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>
45 #include <sys/limits.h>
47 #include <sys/mutex.h>
48 #include <sys/malloc.h>
51 #include <sys/proc.h> /* for proc0 declaration */
52 #include <sys/random.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/syslog.h>
56 #include <sys/ucred.h>
57 #include <sys/vimage.h>
62 #include <net/route.h>
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/ip.h>
67 #include <netinet/in_var.h>
68 #include <netinet/in_pcb.h>
69 #include <netinet/ip_var.h>
70 #include <netinet/ip_options.h>
72 #include <netinet/ip6.h>
73 #include <netinet/icmp6.h>
74 #include <netinet6/nd6.h>
75 #include <netinet6/ip6_var.h>
76 #include <netinet6/in6_pcb.h>
78 #include <netinet/tcp.h>
79 #include <netinet/tcp_fsm.h>
80 #include <netinet/tcp_seq.h>
81 #include <netinet/tcp_timer.h>
82 #include <netinet/tcp_var.h>
83 #include <netinet/tcp_syncache.h>
84 #include <netinet/tcp_offload.h>
86 #include <netinet6/tcp6_var.h>
90 #include <netipsec/ipsec.h>
92 #include <netipsec/ipsec6.h>
94 #include <netipsec/key.h>
97 #include <machine/in_cksum.h>
99 #include <security/mac/mac_framework.h>
101 static int tcp_syncookies = 1;
102 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
104 "Use TCP SYN cookies if the syncache overflows");
106 static int tcp_syncookiesonly = 0;
107 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW,
108 &tcp_syncookiesonly, 0,
109 "Use only TCP SYN cookies");
111 #ifdef TCP_OFFLOAD_DISABLE
112 #define TOEPCB_ISSET(sc) (0)
114 #define TOEPCB_ISSET(sc) ((sc)->sc_toepcb != NULL)
117 static void syncache_drop(struct syncache *, struct syncache_head *);
118 static void syncache_free(struct syncache *);
119 static void syncache_insert(struct syncache *, struct syncache_head *);
120 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
121 static int syncache_respond(struct syncache *);
122 static struct socket *syncache_socket(struct syncache *, struct socket *,
124 static void syncache_timeout(struct syncache *sc, struct syncache_head *sch,
126 static void syncache_timer(void *);
127 static void syncookie_generate(struct syncache_head *, struct syncache *,
129 static struct syncache
130 *syncookie_lookup(struct in_conninfo *, struct syncache_head *,
131 struct syncache *, struct tcpopt *, struct tcphdr *,
135 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
136 * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds,
137 * the odds are that the user has given up attempting to connect by then.
139 #define SYNCACHE_MAXREXMTS 3
141 /* Arbitrary values */
142 #define TCP_SYNCACHE_HASHSIZE 512
143 #define TCP_SYNCACHE_BUCKETLIMIT 30
145 static struct tcp_syncache tcp_syncache;
147 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
149 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
150 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
152 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
153 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
155 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
156 &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
158 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
159 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
161 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
162 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
164 int tcp_sc_rst_sock_fail = 1;
165 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail, CTLFLAG_RW,
166 &tcp_sc_rst_sock_fail, 0, "Send reset on socket allocation failure");
168 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
170 #define SYNCACHE_HASH(inc, mask) \
171 ((V_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 ((V_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 SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx)
192 #define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx)
193 #define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED)
196 * Requires the syncache entry to be already removed from the bucket list.
199 syncache_free(struct syncache *sc)
202 (void) m_free(sc->sc_ipopts);
206 mac_syncache_destroy(&sc->sc_label);
209 uma_zfree(V_tcp_syncache.zone, sc);
217 V_tcp_syncache.cache_count = 0;
218 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
219 V_tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
220 V_tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
221 V_tcp_syncache.hash_secret = arc4random();
223 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
224 &V_tcp_syncache.hashsize);
225 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
226 &V_tcp_syncache.bucket_limit);
227 if (!powerof2(V_tcp_syncache.hashsize) ||
228 V_tcp_syncache.hashsize == 0) {
229 printf("WARNING: syncache hash size is not a power of 2.\n");
230 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
232 V_tcp_syncache.hashmask = V_tcp_syncache.hashsize - 1;
235 V_tcp_syncache.cache_limit =
236 V_tcp_syncache.hashsize * V_tcp_syncache.bucket_limit;
237 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
238 &V_tcp_syncache.cache_limit);
240 /* Allocate the hash table. */
241 MALLOC(V_tcp_syncache.hashbase, struct syncache_head *,
242 V_tcp_syncache.hashsize * sizeof(struct syncache_head),
243 M_SYNCACHE, M_WAITOK | M_ZERO);
245 /* Initialize the hash buckets. */
246 for (i = 0; i < V_tcp_syncache.hashsize; i++) {
247 TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket);
248 mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
250 callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer,
251 &V_tcp_syncache.hashbase[i].sch_mtx, 0);
252 V_tcp_syncache.hashbase[i].sch_length = 0;
255 /* Create the syncache entry zone. */
256 V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
257 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
258 uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit);
262 * Inserts a syncache entry into the specified bucket row.
263 * Locks and unlocks the syncache_head autonomously.
266 syncache_insert(struct syncache *sc, struct syncache_head *sch)
268 struct syncache *sc2;
273 * Make sure that we don't overflow the per-bucket limit.
274 * If the bucket is full, toss the oldest element.
276 if (sch->sch_length >= V_tcp_syncache.bucket_limit) {
277 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
278 ("sch->sch_length incorrect"));
279 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
280 syncache_drop(sc2, sch);
281 V_tcpstat.tcps_sc_bucketoverflow++;
284 /* Put it into the bucket. */
285 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
288 /* Reinitialize the bucket row's timer. */
289 if (sch->sch_length == 1)
290 sch->sch_nextc = ticks + INT_MAX;
291 syncache_timeout(sc, sch, 1);
295 V_tcp_syncache.cache_count++;
296 V_tcpstat.tcps_sc_added++;
300 * Remove and free entry from syncache bucket row.
301 * Expects locked syncache head.
304 syncache_drop(struct syncache *sc, struct syncache_head *sch)
307 SCH_LOCK_ASSERT(sch);
309 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
312 #ifndef TCP_OFFLOAD_DISABLE
314 sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb);
317 V_tcp_syncache.cache_count--;
321 * Engage/reengage time on bucket row.
324 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
326 sc->sc_rxttime = ticks +
327 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
329 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
330 sch->sch_nextc = sc->sc_rxttime;
332 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
333 syncache_timer, (void *)sch);
338 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
339 * If we have retransmitted an entry the maximum number of times, expire it.
340 * One separate timer for each bucket row.
343 syncache_timer(void *xsch)
345 struct syncache_head *sch = (struct syncache_head *)xsch;
346 struct syncache *sc, *nsc;
350 /* NB: syncache_head has already been locked by the callout. */
351 SCH_LOCK_ASSERT(sch);
354 * In the following cycle we may remove some entries and/or
355 * advance some timeouts, so re-initialize the bucket timer.
357 sch->sch_nextc = tick + INT_MAX;
359 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
361 * We do not check if the listen socket still exists
362 * and accept the case where the listen socket may be
363 * gone by the time we resend the SYN/ACK. We do
364 * not expect this to happens often. If it does,
365 * then the RST will be sent by the time the remote
366 * host does the SYN/ACK->ACK.
368 if (TSTMP_GT(sc->sc_rxttime, tick)) {
369 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
370 sch->sch_nextc = sc->sc_rxttime;
373 if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) {
374 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
375 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
376 "giving up and removing syncache entry\n",
380 syncache_drop(sc, sch);
381 V_tcpstat.tcps_sc_stale++;
384 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
385 log(LOG_DEBUG, "%s; %s: Response timeout, "
386 "retransmitting (%u) SYN|ACK\n",
387 s, __func__, sc->sc_rxmits);
391 (void) syncache_respond(sc);
392 V_tcpstat.tcps_sc_retransmitted++;
393 syncache_timeout(sc, sch, 0);
395 if (!TAILQ_EMPTY(&(sch)->sch_bucket))
396 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
397 syncache_timer, (void *)(sch));
401 * Find an entry in the syncache.
402 * Returns always with locked syncache_head plus a matching entry or NULL.
405 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
408 struct syncache_head *sch;
411 if (inc->inc_isipv6) {
412 sch = &V_tcp_syncache.hashbase[
413 SYNCACHE_HASH6(inc, V_tcp_syncache.hashmask)];
418 /* Circle through bucket row to find matching entry. */
419 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
420 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
426 sch = &V_tcp_syncache.hashbase[
427 SYNCACHE_HASH(inc, V_tcp_syncache.hashmask)];
432 /* Circle through bucket row to find matching entry. */
433 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
435 if (sc->sc_inc.inc_isipv6)
438 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
442 SCH_LOCK_ASSERT(*schp);
443 return (NULL); /* always returns with locked sch */
447 * This function is called when we get a RST for a
448 * non-existent connection, so that we can see if the
449 * connection is in the syn cache. If it is, zap it.
452 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
455 struct syncache_head *sch;
458 sc = syncache_lookup(inc, &sch); /* returns locked sch */
459 SCH_LOCK_ASSERT(sch);
462 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
463 * See RFC 793 page 65, section SEGMENT ARRIVES.
465 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
466 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
467 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
468 "FIN flag set, segment ignored\n", s, __func__);
469 V_tcpstat.tcps_badrst++;
474 * No corresponding connection was found in syncache.
475 * If syncookies are enabled and possibly exclusively
476 * used, or we are under memory pressure, a valid RST
477 * may not find a syncache entry. In that case we're
478 * done and no SYN|ACK retransmissions will happen.
479 * Otherwise the the RST was misdirected or spoofed.
482 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
483 log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
484 "syncache entry (possibly syncookie only), "
485 "segment ignored\n", s, __func__);
486 V_tcpstat.tcps_badrst++;
491 * If the RST bit is set, check the sequence number to see
492 * if this is a valid reset segment.
494 * In all states except SYN-SENT, all reset (RST) segments
495 * are validated by checking their SEQ-fields. A reset is
496 * valid if its sequence number is in the window.
498 * The sequence number in the reset segment is normally an
499 * echo of our outgoing acknowlegement numbers, but some hosts
500 * send a reset with the sequence number at the rightmost edge
501 * of our receive window, and we have to handle this case.
503 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
504 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
505 syncache_drop(sc, sch);
506 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
507 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
508 "connection attempt aborted by remote endpoint\n",
510 V_tcpstat.tcps_sc_reset++;
512 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
513 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != "
514 "IRS %u (+WND %u), segment ignored\n",
515 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
516 V_tcpstat.tcps_badrst++;
526 syncache_badack(struct in_conninfo *inc)
529 struct syncache_head *sch;
531 sc = syncache_lookup(inc, &sch); /* returns locked sch */
532 SCH_LOCK_ASSERT(sch);
534 syncache_drop(sc, sch);
535 V_tcpstat.tcps_sc_badack++;
541 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
544 struct syncache_head *sch;
546 sc = syncache_lookup(inc, &sch); /* returns locked sch */
547 SCH_LOCK_ASSERT(sch);
551 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
552 if (ntohl(th->th_seq) != sc->sc_iss)
556 * If we've rertransmitted 3 times and this is our second error,
557 * we remove the entry. Otherwise, we allow it to continue on.
558 * This prevents us from incorrectly nuking an entry during a
559 * spurious network outage.
563 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
564 sc->sc_flags |= SCF_UNREACH;
567 syncache_drop(sc, sch);
568 V_tcpstat.tcps_sc_unreach++;
574 * Build a new TCP socket structure from a syncache entry.
576 static struct socket *
577 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
579 struct inpcb *inp = NULL;
584 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
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 either send a RST or
596 * have the peer retransmit its SYN again after its
599 V_tcpstat.tcps_listendrop++;
600 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
601 log(LOG_DEBUG, "%s; %s: Socket create failed "
602 "due to limits or memory shortage\n",
610 mac_socketpeer_set_from_mbuf(m, so);
615 inp->inp_inc.inc_fibnum = sc->sc_inc.inc_fibnum;
616 so->so_fibnum = sc->sc_inc.inc_fibnum;
619 /* Insert new socket into PCB hash list. */
620 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
622 if (sc->sc_inc.inc_isipv6) {
623 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
625 inp->inp_vflag &= ~INP_IPV6;
626 inp->inp_vflag |= INP_IPV4;
628 inp->inp_laddr = sc->sc_inc.inc_laddr;
632 inp->inp_lport = sc->sc_inc.inc_lport;
633 if (in_pcbinshash(inp) != 0) {
635 * Undo the assignments above if we failed to
636 * put the PCB on the hash lists.
639 if (sc->sc_inc.inc_isipv6)
640 inp->in6p_laddr = in6addr_any;
643 inp->inp_laddr.s_addr = INADDR_ANY;
648 /* Copy old policy into new socket's. */
649 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
650 printf("syncache_socket: could not copy policy\n");
653 if (sc->sc_inc.inc_isipv6) {
654 struct inpcb *oinp = sotoinpcb(lso);
655 struct in6_addr laddr6;
656 struct sockaddr_in6 sin6;
658 * Inherit socket options from the listening socket.
659 * Note that in6p_inputopts are not (and should not be)
660 * copied, since it stores previously received options and is
661 * used to detect if each new option is different than the
662 * previous one and hence should be passed to a user.
663 * If we copied in6p_inputopts, a user would not be able to
664 * receive options just after calling the accept system call.
666 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
667 if (oinp->in6p_outputopts)
668 inp->in6p_outputopts =
669 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
671 sin6.sin6_family = AF_INET6;
672 sin6.sin6_len = sizeof(sin6);
673 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
674 sin6.sin6_port = sc->sc_inc.inc_fport;
675 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
676 laddr6 = inp->in6p_laddr;
677 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
678 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
679 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
681 inp->in6p_laddr = laddr6;
684 /* Override flowlabel from in6_pcbconnect. */
685 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK;
686 inp->in6p_flowinfo |= sc->sc_flowlabel;
690 struct in_addr laddr;
691 struct sockaddr_in sin;
693 inp->inp_options = (m) ? ip_srcroute(m) : NULL;
695 if (inp->inp_options == NULL) {
696 inp->inp_options = sc->sc_ipopts;
697 sc->sc_ipopts = NULL;
700 sin.sin_family = AF_INET;
701 sin.sin_len = sizeof(sin);
702 sin.sin_addr = sc->sc_inc.inc_faddr;
703 sin.sin_port = sc->sc_inc.inc_fport;
704 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
705 laddr = inp->inp_laddr;
706 if (inp->inp_laddr.s_addr == INADDR_ANY)
707 inp->inp_laddr = sc->sc_inc.inc_laddr;
708 if (in_pcbconnect(inp, (struct sockaddr *)&sin,
710 inp->inp_laddr = laddr;
715 tp->t_state = TCPS_SYN_RECEIVED;
716 tp->iss = sc->sc_iss;
717 tp->irs = sc->sc_irs;
720 tp->snd_wl1 = sc->sc_irs;
721 tp->snd_max = tp->iss + 1;
722 tp->snd_nxt = tp->iss + 1;
723 tp->rcv_up = sc->sc_irs + 1;
724 tp->rcv_wnd = sc->sc_wnd;
725 tp->rcv_adv += tp->rcv_wnd;
726 tp->last_ack_sent = tp->rcv_nxt;
728 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
729 if (sc->sc_flags & SCF_NOOPT)
730 tp->t_flags |= TF_NOOPT;
732 if (sc->sc_flags & SCF_WINSCALE) {
733 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
734 tp->snd_scale = sc->sc_requested_s_scale;
735 tp->request_r_scale = sc->sc_requested_r_scale;
737 if (sc->sc_flags & SCF_TIMESTAMP) {
738 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
739 tp->ts_recent = sc->sc_tsreflect;
740 tp->ts_recent_age = ticks;
741 tp->ts_offset = sc->sc_tsoff;
744 if (sc->sc_flags & SCF_SIGNATURE)
745 tp->t_flags |= TF_SIGNATURE;
747 if (sc->sc_flags & SCF_SACK)
748 tp->t_flags |= TF_SACK_PERMIT;
751 if (sc->sc_flags & SCF_ECN)
752 tp->t_flags |= TF_ECN_PERMIT;
755 * Set up MSS and get cached values from tcp_hostcache.
756 * This might overwrite some of the defaults we just set.
758 tcp_mss(tp, sc->sc_peer_mss);
761 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
764 tp->snd_cwnd = tp->t_maxseg;
765 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
769 V_tcpstat.tcps_accepts++;
781 * This function gets called when we receive an ACK for a
782 * socket in the LISTEN state. We look up the connection
783 * in the syncache, and if its there, we pull it out of
784 * the cache and turn it into a full-blown connection in
785 * the SYN-RECEIVED state.
788 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
789 struct socket **lsop, struct mbuf *m)
792 struct syncache_head *sch;
797 * Global TCP locks are held because we manipulate the PCB lists
798 * and create a new socket.
800 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
801 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
802 ("%s: can handle only ACK", __func__));
804 sc = syncache_lookup(inc, &sch); /* returns locked sch */
805 SCH_LOCK_ASSERT(sch);
808 * There is no syncache entry, so see if this ACK is
809 * a returning syncookie. To do this, first:
810 * A. See if this socket has had a syncache entry dropped in
811 * the past. We don't want to accept a bogus syncookie
812 * if we've never received a SYN.
813 * B. check that the syncookie is valid. If it is, then
814 * cobble up a fake syncache entry, and return.
816 if (!tcp_syncookies) {
818 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
819 log(LOG_DEBUG, "%s; %s: Spurious ACK, "
820 "segment rejected (syncookies disabled)\n",
824 bzero(&scs, sizeof(scs));
825 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
828 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
829 log(LOG_DEBUG, "%s; %s: Segment failed "
830 "SYNCOOKIE authentication, segment rejected "
831 "(probably spoofed)\n", s, __func__);
835 /* Pull out the entry to unlock the bucket row. */
836 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
838 V_tcp_syncache.cache_count--;
843 * Segment validation:
844 * ACK must match our initial sequence number + 1 (the SYN|ACK).
846 if (th->th_ack != sc->sc_iss + 1 && !TOEPCB_ISSET(sc)) {
847 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
848 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment "
849 "rejected\n", s, __func__, th->th_ack, sc->sc_iss);
854 * The SEQ must fall in the window starting at the received
855 * initial receive sequence number + 1 (the SYN).
857 if ((SEQ_LEQ(th->th_seq, sc->sc_irs) ||
858 SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) &&
860 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
861 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment "
862 "rejected\n", s, __func__, th->th_seq, sc->sc_irs);
866 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
867 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
868 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
869 "segment rejected\n", s, __func__);
873 * If timestamps were negotiated the reflected timestamp
874 * must be equal to what we actually sent in the SYN|ACK.
876 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts &&
878 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
879 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
880 "segment rejected\n",
881 s, __func__, to->to_tsecr, sc->sc_ts);
885 *lsop = syncache_socket(sc, *lsop, m);
888 V_tcpstat.tcps_sc_aborted++;
890 V_tcpstat.tcps_sc_completed++;
892 /* how do we find the inp for the new socket? */
897 if (sc != NULL && sc != &scs)
906 tcp_offload_syncache_expand(struct in_conninfo *inc, struct tcpopt *to,
907 struct tcphdr *th, struct socket **lsop, struct mbuf *m)
911 INP_INFO_WLOCK(&V_tcbinfo);
912 rc = syncache_expand(inc, to, th, lsop, m);
913 INP_INFO_WUNLOCK(&V_tcbinfo);
919 * Given a LISTEN socket and an inbound SYN request, add
920 * this to the syn cache, and send back a segment:
921 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
924 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
925 * Doing so would require that we hold onto the data and deliver it
926 * to the application. However, if we are the target of a SYN-flood
927 * DoS attack, an attacker could send data which would eventually
928 * consume all available buffer space if it were ACKed. By not ACKing
929 * the data, we avoid this DoS scenario.
932 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
933 struct inpcb *inp, struct socket **lsop, struct mbuf *m,
934 struct toe_usrreqs *tu, void *toepcb)
938 struct syncache *sc = NULL;
939 struct syncache_head *sch;
940 struct mbuf *ipopts = NULL;
942 int win, sb_hiwat, ip_ttl, ip_tos, noopt;
945 int autoflowlabel = 0;
948 struct label *maclabel;
953 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
954 INP_WLOCK_ASSERT(inp); /* listen socket */
955 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
956 ("%s: unexpected tcp flags", __func__));
959 * Combine all so/tp operations very early to drop the INP lock as
964 cred = crhold(so->so_cred);
967 if (inc->inc_isipv6 &&
968 (inp->in6p_flags & IN6P_AUTOFLOWLABEL))
971 ip_ttl = inp->inp_ip_ttl;
972 ip_tos = inp->inp_ip_tos;
973 win = sbspace(&so->so_rcv);
974 sb_hiwat = so->so_rcv.sb_hiwat;
975 noopt = (tp->t_flags & TF_NOOPT);
977 /* By the time we drop the lock these should no longer be used. */
982 if (mac_syncache_init(&maclabel) != 0) {
984 INP_INFO_WUNLOCK(&V_tcbinfo);
987 mac_syncache_create(maclabel, inp);
990 INP_INFO_WUNLOCK(&V_tcbinfo);
993 * Remember the IP options, if any.
996 if (!inc->inc_isipv6)
998 ipopts = (m) ? ip_srcroute(m) : NULL;
1001 * See if we already have an entry for this connection.
1002 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1004 * XXX: should the syncache be re-initialized with the contents
1005 * of the new SYN here (which may have different options?)
1007 * XXX: We do not check the sequence number to see if this is a
1008 * real retransmit or a new connection attempt. The question is
1009 * how to handle such a case; either ignore it as spoofed, or
1010 * drop the current entry and create a new one?
1012 sc = syncache_lookup(inc, &sch); /* returns locked entry */
1013 SCH_LOCK_ASSERT(sch);
1015 #ifndef TCP_OFFLOAD_DISABLE
1017 sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT,
1020 V_tcpstat.tcps_sc_dupsyn++;
1023 * If we were remembering a previous source route,
1024 * forget it and use the new one we've been given.
1027 (void) m_free(sc->sc_ipopts);
1028 sc->sc_ipopts = ipopts;
1031 * Update timestamp if present.
1033 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
1034 sc->sc_tsreflect = to->to_tsval;
1036 sc->sc_flags &= ~SCF_TIMESTAMP;
1039 * Since we have already unconditionally allocated label
1040 * storage, free it up. The syncache entry will already
1041 * have an initialized label we can use.
1043 mac_syncache_destroy(&maclabel);
1044 KASSERT(sc->sc_label != NULL,
1045 ("%s: label not initialized", __func__));
1047 /* Retransmit SYN|ACK and reset retransmit count. */
1048 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
1049 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
1050 "resetting timer and retransmitting SYN|ACK\n",
1054 if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) {
1056 syncache_timeout(sc, sch, 1);
1057 V_tcpstat.tcps_sndacks++;
1058 V_tcpstat.tcps_sndtotal++;
1064 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1067 * The zone allocator couldn't provide more entries.
1068 * Treat this as if the cache was full; drop the oldest
1069 * entry and insert the new one.
1071 V_tcpstat.tcps_sc_zonefail++;
1072 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1073 syncache_drop(sc, sch);
1074 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1076 if (tcp_syncookies) {
1077 bzero(&scs, sizeof(scs));
1082 (void) m_free(ipopts);
1089 * Fill in the syncache values.
1092 sc->sc_label = maclabel;
1096 sc->sc_ipopts = ipopts;
1097 sc->sc_inc.inc_fibnum = inp->inp_inc.inc_fibnum;
1098 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1100 if (!inc->inc_isipv6)
1103 sc->sc_ip_tos = ip_tos;
1104 sc->sc_ip_ttl = ip_ttl;
1106 #ifndef TCP_OFFLOAD_DISABLE
1108 sc->sc_toepcb = toepcb;
1110 sc->sc_irs = th->th_seq;
1111 sc->sc_iss = arc4random();
1113 sc->sc_flowlabel = 0;
1116 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
1117 * win was derived from socket earlier in the function.
1120 win = imin(win, TCP_MAXWIN);
1123 if (V_tcp_do_rfc1323) {
1125 * A timestamp received in a SYN makes
1126 * it ok to send timestamp requests and replies.
1128 if (to->to_flags & TOF_TS) {
1129 sc->sc_tsreflect = to->to_tsval;
1131 sc->sc_flags |= SCF_TIMESTAMP;
1133 if (to->to_flags & TOF_SCALE) {
1137 * Pick the smallest possible scaling factor that
1138 * will still allow us to scale up to sb_max, aka
1139 * kern.ipc.maxsockbuf.
1141 * We do this because there are broken firewalls that
1142 * will corrupt the window scale option, leading to
1143 * the other endpoint believing that our advertised
1144 * window is unscaled. At scale factors larger than
1145 * 5 the unscaled window will drop below 1500 bytes,
1146 * leading to serious problems when traversing these
1149 * With the default maxsockbuf of 256K, a scale factor
1150 * of 3 will be chosen by this algorithm. Those who
1151 * choose a larger maxsockbuf should watch out
1152 * for the compatiblity problems mentioned above.
1154 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1155 * or <SYN,ACK>) segment itself is never scaled.
1157 while (wscale < TCP_MAX_WINSHIFT &&
1158 (TCP_MAXWIN << wscale) < sb_max)
1160 sc->sc_requested_r_scale = wscale;
1161 sc->sc_requested_s_scale = to->to_wscale;
1162 sc->sc_flags |= SCF_WINSCALE;
1165 #ifdef TCP_SIGNATURE
1167 * If listening socket requested TCP digests, and received SYN
1168 * contains the option, flag this in the syncache so that
1169 * syncache_respond() will do the right thing with the SYN+ACK.
1170 * XXX: Currently we always record the option by default and will
1171 * attempt to use it in syncache_respond().
1173 if (to->to_flags & TOF_SIGNATURE)
1174 sc->sc_flags |= SCF_SIGNATURE;
1176 if (to->to_flags & TOF_SACKPERM)
1177 sc->sc_flags |= SCF_SACK;
1178 if (to->to_flags & TOF_MSS)
1179 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */
1181 sc->sc_flags |= SCF_NOOPT;
1182 if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn)
1183 sc->sc_flags |= SCF_ECN;
1185 if (tcp_syncookies) {
1186 syncookie_generate(sch, sc, &flowtmp);
1189 sc->sc_flowlabel = flowtmp;
1195 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1201 * Do a standard 3-way handshake.
1203 if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) {
1204 if (tcp_syncookies && tcp_syncookiesonly && sc != &scs)
1206 else if (sc != &scs)
1207 syncache_insert(sc, sch); /* locks and unlocks sch */
1208 V_tcpstat.tcps_sndacks++;
1209 V_tcpstat.tcps_sndtotal++;
1213 V_tcpstat.tcps_sc_dropped++;
1221 mac_syncache_destroy(&maclabel);
1232 syncache_respond(struct syncache *sc)
1234 struct ip *ip = NULL;
1238 u_int16_t hlen, tlen, mssopt;
1241 struct ip6_hdr *ip6 = NULL;
1246 (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) :
1249 tlen = hlen + sizeof(struct tcphdr);
1251 /* Determine MSS we advertize to other end of connection. */
1252 mssopt = tcp_mssopt(&sc->sc_inc);
1253 if (sc->sc_peer_mss)
1254 mssopt = max( min(sc->sc_peer_mss, mssopt), V_tcp_minmss);
1256 /* XXX: Assume that the entire packet will fit in a header mbuf. */
1257 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1258 ("syncache: mbuf too small"));
1260 /* Create the IP+TCP header from scratch. */
1261 m = m_gethdr(M_DONTWAIT, MT_DATA);
1265 mac_syncache_create_mbuf(sc->sc_label, m);
1267 m->m_data += max_linkhdr;
1269 m->m_pkthdr.len = tlen;
1270 m->m_pkthdr.rcvif = NULL;
1273 if (sc->sc_inc.inc_isipv6) {
1274 ip6 = mtod(m, struct ip6_hdr *);
1275 ip6->ip6_vfc = IPV6_VERSION;
1276 ip6->ip6_nxt = IPPROTO_TCP;
1277 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1278 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1279 ip6->ip6_plen = htons(tlen - hlen);
1280 /* ip6_hlim is set after checksum */
1281 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1282 ip6->ip6_flow |= sc->sc_flowlabel;
1284 th = (struct tcphdr *)(ip6 + 1);
1288 ip = mtod(m, struct ip *);
1289 ip->ip_v = IPVERSION;
1290 ip->ip_hl = sizeof(struct ip) >> 2;
1295 ip->ip_p = IPPROTO_TCP;
1296 ip->ip_src = sc->sc_inc.inc_laddr;
1297 ip->ip_dst = sc->sc_inc.inc_faddr;
1298 ip->ip_ttl = sc->sc_ip_ttl;
1299 ip->ip_tos = sc->sc_ip_tos;
1302 * See if we should do MTU discovery. Route lookups are
1303 * expensive, so we will only unset the DF bit if:
1305 * 1) path_mtu_discovery is disabled
1306 * 2) the SCF_UNREACH flag has been set
1308 if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1309 ip->ip_off |= IP_DF;
1311 th = (struct tcphdr *)(ip + 1);
1313 th->th_sport = sc->sc_inc.inc_lport;
1314 th->th_dport = sc->sc_inc.inc_fport;
1316 th->th_seq = htonl(sc->sc_iss);
1317 th->th_ack = htonl(sc->sc_irs + 1);
1318 th->th_off = sizeof(struct tcphdr) >> 2;
1320 th->th_flags = TH_SYN|TH_ACK;
1321 th->th_win = htons(sc->sc_wnd);
1324 if (sc->sc_flags & SCF_ECN) {
1325 th->th_flags |= TH_ECE;
1326 V_tcpstat.tcps_ecn_shs++;
1329 /* Tack on the TCP options. */
1330 if ((sc->sc_flags & SCF_NOOPT) == 0) {
1334 to.to_flags = TOF_MSS;
1335 if (sc->sc_flags & SCF_WINSCALE) {
1336 to.to_wscale = sc->sc_requested_r_scale;
1337 to.to_flags |= TOF_SCALE;
1339 if (sc->sc_flags & SCF_TIMESTAMP) {
1340 /* Virgin timestamp or TCP cookie enhanced one. */
1341 to.to_tsval = sc->sc_ts;
1342 to.to_tsecr = sc->sc_tsreflect;
1343 to.to_flags |= TOF_TS;
1345 if (sc->sc_flags & SCF_SACK)
1346 to.to_flags |= TOF_SACKPERM;
1347 #ifdef TCP_SIGNATURE
1348 if (sc->sc_flags & SCF_SIGNATURE)
1349 to.to_flags |= TOF_SIGNATURE;
1351 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1353 /* Adjust headers by option size. */
1354 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1356 m->m_pkthdr.len += optlen;
1358 #ifdef TCP_SIGNATURE
1359 if (sc->sc_flags & SCF_SIGNATURE)
1360 tcp_signature_compute(m, sizeof(struct ip), 0, optlen,
1361 to.to_signature, IPSEC_DIR_OUTBOUND);
1364 if (sc->sc_inc.inc_isipv6)
1365 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1368 ip->ip_len += optlen;
1373 if (sc->sc_inc.inc_isipv6) {
1375 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
1376 tlen + optlen - hlen);
1377 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1378 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1382 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1383 htons(tlen + optlen - hlen + IPPROTO_TCP));
1384 m->m_pkthdr.csum_flags = CSUM_TCP;
1385 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1386 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1392 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1393 struct inpcb *inp, struct socket **lsop, struct mbuf *m)
1395 _syncache_add(inc, to, th, inp, lsop, m, NULL, NULL);
1399 tcp_offload_syncache_add(struct in_conninfo *inc, struct tcpopt *to,
1400 struct tcphdr *th, struct inpcb *inp, struct socket **lsop,
1401 struct toe_usrreqs *tu, void *toepcb)
1404 INP_INFO_WLOCK(&V_tcbinfo);
1406 _syncache_add(inc, to, th, inp, lsop, NULL, tu, toepcb);
1410 * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1411 * receive and to be able to handle SYN floods from bogus source addresses
1412 * (where we will never receive any reply). SYN floods try to exhaust all
1413 * our memory and available slots in the SYN cache table to cause a denial
1414 * of service to legitimate users of the local host.
1416 * The idea of SYN cookies is to encode and include all necessary information
1417 * about the connection setup state within the SYN-ACK we send back and thus
1418 * to get along without keeping any local state until the ACK to the SYN-ACK
1419 * arrives (if ever). Everything we need to know should be available from
1420 * the information we encoded in the SYN-ACK.
1422 * More information about the theory behind SYN cookies and its first
1423 * discussion and specification can be found at:
1424 * http://cr.yp.to/syncookies.html (overview)
1425 * http://cr.yp.to/syncookies/archive (gory details)
1427 * This implementation extends the orginal idea and first implementation
1428 * of FreeBSD by using not only the initial sequence number field to store
1429 * information but also the timestamp field if present. This way we can
1430 * keep track of the entire state we need to know to recreate the session in
1431 * its original form. Almost all TCP speakers implement RFC1323 timestamps
1432 * these days. For those that do not we still have to live with the known
1433 * shortcomings of the ISN only SYN cookies.
1437 * Initial sequence number we send:
1438 * 31|................................|0
1439 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1440 * D = MD5 Digest (first dword)
1442 * R = Rotation of secret
1443 * P = Odd or Even secret
1445 * The MD5 Digest is computed with over following parameters:
1446 * a) randomly rotated secret
1447 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1448 * c) the received initial sequence number from remote host
1449 * d) the rotation offset and odd/even bit
1451 * Timestamp we send:
1452 * 31|................................|0
1453 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1454 * D = MD5 Digest (third dword) (only as filler)
1455 * S = Requested send window scale
1456 * R = Requested receive window scale
1458 * 5 = TCP-MD5 enabled (not implemented yet)
1459 * XORed with MD5 Digest (forth dword)
1461 * The timestamp isn't cryptographically secure and doesn't need to be.
1462 * The double use of the MD5 digest dwords ties it to a specific remote/
1463 * local host/port, remote initial sequence number and our local time
1464 * limited secret. A received timestamp is reverted (XORed) and then
1465 * the contained MD5 dword is compared to the computed one to ensure the
1466 * timestamp belongs to the SYN-ACK we sent. The other parameters may
1467 * have been tampered with but this isn't different from supplying bogus
1468 * values in the SYN in the first place.
1470 * Some problems with SYN cookies remain however:
1471 * Consider the problem of a recreated (and retransmitted) cookie. If the
1472 * original SYN was accepted, the connection is established. The second
1473 * SYN is inflight, and if it arrives with an ISN that falls within the
1474 * receive window, the connection is killed.
1477 * A heuristic to determine when to accept syn cookies is not necessary.
1478 * An ACK flood would cause the syncookie verification to be attempted,
1479 * but a SYN flood causes syncookies to be generated. Both are of equal
1480 * cost, so there's no point in trying to optimize the ACK flood case.
1481 * Also, if you don't process certain ACKs for some reason, then all someone
1482 * would have to do is launch a SYN and ACK flood at the same time, which
1483 * would stop cookie verification and defeat the entire purpose of syncookies.
1485 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1488 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1489 u_int32_t *flowlabel)
1492 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1495 u_int off, pmss, mss;
1498 SCH_LOCK_ASSERT(sch);
1500 /* Which of the two secrets to use. */
1501 secbits = sch->sch_oddeven ?
1502 sch->sch_secbits_odd : sch->sch_secbits_even;
1504 /* Reseed secret if too old. */
1505 if (sch->sch_reseed < time_uptime) {
1506 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */
1507 secbits = sch->sch_oddeven ?
1508 sch->sch_secbits_odd : sch->sch_secbits_even;
1509 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1510 secbits[i] = arc4random();
1511 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1514 /* Secret rotation offset. */
1515 off = sc->sc_iss & 0x7; /* iss was randomized before */
1517 /* Maximum segment size calculation. */
1519 max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), V_tcp_minmss);
1520 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1521 if (tcp_sc_msstab[mss] <= pmss)
1524 /* Fold parameters and MD5 digest into the ISN we will send. */
1525 data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1526 data |= off << 1; /* secret offset, derived from iss, 3 bits */
1527 data |= mss << 4; /* mss, 3 bits */
1530 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1531 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1532 MD5Update(&ctx, secbits, off);
1533 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1534 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1535 MD5Update(&ctx, &data, sizeof(data));
1536 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1538 data |= (md5_buffer[0] << 7);
1542 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1545 /* Additional parameters are stored in the timestamp if present. */
1546 if (sc->sc_flags & SCF_TIMESTAMP) {
1547 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1548 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1549 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */
1550 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */
1551 data |= md5_buffer[2] << 10; /* more digest bits */
1552 data ^= md5_buffer[3];
1554 sc->sc_tsoff = data - ticks; /* after XOR */
1557 V_tcpstat.tcps_sc_sendcookie++;
1561 static struct syncache *
1562 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1563 struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1567 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1571 int off, mss, wnd, flags;
1573 SCH_LOCK_ASSERT(sch);
1576 * Pull information out of SYN-ACK/ACK and
1577 * revert sequence number advances.
1579 ack = th->th_ack - 1;
1580 seq = th->th_seq - 1;
1581 off = (ack >> 1) & 0x7;
1582 mss = (ack >> 4) & 0x7;
1585 /* Which of the two secrets to use. */
1586 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1589 * The secret wasn't updated for the lifetime of a syncookie,
1590 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1592 if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) {
1596 /* Recompute the digest so we can compare it. */
1598 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1599 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1600 MD5Update(&ctx, secbits, off);
1601 MD5Update(&ctx, inc, sizeof(*inc));
1602 MD5Update(&ctx, &seq, sizeof(seq));
1603 MD5Update(&ctx, &flags, sizeof(flags));
1604 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1606 /* Does the digest part of or ACK'ed ISS match? */
1607 if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1610 /* Does the digest part of our reflected timestamp match? */
1611 if (to->to_flags & TOF_TS) {
1612 data = md5_buffer[3] ^ to->to_tsecr;
1613 if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1617 /* Fill in the syncache values. */
1618 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1619 sc->sc_ipopts = NULL;
1625 if (inc->inc_isipv6) {
1626 if (sotoinpcb(so)->in6p_flags & IN6P_AUTOFLOWLABEL)
1627 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1631 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1632 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1635 /* Additional parameters that were encoded in the timestamp. */
1637 sc->sc_flags |= SCF_TIMESTAMP;
1638 sc->sc_tsreflect = to->to_tsval;
1639 sc->sc_ts = to->to_tsecr;
1640 sc->sc_tsoff = to->to_tsecr - ticks;
1641 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1642 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1643 sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1645 sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1647 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1648 sc->sc_flags |= SCF_WINSCALE;
1650 sc->sc_flags |= SCF_NOOPT;
1652 wnd = sbspace(&so->so_rcv);
1654 wnd = imin(wnd, TCP_MAXWIN);
1658 sc->sc_peer_mss = tcp_sc_msstab[mss];
1660 V_tcpstat.tcps_sc_recvcookie++;
1665 * Returns the current number of syncache entries. This number
1666 * will probably change before you get around to calling
1671 syncache_pcbcount(void)
1673 struct syncache_head *sch;
1676 for (count = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1677 /* No need to lock for a read. */
1678 sch = &V_tcp_syncache.hashbase[i];
1679 count += sch->sch_length;
1685 * Exports the syncache entries to userland so that netstat can display
1686 * them alongside the other sockets. This function is intended to be
1687 * called only from tcp_pcblist.
1689 * Due to concurrency on an active system, the number of pcbs exported
1690 * may have no relation to max_pcbs. max_pcbs merely indicates the
1691 * amount of space the caller allocated for this function to use.
1694 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
1697 struct syncache *sc;
1698 struct syncache_head *sch;
1699 int count, error, i;
1701 for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1702 sch = &V_tcp_syncache.hashbase[i];
1704 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
1705 if (count >= max_pcbs) {
1709 if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0)
1711 bzero(&xt, sizeof(xt));
1712 xt.xt_len = sizeof(xt);
1713 if (sc->sc_inc.inc_isipv6)
1714 xt.xt_inp.inp_vflag = INP_IPV6;
1716 xt.xt_inp.inp_vflag = INP_IPV4;
1717 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
1718 xt.xt_tp.t_inpcb = &xt.xt_inp;
1719 xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
1720 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1721 xt.xt_socket.xso_len = sizeof (struct xsocket);
1722 xt.xt_socket.so_type = SOCK_STREAM;
1723 xt.xt_socket.so_state = SS_ISCONNECTING;
1724 error = SYSCTL_OUT(req, &xt, sizeof xt);
1734 *pcbs_exported = count;