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 1.31 2007/05/24 05:51:29 dillon 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 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 struct vm_zone *zone; 178 u_int hashsize; 179 u_int hashmask; 180 u_int bucket_limit; 181 u_int cache_limit; 182 u_int rexmt_limit; 183 u_int hash_secret; 184 }; 185 static struct tcp_syncache tcp_syncache; 186 187 struct tcp_syncache_percpu { 188 struct syncache_head *hashbase; 189 u_int cache_count; 190 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1]; 191 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1]; 192 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1]; 193 }; 194 static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU]; 195 196 static struct lwkt_port syncache_null_rport; 197 198 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 199 200 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD, 201 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 202 203 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD, 204 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 205 206 /* XXX JH */ 207 #if 0 208 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 209 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 210 #endif 211 212 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD, 213 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 214 215 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 216 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 217 218 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 219 220 #define SYNCACHE_HASH(inc, mask) \ 221 ((tcp_syncache.hash_secret ^ \ 222 (inc)->inc_faddr.s_addr ^ \ 223 ((inc)->inc_faddr.s_addr >> 16) ^ \ 224 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 225 226 #define SYNCACHE_HASH6(inc, mask) \ 227 ((tcp_syncache.hash_secret ^ \ 228 (inc)->inc6_faddr.s6_addr32[0] ^ \ 229 (inc)->inc6_faddr.s6_addr32[3] ^ \ 230 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 231 232 #define ENDPTS_EQ(a, b) ( \ 233 (a)->ie_fport == (b)->ie_fport && \ 234 (a)->ie_lport == (b)->ie_lport && \ 235 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 236 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 237 ) 238 239 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 240 241 static __inline void 242 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu, 243 struct syncache *sc, int slot) 244 { 245 sc->sc_rxtslot = slot; 246 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[slot]; 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 } 255 256 static void 257 syncache_free(struct syncache *sc) 258 { 259 struct rtentry *rt; 260 #ifdef INET6 261 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 262 #else 263 const boolean_t isipv6 = FALSE; 264 #endif 265 266 if (sc->sc_ipopts) 267 m_free(sc->sc_ipopts); 268 269 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt; 270 if (rt != NULL) { 271 /* 272 * If this is the only reference to a protocol-cloned 273 * route, remove it immediately. 274 */ 275 if ((rt->rt_flags & RTF_WASCLONED) && rt->rt_refcnt == 1) 276 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 277 rt_mask(rt), rt->rt_flags, NULL); 278 RTFREE(rt); 279 } 280 281 zfree(tcp_syncache.zone, sc); 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 &syncache_null_rport, 0, 339 syncache_timer_handler); 340 } 341 } 342 343 /* 344 * Allocate the syncache entries. Allow the zone to allocate one 345 * more entry than cache limit, so a new entry can bump out an 346 * older one. 347 */ 348 tcp_syncache.zone = zinit("syncache", sizeof(struct syncache), 349 tcp_syncache.cache_limit * ncpus2, ZONE_INTERRUPT, 0); 350 tcp_syncache.cache_limit -= 1; 351 } 352 353 static void 354 syncache_insert(struct syncache *sc, struct syncache_head *sch) 355 { 356 struct tcp_syncache_percpu *syncache_percpu; 357 struct syncache *sc2; 358 int i; 359 360 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 361 362 /* 363 * Make sure that we don't overflow the per-bucket 364 * limit or the total cache size limit. 365 */ 366 if (sch->sch_length >= tcp_syncache.bucket_limit) { 367 /* 368 * The bucket is full, toss the oldest element. 369 */ 370 sc2 = TAILQ_FIRST(&sch->sch_bucket); 371 sc2->sc_tp->ts_recent = ticks; 372 syncache_drop(sc2, sch); 373 tcpstat.tcps_sc_bucketoverflow++; 374 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) { 375 /* 376 * The cache is full. Toss the oldest entry in the 377 * entire cache. This is the front entry in the 378 * first non-empty timer queue with the largest 379 * timeout value. 380 */ 381 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 382 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]); 383 if (sc2 != NULL) 384 break; 385 } 386 sc2->sc_tp->ts_recent = ticks; 387 syncache_drop(sc2, NULL); 388 tcpstat.tcps_sc_cacheoverflow++; 389 } 390 391 /* Initialize the entry's timer. */ 392 syncache_timeout(syncache_percpu, sc, 0); 393 394 /* Put it into the bucket. */ 395 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 396 sch->sch_length++; 397 syncache_percpu->cache_count++; 398 tcpstat.tcps_sc_added++; 399 } 400 401 static void 402 syncache_drop(struct syncache *sc, struct syncache_head *sch) 403 { 404 struct tcp_syncache_percpu *syncache_percpu; 405 #ifdef INET6 406 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 407 #else 408 const boolean_t isipv6 = FALSE; 409 #endif 410 411 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 412 413 if (sch == NULL) { 414 if (isipv6) { 415 sch = &syncache_percpu->hashbase[ 416 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 417 } else { 418 sch = &syncache_percpu->hashbase[ 419 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 420 } 421 } 422 423 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 424 sch->sch_length--; 425 syncache_percpu->cache_count--; 426 427 /* 428 * Remove the entry from the syncache timer/timeout queue. Note 429 * that we do not try to stop any running timer since we do not know 430 * whether the timer's message is in-transit or not. Since timeouts 431 * are fairly long, taking an unneeded callout does not detrimentally 432 * effect performance. 433 */ 434 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq); 435 436 syncache_free(sc); 437 } 438 439 /* 440 * Place a timeout message on the TCP thread's message queue. 441 * This routine runs in soft interrupt context. 442 * 443 * An invariant is for this routine to be called, the callout must 444 * have been active. Note that the callout is not deactivated until 445 * after the message has been processed in syncache_timer_handler() below. 446 */ 447 static void 448 syncache_timer(void *p) 449 { 450 struct netmsg_sc_timer *msg = p; 451 452 lwkt_sendmsg(msg->nm_mrec->port, &msg->nm_netmsg.nm_lmsg); 453 } 454 455 /* 456 * Service a timer message queued by timer expiration. 457 * This routine runs in the TCP protocol thread. 458 * 459 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 460 * If we have retransmitted an entry the maximum number of times, expire it. 461 * 462 * When we finish processing timed-out entries, we restart the timer if there 463 * are any entries still on the queue and deactivate it otherwise. Only after 464 * a timer has been deactivated here can it be restarted by syncache_timeout(). 465 */ 466 static void 467 syncache_timer_handler(netmsg_t netmsg) 468 { 469 struct tcp_syncache_percpu *syncache_percpu; 470 struct syncache *sc, *nsc; 471 struct inpcb *inp; 472 int slot; 473 474 slot = ((struct netmsg_sc_timer *)netmsg)->nm_mrec->slot; 475 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 476 477 nsc = TAILQ_FIRST(&syncache_percpu->timerq[slot]); 478 while (nsc != NULL) { 479 if (ticks < nsc->sc_rxttime) 480 break; /* finished because timerq sorted by time */ 481 sc = nsc; 482 inp = sc->sc_tp->t_inpcb; 483 if (slot == SYNCACHE_MAXREXMTS || 484 slot >= tcp_syncache.rexmt_limit || 485 inp->inp_gencnt != sc->sc_inp_gencnt) { 486 nsc = TAILQ_NEXT(sc, sc_timerq); 487 syncache_drop(sc, NULL); 488 tcpstat.tcps_sc_stale++; 489 continue; 490 } 491 /* 492 * syncache_respond() may call back into the syncache to 493 * to modify another entry, so do not obtain the next 494 * entry on the timer chain until it has completed. 495 */ 496 syncache_respond(sc, NULL); 497 nsc = TAILQ_NEXT(sc, sc_timerq); 498 tcpstat.tcps_sc_retransmitted++; 499 TAILQ_REMOVE(&syncache_percpu->timerq[slot], sc, sc_timerq); 500 syncache_timeout(syncache_percpu, sc, slot + 1); 501 } 502 if (nsc != NULL) 503 callout_reset(&syncache_percpu->tt_timerq[slot], 504 nsc->sc_rxttime - ticks, syncache_timer, 505 &syncache_percpu->mrec[slot]); 506 else 507 callout_deactivate(&syncache_percpu->tt_timerq[slot]); 508 509 lwkt_replymsg(&netmsg->nm_lmsg, 0); 510 } 511 512 /* 513 * Find an entry in the syncache. 514 */ 515 struct syncache * 516 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 517 { 518 struct tcp_syncache_percpu *syncache_percpu; 519 struct syncache *sc; 520 struct syncache_head *sch; 521 522 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 523 #ifdef INET6 524 if (inc->inc_isipv6) { 525 sch = &syncache_percpu->hashbase[ 526 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 527 *schp = sch; 528 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) 529 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 530 return (sc); 531 } else 532 #endif 533 { 534 sch = &syncache_percpu->hashbase[ 535 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 536 *schp = sch; 537 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 538 #ifdef INET6 539 if (sc->sc_inc.inc_isipv6) 540 continue; 541 #endif 542 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 543 return (sc); 544 } 545 } 546 return (NULL); 547 } 548 549 /* 550 * This function is called when we get a RST for a 551 * non-existent connection, so that we can see if the 552 * connection is in the syn cache. If it is, zap it. 553 */ 554 void 555 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 556 { 557 struct syncache *sc; 558 struct syncache_head *sch; 559 560 sc = syncache_lookup(inc, &sch); 561 if (sc == NULL) 562 return; 563 /* 564 * If the RST bit is set, check the sequence number to see 565 * if this is a valid reset segment. 566 * RFC 793 page 37: 567 * In all states except SYN-SENT, all reset (RST) segments 568 * are validated by checking their SEQ-fields. A reset is 569 * valid if its sequence number is in the window. 570 * 571 * The sequence number in the reset segment is normally an 572 * echo of our outgoing acknowlegement numbers, but some hosts 573 * send a reset with the sequence number at the rightmost edge 574 * of our receive window, and we have to handle this case. 575 */ 576 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 577 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 578 syncache_drop(sc, sch); 579 tcpstat.tcps_sc_reset++; 580 } 581 } 582 583 void 584 syncache_badack(struct in_conninfo *inc) 585 { 586 struct syncache *sc; 587 struct syncache_head *sch; 588 589 sc = syncache_lookup(inc, &sch); 590 if (sc != NULL) { 591 syncache_drop(sc, sch); 592 tcpstat.tcps_sc_badack++; 593 } 594 } 595 596 void 597 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 598 { 599 struct syncache *sc; 600 struct syncache_head *sch; 601 602 /* we are called at splnet() here */ 603 sc = syncache_lookup(inc, &sch); 604 if (sc == NULL) 605 return; 606 607 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 608 if (ntohl(th->th_seq) != sc->sc_iss) 609 return; 610 611 /* 612 * If we've rertransmitted 3 times and this is our second error, 613 * we remove the entry. Otherwise, we allow it to continue on. 614 * This prevents us from incorrectly nuking an entry during a 615 * spurious network outage. 616 * 617 * See tcp_notify(). 618 */ 619 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 620 sc->sc_flags |= SCF_UNREACH; 621 return; 622 } 623 syncache_drop(sc, sch); 624 tcpstat.tcps_sc_unreach++; 625 } 626 627 /* 628 * Build a new TCP socket structure from a syncache entry. 629 */ 630 static struct socket * 631 syncache_socket(struct syncache *sc, struct socket *lso) 632 { 633 struct inpcb *inp = NULL, *linp; 634 struct socket *so; 635 struct tcpcb *tp; 636 #ifdef INET6 637 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 638 #else 639 const boolean_t isipv6 = FALSE; 640 #endif 641 642 /* 643 * Ok, create the full blown connection, and set things up 644 * as they would have been set up if we had created the 645 * connection when the SYN arrived. If we can't create 646 * the connection, abort it. 647 */ 648 so = sonewconn(lso, SS_ISCONNECTED); 649 if (so == NULL) { 650 /* 651 * Drop the connection; we will send a RST if the peer 652 * retransmits the ACK, 653 */ 654 tcpstat.tcps_listendrop++; 655 goto abort; 656 } 657 658 inp = so->so_pcb; 659 660 /* 661 * Insert new socket into hash list. 662 */ 663 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 664 if (isipv6) { 665 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 666 } else { 667 #ifdef INET6 668 inp->inp_vflag &= ~INP_IPV6; 669 inp->inp_vflag |= INP_IPV4; 670 #endif 671 inp->inp_laddr = sc->sc_inc.inc_laddr; 672 } 673 inp->inp_lport = sc->sc_inc.inc_lport; 674 if (in_pcbinsporthash(inp) != 0) { 675 /* 676 * Undo the assignments above if we failed to 677 * put the PCB on the hash lists. 678 */ 679 if (isipv6) 680 inp->in6p_laddr = kin6addr_any; 681 else 682 inp->inp_laddr.s_addr = INADDR_ANY; 683 inp->inp_lport = 0; 684 goto abort; 685 } 686 linp = so->so_pcb; 687 #ifdef IPSEC 688 /* copy old policy into new socket's */ 689 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 690 kprintf("syncache_expand: could not copy policy\n"); 691 #endif 692 if (isipv6) { 693 struct in6_addr laddr6; 694 struct sockaddr_in6 sin6; 695 /* 696 * Inherit socket options from the listening socket. 697 * Note that in6p_inputopts are not (and should not be) 698 * copied, since it stores previously received options and is 699 * used to detect if each new option is different than the 700 * previous one and hence should be passed to a user. 701 * If we copied in6p_inputopts, a user would not be able to 702 * receive options just after calling the accept system call. 703 */ 704 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 705 if (linp->in6p_outputopts) 706 inp->in6p_outputopts = 707 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 708 inp->in6p_route = sc->sc_route6; 709 sc->sc_route6.ro_rt = NULL; 710 711 sin6.sin6_family = AF_INET6; 712 sin6.sin6_len = sizeof sin6; 713 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 714 sin6.sin6_port = sc->sc_inc.inc_fport; 715 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 716 laddr6 = inp->in6p_laddr; 717 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 718 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 719 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) { 720 inp->in6p_laddr = laddr6; 721 goto abort; 722 } 723 } else { 724 struct in_addr laddr; 725 struct sockaddr_in sin; 726 727 inp->inp_options = ip_srcroute(); 728 if (inp->inp_options == NULL) { 729 inp->inp_options = sc->sc_ipopts; 730 sc->sc_ipopts = NULL; 731 } 732 inp->inp_route = sc->sc_route; 733 sc->sc_route.ro_rt = NULL; 734 735 sin.sin_family = AF_INET; 736 sin.sin_len = sizeof sin; 737 sin.sin_addr = sc->sc_inc.inc_faddr; 738 sin.sin_port = sc->sc_inc.inc_fport; 739 bzero(sin.sin_zero, sizeof sin.sin_zero); 740 laddr = inp->inp_laddr; 741 if (inp->inp_laddr.s_addr == INADDR_ANY) 742 inp->inp_laddr = sc->sc_inc.inc_laddr; 743 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) { 744 inp->inp_laddr = laddr; 745 goto abort; 746 } 747 } 748 749 tp = intotcpcb(inp); 750 tp->t_state = TCPS_SYN_RECEIVED; 751 tp->iss = sc->sc_iss; 752 tp->irs = sc->sc_irs; 753 tcp_rcvseqinit(tp); 754 tcp_sendseqinit(tp); 755 tp->snd_wl1 = sc->sc_irs; 756 tp->rcv_up = sc->sc_irs + 1; 757 tp->rcv_wnd = sc->sc_wnd; 758 tp->rcv_adv += tp->rcv_wnd; 759 760 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 761 if (sc->sc_flags & SCF_NOOPT) 762 tp->t_flags |= TF_NOOPT; 763 if (sc->sc_flags & SCF_WINSCALE) { 764 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 765 tp->requested_s_scale = sc->sc_requested_s_scale; 766 tp->request_r_scale = sc->sc_request_r_scale; 767 } 768 if (sc->sc_flags & SCF_TIMESTAMP) { 769 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 770 tp->ts_recent = sc->sc_tsrecent; 771 tp->ts_recent_age = ticks; 772 } 773 if (sc->sc_flags & SCF_CC) { 774 /* 775 * Initialization of the tcpcb for transaction; 776 * set SND.WND = SEG.WND, 777 * initialize CCsend and CCrecv. 778 */ 779 tp->t_flags |= TF_REQ_CC | TF_RCVD_CC; 780 tp->cc_send = sc->sc_cc_send; 781 tp->cc_recv = sc->sc_cc_recv; 782 } 783 if (sc->sc_flags & SCF_SACK_PERMITTED) 784 tp->t_flags |= TF_SACK_PERMITTED; 785 786 tcp_mss(tp, sc->sc_peer_mss); 787 788 /* 789 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 790 */ 791 if (sc->sc_rxtslot != 0) 792 tp->snd_cwnd = tp->t_maxseg; 793 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 794 795 tcpstat.tcps_accepts++; 796 return (so); 797 798 abort: 799 if (so != NULL) 800 soabort(so); 801 return (NULL); 802 } 803 804 /* 805 * This function gets called when we receive an ACK for a 806 * socket in the LISTEN state. We look up the connection 807 * in the syncache, and if its there, we pull it out of 808 * the cache and turn it into a full-blown connection in 809 * the SYN-RECEIVED state. 810 */ 811 int 812 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop, 813 struct mbuf *m) 814 { 815 struct syncache *sc; 816 struct syncache_head *sch; 817 struct socket *so; 818 819 sc = syncache_lookup(inc, &sch); 820 if (sc == NULL) { 821 /* 822 * There is no syncache entry, so see if this ACK is 823 * a returning syncookie. To do this, first: 824 * A. See if this socket has had a syncache entry dropped in 825 * the past. We don't want to accept a bogus syncookie 826 * if we've never received a SYN. 827 * B. check that the syncookie is valid. If it is, then 828 * cobble up a fake syncache entry, and return. 829 */ 830 if (!tcp_syncookies) 831 return (0); 832 sc = syncookie_lookup(inc, th, *sop); 833 if (sc == NULL) 834 return (0); 835 sch = NULL; 836 tcpstat.tcps_sc_recvcookie++; 837 } 838 839 /* 840 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 841 */ 842 if (th->th_ack != sc->sc_iss + 1) 843 return (0); 844 845 so = syncache_socket(sc, *sop); 846 if (so == NULL) { 847 #if 0 848 resetandabort: 849 /* XXXjlemon check this - is this correct? */ 850 tcp_respond(NULL, m, m, th, 851 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 852 #endif 853 m_freem(m); /* XXX only needed for above */ 854 tcpstat.tcps_sc_aborted++; 855 } else { 856 tcpstat.tcps_sc_completed++; 857 } 858 if (sch == NULL) 859 syncache_free(sc); 860 else 861 syncache_drop(sc, sch); 862 *sop = so; 863 return (1); 864 } 865 866 /* 867 * Given a LISTEN socket and an inbound SYN request, add 868 * this to the syn cache, and send back a segment: 869 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 870 * to the source. 871 * 872 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 873 * Doing so would require that we hold onto the data and deliver it 874 * to the application. However, if we are the target of a SYN-flood 875 * DoS attack, an attacker could send data which would eventually 876 * consume all available buffer space if it were ACKed. By not ACKing 877 * the data, we avoid this DoS scenario. 878 */ 879 int 880 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 881 struct socket **sop, struct mbuf *m) 882 { 883 struct tcp_syncache_percpu *syncache_percpu; 884 struct tcpcb *tp; 885 struct socket *so; 886 struct syncache *sc = NULL; 887 struct syncache_head *sch; 888 struct mbuf *ipopts = NULL; 889 struct rmxp_tao *taop; 890 int win; 891 892 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 893 so = *sop; 894 tp = sototcpcb(so); 895 896 /* 897 * Remember the IP options, if any. 898 */ 899 #ifdef INET6 900 if (!inc->inc_isipv6) 901 #endif 902 ipopts = ip_srcroute(); 903 904 /* 905 * See if we already have an entry for this connection. 906 * If we do, resend the SYN,ACK, and reset the retransmit timer. 907 * 908 * XXX 909 * The syncache should be re-initialized with the contents 910 * of the new SYN which may have different options. 911 */ 912 sc = syncache_lookup(inc, &sch); 913 if (sc != NULL) { 914 tcpstat.tcps_sc_dupsyn++; 915 if (ipopts) { 916 /* 917 * If we were remembering a previous source route, 918 * forget it and use the new one we've been given. 919 */ 920 if (sc->sc_ipopts) 921 m_free(sc->sc_ipopts); 922 sc->sc_ipopts = ipopts; 923 } 924 /* 925 * Update timestamp if present. 926 */ 927 if (sc->sc_flags & SCF_TIMESTAMP) 928 sc->sc_tsrecent = to->to_tsval; 929 930 /* Just update the TOF_SACK_PERMITTED for now. */ 931 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 932 sc->sc_flags |= SCF_SACK_PERMITTED; 933 else 934 sc->sc_flags &= ~SCF_SACK_PERMITTED; 935 936 /* 937 * PCB may have changed, pick up new values. 938 */ 939 sc->sc_tp = tp; 940 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 941 if (syncache_respond(sc, m) == 0) { 942 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 943 sc, sc_timerq); 944 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 945 tcpstat.tcps_sndacks++; 946 tcpstat.tcps_sndtotal++; 947 } 948 *sop = NULL; 949 return (1); 950 } 951 952 /* 953 * This allocation is guaranteed to succeed because we 954 * preallocate one more syncache entry than cache_limit. 955 */ 956 sc = zalloc(tcp_syncache.zone); 957 958 /* 959 * Fill in the syncache values. 960 */ 961 sc->sc_tp = tp; 962 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 963 sc->sc_ipopts = ipopts; 964 sc->sc_inc.inc_fport = inc->inc_fport; 965 sc->sc_inc.inc_lport = inc->inc_lport; 966 #ifdef INET6 967 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 968 if (inc->inc_isipv6) { 969 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 970 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 971 sc->sc_route6.ro_rt = NULL; 972 } else 973 #endif 974 { 975 sc->sc_inc.inc_faddr = inc->inc_faddr; 976 sc->sc_inc.inc_laddr = inc->inc_laddr; 977 sc->sc_route.ro_rt = NULL; 978 } 979 sc->sc_irs = th->th_seq; 980 sc->sc_flags = 0; 981 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 982 if (tcp_syncookies) 983 sc->sc_iss = syncookie_generate(sc); 984 else 985 sc->sc_iss = karc4random(); 986 987 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */ 988 win = ssb_space(&so->so_rcv); 989 win = imax(win, 0); 990 win = imin(win, TCP_MAXWIN); 991 sc->sc_wnd = win; 992 993 if (tcp_do_rfc1323) { 994 /* 995 * A timestamp received in a SYN makes 996 * it ok to send timestamp requests and replies. 997 */ 998 if (to->to_flags & TOF_TS) { 999 sc->sc_tsrecent = to->to_tsval; 1000 sc->sc_flags |= SCF_TIMESTAMP; 1001 } 1002 if (to->to_flags & TOF_SCALE) { 1003 int wscale = 0; 1004 1005 /* Compute proper scaling value from buffer space */ 1006 while (wscale < TCP_MAX_WINSHIFT && 1007 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) 1008 wscale++; 1009 sc->sc_request_r_scale = wscale; 1010 sc->sc_requested_s_scale = to->to_requested_s_scale; 1011 sc->sc_flags |= SCF_WINSCALE; 1012 } 1013 } 1014 if (tcp_do_rfc1644) { 1015 /* 1016 * A CC or CC.new option received in a SYN makes 1017 * it ok to send CC in subsequent segments. 1018 */ 1019 if (to->to_flags & (TOF_CC | TOF_CCNEW)) { 1020 sc->sc_cc_recv = to->to_cc; 1021 sc->sc_cc_send = CC_INC(tcp_ccgen); 1022 sc->sc_flags |= SCF_CC; 1023 } 1024 } 1025 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1026 sc->sc_flags |= SCF_SACK_PERMITTED; 1027 if (tp->t_flags & TF_NOOPT) 1028 sc->sc_flags = SCF_NOOPT; 1029 1030 /* 1031 * XXX 1032 * We have the option here of not doing TAO (even if the segment 1033 * qualifies) and instead fall back to a normal 3WHS via the syncache. 1034 * This allows us to apply synflood protection to TAO-qualifying SYNs 1035 * also. However, there should be a hueristic to determine when to 1036 * do this, and is not present at the moment. 1037 */ 1038 1039 /* 1040 * Perform TAO test on incoming CC (SEG.CC) option, if any. 1041 * - compare SEG.CC against cached CC from the same host, if any. 1042 * - if SEG.CC > chached value, SYN must be new and is accepted 1043 * immediately: save new CC in the cache, mark the socket 1044 * connected, enter ESTABLISHED state, turn on flag to 1045 * send a SYN in the next segment. 1046 * A virtual advertised window is set in rcv_adv to 1047 * initialize SWS prevention. Then enter normal segment 1048 * processing: drop SYN, process data and FIN. 1049 * - otherwise do a normal 3-way handshake. 1050 */ 1051 taop = tcp_gettaocache(&sc->sc_inc); 1052 if (to->to_flags & TOF_CC) { 1053 if ((tp->t_flags & TF_NOPUSH) && 1054 sc->sc_flags & SCF_CC && 1055 taop != NULL && taop->tao_cc != 0 && 1056 CC_GT(to->to_cc, taop->tao_cc)) { 1057 sc->sc_rxtslot = 0; 1058 so = syncache_socket(sc, *sop); 1059 if (so != NULL) { 1060 taop->tao_cc = to->to_cc; 1061 *sop = so; 1062 } 1063 syncache_free(sc); 1064 return (so != NULL); 1065 } 1066 } else { 1067 /* 1068 * No CC option, but maybe CC.NEW: invalidate cached value. 1069 */ 1070 if (taop != NULL) 1071 taop->tao_cc = 0; 1072 } 1073 /* 1074 * TAO test failed or there was no CC option, 1075 * do a standard 3-way handshake. 1076 */ 1077 if (syncache_respond(sc, m) == 0) { 1078 syncache_insert(sc, sch); 1079 tcpstat.tcps_sndacks++; 1080 tcpstat.tcps_sndtotal++; 1081 } else { 1082 syncache_free(sc); 1083 tcpstat.tcps_sc_dropped++; 1084 } 1085 *sop = NULL; 1086 return (1); 1087 } 1088 1089 static int 1090 syncache_respond(struct syncache *sc, struct mbuf *m) 1091 { 1092 u_int8_t *optp; 1093 int optlen, error; 1094 u_int16_t tlen, hlen, mssopt; 1095 struct ip *ip = NULL; 1096 struct rtentry *rt; 1097 struct tcphdr *th; 1098 struct ip6_hdr *ip6 = NULL; 1099 #ifdef INET6 1100 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1101 #else 1102 const boolean_t isipv6 = FALSE; 1103 #endif 1104 1105 if (isipv6) { 1106 rt = tcp_rtlookup6(&sc->sc_inc); 1107 if (rt != NULL) 1108 mssopt = rt->rt_ifp->if_mtu - 1109 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1110 else 1111 mssopt = tcp_v6mssdflt; 1112 hlen = sizeof(struct ip6_hdr); 1113 } else { 1114 rt = tcp_rtlookup(&sc->sc_inc); 1115 if (rt != NULL) 1116 mssopt = rt->rt_ifp->if_mtu - 1117 (sizeof(struct ip) + sizeof(struct tcphdr)); 1118 else 1119 mssopt = tcp_mssdflt; 1120 hlen = sizeof(struct ip); 1121 } 1122 1123 /* Compute the size of the TCP options. */ 1124 if (sc->sc_flags & SCF_NOOPT) { 1125 optlen = 0; 1126 } else { 1127 optlen = TCPOLEN_MAXSEG + 1128 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1129 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1130 ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0) + 1131 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1132 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1133 } 1134 tlen = hlen + sizeof(struct tcphdr) + optlen; 1135 1136 /* 1137 * XXX 1138 * assume that the entire packet will fit in a header mbuf 1139 */ 1140 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1141 1142 /* 1143 * XXX shouldn't this reuse the mbuf if possible ? 1144 * Create the IP+TCP header from scratch. 1145 */ 1146 if (m) 1147 m_freem(m); 1148 1149 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 1150 if (m == NULL) 1151 return (ENOBUFS); 1152 m->m_data += max_linkhdr; 1153 m->m_len = tlen; 1154 m->m_pkthdr.len = tlen; 1155 m->m_pkthdr.rcvif = NULL; 1156 1157 if (isipv6) { 1158 ip6 = mtod(m, struct ip6_hdr *); 1159 ip6->ip6_vfc = IPV6_VERSION; 1160 ip6->ip6_nxt = IPPROTO_TCP; 1161 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1162 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1163 ip6->ip6_plen = htons(tlen - hlen); 1164 /* ip6_hlim is set after checksum */ 1165 /* ip6_flow = ??? */ 1166 1167 th = (struct tcphdr *)(ip6 + 1); 1168 } else { 1169 ip = mtod(m, struct ip *); 1170 ip->ip_v = IPVERSION; 1171 ip->ip_hl = sizeof(struct ip) >> 2; 1172 ip->ip_len = tlen; 1173 ip->ip_id = 0; 1174 ip->ip_off = 0; 1175 ip->ip_sum = 0; 1176 ip->ip_p = IPPROTO_TCP; 1177 ip->ip_src = sc->sc_inc.inc_laddr; 1178 ip->ip_dst = sc->sc_inc.inc_faddr; 1179 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1180 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1181 1182 /* 1183 * See if we should do MTU discovery. Route lookups are 1184 * expensive, so we will only unset the DF bit if: 1185 * 1186 * 1) path_mtu_discovery is disabled 1187 * 2) the SCF_UNREACH flag has been set 1188 */ 1189 if (path_mtu_discovery 1190 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1191 ip->ip_off |= IP_DF; 1192 } 1193 1194 th = (struct tcphdr *)(ip + 1); 1195 } 1196 th->th_sport = sc->sc_inc.inc_lport; 1197 th->th_dport = sc->sc_inc.inc_fport; 1198 1199 th->th_seq = htonl(sc->sc_iss); 1200 th->th_ack = htonl(sc->sc_irs + 1); 1201 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1202 th->th_x2 = 0; 1203 th->th_flags = TH_SYN | TH_ACK; 1204 th->th_win = htons(sc->sc_wnd); 1205 th->th_urp = 0; 1206 1207 /* Tack on the TCP options. */ 1208 if (optlen == 0) 1209 goto no_options; 1210 optp = (u_int8_t *)(th + 1); 1211 *optp++ = TCPOPT_MAXSEG; 1212 *optp++ = TCPOLEN_MAXSEG; 1213 *optp++ = (mssopt >> 8) & 0xff; 1214 *optp++ = mssopt & 0xff; 1215 1216 if (sc->sc_flags & SCF_WINSCALE) { 1217 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1218 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1219 sc->sc_request_r_scale); 1220 optp += 4; 1221 } 1222 1223 if (sc->sc_flags & SCF_TIMESTAMP) { 1224 u_int32_t *lp = (u_int32_t *)(optp); 1225 1226 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1227 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1228 *lp++ = htonl(ticks); 1229 *lp = htonl(sc->sc_tsrecent); 1230 optp += TCPOLEN_TSTAMP_APPA; 1231 } 1232 1233 /* 1234 * Send CC and CC.echo if we received CC from our peer. 1235 */ 1236 if (sc->sc_flags & SCF_CC) { 1237 u_int32_t *lp = (u_int32_t *)(optp); 1238 1239 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC)); 1240 *lp++ = htonl(sc->sc_cc_send); 1241 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO)); 1242 *lp = htonl(sc->sc_cc_recv); 1243 optp += TCPOLEN_CC_APPA * 2; 1244 } 1245 1246 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1247 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1248 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1249 } 1250 1251 no_options: 1252 if (isipv6) { 1253 struct route_in6 *ro6 = &sc->sc_route6; 1254 1255 th->th_sum = 0; 1256 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1257 ip6->ip6_hlim = in6_selecthlim(NULL, 1258 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1259 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1260 sc->sc_tp->t_inpcb); 1261 } else { 1262 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1263 htons(tlen - hlen + IPPROTO_TCP)); 1264 m->m_pkthdr.csum_flags = CSUM_TCP; 1265 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1266 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 0, NULL, 1267 sc->sc_tp->t_inpcb); 1268 } 1269 return (error); 1270 } 1271 1272 /* 1273 * cookie layers: 1274 * 1275 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1276 * | peer iss | 1277 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1278 * | 0 |(A)| | 1279 * (A): peer mss index 1280 */ 1281 1282 /* 1283 * The values below are chosen to minimize the size of the tcp_secret 1284 * table, as well as providing roughly a 16 second lifetime for the cookie. 1285 */ 1286 1287 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1288 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1289 1290 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1291 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1292 #define SYNCOOKIE_TIMEOUT \ 1293 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1294 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1295 1296 static struct { 1297 u_int32_t ts_secbits[4]; 1298 u_int ts_expire; 1299 } tcp_secret[SYNCOOKIE_NSECRETS]; 1300 1301 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1302 1303 static MD5_CTX syn_ctx; 1304 1305 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1306 1307 struct md5_add { 1308 u_int32_t laddr, faddr; 1309 u_int32_t secbits[4]; 1310 u_int16_t lport, fport; 1311 }; 1312 1313 #ifdef CTASSERT 1314 CTASSERT(sizeof(struct md5_add) == 28); 1315 #endif 1316 1317 /* 1318 * Consider the problem of a recreated (and retransmitted) cookie. If the 1319 * original SYN was accepted, the connection is established. The second 1320 * SYN is inflight, and if it arrives with an ISN that falls within the 1321 * receive window, the connection is killed. 1322 * 1323 * However, since cookies have other problems, this may not be worth 1324 * worrying about. 1325 */ 1326 1327 static u_int32_t 1328 syncookie_generate(struct syncache *sc) 1329 { 1330 u_int32_t md5_buffer[4]; 1331 u_int32_t data; 1332 int idx, i; 1333 struct md5_add add; 1334 #ifdef INET6 1335 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1336 #else 1337 const boolean_t isipv6 = FALSE; 1338 #endif 1339 1340 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1341 if (tcp_secret[idx].ts_expire < ticks) { 1342 for (i = 0; i < 4; i++) 1343 tcp_secret[idx].ts_secbits[i] = karc4random(); 1344 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1345 } 1346 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1347 if (tcp_msstab[data] <= sc->sc_peer_mss) 1348 break; 1349 data = (data << SYNCOOKIE_WNDBITS) | idx; 1350 data ^= sc->sc_irs; /* peer's iss */ 1351 MD5Init(&syn_ctx); 1352 if (isipv6) { 1353 MD5Add(sc->sc_inc.inc6_laddr); 1354 MD5Add(sc->sc_inc.inc6_faddr); 1355 add.laddr = 0; 1356 add.faddr = 0; 1357 } else { 1358 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1359 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1360 } 1361 add.lport = sc->sc_inc.inc_lport; 1362 add.fport = sc->sc_inc.inc_fport; 1363 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1364 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1365 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1366 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1367 MD5Add(add); 1368 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1369 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1370 return (data); 1371 } 1372 1373 static struct syncache * 1374 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so) 1375 { 1376 u_int32_t md5_buffer[4]; 1377 struct syncache *sc; 1378 u_int32_t data; 1379 int wnd, idx; 1380 struct md5_add add; 1381 1382 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1383 idx = data & SYNCOOKIE_WNDMASK; 1384 if (tcp_secret[idx].ts_expire < ticks || 1385 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1386 return (NULL); 1387 MD5Init(&syn_ctx); 1388 #ifdef INET6 1389 if (inc->inc_isipv6) { 1390 MD5Add(inc->inc6_laddr); 1391 MD5Add(inc->inc6_faddr); 1392 add.laddr = 0; 1393 add.faddr = 0; 1394 } else 1395 #endif 1396 { 1397 add.laddr = inc->inc_laddr.s_addr; 1398 add.faddr = inc->inc_faddr.s_addr; 1399 } 1400 add.lport = inc->inc_lport; 1401 add.fport = inc->inc_fport; 1402 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1403 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1404 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1405 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1406 MD5Add(add); 1407 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1408 data ^= md5_buffer[0]; 1409 if (data & ~SYNCOOKIE_DATAMASK) 1410 return (NULL); 1411 data = data >> SYNCOOKIE_WNDBITS; 1412 1413 /* 1414 * This allocation is guaranteed to succeed because we 1415 * preallocate one more syncache entry than cache_limit. 1416 */ 1417 sc = zalloc(tcp_syncache.zone); 1418 1419 /* 1420 * Fill in the syncache values. 1421 * XXX duplicate code from syncache_add 1422 */ 1423 sc->sc_ipopts = NULL; 1424 sc->sc_inc.inc_fport = inc->inc_fport; 1425 sc->sc_inc.inc_lport = inc->inc_lport; 1426 #ifdef INET6 1427 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1428 if (inc->inc_isipv6) { 1429 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1430 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1431 sc->sc_route6.ro_rt = NULL; 1432 } else 1433 #endif 1434 { 1435 sc->sc_inc.inc_faddr = inc->inc_faddr; 1436 sc->sc_inc.inc_laddr = inc->inc_laddr; 1437 sc->sc_route.ro_rt = NULL; 1438 } 1439 sc->sc_irs = th->th_seq - 1; 1440 sc->sc_iss = th->th_ack - 1; 1441 wnd = ssb_space(&so->so_rcv); 1442 wnd = imax(wnd, 0); 1443 wnd = imin(wnd, TCP_MAXWIN); 1444 sc->sc_wnd = wnd; 1445 sc->sc_flags = 0; 1446 sc->sc_rxtslot = 0; 1447 sc->sc_peer_mss = tcp_msstab[data]; 1448 return (sc); 1449 } 1450