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 MALLOC(syncache_percpu->hashbase, struct syncache_head *, 316 tcp_syncache.hashsize * sizeof(struct syncache_head), 317 M_SYNCACHE, M_WAITOK); 318 319 /* Initialize the hash buckets. */ 320 for (i = 0; i < tcp_syncache.hashsize; i++) { 321 struct syncache_head *bucket; 322 323 bucket = &syncache_percpu->hashbase[i]; 324 TAILQ_INIT(&bucket->sch_bucket); 325 bucket->sch_length = 0; 326 } 327 328 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 329 /* Initialize the timer queues. */ 330 TAILQ_INIT(&syncache_percpu->timerq[i]); 331 callout_init_mp(&syncache_percpu->tt_timerq[i]); 332 333 syncache_percpu->mrec[i].slot = i; 334 syncache_percpu->mrec[i].port = cpu_portfn(cpu); 335 syncache_percpu->mrec[i].msg.nm_mrec = 336 &syncache_percpu->mrec[i]; 337 netmsg_init(&syncache_percpu->mrec[i].msg.base, 338 NULL, &syncache_null_rport, 339 0, syncache_timer_handler); 340 } 341 } 342 } 343 344 static void 345 syncache_insert(struct syncache *sc, struct syncache_head *sch) 346 { 347 struct tcp_syncache_percpu *syncache_percpu; 348 struct syncache *sc2; 349 int i; 350 351 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 352 353 /* 354 * Make sure that we don't overflow the per-bucket 355 * limit or the total cache size limit. 356 */ 357 if (sch->sch_length >= tcp_syncache.bucket_limit) { 358 /* 359 * The bucket is full, toss the oldest element. 360 */ 361 sc2 = TAILQ_FIRST(&sch->sch_bucket); 362 sc2->sc_tp->ts_recent = ticks; 363 syncache_drop(sc2, sch); 364 tcpstat.tcps_sc_bucketoverflow++; 365 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) { 366 /* 367 * The cache is full. Toss the oldest entry in the 368 * entire cache. This is the front entry in the 369 * first non-empty timer queue with the largest 370 * timeout value. 371 */ 372 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 373 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]); 374 while (sc2 && (sc2->sc_flags & SCF_MARKER)) 375 sc2 = TAILQ_NEXT(sc2, sc_timerq); 376 if (sc2 != NULL) 377 break; 378 } 379 sc2->sc_tp->ts_recent = ticks; 380 syncache_drop(sc2, NULL); 381 tcpstat.tcps_sc_cacheoverflow++; 382 } 383 384 /* Initialize the entry's timer. */ 385 syncache_timeout(syncache_percpu, sc, 0); 386 387 /* Put it into the bucket. */ 388 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 389 sch->sch_length++; 390 syncache_percpu->cache_count++; 391 tcpstat.tcps_sc_added++; 392 } 393 394 void 395 syncache_destroy(struct tcpcb *tp) 396 { 397 struct tcp_syncache_percpu *syncache_percpu; 398 struct syncache_head *bucket; 399 struct syncache *sc; 400 int i; 401 402 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 403 sc = NULL; 404 405 for (i = 0; i < tcp_syncache.hashsize; i++) { 406 bucket = &syncache_percpu->hashbase[i]; 407 TAILQ_FOREACH(sc, &bucket->sch_bucket, sc_hash) { 408 if (sc->sc_tp == tp) 409 sc->sc_tp = NULL; 410 } 411 } 412 } 413 414 static void 415 syncache_drop(struct syncache *sc, struct syncache_head *sch) 416 { 417 struct tcp_syncache_percpu *syncache_percpu; 418 #ifdef INET6 419 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 420 #else 421 const boolean_t isipv6 = FALSE; 422 #endif 423 424 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 425 426 if (sch == NULL) { 427 if (isipv6) { 428 sch = &syncache_percpu->hashbase[ 429 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 430 } else { 431 sch = &syncache_percpu->hashbase[ 432 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 433 } 434 } 435 436 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 437 sch->sch_length--; 438 syncache_percpu->cache_count--; 439 440 /* 441 * Cleanup 442 */ 443 if (sc->sc_tp) 444 sc->sc_tp = NULL; 445 446 /* 447 * Remove the entry from the syncache timer/timeout queue. Note 448 * that we do not try to stop any running timer since we do not know 449 * whether the timer's message is in-transit or not. Since timeouts 450 * are fairly long, taking an unneeded callout does not detrimentally 451 * effect performance. 452 */ 453 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq); 454 455 syncache_free(sc); 456 } 457 458 /* 459 * Place a timeout message on the TCP thread's message queue. 460 * This routine runs in soft interrupt context. 461 * 462 * An invariant is for this routine to be called, the callout must 463 * have been active. Note that the callout is not deactivated until 464 * after the message has been processed in syncache_timer_handler() below. 465 */ 466 static void 467 syncache_timer(void *p) 468 { 469 struct netmsg_sc_timer *msg = p; 470 471 lwkt_sendmsg(msg->nm_mrec->port, &msg->base.lmsg); 472 } 473 474 /* 475 * Service a timer message queued by timer expiration. 476 * This routine runs in the TCP protocol thread. 477 * 478 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 479 * If we have retransmitted an entry the maximum number of times, expire it. 480 * 481 * When we finish processing timed-out entries, we restart the timer if there 482 * are any entries still on the queue and deactivate it otherwise. Only after 483 * a timer has been deactivated here can it be restarted by syncache_timeout(). 484 */ 485 static void 486 syncache_timer_handler(netmsg_t msg) 487 { 488 struct tcp_syncache_percpu *syncache_percpu; 489 struct syncache *sc; 490 struct syncache marker; 491 struct syncache_list *list; 492 struct inpcb *inp; 493 int slot; 494 495 slot = ((struct netmsg_sc_timer *)msg)->nm_mrec->slot; 496 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 497 498 list = &syncache_percpu->timerq[slot]; 499 500 /* 501 * Use a marker to keep our place in the scan. syncache_drop() 502 * can block and cause any next pointer we cache to become stale. 503 */ 504 marker.sc_flags = SCF_MARKER; 505 TAILQ_INSERT_HEAD(list, &marker, sc_timerq); 506 507 while ((sc = TAILQ_NEXT(&marker, sc_timerq)) != NULL) { 508 /* 509 * Move the marker. 510 */ 511 TAILQ_REMOVE(list, &marker, sc_timerq); 512 TAILQ_INSERT_AFTER(list, sc, &marker, sc_timerq); 513 514 if (sc->sc_flags & SCF_MARKER) 515 continue; 516 517 if (ticks < sc->sc_rxttime) 518 break; /* finished because timerq sorted by time */ 519 if (sc->sc_tp == NULL) { 520 syncache_drop(sc, NULL); 521 tcpstat.tcps_sc_stale++; 522 continue; 523 } 524 inp = sc->sc_tp->t_inpcb; 525 if (slot == SYNCACHE_MAXREXMTS || 526 slot >= tcp_syncache.rexmt_limit || 527 inp == NULL || 528 inp->inp_gencnt != sc->sc_inp_gencnt) { 529 syncache_drop(sc, NULL); 530 tcpstat.tcps_sc_stale++; 531 continue; 532 } 533 /* 534 * syncache_respond() may call back into the syncache to 535 * to modify another entry, so do not obtain the next 536 * entry on the timer chain until it has completed. 537 */ 538 syncache_respond(sc, NULL); 539 tcpstat.tcps_sc_retransmitted++; 540 TAILQ_REMOVE(list, sc, sc_timerq); 541 syncache_timeout(syncache_percpu, sc, slot + 1); 542 } 543 TAILQ_REMOVE(list, &marker, sc_timerq); 544 545 if (sc != NULL) { 546 callout_reset(&syncache_percpu->tt_timerq[slot], 547 sc->sc_rxttime - ticks, syncache_timer, 548 &syncache_percpu->mrec[slot]); 549 } else { 550 callout_deactivate(&syncache_percpu->tt_timerq[slot]); 551 } 552 lwkt_replymsg(&msg->base.lmsg, 0); 553 } 554 555 /* 556 * Find an entry in the syncache. 557 */ 558 struct syncache * 559 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 560 { 561 struct tcp_syncache_percpu *syncache_percpu; 562 struct syncache *sc; 563 struct syncache_head *sch; 564 565 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 566 #ifdef INET6 567 if (inc->inc_isipv6) { 568 sch = &syncache_percpu->hashbase[ 569 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 570 *schp = sch; 571 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) 572 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 573 return (sc); 574 } else 575 #endif 576 { 577 sch = &syncache_percpu->hashbase[ 578 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 579 *schp = sch; 580 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 581 #ifdef INET6 582 if (sc->sc_inc.inc_isipv6) 583 continue; 584 #endif 585 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 586 return (sc); 587 } 588 } 589 return (NULL); 590 } 591 592 /* 593 * This function is called when we get a RST for a 594 * non-existent connection, so that we can see if the 595 * connection is in the syn cache. If it is, zap it. 596 */ 597 void 598 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 599 { 600 struct syncache *sc; 601 struct syncache_head *sch; 602 603 sc = syncache_lookup(inc, &sch); 604 if (sc == NULL) { 605 return; 606 } 607 /* 608 * If the RST bit is set, check the sequence number to see 609 * if this is a valid reset segment. 610 * RFC 793 page 37: 611 * In all states except SYN-SENT, all reset (RST) segments 612 * are validated by checking their SEQ-fields. A reset is 613 * valid if its sequence number is in the window. 614 * 615 * The sequence number in the reset segment is normally an 616 * echo of our outgoing acknowlegement numbers, but some hosts 617 * send a reset with the sequence number at the rightmost edge 618 * of our receive window, and we have to handle this case. 619 */ 620 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 621 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 622 syncache_drop(sc, sch); 623 tcpstat.tcps_sc_reset++; 624 } 625 } 626 627 void 628 syncache_badack(struct in_conninfo *inc) 629 { 630 struct syncache *sc; 631 struct syncache_head *sch; 632 633 sc = syncache_lookup(inc, &sch); 634 if (sc != NULL) { 635 syncache_drop(sc, sch); 636 tcpstat.tcps_sc_badack++; 637 } 638 } 639 640 void 641 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 642 { 643 struct syncache *sc; 644 struct syncache_head *sch; 645 646 /* we are called at splnet() here */ 647 sc = syncache_lookup(inc, &sch); 648 if (sc == NULL) 649 return; 650 651 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 652 if (ntohl(th->th_seq) != sc->sc_iss) 653 return; 654 655 /* 656 * If we've rertransmitted 3 times and this is our second error, 657 * we remove the entry. Otherwise, we allow it to continue on. 658 * This prevents us from incorrectly nuking an entry during a 659 * spurious network outage. 660 * 661 * See tcp_notify(). 662 */ 663 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 664 sc->sc_flags |= SCF_UNREACH; 665 return; 666 } 667 syncache_drop(sc, sch); 668 tcpstat.tcps_sc_unreach++; 669 } 670 671 /* 672 * Build a new TCP socket structure from a syncache entry. 673 * 674 * This is called from the context of the SYN+ACK 675 */ 676 static struct socket * 677 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 678 { 679 struct inpcb *inp = NULL, *linp; 680 struct socket *so; 681 struct tcpcb *tp, *ltp; 682 lwkt_port_t port; 683 #ifdef INET6 684 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 685 #else 686 const boolean_t isipv6 = FALSE; 687 #endif 688 struct sockaddr_in sin_faddr; 689 struct sockaddr_in6 sin6_faddr; 690 struct sockaddr *faddr; 691 692 if (isipv6) { 693 faddr = (struct sockaddr *)&sin6_faddr; 694 sin6_faddr.sin6_family = AF_INET6; 695 sin6_faddr.sin6_len = sizeof(sin6_faddr); 696 sin6_faddr.sin6_addr = sc->sc_inc.inc6_faddr; 697 sin6_faddr.sin6_port = sc->sc_inc.inc_fport; 698 sin6_faddr.sin6_flowinfo = sin6_faddr.sin6_scope_id = 0; 699 } else { 700 faddr = (struct sockaddr *)&sin_faddr; 701 sin_faddr.sin_family = AF_INET; 702 sin_faddr.sin_len = sizeof(sin_faddr); 703 sin_faddr.sin_addr = sc->sc_inc.inc_faddr; 704 sin_faddr.sin_port = sc->sc_inc.inc_fport; 705 bzero(sin_faddr.sin_zero, sizeof(sin_faddr.sin_zero)); 706 } 707 708 /* 709 * Ok, create the full blown connection, and set things up 710 * as they would have been set up if we had created the 711 * connection when the SYN arrived. If we can't create 712 * the connection, abort it. 713 * 714 * Set the protocol processing port for the socket to the current 715 * port (that the connection came in on). 716 */ 717 so = sonewconn_faddr(lso, SS_ISCONNECTED, faddr); 718 if (so == NULL) { 719 /* 720 * Drop the connection; we will send a RST if the peer 721 * retransmits the ACK, 722 */ 723 tcpstat.tcps_listendrop++; 724 goto abort; 725 } 726 727 /* 728 * Insert new socket into hash list. 729 */ 730 inp = so->so_pcb; 731 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 732 if (isipv6) { 733 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 734 } else { 735 #ifdef INET6 736 inp->inp_vflag &= ~INP_IPV6; 737 inp->inp_vflag |= INP_IPV4; 738 inp->inp_flags &= ~IN6P_IPV6_V6ONLY; 739 #endif 740 inp->inp_laddr = sc->sc_inc.inc_laddr; 741 } 742 inp->inp_lport = sc->sc_inc.inc_lport; 743 if (in_pcbinsporthash(inp) != 0) { 744 /* 745 * Undo the assignments above if we failed to 746 * put the PCB on the hash lists. 747 */ 748 if (isipv6) 749 inp->in6p_laddr = kin6addr_any; 750 else 751 inp->inp_laddr.s_addr = INADDR_ANY; 752 inp->inp_lport = 0; 753 goto abort; 754 } 755 linp = lso->so_pcb; 756 #ifdef IPSEC 757 /* copy old policy into new socket's */ 758 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 759 kprintf("syncache_expand: could not copy policy\n"); 760 #endif 761 if (isipv6) { 762 struct in6_addr laddr6; 763 /* 764 * Inherit socket options from the listening socket. 765 * Note that in6p_inputopts are not (and should not be) 766 * copied, since it stores previously received options and is 767 * used to detect if each new option is different than the 768 * previous one and hence should be passed to a user. 769 * If we copied in6p_inputopts, a user would not be able to 770 * receive options just after calling the accept system call. 771 */ 772 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 773 if (linp->in6p_outputopts) 774 inp->in6p_outputopts = 775 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 776 inp->in6p_route = sc->sc_route6; 777 sc->sc_route6.ro_rt = NULL; 778 779 laddr6 = inp->in6p_laddr; 780 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 781 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 782 if (in6_pcbconnect(inp, faddr, &thread0)) { 783 inp->in6p_laddr = laddr6; 784 goto abort; 785 } 786 } else { 787 struct in_addr laddr; 788 789 inp->inp_options = ip_srcroute(m); 790 if (inp->inp_options == NULL) { 791 inp->inp_options = sc->sc_ipopts; 792 sc->sc_ipopts = NULL; 793 } 794 inp->inp_route = sc->sc_route; 795 sc->sc_route.ro_rt = NULL; 796 797 laddr = inp->inp_laddr; 798 if (inp->inp_laddr.s_addr == INADDR_ANY) 799 inp->inp_laddr = sc->sc_inc.inc_laddr; 800 if (in_pcbconnect(inp, faddr, &thread0)) { 801 inp->inp_laddr = laddr; 802 goto abort; 803 } 804 } 805 806 /* 807 * The current port should be in the context of the SYN+ACK and 808 * so should match the tcp address port. 809 * 810 * XXX we may be running on the netisr thread instead of a tcp 811 * thread, in which case port will not match 812 * curthread->td_msgport. 813 */ 814 if (isipv6) { 815 port = tcp6_addrport(); 816 } else { 817 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport, 818 inp->inp_laddr.s_addr, inp->inp_lport); 819 } 820 if (port != &curthread->td_msgport) { 821 print_backtrace(-1); 822 kprintf("TCP PORT MISMATCH %p vs %p\n", 823 port, &curthread->td_msgport); 824 } 825 /*KKASSERT(port == &curthread->td_msgport);*/ 826 827 tp = intotcpcb(inp); 828 tp->t_state = TCPS_SYN_RECEIVED; 829 tp->iss = sc->sc_iss; 830 tp->irs = sc->sc_irs; 831 tcp_rcvseqinit(tp); 832 tcp_sendseqinit(tp); 833 tp->snd_wl1 = sc->sc_irs; 834 tp->rcv_up = sc->sc_irs + 1; 835 tp->rcv_wnd = sc->sc_wnd; 836 tp->rcv_adv += tp->rcv_wnd; 837 838 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 839 if (sc->sc_flags & SCF_NOOPT) 840 tp->t_flags |= TF_NOOPT; 841 if (sc->sc_flags & SCF_WINSCALE) { 842 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 843 tp->requested_s_scale = sc->sc_requested_s_scale; 844 tp->request_r_scale = sc->sc_request_r_scale; 845 } 846 if (sc->sc_flags & SCF_TIMESTAMP) { 847 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 848 tp->ts_recent = sc->sc_tsrecent; 849 tp->ts_recent_age = ticks; 850 } 851 if (sc->sc_flags & SCF_SACK_PERMITTED) 852 tp->t_flags |= TF_SACK_PERMITTED; 853 854 #ifdef TCP_SIGNATURE 855 if (sc->sc_flags & SCF_SIGNATURE) 856 tp->t_flags |= TF_SIGNATURE; 857 #endif /* TCP_SIGNATURE */ 858 859 860 tcp_mss(tp, sc->sc_peer_mss); 861 862 /* 863 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 864 */ 865 if (sc->sc_rxtslot != 0) 866 tp->snd_cwnd = tp->t_maxseg; 867 868 /* 869 * Inherit some properties from the listen socket 870 */ 871 ltp = intotcpcb(linp); 872 tp->t_keepinit = ltp->t_keepinit; 873 tp->t_keepidle = ltp->t_keepidle; 874 tp->t_keepintvl = ltp->t_keepintvl; 875 tp->t_keepcnt = ltp->t_keepcnt; 876 tp->t_maxidle = ltp->t_maxidle; 877 878 tcp_create_timermsg(tp, port); 879 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep); 880 881 tcpstat.tcps_accepts++; 882 return (so); 883 884 abort: 885 if (so != NULL) 886 soabort_oncpu(so); 887 return (NULL); 888 } 889 890 /* 891 * This function gets called when we receive an ACK for a 892 * socket in the LISTEN state. We look up the connection 893 * in the syncache, and if its there, we pull it out of 894 * the cache and turn it into a full-blown connection in 895 * the SYN-RECEIVED state. 896 */ 897 int 898 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop, 899 struct mbuf *m) 900 { 901 struct syncache *sc; 902 struct syncache_head *sch; 903 struct socket *so; 904 905 sc = syncache_lookup(inc, &sch); 906 if (sc == NULL) { 907 /* 908 * There is no syncache entry, so see if this ACK is 909 * a returning syncookie. To do this, first: 910 * A. See if this socket has had a syncache entry dropped in 911 * the past. We don't want to accept a bogus syncookie 912 * if we've never received a SYN. 913 * B. check that the syncookie is valid. If it is, then 914 * cobble up a fake syncache entry, and return. 915 */ 916 if (!tcp_syncookies) 917 return (0); 918 sc = syncookie_lookup(inc, th, *sop); 919 if (sc == NULL) 920 return (0); 921 sch = NULL; 922 tcpstat.tcps_sc_recvcookie++; 923 } 924 925 /* 926 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 927 */ 928 if (th->th_ack != sc->sc_iss + 1) 929 return (0); 930 931 so = syncache_socket(sc, *sop, m); 932 if (so == NULL) { 933 #if 0 934 resetandabort: 935 /* XXXjlemon check this - is this correct? */ 936 tcp_respond(NULL, m, m, th, 937 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 938 #endif 939 m_freem(m); /* XXX only needed for above */ 940 tcpstat.tcps_sc_aborted++; 941 } else { 942 tcpstat.tcps_sc_completed++; 943 } 944 if (sch == NULL) 945 syncache_free(sc); 946 else 947 syncache_drop(sc, sch); 948 *sop = so; 949 return (1); 950 } 951 952 /* 953 * Given a LISTEN socket and an inbound SYN request, add 954 * this to the syn cache, and send back a segment: 955 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 956 * to the source. 957 * 958 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 959 * Doing so would require that we hold onto the data and deliver it 960 * to the application. However, if we are the target of a SYN-flood 961 * DoS attack, an attacker could send data which would eventually 962 * consume all available buffer space if it were ACKed. By not ACKing 963 * the data, we avoid this DoS scenario. 964 */ 965 int 966 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 967 struct socket **sop, struct mbuf *m) 968 { 969 struct tcp_syncache_percpu *syncache_percpu; 970 struct tcpcb *tp; 971 struct socket *so; 972 struct syncache *sc = NULL; 973 struct syncache_head *sch; 974 struct mbuf *ipopts = NULL; 975 int win; 976 977 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 978 so = *sop; 979 tp = sototcpcb(so); 980 981 /* 982 * Remember the IP options, if any. 983 */ 984 #ifdef INET6 985 if (!inc->inc_isipv6) 986 #endif 987 ipopts = ip_srcroute(m); 988 989 /* 990 * See if we already have an entry for this connection. 991 * If we do, resend the SYN,ACK, and reset the retransmit timer. 992 * 993 * XXX 994 * The syncache should be re-initialized with the contents 995 * of the new SYN which may have different options. 996 */ 997 sc = syncache_lookup(inc, &sch); 998 if (sc != NULL) { 999 tcpstat.tcps_sc_dupsyn++; 1000 if (ipopts) { 1001 /* 1002 * If we were remembering a previous source route, 1003 * forget it and use the new one we've been given. 1004 */ 1005 if (sc->sc_ipopts) 1006 m_free(sc->sc_ipopts); 1007 sc->sc_ipopts = ipopts; 1008 } 1009 /* 1010 * Update timestamp if present. 1011 */ 1012 if (sc->sc_flags & SCF_TIMESTAMP) 1013 sc->sc_tsrecent = to->to_tsval; 1014 1015 /* Just update the TOF_SACK_PERMITTED for now. */ 1016 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1017 sc->sc_flags |= SCF_SACK_PERMITTED; 1018 else 1019 sc->sc_flags &= ~SCF_SACK_PERMITTED; 1020 1021 /* 1022 * PCB may have changed, pick up new values. 1023 */ 1024 sc->sc_tp = tp; 1025 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1026 if (syncache_respond(sc, m) == 0) { 1027 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 1028 sc, sc_timerq); 1029 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 1030 tcpstat.tcps_sndacks++; 1031 tcpstat.tcps_sndtotal++; 1032 } 1033 *sop = NULL; 1034 return (1); 1035 } 1036 1037 /* 1038 * Fill in the syncache values. 1039 */ 1040 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1041 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1042 sc->sc_ipopts = ipopts; 1043 sc->sc_inc.inc_fport = inc->inc_fport; 1044 sc->sc_inc.inc_lport = inc->inc_lport; 1045 sc->sc_tp = tp; 1046 #ifdef INET6 1047 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1048 if (inc->inc_isipv6) { 1049 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1050 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1051 sc->sc_route6.ro_rt = NULL; 1052 } else 1053 #endif 1054 { 1055 sc->sc_inc.inc_faddr = inc->inc_faddr; 1056 sc->sc_inc.inc_laddr = inc->inc_laddr; 1057 sc->sc_route.ro_rt = NULL; 1058 } 1059 sc->sc_irs = th->th_seq; 1060 sc->sc_flags = 0; 1061 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 1062 if (tcp_syncookies) 1063 sc->sc_iss = syncookie_generate(sc); 1064 else 1065 sc->sc_iss = karc4random(); 1066 1067 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */ 1068 win = ssb_space(&so->so_rcv); 1069 win = imax(win, 0); 1070 win = imin(win, TCP_MAXWIN); 1071 sc->sc_wnd = win; 1072 1073 if (tcp_do_rfc1323) { 1074 /* 1075 * A timestamp received in a SYN makes 1076 * it ok to send timestamp requests and replies. 1077 */ 1078 if (to->to_flags & TOF_TS) { 1079 sc->sc_tsrecent = to->to_tsval; 1080 sc->sc_flags |= SCF_TIMESTAMP; 1081 } 1082 if (to->to_flags & TOF_SCALE) { 1083 int wscale = TCP_MIN_WINSHIFT; 1084 1085 /* Compute proper scaling value from buffer space */ 1086 while (wscale < TCP_MAX_WINSHIFT && 1087 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) { 1088 wscale++; 1089 } 1090 sc->sc_request_r_scale = wscale; 1091 sc->sc_requested_s_scale = to->to_requested_s_scale; 1092 sc->sc_flags |= SCF_WINSCALE; 1093 } 1094 } 1095 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1096 sc->sc_flags |= SCF_SACK_PERMITTED; 1097 if (tp->t_flags & TF_NOOPT) 1098 sc->sc_flags = SCF_NOOPT; 1099 #ifdef TCP_SIGNATURE 1100 /* 1101 * If listening socket requested TCP digests, and received SYN 1102 * contains the option, flag this in the syncache so that 1103 * syncache_respond() will do the right thing with the SYN+ACK. 1104 * XXX Currently we always record the option by default and will 1105 * attempt to use it in syncache_respond(). 1106 */ 1107 if (to->to_flags & TOF_SIGNATURE) 1108 sc->sc_flags = SCF_SIGNATURE; 1109 #endif /* TCP_SIGNATURE */ 1110 1111 if (syncache_respond(sc, m) == 0) { 1112 syncache_insert(sc, sch); 1113 tcpstat.tcps_sndacks++; 1114 tcpstat.tcps_sndtotal++; 1115 } else { 1116 syncache_free(sc); 1117 tcpstat.tcps_sc_dropped++; 1118 } 1119 *sop = NULL; 1120 return (1); 1121 } 1122 1123 static int 1124 syncache_respond(struct syncache *sc, struct mbuf *m) 1125 { 1126 u_int8_t *optp; 1127 int optlen, error; 1128 u_int16_t tlen, hlen, mssopt; 1129 struct ip *ip = NULL; 1130 struct rtentry *rt; 1131 struct tcphdr *th; 1132 struct ip6_hdr *ip6 = NULL; 1133 #ifdef INET6 1134 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1135 #else 1136 const boolean_t isipv6 = FALSE; 1137 #endif 1138 1139 if (isipv6) { 1140 rt = tcp_rtlookup6(&sc->sc_inc); 1141 if (rt != NULL) 1142 mssopt = rt->rt_ifp->if_mtu - 1143 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1144 else 1145 mssopt = tcp_v6mssdflt; 1146 hlen = sizeof(struct ip6_hdr); 1147 } else { 1148 rt = tcp_rtlookup(&sc->sc_inc); 1149 if (rt != NULL) 1150 mssopt = rt->rt_ifp->if_mtu - 1151 (sizeof(struct ip) + sizeof(struct tcphdr)); 1152 else 1153 mssopt = tcp_mssdflt; 1154 hlen = sizeof(struct ip); 1155 } 1156 1157 /* Compute the size of the TCP options. */ 1158 if (sc->sc_flags & SCF_NOOPT) { 1159 optlen = 0; 1160 } else { 1161 optlen = TCPOLEN_MAXSEG + 1162 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1163 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1164 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1165 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1166 #ifdef TCP_SIGNATURE 1167 optlen += ((sc->sc_flags & SCF_SIGNATURE) ? 1168 (TCPOLEN_SIGNATURE + 2) : 0); 1169 #endif /* TCP_SIGNATURE */ 1170 } 1171 tlen = hlen + sizeof(struct tcphdr) + optlen; 1172 1173 /* 1174 * XXX 1175 * assume that the entire packet will fit in a header mbuf 1176 */ 1177 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1178 1179 /* 1180 * XXX shouldn't this reuse the mbuf if possible ? 1181 * Create the IP+TCP header from scratch. 1182 */ 1183 if (m) 1184 m_freem(m); 1185 1186 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 1187 if (m == NULL) 1188 return (ENOBUFS); 1189 m->m_data += max_linkhdr; 1190 m->m_len = tlen; 1191 m->m_pkthdr.len = tlen; 1192 m->m_pkthdr.rcvif = NULL; 1193 1194 if (isipv6) { 1195 ip6 = mtod(m, struct ip6_hdr *); 1196 ip6->ip6_vfc = IPV6_VERSION; 1197 ip6->ip6_nxt = IPPROTO_TCP; 1198 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1199 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1200 ip6->ip6_plen = htons(tlen - hlen); 1201 /* ip6_hlim is set after checksum */ 1202 /* ip6_flow = ??? */ 1203 1204 th = (struct tcphdr *)(ip6 + 1); 1205 } else { 1206 ip = mtod(m, struct ip *); 1207 ip->ip_v = IPVERSION; 1208 ip->ip_hl = sizeof(struct ip) >> 2; 1209 ip->ip_len = tlen; 1210 ip->ip_id = 0; 1211 ip->ip_off = 0; 1212 ip->ip_sum = 0; 1213 ip->ip_p = IPPROTO_TCP; 1214 ip->ip_src = sc->sc_inc.inc_laddr; 1215 ip->ip_dst = sc->sc_inc.inc_faddr; 1216 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1217 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1218 1219 /* 1220 * See if we should do MTU discovery. Route lookups are 1221 * expensive, so we will only unset the DF bit if: 1222 * 1223 * 1) path_mtu_discovery is disabled 1224 * 2) the SCF_UNREACH flag has been set 1225 */ 1226 if (path_mtu_discovery 1227 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1228 ip->ip_off |= IP_DF; 1229 } 1230 1231 th = (struct tcphdr *)(ip + 1); 1232 } 1233 th->th_sport = sc->sc_inc.inc_lport; 1234 th->th_dport = sc->sc_inc.inc_fport; 1235 1236 th->th_seq = htonl(sc->sc_iss); 1237 th->th_ack = htonl(sc->sc_irs + 1); 1238 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1239 th->th_x2 = 0; 1240 th->th_flags = TH_SYN | TH_ACK; 1241 th->th_win = htons(sc->sc_wnd); 1242 th->th_urp = 0; 1243 1244 /* Tack on the TCP options. */ 1245 if (optlen == 0) 1246 goto no_options; 1247 optp = (u_int8_t *)(th + 1); 1248 *optp++ = TCPOPT_MAXSEG; 1249 *optp++ = TCPOLEN_MAXSEG; 1250 *optp++ = (mssopt >> 8) & 0xff; 1251 *optp++ = mssopt & 0xff; 1252 1253 if (sc->sc_flags & SCF_WINSCALE) { 1254 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1255 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1256 sc->sc_request_r_scale); 1257 optp += 4; 1258 } 1259 1260 if (sc->sc_flags & SCF_TIMESTAMP) { 1261 u_int32_t *lp = (u_int32_t *)(optp); 1262 1263 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1264 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1265 *lp++ = htonl(ticks); 1266 *lp = htonl(sc->sc_tsrecent); 1267 optp += TCPOLEN_TSTAMP_APPA; 1268 } 1269 1270 #ifdef TCP_SIGNATURE 1271 /* 1272 * Handle TCP-MD5 passive opener response. 1273 */ 1274 if (sc->sc_flags & SCF_SIGNATURE) { 1275 u_int8_t *bp = optp; 1276 int i; 1277 1278 *bp++ = TCPOPT_SIGNATURE; 1279 *bp++ = TCPOLEN_SIGNATURE; 1280 for (i = 0; i < TCP_SIGLEN; i++) 1281 *bp++ = 0; 1282 tcpsignature_compute(m, 0, optlen, 1283 optp + 2, IPSEC_DIR_OUTBOUND); 1284 *bp++ = TCPOPT_NOP; 1285 *bp++ = TCPOPT_EOL; 1286 optp += TCPOLEN_SIGNATURE + 2; 1287 } 1288 #endif /* TCP_SIGNATURE */ 1289 1290 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1291 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1292 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1293 } 1294 1295 no_options: 1296 if (isipv6) { 1297 struct route_in6 *ro6 = &sc->sc_route6; 1298 1299 th->th_sum = 0; 1300 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1301 ip6->ip6_hlim = in6_selecthlim(NULL, 1302 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1303 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1304 sc->sc_tp->t_inpcb); 1305 } else { 1306 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1307 htons(tlen - hlen + IPPROTO_TCP)); 1308 m->m_pkthdr.csum_flags = CSUM_TCP; 1309 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1310 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 1311 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb); 1312 } 1313 return (error); 1314 } 1315 1316 /* 1317 * cookie layers: 1318 * 1319 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1320 * | peer iss | 1321 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1322 * | 0 |(A)| | 1323 * (A): peer mss index 1324 */ 1325 1326 /* 1327 * The values below are chosen to minimize the size of the tcp_secret 1328 * table, as well as providing roughly a 16 second lifetime for the cookie. 1329 */ 1330 1331 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1332 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1333 1334 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1335 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1336 #define SYNCOOKIE_TIMEOUT \ 1337 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1338 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1339 1340 static struct { 1341 u_int32_t ts_secbits[4]; 1342 u_int ts_expire; 1343 } tcp_secret[SYNCOOKIE_NSECRETS]; 1344 1345 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1346 1347 static MD5_CTX syn_ctx; 1348 1349 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1350 1351 struct md5_add { 1352 u_int32_t laddr, faddr; 1353 u_int32_t secbits[4]; 1354 u_int16_t lport, fport; 1355 }; 1356 1357 #ifdef CTASSERT 1358 CTASSERT(sizeof(struct md5_add) == 28); 1359 #endif 1360 1361 /* 1362 * Consider the problem of a recreated (and retransmitted) cookie. If the 1363 * original SYN was accepted, the connection is established. The second 1364 * SYN is inflight, and if it arrives with an ISN that falls within the 1365 * receive window, the connection is killed. 1366 * 1367 * However, since cookies have other problems, this may not be worth 1368 * worrying about. 1369 */ 1370 1371 static u_int32_t 1372 syncookie_generate(struct syncache *sc) 1373 { 1374 u_int32_t md5_buffer[4]; 1375 u_int32_t data; 1376 int idx, i; 1377 struct md5_add add; 1378 #ifdef INET6 1379 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1380 #else 1381 const boolean_t isipv6 = FALSE; 1382 #endif 1383 1384 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1385 if (tcp_secret[idx].ts_expire < ticks) { 1386 for (i = 0; i < 4; i++) 1387 tcp_secret[idx].ts_secbits[i] = karc4random(); 1388 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1389 } 1390 for (data = NELEM(tcp_msstab) - 1; data > 0; data--) 1391 if (tcp_msstab[data] <= sc->sc_peer_mss) 1392 break; 1393 data = (data << SYNCOOKIE_WNDBITS) | idx; 1394 data ^= sc->sc_irs; /* peer's iss */ 1395 MD5Init(&syn_ctx); 1396 if (isipv6) { 1397 MD5Add(sc->sc_inc.inc6_laddr); 1398 MD5Add(sc->sc_inc.inc6_faddr); 1399 add.laddr = 0; 1400 add.faddr = 0; 1401 } else { 1402 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1403 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1404 } 1405 add.lport = sc->sc_inc.inc_lport; 1406 add.fport = sc->sc_inc.inc_fport; 1407 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1408 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1409 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1410 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1411 MD5Add(add); 1412 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1413 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1414 return (data); 1415 } 1416 1417 static struct syncache * 1418 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so) 1419 { 1420 u_int32_t md5_buffer[4]; 1421 struct syncache *sc; 1422 u_int32_t data; 1423 int wnd, idx; 1424 struct md5_add add; 1425 1426 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1427 idx = data & SYNCOOKIE_WNDMASK; 1428 if (tcp_secret[idx].ts_expire < ticks || 1429 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1430 return (NULL); 1431 MD5Init(&syn_ctx); 1432 #ifdef INET6 1433 if (inc->inc_isipv6) { 1434 MD5Add(inc->inc6_laddr); 1435 MD5Add(inc->inc6_faddr); 1436 add.laddr = 0; 1437 add.faddr = 0; 1438 } else 1439 #endif 1440 { 1441 add.laddr = inc->inc_laddr.s_addr; 1442 add.faddr = inc->inc_faddr.s_addr; 1443 } 1444 add.lport = inc->inc_lport; 1445 add.fport = inc->inc_fport; 1446 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1447 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1448 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1449 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1450 MD5Add(add); 1451 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1452 data ^= md5_buffer[0]; 1453 if (data & ~SYNCOOKIE_DATAMASK) 1454 return (NULL); 1455 data = data >> SYNCOOKIE_WNDBITS; 1456 1457 /* 1458 * Fill in the syncache values. 1459 * XXX duplicate code from syncache_add 1460 */ 1461 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1462 sc->sc_ipopts = NULL; 1463 sc->sc_inc.inc_fport = inc->inc_fport; 1464 sc->sc_inc.inc_lport = inc->inc_lport; 1465 #ifdef INET6 1466 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1467 if (inc->inc_isipv6) { 1468 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1469 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1470 sc->sc_route6.ro_rt = NULL; 1471 } else 1472 #endif 1473 { 1474 sc->sc_inc.inc_faddr = inc->inc_faddr; 1475 sc->sc_inc.inc_laddr = inc->inc_laddr; 1476 sc->sc_route.ro_rt = NULL; 1477 } 1478 sc->sc_irs = th->th_seq - 1; 1479 sc->sc_iss = th->th_ack - 1; 1480 wnd = ssb_space(&so->so_rcv); 1481 wnd = imax(wnd, 0); 1482 wnd = imin(wnd, TCP_MAXWIN); 1483 sc->sc_wnd = wnd; 1484 sc->sc_flags = 0; 1485 sc->sc_rxtslot = 0; 1486 sc->sc_peer_mss = tcp_msstab[data]; 1487 return (sc); 1488 } 1489