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"
39 #include "opt_pcbgroup.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>
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>
85 #include <netinet6/tcp6_var.h>
88 #include <netinet/toecore.h>
92 #include <netipsec/ipsec.h>
94 #include <netipsec/ipsec6.h>
96 #include <netipsec/key.h>
99 #include <machine/in_cksum.h>
101 #include <security/mac/mac_framework.h>
103 static VNET_DEFINE(int, tcp_syncookies) = 1;
104 #define V_tcp_syncookies VNET(tcp_syncookies)
105 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
106 &VNET_NAME(tcp_syncookies), 0,
107 "Use TCP SYN cookies if the syncache overflows");
109 static VNET_DEFINE(int, tcp_syncookiesonly) = 0;
110 #define V_tcp_syncookiesonly VNET(tcp_syncookiesonly)
111 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW,
112 &VNET_NAME(tcp_syncookiesonly), 0,
113 "Use only TCP SYN cookies");
116 #define ADDED_BY_TOE(sc) ((sc)->sc_tod != NULL)
119 static void syncache_drop(struct syncache *, struct syncache_head *);
120 static void syncache_free(struct syncache *);
121 static void syncache_insert(struct syncache *, struct syncache_head *);
122 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
123 static int syncache_respond(struct syncache *);
124 static struct socket *syncache_socket(struct syncache *, struct socket *,
126 static int syncache_sysctl_count(SYSCTL_HANDLER_ARGS);
127 static void syncache_timeout(struct syncache *sc, struct syncache_head *sch,
129 static void syncache_timer(void *);
130 static void syncookie_generate(struct syncache_head *, struct syncache *,
132 static struct syncache
133 *syncookie_lookup(struct in_conninfo *, struct syncache_head *,
134 struct syncache *, struct tcpopt *, struct tcphdr *,
138 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
139 * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds,
140 * the odds are that the user has given up attempting to connect by then.
142 #define SYNCACHE_MAXREXMTS 3
144 /* Arbitrary values */
145 #define TCP_SYNCACHE_HASHSIZE 512
146 #define TCP_SYNCACHE_BUCKETLIMIT 30
148 static VNET_DEFINE(struct tcp_syncache, tcp_syncache);
149 #define V_tcp_syncache VNET(tcp_syncache)
151 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0,
154 SYSCTL_VNET_UINT(_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_UINT(_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_PROC(_net_inet_tcp_syncache, OID_AUTO, count, (CTLTYPE_UINT|CTLFLAG_RD),
163 NULL, 0, &syncache_sysctl_count, "IU",
164 "Current number of entries in syncache");
166 SYSCTL_VNET_UINT(_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_UINT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
171 &VNET_NAME(tcp_syncache.rexmt_limit), 0,
172 "Limit on SYN/ACK retransmissions");
174 VNET_DEFINE(int, tcp_sc_rst_sock_fail) = 1;
175 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail,
176 CTLFLAG_RW, &VNET_NAME(tcp_sc_rst_sock_fail), 0,
177 "Send reset on socket allocation failure");
179 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
181 #define SYNCACHE_HASH(inc, mask) \
182 ((V_tcp_syncache.hash_secret ^ \
183 (inc)->inc_faddr.s_addr ^ \
184 ((inc)->inc_faddr.s_addr >> 16) ^ \
185 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
187 #define SYNCACHE_HASH6(inc, mask) \
188 ((V_tcp_syncache.hash_secret ^ \
189 (inc)->inc6_faddr.s6_addr32[0] ^ \
190 (inc)->inc6_faddr.s6_addr32[3] ^ \
191 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
193 #define ENDPTS_EQ(a, b) ( \
194 (a)->ie_fport == (b)->ie_fport && \
195 (a)->ie_lport == (b)->ie_lport && \
196 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
197 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
200 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
202 #define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx)
203 #define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx)
204 #define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED)
207 * Requires the syncache entry to be already removed from the bucket list.
210 syncache_free(struct syncache *sc)
214 (void) m_free(sc->sc_ipopts);
218 mac_syncache_destroy(&sc->sc_label);
221 uma_zfree(V_tcp_syncache.zone, sc);
229 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
230 V_tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
231 V_tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
232 V_tcp_syncache.hash_secret = arc4random();
234 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
235 &V_tcp_syncache.hashsize);
236 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
237 &V_tcp_syncache.bucket_limit);
238 if (!powerof2(V_tcp_syncache.hashsize) ||
239 V_tcp_syncache.hashsize == 0) {
240 printf("WARNING: syncache hash size is not a power of 2.\n");
241 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
243 V_tcp_syncache.hashmask = V_tcp_syncache.hashsize - 1;
246 V_tcp_syncache.cache_limit =
247 V_tcp_syncache.hashsize * V_tcp_syncache.bucket_limit;
248 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
249 &V_tcp_syncache.cache_limit);
251 /* Allocate the hash table. */
252 V_tcp_syncache.hashbase = malloc(V_tcp_syncache.hashsize *
253 sizeof(struct syncache_head), M_SYNCACHE, M_WAITOK | M_ZERO);
255 /* Initialize the hash buckets. */
256 for (i = 0; i < V_tcp_syncache.hashsize; i++) {
258 V_tcp_syncache.hashbase[i].sch_vnet = curvnet;
260 TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket);
261 mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
263 callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer,
264 &V_tcp_syncache.hashbase[i].sch_mtx, 0);
265 V_tcp_syncache.hashbase[i].sch_length = 0;
268 /* Create the syncache entry zone. */
269 V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
270 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
271 V_tcp_syncache.cache_limit = uma_zone_set_max(V_tcp_syncache.zone,
272 V_tcp_syncache.cache_limit);
277 syncache_destroy(void)
279 struct syncache_head *sch;
280 struct syncache *sc, *nsc;
283 /* Cleanup hash buckets: stop timers, free entries, destroy locks. */
284 for (i = 0; i < V_tcp_syncache.hashsize; i++) {
286 sch = &V_tcp_syncache.hashbase[i];
287 callout_drain(&sch->sch_timer);
290 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc)
291 syncache_drop(sc, sch);
293 KASSERT(TAILQ_EMPTY(&sch->sch_bucket),
294 ("%s: sch->sch_bucket not empty", __func__));
295 KASSERT(sch->sch_length == 0, ("%s: sch->sch_length %d not 0",
296 __func__, sch->sch_length));
297 mtx_destroy(&sch->sch_mtx);
300 KASSERT(uma_zone_get_cur(V_tcp_syncache.zone) == 0,
301 ("%s: cache_count not 0", __func__));
303 /* Free the allocated global resources. */
304 uma_zdestroy(V_tcp_syncache.zone);
305 free(V_tcp_syncache.hashbase, M_SYNCACHE);
310 syncache_sysctl_count(SYSCTL_HANDLER_ARGS)
314 count = uma_zone_get_cur(V_tcp_syncache.zone);
315 return (sysctl_handle_int(oidp, &count, 0, req));
319 * Inserts a syncache entry into the specified bucket row.
320 * Locks and unlocks the syncache_head autonomously.
323 syncache_insert(struct syncache *sc, struct syncache_head *sch)
325 struct syncache *sc2;
330 * Make sure that we don't overflow the per-bucket limit.
331 * If the bucket is full, toss the oldest element.
333 if (sch->sch_length >= V_tcp_syncache.bucket_limit) {
334 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
335 ("sch->sch_length incorrect"));
336 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
337 syncache_drop(sc2, sch);
338 TCPSTAT_INC(tcps_sc_bucketoverflow);
341 /* Put it into the bucket. */
342 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
346 if (ADDED_BY_TOE(sc)) {
347 struct toedev *tod = sc->sc_tod;
349 tod->tod_syncache_added(tod, sc->sc_todctx);
353 /* Reinitialize the bucket row's timer. */
354 if (sch->sch_length == 1)
355 sch->sch_nextc = ticks + INT_MAX;
356 syncache_timeout(sc, sch, 1);
360 TCPSTAT_INC(tcps_sc_added);
364 * Remove and free entry from syncache bucket row.
365 * Expects locked syncache head.
368 syncache_drop(struct syncache *sc, struct syncache_head *sch)
371 SCH_LOCK_ASSERT(sch);
373 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
377 if (ADDED_BY_TOE(sc)) {
378 struct toedev *tod = sc->sc_tod;
380 tod->tod_syncache_removed(tod, sc->sc_todctx);
388 * Engage/reengage time on bucket row.
391 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
393 sc->sc_rxttime = ticks +
394 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
396 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
397 sch->sch_nextc = sc->sc_rxttime;
399 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
400 syncache_timer, (void *)sch);
405 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
406 * If we have retransmitted an entry the maximum number of times, expire it.
407 * One separate timer for each bucket row.
410 syncache_timer(void *xsch)
412 struct syncache_head *sch = (struct syncache_head *)xsch;
413 struct syncache *sc, *nsc;
417 CURVNET_SET(sch->sch_vnet);
419 /* NB: syncache_head has already been locked by the callout. */
420 SCH_LOCK_ASSERT(sch);
423 * In the following cycle we may remove some entries and/or
424 * advance some timeouts, so re-initialize the bucket timer.
426 sch->sch_nextc = tick + INT_MAX;
428 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
430 * We do not check if the listen socket still exists
431 * and accept the case where the listen socket may be
432 * gone by the time we resend the SYN/ACK. We do
433 * not expect this to happens often. If it does,
434 * then the RST will be sent by the time the remote
435 * host does the SYN/ACK->ACK.
437 if (TSTMP_GT(sc->sc_rxttime, tick)) {
438 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
439 sch->sch_nextc = sc->sc_rxttime;
442 if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) {
443 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
444 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
445 "giving up and removing syncache entry\n",
449 syncache_drop(sc, sch);
450 TCPSTAT_INC(tcps_sc_stale);
453 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
454 log(LOG_DEBUG, "%s; %s: Response timeout, "
455 "retransmitting (%u) SYN|ACK\n",
456 s, __func__, sc->sc_rxmits);
460 (void) syncache_respond(sc);
461 TCPSTAT_INC(tcps_sc_retransmitted);
462 syncache_timeout(sc, sch, 0);
464 if (!TAILQ_EMPTY(&(sch)->sch_bucket))
465 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
466 syncache_timer, (void *)(sch));
471 * Find an entry in the syncache.
472 * Returns always with locked syncache_head plus a matching entry or NULL.
475 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
478 struct syncache_head *sch;
481 if (inc->inc_flags & INC_ISIPV6) {
482 sch = &V_tcp_syncache.hashbase[
483 SYNCACHE_HASH6(inc, V_tcp_syncache.hashmask)];
488 /* Circle through bucket row to find matching entry. */
489 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
490 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
496 sch = &V_tcp_syncache.hashbase[
497 SYNCACHE_HASH(inc, V_tcp_syncache.hashmask)];
502 /* Circle through bucket row to find matching entry. */
503 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
505 if (sc->sc_inc.inc_flags & INC_ISIPV6)
508 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
512 SCH_LOCK_ASSERT(*schp);
513 return (NULL); /* always returns with locked sch */
517 * This function is called when we get a RST for a
518 * non-existent connection, so that we can see if the
519 * connection is in the syn cache. If it is, zap it.
522 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
525 struct syncache_head *sch;
528 sc = syncache_lookup(inc, &sch); /* returns locked sch */
529 SCH_LOCK_ASSERT(sch);
532 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
533 * See RFC 793 page 65, section SEGMENT ARRIVES.
535 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
536 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
537 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
538 "FIN flag set, segment ignored\n", s, __func__);
539 TCPSTAT_INC(tcps_badrst);
544 * No corresponding connection was found in syncache.
545 * If syncookies are enabled and possibly exclusively
546 * used, or we are under memory pressure, a valid RST
547 * may not find a syncache entry. In that case we're
548 * done and no SYN|ACK retransmissions will happen.
549 * Otherwise the RST was misdirected or spoofed.
552 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
553 log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
554 "syncache entry (possibly syncookie only), "
555 "segment ignored\n", s, __func__);
556 TCPSTAT_INC(tcps_badrst);
561 * If the RST bit is set, check the sequence number to see
562 * if this is a valid reset segment.
564 * In all states except SYN-SENT, all reset (RST) segments
565 * are validated by checking their SEQ-fields. A reset is
566 * valid if its sequence number is in the window.
568 * The sequence number in the reset segment is normally an
569 * echo of our outgoing acknowlegement numbers, but some hosts
570 * send a reset with the sequence number at the rightmost edge
571 * of our receive window, and we have to handle this case.
573 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
574 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
575 syncache_drop(sc, sch);
576 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
577 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
578 "connection attempt aborted by remote endpoint\n",
580 TCPSTAT_INC(tcps_sc_reset);
582 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
583 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != "
584 "IRS %u (+WND %u), segment ignored\n",
585 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
586 TCPSTAT_INC(tcps_badrst);
596 syncache_badack(struct in_conninfo *inc)
599 struct syncache_head *sch;
601 sc = syncache_lookup(inc, &sch); /* returns locked sch */
602 SCH_LOCK_ASSERT(sch);
604 syncache_drop(sc, sch);
605 TCPSTAT_INC(tcps_sc_badack);
611 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
614 struct syncache_head *sch;
616 sc = syncache_lookup(inc, &sch); /* returns locked sch */
617 SCH_LOCK_ASSERT(sch);
621 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
622 if (ntohl(th->th_seq) != sc->sc_iss)
626 * If we've rertransmitted 3 times and this is our second error,
627 * we remove the entry. Otherwise, we allow it to continue on.
628 * This prevents us from incorrectly nuking an entry during a
629 * spurious network outage.
633 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
634 sc->sc_flags |= SCF_UNREACH;
637 syncache_drop(sc, sch);
638 TCPSTAT_INC(tcps_sc_unreach);
644 * Build a new TCP socket structure from a syncache entry.
646 static struct socket *
647 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
649 struct inpcb *inp = NULL;
655 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
658 * Ok, create the full blown connection, and set things up
659 * as they would have been set up if we had created the
660 * connection when the SYN arrived. If we can't create
661 * the connection, abort it.
663 so = sonewconn(lso, SS_ISCONNECTED);
666 * Drop the connection; we will either send a RST or
667 * have the peer retransmit its SYN again after its
670 TCPSTAT_INC(tcps_listendrop);
671 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
672 log(LOG_DEBUG, "%s; %s: Socket create failed "
673 "due to limits or memory shortage\n",
680 mac_socketpeer_set_from_mbuf(m, so);
684 inp->inp_inc.inc_fibnum = so->so_fibnum;
686 INP_HASH_WLOCK(&V_tcbinfo);
688 /* Insert new socket into PCB hash list. */
689 inp->inp_inc.inc_flags = sc->sc_inc.inc_flags;
691 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
692 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
694 inp->inp_vflag &= ~INP_IPV6;
695 inp->inp_vflag |= INP_IPV4;
697 inp->inp_laddr = sc->sc_inc.inc_laddr;
703 * Install in the reservation hash table for now, but don't yet
704 * install a connection group since the full 4-tuple isn't yet
707 inp->inp_lport = sc->sc_inc.inc_lport;
708 if ((error = in_pcbinshash_nopcbgroup(inp)) != 0) {
710 * Undo the assignments above if we failed to
711 * put the PCB on the hash lists.
714 if (sc->sc_inc.inc_flags & INC_ISIPV6)
715 inp->in6p_laddr = in6addr_any;
718 inp->inp_laddr.s_addr = INADDR_ANY;
720 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
721 log(LOG_DEBUG, "%s; %s: in_pcbinshash failed "
726 INP_HASH_WUNLOCK(&V_tcbinfo);
730 /* Copy old policy into new socket's. */
731 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
732 printf("syncache_socket: could not copy policy\n");
735 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
736 struct inpcb *oinp = sotoinpcb(lso);
737 struct in6_addr laddr6;
738 struct sockaddr_in6 sin6;
740 * Inherit socket options from the listening socket.
741 * Note that in6p_inputopts are not (and should not be)
742 * copied, since it stores previously received options and is
743 * used to detect if each new option is different than the
744 * previous one and hence should be passed to a user.
745 * If we copied in6p_inputopts, a user would not be able to
746 * receive options just after calling the accept system call.
748 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
749 if (oinp->in6p_outputopts)
750 inp->in6p_outputopts =
751 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
753 sin6.sin6_family = AF_INET6;
754 sin6.sin6_len = sizeof(sin6);
755 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
756 sin6.sin6_port = sc->sc_inc.inc_fport;
757 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
758 laddr6 = inp->in6p_laddr;
759 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
760 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
761 if ((error = in6_pcbconnect_mbuf(inp, (struct sockaddr *)&sin6,
762 thread0.td_ucred, m)) != 0) {
763 inp->in6p_laddr = laddr6;
764 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
765 log(LOG_DEBUG, "%s; %s: in6_pcbconnect failed "
770 INP_HASH_WUNLOCK(&V_tcbinfo);
773 /* Override flowlabel from in6_pcbconnect. */
774 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
775 inp->inp_flow |= sc->sc_flowlabel;
778 #if defined(INET) && defined(INET6)
783 struct in_addr laddr;
784 struct sockaddr_in sin;
786 inp->inp_options = (m) ? ip_srcroute(m) : NULL;
788 if (inp->inp_options == NULL) {
789 inp->inp_options = sc->sc_ipopts;
790 sc->sc_ipopts = NULL;
793 sin.sin_family = AF_INET;
794 sin.sin_len = sizeof(sin);
795 sin.sin_addr = sc->sc_inc.inc_faddr;
796 sin.sin_port = sc->sc_inc.inc_fport;
797 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
798 laddr = inp->inp_laddr;
799 if (inp->inp_laddr.s_addr == INADDR_ANY)
800 inp->inp_laddr = sc->sc_inc.inc_laddr;
801 if ((error = in_pcbconnect_mbuf(inp, (struct sockaddr *)&sin,
802 thread0.td_ucred, m)) != 0) {
803 inp->inp_laddr = laddr;
804 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
805 log(LOG_DEBUG, "%s; %s: in_pcbconnect failed "
810 INP_HASH_WUNLOCK(&V_tcbinfo);
815 INP_HASH_WUNLOCK(&V_tcbinfo);
817 tp->t_state = TCPS_SYN_RECEIVED;
818 tp->iss = sc->sc_iss;
819 tp->irs = sc->sc_irs;
822 tp->snd_wl1 = sc->sc_irs;
823 tp->snd_max = tp->iss + 1;
824 tp->snd_nxt = tp->iss + 1;
825 tp->rcv_up = sc->sc_irs + 1;
826 tp->rcv_wnd = sc->sc_wnd;
827 tp->rcv_adv += tp->rcv_wnd;
828 tp->last_ack_sent = tp->rcv_nxt;
830 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
831 if (sc->sc_flags & SCF_NOOPT)
832 tp->t_flags |= TF_NOOPT;
834 if (sc->sc_flags & SCF_WINSCALE) {
835 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
836 tp->snd_scale = sc->sc_requested_s_scale;
837 tp->request_r_scale = sc->sc_requested_r_scale;
839 if (sc->sc_flags & SCF_TIMESTAMP) {
840 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
841 tp->ts_recent = sc->sc_tsreflect;
842 tp->ts_recent_age = tcp_ts_getticks();
843 tp->ts_offset = sc->sc_tsoff;
846 if (sc->sc_flags & SCF_SIGNATURE)
847 tp->t_flags |= TF_SIGNATURE;
849 if (sc->sc_flags & SCF_SACK)
850 tp->t_flags |= TF_SACK_PERMIT;
853 if (sc->sc_flags & SCF_ECN)
854 tp->t_flags |= TF_ECN_PERMIT;
857 * Set up MSS and get cached values from tcp_hostcache.
858 * This might overwrite some of the defaults we just set.
860 tcp_mss(tp, sc->sc_peer_mss);
863 * If the SYN,ACK was retransmitted, indicate that CWND to be
864 * limited to one segment in cc_conn_init().
865 * NB: sc_rxmits counts all SYN,ACK transmits, not just retransmits.
867 if (sc->sc_rxmits > 1)
872 * Allow a TOE driver to install its hooks. Note that we hold the
873 * pcbinfo lock too and that prevents tcp_usr_accept from accepting a
874 * new connection before the TOE driver has done its thing.
876 if (ADDED_BY_TOE(sc)) {
877 struct toedev *tod = sc->sc_tod;
879 tod->tod_offload_socket(tod, sc->sc_todctx, so);
883 * Copy and activate timers.
885 tp->t_keepinit = sototcpcb(lso)->t_keepinit;
886 tp->t_keepidle = sototcpcb(lso)->t_keepidle;
887 tp->t_keepintvl = sototcpcb(lso)->t_keepintvl;
888 tp->t_keepcnt = sototcpcb(lso)->t_keepcnt;
889 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
893 TCPSTAT_INC(tcps_accepts);
905 * This function gets called when we receive an ACK for a
906 * socket in the LISTEN state. We look up the connection
907 * in the syncache, and if its there, we pull it out of
908 * the cache and turn it into a full-blown connection in
909 * the SYN-RECEIVED state.
912 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
913 struct socket **lsop, struct mbuf *m)
916 struct syncache_head *sch;
921 * Global TCP locks are held because we manipulate the PCB lists
922 * and create a new socket.
924 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
925 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
926 ("%s: can handle only ACK", __func__));
928 sc = syncache_lookup(inc, &sch); /* returns locked sch */
929 SCH_LOCK_ASSERT(sch);
932 * There is no syncache entry, so see if this ACK is
933 * a returning syncookie. To do this, first:
934 * A. See if this socket has had a syncache entry dropped in
935 * the past. We don't want to accept a bogus syncookie
936 * if we've never received a SYN.
937 * B. check that the syncookie is valid. If it is, then
938 * cobble up a fake syncache entry, and return.
940 if (!V_tcp_syncookies) {
942 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
943 log(LOG_DEBUG, "%s; %s: Spurious ACK, "
944 "segment rejected (syncookies disabled)\n",
948 bzero(&scs, sizeof(scs));
949 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
952 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
953 log(LOG_DEBUG, "%s; %s: Segment failed "
954 "SYNCOOKIE authentication, segment rejected "
955 "(probably spoofed)\n", s, __func__);
959 /* Pull out the entry to unlock the bucket row. */
960 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
963 if (ADDED_BY_TOE(sc)) {
964 struct toedev *tod = sc->sc_tod;
966 tod->tod_syncache_removed(tod, sc->sc_todctx);
973 * Segment validation:
974 * ACK must match our initial sequence number + 1 (the SYN|ACK).
976 if (th->th_ack != sc->sc_iss + 1) {
977 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
978 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment "
979 "rejected\n", s, __func__, th->th_ack, sc->sc_iss);
984 * The SEQ must fall in the window starting at the received
985 * initial receive sequence number + 1 (the SYN).
987 if (SEQ_LEQ(th->th_seq, sc->sc_irs) ||
988 SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
989 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
990 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment "
991 "rejected\n", s, __func__, th->th_seq, sc->sc_irs);
995 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
996 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
997 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
998 "segment rejected\n", s, __func__);
1002 * If timestamps were negotiated the reflected timestamp
1003 * must be equal to what we actually sent in the SYN|ACK.
1005 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts) {
1006 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
1007 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
1008 "segment rejected\n",
1009 s, __func__, to->to_tsecr, sc->sc_ts);
1013 *lsop = syncache_socket(sc, *lsop, m);
1016 TCPSTAT_INC(tcps_sc_aborted);
1018 TCPSTAT_INC(tcps_sc_completed);
1020 /* how do we find the inp for the new socket? */
1025 if (sc != NULL && sc != &scs)
1034 * Given a LISTEN socket and an inbound SYN request, add
1035 * this to the syn cache, and send back a segment:
1036 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
1039 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
1040 * Doing so would require that we hold onto the data and deliver it
1041 * to the application. However, if we are the target of a SYN-flood
1042 * DoS attack, an attacker could send data which would eventually
1043 * consume all available buffer space if it were ACKed. By not ACKing
1044 * the data, we avoid this DoS scenario.
1047 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1048 struct inpcb *inp, struct socket **lsop, struct mbuf *m, void *tod,
1053 struct syncache *sc = NULL;
1054 struct syncache_head *sch;
1055 struct mbuf *ipopts = NULL;
1058 int win, sb_hiwat, ip_ttl, ip_tos;
1061 int autoflowlabel = 0;
1064 struct label *maclabel;
1066 struct syncache scs;
1069 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1070 INP_WLOCK_ASSERT(inp); /* listen socket */
1071 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
1072 ("%s: unexpected tcp flags", __func__));
1075 * Combine all so/tp operations very early to drop the INP lock as
1080 cred = crhold(so->so_cred);
1083 if ((inc->inc_flags & INC_ISIPV6) &&
1084 (inp->inp_flags & IN6P_AUTOFLOWLABEL))
1087 ip_ttl = inp->inp_ip_ttl;
1088 ip_tos = inp->inp_ip_tos;
1089 win = sbspace(&so->so_rcv);
1090 sb_hiwat = so->so_rcv.sb_hiwat;
1091 ltflags = (tp->t_flags & (TF_NOOPT | TF_SIGNATURE));
1093 /* By the time we drop the lock these should no longer be used. */
1098 if (mac_syncache_init(&maclabel) != 0) {
1100 INP_INFO_WUNLOCK(&V_tcbinfo);
1103 mac_syncache_create(maclabel, inp);
1106 INP_INFO_WUNLOCK(&V_tcbinfo);
1109 * Remember the IP options, if any.
1112 if (!(inc->inc_flags & INC_ISIPV6))
1115 ipopts = (m) ? ip_srcroute(m) : NULL;
1121 * See if we already have an entry for this connection.
1122 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1124 * XXX: should the syncache be re-initialized with the contents
1125 * of the new SYN here (which may have different options?)
1127 * XXX: We do not check the sequence number to see if this is a
1128 * real retransmit or a new connection attempt. The question is
1129 * how to handle such a case; either ignore it as spoofed, or
1130 * drop the current entry and create a new one?
1132 sc = syncache_lookup(inc, &sch); /* returns locked entry */
1133 SCH_LOCK_ASSERT(sch);
1135 TCPSTAT_INC(tcps_sc_dupsyn);
1138 * If we were remembering a previous source route,
1139 * forget it and use the new one we've been given.
1142 (void) m_free(sc->sc_ipopts);
1143 sc->sc_ipopts = ipopts;
1146 * Update timestamp if present.
1148 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
1149 sc->sc_tsreflect = to->to_tsval;
1151 sc->sc_flags &= ~SCF_TIMESTAMP;
1154 * Since we have already unconditionally allocated label
1155 * storage, free it up. The syncache entry will already
1156 * have an initialized label we can use.
1158 mac_syncache_destroy(&maclabel);
1160 /* Retransmit SYN|ACK and reset retransmit count. */
1161 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
1162 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
1163 "resetting timer and retransmitting SYN|ACK\n",
1167 if (syncache_respond(sc) == 0) {
1169 syncache_timeout(sc, sch, 1);
1170 TCPSTAT_INC(tcps_sndacks);
1171 TCPSTAT_INC(tcps_sndtotal);
1177 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1180 * The zone allocator couldn't provide more entries.
1181 * Treat this as if the cache was full; drop the oldest
1182 * entry and insert the new one.
1184 TCPSTAT_INC(tcps_sc_zonefail);
1185 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1186 syncache_drop(sc, sch);
1187 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1189 if (V_tcp_syncookies) {
1190 bzero(&scs, sizeof(scs));
1195 (void) m_free(ipopts);
1202 * Fill in the syncache values.
1205 sc->sc_label = maclabel;
1209 sc->sc_ipopts = ipopts;
1210 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1212 if (!(inc->inc_flags & INC_ISIPV6))
1215 sc->sc_ip_tos = ip_tos;
1216 sc->sc_ip_ttl = ip_ttl;
1220 sc->sc_todctx = todctx;
1222 sc->sc_irs = th->th_seq;
1223 sc->sc_iss = arc4random();
1225 sc->sc_flowlabel = 0;
1228 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
1229 * win was derived from socket earlier in the function.
1232 win = imin(win, TCP_MAXWIN);
1235 if (V_tcp_do_rfc1323) {
1237 * A timestamp received in a SYN makes
1238 * it ok to send timestamp requests and replies.
1240 if (to->to_flags & TOF_TS) {
1241 sc->sc_tsreflect = to->to_tsval;
1242 sc->sc_ts = tcp_ts_getticks();
1243 sc->sc_flags |= SCF_TIMESTAMP;
1245 if (to->to_flags & TOF_SCALE) {
1249 * Pick the smallest possible scaling factor that
1250 * will still allow us to scale up to sb_max, aka
1251 * kern.ipc.maxsockbuf.
1253 * We do this because there are broken firewalls that
1254 * will corrupt the window scale option, leading to
1255 * the other endpoint believing that our advertised
1256 * window is unscaled. At scale factors larger than
1257 * 5 the unscaled window will drop below 1500 bytes,
1258 * leading to serious problems when traversing these
1261 * With the default maxsockbuf of 256K, a scale factor
1262 * of 3 will be chosen by this algorithm. Those who
1263 * choose a larger maxsockbuf should watch out
1264 * for the compatiblity problems mentioned above.
1266 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1267 * or <SYN,ACK>) segment itself is never scaled.
1269 while (wscale < TCP_MAX_WINSHIFT &&
1270 (TCP_MAXWIN << wscale) < sb_max)
1272 sc->sc_requested_r_scale = wscale;
1273 sc->sc_requested_s_scale = to->to_wscale;
1274 sc->sc_flags |= SCF_WINSCALE;
1277 #ifdef TCP_SIGNATURE
1279 * If listening socket requested TCP digests, and received SYN
1280 * contains the option, flag this in the syncache so that
1281 * syncache_respond() will do the right thing with the SYN+ACK.
1282 * XXX: Currently we always record the option by default and will
1283 * attempt to use it in syncache_respond().
1285 if (to->to_flags & TOF_SIGNATURE || ltflags & TF_SIGNATURE)
1286 sc->sc_flags |= SCF_SIGNATURE;
1288 if (to->to_flags & TOF_SACKPERM)
1289 sc->sc_flags |= SCF_SACK;
1290 if (to->to_flags & TOF_MSS)
1291 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */
1292 if (ltflags & TF_NOOPT)
1293 sc->sc_flags |= SCF_NOOPT;
1294 if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn)
1295 sc->sc_flags |= SCF_ECN;
1297 if (V_tcp_syncookies) {
1298 syncookie_generate(sch, sc, &flowtmp);
1301 sc->sc_flowlabel = flowtmp;
1307 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1313 * Do a standard 3-way handshake.
1315 if (syncache_respond(sc) == 0) {
1316 if (V_tcp_syncookies && V_tcp_syncookiesonly && sc != &scs)
1318 else if (sc != &scs)
1319 syncache_insert(sc, sch); /* locks and unlocks sch */
1320 TCPSTAT_INC(tcps_sndacks);
1321 TCPSTAT_INC(tcps_sndtotal);
1325 TCPSTAT_INC(tcps_sc_dropped);
1333 mac_syncache_destroy(&maclabel);
1343 syncache_respond(struct syncache *sc)
1345 struct ip *ip = NULL;
1347 struct tcphdr *th = NULL;
1348 int optlen, error = 0; /* Make compiler happy */
1349 u_int16_t hlen, tlen, mssopt;
1352 struct ip6_hdr *ip6 = NULL;
1357 (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) :
1360 tlen = hlen + sizeof(struct tcphdr);
1362 /* Determine MSS we advertize to other end of connection. */
1363 mssopt = tcp_mssopt(&sc->sc_inc);
1364 if (sc->sc_peer_mss)
1365 mssopt = max( min(sc->sc_peer_mss, mssopt), V_tcp_minmss);
1367 /* XXX: Assume that the entire packet will fit in a header mbuf. */
1368 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1369 ("syncache: mbuf too small"));
1371 /* Create the IP+TCP header from scratch. */
1372 m = m_gethdr(M_DONTWAIT, MT_DATA);
1376 mac_syncache_create_mbuf(sc->sc_label, m);
1378 m->m_data += max_linkhdr;
1380 m->m_pkthdr.len = tlen;
1381 m->m_pkthdr.rcvif = NULL;
1384 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1385 ip6 = mtod(m, struct ip6_hdr *);
1386 ip6->ip6_vfc = IPV6_VERSION;
1387 ip6->ip6_nxt = IPPROTO_TCP;
1388 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1389 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1390 ip6->ip6_plen = htons(tlen - hlen);
1391 /* ip6_hlim is set after checksum */
1392 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1393 ip6->ip6_flow |= sc->sc_flowlabel;
1395 th = (struct tcphdr *)(ip6 + 1);
1398 #if defined(INET6) && defined(INET)
1403 ip = mtod(m, struct ip *);
1404 ip->ip_v = IPVERSION;
1405 ip->ip_hl = sizeof(struct ip) >> 2;
1410 ip->ip_p = IPPROTO_TCP;
1411 ip->ip_src = sc->sc_inc.inc_laddr;
1412 ip->ip_dst = sc->sc_inc.inc_faddr;
1413 ip->ip_ttl = sc->sc_ip_ttl;
1414 ip->ip_tos = sc->sc_ip_tos;
1417 * See if we should do MTU discovery. Route lookups are
1418 * expensive, so we will only unset the DF bit if:
1420 * 1) path_mtu_discovery is disabled
1421 * 2) the SCF_UNREACH flag has been set
1423 if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1424 ip->ip_off |= IP_DF;
1426 th = (struct tcphdr *)(ip + 1);
1429 th->th_sport = sc->sc_inc.inc_lport;
1430 th->th_dport = sc->sc_inc.inc_fport;
1432 th->th_seq = htonl(sc->sc_iss);
1433 th->th_ack = htonl(sc->sc_irs + 1);
1434 th->th_off = sizeof(struct tcphdr) >> 2;
1436 th->th_flags = TH_SYN|TH_ACK;
1437 th->th_win = htons(sc->sc_wnd);
1440 if (sc->sc_flags & SCF_ECN) {
1441 th->th_flags |= TH_ECE;
1442 TCPSTAT_INC(tcps_ecn_shs);
1445 /* Tack on the TCP options. */
1446 if ((sc->sc_flags & SCF_NOOPT) == 0) {
1450 to.to_flags = TOF_MSS;
1451 if (sc->sc_flags & SCF_WINSCALE) {
1452 to.to_wscale = sc->sc_requested_r_scale;
1453 to.to_flags |= TOF_SCALE;
1455 if (sc->sc_flags & SCF_TIMESTAMP) {
1456 /* Virgin timestamp or TCP cookie enhanced one. */
1457 to.to_tsval = sc->sc_ts;
1458 to.to_tsecr = sc->sc_tsreflect;
1459 to.to_flags |= TOF_TS;
1461 if (sc->sc_flags & SCF_SACK)
1462 to.to_flags |= TOF_SACKPERM;
1463 #ifdef TCP_SIGNATURE
1464 if (sc->sc_flags & SCF_SIGNATURE)
1465 to.to_flags |= TOF_SIGNATURE;
1467 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1469 /* Adjust headers by option size. */
1470 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1472 m->m_pkthdr.len += optlen;
1474 #ifdef TCP_SIGNATURE
1475 if (sc->sc_flags & SCF_SIGNATURE)
1476 tcp_signature_compute(m, 0, 0, optlen,
1477 to.to_signature, IPSEC_DIR_OUTBOUND);
1480 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1481 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1484 ip->ip_len += optlen;
1488 M_SETFIB(m, sc->sc_inc.inc_fibnum);
1489 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1491 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1492 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1493 th->th_sum = in6_cksum_pseudo(ip6, tlen + optlen - hlen,
1495 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1497 if (ADDED_BY_TOE(sc)) {
1498 struct toedev *tod = sc->sc_tod;
1500 error = tod->tod_syncache_respond(tod, sc->sc_todctx, m);
1505 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1508 #if defined(INET6) && defined(INET)
1513 m->m_pkthdr.csum_flags = CSUM_TCP;
1514 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1515 htons(tlen + optlen - hlen + IPPROTO_TCP));
1517 if (ADDED_BY_TOE(sc)) {
1518 struct toedev *tod = sc->sc_tod;
1520 error = tod->tod_syncache_respond(tod, sc->sc_todctx, m);
1525 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1532 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1533 struct inpcb *inp, struct socket **lsop, struct mbuf *m)
1535 _syncache_add(inc, to, th, inp, lsop, m, NULL, NULL);
1539 tcp_offload_syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1540 struct inpcb *inp, struct socket **lsop, void *tod, void *todctx)
1543 _syncache_add(inc, to, th, inp, lsop, NULL, tod, todctx);
1546 * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1547 * receive and to be able to handle SYN floods from bogus source addresses
1548 * (where we will never receive any reply). SYN floods try to exhaust all
1549 * our memory and available slots in the SYN cache table to cause a denial
1550 * of service to legitimate users of the local host.
1552 * The idea of SYN cookies is to encode and include all necessary information
1553 * about the connection setup state within the SYN-ACK we send back and thus
1554 * to get along without keeping any local state until the ACK to the SYN-ACK
1555 * arrives (if ever). Everything we need to know should be available from
1556 * the information we encoded in the SYN-ACK.
1558 * More information about the theory behind SYN cookies and its first
1559 * discussion and specification can be found at:
1560 * http://cr.yp.to/syncookies.html (overview)
1561 * http://cr.yp.to/syncookies/archive (gory details)
1563 * This implementation extends the orginal idea and first implementation
1564 * of FreeBSD by using not only the initial sequence number field to store
1565 * information but also the timestamp field if present. This way we can
1566 * keep track of the entire state we need to know to recreate the session in
1567 * its original form. Almost all TCP speakers implement RFC1323 timestamps
1568 * these days. For those that do not we still have to live with the known
1569 * shortcomings of the ISN only SYN cookies.
1573 * Initial sequence number we send:
1574 * 31|................................|0
1575 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1576 * D = MD5 Digest (first dword)
1578 * R = Rotation of secret
1579 * P = Odd or Even secret
1581 * The MD5 Digest is computed with over following parameters:
1582 * a) randomly rotated secret
1583 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1584 * c) the received initial sequence number from remote host
1585 * d) the rotation offset and odd/even bit
1587 * Timestamp we send:
1588 * 31|................................|0
1589 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1590 * D = MD5 Digest (third dword) (only as filler)
1591 * S = Requested send window scale
1592 * R = Requested receive window scale
1594 * 5 = TCP-MD5 enabled (not implemented yet)
1595 * XORed with MD5 Digest (forth dword)
1597 * The timestamp isn't cryptographically secure and doesn't need to be.
1598 * The double use of the MD5 digest dwords ties it to a specific remote/
1599 * local host/port, remote initial sequence number and our local time
1600 * limited secret. A received timestamp is reverted (XORed) and then
1601 * the contained MD5 dword is compared to the computed one to ensure the
1602 * timestamp belongs to the SYN-ACK we sent. The other parameters may
1603 * have been tampered with but this isn't different from supplying bogus
1604 * values in the SYN in the first place.
1606 * Some problems with SYN cookies remain however:
1607 * Consider the problem of a recreated (and retransmitted) cookie. If the
1608 * original SYN was accepted, the connection is established. The second
1609 * SYN is inflight, and if it arrives with an ISN that falls within the
1610 * receive window, the connection is killed.
1613 * A heuristic to determine when to accept syn cookies is not necessary.
1614 * An ACK flood would cause the syncookie verification to be attempted,
1615 * but a SYN flood causes syncookies to be generated. Both are of equal
1616 * cost, so there's no point in trying to optimize the ACK flood case.
1617 * Also, if you don't process certain ACKs for some reason, then all someone
1618 * would have to do is launch a SYN and ACK flood at the same time, which
1619 * would stop cookie verification and defeat the entire purpose of syncookies.
1621 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1624 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1625 u_int32_t *flowlabel)
1628 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1631 u_int off, pmss, mss;
1634 SCH_LOCK_ASSERT(sch);
1636 /* Which of the two secrets to use. */
1637 secbits = sch->sch_oddeven ?
1638 sch->sch_secbits_odd : sch->sch_secbits_even;
1640 /* Reseed secret if too old. */
1641 if (sch->sch_reseed < time_uptime) {
1642 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */
1643 secbits = sch->sch_oddeven ?
1644 sch->sch_secbits_odd : sch->sch_secbits_even;
1645 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1646 secbits[i] = arc4random();
1647 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1650 /* Secret rotation offset. */
1651 off = sc->sc_iss & 0x7; /* iss was randomized before */
1653 /* Maximum segment size calculation. */
1655 max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), V_tcp_minmss);
1656 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1657 if (tcp_sc_msstab[mss] <= pmss)
1660 /* Fold parameters and MD5 digest into the ISN we will send. */
1661 data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1662 data |= off << 1; /* secret offset, derived from iss, 3 bits */
1663 data |= mss << 4; /* mss, 3 bits */
1666 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1667 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1668 MD5Update(&ctx, secbits, off);
1669 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1670 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1671 MD5Update(&ctx, &data, sizeof(data));
1672 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1674 data |= (md5_buffer[0] << 7);
1678 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1681 /* Additional parameters are stored in the timestamp if present. */
1682 if (sc->sc_flags & SCF_TIMESTAMP) {
1683 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1684 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1685 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */
1686 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */
1687 data |= md5_buffer[2] << 10; /* more digest bits */
1688 data ^= md5_buffer[3];
1690 sc->sc_tsoff = data - tcp_ts_getticks(); /* after XOR */
1693 TCPSTAT_INC(tcps_sc_sendcookie);
1696 static struct syncache *
1697 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1698 struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1702 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1706 int off, mss, wnd, flags;
1708 SCH_LOCK_ASSERT(sch);
1711 * Pull information out of SYN-ACK/ACK and
1712 * revert sequence number advances.
1714 ack = th->th_ack - 1;
1715 seq = th->th_seq - 1;
1716 off = (ack >> 1) & 0x7;
1717 mss = (ack >> 4) & 0x7;
1720 /* Which of the two secrets to use. */
1721 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1724 * The secret wasn't updated for the lifetime of a syncookie,
1725 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1727 if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) {
1731 /* Recompute the digest so we can compare it. */
1733 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1734 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1735 MD5Update(&ctx, secbits, off);
1736 MD5Update(&ctx, inc, sizeof(*inc));
1737 MD5Update(&ctx, &seq, sizeof(seq));
1738 MD5Update(&ctx, &flags, sizeof(flags));
1739 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1741 /* Does the digest part of or ACK'ed ISS match? */
1742 if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1745 /* Does the digest part of our reflected timestamp match? */
1746 if (to->to_flags & TOF_TS) {
1747 data = md5_buffer[3] ^ to->to_tsecr;
1748 if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1752 /* Fill in the syncache values. */
1753 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1754 sc->sc_ipopts = NULL;
1760 if (inc->inc_flags & INC_ISIPV6) {
1761 if (sotoinpcb(so)->inp_flags & IN6P_AUTOFLOWLABEL)
1762 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1766 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1767 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1770 /* Additional parameters that were encoded in the timestamp. */
1772 sc->sc_flags |= SCF_TIMESTAMP;
1773 sc->sc_tsreflect = to->to_tsval;
1774 sc->sc_ts = to->to_tsecr;
1775 sc->sc_tsoff = to->to_tsecr - tcp_ts_getticks();
1776 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1777 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1778 sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1780 sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1782 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1783 sc->sc_flags |= SCF_WINSCALE;
1785 sc->sc_flags |= SCF_NOOPT;
1787 wnd = sbspace(&so->so_rcv);
1789 wnd = imin(wnd, TCP_MAXWIN);
1793 sc->sc_peer_mss = tcp_sc_msstab[mss];
1795 TCPSTAT_INC(tcps_sc_recvcookie);
1800 * Returns the current number of syncache entries. This number
1801 * will probably change before you get around to calling
1806 syncache_pcbcount(void)
1808 struct syncache_head *sch;
1811 for (count = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1812 /* No need to lock for a read. */
1813 sch = &V_tcp_syncache.hashbase[i];
1814 count += sch->sch_length;
1820 * Exports the syncache entries to userland so that netstat can display
1821 * them alongside the other sockets. This function is intended to be
1822 * called only from tcp_pcblist.
1824 * Due to concurrency on an active system, the number of pcbs exported
1825 * may have no relation to max_pcbs. max_pcbs merely indicates the
1826 * amount of space the caller allocated for this function to use.
1829 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
1832 struct syncache *sc;
1833 struct syncache_head *sch;
1834 int count, error, i;
1836 for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1837 sch = &V_tcp_syncache.hashbase[i];
1839 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
1840 if (count >= max_pcbs) {
1844 if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0)
1846 bzero(&xt, sizeof(xt));
1847 xt.xt_len = sizeof(xt);
1848 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1849 xt.xt_inp.inp_vflag = INP_IPV6;
1851 xt.xt_inp.inp_vflag = INP_IPV4;
1852 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
1853 xt.xt_tp.t_inpcb = &xt.xt_inp;
1854 xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
1855 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1856 xt.xt_socket.xso_len = sizeof (struct xsocket);
1857 xt.xt_socket.so_type = SOCK_STREAM;
1858 xt.xt_socket.so_state = SS_ISCONNECTING;
1859 error = SYSCTL_OUT(req, &xt, sizeof xt);
1869 *pcbs_exported = count;