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>
93 #include <net/netisr2.h>
96 #include <net/route.h>
98 #include <netinet/in.h>
99 #include <netinet/in_systm.h>
100 #include <netinet/ip.h>
101 #include <netinet/in_var.h>
102 #include <netinet/in_pcb.h>
103 #include <netinet/ip_var.h>
104 #include <netinet/ip6.h>
106 #include <netinet/icmp6.h>
107 #include <netinet6/nd6.h>
109 #include <netinet6/ip6_var.h>
110 #include <netinet6/in6_pcb.h>
111 #include <netinet/tcp.h>
112 #include <netinet/tcp_fsm.h>
113 #include <netinet/tcp_seq.h>
114 #include <netinet/tcp_timer.h>
115 #include <netinet/tcp_timer2.h>
116 #include <netinet/tcp_var.h>
117 #include <netinet6/tcp6_var.h>
120 #include <netinet6/ipsec.h>
122 #include <netinet6/ipsec6.h>
124 #include <netproto/key/key.h>
128 #include <netproto/ipsec/ipsec.h>
130 #include <netproto/ipsec/ipsec6.h>
132 #include <netproto/ipsec/key.h>
134 #endif /*FAST_IPSEC*/
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 * 4 retransmits corresponds to a timeout of (3 + 3 + 3 + 3 + 3 == 15) seconds
156 * or (1 + 1 + 2 + 4 + 8 == 16) seconds if RFC6298 is used, the odds are that
157 * the user has given up attempting to connect by then.
159 #define SYNCACHE_MAXREXMTS 4
161 /* Arbitrary values */
162 #define TCP_SYNCACHE_HASHSIZE 512
163 #define TCP_SYNCACHE_BUCKETLIMIT 30
165 struct netmsg_sc_timer {
166 struct netmsg_base base;
167 struct msgrec *nm_mrec; /* back pointer to containing msgrec */
171 struct netmsg_sc_timer msg;
172 lwkt_port_t port; /* constant after init */
173 int slot; /* constant after init */
176 static void syncache_timer_handler(netmsg_t);
178 struct tcp_syncache {
186 static struct tcp_syncache tcp_syncache;
188 TAILQ_HEAD(syncache_list, syncache);
190 struct tcp_syncache_percpu {
191 struct syncache_head *hashbase;
193 struct syncache_list timerq[SYNCACHE_MAXREXMTS + 1];
194 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1];
195 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1];
197 static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU];
199 static struct lwkt_port syncache_null_rport;
201 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
203 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD,
204 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
206 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD,
207 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
211 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
212 &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
215 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD,
216 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
218 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
219 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
221 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
223 #define SYNCACHE_HASH(inc, mask) \
224 ((tcp_syncache.hash_secret ^ \
225 (inc)->inc_faddr.s_addr ^ \
226 ((inc)->inc_faddr.s_addr >> 16) ^ \
227 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
229 #define SYNCACHE_HASH6(inc, mask) \
230 ((tcp_syncache.hash_secret ^ \
231 (inc)->inc6_faddr.s6_addr32[0] ^ \
232 (inc)->inc6_faddr.s6_addr32[3] ^ \
233 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
235 #define ENDPTS_EQ(a, b) ( \
236 (a)->ie_fport == (b)->ie_fport && \
237 (a)->ie_lport == (b)->ie_lport && \
238 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
239 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
242 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
245 syncache_rto(int slot)
248 return (TCPTV_RTOBASE * tcp_syn_backoff_low[slot]);
250 return (TCPTV_RTOBASE * tcp_syn_backoff[slot]);
254 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu,
255 struct syncache *sc, int slot)
261 * Record the time that we spent in SYN|ACK
264 * Needed by RFC3390 and RFC6298.
266 sc->sc_rxtused += syncache_rto(slot - 1);
268 sc->sc_rxtslot = slot;
270 rto = syncache_rto(slot);
271 sc->sc_rxttime = ticks + rto;
273 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq);
274 if (!callout_active(&syncache_percpu->tt_timerq[slot])) {
275 callout_reset(&syncache_percpu->tt_timerq[slot], rto,
276 syncache_timer, &syncache_percpu->mrec[slot]);
281 syncache_free(struct syncache *sc)
285 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
287 const boolean_t isipv6 = FALSE;
291 m_free(sc->sc_ipopts);
293 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt;
296 * If this is the only reference to a protocol-cloned
297 * route, remove it immediately.
299 if ((rt->rt_flags & RTF_WASCLONED) && rt->rt_refcnt == 1)
300 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
301 rt_mask(rt), rt->rt_flags, NULL);
304 kfree(sc, M_SYNCACHE);
312 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
313 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
314 tcp_syncache.cache_limit =
315 tcp_syncache.hashsize * tcp_syncache.bucket_limit;
316 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
317 tcp_syncache.hash_secret = karc4random();
319 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
320 &tcp_syncache.hashsize);
321 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
322 &tcp_syncache.cache_limit);
323 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
324 &tcp_syncache.bucket_limit);
325 if (!powerof2(tcp_syncache.hashsize)) {
326 kprintf("WARNING: syncache hash size is not a power of 2.\n");
327 tcp_syncache.hashsize = 512; /* safe default */
329 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
331 lwkt_initport_replyonly_null(&syncache_null_rport);
333 for (cpu = 0; cpu < ncpus2; cpu++) {
334 struct tcp_syncache_percpu *syncache_percpu;
336 syncache_percpu = &tcp_syncache_percpu[cpu];
337 /* Allocate the hash table. */
338 syncache_percpu->hashbase = kmalloc(tcp_syncache.hashsize * sizeof(struct syncache_head),
339 M_SYNCACHE, M_WAITOK);
341 /* Initialize the hash buckets. */
342 for (i = 0; i < tcp_syncache.hashsize; i++) {
343 struct syncache_head *bucket;
345 bucket = &syncache_percpu->hashbase[i];
346 TAILQ_INIT(&bucket->sch_bucket);
347 bucket->sch_length = 0;
350 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) {
351 /* Initialize the timer queues. */
352 TAILQ_INIT(&syncache_percpu->timerq[i]);
353 callout_init_mp(&syncache_percpu->tt_timerq[i]);
355 syncache_percpu->mrec[i].slot = i;
356 syncache_percpu->mrec[i].port = netisr_cpuport(cpu);
357 syncache_percpu->mrec[i].msg.nm_mrec =
358 &syncache_percpu->mrec[i];
359 netmsg_init(&syncache_percpu->mrec[i].msg.base,
360 NULL, &syncache_null_rport,
361 0, syncache_timer_handler);
367 syncache_insert(struct syncache *sc, struct syncache_head *sch)
369 struct tcp_syncache_percpu *syncache_percpu;
370 struct syncache *sc2;
373 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
376 * Make sure that we don't overflow the per-bucket
377 * limit or the total cache size limit.
379 if (sch->sch_length >= tcp_syncache.bucket_limit) {
381 * The bucket is full, toss the oldest element.
383 sc2 = TAILQ_FIRST(&sch->sch_bucket);
384 if (sc2->sc_tp != NULL)
385 sc2->sc_tp->ts_recent = ticks;
386 syncache_drop(sc2, sch);
387 tcpstat.tcps_sc_bucketoverflow++;
388 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) {
390 * The cache is full. Toss the oldest entry in the
391 * entire cache. This is the front entry in the
392 * first non-empty timer queue with the largest
395 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
396 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]);
397 while (sc2 && (sc2->sc_flags & SCF_MARKER))
398 sc2 = TAILQ_NEXT(sc2, sc_timerq);
402 if (sc2->sc_tp != NULL)
403 sc2->sc_tp->ts_recent = ticks;
404 syncache_drop(sc2, NULL);
405 tcpstat.tcps_sc_cacheoverflow++;
408 /* Initialize the entry's timer. */
409 syncache_timeout(syncache_percpu, sc, 0);
411 /* Put it into the bucket. */
412 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash);
414 syncache_percpu->cache_count++;
415 tcpstat.tcps_sc_added++;
419 syncache_destroy(struct tcpcb *tp)
421 struct tcp_syncache_percpu *syncache_percpu;
422 struct syncache_head *bucket;
426 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
429 for (i = 0; i < tcp_syncache.hashsize; i++) {
430 bucket = &syncache_percpu->hashbase[i];
431 TAILQ_FOREACH(sc, &bucket->sch_bucket, sc_hash) {
439 syncache_drop(struct syncache *sc, struct syncache_head *sch)
441 struct tcp_syncache_percpu *syncache_percpu;
443 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
445 const boolean_t isipv6 = FALSE;
448 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
452 sch = &syncache_percpu->hashbase[
453 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)];
455 sch = &syncache_percpu->hashbase[
456 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)];
460 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
462 syncache_percpu->cache_count--;
471 * Remove the entry from the syncache timer/timeout queue. Note
472 * that we do not try to stop any running timer since we do not know
473 * whether the timer's message is in-transit or not. Since timeouts
474 * are fairly long, taking an unneeded callout does not detrimentally
475 * effect performance.
477 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq);
483 * Place a timeout message on the TCP thread's message queue.
484 * This routine runs in soft interrupt context.
486 * An invariant is for this routine to be called, the callout must
487 * have been active. Note that the callout is not deactivated until
488 * after the message has been processed in syncache_timer_handler() below.
491 syncache_timer(void *p)
493 struct netmsg_sc_timer *msg = p;
495 lwkt_sendmsg(msg->nm_mrec->port, &msg->base.lmsg);
499 * Service a timer message queued by timer expiration.
500 * This routine runs in the TCP protocol thread.
502 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
503 * If we have retransmitted an entry the maximum number of times, expire it.
505 * When we finish processing timed-out entries, we restart the timer if there
506 * are any entries still on the queue and deactivate it otherwise. Only after
507 * a timer has been deactivated here can it be restarted by syncache_timeout().
510 syncache_timer_handler(netmsg_t msg)
512 struct tcp_syncache_percpu *syncache_percpu;
514 struct syncache marker;
515 struct syncache_list *list;
519 slot = ((struct netmsg_sc_timer *)msg)->nm_mrec->slot;
520 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
522 list = &syncache_percpu->timerq[slot];
525 * Use a marker to keep our place in the scan. syncache_drop()
526 * can block and cause any next pointer we cache to become stale.
528 marker.sc_flags = SCF_MARKER;
529 TAILQ_INSERT_HEAD(list, &marker, sc_timerq);
531 while ((sc = TAILQ_NEXT(&marker, sc_timerq)) != NULL) {
535 TAILQ_REMOVE(list, &marker, sc_timerq);
536 TAILQ_INSERT_AFTER(list, sc, &marker, sc_timerq);
538 if (sc->sc_flags & SCF_MARKER)
541 if (ticks < sc->sc_rxttime)
542 break; /* finished because timerq sorted by time */
543 if (sc->sc_tp == NULL) {
544 syncache_drop(sc, NULL);
545 tcpstat.tcps_sc_stale++;
548 inp = sc->sc_tp->t_inpcb;
549 if (slot == SYNCACHE_MAXREXMTS ||
550 slot >= tcp_syncache.rexmt_limit ||
552 inp->inp_gencnt != sc->sc_inp_gencnt) {
553 syncache_drop(sc, NULL);
554 tcpstat.tcps_sc_stale++;
558 * syncache_respond() may call back into the syncache to
559 * to modify another entry, so do not obtain the next
560 * entry on the timer chain until it has completed.
562 syncache_respond(sc, NULL);
563 tcpstat.tcps_sc_retransmitted++;
564 TAILQ_REMOVE(list, sc, sc_timerq);
565 syncache_timeout(syncache_percpu, sc, slot + 1);
567 TAILQ_REMOVE(list, &marker, sc_timerq);
570 callout_reset(&syncache_percpu->tt_timerq[slot],
571 sc->sc_rxttime - ticks, syncache_timer,
572 &syncache_percpu->mrec[slot]);
574 callout_deactivate(&syncache_percpu->tt_timerq[slot]);
576 lwkt_replymsg(&msg->base.lmsg, 0);
580 * Find an entry in the syncache.
583 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
585 struct tcp_syncache_percpu *syncache_percpu;
587 struct syncache_head *sch;
589 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
591 if (inc->inc_isipv6) {
592 sch = &syncache_percpu->hashbase[
593 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
595 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash)
596 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
601 sch = &syncache_percpu->hashbase[
602 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
604 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
606 if (sc->sc_inc.inc_isipv6)
609 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
617 * This function is called when we get a RST for a
618 * non-existent connection, so that we can see if the
619 * connection is in the syn cache. If it is, zap it.
622 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
625 struct syncache_head *sch;
627 sc = syncache_lookup(inc, &sch);
632 * If the RST bit is set, check the sequence number to see
633 * if this is a valid reset segment.
635 * In all states except SYN-SENT, all reset (RST) segments
636 * are validated by checking their SEQ-fields. A reset is
637 * valid if its sequence number is in the window.
639 * The sequence number in the reset segment is normally an
640 * echo of our outgoing acknowlegement numbers, but some hosts
641 * send a reset with the sequence number at the rightmost edge
642 * of our receive window, and we have to handle this case.
644 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
645 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
646 syncache_drop(sc, sch);
647 tcpstat.tcps_sc_reset++;
652 syncache_badack(struct in_conninfo *inc)
655 struct syncache_head *sch;
657 sc = syncache_lookup(inc, &sch);
659 syncache_drop(sc, sch);
660 tcpstat.tcps_sc_badack++;
665 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
668 struct syncache_head *sch;
670 /* we are called at splnet() here */
671 sc = syncache_lookup(inc, &sch);
675 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
676 if (ntohl(th->th_seq) != sc->sc_iss)
680 * If we've rertransmitted 3 times and this is our second error,
681 * we remove the entry. Otherwise, we allow it to continue on.
682 * This prevents us from incorrectly nuking an entry during a
683 * spurious network outage.
687 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) {
688 sc->sc_flags |= SCF_UNREACH;
691 syncache_drop(sc, sch);
692 tcpstat.tcps_sc_unreach++;
696 * Build a new TCP socket structure from a syncache entry.
698 * This is called from the context of the SYN+ACK
700 static struct socket *
701 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
703 struct inpcb *inp = NULL, *linp;
705 struct tcpcb *tp, *ltp;
708 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
710 const boolean_t isipv6 = FALSE;
712 struct sockaddr_in sin_faddr;
713 struct sockaddr_in6 sin6_faddr;
714 struct sockaddr *faddr;
717 faddr = (struct sockaddr *)&sin6_faddr;
718 sin6_faddr.sin6_family = AF_INET6;
719 sin6_faddr.sin6_len = sizeof(sin6_faddr);
720 sin6_faddr.sin6_addr = sc->sc_inc.inc6_faddr;
721 sin6_faddr.sin6_port = sc->sc_inc.inc_fport;
722 sin6_faddr.sin6_flowinfo = sin6_faddr.sin6_scope_id = 0;
724 faddr = (struct sockaddr *)&sin_faddr;
725 sin_faddr.sin_family = AF_INET;
726 sin_faddr.sin_len = sizeof(sin_faddr);
727 sin_faddr.sin_addr = sc->sc_inc.inc_faddr;
728 sin_faddr.sin_port = sc->sc_inc.inc_fport;
729 bzero(sin_faddr.sin_zero, sizeof(sin_faddr.sin_zero));
733 * Ok, create the full blown connection, and set things up
734 * as they would have been set up if we had created the
735 * connection when the SYN arrived. If we can't create
736 * the connection, abort it.
738 * Set the protocol processing port for the socket to the current
739 * port (that the connection came in on).
741 so = sonewconn_faddr(lso, SS_ISCONNECTED, faddr);
744 * Drop the connection; we will send a RST if the peer
745 * retransmits the ACK,
747 tcpstat.tcps_listendrop++;
752 * Insert new socket into hash list.
755 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
757 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
760 inp->inp_vflag &= ~INP_IPV6;
761 inp->inp_vflag |= INP_IPV4;
762 inp->inp_flags &= ~IN6P_IPV6_V6ONLY;
764 inp->inp_laddr = sc->sc_inc.inc_laddr;
766 inp->inp_lport = sc->sc_inc.inc_lport;
767 if (in_pcbinsporthash(inp) != 0) {
769 * Undo the assignments above if we failed to
770 * put the PCB on the hash lists.
773 inp->in6p_laddr = kin6addr_any;
775 inp->inp_laddr.s_addr = INADDR_ANY;
781 /* copy old policy into new socket's */
782 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp))
783 kprintf("syncache_expand: could not copy policy\n");
786 struct in6_addr laddr6;
788 * Inherit socket options from the listening socket.
789 * Note that in6p_inputopts are not (and should not be)
790 * copied, since it stores previously received options and is
791 * used to detect if each new option is different than the
792 * previous one and hence should be passed to a user.
793 * If we copied in6p_inputopts, a user would not be able to
794 * receive options just after calling the accept system call.
796 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS;
797 if (linp->in6p_outputopts)
798 inp->in6p_outputopts =
799 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT);
800 inp->in6p_route = sc->sc_route6;
801 sc->sc_route6.ro_rt = NULL;
803 laddr6 = inp->in6p_laddr;
804 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
805 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
806 if (in6_pcbconnect(inp, faddr, &thread0)) {
807 inp->in6p_laddr = laddr6;
811 struct in_addr laddr;
813 inp->inp_options = ip_srcroute(m);
814 if (inp->inp_options == NULL) {
815 inp->inp_options = sc->sc_ipopts;
816 sc->sc_ipopts = NULL;
818 inp->inp_route = sc->sc_route;
819 sc->sc_route.ro_rt = NULL;
821 laddr = inp->inp_laddr;
822 if (inp->inp_laddr.s_addr == INADDR_ANY)
823 inp->inp_laddr = sc->sc_inc.inc_laddr;
824 if (in_pcbconnect(inp, faddr, &thread0)) {
825 inp->inp_laddr = laddr;
831 * The current port should be in the context of the SYN+ACK and
832 * so should match the tcp address port.
835 port = tcp6_addrport();
837 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport,
838 inp->inp_laddr.s_addr, inp->inp_lport);
840 KASSERT(port == &curthread->td_msgport,
841 ("TCP PORT MISMATCH %p vs %p\n", port, &curthread->td_msgport));
844 tp->t_state = TCPS_SYN_RECEIVED;
845 tp->iss = sc->sc_iss;
846 tp->irs = sc->sc_irs;
849 tp->snd_wnd = sc->sc_sndwnd;
850 tp->snd_wl1 = sc->sc_irs;
851 tp->rcv_up = sc->sc_irs + 1;
852 tp->rcv_wnd = sc->sc_wnd;
853 tp->rcv_adv += tp->rcv_wnd;
855 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY);
856 if (sc->sc_flags & SCF_NOOPT)
857 tp->t_flags |= TF_NOOPT;
858 if (sc->sc_flags & SCF_WINSCALE) {
859 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE;
860 tp->snd_scale = sc->sc_requested_s_scale;
861 tp->request_r_scale = sc->sc_request_r_scale;
863 if (sc->sc_flags & SCF_TIMESTAMP) {
864 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP;
865 tp->ts_recent = sc->sc_tsrecent;
866 tp->ts_recent_age = ticks;
868 if (sc->sc_flags & SCF_SACK_PERMITTED)
869 tp->t_flags |= TF_SACK_PERMITTED;
872 if (sc->sc_flags & SCF_SIGNATURE)
873 tp->t_flags |= TF_SIGNATURE;
874 #endif /* TCP_SIGNATURE */
876 tp->t_rxtsyn = sc->sc_rxtused;
877 tcp_mss(tp, sc->sc_peer_mss);
880 * Inherit some properties from the listen socket
882 ltp = intotcpcb(linp);
883 tp->t_keepinit = ltp->t_keepinit;
884 tp->t_keepidle = ltp->t_keepidle;
885 tp->t_keepintvl = ltp->t_keepintvl;
886 tp->t_keepcnt = ltp->t_keepcnt;
887 tp->t_maxidle = ltp->t_maxidle;
889 tcp_create_timermsg(tp, port);
890 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
892 tcpstat.tcps_accepts++;
902 * This function gets called when we receive an ACK for a
903 * socket in the LISTEN state. We look up the connection
904 * in the syncache, and if its there, we pull it out of
905 * the cache and turn it into a full-blown connection in
906 * the SYN-RECEIVED state.
909 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop,
913 struct syncache_head *sch;
916 sc = syncache_lookup(inc, &sch);
919 * There is no syncache entry, so see if this ACK is
920 * a returning syncookie. To do this, first:
921 * A. See if this socket has had a syncache entry dropped in
922 * the past. We don't want to accept a bogus syncookie
923 * if we've never received a SYN.
924 * B. check that the syncookie is valid. If it is, then
925 * cobble up a fake syncache entry, and return.
929 sc = syncookie_lookup(inc, th, *sop);
933 tcpstat.tcps_sc_recvcookie++;
937 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
939 if (th->th_ack != sc->sc_iss + 1)
942 so = syncache_socket(sc, *sop, m);
946 /* XXXjlemon check this - is this correct? */
947 tcp_respond(NULL, m, m, th,
948 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK);
950 m_freem(m); /* XXX only needed for above */
951 tcpstat.tcps_sc_aborted++;
953 tcpstat.tcps_sc_completed++;
958 syncache_drop(sc, sch);
964 * Given a LISTEN socket and an inbound SYN request, add
965 * this to the syn cache, and send back a segment:
966 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
969 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
970 * Doing so would require that we hold onto the data and deliver it
971 * to the application. However, if we are the target of a SYN-flood
972 * DoS attack, an attacker could send data which would eventually
973 * consume all available buffer space if it were ACKed. By not ACKing
974 * the data, we avoid this DoS scenario.
977 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
978 struct socket *so, struct mbuf *m)
980 struct tcp_syncache_percpu *syncache_percpu;
982 struct syncache *sc = NULL;
983 struct syncache_head *sch;
984 struct mbuf *ipopts = NULL;
987 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
991 * Remember the IP options, if any.
994 if (!inc->inc_isipv6)
996 ipopts = ip_srcroute(m);
999 * See if we already have an entry for this connection.
1000 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1003 * The syncache should be re-initialized with the contents
1004 * of the new SYN which may have different options.
1006 sc = syncache_lookup(inc, &sch);
1008 tcpstat.tcps_sc_dupsyn++;
1011 * If we were remembering a previous source route,
1012 * forget it and use the new one we've been given.
1015 m_free(sc->sc_ipopts);
1016 sc->sc_ipopts = ipopts;
1019 * Update timestamp if present.
1021 if (sc->sc_flags & SCF_TIMESTAMP)
1022 sc->sc_tsrecent = to->to_tsval;
1024 /* Just update the TOF_SACK_PERMITTED for now. */
1025 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1026 sc->sc_flags |= SCF_SACK_PERMITTED;
1028 sc->sc_flags &= ~SCF_SACK_PERMITTED;
1030 /* Update initial send window */
1031 sc->sc_sndwnd = th->th_win;
1034 * PCB may have changed, pick up new values.
1037 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
1038 if (syncache_respond(sc, m) == 0) {
1039 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot],
1041 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot);
1042 tcpstat.tcps_sndacks++;
1043 tcpstat.tcps_sndtotal++;
1049 * Fill in the syncache values.
1051 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO);
1052 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
1053 sc->sc_ipopts = ipopts;
1054 sc->sc_inc.inc_fport = inc->inc_fport;
1055 sc->sc_inc.inc_lport = inc->inc_lport;
1058 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1059 if (inc->inc_isipv6) {
1060 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1061 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1062 sc->sc_route6.ro_rt = NULL;
1066 sc->sc_inc.inc_faddr = inc->inc_faddr;
1067 sc->sc_inc.inc_laddr = inc->inc_laddr;
1068 sc->sc_route.ro_rt = NULL;
1070 sc->sc_irs = th->th_seq;
1072 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0;
1074 sc->sc_iss = syncookie_generate(sc);
1076 sc->sc_iss = karc4random();
1078 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */
1079 win = ssb_space(&so->so_rcv);
1081 win = imin(win, TCP_MAXWIN);
1084 if (tcp_do_rfc1323) {
1086 * A timestamp received in a SYN makes
1087 * it ok to send timestamp requests and replies.
1089 if (to->to_flags & TOF_TS) {
1090 sc->sc_tsrecent = to->to_tsval;
1091 sc->sc_flags |= SCF_TIMESTAMP;
1093 if (to->to_flags & TOF_SCALE) {
1094 int wscale = TCP_MIN_WINSHIFT;
1096 /* Compute proper scaling value from buffer space */
1097 while (wscale < TCP_MAX_WINSHIFT &&
1098 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) {
1101 sc->sc_request_r_scale = wscale;
1102 sc->sc_requested_s_scale = to->to_requested_s_scale;
1103 sc->sc_flags |= SCF_WINSCALE;
1106 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1107 sc->sc_flags |= SCF_SACK_PERMITTED;
1108 if (tp->t_flags & TF_NOOPT)
1109 sc->sc_flags = SCF_NOOPT;
1110 #ifdef TCP_SIGNATURE
1112 * If listening socket requested TCP digests, and received SYN
1113 * contains the option, flag this in the syncache so that
1114 * syncache_respond() will do the right thing with the SYN+ACK.
1115 * XXX Currently we always record the option by default and will
1116 * attempt to use it in syncache_respond().
1118 if (to->to_flags & TOF_SIGNATURE)
1119 sc->sc_flags = SCF_SIGNATURE;
1120 #endif /* TCP_SIGNATURE */
1121 sc->sc_sndwnd = th->th_win;
1123 if (syncache_respond(sc, m) == 0) {
1124 syncache_insert(sc, sch);
1125 tcpstat.tcps_sndacks++;
1126 tcpstat.tcps_sndtotal++;
1129 tcpstat.tcps_sc_dropped++;
1135 syncache_respond(struct syncache *sc, struct mbuf *m)
1139 u_int16_t tlen, hlen, mssopt;
1140 struct ip *ip = NULL;
1143 struct ip6_hdr *ip6 = NULL;
1145 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1147 const boolean_t isipv6 = FALSE;
1151 rt = tcp_rtlookup6(&sc->sc_inc);
1153 mssopt = rt->rt_ifp->if_mtu -
1154 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1156 mssopt = tcp_v6mssdflt;
1157 hlen = sizeof(struct ip6_hdr);
1159 rt = tcp_rtlookup(&sc->sc_inc);
1161 mssopt = rt->rt_ifp->if_mtu -
1162 (sizeof(struct ip) + sizeof(struct tcphdr));
1164 mssopt = tcp_mssdflt;
1165 hlen = sizeof(struct ip);
1168 /* Compute the size of the TCP options. */
1169 if (sc->sc_flags & SCF_NOOPT) {
1172 optlen = TCPOLEN_MAXSEG +
1173 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) +
1174 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) +
1175 ((sc->sc_flags & SCF_SACK_PERMITTED) ?
1176 TCPOLEN_SACK_PERMITTED_ALIGNED : 0);
1177 #ifdef TCP_SIGNATURE
1178 optlen += ((sc->sc_flags & SCF_SIGNATURE) ?
1179 (TCPOLEN_SIGNATURE + 2) : 0);
1180 #endif /* TCP_SIGNATURE */
1182 tlen = hlen + sizeof(struct tcphdr) + optlen;
1186 * assume that the entire packet will fit in a header mbuf
1188 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small"));
1191 * XXX shouldn't this reuse the mbuf if possible ?
1192 * Create the IP+TCP header from scratch.
1197 m = m_gethdr(MB_DONTWAIT, MT_HEADER);
1200 m->m_data += max_linkhdr;
1202 m->m_pkthdr.len = tlen;
1203 m->m_pkthdr.rcvif = NULL;
1206 ip6 = mtod(m, struct ip6_hdr *);
1207 ip6->ip6_vfc = IPV6_VERSION;
1208 ip6->ip6_nxt = IPPROTO_TCP;
1209 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1210 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1211 ip6->ip6_plen = htons(tlen - hlen);
1212 /* ip6_hlim is set after checksum */
1213 /* ip6_flow = ??? */
1215 th = (struct tcphdr *)(ip6 + 1);
1217 ip = mtod(m, struct ip *);
1218 ip->ip_v = IPVERSION;
1219 ip->ip_hl = sizeof(struct ip) >> 2;
1224 ip->ip_p = IPPROTO_TCP;
1225 ip->ip_src = sc->sc_inc.inc_laddr;
1226 ip->ip_dst = sc->sc_inc.inc_faddr;
1227 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */
1228 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */
1231 * See if we should do MTU discovery. Route lookups are
1232 * expensive, so we will only unset the DF bit if:
1234 * 1) path_mtu_discovery is disabled
1235 * 2) the SCF_UNREACH flag has been set
1237 if (path_mtu_discovery
1238 && ((sc->sc_flags & SCF_UNREACH) == 0)) {
1239 ip->ip_off |= IP_DF;
1242 th = (struct tcphdr *)(ip + 1);
1244 th->th_sport = sc->sc_inc.inc_lport;
1245 th->th_dport = sc->sc_inc.inc_fport;
1247 th->th_seq = htonl(sc->sc_iss);
1248 th->th_ack = htonl(sc->sc_irs + 1);
1249 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1251 th->th_flags = TH_SYN | TH_ACK;
1252 th->th_win = htons(sc->sc_wnd);
1255 /* Tack on the TCP options. */
1258 optp = (u_int8_t *)(th + 1);
1259 *optp++ = TCPOPT_MAXSEG;
1260 *optp++ = TCPOLEN_MAXSEG;
1261 *optp++ = (mssopt >> 8) & 0xff;
1262 *optp++ = mssopt & 0xff;
1264 if (sc->sc_flags & SCF_WINSCALE) {
1265 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
1266 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
1267 sc->sc_request_r_scale);
1271 if (sc->sc_flags & SCF_TIMESTAMP) {
1272 u_int32_t *lp = (u_int32_t *)(optp);
1274 /* Form timestamp option as shown in appendix A of RFC 1323. */
1275 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1276 *lp++ = htonl(ticks);
1277 *lp = htonl(sc->sc_tsrecent);
1278 optp += TCPOLEN_TSTAMP_APPA;
1281 #ifdef TCP_SIGNATURE
1283 * Handle TCP-MD5 passive opener response.
1285 if (sc->sc_flags & SCF_SIGNATURE) {
1286 u_int8_t *bp = optp;
1289 *bp++ = TCPOPT_SIGNATURE;
1290 *bp++ = TCPOLEN_SIGNATURE;
1291 for (i = 0; i < TCP_SIGLEN; i++)
1293 tcpsignature_compute(m, 0, optlen,
1294 optp + 2, IPSEC_DIR_OUTBOUND);
1297 optp += TCPOLEN_SIGNATURE + 2;
1299 #endif /* TCP_SIGNATURE */
1301 if (sc->sc_flags & SCF_SACK_PERMITTED) {
1302 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED);
1303 optp += TCPOLEN_SACK_PERMITTED_ALIGNED;
1308 struct route_in6 *ro6 = &sc->sc_route6;
1311 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
1312 ip6->ip6_hlim = in6_selecthlim(NULL,
1313 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL);
1314 error = ip6_output(m, NULL, ro6, 0, NULL, NULL,
1315 sc->sc_tp->t_inpcb);
1317 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1318 htons(tlen - hlen + IPPROTO_TCP));
1319 m->m_pkthdr.csum_flags = CSUM_TCP;
1320 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1321 m->m_pkthdr.csum_thlen = sizeof(struct tcphdr) + optlen;
1322 error = ip_output(m, sc->sc_ipopts, &sc->sc_route,
1323 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb);
1331 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .|
1333 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .|
1335 * (A): peer mss index
1339 * The values below are chosen to minimize the size of the tcp_secret
1340 * table, as well as providing roughly a 16 second lifetime for the cookie.
1343 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */
1344 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */
1346 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1)
1347 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS)
1348 #define SYNCOOKIE_TIMEOUT \
1349 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT))
1350 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK)
1353 u_int32_t ts_secbits[4];
1355 } tcp_secret[SYNCOOKIE_NSECRETS];
1357 static int tcp_msstab[] = { 0, 536, 1460, 8960 };
1359 static MD5_CTX syn_ctx;
1361 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v))
1364 u_int32_t laddr, faddr;
1365 u_int32_t secbits[4];
1366 u_int16_t lport, fport;
1370 CTASSERT(sizeof(struct md5_add) == 28);
1374 * Consider the problem of a recreated (and retransmitted) cookie. If the
1375 * original SYN was accepted, the connection is established. The second
1376 * SYN is inflight, and if it arrives with an ISN that falls within the
1377 * receive window, the connection is killed.
1379 * However, since cookies have other problems, this may not be worth
1384 syncookie_generate(struct syncache *sc)
1386 u_int32_t md5_buffer[4];
1391 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1393 const boolean_t isipv6 = FALSE;
1396 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK;
1397 if (tcp_secret[idx].ts_expire < ticks) {
1398 for (i = 0; i < 4; i++)
1399 tcp_secret[idx].ts_secbits[i] = karc4random();
1400 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT;
1402 for (data = NELEM(tcp_msstab) - 1; data > 0; data--)
1403 if (tcp_msstab[data] <= sc->sc_peer_mss)
1405 data = (data << SYNCOOKIE_WNDBITS) | idx;
1406 data ^= sc->sc_irs; /* peer's iss */
1409 MD5Add(sc->sc_inc.inc6_laddr);
1410 MD5Add(sc->sc_inc.inc6_faddr);
1414 add.laddr = sc->sc_inc.inc_laddr.s_addr;
1415 add.faddr = sc->sc_inc.inc_faddr.s_addr;
1417 add.lport = sc->sc_inc.inc_lport;
1418 add.fport = sc->sc_inc.inc_fport;
1419 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1420 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1421 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1422 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1424 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1425 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK);
1429 static struct syncache *
1430 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so)
1432 u_int32_t md5_buffer[4];
1433 struct syncache *sc;
1438 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */
1439 idx = data & SYNCOOKIE_WNDMASK;
1440 if (tcp_secret[idx].ts_expire < ticks ||
1441 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks)
1445 if (inc->inc_isipv6) {
1446 MD5Add(inc->inc6_laddr);
1447 MD5Add(inc->inc6_faddr);
1453 add.laddr = inc->inc_laddr.s_addr;
1454 add.faddr = inc->inc_faddr.s_addr;
1456 add.lport = inc->inc_lport;
1457 add.fport = inc->inc_fport;
1458 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1459 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1460 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1461 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1463 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1464 data ^= md5_buffer[0];
1465 if (data & ~SYNCOOKIE_DATAMASK)
1467 data = data >> SYNCOOKIE_WNDBITS;
1470 * Fill in the syncache values.
1471 * XXX duplicate code from syncache_add
1473 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO);
1474 sc->sc_ipopts = NULL;
1475 sc->sc_inc.inc_fport = inc->inc_fport;
1476 sc->sc_inc.inc_lport = inc->inc_lport;
1478 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1479 if (inc->inc_isipv6) {
1480 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1481 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1482 sc->sc_route6.ro_rt = NULL;
1486 sc->sc_inc.inc_faddr = inc->inc_faddr;
1487 sc->sc_inc.inc_laddr = inc->inc_laddr;
1488 sc->sc_route.ro_rt = NULL;
1490 sc->sc_irs = th->th_seq - 1;
1491 sc->sc_iss = th->th_ack - 1;
1492 wnd = ssb_space(&so->so_rcv);
1494 wnd = imin(wnd, TCP_MAXWIN);
1498 sc->sc_peer_mss = tcp_msstab[data];