mbuf: Save linker layer, IP and TCP/UDP header length
[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 $
72 */
73
b1992928 74#include "opt_inet.h"
984263bc
MD
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
135static int tcp_syncookies = 1;
136SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
f23061d4 137 &tcp_syncookies, 0,
984263bc
MD
138 "Use TCP SYN cookies if the syncache overflows");
139
140static void syncache_drop(struct syncache *, struct syncache_head *);
141static void syncache_free(struct syncache *);
142static void syncache_insert(struct syncache *, struct syncache_head *);
143struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
144static int syncache_respond(struct syncache *, struct mbuf *);
7e31206a
SZ
145static struct socket *syncache_socket(struct syncache *, struct socket *,
146 struct mbuf *);
984263bc
MD
147static void syncache_timer(void *);
148static u_int32_t syncookie_generate(struct syncache *);
149static struct syncache *syncookie_lookup(struct in_conninfo *,
150 struct tcphdr *, struct socket *);
151
152/*
153 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
60536dc2
SZ
154 * 4 retransmits corresponds to a timeout of (3 + 3 + 3 + 3 + 3 == 15) seconds
155 * or (1 + 1 + 2 + 4 + 8 == 16) seconds if RFC6298 is used, the odds are that
156 * the user has given up attempting to connect by then.
984263bc 157 */
60536dc2 158#define SYNCACHE_MAXREXMTS 4
984263bc
MD
159
160/* Arbitrary values */
161#define TCP_SYNCACHE_HASHSIZE 512
162#define TCP_SYNCACHE_BUCKETLIMIT 30
163
00943fd6 164struct netmsg_sc_timer {
002c1265 165 struct netmsg_base base;
00943fd6
JH
166 struct msgrec *nm_mrec; /* back pointer to containing msgrec */
167};
168
169struct msgrec {
170 struct netmsg_sc_timer msg;
171 lwkt_port_t port; /* constant after init */
172 int slot; /* constant after init */
173};
174
4599cf19 175static void syncache_timer_handler(netmsg_t);
00943fd6 176
984263bc 177struct tcp_syncache {
984263bc
MD
178 u_int hashsize;
179 u_int hashmask;
180 u_int bucket_limit;
984263bc
MD
181 u_int cache_limit;
182 u_int rexmt_limit;
183 u_int hash_secret;
984263bc
MD
184};
185static struct tcp_syncache tcp_syncache;
186
c1d0893d
MD
187TAILQ_HEAD(syncache_list, syncache);
188
00943fd6
JH
189struct tcp_syncache_percpu {
190 struct syncache_head *hashbase;
191 u_int cache_count;
c1d0893d 192 struct syncache_list timerq[SYNCACHE_MAXREXMTS + 1];
00943fd6
JH
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
d5082e3d
SZ
243static __inline int
244syncache_rto(int slot)
245{
246 if (tcp_low_rtobase)
247 return (TCPTV_RTOBASE * tcp_syn_backoff_low[slot]);
248 else
249 return (TCPTV_RTOBASE * tcp_syn_backoff[slot]);
250}
251
00943fd6
JH
252static __inline void
253syncache_timeout(struct tcp_syncache_percpu *syncache_percpu,
254 struct syncache *sc, int slot)
255{
9d173e54
SZ
256 int rto;
257
be34e534
SZ
258 if (slot > 0) {
259 /*
d5082e3d
SZ
260 * Record the time that we spent in SYN|ACK
261 * retransmition.
262 *
be34e534
SZ
263 * Needed by RFC3390 and RFC6298.
264 */
d5082e3d 265 sc->sc_rxtused += syncache_rto(slot - 1);
be34e534 266 }
00943fd6 267 sc->sc_rxtslot = slot;
9d173e54 268
d5082e3d 269 rto = syncache_rto(slot);
9d173e54
SZ
270 sc->sc_rxttime = ticks + rto;
271
00943fd6
JH
272 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq);
273 if (!callout_active(&syncache_percpu->tt_timerq[slot])) {
d60606dd
SZ
274 callout_reset(&syncache_percpu->tt_timerq[slot], rto,
275 syncache_timer, &syncache_percpu->mrec[slot]);
00943fd6
JH
276 }
277}
984263bc
MD
278
279static void
280syncache_free(struct syncache *sc)
281{
282 struct rtentry *rt;
61896e3c
JH
283#ifdef INET6
284 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
285#else
286 const boolean_t isipv6 = FALSE;
287#endif
984263bc
MD
288
289 if (sc->sc_ipopts)
f23061d4 290 m_free(sc->sc_ipopts);
a5263048
JH
291
292 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt;
984263bc
MD
293 if (rt != NULL) {
294 /*
a5263048 295 * If this is the only reference to a protocol-cloned
984263bc
MD
296 * route, remove it immediately.
297 */
a5263048 298 if ((rt->rt_flags & RTF_WASCLONED) && rt->rt_refcnt == 1)
f23061d4
JH
299 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
300 rt_mask(rt), rt->rt_flags, NULL);
984263bc
MD
301 RTFREE(rt);
302 }
9f42c129 303 kfree(sc, M_SYNCACHE);
984263bc
MD
304}
305
306void
307syncache_init(void)
308{
00943fd6 309 int i, cpu;
984263bc 310
984263bc
MD
311 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
312 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
313 tcp_syncache.cache_limit =
314 tcp_syncache.hashsize * tcp_syncache.bucket_limit;
315 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
0ced1954 316 tcp_syncache.hash_secret = karc4random();
984263bc 317
f23061d4 318 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
984263bc 319 &tcp_syncache.hashsize);
f23061d4 320 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
984263bc 321 &tcp_syncache.cache_limit);
f23061d4 322 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
984263bc
MD
323 &tcp_syncache.bucket_limit);
324 if (!powerof2(tcp_syncache.hashsize)) {
a6ec04bc 325 kprintf("WARNING: syncache hash size is not a power of 2.\n");
984263bc 326 tcp_syncache.hashsize = 512; /* safe default */
f23061d4 327 }
984263bc
MD
328 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
329
fb0f29c4 330 lwkt_initport_replyonly_null(&syncache_null_rport);
984263bc 331
00943fd6
JH
332 for (cpu = 0; cpu < ncpus2; cpu++) {
333 struct tcp_syncache_percpu *syncache_percpu;
334
335 syncache_percpu = &tcp_syncache_percpu[cpu];
336 /* Allocate the hash table. */
884717e1
SW
337 syncache_percpu->hashbase = kmalloc(tcp_syncache.hashsize * sizeof(struct syncache_head),
338 M_SYNCACHE, M_WAITOK);
00943fd6
JH
339
340 /* Initialize the hash buckets. */
341 for (i = 0; i < tcp_syncache.hashsize; i++) {
342 struct syncache_head *bucket;
343
344 bucket = &syncache_percpu->hashbase[i];
345 TAILQ_INIT(&bucket->sch_bucket);
346 bucket->sch_length = 0;
347 }
984263bc 348
00943fd6
JH
349 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) {
350 /* Initialize the timer queues. */
351 TAILQ_INIT(&syncache_percpu->timerq[i]);
608499ad 352 callout_init_mp(&syncache_percpu->tt_timerq[i]);
00943fd6
JH
353
354 syncache_percpu->mrec[i].slot = i;
c3c96e44 355 syncache_percpu->mrec[i].port = cpu_portfn(cpu);
00943fd6 356 syncache_percpu->mrec[i].msg.nm_mrec =
c3c96e44 357 &syncache_percpu->mrec[i];
002c1265 358 netmsg_init(&syncache_percpu->mrec[i].msg.base,
48e7b118
MD
359 NULL, &syncache_null_rport,
360 0, syncache_timer_handler);
00943fd6 361 }
984263bc 362 }
984263bc
MD
363}
364
365static void
f3f70f0d 366syncache_insert(struct syncache *sc, struct syncache_head *sch)
984263bc 367{
00943fd6 368 struct tcp_syncache_percpu *syncache_percpu;
984263bc 369 struct syncache *sc2;
d982be66 370 int i;
984263bc 371
00943fd6
JH
372 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
373
984263bc
MD
374 /*
375 * Make sure that we don't overflow the per-bucket
376 * limit or the total cache size limit.
377 */
984263bc
MD
378 if (sch->sch_length >= tcp_syncache.bucket_limit) {
379 /*
380 * The bucket is full, toss the oldest element.
381 */
382 sc2 = TAILQ_FIRST(&sch->sch_bucket);
383 sc2->sc_tp->ts_recent = ticks;
384 syncache_drop(sc2, sch);
385 tcpstat.tcps_sc_bucketoverflow++;
00943fd6 386 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) {
984263bc
MD
387 /*
388 * The cache is full. Toss the oldest entry in the
389 * entire cache. This is the front entry in the
390 * first non-empty timer queue with the largest
391 * timeout value.
392 */
393 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
00943fd6 394 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]);
c1d0893d
MD
395 while (sc2 && (sc2->sc_flags & SCF_MARKER))
396 sc2 = TAILQ_NEXT(sc2, sc_timerq);
984263bc
MD
397 if (sc2 != NULL)
398 break;
399 }
400 sc2->sc_tp->ts_recent = ticks;
401 syncache_drop(sc2, NULL);
402 tcpstat.tcps_sc_cacheoverflow++;
403 }
404
405 /* Initialize the entry's timer. */
00943fd6 406 syncache_timeout(syncache_percpu, sc, 0);
984263bc
MD
407
408 /* Put it into the bucket. */
409 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash);
410 sch->sch_length++;
00943fd6 411 syncache_percpu->cache_count++;
984263bc 412 tcpstat.tcps_sc_added++;
984263bc
MD
413}
414
e5fe3477
MD
415void
416syncache_destroy(struct tcpcb *tp)
417{
418 struct tcp_syncache_percpu *syncache_percpu;
419 struct syncache_head *bucket;
420 struct syncache *sc;
421 int i;
422
423 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
424 sc = NULL;
c1d0893d 425
e5fe3477
MD
426 for (i = 0; i < tcp_syncache.hashsize; i++) {
427 bucket = &syncache_percpu->hashbase[i];
428 TAILQ_FOREACH(sc, &bucket->sch_bucket, sc_hash) {
7123bbff 429 if (sc->sc_tp == tp)
e5fe3477 430 sc->sc_tp = NULL;
e5fe3477
MD
431 }
432 }
e5fe3477
MD
433}
434
984263bc 435static void
f3f70f0d 436syncache_drop(struct syncache *sc, struct syncache_head *sch)
984263bc 437{
00943fd6 438 struct tcp_syncache_percpu *syncache_percpu;
61896e3c
JH
439#ifdef INET6
440 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
441#else
442 const boolean_t isipv6 = FALSE;
443#endif
984263bc 444
00943fd6
JH
445 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
446
984263bc 447 if (sch == NULL) {
61896e3c 448 if (isipv6) {
00943fd6 449 sch = &syncache_percpu->hashbase[
984263bc 450 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)];
61896e3c 451 } else {
00943fd6 452 sch = &syncache_percpu->hashbase[
984263bc
MD
453 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)];
454 }
455 }
456
984263bc
MD
457 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
458 sch->sch_length--;
00943fd6 459 syncache_percpu->cache_count--;
984263bc 460
e5fe3477
MD
461 /*
462 * Cleanup
463 */
7123bbff 464 if (sc->sc_tp)
e5fe3477 465 sc->sc_tp = NULL;
e5fe3477 466
00943fd6
JH
467 /*
468 * Remove the entry from the syncache timer/timeout queue. Note
469 * that we do not try to stop any running timer since we do not know
470 * whether the timer's message is in-transit or not. Since timeouts
471 * are fairly long, taking an unneeded callout does not detrimentally
472 * effect performance.
473 */
474 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq);
984263bc
MD
475
476 syncache_free(sc);
477}
478
479/*
00943fd6
JH
480 * Place a timeout message on the TCP thread's message queue.
481 * This routine runs in soft interrupt context.
482 *
483 * An invariant is for this routine to be called, the callout must
484 * have been active. Note that the callout is not deactivated until
485 * after the message has been processed in syncache_timer_handler() below.
486 */
487static void
488syncache_timer(void *p)
489{
490 struct netmsg_sc_timer *msg = p;
491
002c1265 492 lwkt_sendmsg(msg->nm_mrec->port, &msg->base.lmsg);
00943fd6
JH
493}
494
495/*
496 * Service a timer message queued by timer expiration.
497 * This routine runs in the TCP protocol thread.
498 *
984263bc
MD
499 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
500 * If we have retransmitted an entry the maximum number of times, expire it.
00943fd6
JH
501 *
502 * When we finish processing timed-out entries, we restart the timer if there
503 * are any entries still on the queue and deactivate it otherwise. Only after
504 * a timer has been deactivated here can it be restarted by syncache_timeout().
984263bc 505 */
4599cf19 506static void
002c1265 507syncache_timer_handler(netmsg_t msg)
984263bc 508{
00943fd6 509 struct tcp_syncache_percpu *syncache_percpu;
c1d0893d
MD
510 struct syncache *sc;
511 struct syncache marker;
512 struct syncache_list *list;
984263bc 513 struct inpcb *inp;
00943fd6 514 int slot;
984263bc 515
002c1265 516 slot = ((struct netmsg_sc_timer *)msg)->nm_mrec->slot;
00943fd6 517 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
984263bc 518
c1d0893d
MD
519 list = &syncache_percpu->timerq[slot];
520
521 /*
522 * Use a marker to keep our place in the scan. syncache_drop()
523 * can block and cause any next pointer we cache to become stale.
524 */
525 marker.sc_flags = SCF_MARKER;
526 TAILQ_INSERT_HEAD(list, &marker, sc_timerq);
527
528 while ((sc = TAILQ_NEXT(&marker, sc_timerq)) != NULL) {
529 /*
530 * Move the marker.
531 */
532 TAILQ_REMOVE(list, &marker, sc_timerq);
533 TAILQ_INSERT_AFTER(list, sc, &marker, sc_timerq);
534
535 if (sc->sc_flags & SCF_MARKER)
536 continue;
537
538 if (ticks < sc->sc_rxttime)
00943fd6 539 break; /* finished because timerq sorted by time */
e5fe3477 540 if (sc->sc_tp == NULL) {
e5fe3477
MD
541 syncache_drop(sc, NULL);
542 tcpstat.tcps_sc_stale++;
543 continue;
544 }
984263bc
MD
545 inp = sc->sc_tp->t_inpcb;
546 if (slot == SYNCACHE_MAXREXMTS ||
547 slot >= tcp_syncache.rexmt_limit ||
16961763 548 inp == NULL ||
984263bc 549 inp->inp_gencnt != sc->sc_inp_gencnt) {
984263bc
MD
550 syncache_drop(sc, NULL);
551 tcpstat.tcps_sc_stale++;
552 continue;
553 }
554 /*
555 * syncache_respond() may call back into the syncache to
556 * to modify another entry, so do not obtain the next
557 * entry on the timer chain until it has completed.
558 */
f23061d4 559 syncache_respond(sc, NULL);
984263bc 560 tcpstat.tcps_sc_retransmitted++;
c1d0893d 561 TAILQ_REMOVE(list, sc, sc_timerq);
00943fd6 562 syncache_timeout(syncache_percpu, sc, slot + 1);
984263bc 563 }
c1d0893d
MD
564 TAILQ_REMOVE(list, &marker, sc_timerq);
565
566 if (sc != NULL) {
00943fd6 567 callout_reset(&syncache_percpu->tt_timerq[slot],
c1d0893d
MD
568 sc->sc_rxttime - ticks, syncache_timer,
569 &syncache_percpu->mrec[slot]);
570 } else {
00943fd6 571 callout_deactivate(&syncache_percpu->tt_timerq[slot]);
c1d0893d 572 }
002c1265 573 lwkt_replymsg(&msg->base.lmsg, 0);
984263bc
MD
574}
575
576/*
577 * Find an entry in the syncache.
578 */
579struct syncache *
f3f70f0d 580syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
984263bc 581{
00943fd6 582 struct tcp_syncache_percpu *syncache_percpu;
984263bc
MD
583 struct syncache *sc;
584 struct syncache_head *sch;
984263bc 585
00943fd6 586 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
984263bc
MD
587#ifdef INET6
588 if (inc->inc_isipv6) {
00943fd6 589 sch = &syncache_percpu->hashbase[
984263bc
MD
590 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
591 *schp = sch;
d982be66
JH
592 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash)
593 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
984263bc 594 return (sc);
984263bc
MD
595 } else
596#endif
597 {
00943fd6 598 sch = &syncache_percpu->hashbase[
984263bc
MD
599 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
600 *schp = sch;
984263bc
MD
601 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
602#ifdef INET6
603 if (sc->sc_inc.inc_isipv6)
604 continue;
605#endif
d982be66 606 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
984263bc 607 return (sc);
984263bc 608 }
984263bc
MD
609 }
610 return (NULL);
611}
612
613/*
614 * This function is called when we get a RST for a
615 * non-existent connection, so that we can see if the
616 * connection is in the syn cache. If it is, zap it.
617 */
618void
f3f70f0d 619syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
984263bc
MD
620{
621 struct syncache *sc;
622 struct syncache_head *sch;
623
624 sc = syncache_lookup(inc, &sch);
c1d0893d 625 if (sc == NULL) {
984263bc 626 return;
c1d0893d 627 }
984263bc
MD
628 /*
629 * If the RST bit is set, check the sequence number to see
630 * if this is a valid reset segment.
631 * RFC 793 page 37:
632 * In all states except SYN-SENT, all reset (RST) segments
633 * are validated by checking their SEQ-fields. A reset is
634 * valid if its sequence number is in the window.
635 *
636 * The sequence number in the reset segment is normally an
637 * echo of our outgoing acknowlegement numbers, but some hosts
638 * send a reset with the sequence number at the rightmost edge
639 * of our receive window, and we have to handle this case.
640 */
641 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
642 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
643 syncache_drop(sc, sch);
644 tcpstat.tcps_sc_reset++;
645 }
646}
647
648void
f3f70f0d 649syncache_badack(struct in_conninfo *inc)
984263bc
MD
650{
651 struct syncache *sc;
652 struct syncache_head *sch;
653
654 sc = syncache_lookup(inc, &sch);
655 if (sc != NULL) {
656 syncache_drop(sc, sch);
657 tcpstat.tcps_sc_badack++;
658 }
659}
660
661void
f3f70f0d 662syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
984263bc
MD
663{
664 struct syncache *sc;
665 struct syncache_head *sch;
666
667 /* we are called at splnet() here */
668 sc = syncache_lookup(inc, &sch);
669 if (sc == NULL)
670 return;
671
672 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
673 if (ntohl(th->th_seq) != sc->sc_iss)
674 return;
675
676 /*
677 * If we've rertransmitted 3 times and this is our second error,
678 * we remove the entry. Otherwise, we allow it to continue on.
679 * This prevents us from incorrectly nuking an entry during a
680 * spurious network outage.
681 *
682 * See tcp_notify().
683 */
684 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) {
685 sc->sc_flags |= SCF_UNREACH;
686 return;
687 }
688 syncache_drop(sc, sch);
689 tcpstat.tcps_sc_unreach++;
690}
691
692/*
693 * Build a new TCP socket structure from a syncache entry.
48e7b118
MD
694 *
695 * This is called from the context of the SYN+ACK
984263bc
MD
696 */
697static struct socket *
7e31206a 698syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
984263bc 699{
ed894f8c 700 struct inpcb *inp = NULL, *linp;
984263bc 701 struct socket *so;
2ce132be 702 struct tcpcb *tp, *ltp;
48e7b118 703 lwkt_port_t port;
61896e3c
JH
704#ifdef INET6
705 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
706#else
707 const boolean_t isipv6 = FALSE;
708#endif
88da6203 709 struct sockaddr_in sin_faddr;
68dc4251 710 struct sockaddr_in6 sin6_faddr;
88da6203
SZ
711 struct sockaddr *faddr;
712
713 if (isipv6) {
68dc4251
SZ
714 faddr = (struct sockaddr *)&sin6_faddr;
715 sin6_faddr.sin6_family = AF_INET6;
716 sin6_faddr.sin6_len = sizeof(sin6_faddr);
717 sin6_faddr.sin6_addr = sc->sc_inc.inc6_faddr;
718 sin6_faddr.sin6_port = sc->sc_inc.inc_fport;
719 sin6_faddr.sin6_flowinfo = sin6_faddr.sin6_scope_id = 0;
88da6203 720 } else {
88da6203
SZ
721 faddr = (struct sockaddr *)&sin_faddr;
722 sin_faddr.sin_family = AF_INET;
723 sin_faddr.sin_len = sizeof(sin_faddr);
724 sin_faddr.sin_addr = sc->sc_inc.inc_faddr;
725 sin_faddr.sin_port = sc->sc_inc.inc_fport;
726 bzero(sin_faddr.sin_zero, sizeof(sin_faddr.sin_zero));
727 }
984263bc
MD
728
729 /*
730 * Ok, create the full blown connection, and set things up
731 * as they would have been set up if we had created the
732 * connection when the SYN arrived. If we can't create
733 * the connection, abort it.
0ce0603e
MD
734 *
735 * Set the protocol processing port for the socket to the current
736 * port (that the connection came in on).
984263bc 737 */
88da6203 738 so = sonewconn_faddr(lso, SS_ISCONNECTED, faddr);
984263bc
MD
739 if (so == NULL) {
740 /*
741 * Drop the connection; we will send a RST if the peer
742 * retransmits the ACK,
743 */
744 tcpstat.tcps_listendrop++;
745 goto abort;
746 }
747
984263bc
MD
748 /*
749 * Insert new socket into hash list.
750 */
48e7b118 751 inp = so->so_pcb;
984263bc 752 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
61896e3c 753 if (isipv6) {
984263bc
MD
754 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
755 } else {
61896e3c 756#ifdef INET6
984263bc
MD
757 inp->inp_vflag &= ~INP_IPV6;
758 inp->inp_vflag |= INP_IPV4;
65f3e756 759 inp->inp_flags &= ~IN6P_IPV6_V6ONLY;
984263bc
MD
760#endif
761 inp->inp_laddr = sc->sc_inc.inc_laddr;
984263bc 762 }
984263bc 763 inp->inp_lport = sc->sc_inc.inc_lport;
13d8907a 764 if (in_pcbinsporthash(inp) != 0) {
984263bc
MD
765 /*
766 * Undo the assignments above if we failed to
767 * put the PCB on the hash lists.
768 */
61896e3c 769 if (isipv6)
84204577 770 inp->in6p_laddr = kin6addr_any;
f23061d4 771 else
984263bc
MD
772 inp->inp_laddr.s_addr = INADDR_ANY;
773 inp->inp_lport = 0;
774 goto abort;
775 }
0ce0603e 776 linp = lso->so_pcb;
984263bc
MD
777#ifdef IPSEC
778 /* copy old policy into new socket's */
ed894f8c 779 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp))
a6ec04bc 780 kprintf("syncache_expand: could not copy policy\n");
984263bc 781#endif
61896e3c 782 if (isipv6) {
984263bc 783 struct in6_addr laddr6;
984263bc
MD
784 /*
785 * Inherit socket options from the listening socket.
786 * Note that in6p_inputopts are not (and should not be)
787 * copied, since it stores previously received options and is
788 * used to detect if each new option is different than the
789 * previous one and hence should be passed to a user.
f23061d4 790 * If we copied in6p_inputopts, a user would not be able to
984263bc
MD
791 * receive options just after calling the accept system call.
792 */
ed894f8c
JH
793 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS;
794 if (linp->in6p_outputopts)
984263bc 795 inp->in6p_outputopts =
ed894f8c 796 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT);
984263bc
MD
797 inp->in6p_route = sc->sc_route6;
798 sc->sc_route6.ro_rt = NULL;
799
984263bc
MD
800 laddr6 = inp->in6p_laddr;
801 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
802 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
68dc4251 803 if (in6_pcbconnect(inp, faddr, &thread0)) {
984263bc 804 inp->in6p_laddr = laddr6;
984263bc
MD
805 goto abort;
806 }
61896e3c 807 } else {
984263bc 808 struct in_addr laddr;
984263bc 809
7e31206a 810 inp->inp_options = ip_srcroute(m);
984263bc
MD
811 if (inp->inp_options == NULL) {
812 inp->inp_options = sc->sc_ipopts;
813 sc->sc_ipopts = NULL;
814 }
815 inp->inp_route = sc->sc_route;
816 sc->sc_route.ro_rt = NULL;
817
984263bc
MD
818 laddr = inp->inp_laddr;
819 if (inp->inp_laddr.s_addr == INADDR_ANY)
820 inp->inp_laddr = sc->sc_inc.inc_laddr;
88da6203 821 if (in_pcbconnect(inp, faddr, &thread0)) {
984263bc 822 inp->inp_laddr = laddr;
984263bc
MD
823 goto abort;
824 }
984263bc
MD
825 }
826
48e7b118
MD
827 /*
828 * The current port should be in the context of the SYN+ACK and
829 * so should match the tcp address port.
830 *
831 * XXX we may be running on the netisr thread instead of a tcp
832 * thread, in which case port will not match
833 * curthread->td_msgport.
834 */
835 if (isipv6) {
836 port = tcp6_addrport();
837 } else {
838 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport,
839 inp->inp_laddr.s_addr, inp->inp_lport);
840 }
0ce0603e
MD
841 if (port != &curthread->td_msgport) {
842 print_backtrace(-1);
843 kprintf("TCP PORT MISMATCH %p vs %p\n",
844 port, &curthread->td_msgport);
845 }
48e7b118
MD
846 /*KKASSERT(port == &curthread->td_msgport);*/
847
984263bc
MD
848 tp = intotcpcb(inp);
849 tp->t_state = TCPS_SYN_RECEIVED;
850 tp->iss = sc->sc_iss;
851 tp->irs = sc->sc_irs;
852 tcp_rcvseqinit(tp);
853 tcp_sendseqinit(tp);
df1d2774 854 tp->snd_wnd = sc->sc_sndwnd;
984263bc
MD
855 tp->snd_wl1 = sc->sc_irs;
856 tp->rcv_up = sc->sc_irs + 1;
857 tp->rcv_wnd = sc->sc_wnd;
858 tp->rcv_adv += tp->rcv_wnd;
859
61896e3c 860 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY);
984263bc
MD
861 if (sc->sc_flags & SCF_NOOPT)
862 tp->t_flags |= TF_NOOPT;
863 if (sc->sc_flags & SCF_WINSCALE) {
61896e3c 864 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE;
df1d2774 865 tp->snd_scale = sc->sc_requested_s_scale;
984263bc
MD
866 tp->request_r_scale = sc->sc_request_r_scale;
867 }
868 if (sc->sc_flags & SCF_TIMESTAMP) {
61896e3c 869 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP;
984263bc
MD
870 tp->ts_recent = sc->sc_tsrecent;
871 tp->ts_recent_age = ticks;
872 }
91489f6b
JH
873 if (sc->sc_flags & SCF_SACK_PERMITTED)
874 tp->t_flags |= TF_SACK_PERMITTED;
984263bc 875
b1992928
MD
876#ifdef TCP_SIGNATURE
877 if (sc->sc_flags & SCF_SIGNATURE)
878 tp->t_flags |= TF_SIGNATURE;
879#endif /* TCP_SIGNATURE */
880
d5082e3d 881 tp->t_rxtsyn = sc->sc_rxtused;
984263bc
MD
882 tcp_mss(tp, sc->sc_peer_mss);
883
2ce132be
SZ
884 /*
885 * Inherit some properties from the listen socket
886 */
887 ltp = intotcpcb(linp);
888 tp->t_keepinit = ltp->t_keepinit;
7ea3a353 889 tp->t_keepidle = ltp->t_keepidle;
5d61ded3
SZ
890 tp->t_keepintvl = ltp->t_keepintvl;
891 tp->t_keepcnt = ltp->t_keepcnt;
892 tp->t_maxidle = ltp->t_maxidle;
2ce132be 893
48e7b118 894 tcp_create_timermsg(tp, port);
2ce132be 895 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
984263bc
MD
896
897 tcpstat.tcps_accepts++;
898 return (so);
899
900abort:
901 if (so != NULL)
fd86a41c 902 soabort_oncpu(so);
984263bc
MD
903 return (NULL);
904}
905
906/*
907 * This function gets called when we receive an ACK for a
908 * socket in the LISTEN state. We look up the connection
909 * in the syncache, and if its there, we pull it out of
910 * the cache and turn it into a full-blown connection in
911 * the SYN-RECEIVED state.
912 */
913int
f3f70f0d
SW
914syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop,
915 struct mbuf *m)
984263bc
MD
916{
917 struct syncache *sc;
918 struct syncache_head *sch;
919 struct socket *so;
920
921 sc = syncache_lookup(inc, &sch);
922 if (sc == NULL) {
923 /*
f23061d4 924 * There is no syncache entry, so see if this ACK is
984263bc
MD
925 * a returning syncookie. To do this, first:
926 * A. See if this socket has had a syncache entry dropped in
927 * the past. We don't want to accept a bogus syncookie
f23061d4 928 * if we've never received a SYN.
984263bc
MD
929 * B. check that the syncookie is valid. If it is, then
930 * cobble up a fake syncache entry, and return.
931 */
932 if (!tcp_syncookies)
933 return (0);
934 sc = syncookie_lookup(inc, th, *sop);
935 if (sc == NULL)
936 return (0);
937 sch = NULL;
938 tcpstat.tcps_sc_recvcookie++;
939 }
940
941 /*
942 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
943 */
944 if (th->th_ack != sc->sc_iss + 1)
945 return (0);
946
7e31206a 947 so = syncache_socket(sc, *sop, m);
984263bc
MD
948 if (so == NULL) {
949#if 0
950resetandabort:
951 /* XXXjlemon check this - is this correct? */
f23061d4 952 tcp_respond(NULL, m, m, th,
61896e3c 953 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK);
984263bc
MD
954#endif
955 m_freem(m); /* XXX only needed for above */
956 tcpstat.tcps_sc_aborted++;
957 } else {
984263bc
MD
958 tcpstat.tcps_sc_completed++;
959 }
960 if (sch == NULL)
961 syncache_free(sc);
962 else
963 syncache_drop(sc, sch);
964 *sop = so;
965 return (1);
966}
967
968/*
969 * Given a LISTEN socket and an inbound SYN request, add
970 * this to the syn cache, and send back a segment:
971 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
972 * to the source.
973 *
974 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
975 * Doing so would require that we hold onto the data and deliver it
976 * to the application. However, if we are the target of a SYN-flood
977 * DoS attack, an attacker could send data which would eventually
978 * consume all available buffer space if it were ACKed. By not ACKing
979 * the data, we avoid this DoS scenario.
980 */
981int
f3f70f0d 982syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
b09567cc 983 struct socket *so, struct mbuf *m)
984263bc 984{
00943fd6 985 struct tcp_syncache_percpu *syncache_percpu;
984263bc 986 struct tcpcb *tp;
984263bc
MD
987 struct syncache *sc = NULL;
988 struct syncache_head *sch;
989 struct mbuf *ipopts = NULL;
913d40d1 990 int win;
984263bc 991
00943fd6 992 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
984263bc
MD
993 tp = sototcpcb(so);
994
995 /*
996 * Remember the IP options, if any.
997 */
998#ifdef INET6
999 if (!inc->inc_isipv6)
1000#endif
7e31206a 1001 ipopts = ip_srcroute(m);
984263bc
MD
1002
1003 /*
1004 * See if we already have an entry for this connection.
1005 * If we do, resend the SYN,ACK, and reset the retransmit timer.
1006 *
1007 * XXX
91489f6b
JH
1008 * The syncache should be re-initialized with the contents
1009 * of the new SYN which may have different options.
984263bc
MD
1010 */
1011 sc = syncache_lookup(inc, &sch);
1012 if (sc != NULL) {
1013 tcpstat.tcps_sc_dupsyn++;
1014 if (ipopts) {
1015 /*
1016 * If we were remembering a previous source route,
1017 * forget it and use the new one we've been given.
1018 */
1019 if (sc->sc_ipopts)
f23061d4 1020 m_free(sc->sc_ipopts);
984263bc
MD
1021 sc->sc_ipopts = ipopts;
1022 }
1023 /*
1024 * Update timestamp if present.
1025 */
1026 if (sc->sc_flags & SCF_TIMESTAMP)
1027 sc->sc_tsrecent = to->to_tsval;
91489f6b
JH
1028
1029 /* Just update the TOF_SACK_PERMITTED for now. */
1030 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1031 sc->sc_flags |= SCF_SACK_PERMITTED;
1032 else
1033 sc->sc_flags &= ~SCF_SACK_PERMITTED;
1034
df1d2774
SZ
1035 /* Update initial send window */
1036 sc->sc_sndwnd = th->th_win;
1037
984263bc
MD
1038 /*
1039 * PCB may have changed, pick up new values.
1040 */
1041 sc->sc_tp = tp;
1042 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
1043 if (syncache_respond(sc, m) == 0) {
00943fd6 1044 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot],
e5fe3477 1045 sc, sc_timerq);
00943fd6 1046 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot);
f23061d4 1047 tcpstat.tcps_sndacks++;
984263bc
MD
1048 tcpstat.tcps_sndtotal++;
1049 }
984263bc
MD
1050 return (1);
1051 }
1052
984263bc
MD
1053 /*
1054 * Fill in the syncache values.
1055 */
9f42c129 1056 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO);
984263bc
MD
1057 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
1058 sc->sc_ipopts = ipopts;
1059 sc->sc_inc.inc_fport = inc->inc_fport;
1060 sc->sc_inc.inc_lport = inc->inc_lport;
e5fe3477 1061 sc->sc_tp = tp;
984263bc
MD
1062#ifdef INET6
1063 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1064 if (inc->inc_isipv6) {
1065 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1066 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1067 sc->sc_route6.ro_rt = NULL;
1068 } else
1069#endif
1070 {
1071 sc->sc_inc.inc_faddr = inc->inc_faddr;
1072 sc->sc_inc.inc_laddr = inc->inc_laddr;
1073 sc->sc_route.ro_rt = NULL;
1074 }
1075 sc->sc_irs = th->th_seq;
1076 sc->sc_flags = 0;
1077 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0;
1078 if (tcp_syncookies)
1079 sc->sc_iss = syncookie_generate(sc);
1080 else
0ced1954 1081 sc->sc_iss = karc4random();
984263bc 1082
6d49aa6f
MD
1083 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */
1084 win = ssb_space(&so->so_rcv);
984263bc
MD
1085 win = imax(win, 0);
1086 win = imin(win, TCP_MAXWIN);
1087 sc->sc_wnd = win;
1088
1089 if (tcp_do_rfc1323) {
1090 /*
1091 * A timestamp received in a SYN makes
1092 * it ok to send timestamp requests and replies.
1093 */
1094 if (to->to_flags & TOF_TS) {
1095 sc->sc_tsrecent = to->to_tsval;
1096 sc->sc_flags |= SCF_TIMESTAMP;
1097 }
1098 if (to->to_flags & TOF_SCALE) {
46e92930 1099 int wscale = TCP_MIN_WINSHIFT;
984263bc
MD
1100
1101 /* Compute proper scaling value from buffer space */
1102 while (wscale < TCP_MAX_WINSHIFT &&
46e92930 1103 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) {
984263bc 1104 wscale++;
46e92930 1105 }
984263bc
MD
1106 sc->sc_request_r_scale = wscale;
1107 sc->sc_requested_s_scale = to->to_requested_s_scale;
1108 sc->sc_flags |= SCF_WINSCALE;
1109 }
1110 }
91489f6b
JH
1111 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1112 sc->sc_flags |= SCF_SACK_PERMITTED;
984263bc
MD
1113 if (tp->t_flags & TF_NOOPT)
1114 sc->sc_flags = SCF_NOOPT;
b1992928
MD
1115#ifdef TCP_SIGNATURE
1116 /*
1117 * If listening socket requested TCP digests, and received SYN
1118 * contains the option, flag this in the syncache so that
1119 * syncache_respond() will do the right thing with the SYN+ACK.
1120 * XXX Currently we always record the option by default and will
1121 * attempt to use it in syncache_respond().
1122 */
1123 if (to->to_flags & TOF_SIGNATURE)
1124 sc->sc_flags = SCF_SIGNATURE;
1125#endif /* TCP_SIGNATURE */
df1d2774 1126 sc->sc_sndwnd = th->th_win;
984263bc 1127
984263bc
MD
1128 if (syncache_respond(sc, m) == 0) {
1129 syncache_insert(sc, sch);
1130 tcpstat.tcps_sndacks++;
1131 tcpstat.tcps_sndtotal++;
1132 } else {
1133 syncache_free(sc);
1134 tcpstat.tcps_sc_dropped++;
1135 }
984263bc
MD
1136 return (1);
1137}
1138
1139static int
f3f70f0d 1140syncache_respond(struct syncache *sc, struct mbuf *m)
984263bc
MD
1141{
1142 u_int8_t *optp;
1143 int optlen, error;
1144 u_int16_t tlen, hlen, mssopt;
1145 struct ip *ip = NULL;
1146 struct rtentry *rt;
1147 struct tcphdr *th;
984263bc 1148 struct ip6_hdr *ip6 = NULL;
61896e3c
JH
1149#ifdef INET6
1150 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1151#else
1152 const boolean_t isipv6 = FALSE;
984263bc
MD
1153#endif
1154
61896e3c 1155 if (isipv6) {
984263bc
MD
1156 rt = tcp_rtlookup6(&sc->sc_inc);
1157 if (rt != NULL)
1158 mssopt = rt->rt_ifp->if_mtu -
1159 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
f23061d4 1160 else
984263bc
MD
1161 mssopt = tcp_v6mssdflt;
1162 hlen = sizeof(struct ip6_hdr);
61896e3c 1163 } else {
984263bc
MD
1164 rt = tcp_rtlookup(&sc->sc_inc);
1165 if (rt != NULL)
1166 mssopt = rt->rt_ifp->if_mtu -
1167 (sizeof(struct ip) + sizeof(struct tcphdr));
f23061d4 1168 else
984263bc
MD
1169 mssopt = tcp_mssdflt;
1170 hlen = sizeof(struct ip);
1171 }
1172
1173 /* Compute the size of the TCP options. */
1174 if (sc->sc_flags & SCF_NOOPT) {
1175 optlen = 0;
1176 } else {
1177 optlen = TCPOLEN_MAXSEG +
1178 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) +
1179 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) +
91489f6b 1180 ((sc->sc_flags & SCF_SACK_PERMITTED) ?
f23061d4 1181 TCPOLEN_SACK_PERMITTED_ALIGNED : 0);
b1992928 1182#ifdef TCP_SIGNATURE
4931a889
SZ
1183 optlen += ((sc->sc_flags & SCF_SIGNATURE) ?
1184 (TCPOLEN_SIGNATURE + 2) : 0);
b1992928 1185#endif /* TCP_SIGNATURE */
984263bc
MD
1186 }
1187 tlen = hlen + sizeof(struct tcphdr) + optlen;
1188
1189 /*
1190 * XXX
1191 * assume that the entire packet will fit in a header mbuf
1192 */
1193 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small"));
1194
1195 /*
1196 * XXX shouldn't this reuse the mbuf if possible ?
1197 * Create the IP+TCP header from scratch.
1198 */
1199 if (m)
1200 m_freem(m);
1201
74f1caca 1202 m = m_gethdr(MB_DONTWAIT, MT_HEADER);
984263bc
MD
1203 if (m == NULL)
1204 return (ENOBUFS);
1205 m->m_data += max_linkhdr;
1206 m->m_len = tlen;
1207 m->m_pkthdr.len = tlen;
1208 m->m_pkthdr.rcvif = NULL;
1209
61896e3c 1210 if (isipv6) {
984263bc
MD
1211 ip6 = mtod(m, struct ip6_hdr *);
1212 ip6->ip6_vfc = IPV6_VERSION;
1213 ip6->ip6_nxt = IPPROTO_TCP;
1214 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1215 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1216 ip6->ip6_plen = htons(tlen - hlen);
1217 /* ip6_hlim is set after checksum */
1218 /* ip6_flow = ??? */
1219
1220 th = (struct tcphdr *)(ip6 + 1);
61896e3c 1221 } else {
984263bc
MD
1222 ip = mtod(m, struct ip *);
1223 ip->ip_v = IPVERSION;
1224 ip->ip_hl = sizeof(struct ip) >> 2;
1225 ip->ip_len = tlen;
1226 ip->ip_id = 0;
1227 ip->ip_off = 0;
1228 ip->ip_sum = 0;
1229 ip->ip_p = IPPROTO_TCP;
1230 ip->ip_src = sc->sc_inc.inc_laddr;
1231 ip->ip_dst = sc->sc_inc.inc_faddr;
1232 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */
1233 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */
1234
1235 /*
61896e3c
JH
1236 * See if we should do MTU discovery. Route lookups are
1237 * expensive, so we will only unset the DF bit if:
984263bc
MD
1238 *
1239 * 1) path_mtu_discovery is disabled
1240 * 2) the SCF_UNREACH flag has been set
1241 */
1242 if (path_mtu_discovery
1243 && ((sc->sc_flags & SCF_UNREACH) == 0)) {
1244 ip->ip_off |= IP_DF;
1245 }
1246
1247 th = (struct tcphdr *)(ip + 1);
1248 }
1249 th->th_sport = sc->sc_inc.inc_lport;
1250 th->th_dport = sc->sc_inc.inc_fport;
1251
1252 th->th_seq = htonl(sc->sc_iss);
1253 th->th_ack = htonl(sc->sc_irs + 1);
1254 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1255 th->th_x2 = 0;
61896e3c 1256 th->th_flags = TH_SYN | TH_ACK;
984263bc
MD
1257 th->th_win = htons(sc->sc_wnd);
1258 th->th_urp = 0;
1259
1260 /* Tack on the TCP options. */
1261 if (optlen == 0)
1262 goto no_options;
1263 optp = (u_int8_t *)(th + 1);
1264 *optp++ = TCPOPT_MAXSEG;
1265 *optp++ = TCPOLEN_MAXSEG;
1266 *optp++ = (mssopt >> 8) & 0xff;
1267 *optp++ = mssopt & 0xff;
1268
1269 if (sc->sc_flags & SCF_WINSCALE) {
1270 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
1271 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
1272 sc->sc_request_r_scale);
1273 optp += 4;
1274 }
1275
1276 if (sc->sc_flags & SCF_TIMESTAMP) {
1277 u_int32_t *lp = (u_int32_t *)(optp);
1278
1279 /* Form timestamp option as shown in appendix A of RFC 1323. */
1280 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1281 *lp++ = htonl(ticks);
1282 *lp = htonl(sc->sc_tsrecent);
1283 optp += TCPOLEN_TSTAMP_APPA;
1284 }
1285
b1992928
MD
1286#ifdef TCP_SIGNATURE
1287 /*
1288 * Handle TCP-MD5 passive opener response.
1289 */
1290 if (sc->sc_flags & SCF_SIGNATURE) {
1291 u_int8_t *bp = optp;
1292 int i;
1293
1294 *bp++ = TCPOPT_SIGNATURE;
1295 *bp++ = TCPOLEN_SIGNATURE;
1296 for (i = 0; i < TCP_SIGLEN; i++)
1297 *bp++ = 0;
1298 tcpsignature_compute(m, 0, optlen,
1299 optp + 2, IPSEC_DIR_OUTBOUND);
1300 *bp++ = TCPOPT_NOP;
1301 *bp++ = TCPOPT_EOL;
1302 optp += TCPOLEN_SIGNATURE + 2;
4931a889 1303 }
b1992928
MD
1304#endif /* TCP_SIGNATURE */
1305
91489f6b
JH
1306 if (sc->sc_flags & SCF_SACK_PERMITTED) {
1307 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED);
1308 optp += TCPOLEN_SACK_PERMITTED_ALIGNED;
1309 }
1310
61896e3c
JH
1311no_options:
1312 if (isipv6) {
984263bc
MD
1313 struct route_in6 *ro6 = &sc->sc_route6;
1314
1315 th->th_sum = 0;
1316 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
1317 ip6->ip6_hlim = in6_selecthlim(NULL,
1318 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL);
1319 error = ip6_output(m, NULL, ro6, 0, NULL, NULL,
1320 sc->sc_tp->t_inpcb);
61896e3c 1321 } else {
f23061d4
JH
1322 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1323 htons(tlen - hlen + IPPROTO_TCP));
984263bc
MD
1324 m->m_pkthdr.csum_flags = CSUM_TCP;
1325 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
7df36335 1326 m->m_pkthdr.csum_thlen = sizeof(struct tcphdr) + optlen;
1dbb3516
SZ
1327 error = ip_output(m, sc->sc_ipopts, &sc->sc_route,
1328 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb);
984263bc
MD
1329 }
1330 return (error);
1331}
1332
1333/*
1334 * cookie layers:
1335 *
1336 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .|
1337 * | peer iss |
1338 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .|
1339 * | 0 |(A)| |
1340 * (A): peer mss index
1341 */
1342
1343/*
1344 * The values below are chosen to minimize the size of the tcp_secret
1345 * table, as well as providing roughly a 16 second lifetime for the cookie.
1346 */
1347
1348#define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */
1349#define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */
1350
1351#define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1)
1352#define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS)
1353#define SYNCOOKIE_TIMEOUT \
1354 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT))
f23061d4 1355#define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK)
984263bc
MD
1356
1357static struct {
1358 u_int32_t ts_secbits[4];
1359 u_int ts_expire;
1360} tcp_secret[SYNCOOKIE_NSECRETS];
1361
1362static int tcp_msstab[] = { 0, 536, 1460, 8960 };
1363
1364static MD5_CTX syn_ctx;
1365
1366#define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v))
1367
1368struct md5_add {
1369 u_int32_t laddr, faddr;
1370 u_int32_t secbits[4];
1371 u_int16_t lport, fport;
1372};
1373
1374#ifdef CTASSERT
1375CTASSERT(sizeof(struct md5_add) == 28);
1376#endif
1377
1378/*
1379 * Consider the problem of a recreated (and retransmitted) cookie. If the
f23061d4
JH
1380 * original SYN was accepted, the connection is established. The second
1381 * SYN is inflight, and if it arrives with an ISN that falls within the
1382 * receive window, the connection is killed.
984263bc
MD
1383 *
1384 * However, since cookies have other problems, this may not be worth
1385 * worrying about.
1386 */
1387
1388static u_int32_t
1389syncookie_generate(struct syncache *sc)
1390{
1391 u_int32_t md5_buffer[4];
1392 u_int32_t data;
1393 int idx, i;
1394 struct md5_add add;
61896e3c
JH
1395#ifdef INET6
1396 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1397#else
1398 const boolean_t isipv6 = FALSE;
1399#endif
984263bc
MD
1400
1401 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK;
1402 if (tcp_secret[idx].ts_expire < ticks) {
1403 for (i = 0; i < 4; i++)
0ced1954 1404 tcp_secret[idx].ts_secbits[i] = karc4random();
984263bc
MD
1405 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT;
1406 }
b370aff7 1407 for (data = NELEM(tcp_msstab) - 1; data > 0; data--)
984263bc
MD
1408 if (tcp_msstab[data] <= sc->sc_peer_mss)
1409 break;
1410 data = (data << SYNCOOKIE_WNDBITS) | idx;
1411 data ^= sc->sc_irs; /* peer's iss */
1412 MD5Init(&syn_ctx);
61896e3c 1413 if (isipv6) {
984263bc
MD
1414 MD5Add(sc->sc_inc.inc6_laddr);
1415 MD5Add(sc->sc_inc.inc6_faddr);
1416 add.laddr = 0;
1417 add.faddr = 0;
61896e3c 1418 } else {
984263bc
MD
1419 add.laddr = sc->sc_inc.inc_laddr.s_addr;
1420 add.faddr = sc->sc_inc.inc_faddr.s_addr;
1421 }
1422 add.lport = sc->sc_inc.inc_lport;
1423 add.fport = sc->sc_inc.inc_fport;
1424 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1425 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1426 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1427 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1428 MD5Add(add);
1429 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1430 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK);
1431 return (data);
1432}
1433
1434static struct syncache *
f3f70f0d 1435syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so)
984263bc
MD
1436{
1437 u_int32_t md5_buffer[4];
1438 struct syncache *sc;
1439 u_int32_t data;
1440 int wnd, idx;
1441 struct md5_add add;
1442
1443 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */
1444 idx = data & SYNCOOKIE_WNDMASK;
1445 if (tcp_secret[idx].ts_expire < ticks ||
1446 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks)
1447 return (NULL);
1448 MD5Init(&syn_ctx);
1449#ifdef INET6
1450 if (inc->inc_isipv6) {
1451 MD5Add(inc->inc6_laddr);
1452 MD5Add(inc->inc6_faddr);
1453 add.laddr = 0;
1454 add.faddr = 0;
1455 } else
1456#endif
1457 {
1458 add.laddr = inc->inc_laddr.s_addr;
1459 add.faddr = inc->inc_faddr.s_addr;
1460 }
1461 add.lport = inc->inc_lport;
1462 add.fport = inc->inc_fport;
1463 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1464 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1465 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1466 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1467 MD5Add(add);
1468 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1469 data ^= md5_buffer[0];
f23061d4 1470 if (data & ~SYNCOOKIE_DATAMASK)
984263bc
MD
1471 return (NULL);
1472 data = data >> SYNCOOKIE_WNDBITS;
1473
984263bc
MD
1474 /*
1475 * Fill in the syncache values.
1476 * XXX duplicate code from syncache_add
1477 */
9f42c129 1478 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO);
984263bc
MD
1479 sc->sc_ipopts = NULL;
1480 sc->sc_inc.inc_fport = inc->inc_fport;
1481 sc->sc_inc.inc_lport = inc->inc_lport;
1482#ifdef INET6
1483 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1484 if (inc->inc_isipv6) {
1485 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1486 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1487 sc->sc_route6.ro_rt = NULL;
1488 } else
1489#endif
1490 {
1491 sc->sc_inc.inc_faddr = inc->inc_faddr;
1492 sc->sc_inc.inc_laddr = inc->inc_laddr;
1493 sc->sc_route.ro_rt = NULL;
1494 }
1495 sc->sc_irs = th->th_seq - 1;
1496 sc->sc_iss = th->th_ack - 1;
6d49aa6f 1497 wnd = ssb_space(&so->so_rcv);
984263bc
MD
1498 wnd = imax(wnd, 0);
1499 wnd = imin(wnd, TCP_MAXWIN);
1500 sc->sc_wnd = wnd;
1501 sc->sc_flags = 0;
1502 sc->sc_rxtslot = 0;
1503 sc->sc_peer_mss = tcp_msstab[data];
1504 return (sc);
1505}