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++) {
262 TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket);
263 mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
265 callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer,
266 &V_tcp_syncache.hashbase[i].sch_mtx, 0);
267 V_tcp_syncache.hashbase[i].sch_length = 0;
270 /* Create the syncache entry zone. */
271 V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
272 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
273 uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit);
277 * Inserts a syncache entry into the specified bucket row.
278 * Locks and unlocks the syncache_head autonomously.
281 syncache_insert(struct syncache *sc, struct syncache_head *sch)
283 INIT_VNET_INET(sch->sch_vnet);
284 struct syncache *sc2;
289 * Make sure that we don't overflow the per-bucket limit.
290 * If the bucket is full, toss the oldest element.
292 if (sch->sch_length >= V_tcp_syncache.bucket_limit) {
293 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
294 ("sch->sch_length incorrect"));
295 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
296 syncache_drop(sc2, sch);
297 V_tcpstat.tcps_sc_bucketoverflow++;
300 /* Put it into the bucket. */
301 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
304 /* Reinitialize the bucket row's timer. */
305 if (sch->sch_length == 1)
306 sch->sch_nextc = ticks + INT_MAX;
307 syncache_timeout(sc, sch, 1);
311 V_tcp_syncache.cache_count++;
312 V_tcpstat.tcps_sc_added++;
316 * Remove and free entry from syncache bucket row.
317 * Expects locked syncache head.
320 syncache_drop(struct syncache *sc, struct syncache_head *sch)
322 INIT_VNET_INET(sch->sch_vnet);
324 SCH_LOCK_ASSERT(sch);
326 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
329 #ifndef TCP_OFFLOAD_DISABLE
331 sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb);
334 V_tcp_syncache.cache_count--;
338 * Engage/reengage time on bucket row.
341 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
343 sc->sc_rxttime = ticks +
344 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
346 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
347 sch->sch_nextc = sc->sc_rxttime;
349 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
350 syncache_timer, (void *)sch);
355 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
356 * If we have retransmitted an entry the maximum number of times, expire it.
357 * One separate timer for each bucket row.
360 syncache_timer(void *xsch)
362 struct syncache_head *sch = (struct syncache_head *)xsch;
363 struct syncache *sc, *nsc;
367 CURVNET_SET(sch->sch_vnet);
368 INIT_VNET_INET(sch->sch_vnet);
370 /* NB: syncache_head has already been locked by the callout. */
371 SCH_LOCK_ASSERT(sch);
374 * In the following cycle we may remove some entries and/or
375 * advance some timeouts, so re-initialize the bucket timer.
377 sch->sch_nextc = tick + INT_MAX;
379 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
381 * We do not check if the listen socket still exists
382 * and accept the case where the listen socket may be
383 * gone by the time we resend the SYN/ACK. We do
384 * not expect this to happens often. If it does,
385 * then the RST will be sent by the time the remote
386 * host does the SYN/ACK->ACK.
388 if (TSTMP_GT(sc->sc_rxttime, tick)) {
389 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
390 sch->sch_nextc = sc->sc_rxttime;
393 if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) {
394 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
395 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
396 "giving up and removing syncache entry\n",
400 syncache_drop(sc, sch);
401 V_tcpstat.tcps_sc_stale++;
404 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
405 log(LOG_DEBUG, "%s; %s: Response timeout, "
406 "retransmitting (%u) SYN|ACK\n",
407 s, __func__, sc->sc_rxmits);
411 (void) syncache_respond(sc);
412 V_tcpstat.tcps_sc_retransmitted++;
413 syncache_timeout(sc, sch, 0);
415 if (!TAILQ_EMPTY(&(sch)->sch_bucket))
416 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
417 syncache_timer, (void *)(sch));
422 * Find an entry in the syncache.
423 * Returns always with locked syncache_head plus a matching entry or NULL.
426 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
428 INIT_VNET_INET(curvnet);
430 struct syncache_head *sch;
433 if (inc->inc_flags & INC_ISIPV6) {
434 sch = &V_tcp_syncache.hashbase[
435 SYNCACHE_HASH6(inc, V_tcp_syncache.hashmask)];
440 /* Circle through bucket row to find matching entry. */
441 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
442 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
448 sch = &V_tcp_syncache.hashbase[
449 SYNCACHE_HASH(inc, V_tcp_syncache.hashmask)];
454 /* Circle through bucket row to find matching entry. */
455 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
457 if (sc->sc_inc.inc_flags & INC_ISIPV6)
460 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
464 SCH_LOCK_ASSERT(*schp);
465 return (NULL); /* always returns with locked sch */
469 * This function is called when we get a RST for a
470 * non-existent connection, so that we can see if the
471 * connection is in the syn cache. If it is, zap it.
474 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
476 INIT_VNET_INET(curvnet);
478 struct syncache_head *sch;
481 sc = syncache_lookup(inc, &sch); /* returns locked sch */
482 SCH_LOCK_ASSERT(sch);
485 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
486 * See RFC 793 page 65, section SEGMENT ARRIVES.
488 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
489 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
490 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
491 "FIN flag set, segment ignored\n", s, __func__);
492 V_tcpstat.tcps_badrst++;
497 * No corresponding connection was found in syncache.
498 * If syncookies are enabled and possibly exclusively
499 * used, or we are under memory pressure, a valid RST
500 * may not find a syncache entry. In that case we're
501 * done and no SYN|ACK retransmissions will happen.
502 * Otherwise the the RST was misdirected or spoofed.
505 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
506 log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
507 "syncache entry (possibly syncookie only), "
508 "segment ignored\n", s, __func__);
509 V_tcpstat.tcps_badrst++;
514 * If the RST bit is set, check the sequence number to see
515 * if this is a valid reset segment.
517 * In all states except SYN-SENT, all reset (RST) segments
518 * are validated by checking their SEQ-fields. A reset is
519 * valid if its sequence number is in the window.
521 * The sequence number in the reset segment is normally an
522 * echo of our outgoing acknowlegement numbers, but some hosts
523 * send a reset with the sequence number at the rightmost edge
524 * of our receive window, and we have to handle this case.
526 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
527 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
528 syncache_drop(sc, sch);
529 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
530 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
531 "connection attempt aborted by remote endpoint\n",
533 V_tcpstat.tcps_sc_reset++;
535 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
536 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != "
537 "IRS %u (+WND %u), segment ignored\n",
538 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
539 V_tcpstat.tcps_badrst++;
549 syncache_badack(struct in_conninfo *inc)
551 INIT_VNET_INET(curvnet);
553 struct syncache_head *sch;
555 sc = syncache_lookup(inc, &sch); /* returns locked sch */
556 SCH_LOCK_ASSERT(sch);
558 syncache_drop(sc, sch);
559 V_tcpstat.tcps_sc_badack++;
565 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
567 INIT_VNET_INET(curvnet);
569 struct syncache_head *sch;
571 sc = syncache_lookup(inc, &sch); /* returns locked sch */
572 SCH_LOCK_ASSERT(sch);
576 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
577 if (ntohl(th->th_seq) != sc->sc_iss)
581 * If we've rertransmitted 3 times and this is our second error,
582 * we remove the entry. Otherwise, we allow it to continue on.
583 * This prevents us from incorrectly nuking an entry during a
584 * spurious network outage.
588 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
589 sc->sc_flags |= SCF_UNREACH;
592 syncache_drop(sc, sch);
593 V_tcpstat.tcps_sc_unreach++;
599 * Build a new TCP socket structure from a syncache entry.
601 static struct socket *
602 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
604 INIT_VNET_INET(lso->so_vnet);
605 struct inpcb *inp = NULL;
610 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
613 * Ok, create the full blown connection, and set things up
614 * as they would have been set up if we had created the
615 * connection when the SYN arrived. If we can't create
616 * the connection, abort it.
618 so = sonewconn(lso, SS_ISCONNECTED);
621 * Drop the connection; we will either send a RST or
622 * have the peer retransmit its SYN again after its
625 V_tcpstat.tcps_listendrop++;
626 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
627 log(LOG_DEBUG, "%s; %s: Socket create failed "
628 "due to limits or memory shortage\n",
636 mac_socketpeer_set_from_mbuf(m, so);
641 inp->inp_inc.inc_fibnum = sc->sc_inc.inc_fibnum;
642 so->so_fibnum = sc->sc_inc.inc_fibnum;
645 /* Insert new socket into PCB hash list. */
646 inp->inp_inc.inc_flags = sc->sc_inc.inc_flags;
648 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
649 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
651 inp->inp_vflag &= ~INP_IPV6;
652 inp->inp_vflag |= INP_IPV4;
654 inp->inp_laddr = sc->sc_inc.inc_laddr;
658 inp->inp_lport = sc->sc_inc.inc_lport;
659 if (in_pcbinshash(inp) != 0) {
661 * Undo the assignments above if we failed to
662 * put the PCB on the hash lists.
665 if (sc->sc_inc.inc_flags & INC_ISIPV6)
666 inp->in6p_laddr = in6addr_any;
669 inp->inp_laddr.s_addr = INADDR_ANY;
674 /* Copy old policy into new socket's. */
675 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
676 printf("syncache_socket: could not copy policy\n");
679 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
680 struct inpcb *oinp = sotoinpcb(lso);
681 struct in6_addr laddr6;
682 struct sockaddr_in6 sin6;
684 * Inherit socket options from the listening socket.
685 * Note that in6p_inputopts are not (and should not be)
686 * copied, since it stores previously received options and is
687 * used to detect if each new option is different than the
688 * previous one and hence should be passed to a user.
689 * If we copied in6p_inputopts, a user would not be able to
690 * receive options just after calling the accept system call.
692 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
693 if (oinp->in6p_outputopts)
694 inp->in6p_outputopts =
695 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
697 sin6.sin6_family = AF_INET6;
698 sin6.sin6_len = sizeof(sin6);
699 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
700 sin6.sin6_port = sc->sc_inc.inc_fport;
701 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
702 laddr6 = inp->in6p_laddr;
703 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
704 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
705 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
707 inp->in6p_laddr = laddr6;
710 /* Override flowlabel from in6_pcbconnect. */
711 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
712 inp->inp_flow |= sc->sc_flowlabel;
716 struct in_addr laddr;
717 struct sockaddr_in sin;
719 inp->inp_options = (m) ? ip_srcroute(m) : NULL;
721 if (inp->inp_options == NULL) {
722 inp->inp_options = sc->sc_ipopts;
723 sc->sc_ipopts = NULL;
726 sin.sin_family = AF_INET;
727 sin.sin_len = sizeof(sin);
728 sin.sin_addr = sc->sc_inc.inc_faddr;
729 sin.sin_port = sc->sc_inc.inc_fport;
730 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
731 laddr = inp->inp_laddr;
732 if (inp->inp_laddr.s_addr == INADDR_ANY)
733 inp->inp_laddr = sc->sc_inc.inc_laddr;
734 if (in_pcbconnect(inp, (struct sockaddr *)&sin,
736 inp->inp_laddr = laddr;
741 tp->t_state = TCPS_SYN_RECEIVED;
742 tp->iss = sc->sc_iss;
743 tp->irs = sc->sc_irs;
746 tp->snd_wl1 = sc->sc_irs;
747 tp->snd_max = tp->iss + 1;
748 tp->snd_nxt = tp->iss + 1;
749 tp->rcv_up = sc->sc_irs + 1;
750 tp->rcv_wnd = sc->sc_wnd;
751 tp->rcv_adv += tp->rcv_wnd;
752 tp->last_ack_sent = tp->rcv_nxt;
754 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
755 if (sc->sc_flags & SCF_NOOPT)
756 tp->t_flags |= TF_NOOPT;
758 if (sc->sc_flags & SCF_WINSCALE) {
759 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
760 tp->snd_scale = sc->sc_requested_s_scale;
761 tp->request_r_scale = sc->sc_requested_r_scale;
763 if (sc->sc_flags & SCF_TIMESTAMP) {
764 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
765 tp->ts_recent = sc->sc_tsreflect;
766 tp->ts_recent_age = ticks;
767 tp->ts_offset = sc->sc_tsoff;
770 if (sc->sc_flags & SCF_SIGNATURE)
771 tp->t_flags |= TF_SIGNATURE;
773 if (sc->sc_flags & SCF_SACK)
774 tp->t_flags |= TF_SACK_PERMIT;
777 if (sc->sc_flags & SCF_ECN)
778 tp->t_flags |= TF_ECN_PERMIT;
781 * Set up MSS and get cached values from tcp_hostcache.
782 * This might overwrite some of the defaults we just set.
784 tcp_mss(tp, sc->sc_peer_mss);
787 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
790 tp->snd_cwnd = tp->t_maxseg;
791 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
795 V_tcpstat.tcps_accepts++;
807 * This function gets called when we receive an ACK for a
808 * socket in the LISTEN state. We look up the connection
809 * in the syncache, and if its there, we pull it out of
810 * the cache and turn it into a full-blown connection in
811 * the SYN-RECEIVED state.
814 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
815 struct socket **lsop, struct mbuf *m)
817 INIT_VNET_INET(curvnet);
819 struct syncache_head *sch;
824 * Global TCP locks are held because we manipulate the PCB lists
825 * and create a new socket.
827 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
828 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
829 ("%s: can handle only ACK", __func__));
831 sc = syncache_lookup(inc, &sch); /* returns locked sch */
832 SCH_LOCK_ASSERT(sch);
835 * There is no syncache entry, so see if this ACK is
836 * a returning syncookie. To do this, first:
837 * A. See if this socket has had a syncache entry dropped in
838 * the past. We don't want to accept a bogus syncookie
839 * if we've never received a SYN.
840 * B. check that the syncookie is valid. If it is, then
841 * cobble up a fake syncache entry, and return.
843 if (!V_tcp_syncookies) {
845 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
846 log(LOG_DEBUG, "%s; %s: Spurious ACK, "
847 "segment rejected (syncookies disabled)\n",
851 bzero(&scs, sizeof(scs));
852 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
855 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
856 log(LOG_DEBUG, "%s; %s: Segment failed "
857 "SYNCOOKIE authentication, segment rejected "
858 "(probably spoofed)\n", s, __func__);
862 /* Pull out the entry to unlock the bucket row. */
863 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
865 V_tcp_syncache.cache_count--;
870 * Segment validation:
871 * ACK must match our initial sequence number + 1 (the SYN|ACK).
873 if (th->th_ack != sc->sc_iss + 1 && !TOEPCB_ISSET(sc)) {
874 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
875 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment "
876 "rejected\n", s, __func__, th->th_ack, sc->sc_iss);
881 * The SEQ must fall in the window starting at the received
882 * initial receive sequence number + 1 (the SYN).
884 if ((SEQ_LEQ(th->th_seq, sc->sc_irs) ||
885 SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) &&
887 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
888 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment "
889 "rejected\n", s, __func__, th->th_seq, sc->sc_irs);
893 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
894 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
895 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
896 "segment rejected\n", s, __func__);
900 * If timestamps were negotiated the reflected timestamp
901 * must be equal to what we actually sent in the SYN|ACK.
903 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts &&
905 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
906 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
907 "segment rejected\n",
908 s, __func__, to->to_tsecr, sc->sc_ts);
912 *lsop = syncache_socket(sc, *lsop, m);
915 V_tcpstat.tcps_sc_aborted++;
917 V_tcpstat.tcps_sc_completed++;
919 /* how do we find the inp for the new socket? */
924 if (sc != NULL && sc != &scs)
933 tcp_offload_syncache_expand(struct in_conninfo *inc, struct tcpopt *to,
934 struct tcphdr *th, struct socket **lsop, struct mbuf *m)
936 INIT_VNET_INET(curvnet);
939 INP_INFO_WLOCK(&V_tcbinfo);
940 rc = syncache_expand(inc, to, th, lsop, m);
941 INP_INFO_WUNLOCK(&V_tcbinfo);
947 * Given a LISTEN socket and an inbound SYN request, add
948 * this to the syn cache, and send back a segment:
949 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
952 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
953 * Doing so would require that we hold onto the data and deliver it
954 * to the application. However, if we are the target of a SYN-flood
955 * DoS attack, an attacker could send data which would eventually
956 * consume all available buffer space if it were ACKed. By not ACKing
957 * the data, we avoid this DoS scenario.
960 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
961 struct inpcb *inp, struct socket **lsop, struct mbuf *m,
962 struct toe_usrreqs *tu, void *toepcb)
964 INIT_VNET_INET(inp->inp_vnet);
967 struct syncache *sc = NULL;
968 struct syncache_head *sch;
969 struct mbuf *ipopts = NULL;
971 int win, sb_hiwat, ip_ttl, ip_tos, noopt;
974 int autoflowlabel = 0;
977 struct label *maclabel;
982 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
983 INP_WLOCK_ASSERT(inp); /* listen socket */
984 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
985 ("%s: unexpected tcp flags", __func__));
988 * Combine all so/tp operations very early to drop the INP lock as
993 cred = crhold(so->so_cred);
996 if ((inc->inc_flags & INC_ISIPV6) &&
997 (inp->inp_flags & IN6P_AUTOFLOWLABEL))
1000 ip_ttl = inp->inp_ip_ttl;
1001 ip_tos = inp->inp_ip_tos;
1002 win = sbspace(&so->so_rcv);
1003 sb_hiwat = so->so_rcv.sb_hiwat;
1004 noopt = (tp->t_flags & TF_NOOPT);
1006 /* By the time we drop the lock these should no longer be used. */
1011 if (mac_syncache_init(&maclabel) != 0) {
1013 INP_INFO_WUNLOCK(&V_tcbinfo);
1016 mac_syncache_create(maclabel, inp);
1019 INP_INFO_WUNLOCK(&V_tcbinfo);
1022 * Remember the IP options, if any.
1025 if (!(inc->inc_flags & INC_ISIPV6))
1027 ipopts = (m) ? ip_srcroute(m) : NULL;
1030 * See if we already have an entry for this connection.
1031 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1033 * XXX: should the syncache be re-initialized with the contents
1034 * of the new SYN here (which may have different options?)
1036 * XXX: We do not check the sequence number to see if this is a
1037 * real retransmit or a new connection attempt. The question is
1038 * how to handle such a case; either ignore it as spoofed, or
1039 * drop the current entry and create a new one?
1041 sc = syncache_lookup(inc, &sch); /* returns locked entry */
1042 SCH_LOCK_ASSERT(sch);
1044 #ifndef TCP_OFFLOAD_DISABLE
1046 sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT,
1049 V_tcpstat.tcps_sc_dupsyn++;
1052 * If we were remembering a previous source route,
1053 * forget it and use the new one we've been given.
1056 (void) m_free(sc->sc_ipopts);
1057 sc->sc_ipopts = ipopts;
1060 * Update timestamp if present.
1062 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
1063 sc->sc_tsreflect = to->to_tsval;
1065 sc->sc_flags &= ~SCF_TIMESTAMP;
1068 * Since we have already unconditionally allocated label
1069 * storage, free it up. The syncache entry will already
1070 * have an initialized label we can use.
1072 mac_syncache_destroy(&maclabel);
1074 /* Retransmit SYN|ACK and reset retransmit count. */
1075 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
1076 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
1077 "resetting timer and retransmitting SYN|ACK\n",
1081 if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) {
1083 syncache_timeout(sc, sch, 1);
1084 V_tcpstat.tcps_sndacks++;
1085 V_tcpstat.tcps_sndtotal++;
1091 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1094 * The zone allocator couldn't provide more entries.
1095 * Treat this as if the cache was full; drop the oldest
1096 * entry and insert the new one.
1098 V_tcpstat.tcps_sc_zonefail++;
1099 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1100 syncache_drop(sc, sch);
1101 sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO);
1103 if (V_tcp_syncookies) {
1104 bzero(&scs, sizeof(scs));
1109 (void) m_free(ipopts);
1116 * Fill in the syncache values.
1119 sc->sc_label = maclabel;
1123 sc->sc_ipopts = ipopts;
1124 /* XXX-BZ this fib assignment is just useless. */
1125 sc->sc_inc.inc_fibnum = inp->inp_inc.inc_fibnum;
1126 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1128 if (!(inc->inc_flags & INC_ISIPV6))
1131 sc->sc_ip_tos = ip_tos;
1132 sc->sc_ip_ttl = ip_ttl;
1134 #ifndef TCP_OFFLOAD_DISABLE
1136 sc->sc_toepcb = toepcb;
1138 sc->sc_irs = th->th_seq;
1139 sc->sc_iss = arc4random();
1141 sc->sc_flowlabel = 0;
1144 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
1145 * win was derived from socket earlier in the function.
1148 win = imin(win, TCP_MAXWIN);
1151 if (V_tcp_do_rfc1323) {
1153 * A timestamp received in a SYN makes
1154 * it ok to send timestamp requests and replies.
1156 if (to->to_flags & TOF_TS) {
1157 sc->sc_tsreflect = to->to_tsval;
1159 sc->sc_flags |= SCF_TIMESTAMP;
1161 if (to->to_flags & TOF_SCALE) {
1165 * Pick the smallest possible scaling factor that
1166 * will still allow us to scale up to sb_max, aka
1167 * kern.ipc.maxsockbuf.
1169 * We do this because there are broken firewalls that
1170 * will corrupt the window scale option, leading to
1171 * the other endpoint believing that our advertised
1172 * window is unscaled. At scale factors larger than
1173 * 5 the unscaled window will drop below 1500 bytes,
1174 * leading to serious problems when traversing these
1177 * With the default maxsockbuf of 256K, a scale factor
1178 * of 3 will be chosen by this algorithm. Those who
1179 * choose a larger maxsockbuf should watch out
1180 * for the compatiblity problems mentioned above.
1182 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1183 * or <SYN,ACK>) segment itself is never scaled.
1185 while (wscale < TCP_MAX_WINSHIFT &&
1186 (TCP_MAXWIN << wscale) < sb_max)
1188 sc->sc_requested_r_scale = wscale;
1189 sc->sc_requested_s_scale = to->to_wscale;
1190 sc->sc_flags |= SCF_WINSCALE;
1193 #ifdef TCP_SIGNATURE
1195 * If listening socket requested TCP digests, and received SYN
1196 * contains the option, flag this in the syncache so that
1197 * syncache_respond() will do the right thing with the SYN+ACK.
1198 * XXX: Currently we always record the option by default and will
1199 * attempt to use it in syncache_respond().
1201 if (to->to_flags & TOF_SIGNATURE)
1202 sc->sc_flags |= SCF_SIGNATURE;
1204 if (to->to_flags & TOF_SACKPERM)
1205 sc->sc_flags |= SCF_SACK;
1206 if (to->to_flags & TOF_MSS)
1207 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */
1209 sc->sc_flags |= SCF_NOOPT;
1210 if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn)
1211 sc->sc_flags |= SCF_ECN;
1213 if (V_tcp_syncookies) {
1214 syncookie_generate(sch, sc, &flowtmp);
1217 sc->sc_flowlabel = flowtmp;
1223 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1229 * Do a standard 3-way handshake.
1231 if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) {
1232 if (V_tcp_syncookies && V_tcp_syncookiesonly && sc != &scs)
1234 else if (sc != &scs)
1235 syncache_insert(sc, sch); /* locks and unlocks sch */
1236 V_tcpstat.tcps_sndacks++;
1237 V_tcpstat.tcps_sndtotal++;
1241 V_tcpstat.tcps_sc_dropped++;
1249 mac_syncache_destroy(&maclabel);
1259 syncache_respond(struct syncache *sc)
1261 INIT_VNET_INET(curvnet);
1262 struct ip *ip = NULL;
1266 u_int16_t hlen, tlen, mssopt;
1269 struct ip6_hdr *ip6 = NULL;
1274 (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) :
1277 tlen = hlen + sizeof(struct tcphdr);
1279 /* Determine MSS we advertize to other end of connection. */
1280 mssopt = tcp_mssopt(&sc->sc_inc);
1281 if (sc->sc_peer_mss)
1282 mssopt = max( min(sc->sc_peer_mss, mssopt), V_tcp_minmss);
1284 /* XXX: Assume that the entire packet will fit in a header mbuf. */
1285 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1286 ("syncache: mbuf too small"));
1288 /* Create the IP+TCP header from scratch. */
1289 m = m_gethdr(M_DONTWAIT, MT_DATA);
1293 mac_syncache_create_mbuf(sc->sc_label, m);
1295 m->m_data += max_linkhdr;
1297 m->m_pkthdr.len = tlen;
1298 m->m_pkthdr.rcvif = NULL;
1301 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1302 ip6 = mtod(m, struct ip6_hdr *);
1303 ip6->ip6_vfc = IPV6_VERSION;
1304 ip6->ip6_nxt = IPPROTO_TCP;
1305 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1306 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1307 ip6->ip6_plen = htons(tlen - hlen);
1308 /* ip6_hlim is set after checksum */
1309 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1310 ip6->ip6_flow |= sc->sc_flowlabel;
1312 th = (struct tcphdr *)(ip6 + 1);
1316 ip = mtod(m, struct ip *);
1317 ip->ip_v = IPVERSION;
1318 ip->ip_hl = sizeof(struct ip) >> 2;
1323 ip->ip_p = IPPROTO_TCP;
1324 ip->ip_src = sc->sc_inc.inc_laddr;
1325 ip->ip_dst = sc->sc_inc.inc_faddr;
1326 ip->ip_ttl = sc->sc_ip_ttl;
1327 ip->ip_tos = sc->sc_ip_tos;
1330 * See if we should do MTU discovery. Route lookups are
1331 * expensive, so we will only unset the DF bit if:
1333 * 1) path_mtu_discovery is disabled
1334 * 2) the SCF_UNREACH flag has been set
1336 if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1337 ip->ip_off |= IP_DF;
1339 th = (struct tcphdr *)(ip + 1);
1341 th->th_sport = sc->sc_inc.inc_lport;
1342 th->th_dport = sc->sc_inc.inc_fport;
1344 th->th_seq = htonl(sc->sc_iss);
1345 th->th_ack = htonl(sc->sc_irs + 1);
1346 th->th_off = sizeof(struct tcphdr) >> 2;
1348 th->th_flags = TH_SYN|TH_ACK;
1349 th->th_win = htons(sc->sc_wnd);
1352 if (sc->sc_flags & SCF_ECN) {
1353 th->th_flags |= TH_ECE;
1354 V_tcpstat.tcps_ecn_shs++;
1357 /* Tack on the TCP options. */
1358 if ((sc->sc_flags & SCF_NOOPT) == 0) {
1362 to.to_flags = TOF_MSS;
1363 if (sc->sc_flags & SCF_WINSCALE) {
1364 to.to_wscale = sc->sc_requested_r_scale;
1365 to.to_flags |= TOF_SCALE;
1367 if (sc->sc_flags & SCF_TIMESTAMP) {
1368 /* Virgin timestamp or TCP cookie enhanced one. */
1369 to.to_tsval = sc->sc_ts;
1370 to.to_tsecr = sc->sc_tsreflect;
1371 to.to_flags |= TOF_TS;
1373 if (sc->sc_flags & SCF_SACK)
1374 to.to_flags |= TOF_SACKPERM;
1375 #ifdef TCP_SIGNATURE
1376 if (sc->sc_flags & SCF_SIGNATURE)
1377 to.to_flags |= TOF_SIGNATURE;
1379 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1381 /* Adjust headers by option size. */
1382 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1384 m->m_pkthdr.len += optlen;
1386 #ifdef TCP_SIGNATURE
1387 if (sc->sc_flags & SCF_SIGNATURE)
1388 tcp_signature_compute(m, 0, 0, optlen,
1389 to.to_signature, IPSEC_DIR_OUTBOUND);
1392 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1393 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1396 ip->ip_len += optlen;
1401 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1403 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
1404 tlen + optlen - hlen);
1405 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1406 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1410 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1411 htons(tlen + optlen - hlen + IPPROTO_TCP));
1412 m->m_pkthdr.csum_flags = CSUM_TCP;
1413 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1414 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1420 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1421 struct inpcb *inp, struct socket **lsop, struct mbuf *m)
1423 _syncache_add(inc, to, th, inp, lsop, m, NULL, NULL);
1427 tcp_offload_syncache_add(struct in_conninfo *inc, struct tcpopt *to,
1428 struct tcphdr *th, struct inpcb *inp, struct socket **lsop,
1429 struct toe_usrreqs *tu, void *toepcb)
1431 INIT_VNET_INET(curvnet);
1433 INP_INFO_WLOCK(&V_tcbinfo);
1435 _syncache_add(inc, to, th, inp, lsop, NULL, tu, toepcb);
1439 * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1440 * receive and to be able to handle SYN floods from bogus source addresses
1441 * (where we will never receive any reply). SYN floods try to exhaust all
1442 * our memory and available slots in the SYN cache table to cause a denial
1443 * of service to legitimate users of the local host.
1445 * The idea of SYN cookies is to encode and include all necessary information
1446 * about the connection setup state within the SYN-ACK we send back and thus
1447 * to get along without keeping any local state until the ACK to the SYN-ACK
1448 * arrives (if ever). Everything we need to know should be available from
1449 * the information we encoded in the SYN-ACK.
1451 * More information about the theory behind SYN cookies and its first
1452 * discussion and specification can be found at:
1453 * http://cr.yp.to/syncookies.html (overview)
1454 * http://cr.yp.to/syncookies/archive (gory details)
1456 * This implementation extends the orginal idea and first implementation
1457 * of FreeBSD by using not only the initial sequence number field to store
1458 * information but also the timestamp field if present. This way we can
1459 * keep track of the entire state we need to know to recreate the session in
1460 * its original form. Almost all TCP speakers implement RFC1323 timestamps
1461 * these days. For those that do not we still have to live with the known
1462 * shortcomings of the ISN only SYN cookies.
1466 * Initial sequence number we send:
1467 * 31|................................|0
1468 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1469 * D = MD5 Digest (first dword)
1471 * R = Rotation of secret
1472 * P = Odd or Even secret
1474 * The MD5 Digest is computed with over following parameters:
1475 * a) randomly rotated secret
1476 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1477 * c) the received initial sequence number from remote host
1478 * d) the rotation offset and odd/even bit
1480 * Timestamp we send:
1481 * 31|................................|0
1482 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1483 * D = MD5 Digest (third dword) (only as filler)
1484 * S = Requested send window scale
1485 * R = Requested receive window scale
1487 * 5 = TCP-MD5 enabled (not implemented yet)
1488 * XORed with MD5 Digest (forth dword)
1490 * The timestamp isn't cryptographically secure and doesn't need to be.
1491 * The double use of the MD5 digest dwords ties it to a specific remote/
1492 * local host/port, remote initial sequence number and our local time
1493 * limited secret. A received timestamp is reverted (XORed) and then
1494 * the contained MD5 dword is compared to the computed one to ensure the
1495 * timestamp belongs to the SYN-ACK we sent. The other parameters may
1496 * have been tampered with but this isn't different from supplying bogus
1497 * values in the SYN in the first place.
1499 * Some problems with SYN cookies remain however:
1500 * Consider the problem of a recreated (and retransmitted) cookie. If the
1501 * original SYN was accepted, the connection is established. The second
1502 * SYN is inflight, and if it arrives with an ISN that falls within the
1503 * receive window, the connection is killed.
1506 * A heuristic to determine when to accept syn cookies is not necessary.
1507 * An ACK flood would cause the syncookie verification to be attempted,
1508 * but a SYN flood causes syncookies to be generated. Both are of equal
1509 * cost, so there's no point in trying to optimize the ACK flood case.
1510 * Also, if you don't process certain ACKs for some reason, then all someone
1511 * would have to do is launch a SYN and ACK flood at the same time, which
1512 * would stop cookie verification and defeat the entire purpose of syncookies.
1514 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1517 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1518 u_int32_t *flowlabel)
1520 INIT_VNET_INET(curvnet);
1522 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1525 u_int off, pmss, mss;
1528 SCH_LOCK_ASSERT(sch);
1530 /* Which of the two secrets to use. */
1531 secbits = sch->sch_oddeven ?
1532 sch->sch_secbits_odd : sch->sch_secbits_even;
1534 /* Reseed secret if too old. */
1535 if (sch->sch_reseed < time_uptime) {
1536 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */
1537 secbits = sch->sch_oddeven ?
1538 sch->sch_secbits_odd : sch->sch_secbits_even;
1539 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1540 secbits[i] = arc4random();
1541 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1544 /* Secret rotation offset. */
1545 off = sc->sc_iss & 0x7; /* iss was randomized before */
1547 /* Maximum segment size calculation. */
1549 max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), V_tcp_minmss);
1550 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1551 if (tcp_sc_msstab[mss] <= pmss)
1554 /* Fold parameters and MD5 digest into the ISN we will send. */
1555 data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1556 data |= off << 1; /* secret offset, derived from iss, 3 bits */
1557 data |= mss << 4; /* mss, 3 bits */
1560 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1561 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1562 MD5Update(&ctx, secbits, off);
1563 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1564 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1565 MD5Update(&ctx, &data, sizeof(data));
1566 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1568 data |= (md5_buffer[0] << 7);
1572 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1575 /* Additional parameters are stored in the timestamp if present. */
1576 if (sc->sc_flags & SCF_TIMESTAMP) {
1577 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1578 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1579 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */
1580 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */
1581 data |= md5_buffer[2] << 10; /* more digest bits */
1582 data ^= md5_buffer[3];
1584 sc->sc_tsoff = data - ticks; /* after XOR */
1587 V_tcpstat.tcps_sc_sendcookie++;
1590 static struct syncache *
1591 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1592 struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1595 INIT_VNET_INET(curvnet);
1597 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1601 int off, mss, wnd, flags;
1603 SCH_LOCK_ASSERT(sch);
1606 * Pull information out of SYN-ACK/ACK and
1607 * revert sequence number advances.
1609 ack = th->th_ack - 1;
1610 seq = th->th_seq - 1;
1611 off = (ack >> 1) & 0x7;
1612 mss = (ack >> 4) & 0x7;
1615 /* Which of the two secrets to use. */
1616 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1619 * The secret wasn't updated for the lifetime of a syncookie,
1620 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1622 if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) {
1626 /* Recompute the digest so we can compare it. */
1628 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1629 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1630 MD5Update(&ctx, secbits, off);
1631 MD5Update(&ctx, inc, sizeof(*inc));
1632 MD5Update(&ctx, &seq, sizeof(seq));
1633 MD5Update(&ctx, &flags, sizeof(flags));
1634 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1636 /* Does the digest part of or ACK'ed ISS match? */
1637 if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1640 /* Does the digest part of our reflected timestamp match? */
1641 if (to->to_flags & TOF_TS) {
1642 data = md5_buffer[3] ^ to->to_tsecr;
1643 if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1647 /* Fill in the syncache values. */
1648 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1649 sc->sc_ipopts = NULL;
1655 if (inc->inc_flags & INC_ISIPV6) {
1656 if (sotoinpcb(so)->inp_flags & IN6P_AUTOFLOWLABEL)
1657 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1661 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1662 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1665 /* Additional parameters that were encoded in the timestamp. */
1667 sc->sc_flags |= SCF_TIMESTAMP;
1668 sc->sc_tsreflect = to->to_tsval;
1669 sc->sc_ts = to->to_tsecr;
1670 sc->sc_tsoff = to->to_tsecr - ticks;
1671 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1672 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1673 sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1675 sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1677 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1678 sc->sc_flags |= SCF_WINSCALE;
1680 sc->sc_flags |= SCF_NOOPT;
1682 wnd = sbspace(&so->so_rcv);
1684 wnd = imin(wnd, TCP_MAXWIN);
1688 sc->sc_peer_mss = tcp_sc_msstab[mss];
1690 V_tcpstat.tcps_sc_recvcookie++;
1695 * Returns the current number of syncache entries. This number
1696 * will probably change before you get around to calling
1701 syncache_pcbcount(void)
1703 INIT_VNET_INET(curvnet);
1704 struct syncache_head *sch;
1707 for (count = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1708 /* No need to lock for a read. */
1709 sch = &V_tcp_syncache.hashbase[i];
1710 count += sch->sch_length;
1716 * Exports the syncache entries to userland so that netstat can display
1717 * them alongside the other sockets. This function is intended to be
1718 * called only from tcp_pcblist.
1720 * Due to concurrency on an active system, the number of pcbs exported
1721 * may have no relation to max_pcbs. max_pcbs merely indicates the
1722 * amount of space the caller allocated for this function to use.
1725 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
1727 INIT_VNET_INET(curvnet);
1729 struct syncache *sc;
1730 struct syncache_head *sch;
1731 int count, error, i;
1733 for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) {
1734 sch = &V_tcp_syncache.hashbase[i];
1736 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
1737 if (count >= max_pcbs) {
1741 if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0)
1743 bzero(&xt, sizeof(xt));
1744 xt.xt_len = sizeof(xt);
1745 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1746 xt.xt_inp.inp_vflag = INP_IPV6;
1748 xt.xt_inp.inp_vflag = INP_IPV4;
1749 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
1750 xt.xt_tp.t_inpcb = &xt.xt_inp;
1751 xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
1752 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1753 xt.xt_socket.xso_len = sizeof (struct xsocket);
1754 xt.xt_socket.so_type = SOCK_STREAM;
1755 xt.xt_socket.so_state = SS_ISCONNECTING;
1756 error = SYSCTL_OUT(req, &xt, sizeof xt);
1766 *pcbs_exported = count;