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>
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(int, tcp_syncookies) = 1;
102 #define V_tcp_syncookies VNET(tcp_syncookies)
103 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
104 &VNET_NAME(tcp_syncookies), 0,
105 "Use TCP SYN cookies if the syncache overflows");
107 static VNET_DEFINE(int, tcp_syncookiesonly) = 0;
108 #define V_tcp_syncookiesonly VNET(tcp_syncookiesonly)
109 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW,
110 &VNET_NAME(tcp_syncookiesonly), 0,
111 "Use only TCP SYN cookies");
113 #ifdef TCP_OFFLOAD_DISABLE
114 #define TOEPCB_ISSET(sc) (0)
116 #define TOEPCB_ISSET(sc) ((sc)->sc_toepcb != 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 void syncache_timeout(struct syncache *sc, struct syncache_head *sch,
128 static void syncache_timer(void *);
129 static void syncookie_generate(struct syncache_head *, struct syncache *,
131 static struct syncache
132 *syncookie_lookup(struct in_conninfo *, struct syncache_head *,
133 struct syncache *, struct tcpopt *, struct tcphdr *,
137 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
138 * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds,
139 * the odds are that the user has given up attempting to connect by then.
141 #define SYNCACHE_MAXREXMTS 3
143 /* Arbitrary values */
144 #define TCP_SYNCACHE_HASHSIZE 512
145 #define TCP_SYNCACHE_BUCKETLIMIT 30
147 static VNET_DEFINE(struct tcp_syncache, tcp_syncache);
148 #define V_tcp_syncache VNET(tcp_syncache)
150 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0,
153 SYSCTL_VNET_UINT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
154 &VNET_NAME(tcp_syncache.bucket_limit), 0,
155 "Per-bucket hash limit for syncache");
157 SYSCTL_VNET_UINT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
158 &VNET_NAME(tcp_syncache.cache_limit), 0,
159 "Overall entry limit for syncache");
161 SYSCTL_VNET_UINT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
162 &VNET_NAME(tcp_syncache.cache_count), 0,
163 "Current number of entries in syncache");
165 SYSCTL_VNET_UINT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
166 &VNET_NAME(tcp_syncache.hashsize), 0,
167 "Size of TCP syncache hashtable");
169 SYSCTL_VNET_UINT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
170 &VNET_NAME(tcp_syncache.rexmt_limit), 0,
171 "Limit on SYN/ACK retransmissions");
173 VNET_DEFINE(int, tcp_sc_rst_sock_fail) = 1;
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_syncache.cache_count = 0;
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 uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit);
276 syncache_destroy(void)
278 struct syncache_head *sch;
279 struct syncache *sc, *nsc;
282 /* Cleanup hash buckets: stop timers, free entries, destroy locks. */
283 for (i = 0; i < V_tcp_syncache.hashsize; i++) {
285 sch = &V_tcp_syncache.hashbase[i];
286 callout_drain(&sch->sch_timer);
289 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc)
290 syncache_drop(sc, sch);
292 KASSERT(TAILQ_EMPTY(&sch->sch_bucket),
293 ("%s: sch->sch_bucket not empty", __func__));
294 KASSERT(sch->sch_length == 0, ("%s: sch->sch_length %d not 0",
295 __func__, sch->sch_length));
296 mtx_destroy(&sch->sch_mtx);
299 KASSERT(V_tcp_syncache.cache_count == 0, ("%s: cache_count %d not 0",
300 __func__, V_tcp_syncache.cache_count));
302 /* Free the allocated global resources. */
303 uma_zdestroy(V_tcp_syncache.zone);
304 free(V_tcp_syncache.hashbase, M_SYNCACHE);
309 * Inserts a syncache entry into the specified bucket row.
310 * Locks and unlocks the syncache_head autonomously.
313 syncache_insert(struct syncache *sc, struct syncache_head *sch)
315 struct syncache *sc2;
320 * Make sure that we don't overflow the per-bucket limit.
321 * If the bucket is full, toss the oldest element.
323 if (sch->sch_length >= V_tcp_syncache.bucket_limit) {
324 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
325 ("sch->sch_length incorrect"));
326 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
327 syncache_drop(sc2, sch);
328 TCPSTAT_INC(tcps_sc_bucketoverflow);
331 /* Put it into the bucket. */
332 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
335 /* Reinitialize the bucket row's timer. */
336 if (sch->sch_length == 1)
337 sch->sch_nextc = ticks + INT_MAX;
338 syncache_timeout(sc, sch, 1);
342 V_tcp_syncache.cache_count++;
343 TCPSTAT_INC(tcps_sc_added);
347 * Remove and free entry from syncache bucket row.
348 * Expects locked syncache head.
351 syncache_drop(struct syncache *sc, struct syncache_head *sch)
354 SCH_LOCK_ASSERT(sch);
356 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
359 #ifndef TCP_OFFLOAD_DISABLE
361 sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb);
364 V_tcp_syncache.cache_count--;
368 * Engage/reengage time on bucket row.
371 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
373 sc->sc_rxttime = ticks +
374 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
376 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
377 sch->sch_nextc = sc->sc_rxttime;
379 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
380 syncache_timer, (void *)sch);
385 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
386 * If we have retransmitted an entry the maximum number of times, expire it.
387 * One separate timer for each bucket row.
390 syncache_timer(void *xsch)
392 struct syncache_head *sch = (struct syncache_head *)xsch;
393 struct syncache *sc, *nsc;
397 CURVNET_SET(sch->sch_vnet);
399 /* NB: syncache_head has already been locked by the callout. */
400 SCH_LOCK_ASSERT(sch);
403 * In the following cycle we may remove some entries and/or
404 * advance some timeouts, so re-initialize the bucket timer.
406 sch->sch_nextc = tick + INT_MAX;
408 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
410 * We do not check if the listen socket still exists
411 * and accept the case where the listen socket may be
412 * gone by the time we resend the SYN/ACK. We do
413 * not expect this to happens often. If it does,
414 * then the RST will be sent by the time the remote
415 * host does the SYN/ACK->ACK.
417 if (TSTMP_GT(sc->sc_rxttime, tick)) {
418 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
419 sch->sch_nextc = sc->sc_rxttime;
422 if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) {
423 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
424 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
425 "giving up and removing syncache entry\n",
429 syncache_drop(sc, sch);
430 TCPSTAT_INC(tcps_sc_stale);
433 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
434 log(LOG_DEBUG, "%s; %s: Response timeout, "
435 "retransmitting (%u) SYN|ACK\n",
436 s, __func__, sc->sc_rxmits);
440 (void) syncache_respond(sc);
441 TCPSTAT_INC(tcps_sc_retransmitted);
442 syncache_timeout(sc, sch, 0);
444 if (!TAILQ_EMPTY(&(sch)->sch_bucket))
445 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
446 syncache_timer, (void *)(sch));
451 * Find an entry in the syncache.
452 * Returns always with locked syncache_head plus a matching entry or NULL.
455 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
458 struct syncache_head *sch;
461 if (inc->inc_flags & INC_ISIPV6) {
462 sch = &V_tcp_syncache.hashbase[
463 SYNCACHE_HASH6(inc, V_tcp_syncache.hashmask)];
468 /* Circle through bucket row to find matching entry. */
469 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
470 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
476 sch = &V_tcp_syncache.hashbase[
477 SYNCACHE_HASH(inc, V_tcp_syncache.hashmask)];
482 /* Circle through bucket row to find matching entry. */
483 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
485 if (sc->sc_inc.inc_flags & INC_ISIPV6)
488 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
492 SCH_LOCK_ASSERT(*schp);
493 return (NULL); /* always returns with locked sch */
497 * This function is called when we get a RST for a
498 * non-existent connection, so that we can see if the
499 * connection is in the syn cache. If it is, zap it.
502 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
505 struct syncache_head *sch;
508 sc = syncache_lookup(inc, &sch); /* returns locked sch */
509 SCH_LOCK_ASSERT(sch);
512 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
513 * See RFC 793 page 65, section SEGMENT ARRIVES.
515 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
516 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
517 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
518 "FIN flag set, segment ignored\n", s, __func__);
519 TCPSTAT_INC(tcps_badrst);
524 * No corresponding connection was found in syncache.
525 * If syncookies are enabled and possibly exclusively
526 * used, or we are under memory pressure, a valid RST
527 * may not find a syncache entry. In that case we're
528 * done and no SYN|ACK retransmissions will happen.
529 * Otherwise the RST was misdirected or spoofed.
532 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
533 log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
534 "syncache entry (possibly syncookie only), "
535 "segment ignored\n", s, __func__);
536 TCPSTAT_INC(tcps_badrst);
541 * If the RST bit is set, check the sequence number to see
542 * if this is a valid reset segment.
544 * In all states except SYN-SENT, all reset (RST) segments
545 * are validated by checking their SEQ-fields. A reset is
546 * valid if its sequence number is in the window.
548 * The sequence number in the reset segment is normally an
549 * echo of our outgoing acknowlegement numbers, but some hosts
550 * send a reset with the sequence number at the rightmost edge
551 * of our receive window, and we have to handle this case.
553 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
554 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
555 syncache_drop(sc, sch);
556 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
557 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
558 "connection attempt aborted by remote endpoint\n",
560 TCPSTAT_INC(tcps_sc_reset);
562 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
563 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != "
564 "IRS %u (+WND %u), segment ignored\n",
565 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
566 TCPSTAT_INC(tcps_badrst);
576 syncache_badack(struct in_conninfo *inc)
579 struct syncache_head *sch;
581 sc = syncache_lookup(inc, &sch); /* returns locked sch */
582 SCH_LOCK_ASSERT(sch);
584 syncache_drop(sc, sch);
585 TCPSTAT_INC(tcps_sc_badack);
591 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
594 struct syncache_head *sch;
596 sc = syncache_lookup(inc, &sch); /* returns locked sch */
597 SCH_LOCK_ASSERT(sch);
601 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
602 if (ntohl(th->th_seq) != sc->sc_iss)
606 * If we've rertransmitted 3 times and this is our second error,
607 * we remove the entry. Otherwise, we allow it to continue on.
608 * This prevents us from incorrectly nuking an entry during a
609 * spurious network outage.
613 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
614 sc->sc_flags |= SCF_UNREACH;
617 syncache_drop(sc, sch);
618 TCPSTAT_INC(tcps_sc_unreach);
624 * Build a new TCP socket structure from a syncache entry.
626 static struct socket *
627 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
629 struct inpcb *inp = NULL;
635 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
638 * Ok, create the full blown connection, and set things up
639 * as they would have been set up if we had created the
640 * connection when the SYN arrived. If we can't create
641 * the connection, abort it.
643 so = sonewconn(lso, SS_ISCONNECTED);
646 * Drop the connection; we will either send a RST or
647 * have the peer retransmit its SYN again after its
650 TCPSTAT_INC(tcps_listendrop);
651 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
652 log(LOG_DEBUG, "%s; %s: Socket create failed "
653 "due to limits or memory shortage\n",
660 mac_socketpeer_set_from_mbuf(m, so);
664 inp->inp_inc.inc_fibnum = so->so_fibnum;
666 INP_HASH_WLOCK(&V_tcbinfo);
668 /* Insert new socket into PCB hash list. */
669 inp->inp_inc.inc_flags = sc->sc_inc.inc_flags;
671 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
672 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
674 inp->inp_vflag &= ~INP_IPV6;
675 inp->inp_vflag |= INP_IPV4;
677 inp->inp_laddr = sc->sc_inc.inc_laddr;
683 * Install in the reservation hash table for now, but don't yet
684 * install a connection group since the full 4-tuple isn't yet
687 inp->inp_lport = sc->sc_inc.inc_lport;
688 if ((error = in_pcbinshash_nopcbgroup(inp)) != 0) {
690 * Undo the assignments above if we failed to
691 * put the PCB on the hash lists.
694 if (sc->sc_inc.inc_flags & INC_ISIPV6)
695 inp->in6p_laddr = in6addr_any;
698 inp->inp_laddr.s_addr = INADDR_ANY;
700 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
701 log(LOG_DEBUG, "%s; %s: in_pcbinshash failed "
706 INP_HASH_WUNLOCK(&V_tcbinfo);
710 /* Copy old policy into new socket's. */
711 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
712 printf("syncache_socket: could not copy policy\n");
715 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
716 struct inpcb *oinp = sotoinpcb(lso);
717 struct in6_addr laddr6;
718 struct sockaddr_in6 sin6;
720 * Inherit socket options from the listening socket.
721 * Note that in6p_inputopts are not (and should not be)
722 * copied, since it stores previously received options and is
723 * used to detect if each new option is different than the
724 * previous one and hence should be passed to a user.
725 * If we copied in6p_inputopts, a user would not be able to
726 * receive options just after calling the accept system call.
728 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
729 if (oinp->in6p_outputopts)
730 inp->in6p_outputopts =
731 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
733 sin6.sin6_family = AF_INET6;
734 sin6.sin6_len = sizeof(sin6);
735 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
736 sin6.sin6_port = sc->sc_inc.inc_fport;
737 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
738 laddr6 = inp->in6p_laddr;
739 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
740 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
741 if ((error = in6_pcbconnect_mbuf(inp, (struct sockaddr *)&sin6,
742 thread0.td_ucred, m)) != 0) {
743 inp->in6p_laddr = laddr6;
744 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
745 log(LOG_DEBUG, "%s; %s: in6_pcbconnect failed "
750 INP_HASH_WUNLOCK(&V_tcbinfo);
753 /* Override flowlabel from in6_pcbconnect. */
754 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
755 inp->inp_flow |= sc->sc_flowlabel;
758 #if defined(INET) && defined(INET6)
763 struct in_addr laddr;
764 struct sockaddr_in sin;
766 inp->inp_options = (m) ? ip_srcroute(m) : NULL;
768 if (inp->inp_options == NULL) {
769 inp->inp_options = sc->sc_ipopts;
770 sc->sc_ipopts = NULL;
773 sin.sin_family = AF_INET;
774 sin.sin_len = sizeof(sin);
775 sin.sin_addr = sc->sc_inc.inc_faddr;
776 sin.sin_port = sc->sc_inc.inc_fport;
777 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
778 laddr = inp->inp_laddr;
779 if (inp->inp_laddr.s_addr == INADDR_ANY)
780 inp->inp_laddr = sc->sc_inc.inc_laddr;
781 if ((error = in_pcbconnect_mbuf(inp, (struct sockaddr *)&sin,
782 thread0.td_ucred, m)) != 0) {
783 inp->inp_laddr = laddr;
784 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
785 log(LOG_DEBUG, "%s; %s: in_pcbconnect failed "
790 INP_HASH_WUNLOCK(&V_tcbinfo);
795 INP_HASH_WUNLOCK(&V_tcbinfo);
797 tp->t_state = TCPS_SYN_RECEIVED;
798 tp->iss = sc->sc_iss;
799 tp->irs = sc->sc_irs;
802 tp->snd_wl1 = sc->sc_irs;
803 tp->snd_max = tp->iss + 1;
804 tp->snd_nxt = tp->iss + 1;
805 tp->rcv_up = sc->sc_irs + 1;
806 tp->rcv_wnd = sc->sc_wnd;
807 tp->rcv_adv += tp->rcv_wnd;
808 tp->last_ack_sent = tp->rcv_nxt;
810 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
811 if (sc->sc_flags & SCF_NOOPT)
812 tp->t_flags |= TF_NOOPT;
814 if (sc->sc_flags & SCF_WINSCALE) {
815 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
816 tp->snd_scale = sc->sc_requested_s_scale;
817 tp->request_r_scale = sc->sc_requested_r_scale;
819 if (sc->sc_flags & SCF_TIMESTAMP) {
820 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
821 tp->ts_recent = sc->sc_tsreflect;
822 tp->ts_recent_age = tcp_ts_getticks();
823 tp->ts_offset = sc->sc_tsoff;
826 if (sc->sc_flags & SCF_SIGNATURE)
827 tp->t_flags |= TF_SIGNATURE;
829 if (sc->sc_flags & SCF_SACK)
830 tp->t_flags |= TF_SACK_PERMIT;
833 if (sc->sc_flags & SCF_ECN)
834 tp->t_flags |= TF_ECN_PERMIT;
837 * Set up MSS and get cached values from tcp_hostcache.
838 * This might overwrite some of the defaults we just set.
840 tcp_mss(tp, sc->sc_peer_mss);
843 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
844 * NB: sc_rxmits counts all SYN,ACK transmits, not just retransmits.
846 if (sc->sc_rxmits > 1)
847 tp->snd_cwnd = tp->t_maxseg;
850 * Copy and activate timers.
852 tp->t_keepinit = sototcpcb(lso)->t_keepinit;
853 tp->t_keepidle = sototcpcb(lso)->t_keepidle;
854 tp->t_keepintvl = sototcpcb(lso)->t_keepintvl;
855 tp->t_keepcnt = sototcpcb(lso)->t_keepcnt;
856 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
860 TCPSTAT_INC(tcps_accepts);
872 * This function gets called when we receive an ACK for a
873 * socket in the LISTEN state. We look up the connection
874 * in the syncache, and if its there, we pull it out of
875 * the cache and turn it into a full-blown connection in
876 * the SYN-RECEIVED state.
879 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
880 struct socket **lsop, struct mbuf *m)
883 struct syncache_head *sch;
888 * Global TCP locks are held because we manipulate the PCB lists
889 * and create a new socket.
891 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
892 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
893 ("%s: can handle only ACK", __func__));
895 sc = syncache_lookup(inc, &sch); /* returns locked sch */
896 SCH_LOCK_ASSERT(sch);
899 * There is no syncache entry, so see if this ACK is
900 * a returning syncookie. To do this, first:
901 * A. See if this socket has had a syncache entry dropped in
902 * the past. We don't want to accept a bogus syncookie
903 * if we've never received a SYN.
904 * B. check that the syncookie is valid. If it is, then
905 * cobble up a fake syncache entry, and return.
907 if (!V_tcp_syncookies) {
909 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
910 log(LOG_DEBUG, "%s; %s: Spurious ACK, "
911 "segment rejected (syncookies disabled)\n",
915 bzero(&scs, sizeof(scs));
916 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
919 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
920 log(LOG_DEBUG, "%s; %s: Segment failed "
921 "SYNCOOKIE authentication, segment rejected "
922 "(probably spoofed)\n", s, __func__);
926 /* Pull out the entry to unlock the bucket row. */
927 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
929 V_tcp_syncache.cache_count--;
934 * Segment validation:
935 * ACK must match our initial sequence number + 1 (the SYN|ACK).
937 if (th->th_ack != sc->sc_iss + 1 && !TOEPCB_ISSET(sc)) {
938 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
939 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment "
940 "rejected\n", s, __func__, th->th_ack, sc->sc_iss);
945 * The SEQ must fall in the window starting at the received
946 * initial receive sequence number + 1 (the SYN).
948 if ((SEQ_LEQ(th->th_seq, sc->sc_irs) ||
949 SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) &&
951 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
952 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment "
953 "rejected\n", s, __func__, th->th_seq, sc->sc_irs);
957 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
958 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
959 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
960 "segment rejected\n", s, __func__);
964 * If timestamps were negotiated the reflected timestamp
965 * must be equal to what we actually sent in the SYN|ACK.
967 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts &&
969 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
970 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
971 "segment rejected\n",
972 s, __func__, to->to_tsecr, sc->sc_ts);
976 *lsop = syncache_socket(sc, *lsop, m);
979 TCPSTAT_INC(tcps_sc_aborted);
981 TCPSTAT_INC(tcps_sc_completed);
983 /* how do we find the inp for the new socket? */
988 if (sc != NULL && sc != &scs)
997 tcp_offload_syncache_expand(struct in_conninfo *inc, struct toeopt *toeo,
998 struct tcphdr *th, struct socket **lsop, struct mbuf *m)
1003 bzero(&to, sizeof(struct tcpopt));
1004 to.to_mss = toeo->to_mss;
1005 to.to_wscale = toeo->to_wscale;
1006 to.to_flags = toeo->to_flags;
1008 INP_INFO_WLOCK(&V_tcbinfo);
1009 rc = syncache_expand(inc, &to, th, lsop, m);
1010 INP_INFO_WUNLOCK(&V_tcbinfo);
1016 * Given a LISTEN socket and an inbound SYN request, add
1017 * this to the syn cache, and send back a segment:
1018 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
1021 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
1022 * Doing so would require that we hold onto the data and deliver it
1023 * to the application. However, if we are the target of a SYN-flood
1024 * DoS attack, an attacker could send data which would eventually
1025 * consume all available buffer space if it were ACKed. By not ACKing
1026 * the data, we avoid this DoS scenario.
1029 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1030 struct inpcb *inp, struct socket **lsop, struct mbuf *m,
1031 struct toe_usrreqs *tu, void *toepcb)
1035 struct syncache *sc = NULL;
1036 struct syncache_head *sch;
1037 struct mbuf *ipopts = NULL;
1040 int win, sb_hiwat, ip_ttl, ip_tos;
1043 int autoflowlabel = 0;
1046 struct label *maclabel;
1048 struct syncache scs;
1051 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1052 INP_WLOCK_ASSERT(inp); /* listen socket */
1053 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
1054 ("%s: unexpected tcp flags", __func__));
1057 * Combine all so/tp operations very early to drop the INP lock as
1062 cred = crhold(so->so_cred);
1065 if ((inc->inc_flags & INC_ISIPV6) &&
1066 (inp->inp_flags & IN6P_AUTOFLOWLABEL))
1069 ip_ttl = inp->inp_ip_ttl;
1070 ip_tos = inp->inp_ip_tos;
1071 win = sbspace(&so->so_rcv);
1072 sb_hiwat = so->so_rcv.sb_hiwat;
1073 ltflags = (tp->t_flags & (TF_NOOPT | TF_SIGNATURE));
1075 /* By the time we drop the lock these should no longer be used. */
1080 if (mac_syncache_init(&maclabel) != 0) {
1082 INP_INFO_WUNLOCK(&V_tcbinfo);
1085 mac_syncache_create(maclabel, inp);
1088 INP_INFO_WUNLOCK(&V_tcbinfo);
1091 * Remember the IP options, if any.
1094 if (!(inc->inc_flags & INC_ISIPV6))
1097 ipopts = (m) ? ip_srcroute(m) : NULL;
1103 * See if we already have an entry for this connection.
1104 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1106 * XXX: should the syncache be re-initialized with the contents
1107 * of the new SYN here (which may have different options?)
1109 * XXX: We do not check the sequence number to see if this is a
1110 * real retransmit or a new connection attempt. The question is
1111 * how to handle such a case; either ignore it as spoofed, or
1112 * drop the current entry and create a new one?
1114 sc = syncache_lookup(inc, &sch); /* returns locked entry */
1115 SCH_LOCK_ASSERT(sch);
1117 #ifndef TCP_OFFLOAD_DISABLE
1119 sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT,
1122 TCPSTAT_INC(tcps_sc_dupsyn);
1125 * If we were remembering a previous source route,
1126 * forget it and use the new one we've been given.
1129 (void) m_free(sc->sc_ipopts);
1130 sc->sc_ipopts = ipopts;
1133 * Update timestamp if present.
1135 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
1136 sc->sc_tsreflect = to->to_tsval;
1138 sc->sc_flags &= ~SCF_TIMESTAMP;
1141 * Since we have already unconditionally allocated label
1142 * storage, free it up. The syncache entry will already
1143 * have an initialized label we can use.
1145 mac_syncache_destroy(&maclabel);
1147 /* Retransmit SYN|ACK and reset retransmit count. */
1148 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
1149 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
1150 "resetting timer and retransmitting SYN|ACK\n",
1154 if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) {
1156 syncache_timeout(sc, sch, 1);
1157 TCPSTAT_INC(tcps_sndacks);
1158 TCPSTAT_INC(tcps_sndtotal);
1164 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1167 * The zone allocator couldn't provide more entries.
1168 * Treat this as if the cache was full; drop the oldest
1169 * entry and insert the new one.
1171 TCPSTAT_INC(tcps_sc_zonefail);
1172 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1173 syncache_drop(sc, sch);
1174 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1176 if (V_tcp_syncookies) {
1177 bzero(&scs, sizeof(scs));
1182 (void) m_free(ipopts);
1189 * Fill in the syncache values.
1192 sc->sc_label = maclabel;
1196 sc->sc_ipopts = ipopts;
1197 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1199 if (!(inc->inc_flags & INC_ISIPV6))
1202 sc->sc_ip_tos = ip_tos;
1203 sc->sc_ip_ttl = ip_ttl;
1205 #ifndef TCP_OFFLOAD_DISABLE
1207 sc->sc_toepcb = toepcb;
1209 sc->sc_irs = th->th_seq;
1210 sc->sc_iss = arc4random();
1212 sc->sc_flowlabel = 0;
1215 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
1216 * win was derived from socket earlier in the function.
1219 win = imin(win, TCP_MAXWIN);
1222 if (V_tcp_do_rfc1323) {
1224 * A timestamp received in a SYN makes
1225 * it ok to send timestamp requests and replies.
1227 if (to->to_flags & TOF_TS) {
1228 sc->sc_tsreflect = to->to_tsval;
1229 sc->sc_ts = tcp_ts_getticks();
1230 sc->sc_flags |= SCF_TIMESTAMP;
1232 if (to->to_flags & TOF_SCALE) {
1236 * Pick the smallest possible scaling factor that
1237 * will still allow us to scale up to sb_max, aka
1238 * kern.ipc.maxsockbuf.
1240 * We do this because there are broken firewalls that
1241 * will corrupt the window scale option, leading to
1242 * the other endpoint believing that our advertised
1243 * window is unscaled. At scale factors larger than
1244 * 5 the unscaled window will drop below 1500 bytes,
1245 * leading to serious problems when traversing these
1248 * With the default maxsockbuf of 256K, a scale factor
1249 * of 3 will be chosen by this algorithm. Those who
1250 * choose a larger maxsockbuf should watch out
1251 * for the compatiblity problems mentioned above.
1253 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1254 * or <SYN,ACK>) segment itself is never scaled.
1256 while (wscale < TCP_MAX_WINSHIFT &&
1257 (TCP_MAXWIN << wscale) < sb_max)
1259 sc->sc_requested_r_scale = wscale;
1260 sc->sc_requested_s_scale = to->to_wscale;
1261 sc->sc_flags |= SCF_WINSCALE;
1264 #ifdef TCP_SIGNATURE
1266 * If listening socket requested TCP digests, and received SYN
1267 * contains the option, flag this in the syncache so that
1268 * syncache_respond() will do the right thing with the SYN+ACK.
1269 * XXX: Currently we always record the option by default and will
1270 * attempt to use it in syncache_respond().
1272 if (to->to_flags & TOF_SIGNATURE || ltflags & TF_SIGNATURE)
1273 sc->sc_flags |= SCF_SIGNATURE;
1275 if (to->to_flags & TOF_SACKPERM)
1276 sc->sc_flags |= SCF_SACK;
1277 if (to->to_flags & TOF_MSS)
1278 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */
1279 if (ltflags & TF_NOOPT)
1280 sc->sc_flags |= SCF_NOOPT;
1281 if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn)
1282 sc->sc_flags |= SCF_ECN;
1284 if (V_tcp_syncookies) {
1285 syncookie_generate(sch, sc, &flowtmp);
1288 sc->sc_flowlabel = flowtmp;
1294 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1300 * Do a standard 3-way handshake.
1302 if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) {
1303 if (V_tcp_syncookies && V_tcp_syncookiesonly && sc != &scs)
1305 else if (sc != &scs)
1306 syncache_insert(sc, sch); /* locks and unlocks sch */
1307 TCPSTAT_INC(tcps_sndacks);
1308 TCPSTAT_INC(tcps_sndtotal);
1312 TCPSTAT_INC(tcps_sc_dropped);
1320 mac_syncache_destroy(&maclabel);
1330 syncache_respond(struct syncache *sc)
1332 struct ip *ip = NULL;
1334 struct tcphdr *th = NULL;
1335 int optlen, error = 0; /* Make compiler happy */
1336 u_int16_t hlen, tlen, mssopt;
1339 struct ip6_hdr *ip6 = NULL;
1344 (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) :
1347 tlen = hlen + sizeof(struct tcphdr);
1349 /* Determine MSS we advertize to other end of connection. */
1350 mssopt = tcp_mssopt(&sc->sc_inc);
1351 if (sc->sc_peer_mss)
1352 mssopt = max( min(sc->sc_peer_mss, mssopt), V_tcp_minmss);
1354 /* XXX: Assume that the entire packet will fit in a header mbuf. */
1355 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1356 ("syncache: mbuf too small"));
1358 /* Create the IP+TCP header from scratch. */
1359 m = m_gethdr(M_DONTWAIT, MT_DATA);
1363 mac_syncache_create_mbuf(sc->sc_label, m);
1365 m->m_data += max_linkhdr;
1367 m->m_pkthdr.len = tlen;
1368 m->m_pkthdr.rcvif = NULL;
1371 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1372 ip6 = mtod(m, struct ip6_hdr *);
1373 ip6->ip6_vfc = IPV6_VERSION;
1374 ip6->ip6_nxt = IPPROTO_TCP;
1375 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1376 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1377 ip6->ip6_plen = htons(tlen - hlen);
1378 /* ip6_hlim is set after checksum */
1379 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1380 ip6->ip6_flow |= sc->sc_flowlabel;
1382 th = (struct tcphdr *)(ip6 + 1);
1385 #if defined(INET6) && defined(INET)
1390 ip = mtod(m, struct ip *);
1391 ip->ip_v = IPVERSION;
1392 ip->ip_hl = sizeof(struct ip) >> 2;
1397 ip->ip_p = IPPROTO_TCP;
1398 ip->ip_src = sc->sc_inc.inc_laddr;
1399 ip->ip_dst = sc->sc_inc.inc_faddr;
1400 ip->ip_ttl = sc->sc_ip_ttl;
1401 ip->ip_tos = sc->sc_ip_tos;
1404 * See if we should do MTU discovery. Route lookups are
1405 * expensive, so we will only unset the DF bit if:
1407 * 1) path_mtu_discovery is disabled
1408 * 2) the SCF_UNREACH flag has been set
1410 if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1411 ip->ip_off |= IP_DF;
1413 th = (struct tcphdr *)(ip + 1);
1416 th->th_sport = sc->sc_inc.inc_lport;
1417 th->th_dport = sc->sc_inc.inc_fport;
1419 th->th_seq = htonl(sc->sc_iss);
1420 th->th_ack = htonl(sc->sc_irs + 1);
1421 th->th_off = sizeof(struct tcphdr) >> 2;
1423 th->th_flags = TH_SYN|TH_ACK;
1424 th->th_win = htons(sc->sc_wnd);
1427 if (sc->sc_flags & SCF_ECN) {
1428 th->th_flags |= TH_ECE;
1429 TCPSTAT_INC(tcps_ecn_shs);
1432 /* Tack on the TCP options. */
1433 if ((sc->sc_flags & SCF_NOOPT) == 0) {
1437 to.to_flags = TOF_MSS;
1438 if (sc->sc_flags & SCF_WINSCALE) {
1439 to.to_wscale = sc->sc_requested_r_scale;
1440 to.to_flags |= TOF_SCALE;
1442 if (sc->sc_flags & SCF_TIMESTAMP) {
1443 /* Virgin timestamp or TCP cookie enhanced one. */
1444 to.to_tsval = sc->sc_ts;
1445 to.to_tsecr = sc->sc_tsreflect;
1446 to.to_flags |= TOF_TS;
1448 if (sc->sc_flags & SCF_SACK)
1449 to.to_flags |= TOF_SACKPERM;
1450 #ifdef TCP_SIGNATURE
1451 if (sc->sc_flags & SCF_SIGNATURE)
1452 to.to_flags |= TOF_SIGNATURE;
1454 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1456 /* Adjust headers by option size. */
1457 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1459 m->m_pkthdr.len += optlen;
1461 #ifdef TCP_SIGNATURE
1462 if (sc->sc_flags & SCF_SIGNATURE)
1463 tcp_signature_compute(m, 0, 0, optlen,
1464 to.to_signature, IPSEC_DIR_OUTBOUND);
1467 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1468 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1471 ip->ip_len += optlen;
1475 M_SETFIB(m, sc->sc_inc.inc_fibnum);
1476 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1478 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1479 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1480 th->th_sum = in6_cksum_pseudo(ip6, tlen + optlen - hlen,
1482 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1483 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1486 #if defined(INET6) && defined(INET)
1491 m->m_pkthdr.csum_flags = CSUM_TCP;
1492 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1493 htons(tlen + optlen - hlen + IPPROTO_TCP));
1494 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1501 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1502 struct inpcb *inp, struct socket **lsop, struct mbuf *m)
1504 _syncache_add(inc, to, th, inp, lsop, m, NULL, NULL);
1508 tcp_offload_syncache_add(struct in_conninfo *inc, struct toeopt *toeo,
1509 struct tcphdr *th, struct inpcb *inp, struct socket **lsop,
1510 struct toe_usrreqs *tu, void *toepcb)
1514 bzero(&to, sizeof(struct tcpopt));
1515 to.to_mss = toeo->to_mss;
1516 to.to_wscale = toeo->to_wscale;
1517 to.to_flags = toeo->to_flags;
1519 INP_INFO_WLOCK(&V_tcbinfo);
1522 _syncache_add(inc, &to, th, inp, lsop, NULL, tu, toepcb);
1526 * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1527 * receive and to be able to handle SYN floods from bogus source addresses
1528 * (where we will never receive any reply). SYN floods try to exhaust all
1529 * our memory and available slots in the SYN cache table to cause a denial
1530 * of service to legitimate users of the local host.
1532 * The idea of SYN cookies is to encode and include all necessary information
1533 * about the connection setup state within the SYN-ACK we send back and thus
1534 * to get along without keeping any local state until the ACK to the SYN-ACK
1535 * arrives (if ever). Everything we need to know should be available from
1536 * the information we encoded in the SYN-ACK.
1538 * More information about the theory behind SYN cookies and its first
1539 * discussion and specification can be found at:
1540 * http://cr.yp.to/syncookies.html (overview)
1541 * http://cr.yp.to/syncookies/archive (gory details)
1543 * This implementation extends the orginal idea and first implementation
1544 * of FreeBSD by using not only the initial sequence number field to store
1545 * information but also the timestamp field if present. This way we can
1546 * keep track of the entire state we need to know to recreate the session in
1547 * its original form. Almost all TCP speakers implement RFC1323 timestamps
1548 * these days. For those that do not we still have to live with the known
1549 * shortcomings of the ISN only SYN cookies.
1553 * Initial sequence number we send:
1554 * 31|................................|0
1555 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1556 * D = MD5 Digest (first dword)
1558 * R = Rotation of secret
1559 * P = Odd or Even secret
1561 * The MD5 Digest is computed with over following parameters:
1562 * a) randomly rotated secret
1563 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1564 * c) the received initial sequence number from remote host
1565 * d) the rotation offset and odd/even bit
1567 * Timestamp we send:
1568 * 31|................................|0
1569 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1570 * D = MD5 Digest (third dword) (only as filler)
1571 * S = Requested send window scale
1572 * R = Requested receive window scale
1574 * 5 = TCP-MD5 enabled (not implemented yet)
1575 * XORed with MD5 Digest (forth dword)
1577 * The timestamp isn't cryptographically secure and doesn't need to be.
1578 * The double use of the MD5 digest dwords ties it to a specific remote/
1579 * local host/port, remote initial sequence number and our local time
1580 * limited secret. A received timestamp is reverted (XORed) and then
1581 * the contained MD5 dword is compared to the computed one to ensure the
1582 * timestamp belongs to the SYN-ACK we sent. The other parameters may
1583 * have been tampered with but this isn't different from supplying bogus
1584 * values in the SYN in the first place.
1586 * Some problems with SYN cookies remain however:
1587 * Consider the problem of a recreated (and retransmitted) cookie. If the
1588 * original SYN was accepted, the connection is established. The second
1589 * SYN is inflight, and if it arrives with an ISN that falls within the
1590 * receive window, the connection is killed.
1593 * A heuristic to determine when to accept syn cookies is not necessary.
1594 * An ACK flood would cause the syncookie verification to be attempted,
1595 * but a SYN flood causes syncookies to be generated. Both are of equal
1596 * cost, so there's no point in trying to optimize the ACK flood case.
1597 * Also, if you don't process certain ACKs for some reason, then all someone
1598 * would have to do is launch a SYN and ACK flood at the same time, which
1599 * would stop cookie verification and defeat the entire purpose of syncookies.
1601 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1604 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1605 u_int32_t *flowlabel)
1608 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1611 u_int off, pmss, mss;
1614 SCH_LOCK_ASSERT(sch);
1616 /* Which of the two secrets to use. */
1617 secbits = sch->sch_oddeven ?
1618 sch->sch_secbits_odd : sch->sch_secbits_even;
1620 /* Reseed secret if too old. */
1621 if (sch->sch_reseed < time_uptime) {
1622 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */
1623 secbits = sch->sch_oddeven ?
1624 sch->sch_secbits_odd : sch->sch_secbits_even;
1625 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1626 secbits[i] = arc4random();
1627 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1630 /* Secret rotation offset. */
1631 off = sc->sc_iss & 0x7; /* iss was randomized before */
1633 /* Maximum segment size calculation. */
1635 max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), V_tcp_minmss);
1636 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1637 if (tcp_sc_msstab[mss] <= pmss)
1640 /* Fold parameters and MD5 digest into the ISN we will send. */
1641 data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1642 data |= off << 1; /* secret offset, derived from iss, 3 bits */
1643 data |= mss << 4; /* mss, 3 bits */
1646 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1647 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1648 MD5Update(&ctx, secbits, off);
1649 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1650 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1651 MD5Update(&ctx, &data, sizeof(data));
1652 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1654 data |= (md5_buffer[0] << 7);
1658 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1661 /* Additional parameters are stored in the timestamp if present. */
1662 if (sc->sc_flags & SCF_TIMESTAMP) {
1663 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1664 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1665 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */
1666 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */
1667 data |= md5_buffer[2] << 10; /* more digest bits */
1668 data ^= md5_buffer[3];
1670 sc->sc_tsoff = data - tcp_ts_getticks(); /* after XOR */
1673 TCPSTAT_INC(tcps_sc_sendcookie);
1676 static struct syncache *
1677 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1678 struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1682 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1686 int off, mss, wnd, flags;
1688 SCH_LOCK_ASSERT(sch);
1691 * Pull information out of SYN-ACK/ACK and
1692 * revert sequence number advances.
1694 ack = th->th_ack - 1;
1695 seq = th->th_seq - 1;
1696 off = (ack >> 1) & 0x7;
1697 mss = (ack >> 4) & 0x7;
1700 /* Which of the two secrets to use. */
1701 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1704 * The secret wasn't updated for the lifetime of a syncookie,
1705 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1707 if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) {
1711 /* Recompute the digest so we can compare it. */
1713 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1714 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1715 MD5Update(&ctx, secbits, off);
1716 MD5Update(&ctx, inc, sizeof(*inc));
1717 MD5Update(&ctx, &seq, sizeof(seq));
1718 MD5Update(&ctx, &flags, sizeof(flags));
1719 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1721 /* Does the digest part of or ACK'ed ISS match? */
1722 if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1725 /* Does the digest part of our reflected timestamp match? */
1726 if (to->to_flags & TOF_TS) {
1727 data = md5_buffer[3] ^ to->to_tsecr;
1728 if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1732 /* Fill in the syncache values. */
1733 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1734 sc->sc_ipopts = NULL;
1740 if (inc->inc_flags & INC_ISIPV6) {
1741 if (sotoinpcb(so)->inp_flags & IN6P_AUTOFLOWLABEL)
1742 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1746 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1747 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1750 /* Additional parameters that were encoded in the timestamp. */
1752 sc->sc_flags |= SCF_TIMESTAMP;
1753 sc->sc_tsreflect = to->to_tsval;
1754 sc->sc_ts = to->to_tsecr;
1755 sc->sc_tsoff = to->to_tsecr - tcp_ts_getticks();
1756 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1757 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1758 sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1760 sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1762 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1763 sc->sc_flags |= SCF_WINSCALE;
1765 sc->sc_flags |= SCF_NOOPT;
1767 wnd = sbspace(&so->so_rcv);
1769 wnd = imin(wnd, TCP_MAXWIN);
1773 sc->sc_peer_mss = tcp_sc_msstab[mss];
1775 TCPSTAT_INC(tcps_sc_recvcookie);
1780 * Returns the current number of syncache entries. This number
1781 * will probably change before you get around to calling
1786 syncache_pcbcount(void)
1788 struct syncache_head *sch;
1791 for (count = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1792 /* No need to lock for a read. */
1793 sch = &V_tcp_syncache.hashbase[i];
1794 count += sch->sch_length;
1800 * Exports the syncache entries to userland so that netstat can display
1801 * them alongside the other sockets. This function is intended to be
1802 * called only from tcp_pcblist.
1804 * Due to concurrency on an active system, the number of pcbs exported
1805 * may have no relation to max_pcbs. max_pcbs merely indicates the
1806 * amount of space the caller allocated for this function to use.
1809 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
1812 struct syncache *sc;
1813 struct syncache_head *sch;
1814 int count, error, i;
1816 for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1817 sch = &V_tcp_syncache.hashbase[i];
1819 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
1820 if (count >= max_pcbs) {
1824 if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0)
1826 bzero(&xt, sizeof(xt));
1827 xt.xt_len = sizeof(xt);
1828 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1829 xt.xt_inp.inp_vflag = INP_IPV6;
1831 xt.xt_inp.inp_vflag = INP_IPV4;
1832 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
1833 xt.xt_tp.t_inpcb = &xt.xt_inp;
1834 xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
1835 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1836 xt.xt_socket.xso_len = sizeof (struct xsocket);
1837 xt.xt_socket.so_type = SOCK_STREAM;
1838 xt.xt_socket.so_state = SS_ISCONNECTING;
1839 error = SYSCTL_OUT(req, &xt, sizeof xt);
1849 *pcbs_exported = count;