to handle this case.
577 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
578 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
579 syncache_drop(sc, sch);
580 tcpstat.tcps_sc_reset++;
585 syncache_badack(struct in_conninfo *inc)
588 struct syncache_head *sch;
590 sc = syncache_lookup(inc, &sch);
592 syncache_drop(sc, sch);
593 tcpstat.tcps_sc_badack++;
598 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
601 struct syncache_head *sch;
603 /* we are called at splnet() here */
604 sc = syncache_lookup(inc, &sch);
608 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
609 if (ntohl(th->th_seq) != sc->sc_iss)
613 * If we've rertransmitted 3 times and this is our second error,
614 * we remove the entry. Otherwise, we allow it to continue on.
615 * This prevents us from incorrectly nuking an entry during a
616 * spurious network outage.
620 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) {
621 sc->sc_flags |= SCF_UNREACH;
624 syncache_drop(sc, sch);
625 tcpstat.tcps_sc_unreach++;
629 * Build a new TCP socket structure from a syncache entry.
631 static struct socket *
632 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
634 struct inpcb *inp = NULL, *linp;
638 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
640 const boolean_t isipv6 = FALSE;
644 * Ok, create the full blown connection, and set things up
645 * as they would have been set up if we had created the
646 * connection when the SYN arrived. If we can't create
647 * the connection, abort it.
649 so = sonewconn(lso, SS_ISCONNECTED);
652 * Drop the connection; we will send a RST if the peer
653 * retransmits the ACK,
655 tcpstat.tcps_listendrop++;
662 * Insert new socket into hash list.
664 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
666 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
669 inp->inp_vflag &= ~INP_IPV6;
670 inp->inp_vflag |= INP_IPV4;
672 inp->inp_laddr = sc->sc_inc.inc_laddr;
674 inp->inp_lport = sc->sc_inc.inc_lport;
675 if (in_pcbinsporthash(inp) != 0) {
677 * Undo the assignments above if we failed to
678 * put the PCB on the hash lists.
681 inp->in6p_laddr = kin6addr_any;
683 inp->inp_laddr.s_addr = INADDR_ANY;
689 /* copy old policy into new socket's */
690 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp))
691 kprintf("syncache_expand: could not copy policy\n");
694 struct in6_addr laddr6;
695 struct sockaddr_in6 sin6;
697 * Inherit socket options from the listening socket.
698 * Note that in6p_inputopts are not (and should not be)
699 * copied, since it stores previously received options and is
700 * used to detect if each new option is different than the
701 * previous one and hence should be passed to a user.
702 * If we copied in6p_inputopts, a user would not be able to
703 * receive options just after calling the accept system call.
705 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS;
706 if (linp->in6p_outputopts)
707 inp->in6p_outputopts =
708 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT);
709 inp->in6p_route = sc->sc_route6;
710 sc->sc_route6.ro_rt = NULL;
712 sin6.sin6_family = AF_INET6;
713 sin6.sin6_len = sizeof sin6;
714 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
715 sin6.sin6_port = sc->sc_inc.inc_fport;
716 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
717 laddr6 = inp->in6p_laddr;
718 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
719 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
720 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) {
721 inp->in6p_laddr = laddr6;
725 struct in_addr laddr;
726 struct sockaddr_in sin;
728 inp->inp_options = ip_srcroute(m);
729 if (inp->inp_options == NULL) {
730 inp->inp_options = sc->sc_ipopts;
731 sc->sc_ipopts = NULL;
733 inp->inp_route = sc->sc_route;
734 sc->sc_route.ro_rt = NULL;
736 sin.sin_family = AF_INET;
737 sin.sin_len = sizeof sin;
738 sin.sin_addr = sc->sc_inc.inc_faddr;
739 sin.sin_port = sc->sc_inc.inc_fport;
740 bzero(sin.sin_zero, sizeof sin.sin_zero);
741 laddr = inp->inp_laddr;
742 if (inp->inp_laddr.s_addr == INADDR_ANY)
743 inp->inp_laddr = sc->sc_inc.inc_laddr;
744 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) {
745 inp->inp_laddr = laddr;
751 tp->t_state = TCPS_SYN_RECEIVED;
752 tp->iss = sc->sc_iss;
753 tp->irs = sc->sc_irs;
756 tp->snd_wl1 = sc->sc_irs;
757 tp->rcv_up = sc->sc_irs + 1;
758 tp->rcv_wnd = sc->sc_wnd;
759 tp->rcv_adv += tp->rcv_wnd;
761 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY);
762 if (sc->sc_flags & SCF_NOOPT)
763 tp->t_flags |= TF_NOOPT;
764 if (sc->sc_flags & SCF_WINSCALE) {
765 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE;
766 tp->requested_s_scale = sc->sc_requested_s_scale;
767 tp->request_r_scale = sc->sc_request_r_scale;
769 if (sc->sc_flags & SCF_TIMESTAMP) {
770 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP;
771 tp->ts_recent = sc->sc_tsrecent;
772 tp->ts_recent_age = ticks;
774 if (sc->sc_flags & SCF_CC) {
776 * Initialization of the tcpcb for transaction;
777 * set SND.WND = SEG.WND,
778 * initialize CCsend and CCrecv.
780 tp->t_flags |= TF_REQ_CC | TF_RCVD_CC;
781 tp->cc_send = sc->sc_cc_send;
782 tp->cc_recv = sc->sc_cc_recv;
784 if (sc->sc_flags & SCF_SACK_PERMITTED)
785 tp->t_flags |= TF_SACK_PERMITTED;
787 tcp_mss(tp, sc->sc_peer_mss);
790 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
792 if (sc->sc_rxtslot != 0)
793 tp->snd_cwnd = tp->t_maxseg;
794 tcp_create_timermsg(tp);
795 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
797 tcpstat.tcps_accepts++;
807 * This function gets called when we receive an ACK for a
808 * socket in the LISTEN state. We look up the connection
809 * in the syncache, and if its there, we pull it out of
810 * the cache and turn it into a full-blown connection in
811 * the SYN-RECEIVED state.
814 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop,
818 struct syncache_head *sch;
821 sc = syncache_lookup(inc, &sch);
824 * There is no syncache entry, so see if this ACK is
825 * a returning syncookie. To do this, first:
826 * A. See if this socket has had a syncache entry dropped in
827 * the past. We don't want to accept a bogus syncookie
828 * if we've never received a SYN.
829 * B. check that the syncookie is valid. If it is, then
830 * cobble up a fake syncache entry, and return.
834 sc = syncookie_lookup(inc, th, *sop);
838 tcpstat.tcps_sc_recvcookie++;
842 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
844 if (th->th_ack != sc->sc_iss + 1)
847 so = syncache_socket(sc, *sop, m);
851 /* XXXjlemon check this - is this correct? */
852 tcp_respond(NULL, m, m, th,
853 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK);
855 m_freem(m); /* XXX only needed for above */
856 tcpstat.tcps_sc_aborted++;
858 tcpstat.tcps_sc_completed++;
863 syncache_drop(sc, sch);
869 * Given a LISTEN socket and an inbound SYN request, add
870 * this to the syn cache, and send back a segment:
871 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
874 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
875 * Doing so would require that we hold onto the data and deliver it
876 * to the application. However, if we are the target of a SYN-flood
877 * DoS attack, an attacker could send data which would eventually
878 * consume all available buffer space if it were ACKed. By not ACKing
879 * the data, we avoid this DoS scenario.
882 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
883 struct socket **sop, struct mbuf *m)
885 struct tcp_syncache_percpu *syncache_percpu;
888 struct syncache *sc = NULL;
889 struct syncache_head *sch;
890 struct mbuf *ipopts = NULL;
891 struct rmxp_tao *taop;
894 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
899 * Remember the IP options, if any.
902 if (!inc->inc_isipv6)
904 ipopts = ip_srcroute(m);
907 * See if we already have an entry for this connection.
908 * If we do, resend the SYN,ACK, and reset the retransmit timer.
911 * The syncache should be re-initialized with the contents
912 * of the new SYN which may have different options.
914 sc = syncache_lookup(inc, &sch);
916 tcpstat.tcps_sc_dupsyn++;
919 * If we were remembering a previous source route,
920 * forget it and use the new one we've been given.
923 m_free(sc->sc_ipopts);
924 sc->sc_ipopts = ipopts;
927 * Update timestamp if present.
929 if (sc->sc_flags & SCF_TIMESTAMP)
930 sc->sc_tsrecent = to->to_tsval;
932 /* Just update the TOF_SACK_PERMITTED for now. */
933 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
934 sc->sc_flags |= SCF_SACK_PERMITTED;
936 sc->sc_flags &= ~SCF_SACK_PERMITTED;
939 * PCB may have changed, pick up new values.
942 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
943 if (syncache_respond(sc, m) == 0) {
944 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot],
946 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot);
947 tcpstat.tcps_sndacks++;
948 tcpstat.tcps_sndtotal++;
955 * This allocation is guaranteed to succeed because we
956 * preallocate one more syncache entry than cache_limit.
958 sc = zalloc(tcp_syncache.zone);
961 * Fill in the syncache values.
964 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
965 sc->sc_ipopts = ipopts;
966 sc->sc_inc.inc_fport = inc->inc_fport;
967 sc->sc_inc.inc_lport = inc->inc_lport;
969 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
970 if (inc->inc_isipv6) {
971 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
972 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
973 sc->sc_route6.ro_rt = NULL;
977 sc->sc_inc.inc_faddr = inc->inc_faddr;
978 sc->sc_inc.inc_laddr = inc->inc_laddr;
979 sc->sc_route.ro_rt = NULL;
981 sc->sc_irs = th->th_seq;
983 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0;
985 sc->sc_iss = syncookie_generate(sc);
987 sc->sc_iss = karc4random();
989 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */
990 win = ssb_space(&so->so_rcv);
992 win = imin(win, TCP_MAXWIN);
995 if (tcp_do_rfc1323) {
997 * A timestamp received in a SYN makes
998 * it ok to send timestamp requests and replies.
1000 if (to->to_flags & TOF_TS) {
1001 sc->sc_tsrecent = to->to_tsval;
1002 sc->sc_flags |= SCF_TIMESTAMP;
1004 if (to->to_flags & TOF_SCALE) {
1007 /* Compute proper scaling value from buffer space */
1008 while (wscale < TCP_MAX_WINSHIFT &&
1009 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat)
1011 sc->sc_request_r_scale = wscale;
1012 sc->sc_requested_s_scale = to->to_requested_s_scale;
1013 sc->sc_flags |= SCF_WINSCALE;
1016 if (tcp_do_rfc1644) {
1018 * A CC or CC.new option received in a SYN makes
1019 * it ok to send CC in subsequent segments.
1021 if (to->to_flags & (TOF_CC | TOF_CCNEW)) {
1022 sc->sc_cc_recv = to->to_cc;
1023 sc->sc_cc_send = CC_INC(tcp_ccgen);
1024 sc->sc_flags |= SCF_CC;
1027 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1028 sc->sc_flags |= SCF_SACK_PERMITTED;
1029 if (tp->t_flags & TF_NOOPT)
1030 sc->sc_flags = SCF_NOOPT;
1034 * We have the option here of not doing TAO (even if the segment
1035 * qualifies) and instead fall back to a normal 3WHS via the syncache.
1036 * This allows us to apply synflood protection to TAO-qualifying SYNs
1037 * also. However, there should be a hueristic to determine when to
1038 * do this, and is not present at the moment.
1042 * Perform TAO test on incoming CC (SEG.CC) option, if any.
1043 * - compare SEG.CC against cached CC from the same host, if any.
1044 * - if SEG.CC > chached value, SYN must be new and is accepted
1045 * immediately: save new CC in the cache, mark the socket
1046 * connected, enter ESTABLISHED state, turn on flag to
1047 * send a SYN in the next segment.
1048 * A virtual advertised window is set in rcv_adv to
1049 * initialize SWS prevention. Then enter normal segment
1050 * processing: drop SYN, process data and FIN.
1051 * - otherwise do a normal 3-way handshake.
1053 taop = tcp_gettaocache(&sc->sc_inc);
1054 if (to->to_flags & TOF_CC) {
1055 if ((tp->t_flags & TF_NOPUSH) &&
1056 sc->sc_flags & SCF_CC &&
1057 taop != NULL && taop->tao_cc != 0 &&
1058 CC_GT(to->to_cc, taop->tao_cc)) {
1060 so = syncache_socket(sc, *sop, m);
1062 taop->tao_cc = to->to_cc;
1066 return (so != NULL);
1070 * No CC option, but maybe CC.NEW: invalidate cached value.
1076 * TAO test failed or there was no CC option,
1077 * do a standard 3-way handshake.
1079 if (syncache_respond(sc, m) == 0) {
1080 syncache_insert(sc, sch);
1081 tcpstat.tcps_sndacks++;
1082 tcpstat.tcps_sndtotal++;
1085 tcpstat.tcps_sc_dropped++;
1092 syncache_respond(struct syncache *sc, struct mbuf *m)
1096 u_int16_t tlen, hlen, mssopt;
1097 struct ip *ip = NULL;
1100 struct ip6_hdr *ip6 = NULL;
1102 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1104 const boolean_t isipv6 = FALSE;
1108 rt = tcp_rtlookup6(&sc->sc_inc);
1110 mssopt = rt->rt_ifp->if_mtu -
1111 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1113 mssopt = tcp_v6mssdflt;
1114 hlen = sizeof(struct ip6_hdr);
1116 rt = tcp_rtlookup(&sc->sc_inc);
1118 mssopt = rt->rt_ifp->if_mtu -
1119 (sizeof(struct ip) + sizeof(struct tcphdr));
1121 mssopt = tcp_mssdflt;
1122 hlen = sizeof(struct ip);
1125 /* Compute the size of the TCP options. */
1126 if (sc->sc_flags & SCF_NOOPT) {
1129 optlen = TCPOLEN_MAXSEG +
1130 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) +
1131 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) +
1132 ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0) +
1133 ((sc->sc_flags & SCF_SACK_PERMITTED) ?
1134 TCPOLEN_SACK_PERMITTED_ALIGNED : 0);
1136 tlen = hlen + sizeof(struct tcphdr) + optlen;
1140 * assume that the entire packet will fit in a header mbuf
1142 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small"));
1145 * XXX shouldn't this reuse the mbuf if possible ?
1146 * Create the IP+TCP header from scratch.
1151 m = m_gethdr(MB_DONTWAIT, MT_HEADER);
1154 m->m_data += max_linkhdr;
1156 m->m_pkthdr.len = tlen;
1157 m->m_pkthdr.rcvif = NULL;
1160 ip6 = mtod(m, struct ip6_hdr *);
1161 ip6->ip6_vfc = IPV6_VERSION;
1162 ip6->ip6_nxt = IPPROTO_TCP;
1163 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1164 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1165 ip6->ip6_plen = htons(tlen - hlen);
1166 /* ip6_hlim is set after checksum */
1167 /* ip6_flow = ??? */
1169 th = (struct tcphdr *)(ip6 + 1);
1171 ip = mtod(m, struct ip *);
1172 ip->ip_v = IPVERSION;
1173 ip->ip_hl = sizeof(struct ip) >> 2;
1178 ip->ip_p = IPPROTO_TCP;
1179 ip->ip_src = sc->sc_inc.inc_laddr;
1180 ip->ip_dst = sc->sc_inc.inc_faddr;
1181 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */
1182 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */
1185 * See if we should do MTU discovery. Route lookups are
1186 * expensive, so we will only unset the DF bit if:
1188 * 1) path_mtu_discovery is disabled
1189 * 2) the SCF_UNREACH flag has been set
1191 if (path_mtu_discovery
1192 && ((sc->sc_flags & SCF_UNREACH) == 0)) {
1193 ip->ip_off |= IP_DF;
1196 th = (struct tcphdr *)(ip + 1);
1198 th->th_sport = sc->sc_inc.inc_lport;
1199 th->th_dport = sc->sc_inc.inc_fport;
1201 th->th_seq = htonl(sc->sc_iss);
1202 th->th_ack = htonl(sc->sc_irs + 1);
1203 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1205 th->th_flags = TH_SYN | TH_ACK;
1206 th->th_win = htons(sc->sc_wnd);
1209 /* Tack on the TCP options. */
1212 optp = (u_int8_t *)(th + 1);
1213 *optp++ = TCPOPT_MAXSEG;
1214 *optp++ = TCPOLEN_MAXSEG;
1215 *optp++ = (mssopt >> 8) & 0xff;
1216 *optp++ = mssopt & 0xff;
1218 if (sc->sc_flags & SCF_WINSCALE) {
1219 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
1220 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
1221 sc->sc_request_r_scale);
1225 if (sc->sc_flags & SCF_TIMESTAMP) {
1226 u_int32_t *lp = (u_int32_t *)(optp);
1228 /* Form timestamp option as shown in appendix A of RFC 1323. */
1229 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1230 *lp++ = htonl(ticks);
1231 *lp = htonl(sc->sc_tsrecent);
1232 optp += TCPOLEN_TSTAMP_APPA;
1236 * Send CC and CC.echo if we received CC from our peer.
1238 if (sc->sc_flags & SCF_CC) {
1239 u_int32_t *lp = (u_int32_t *)(optp);
1241 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC));
1242 *lp++ = htonl(sc->sc_cc_send);
1243 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO));
1244 *lp = htonl(sc->sc_cc_recv);
1245 optp += TCPOLEN_CC_APPA * 2;
1248 if (sc->sc_flags & SCF_SACK_PERMITTED) {
1249 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED);
1250 optp += TCPOLEN_SACK_PERMITTED_ALIGNED;
1255 struct route_in6 *ro6 = &sc->sc_route6;
1258 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
1259 ip6->ip6_hlim = in6_selecthlim(NULL,
1260 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL);
1261 error = ip6_output(m, NULL, ro6, 0, NULL, NULL,
1262 sc->sc_tp->t_inpcb);
1264 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1265 htons(tlen - hlen + IPPROTO_TCP));
1266 m->m_pkthdr.csum_flags = CSUM_TCP;
1267 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1268 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 0, NULL,
1269 sc->sc_tp->t_inpcb);
1277 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .|
1279 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .|
1281 * (A): peer mss index
1285 * The values below are chosen to minimize the size of the tcp_secret
1286 * table, as well as providing roughly a 16 second lifetime for the cookie.
1289 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */
1290 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */
1292 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1)
1293 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS)
1294 #define SYNCOOKIE_TIMEOUT \
1295 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT))
1296 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK)
1299 u_int32_t ts_secbits[4];
1301 } tcp_secret[SYNCOOKIE_NSECRETS];
1303 static int tcp_msstab[] = { 0, 536, 1460, 8960 };
1305 static MD5_CTX syn_ctx;
1307 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v))
1310 u_int32_t laddr, faddr;
1311 u_int32_t secbits[4];
1312 u_int16_t lport, fport;
1316 CTASSERT(sizeof(struct md5_add) == 28);
1320 * Consider the problem of a recreated (and retransmitted) cookie. If the
1321 * original SYN was accepted, the connection is established. The second
1322 * SYN is inflight, and if it arrives with an ISN that falls within the
1323 * receive window, the connection is killed.
1325 * However, since cookies have other problems, this may not be worth
1330 syncookie_generate(struct syncache *sc)
1332 u_int32_t md5_buffer[4];
1337 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1339 const boolean_t isipv6 = FALSE;
1342 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK;
1343 if (tcp_secret[idx].ts_expire < ticks) {
1344 for (i = 0; i < 4; i++)
1345 tcp_secret[idx].ts_secbits[i] = karc4random();
1346 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT;
1348 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--)
1349 if (tcp_msstab[data] <= sc->sc_peer_mss)
1351 data = (data << SYNCOOKIE_WNDBITS) | idx;
1352 data ^= sc->sc_irs; /* peer's iss */
1355 MD5Add(sc->sc_inc.inc6_laddr);
1356 MD5Add(sc->sc_inc.inc6_faddr);
1360 add.laddr = sc->sc_inc.inc_laddr.s_addr;
1361 add.faddr = sc->sc_inc.inc_faddr.s_addr;
1363 add.lport = sc->sc_inc.inc_lport;
1364 add.fport = sc->sc_inc.inc_fport;
1365 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1366 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1367 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1368 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1370 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1371 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK);
1375 static struct syncache *
1376 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so)
1378 u_int32_t md5_buffer[4];
1379 struct syncache *sc;
1384 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */
1385 idx = data & SYNCOOKIE_WNDMASK;
1386 if (tcp_secret[idx].ts_expire < ticks ||
1387 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks)
1391 if (inc->inc_isipv6) {
1392 MD5Add(inc->inc6_laddr);
1393 MD5Add(inc->inc6_faddr);
1399 add.laddr = inc->inc_laddr.s_addr;
1400 add.faddr = inc->inc_faddr.s_addr;
1402 add.lport = inc->inc_lport;
1403 add.fport = inc->inc_fport;
1404 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1405 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1406 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1407 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1409 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1410 data ^= md5_buffer[0];
1411 if (data & ~SYNCOOKIE_DATAMASK)
1413 data = data >> SYNCOOKIE_WNDBITS;
1416 * This allocation is guaranteed to succeed because we
1417 * preallocate one more syncache entry than cache_limit.
1419 sc = zalloc(tcp_syncache.zone);
1422 * Fill in the syncache values.
1423 * XXX duplicate code from syncache_add
1425 sc->sc_ipopts = NULL;
1426 sc->sc_inc.inc_fport = inc->inc_fport;
1427 sc->sc_inc.inc_lport = inc->inc_lport;
1429 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1430 if (inc->inc_isipv6) {
1431 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1432 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1433 sc->sc_route6.ro_rt = NULL;
1437 sc->sc_inc.inc_faddr = inc->inc_faddr;
1438 sc->sc_inc.inc_laddr = inc->inc_laddr;
1439 sc->sc_route.ro_rt = NULL;
1441 sc->sc_irs = th->th_seq - 1;
1442 sc->sc_iss = th->th_ack - 1;
1443 wnd = ssb_space(&so->so_rcv);
1445 wnd = imin(wnd, TCP_MAXWIN);
1449 sc->sc_peer_mss = tcp_msstab[data];