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 == NULL || 517 inp->inp_gencnt != sc->sc_inp_gencnt) { 518 nsc = TAILQ_NEXT(sc, sc_timerq); 519 syncache_drop(sc, NULL); 520 tcpstat.tcps_sc_stale++; 521 continue; 522 } 523 /* 524 * syncache_respond() may call back into the syncache to 525 * to modify another entry, so do not obtain the next 526 * entry on the timer chain until it has completed. 527 */ 528 syncache_respond(sc, NULL); 529 nsc = TAILQ_NEXT(sc, sc_timerq); 530 tcpstat.tcps_sc_retransmitted++; 531 TAILQ_REMOVE(&syncache_percpu->timerq[slot], sc, sc_timerq); 532 syncache_timeout(syncache_percpu, sc, slot + 1); 533 } 534 if (nsc != NULL) 535 callout_reset(&syncache_percpu->tt_timerq[slot], 536 nsc->sc_rxttime - ticks, syncache_timer, 537 &syncache_percpu->mrec[slot]); 538 else 539 callout_deactivate(&syncache_percpu->tt_timerq[slot]); 540 crit_exit(); 541 542 lwkt_replymsg(&netmsg->nm_lmsg, 0); 543 } 544 545 /* 546 * Find an entry in the syncache. 547 */ 548 struct syncache * 549 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 550 { 551 struct tcp_syncache_percpu *syncache_percpu; 552 struct syncache *sc; 553 struct syncache_head *sch; 554 555 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 556 #ifdef INET6 557 if (inc->inc_isipv6) { 558 sch = &syncache_percpu->hashbase[ 559 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 560 *schp = sch; 561 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) 562 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 563 return (sc); 564 } else 565 #endif 566 { 567 sch = &syncache_percpu->hashbase[ 568 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 569 *schp = sch; 570 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 571 #ifdef INET6 572 if (sc->sc_inc.inc_isipv6) 573 continue; 574 #endif 575 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 576 return (sc); 577 } 578 } 579 return (NULL); 580 } 581 582 /* 583 * This function is called when we get a RST for a 584 * non-existent connection, so that we can see if the 585 * connection is in the syn cache. If it is, zap it. 586 */ 587 void 588 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 589 { 590 struct syncache *sc; 591 struct syncache_head *sch; 592 593 sc = syncache_lookup(inc, &sch); 594 if (sc == NULL) 595 return; 596 /* 597 * If the RST bit is set, check the sequence number to see 598 * if this is a valid reset segment. 599 * RFC 793 page 37: 600 * In all states except SYN-SENT, all reset (RST) segments 601 * are validated by checking their SEQ-fields. A reset is 602 * valid if its sequence number is in the window. 603 * 604 * The sequence number in the reset segment is normally an 605 * echo of our outgoing acknowlegement numbers, but some hosts 606 * send a reset with the sequence number at the rightmost edge 607 * of our receive window, and we have to handle this case. 608 */ 609 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 610 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 611 syncache_drop(sc, sch); 612 tcpstat.tcps_sc_reset++; 613 } 614 } 615 616 void 617 syncache_badack(struct in_conninfo *inc) 618 { 619 struct syncache *sc; 620 struct syncache_head *sch; 621 622 sc = syncache_lookup(inc, &sch); 623 if (sc != NULL) { 624 syncache_drop(sc, sch); 625 tcpstat.tcps_sc_badack++; 626 } 627 } 628 629 void 630 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 631 { 632 struct syncache *sc; 633 struct syncache_head *sch; 634 635 /* we are called at splnet() here */ 636 sc = syncache_lookup(inc, &sch); 637 if (sc == NULL) 638 return; 639 640 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 641 if (ntohl(th->th_seq) != sc->sc_iss) 642 return; 643 644 /* 645 * If we've rertransmitted 3 times and this is our second error, 646 * we remove the entry. Otherwise, we allow it to continue on. 647 * This prevents us from incorrectly nuking an entry during a 648 * spurious network outage. 649 * 650 * See tcp_notify(). 651 */ 652 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 653 sc->sc_flags |= SCF_UNREACH; 654 return; 655 } 656 syncache_drop(sc, sch); 657 tcpstat.tcps_sc_unreach++; 658 } 659 660 /* 661 * Build a new TCP socket structure from a syncache entry. 662 * 663 * This is called from the context of the SYN+ACK 664 */ 665 static struct socket * 666 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 667 { 668 struct inpcb *inp = NULL, *linp; 669 struct socket *so; 670 struct tcpcb *tp; 671 lwkt_port_t port; 672 #ifdef INET6 673 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 674 #else 675 const boolean_t isipv6 = FALSE; 676 #endif 677 678 /* 679 * Ok, create the full blown connection, and set things up 680 * as they would have been set up if we had created the 681 * connection when the SYN arrived. If we can't create 682 * the connection, abort it. 683 */ 684 so = sonewconn(lso, SS_ISCONNECTED); 685 if (so == NULL) { 686 /* 687 * Drop the connection; we will send a RST if the peer 688 * retransmits the ACK, 689 */ 690 tcpstat.tcps_listendrop++; 691 goto abort; 692 } 693 694 /* 695 * Set the protocol processing port for the socket to the current 696 * port (that the connection came in on). 697 */ 698 sosetport(so, &curthread->td_msgport); 699 700 /* 701 * Insert new socket into hash list. 702 */ 703 inp = so->so_pcb; 704 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 705 if (isipv6) { 706 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 707 } else { 708 #ifdef INET6 709 inp->inp_vflag &= ~INP_IPV6; 710 inp->inp_vflag |= INP_IPV4; 711 inp->inp_flags &= ~IN6P_IPV6_V6ONLY; 712 #endif 713 inp->inp_laddr = sc->sc_inc.inc_laddr; 714 } 715 inp->inp_lport = sc->sc_inc.inc_lport; 716 if (in_pcbinsporthash(inp) != 0) { 717 /* 718 * Undo the assignments above if we failed to 719 * put the PCB on the hash lists. 720 */ 721 if (isipv6) 722 inp->in6p_laddr = kin6addr_any; 723 else 724 inp->inp_laddr.s_addr = INADDR_ANY; 725 inp->inp_lport = 0; 726 goto abort; 727 } 728 linp = so->so_pcb; 729 #ifdef IPSEC 730 /* copy old policy into new socket's */ 731 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 732 kprintf("syncache_expand: could not copy policy\n"); 733 #endif 734 if (isipv6) { 735 struct in6_addr laddr6; 736 struct sockaddr_in6 sin6; 737 /* 738 * Inherit socket options from the listening socket. 739 * Note that in6p_inputopts are not (and should not be) 740 * copied, since it stores previously received options and is 741 * used to detect if each new option is different than the 742 * previous one and hence should be passed to a user. 743 * If we copied in6p_inputopts, a user would not be able to 744 * receive options just after calling the accept system call. 745 */ 746 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 747 if (linp->in6p_outputopts) 748 inp->in6p_outputopts = 749 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 750 inp->in6p_route = sc->sc_route6; 751 sc->sc_route6.ro_rt = NULL; 752 753 sin6.sin6_family = AF_INET6; 754 sin6.sin6_len = sizeof sin6; 755 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 756 sin6.sin6_port = sc->sc_inc.inc_fport; 757 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 758 laddr6 = inp->in6p_laddr; 759 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 760 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 761 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) { 762 inp->in6p_laddr = laddr6; 763 goto abort; 764 } 765 } else { 766 struct in_addr laddr; 767 struct sockaddr_in sin; 768 769 inp->inp_options = ip_srcroute(m); 770 if (inp->inp_options == NULL) { 771 inp->inp_options = sc->sc_ipopts; 772 sc->sc_ipopts = NULL; 773 } 774 inp->inp_route = sc->sc_route; 775 sc->sc_route.ro_rt = NULL; 776 777 sin.sin_family = AF_INET; 778 sin.sin_len = sizeof sin; 779 sin.sin_addr = sc->sc_inc.inc_faddr; 780 sin.sin_port = sc->sc_inc.inc_fport; 781 bzero(sin.sin_zero, sizeof sin.sin_zero); 782 laddr = inp->inp_laddr; 783 if (inp->inp_laddr.s_addr == INADDR_ANY) 784 inp->inp_laddr = sc->sc_inc.inc_laddr; 785 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) { 786 inp->inp_laddr = laddr; 787 goto abort; 788 } 789 } 790 791 /* 792 * The current port should be in the context of the SYN+ACK and 793 * so should match the tcp address port. 794 * 795 * XXX we may be running on the netisr thread instead of a tcp 796 * thread, in which case port will not match 797 * curthread->td_msgport. 798 */ 799 if (isipv6) { 800 port = tcp6_addrport(); 801 } else { 802 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport, 803 inp->inp_laddr.s_addr, inp->inp_lport); 804 } 805 /*KKASSERT(port == &curthread->td_msgport);*/ 806 807 tp = intotcpcb(inp); 808 tp->t_state = TCPS_SYN_RECEIVED; 809 tp->iss = sc->sc_iss; 810 tp->irs = sc->sc_irs; 811 tcp_rcvseqinit(tp); 812 tcp_sendseqinit(tp); 813 tp->snd_wl1 = sc->sc_irs; 814 tp->rcv_up = sc->sc_irs + 1; 815 tp->rcv_wnd = sc->sc_wnd; 816 tp->rcv_adv += tp->rcv_wnd; 817 818 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 819 if (sc->sc_flags & SCF_NOOPT) 820 tp->t_flags |= TF_NOOPT; 821 if (sc->sc_flags & SCF_WINSCALE) { 822 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 823 tp->requested_s_scale = sc->sc_requested_s_scale; 824 tp->request_r_scale = sc->sc_request_r_scale; 825 } 826 if (sc->sc_flags & SCF_TIMESTAMP) { 827 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 828 tp->ts_recent = sc->sc_tsrecent; 829 tp->ts_recent_age = ticks; 830 } 831 if (sc->sc_flags & SCF_SACK_PERMITTED) 832 tp->t_flags |= TF_SACK_PERMITTED; 833 834 tcp_mss(tp, sc->sc_peer_mss); 835 836 /* 837 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 838 */ 839 if (sc->sc_rxtslot != 0) 840 tp->snd_cwnd = tp->t_maxseg; 841 tcp_create_timermsg(tp, port); 842 tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep); 843 844 tcpstat.tcps_accepts++; 845 return (so); 846 847 abort: 848 if (so != NULL) 849 soabort_oncpu(so); 850 return (NULL); 851 } 852 853 /* 854 * This function gets called when we receive an ACK for a 855 * socket in the LISTEN state. We look up the connection 856 * in the syncache, and if its there, we pull it out of 857 * the cache and turn it into a full-blown connection in 858 * the SYN-RECEIVED state. 859 */ 860 int 861 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop, 862 struct mbuf *m) 863 { 864 struct syncache *sc; 865 struct syncache_head *sch; 866 struct socket *so; 867 868 sc = syncache_lookup(inc, &sch); 869 if (sc == NULL) { 870 /* 871 * There is no syncache entry, so see if this ACK is 872 * a returning syncookie. To do this, first: 873 * A. See if this socket has had a syncache entry dropped in 874 * the past. We don't want to accept a bogus syncookie 875 * if we've never received a SYN. 876 * B. check that the syncookie is valid. If it is, then 877 * cobble up a fake syncache entry, and return. 878 */ 879 if (!tcp_syncookies) 880 return (0); 881 sc = syncookie_lookup(inc, th, *sop); 882 if (sc == NULL) 883 return (0); 884 sch = NULL; 885 tcpstat.tcps_sc_recvcookie++; 886 } 887 888 /* 889 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 890 */ 891 if (th->th_ack != sc->sc_iss + 1) 892 return (0); 893 894 so = syncache_socket(sc, *sop, m); 895 if (so == NULL) { 896 #if 0 897 resetandabort: 898 /* XXXjlemon check this - is this correct? */ 899 tcp_respond(NULL, m, m, th, 900 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 901 #endif 902 m_freem(m); /* XXX only needed for above */ 903 tcpstat.tcps_sc_aborted++; 904 } else { 905 tcpstat.tcps_sc_completed++; 906 } 907 if (sch == NULL) 908 syncache_free(sc); 909 else 910 syncache_drop(sc, sch); 911 *sop = so; 912 return (1); 913 } 914 915 /* 916 * Given a LISTEN socket and an inbound SYN request, add 917 * this to the syn cache, and send back a segment: 918 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 919 * to the source. 920 * 921 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 922 * Doing so would require that we hold onto the data and deliver it 923 * to the application. However, if we are the target of a SYN-flood 924 * DoS attack, an attacker could send data which would eventually 925 * consume all available buffer space if it were ACKed. By not ACKing 926 * the data, we avoid this DoS scenario. 927 */ 928 int 929 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 930 struct socket **sop, struct mbuf *m) 931 { 932 struct tcp_syncache_percpu *syncache_percpu; 933 struct tcpcb *tp; 934 struct socket *so; 935 struct syncache *sc = NULL; 936 struct syncache_head *sch; 937 struct mbuf *ipopts = NULL; 938 int win; 939 940 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 941 so = *sop; 942 tp = sototcpcb(so); 943 944 /* 945 * Remember the IP options, if any. 946 */ 947 #ifdef INET6 948 if (!inc->inc_isipv6) 949 #endif 950 ipopts = ip_srcroute(m); 951 952 /* 953 * See if we already have an entry for this connection. 954 * If we do, resend the SYN,ACK, and reset the retransmit timer. 955 * 956 * XXX 957 * The syncache should be re-initialized with the contents 958 * of the new SYN which may have different options. 959 */ 960 sc = syncache_lookup(inc, &sch); 961 if (sc != NULL) { 962 tcpstat.tcps_sc_dupsyn++; 963 if (ipopts) { 964 /* 965 * If we were remembering a previous source route, 966 * forget it and use the new one we've been given. 967 */ 968 if (sc->sc_ipopts) 969 m_free(sc->sc_ipopts); 970 sc->sc_ipopts = ipopts; 971 } 972 /* 973 * Update timestamp if present. 974 */ 975 if (sc->sc_flags & SCF_TIMESTAMP) 976 sc->sc_tsrecent = to->to_tsval; 977 978 /* Just update the TOF_SACK_PERMITTED for now. */ 979 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 980 sc->sc_flags |= SCF_SACK_PERMITTED; 981 else 982 sc->sc_flags &= ~SCF_SACK_PERMITTED; 983 984 /* 985 * PCB may have changed, pick up new values. 986 */ 987 if (sc->sc_tp) { 988 sc->sc_tp->t_flags &= ~TF_SYNCACHE; 989 tp->t_flags |= TF_SYNCACHE; 990 } 991 sc->sc_tp = tp; 992 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 993 if (syncache_respond(sc, m) == 0) { 994 crit_enter(); 995 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 996 sc, sc_timerq); 997 crit_exit(); 998 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 999 tcpstat.tcps_sndacks++; 1000 tcpstat.tcps_sndtotal++; 1001 } 1002 *sop = NULL; 1003 return (1); 1004 } 1005 1006 /* 1007 * Fill in the syncache values. 1008 */ 1009 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1010 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1011 sc->sc_ipopts = ipopts; 1012 sc->sc_inc.inc_fport = inc->inc_fport; 1013 sc->sc_inc.inc_lport = inc->inc_lport; 1014 sc->sc_tp = tp; 1015 tp->t_flags |= TF_SYNCACHE; 1016 #ifdef INET6 1017 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1018 if (inc->inc_isipv6) { 1019 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1020 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1021 sc->sc_route6.ro_rt = NULL; 1022 } else 1023 #endif 1024 { 1025 sc->sc_inc.inc_faddr = inc->inc_faddr; 1026 sc->sc_inc.inc_laddr = inc->inc_laddr; 1027 sc->sc_route.ro_rt = NULL; 1028 } 1029 sc->sc_irs = th->th_seq; 1030 sc->sc_flags = 0; 1031 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 1032 if (tcp_syncookies) 1033 sc->sc_iss = syncookie_generate(sc); 1034 else 1035 sc->sc_iss = karc4random(); 1036 1037 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */ 1038 win = ssb_space(&so->so_rcv); 1039 win = imax(win, 0); 1040 win = imin(win, TCP_MAXWIN); 1041 sc->sc_wnd = win; 1042 1043 if (tcp_do_rfc1323) { 1044 /* 1045 * A timestamp received in a SYN makes 1046 * it ok to send timestamp requests and replies. 1047 */ 1048 if (to->to_flags & TOF_TS) { 1049 sc->sc_tsrecent = to->to_tsval; 1050 sc->sc_flags |= SCF_TIMESTAMP; 1051 } 1052 if (to->to_flags & TOF_SCALE) { 1053 int wscale = TCP_MIN_WINSHIFT; 1054 1055 /* Compute proper scaling value from buffer space */ 1056 while (wscale < TCP_MAX_WINSHIFT && 1057 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) { 1058 wscale++; 1059 } 1060 sc->sc_request_r_scale = wscale; 1061 sc->sc_requested_s_scale = to->to_requested_s_scale; 1062 sc->sc_flags |= SCF_WINSCALE; 1063 } 1064 } 1065 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1066 sc->sc_flags |= SCF_SACK_PERMITTED; 1067 if (tp->t_flags & TF_NOOPT) 1068 sc->sc_flags = SCF_NOOPT; 1069 1070 if (syncache_respond(sc, m) == 0) { 1071 syncache_insert(sc, sch); 1072 tcpstat.tcps_sndacks++; 1073 tcpstat.tcps_sndtotal++; 1074 } else { 1075 syncache_free(sc); 1076 tcpstat.tcps_sc_dropped++; 1077 } 1078 *sop = NULL; 1079 return (1); 1080 } 1081 1082 static int 1083 syncache_respond(struct syncache *sc, struct mbuf *m) 1084 { 1085 u_int8_t *optp; 1086 int optlen, error; 1087 u_int16_t tlen, hlen, mssopt; 1088 struct ip *ip = NULL; 1089 struct rtentry *rt; 1090 struct tcphdr *th; 1091 struct ip6_hdr *ip6 = NULL; 1092 #ifdef INET6 1093 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1094 #else 1095 const boolean_t isipv6 = FALSE; 1096 #endif 1097 1098 if (isipv6) { 1099 rt = tcp_rtlookup6(&sc->sc_inc); 1100 if (rt != NULL) 1101 mssopt = rt->rt_ifp->if_mtu - 1102 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1103 else 1104 mssopt = tcp_v6mssdflt; 1105 hlen = sizeof(struct ip6_hdr); 1106 } else { 1107 rt = tcp_rtlookup(&sc->sc_inc); 1108 if (rt != NULL) 1109 mssopt = rt->rt_ifp->if_mtu - 1110 (sizeof(struct ip) + sizeof(struct tcphdr)); 1111 else 1112 mssopt = tcp_mssdflt; 1113 hlen = sizeof(struct ip); 1114 } 1115 1116 /* Compute the size of the TCP options. */ 1117 if (sc->sc_flags & SCF_NOOPT) { 1118 optlen = 0; 1119 } else { 1120 optlen = TCPOLEN_MAXSEG + 1121 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1122 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1123 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1124 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1125 } 1126 tlen = hlen + sizeof(struct tcphdr) + optlen; 1127 1128 /* 1129 * XXX 1130 * assume that the entire packet will fit in a header mbuf 1131 */ 1132 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1133 1134 /* 1135 * XXX shouldn't this reuse the mbuf if possible ? 1136 * Create the IP+TCP header from scratch. 1137 */ 1138 if (m) 1139 m_freem(m); 1140 1141 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 1142 if (m == NULL) 1143 return (ENOBUFS); 1144 m->m_data += max_linkhdr; 1145 m->m_len = tlen; 1146 m->m_pkthdr.len = tlen; 1147 m->m_pkthdr.rcvif = NULL; 1148 1149 if (isipv6) { 1150 ip6 = mtod(m, struct ip6_hdr *); 1151 ip6->ip6_vfc = IPV6_VERSION; 1152 ip6->ip6_nxt = IPPROTO_TCP; 1153 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1154 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1155 ip6->ip6_plen = htons(tlen - hlen); 1156 /* ip6_hlim is set after checksum */ 1157 /* ip6_flow = ??? */ 1158 1159 th = (struct tcphdr *)(ip6 + 1); 1160 } else { 1161 ip = mtod(m, struct ip *); 1162 ip->ip_v = IPVERSION; 1163 ip->ip_hl = sizeof(struct ip) >> 2; 1164 ip->ip_len = tlen; 1165 ip->ip_id = 0; 1166 ip->ip_off = 0; 1167 ip->ip_sum = 0; 1168 ip->ip_p = IPPROTO_TCP; 1169 ip->ip_src = sc->sc_inc.inc_laddr; 1170 ip->ip_dst = sc->sc_inc.inc_faddr; 1171 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1172 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1173 1174 /* 1175 * See if we should do MTU discovery. Route lookups are 1176 * expensive, so we will only unset the DF bit if: 1177 * 1178 * 1) path_mtu_discovery is disabled 1179 * 2) the SCF_UNREACH flag has been set 1180 */ 1181 if (path_mtu_discovery 1182 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1183 ip->ip_off |= IP_DF; 1184 } 1185 1186 th = (struct tcphdr *)(ip + 1); 1187 } 1188 th->th_sport = sc->sc_inc.inc_lport; 1189 th->th_dport = sc->sc_inc.inc_fport; 1190 1191 th->th_seq = htonl(sc->sc_iss); 1192 th->th_ack = htonl(sc->sc_irs + 1); 1193 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1194 th->th_x2 = 0; 1195 th->th_flags = TH_SYN | TH_ACK; 1196 th->th_win = htons(sc->sc_wnd); 1197 th->th_urp = 0; 1198 1199 /* Tack on the TCP options. */ 1200 if (optlen == 0) 1201 goto no_options; 1202 optp = (u_int8_t *)(th + 1); 1203 *optp++ = TCPOPT_MAXSEG; 1204 *optp++ = TCPOLEN_MAXSEG; 1205 *optp++ = (mssopt >> 8) & 0xff; 1206 *optp++ = mssopt & 0xff; 1207 1208 if (sc->sc_flags & SCF_WINSCALE) { 1209 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1210 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1211 sc->sc_request_r_scale); 1212 optp += 4; 1213 } 1214 1215 if (sc->sc_flags & SCF_TIMESTAMP) { 1216 u_int32_t *lp = (u_int32_t *)(optp); 1217 1218 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1219 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1220 *lp++ = htonl(ticks); 1221 *lp = htonl(sc->sc_tsrecent); 1222 optp += TCPOLEN_TSTAMP_APPA; 1223 } 1224 1225 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1226 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1227 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1228 } 1229 1230 no_options: 1231 if (isipv6) { 1232 struct route_in6 *ro6 = &sc->sc_route6; 1233 1234 th->th_sum = 0; 1235 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1236 ip6->ip6_hlim = in6_selecthlim(NULL, 1237 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1238 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1239 sc->sc_tp->t_inpcb); 1240 } else { 1241 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1242 htons(tlen - hlen + IPPROTO_TCP)); 1243 m->m_pkthdr.csum_flags = CSUM_TCP; 1244 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1245 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 1246 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb); 1247 } 1248 return (error); 1249 } 1250 1251 /* 1252 * cookie layers: 1253 * 1254 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1255 * | peer iss | 1256 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1257 * | 0 |(A)| | 1258 * (A): peer mss index 1259 */ 1260 1261 /* 1262 * The values below are chosen to minimize the size of the tcp_secret 1263 * table, as well as providing roughly a 16 second lifetime for the cookie. 1264 */ 1265 1266 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1267 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1268 1269 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1270 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1271 #define SYNCOOKIE_TIMEOUT \ 1272 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1273 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1274 1275 static struct { 1276 u_int32_t ts_secbits[4]; 1277 u_int ts_expire; 1278 } tcp_secret[SYNCOOKIE_NSECRETS]; 1279 1280 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1281 1282 static MD5_CTX syn_ctx; 1283 1284 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1285 1286 struct md5_add { 1287 u_int32_t laddr, faddr; 1288 u_int32_t secbits[4]; 1289 u_int16_t lport, fport; 1290 }; 1291 1292 #ifdef CTASSERT 1293 CTASSERT(sizeof(struct md5_add) == 28); 1294 #endif 1295 1296 /* 1297 * Consider the problem of a recreated (and retransmitted) cookie. If the 1298 * original SYN was accepted, the connection is established. The second 1299 * SYN is inflight, and if it arrives with an ISN that falls within the 1300 * receive window, the connection is killed. 1301 * 1302 * However, since cookies have other problems, this may not be worth 1303 * worrying about. 1304 */ 1305 1306 static u_int32_t 1307 syncookie_generate(struct syncache *sc) 1308 { 1309 u_int32_t md5_buffer[4]; 1310 u_int32_t data; 1311 int idx, i; 1312 struct md5_add add; 1313 #ifdef INET6 1314 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1315 #else 1316 const boolean_t isipv6 = FALSE; 1317 #endif 1318 1319 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1320 if (tcp_secret[idx].ts_expire < ticks) { 1321 for (i = 0; i < 4; i++) 1322 tcp_secret[idx].ts_secbits[i] = karc4random(); 1323 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1324 } 1325 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1326 if (tcp_msstab[data] <= sc->sc_peer_mss) 1327 break; 1328 data = (data << SYNCOOKIE_WNDBITS) | idx; 1329 data ^= sc->sc_irs; /* peer's iss */ 1330 MD5Init(&syn_ctx); 1331 if (isipv6) { 1332 MD5Add(sc->sc_inc.inc6_laddr); 1333 MD5Add(sc->sc_inc.inc6_faddr); 1334 add.laddr = 0; 1335 add.faddr = 0; 1336 } else { 1337 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1338 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1339 } 1340 add.lport = sc->sc_inc.inc_lport; 1341 add.fport = sc->sc_inc.inc_fport; 1342 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1343 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1344 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1345 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1346 MD5Add(add); 1347 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1348 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1349 return (data); 1350 } 1351 1352 static struct syncache * 1353 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so) 1354 { 1355 u_int32_t md5_buffer[4]; 1356 struct syncache *sc; 1357 u_int32_t data; 1358 int wnd, idx; 1359 struct md5_add add; 1360 1361 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1362 idx = data & SYNCOOKIE_WNDMASK; 1363 if (tcp_secret[idx].ts_expire < ticks || 1364 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1365 return (NULL); 1366 MD5Init(&syn_ctx); 1367 #ifdef INET6 1368 if (inc->inc_isipv6) { 1369 MD5Add(inc->inc6_laddr); 1370 MD5Add(inc->inc6_faddr); 1371 add.laddr = 0; 1372 add.faddr = 0; 1373 } else 1374 #endif 1375 { 1376 add.laddr = inc->inc_laddr.s_addr; 1377 add.faddr = inc->inc_faddr.s_addr; 1378 } 1379 add.lport = inc->inc_lport; 1380 add.fport = inc->inc_fport; 1381 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1382 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1383 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1384 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1385 MD5Add(add); 1386 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1387 data ^= md5_buffer[0]; 1388 if (data & ~SYNCOOKIE_DATAMASK) 1389 return (NULL); 1390 data = data >> SYNCOOKIE_WNDBITS; 1391 1392 /* 1393 * Fill in the syncache values. 1394 * XXX duplicate code from syncache_add 1395 */ 1396 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1397 sc->sc_ipopts = NULL; 1398 sc->sc_inc.inc_fport = inc->inc_fport; 1399 sc->sc_inc.inc_lport = inc->inc_lport; 1400 #ifdef INET6 1401 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1402 if (inc->inc_isipv6) { 1403 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1404 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1405 sc->sc_route6.ro_rt = NULL; 1406 } else 1407 #endif 1408 { 1409 sc->sc_inc.inc_faddr = inc->inc_faddr; 1410 sc->sc_inc.inc_laddr = inc->inc_laddr; 1411 sc->sc_route.ro_rt = NULL; 1412 } 1413 sc->sc_irs = th->th_seq - 1; 1414 sc->sc_iss = th->th_ack - 1; 1415 wnd = ssb_space(&so->so_rcv); 1416 wnd = imax(wnd, 0); 1417 wnd = imin(wnd, TCP_MAXWIN); 1418 sc->sc_wnd = wnd; 1419 sc->sc_flags = 0; 1420 sc->sc_rxtslot = 0; 1421 sc->sc_peer_mss = tcp_msstab[data]; 1422 return (sc); 1423 } 1424