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 $
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_timer2.h>
115 #include <netinet/tcp_var.h>
116 #include <netinet6/tcp6_var.h>
119 #include <netinet6/ipsec.h>
121 #include <netinet6/ipsec6.h>
123 #include <netproto/key/key.h>
127 #include <netproto/ipsec/ipsec.h>
129 #include <netproto/ipsec/ipsec6.h>
131 #include <netproto/ipsec/key.h>
133 #endif /*FAST_IPSEC*/
135 static int tcp_syncookies = 1;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
138 "Use TCP SYN cookies if the syncache overflows");
140 static void syncache_drop(struct syncache *, struct syncache_head *);
141 static void syncache_free(struct syncache *);
142 static void syncache_insert(struct syncache *, struct syncache_head *);
143 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
144 static int syncache_respond(struct syncache *, struct mbuf *);
145 static struct socket *syncache_socket(struct syncache *, struct socket *,
147 static void syncache_timer(void *);
148 static u_int32_t syncookie_generate(struct syncache *);
149 static struct syncache *syncookie_lookup(struct in_conninfo *,
150 struct tcphdr *, struct socket *);
153 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
154 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds,
155 * the odds are that the user has given up attempting to connect by then.
157 #define SYNCACHE_MAXREXMTS 3
159 /* Arbitrary values */
160 #define TCP_SYNCACHE_HASHSIZE 512
161 #define TCP_SYNCACHE_BUCKETLIMIT 30
163 struct netmsg_sc_timer {
164 struct netmsg_base base;
165 struct msgrec *nm_mrec; /* back pointer to containing msgrec */
169 struct netmsg_sc_timer msg;
170 lwkt_port_t port; /* constant after init */
171 int slot; /* constant after init */
174 static void syncache_timer_handler(netmsg_t);
176 struct tcp_syncache {
184 static struct tcp_syncache tcp_syncache;
186 TAILQ_HEAD(syncache_list, syncache);
188 struct tcp_syncache_percpu {
189 struct syncache_head *hashbase;
191 struct syncache_list 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);
281 kfree(sc, M_SYNCACHE);
289 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
290 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
291 tcp_syncache.cache_limit =
292 tcp_syncache.hashsize * tcp_syncache.bucket_limit;
293 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
294 tcp_syncache.hash_secret = karc4random();
296 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
297 &tcp_syncache.hashsize);
298 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
299 &tcp_syncache.cache_limit);
300 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
301 &tcp_syncache.bucket_limit);
302 if (!powerof2(tcp_syncache.hashsize)) {
303 kprintf("WARNING: syncache hash size is not a power of 2.\n");
304 tcp_syncache.hashsize = 512; /* safe default */
306 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
308 lwkt_initport_replyonly_null(&syncache_null_rport);
310 for (cpu = 0; cpu < ncpus2; cpu++) {
311 struct tcp_syncache_percpu *syncache_percpu;
313 syncache_percpu = &tcp_syncache_percpu[cpu];
314 /* Allocate the hash table. */
315 syncache_percpu->hashbase = kmalloc(tcp_syncache.hashsize * sizeof(struct syncache_head),
316 M_SYNCACHE, M_WAITOK);
318 /* Initialize the hash buckets. */
319 for (i = 0; i < tcp_syncache.hashsize; i++) {
320 struct syncache_head *bucket;
322 bucket = &syncache_percpu->hashbase[i];
323 TAILQ_INIT(&bucket->sch_bucket);
324 bucket->sch_length = 0;
327 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) {
328 /* Initialize the timer queues. */
329 TAILQ_INIT(&syncache_percpu->timerq[i]);
330 callout_init_mp(&syncache_percpu->tt_timerq[i]);
332 syncache_percpu->mrec[i].slot = i;
333 syncache_percpu->mrec[i].port = cpu_portfn(cpu);
334 syncache_percpu->mrec[i].msg.nm_mrec =
335 &syncache_percpu->mrec[i];
336 netmsg_init(&syncache_percpu->mrec[i].msg.base,
337 NULL, &syncache_null_rport,
338 0, syncache_timer_handler);
344 syncache_insert(struct syncache *sc, struct syncache_head *sch)
346 struct tcp_syncache_percpu *syncache_percpu;
347 struct syncache *sc2;
350 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
353 * Make sure that we don't overflow the per-bucket
354 * limit or the total cache size limit.
356 if (sch->sch_length >= tcp_syncache.bucket_limit) {
358 * The bucket is full, toss the oldest element.
360 sc2 = TAILQ_FIRST(&sch->sch_bucket);
361 sc2->sc_tp->ts_recent = ticks;
362 syncache_drop(sc2, sch);
363 tcpstat.tcps_sc_bucketoverflow++;
364 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) {
366 * The cache is full. Toss the oldest entry in the
367 * entire cache. This is the front entry in the
368 * first non-empty timer queue with the largest
371 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
372 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]);
373 while (sc2 && (sc2->sc_flags & SCF_MARKER))
374 sc2 = TAILQ_NEXT(sc2, sc_timerq);
378 sc2->sc_tp->ts_recent = ticks;
379 syncache_drop(sc2, NULL);
380 tcpstat.tcps_sc_cacheoverflow++;
383 /* Initialize the entry's timer. */
384 syncache_timeout(syncache_percpu, sc, 0);
386 /* Put it into the bucket. */
387 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash);
389 syncache_percpu->cache_count++;
390 tcpstat.tcps_sc_added++;
394 syncache_destroy(struct tcpcb *tp)
396 struct tcp_syncache_percpu *syncache_percpu;
397 struct syncache_head *bucket;
401 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
404 for (i = 0; i < tcp_syncache.hashsize; i++) {
405 bucket = &syncache_percpu->hashbase[i];
406 TAILQ_FOREACH(sc, &bucket->sch_bucket, sc_hash) {
414 syncache_drop(struct syncache *sc, struct syncache_head *sch)
416 struct tcp_syncache_percpu *syncache_percpu;
418 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
420 const boolean_t isipv6 = FALSE;
423 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
427 sch = &syncache_percpu->hashbase[
428 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)];
430 sch = &syncache_percpu->hashbase[
431 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)];
435 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
437 syncache_percpu->cache_count--;
446 * Remove the entry from the syncache timer/timeout queue. Note
447 * that we do not try to stop any running timer since we do not know
448 * whether the timer's message is in-transit or not. Since timeouts
449 * are fairly long, taking an unneeded callout does not detrimentally
450 * effect performance.
452 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq);
458 * Place a timeout message on the TCP thread's message queue.
459 * This routine runs in soft interrupt context.
461 * An invariant is for this routine to be called, the callout must
462 * have been active. Note that the callout is not deactivated until
463 * after the message has been processed in syncache_timer_handler() below.
466 syncache_timer(void *p)
468 struct netmsg_sc_timer *msg = p;
470 lwkt_sendmsg(msg->nm_mrec->port, &msg->base.lmsg);
474 * Service a timer message queued by timer expiration.
475 * This routine runs in the TCP protocol thread.
477 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
478 * If we have retransmitted an entry the maximum number of times, expire it.
480 * When we finish processing timed-out entries, we restart the timer if there
481 * are any entries still on the queue and deactivate it otherwise. Only after
482 * a timer has been deactivated here can it be restarted by syncache_timeout().
485 syncache_timer_handler(netmsg_t msg)
487 struct tcp_syncache_percpu *syncache_percpu;
489 struct syncache marker;
490 struct syncache_list *list;
494 slot = ((struct netmsg_sc_timer *)msg)->nm_mrec->slot;
495 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
497 list = &syncache_percpu->timerq[slot];
500 * Use a marker to keep our place in the scan. syncache_drop()
501 * can block and cause any next pointer we cache to become stale.
503 marker.sc_flags = SCF_MARKER;
504 TAILQ_INSERT_HEAD(list, &marker, sc_timerq);
506 while ((sc = TAILQ_NEXT(&marker, sc_timerq)) != NULL) {
510 TAILQ_REMOVE(list, &marker, sc_timerq);
511 TAILQ_INSERT_AFTER(list, sc, &marker, sc_timerq);
513 if (sc->sc_flags & SCF_MARKER)
516 if (ticks < sc->sc_rxttime)
517 break; /* finished because timerq sorted by time */
518 if (sc->sc_tp == NULL) {
519 syncache_drop(sc, NULL);
520 tcpstat.tcps_sc_stale++;
523 inp = sc->sc_tp->t_inpcb;
524 if (slot == SYNCACHE_MAXREXMTS ||
525 slot >= tcp_syncache.rexmt_limit ||
527 inp->inp_gencnt != sc->sc_inp_gencnt) {
528 syncache_drop(sc, NULL);
529 tcpstat.tcps_sc_stale++;
533 * syncache_respond() may call back into the syncache to
534 * to modify another entry, so do not obtain the next
535 * entry on the timer chain until it has completed.
537 syncache_respond(sc, NULL);
538 tcpstat.tcps_sc_retransmitted++;
539 TAILQ_REMOVE(list, sc, sc_timerq);
540 syncache_timeout(syncache_percpu, sc, slot + 1);
542 TAILQ_REMOVE(list, &marker, sc_timerq);
545 callout_reset(&syncache_percpu->tt_timerq[slot],
546 sc->sc_rxttime - ticks, syncache_timer,
547 &syncache_percpu->mrec[slot]);
549 callout_deactivate(&syncache_percpu->tt_timerq[slot]);
551 lwkt_replymsg(&msg->base.lmsg, 0);
555 * Find an entry in the syncache.
558 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
560 struct tcp_syncache_percpu *syncache_percpu;
562 struct syncache_head *sch;
564 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
566 if (inc->inc_isipv6) {
567 sch = &syncache_percpu->hashbase[
568 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
570 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash)
571 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
576 sch = &syncache_percpu->hashbase[
577 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
579 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
581 if (sc->sc_inc.inc_isipv6)
584 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
592 * This function is called when we get a RST for a
593 * non-existent connection, so that we can see if the
594 * connection is in the syn cache. If it is, zap it.
597 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
600 struct syncache_head *sch;
602 sc = syncache_lookup(inc, &sch);
607 * If the RST bit is set, check the sequence number to see
608 * if this is a valid reset segment.
610 * In all states except SYN-SENT, all reset (RST) segments
611 * are validated by checking their SEQ-fields. A reset is
612 * valid if its sequence number is in the window.
614 * The sequence number in the reset segment is normally an
615 * echo of our outgoing acknowlegement numbers, but some hosts
616 * send a reset with the sequence number at the rightmost edge
617 * of our receive window, and we have to handle this case.
619 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
620 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
621 syncache_drop(sc, sch);
622 tcpstat.tcps_sc_reset++;
627 syncache_badack(struct in_conninfo *inc)
630 struct syncache_head *sch;
632 sc = syncache_lookup(inc, &sch);
634 syncache_drop(sc, sch);
635 tcpstat.tcps_sc_badack++;
640 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
643 struct syncache_head *sch;
645 /* we are called at splnet() here */
646 sc = syncache_lookup(inc, &sch);
650 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
651 if (ntohl(th->th_seq) != sc->sc_iss)
655 * If we've rertransmitted 3 times and this is our second error,
656 * we remove the entry. Otherwise, we allow it to continue on.
657 * This prevents us from incorrectly nuking an entry during a
658 * spurious network outage.
662 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) {
663 sc->sc_flags |= SCF_UNREACH;
666 syncache_drop(sc, sch);
667 tcpstat.tcps_sc_unreach++;
671 * Build a new TCP socket structure from a syncache entry.
673 * This is called from the context of the SYN+ACK
675 static struct socket *
676 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
678 struct inpcb *inp = NULL, *linp;
680 struct tcpcb *tp, *ltp;
683 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
685 const boolean_t isipv6 = FALSE;
687 struct sockaddr_in sin_faddr;
688 struct sockaddr_in6 sin6_faddr;
689 struct sockaddr *faddr;
692 faddr = (struct sockaddr *)&sin6_faddr;
693 sin6_faddr.sin6_family = AF_INET6;
694 sin6_faddr.sin6_len = sizeof(sin6_faddr);
695 sin6_faddr.sin6_addr = sc->sc_inc.inc6_faddr;
696 sin6_faddr.sin6_port = sc->sc_inc.inc_fport;
697 sin6_faddr.sin6_flowinfo = sin6_faddr.sin6_scope_id = 0;
699 faddr = (struct sockaddr *)&sin_faddr;
700 sin_faddr.sin_family = AF_INET;
701 sin_faddr.sin_len = sizeof(sin_faddr);
702 sin_faddr.sin_addr = sc->sc_inc.inc_faddr;
703 sin_faddr.sin_port = sc->sc_inc.inc_fport;
704 bzero(sin_faddr.sin_zero, sizeof(sin_faddr.sin_zero));
708 * Ok, create the full blown connection, and set things up
709 * as they would have been set up if we had created the
710 * connection when the SYN arrived. If we can't create
711 * the connection, abort it.
713 * Set the protocol processing port for the socket to the current
714 * port (that the connection came in on).
716 so = sonewconn_faddr(lso, SS_ISCONNECTED, faddr);
719 * Drop the connection; we will send a RST if the peer
720 * retransmits the ACK,
722 tcpstat.tcps_listendrop++;
727 * Insert new socket into hash list.
730 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
732 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
735 inp->inp_vflag &= ~INP_IPV6;
736 inp->inp_vflag |= INP_IPV4;
737 inp->inp_flags &= ~IN6P_IPV6_V6ONLY;
739 inp->inp_laddr = sc->sc_inc.inc_laddr;
741 inp->inp_lport = sc->sc_inc.inc_lport;
742 if (in_pcbinsporthash(inp) != 0) {
744 * Undo the assignments above if we failed to
745 * put the PCB on the hash lists.
748 inp->in6p_laddr = kin6addr_any;
750 inp->inp_laddr.s_addr = INADDR_ANY;
756 /* copy old policy into new socket's */
757 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp))
758 kprintf("syncache_expand: could not copy policy\n");
761 struct in6_addr laddr6;
763 * Inherit socket options from the listening socket.
764 * Note that in6p_inputopts are not (and should not be)
765 * copied, since it stores previously received options and is
766 * used to detect if each new option is different than the
767 * previous one and hence should be passed to a user.
768 * If we copied in6p_inputopts, a user would not be able to
769 * receive options just after calling the accept system call.
771 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS;
772 if (linp->in6p_outputopts)
773 inp->in6p_outputopts =
774 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT);
775 inp->in6p_route = sc->sc_route6;
776 sc->sc_route6.ro_rt = NULL;
778 laddr6 = inp->in6p_laddr;
779 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
780 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
781 if (in6_pcbconnect(inp, faddr, &thread0)) {
782 inp->in6p_laddr = laddr6;
786 struct in_addr laddr;
788 inp->inp_options = ip_srcroute(m);
789 if (inp->inp_options == NULL) {
790 inp->inp_options = sc->sc_ipopts;
791 sc->sc_ipopts = NULL;
793 inp->inp_route = sc->sc_route;
794 sc->sc_route.ro_rt = NULL;
796 laddr = inp->inp_laddr;
797 if (inp->inp_laddr.s_addr == INADDR_ANY)
798 inp->inp_laddr = sc->sc_inc.inc_laddr;
799 if (in_pcbconnect(inp, faddr, &thread0)) {
800 inp->inp_laddr = laddr;
806 * The current port should be in the context of the SYN+ACK and
807 * so should match the tcp address port.
809 * XXX we may be running on the netisr thread instead of a tcp
810 * thread, in which case port will not match
811 * curthread->td_msgport.
814 port = tcp6_addrport();
816 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport,
817 inp->inp_laddr.s_addr, inp->inp_lport);
819 if (port != &curthread->td_msgport) {
821 kprintf("TCP PORT MISMATCH %p vs %p\n",
822 port, &curthread->td_msgport);
824 /*KKASSERT(port == &curthread->td_msgport);*/
827 tp->t_state = TCPS_SYN_RECEIVED;
828 tp->iss = sc->sc_iss;
829 tp->irs = sc->sc_irs;
832 tp->snd_wl1 = sc->sc_irs;
833 tp->rcv_up = sc->sc_irs + 1;
834 tp->rcv_wnd = sc->sc_wnd;
835 tp->rcv_adv += tp->rcv_wnd;
837 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY);
838 if (sc->sc_flags & SCF_NOOPT)
839 tp->t_flags |= TF_NOOPT;
840 if (sc->sc_flags & SCF_WINSCALE) {
841 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE;
842 tp->requested_s_scale = sc->sc_requested_s_scale;
843 tp->request_r_scale = sc->sc_request_r_scale;
845 if (sc->sc_flags & SCF_TIMESTAMP) {
846 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP;
847 tp->ts_recent = sc->sc_tsrecent;
848 tp->ts_recent_age = ticks;
850 if (sc->sc_flags & SCF_SACK_PERMITTED)
851 tp->t_flags |= TF_SACK_PERMITTED;
854 if (sc->sc_flags & SCF_SIGNATURE)
855 tp->t_flags |= TF_SIGNATURE;
856 #endif /* TCP_SIGNATURE */
859 tcp_mss(tp, sc->sc_peer_mss);
862 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
864 if (sc->sc_rxtslot != 0)
865 tp->snd_cwnd = tp->t_maxseg;
868 * Inherit some properties from the listen socket
870 ltp = intotcpcb(linp);
871 tp->t_keepinit = ltp->t_keepinit;
872 tp->t_keepidle = ltp->t_keepidle;
873 tp->t_keepintvl = ltp->t_keepintvl;
874 tp->t_keepcnt = ltp->t_keepcnt;
875 tp->t_maxidle = ltp->t_maxidle;
877 tcp_create_timermsg(tp, port);
878 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
880 tcpstat.tcps_accepts++;
890 * This function gets called when we receive an ACK for a
891 * socket in the LISTEN state. We look up the connection
892 * in the syncache, and if its there, we pull it out of
893 * the cache and turn it into a full-blown connection in
894 * the SYN-RECEIVED state.
897 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop,
901 struct syncache_head *sch;
904 sc = syncache_lookup(inc, &sch);
907 * There is no syncache entry, so see if this ACK is
908 * a returning syncookie. To do this, first:
909 * A. See if this socket has had a syncache entry dropped in
910 * the past. We don't want to accept a bogus syncookie
911 * if we've never received a SYN.
912 * B. check that the syncookie is valid. If it is, then
913 * cobble up a fake syncache entry, and return.
917 sc = syncookie_lookup(inc, th, *sop);
921 tcpstat.tcps_sc_recvcookie++;
925 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
927 if (th->th_ack != sc->sc_iss + 1)
930 so = syncache_socket(sc, *sop, m);
934 /* XXXjlemon check this - is this correct? */
935 tcp_respond(NULL, m, m, th,
936 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK);
938 m_freem(m); /* XXX only needed for above */
939 tcpstat.tcps_sc_aborted++;
941 tcpstat.tcps_sc_completed++;
946 syncache_drop(sc, sch);
952 * Given a LISTEN socket and an inbound SYN request, add
953 * this to the syn cache, and send back a segment:
954 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
957 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
958 * Doing so would require that we hold onto the data and deliver it
959 * to the application. However, if we are the target of a SYN-flood
960 * DoS attack, an attacker could send data which would eventually
961 * consume all available buffer space if it were ACKed. By not ACKing
962 * the data, we avoid this DoS scenario.
965 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
966 struct socket **sop, struct mbuf *m)
968 struct tcp_syncache_percpu *syncache_percpu;
971 struct syncache *sc = NULL;
972 struct syncache_head *sch;
973 struct mbuf *ipopts = NULL;
976 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
981 * Remember the IP options, if any.
984 if (!inc->inc_isipv6)
986 ipopts = ip_srcroute(m);
989 * See if we already have an entry for this connection.
990 * If we do, resend the SYN,ACK, and reset the retransmit timer.
993 * The syncache should be re-initialized with the contents
994 * of the new SYN which may have different options.
996 sc = syncache_lookup(inc, &sch);
998 tcpstat.tcps_sc_dupsyn++;
1001 * If we were remembering a previous source route,
1002 * forget it and use the new one we've been given.
1005 m_free(sc->sc_ipopts);
1006 sc->sc_ipopts = ipopts;
1009 * Update timestamp if present.
1011 if (sc->sc_flags & SCF_TIMESTAMP)
1012 sc->sc_tsrecent = to->to_tsval;
1014 /* Just update the TOF_SACK_PERMITTED for now. */
1015 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1016 sc->sc_flags |= SCF_SACK_PERMITTED;
1018 sc->sc_flags &= ~SCF_SACK_PERMITTED;
1021 * PCB may have changed, pick up new values.
1024 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
1025 if (syncache_respond(sc, m) == 0) {
1026 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot],
1028 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot);
1029 tcpstat.tcps_sndacks++;
1030 tcpstat.tcps_sndtotal++;
1037 * Fill in the syncache values.
1039 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO);
1040 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
1041 sc->sc_ipopts = ipopts;
1042 sc->sc_inc.inc_fport = inc->inc_fport;
1043 sc->sc_inc.inc_lport = inc->inc_lport;
1046 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1047 if (inc->inc_isipv6) {
1048 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1049 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1050 sc->sc_route6.ro_rt = NULL;
1054 sc->sc_inc.inc_faddr = inc->inc_faddr;
1055 sc->sc_inc.inc_laddr = inc->inc_laddr;
1056 sc->sc_route.ro_rt = NULL;
1058 sc->sc_irs = th->th_seq;
1060 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0;
1062 sc->sc_iss = syncookie_generate(sc);
1064 sc->sc_iss = karc4random();
1066 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */
1067 win = ssb_space(&so->so_rcv);
1069 win = imin(win, TCP_MAXWIN);
1072 if (tcp_do_rfc1323) {
1074 * A timestamp received in a SYN makes
1075 * it ok to send timestamp requests and replies.
1077 if (to->to_flags & TOF_TS) {
1078 sc->sc_tsrecent = to->to_tsval;
1079 sc->sc_flags |= SCF_TIMESTAMP;
1081 if (to->to_flags & TOF_SCALE) {
1082 int wscale = TCP_MIN_WINSHIFT;
1084 /* Compute proper scaling value from buffer space */
1085 while (wscale < TCP_MAX_WINSHIFT &&
1086 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) {
1089 sc->sc_request_r_scale = wscale;
1090 sc->sc_requested_s_scale = to->to_requested_s_scale;
1091 sc->sc_flags |= SCF_WINSCALE;
1094 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1095 sc->sc_flags |= SCF_SACK_PERMITTED;
1096 if (tp->t_flags & TF_NOOPT)
1097 sc->sc_flags = SCF_NOOPT;
1098 #ifdef TCP_SIGNATURE
1100 * If listening socket requested TCP digests, and received SYN
1101 * contains the option, flag this in the syncache so that
1102 * syncache_respond() will do the right thing with the SYN+ACK.
1103 * XXX Currently we always record the option by default and will
1104 * attempt to use it in syncache_respond().
1106 if (to->to_flags & TOF_SIGNATURE)
1107 sc->sc_flags = SCF_SIGNATURE;
1108 #endif /* TCP_SIGNATURE */
1110 if (syncache_respond(sc, m) == 0) {
1111 syncache_insert(sc, sch);
1112 tcpstat.tcps_sndacks++;
1113 tcpstat.tcps_sndtotal++;
1116 tcpstat.tcps_sc_dropped++;
1123 syncache_respond(struct syncache *sc, struct mbuf *m)
1127 u_int16_t tlen, hlen, mssopt;
1128 struct ip *ip = NULL;
1131 struct ip6_hdr *ip6 = NULL;
1133 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1135 const boolean_t isipv6 = FALSE;
1139 rt = tcp_rtlookup6(&sc->sc_inc);
1141 mssopt = rt->rt_ifp->if_mtu -
1142 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1144 mssopt = tcp_v6mssdflt;
1145 hlen = sizeof(struct ip6_hdr);
1147 rt = tcp_rtlookup(&sc->sc_inc);
1149 mssopt = rt->rt_ifp->if_mtu -
1150 (sizeof(struct ip) + sizeof(struct tcphdr));
1152 mssopt = tcp_mssdflt;
1153 hlen = sizeof(struct ip);
1156 /* Compute the size of the TCP options. */
1157 if (sc->sc_flags & SCF_NOOPT) {
1160 optlen = TCPOLEN_MAXSEG +
1161 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) +
1162 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) +
1163 ((sc->sc_flags & SCF_SACK_PERMITTED) ?
1164 TCPOLEN_SACK_PERMITTED_ALIGNED : 0);
1165 #ifdef TCP_SIGNATURE
1166 optlen += ((sc->sc_flags & SCF_SIGNATURE) ?
1167 (TCPOLEN_SIGNATURE + 2) : 0);
1168 #endif /* TCP_SIGNATURE */
1170 tlen = hlen + sizeof(struct tcphdr) + optlen;
1174 * assume that the entire packet will fit in a header mbuf
1176 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small"));
1179 * XXX shouldn't this reuse the mbuf if possible ?
1180 * Create the IP+TCP header from scratch.
1185 m = m_gethdr(MB_DONTWAIT, MT_HEADER);
1188 m->m_data += max_linkhdr;
1190 m->m_pkthdr.len = tlen;
1191 m->m_pkthdr.rcvif = NULL;
1194 ip6 = mtod(m, struct ip6_hdr *);
1195 ip6->ip6_vfc = IPV6_VERSION;
1196 ip6->ip6_nxt = IPPROTO_TCP;
1197 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1198 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1199 ip6->ip6_plen = htons(tlen - hlen);
1200 /* ip6_hlim is set after checksum */
1201 /* ip6_flow = ??? */
1203 th = (struct tcphdr *)(ip6 + 1);
1205 ip = mtod(m, struct ip *);
1206 ip->ip_v = IPVERSION;
1207 ip->ip_hl = sizeof(struct ip) >> 2;
1212 ip->ip_p = IPPROTO_TCP;
1213 ip->ip_src = sc->sc_inc.inc_laddr;
1214 ip->ip_dst = sc->sc_inc.inc_faddr;
1215 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */
1216 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */
1219 * See if we should do MTU discovery. Route lookups are
1220 * expensive, so we will only unset the DF bit if:
1222 * 1) path_mtu_discovery is disabled
1223 * 2) the SCF_UNREACH flag has been set
1225 if (path_mtu_discovery
1226 && ((sc->sc_flags & SCF_UNREACH) == 0)) {
1227 ip->ip_off |= IP_DF;
1230 th = (struct tcphdr *)(ip + 1);
1232 th->th_sport = sc->sc_inc.inc_lport;
1233 th->th_dport = sc->sc_inc.inc_fport;
1235 th->th_seq = htonl(sc->sc_iss);
1236 th->th_ack = htonl(sc->sc_irs + 1);
1237 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1239 th->th_flags = TH_SYN | TH_ACK;
1240 th->th_win = htons(sc->sc_wnd);
1243 /* Tack on the TCP options. */
1246 optp = (u_int8_t *)(th + 1);
1247 *optp++ = TCPOPT_MAXSEG;
1248 *optp++ = TCPOLEN_MAXSEG;
1249 *optp++ = (mssopt >> 8) & 0xff;
1250 *optp++ = mssopt & 0xff;
1252 if (sc->sc_flags & SCF_WINSCALE) {
1253 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
1254 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
1255 sc->sc_request_r_scale);
1259 if (sc->sc_flags & SCF_TIMESTAMP) {
1260 u_int32_t *lp = (u_int32_t *)(optp);
1262 /* Form timestamp option as shown in appendix A of RFC 1323. */
1263 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1264 *lp++ = htonl(ticks);
1265 *lp = htonl(sc->sc_tsrecent);
1266 optp += TCPOLEN_TSTAMP_APPA;
1269 #ifdef TCP_SIGNATURE
1271 * Handle TCP-MD5 passive opener response.
1273 if (sc->sc_flags & SCF_SIGNATURE) {
1274 u_int8_t *bp = optp;
1277 *bp++ = TCPOPT_SIGNATURE;
1278 *bp++ = TCPOLEN_SIGNATURE;
1279 for (i = 0; i < TCP_SIGLEN; i++)
1281 tcpsignature_compute(m, 0, optlen,
1282 optp + 2, IPSEC_DIR_OUTBOUND);
1285 optp += TCPOLEN_SIGNATURE + 2;
1287 #endif /* TCP_SIGNATURE */
1289 if (sc->sc_flags & SCF_SACK_PERMITTED) {
1290 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED);
1291 optp += TCPOLEN_SACK_PERMITTED_ALIGNED;
1296 struct route_in6 *ro6 = &sc->sc_route6;
1299 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
1300 ip6->ip6_hlim = in6_selecthlim(NULL,
1301 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL);
1302 error = ip6_output(m, NULL, ro6, 0, NULL, NULL,
1303 sc->sc_tp->t_inpcb);
1305 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1306 htons(tlen - hlen + IPPROTO_TCP));
1307 m->m_pkthdr.csum_flags = CSUM_TCP;
1308 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1309 error = ip_output(m, sc->sc_ipopts, &sc->sc_route,
1310 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb);
1318 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .|
1320 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .|
1322 * (A): peer mss index
1326 * The values below are chosen to minimize the size of the tcp_secret
1327 * table, as well as providing roughly a 16 second lifetime for the cookie.
1330 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */
1331 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */
1333 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1)
1334 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS)
1335 #define SYNCOOKIE_TIMEOUT \
1336 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT))
1337 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK)
1340 u_int32_t ts_secbits[4];
1342 } tcp_secret[SYNCOOKIE_NSECRETS];
1344 static int tcp_msstab[] = { 0, 536, 1460, 8960 };
1346 static MD5_CTX syn_ctx;
1348 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v))
1351 u_int32_t laddr, faddr;
1352 u_int32_t secbits[4];
1353 u_int16_t lport, fport;
1357 CTASSERT(sizeof(struct md5_add) == 28);
1361 * Consider the problem of a recreated (and retransmitted) cookie. If the
1362 * original SYN was accepted, the connection is established. The second
1363 * SYN is inflight, and if it arrives with an ISN that falls within the
1364 * receive window, the connection is killed.
1366 * However, since cookies have other problems, this may not be worth
1371 syncookie_generate(struct syncache *sc)
1373 u_int32_t md5_buffer[4];
1378 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1380 const boolean_t isipv6 = FALSE;
1383 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK;
1384 if (tcp_secret[idx].ts_expire < ticks) {
1385 for (i = 0; i < 4; i++)
1386 tcp_secret[idx].ts_secbits[i] = karc4random();
1387 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT;
1389 for (data = NELEM(tcp_msstab) - 1; data > 0; data--)
1390 if (tcp_msstab[data] <= sc->sc_peer_mss)
1392 data = (data << SYNCOOKIE_WNDBITS) | idx;
1393 data ^= sc->sc_irs; /* peer's iss */
1396 MD5Add(sc->sc_inc.inc6_laddr);
1397 MD5Add(sc->sc_inc.inc6_faddr);
1401 add.laddr = sc->sc_inc.inc_laddr.s_addr;
1402 add.faddr = sc->sc_inc.inc_faddr.s_addr;
1404 add.lport = sc->sc_inc.inc_lport;
1405 add.fport = sc->sc_inc.inc_fport;
1406 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1407 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1408 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1409 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1411 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1412 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK);
1416 static struct syncache *
1417 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so)
1419 u_int32_t md5_buffer[4];
1420 struct syncache *sc;
1425 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */
1426 idx = data & SYNCOOKIE_WNDMASK;
1427 if (tcp_secret[idx].ts_expire < ticks ||
1428 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks)
1432 if (inc->inc_isipv6) {
1433 MD5Add(inc->inc6_laddr);
1434 MD5Add(inc->inc6_faddr);
1440 add.laddr = inc->inc_laddr.s_addr;
1441 add.faddr = inc->inc_faddr.s_addr;
1443 add.lport = inc->inc_lport;
1444 add.fport = inc->inc_fport;
1445 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1446 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1447 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1448 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1450 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1451 data ^= md5_buffer[0];
1452 if (data & ~SYNCOOKIE_DATAMASK)
1454 data = data >> SYNCOOKIE_WNDBITS;
1457 * Fill in the syncache values.
1458 * XXX duplicate code from syncache_add
1460 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO);
1461 sc->sc_ipopts = NULL;
1462 sc->sc_inc.inc_fport = inc->inc_fport;
1463 sc->sc_inc.inc_lport = inc->inc_lport;
1465 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1466 if (inc->inc_isipv6) {
1467 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1468 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1469 sc->sc_route6.ro_rt = NULL;
1473 sc->sc_inc.inc_faddr = inc->inc_faddr;
1474 sc->sc_inc.inc_laddr = inc->inc_laddr;
1475 sc->sc_route.ro_rt = NULL;
1477 sc->sc_irs = th->th_seq - 1;
1478 sc->sc_iss = th->th_ack - 1;
1479 wnd = ssb_space(&so->so_rcv);
1481 wnd = imin(wnd, TCP_MAXWIN);
1485 sc->sc_peer_mss = tcp_msstab[data];