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 1.22 2005/03/04 05:57:50 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 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, ZONE_INTERRUPT, 0); 367 tcp_syncache.cache_limit -= 1; 368 } 369 370 static void 371 syncache_insert(sc, sch) 372 struct syncache *sc; 373 struct syncache_head *sch; 374 { 375 struct tcp_syncache_percpu *syncache_percpu; 376 struct syncache *sc2; 377 int i; 378 379 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 380 381 /* 382 * Make sure that we don't overflow the per-bucket 383 * limit or the total cache size limit. 384 */ 385 if (sch->sch_length >= tcp_syncache.bucket_limit) { 386 /* 387 * The bucket is full, toss the oldest element. 388 */ 389 sc2 = TAILQ_FIRST(&sch->sch_bucket); 390 sc2->sc_tp->ts_recent = ticks; 391 syncache_drop(sc2, sch); 392 tcpstat.tcps_sc_bucketoverflow++; 393 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) { 394 /* 395 * The cache is full. Toss the oldest entry in the 396 * entire cache. This is the front entry in the 397 * first non-empty timer queue with the largest 398 * timeout value. 399 */ 400 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 401 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]); 402 if (sc2 != NULL) 403 break; 404 } 405 sc2->sc_tp->ts_recent = ticks; 406 syncache_drop(sc2, NULL); 407 tcpstat.tcps_sc_cacheoverflow++; 408 } 409 410 /* Initialize the entry's timer. */ 411 syncache_timeout(syncache_percpu, sc, 0); 412 413 /* Put it into the bucket. */ 414 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 415 sch->sch_length++; 416 syncache_percpu->cache_count++; 417 tcpstat.tcps_sc_added++; 418 } 419 420 static void 421 syncache_drop(sc, sch) 422 struct syncache *sc; 423 struct syncache_head *sch; 424 { 425 struct tcp_syncache_percpu *syncache_percpu; 426 #ifdef INET6 427 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 428 #else 429 const boolean_t isipv6 = FALSE; 430 #endif 431 432 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 433 434 if (sch == NULL) { 435 if (isipv6) { 436 sch = &syncache_percpu->hashbase[ 437 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 438 } else { 439 sch = &syncache_percpu->hashbase[ 440 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 441 } 442 } 443 444 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 445 sch->sch_length--; 446 syncache_percpu->cache_count--; 447 448 /* 449 * Remove the entry from the syncache timer/timeout queue. Note 450 * that we do not try to stop any running timer since we do not know 451 * whether the timer's message is in-transit or not. Since timeouts 452 * are fairly long, taking an unneeded callout does not detrimentally 453 * effect performance. 454 */ 455 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq); 456 457 syncache_free(sc); 458 } 459 460 /* 461 * Place a timeout message on the TCP thread's message queue. 462 * This routine runs in soft interrupt context. 463 * 464 * An invariant is for this routine to be called, the callout must 465 * have been active. Note that the callout is not deactivated until 466 * after the message has been processed in syncache_timer_handler() below. 467 */ 468 static void 469 syncache_timer(void *p) 470 { 471 struct netmsg_sc_timer *msg = p; 472 473 lwkt_sendmsg(msg->nm_mrec->port, &msg->nm_lmsg); 474 } 475 476 /* 477 * Service a timer message queued by timer expiration. 478 * This routine runs in the TCP protocol thread. 479 * 480 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 481 * If we have retransmitted an entry the maximum number of times, expire it. 482 * 483 * When we finish processing timed-out entries, we restart the timer if there 484 * are any entries still on the queue and deactivate it otherwise. Only after 485 * a timer has been deactivated here can it be restarted by syncache_timeout(). 486 */ 487 static int 488 syncache_timer_handler(lwkt_msg_t msg) 489 { 490 struct tcp_syncache_percpu *syncache_percpu; 491 struct syncache *sc, *nsc; 492 struct inpcb *inp; 493 int slot; 494 495 slot = ((struct netmsg_sc_timer *)msg)->nm_mrec->slot; 496 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 497 498 nsc = TAILQ_FIRST(&syncache_percpu->timerq[slot]); 499 while (nsc != NULL) { 500 if (ticks < nsc->sc_rxttime) 501 break; /* finished because timerq sorted by time */ 502 sc = nsc; 503 inp = sc->sc_tp->t_inpcb; 504 if (slot == SYNCACHE_MAXREXMTS || 505 slot >= tcp_syncache.rexmt_limit || 506 inp->inp_gencnt != sc->sc_inp_gencnt) { 507 nsc = TAILQ_NEXT(sc, sc_timerq); 508 syncache_drop(sc, NULL); 509 tcpstat.tcps_sc_stale++; 510 continue; 511 } 512 /* 513 * syncache_respond() may call back into the syncache to 514 * to modify another entry, so do not obtain the next 515 * entry on the timer chain until it has completed. 516 */ 517 syncache_respond(sc, NULL); 518 nsc = TAILQ_NEXT(sc, sc_timerq); 519 tcpstat.tcps_sc_retransmitted++; 520 TAILQ_REMOVE(&syncache_percpu->timerq[slot], sc, sc_timerq); 521 syncache_timeout(syncache_percpu, sc, slot + 1); 522 } 523 if (nsc != NULL) 524 callout_reset(&syncache_percpu->tt_timerq[slot], 525 nsc->sc_rxttime - ticks, syncache_timer, 526 &syncache_percpu->mrec[slot]); 527 else 528 callout_deactivate(&syncache_percpu->tt_timerq[slot]); 529 530 lwkt_replymsg(msg, 0); 531 return (EASYNC); 532 } 533 534 /* 535 * Find an entry in the syncache. 536 */ 537 struct syncache * 538 syncache_lookup(inc, schp) 539 struct in_conninfo *inc; 540 struct syncache_head **schp; 541 { 542 struct tcp_syncache_percpu *syncache_percpu; 543 struct syncache *sc; 544 struct syncache_head *sch; 545 546 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 547 #ifdef INET6 548 if (inc->inc_isipv6) { 549 sch = &syncache_percpu->hashbase[ 550 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 551 *schp = sch; 552 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) 553 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 554 return (sc); 555 } else 556 #endif 557 { 558 sch = &syncache_percpu->hashbase[ 559 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 560 *schp = sch; 561 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 562 #ifdef INET6 563 if (sc->sc_inc.inc_isipv6) 564 continue; 565 #endif 566 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 567 return (sc); 568 } 569 } 570 return (NULL); 571 } 572 573 /* 574 * This function is called when we get a RST for a 575 * non-existent connection, so that we can see if the 576 * connection is in the syn cache. If it is, zap it. 577 */ 578 void 579 syncache_chkrst(inc, th) 580 struct in_conninfo *inc; 581 struct tcphdr *th; 582 { 583 struct syncache *sc; 584 struct syncache_head *sch; 585 586 sc = syncache_lookup(inc, &sch); 587 if (sc == NULL) 588 return; 589 /* 590 * If the RST bit is set, check the sequence number to see 591 * if this is a valid reset segment. 592 * RFC 793 page 37: 593 * In all states except SYN-SENT, all reset (RST) segments 594 * are validated by checking their SEQ-fields. A reset is 595 * valid if its sequence number is in the window. 596 * 597 * The sequence number in the reset segment is normally an 598 * echo of our outgoing acknowlegement numbers, but some hosts 599 * send a reset with the sequence number at the rightmost edge 600 * of our receive window, and we have to handle this case. 601 */ 602 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 603 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 604 syncache_drop(sc, sch); 605 tcpstat.tcps_sc_reset++; 606 } 607 } 608 609 void 610 syncache_badack(inc) 611 struct in_conninfo *inc; 612 { 613 struct syncache *sc; 614 struct syncache_head *sch; 615 616 sc = syncache_lookup(inc, &sch); 617 if (sc != NULL) { 618 syncache_drop(sc, sch); 619 tcpstat.tcps_sc_badack++; 620 } 621 } 622 623 void 624 syncache_unreach(inc, th) 625 struct in_conninfo *inc; 626 struct tcphdr *th; 627 { 628 struct syncache *sc; 629 struct syncache_head *sch; 630 631 /* we are called at splnet() here */ 632 sc = syncache_lookup(inc, &sch); 633 if (sc == NULL) 634 return; 635 636 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 637 if (ntohl(th->th_seq) != sc->sc_iss) 638 return; 639 640 /* 641 * If we've rertransmitted 3 times and this is our second error, 642 * we remove the entry. Otherwise, we allow it to continue on. 643 * This prevents us from incorrectly nuking an entry during a 644 * spurious network outage. 645 * 646 * See tcp_notify(). 647 */ 648 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 649 sc->sc_flags |= SCF_UNREACH; 650 return; 651 } 652 syncache_drop(sc, sch); 653 tcpstat.tcps_sc_unreach++; 654 } 655 656 /* 657 * Build a new TCP socket structure from a syncache entry. 658 */ 659 static struct socket * 660 syncache_socket(sc, lso) 661 struct syncache *sc; 662 struct socket *lso; 663 { 664 struct inpcb *inp = NULL, *linp; 665 struct socket *so; 666 struct tcpcb *tp; 667 #ifdef INET6 668 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 669 #else 670 const boolean_t isipv6 = FALSE; 671 #endif 672 673 /* 674 * Ok, create the full blown connection, and set things up 675 * as they would have been set up if we had created the 676 * connection when the SYN arrived. If we can't create 677 * the connection, abort it. 678 */ 679 so = sonewconn(lso, SS_ISCONNECTED); 680 if (so == NULL) { 681 /* 682 * Drop the connection; we will send a RST if the peer 683 * retransmits the ACK, 684 */ 685 tcpstat.tcps_listendrop++; 686 goto abort; 687 } 688 689 inp = so->so_pcb; 690 691 /* 692 * Insert new socket into hash list. 693 */ 694 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 695 if (isipv6) { 696 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 697 } else { 698 #ifdef INET6 699 inp->inp_vflag &= ~INP_IPV6; 700 inp->inp_vflag |= INP_IPV4; 701 #endif 702 inp->inp_laddr = sc->sc_inc.inc_laddr; 703 } 704 inp->inp_lport = sc->sc_inc.inc_lport; 705 if (in_pcbinsporthash(inp) != 0) { 706 /* 707 * Undo the assignments above if we failed to 708 * put the PCB on the hash lists. 709 */ 710 if (isipv6) 711 inp->in6p_laddr = in6addr_any; 712 else 713 inp->inp_laddr.s_addr = INADDR_ANY; 714 inp->inp_lport = 0; 715 goto abort; 716 } 717 linp = so->so_pcb; 718 #ifdef IPSEC 719 /* copy old policy into new socket's */ 720 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 721 printf("syncache_expand: could not copy policy\n"); 722 #endif 723 if (isipv6) { 724 struct in6_addr laddr6; 725 struct sockaddr_in6 sin6; 726 /* 727 * Inherit socket options from the listening socket. 728 * Note that in6p_inputopts are not (and should not be) 729 * copied, since it stores previously received options and is 730 * used to detect if each new option is different than the 731 * previous one and hence should be passed to a user. 732 * If we copied in6p_inputopts, a user would not be able to 733 * receive options just after calling the accept system call. 734 */ 735 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 736 if (linp->in6p_outputopts) 737 inp->in6p_outputopts = 738 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 739 inp->in6p_route = sc->sc_route6; 740 sc->sc_route6.ro_rt = NULL; 741 742 sin6.sin6_family = AF_INET6; 743 sin6.sin6_len = sizeof sin6; 744 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 745 sin6.sin6_port = sc->sc_inc.inc_fport; 746 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 747 laddr6 = inp->in6p_laddr; 748 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 749 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 750 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) { 751 inp->in6p_laddr = laddr6; 752 goto abort; 753 } 754 } else { 755 struct in_addr laddr; 756 struct sockaddr_in sin; 757 758 inp->inp_options = ip_srcroute(); 759 if (inp->inp_options == NULL) { 760 inp->inp_options = sc->sc_ipopts; 761 sc->sc_ipopts = NULL; 762 } 763 inp->inp_route = sc->sc_route; 764 sc->sc_route.ro_rt = NULL; 765 766 sin.sin_family = AF_INET; 767 sin.sin_len = sizeof sin; 768 sin.sin_addr = sc->sc_inc.inc_faddr; 769 sin.sin_port = sc->sc_inc.inc_fport; 770 bzero(sin.sin_zero, sizeof sin.sin_zero); 771 laddr = inp->inp_laddr; 772 if (inp->inp_laddr.s_addr == INADDR_ANY) 773 inp->inp_laddr = sc->sc_inc.inc_laddr; 774 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) { 775 inp->inp_laddr = laddr; 776 goto abort; 777 } 778 } 779 780 tp = intotcpcb(inp); 781 tp->t_state = TCPS_SYN_RECEIVED; 782 tp->iss = sc->sc_iss; 783 tp->irs = sc->sc_irs; 784 tcp_rcvseqinit(tp); 785 tcp_sendseqinit(tp); 786 tp->snd_wl1 = sc->sc_irs; 787 tp->rcv_up = sc->sc_irs + 1; 788 tp->rcv_wnd = sc->sc_wnd; 789 tp->rcv_adv += tp->rcv_wnd; 790 791 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 792 if (sc->sc_flags & SCF_NOOPT) 793 tp->t_flags |= TF_NOOPT; 794 if (sc->sc_flags & SCF_WINSCALE) { 795 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 796 tp->requested_s_scale = sc->sc_requested_s_scale; 797 tp->request_r_scale = sc->sc_request_r_scale; 798 } 799 if (sc->sc_flags & SCF_TIMESTAMP) { 800 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 801 tp->ts_recent = sc->sc_tsrecent; 802 tp->ts_recent_age = ticks; 803 } 804 if (sc->sc_flags & SCF_CC) { 805 /* 806 * Initialization of the tcpcb for transaction; 807 * set SND.WND = SEG.WND, 808 * initialize CCsend and CCrecv. 809 */ 810 tp->t_flags |= TF_REQ_CC | TF_RCVD_CC; 811 tp->cc_send = sc->sc_cc_send; 812 tp->cc_recv = sc->sc_cc_recv; 813 } 814 if (sc->sc_flags & SCF_SACK_PERMITTED) 815 tp->t_flags |= TF_SACK_PERMITTED; 816 817 tcp_mss(tp, sc->sc_peer_mss); 818 819 /* 820 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 821 */ 822 if (sc->sc_rxtslot != 0) 823 tp->snd_cwnd = tp->t_maxseg; 824 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 825 826 tcpstat.tcps_accepts++; 827 return (so); 828 829 abort: 830 if (so != NULL) 831 soabort(so); 832 return (NULL); 833 } 834 835 /* 836 * This function gets called when we receive an ACK for a 837 * socket in the LISTEN state. We look up the connection 838 * in the syncache, and if its there, we pull it out of 839 * the cache and turn it into a full-blown connection in 840 * the SYN-RECEIVED state. 841 */ 842 int 843 syncache_expand(inc, th, sop, m) 844 struct in_conninfo *inc; 845 struct tcphdr *th; 846 struct socket **sop; 847 struct mbuf *m; 848 { 849 struct syncache *sc; 850 struct syncache_head *sch; 851 struct socket *so; 852 853 sc = syncache_lookup(inc, &sch); 854 if (sc == NULL) { 855 /* 856 * There is no syncache entry, so see if this ACK is 857 * a returning syncookie. To do this, first: 858 * A. See if this socket has had a syncache entry dropped in 859 * the past. We don't want to accept a bogus syncookie 860 * if we've never received a SYN. 861 * B. check that the syncookie is valid. If it is, then 862 * cobble up a fake syncache entry, and return. 863 */ 864 if (!tcp_syncookies) 865 return (0); 866 sc = syncookie_lookup(inc, th, *sop); 867 if (sc == NULL) 868 return (0); 869 sch = NULL; 870 tcpstat.tcps_sc_recvcookie++; 871 } 872 873 /* 874 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 875 */ 876 if (th->th_ack != sc->sc_iss + 1) 877 return (0); 878 879 so = syncache_socket(sc, *sop); 880 if (so == NULL) { 881 #if 0 882 resetandabort: 883 /* XXXjlemon check this - is this correct? */ 884 tcp_respond(NULL, m, m, th, 885 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 886 #endif 887 m_freem(m); /* XXX only needed for above */ 888 tcpstat.tcps_sc_aborted++; 889 } else { 890 tcpstat.tcps_sc_completed++; 891 } 892 if (sch == NULL) 893 syncache_free(sc); 894 else 895 syncache_drop(sc, sch); 896 *sop = so; 897 return (1); 898 } 899 900 /* 901 * Given a LISTEN socket and an inbound SYN request, add 902 * this to the syn cache, and send back a segment: 903 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 904 * to the source. 905 * 906 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 907 * Doing so would require that we hold onto the data and deliver it 908 * to the application. However, if we are the target of a SYN-flood 909 * DoS attack, an attacker could send data which would eventually 910 * consume all available buffer space if it were ACKed. By not ACKing 911 * the data, we avoid this DoS scenario. 912 */ 913 int 914 syncache_add(inc, to, th, sop, m) 915 struct in_conninfo *inc; 916 struct tcpopt *to; 917 struct tcphdr *th; 918 struct socket **sop; 919 struct mbuf *m; 920 { 921 struct tcp_syncache_percpu *syncache_percpu; 922 struct tcpcb *tp; 923 struct socket *so; 924 struct syncache *sc = NULL; 925 struct syncache_head *sch; 926 struct mbuf *ipopts = NULL; 927 struct rmxp_tao *taop; 928 int win; 929 930 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 931 so = *sop; 932 tp = sototcpcb(so); 933 934 /* 935 * Remember the IP options, if any. 936 */ 937 #ifdef INET6 938 if (!inc->inc_isipv6) 939 #endif 940 ipopts = ip_srcroute(); 941 942 /* 943 * See if we already have an entry for this connection. 944 * If we do, resend the SYN,ACK, and reset the retransmit timer. 945 * 946 * XXX 947 * The syncache should be re-initialized with the contents 948 * of the new SYN which may have different options. 949 */ 950 sc = syncache_lookup(inc, &sch); 951 if (sc != NULL) { 952 tcpstat.tcps_sc_dupsyn++; 953 if (ipopts) { 954 /* 955 * If we were remembering a previous source route, 956 * forget it and use the new one we've been given. 957 */ 958 if (sc->sc_ipopts) 959 m_free(sc->sc_ipopts); 960 sc->sc_ipopts = ipopts; 961 } 962 /* 963 * Update timestamp if present. 964 */ 965 if (sc->sc_flags & SCF_TIMESTAMP) 966 sc->sc_tsrecent = to->to_tsval; 967 968 /* Just update the TOF_SACK_PERMITTED for now. */ 969 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 970 sc->sc_flags |= SCF_SACK_PERMITTED; 971 else 972 sc->sc_flags &= ~SCF_SACK_PERMITTED; 973 974 /* 975 * PCB may have changed, pick up new values. 976 */ 977 sc->sc_tp = tp; 978 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 979 if (syncache_respond(sc, m) == 0) { 980 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 981 sc, sc_timerq); 982 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 983 tcpstat.tcps_sndacks++; 984 tcpstat.tcps_sndtotal++; 985 } 986 *sop = NULL; 987 return (1); 988 } 989 990 /* 991 * This allocation is guaranteed to succeed because we 992 * preallocate one more syncache entry than cache_limit. 993 */ 994 sc = zalloc(tcp_syncache.zone); 995 996 /* 997 * Fill in the syncache values. 998 */ 999 sc->sc_tp = tp; 1000 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1001 sc->sc_ipopts = ipopts; 1002 sc->sc_inc.inc_fport = inc->inc_fport; 1003 sc->sc_inc.inc_lport = inc->inc_lport; 1004 #ifdef INET6 1005 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1006 if (inc->inc_isipv6) { 1007 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1008 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1009 sc->sc_route6.ro_rt = NULL; 1010 } else 1011 #endif 1012 { 1013 sc->sc_inc.inc_faddr = inc->inc_faddr; 1014 sc->sc_inc.inc_laddr = inc->inc_laddr; 1015 sc->sc_route.ro_rt = NULL; 1016 } 1017 sc->sc_irs = th->th_seq; 1018 sc->sc_flags = 0; 1019 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 1020 if (tcp_syncookies) 1021 sc->sc_iss = syncookie_generate(sc); 1022 else 1023 sc->sc_iss = arc4random(); 1024 1025 /* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */ 1026 win = sbspace(&so->so_rcv); 1027 win = imax(win, 0); 1028 win = imin(win, TCP_MAXWIN); 1029 sc->sc_wnd = win; 1030 1031 if (tcp_do_rfc1323) { 1032 /* 1033 * A timestamp received in a SYN makes 1034 * it ok to send timestamp requests and replies. 1035 */ 1036 if (to->to_flags & TOF_TS) { 1037 sc->sc_tsrecent = to->to_tsval; 1038 sc->sc_flags |= SCF_TIMESTAMP; 1039 } 1040 if (to->to_flags & TOF_SCALE) { 1041 int wscale = 0; 1042 1043 /* Compute proper scaling value from buffer space */ 1044 while (wscale < TCP_MAX_WINSHIFT && 1045 (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat) 1046 wscale++; 1047 sc->sc_request_r_scale = wscale; 1048 sc->sc_requested_s_scale = to->to_requested_s_scale; 1049 sc->sc_flags |= SCF_WINSCALE; 1050 } 1051 } 1052 if (tcp_do_rfc1644) { 1053 /* 1054 * A CC or CC.new option received in a SYN makes 1055 * it ok to send CC in subsequent segments. 1056 */ 1057 if (to->to_flags & (TOF_CC | TOF_CCNEW)) { 1058 sc->sc_cc_recv = to->to_cc; 1059 sc->sc_cc_send = CC_INC(tcp_ccgen); 1060 sc->sc_flags |= SCF_CC; 1061 } 1062 } 1063 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1064 sc->sc_flags |= SCF_SACK_PERMITTED; 1065 if (tp->t_flags & TF_NOOPT) 1066 sc->sc_flags = SCF_NOOPT; 1067 1068 /* 1069 * XXX 1070 * We have the option here of not doing TAO (even if the segment 1071 * qualifies) and instead fall back to a normal 3WHS via the syncache. 1072 * This allows us to apply synflood protection to TAO-qualifying SYNs 1073 * also. However, there should be a hueristic to determine when to 1074 * do this, and is not present at the moment. 1075 */ 1076 1077 /* 1078 * Perform TAO test on incoming CC (SEG.CC) option, if any. 1079 * - compare SEG.CC against cached CC from the same host, if any. 1080 * - if SEG.CC > chached value, SYN must be new and is accepted 1081 * immediately: save new CC in the cache, mark the socket 1082 * connected, enter ESTABLISHED state, turn on flag to 1083 * send a SYN in the next segment. 1084 * A virtual advertised window is set in rcv_adv to 1085 * initialize SWS prevention. Then enter normal segment 1086 * processing: drop SYN, process data and FIN. 1087 * - otherwise do a normal 3-way handshake. 1088 */ 1089 taop = tcp_gettaocache(&sc->sc_inc); 1090 if (to->to_flags & TOF_CC) { 1091 if ((tp->t_flags & TF_NOPUSH) && 1092 sc->sc_flags & SCF_CC && 1093 taop != NULL && taop->tao_cc != 0 && 1094 CC_GT(to->to_cc, taop->tao_cc)) { 1095 sc->sc_rxtslot = 0; 1096 so = syncache_socket(sc, *sop); 1097 if (so != NULL) { 1098 taop->tao_cc = to->to_cc; 1099 *sop = so; 1100 } 1101 syncache_free(sc); 1102 return (so != NULL); 1103 } 1104 } else { 1105 /* 1106 * No CC option, but maybe CC.NEW: invalidate cached value. 1107 */ 1108 if (taop != NULL) 1109 taop->tao_cc = 0; 1110 } 1111 /* 1112 * TAO test failed or there was no CC option, 1113 * do a standard 3-way handshake. 1114 */ 1115 if (syncache_respond(sc, m) == 0) { 1116 syncache_insert(sc, sch); 1117 tcpstat.tcps_sndacks++; 1118 tcpstat.tcps_sndtotal++; 1119 } else { 1120 syncache_free(sc); 1121 tcpstat.tcps_sc_dropped++; 1122 } 1123 *sop = NULL; 1124 return (1); 1125 } 1126 1127 static int 1128 syncache_respond(sc, m) 1129 struct syncache *sc; 1130 struct mbuf *m; 1131 { 1132 u_int8_t *optp; 1133 int optlen, error; 1134 u_int16_t tlen, hlen, mssopt; 1135 struct ip *ip = NULL; 1136 struct rtentry *rt; 1137 struct tcphdr *th; 1138 struct ip6_hdr *ip6 = NULL; 1139 #ifdef INET6 1140 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1141 #else 1142 const boolean_t isipv6 = FALSE; 1143 #endif 1144 1145 if (isipv6) { 1146 rt = tcp_rtlookup6(&sc->sc_inc); 1147 if (rt != NULL) 1148 mssopt = rt->rt_ifp->if_mtu - 1149 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1150 else 1151 mssopt = tcp_v6mssdflt; 1152 hlen = sizeof(struct ip6_hdr); 1153 } else { 1154 rt = tcp_rtlookup(&sc->sc_inc); 1155 if (rt != NULL) 1156 mssopt = rt->rt_ifp->if_mtu - 1157 (sizeof(struct ip) + sizeof(struct tcphdr)); 1158 else 1159 mssopt = tcp_mssdflt; 1160 hlen = sizeof(struct ip); 1161 } 1162 1163 /* Compute the size of the TCP options. */ 1164 if (sc->sc_flags & SCF_NOOPT) { 1165 optlen = 0; 1166 } else { 1167 optlen = TCPOLEN_MAXSEG + 1168 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1169 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1170 ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0) + 1171 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1172 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1173 } 1174 tlen = hlen + sizeof(struct tcphdr) + optlen; 1175 1176 /* 1177 * XXX 1178 * assume that the entire packet will fit in a header mbuf 1179 */ 1180 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1181 1182 /* 1183 * XXX shouldn't this reuse the mbuf if possible ? 1184 * Create the IP+TCP header from scratch. 1185 */ 1186 if (m) 1187 m_freem(m); 1188 1189 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 1190 if (m == NULL) 1191 return (ENOBUFS); 1192 m->m_data += max_linkhdr; 1193 m->m_len = tlen; 1194 m->m_pkthdr.len = tlen; 1195 m->m_pkthdr.rcvif = NULL; 1196 1197 if (isipv6) { 1198 ip6 = mtod(m, struct ip6_hdr *); 1199 ip6->ip6_vfc = IPV6_VERSION; 1200 ip6->ip6_nxt = IPPROTO_TCP; 1201 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1202 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1203 ip6->ip6_plen = htons(tlen - hlen); 1204 /* ip6_hlim is set after checksum */ 1205 /* ip6_flow = ??? */ 1206 1207 th = (struct tcphdr *)(ip6 + 1); 1208 } else { 1209 ip = mtod(m, struct ip *); 1210 ip->ip_v = IPVERSION; 1211 ip->ip_hl = sizeof(struct ip) >> 2; 1212 ip->ip_len = tlen; 1213 ip->ip_id = 0; 1214 ip->ip_off = 0; 1215 ip->ip_sum = 0; 1216 ip->ip_p = IPPROTO_TCP; 1217 ip->ip_src = sc->sc_inc.inc_laddr; 1218 ip->ip_dst = sc->sc_inc.inc_faddr; 1219 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1220 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1221 1222 /* 1223 * See if we should do MTU discovery. Route lookups are 1224 * expensive, so we will only unset the DF bit if: 1225 * 1226 * 1) path_mtu_discovery is disabled 1227 * 2) the SCF_UNREACH flag has been set 1228 */ 1229 if (path_mtu_discovery 1230 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1231 ip->ip_off |= IP_DF; 1232 } 1233 1234 th = (struct tcphdr *)(ip + 1); 1235 } 1236 th->th_sport = sc->sc_inc.inc_lport; 1237 th->th_dport = sc->sc_inc.inc_fport; 1238 1239 th->th_seq = htonl(sc->sc_iss); 1240 th->th_ack = htonl(sc->sc_irs + 1); 1241 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1242 th->th_x2 = 0; 1243 th->th_flags = TH_SYN | TH_ACK; 1244 th->th_win = htons(sc->sc_wnd); 1245 th->th_urp = 0; 1246 1247 /* Tack on the TCP options. */ 1248 if (optlen == 0) 1249 goto no_options; 1250 optp = (u_int8_t *)(th + 1); 1251 *optp++ = TCPOPT_MAXSEG; 1252 *optp++ = TCPOLEN_MAXSEG; 1253 *optp++ = (mssopt >> 8) & 0xff; 1254 *optp++ = mssopt & 0xff; 1255 1256 if (sc->sc_flags & SCF_WINSCALE) { 1257 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1258 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1259 sc->sc_request_r_scale); 1260 optp += 4; 1261 } 1262 1263 if (sc->sc_flags & SCF_TIMESTAMP) { 1264 u_int32_t *lp = (u_int32_t *)(optp); 1265 1266 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1267 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1268 *lp++ = htonl(ticks); 1269 *lp = htonl(sc->sc_tsrecent); 1270 optp += TCPOLEN_TSTAMP_APPA; 1271 } 1272 1273 /* 1274 * Send CC and CC.echo if we received CC from our peer. 1275 */ 1276 if (sc->sc_flags & SCF_CC) { 1277 u_int32_t *lp = (u_int32_t *)(optp); 1278 1279 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC)); 1280 *lp++ = htonl(sc->sc_cc_send); 1281 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO)); 1282 *lp = htonl(sc->sc_cc_recv); 1283 optp += TCPOLEN_CC_APPA * 2; 1284 } 1285 1286 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1287 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1288 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1289 } 1290 1291 no_options: 1292 if (isipv6) { 1293 struct route_in6 *ro6 = &sc->sc_route6; 1294 1295 th->th_sum = 0; 1296 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1297 ip6->ip6_hlim = in6_selecthlim(NULL, 1298 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1299 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1300 sc->sc_tp->t_inpcb); 1301 } else { 1302 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1303 htons(tlen - hlen + IPPROTO_TCP)); 1304 m->m_pkthdr.csum_flags = CSUM_TCP; 1305 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1306 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 0, NULL, 1307 sc->sc_tp->t_inpcb); 1308 } 1309 return (error); 1310 } 1311 1312 /* 1313 * cookie layers: 1314 * 1315 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1316 * | peer iss | 1317 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1318 * | 0 |(A)| | 1319 * (A): peer mss index 1320 */ 1321 1322 /* 1323 * The values below are chosen to minimize the size of the tcp_secret 1324 * table, as well as providing roughly a 16 second lifetime for the cookie. 1325 */ 1326 1327 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1328 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1329 1330 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1331 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1332 #define SYNCOOKIE_TIMEOUT \ 1333 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1334 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1335 1336 static struct { 1337 u_int32_t ts_secbits[4]; 1338 u_int ts_expire; 1339 } tcp_secret[SYNCOOKIE_NSECRETS]; 1340 1341 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1342 1343 static MD5_CTX syn_ctx; 1344 1345 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1346 1347 struct md5_add { 1348 u_int32_t laddr, faddr; 1349 u_int32_t secbits[4]; 1350 u_int16_t lport, fport; 1351 }; 1352 1353 #ifdef CTASSERT 1354 CTASSERT(sizeof(struct md5_add) == 28); 1355 #endif 1356 1357 /* 1358 * Consider the problem of a recreated (and retransmitted) cookie. If the 1359 * original SYN was accepted, the connection is established. The second 1360 * SYN is inflight, and if it arrives with an ISN that falls within the 1361 * receive window, the connection is killed. 1362 * 1363 * However, since cookies have other problems, this may not be worth 1364 * worrying about. 1365 */ 1366 1367 static u_int32_t 1368 syncookie_generate(struct syncache *sc) 1369 { 1370 u_int32_t md5_buffer[4]; 1371 u_int32_t data; 1372 int idx, i; 1373 struct md5_add add; 1374 #ifdef INET6 1375 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1376 #else 1377 const boolean_t isipv6 = FALSE; 1378 #endif 1379 1380 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1381 if (tcp_secret[idx].ts_expire < ticks) { 1382 for (i = 0; i < 4; i++) 1383 tcp_secret[idx].ts_secbits[i] = arc4random(); 1384 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1385 } 1386 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1387 if (tcp_msstab[data] <= sc->sc_peer_mss) 1388 break; 1389 data = (data << SYNCOOKIE_WNDBITS) | idx; 1390 data ^= sc->sc_irs; /* peer's iss */ 1391 MD5Init(&syn_ctx); 1392 if (isipv6) { 1393 MD5Add(sc->sc_inc.inc6_laddr); 1394 MD5Add(sc->sc_inc.inc6_faddr); 1395 add.laddr = 0; 1396 add.faddr = 0; 1397 } else { 1398 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1399 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1400 } 1401 add.lport = sc->sc_inc.inc_lport; 1402 add.fport = sc->sc_inc.inc_fport; 1403 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1404 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1405 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1406 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1407 MD5Add(add); 1408 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1409 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1410 return (data); 1411 } 1412 1413 static struct syncache * 1414 syncookie_lookup(inc, th, so) 1415 struct in_conninfo *inc; 1416 struct tcphdr *th; 1417 struct socket *so; 1418 { 1419 u_int32_t md5_buffer[4]; 1420 struct syncache *sc; 1421 u_int32_t data; 1422 int wnd, idx; 1423 struct md5_add add; 1424 1425 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1426 idx = data & SYNCOOKIE_WNDMASK; 1427 if (tcp_secret[idx].ts_expire < ticks || 1428 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1429 return (NULL); 1430 MD5Init(&syn_ctx); 1431 #ifdef INET6 1432 if (inc->inc_isipv6) { 1433 MD5Add(inc->inc6_laddr); 1434 MD5Add(inc->inc6_faddr); 1435 add.laddr = 0; 1436 add.faddr = 0; 1437 } else 1438 #endif 1439 { 1440 add.laddr = inc->inc_laddr.s_addr; 1441 add.faddr = inc->inc_faddr.s_addr; 1442 } 1443 add.lport = inc->inc_lport; 1444 add.fport = inc->inc_fport; 1445 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1446 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1447 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1448 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1449 MD5Add(add); 1450 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1451 data ^= md5_buffer[0]; 1452 if (data & ~SYNCOOKIE_DATAMASK) 1453 return (NULL); 1454 data = data >> SYNCOOKIE_WNDBITS; 1455 1456 /* 1457 * This allocation is guaranteed to succeed because we 1458 * preallocate one more syncache entry than cache_limit. 1459 */ 1460 sc = zalloc(tcp_syncache.zone); 1461 1462 /* 1463 * Fill in the syncache values. 1464 * XXX duplicate code from syncache_add 1465 */ 1466 sc->sc_ipopts = NULL; 1467 sc->sc_inc.inc_fport = inc->inc_fport; 1468 sc->sc_inc.inc_lport = inc->inc_lport; 1469 #ifdef INET6 1470 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1471 if (inc->inc_isipv6) { 1472 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1473 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1474 sc->sc_route6.ro_rt = NULL; 1475 } else 1476 #endif 1477 { 1478 sc->sc_inc.inc_faddr = inc->inc_faddr; 1479 sc->sc_inc.inc_laddr = inc->inc_laddr; 1480 sc->sc_route.ro_rt = NULL; 1481 } 1482 sc->sc_irs = th->th_seq - 1; 1483 sc->sc_iss = th->th_ack - 1; 1484 wnd = sbspace(&so->so_rcv); 1485 wnd = imax(wnd, 0); 1486 wnd = imin(wnd, TCP_MAXWIN); 1487 sc->sc_wnd = wnd; 1488 sc->sc_flags = 0; 1489 sc->sc_rxtslot = 0; 1490 sc->sc_peer_mss = tcp_msstab[data]; 1491 return (sc); 1492 } 1493