1 /* 2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 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. 19 * 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 /* 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 * 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 * $DragonFly: src/sys/netinet/tcp_syncache.c,v 1.35 2008/11/22 11:03:35 sephe Exp $ 73 */ 74 75 #include "opt_inet6.h" 76 #include "opt_ipsec.h" 77 78 #include <sys/param.h> 79 #include <sys/systm.h> 80 #include <sys/kernel.h> 81 #include <sys/sysctl.h> 82 #include <sys/malloc.h> 83 #include <sys/mbuf.h> 84 #include <sys/md5.h> 85 #include <sys/proc.h> /* for proc0 declaration */ 86 #include <sys/random.h> 87 #include <sys/socket.h> 88 #include <sys/socketvar.h> 89 #include <sys/in_cksum.h> 90 91 #include <sys/msgport2.h> 92 #include <net/netmsg2.h> 93 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> 103 #include <netinet/ip6.h> 104 #ifdef INET6 105 #include <netinet/icmp6.h> 106 #include <netinet6/nd6.h> 107 #endif 108 #include <netinet6/ip6_var.h> 109 #include <netinet6/in6_pcb.h> 110 #include <netinet/tcp.h> 111 #include <netinet/tcp_fsm.h> 112 #include <netinet/tcp_seq.h> 113 #include <netinet/tcp_timer.h> 114 #include <netinet/tcp_timer2.h> 115 #include <netinet/tcp_var.h> 116 #include <netinet6/tcp6_var.h> 117 118 #ifdef IPSEC 119 #include <netinet6/ipsec.h> 120 #ifdef INET6 121 #include <netinet6/ipsec6.h> 122 #endif 123 #include <netproto/key/key.h> 124 #endif /*IPSEC*/ 125 126 #ifdef FAST_IPSEC 127 #include <netproto/ipsec/ipsec.h> 128 #ifdef INET6 129 #include <netproto/ipsec/ipsec6.h> 130 #endif 131 #include <netproto/ipsec/key.h> 132 #define IPSEC 133 #endif /*FAST_IPSEC*/ 134 135 static int tcp_syncookies = 1; 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 137 &tcp_syncookies, 0, 138 "Use TCP SYN cookies if the syncache overflows"); 139 140 static void syncache_drop(struct syncache *, struct syncache_head *); 141 static void syncache_free(struct syncache *); 142 static void syncache_insert(struct syncache *, struct syncache_head *); 143 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 144 static int syncache_respond(struct syncache *, struct mbuf *); 145 static struct socket *syncache_socket(struct syncache *, struct socket *, 146 struct mbuf *); 147 static void syncache_timer(void *); 148 static u_int32_t syncookie_generate(struct syncache *); 149 static struct syncache *syncookie_lookup(struct in_conninfo *, 150 struct tcphdr *, struct socket *); 151 152 /* 153 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 154 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 155 * the odds are that the user has given up attempting to connect by then. 156 */ 157 #define SYNCACHE_MAXREXMTS 3 158 159 /* Arbitrary values */ 160 #define TCP_SYNCACHE_HASHSIZE 512 161 #define TCP_SYNCACHE_BUCKETLIMIT 30 162 163 struct netmsg_sc_timer { 164 struct netmsg nm_netmsg; 165 struct msgrec *nm_mrec; /* back pointer to containing msgrec */ 166 }; 167 168 struct msgrec { 169 struct netmsg_sc_timer msg; 170 lwkt_port_t port; /* constant after init */ 171 int slot; /* constant after init */ 172 }; 173 174 static void syncache_timer_handler(netmsg_t); 175 176 struct tcp_syncache { 177 u_int hashsize; 178 u_int hashmask; 179 u_int bucket_limit; 180 u_int cache_limit; 181 u_int rexmt_limit; 182 u_int hash_secret; 183 }; 184 static struct tcp_syncache tcp_syncache; 185 186 struct tcp_syncache_percpu { 187 struct syncache_head *hashbase; 188 u_int cache_count; 189 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1]; 190 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1]; 191 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1]; 192 }; 193 static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU]; 194 195 static struct lwkt_port syncache_null_rport; 196 197 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 198 199 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD, 200 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 201 202 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD, 203 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 204 205 /* XXX JH */ 206 #if 0 207 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 208 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 209 #endif 210 211 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD, 212 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 213 214 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 215 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 216 217 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 218 219 #define SYNCACHE_HASH(inc, mask) \ 220 ((tcp_syncache.hash_secret ^ \ 221 (inc)->inc_faddr.s_addr ^ \ 222 ((inc)->inc_faddr.s_addr >> 16) ^ \ 223 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 224 225 #define SYNCACHE_HASH6(inc, mask) \ 226 ((tcp_syncache.hash_secret ^ \ 227 (inc)->inc6_faddr.s6_addr32[0] ^ \ 228 (inc)->inc6_faddr.s6_addr32[3] ^ \ 229 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 230 231 #define ENDPTS_EQ(a, b) ( \ 232 (a)->ie_fport == (b)->ie_fport && \ 233 (a)->ie_lport == (b)->ie_lport && \ 234 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 235 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 236 ) 237 238 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 239 240 static __inline void 241 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu, 242 struct syncache *sc, int slot) 243 { 244 sc->sc_rxtslot = slot; 245 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[slot]; 246 crit_enter(); 247 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq); 248 if (!callout_active(&syncache_percpu->tt_timerq[slot])) { 249 callout_reset(&syncache_percpu->tt_timerq[slot], 250 TCPTV_RTOBASE * tcp_backoff[slot], 251 syncache_timer, 252 &syncache_percpu->mrec[slot]); 253 } 254 crit_exit(); 255 } 256 257 static void 258 syncache_free(struct syncache *sc) 259 { 260 struct rtentry *rt; 261 #ifdef INET6 262 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 263 #else 264 const boolean_t isipv6 = FALSE; 265 #endif 266 267 if (sc->sc_ipopts) 268 m_free(sc->sc_ipopts); 269 270 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt; 271 if (rt != NULL) { 272 /* 273 * If this is the only reference to a protocol-cloned 274 * route, remove it immediately. 275 */ 276 if ((rt->rt_flags & RTF_WASCLONED) && rt->rt_refcnt == 1) 277 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 278 rt_mask(rt), rt->rt_flags, NULL); 279 RTFREE(rt); 280 } 281 kfree(sc, M_SYNCACHE); 282 } 283 284 void 285 syncache_init(void) 286 { 287 int i, cpu; 288 289 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 290 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 291 tcp_syncache.cache_limit = 292 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 293 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 294 tcp_syncache.hash_secret = karc4random(); 295 296 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 297 &tcp_syncache.hashsize); 298 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 299 &tcp_syncache.cache_limit); 300 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 301 &tcp_syncache.bucket_limit); 302 if (!powerof2(tcp_syncache.hashsize)) { 303 kprintf("WARNING: syncache hash size is not a power of 2.\n"); 304 tcp_syncache.hashsize = 512; /* safe default */ 305 } 306 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 307 308 lwkt_initport_replyonly_null(&syncache_null_rport); 309 310 for (cpu = 0; cpu < ncpus2; cpu++) { 311 struct tcp_syncache_percpu *syncache_percpu; 312 313 syncache_percpu = &tcp_syncache_percpu[cpu]; 314 /* Allocate the hash table. */ 315 MALLOC(syncache_percpu->hashbase, struct syncache_head *, 316 tcp_syncache.hashsize * sizeof(struct syncache_head), 317 M_SYNCACHE, M_WAITOK); 318 319 /* Initialize the hash buckets. */ 320 for (i = 0; i < tcp_syncache.hashsize; i++) { 321 struct syncache_head *bucket; 322 323 bucket = &syncache_percpu->hashbase[i]; 324 TAILQ_INIT(&bucket->sch_bucket); 325 bucket->sch_length = 0; 326 } 327 328 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 329 /* Initialize the timer queues. */ 330 TAILQ_INIT(&syncache_percpu->timerq[i]); 331 callout_init(&syncache_percpu->tt_timerq[i]); 332 333 syncache_percpu->mrec[i].slot = i; 334 syncache_percpu->mrec[i].port = tcp_cport(cpu); 335 syncache_percpu->mrec[i].msg.nm_mrec = 336 &syncache_percpu->mrec[i]; 337 netmsg_init(&syncache_percpu->mrec[i].msg.nm_netmsg, 338 NULL, &syncache_null_rport, 339 0, syncache_timer_handler); 340 } 341 } 342 } 343 344 static void 345 syncache_insert(struct syncache *sc, struct syncache_head *sch) 346 { 347 struct tcp_syncache_percpu *syncache_percpu; 348 struct syncache *sc2; 349 int i; 350 351 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 352 353 /* 354 * Make sure that we don't overflow the per-bucket 355 * limit or the total cache size limit. 356 */ 357 if (sch->sch_length >= tcp_syncache.bucket_limit) { 358 /* 359 * The bucket is full, toss the oldest element. 360 */ 361 sc2 = TAILQ_FIRST(&sch->sch_bucket); 362 sc2->sc_tp->ts_recent = ticks; 363 syncache_drop(sc2, sch); 364 tcpstat.tcps_sc_bucketoverflow++; 365 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) { 366 /* 367 * The cache is full. Toss the oldest entry in the 368 * entire cache. This is the front entry in the 369 * first non-empty timer queue with the largest 370 * timeout value. 371 */ 372 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 373 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]); 374 if (sc2 != NULL) 375 break; 376 } 377 sc2->sc_tp->ts_recent = ticks; 378 syncache_drop(sc2, NULL); 379 tcpstat.tcps_sc_cacheoverflow++; 380 } 381 382 /* Initialize the entry's timer. */ 383 syncache_timeout(syncache_percpu, sc, 0); 384 385 /* Put it into the bucket. */ 386 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 387 sch->sch_length++; 388 syncache_percpu->cache_count++; 389 tcpstat.tcps_sc_added++; 390 } 391 392 void 393 syncache_destroy(struct tcpcb *tp) 394 { 395 struct tcp_syncache_percpu *syncache_percpu; 396 struct syncache_head *bucket; 397 struct syncache *sc; 398 int i; 399 400 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 401 sc = NULL; 402 for (i = 0; i < tcp_syncache.hashsize; i++) { 403 bucket = &syncache_percpu->hashbase[i]; 404 TAILQ_FOREACH(sc, &bucket->sch_bucket, sc_hash) { 405 if (sc->sc_tp == tp) { 406 sc->sc_tp = NULL; 407 tp->t_flags &= ~TF_SYNCACHE; 408 break; 409 } 410 } 411 } 412 kprintf("Warning: delete stale syncache for tp=%p, sc=%p\n", tp, sc); 413 } 414 415 static void 416 syncache_drop(struct syncache *sc, struct syncache_head *sch) 417 { 418 struct tcp_syncache_percpu *syncache_percpu; 419 #ifdef INET6 420 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 421 #else 422 const boolean_t isipv6 = FALSE; 423 #endif 424 425 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 426 427 if (sch == NULL) { 428 if (isipv6) { 429 sch = &syncache_percpu->hashbase[ 430 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 431 } else { 432 sch = &syncache_percpu->hashbase[ 433 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 434 } 435 } 436 437 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 438 sch->sch_length--; 439 syncache_percpu->cache_count--; 440 441 /* 442 * Cleanup 443 */ 444 if (sc->sc_tp) { 445 sc->sc_tp->t_flags &= ~TF_SYNCACHE; 446 sc->sc_tp = NULL; 447 } 448 449 /* 450 * Remove the entry from the syncache timer/timeout queue. Note 451 * that we do not try to stop any running timer since we do not know 452 * whether the timer's message is in-transit or not. Since timeouts 453 * are fairly long, taking an unneeded callout does not detrimentally 454 * effect performance. 455 */ 456 crit_enter(); 457 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq); 458 crit_exit(); 459 460 syncache_free(sc); 461 } 462 463 /* 464 * Place a timeout message on the TCP thread's message queue. 465 * This routine runs in soft interrupt context. 466 * 467 * An invariant is for this routine to be called, the callout must 468 * have been active. Note that the callout is not deactivated until 469 * after the message has been processed in syncache_timer_handler() below. 470 */ 471 static void 472 syncache_timer(void *p) 473 { 474 struct netmsg_sc_timer *msg = p; 475 476 lwkt_sendmsg(msg->nm_mrec->port, &msg->nm_netmsg.nm_lmsg); 477 } 478 479 /* 480 * Service a timer message queued by timer expiration. 481 * This routine runs in the TCP protocol thread. 482 * 483 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 484 * If we have retransmitted an entry the maximum number of times, expire it. 485 * 486 * When we finish processing timed-out entries, we restart the timer if there 487 * are any entries still on the queue and deactivate it otherwise. Only after 488 * a timer has been deactivated here can it be restarted by syncache_timeout(). 489 */ 490 static void 491 syncache_timer_handler(netmsg_t netmsg) 492 { 493 struct tcp_syncache_percpu *syncache_percpu; 494 struct syncache *sc, *nsc; 495 struct inpcb *inp; 496 int slot; 497 498 slot = ((struct netmsg_sc_timer *)netmsg)->nm_mrec->slot; 499 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 500 501 crit_enter(); 502 nsc = TAILQ_FIRST(&syncache_percpu->timerq[slot]); 503 while (nsc != NULL) { 504 if (ticks < nsc->sc_rxttime) 505 break; /* finished because timerq sorted by time */ 506 sc = nsc; 507 if (sc->sc_tp == NULL) { 508 nsc = TAILQ_NEXT(sc, sc_timerq); 509 syncache_drop(sc, NULL); 510 tcpstat.tcps_sc_stale++; 511 continue; 512 } 513 inp = sc->sc_tp->t_inpcb; 514 if (slot == SYNCACHE_MAXREXMTS || 515 slot >= tcp_syncache.rexmt_limit || 516 inp->inp_gencnt != sc->sc_inp_gencnt) { 517 nsc = TAILQ_NEXT(sc, sc_timerq); 518 syncache_drop(sc, NULL); 519 tcpstat.tcps_sc_stale++; 520 continue; 521 } 522 /* 523 * syncache_respond() may call back into the syncache to 524 * to modify another entry, so do not obtain the next 525 * entry on the timer chain until it has completed. 526 */ 527 syncache_respond(sc, NULL); 528 nsc = TAILQ_NEXT(sc, sc_timerq); 529 tcpstat.tcps_sc_retransmitted++; 530 TAILQ_REMOVE(&syncache_percpu->timerq[slot], sc, sc_timerq); 531 syncache_timeout(syncache_percpu, sc, slot + 1); 532 } 533 if (nsc != NULL) 534 callout_reset(&syncache_percpu->tt_timerq[slot], 535 nsc->sc_rxttime - ticks, syncache_timer, 536 &syncache_percpu->mrec[slot]); 537 else 538 callout_deactivate(&syncache_percpu->tt_timerq[slot]); 539 crit_exit(); 540 541 lwkt_replymsg(&netmsg->nm_lmsg, 0); 542 } 543 544 /* 545 * Find an entry in the syncache. 546 */ 547 struct syncache * 548 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 549 { 550 struct tcp_syncache_percpu *syncache_percpu; 551 struct syncache *sc; 552 struct syncache_head *sch; 553 554 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 555 #ifdef INET6 556 if (inc->inc_isipv6) { 557 sch = &syncache_percpu->hashbase[ 558 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 559 *schp = sch; 560 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) 561 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 562 return (sc); 563 } else 564 #endif 565 { 566 sch = &syncache_percpu->hashbase[ 567 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 568 *schp = sch; 569 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 570 #ifdef INET6 571 if (sc->sc_inc.inc_isipv6) 572 continue; 573 #endif 574 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 575 return (sc); 576 } 577 } 578 return (NULL); 579 } 580 581 /* 582 * This function is called when we get a RST for a 583 * non-existent connection, so that we can see if the 584 * connection is in the syn cache. If it is, zap it. 585 */ 586 void 587 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 588 { 589 struct syncache *sc; 590 struct syncache_head *sch; 591 592 sc = syncache_lookup(inc, &sch); 593 if (sc == NULL) 594 return; 595 /* 596 * If the RST bit is set, check the sequence number to see 597 * if this is a valid reset segment. 598 * RFC 793 page 37: 599 * In all states except SYN-SENT, all reset (RST) segments 600 * are validated by checking their SEQ-fields. A reset is 601 * valid if its sequence number is in the window. 602 * 603 * The sequence number in the reset segment is normally an 604 * echo of our outgoing acknowlegement numbers, but some hosts 605 * send a reset with the sequence number at the rightmost edge 606 * of our receive window, and we have to handle this case. 607 */ 608 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 609 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 610 syncache_drop(sc, sch); 611 tcpstat.tcps_sc_reset++; 612 } 613 } 614 615 void 616 syncache_badack(struct in_conninfo *inc) 617 { 618 struct syncache *sc; 619 struct syncache_head *sch; 620 621 sc = syncache_lookup(inc, &sch); 622 if (sc != NULL) { 623 syncache_drop(sc, sch); 624 tcpstat.tcps_sc_badack++; 625 } 626 } 627 628 void 629 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 630 { 631 struct syncache *sc; 632 struct syncache_head *sch; 633 634 /* we are called at splnet() here */ 635 sc = syncache_lookup(inc, &sch); 636 if (sc == NULL) 637 return; 638 639 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 640 if (ntohl(th->th_seq) != sc->sc_iss) 641 return; 642 643 /* 644 * If we've rertransmitted 3 times and this is our second error, 645 * we remove the entry. Otherwise, we allow it to continue on. 646 * This prevents us from incorrectly nuking an entry during a 647 * spurious network outage. 648 * 649 * See tcp_notify(). 650 */ 651 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 652 sc->sc_flags |= SCF_UNREACH; 653 return; 654 } 655 syncache_drop(sc, sch); 656 tcpstat.tcps_sc_unreach++; 657 } 658 659 /* 660 * Build a new TCP socket structure from a syncache entry. 661 * 662 * This is called from the context of the SYN+ACK 663 */ 664 static struct socket * 665 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 666 { 667 struct inpcb *inp = NULL, *linp; 668 struct socket *so; 669 struct tcpcb *tp; 670 lwkt_port_t port; 671 #ifdef INET6 672 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 673 #else 674 const boolean_t isipv6 = FALSE; 675 #endif 676 677 /* 678 * Ok, create the full blown connection, and set things up 679 * as they would have been set up if we had created the 680 * connection when the SYN arrived. If we can't create 681 * the connection, abort it. 682 */ 683 so = sonewconn(lso, SS_ISCONNECTED); 684 if (so == NULL) { 685 /* 686 * Drop the connection; we will send a RST if the peer 687 * retransmits the ACK, 688 */ 689 tcpstat.tcps_listendrop++; 690 goto abort; 691 } 692 693 /* 694 * Set the protocol processing port for the socket to the current 695 * port (that the connection came in on). 696 */ 697 sosetport(so, &curthread->td_msgport); 698 699 /* 700 * Insert new socket into hash list. 701 */ 702 inp = so->so_pcb; 703 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 704 if (isipv6) { 705 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 706 } else { 707 #ifdef INET6 708 inp->inp_vflag &= ~INP_IPV6; 709 inp->inp_vflag |= INP_IPV4; 710 inp->inp_flags &= ~IN6P_IPV6_V6ONLY; 711 #endif 712 inp->inp_laddr = sc->sc_inc.inc_laddr; 713 } 714 inp->inp_lport = sc->sc_inc.inc_lport; 715 if (in_pcbinsporthash(inp) != 0) { 716 /* 717 * Undo the assignments above if we failed to 718 * put the PCB on the hash lists. 719 */ 720 if (isipv6) 721 inp->in6p_laddr = kin6addr_any; 722 else 723 inp->inp_laddr.s_addr = INADDR_ANY; 724 inp->inp_lport = 0; 725 goto abort; 726 } 727 linp = so->so_pcb; 728 #ifdef IPSEC 729 /* copy old policy into new socket's */ 730 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 731 kprintf("syncache_expand: could not copy policy\n"); 732 #endif 733 if (isipv6) { 734 struct in6_addr laddr6; 735 struct sockaddr_in6 sin6; 736 /* 737 * Inherit socket options from the listening socket. 738 * Note that in6p_inputopts are not (and should not be) 739 * copied, since it stores previously received options and is 740 * used to detect if each new option is different than the 741 * previous one and hence should be passed to a user. 742 * If we copied in6p_inputopts, a user would not be able to 743 * receive options just after calling the accept system call. 744 */ 745 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 746 if (linp->in6p_outputopts) 747 inp->in6p_outputopts = 748 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 749 inp->in6p_route = sc->sc_route6; 750 sc->sc_route6.ro_rt = NULL; 751 752 sin6.sin6_family = AF_INET6; 753 sin6.sin6_len = sizeof sin6; 754 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 755 sin6.sin6_port = sc->sc_inc.inc_fport; 756 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 757 laddr6 = inp->in6p_laddr; 758 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 759 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 760 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) { 761 inp->in6p_laddr = laddr6; 762 goto abort; 763 } 764 } else { 765 struct in_addr laddr; 766 struct sockaddr_in sin; 767 768 inp->inp_options = ip_srcroute(m); 769 if (inp->inp_options == NULL) { 770 inp->inp_options = sc->sc_ipopts; 771 sc->sc_ipopts = NULL; 772 } 773 inp->inp_route = sc->sc_route; 774 sc->sc_route.ro_rt = NULL; 775 776 sin.sin_family = AF_INET; 777 sin.sin_len = sizeof sin; 778 sin.sin_addr = sc->sc_inc.inc_faddr; 779 sin.sin_port = sc->sc_inc.inc_fport; 780 bzero(sin.sin_zero, sizeof sin.sin_zero); 781 laddr = inp->inp_laddr; 782 if (inp->inp_laddr.s_addr == INADDR_ANY) 783 inp->inp_laddr = sc->sc_inc.inc_laddr; 784 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) { 785 inp->inp_laddr = laddr; 786 goto abort; 787 } 788 } 789 790 /* 791 * The current port should be in the context of the SYN+ACK and 792 * so should match the tcp address port. 793 * 794 * XXX we may be running on the netisr thread instead of a tcp 795 * thread, in which case port will not match 796 * curthread->td_msgport. 797 */ 798 if (isipv6) { 799 port = tcp6_addrport(); 800 } else { 801 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport, 802 inp->inp_laddr.s_addr, inp->inp_lport); 803 } 804 /*KKASSERT(port == &curthread->td_msgport);*/ 805 806 tp = intotcpcb(inp); 807 tp->t_state = TCPS_SYN_RECEIVED; 808 tp->iss = sc->sc_iss; 809 tp->irs = sc->sc_irs; 810 tcp_rcvseqinit(tp); 811 tcp_sendseqinit(tp); 812 tp->snd_wl1 = sc->sc_irs; 813 tp->rcv_up = sc->sc_irs + 1; 814 tp->rcv_wnd = sc->sc_wnd; 815 tp->rcv_adv += tp->rcv_wnd; 816 817 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 818 if (sc->sc_flags & SCF_NOOPT) 819 tp->t_flags |= TF_NOOPT; 820 if (sc->sc_flags & SCF_WINSCALE) { 821 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 822 tp->requested_s_scale = sc->sc_requested_s_scale; 823 tp->request_r_scale = sc->sc_request_r_scale; 824 } 825 if (sc->sc_flags & SCF_TIMESTAMP) { 826 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 827 tp->ts_recent = sc->sc_tsrecent; 828 tp->ts_recent_age = ticks; 829 } 830 if (sc->sc_flags & SCF_SACK_PERMITTED) 831 tp->t_flags |= TF_SACK_PERMITTED; 832 833 tcp_mss(tp, sc->sc_peer_mss); 834 835 /* 836 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 837 */ 838 if (sc->sc_rxtslot != 0) 839 tp->snd_cwnd = tp->t_maxseg; 840 tcp_create_timermsg(tp, port); 841 tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep); 842 843 tcpstat.tcps_accepts++; 844 return (so); 845 846 abort: 847 if (so != NULL) 848 soabort_oncpu(so); 849 return (NULL); 850 } 851 852 /* 853 * This function gets called when we receive an ACK for a 854 * socket in the LISTEN state. We look up the connection 855 * in the syncache, and if its there, we pull it out of 856 * the cache and turn it into a full-blown connection in 857 * the SYN-RECEIVED state. 858 */ 859 int 860 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop, 861 struct mbuf *m) 862 { 863 struct syncache *sc; 864 struct syncache_head *sch; 865 struct socket *so; 866 867 sc = syncache_lookup(inc, &sch); 868 if (sc == NULL) { 869 /* 870 * There is no syncache entry, so see if this ACK is 871 * a returning syncookie. To do this, first: 872 * A. See if this socket has had a syncache entry dropped in 873 * the past. We don't want to accept a bogus syncookie 874 * if we've never received a SYN. 875 * B. check that the syncookie is valid. If it is, then 876 * cobble up a fake syncache entry, and return. 877 */ 878 if (!tcp_syncookies) 879 return (0); 880 sc = syncookie_lookup(inc, th, *sop); 881 if (sc == NULL) 882 return (0); 883 sch = NULL; 884 tcpstat.tcps_sc_recvcookie++; 885 } 886 887 /* 888 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 889 */ 890 if (th->th_ack != sc->sc_iss + 1) 891 return (0); 892 893 so = syncache_socket(sc, *sop, m); 894 if (so == NULL) { 895 #if 0 896 resetandabort: 897 /* XXXjlemon check this - is this correct? */ 898 tcp_respond(NULL, m, m, th, 899 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 900 #endif 901 m_freem(m); /* XXX only needed for above */ 902 tcpstat.tcps_sc_aborted++; 903 } else { 904 tcpstat.tcps_sc_completed++; 905 } 906 if (sch == NULL) 907 syncache_free(sc); 908 else 909 syncache_drop(sc, sch); 910 *sop = so; 911 return (1); 912 } 913 914 /* 915 * Given a LISTEN socket and an inbound SYN request, add 916 * this to the syn cache, and send back a segment: 917 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 918 * to the source. 919 * 920 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 921 * Doing so would require that we hold onto the data and deliver it 922 * to the application. However, if we are the target of a SYN-flood 923 * DoS attack, an attacker could send data which would eventually 924 * consume all available buffer space if it were ACKed. By not ACKing 925 * the data, we avoid this DoS scenario. 926 */ 927 int 928 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 929 struct socket **sop, struct mbuf *m) 930 { 931 struct tcp_syncache_percpu *syncache_percpu; 932 struct tcpcb *tp; 933 struct socket *so; 934 struct syncache *sc = NULL; 935 struct syncache_head *sch; 936 struct mbuf *ipopts = NULL; 937 int win; 938 939 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 940 so = *sop; 941 tp = sototcpcb(so); 942 943 /* 944 * Remember the IP options, if any. 945 */ 946 #ifdef INET6 947 if (!inc->inc_isipv6) 948 #endif 949 ipopts = ip_srcroute(m); 950 951 /* 952 * See if we already have an entry for this connection. 953 * If we do, resend the SYN,ACK, and reset the retransmit timer. 954 * 955 * XXX 956 * The syncache should be re-initialized with the contents 957 * of the new SYN which may have different options. 958 */ 959 sc = syncache_lookup(inc, &sch); 960 if (sc != NULL) { 961 tcpstat.tcps_sc_dupsyn++; 962 if (ipopts) { 963 /* 964 * If we were remembering a previous source route, 965 * forget it and use the new one we've been given. 966 */ 967 if (sc->sc_ipopts) 968 m_free(sc->sc_ipopts); 969 sc->sc_ipopts = ipopts; 970 } 971 /* 972 * Update timestamp if present. 973 */ 974 if (sc->sc_flags & SCF_TIMESTAMP) 975 sc->sc_tsrecent = to->to_tsval; 976 977 /* Just update the TOF_SACK_PERMITTED for now. */ 978 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 979 sc->sc_flags |= SCF_SACK_PERMITTED; 980 else 981 sc->sc_flags &= ~SCF_SACK_PERMITTED; 982 983 /* 984 * PCB may have changed, pick up new values. 985 */ 986 if (sc->sc_tp) { 987 sc->sc_tp->t_flags &= ~TF_SYNCACHE; 988 tp->t_flags |= TF_SYNCACHE; 989 } 990 sc->sc_tp = tp; 991 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 992 if (syncache_respond(sc, m) == 0) { 993 crit_enter(); 994 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 995 sc, sc_timerq); 996 crit_exit(); 997 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 998 tcpstat.tcps_sndacks++; 999 tcpstat.tcps_sndtotal++; 1000 } 1001 *sop = NULL; 1002 return (1); 1003 } 1004 1005 /* 1006 * Fill in the syncache values. 1007 */ 1008 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1009 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1010 sc->sc_ipopts = ipopts; 1011 sc->sc_inc.inc_fport = inc->inc_fport; 1012 sc->sc_inc.inc_lport = inc->inc_lport; 1013 sc->sc_tp = tp; 1014 tp->t_flags |= TF_SYNCACHE; 1015 #ifdef INET6 1016 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1017 if (inc->inc_isipv6) { 1018 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1019 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1020 sc->sc_route6.ro_rt = NULL; 1021 } else 1022 #endif 1023 { 1024 sc->sc_inc.inc_faddr = inc->inc_faddr; 1025 sc->sc_inc.inc_laddr = inc->inc_laddr; 1026 sc->sc_route.ro_rt = NULL; 1027 } 1028 sc->sc_irs = th->th_seq; 1029 sc->sc_flags = 0; 1030 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 1031 if (tcp_syncookies) 1032 sc->sc_iss = syncookie_generate(sc); 1033 else 1034 sc->sc_iss = karc4random(); 1035 1036 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */ 1037 win = ssb_space(&so->so_rcv); 1038 win = imax(win, 0); 1039 win = imin(win, TCP_MAXWIN); 1040 sc->sc_wnd = win; 1041 1042 if (tcp_do_rfc1323) { 1043 /* 1044 * A timestamp received in a SYN makes 1045 * it ok to send timestamp requests and replies. 1046 */ 1047 if (to->to_flags & TOF_TS) { 1048 sc->sc_tsrecent = to->to_tsval; 1049 sc->sc_flags |= SCF_TIMESTAMP; 1050 } 1051 if (to->to_flags & TOF_SCALE) { 1052 int wscale = TCP_MIN_WINSHIFT; 1053 1054 /* Compute proper scaling value from buffer space */ 1055 while (wscale < TCP_MAX_WINSHIFT && 1056 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) { 1057 wscale++; 1058 } 1059 sc->sc_request_r_scale = wscale; 1060 sc->sc_requested_s_scale = to->to_requested_s_scale; 1061 sc->sc_flags |= SCF_WINSCALE; 1062 } 1063 } 1064 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1065 sc->sc_flags |= SCF_SACK_PERMITTED; 1066 if (tp->t_flags & TF_NOOPT) 1067 sc->sc_flags = SCF_NOOPT; 1068 1069 if (syncache_respond(sc, m) == 0) { 1070 syncache_insert(sc, sch); 1071 tcpstat.tcps_sndacks++; 1072 tcpstat.tcps_sndtotal++; 1073 } else { 1074 syncache_free(sc); 1075 tcpstat.tcps_sc_dropped++; 1076 } 1077 *sop = NULL; 1078 return (1); 1079 } 1080 1081 static int 1082 syncache_respond(struct syncache *sc, struct mbuf *m) 1083 { 1084 u_int8_t *optp; 1085 int optlen, error; 1086 u_int16_t tlen, hlen, mssopt; 1087 struct ip *ip = NULL; 1088 struct rtentry *rt; 1089 struct tcphdr *th; 1090 struct ip6_hdr *ip6 = NULL; 1091 #ifdef INET6 1092 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1093 #else 1094 const boolean_t isipv6 = FALSE; 1095 #endif 1096 1097 if (isipv6) { 1098 rt = tcp_rtlookup6(&sc->sc_inc); 1099 if (rt != NULL) 1100 mssopt = rt->rt_ifp->if_mtu - 1101 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1102 else 1103 mssopt = tcp_v6mssdflt; 1104 hlen = sizeof(struct ip6_hdr); 1105 } else { 1106 rt = tcp_rtlookup(&sc->sc_inc); 1107 if (rt != NULL) 1108 mssopt = rt->rt_ifp->if_mtu - 1109 (sizeof(struct ip) + sizeof(struct tcphdr)); 1110 else 1111 mssopt = tcp_mssdflt; 1112 hlen = sizeof(struct ip); 1113 } 1114 1115 /* Compute the size of the TCP options. */ 1116 if (sc->sc_flags & SCF_NOOPT) { 1117 optlen = 0; 1118 } else { 1119 optlen = TCPOLEN_MAXSEG + 1120 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1121 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1122 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1123 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1124 } 1125 tlen = hlen + sizeof(struct tcphdr) + optlen; 1126 1127 /* 1128 * XXX 1129 * assume that the entire packet will fit in a header mbuf 1130 */ 1131 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1132 1133 /* 1134 * XXX shouldn't this reuse the mbuf if possible ? 1135 * Create the IP+TCP header from scratch. 1136 */ 1137 if (m) 1138 m_freem(m); 1139 1140 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 1141 if (m == NULL) 1142 return (ENOBUFS); 1143 m->m_data += max_linkhdr; 1144 m->m_len = tlen; 1145 m->m_pkthdr.len = tlen; 1146 m->m_pkthdr.rcvif = NULL; 1147 1148 if (isipv6) { 1149 ip6 = mtod(m, struct ip6_hdr *); 1150 ip6->ip6_vfc = IPV6_VERSION; 1151 ip6->ip6_nxt = IPPROTO_TCP; 1152 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1153 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1154 ip6->ip6_plen = htons(tlen - hlen); 1155 /* ip6_hlim is set after checksum */ 1156 /* ip6_flow = ??? */ 1157 1158 th = (struct tcphdr *)(ip6 + 1); 1159 } else { 1160 ip = mtod(m, struct ip *); 1161 ip->ip_v = IPVERSION; 1162 ip->ip_hl = sizeof(struct ip) >> 2; 1163 ip->ip_len = tlen; 1164 ip->ip_id = 0; 1165 ip->ip_off = 0; 1166 ip->ip_sum = 0; 1167 ip->ip_p = IPPROTO_TCP; 1168 ip->ip_src = sc->sc_inc.inc_laddr; 1169 ip->ip_dst = sc->sc_inc.inc_faddr; 1170 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1171 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1172 1173 /* 1174 * See if we should do MTU discovery. Route lookups are 1175 * expensive, so we will only unset the DF bit if: 1176 * 1177 * 1) path_mtu_discovery is disabled 1178 * 2) the SCF_UNREACH flag has been set 1179 */ 1180 if (path_mtu_discovery 1181 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1182 ip->ip_off |= IP_DF; 1183 } 1184 1185 th = (struct tcphdr *)(ip + 1); 1186 } 1187 th->th_sport = sc->sc_inc.inc_lport; 1188 th->th_dport = sc->sc_inc.inc_fport; 1189 1190 th->th_seq = htonl(sc->sc_iss); 1191 th->th_ack = htonl(sc->sc_irs + 1); 1192 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1193 th->th_x2 = 0; 1194 th->th_flags = TH_SYN | TH_ACK; 1195 th->th_win = htons(sc->sc_wnd); 1196 th->th_urp = 0; 1197 1198 /* Tack on the TCP options. */ 1199 if (optlen == 0) 1200 goto no_options; 1201 optp = (u_int8_t *)(th + 1); 1202 *optp++ = TCPOPT_MAXSEG; 1203 *optp++ = TCPOLEN_MAXSEG; 1204 *optp++ = (mssopt >> 8) & 0xff; 1205 *optp++ = mssopt & 0xff; 1206 1207 if (sc->sc_flags & SCF_WINSCALE) { 1208 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1209 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1210 sc->sc_request_r_scale); 1211 optp += 4; 1212 } 1213 1214 if (sc->sc_flags & SCF_TIMESTAMP) { 1215 u_int32_t *lp = (u_int32_t *)(optp); 1216 1217 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1218 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1219 *lp++ = htonl(ticks); 1220 *lp = htonl(sc->sc_tsrecent); 1221 optp += TCPOLEN_TSTAMP_APPA; 1222 } 1223 1224 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1225 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1226 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1227 } 1228 1229 no_options: 1230 if (isipv6) { 1231 struct route_in6 *ro6 = &sc->sc_route6; 1232 1233 th->th_sum = 0; 1234 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1235 ip6->ip6_hlim = in6_selecthlim(NULL, 1236 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1237 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1238 sc->sc_tp->t_inpcb); 1239 } else { 1240 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1241 htons(tlen - hlen + IPPROTO_TCP)); 1242 m->m_pkthdr.csum_flags = CSUM_TCP; 1243 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1244 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 1245 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb); 1246 } 1247 return (error); 1248 } 1249 1250 /* 1251 * cookie layers: 1252 * 1253 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1254 * | peer iss | 1255 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1256 * | 0 |(A)| | 1257 * (A): peer mss index 1258 */ 1259 1260 /* 1261 * The values below are chosen to minimize the size of the tcp_secret 1262 * table, as well as providing roughly a 16 second lifetime for the cookie. 1263 */ 1264 1265 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1266 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1267 1268 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1269 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1270 #define SYNCOOKIE_TIMEOUT \ 1271 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1272 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1273 1274 static struct { 1275 u_int32_t ts_secbits[4]; 1276 u_int ts_expire; 1277 } tcp_secret[SYNCOOKIE_NSECRETS]; 1278 1279 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1280 1281 static MD5_CTX syn_ctx; 1282 1283 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1284 1285 struct md5_add { 1286 u_int32_t laddr, faddr; 1287 u_int32_t secbits[4]; 1288 u_int16_t lport, fport; 1289 }; 1290 1291 #ifdef CTASSERT 1292 CTASSERT(sizeof(struct md5_add) == 28); 1293 #endif 1294 1295 /* 1296 * Consider the problem of a recreated (and retransmitted) cookie. If the 1297 * original SYN was accepted, the connection is established. The second 1298 * SYN is inflight, and if it arrives with an ISN that falls within the 1299 * receive window, the connection is killed. 1300 * 1301 * However, since cookies have other problems, this may not be worth 1302 * worrying about. 1303 */ 1304 1305 static u_int32_t 1306 syncookie_generate(struct syncache *sc) 1307 { 1308 u_int32_t md5_buffer[4]; 1309 u_int32_t data; 1310 int idx, i; 1311 struct md5_add add; 1312 #ifdef INET6 1313 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1314 #else 1315 const boolean_t isipv6 = FALSE; 1316 #endif 1317 1318 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1319 if (tcp_secret[idx].ts_expire < ticks) { 1320 for (i = 0; i < 4; i++) 1321 tcp_secret[idx].ts_secbits[i] = karc4random(); 1322 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1323 } 1324 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1325 if (tcp_msstab[data] <= sc->sc_peer_mss) 1326 break; 1327 data = (data << SYNCOOKIE_WNDBITS) | idx; 1328 data ^= sc->sc_irs; /* peer's iss */ 1329 MD5Init(&syn_ctx); 1330 if (isipv6) { 1331 MD5Add(sc->sc_inc.inc6_laddr); 1332 MD5Add(sc->sc_inc.inc6_faddr); 1333 add.laddr = 0; 1334 add.faddr = 0; 1335 } else { 1336 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1337 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1338 } 1339 add.lport = sc->sc_inc.inc_lport; 1340 add.fport = sc->sc_inc.inc_fport; 1341 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1342 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1343 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1344 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1345 MD5Add(add); 1346 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1347 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1348 return (data); 1349 } 1350 1351 static struct syncache * 1352 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so) 1353 { 1354 u_int32_t md5_buffer[4]; 1355 struct syncache *sc; 1356 u_int32_t data; 1357 int wnd, idx; 1358 struct md5_add add; 1359 1360 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1361 idx = data & SYNCOOKIE_WNDMASK; 1362 if (tcp_secret[idx].ts_expire < ticks || 1363 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1364 return (NULL); 1365 MD5Init(&syn_ctx); 1366 #ifdef INET6 1367 if (inc->inc_isipv6) { 1368 MD5Add(inc->inc6_laddr); 1369 MD5Add(inc->inc6_faddr); 1370 add.laddr = 0; 1371 add.faddr = 0; 1372 } else 1373 #endif 1374 { 1375 add.laddr = inc->inc_laddr.s_addr; 1376 add.faddr = inc->inc_faddr.s_addr; 1377 } 1378 add.lport = inc->inc_lport; 1379 add.fport = inc->inc_fport; 1380 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1381 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1382 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1383 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1384 MD5Add(add); 1385 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1386 data ^= md5_buffer[0]; 1387 if (data & ~SYNCOOKIE_DATAMASK) 1388 return (NULL); 1389 data = data >> SYNCOOKIE_WNDBITS; 1390 1391 /* 1392 * Fill in the syncache values. 1393 * XXX duplicate code from syncache_add 1394 */ 1395 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1396 sc->sc_ipopts = NULL; 1397 sc->sc_inc.inc_fport = inc->inc_fport; 1398 sc->sc_inc.inc_lport = inc->inc_lport; 1399 #ifdef INET6 1400 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1401 if (inc->inc_isipv6) { 1402 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1403 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1404 sc->sc_route6.ro_rt = NULL; 1405 } else 1406 #endif 1407 { 1408 sc->sc_inc.inc_faddr = inc->inc_faddr; 1409 sc->sc_inc.inc_laddr = inc->inc_laddr; 1410 sc->sc_route.ro_rt = NULL; 1411 } 1412 sc->sc_irs = th->th_seq - 1; 1413 sc->sc_iss = th->th_ack - 1; 1414 wnd = ssb_space(&so->so_rcv); 1415 wnd = imax(wnd, 0); 1416 wnd = imin(wnd, TCP_MAXWIN); 1417 sc->sc_wnd = wnd; 1418 sc->sc_flags = 0; 1419 sc->sc_rxtslot = 0; 1420 sc->sc_peer_mss = tcp_msstab[data]; 1421 return (sc); 1422 } 1423