2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * All advertising materials mentioning features or use of this software
36 * must display the following acknowledgement:
37 * This product includes software developed by Jeffrey M. Hsu.
39 * Copyright (c) 2001 Networks Associates Technologies, Inc.
40 * All rights reserved.
42 * This software was developed for the FreeBSD Project by Jonathan Lemon
43 * and NAI Labs, the Security Research Division of Network Associates, Inc.
44 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
45 * DARPA CHATS research program.
47 * Redistribution and use in source and binary forms, with or without
48 * modification, are permitted provided that the following conditions
50 * 1. Redistributions of source code must retain the above copyright
51 * notice, this list of conditions and the following disclaimer.
52 * 2. Redistributions in binary form must reproduce the above copyright
53 * notice, this list of conditions and the following disclaimer in the
54 * documentation and/or other materials provided with the distribution.
55 * 3. The name of the author may not be used to endorse or promote
56 * products derived from this software without specific prior written
59 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
71 * $FreeBSD: src/sys/netinet/tcp_syncache.c,v 1.5.2.14 2003/02/24 04:02:27 silby Exp $
72 * $DragonFly: src/sys/netinet/tcp_syncache.c,v 1.35 2008/11/22 11:03:35 sephe Exp $
75 #include "opt_inet6.h"
76 #include "opt_ipsec.h"
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/kernel.h>
81 #include <sys/sysctl.h>
82 #include <sys/malloc.h>
85 #include <sys/proc.h> /* for proc0 declaration */
86 #include <sys/random.h>
87 #include <sys/socket.h>
88 #include <sys/socketvar.h>
89 #include <sys/in_cksum.h>
91 #include <sys/msgport2.h>
92 #include <net/netmsg2.h>
95 #include <net/route.h>
97 #include <netinet/in.h>
98 #include <netinet/in_systm.h>
99 #include <netinet/ip.h>
100 #include <netinet/in_var.h>
101 #include <netinet/in_pcb.h>
102 #include <netinet/ip_var.h>
103 #include <netinet/ip6.h>
105 #include <netinet/icmp6.h>
106 #include <netinet6/nd6.h>
108 #include <netinet6/ip6_var.h>
109 #include <netinet6/in6_pcb.h>
110 #include <netinet/tcp.h>
111 #include <netinet/tcp_fsm.h>
112 #include <netinet/tcp_seq.h>
113 #include <netinet/tcp_timer.h>
114 #include <netinet/tcp_timer2.h>
115 #include <netinet/tcp_var.h>
116 #include <netinet6/tcp6_var.h>
119 #include <netinet6/ipsec.h>
121 #include <netinet6/ipsec6.h>
123 #include <netproto/key/key.h>
127 #include <netproto/ipsec/ipsec.h>
129 #include <netproto/ipsec/ipsec6.h>
131 #include <netproto/ipsec/key.h>
133 #endif /*FAST_IPSEC*/
135 #include <vm/vm_zone.h>
137 static int tcp_syncookies = 1;
138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
140 "Use TCP SYN cookies if the syncache overflows");
142 static void syncache_drop(struct syncache *, struct syncache_head *);
143 static void syncache_free(struct syncache *);
144 static void syncache_insert(struct syncache *, struct syncache_head *);
145 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
146 static int syncache_respond(struct syncache *, struct mbuf *);
147 static struct socket *syncache_socket(struct syncache *, struct socket *,
149 static void syncache_timer(void *);
150 static u_int32_t syncookie_generate(struct syncache *);
151 static struct syncache *syncookie_lookup(struct in_conninfo *,
152 struct tcphdr *, struct socket *);
155 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
156 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds,
157 * the odds are that the user has given up attempting to connect by then.
159 #define SYNCACHE_MAXREXMTS 3
161 /* Arbitrary values */
162 #define TCP_SYNCACHE_HASHSIZE 512
163 #define TCP_SYNCACHE_BUCKETLIMIT 30
165 struct netmsg_sc_timer {
166 struct netmsg nm_netmsg;
167 struct msgrec *nm_mrec; /* back pointer to containing msgrec */
171 struct netmsg_sc_timer msg;
172 lwkt_port_t port; /* constant after init */
173 int slot; /* constant after init */
176 static void syncache_timer_handler(netmsg_t);
178 struct tcp_syncache {
179 struct vm_zone *zone;
187 static struct tcp_syncache tcp_syncache;
189 struct tcp_syncache_percpu {
190 struct syncache_head *hashbase;
192 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1];
193 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1];
194 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1];
196 static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU];
198 static struct lwkt_port syncache_null_rport;
200 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
202 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD,
203 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
205 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD,
206 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
210 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
211 &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
214 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD,
215 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
217 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
218 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
220 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
222 #define SYNCACHE_HASH(inc, mask) \
223 ((tcp_syncache.hash_secret ^ \
224 (inc)->inc_faddr.s_addr ^ \
225 ((inc)->inc_faddr.s_addr >> 16) ^ \
226 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
228 #define SYNCACHE_HASH6(inc, mask) \
229 ((tcp_syncache.hash_secret ^ \
230 (inc)->inc6_faddr.s6_addr32[0] ^ \
231 (inc)->inc6_faddr.s6_addr32[3] ^ \
232 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
234 #define ENDPTS_EQ(a, b) ( \
235 (a)->ie_fport == (b)->ie_fport && \
236 (a)->ie_lport == (b)->ie_lport && \
237 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
238 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
241 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
244 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu,
245 struct syncache *sc, int slot)
247 sc->sc_rxtslot = slot;
248 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[slot];
249 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq);
250 if (!callout_active(&syncache_percpu->tt_timerq[slot])) {
251 callout_reset(&syncache_percpu->tt_timerq[slot],
252 TCPTV_RTOBASE * tcp_backoff[slot],
254 &syncache_percpu->mrec[slot]);
259 syncache_free(struct syncache *sc)
263 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
265 const boolean_t isipv6 = FALSE;
269 m_free(sc->sc_ipopts);
271 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt;
274 * If this is the only reference to a protocol-cloned
275 * route, remove it immediately.
277 if ((rt->rt_flags & RTF_WASCLONED) && rt->rt_refcnt == 1)
278 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
279 rt_mask(rt), rt->rt_flags, NULL);
283 zfree(tcp_syncache.zone, sc);
291 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
292 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
293 tcp_syncache.cache_limit =
294 tcp_syncache.hashsize * tcp_syncache.bucket_limit;
295 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
296 tcp_syncache.hash_secret = karc4random();
298 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
299 &tcp_syncache.hashsize);
300 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
301 &tcp_syncache.cache_limit);
302 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
303 &tcp_syncache.bucket_limit);
304 if (!powerof2(tcp_syncache.hashsize)) {
305 kprintf("WARNING: syncache hash size is not a power of 2.\n");
306 tcp_syncache.hashsize = 512; /* safe default */
308 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
310 lwkt_initport_replyonly_null(&syncache_null_rport);
312 for (cpu = 0; cpu < ncpus2; cpu++) {
313 struct tcp_syncache_percpu *syncache_percpu;
315 syncache_percpu = &tcp_syncache_percpu[cpu];
316 /* Allocate the hash table. */
317 MALLOC(syncache_percpu->hashbase, struct syncache_head *,
318 tcp_syncache.hashsize * sizeof(struct syncache_head),
319 M_SYNCACHE, M_WAITOK);
321 /* Initialize the hash buckets. */
322 for (i = 0; i < tcp_syncache.hashsize; i++) {
323 struct syncache_head *bucket;
325 bucket = &syncache_percpu->hashbase[i];
326 TAILQ_INIT(&bucket->sch_bucket);
327 bucket->sch_length = 0;
330 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) {
331 /* Initialize the timer queues. */
332 TAILQ_INIT(&syncache_percpu->timerq[i]);
333 callout_init(&syncache_percpu->tt_timerq[i]);
335 syncache_percpu->mrec[i].slot = i;
336 syncache_percpu->mrec[i].port = tcp_cport(cpu);
337 syncache_percpu->mrec[i].msg.nm_mrec =
338 &syncache_percpu->mrec[i];
339 netmsg_init(&syncache_percpu->mrec[i].msg.nm_netmsg,
340 NULL, &syncache_null_rport,
341 0, syncache_timer_handler);
346 * Allocate the syncache entries. Allow the zone to allocate one
347 * more entry than cache limit, so a new entry can bump out an
350 tcp_syncache.zone = zinit("syncache", sizeof(struct syncache),
351 tcp_syncache.cache_limit * ncpus2, ZONE_INTERRUPT, 0);
352 tcp_syncache.cache_limit -= 1;
356 syncache_insert(struct syncache *sc, struct syncache_head *sch)
358 struct tcp_syncache_percpu *syncache_percpu;
359 struct syncache *sc2;
362 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
365 * Make sure that we don't overflow the per-bucket
366 * limit or the total cache size limit.
368 if (sch->sch_length >= tcp_syncache.bucket_limit) {
370 * The bucket is full, toss the oldest element.
372 sc2 = TAILQ_FIRST(&sch->sch_bucket);
373 sc2->sc_tp->ts_recent = ticks;
374 syncache_drop(sc2, sch);
375 tcpstat.tcps_sc_bucketoverflow++;
376 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) {
378 * The cache is full. Toss the oldest entry in the
379 * entire cache. This is the front entry in the
380 * first non-empty timer queue with the largest
383 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
384 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]);
388 sc2->sc_tp->ts_recent = ticks;
389 syncache_drop(sc2, NULL);
390 tcpstat.tcps_sc_cacheoverflow++;
393 /* Initialize the entry's timer. */
394 syncache_timeout(syncache_percpu, sc, 0);
396 /* Put it into the bucket. */
397 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash);
399 syncache_percpu->cache_count++;
400 tcpstat.tcps_sc_added++;
404 syncache_drop(struct syncache *sc, struct syncache_head *sch)
406 struct tcp_syncache_percpu *syncache_percpu;
408 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
410 const boolean_t isipv6 = FALSE;
413 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
417 sch = &syncache_percpu->hashbase[
418 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)];
420 sch = &syncache_percpu->hashbase[
421 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)];
425 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
427 syncache_percpu->cache_count--;
430 * Remove the entry from the syncache timer/timeout queue. Note
431 * that we do not try to stop any running timer since we do not know
432 * whether the timer's message is in-transit or not. Since timeouts
433 * are fairly long, taking an unneeded callout does not detrimentally
434 * effect performance.
436 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq);
442 * Place a timeout message on the TCP thread's message queue.
443 * This routine runs in soft interrupt context.
445 * An invariant is for this routine to be called, the callout must
446 * have been active. Note that the callout is not deactivated until
447 * after the message has been processed in syncache_timer_handler() below.
450 syncache_timer(void *p)
452 struct netmsg_sc_timer *msg = p;
454 lwkt_sendmsg(msg->nm_mrec->port, &msg->nm_netmsg.nm_lmsg);
458 * Service a timer message queued by timer expiration.
459 * This routine runs in the TCP protocol thread.
461 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
462 * If we have retransmitted an entry the maximum number of times, expire it.
464 * When we finish processing timed-out entries, we restart the timer if there
465 * are any entries still on the queue and deactivate it otherwise. Only after
466 * a timer has been deactivated here can it be restarted by syncache_timeout().
469 syncache_timer_handler(netmsg_t netmsg)
471 struct tcp_syncache_percpu *syncache_percpu;
472 struct syncache *sc, *nsc;
476 slot = ((struct netmsg_sc_timer *)netmsg)->nm_mrec->slot;
477 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
479 nsc = TAILQ_FIRST(&syncache_percpu->timerq[slot]);
480 while (nsc != NULL) {
481 if (ticks < nsc->sc_rxttime)
482 break; /* finished because timerq sorted by time */
484 inp = sc->sc_tp->t_inpcb;
485 if (slot == SYNCACHE_MAXREXMTS ||
486 slot >= tcp_syncache.rexmt_limit ||
487 inp->inp_gencnt != sc->sc_inp_gencnt) {
488 nsc = TAILQ_NEXT(sc, sc_timerq);
489 syncache_drop(sc, NULL);
490 tcpstat.tcps_sc_stale++;
494 * syncache_respond() may call back into the syncache to
495 * to modify another entry, so do not obtain the next
496 * entry on the timer chain until it has completed.
498 syncache_respond(sc, NULL);
499 nsc = TAILQ_NEXT(sc, sc_timerq);
500 tcpstat.tcps_sc_retransmitted++;
501 TAILQ_REMOVE(&syncache_percpu->timerq[slot], sc, sc_timerq);
502 syncache_timeout(syncache_percpu, sc, slot + 1);
505 callout_reset(&syncache_percpu->tt_timerq[slot],
506 nsc->sc_rxttime - ticks, syncache_timer,
507 &syncache_percpu->mrec[slot]);
509 callout_deactivate(&syncache_percpu->tt_timerq[slot]);
511 lwkt_replymsg(&netmsg->nm_lmsg, 0);
515 * Find an entry in the syncache.
518 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
520 struct tcp_syncache_percpu *syncache_percpu;
522 struct syncache_head *sch;
524 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
526 if (inc->inc_isipv6) {
527 sch = &syncache_percpu->hashbase[
528 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
530 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash)
531 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
536 sch = &syncache_percpu->hashbase[
537 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
539 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
541 if (sc->sc_inc.inc_isipv6)
544 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
552 * This function is called when we get a RST for a
553 * non-existent connection, so that we can see if the
554 * connection is in the syn cache. If it is, zap it.
557 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
560 struct syncache_head *sch;
562 sc = syncache_lookup(inc, &sch);
566 * If the RST bit is set, check the sequence number to see
567 * if this is a valid reset segment.
569 * In all states except SYN-SENT, all reset (RST) segments
570 * are validated by checking their SEQ-fields. A reset is
571 * valid if its sequence number is in the window.
573 * The sequence number in the reset segment is normally an
574 * echo of our outgoing acknowlegement numbers, but some hosts
575 * send a reset with the sequence number at the rightmost edge
576 * of our receive window, and we have to handle this case.
578 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
579 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
580 syncache_drop(sc, sch);
581 tcpstat.tcps_sc_reset++;
586 syncache_badack(struct in_conninfo *inc)
589 struct syncache_head *sch;
591 sc = syncache_lookup(inc, &sch);
593 syncache_drop(sc, sch);
594 tcpstat.tcps_sc_badack++;
599 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
602 struct syncache_head *sch;
604 /* we are called at splnet() here */
605 sc = syncache_lookup(inc, &sch);
609 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
610 if (ntohl(th->th_seq) != sc->sc_iss)
614 * If we've rertransmitted 3 times and this is our second error,
615 * we remove the entry. Otherwise, we allow it to continue on.
616 * This prevents us from incorrectly nuking an entry during a
617 * spurious network outage.
621 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) {
622 sc->sc_flags |= SCF_UNREACH;
625 syncache_drop(sc, sch);
626 tcpstat.tcps_sc_unreach++;
630 * Build a new TCP socket structure from a syncache entry.
632 * This is called from the context of the SYN+ACK
634 static struct socket *
635 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
637 struct inpcb *inp = NULL, *linp;
642 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
644 const boolean_t isipv6 = FALSE;
648 * Ok, create the full blown connection, and set things up
649 * as they would have been set up if we had created the
650 * connection when the SYN arrived. If we can't create
651 * the connection, abort it.
653 so = sonewconn(lso, SS_ISCONNECTED);
656 * Drop the connection; we will send a RST if the peer
657 * retransmits the ACK,
659 tcpstat.tcps_listendrop++;
664 * Set the protocol processing port for the socket to the current
665 * port (that the connection came in on).
667 sosetport(so, &curthread->td_msgport);
670 * Insert new socket into hash list.
673 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
675 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
678 inp->inp_vflag &= ~INP_IPV6;
679 inp->inp_vflag |= INP_IPV4;
680 inp->inp_flags &= ~IN6P_IPV6_V6ONLY;
682 inp->inp_laddr = sc->sc_inc.inc_laddr;
684 inp->inp_lport = sc->sc_inc.inc_lport;
685 if (in_pcbinsporthash(inp) != 0) {
687 * Undo the assignments above if we failed to
688 * put the PCB on the hash lists.
691 inp->in6p_laddr = kin6addr_any;
693 inp->inp_laddr.s_addr = INADDR_ANY;
699 /* copy old policy into new socket's */
700 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp))
701 kprintf("syncache_expand: could not copy policy\n");
704 struct in6_addr laddr6;
705 struct sockaddr_in6 sin6;
707 * Inherit socket options from the listening socket.
708 * Note that in6p_inputopts are not (and should not be)
709 * copied, since it stores previously received options and is
710 * used to detect if each new option is different than the
711 * previous one and hence should be passed to a user.
712 * If we copied in6p_inputopts, a user would not be able to
713 * receive options just after calling the accept system call.
715 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS;
716 if (linp->in6p_outputopts)
717 inp->in6p_outputopts =
718 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT);
719 inp->in6p_route = sc->sc_route6;
720 sc->sc_route6.ro_rt = NULL;
722 sin6.sin6_family = AF_INET6;
723 sin6.sin6_len = sizeof sin6;
724 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
725 sin6.sin6_port = sc->sc_inc.inc_fport;
726 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
727 laddr6 = inp->in6p_laddr;
728 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
729 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
730 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) {
731 inp->in6p_laddr = laddr6;
735 struct in_addr laddr;
736 struct sockaddr_in sin;
738 inp->inp_options = ip_srcroute(m);
739 if (inp->inp_options == NULL) {
740 inp->inp_options = sc->sc_ipopts;
741 sc->sc_ipopts = NULL;
743 inp->inp_route = sc->sc_route;
744 sc->sc_route.ro_rt = NULL;
746 sin.sin_family = AF_INET;
747 sin.sin_len = sizeof sin;
748 sin.sin_addr = sc->sc_inc.inc_faddr;
749 sin.sin_port = sc->sc_inc.inc_fport;
750 bzero(sin.sin_zero, sizeof sin.sin_zero);
751 laddr = inp->inp_laddr;
752 if (inp->inp_laddr.s_addr == INADDR_ANY)
753 inp->inp_laddr = sc->sc_inc.inc_laddr;
754 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) {
755 inp->inp_laddr = laddr;
761 * The current port should be in the context of the SYN+ACK and
762 * so should match the tcp address port.
764 * XXX we may be running on the netisr thread instead of a tcp
765 * thread, in which case port will not match
766 * curthread->td_msgport.
769 port = tcp6_addrport();
771 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport,
772 inp->inp_laddr.s_addr, inp->inp_lport);
774 /*KKASSERT(port == &curthread->td_msgport);*/
777 tp->t_state = TCPS_SYN_RECEIVED;
778 tp->iss = sc->sc_iss;
779 tp->irs = sc->sc_irs;
782 tp->snd_wl1 = sc->sc_irs;
783 tp->rcv_up = sc->sc_irs + 1;
784 tp->rcv_wnd = sc->sc_wnd;
785 tp->rcv_adv += tp->rcv_wnd;
787 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY);
788 if (sc->sc_flags & SCF_NOOPT)
789 tp->t_flags |= TF_NOOPT;
790 if (sc->sc_flags & SCF_WINSCALE) {
791 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE;
792 tp->requested_s_scale = sc->sc_requested_s_scale;
793 tp->request_r_scale = sc->sc_request_r_scale;
795 if (sc->sc_flags & SCF_TIMESTAMP) {
796 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP;
797 tp->ts_recent = sc->sc_tsrecent;
798 tp->ts_recent_age = ticks;
800 if (sc->sc_flags & SCF_SACK_PERMITTED)
801 tp->t_flags |= TF_SACK_PERMITTED;
803 tcp_mss(tp, sc->sc_peer_mss);
806 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
808 if (sc->sc_rxtslot != 0)
809 tp->snd_cwnd = tp->t_maxseg;
810 tcp_create_timermsg(tp, port);
811 tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep);
813 tcpstat.tcps_accepts++;
823 * This function gets called when we receive an ACK for a
824 * socket in the LISTEN state. We look up the connection
825 * in the syncache, and if its there, we pull it out of
826 * the cache and turn it into a full-blown connection in
827 * the SYN-RECEIVED state.
830 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop,
834 struct syncache_head *sch;
837 sc = syncache_lookup(inc, &sch);
840 * There is no syncache entry, so see if this ACK is
841 * a returning syncookie. To do this, first:
842 * A. See if this socket has had a syncache entry dropped in
843 * the past. We don't want to accept a bogus syncookie
844 * if we've never received a SYN.
845 * B. check that the syncookie is valid. If it is, then
846 * cobble up a fake syncache entry, and return.
850 sc = syncookie_lookup(inc, th, *sop);
854 tcpstat.tcps_sc_recvcookie++;
858 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
860 if (th->th_ack != sc->sc_iss + 1)
863 so = syncache_socket(sc, *sop, m);
867 /* XXXjlemon check this - is this correct? */
868 tcp_respond(NULL, m, m, th,
869 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK);
871 m_freem(m); /* XXX only needed for above */
872 tcpstat.tcps_sc_aborted++;
874 tcpstat.tcps_sc_completed++;
879 syncache_drop(sc, sch);
885 * Given a LISTEN socket and an inbound SYN request, add
886 * this to the syn cache, and send back a segment:
887 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
890 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
891 * Doing so would require that we hold onto the data and deliver it
892 * to the application. However, if we are the target of a SYN-flood
893 * DoS attack, an attacker could send data which would eventually
894 * consume all available buffer space if it were ACKed. By not ACKing
895 * the data, we avoid this DoS scenario.
898 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
899 struct socket **sop, struct mbuf *m)
901 struct tcp_syncache_percpu *syncache_percpu;
904 struct syncache *sc = NULL;
905 struct syncache_head *sch;
906 struct mbuf *ipopts = NULL;
909 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
914 * Remember the IP options, if any.
917 if (!inc->inc_isipv6)
919 ipopts = ip_srcroute(m);
922 * See if we already have an entry for this connection.
923 * If we do, resend the SYN,ACK, and reset the retransmit timer.
926 * The syncache should be re-initialized with the contents
927 * of the new SYN which may have different options.
929 sc = syncache_lookup(inc, &sch);
931 tcpstat.tcps_sc_dupsyn++;
934 * If we were remembering a previous source route,
935 * forget it and use the new one we've been given.
938 m_free(sc->sc_ipopts);
939 sc->sc_ipopts = ipopts;
942 * Update timestamp if present.
944 if (sc->sc_flags & SCF_TIMESTAMP)
945 sc->sc_tsrecent = to->to_tsval;
947 /* Just update the TOF_SACK_PERMITTED for now. */
948 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
949 sc->sc_flags |= SCF_SACK_PERMITTED;
951 sc->sc_flags &= ~SCF_SACK_PERMITTED;
954 * PCB may have changed, pick up new values.
957 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
958 if (syncache_respond(sc, m) == 0) {
959 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot],
961 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot);
962 tcpstat.tcps_sndacks++;
963 tcpstat.tcps_sndtotal++;
970 * This allocation is guaranteed to succeed because we
971 * preallocate one more syncache entry than cache_limit.
973 sc = zalloc(tcp_syncache.zone);
976 * Fill in the syncache values.
979 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
980 sc->sc_ipopts = ipopts;
981 sc->sc_inc.inc_fport = inc->inc_fport;
982 sc->sc_inc.inc_lport = inc->inc_lport;
984 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
985 if (inc->inc_isipv6) {
986 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
987 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
988 sc->sc_route6.ro_rt = NULL;
992 sc->sc_inc.inc_faddr = inc->inc_faddr;
993 sc->sc_inc.inc_laddr = inc->inc_laddr;
994 sc->sc_route.ro_rt = NULL;
996 sc->sc_irs = th->th_seq;
998 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0;
1000 sc->sc_iss = syncookie_generate(sc);
1002 sc->sc_iss = karc4random();
1004 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */
1005 win = ssb_space(&so->so_rcv);
1007 win = imin(win, TCP_MAXWIN);
1010 if (tcp_do_rfc1323) {
1012 * A timestamp received in a SYN makes
1013 * it ok to send timestamp requests and replies.
1015 if (to->to_flags & TOF_TS) {
1016 sc->sc_tsrecent = to->to_tsval;
1017 sc->sc_flags |= SCF_TIMESTAMP;
1019 if (to->to_flags & TOF_SCALE) {
1020 int wscale = TCP_MIN_WINSHIFT;
1022 /* Compute proper scaling value from buffer space */
1023 while (wscale < TCP_MAX_WINSHIFT &&
1024 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) {
1027 sc->sc_request_r_scale = wscale;
1028 sc->sc_requested_s_scale = to->to_requested_s_scale;
1029 sc->sc_flags |= SCF_WINSCALE;
1032 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1033 sc->sc_flags |= SCF_SACK_PERMITTED;
1034 if (tp->t_flags & TF_NOOPT)
1035 sc->sc_flags = SCF_NOOPT;
1037 if (syncache_respond(sc, m) == 0) {
1038 syncache_insert(sc, sch);
1039 tcpstat.tcps_sndacks++;
1040 tcpstat.tcps_sndtotal++;
1043 tcpstat.tcps_sc_dropped++;
1050 syncache_respond(struct syncache *sc, struct mbuf *m)
1054 u_int16_t tlen, hlen, mssopt;
1055 struct ip *ip = NULL;
1058 struct ip6_hdr *ip6 = NULL;
1060 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1062 const boolean_t isipv6 = FALSE;
1066 rt = tcp_rtlookup6(&sc->sc_inc);
1068 mssopt = rt->rt_ifp->if_mtu -
1069 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1071 mssopt = tcp_v6mssdflt;
1072 hlen = sizeof(struct ip6_hdr);
1074 rt = tcp_rtlookup(&sc->sc_inc);
1076 mssopt = rt->rt_ifp->if_mtu -
1077 (sizeof(struct ip) + sizeof(struct tcphdr));
1079 mssopt = tcp_mssdflt;
1080 hlen = sizeof(struct ip);
1083 /* Compute the size of the TCP options. */
1084 if (sc->sc_flags & SCF_NOOPT) {
1087 optlen = TCPOLEN_MAXSEG +
1088 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) +
1089 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) +
1090 ((sc->sc_flags & SCF_SACK_PERMITTED) ?
1091 TCPOLEN_SACK_PERMITTED_ALIGNED : 0);
1093 tlen = hlen + sizeof(struct tcphdr) + optlen;
1097 * assume that the entire packet will fit in a header mbuf
1099 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small"));
1102 * XXX shouldn't this reuse the mbuf if possible ?
1103 * Create the IP+TCP header from scratch.
1108 m = m_gethdr(MB_DONTWAIT, MT_HEADER);
1111 m->m_data += max_linkhdr;
1113 m->m_pkthdr.len = tlen;
1114 m->m_pkthdr.rcvif = NULL;
1117 ip6 = mtod(m, struct ip6_hdr *);
1118 ip6->ip6_vfc = IPV6_VERSION;
1119 ip6->ip6_nxt = IPPROTO_TCP;
1120 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1121 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1122 ip6->ip6_plen = htons(tlen - hlen);
1123 /* ip6_hlim is set after checksum */
1124 /* ip6_flow = ??? */
1126 th = (struct tcphdr *)(ip6 + 1);
1128 ip = mtod(m, struct ip *);
1129 ip->ip_v = IPVERSION;
1130 ip->ip_hl = sizeof(struct ip) >> 2;
1135 ip->ip_p = IPPROTO_TCP;
1136 ip->ip_src = sc->sc_inc.inc_laddr;
1137 ip->ip_dst = sc->sc_inc.inc_faddr;
1138 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */
1139 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */
1142 * See if we should do MTU discovery. Route lookups are
1143 * expensive, so we will only unset the DF bit if:
1145 * 1) path_mtu_discovery is disabled
1146 * 2) the SCF_UNREACH flag has been set
1148 if (path_mtu_discovery
1149 && ((sc->sc_flags & SCF_UNREACH) == 0)) {
1150 ip->ip_off |= IP_DF;
1153 th = (struct tcphdr *)(ip + 1);
1155 th->th_sport = sc->sc_inc.inc_lport;
1156 th->th_dport = sc->sc_inc.inc_fport;
1158 th->th_seq = htonl(sc->sc_iss);
1159 th->th_ack = htonl(sc->sc_irs + 1);
1160 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1162 th->th_flags = TH_SYN | TH_ACK;
1163 th->th_win = htons(sc->sc_wnd);
1166 /* Tack on the TCP options. */
1169 optp = (u_int8_t *)(th + 1);
1170 *optp++ = TCPOPT_MAXSEG;
1171 *optp++ = TCPOLEN_MAXSEG;
1172 *optp++ = (mssopt >> 8) & 0xff;
1173 *optp++ = mssopt & 0xff;
1175 if (sc->sc_flags & SCF_WINSCALE) {
1176 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
1177 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
1178 sc->sc_request_r_scale);
1182 if (sc->sc_flags & SCF_TIMESTAMP) {
1183 u_int32_t *lp = (u_int32_t *)(optp);
1185 /* Form timestamp option as shown in appendix A of RFC 1323. */
1186 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1187 *lp++ = htonl(ticks);
1188 *lp = htonl(sc->sc_tsrecent);
1189 optp += TCPOLEN_TSTAMP_APPA;
1192 if (sc->sc_flags & SCF_SACK_PERMITTED) {
1193 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED);
1194 optp += TCPOLEN_SACK_PERMITTED_ALIGNED;
1199 struct route_in6 *ro6 = &sc->sc_route6;
1202 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
1203 ip6->ip6_hlim = in6_selecthlim(NULL,
1204 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL);
1205 error = ip6_output(m, NULL, ro6, 0, NULL, NULL,
1206 sc->sc_tp->t_inpcb);
1208 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1209 htons(tlen - hlen + IPPROTO_TCP));
1210 m->m_pkthdr.csum_flags = CSUM_TCP;
1211 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1212 error = ip_output(m, sc->sc_ipopts, &sc->sc_route,
1213 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb);
1221 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .|
1223 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .|
1225 * (A): peer mss index
1229 * The values below are chosen to minimize the size of the tcp_secret
1230 * table, as well as providing roughly a 16 second lifetime for the cookie.
1233 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */
1234 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */
1236 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1)
1237 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS)
1238 #define SYNCOOKIE_TIMEOUT \
1239 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT))
1240 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK)
1243 u_int32_t ts_secbits[4];
1245 } tcp_secret[SYNCOOKIE_NSECRETS];
1247 static int tcp_msstab[] = { 0, 536, 1460, 8960 };
1249 static MD5_CTX syn_ctx;
1251 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v))
1254 u_int32_t laddr, faddr;
1255 u_int32_t secbits[4];
1256 u_int16_t lport, fport;
1260 CTASSERT(sizeof(struct md5_add) == 28);
1264 * Consider the problem of a recreated (and retransmitted) cookie. If the
1265 * original SYN was accepted, the connection is established. The second
1266 * SYN is inflight, and if it arrives with an ISN that falls within the
1267 * receive window, the connection is killed.
1269 * However, since cookies have other problems, this may not be worth
1274 syncookie_generate(struct syncache *sc)
1276 u_int32_t md5_buffer[4];
1281 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1283 const boolean_t isipv6 = FALSE;
1286 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK;
1287 if (tcp_secret[idx].ts_expire < ticks) {
1288 for (i = 0; i < 4; i++)
1289 tcp_secret[idx].ts_secbits[i] = karc4random();
1290 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT;
1292 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--)
1293 if (tcp_msstab[data] <= sc->sc_peer_mss)
1295 data = (data << SYNCOOKIE_WNDBITS) | idx;
1296 data ^= sc->sc_irs; /* peer's iss */
1299 MD5Add(sc->sc_inc.inc6_laddr);
1300 MD5Add(sc->sc_inc.inc6_faddr);
1304 add.laddr = sc->sc_inc.inc_laddr.s_addr;
1305 add.faddr = sc->sc_inc.inc_faddr.s_addr;
1307 add.lport = sc->sc_inc.inc_lport;
1308 add.fport = sc->sc_inc.inc_fport;
1309 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1310 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1311 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1312 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1314 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1315 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK);
1319 static struct syncache *
1320 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so)
1322 u_int32_t md5_buffer[4];
1323 struct syncache *sc;
1328 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */
1329 idx = data & SYNCOOKIE_WNDMASK;
1330 if (tcp_secret[idx].ts_expire < ticks ||
1331 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks)
1335 if (inc->inc_isipv6) {
1336 MD5Add(inc->inc6_laddr);
1337 MD5Add(inc->inc6_faddr);
1343 add.laddr = inc->inc_laddr.s_addr;
1344 add.faddr = inc->inc_faddr.s_addr;
1346 add.lport = inc->inc_lport;
1347 add.fport = inc->inc_fport;
1348 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1349 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1350 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1351 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1353 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1354 data ^= md5_buffer[0];
1355 if (data & ~SYNCOOKIE_DATAMASK)
1357 data = data >> SYNCOOKIE_WNDBITS;
1360 * This allocation is guaranteed to succeed because we
1361 * preallocate one more syncache entry than cache_limit.
1363 sc = zalloc(tcp_syncache.zone);
1366 * Fill in the syncache values.
1367 * XXX duplicate code from syncache_add
1369 sc->sc_ipopts = NULL;
1370 sc->sc_inc.inc_fport = inc->inc_fport;
1371 sc->sc_inc.inc_lport = inc->inc_lport;
1373 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1374 if (inc->inc_isipv6) {
1375 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1376 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1377 sc->sc_route6.ro_rt = NULL;
1381 sc->sc_inc.inc_faddr = inc->inc_faddr;
1382 sc->sc_inc.inc_laddr = inc->inc_laddr;
1383 sc->sc_route.ro_rt = NULL;
1385 sc->sc_irs = th->th_seq - 1;
1386 sc->sc_iss = th->th_ack - 1;
1387 wnd = ssb_space(&so->so_rcv);
1389 wnd = imin(wnd, TCP_MAXWIN);
1393 sc->sc_peer_mss = tcp_msstab[data];