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 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
37 * License terms: all terms for the DragonFly license above plus the following:
39 * 4. All advertising materials mentioning features or use of this software
40 * must display the following acknowledgement:
42 * This product includes software developed by Jeffrey M. Hsu
43 * for the DragonFly Project.
45 * This requirement may be waived with permission from Jeffrey Hsu.
46 * This requirement will sunset and may be removed on July 8 2005,
47 * after which the standard DragonFly license (as shown above) will
52 * All advertising materials mentioning features or use of this software
53 * must display the following acknowledgement:
54 * This product includes software developed by Jeffrey M. Hsu.
56 * Copyright (c) 2001 Networks Associates Technologies, Inc.
57 * All rights reserved.
59 * This software was developed for the FreeBSD Project by Jonathan Lemon
60 * and NAI Labs, the Security Research Division of Network Associates, Inc.
61 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
62 * DARPA CHATS research program.
64 * Redistribution and use in source and binary forms, with or without
65 * modification, are permitted provided that the following conditions
67 * 1. Redistributions of source code must retain the above copyright
68 * notice, this list of conditions and the following disclaimer.
69 * 2. Redistributions in binary form must reproduce the above copyright
70 * notice, this list of conditions and the following disclaimer in the
71 * documentation and/or other materials provided with the distribution.
72 * 3. The name of the author may not be used to endorse or promote
73 * products derived from this software without specific prior written
76 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
77 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
78 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
79 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
80 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
81 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
82 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
83 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
84 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
85 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
88 * $FreeBSD: src/sys/netinet/tcp_syncache.c,v 1.5.2.14 2003/02/24 04:02:27 silby Exp $
89 * $DragonFly: src/sys/netinet/tcp_syncache.c,v 1.15 2004/07/08 22:07:35 hsu Exp $
92 #include "opt_inet6.h"
93 #include "opt_ipsec.h"
95 #include <sys/param.h>
96 #include <sys/systm.h>
97 #include <sys/kernel.h>
98 #include <sys/sysctl.h>
99 #include <sys/malloc.h>
100 #include <sys/mbuf.h>
102 #include <sys/proc.h> /* for proc0 declaration */
103 #include <sys/random.h>
104 #include <sys/socket.h>
105 #include <sys/socketvar.h>
106 #include <sys/in_cksum.h>
109 #include <net/route.h>
111 #include <netinet/in.h>
112 #include <netinet/in_systm.h>
113 #include <netinet/ip.h>
114 #include <netinet/in_var.h>
115 #include <netinet/in_pcb.h>
116 #include <netinet/ip_var.h>
118 #include <netinet/ip6.h>
119 #include <netinet/icmp6.h>
120 #include <netinet6/nd6.h>
121 #include <netinet6/ip6_var.h>
122 #include <netinet6/in6_pcb.h>
124 #include <netinet/tcp.h>
125 #include <netinet/tcp_fsm.h>
126 #include <netinet/tcp_seq.h>
127 #include <netinet/tcp_timer.h>
128 #include <netinet/tcp_var.h>
130 #include <netinet6/tcp6_var.h>
134 #include <netinet6/ipsec.h>
136 #include <netinet6/ipsec6.h>
138 #include <netproto/key/key.h>
142 #include <netipsec/ipsec.h>
144 #include <netipsec/ipsec6.h>
146 #include <netipsec/key.h>
148 #endif /*FAST_IPSEC*/
150 #include <vm/vm_zone.h>
152 static int tcp_syncookies = 1;
153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
155 "Use TCP SYN cookies if the syncache overflows");
157 static void syncache_drop(struct syncache *, struct syncache_head *);
158 static void syncache_free(struct syncache *);
159 static void syncache_insert(struct syncache *, struct syncache_head *);
160 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
161 static int syncache_respond(struct syncache *, struct mbuf *);
162 static struct socket *syncache_socket(struct syncache *, struct socket *);
163 static void syncache_timer(void *);
164 static u_int32_t syncookie_generate(struct syncache *);
165 static struct syncache *syncookie_lookup(struct in_conninfo *,
166 struct tcphdr *, struct socket *);
169 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
170 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds,
171 * the odds are that the user has given up attempting to connect by then.
173 #define SYNCACHE_MAXREXMTS 3
175 /* Arbitrary values */
176 #define TCP_SYNCACHE_HASHSIZE 512
177 #define TCP_SYNCACHE_BUCKETLIMIT 30
179 struct tcp_syncache {
180 struct syncache_head *hashbase;
181 struct vm_zone *zone;
189 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1];
190 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1];
192 static struct tcp_syncache tcp_syncache;
194 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
196 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD,
197 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
199 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD,
200 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
202 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
203 &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
205 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD,
206 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
208 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
209 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
211 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
213 #define SYNCACHE_HASH(inc, mask) \
214 ((tcp_syncache.hash_secret ^ \
215 (inc)->inc_faddr.s_addr ^ \
216 ((inc)->inc_faddr.s_addr >> 16) ^ \
217 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
219 #define SYNCACHE_HASH6(inc, mask) \
220 ((tcp_syncache.hash_secret ^ \
221 (inc)->inc6_faddr.s6_addr32[0] ^ \
222 (inc)->inc6_faddr.s6_addr32[3] ^ \
223 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
225 #define ENDPTS_EQ(a, b) ( \
226 (a)->ie_fport == (b)->ie_fport && \
227 (a)->ie_lport == (b)->ie_lport && \
228 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
229 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
232 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
234 #define SYNCACHE_TIMEOUT(sc, slot) do { \
235 sc->sc_rxtslot = slot; \
236 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[slot]; \
237 TAILQ_INSERT_TAIL(&tcp_syncache.timerq[slot], sc, sc_timerq); \
238 if (!callout_active(&tcp_syncache.tt_timerq[slot])) \
239 callout_reset(&tcp_syncache.tt_timerq[slot], \
240 TCPTV_RTOBASE * tcp_backoff[slot], \
241 syncache_timer, (void *)((intptr_t)slot)); \
245 syncache_free(struct syncache *sc)
250 (void) m_free(sc->sc_ipopts);
252 if (sc->sc_inc.inc_isipv6)
253 rt = sc->sc_route6.ro_rt;
256 rt = sc->sc_route.ro_rt;
259 * If this is the only reference to a protocol cloned
260 * route, remove it immediately.
262 if (rt->rt_flags & RTF_WASCLONED &&
263 (sc->sc_flags & SCF_KEEPROUTE) == 0 &&
265 rtrequest(RTM_DELETE, rt_key(rt),
266 rt->rt_gateway, rt_mask(rt),
270 zfree(tcp_syncache.zone, sc);
278 tcp_syncache.cache_count = 0;
279 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
280 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
281 tcp_syncache.cache_limit =
282 tcp_syncache.hashsize * tcp_syncache.bucket_limit;
283 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
284 tcp_syncache.hash_secret = arc4random();
286 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
287 &tcp_syncache.hashsize);
288 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
289 &tcp_syncache.cache_limit);
290 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
291 &tcp_syncache.bucket_limit);
292 if (!powerof2(tcp_syncache.hashsize)) {
293 printf("WARNING: syncache hash size is not a power of 2.\n");
294 tcp_syncache.hashsize = 512; /* safe default */
296 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
298 /* Allocate the hash table. */
299 MALLOC(tcp_syncache.hashbase, struct syncache_head *,
300 tcp_syncache.hashsize * sizeof(struct syncache_head),
301 M_SYNCACHE, M_WAITOK);
303 /* Initialize the hash buckets. */
304 for (i = 0; i < tcp_syncache.hashsize; i++) {
305 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket);
306 tcp_syncache.hashbase[i].sch_length = 0;
309 /* Initialize the timer queues. */
310 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) {
311 TAILQ_INIT(&tcp_syncache.timerq[i]);
312 callout_init(&tcp_syncache.tt_timerq[i]);
316 * Allocate the syncache entries. Allow the zone to allocate one
317 * more entry than cache limit, so a new entry can bump out an
320 tcp_syncache.zone = zinit("syncache", sizeof(struct syncache),
321 tcp_syncache.cache_limit, ZONE_INTERRUPT, 0);
322 tcp_syncache.cache_limit -= 1;
326 syncache_insert(sc, sch)
328 struct syncache_head *sch;
330 struct syncache *sc2;
334 * Make sure that we don't overflow the per-bucket
335 * limit or the total cache size limit.
337 if (sch->sch_length >= tcp_syncache.bucket_limit) {
339 * The bucket is full, toss the oldest element.
341 sc2 = TAILQ_FIRST(&sch->sch_bucket);
342 sc2->sc_tp->ts_recent = ticks;
343 syncache_drop(sc2, sch);
344 tcpstat.tcps_sc_bucketoverflow++;
345 } else if (tcp_syncache.cache_count >= tcp_syncache.cache_limit) {
347 * The cache is full. Toss the oldest entry in the
348 * entire cache. This is the front entry in the
349 * first non-empty timer queue with the largest
352 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
353 sc2 = TAILQ_FIRST(&tcp_syncache.timerq[i]);
357 sc2->sc_tp->ts_recent = ticks;
358 syncache_drop(sc2, NULL);
359 tcpstat.tcps_sc_cacheoverflow++;
362 /* Initialize the entry's timer. */
363 SYNCACHE_TIMEOUT(sc, 0);
365 /* Put it into the bucket. */
366 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash);
368 tcp_syncache.cache_count++;
369 tcpstat.tcps_sc_added++;
373 syncache_drop(sc, sch)
375 struct syncache_head *sch;
380 if (sc->sc_inc.inc_isipv6) {
381 sch = &tcp_syncache.hashbase[
382 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)];
386 sch = &tcp_syncache.hashbase[
387 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)];
391 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
393 tcp_syncache.cache_count--;
395 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], sc, sc_timerq);
396 if (TAILQ_EMPTY(&tcp_syncache.timerq[sc->sc_rxtslot]))
397 callout_stop(&tcp_syncache.tt_timerq[sc->sc_rxtslot]);
403 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
404 * If we have retransmitted an entry the maximum number of times, expire it.
407 syncache_timer(xslot)
410 intptr_t slot = (intptr_t)xslot;
411 struct syncache *sc, *nsc;
416 if (callout_pending(&tcp_syncache.tt_timerq[slot]) ||
417 !callout_active(&tcp_syncache.tt_timerq[slot])) {
421 callout_deactivate(&tcp_syncache.tt_timerq[slot]);
423 nsc = TAILQ_FIRST(&tcp_syncache.timerq[slot]);
424 while (nsc != NULL) {
425 if (ticks < nsc->sc_rxttime)
428 inp = sc->sc_tp->t_inpcb;
429 if (slot == SYNCACHE_MAXREXMTS ||
430 slot >= tcp_syncache.rexmt_limit ||
431 inp->inp_gencnt != sc->sc_inp_gencnt) {
432 nsc = TAILQ_NEXT(sc, sc_timerq);
433 syncache_drop(sc, NULL);
434 tcpstat.tcps_sc_stale++;
438 * syncache_respond() may call back into the syncache to
439 * to modify another entry, so do not obtain the next
440 * entry on the timer chain until it has completed.
442 (void) syncache_respond(sc, NULL);
443 nsc = TAILQ_NEXT(sc, sc_timerq);
444 tcpstat.tcps_sc_retransmitted++;
445 TAILQ_REMOVE(&tcp_syncache.timerq[slot], sc, sc_timerq);
446 SYNCACHE_TIMEOUT(sc, slot + 1);
449 callout_reset(&tcp_syncache.tt_timerq[slot],
450 nsc->sc_rxttime - ticks, syncache_timer, (void *)(slot));
455 * Find an entry in the syncache.
458 syncache_lookup(inc, schp)
459 struct in_conninfo *inc;
460 struct syncache_head **schp;
463 struct syncache_head *sch;
466 if (inc->inc_isipv6) {
467 sch = &tcp_syncache.hashbase[
468 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
470 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash)
471 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
476 sch = &tcp_syncache.hashbase[
477 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
479 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
481 if (sc->sc_inc.inc_isipv6)
484 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
492 * This function is called when we get a RST for a
493 * non-existent connection, so that we can see if the
494 * connection is in the syn cache. If it is, zap it.
497 syncache_chkrst(inc, th)
498 struct in_conninfo *inc;
502 struct syncache_head *sch;
504 sc = syncache_lookup(inc, &sch);
508 * If the RST bit is set, check the sequence number to see
509 * if this is a valid reset segment.
511 * In all states except SYN-SENT, all reset (RST) segments
512 * are validated by checking their SEQ-fields. A reset is
513 * valid if its sequence number is in the window.
515 * The sequence number in the reset segment is normally an
516 * echo of our outgoing acknowlegement numbers, but some hosts
517 * send a reset with the sequence number at the rightmost edge
518 * of our receive window, and we have to handle this case.
520 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
521 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
522 syncache_drop(sc, sch);
523 tcpstat.tcps_sc_reset++;
529 struct in_conninfo *inc;
532 struct syncache_head *sch;
534 sc = syncache_lookup(inc, &sch);
536 syncache_drop(sc, sch);
537 tcpstat.tcps_sc_badack++;
542 syncache_unreach(inc, th)
543 struct in_conninfo *inc;
547 struct syncache_head *sch;
549 /* we are called at splnet() here */
550 sc = syncache_lookup(inc, &sch);
554 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
555 if (ntohl(th->th_seq) != sc->sc_iss)
559 * If we've rertransmitted 3 times and this is our second error,
560 * we remove the entry. Otherwise, we allow it to continue on.
561 * This prevents us from incorrectly nuking an entry during a
562 * spurious network outage.
566 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) {
567 sc->sc_flags |= SCF_UNREACH;
570 syncache_drop(sc, sch);
571 tcpstat.tcps_sc_unreach++;
575 * Build a new TCP socket structure from a syncache entry.
577 static struct socket *
578 syncache_socket(sc, lso)
582 struct inpcb *inp = NULL;
587 * Ok, create the full blown connection, and set things up
588 * as they would have been set up if we had created the
589 * connection when the SYN arrived. If we can't create
590 * the connection, abort it.
592 so = sonewconn(lso, SS_ISCONNECTED);
595 * Drop the connection; we will send a RST if the peer
596 * retransmits the ACK,
598 tcpstat.tcps_listendrop++;
605 * Insert new socket into hash list.
607 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
609 if (sc->sc_inc.inc_isipv6) {
610 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
612 inp->inp_vflag &= ~INP_IPV6;
613 inp->inp_vflag |= INP_IPV4;
615 inp->inp_laddr = sc->sc_inc.inc_laddr;
619 inp->inp_lport = sc->sc_inc.inc_lport;
620 if (in_pcbinsporthash(inp) != 0) {
622 * Undo the assignments above if we failed to
623 * put the PCB on the hash lists.
626 if (sc->sc_inc.inc_isipv6)
627 inp->in6p_laddr = in6addr_any;
630 inp->inp_laddr.s_addr = INADDR_ANY;
635 /* copy old policy into new socket's */
636 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
637 printf("syncache_expand: could not copy policy\n");
640 if (sc->sc_inc.inc_isipv6) {
641 struct inpcb *oinp = sotoinpcb(lso);
642 struct in6_addr laddr6;
643 struct sockaddr_in6 sin6;
645 * Inherit socket options from the listening socket.
646 * Note that in6p_inputopts are not (and should not be)
647 * copied, since it stores previously received options and is
648 * used to detect if each new option is different than the
649 * previous one and hence should be passed to a user.
650 * If we copied in6p_inputopts, a user would not be able to
651 * receive options just after calling the accept system call.
653 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
654 if (oinp->in6p_outputopts)
655 inp->in6p_outputopts =
656 ip6_copypktopts(oinp->in6p_outputopts, M_INTWAIT);
657 inp->in6p_route = sc->sc_route6;
658 sc->sc_route6.ro_rt = NULL;
660 sin6.sin6_family = AF_INET6;
661 sin6.sin6_len = sizeof sin6;
662 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
663 sin6.sin6_port = sc->sc_inc.inc_fport;
664 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
665 laddr6 = inp->in6p_laddr;
666 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
667 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
668 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) {
669 inp->in6p_laddr = laddr6;
675 struct in_addr laddr;
676 struct sockaddr_in sin;
678 inp->inp_options = ip_srcroute();
679 if (inp->inp_options == NULL) {
680 inp->inp_options = sc->sc_ipopts;
681 sc->sc_ipopts = NULL;
683 inp->inp_route = sc->sc_route;
684 sc->sc_route.ro_rt = NULL;
686 sin.sin_family = AF_INET;
687 sin.sin_len = sizeof sin;
688 sin.sin_addr = sc->sc_inc.inc_faddr;
689 sin.sin_port = sc->sc_inc.inc_fport;
690 bzero(sin.sin_zero, sizeof sin.sin_zero);
691 laddr = inp->inp_laddr;
692 if (inp->inp_laddr.s_addr == INADDR_ANY)
693 inp->inp_laddr = sc->sc_inc.inc_laddr;
694 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) {
695 inp->inp_laddr = laddr;
701 tp->t_state = TCPS_SYN_RECEIVED;
702 tp->iss = sc->sc_iss;
703 tp->irs = sc->sc_irs;
706 tp->snd_wl1 = sc->sc_irs;
707 tp->rcv_up = sc->sc_irs + 1;
708 tp->rcv_wnd = sc->sc_wnd;
709 tp->rcv_adv += tp->rcv_wnd;
711 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
712 if (sc->sc_flags & SCF_NOOPT)
713 tp->t_flags |= TF_NOOPT;
714 if (sc->sc_flags & SCF_WINSCALE) {
715 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
716 tp->requested_s_scale = sc->sc_requested_s_scale;
717 tp->request_r_scale = sc->sc_request_r_scale;
719 if (sc->sc_flags & SCF_TIMESTAMP) {
720 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
721 tp->ts_recent = sc->sc_tsrecent;
722 tp->ts_recent_age = ticks;
724 if (sc->sc_flags & SCF_CC) {
726 * Initialization of the tcpcb for transaction;
727 * set SND.WND = SEG.WND,
728 * initialize CCsend and CCrecv.
730 tp->t_flags |= TF_REQ_CC|TF_RCVD_CC;
731 tp->cc_send = sc->sc_cc_send;
732 tp->cc_recv = sc->sc_cc_recv;
735 tcp_mss(tp, sc->sc_peer_mss);
738 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
740 if (sc->sc_rxtslot != 0)
741 tp->snd_cwnd = tp->t_maxseg;
742 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
744 tcpstat.tcps_accepts++;
754 * This function gets called when we receive an ACK for a
755 * socket in the LISTEN state. We look up the connection
756 * in the syncache, and if its there, we pull it out of
757 * the cache and turn it into a full-blown connection in
758 * the SYN-RECEIVED state.
761 syncache_expand(inc, th, sop, m)
762 struct in_conninfo *inc;
768 struct syncache_head *sch;
771 sc = syncache_lookup(inc, &sch);
774 * There is no syncache entry, so see if this ACK is
775 * a returning syncookie. To do this, first:
776 * A. See if this socket has had a syncache entry dropped in
777 * the past. We don't want to accept a bogus syncookie
778 * if we've never received a SYN.
779 * B. check that the syncookie is valid. If it is, then
780 * cobble up a fake syncache entry, and return.
784 sc = syncookie_lookup(inc, th, *sop);
788 tcpstat.tcps_sc_recvcookie++;
792 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
794 if (th->th_ack != sc->sc_iss + 1)
797 so = syncache_socket(sc, *sop);
801 /* XXXjlemon check this - is this correct? */
802 (void) tcp_respond(NULL, m, m, th,
803 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK);
805 m_freem(m); /* XXX only needed for above */
806 tcpstat.tcps_sc_aborted++;
808 sc->sc_flags |= SCF_KEEPROUTE;
809 tcpstat.tcps_sc_completed++;
814 syncache_drop(sc, sch);
820 * Given a LISTEN socket and an inbound SYN request, add
821 * this to the syn cache, and send back a segment:
822 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
825 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
826 * Doing so would require that we hold onto the data and deliver it
827 * to the application. However, if we are the target of a SYN-flood
828 * DoS attack, an attacker could send data which would eventually
829 * consume all available buffer space if it were ACKed. By not ACKing
830 * the data, we avoid this DoS scenario.
833 syncache_add(inc, to, th, sop, m)
834 struct in_conninfo *inc;
842 struct syncache *sc = NULL;
843 struct syncache_head *sch;
844 struct mbuf *ipopts = NULL;
845 struct rmxp_tao *taop;
852 * Remember the IP options, if any.
855 if (!inc->inc_isipv6)
857 ipopts = ip_srcroute();
860 * See if we already have an entry for this connection.
861 * If we do, resend the SYN,ACK, and reset the retransmit timer.
864 * should the syncache be re-initialized with the contents
865 * of the new SYN here (which may have different options?)
867 sc = syncache_lookup(inc, &sch);
869 tcpstat.tcps_sc_dupsyn++;
872 * If we were remembering a previous source route,
873 * forget it and use the new one we've been given.
876 (void) m_free(sc->sc_ipopts);
877 sc->sc_ipopts = ipopts;
880 * Update timestamp if present.
882 if (sc->sc_flags & SCF_TIMESTAMP)
883 sc->sc_tsrecent = to->to_tsval;
885 * PCB may have changed, pick up new values.
888 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
889 if (syncache_respond(sc, m) == 0) {
890 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot],
892 SYNCACHE_TIMEOUT(sc, sc->sc_rxtslot);
893 tcpstat.tcps_sndacks++;
894 tcpstat.tcps_sndtotal++;
901 * This allocation is guaranteed to succeed because we
902 * preallocate one more syncache entry than cache_limit.
904 sc = zalloc(tcp_syncache.zone);
907 * Fill in the syncache values.
910 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
911 sc->sc_ipopts = ipopts;
912 sc->sc_inc.inc_fport = inc->inc_fport;
913 sc->sc_inc.inc_lport = inc->inc_lport;
915 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
916 if (inc->inc_isipv6) {
917 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
918 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
919 sc->sc_route6.ro_rt = NULL;
923 sc->sc_inc.inc_faddr = inc->inc_faddr;
924 sc->sc_inc.inc_laddr = inc->inc_laddr;
925 sc->sc_route.ro_rt = NULL;
927 sc->sc_irs = th->th_seq;
929 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0;
931 sc->sc_iss = syncookie_generate(sc);
933 sc->sc_iss = arc4random();
935 /* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */
936 win = sbspace(&so->so_rcv);
938 win = imin(win, TCP_MAXWIN);
941 if (tcp_do_rfc1323) {
943 * A timestamp received in a SYN makes
944 * it ok to send timestamp requests and replies.
946 if (to->to_flags & TOF_TS) {
947 sc->sc_tsrecent = to->to_tsval;
948 sc->sc_flags |= SCF_TIMESTAMP;
950 if (to->to_flags & TOF_SCALE) {
953 /* Compute proper scaling value from buffer space */
954 while (wscale < TCP_MAX_WINSHIFT &&
955 (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat)
957 sc->sc_request_r_scale = wscale;
958 sc->sc_requested_s_scale = to->to_requested_s_scale;
959 sc->sc_flags |= SCF_WINSCALE;
962 if (tcp_do_rfc1644) {
964 * A CC or CC.new option received in a SYN makes
965 * it ok to send CC in subsequent segments.
967 if (to->to_flags & (TOF_CC|TOF_CCNEW)) {
968 sc->sc_cc_recv = to->to_cc;
969 sc->sc_cc_send = CC_INC(tcp_ccgen);
970 sc->sc_flags |= SCF_CC;
973 if (tp->t_flags & TF_NOOPT)
974 sc->sc_flags = SCF_NOOPT;
978 * We have the option here of not doing TAO (even if the segment
979 * qualifies) and instead fall back to a normal 3WHS via the syncache.
980 * This allows us to apply synflood protection to TAO-qualifying SYNs
981 * also. However, there should be a hueristic to determine when to
982 * do this, and is not present at the moment.
986 * Perform TAO test on incoming CC (SEG.CC) option, if any.
987 * - compare SEG.CC against cached CC from the same host, if any.
988 * - if SEG.CC > chached value, SYN must be new and is accepted
989 * immediately: save new CC in the cache, mark the socket
990 * connected, enter ESTABLISHED state, turn on flag to
991 * send a SYN in the next segment.
992 * A virtual advertised window is set in rcv_adv to
993 * initialize SWS prevention. Then enter normal segment
994 * processing: drop SYN, process data and FIN.
995 * - otherwise do a normal 3-way handshake.
997 taop = tcp_gettaocache(&sc->sc_inc);
998 if ((to->to_flags & TOF_CC) != 0) {
999 if (((tp->t_flags & TF_NOPUSH) != 0) &&
1000 sc->sc_flags & SCF_CC &&
1001 taop != NULL && taop->tao_cc != 0 &&
1002 CC_GT(to->to_cc, taop->tao_cc)) {
1004 so = syncache_socket(sc, *sop);
1006 sc->sc_flags |= SCF_KEEPROUTE;
1007 taop->tao_cc = to->to_cc;
1011 return (so != NULL);
1015 * No CC option, but maybe CC.NEW: invalidate cached value.
1021 * TAO test failed or there was no CC option,
1022 * do a standard 3-way handshake.
1024 if (syncache_respond(sc, m) == 0) {
1025 syncache_insert(sc, sch);
1026 tcpstat.tcps_sndacks++;
1027 tcpstat.tcps_sndtotal++;
1030 tcpstat.tcps_sc_dropped++;
1037 syncache_respond(sc, m)
1038 struct syncache *sc;
1043 u_int16_t tlen, hlen, mssopt;
1044 struct ip *ip = NULL;
1048 struct ip6_hdr *ip6 = NULL;
1052 if (sc->sc_inc.inc_isipv6) {
1053 rt = tcp_rtlookup6(&sc->sc_inc);
1055 mssopt = rt->rt_ifp->if_mtu -
1056 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1058 mssopt = tcp_v6mssdflt;
1059 hlen = sizeof(struct ip6_hdr);
1063 rt = tcp_rtlookup(&sc->sc_inc);
1065 mssopt = rt->rt_ifp->if_mtu -
1066 (sizeof(struct ip) + sizeof(struct tcphdr));
1068 mssopt = tcp_mssdflt;
1069 hlen = sizeof(struct ip);
1072 /* Compute the size of the TCP options. */
1073 if (sc->sc_flags & SCF_NOOPT) {
1076 optlen = TCPOLEN_MAXSEG +
1077 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) +
1078 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) +
1079 ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0);
1081 tlen = hlen + sizeof(struct tcphdr) + optlen;
1085 * assume that the entire packet will fit in a header mbuf
1087 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small"));
1090 * XXX shouldn't this reuse the mbuf if possible ?
1091 * Create the IP+TCP header from scratch.
1096 m = m_gethdr(MB_DONTWAIT, MT_HEADER);
1099 m->m_data += max_linkhdr;
1101 m->m_pkthdr.len = tlen;
1102 m->m_pkthdr.rcvif = NULL;
1105 if (sc->sc_inc.inc_isipv6) {
1106 ip6 = mtod(m, struct ip6_hdr *);
1107 ip6->ip6_vfc = IPV6_VERSION;
1108 ip6->ip6_nxt = IPPROTO_TCP;
1109 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1110 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1111 ip6->ip6_plen = htons(tlen - hlen);
1112 /* ip6_hlim is set after checksum */
1113 /* ip6_flow = ??? */
1115 th = (struct tcphdr *)(ip6 + 1);
1119 ip = mtod(m, struct ip *);
1120 ip->ip_v = IPVERSION;
1121 ip->ip_hl = sizeof(struct ip) >> 2;
1126 ip->ip_p = IPPROTO_TCP;
1127 ip->ip_src = sc->sc_inc.inc_laddr;
1128 ip->ip_dst = sc->sc_inc.inc_faddr;
1129 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */
1130 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */
1133 * See if we should do MTU discovery. Route lookups are expensive,
1134 * so we will only unset the DF bit if:
1136 * 1) path_mtu_discovery is disabled
1137 * 2) the SCF_UNREACH flag has been set
1139 if (path_mtu_discovery
1140 && ((sc->sc_flags & SCF_UNREACH) == 0)) {
1141 ip->ip_off |= IP_DF;
1144 th = (struct tcphdr *)(ip + 1);
1146 th->th_sport = sc->sc_inc.inc_lport;
1147 th->th_dport = sc->sc_inc.inc_fport;
1149 th->th_seq = htonl(sc->sc_iss);
1150 th->th_ack = htonl(sc->sc_irs + 1);
1151 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1153 th->th_flags = TH_SYN|TH_ACK;
1154 th->th_win = htons(sc->sc_wnd);
1157 /* Tack on the TCP options. */
1160 optp = (u_int8_t *)(th + 1);
1161 *optp++ = TCPOPT_MAXSEG;
1162 *optp++ = TCPOLEN_MAXSEG;
1163 *optp++ = (mssopt >> 8) & 0xff;
1164 *optp++ = mssopt & 0xff;
1166 if (sc->sc_flags & SCF_WINSCALE) {
1167 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
1168 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
1169 sc->sc_request_r_scale);
1173 if (sc->sc_flags & SCF_TIMESTAMP) {
1174 u_int32_t *lp = (u_int32_t *)(optp);
1176 /* Form timestamp option as shown in appendix A of RFC 1323. */
1177 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1178 *lp++ = htonl(ticks);
1179 *lp = htonl(sc->sc_tsrecent);
1180 optp += TCPOLEN_TSTAMP_APPA;
1184 * Send CC and CC.echo if we received CC from our peer.
1186 if (sc->sc_flags & SCF_CC) {
1187 u_int32_t *lp = (u_int32_t *)(optp);
1189 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC));
1190 *lp++ = htonl(sc->sc_cc_send);
1191 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO));
1192 *lp = htonl(sc->sc_cc_recv);
1193 optp += TCPOLEN_CC_APPA * 2;
1198 if (sc->sc_inc.inc_isipv6) {
1199 struct route_in6 *ro6 = &sc->sc_route6;
1202 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
1203 ip6->ip6_hlim = in6_selecthlim(NULL,
1204 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL);
1205 error = ip6_output(m, NULL, ro6, 0, NULL, NULL,
1206 sc->sc_tp->t_inpcb);
1210 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1211 htons(tlen - hlen + IPPROTO_TCP));
1212 m->m_pkthdr.csum_flags = CSUM_TCP;
1213 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1214 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 0, NULL,
1215 sc->sc_tp->t_inpcb);
1223 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .|
1225 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .|
1227 * (A): peer mss index
1231 * The values below are chosen to minimize the size of the tcp_secret
1232 * table, as well as providing roughly a 16 second lifetime for the cookie.
1235 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */
1236 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */
1238 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1)
1239 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS)
1240 #define SYNCOOKIE_TIMEOUT \
1241 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT))
1242 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK)
1245 u_int32_t ts_secbits[4];
1247 } tcp_secret[SYNCOOKIE_NSECRETS];
1249 static int tcp_msstab[] = { 0, 536, 1460, 8960 };
1251 static MD5_CTX syn_ctx;
1253 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v))
1256 u_int32_t laddr, faddr;
1257 u_int32_t secbits[4];
1258 u_int16_t lport, fport;
1262 CTASSERT(sizeof(struct md5_add) == 28);
1266 * Consider the problem of a recreated (and retransmitted) cookie. If the
1267 * original SYN was accepted, the connection is established. The second
1268 * SYN is inflight, and if it arrives with an ISN that falls within the
1269 * receive window, the connection is killed.
1271 * However, since cookies have other problems, this may not be worth
1276 syncookie_generate(struct syncache *sc)
1278 u_int32_t md5_buffer[4];
1283 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK;
1284 if (tcp_secret[idx].ts_expire < ticks) {
1285 for (i = 0; i < 4; i++)
1286 tcp_secret[idx].ts_secbits[i] = arc4random();
1287 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT;
1289 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--)
1290 if (tcp_msstab[data] <= sc->sc_peer_mss)
1292 data = (data << SYNCOOKIE_WNDBITS) | idx;
1293 data ^= sc->sc_irs; /* peer's iss */
1296 if (sc->sc_inc.inc_isipv6) {
1297 MD5Add(sc->sc_inc.inc6_laddr);
1298 MD5Add(sc->sc_inc.inc6_faddr);
1304 add.laddr = sc->sc_inc.inc_laddr.s_addr;
1305 add.faddr = sc->sc_inc.inc_faddr.s_addr;
1307 add.lport = sc->sc_inc.inc_lport;
1308 add.fport = sc->sc_inc.inc_fport;
1309 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1310 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1311 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1312 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1314 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1315 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK);
1319 static struct syncache *
1320 syncookie_lookup(inc, th, so)
1321 struct in_conninfo *inc;
1325 u_int32_t md5_buffer[4];
1326 struct syncache *sc;
1331 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */
1332 idx = data & SYNCOOKIE_WNDMASK;
1333 if (tcp_secret[idx].ts_expire < ticks ||
1334 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks)
1338 if (inc->inc_isipv6) {
1339 MD5Add(inc->inc6_laddr);
1340 MD5Add(inc->inc6_faddr);
1346 add.laddr = inc->inc_laddr.s_addr;
1347 add.faddr = inc->inc_faddr.s_addr;
1349 add.lport = inc->inc_lport;
1350 add.fport = inc->inc_fport;
1351 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1352 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1353 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1354 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1356 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1357 data ^= md5_buffer[0];
1358 if ((data & ~SYNCOOKIE_DATAMASK) != 0)
1360 data = data >> SYNCOOKIE_WNDBITS;
1363 * This allocation is guaranteed to succeed because we
1364 * preallocate one more syncache entry than cache_limit.
1366 sc = zalloc(tcp_syncache.zone);
1369 * Fill in the syncache values.
1370 * XXX duplicate code from syncache_add
1372 sc->sc_ipopts = NULL;
1373 sc->sc_inc.inc_fport = inc->inc_fport;
1374 sc->sc_inc.inc_lport = inc->inc_lport;
1376 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1377 if (inc->inc_isipv6) {
1378 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1379 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1380 sc->sc_route6.ro_rt = NULL;
1384 sc->sc_inc.inc_faddr = inc->inc_faddr;
1385 sc->sc_inc.inc_laddr = inc->inc_laddr;
1386 sc->sc_route.ro_rt = NULL;
1388 sc->sc_irs = th->th_seq - 1;
1389 sc->sc_iss = th->th_ack - 1;
1390 wnd = sbspace(&so->so_rcv);
1392 wnd = imin(wnd, TCP_MAXWIN);
1396 sc->sc_peer_mss = tcp_msstab[data];