IPV6 - Attempt to fix tcp46 compatibility listen sockets
[dragonfly.git] / sys / netinet / tcp_syncache.c
CommitLineData
86e10434 1/*
66d6c637
JH
2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
f23061d4 4 *
66d6c637
JH
5 * This code is derived from software contributed to The DragonFly Project
6 * by Jeffrey M. Hsu.
f23061d4 7 *
66d6c637
JH
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
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.
f23061d4 19 *
66d6c637
JH
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
31 * SUCH DAMAGE.
32 */
33
66d6c637 34/*
86e10434
JH
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.
38 *
984263bc
MD
39 * Copyright (c) 2001 Networks Associates Technologies, Inc.
40 * All rights reserved.
41 *
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.
46 *
47 * Redistribution and use in source and binary forms, with or without
48 * modification, are permitted provided that the following conditions
49 * are met:
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
57 * permission.
58 *
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
69 * SUCH DAMAGE.
70 *
71 * $FreeBSD: src/sys/netinet/tcp_syncache.c,v 1.5.2.14 2003/02/24 04:02:27 silby Exp $
b95665d8 72 * $DragonFly: src/sys/netinet/tcp_syncache.c,v 1.35 2008/11/22 11:03:35 sephe Exp $
984263bc
MD
73 */
74
75#include "opt_inet6.h"
76#include "opt_ipsec.h"
77
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>
83#include <sys/mbuf.h>
84#include <sys/md5.h>
85#include <sys/proc.h> /* for proc0 declaration */
86#include <sys/random.h>
87#include <sys/socket.h>
88#include <sys/socketvar.h>
3f9db7f8 89#include <sys/in_cksum.h>
984263bc 90
00943fd6 91#include <sys/msgport2.h>
4599cf19 92#include <net/netmsg2.h>
00943fd6 93
984263bc
MD
94#include <net/if.h>
95#include <net/route.h>
96
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>
984263bc 103#include <netinet/ip6.h>
61896e3c 104#ifdef INET6
984263bc
MD
105#include <netinet/icmp6.h>
106#include <netinet6/nd6.h>
61896e3c 107#endif
984263bc
MD
108#include <netinet6/ip6_var.h>
109#include <netinet6/in6_pcb.h>
984263bc
MD
110#include <netinet/tcp.h>
111#include <netinet/tcp_fsm.h>
112#include <netinet/tcp_seq.h>
113#include <netinet/tcp_timer.h>
a48c5dd5 114#include <netinet/tcp_timer2.h>
984263bc 115#include <netinet/tcp_var.h>
984263bc 116#include <netinet6/tcp6_var.h>
984263bc
MD
117
118#ifdef IPSEC
119#include <netinet6/ipsec.h>
120#ifdef INET6
121#include <netinet6/ipsec6.h>
122#endif
d2438d69 123#include <netproto/key/key.h>
984263bc
MD
124#endif /*IPSEC*/
125
126#ifdef FAST_IPSEC
bf844ffa 127#include <netproto/ipsec/ipsec.h>
984263bc 128#ifdef INET6
bf844ffa 129#include <netproto/ipsec/ipsec6.h>
984263bc 130#endif
bf844ffa 131#include <netproto/ipsec/key.h>
984263bc
MD
132#define IPSEC
133#endif /*FAST_IPSEC*/
134
984263bc
MD
135#include <vm/vm_zone.h>
136
137static int tcp_syncookies = 1;
138SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
f23061d4 139 &tcp_syncookies, 0,
984263bc
MD
140 "Use TCP SYN cookies if the syncache overflows");
141
142static void syncache_drop(struct syncache *, struct syncache_head *);
143static void syncache_free(struct syncache *);
144static void syncache_insert(struct syncache *, struct syncache_head *);
145struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
146static int syncache_respond(struct syncache *, struct mbuf *);
7e31206a
SZ
147static struct socket *syncache_socket(struct syncache *, struct socket *,
148 struct mbuf *);
984263bc
MD
149static void syncache_timer(void *);
150static u_int32_t syncookie_generate(struct syncache *);
151static struct syncache *syncookie_lookup(struct in_conninfo *,
152 struct tcphdr *, struct socket *);
153
154/*
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.
158 */
159#define SYNCACHE_MAXREXMTS 3
160
161/* Arbitrary values */
162#define TCP_SYNCACHE_HASHSIZE 512
163#define TCP_SYNCACHE_BUCKETLIMIT 30
164
00943fd6 165struct netmsg_sc_timer {
4599cf19 166 struct netmsg nm_netmsg;
00943fd6
JH
167 struct msgrec *nm_mrec; /* back pointer to containing msgrec */
168};
169
170struct msgrec {
171 struct netmsg_sc_timer msg;
172 lwkt_port_t port; /* constant after init */
173 int slot; /* constant after init */
174};
175
4599cf19 176static void syncache_timer_handler(netmsg_t);
00943fd6 177
984263bc 178struct tcp_syncache {
984263bc
MD
179 struct vm_zone *zone;
180 u_int hashsize;
181 u_int hashmask;
182 u_int bucket_limit;
984263bc
MD
183 u_int cache_limit;
184 u_int rexmt_limit;
185 u_int hash_secret;
984263bc
MD
186};
187static struct tcp_syncache tcp_syncache;
188
00943fd6
JH
189struct tcp_syncache_percpu {
190 struct syncache_head *hashbase;
191 u_int cache_count;
192 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1];
193 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1];
194 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1];
195};
196static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU];
197
198static struct lwkt_port syncache_null_rport;
199
984263bc
MD
200SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache");
201
202SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD,
203 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache");
204
205SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD,
206 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache");
207
00943fd6
JH
208/* XXX JH */
209#if 0
984263bc
MD
210SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD,
211 &tcp_syncache.cache_count, 0, "Current number of entries in syncache");
00943fd6 212#endif
984263bc
MD
213
214SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD,
215 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable");
216
217SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW,
218 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions");
219
220static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
221
f23061d4 222#define SYNCACHE_HASH(inc, mask) \
984263bc
MD
223 ((tcp_syncache.hash_secret ^ \
224 (inc)->inc_faddr.s_addr ^ \
f23061d4 225 ((inc)->inc_faddr.s_addr >> 16) ^ \
984263bc
MD
226 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
227
f23061d4 228#define SYNCACHE_HASH6(inc, mask) \
984263bc 229 ((tcp_syncache.hash_secret ^ \
f23061d4
JH
230 (inc)->inc6_faddr.s6_addr32[0] ^ \
231 (inc)->inc6_faddr.s6_addr32[3] ^ \
984263bc
MD
232 (inc)->inc_fport ^ (inc)->inc_lport) & mask)
233
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 \
239)
240
241#define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0)
242
00943fd6
JH
243static __inline void
244syncache_timeout(struct tcp_syncache_percpu *syncache_percpu,
245 struct syncache *sc, int slot)
246{
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],
253 syncache_timer,
254 &syncache_percpu->mrec[slot]);
255 }
256}
984263bc
MD
257
258static void
259syncache_free(struct syncache *sc)
260{
261 struct rtentry *rt;
61896e3c
JH
262#ifdef INET6
263 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
264#else
265 const boolean_t isipv6 = FALSE;
266#endif
984263bc
MD
267
268 if (sc->sc_ipopts)
f23061d4 269 m_free(sc->sc_ipopts);
a5263048
JH
270
271 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt;
984263bc
MD
272 if (rt != NULL) {
273 /*
a5263048 274 * If this is the only reference to a protocol-cloned
984263bc
MD
275 * route, remove it immediately.
276 */
a5263048 277 if ((rt->rt_flags & RTF_WASCLONED) && rt->rt_refcnt == 1)
f23061d4
JH
278 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
279 rt_mask(rt), rt->rt_flags, NULL);
984263bc
MD
280 RTFREE(rt);
281 }
a5263048 282
984263bc
MD
283 zfree(tcp_syncache.zone, sc);
284}
285
286void
287syncache_init(void)
288{
00943fd6 289 int i, cpu;
984263bc 290
984263bc
MD
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;
0ced1954 296 tcp_syncache.hash_secret = karc4random();
984263bc 297
f23061d4 298 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
984263bc 299 &tcp_syncache.hashsize);
f23061d4 300 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
984263bc 301 &tcp_syncache.cache_limit);
f23061d4 302 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
984263bc
MD
303 &tcp_syncache.bucket_limit);
304 if (!powerof2(tcp_syncache.hashsize)) {
a6ec04bc 305 kprintf("WARNING: syncache hash size is not a power of 2.\n");
984263bc 306 tcp_syncache.hashsize = 512; /* safe default */
f23061d4 307 }
984263bc
MD
308 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
309
fb0f29c4 310 lwkt_initport_replyonly_null(&syncache_null_rport);
984263bc 311
00943fd6
JH
312 for (cpu = 0; cpu < ncpus2; cpu++) {
313 struct tcp_syncache_percpu *syncache_percpu;
314
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);
320
321 /* Initialize the hash buckets. */
322 for (i = 0; i < tcp_syncache.hashsize; i++) {
323 struct syncache_head *bucket;
324
325 bucket = &syncache_percpu->hashbase[i];
326 TAILQ_INIT(&bucket->sch_bucket);
327 bucket->sch_length = 0;
328 }
984263bc 329
00943fd6
JH
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]);
334
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];
4599cf19 339 netmsg_init(&syncache_percpu->mrec[i].msg.nm_netmsg,
b95665d8 340 &syncache_null_rport, 0,
4599cf19 341 syncache_timer_handler);
00943fd6 342 }
984263bc
MD
343 }
344
345 /*
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
348 * older one.
349 */
984263bc 350 tcp_syncache.zone = zinit("syncache", sizeof(struct syncache),
7795c2fe 351 tcp_syncache.cache_limit * ncpus2, ZONE_INTERRUPT, 0);
d982be66 352 tcp_syncache.cache_limit -= 1;
984263bc
MD
353}
354
355static void
f3f70f0d 356syncache_insert(struct syncache *sc, struct syncache_head *sch)
984263bc 357{
00943fd6 358 struct tcp_syncache_percpu *syncache_percpu;
984263bc 359 struct syncache *sc2;
d982be66 360 int i;
984263bc 361
00943fd6
JH
362 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
363
984263bc
MD
364 /*
365 * Make sure that we don't overflow the per-bucket
366 * limit or the total cache size limit.
367 */
984263bc
MD
368 if (sch->sch_length >= tcp_syncache.bucket_limit) {
369 /*
370 * The bucket is full, toss the oldest element.
371 */
372 sc2 = TAILQ_FIRST(&sch->sch_bucket);
373 sc2->sc_tp->ts_recent = ticks;
374 syncache_drop(sc2, sch);
375 tcpstat.tcps_sc_bucketoverflow++;
00943fd6 376 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) {
984263bc
MD
377 /*
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
381 * timeout value.
382 */
383 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
00943fd6 384 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]);
984263bc
MD
385 if (sc2 != NULL)
386 break;
387 }
388 sc2->sc_tp->ts_recent = ticks;
389 syncache_drop(sc2, NULL);
390 tcpstat.tcps_sc_cacheoverflow++;
391 }
392
393 /* Initialize the entry's timer. */
00943fd6 394 syncache_timeout(syncache_percpu, sc, 0);
984263bc
MD
395
396 /* Put it into the bucket. */
397 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash);
398 sch->sch_length++;
00943fd6 399 syncache_percpu->cache_count++;
984263bc 400 tcpstat.tcps_sc_added++;
984263bc
MD
401}
402
403static void
f3f70f0d 404syncache_drop(struct syncache *sc, struct syncache_head *sch)
984263bc 405{
00943fd6 406 struct tcp_syncache_percpu *syncache_percpu;
61896e3c
JH
407#ifdef INET6
408 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
409#else
410 const boolean_t isipv6 = FALSE;
411#endif
984263bc 412
00943fd6
JH
413 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
414
984263bc 415 if (sch == NULL) {
61896e3c 416 if (isipv6) {
00943fd6 417 sch = &syncache_percpu->hashbase[
984263bc 418 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)];
61896e3c 419 } else {
00943fd6 420 sch = &syncache_percpu->hashbase[
984263bc
MD
421 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)];
422 }
423 }
424
984263bc
MD
425 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
426 sch->sch_length--;
00943fd6 427 syncache_percpu->cache_count--;
984263bc 428
00943fd6
JH
429 /*
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.
435 */
436 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq);
984263bc
MD
437
438 syncache_free(sc);
439}
440
441/*
00943fd6
JH
442 * Place a timeout message on the TCP thread's message queue.
443 * This routine runs in soft interrupt context.
444 *
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.
448 */
449static void
450syncache_timer(void *p)
451{
452 struct netmsg_sc_timer *msg = p;
453
4599cf19 454 lwkt_sendmsg(msg->nm_mrec->port, &msg->nm_netmsg.nm_lmsg);
00943fd6
JH
455}
456
457/*
458 * Service a timer message queued by timer expiration.
459 * This routine runs in the TCP protocol thread.
460 *
984263bc
MD
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.
00943fd6
JH
463 *
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().
984263bc 467 */
4599cf19
MD
468static void
469syncache_timer_handler(netmsg_t netmsg)
984263bc 470{
00943fd6 471 struct tcp_syncache_percpu *syncache_percpu;
984263bc
MD
472 struct syncache *sc, *nsc;
473 struct inpcb *inp;
00943fd6 474 int slot;
984263bc 475
4599cf19 476 slot = ((struct netmsg_sc_timer *)netmsg)->nm_mrec->slot;
00943fd6 477 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
984263bc 478
f23061d4 479 nsc = TAILQ_FIRST(&syncache_percpu->timerq[slot]);
984263bc
MD
480 while (nsc != NULL) {
481 if (ticks < nsc->sc_rxttime)
00943fd6 482 break; /* finished because timerq sorted by time */
984263bc
MD
483 sc = nsc;
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++;
491 continue;
492 }
493 /*
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.
497 */
f23061d4 498 syncache_respond(sc, NULL);
984263bc
MD
499 nsc = TAILQ_NEXT(sc, sc_timerq);
500 tcpstat.tcps_sc_retransmitted++;
00943fd6
JH
501 TAILQ_REMOVE(&syncache_percpu->timerq[slot], sc, sc_timerq);
502 syncache_timeout(syncache_percpu, sc, slot + 1);
984263bc
MD
503 }
504 if (nsc != NULL)
00943fd6
JH
505 callout_reset(&syncache_percpu->tt_timerq[slot],
506 nsc->sc_rxttime - ticks, syncache_timer,
507 &syncache_percpu->mrec[slot]);
508 else
509 callout_deactivate(&syncache_percpu->tt_timerq[slot]);
510
4599cf19 511 lwkt_replymsg(&netmsg->nm_lmsg, 0);
984263bc
MD
512}
513
514/*
515 * Find an entry in the syncache.
516 */
517struct syncache *
f3f70f0d 518syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
984263bc 519{
00943fd6 520 struct tcp_syncache_percpu *syncache_percpu;
984263bc
MD
521 struct syncache *sc;
522 struct syncache_head *sch;
984263bc 523
00943fd6 524 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
984263bc
MD
525#ifdef INET6
526 if (inc->inc_isipv6) {
00943fd6 527 sch = &syncache_percpu->hashbase[
984263bc
MD
528 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
529 *schp = sch;
d982be66
JH
530 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash)
531 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
984263bc 532 return (sc);
984263bc
MD
533 } else
534#endif
535 {
00943fd6 536 sch = &syncache_percpu->hashbase[
984263bc
MD
537 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
538 *schp = sch;
984263bc
MD
539 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
540#ifdef INET6
541 if (sc->sc_inc.inc_isipv6)
542 continue;
543#endif
d982be66 544 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
984263bc 545 return (sc);
984263bc 546 }
984263bc
MD
547 }
548 return (NULL);
549}
550
551/*
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.
555 */
556void
f3f70f0d 557syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
984263bc
MD
558{
559 struct syncache *sc;
560 struct syncache_head *sch;
561
562 sc = syncache_lookup(inc, &sch);
563 if (sc == NULL)
564 return;
565 /*
566 * If the RST bit is set, check the sequence number to see
567 * if this is a valid reset segment.
568 * RFC 793 page 37:
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.
572 *
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.
577 */
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++;
582 }
583}
584
585void
f3f70f0d 586syncache_badack(struct in_conninfo *inc)
984263bc
MD
587{
588 struct syncache *sc;
589 struct syncache_head *sch;
590
591 sc = syncache_lookup(inc, &sch);
592 if (sc != NULL) {
593 syncache_drop(sc, sch);
594 tcpstat.tcps_sc_badack++;
595 }
596}
597
598void
f3f70f0d 599syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
984263bc
MD
600{
601 struct syncache *sc;
602 struct syncache_head *sch;
603
604 /* we are called at splnet() here */
605 sc = syncache_lookup(inc, &sch);
606 if (sc == NULL)
607 return;
608
609 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
610 if (ntohl(th->th_seq) != sc->sc_iss)
611 return;
612
613 /*
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.
618 *
619 * See tcp_notify().
620 */
621 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) {
622 sc->sc_flags |= SCF_UNREACH;
623 return;
624 }
625 syncache_drop(sc, sch);
626 tcpstat.tcps_sc_unreach++;
627}
628
629/*
630 * Build a new TCP socket structure from a syncache entry.
631 */
632static struct socket *
7e31206a 633syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
984263bc 634{
ed894f8c 635 struct inpcb *inp = NULL, *linp;
984263bc
MD
636 struct socket *so;
637 struct tcpcb *tp;
61896e3c
JH
638#ifdef INET6
639 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
640#else
641 const boolean_t isipv6 = FALSE;
642#endif
984263bc
MD
643
644 /*
645 * Ok, create the full blown connection, and set things up
646 * as they would have been set up if we had created the
647 * connection when the SYN arrived. If we can't create
648 * the connection, abort it.
649 */
650 so = sonewconn(lso, SS_ISCONNECTED);
651 if (so == NULL) {
652 /*
653 * Drop the connection; we will send a RST if the peer
654 * retransmits the ACK,
655 */
656 tcpstat.tcps_listendrop++;
657 goto abort;
658 }
659
ed894f8c 660 inp = so->so_pcb;
984263bc
MD
661
662 /*
663 * Insert new socket into hash list.
664 */
665 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
61896e3c 666 if (isipv6) {
984263bc
MD
667 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
668 } else {
61896e3c 669#ifdef INET6
984263bc
MD
670 inp->inp_vflag &= ~INP_IPV6;
671 inp->inp_vflag |= INP_IPV4;
65f3e756 672 inp->inp_flags &= ~IN6P_IPV6_V6ONLY;
984263bc
MD
673#endif
674 inp->inp_laddr = sc->sc_inc.inc_laddr;
984263bc 675 }
984263bc 676 inp->inp_lport = sc->sc_inc.inc_lport;
13d8907a 677 if (in_pcbinsporthash(inp) != 0) {
984263bc
MD
678 /*
679 * Undo the assignments above if we failed to
680 * put the PCB on the hash lists.
681 */
61896e3c 682 if (isipv6)
84204577 683 inp->in6p_laddr = kin6addr_any;
f23061d4 684 else
984263bc
MD
685 inp->inp_laddr.s_addr = INADDR_ANY;
686 inp->inp_lport = 0;
687 goto abort;
688 }
ed894f8c 689 linp = so->so_pcb;
984263bc
MD
690#ifdef IPSEC
691 /* copy old policy into new socket's */
ed894f8c 692 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp))
a6ec04bc 693 kprintf("syncache_expand: could not copy policy\n");
984263bc 694#endif
61896e3c 695 if (isipv6) {
984263bc 696 struct in6_addr laddr6;
d982be66 697 struct sockaddr_in6 sin6;
984263bc
MD
698 /*
699 * Inherit socket options from the listening socket.
700 * Note that in6p_inputopts are not (and should not be)
701 * copied, since it stores previously received options and is
702 * used to detect if each new option is different than the
703 * previous one and hence should be passed to a user.
f23061d4 704 * If we copied in6p_inputopts, a user would not be able to
984263bc
MD
705 * receive options just after calling the accept system call.
706 */
ed894f8c
JH
707 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS;
708 if (linp->in6p_outputopts)
984263bc 709 inp->in6p_outputopts =
ed894f8c 710 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT);
984263bc
MD
711 inp->in6p_route = sc->sc_route6;
712 sc->sc_route6.ro_rt = NULL;
713
d982be66
JH
714 sin6.sin6_family = AF_INET6;
715 sin6.sin6_len = sizeof sin6;
716 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
717 sin6.sin6_port = sc->sc_inc.inc_fport;
718 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
984263bc
MD
719 laddr6 = inp->in6p_laddr;
720 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
721 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
d982be66 722 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) {
984263bc 723 inp->in6p_laddr = laddr6;
984263bc
MD
724 goto abort;
725 }
61896e3c 726 } else {
984263bc 727 struct in_addr laddr;
d982be66 728 struct sockaddr_in sin;
984263bc 729
7e31206a 730 inp->inp_options = ip_srcroute(m);
984263bc
MD
731 if (inp->inp_options == NULL) {
732 inp->inp_options = sc->sc_ipopts;
733 sc->sc_ipopts = NULL;
734 }
735 inp->inp_route = sc->sc_route;
736 sc->sc_route.ro_rt = NULL;
737
d982be66
JH
738 sin.sin_family = AF_INET;
739 sin.sin_len = sizeof sin;
740 sin.sin_addr = sc->sc_inc.inc_faddr;
741 sin.sin_port = sc->sc_inc.inc_fport;
742 bzero(sin.sin_zero, sizeof sin.sin_zero);
984263bc
MD
743 laddr = inp->inp_laddr;
744 if (inp->inp_laddr.s_addr == INADDR_ANY)
745 inp->inp_laddr = sc->sc_inc.inc_laddr;
d982be66 746 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) {
984263bc 747 inp->inp_laddr = laddr;
984263bc
MD
748 goto abort;
749 }
984263bc
MD
750 }
751
752 tp = intotcpcb(inp);
753 tp->t_state = TCPS_SYN_RECEIVED;
754 tp->iss = sc->sc_iss;
755 tp->irs = sc->sc_irs;
756 tcp_rcvseqinit(tp);
757 tcp_sendseqinit(tp);
758 tp->snd_wl1 = sc->sc_irs;
759 tp->rcv_up = sc->sc_irs + 1;
760 tp->rcv_wnd = sc->sc_wnd;
761 tp->rcv_adv += tp->rcv_wnd;
762
61896e3c 763 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY);
984263bc
MD
764 if (sc->sc_flags & SCF_NOOPT)
765 tp->t_flags |= TF_NOOPT;
766 if (sc->sc_flags & SCF_WINSCALE) {
61896e3c 767 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE;
984263bc
MD
768 tp->requested_s_scale = sc->sc_requested_s_scale;
769 tp->request_r_scale = sc->sc_request_r_scale;
770 }
771 if (sc->sc_flags & SCF_TIMESTAMP) {
61896e3c 772 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP;
984263bc
MD
773 tp->ts_recent = sc->sc_tsrecent;
774 tp->ts_recent_age = ticks;
775 }
776 if (sc->sc_flags & SCF_CC) {
777 /*
778 * Initialization of the tcpcb for transaction;
779 * set SND.WND = SEG.WND,
780 * initialize CCsend and CCrecv.
781 */
61896e3c 782 tp->t_flags |= TF_REQ_CC | TF_RCVD_CC;
984263bc
MD
783 tp->cc_send = sc->sc_cc_send;
784 tp->cc_recv = sc->sc_cc_recv;
785 }
91489f6b
JH
786 if (sc->sc_flags & SCF_SACK_PERMITTED)
787 tp->t_flags |= TF_SACK_PERMITTED;
984263bc
MD
788
789 tcp_mss(tp, sc->sc_peer_mss);
790
791 /*
792 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
793 */
794 if (sc->sc_rxtslot != 0)
f23061d4 795 tp->snd_cwnd = tp->t_maxseg;
3db1c8a3 796 tcp_create_timermsg(tp, &curthread->td_msgport);
a48c5dd5 797 tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep);
984263bc
MD
798
799 tcpstat.tcps_accepts++;
800 return (so);
801
802abort:
803 if (so != NULL)
fd86a41c 804 soabort_oncpu(so);
984263bc
MD
805 return (NULL);
806}
807
808/*
809 * This function gets called when we receive an ACK for a
810 * socket in the LISTEN state. We look up the connection
811 * in the syncache, and if its there, we pull it out of
812 * the cache and turn it into a full-blown connection in
813 * the SYN-RECEIVED state.
814 */
815int
f3f70f0d
SW
816syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop,
817 struct mbuf *m)
984263bc
MD
818{
819 struct syncache *sc;
820 struct syncache_head *sch;
821 struct socket *so;
822
823 sc = syncache_lookup(inc, &sch);
824 if (sc == NULL) {
825 /*
f23061d4 826 * There is no syncache entry, so see if this ACK is
984263bc
MD
827 * a returning syncookie. To do this, first:
828 * A. See if this socket has had a syncache entry dropped in
829 * the past. We don't want to accept a bogus syncookie
f23061d4 830 * if we've never received a SYN.
984263bc
MD
831 * B. check that the syncookie is valid. If it is, then
832 * cobble up a fake syncache entry, and return.
833 */
834 if (!tcp_syncookies)
835 return (0);
836 sc = syncookie_lookup(inc, th, *sop);
837 if (sc == NULL)
838 return (0);
839 sch = NULL;
840 tcpstat.tcps_sc_recvcookie++;
841 }
842
843 /*
844 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
845 */
846 if (th->th_ack != sc->sc_iss + 1)
847 return (0);
848
7e31206a 849 so = syncache_socket(sc, *sop, m);
984263bc
MD
850 if (so == NULL) {
851#if 0
852resetandabort:
853 /* XXXjlemon check this - is this correct? */
f23061d4 854 tcp_respond(NULL, m, m, th,
61896e3c 855 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK);
984263bc
MD
856#endif
857 m_freem(m); /* XXX only needed for above */
858 tcpstat.tcps_sc_aborted++;
859 } else {
984263bc
MD
860 tcpstat.tcps_sc_completed++;
861 }
862 if (sch == NULL)
863 syncache_free(sc);
864 else
865 syncache_drop(sc, sch);
866 *sop = so;
867 return (1);
868}
869
870/*
871 * Given a LISTEN socket and an inbound SYN request, add
872 * this to the syn cache, and send back a segment:
873 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
874 * to the source.
875 *
876 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
877 * Doing so would require that we hold onto the data and deliver it
878 * to the application. However, if we are the target of a SYN-flood
879 * DoS attack, an attacker could send data which would eventually
880 * consume all available buffer space if it were ACKed. By not ACKing
881 * the data, we avoid this DoS scenario.
882 */
883int
f3f70f0d
SW
884syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
885 struct socket **sop, struct mbuf *m)
984263bc 886{
00943fd6 887 struct tcp_syncache_percpu *syncache_percpu;
984263bc
MD
888 struct tcpcb *tp;
889 struct socket *so;
890 struct syncache *sc = NULL;
891 struct syncache_head *sch;
892 struct mbuf *ipopts = NULL;
893 struct rmxp_tao *taop;
913d40d1 894 int win;
984263bc 895
00943fd6 896 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
984263bc
MD
897 so = *sop;
898 tp = sototcpcb(so);
899
900 /*
901 * Remember the IP options, if any.
902 */
903#ifdef INET6
904 if (!inc->inc_isipv6)
905#endif
7e31206a 906 ipopts = ip_srcroute(m);
984263bc
MD
907
908 /*
909 * See if we already have an entry for this connection.
910 * If we do, resend the SYN,ACK, and reset the retransmit timer.
911 *
912 * XXX
91489f6b
JH
913 * The syncache should be re-initialized with the contents
914 * of the new SYN which may have different options.
984263bc
MD
915 */
916 sc = syncache_lookup(inc, &sch);
917 if (sc != NULL) {
918 tcpstat.tcps_sc_dupsyn++;
919 if (ipopts) {
920 /*
921 * If we were remembering a previous source route,
922 * forget it and use the new one we've been given.
923 */
924 if (sc->sc_ipopts)
f23061d4 925 m_free(sc->sc_ipopts);
984263bc
MD
926 sc->sc_ipopts = ipopts;
927 }
928 /*
929 * Update timestamp if present.
930 */
931 if (sc->sc_flags & SCF_TIMESTAMP)
932 sc->sc_tsrecent = to->to_tsval;
91489f6b
JH
933
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;
937 else
938 sc->sc_flags &= ~SCF_SACK_PERMITTED;
939
984263bc
MD
940 /*
941 * PCB may have changed, pick up new values.
942 */
943 sc->sc_tp = tp;
944 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
945 if (syncache_respond(sc, m) == 0) {
00943fd6 946 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot],
984263bc 947 sc, sc_timerq);
00943fd6 948 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot);
f23061d4 949 tcpstat.tcps_sndacks++;
984263bc
MD
950 tcpstat.tcps_sndtotal++;
951 }
952 *sop = NULL;
953 return (1);
954 }
955
d191a96b
JH
956 /*
957 * This allocation is guaranteed to succeed because we
958 * preallocate one more syncache entry than cache_limit.
959 */
984263bc 960 sc = zalloc(tcp_syncache.zone);
984263bc
MD
961
962 /*
963 * Fill in the syncache values.
964 */
984263bc
MD
965 sc->sc_tp = tp;
966 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
967 sc->sc_ipopts = ipopts;
968 sc->sc_inc.inc_fport = inc->inc_fport;
969 sc->sc_inc.inc_lport = inc->inc_lport;
970#ifdef INET6
971 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
972 if (inc->inc_isipv6) {
973 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
974 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
975 sc->sc_route6.ro_rt = NULL;
976 } else
977#endif
978 {
979 sc->sc_inc.inc_faddr = inc->inc_faddr;
980 sc->sc_inc.inc_laddr = inc->inc_laddr;
981 sc->sc_route.ro_rt = NULL;
982 }
983 sc->sc_irs = th->th_seq;
984 sc->sc_flags = 0;
985 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0;
986 if (tcp_syncookies)
987 sc->sc_iss = syncookie_generate(sc);
988 else
0ced1954 989 sc->sc_iss = karc4random();
984263bc 990
6d49aa6f
MD
991 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */
992 win = ssb_space(&so->so_rcv);
984263bc
MD
993 win = imax(win, 0);
994 win = imin(win, TCP_MAXWIN);
995 sc->sc_wnd = win;
996
997 if (tcp_do_rfc1323) {
998 /*
999 * A timestamp received in a SYN makes
1000 * it ok to send timestamp requests and replies.
1001 */
1002 if (to->to_flags & TOF_TS) {
1003 sc->sc_tsrecent = to->to_tsval;
1004 sc->sc_flags |= SCF_TIMESTAMP;
1005 }
1006 if (to->to_flags & TOF_SCALE) {
46e92930 1007 int wscale = TCP_MIN_WINSHIFT;
984263bc
MD
1008
1009 /* Compute proper scaling value from buffer space */
1010 while (wscale < TCP_MAX_WINSHIFT &&
46e92930 1011 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) {
984263bc 1012 wscale++;
46e92930 1013 }
984263bc
MD
1014 sc->sc_request_r_scale = wscale;
1015 sc->sc_requested_s_scale = to->to_requested_s_scale;
1016 sc->sc_flags |= SCF_WINSCALE;
1017 }
1018 }
1019 if (tcp_do_rfc1644) {
1020 /*
1021 * A CC or CC.new option received in a SYN makes
1022 * it ok to send CC in subsequent segments.
1023 */
61896e3c 1024 if (to->to_flags & (TOF_CC | TOF_CCNEW)) {
984263bc
MD
1025 sc->sc_cc_recv = to->to_cc;
1026 sc->sc_cc_send = CC_INC(tcp_ccgen);
1027 sc->sc_flags |= SCF_CC;
1028 }
1029 }
91489f6b
JH
1030 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1031 sc->sc_flags |= SCF_SACK_PERMITTED;
984263bc
MD
1032 if (tp->t_flags & TF_NOOPT)
1033 sc->sc_flags = SCF_NOOPT;
1034
1035 /*
1036 * XXX
1037 * We have the option here of not doing TAO (even if the segment
1038 * qualifies) and instead fall back to a normal 3WHS via the syncache.
1039 * This allows us to apply synflood protection to TAO-qualifying SYNs
1040 * also. However, there should be a hueristic to determine when to
1041 * do this, and is not present at the moment.
1042 */
1043
1044 /*
1045 * Perform TAO test on incoming CC (SEG.CC) option, if any.
1046 * - compare SEG.CC against cached CC from the same host, if any.
1047 * - if SEG.CC > chached value, SYN must be new and is accepted
1048 * immediately: save new CC in the cache, mark the socket
1049 * connected, enter ESTABLISHED state, turn on flag to
1050 * send a SYN in the next segment.
1051 * A virtual advertised window is set in rcv_adv to
1052 * initialize SWS prevention. Then enter normal segment
1053 * processing: drop SYN, process data and FIN.
1054 * - otherwise do a normal 3-way handshake.
1055 */
1056 taop = tcp_gettaocache(&sc->sc_inc);
f23061d4
JH
1057 if (to->to_flags & TOF_CC) {
1058 if ((tp->t_flags & TF_NOPUSH) &&
1059 sc->sc_flags & SCF_CC &&
984263bc
MD
1060 taop != NULL && taop->tao_cc != 0 &&
1061 CC_GT(to->to_cc, taop->tao_cc)) {
1062 sc->sc_rxtslot = 0;
7e31206a 1063 so = syncache_socket(sc, *sop, m);
984263bc 1064 if (so != NULL) {
984263bc
MD
1065 taop->tao_cc = to->to_cc;
1066 *sop = so;
1067 }
1068 syncache_free(sc);
1069 return (so != NULL);
1070 }
1071 } else {
1072 /*
1073 * No CC option, but maybe CC.NEW: invalidate cached value.
1074 */
1075 if (taop != NULL)
1076 taop->tao_cc = 0;
1077 }
1078 /*
1079 * TAO test failed or there was no CC option,
1080 * do a standard 3-way handshake.
1081 */
1082 if (syncache_respond(sc, m) == 0) {
1083 syncache_insert(sc, sch);
1084 tcpstat.tcps_sndacks++;
1085 tcpstat.tcps_sndtotal++;
1086 } else {
1087 syncache_free(sc);
1088 tcpstat.tcps_sc_dropped++;
1089 }
1090 *sop = NULL;
1091 return (1);
1092}
1093
1094static int
f3f70f0d 1095syncache_respond(struct syncache *sc, struct mbuf *m)
984263bc
MD
1096{
1097 u_int8_t *optp;
1098 int optlen, error;
1099 u_int16_t tlen, hlen, mssopt;
1100 struct ip *ip = NULL;
1101 struct rtentry *rt;
1102 struct tcphdr *th;
984263bc 1103 struct ip6_hdr *ip6 = NULL;
61896e3c
JH
1104#ifdef INET6
1105 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1106#else
1107 const boolean_t isipv6 = FALSE;
984263bc
MD
1108#endif
1109
61896e3c 1110 if (isipv6) {
984263bc
MD
1111 rt = tcp_rtlookup6(&sc->sc_inc);
1112 if (rt != NULL)
1113 mssopt = rt->rt_ifp->if_mtu -
1114 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
f23061d4 1115 else
984263bc
MD
1116 mssopt = tcp_v6mssdflt;
1117 hlen = sizeof(struct ip6_hdr);
61896e3c 1118 } else {
984263bc
MD
1119 rt = tcp_rtlookup(&sc->sc_inc);
1120 if (rt != NULL)
1121 mssopt = rt->rt_ifp->if_mtu -
1122 (sizeof(struct ip) + sizeof(struct tcphdr));
f23061d4 1123 else
984263bc
MD
1124 mssopt = tcp_mssdflt;
1125 hlen = sizeof(struct ip);
1126 }
1127
1128 /* Compute the size of the TCP options. */
1129 if (sc->sc_flags & SCF_NOOPT) {
1130 optlen = 0;
1131 } else {
1132 optlen = TCPOLEN_MAXSEG +
1133 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) +
1134 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) +
91489f6b
JH
1135 ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0) +
1136 ((sc->sc_flags & SCF_SACK_PERMITTED) ?
f23061d4 1137 TCPOLEN_SACK_PERMITTED_ALIGNED : 0);
984263bc
MD
1138 }
1139 tlen = hlen + sizeof(struct tcphdr) + optlen;
1140
1141 /*
1142 * XXX
1143 * assume that the entire packet will fit in a header mbuf
1144 */
1145 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small"));
1146
1147 /*
1148 * XXX shouldn't this reuse the mbuf if possible ?
1149 * Create the IP+TCP header from scratch.
1150 */
1151 if (m)
1152 m_freem(m);
1153
74f1caca 1154 m = m_gethdr(MB_DONTWAIT, MT_HEADER);
984263bc
MD
1155 if (m == NULL)
1156 return (ENOBUFS);
1157 m->m_data += max_linkhdr;
1158 m->m_len = tlen;
1159 m->m_pkthdr.len = tlen;
1160 m->m_pkthdr.rcvif = NULL;
1161
61896e3c 1162 if (isipv6) {
984263bc
MD
1163 ip6 = mtod(m, struct ip6_hdr *);
1164 ip6->ip6_vfc = IPV6_VERSION;
1165 ip6->ip6_nxt = IPPROTO_TCP;
1166 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1167 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1168 ip6->ip6_plen = htons(tlen - hlen);
1169 /* ip6_hlim is set after checksum */
1170 /* ip6_flow = ??? */
1171
1172 th = (struct tcphdr *)(ip6 + 1);
61896e3c 1173 } else {
984263bc
MD
1174 ip = mtod(m, struct ip *);
1175 ip->ip_v = IPVERSION;
1176 ip->ip_hl = sizeof(struct ip) >> 2;
1177 ip->ip_len = tlen;
1178 ip->ip_id = 0;
1179 ip->ip_off = 0;
1180 ip->ip_sum = 0;
1181 ip->ip_p = IPPROTO_TCP;
1182 ip->ip_src = sc->sc_inc.inc_laddr;
1183 ip->ip_dst = sc->sc_inc.inc_faddr;
1184 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */
1185 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */
1186
1187 /*
61896e3c
JH
1188 * See if we should do MTU discovery. Route lookups are
1189 * expensive, so we will only unset the DF bit if:
984263bc
MD
1190 *
1191 * 1) path_mtu_discovery is disabled
1192 * 2) the SCF_UNREACH flag has been set
1193 */
1194 if (path_mtu_discovery
1195 && ((sc->sc_flags & SCF_UNREACH) == 0)) {
1196 ip->ip_off |= IP_DF;
1197 }
1198
1199 th = (struct tcphdr *)(ip + 1);
1200 }
1201 th->th_sport = sc->sc_inc.inc_lport;
1202 th->th_dport = sc->sc_inc.inc_fport;
1203
1204 th->th_seq = htonl(sc->sc_iss);
1205 th->th_ack = htonl(sc->sc_irs + 1);
1206 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1207 th->th_x2 = 0;
61896e3c 1208 th->th_flags = TH_SYN | TH_ACK;
984263bc
MD
1209 th->th_win = htons(sc->sc_wnd);
1210 th->th_urp = 0;
1211
1212 /* Tack on the TCP options. */
1213 if (optlen == 0)
1214 goto no_options;
1215 optp = (u_int8_t *)(th + 1);
1216 *optp++ = TCPOPT_MAXSEG;
1217 *optp++ = TCPOLEN_MAXSEG;
1218 *optp++ = (mssopt >> 8) & 0xff;
1219 *optp++ = mssopt & 0xff;
1220
1221 if (sc->sc_flags & SCF_WINSCALE) {
1222 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
1223 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
1224 sc->sc_request_r_scale);
1225 optp += 4;
1226 }
1227
1228 if (sc->sc_flags & SCF_TIMESTAMP) {
1229 u_int32_t *lp = (u_int32_t *)(optp);
1230
1231 /* Form timestamp option as shown in appendix A of RFC 1323. */
1232 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1233 *lp++ = htonl(ticks);
1234 *lp = htonl(sc->sc_tsrecent);
1235 optp += TCPOLEN_TSTAMP_APPA;
1236 }
1237
1238 /*
f23061d4
JH
1239 * Send CC and CC.echo if we received CC from our peer.
1240 */
1241 if (sc->sc_flags & SCF_CC) {
984263bc
MD
1242 u_int32_t *lp = (u_int32_t *)(optp);
1243
1244 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC));
1245 *lp++ = htonl(sc->sc_cc_send);
1246 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO));
1247 *lp = htonl(sc->sc_cc_recv);
1248 optp += TCPOLEN_CC_APPA * 2;
1249 }
984263bc 1250
91489f6b
JH
1251 if (sc->sc_flags & SCF_SACK_PERMITTED) {
1252 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED);
1253 optp += TCPOLEN_SACK_PERMITTED_ALIGNED;
1254 }
1255
61896e3c
JH
1256no_options:
1257 if (isipv6) {
984263bc
MD
1258 struct route_in6 *ro6 = &sc->sc_route6;
1259
1260 th->th_sum = 0;
1261 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
1262 ip6->ip6_hlim = in6_selecthlim(NULL,
1263 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL);
1264 error = ip6_output(m, NULL, ro6, 0, NULL, NULL,
1265 sc->sc_tp->t_inpcb);
61896e3c 1266 } else {
f23061d4
JH
1267 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1268 htons(tlen - hlen + IPPROTO_TCP));
984263bc
MD
1269 m->m_pkthdr.csum_flags = CSUM_TCP;
1270 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1dbb3516
SZ
1271 error = ip_output(m, sc->sc_ipopts, &sc->sc_route,
1272 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb);
984263bc
MD
1273 }
1274 return (error);
1275}
1276
1277/*
1278 * cookie layers:
1279 *
1280 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .|
1281 * | peer iss |
1282 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .|
1283 * | 0 |(A)| |
1284 * (A): peer mss index
1285 */
1286
1287/*
1288 * The values below are chosen to minimize the size of the tcp_secret
1289 * table, as well as providing roughly a 16 second lifetime for the cookie.
1290 */
1291
1292#define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */
1293#define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */
1294
1295#define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1)
1296#define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS)
1297#define SYNCOOKIE_TIMEOUT \
1298 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT))
f23061d4 1299#define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK)
984263bc
MD
1300
1301static struct {
1302 u_int32_t ts_secbits[4];
1303 u_int ts_expire;
1304} tcp_secret[SYNCOOKIE_NSECRETS];
1305
1306static int tcp_msstab[] = { 0, 536, 1460, 8960 };
1307
1308static MD5_CTX syn_ctx;
1309
1310#define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v))
1311
1312struct md5_add {
1313 u_int32_t laddr, faddr;
1314 u_int32_t secbits[4];
1315 u_int16_t lport, fport;
1316};
1317
1318#ifdef CTASSERT
1319CTASSERT(sizeof(struct md5_add) == 28);
1320#endif
1321
1322/*
1323 * Consider the problem of a recreated (and retransmitted) cookie. If the
f23061d4
JH
1324 * original SYN was accepted, the connection is established. The second
1325 * SYN is inflight, and if it arrives with an ISN that falls within the
1326 * receive window, the connection is killed.
984263bc
MD
1327 *
1328 * However, since cookies have other problems, this may not be worth
1329 * worrying about.
1330 */
1331
1332static u_int32_t
1333syncookie_generate(struct syncache *sc)
1334{
1335 u_int32_t md5_buffer[4];
1336 u_int32_t data;
1337 int idx, i;
1338 struct md5_add add;
61896e3c
JH
1339#ifdef INET6
1340 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1341#else
1342 const boolean_t isipv6 = FALSE;
1343#endif
984263bc
MD
1344
1345 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK;
1346 if (tcp_secret[idx].ts_expire < ticks) {
1347 for (i = 0; i < 4; i++)
0ced1954 1348 tcp_secret[idx].ts_secbits[i] = karc4random();
984263bc
MD
1349 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT;
1350 }
1351 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--)
1352 if (tcp_msstab[data] <= sc->sc_peer_mss)
1353 break;
1354 data = (data << SYNCOOKIE_WNDBITS) | idx;
1355 data ^= sc->sc_irs; /* peer's iss */
1356 MD5Init(&syn_ctx);
61896e3c 1357 if (isipv6) {
984263bc
MD
1358 MD5Add(sc->sc_inc.inc6_laddr);
1359 MD5Add(sc->sc_inc.inc6_faddr);
1360 add.laddr = 0;
1361 add.faddr = 0;
61896e3c 1362 } else {
984263bc
MD
1363 add.laddr = sc->sc_inc.inc_laddr.s_addr;
1364 add.faddr = sc->sc_inc.inc_faddr.s_addr;
1365 }
1366 add.lport = sc->sc_inc.inc_lport;
1367 add.fport = sc->sc_inc.inc_fport;
1368 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1369 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1370 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1371 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1372 MD5Add(add);
1373 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1374 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK);
1375 return (data);
1376}
1377
1378static struct syncache *
f3f70f0d 1379syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so)
984263bc
MD
1380{
1381 u_int32_t md5_buffer[4];
1382 struct syncache *sc;
1383 u_int32_t data;
1384 int wnd, idx;
1385 struct md5_add add;
1386
1387 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */
1388 idx = data & SYNCOOKIE_WNDMASK;
1389 if (tcp_secret[idx].ts_expire < ticks ||
1390 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks)
1391 return (NULL);
1392 MD5Init(&syn_ctx);
1393#ifdef INET6
1394 if (inc->inc_isipv6) {
1395 MD5Add(inc->inc6_laddr);
1396 MD5Add(inc->inc6_faddr);
1397 add.laddr = 0;
1398 add.faddr = 0;
1399 } else
1400#endif
1401 {
1402 add.laddr = inc->inc_laddr.s_addr;
1403 add.faddr = inc->inc_faddr.s_addr;
1404 }
1405 add.lport = inc->inc_lport;
1406 add.fport = inc->inc_fport;
1407 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1408 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1409 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1410 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1411 MD5Add(add);
1412 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1413 data ^= md5_buffer[0];
f23061d4 1414 if (data & ~SYNCOOKIE_DATAMASK)
984263bc
MD
1415 return (NULL);
1416 data = data >> SYNCOOKIE_WNDBITS;
1417
d191a96b
JH
1418 /*
1419 * This allocation is guaranteed to succeed because we
1420 * preallocate one more syncache entry than cache_limit.
1421 */
984263bc 1422 sc = zalloc(tcp_syncache.zone);
d191a96b 1423
984263bc
MD
1424 /*
1425 * Fill in the syncache values.
1426 * XXX duplicate code from syncache_add
1427 */
1428 sc->sc_ipopts = NULL;
1429 sc->sc_inc.inc_fport = inc->inc_fport;
1430 sc->sc_inc.inc_lport = inc->inc_lport;
1431#ifdef INET6
1432 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1433 if (inc->inc_isipv6) {
1434 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1435 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1436 sc->sc_route6.ro_rt = NULL;
1437 } else
1438#endif
1439 {
1440 sc->sc_inc.inc_faddr = inc->inc_faddr;
1441 sc->sc_inc.inc_laddr = inc->inc_laddr;
1442 sc->sc_route.ro_rt = NULL;
1443 }
1444 sc->sc_irs = th->th_seq - 1;
1445 sc->sc_iss = th->th_ack - 1;
6d49aa6f 1446 wnd = ssb_space(&so->so_rcv);
984263bc
MD
1447 wnd = imax(wnd, 0);
1448 wnd = imin(wnd, TCP_MAXWIN);
1449 sc->sc_wnd = wnd;
1450 sc->sc_flags = 0;
1451 sc->sc_rxtslot = 0;
1452 sc->sc_peer_mss = tcp_msstab[data];
1453 return (sc);
1454}