2 * Copyright (c) 2001 McAfee, Inc.
3 * Copyright (c) 2006 Andre Oppermann, Internet Business Solutions AG
6 * This software was developed for the FreeBSD Project by Jonathan Lemon
7 * and McAfee Research, the Security Research Division of McAfee, Inc. under
8 * DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
9 * DARPA CHATS research program.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
46 #include <sys/mutex.h>
47 #include <sys/malloc.h>
50 #include <sys/proc.h> /* for proc0 declaration */
51 #include <sys/random.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
54 #include <sys/syslog.h>
59 #include <net/route.h>
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/ip.h>
64 #include <netinet/in_var.h>
65 #include <netinet/in_pcb.h>
66 #include <netinet/ip_var.h>
67 #include <netinet/ip_options.h>
69 #include <netinet/ip6.h>
70 #include <netinet/icmp6.h>
71 #include <netinet6/nd6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/in6_pcb.h>
75 #include <netinet/tcp.h>
76 #include <netinet/tcp_fsm.h>
77 #include <netinet/tcp_seq.h>
78 #include <netinet/tcp_timer.h>
79 #include <netinet/tcp_var.h>
80 #include <netinet/tcp_syncache.h>
82 #include <netinet6/tcp6_var.h>
86 #include <netipsec/ipsec.h>
88 #include <netipsec/ipsec6.h>
90 #include <netipsec/key.h>
93 #include <machine/in_cksum.h>
95 #include <security/mac/mac_framework.h>
97 static int tcp_syncookies = 1;
98 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
100 "Use TCP SYN cookies if the syncache overflows");
102 static int tcp_syncookiesonly = 0;
103 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW,
104 &tcp_syncookiesonly, 0,
105 "Use only TCP SYN cookies");
107 #define SYNCOOKIE_SECRET_SIZE 8 /* dwords */
108 #define SYNCOOKIE_LIFETIME 16 /* seconds */
111 TAILQ_ENTRY(syncache) sc_hash;
112 struct in_conninfo sc_inc; /* addresses */
113 u_long sc_rxttime; /* retransmit time */
114 u_int16_t sc_rxmits; /* retransmit counter */
116 u_int32_t sc_tsreflect; /* timestamp to reflect */
117 u_int32_t sc_ts; /* our timestamp to send */
118 u_int32_t sc_tsoff; /* ts offset w/ syncookies */
119 u_int32_t sc_flowlabel; /* IPv6 flowlabel */
120 tcp_seq sc_irs; /* seq from peer */
121 tcp_seq sc_iss; /* our ISS */
122 struct mbuf *sc_ipopts; /* source route */
124 u_int16_t sc_peer_mss; /* peer's MSS */
125 u_int16_t sc_wnd; /* advertised window */
126 u_int8_t sc_ip_ttl; /* IPv4 TTL */
127 u_int8_t sc_ip_tos; /* IPv4 TOS */
128 u_int8_t sc_requested_s_scale:4,
129 sc_requested_r_scale:4;
131 #define SCF_NOOPT 0x01 /* no TCP options */
132 #define SCF_WINSCALE 0x02 /* negotiated window scaling */
133 #define SCF_TIMESTAMP 0x04 /* negotiated timestamps */
134 /* MSS is implicit */
135 #define SCF_UNREACH 0x10 /* icmp unreachable received */
136 #define SCF_SIGNATURE 0x20 /* send MD5 digests */
137 #define SCF_SACK 0x80 /* send SACK option */
139 struct label *sc_label; /* MAC label reference */
143 struct syncache_head {
145 TAILQ_HEAD(sch_head, syncache) sch_bucket;
146 struct callout sch_timer;
150 u_int32_t sch_secbits_odd[SYNCOOKIE_SECRET_SIZE];
151 u_int32_t sch_secbits_even[SYNCOOKIE_SECRET_SIZE];
152 u_int sch_reseed; /* time_uptime, seconds */
155 static void syncache_drop(struct syncache *, struct syncache_head *);
156 static void syncache_free(struct syncache *);
157 static void syncache_insert(struct syncache *, struct syncache_head *);
158 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
159 static int syncache_respond(struct syncache *);
160 static struct socket *syncache_socket(struct syncache *, struct socket *,
162 static void syncache_timeout(struct syncache *sc, struct syncache_head *sch,
164 static void syncache_timer(void *);
165 static void syncookie_generate(struct syncache_head *, struct syncache *,
167 static struct syncache
168 *syncookie_lookup(struct in_conninfo *, struct syncache_head *,
169 struct syncache *, struct tcpopt *, struct tcphdr *,
173 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
174 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds,
175 * the odds are that the user has given up attempting to connect by then.
177 #define SYNCACHE_MAXREXMTS 3
179 /* Arbitrary values */
180 #define TCP_SYNCACHE_HASHSIZE 512
181 #define TCP_SYNCACHE_BUCKETLIMIT 30
183 struct tcp_syncache {
184 struct syncache_head *hashbase;
189 u_int cache_count; /* XXX: unprotected */
194 static struct tcp_syncache tcp_syncache;
196 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
198 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
199 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
201 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
202 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
204 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
205 &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
207 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
208 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
210 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
211 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
213 int tcp_sc_rst_sock_fail = 1;
214 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail, CTLFLAG_RW,
215 &tcp_sc_rst_sock_fail, 0, "Send reset on socket allocation failure");
217 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
219 #define SYNCACHE_HASH(inc, mask) \
220 ((tcp_syncache.hash_secret ^ \
221 (inc)->inc_faddr.s_addr ^ \
222 ((inc)->inc_faddr.s_addr >> 16) ^ \
223 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
225 #define SYNCACHE_HASH6(inc, mask) \
226 ((tcp_syncache.hash_secret ^ \
227 (inc)->inc6_faddr.s6_addr32[0] ^ \
228 (inc)->inc6_faddr.s6_addr32[3] ^ \
229 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
231 #define ENDPTS_EQ(a, b) ( \
232 (a)->ie_fport == (b)->ie_fport && \
233 (a)->ie_lport == (b)->ie_lport && \
234 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
235 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
238 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
240 #define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx)
241 #define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx)
242 #define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED)
245 * Requires the syncache entry to be already removed from the bucket list.
248 syncache_free(struct syncache *sc)
251 (void) m_free(sc->sc_ipopts);
253 mac_syncache_destroy(&sc->sc_label);
256 uma_zfree(tcp_syncache.zone, sc);
264 tcp_syncache.cache_count = 0;
265 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
266 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
267 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
268 tcp_syncache.hash_secret = arc4random();
270 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
271 &tcp_syncache.hashsize);
272 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
273 &tcp_syncache.bucket_limit);
274 if (!powerof2(tcp_syncache.hashsize) || tcp_syncache.hashsize == 0) {
275 printf("WARNING: syncache hash size is not a power of 2.\n");
276 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
278 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
281 tcp_syncache.cache_limit =
282 tcp_syncache.hashsize * tcp_syncache.bucket_limit;
283 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
284 &tcp_syncache.cache_limit);
286 /* Allocate the hash table. */
287 MALLOC(tcp_syncache.hashbase, struct syncache_head *,
288 tcp_syncache.hashsize * sizeof(struct syncache_head),
289 M_SYNCACHE, M_WAITOK | M_ZERO);
291 /* Initialize the hash buckets. */
292 for (i = 0; i < tcp_syncache.hashsize; i++) {
293 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket);
294 mtx_init(&tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
296 callout_init_mtx(&tcp_syncache.hashbase[i].sch_timer,
297 &tcp_syncache.hashbase[i].sch_mtx, 0);
298 tcp_syncache.hashbase[i].sch_length = 0;
301 /* Create the syncache entry zone. */
302 tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
303 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
304 uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit);
308 * Inserts a syncache entry into the specified bucket row.
309 * Locks and unlocks the syncache_head autonomously.
312 syncache_insert(struct syncache *sc, struct syncache_head *sch)
314 struct syncache *sc2;
319 * Make sure that we don't overflow the per-bucket limit.
320 * If the bucket is full, toss the oldest element.
322 if (sch->sch_length >= tcp_syncache.bucket_limit) {
323 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
324 ("sch->sch_length incorrect"));
325 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
326 syncache_drop(sc2, sch);
327 tcpstat.tcps_sc_bucketoverflow++;
330 /* Put it into the bucket. */
331 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
334 /* Reinitialize the bucket row's timer. */
335 syncache_timeout(sc, sch, 1);
339 tcp_syncache.cache_count++;
340 tcpstat.tcps_sc_added++;
344 * Remove and free entry from syncache bucket row.
345 * Expects locked syncache head.
348 syncache_drop(struct syncache *sc, struct syncache_head *sch)
351 SCH_LOCK_ASSERT(sch);
353 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
357 tcp_syncache.cache_count--;
361 * Engage/reengage time on bucket row.
364 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
366 sc->sc_rxttime = ticks +
367 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
369 if (sch->sch_nextc > sc->sc_rxttime)
370 sch->sch_nextc = sc->sc_rxttime;
371 if (!TAILQ_EMPTY(&sch->sch_bucket) && docallout)
372 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
373 syncache_timer, (void *)sch);
377 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
378 * If we have retransmitted an entry the maximum number of times, expire it.
379 * One separate timer for each bucket row.
382 syncache_timer(void *xsch)
384 struct syncache_head *sch = (struct syncache_head *)xsch;
385 struct syncache *sc, *nsc;
389 /* NB: syncache_head has already been locked by the callout. */
390 SCH_LOCK_ASSERT(sch);
392 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
394 * We do not check if the listen socket still exists
395 * and accept the case where the listen socket may be
396 * gone by the time we resend the SYN/ACK. We do
397 * not expect this to happens often. If it does,
398 * then the RST will be sent by the time the remote
399 * host does the SYN/ACK->ACK.
401 if (sc->sc_rxttime > tick) {
402 if (sc->sc_rxttime < sch->sch_nextc)
403 sch->sch_nextc = sc->sc_rxttime;
407 if (sc->sc_rxmits > tcp_syncache.rexmt_limit) {
408 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
409 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
410 "giving up and removing syncache entry\n",
414 syncache_drop(sc, sch);
415 tcpstat.tcps_sc_stale++;
418 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
419 log(LOG_DEBUG, "%s; %s: Response timeout, "
420 "retransmitting (%u) SYN|ACK\n",
421 s, __func__, sc->sc_rxmits);
425 (void) syncache_respond(sc);
426 tcpstat.tcps_sc_retransmitted++;
427 syncache_timeout(sc, sch, 0);
429 if (!TAILQ_EMPTY(&(sch)->sch_bucket))
430 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
431 syncache_timer, (void *)(sch));
435 * Find an entry in the syncache.
436 * Returns always with locked syncache_head plus a matching entry or NULL.
439 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
442 struct syncache_head *sch;
445 if (inc->inc_isipv6) {
446 sch = &tcp_syncache.hashbase[
447 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
452 /* Circle through bucket row to find matching entry. */
453 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
454 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
460 sch = &tcp_syncache.hashbase[
461 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
466 /* Circle through bucket row to find matching entry. */
467 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
469 if (sc->sc_inc.inc_isipv6)
472 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
476 SCH_LOCK_ASSERT(*schp);
477 return (NULL); /* always returns with locked sch */
481 * This function is called when we get a RST for a
482 * non-existent connection, so that we can see if the
483 * connection is in the syn cache. If it is, zap it.
486 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
489 struct syncache_head *sch;
492 sc = syncache_lookup(inc, &sch); /* returns locked sch */
493 SCH_LOCK_ASSERT(sch);
496 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
497 * See RFC 793 page 65, section SEGMENT ARRIVES.
499 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
500 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
501 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
502 "FIN flag set, segment ignored\n", s, __func__);
503 tcpstat.tcps_badrst++;
508 * No corresponding connection was found in syncache.
509 * If syncookies are enabled and possibly exclusively
510 * used, or we are under memory pressure, a valid RST
511 * may not find a syncache entry. In that case we're
512 * done and no SYN|ACK retransmissions will happen.
513 * Otherwise the the RST was misdirected or spoofed.
516 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
517 log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
518 "syncache entry (possibly syncookie only), "
519 "segment ignored\n", s, __func__);
520 tcpstat.tcps_badrst++;
525 * If the RST bit is set, check the sequence number to see
526 * if this is a valid reset segment.
528 * In all states except SYN-SENT, all reset (RST) segments
529 * are validated by checking their SEQ-fields. A reset is
530 * valid if its sequence number is in the window.
532 * The sequence number in the reset segment is normally an
533 * echo of our outgoing acknowlegement numbers, but some hosts
534 * send a reset with the sequence number at the rightmost edge
535 * of our receive window, and we have to handle this case.
537 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
538 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
539 syncache_drop(sc, sch);
540 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
541 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
542 "connection attempt aborted by remote endpoint\n",
544 tcpstat.tcps_sc_reset++;
545 } else if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
546 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != IRS %u "
547 "(+WND %u), segment ignored\n",
548 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
549 tcpstat.tcps_badrst++;
559 syncache_badack(struct in_conninfo *inc)
562 struct syncache_head *sch;
564 sc = syncache_lookup(inc, &sch); /* returns locked sch */
565 SCH_LOCK_ASSERT(sch);
567 syncache_drop(sc, sch);
568 tcpstat.tcps_sc_badack++;
574 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
577 struct syncache_head *sch;
579 sc = syncache_lookup(inc, &sch); /* returns locked sch */
580 SCH_LOCK_ASSERT(sch);
584 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
585 if (ntohl(th->th_seq) != sc->sc_iss)
589 * If we've rertransmitted 3 times and this is our second error,
590 * we remove the entry. Otherwise, we allow it to continue on.
591 * This prevents us from incorrectly nuking an entry during a
592 * spurious network outage.
596 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
597 sc->sc_flags |= SCF_UNREACH;
600 syncache_drop(sc, sch);
601 tcpstat.tcps_sc_unreach++;
607 * Build a new TCP socket structure from a syncache entry.
609 static struct socket *
610 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
612 struct inpcb *inp = NULL;
617 INP_INFO_WLOCK_ASSERT(&tcbinfo);
620 * Ok, create the full blown connection, and set things up
621 * as they would have been set up if we had created the
622 * connection when the SYN arrived. If we can't create
623 * the connection, abort it.
625 so = sonewconn(lso, SS_ISCONNECTED);
628 * Drop the connection; we will either send a RST or
629 * have the peer retransmit its SYN again after its
632 tcpstat.tcps_listendrop++;
633 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
634 log(LOG_DEBUG, "%s; %s: Socket create failed "
635 "due to limits or memory shortage\n",
643 mac_socketpeer_set_from_mbuf(m, so);
650 /* Insert new socket into PCB hash list. */
651 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
653 if (sc->sc_inc.inc_isipv6) {
654 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
656 inp->inp_vflag &= ~INP_IPV6;
657 inp->inp_vflag |= INP_IPV4;
659 inp->inp_laddr = sc->sc_inc.inc_laddr;
663 inp->inp_lport = sc->sc_inc.inc_lport;
664 if (in_pcbinshash(inp) != 0) {
666 * Undo the assignments above if we failed to
667 * put the PCB on the hash lists.
670 if (sc->sc_inc.inc_isipv6)
671 inp->in6p_laddr = in6addr_any;
674 inp->inp_laddr.s_addr = INADDR_ANY;
679 /* Copy old policy into new socket's. */
680 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
681 printf("syncache_socket: could not copy policy\n");
684 if (sc->sc_inc.inc_isipv6) {
685 struct inpcb *oinp = sotoinpcb(lso);
686 struct in6_addr laddr6;
687 struct sockaddr_in6 sin6;
689 * Inherit socket options from the listening socket.
690 * Note that in6p_inputopts are not (and should not be)
691 * copied, since it stores previously received options and is
692 * used to detect if each new option is different than the
693 * previous one and hence should be passed to a user.
694 * If we copied in6p_inputopts, a user would not be able to
695 * receive options just after calling the accept system call.
697 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
698 if (oinp->in6p_outputopts)
699 inp->in6p_outputopts =
700 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
702 sin6.sin6_family = AF_INET6;
703 sin6.sin6_len = sizeof(sin6);
704 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
705 sin6.sin6_port = sc->sc_inc.inc_fport;
706 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
707 laddr6 = inp->in6p_laddr;
708 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
709 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
710 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
712 inp->in6p_laddr = laddr6;
715 /* Override flowlabel from in6_pcbconnect. */
716 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK;
717 inp->in6p_flowinfo |= sc->sc_flowlabel;
721 struct in_addr laddr;
722 struct sockaddr_in sin;
724 inp->inp_options = ip_srcroute(m);
725 if (inp->inp_options == NULL) {
726 inp->inp_options = sc->sc_ipopts;
727 sc->sc_ipopts = NULL;
730 sin.sin_family = AF_INET;
731 sin.sin_len = sizeof(sin);
732 sin.sin_addr = sc->sc_inc.inc_faddr;
733 sin.sin_port = sc->sc_inc.inc_fport;
734 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
735 laddr = inp->inp_laddr;
736 if (inp->inp_laddr.s_addr == INADDR_ANY)
737 inp->inp_laddr = sc->sc_inc.inc_laddr;
738 if (in_pcbconnect(inp, (struct sockaddr *)&sin,
740 inp->inp_laddr = laddr;
745 tp->t_state = TCPS_SYN_RECEIVED;
746 tp->iss = sc->sc_iss;
747 tp->irs = sc->sc_irs;
750 tp->snd_wl1 = sc->sc_irs;
751 tp->snd_max = tp->iss + 1;
752 tp->snd_nxt = tp->iss + 1;
753 tp->rcv_up = sc->sc_irs + 1;
754 tp->rcv_wnd = sc->sc_wnd;
755 tp->rcv_adv += tp->rcv_wnd;
756 tp->last_ack_sent = tp->rcv_nxt;
758 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
759 if (sc->sc_flags & SCF_NOOPT)
760 tp->t_flags |= TF_NOOPT;
762 if (sc->sc_flags & SCF_WINSCALE) {
763 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
764 tp->snd_scale = sc->sc_requested_s_scale;
765 tp->request_r_scale = sc->sc_requested_r_scale;
767 if (sc->sc_flags & SCF_TIMESTAMP) {
768 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
769 tp->ts_recent = sc->sc_tsreflect;
770 tp->ts_recent_age = ticks;
771 tp->ts_offset = sc->sc_tsoff;
774 if (sc->sc_flags & SCF_SIGNATURE)
775 tp->t_flags |= TF_SIGNATURE;
777 if (sc->sc_flags & SCF_SACK)
778 tp->t_flags |= TF_SACK_PERMIT;
782 * Set up MSS and get cached values from tcp_hostcache.
783 * This might overwrite some of the defaults we just set.
785 tcp_mss(tp, sc->sc_peer_mss);
788 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
791 tp->snd_cwnd = tp->t_maxseg;
792 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
796 tcpstat.tcps_accepts++;
808 * This function gets called when we receive an ACK for a
809 * socket in the LISTEN state. We look up the connection
810 * in the syncache, and if its there, we pull it out of
811 * the cache and turn it into a full-blown connection in
812 * the SYN-RECEIVED state.
815 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
816 struct socket **lsop, struct mbuf *m)
819 struct syncache_head *sch;
824 * Global TCP locks are held because we manipulate the PCB lists
825 * and create a new socket.
827 INP_INFO_WLOCK_ASSERT(&tcbinfo);
828 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
829 ("%s: can handle only ACK", __func__));
831 sc = syncache_lookup(inc, &sch); /* returns locked sch */
832 SCH_LOCK_ASSERT(sch);
835 * There is no syncache entry, so see if this ACK is
836 * a returning syncookie. To do this, first:
837 * A. See if this socket has had a syncache entry dropped in
838 * the past. We don't want to accept a bogus syncookie
839 * if we've never received a SYN.
840 * B. check that the syncookie is valid. If it is, then
841 * cobble up a fake syncache entry, and return.
843 if (!tcp_syncookies) {
845 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
846 log(LOG_DEBUG, "%s; %s: Spurious ACK, "
847 "segment rejected (syncookies disabled)\n",
851 bzero(&scs, sizeof(scs));
852 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
855 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
856 log(LOG_DEBUG, "%s; %s: Segment failed "
857 "SYNCOOKIE authentication, segment rejected "
858 "(probably spoofed)\n", s, __func__);
862 /* Pull out the entry to unlock the bucket row. */
863 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
865 tcp_syncache.cache_count--;
870 * Segment validation:
871 * ACK must match our initial sequence number + 1 (the SYN|ACK).
873 if (th->th_ack != sc->sc_iss + 1) {
874 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
875 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment "
876 "rejected\n", s, __func__, th->th_ack, sc->sc_iss);
880 * The SEQ must match the received initial receive sequence
881 * number + 1 (the SYN) because we didn't ACK any data that
882 * may have come with the SYN.
884 if (th->th_seq != sc->sc_irs + 1) {
885 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
886 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment "
887 "rejected\n", s, __func__, th->th_seq, sc->sc_irs);
892 * If timestamps were present in the SYN and we accepted
893 * them in our SYN|ACK we require them to be present from
894 * now on. And vice versa.
896 * Unfortunately, during testing of 7.0 some users found
897 * network devices that violate this constraint, so it must
900 if ((sc->sc_flags & SCF_TIMESTAMP) && !(to->to_flags & TOF_TS)) {
901 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
902 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
903 "segment rejected\n", s, __func__);
907 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
908 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
909 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
910 "segment rejected\n", s, __func__);
914 * If timestamps were negotiated the reflected timestamp
915 * must be equal to what we actually sent in the SYN|ACK.
917 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts) {
918 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
919 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
920 "segment rejected\n",
921 s, __func__, to->to_tsecr, sc->sc_ts);
925 *lsop = syncache_socket(sc, *lsop, m);
928 tcpstat.tcps_sc_aborted++;
930 tcpstat.tcps_sc_completed++;
936 if (sc != NULL && sc != &scs)
945 * Given a LISTEN socket and an inbound SYN request, add
946 * this to the syn cache, and send back a segment:
947 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
950 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
951 * Doing so would require that we hold onto the data and deliver it
952 * to the application. However, if we are the target of a SYN-flood
953 * DoS attack, an attacker could send data which would eventually
954 * consume all available buffer space if it were ACKed. By not ACKing
955 * the data, we avoid this DoS scenario.
958 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
959 struct inpcb *inp, struct socket **lsop, struct mbuf *m)
963 struct syncache *sc = NULL;
964 struct syncache_head *sch;
965 struct mbuf *ipopts = NULL;
967 int win, sb_hiwat, ip_ttl, ip_tos, noopt;
970 int autoflowlabel = 0;
973 struct label *maclabel;
977 INP_INFO_WLOCK_ASSERT(&tcbinfo);
978 INP_LOCK_ASSERT(inp); /* listen socket */
979 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
980 ("%s: unexpected tcp flags", __func__));
983 * Combine all so/tp operations very early to drop the INP lock as
990 if (inc->inc_isipv6 &&
991 (inp->in6p_flags & IN6P_AUTOFLOWLABEL))
994 ip_ttl = inp->inp_ip_ttl;
995 ip_tos = inp->inp_ip_tos;
996 win = sbspace(&so->so_rcv);
997 sb_hiwat = so->so_rcv.sb_hiwat;
998 noopt = (tp->t_flags & TF_NOOPT);
1004 if (mac_syncache_init(&maclabel) != 0) {
1006 INP_INFO_WUNLOCK(&tcbinfo);
1009 mac_syncache_create(maclabel, inp);
1012 INP_INFO_WUNLOCK(&tcbinfo);
1015 * Remember the IP options, if any.
1018 if (!inc->inc_isipv6)
1020 ipopts = ip_srcroute(m);
1023 * See if we already have an entry for this connection.
1024 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1026 * XXX: should the syncache be re-initialized with the contents
1027 * of the new SYN here (which may have different options?)
1029 * XXX: We do not check the sequence number to see if this is a
1030 * real retransmit or a new connection attempt. The question is
1031 * how to handle such a case; either ignore it as spoofed, or
1032 * drop the current entry and create a new one?
1034 sc = syncache_lookup(inc, &sch); /* returns locked entry */
1035 SCH_LOCK_ASSERT(sch);
1037 tcpstat.tcps_sc_dupsyn++;
1040 * If we were remembering a previous source route,
1041 * forget it and use the new one we've been given.
1044 (void) m_free(sc->sc_ipopts);
1045 sc->sc_ipopts = ipopts;
1048 * Update timestamp if present.
1050 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
1051 sc->sc_tsreflect = to->to_tsval;
1053 sc->sc_flags &= ~SCF_TIMESTAMP;
1056 * Since we have already unconditionally allocated label
1057 * storage, free it up. The syncache entry will already
1058 * have an initialized label we can use.
1060 mac_syncache_destroy(&maclabel);
1061 KASSERT(sc->sc_label != NULL,
1062 ("%s: label not initialized", __func__));
1064 /* Retransmit SYN|ACK and reset retransmit count. */
1065 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
1066 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
1067 "resetting timer and retransmitting SYN|ACK\n",
1071 if (syncache_respond(sc) == 0) {
1073 syncache_timeout(sc, sch, 1);
1074 tcpstat.tcps_sndacks++;
1075 tcpstat.tcps_sndtotal++;
1081 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
1084 * The zone allocator couldn't provide more entries.
1085 * Treat this as if the cache was full; drop the oldest
1086 * entry and insert the new one.
1088 tcpstat.tcps_sc_zonefail++;
1089 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1090 syncache_drop(sc, sch);
1091 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
1093 if (tcp_syncookies) {
1094 bzero(&scs, sizeof(scs));
1099 (void) m_free(ipopts);
1106 * Fill in the syncache values.
1109 sc->sc_label = maclabel;
1111 sc->sc_ipopts = ipopts;
1112 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1114 if (!inc->inc_isipv6)
1117 sc->sc_ip_tos = ip_tos;
1118 sc->sc_ip_ttl = ip_ttl;
1121 sc->sc_irs = th->th_seq;
1122 sc->sc_iss = arc4random();
1124 sc->sc_flowlabel = 0;
1127 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
1128 * win was derived from socket earlier in the function.
1131 win = imin(win, TCP_MAXWIN);
1134 if (tcp_do_rfc1323) {
1136 * A timestamp received in a SYN makes
1137 * it ok to send timestamp requests and replies.
1139 if (to->to_flags & TOF_TS) {
1140 sc->sc_tsreflect = to->to_tsval;
1142 sc->sc_flags |= SCF_TIMESTAMP;
1144 if (to->to_flags & TOF_SCALE) {
1148 * Pick the smallest possible scaling factor that
1149 * will still allow us to scale up to sb_max, aka
1150 * kern.ipc.maxsockbuf.
1152 * We do this because there are broken firewalls that
1153 * will corrupt the window scale option, leading to
1154 * the other endpoint believing that our advertised
1155 * window is unscaled. At scale factors larger than
1156 * 5 the unscaled window will drop below 1500 bytes,
1157 * leading to serious problems when traversing these
1160 * With the default maxsockbuf of 256K, a scale factor
1161 * of 3 will be chosen by this algorithm. Those who
1162 * choose a larger maxsockbuf should watch out
1163 * for the compatiblity problems mentioned above.
1165 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1166 * or <SYN,ACK>) segment itself is never scaled.
1168 while (wscale < TCP_MAX_WINSHIFT &&
1169 (TCP_MAXWIN << wscale) < sb_max)
1171 sc->sc_requested_r_scale = wscale;
1172 sc->sc_requested_s_scale = to->to_wscale;
1173 sc->sc_flags |= SCF_WINSCALE;
1176 #ifdef TCP_SIGNATURE
1178 * If listening socket requested TCP digests, and received SYN
1179 * contains the option, flag this in the syncache so that
1180 * syncache_respond() will do the right thing with the SYN+ACK.
1181 * XXX: Currently we always record the option by default and will
1182 * attempt to use it in syncache_respond().
1184 if (to->to_flags & TOF_SIGNATURE)
1185 sc->sc_flags |= SCF_SIGNATURE;
1187 if (to->to_flags & TOF_SACKPERM)
1188 sc->sc_flags |= SCF_SACK;
1189 if (to->to_flags & TOF_MSS)
1190 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */
1192 sc->sc_flags |= SCF_NOOPT;
1194 if (tcp_syncookies) {
1195 syncookie_generate(sch, sc, &flowtmp);
1198 sc->sc_flowlabel = flowtmp;
1204 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1210 * Do a standard 3-way handshake.
1212 if (syncache_respond(sc) == 0) {
1213 if (tcp_syncookies && tcp_syncookiesonly && sc != &scs)
1215 else if (sc != &scs)
1216 syncache_insert(sc, sch); /* locks and unlocks sch */
1217 tcpstat.tcps_sndacks++;
1218 tcpstat.tcps_sndtotal++;
1222 tcpstat.tcps_sc_dropped++;
1228 mac_syncache_destroy(&maclabel);
1236 syncache_respond(struct syncache *sc)
1238 struct ip *ip = NULL;
1242 u_int16_t hlen, tlen, mssopt;
1245 struct ip6_hdr *ip6 = NULL;
1250 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) :
1253 tlen = hlen + sizeof(struct tcphdr);
1255 /* Determine MSS we advertize to other end of connection. */
1256 mssopt = tcp_mssopt(&sc->sc_inc);
1257 if (sc->sc_peer_mss)
1258 mssopt = max( min(sc->sc_peer_mss, mssopt), tcp_minmss);
1260 /* XXX: Assume that the entire packet will fit in a header mbuf. */
1261 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1262 ("syncache: mbuf too small"));
1264 /* Create the IP+TCP header from scratch. */
1265 m = m_gethdr(M_DONTWAIT, MT_DATA);
1269 mac_syncache_create_mbuf(sc->sc_label, m);
1271 m->m_data += max_linkhdr;
1273 m->m_pkthdr.len = tlen;
1274 m->m_pkthdr.rcvif = NULL;
1277 if (sc->sc_inc.inc_isipv6) {
1278 ip6 = mtod(m, struct ip6_hdr *);
1279 ip6->ip6_vfc = IPV6_VERSION;
1280 ip6->ip6_nxt = IPPROTO_TCP;
1281 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1282 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1283 ip6->ip6_plen = htons(tlen - hlen);
1284 /* ip6_hlim is set after checksum */
1285 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1286 ip6->ip6_flow |= sc->sc_flowlabel;
1288 th = (struct tcphdr *)(ip6 + 1);
1292 ip = mtod(m, struct ip *);
1293 ip->ip_v = IPVERSION;
1294 ip->ip_hl = sizeof(struct ip) >> 2;
1299 ip->ip_p = IPPROTO_TCP;
1300 ip->ip_src = sc->sc_inc.inc_laddr;
1301 ip->ip_dst = sc->sc_inc.inc_faddr;
1302 ip->ip_ttl = sc->sc_ip_ttl;
1303 ip->ip_tos = sc->sc_ip_tos;
1306 * See if we should do MTU discovery. Route lookups are
1307 * expensive, so we will only unset the DF bit if:
1309 * 1) path_mtu_discovery is disabled
1310 * 2) the SCF_UNREACH flag has been set
1312 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1313 ip->ip_off |= IP_DF;
1315 th = (struct tcphdr *)(ip + 1);
1317 th->th_sport = sc->sc_inc.inc_lport;
1318 th->th_dport = sc->sc_inc.inc_fport;
1320 th->th_seq = htonl(sc->sc_iss);
1321 th->th_ack = htonl(sc->sc_irs + 1);
1322 th->th_off = sizeof(struct tcphdr) >> 2;
1324 th->th_flags = TH_SYN|TH_ACK;
1325 th->th_win = htons(sc->sc_wnd);
1328 /* Tack on the TCP options. */
1329 if ((sc->sc_flags & SCF_NOOPT) == 0) {
1333 to.to_flags = TOF_MSS;
1334 if (sc->sc_flags & SCF_WINSCALE) {
1335 to.to_wscale = sc->sc_requested_r_scale;
1336 to.to_flags |= TOF_SCALE;
1338 if (sc->sc_flags & SCF_TIMESTAMP) {
1339 /* Virgin timestamp or TCP cookie enhanced one. */
1340 to.to_tsval = sc->sc_ts;
1341 to.to_tsecr = sc->sc_tsreflect;
1342 to.to_flags |= TOF_TS;
1344 if (sc->sc_flags & SCF_SACK)
1345 to.to_flags |= TOF_SACKPERM;
1346 #ifdef TCP_SIGNATURE
1347 if (sc->sc_flags & SCF_SIGNATURE)
1348 to.to_flags |= TOF_SIGNATURE;
1350 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1352 /* Adjust headers by option size. */
1353 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1355 m->m_pkthdr.len += optlen;
1357 #ifdef TCP_SIGNATURE
1358 if (sc->sc_flags & SCF_SIGNATURE)
1359 tcp_signature_compute(m, sizeof(struct ip), 0, optlen,
1360 to.to_signature, IPSEC_DIR_OUTBOUND);
1363 if (sc->sc_inc.inc_isipv6)
1364 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1367 ip->ip_len += optlen;
1372 if (sc->sc_inc.inc_isipv6) {
1374 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
1375 tlen + optlen - hlen);
1376 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1377 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1381 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1382 htons(tlen + optlen - hlen + IPPROTO_TCP));
1383 m->m_pkthdr.csum_flags = CSUM_TCP;
1384 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1385 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1391 * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1392 * receive and to be able to handle SYN floods from bogus source addresses
1393 * (where we will never receive any reply). SYN floods try to exhaust all
1394 * our memory and available slots in the SYN cache table to cause a denial
1395 * of service to legitimate users of the local host.
1397 * The idea of SYN cookies is to encode and include all necessary information
1398 * about the connection setup state within the SYN-ACK we send back and thus
1399 * to get along without keeping any local state until the ACK to the SYN-ACK
1400 * arrives (if ever). Everything we need to know should be available from
1401 * the information we encoded in the SYN-ACK.
1403 * More information about the theory behind SYN cookies and its first
1404 * discussion and specification can be found at:
1405 * http://cr.yp.to/syncookies.html (overview)
1406 * http://cr.yp.to/syncookies/archive (gory details)
1408 * This implementation extends the orginal idea and first implementation
1409 * of FreeBSD by using not only the initial sequence number field to store
1410 * information but also the timestamp field if present. This way we can
1411 * keep track of the entire state we need to know to recreate the session in
1412 * its original form. Almost all TCP speakers implement RFC1323 timestamps
1413 * these days. For those that do not we still have to live with the known
1414 * shortcomings of the ISN only SYN cookies.
1418 * Initial sequence number we send:
1419 * 31|................................|0
1420 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1421 * D = MD5 Digest (first dword)
1423 * R = Rotation of secret
1424 * P = Odd or Even secret
1426 * The MD5 Digest is computed with over following parameters:
1427 * a) randomly rotated secret
1428 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1429 * c) the received initial sequence number from remote host
1430 * d) the rotation offset and odd/even bit
1432 * Timestamp we send:
1433 * 31|................................|0
1434 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1435 * D = MD5 Digest (third dword) (only as filler)
1436 * S = Requested send window scale
1437 * R = Requested receive window scale
1439 * 5 = TCP-MD5 enabled (not implemented yet)
1440 * XORed with MD5 Digest (forth dword)
1442 * The timestamp isn't cryptographically secure and doesn't need to be.
1443 * The double use of the MD5 digest dwords ties it to a specific remote/
1444 * local host/port, remote initial sequence number and our local time
1445 * limited secret. A received timestamp is reverted (XORed) and then
1446 * the contained MD5 dword is compared to the computed one to ensure the
1447 * timestamp belongs to the SYN-ACK we sent. The other parameters may
1448 * have been tampered with but this isn't different from supplying bogus
1449 * values in the SYN in the first place.
1451 * Some problems with SYN cookies remain however:
1452 * Consider the problem of a recreated (and retransmitted) cookie. If the
1453 * original SYN was accepted, the connection is established. The second
1454 * SYN is inflight, and if it arrives with an ISN that falls within the
1455 * receive window, the connection is killed.
1458 * A heuristic to determine when to accept syn cookies is not necessary.
1459 * An ACK flood would cause the syncookie verification to be attempted,
1460 * but a SYN flood causes syncookies to be generated. Both are of equal
1461 * cost, so there's no point in trying to optimize the ACK flood case.
1462 * Also, if you don't process certain ACKs for some reason, then all someone
1463 * would have to do is launch a SYN and ACK flood at the same time, which
1464 * would stop cookie verification and defeat the entire purpose of syncookies.
1466 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1469 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1470 u_int32_t *flowlabel)
1473 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1476 u_int off, pmss, mss;
1479 SCH_LOCK_ASSERT(sch);
1481 /* Which of the two secrets to use. */
1482 secbits = sch->sch_oddeven ?
1483 sch->sch_secbits_odd : sch->sch_secbits_even;
1485 /* Reseed secret if too old. */
1486 if (sch->sch_reseed < time_uptime) {
1487 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */
1488 secbits = sch->sch_oddeven ?
1489 sch->sch_secbits_odd : sch->sch_secbits_even;
1490 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1491 secbits[i] = arc4random();
1492 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1495 /* Secret rotation offset. */
1496 off = sc->sc_iss & 0x7; /* iss was randomized before */
1498 /* Maximum segment size calculation. */
1499 pmss = max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), tcp_minmss);
1500 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1501 if (tcp_sc_msstab[mss] <= pmss)
1504 /* Fold parameters and MD5 digest into the ISN we will send. */
1505 data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1506 data |= off << 1; /* secret offset, derived from iss, 3 bits */
1507 data |= mss << 4; /* mss, 3 bits */
1510 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1511 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1512 MD5Update(&ctx, secbits, off);
1513 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1514 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1515 MD5Update(&ctx, &data, sizeof(data));
1516 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1518 data |= (md5_buffer[0] << 7);
1522 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1525 /* Additional parameters are stored in the timestamp if present. */
1526 if (sc->sc_flags & SCF_TIMESTAMP) {
1527 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1528 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1529 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */
1530 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */
1531 data |= md5_buffer[2] << 10; /* more digest bits */
1532 data ^= md5_buffer[3];
1534 sc->sc_tsoff = data - ticks; /* after XOR */
1537 tcpstat.tcps_sc_sendcookie++;
1541 static struct syncache *
1542 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1543 struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1547 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1551 int off, mss, wnd, flags;
1553 SCH_LOCK_ASSERT(sch);
1556 * Pull information out of SYN-ACK/ACK and
1557 * revert sequence number advances.
1559 ack = th->th_ack - 1;
1560 seq = th->th_seq - 1;
1561 off = (ack >> 1) & 0x7;
1562 mss = (ack >> 4) & 0x7;
1565 /* Which of the two secrets to use. */
1566 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1569 * The secret wasn't updated for the lifetime of a syncookie,
1570 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1572 if (sch->sch_reseed < time_uptime) {
1576 /* Recompute the digest so we can compare it. */
1578 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1579 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1580 MD5Update(&ctx, secbits, off);
1581 MD5Update(&ctx, inc, sizeof(*inc));
1582 MD5Update(&ctx, &seq, sizeof(seq));
1583 MD5Update(&ctx, &flags, sizeof(flags));
1584 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1586 /* Does the digest part of or ACK'ed ISS match? */
1587 if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1590 /* Does the digest part of our reflected timestamp match? */
1591 if (to->to_flags & TOF_TS) {
1592 data = md5_buffer[3] ^ to->to_tsecr;
1593 if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1597 /* Fill in the syncache values. */
1598 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1599 sc->sc_ipopts = NULL;
1605 if (inc->inc_isipv6) {
1606 if (sotoinpcb(so)->in6p_flags & IN6P_AUTOFLOWLABEL)
1607 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1611 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1612 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1615 /* Additional parameters that were encoded in the timestamp. */
1617 sc->sc_flags |= SCF_TIMESTAMP;
1618 sc->sc_tsreflect = to->to_tsval;
1619 sc->sc_ts = to->to_tsecr;
1620 sc->sc_tsoff = to->to_tsecr - ticks;
1621 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1622 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1623 sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1625 sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1627 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1628 sc->sc_flags |= SCF_WINSCALE;
1630 sc->sc_flags |= SCF_NOOPT;
1632 wnd = sbspace(&so->so_rcv);
1634 wnd = imin(wnd, TCP_MAXWIN);
1638 sc->sc_peer_mss = tcp_sc_msstab[mss];
1640 tcpstat.tcps_sc_recvcookie++;
1645 * Returns the current number of syncache entries. This number
1646 * will probably change before you get around to calling
1651 syncache_pcbcount(void)
1653 struct syncache_head *sch;
1656 for (count = 0, i = 0; i < tcp_syncache.hashsize; i++) {
1657 /* No need to lock for a read. */
1658 sch = &tcp_syncache.hashbase[i];
1659 count += sch->sch_length;
1665 * Exports the syncache entries to userland so that netstat can display
1666 * them alongside the other sockets. This function is intended to be
1667 * called only from tcp_pcblist.
1669 * Due to concurrency on an active system, the number of pcbs exported
1670 * may have no relation to max_pcbs. max_pcbs merely indicates the
1671 * amount of space the caller allocated for this function to use.
1674 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
1677 struct syncache *sc;
1678 struct syncache_head *sch;
1679 int count, error, i;
1681 for (count = 0, error = 0, i = 0; i < tcp_syncache.hashsize; i++) {
1682 sch = &tcp_syncache.hashbase[i];
1684 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
1685 if (count >= max_pcbs) {
1689 bzero(&xt, sizeof(xt));
1690 xt.xt_len = sizeof(xt);
1691 if (sc->sc_inc.inc_isipv6)
1692 xt.xt_inp.inp_vflag = INP_IPV6;
1694 xt.xt_inp.inp_vflag = INP_IPV4;
1695 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
1696 xt.xt_tp.t_inpcb = &xt.xt_inp;
1697 xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
1698 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1699 xt.xt_socket.xso_len = sizeof (struct xsocket);
1700 xt.xt_socket.so_type = SOCK_STREAM;
1701 xt.xt_socket.so_state = SS_ISCONNECTING;
1702 error = SYSCTL_OUT(req, &xt, sizeof xt);
1712 *pcbs_exported = count;