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(struct tcp_syncache, tcp_syncache);
101 static VNET_DEFINE(int, tcp_syncookies);
102 static VNET_DEFINE(int, tcp_syncookiesonly);
103 VNET_DEFINE(int, tcp_sc_rst_sock_fail);
105 #define V_tcp_syncache VNET(tcp_syncache)
106 #define V_tcp_syncookies VNET(tcp_syncookies)
107 #define V_tcp_syncookiesonly VNET(tcp_syncookiesonly)
109 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
110 &VNET_NAME(tcp_syncookies), 0,
111 "Use TCP SYN cookies if the syncache overflows");
113 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW,
114 &VNET_NAME(tcp_syncookiesonly), 0,
115 "Use only TCP SYN cookies");
117 #ifdef TCP_OFFLOAD_DISABLE
118 #define TOEPCB_ISSET(sc) (0)
120 #define TOEPCB_ISSET(sc) ((sc)->sc_toepcb != NULL)
123 static void syncache_drop(struct syncache *, struct syncache_head *);
124 static void syncache_free(struct syncache *);
125 static void syncache_insert(struct syncache *, struct syncache_head *);
126 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
127 static int syncache_respond(struct syncache *);
128 static struct socket *syncache_socket(struct syncache *, struct socket *,
130 static void syncache_timeout(struct syncache *sc, struct syncache_head *sch,
132 static void syncache_timer(void *);
133 static void syncookie_generate(struct syncache_head *, struct syncache *,
135 static struct syncache
136 *syncookie_lookup(struct in_conninfo *, struct syncache_head *,
137 struct syncache *, struct tcpopt *, struct tcphdr *,
141 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
142 * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds,
143 * the odds are that the user has given up attempting to connect by then.
145 #define SYNCACHE_MAXREXMTS 3
147 /* Arbitrary values */
148 #define TCP_SYNCACHE_HASHSIZE 512
149 #define TCP_SYNCACHE_BUCKETLIMIT 30
151 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
153 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
154 &VNET_NAME(tcp_syncache.bucket_limit), 0,
155 "Per-bucket hash limit for syncache");
157 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
158 &VNET_NAME(tcp_syncache.cache_limit), 0,
159 "Overall entry limit for syncache");
161 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
162 &VNET_NAME(tcp_syncache.cache_count), 0,
163 "Current number of entries in syncache");
165 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
166 &VNET_NAME(tcp_syncache.hashsize), 0,
167 "Size of TCP syncache hashtable");
169 SYSCTL_VNET_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
170 &VNET_NAME(tcp_syncache.rexmt_limit), 0,
171 "Limit on SYN/ACK retransmissions");
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_syncookies = 1;
228 V_tcp_syncookiesonly = 0;
229 V_tcp_sc_rst_sock_fail = 1;
231 V_tcp_syncache.cache_count = 0;
232 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
233 V_tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
234 V_tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
235 V_tcp_syncache.hash_secret = arc4random();
237 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
238 &V_tcp_syncache.hashsize);
239 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
240 &V_tcp_syncache.bucket_limit);
241 if (!powerof2(V_tcp_syncache.hashsize) ||
242 V_tcp_syncache.hashsize == 0) {
243 printf("WARNING: syncache hash size is not a power of 2.\n");
244 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
246 V_tcp_syncache.hashmask = V_tcp_syncache.hashsize - 1;
249 V_tcp_syncache.cache_limit =
250 V_tcp_syncache.hashsize * V_tcp_syncache.bucket_limit;
251 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
252 &V_tcp_syncache.cache_limit);
254 /* Allocate the hash table. */
255 V_tcp_syncache.hashbase = malloc(V_tcp_syncache.hashsize *
256 sizeof(struct syncache_head), M_SYNCACHE, M_WAITOK | M_ZERO);
258 /* Initialize the hash buckets. */
259 for (i = 0; i < V_tcp_syncache.hashsize; i++) {
261 V_tcp_syncache.hashbase[i].sch_vnet = curvnet;
263 TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket);
264 mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
266 callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer,
267 &V_tcp_syncache.hashbase[i].sch_mtx, 0);
268 V_tcp_syncache.hashbase[i].sch_length = 0;
271 /* Create the syncache entry zone. */
272 V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
273 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
274 uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit);
279 syncache_destroy(void)
282 /* XXX walk the cache, free remaining objects, stop timers */
284 uma_zdestroy(V_tcp_syncache.zone);
285 FREE(V_tcp_syncache.hashbase, M_SYNCACHE);
290 * Inserts a syncache entry into the specified bucket row.
291 * Locks and unlocks the syncache_head autonomously.
294 syncache_insert(struct syncache *sc, struct syncache_head *sch)
296 struct syncache *sc2;
301 * Make sure that we don't overflow the per-bucket limit.
302 * If the bucket is full, toss the oldest element.
304 if (sch->sch_length >= V_tcp_syncache.bucket_limit) {
305 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
306 ("sch->sch_length incorrect"));
307 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
308 syncache_drop(sc2, sch);
309 TCPSTAT_INC(tcps_sc_bucketoverflow);
312 /* Put it into the bucket. */
313 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
316 /* Reinitialize the bucket row's timer. */
317 if (sch->sch_length == 1)
318 sch->sch_nextc = ticks + INT_MAX;
319 syncache_timeout(sc, sch, 1);
323 V_tcp_syncache.cache_count++;
324 TCPSTAT_INC(tcps_sc_added);
328 * Remove and free entry from syncache bucket row.
329 * Expects locked syncache head.
332 syncache_drop(struct syncache *sc, struct syncache_head *sch)
335 SCH_LOCK_ASSERT(sch);
337 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
340 #ifndef TCP_OFFLOAD_DISABLE
342 sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb);
345 V_tcp_syncache.cache_count--;
349 * Engage/reengage time on bucket row.
352 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
354 sc->sc_rxttime = ticks +
355 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
357 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
358 sch->sch_nextc = sc->sc_rxttime;
360 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
361 syncache_timer, (void *)sch);
366 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
367 * If we have retransmitted an entry the maximum number of times, expire it.
368 * One separate timer for each bucket row.
371 syncache_timer(void *xsch)
373 struct syncache_head *sch = (struct syncache_head *)xsch;
374 struct syncache *sc, *nsc;
378 CURVNET_SET(sch->sch_vnet);
380 /* NB: syncache_head has already been locked by the callout. */
381 SCH_LOCK_ASSERT(sch);
384 * In the following cycle we may remove some entries and/or
385 * advance some timeouts, so re-initialize the bucket timer.
387 sch->sch_nextc = tick + INT_MAX;
389 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
391 * We do not check if the listen socket still exists
392 * and accept the case where the listen socket may be
393 * gone by the time we resend the SYN/ACK. We do
394 * not expect this to happens often. If it does,
395 * then the RST will be sent by the time the remote
396 * host does the SYN/ACK->ACK.
398 if (TSTMP_GT(sc->sc_rxttime, tick)) {
399 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
400 sch->sch_nextc = sc->sc_rxttime;
403 if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) {
404 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
405 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
406 "giving up and removing syncache entry\n",
410 syncache_drop(sc, sch);
411 TCPSTAT_INC(tcps_sc_stale);
414 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
415 log(LOG_DEBUG, "%s; %s: Response timeout, "
416 "retransmitting (%u) SYN|ACK\n",
417 s, __func__, sc->sc_rxmits);
421 (void) syncache_respond(sc);
422 TCPSTAT_INC(tcps_sc_retransmitted);
423 syncache_timeout(sc, sch, 0);
425 if (!TAILQ_EMPTY(&(sch)->sch_bucket))
426 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
427 syncache_timer, (void *)(sch));
432 * Find an entry in the syncache.
433 * Returns always with locked syncache_head plus a matching entry or NULL.
436 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
439 struct syncache_head *sch;
442 if (inc->inc_flags & INC_ISIPV6) {
443 sch = &V_tcp_syncache.hashbase[
444 SYNCACHE_HASH6(inc, V_tcp_syncache.hashmask)];
449 /* Circle through bucket row to find matching entry. */
450 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
451 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
457 sch = &V_tcp_syncache.hashbase[
458 SYNCACHE_HASH(inc, V_tcp_syncache.hashmask)];
463 /* Circle through bucket row to find matching entry. */
464 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
466 if (sc->sc_inc.inc_flags & INC_ISIPV6)
469 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
473 SCH_LOCK_ASSERT(*schp);
474 return (NULL); /* always returns with locked sch */
478 * This function is called when we get a RST for a
479 * non-existent connection, so that we can see if the
480 * connection is in the syn cache. If it is, zap it.
483 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
486 struct syncache_head *sch;
489 sc = syncache_lookup(inc, &sch); /* returns locked sch */
490 SCH_LOCK_ASSERT(sch);
493 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
494 * See RFC 793 page 65, section SEGMENT ARRIVES.
496 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
497 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
498 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
499 "FIN flag set, segment ignored\n", s, __func__);
500 TCPSTAT_INC(tcps_badrst);
505 * No corresponding connection was found in syncache.
506 * If syncookies are enabled and possibly exclusively
507 * used, or we are under memory pressure, a valid RST
508 * may not find a syncache entry. In that case we're
509 * done and no SYN|ACK retransmissions will happen.
510 * Otherwise the the RST was misdirected or spoofed.
513 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
514 log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
515 "syncache entry (possibly syncookie only), "
516 "segment ignored\n", s, __func__);
517 TCPSTAT_INC(tcps_badrst);
522 * If the RST bit is set, check the sequence number to see
523 * if this is a valid reset segment.
525 * In all states except SYN-SENT, all reset (RST) segments
526 * are validated by checking their SEQ-fields. A reset is
527 * valid if its sequence number is in the window.
529 * The sequence number in the reset segment is normally an
530 * echo of our outgoing acknowlegement numbers, but some hosts
531 * send a reset with the sequence number at the rightmost edge
532 * of our receive window, and we have to handle this case.
534 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
535 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
536 syncache_drop(sc, sch);
537 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
538 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
539 "connection attempt aborted by remote endpoint\n",
541 TCPSTAT_INC(tcps_sc_reset);
543 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
544 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != "
545 "IRS %u (+WND %u), segment ignored\n",
546 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
547 TCPSTAT_INC(tcps_badrst);
557 syncache_badack(struct in_conninfo *inc)
560 struct syncache_head *sch;
562 sc = syncache_lookup(inc, &sch); /* returns locked sch */
563 SCH_LOCK_ASSERT(sch);
565 syncache_drop(sc, sch);
566 TCPSTAT_INC(tcps_sc_badack);
572 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
575 struct syncache_head *sch;
577 sc = syncache_lookup(inc, &sch); /* returns locked sch */
578 SCH_LOCK_ASSERT(sch);
582 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
583 if (ntohl(th->th_seq) != sc->sc_iss)
587 * If we've rertransmitted 3 times and this is our second error,
588 * we remove the entry. Otherwise, we allow it to continue on.
589 * This prevents us from incorrectly nuking an entry during a
590 * spurious network outage.
594 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
595 sc->sc_flags |= SCF_UNREACH;
598 syncache_drop(sc, sch);
599 TCPSTAT_INC(tcps_sc_unreach);
605 * Build a new TCP socket structure from a syncache entry.
607 static struct socket *
608 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
610 struct inpcb *inp = NULL;
615 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
618 * Ok, create the full blown connection, and set things up
619 * as they would have been set up if we had created the
620 * connection when the SYN arrived. If we can't create
621 * the connection, abort it.
623 so = sonewconn(lso, SS_ISCONNECTED);
626 * Drop the connection; we will either send a RST or
627 * have the peer retransmit its SYN again after its
630 TCPSTAT_INC(tcps_listendrop);
631 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
632 log(LOG_DEBUG, "%s; %s: Socket create failed "
633 "due to limits or memory shortage\n",
640 mac_socketpeer_set_from_mbuf(m, so);
644 inp->inp_inc.inc_fibnum = so->so_fibnum;
647 /* Insert new socket into PCB hash list. */
648 inp->inp_inc.inc_flags = sc->sc_inc.inc_flags;
650 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
651 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
653 inp->inp_vflag &= ~INP_IPV6;
654 inp->inp_vflag |= INP_IPV4;
656 inp->inp_laddr = sc->sc_inc.inc_laddr;
660 inp->inp_lport = sc->sc_inc.inc_lport;
661 if (in_pcbinshash(inp) != 0) {
663 * Undo the assignments above if we failed to
664 * put the PCB on the hash lists.
667 if (sc->sc_inc.inc_flags & INC_ISIPV6)
668 inp->in6p_laddr = in6addr_any;
671 inp->inp_laddr.s_addr = INADDR_ANY;
676 /* Copy old policy into new socket's. */
677 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
678 printf("syncache_socket: could not copy policy\n");
681 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
682 struct inpcb *oinp = sotoinpcb(lso);
683 struct in6_addr laddr6;
684 struct sockaddr_in6 sin6;
686 * Inherit socket options from the listening socket.
687 * Note that in6p_inputopts are not (and should not be)
688 * copied, since it stores previously received options and is
689 * used to detect if each new option is different than the
690 * previous one and hence should be passed to a user.
691 * If we copied in6p_inputopts, a user would not be able to
692 * receive options just after calling the accept system call.
694 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
695 if (oinp->in6p_outputopts)
696 inp->in6p_outputopts =
697 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
699 sin6.sin6_family = AF_INET6;
700 sin6.sin6_len = sizeof(sin6);
701 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
702 sin6.sin6_port = sc->sc_inc.inc_fport;
703 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
704 laddr6 = inp->in6p_laddr;
705 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
706 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
707 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
709 inp->in6p_laddr = laddr6;
712 /* Override flowlabel from in6_pcbconnect. */
713 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
714 inp->inp_flow |= sc->sc_flowlabel;
718 struct in_addr laddr;
719 struct sockaddr_in sin;
721 inp->inp_options = (m) ? ip_srcroute(m) : NULL;
723 if (inp->inp_options == NULL) {
724 inp->inp_options = sc->sc_ipopts;
725 sc->sc_ipopts = NULL;
728 sin.sin_family = AF_INET;
729 sin.sin_len = sizeof(sin);
730 sin.sin_addr = sc->sc_inc.inc_faddr;
731 sin.sin_port = sc->sc_inc.inc_fport;
732 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
733 laddr = inp->inp_laddr;
734 if (inp->inp_laddr.s_addr == INADDR_ANY)
735 inp->inp_laddr = sc->sc_inc.inc_laddr;
736 if (in_pcbconnect(inp, (struct sockaddr *)&sin,
738 inp->inp_laddr = laddr;
743 tp->t_state = TCPS_SYN_RECEIVED;
744 tp->iss = sc->sc_iss;
745 tp->irs = sc->sc_irs;
748 tp->snd_wl1 = sc->sc_irs;
749 tp->snd_max = tp->iss + 1;
750 tp->snd_nxt = tp->iss + 1;
751 tp->rcv_up = sc->sc_irs + 1;
752 tp->rcv_wnd = sc->sc_wnd;
753 tp->rcv_adv += tp->rcv_wnd;
754 tp->last_ack_sent = tp->rcv_nxt;
756 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
757 if (sc->sc_flags & SCF_NOOPT)
758 tp->t_flags |= TF_NOOPT;
760 if (sc->sc_flags & SCF_WINSCALE) {
761 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
762 tp->snd_scale = sc->sc_requested_s_scale;
763 tp->request_r_scale = sc->sc_requested_r_scale;
765 if (sc->sc_flags & SCF_TIMESTAMP) {
766 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
767 tp->ts_recent = sc->sc_tsreflect;
768 tp->ts_recent_age = ticks;
769 tp->ts_offset = sc->sc_tsoff;
772 if (sc->sc_flags & SCF_SIGNATURE)
773 tp->t_flags |= TF_SIGNATURE;
775 if (sc->sc_flags & SCF_SACK)
776 tp->t_flags |= TF_SACK_PERMIT;
779 if (sc->sc_flags & SCF_ECN)
780 tp->t_flags |= TF_ECN_PERMIT;
783 * Set up MSS and get cached values from tcp_hostcache.
784 * This might overwrite some of the defaults we just set.
786 tcp_mss(tp, sc->sc_peer_mss);
789 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
792 tp->snd_cwnd = tp->t_maxseg;
793 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
797 TCPSTAT_INC(tcps_accepts);
809 * This function gets called when we receive an ACK for a
810 * socket in the LISTEN state. We look up the connection
811 * in the syncache, and if its there, we pull it out of
812 * the cache and turn it into a full-blown connection in
813 * the SYN-RECEIVED state.
816 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
817 struct socket **lsop, struct mbuf *m)
820 struct syncache_head *sch;
825 * Global TCP locks are held because we manipulate the PCB lists
826 * and create a new socket.
828 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
829 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
830 ("%s: can handle only ACK", __func__));
832 sc = syncache_lookup(inc, &sch); /* returns locked sch */
833 SCH_LOCK_ASSERT(sch);
836 * There is no syncache entry, so see if this ACK is
837 * a returning syncookie. To do this, first:
838 * A. See if this socket has had a syncache entry dropped in
839 * the past. We don't want to accept a bogus syncookie
840 * if we've never received a SYN.
841 * B. check that the syncookie is valid. If it is, then
842 * cobble up a fake syncache entry, and return.
844 if (!V_tcp_syncookies) {
846 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
847 log(LOG_DEBUG, "%s; %s: Spurious ACK, "
848 "segment rejected (syncookies disabled)\n",
852 bzero(&scs, sizeof(scs));
853 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
856 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
857 log(LOG_DEBUG, "%s; %s: Segment failed "
858 "SYNCOOKIE authentication, segment rejected "
859 "(probably spoofed)\n", s, __func__);
863 /* Pull out the entry to unlock the bucket row. */
864 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
866 V_tcp_syncache.cache_count--;
871 * Segment validation:
872 * ACK must match our initial sequence number + 1 (the SYN|ACK).
874 if (th->th_ack != sc->sc_iss + 1 && !TOEPCB_ISSET(sc)) {
875 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
876 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment "
877 "rejected\n", s, __func__, th->th_ack, sc->sc_iss);
882 * The SEQ must fall in the window starting at the received
883 * initial receive sequence number + 1 (the SYN).
885 if ((SEQ_LEQ(th->th_seq, sc->sc_irs) ||
886 SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) &&
888 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
889 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment "
890 "rejected\n", s, __func__, th->th_seq, sc->sc_irs);
894 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
895 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
896 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
897 "segment rejected\n", s, __func__);
901 * If timestamps were negotiated the reflected timestamp
902 * must be equal to what we actually sent in the SYN|ACK.
904 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts &&
906 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
907 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
908 "segment rejected\n",
909 s, __func__, to->to_tsecr, sc->sc_ts);
913 *lsop = syncache_socket(sc, *lsop, m);
916 TCPSTAT_INC(tcps_sc_aborted);
918 TCPSTAT_INC(tcps_sc_completed);
920 /* how do we find the inp for the new socket? */
925 if (sc != NULL && sc != &scs)
934 tcp_offload_syncache_expand(struct in_conninfo *inc, struct toeopt *toeo,
935 struct tcphdr *th, struct socket **lsop, struct mbuf *m)
940 bzero(&to, sizeof(struct tcpopt));
941 to.to_mss = toeo->to_mss;
942 to.to_wscale = toeo->to_wscale;
943 to.to_flags = toeo->to_flags;
945 INP_INFO_WLOCK(&V_tcbinfo);
946 rc = syncache_expand(inc, &to, th, lsop, m);
947 INP_INFO_WUNLOCK(&V_tcbinfo);
953 * Given a LISTEN socket and an inbound SYN request, add
954 * this to the syn cache, and send back a segment:
955 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
958 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
959 * Doing so would require that we hold onto the data and deliver it
960 * to the application. However, if we are the target of a SYN-flood
961 * DoS attack, an attacker could send data which would eventually
962 * consume all available buffer space if it were ACKed. By not ACKing
963 * the data, we avoid this DoS scenario.
966 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
967 struct inpcb *inp, struct socket **lsop, struct mbuf *m,
968 struct toe_usrreqs *tu, void *toepcb)
972 struct syncache *sc = NULL;
973 struct syncache_head *sch;
974 struct mbuf *ipopts = NULL;
976 int win, sb_hiwat, ip_ttl, ip_tos, noopt;
979 int autoflowlabel = 0;
982 struct label *maclabel;
987 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
988 INP_WLOCK_ASSERT(inp); /* listen socket */
989 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
990 ("%s: unexpected tcp flags", __func__));
993 * Combine all so/tp operations very early to drop the INP lock as
998 cred = crhold(so->so_cred);
1001 if ((inc->inc_flags & INC_ISIPV6) &&
1002 (inp->inp_flags & IN6P_AUTOFLOWLABEL))
1005 ip_ttl = inp->inp_ip_ttl;
1006 ip_tos = inp->inp_ip_tos;
1007 win = sbspace(&so->so_rcv);
1008 sb_hiwat = so->so_rcv.sb_hiwat;
1009 noopt = (tp->t_flags & TF_NOOPT);
1011 /* By the time we drop the lock these should no longer be used. */
1016 if (mac_syncache_init(&maclabel) != 0) {
1018 INP_INFO_WUNLOCK(&V_tcbinfo);
1021 mac_syncache_create(maclabel, inp);
1024 INP_INFO_WUNLOCK(&V_tcbinfo);
1027 * Remember the IP options, if any.
1030 if (!(inc->inc_flags & INC_ISIPV6))
1032 ipopts = (m) ? ip_srcroute(m) : NULL;
1035 * See if we already have an entry for this connection.
1036 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1038 * XXX: should the syncache be re-initialized with the contents
1039 * of the new SYN here (which may have different options?)
1041 * XXX: We do not check the sequence number to see if this is a
1042 * real retransmit or a new connection attempt. The question is
1043 * how to handle such a case; either ignore it as spoofed, or
1044 * drop the current entry and create a new one?
1046 sc = syncache_lookup(inc, &sch); /* returns locked entry */
1047 SCH_LOCK_ASSERT(sch);
1049 #ifndef TCP_OFFLOAD_DISABLE
1051 sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT,
1054 TCPSTAT_INC(tcps_sc_dupsyn);
1057 * If we were remembering a previous source route,
1058 * forget it and use the new one we've been given.
1061 (void) m_free(sc->sc_ipopts);
1062 sc->sc_ipopts = ipopts;
1065 * Update timestamp if present.
1067 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
1068 sc->sc_tsreflect = to->to_tsval;
1070 sc->sc_flags &= ~SCF_TIMESTAMP;
1073 * Since we have already unconditionally allocated label
1074 * storage, free it up. The syncache entry will already
1075 * have an initialized label we can use.
1077 mac_syncache_destroy(&maclabel);
1079 /* Retransmit SYN|ACK and reset retransmit count. */
1080 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
1081 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
1082 "resetting timer and retransmitting SYN|ACK\n",
1086 if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) {
1088 syncache_timeout(sc, sch, 1);
1089 TCPSTAT_INC(tcps_sndacks);
1090 TCPSTAT_INC(tcps_sndtotal);
1096 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1099 * The zone allocator couldn't provide more entries.
1100 * Treat this as if the cache was full; drop the oldest
1101 * entry and insert the new one.
1103 TCPSTAT_INC(tcps_sc_zonefail);
1104 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1105 syncache_drop(sc, sch);
1106 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1108 if (V_tcp_syncookies) {
1109 bzero(&scs, sizeof(scs));
1114 (void) m_free(ipopts);
1121 * Fill in the syncache values.
1124 sc->sc_label = maclabel;
1128 sc->sc_ipopts = ipopts;
1129 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1131 if (!(inc->inc_flags & INC_ISIPV6))
1134 sc->sc_ip_tos = ip_tos;
1135 sc->sc_ip_ttl = ip_ttl;
1137 #ifndef TCP_OFFLOAD_DISABLE
1139 sc->sc_toepcb = toepcb;
1141 sc->sc_irs = th->th_seq;
1142 sc->sc_iss = arc4random();
1144 sc->sc_flowlabel = 0;
1147 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
1148 * win was derived from socket earlier in the function.
1151 win = imin(win, TCP_MAXWIN);
1154 if (V_tcp_do_rfc1323) {
1156 * A timestamp received in a SYN makes
1157 * it ok to send timestamp requests and replies.
1159 if (to->to_flags & TOF_TS) {
1160 sc->sc_tsreflect = to->to_tsval;
1162 sc->sc_flags |= SCF_TIMESTAMP;
1164 if (to->to_flags & TOF_SCALE) {
1168 * Pick the smallest possible scaling factor that
1169 * will still allow us to scale up to sb_max, aka
1170 * kern.ipc.maxsockbuf.
1172 * We do this because there are broken firewalls that
1173 * will corrupt the window scale option, leading to
1174 * the other endpoint believing that our advertised
1175 * window is unscaled. At scale factors larger than
1176 * 5 the unscaled window will drop below 1500 bytes,
1177 * leading to serious problems when traversing these
1180 * With the default maxsockbuf of 256K, a scale factor
1181 * of 3 will be chosen by this algorithm. Those who
1182 * choose a larger maxsockbuf should watch out
1183 * for the compatiblity problems mentioned above.
1185 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1186 * or <SYN,ACK>) segment itself is never scaled.
1188 while (wscale < TCP_MAX_WINSHIFT &&
1189 (TCP_MAXWIN << wscale) < sb_max)
1191 sc->sc_requested_r_scale = wscale;
1192 sc->sc_requested_s_scale = to->to_wscale;
1193 sc->sc_flags |= SCF_WINSCALE;
1196 #ifdef TCP_SIGNATURE
1198 * If listening socket requested TCP digests, and received SYN
1199 * contains the option, flag this in the syncache so that
1200 * syncache_respond() will do the right thing with the SYN+ACK.
1201 * XXX: Currently we always record the option by default and will
1202 * attempt to use it in syncache_respond().
1204 if (to->to_flags & TOF_SIGNATURE)
1205 sc->sc_flags |= SCF_SIGNATURE;
1207 if (to->to_flags & TOF_SACKPERM)
1208 sc->sc_flags |= SCF_SACK;
1209 if (to->to_flags & TOF_MSS)
1210 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */
1212 sc->sc_flags |= SCF_NOOPT;
1213 if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn)
1214 sc->sc_flags |= SCF_ECN;
1216 if (V_tcp_syncookies) {
1217 syncookie_generate(sch, sc, &flowtmp);
1220 sc->sc_flowlabel = flowtmp;
1226 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1232 * Do a standard 3-way handshake.
1234 if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) {
1235 if (V_tcp_syncookies && V_tcp_syncookiesonly && sc != &scs)
1237 else if (sc != &scs)
1238 syncache_insert(sc, sch); /* locks and unlocks sch */
1239 TCPSTAT_INC(tcps_sndacks);
1240 TCPSTAT_INC(tcps_sndtotal);
1244 TCPSTAT_INC(tcps_sc_dropped);
1252 mac_syncache_destroy(&maclabel);
1262 syncache_respond(struct syncache *sc)
1264 struct ip *ip = NULL;
1268 u_int16_t hlen, tlen, mssopt;
1271 struct ip6_hdr *ip6 = NULL;
1276 (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) :
1279 tlen = hlen + sizeof(struct tcphdr);
1281 /* Determine MSS we advertize to other end of connection. */
1282 mssopt = tcp_mssopt(&sc->sc_inc);
1283 if (sc->sc_peer_mss)
1284 mssopt = max( min(sc->sc_peer_mss, mssopt), V_tcp_minmss);
1286 /* XXX: Assume that the entire packet will fit in a header mbuf. */
1287 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1288 ("syncache: mbuf too small"));
1290 /* Create the IP+TCP header from scratch. */
1291 m = m_gethdr(M_DONTWAIT, MT_DATA);
1295 mac_syncache_create_mbuf(sc->sc_label, m);
1297 m->m_data += max_linkhdr;
1299 m->m_pkthdr.len = tlen;
1300 m->m_pkthdr.rcvif = NULL;
1303 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1304 ip6 = mtod(m, struct ip6_hdr *);
1305 ip6->ip6_vfc = IPV6_VERSION;
1306 ip6->ip6_nxt = IPPROTO_TCP;
1307 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1308 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1309 ip6->ip6_plen = htons(tlen - hlen);
1310 /* ip6_hlim is set after checksum */
1311 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1312 ip6->ip6_flow |= sc->sc_flowlabel;
1314 th = (struct tcphdr *)(ip6 + 1);
1318 ip = mtod(m, struct ip *);
1319 ip->ip_v = IPVERSION;
1320 ip->ip_hl = sizeof(struct ip) >> 2;
1325 ip->ip_p = IPPROTO_TCP;
1326 ip->ip_src = sc->sc_inc.inc_laddr;
1327 ip->ip_dst = sc->sc_inc.inc_faddr;
1328 ip->ip_ttl = sc->sc_ip_ttl;
1329 ip->ip_tos = sc->sc_ip_tos;
1332 * See if we should do MTU discovery. Route lookups are
1333 * expensive, so we will only unset the DF bit if:
1335 * 1) path_mtu_discovery is disabled
1336 * 2) the SCF_UNREACH flag has been set
1338 if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1339 ip->ip_off |= IP_DF;
1341 th = (struct tcphdr *)(ip + 1);
1343 th->th_sport = sc->sc_inc.inc_lport;
1344 th->th_dport = sc->sc_inc.inc_fport;
1346 th->th_seq = htonl(sc->sc_iss);
1347 th->th_ack = htonl(sc->sc_irs + 1);
1348 th->th_off = sizeof(struct tcphdr) >> 2;
1350 th->th_flags = TH_SYN|TH_ACK;
1351 th->th_win = htons(sc->sc_wnd);
1354 if (sc->sc_flags & SCF_ECN) {
1355 th->th_flags |= TH_ECE;
1356 TCPSTAT_INC(tcps_ecn_shs);
1359 /* Tack on the TCP options. */
1360 if ((sc->sc_flags & SCF_NOOPT) == 0) {
1364 to.to_flags = TOF_MSS;
1365 if (sc->sc_flags & SCF_WINSCALE) {
1366 to.to_wscale = sc->sc_requested_r_scale;
1367 to.to_flags |= TOF_SCALE;
1369 if (sc->sc_flags & SCF_TIMESTAMP) {
1370 /* Virgin timestamp or TCP cookie enhanced one. */
1371 to.to_tsval = sc->sc_ts;
1372 to.to_tsecr = sc->sc_tsreflect;
1373 to.to_flags |= TOF_TS;
1375 if (sc->sc_flags & SCF_SACK)
1376 to.to_flags |= TOF_SACKPERM;
1377 #ifdef TCP_SIGNATURE
1378 if (sc->sc_flags & SCF_SIGNATURE)
1379 to.to_flags |= TOF_SIGNATURE;
1381 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1383 /* Adjust headers by option size. */
1384 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1386 m->m_pkthdr.len += optlen;
1388 #ifdef TCP_SIGNATURE
1389 if (sc->sc_flags & SCF_SIGNATURE)
1390 tcp_signature_compute(m, 0, 0, optlen,
1391 to.to_signature, IPSEC_DIR_OUTBOUND);
1394 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1395 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1398 ip->ip_len += optlen;
1402 M_SETFIB(m, sc->sc_inc.inc_fibnum);
1404 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1406 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
1407 tlen + optlen - hlen);
1408 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1409 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1413 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1414 htons(tlen + optlen - hlen + IPPROTO_TCP));
1415 m->m_pkthdr.csum_flags = CSUM_TCP;
1416 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1417 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1423 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1424 struct inpcb *inp, struct socket **lsop, struct mbuf *m)
1426 _syncache_add(inc, to, th, inp, lsop, m, NULL, NULL);
1430 tcp_offload_syncache_add(struct in_conninfo *inc, struct toeopt *toeo,
1431 struct tcphdr *th, struct inpcb *inp, struct socket **lsop,
1432 struct toe_usrreqs *tu, void *toepcb)
1436 bzero(&to, sizeof(struct tcpopt));
1437 to.to_mss = toeo->to_mss;
1438 to.to_wscale = toeo->to_wscale;
1439 to.to_flags = toeo->to_flags;
1441 INP_INFO_WLOCK(&V_tcbinfo);
1444 _syncache_add(inc, &to, th, inp, lsop, NULL, tu, toepcb);
1448 * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1449 * receive and to be able to handle SYN floods from bogus source addresses
1450 * (where we will never receive any reply). SYN floods try to exhaust all
1451 * our memory and available slots in the SYN cache table to cause a denial
1452 * of service to legitimate users of the local host.
1454 * The idea of SYN cookies is to encode and include all necessary information
1455 * about the connection setup state within the SYN-ACK we send back and thus
1456 * to get along without keeping any local state until the ACK to the SYN-ACK
1457 * arrives (if ever). Everything we need to know should be available from
1458 * the information we encoded in the SYN-ACK.
1460 * More information about the theory behind SYN cookies and its first
1461 * discussion and specification can be found at:
1462 * http://cr.yp.to/syncookies.html (overview)
1463 * http://cr.yp.to/syncookies/archive (gory details)
1465 * This implementation extends the orginal idea and first implementation
1466 * of FreeBSD by using not only the initial sequence number field to store
1467 * information but also the timestamp field if present. This way we can
1468 * keep track of the entire state we need to know to recreate the session in
1469 * its original form. Almost all TCP speakers implement RFC1323 timestamps
1470 * these days. For those that do not we still have to live with the known
1471 * shortcomings of the ISN only SYN cookies.
1475 * Initial sequence number we send:
1476 * 31|................................|0
1477 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1478 * D = MD5 Digest (first dword)
1480 * R = Rotation of secret
1481 * P = Odd or Even secret
1483 * The MD5 Digest is computed with over following parameters:
1484 * a) randomly rotated secret
1485 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1486 * c) the received initial sequence number from remote host
1487 * d) the rotation offset and odd/even bit
1489 * Timestamp we send:
1490 * 31|................................|0
1491 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1492 * D = MD5 Digest (third dword) (only as filler)
1493 * S = Requested send window scale
1494 * R = Requested receive window scale
1496 * 5 = TCP-MD5 enabled (not implemented yet)
1497 * XORed with MD5 Digest (forth dword)
1499 * The timestamp isn't cryptographically secure and doesn't need to be.
1500 * The double use of the MD5 digest dwords ties it to a specific remote/
1501 * local host/port, remote initial sequence number and our local time
1502 * limited secret. A received timestamp is reverted (XORed) and then
1503 * the contained MD5 dword is compared to the computed one to ensure the
1504 * timestamp belongs to the SYN-ACK we sent. The other parameters may
1505 * have been tampered with but this isn't different from supplying bogus
1506 * values in the SYN in the first place.
1508 * Some problems with SYN cookies remain however:
1509 * Consider the problem of a recreated (and retransmitted) cookie. If the
1510 * original SYN was accepted, the connection is established. The second
1511 * SYN is inflight, and if it arrives with an ISN that falls within the
1512 * receive window, the connection is killed.
1515 * A heuristic to determine when to accept syn cookies is not necessary.
1516 * An ACK flood would cause the syncookie verification to be attempted,
1517 * but a SYN flood causes syncookies to be generated. Both are of equal
1518 * cost, so there's no point in trying to optimize the ACK flood case.
1519 * Also, if you don't process certain ACKs for some reason, then all someone
1520 * would have to do is launch a SYN and ACK flood at the same time, which
1521 * would stop cookie verification and defeat the entire purpose of syncookies.
1523 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1526 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1527 u_int32_t *flowlabel)
1530 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1533 u_int off, pmss, mss;
1536 SCH_LOCK_ASSERT(sch);
1538 /* Which of the two secrets to use. */
1539 secbits = sch->sch_oddeven ?
1540 sch->sch_secbits_odd : sch->sch_secbits_even;
1542 /* Reseed secret if too old. */
1543 if (sch->sch_reseed < time_uptime) {
1544 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */
1545 secbits = sch->sch_oddeven ?
1546 sch->sch_secbits_odd : sch->sch_secbits_even;
1547 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1548 secbits[i] = arc4random();
1549 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1552 /* Secret rotation offset. */
1553 off = sc->sc_iss & 0x7; /* iss was randomized before */
1555 /* Maximum segment size calculation. */
1557 max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), V_tcp_minmss);
1558 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1559 if (tcp_sc_msstab[mss] <= pmss)
1562 /* Fold parameters and MD5 digest into the ISN we will send. */
1563 data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1564 data |= off << 1; /* secret offset, derived from iss, 3 bits */
1565 data |= mss << 4; /* mss, 3 bits */
1568 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1569 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1570 MD5Update(&ctx, secbits, off);
1571 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1572 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1573 MD5Update(&ctx, &data, sizeof(data));
1574 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1576 data |= (md5_buffer[0] << 7);
1580 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1583 /* Additional parameters are stored in the timestamp if present. */
1584 if (sc->sc_flags & SCF_TIMESTAMP) {
1585 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1586 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1587 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */
1588 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */
1589 data |= md5_buffer[2] << 10; /* more digest bits */
1590 data ^= md5_buffer[3];
1592 sc->sc_tsoff = data - ticks; /* after XOR */
1595 TCPSTAT_INC(tcps_sc_sendcookie);
1598 static struct syncache *
1599 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1600 struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1604 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1608 int off, mss, wnd, flags;
1610 SCH_LOCK_ASSERT(sch);
1613 * Pull information out of SYN-ACK/ACK and
1614 * revert sequence number advances.
1616 ack = th->th_ack - 1;
1617 seq = th->th_seq - 1;
1618 off = (ack >> 1) & 0x7;
1619 mss = (ack >> 4) & 0x7;
1622 /* Which of the two secrets to use. */
1623 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1626 * The secret wasn't updated for the lifetime of a syncookie,
1627 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1629 if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) {
1633 /* Recompute the digest so we can compare it. */
1635 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1636 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1637 MD5Update(&ctx, secbits, off);
1638 MD5Update(&ctx, inc, sizeof(*inc));
1639 MD5Update(&ctx, &seq, sizeof(seq));
1640 MD5Update(&ctx, &flags, sizeof(flags));
1641 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1643 /* Does the digest part of or ACK'ed ISS match? */
1644 if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1647 /* Does the digest part of our reflected timestamp match? */
1648 if (to->to_flags & TOF_TS) {
1649 data = md5_buffer[3] ^ to->to_tsecr;
1650 if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1654 /* Fill in the syncache values. */
1655 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1656 sc->sc_ipopts = NULL;
1662 if (inc->inc_flags & INC_ISIPV6) {
1663 if (sotoinpcb(so)->inp_flags & IN6P_AUTOFLOWLABEL)
1664 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1668 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1669 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1672 /* Additional parameters that were encoded in the timestamp. */
1674 sc->sc_flags |= SCF_TIMESTAMP;
1675 sc->sc_tsreflect = to->to_tsval;
1676 sc->sc_ts = to->to_tsecr;
1677 sc->sc_tsoff = to->to_tsecr - ticks;
1678 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1679 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1680 sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1682 sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1684 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1685 sc->sc_flags |= SCF_WINSCALE;
1687 sc->sc_flags |= SCF_NOOPT;
1689 wnd = sbspace(&so->so_rcv);
1691 wnd = imin(wnd, TCP_MAXWIN);
1695 sc->sc_peer_mss = tcp_sc_msstab[mss];
1697 TCPSTAT_INC(tcps_sc_recvcookie);
1702 * Returns the current number of syncache entries. This number
1703 * will probably change before you get around to calling
1708 syncache_pcbcount(void)
1710 struct syncache_head *sch;
1713 for (count = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1714 /* No need to lock for a read. */
1715 sch = &V_tcp_syncache.hashbase[i];
1716 count += sch->sch_length;
1722 * Exports the syncache entries to userland so that netstat can display
1723 * them alongside the other sockets. This function is intended to be
1724 * called only from tcp_pcblist.
1726 * Due to concurrency on an active system, the number of pcbs exported
1727 * may have no relation to max_pcbs. max_pcbs merely indicates the
1728 * amount of space the caller allocated for this function to use.
1731 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
1734 struct syncache *sc;
1735 struct syncache_head *sch;
1736 int count, error, i;
1738 for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1739 sch = &V_tcp_syncache.hashbase[i];
1741 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
1742 if (count >= max_pcbs) {
1746 if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0)
1748 bzero(&xt, sizeof(xt));
1749 xt.xt_len = sizeof(xt);
1750 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1751 xt.xt_inp.inp_vflag = INP_IPV6;
1753 xt.xt_inp.inp_vflag = INP_IPV4;
1754 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
1755 xt.xt_tp.t_inpcb = &xt.xt_inp;
1756 xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
1757 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1758 xt.xt_socket.xso_len = sizeof (struct xsocket);
1759 xt.xt_socket.so_type = SOCK_STREAM;
1760 xt.xt_socket.so_state = SS_ISCONNECTING;
1761 error = SYSCTL_OUT(req, &xt, sizeof xt);
1771 *pcbs_exported = count;