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"
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/sysctl.h>
44 #include <sys/limits.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>
55 #include <sys/ucred.h>
56 #include <sys/vimage.h>
61 #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 VNET_DEFINE(struct tcp_syncache, tcp_syncache);
102 static VNET_DEFINE(int, tcp_syncookies);
103 static VNET_DEFINE(int, tcp_syncookiesonly);
104 VNET_DEFINE(int, tcp_sc_rst_sock_fail);
106 #define V_tcp_syncache VNET(tcp_syncache)
107 #define V_tcp_syncookies VNET(tcp_syncookies)
108 #define V_tcp_syncookiesonly VNET(tcp_syncookiesonly)
110 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
111 &VNET_NAME(tcp_syncookies), 0,
112 "Use TCP SYN cookies if the syncache overflows");
114 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW,
115 &VNET_NAME(tcp_syncookiesonly), 0,
116 "Use only TCP SYN cookies");
118 #ifdef TCP_OFFLOAD_DISABLE
119 #define TOEPCB_ISSET(sc) (0)
121 #define TOEPCB_ISSET(sc) ((sc)->sc_toepcb != NULL)
124 static void syncache_drop(struct syncache *, struct syncache_head *);
125 static void syncache_free(struct syncache *);
126 static void syncache_insert(struct syncache *, struct syncache_head *);
127 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
128 static int syncache_respond(struct syncache *);
129 static struct socket *syncache_socket(struct syncache *, struct socket *,
131 static void syncache_timeout(struct syncache *sc, struct syncache_head *sch,
133 static void syncache_timer(void *);
134 static void syncookie_generate(struct syncache_head *, struct syncache *,
136 static struct syncache
137 *syncookie_lookup(struct in_conninfo *, struct syncache_head *,
138 struct syncache *, struct tcpopt *, struct tcphdr *,
142 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
143 * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds,
144 * the odds are that the user has given up attempting to connect by then.
146 #define SYNCACHE_MAXREXMTS 3
148 /* Arbitrary values */
149 #define TCP_SYNCACHE_HASHSIZE 512
150 #define TCP_SYNCACHE_BUCKETLIMIT 30
152 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
154 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
155 &VNET_NAME(tcp_syncache.bucket_limit), 0,
156 "Per-bucket hash limit for syncache");
158 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
159 &VNET_NAME(tcp_syncache.cache_limit), 0,
160 "Overall entry limit for syncache");
162 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
163 &VNET_NAME(tcp_syncache.cache_count), 0,
164 "Current number of entries in syncache");
166 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
167 &VNET_NAME(tcp_syncache.hashsize), 0,
168 "Size of TCP syncache hashtable");
170 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
171 &VNET_NAME(tcp_syncache.rexmt_limit), 0,
172 "Limit on SYN/ACK retransmissions");
174 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail,
175 CTLFLAG_RW, &VNET_NAME(tcp_sc_rst_sock_fail), 0,
176 "Send reset on socket allocation failure");
178 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
180 #define SYNCACHE_HASH(inc, mask) \
181 ((V_tcp_syncache.hash_secret ^ \
182 (inc)->inc_faddr.s_addr ^ \
183 ((inc)->inc_faddr.s_addr >> 16) ^ \
184 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
186 #define SYNCACHE_HASH6(inc, mask) \
187 ((V_tcp_syncache.hash_secret ^ \
188 (inc)->inc6_faddr.s6_addr32[0] ^ \
189 (inc)->inc6_faddr.s6_addr32[3] ^ \
190 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
192 #define ENDPTS_EQ(a, b) ( \
193 (a)->ie_fport == (b)->ie_fport && \
194 (a)->ie_lport == (b)->ie_lport && \
195 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
196 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
199 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
201 #define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx)
202 #define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx)
203 #define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED)
206 * Requires the syncache entry to be already removed from the bucket list.
209 syncache_free(struct syncache *sc)
213 (void) m_free(sc->sc_ipopts);
217 mac_syncache_destroy(&sc->sc_label);
220 uma_zfree(V_tcp_syncache.zone, sc);
228 V_tcp_syncookies = 1;
229 V_tcp_syncookiesonly = 0;
230 V_tcp_sc_rst_sock_fail = 1;
232 V_tcp_syncache.cache_count = 0;
233 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
234 V_tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
235 V_tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
236 V_tcp_syncache.hash_secret = arc4random();
238 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
239 &V_tcp_syncache.hashsize);
240 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
241 &V_tcp_syncache.bucket_limit);
242 if (!powerof2(V_tcp_syncache.hashsize) ||
243 V_tcp_syncache.hashsize == 0) {
244 printf("WARNING: syncache hash size is not a power of 2.\n");
245 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
247 V_tcp_syncache.hashmask = V_tcp_syncache.hashsize - 1;
250 V_tcp_syncache.cache_limit =
251 V_tcp_syncache.hashsize * V_tcp_syncache.bucket_limit;
252 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
253 &V_tcp_syncache.cache_limit);
255 /* Allocate the hash table. */
256 V_tcp_syncache.hashbase = malloc(V_tcp_syncache.hashsize *
257 sizeof(struct syncache_head), M_SYNCACHE, M_WAITOK | M_ZERO);
259 /* Initialize the hash buckets. */
260 for (i = 0; i < V_tcp_syncache.hashsize; i++) {
262 V_tcp_syncache.hashbase[i].sch_vnet = curvnet;
264 TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket);
265 mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
267 callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer,
268 &V_tcp_syncache.hashbase[i].sch_mtx, 0);
269 V_tcp_syncache.hashbase[i].sch_length = 0;
272 /* Create the syncache entry zone. */
273 V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
274 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
275 uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit);
280 syncache_destroy(void)
283 /* XXX walk the cache, free remaining objects, stop timers */
285 uma_zdestroy(V_tcp_syncache.zone);
286 FREE(V_tcp_syncache.hashbase, M_SYNCACHE);
291 * Inserts a syncache entry into the specified bucket row.
292 * Locks and unlocks the syncache_head autonomously.
295 syncache_insert(struct syncache *sc, struct syncache_head *sch)
297 struct syncache *sc2;
302 * Make sure that we don't overflow the per-bucket limit.
303 * If the bucket is full, toss the oldest element.
305 if (sch->sch_length >= V_tcp_syncache.bucket_limit) {
306 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
307 ("sch->sch_length incorrect"));
308 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
309 syncache_drop(sc2, sch);
310 TCPSTAT_INC(tcps_sc_bucketoverflow);
313 /* Put it into the bucket. */
314 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
317 /* Reinitialize the bucket row's timer. */
318 if (sch->sch_length == 1)
319 sch->sch_nextc = ticks + INT_MAX;
320 syncache_timeout(sc, sch, 1);
324 V_tcp_syncache.cache_count++;
325 TCPSTAT_INC(tcps_sc_added);
329 * Remove and free entry from syncache bucket row.
330 * Expects locked syncache head.
333 syncache_drop(struct syncache *sc, struct syncache_head *sch)
336 SCH_LOCK_ASSERT(sch);
338 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
341 #ifndef TCP_OFFLOAD_DISABLE
343 sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb);
346 V_tcp_syncache.cache_count--;
350 * Engage/reengage time on bucket row.
353 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
355 sc->sc_rxttime = ticks +
356 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
358 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
359 sch->sch_nextc = sc->sc_rxttime;
361 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
362 syncache_timer, (void *)sch);
367 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
368 * If we have retransmitted an entry the maximum number of times, expire it.
369 * One separate timer for each bucket row.
372 syncache_timer(void *xsch)
374 struct syncache_head *sch = (struct syncache_head *)xsch;
375 struct syncache *sc, *nsc;
379 CURVNET_SET(sch->sch_vnet);
381 /* NB: syncache_head has already been locked by the callout. */
382 SCH_LOCK_ASSERT(sch);
385 * In the following cycle we may remove some entries and/or
386 * advance some timeouts, so re-initialize the bucket timer.
388 sch->sch_nextc = tick + INT_MAX;
390 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
392 * We do not check if the listen socket still exists
393 * and accept the case where the listen socket may be
394 * gone by the time we resend the SYN/ACK. We do
395 * not expect this to happens often. If it does,
396 * then the RST will be sent by the time the remote
397 * host does the SYN/ACK->ACK.
399 if (TSTMP_GT(sc->sc_rxttime, tick)) {
400 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
401 sch->sch_nextc = sc->sc_rxttime;
404 if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) {
405 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
406 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
407 "giving up and removing syncache entry\n",
411 syncache_drop(sc, sch);
412 TCPSTAT_INC(tcps_sc_stale);
415 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
416 log(LOG_DEBUG, "%s; %s: Response timeout, "
417 "retransmitting (%u) SYN|ACK\n",
418 s, __func__, sc->sc_rxmits);
422 (void) syncache_respond(sc);
423 TCPSTAT_INC(tcps_sc_retransmitted);
424 syncache_timeout(sc, sch, 0);
426 if (!TAILQ_EMPTY(&(sch)->sch_bucket))
427 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
428 syncache_timer, (void *)(sch));
433 * Find an entry in the syncache.
434 * Returns always with locked syncache_head plus a matching entry or NULL.
437 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
440 struct syncache_head *sch;
443 if (inc->inc_flags & INC_ISIPV6) {
444 sch = &V_tcp_syncache.hashbase[
445 SYNCACHE_HASH6(inc, V_tcp_syncache.hashmask)];
450 /* Circle through bucket row to find matching entry. */
451 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
452 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
458 sch = &V_tcp_syncache.hashbase[
459 SYNCACHE_HASH(inc, V_tcp_syncache.hashmask)];
464 /* Circle through bucket row to find matching entry. */
465 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
467 if (sc->sc_inc.inc_flags & INC_ISIPV6)
470 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
474 SCH_LOCK_ASSERT(*schp);
475 return (NULL); /* always returns with locked sch */
479 * This function is called when we get a RST for a
480 * non-existent connection, so that we can see if the
481 * connection is in the syn cache. If it is, zap it.
484 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
487 struct syncache_head *sch;
490 sc = syncache_lookup(inc, &sch); /* returns locked sch */
491 SCH_LOCK_ASSERT(sch);
494 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
495 * See RFC 793 page 65, section SEGMENT ARRIVES.
497 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
498 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
499 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
500 "FIN flag set, segment ignored\n", s, __func__);
501 TCPSTAT_INC(tcps_badrst);
506 * No corresponding connection was found in syncache.
507 * If syncookies are enabled and possibly exclusively
508 * used, or we are under memory pressure, a valid RST
509 * may not find a syncache entry. In that case we're
510 * done and no SYN|ACK retransmissions will happen.
511 * Otherwise the the RST was misdirected or spoofed.
514 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
515 log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
516 "syncache entry (possibly syncookie only), "
517 "segment ignored\n", s, __func__);
518 TCPSTAT_INC(tcps_badrst);
523 * If the RST bit is set, check the sequence number to see
524 * if this is a valid reset segment.
526 * In all states except SYN-SENT, all reset (RST) segments
527 * are validated by checking their SEQ-fields. A reset is
528 * valid if its sequence number is in the window.
530 * The sequence number in the reset segment is normally an
531 * echo of our outgoing acknowlegement numbers, but some hosts
532 * send a reset with the sequence number at the rightmost edge
533 * of our receive window, and we have to handle this case.
535 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
536 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
537 syncache_drop(sc, sch);
538 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
539 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
540 "connection attempt aborted by remote endpoint\n",
542 TCPSTAT_INC(tcps_sc_reset);
544 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
545 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != "
546 "IRS %u (+WND %u), segment ignored\n",
547 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
548 TCPSTAT_INC(tcps_badrst);
558 syncache_badack(struct in_conninfo *inc)
561 struct syncache_head *sch;
563 sc = syncache_lookup(inc, &sch); /* returns locked sch */
564 SCH_LOCK_ASSERT(sch);
566 syncache_drop(sc, sch);
567 TCPSTAT_INC(tcps_sc_badack);
573 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
576 struct syncache_head *sch;
578 sc = syncache_lookup(inc, &sch); /* returns locked sch */
579 SCH_LOCK_ASSERT(sch);
583 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
584 if (ntohl(th->th_seq) != sc->sc_iss)
588 * If we've rertransmitted 3 times and this is our second error,
589 * we remove the entry. Otherwise, we allow it to continue on.
590 * This prevents us from incorrectly nuking an entry during a
591 * spurious network outage.
595 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
596 sc->sc_flags |= SCF_UNREACH;
599 syncache_drop(sc, sch);
600 TCPSTAT_INC(tcps_sc_unreach);
606 * Build a new TCP socket structure from a syncache entry.
608 static struct socket *
609 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
611 struct inpcb *inp = NULL;
616 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
619 * Ok, create the full blown connection, and set things up
620 * as they would have been set up if we had created the
621 * connection when the SYN arrived. If we can't create
622 * the connection, abort it.
624 so = sonewconn(lso, SS_ISCONNECTED);
627 * Drop the connection; we will either send a RST or
628 * have the peer retransmit its SYN again after its
631 TCPSTAT_INC(tcps_listendrop);
632 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
633 log(LOG_DEBUG, "%s; %s: Socket create failed "
634 "due to limits or memory shortage\n",
641 mac_socketpeer_set_from_mbuf(m, so);
645 inp->inp_inc.inc_fibnum = sc->sc_inc.inc_fibnum;
646 so->so_fibnum = sc->sc_inc.inc_fibnum;
649 /* Insert new socket into PCB hash list. */
650 inp->inp_inc.inc_flags = sc->sc_inc.inc_flags;
652 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
653 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
655 inp->inp_vflag &= ~INP_IPV6;
656 inp->inp_vflag |= INP_IPV4;
658 inp->inp_laddr = sc->sc_inc.inc_laddr;
662 inp->inp_lport = sc->sc_inc.inc_lport;
663 if (in_pcbinshash(inp) != 0) {
665 * Undo the assignments above if we failed to
666 * put the PCB on the hash lists.
669 if (sc->sc_inc.inc_flags & INC_ISIPV6)
670 inp->in6p_laddr = in6addr_any;
673 inp->inp_laddr.s_addr = INADDR_ANY;
678 /* Copy old policy into new socket's. */
679 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
680 printf("syncache_socket: could not copy policy\n");
683 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
684 struct inpcb *oinp = sotoinpcb(lso);
685 struct in6_addr laddr6;
686 struct sockaddr_in6 sin6;
688 * Inherit socket options from the listening socket.
689 * Note that in6p_inputopts are not (and should not be)
690 * copied, since it stores previously received options and is
691 * used to detect if each new option is different than the
692 * previous one and hence should be passed to a user.
693 * If we copied in6p_inputopts, a user would not be able to
694 * receive options just after calling the accept system call.
696 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
697 if (oinp->in6p_outputopts)
698 inp->in6p_outputopts =
699 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
701 sin6.sin6_family = AF_INET6;
702 sin6.sin6_len = sizeof(sin6);
703 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
704 sin6.sin6_port = sc->sc_inc.inc_fport;
705 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
706 laddr6 = inp->in6p_laddr;
707 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
708 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
709 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
711 inp->in6p_laddr = laddr6;
714 /* Override flowlabel from in6_pcbconnect. */
715 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
716 inp->inp_flow |= sc->sc_flowlabel;
720 struct in_addr laddr;
721 struct sockaddr_in sin;
723 inp->inp_options = (m) ? ip_srcroute(m) : NULL;
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;
781 if (sc->sc_flags & SCF_ECN)
782 tp->t_flags |= TF_ECN_PERMIT;
785 * Set up MSS and get cached values from tcp_hostcache.
786 * This might overwrite some of the defaults we just set.
788 tcp_mss(tp, sc->sc_peer_mss);
791 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
794 tp->snd_cwnd = tp->t_maxseg;
795 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
799 TCPSTAT_INC(tcps_accepts);
811 * This function gets called when we receive an ACK for a
812 * socket in the LISTEN state. We look up the connection
813 * in the syncache, and if its there, we pull it out of
814 * the cache and turn it into a full-blown connection in
815 * the SYN-RECEIVED state.
818 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
819 struct socket **lsop, struct mbuf *m)
822 struct syncache_head *sch;
827 * Global TCP locks are held because we manipulate the PCB lists
828 * and create a new socket.
830 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
831 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
832 ("%s: can handle only ACK", __func__));
834 sc = syncache_lookup(inc, &sch); /* returns locked sch */
835 SCH_LOCK_ASSERT(sch);
838 * There is no syncache entry, so see if this ACK is
839 * a returning syncookie. To do this, first:
840 * A. See if this socket has had a syncache entry dropped in
841 * the past. We don't want to accept a bogus syncookie
842 * if we've never received a SYN.
843 * B. check that the syncookie is valid. If it is, then
844 * cobble up a fake syncache entry, and return.
846 if (!V_tcp_syncookies) {
848 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
849 log(LOG_DEBUG, "%s; %s: Spurious ACK, "
850 "segment rejected (syncookies disabled)\n",
854 bzero(&scs, sizeof(scs));
855 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
858 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
859 log(LOG_DEBUG, "%s; %s: Segment failed "
860 "SYNCOOKIE authentication, segment rejected "
861 "(probably spoofed)\n", s, __func__);
865 /* Pull out the entry to unlock the bucket row. */
866 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
868 V_tcp_syncache.cache_count--;
873 * Segment validation:
874 * ACK must match our initial sequence number + 1 (the SYN|ACK).
876 if (th->th_ack != sc->sc_iss + 1 && !TOEPCB_ISSET(sc)) {
877 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
878 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment "
879 "rejected\n", s, __func__, th->th_ack, sc->sc_iss);
884 * The SEQ must fall in the window starting at the received
885 * initial receive sequence number + 1 (the SYN).
887 if ((SEQ_LEQ(th->th_seq, sc->sc_irs) ||
888 SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) &&
890 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
891 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment "
892 "rejected\n", s, __func__, th->th_seq, sc->sc_irs);
896 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
897 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
898 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
899 "segment rejected\n", s, __func__);
903 * If timestamps were negotiated the reflected timestamp
904 * must be equal to what we actually sent in the SYN|ACK.
906 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts &&
908 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
909 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
910 "segment rejected\n",
911 s, __func__, to->to_tsecr, sc->sc_ts);
915 *lsop = syncache_socket(sc, *lsop, m);
918 TCPSTAT_INC(tcps_sc_aborted);
920 TCPSTAT_INC(tcps_sc_completed);
922 /* how do we find the inp for the new socket? */
927 if (sc != NULL && sc != &scs)
936 tcp_offload_syncache_expand(struct in_conninfo *inc, struct toeopt *toeo,
937 struct tcphdr *th, struct socket **lsop, struct mbuf *m)
942 bzero(&to, sizeof(struct tcpopt));
943 to.to_mss = toeo->to_mss;
944 to.to_wscale = toeo->to_wscale;
945 to.to_flags = toeo->to_flags;
947 INP_INFO_WLOCK(&V_tcbinfo);
948 rc = syncache_expand(inc, &to, th, lsop, m);
949 INP_INFO_WUNLOCK(&V_tcbinfo);
955 * Given a LISTEN socket and an inbound SYN request, add
956 * this to the syn cache, and send back a segment:
957 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
960 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
961 * Doing so would require that we hold onto the data and deliver it
962 * to the application. However, if we are the target of a SYN-flood
963 * DoS attack, an attacker could send data which would eventually
964 * consume all available buffer space if it were ACKed. By not ACKing
965 * the data, we avoid this DoS scenario.
968 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
969 struct inpcb *inp, struct socket **lsop, struct mbuf *m,
970 struct toe_usrreqs *tu, void *toepcb)
974 struct syncache *sc = NULL;
975 struct syncache_head *sch;
976 struct mbuf *ipopts = NULL;
978 int win, sb_hiwat, ip_ttl, ip_tos, noopt;
981 int autoflowlabel = 0;
984 struct label *maclabel;
989 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
990 INP_WLOCK_ASSERT(inp); /* listen socket */
991 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
992 ("%s: unexpected tcp flags", __func__));
995 * Combine all so/tp operations very early to drop the INP lock as
1000 cred = crhold(so->so_cred);
1003 if ((inc->inc_flags & INC_ISIPV6) &&
1004 (inp->inp_flags & IN6P_AUTOFLOWLABEL))
1007 ip_ttl = inp->inp_ip_ttl;
1008 ip_tos = inp->inp_ip_tos;
1009 win = sbspace(&so->so_rcv);
1010 sb_hiwat = so->so_rcv.sb_hiwat;
1011 noopt = (tp->t_flags & TF_NOOPT);
1013 /* By the time we drop the lock these should no longer be used. */
1018 if (mac_syncache_init(&maclabel) != 0) {
1020 INP_INFO_WUNLOCK(&V_tcbinfo);
1023 mac_syncache_create(maclabel, inp);
1026 INP_INFO_WUNLOCK(&V_tcbinfo);
1029 * Remember the IP options, if any.
1032 if (!(inc->inc_flags & INC_ISIPV6))
1034 ipopts = (m) ? ip_srcroute(m) : NULL;
1037 * See if we already have an entry for this connection.
1038 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1040 * XXX: should the syncache be re-initialized with the contents
1041 * of the new SYN here (which may have different options?)
1043 * XXX: We do not check the sequence number to see if this is a
1044 * real retransmit or a new connection attempt. The question is
1045 * how to handle such a case; either ignore it as spoofed, or
1046 * drop the current entry and create a new one?
1048 sc = syncache_lookup(inc, &sch); /* returns locked entry */
1049 SCH_LOCK_ASSERT(sch);
1051 #ifndef TCP_OFFLOAD_DISABLE
1053 sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT,
1056 TCPSTAT_INC(tcps_sc_dupsyn);
1059 * If we were remembering a previous source route,
1060 * forget it and use the new one we've been given.
1063 (void) m_free(sc->sc_ipopts);
1064 sc->sc_ipopts = ipopts;
1067 * Update timestamp if present.
1069 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
1070 sc->sc_tsreflect = to->to_tsval;
1072 sc->sc_flags &= ~SCF_TIMESTAMP;
1075 * Since we have already unconditionally allocated label
1076 * storage, free it up. The syncache entry will already
1077 * have an initialized label we can use.
1079 mac_syncache_destroy(&maclabel);
1081 /* Retransmit SYN|ACK and reset retransmit count. */
1082 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
1083 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
1084 "resetting timer and retransmitting SYN|ACK\n",
1088 if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) {
1090 syncache_timeout(sc, sch, 1);
1091 TCPSTAT_INC(tcps_sndacks);
1092 TCPSTAT_INC(tcps_sndtotal);
1098 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1101 * The zone allocator couldn't provide more entries.
1102 * Treat this as if the cache was full; drop the oldest
1103 * entry and insert the new one.
1105 TCPSTAT_INC(tcps_sc_zonefail);
1106 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1107 syncache_drop(sc, sch);
1108 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1110 if (V_tcp_syncookies) {
1111 bzero(&scs, sizeof(scs));
1116 (void) m_free(ipopts);
1123 * Fill in the syncache values.
1126 sc->sc_label = maclabel;
1130 sc->sc_ipopts = ipopts;
1131 /* XXX-BZ this fib assignment is just useless. */
1132 sc->sc_inc.inc_fibnum = inp->inp_inc.inc_fibnum;
1133 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1135 if (!(inc->inc_flags & INC_ISIPV6))
1138 sc->sc_ip_tos = ip_tos;
1139 sc->sc_ip_ttl = ip_ttl;
1141 #ifndef TCP_OFFLOAD_DISABLE
1143 sc->sc_toepcb = toepcb;
1145 sc->sc_irs = th->th_seq;
1146 sc->sc_iss = arc4random();
1148 sc->sc_flowlabel = 0;
1151 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
1152 * win was derived from socket earlier in the function.
1155 win = imin(win, TCP_MAXWIN);
1158 if (V_tcp_do_rfc1323) {
1160 * A timestamp received in a SYN makes
1161 * it ok to send timestamp requests and replies.
1163 if (to->to_flags & TOF_TS) {
1164 sc->sc_tsreflect = to->to_tsval;
1166 sc->sc_flags |= SCF_TIMESTAMP;
1168 if (to->to_flags & TOF_SCALE) {
1172 * Pick the smallest possible scaling factor that
1173 * will still allow us to scale up to sb_max, aka
1174 * kern.ipc.maxsockbuf.
1176 * We do this because there are broken firewalls that
1177 * will corrupt the window scale option, leading to
1178 * the other endpoint believing that our advertised
1179 * window is unscaled. At scale factors larger than
1180 * 5 the unscaled window will drop below 1500 bytes,
1181 * leading to serious problems when traversing these
1184 * With the default maxsockbuf of 256K, a scale factor
1185 * of 3 will be chosen by this algorithm. Those who
1186 * choose a larger maxsockbuf should watch out
1187 * for the compatiblity problems mentioned above.
1189 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1190 * or <SYN,ACK>) segment itself is never scaled.
1192 while (wscale < TCP_MAX_WINSHIFT &&
1193 (TCP_MAXWIN << wscale) < sb_max)
1195 sc->sc_requested_r_scale = wscale;
1196 sc->sc_requested_s_scale = to->to_wscale;
1197 sc->sc_flags |= SCF_WINSCALE;
1200 #ifdef TCP_SIGNATURE
1202 * If listening socket requested TCP digests, and received SYN
1203 * contains the option, flag this in the syncache so that
1204 * syncache_respond() will do the right thing with the SYN+ACK.
1205 * XXX: Currently we always record the option by default and will
1206 * attempt to use it in syncache_respond().
1208 if (to->to_flags & TOF_SIGNATURE)
1209 sc->sc_flags |= SCF_SIGNATURE;
1211 if (to->to_flags & TOF_SACKPERM)
1212 sc->sc_flags |= SCF_SACK;
1213 if (to->to_flags & TOF_MSS)
1214 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */
1216 sc->sc_flags |= SCF_NOOPT;
1217 if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn)
1218 sc->sc_flags |= SCF_ECN;
1220 if (V_tcp_syncookies) {
1221 syncookie_generate(sch, sc, &flowtmp);
1224 sc->sc_flowlabel = flowtmp;
1230 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1236 * Do a standard 3-way handshake.
1238 if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) {
1239 if (V_tcp_syncookies && V_tcp_syncookiesonly && sc != &scs)
1241 else if (sc != &scs)
1242 syncache_insert(sc, sch); /* locks and unlocks sch */
1243 TCPSTAT_INC(tcps_sndacks);
1244 TCPSTAT_INC(tcps_sndtotal);
1248 TCPSTAT_INC(tcps_sc_dropped);
1256 mac_syncache_destroy(&maclabel);
1266 syncache_respond(struct syncache *sc)
1268 struct ip *ip = NULL;
1272 u_int16_t hlen, tlen, mssopt;
1275 struct ip6_hdr *ip6 = NULL;
1280 (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) :
1283 tlen = hlen + sizeof(struct tcphdr);
1285 /* Determine MSS we advertize to other end of connection. */
1286 mssopt = tcp_mssopt(&sc->sc_inc);
1287 if (sc->sc_peer_mss)
1288 mssopt = max( min(sc->sc_peer_mss, mssopt), V_tcp_minmss);
1290 /* XXX: Assume that the entire packet will fit in a header mbuf. */
1291 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1292 ("syncache: mbuf too small"));
1294 /* Create the IP+TCP header from scratch. */
1295 m = m_gethdr(M_DONTWAIT, MT_DATA);
1299 mac_syncache_create_mbuf(sc->sc_label, m);
1301 m->m_data += max_linkhdr;
1303 m->m_pkthdr.len = tlen;
1304 m->m_pkthdr.rcvif = NULL;
1307 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1308 ip6 = mtod(m, struct ip6_hdr *);
1309 ip6->ip6_vfc = IPV6_VERSION;
1310 ip6->ip6_nxt = IPPROTO_TCP;
1311 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1312 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1313 ip6->ip6_plen = htons(tlen - hlen);
1314 /* ip6_hlim is set after checksum */
1315 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1316 ip6->ip6_flow |= sc->sc_flowlabel;
1318 th = (struct tcphdr *)(ip6 + 1);
1322 ip = mtod(m, struct ip *);
1323 ip->ip_v = IPVERSION;
1324 ip->ip_hl = sizeof(struct ip) >> 2;
1329 ip->ip_p = IPPROTO_TCP;
1330 ip->ip_src = sc->sc_inc.inc_laddr;
1331 ip->ip_dst = sc->sc_inc.inc_faddr;
1332 ip->ip_ttl = sc->sc_ip_ttl;
1333 ip->ip_tos = sc->sc_ip_tos;
1336 * See if we should do MTU discovery. Route lookups are
1337 * expensive, so we will only unset the DF bit if:
1339 * 1) path_mtu_discovery is disabled
1340 * 2) the SCF_UNREACH flag has been set
1342 if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1343 ip->ip_off |= IP_DF;
1345 th = (struct tcphdr *)(ip + 1);
1347 th->th_sport = sc->sc_inc.inc_lport;
1348 th->th_dport = sc->sc_inc.inc_fport;
1350 th->th_seq = htonl(sc->sc_iss);
1351 th->th_ack = htonl(sc->sc_irs + 1);
1352 th->th_off = sizeof(struct tcphdr) >> 2;
1354 th->th_flags = TH_SYN|TH_ACK;
1355 th->th_win = htons(sc->sc_wnd);
1358 if (sc->sc_flags & SCF_ECN) {
1359 th->th_flags |= TH_ECE;
1360 TCPSTAT_INC(tcps_ecn_shs);
1363 /* Tack on the TCP options. */
1364 if ((sc->sc_flags & SCF_NOOPT) == 0) {
1368 to.to_flags = TOF_MSS;
1369 if (sc->sc_flags & SCF_WINSCALE) {
1370 to.to_wscale = sc->sc_requested_r_scale;
1371 to.to_flags |= TOF_SCALE;
1373 if (sc->sc_flags & SCF_TIMESTAMP) {
1374 /* Virgin timestamp or TCP cookie enhanced one. */
1375 to.to_tsval = sc->sc_ts;
1376 to.to_tsecr = sc->sc_tsreflect;
1377 to.to_flags |= TOF_TS;
1379 if (sc->sc_flags & SCF_SACK)
1380 to.to_flags |= TOF_SACKPERM;
1381 #ifdef TCP_SIGNATURE
1382 if (sc->sc_flags & SCF_SIGNATURE)
1383 to.to_flags |= TOF_SIGNATURE;
1385 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1387 /* Adjust headers by option size. */
1388 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1390 m->m_pkthdr.len += optlen;
1392 #ifdef TCP_SIGNATURE
1393 if (sc->sc_flags & SCF_SIGNATURE)
1394 tcp_signature_compute(m, 0, 0, optlen,
1395 to.to_signature, IPSEC_DIR_OUTBOUND);
1398 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1399 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1402 ip->ip_len += optlen;
1407 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1409 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
1410 tlen + optlen - hlen);
1411 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1412 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1416 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1417 htons(tlen + optlen - hlen + IPPROTO_TCP));
1418 m->m_pkthdr.csum_flags = CSUM_TCP;
1419 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1420 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1426 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1427 struct inpcb *inp, struct socket **lsop, struct mbuf *m)
1429 _syncache_add(inc, to, th, inp, lsop, m, NULL, NULL);
1433 tcp_offload_syncache_add(struct in_conninfo *inc, struct toeopt *toeo,
1434 struct tcphdr *th, struct inpcb *inp, struct socket **lsop,
1435 struct toe_usrreqs *tu, void *toepcb)
1439 bzero(&to, sizeof(struct tcpopt));
1440 to.to_mss = toeo->to_mss;
1441 to.to_wscale = toeo->to_wscale;
1442 to.to_flags = toeo->to_flags;
1444 INP_INFO_WLOCK(&V_tcbinfo);
1447 _syncache_add(inc, &to, th, inp, lsop, NULL, tu, toepcb);
1451 * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1452 * receive and to be able to handle SYN floods from bogus source addresses
1453 * (where we will never receive any reply). SYN floods try to exhaust all
1454 * our memory and available slots in the SYN cache table to cause a denial
1455 * of service to legitimate users of the local host.
1457 * The idea of SYN cookies is to encode and include all necessary information
1458 * about the connection setup state within the SYN-ACK we send back and thus
1459 * to get along without keeping any local state until the ACK to the SYN-ACK
1460 * arrives (if ever). Everything we need to know should be available from
1461 * the information we encoded in the SYN-ACK.
1463 * More information about the theory behind SYN cookies and its first
1464 * discussion and specification can be found at:
1465 * http://cr.yp.to/syncookies.html (overview)
1466 * http://cr.yp.to/syncookies/archive (gory details)
1468 * This implementation extends the orginal idea and first implementation
1469 * of FreeBSD by using not only the initial sequence number field to store
1470 * information but also the timestamp field if present. This way we can
1471 * keep track of the entire state we need to know to recreate the session in
1472 * its original form. Almost all TCP speakers implement RFC1323 timestamps
1473 * these days. For those that do not we still have to live with the known
1474 * shortcomings of the ISN only SYN cookies.
1478 * Initial sequence number we send:
1479 * 31|................................|0
1480 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1481 * D = MD5 Digest (first dword)
1483 * R = Rotation of secret
1484 * P = Odd or Even secret
1486 * The MD5 Digest is computed with over following parameters:
1487 * a) randomly rotated secret
1488 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1489 * c) the received initial sequence number from remote host
1490 * d) the rotation offset and odd/even bit
1492 * Timestamp we send:
1493 * 31|................................|0
1494 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1495 * D = MD5 Digest (third dword) (only as filler)
1496 * S = Requested send window scale
1497 * R = Requested receive window scale
1499 * 5 = TCP-MD5 enabled (not implemented yet)
1500 * XORed with MD5 Digest (forth dword)
1502 * The timestamp isn't cryptographically secure and doesn't need to be.
1503 * The double use of the MD5 digest dwords ties it to a specific remote/
1504 * local host/port, remote initial sequence number and our local time
1505 * limited secret. A received timestamp is reverted (XORed) and then
1506 * the contained MD5 dword is compared to the computed one to ensure the
1507 * timestamp belongs to the SYN-ACK we sent. The other parameters may
1508 * have been tampered with but this isn't different from supplying bogus
1509 * values in the SYN in the first place.
1511 * Some problems with SYN cookies remain however:
1512 * Consider the problem of a recreated (and retransmitted) cookie. If the
1513 * original SYN was accepted, the connection is established. The second
1514 * SYN is inflight, and if it arrives with an ISN that falls within the
1515 * receive window, the connection is killed.
1518 * A heuristic to determine when to accept syn cookies is not necessary.
1519 * An ACK flood would cause the syncookie verification to be attempted,
1520 * but a SYN flood causes syncookies to be generated. Both are of equal
1521 * cost, so there's no point in trying to optimize the ACK flood case.
1522 * Also, if you don't process certain ACKs for some reason, then all someone
1523 * would have to do is launch a SYN and ACK flood at the same time, which
1524 * would stop cookie verification and defeat the entire purpose of syncookies.
1526 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1529 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1530 u_int32_t *flowlabel)
1533 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1536 u_int off, pmss, mss;
1539 SCH_LOCK_ASSERT(sch);
1541 /* Which of the two secrets to use. */
1542 secbits = sch->sch_oddeven ?
1543 sch->sch_secbits_odd : sch->sch_secbits_even;
1545 /* Reseed secret if too old. */
1546 if (sch->sch_reseed < time_uptime) {
1547 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */
1548 secbits = sch->sch_oddeven ?
1549 sch->sch_secbits_odd : sch->sch_secbits_even;
1550 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1551 secbits[i] = arc4random();
1552 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1555 /* Secret rotation offset. */
1556 off = sc->sc_iss & 0x7; /* iss was randomized before */
1558 /* Maximum segment size calculation. */
1560 max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), V_tcp_minmss);
1561 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1562 if (tcp_sc_msstab[mss] <= pmss)
1565 /* Fold parameters and MD5 digest into the ISN we will send. */
1566 data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1567 data |= off << 1; /* secret offset, derived from iss, 3 bits */
1568 data |= mss << 4; /* mss, 3 bits */
1571 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1572 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1573 MD5Update(&ctx, secbits, off);
1574 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1575 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1576 MD5Update(&ctx, &data, sizeof(data));
1577 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1579 data |= (md5_buffer[0] << 7);
1583 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1586 /* Additional parameters are stored in the timestamp if present. */
1587 if (sc->sc_flags & SCF_TIMESTAMP) {
1588 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1589 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1590 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */
1591 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */
1592 data |= md5_buffer[2] << 10; /* more digest bits */
1593 data ^= md5_buffer[3];
1595 sc->sc_tsoff = data - ticks; /* after XOR */
1598 TCPSTAT_INC(tcps_sc_sendcookie);
1601 static struct syncache *
1602 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1603 struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1607 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1611 int off, mss, wnd, flags;
1613 SCH_LOCK_ASSERT(sch);
1616 * Pull information out of SYN-ACK/ACK and
1617 * revert sequence number advances.
1619 ack = th->th_ack - 1;
1620 seq = th->th_seq - 1;
1621 off = (ack >> 1) & 0x7;
1622 mss = (ack >> 4) & 0x7;
1625 /* Which of the two secrets to use. */
1626 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1629 * The secret wasn't updated for the lifetime of a syncookie,
1630 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1632 if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) {
1636 /* Recompute the digest so we can compare it. */
1638 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1639 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1640 MD5Update(&ctx, secbits, off);
1641 MD5Update(&ctx, inc, sizeof(*inc));
1642 MD5Update(&ctx, &seq, sizeof(seq));
1643 MD5Update(&ctx, &flags, sizeof(flags));
1644 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1646 /* Does the digest part of or ACK'ed ISS match? */
1647 if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1650 /* Does the digest part of our reflected timestamp match? */
1651 if (to->to_flags & TOF_TS) {
1652 data = md5_buffer[3] ^ to->to_tsecr;
1653 if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1657 /* Fill in the syncache values. */
1658 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1659 sc->sc_ipopts = NULL;
1665 if (inc->inc_flags & INC_ISIPV6) {
1666 if (sotoinpcb(so)->inp_flags & IN6P_AUTOFLOWLABEL)
1667 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1671 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1672 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1675 /* Additional parameters that were encoded in the timestamp. */
1677 sc->sc_flags |= SCF_TIMESTAMP;
1678 sc->sc_tsreflect = to->to_tsval;
1679 sc->sc_ts = to->to_tsecr;
1680 sc->sc_tsoff = to->to_tsecr - ticks;
1681 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1682 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1683 sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1685 sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1687 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1688 sc->sc_flags |= SCF_WINSCALE;
1690 sc->sc_flags |= SCF_NOOPT;
1692 wnd = sbspace(&so->so_rcv);
1694 wnd = imin(wnd, TCP_MAXWIN);
1698 sc->sc_peer_mss = tcp_sc_msstab[mss];
1700 TCPSTAT_INC(tcps_sc_recvcookie);
1705 * Returns the current number of syncache entries. This number
1706 * will probably change before you get around to calling
1711 syncache_pcbcount(void)
1713 struct syncache_head *sch;
1716 for (count = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1717 /* No need to lock for a read. */
1718 sch = &V_tcp_syncache.hashbase[i];
1719 count += sch->sch_length;
1725 * Exports the syncache entries to userland so that netstat can display
1726 * them alongside the other sockets. This function is intended to be
1727 * called only from tcp_pcblist.
1729 * Due to concurrency on an active system, the number of pcbs exported
1730 * may have no relation to max_pcbs. max_pcbs merely indicates the
1731 * amount of space the caller allocated for this function to use.
1734 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
1737 struct syncache *sc;
1738 struct syncache_head *sch;
1739 int count, error, i;
1741 for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1742 sch = &V_tcp_syncache.hashbase[i];
1744 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
1745 if (count >= max_pcbs) {
1749 if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0)
1751 bzero(&xt, sizeof(xt));
1752 xt.xt_len = sizeof(xt);
1753 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1754 xt.xt_inp.inp_vflag = INP_IPV6;
1756 xt.xt_inp.inp_vflag = INP_IPV4;
1757 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
1758 xt.xt_tp.t_inpcb = &xt.xt_inp;
1759 xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
1760 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1761 xt.xt_socket.xso_len = sizeof (struct xsocket);
1762 xt.xt_socket.so_type = SOCK_STREAM;
1763 xt.xt_socket.so_state = SS_ISCONNECTING;
1764 error = SYSCTL_OUT(req, &xt, sizeof xt);
1774 *pcbs_exported = count;