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 1.24 2006/01/14 11:33:50 swildner 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 m_free(sc->sc_ipopts); 284 285 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt; 286 if (rt != NULL) { 287 /* 288 * If this is the only reference to a protocol-cloned 289 * route, remove it immediately. 290 */ 291 if ((rt->rt_flags & RTF_WASCLONED) && rt->rt_refcnt == 1) 292 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 293 rt_mask(rt), rt->rt_flags, NULL); 294 RTFREE(rt); 295 } 296 297 zfree(tcp_syncache.zone, sc); 298 } 299 300 void 301 syncache_init(void) 302 { 303 int i, cpu; 304 305 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 306 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 307 tcp_syncache.cache_limit = 308 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 309 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 310 tcp_syncache.hash_secret = arc4random(); 311 312 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 313 &tcp_syncache.hashsize); 314 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 315 &tcp_syncache.cache_limit); 316 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 317 &tcp_syncache.bucket_limit); 318 if (!powerof2(tcp_syncache.hashsize)) { 319 printf("WARNING: syncache hash size is not a power of 2.\n"); 320 tcp_syncache.hashsize = 512; /* safe default */ 321 } 322 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 323 324 lwkt_initport_null_rport(&syncache_null_rport, NULL); 325 326 for (cpu = 0; cpu < ncpus2; cpu++) { 327 struct tcp_syncache_percpu *syncache_percpu; 328 329 syncache_percpu = &tcp_syncache_percpu[cpu]; 330 /* Allocate the hash table. */ 331 MALLOC(syncache_percpu->hashbase, struct syncache_head *, 332 tcp_syncache.hashsize * sizeof(struct syncache_head), 333 M_SYNCACHE, M_WAITOK); 334 335 /* Initialize the hash buckets. */ 336 for (i = 0; i < tcp_syncache.hashsize; i++) { 337 struct syncache_head *bucket; 338 339 bucket = &syncache_percpu->hashbase[i]; 340 TAILQ_INIT(&bucket->sch_bucket); 341 bucket->sch_length = 0; 342 } 343 344 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 345 /* Initialize the timer queues. */ 346 TAILQ_INIT(&syncache_percpu->timerq[i]); 347 callout_init(&syncache_percpu->tt_timerq[i]); 348 349 syncache_percpu->mrec[i].slot = i; 350 syncache_percpu->mrec[i].port = tcp_cport(cpu); 351 syncache_percpu->mrec[i].msg.nm_mrec = 352 &syncache_percpu->mrec[i]; 353 lwkt_initmsg(&syncache_percpu->mrec[i].msg.nm_lmsg, 354 &syncache_null_rport, 0, 355 lwkt_cmd_func(syncache_timer_handler), 356 lwkt_cmd_op_none); 357 } 358 } 359 360 /* 361 * Allocate the syncache entries. Allow the zone to allocate one 362 * more entry than cache limit, so a new entry can bump out an 363 * older one. 364 */ 365 tcp_syncache.zone = zinit("syncache", sizeof(struct syncache), 366 tcp_syncache.cache_limit * ncpus2, ZONE_INTERRUPT, 0); 367 tcp_syncache.cache_limit -= 1; 368 } 369 370 static void 371 syncache_insert(struct syncache *sc, struct syncache_head *sch) 372 { 373 struct tcp_syncache_percpu *syncache_percpu; 374 struct syncache *sc2; 375 int i; 376 377 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 378 379 /* 380 * Make sure that we don't overflow the per-bucket 381 * limit or the total cache size limit. 382 */ 383 if (sch->sch_length >= tcp_syncache.bucket_limit) { 384 /* 385 * The bucket is full, toss the oldest element. 386 */ 387 sc2 = TAILQ_FIRST(&sch->sch_bucket); 388 sc2->sc_tp->ts_recent = ticks; 389 syncache_drop(sc2, sch); 390 tcpstat.tcps_sc_bucketoverflow++; 391 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) { 392 /* 393 * The cache is full. Toss the oldest entry in the 394 * entire cache. This is the front entry in the 395 * first non-empty timer queue with the largest 396 * timeout value. 397 */ 398 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 399 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]); 400 if (sc2 != NULL) 401 break; 402 } 403 sc2->sc_tp->ts_recent = ticks; 404 syncache_drop(sc2, NULL); 405 tcpstat.tcps_sc_cacheoverflow++; 406 } 407 408 /* Initialize the entry's timer. */ 409 syncache_timeout(syncache_percpu, sc, 0); 410 411 /* Put it into the bucket. */ 412 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 413 sch->sch_length++; 414 syncache_percpu->cache_count++; 415 tcpstat.tcps_sc_added++; 416 } 417 418 static void 419 syncache_drop(struct syncache *sc, struct syncache_head *sch) 420 { 421 struct tcp_syncache_percpu *syncache_percpu; 422 #ifdef INET6 423 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 424 #else 425 const boolean_t isipv6 = FALSE; 426 #endif 427 428 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 429 430 if (sch == NULL) { 431 if (isipv6) { 432 sch = &syncache_percpu->hashbase[ 433 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 434 } else { 435 sch = &syncache_percpu->hashbase[ 436 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 437 } 438 } 439 440 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 441 sch->sch_length--; 442 syncache_percpu->cache_count--; 443 444 /* 445 * Remove the entry from the syncache timer/timeout queue. Note 446 * that we do not try to stop any running timer since we do not know 447 * whether the timer's message is in-transit or not. Since timeouts 448 * are fairly long, taking an unneeded callout does not detrimentally 449 * effect performance. 450 */ 451 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq); 452 453 syncache_free(sc); 454 } 455 456 /* 457 * Place a timeout message on the TCP thread's message queue. 458 * This routine runs in soft interrupt context. 459 * 460 * An invariant is for this routine to be called, the callout must 461 * have been active. Note that the callout is not deactivated until 462 * after the message has been processed in syncache_timer_handler() below. 463 */ 464 static void 465 syncache_timer(void *p) 466 { 467 struct netmsg_sc_timer *msg = p; 468 469 lwkt_sendmsg(msg->nm_mrec->port, &msg->nm_lmsg); 470 } 471 472 /* 473 * Service a timer message queued by timer expiration. 474 * This routine runs in the TCP protocol thread. 475 * 476 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 477 * If we have retransmitted an entry the maximum number of times, expire it. 478 * 479 * When we finish processing timed-out entries, we restart the timer if there 480 * are any entries still on the queue and deactivate it otherwise. Only after 481 * a timer has been deactivated here can it be restarted by syncache_timeout(). 482 */ 483 static int 484 syncache_timer_handler(lwkt_msg_t msg) 485 { 486 struct tcp_syncache_percpu *syncache_percpu; 487 struct syncache *sc, *nsc; 488 struct inpcb *inp; 489 int slot; 490 491 slot = ((struct netmsg_sc_timer *)msg)->nm_mrec->slot; 492 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 493 494 nsc = TAILQ_FIRST(&syncache_percpu->timerq[slot]); 495 while (nsc != NULL) { 496 if (ticks < nsc->sc_rxttime) 497 break; /* finished because timerq sorted by time */ 498 sc = nsc; 499 inp = sc->sc_tp->t_inpcb; 500 if (slot == SYNCACHE_MAXREXMTS || 501 slot >= tcp_syncache.rexmt_limit || 502 inp->inp_gencnt != sc->sc_inp_gencnt) { 503 nsc = TAILQ_NEXT(sc, sc_timerq); 504 syncache_drop(sc, NULL); 505 tcpstat.tcps_sc_stale++; 506 continue; 507 } 508 /* 509 * syncache_respond() may call back into the syncache to 510 * to modify another entry, so do not obtain the next 511 * entry on the timer chain until it has completed. 512 */ 513 syncache_respond(sc, NULL); 514 nsc = TAILQ_NEXT(sc, sc_timerq); 515 tcpstat.tcps_sc_retransmitted++; 516 TAILQ_REMOVE(&syncache_percpu->timerq[slot], sc, sc_timerq); 517 syncache_timeout(syncache_percpu, sc, slot + 1); 518 } 519 if (nsc != NULL) 520 callout_reset(&syncache_percpu->tt_timerq[slot], 521 nsc->sc_rxttime - ticks, syncache_timer, 522 &syncache_percpu->mrec[slot]); 523 else 524 callout_deactivate(&syncache_percpu->tt_timerq[slot]); 525 526 lwkt_replymsg(msg, 0); 527 return (EASYNC); 528 } 529 530 /* 531 * Find an entry in the syncache. 532 */ 533 struct syncache * 534 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 535 { 536 struct tcp_syncache_percpu *syncache_percpu; 537 struct syncache *sc; 538 struct syncache_head *sch; 539 540 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 541 #ifdef INET6 542 if (inc->inc_isipv6) { 543 sch = &syncache_percpu->hashbase[ 544 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 545 *schp = sch; 546 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) 547 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 548 return (sc); 549 } else 550 #endif 551 { 552 sch = &syncache_percpu->hashbase[ 553 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 554 *schp = sch; 555 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 556 #ifdef INET6 557 if (sc->sc_inc.inc_isipv6) 558 continue; 559 #endif 560 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 561 return (sc); 562 } 563 } 564 return (NULL); 565 } 566 567 /* 568 * This function is called when we get a RST for a 569 * non-existent connection, so that we can see if the 570 * connection is in the syn cache. If it is, zap it. 571 */ 572 void 573 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 574 { 575 struct syncache *sc; 576 struct syncache_head *sch; 577 578 sc = syncache_lookup(inc, &sch); 579 if (sc == NULL) 580 return; 581 /* 582 * If the RST bit is set, check the sequence number to see 583 * if this is a valid reset segment. 584 * RFC 793 page 37: 585 * In all states except SYN-SENT, all reset (RST) segments 586 * are validated by checking their SEQ-fields. A reset is 587 * valid if its sequence number is in the window. 588 * 589 * The sequence number in the reset segment is normally an 590 * echo of our outgoing acknowlegement numbers, but some hosts 591 * send a reset with the sequence number at the rightmost edge 592 * of our receive window, and we have to handle this case. 593 */ 594 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 595 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 596 syncache_drop(sc, sch); 597 tcpstat.tcps_sc_reset++; 598 } 599 } 600 601 void 602 syncache_badack(struct in_conninfo *inc) 603 { 604 struct syncache *sc; 605 struct syncache_head *sch; 606 607 sc = syncache_lookup(inc, &sch); 608 if (sc != NULL) { 609 syncache_drop(sc, sch); 610 tcpstat.tcps_sc_badack++; 611 } 612 } 613 614 void 615 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 616 { 617 struct syncache *sc; 618 struct syncache_head *sch; 619 620 /* we are called at splnet() here */ 621 sc = syncache_lookup(inc, &sch); 622 if (sc == NULL) 623 return; 624 625 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 626 if (ntohl(th->th_seq) != sc->sc_iss) 627 return; 628 629 /* 630 * If we've rertransmitted 3 times and this is our second error, 631 * we remove the entry. Otherwise, we allow it to continue on. 632 * This prevents us from incorrectly nuking an entry during a 633 * spurious network outage. 634 * 635 * See tcp_notify(). 636 */ 637 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 638 sc->sc_flags |= SCF_UNREACH; 639 return; 640 } 641 syncache_drop(sc, sch); 642 tcpstat.tcps_sc_unreach++; 643 } 644 645 /* 646 * Build a new TCP socket structure from a syncache entry. 647 */ 648 static struct socket * 649 syncache_socket(struct syncache *sc, struct socket *lso) 650 { 651 struct inpcb *inp = NULL, *linp; 652 struct socket *so; 653 struct tcpcb *tp; 654 #ifdef INET6 655 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 656 #else 657 const boolean_t isipv6 = FALSE; 658 #endif 659 660 /* 661 * Ok, create the full blown connection, and set things up 662 * as they would have been set up if we had created the 663 * connection when the SYN arrived. If we can't create 664 * the connection, abort it. 665 */ 666 so = sonewconn(lso, SS_ISCONNECTED); 667 if (so == NULL) { 668 /* 669 * Drop the connection; we will send a RST if the peer 670 * retransmits the ACK, 671 */ 672 tcpstat.tcps_listendrop++; 673 goto abort; 674 } 675 676 inp = so->so_pcb; 677 678 /* 679 * Insert new socket into hash list. 680 */ 681 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 682 if (isipv6) { 683 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 684 } else { 685 #ifdef INET6 686 inp->inp_vflag &= ~INP_IPV6; 687 inp->inp_vflag |= INP_IPV4; 688 #endif 689 inp->inp_laddr = sc->sc_inc.inc_laddr; 690 } 691 inp->inp_lport = sc->sc_inc.inc_lport; 692 if (in_pcbinsporthash(inp) != 0) { 693 /* 694 * Undo the assignments above if we failed to 695 * put the PCB on the hash lists. 696 */ 697 if (isipv6) 698 inp->in6p_laddr = in6addr_any; 699 else 700 inp->inp_laddr.s_addr = INADDR_ANY; 701 inp->inp_lport = 0; 702 goto abort; 703 } 704 linp = so->so_pcb; 705 #ifdef IPSEC 706 /* copy old policy into new socket's */ 707 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 708 printf("syncache_expand: could not copy policy\n"); 709 #endif 710 if (isipv6) { 711 struct in6_addr laddr6; 712 struct sockaddr_in6 sin6; 713 /* 714 * Inherit socket options from the listening socket. 715 * Note that in6p_inputopts are not (and should not be) 716 * copied, since it stores previously received options and is 717 * used to detect if each new option is different than the 718 * previous one and hence should be passed to a user. 719 * If we copied in6p_inputopts, a user would not be able to 720 * receive options just after calling the accept system call. 721 */ 722 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 723 if (linp->in6p_outputopts) 724 inp->in6p_outputopts = 725 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 726 inp->in6p_route = sc->sc_route6; 727 sc->sc_route6.ro_rt = NULL; 728 729 sin6.sin6_family = AF_INET6; 730 sin6.sin6_len = sizeof sin6; 731 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 732 sin6.sin6_port = sc->sc_inc.inc_fport; 733 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 734 laddr6 = inp->in6p_laddr; 735 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 736 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 737 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) { 738 inp->in6p_laddr = laddr6; 739 goto abort; 740 } 741 } else { 742 struct in_addr laddr; 743 struct sockaddr_in sin; 744 745 inp->inp_options = ip_srcroute(); 746 if (inp->inp_options == NULL) { 747 inp->inp_options = sc->sc_ipopts; 748 sc->sc_ipopts = NULL; 749 } 750 inp->inp_route = sc->sc_route; 751 sc->sc_route.ro_rt = NULL; 752 753 sin.sin_family = AF_INET; 754 sin.sin_len = sizeof sin; 755 sin.sin_addr = sc->sc_inc.inc_faddr; 756 sin.sin_port = sc->sc_inc.inc_fport; 757 bzero(sin.sin_zero, sizeof sin.sin_zero); 758 laddr = inp->inp_laddr; 759 if (inp->inp_laddr.s_addr == INADDR_ANY) 760 inp->inp_laddr = sc->sc_inc.inc_laddr; 761 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) { 762 inp->inp_laddr = laddr; 763 goto abort; 764 } 765 } 766 767 tp = intotcpcb(inp); 768 tp->t_state = TCPS_SYN_RECEIVED; 769 tp->iss = sc->sc_iss; 770 tp->irs = sc->sc_irs; 771 tcp_rcvseqinit(tp); 772 tcp_sendseqinit(tp); 773 tp->snd_wl1 = sc->sc_irs; 774 tp->rcv_up = sc->sc_irs + 1; 775 tp->rcv_wnd = sc->sc_wnd; 776 tp->rcv_adv += tp->rcv_wnd; 777 778 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 779 if (sc->sc_flags & SCF_NOOPT) 780 tp->t_flags |= TF_NOOPT; 781 if (sc->sc_flags & SCF_WINSCALE) { 782 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 783 tp->requested_s_scale = sc->sc_requested_s_scale; 784 tp->request_r_scale = sc->sc_request_r_scale; 785 } 786 if (sc->sc_flags & SCF_TIMESTAMP) { 787 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 788 tp->ts_recent = sc->sc_tsrecent; 789 tp->ts_recent_age = ticks; 790 } 791 if (sc->sc_flags & SCF_CC) { 792 /* 793 * Initialization of the tcpcb for transaction; 794 * set SND.WND = SEG.WND, 795 * initialize CCsend and CCrecv. 796 */ 797 tp->t_flags |= TF_REQ_CC | TF_RCVD_CC; 798 tp->cc_send = sc->sc_cc_send; 799 tp->cc_recv = sc->sc_cc_recv; 800 } 801 if (sc->sc_flags & SCF_SACK_PERMITTED) 802 tp->t_flags |= TF_SACK_PERMITTED; 803 804 tcp_mss(tp, sc->sc_peer_mss); 805 806 /* 807 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 808 */ 809 if (sc->sc_rxtslot != 0) 810 tp->snd_cwnd = tp->t_maxseg; 811 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 812 813 tcpstat.tcps_accepts++; 814 return (so); 815 816 abort: 817 if (so != NULL) 818 soabort(so); 819 return (NULL); 820 } 821 822 /* 823 * This function gets called when we receive an ACK for a 824 * socket in the LISTEN state. We look up the connection 825 * in the syncache, and if its there, we pull it out of 826 * the cache and turn it into a full-blown connection in 827 * the SYN-RECEIVED state. 828 */ 829 int 830 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop, 831 struct mbuf *m) 832 { 833 struct syncache *sc; 834 struct syncache_head *sch; 835 struct socket *so; 836 837 sc = syncache_lookup(inc, &sch); 838 if (sc == NULL) { 839 /* 840 * There is no syncache entry, so see if this ACK is 841 * a returning syncookie. To do this, first: 842 * A. See if this socket has had a syncache entry dropped in 843 * the past. We don't want to accept a bogus syncookie 844 * if we've never received a SYN. 845 * B. check that the syncookie is valid. If it is, then 846 * cobble up a fake syncache entry, and return. 847 */ 848 if (!tcp_syncookies) 849 return (0); 850 sc = syncookie_lookup(inc, th, *sop); 851 if (sc == NULL) 852 return (0); 853 sch = NULL; 854 tcpstat.tcps_sc_recvcookie++; 855 } 856 857 /* 858 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 859 */ 860 if (th->th_ack != sc->sc_iss + 1) 861 return (0); 862 863 so = syncache_socket(sc, *sop); 864 if (so == NULL) { 865 #if 0 866 resetandabort: 867 /* XXXjlemon check this - is this correct? */ 868 tcp_respond(NULL, m, m, th, 869 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 870 #endif 871 m_freem(m); /* XXX only needed for above */ 872 tcpstat.tcps_sc_aborted++; 873 } else { 874 tcpstat.tcps_sc_completed++; 875 } 876 if (sch == NULL) 877 syncache_free(sc); 878 else 879 syncache_drop(sc, sch); 880 *sop = so; 881 return (1); 882 } 883 884 /* 885 * Given a LISTEN socket and an inbound SYN request, add 886 * this to the syn cache, and send back a segment: 887 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 888 * to the source. 889 * 890 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 891 * Doing so would require that we hold onto the data and deliver it 892 * to the application. However, if we are the target of a SYN-flood 893 * DoS attack, an attacker could send data which would eventually 894 * consume all available buffer space if it were ACKed. By not ACKing 895 * the data, we avoid this DoS scenario. 896 */ 897 int 898 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 899 struct socket **sop, struct mbuf *m) 900 { 901 struct tcp_syncache_percpu *syncache_percpu; 902 struct tcpcb *tp; 903 struct socket *so; 904 struct syncache *sc = NULL; 905 struct syncache_head *sch; 906 struct mbuf *ipopts = NULL; 907 struct rmxp_tao *taop; 908 int win; 909 910 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 911 so = *sop; 912 tp = sototcpcb(so); 913 914 /* 915 * Remember the IP options, if any. 916 */ 917 #ifdef INET6 918 if (!inc->inc_isipv6) 919 #endif 920 ipopts = ip_srcroute(); 921 922 /* 923 * See if we already have an entry for this connection. 924 * If we do, resend the SYN,ACK, and reset the retransmit timer. 925 * 926 * XXX 927 * The syncache should be re-initialized with the contents 928 * of the new SYN which may have different options. 929 */ 930 sc = syncache_lookup(inc, &sch); 931 if (sc != NULL) { 932 tcpstat.tcps_sc_dupsyn++; 933 if (ipopts) { 934 /* 935 * If we were remembering a previous source route, 936 * forget it and use the new one we've been given. 937 */ 938 if (sc->sc_ipopts) 939 m_free(sc->sc_ipopts); 940 sc->sc_ipopts = ipopts; 941 } 942 /* 943 * Update timestamp if present. 944 */ 945 if (sc->sc_flags & SCF_TIMESTAMP) 946 sc->sc_tsrecent = to->to_tsval; 947 948 /* Just update the TOF_SACK_PERMITTED for now. */ 949 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 950 sc->sc_flags |= SCF_SACK_PERMITTED; 951 else 952 sc->sc_flags &= ~SCF_SACK_PERMITTED; 953 954 /* 955 * PCB may have changed, pick up new values. 956 */ 957 sc->sc_tp = tp; 958 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 959 if (syncache_respond(sc, m) == 0) { 960 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 961 sc, sc_timerq); 962 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 963 tcpstat.tcps_sndacks++; 964 tcpstat.tcps_sndtotal++; 965 } 966 *sop = NULL; 967 return (1); 968 } 969 970 /* 971 * This allocation is guaranteed to succeed because we 972 * preallocate one more syncache entry than cache_limit. 973 */ 974 sc = zalloc(tcp_syncache.zone); 975 976 /* 977 * Fill in the syncache values. 978 */ 979 sc->sc_tp = tp; 980 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 981 sc->sc_ipopts = ipopts; 982 sc->sc_inc.inc_fport = inc->inc_fport; 983 sc->sc_inc.inc_lport = inc->inc_lport; 984 #ifdef INET6 985 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 986 if (inc->inc_isipv6) { 987 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 988 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 989 sc->sc_route6.ro_rt = NULL; 990 } else 991 #endif 992 { 993 sc->sc_inc.inc_faddr = inc->inc_faddr; 994 sc->sc_inc.inc_laddr = inc->inc_laddr; 995 sc->sc_route.ro_rt = NULL; 996 } 997 sc->sc_irs = th->th_seq; 998 sc->sc_flags = 0; 999 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 1000 if (tcp_syncookies) 1001 sc->sc_iss = syncookie_generate(sc); 1002 else 1003 sc->sc_iss = arc4random(); 1004 1005 /* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */ 1006 win = sbspace(&so->so_rcv); 1007 win = imax(win, 0); 1008 win = imin(win, TCP_MAXWIN); 1009 sc->sc_wnd = win; 1010 1011 if (tcp_do_rfc1323) { 1012 /* 1013 * A timestamp received in a SYN makes 1014 * it ok to send timestamp requests and replies. 1015 */ 1016 if (to->to_flags & TOF_TS) { 1017 sc->sc_tsrecent = to->to_tsval; 1018 sc->sc_flags |= SCF_TIMESTAMP; 1019 } 1020 if (to->to_flags & TOF_SCALE) { 1021 int wscale = 0; 1022 1023 /* Compute proper scaling value from buffer space */ 1024 while (wscale < TCP_MAX_WINSHIFT && 1025 (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat) 1026 wscale++; 1027 sc->sc_request_r_scale = wscale; 1028 sc->sc_requested_s_scale = to->to_requested_s_scale; 1029 sc->sc_flags |= SCF_WINSCALE; 1030 } 1031 } 1032 if (tcp_do_rfc1644) { 1033 /* 1034 * A CC or CC.new option received in a SYN makes 1035 * it ok to send CC in subsequent segments. 1036 */ 1037 if (to->to_flags & (TOF_CC | TOF_CCNEW)) { 1038 sc->sc_cc_recv = to->to_cc; 1039 sc->sc_cc_send = CC_INC(tcp_ccgen); 1040 sc->sc_flags |= SCF_CC; 1041 } 1042 } 1043 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1044 sc->sc_flags |= SCF_SACK_PERMITTED; 1045 if (tp->t_flags & TF_NOOPT) 1046 sc->sc_flags = SCF_NOOPT; 1047 1048 /* 1049 * XXX 1050 * We have the option here of not doing TAO (even if the segment 1051 * qualifies) and instead fall back to a normal 3WHS via the syncache. 1052 * This allows us to apply synflood protection to TAO-qualifying SYNs 1053 * also. However, there should be a hueristic to determine when to 1054 * do this, and is not present at the moment. 1055 */ 1056 1057 /* 1058 * Perform TAO test on incoming CC (SEG.CC) option, if any. 1059 * - compare SEG.CC against cached CC from the same host, if any. 1060 * - if SEG.CC > chached value, SYN must be new and is accepted 1061 * immediately: save new CC in the cache, mark the socket 1062 * connected, enter ESTABLISHED state, turn on flag to 1063 * send a SYN in the next segment. 1064 * A virtual advertised window is set in rcv_adv to 1065 * initialize SWS prevention. Then enter normal segment 1066 * processing: drop SYN, process data and FIN. 1067 * - otherwise do a normal 3-way handshake. 1068 */ 1069 taop = tcp_gettaocache(&sc->sc_inc); 1070 if (to->to_flags & TOF_CC) { 1071 if ((tp->t_flags & TF_NOPUSH) && 1072 sc->sc_flags & SCF_CC && 1073 taop != NULL && taop->tao_cc != 0 && 1074 CC_GT(to->to_cc, taop->tao_cc)) { 1075 sc->sc_rxtslot = 0; 1076 so = syncache_socket(sc, *sop); 1077 if (so != NULL) { 1078 taop->tao_cc = to->to_cc; 1079 *sop = so; 1080 } 1081 syncache_free(sc); 1082 return (so != NULL); 1083 } 1084 } else { 1085 /* 1086 * No CC option, but maybe CC.NEW: invalidate cached value. 1087 */ 1088 if (taop != NULL) 1089 taop->tao_cc = 0; 1090 } 1091 /* 1092 * TAO test failed or there was no CC option, 1093 * do a standard 3-way handshake. 1094 */ 1095 if (syncache_respond(sc, m) == 0) { 1096 syncache_insert(sc, sch); 1097 tcpstat.tcps_sndacks++; 1098 tcpstat.tcps_sndtotal++; 1099 } else { 1100 syncache_free(sc); 1101 tcpstat.tcps_sc_dropped++; 1102 } 1103 *sop = NULL; 1104 return (1); 1105 } 1106 1107 static int 1108 syncache_respond(struct syncache *sc, struct mbuf *m) 1109 { 1110 u_int8_t *optp; 1111 int optlen, error; 1112 u_int16_t tlen, hlen, mssopt; 1113 struct ip *ip = NULL; 1114 struct rtentry *rt; 1115 struct tcphdr *th; 1116 struct ip6_hdr *ip6 = NULL; 1117 #ifdef INET6 1118 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1119 #else 1120 const boolean_t isipv6 = FALSE; 1121 #endif 1122 1123 if (isipv6) { 1124 rt = tcp_rtlookup6(&sc->sc_inc); 1125 if (rt != NULL) 1126 mssopt = rt->rt_ifp->if_mtu - 1127 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1128 else 1129 mssopt = tcp_v6mssdflt; 1130 hlen = sizeof(struct ip6_hdr); 1131 } else { 1132 rt = tcp_rtlookup(&sc->sc_inc); 1133 if (rt != NULL) 1134 mssopt = rt->rt_ifp->if_mtu - 1135 (sizeof(struct ip) + sizeof(struct tcphdr)); 1136 else 1137 mssopt = tcp_mssdflt; 1138 hlen = sizeof(struct ip); 1139 } 1140 1141 /* Compute the size of the TCP options. */ 1142 if (sc->sc_flags & SCF_NOOPT) { 1143 optlen = 0; 1144 } else { 1145 optlen = TCPOLEN_MAXSEG + 1146 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1147 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1148 ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0) + 1149 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1150 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1151 } 1152 tlen = hlen + sizeof(struct tcphdr) + optlen; 1153 1154 /* 1155 * XXX 1156 * assume that the entire packet will fit in a header mbuf 1157 */ 1158 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1159 1160 /* 1161 * XXX shouldn't this reuse the mbuf if possible ? 1162 * Create the IP+TCP header from scratch. 1163 */ 1164 if (m) 1165 m_freem(m); 1166 1167 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 1168 if (m == NULL) 1169 return (ENOBUFS); 1170 m->m_data += max_linkhdr; 1171 m->m_len = tlen; 1172 m->m_pkthdr.len = tlen; 1173 m->m_pkthdr.rcvif = NULL; 1174 1175 if (isipv6) { 1176 ip6 = mtod(m, struct ip6_hdr *); 1177 ip6->ip6_vfc = IPV6_VERSION; 1178 ip6->ip6_nxt = IPPROTO_TCP; 1179 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1180 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1181 ip6->ip6_plen = htons(tlen - hlen); 1182 /* ip6_hlim is set after checksum */ 1183 /* ip6_flow = ??? */ 1184 1185 th = (struct tcphdr *)(ip6 + 1); 1186 } else { 1187 ip = mtod(m, struct ip *); 1188 ip->ip_v = IPVERSION; 1189 ip->ip_hl = sizeof(struct ip) >> 2; 1190 ip->ip_len = tlen; 1191 ip->ip_id = 0; 1192 ip->ip_off = 0; 1193 ip->ip_sum = 0; 1194 ip->ip_p = IPPROTO_TCP; 1195 ip->ip_src = sc->sc_inc.inc_laddr; 1196 ip->ip_dst = sc->sc_inc.inc_faddr; 1197 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1198 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1199 1200 /* 1201 * See if we should do MTU discovery. Route lookups are 1202 * expensive, so we will only unset the DF bit if: 1203 * 1204 * 1) path_mtu_discovery is disabled 1205 * 2) the SCF_UNREACH flag has been set 1206 */ 1207 if (path_mtu_discovery 1208 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1209 ip->ip_off |= IP_DF; 1210 } 1211 1212 th = (struct tcphdr *)(ip + 1); 1213 } 1214 th->th_sport = sc->sc_inc.inc_lport; 1215 th->th_dport = sc->sc_inc.inc_fport; 1216 1217 th->th_seq = htonl(sc->sc_iss); 1218 th->th_ack = htonl(sc->sc_irs + 1); 1219 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1220 th->th_x2 = 0; 1221 th->th_flags = TH_SYN | TH_ACK; 1222 th->th_win = htons(sc->sc_wnd); 1223 th->th_urp = 0; 1224 1225 /* Tack on the TCP options. */ 1226 if (optlen == 0) 1227 goto no_options; 1228 optp = (u_int8_t *)(th + 1); 1229 *optp++ = TCPOPT_MAXSEG; 1230 *optp++ = TCPOLEN_MAXSEG; 1231 *optp++ = (mssopt >> 8) & 0xff; 1232 *optp++ = mssopt & 0xff; 1233 1234 if (sc->sc_flags & SCF_WINSCALE) { 1235 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1236 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1237 sc->sc_request_r_scale); 1238 optp += 4; 1239 } 1240 1241 if (sc->sc_flags & SCF_TIMESTAMP) { 1242 u_int32_t *lp = (u_int32_t *)(optp); 1243 1244 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1245 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1246 *lp++ = htonl(ticks); 1247 *lp = htonl(sc->sc_tsrecent); 1248 optp += TCPOLEN_TSTAMP_APPA; 1249 } 1250 1251 /* 1252 * Send CC and CC.echo if we received CC from our peer. 1253 */ 1254 if (sc->sc_flags & SCF_CC) { 1255 u_int32_t *lp = (u_int32_t *)(optp); 1256 1257 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC)); 1258 *lp++ = htonl(sc->sc_cc_send); 1259 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO)); 1260 *lp = htonl(sc->sc_cc_recv); 1261 optp += TCPOLEN_CC_APPA * 2; 1262 } 1263 1264 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1265 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1266 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1267 } 1268 1269 no_options: 1270 if (isipv6) { 1271 struct route_in6 *ro6 = &sc->sc_route6; 1272 1273 th->th_sum = 0; 1274 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1275 ip6->ip6_hlim = in6_selecthlim(NULL, 1276 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1277 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1278 sc->sc_tp->t_inpcb); 1279 } else { 1280 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1281 htons(tlen - hlen + IPPROTO_TCP)); 1282 m->m_pkthdr.csum_flags = CSUM_TCP; 1283 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1284 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 0, NULL, 1285 sc->sc_tp->t_inpcb); 1286 } 1287 return (error); 1288 } 1289 1290 /* 1291 * cookie layers: 1292 * 1293 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1294 * | peer iss | 1295 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1296 * | 0 |(A)| | 1297 * (A): peer mss index 1298 */ 1299 1300 /* 1301 * The values below are chosen to minimize the size of the tcp_secret 1302 * table, as well as providing roughly a 16 second lifetime for the cookie. 1303 */ 1304 1305 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1306 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1307 1308 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1309 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1310 #define SYNCOOKIE_TIMEOUT \ 1311 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1312 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1313 1314 static struct { 1315 u_int32_t ts_secbits[4]; 1316 u_int ts_expire; 1317 } tcp_secret[SYNCOOKIE_NSECRETS]; 1318 1319 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1320 1321 static MD5_CTX syn_ctx; 1322 1323 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1324 1325 struct md5_add { 1326 u_int32_t laddr, faddr; 1327 u_int32_t secbits[4]; 1328 u_int16_t lport, fport; 1329 }; 1330 1331 #ifdef CTASSERT 1332 CTASSERT(sizeof(struct md5_add) == 28); 1333 #endif 1334 1335 /* 1336 * Consider the problem of a recreated (and retransmitted) cookie. If the 1337 * original SYN was accepted, the connection is established. The second 1338 * SYN is inflight, and if it arrives with an ISN that falls within the 1339 * receive window, the connection is killed. 1340 * 1341 * However, since cookies have other problems, this may not be worth 1342 * worrying about. 1343 */ 1344 1345 static u_int32_t 1346 syncookie_generate(struct syncache *sc) 1347 { 1348 u_int32_t md5_buffer[4]; 1349 u_int32_t data; 1350 int idx, i; 1351 struct md5_add add; 1352 #ifdef INET6 1353 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1354 #else 1355 const boolean_t isipv6 = FALSE; 1356 #endif 1357 1358 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1359 if (tcp_secret[idx].ts_expire < ticks) { 1360 for (i = 0; i < 4; i++) 1361 tcp_secret[idx].ts_secbits[i] = arc4random(); 1362 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1363 } 1364 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1365 if (tcp_msstab[data] <= sc->sc_peer_mss) 1366 break; 1367 data = (data << SYNCOOKIE_WNDBITS) | idx; 1368 data ^= sc->sc_irs; /* peer's iss */ 1369 MD5Init(&syn_ctx); 1370 if (isipv6) { 1371 MD5Add(sc->sc_inc.inc6_laddr); 1372 MD5Add(sc->sc_inc.inc6_faddr); 1373 add.laddr = 0; 1374 add.faddr = 0; 1375 } else { 1376 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1377 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1378 } 1379 add.lport = sc->sc_inc.inc_lport; 1380 add.fport = sc->sc_inc.inc_fport; 1381 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1382 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1383 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1384 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1385 MD5Add(add); 1386 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1387 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1388 return (data); 1389 } 1390 1391 static struct syncache * 1392 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so) 1393 { 1394 u_int32_t md5_buffer[4]; 1395 struct syncache *sc; 1396 u_int32_t data; 1397 int wnd, idx; 1398 struct md5_add add; 1399 1400 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1401 idx = data & SYNCOOKIE_WNDMASK; 1402 if (tcp_secret[idx].ts_expire < ticks || 1403 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1404 return (NULL); 1405 MD5Init(&syn_ctx); 1406 #ifdef INET6 1407 if (inc->inc_isipv6) { 1408 MD5Add(inc->inc6_laddr); 1409 MD5Add(inc->inc6_faddr); 1410 add.laddr = 0; 1411 add.faddr = 0; 1412 } else 1413 #endif 1414 { 1415 add.laddr = inc->inc_laddr.s_addr; 1416 add.faddr = inc->inc_faddr.s_addr; 1417 } 1418 add.lport = inc->inc_lport; 1419 add.fport = inc->inc_fport; 1420 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1421 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1422 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1423 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1424 MD5Add(add); 1425 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1426 data ^= md5_buffer[0]; 1427 if (data & ~SYNCOOKIE_DATAMASK) 1428 return (NULL); 1429 data = data >> SYNCOOKIE_WNDBITS; 1430 1431 /* 1432 * This allocation is guaranteed to succeed because we 1433 * preallocate one more syncache entry than cache_limit. 1434 */ 1435 sc = zalloc(tcp_syncache.zone); 1436 1437 /* 1438 * Fill in the syncache values. 1439 * XXX duplicate code from syncache_add 1440 */ 1441 sc->sc_ipopts = NULL; 1442 sc->sc_inc.inc_fport = inc->inc_fport; 1443 sc->sc_inc.inc_lport = inc->inc_lport; 1444 #ifdef INET6 1445 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1446 if (inc->inc_isipv6) { 1447 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1448 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1449 sc->sc_route6.ro_rt = NULL; 1450 } else 1451 #endif 1452 { 1453 sc->sc_inc.inc_faddr = inc->inc_faddr; 1454 sc->sc_inc.inc_laddr = inc->inc_laddr; 1455 sc->sc_route.ro_rt = NULL; 1456 } 1457 sc->sc_irs = th->th_seq - 1; 1458 sc->sc_iss = th->th_ack - 1; 1459 wnd = sbspace(&so->so_rcv); 1460 wnd = imax(wnd, 0); 1461 wnd = imin(wnd, TCP_MAXWIN); 1462 sc->sc_wnd = wnd; 1463 sc->sc_flags = 0; 1464 sc->sc_rxtslot = 0; 1465 sc->sc_peer_mss = tcp_msstab[data]; 1466 return (sc); 1467 } 1468