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 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
152 SYSCTL_VNET_UINT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
153 &VNET_NAME(tcp_syncache.bucket_limit), 0,
154 "Per-bucket hash limit for syncache");
156 SYSCTL_VNET_UINT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
157 &VNET_NAME(tcp_syncache.cache_limit), 0,
158 "Overall entry limit for syncache");
160 SYSCTL_VNET_UINT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
161 &VNET_NAME(tcp_syncache.cache_count), 0,
162 "Current number of entries in syncache");
164 SYSCTL_VNET_UINT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
165 &VNET_NAME(tcp_syncache.hashsize), 0,
166 "Size of TCP syncache hashtable");
168 SYSCTL_VNET_UINT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
169 &VNET_NAME(tcp_syncache.rexmt_limit), 0,
170 "Limit on SYN/ACK retransmissions");
172 VNET_DEFINE(int, tcp_sc_rst_sock_fail) = 1;
173 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail,
174 CTLFLAG_RW, &VNET_NAME(tcp_sc_rst_sock_fail), 0,
175 "Send reset on socket allocation failure");
177 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
179 #define SYNCACHE_HASH(inc, mask) \
180 ((V_tcp_syncache.hash_secret ^ \
181 (inc)->inc_faddr.s_addr ^ \
182 ((inc)->inc_faddr.s_addr >> 16) ^ \
183 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
185 #define SYNCACHE_HASH6(inc, mask) \
186 ((V_tcp_syncache.hash_secret ^ \
187 (inc)->inc6_faddr.s6_addr32[0] ^ \
188 (inc)->inc6_faddr.s6_addr32[3] ^ \
189 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
191 #define ENDPTS_EQ(a, b) ( \
192 (a)->ie_fport == (b)->ie_fport && \
193 (a)->ie_lport == (b)->ie_lport && \
194 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
195 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
198 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
200 #define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx)
201 #define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx)
202 #define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED)
205 * Requires the syncache entry to be already removed from the bucket list.
208 syncache_free(struct syncache *sc)
212 (void) m_free(sc->sc_ipopts);
216 mac_syncache_destroy(&sc->sc_label);
219 uma_zfree(V_tcp_syncache.zone, sc);
227 V_tcp_syncache.cache_count = 0;
228 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
229 V_tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
230 V_tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
231 V_tcp_syncache.hash_secret = arc4random();
233 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
234 &V_tcp_syncache.hashsize);
235 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
236 &V_tcp_syncache.bucket_limit);
237 if (!powerof2(V_tcp_syncache.hashsize) ||
238 V_tcp_syncache.hashsize == 0) {
239 printf("WARNING: syncache hash size is not a power of 2.\n");
240 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
242 V_tcp_syncache.hashmask = V_tcp_syncache.hashsize - 1;
245 V_tcp_syncache.cache_limit =
246 V_tcp_syncache.hashsize * V_tcp_syncache.bucket_limit;
247 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
248 &V_tcp_syncache.cache_limit);
250 /* Allocate the hash table. */
251 V_tcp_syncache.hashbase = malloc(V_tcp_syncache.hashsize *
252 sizeof(struct syncache_head), M_SYNCACHE, M_WAITOK | M_ZERO);
254 /* Initialize the hash buckets. */
255 for (i = 0; i < V_tcp_syncache.hashsize; i++) {
257 V_tcp_syncache.hashbase[i].sch_vnet = curvnet;
259 TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket);
260 mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
262 callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer,
263 &V_tcp_syncache.hashbase[i].sch_mtx, 0);
264 V_tcp_syncache.hashbase[i].sch_length = 0;
267 /* Create the syncache entry zone. */
268 V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
269 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
270 uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit);
275 syncache_destroy(void)
277 struct syncache_head *sch;
278 struct syncache *sc, *nsc;
281 /* Cleanup hash buckets: stop timers, free entries, destroy locks. */
282 for (i = 0; i < V_tcp_syncache.hashsize; i++) {
284 sch = &V_tcp_syncache.hashbase[i];
285 callout_drain(&sch->sch_timer);
288 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc)
289 syncache_drop(sc, sch);
291 KASSERT(TAILQ_EMPTY(&sch->sch_bucket),
292 ("%s: sch->sch_bucket not empty", __func__));
293 KASSERT(sch->sch_length == 0, ("%s: sch->sch_length %d not 0",
294 __func__, sch->sch_length));
295 mtx_destroy(&sch->sch_mtx);
298 KASSERT(V_tcp_syncache.cache_count == 0, ("%s: cache_count %d not 0",
299 __func__, V_tcp_syncache.cache_count));
301 /* Free the allocated global resources. */
302 uma_zdestroy(V_tcp_syncache.zone);
303 free(V_tcp_syncache.hashbase, M_SYNCACHE);
308 * Inserts a syncache entry into the specified bucket row.
309 * Locks and unlocks the syncache_head autonomously.
312 syncache_insert(struct syncache *sc, struct syncache_head *sch)
314 struct syncache *sc2;
319 * Make sure that we don't overflow the per-bucket limit.
320 * If the bucket is full, toss the oldest element.
322 if (sch->sch_length >= V_tcp_syncache.bucket_limit) {
323 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
324 ("sch->sch_length incorrect"));
325 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
326 syncache_drop(sc2, sch);
327 TCPSTAT_INC(tcps_sc_bucketoverflow);
330 /* Put it into the bucket. */
331 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
334 /* Reinitialize the bucket row's timer. */
335 if (sch->sch_length == 1)
336 sch->sch_nextc = ticks + INT_MAX;
337 syncache_timeout(sc, sch, 1);
341 V_tcp_syncache.cache_count++;
342 TCPSTAT_INC(tcps_sc_added);
346 * Remove and free entry from syncache bucket row.
347 * Expects locked syncache head.
350 syncache_drop(struct syncache *sc, struct syncache_head *sch)
353 SCH_LOCK_ASSERT(sch);
355 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
358 #ifndef TCP_OFFLOAD_DISABLE
360 sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb);
363 V_tcp_syncache.cache_count--;
367 * Engage/reengage time on bucket row.
370 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
372 sc->sc_rxttime = ticks +
373 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
375 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
376 sch->sch_nextc = sc->sc_rxttime;
378 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
379 syncache_timer, (void *)sch);
384 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
385 * If we have retransmitted an entry the maximum number of times, expire it.
386 * One separate timer for each bucket row.
389 syncache_timer(void *xsch)
391 struct syncache_head *sch = (struct syncache_head *)xsch;
392 struct syncache *sc, *nsc;
396 CURVNET_SET(sch->sch_vnet);
398 /* NB: syncache_head has already been locked by the callout. */
399 SCH_LOCK_ASSERT(sch);
402 * In the following cycle we may remove some entries and/or
403 * advance some timeouts, so re-initialize the bucket timer.
405 sch->sch_nextc = tick + INT_MAX;
407 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
409 * We do not check if the listen socket still exists
410 * and accept the case where the listen socket may be
411 * gone by the time we resend the SYN/ACK. We do
412 * not expect this to happens often. If it does,
413 * then the RST will be sent by the time the remote
414 * host does the SYN/ACK->ACK.
416 if (TSTMP_GT(sc->sc_rxttime, tick)) {
417 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
418 sch->sch_nextc = sc->sc_rxttime;
421 if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) {
422 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
423 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
424 "giving up and removing syncache entry\n",
428 syncache_drop(sc, sch);
429 TCPSTAT_INC(tcps_sc_stale);
432 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
433 log(LOG_DEBUG, "%s; %s: Response timeout, "
434 "retransmitting (%u) SYN|ACK\n",
435 s, __func__, sc->sc_rxmits);
439 (void) syncache_respond(sc);
440 TCPSTAT_INC(tcps_sc_retransmitted);
441 syncache_timeout(sc, sch, 0);
443 if (!TAILQ_EMPTY(&(sch)->sch_bucket))
444 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
445 syncache_timer, (void *)(sch));
450 * Find an entry in the syncache.
451 * Returns always with locked syncache_head plus a matching entry or NULL.
454 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
457 struct syncache_head *sch;
460 if (inc->inc_flags & INC_ISIPV6) {
461 sch = &V_tcp_syncache.hashbase[
462 SYNCACHE_HASH6(inc, V_tcp_syncache.hashmask)];
467 /* Circle through bucket row to find matching entry. */
468 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
469 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
475 sch = &V_tcp_syncache.hashbase[
476 SYNCACHE_HASH(inc, V_tcp_syncache.hashmask)];
481 /* Circle through bucket row to find matching entry. */
482 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
484 if (sc->sc_inc.inc_flags & INC_ISIPV6)
487 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
491 SCH_LOCK_ASSERT(*schp);
492 return (NULL); /* always returns with locked sch */
496 * This function is called when we get a RST for a
497 * non-existent connection, so that we can see if the
498 * connection is in the syn cache. If it is, zap it.
501 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
504 struct syncache_head *sch;
507 sc = syncache_lookup(inc, &sch); /* returns locked sch */
508 SCH_LOCK_ASSERT(sch);
511 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
512 * See RFC 793 page 65, section SEGMENT ARRIVES.
514 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
515 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
516 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
517 "FIN flag set, segment ignored\n", s, __func__);
518 TCPSTAT_INC(tcps_badrst);
523 * No corresponding connection was found in syncache.
524 * If syncookies are enabled and possibly exclusively
525 * used, or we are under memory pressure, a valid RST
526 * may not find a syncache entry. In that case we're
527 * done and no SYN|ACK retransmissions will happen.
528 * Otherwise the RST was misdirected or spoofed.
531 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
532 log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
533 "syncache entry (possibly syncookie only), "
534 "segment ignored\n", s, __func__);
535 TCPSTAT_INC(tcps_badrst);
540 * If the RST bit is set, check the sequence number to see
541 * if this is a valid reset segment.
543 * In all states except SYN-SENT, all reset (RST) segments
544 * are validated by checking their SEQ-fields. A reset is
545 * valid if its sequence number is in the window.
547 * The sequence number in the reset segment is normally an
548 * echo of our outgoing acknowlegement numbers, but some hosts
549 * send a reset with the sequence number at the rightmost edge
550 * of our receive window, and we have to handle this case.
552 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
553 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
554 syncache_drop(sc, sch);
555 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
556 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
557 "connection attempt aborted by remote endpoint\n",
559 TCPSTAT_INC(tcps_sc_reset);
561 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
562 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != "
563 "IRS %u (+WND %u), segment ignored\n",
564 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
565 TCPSTAT_INC(tcps_badrst);
575 syncache_badack(struct in_conninfo *inc)
578 struct syncache_head *sch;
580 sc = syncache_lookup(inc, &sch); /* returns locked sch */
581 SCH_LOCK_ASSERT(sch);
583 syncache_drop(sc, sch);
584 TCPSTAT_INC(tcps_sc_badack);
590 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
593 struct syncache_head *sch;
595 sc = syncache_lookup(inc, &sch); /* returns locked sch */
596 SCH_LOCK_ASSERT(sch);
600 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
601 if (ntohl(th->th_seq) != sc->sc_iss)
605 * If we've rertransmitted 3 times and this is our second error,
606 * we remove the entry. Otherwise, we allow it to continue on.
607 * This prevents us from incorrectly nuking an entry during a
608 * spurious network outage.
612 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
613 sc->sc_flags |= SCF_UNREACH;
616 syncache_drop(sc, sch);
617 TCPSTAT_INC(tcps_sc_unreach);
623 * Build a new TCP socket structure from a syncache entry.
625 static struct socket *
626 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
628 struct inpcb *inp = NULL;
634 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
637 * Ok, create the full blown connection, and set things up
638 * as they would have been set up if we had created the
639 * connection when the SYN arrived. If we can't create
640 * the connection, abort it.
642 so = sonewconn(lso, SS_ISCONNECTED);
645 * Drop the connection; we will either send a RST or
646 * have the peer retransmit its SYN again after its
649 TCPSTAT_INC(tcps_listendrop);
650 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
651 log(LOG_DEBUG, "%s; %s: Socket create failed "
652 "due to limits or memory shortage\n",
659 mac_socketpeer_set_from_mbuf(m, so);
663 inp->inp_inc.inc_fibnum = so->so_fibnum;
665 INP_HASH_WLOCK(&V_tcbinfo);
667 /* Insert new socket into PCB hash list. */
668 inp->inp_inc.inc_flags = sc->sc_inc.inc_flags;
670 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
671 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
673 inp->inp_vflag &= ~INP_IPV6;
674 inp->inp_vflag |= INP_IPV4;
676 inp->inp_laddr = sc->sc_inc.inc_laddr;
682 * Install in the reservation hash table for now, but don't yet
683 * install a connection group since the full 4-tuple isn't yet
686 inp->inp_lport = sc->sc_inc.inc_lport;
687 if ((error = in_pcbinshash_nopcbgroup(inp)) != 0) {
689 * Undo the assignments above if we failed to
690 * put the PCB on the hash lists.
693 if (sc->sc_inc.inc_flags & INC_ISIPV6)
694 inp->in6p_laddr = in6addr_any;
697 inp->inp_laddr.s_addr = INADDR_ANY;
699 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
700 log(LOG_DEBUG, "%s; %s: in_pcbinshash failed "
705 INP_HASH_WUNLOCK(&V_tcbinfo);
709 /* Copy old policy into new socket's. */
710 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
711 printf("syncache_socket: could not copy policy\n");
714 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
715 struct inpcb *oinp = sotoinpcb(lso);
716 struct in6_addr laddr6;
717 struct sockaddr_in6 sin6;
719 * Inherit socket options from the listening socket.
720 * Note that in6p_inputopts are not (and should not be)
721 * copied, since it stores previously received options and is
722 * used to detect if each new option is different than the
723 * previous one and hence should be passed to a user.
724 * If we copied in6p_inputopts, a user would not be able to
725 * receive options just after calling the accept system call.
727 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
728 if (oinp->in6p_outputopts)
729 inp->in6p_outputopts =
730 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
732 sin6.sin6_family = AF_INET6;
733 sin6.sin6_len = sizeof(sin6);
734 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
735 sin6.sin6_port = sc->sc_inc.inc_fport;
736 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
737 laddr6 = inp->in6p_laddr;
738 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
739 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
740 if ((error = in6_pcbconnect_mbuf(inp, (struct sockaddr *)&sin6,
741 thread0.td_ucred, m)) != 0) {
742 inp->in6p_laddr = laddr6;
743 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
744 log(LOG_DEBUG, "%s; %s: in6_pcbconnect failed "
749 INP_HASH_WUNLOCK(&V_tcbinfo);
752 /* Override flowlabel from in6_pcbconnect. */
753 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
754 inp->inp_flow |= sc->sc_flowlabel;
757 #if defined(INET) && defined(INET6)
762 struct in_addr laddr;
763 struct sockaddr_in sin;
765 inp->inp_options = (m) ? ip_srcroute(m) : NULL;
767 if (inp->inp_options == NULL) {
768 inp->inp_options = sc->sc_ipopts;
769 sc->sc_ipopts = NULL;
772 sin.sin_family = AF_INET;
773 sin.sin_len = sizeof(sin);
774 sin.sin_addr = sc->sc_inc.inc_faddr;
775 sin.sin_port = sc->sc_inc.inc_fport;
776 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
777 laddr = inp->inp_laddr;
778 if (inp->inp_laddr.s_addr == INADDR_ANY)
779 inp->inp_laddr = sc->sc_inc.inc_laddr;
780 if ((error = in_pcbconnect_mbuf(inp, (struct sockaddr *)&sin,
781 thread0.td_ucred, m)) != 0) {
782 inp->inp_laddr = laddr;
783 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
784 log(LOG_DEBUG, "%s; %s: in_pcbconnect failed "
789 INP_HASH_WUNLOCK(&V_tcbinfo);
794 INP_HASH_WUNLOCK(&V_tcbinfo);
796 tp->t_state = TCPS_SYN_RECEIVED;
797 tp->iss = sc->sc_iss;
798 tp->irs = sc->sc_irs;
801 tp->snd_wl1 = sc->sc_irs;
802 tp->snd_max = tp->iss + 1;
803 tp->snd_nxt = tp->iss + 1;
804 tp->rcv_up = sc->sc_irs + 1;
805 tp->rcv_wnd = sc->sc_wnd;
806 tp->rcv_adv += tp->rcv_wnd;
807 tp->last_ack_sent = tp->rcv_nxt;
809 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
810 if (sc->sc_flags & SCF_NOOPT)
811 tp->t_flags |= TF_NOOPT;
813 if (sc->sc_flags & SCF_WINSCALE) {
814 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
815 tp->snd_scale = sc->sc_requested_s_scale;
816 tp->request_r_scale = sc->sc_requested_r_scale;
818 if (sc->sc_flags & SCF_TIMESTAMP) {
819 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
820 tp->ts_recent = sc->sc_tsreflect;
821 tp->ts_recent_age = ticks;
822 tp->ts_offset = sc->sc_tsoff;
825 if (sc->sc_flags & SCF_SIGNATURE)
826 tp->t_flags |= TF_SIGNATURE;
828 if (sc->sc_flags & SCF_SACK)
829 tp->t_flags |= TF_SACK_PERMIT;
832 if (sc->sc_flags & SCF_ECN)
833 tp->t_flags |= TF_ECN_PERMIT;
836 * Set up MSS and get cached values from tcp_hostcache.
837 * This might overwrite some of the defaults we just set.
839 tcp_mss(tp, sc->sc_peer_mss);
842 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
843 * NB: sc_rxmits counts all SYN,ACK transmits, not just retransmits.
845 if (sc->sc_rxmits > 1)
846 tp->snd_cwnd = tp->t_maxseg;
849 * Copy and activate timers.
851 tp->t_keepinit = sototcpcb(lso)->t_keepinit;
852 tp->t_keepidle = sototcpcb(lso)->t_keepidle;
853 tp->t_keepintvl = sototcpcb(lso)->t_keepintvl;
854 tp->t_keepcnt = sototcpcb(lso)->t_keepcnt;
855 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
859 TCPSTAT_INC(tcps_accepts);
871 * This function gets called when we receive an ACK for a
872 * socket in the LISTEN state. We look up the connection
873 * in the syncache, and if its there, we pull it out of
874 * the cache and turn it into a full-blown connection in
875 * the SYN-RECEIVED state.
878 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
879 struct socket **lsop, struct mbuf *m)
882 struct syncache_head *sch;
887 * Global TCP locks are held because we manipulate the PCB lists
888 * and create a new socket.
890 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
891 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
892 ("%s: can handle only ACK", __func__));
894 sc = syncache_lookup(inc, &sch); /* returns locked sch */
895 SCH_LOCK_ASSERT(sch);
898 * There is no syncache entry, so see if this ACK is
899 * a returning syncookie. To do this, first:
900 * A. See if this socket has had a syncache entry dropped in
901 * the past. We don't want to accept a bogus syncookie
902 * if we've never received a SYN.
903 * B. check that the syncookie is valid. If it is, then
904 * cobble up a fake syncache entry, and return.
906 if (!V_tcp_syncookies) {
908 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
909 log(LOG_DEBUG, "%s; %s: Spurious ACK, "
910 "segment rejected (syncookies disabled)\n",
914 bzero(&scs, sizeof(scs));
915 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
918 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
919 log(LOG_DEBUG, "%s; %s: Segment failed "
920 "SYNCOOKIE authentication, segment rejected "
921 "(probably spoofed)\n", s, __func__);
925 /* Pull out the entry to unlock the bucket row. */
926 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
928 V_tcp_syncache.cache_count--;
933 * Segment validation:
934 * ACK must match our initial sequence number + 1 (the SYN|ACK).
936 if (th->th_ack != sc->sc_iss + 1 && !TOEPCB_ISSET(sc)) {
937 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
938 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment "
939 "rejected\n", s, __func__, th->th_ack, sc->sc_iss);
944 * The SEQ must fall in the window starting at the received
945 * initial receive sequence number + 1 (the SYN).
947 if ((SEQ_LEQ(th->th_seq, sc->sc_irs) ||
948 SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) &&
950 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
951 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment "
952 "rejected\n", s, __func__, th->th_seq, sc->sc_irs);
956 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
957 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
958 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
959 "segment rejected\n", s, __func__);
963 * If timestamps were negotiated the reflected timestamp
964 * must be equal to what we actually sent in the SYN|ACK.
966 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts &&
968 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
969 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
970 "segment rejected\n",
971 s, __func__, to->to_tsecr, sc->sc_ts);
975 *lsop = syncache_socket(sc, *lsop, m);
978 TCPSTAT_INC(tcps_sc_aborted);
980 TCPSTAT_INC(tcps_sc_completed);
982 /* how do we find the inp for the new socket? */
987 if (sc != NULL && sc != &scs)
996 tcp_offload_syncache_expand(struct in_conninfo *inc, struct toeopt *toeo,
997 struct tcphdr *th, struct socket **lsop, struct mbuf *m)
1002 bzero(&to, sizeof(struct tcpopt));
1003 to.to_mss = toeo->to_mss;
1004 to.to_wscale = toeo->to_wscale;
1005 to.to_flags = toeo->to_flags;
1007 INP_INFO_WLOCK(&V_tcbinfo);
1008 rc = syncache_expand(inc, &to, th, lsop, m);
1009 INP_INFO_WUNLOCK(&V_tcbinfo);
1015 * Given a LISTEN socket and an inbound SYN request, add
1016 * this to the syn cache, and send back a segment:
1017 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
1020 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
1021 * Doing so would require that we hold onto the data and deliver it
1022 * to the application. However, if we are the target of a SYN-flood
1023 * DoS attack, an attacker could send data which would eventually
1024 * consume all available buffer space if it were ACKed. By not ACKing
1025 * the data, we avoid this DoS scenario.
1028 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1029 struct inpcb *inp, struct socket **lsop, struct mbuf *m,
1030 struct toe_usrreqs *tu, void *toepcb)
1034 struct syncache *sc = NULL;
1035 struct syncache_head *sch;
1036 struct mbuf *ipopts = NULL;
1039 int win, sb_hiwat, ip_ttl, ip_tos;
1042 int autoflowlabel = 0;
1045 struct label *maclabel;
1047 struct syncache scs;
1050 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1051 INP_WLOCK_ASSERT(inp); /* listen socket */
1052 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
1053 ("%s: unexpected tcp flags", __func__));
1056 * Combine all so/tp operations very early to drop the INP lock as
1061 cred = crhold(so->so_cred);
1064 if ((inc->inc_flags & INC_ISIPV6) &&
1065 (inp->inp_flags & IN6P_AUTOFLOWLABEL))
1068 ip_ttl = inp->inp_ip_ttl;
1069 ip_tos = inp->inp_ip_tos;
1070 win = sbspace(&so->so_rcv);
1071 sb_hiwat = so->so_rcv.sb_hiwat;
1072 ltflags = (tp->t_flags & (TF_NOOPT | TF_SIGNATURE));
1074 /* By the time we drop the lock these should no longer be used. */
1079 if (mac_syncache_init(&maclabel) != 0) {
1081 INP_INFO_WUNLOCK(&V_tcbinfo);
1084 mac_syncache_create(maclabel, inp);
1087 INP_INFO_WUNLOCK(&V_tcbinfo);
1090 * Remember the IP options, if any.
1093 if (!(inc->inc_flags & INC_ISIPV6))
1096 ipopts = (m) ? ip_srcroute(m) : NULL;
1102 * See if we already have an entry for this connection.
1103 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1105 * XXX: should the syncache be re-initialized with the contents
1106 * of the new SYN here (which may have different options?)
1108 * XXX: We do not check the sequence number to see if this is a
1109 * real retransmit or a new connection attempt. The question is
1110 * how to handle such a case; either ignore it as spoofed, or
1111 * drop the current entry and create a new one?
1113 sc = syncache_lookup(inc, &sch); /* returns locked entry */
1114 SCH_LOCK_ASSERT(sch);
1116 #ifndef TCP_OFFLOAD_DISABLE
1118 sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT,
1121 TCPSTAT_INC(tcps_sc_dupsyn);
1124 * If we were remembering a previous source route,
1125 * forget it and use the new one we've been given.
1128 (void) m_free(sc->sc_ipopts);
1129 sc->sc_ipopts = ipopts;
1132 * Update timestamp if present.
1134 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
1135 sc->sc_tsreflect = to->to_tsval;
1137 sc->sc_flags &= ~SCF_TIMESTAMP;
1140 * Since we have already unconditionally allocated label
1141 * storage, free it up. The syncache entry will already
1142 * have an initialized label we can use.
1144 mac_syncache_destroy(&maclabel);
1146 /* Retransmit SYN|ACK and reset retransmit count. */
1147 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
1148 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
1149 "resetting timer and retransmitting SYN|ACK\n",
1153 if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) {
1155 syncache_timeout(sc, sch, 1);
1156 TCPSTAT_INC(tcps_sndacks);
1157 TCPSTAT_INC(tcps_sndtotal);
1163 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1166 * The zone allocator couldn't provide more entries.
1167 * Treat this as if the cache was full; drop the oldest
1168 * entry and insert the new one.
1170 TCPSTAT_INC(tcps_sc_zonefail);
1171 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1172 syncache_drop(sc, sch);
1173 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1175 if (V_tcp_syncookies) {
1176 bzero(&scs, sizeof(scs));
1181 (void) m_free(ipopts);
1188 * Fill in the syncache values.
1191 sc->sc_label = maclabel;
1195 sc->sc_ipopts = ipopts;
1196 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1198 if (!(inc->inc_flags & INC_ISIPV6))
1201 sc->sc_ip_tos = ip_tos;
1202 sc->sc_ip_ttl = ip_ttl;
1204 #ifndef TCP_OFFLOAD_DISABLE
1206 sc->sc_toepcb = toepcb;
1208 sc->sc_irs = th->th_seq;
1209 sc->sc_iss = arc4random();
1211 sc->sc_flowlabel = 0;
1214 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
1215 * win was derived from socket earlier in the function.
1218 win = imin(win, TCP_MAXWIN);
1221 if (V_tcp_do_rfc1323) {
1223 * A timestamp received in a SYN makes
1224 * it ok to send timestamp requests and replies.
1226 if (to->to_flags & TOF_TS) {
1227 sc->sc_tsreflect = to->to_tsval;
1229 sc->sc_flags |= SCF_TIMESTAMP;
1231 if (to->to_flags & TOF_SCALE) {
1235 * Pick the smallest possible scaling factor that
1236 * will still allow us to scale up to sb_max, aka
1237 * kern.ipc.maxsockbuf.
1239 * We do this because there are broken firewalls that
1240 * will corrupt the window scale option, leading to
1241 * the other endpoint believing that our advertised
1242 * window is unscaled. At scale factors larger than
1243 * 5 the unscaled window will drop below 1500 bytes,
1244 * leading to serious problems when traversing these
1247 * With the default maxsockbuf of 256K, a scale factor
1248 * of 3 will be chosen by this algorithm. Those who
1249 * choose a larger maxsockbuf should watch out
1250 * for the compatiblity problems mentioned above.
1252 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1253 * or <SYN,ACK>) segment itself is never scaled.
1255 while (wscale < TCP_MAX_WINSHIFT &&
1256 (TCP_MAXWIN << wscale) < sb_max)
1258 sc->sc_requested_r_scale = wscale;
1259 sc->sc_requested_s_scale = to->to_wscale;
1260 sc->sc_flags |= SCF_WINSCALE;
1263 #ifdef TCP_SIGNATURE
1265 * If listening socket requested TCP digests, and received SYN
1266 * contains the option, flag this in the syncache so that
1267 * syncache_respond() will do the right thing with the SYN+ACK.
1268 * XXX: Currently we always record the option by default and will
1269 * attempt to use it in syncache_respond().
1271 if (to->to_flags & TOF_SIGNATURE || ltflags & TF_SIGNATURE)
1272 sc->sc_flags |= SCF_SIGNATURE;
1274 if (to->to_flags & TOF_SACKPERM)
1275 sc->sc_flags |= SCF_SACK;
1276 if (to->to_flags & TOF_MSS)
1277 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */
1278 if (ltflags & TF_NOOPT)
1279 sc->sc_flags |= SCF_NOOPT;
1280 if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn)
1281 sc->sc_flags |= SCF_ECN;
1283 if (V_tcp_syncookies) {
1284 syncookie_generate(sch, sc, &flowtmp);
1287 sc->sc_flowlabel = flowtmp;
1293 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1299 * Do a standard 3-way handshake.
1301 if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) {
1302 if (V_tcp_syncookies && V_tcp_syncookiesonly && sc != &scs)
1304 else if (sc != &scs)
1305 syncache_insert(sc, sch); /* locks and unlocks sch */
1306 TCPSTAT_INC(tcps_sndacks);
1307 TCPSTAT_INC(tcps_sndtotal);
1311 TCPSTAT_INC(tcps_sc_dropped);
1319 mac_syncache_destroy(&maclabel);
1329 syncache_respond(struct syncache *sc)
1331 struct ip *ip = NULL;
1333 struct tcphdr *th = NULL;
1334 int optlen, error = 0; /* Make compiler happy */
1335 u_int16_t hlen, tlen, mssopt;
1338 struct ip6_hdr *ip6 = NULL;
1343 (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) :
1346 tlen = hlen + sizeof(struct tcphdr);
1348 /* Determine MSS we advertize to other end of connection. */
1349 mssopt = tcp_mssopt(&sc->sc_inc);
1350 if (sc->sc_peer_mss)
1351 mssopt = max( min(sc->sc_peer_mss, mssopt), V_tcp_minmss);
1353 /* XXX: Assume that the entire packet will fit in a header mbuf. */
1354 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1355 ("syncache: mbuf too small"));
1357 /* Create the IP+TCP header from scratch. */
1358 m = m_gethdr(M_DONTWAIT, MT_DATA);
1362 mac_syncache_create_mbuf(sc->sc_label, m);
1364 m->m_data += max_linkhdr;
1366 m->m_pkthdr.len = tlen;
1367 m->m_pkthdr.rcvif = NULL;
1370 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1371 ip6 = mtod(m, struct ip6_hdr *);
1372 ip6->ip6_vfc = IPV6_VERSION;
1373 ip6->ip6_nxt = IPPROTO_TCP;
1374 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1375 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1376 ip6->ip6_plen = htons(tlen - hlen);
1377 /* ip6_hlim is set after checksum */
1378 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1379 ip6->ip6_flow |= sc->sc_flowlabel;
1381 th = (struct tcphdr *)(ip6 + 1);
1384 #if defined(INET6) && defined(INET)
1389 ip = mtod(m, struct ip *);
1390 ip->ip_v = IPVERSION;
1391 ip->ip_hl = sizeof(struct ip) >> 2;
1396 ip->ip_p = IPPROTO_TCP;
1397 ip->ip_src = sc->sc_inc.inc_laddr;
1398 ip->ip_dst = sc->sc_inc.inc_faddr;
1399 ip->ip_ttl = sc->sc_ip_ttl;
1400 ip->ip_tos = sc->sc_ip_tos;
1403 * See if we should do MTU discovery. Route lookups are
1404 * expensive, so we will only unset the DF bit if:
1406 * 1) path_mtu_discovery is disabled
1407 * 2) the SCF_UNREACH flag has been set
1409 if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1410 ip->ip_off |= IP_DF;
1412 th = (struct tcphdr *)(ip + 1);
1415 th->th_sport = sc->sc_inc.inc_lport;
1416 th->th_dport = sc->sc_inc.inc_fport;
1418 th->th_seq = htonl(sc->sc_iss);
1419 th->th_ack = htonl(sc->sc_irs + 1);
1420 th->th_off = sizeof(struct tcphdr) >> 2;
1422 th->th_flags = TH_SYN|TH_ACK;
1423 th->th_win = htons(sc->sc_wnd);
1426 if (sc->sc_flags & SCF_ECN) {
1427 th->th_flags |= TH_ECE;
1428 TCPSTAT_INC(tcps_ecn_shs);
1431 /* Tack on the TCP options. */
1432 if ((sc->sc_flags & SCF_NOOPT) == 0) {
1436 to.to_flags = TOF_MSS;
1437 if (sc->sc_flags & SCF_WINSCALE) {
1438 to.to_wscale = sc->sc_requested_r_scale;
1439 to.to_flags |= TOF_SCALE;
1441 if (sc->sc_flags & SCF_TIMESTAMP) {
1442 /* Virgin timestamp or TCP cookie enhanced one. */
1443 to.to_tsval = sc->sc_ts;
1444 to.to_tsecr = sc->sc_tsreflect;
1445 to.to_flags |= TOF_TS;
1447 if (sc->sc_flags & SCF_SACK)
1448 to.to_flags |= TOF_SACKPERM;
1449 #ifdef TCP_SIGNATURE
1450 if (sc->sc_flags & SCF_SIGNATURE)
1451 to.to_flags |= TOF_SIGNATURE;
1453 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1455 /* Adjust headers by option size. */
1456 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1458 m->m_pkthdr.len += optlen;
1460 #ifdef TCP_SIGNATURE
1461 if (sc->sc_flags & SCF_SIGNATURE)
1462 tcp_signature_compute(m, 0, 0, optlen,
1463 to.to_signature, IPSEC_DIR_OUTBOUND);
1466 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1467 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1470 ip->ip_len += optlen;
1474 M_SETFIB(m, sc->sc_inc.inc_fibnum);
1476 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1478 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
1479 tlen + optlen - hlen);
1480 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1481 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1484 #if defined(INET6) && defined(INET)
1489 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1490 htons(tlen + optlen - hlen + IPPROTO_TCP));
1491 m->m_pkthdr.csum_flags = CSUM_TCP;
1492 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1493 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1500 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1501 struct inpcb *inp, struct socket **lsop, struct mbuf *m)
1503 _syncache_add(inc, to, th, inp, lsop, m, NULL, NULL);
1507 tcp_offload_syncache_add(struct in_conninfo *inc, struct toeopt *toeo,
1508 struct tcphdr *th, struct inpcb *inp, struct socket **lsop,
1509 struct toe_usrreqs *tu, void *toepcb)
1513 bzero(&to, sizeof(struct tcpopt));
1514 to.to_mss = toeo->to_mss;
1515 to.to_wscale = toeo->to_wscale;
1516 to.to_flags = toeo->to_flags;
1518 INP_INFO_WLOCK(&V_tcbinfo);
1521 _syncache_add(inc, &to, th, inp, lsop, NULL, tu, toepcb);
1525 * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1526 * receive and to be able to handle SYN floods from bogus source addresses
1527 * (where we will never receive any reply). SYN floods try to exhaust all
1528 * our memory and available slots in the SYN cache table to cause a denial
1529 * of service to legitimate users of the local host.
1531 * The idea of SYN cookies is to encode and include all necessary information
1532 * about the connection setup state within the SYN-ACK we send back and thus
1533 * to get along without keeping any local state until the ACK to the SYN-ACK
1534 * arrives (if ever). Everything we need to know should be available from
1535 * the information we encoded in the SYN-ACK.
1537 * More information about the theory behind SYN cookies and its first
1538 * discussion and specification can be found at:
1539 * http://cr.yp.to/syncookies.html (overview)
1540 * http://cr.yp.to/syncookies/archive (gory details)
1542 * This implementation extends the orginal idea and first implementation
1543 * of FreeBSD by using not only the initial sequence number field to store
1544 * information but also the timestamp field if present. This way we can
1545 * keep track of the entire state we need to know to recreate the session in
1546 * its original form. Almost all TCP speakers implement RFC1323 timestamps
1547 * these days. For those that do not we still have to live with the known
1548 * shortcomings of the ISN only SYN cookies.
1552 * Initial sequence number we send:
1553 * 31|................................|0
1554 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1555 * D = MD5 Digest (first dword)
1557 * R = Rotation of secret
1558 * P = Odd or Even secret
1560 * The MD5 Digest is computed with over following parameters:
1561 * a) randomly rotated secret
1562 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1563 * c) the received initial sequence number from remote host
1564 * d) the rotation offset and odd/even bit
1566 * Timestamp we send:
1567 * 31|................................|0
1568 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1569 * D = MD5 Digest (third dword) (only as filler)
1570 * S = Requested send window scale
1571 * R = Requested receive window scale
1573 * 5 = TCP-MD5 enabled (not implemented yet)
1574 * XORed with MD5 Digest (forth dword)
1576 * The timestamp isn't cryptographically secure and doesn't need to be.
1577 * The double use of the MD5 digest dwords ties it to a specific remote/
1578 * local host/port, remote initial sequence number and our local time
1579 * limited secret. A received timestamp is reverted (XORed) and then
1580 * the contained MD5 dword is compared to the computed one to ensure the
1581 * timestamp belongs to the SYN-ACK we sent. The other parameters may
1582 * have been tampered with but this isn't different from supplying bogus
1583 * values in the SYN in the first place.
1585 * Some problems with SYN cookies remain however:
1586 * Consider the problem of a recreated (and retransmitted) cookie. If the
1587 * original SYN was accepted, the connection is established. The second
1588 * SYN is inflight, and if it arrives with an ISN that falls within the
1589 * receive window, the connection is killed.
1592 * A heuristic to determine when to accept syn cookies is not necessary.
1593 * An ACK flood would cause the syncookie verification to be attempted,
1594 * but a SYN flood causes syncookies to be generated. Both are of equal
1595 * cost, so there's no point in trying to optimize the ACK flood case.
1596 * Also, if you don't process certain ACKs for some reason, then all someone
1597 * would have to do is launch a SYN and ACK flood at the same time, which
1598 * would stop cookie verification and defeat the entire purpose of syncookies.
1600 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1603 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1604 u_int32_t *flowlabel)
1607 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1610 u_int off, pmss, mss;
1613 SCH_LOCK_ASSERT(sch);
1615 /* Which of the two secrets to use. */
1616 secbits = sch->sch_oddeven ?
1617 sch->sch_secbits_odd : sch->sch_secbits_even;
1619 /* Reseed secret if too old. */
1620 if (sch->sch_reseed < time_uptime) {
1621 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */
1622 secbits = sch->sch_oddeven ?
1623 sch->sch_secbits_odd : sch->sch_secbits_even;
1624 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1625 secbits[i] = arc4random();
1626 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1629 /* Secret rotation offset. */
1630 off = sc->sc_iss & 0x7; /* iss was randomized before */
1632 /* Maximum segment size calculation. */
1634 max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), V_tcp_minmss);
1635 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1636 if (tcp_sc_msstab[mss] <= pmss)
1639 /* Fold parameters and MD5 digest into the ISN we will send. */
1640 data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1641 data |= off << 1; /* secret offset, derived from iss, 3 bits */
1642 data |= mss << 4; /* mss, 3 bits */
1645 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1646 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1647 MD5Update(&ctx, secbits, off);
1648 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1649 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1650 MD5Update(&ctx, &data, sizeof(data));
1651 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1653 data |= (md5_buffer[0] << 7);
1657 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1660 /* Additional parameters are stored in the timestamp if present. */
1661 if (sc->sc_flags & SCF_TIMESTAMP) {
1662 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1663 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1664 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */
1665 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */
1666 data |= md5_buffer[2] << 10; /* more digest bits */
1667 data ^= md5_buffer[3];
1669 sc->sc_tsoff = data - ticks; /* after XOR */
1672 TCPSTAT_INC(tcps_sc_sendcookie);
1675 static struct syncache *
1676 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1677 struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1681 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1685 int off, mss, wnd, flags;
1687 SCH_LOCK_ASSERT(sch);
1690 * Pull information out of SYN-ACK/ACK and
1691 * revert sequence number advances.
1693 ack = th->th_ack - 1;
1694 seq = th->th_seq - 1;
1695 off = (ack >> 1) & 0x7;
1696 mss = (ack >> 4) & 0x7;
1699 /* Which of the two secrets to use. */
1700 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1703 * The secret wasn't updated for the lifetime of a syncookie,
1704 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1706 if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) {
1710 /* Recompute the digest so we can compare it. */
1712 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1713 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1714 MD5Update(&ctx, secbits, off);
1715 MD5Update(&ctx, inc, sizeof(*inc));
1716 MD5Update(&ctx, &seq, sizeof(seq));
1717 MD5Update(&ctx, &flags, sizeof(flags));
1718 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1720 /* Does the digest part of or ACK'ed ISS match? */
1721 if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1724 /* Does the digest part of our reflected timestamp match? */
1725 if (to->to_flags & TOF_TS) {
1726 data = md5_buffer[3] ^ to->to_tsecr;
1727 if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1731 /* Fill in the syncache values. */
1732 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1733 sc->sc_ipopts = NULL;
1739 if (inc->inc_flags & INC_ISIPV6) {
1740 if (sotoinpcb(so)->inp_flags & IN6P_AUTOFLOWLABEL)
1741 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1745 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1746 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1749 /* Additional parameters that were encoded in the timestamp. */
1751 sc->sc_flags |= SCF_TIMESTAMP;
1752 sc->sc_tsreflect = to->to_tsval;
1753 sc->sc_ts = to->to_tsecr;
1754 sc->sc_tsoff = to->to_tsecr - ticks;
1755 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1756 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1757 sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1759 sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1761 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1762 sc->sc_flags |= SCF_WINSCALE;
1764 sc->sc_flags |= SCF_NOOPT;
1766 wnd = sbspace(&so->so_rcv);
1768 wnd = imin(wnd, TCP_MAXWIN);
1772 sc->sc_peer_mss = tcp_sc_msstab[mss];
1774 TCPSTAT_INC(tcps_sc_recvcookie);
1779 * Returns the current number of syncache entries. This number
1780 * will probably change before you get around to calling
1785 syncache_pcbcount(void)
1787 struct syncache_head *sch;
1790 for (count = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1791 /* No need to lock for a read. */
1792 sch = &V_tcp_syncache.hashbase[i];
1793 count += sch->sch_length;
1799 * Exports the syncache entries to userland so that netstat can display
1800 * them alongside the other sockets. This function is intended to be
1801 * called only from tcp_pcblist.
1803 * Due to concurrency on an active system, the number of pcbs exported
1804 * may have no relation to max_pcbs. max_pcbs merely indicates the
1805 * amount of space the caller allocated for this function to use.
1808 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
1811 struct syncache *sc;
1812 struct syncache_head *sch;
1813 int count, error, i;
1815 for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1816 sch = &V_tcp_syncache.hashbase[i];
1818 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
1819 if (count >= max_pcbs) {
1823 if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0)
1825 bzero(&xt, sizeof(xt));
1826 xt.xt_len = sizeof(xt);
1827 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1828 xt.xt_inp.inp_vflag = INP_IPV6;
1830 xt.xt_inp.inp_vflag = INP_IPV4;
1831 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
1832 xt.xt_tp.t_inpcb = &xt.xt_inp;
1833 xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
1834 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1835 xt.xt_socket.xso_len = sizeof (struct xsocket);
1836 xt.xt_socket.so_type = SOCK_STREAM;
1837 xt.xt_socket.so_state = SS_ISCONNECTING;
1838 error = SYSCTL_OUT(req, &xt, sizeof xt);
1848 *pcbs_exported = count;