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 */ 73 74 #include "opt_inet.h" 75 #include "opt_inet6.h" 76 #include "opt_ipsec.h" 77 78 #include <sys/param.h> 79 #include <sys/systm.h> 80 #include <sys/kernel.h> 81 #include <sys/sysctl.h> 82 #include <sys/malloc.h> 83 #include <sys/mbuf.h> 84 #include <sys/md5.h> 85 #include <sys/proc.h> /* for proc0 declaration */ 86 #include <sys/random.h> 87 #include <sys/socket.h> 88 #include <sys/socketvar.h> 89 #include <sys/in_cksum.h> 90 91 #include <sys/msgport2.h> 92 #include <net/netmsg2.h> 93 94 #include <net/if.h> 95 #include <net/route.h> 96 97 #include <netinet/in.h> 98 #include <netinet/in_systm.h> 99 #include <netinet/ip.h> 100 #include <netinet/in_var.h> 101 #include <netinet/in_pcb.h> 102 #include <netinet/ip_var.h> 103 #include <netinet/ip6.h> 104 #ifdef INET6 105 #include <netinet/icmp6.h> 106 #include <netinet6/nd6.h> 107 #endif 108 #include <netinet6/ip6_var.h> 109 #include <netinet6/in6_pcb.h> 110 #include <netinet/tcp.h> 111 #include <netinet/tcp_fsm.h> 112 #include <netinet/tcp_seq.h> 113 #include <netinet/tcp_timer.h> 114 #include <netinet/tcp_timer2.h> 115 #include <netinet/tcp_var.h> 116 #include <netinet6/tcp6_var.h> 117 118 #ifdef IPSEC 119 #include <netinet6/ipsec.h> 120 #ifdef INET6 121 #include <netinet6/ipsec6.h> 122 #endif 123 #include <netproto/key/key.h> 124 #endif /*IPSEC*/ 125 126 #ifdef FAST_IPSEC 127 #include <netproto/ipsec/ipsec.h> 128 #ifdef INET6 129 #include <netproto/ipsec/ipsec6.h> 130 #endif 131 #include <netproto/ipsec/key.h> 132 #define IPSEC 133 #endif /*FAST_IPSEC*/ 134 135 static int tcp_syncookies = 1; 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 137 &tcp_syncookies, 0, 138 "Use TCP SYN cookies if the syncache overflows"); 139 140 static void syncache_drop(struct syncache *, struct syncache_head *); 141 static void syncache_free(struct syncache *); 142 static void syncache_insert(struct syncache *, struct syncache_head *); 143 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 144 static int syncache_respond(struct syncache *, struct mbuf *); 145 static struct socket *syncache_socket(struct syncache *, struct socket *, 146 struct mbuf *); 147 static void syncache_timer(void *); 148 static u_int32_t syncookie_generate(struct syncache *); 149 static struct syncache *syncookie_lookup(struct in_conninfo *, 150 struct tcphdr *, struct socket *); 151 152 /* 153 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 154 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 155 * the odds are that the user has given up attempting to connect by then. 156 */ 157 #define SYNCACHE_MAXREXMTS 3 158 159 /* Arbitrary values */ 160 #define TCP_SYNCACHE_HASHSIZE 512 161 #define TCP_SYNCACHE_BUCKETLIMIT 30 162 163 struct netmsg_sc_timer { 164 struct netmsg_base base; 165 struct msgrec *nm_mrec; /* back pointer to containing msgrec */ 166 }; 167 168 struct msgrec { 169 struct netmsg_sc_timer msg; 170 lwkt_port_t port; /* constant after init */ 171 int slot; /* constant after init */ 172 }; 173 174 static void syncache_timer_handler(netmsg_t); 175 176 struct tcp_syncache { 177 u_int hashsize; 178 u_int hashmask; 179 u_int bucket_limit; 180 u_int cache_limit; 181 u_int rexmt_limit; 182 u_int hash_secret; 183 }; 184 static struct tcp_syncache tcp_syncache; 185 186 TAILQ_HEAD(syncache_list, syncache); 187 188 struct tcp_syncache_percpu { 189 struct syncache_head *hashbase; 190 u_int cache_count; 191 struct syncache_list timerq[SYNCACHE_MAXREXMTS + 1]; 192 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1]; 193 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1]; 194 }; 195 static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU]; 196 197 static struct lwkt_port syncache_null_rport; 198 199 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 200 201 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD, 202 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 203 204 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD, 205 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 206 207 /* XXX JH */ 208 #if 0 209 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 210 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 211 #endif 212 213 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD, 214 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 215 216 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 217 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 218 219 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 220 221 #define SYNCACHE_HASH(inc, mask) \ 222 ((tcp_syncache.hash_secret ^ \ 223 (inc)->inc_faddr.s_addr ^ \ 224 ((inc)->inc_faddr.s_addr >> 16) ^ \ 225 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 226 227 #define SYNCACHE_HASH6(inc, mask) \ 228 ((tcp_syncache.hash_secret ^ \ 229 (inc)->inc6_faddr.s6_addr32[0] ^ \ 230 (inc)->inc6_faddr.s6_addr32[3] ^ \ 231 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 232 233 #define ENDPTS_EQ(a, b) ( \ 234 (a)->ie_fport == (b)->ie_fport && \ 235 (a)->ie_lport == (b)->ie_lport && \ 236 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 237 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 238 ) 239 240 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 241 242 static __inline void 243 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu, 244 struct syncache *sc, int slot) 245 { 246 sc->sc_rxtslot = slot; 247 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[slot]; 248 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq); 249 if (!callout_active(&syncache_percpu->tt_timerq[slot])) { 250 callout_reset(&syncache_percpu->tt_timerq[slot], 251 TCPTV_RTOBASE * tcp_backoff[slot], 252 syncache_timer, 253 &syncache_percpu->mrec[slot]); 254 } 255 } 256 257 static void 258 syncache_free(struct syncache *sc) 259 { 260 struct rtentry *rt; 261 #ifdef INET6 262 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 263 #else 264 const boolean_t isipv6 = FALSE; 265 #endif 266 267 if (sc->sc_ipopts) 268 m_free(sc->sc_ipopts); 269 270 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt; 271 if (rt != NULL) { 272 /* 273 * If this is the only reference to a protocol-cloned 274 * route, remove it immediately. 275 */ 276 if ((rt->rt_flags & RTF_WASCLONED) && rt->rt_refcnt == 1) 277 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 278 rt_mask(rt), rt->rt_flags, NULL); 279 RTFREE(rt); 280 } 281 kfree(sc, M_SYNCACHE); 282 } 283 284 void 285 syncache_init(void) 286 { 287 int i, cpu; 288 289 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 290 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 291 tcp_syncache.cache_limit = 292 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 293 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 294 tcp_syncache.hash_secret = karc4random(); 295 296 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 297 &tcp_syncache.hashsize); 298 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 299 &tcp_syncache.cache_limit); 300 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 301 &tcp_syncache.bucket_limit); 302 if (!powerof2(tcp_syncache.hashsize)) { 303 kprintf("WARNING: syncache hash size is not a power of 2.\n"); 304 tcp_syncache.hashsize = 512; /* safe default */ 305 } 306 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 307 308 lwkt_initport_replyonly_null(&syncache_null_rport); 309 310 for (cpu = 0; cpu < ncpus2; cpu++) { 311 struct tcp_syncache_percpu *syncache_percpu; 312 313 syncache_percpu = &tcp_syncache_percpu[cpu]; 314 /* Allocate the hash table. */ 315 syncache_percpu->hashbase = kmalloc(tcp_syncache.hashsize * sizeof(struct syncache_head), 316 M_SYNCACHE, M_WAITOK); 317 318 /* Initialize the hash buckets. */ 319 for (i = 0; i < tcp_syncache.hashsize; i++) { 320 struct syncache_head *bucket; 321 322 bucket = &syncache_percpu->hashbase[i]; 323 TAILQ_INIT(&bucket->sch_bucket); 324 bucket->sch_length = 0; 325 } 326 327 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 328 /* Initialize the timer queues. */ 329 TAILQ_INIT(&syncache_percpu->timerq[i]); 330 callout_init_mp(&syncache_percpu->tt_timerq[i]); 331 332 syncache_percpu->mrec[i].slot = i; 333 syncache_percpu->mrec[i].port = cpu_portfn(cpu); 334 syncache_percpu->mrec[i].msg.nm_mrec = 335 &syncache_percpu->mrec[i]; 336 netmsg_init(&syncache_percpu->mrec[i].msg.base, 337 NULL, &syncache_null_rport, 338 0, syncache_timer_handler); 339 } 340 } 341 } 342 343 static void 344 syncache_insert(struct syncache *sc, struct syncache_head *sch) 345 { 346 struct tcp_syncache_percpu *syncache_percpu; 347 struct syncache *sc2; 348 int i; 349 350 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 351 352 /* 353 * Make sure that we don't overflow the per-bucket 354 * limit or the total cache size limit. 355 */ 356 if (sch->sch_length >= tcp_syncache.bucket_limit) { 357 /* 358 * The bucket is full, toss the oldest element. 359 */ 360 sc2 = TAILQ_FIRST(&sch->sch_bucket); 361 sc2->sc_tp->ts_recent = ticks; 362 syncache_drop(sc2, sch); 363 tcpstat.tcps_sc_bucketoverflow++; 364 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) { 365 /* 366 * The cache is full. Toss the oldest entry in the 367 * entire cache. This is the front entry in the 368 * first non-empty timer queue with the largest 369 * timeout value. 370 */ 371 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 372 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]); 373 while (sc2 && (sc2->sc_flags & SCF_MARKER)) 374 sc2 = TAILQ_NEXT(sc2, sc_timerq); 375 if (sc2 != NULL) 376 break; 377 } 378 sc2->sc_tp->ts_recent = ticks; 379 syncache_drop(sc2, NULL); 380 tcpstat.tcps_sc_cacheoverflow++; 381 } 382 383 /* Initialize the entry's timer. */ 384 syncache_timeout(syncache_percpu, sc, 0); 385 386 /* Put it into the bucket. */ 387 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 388 sch->sch_length++; 389 syncache_percpu->cache_count++; 390 tcpstat.tcps_sc_added++; 391 } 392 393 void 394 syncache_destroy(struct tcpcb *tp) 395 { 396 struct tcp_syncache_percpu *syncache_percpu; 397 struct syncache_head *bucket; 398 struct syncache *sc; 399 int i; 400 401 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 402 sc = NULL; 403 404 for (i = 0; i < tcp_syncache.hashsize; i++) { 405 bucket = &syncache_percpu->hashbase[i]; 406 TAILQ_FOREACH(sc, &bucket->sch_bucket, sc_hash) { 407 if (sc->sc_tp == tp) 408 sc->sc_tp = NULL; 409 } 410 } 411 } 412 413 static void 414 syncache_drop(struct syncache *sc, struct syncache_head *sch) 415 { 416 struct tcp_syncache_percpu *syncache_percpu; 417 #ifdef INET6 418 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 419 #else 420 const boolean_t isipv6 = FALSE; 421 #endif 422 423 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 424 425 if (sch == NULL) { 426 if (isipv6) { 427 sch = &syncache_percpu->hashbase[ 428 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 429 } else { 430 sch = &syncache_percpu->hashbase[ 431 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 432 } 433 } 434 435 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 436 sch->sch_length--; 437 syncache_percpu->cache_count--; 438 439 /* 440 * Cleanup 441 */ 442 if (sc->sc_tp) 443 sc->sc_tp = NULL; 444 445 /* 446 * Remove the entry from the syncache timer/timeout queue. Note 447 * that we do not try to stop any running timer since we do not know 448 * whether the timer's message is in-transit or not. Since timeouts 449 * are fairly long, taking an unneeded callout does not detrimentally 450 * effect performance. 451 */ 452 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq); 453 454 syncache_free(sc); 455 } 456 457 /* 458 * Place a timeout message on the TCP thread's message queue. 459 * This routine runs in soft interrupt context. 460 * 461 * An invariant is for this routine to be called, the callout must 462 * have been active. Note that the callout is not deactivated until 463 * after the message has been processed in syncache_timer_handler() below. 464 */ 465 static void 466 syncache_timer(void *p) 467 { 468 struct netmsg_sc_timer *msg = p; 469 470 lwkt_sendmsg(msg->nm_mrec->port, &msg->base.lmsg); 471 } 472 473 /* 474 * Service a timer message queued by timer expiration. 475 * This routine runs in the TCP protocol thread. 476 * 477 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 478 * If we have retransmitted an entry the maximum number of times, expire it. 479 * 480 * When we finish processing timed-out entries, we restart the timer if there 481 * are any entries still on the queue and deactivate it otherwise. Only after 482 * a timer has been deactivated here can it be restarted by syncache_timeout(). 483 */ 484 static void 485 syncache_timer_handler(netmsg_t msg) 486 { 487 struct tcp_syncache_percpu *syncache_percpu; 488 struct syncache *sc; 489 struct syncache marker; 490 struct syncache_list *list; 491 struct inpcb *inp; 492 int slot; 493 494 slot = ((struct netmsg_sc_timer *)msg)->nm_mrec->slot; 495 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 496 497 list = &syncache_percpu->timerq[slot]; 498 499 /* 500 * Use a marker to keep our place in the scan. syncache_drop() 501 * can block and cause any next pointer we cache to become stale. 502 */ 503 marker.sc_flags = SCF_MARKER; 504 TAILQ_INSERT_HEAD(list, &marker, sc_timerq); 505 506 while ((sc = TAILQ_NEXT(&marker, sc_timerq)) != NULL) { 507 /* 508 * Move the marker. 509 */ 510 TAILQ_REMOVE(list, &marker, sc_timerq); 511 TAILQ_INSERT_AFTER(list, sc, &marker, sc_timerq); 512 513 if (sc->sc_flags & SCF_MARKER) 514 continue; 515 516 if (ticks < sc->sc_rxttime) 517 break; /* finished because timerq sorted by time */ 518 if (sc->sc_tp == NULL) { 519 syncache_drop(sc, NULL); 520 tcpstat.tcps_sc_stale++; 521 continue; 522 } 523 inp = sc->sc_tp->t_inpcb; 524 if (slot == SYNCACHE_MAXREXMTS || 525 slot >= tcp_syncache.rexmt_limit || 526 inp == NULL || 527 inp->inp_gencnt != sc->sc_inp_gencnt) { 528 syncache_drop(sc, NULL); 529 tcpstat.tcps_sc_stale++; 530 continue; 531 } 532 /* 533 * syncache_respond() may call back into the syncache to 534 * to modify another entry, so do not obtain the next 535 * entry on the timer chain until it has completed. 536 */ 537 syncache_respond(sc, NULL); 538 tcpstat.tcps_sc_retransmitted++; 539 TAILQ_REMOVE(list, sc, sc_timerq); 540 syncache_timeout(syncache_percpu, sc, slot + 1); 541 } 542 TAILQ_REMOVE(list, &marker, sc_timerq); 543 544 if (sc != NULL) { 545 callout_reset(&syncache_percpu->tt_timerq[slot], 546 sc->sc_rxttime - ticks, syncache_timer, 547 &syncache_percpu->mrec[slot]); 548 } else { 549 callout_deactivate(&syncache_percpu->tt_timerq[slot]); 550 } 551 lwkt_replymsg(&msg->base.lmsg, 0); 552 } 553 554 /* 555 * Find an entry in the syncache. 556 */ 557 struct syncache * 558 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 559 { 560 struct tcp_syncache_percpu *syncache_percpu; 561 struct syncache *sc; 562 struct syncache_head *sch; 563 564 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 565 #ifdef INET6 566 if (inc->inc_isipv6) { 567 sch = &syncache_percpu->hashbase[ 568 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 569 *schp = sch; 570 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) 571 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 572 return (sc); 573 } else 574 #endif 575 { 576 sch = &syncache_percpu->hashbase[ 577 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 578 *schp = sch; 579 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 580 #ifdef INET6 581 if (sc->sc_inc.inc_isipv6) 582 continue; 583 #endif 584 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 585 return (sc); 586 } 587 } 588 return (NULL); 589 } 590 591 /* 592 * This function is called when we get a RST for a 593 * non-existent connection, so that we can see if the 594 * connection is in the syn cache. If it is, zap it. 595 */ 596 void 597 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 598 { 599 struct syncache *sc; 600 struct syncache_head *sch; 601 602 sc = syncache_lookup(inc, &sch); 603 if (sc == NULL) { 604 return; 605 } 606 /* 607 * If the RST bit is set, check the sequence number to see 608 * if this is a valid reset segment. 609 * RFC 793 page 37: 610 * In all states except SYN-SENT, all reset (RST) segments 611 * are validated by checking their SEQ-fields. A reset is 612 * valid if its sequence number is in the window. 613 * 614 * The sequence number in the reset segment is normally an 615 * echo of our outgoing acknowlegement numbers, but some hosts 616 * send a reset with the sequence number at the rightmost edge 617 * of our receive window, and we have to handle this case. 618 */ 619 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 620 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 621 syncache_drop(sc, sch); 622 tcpstat.tcps_sc_reset++; 623 } 624 } 625 626 void 627 syncache_badack(struct in_conninfo *inc) 628 { 629 struct syncache *sc; 630 struct syncache_head *sch; 631 632 sc = syncache_lookup(inc, &sch); 633 if (sc != NULL) { 634 syncache_drop(sc, sch); 635 tcpstat.tcps_sc_badack++; 636 } 637 } 638 639 void 640 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 641 { 642 struct syncache *sc; 643 struct syncache_head *sch; 644 645 /* we are called at splnet() here */ 646 sc = syncache_lookup(inc, &sch); 647 if (sc == NULL) 648 return; 649 650 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 651 if (ntohl(th->th_seq) != sc->sc_iss) 652 return; 653 654 /* 655 * If we've rertransmitted 3 times and this is our second error, 656 * we remove the entry. Otherwise, we allow it to continue on. 657 * This prevents us from incorrectly nuking an entry during a 658 * spurious network outage. 659 * 660 * See tcp_notify(). 661 */ 662 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 663 sc->sc_flags |= SCF_UNREACH; 664 return; 665 } 666 syncache_drop(sc, sch); 667 tcpstat.tcps_sc_unreach++; 668 } 669 670 /* 671 * Build a new TCP socket structure from a syncache entry. 672 * 673 * This is called from the context of the SYN+ACK 674 */ 675 static struct socket * 676 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 677 { 678 struct inpcb *inp = NULL, *linp; 679 struct socket *so; 680 struct tcpcb *tp, *ltp; 681 lwkt_port_t port; 682 #ifdef INET6 683 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 684 #else 685 const boolean_t isipv6 = FALSE; 686 #endif 687 struct sockaddr_in sin_faddr; 688 struct sockaddr_in6 sin6_faddr; 689 struct sockaddr *faddr; 690 691 if (isipv6) { 692 faddr = (struct sockaddr *)&sin6_faddr; 693 sin6_faddr.sin6_family = AF_INET6; 694 sin6_faddr.sin6_len = sizeof(sin6_faddr); 695 sin6_faddr.sin6_addr = sc->sc_inc.inc6_faddr; 696 sin6_faddr.sin6_port = sc->sc_inc.inc_fport; 697 sin6_faddr.sin6_flowinfo = sin6_faddr.sin6_scope_id = 0; 698 } else { 699 faddr = (struct sockaddr *)&sin_faddr; 700 sin_faddr.sin_family = AF_INET; 701 sin_faddr.sin_len = sizeof(sin_faddr); 702 sin_faddr.sin_addr = sc->sc_inc.inc_faddr; 703 sin_faddr.sin_port = sc->sc_inc.inc_fport; 704 bzero(sin_faddr.sin_zero, sizeof(sin_faddr.sin_zero)); 705 } 706 707 /* 708 * Ok, create the full blown connection, and set things up 709 * as they would have been set up if we had created the 710 * connection when the SYN arrived. If we can't create 711 * the connection, abort it. 712 * 713 * Set the protocol processing port for the socket to the current 714 * port (that the connection came in on). 715 */ 716 so = sonewconn_faddr(lso, SS_ISCONNECTED, faddr); 717 if (so == NULL) { 718 /* 719 * Drop the connection; we will send a RST if the peer 720 * retransmits the ACK, 721 */ 722 tcpstat.tcps_listendrop++; 723 goto abort; 724 } 725 726 /* 727 * Insert new socket into hash list. 728 */ 729 inp = so->so_pcb; 730 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 731 if (isipv6) { 732 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 733 } else { 734 #ifdef INET6 735 inp->inp_vflag &= ~INP_IPV6; 736 inp->inp_vflag |= INP_IPV4; 737 inp->inp_flags &= ~IN6P_IPV6_V6ONLY; 738 #endif 739 inp->inp_laddr = sc->sc_inc.inc_laddr; 740 } 741 inp->inp_lport = sc->sc_inc.inc_lport; 742 if (in_pcbinsporthash(inp) != 0) { 743 /* 744 * Undo the assignments above if we failed to 745 * put the PCB on the hash lists. 746 */ 747 if (isipv6) 748 inp->in6p_laddr = kin6addr_any; 749 else 750 inp->inp_laddr.s_addr = INADDR_ANY; 751 inp->inp_lport = 0; 752 goto abort; 753 } 754 linp = lso->so_pcb; 755 #ifdef IPSEC 756 /* copy old policy into new socket's */ 757 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 758 kprintf("syncache_expand: could not copy policy\n"); 759 #endif 760 if (isipv6) { 761 struct in6_addr laddr6; 762 /* 763 * Inherit socket options from the listening socket. 764 * Note that in6p_inputopts are not (and should not be) 765 * copied, since it stores previously received options and is 766 * used to detect if each new option is different than the 767 * previous one and hence should be passed to a user. 768 * If we copied in6p_inputopts, a user would not be able to 769 * receive options just after calling the accept system call. 770 */ 771 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 772 if (linp->in6p_outputopts) 773 inp->in6p_outputopts = 774 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 775 inp->in6p_route = sc->sc_route6; 776 sc->sc_route6.ro_rt = NULL; 777 778 laddr6 = inp->in6p_laddr; 779 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 780 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 781 if (in6_pcbconnect(inp, faddr, &thread0)) { 782 inp->in6p_laddr = laddr6; 783 goto abort; 784 } 785 } else { 786 struct in_addr laddr; 787 788 inp->inp_options = ip_srcroute(m); 789 if (inp->inp_options == NULL) { 790 inp->inp_options = sc->sc_ipopts; 791 sc->sc_ipopts = NULL; 792 } 793 inp->inp_route = sc->sc_route; 794 sc->sc_route.ro_rt = NULL; 795 796 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, faddr, &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; 866 867 /* 868 * Inherit some properties from the listen socket 869 */ 870 ltp = intotcpcb(linp); 871 tp->t_keepinit = ltp->t_keepinit; 872 tp->t_keepidle = ltp->t_keepidle; 873 tp->t_keepintvl = ltp->t_keepintvl; 874 tp->t_keepcnt = ltp->t_keepcnt; 875 tp->t_maxidle = ltp->t_maxidle; 876 877 tcp_create_timermsg(tp, port); 878 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep); 879 880 tcpstat.tcps_accepts++; 881 return (so); 882 883 abort: 884 if (so != NULL) 885 soabort_oncpu(so); 886 return (NULL); 887 } 888 889 /* 890 * This function gets called when we receive an ACK for a 891 * socket in the LISTEN state. We look up the connection 892 * in the syncache, and if its there, we pull it out of 893 * the cache and turn it into a full-blown connection in 894 * the SYN-RECEIVED state. 895 */ 896 int 897 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop, 898 struct mbuf *m) 899 { 900 struct syncache *sc; 901 struct syncache_head *sch; 902 struct socket *so; 903 904 sc = syncache_lookup(inc, &sch); 905 if (sc == NULL) { 906 /* 907 * There is no syncache entry, so see if this ACK is 908 * a returning syncookie. To do this, first: 909 * A. See if this socket has had a syncache entry dropped in 910 * the past. We don't want to accept a bogus syncookie 911 * if we've never received a SYN. 912 * B. check that the syncookie is valid. If it is, then 913 * cobble up a fake syncache entry, and return. 914 */ 915 if (!tcp_syncookies) 916 return (0); 917 sc = syncookie_lookup(inc, th, *sop); 918 if (sc == NULL) 919 return (0); 920 sch = NULL; 921 tcpstat.tcps_sc_recvcookie++; 922 } 923 924 /* 925 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 926 */ 927 if (th->th_ack != sc->sc_iss + 1) 928 return (0); 929 930 so = syncache_socket(sc, *sop, m); 931 if (so == NULL) { 932 #if 0 933 resetandabort: 934 /* XXXjlemon check this - is this correct? */ 935 tcp_respond(NULL, m, m, th, 936 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 937 #endif 938 m_freem(m); /* XXX only needed for above */ 939 tcpstat.tcps_sc_aborted++; 940 } else { 941 tcpstat.tcps_sc_completed++; 942 } 943 if (sch == NULL) 944 syncache_free(sc); 945 else 946 syncache_drop(sc, sch); 947 *sop = so; 948 return (1); 949 } 950 951 /* 952 * Given a LISTEN socket and an inbound SYN request, add 953 * this to the syn cache, and send back a segment: 954 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 955 * to the source. 956 * 957 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 958 * Doing so would require that we hold onto the data and deliver it 959 * to the application. However, if we are the target of a SYN-flood 960 * DoS attack, an attacker could send data which would eventually 961 * consume all available buffer space if it were ACKed. By not ACKing 962 * the data, we avoid this DoS scenario. 963 */ 964 int 965 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 966 struct socket **sop, struct mbuf *m) 967 { 968 struct tcp_syncache_percpu *syncache_percpu; 969 struct tcpcb *tp; 970 struct socket *so; 971 struct syncache *sc = NULL; 972 struct syncache_head *sch; 973 struct mbuf *ipopts = NULL; 974 int win; 975 976 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 977 so = *sop; 978 tp = sototcpcb(so); 979 980 /* 981 * Remember the IP options, if any. 982 */ 983 #ifdef INET6 984 if (!inc->inc_isipv6) 985 #endif 986 ipopts = ip_srcroute(m); 987 988 /* 989 * See if we already have an entry for this connection. 990 * If we do, resend the SYN,ACK, and reset the retransmit timer. 991 * 992 * XXX 993 * The syncache should be re-initialized with the contents 994 * of the new SYN which may have different options. 995 */ 996 sc = syncache_lookup(inc, &sch); 997 if (sc != NULL) { 998 tcpstat.tcps_sc_dupsyn++; 999 if (ipopts) { 1000 /* 1001 * If we were remembering a previous source route, 1002 * forget it and use the new one we've been given. 1003 */ 1004 if (sc->sc_ipopts) 1005 m_free(sc->sc_ipopts); 1006 sc->sc_ipopts = ipopts; 1007 } 1008 /* 1009 * Update timestamp if present. 1010 */ 1011 if (sc->sc_flags & SCF_TIMESTAMP) 1012 sc->sc_tsrecent = to->to_tsval; 1013 1014 /* Just update the TOF_SACK_PERMITTED for now. */ 1015 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1016 sc->sc_flags |= SCF_SACK_PERMITTED; 1017 else 1018 sc->sc_flags &= ~SCF_SACK_PERMITTED; 1019 1020 /* 1021 * PCB may have changed, pick up new values. 1022 */ 1023 sc->sc_tp = tp; 1024 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1025 if (syncache_respond(sc, m) == 0) { 1026 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 1027 sc, sc_timerq); 1028 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 1029 tcpstat.tcps_sndacks++; 1030 tcpstat.tcps_sndtotal++; 1031 } 1032 *sop = NULL; 1033 return (1); 1034 } 1035 1036 /* 1037 * Fill in the syncache values. 1038 */ 1039 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1040 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1041 sc->sc_ipopts = ipopts; 1042 sc->sc_inc.inc_fport = inc->inc_fport; 1043 sc->sc_inc.inc_lport = inc->inc_lport; 1044 sc->sc_tp = tp; 1045 #ifdef INET6 1046 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1047 if (inc->inc_isipv6) { 1048 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1049 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1050 sc->sc_route6.ro_rt = NULL; 1051 } else 1052 #endif 1053 { 1054 sc->sc_inc.inc_faddr = inc->inc_faddr; 1055 sc->sc_inc.inc_laddr = inc->inc_laddr; 1056 sc->sc_route.ro_rt = NULL; 1057 } 1058 sc->sc_irs = th->th_seq; 1059 sc->sc_flags = 0; 1060 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 1061 if (tcp_syncookies) 1062 sc->sc_iss = syncookie_generate(sc); 1063 else 1064 sc->sc_iss = karc4random(); 1065 1066 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */ 1067 win = ssb_space(&so->so_rcv); 1068 win = imax(win, 0); 1069 win = imin(win, TCP_MAXWIN); 1070 sc->sc_wnd = win; 1071 1072 if (tcp_do_rfc1323) { 1073 /* 1074 * A timestamp received in a SYN makes 1075 * it ok to send timestamp requests and replies. 1076 */ 1077 if (to->to_flags & TOF_TS) { 1078 sc->sc_tsrecent = to->to_tsval; 1079 sc->sc_flags |= SCF_TIMESTAMP; 1080 } 1081 if (to->to_flags & TOF_SCALE) { 1082 int wscale = TCP_MIN_WINSHIFT; 1083 1084 /* Compute proper scaling value from buffer space */ 1085 while (wscale < TCP_MAX_WINSHIFT && 1086 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) { 1087 wscale++; 1088 } 1089 sc->sc_request_r_scale = wscale; 1090 sc->sc_requested_s_scale = to->to_requested_s_scale; 1091 sc->sc_flags |= SCF_WINSCALE; 1092 } 1093 } 1094 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1095 sc->sc_flags |= SCF_SACK_PERMITTED; 1096 if (tp->t_flags & TF_NOOPT) 1097 sc->sc_flags = SCF_NOOPT; 1098 #ifdef TCP_SIGNATURE 1099 /* 1100 * If listening socket requested TCP digests, and received SYN 1101 * contains the option, flag this in the syncache so that 1102 * syncache_respond() will do the right thing with the SYN+ACK. 1103 * XXX Currently we always record the option by default and will 1104 * attempt to use it in syncache_respond(). 1105 */ 1106 if (to->to_flags & TOF_SIGNATURE) 1107 sc->sc_flags = SCF_SIGNATURE; 1108 #endif /* TCP_SIGNATURE */ 1109 1110 if (syncache_respond(sc, m) == 0) { 1111 syncache_insert(sc, sch); 1112 tcpstat.tcps_sndacks++; 1113 tcpstat.tcps_sndtotal++; 1114 } else { 1115 syncache_free(sc); 1116 tcpstat.tcps_sc_dropped++; 1117 } 1118 *sop = NULL; 1119 return (1); 1120 } 1121 1122 static int 1123 syncache_respond(struct syncache *sc, struct mbuf *m) 1124 { 1125 u_int8_t *optp; 1126 int optlen, error; 1127 u_int16_t tlen, hlen, mssopt; 1128 struct ip *ip = NULL; 1129 struct rtentry *rt; 1130 struct tcphdr *th; 1131 struct ip6_hdr *ip6 = NULL; 1132 #ifdef INET6 1133 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1134 #else 1135 const boolean_t isipv6 = FALSE; 1136 #endif 1137 1138 if (isipv6) { 1139 rt = tcp_rtlookup6(&sc->sc_inc); 1140 if (rt != NULL) 1141 mssopt = rt->rt_ifp->if_mtu - 1142 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1143 else 1144 mssopt = tcp_v6mssdflt; 1145 hlen = sizeof(struct ip6_hdr); 1146 } else { 1147 rt = tcp_rtlookup(&sc->sc_inc); 1148 if (rt != NULL) 1149 mssopt = rt->rt_ifp->if_mtu - 1150 (sizeof(struct ip) + sizeof(struct tcphdr)); 1151 else 1152 mssopt = tcp_mssdflt; 1153 hlen = sizeof(struct ip); 1154 } 1155 1156 /* Compute the size of the TCP options. */ 1157 if (sc->sc_flags & SCF_NOOPT) { 1158 optlen = 0; 1159 } else { 1160 optlen = TCPOLEN_MAXSEG + 1161 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1162 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1163 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1164 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1165 #ifdef TCP_SIGNATURE 1166 optlen += ((sc->sc_flags & SCF_SIGNATURE) ? 1167 (TCPOLEN_SIGNATURE + 2) : 0); 1168 #endif /* TCP_SIGNATURE */ 1169 } 1170 tlen = hlen + sizeof(struct tcphdr) + optlen; 1171 1172 /* 1173 * XXX 1174 * assume that the entire packet will fit in a header mbuf 1175 */ 1176 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1177 1178 /* 1179 * XXX shouldn't this reuse the mbuf if possible ? 1180 * Create the IP+TCP header from scratch. 1181 */ 1182 if (m) 1183 m_freem(m); 1184 1185 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 1186 if (m == NULL) 1187 return (ENOBUFS); 1188 m->m_data += max_linkhdr; 1189 m->m_len = tlen; 1190 m->m_pkthdr.len = tlen; 1191 m->m_pkthdr.rcvif = NULL; 1192 1193 if (isipv6) { 1194 ip6 = mtod(m, struct ip6_hdr *); 1195 ip6->ip6_vfc = IPV6_VERSION; 1196 ip6->ip6_nxt = IPPROTO_TCP; 1197 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1198 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1199 ip6->ip6_plen = htons(tlen - hlen); 1200 /* ip6_hlim is set after checksum */ 1201 /* ip6_flow = ??? */ 1202 1203 th = (struct tcphdr *)(ip6 + 1); 1204 } else { 1205 ip = mtod(m, struct ip *); 1206 ip->ip_v = IPVERSION; 1207 ip->ip_hl = sizeof(struct ip) >> 2; 1208 ip->ip_len = tlen; 1209 ip->ip_id = 0; 1210 ip->ip_off = 0; 1211 ip->ip_sum = 0; 1212 ip->ip_p = IPPROTO_TCP; 1213 ip->ip_src = sc->sc_inc.inc_laddr; 1214 ip->ip_dst = sc->sc_inc.inc_faddr; 1215 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1216 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1217 1218 /* 1219 * See if we should do MTU discovery. Route lookups are 1220 * expensive, so we will only unset the DF bit if: 1221 * 1222 * 1) path_mtu_discovery is disabled 1223 * 2) the SCF_UNREACH flag has been set 1224 */ 1225 if (path_mtu_discovery 1226 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1227 ip->ip_off |= IP_DF; 1228 } 1229 1230 th = (struct tcphdr *)(ip + 1); 1231 } 1232 th->th_sport = sc->sc_inc.inc_lport; 1233 th->th_dport = sc->sc_inc.inc_fport; 1234 1235 th->th_seq = htonl(sc->sc_iss); 1236 th->th_ack = htonl(sc->sc_irs + 1); 1237 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1238 th->th_x2 = 0; 1239 th->th_flags = TH_SYN | TH_ACK; 1240 th->th_win = htons(sc->sc_wnd); 1241 th->th_urp = 0; 1242 1243 /* Tack on the TCP options. */ 1244 if (optlen == 0) 1245 goto no_options; 1246 optp = (u_int8_t *)(th + 1); 1247 *optp++ = TCPOPT_MAXSEG; 1248 *optp++ = TCPOLEN_MAXSEG; 1249 *optp++ = (mssopt >> 8) & 0xff; 1250 *optp++ = mssopt & 0xff; 1251 1252 if (sc->sc_flags & SCF_WINSCALE) { 1253 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1254 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1255 sc->sc_request_r_scale); 1256 optp += 4; 1257 } 1258 1259 if (sc->sc_flags & SCF_TIMESTAMP) { 1260 u_int32_t *lp = (u_int32_t *)(optp); 1261 1262 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1263 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1264 *lp++ = htonl(ticks); 1265 *lp = htonl(sc->sc_tsrecent); 1266 optp += TCPOLEN_TSTAMP_APPA; 1267 } 1268 1269 #ifdef TCP_SIGNATURE 1270 /* 1271 * Handle TCP-MD5 passive opener response. 1272 */ 1273 if (sc->sc_flags & SCF_SIGNATURE) { 1274 u_int8_t *bp = optp; 1275 int i; 1276 1277 *bp++ = TCPOPT_SIGNATURE; 1278 *bp++ = TCPOLEN_SIGNATURE; 1279 for (i = 0; i < TCP_SIGLEN; i++) 1280 *bp++ = 0; 1281 tcpsignature_compute(m, 0, optlen, 1282 optp + 2, IPSEC_DIR_OUTBOUND); 1283 *bp++ = TCPOPT_NOP; 1284 *bp++ = TCPOPT_EOL; 1285 optp += TCPOLEN_SIGNATURE + 2; 1286 } 1287 #endif /* TCP_SIGNATURE */ 1288 1289 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1290 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1291 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1292 } 1293 1294 no_options: 1295 if (isipv6) { 1296 struct route_in6 *ro6 = &sc->sc_route6; 1297 1298 th->th_sum = 0; 1299 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1300 ip6->ip6_hlim = in6_selecthlim(NULL, 1301 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1302 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1303 sc->sc_tp->t_inpcb); 1304 } else { 1305 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1306 htons(tlen - hlen + IPPROTO_TCP)); 1307 m->m_pkthdr.csum_flags = CSUM_TCP; 1308 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1309 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 1310 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb); 1311 } 1312 return (error); 1313 } 1314 1315 /* 1316 * cookie layers: 1317 * 1318 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1319 * | peer iss | 1320 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1321 * | 0 |(A)| | 1322 * (A): peer mss index 1323 */ 1324 1325 /* 1326 * The values below are chosen to minimize the size of the tcp_secret 1327 * table, as well as providing roughly a 16 second lifetime for the cookie. 1328 */ 1329 1330 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1331 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1332 1333 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1334 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1335 #define SYNCOOKIE_TIMEOUT \ 1336 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1337 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1338 1339 static struct { 1340 u_int32_t ts_secbits[4]; 1341 u_int ts_expire; 1342 } tcp_secret[SYNCOOKIE_NSECRETS]; 1343 1344 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1345 1346 static MD5_CTX syn_ctx; 1347 1348 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1349 1350 struct md5_add { 1351 u_int32_t laddr, faddr; 1352 u_int32_t secbits[4]; 1353 u_int16_t lport, fport; 1354 }; 1355 1356 #ifdef CTASSERT 1357 CTASSERT(sizeof(struct md5_add) == 28); 1358 #endif 1359 1360 /* 1361 * Consider the problem of a recreated (and retransmitted) cookie. If the 1362 * original SYN was accepted, the connection is established. The second 1363 * SYN is inflight, and if it arrives with an ISN that falls within the 1364 * receive window, the connection is killed. 1365 * 1366 * However, since cookies have other problems, this may not be worth 1367 * worrying about. 1368 */ 1369 1370 static u_int32_t 1371 syncookie_generate(struct syncache *sc) 1372 { 1373 u_int32_t md5_buffer[4]; 1374 u_int32_t data; 1375 int idx, i; 1376 struct md5_add add; 1377 #ifdef INET6 1378 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1379 #else 1380 const boolean_t isipv6 = FALSE; 1381 #endif 1382 1383 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1384 if (tcp_secret[idx].ts_expire < ticks) { 1385 for (i = 0; i < 4; i++) 1386 tcp_secret[idx].ts_secbits[i] = karc4random(); 1387 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1388 } 1389 for (data = NELEM(tcp_msstab) - 1; data > 0; data--) 1390 if (tcp_msstab[data] <= sc->sc_peer_mss) 1391 break; 1392 data = (data << SYNCOOKIE_WNDBITS) | idx; 1393 data ^= sc->sc_irs; /* peer's iss */ 1394 MD5Init(&syn_ctx); 1395 if (isipv6) { 1396 MD5Add(sc->sc_inc.inc6_laddr); 1397 MD5Add(sc->sc_inc.inc6_faddr); 1398 add.laddr = 0; 1399 add.faddr = 0; 1400 } else { 1401 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1402 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1403 } 1404 add.lport = sc->sc_inc.inc_lport; 1405 add.fport = sc->sc_inc.inc_fport; 1406 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1407 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1408 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1409 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1410 MD5Add(add); 1411 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1412 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1413 return (data); 1414 } 1415 1416 static struct syncache * 1417 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, 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 * Fill in the syncache values. 1458 * XXX duplicate code from syncache_add 1459 */ 1460 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1461 sc->sc_ipopts = NULL; 1462 sc->sc_inc.inc_fport = inc->inc_fport; 1463 sc->sc_inc.inc_lport = inc->inc_lport; 1464 #ifdef INET6 1465 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1466 if (inc->inc_isipv6) { 1467 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1468 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1469 sc->sc_route6.ro_rt = NULL; 1470 } else 1471 #endif 1472 { 1473 sc->sc_inc.inc_faddr = inc->inc_faddr; 1474 sc->sc_inc.inc_laddr = inc->inc_laddr; 1475 sc->sc_route.ro_rt = NULL; 1476 } 1477 sc->sc_irs = th->th_seq - 1; 1478 sc->sc_iss = th->th_ack - 1; 1479 wnd = ssb_space(&so->so_rcv); 1480 wnd = imax(wnd, 0); 1481 wnd = imin(wnd, TCP_MAXWIN); 1482 sc->sc_wnd = wnd; 1483 sc->sc_flags = 0; 1484 sc->sc_rxtslot = 0; 1485 sc->sc_peer_mss = tcp_msstab[data]; 1486 return (sc); 1487 } 1488