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 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved. 36 * 37 * License terms: all terms for the DragonFly license above plus the following: 38 * 39 * 4. All advertising materials mentioning features or use of this software 40 * must display the following acknowledgement: 41 * 42 * This product includes software developed by Jeffrey M. Hsu 43 * for the DragonFly Project. 44 * 45 * This requirement may be waived with permission from Jeffrey Hsu. 46 * This requirement will sunset and may be removed on July 8 2005, 47 * after which the standard DragonFly license (as shown above) will 48 * apply. 49 */ 50 51 /* 52 * All advertising materials mentioning features or use of this software 53 * must display the following acknowledgement: 54 * This product includes software developed by Jeffrey M. Hsu. 55 * 56 * Copyright (c) 2001 Networks Associates Technologies, Inc. 57 * All rights reserved. 58 * 59 * This software was developed for the FreeBSD Project by Jonathan Lemon 60 * and NAI Labs, the Security Research Division of Network Associates, Inc. 61 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 62 * DARPA CHATS research program. 63 * 64 * Redistribution and use in source and binary forms, with or without 65 * modification, are permitted provided that the following conditions 66 * are met: 67 * 1. Redistributions of source code must retain the above copyright 68 * notice, this list of conditions and the following disclaimer. 69 * 2. Redistributions in binary form must reproduce the above copyright 70 * notice, this list of conditions and the following disclaimer in the 71 * documentation and/or other materials provided with the distribution. 72 * 3. The name of the author may not be used to endorse or promote 73 * products derived from this software without specific prior written 74 * permission. 75 * 76 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 77 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 78 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 79 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 80 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 81 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 82 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 83 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 84 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 85 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 86 * SUCH DAMAGE. 87 * 88 * $FreeBSD: src/sys/netinet/tcp_syncache.c,v 1.5.2.14 2003/02/24 04:02:27 silby Exp $ 89 * $DragonFly: src/sys/netinet/tcp_syncache.c,v 1.19 2004/11/14 00:49:08 hsu Exp $ 90 */ 91 92 #include "opt_inet6.h" 93 #include "opt_ipsec.h" 94 95 #include <sys/param.h> 96 #include <sys/systm.h> 97 #include <sys/kernel.h> 98 #include <sys/sysctl.h> 99 #include <sys/malloc.h> 100 #include <sys/mbuf.h> 101 #include <sys/md5.h> 102 #include <sys/proc.h> /* for proc0 declaration */ 103 #include <sys/random.h> 104 #include <sys/socket.h> 105 #include <sys/socketvar.h> 106 #include <sys/in_cksum.h> 107 108 #include <sys/msgport2.h> 109 110 #include <net/if.h> 111 #include <net/route.h> 112 113 #include <netinet/in.h> 114 #include <netinet/in_systm.h> 115 #include <netinet/ip.h> 116 #include <netinet/in_var.h> 117 #include <netinet/in_pcb.h> 118 #include <netinet/ip_var.h> 119 #include <netinet/ip6.h> 120 #ifdef INET6 121 #include <netinet/icmp6.h> 122 #include <netinet6/nd6.h> 123 #endif 124 #include <netinet6/ip6_var.h> 125 #include <netinet6/in6_pcb.h> 126 #include <netinet/tcp.h> 127 #include <netinet/tcp_fsm.h> 128 #include <netinet/tcp_seq.h> 129 #include <netinet/tcp_timer.h> 130 #include <netinet/tcp_var.h> 131 #include <netinet6/tcp6_var.h> 132 133 #ifdef IPSEC 134 #include <netinet6/ipsec.h> 135 #ifdef INET6 136 #include <netinet6/ipsec6.h> 137 #endif 138 #include <netproto/key/key.h> 139 #endif /*IPSEC*/ 140 141 #ifdef FAST_IPSEC 142 #include <netproto/ipsec/ipsec.h> 143 #ifdef INET6 144 #include <netproto/ipsec/ipsec6.h> 145 #endif 146 #include <netproto/ipsec/key.h> 147 #define IPSEC 148 #endif /*FAST_IPSEC*/ 149 150 #include <vm/vm_zone.h> 151 152 static int tcp_syncookies = 1; 153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 154 &tcp_syncookies, 0, 155 "Use TCP SYN cookies if the syncache overflows"); 156 157 static void syncache_drop(struct syncache *, struct syncache_head *); 158 static void syncache_free(struct syncache *); 159 static void syncache_insert(struct syncache *, struct syncache_head *); 160 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 161 static int syncache_respond(struct syncache *, struct mbuf *); 162 static struct socket *syncache_socket(struct syncache *, struct socket *); 163 static void syncache_timer(void *); 164 static u_int32_t syncookie_generate(struct syncache *); 165 static struct syncache *syncookie_lookup(struct in_conninfo *, 166 struct tcphdr *, struct socket *); 167 168 /* 169 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 170 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 171 * the odds are that the user has given up attempting to connect by then. 172 */ 173 #define SYNCACHE_MAXREXMTS 3 174 175 /* Arbitrary values */ 176 #define TCP_SYNCACHE_HASHSIZE 512 177 #define TCP_SYNCACHE_BUCKETLIMIT 30 178 179 struct netmsg_sc_timer { 180 struct lwkt_msg nm_lmsg; 181 struct msgrec *nm_mrec; /* back pointer to containing msgrec */ 182 }; 183 184 struct msgrec { 185 struct netmsg_sc_timer msg; 186 lwkt_port_t port; /* constant after init */ 187 int slot; /* constant after init */ 188 }; 189 190 static int syncache_timer_handler(lwkt_msg_t); 191 192 struct tcp_syncache { 193 struct vm_zone *zone; 194 u_int hashsize; 195 u_int hashmask; 196 u_int bucket_limit; 197 u_int cache_limit; 198 u_int rexmt_limit; 199 u_int hash_secret; 200 }; 201 static struct tcp_syncache tcp_syncache; 202 203 struct tcp_syncache_percpu { 204 struct syncache_head *hashbase; 205 u_int cache_count; 206 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1]; 207 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1]; 208 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1]; 209 }; 210 static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU]; 211 212 static struct lwkt_port syncache_null_rport; 213 214 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 215 216 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD, 217 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 218 219 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD, 220 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 221 222 /* XXX JH */ 223 #if 0 224 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 225 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 226 #endif 227 228 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD, 229 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 230 231 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 232 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 233 234 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 235 236 #define SYNCACHE_HASH(inc, mask) \ 237 ((tcp_syncache.hash_secret ^ \ 238 (inc)->inc_faddr.s_addr ^ \ 239 ((inc)->inc_faddr.s_addr >> 16) ^ \ 240 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 241 242 #define SYNCACHE_HASH6(inc, mask) \ 243 ((tcp_syncache.hash_secret ^ \ 244 (inc)->inc6_faddr.s6_addr32[0] ^ \ 245 (inc)->inc6_faddr.s6_addr32[3] ^ \ 246 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 247 248 #define ENDPTS_EQ(a, b) ( \ 249 (a)->ie_fport == (b)->ie_fport && \ 250 (a)->ie_lport == (b)->ie_lport && \ 251 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 252 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 253 ) 254 255 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 256 257 static __inline void 258 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu, 259 struct syncache *sc, int slot) 260 { 261 sc->sc_rxtslot = slot; 262 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[slot]; 263 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq); 264 if (!callout_active(&syncache_percpu->tt_timerq[slot])) { 265 callout_reset(&syncache_percpu->tt_timerq[slot], 266 TCPTV_RTOBASE * tcp_backoff[slot], 267 syncache_timer, 268 &syncache_percpu->mrec[slot]); 269 } 270 } 271 272 static void 273 syncache_free(struct syncache *sc) 274 { 275 struct rtentry *rt; 276 #ifdef INET6 277 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 278 #else 279 const boolean_t isipv6 = FALSE; 280 #endif 281 282 if (sc->sc_ipopts) 283 (void) m_free(sc->sc_ipopts); 284 if (isipv6) 285 rt = sc->sc_route6.ro_rt; 286 else 287 rt = sc->sc_route.ro_rt; 288 if (rt != NULL) { 289 /* 290 * If this is the only reference to a protocol cloned 291 * route, remove it immediately. 292 */ 293 if (rt->rt_flags & RTF_WASCLONED && 294 (sc->sc_flags & SCF_KEEPROUTE) == 0 && 295 rt->rt_refcnt == 1) { 296 rtrequest(RTM_DELETE, rt_key(rt), 297 rt->rt_gateway, rt_mask(rt), 298 rt->rt_flags, NULL); 299 } 300 RTFREE(rt); 301 } 302 zfree(tcp_syncache.zone, sc); 303 } 304 305 void 306 syncache_init(void) 307 { 308 int i, cpu; 309 310 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 311 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 312 tcp_syncache.cache_limit = 313 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 314 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 315 tcp_syncache.hash_secret = arc4random(); 316 317 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 318 &tcp_syncache.hashsize); 319 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 320 &tcp_syncache.cache_limit); 321 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 322 &tcp_syncache.bucket_limit); 323 if (!powerof2(tcp_syncache.hashsize)) { 324 printf("WARNING: syncache hash size is not a power of 2.\n"); 325 tcp_syncache.hashsize = 512; /* safe default */ 326 } 327 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 328 329 lwkt_initport_null_rport(&syncache_null_rport, NULL); 330 331 for (cpu = 0; cpu < ncpus2; cpu++) { 332 struct tcp_syncache_percpu *syncache_percpu; 333 334 syncache_percpu = &tcp_syncache_percpu[cpu]; 335 /* Allocate the hash table. */ 336 MALLOC(syncache_percpu->hashbase, struct syncache_head *, 337 tcp_syncache.hashsize * sizeof(struct syncache_head), 338 M_SYNCACHE, M_WAITOK); 339 340 /* Initialize the hash buckets. */ 341 for (i = 0; i < tcp_syncache.hashsize; i++) { 342 struct syncache_head *bucket; 343 344 bucket = &syncache_percpu->hashbase[i]; 345 TAILQ_INIT(&bucket->sch_bucket); 346 bucket->sch_length = 0; 347 } 348 349 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 350 /* Initialize the timer queues. */ 351 TAILQ_INIT(&syncache_percpu->timerq[i]); 352 callout_init(&syncache_percpu->tt_timerq[i]); 353 354 syncache_percpu->mrec[i].slot = i; 355 syncache_percpu->mrec[i].port = tcp_cport(cpu); 356 syncache_percpu->mrec[i].msg.nm_mrec = 357 &syncache_percpu->mrec[i]; 358 lwkt_initmsg(&syncache_percpu->mrec[i].msg.nm_lmsg, 359 &syncache_null_rport, 0, 360 lwkt_cmd_func(syncache_timer_handler), 361 lwkt_cmd_op_none); 362 } 363 } 364 365 /* 366 * Allocate the syncache entries. Allow the zone to allocate one 367 * more entry than cache limit, so a new entry can bump out an 368 * older one. 369 */ 370 tcp_syncache.zone = zinit("syncache", sizeof(struct syncache), 371 tcp_syncache.cache_limit, ZONE_INTERRUPT, 0); 372 tcp_syncache.cache_limit -= 1; 373 } 374 375 static void 376 syncache_insert(sc, sch) 377 struct syncache *sc; 378 struct syncache_head *sch; 379 { 380 struct tcp_syncache_percpu *syncache_percpu; 381 struct syncache *sc2; 382 int i; 383 384 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 385 386 /* 387 * Make sure that we don't overflow the per-bucket 388 * limit or the total cache size limit. 389 */ 390 if (sch->sch_length >= tcp_syncache.bucket_limit) { 391 /* 392 * The bucket is full, toss the oldest element. 393 */ 394 sc2 = TAILQ_FIRST(&sch->sch_bucket); 395 sc2->sc_tp->ts_recent = ticks; 396 syncache_drop(sc2, sch); 397 tcpstat.tcps_sc_bucketoverflow++; 398 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) { 399 /* 400 * The cache is full. Toss the oldest entry in the 401 * entire cache. This is the front entry in the 402 * first non-empty timer queue with the largest 403 * timeout value. 404 */ 405 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 406 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]); 407 if (sc2 != NULL) 408 break; 409 } 410 sc2->sc_tp->ts_recent = ticks; 411 syncache_drop(sc2, NULL); 412 tcpstat.tcps_sc_cacheoverflow++; 413 } 414 415 /* Initialize the entry's timer. */ 416 syncache_timeout(syncache_percpu, sc, 0); 417 418 /* Put it into the bucket. */ 419 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 420 sch->sch_length++; 421 syncache_percpu->cache_count++; 422 tcpstat.tcps_sc_added++; 423 } 424 425 static void 426 syncache_drop(sc, sch) 427 struct syncache *sc; 428 struct syncache_head *sch; 429 { 430 struct tcp_syncache_percpu *syncache_percpu; 431 #ifdef INET6 432 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 433 #else 434 const boolean_t isipv6 = FALSE; 435 #endif 436 437 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 438 439 if (sch == NULL) { 440 if (isipv6) { 441 sch = &syncache_percpu->hashbase[ 442 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 443 } else { 444 sch = &syncache_percpu->hashbase[ 445 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 446 } 447 } 448 449 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 450 sch->sch_length--; 451 syncache_percpu->cache_count--; 452 453 /* 454 * Remove the entry from the syncache timer/timeout queue. Note 455 * that we do not try to stop any running timer since we do not know 456 * whether the timer's message is in-transit or not. Since timeouts 457 * are fairly long, taking an unneeded callout does not detrimentally 458 * effect performance. 459 */ 460 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq); 461 462 syncache_free(sc); 463 } 464 465 /* 466 * Place a timeout message on the TCP thread's message queue. 467 * This routine runs in soft interrupt context. 468 * 469 * An invariant is for this routine to be called, the callout must 470 * have been active. Note that the callout is not deactivated until 471 * after the message has been processed in syncache_timer_handler() below. 472 */ 473 static void 474 syncache_timer(void *p) 475 { 476 struct netmsg_sc_timer *msg = p; 477 478 lwkt_sendmsg(msg->nm_mrec->port, &msg->nm_lmsg); 479 } 480 481 /* 482 * Service a timer message queued by timer expiration. 483 * This routine runs in the TCP protocol thread. 484 * 485 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 486 * If we have retransmitted an entry the maximum number of times, expire it. 487 * 488 * When we finish processing timed-out entries, we restart the timer if there 489 * are any entries still on the queue and deactivate it otherwise. Only after 490 * a timer has been deactivated here can it be restarted by syncache_timeout(). 491 */ 492 static int 493 syncache_timer_handler(lwkt_msg_t msg) 494 { 495 struct tcp_syncache_percpu *syncache_percpu; 496 struct syncache *sc, *nsc; 497 struct inpcb *inp; 498 int slot; 499 500 slot = ((struct netmsg_sc_timer *)msg)->nm_mrec->slot; 501 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 502 503 nsc = TAILQ_FIRST(&syncache_percpu->timerq[slot]); 504 while (nsc != NULL) { 505 if (ticks < nsc->sc_rxttime) 506 break; /* finished because timerq sorted by time */ 507 sc = nsc; 508 inp = sc->sc_tp->t_inpcb; 509 if (slot == SYNCACHE_MAXREXMTS || 510 slot >= tcp_syncache.rexmt_limit || 511 inp->inp_gencnt != sc->sc_inp_gencnt) { 512 nsc = TAILQ_NEXT(sc, sc_timerq); 513 syncache_drop(sc, NULL); 514 tcpstat.tcps_sc_stale++; 515 continue; 516 } 517 /* 518 * syncache_respond() may call back into the syncache to 519 * to modify another entry, so do not obtain the next 520 * entry on the timer chain until it has completed. 521 */ 522 (void) syncache_respond(sc, NULL); 523 nsc = TAILQ_NEXT(sc, sc_timerq); 524 tcpstat.tcps_sc_retransmitted++; 525 TAILQ_REMOVE(&syncache_percpu->timerq[slot], sc, sc_timerq); 526 syncache_timeout(syncache_percpu, sc, slot + 1); 527 } 528 if (nsc != NULL) 529 callout_reset(&syncache_percpu->tt_timerq[slot], 530 nsc->sc_rxttime - ticks, syncache_timer, 531 &syncache_percpu->mrec[slot]); 532 else 533 callout_deactivate(&syncache_percpu->tt_timerq[slot]); 534 535 lwkt_replymsg(msg, 0); 536 return (EASYNC); 537 } 538 539 /* 540 * Find an entry in the syncache. 541 */ 542 struct syncache * 543 syncache_lookup(inc, schp) 544 struct in_conninfo *inc; 545 struct syncache_head **schp; 546 { 547 struct tcp_syncache_percpu *syncache_percpu; 548 struct syncache *sc; 549 struct syncache_head *sch; 550 551 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 552 #ifdef INET6 553 if (inc->inc_isipv6) { 554 sch = &syncache_percpu->hashbase[ 555 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 556 *schp = sch; 557 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) 558 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 559 return (sc); 560 } else 561 #endif 562 { 563 sch = &syncache_percpu->hashbase[ 564 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 565 *schp = sch; 566 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 567 #ifdef INET6 568 if (sc->sc_inc.inc_isipv6) 569 continue; 570 #endif 571 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 572 return (sc); 573 } 574 } 575 return (NULL); 576 } 577 578 /* 579 * This function is called when we get a RST for a 580 * non-existent connection, so that we can see if the 581 * connection is in the syn cache. If it is, zap it. 582 */ 583 void 584 syncache_chkrst(inc, th) 585 struct in_conninfo *inc; 586 struct tcphdr *th; 587 { 588 struct syncache *sc; 589 struct syncache_head *sch; 590 591 sc = syncache_lookup(inc, &sch); 592 if (sc == NULL) 593 return; 594 /* 595 * If the RST bit is set, check the sequence number to see 596 * if this is a valid reset segment. 597 * RFC 793 page 37: 598 * In all states except SYN-SENT, all reset (RST) segments 599 * are validated by checking their SEQ-fields. A reset is 600 * valid if its sequence number is in the window. 601 * 602 * The sequence number in the reset segment is normally an 603 * echo of our outgoing acknowlegement numbers, but some hosts 604 * send a reset with the sequence number at the rightmost edge 605 * of our receive window, and we have to handle this case. 606 */ 607 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 608 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 609 syncache_drop(sc, sch); 610 tcpstat.tcps_sc_reset++; 611 } 612 } 613 614 void 615 syncache_badack(inc) 616 struct in_conninfo *inc; 617 { 618 struct syncache *sc; 619 struct syncache_head *sch; 620 621 sc = syncache_lookup(inc, &sch); 622 if (sc != NULL) { 623 syncache_drop(sc, sch); 624 tcpstat.tcps_sc_badack++; 625 } 626 } 627 628 void 629 syncache_unreach(inc, th) 630 struct in_conninfo *inc; 631 struct tcphdr *th; 632 { 633 struct syncache *sc; 634 struct syncache_head *sch; 635 636 /* we are called at splnet() here */ 637 sc = syncache_lookup(inc, &sch); 638 if (sc == NULL) 639 return; 640 641 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 642 if (ntohl(th->th_seq) != sc->sc_iss) 643 return; 644 645 /* 646 * If we've rertransmitted 3 times and this is our second error, 647 * we remove the entry. Otherwise, we allow it to continue on. 648 * This prevents us from incorrectly nuking an entry during a 649 * spurious network outage. 650 * 651 * See tcp_notify(). 652 */ 653 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 654 sc->sc_flags |= SCF_UNREACH; 655 return; 656 } 657 syncache_drop(sc, sch); 658 tcpstat.tcps_sc_unreach++; 659 } 660 661 /* 662 * Build a new TCP socket structure from a syncache entry. 663 */ 664 static struct socket * 665 syncache_socket(sc, lso) 666 struct syncache *sc; 667 struct socket *lso; 668 { 669 struct inpcb *inp = NULL; 670 struct socket *so; 671 struct tcpcb *tp; 672 #ifdef INET6 673 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 674 #else 675 const boolean_t isipv6 = FALSE; 676 #endif 677 678 /* 679 * Ok, create the full blown connection, and set things up 680 * as they would have been set up if we had created the 681 * connection when the SYN arrived. If we can't create 682 * the connection, abort it. 683 */ 684 so = sonewconn(lso, SS_ISCONNECTED); 685 if (so == NULL) { 686 /* 687 * Drop the connection; we will send a RST if the peer 688 * retransmits the ACK, 689 */ 690 tcpstat.tcps_listendrop++; 691 goto abort; 692 } 693 694 inp = sotoinpcb(so); 695 696 /* 697 * Insert new socket into hash list. 698 */ 699 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 700 if (isipv6) { 701 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 702 } else { 703 #ifdef INET6 704 inp->inp_vflag &= ~INP_IPV6; 705 inp->inp_vflag |= INP_IPV4; 706 #endif 707 inp->inp_laddr = sc->sc_inc.inc_laddr; 708 } 709 inp->inp_lport = sc->sc_inc.inc_lport; 710 if (in_pcbinsporthash(inp) != 0) { 711 /* 712 * Undo the assignments above if we failed to 713 * put the PCB on the hash lists. 714 */ 715 if (isipv6) 716 inp->in6p_laddr = in6addr_any; 717 else 718 inp->inp_laddr.s_addr = INADDR_ANY; 719 inp->inp_lport = 0; 720 goto abort; 721 } 722 #ifdef IPSEC 723 /* copy old policy into new socket's */ 724 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 725 printf("syncache_expand: could not copy policy\n"); 726 #endif 727 if (isipv6) { 728 struct inpcb *oinp = sotoinpcb(lso); 729 struct in6_addr laddr6; 730 struct sockaddr_in6 sin6; 731 /* 732 * Inherit socket options from the listening socket. 733 * Note that in6p_inputopts are not (and should not be) 734 * copied, since it stores previously received options and is 735 * used to detect if each new option is different than the 736 * previous one and hence should be passed to a user. 737 * If we copied in6p_inputopts, a user would not be able to 738 * receive options just after calling the accept system call. 739 */ 740 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 741 if (oinp->in6p_outputopts) 742 inp->in6p_outputopts = 743 ip6_copypktopts(oinp->in6p_outputopts, M_INTWAIT); 744 inp->in6p_route = sc->sc_route6; 745 sc->sc_route6.ro_rt = NULL; 746 747 sin6.sin6_family = AF_INET6; 748 sin6.sin6_len = sizeof sin6; 749 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 750 sin6.sin6_port = sc->sc_inc.inc_fport; 751 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 752 laddr6 = inp->in6p_laddr; 753 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 754 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 755 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) { 756 inp->in6p_laddr = laddr6; 757 goto abort; 758 } 759 } else { 760 struct in_addr laddr; 761 struct sockaddr_in sin; 762 763 inp->inp_options = ip_srcroute(); 764 if (inp->inp_options == NULL) { 765 inp->inp_options = sc->sc_ipopts; 766 sc->sc_ipopts = NULL; 767 } 768 inp->inp_route = sc->sc_route; 769 sc->sc_route.ro_rt = NULL; 770 771 sin.sin_family = AF_INET; 772 sin.sin_len = sizeof sin; 773 sin.sin_addr = sc->sc_inc.inc_faddr; 774 sin.sin_port = sc->sc_inc.inc_fport; 775 bzero(sin.sin_zero, sizeof sin.sin_zero); 776 laddr = inp->inp_laddr; 777 if (inp->inp_laddr.s_addr == INADDR_ANY) 778 inp->inp_laddr = sc->sc_inc.inc_laddr; 779 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) { 780 inp->inp_laddr = laddr; 781 goto abort; 782 } 783 } 784 785 tp = intotcpcb(inp); 786 tp->t_state = TCPS_SYN_RECEIVED; 787 tp->iss = sc->sc_iss; 788 tp->irs = sc->sc_irs; 789 tcp_rcvseqinit(tp); 790 tcp_sendseqinit(tp); 791 tp->snd_wl1 = sc->sc_irs; 792 tp->rcv_up = sc->sc_irs + 1; 793 tp->rcv_wnd = sc->sc_wnd; 794 tp->rcv_adv += tp->rcv_wnd; 795 796 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 797 if (sc->sc_flags & SCF_NOOPT) 798 tp->t_flags |= TF_NOOPT; 799 if (sc->sc_flags & SCF_WINSCALE) { 800 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 801 tp->requested_s_scale = sc->sc_requested_s_scale; 802 tp->request_r_scale = sc->sc_request_r_scale; 803 } 804 if (sc->sc_flags & SCF_TIMESTAMP) { 805 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 806 tp->ts_recent = sc->sc_tsrecent; 807 tp->ts_recent_age = ticks; 808 } 809 if (sc->sc_flags & SCF_CC) { 810 /* 811 * Initialization of the tcpcb for transaction; 812 * set SND.WND = SEG.WND, 813 * initialize CCsend and CCrecv. 814 */ 815 tp->t_flags |= TF_REQ_CC | TF_RCVD_CC; 816 tp->cc_send = sc->sc_cc_send; 817 tp->cc_recv = sc->sc_cc_recv; 818 } 819 if (sc->sc_flags & SCF_SACK_PERMITTED) 820 tp->t_flags |= TF_SACK_PERMITTED; 821 822 tcp_mss(tp, sc->sc_peer_mss); 823 824 /* 825 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 826 */ 827 if (sc->sc_rxtslot != 0) 828 tp->snd_cwnd = tp->t_maxseg; 829 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 830 831 tcpstat.tcps_accepts++; 832 return (so); 833 834 abort: 835 if (so != NULL) 836 (void) soabort(so); 837 return (NULL); 838 } 839 840 /* 841 * This function gets called when we receive an ACK for a 842 * socket in the LISTEN state. We look up the connection 843 * in the syncache, and if its there, we pull it out of 844 * the cache and turn it into a full-blown connection in 845 * the SYN-RECEIVED state. 846 */ 847 int 848 syncache_expand(inc, th, sop, m) 849 struct in_conninfo *inc; 850 struct tcphdr *th; 851 struct socket **sop; 852 struct mbuf *m; 853 { 854 struct syncache *sc; 855 struct syncache_head *sch; 856 struct socket *so; 857 858 sc = syncache_lookup(inc, &sch); 859 if (sc == NULL) { 860 /* 861 * There is no syncache entry, so see if this ACK is 862 * a returning syncookie. To do this, first: 863 * A. See if this socket has had a syncache entry dropped in 864 * the past. We don't want to accept a bogus syncookie 865 * if we've never received a SYN. 866 * B. check that the syncookie is valid. If it is, then 867 * cobble up a fake syncache entry, and return. 868 */ 869 if (!tcp_syncookies) 870 return (0); 871 sc = syncookie_lookup(inc, th, *sop); 872 if (sc == NULL) 873 return (0); 874 sch = NULL; 875 tcpstat.tcps_sc_recvcookie++; 876 } 877 878 /* 879 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 880 */ 881 if (th->th_ack != sc->sc_iss + 1) 882 return (0); 883 884 so = syncache_socket(sc, *sop); 885 if (so == NULL) { 886 #if 0 887 resetandabort: 888 /* XXXjlemon check this - is this correct? */ 889 (void) tcp_respond(NULL, m, m, th, 890 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 891 #endif 892 m_freem(m); /* XXX only needed for above */ 893 tcpstat.tcps_sc_aborted++; 894 } else { 895 sc->sc_flags |= SCF_KEEPROUTE; 896 tcpstat.tcps_sc_completed++; 897 } 898 if (sch == NULL) 899 syncache_free(sc); 900 else 901 syncache_drop(sc, sch); 902 *sop = so; 903 return (1); 904 } 905 906 /* 907 * Given a LISTEN socket and an inbound SYN request, add 908 * this to the syn cache, and send back a segment: 909 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 910 * to the source. 911 * 912 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 913 * Doing so would require that we hold onto the data and deliver it 914 * to the application. However, if we are the target of a SYN-flood 915 * DoS attack, an attacker could send data which would eventually 916 * consume all available buffer space if it were ACKed. By not ACKing 917 * the data, we avoid this DoS scenario. 918 */ 919 int 920 syncache_add(inc, to, th, sop, m) 921 struct in_conninfo *inc; 922 struct tcpopt *to; 923 struct tcphdr *th; 924 struct socket **sop; 925 struct mbuf *m; 926 { 927 struct tcp_syncache_percpu *syncache_percpu; 928 struct tcpcb *tp; 929 struct socket *so; 930 struct syncache *sc = NULL; 931 struct syncache_head *sch; 932 struct mbuf *ipopts = NULL; 933 struct rmxp_tao *taop; 934 int win; 935 936 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 937 so = *sop; 938 tp = sototcpcb(so); 939 940 /* 941 * Remember the IP options, if any. 942 */ 943 #ifdef INET6 944 if (!inc->inc_isipv6) 945 #endif 946 ipopts = ip_srcroute(); 947 948 /* 949 * See if we already have an entry for this connection. 950 * If we do, resend the SYN,ACK, and reset the retransmit timer. 951 * 952 * XXX 953 * The syncache should be re-initialized with the contents 954 * of the new SYN which may have different options. 955 */ 956 sc = syncache_lookup(inc, &sch); 957 if (sc != NULL) { 958 tcpstat.tcps_sc_dupsyn++; 959 if (ipopts) { 960 /* 961 * If we were remembering a previous source route, 962 * forget it and use the new one we've been given. 963 */ 964 if (sc->sc_ipopts) 965 (void) m_free(sc->sc_ipopts); 966 sc->sc_ipopts = ipopts; 967 } 968 /* 969 * Update timestamp if present. 970 */ 971 if (sc->sc_flags & SCF_TIMESTAMP) 972 sc->sc_tsrecent = to->to_tsval; 973 974 /* Just update the TOF_SACK_PERMITTED for now. */ 975 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 976 sc->sc_flags |= SCF_SACK_PERMITTED; 977 else 978 sc->sc_flags &= ~SCF_SACK_PERMITTED; 979 980 /* 981 * PCB may have changed, pick up new values. 982 */ 983 sc->sc_tp = tp; 984 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 985 if (syncache_respond(sc, m) == 0) { 986 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 987 sc, sc_timerq); 988 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 989 tcpstat.tcps_sndacks++; 990 tcpstat.tcps_sndtotal++; 991 } 992 *sop = NULL; 993 return (1); 994 } 995 996 /* 997 * This allocation is guaranteed to succeed because we 998 * preallocate one more syncache entry than cache_limit. 999 */ 1000 sc = zalloc(tcp_syncache.zone); 1001 1002 /* 1003 * Fill in the syncache values. 1004 */ 1005 sc->sc_tp = tp; 1006 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1007 sc->sc_ipopts = ipopts; 1008 sc->sc_inc.inc_fport = inc->inc_fport; 1009 sc->sc_inc.inc_lport = inc->inc_lport; 1010 #ifdef INET6 1011 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1012 if (inc->inc_isipv6) { 1013 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1014 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1015 sc->sc_route6.ro_rt = NULL; 1016 } else 1017 #endif 1018 { 1019 sc->sc_inc.inc_faddr = inc->inc_faddr; 1020 sc->sc_inc.inc_laddr = inc->inc_laddr; 1021 sc->sc_route.ro_rt = NULL; 1022 } 1023 sc->sc_irs = th->th_seq; 1024 sc->sc_flags = 0; 1025 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 1026 if (tcp_syncookies) 1027 sc->sc_iss = syncookie_generate(sc); 1028 else 1029 sc->sc_iss = arc4random(); 1030 1031 /* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */ 1032 win = sbspace(&so->so_rcv); 1033 win = imax(win, 0); 1034 win = imin(win, TCP_MAXWIN); 1035 sc->sc_wnd = win; 1036 1037 if (tcp_do_rfc1323) { 1038 /* 1039 * A timestamp received in a SYN makes 1040 * it ok to send timestamp requests and replies. 1041 */ 1042 if (to->to_flags & TOF_TS) { 1043 sc->sc_tsrecent = to->to_tsval; 1044 sc->sc_flags |= SCF_TIMESTAMP; 1045 } 1046 if (to->to_flags & TOF_SCALE) { 1047 int wscale = 0; 1048 1049 /* Compute proper scaling value from buffer space */ 1050 while (wscale < TCP_MAX_WINSHIFT && 1051 (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat) 1052 wscale++; 1053 sc->sc_request_r_scale = wscale; 1054 sc->sc_requested_s_scale = to->to_requested_s_scale; 1055 sc->sc_flags |= SCF_WINSCALE; 1056 } 1057 } 1058 if (tcp_do_rfc1644) { 1059 /* 1060 * A CC or CC.new option received in a SYN makes 1061 * it ok to send CC in subsequent segments. 1062 */ 1063 if (to->to_flags & (TOF_CC | TOF_CCNEW)) { 1064 sc->sc_cc_recv = to->to_cc; 1065 sc->sc_cc_send = CC_INC(tcp_ccgen); 1066 sc->sc_flags |= SCF_CC; 1067 } 1068 } 1069 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1070 sc->sc_flags |= SCF_SACK_PERMITTED; 1071 if (tp->t_flags & TF_NOOPT) 1072 sc->sc_flags = SCF_NOOPT; 1073 1074 /* 1075 * XXX 1076 * We have the option here of not doing TAO (even if the segment 1077 * qualifies) and instead fall back to a normal 3WHS via the syncache. 1078 * This allows us to apply synflood protection to TAO-qualifying SYNs 1079 * also. However, there should be a hueristic to determine when to 1080 * do this, and is not present at the moment. 1081 */ 1082 1083 /* 1084 * Perform TAO test on incoming CC (SEG.CC) option, if any. 1085 * - compare SEG.CC against cached CC from the same host, if any. 1086 * - if SEG.CC > chached value, SYN must be new and is accepted 1087 * immediately: save new CC in the cache, mark the socket 1088 * connected, enter ESTABLISHED state, turn on flag to 1089 * send a SYN in the next segment. 1090 * A virtual advertised window is set in rcv_adv to 1091 * initialize SWS prevention. Then enter normal segment 1092 * processing: drop SYN, process data and FIN. 1093 * - otherwise do a normal 3-way handshake. 1094 */ 1095 taop = tcp_gettaocache(&sc->sc_inc); 1096 if ((to->to_flags & TOF_CC) != 0) { 1097 if (((tp->t_flags & TF_NOPUSH) != 0) && 1098 sc->sc_flags & SCF_CC && 1099 taop != NULL && taop->tao_cc != 0 && 1100 CC_GT(to->to_cc, taop->tao_cc)) { 1101 sc->sc_rxtslot = 0; 1102 so = syncache_socket(sc, *sop); 1103 if (so != NULL) { 1104 sc->sc_flags |= SCF_KEEPROUTE; 1105 taop->tao_cc = to->to_cc; 1106 *sop = so; 1107 } 1108 syncache_free(sc); 1109 return (so != NULL); 1110 } 1111 } else { 1112 /* 1113 * No CC option, but maybe CC.NEW: invalidate cached value. 1114 */ 1115 if (taop != NULL) 1116 taop->tao_cc = 0; 1117 } 1118 /* 1119 * TAO test failed or there was no CC option, 1120 * do a standard 3-way handshake. 1121 */ 1122 if (syncache_respond(sc, m) == 0) { 1123 syncache_insert(sc, sch); 1124 tcpstat.tcps_sndacks++; 1125 tcpstat.tcps_sndtotal++; 1126 } else { 1127 syncache_free(sc); 1128 tcpstat.tcps_sc_dropped++; 1129 } 1130 *sop = NULL; 1131 return (1); 1132 } 1133 1134 static int 1135 syncache_respond(sc, m) 1136 struct syncache *sc; 1137 struct mbuf *m; 1138 { 1139 u_int8_t *optp; 1140 int optlen, error; 1141 u_int16_t tlen, hlen, mssopt; 1142 struct ip *ip = NULL; 1143 struct rtentry *rt; 1144 struct tcphdr *th; 1145 struct ip6_hdr *ip6 = NULL; 1146 #ifdef INET6 1147 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1148 #else 1149 const boolean_t isipv6 = FALSE; 1150 #endif 1151 1152 if (isipv6) { 1153 rt = tcp_rtlookup6(&sc->sc_inc); 1154 if (rt != NULL) 1155 mssopt = rt->rt_ifp->if_mtu - 1156 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1157 else 1158 mssopt = tcp_v6mssdflt; 1159 hlen = sizeof(struct ip6_hdr); 1160 } else { 1161 rt = tcp_rtlookup(&sc->sc_inc); 1162 if (rt != NULL) 1163 mssopt = rt->rt_ifp->if_mtu - 1164 (sizeof(struct ip) + sizeof(struct tcphdr)); 1165 else 1166 mssopt = tcp_mssdflt; 1167 hlen = sizeof(struct ip); 1168 } 1169 1170 /* Compute the size of the TCP options. */ 1171 if (sc->sc_flags & SCF_NOOPT) { 1172 optlen = 0; 1173 } else { 1174 optlen = TCPOLEN_MAXSEG + 1175 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1176 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1177 ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0) + 1178 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1179 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1180 } 1181 tlen = hlen + sizeof(struct tcphdr) + optlen; 1182 1183 /* 1184 * XXX 1185 * assume that the entire packet will fit in a header mbuf 1186 */ 1187 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1188 1189 /* 1190 * XXX shouldn't this reuse the mbuf if possible ? 1191 * Create the IP+TCP header from scratch. 1192 */ 1193 if (m) 1194 m_freem(m); 1195 1196 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 1197 if (m == NULL) 1198 return (ENOBUFS); 1199 m->m_data += max_linkhdr; 1200 m->m_len = tlen; 1201 m->m_pkthdr.len = tlen; 1202 m->m_pkthdr.rcvif = NULL; 1203 1204 if (isipv6) { 1205 ip6 = mtod(m, struct ip6_hdr *); 1206 ip6->ip6_vfc = IPV6_VERSION; 1207 ip6->ip6_nxt = IPPROTO_TCP; 1208 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1209 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1210 ip6->ip6_plen = htons(tlen - hlen); 1211 /* ip6_hlim is set after checksum */ 1212 /* ip6_flow = ??? */ 1213 1214 th = (struct tcphdr *)(ip6 + 1); 1215 } else { 1216 ip = mtod(m, struct ip *); 1217 ip->ip_v = IPVERSION; 1218 ip->ip_hl = sizeof(struct ip) >> 2; 1219 ip->ip_len = tlen; 1220 ip->ip_id = 0; 1221 ip->ip_off = 0; 1222 ip->ip_sum = 0; 1223 ip->ip_p = IPPROTO_TCP; 1224 ip->ip_src = sc->sc_inc.inc_laddr; 1225 ip->ip_dst = sc->sc_inc.inc_faddr; 1226 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1227 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1228 1229 /* 1230 * See if we should do MTU discovery. Route lookups are 1231 * expensive, so we will only unset the DF bit if: 1232 * 1233 * 1) path_mtu_discovery is disabled 1234 * 2) the SCF_UNREACH flag has been set 1235 */ 1236 if (path_mtu_discovery 1237 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1238 ip->ip_off |= IP_DF; 1239 } 1240 1241 th = (struct tcphdr *)(ip + 1); 1242 } 1243 th->th_sport = sc->sc_inc.inc_lport; 1244 th->th_dport = sc->sc_inc.inc_fport; 1245 1246 th->th_seq = htonl(sc->sc_iss); 1247 th->th_ack = htonl(sc->sc_irs + 1); 1248 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1249 th->th_x2 = 0; 1250 th->th_flags = TH_SYN | TH_ACK; 1251 th->th_win = htons(sc->sc_wnd); 1252 th->th_urp = 0; 1253 1254 /* Tack on the TCP options. */ 1255 if (optlen == 0) 1256 goto no_options; 1257 optp = (u_int8_t *)(th + 1); 1258 *optp++ = TCPOPT_MAXSEG; 1259 *optp++ = TCPOLEN_MAXSEG; 1260 *optp++ = (mssopt >> 8) & 0xff; 1261 *optp++ = mssopt & 0xff; 1262 1263 if (sc->sc_flags & SCF_WINSCALE) { 1264 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1265 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1266 sc->sc_request_r_scale); 1267 optp += 4; 1268 } 1269 1270 if (sc->sc_flags & SCF_TIMESTAMP) { 1271 u_int32_t *lp = (u_int32_t *)(optp); 1272 1273 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1274 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1275 *lp++ = htonl(ticks); 1276 *lp = htonl(sc->sc_tsrecent); 1277 optp += TCPOLEN_TSTAMP_APPA; 1278 } 1279 1280 /* 1281 * Send CC and CC.echo if we received CC from our peer. 1282 */ 1283 if (sc->sc_flags & SCF_CC) { 1284 u_int32_t *lp = (u_int32_t *)(optp); 1285 1286 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC)); 1287 *lp++ = htonl(sc->sc_cc_send); 1288 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO)); 1289 *lp = htonl(sc->sc_cc_recv); 1290 optp += TCPOLEN_CC_APPA * 2; 1291 } 1292 1293 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1294 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1295 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1296 } 1297 1298 no_options: 1299 if (isipv6) { 1300 struct route_in6 *ro6 = &sc->sc_route6; 1301 1302 th->th_sum = 0; 1303 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1304 ip6->ip6_hlim = in6_selecthlim(NULL, 1305 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1306 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1307 sc->sc_tp->t_inpcb); 1308 } else { 1309 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1310 htons(tlen - hlen + IPPROTO_TCP)); 1311 m->m_pkthdr.csum_flags = CSUM_TCP; 1312 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1313 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 0, NULL, 1314 sc->sc_tp->t_inpcb); 1315 } 1316 return (error); 1317 } 1318 1319 /* 1320 * cookie layers: 1321 * 1322 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1323 * | peer iss | 1324 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1325 * | 0 |(A)| | 1326 * (A): peer mss index 1327 */ 1328 1329 /* 1330 * The values below are chosen to minimize the size of the tcp_secret 1331 * table, as well as providing roughly a 16 second lifetime for the cookie. 1332 */ 1333 1334 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1335 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1336 1337 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1338 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1339 #define SYNCOOKIE_TIMEOUT \ 1340 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1341 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1342 1343 static struct { 1344 u_int32_t ts_secbits[4]; 1345 u_int ts_expire; 1346 } tcp_secret[SYNCOOKIE_NSECRETS]; 1347 1348 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1349 1350 static MD5_CTX syn_ctx; 1351 1352 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1353 1354 struct md5_add { 1355 u_int32_t laddr, faddr; 1356 u_int32_t secbits[4]; 1357 u_int16_t lport, fport; 1358 }; 1359 1360 #ifdef CTASSERT 1361 CTASSERT(sizeof(struct md5_add) == 28); 1362 #endif 1363 1364 /* 1365 * Consider the problem of a recreated (and retransmitted) cookie. If the 1366 * original SYN was accepted, the connection is established. The second 1367 * SYN is inflight, and if it arrives with an ISN that falls within the 1368 * receive window, the connection is killed. 1369 * 1370 * However, since cookies have other problems, this may not be worth 1371 * worrying about. 1372 */ 1373 1374 static u_int32_t 1375 syncookie_generate(struct syncache *sc) 1376 { 1377 u_int32_t md5_buffer[4]; 1378 u_int32_t data; 1379 int idx, i; 1380 struct md5_add add; 1381 #ifdef INET6 1382 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1383 #else 1384 const boolean_t isipv6 = FALSE; 1385 #endif 1386 1387 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1388 if (tcp_secret[idx].ts_expire < ticks) { 1389 for (i = 0; i < 4; i++) 1390 tcp_secret[idx].ts_secbits[i] = arc4random(); 1391 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1392 } 1393 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1394 if (tcp_msstab[data] <= sc->sc_peer_mss) 1395 break; 1396 data = (data << SYNCOOKIE_WNDBITS) | idx; 1397 data ^= sc->sc_irs; /* peer's iss */ 1398 MD5Init(&syn_ctx); 1399 if (isipv6) { 1400 MD5Add(sc->sc_inc.inc6_laddr); 1401 MD5Add(sc->sc_inc.inc6_faddr); 1402 add.laddr = 0; 1403 add.faddr = 0; 1404 } else { 1405 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1406 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1407 } 1408 add.lport = sc->sc_inc.inc_lport; 1409 add.fport = sc->sc_inc.inc_fport; 1410 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1411 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1412 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1413 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1414 MD5Add(add); 1415 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1416 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1417 return (data); 1418 } 1419 1420 static struct syncache * 1421 syncookie_lookup(inc, th, so) 1422 struct in_conninfo *inc; 1423 struct tcphdr *th; 1424 struct socket *so; 1425 { 1426 u_int32_t md5_buffer[4]; 1427 struct syncache *sc; 1428 u_int32_t data; 1429 int wnd, idx; 1430 struct md5_add add; 1431 1432 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1433 idx = data & SYNCOOKIE_WNDMASK; 1434 if (tcp_secret[idx].ts_expire < ticks || 1435 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1436 return (NULL); 1437 MD5Init(&syn_ctx); 1438 #ifdef INET6 1439 if (inc->inc_isipv6) { 1440 MD5Add(inc->inc6_laddr); 1441 MD5Add(inc->inc6_faddr); 1442 add.laddr = 0; 1443 add.faddr = 0; 1444 } else 1445 #endif 1446 { 1447 add.laddr = inc->inc_laddr.s_addr; 1448 add.faddr = inc->inc_faddr.s_addr; 1449 } 1450 add.lport = inc->inc_lport; 1451 add.fport = inc->inc_fport; 1452 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1453 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1454 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1455 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1456 MD5Add(add); 1457 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1458 data ^= md5_buffer[0]; 1459 if ((data & ~SYNCOOKIE_DATAMASK) != 0) 1460 return (NULL); 1461 data = data >> SYNCOOKIE_WNDBITS; 1462 1463 /* 1464 * This allocation is guaranteed to succeed because we 1465 * preallocate one more syncache entry than cache_limit. 1466 */ 1467 sc = zalloc(tcp_syncache.zone); 1468 1469 /* 1470 * Fill in the syncache values. 1471 * XXX duplicate code from syncache_add 1472 */ 1473 sc->sc_ipopts = NULL; 1474 sc->sc_inc.inc_fport = inc->inc_fport; 1475 sc->sc_inc.inc_lport = inc->inc_lport; 1476 #ifdef INET6 1477 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1478 if (inc->inc_isipv6) { 1479 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1480 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1481 sc->sc_route6.ro_rt = NULL; 1482 } else 1483 #endif 1484 { 1485 sc->sc_inc.inc_faddr = inc->inc_faddr; 1486 sc->sc_inc.inc_laddr = inc->inc_laddr; 1487 sc->sc_route.ro_rt = NULL; 1488 } 1489 sc->sc_irs = th->th_seq - 1; 1490 sc->sc_iss = th->th_ack - 1; 1491 wnd = sbspace(&so->so_rcv); 1492 wnd = imax(wnd, 0); 1493 wnd = imin(wnd, TCP_MAXWIN); 1494 sc->sc_wnd = wnd; 1495 sc->sc_flags = 0; 1496 sc->sc_rxtslot = 0; 1497 sc->sc_peer_mss = tcp_msstab[data]; 1498 return (sc); 1499 } 1500