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 1.34 2008/11/09 10:32:16 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_var.h> 115 #include <netinet6/tcp6_var.h> 116 117 #ifdef IPSEC 118 #include <netinet6/ipsec.h> 119 #ifdef INET6 120 #include <netinet6/ipsec6.h> 121 #endif 122 #include <netproto/key/key.h> 123 #endif /*IPSEC*/ 124 125 #ifdef FAST_IPSEC 126 #include <netproto/ipsec/ipsec.h> 127 #ifdef INET6 128 #include <netproto/ipsec/ipsec6.h> 129 #endif 130 #include <netproto/ipsec/key.h> 131 #define IPSEC 132 #endif /*FAST_IPSEC*/ 133 134 #include <vm/vm_zone.h> 135 136 static int tcp_syncookies = 1; 137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 138 &tcp_syncookies, 0, 139 "Use TCP SYN cookies if the syncache overflows"); 140 141 static void syncache_drop(struct syncache *, struct syncache_head *); 142 static void syncache_free(struct syncache *); 143 static void syncache_insert(struct syncache *, struct syncache_head *); 144 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 145 static int syncache_respond(struct syncache *, struct mbuf *); 146 static struct socket *syncache_socket(struct syncache *, struct socket *, 147 struct mbuf *); 148 static void syncache_timer(void *); 149 static u_int32_t syncookie_generate(struct syncache *); 150 static 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 30 163 164 struct netmsg_sc_timer { 165 struct netmsg nm_netmsg; 166 struct msgrec *nm_mrec; /* back pointer to containing msgrec */ 167 }; 168 169 struct msgrec { 170 struct netmsg_sc_timer msg; 171 lwkt_port_t port; /* constant after init */ 172 int slot; /* constant after init */ 173 }; 174 175 static void syncache_timer_handler(netmsg_t); 176 177 struct tcp_syncache { 178 struct vm_zone *zone; 179 u_int hashsize; 180 u_int hashmask; 181 u_int bucket_limit; 182 u_int cache_limit; 183 u_int rexmt_limit; 184 u_int hash_secret; 185 }; 186 static struct tcp_syncache tcp_syncache; 187 188 struct tcp_syncache_percpu { 189 struct syncache_head *hashbase; 190 u_int cache_count; 191 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1]; 192 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1]; 193 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1]; 194 }; 195 static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU]; 196 197 static struct lwkt_port syncache_null_rport; 198 199 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 200 201 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD, 202 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 203 204 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD, 205 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 206 207 /* XXX JH */ 208 #if 0 209 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 210 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 211 #endif 212 213 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD, 214 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 215 216 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 217 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 218 219 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 220 221 #define SYNCACHE_HASH(inc, mask) \ 222 ((tcp_syncache.hash_secret ^ \ 223 (inc)->inc_faddr.s_addr ^ \ 224 ((inc)->inc_faddr.s_addr >> 16) ^ \ 225 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 226 227 #define SYNCACHE_HASH6(inc, mask) \ 228 ((tcp_syncache.hash_secret ^ \ 229 (inc)->inc6_faddr.s6_addr32[0] ^ \ 230 (inc)->inc6_faddr.s6_addr32[3] ^ \ 231 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 232 233 #define ENDPTS_EQ(a, b) ( \ 234 (a)->ie_fport == (b)->ie_fport && \ 235 (a)->ie_lport == (b)->ie_lport && \ 236 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 237 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 238 ) 239 240 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 241 242 static __inline void 243 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu, 244 struct syncache *sc, int slot) 245 { 246 sc->sc_rxtslot = slot; 247 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[slot]; 248 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq); 249 if (!callout_active(&syncache_percpu->tt_timerq[slot])) { 250 callout_reset(&syncache_percpu->tt_timerq[slot], 251 TCPTV_RTOBASE * tcp_backoff[slot], 252 syncache_timer, 253 &syncache_percpu->mrec[slot]); 254 } 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 } 281 282 zfree(tcp_syncache.zone, sc); 283 } 284 285 void 286 syncache_init(void) 287 { 288 int i, cpu; 289 290 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; 295 tcp_syncache.hash_secret = karc4random(); 296 297 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 298 &tcp_syncache.hashsize); 299 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 300 &tcp_syncache.cache_limit); 301 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 302 &tcp_syncache.bucket_limit); 303 if (!powerof2(tcp_syncache.hashsize)) { 304 kprintf("WARNING: syncache hash size is not a power of 2.\n"); 305 tcp_syncache.hashsize = 512; /* safe default */ 306 } 307 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 308 309 lwkt_initport_replyonly_null(&syncache_null_rport); 310 311 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 } 328 329 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; 335 syncache_percpu->mrec[i].port = tcp_cport(cpu); 336 syncache_percpu->mrec[i].msg.nm_mrec = 337 &syncache_percpu->mrec[i]; 338 netmsg_init(&syncache_percpu->mrec[i].msg.nm_netmsg, 339 &syncache_null_rport, MSGF_PRIORITY, 340 syncache_timer_handler); 341 } 342 } 343 344 /* 345 * Allocate the syncache entries. Allow the zone to allocate one 346 * more entry than cache limit, so a new entry can bump out an 347 * older one. 348 */ 349 tcp_syncache.zone = zinit("syncache", sizeof(struct syncache), 350 tcp_syncache.cache_limit * ncpus2, ZONE_INTERRUPT, 0); 351 tcp_syncache.cache_limit -= 1; 352 } 353 354 static void 355 syncache_insert(struct syncache *sc, struct syncache_head *sch) 356 { 357 struct tcp_syncache_percpu *syncache_percpu; 358 struct syncache *sc2; 359 int i; 360 361 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 362 363 /* 364 * Make sure that we don't overflow the per-bucket 365 * limit or the total cache size limit. 366 */ 367 if (sch->sch_length >= tcp_syncache.bucket_limit) { 368 /* 369 * The bucket is full, toss the oldest element. 370 */ 371 sc2 = TAILQ_FIRST(&sch->sch_bucket); 372 sc2->sc_tp->ts_recent = ticks; 373 syncache_drop(sc2, sch); 374 tcpstat.tcps_sc_bucketoverflow++; 375 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) { 376 /* 377 * The cache is full. Toss the oldest entry in the 378 * entire cache. This is the front entry in the 379 * first non-empty timer queue with the largest 380 * timeout value. 381 */ 382 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 383 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]); 384 if (sc2 != NULL) 385 break; 386 } 387 sc2->sc_tp->ts_recent = ticks; 388 syncache_drop(sc2, NULL); 389 tcpstat.tcps_sc_cacheoverflow++; 390 } 391 392 /* Initialize the entry's timer. */ 393 syncache_timeout(syncache_percpu, sc, 0); 394 395 /* Put it into the bucket. */ 396 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 397 sch->sch_length++; 398 syncache_percpu->cache_count++; 399 tcpstat.tcps_sc_added++; 400 } 401 402 static void 403 syncache_drop(struct syncache *sc, struct syncache_head *sch) 404 { 405 struct tcp_syncache_percpu *syncache_percpu; 406 #ifdef INET6 407 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 408 #else 409 const boolean_t isipv6 = FALSE; 410 #endif 411 412 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 413 414 if (sch == NULL) { 415 if (isipv6) { 416 sch = &syncache_percpu->hashbase[ 417 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 418 } else { 419 sch = &syncache_percpu->hashbase[ 420 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 421 } 422 } 423 424 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 425 sch->sch_length--; 426 syncache_percpu->cache_count--; 427 428 /* 429 * Remove the entry from the syncache timer/timeout queue. Note 430 * that we do not try to stop any running timer since we do not know 431 * whether the timer's message is in-transit or not. Since timeouts 432 * are fairly long, taking an unneeded callout does not detrimentally 433 * effect performance. 434 */ 435 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq); 436 437 syncache_free(sc); 438 } 439 440 /* 441 * Place a timeout message on the TCP thread's message queue. 442 * This routine runs in soft interrupt context. 443 * 444 * An invariant is for this routine to be called, the callout must 445 * have been active. Note that the callout is not deactivated until 446 * after the message has been processed in syncache_timer_handler() below. 447 */ 448 static void 449 syncache_timer(void *p) 450 { 451 struct netmsg_sc_timer *msg = p; 452 453 lwkt_sendmsg(msg->nm_mrec->port, &msg->nm_netmsg.nm_lmsg); 454 } 455 456 /* 457 * Service a timer message queued by timer expiration. 458 * This routine runs in the TCP protocol thread. 459 * 460 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 461 * If we have retransmitted an entry the maximum number of times, expire it. 462 * 463 * When we finish processing timed-out entries, we restart the timer if there 464 * are any entries still on the queue and deactivate it otherwise. Only after 465 * a timer has been deactivated here can it be restarted by syncache_timeout(). 466 */ 467 static void 468 syncache_timer_handler(netmsg_t netmsg) 469 { 470 struct tcp_syncache_percpu *syncache_percpu; 471 struct syncache *sc, *nsc; 472 struct inpcb *inp; 473 int slot; 474 475 slot = ((struct netmsg_sc_timer *)netmsg)->nm_mrec->slot; 476 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 477 478 nsc = TAILQ_FIRST(&syncache_percpu->timerq[slot]); 479 while (nsc != NULL) { 480 if (ticks < nsc->sc_rxttime) 481 break; /* finished because timerq sorted by time */ 482 sc = nsc; 483 inp = sc->sc_tp->t_inpcb; 484 if (slot == SYNCACHE_MAXREXMTS || 485 slot >= tcp_syncache.rexmt_limit || 486 inp->inp_gencnt != sc->sc_inp_gencnt) { 487 nsc = TAILQ_NEXT(sc, sc_timerq); 488 syncache_drop(sc, NULL); 489 tcpstat.tcps_sc_stale++; 490 continue; 491 } 492 /* 493 * syncache_respond() may call back into the syncache to 494 * to modify another entry, so do not obtain the next 495 * entry on the timer chain until it has completed. 496 */ 497 syncache_respond(sc, NULL); 498 nsc = TAILQ_NEXT(sc, sc_timerq); 499 tcpstat.tcps_sc_retransmitted++; 500 TAILQ_REMOVE(&syncache_percpu->timerq[slot], sc, sc_timerq); 501 syncache_timeout(syncache_percpu, sc, slot + 1); 502 } 503 if (nsc != NULL) 504 callout_reset(&syncache_percpu->tt_timerq[slot], 505 nsc->sc_rxttime - ticks, syncache_timer, 506 &syncache_percpu->mrec[slot]); 507 else 508 callout_deactivate(&syncache_percpu->tt_timerq[slot]); 509 510 lwkt_replymsg(&netmsg->nm_lmsg, 0); 511 } 512 513 /* 514 * Find an entry in the syncache. 515 */ 516 struct syncache * 517 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 518 { 519 struct tcp_syncache_percpu *syncache_percpu; 520 struct syncache *sc; 521 struct syncache_head *sch; 522 523 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 524 #ifdef INET6 525 if (inc->inc_isipv6) { 526 sch = &syncache_percpu->hashbase[ 527 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 528 *schp = sch; 529 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) 530 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 531 return (sc); 532 } else 533 #endif 534 { 535 sch = &syncache_percpu->hashbase[ 536 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 537 *schp = sch; 538 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 539 #ifdef INET6 540 if (sc->sc_inc.inc_isipv6) 541 continue; 542 #endif 543 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 544 return (sc); 545 } 546 } 547 return (NULL); 548 } 549 550 /* 551 * This function is called when we get a RST for a 552 * non-existent connection, so that we can see if the 553 * connection is in the syn cache. If it is, zap it. 554 */ 555 void 556 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 557 { 558 struct syncache *sc; 559 struct syncache_head *sch; 560 561 sc = syncache_lookup(inc, &sch); 562 if (sc == NULL) 563 return; 564 /* 565 * If the RST bit is set, check the sequence number to see 566 * if this is a valid reset segment. 567 * RFC 793 page 37: 568 * In all states except SYN-SENT, all reset (RST) segments 569 * are validated by checking their SEQ-fields. A reset is 570 * valid if its sequence number is in the window. 571 * 572 * The sequence number in the reset segment is normally an 573 * echo of our outgoing acknowlegement numbers, but some hosts 574 * send a reset with the sequence number at the rightmost edge 575 * of our receive window, and we have to handle this case. 576 */ 577 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 578 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 579 syncache_drop(sc, sch); 580 tcpstat.tcps_sc_reset++; 581 } 582 } 583 584 void 585 syncache_badack(struct in_conninfo *inc) 586 { 587 struct syncache *sc; 588 struct syncache_head *sch; 589 590 sc = syncache_lookup(inc, &sch); 591 if (sc != NULL) { 592 syncache_drop(sc, sch); 593 tcpstat.tcps_sc_badack++; 594 } 595 } 596 597 void 598 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 599 { 600 struct syncache *sc; 601 struct syncache_head *sch; 602 603 /* we are called at splnet() here */ 604 sc = syncache_lookup(inc, &sch); 605 if (sc == NULL) 606 return; 607 608 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 609 if (ntohl(th->th_seq) != sc->sc_iss) 610 return; 611 612 /* 613 * If we've rertransmitted 3 times and this is our second error, 614 * we remove the entry. Otherwise, we allow it to continue on. 615 * This prevents us from incorrectly nuking an entry during a 616 * spurious network outage. 617 * 618 * See tcp_notify(). 619 */ 620 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 621 sc->sc_flags |= SCF_UNREACH; 622 return; 623 } 624 syncache_drop(sc, sch); 625 tcpstat.tcps_sc_unreach++; 626 } 627 628 /* 629 * Build a new TCP socket structure from a syncache entry. 630 */ 631 static struct socket * 632 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 633 { 634 struct inpcb *inp = NULL, *linp; 635 struct socket *so; 636 struct tcpcb *tp; 637 #ifdef INET6 638 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 639 #else 640 const boolean_t isipv6 = FALSE; 641 #endif 642 643 /* 644 * Ok, create the full blown connection, and set things up 645 * as they would have been set up if we had created the 646 * connection when the SYN arrived. If we can't create 647 * the connection, abort it. 648 */ 649 so = sonewconn(lso, SS_ISCONNECTED); 650 if (so == NULL) { 651 /* 652 * Drop the connection; we will send a RST if the peer 653 * retransmits the ACK, 654 */ 655 tcpstat.tcps_listendrop++; 656 goto abort; 657 } 658 659 inp = so->so_pcb; 660 661 /* 662 * Insert new socket into hash list. 663 */ 664 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 665 if (isipv6) { 666 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 667 } else { 668 #ifdef INET6 669 inp->inp_vflag &= ~INP_IPV6; 670 inp->inp_vflag |= INP_IPV4; 671 #endif 672 inp->inp_laddr = sc->sc_inc.inc_laddr; 673 } 674 inp->inp_lport = sc->sc_inc.inc_lport; 675 if (in_pcbinsporthash(inp) != 0) { 676 /* 677 * Undo the assignments above if we failed to 678 * put the PCB on the hash lists. 679 */ 680 if (isipv6) 681 inp->in6p_laddr = kin6addr_any; 682 else 683 inp->inp_laddr.s_addr = INADDR_ANY; 684 inp->inp_lport = 0; 685 goto abort; 686 } 687 linp = so->so_pcb; 688 #ifdef IPSEC 689 /* copy old policy into new socket's */ 690 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 691 kprintf("syncache_expand: could not copy policy\n"); 692 #endif 693 if (isipv6) { 694 struct in6_addr laddr6; 695 struct sockaddr_in6 sin6; 696 /* 697 * Inherit socket options from the listening socket. 698 * Note that in6p_inputopts are not (and should not be) 699 * copied, since it stores previously received options and is 700 * used to detect if each new option is different than the 701 * previous one and hence should be passed to a user. 702 * If we copied in6p_inputopts, a user would not be able to 703 * receive options just after calling the accept system call. 704 */ 705 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 706 if (linp->in6p_outputopts) 707 inp->in6p_outputopts = 708 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 709 inp->in6p_route = sc->sc_route6; 710 sc->sc_route6.ro_rt = NULL; 711 712 sin6.sin6_family = AF_INET6; 713 sin6.sin6_len = sizeof sin6; 714 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 715 sin6.sin6_port = sc->sc_inc.inc_fport; 716 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 717 laddr6 = inp->in6p_laddr; 718 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 719 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 720 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) { 721 inp->in6p_laddr = laddr6; 722 goto abort; 723 } 724 } else { 725 struct in_addr laddr; 726 struct sockaddr_in sin; 727 728 inp->inp_options = ip_srcroute(m); 729 if (inp->inp_options == NULL) { 730 inp->inp_options = sc->sc_ipopts; 731 sc->sc_ipopts = NULL; 732 } 733 inp->inp_route = sc->sc_route; 734 sc->sc_route.ro_rt = NULL; 735 736 sin.sin_family = AF_INET; 737 sin.sin_len = sizeof sin; 738 sin.sin_addr = sc->sc_inc.inc_faddr; 739 sin.sin_port = sc->sc_inc.inc_fport; 740 bzero(sin.sin_zero, sizeof sin.sin_zero); 741 laddr = inp->inp_laddr; 742 if (inp->inp_laddr.s_addr == INADDR_ANY) 743 inp->inp_laddr = sc->sc_inc.inc_laddr; 744 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) { 745 inp->inp_laddr = laddr; 746 goto abort; 747 } 748 } 749 750 tp = intotcpcb(inp); 751 tp->t_state = TCPS_SYN_RECEIVED; 752 tp->iss = sc->sc_iss; 753 tp->irs = sc->sc_irs; 754 tcp_rcvseqinit(tp); 755 tcp_sendseqinit(tp); 756 tp->snd_wl1 = sc->sc_irs; 757 tp->rcv_up = sc->sc_irs + 1; 758 tp->rcv_wnd = sc->sc_wnd; 759 tp->rcv_adv += tp->rcv_wnd; 760 761 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 762 if (sc->sc_flags & SCF_NOOPT) 763 tp->t_flags |= TF_NOOPT; 764 if (sc->sc_flags & SCF_WINSCALE) { 765 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 766 tp->requested_s_scale = sc->sc_requested_s_scale; 767 tp->request_r_scale = sc->sc_request_r_scale; 768 } 769 if (sc->sc_flags & SCF_TIMESTAMP) { 770 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 771 tp->ts_recent = sc->sc_tsrecent; 772 tp->ts_recent_age = ticks; 773 } 774 if (sc->sc_flags & SCF_CC) { 775 /* 776 * Initialization of the tcpcb for transaction; 777 * set SND.WND = SEG.WND, 778 * initialize CCsend and CCrecv. 779 */ 780 tp->t_flags |= TF_REQ_CC | TF_RCVD_CC; 781 tp->cc_send = sc->sc_cc_send; 782 tp->cc_recv = sc->sc_cc_recv; 783 } 784 if (sc->sc_flags & SCF_SACK_PERMITTED) 785 tp->t_flags |= TF_SACK_PERMITTED; 786 787 tcp_mss(tp, sc->sc_peer_mss); 788 789 /* 790 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 791 */ 792 if (sc->sc_rxtslot != 0) 793 tp->snd_cwnd = tp->t_maxseg; 794 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 795 796 tcpstat.tcps_accepts++; 797 return (so); 798 799 abort: 800 if (so != NULL) 801 soaborta(so); 802 return (NULL); 803 } 804 805 /* 806 * This function gets called when we receive an ACK for a 807 * socket in the LISTEN state. We look up the connection 808 * in the syncache, and if its there, we pull it out of 809 * the cache and turn it into a full-blown connection in 810 * the SYN-RECEIVED state. 811 */ 812 int 813 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop, 814 struct mbuf *m) 815 { 816 struct syncache *sc; 817 struct syncache_head *sch; 818 struct socket *so; 819 820 sc = syncache_lookup(inc, &sch); 821 if (sc == NULL) { 822 /* 823 * There is no syncache entry, so see if this ACK is 824 * a returning syncookie. To do this, first: 825 * A. See if this socket has had a syncache entry dropped in 826 * the past. We don't want to accept a bogus syncookie 827 * if we've never received a SYN. 828 * B. check that the syncookie is valid. If it is, then 829 * cobble up a fake syncache entry, and return. 830 */ 831 if (!tcp_syncookies) 832 return (0); 833 sc = syncookie_lookup(inc, th, *sop); 834 if (sc == NULL) 835 return (0); 836 sch = NULL; 837 tcpstat.tcps_sc_recvcookie++; 838 } 839 840 /* 841 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 842 */ 843 if (th->th_ack != sc->sc_iss + 1) 844 return (0); 845 846 so = syncache_socket(sc, *sop, m); 847 if (so == NULL) { 848 #if 0 849 resetandabort: 850 /* XXXjlemon check this - is this correct? */ 851 tcp_respond(NULL, m, m, th, 852 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 853 #endif 854 m_freem(m); /* XXX only needed for above */ 855 tcpstat.tcps_sc_aborted++; 856 } else { 857 tcpstat.tcps_sc_completed++; 858 } 859 if (sch == NULL) 860 syncache_free(sc); 861 else 862 syncache_drop(sc, sch); 863 *sop = so; 864 return (1); 865 } 866 867 /* 868 * Given a LISTEN socket and an inbound SYN request, add 869 * this to the syn cache, and send back a segment: 870 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 871 * to the source. 872 * 873 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 874 * Doing so would require that we hold onto the data and deliver it 875 * to the application. However, if we are the target of a SYN-flood 876 * DoS attack, an attacker could send data which would eventually 877 * consume all available buffer space if it were ACKed. By not ACKing 878 * the data, we avoid this DoS scenario. 879 */ 880 int 881 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 882 struct socket **sop, struct mbuf *m) 883 { 884 struct tcp_syncache_percpu *syncache_percpu; 885 struct tcpcb *tp; 886 struct socket *so; 887 struct syncache *sc = NULL; 888 struct syncache_head *sch; 889 struct mbuf *ipopts = NULL; 890 struct rmxp_tao *taop; 891 int win; 892 893 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 894 so = *sop; 895 tp = sototcpcb(so); 896 897 /* 898 * Remember the IP options, if any. 899 */ 900 #ifdef INET6 901 if (!inc->inc_isipv6) 902 #endif 903 ipopts = ip_srcroute(m); 904 905 /* 906 * See if we already have an entry for this connection. 907 * If we do, resend the SYN,ACK, and reset the retransmit timer. 908 * 909 * XXX 910 * The syncache should be re-initialized with the contents 911 * of the new SYN which may have different options. 912 */ 913 sc = syncache_lookup(inc, &sch); 914 if (sc != NULL) { 915 tcpstat.tcps_sc_dupsyn++; 916 if (ipopts) { 917 /* 918 * If we were remembering a previous source route, 919 * forget it and use the new one we've been given. 920 */ 921 if (sc->sc_ipopts) 922 m_free(sc->sc_ipopts); 923 sc->sc_ipopts = ipopts; 924 } 925 /* 926 * Update timestamp if present. 927 */ 928 if (sc->sc_flags & SCF_TIMESTAMP) 929 sc->sc_tsrecent = to->to_tsval; 930 931 /* Just update the TOF_SACK_PERMITTED for now. */ 932 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 933 sc->sc_flags |= SCF_SACK_PERMITTED; 934 else 935 sc->sc_flags &= ~SCF_SACK_PERMITTED; 936 937 /* 938 * PCB may have changed, pick up new values. 939 */ 940 sc->sc_tp = tp; 941 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 942 if (syncache_respond(sc, m) == 0) { 943 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 944 sc, sc_timerq); 945 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 946 tcpstat.tcps_sndacks++; 947 tcpstat.tcps_sndtotal++; 948 } 949 *sop = NULL; 950 return (1); 951 } 952 953 /* 954 * This allocation is guaranteed to succeed because we 955 * preallocate one more syncache entry than cache_limit. 956 */ 957 sc = zalloc(tcp_syncache.zone); 958 959 /* 960 * Fill in the syncache values. 961 */ 962 sc->sc_tp = tp; 963 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 964 sc->sc_ipopts = ipopts; 965 sc->sc_inc.inc_fport = inc->inc_fport; 966 sc->sc_inc.inc_lport = inc->inc_lport; 967 #ifdef INET6 968 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 969 if (inc->inc_isipv6) { 970 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 971 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 972 sc->sc_route6.ro_rt = NULL; 973 } else 974 #endif 975 { 976 sc->sc_inc.inc_faddr = inc->inc_faddr; 977 sc->sc_inc.inc_laddr = inc->inc_laddr; 978 sc->sc_route.ro_rt = NULL; 979 } 980 sc->sc_irs = th->th_seq; 981 sc->sc_flags = 0; 982 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 983 if (tcp_syncookies) 984 sc->sc_iss = syncookie_generate(sc); 985 else 986 sc->sc_iss = karc4random(); 987 988 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */ 989 win = ssb_space(&so->so_rcv); 990 win = imax(win, 0); 991 win = imin(win, TCP_MAXWIN); 992 sc->sc_wnd = win; 993 994 if (tcp_do_rfc1323) { 995 /* 996 * A timestamp received in a SYN makes 997 * it ok to send timestamp requests and replies. 998 */ 999 if (to->to_flags & TOF_TS) { 1000 sc->sc_tsrecent = to->to_tsval; 1001 sc->sc_flags |= SCF_TIMESTAMP; 1002 } 1003 if (to->to_flags & TOF_SCALE) { 1004 int wscale = 0; 1005 1006 /* Compute proper scaling value from buffer space */ 1007 while (wscale < TCP_MAX_WINSHIFT && 1008 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) 1009 wscale++; 1010 sc->sc_request_r_scale = wscale; 1011 sc->sc_requested_s_scale = to->to_requested_s_scale; 1012 sc->sc_flags |= SCF_WINSCALE; 1013 } 1014 } 1015 if (tcp_do_rfc1644) { 1016 /* 1017 * A CC or CC.new option received in a SYN makes 1018 * it ok to send CC in subsequent segments. 1019 */ 1020 if (to->to_flags & (TOF_CC | TOF_CCNEW)) { 1021 sc->sc_cc_recv = to->to_cc; 1022 sc->sc_cc_send = CC_INC(tcp_ccgen); 1023 sc->sc_flags |= SCF_CC; 1024 } 1025 } 1026 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1027 sc->sc_flags |= SCF_SACK_PERMITTED; 1028 if (tp->t_flags & TF_NOOPT) 1029 sc->sc_flags = SCF_NOOPT; 1030 1031 /* 1032 * XXX 1033 * We have the option here of not doing TAO (even if the segment 1034 * qualifies) and instead fall back to a normal 3WHS via the syncache. 1035 * This allows us to apply synflood protection to TAO-qualifying SYNs 1036 * also. However, there should be a hueristic to determine when to 1037 * do this, and is not present at the moment. 1038 */ 1039 1040 /* 1041 * Perform TAO test on incoming CC (SEG.CC) option, if any. 1042 * - compare SEG.CC against cached CC from the same host, if any. 1043 * - if SEG.CC > chached value, SYN must be new and is accepted 1044 * immediately: save new CC in the cache, mark the socket 1045 * connected, enter ESTABLISHED state, turn on flag to 1046 * send a SYN in the next segment. 1047 * A virtual advertised window is set in rcv_adv to 1048 * initialize SWS prevention. Then enter normal segment 1049 * processing: drop SYN, process data and FIN. 1050 * - otherwise do a normal 3-way handshake. 1051 */ 1052 taop = tcp_gettaocache(&sc->sc_inc); 1053 if (to->to_flags & TOF_CC) { 1054 if ((tp->t_flags & TF_NOPUSH) && 1055 sc->sc_flags & SCF_CC && 1056 taop != NULL && taop->tao_cc != 0 && 1057 CC_GT(to->to_cc, taop->tao_cc)) { 1058 sc->sc_rxtslot = 0; 1059 so = syncache_socket(sc, *sop, m); 1060 if (so != NULL) { 1061 taop->tao_cc = to->to_cc; 1062 *sop = so; 1063 } 1064 syncache_free(sc); 1065 return (so != NULL); 1066 } 1067 } else { 1068 /* 1069 * No CC option, but maybe CC.NEW: invalidate cached value. 1070 */ 1071 if (taop != NULL) 1072 taop->tao_cc = 0; 1073 } 1074 /* 1075 * TAO test failed or there was no CC option, 1076 * do a standard 3-way handshake. 1077 */ 1078 if (syncache_respond(sc, m) == 0) { 1079 syncache_insert(sc, sch); 1080 tcpstat.tcps_sndacks++; 1081 tcpstat.tcps_sndtotal++; 1082 } else { 1083 syncache_free(sc); 1084 tcpstat.tcps_sc_dropped++; 1085 } 1086 *sop = NULL; 1087 return (1); 1088 } 1089 1090 static int 1091 syncache_respond(struct syncache *sc, struct mbuf *m) 1092 { 1093 u_int8_t *optp; 1094 int optlen, error; 1095 u_int16_t tlen, hlen, mssopt; 1096 struct ip *ip = NULL; 1097 struct rtentry *rt; 1098 struct tcphdr *th; 1099 struct ip6_hdr *ip6 = NULL; 1100 #ifdef INET6 1101 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1102 #else 1103 const boolean_t isipv6 = FALSE; 1104 #endif 1105 1106 if (isipv6) { 1107 rt = tcp_rtlookup6(&sc->sc_inc); 1108 if (rt != NULL) 1109 mssopt = rt->rt_ifp->if_mtu - 1110 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1111 else 1112 mssopt = tcp_v6mssdflt; 1113 hlen = sizeof(struct ip6_hdr); 1114 } else { 1115 rt = tcp_rtlookup(&sc->sc_inc); 1116 if (rt != NULL) 1117 mssopt = rt->rt_ifp->if_mtu - 1118 (sizeof(struct ip) + sizeof(struct tcphdr)); 1119 else 1120 mssopt = tcp_mssdflt; 1121 hlen = sizeof(struct ip); 1122 } 1123 1124 /* Compute the size of the TCP options. */ 1125 if (sc->sc_flags & SCF_NOOPT) { 1126 optlen = 0; 1127 } else { 1128 optlen = TCPOLEN_MAXSEG + 1129 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1130 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1131 ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0) + 1132 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1133 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1134 } 1135 tlen = hlen + sizeof(struct tcphdr) + optlen; 1136 1137 /* 1138 * XXX 1139 * assume that the entire packet will fit in a header mbuf 1140 */ 1141 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1142 1143 /* 1144 * XXX shouldn't this reuse the mbuf if possible ? 1145 * Create the IP+TCP header from scratch. 1146 */ 1147 if (m) 1148 m_freem(m); 1149 1150 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 1151 if (m == NULL) 1152 return (ENOBUFS); 1153 m->m_data += max_linkhdr; 1154 m->m_len = tlen; 1155 m->m_pkthdr.len = tlen; 1156 m->m_pkthdr.rcvif = NULL; 1157 1158 if (isipv6) { 1159 ip6 = mtod(m, struct ip6_hdr *); 1160 ip6->ip6_vfc = IPV6_VERSION; 1161 ip6->ip6_nxt = IPPROTO_TCP; 1162 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1163 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1164 ip6->ip6_plen = htons(tlen - hlen); 1165 /* ip6_hlim is set after checksum */ 1166 /* ip6_flow = ??? */ 1167 1168 th = (struct tcphdr *)(ip6 + 1); 1169 } else { 1170 ip = mtod(m, struct ip *); 1171 ip->ip_v = IPVERSION; 1172 ip->ip_hl = sizeof(struct ip) >> 2; 1173 ip->ip_len = tlen; 1174 ip->ip_id = 0; 1175 ip->ip_off = 0; 1176 ip->ip_sum = 0; 1177 ip->ip_p = IPPROTO_TCP; 1178 ip->ip_src = sc->sc_inc.inc_laddr; 1179 ip->ip_dst = sc->sc_inc.inc_faddr; 1180 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1181 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1182 1183 /* 1184 * See if we should do MTU discovery. Route lookups are 1185 * expensive, so we will only unset the DF bit if: 1186 * 1187 * 1) path_mtu_discovery is disabled 1188 * 2) the SCF_UNREACH flag has been set 1189 */ 1190 if (path_mtu_discovery 1191 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1192 ip->ip_off |= IP_DF; 1193 } 1194 1195 th = (struct tcphdr *)(ip + 1); 1196 } 1197 th->th_sport = sc->sc_inc.inc_lport; 1198 th->th_dport = sc->sc_inc.inc_fport; 1199 1200 th->th_seq = htonl(sc->sc_iss); 1201 th->th_ack = htonl(sc->sc_irs + 1); 1202 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1203 th->th_x2 = 0; 1204 th->th_flags = TH_SYN | TH_ACK; 1205 th->th_win = htons(sc->sc_wnd); 1206 th->th_urp = 0; 1207 1208 /* Tack on the TCP options. */ 1209 if (optlen == 0) 1210 goto no_options; 1211 optp = (u_int8_t *)(th + 1); 1212 *optp++ = TCPOPT_MAXSEG; 1213 *optp++ = TCPOLEN_MAXSEG; 1214 *optp++ = (mssopt >> 8) & 0xff; 1215 *optp++ = mssopt & 0xff; 1216 1217 if (sc->sc_flags & SCF_WINSCALE) { 1218 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1219 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1220 sc->sc_request_r_scale); 1221 optp += 4; 1222 } 1223 1224 if (sc->sc_flags & SCF_TIMESTAMP) { 1225 u_int32_t *lp = (u_int32_t *)(optp); 1226 1227 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1228 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1229 *lp++ = htonl(ticks); 1230 *lp = htonl(sc->sc_tsrecent); 1231 optp += TCPOLEN_TSTAMP_APPA; 1232 } 1233 1234 /* 1235 * Send CC and CC.echo if we received CC from our peer. 1236 */ 1237 if (sc->sc_flags & SCF_CC) { 1238 u_int32_t *lp = (u_int32_t *)(optp); 1239 1240 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC)); 1241 *lp++ = htonl(sc->sc_cc_send); 1242 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO)); 1243 *lp = htonl(sc->sc_cc_recv); 1244 optp += TCPOLEN_CC_APPA * 2; 1245 } 1246 1247 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1248 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1249 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1250 } 1251 1252 no_options: 1253 if (isipv6) { 1254 struct route_in6 *ro6 = &sc->sc_route6; 1255 1256 th->th_sum = 0; 1257 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1258 ip6->ip6_hlim = in6_selecthlim(NULL, 1259 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1260 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1261 sc->sc_tp->t_inpcb); 1262 } else { 1263 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1264 htons(tlen - hlen + IPPROTO_TCP)); 1265 m->m_pkthdr.csum_flags = CSUM_TCP; 1266 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1267 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 0, NULL, 1268 sc->sc_tp->t_inpcb); 1269 } 1270 return (error); 1271 } 1272 1273 /* 1274 * cookie layers: 1275 * 1276 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1277 * | peer iss | 1278 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1279 * | 0 |(A)| | 1280 * (A): peer mss index 1281 */ 1282 1283 /* 1284 * The values below are chosen to minimize the size of the tcp_secret 1285 * table, as well as providing roughly a 16 second lifetime for the cookie. 1286 */ 1287 1288 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1289 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1290 1291 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1292 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1293 #define SYNCOOKIE_TIMEOUT \ 1294 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1295 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1296 1297 static struct { 1298 u_int32_t ts_secbits[4]; 1299 u_int ts_expire; 1300 } tcp_secret[SYNCOOKIE_NSECRETS]; 1301 1302 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1303 1304 static MD5_CTX syn_ctx; 1305 1306 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1307 1308 struct md5_add { 1309 u_int32_t laddr, faddr; 1310 u_int32_t secbits[4]; 1311 u_int16_t lport, fport; 1312 }; 1313 1314 #ifdef CTASSERT 1315 CTASSERT(sizeof(struct md5_add) == 28); 1316 #endif 1317 1318 /* 1319 * Consider the problem of a recreated (and retransmitted) cookie. If the 1320 * original SYN was accepted, the connection is established. The second 1321 * SYN is inflight, and if it arrives with an ISN that falls within the 1322 * receive window, the connection is killed. 1323 * 1324 * However, since cookies have other problems, this may not be worth 1325 * worrying about. 1326 */ 1327 1328 static u_int32_t 1329 syncookie_generate(struct syncache *sc) 1330 { 1331 u_int32_t md5_buffer[4]; 1332 u_int32_t data; 1333 int idx, i; 1334 struct md5_add add; 1335 #ifdef INET6 1336 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1337 #else 1338 const boolean_t isipv6 = FALSE; 1339 #endif 1340 1341 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1342 if (tcp_secret[idx].ts_expire < ticks) { 1343 for (i = 0; i < 4; i++) 1344 tcp_secret[idx].ts_secbits[i] = karc4random(); 1345 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1346 } 1347 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1348 if (tcp_msstab[data] <= sc->sc_peer_mss) 1349 break; 1350 data = (data << SYNCOOKIE_WNDBITS) | idx; 1351 data ^= sc->sc_irs; /* peer's iss */ 1352 MD5Init(&syn_ctx); 1353 if (isipv6) { 1354 MD5Add(sc->sc_inc.inc6_laddr); 1355 MD5Add(sc->sc_inc.inc6_faddr); 1356 add.laddr = 0; 1357 add.faddr = 0; 1358 } else { 1359 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1360 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1361 } 1362 add.lport = sc->sc_inc.inc_lport; 1363 add.fport = sc->sc_inc.inc_fport; 1364 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1365 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1366 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1367 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1368 MD5Add(add); 1369 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1370 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1371 return (data); 1372 } 1373 1374 static struct syncache * 1375 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so) 1376 { 1377 u_int32_t md5_buffer[4]; 1378 struct syncache *sc; 1379 u_int32_t data; 1380 int wnd, idx; 1381 struct md5_add add; 1382 1383 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1384 idx = data & SYNCOOKIE_WNDMASK; 1385 if (tcp_secret[idx].ts_expire < ticks || 1386 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1387 return (NULL); 1388 MD5Init(&syn_ctx); 1389 #ifdef INET6 1390 if (inc->inc_isipv6) { 1391 MD5Add(inc->inc6_laddr); 1392 MD5Add(inc->inc6_faddr); 1393 add.laddr = 0; 1394 add.faddr = 0; 1395 } else 1396 #endif 1397 { 1398 add.laddr = inc->inc_laddr.s_addr; 1399 add.faddr = inc->inc_faddr.s_addr; 1400 } 1401 add.lport = inc->inc_lport; 1402 add.fport = inc->inc_fport; 1403 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1404 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1405 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1406 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1407 MD5Add(add); 1408 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1409 data ^= md5_buffer[0]; 1410 if (data & ~SYNCOOKIE_DATAMASK) 1411 return (NULL); 1412 data = data >> SYNCOOKIE_WNDBITS; 1413 1414 /* 1415 * This allocation is guaranteed to succeed because we 1416 * preallocate one more syncache entry than cache_limit. 1417 */ 1418 sc = zalloc(tcp_syncache.zone); 1419 1420 /* 1421 * Fill in the syncache values. 1422 * XXX duplicate code from syncache_add 1423 */ 1424 sc->sc_ipopts = NULL; 1425 sc->sc_inc.inc_fport = inc->inc_fport; 1426 sc->sc_inc.inc_lport = inc->inc_lport; 1427 #ifdef INET6 1428 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1429 if (inc->inc_isipv6) { 1430 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1431 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1432 sc->sc_route6.ro_rt = NULL; 1433 } else 1434 #endif 1435 { 1436 sc->sc_inc.inc_faddr = inc->inc_faddr; 1437 sc->sc_inc.inc_laddr = inc->inc_laddr; 1438 sc->sc_route.ro_rt = NULL; 1439 } 1440 sc->sc_irs = th->th_seq - 1; 1441 sc->sc_iss = th->th_ack - 1; 1442 wnd = ssb_space(&so->so_rcv); 1443 wnd = imax(wnd, 0); 1444 wnd = imin(wnd, TCP_MAXWIN); 1445 sc->sc_wnd = wnd; 1446 sc->sc_flags = 0; 1447 sc->sc_rxtslot = 0; 1448 sc->sc_peer_mss = tcp_msstab[data]; 1449 return (sc); 1450 } 1451