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"
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>
57 #include <sys/vimage.h>
62 #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>
88 #include <netinet/vinet.h>
91 #include <netipsec/ipsec.h>
93 #include <netipsec/ipsec6.h>
95 #include <netipsec/key.h>
98 #include <machine/in_cksum.h>
100 #include <security/mac/mac_framework.h>
102 #ifdef VIMAGE_GLOBALS
103 static struct tcp_syncache tcp_syncache;
104 static int tcp_syncookies;
105 static int tcp_syncookiesonly;
106 int tcp_sc_rst_sock_fail;
109 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, syncookies,
110 CTLFLAG_RW, tcp_syncookies, 0,
111 "Use TCP SYN cookies if the syncache overflows");
113 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, syncookies_only,
114 CTLFLAG_RW, 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_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO,
154 bucketlimit, CTLFLAG_RDTUN,
155 tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
157 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO,
158 cachelimit, CTLFLAG_RDTUN,
159 tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
161 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO,
163 tcp_syncache.cache_count, 0, "Current number of entries in syncache");
165 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO,
166 hashsize, CTLFLAG_RDTUN,
167 tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
169 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO,
170 rexmtlimit, CTLFLAG_RW,
171 tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
173 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_syncache, OID_AUTO,
174 rst_on_sock_fail, CTLFLAG_RW,
175 tcp_sc_rst_sock_fail, 0, "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)
210 INIT_VNET_INET(curvnet);
213 (void) m_free(sc->sc_ipopts);
217 mac_syncache_destroy(&sc->sc_label);
220 uma_zfree(V_tcp_syncache.zone, sc);
226 INIT_VNET_INET(curvnet);
229 V_tcp_syncookies = 1;
230 V_tcp_syncookiesonly = 0;
231 V_tcp_sc_rst_sock_fail = 1;
233 V_tcp_syncache.cache_count = 0;
234 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
235 V_tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
236 V_tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
237 V_tcp_syncache.hash_secret = arc4random();
239 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
240 &V_tcp_syncache.hashsize);
241 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
242 &V_tcp_syncache.bucket_limit);
243 if (!powerof2(V_tcp_syncache.hashsize) ||
244 V_tcp_syncache.hashsize == 0) {
245 printf("WARNING: syncache hash size is not a power of 2.\n");
246 V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
248 V_tcp_syncache.hashmask = V_tcp_syncache.hashsize - 1;
251 V_tcp_syncache.cache_limit =
252 V_tcp_syncache.hashsize * V_tcp_syncache.bucket_limit;
253 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
254 &V_tcp_syncache.cache_limit);
256 /* Allocate the hash table. */
257 V_tcp_syncache.hashbase = malloc(V_tcp_syncache.hashsize *
258 sizeof(struct syncache_head), M_SYNCACHE, M_WAITOK | M_ZERO);
260 /* Initialize the hash buckets. */
261 for (i = 0; i < V_tcp_syncache.hashsize; i++) {
263 V_tcp_syncache.hashbase[i].sch_vnet = curvnet;
265 TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket);
266 mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
268 callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer,
269 &V_tcp_syncache.hashbase[i].sch_mtx, 0);
270 V_tcp_syncache.hashbase[i].sch_length = 0;
273 /* Create the syncache entry zone. */
274 V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
275 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
276 uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit);
280 * Inserts a syncache entry into the specified bucket row.
281 * Locks and unlocks the syncache_head autonomously.
284 syncache_insert(struct syncache *sc, struct syncache_head *sch)
286 INIT_VNET_INET(sch->sch_vnet);
287 struct syncache *sc2;
292 * Make sure that we don't overflow the per-bucket limit.
293 * If the bucket is full, toss the oldest element.
295 if (sch->sch_length >= V_tcp_syncache.bucket_limit) {
296 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
297 ("sch->sch_length incorrect"));
298 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
299 syncache_drop(sc2, sch);
300 TCPSTAT_INC(tcps_sc_bucketoverflow);
303 /* Put it into the bucket. */
304 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
307 /* Reinitialize the bucket row's timer. */
308 if (sch->sch_length == 1)
309 sch->sch_nextc = ticks + INT_MAX;
310 syncache_timeout(sc, sch, 1);
314 V_tcp_syncache.cache_count++;
315 TCPSTAT_INC(tcps_sc_added);
319 * Remove and free entry from syncache bucket row.
320 * Expects locked syncache head.
323 syncache_drop(struct syncache *sc, struct syncache_head *sch)
325 INIT_VNET_INET(sch->sch_vnet);
327 SCH_LOCK_ASSERT(sch);
329 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
332 #ifndef TCP_OFFLOAD_DISABLE
334 sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb);
337 V_tcp_syncache.cache_count--;
341 * Engage/reengage time on bucket row.
344 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
346 sc->sc_rxttime = ticks +
347 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
349 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
350 sch->sch_nextc = sc->sc_rxttime;
352 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
353 syncache_timer, (void *)sch);
358 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
359 * If we have retransmitted an entry the maximum number of times, expire it.
360 * One separate timer for each bucket row.
363 syncache_timer(void *xsch)
365 struct syncache_head *sch = (struct syncache_head *)xsch;
366 struct syncache *sc, *nsc;
370 CURVNET_SET(sch->sch_vnet);
371 INIT_VNET_INET(sch->sch_vnet);
373 /* NB: syncache_head has already been locked by the callout. */
374 SCH_LOCK_ASSERT(sch);
377 * In the following cycle we may remove some entries and/or
378 * advance some timeouts, so re-initialize the bucket timer.
380 sch->sch_nextc = tick + INT_MAX;
382 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
384 * We do not check if the listen socket still exists
385 * and accept the case where the listen socket may be
386 * gone by the time we resend the SYN/ACK. We do
387 * not expect this to happens often. If it does,
388 * then the RST will be sent by the time the remote
389 * host does the SYN/ACK->ACK.
391 if (TSTMP_GT(sc->sc_rxttime, tick)) {
392 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
393 sch->sch_nextc = sc->sc_rxttime;
396 if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) {
397 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
398 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
399 "giving up and removing syncache entry\n",
403 syncache_drop(sc, sch);
404 TCPSTAT_INC(tcps_sc_stale);
407 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
408 log(LOG_DEBUG, "%s; %s: Response timeout, "
409 "retransmitting (%u) SYN|ACK\n",
410 s, __func__, sc->sc_rxmits);
414 (void) syncache_respond(sc);
415 TCPSTAT_INC(tcps_sc_retransmitted);
416 syncache_timeout(sc, sch, 0);
418 if (!TAILQ_EMPTY(&(sch)->sch_bucket))
419 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
420 syncache_timer, (void *)(sch));
425 * Find an entry in the syncache.
426 * Returns always with locked syncache_head plus a matching entry or NULL.
429 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
431 INIT_VNET_INET(curvnet);
433 struct syncache_head *sch;
436 if (inc->inc_flags & INC_ISIPV6) {
437 sch = &V_tcp_syncache.hashbase[
438 SYNCACHE_HASH6(inc, V_tcp_syncache.hashmask)];
443 /* Circle through bucket row to find matching entry. */
444 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
445 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
451 sch = &V_tcp_syncache.hashbase[
452 SYNCACHE_HASH(inc, V_tcp_syncache.hashmask)];
457 /* Circle through bucket row to find matching entry. */
458 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
460 if (sc->sc_inc.inc_flags & INC_ISIPV6)
463 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
467 SCH_LOCK_ASSERT(*schp);
468 return (NULL); /* always returns with locked sch */
472 * This function is called when we get a RST for a
473 * non-existent connection, so that we can see if the
474 * connection is in the syn cache. If it is, zap it.
477 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
479 INIT_VNET_INET(curvnet);
481 struct syncache_head *sch;
484 sc = syncache_lookup(inc, &sch); /* returns locked sch */
485 SCH_LOCK_ASSERT(sch);
488 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
489 * See RFC 793 page 65, section SEGMENT ARRIVES.
491 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
492 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
493 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
494 "FIN flag set, segment ignored\n", s, __func__);
495 TCPSTAT_INC(tcps_badrst);
500 * No corresponding connection was found in syncache.
501 * If syncookies are enabled and possibly exclusively
502 * used, or we are under memory pressure, a valid RST
503 * may not find a syncache entry. In that case we're
504 * done and no SYN|ACK retransmissions will happen.
505 * Otherwise the the RST was misdirected or spoofed.
508 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
509 log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
510 "syncache entry (possibly syncookie only), "
511 "segment ignored\n", s, __func__);
512 TCPSTAT_INC(tcps_badrst);
517 * If the RST bit is set, check the sequence number to see
518 * if this is a valid reset segment.
520 * In all states except SYN-SENT, all reset (RST) segments
521 * are validated by checking their SEQ-fields. A reset is
522 * valid if its sequence number is in the window.
524 * The sequence number in the reset segment is normally an
525 * echo of our outgoing acknowlegement numbers, but some hosts
526 * send a reset with the sequence number at the rightmost edge
527 * of our receive window, and we have to handle this case.
529 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
530 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
531 syncache_drop(sc, sch);
532 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
533 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
534 "connection attempt aborted by remote endpoint\n",
536 TCPSTAT_INC(tcps_sc_reset);
538 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
539 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != "
540 "IRS %u (+WND %u), segment ignored\n",
541 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
542 TCPSTAT_INC(tcps_badrst);
552 syncache_badack(struct in_conninfo *inc)
554 INIT_VNET_INET(curvnet);
556 struct syncache_head *sch;
558 sc = syncache_lookup(inc, &sch); /* returns locked sch */
559 SCH_LOCK_ASSERT(sch);
561 syncache_drop(sc, sch);
562 TCPSTAT_INC(tcps_sc_badack);
568 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
570 INIT_VNET_INET(curvnet);
572 struct syncache_head *sch;
574 sc = syncache_lookup(inc, &sch); /* returns locked sch */
575 SCH_LOCK_ASSERT(sch);
579 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
580 if (ntohl(th->th_seq) != sc->sc_iss)
584 * If we've rertransmitted 3 times and this is our second error,
585 * we remove the entry. Otherwise, we allow it to continue on.
586 * This prevents us from incorrectly nuking an entry during a
587 * spurious network outage.
591 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
592 sc->sc_flags |= SCF_UNREACH;
595 syncache_drop(sc, sch);
596 TCPSTAT_INC(tcps_sc_unreach);
602 * Build a new TCP socket structure from a syncache entry.
604 static struct socket *
605 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
607 INIT_VNET_INET(lso->so_vnet);
608 struct inpcb *inp = NULL;
613 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
616 * Ok, create the full blown connection, and set things up
617 * as they would have been set up if we had created the
618 * connection when the SYN arrived. If we can't create
619 * the connection, abort it.
621 so = sonewconn(lso, SS_ISCONNECTED);
624 * Drop the connection; we will either send a RST or
625 * have the peer retransmit its SYN again after its
628 TCPSTAT_INC(tcps_listendrop);
629 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
630 log(LOG_DEBUG, "%s; %s: Socket create failed "
631 "due to limits or memory shortage\n",
638 mac_socketpeer_set_from_mbuf(m, so);
642 inp->inp_inc.inc_fibnum = sc->sc_inc.inc_fibnum;
643 so->so_fibnum = sc->sc_inc.inc_fibnum;
646 /* Insert new socket into PCB hash list. */
647 inp->inp_inc.inc_flags = sc->sc_inc.inc_flags;
649 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
650 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
652 inp->inp_vflag &= ~INP_IPV6;
653 inp->inp_vflag |= INP_IPV4;
655 inp->inp_laddr = sc->sc_inc.inc_laddr;
659 inp->inp_lport = sc->sc_inc.inc_lport;
660 if (in_pcbinshash(inp) != 0) {
662 * Undo the assignments above if we failed to
663 * put the PCB on the hash lists.
666 if (sc->sc_inc.inc_flags & INC_ISIPV6)
667 inp->in6p_laddr = in6addr_any;
670 inp->inp_laddr.s_addr = INADDR_ANY;
675 /* Copy old policy into new socket's. */
676 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
677 printf("syncache_socket: could not copy policy\n");
680 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
681 struct inpcb *oinp = sotoinpcb(lso);
682 struct in6_addr laddr6;
683 struct sockaddr_in6 sin6;
685 * Inherit socket options from the listening socket.
686 * Note that in6p_inputopts are not (and should not be)
687 * copied, since it stores previously received options and is
688 * used to detect if each new option is different than the
689 * previous one and hence should be passed to a user.
690 * If we copied in6p_inputopts, a user would not be able to
691 * receive options just after calling the accept system call.
693 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
694 if (oinp->in6p_outputopts)
695 inp->in6p_outputopts =
696 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
698 sin6.sin6_family = AF_INET6;
699 sin6.sin6_len = sizeof(sin6);
700 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
701 sin6.sin6_port = sc->sc_inc.inc_fport;
702 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
703 laddr6 = inp->in6p_laddr;
704 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
705 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
706 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
708 inp->in6p_laddr = laddr6;
711 /* Override flowlabel from in6_pcbconnect. */
712 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
713 inp->inp_flow |= sc->sc_flowlabel;
717 struct in_addr laddr;
718 struct sockaddr_in sin;
720 inp->inp_options = (m) ? ip_srcroute(m) : NULL;
722 if (inp->inp_options == NULL) {
723 inp->inp_options = sc->sc_ipopts;
724 sc->sc_ipopts = NULL;
727 sin.sin_family = AF_INET;
728 sin.sin_len = sizeof(sin);
729 sin.sin_addr = sc->sc_inc.inc_faddr;
730 sin.sin_port = sc->sc_inc.inc_fport;
731 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
732 laddr = inp->inp_laddr;
733 if (inp->inp_laddr.s_addr == INADDR_ANY)
734 inp->inp_laddr = sc->sc_inc.inc_laddr;
735 if (in_pcbconnect(inp, (struct sockaddr *)&sin,
737 inp->inp_laddr = laddr;
742 tp->t_state = TCPS_SYN_RECEIVED;
743 tp->iss = sc->sc_iss;
744 tp->irs = sc->sc_irs;
747 tp->snd_wl1 = sc->sc_irs;
748 tp->snd_max = tp->iss + 1;
749 tp->snd_nxt = tp->iss + 1;
750 tp->rcv_up = sc->sc_irs + 1;
751 tp->rcv_wnd = sc->sc_wnd;
752 tp->rcv_adv += tp->rcv_wnd;
753 tp->last_ack_sent = tp->rcv_nxt;
755 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
756 if (sc->sc_flags & SCF_NOOPT)
757 tp->t_flags |= TF_NOOPT;
759 if (sc->sc_flags & SCF_WINSCALE) {
760 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
761 tp->snd_scale = sc->sc_requested_s_scale;
762 tp->request_r_scale = sc->sc_requested_r_scale;
764 if (sc->sc_flags & SCF_TIMESTAMP) {
765 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
766 tp->ts_recent = sc->sc_tsreflect;
767 tp->ts_recent_age = ticks;
768 tp->ts_offset = sc->sc_tsoff;
771 if (sc->sc_flags & SCF_SIGNATURE)
772 tp->t_flags |= TF_SIGNATURE;
774 if (sc->sc_flags & SCF_SACK)
775 tp->t_flags |= TF_SACK_PERMIT;
778 if (sc->sc_flags & SCF_ECN)
779 tp->t_flags |= TF_ECN_PERMIT;
782 * Set up MSS and get cached values from tcp_hostcache.
783 * This might overwrite some of the defaults we just set.
785 tcp_mss(tp, sc->sc_peer_mss);
788 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
791 tp->snd_cwnd = tp->t_maxseg;
792 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
796 TCPSTAT_INC(tcps_accepts);
808 * This function gets called when we receive an ACK for a
809 * socket in the LISTEN state. We look up the connection
810 * in the syncache, and if its there, we pull it out of
811 * the cache and turn it into a full-blown connection in
812 * the SYN-RECEIVED state.
815 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
816 struct socket **lsop, struct mbuf *m)
818 INIT_VNET_INET(curvnet);
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 tcpopt *to,
935 struct tcphdr *th, struct socket **lsop, struct mbuf *m)
937 INIT_VNET_INET(curvnet);
940 INP_INFO_WLOCK(&V_tcbinfo);
941 rc = syncache_expand(inc, to, th, lsop, m);
942 INP_INFO_WUNLOCK(&V_tcbinfo);
948 * Given a LISTEN socket and an inbound SYN request, add
949 * this to the syn cache, and send back a segment:
950 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
953 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
954 * Doing so would require that we hold onto the data and deliver it
955 * to the application. However, if we are the target of a SYN-flood
956 * DoS attack, an attacker could send data which would eventually
957 * consume all available buffer space if it were ACKed. By not ACKing
958 * the data, we avoid this DoS scenario.
961 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
962 struct inpcb *inp, struct socket **lsop, struct mbuf *m,
963 struct toe_usrreqs *tu, void *toepcb)
965 INIT_VNET_INET(inp->inp_vnet);
968 struct syncache *sc = NULL;
969 struct syncache_head *sch;
970 struct mbuf *ipopts = NULL;
972 int win, sb_hiwat, ip_ttl, ip_tos, noopt;
975 int autoflowlabel = 0;
978 struct label *maclabel;
983 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
984 INP_WLOCK_ASSERT(inp); /* listen socket */
985 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
986 ("%s: unexpected tcp flags", __func__));
989 * Combine all so/tp operations very early to drop the INP lock as
994 cred = crhold(so->so_cred);
997 if ((inc->inc_flags & INC_ISIPV6) &&
998 (inp->inp_flags & IN6P_AUTOFLOWLABEL))
1001 ip_ttl = inp->inp_ip_ttl;
1002 ip_tos = inp->inp_ip_tos;
1003 win = sbspace(&so->so_rcv);
1004 sb_hiwat = so->so_rcv.sb_hiwat;
1005 noopt = (tp->t_flags & TF_NOOPT);
1007 /* By the time we drop the lock these should no longer be used. */
1012 if (mac_syncache_init(&maclabel) != 0) {
1014 INP_INFO_WUNLOCK(&V_tcbinfo);
1017 mac_syncache_create(maclabel, inp);
1020 INP_INFO_WUNLOCK(&V_tcbinfo);
1023 * Remember the IP options, if any.
1026 if (!(inc->inc_flags & INC_ISIPV6))
1028 ipopts = (m) ? ip_srcroute(m) : NULL;
1031 * See if we already have an entry for this connection.
1032 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1034 * XXX: should the syncache be re-initialized with the contents
1035 * of the new SYN here (which may have different options?)
1037 * XXX: We do not check the sequence number to see if this is a
1038 * real retransmit or a new connection attempt. The question is
1039 * how to handle such a case; either ignore it as spoofed, or
1040 * drop the current entry and create a new one?
1042 sc = syncache_lookup(inc, &sch); /* returns locked entry */
1043 SCH_LOCK_ASSERT(sch);
1045 #ifndef TCP_OFFLOAD_DISABLE
1047 sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT,
1050 TCPSTAT_INC(tcps_sc_dupsyn);
1053 * If we were remembering a previous source route,
1054 * forget it and use the new one we've been given.
1057 (void) m_free(sc->sc_ipopts);
1058 sc->sc_ipopts = ipopts;
1061 * Update timestamp if present.
1063 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
1064 sc->sc_tsreflect = to->to_tsval;
1066 sc->sc_flags &= ~SCF_TIMESTAMP;
1069 * Since we have already unconditionally allocated label
1070 * storage, free it up. The syncache entry will already
1071 * have an initialized label we can use.
1073 mac_syncache_destroy(&maclabel);
1075 /* Retransmit SYN|ACK and reset retransmit count. */
1076 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
1077 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
1078 "resetting timer and retransmitting SYN|ACK\n",
1082 if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) {
1084 syncache_timeout(sc, sch, 1);
1085 TCPSTAT_INC(tcps_sndacks);
1086 TCPSTAT_INC(tcps_sndtotal);
1092 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1095 * The zone allocator couldn't provide more entries.
1096 * Treat this as if the cache was full; drop the oldest
1097 * entry and insert the new one.
1099 TCPSTAT_INC(tcps_sc_zonefail);
1100 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1101 syncache_drop(sc, sch);
1102 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1104 if (V_tcp_syncookies) {
1105 bzero(&scs, sizeof(scs));
1110 (void) m_free(ipopts);
1117 * Fill in the syncache values.
1120 sc->sc_label = maclabel;
1124 sc->sc_ipopts = ipopts;
1125 /* XXX-BZ this fib assignment is just useless. */
1126 sc->sc_inc.inc_fibnum = inp->inp_inc.inc_fibnum;
1127 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1129 if (!(inc->inc_flags & INC_ISIPV6))
1132 sc->sc_ip_tos = ip_tos;
1133 sc->sc_ip_ttl = ip_ttl;
1135 #ifndef TCP_OFFLOAD_DISABLE
1137 sc->sc_toepcb = toepcb;
1139 sc->sc_irs = th->th_seq;
1140 sc->sc_iss = arc4random();
1142 sc->sc_flowlabel = 0;
1145 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
1146 * win was derived from socket earlier in the function.
1149 win = imin(win, TCP_MAXWIN);
1152 if (V_tcp_do_rfc1323) {
1154 * A timestamp received in a SYN makes
1155 * it ok to send timestamp requests and replies.
1157 if (to->to_flags & TOF_TS) {
1158 sc->sc_tsreflect = to->to_tsval;
1160 sc->sc_flags |= SCF_TIMESTAMP;
1162 if (to->to_flags & TOF_SCALE) {
1166 * Pick the smallest possible scaling factor that
1167 * will still allow us to scale up to sb_max, aka
1168 * kern.ipc.maxsockbuf.
1170 * We do this because there are broken firewalls that
1171 * will corrupt the window scale option, leading to
1172 * the other endpoint believing that our advertised
1173 * window is unscaled. At scale factors larger than
1174 * 5 the unscaled window will drop below 1500 bytes,
1175 * leading to serious problems when traversing these
1178 * With the default maxsockbuf of 256K, a scale factor
1179 * of 3 will be chosen by this algorithm. Those who
1180 * choose a larger maxsockbuf should watch out
1181 * for the compatiblity problems mentioned above.
1183 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1184 * or <SYN,ACK>) segment itself is never scaled.
1186 while (wscale < TCP_MAX_WINSHIFT &&
1187 (TCP_MAXWIN << wscale) < sb_max)
1189 sc->sc_requested_r_scale = wscale;
1190 sc->sc_requested_s_scale = to->to_wscale;
1191 sc->sc_flags |= SCF_WINSCALE;
1194 #ifdef TCP_SIGNATURE
1196 * If listening socket requested TCP digests, and received SYN
1197 * contains the option, flag this in the syncache so that
1198 * syncache_respond() will do the right thing with the SYN+ACK.
1199 * XXX: Currently we always record the option by default and will
1200 * attempt to use it in syncache_respond().
1202 if (to->to_flags & TOF_SIGNATURE)
1203 sc->sc_flags |= SCF_SIGNATURE;
1205 if (to->to_flags & TOF_SACKPERM)
1206 sc->sc_flags |= SCF_SACK;
1207 if (to->to_flags & TOF_MSS)
1208 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */
1210 sc->sc_flags |= SCF_NOOPT;
1211 if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn)
1212 sc->sc_flags |= SCF_ECN;
1214 if (V_tcp_syncookies) {
1215 syncookie_generate(sch, sc, &flowtmp);
1218 sc->sc_flowlabel = flowtmp;
1224 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1230 * Do a standard 3-way handshake.
1232 if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) {
1233 if (V_tcp_syncookies && V_tcp_syncookiesonly && sc != &scs)
1235 else if (sc != &scs)
1236 syncache_insert(sc, sch); /* locks and unlocks sch */
1237 TCPSTAT_INC(tcps_sndacks);
1238 TCPSTAT_INC(tcps_sndtotal);
1242 TCPSTAT_INC(tcps_sc_dropped);
1250 mac_syncache_destroy(&maclabel);
1260 syncache_respond(struct syncache *sc)
1262 INIT_VNET_INET(curvnet);
1263 struct ip *ip = NULL;
1267 u_int16_t hlen, tlen, mssopt;
1270 struct ip6_hdr *ip6 = NULL;
1275 (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) :
1278 tlen = hlen + sizeof(struct tcphdr);
1280 /* Determine MSS we advertize to other end of connection. */
1281 mssopt = tcp_mssopt(&sc->sc_inc);
1282 if (sc->sc_peer_mss)
1283 mssopt = max( min(sc->sc_peer_mss, mssopt), V_tcp_minmss);
1285 /* XXX: Assume that the entire packet will fit in a header mbuf. */
1286 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1287 ("syncache: mbuf too small"));
1289 /* Create the IP+TCP header from scratch. */
1290 m = m_gethdr(M_DONTWAIT, MT_DATA);
1294 mac_syncache_create_mbuf(sc->sc_label, m);
1296 m->m_data += max_linkhdr;
1298 m->m_pkthdr.len = tlen;
1299 m->m_pkthdr.rcvif = NULL;
1302 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1303 ip6 = mtod(m, struct ip6_hdr *);
1304 ip6->ip6_vfc = IPV6_VERSION;
1305 ip6->ip6_nxt = IPPROTO_TCP;
1306 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1307 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1308 ip6->ip6_plen = htons(tlen - hlen);
1309 /* ip6_hlim is set after checksum */
1310 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1311 ip6->ip6_flow |= sc->sc_flowlabel;
1313 th = (struct tcphdr *)(ip6 + 1);
1317 ip = mtod(m, struct ip *);
1318 ip->ip_v = IPVERSION;
1319 ip->ip_hl = sizeof(struct ip) >> 2;
1324 ip->ip_p = IPPROTO_TCP;
1325 ip->ip_src = sc->sc_inc.inc_laddr;
1326 ip->ip_dst = sc->sc_inc.inc_faddr;
1327 ip->ip_ttl = sc->sc_ip_ttl;
1328 ip->ip_tos = sc->sc_ip_tos;
1331 * See if we should do MTU discovery. Route lookups are
1332 * expensive, so we will only unset the DF bit if:
1334 * 1) path_mtu_discovery is disabled
1335 * 2) the SCF_UNREACH flag has been set
1337 if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1338 ip->ip_off |= IP_DF;
1340 th = (struct tcphdr *)(ip + 1);
1342 th->th_sport = sc->sc_inc.inc_lport;
1343 th->th_dport = sc->sc_inc.inc_fport;
1345 th->th_seq = htonl(sc->sc_iss);
1346 th->th_ack = htonl(sc->sc_irs + 1);
1347 th->th_off = sizeof(struct tcphdr) >> 2;
1349 th->th_flags = TH_SYN|TH_ACK;
1350 th->th_win = htons(sc->sc_wnd);
1353 if (sc->sc_flags & SCF_ECN) {
1354 th->th_flags |= TH_ECE;
1355 TCPSTAT_INC(tcps_ecn_shs);
1358 /* Tack on the TCP options. */
1359 if ((sc->sc_flags & SCF_NOOPT) == 0) {
1363 to.to_flags = TOF_MSS;
1364 if (sc->sc_flags & SCF_WINSCALE) {
1365 to.to_wscale = sc->sc_requested_r_scale;
1366 to.to_flags |= TOF_SCALE;
1368 if (sc->sc_flags & SCF_TIMESTAMP) {
1369 /* Virgin timestamp or TCP cookie enhanced one. */
1370 to.to_tsval = sc->sc_ts;
1371 to.to_tsecr = sc->sc_tsreflect;
1372 to.to_flags |= TOF_TS;
1374 if (sc->sc_flags & SCF_SACK)
1375 to.to_flags |= TOF_SACKPERM;
1376 #ifdef TCP_SIGNATURE
1377 if (sc->sc_flags & SCF_SIGNATURE)
1378 to.to_flags |= TOF_SIGNATURE;
1380 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1382 /* Adjust headers by option size. */
1383 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1385 m->m_pkthdr.len += optlen;
1387 #ifdef TCP_SIGNATURE
1388 if (sc->sc_flags & SCF_SIGNATURE)
1389 tcp_signature_compute(m, 0, 0, optlen,
1390 to.to_signature, IPSEC_DIR_OUTBOUND);
1393 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1394 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1397 ip->ip_len += optlen;
1402 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1404 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
1405 tlen + optlen - hlen);
1406 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1407 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1411 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1412 htons(tlen + optlen - hlen + IPPROTO_TCP));
1413 m->m_pkthdr.csum_flags = CSUM_TCP;
1414 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1415 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1421 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1422 struct inpcb *inp, struct socket **lsop, struct mbuf *m)
1424 _syncache_add(inc, to, th, inp, lsop, m, NULL, NULL);
1428 tcp_offload_syncache_add(struct in_conninfo *inc, struct tcpopt *to,
1429 struct tcphdr *th, struct inpcb *inp, struct socket **lsop,
1430 struct toe_usrreqs *tu, void *toepcb)
1432 INIT_VNET_INET(curvnet);
1434 INP_INFO_WLOCK(&V_tcbinfo);
1436 _syncache_add(inc, to, th, inp, lsop, NULL, tu, toepcb);
1440 * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1441 * receive and to be able to handle SYN floods from bogus source addresses
1442 * (where we will never receive any reply). SYN floods try to exhaust all
1443 * our memory and available slots in the SYN cache table to cause a denial
1444 * of service to legitimate users of the local host.
1446 * The idea of SYN cookies is to encode and include all necessary information
1447 * about the connection setup state within the SYN-ACK we send back and thus
1448 * to get along without keeping any local state until the ACK to the SYN-ACK
1449 * arrives (if ever). Everything we need to know should be available from
1450 * the information we encoded in the SYN-ACK.
1452 * More information about the theory behind SYN cookies and its first
1453 * discussion and specification can be found at:
1454 * http://cr.yp.to/syncookies.html (overview)
1455 * http://cr.yp.to/syncookies/archive (gory details)
1457 * This implementation extends the orginal idea and first implementation
1458 * of FreeBSD by using not only the initial sequence number field to store
1459 * information but also the timestamp field if present. This way we can
1460 * keep track of the entire state we need to know to recreate the session in
1461 * its original form. Almost all TCP speakers implement RFC1323 timestamps
1462 * these days. For those that do not we still have to live with the known
1463 * shortcomings of the ISN only SYN cookies.
1467 * Initial sequence number we send:
1468 * 31|................................|0
1469 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1470 * D = MD5 Digest (first dword)
1472 * R = Rotation of secret
1473 * P = Odd or Even secret
1475 * The MD5 Digest is computed with over following parameters:
1476 * a) randomly rotated secret
1477 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1478 * c) the received initial sequence number from remote host
1479 * d) the rotation offset and odd/even bit
1481 * Timestamp we send:
1482 * 31|................................|0
1483 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1484 * D = MD5 Digest (third dword) (only as filler)
1485 * S = Requested send window scale
1486 * R = Requested receive window scale
1488 * 5 = TCP-MD5 enabled (not implemented yet)
1489 * XORed with MD5 Digest (forth dword)
1491 * The timestamp isn't cryptographically secure and doesn't need to be.
1492 * The double use of the MD5 digest dwords ties it to a specific remote/
1493 * local host/port, remote initial sequence number and our local time
1494 * limited secret. A received timestamp is reverted (XORed) and then
1495 * the contained MD5 dword is compared to the computed one to ensure the
1496 * timestamp belongs to the SYN-ACK we sent. The other parameters may
1497 * have been tampered with but this isn't different from supplying bogus
1498 * values in the SYN in the first place.
1500 * Some problems with SYN cookies remain however:
1501 * Consider the problem of a recreated (and retransmitted) cookie. If the
1502 * original SYN was accepted, the connection is established. The second
1503 * SYN is inflight, and if it arrives with an ISN that falls within the
1504 * receive window, the connection is killed.
1507 * A heuristic to determine when to accept syn cookies is not necessary.
1508 * An ACK flood would cause the syncookie verification to be attempted,
1509 * but a SYN flood causes syncookies to be generated. Both are of equal
1510 * cost, so there's no point in trying to optimize the ACK flood case.
1511 * Also, if you don't process certain ACKs for some reason, then all someone
1512 * would have to do is launch a SYN and ACK flood at the same time, which
1513 * would stop cookie verification and defeat the entire purpose of syncookies.
1515 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1518 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1519 u_int32_t *flowlabel)
1521 INIT_VNET_INET(curvnet);
1523 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1526 u_int off, pmss, mss;
1529 SCH_LOCK_ASSERT(sch);
1531 /* Which of the two secrets to use. */
1532 secbits = sch->sch_oddeven ?
1533 sch->sch_secbits_odd : sch->sch_secbits_even;
1535 /* Reseed secret if too old. */
1536 if (sch->sch_reseed < time_uptime) {
1537 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */
1538 secbits = sch->sch_oddeven ?
1539 sch->sch_secbits_odd : sch->sch_secbits_even;
1540 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1541 secbits[i] = arc4random();
1542 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1545 /* Secret rotation offset. */
1546 off = sc->sc_iss & 0x7; /* iss was randomized before */
1548 /* Maximum segment size calculation. */
1550 max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), V_tcp_minmss);
1551 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1552 if (tcp_sc_msstab[mss] <= pmss)
1555 /* Fold parameters and MD5 digest into the ISN we will send. */
1556 data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1557 data |= off << 1; /* secret offset, derived from iss, 3 bits */
1558 data |= mss << 4; /* mss, 3 bits */
1561 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1562 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1563 MD5Update(&ctx, secbits, off);
1564 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1565 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1566 MD5Update(&ctx, &data, sizeof(data));
1567 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1569 data |= (md5_buffer[0] << 7);
1573 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1576 /* Additional parameters are stored in the timestamp if present. */
1577 if (sc->sc_flags & SCF_TIMESTAMP) {
1578 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1579 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1580 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */
1581 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */
1582 data |= md5_buffer[2] << 10; /* more digest bits */
1583 data ^= md5_buffer[3];
1585 sc->sc_tsoff = data - ticks; /* after XOR */
1588 TCPSTAT_INC(tcps_sc_sendcookie);
1591 static struct syncache *
1592 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1593 struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1596 INIT_VNET_INET(curvnet);
1598 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1602 int off, mss, wnd, flags;
1604 SCH_LOCK_ASSERT(sch);
1607 * Pull information out of SYN-ACK/ACK and
1608 * revert sequence number advances.
1610 ack = th->th_ack - 1;
1611 seq = th->th_seq - 1;
1612 off = (ack >> 1) & 0x7;
1613 mss = (ack >> 4) & 0x7;
1616 /* Which of the two secrets to use. */
1617 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1620 * The secret wasn't updated for the lifetime of a syncookie,
1621 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1623 if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) {
1627 /* Recompute the digest so we can compare it. */
1629 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1630 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1631 MD5Update(&ctx, secbits, off);
1632 MD5Update(&ctx, inc, sizeof(*inc));
1633 MD5Update(&ctx, &seq, sizeof(seq));
1634 MD5Update(&ctx, &flags, sizeof(flags));
1635 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1637 /* Does the digest part of or ACK'ed ISS match? */
1638 if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1641 /* Does the digest part of our reflected timestamp match? */
1642 if (to->to_flags & TOF_TS) {
1643 data = md5_buffer[3] ^ to->to_tsecr;
1644 if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1648 /* Fill in the syncache values. */
1649 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1650 sc->sc_ipopts = NULL;
1656 if (inc->inc_flags & INC_ISIPV6) {
1657 if (sotoinpcb(so)->inp_flags & IN6P_AUTOFLOWLABEL)
1658 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1662 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1663 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1666 /* Additional parameters that were encoded in the timestamp. */
1668 sc->sc_flags |= SCF_TIMESTAMP;
1669 sc->sc_tsreflect = to->to_tsval;
1670 sc->sc_ts = to->to_tsecr;
1671 sc->sc_tsoff = to->to_tsecr - ticks;
1672 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1673 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1674 sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1676 sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1678 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1679 sc->sc_flags |= SCF_WINSCALE;
1681 sc->sc_flags |= SCF_NOOPT;
1683 wnd = sbspace(&so->so_rcv);
1685 wnd = imin(wnd, TCP_MAXWIN);
1689 sc->sc_peer_mss = tcp_sc_msstab[mss];
1691 TCPSTAT_INC(tcps_sc_recvcookie);
1696 * Returns the current number of syncache entries. This number
1697 * will probably change before you get around to calling
1702 syncache_pcbcount(void)
1704 INIT_VNET_INET(curvnet);
1705 struct syncache_head *sch;
1708 for (count = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1709 /* No need to lock for a read. */
1710 sch = &V_tcp_syncache.hashbase[i];
1711 count += sch->sch_length;
1717 * Exports the syncache entries to userland so that netstat can display
1718 * them alongside the other sockets. This function is intended to be
1719 * called only from tcp_pcblist.
1721 * Due to concurrency on an active system, the number of pcbs exported
1722 * may have no relation to max_pcbs. max_pcbs merely indicates the
1723 * amount of space the caller allocated for this function to use.
1726 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
1728 INIT_VNET_INET(curvnet);
1730 struct syncache *sc;
1731 struct syncache_head *sch;
1732 int count, error, i;
1734 for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1735 sch = &V_tcp_syncache.hashbase[i];
1737 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
1738 if (count >= max_pcbs) {
1742 if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0)
1744 bzero(&xt, sizeof(xt));
1745 xt.xt_len = sizeof(xt);
1746 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1747 xt.xt_inp.inp_vflag = INP_IPV6;
1749 xt.xt_inp.inp_vflag = INP_IPV4;
1750 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
1751 xt.xt_tp.t_inpcb = &xt.xt_inp;
1752 xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
1753 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1754 xt.xt_socket.xso_len = sizeof (struct xsocket);
1755 xt.xt_socket.so_type = SOCK_STREAM;
1756 xt.xt_socket.so_state = SS_ISCONNECTING;
1757 error = SYSCTL_OUT(req, &xt, sizeof xt);
1767 *pcbs_exported = count;