2 * Copyright (c) 2001 McAfee, Inc.
5 * This software was developed for the FreeBSD Project by Jonathan Lemon
6 * and McAfee Research, the Security Research Division of McAfee, Inc. under
7 * DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
8 * DARPA CHATS research program.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 #include "opt_inet6.h"
36 #include "opt_ipsec.h"
38 #include "opt_tcpdebug.h"
39 #include "opt_tcp_sack.h"
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
45 #include <sys/malloc.h>
49 #include <sys/proc.h> /* for proc0 declaration */
50 #include <sys/random.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
55 #include <net/route.h>
57 #include <netinet/in.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/ip.h>
60 #include <netinet/in_var.h>
61 #include <netinet/in_pcb.h>
62 #include <netinet/ip_var.h>
63 #include <netinet/ip_options.h>
65 #include <netinet/ip6.h>
66 #include <netinet/icmp6.h>
67 #include <netinet6/nd6.h>
68 #include <netinet6/ip6_var.h>
69 #include <netinet6/in6_pcb.h>
71 #include <netinet/tcp.h>
73 #include <netinet/tcpip.h>
75 #include <netinet/tcp_fsm.h>
76 #include <netinet/tcp_seq.h>
77 #include <netinet/tcp_timer.h>
78 #include <netinet/tcp_var.h>
80 #include <netinet/tcp_debug.h>
83 #include <netinet6/tcp6_var.h>
87 #include <netinet6/ipsec.h>
89 #include <netinet6/ipsec6.h>
94 #include <netipsec/ipsec.h>
96 #include <netipsec/ipsec6.h>
98 #include <netipsec/key.h>
101 #include <machine/in_cksum.h>
104 static int tcp_syncookies = 1;
105 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
107 "Use TCP SYN cookies if the syncache overflows");
109 static void syncache_drop(struct syncache *, struct syncache_head *);
110 static void syncache_free(struct syncache *);
111 static void syncache_insert(struct syncache *, struct syncache_head *);
112 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
114 static int syncache_respond(struct syncache *, struct mbuf *, struct socket *);
116 static int syncache_respond(struct syncache *, struct mbuf *);
118 static struct socket *syncache_socket(struct syncache *, struct socket *,
120 static void syncache_timer(void *);
121 static u_int32_t syncookie_generate(struct syncache *, u_int32_t *);
122 static struct syncache *syncookie_lookup(struct in_conninfo *,
123 struct tcphdr *, struct socket *);
126 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
127 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds,
128 * the odds are that the user has given up attempting to connect by then.
130 #define SYNCACHE_MAXREXMTS 3
132 /* Arbitrary values */
133 #define TCP_SYNCACHE_HASHSIZE 512
134 #define TCP_SYNCACHE_BUCKETLIMIT 30
136 struct tcp_syncache {
137 struct syncache_head *hashbase;
146 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1];
147 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1];
149 static struct tcp_syncache tcp_syncache;
151 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
153 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
154 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
156 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
157 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
159 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
160 &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
162 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
163 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
165 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
166 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
168 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
170 #define SYNCACHE_HASH(inc, mask) \
171 ((tcp_syncache.hash_secret ^ \
172 (inc)->inc_faddr.s_addr ^ \
173 ((inc)->inc_faddr.s_addr >> 16) ^ \
174 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
176 #define SYNCACHE_HASH6(inc, mask) \
177 ((tcp_syncache.hash_secret ^ \
178 (inc)->inc6_faddr.s6_addr32[0] ^ \
179 (inc)->inc6_faddr.s6_addr32[3] ^ \
180 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
182 #define ENDPTS_EQ(a, b) ( \
183 (a)->ie_fport == (b)->ie_fport && \
184 (a)->ie_lport == (b)->ie_lport && \
185 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
186 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
189 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
191 #define SYNCACHE_TIMEOUT(sc, slot) do { \
192 sc->sc_rxtslot = (slot); \
193 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[(slot)]; \
194 TAILQ_INSERT_TAIL(&tcp_syncache.timerq[(slot)], sc, sc_timerq); \
195 if (!callout_active(&tcp_syncache.tt_timerq[(slot)])) \
196 callout_reset(&tcp_syncache.tt_timerq[(slot)], \
197 TCPTV_RTOBASE * tcp_backoff[(slot)], \
198 syncache_timer, (void *)((intptr_t)(slot))); \
202 syncache_free(struct syncache *sc)
205 (void) m_free(sc->sc_ipopts);
207 uma_zfree(tcp_syncache.zone, sc);
215 tcp_syncache.cache_count = 0;
216 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
217 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
218 tcp_syncache.cache_limit =
219 tcp_syncache.hashsize * tcp_syncache.bucket_limit;
220 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
221 tcp_syncache.hash_secret = arc4random();
223 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
224 &tcp_syncache.hashsize);
225 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
226 &tcp_syncache.cache_limit);
227 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
228 &tcp_syncache.bucket_limit);
229 if (!powerof2(tcp_syncache.hashsize) || tcp_syncache.hashsize == 0) {
230 printf("WARNING: syncache hash size is not a power of 2.\n");
231 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
233 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
235 /* Allocate the hash table. */
236 MALLOC(tcp_syncache.hashbase, struct syncache_head *,
237 tcp_syncache.hashsize * sizeof(struct syncache_head),
238 M_SYNCACHE, M_WAITOK);
240 /* Initialize the hash buckets. */
241 for (i = 0; i < tcp_syncache.hashsize; i++) {
242 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket);
243 tcp_syncache.hashbase[i].sch_length = 0;
246 /* Initialize the timer queues. */
247 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) {
248 TAILQ_INIT(&tcp_syncache.timerq[i]);
249 callout_init(&tcp_syncache.tt_timerq[i], NET_CALLOUT_MPSAFE);
253 * Allocate the syncache entries. Allow the zone to allocate one
254 * more entry than cache limit, so a new entry can bump out an
257 tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
258 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
259 uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit);
260 tcp_syncache.cache_limit -= 1;
264 syncache_insert(sc, sch)
266 struct syncache_head *sch;
268 struct syncache *sc2;
271 INP_INFO_WLOCK_ASSERT(&tcbinfo);
274 * Make sure that we don't overflow the per-bucket
275 * limit or the total cache size limit.
277 if (sch->sch_length >= tcp_syncache.bucket_limit) {
279 * The bucket is full, toss the oldest element.
281 sc2 = TAILQ_FIRST(&sch->sch_bucket);
282 sc2->sc_tp->ts_recent = ticks;
283 syncache_drop(sc2, sch);
284 tcpstat.tcps_sc_bucketoverflow++;
285 } else if (tcp_syncache.cache_count >= tcp_syncache.cache_limit) {
287 * The cache is full. Toss the oldest entry in the
288 * entire cache. This is the front entry in the
289 * first non-empty timer queue with the largest
292 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
293 sc2 = TAILQ_FIRST(&tcp_syncache.timerq[i]);
297 sc2->sc_tp->ts_recent = ticks;
298 syncache_drop(sc2, NULL);
299 tcpstat.tcps_sc_cacheoverflow++;
302 /* Initialize the entry's timer. */
303 SYNCACHE_TIMEOUT(sc, 0);
305 /* Put it into the bucket. */
306 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash);
308 tcp_syncache.cache_count++;
309 tcpstat.tcps_sc_added++;
313 syncache_drop(sc, sch)
315 struct syncache_head *sch;
317 INP_INFO_WLOCK_ASSERT(&tcbinfo);
321 if (sc->sc_inc.inc_isipv6) {
322 sch = &tcp_syncache.hashbase[
323 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)];
327 sch = &tcp_syncache.hashbase[
328 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)];
332 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
334 tcp_syncache.cache_count--;
336 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], sc, sc_timerq);
337 if (TAILQ_EMPTY(&tcp_syncache.timerq[sc->sc_rxtslot]))
338 callout_stop(&tcp_syncache.tt_timerq[sc->sc_rxtslot]);
344 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
345 * If we have retransmitted an entry the maximum number of times, expire it.
348 syncache_timer(xslot)
351 intptr_t slot = (intptr_t)xslot;
352 struct syncache *sc, *nsc;
355 INP_INFO_WLOCK(&tcbinfo);
356 if (callout_pending(&tcp_syncache.tt_timerq[slot]) ||
357 !callout_active(&tcp_syncache.tt_timerq[slot])) {
358 /* XXX can this happen? */
359 INP_INFO_WUNLOCK(&tcbinfo);
362 callout_deactivate(&tcp_syncache.tt_timerq[slot]);
364 nsc = TAILQ_FIRST(&tcp_syncache.timerq[slot]);
365 while (nsc != NULL) {
366 if (ticks < nsc->sc_rxttime)
369 inp = sc->sc_tp->t_inpcb;
370 if (slot == SYNCACHE_MAXREXMTS ||
371 slot >= tcp_syncache.rexmt_limit ||
372 inp == NULL || inp->inp_gencnt != sc->sc_inp_gencnt) {
373 nsc = TAILQ_NEXT(sc, sc_timerq);
374 syncache_drop(sc, NULL);
375 tcpstat.tcps_sc_stale++;
379 * syncache_respond() may call back into the syncache to
380 * to modify another entry, so do not obtain the next
381 * entry on the timer chain until it has completed.
384 (void) syncache_respond(sc, NULL, NULL);
386 (void) syncache_respond(sc, NULL);
388 nsc = TAILQ_NEXT(sc, sc_timerq);
389 tcpstat.tcps_sc_retransmitted++;
390 TAILQ_REMOVE(&tcp_syncache.timerq[slot], sc, sc_timerq);
391 SYNCACHE_TIMEOUT(sc, slot + 1);
394 callout_reset(&tcp_syncache.tt_timerq[slot],
395 nsc->sc_rxttime - ticks, syncache_timer, (void *)(slot));
396 INP_INFO_WUNLOCK(&tcbinfo);
400 * Find an entry in the syncache.
403 syncache_lookup(inc, schp)
404 struct in_conninfo *inc;
405 struct syncache_head **schp;
408 struct syncache_head *sch;
410 INP_INFO_WLOCK_ASSERT(&tcbinfo);
413 if (inc->inc_isipv6) {
414 sch = &tcp_syncache.hashbase[
415 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
417 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
418 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
424 sch = &tcp_syncache.hashbase[
425 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
427 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
429 if (sc->sc_inc.inc_isipv6)
432 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
440 * This function is called when we get a RST for a
441 * non-existent connection, so that we can see if the
442 * connection is in the syn cache. If it is, zap it.
445 syncache_chkrst(inc, th)
446 struct in_conninfo *inc;
450 struct syncache_head *sch;
452 INP_INFO_WLOCK_ASSERT(&tcbinfo);
454 sc = syncache_lookup(inc, &sch);
458 * If the RST bit is set, check the sequence number to see
459 * if this is a valid reset segment.
461 * In all states except SYN-SENT, all reset (RST) segments
462 * are validated by checking their SEQ-fields. A reset is
463 * valid if its sequence number is in the window.
465 * The sequence number in the reset segment is normally an
466 * echo of our outgoing acknowlegement numbers, but some hosts
467 * send a reset with the sequence number at the rightmost edge
468 * of our receive window, and we have to handle this case.
470 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
471 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
472 syncache_drop(sc, sch);
473 tcpstat.tcps_sc_reset++;
479 struct in_conninfo *inc;
482 struct syncache_head *sch;
484 INP_INFO_WLOCK_ASSERT(&tcbinfo);
486 sc = syncache_lookup(inc, &sch);
488 syncache_drop(sc, sch);
489 tcpstat.tcps_sc_badack++;
494 syncache_unreach(inc, th)
495 struct in_conninfo *inc;
499 struct syncache_head *sch;
501 INP_INFO_WLOCK_ASSERT(&tcbinfo);
503 sc = syncache_lookup(inc, &sch);
507 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
508 if (ntohl(th->th_seq) != sc->sc_iss)
512 * If we've rertransmitted 3 times and this is our second error,
513 * we remove the entry. Otherwise, we allow it to continue on.
514 * This prevents us from incorrectly nuking an entry during a
515 * spurious network outage.
519 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) {
520 sc->sc_flags |= SCF_UNREACH;
523 syncache_drop(sc, sch);
524 tcpstat.tcps_sc_unreach++;
528 * Build a new TCP socket structure from a syncache entry.
530 static struct socket *
531 syncache_socket(sc, lso, m)
536 struct inpcb *inp = NULL;
541 INP_INFO_WLOCK_ASSERT(&tcbinfo);
544 * Ok, create the full blown connection, and set things up
545 * as they would have been set up if we had created the
546 * connection when the SYN arrived. If we can't create
547 * the connection, abort it.
549 so = sonewconn(lso, SS_ISCONNECTED);
552 * Drop the connection; we will send a RST if the peer
553 * retransmits the ACK,
555 tcpstat.tcps_listendrop++;
560 mac_set_socket_peer_from_mbuf(m, so);
568 * Insert new socket into hash list.
570 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
572 if (sc->sc_inc.inc_isipv6) {
573 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
575 inp->inp_vflag &= ~INP_IPV6;
576 inp->inp_vflag |= INP_IPV4;
578 inp->inp_laddr = sc->sc_inc.inc_laddr;
582 inp->inp_lport = sc->sc_inc.inc_lport;
583 if (in_pcbinshash(inp) != 0) {
585 * Undo the assignments above if we failed to
586 * put the PCB on the hash lists.
589 if (sc->sc_inc.inc_isipv6)
590 inp->in6p_laddr = in6addr_any;
593 inp->inp_laddr.s_addr = INADDR_ANY;
598 /* copy old policy into new socket's */
599 if (ipsec_copy_pcbpolicy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
600 printf("syncache_expand: could not copy policy\n");
603 /* copy old policy into new socket's */
604 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
605 printf("syncache_expand: could not copy policy\n");
608 if (sc->sc_inc.inc_isipv6) {
609 struct inpcb *oinp = sotoinpcb(lso);
610 struct in6_addr laddr6;
611 struct sockaddr_in6 sin6;
613 * Inherit socket options from the listening socket.
614 * Note that in6p_inputopts are not (and should not be)
615 * copied, since it stores previously received options and is
616 * used to detect if each new option is different than the
617 * previous one and hence should be passed to a user.
618 * If we copied in6p_inputopts, a user would not be able to
619 * receive options just after calling the accept system call.
621 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
622 if (oinp->in6p_outputopts)
623 inp->in6p_outputopts =
624 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
626 sin6.sin6_family = AF_INET6;
627 sin6.sin6_len = sizeof(sin6);
628 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
629 sin6.sin6_port = sc->sc_inc.inc_fport;
630 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
631 laddr6 = inp->in6p_laddr;
632 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
633 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
634 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
636 inp->in6p_laddr = laddr6;
639 /* Override flowlabel from in6_pcbconnect. */
640 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK;
641 inp->in6p_flowinfo |= sc->sc_flowlabel;
645 struct in_addr laddr;
646 struct sockaddr_in sin;
648 inp->inp_options = ip_srcroute(m);
649 if (inp->inp_options == NULL) {
650 inp->inp_options = sc->sc_ipopts;
651 sc->sc_ipopts = NULL;
654 sin.sin_family = AF_INET;
655 sin.sin_len = sizeof(sin);
656 sin.sin_addr = sc->sc_inc.inc_faddr;
657 sin.sin_port = sc->sc_inc.inc_fport;
658 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
659 laddr = inp->inp_laddr;
660 if (inp->inp_laddr.s_addr == INADDR_ANY)
661 inp->inp_laddr = sc->sc_inc.inc_laddr;
662 if (in_pcbconnect(inp, (struct sockaddr *)&sin,
664 inp->inp_laddr = laddr;
670 tp->t_state = TCPS_SYN_RECEIVED;
671 tp->iss = sc->sc_iss;
672 tp->irs = sc->sc_irs;
675 tp->snd_wl1 = sc->sc_irs;
676 tp->rcv_up = sc->sc_irs + 1;
677 tp->rcv_wnd = sc->sc_wnd;
678 tp->rcv_adv += tp->rcv_wnd;
680 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
681 if (sc->sc_flags & SCF_NOOPT)
682 tp->t_flags |= TF_NOOPT;
683 if (sc->sc_flags & SCF_WINSCALE) {
684 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
685 tp->snd_scale = sc->sc_requested_s_scale;
686 tp->request_r_scale = sc->sc_request_r_scale;
688 if (sc->sc_flags & SCF_TIMESTAMP) {
689 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
690 tp->ts_recent = sc->sc_tsrecent;
691 tp->ts_recent_age = ticks;
694 if (sc->sc_flags & SCF_SIGNATURE)
695 tp->t_flags |= TF_SIGNATURE;
697 if (sc->sc_flags & SCF_SACK) {
699 tp->t_flags |= TF_SACK_PERMIT;
702 * Set up MSS and get cached values from tcp_hostcache.
703 * This might overwrite some of the defaults we just set.
705 tcp_mss(tp, sc->sc_peer_mss);
708 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
710 if (sc->sc_rxtslot != 0)
711 tp->snd_cwnd = tp->t_maxseg;
712 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
716 tcpstat.tcps_accepts++;
728 * This function gets called when we receive an ACK for a
729 * socket in the LISTEN state. We look up the connection
730 * in the syncache, and if its there, we pull it out of
731 * the cache and turn it into a full-blown connection in
732 * the SYN-RECEIVED state.
735 syncache_expand(inc, th, sop, m)
736 struct in_conninfo *inc;
742 struct syncache_head *sch;
745 INP_INFO_WLOCK_ASSERT(&tcbinfo);
747 sc = syncache_lookup(inc, &sch);
750 * There is no syncache entry, so see if this ACK is
751 * a returning syncookie. To do this, first:
752 * A. See if this socket has had a syncache entry dropped in
753 * the past. We don't want to accept a bogus syncookie
754 * if we've never received a SYN.
755 * B. check that the syncookie is valid. If it is, then
756 * cobble up a fake syncache entry, and return.
760 sc = syncookie_lookup(inc, th, *sop);
764 tcpstat.tcps_sc_recvcookie++;
768 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
770 if (th->th_ack != sc->sc_iss + 1) {
776 so = syncache_socket(sc, *sop, m);
780 /* XXXjlemon check this - is this correct? */
781 (void) tcp_respond(NULL, m, m, th,
782 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK);
784 m_freem(m); /* XXX only needed for above */
785 tcpstat.tcps_sc_aborted++;
787 tcpstat.tcps_sc_completed++;
792 syncache_drop(sc, sch);
798 * Given a LISTEN socket and an inbound SYN request, add
799 * this to the syn cache, and send back a segment:
800 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
803 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
804 * Doing so would require that we hold onto the data and deliver it
805 * to the application. However, if we are the target of a SYN-flood
806 * DoS attack, an attacker could send data which would eventually
807 * consume all available buffer space if it were ACKed. By not ACKing
808 * the data, we avoid this DoS scenario.
811 syncache_add(inc, to, th, sop, m)
812 struct in_conninfo *inc;
820 struct syncache *sc = NULL;
821 struct syncache_head *sch;
822 struct mbuf *ipopts = NULL;
826 INP_INFO_WLOCK_ASSERT(&tcbinfo);
832 * Remember the IP options, if any.
835 if (!inc->inc_isipv6)
837 ipopts = ip_srcroute(m);
840 * See if we already have an entry for this connection.
841 * If we do, resend the SYN,ACK, and reset the retransmit timer.
844 * should the syncache be re-initialized with the contents
845 * of the new SYN here (which may have different options?)
847 sc = syncache_lookup(inc, &sch);
849 tcpstat.tcps_sc_dupsyn++;
852 * If we were remembering a previous source route,
853 * forget it and use the new one we've been given.
856 (void) m_free(sc->sc_ipopts);
857 sc->sc_ipopts = ipopts;
860 * Update timestamp if present.
862 if (sc->sc_flags & SCF_TIMESTAMP)
863 sc->sc_tsrecent = to->to_tsval;
865 * PCB may have changed, pick up new values.
868 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
870 if (syncache_respond(sc, m, so) == 0) {
872 if (syncache_respond(sc, m) == 0) {
874 /* NB: guarded by INP_INFO_WLOCK(&tcbinfo) */
875 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot],
877 SYNCACHE_TIMEOUT(sc, sc->sc_rxtslot);
878 tcpstat.tcps_sndacks++;
879 tcpstat.tcps_sndtotal++;
885 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
888 * The zone allocator couldn't provide more entries.
889 * Treat this as if the cache was full; drop the oldest
890 * entry and insert the new one.
892 tcpstat.tcps_sc_zonefail++;
893 /* NB: guarded by INP_INFO_WLOCK(&tcbinfo) */
894 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
895 sc = TAILQ_FIRST(&tcp_syncache.timerq[i]);
900 /* Generic memory failure. */
902 (void) m_free(ipopts);
905 sc->sc_tp->ts_recent = ticks;
906 syncache_drop(sc, NULL);
907 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
910 (void) m_free(ipopts);
916 * Fill in the syncache values.
919 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
920 sc->sc_ipopts = ipopts;
921 sc->sc_inc.inc_fport = inc->inc_fport;
922 sc->sc_inc.inc_lport = inc->inc_lport;
924 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
925 if (inc->inc_isipv6) {
926 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
927 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
931 sc->sc_inc.inc_faddr = inc->inc_faddr;
932 sc->sc_inc.inc_laddr = inc->inc_laddr;
934 sc->sc_irs = th->th_seq;
936 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0;
937 sc->sc_flowlabel = 0;
938 if (tcp_syncookies) {
939 sc->sc_iss = syncookie_generate(sc, &flowtmp);
941 if (inc->inc_isipv6 &&
942 (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL)) {
943 sc->sc_flowlabel = flowtmp & IPV6_FLOWLABEL_MASK;
947 sc->sc_iss = arc4random();
949 if (inc->inc_isipv6 &&
950 (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL)) {
952 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
957 /* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */
958 win = sbspace(&so->so_rcv);
960 win = imin(win, TCP_MAXWIN);
963 if (tcp_do_rfc1323) {
965 * A timestamp received in a SYN makes
966 * it ok to send timestamp requests and replies.
968 if (to->to_flags & TOF_TS) {
969 sc->sc_tsrecent = to->to_tsval;
970 sc->sc_flags |= SCF_TIMESTAMP;
972 if (to->to_flags & TOF_SCALE) {
975 /* Compute proper scaling value from buffer space */
976 while (wscale < TCP_MAX_WINSHIFT &&
977 (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat)
979 sc->sc_request_r_scale = wscale;
980 sc->sc_requested_s_scale = to->to_requested_s_scale;
981 sc->sc_flags |= SCF_WINSCALE;
984 if (tp->t_flags & TF_NOOPT)
985 sc->sc_flags = SCF_NOOPT;
988 * If listening socket requested TCP digests, and received SYN
989 * contains the option, flag this in the syncache so that
990 * syncache_respond() will do the right thing with the SYN+ACK.
991 * XXX Currently we always record the option by default and will
992 * attempt to use it in syncache_respond().
994 if (to->to_flags & TOF_SIGNATURE)
995 sc->sc_flags |= SCF_SIGNATURE;
998 if (to->to_flags & TOF_SACK)
999 sc->sc_flags |= SCF_SACK;
1002 * Do a standard 3-way handshake.
1005 if (syncache_respond(sc, m, so) == 0) {
1007 if (syncache_respond(sc, m) == 0) {
1009 syncache_insert(sc, sch);
1010 tcpstat.tcps_sndacks++;
1011 tcpstat.tcps_sndtotal++;
1014 tcpstat.tcps_sc_dropped++;
1022 syncache_respond(sc, m, so)
1023 struct syncache *sc;
1028 syncache_respond(sc, m)
1029 struct syncache *sc;
1035 u_int16_t tlen, hlen, mssopt;
1036 struct ip *ip = NULL;
1040 struct ip6_hdr *ip6 = NULL;
1045 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) :
1049 KASSERT((&sc->sc_inc) != NULL, ("syncache_respond with NULL in_conninfo pointer"));
1051 /* Determine MSS we advertize to other end of connection */
1052 mssopt = tcp_mssopt(&sc->sc_inc);
1054 /* Compute the size of the TCP options. */
1055 if (sc->sc_flags & SCF_NOOPT) {
1058 optlen = TCPOLEN_MAXSEG +
1059 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) +
1060 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
1061 #ifdef TCP_SIGNATURE
1062 if (sc->sc_flags & SCF_SIGNATURE)
1063 optlen += TCPOLEN_SIGNATURE;
1065 if (sc->sc_flags & SCF_SACK)
1066 optlen += TCPOLEN_SACK_PERMITTED;
1067 optlen = roundup2(optlen, 4);
1069 tlen = hlen + sizeof(struct tcphdr) + optlen;
1073 * assume that the entire packet will fit in a header mbuf
1075 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small"));
1078 * XXX shouldn't this reuse the mbuf if possible ?
1079 * Create the IP+TCP header from scratch.
1084 m = m_gethdr(M_DONTWAIT, MT_DATA);
1087 m->m_data += max_linkhdr;
1089 m->m_pkthdr.len = tlen;
1090 m->m_pkthdr.rcvif = NULL;
1091 inp = sc->sc_tp->t_inpcb;
1094 mac_create_mbuf_from_inpcb(inp, m);
1098 if (sc->sc_inc.inc_isipv6) {
1099 ip6 = mtod(m, struct ip6_hdr *);
1100 ip6->ip6_vfc = IPV6_VERSION;
1101 ip6->ip6_nxt = IPPROTO_TCP;
1102 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1103 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1104 ip6->ip6_plen = htons(tlen - hlen);
1105 /* ip6_hlim is set after checksum */
1106 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1107 ip6->ip6_flow |= sc->sc_flowlabel;
1109 th = (struct tcphdr *)(ip6 + 1);
1113 ip = mtod(m, struct ip *);
1114 ip->ip_v = IPVERSION;
1115 ip->ip_hl = sizeof(struct ip) >> 2;
1120 ip->ip_p = IPPROTO_TCP;
1121 ip->ip_src = sc->sc_inc.inc_laddr;
1122 ip->ip_dst = sc->sc_inc.inc_faddr;
1123 ip->ip_ttl = inp->inp_ip_ttl; /* XXX */
1124 ip->ip_tos = inp->inp_ip_tos; /* XXX */
1127 * See if we should do MTU discovery. Route lookups are
1128 * expensive, so we will only unset the DF bit if:
1130 * 1) path_mtu_discovery is disabled
1131 * 2) the SCF_UNREACH flag has been set
1133 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1134 ip->ip_off |= IP_DF;
1136 th = (struct tcphdr *)(ip + 1);
1138 th->th_sport = sc->sc_inc.inc_lport;
1139 th->th_dport = sc->sc_inc.inc_fport;
1141 th->th_seq = htonl(sc->sc_iss);
1142 th->th_ack = htonl(sc->sc_irs + 1);
1143 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1145 th->th_flags = TH_SYN|TH_ACK;
1146 th->th_win = htons(sc->sc_wnd);
1149 /* Tack on the TCP options. */
1151 optp = (u_int8_t *)(th + 1);
1152 *optp++ = TCPOPT_MAXSEG;
1153 *optp++ = TCPOLEN_MAXSEG;
1154 *optp++ = (mssopt >> 8) & 0xff;
1155 *optp++ = mssopt & 0xff;
1157 if (sc->sc_flags & SCF_WINSCALE) {
1158 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
1159 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
1160 sc->sc_request_r_scale);
1164 if (sc->sc_flags & SCF_TIMESTAMP) {
1165 u_int32_t *lp = (u_int32_t *)(optp);
1167 /* Form timestamp option per appendix A of RFC 1323. */
1168 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1169 *lp++ = htonl(ticks);
1170 *lp = htonl(sc->sc_tsrecent);
1171 optp += TCPOLEN_TSTAMP_APPA;
1174 #ifdef TCP_SIGNATURE
1176 * Handle TCP-MD5 passive opener response.
1178 if (sc->sc_flags & SCF_SIGNATURE) {
1179 u_int8_t *bp = optp;
1182 *bp++ = TCPOPT_SIGNATURE;
1183 *bp++ = TCPOLEN_SIGNATURE;
1184 for (i = 0; i < TCP_SIGLEN; i++)
1186 tcp_signature_compute(m, sizeof(struct ip), 0, optlen,
1187 optp + 2, IPSEC_DIR_OUTBOUND);
1188 optp += TCPOLEN_SIGNATURE;
1190 #endif /* TCP_SIGNATURE */
1192 if (sc->sc_flags & SCF_SACK) {
1193 *optp++ = TCPOPT_SACK_PERMITTED;
1194 *optp++ = TCPOLEN_SACK_PERMITTED;
1198 /* Pad TCP options to a 4 byte boundary */
1199 int padlen = optlen - (optp - (u_int8_t *)(th + 1));
1200 while (padlen-- > 0)
1201 *optp++ = TCPOPT_EOL;
1206 if (sc->sc_inc.inc_isipv6) {
1208 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
1209 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1210 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1214 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1215 htons(tlen - hlen + IPPROTO_TCP));
1216 m->m_pkthdr.csum_flags = CSUM_TCP;
1217 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1222 if (so != NULL && so->so_options & SO_DEBUG) {
1223 struct tcpcb *tp = sototcpcb(so);
1224 tcp_trace(TA_OUTPUT, tp->t_state, tp,
1225 mtod(m, void *), th, 0);
1228 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, inp);
1237 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .|
1239 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .|
1241 * (A): peer mss index
1245 * The values below are chosen to minimize the size of the tcp_secret
1246 * table, as well as providing roughly a 16 second lifetime for the cookie.
1249 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */
1250 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */
1252 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1)
1253 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS)
1254 #define SYNCOOKIE_TIMEOUT \
1255 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT))
1256 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK)
1259 u_int32_t ts_secbits[4];
1261 } tcp_secret[SYNCOOKIE_NSECRETS];
1263 static int tcp_msstab[] = { 0, 536, 1460, 8960 };
1265 static MD5_CTX syn_ctx;
1267 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v))
1270 u_int32_t laddr, faddr;
1271 u_int32_t secbits[4];
1272 u_int16_t lport, fport;
1276 CTASSERT(sizeof(struct md5_add) == 28);
1280 * Consider the problem of a recreated (and retransmitted) cookie. If the
1281 * original SYN was accepted, the connection is established. The second
1282 * SYN is inflight, and if it arrives with an ISN that falls within the
1283 * receive window, the connection is killed.
1285 * However, since cookies have other problems, this may not be worth
1290 syncookie_generate(struct syncache *sc, u_int32_t *flowid)
1292 u_int32_t md5_buffer[4];
1297 /* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */
1299 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK;
1300 if (tcp_secret[idx].ts_expire < ticks) {
1301 for (i = 0; i < 4; i++)
1302 tcp_secret[idx].ts_secbits[i] = arc4random();
1303 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT;
1305 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--)
1306 if (tcp_msstab[data] <= sc->sc_peer_mss)
1308 data = (data << SYNCOOKIE_WNDBITS) | idx;
1309 data ^= sc->sc_irs; /* peer's iss */
1312 if (sc->sc_inc.inc_isipv6) {
1313 MD5Add(sc->sc_inc.inc6_laddr);
1314 MD5Add(sc->sc_inc.inc6_faddr);
1320 add.laddr = sc->sc_inc.inc_laddr.s_addr;
1321 add.faddr = sc->sc_inc.inc_faddr.s_addr;
1323 add.lport = sc->sc_inc.inc_lport;
1324 add.fport = sc->sc_inc.inc_fport;
1325 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1326 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1327 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1328 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1330 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1331 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK);
1332 *flowid = md5_buffer[1];
1336 static struct syncache *
1337 syncookie_lookup(inc, th, so)
1338 struct in_conninfo *inc;
1342 u_int32_t md5_buffer[4];
1343 struct syncache *sc;
1348 /* NB: single threaded; could add INP_INFO_WLOCK_ASSERT(&tcbinfo) */
1350 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */
1351 idx = data & SYNCOOKIE_WNDMASK;
1352 if (tcp_secret[idx].ts_expire < ticks ||
1353 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks)
1357 if (inc->inc_isipv6) {
1358 MD5Add(inc->inc6_laddr);
1359 MD5Add(inc->inc6_faddr);
1365 add.laddr = inc->inc_laddr.s_addr;
1366 add.faddr = inc->inc_faddr.s_addr;
1368 add.lport = inc->inc_lport;
1369 add.fport = inc->inc_fport;
1370 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1371 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1372 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1373 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1375 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1376 data ^= md5_buffer[0];
1377 if ((data & ~SYNCOOKIE_DATAMASK) != 0)
1379 data = data >> SYNCOOKIE_WNDBITS;
1381 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
1385 * Fill in the syncache values.
1386 * XXX duplicate code from syncache_add
1388 sc->sc_ipopts = NULL;
1389 sc->sc_inc.inc_fport = inc->inc_fport;
1390 sc->sc_inc.inc_lport = inc->inc_lport;
1391 sc->sc_tp = sototcpcb(so);
1393 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1394 if (inc->inc_isipv6) {
1395 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1396 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1397 if (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL)
1398 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1402 sc->sc_inc.inc_faddr = inc->inc_faddr;
1403 sc->sc_inc.inc_laddr = inc->inc_laddr;
1405 sc->sc_irs = th->th_seq - 1;
1406 sc->sc_iss = th->th_ack - 1;
1407 wnd = sbspace(&so->so_rcv);
1409 wnd = imin(wnd, TCP_MAXWIN);
1413 sc->sc_peer_mss = tcp_msstab[data];