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>
61 #include <net/route.h>
63 #include <netinet/in.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/ip.h>
66 #include <netinet/in_var.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/ip_var.h>
69 #include <netinet/ip_options.h>
71 #include <netinet/ip6.h>
72 #include <netinet/icmp6.h>
73 #include <netinet6/nd6.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/in6_pcb.h>
77 #include <netinet/tcp.h>
78 #include <netinet/tcp_fsm.h>
79 #include <netinet/tcp_seq.h>
80 #include <netinet/tcp_timer.h>
81 #include <netinet/tcp_var.h>
82 #include <netinet/tcp_syncache.h>
83 #include <netinet/tcp_offload.h>
85 #include <netinet6/tcp6_var.h>
89 #include <netipsec/ipsec.h>
91 #include <netipsec/ipsec6.h>
93 #include <netipsec/key.h>
96 #include <machine/in_cksum.h>
98 #include <security/mac/mac_framework.h>
100 static int tcp_syncookies = 1;
101 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
103 "Use TCP SYN cookies if the syncache overflows");
105 static int tcp_syncookiesonly = 0;
106 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_RW,
107 &tcp_syncookiesonly, 0,
108 "Use only TCP SYN cookies");
110 #define SYNCOOKIE_SECRET_SIZE 8 /* dwords */
111 #define SYNCOOKIE_LIFETIME 16 /* seconds */
114 TAILQ_ENTRY(syncache) sc_hash;
115 struct in_conninfo sc_inc; /* addresses */
116 int sc_rxttime; /* retransmit time */
117 u_int16_t sc_rxmits; /* retransmit counter */
118 u_int32_t sc_tsreflect; /* timestamp to reflect */
119 u_int32_t sc_ts; /* our timestamp to send */
120 u_int32_t sc_tsoff; /* ts offset w/ syncookies */
121 u_int32_t sc_flowlabel; /* IPv6 flowlabel */
122 tcp_seq sc_irs; /* seq from peer */
123 tcp_seq sc_iss; /* our ISS */
124 struct mbuf *sc_ipopts; /* source route */
125 u_int16_t sc_peer_mss; /* peer's MSS */
126 u_int16_t sc_wnd; /* advertised window */
127 u_int8_t sc_ip_ttl; /* IPv4 TTL */
128 u_int8_t sc_ip_tos; /* IPv4 TOS */
129 u_int8_t sc_requested_s_scale:4,
130 sc_requested_r_scale:4;
132 #ifndef TCP_OFFLOAD_DISABLE
133 struct toe_usrreqs *sc_tu; /* TOE operations */
134 void *sc_toepcb; /* TOE protocol block */
137 struct label *sc_label; /* MAC label reference */
139 struct ucred *sc_cred; /* cred cache for jail checks */
143 * Flags for the sc_flags field.
145 #define SCF_NOOPT 0x01 /* no TCP options */
146 #define SCF_WINSCALE 0x02 /* negotiated window scaling */
147 #define SCF_TIMESTAMP 0x04 /* negotiated timestamps */
148 /* MSS is implicit */
149 #define SCF_UNREACH 0x10 /* icmp unreachable received */
150 #define SCF_SIGNATURE 0x20 /* send MD5 digests */
151 #define SCF_SACK 0x80 /* send SACK option */
153 #ifdef TCP_OFFLOAD_DISABLE
154 #define TOEPCB_ISSET(sc) (0)
156 #define TOEPCB_ISSET(sc) ((sc)->sc_toepcb != NULL)
160 struct syncache_head {
162 TAILQ_HEAD(sch_head, syncache) sch_bucket;
163 struct callout sch_timer;
167 u_int32_t sch_secbits_odd[SYNCOOKIE_SECRET_SIZE];
168 u_int32_t sch_secbits_even[SYNCOOKIE_SECRET_SIZE];
169 u_int sch_reseed; /* time_uptime, seconds */
172 static void syncache_drop(struct syncache *, struct syncache_head *);
173 static void syncache_free(struct syncache *);
174 static void syncache_insert(struct syncache *, struct syncache_head *);
175 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
176 static int syncache_respond(struct syncache *);
177 static struct socket *syncache_socket(struct syncache *, struct socket *,
179 static void syncache_timeout(struct syncache *sc, struct syncache_head *sch,
181 static void syncache_timer(void *);
182 static void syncookie_generate(struct syncache_head *, struct syncache *,
184 static struct syncache
185 *syncookie_lookup(struct in_conninfo *, struct syncache_head *,
186 struct syncache *, struct tcpopt *, struct tcphdr *,
190 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
191 * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds,
192 * the odds are that the user has given up attempting to connect by then.
194 #define SYNCACHE_MAXREXMTS 3
196 /* Arbitrary values */
197 #define TCP_SYNCACHE_HASHSIZE 512
198 #define TCP_SYNCACHE_BUCKETLIMIT 30
200 struct tcp_syncache {
201 struct syncache_head *hashbase;
206 u_int cache_count; /* XXX: unprotected */
211 static struct tcp_syncache tcp_syncache;
213 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
215 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RDTUN,
216 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
218 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RDTUN,
219 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
221 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
222 &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
224 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RDTUN,
225 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
227 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
228 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
230 int tcp_sc_rst_sock_fail = 1;
231 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail, CTLFLAG_RW,
232 &tcp_sc_rst_sock_fail, 0, "Send reset on socket allocation failure");
234 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
236 #define SYNCACHE_HASH(inc, mask) \
237 ((tcp_syncache.hash_secret ^ \
238 (inc)->inc_faddr.s_addr ^ \
239 ((inc)->inc_faddr.s_addr >> 16) ^ \
240 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
242 #define SYNCACHE_HASH6(inc, mask) \
243 ((tcp_syncache.hash_secret ^ \
244 (inc)->inc6_faddr.s6_addr32[0] ^ \
245 (inc)->inc6_faddr.s6_addr32[3] ^ \
246 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
248 #define ENDPTS_EQ(a, b) ( \
249 (a)->ie_fport == (b)->ie_fport && \
250 (a)->ie_lport == (b)->ie_lport && \
251 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
252 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
255 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
257 #define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx)
258 #define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx)
259 #define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED)
262 * Requires the syncache entry to be already removed from the bucket list.
265 syncache_free(struct syncache *sc)
268 (void) m_free(sc->sc_ipopts);
272 mac_destroy_syncache(&sc->sc_label);
275 uma_zfree(tcp_syncache.zone, sc);
283 tcp_syncache.cache_count = 0;
284 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
285 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
286 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
287 tcp_syncache.hash_secret = arc4random();
289 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
290 &tcp_syncache.hashsize);
291 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
292 &tcp_syncache.bucket_limit);
293 if (!powerof2(tcp_syncache.hashsize) || tcp_syncache.hashsize == 0) {
294 printf("WARNING: syncache hash size is not a power of 2.\n");
295 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
297 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
300 tcp_syncache.cache_limit =
301 tcp_syncache.hashsize * tcp_syncache.bucket_limit;
302 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
303 &tcp_syncache.cache_limit);
305 /* Allocate the hash table. */
306 MALLOC(tcp_syncache.hashbase, struct syncache_head *,
307 tcp_syncache.hashsize * sizeof(struct syncache_head),
308 M_SYNCACHE, M_WAITOK | M_ZERO);
310 /* Initialize the hash buckets. */
311 for (i = 0; i < tcp_syncache.hashsize; i++) {
312 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket);
313 mtx_init(&tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head",
315 callout_init_mtx(&tcp_syncache.hashbase[i].sch_timer,
316 &tcp_syncache.hashbase[i].sch_mtx, 0);
317 tcp_syncache.hashbase[i].sch_length = 0;
320 /* Create the syncache entry zone. */
321 tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache),
322 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
323 uma_zone_set_max(tcp_syncache.zone, tcp_syncache.cache_limit);
327 * Inserts a syncache entry into the specified bucket row.
328 * Locks and unlocks the syncache_head autonomously.
331 syncache_insert(struct syncache *sc, struct syncache_head *sch)
333 struct syncache *sc2;
338 * Make sure that we don't overflow the per-bucket limit.
339 * If the bucket is full, toss the oldest element.
341 if (sch->sch_length >= tcp_syncache.bucket_limit) {
342 KASSERT(!TAILQ_EMPTY(&sch->sch_bucket),
343 ("sch->sch_length incorrect"));
344 sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head);
345 syncache_drop(sc2, sch);
346 tcpstat.tcps_sc_bucketoverflow++;
349 /* Put it into the bucket. */
350 TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash);
353 /* Reinitialize the bucket row's timer. */
354 if (sch->sch_length == 1)
355 sch->sch_nextc = ticks + INT_MAX;
356 syncache_timeout(sc, sch, 1);
360 tcp_syncache.cache_count++;
361 tcpstat.tcps_sc_added++;
365 * Remove and free entry from syncache bucket row.
366 * Expects locked syncache head.
369 syncache_drop(struct syncache *sc, struct syncache_head *sch)
372 SCH_LOCK_ASSERT(sch);
374 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
377 #ifndef TCP_OFFLOAD_DISABLE
379 sc->sc_tu->tu_syncache_event(TOE_SC_DROP, sc->sc_toepcb);
382 tcp_syncache.cache_count--;
386 * Engage/reengage time on bucket row.
389 syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout)
391 sc->sc_rxttime = ticks +
392 TCPTV_RTOBASE * (tcp_backoff[sc->sc_rxmits]);
394 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) {
395 sch->sch_nextc = sc->sc_rxttime;
397 callout_reset(&sch->sch_timer, sch->sch_nextc - ticks,
398 syncache_timer, (void *)sch);
403 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
404 * If we have retransmitted an entry the maximum number of times, expire it.
405 * One separate timer for each bucket row.
408 syncache_timer(void *xsch)
410 struct syncache_head *sch = (struct syncache_head *)xsch;
411 struct syncache *sc, *nsc;
415 /* NB: syncache_head has already been locked by the callout. */
416 SCH_LOCK_ASSERT(sch);
419 * In the following cycle we may remove some entries and/or
420 * advance some timeouts, so re-initialize the bucket timer.
422 sch->sch_nextc = tick + INT_MAX;
424 TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) {
426 * We do not check if the listen socket still exists
427 * and accept the case where the listen socket may be
428 * gone by the time we resend the SYN/ACK. We do
429 * not expect this to happens often. If it does,
430 * then the RST will be sent by the time the remote
431 * host does the SYN/ACK->ACK.
433 if (TSTMP_GT(sc->sc_rxttime, tick)) {
434 if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc))
435 sch->sch_nextc = sc->sc_rxttime;
439 if (sc->sc_rxmits > tcp_syncache.rexmt_limit) {
440 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
441 log(LOG_DEBUG, "%s; %s: Retransmits exhausted, "
442 "giving up and removing syncache entry\n",
446 syncache_drop(sc, sch);
447 tcpstat.tcps_sc_stale++;
450 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
451 log(LOG_DEBUG, "%s; %s: Response timeout, "
452 "retransmitting (%u) SYN|ACK\n",
453 s, __func__, sc->sc_rxmits);
457 (void) syncache_respond(sc);
458 tcpstat.tcps_sc_retransmitted++;
459 syncache_timeout(sc, sch, 0);
461 if (!TAILQ_EMPTY(&(sch)->sch_bucket))
462 callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick,
463 syncache_timer, (void *)(sch));
467 * Find an entry in the syncache.
468 * Returns always with locked syncache_head plus a matching entry or NULL.
471 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
474 struct syncache_head *sch;
477 if (inc->inc_flags & INC_ISIPV6) {
478 sch = &tcp_syncache.hashbase[
479 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
484 /* Circle through bucket row to find matching entry. */
485 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
486 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
492 sch = &tcp_syncache.hashbase[
493 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
498 /* Circle through bucket row to find matching entry. */
499 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
501 if (sc->sc_inc.inc_flags & INC_ISIPV6)
504 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
508 SCH_LOCK_ASSERT(*schp);
509 return (NULL); /* always returns with locked sch */
513 * This function is called when we get a RST for a
514 * non-existent connection, so that we can see if the
515 * connection is in the syn cache. If it is, zap it.
518 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
521 struct syncache_head *sch;
524 sc = syncache_lookup(inc, &sch); /* returns locked sch */
525 SCH_LOCK_ASSERT(sch);
528 * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags.
529 * See RFC 793 page 65, section SEGMENT ARRIVES.
531 if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) {
532 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
533 log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or "
534 "FIN flag set, segment ignored\n", s, __func__);
535 tcpstat.tcps_badrst++;
540 * No corresponding connection was found in syncache.
541 * If syncookies are enabled and possibly exclusively
542 * used, or we are under memory pressure, a valid RST
543 * may not find a syncache entry. In that case we're
544 * done and no SYN|ACK retransmissions will happen.
545 * Otherwise the the RST was misdirected or spoofed.
548 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
549 log(LOG_DEBUG, "%s; %s: Spurious RST without matching "
550 "syncache entry (possibly syncookie only), "
551 "segment ignored\n", s, __func__);
552 tcpstat.tcps_badrst++;
557 * If the RST bit is set, check the sequence number to see
558 * if this is a valid reset segment.
560 * In all states except SYN-SENT, all reset (RST) segments
561 * are validated by checking their SEQ-fields. A reset is
562 * valid if its sequence number is in the window.
564 * The sequence number in the reset segment is normally an
565 * echo of our outgoing acknowlegement numbers, but some hosts
566 * send a reset with the sequence number at the rightmost edge
567 * of our receive window, and we have to handle this case.
569 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
570 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
571 syncache_drop(sc, sch);
572 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
573 log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, "
574 "connection attempt aborted by remote endpoint\n",
576 tcpstat.tcps_sc_reset++;
578 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
579 log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != "
580 "IRS %u (+WND %u), segment ignored\n",
581 s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd);
582 tcpstat.tcps_badrst++;
592 syncache_badack(struct in_conninfo *inc)
595 struct syncache_head *sch;
597 sc = syncache_lookup(inc, &sch); /* returns locked sch */
598 SCH_LOCK_ASSERT(sch);
600 syncache_drop(sc, sch);
601 tcpstat.tcps_sc_badack++;
607 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
610 struct syncache_head *sch;
612 sc = syncache_lookup(inc, &sch); /* returns locked sch */
613 SCH_LOCK_ASSERT(sch);
617 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
618 if (ntohl(th->th_seq) != sc->sc_iss)
622 * If we've rertransmitted 3 times and this is our second error,
623 * we remove the entry. Otherwise, we allow it to continue on.
624 * This prevents us from incorrectly nuking an entry during a
625 * spurious network outage.
629 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) {
630 sc->sc_flags |= SCF_UNREACH;
633 syncache_drop(sc, sch);
634 tcpstat.tcps_sc_unreach++;
640 * Build a new TCP socket structure from a syncache entry.
642 static struct socket *
643 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
645 struct inpcb *inp = NULL;
650 INP_INFO_WLOCK_ASSERT(&tcbinfo);
653 * Ok, create the full blown connection, and set things up
654 * as they would have been set up if we had created the
655 * connection when the SYN arrived. If we can't create
656 * the connection, abort it.
658 so = sonewconn(lso, SS_ISCONNECTED);
661 * Drop the connection; we will either send a RST or
662 * have the peer retransmit its SYN again after its
665 tcpstat.tcps_listendrop++;
666 if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
667 log(LOG_DEBUG, "%s; %s: Socket create failed "
668 "due to limits or memory shortage\n",
676 mac_set_socket_peer_from_mbuf(m, so);
681 inp->inp_inc.inc_fibnum = sc->sc_inc.inc_fibnum;
682 so->so_fibnum = sc->sc_inc.inc_fibnum;
685 /* Insert new socket into PCB hash list. */
686 inp->inp_inc.inc_flags = sc->sc_inc.inc_flags;
688 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
689 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
691 inp->inp_vflag &= ~INP_IPV6;
692 inp->inp_vflag |= INP_IPV4;
694 inp->inp_laddr = sc->sc_inc.inc_laddr;
698 inp->inp_lport = sc->sc_inc.inc_lport;
699 if (in_pcbinshash(inp) != 0) {
701 * Undo the assignments above if we failed to
702 * put the PCB on the hash lists.
705 if (sc->sc_inc.inc_flags & INC_ISIPV6)
706 inp->in6p_laddr = in6addr_any;
709 inp->inp_laddr.s_addr = INADDR_ANY;
714 /* Copy old policy into new socket's. */
715 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp))
716 printf("syncache_socket: could not copy policy\n");
719 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
720 struct inpcb *oinp = sotoinpcb(lso);
721 struct in6_addr laddr6;
722 struct sockaddr_in6 sin6;
724 * Inherit socket options from the listening socket.
725 * Note that in6p_inputopts are not (and should not be)
726 * copied, since it stores previously received options and is
727 * used to detect if each new option is different than the
728 * previous one and hence should be passed to a user.
729 * If we copied in6p_inputopts, a user would not be able to
730 * receive options just after calling the accept system call.
732 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS;
733 if (oinp->in6p_outputopts)
734 inp->in6p_outputopts =
735 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
737 sin6.sin6_family = AF_INET6;
738 sin6.sin6_len = sizeof(sin6);
739 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
740 sin6.sin6_port = sc->sc_inc.inc_fport;
741 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
742 laddr6 = inp->in6p_laddr;
743 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
744 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
745 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6,
747 inp->in6p_laddr = laddr6;
750 /* Override flowlabel from in6_pcbconnect. */
751 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
752 inp->inp_flow |= sc->sc_flowlabel;
756 struct in_addr laddr;
757 struct sockaddr_in sin;
759 inp->inp_options = (m) ? ip_srcroute(m) : NULL;
761 if (inp->inp_options == NULL) {
762 inp->inp_options = sc->sc_ipopts;
763 sc->sc_ipopts = NULL;
766 sin.sin_family = AF_INET;
767 sin.sin_len = sizeof(sin);
768 sin.sin_addr = sc->sc_inc.inc_faddr;
769 sin.sin_port = sc->sc_inc.inc_fport;
770 bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero));
771 laddr = inp->inp_laddr;
772 if (inp->inp_laddr.s_addr == INADDR_ANY)
773 inp->inp_laddr = sc->sc_inc.inc_laddr;
774 if (in_pcbconnect(inp, (struct sockaddr *)&sin,
776 inp->inp_laddr = laddr;
781 tp->t_state = TCPS_SYN_RECEIVED;
782 tp->iss = sc->sc_iss;
783 tp->irs = sc->sc_irs;
786 tp->snd_wl1 = sc->sc_irs;
787 tp->snd_max = tp->iss + 1;
788 tp->snd_nxt = tp->iss + 1;
789 tp->rcv_up = sc->sc_irs + 1;
790 tp->rcv_wnd = sc->sc_wnd;
791 tp->rcv_adv += tp->rcv_wnd;
792 tp->last_ack_sent = tp->rcv_nxt;
794 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY);
795 if (sc->sc_flags & SCF_NOOPT)
796 tp->t_flags |= TF_NOOPT;
798 if (sc->sc_flags & SCF_WINSCALE) {
799 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
800 tp->snd_scale = sc->sc_requested_s_scale;
801 tp->request_r_scale = sc->sc_requested_r_scale;
803 if (sc->sc_flags & SCF_TIMESTAMP) {
804 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
805 tp->ts_recent = sc->sc_tsreflect;
806 tp->ts_recent_age = ticks;
807 tp->ts_offset = sc->sc_tsoff;
810 if (sc->sc_flags & SCF_SIGNATURE)
811 tp->t_flags |= TF_SIGNATURE;
813 if (sc->sc_flags & SCF_SACK)
814 tp->t_flags |= TF_SACK_PERMIT;
818 * Set up MSS and get cached values from tcp_hostcache.
819 * This might overwrite some of the defaults we just set.
821 tcp_mss(tp, sc->sc_peer_mss);
824 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
827 tp->snd_cwnd = tp->t_maxseg;
828 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
832 tcpstat.tcps_accepts++;
844 * This function gets called when we receive an ACK for a
845 * socket in the LISTEN state. We look up the connection
846 * in the syncache, and if its there, we pull it out of
847 * the cache and turn it into a full-blown connection in
848 * the SYN-RECEIVED state.
851 syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
852 struct socket **lsop, struct mbuf *m)
855 struct syncache_head *sch;
860 * Global TCP locks are held because we manipulate the PCB lists
861 * and create a new socket.
863 INP_INFO_WLOCK_ASSERT(&tcbinfo);
864 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK,
865 ("%s: can handle only ACK", __func__));
867 sc = syncache_lookup(inc, &sch); /* returns locked sch */
868 SCH_LOCK_ASSERT(sch);
871 * There is no syncache entry, so see if this ACK is
872 * a returning syncookie. To do this, first:
873 * A. See if this socket has had a syncache entry dropped in
874 * the past. We don't want to accept a bogus syncookie
875 * if we've never received a SYN.
876 * B. check that the syncookie is valid. If it is, then
877 * cobble up a fake syncache entry, and return.
879 if (!tcp_syncookies) {
881 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
882 log(LOG_DEBUG, "%s; %s: Spurious ACK, "
883 "segment rejected (syncookies disabled)\n",
887 bzero(&scs, sizeof(scs));
888 sc = syncookie_lookup(inc, sch, &scs, to, th, *lsop);
891 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
892 log(LOG_DEBUG, "%s; %s: Segment failed "
893 "SYNCOOKIE authentication, segment rejected "
894 "(probably spoofed)\n", s, __func__);
898 /* Pull out the entry to unlock the bucket row. */
899 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
901 tcp_syncache.cache_count--;
906 * Segment validation:
907 * ACK must match our initial sequence number + 1 (the SYN|ACK).
909 if (th->th_ack != sc->sc_iss + 1 && !TOEPCB_ISSET(sc)) {
910 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
911 log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment "
912 "rejected\n", s, __func__, th->th_ack, sc->sc_iss);
917 * The SEQ must fall in the window starting at the received
918 * initial receive sequence number + 1 (the SYN).
920 if ((SEQ_LEQ(th->th_seq, sc->sc_irs) ||
921 SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) &&
923 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
924 log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment "
925 "rejected\n", s, __func__, th->th_seq, sc->sc_irs);
930 * If timestamps were present in the SYN and we accepted
931 * them in our SYN|ACK we require them to be present from
932 * now on. And vice versa.
934 * Unfortunately, during testing of 7.0 some users found
935 * network devices that violate this constraint, so it must
938 if ((sc->sc_flags & SCF_TIMESTAMP) && !(to->to_flags & TOF_TS)) {
939 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
940 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
941 "segment rejected\n", s, __func__);
945 if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) {
946 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
947 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
948 "segment rejected\n", s, __func__);
952 * If timestamps were negotiated the reflected timestamp
953 * must be equal to what we actually sent in the SYN|ACK.
955 if ((to->to_flags & TOF_TS) && to->to_tsecr != sc->sc_ts &&
957 if ((s = tcp_log_addrs(inc, th, NULL, NULL)))
958 log(LOG_DEBUG, "%s; %s: TSECR %u != TS %u, "
959 "segment rejected\n",
960 s, __func__, to->to_tsecr, sc->sc_ts);
964 *lsop = syncache_socket(sc, *lsop, m);
967 tcpstat.tcps_sc_aborted++;
969 tcpstat.tcps_sc_completed++;
971 /* how do we find the inp for the new socket? */
976 if (sc != NULL && sc != &scs)
985 tcp_offload_syncache_expand(struct in_conninfo *inc, struct tcpopt *to,
986 struct tcphdr *th, struct socket **lsop, struct mbuf *m)
990 INP_INFO_WLOCK(&tcbinfo);
991 rc = syncache_expand(inc, to, th, lsop, m);
992 INP_INFO_WUNLOCK(&tcbinfo);
998 * Given a LISTEN socket and an inbound SYN request, add
999 * this to the syn cache, and send back a segment:
1000 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
1003 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
1004 * Doing so would require that we hold onto the data and deliver it
1005 * to the application. However, if we are the target of a SYN-flood
1006 * DoS attack, an attacker could send data which would eventually
1007 * consume all available buffer space if it were ACKed. By not ACKing
1008 * the data, we avoid this DoS scenario.
1011 _syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1012 struct inpcb *inp, struct socket **lsop, struct mbuf *m,
1013 struct toe_usrreqs *tu, void *toepcb)
1017 struct syncache *sc = NULL;
1018 struct syncache_head *sch;
1019 struct mbuf *ipopts = NULL;
1021 int win, sb_hiwat, ip_ttl, ip_tos, noopt;
1024 int autoflowlabel = 0;
1027 struct label *maclabel;
1029 struct syncache scs;
1032 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1033 INP_WLOCK_ASSERT(inp); /* listen socket */
1034 KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN,
1035 ("%s: unexpected tcp flags", __func__));
1038 * Combine all so/tp operations very early to drop the INP lock as
1043 cred = crhold(so->so_cred);
1046 if ((inc->inc_flags & INC_ISIPV6) &&
1047 (inp->inp_flags & IN6P_AUTOFLOWLABEL))
1050 ip_ttl = inp->inp_ip_ttl;
1051 ip_tos = inp->inp_ip_tos;
1052 win = sbspace(&so->so_rcv);
1053 sb_hiwat = so->so_rcv.sb_hiwat;
1054 noopt = (tp->t_flags & TF_NOOPT);
1056 /* By the time we drop the lock these should no longer be used. */
1061 if (mac_init_syncache(&maclabel) != 0) {
1063 INP_INFO_WUNLOCK(&tcbinfo);
1066 mac_init_syncache_from_inpcb(maclabel, inp);
1069 INP_INFO_WUNLOCK(&tcbinfo);
1072 * Remember the IP options, if any.
1075 if (!(inc->inc_flags & INC_ISIPV6))
1077 ipopts = (m) ? ip_srcroute(m) : NULL;
1080 * See if we already have an entry for this connection.
1081 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1083 * XXX: should the syncache be re-initialized with the contents
1084 * of the new SYN here (which may have different options?)
1086 * XXX: We do not check the sequence number to see if this is a
1087 * real retransmit or a new connection attempt. The question is
1088 * how to handle such a case; either ignore it as spoofed, or
1089 * drop the current entry and create a new one?
1091 sc = syncache_lookup(inc, &sch); /* returns locked entry */
1092 SCH_LOCK_ASSERT(sch);
1094 #ifndef TCP_OFFLOAD_DISABLE
1096 sc->sc_tu->tu_syncache_event(TOE_SC_ENTRY_PRESENT,
1099 tcpstat.tcps_sc_dupsyn++;
1102 * If we were remembering a previous source route,
1103 * forget it and use the new one we've been given.
1106 (void) m_free(sc->sc_ipopts);
1107 sc->sc_ipopts = ipopts;
1110 * Update timestamp if present.
1112 if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS))
1113 sc->sc_tsreflect = to->to_tsval;
1115 sc->sc_flags &= ~SCF_TIMESTAMP;
1118 * Since we have already unconditionally allocated label
1119 * storage, free it up. The syncache entry will already
1120 * have an initialized label we can use.
1122 mac_destroy_syncache(&maclabel);
1123 KASSERT(sc->sc_label != NULL,
1124 ("%s: label not initialized", __func__));
1126 /* Retransmit SYN|ACK and reset retransmit count. */
1127 if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) {
1128 log(LOG_DEBUG, "%s; %s: Received duplicate SYN, "
1129 "resetting timer and retransmitting SYN|ACK\n",
1133 if (!TOEPCB_ISSET(sc) && syncache_respond(sc) == 0) {
1135 syncache_timeout(sc, sch, 1);
1136 tcpstat.tcps_sndacks++;
1137 tcpstat.tcps_sndtotal++;
1143 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
1146 * The zone allocator couldn't provide more entries.
1147 * Treat this as if the cache was full; drop the oldest
1148 * entry and insert the new one.
1150 tcpstat.tcps_sc_zonefail++;
1151 if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL)
1152 syncache_drop(sc, sch);
1153 sc = uma_zalloc(tcp_syncache.zone, M_NOWAIT | M_ZERO);
1155 if (tcp_syncookies) {
1156 bzero(&scs, sizeof(scs));
1161 (void) m_free(ipopts);
1168 * Fill in the syncache values.
1171 sc->sc_label = maclabel;
1175 sc->sc_ipopts = ipopts;
1176 /* XXX-BZ this fib assignment is just useless. */
1177 sc->sc_inc.inc_fibnum = inp->inp_inc.inc_fibnum;
1178 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1180 if (!(inc->inc_flags & INC_ISIPV6))
1183 sc->sc_ip_tos = ip_tos;
1184 sc->sc_ip_ttl = ip_ttl;
1186 #ifndef TCP_OFFLOAD_DISABLE
1188 sc->sc_toepcb = toepcb;
1190 sc->sc_irs = th->th_seq;
1191 sc->sc_iss = arc4random();
1193 sc->sc_flowlabel = 0;
1196 * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN].
1197 * win was derived from socket earlier in the function.
1200 win = imin(win, TCP_MAXWIN);
1203 if (tcp_do_rfc1323) {
1205 * A timestamp received in a SYN makes
1206 * it ok to send timestamp requests and replies.
1208 if (to->to_flags & TOF_TS) {
1209 sc->sc_tsreflect = to->to_tsval;
1211 sc->sc_flags |= SCF_TIMESTAMP;
1213 if (to->to_flags & TOF_SCALE) {
1217 * Pick the smallest possible scaling factor that
1218 * will still allow us to scale up to sb_max, aka
1219 * kern.ipc.maxsockbuf.
1221 * We do this because there are broken firewalls that
1222 * will corrupt the window scale option, leading to
1223 * the other endpoint believing that our advertised
1224 * window is unscaled. At scale factors larger than
1225 * 5 the unscaled window will drop below 1500 bytes,
1226 * leading to serious problems when traversing these
1229 * With the default maxsockbuf of 256K, a scale factor
1230 * of 3 will be chosen by this algorithm. Those who
1231 * choose a larger maxsockbuf should watch out
1232 * for the compatiblity problems mentioned above.
1234 * RFC1323: The Window field in a SYN (i.e., a <SYN>
1235 * or <SYN,ACK>) segment itself is never scaled.
1237 while (wscale < TCP_MAX_WINSHIFT &&
1238 (TCP_MAXWIN << wscale) < sb_max)
1240 sc->sc_requested_r_scale = wscale;
1241 sc->sc_requested_s_scale = to->to_wscale;
1242 sc->sc_flags |= SCF_WINSCALE;
1245 #ifdef TCP_SIGNATURE
1247 * If listening socket requested TCP digests, and received SYN
1248 * contains the option, flag this in the syncache so that
1249 * syncache_respond() will do the right thing with the SYN+ACK.
1250 * XXX: Currently we always record the option by default and will
1251 * attempt to use it in syncache_respond().
1253 if (to->to_flags & TOF_SIGNATURE)
1254 sc->sc_flags |= SCF_SIGNATURE;
1256 if (to->to_flags & TOF_SACKPERM)
1257 sc->sc_flags |= SCF_SACK;
1258 if (to->to_flags & TOF_MSS)
1259 sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */
1261 sc->sc_flags |= SCF_NOOPT;
1263 if (tcp_syncookies) {
1264 syncookie_generate(sch, sc, &flowtmp);
1267 sc->sc_flowlabel = flowtmp;
1273 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1279 * Do a standard 3-way handshake.
1281 if (TOEPCB_ISSET(sc) || syncache_respond(sc) == 0) {
1282 if (tcp_syncookies && tcp_syncookiesonly && sc != &scs)
1284 else if (sc != &scs)
1285 syncache_insert(sc, sch); /* locks and unlocks sch */
1286 tcpstat.tcps_sndacks++;
1287 tcpstat.tcps_sndtotal++;
1291 tcpstat.tcps_sc_dropped++;
1299 mac_destroy_syncache(&maclabel);
1306 syncache_respond(struct syncache *sc)
1308 struct ip *ip = NULL;
1312 u_int16_t hlen, tlen, mssopt;
1315 struct ip6_hdr *ip6 = NULL;
1320 (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) :
1323 tlen = hlen + sizeof(struct tcphdr);
1325 /* Determine MSS we advertize to other end of connection. */
1326 mssopt = tcp_mssopt(&sc->sc_inc);
1327 if (sc->sc_peer_mss)
1328 mssopt = max( min(sc->sc_peer_mss, mssopt), tcp_minmss);
1330 /* XXX: Assume that the entire packet will fit in a header mbuf. */
1331 KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN,
1332 ("syncache: mbuf too small"));
1334 /* Create the IP+TCP header from scratch. */
1335 m = m_gethdr(M_DONTWAIT, MT_DATA);
1339 mac_create_mbuf_from_syncache(sc->sc_label, m);
1341 m->m_data += max_linkhdr;
1343 m->m_pkthdr.len = tlen;
1344 m->m_pkthdr.rcvif = NULL;
1347 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1348 ip6 = mtod(m, struct ip6_hdr *);
1349 ip6->ip6_vfc = IPV6_VERSION;
1350 ip6->ip6_nxt = IPPROTO_TCP;
1351 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1352 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1353 ip6->ip6_plen = htons(tlen - hlen);
1354 /* ip6_hlim is set after checksum */
1355 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
1356 ip6->ip6_flow |= sc->sc_flowlabel;
1358 th = (struct tcphdr *)(ip6 + 1);
1362 ip = mtod(m, struct ip *);
1363 ip->ip_v = IPVERSION;
1364 ip->ip_hl = sizeof(struct ip) >> 2;
1369 ip->ip_p = IPPROTO_TCP;
1370 ip->ip_src = sc->sc_inc.inc_laddr;
1371 ip->ip_dst = sc->sc_inc.inc_faddr;
1372 ip->ip_ttl = sc->sc_ip_ttl;
1373 ip->ip_tos = sc->sc_ip_tos;
1376 * See if we should do MTU discovery. Route lookups are
1377 * expensive, so we will only unset the DF bit if:
1379 * 1) path_mtu_discovery is disabled
1380 * 2) the SCF_UNREACH flag has been set
1382 if (path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0))
1383 ip->ip_off |= IP_DF;
1385 th = (struct tcphdr *)(ip + 1);
1387 th->th_sport = sc->sc_inc.inc_lport;
1388 th->th_dport = sc->sc_inc.inc_fport;
1390 th->th_seq = htonl(sc->sc_iss);
1391 th->th_ack = htonl(sc->sc_irs + 1);
1392 th->th_off = sizeof(struct tcphdr) >> 2;
1394 th->th_flags = TH_SYN|TH_ACK;
1395 th->th_win = htons(sc->sc_wnd);
1398 /* Tack on the TCP options. */
1399 if ((sc->sc_flags & SCF_NOOPT) == 0) {
1403 to.to_flags = TOF_MSS;
1404 if (sc->sc_flags & SCF_WINSCALE) {
1405 to.to_wscale = sc->sc_requested_r_scale;
1406 to.to_flags |= TOF_SCALE;
1408 if (sc->sc_flags & SCF_TIMESTAMP) {
1409 /* Virgin timestamp or TCP cookie enhanced one. */
1410 to.to_tsval = sc->sc_ts;
1411 to.to_tsecr = sc->sc_tsreflect;
1412 to.to_flags |= TOF_TS;
1414 if (sc->sc_flags & SCF_SACK)
1415 to.to_flags |= TOF_SACKPERM;
1416 #ifdef TCP_SIGNATURE
1417 if (sc->sc_flags & SCF_SIGNATURE)
1418 to.to_flags |= TOF_SIGNATURE;
1420 optlen = tcp_addoptions(&to, (u_char *)(th + 1));
1422 /* Adjust headers by option size. */
1423 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1425 m->m_pkthdr.len += optlen;
1427 #ifdef TCP_SIGNATURE
1428 if (sc->sc_flags & SCF_SIGNATURE)
1429 tcp_signature_compute(m, 0, 0, optlen,
1430 to.to_signature, IPSEC_DIR_OUTBOUND);
1433 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1434 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen);
1437 ip->ip_len += optlen;
1442 if (sc->sc_inc.inc_flags & INC_ISIPV6) {
1444 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen,
1445 tlen + optlen - hlen);
1446 ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
1447 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1451 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1452 htons(tlen + optlen - hlen + IPPROTO_TCP));
1453 m->m_pkthdr.csum_flags = CSUM_TCP;
1454 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1455 error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL);
1461 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
1462 struct inpcb *inp, struct socket **lsop, struct mbuf *m)
1465 _syncache_add(inc, to, th, inp, lsop, m, NULL, NULL);
1469 tcp_offload_syncache_add(struct in_conninfo *inc, struct tcpopt *to,
1470 struct tcphdr *th, struct inpcb *inp, struct socket **lsop,
1471 struct toe_usrreqs *tu, void *toepcb)
1475 INP_INFO_WLOCK(&tcbinfo);
1477 _syncache_add(inc, to, th, inp, lsop, NULL, tu, toepcb);
1482 * The purpose of SYN cookies is to avoid keeping track of all SYN's we
1483 * receive and to be able to handle SYN floods from bogus source addresses
1484 * (where we will never receive any reply). SYN floods try to exhaust all
1485 * our memory and available slots in the SYN cache table to cause a denial
1486 * of service to legitimate users of the local host.
1488 * The idea of SYN cookies is to encode and include all necessary information
1489 * about the connection setup state within the SYN-ACK we send back and thus
1490 * to get along without keeping any local state until the ACK to the SYN-ACK
1491 * arrives (if ever). Everything we need to know should be available from
1492 * the information we encoded in the SYN-ACK.
1494 * More information about the theory behind SYN cookies and its first
1495 * discussion and specification can be found at:
1496 * http://cr.yp.to/syncookies.html (overview)
1497 * http://cr.yp.to/syncookies/archive (gory details)
1499 * This implementation extends the orginal idea and first implementation
1500 * of FreeBSD by using not only the initial sequence number field to store
1501 * information but also the timestamp field if present. This way we can
1502 * keep track of the entire state we need to know to recreate the session in
1503 * its original form. Almost all TCP speakers implement RFC1323 timestamps
1504 * these days. For those that do not we still have to live with the known
1505 * shortcomings of the ISN only SYN cookies.
1509 * Initial sequence number we send:
1510 * 31|................................|0
1511 * DDDDDDDDDDDDDDDDDDDDDDDDDMMMRRRP
1512 * D = MD5 Digest (first dword)
1514 * R = Rotation of secret
1515 * P = Odd or Even secret
1517 * The MD5 Digest is computed with over following parameters:
1518 * a) randomly rotated secret
1519 * b) struct in_conninfo containing the remote/local ip/port (IPv4&IPv6)
1520 * c) the received initial sequence number from remote host
1521 * d) the rotation offset and odd/even bit
1523 * Timestamp we send:
1524 * 31|................................|0
1525 * DDDDDDDDDDDDDDDDDDDDDDSSSSRRRRA5
1526 * D = MD5 Digest (third dword) (only as filler)
1527 * S = Requested send window scale
1528 * R = Requested receive window scale
1530 * 5 = TCP-MD5 enabled (not implemented yet)
1531 * XORed with MD5 Digest (forth dword)
1533 * The timestamp isn't cryptographically secure and doesn't need to be.
1534 * The double use of the MD5 digest dwords ties it to a specific remote/
1535 * local host/port, remote initial sequence number and our local time
1536 * limited secret. A received timestamp is reverted (XORed) and then
1537 * the contained MD5 dword is compared to the computed one to ensure the
1538 * timestamp belongs to the SYN-ACK we sent. The other parameters may
1539 * have been tampered with but this isn't different from supplying bogus
1540 * values in the SYN in the first place.
1542 * Some problems with SYN cookies remain however:
1543 * Consider the problem of a recreated (and retransmitted) cookie. If the
1544 * original SYN was accepted, the connection is established. The second
1545 * SYN is inflight, and if it arrives with an ISN that falls within the
1546 * receive window, the connection is killed.
1549 * A heuristic to determine when to accept syn cookies is not necessary.
1550 * An ACK flood would cause the syncookie verification to be attempted,
1551 * but a SYN flood causes syncookies to be generated. Both are of equal
1552 * cost, so there's no point in trying to optimize the ACK flood case.
1553 * Also, if you don't process certain ACKs for some reason, then all someone
1554 * would have to do is launch a SYN and ACK flood at the same time, which
1555 * would stop cookie verification and defeat the entire purpose of syncookies.
1557 static int tcp_sc_msstab[] = { 0, 256, 468, 536, 996, 1452, 1460, 8960 };
1560 syncookie_generate(struct syncache_head *sch, struct syncache *sc,
1561 u_int32_t *flowlabel)
1564 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1567 u_int off, pmss, mss;
1570 SCH_LOCK_ASSERT(sch);
1572 /* Which of the two secrets to use. */
1573 secbits = sch->sch_oddeven ?
1574 sch->sch_secbits_odd : sch->sch_secbits_even;
1576 /* Reseed secret if too old. */
1577 if (sch->sch_reseed < time_uptime) {
1578 sch->sch_oddeven = sch->sch_oddeven ? 0 : 1; /* toggle */
1579 secbits = sch->sch_oddeven ?
1580 sch->sch_secbits_odd : sch->sch_secbits_even;
1581 for (i = 0; i < SYNCOOKIE_SECRET_SIZE; i++)
1582 secbits[i] = arc4random();
1583 sch->sch_reseed = time_uptime + SYNCOOKIE_LIFETIME;
1586 /* Secret rotation offset. */
1587 off = sc->sc_iss & 0x7; /* iss was randomized before */
1589 /* Maximum segment size calculation. */
1590 pmss = max( min(sc->sc_peer_mss, tcp_mssopt(&sc->sc_inc)), tcp_minmss);
1591 for (mss = sizeof(tcp_sc_msstab) / sizeof(int) - 1; mss > 0; mss--)
1592 if (tcp_sc_msstab[mss] <= pmss)
1595 /* Fold parameters and MD5 digest into the ISN we will send. */
1596 data = sch->sch_oddeven;/* odd or even secret, 1 bit */
1597 data |= off << 1; /* secret offset, derived from iss, 3 bits */
1598 data |= mss << 4; /* mss, 3 bits */
1601 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1602 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1603 MD5Update(&ctx, secbits, off);
1604 MD5Update(&ctx, &sc->sc_inc, sizeof(sc->sc_inc));
1605 MD5Update(&ctx, &sc->sc_irs, sizeof(sc->sc_irs));
1606 MD5Update(&ctx, &data, sizeof(data));
1607 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1609 data |= (md5_buffer[0] << 7);
1613 *flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1616 /* Additional parameters are stored in the timestamp if present. */
1617 if (sc->sc_flags & SCF_TIMESTAMP) {
1618 data = ((sc->sc_flags & SCF_SIGNATURE) ? 1 : 0); /* TCP-MD5, 1 bit */
1619 data |= ((sc->sc_flags & SCF_SACK) ? 1 : 0) << 1; /* SACK, 1 bit */
1620 data |= sc->sc_requested_s_scale << 2; /* SWIN scale, 4 bits */
1621 data |= sc->sc_requested_r_scale << 6; /* RWIN scale, 4 bits */
1622 data |= md5_buffer[2] << 10; /* more digest bits */
1623 data ^= md5_buffer[3];
1625 sc->sc_tsoff = data - ticks; /* after XOR */
1628 tcpstat.tcps_sc_sendcookie++;
1631 static struct syncache *
1632 syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch,
1633 struct syncache *sc, struct tcpopt *to, struct tcphdr *th,
1637 u_int32_t md5_buffer[MD5_DIGEST_LENGTH / sizeof(u_int32_t)];
1641 int off, mss, wnd, flags;
1643 SCH_LOCK_ASSERT(sch);
1646 * Pull information out of SYN-ACK/ACK and
1647 * revert sequence number advances.
1649 ack = th->th_ack - 1;
1650 seq = th->th_seq - 1;
1651 off = (ack >> 1) & 0x7;
1652 mss = (ack >> 4) & 0x7;
1655 /* Which of the two secrets to use. */
1656 secbits = (flags & 0x1) ? sch->sch_secbits_odd : sch->sch_secbits_even;
1659 * The secret wasn't updated for the lifetime of a syncookie,
1660 * so this SYN-ACK/ACK is either too old (replay) or totally bogus.
1662 if (sch->sch_reseed + SYNCOOKIE_LIFETIME < time_uptime) {
1666 /* Recompute the digest so we can compare it. */
1668 MD5Update(&ctx, ((u_int8_t *)secbits) + off,
1669 SYNCOOKIE_SECRET_SIZE * sizeof(*secbits) - off);
1670 MD5Update(&ctx, secbits, off);
1671 MD5Update(&ctx, inc, sizeof(*inc));
1672 MD5Update(&ctx, &seq, sizeof(seq));
1673 MD5Update(&ctx, &flags, sizeof(flags));
1674 MD5Final((u_int8_t *)&md5_buffer, &ctx);
1676 /* Does the digest part of or ACK'ed ISS match? */
1677 if ((ack & (~0x7f)) != (md5_buffer[0] << 7))
1680 /* Does the digest part of our reflected timestamp match? */
1681 if (to->to_flags & TOF_TS) {
1682 data = md5_buffer[3] ^ to->to_tsecr;
1683 if ((data & (~0x3ff)) != (md5_buffer[2] << 10))
1687 /* Fill in the syncache values. */
1688 bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo));
1689 sc->sc_ipopts = NULL;
1695 if (inc->inc_flags & INC_ISIPV6) {
1696 if (sotoinpcb(so)->inp_flags & IN6P_AUTOFLOWLABEL)
1697 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK;
1701 sc->sc_ip_ttl = sotoinpcb(so)->inp_ip_ttl;
1702 sc->sc_ip_tos = sotoinpcb(so)->inp_ip_tos;
1705 /* Additional parameters that were encoded in the timestamp. */
1707 sc->sc_flags |= SCF_TIMESTAMP;
1708 sc->sc_tsreflect = to->to_tsval;
1709 sc->sc_ts = to->to_tsecr;
1710 sc->sc_tsoff = to->to_tsecr - ticks;
1711 sc->sc_flags |= (data & 0x1) ? SCF_SIGNATURE : 0;
1712 sc->sc_flags |= ((data >> 1) & 0x1) ? SCF_SACK : 0;
1713 sc->sc_requested_s_scale = min((data >> 2) & 0xf,
1715 sc->sc_requested_r_scale = min((data >> 6) & 0xf,
1717 if (sc->sc_requested_s_scale || sc->sc_requested_r_scale)
1718 sc->sc_flags |= SCF_WINSCALE;
1720 sc->sc_flags |= SCF_NOOPT;
1722 wnd = sbspace(&so->so_rcv);
1724 wnd = imin(wnd, TCP_MAXWIN);
1728 sc->sc_peer_mss = tcp_sc_msstab[mss];
1730 tcpstat.tcps_sc_recvcookie++;
1735 * Returns the current number of syncache entries. This number
1736 * will probably change before you get around to calling
1741 syncache_pcbcount(void)
1743 struct syncache_head *sch;
1746 for (count = 0, i = 0; i < tcp_syncache.hashsize; i++) {
1747 /* No need to lock for a read. */
1748 sch = &tcp_syncache.hashbase[i];
1749 count += sch->sch_length;
1755 * Exports the syncache entries to userland so that netstat can display
1756 * them alongside the other sockets. This function is intended to be
1757 * called only from tcp_pcblist.
1759 * Due to concurrency on an active system, the number of pcbs exported
1760 * may have no relation to max_pcbs. max_pcbs merely indicates the
1761 * amount of space the caller allocated for this function to use.
1764 syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported)
1767 struct syncache *sc;
1768 struct syncache_head *sch;
1769 int count, error, i;
1771 for (count = 0, error = 0, i = 0; i < tcp_syncache.hashsize; i++) {
1772 sch = &tcp_syncache.hashbase[i];
1774 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
1775 if (count >= max_pcbs) {
1779 if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0)
1781 bzero(&xt, sizeof(xt));
1782 xt.xt_len = sizeof(xt);
1783 if (sc->sc_inc.inc_flags & INC_ISIPV6)
1784 xt.xt_inp.inp_vflag = INP_IPV6;
1786 xt.xt_inp.inp_vflag = INP_IPV4;
1787 bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo));
1788 xt.xt_tp.t_inpcb = &xt.xt_inp;
1789 xt.xt_tp.t_state = TCPS_SYN_RECEIVED;
1790 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1791 xt.xt_socket.xso_len = sizeof (struct xsocket);
1792 xt.xt_socket.so_type = SOCK_STREAM;
1793 xt.xt_socket.so_state = SS_ISCONNECTING;
1794 error = SYSCTL_OUT(req, &xt, sizeof xt);
1804 *pcbs_exported = count;