tcp: Allow per-tcpcb keepintvl and keepcnt
[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
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
b1992928 75#include "opt_inet.h"
984263bc
MD
76#include "opt_inet6.h"
77#include "opt_ipsec.h"
78
79#include <sys/param.h>
80#include <sys/systm.h>
81#include <sys/kernel.h>
82#include <sys/sysctl.h>
83#include <sys/malloc.h>
84#include <sys/mbuf.h>
85#include <sys/md5.h>
86#include <sys/proc.h> /* for proc0 declaration */
87#include <sys/random.h>
88#include <sys/socket.h>
89#include <sys/socketvar.h>
3f9db7f8 90#include <sys/in_cksum.h>
984263bc 91
00943fd6 92#include <sys/msgport2.h>
4599cf19 93#include <net/netmsg2.h>
00943fd6 94
984263bc
MD
95#include <net/if.h>
96#include <net/route.h>
97
98#include <netinet/in.h>
99#include <netinet/in_systm.h>
100#include <netinet/ip.h>
101#include <netinet/in_var.h>
102#include <netinet/in_pcb.h>
103#include <netinet/ip_var.h>
984263bc 104#include <netinet/ip6.h>
61896e3c 105#ifdef INET6
984263bc
MD
106#include <netinet/icmp6.h>
107#include <netinet6/nd6.h>
61896e3c 108#endif
984263bc
MD
109#include <netinet6/ip6_var.h>
110#include <netinet6/in6_pcb.h>
984263bc
MD
111#include <netinet/tcp.h>
112#include <netinet/tcp_fsm.h>
113#include <netinet/tcp_seq.h>
114#include <netinet/tcp_timer.h>
a48c5dd5 115#include <netinet/tcp_timer2.h>
984263bc 116#include <netinet/tcp_var.h>
984263bc 117#include <netinet6/tcp6_var.h>
984263bc
MD
118
119#ifdef IPSEC
120#include <netinet6/ipsec.h>
121#ifdef INET6
122#include <netinet6/ipsec6.h>
123#endif
d2438d69 124#include <netproto/key/key.h>
984263bc
MD
125#endif /*IPSEC*/
126
127#ifdef FAST_IPSEC
bf844ffa 128#include <netproto/ipsec/ipsec.h>
984263bc 129#ifdef INET6
bf844ffa 130#include <netproto/ipsec/ipsec6.h>
984263bc 131#endif
bf844ffa 132#include <netproto/ipsec/key.h>
984263bc
MD
133#define IPSEC
134#endif /*FAST_IPSEC*/
135
984263bc
MD
136static int tcp_syncookies = 1;
137SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW,
f23061d4 138 &tcp_syncookies, 0,
984263bc
MD
139 "Use TCP SYN cookies if the syncache overflows");
140
141static void syncache_drop(struct syncache *, struct syncache_head *);
142static void syncache_free(struct syncache *);
143static void syncache_insert(struct syncache *, struct syncache_head *);
144struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **);
145static int syncache_respond(struct syncache *, struct mbuf *);
7e31206a
SZ
146static struct socket *syncache_socket(struct syncache *, struct socket *,
147 struct mbuf *);
984263bc
MD
148static void syncache_timer(void *);
149static u_int32_t syncookie_generate(struct syncache *);
150static struct syncache *syncookie_lookup(struct in_conninfo *,
151 struct tcphdr *, struct socket *);
152
153/*
154 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies.
155 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds,
156 * the odds are that the user has given up attempting to connect by then.
157 */
158#define SYNCACHE_MAXREXMTS 3
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
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 }
9f42c129 282 kfree(sc, M_SYNCACHE);
984263bc
MD
283}
284
285void
286syncache_init(void)
287{
00943fd6 288 int i, cpu;
984263bc 289
984263bc
MD
290 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE;
291 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT;
292 tcp_syncache.cache_limit =
293 tcp_syncache.hashsize * tcp_syncache.bucket_limit;
294 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS;
0ced1954 295 tcp_syncache.hash_secret = karc4random();
984263bc 296
f23061d4 297 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize",
984263bc 298 &tcp_syncache.hashsize);
f23061d4 299 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit",
984263bc 300 &tcp_syncache.cache_limit);
f23061d4 301 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit",
984263bc
MD
302 &tcp_syncache.bucket_limit);
303 if (!powerof2(tcp_syncache.hashsize)) {
a6ec04bc 304 kprintf("WARNING: syncache hash size is not a power of 2.\n");
984263bc 305 tcp_syncache.hashsize = 512; /* safe default */
f23061d4 306 }
984263bc
MD
307 tcp_syncache.hashmask = tcp_syncache.hashsize - 1;
308
fb0f29c4 309 lwkt_initport_replyonly_null(&syncache_null_rport);
984263bc 310
00943fd6
JH
311 for (cpu = 0; cpu < ncpus2; cpu++) {
312 struct tcp_syncache_percpu *syncache_percpu;
313
314 syncache_percpu = &tcp_syncache_percpu[cpu];
315 /* Allocate the hash table. */
316 MALLOC(syncache_percpu->hashbase, struct syncache_head *,
317 tcp_syncache.hashsize * sizeof(struct syncache_head),
318 M_SYNCACHE, M_WAITOK);
319
320 /* Initialize the hash buckets. */
321 for (i = 0; i < tcp_syncache.hashsize; i++) {
322 struct syncache_head *bucket;
323
324 bucket = &syncache_percpu->hashbase[i];
325 TAILQ_INIT(&bucket->sch_bucket);
326 bucket->sch_length = 0;
327 }
984263bc 328
00943fd6
JH
329 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) {
330 /* Initialize the timer queues. */
331 TAILQ_INIT(&syncache_percpu->timerq[i]);
332 callout_init(&syncache_percpu->tt_timerq[i]);
333
334 syncache_percpu->mrec[i].slot = i;
c3c96e44 335 syncache_percpu->mrec[i].port = cpu_portfn(cpu);
00943fd6 336 syncache_percpu->mrec[i].msg.nm_mrec =
c3c96e44 337 &syncache_percpu->mrec[i];
002c1265 338 netmsg_init(&syncache_percpu->mrec[i].msg.base,
48e7b118
MD
339 NULL, &syncache_null_rport,
340 0, syncache_timer_handler);
00943fd6 341 }
984263bc 342 }
984263bc
MD
343}
344
345static void
f3f70f0d 346syncache_insert(struct syncache *sc, struct syncache_head *sch)
984263bc 347{
00943fd6 348 struct tcp_syncache_percpu *syncache_percpu;
984263bc 349 struct syncache *sc2;
d982be66 350 int i;
984263bc 351
00943fd6
JH
352 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
353
984263bc
MD
354 /*
355 * Make sure that we don't overflow the per-bucket
356 * limit or the total cache size limit.
357 */
984263bc
MD
358 if (sch->sch_length >= tcp_syncache.bucket_limit) {
359 /*
360 * The bucket is full, toss the oldest element.
361 */
362 sc2 = TAILQ_FIRST(&sch->sch_bucket);
363 sc2->sc_tp->ts_recent = ticks;
364 syncache_drop(sc2, sch);
365 tcpstat.tcps_sc_bucketoverflow++;
00943fd6 366 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) {
984263bc
MD
367 /*
368 * The cache is full. Toss the oldest entry in the
369 * entire cache. This is the front entry in the
370 * first non-empty timer queue with the largest
371 * timeout value.
372 */
373 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) {
00943fd6 374 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]);
c1d0893d
MD
375 while (sc2 && (sc2->sc_flags & SCF_MARKER))
376 sc2 = TAILQ_NEXT(sc2, sc_timerq);
984263bc
MD
377 if (sc2 != NULL)
378 break;
379 }
380 sc2->sc_tp->ts_recent = ticks;
381 syncache_drop(sc2, NULL);
382 tcpstat.tcps_sc_cacheoverflow++;
383 }
384
385 /* Initialize the entry's timer. */
00943fd6 386 syncache_timeout(syncache_percpu, sc, 0);
984263bc
MD
387
388 /* Put it into the bucket. */
389 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash);
390 sch->sch_length++;
00943fd6 391 syncache_percpu->cache_count++;
984263bc 392 tcpstat.tcps_sc_added++;
984263bc
MD
393}
394
e5fe3477
MD
395void
396syncache_destroy(struct tcpcb *tp)
397{
398 struct tcp_syncache_percpu *syncache_percpu;
399 struct syncache_head *bucket;
400 struct syncache *sc;
401 int i;
402
403 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
404 sc = NULL;
c1d0893d 405
e5fe3477
MD
406 for (i = 0; i < tcp_syncache.hashsize; i++) {
407 bucket = &syncache_percpu->hashbase[i];
408 TAILQ_FOREACH(sc, &bucket->sch_bucket, sc_hash) {
7123bbff 409 if (sc->sc_tp == tp)
e5fe3477 410 sc->sc_tp = NULL;
e5fe3477
MD
411 }
412 }
e5fe3477
MD
413}
414
984263bc 415static void
f3f70f0d 416syncache_drop(struct syncache *sc, struct syncache_head *sch)
984263bc 417{
00943fd6 418 struct tcp_syncache_percpu *syncache_percpu;
61896e3c
JH
419#ifdef INET6
420 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
421#else
422 const boolean_t isipv6 = FALSE;
423#endif
984263bc 424
00943fd6
JH
425 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
426
984263bc 427 if (sch == NULL) {
61896e3c 428 if (isipv6) {
00943fd6 429 sch = &syncache_percpu->hashbase[
984263bc 430 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)];
61896e3c 431 } else {
00943fd6 432 sch = &syncache_percpu->hashbase[
984263bc
MD
433 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)];
434 }
435 }
436
984263bc
MD
437 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash);
438 sch->sch_length--;
00943fd6 439 syncache_percpu->cache_count--;
984263bc 440
00943fd6 441 /*
e5fe3477
MD
442 * Cleanup
443 */
7123bbff 444 if (sc->sc_tp)
e5fe3477 445 sc->sc_tp = NULL;
e5fe3477
MD
446
447 /*
00943fd6
JH
448 * Remove the entry from the syncache timer/timeout queue. Note
449 * that we do not try to stop any running timer since we do not know
450 * whether the timer's message is in-transit or not. Since timeouts
451 * are fairly long, taking an unneeded callout does not detrimentally
452 * effect performance.
453 */
454 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq);
984263bc
MD
455
456 syncache_free(sc);
457}
458
459/*
00943fd6
JH
460 * Place a timeout message on the TCP thread's message queue.
461 * This routine runs in soft interrupt context.
462 *
463 * An invariant is for this routine to be called, the callout must
464 * have been active. Note that the callout is not deactivated until
465 * after the message has been processed in syncache_timer_handler() below.
466 */
467static void
468syncache_timer(void *p)
469{
470 struct netmsg_sc_timer *msg = p;
471
002c1265 472 lwkt_sendmsg(msg->nm_mrec->port, &msg->base.lmsg);
00943fd6
JH
473}
474
475/*
476 * Service a timer message queued by timer expiration.
477 * This routine runs in the TCP protocol thread.
478 *
984263bc
MD
479 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
480 * If we have retransmitted an entry the maximum number of times, expire it.
00943fd6
JH
481 *
482 * When we finish processing timed-out entries, we restart the timer if there
483 * are any entries still on the queue and deactivate it otherwise. Only after
484 * a timer has been deactivated here can it be restarted by syncache_timeout().
984263bc 485 */
4599cf19 486static void
002c1265 487syncache_timer_handler(netmsg_t msg)
984263bc 488{
00943fd6 489 struct tcp_syncache_percpu *syncache_percpu;
c1d0893d
MD
490 struct syncache *sc;
491 struct syncache marker;
492 struct syncache_list *list;
984263bc 493 struct inpcb *inp;
00943fd6 494 int slot;
984263bc 495
002c1265 496 slot = ((struct netmsg_sc_timer *)msg)->nm_mrec->slot;
00943fd6 497 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
984263bc 498
c1d0893d
MD
499 list = &syncache_percpu->timerq[slot];
500
501 /*
502 * Use a marker to keep our place in the scan. syncache_drop()
503 * can block and cause any next pointer we cache to become stale.
504 */
505 marker.sc_flags = SCF_MARKER;
506 TAILQ_INSERT_HEAD(list, &marker, sc_timerq);
507
508 while ((sc = TAILQ_NEXT(&marker, sc_timerq)) != NULL) {
509 /*
510 * Move the marker.
511 */
512 TAILQ_REMOVE(list, &marker, sc_timerq);
513 TAILQ_INSERT_AFTER(list, sc, &marker, sc_timerq);
514
515 if (sc->sc_flags & SCF_MARKER)
516 continue;
517
518 if (ticks < sc->sc_rxttime)
00943fd6 519 break; /* finished because timerq sorted by time */
e5fe3477 520 if (sc->sc_tp == NULL) {
e5fe3477
MD
521 syncache_drop(sc, NULL);
522 tcpstat.tcps_sc_stale++;
523 continue;
524 }
984263bc
MD
525 inp = sc->sc_tp->t_inpcb;
526 if (slot == SYNCACHE_MAXREXMTS ||
527 slot >= tcp_syncache.rexmt_limit ||
16961763 528 inp == NULL ||
984263bc 529 inp->inp_gencnt != sc->sc_inp_gencnt) {
984263bc
MD
530 syncache_drop(sc, NULL);
531 tcpstat.tcps_sc_stale++;
532 continue;
533 }
534 /*
535 * syncache_respond() may call back into the syncache to
536 * to modify another entry, so do not obtain the next
537 * entry on the timer chain until it has completed.
538 */
f23061d4 539 syncache_respond(sc, NULL);
984263bc 540 tcpstat.tcps_sc_retransmitted++;
c1d0893d 541 TAILQ_REMOVE(list, sc, sc_timerq);
00943fd6 542 syncache_timeout(syncache_percpu, sc, slot + 1);
984263bc 543 }
c1d0893d
MD
544 TAILQ_REMOVE(list, &marker, sc_timerq);
545
546 if (sc != NULL) {
00943fd6 547 callout_reset(&syncache_percpu->tt_timerq[slot],
c1d0893d
MD
548 sc->sc_rxttime - ticks, syncache_timer,
549 &syncache_percpu->mrec[slot]);
550 } else {
00943fd6 551 callout_deactivate(&syncache_percpu->tt_timerq[slot]);
c1d0893d 552 }
002c1265 553 lwkt_replymsg(&msg->base.lmsg, 0);
984263bc
MD
554}
555
556/*
557 * Find an entry in the syncache.
558 */
559struct syncache *
f3f70f0d 560syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp)
984263bc 561{
00943fd6 562 struct tcp_syncache_percpu *syncache_percpu;
984263bc
MD
563 struct syncache *sc;
564 struct syncache_head *sch;
984263bc 565
00943fd6 566 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
984263bc
MD
567#ifdef INET6
568 if (inc->inc_isipv6) {
00943fd6 569 sch = &syncache_percpu->hashbase[
984263bc
MD
570 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)];
571 *schp = sch;
d982be66
JH
572 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash)
573 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
984263bc 574 return (sc);
984263bc
MD
575 } else
576#endif
577 {
00943fd6 578 sch = &syncache_percpu->hashbase[
984263bc
MD
579 SYNCACHE_HASH(inc, tcp_syncache.hashmask)];
580 *schp = sch;
984263bc
MD
581 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) {
582#ifdef INET6
583 if (sc->sc_inc.inc_isipv6)
584 continue;
585#endif
d982be66 586 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie))
984263bc 587 return (sc);
984263bc 588 }
984263bc
MD
589 }
590 return (NULL);
591}
592
593/*
594 * This function is called when we get a RST for a
595 * non-existent connection, so that we can see if the
596 * connection is in the syn cache. If it is, zap it.
597 */
598void
f3f70f0d 599syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th)
984263bc
MD
600{
601 struct syncache *sc;
602 struct syncache_head *sch;
603
604 sc = syncache_lookup(inc, &sch);
c1d0893d 605 if (sc == NULL) {
984263bc 606 return;
c1d0893d 607 }
984263bc
MD
608 /*
609 * If the RST bit is set, check the sequence number to see
610 * if this is a valid reset segment.
611 * RFC 793 page 37:
612 * In all states except SYN-SENT, all reset (RST) segments
613 * are validated by checking their SEQ-fields. A reset is
614 * valid if its sequence number is in the window.
615 *
616 * The sequence number in the reset segment is normally an
617 * echo of our outgoing acknowlegement numbers, but some hosts
618 * send a reset with the sequence number at the rightmost edge
619 * of our receive window, and we have to handle this case.
620 */
621 if (SEQ_GEQ(th->th_seq, sc->sc_irs) &&
622 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) {
623 syncache_drop(sc, sch);
624 tcpstat.tcps_sc_reset++;
625 }
626}
627
628void
f3f70f0d 629syncache_badack(struct in_conninfo *inc)
984263bc
MD
630{
631 struct syncache *sc;
632 struct syncache_head *sch;
633
634 sc = syncache_lookup(inc, &sch);
635 if (sc != NULL) {
636 syncache_drop(sc, sch);
637 tcpstat.tcps_sc_badack++;
638 }
639}
640
641void
f3f70f0d 642syncache_unreach(struct in_conninfo *inc, struct tcphdr *th)
984263bc
MD
643{
644 struct syncache *sc;
645 struct syncache_head *sch;
646
647 /* we are called at splnet() here */
648 sc = syncache_lookup(inc, &sch);
649 if (sc == NULL)
650 return;
651
652 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */
653 if (ntohl(th->th_seq) != sc->sc_iss)
654 return;
655
656 /*
657 * If we've rertransmitted 3 times and this is our second error,
658 * we remove the entry. Otherwise, we allow it to continue on.
659 * This prevents us from incorrectly nuking an entry during a
660 * spurious network outage.
661 *
662 * See tcp_notify().
663 */
664 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) {
665 sc->sc_flags |= SCF_UNREACH;
666 return;
667 }
668 syncache_drop(sc, sch);
669 tcpstat.tcps_sc_unreach++;
670}
671
672/*
673 * Build a new TCP socket structure from a syncache entry.
48e7b118
MD
674 *
675 * This is called from the context of the SYN+ACK
984263bc
MD
676 */
677static struct socket *
7e31206a 678syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
984263bc 679{
ed894f8c 680 struct inpcb *inp = NULL, *linp;
984263bc 681 struct socket *so;
2ce132be 682 struct tcpcb *tp, *ltp;
48e7b118 683 lwkt_port_t port;
61896e3c
JH
684#ifdef INET6
685 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
686#else
687 const boolean_t isipv6 = FALSE;
688#endif
984263bc
MD
689
690 /*
691 * Ok, create the full blown connection, and set things up
692 * as they would have been set up if we had created the
693 * connection when the SYN arrived. If we can't create
694 * the connection, abort it.
0ce0603e
MD
695 *
696 * Set the protocol processing port for the socket to the current
697 * port (that the connection came in on).
984263bc
MD
698 */
699 so = sonewconn(lso, SS_ISCONNECTED);
700 if (so == NULL) {
701 /*
702 * Drop the connection; we will send a RST if the peer
703 * retransmits the ACK,
704 */
705 tcpstat.tcps_listendrop++;
706 goto abort;
707 }
708
48e7b118 709 /*
984263bc
MD
710 * Insert new socket into hash list.
711 */
48e7b118 712 inp = so->so_pcb;
984263bc 713 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6;
61896e3c 714 if (isipv6) {
984263bc
MD
715 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
716 } else {
61896e3c 717#ifdef INET6
984263bc
MD
718 inp->inp_vflag &= ~INP_IPV6;
719 inp->inp_vflag |= INP_IPV4;
65f3e756 720 inp->inp_flags &= ~IN6P_IPV6_V6ONLY;
984263bc
MD
721#endif
722 inp->inp_laddr = sc->sc_inc.inc_laddr;
984263bc 723 }
984263bc 724 inp->inp_lport = sc->sc_inc.inc_lport;
13d8907a 725 if (in_pcbinsporthash(inp) != 0) {
984263bc
MD
726 /*
727 * Undo the assignments above if we failed to
728 * put the PCB on the hash lists.
729 */
61896e3c 730 if (isipv6)
84204577 731 inp->in6p_laddr = kin6addr_any;
f23061d4 732 else
984263bc
MD
733 inp->inp_laddr.s_addr = INADDR_ANY;
734 inp->inp_lport = 0;
735 goto abort;
736 }
0ce0603e 737 linp = lso->so_pcb;
984263bc
MD
738#ifdef IPSEC
739 /* copy old policy into new socket's */
ed894f8c 740 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp))
a6ec04bc 741 kprintf("syncache_expand: could not copy policy\n");
984263bc 742#endif
61896e3c 743 if (isipv6) {
984263bc 744 struct in6_addr laddr6;
d982be66 745 struct sockaddr_in6 sin6;
984263bc
MD
746 /*
747 * Inherit socket options from the listening socket.
748 * Note that in6p_inputopts are not (and should not be)
749 * copied, since it stores previously received options and is
750 * used to detect if each new option is different than the
751 * previous one and hence should be passed to a user.
f23061d4 752 * If we copied in6p_inputopts, a user would not be able to
984263bc
MD
753 * receive options just after calling the accept system call.
754 */
ed894f8c
JH
755 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS;
756 if (linp->in6p_outputopts)
984263bc 757 inp->in6p_outputopts =
ed894f8c 758 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT);
984263bc
MD
759 inp->in6p_route = sc->sc_route6;
760 sc->sc_route6.ro_rt = NULL;
761
d982be66
JH
762 sin6.sin6_family = AF_INET6;
763 sin6.sin6_len = sizeof sin6;
764 sin6.sin6_addr = sc->sc_inc.inc6_faddr;
765 sin6.sin6_port = sc->sc_inc.inc_fport;
766 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0;
984263bc
MD
767 laddr6 = inp->in6p_laddr;
768 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
769 inp->in6p_laddr = sc->sc_inc.inc6_laddr;
d982be66 770 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) {
984263bc 771 inp->in6p_laddr = laddr6;
984263bc
MD
772 goto abort;
773 }
61896e3c 774 } else {
984263bc 775 struct in_addr laddr;
d982be66 776 struct sockaddr_in sin;
984263bc 777
7e31206a 778 inp->inp_options = ip_srcroute(m);
984263bc
MD
779 if (inp->inp_options == NULL) {
780 inp->inp_options = sc->sc_ipopts;
781 sc->sc_ipopts = NULL;
782 }
783 inp->inp_route = sc->sc_route;
784 sc->sc_route.ro_rt = NULL;
785
d982be66
JH
786 sin.sin_family = AF_INET;
787 sin.sin_len = sizeof sin;
788 sin.sin_addr = sc->sc_inc.inc_faddr;
789 sin.sin_port = sc->sc_inc.inc_fport;
790 bzero(sin.sin_zero, sizeof sin.sin_zero);
984263bc
MD
791 laddr = inp->inp_laddr;
792 if (inp->inp_laddr.s_addr == INADDR_ANY)
793 inp->inp_laddr = sc->sc_inc.inc_laddr;
d982be66 794 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) {
984263bc 795 inp->inp_laddr = laddr;
984263bc
MD
796 goto abort;
797 }
984263bc
MD
798 }
799
48e7b118
MD
800 /*
801 * The current port should be in the context of the SYN+ACK and
802 * so should match the tcp address port.
803 *
804 * XXX we may be running on the netisr thread instead of a tcp
805 * thread, in which case port will not match
806 * curthread->td_msgport.
807 */
808 if (isipv6) {
809 port = tcp6_addrport();
810 } else {
811 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport,
812 inp->inp_laddr.s_addr, inp->inp_lport);
813 }
0ce0603e
MD
814 if (port != &curthread->td_msgport) {
815 print_backtrace(-1);
816 kprintf("TCP PORT MISMATCH %p vs %p\n",
817 port, &curthread->td_msgport);
818 }
48e7b118
MD
819 /*KKASSERT(port == &curthread->td_msgport);*/
820
984263bc
MD
821 tp = intotcpcb(inp);
822 tp->t_state = TCPS_SYN_RECEIVED;
823 tp->iss = sc->sc_iss;
824 tp->irs = sc->sc_irs;
825 tcp_rcvseqinit(tp);
826 tcp_sendseqinit(tp);
827 tp->snd_wl1 = sc->sc_irs;
828 tp->rcv_up = sc->sc_irs + 1;
829 tp->rcv_wnd = sc->sc_wnd;
830 tp->rcv_adv += tp->rcv_wnd;
831
61896e3c 832 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY);
984263bc
MD
833 if (sc->sc_flags & SCF_NOOPT)
834 tp->t_flags |= TF_NOOPT;
835 if (sc->sc_flags & SCF_WINSCALE) {
61896e3c 836 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE;
984263bc
MD
837 tp->requested_s_scale = sc->sc_requested_s_scale;
838 tp->request_r_scale = sc->sc_request_r_scale;
839 }
840 if (sc->sc_flags & SCF_TIMESTAMP) {
61896e3c 841 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP;
984263bc
MD
842 tp->ts_recent = sc->sc_tsrecent;
843 tp->ts_recent_age = ticks;
844 }
91489f6b
JH
845 if (sc->sc_flags & SCF_SACK_PERMITTED)
846 tp->t_flags |= TF_SACK_PERMITTED;
984263bc 847
b1992928
MD
848#ifdef TCP_SIGNATURE
849 if (sc->sc_flags & SCF_SIGNATURE)
850 tp->t_flags |= TF_SIGNATURE;
851#endif /* TCP_SIGNATURE */
852
853
984263bc
MD
854 tcp_mss(tp, sc->sc_peer_mss);
855
856 /*
857 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment.
858 */
859 if (sc->sc_rxtslot != 0)
f23061d4 860 tp->snd_cwnd = tp->t_maxseg;
2ce132be
SZ
861
862 /*
863 * Inherit some properties from the listen socket
864 */
865 ltp = intotcpcb(linp);
866 tp->t_keepinit = ltp->t_keepinit;
7ea3a353 867 tp->t_keepidle = ltp->t_keepidle;
5d61ded3
SZ
868 tp->t_keepintvl = ltp->t_keepintvl;
869 tp->t_keepcnt = ltp->t_keepcnt;
870 tp->t_maxidle = ltp->t_maxidle;
2ce132be 871
48e7b118 872 tcp_create_timermsg(tp, port);
2ce132be 873 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
984263bc
MD
874
875 tcpstat.tcps_accepts++;
876 return (so);
877
878abort:
879 if (so != NULL)
fd86a41c 880 soabort_oncpu(so);
984263bc
MD
881 return (NULL);
882}
883
884/*
885 * This function gets called when we receive an ACK for a
886 * socket in the LISTEN state. We look up the connection
887 * in the syncache, and if its there, we pull it out of
888 * the cache and turn it into a full-blown connection in
889 * the SYN-RECEIVED state.
890 */
891int
f3f70f0d
SW
892syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop,
893 struct mbuf *m)
984263bc
MD
894{
895 struct syncache *sc;
896 struct syncache_head *sch;
897 struct socket *so;
898
899 sc = syncache_lookup(inc, &sch);
900 if (sc == NULL) {
901 /*
f23061d4 902 * There is no syncache entry, so see if this ACK is
984263bc
MD
903 * a returning syncookie. To do this, first:
904 * A. See if this socket has had a syncache entry dropped in
905 * the past. We don't want to accept a bogus syncookie
f23061d4 906 * if we've never received a SYN.
984263bc
MD
907 * B. check that the syncookie is valid. If it is, then
908 * cobble up a fake syncache entry, and return.
909 */
910 if (!tcp_syncookies)
911 return (0);
912 sc = syncookie_lookup(inc, th, *sop);
913 if (sc == NULL)
914 return (0);
915 sch = NULL;
916 tcpstat.tcps_sc_recvcookie++;
917 }
918
919 /*
920 * If seg contains an ACK, but not for our SYN/ACK, send a RST.
921 */
922 if (th->th_ack != sc->sc_iss + 1)
923 return (0);
924
7e31206a 925 so = syncache_socket(sc, *sop, m);
984263bc
MD
926 if (so == NULL) {
927#if 0
928resetandabort:
929 /* XXXjlemon check this - is this correct? */
f23061d4 930 tcp_respond(NULL, m, m, th,
61896e3c 931 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK);
984263bc
MD
932#endif
933 m_freem(m); /* XXX only needed for above */
934 tcpstat.tcps_sc_aborted++;
935 } else {
984263bc
MD
936 tcpstat.tcps_sc_completed++;
937 }
938 if (sch == NULL)
939 syncache_free(sc);
940 else
941 syncache_drop(sc, sch);
942 *sop = so;
943 return (1);
944}
945
946/*
947 * Given a LISTEN socket and an inbound SYN request, add
948 * this to the syn cache, and send back a segment:
949 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
950 * to the source.
951 *
952 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
953 * Doing so would require that we hold onto the data and deliver it
954 * to the application. However, if we are the target of a SYN-flood
955 * DoS attack, an attacker could send data which would eventually
956 * consume all available buffer space if it were ACKed. By not ACKing
957 * the data, we avoid this DoS scenario.
958 */
959int
f3f70f0d
SW
960syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
961 struct socket **sop, struct mbuf *m)
984263bc 962{
00943fd6 963 struct tcp_syncache_percpu *syncache_percpu;
984263bc
MD
964 struct tcpcb *tp;
965 struct socket *so;
966 struct syncache *sc = NULL;
967 struct syncache_head *sch;
968 struct mbuf *ipopts = NULL;
913d40d1 969 int win;
984263bc 970
00943fd6 971 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid];
984263bc
MD
972 so = *sop;
973 tp = sototcpcb(so);
974
975 /*
976 * Remember the IP options, if any.
977 */
978#ifdef INET6
979 if (!inc->inc_isipv6)
980#endif
7e31206a 981 ipopts = ip_srcroute(m);
984263bc
MD
982
983 /*
984 * See if we already have an entry for this connection.
985 * If we do, resend the SYN,ACK, and reset the retransmit timer.
986 *
987 * XXX
91489f6b
JH
988 * The syncache should be re-initialized with the contents
989 * of the new SYN which may have different options.
984263bc
MD
990 */
991 sc = syncache_lookup(inc, &sch);
992 if (sc != NULL) {
993 tcpstat.tcps_sc_dupsyn++;
994 if (ipopts) {
995 /*
996 * If we were remembering a previous source route,
997 * forget it and use the new one we've been given.
998 */
999 if (sc->sc_ipopts)
f23061d4 1000 m_free(sc->sc_ipopts);
984263bc
MD
1001 sc->sc_ipopts = ipopts;
1002 }
1003 /*
1004 * Update timestamp if present.
1005 */
1006 if (sc->sc_flags & SCF_TIMESTAMP)
1007 sc->sc_tsrecent = to->to_tsval;
91489f6b
JH
1008
1009 /* Just update the TOF_SACK_PERMITTED for now. */
1010 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1011 sc->sc_flags |= SCF_SACK_PERMITTED;
1012 else
1013 sc->sc_flags &= ~SCF_SACK_PERMITTED;
1014
984263bc
MD
1015 /*
1016 * PCB may have changed, pick up new values.
1017 */
1018 sc->sc_tp = tp;
1019 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
1020 if (syncache_respond(sc, m) == 0) {
00943fd6 1021 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot],
e5fe3477 1022 sc, sc_timerq);
00943fd6 1023 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot);
f23061d4 1024 tcpstat.tcps_sndacks++;
984263bc
MD
1025 tcpstat.tcps_sndtotal++;
1026 }
1027 *sop = NULL;
1028 return (1);
1029 }
1030
d191a96b 1031 /*
984263bc
MD
1032 * Fill in the syncache values.
1033 */
9f42c129 1034 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO);
984263bc
MD
1035 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt;
1036 sc->sc_ipopts = ipopts;
1037 sc->sc_inc.inc_fport = inc->inc_fport;
1038 sc->sc_inc.inc_lport = inc->inc_lport;
e5fe3477 1039 sc->sc_tp = tp;
984263bc
MD
1040#ifdef INET6
1041 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1042 if (inc->inc_isipv6) {
1043 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1044 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1045 sc->sc_route6.ro_rt = NULL;
1046 } else
1047#endif
1048 {
1049 sc->sc_inc.inc_faddr = inc->inc_faddr;
1050 sc->sc_inc.inc_laddr = inc->inc_laddr;
1051 sc->sc_route.ro_rt = NULL;
1052 }
1053 sc->sc_irs = th->th_seq;
1054 sc->sc_flags = 0;
1055 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0;
1056 if (tcp_syncookies)
1057 sc->sc_iss = syncookie_generate(sc);
1058 else
0ced1954 1059 sc->sc_iss = karc4random();
984263bc 1060
6d49aa6f
MD
1061 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */
1062 win = ssb_space(&so->so_rcv);
984263bc
MD
1063 win = imax(win, 0);
1064 win = imin(win, TCP_MAXWIN);
1065 sc->sc_wnd = win;
1066
1067 if (tcp_do_rfc1323) {
1068 /*
1069 * A timestamp received in a SYN makes
1070 * it ok to send timestamp requests and replies.
1071 */
1072 if (to->to_flags & TOF_TS) {
1073 sc->sc_tsrecent = to->to_tsval;
1074 sc->sc_flags |= SCF_TIMESTAMP;
1075 }
1076 if (to->to_flags & TOF_SCALE) {
46e92930 1077 int wscale = TCP_MIN_WINSHIFT;
984263bc
MD
1078
1079 /* Compute proper scaling value from buffer space */
1080 while (wscale < TCP_MAX_WINSHIFT &&
46e92930 1081 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) {
984263bc 1082 wscale++;
46e92930 1083 }
984263bc
MD
1084 sc->sc_request_r_scale = wscale;
1085 sc->sc_requested_s_scale = to->to_requested_s_scale;
1086 sc->sc_flags |= SCF_WINSCALE;
1087 }
1088 }
91489f6b
JH
1089 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED))
1090 sc->sc_flags |= SCF_SACK_PERMITTED;
984263bc
MD
1091 if (tp->t_flags & TF_NOOPT)
1092 sc->sc_flags = SCF_NOOPT;
b1992928
MD
1093#ifdef TCP_SIGNATURE
1094 /*
1095 * If listening socket requested TCP digests, and received SYN
1096 * contains the option, flag this in the syncache so that
1097 * syncache_respond() will do the right thing with the SYN+ACK.
1098 * XXX Currently we always record the option by default and will
1099 * attempt to use it in syncache_respond().
1100 */
1101 if (to->to_flags & TOF_SIGNATURE)
1102 sc->sc_flags = SCF_SIGNATURE;
1103#endif /* TCP_SIGNATURE */
984263bc 1104
984263bc
MD
1105 if (syncache_respond(sc, m) == 0) {
1106 syncache_insert(sc, sch);
1107 tcpstat.tcps_sndacks++;
1108 tcpstat.tcps_sndtotal++;
1109 } else {
1110 syncache_free(sc);
1111 tcpstat.tcps_sc_dropped++;
1112 }
1113 *sop = NULL;
1114 return (1);
1115}
1116
1117static int
f3f70f0d 1118syncache_respond(struct syncache *sc, struct mbuf *m)
984263bc
MD
1119{
1120 u_int8_t *optp;
1121 int optlen, error;
1122 u_int16_t tlen, hlen, mssopt;
1123 struct ip *ip = NULL;
1124 struct rtentry *rt;
1125 struct tcphdr *th;
984263bc 1126 struct ip6_hdr *ip6 = NULL;
61896e3c
JH
1127#ifdef INET6
1128 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1129#else
1130 const boolean_t isipv6 = FALSE;
984263bc
MD
1131#endif
1132
61896e3c 1133 if (isipv6) {
984263bc
MD
1134 rt = tcp_rtlookup6(&sc->sc_inc);
1135 if (rt != NULL)
1136 mssopt = rt->rt_ifp->if_mtu -
1137 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
f23061d4 1138 else
984263bc
MD
1139 mssopt = tcp_v6mssdflt;
1140 hlen = sizeof(struct ip6_hdr);
61896e3c 1141 } else {
984263bc
MD
1142 rt = tcp_rtlookup(&sc->sc_inc);
1143 if (rt != NULL)
1144 mssopt = rt->rt_ifp->if_mtu -
1145 (sizeof(struct ip) + sizeof(struct tcphdr));
f23061d4 1146 else
984263bc
MD
1147 mssopt = tcp_mssdflt;
1148 hlen = sizeof(struct ip);
1149 }
1150
1151 /* Compute the size of the TCP options. */
1152 if (sc->sc_flags & SCF_NOOPT) {
1153 optlen = 0;
1154 } else {
1155 optlen = TCPOLEN_MAXSEG +
1156 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) +
1157 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) +
91489f6b 1158 ((sc->sc_flags & SCF_SACK_PERMITTED) ?
f23061d4 1159 TCPOLEN_SACK_PERMITTED_ALIGNED : 0);
b1992928
MD
1160#ifdef TCP_SIGNATURE
1161 optlen += ((sc->sc_flags & SCF_SIGNATURE) ?
1162 (TCPOLEN_SIGNATURE + 2) : 0);
1163#endif /* TCP_SIGNATURE */
984263bc
MD
1164 }
1165 tlen = hlen + sizeof(struct tcphdr) + optlen;
1166
1167 /*
1168 * XXX
1169 * assume that the entire packet will fit in a header mbuf
1170 */
1171 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small"));
1172
1173 /*
1174 * XXX shouldn't this reuse the mbuf if possible ?
1175 * Create the IP+TCP header from scratch.
1176 */
1177 if (m)
1178 m_freem(m);
1179
74f1caca 1180 m = m_gethdr(MB_DONTWAIT, MT_HEADER);
984263bc
MD
1181 if (m == NULL)
1182 return (ENOBUFS);
1183 m->m_data += max_linkhdr;
1184 m->m_len = tlen;
1185 m->m_pkthdr.len = tlen;
1186 m->m_pkthdr.rcvif = NULL;
1187
61896e3c 1188 if (isipv6) {
984263bc
MD
1189 ip6 = mtod(m, struct ip6_hdr *);
1190 ip6->ip6_vfc = IPV6_VERSION;
1191 ip6->ip6_nxt = IPPROTO_TCP;
1192 ip6->ip6_src = sc->sc_inc.inc6_laddr;
1193 ip6->ip6_dst = sc->sc_inc.inc6_faddr;
1194 ip6->ip6_plen = htons(tlen - hlen);
1195 /* ip6_hlim is set after checksum */
1196 /* ip6_flow = ??? */
1197
1198 th = (struct tcphdr *)(ip6 + 1);
61896e3c 1199 } else {
984263bc
MD
1200 ip = mtod(m, struct ip *);
1201 ip->ip_v = IPVERSION;
1202 ip->ip_hl = sizeof(struct ip) >> 2;
1203 ip->ip_len = tlen;
1204 ip->ip_id = 0;
1205 ip->ip_off = 0;
1206 ip->ip_sum = 0;
1207 ip->ip_p = IPPROTO_TCP;
1208 ip->ip_src = sc->sc_inc.inc_laddr;
1209 ip->ip_dst = sc->sc_inc.inc_faddr;
1210 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */
1211 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */
1212
1213 /*
61896e3c
JH
1214 * See if we should do MTU discovery. Route lookups are
1215 * expensive, so we will only unset the DF bit if:
984263bc
MD
1216 *
1217 * 1) path_mtu_discovery is disabled
1218 * 2) the SCF_UNREACH flag has been set
1219 */
1220 if (path_mtu_discovery
1221 && ((sc->sc_flags & SCF_UNREACH) == 0)) {
1222 ip->ip_off |= IP_DF;
1223 }
1224
1225 th = (struct tcphdr *)(ip + 1);
1226 }
1227 th->th_sport = sc->sc_inc.inc_lport;
1228 th->th_dport = sc->sc_inc.inc_fport;
1229
1230 th->th_seq = htonl(sc->sc_iss);
1231 th->th_ack = htonl(sc->sc_irs + 1);
1232 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1233 th->th_x2 = 0;
61896e3c 1234 th->th_flags = TH_SYN | TH_ACK;
984263bc
MD
1235 th->th_win = htons(sc->sc_wnd);
1236 th->th_urp = 0;
1237
1238 /* Tack on the TCP options. */
1239 if (optlen == 0)
1240 goto no_options;
1241 optp = (u_int8_t *)(th + 1);
1242 *optp++ = TCPOPT_MAXSEG;
1243 *optp++ = TCPOLEN_MAXSEG;
1244 *optp++ = (mssopt >> 8) & 0xff;
1245 *optp++ = mssopt & 0xff;
1246
1247 if (sc->sc_flags & SCF_WINSCALE) {
1248 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
1249 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
1250 sc->sc_request_r_scale);
1251 optp += 4;
1252 }
1253
1254 if (sc->sc_flags & SCF_TIMESTAMP) {
1255 u_int32_t *lp = (u_int32_t *)(optp);
1256
1257 /* Form timestamp option as shown in appendix A of RFC 1323. */
1258 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1259 *lp++ = htonl(ticks);
1260 *lp = htonl(sc->sc_tsrecent);
1261 optp += TCPOLEN_TSTAMP_APPA;
1262 }
1263
b1992928
MD
1264#ifdef TCP_SIGNATURE
1265 /*
1266 * Handle TCP-MD5 passive opener response.
1267 */
1268 if (sc->sc_flags & SCF_SIGNATURE) {
1269 u_int8_t *bp = optp;
1270 int i;
1271
1272 *bp++ = TCPOPT_SIGNATURE;
1273 *bp++ = TCPOLEN_SIGNATURE;
1274 for (i = 0; i < TCP_SIGLEN; i++)
1275 *bp++ = 0;
1276 tcpsignature_compute(m, 0, optlen,
1277 optp + 2, IPSEC_DIR_OUTBOUND);
1278 *bp++ = TCPOPT_NOP;
1279 *bp++ = TCPOPT_EOL;
1280 optp += TCPOLEN_SIGNATURE + 2;
1281}
1282#endif /* TCP_SIGNATURE */
1283
91489f6b
JH
1284 if (sc->sc_flags & SCF_SACK_PERMITTED) {
1285 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED);
1286 optp += TCPOLEN_SACK_PERMITTED_ALIGNED;
1287 }
1288
61896e3c
JH
1289no_options:
1290 if (isipv6) {
984263bc
MD
1291 struct route_in6 *ro6 = &sc->sc_route6;
1292
1293 th->th_sum = 0;
1294 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
1295 ip6->ip6_hlim = in6_selecthlim(NULL,
1296 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL);
1297 error = ip6_output(m, NULL, ro6, 0, NULL, NULL,
1298 sc->sc_tp->t_inpcb);
61896e3c 1299 } else {
f23061d4
JH
1300 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1301 htons(tlen - hlen + IPPROTO_TCP));
984263bc
MD
1302 m->m_pkthdr.csum_flags = CSUM_TCP;
1303 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1dbb3516
SZ
1304 error = ip_output(m, sc->sc_ipopts, &sc->sc_route,
1305 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb);
984263bc
MD
1306 }
1307 return (error);
1308}
1309
1310/*
1311 * cookie layers:
1312 *
1313 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .|
1314 * | peer iss |
1315 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .|
1316 * | 0 |(A)| |
1317 * (A): peer mss index
1318 */
1319
1320/*
1321 * The values below are chosen to minimize the size of the tcp_secret
1322 * table, as well as providing roughly a 16 second lifetime for the cookie.
1323 */
1324
1325#define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */
1326#define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */
1327
1328#define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1)
1329#define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS)
1330#define SYNCOOKIE_TIMEOUT \
1331 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT))
f23061d4 1332#define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK)
984263bc
MD
1333
1334static struct {
1335 u_int32_t ts_secbits[4];
1336 u_int ts_expire;
1337} tcp_secret[SYNCOOKIE_NSECRETS];
1338
1339static int tcp_msstab[] = { 0, 536, 1460, 8960 };
1340
1341static MD5_CTX syn_ctx;
1342
1343#define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v))
1344
1345struct md5_add {
1346 u_int32_t laddr, faddr;
1347 u_int32_t secbits[4];
1348 u_int16_t lport, fport;
1349};
1350
1351#ifdef CTASSERT
1352CTASSERT(sizeof(struct md5_add) == 28);
1353#endif
1354
1355/*
1356 * Consider the problem of a recreated (and retransmitted) cookie. If the
f23061d4
JH
1357 * original SYN was accepted, the connection is established. The second
1358 * SYN is inflight, and if it arrives with an ISN that falls within the
1359 * receive window, the connection is killed.
984263bc
MD
1360 *
1361 * However, since cookies have other problems, this may not be worth
1362 * worrying about.
1363 */
1364
1365static u_int32_t
1366syncookie_generate(struct syncache *sc)
1367{
1368 u_int32_t md5_buffer[4];
1369 u_int32_t data;
1370 int idx, i;
1371 struct md5_add add;
61896e3c
JH
1372#ifdef INET6
1373 const boolean_t isipv6 = sc->sc_inc.inc_isipv6;
1374#else
1375 const boolean_t isipv6 = FALSE;
1376#endif
984263bc
MD
1377
1378 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK;
1379 if (tcp_secret[idx].ts_expire < ticks) {
1380 for (i = 0; i < 4; i++)
0ced1954 1381 tcp_secret[idx].ts_secbits[i] = karc4random();
984263bc
MD
1382 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT;
1383 }
1384 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--)
1385 if (tcp_msstab[data] <= sc->sc_peer_mss)
1386 break;
1387 data = (data << SYNCOOKIE_WNDBITS) | idx;
1388 data ^= sc->sc_irs; /* peer's iss */
1389 MD5Init(&syn_ctx);
61896e3c 1390 if (isipv6) {
984263bc
MD
1391 MD5Add(sc->sc_inc.inc6_laddr);
1392 MD5Add(sc->sc_inc.inc6_faddr);
1393 add.laddr = 0;
1394 add.faddr = 0;
61896e3c 1395 } else {
984263bc
MD
1396 add.laddr = sc->sc_inc.inc_laddr.s_addr;
1397 add.faddr = sc->sc_inc.inc_faddr.s_addr;
1398 }
1399 add.lport = sc->sc_inc.inc_lport;
1400 add.fport = sc->sc_inc.inc_fport;
1401 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1402 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1403 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1404 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1405 MD5Add(add);
1406 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1407 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK);
1408 return (data);
1409}
1410
1411static struct syncache *
f3f70f0d 1412syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so)
984263bc
MD
1413{
1414 u_int32_t md5_buffer[4];
1415 struct syncache *sc;
1416 u_int32_t data;
1417 int wnd, idx;
1418 struct md5_add add;
1419
1420 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */
1421 idx = data & SYNCOOKIE_WNDMASK;
1422 if (tcp_secret[idx].ts_expire < ticks ||
1423 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks)
1424 return (NULL);
1425 MD5Init(&syn_ctx);
1426#ifdef INET6
1427 if (inc->inc_isipv6) {
1428 MD5Add(inc->inc6_laddr);
1429 MD5Add(inc->inc6_faddr);
1430 add.laddr = 0;
1431 add.faddr = 0;
1432 } else
1433#endif
1434 {
1435 add.laddr = inc->inc_laddr.s_addr;
1436 add.faddr = inc->inc_faddr.s_addr;
1437 }
1438 add.lport = inc->inc_lport;
1439 add.fport = inc->inc_fport;
1440 add.secbits[0] = tcp_secret[idx].ts_secbits[0];
1441 add.secbits[1] = tcp_secret[idx].ts_secbits[1];
1442 add.secbits[2] = tcp_secret[idx].ts_secbits[2];
1443 add.secbits[3] = tcp_secret[idx].ts_secbits[3];
1444 MD5Add(add);
1445 MD5Final((u_char *)&md5_buffer, &syn_ctx);
1446 data ^= md5_buffer[0];
f23061d4 1447 if (data & ~SYNCOOKIE_DATAMASK)
984263bc
MD
1448 return (NULL);
1449 data = data >> SYNCOOKIE_WNDBITS;
1450
d191a96b 1451 /*
984263bc
MD
1452 * Fill in the syncache values.
1453 * XXX duplicate code from syncache_add
1454 */
9f42c129 1455 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO);
984263bc
MD
1456 sc->sc_ipopts = NULL;
1457 sc->sc_inc.inc_fport = inc->inc_fport;
1458 sc->sc_inc.inc_lport = inc->inc_lport;
1459#ifdef INET6
1460 sc->sc_inc.inc_isipv6 = inc->inc_isipv6;
1461 if (inc->inc_isipv6) {
1462 sc->sc_inc.inc6_faddr = inc->inc6_faddr;
1463 sc->sc_inc.inc6_laddr = inc->inc6_laddr;
1464 sc->sc_route6.ro_rt = NULL;
1465 } else
1466#endif
1467 {
1468 sc->sc_inc.inc_faddr = inc->inc_faddr;
1469 sc->sc_inc.inc_laddr = inc->inc_laddr;
1470 sc->sc_route.ro_rt = NULL;
1471 }
1472 sc->sc_irs = th->th_seq - 1;
1473 sc->sc_iss = th->th_ack - 1;
6d49aa6f 1474 wnd = ssb_space(&so->so_rcv);
984263bc
MD
1475 wnd = imax(wnd, 0);
1476 wnd = imin(wnd, TCP_MAXWIN);
1477 sc->sc_wnd = wnd;
1478 sc->sc_flags = 0;
1479 sc->sc_rxtslot = 0;
1480 sc->sc_peer_mss = tcp_msstab[data];
1481 return (sc);
1482}