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 * 4 retransmits corresponds to a timeout of (3 + 3 + 3 + 3 + 3 == 15) seconds
155 * or (1 + 1 + 2 + 4 + 8 == 16) seconds if RFC6298 is used, the odds are that
156 * the user has given up attempting to connect by then.
158 #define SYNCACHE_MAXREXMTS 4
160 /* Arbitrary values */
161 #define TCP_SYNCACHE_HASHSIZE 512
162 #define TCP_SYNCACHE_BUCKETLIMIT 30
164 struct netmsg_sc_timer {
165 struct netmsg_base base;
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 {
185 static struct tcp_syncache tcp_syncache;
187 TAILQ_HEAD(syncache_list, syncache);
189 struct tcp_syncache_percpu {
190 struct syncache_head *hashbase;
192 struct syncache_list timerq[SYNCACHE_MAXREXMTS + 1];
193 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1];
194 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1];
196 static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU];
198 static struct lwkt_port syncache_null_rport;
200 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
202 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD,
203 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
205 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD,
206 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
210 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
211 &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
214 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD,
215 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
217 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
218 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
220 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
222 #define SYNCACHE_HASH(inc, mask) \
223 ((tcp_syncache.hash_secret ^ \
224 (inc)->inc_faddr.s_addr ^ \
225 ((inc)->inc_faddr.s_addr >> 16) ^ \
226 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
228 #define SYNCACHE_HASH6(inc, mask) \
229 ((tcp_syncache.hash_secret ^ \
230 (inc)->inc6_faddr.s6_addr32[0] ^ \
231 (inc)->inc6_faddr.s6_addr32[3] ^ \
232 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
234 #define ENDPTS_EQ(a, b) ( \
235 (a)->ie_fport == (b)->ie_fport && \
236 (a)->ie_lport == (b)->ie_lport && \
237 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
238 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
241 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
244 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu,
245 struct syncache *sc, int slot)
251 * Record that SYN|ACK was lost.
252 * Needed by RFC3390 and RFC6298.
254 sc->sc_flags |= SCF_SYN_WASLOST;
256 sc->sc_rxtslot = slot;
259 rto = TCPTV_RTOBASE * tcp_syn_backoff_low[slot];
261 rto = TCPTV_RTOBASE * tcp_syn_backoff[slot];
262 sc->sc_rxttime = ticks + rto;
264 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq);
265 if (!callout_active(&syncache_percpu->tt_timerq[slot])) {
266 callout_reset(&syncache_percpu->tt_timerq[slot], rto,
267 syncache_timer, &syncache_percpu->mrec[slot]);
272 syncache_free(struct syncache *sc)
276 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
278 const boolean_t isipv6 = FALSE;
282 m_free(sc->sc_ipopts);
284 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt;
287 * If this is the only reference to a protocol-cloned
288 * route, remove it immediately.
290 if ((rt->rt_flags & RTF_WASCLONED) && rt->rt_refcnt == 1)
291 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
292 rt_mask(rt), rt->rt_flags, NULL);
295 kfree(sc, M_SYNCACHE);
303 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
304 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
305 tcp_syncache.cache_limit =
306 tcp_syncache.hashsize * tcp_syncache.bucket_limit;
307 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
308 tcp_syncache.hash_secret = karc4random();
310 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
311 &tcp_syncache.hashsize);
312 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
313 &tcp_syncache.cache_limit);
314 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
315 &tcp_syncache.bucket_limit);
316 if (!powerof2(tcp_syncache.hashsize)) {
317 kprintf("WARNING: syncache hash size is not a power of 2.\n");
318 tcp_syncache.hashsize = 512; /* safe default */
320 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
322 lwkt_initport_replyonly_null(&syncache_null_rport);
324 for (cpu = 0; cpu < ncpus2; cpu++) {
325 struct tcp_syncache_percpu *syncache_percpu;
327 syncache_percpu = &tcp_syncache_percpu[cpu];
328 /* Allocate the hash table. */
329 syncache_percpu->hashbase = kmalloc(tcp_syncache.hashsize * sizeof(struct syncache_head),
330 M_SYNCACHE, M_WAITOK);
332 /* Initialize the hash buckets. */
333 for (i = 0; i < tcp_syncache.hashsize; i++) {
334 struct syncache_head *bucket;
336 bucket = &syncache_percpu->hashbase[i];
337 TAILQ_INIT(&bucket->sch_bucket);
338 bucket->sch_length = 0;
341 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) {
342 /* Initialize the timer queues. */
343 TAILQ_INIT(&syncache_percpu->timerq[i]);
344 callout_init_mp(&syncache_percpu->tt_timerq[i]);
346 syncache_percpu->mrec[i].slot = i;
347 syncache_percpu->mrec[i].port = cpu_portfn(cpu);
348 syncache_percpu->mrec[i].msg.nm_mrec =
349 &syncache_percpu->mrec[i];
350 netmsg_init(&syncache_percpu->mrec[i].msg.base,
351 NULL, &syncache_null_rport,
352 0, syncache_timer_handler);
358 syncache_insert(struct syncache *sc, struct syncache_head *sch)
360 struct tcp_syncache_percpu *syncache_percpu;
361 struct syncache *sc2;
364 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
367 * Make sure that we don't overflow the per-bucket
368 * limit or the total cache size limit.
370 if (sch->sch_length >= tcp_syncache.bucket_limit) {
372 * The bucket is full, toss the oldest element.
374 sc2 = TAILQ_FIRST(&sch->sch_bucket);
375 sc2->sc_tp->ts_recent = ticks;
376 syncache_drop(sc2, sch);
377 tcpstat.tcps_sc_bucketoverflow++;
378 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) {
380 * The cache is full. Toss the oldest entry in the
381 * entire cache. This is the front entry in the
382 * first non-empty timer queue with the largest
385 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
386 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]);
387 while (sc2 && (sc2->sc_flags & SCF_MARKER))
388 sc2 = TAILQ_NEXT(sc2, sc_timerq);
392 sc2->sc_tp->ts_recent = ticks;
393 syncache_drop(sc2, NULL);
394 tcpstat.tcps_sc_cacheoverflow++;
397 /* Initialize the entry's timer. */
398 syncache_timeout(syncache_percpu, sc, 0);
400 /* Put it into the bucket. */
401 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash);
403 syncache_percpu->cache_count++;
404 tcpstat.tcps_sc_added++;
408 syncache_destroy(struct tcpcb *tp)
410 struct tcp_syncache_percpu *syncache_percpu;
411 struct syncache_head *bucket;
415 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
418 for (i = 0; i < tcp_syncache.hashsize; i++) {
419 bucket = &syncache_percpu->hashbase[i];
420 TAILQ_FOREACH(sc, &bucket->sch_bucket, sc_hash) {
428 syncache_drop(struct syncache *sc, struct syncache_head *sch)
430 struct tcp_syncache_percpu *syncache_percpu;
432 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
434 const boolean_t isipv6 = FALSE;
437 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
441 sch = &syncache_percpu->hashbase[
442 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)];
444 sch = &syncache_percpu->hashbase[
445 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)];
449 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
451 syncache_percpu->cache_count--;
460 * Remove the entry from the syncache timer/timeout queue. Note
461 * that we do not try to stop any running timer since we do not know
462 * whether the timer's message is in-transit or not. Since timeouts
463 * are fairly long, taking an unneeded callout does not detrimentally
464 * effect performance.
466 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq);
472 * Place a timeout message on the TCP thread's message queue.
473 * This routine runs in soft interrupt context.
475 * An invariant is for this routine to be called, the callout must
476 * have been active. Note that the callout is not deactivated until
477 * after the message has been processed in syncache_timer_handler() below.
480 syncache_timer(void *p)
482 struct netmsg_sc_timer *msg = p;
484 lwkt_sendmsg(msg->nm_mrec->port, &msg->base.lmsg);
488 * Service a timer message queued by timer expiration.
489 * This routine runs in the TCP protocol thread.
491 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
492 * If we have retransmitted an entry the maximum number of times, expire it.
494 * When we finish processing timed-out entries, we restart the timer if there
495 * are any entries still on the queue and deactivate it otherwise. Only after
496 * a timer has been deactivated here can it be restarted by syncache_timeout().
499 syncache_timer_handler(netmsg_t msg)
501 struct tcp_syncache_percpu *syncache_percpu;
503 struct syncache marker;
504 struct syncache_list *list;
508 slot = ((struct netmsg_sc_timer *)msg)->nm_mrec->slot;
509 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
511 list = &syncache_percpu->timerq[slot];
514 * Use a marker to keep our place in the scan. syncache_drop()
515 * can block and cause any next pointer we cache to become stale.
517 marker.sc_flags = SCF_MARKER;
518 TAILQ_INSERT_HEAD(list, &marker, sc_timerq);
520 while ((sc = TAILQ_NEXT(&marker, sc_timerq)) != NULL) {
524 TAILQ_REMOVE(list, &marker, sc_timerq);
525 TAILQ_INSERT_AFTER(list, sc, &marker, sc_timerq);
527 if (sc->sc_flags & SCF_MARKER)
530 if (ticks < sc->sc_rxttime)
531 break; /* finished because timerq sorted by time */
532 if (sc->sc_tp == NULL) {
533 syncache_drop(sc, NULL);
534 tcpstat.tcps_sc_stale++;
537 inp = sc->sc_tp->t_inpcb;
538 if (slot == SYNCACHE_MAXREXMTS ||
539 slot >= tcp_syncache.rexmt_limit ||
541 inp->inp_gencnt != sc->sc_inp_gencnt) {
542 syncache_drop(sc, NULL);
543 tcpstat.tcps_sc_stale++;
547 * syncache_respond() may call back into the syncache to
548 * to modify another entry, so do not obtain the next
549 * entry on the timer chain until it has completed.
551 syncache_respond(sc, NULL);
552 tcpstat.tcps_sc_retransmitted++;
553 TAILQ_REMOVE(list, sc, sc_timerq);
554 syncache_timeout(syncache_percpu, sc, slot + 1);
556 TAILQ_REMOVE(list, &marker, sc_timerq);
559 callout_reset(&syncache_percpu->tt_timerq[slot],
560 sc->sc_rxttime - ticks, syncache_timer,
561 &syncache_percpu->mrec[slot]);
563 callout_deactivate(&syncache_percpu->tt_timerq[slot]);
565 lwkt_replymsg(&msg->base.lmsg, 0);
569 * Find an entry in the syncache.
572 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
574 struct tcp_syncache_percpu *syncache_percpu;
576 struct syncache_head *sch;
578 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
580 if (inc->inc_isipv6) {
581 sch = &syncache_percpu->hashbase[
582 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
584 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash)
585 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
590 sch = &syncache_percpu->hashbase[
591 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
593 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
595 if (sc->sc_inc.inc_isipv6)
598 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
606 * This function is called when we get a RST for a
607 * non-existent connection, so that we can see if the
608 * connection is in the syn cache. If it is, zap it.
611 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
614 struct syncache_head *sch;
616 sc = syncache_lookup(inc, &sch);
621 * If the RST bit is set, check the sequence number to see
622 * if this is a valid reset segment.
624 * In all states except SYN-SENT, all reset (RST) segments
625 * are validated by checking their SEQ-fields. A reset is
626 * valid if its sequence number is in the window.
628 * The sequence number in the reset segment is normally an
629 * echo of our outgoing acknowlegement numbers, but some hosts
630 * send a reset with the sequence number at the rightmost edge
631 * of our receive window, and we have to handle this case.
633 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
634 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
635 syncache_drop(sc, sch);
636 tcpstat.tcps_sc_reset++;
641 syncache_badack(struct in_conninfo *inc)
644 struct syncache_head *sch;
646 sc = syncache_lookup(inc, &sch);
648 syncache_drop(sc, sch);
649 tcpstat.tcps_sc_badack++;
654 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
657 struct syncache_head *sch;
659 /* we are called at splnet() here */
660 sc = syncache_lookup(inc, &sch);
664 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
665 if (ntohl(th->th_seq) != sc->sc_iss)
669 * If we've rertransmitted 3 times and this is our second error,
670 * we remove the entry. Otherwise, we allow it to continue on.
671 * This prevents us from incorrectly nuking an entry during a
672 * spurious network outage.
676 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) {
677 sc->sc_flags |= SCF_UNREACH;
680 syncache_drop(sc, sch);
681 tcpstat.tcps_sc_unreach++;
685 * Build a new TCP socket structure from a syncache entry.
687 * This is called from the context of the SYN+ACK
689 static struct socket *
690 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
692 struct inpcb *inp = NULL, *linp;
694 struct tcpcb *tp, *ltp;
697 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
699 const boolean_t isipv6 = FALSE;
701 struct sockaddr_in sin_faddr;
702 struct sockaddr_in6 sin6_faddr;
703 struct sockaddr *faddr;
706 faddr = (struct sockaddr *)&sin6_faddr;
707 sin6_faddr.sin6_family = AF_INET6;
708 sin6_faddr.sin6_len = sizeof(sin6_faddr);
709 sin6_faddr.sin6_addr = sc->sc_inc.inc6_faddr;
710 sin6_faddr.sin6_port = sc->sc_inc.inc_fport;
711 sin6_faddr.sin6_flowinfo = sin6_faddr.sin6_scope_id = 0;
713 faddr = (struct sockaddr *)&sin_faddr;
714 sin_faddr.sin_family = AF_INET;
715 sin_faddr.sin_len = sizeof(sin_faddr);
716 sin_faddr.sin_addr = sc->sc_inc.inc_faddr;
717 sin_faddr.sin_port = sc->sc_inc.inc_fport;
718 bzero(sin_faddr.sin_zero, sizeof(sin_faddr.sin_zero));
722 * Ok, create the full blown connection, and set things up
723 * as they would have been set up if we had created the
724 * connection when the SYN arrived. If we can't create
725 * the connection, abort it.
727 * Set the protocol processing port for the socket to the current
728 * port (that the connection came in on).
730 so = sonewconn_faddr(lso, SS_ISCONNECTED, faddr);
733 * Drop the connection; we will send a RST if the peer
734 * retransmits the ACK,
736 tcpstat.tcps_listendrop++;
741 * Insert new socket into hash list.
744 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
746 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
749 inp->inp_vflag &= ~INP_IPV6;
750 inp->inp_vflag |= INP_IPV4;
751 inp->inp_flags &= ~IN6P_IPV6_V6ONLY;
753 inp->inp_laddr = sc->sc_inc.inc_laddr;
755 inp->inp_lport = sc->sc_inc.inc_lport;
756 if (in_pcbinsporthash(inp) != 0) {
758 * Undo the assignments above if we failed to
759 * put the PCB on the hash lists.
762 inp->in6p_laddr = kin6addr_any;
764 inp->inp_laddr.s_addr = INADDR_ANY;
770 /* copy old policy into new socket's */
771 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp))
772 kprintf("syncache_expand: could not copy policy\n");
775 struct in6_addr laddr6;
777 * Inherit socket options from the listening socket.
778 * Note that in6p_inputopts are not (and should not be)
779 * copied, since it stores previously received options and is
780 * used to detect if each new option is different than the
781 * previous one and hence should be passed to a user.
782 * If we copied in6p_inputopts, a user would not be able to
783 * receive options just after calling the accept system call.
785 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS;
786 if (linp->in6p_outputopts)
787 inp->in6p_outputopts =
788 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT);
789 inp->in6p_route = sc->sc_route6;
790 sc->sc_route6.ro_rt = NULL;
792 laddr6 = inp->in6p_laddr;
793 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
794 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
795 if (in6_pcbconnect(inp, faddr, &thread0)) {
796 inp->in6p_laddr = laddr6;
800 struct in_addr laddr;
802 inp->inp_options = ip_srcroute(m);
803 if (inp->inp_options == NULL) {
804 inp->inp_options = sc->sc_ipopts;
805 sc->sc_ipopts = NULL;
807 inp->inp_route = sc->sc_route;
808 sc->sc_route.ro_rt = NULL;
810 laddr = inp->inp_laddr;
811 if (inp->inp_laddr.s_addr == INADDR_ANY)
812 inp->inp_laddr = sc->sc_inc.inc_laddr;
813 if (in_pcbconnect(inp, faddr, &thread0)) {
814 inp->inp_laddr = laddr;
820 * The current port should be in the context of the SYN+ACK and
821 * so should match the tcp address port.
823 * XXX we may be running on the netisr thread instead of a tcp
824 * thread, in which case port will not match
825 * curthread->td_msgport.
828 port = tcp6_addrport();
830 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport,
831 inp->inp_laddr.s_addr, inp->inp_lport);
833 if (port != &curthread->td_msgport) {
835 kprintf("TCP PORT MISMATCH %p vs %p\n",
836 port, &curthread->td_msgport);
838 /*KKASSERT(port == &curthread->td_msgport);*/
841 tp->t_state = TCPS_SYN_RECEIVED;
842 tp->iss = sc->sc_iss;
843 tp->irs = sc->sc_irs;
846 tp->snd_wl1 = sc->sc_irs;
847 tp->rcv_up = sc->sc_irs + 1;
848 tp->rcv_wnd = sc->sc_wnd;
849 tp->rcv_adv += tp->rcv_wnd;
851 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY);
852 if (sc->sc_flags & SCF_NOOPT)
853 tp->t_flags |= TF_NOOPT;
854 if (sc->sc_flags & SCF_WINSCALE) {
855 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE;
856 tp->requested_s_scale = sc->sc_requested_s_scale;
857 tp->request_r_scale = sc->sc_request_r_scale;
859 if (sc->sc_flags & SCF_TIMESTAMP) {
860 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP;
861 tp->ts_recent = sc->sc_tsrecent;
862 tp->ts_recent_age = ticks;
864 if (sc->sc_flags & SCF_SACK_PERMITTED)
865 tp->t_flags |= TF_SACK_PERMITTED;
866 if (sc->sc_flags & SCF_SYN_WASLOST)
867 tp->t_flags |= TF_SYN_WASLOST;
870 if (sc->sc_flags & SCF_SIGNATURE)
871 tp->t_flags |= TF_SIGNATURE;
872 #endif /* TCP_SIGNATURE */
874 tcp_mss(tp, sc->sc_peer_mss);
877 * Inherit some properties from the listen socket
879 ltp = intotcpcb(linp);
880 tp->t_keepinit = ltp->t_keepinit;
881 tp->t_keepidle = ltp->t_keepidle;
882 tp->t_keepintvl = ltp->t_keepintvl;
883 tp->t_keepcnt = ltp->t_keepcnt;
884 tp->t_maxidle = ltp->t_maxidle;
886 tcp_create_timermsg(tp, port);
887 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
889 tcpstat.tcps_accepts++;
899 * This function gets called when we receive an ACK for a
900 * socket in the LISTEN state. We look up the connection
901 * in the syncache, and if its there, we pull it out of
902 * the cache and turn it into a full-blown connection in
903 * the SYN-RECEIVED state.
906 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop,
910 struct syncache_head *sch;
913 sc = syncache_lookup(inc, &sch);
916 * There is no syncache entry, so see if this ACK is
917 * a returning syncookie. To do this, first:
918 * A. See if this socket has had a syncache entry dropped in
919 * the past. We don't want to accept a bogus syncookie
920 * if we've never received a SYN.
921 * B. check that the syncookie is valid. If it is, then
922 * cobble up a fake syncache entry, and return.
926 sc = syncookie_lookup(inc, th, *sop);
930 tcpstat.tcps_sc_recvcookie++;
934 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
936 if (th->th_ack != sc->sc_iss + 1)
939 so = syncache_socket(sc, *sop, m);
943 /* XXXjlemon check this - is this correct? */
944 tcp_respond(NULL, m, m, th,
945 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK);
947 m_freem(m); /* XXX only needed for above */
948 tcpstat.tcps_sc_aborted++;
950 tcpstat.tcps_sc_completed++;
955 syncache_drop(sc, sch);
961 * Given a LISTEN socket and an inbound SYN request, add
962 * this to the syn cache, and send back a segment:
963 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
966 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
967 * Doing so would require that we hold onto the data and deliver it
968 * to the application. However, if we are the target of a SYN-flood
969 * DoS attack, an attacker could send data which would eventually
970 * consume all available buffer space if it were ACKed. By not ACKing
971 * the data, we avoid this DoS scenario.
974 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
975 struct socket *so, struct mbuf *m)
977 struct tcp_syncache_percpu *syncache_percpu;
979 struct syncache *sc = NULL;
980 struct syncache_head *sch;
981 struct mbuf *ipopts = NULL;
984 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
988 * Remember the IP options, if any.
991 if (!inc->inc_isipv6)
993 ipopts = ip_srcroute(m);
996 * See if we already have an entry for this connection.
997 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1000 * The syncache should be re-initialized with the contents
1001 * of the new SYN which may have different options.
1003 sc = syncache_lookup(inc, &sch);
1005 tcpstat.tcps_sc_dupsyn++;
1008 * If we were remembering a previous source route,
1009 * forget it and use the new one we've been given.
1012 m_free(sc->sc_ipopts);
1013 sc->sc_ipopts = ipopts;
1016 * Update timestamp if present.
1018 if (sc->sc_flags & SCF_TIMESTAMP)
1019 sc->sc_tsrecent = to->to_tsval;
1021 /* Just update the TOF_SACK_PERMITTED for now. */
1022 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1023 sc->sc_flags |= SCF_SACK_PERMITTED;
1025 sc->sc_flags &= ~SCF_SACK_PERMITTED;
1028 * PCB may have changed, pick up new values.
1031 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
1032 if (syncache_respond(sc, m) == 0) {
1033 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot],
1035 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot);
1036 tcpstat.tcps_sndacks++;
1037 tcpstat.tcps_sndtotal++;
1043 * Fill in the syncache values.
1045 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO);
1046 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
1047 sc->sc_ipopts = ipopts;
1048 sc->sc_inc.inc_fport = inc->inc_fport;
1049 sc->sc_inc.inc_lport = inc->inc_lport;
1052 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1053 if (inc->inc_isipv6) {
1054 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1055 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1056 sc->sc_route6.ro_rt = NULL;
1060 sc->sc_inc.inc_faddr = inc->inc_faddr;
1061 sc->sc_inc.inc_laddr = inc->inc_laddr;
1062 sc->sc_route.ro_rt = NULL;
1064 sc->sc_irs = th->th_seq;
1066 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0;
1068 sc->sc_iss = syncookie_generate(sc);
1070 sc->sc_iss = karc4random();
1072 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */
1073 win = ssb_space(&so->so_rcv);
1075 win = imin(win, TCP_MAXWIN);
1078 if (tcp_do_rfc1323) {
1080 * A timestamp received in a SYN makes
1081 * it ok to send timestamp requests and replies.
1083 if (to->to_flags & TOF_TS) {
1084 sc->sc_tsrecent = to->to_tsval;
1085 sc->sc_flags |= SCF_TIMESTAMP;
1087 if (to->to_flags & TOF_SCALE) {
1088 int wscale = TCP_MIN_WINSHIFT;
1090 /* Compute proper scaling value from buffer space */
1091 while (wscale < TCP_MAX_WINSHIFT &&
1092 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) {
1095 sc->sc_request_r_scale = wscale;
1096 sc->sc_requested_s_scale = to->to_requested_s_scale;
1097 sc->sc_flags |= SCF_WINSCALE;
1100 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1101 sc->sc_flags |= SCF_SACK_PERMITTED;
1102 if (tp->t_flags & TF_NOOPT)
1103 sc->sc_flags = SCF_NOOPT;
1104 #ifdef TCP_SIGNATURE
1106 * If listening socket requested TCP digests, and received SYN
1107 * contains the option, flag this in the syncache so that
1108 * syncache_respond() will do the right thing with the SYN+ACK.
1109 * XXX Currently we always record the option by default and will
1110 * attempt to use it in syncache_respond().
1112 if (to->to_flags & TOF_SIGNATURE)
1113 sc->sc_flags = SCF_SIGNATURE;
1114 #endif /* TCP_SIGNATURE */
1116 if (syncache_respond(sc, m) == 0) {
1117 syncache_insert(sc, sch);
1118 tcpstat.tcps_sndacks++;
1119 tcpstat.tcps_sndtotal++;
1122 tcpstat.tcps_sc_dropped++;
1128 syncache_respond(struct syncache *sc, struct mbuf *m)
1132 u_int16_t tlen, hlen, mssopt;
1133 struct ip *ip = NULL;
1136 struct ip6_hdr *ip6 = NULL;
1138 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1140 const boolean_t isipv6 = FALSE;
1144 rt = tcp_rtlookup6(&sc->sc_inc);
1146 mssopt = rt->rt_ifp->if_mtu -
1147 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1149 mssopt = tcp_v6mssdflt;
1150 hlen = sizeof(struct ip6_hdr);
1152 rt = tcp_rtlookup(&sc->sc_inc);
1154 mssopt = rt->rt_ifp->if_mtu -
1155 (sizeof(struct ip) + sizeof(struct tcphdr));
1157 mssopt = tcp_mssdflt;
1158 hlen = sizeof(struct ip);
1161 /* Compute the size of the TCP options. */
1162 if (sc->sc_flags & SCF_NOOPT) {
1165 optlen = TCPOLEN_MAXSEG +
1166 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) +
1167 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) +
1168 ((sc->sc_flags & SCF_SACK_PERMITTED) ?
1169 TCPOLEN_SACK_PERMITTED_ALIGNED : 0);
1170 #ifdef TCP_SIGNATURE
1171 optlen += ((sc->sc_flags & SCF_SIGNATURE) ?
1172 (TCPOLEN_SIGNATURE + 2) : 0);
1173 #endif /* TCP_SIGNATURE */
1175 tlen = hlen + sizeof(struct tcphdr) + optlen;
1179 * assume that the entire packet will fit in a header mbuf
1181 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small"));
1184 * XXX shouldn't this reuse the mbuf if possible ?
1185 * Create the IP+TCP header from scratch.
1190 m = m_gethdr(MB_DONTWAIT, MT_HEADER);
1193 m->m_data += max_linkhdr;
1195 m->m_pkthdr.len = tlen;
1196 m->m_pkthdr.rcvif = NULL;
1199 ip6 = mtod(m, struct ip6_hdr *);
1200 ip6->ip6_vfc = IPV6_VERSION;
1201 ip6->ip6_nxt = IPPROTO_TCP;
1202 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1203 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1204 ip6->ip6_plen = htons(tlen - hlen);
1205 /* ip6_hlim is set after checksum */
1206 /* ip6_flow = ??? */
1208 th = (struct tcphdr *)(ip6 + 1);
1210 ip = mtod(m, struct ip *);
1211 ip->ip_v = IPVERSION;
1212 ip->ip_hl = sizeof(struct ip) >> 2;
1217 ip->ip_p = IPPROTO_TCP;
1218 ip->ip_src = sc->sc_inc.inc_laddr;
1219 ip->ip_dst = sc->sc_inc.inc_faddr;
1220 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */
1221 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */
1224 * See if we should do MTU discovery. Route lookups are
1225 * expensive, so we will only unset the DF bit if:
1227 * 1) path_mtu_discovery is disabled
1228 * 2) the SCF_UNREACH flag has been set
1230 if (path_mtu_discovery
1231 && ((sc->sc_flags & SCF_UNREACH) == 0)) {
1232 ip->ip_off |= IP_DF;
1235 th = (struct tcphdr *)(ip + 1);
1237 th->th_sport = sc->sc_inc.inc_lport;
1238 th->th_dport = sc->sc_inc.inc_fport;
1240 th->th_seq = htonl(sc->sc_iss);
1241 th->th_ack = htonl(sc->sc_irs + 1);
1242 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1244 th->th_flags = TH_SYN | TH_ACK;
1245 th->th_win = htons(sc->sc_wnd);
1248 /* Tack on the TCP options. */
1251 optp = (u_int8_t *)(th + 1);
1252 *optp++ = TCPOPT_MAXSEG;
1253 *optp++ = TCPOLEN_MAXSEG;
1254 *optp++ = (mssopt >> 8) & 0xff;
1255 *optp++ = mssopt & 0xff;
1257 if (sc->sc_flags & SCF_WINSCALE) {
1258 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
1259 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
1260 sc->sc_request_r_scale);
1264 if (sc->sc_flags & SCF_TIMESTAMP) {
1265 u_int32_t *lp = (u_int32_t *)(optp);
1267 /* Form timestamp option as shown in appendix A of RFC 1323. */
1268 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1269 *lp++ = htonl(ticks);
1270 *lp = htonl(sc->sc_tsrecent);
1271 optp += TCPOLEN_TSTAMP_APPA;
1274 #ifdef TCP_SIGNATURE
1276 * Handle TCP-MD5 passive opener response.
1278 if (sc->sc_flags & SCF_SIGNATURE) {
1279 u_int8_t *bp = optp;
1282 *bp++ = TCPOPT_SIGNATURE;
1283 *bp++ = TCPOLEN_SIGNATURE;
1284 for (i = 0; i < TCP_SIGLEN; i++)
1286 tcpsignature_compute(m, 0, optlen,
1287 optp + 2, IPSEC_DIR_OUTBOUND);
1290 optp += TCPOLEN_SIGNATURE + 2;
1292 #endif /* TCP_SIGNATURE */
1294 if (sc->sc_flags & SCF_SACK_PERMITTED) {
1295 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED);
1296 optp += TCPOLEN_SACK_PERMITTED_ALIGNED;
1301 struct route_in6 *ro6 = &sc->sc_route6;
1304 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
1305 ip6->ip6_hlim = in6_selecthlim(NULL,
1306 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL);
1307 error = ip6_output(m, NULL, ro6, 0, NULL, NULL,
1308 sc->sc_tp->t_inpcb);
1310 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1311 htons(tlen - hlen + IPPROTO_TCP));
1312 m->m_pkthdr.csum_flags = CSUM_TCP;
1313 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1314 error = ip_output(m, sc->sc_ipopts, &sc->sc_route,
1315 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb);
1323 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .|
1325 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .|
1327 * (A): peer mss index
1331 * The values below are chosen to minimize the size of the tcp_secret
1332 * table, as well as providing roughly a 16 second lifetime for the cookie.
1335 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */
1336 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */
1338 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1)
1339 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS)
1340 #define SYNCOOKIE_TIMEOUT \
1341 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT))
1342 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK)
1345 u_int32_t ts_secbits[4];
1347 } tcp_secret[SYNCOOKIE_NSECRETS];
1349 static int tcp_msstab[] = { 0, 536, 1460, 8960 };
1351 static MD5_CTX syn_ctx;
1353 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v))
1356 u_int32_t laddr, faddr;
1357 u_int32_t secbits[4];
1358 u_int16_t lport, fport;
1362 CTASSERT(sizeof(struct md5_add) == 28);
1366 * Consider the problem of a recreated (and retransmitted) cookie. If the
1367 * original SYN was accepted, the connection is established. The second
1368 * SYN is inflight, and if it arrives with an ISN that falls within the
1369 * receive window, the connection is killed.
1371 * However, since cookies have other problems, this may not be worth
1376 syncookie_generate(struct syncache *sc)
1378 u_int32_t md5_buffer[4];
1383 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1385 const boolean_t isipv6 = FALSE;
1388 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK;
1389 if (tcp_secret[idx].ts_expire < ticks) {
1390 for (i = 0; i < 4; i++)
1391 tcp_secret[idx].ts_secbits[i] = karc4random();
1392 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT;
1394 for (data = NELEM(tcp_msstab) - 1; data > 0; data--)
1395 if (tcp_msstab[data] <= sc->sc_peer_mss)
1397 data = (data << SYNCOOKIE_WNDBITS) | idx;
1398 data ^= sc->sc_irs; /* peer's iss */
1401 MD5Add(sc->sc_inc.inc6_laddr);
1402 MD5Add(sc->sc_inc.inc6_faddr);
1406 add.laddr = sc->sc_inc.inc_laddr.s_addr;
1407 add.faddr = sc->sc_inc.inc_faddr.s_addr;
1409 add.lport = sc->sc_inc.inc_lport;
1410 add.fport = sc->sc_inc.inc_fport;
1411 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1412 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1413 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1414 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1416 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1417 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK);
1421 static struct syncache *
1422 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so)
1424 u_int32_t md5_buffer[4];
1425 struct syncache *sc;
1430 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */
1431 idx = data & SYNCOOKIE_WNDMASK;
1432 if (tcp_secret[idx].ts_expire < ticks ||
1433 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks)
1437 if (inc->inc_isipv6) {
1438 MD5Add(inc->inc6_laddr);
1439 MD5Add(inc->inc6_faddr);
1445 add.laddr = inc->inc_laddr.s_addr;
1446 add.faddr = inc->inc_faddr.s_addr;
1448 add.lport = inc->inc_lport;
1449 add.fport = inc->inc_fport;
1450 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1451 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1452 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1453 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1455 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1456 data ^= md5_buffer[0];
1457 if (data & ~SYNCOOKIE_DATAMASK)
1459 data = data >> SYNCOOKIE_WNDBITS;
1462 * Fill in the syncache values.
1463 * XXX duplicate code from syncache_add
1465 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO);
1466 sc->sc_ipopts = NULL;
1467 sc->sc_inc.inc_fport = inc->inc_fport;
1468 sc->sc_inc.inc_lport = inc->inc_lport;
1470 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1471 if (inc->inc_isipv6) {
1472 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1473 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1474 sc->sc_route6.ro_rt = NULL;
1478 sc->sc_inc.inc_faddr = inc->inc_faddr;
1479 sc->sc_inc.inc_laddr = inc->inc_laddr;
1480 sc->sc_route.ro_rt = NULL;
1482 sc->sc_irs = th->th_seq - 1;
1483 sc->sc_iss = th->th_ack - 1;
1484 wnd = ssb_space(&so->so_rcv);
1486 wnd = imin(wnd, TCP_MAXWIN);
1490 sc->sc_peer_mss = tcp_msstab[data];