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 #include <vm/vm_zone.h> 136 137 static int tcp_syncookies = 1; 138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 139 &tcp_syncookies, 0, 140 "Use TCP SYN cookies if the syncache overflows"); 141 142 static void syncache_drop(struct syncache *, struct syncache_head *); 143 static void syncache_free(struct syncache *); 144 static void syncache_insert(struct syncache *, struct syncache_head *); 145 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 146 static int syncache_respond(struct syncache *, struct mbuf *); 147 static struct socket *syncache_socket(struct syncache *, struct socket *, 148 struct mbuf *); 149 static void syncache_timer(void *); 150 static u_int32_t syncookie_generate(struct syncache *); 151 static struct syncache *syncookie_lookup(struct in_conninfo *, 152 struct tcphdr *, struct socket *); 153 154 /* 155 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 156 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 157 * the odds are that the user has given up attempting to connect by then. 158 */ 159 #define SYNCACHE_MAXREXMTS 3 160 161 /* Arbitrary values */ 162 #define TCP_SYNCACHE_HASHSIZE 512 163 #define TCP_SYNCACHE_BUCKETLIMIT 30 164 165 struct netmsg_sc_timer { 166 struct netmsg nm_netmsg; 167 struct msgrec *nm_mrec; /* back pointer to containing msgrec */ 168 }; 169 170 struct msgrec { 171 struct netmsg_sc_timer msg; 172 lwkt_port_t port; /* constant after init */ 173 int slot; /* constant after init */ 174 }; 175 176 static void syncache_timer_handler(netmsg_t); 177 178 struct tcp_syncache { 179 struct vm_zone *zone; 180 u_int hashsize; 181 u_int hashmask; 182 u_int bucket_limit; 183 u_int cache_limit; 184 u_int rexmt_limit; 185 u_int hash_secret; 186 }; 187 static struct tcp_syncache tcp_syncache; 188 189 struct tcp_syncache_percpu { 190 struct syncache_head *hashbase; 191 u_int cache_count; 192 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1]; 193 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1]; 194 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1]; 195 }; 196 static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU]; 197 198 static struct lwkt_port syncache_null_rport; 199 200 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 201 202 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD, 203 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 204 205 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD, 206 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 207 208 /* XXX JH */ 209 #if 0 210 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 211 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 212 #endif 213 214 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD, 215 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 216 217 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 218 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 219 220 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 221 222 #define SYNCACHE_HASH(inc, mask) \ 223 ((tcp_syncache.hash_secret ^ \ 224 (inc)->inc_faddr.s_addr ^ \ 225 ((inc)->inc_faddr.s_addr >> 16) ^ \ 226 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 227 228 #define SYNCACHE_HASH6(inc, mask) \ 229 ((tcp_syncache.hash_secret ^ \ 230 (inc)->inc6_faddr.s6_addr32[0] ^ \ 231 (inc)->inc6_faddr.s6_addr32[3] ^ \ 232 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 233 234 #define ENDPTS_EQ(a, b) ( \ 235 (a)->ie_fport == (b)->ie_fport && \ 236 (a)->ie_lport == (b)->ie_lport && \ 237 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 238 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 239 ) 240 241 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 242 243 static __inline void 244 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu, 245 struct syncache *sc, int slot) 246 { 247 sc->sc_rxtslot = slot; 248 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[slot]; 249 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq); 250 if (!callout_active(&syncache_percpu->tt_timerq[slot])) { 251 callout_reset(&syncache_percpu->tt_timerq[slot], 252 TCPTV_RTOBASE * tcp_backoff[slot], 253 syncache_timer, 254 &syncache_percpu->mrec[slot]); 255 } 256 } 257 258 static void 259 syncache_free(struct syncache *sc) 260 { 261 struct rtentry *rt; 262 #ifdef INET6 263 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 264 #else 265 const boolean_t isipv6 = FALSE; 266 #endif 267 268 if (sc->sc_ipopts) 269 m_free(sc->sc_ipopts); 270 271 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt; 272 if (rt != NULL) { 273 /* 274 * If this is the only reference to a protocol-cloned 275 * route, remove it immediately. 276 */ 277 if ((rt->rt_flags & RTF_WASCLONED) && rt->rt_refcnt == 1) 278 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 279 rt_mask(rt), rt->rt_flags, NULL); 280 RTFREE(rt); 281 } 282 283 zfree(tcp_syncache.zone, sc); 284 } 285 286 void 287 syncache_init(void) 288 { 289 int i, cpu; 290 291 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 292 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 293 tcp_syncache.cache_limit = 294 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 295 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 296 tcp_syncache.hash_secret = karc4random(); 297 298 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 299 &tcp_syncache.hashsize); 300 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 301 &tcp_syncache.cache_limit); 302 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 303 &tcp_syncache.bucket_limit); 304 if (!powerof2(tcp_syncache.hashsize)) { 305 kprintf("WARNING: syncache hash size is not a power of 2.\n"); 306 tcp_syncache.hashsize = 512; /* safe default */ 307 } 308 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 309 310 lwkt_initport_replyonly_null(&syncache_null_rport); 311 312 for (cpu = 0; cpu < ncpus2; cpu++) { 313 struct tcp_syncache_percpu *syncache_percpu; 314 315 syncache_percpu = &tcp_syncache_percpu[cpu]; 316 /* Allocate the hash table. */ 317 MALLOC(syncache_percpu->hashbase, struct syncache_head *, 318 tcp_syncache.hashsize * sizeof(struct syncache_head), 319 M_SYNCACHE, M_WAITOK); 320 321 /* Initialize the hash buckets. */ 322 for (i = 0; i < tcp_syncache.hashsize; i++) { 323 struct syncache_head *bucket; 324 325 bucket = &syncache_percpu->hashbase[i]; 326 TAILQ_INIT(&bucket->sch_bucket); 327 bucket->sch_length = 0; 328 } 329 330 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 331 /* Initialize the timer queues. */ 332 TAILQ_INIT(&syncache_percpu->timerq[i]); 333 callout_init(&syncache_percpu->tt_timerq[i]); 334 335 syncache_percpu->mrec[i].slot = i; 336 syncache_percpu->mrec[i].port = tcp_cport(cpu); 337 syncache_percpu->mrec[i].msg.nm_mrec = 338 &syncache_percpu->mrec[i]; 339 netmsg_init(&syncache_percpu->mrec[i].msg.nm_netmsg, 340 &syncache_null_rport, 0, 341 syncache_timer_handler); 342 } 343 } 344 345 /* 346 * Allocate the syncache entries. Allow the zone to allocate one 347 * more entry than cache limit, so a new entry can bump out an 348 * older one. 349 */ 350 tcp_syncache.zone = zinit("syncache", sizeof(struct syncache), 351 tcp_syncache.cache_limit * ncpus2, ZONE_INTERRUPT, 0); 352 tcp_syncache.cache_limit -= 1; 353 } 354 355 static void 356 syncache_insert(struct syncache *sc, struct syncache_head *sch) 357 { 358 struct tcp_syncache_percpu *syncache_percpu; 359 struct syncache *sc2; 360 int i; 361 362 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 363 364 /* 365 * Make sure that we don't overflow the per-bucket 366 * limit or the total cache size limit. 367 */ 368 if (sch->sch_length >= tcp_syncache.bucket_limit) { 369 /* 370 * The bucket is full, toss the oldest element. 371 */ 372 sc2 = TAILQ_FIRST(&sch->sch_bucket); 373 sc2->sc_tp->ts_recent = ticks; 374 syncache_drop(sc2, sch); 375 tcpstat.tcps_sc_bucketoverflow++; 376 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) { 377 /* 378 * The cache is full. Toss the oldest entry in the 379 * entire cache. This is the front entry in the 380 * first non-empty timer queue with the largest 381 * timeout value. 382 */ 383 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 384 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]); 385 if (sc2 != NULL) 386 break; 387 } 388 sc2->sc_tp->ts_recent = ticks; 389 syncache_drop(sc2, NULL); 390 tcpstat.tcps_sc_cacheoverflow++; 391 } 392 393 /* Initialize the entry's timer. */ 394 syncache_timeout(syncache_percpu, sc, 0); 395 396 /* Put it into the bucket. */ 397 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 398 sch->sch_length++; 399 syncache_percpu->cache_count++; 400 tcpstat.tcps_sc_added++; 401 } 402 403 static void 404 syncache_drop(struct syncache *sc, struct syncache_head *sch) 405 { 406 struct tcp_syncache_percpu *syncache_percpu; 407 #ifdef INET6 408 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 409 #else 410 const boolean_t isipv6 = FALSE; 411 #endif 412 413 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 414 415 if (sch == NULL) { 416 if (isipv6) { 417 sch = &syncache_percpu->hashbase[ 418 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 419 } else { 420 sch = &syncache_percpu->hashbase[ 421 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 422 } 423 } 424 425 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 426 sch->sch_length--; 427 syncache_percpu->cache_count--; 428 429 /* 430 * Remove the entry from the syncache timer/timeout queue. Note 431 * that we do not try to stop any running timer since we do not know 432 * whether the timer's message is in-transit or not. Since timeouts 433 * are fairly long, taking an unneeded callout does not detrimentally 434 * effect performance. 435 */ 436 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq); 437 438 syncache_free(sc); 439 } 440 441 /* 442 * Place a timeout message on the TCP thread's message queue. 443 * This routine runs in soft interrupt context. 444 * 445 * An invariant is for this routine to be called, the callout must 446 * have been active. Note that the callout is not deactivated until 447 * after the message has been processed in syncache_timer_handler() below. 448 */ 449 static void 450 syncache_timer(void *p) 451 { 452 struct netmsg_sc_timer *msg = p; 453 454 lwkt_sendmsg(msg->nm_mrec->port, &msg->nm_netmsg.nm_lmsg); 455 } 456 457 /* 458 * Service a timer message queued by timer expiration. 459 * This routine runs in the TCP protocol thread. 460 * 461 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 462 * If we have retransmitted an entry the maximum number of times, expire it. 463 * 464 * When we finish processing timed-out entries, we restart the timer if there 465 * are any entries still on the queue and deactivate it otherwise. Only after 466 * a timer has been deactivated here can it be restarted by syncache_timeout(). 467 */ 468 static void 469 syncache_timer_handler(netmsg_t netmsg) 470 { 471 struct tcp_syncache_percpu *syncache_percpu; 472 struct syncache *sc, *nsc; 473 struct inpcb *inp; 474 int slot; 475 476 slot = ((struct netmsg_sc_timer *)netmsg)->nm_mrec->slot; 477 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 478 479 nsc = TAILQ_FIRST(&syncache_percpu->timerq[slot]); 480 while (nsc != NULL) { 481 if (ticks < nsc->sc_rxttime) 482 break; /* finished because timerq sorted by time */ 483 sc = nsc; 484 inp = sc->sc_tp->t_inpcb; 485 if (slot == SYNCACHE_MAXREXMTS || 486 slot >= tcp_syncache.rexmt_limit || 487 inp->inp_gencnt != sc->sc_inp_gencnt) { 488 nsc = TAILQ_NEXT(sc, sc_timerq); 489 syncache_drop(sc, NULL); 490 tcpstat.tcps_sc_stale++; 491 continue; 492 } 493 /* 494 * syncache_respond() may call back into the syncache to 495 * to modify another entry, so do not obtain the next 496 * entry on the timer chain until it has completed. 497 */ 498 syncache_respond(sc, NULL); 499 nsc = TAILQ_NEXT(sc, sc_timerq); 500 tcpstat.tcps_sc_retransmitted++; 501 TAILQ_REMOVE(&syncache_percpu->timerq[slot], sc, sc_timerq); 502 syncache_timeout(syncache_percpu, sc, slot + 1); 503 } 504 if (nsc != NULL) 505 callout_reset(&syncache_percpu->tt_timerq[slot], 506 nsc->sc_rxttime - ticks, syncache_timer, 507 &syncache_percpu->mrec[slot]); 508 else 509 callout_deactivate(&syncache_percpu->tt_timerq[slot]); 510 511 lwkt_replymsg(&netmsg->nm_lmsg, 0); 512 } 513 514 /* 515 * Find an entry in the syncache. 516 */ 517 struct syncache * 518 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 519 { 520 struct tcp_syncache_percpu *syncache_percpu; 521 struct syncache *sc; 522 struct syncache_head *sch; 523 524 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 525 #ifdef INET6 526 if (inc->inc_isipv6) { 527 sch = &syncache_percpu->hashbase[ 528 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 529 *schp = sch; 530 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) 531 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 532 return (sc); 533 } else 534 #endif 535 { 536 sch = &syncache_percpu->hashbase[ 537 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 538 *schp = sch; 539 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 540 #ifdef INET6 541 if (sc->sc_inc.inc_isipv6) 542 continue; 543 #endif 544 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 545 return (sc); 546 } 547 } 548 return (NULL); 549 } 550 551 /* 552 * This function is called when we get a RST for a 553 * non-existent connection, so that we can see if the 554 * connection is in the syn cache. If it is, zap it. 555 */ 556 void 557 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 558 { 559 struct syncache *sc; 560 struct syncache_head *sch; 561 562 sc = syncache_lookup(inc, &sch); 563 if (sc == NULL) 564 return; 565 /* 566 * If the RST bit is set, check the sequence number to see 567 * if this is a valid reset segment. 568 * RFC 793 page 37: 569 * In all states except SYN-SENT, all reset (RST) segments 570 * are validated by checking their SEQ-fields. A reset is 571 * valid if its sequence number is in the window. 572 * 573 * The sequence number in the reset segment is normally an 574 * echo of our outgoing acknowlegement numbers, but some hosts 575 * send a reset with the sequence number at the rightmost edge 576 * of our receive window, and we have to handle this case. 577 */ 578 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 579 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 580 syncache_drop(sc, sch); 581 tcpstat.tcps_sc_reset++; 582 } 583 } 584 585 void 586 syncache_badack(struct in_conninfo *inc) 587 { 588 struct syncache *sc; 589 struct syncache_head *sch; 590 591 sc = syncache_lookup(inc, &sch); 592 if (sc != NULL) { 593 syncache_drop(sc, sch); 594 tcpstat.tcps_sc_badack++; 595 } 596 } 597 598 void 599 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 600 { 601 struct syncache *sc; 602 struct syncache_head *sch; 603 604 /* we are called at splnet() here */ 605 sc = syncache_lookup(inc, &sch); 606 if (sc == NULL) 607 return; 608 609 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 610 if (ntohl(th->th_seq) != sc->sc_iss) 611 return; 612 613 /* 614 * If we've rertransmitted 3 times and this is our second error, 615 * we remove the entry. Otherwise, we allow it to continue on. 616 * This prevents us from incorrectly nuking an entry during a 617 * spurious network outage. 618 * 619 * See tcp_notify(). 620 */ 621 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 622 sc->sc_flags |= SCF_UNREACH; 623 return; 624 } 625 syncache_drop(sc, sch); 626 tcpstat.tcps_sc_unreach++; 627 } 628 629 /* 630 * Build a new TCP socket structure from a syncache entry. 631 */ 632 static struct socket * 633 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 634 { 635 struct inpcb *inp = NULL, *linp; 636 struct socket *so; 637 struct tcpcb *tp; 638 #ifdef INET6 639 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 640 #else 641 const boolean_t isipv6 = FALSE; 642 #endif 643 644 /* 645 * Ok, create the full blown connection, and set things up 646 * as they would have been set up if we had created the 647 * connection when the SYN arrived. If we can't create 648 * the connection, abort it. 649 */ 650 so = sonewconn(lso, SS_ISCONNECTED); 651 if (so == NULL) { 652 /* 653 * Drop the connection; we will send a RST if the peer 654 * retransmits the ACK, 655 */ 656 tcpstat.tcps_listendrop++; 657 goto abort; 658 } 659 660 inp = so->so_pcb; 661 662 /* 663 * Insert new socket into hash list. 664 */ 665 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 666 if (isipv6) { 667 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 668 } else { 669 #ifdef INET6 670 inp->inp_vflag &= ~INP_IPV6; 671 inp->inp_vflag |= INP_IPV4; 672 #endif 673 inp->inp_laddr = sc->sc_inc.inc_laddr; 674 } 675 inp->inp_lport = sc->sc_inc.inc_lport; 676 if (in_pcbinsporthash(inp) != 0) { 677 /* 678 * Undo the assignments above if we failed to 679 * put the PCB on the hash lists. 680 */ 681 if (isipv6) 682 inp->in6p_laddr = kin6addr_any; 683 else 684 inp->inp_laddr.s_addr = INADDR_ANY; 685 inp->inp_lport = 0; 686 goto abort; 687 } 688 linp = so->so_pcb; 689 #ifdef IPSEC 690 /* copy old policy into new socket's */ 691 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 692 kprintf("syncache_expand: could not copy policy\n"); 693 #endif 694 if (isipv6) { 695 struct in6_addr laddr6; 696 struct sockaddr_in6 sin6; 697 /* 698 * Inherit socket options from the listening socket. 699 * Note that in6p_inputopts are not (and should not be) 700 * copied, since it stores previously received options and is 701 * used to detect if each new option is different than the 702 * previous one and hence should be passed to a user. 703 * If we copied in6p_inputopts, a user would not be able to 704 * receive options just after calling the accept system call. 705 */ 706 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 707 if (linp->in6p_outputopts) 708 inp->in6p_outputopts = 709 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 710 inp->in6p_route = sc->sc_route6; 711 sc->sc_route6.ro_rt = NULL; 712 713 sin6.sin6_family = AF_INET6; 714 sin6.sin6_len = sizeof sin6; 715 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 716 sin6.sin6_port = sc->sc_inc.inc_fport; 717 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 718 laddr6 = inp->in6p_laddr; 719 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 720 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 721 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) { 722 inp->in6p_laddr = laddr6; 723 goto abort; 724 } 725 } else { 726 struct in_addr laddr; 727 struct sockaddr_in sin; 728 729 inp->inp_options = ip_srcroute(m); 730 if (inp->inp_options == NULL) { 731 inp->inp_options = sc->sc_ipopts; 732 sc->sc_ipopts = NULL; 733 } 734 inp->inp_route = sc->sc_route; 735 sc->sc_route.ro_rt = NULL; 736 737 sin.sin_family = AF_INET; 738 sin.sin_len = sizeof sin; 739 sin.sin_addr = sc->sc_inc.inc_faddr; 740 sin.sin_port = sc->sc_inc.inc_fport; 741 bzero(sin.sin_zero, sizeof sin.sin_zero); 742 laddr = inp->inp_laddr; 743 if (inp->inp_laddr.s_addr == INADDR_ANY) 744 inp->inp_laddr = sc->sc_inc.inc_laddr; 745 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) { 746 inp->inp_laddr = laddr; 747 goto abort; 748 } 749 } 750 751 tp = intotcpcb(inp); 752 tp->t_state = TCPS_SYN_RECEIVED; 753 tp->iss = sc->sc_iss; 754 tp->irs = sc->sc_irs; 755 tcp_rcvseqinit(tp); 756 tcp_sendseqinit(tp); 757 tp->snd_wl1 = sc->sc_irs; 758 tp->rcv_up = sc->sc_irs + 1; 759 tp->rcv_wnd = sc->sc_wnd; 760 tp->rcv_adv += tp->rcv_wnd; 761 762 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 763 if (sc->sc_flags & SCF_NOOPT) 764 tp->t_flags |= TF_NOOPT; 765 if (sc->sc_flags & SCF_WINSCALE) { 766 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 767 tp->requested_s_scale = sc->sc_requested_s_scale; 768 tp->request_r_scale = sc->sc_request_r_scale; 769 } 770 if (sc->sc_flags & SCF_TIMESTAMP) { 771 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 772 tp->ts_recent = sc->sc_tsrecent; 773 tp->ts_recent_age = ticks; 774 } 775 if (sc->sc_flags & SCF_CC) { 776 /* 777 * Initialization of the tcpcb for transaction; 778 * set SND.WND = SEG.WND, 779 * initialize CCsend and CCrecv. 780 */ 781 tp->t_flags |= TF_REQ_CC | TF_RCVD_CC; 782 tp->cc_send = sc->sc_cc_send; 783 tp->cc_recv = sc->sc_cc_recv; 784 } 785 if (sc->sc_flags & SCF_SACK_PERMITTED) 786 tp->t_flags |= TF_SACK_PERMITTED; 787 788 tcp_mss(tp, sc->sc_peer_mss); 789 790 /* 791 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 792 */ 793 if (sc->sc_rxtslot != 0) 794 tp->snd_cwnd = tp->t_maxseg; 795 tcp_create_timermsg(tp); 796 tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep); 797 798 tcpstat.tcps_accepts++; 799 return (so); 800 801 abort: 802 if (so != NULL) 803 soabort_oncpu(so); 804 return (NULL); 805 } 806 807 /* 808 * This function gets called when we receive an ACK for a 809 * socket in the LISTEN state. We look up the connection 810 * in the syncache, and if its there, we pull it out of 811 * the cache and turn it into a full-blown connection in 812 * the SYN-RECEIVED state. 813 */ 814 int 815 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop, 816 struct mbuf *m) 817 { 818 struct syncache *sc; 819 struct syncache_head *sch; 820 struct socket *so; 821 822 sc = syncache_lookup(inc, &sch); 823 if (sc == NULL) { 824 /* 825 * There is no syncache entry, so see if this ACK is 826 * a returning syncookie. To do this, first: 827 * A. See if this socket has had a syncache entry dropped in 828 * the past. We don't want to accept a bogus syncookie 829 * if we've never received a SYN. 830 * B. check that the syncookie is valid. If it is, then 831 * cobble up a fake syncache entry, and return. 832 */ 833 if (!tcp_syncookies) 834 return (0); 835 sc = syncookie_lookup(inc, th, *sop); 836 if (sc == NULL) 837 return (0); 838 sch = NULL; 839 tcpstat.tcps_sc_recvcookie++; 840 } 841 842 /* 843 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 844 */ 845 if (th->th_ack != sc->sc_iss + 1) 846 return (0); 847 848 so = syncache_socket(sc, *sop, m); 849 if (so == NULL) { 850 #if 0 851 resetandabort: 852 /* XXXjlemon check this - is this correct? */ 853 tcp_respond(NULL, m, m, th, 854 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 855 #endif 856 m_freem(m); /* XXX only needed for above */ 857 tcpstat.tcps_sc_aborted++; 858 } else { 859 tcpstat.tcps_sc_completed++; 860 } 861 if (sch == NULL) 862 syncache_free(sc); 863 else 864 syncache_drop(sc, sch); 865 *sop = so; 866 return (1); 867 } 868 869 /* 870 * Given a LISTEN socket and an inbound SYN request, add 871 * this to the syn cache, and send back a segment: 872 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 873 * to the source. 874 * 875 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 876 * Doing so would require that we hold onto the data and deliver it 877 * to the application. However, if we are the target of a SYN-flood 878 * DoS attack, an attacker could send data which would eventually 879 * consume all available buffer space if it were ACKed. By not ACKing 880 * the data, we avoid this DoS scenario. 881 */ 882 int 883 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 884 struct socket **sop, struct mbuf *m) 885 { 886 struct tcp_syncache_percpu *syncache_percpu; 887 struct tcpcb *tp; 888 struct socket *so; 889 struct syncache *sc = NULL; 890 struct syncache_head *sch; 891 struct mbuf *ipopts = NULL; 892 struct rmxp_tao *taop; 893 int win; 894 895 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 896 so = *sop; 897 tp = sototcpcb(so); 898 899 /* 900 * Remember the IP options, if any. 901 */ 902 #ifdef INET6 903 if (!inc->inc_isipv6) 904 #endif 905 ipopts = ip_srcroute(m); 906 907 /* 908 * See if we already have an entry for this connection. 909 * If we do, resend the SYN,ACK, and reset the retransmit timer. 910 * 911 * XXX 912 * The syncache should be re-initialized with the contents 913 * of the new SYN which may have different options. 914 */ 915 sc = syncache_lookup(inc, &sch); 916 if (sc != NULL) { 917 tcpstat.tcps_sc_dupsyn++; 918 if (ipopts) { 919 /* 920 * If we were remembering a previous source route, 921 * forget it and use the new one we've been given. 922 */ 923 if (sc->sc_ipopts) 924 m_free(sc->sc_ipopts); 925 sc->sc_ipopts = ipopts; 926 } 927 /* 928 * Update timestamp if present. 929 */ 930 if (sc->sc_flags & SCF_TIMESTAMP) 931 sc->sc_tsrecent = to->to_tsval; 932 933 /* Just update the TOF_SACK_PERMITTED for now. */ 934 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 935 sc->sc_flags |= SCF_SACK_PERMITTED; 936 else 937 sc->sc_flags &= ~SCF_SACK_PERMITTED; 938 939 /* 940 * PCB may have changed, pick up new values. 941 */ 942 sc->sc_tp = tp; 943 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 944 if (syncache_respond(sc, m) == 0) { 945 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 946 sc, sc_timerq); 947 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 948 tcpstat.tcps_sndacks++; 949 tcpstat.tcps_sndtotal++; 950 } 951 *sop = NULL; 952 return (1); 953 } 954 955 /* 956 * This allocation is guaranteed to succeed because we 957 * preallocate one more syncache entry than cache_limit. 958 */ 959 sc = zalloc(tcp_syncache.zone); 960 961 /* 962 * Fill in the syncache values. 963 */ 964 sc->sc_tp = tp; 965 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 966 sc->sc_ipopts = ipopts; 967 sc->sc_inc.inc_fport = inc->inc_fport; 968 sc->sc_inc.inc_lport = inc->inc_lport; 969 #ifdef INET6 970 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 971 if (inc->inc_isipv6) { 972 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 973 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 974 sc->sc_route6.ro_rt = NULL; 975 } else 976 #endif 977 { 978 sc->sc_inc.inc_faddr = inc->inc_faddr; 979 sc->sc_inc.inc_laddr = inc->inc_laddr; 980 sc->sc_route.ro_rt = NULL; 981 } 982 sc->sc_irs = th->th_seq; 983 sc->sc_flags = 0; 984 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 985 if (tcp_syncookies) 986 sc->sc_iss = syncookie_generate(sc); 987 else 988 sc->sc_iss = karc4random(); 989 990 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */ 991 win = ssb_space(&so->so_rcv); 992 win = imax(win, 0); 993 win = imin(win, TCP_MAXWIN); 994 sc->sc_wnd = win; 995 996 if (tcp_do_rfc1323) { 997 /* 998 * A timestamp received in a SYN makes 999 * it ok to send timestamp requests and replies. 1000 */ 1001 if (to->to_flags & TOF_TS) { 1002 sc->sc_tsrecent = to->to_tsval; 1003 sc->sc_flags |= SCF_TIMESTAMP; 1004 } 1005 if (to->to_flags & TOF_SCALE) { 1006 int wscale = 0; 1007 1008 /* Compute proper scaling value from buffer space */ 1009 while (wscale < TCP_MAX_WINSHIFT && 1010 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) 1011 wscale++; 1012 sc->sc_request_r_scale = wscale; 1013 sc->sc_requested_s_scale = to->to_requested_s_scale; 1014 sc->sc_flags |= SCF_WINSCALE; 1015 } 1016 } 1017 if (tcp_do_rfc1644) { 1018 /* 1019 * A CC or CC.new option received in a SYN makes 1020 * it ok to send CC in subsequent segments. 1021 */ 1022 if (to->to_flags & (TOF_CC | TOF_CCNEW)) { 1023 sc->sc_cc_recv = to->to_cc; 1024 sc->sc_cc_send = CC_INC(tcp_ccgen); 1025 sc->sc_flags |= SCF_CC; 1026 } 1027 } 1028 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1029 sc->sc_flags |= SCF_SACK_PERMITTED; 1030 if (tp->t_flags & TF_NOOPT) 1031 sc->sc_flags = SCF_NOOPT; 1032 1033 /* 1034 * XXX 1035 * We have the option here of not doing TAO (even if the segment 1036 * qualifies) and instead fall back to a normal 3WHS via the syncache. 1037 * This allows us to apply synflood protection to TAO-qualifying SYNs 1038 * also. However, there should be a hueristic to determine when to 1039 * do this, and is not present at the moment. 1040 */ 1041 1042 /* 1043 * Perform TAO test on incoming CC (SEG.CC) option, if any. 1044 * - compare SEG.CC against cached CC from the same host, if any. 1045 * - if SEG.CC > chached value, SYN must be new and is accepted 1046 * immediately: save new CC in the cache, mark the socket 1047 * connected, enter ESTABLISHED state, turn on flag to 1048 * send a SYN in the next segment. 1049 * A virtual advertised window is set in rcv_adv to 1050 * initialize SWS prevention. Then enter normal segment 1051 * processing: drop SYN, process data and FIN. 1052 * - otherwise do a normal 3-way handshake. 1053 */ 1054 taop = tcp_gettaocache(&sc->sc_inc); 1055 if (to->to_flags & TOF_CC) { 1056 if ((tp->t_flags & TF_NOPUSH) && 1057 sc->sc_flags & SCF_CC && 1058 taop != NULL && taop->tao_cc != 0 && 1059 CC_GT(to->to_cc, taop->tao_cc)) { 1060 sc->sc_rxtslot = 0; 1061 so = syncache_socket(sc, *sop, m); 1062 if (so != NULL) { 1063 taop->tao_cc = to->to_cc; 1064 *sop = so; 1065 } 1066 syncache_free(sc); 1067 return (so != NULL); 1068 } 1069 } else { 1070 /* 1071 * No CC option, but maybe CC.NEW: invalidate cached value. 1072 */ 1073 if (taop != NULL) 1074 taop->tao_cc = 0; 1075 } 1076 /* 1077 * TAO test failed or there was no CC option, 1078 * do a standard 3-way handshake. 1079 */ 1080 if (syncache_respond(sc, m) == 0) { 1081 syncache_insert(sc, sch); 1082 tcpstat.tcps_sndacks++; 1083 tcpstat.tcps_sndtotal++; 1084 } else { 1085 syncache_free(sc); 1086 tcpstat.tcps_sc_dropped++; 1087 } 1088 *sop = NULL; 1089 return (1); 1090 } 1091 1092 static int 1093 syncache_respond(struct syncache *sc, struct mbuf *m) 1094 { 1095 u_int8_t *optp; 1096 int optlen, error; 1097 u_int16_t tlen, hlen, mssopt; 1098 struct ip *ip = NULL; 1099 struct rtentry *rt; 1100 struct tcphdr *th; 1101 struct ip6_hdr *ip6 = NULL; 1102 #ifdef INET6 1103 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1104 #else 1105 const boolean_t isipv6 = FALSE; 1106 #endif 1107 1108 if (isipv6) { 1109 rt = tcp_rtlookup6(&sc->sc_inc); 1110 if (rt != NULL) 1111 mssopt = rt->rt_ifp->if_mtu - 1112 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1113 else 1114 mssopt = tcp_v6mssdflt; 1115 hlen = sizeof(struct ip6_hdr); 1116 } else { 1117 rt = tcp_rtlookup(&sc->sc_inc); 1118 if (rt != NULL) 1119 mssopt = rt->rt_ifp->if_mtu - 1120 (sizeof(struct ip) + sizeof(struct tcphdr)); 1121 else 1122 mssopt = tcp_mssdflt; 1123 hlen = sizeof(struct ip); 1124 } 1125 1126 /* Compute the size of the TCP options. */ 1127 if (sc->sc_flags & SCF_NOOPT) { 1128 optlen = 0; 1129 } else { 1130 optlen = TCPOLEN_MAXSEG + 1131 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1132 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1133 ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0) + 1134 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1135 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1136 } 1137 tlen = hlen + sizeof(struct tcphdr) + optlen; 1138 1139 /* 1140 * XXX 1141 * assume that the entire packet will fit in a header mbuf 1142 */ 1143 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1144 1145 /* 1146 * XXX shouldn't this reuse the mbuf if possible ? 1147 * Create the IP+TCP header from scratch. 1148 */ 1149 if (m) 1150 m_freem(m); 1151 1152 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 1153 if (m == NULL) 1154 return (ENOBUFS); 1155 m->m_data += max_linkhdr; 1156 m->m_len = tlen; 1157 m->m_pkthdr.len = tlen; 1158 m->m_pkthdr.rcvif = NULL; 1159 1160 if (isipv6) { 1161 ip6 = mtod(m, struct ip6_hdr *); 1162 ip6->ip6_vfc = IPV6_VERSION; 1163 ip6->ip6_nxt = IPPROTO_TCP; 1164 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1165 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1166 ip6->ip6_plen = htons(tlen - hlen); 1167 /* ip6_hlim is set after checksum */ 1168 /* ip6_flow = ??? */ 1169 1170 th = (struct tcphdr *)(ip6 + 1); 1171 } else { 1172 ip = mtod(m, struct ip *); 1173 ip->ip_v = IPVERSION; 1174 ip->ip_hl = sizeof(struct ip) >> 2; 1175 ip->ip_len = tlen; 1176 ip->ip_id = 0; 1177 ip->ip_off = 0; 1178 ip->ip_sum = 0; 1179 ip->ip_p = IPPROTO_TCP; 1180 ip->ip_src = sc->sc_inc.inc_laddr; 1181 ip->ip_dst = sc->sc_inc.inc_faddr; 1182 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1183 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1184 1185 /* 1186 * See if we should do MTU discovery. Route lookups are 1187 * expensive, so we will only unset the DF bit if: 1188 * 1189 * 1) path_mtu_discovery is disabled 1190 * 2) the SCF_UNREACH flag has been set 1191 */ 1192 if (path_mtu_discovery 1193 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1194 ip->ip_off |= IP_DF; 1195 } 1196 1197 th = (struct tcphdr *)(ip + 1); 1198 } 1199 th->th_sport = sc->sc_inc.inc_lport; 1200 th->th_dport = sc->sc_inc.inc_fport; 1201 1202 th->th_seq = htonl(sc->sc_iss); 1203 th->th_ack = htonl(sc->sc_irs + 1); 1204 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1205 th->th_x2 = 0; 1206 th->th_flags = TH_SYN | TH_ACK; 1207 th->th_win = htons(sc->sc_wnd); 1208 th->th_urp = 0; 1209 1210 /* Tack on the TCP options. */ 1211 if (optlen == 0) 1212 goto no_options; 1213 optp = (u_int8_t *)(th + 1); 1214 *optp++ = TCPOPT_MAXSEG; 1215 *optp++ = TCPOLEN_MAXSEG; 1216 *optp++ = (mssopt >> 8) & 0xff; 1217 *optp++ = mssopt & 0xff; 1218 1219 if (sc->sc_flags & SCF_WINSCALE) { 1220 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1221 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1222 sc->sc_request_r_scale); 1223 optp += 4; 1224 } 1225 1226 if (sc->sc_flags & SCF_TIMESTAMP) { 1227 u_int32_t *lp = (u_int32_t *)(optp); 1228 1229 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1230 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1231 *lp++ = htonl(ticks); 1232 *lp = htonl(sc->sc_tsrecent); 1233 optp += TCPOLEN_TSTAMP_APPA; 1234 } 1235 1236 /* 1237 * Send CC and CC.echo if we received CC from our peer. 1238 */ 1239 if (sc->sc_flags & SCF_CC) { 1240 u_int32_t *lp = (u_int32_t *)(optp); 1241 1242 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC)); 1243 *lp++ = htonl(sc->sc_cc_send); 1244 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO)); 1245 *lp = htonl(sc->sc_cc_recv); 1246 optp += TCPOLEN_CC_APPA * 2; 1247 } 1248 1249 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1250 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1251 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1252 } 1253 1254 no_options: 1255 if (isipv6) { 1256 struct route_in6 *ro6 = &sc->sc_route6; 1257 1258 th->th_sum = 0; 1259 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1260 ip6->ip6_hlim = in6_selecthlim(NULL, 1261 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1262 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1263 sc->sc_tp->t_inpcb); 1264 } else { 1265 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1266 htons(tlen - hlen + IPPROTO_TCP)); 1267 m->m_pkthdr.csum_flags = CSUM_TCP; 1268 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1269 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 1270 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb); 1271 } 1272 return (error); 1273 } 1274 1275 /* 1276 * cookie layers: 1277 * 1278 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1279 * | peer iss | 1280 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1281 * | 0 |(A)| | 1282 * (A): peer mss index 1283 */ 1284 1285 /* 1286 * The values below are chosen to minimize the size of the tcp_secret 1287 * table, as well as providing roughly a 16 second lifetime for the cookie. 1288 */ 1289 1290 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1291 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1292 1293 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1294 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1295 #define SYNCOOKIE_TIMEOUT \ 1296 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1297 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1298 1299 static struct { 1300 u_int32_t ts_secbits[4]; 1301 u_int ts_expire; 1302 } tcp_secret[SYNCOOKIE_NSECRETS]; 1303 1304 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1305 1306 static MD5_CTX syn_ctx; 1307 1308 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1309 1310 struct md5_add { 1311 u_int32_t laddr, faddr; 1312 u_int32_t secbits[4]; 1313 u_int16_t lport, fport; 1314 }; 1315 1316 #ifdef CTASSERT 1317 CTASSERT(sizeof(struct md5_add) == 28); 1318 #endif 1319 1320 /* 1321 * Consider the problem of a recreated (and retransmitted) cookie. If the 1322 * original SYN was accepted, the connection is established. The second 1323 * SYN is inflight, and if it arrives with an ISN that falls within the 1324 * receive window, the connection is killed. 1325 * 1326 * However, since cookies have other problems, this may not be worth 1327 * worrying about. 1328 */ 1329 1330 static u_int32_t 1331 syncookie_generate(struct syncache *sc) 1332 { 1333 u_int32_t md5_buffer[4]; 1334 u_int32_t data; 1335 int idx, i; 1336 struct md5_add add; 1337 #ifdef INET6 1338 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1339 #else 1340 const boolean_t isipv6 = FALSE; 1341 #endif 1342 1343 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1344 if (tcp_secret[idx].ts_expire < ticks) { 1345 for (i = 0; i < 4; i++) 1346 tcp_secret[idx].ts_secbits[i] = karc4random(); 1347 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1348 } 1349 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1350 if (tcp_msstab[data] <= sc->sc_peer_mss) 1351 break; 1352 data = (data << SYNCOOKIE_WNDBITS) | idx; 1353 data ^= sc->sc_irs; /* peer's iss */ 1354 MD5Init(&syn_ctx); 1355 if (isipv6) { 1356 MD5Add(sc->sc_inc.inc6_laddr); 1357 MD5Add(sc->sc_inc.inc6_faddr); 1358 add.laddr = 0; 1359 add.faddr = 0; 1360 } else { 1361 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1362 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1363 } 1364 add.lport = sc->sc_inc.inc_lport; 1365 add.fport = sc->sc_inc.inc_fport; 1366 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1367 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1368 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1369 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1370 MD5Add(add); 1371 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1372 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1373 return (data); 1374 } 1375 1376 static struct syncache * 1377 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so) 1378 { 1379 u_int32_t md5_buffer[4]; 1380 struct syncache *sc; 1381 u_int32_t data; 1382 int wnd, idx; 1383 struct md5_add add; 1384 1385 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1386 idx = data & SYNCOOKIE_WNDMASK; 1387 if (tcp_secret[idx].ts_expire < ticks || 1388 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1389 return (NULL); 1390 MD5Init(&syn_ctx); 1391 #ifdef INET6 1392 if (inc->inc_isipv6) { 1393 MD5Add(inc->inc6_laddr); 1394 MD5Add(inc->inc6_faddr); 1395 add.laddr = 0; 1396 add.faddr = 0; 1397 } else 1398 #endif 1399 { 1400 add.laddr = inc->inc_laddr.s_addr; 1401 add.faddr = inc->inc_faddr.s_addr; 1402 } 1403 add.lport = inc->inc_lport; 1404 add.fport = inc->inc_fport; 1405 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1406 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1407 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1408 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1409 MD5Add(add); 1410 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1411 data ^= md5_buffer[0]; 1412 if (data & ~SYNCOOKIE_DATAMASK) 1413 return (NULL); 1414 data = data >> SYNCOOKIE_WNDBITS; 1415 1416 /* 1417 * This allocation is guaranteed to succeed because we 1418 * preallocate one more syncache entry than cache_limit. 1419 */ 1420 sc = zalloc(tcp_syncache.zone); 1421 1422 /* 1423 * Fill in the syncache values. 1424 * XXX duplicate code from syncache_add 1425 */ 1426 sc->sc_ipopts = NULL; 1427 sc->sc_inc.inc_fport = inc->inc_fport; 1428 sc->sc_inc.inc_lport = inc->inc_lport; 1429 #ifdef INET6 1430 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1431 if (inc->inc_isipv6) { 1432 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1433 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1434 sc->sc_route6.ro_rt = NULL; 1435 } else 1436 #endif 1437 { 1438 sc->sc_inc.inc_faddr = inc->inc_faddr; 1439 sc->sc_inc.inc_laddr = inc->inc_laddr; 1440 sc->sc_route.ro_rt = NULL; 1441 } 1442 sc->sc_irs = th->th_seq - 1; 1443 sc->sc_iss = th->th_ack - 1; 1444 wnd = ssb_space(&so->so_rcv); 1445 wnd = imax(wnd, 0); 1446 wnd = imin(wnd, TCP_MAXWIN); 1447 sc->sc_wnd = wnd; 1448 sc->sc_flags = 0; 1449 sc->sc_rxtslot = 0; 1450 sc->sc_peer_mss = tcp_msstab[data]; 1451 return (sc); 1452 } 1453