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 static int tcp_syncookies = 1;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
138 "Use TCP SYN cookies if the syncache overflows");
140 static void syncache_drop(struct syncache *, struct syncache_head *);
141 static void syncache_free(struct syncache *);
142 static void syncache_insert(struct syncache *, struct syncache_head *);
143 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
144 static int syncache_respond(struct syncache *, struct mbuf *);
145 static struct socket *syncache_socket(struct syncache *, struct socket *,
147 static void syncache_timer(void *);
148 static u_int32_t syncookie_generate(struct syncache *);
149 static struct syncache *syncookie_lookup(struct in_conninfo *,
150 struct tcphdr *, struct socket *);
153 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
154 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds,
155 * the odds are that the user has given up attempting to connect by then.
157 #define SYNCACHE_MAXREXMTS 3
159 /* Arbitrary values */
160 #define TCP_SYNCACHE_HASHSIZE 512
161 #define TCP_SYNCACHE_BUCKETLIMIT 30
163 struct netmsg_sc_timer {
164 struct netmsg nm_netmsg;
165 struct msgrec *nm_mrec; /* back pointer to containing msgrec */
169 struct netmsg_sc_timer msg;
170 lwkt_port_t port; /* constant after init */
171 int slot; /* constant after init */
174 static void syncache_timer_handler(netmsg_t);
176 struct tcp_syncache {
184 static struct tcp_syncache tcp_syncache;
186 struct tcp_syncache_percpu {
187 struct syncache_head *hashbase;
189 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1];
190 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1];
191 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1];
193 static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU];
195 static struct lwkt_port syncache_null_rport;
197 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
199 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD,
200 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
202 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD,
203 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
207 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
208 &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
211 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD,
212 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
214 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
215 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
217 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
219 #define SYNCACHE_HASH(inc, mask) \
220 ((tcp_syncache.hash_secret ^ \
221 (inc)->inc_faddr.s_addr ^ \
222 ((inc)->inc_faddr.s_addr >> 16) ^ \
223 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
225 #define SYNCACHE_HASH6(inc, mask) \
226 ((tcp_syncache.hash_secret ^ \
227 (inc)->inc6_faddr.s6_addr32[0] ^ \
228 (inc)->inc6_faddr.s6_addr32[3] ^ \
229 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
231 #define ENDPTS_EQ(a, b) ( \
232 (a)->ie_fport == (b)->ie_fport && \
233 (a)->ie_lport == (b)->ie_lport && \
234 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \
235 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \
238 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
241 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu,
242 struct syncache *sc, int slot)
244 sc->sc_rxtslot = slot;
245 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[slot];
246 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq);
247 if (!callout_active(&syncache_percpu->tt_timerq[slot])) {
248 callout_reset(&syncache_percpu->tt_timerq[slot],
249 TCPTV_RTOBASE * tcp_backoff[slot],
251 &syncache_percpu->mrec[slot]);
256 syncache_free(struct syncache *sc)
260 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
262 const boolean_t isipv6 = FALSE;
266 m_free(sc->sc_ipopts);
268 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt;
271 * If this is the only reference to a protocol-cloned
272 * route, remove it immediately.
274 if ((rt->rt_flags & RTF_WASCLONED) && rt->rt_refcnt == 1)
275 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
276 rt_mask(rt), rt->rt_flags, NULL);
279 kfree(sc, M_SYNCACHE);
287 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
288 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
289 tcp_syncache.cache_limit =
290 tcp_syncache.hashsize * tcp_syncache.bucket_limit;
291 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
292 tcp_syncache.hash_secret = karc4random();
294 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
295 &tcp_syncache.hashsize);
296 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
297 &tcp_syncache.cache_limit);
298 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
299 &tcp_syncache.bucket_limit);
300 if (!powerof2(tcp_syncache.hashsize)) {
301 kprintf("WARNING: syncache hash size is not a power of 2.\n");
302 tcp_syncache.hashsize = 512; /* safe default */
304 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
306 lwkt_initport_replyonly_null(&syncache_null_rport);
308 for (cpu = 0; cpu < ncpus2; cpu++) {
309 struct tcp_syncache_percpu *syncache_percpu;
311 syncache_percpu = &tcp_syncache_percpu[cpu];
312 /* Allocate the hash table. */
313 MALLOC(syncache_percpu->hashbase, struct syncache_head *,
314 tcp_syncache.hashsize * sizeof(struct syncache_head),
315 M_SYNCACHE, M_WAITOK);
317 /* Initialize the hash buckets. */
318 for (i = 0; i < tcp_syncache.hashsize; i++) {
319 struct syncache_head *bucket;
321 bucket = &syncache_percpu->hashbase[i];
322 TAILQ_INIT(&bucket->sch_bucket);
323 bucket->sch_length = 0;
326 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) {
327 /* Initialize the timer queues. */
328 TAILQ_INIT(&syncache_percpu->timerq[i]);
329 callout_init(&syncache_percpu->tt_timerq[i]);
331 syncache_percpu->mrec[i].slot = i;
332 syncache_percpu->mrec[i].port = tcp_cport(cpu);
333 syncache_percpu->mrec[i].msg.nm_mrec =
334 &syncache_percpu->mrec[i];
335 netmsg_init(&syncache_percpu->mrec[i].msg.nm_netmsg,
336 NULL, &syncache_null_rport,
337 0, syncache_timer_handler);
343 syncache_insert(struct syncache *sc, struct syncache_head *sch)
345 struct tcp_syncache_percpu *syncache_percpu;
346 struct syncache *sc2;
349 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
352 * Make sure that we don't overflow the per-bucket
353 * limit or the total cache size limit.
355 if (sch->sch_length >= tcp_syncache.bucket_limit) {
357 * The bucket is full, toss the oldest element.
359 sc2 = TAILQ_FIRST(&sch->sch_bucket);
360 sc2->sc_tp->ts_recent = ticks;
361 syncache_drop(sc2, sch);
362 tcpstat.tcps_sc_bucketoverflow++;
363 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) {
365 * The cache is full. Toss the oldest entry in the
366 * entire cache. This is the front entry in the
367 * first non-empty timer queue with the largest
370 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
371 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]);
375 sc2->sc_tp->ts_recent = ticks;
376 syncache_drop(sc2, NULL);
377 tcpstat.tcps_sc_cacheoverflow++;
380 /* Initialize the entry's timer. */
381 syncache_timeout(syncache_percpu, sc, 0);
383 /* Put it into the bucket. */
384 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash);
386 syncache_percpu->cache_count++;
387 tcpstat.tcps_sc_added++;
391 syncache_drop(struct syncache *sc, struct syncache_head *sch)
393 struct tcp_syncache_percpu *syncache_percpu;
395 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
397 const boolean_t isipv6 = FALSE;
400 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
404 sch = &syncache_percpu->hashbase[
405 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)];
407 sch = &syncache_percpu->hashbase[
408 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)];
412 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
414 syncache_percpu->cache_count--;
417 * Remove the entry from the syncache timer/timeout queue. Note
418 * that we do not try to stop any running timer since we do not know
419 * whether the timer's message is in-transit or not. Since timeouts
420 * are fairly long, taking an unneeded callout does not detrimentally
421 * effect performance.
423 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq);
429 * Place a timeout message on the TCP thread's message queue.
430 * This routine runs in soft interrupt context.
432 * An invariant is for this routine to be called, the callout must
433 * have been active. Note that the callout is not deactivated until
434 * after the message has been processed in syncache_timer_handler() below.
437 syncache_timer(void *p)
439 struct netmsg_sc_timer *msg = p;
441 lwkt_sendmsg(msg->nm_mrec->port, &msg->nm_netmsg.nm_lmsg);
445 * Service a timer message queued by timer expiration.
446 * This routine runs in the TCP protocol thread.
448 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
449 * If we have retransmitted an entry the maximum number of times, expire it.
451 * When we finish processing timed-out entries, we restart the timer if there
452 * are any entries still on the queue and deactivate it otherwise. Only after
453 * a timer has been deactivated here can it be restarted by syncache_timeout().
456 syncache_timer_handler(netmsg_t netmsg)
458 struct tcp_syncache_percpu *syncache_percpu;
459 struct syncache *sc, *nsc;
463 slot = ((struct netmsg_sc_timer *)netmsg)->nm_mrec->slot;
464 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
466 nsc = TAILQ_FIRST(&syncache_percpu->timerq[slot]);
467 while (nsc != NULL) {
468 if (ticks < nsc->sc_rxttime)
469 break; /* finished because timerq sorted by time */
471 inp = sc->sc_tp->t_inpcb;
472 if (slot == SYNCACHE_MAXREXMTS ||
473 slot >= tcp_syncache.rexmt_limit ||
474 inp->inp_gencnt != sc->sc_inp_gencnt) {
475 nsc = TAILQ_NEXT(sc, sc_timerq);
476 syncache_drop(sc, NULL);
477 tcpstat.tcps_sc_stale++;
481 * syncache_respond() may call back into the syncache to
482 * to modify another entry, so do not obtain the next
483 * entry on the timer chain until it has completed.
485 syncache_respond(sc, NULL);
486 nsc = TAILQ_NEXT(sc, sc_timerq);
487 tcpstat.tcps_sc_retransmitted++;
488 TAILQ_REMOVE(&syncache_percpu->timerq[slot], sc, sc_timerq);
489 syncache_timeout(syncache_percpu, sc, slot + 1);
492 callout_reset(&syncache_percpu->tt_timerq[slot],
493 nsc->sc_rxttime - ticks, syncache_timer,
494 &syncache_percpu->mrec[slot]);
496 callout_deactivate(&syncache_percpu->tt_timerq[slot]);
498 lwkt_replymsg(&netmsg->nm_lmsg, 0);
502 * Find an entry in the syncache.
505 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
507 struct tcp_syncache_percpu *syncache_percpu;
509 struct syncache_head *sch;
511 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
513 if (inc->inc_isipv6) {
514 sch = &syncache_percpu->hashbase[
515 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
517 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash)
518 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
523 sch = &syncache_percpu->hashbase[
524 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
526 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
528 if (sc->sc_inc.inc_isipv6)
531 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
539 * This function is called when we get a RST for a
540 * non-existent connection, so that we can see if the
541 * connection is in the syn cache. If it is, zap it.
544 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
547 struct syncache_head *sch;
549 sc = syncache_lookup(inc, &sch);
553 * If the RST bit is set, check the sequence number to see
554 * if this is a valid reset segment.
556 * In all states except SYN-SENT, all reset (RST) segments
557 * are validated by checking their SEQ-fields. A reset is
558 * valid if its sequence number is in the window.
560 * The sequence number in the reset segment is normally an
561 * echo of our outgoing acknowlegement numbers, but some hosts
562 * send a reset with the sequence number at the rightmost edge
563 * of our receive window, and we have to handle this case.
565 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
566 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
567 syncache_drop(sc, sch);
568 tcpstat.tcps_sc_reset++;
573 syncache_badack(struct in_conninfo *inc)
576 struct syncache_head *sch;
578 sc = syncache_lookup(inc, &sch);
580 syncache_drop(sc, sch);
581 tcpstat.tcps_sc_badack++;
586 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
589 struct syncache_head *sch;
591 /* we are called at splnet() here */
592 sc = syncache_lookup(inc, &sch);
596 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
597 if (ntohl(th->th_seq) != sc->sc_iss)
601 * If we've rertransmitted 3 times and this is our second error,
602 * we remove the entry. Otherwise, we allow it to continue on.
603 * This prevents us from incorrectly nuking an entry during a
604 * spurious network outage.
608 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) {
609 sc->sc_flags |= SCF_UNREACH;
612 syncache_drop(sc, sch);
613 tcpstat.tcps_sc_unreach++;
617 * Build a new TCP socket structure from a syncache entry.
619 * This is called from the context of the SYN+ACK
621 static struct socket *
622 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
624 struct inpcb *inp = NULL, *linp;
629 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
631 const boolean_t isipv6 = FALSE;
635 * Ok, create the full blown connection, and set things up
636 * as they would have been set up if we had created the
637 * connection when the SYN arrived. If we can't create
638 * the connection, abort it.
640 so = sonewconn(lso, SS_ISCONNECTED);
643 * Drop the connection; we will send a RST if the peer
644 * retransmits the ACK,
646 tcpstat.tcps_listendrop++;
651 * Set the protocol processing port for the socket to the current
652 * port (that the connection came in on).
654 sosetport(so, &curthread->td_msgport);
657 * Insert new socket into hash list.
660 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
662 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
665 inp->inp_vflag &= ~INP_IPV6;
666 inp->inp_vflag |= INP_IPV4;
667 inp->inp_flags &= ~IN6P_IPV6_V6ONLY;
669 inp->inp_laddr = sc->sc_inc.inc_laddr;
671 inp->inp_lport = sc->sc_inc.inc_lport;
672 if (in_pcbinsporthash(inp) != 0) {
674 * Undo the assignments above if we failed to
675 * put the PCB on the hash lists.
678 inp->in6p_laddr = kin6addr_any;
680 inp->inp_laddr.s_addr = INADDR_ANY;
686 /* copy old policy into new socket's */
687 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp))
688 kprintf("syncache_expand: could not copy policy\n");
691 struct in6_addr laddr6;
692 struct sockaddr_in6 sin6;
694 * Inherit socket options from the listening socket.
695 * Note that in6p_inputopts are not (and should not be)
696 * copied, since it stores previously received options and is
697 * used to detect if each new option is different than the
698 * previous one and hence should be passed to a user.
699 * If we copied in6p_inputopts, a user would not be able to
700 * receive options just after calling the accept system call.
702 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS;
703 if (linp->in6p_outputopts)
704 inp->in6p_outputopts =
705 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT);
706 inp->in6p_route = sc->sc_route6;
707 sc->sc_route6.ro_rt = NULL;
709 sin6.sin6_family = AF_INET6;
710 sin6.sin6_len = sizeof sin6;
711 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
712 sin6.sin6_port = sc->sc_inc.inc_fport;
713 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
714 laddr6 = inp->in6p_laddr;
715 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
716 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
717 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) {
718 inp->in6p_laddr = laddr6;
722 struct in_addr laddr;
723 struct sockaddr_in sin;
725 inp->inp_options = ip_srcroute(m);
726 if (inp->inp_options == NULL) {
727 inp->inp_options = sc->sc_ipopts;
728 sc->sc_ipopts = NULL;
730 inp->inp_route = sc->sc_route;
731 sc->sc_route.ro_rt = NULL;
733 sin.sin_family = AF_INET;
734 sin.sin_len = sizeof sin;
735 sin.sin_addr = sc->sc_inc.inc_faddr;
736 sin.sin_port = sc->sc_inc.inc_fport;
737 bzero(sin.sin_zero, sizeof sin.sin_zero);
738 laddr = inp->inp_laddr;
739 if (inp->inp_laddr.s_addr == INADDR_ANY)
740 inp->inp_laddr = sc->sc_inc.inc_laddr;
741 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) {
742 inp->inp_laddr = laddr;
748 * The current port should be in the context of the SYN+ACK and
749 * so should match the tcp address port.
751 * XXX we may be running on the netisr thread instead of a tcp
752 * thread, in which case port will not match
753 * curthread->td_msgport.
756 port = tcp6_addrport();
758 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport,
759 inp->inp_laddr.s_addr, inp->inp_lport);
761 /*KKASSERT(port == &curthread->td_msgport);*/
764 tp->t_state = TCPS_SYN_RECEIVED;
765 tp->iss = sc->sc_iss;
766 tp->irs = sc->sc_irs;
769 tp->snd_wl1 = sc->sc_irs;
770 tp->rcv_up = sc->sc_irs + 1;
771 tp->rcv_wnd = sc->sc_wnd;
772 tp->rcv_adv += tp->rcv_wnd;
774 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY);
775 if (sc->sc_flags & SCF_NOOPT)
776 tp->t_flags |= TF_NOOPT;
777 if (sc->sc_flags & SCF_WINSCALE) {
778 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE;
779 tp->requested_s_scale = sc->sc_requested_s_scale;
780 tp->request_r_scale = sc->sc_request_r_scale;
782 if (sc->sc_flags & SCF_TIMESTAMP) {
783 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP;
784 tp->ts_recent = sc->sc_tsrecent;
785 tp->ts_recent_age = ticks;
787 if (sc->sc_flags & SCF_SACK_PERMITTED)
788 tp->t_flags |= TF_SACK_PERMITTED;
790 tcp_mss(tp, sc->sc_peer_mss);
793 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
795 if (sc->sc_rxtslot != 0)
796 tp->snd_cwnd = tp->t_maxseg;
797 tcp_create_timermsg(tp, port);
798 tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep);
800 tcpstat.tcps_accepts++;
810 * This function gets called when we receive an ACK for a
811 * socket in the LISTEN state. We look up the connection
812 * in the syncache, and if its there, we pull it out of
813 * the cache and turn it into a full-blown connection in
814 * the SYN-RECEIVED state.
817 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop,
821 struct syncache_head *sch;
824 sc = syncache_lookup(inc, &sch);
827 * There is no syncache entry, so see if this ACK is
828 * a returning syncookie. To do this, first:
829 * A. See if this socket has had a syncache entry dropped in
830 * the past. We don't want to accept a bogus syncookie
831 * if we've never received a SYN.
832 * B. check that the syncookie is valid. If it is, then
833 * cobble up a fake syncache entry, and return.
837 sc = syncookie_lookup(inc, th, *sop);
841 tcpstat.tcps_sc_recvcookie++;
845 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
847 if (th->th_ack != sc->sc_iss + 1)
850 so = syncache_socket(sc, *sop, m);
854 /* XXXjlemon check this - is this correct? */
855 tcp_respond(NULL, m, m, th,
856 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK);
858 m_freem(m); /* XXX only needed for above */
859 tcpstat.tcps_sc_aborted++;
861 tcpstat.tcps_sc_completed++;
866 syncache_drop(sc, sch);
872 * Given a LISTEN socket and an inbound SYN request, add
873 * this to the syn cache, and send back a segment:
874 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
877 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
878 * Doing so would require that we hold onto the data and deliver it
879 * to the application. However, if we are the target of a SYN-flood
880 * DoS attack, an attacker could send data which would eventually
881 * consume all available buffer space if it were ACKed. By not ACKing
882 * the data, we avoid this DoS scenario.
885 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
886 struct socket **sop, struct mbuf *m)
888 struct tcp_syncache_percpu *syncache_percpu;
891 struct syncache *sc = NULL;
892 struct syncache_head *sch;
893 struct mbuf *ipopts = NULL;
896 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
901 * Remember the IP options, if any.
904 if (!inc->inc_isipv6)
906 ipopts = ip_srcroute(m);
909 * See if we already have an entry for this connection.
910 * If we do, resend the SYN,ACK, and reset the retransmit timer.
913 * The syncache should be re-initialized with the contents
914 * of the new SYN which may have different options.
916 sc = syncache_lookup(inc, &sch);
918 tcpstat.tcps_sc_dupsyn++;
921 * If we were remembering a previous source route,
922 * forget it and use the new one we've been given.
925 m_free(sc->sc_ipopts);
926 sc->sc_ipopts = ipopts;
929 * Update timestamp if present.
931 if (sc->sc_flags & SCF_TIMESTAMP)
932 sc->sc_tsrecent = to->to_tsval;
934 /* Just update the TOF_SACK_PERMITTED for now. */
935 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
936 sc->sc_flags |= SCF_SACK_PERMITTED;
938 sc->sc_flags &= ~SCF_SACK_PERMITTED;
941 * PCB may have changed, pick up new values.
944 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
945 if (syncache_respond(sc, m) == 0) {
946 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot],
948 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot);
949 tcpstat.tcps_sndacks++;
950 tcpstat.tcps_sndtotal++;
957 * Fill in the syncache values.
959 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO);
961 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
962 sc->sc_ipopts = ipopts;
963 sc->sc_inc.inc_fport = inc->inc_fport;
964 sc->sc_inc.inc_lport = inc->inc_lport;
966 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
967 if (inc->inc_isipv6) {
968 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
969 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
970 sc->sc_route6.ro_rt = NULL;
974 sc->sc_inc.inc_faddr = inc->inc_faddr;
975 sc->sc_inc.inc_laddr = inc->inc_laddr;
976 sc->sc_route.ro_rt = NULL;
978 sc->sc_irs = th->th_seq;
980 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0;
982 sc->sc_iss = syncookie_generate(sc);
984 sc->sc_iss = karc4random();
986 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */
987 win = ssb_space(&so->so_rcv);
989 win = imin(win, TCP_MAXWIN);
992 if (tcp_do_rfc1323) {
994 * A timestamp received in a SYN makes
995 * it ok to send timestamp requests and replies.
997 if (to->to_flags & TOF_TS) {
998 sc->sc_tsrecent = to->to_tsval;
999 sc->sc_flags |= SCF_TIMESTAMP;
1001 if (to->to_flags & TOF_SCALE) {
1002 int wscale = TCP_MIN_WINSHIFT;
1004 /* Compute proper scaling value from buffer space */
1005 while (wscale < TCP_MAX_WINSHIFT &&
1006 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) {
1009 sc->sc_request_r_scale = wscale;
1010 sc->sc_requested_s_scale = to->to_requested_s_scale;
1011 sc->sc_flags |= SCF_WINSCALE;
1014 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1015 sc->sc_flags |= SCF_SACK_PERMITTED;
1016 if (tp->t_flags & TF_NOOPT)
1017 sc->sc_flags = SCF_NOOPT;
1019 if (syncache_respond(sc, m) == 0) {
1020 syncache_insert(sc, sch);
1021 tcpstat.tcps_sndacks++;
1022 tcpstat.tcps_sndtotal++;
1025 tcpstat.tcps_sc_dropped++;
1032 syncache_respond(struct syncache *sc, struct mbuf *m)
1036 u_int16_t tlen, hlen, mssopt;
1037 struct ip *ip = NULL;
1040 struct ip6_hdr *ip6 = NULL;
1042 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1044 const boolean_t isipv6 = FALSE;
1048 rt = tcp_rtlookup6(&sc->sc_inc);
1050 mssopt = rt->rt_ifp->if_mtu -
1051 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1053 mssopt = tcp_v6mssdflt;
1054 hlen = sizeof(struct ip6_hdr);
1056 rt = tcp_rtlookup(&sc->sc_inc);
1058 mssopt = rt->rt_ifp->if_mtu -
1059 (sizeof(struct ip) + sizeof(struct tcphdr));
1061 mssopt = tcp_mssdflt;
1062 hlen = sizeof(struct ip);
1065 /* Compute the size of the TCP options. */
1066 if (sc->sc_flags & SCF_NOOPT) {
1069 optlen = TCPOLEN_MAXSEG +
1070 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) +
1071 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) +
1072 ((sc->sc_flags & SCF_SACK_PERMITTED) ?
1073 TCPOLEN_SACK_PERMITTED_ALIGNED : 0);
1075 tlen = hlen + sizeof(struct tcphdr) + optlen;
1079 * assume that the entire packet will fit in a header mbuf
1081 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small"));
1084 * XXX shouldn't this reuse the mbuf if possible ?
1085 * Create the IP+TCP header from scratch.
1090 m = m_gethdr(MB_DONTWAIT, MT_HEADER);
1093 m->m_data += max_linkhdr;
1095 m->m_pkthdr.len = tlen;
1096 m->m_pkthdr.rcvif = NULL;
1099 ip6 = mtod(m, struct ip6_hdr *);
1100 ip6->ip6_vfc = IPV6_VERSION;
1101 ip6->ip6_nxt = IPPROTO_TCP;
1102 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1103 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1104 ip6->ip6_plen = htons(tlen - hlen);
1105 /* ip6_hlim is set after checksum */
1106 /* ip6_flow = ??? */
1108 th = (struct tcphdr *)(ip6 + 1);
1110 ip = mtod(m, struct ip *);
1111 ip->ip_v = IPVERSION;
1112 ip->ip_hl = sizeof(struct ip) >> 2;
1117 ip->ip_p = IPPROTO_TCP;
1118 ip->ip_src = sc->sc_inc.inc_laddr;
1119 ip->ip_dst = sc->sc_inc.inc_faddr;
1120 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */
1121 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */
1124 * See if we should do MTU discovery. Route lookups are
1125 * expensive, so we will only unset the DF bit if:
1127 * 1) path_mtu_discovery is disabled
1128 * 2) the SCF_UNREACH flag has been set
1130 if (path_mtu_discovery
1131 && ((sc->sc_flags & SCF_UNREACH) == 0)) {
1132 ip->ip_off |= IP_DF;
1135 th = (struct tcphdr *)(ip + 1);
1137 th->th_sport = sc->sc_inc.inc_lport;
1138 th->th_dport = sc->sc_inc.inc_fport;
1140 th->th_seq = htonl(sc->sc_iss);
1141 th->th_ack = htonl(sc->sc_irs + 1);
1142 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1144 th->th_flags = TH_SYN | TH_ACK;
1145 th->th_win = htons(sc->sc_wnd);
1148 /* Tack on the TCP options. */
1151 optp = (u_int8_t *)(th + 1);
1152 *optp++ = TCPOPT_MAXSEG;
1153 *optp++ = TCPOLEN_MAXSEG;
1154 *optp++ = (mssopt >> 8) & 0xff;
1155 *optp++ = mssopt & 0xff;
1157 if (sc->sc_flags & SCF_WINSCALE) {
1158 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
1159 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
1160 sc->sc_request_r_scale);
1164 if (sc->sc_flags & SCF_TIMESTAMP) {
1165 u_int32_t *lp = (u_int32_t *)(optp);
1167 /* Form timestamp option as shown in appendix A of RFC 1323. */
1168 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1169 *lp++ = htonl(ticks);
1170 *lp = htonl(sc->sc_tsrecent);
1171 optp += TCPOLEN_TSTAMP_APPA;
1174 if (sc->sc_flags & SCF_SACK_PERMITTED) {
1175 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED);
1176 optp += TCPOLEN_SACK_PERMITTED_ALIGNED;
1181 struct route_in6 *ro6 = &sc->sc_route6;
1184 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
1185 ip6->ip6_hlim = in6_selecthlim(NULL,
1186 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL);
1187 error = ip6_output(m, NULL, ro6, 0, NULL, NULL,
1188 sc->sc_tp->t_inpcb);
1190 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1191 htons(tlen - hlen + IPPROTO_TCP));
1192 m->m_pkthdr.csum_flags = CSUM_TCP;
1193 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1194 error = ip_output(m, sc->sc_ipopts, &sc->sc_route,
1195 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb);
1203 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .|
1205 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .|
1207 * (A): peer mss index
1211 * The values below are chosen to minimize the size of the tcp_secret
1212 * table, as well as providing roughly a 16 second lifetime for the cookie.
1215 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */
1216 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */
1218 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1)
1219 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS)
1220 #define SYNCOOKIE_TIMEOUT \
1221 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT))
1222 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK)
1225 u_int32_t ts_secbits[4];
1227 } tcp_secret[SYNCOOKIE_NSECRETS];
1229 static int tcp_msstab[] = { 0, 536, 1460, 8960 };
1231 static MD5_CTX syn_ctx;
1233 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v))
1236 u_int32_t laddr, faddr;
1237 u_int32_t secbits[4];
1238 u_int16_t lport, fport;
1242 CTASSERT(sizeof(struct md5_add) == 28);
1246 * Consider the problem of a recreated (and retransmitted) cookie. If the
1247 * original SYN was accepted, the connection is established. The second
1248 * SYN is inflight, and if it arrives with an ISN that falls within the
1249 * receive window, the connection is killed.
1251 * However, since cookies have other problems, this may not be worth
1256 syncookie_generate(struct syncache *sc)
1258 u_int32_t md5_buffer[4];
1263 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1265 const boolean_t isipv6 = FALSE;
1268 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK;
1269 if (tcp_secret[idx].ts_expire < ticks) {
1270 for (i = 0; i < 4; i++)
1271 tcp_secret[idx].ts_secbits[i] = karc4random();
1272 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT;
1274 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--)
1275 if (tcp_msstab[data] <= sc->sc_peer_mss)
1277 data = (data << SYNCOOKIE_WNDBITS) | idx;
1278 data ^= sc->sc_irs; /* peer's iss */
1281 MD5Add(sc->sc_inc.inc6_laddr);
1282 MD5Add(sc->sc_inc.inc6_faddr);
1286 add.laddr = sc->sc_inc.inc_laddr.s_addr;
1287 add.faddr = sc->sc_inc.inc_faddr.s_addr;
1289 add.lport = sc->sc_inc.inc_lport;
1290 add.fport = sc->sc_inc.inc_fport;
1291 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1292 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1293 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1294 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1296 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1297 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK);
1301 static struct syncache *
1302 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so)
1304 u_int32_t md5_buffer[4];
1305 struct syncache *sc;
1310 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */
1311 idx = data & SYNCOOKIE_WNDMASK;
1312 if (tcp_secret[idx].ts_expire < ticks ||
1313 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks)
1317 if (inc->inc_isipv6) {
1318 MD5Add(inc->inc6_laddr);
1319 MD5Add(inc->inc6_faddr);
1325 add.laddr = inc->inc_laddr.s_addr;
1326 add.faddr = inc->inc_faddr.s_addr;
1328 add.lport = inc->inc_lport;
1329 add.fport = inc->inc_fport;
1330 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1331 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1332 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1333 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1335 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1336 data ^= md5_buffer[0];
1337 if (data & ~SYNCOOKIE_DATAMASK)
1339 data = data >> SYNCOOKIE_WNDBITS;
1342 * Fill in the syncache values.
1343 * XXX duplicate code from syncache_add
1345 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO);
1346 sc->sc_ipopts = NULL;
1347 sc->sc_inc.inc_fport = inc->inc_fport;
1348 sc->sc_inc.inc_lport = inc->inc_lport;
1350 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1351 if (inc->inc_isipv6) {
1352 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1353 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1354 sc->sc_route6.ro_rt = NULL;
1358 sc->sc_inc.inc_faddr = inc->inc_faddr;
1359 sc->sc_inc.inc_laddr = inc->inc_laddr;
1360 sc->sc_route.ro_rt = NULL;
1362 sc->sc_irs = th->th_seq - 1;
1363 sc->sc_iss = th->th_ack - 1;
1364 wnd = ssb_space(&so->so_rcv);
1366 wnd = imin(wnd, TCP_MAXWIN);
1370 sc->sc_peer_mss = tcp_msstab[data];