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],
267 TCPTV_RTOBASE * tcp_backoff[slot],
269 &syncache_percpu->mrec[slot]);
274 syncache_free(struct syncache *sc)
278 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
280 const boolean_t isipv6 = FALSE;
284 m_free(sc->sc_ipopts);
286 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt;
289 * If this is the only reference to a protocol-cloned
290 * route, remove it immediately.
292 if ((rt->rt_flags & RTF_WASCLONED) && rt->rt_refcnt == 1)
293 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
294 rt_mask(rt), rt->rt_flags, NULL);
297 kfree(sc, M_SYNCACHE);
305 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
306 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
307 tcp_syncache.cache_limit =
308 tcp_syncache.hashsize * tcp_syncache.bucket_limit;
309 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
310 tcp_syncache.hash_secret = karc4random();
312 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
313 &tcp_syncache.hashsize);
314 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
315 &tcp_syncache.cache_limit);
316 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
317 &tcp_syncache.bucket_limit);
318 if (!powerof2(tcp_syncache.hashsize)) {
319 kprintf("WARNING: syncache hash size is not a power of 2.\n");
320 tcp_syncache.hashsize = 512; /* safe default */
322 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
324 lwkt_initport_replyonly_null(&syncache_null_rport);
326 for (cpu = 0; cpu < ncpus2; cpu++) {
327 struct tcp_syncache_percpu *syncache_percpu;
329 syncache_percpu = &tcp_syncache_percpu[cpu];
330 /* Allocate the hash table. */
331 syncache_percpu->hashbase = kmalloc(tcp_syncache.hashsize * sizeof(struct syncache_head),
332 M_SYNCACHE, M_WAITOK);
334 /* Initialize the hash buckets. */
335 for (i = 0; i < tcp_syncache.hashsize; i++) {
336 struct syncache_head *bucket;
338 bucket = &syncache_percpu->hashbase[i];
339 TAILQ_INIT(&bucket->sch_bucket);
340 bucket->sch_length = 0;
343 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) {
344 /* Initialize the timer queues. */
345 TAILQ_INIT(&syncache_percpu->timerq[i]);
346 callout_init_mp(&syncache_percpu->tt_timerq[i]);
348 syncache_percpu->mrec[i].slot = i;
349 syncache_percpu->mrec[i].port = cpu_portfn(cpu);
350 syncache_percpu->mrec[i].msg.nm_mrec =
351 &syncache_percpu->mrec[i];
352 netmsg_init(&syncache_percpu->mrec[i].msg.base,
353 NULL, &syncache_null_rport,
354 0, syncache_timer_handler);
360 syncache_insert(struct syncache *sc, struct syncache_head *sch)
362 struct tcp_syncache_percpu *syncache_percpu;
363 struct syncache *sc2;
366 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
369 * Make sure that we don't overflow the per-bucket
370 * limit or the total cache size limit.
372 if (sch->sch_length >= tcp_syncache.bucket_limit) {
374 * The bucket is full, toss the oldest element.
376 sc2 = TAILQ_FIRST(&sch->sch_bucket);
377 sc2->sc_tp->ts_recent = ticks;
378 syncache_drop(sc2, sch);
379 tcpstat.tcps_sc_bucketoverflow++;
380 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) {
382 * The cache is full. Toss the oldest entry in the
383 * entire cache. This is the front entry in the
384 * first non-empty timer queue with the largest
387 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
388 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]);
389 while (sc2 && (sc2->sc_flags & SCF_MARKER))
390 sc2 = TAILQ_NEXT(sc2, sc_timerq);
394 sc2->sc_tp->ts_recent = ticks;
395 syncache_drop(sc2, NULL);
396 tcpstat.tcps_sc_cacheoverflow++;
399 /* Initialize the entry's timer. */
400 syncache_timeout(syncache_percpu, sc, 0);
402 /* Put it into the bucket. */
403 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash);
405 syncache_percpu->cache_count++;
406 tcpstat.tcps_sc_added++;
410 syncache_destroy(struct tcpcb *tp)
412 struct tcp_syncache_percpu *syncache_percpu;
413 struct syncache_head *bucket;
417 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
420 for (i = 0; i < tcp_syncache.hashsize; i++) {
421 bucket = &syncache_percpu->hashbase[i];
422 TAILQ_FOREACH(sc, &bucket->sch_bucket, sc_hash) {
430 syncache_drop(struct syncache *sc, struct syncache_head *sch)
432 struct tcp_syncache_percpu *syncache_percpu;
434 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
436 const boolean_t isipv6 = FALSE;
439 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
443 sch = &syncache_percpu->hashbase[
444 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)];
446 sch = &syncache_percpu->hashbase[
447 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)];
451 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
453 syncache_percpu->cache_count--;
462 * Remove the entry from the syncache timer/timeout queue. Note
463 * that we do not try to stop any running timer since we do not know
464 * whether the timer's message is in-transit or not. Since timeouts
465 * are fairly long, taking an unneeded callout does not detrimentally
466 * effect performance.
468 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq);
474 * Place a timeout message on the TCP thread's message queue.
475 * This routine runs in soft interrupt context.
477 * An invariant is for this routine to be called, the callout must
478 * have been active. Note that the callout is not deactivated until
479 * after the message has been processed in syncache_timer_handler() below.
482 syncache_timer(void *p)
484 struct netmsg_sc_timer *msg = p;
486 lwkt_sendmsg(msg->nm_mrec->port, &msg->base.lmsg);
490 * Service a timer message queued by timer expiration.
491 * This routine runs in the TCP protocol thread.
493 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
494 * If we have retransmitted an entry the maximum number of times, expire it.
496 * When we finish processing timed-out entries, we restart the timer if there
497 * are any entries still on the queue and deactivate it otherwise. Only after
498 * a timer has been deactivated here can it be restarted by syncache_timeout().
501 syncache_timer_handler(netmsg_t msg)
503 struct tcp_syncache_percpu *syncache_percpu;
505 struct syncache marker;
506 struct syncache_list *list;
510 slot = ((struct netmsg_sc_timer *)msg)->nm_mrec->slot;
511 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
513 list = &syncache_percpu->timerq[slot];
516 * Use a marker to keep our place in the scan. syncache_drop()
517 * can block and cause any next pointer we cache to become stale.
519 marker.sc_flags = SCF_MARKER;
520 TAILQ_INSERT_HEAD(list, &marker, sc_timerq);
522 while ((sc = TAILQ_NEXT(&marker, sc_timerq)) != NULL) {
526 TAILQ_REMOVE(list, &marker, sc_timerq);
527 TAILQ_INSERT_AFTER(list, sc, &marker, sc_timerq);
529 if (sc->sc_flags & SCF_MARKER)
532 if (ticks < sc->sc_rxttime)
533 break; /* finished because timerq sorted by time */
534 if (sc->sc_tp == NULL) {
535 syncache_drop(sc, NULL);
536 tcpstat.tcps_sc_stale++;
539 inp = sc->sc_tp->t_inpcb;
540 if (slot == SYNCACHE_MAXREXMTS ||
541 slot >= tcp_syncache.rexmt_limit ||
543 inp->inp_gencnt != sc->sc_inp_gencnt) {
544 syncache_drop(sc, NULL);
545 tcpstat.tcps_sc_stale++;
549 * syncache_respond() may call back into the syncache to
550 * to modify another entry, so do not obtain the next
551 * entry on the timer chain until it has completed.
553 syncache_respond(sc, NULL);
554 tcpstat.tcps_sc_retransmitted++;
555 TAILQ_REMOVE(list, sc, sc_timerq);
556 syncache_timeout(syncache_percpu, sc, slot + 1);
558 TAILQ_REMOVE(list, &marker, sc_timerq);
561 callout_reset(&syncache_percpu->tt_timerq[slot],
562 sc->sc_rxttime - ticks, syncache_timer,
563 &syncache_percpu->mrec[slot]);
565 callout_deactivate(&syncache_percpu->tt_timerq[slot]);
567 lwkt_replymsg(&msg->base.lmsg, 0);
571 * Find an entry in the syncache.
574 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
576 struct tcp_syncache_percpu *syncache_percpu;
578 struct syncache_head *sch;
580 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
582 if (inc->inc_isipv6) {
583 sch = &syncache_percpu->hashbase[
584 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
586 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash)
587 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
592 sch = &syncache_percpu->hashbase[
593 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
595 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
597 if (sc->sc_inc.inc_isipv6)
600 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
608 * This function is called when we get a RST for a
609 * non-existent connection, so that we can see if the
610 * connection is in the syn cache. If it is, zap it.
613 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
616 struct syncache_head *sch;
618 sc = syncache_lookup(inc, &sch);
623 * If the RST bit is set, check the sequence number to see
624 * if this is a valid reset segment.
626 * In all states except SYN-SENT, all reset (RST) segments
627 * are validated by checking their SEQ-fields. A reset is
628 * valid if its sequence number is in the window.
630 * The sequence number in the reset segment is normally an
631 * echo of our outgoing acknowlegement numbers, but some hosts
632 * send a reset with the sequence number at the rightmost edge
633 * of our receive window, and we have to handle this case.
635 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
636 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
637 syncache_drop(sc, sch);
638 tcpstat.tcps_sc_reset++;
643 syncache_badack(struct in_conninfo *inc)
646 struct syncache_head *sch;
648 sc = syncache_lookup(inc, &sch);
650 syncache_drop(sc, sch);
651 tcpstat.tcps_sc_badack++;
656 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
659 struct syncache_head *sch;
661 /* we are called at splnet() here */
662 sc = syncache_lookup(inc, &sch);
666 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
667 if (ntohl(th->th_seq) != sc->sc_iss)
671 * If we've rertransmitted 3 times and this is our second error,
672 * we remove the entry. Otherwise, we allow it to continue on.
673 * This prevents us from incorrectly nuking an entry during a
674 * spurious network outage.
678 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) {
679 sc->sc_flags |= SCF_UNREACH;
682 syncache_drop(sc, sch);
683 tcpstat.tcps_sc_unreach++;
687 * Build a new TCP socket structure from a syncache entry.
689 * This is called from the context of the SYN+ACK
691 static struct socket *
692 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
694 struct inpcb *inp = NULL, *linp;
696 struct tcpcb *tp, *ltp;
699 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
701 const boolean_t isipv6 = FALSE;
703 struct sockaddr_in sin_faddr;
704 struct sockaddr_in6 sin6_faddr;
705 struct sockaddr *faddr;
708 faddr = (struct sockaddr *)&sin6_faddr;
709 sin6_faddr.sin6_family = AF_INET6;
710 sin6_faddr.sin6_len = sizeof(sin6_faddr);
711 sin6_faddr.sin6_addr = sc->sc_inc.inc6_faddr;
712 sin6_faddr.sin6_port = sc->sc_inc.inc_fport;
713 sin6_faddr.sin6_flowinfo = sin6_faddr.sin6_scope_id = 0;
715 faddr = (struct sockaddr *)&sin_faddr;
716 sin_faddr.sin_family = AF_INET;
717 sin_faddr.sin_len = sizeof(sin_faddr);
718 sin_faddr.sin_addr = sc->sc_inc.inc_faddr;
719 sin_faddr.sin_port = sc->sc_inc.inc_fport;
720 bzero(sin_faddr.sin_zero, sizeof(sin_faddr.sin_zero));
724 * Ok, create the full blown connection, and set things up
725 * as they would have been set up if we had created the
726 * connection when the SYN arrived. If we can't create
727 * the connection, abort it.
729 * Set the protocol processing port for the socket to the current
730 * port (that the connection came in on).
732 so = sonewconn_faddr(lso, SS_ISCONNECTED, faddr);
735 * Drop the connection; we will send a RST if the peer
736 * retransmits the ACK,
738 tcpstat.tcps_listendrop++;
743 * Insert new socket into hash list.
746 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
748 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
751 inp->inp_vflag &= ~INP_IPV6;
752 inp->inp_vflag |= INP_IPV4;
753 inp->inp_flags &= ~IN6P_IPV6_V6ONLY;
755 inp->inp_laddr = sc->sc_inc.inc_laddr;
757 inp->inp_lport = sc->sc_inc.inc_lport;
758 if (in_pcbinsporthash(inp) != 0) {
760 * Undo the assignments above if we failed to
761 * put the PCB on the hash lists.
764 inp->in6p_laddr = kin6addr_any;
766 inp->inp_laddr.s_addr = INADDR_ANY;
772 /* copy old policy into new socket's */
773 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp))
774 kprintf("syncache_expand: could not copy policy\n");
777 struct in6_addr laddr6;
779 * Inherit socket options from the listening socket.
780 * Note that in6p_inputopts are not (and should not be)
781 * copied, since it stores previously received options and is
782 * used to detect if each new option is different than the
783 * previous one and hence should be passed to a user.
784 * If we copied in6p_inputopts, a user would not be able to
785 * receive options just after calling the accept system call.
787 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS;
788 if (linp->in6p_outputopts)
789 inp->in6p_outputopts =
790 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT);
791 inp->in6p_route = sc->sc_route6;
792 sc->sc_route6.ro_rt = NULL;
794 laddr6 = inp->in6p_laddr;
795 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
796 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
797 if (in6_pcbconnect(inp, faddr, &thread0)) {
798 inp->in6p_laddr = laddr6;
802 struct in_addr laddr;
804 inp->inp_options = ip_srcroute(m);
805 if (inp->inp_options == NULL) {
806 inp->inp_options = sc->sc_ipopts;
807 sc->sc_ipopts = NULL;
809 inp->inp_route = sc->sc_route;
810 sc->sc_route.ro_rt = NULL;
812 laddr = inp->inp_laddr;
813 if (inp->inp_laddr.s_addr == INADDR_ANY)
814 inp->inp_laddr = sc->sc_inc.inc_laddr;
815 if (in_pcbconnect(inp, faddr, &thread0)) {
816 inp->inp_laddr = laddr;
822 * The current port should be in the context of the SYN+ACK and
823 * so should match the tcp address port.
825 * XXX we may be running on the netisr thread instead of a tcp
826 * thread, in which case port will not match
827 * curthread->td_msgport.
830 port = tcp6_addrport();
832 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport,
833 inp->inp_laddr.s_addr, inp->inp_lport);
835 if (port != &curthread->td_msgport) {
837 kprintf("TCP PORT MISMATCH %p vs %p\n",
838 port, &curthread->td_msgport);
840 /*KKASSERT(port == &curthread->td_msgport);*/
843 tp->t_state = TCPS_SYN_RECEIVED;
844 tp->iss = sc->sc_iss;
845 tp->irs = sc->sc_irs;
848 tp->snd_wl1 = sc->sc_irs;
849 tp->rcv_up = sc->sc_irs + 1;
850 tp->rcv_wnd = sc->sc_wnd;
851 tp->rcv_adv += tp->rcv_wnd;
853 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY);
854 if (sc->sc_flags & SCF_NOOPT)
855 tp->t_flags |= TF_NOOPT;
856 if (sc->sc_flags & SCF_WINSCALE) {
857 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE;
858 tp->requested_s_scale = sc->sc_requested_s_scale;
859 tp->request_r_scale = sc->sc_request_r_scale;
861 if (sc->sc_flags & SCF_TIMESTAMP) {
862 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP;
863 tp->ts_recent = sc->sc_tsrecent;
864 tp->ts_recent_age = ticks;
866 if (sc->sc_flags & SCF_SACK_PERMITTED)
867 tp->t_flags |= TF_SACK_PERMITTED;
868 if (sc->sc_flags & SCF_SYN_WASLOST)
869 tp->t_flags |= TF_SYN_WASLOST;
872 if (sc->sc_flags & SCF_SIGNATURE)
873 tp->t_flags |= TF_SIGNATURE;
874 #endif /* TCP_SIGNATURE */
876 tcp_mss(tp, sc->sc_peer_mss);
879 * Inherit some properties from the listen socket
881 ltp = intotcpcb(linp);
882 tp->t_keepinit = ltp->t_keepinit;
883 tp->t_keepidle = ltp->t_keepidle;
884 tp->t_keepintvl = ltp->t_keepintvl;
885 tp->t_keepcnt = ltp->t_keepcnt;
886 tp->t_maxidle = ltp->t_maxidle;
888 tcp_create_timermsg(tp, port);
889 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
891 tcpstat.tcps_accepts++;
901 * This function gets called when we receive an ACK for a
902 * socket in the LISTEN state. We look up the connection
903 * in the syncache, and if its there, we pull it out of
904 * the cache and turn it into a full-blown connection in
905 * the SYN-RECEIVED state.
908 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop,
912 struct syncache_head *sch;
915 sc = syncache_lookup(inc, &sch);
918 * There is no syncache entry, so see if this ACK is
919 * a returning syncookie. To do this, first:
920 * A. See if this socket has had a syncache entry dropped in
921 * the past. We don't want to accept a bogus syncookie
922 * if we've never received a SYN.
923 * B. check that the syncookie is valid. If it is, then
924 * cobble up a fake syncache entry, and return.
928 sc = syncookie_lookup(inc, th, *sop);
932 tcpstat.tcps_sc_recvcookie++;
936 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
938 if (th->th_ack != sc->sc_iss + 1)
941 so = syncache_socket(sc, *sop, m);
945 /* XXXjlemon check this - is this correct? */
946 tcp_respond(NULL, m, m, th,
947 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK);
949 m_freem(m); /* XXX only needed for above */
950 tcpstat.tcps_sc_aborted++;
952 tcpstat.tcps_sc_completed++;
957 syncache_drop(sc, sch);
963 * Given a LISTEN socket and an inbound SYN request, add
964 * this to the syn cache, and send back a segment:
965 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
968 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
969 * Doing so would require that we hold onto the data and deliver it
970 * to the application. However, if we are the target of a SYN-flood
971 * DoS attack, an attacker could send data which would eventually
972 * consume all available buffer space if it were ACKed. By not ACKing
973 * the data, we avoid this DoS scenario.
976 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
977 struct socket *so, struct mbuf *m)
979 struct tcp_syncache_percpu *syncache_percpu;
981 struct syncache *sc = NULL;
982 struct syncache_head *sch;
983 struct mbuf *ipopts = NULL;
986 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
990 * Remember the IP options, if any.
993 if (!inc->inc_isipv6)
995 ipopts = ip_srcroute(m);
998 * See if we already have an entry for this connection.
999 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1002 * The syncache should be re-initialized with the contents
1003 * of the new SYN which may have different options.
1005 sc = syncache_lookup(inc, &sch);
1007 tcpstat.tcps_sc_dupsyn++;
1010 * If we were remembering a previous source route,
1011 * forget it and use the new one we've been given.
1014 m_free(sc->sc_ipopts);
1015 sc->sc_ipopts = ipopts;
1018 * Update timestamp if present.
1020 if (sc->sc_flags & SCF_TIMESTAMP)
1021 sc->sc_tsrecent = to->to_tsval;
1023 /* Just update the TOF_SACK_PERMITTED for now. */
1024 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1025 sc->sc_flags |= SCF_SACK_PERMITTED;
1027 sc->sc_flags &= ~SCF_SACK_PERMITTED;
1030 * PCB may have changed, pick up new values.
1033 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
1034 if (syncache_respond(sc, m) == 0) {
1035 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot],
1037 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot);
1038 tcpstat.tcps_sndacks++;
1039 tcpstat.tcps_sndtotal++;
1045 * Fill in the syncache values.
1047 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO);
1048 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
1049 sc->sc_ipopts = ipopts;
1050 sc->sc_inc.inc_fport = inc->inc_fport;
1051 sc->sc_inc.inc_lport = inc->inc_lport;
1054 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1055 if (inc->inc_isipv6) {
1056 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1057 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1058 sc->sc_route6.ro_rt = NULL;
1062 sc->sc_inc.inc_faddr = inc->inc_faddr;
1063 sc->sc_inc.inc_laddr = inc->inc_laddr;
1064 sc->sc_route.ro_rt = NULL;
1066 sc->sc_irs = th->th_seq;
1068 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0;
1070 sc->sc_iss = syncookie_generate(sc);
1072 sc->sc_iss = karc4random();
1074 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */
1075 win = ssb_space(&so->so_rcv);
1077 win = imin(win, TCP_MAXWIN);
1080 if (tcp_do_rfc1323) {
1082 * A timestamp received in a SYN makes
1083 * it ok to send timestamp requests and replies.
1085 if (to->to_flags & TOF_TS) {
1086 sc->sc_tsrecent = to->to_tsval;
1087 sc->sc_flags |= SCF_TIMESTAMP;
1089 if (to->to_flags & TOF_SCALE) {
1090 int wscale = TCP_MIN_WINSHIFT;
1092 /* Compute proper scaling value from buffer space */
1093 while (wscale < TCP_MAX_WINSHIFT &&
1094 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) {
1097 sc->sc_request_r_scale = wscale;
1098 sc->sc_requested_s_scale = to->to_requested_s_scale;
1099 sc->sc_flags |= SCF_WINSCALE;
1102 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1103 sc->sc_flags |= SCF_SACK_PERMITTED;
1104 if (tp->t_flags & TF_NOOPT)
1105 sc->sc_flags = SCF_NOOPT;
1106 #ifdef TCP_SIGNATURE
1108 * If listening socket requested TCP digests, and received SYN
1109 * contains the option, flag this in the syncache so that
1110 * syncache_respond() will do the right thing with the SYN+ACK.
1111 * XXX Currently we always record the option by default and will
1112 * attempt to use it in syncache_respond().
1114 if (to->to_flags & TOF_SIGNATURE)
1115 sc->sc_flags = SCF_SIGNATURE;
1116 #endif /* TCP_SIGNATURE */
1118 if (syncache_respond(sc, m) == 0) {
1119 syncache_insert(sc, sch);
1120 tcpstat.tcps_sndacks++;
1121 tcpstat.tcps_sndtotal++;
1124 tcpstat.tcps_sc_dropped++;
1130 syncache_respond(struct syncache *sc, struct mbuf *m)
1134 u_int16_t tlen, hlen, mssopt;
1135 struct ip *ip = NULL;
1138 struct ip6_hdr *ip6 = NULL;
1140 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1142 const boolean_t isipv6 = FALSE;
1146 rt = tcp_rtlookup6(&sc->sc_inc);
1148 mssopt = rt->rt_ifp->if_mtu -
1149 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1151 mssopt = tcp_v6mssdflt;
1152 hlen = sizeof(struct ip6_hdr);
1154 rt = tcp_rtlookup(&sc->sc_inc);
1156 mssopt = rt->rt_ifp->if_mtu -
1157 (sizeof(struct ip) + sizeof(struct tcphdr));
1159 mssopt = tcp_mssdflt;
1160 hlen = sizeof(struct ip);
1163 /* Compute the size of the TCP options. */
1164 if (sc->sc_flags & SCF_NOOPT) {
1167 optlen = TCPOLEN_MAXSEG +
1168 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) +
1169 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) +
1170 ((sc->sc_flags & SCF_SACK_PERMITTED) ?
1171 TCPOLEN_SACK_PERMITTED_ALIGNED : 0);
1172 #ifdef TCP_SIGNATURE
1173 optlen += ((sc->sc_flags & SCF_SIGNATURE) ?
1174 (TCPOLEN_SIGNATURE + 2) : 0);
1175 #endif /* TCP_SIGNATURE */
1177 tlen = hlen + sizeof(struct tcphdr) + optlen;
1181 * assume that the entire packet will fit in a header mbuf
1183 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small"));
1186 * XXX shouldn't this reuse the mbuf if possible ?
1187 * Create the IP+TCP header from scratch.
1192 m = m_gethdr(MB_DONTWAIT, MT_HEADER);
1195 m->m_data += max_linkhdr;
1197 m->m_pkthdr.len = tlen;
1198 m->m_pkthdr.rcvif = NULL;
1201 ip6 = mtod(m, struct ip6_hdr *);
1202 ip6->ip6_vfc = IPV6_VERSION;
1203 ip6->ip6_nxt = IPPROTO_TCP;
1204 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1205 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1206 ip6->ip6_plen = htons(tlen - hlen);
1207 /* ip6_hlim is set after checksum */
1208 /* ip6_flow = ??? */
1210 th = (struct tcphdr *)(ip6 + 1);
1212 ip = mtod(m, struct ip *);
1213 ip->ip_v = IPVERSION;
1214 ip->ip_hl = sizeof(struct ip) >> 2;
1219 ip->ip_p = IPPROTO_TCP;
1220 ip->ip_src = sc->sc_inc.inc_laddr;
1221 ip->ip_dst = sc->sc_inc.inc_faddr;
1222 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */
1223 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */
1226 * See if we should do MTU discovery. Route lookups are
1227 * expensive, so we will only unset the DF bit if:
1229 * 1) path_mtu_discovery is disabled
1230 * 2) the SCF_UNREACH flag has been set
1232 if (path_mtu_discovery
1233 && ((sc->sc_flags & SCF_UNREACH) == 0)) {
1234 ip->ip_off |= IP_DF;
1237 th = (struct tcphdr *)(ip + 1);
1239 th->th_sport = sc->sc_inc.inc_lport;
1240 th->th_dport = sc->sc_inc.inc_fport;
1242 th->th_seq = htonl(sc->sc_iss);
1243 th->th_ack = htonl(sc->sc_irs + 1);
1244 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1246 th->th_flags = TH_SYN | TH_ACK;
1247 th->th_win = htons(sc->sc_wnd);
1250 /* Tack on the TCP options. */
1253 optp = (u_int8_t *)(th + 1);
1254 *optp++ = TCPOPT_MAXSEG;
1255 *optp++ = TCPOLEN_MAXSEG;
1256 *optp++ = (mssopt >> 8) & 0xff;
1257 *optp++ = mssopt & 0xff;
1259 if (sc->sc_flags & SCF_WINSCALE) {
1260 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
1261 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
1262 sc->sc_request_r_scale);
1266 if (sc->sc_flags & SCF_TIMESTAMP) {
1267 u_int32_t *lp = (u_int32_t *)(optp);
1269 /* Form timestamp option as shown in appendix A of RFC 1323. */
1270 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1271 *lp++ = htonl(ticks);
1272 *lp = htonl(sc->sc_tsrecent);
1273 optp += TCPOLEN_TSTAMP_APPA;
1276 #ifdef TCP_SIGNATURE
1278 * Handle TCP-MD5 passive opener response.
1280 if (sc->sc_flags & SCF_SIGNATURE) {
1281 u_int8_t *bp = optp;
1284 *bp++ = TCPOPT_SIGNATURE;
1285 *bp++ = TCPOLEN_SIGNATURE;
1286 for (i = 0; i < TCP_SIGLEN; i++)
1288 tcpsignature_compute(m, 0, optlen,
1289 optp + 2, IPSEC_DIR_OUTBOUND);
1292 optp += TCPOLEN_SIGNATURE + 2;
1294 #endif /* TCP_SIGNATURE */
1296 if (sc->sc_flags & SCF_SACK_PERMITTED) {
1297 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED);
1298 optp += TCPOLEN_SACK_PERMITTED_ALIGNED;
1303 struct route_in6 *ro6 = &sc->sc_route6;
1306 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
1307 ip6->ip6_hlim = in6_selecthlim(NULL,
1308 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL);
1309 error = ip6_output(m, NULL, ro6, 0, NULL, NULL,
1310 sc->sc_tp->t_inpcb);
1312 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1313 htons(tlen - hlen + IPPROTO_TCP));
1314 m->m_pkthdr.csum_flags = CSUM_TCP;
1315 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1316 error = ip_output(m, sc->sc_ipopts, &sc->sc_route,
1317 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb);
1325 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .|
1327 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .|
1329 * (A): peer mss index
1333 * The values below are chosen to minimize the size of the tcp_secret
1334 * table, as well as providing roughly a 16 second lifetime for the cookie.
1337 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */
1338 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */
1340 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1)
1341 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS)
1342 #define SYNCOOKIE_TIMEOUT \
1343 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT))
1344 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK)
1347 u_int32_t ts_secbits[4];
1349 } tcp_secret[SYNCOOKIE_NSECRETS];
1351 static int tcp_msstab[] = { 0, 536, 1460, 8960 };
1353 static MD5_CTX syn_ctx;
1355 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v))
1358 u_int32_t laddr, faddr;
1359 u_int32_t secbits[4];
1360 u_int16_t lport, fport;
1364 CTASSERT(sizeof(struct md5_add) == 28);
1368 * Consider the problem of a recreated (and retransmitted) cookie. If the
1369 * original SYN was accepted, the connection is established. The second
1370 * SYN is inflight, and if it arrives with an ISN that falls within the
1371 * receive window, the connection is killed.
1373 * However, since cookies have other problems, this may not be worth
1378 syncookie_generate(struct syncache *sc)
1380 u_int32_t md5_buffer[4];
1385 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1387 const boolean_t isipv6 = FALSE;
1390 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK;
1391 if (tcp_secret[idx].ts_expire < ticks) {
1392 for (i = 0; i < 4; i++)
1393 tcp_secret[idx].ts_secbits[i] = karc4random();
1394 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT;
1396 for (data = NELEM(tcp_msstab) - 1; data > 0; data--)
1397 if (tcp_msstab[data] <= sc->sc_peer_mss)
1399 data = (data << SYNCOOKIE_WNDBITS) | idx;
1400 data ^= sc->sc_irs; /* peer's iss */
1403 MD5Add(sc->sc_inc.inc6_laddr);
1404 MD5Add(sc->sc_inc.inc6_faddr);
1408 add.laddr = sc->sc_inc.inc_laddr.s_addr;
1409 add.faddr = sc->sc_inc.inc_faddr.s_addr;
1411 add.lport = sc->sc_inc.inc_lport;
1412 add.fport = sc->sc_inc.inc_fport;
1413 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1414 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1415 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1416 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1418 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1419 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK);
1423 static struct syncache *
1424 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so)
1426 u_int32_t md5_buffer[4];
1427 struct syncache *sc;
1432 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */
1433 idx = data & SYNCOOKIE_WNDMASK;
1434 if (tcp_secret[idx].ts_expire < ticks ||
1435 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks)
1439 if (inc->inc_isipv6) {
1440 MD5Add(inc->inc6_laddr);
1441 MD5Add(inc->inc6_faddr);
1447 add.laddr = inc->inc_laddr.s_addr;
1448 add.faddr = inc->inc_faddr.s_addr;
1450 add.lport = inc->inc_lport;
1451 add.fport = inc->inc_fport;
1452 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1453 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1454 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1455 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1457 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1458 data ^= md5_buffer[0];
1459 if (data & ~SYNCOOKIE_DATAMASK)
1461 data = data >> SYNCOOKIE_WNDBITS;
1464 * Fill in the syncache values.
1465 * XXX duplicate code from syncache_add
1467 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO);
1468 sc->sc_ipopts = NULL;
1469 sc->sc_inc.inc_fport = inc->inc_fport;
1470 sc->sc_inc.inc_lport = inc->inc_lport;
1472 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1473 if (inc->inc_isipv6) {
1474 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1475 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1476 sc->sc_route6.ro_rt = NULL;
1480 sc->sc_inc.inc_faddr = inc->inc_faddr;
1481 sc->sc_inc.inc_laddr = inc->inc_laddr;
1482 sc->sc_route.ro_rt = NULL;
1484 sc->sc_irs = th->th_seq - 1;
1485 sc->sc_iss = th->th_ack - 1;
1486 wnd = ssb_space(&so->so_rcv);
1488 wnd = imin(wnd, TCP_MAXWIN);
1492 sc->sc_peer_mss = tcp_msstab[data];