2 * Copyright (c) 2001 McAfee, Inc.
3 * Copyright (c) 2006 Andre Oppermann, Internet Business Solutions AG
6 * This software was developed for the FreeBSD Project by Jonathan Lemon
7 * and McAfee Research, the Security Research Division of McAfee, Inc. under
8 * DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
9 * DARPA CHATS research program.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/sysctl.h>
44 #include <sys/limits.h>
46 #include <sys/mutex.h>
47 #include <sys/malloc.h>
50 #include <sys/proc.h> /* for proc0 declaration */
51 #include <sys/random.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
54 #include <sys/syslog.h>
55 #include <sys/ucred.h>
60 #include <net/route.h>
63 #include <netinet/in.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/ip.h>
66 #include <netinet/in_var.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/ip_var.h>
69 #include <netinet/ip_options.h>
71 #include <netinet/ip6.h>
72 #include <netinet/icmp6.h>
73 #include <netinet6/nd6.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/in6_pcb.h>
77 #include <netinet/tcp.h>
78 #include <netinet/tcp_fsm.h>
79 #include <netinet/tcp_seq.h>
80 #include <netinet/tcp_timer.h>
81 #include <netinet/tcp_var.h>
82 #include <netinet/tcp_syncache.h>
83 #include <netinet/tcp_offload.h>
85 #include <netinet6/tcp6_var.h>
89 #include <netipsec/ipsec.h>
91 #include <netipsec/ipsec6.h>
93 #include <netipsec/key.h>
96 #include <machine/in_cksum.h>
98 #include <security/mac/mac_framework.h>
100 static VNET_DEFINE(int, tcp_syncookies) = 1;
101 #define V_tcp_syncookies VNET(tcp_syncookies)
102 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
103 &VNET_NAME(tcp_syncookies), 0,
104 "Use TCP SYN cookies if the syncache overflows");
106 static VNET_DEFINE(int, tcp_syncookiesonly) = 0;
107 #define V_tcp_syncookiesonly VNET(tcp_syncookiesonly)
108 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW,
109 &VNET_NAME(tcp_syncookiesonly), 0,
110 "Use only TCP SYN cookies");
112 #ifdef TCP_OFFLOAD_DISABLE
113 #define TOEPCB_ISSET(sc) (0)
115 #define TOEPCB_ISSET(sc) ((sc)->sc_toepcb != NULL)
118 static void syncache_drop(struct syncache *, struct syncache_head *);
119 static void syncache_free(struct syncache *);
120 static void syncache_insert(struct syncache *, struct syncache_head *);
121 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
122 static int syncache_respond(struct syncache *);
123 static struct socket *syncache_socket(struct syncache *, struct socket *,
125 static void syncache_timeout(struct syncache *sc, struct syncache_head *sch,
127 static void syncache_timer(void *);
128 static void syncookie_generate(struct syncache_head *, struct syncache *,
130 static struct syncache
131 *syncookie_lookup(struct in_conninfo *, struct syncache_head *,
132 struct syncache *, struct tcpopt *, struct tcphdr *,
136 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
137 * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds,
138 * the odds are that the user has given up attempting to connect by then.
140 #define SYNCACHE_MAXREXMTS 3
142 /* Arbitrary values */
143 #define TCP_SYNCACHE_HASHSIZE 512
144 #define TCP_SYNCACHE_BUCKETLIMIT 30
146 static VNET_DEFINE(struct tcp_syncache, tcp_syncache);
147 #define V_tcp_syncache VNET(tcp_syncache)
149 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
151 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
152 &VNET_NAME(tcp_syncache.bucket_limit), 0,
153 "Per-bucket hash limit for syncache");
155 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
156 &VNET_NAME(tcp_syncache.cache_limit), 0,
157 "Overall entry limit for syncache");
159 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
160 &VNET_NAME(tcp_syncache.cache_count), 0,
161 "Current number of entries in syncache");
163 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
164 &VNET_NAME(tcp_syncache.hashsize), 0,
165 "Size of TCP syncache hashtable");
167 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
168 &VNET_NAME(tcp_syncache.rexmt_limit), 0,
169 "Limit on SYN/ACK retransmissions");
171 VNET_DEFINE(int, tcp_sc_rst_sock_fail) = 1;
172 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail,
173 CTLFLAG_RW, &VNET_NAME(tcp_sc_rst_sock_fail), 0,
174 "Send reset on socket allocation failure");
176 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
178 #define SYNCACHE_HASH(inc, mask) \
179 ((V_tcp_syncache.hash_secret ^ \
180 (inc)->inc_faddr.s_addr ^ \
181 ((inc)->inc_faddr.s_addr >> 16) ^ \
182 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
184 #define SYNCACHE_HASH6(inc, mask) \
185 ((V_tcp_syncache.hash_secret ^ \
186 (inc)->inc6_faddr.s6_addr32[0] ^ \
187 (inc)->inc6_faddr.s6_addr32[3] ^ \
188 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
190 #define ENDPTS_EQ(a, b) ( \
191 (a)->ie_fport == (b)->ie_fport && \
192 (a)->ie_lport == (b)->ie_lport && \
193 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
194 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
197 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
199 #define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx)
200 #define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx)
201 #define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED)
204 * Requires the syncache entry to be already removed from the bucket list.
207 syncache_free(struct syncache *sc)
211 (void) m_free(sc->sc_ipopts);
215 mac_syncache_destroy(&sc->sc_label);
218 uma_zfree(V_tcp_syncache.zone, sc);
226 V_tcp_syncache.cache_count = 0;
227 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
228 V_tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
229 V_tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
230 V_tcp_syncache.hash_secret = arc4random();
232 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
233 &V_tcp_syncache.hashsize);
234 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
235 &V_tcp_syncache.bucket_limit);
236 if (!powerof2(V_tcp_syncache.hashsize) ||
237 V_tcp_syncache.hashsize == 0) {
238 printf("WARNING: syncache hash size is not a power of 2.\n");
239 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
241 V_tcp_syncache.hashmask = V_tcp_syncache.hashsize - 1;
244 V_tcp_syncache.cache_limit =
245 V_tcp_syncache.hashsize * V_tcp_syncache.bucket_limit;
246 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
247 &V_tcp_syncache.cache_limit);
249 /* Allocate the hash table. */
250 V_tcp_syncache.hashbase = malloc(V_tcp_syncache.hashsize *
251 sizeof(struct syncache_head), M_SYNCACHE, M_WAITOK | M_ZERO);
253 /* Initialize the hash buckets. */
254 for (i = 0; i < V_tcp_syncache.hashsize; i++) {
256 V_tcp_syncache.hashbase[i].sch_vnet = curvnet;
258 TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket);
259 mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
261 callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer,
262 &V_tcp_syncache.hashbase[i].sch_mtx, 0);
263 V_tcp_syncache.hashbase[i].sch_length = 0;
266 /* Create the syncache entry zone. */
267 V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
268 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
269 uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit);
274 syncache_destroy(void)
276 struct syncache_head *sch;
277 struct syncache *sc, *nsc;
280 /* Cleanup hash buckets: stop timers, free entries, destroy locks. */
281 for (i = 0; i < V_tcp_syncache.hashsize; i++) {
283 sch = &V_tcp_syncache.hashbase[i];
284 callout_drain(&sch->sch_timer);
287 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc)
288 syncache_drop(sc, sch);
290 KASSERT(TAILQ_EMPTY(&sch->sch_bucket),
291 ("%s: sch->sch_bucket not empty", __func__));
292 KASSERT(sch->sch_length == 0, ("%s: sch->sch_length %d not 0",
293 __func__, sch->sch_length));
294 mtx_destroy(&sch->sch_mtx);
297 KASSERT(V_tcp_syncache.cache_count == 0, ("%s: cache_count %d not 0",
298 __func__, V_tcp_syncache.cache_count));
300 /* Free the allocated global resources. */
301 uma_zdestroy(V_tcp_syncache.zone);
302 free(V_tcp_syncache.hashbase, M_SYNCACHE);
307 * Inserts a syncache entry into the specified bucket row.
308 * Locks and unlocks the syncache_head autonomously.
311 syncache_insert(struct syncache *sc, struct syncache_head *sch)
313 struct syncache *sc2;
318 * Make sure that we don't overflow the per-bucket limit.
319 * If the bucket is full, toss the oldest element.
321 if (sch->sch_length >= V_tcp_syncache.bucket_limit) {
322 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
323 ("sch->sch_length incorrect"));
324 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
325 syncache_drop(sc2, sch);
326 TCPSTAT_INC(tcps_sc_bucketoverflow);
329 /* Put it into the bucket. */
330 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
333 /* Reinitialize the bucket row's timer. */
334 if (sch->sch_length == 1)
335 sch->sch_nextc = ticks + INT_MAX;
336 syncache_timeout(sc, sch, 1);
340 V_tcp_syncache.cache_count++;
341 TCPSTAT_INC(tcps_sc_added);
345 * Remove and free entry from syncache bucket row.
346 * Expects locked syncache head.
349 syncache_drop(struct syncache *sc, struct syncache_head *sch)
352 SCH_LOCK_ASSERT(sch);
354 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
357 #ifndef TCP_OFFLOAD_DISABLE
359 sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb);
362 V_tcp_syncache.cache_count--;
366 * Engage/reengage time on bucket row.
369 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
371 sc->sc_rxttime = ticks +
372 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
374 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
375 sch->sch_nextc = sc->sc_rxttime;
377 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
378 syncache_timer, (void *)sch);
383 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
384 * If we have retransmitted an entry the maximum number of times, expire it.
385 * One separate timer for each bucket row.
388 syncache_timer(void *xsch)
390 struct syncache_head *sch = (struct syncache_head *)xsch;
391 struct syncache *sc, *nsc;
395 CURVNET_SET(sch->sch_vnet);
397 /* NB: syncache_head has already been locked by the callout. */
398 SCH_LOCK_ASSERT(sch);
401 * In the following cycle we may remove some entries and/or
402 * advance some timeouts, so re-initialize the bucket timer.
404 sch->sch_nextc = tick + INT_MAX;
406 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
408 * We do not check if the listen socket still exists
409 * and accept the case where the listen socket may be
410 * gone by the time we resend the SYN/ACK. We do
411 * not expect this to happens often. If it does,
412 * then the RST will be sent by the time the remote
413 * host does the SYN/ACK->ACK.
415 if (TSTMP_GT(sc->sc_rxttime, tick)) {
416 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
417 sch->sch_nextc = sc->sc_rxttime;
420 if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) {
421 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
422 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
423 "giving up and removing syncache entry\n",
427 syncache_drop(sc, sch);
428 TCPSTAT_INC(tcps_sc_stale);
431 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
432 log(LOG_DEBUG, "%s; %s: Response timeout, "
433 "retransmitting (%u) SYN|ACK\n",
434 s, __func__, sc->sc_rxmits);
438 (void) syncache_respond(sc);
439 TCPSTAT_INC(tcps_sc_retransmitted);
440 syncache_timeout(sc, sch, 0);
442 if (!TAILQ_EMPTY(&(sch)->sch_bucket))
443 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
444 syncache_timer, (void *)(sch));
449 * Find an entry in the syncache.
450 * Returns always with locked syncache_head plus a matching entry or NULL.
453 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
456 struct syncache_head *sch;
459 if (inc->inc_flags & INC_ISIPV6) {
460 sch = &V_tcp_syncache.hashbase[
461 SYNCACHE_HASH6(inc, V_tcp_syncache.hashmask)];
466 /* Circle through bucket row to find matching entry. */
467 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
468 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
474 sch = &V_tcp_syncache.hashbase[
475 SYNCACHE_HASH(inc, V_tcp_syncache.hashmask)];
480 /* Circle through bucket row to find matching entry. */
481 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
483 if (sc->sc_inc.inc_flags & INC_ISIPV6)
486 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
490 SCH_LOCK_ASSERT(*schp);
491 return (NULL); /* always returns with locked sch */
495 * This function is called when we get a RST for a
496 * non-existent connection, so that we can see if the
497 * connection is in the syn cache. If it is, zap it.
500 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
503 struct syncache_head *sch;
506 sc = syncache_lookup(inc, &sch); /* returns locked sch */
507 SCH_LOCK_ASSERT(sch);
510 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
511 * See RFC 793 page 65, section SEGMENT ARRIVES.
513 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
514 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
515 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
516 "FIN flag set, segment ignored\n", s, __func__);
517 TCPSTAT_INC(tcps_badrst);
522 * No corresponding connection was found in syncache.
523 * If syncookies are enabled and possibly exclusively
524 * used, or we are under memory pressure, a valid RST
525 * may not find a syncache entry. In that case we're
526 * done and no SYN|ACK retransmissions will happen.
527 * Otherwise the the RST was misdirected or spoofed.
530 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
531 log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
532 "syncache entry (possibly syncookie only), "
533 "segment ignored\n", s, __func__);
534 TCPSTAT_INC(tcps_badrst);
539 * If the RST bit is set, check the sequence number to see
540 * if this is a valid reset segment.
542 * In all states except SYN-SENT, all reset (RST) segments
543 * are validated by checking their SEQ-fields. A reset is
544 * valid if its sequence number is in the window.
546 * The sequence number in the reset segment is normally an
547 * echo of our outgoing acknowlegement numbers, but some hosts
548 * send a reset with the sequence number at the rightmost edge
549 * of our receive window, and we have to handle this case.
551 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
552 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
553 syncache_drop(sc, sch);
554 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
555 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
556 "connection attempt aborted by remote endpoint\n",
558 TCPSTAT_INC(tcps_sc_reset);
560 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
561 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != "
562 "IRS %u (+WND %u), segment ignored\n",
563 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
564 TCPSTAT_INC(tcps_badrst);
574 syncache_badack(struct in_conninfo *inc)
577 struct syncache_head *sch;
579 sc = syncache_lookup(inc, &sch); /* returns locked sch */
580 SCH_LOCK_ASSERT(sch);
582 syncache_drop(sc, sch);
583 TCPSTAT_INC(tcps_sc_badack);
589 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
592 struct syncache_head *sch;
594 sc = syncache_lookup(inc, &sch); /* returns locked sch */
595 SCH_LOCK_ASSERT(sch);
599 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
600 if (ntohl(th->th_seq) != sc->sc_iss)
604 * If we've rertransmitted 3 times and this is our second error,
605 * we remove the entry. Otherwise, we allow it to continue on.
606 * This prevents us from incorrectly nuking an entry during a
607 * spurious network outage.
611 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
612 sc->sc_flags |= SCF_UNREACH;
615 syncache_drop(sc, sch);
616 TCPSTAT_INC(tcps_sc_unreach);
622 * Build a new TCP socket structure from a syncache entry.
624 static struct socket *
625 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
627 struct inpcb *inp = NULL;
632 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
635 * Ok, create the full blown connection, and set things up
636 * as they would have been set up if we had created the
637 * connection when the SYN arrived. If we can't create
638 * the connection, abort it.
640 so = sonewconn(lso, SS_ISCONNECTED);
643 * Drop the connection; we will either send a RST or
644 * have the peer retransmit its SYN again after its
647 TCPSTAT_INC(tcps_listendrop);
648 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
649 log(LOG_DEBUG, "%s; %s: Socket create failed "
650 "due to limits or memory shortage\n",
657 mac_socketpeer_set_from_mbuf(m, so);
661 inp->inp_inc.inc_fibnum = so->so_fibnum;
664 /* Insert new socket into PCB hash list. */
665 inp->inp_inc.inc_flags = sc->sc_inc.inc_flags;
667 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
668 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
670 inp->inp_vflag &= ~INP_IPV6;
671 inp->inp_vflag |= INP_IPV4;
673 inp->inp_laddr = sc->sc_inc.inc_laddr;
677 inp->inp_lport = sc->sc_inc.inc_lport;
678 if (in_pcbinshash(inp) != 0) {
680 * Undo the assignments above if we failed to
681 * put the PCB on the hash lists.
684 if (sc->sc_inc.inc_flags & INC_ISIPV6)
685 inp->in6p_laddr = in6addr_any;
688 inp->inp_laddr.s_addr = INADDR_ANY;
693 /* Copy old policy into new socket's. */
694 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
695 printf("syncache_socket: could not copy policy\n");
698 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
699 struct inpcb *oinp = sotoinpcb(lso);
700 struct in6_addr laddr6;
701 struct sockaddr_in6 sin6;
703 * Inherit socket options from the listening socket.
704 * Note that in6p_inputopts are not (and should not be)
705 * copied, since it stores previously received options and is
706 * used to detect if each new option is different than the
707 * previous one and hence should be passed to a user.
708 * If we copied in6p_inputopts, a user would not be able to
709 * receive options just after calling the accept system call.
711 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
712 if (oinp->in6p_outputopts)
713 inp->in6p_outputopts =
714 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
716 sin6.sin6_family = AF_INET6;
717 sin6.sin6_len = sizeof(sin6);
718 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
719 sin6.sin6_port = sc->sc_inc.inc_fport;
720 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
721 laddr6 = inp->in6p_laddr;
722 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
723 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
724 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
726 inp->in6p_laddr = laddr6;
729 /* Override flowlabel from in6_pcbconnect. */
730 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
731 inp->inp_flow |= sc->sc_flowlabel;
735 struct in_addr laddr;
736 struct sockaddr_in sin;
738 inp->inp_options = (m) ? ip_srcroute(m) : NULL;
740 if (inp->inp_options == NULL) {
741 inp->inp_options = sc->sc_ipopts;
742 sc->sc_ipopts = NULL;
745 sin.sin_family = AF_INET;
746 sin.sin_len = sizeof(sin);
747 sin.sin_addr = sc->sc_inc.inc_faddr;
748 sin.sin_port = sc->sc_inc.inc_fport;
749 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
750 laddr = inp->inp_laddr;
751 if (inp->inp_laddr.s_addr == INADDR_ANY)
752 inp->inp_laddr = sc->sc_inc.inc_laddr;
753 if (in_pcbconnect(inp, (struct sockaddr *)&sin,
755 inp->inp_laddr = laddr;
760 tp->t_state = TCPS_SYN_RECEIVED;
761 tp->iss = sc->sc_iss;
762 tp->irs = sc->sc_irs;
765 tp->snd_wl1 = sc->sc_irs;
766 tp->snd_max = tp->iss + 1;
767 tp->snd_nxt = tp->iss + 1;
768 tp->rcv_up = sc->sc_irs + 1;
769 tp->rcv_wnd = sc->sc_wnd;
770 tp->rcv_adv += tp->rcv_wnd;
771 tp->last_ack_sent = tp->rcv_nxt;
773 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
774 if (sc->sc_flags & SCF_NOOPT)
775 tp->t_flags |= TF_NOOPT;
777 if (sc->sc_flags & SCF_WINSCALE) {
778 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
779 tp->snd_scale = sc->sc_requested_s_scale;
780 tp->request_r_scale = sc->sc_requested_r_scale;
782 if (sc->sc_flags & SCF_TIMESTAMP) {
783 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
784 tp->ts_recent = sc->sc_tsreflect;
785 tp->ts_recent_age = ticks;
786 tp->ts_offset = sc->sc_tsoff;
789 if (sc->sc_flags & SCF_SIGNATURE)
790 tp->t_flags |= TF_SIGNATURE;
792 if (sc->sc_flags & SCF_SACK)
793 tp->t_flags |= TF_SACK_PERMIT;
796 if (sc->sc_flags & SCF_ECN)
797 tp->t_flags |= TF_ECN_PERMIT;
800 * Set up MSS and get cached values from tcp_hostcache.
801 * This might overwrite some of the defaults we just set.
803 tcp_mss(tp, sc->sc_peer_mss);
806 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
809 tp->snd_cwnd = tp->t_maxseg;
810 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
814 TCPSTAT_INC(tcps_accepts);
826 * This function gets called when we receive an ACK for a
827 * socket in the LISTEN state. We look up the connection
828 * in the syncache, and if its there, we pull it out of
829 * the cache and turn it into a full-blown connection in
830 * the SYN-RECEIVED state.
833 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
834 struct socket **lsop, struct mbuf *m)
837 struct syncache_head *sch;
842 * Global TCP locks are held because we manipulate the PCB lists
843 * and create a new socket.
845 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
846 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
847 ("%s: can handle only ACK", __func__));
849 sc = syncache_lookup(inc, &sch); /* returns locked sch */
850 SCH_LOCK_ASSERT(sch);
853 * There is no syncache entry, so see if this ACK is
854 * a returning syncookie. To do this, first:
855 * A. See if this socket has had a syncache entry dropped in
856 * the past. We don't want to accept a bogus syncookie
857 * if we've never received a SYN.
858 * B. check that the syncookie is valid. If it is, then
859 * cobble up a fake syncache entry, and return.
861 if (!V_tcp_syncookies) {
863 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
864 log(LOG_DEBUG, "%s; %s: Spurious ACK, "
865 "segment rejected (syncookies disabled)\n",
869 bzero(&scs, sizeof(scs));
870 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
873 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
874 log(LOG_DEBUG, "%s; %s: Segment failed "
875 "SYNCOOKIE authentication, segment rejected "
876 "(probably spoofed)\n", s, __func__);
880 /* Pull out the entry to unlock the bucket row. */
881 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
883 V_tcp_syncache.cache_count--;
888 * Segment validation:
889 * ACK must match our initial sequence number + 1 (the SYN|ACK).
891 if (th->th_ack != sc->sc_iss + 1 && !TOEPCB_ISSET(sc)) {
892 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
893 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment "
894 "rejected\n", s, __func__, th->th_ack, sc->sc_iss);
899 * The SEQ must fall in the window starting at the received
900 * initial receive sequence number + 1 (the SYN).
902 if ((SEQ_LEQ(th->th_seq, sc->sc_irs) ||
903 SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) &&
905 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
906 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment "
907 "rejected\n", s, __func__, th->th_seq, sc->sc_irs);
911 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
912 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
913 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
914 "segment rejected\n", s, __func__);
918 * If timestamps were negotiated the reflected timestamp
919 * must be equal to what we actually sent in the SYN|ACK.
921 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts &&
923 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
924 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
925 "segment rejected\n",
926 s, __func__, to->to_tsecr, sc->sc_ts);
930 *lsop = syncache_socket(sc, *lsop, m);
933 TCPSTAT_INC(tcps_sc_aborted);
935 TCPSTAT_INC(tcps_sc_completed);
937 /* how do we find the inp for the new socket? */
942 if (sc != NULL && sc != &scs)
951 tcp_offload_syncache_expand(struct in_conninfo *inc, struct toeopt *toeo,
952 struct tcphdr *th, struct socket **lsop, struct mbuf *m)
957 bzero(&to, sizeof(struct tcpopt));
958 to.to_mss = toeo->to_mss;
959 to.to_wscale = toeo->to_wscale;
960 to.to_flags = toeo->to_flags;
962 INP_INFO_WLOCK(&V_tcbinfo);
963 rc = syncache_expand(inc, &to, th, lsop, m);
964 INP_INFO_WUNLOCK(&V_tcbinfo);
970 * Given a LISTEN socket and an inbound SYN request, add
971 * this to the syn cache, and send back a segment:
972 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
975 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
976 * Doing so would require that we hold onto the data and deliver it
977 * to the application. However, if we are the target of a SYN-flood
978 * DoS attack, an attacker could send data which would eventually
979 * consume all available buffer space if it were ACKed. By not ACKing
980 * the data, we avoid this DoS scenario.
983 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
984 struct inpcb *inp, struct socket **lsop, struct mbuf *m,
985 struct toe_usrreqs *tu, void *toepcb)
989 struct syncache *sc = NULL;
990 struct syncache_head *sch;
991 struct mbuf *ipopts = NULL;
993 int win, sb_hiwat, ip_ttl, ip_tos, noopt;
996 int autoflowlabel = 0;
999 struct label *maclabel;
1001 struct syncache scs;
1004 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1005 INP_WLOCK_ASSERT(inp); /* listen socket */
1006 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
1007 ("%s: unexpected tcp flags", __func__));
1010 * Combine all so/tp operations very early to drop the INP lock as
1015 cred = crhold(so->so_cred);
1018 if ((inc->inc_flags & INC_ISIPV6) &&
1019 (inp->inp_flags & IN6P_AUTOFLOWLABEL))
1022 ip_ttl = inp->inp_ip_ttl;
1023 ip_tos = inp->inp_ip_tos;
1024 win = sbspace(&so->so_rcv);
1025 sb_hiwat = so->so_rcv.sb_hiwat;
1026 noopt = (tp->t_flags & TF_NOOPT);
1028 /* By the time we drop the lock these should no longer be used. */
1033 if (mac_syncache_init(&maclabel) != 0) {
1035 INP_INFO_WUNLOCK(&V_tcbinfo);
1038 mac_syncache_create(maclabel, inp);
1041 INP_INFO_WUNLOCK(&V_tcbinfo);
1044 * Remember the IP options, if any.
1047 if (!(inc->inc_flags & INC_ISIPV6))
1049 ipopts = (m) ? ip_srcroute(m) : NULL;
1052 * See if we already have an entry for this connection.
1053 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1055 * XXX: should the syncache be re-initialized with the contents
1056 * of the new SYN here (which may have different options?)
1058 * XXX: We do not check the sequence number to see if this is a
1059 * real retransmit or a new connection attempt. The question is
1060 * how to handle such a case; either ignore it as spoofed, or
1061 * drop the current entry and create a new one?
1063 sc = syncache_lookup(inc, &sch); /* returns locked entry */
1064 SCH_LOCK_ASSERT(sch);
1066 #ifndef TCP_OFFLOAD_DISABLE
1068 sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT,
1071 TCPSTAT_INC(tcps_sc_dupsyn);
1074 * If we were remembering a previous source route,
1075 * forget it and use the new one we've been given.
1078 (void) m_free(sc->sc_ipopts);
1079 sc->sc_ipopts = ipopts;
1082 * Update timestamp if present.
1084 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
1085 sc->sc_tsreflect = to->to_tsval;
1087 sc->sc_flags &= ~SCF_TIMESTAMP;
1090 * Since we have already unconditionally allocated label
1091 * storage, free it up. The syncache entry will already
1092 * have an initialized label we can use.
1094 mac_syncache_destroy(&maclabel);
1096 /* Retransmit SYN|ACK and reset retransmit count. */
1097 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
1098 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
1099 "resetting timer and retransmitting SYN|ACK\n",
1103 if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) {
1105 syncache_timeout(sc, sch, 1);
1106 TCPSTAT_INC(tcps_sndacks);
1107 TCPSTAT_INC(tcps_sndtotal);
1113 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1116 * The zone allocator couldn't provide more entries.
1117 * Treat this as if the cache was full; drop the oldest
1118 * entry and insert the new one.
1120 TCPSTAT_INC(tcps_sc_zonefail);
1121 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1122 syncache_drop(sc, sch);
1123 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1125 if (V_tcp_syncookies) {
1126 bzero(&scs, sizeof(scs));
1131 (void) m_free(ipopts);
1138 * Fill in the syncache values.
1141 sc->sc_label = maclabel;
1145 sc->sc_ipopts = ipopts;
1146 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1148 if (!(inc->inc_flags & INC_ISIPV6))
1151 sc->sc_ip_tos = ip_tos;
1152 sc->sc_ip_ttl = ip_ttl;
1154 #ifndef TCP_OFFLOAD_DISABLE
1156 sc->sc_toepcb = toepcb;
1158 sc->sc_irs = th->th_seq;
1159 sc->sc_iss = arc4random();
1161 sc->sc_flowlabel = 0;
1164 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
1165 * win was derived from socket earlier in the function.
1168 win = imin(win, TCP_MAXWIN);
1171 if (V_tcp_do_rfc1323) {
1173 * A timestamp received in a SYN makes
1174 * it ok to send timestamp requests and replies.
1176 if (to->to_flags & TOF_TS) {
1177 sc->sc_tsreflect = to->to_tsval;
1179 sc->sc_flags |= SCF_TIMESTAMP;
1181 if (to->to_flags & TOF_SCALE) {
1185 * Pick the smallest possible scaling factor that
1186 * will still allow us to scale up to sb_max, aka
1187 * kern.ipc.maxsockbuf.
1189 * We do this because there are broken firewalls that
1190 * will corrupt the window scale option, leading to
1191 * the other endpoint believing that our advertised
1192 * window is unscaled. At scale factors larger than
1193 * 5 the unscaled window will drop below 1500 bytes,
1194 * leading to serious problems when traversing these
1197 * With the default maxsockbuf of 256K, a scale factor
1198 * of 3 will be chosen by this algorithm. Those who
1199 * choose a larger maxsockbuf should watch out
1200 * for the compatiblity problems mentioned above.
1202 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1203 * or <SYN,ACK>) segment itself is never scaled.
1205 while (wscale < TCP_MAX_WINSHIFT &&
1206 (TCP_MAXWIN << wscale) < sb_max)
1208 sc->sc_requested_r_scale = wscale;
1209 sc->sc_requested_s_scale = to->to_wscale;
1210 sc->sc_flags |= SCF_WINSCALE;
1213 #ifdef TCP_SIGNATURE
1215 * If listening socket requested TCP digests, and received SYN
1216 * contains the option, flag this in the syncache so that
1217 * syncache_respond() will do the right thing with the SYN+ACK.
1218 * XXX: Currently we always record the option by default and will
1219 * attempt to use it in syncache_respond().
1221 if (to->to_flags & TOF_SIGNATURE)
1222 sc->sc_flags |= SCF_SIGNATURE;
1224 if (to->to_flags & TOF_SACKPERM)
1225 sc->sc_flags |= SCF_SACK;
1226 if (to->to_flags & TOF_MSS)
1227 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */
1229 sc->sc_flags |= SCF_NOOPT;
1230 if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn)
1231 sc->sc_flags |= SCF_ECN;
1233 if (V_tcp_syncookies) {
1234 syncookie_generate(sch, sc, &flowtmp);
1237 sc->sc_flowlabel = flowtmp;
1243 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1249 * Do a standard 3-way handshake.
1251 if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) {
1252 if (V_tcp_syncookies && V_tcp_syncookiesonly && sc != &scs)
1254 else if (sc != &scs)
1255 syncache_insert(sc, sch); /* locks and unlocks sch */
1256 TCPSTAT_INC(tcps_sndacks);
1257 TCPSTAT_INC(tcps_sndtotal);
1261 TCPSTAT_INC(tcps_sc_dropped);
1269 mac_syncache_destroy(&maclabel);
1279 syncache_respond(struct syncache *sc)
1281 struct ip *ip = NULL;
1285 u_int16_t hlen, tlen, mssopt;
1288 struct ip6_hdr *ip6 = NULL;
1293 (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) :
1296 tlen = hlen + sizeof(struct tcphdr);
1298 /* Determine MSS we advertize to other end of connection. */
1299 mssopt = tcp_mssopt(&sc->sc_inc);
1300 if (sc->sc_peer_mss)
1301 mssopt = max( min(sc->sc_peer_mss, mssopt), V_tcp_minmss);
1303 /* XXX: Assume that the entire packet will fit in a header mbuf. */
1304 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1305 ("syncache: mbuf too small"));
1307 /* Create the IP+TCP header from scratch. */
1308 m = m_gethdr(M_DONTWAIT, MT_DATA);
1312 mac_syncache_create_mbuf(sc->sc_label, m);
1314 m->m_data += max_linkhdr;
1316 m->m_pkthdr.len = tlen;
1317 m->m_pkthdr.rcvif = NULL;
1320 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1321 ip6 = mtod(m, struct ip6_hdr *);
1322 ip6->ip6_vfc = IPV6_VERSION;
1323 ip6->ip6_nxt = IPPROTO_TCP;
1324 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1325 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1326 ip6->ip6_plen = htons(tlen - hlen);
1327 /* ip6_hlim is set after checksum */
1328 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1329 ip6->ip6_flow |= sc->sc_flowlabel;
1331 th = (struct tcphdr *)(ip6 + 1);
1335 ip = mtod(m, struct ip *);
1336 ip->ip_v = IPVERSION;
1337 ip->ip_hl = sizeof(struct ip) >> 2;
1342 ip->ip_p = IPPROTO_TCP;
1343 ip->ip_src = sc->sc_inc.inc_laddr;
1344 ip->ip_dst = sc->sc_inc.inc_faddr;
1345 ip->ip_ttl = sc->sc_ip_ttl;
1346 ip->ip_tos = sc->sc_ip_tos;
1349 * See if we should do MTU discovery. Route lookups are
1350 * expensive, so we will only unset the DF bit if:
1352 * 1) path_mtu_discovery is disabled
1353 * 2) the SCF_UNREACH flag has been set
1355 if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1356 ip->ip_off |= IP_DF;
1358 th = (struct tcphdr *)(ip + 1);
1360 th->th_sport = sc->sc_inc.inc_lport;
1361 th->th_dport = sc->sc_inc.inc_fport;
1363 th->th_seq = htonl(sc->sc_iss);
1364 th->th_ack = htonl(sc->sc_irs + 1);
1365 th->th_off = sizeof(struct tcphdr) >> 2;
1367 th->th_flags = TH_SYN|TH_ACK;
1368 th->th_win = htons(sc->sc_wnd);
1371 if (sc->sc_flags & SCF_ECN) {
1372 th->th_flags |= TH_ECE;
1373 TCPSTAT_INC(tcps_ecn_shs);
1376 /* Tack on the TCP options. */
1377 if ((sc->sc_flags & SCF_NOOPT) == 0) {
1381 to.to_flags = TOF_MSS;
1382 if (sc->sc_flags & SCF_WINSCALE) {
1383 to.to_wscale = sc->sc_requested_r_scale;
1384 to.to_flags |= TOF_SCALE;
1386 if (sc->sc_flags & SCF_TIMESTAMP) {
1387 /* Virgin timestamp or TCP cookie enhanced one. */
1388 to.to_tsval = sc->sc_ts;
1389 to.to_tsecr = sc->sc_tsreflect;
1390 to.to_flags |= TOF_TS;
1392 if (sc->sc_flags & SCF_SACK)
1393 to.to_flags |= TOF_SACKPERM;
1394 #ifdef TCP_SIGNATURE
1395 if (sc->sc_flags & SCF_SIGNATURE)
1396 to.to_flags |= TOF_SIGNATURE;
1398 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1400 /* Adjust headers by option size. */
1401 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1403 m->m_pkthdr.len += optlen;
1405 #ifdef TCP_SIGNATURE
1406 if (sc->sc_flags & SCF_SIGNATURE)
1407 tcp_signature_compute(m, 0, 0, optlen,
1408 to.to_signature, IPSEC_DIR_OUTBOUND);
1411 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1412 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1415 ip->ip_len += optlen;
1419 M_SETFIB(m, sc->sc_inc.inc_fibnum);
1421 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1423 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
1424 tlen + optlen - hlen);
1425 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1426 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1430 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1431 htons(tlen + optlen - hlen + IPPROTO_TCP));
1432 m->m_pkthdr.csum_flags = CSUM_TCP;
1433 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1434 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1440 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1441 struct inpcb *inp, struct socket **lsop, struct mbuf *m)
1443 _syncache_add(inc, to, th, inp, lsop, m, NULL, NULL);
1447 tcp_offload_syncache_add(struct in_conninfo *inc, struct toeopt *toeo,
1448 struct tcphdr *th, struct inpcb *inp, struct socket **lsop,
1449 struct toe_usrreqs *tu, void *toepcb)
1453 bzero(&to, sizeof(struct tcpopt));
1454 to.to_mss = toeo->to_mss;
1455 to.to_wscale = toeo->to_wscale;
1456 to.to_flags = toeo->to_flags;
1458 INP_INFO_WLOCK(&V_tcbinfo);
1461 _syncache_add(inc, &to, th, inp, lsop, NULL, tu, toepcb);
1465 * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1466 * receive and to be able to handle SYN floods from bogus source addresses
1467 * (where we will never receive any reply). SYN floods try to exhaust all
1468 * our memory and available slots in the SYN cache table to cause a denial
1469 * of service to legitimate users of the local host.
1471 * The idea of SYN cookies is to encode and include all necessary information
1472 * about the connection setup state within the SYN-ACK we send back and thus
1473 * to get along without keeping any local state until the ACK to the SYN-ACK
1474 * arrives (if ever). Everything we need to know should be available from
1475 * the information we encoded in the SYN-ACK.
1477 * More information about the theory behind SYN cookies and its first
1478 * discussion and specification can be found at:
1479 * http://cr.yp.to/syncookies.html (overview)
1480 * http://cr.yp.to/syncookies/archive (gory details)
1482 * This implementation extends the orginal idea and first implementation
1483 * of FreeBSD by using not only the initial sequence number field to store
1484 * information but also the timestamp field if present. This way we can
1485 * keep track of the entire state we need to know to recreate the session in
1486 * its original form. Almost all TCP speakers implement RFC1323 timestamps
1487 * these days. For those that do not we still have to live with the known
1488 * shortcomings of the ISN only SYN cookies.
1492 * Initial sequence number we send:
1493 * 31|................................|0
1494 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1495 * D = MD5 Digest (first dword)
1497 * R = Rotation of secret
1498 * P = Odd or Even secret
1500 * The MD5 Digest is computed with over following parameters:
1501 * a) randomly rotated secret
1502 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1503 * c) the received initial sequence number from remote host
1504 * d) the rotation offset and odd/even bit
1506 * Timestamp we send:
1507 * 31|................................|0
1508 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1509 * D = MD5 Digest (third dword) (only as filler)
1510 * S = Requested send window scale
1511 * R = Requested receive window scale
1513 * 5 = TCP-MD5 enabled (not implemented yet)
1514 * XORed with MD5 Digest (forth dword)
1516 * The timestamp isn't cryptographically secure and doesn't need to be.
1517 * The double use of the MD5 digest dwords ties it to a specific remote/
1518 * local host/port, remote initial sequence number and our local time
1519 * limited secret. A received timestamp is reverted (XORed) and then
1520 * the contained MD5 dword is compared to the computed one to ensure the
1521 * timestamp belongs to the SYN-ACK we sent. The other parameters may
1522 * have been tampered with but this isn't different from supplying bogus
1523 * values in the SYN in the first place.
1525 * Some problems with SYN cookies remain however:
1526 * Consider the problem of a recreated (and retransmitted) cookie. If the
1527 * original SYN was accepted, the connection is established. The second
1528 * SYN is inflight, and if it arrives with an ISN that falls within the
1529 * receive window, the connection is killed.
1532 * A heuristic to determine when to accept syn cookies is not necessary.
1533 * An ACK flood would cause the syncookie verification to be attempted,
1534 * but a SYN flood causes syncookies to be generated. Both are of equal
1535 * cost, so there's no point in trying to optimize the ACK flood case.
1536 * Also, if you don't process certain ACKs for some reason, then all someone
1537 * would have to do is launch a SYN and ACK flood at the same time, which
1538 * would stop cookie verification and defeat the entire purpose of syncookies.
1540 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1543 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1544 u_int32_t *flowlabel)
1547 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1550 u_int off, pmss, mss;
1553 SCH_LOCK_ASSERT(sch);
1555 /* Which of the two secrets to use. */
1556 secbits = sch->sch_oddeven ?
1557 sch->sch_secbits_odd : sch->sch_secbits_even;
1559 /* Reseed secret if too old. */
1560 if (sch->sch_reseed < time_uptime) {
1561 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */
1562 secbits = sch->sch_oddeven ?
1563 sch->sch_secbits_odd : sch->sch_secbits_even;
1564 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1565 secbits[i] = arc4random();
1566 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1569 /* Secret rotation offset. */
1570 off = sc->sc_iss & 0x7; /* iss was randomized before */
1572 /* Maximum segment size calculation. */
1574 max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), V_tcp_minmss);
1575 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1576 if (tcp_sc_msstab[mss] <= pmss)
1579 /* Fold parameters and MD5 digest into the ISN we will send. */
1580 data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1581 data |= off << 1; /* secret offset, derived from iss, 3 bits */
1582 data |= mss << 4; /* mss, 3 bits */
1585 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1586 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1587 MD5Update(&ctx, secbits, off);
1588 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1589 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1590 MD5Update(&ctx, &data, sizeof(data));
1591 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1593 data |= (md5_buffer[0] << 7);
1597 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1600 /* Additional parameters are stored in the timestamp if present. */
1601 if (sc->sc_flags & SCF_TIMESTAMP) {
1602 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1603 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1604 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */
1605 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */
1606 data |= md5_buffer[2] << 10; /* more digest bits */
1607 data ^= md5_buffer[3];
1609 sc->sc_tsoff = data - ticks; /* after XOR */
1612 TCPSTAT_INC(tcps_sc_sendcookie);
1615 static struct syncache *
1616 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1617 struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1621 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1625 int off, mss, wnd, flags;
1627 SCH_LOCK_ASSERT(sch);
1630 * Pull information out of SYN-ACK/ACK and
1631 * revert sequence number advances.
1633 ack = th->th_ack - 1;
1634 seq = th->th_seq - 1;
1635 off = (ack >> 1) & 0x7;
1636 mss = (ack >> 4) & 0x7;
1639 /* Which of the two secrets to use. */
1640 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1643 * The secret wasn't updated for the lifetime of a syncookie,
1644 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1646 if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) {
1650 /* Recompute the digest so we can compare it. */
1652 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1653 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1654 MD5Update(&ctx, secbits, off);
1655 MD5Update(&ctx, inc, sizeof(*inc));
1656 MD5Update(&ctx, &seq, sizeof(seq));
1657 MD5Update(&ctx, &flags, sizeof(flags));
1658 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1660 /* Does the digest part of or ACK'ed ISS match? */
1661 if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1664 /* Does the digest part of our reflected timestamp match? */
1665 if (to->to_flags & TOF_TS) {
1666 data = md5_buffer[3] ^ to->to_tsecr;
1667 if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1671 /* Fill in the syncache values. */
1672 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1673 sc->sc_ipopts = NULL;
1679 if (inc->inc_flags & INC_ISIPV6) {
1680 if (sotoinpcb(so)->inp_flags & IN6P_AUTOFLOWLABEL)
1681 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1685 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1686 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1689 /* Additional parameters that were encoded in the timestamp. */
1691 sc->sc_flags |= SCF_TIMESTAMP;
1692 sc->sc_tsreflect = to->to_tsval;
1693 sc->sc_ts = to->to_tsecr;
1694 sc->sc_tsoff = to->to_tsecr - ticks;
1695 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1696 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1697 sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1699 sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1701 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1702 sc->sc_flags |= SCF_WINSCALE;
1704 sc->sc_flags |= SCF_NOOPT;
1706 wnd = sbspace(&so->so_rcv);
1708 wnd = imin(wnd, TCP_MAXWIN);
1712 sc->sc_peer_mss = tcp_sc_msstab[mss];
1714 TCPSTAT_INC(tcps_sc_recvcookie);
1719 * Returns the current number of syncache entries. This number
1720 * will probably change before you get around to calling
1725 syncache_pcbcount(void)
1727 struct syncache_head *sch;
1730 for (count = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1731 /* No need to lock for a read. */
1732 sch = &V_tcp_syncache.hashbase[i];
1733 count += sch->sch_length;
1739 * Exports the syncache entries to userland so that netstat can display
1740 * them alongside the other sockets. This function is intended to be
1741 * called only from tcp_pcblist.
1743 * Due to concurrency on an active system, the number of pcbs exported
1744 * may have no relation to max_pcbs. max_pcbs merely indicates the
1745 * amount of space the caller allocated for this function to use.
1748 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
1751 struct syncache *sc;
1752 struct syncache_head *sch;
1753 int count, error, i;
1755 for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1756 sch = &V_tcp_syncache.hashbase[i];
1758 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
1759 if (count >= max_pcbs) {
1763 if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0)
1765 bzero(&xt, sizeof(xt));
1766 xt.xt_len = sizeof(xt);
1767 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1768 xt.xt_inp.inp_vflag = INP_IPV6;
1770 xt.xt_inp.inp_vflag = INP_IPV4;
1771 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
1772 xt.xt_tp.t_inpcb = &xt.xt_inp;
1773 xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
1774 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1775 xt.xt_socket.xso_len = sizeof (struct xsocket);
1776 xt.xt_socket.so_type = SOCK_STREAM;
1777 xt.xt_socket.so_state = SS_ISCONNECTING;
1778 error = SYSCTL_OUT(req, &xt, sizeof xt);
1788 *pcbs_exported = count;