1 /* 2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved. 36 * 37 * License terms: all terms for the DragonFly license above plus the following: 38 * 39 * 4. All advertising materials mentioning features or use of this software 40 * must display the following acknowledgement: 41 * 42 * This product includes software developed by Jeffrey M. Hsu 43 * for the DragonFly Project. 44 * 45 * This requirement may be waived with permission from Jeffrey Hsu. 46 * This requirement will sunset and may be removed on July 8 2005, 47 * after which the standard DragonFly license (as shown above) will 48 * apply. 49 */ 50 51 /* 52 * All advertising materials mentioning features or use of this software 53 * must display the following acknowledgement: 54 * This product includes software developed by Jeffrey M. Hsu. 55 * 56 * Copyright (c) 2001 Networks Associates Technologies, Inc. 57 * All rights reserved. 58 * 59 * This software was developed for the FreeBSD Project by Jonathan Lemon 60 * and NAI Labs, the Security Research Division of Network Associates, Inc. 61 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 62 * DARPA CHATS research program. 63 * 64 * Redistribution and use in source and binary forms, with or without 65 * modification, are permitted provided that the following conditions 66 * are met: 67 * 1. Redistributions of source code must retain the above copyright 68 * notice, this list of conditions and the following disclaimer. 69 * 2. Redistributions in binary form must reproduce the above copyright 70 * notice, this list of conditions and the following disclaimer in the 71 * documentation and/or other materials provided with the distribution. 72 * 3. The name of the author may not be used to endorse or promote 73 * products derived from this software without specific prior written 74 * permission. 75 * 76 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 77 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 78 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 79 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 80 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 81 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 82 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 83 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 84 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 85 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 86 * SUCH DAMAGE. 87 * 88 * $FreeBSD: src/sys/netinet/tcp_syncache.c,v 1.5.2.14 2003/02/24 04:02:27 silby Exp $ 89 * $DragonFly: src/sys/netinet/tcp_syncache.c,v 1.21 2005/02/08 22:56:19 hsu Exp $ 90 */ 91 92 #include "opt_inet6.h" 93 #include "opt_ipsec.h" 94 95 #include <sys/param.h> 96 #include <sys/systm.h> 97 #include <sys/kernel.h> 98 #include <sys/sysctl.h> 99 #include <sys/malloc.h> 100 #include <sys/mbuf.h> 101 #include <sys/md5.h> 102 #include <sys/proc.h> /* for proc0 declaration */ 103 #include <sys/random.h> 104 #include <sys/socket.h> 105 #include <sys/socketvar.h> 106 #include <sys/in_cksum.h> 107 108 #include <sys/msgport2.h> 109 110 #include <net/if.h> 111 #include <net/route.h> 112 113 #include <netinet/in.h> 114 #include <netinet/in_systm.h> 115 #include <netinet/ip.h> 116 #include <netinet/in_var.h> 117 #include <netinet/in_pcb.h> 118 #include <netinet/ip_var.h> 119 #include <netinet/ip6.h> 120 #ifdef INET6 121 #include <netinet/icmp6.h> 122 #include <netinet6/nd6.h> 123 #endif 124 #include <netinet6/ip6_var.h> 125 #include <netinet6/in6_pcb.h> 126 #include <netinet/tcp.h> 127 #include <netinet/tcp_fsm.h> 128 #include <netinet/tcp_seq.h> 129 #include <netinet/tcp_timer.h> 130 #include <netinet/tcp_var.h> 131 #include <netinet6/tcp6_var.h> 132 133 #ifdef IPSEC 134 #include <netinet6/ipsec.h> 135 #ifdef INET6 136 #include <netinet6/ipsec6.h> 137 #endif 138 #include <netproto/key/key.h> 139 #endif /*IPSEC*/ 140 141 #ifdef FAST_IPSEC 142 #include <netproto/ipsec/ipsec.h> 143 #ifdef INET6 144 #include <netproto/ipsec/ipsec6.h> 145 #endif 146 #include <netproto/ipsec/key.h> 147 #define IPSEC 148 #endif /*FAST_IPSEC*/ 149 150 #include <vm/vm_zone.h> 151 152 static int tcp_syncookies = 1; 153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 154 &tcp_syncookies, 0, 155 "Use TCP SYN cookies if the syncache overflows"); 156 157 static void syncache_drop(struct syncache *, struct syncache_head *); 158 static void syncache_free(struct syncache *); 159 static void syncache_insert(struct syncache *, struct syncache_head *); 160 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 161 static int syncache_respond(struct syncache *, struct mbuf *); 162 static struct socket *syncache_socket(struct syncache *, struct socket *); 163 static void syncache_timer(void *); 164 static u_int32_t syncookie_generate(struct syncache *); 165 static struct syncache *syncookie_lookup(struct in_conninfo *, 166 struct tcphdr *, struct socket *); 167 168 /* 169 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 170 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 171 * the odds are that the user has given up attempting to connect by then. 172 */ 173 #define SYNCACHE_MAXREXMTS 3 174 175 /* Arbitrary values */ 176 #define TCP_SYNCACHE_HASHSIZE 512 177 #define TCP_SYNCACHE_BUCKETLIMIT 30 178 179 struct netmsg_sc_timer { 180 struct lwkt_msg nm_lmsg; 181 struct msgrec *nm_mrec; /* back pointer to containing msgrec */ 182 }; 183 184 struct msgrec { 185 struct netmsg_sc_timer msg; 186 lwkt_port_t port; /* constant after init */ 187 int slot; /* constant after init */ 188 }; 189 190 static int syncache_timer_handler(lwkt_msg_t); 191 192 struct tcp_syncache { 193 struct vm_zone *zone; 194 u_int hashsize; 195 u_int hashmask; 196 u_int bucket_limit; 197 u_int cache_limit; 198 u_int rexmt_limit; 199 u_int hash_secret; 200 }; 201 static struct tcp_syncache tcp_syncache; 202 203 struct tcp_syncache_percpu { 204 struct syncache_head *hashbase; 205 u_int cache_count; 206 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1]; 207 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1]; 208 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1]; 209 }; 210 static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU]; 211 212 static struct lwkt_port syncache_null_rport; 213 214 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 215 216 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD, 217 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 218 219 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD, 220 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 221 222 /* XXX JH */ 223 #if 0 224 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 225 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 226 #endif 227 228 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD, 229 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 230 231 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 232 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 233 234 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 235 236 #define SYNCACHE_HASH(inc, mask) \ 237 ((tcp_syncache.hash_secret ^ \ 238 (inc)->inc_faddr.s_addr ^ \ 239 ((inc)->inc_faddr.s_addr >> 16) ^ \ 240 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 241 242 #define SYNCACHE_HASH6(inc, mask) \ 243 ((tcp_syncache.hash_secret ^ \ 244 (inc)->inc6_faddr.s6_addr32[0] ^ \ 245 (inc)->inc6_faddr.s6_addr32[3] ^ \ 246 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 247 248 #define ENDPTS_EQ(a, b) ( \ 249 (a)->ie_fport == (b)->ie_fport && \ 250 (a)->ie_lport == (b)->ie_lport && \ 251 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 252 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 253 ) 254 255 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 256 257 static __inline void 258 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu, 259 struct syncache *sc, int slot) 260 { 261 sc->sc_rxtslot = slot; 262 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[slot]; 263 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq); 264 if (!callout_active(&syncache_percpu->tt_timerq[slot])) { 265 callout_reset(&syncache_percpu->tt_timerq[slot], 266 TCPTV_RTOBASE * tcp_backoff[slot], 267 syncache_timer, 268 &syncache_percpu->mrec[slot]); 269 } 270 } 271 272 static void 273 syncache_free(struct syncache *sc) 274 { 275 struct rtentry *rt; 276 #ifdef INET6 277 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 278 #else 279 const boolean_t isipv6 = FALSE; 280 #endif 281 282 if (sc->sc_ipopts) 283 m_free(sc->sc_ipopts); 284 if (isipv6) 285 rt = sc->sc_route6.ro_rt; 286 else 287 rt = sc->sc_route.ro_rt; 288 if (rt != NULL) { 289 /* 290 * If this is the only reference to a protocol cloned 291 * route, remove it immediately. 292 */ 293 if (rt->rt_flags & RTF_WASCLONED && 294 !(sc->sc_flags & SCF_KEEPROUTE) && 295 rt->rt_refcnt == 1) { 296 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 297 rt_mask(rt), rt->rt_flags, NULL); 298 } 299 RTFREE(rt); 300 } 301 zfree(tcp_syncache.zone, sc); 302 } 303 304 void 305 syncache_init(void) 306 { 307 int i, cpu; 308 309 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 310 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 311 tcp_syncache.cache_limit = 312 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 313 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 314 tcp_syncache.hash_secret = arc4random(); 315 316 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 317 &tcp_syncache.hashsize); 318 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 319 &tcp_syncache.cache_limit); 320 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 321 &tcp_syncache.bucket_limit); 322 if (!powerof2(tcp_syncache.hashsize)) { 323 printf("WARNING: syncache hash size is not a power of 2.\n"); 324 tcp_syncache.hashsize = 512; /* safe default */ 325 } 326 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 327 328 lwkt_initport_null_rport(&syncache_null_rport, NULL); 329 330 for (cpu = 0; cpu < ncpus2; cpu++) { 331 struct tcp_syncache_percpu *syncache_percpu; 332 333 syncache_percpu = &tcp_syncache_percpu[cpu]; 334 /* Allocate the hash table. */ 335 MALLOC(syncache_percpu->hashbase, struct syncache_head *, 336 tcp_syncache.hashsize * sizeof(struct syncache_head), 337 M_SYNCACHE, M_WAITOK); 338 339 /* Initialize the hash buckets. */ 340 for (i = 0; i < tcp_syncache.hashsize; i++) { 341 struct syncache_head *bucket; 342 343 bucket = &syncache_percpu->hashbase[i]; 344 TAILQ_INIT(&bucket->sch_bucket); 345 bucket->sch_length = 0; 346 } 347 348 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 349 /* Initialize the timer queues. */ 350 TAILQ_INIT(&syncache_percpu->timerq[i]); 351 callout_init(&syncache_percpu->tt_timerq[i]); 352 353 syncache_percpu->mrec[i].slot = i; 354 syncache_percpu->mrec[i].port = tcp_cport(cpu); 355 syncache_percpu->mrec[i].msg.nm_mrec = 356 &syncache_percpu->mrec[i]; 357 lwkt_initmsg(&syncache_percpu->mrec[i].msg.nm_lmsg, 358 &syncache_null_rport, 0, 359 lwkt_cmd_func(syncache_timer_handler), 360 lwkt_cmd_op_none); 361 } 362 } 363 364 /* 365 * Allocate the syncache entries. Allow the zone to allocate one 366 * more entry than cache limit, so a new entry can bump out an 367 * older one. 368 */ 369 tcp_syncache.zone = zinit("syncache", sizeof(struct syncache), 370 tcp_syncache.cache_limit, ZONE_INTERRUPT, 0); 371 tcp_syncache.cache_limit -= 1; 372 } 373 374 static void 375 syncache_insert(sc, sch) 376 struct syncache *sc; 377 struct syncache_head *sch; 378 { 379 struct tcp_syncache_percpu *syncache_percpu; 380 struct syncache *sc2; 381 int i; 382 383 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 384 385 /* 386 * Make sure that we don't overflow the per-bucket 387 * limit or the total cache size limit. 388 */ 389 if (sch->sch_length >= tcp_syncache.bucket_limit) { 390 /* 391 * The bucket is full, toss the oldest element. 392 */ 393 sc2 = TAILQ_FIRST(&sch->sch_bucket); 394 sc2->sc_tp->ts_recent = ticks; 395 syncache_drop(sc2, sch); 396 tcpstat.tcps_sc_bucketoverflow++; 397 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) { 398 /* 399 * The cache is full. Toss the oldest entry in the 400 * entire cache. This is the front entry in the 401 * first non-empty timer queue with the largest 402 * timeout value. 403 */ 404 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 405 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]); 406 if (sc2 != NULL) 407 break; 408 } 409 sc2->sc_tp->ts_recent = ticks; 410 syncache_drop(sc2, NULL); 411 tcpstat.tcps_sc_cacheoverflow++; 412 } 413 414 /* Initialize the entry's timer. */ 415 syncache_timeout(syncache_percpu, sc, 0); 416 417 /* Put it into the bucket. */ 418 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 419 sch->sch_length++; 420 syncache_percpu->cache_count++; 421 tcpstat.tcps_sc_added++; 422 } 423 424 static void 425 syncache_drop(sc, sch) 426 struct syncache *sc; 427 struct syncache_head *sch; 428 { 429 struct tcp_syncache_percpu *syncache_percpu; 430 #ifdef INET6 431 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 432 #else 433 const boolean_t isipv6 = FALSE; 434 #endif 435 436 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 437 438 if (sch == NULL) { 439 if (isipv6) { 440 sch = &syncache_percpu->hashbase[ 441 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 442 } else { 443 sch = &syncache_percpu->hashbase[ 444 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 445 } 446 } 447 448 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 449 sch->sch_length--; 450 syncache_percpu->cache_count--; 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->nm_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 int 492 syncache_timer_handler(lwkt_msg_t msg) 493 { 494 struct tcp_syncache_percpu *syncache_percpu; 495 struct syncache *sc, *nsc; 496 struct inpcb *inp; 497 int slot; 498 499 slot = ((struct netmsg_sc_timer *)msg)->nm_mrec->slot; 500 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 501 502 nsc = TAILQ_FIRST(&syncache_percpu->timerq[slot]); 503 while (nsc != NULL) { 504 if (ticks < nsc->sc_rxttime) 505 break; /* finished because timerq sorted by time */ 506 sc = nsc; 507 inp = sc->sc_tp->t_inpcb; 508 if (slot == SYNCACHE_MAXREXMTS || 509 slot >= tcp_syncache.rexmt_limit || 510 inp->inp_gencnt != sc->sc_inp_gencnt) { 511 nsc = TAILQ_NEXT(sc, sc_timerq); 512 syncache_drop(sc, NULL); 513 tcpstat.tcps_sc_stale++; 514 continue; 515 } 516 /* 517 * syncache_respond() may call back into the syncache to 518 * to modify another entry, so do not obtain the next 519 * entry on the timer chain until it has completed. 520 */ 521 syncache_respond(sc, NULL); 522 nsc = TAILQ_NEXT(sc, sc_timerq); 523 tcpstat.tcps_sc_retransmitted++; 524 TAILQ_REMOVE(&syncache_percpu->timerq[slot], sc, sc_timerq); 525 syncache_timeout(syncache_percpu, sc, slot + 1); 526 } 527 if (nsc != NULL) 528 callout_reset(&syncache_percpu->tt_timerq[slot], 529 nsc->sc_rxttime - ticks, syncache_timer, 530 &syncache_percpu->mrec[slot]); 531 else 532 callout_deactivate(&syncache_percpu->tt_timerq[slot]); 533 534 lwkt_replymsg(msg, 0); 535 return (EASYNC); 536 } 537 538 /* 539 * Find an entry in the syncache. 540 */ 541 struct syncache * 542 syncache_lookup(inc, schp) 543 struct in_conninfo *inc; 544 struct syncache_head **schp; 545 { 546 struct tcp_syncache_percpu *syncache_percpu; 547 struct syncache *sc; 548 struct syncache_head *sch; 549 550 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 551 #ifdef INET6 552 if (inc->inc_isipv6) { 553 sch = &syncache_percpu->hashbase[ 554 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 555 *schp = sch; 556 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) 557 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 558 return (sc); 559 } else 560 #endif 561 { 562 sch = &syncache_percpu->hashbase[ 563 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 564 *schp = sch; 565 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 566 #ifdef INET6 567 if (sc->sc_inc.inc_isipv6) 568 continue; 569 #endif 570 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 571 return (sc); 572 } 573 } 574 return (NULL); 575 } 576 577 /* 578 * This function is called when we get a RST for a 579 * non-existent connection, so that we can see if the 580 * connection is in the syn cache. If it is, zap it. 581 */ 582 void 583 syncache_chkrst(inc, th) 584 struct in_conninfo *inc; 585 struct tcphdr *th; 586 { 587 struct syncache *sc; 588 struct syncache_head *sch; 589 590 sc = syncache_lookup(inc, &sch); 591 if (sc == NULL) 592 return; 593 /* 594 * If the RST bit is set, check the sequence number to see 595 * if this is a valid reset segment. 596 * RFC 793 page 37: 597 * In all states except SYN-SENT, all reset (RST) segments 598 * are validated by checking their SEQ-fields. A reset is 599 * valid if its sequence number is in the window. 600 * 601 * The sequence number in the reset segment is normally an 602 * echo of our outgoing acknowlegement numbers, but some hosts 603 * send a reset with the sequence number at the rightmost edge 604 * of our receive window, and we have to handle this case. 605 */ 606 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 607 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 608 syncache_drop(sc, sch); 609 tcpstat.tcps_sc_reset++; 610 } 611 } 612 613 void 614 syncache_badack(inc) 615 struct in_conninfo *inc; 616 { 617 struct syncache *sc; 618 struct syncache_head *sch; 619 620 sc = syncache_lookup(inc, &sch); 621 if (sc != NULL) { 622 syncache_drop(sc, sch); 623 tcpstat.tcps_sc_badack++; 624 } 625 } 626 627 void 628 syncache_unreach(inc, th) 629 struct in_conninfo *inc; 630 struct tcphdr *th; 631 { 632 struct syncache *sc; 633 struct syncache_head *sch; 634 635 /* we are called at splnet() here */ 636 sc = syncache_lookup(inc, &sch); 637 if (sc == NULL) 638 return; 639 640 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 641 if (ntohl(th->th_seq) != sc->sc_iss) 642 return; 643 644 /* 645 * If we've rertransmitted 3 times and this is our second error, 646 * we remove the entry. Otherwise, we allow it to continue on. 647 * This prevents us from incorrectly nuking an entry during a 648 * spurious network outage. 649 * 650 * See tcp_notify(). 651 */ 652 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 653 sc->sc_flags |= SCF_UNREACH; 654 return; 655 } 656 syncache_drop(sc, sch); 657 tcpstat.tcps_sc_unreach++; 658 } 659 660 /* 661 * Build a new TCP socket structure from a syncache entry. 662 */ 663 static struct socket * 664 syncache_socket(sc, lso) 665 struct syncache *sc; 666 struct socket *lso; 667 { 668 struct inpcb *inp = NULL, *linp; 669 struct socket *so; 670 struct tcpcb *tp; 671 #ifdef INET6 672 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 673 #else 674 const boolean_t isipv6 = FALSE; 675 #endif 676 677 /* 678 * Ok, create the full blown connection, and set things up 679 * as they would have been set up if we had created the 680 * connection when the SYN arrived. If we can't create 681 * the connection, abort it. 682 */ 683 so = sonewconn(lso, SS_ISCONNECTED); 684 if (so == NULL) { 685 /* 686 * Drop the connection; we will send a RST if the peer 687 * retransmits the ACK, 688 */ 689 tcpstat.tcps_listendrop++; 690 goto abort; 691 } 692 693 inp = so->so_pcb; 694 695 /* 696 * Insert new socket into hash list. 697 */ 698 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 699 if (isipv6) { 700 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 701 } else { 702 #ifdef INET6 703 inp->inp_vflag &= ~INP_IPV6; 704 inp->inp_vflag |= INP_IPV4; 705 #endif 706 inp->inp_laddr = sc->sc_inc.inc_laddr; 707 } 708 inp->inp_lport = sc->sc_inc.inc_lport; 709 if (in_pcbinsporthash(inp) != 0) { 710 /* 711 * Undo the assignments above if we failed to 712 * put the PCB on the hash lists. 713 */ 714 if (isipv6) 715 inp->in6p_laddr = in6addr_any; 716 else 717 inp->inp_laddr.s_addr = INADDR_ANY; 718 inp->inp_lport = 0; 719 goto abort; 720 } 721 linp = so->so_pcb; 722 #ifdef IPSEC 723 /* copy old policy into new socket's */ 724 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 725 printf("syncache_expand: could not copy policy\n"); 726 #endif 727 if (isipv6) { 728 struct in6_addr laddr6; 729 struct sockaddr_in6 sin6; 730 /* 731 * Inherit socket options from the listening socket. 732 * Note that in6p_inputopts are not (and should not be) 733 * copied, since it stores previously received options and is 734 * used to detect if each new option is different than the 735 * previous one and hence should be passed to a user. 736 * If we copied in6p_inputopts, a user would not be able to 737 * receive options just after calling the accept system call. 738 */ 739 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 740 if (linp->in6p_outputopts) 741 inp->in6p_outputopts = 742 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 743 inp->in6p_route = sc->sc_route6; 744 sc->sc_route6.ro_rt = NULL; 745 746 sin6.sin6_family = AF_INET6; 747 sin6.sin6_len = sizeof sin6; 748 sin6.sin6_addr = sc->sc_inc.inc6_faddr; 749 sin6.sin6_port = sc->sc_inc.inc_fport; 750 sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; 751 laddr6 = inp->in6p_laddr; 752 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 753 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 754 if (in6_pcbconnect(inp, (struct sockaddr *)&sin6, &thread0)) { 755 inp->in6p_laddr = laddr6; 756 goto abort; 757 } 758 } else { 759 struct in_addr laddr; 760 struct sockaddr_in sin; 761 762 inp->inp_options = ip_srcroute(); 763 if (inp->inp_options == NULL) { 764 inp->inp_options = sc->sc_ipopts; 765 sc->sc_ipopts = NULL; 766 } 767 inp->inp_route = sc->sc_route; 768 sc->sc_route.ro_rt = NULL; 769 770 sin.sin_family = AF_INET; 771 sin.sin_len = sizeof sin; 772 sin.sin_addr = sc->sc_inc.inc_faddr; 773 sin.sin_port = sc->sc_inc.inc_fport; 774 bzero(sin.sin_zero, sizeof sin.sin_zero); 775 laddr = inp->inp_laddr; 776 if (inp->inp_laddr.s_addr == INADDR_ANY) 777 inp->inp_laddr = sc->sc_inc.inc_laddr; 778 if (in_pcbconnect(inp, (struct sockaddr *)&sin, &thread0)) { 779 inp->inp_laddr = laddr; 780 goto abort; 781 } 782 } 783 784 tp = intotcpcb(inp); 785 tp->t_state = TCPS_SYN_RECEIVED; 786 tp->iss = sc->sc_iss; 787 tp->irs = sc->sc_irs; 788 tcp_rcvseqinit(tp); 789 tcp_sendseqinit(tp); 790 tp->snd_wl1 = sc->sc_irs; 791 tp->rcv_up = sc->sc_irs + 1; 792 tp->rcv_wnd = sc->sc_wnd; 793 tp->rcv_adv += tp->rcv_wnd; 794 795 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 796 if (sc->sc_flags & SCF_NOOPT) 797 tp->t_flags |= TF_NOOPT; 798 if (sc->sc_flags & SCF_WINSCALE) { 799 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 800 tp->requested_s_scale = sc->sc_requested_s_scale; 801 tp->request_r_scale = sc->sc_request_r_scale; 802 } 803 if (sc->sc_flags & SCF_TIMESTAMP) { 804 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 805 tp->ts_recent = sc->sc_tsrecent; 806 tp->ts_recent_age = ticks; 807 } 808 if (sc->sc_flags & SCF_CC) { 809 /* 810 * Initialization of the tcpcb for transaction; 811 * set SND.WND = SEG.WND, 812 * initialize CCsend and CCrecv. 813 */ 814 tp->t_flags |= TF_REQ_CC | TF_RCVD_CC; 815 tp->cc_send = sc->sc_cc_send; 816 tp->cc_recv = sc->sc_cc_recv; 817 } 818 if (sc->sc_flags & SCF_SACK_PERMITTED) 819 tp->t_flags |= TF_SACK_PERMITTED; 820 821 tcp_mss(tp, sc->sc_peer_mss); 822 823 /* 824 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 825 */ 826 if (sc->sc_rxtslot != 0) 827 tp->snd_cwnd = tp->t_maxseg; 828 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 829 830 tcpstat.tcps_accepts++; 831 return (so); 832 833 abort: 834 if (so != NULL) 835 soabort(so); 836 return (NULL); 837 } 838 839 /* 840 * This function gets called when we receive an ACK for a 841 * socket in the LISTEN state. We look up the connection 842 * in the syncache, and if its there, we pull it out of 843 * the cache and turn it into a full-blown connection in 844 * the SYN-RECEIVED state. 845 */ 846 int 847 syncache_expand(inc, th, sop, m) 848 struct in_conninfo *inc; 849 struct tcphdr *th; 850 struct socket **sop; 851 struct mbuf *m; 852 { 853 struct syncache *sc; 854 struct syncache_head *sch; 855 struct socket *so; 856 857 sc = syncache_lookup(inc, &sch); 858 if (sc == NULL) { 859 /* 860 * There is no syncache entry, so see if this ACK is 861 * a returning syncookie. To do this, first: 862 * A. See if this socket has had a syncache entry dropped in 863 * the past. We don't want to accept a bogus syncookie 864 * if we've never received a SYN. 865 * B. check that the syncookie is valid. If it is, then 866 * cobble up a fake syncache entry, and return. 867 */ 868 if (!tcp_syncookies) 869 return (0); 870 sc = syncookie_lookup(inc, th, *sop); 871 if (sc == NULL) 872 return (0); 873 sch = NULL; 874 tcpstat.tcps_sc_recvcookie++; 875 } 876 877 /* 878 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 879 */ 880 if (th->th_ack != sc->sc_iss + 1) 881 return (0); 882 883 so = syncache_socket(sc, *sop); 884 if (so == NULL) { 885 #if 0 886 resetandabort: 887 /* XXXjlemon check this - is this correct? */ 888 tcp_respond(NULL, m, m, th, 889 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 890 #endif 891 m_freem(m); /* XXX only needed for above */ 892 tcpstat.tcps_sc_aborted++; 893 } else { 894 sc->sc_flags |= SCF_KEEPROUTE; 895 tcpstat.tcps_sc_completed++; 896 } 897 if (sch == NULL) 898 syncache_free(sc); 899 else 900 syncache_drop(sc, sch); 901 *sop = so; 902 return (1); 903 } 904 905 /* 906 * Given a LISTEN socket and an inbound SYN request, add 907 * this to the syn cache, and send back a segment: 908 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 909 * to the source. 910 * 911 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 912 * Doing so would require that we hold onto the data and deliver it 913 * to the application. However, if we are the target of a SYN-flood 914 * DoS attack, an attacker could send data which would eventually 915 * consume all available buffer space if it were ACKed. By not ACKing 916 * the data, we avoid this DoS scenario. 917 */ 918 int 919 syncache_add(inc, to, th, sop, m) 920 struct in_conninfo *inc; 921 struct tcpopt *to; 922 struct tcphdr *th; 923 struct socket **sop; 924 struct mbuf *m; 925 { 926 struct tcp_syncache_percpu *syncache_percpu; 927 struct tcpcb *tp; 928 struct socket *so; 929 struct syncache *sc = NULL; 930 struct syncache_head *sch; 931 struct mbuf *ipopts = NULL; 932 struct rmxp_tao *taop; 933 int win; 934 935 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 936 so = *sop; 937 tp = sototcpcb(so); 938 939 /* 940 * Remember the IP options, if any. 941 */ 942 #ifdef INET6 943 if (!inc->inc_isipv6) 944 #endif 945 ipopts = ip_srcroute(); 946 947 /* 948 * See if we already have an entry for this connection. 949 * If we do, resend the SYN,ACK, and reset the retransmit timer. 950 * 951 * XXX 952 * The syncache should be re-initialized with the contents 953 * of the new SYN which may have different options. 954 */ 955 sc = syncache_lookup(inc, &sch); 956 if (sc != NULL) { 957 tcpstat.tcps_sc_dupsyn++; 958 if (ipopts) { 959 /* 960 * If we were remembering a previous source route, 961 * forget it and use the new one we've been given. 962 */ 963 if (sc->sc_ipopts) 964 m_free(sc->sc_ipopts); 965 sc->sc_ipopts = ipopts; 966 } 967 /* 968 * Update timestamp if present. 969 */ 970 if (sc->sc_flags & SCF_TIMESTAMP) 971 sc->sc_tsrecent = to->to_tsval; 972 973 /* Just update the TOF_SACK_PERMITTED for now. */ 974 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 975 sc->sc_flags |= SCF_SACK_PERMITTED; 976 else 977 sc->sc_flags &= ~SCF_SACK_PERMITTED; 978 979 /* 980 * PCB may have changed, pick up new values. 981 */ 982 sc->sc_tp = tp; 983 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 984 if (syncache_respond(sc, m) == 0) { 985 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 986 sc, sc_timerq); 987 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 988 tcpstat.tcps_sndacks++; 989 tcpstat.tcps_sndtotal++; 990 } 991 *sop = NULL; 992 return (1); 993 } 994 995 /* 996 * This allocation is guaranteed to succeed because we 997 * preallocate one more syncache entry than cache_limit. 998 */ 999 sc = zalloc(tcp_syncache.zone); 1000 1001 /* 1002 * Fill in the syncache values. 1003 */ 1004 sc->sc_tp = tp; 1005 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1006 sc->sc_ipopts = ipopts; 1007 sc->sc_inc.inc_fport = inc->inc_fport; 1008 sc->sc_inc.inc_lport = inc->inc_lport; 1009 #ifdef INET6 1010 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1011 if (inc->inc_isipv6) { 1012 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1013 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1014 sc->sc_route6.ro_rt = NULL; 1015 } else 1016 #endif 1017 { 1018 sc->sc_inc.inc_faddr = inc->inc_faddr; 1019 sc->sc_inc.inc_laddr = inc->inc_laddr; 1020 sc->sc_route.ro_rt = NULL; 1021 } 1022 sc->sc_irs = th->th_seq; 1023 sc->sc_flags = 0; 1024 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 1025 if (tcp_syncookies) 1026 sc->sc_iss = syncookie_generate(sc); 1027 else 1028 sc->sc_iss = arc4random(); 1029 1030 /* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */ 1031 win = sbspace(&so->so_rcv); 1032 win = imax(win, 0); 1033 win = imin(win, TCP_MAXWIN); 1034 sc->sc_wnd = win; 1035 1036 if (tcp_do_rfc1323) { 1037 /* 1038 * A timestamp received in a SYN makes 1039 * it ok to send timestamp requests and replies. 1040 */ 1041 if (to->to_flags & TOF_TS) { 1042 sc->sc_tsrecent = to->to_tsval; 1043 sc->sc_flags |= SCF_TIMESTAMP; 1044 } 1045 if (to->to_flags & TOF_SCALE) { 1046 int wscale = 0; 1047 1048 /* Compute proper scaling value from buffer space */ 1049 while (wscale < TCP_MAX_WINSHIFT && 1050 (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat) 1051 wscale++; 1052 sc->sc_request_r_scale = wscale; 1053 sc->sc_requested_s_scale = to->to_requested_s_scale; 1054 sc->sc_flags |= SCF_WINSCALE; 1055 } 1056 } 1057 if (tcp_do_rfc1644) { 1058 /* 1059 * A CC or CC.new option received in a SYN makes 1060 * it ok to send CC in subsequent segments. 1061 */ 1062 if (to->to_flags & (TOF_CC | TOF_CCNEW)) { 1063 sc->sc_cc_recv = to->to_cc; 1064 sc->sc_cc_send = CC_INC(tcp_ccgen); 1065 sc->sc_flags |= SCF_CC; 1066 } 1067 } 1068 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1069 sc->sc_flags |= SCF_SACK_PERMITTED; 1070 if (tp->t_flags & TF_NOOPT) 1071 sc->sc_flags = SCF_NOOPT; 1072 1073 /* 1074 * XXX 1075 * We have the option here of not doing TAO (even if the segment 1076 * qualifies) and instead fall back to a normal 3WHS via the syncache. 1077 * This allows us to apply synflood protection to TAO-qualifying SYNs 1078 * also. However, there should be a hueristic to determine when to 1079 * do this, and is not present at the moment. 1080 */ 1081 1082 /* 1083 * Perform TAO test on incoming CC (SEG.CC) option, if any. 1084 * - compare SEG.CC against cached CC from the same host, if any. 1085 * - if SEG.CC > chached value, SYN must be new and is accepted 1086 * immediately: save new CC in the cache, mark the socket 1087 * connected, enter ESTABLISHED state, turn on flag to 1088 * send a SYN in the next segment. 1089 * A virtual advertised window is set in rcv_adv to 1090 * initialize SWS prevention. Then enter normal segment 1091 * processing: drop SYN, process data and FIN. 1092 * - otherwise do a normal 3-way handshake. 1093 */ 1094 taop = tcp_gettaocache(&sc->sc_inc); 1095 if (to->to_flags & TOF_CC) { 1096 if ((tp->t_flags & TF_NOPUSH) && 1097 sc->sc_flags & SCF_CC && 1098 taop != NULL && taop->tao_cc != 0 && 1099 CC_GT(to->to_cc, taop->tao_cc)) { 1100 sc->sc_rxtslot = 0; 1101 so = syncache_socket(sc, *sop); 1102 if (so != NULL) { 1103 sc->sc_flags |= SCF_KEEPROUTE; 1104 taop->tao_cc = to->to_cc; 1105 *sop = so; 1106 } 1107 syncache_free(sc); 1108 return (so != NULL); 1109 } 1110 } else { 1111 /* 1112 * No CC option, but maybe CC.NEW: invalidate cached value. 1113 */ 1114 if (taop != NULL) 1115 taop->tao_cc = 0; 1116 } 1117 /* 1118 * TAO test failed or there was no CC option, 1119 * do a standard 3-way handshake. 1120 */ 1121 if (syncache_respond(sc, m) == 0) { 1122 syncache_insert(sc, sch); 1123 tcpstat.tcps_sndacks++; 1124 tcpstat.tcps_sndtotal++; 1125 } else { 1126 syncache_free(sc); 1127 tcpstat.tcps_sc_dropped++; 1128 } 1129 *sop = NULL; 1130 return (1); 1131 } 1132 1133 static int 1134 syncache_respond(sc, m) 1135 struct syncache *sc; 1136 struct mbuf *m; 1137 { 1138 u_int8_t *optp; 1139 int optlen, error; 1140 u_int16_t tlen, hlen, mssopt; 1141 struct ip *ip = NULL; 1142 struct rtentry *rt; 1143 struct tcphdr *th; 1144 struct ip6_hdr *ip6 = NULL; 1145 #ifdef INET6 1146 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1147 #else 1148 const boolean_t isipv6 = FALSE; 1149 #endif 1150 1151 if (isipv6) { 1152 rt = tcp_rtlookup6(&sc->sc_inc); 1153 if (rt != NULL) 1154 mssopt = rt->rt_ifp->if_mtu - 1155 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1156 else 1157 mssopt = tcp_v6mssdflt; 1158 hlen = sizeof(struct ip6_hdr); 1159 } else { 1160 rt = tcp_rtlookup(&sc->sc_inc); 1161 if (rt != NULL) 1162 mssopt = rt->rt_ifp->if_mtu - 1163 (sizeof(struct ip) + sizeof(struct tcphdr)); 1164 else 1165 mssopt = tcp_mssdflt; 1166 hlen = sizeof(struct ip); 1167 } 1168 1169 /* Compute the size of the TCP options. */ 1170 if (sc->sc_flags & SCF_NOOPT) { 1171 optlen = 0; 1172 } else { 1173 optlen = TCPOLEN_MAXSEG + 1174 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1175 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1176 ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0) + 1177 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1178 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1179 } 1180 tlen = hlen + sizeof(struct tcphdr) + optlen; 1181 1182 /* 1183 * XXX 1184 * assume that the entire packet will fit in a header mbuf 1185 */ 1186 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1187 1188 /* 1189 * XXX shouldn't this reuse the mbuf if possible ? 1190 * Create the IP+TCP header from scratch. 1191 */ 1192 if (m) 1193 m_freem(m); 1194 1195 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 1196 if (m == NULL) 1197 return (ENOBUFS); 1198 m->m_data += max_linkhdr; 1199 m->m_len = tlen; 1200 m->m_pkthdr.len = tlen; 1201 m->m_pkthdr.rcvif = NULL; 1202 1203 if (isipv6) { 1204 ip6 = mtod(m, struct ip6_hdr *); 1205 ip6->ip6_vfc = IPV6_VERSION; 1206 ip6->ip6_nxt = IPPROTO_TCP; 1207 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1208 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1209 ip6->ip6_plen = htons(tlen - hlen); 1210 /* ip6_hlim is set after checksum */ 1211 /* ip6_flow = ??? */ 1212 1213 th = (struct tcphdr *)(ip6 + 1); 1214 } else { 1215 ip = mtod(m, struct ip *); 1216 ip->ip_v = IPVERSION; 1217 ip->ip_hl = sizeof(struct ip) >> 2; 1218 ip->ip_len = tlen; 1219 ip->ip_id = 0; 1220 ip->ip_off = 0; 1221 ip->ip_sum = 0; 1222 ip->ip_p = IPPROTO_TCP; 1223 ip->ip_src = sc->sc_inc.inc_laddr; 1224 ip->ip_dst = sc->sc_inc.inc_faddr; 1225 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1226 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1227 1228 /* 1229 * See if we should do MTU discovery. Route lookups are 1230 * expensive, so we will only unset the DF bit if: 1231 * 1232 * 1) path_mtu_discovery is disabled 1233 * 2) the SCF_UNREACH flag has been set 1234 */ 1235 if (path_mtu_discovery 1236 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1237 ip->ip_off |= IP_DF; 1238 } 1239 1240 th = (struct tcphdr *)(ip + 1); 1241 } 1242 th->th_sport = sc->sc_inc.inc_lport; 1243 th->th_dport = sc->sc_inc.inc_fport; 1244 1245 th->th_seq = htonl(sc->sc_iss); 1246 th->th_ack = htonl(sc->sc_irs + 1); 1247 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1248 th->th_x2 = 0; 1249 th->th_flags = TH_SYN | TH_ACK; 1250 th->th_win = htons(sc->sc_wnd); 1251 th->th_urp = 0; 1252 1253 /* Tack on the TCP options. */ 1254 if (optlen == 0) 1255 goto no_options; 1256 optp = (u_int8_t *)(th + 1); 1257 *optp++ = TCPOPT_MAXSEG; 1258 *optp++ = TCPOLEN_MAXSEG; 1259 *optp++ = (mssopt >> 8) & 0xff; 1260 *optp++ = mssopt & 0xff; 1261 1262 if (sc->sc_flags & SCF_WINSCALE) { 1263 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1264 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1265 sc->sc_request_r_scale); 1266 optp += 4; 1267 } 1268 1269 if (sc->sc_flags & SCF_TIMESTAMP) { 1270 u_int32_t *lp = (u_int32_t *)(optp); 1271 1272 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1273 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1274 *lp++ = htonl(ticks); 1275 *lp = htonl(sc->sc_tsrecent); 1276 optp += TCPOLEN_TSTAMP_APPA; 1277 } 1278 1279 /* 1280 * Send CC and CC.echo if we received CC from our peer. 1281 */ 1282 if (sc->sc_flags & SCF_CC) { 1283 u_int32_t *lp = (u_int32_t *)(optp); 1284 1285 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC)); 1286 *lp++ = htonl(sc->sc_cc_send); 1287 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO)); 1288 *lp = htonl(sc->sc_cc_recv); 1289 optp += TCPOLEN_CC_APPA * 2; 1290 } 1291 1292 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1293 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1294 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1295 } 1296 1297 no_options: 1298 if (isipv6) { 1299 struct route_in6 *ro6 = &sc->sc_route6; 1300 1301 th->th_sum = 0; 1302 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1303 ip6->ip6_hlim = in6_selecthlim(NULL, 1304 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1305 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1306 sc->sc_tp->t_inpcb); 1307 } else { 1308 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1309 htons(tlen - hlen + IPPROTO_TCP)); 1310 m->m_pkthdr.csum_flags = CSUM_TCP; 1311 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1312 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 0, NULL, 1313 sc->sc_tp->t_inpcb); 1314 } 1315 return (error); 1316 } 1317 1318 /* 1319 * cookie layers: 1320 * 1321 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1322 * | peer iss | 1323 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1324 * | 0 |(A)| | 1325 * (A): peer mss index 1326 */ 1327 1328 /* 1329 * The values below are chosen to minimize the size of the tcp_secret 1330 * table, as well as providing roughly a 16 second lifetime for the cookie. 1331 */ 1332 1333 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1334 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1335 1336 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1337 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1338 #define SYNCOOKIE_TIMEOUT \ 1339 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1340 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1341 1342 static struct { 1343 u_int32_t ts_secbits[4]; 1344 u_int ts_expire; 1345 } tcp_secret[SYNCOOKIE_NSECRETS]; 1346 1347 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1348 1349 static MD5_CTX syn_ctx; 1350 1351 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1352 1353 struct md5_add { 1354 u_int32_t laddr, faddr; 1355 u_int32_t secbits[4]; 1356 u_int16_t lport, fport; 1357 }; 1358 1359 #ifdef CTASSERT 1360 CTASSERT(sizeof(struct md5_add) == 28); 1361 #endif 1362 1363 /* 1364 * Consider the problem of a recreated (and retransmitted) cookie. If the 1365 * original SYN was accepted, the connection is established. The second 1366 * SYN is inflight, and if it arrives with an ISN that falls within the 1367 * receive window, the connection is killed. 1368 * 1369 * However, since cookies have other problems, this may not be worth 1370 * worrying about. 1371 */ 1372 1373 static u_int32_t 1374 syncookie_generate(struct syncache *sc) 1375 { 1376 u_int32_t md5_buffer[4]; 1377 u_int32_t data; 1378 int idx, i; 1379 struct md5_add add; 1380 #ifdef INET6 1381 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1382 #else 1383 const boolean_t isipv6 = FALSE; 1384 #endif 1385 1386 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1387 if (tcp_secret[idx].ts_expire < ticks) { 1388 for (i = 0; i < 4; i++) 1389 tcp_secret[idx].ts_secbits[i] = arc4random(); 1390 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1391 } 1392 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1393 if (tcp_msstab[data] <= sc->sc_peer_mss) 1394 break; 1395 data = (data << SYNCOOKIE_WNDBITS) | idx; 1396 data ^= sc->sc_irs; /* peer's iss */ 1397 MD5Init(&syn_ctx); 1398 if (isipv6) { 1399 MD5Add(sc->sc_inc.inc6_laddr); 1400 MD5Add(sc->sc_inc.inc6_faddr); 1401 add.laddr = 0; 1402 add.faddr = 0; 1403 } else { 1404 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1405 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1406 } 1407 add.lport = sc->sc_inc.inc_lport; 1408 add.fport = sc->sc_inc.inc_fport; 1409 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1410 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1411 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1412 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1413 MD5Add(add); 1414 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1415 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1416 return (data); 1417 } 1418 1419 static struct syncache * 1420 syncookie_lookup(inc, th, so) 1421 struct in_conninfo *inc; 1422 struct tcphdr *th; 1423 struct socket *so; 1424 { 1425 u_int32_t md5_buffer[4]; 1426 struct syncache *sc; 1427 u_int32_t data; 1428 int wnd, idx; 1429 struct md5_add add; 1430 1431 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1432 idx = data & SYNCOOKIE_WNDMASK; 1433 if (tcp_secret[idx].ts_expire < ticks || 1434 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1435 return (NULL); 1436 MD5Init(&syn_ctx); 1437 #ifdef INET6 1438 if (inc->inc_isipv6) { 1439 MD5Add(inc->inc6_laddr); 1440 MD5Add(inc->inc6_faddr); 1441 add.laddr = 0; 1442 add.faddr = 0; 1443 } else 1444 #endif 1445 { 1446 add.laddr = inc->inc_laddr.s_addr; 1447 add.faddr = inc->inc_faddr.s_addr; 1448 } 1449 add.lport = inc->inc_lport; 1450 add.fport = inc->inc_fport; 1451 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1452 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1453 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1454 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1455 MD5Add(add); 1456 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1457 data ^= md5_buffer[0]; 1458 if (data & ~SYNCOOKIE_DATAMASK) 1459 return (NULL); 1460 data = data >> SYNCOOKIE_WNDBITS; 1461 1462 /* 1463 * This allocation is guaranteed to succeed because we 1464 * preallocate one more syncache entry than cache_limit. 1465 */ 1466 sc = zalloc(tcp_syncache.zone); 1467 1468 /* 1469 * Fill in the syncache values. 1470 * XXX duplicate code from syncache_add 1471 */ 1472 sc->sc_ipopts = NULL; 1473 sc->sc_inc.inc_fport = inc->inc_fport; 1474 sc->sc_inc.inc_lport = inc->inc_lport; 1475 #ifdef INET6 1476 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1477 if (inc->inc_isipv6) { 1478 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1479 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1480 sc->sc_route6.ro_rt = NULL; 1481 } else 1482 #endif 1483 { 1484 sc->sc_inc.inc_faddr = inc->inc_faddr; 1485 sc->sc_inc.inc_laddr = inc->inc_laddr; 1486 sc->sc_route.ro_rt = NULL; 1487 } 1488 sc->sc_irs = th->th_seq - 1; 1489 sc->sc_iss = th->th_ack - 1; 1490 wnd = sbspace(&so->so_rcv); 1491 wnd = imax(wnd, 0); 1492 wnd = imin(wnd, TCP_MAXWIN); 1493 sc->sc_wnd = wnd; 1494 sc->sc_flags = 0; 1495 sc->sc_rxtslot = 0; 1496 sc->sc_peer_mss = tcp_msstab[data]; 1497 return (sc); 1498 } 1499