1 /* 2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * All advertising materials mentioning features or use of this software 36 * must display the following acknowledgement: 37 * This product includes software developed by Jeffrey M. Hsu. 38 * 39 * Copyright (c) 2001 Networks Associates Technologies, Inc. 40 * All rights reserved. 41 * 42 * This software was developed for the FreeBSD Project by Jonathan Lemon 43 * and NAI Labs, the Security Research Division of Network Associates, Inc. 44 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 45 * DARPA CHATS research program. 46 * 47 * Redistribution and use in source and binary forms, with or without 48 * modification, are permitted provided that the following conditions 49 * are met: 50 * 1. Redistributions of source code must retain the above copyright 51 * notice, this list of conditions and the following disclaimer. 52 * 2. Redistributions in binary form must reproduce the above copyright 53 * notice, this list of conditions and the following disclaimer in the 54 * documentation and/or other materials provided with the distribution. 55 * 3. The name of the author may not be used to endorse or promote 56 * products derived from this software without specific prior written 57 * permission. 58 * 59 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 62 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 69 * SUCH DAMAGE. 70 * 71 * $FreeBSD: src/sys/netinet/tcp_syncache.c,v 1.5.2.14 2003/02/24 04:02:27 silby Exp $ 72 * $DragonFly: src/sys/netinet/tcp_syncache.c,v 1.35 2008/11/22 11:03:35 sephe Exp $ 73 */ 74 75 #include "opt_inet.h" 76 #include "opt_inet6.h" 77 #include "opt_ipsec.h" 78 79 #include <sys/param.h> 80 #include <sys/systm.h> 81 #include <sys/kernel.h> 82 #include <sys/sysctl.h> 83 #include <sys/malloc.h> 84 #include <sys/mbuf.h> 85 #include <sys/md5.h> 86 #include <sys/proc.h> /* for proc0 declaration */ 87 #include <sys/random.h> 88 #include <sys/socket.h> 89 #include <sys/socketvar.h> 90 #include <sys/in_cksum.h> 91 92 #include <sys/msgport2.h> 93 #include <net/netmsg2.h> 94 95 #include <net/if.h> 96 #include <net/route.h> 97 98 #include <netinet/in.h> 99 #include <netinet/in_systm.h> 100 #include <netinet/ip.h> 101 #include <netinet/in_var.h> 102 #include <netinet/in_pcb.h> 103 #include <netinet/ip_var.h> 104 #include <netinet/ip6.h> 105 #ifdef INET6 106 #include <netinet/icmp6.h> 107 #include <netinet6/nd6.h> 108 #endif 109 #include <netinet6/ip6_var.h> 110 #include <netinet6/in6_pcb.h> 111 #include <netinet/tcp.h> 112 #include <netinet/tcp_fsm.h> 113 #include <netinet/tcp_seq.h> 114 #include <netinet/tcp_timer.h> 115 #include <netinet/tcp_timer2.h> 116 #include <netinet/tcp_var.h> 117 #include <netinet6/tcp6_var.h> 118 119 #ifdef IPSEC 120 #include <netinet6/ipsec.h> 121 #ifdef INET6 122 #include <netinet6/ipsec6.h> 123 #endif 124 #include <netproto/key/key.h> 125 #endif /*IPSEC*/ 126 127 #ifdef FAST_IPSEC 128 #include <netproto/ipsec/ipsec.h> 129 #ifdef INET6 130 #include <netproto/ipsec/ipsec6.h> 131 #endif 132 #include <netproto/ipsec/key.h> 133 #define IPSEC 134 #endif /*FAST_IPSEC*/ 135 136 static int tcp_syncookies = 1; 137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 138 &tcp_syncookies, 0, 139 "Use TCP SYN cookies if the syncache overflows"); 140 141 static void syncache_drop(struct syncache *, struct syncache_head *); 142 static void syncache_free(struct syncache *); 143 static void syncache_insert(struct syncache *, struct syncache_head *); 144 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 145 static int syncache_respond(struct syncache *, struct mbuf *); 146 static struct socket *syncache_socket(struct syncache *, struct socket *, 147 struct mbuf *); 148 static void syncache_timer(void *); 149 static u_int32_t syncookie_generate(struct syncache *); 150 static struct syncache *syncookie_lookup(struct in_conninfo *, 151 struct tcphdr *, struct socket *); 152 153 /* 154 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 155 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 156 * the odds are that the user has given up attempting to connect by then. 157 */ 158 #define SYNCACHE_MAXREXMTS 3 159 160 /* Arbitrary values */ 161 #define TCP_SYNCACHE_HASHSIZE 512 162 #define TCP_SYNCACHE_BUCKETLIMIT 30 163 164 struct netmsg_sc_timer { 165 struct netmsg_base base; 166 struct msgrec *nm_mrec; /* back pointer to containing msgrec */ 167 }; 168 169 struct msgrec { 170 struct netmsg_sc_timer msg; 171 lwkt_port_t port; /* constant after init */ 172 int slot; /* constant after init */ 173 }; 174 175 static void syncache_timer_handler(netmsg_t); 176 177 struct tcp_syncache { 178 u_int hashsize; 179 u_int hashmask; 180 u_int bucket_limit; 181 u_int cache_limit; 182 u_int rexmt_limit; 183 u_int hash_secret; 184 }; 185 static struct tcp_syncache tcp_syncache; 186 187 TAILQ_HEAD(syncache_list, syncache); 188 189 struct tcp_syncache_percpu { 190 struct syncache_head *hashbase; 191 u_int cache_count; 192 struct syncache_list timerq[SYNCACHE_MAXREXMTS + 1]; 193 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1]; 194 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1]; 195 }; 196 static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU]; 197 198 static struct lwkt_port syncache_null_rport; 199 200 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 201 202 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD, 203 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 204 205 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD, 206 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 207 208 /* XXX JH */ 209 #if 0 210 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 211 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 212 #endif 213 214 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD, 215 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 216 217 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 218 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 219 220 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 221 222 #define SYNCACHE_HASH(inc, mask) \ 223 ((tcp_syncache.hash_secret ^ \ 224 (inc)->inc_faddr.s_addr ^ \ 225 ((inc)->inc_faddr.s_addr >> 16) ^ \ 226 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 227 228 #define SYNCACHE_HASH6(inc, mask) \ 229 ((tcp_syncache.hash_secret ^ \ 230 (inc)->inc6_faddr.s6_addr32[0] ^ \ 231 (inc)->inc6_faddr.s6_addr32[3] ^ \ 232 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 233 234 #define ENDPTS_EQ(a, b) ( \ 235 (a)->ie_fport == (b)->ie_fport && \ 236 (a)->ie_lport == (b)->ie_lport && \ 237 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 238 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 239 ) 240 241 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 242 243 static __inline void 244 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu, 245 struct syncache *sc, int slot) 246 { 247 sc->sc_rxtslot = slot; 248 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[slot]; 249 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq); 250 if (!callout_active(&syncache_percpu->tt_timerq[slot])) { 251 callout_reset(&syncache_percpu->tt_timerq[slot], 252 TCPTV_RTOBASE * tcp_backoff[slot], 253 syncache_timer, 254 &syncache_percpu->mrec[slot]); 255 } 256 } 257 258 static void 259 syncache_free(struct syncache *sc) 260 { 261 struct rtentry *rt; 262 #ifdef INET6 263 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 264 #else 265 const boolean_t isipv6 = FALSE; 266 #endif 267 268 if (sc->sc_ipopts) 269 m_free(sc->sc_ipopts); 270 271 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt; 272 if (rt != NULL) { 273 /* 274 * If this is the only reference to a protocol-cloned 275 * route, remove it immediately. 276 */ 277 if ((rt->rt_flags & RTF_WASCLONED) && rt->rt_refcnt == 1) 278 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 279 rt_mask(rt), rt->rt_flags, NULL); 280 RTFREE(rt); 281 } 282 kfree(sc, M_SYNCACHE); 283 } 284 285 void 286 syncache_init(void) 287 { 288 int i, cpu; 289 290 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 291 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 292 tcp_syncache.cache_limit = 293 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 294 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 295 tcp_syncache.hash_secret = karc4random(); 296 297 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 298 &tcp_syncache.hashsize); 299 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 300 &tcp_syncache.cache_limit); 301 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 302 &tcp_syncache.bucket_limit); 303 if (!powerof2(tcp_syncache.hashsize)) { 304 kprintf("WARNING: syncache hash size is not a power of 2.\n"); 305 tcp_syncache.hashsize = 512; /* safe default */ 306 } 307 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 308 309 lwkt_initport_replyonly_null(&syncache_null_rport); 310 311 for (cpu = 0; cpu < ncpus2; cpu++) { 312 struct tcp_syncache_percpu *syncache_percpu; 313 314 syncache_percpu = &tcp_syncache_percpu[cpu]; 315 /* Allocate the hash table. */ 316 MALLOC(syncache_percpu->hashbase, struct syncache_head *, 317 tcp_syncache.hashsize * sizeof(struct syncache_head), 318 M_SYNCACHE, M_WAITOK); 319 320 /* Initialize the hash buckets. */ 321 for (i = 0; i < tcp_syncache.hashsize; i++) { 322 struct syncache_head *bucket; 323 324 bucket = &syncache_percpu->hashbase[i]; 325 TAILQ_INIT(&bucket->sch_bucket); 326 bucket->sch_length = 0; 327 } 328 329 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 330 /* Initialize the timer queues. */ 331 TAILQ_INIT(&syncache_percpu->timerq[i]); 332 callout_init(&syncache_percpu->tt_timerq[i]); 333 334 syncache_percpu->mrec[i].slot = i; 335 syncache_percpu->mrec[i].port = cpu_portfn(cpu); 336 syncache_percpu->mrec[i].msg.nm_mrec = 337 &syncache_percpu->mrec[i]; 338 netmsg_init(&syncache_percpu->mrec[i].msg.base, 339 NULL, &syncache_null_rport, 340 0, syncache_timer_handler); 341 } 342 } 343 } 344 345 static void 346 syncache_insert(struct syncache *sc, struct syncache_head *sch) 347 { 348 struct tcp_syncache_percpu *syncache_percpu; 349 struct syncache *sc2; 350 int i; 351 352 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 353 354 /* 355 * Make sure that we don't overflow the per-bucket 356 * limit or the total cache size limit. 357 */ 358 if (sch->sch_length >= tcp_syncache.bucket_limit) { 359 /* 360 * The bucket is full, toss the oldest element. 361 */ 362 sc2 = TAILQ_FIRST(&sch->sch_bucket); 363 sc2->sc_tp->ts_recent = ticks; 364 syncache_drop(sc2, sch); 365 tcpstat.tcps_sc_bucketoverflow++; 366 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) { 367 /* 368 * The cache is full. Toss the oldest entry in the 369 * entire cache. This is the front entry in the 370 * first non-empty timer queue with the largest 371 * timeout value. 372 */ 373 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 374 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]); 375 while (sc2 && (sc2->sc_flags & SCF_MARKER)) 376 sc2 = TAILQ_NEXT(sc2, sc_timerq); 377 if (sc2 != NULL) 378 break; 379 } 380 sc2->sc_tp->ts_recent = ticks; 381 syncache_drop(sc2, NULL); 382 tcpstat.tcps_sc_cacheoverflow++; 383 } 384 385 /* Initialize the entry's timer. */ 386 syncache_timeout(syncache_percpu, sc, 0); 387 388 /* Put it into the bucket. */ 389 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 390 sch->sch_length++; 391 syncache_percpu->cache_count++; 392 tcpstat.tcps_sc_added++; 393 } 394 395 void 396 syncache_destroy(struct tcpcb *tp) 397 { 398 struct tcp_syncache_percpu *syncache_percpu; 399 struct syncache_head *bucket; 400 struct syncache *sc; 401 int i; 402 403 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 404 sc = NULL; 405 406 for (i = 0; i < tcp_syncache.hashsize; i++) { 407 bucket = &syncache_percpu->hashbase[i]; 408 TAILQ_FOREACH(sc, &bucket->sch_bucket, sc_hash) { 409 if (sc->sc_tp == tp) { 410 sc->sc_tp = NULL; 411 tp->t_flags &= ~TF_SYNCACHE; 412 break; 413 } 414 } 415 } 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 * Cleanup 446 */ 447 if (sc->sc_tp) { 448 sc->sc_tp->t_flags &= ~TF_SYNCACHE; 449 sc->sc_tp = NULL; 450 } 451 452 /* 453 * Remove the entry from the syncache timer/timeout queue. Note 454 * that we do not try to stop any running timer since we do not know 455 * whether the timer's message is in-transit or not. Since timeouts 456 * are fairly long, taking an unneeded callout does not detrimentally 457 * effect performance. 458 */ 459 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq); 460 461 syncache_free(sc); 462 } 463 464 /* 465 * Place a timeout message on the TCP thread's message queue. 466 * This routine runs in soft interrupt context. 467 * 468 * An invariant is for this routine to be called, the callout must 469 * have been active. Note that the callout is not deactivated until 470 * after the message has been processed in syncache_timer_handler() below. 471 */ 472 static void 473 syncache_timer(void *p) 474 { 475 struct netmsg_sc_timer *msg = p; 476 477 lwkt_sendmsg(msg->nm_mrec->port, &msg->base.lmsg); 478 } 479 480 /* 481 * Service a timer message queued by timer expiration. 482 * This routine runs in the TCP protocol thread. 483 * 484 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 485 * If we have retransmitted an entry the maximum number of times, expire it. 486 * 487 * When we finish processing timed-out entries, we restart the timer if there 488 * are any entries still on the queue and deactivate it otherwise. Only after 489 * a timer has been deactivated here can it be restarted by syncache_timeout(). 490 */ 491 static void 492 syncache_timer_handler(netmsg_t msg) 493 { 494 struct tcp_syncache_percpu *syncache_percpu; 495 struct syncache *sc; 496 struct syncache marker; 497 struct syncache_list *list; 498 struct inpcb *inp; 499 int slot; 500 501 slot = ((struct netmsg_sc_timer *)msg)->nm_mrec->slot; 502 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 503 504 list = &syncache_percpu->timerq[slot]; 505 506 /* 507 * Use a marker to keep our place in the scan. syncache_drop() 508 * can block and cause any next pointer we cache to become stale. 509 */ 510 marker.sc_flags = SCF_MARKER; 511 TAILQ_INSERT_HEAD(list, &marker, sc_timerq); 512 513 while ((sc = TAILQ_NEXT(&marker, sc_timerq)) != NULL) { 514 /* 515 * Move the marker. 516 */ 517 TAILQ_REMOVE(list, &marker, sc_timerq); 518 TAILQ_INSERT_AFTER(list, sc, &marker, sc_timerq); 519 520 if (sc->sc_flags & SCF_MARKER) 521 continue; 522 523 if (ticks < sc->sc_rxttime) 524 break; /* finished because timerq sorted by time */ 525 if (sc->sc_tp == NULL) { 526 syncache_drop(sc, NULL); 527 tcpstat.tcps_sc_stale++; 528 continue; 529 } 530 inp = sc->sc_tp->t_inpcb; 531 if (slot == SYNCACHE_MAXREXMTS || 532 slot >= tcp_syncache.rexmt_limit || 533 inp == NULL || 534 inp->inp_gencnt != sc->sc_inp_gencnt) { 535 syncache_drop(sc, NULL); 536 tcpstat.tcps_sc_stale++; 537 continue; 538 } 539 /* 540 * syncache_respond() may call back into the syncache to 541 * to modify another entry, so do not obtain the next 542 * entry on the timer chain until it has completed. 543 */ 544 syncache_respond(sc, NULL); 545 tcpstat.tcps_sc_retransmitted++; 546 TAILQ_REMOVE(list, sc, sc_timerq); 547 syncache_timeout(syncache_percpu, sc, slot + 1); 548 } 549 TAILQ_REMOVE(list, &marker, sc_timerq); 550 551 if (sc != NULL) { 552 callout_reset(&syncache_percpu->tt_timerq[slot], 553 sc->sc_rxttime - ticks, syncache_timer, 554 &syncache_percpu->mrec[slot]); 555 } else { 556 callout_deactivate(&syncache_percpu->tt_timerq[slot]); 557 } 558 lwkt_replymsg(&msg->base.lmsg, 0); 559 } 560 561 /* 562 * Find an entry in the syncache. 563 */ 564 struct syncache * 565 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 566 { 567 struct tcp_syncache_percpu *syncache_percpu; 568 struct syncache *sc; 569 struct syncache_head *sch; 570 571 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 572 #ifdef INET6 573 if (inc->inc_isipv6) { 574 sch = &syncache_percpu->hashbase[ 575 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 576 *schp = sch; 577 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) 578 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 579 return (sc); 580 } else 581 #endif 582 { 583 sch = &syncache_percpu->hashbase[ 584 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 585 *schp = sch; 586 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 587 #ifdef INET6 588 if (sc->sc_inc.inc_isipv6) 589 continue; 590 #endif 591 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 592 return (sc); 593 } 594 } 595 return (NULL); 596 } 597 598 /* 599 * This function is called when we get a RST for a 600 * non-existent connection, so that we can see if the 601 * connection is in the syn cache. If it is, zap it. 602 */ 603 void 604 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 605 { 606 struct syncache *sc; 607 struct syncache_head *sch; 608 609 sc = syncache_lookup(inc, &sch); 610 if (sc == NULL) { 611 return; 612 } 613 /* 614 * If the RST bit is set, check the sequence number to see 615 * if this is a valid reset segment. 616 * RFC 793 page 37: 617 * In all states except SYN-SENT, all reset (RST) segments 618 * are validated by checking their SEQ-fields. A reset is 619 * valid if its sequence number is in the window. 620 * 621 * The sequence number in the reset segment is normally an 622 * echo of our outgoing acknowlegement numbers, but some hosts 623 * send a reset with the sequence number at the rightmost edge 624 * of our receive window, and we have to handle this case. 625 */ 626 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 627 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 628 syncache_drop(sc, sch); 629 tcpstat.tcps_sc_reset++; 630 } 631 } 632 633 void 634 syncache_badack(struct in_conninfo *inc) 635 { 636 struct syncache *sc; 637 struct syncache_head *sch; 638 639 sc = syncache_lookup(inc, &sch); 640 if (sc != NULL) { 641 syncache_drop(sc, sch); 642 tcpstat.tcps_sc_badack++; 643 } 644 } 645 646 void 647 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 648 { 649 struct syncache *sc; 650 struct syncache_head *sch; 651 652 /* we are called at splnet() here */ 653 sc = syncache_lookup(inc, &sch); 654 if (sc == NULL) 655 return; 656 657 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 658 if (ntohl(th->th_seq) != sc->sc_iss) 659 return; 660 661 /* 662 * If we've rertransmitted 3 times and this is our second error, 663 * we remove the entry. Otherwise, we allow it to continue on. 664 * This prevents us from incorrectly nuking an entry during a 665 * spurious network outage. 666 * 667 * See tcp_notify(). 668 */ 669 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 670 sc->sc_flags |= SCF_UNREACH; 671 return; 672 } 673 syncache_drop(sc, sch); 674 tcpstat.tcps_sc_unreach++; 675 } 676 677 /* 678 * Build a new TCP socket structure from a syncache entry. 679 * 680 * This is called from the context of the SYN+ACK 681 */ 682 static struct socket * 683 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 684 { 685 struct inpcb *inp = NULL, *linp; 686 struct socket *so; 687 struct tcpcb *tp; 688 lwkt_port_t port; 689 #ifdef INET6 690 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 691 #else 692 const boolean_t isipv6 = FALSE; 693 #endif 694 695 /* 696 * Ok, create the full blown connection, and set things up 697 * as they would have been set up if we had created the 698 * connection when the SYN arrived. If we can't create 699 * the connection, abort it. 700 * 701 * Set the protocol processing port for the socket to the current 702 * port (that the connection came in on). 703 */ 704 so = sonewconn(lso, SS_ISCONNECTED); 705 if (so == NULL) { 706 /* 707 * Drop the connection; we will send a RST if the peer 708 * retransmits the ACK, 709 */ 710 tcpstat.tcps_listendrop++; 711 goto abort; 712 } 713 714 /* 715 * Insert new socket into hash list. 716 */ 717 inp = so->so_pcb; 718 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 719 if (isipv6) { 720 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 721 } else { 722 #ifdef INET6 723 inp->inp_vflag &= ~INP_IPV6; 724 inp->inp_vflag |= INP_IPV4; 725 inp->inp_flags &= ~IN6P_IPV6_V6ONLY; 726 #endif 727 inp->inp_laddr = sc->sc_inc.inc_laddr; 728 } 729 inp->inp_lport = sc->sc_inc.inc_lport; 730 if (in_pcbinsporthash(inp) != 0) { 731 /* 732 * Undo the assignments above if we failed to 733 * put the PCB on the hash lists. 734 */ 735 if (isipv6) 736 inp->in6p_laddr = kin6addr_any; 737 else 738 inp->inp_laddr.s_addr = INADDR_ANY; 739 inp->inp_lport = 0; 740 goto abort; 741 } 742 linp = lso->so_pcb; 743 #ifdef IPSEC 744 /* copy old policy into new socket's */ 745 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 746 kprintf("syncache_expand: could not copy policy\n"); 747 #endif 748 if (isipv6) { 749 struct in6_addr laddr6; 750 struct sockaddr_in6 sin6; 751 /* 752 * Inherit socket options from the listening socket. 753 * Note that in6p_inputopts are not (and should not be) 754 * copied, since it stores previously received options and is 755 * used to detect if each new option is different than the 756 * previous one and hence should be passed to a user. 757 * If we copied in6p_inputopts, a user would not be able to 758 * receive options just after calling the accept system call. 759 */ 760 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 761 if (linp->in6p_outputopts) 762 inp->in6p_outputopts = 763 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 764 inp->in6p_route = sc->sc_route6; 765 sc->sc_route6.ro_rt = NULL; 766 767 sin6.sin6_family = AF_INET6; 768 sin6.sin6_len = sizeof sin6; 769 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 770 sin6.sin6_port = sc->sc_inc.inc_fport; 771 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 772 laddr6 = inp->in6p_laddr; 773 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 774 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 775 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) { 776 inp->in6p_laddr = laddr6; 777 goto abort; 778 } 779 } else { 780 struct in_addr laddr; 781 struct sockaddr_in sin; 782 783 inp->inp_options = ip_srcroute(m); 784 if (inp->inp_options == NULL) { 785 inp->inp_options = sc->sc_ipopts; 786 sc->sc_ipopts = NULL; 787 } 788 inp->inp_route = sc->sc_route; 789 sc->sc_route.ro_rt = NULL; 790 791 sin.sin_family = AF_INET; 792 sin.sin_len = sizeof sin; 793 sin.sin_addr = sc->sc_inc.inc_faddr; 794 sin.sin_port = sc->sc_inc.inc_fport; 795 bzero(sin.sin_zero, sizeof sin.sin_zero); 796 laddr = inp->inp_laddr; 797 if (inp->inp_laddr.s_addr == INADDR_ANY) 798 inp->inp_laddr = sc->sc_inc.inc_laddr; 799 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) { 800 inp->inp_laddr = laddr; 801 goto abort; 802 } 803 } 804 805 /* 806 * The current port should be in the context of the SYN+ACK and 807 * so should match the tcp address port. 808 * 809 * XXX we may be running on the netisr thread instead of a tcp 810 * thread, in which case port will not match 811 * curthread->td_msgport. 812 */ 813 if (isipv6) { 814 port = tcp6_addrport(); 815 } else { 816 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport, 817 inp->inp_laddr.s_addr, inp->inp_lport); 818 } 819 if (port != &curthread->td_msgport) { 820 print_backtrace(-1); 821 kprintf("TCP PORT MISMATCH %p vs %p\n", 822 port, &curthread->td_msgport); 823 } 824 /*KKASSERT(port == &curthread->td_msgport);*/ 825 826 tp = intotcpcb(inp); 827 tp->t_state = TCPS_SYN_RECEIVED; 828 tp->iss = sc->sc_iss; 829 tp->irs = sc->sc_irs; 830 tcp_rcvseqinit(tp); 831 tcp_sendseqinit(tp); 832 tp->snd_wl1 = sc->sc_irs; 833 tp->rcv_up = sc->sc_irs + 1; 834 tp->rcv_wnd = sc->sc_wnd; 835 tp->rcv_adv += tp->rcv_wnd; 836 837 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 838 if (sc->sc_flags & SCF_NOOPT) 839 tp->t_flags |= TF_NOOPT; 840 if (sc->sc_flags & SCF_WINSCALE) { 841 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 842 tp->requested_s_scale = sc->sc_requested_s_scale; 843 tp->request_r_scale = sc->sc_request_r_scale; 844 } 845 if (sc->sc_flags & SCF_TIMESTAMP) { 846 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 847 tp->ts_recent = sc->sc_tsrecent; 848 tp->ts_recent_age = ticks; 849 } 850 if (sc->sc_flags & SCF_SACK_PERMITTED) 851 tp->t_flags |= TF_SACK_PERMITTED; 852 853 #ifdef TCP_SIGNATURE 854 if (sc->sc_flags & SCF_SIGNATURE) 855 tp->t_flags |= TF_SIGNATURE; 856 #endif /* TCP_SIGNATURE */ 857 858 859 tcp_mss(tp, sc->sc_peer_mss); 860 861 /* 862 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 863 */ 864 if (sc->sc_rxtslot != 0) 865 tp->snd_cwnd = tp->t_maxseg; 866 tcp_create_timermsg(tp, port); 867 tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep); 868 869 tcpstat.tcps_accepts++; 870 return (so); 871 872 abort: 873 if (so != NULL) 874 soabort_oncpu(so); 875 return (NULL); 876 } 877 878 /* 879 * This function gets called when we receive an ACK for a 880 * socket in the LISTEN state. We look up the connection 881 * in the syncache, and if its there, we pull it out of 882 * the cache and turn it into a full-blown connection in 883 * the SYN-RECEIVED state. 884 */ 885 int 886 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop, 887 struct mbuf *m) 888 { 889 struct syncache *sc; 890 struct syncache_head *sch; 891 struct socket *so; 892 893 sc = syncache_lookup(inc, &sch); 894 if (sc == NULL) { 895 /* 896 * There is no syncache entry, so see if this ACK is 897 * a returning syncookie. To do this, first: 898 * A. See if this socket has had a syncache entry dropped in 899 * the past. We don't want to accept a bogus syncookie 900 * if we've never received a SYN. 901 * B. check that the syncookie is valid. If it is, then 902 * cobble up a fake syncache entry, and return. 903 */ 904 if (!tcp_syncookies) 905 return (0); 906 sc = syncookie_lookup(inc, th, *sop); 907 if (sc == NULL) 908 return (0); 909 sch = NULL; 910 tcpstat.tcps_sc_recvcookie++; 911 } 912 913 /* 914 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 915 */ 916 if (th->th_ack != sc->sc_iss + 1) 917 return (0); 918 919 so = syncache_socket(sc, *sop, m); 920 if (so == NULL) { 921 #if 0 922 resetandabort: 923 /* XXXjlemon check this - is this correct? */ 924 tcp_respond(NULL, m, m, th, 925 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 926 #endif 927 m_freem(m); /* XXX only needed for above */ 928 tcpstat.tcps_sc_aborted++; 929 } else { 930 tcpstat.tcps_sc_completed++; 931 } 932 if (sch == NULL) 933 syncache_free(sc); 934 else 935 syncache_drop(sc, sch); 936 *sop = so; 937 return (1); 938 } 939 940 /* 941 * Given a LISTEN socket and an inbound SYN request, add 942 * this to the syn cache, and send back a segment: 943 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 944 * to the source. 945 * 946 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 947 * Doing so would require that we hold onto the data and deliver it 948 * to the application. However, if we are the target of a SYN-flood 949 * DoS attack, an attacker could send data which would eventually 950 * consume all available buffer space if it were ACKed. By not ACKing 951 * the data, we avoid this DoS scenario. 952 */ 953 int 954 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 955 struct socket **sop, struct mbuf *m) 956 { 957 struct tcp_syncache_percpu *syncache_percpu; 958 struct tcpcb *tp; 959 struct socket *so; 960 struct syncache *sc = NULL; 961 struct syncache_head *sch; 962 struct mbuf *ipopts = NULL; 963 int win; 964 965 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 966 so = *sop; 967 tp = sototcpcb(so); 968 969 /* 970 * Remember the IP options, if any. 971 */ 972 #ifdef INET6 973 if (!inc->inc_isipv6) 974 #endif 975 ipopts = ip_srcroute(m); 976 977 /* 978 * See if we already have an entry for this connection. 979 * If we do, resend the SYN,ACK, and reset the retransmit timer. 980 * 981 * XXX 982 * The syncache should be re-initialized with the contents 983 * of the new SYN which may have different options. 984 */ 985 sc = syncache_lookup(inc, &sch); 986 if (sc != NULL) { 987 tcpstat.tcps_sc_dupsyn++; 988 if (ipopts) { 989 /* 990 * If we were remembering a previous source route, 991 * forget it and use the new one we've been given. 992 */ 993 if (sc->sc_ipopts) 994 m_free(sc->sc_ipopts); 995 sc->sc_ipopts = ipopts; 996 } 997 /* 998 * Update timestamp if present. 999 */ 1000 if (sc->sc_flags & SCF_TIMESTAMP) 1001 sc->sc_tsrecent = to->to_tsval; 1002 1003 /* Just update the TOF_SACK_PERMITTED for now. */ 1004 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1005 sc->sc_flags |= SCF_SACK_PERMITTED; 1006 else 1007 sc->sc_flags &= ~SCF_SACK_PERMITTED; 1008 1009 /* 1010 * PCB may have changed, pick up new values. 1011 */ 1012 if (sc->sc_tp) { 1013 sc->sc_tp->t_flags &= ~TF_SYNCACHE; 1014 tp->t_flags |= TF_SYNCACHE; 1015 } 1016 sc->sc_tp = tp; 1017 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1018 if (syncache_respond(sc, m) == 0) { 1019 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 1020 sc, sc_timerq); 1021 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 1022 tcpstat.tcps_sndacks++; 1023 tcpstat.tcps_sndtotal++; 1024 } 1025 *sop = NULL; 1026 return (1); 1027 } 1028 1029 /* 1030 * Fill in the syncache values. 1031 */ 1032 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1033 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1034 sc->sc_ipopts = ipopts; 1035 sc->sc_inc.inc_fport = inc->inc_fport; 1036 sc->sc_inc.inc_lport = inc->inc_lport; 1037 sc->sc_tp = tp; 1038 tp->t_flags |= TF_SYNCACHE; 1039 #ifdef INET6 1040 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1041 if (inc->inc_isipv6) { 1042 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1043 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1044 sc->sc_route6.ro_rt = NULL; 1045 } else 1046 #endif 1047 { 1048 sc->sc_inc.inc_faddr = inc->inc_faddr; 1049 sc->sc_inc.inc_laddr = inc->inc_laddr; 1050 sc->sc_route.ro_rt = NULL; 1051 } 1052 sc->sc_irs = th->th_seq; 1053 sc->sc_flags = 0; 1054 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 1055 if (tcp_syncookies) 1056 sc->sc_iss = syncookie_generate(sc); 1057 else 1058 sc->sc_iss = karc4random(); 1059 1060 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */ 1061 win = ssb_space(&so->so_rcv); 1062 win = imax(win, 0); 1063 win = imin(win, TCP_MAXWIN); 1064 sc->sc_wnd = win; 1065 1066 if (tcp_do_rfc1323) { 1067 /* 1068 * A timestamp received in a SYN makes 1069 * it ok to send timestamp requests and replies. 1070 */ 1071 if (to->to_flags & TOF_TS) { 1072 sc->sc_tsrecent = to->to_tsval; 1073 sc->sc_flags |= SCF_TIMESTAMP; 1074 } 1075 if (to->to_flags & TOF_SCALE) { 1076 int wscale = TCP_MIN_WINSHIFT; 1077 1078 /* Compute proper scaling value from buffer space */ 1079 while (wscale < TCP_MAX_WINSHIFT && 1080 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) { 1081 wscale++; 1082 } 1083 sc->sc_request_r_scale = wscale; 1084 sc->sc_requested_s_scale = to->to_requested_s_scale; 1085 sc->sc_flags |= SCF_WINSCALE; 1086 } 1087 } 1088 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1089 sc->sc_flags |= SCF_SACK_PERMITTED; 1090 if (tp->t_flags & TF_NOOPT) 1091 sc->sc_flags = SCF_NOOPT; 1092 #ifdef TCP_SIGNATURE 1093 /* 1094 * If listening socket requested TCP digests, and received SYN 1095 * contains the option, flag this in the syncache so that 1096 * syncache_respond() will do the right thing with the SYN+ACK. 1097 * XXX Currently we always record the option by default and will 1098 * attempt to use it in syncache_respond(). 1099 */ 1100 if (to->to_flags & TOF_SIGNATURE) 1101 sc->sc_flags = SCF_SIGNATURE; 1102 #endif /* TCP_SIGNATURE */ 1103 1104 if (syncache_respond(sc, m) == 0) { 1105 syncache_insert(sc, sch); 1106 tcpstat.tcps_sndacks++; 1107 tcpstat.tcps_sndtotal++; 1108 } else { 1109 syncache_free(sc); 1110 tcpstat.tcps_sc_dropped++; 1111 } 1112 *sop = NULL; 1113 return (1); 1114 } 1115 1116 static int 1117 syncache_respond(struct syncache *sc, struct mbuf *m) 1118 { 1119 u_int8_t *optp; 1120 int optlen, error; 1121 u_int16_t tlen, hlen, mssopt; 1122 struct ip *ip = NULL; 1123 struct rtentry *rt; 1124 struct tcphdr *th; 1125 struct ip6_hdr *ip6 = NULL; 1126 #ifdef INET6 1127 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1128 #else 1129 const boolean_t isipv6 = FALSE; 1130 #endif 1131 1132 if (isipv6) { 1133 rt = tcp_rtlookup6(&sc->sc_inc); 1134 if (rt != NULL) 1135 mssopt = rt->rt_ifp->if_mtu - 1136 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1137 else 1138 mssopt = tcp_v6mssdflt; 1139 hlen = sizeof(struct ip6_hdr); 1140 } else { 1141 rt = tcp_rtlookup(&sc->sc_inc); 1142 if (rt != NULL) 1143 mssopt = rt->rt_ifp->if_mtu - 1144 (sizeof(struct ip) + sizeof(struct tcphdr)); 1145 else 1146 mssopt = tcp_mssdflt; 1147 hlen = sizeof(struct ip); 1148 } 1149 1150 /* Compute the size of the TCP options. */ 1151 if (sc->sc_flags & SCF_NOOPT) { 1152 optlen = 0; 1153 } else { 1154 optlen = TCPOLEN_MAXSEG + 1155 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1156 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1157 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1158 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1159 #ifdef TCP_SIGNATURE 1160 optlen += ((sc->sc_flags & SCF_SIGNATURE) ? 1161 (TCPOLEN_SIGNATURE + 2) : 0); 1162 #endif /* TCP_SIGNATURE */ 1163 } 1164 tlen = hlen + sizeof(struct tcphdr) + optlen; 1165 1166 /* 1167 * XXX 1168 * assume that the entire packet will fit in a header mbuf 1169 */ 1170 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1171 1172 /* 1173 * XXX shouldn't this reuse the mbuf if possible ? 1174 * Create the IP+TCP header from scratch. 1175 */ 1176 if (m) 1177 m_freem(m); 1178 1179 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 1180 if (m == NULL) 1181 return (ENOBUFS); 1182 m->m_data += max_linkhdr; 1183 m->m_len = tlen; 1184 m->m_pkthdr.len = tlen; 1185 m->m_pkthdr.rcvif = NULL; 1186 1187 if (isipv6) { 1188 ip6 = mtod(m, struct ip6_hdr *); 1189 ip6->ip6_vfc = IPV6_VERSION; 1190 ip6->ip6_nxt = IPPROTO_TCP; 1191 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1192 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1193 ip6->ip6_plen = htons(tlen - hlen); 1194 /* ip6_hlim is set after checksum */ 1195 /* ip6_flow = ??? */ 1196 1197 th = (struct tcphdr *)(ip6 + 1); 1198 } else { 1199 ip = mtod(m, struct ip *); 1200 ip->ip_v = IPVERSION; 1201 ip->ip_hl = sizeof(struct ip) >> 2; 1202 ip->ip_len = tlen; 1203 ip->ip_id = 0; 1204 ip->ip_off = 0; 1205 ip->ip_sum = 0; 1206 ip->ip_p = IPPROTO_TCP; 1207 ip->ip_src = sc->sc_inc.inc_laddr; 1208 ip->ip_dst = sc->sc_inc.inc_faddr; 1209 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1210 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1211 1212 /* 1213 * See if we should do MTU discovery. Route lookups are 1214 * expensive, so we will only unset the DF bit if: 1215 * 1216 * 1) path_mtu_discovery is disabled 1217 * 2) the SCF_UNREACH flag has been set 1218 */ 1219 if (path_mtu_discovery 1220 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1221 ip->ip_off |= IP_DF; 1222 } 1223 1224 th = (struct tcphdr *)(ip + 1); 1225 } 1226 th->th_sport = sc->sc_inc.inc_lport; 1227 th->th_dport = sc->sc_inc.inc_fport; 1228 1229 th->th_seq = htonl(sc->sc_iss); 1230 th->th_ack = htonl(sc->sc_irs + 1); 1231 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1232 th->th_x2 = 0; 1233 th->th_flags = TH_SYN | TH_ACK; 1234 th->th_win = htons(sc->sc_wnd); 1235 th->th_urp = 0; 1236 1237 /* Tack on the TCP options. */ 1238 if (optlen == 0) 1239 goto no_options; 1240 optp = (u_int8_t *)(th + 1); 1241 *optp++ = TCPOPT_MAXSEG; 1242 *optp++ = TCPOLEN_MAXSEG; 1243 *optp++ = (mssopt >> 8) & 0xff; 1244 *optp++ = mssopt & 0xff; 1245 1246 if (sc->sc_flags & SCF_WINSCALE) { 1247 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1248 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1249 sc->sc_request_r_scale); 1250 optp += 4; 1251 } 1252 1253 if (sc->sc_flags & SCF_TIMESTAMP) { 1254 u_int32_t *lp = (u_int32_t *)(optp); 1255 1256 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1257 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1258 *lp++ = htonl(ticks); 1259 *lp = htonl(sc->sc_tsrecent); 1260 optp += TCPOLEN_TSTAMP_APPA; 1261 } 1262 1263 #ifdef TCP_SIGNATURE 1264 /* 1265 * Handle TCP-MD5 passive opener response. 1266 */ 1267 if (sc->sc_flags & SCF_SIGNATURE) { 1268 u_int8_t *bp = optp; 1269 int i; 1270 1271 *bp++ = TCPOPT_SIGNATURE; 1272 *bp++ = TCPOLEN_SIGNATURE; 1273 for (i = 0; i < TCP_SIGLEN; i++) 1274 *bp++ = 0; 1275 tcpsignature_compute(m, 0, optlen, 1276 optp + 2, IPSEC_DIR_OUTBOUND); 1277 *bp++ = TCPOPT_NOP; 1278 *bp++ = TCPOPT_EOL; 1279 optp += TCPOLEN_SIGNATURE + 2; 1280 } 1281 #endif /* TCP_SIGNATURE */ 1282 1283 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1284 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1285 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1286 } 1287 1288 no_options: 1289 if (isipv6) { 1290 struct route_in6 *ro6 = &sc->sc_route6; 1291 1292 th->th_sum = 0; 1293 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1294 ip6->ip6_hlim = in6_selecthlim(NULL, 1295 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1296 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1297 sc->sc_tp->t_inpcb); 1298 } else { 1299 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1300 htons(tlen - hlen + IPPROTO_TCP)); 1301 m->m_pkthdr.csum_flags = CSUM_TCP; 1302 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1303 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 1304 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb); 1305 } 1306 return (error); 1307 } 1308 1309 /* 1310 * cookie layers: 1311 * 1312 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1313 * | peer iss | 1314 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1315 * | 0 |(A)| | 1316 * (A): peer mss index 1317 */ 1318 1319 /* 1320 * The values below are chosen to minimize the size of the tcp_secret 1321 * table, as well as providing roughly a 16 second lifetime for the cookie. 1322 */ 1323 1324 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1325 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1326 1327 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1328 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1329 #define SYNCOOKIE_TIMEOUT \ 1330 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1331 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1332 1333 static struct { 1334 u_int32_t ts_secbits[4]; 1335 u_int ts_expire; 1336 } tcp_secret[SYNCOOKIE_NSECRETS]; 1337 1338 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1339 1340 static MD5_CTX syn_ctx; 1341 1342 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1343 1344 struct md5_add { 1345 u_int32_t laddr, faddr; 1346 u_int32_t secbits[4]; 1347 u_int16_t lport, fport; 1348 }; 1349 1350 #ifdef CTASSERT 1351 CTASSERT(sizeof(struct md5_add) == 28); 1352 #endif 1353 1354 /* 1355 * Consider the problem of a recreated (and retransmitted) cookie. If the 1356 * original SYN was accepted, the connection is established. The second 1357 * SYN is inflight, and if it arrives with an ISN that falls within the 1358 * receive window, the connection is killed. 1359 * 1360 * However, since cookies have other problems, this may not be worth 1361 * worrying about. 1362 */ 1363 1364 static u_int32_t 1365 syncookie_generate(struct syncache *sc) 1366 { 1367 u_int32_t md5_buffer[4]; 1368 u_int32_t data; 1369 int idx, i; 1370 struct md5_add add; 1371 #ifdef INET6 1372 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1373 #else 1374 const boolean_t isipv6 = FALSE; 1375 #endif 1376 1377 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1378 if (tcp_secret[idx].ts_expire < ticks) { 1379 for (i = 0; i < 4; i++) 1380 tcp_secret[idx].ts_secbits[i] = karc4random(); 1381 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1382 } 1383 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1384 if (tcp_msstab[data] <= sc->sc_peer_mss) 1385 break; 1386 data = (data << SYNCOOKIE_WNDBITS) | idx; 1387 data ^= sc->sc_irs; /* peer's iss */ 1388 MD5Init(&syn_ctx); 1389 if (isipv6) { 1390 MD5Add(sc->sc_inc.inc6_laddr); 1391 MD5Add(sc->sc_inc.inc6_faddr); 1392 add.laddr = 0; 1393 add.faddr = 0; 1394 } else { 1395 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1396 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1397 } 1398 add.lport = sc->sc_inc.inc_lport; 1399 add.fport = sc->sc_inc.inc_fport; 1400 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1401 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1402 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1403 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1404 MD5Add(add); 1405 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1406 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1407 return (data); 1408 } 1409 1410 static struct syncache * 1411 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so) 1412 { 1413 u_int32_t md5_buffer[4]; 1414 struct syncache *sc; 1415 u_int32_t data; 1416 int wnd, idx; 1417 struct md5_add add; 1418 1419 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1420 idx = data & SYNCOOKIE_WNDMASK; 1421 if (tcp_secret[idx].ts_expire < ticks || 1422 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1423 return (NULL); 1424 MD5Init(&syn_ctx); 1425 #ifdef INET6 1426 if (inc->inc_isipv6) { 1427 MD5Add(inc->inc6_laddr); 1428 MD5Add(inc->inc6_faddr); 1429 add.laddr = 0; 1430 add.faddr = 0; 1431 } else 1432 #endif 1433 { 1434 add.laddr = inc->inc_laddr.s_addr; 1435 add.faddr = inc->inc_faddr.s_addr; 1436 } 1437 add.lport = inc->inc_lport; 1438 add.fport = inc->inc_fport; 1439 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1440 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1441 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1442 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1443 MD5Add(add); 1444 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1445 data ^= md5_buffer[0]; 1446 if (data & ~SYNCOOKIE_DATAMASK) 1447 return (NULL); 1448 data = data >> SYNCOOKIE_WNDBITS; 1449 1450 /* 1451 * Fill in the syncache values. 1452 * XXX duplicate code from syncache_add 1453 */ 1454 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1455 sc->sc_ipopts = NULL; 1456 sc->sc_inc.inc_fport = inc->inc_fport; 1457 sc->sc_inc.inc_lport = inc->inc_lport; 1458 #ifdef INET6 1459 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1460 if (inc->inc_isipv6) { 1461 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1462 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1463 sc->sc_route6.ro_rt = NULL; 1464 } else 1465 #endif 1466 { 1467 sc->sc_inc.inc_faddr = inc->inc_faddr; 1468 sc->sc_inc.inc_laddr = inc->inc_laddr; 1469 sc->sc_route.ro_rt = NULL; 1470 } 1471 sc->sc_irs = th->th_seq - 1; 1472 sc->sc_iss = th->th_ack - 1; 1473 wnd = ssb_space(&so->so_rcv); 1474 wnd = imax(wnd, 0); 1475 wnd = imin(wnd, TCP_MAXWIN); 1476 sc->sc_wnd = wnd; 1477 sc->sc_flags = 0; 1478 sc->sc_rxtslot = 0; 1479 sc->sc_peer_mss = tcp_msstab[data]; 1480 return (sc); 1481 } 1482