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 #include <net/netisr2.h> 94 95 #include <net/if.h> 96 #include <net/route.h> 97 98 #include <netinet/in.h> 99 #include <netinet/in_systm.h> 100 #include <netinet/ip.h> 101 #include <netinet/in_var.h> 102 #include <netinet/in_pcb.h> 103 #include <netinet/ip_var.h> 104 #include <netinet/ip6.h> 105 #ifdef INET6 106 #include <netinet/icmp6.h> 107 #include <netinet6/nd6.h> 108 #endif 109 #include <netinet6/ip6_var.h> 110 #include <netinet6/in6_pcb.h> 111 #include <netinet/tcp.h> 112 #include <netinet/tcp_fsm.h> 113 #include <netinet/tcp_seq.h> 114 #include <netinet/tcp_timer.h> 115 #include <netinet/tcp_timer2.h> 116 #include <netinet/tcp_var.h> 117 #include <netinet6/tcp6_var.h> 118 119 #ifdef IPSEC 120 #include <netinet6/ipsec.h> 121 #ifdef INET6 122 #include <netinet6/ipsec6.h> 123 #endif 124 #include <netproto/key/key.h> 125 #endif /*IPSEC*/ 126 127 #ifdef FAST_IPSEC 128 #include <netproto/ipsec/ipsec.h> 129 #ifdef INET6 130 #include <netproto/ipsec/ipsec6.h> 131 #endif 132 #include <netproto/ipsec/key.h> 133 #define IPSEC 134 #endif /*FAST_IPSEC*/ 135 136 static int tcp_syncookies = 1; 137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 138 &tcp_syncookies, 0, 139 "Use TCP SYN cookies if the syncache overflows"); 140 141 static void syncache_drop(struct syncache *, struct syncache_head *); 142 static void syncache_free(struct syncache *); 143 static void syncache_insert(struct syncache *, struct syncache_head *); 144 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 145 static int syncache_respond(struct syncache *, struct mbuf *); 146 static struct socket *syncache_socket(struct syncache *, struct socket *, 147 struct mbuf *); 148 static void syncache_timer(void *); 149 static u_int32_t syncookie_generate(struct syncache *); 150 static struct syncache *syncookie_lookup(struct in_conninfo *, 151 struct tcphdr *, struct socket *); 152 153 /* 154 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 155 * 4 retransmits corresponds to a timeout of (3 + 3 + 3 + 3 + 3 == 15) seconds 156 * or (1 + 1 + 2 + 4 + 8 == 16) seconds if RFC6298 is used, the odds are that 157 * the user has given up attempting to connect by then. 158 */ 159 #define SYNCACHE_MAXREXMTS 4 160 161 /* Arbitrary values */ 162 #define TCP_SYNCACHE_HASHSIZE 512 163 #define TCP_SYNCACHE_BUCKETLIMIT 30 164 165 struct netmsg_sc_timer { 166 struct netmsg_base base; 167 struct msgrec *nm_mrec; /* back pointer to containing msgrec */ 168 }; 169 170 struct msgrec { 171 struct netmsg_sc_timer msg; 172 lwkt_port_t port; /* constant after init */ 173 int slot; /* constant after init */ 174 }; 175 176 static void syncache_timer_handler(netmsg_t); 177 178 struct tcp_syncache { 179 u_int hashsize; 180 u_int hashmask; 181 u_int bucket_limit; 182 u_int cache_limit; 183 u_int rexmt_limit; 184 u_int hash_secret; 185 }; 186 static struct tcp_syncache tcp_syncache; 187 188 TAILQ_HEAD(syncache_list, syncache); 189 190 struct tcp_syncache_percpu { 191 struct syncache_head *hashbase; 192 u_int cache_count; 193 struct syncache_list timerq[SYNCACHE_MAXREXMTS + 1]; 194 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1]; 195 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1]; 196 }; 197 static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU]; 198 199 static struct lwkt_port syncache_null_rport; 200 201 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 202 203 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD, 204 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 205 206 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD, 207 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 208 209 /* XXX JH */ 210 #if 0 211 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 212 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 213 #endif 214 215 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD, 216 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 217 218 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 219 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 220 221 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 222 223 #define SYNCACHE_HASH(inc, mask) \ 224 ((tcp_syncache.hash_secret ^ \ 225 (inc)->inc_faddr.s_addr ^ \ 226 ((inc)->inc_faddr.s_addr >> 16) ^ \ 227 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 228 229 #define SYNCACHE_HASH6(inc, mask) \ 230 ((tcp_syncache.hash_secret ^ \ 231 (inc)->inc6_faddr.s6_addr32[0] ^ \ 232 (inc)->inc6_faddr.s6_addr32[3] ^ \ 233 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 234 235 #define ENDPTS_EQ(a, b) ( \ 236 (a)->ie_fport == (b)->ie_fport && \ 237 (a)->ie_lport == (b)->ie_lport && \ 238 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 239 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 240 ) 241 242 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 243 244 static __inline int 245 syncache_rto(int slot) 246 { 247 if (tcp_low_rtobase) 248 return (TCPTV_RTOBASE * tcp_syn_backoff_low[slot]); 249 else 250 return (TCPTV_RTOBASE * tcp_syn_backoff[slot]); 251 } 252 253 static __inline void 254 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu, 255 struct syncache *sc, int slot) 256 { 257 int rto; 258 259 if (slot > 0) { 260 /* 261 * Record the time that we spent in SYN|ACK 262 * retransmition. 263 * 264 * Needed by RFC3390 and RFC6298. 265 */ 266 sc->sc_rxtused += syncache_rto(slot - 1); 267 } 268 sc->sc_rxtslot = slot; 269 270 rto = syncache_rto(slot); 271 sc->sc_rxttime = ticks + rto; 272 273 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq); 274 if (!callout_active(&syncache_percpu->tt_timerq[slot])) { 275 callout_reset(&syncache_percpu->tt_timerq[slot], rto, 276 syncache_timer, &syncache_percpu->mrec[slot]); 277 } 278 } 279 280 static void 281 syncache_free(struct syncache *sc) 282 { 283 struct rtentry *rt; 284 #ifdef INET6 285 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 286 #else 287 const boolean_t isipv6 = FALSE; 288 #endif 289 290 if (sc->sc_ipopts) 291 m_free(sc->sc_ipopts); 292 293 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt; 294 if (rt != NULL) { 295 /* 296 * If this is the only reference to a protocol-cloned 297 * route, remove it immediately. 298 */ 299 if ((rt->rt_flags & RTF_WASCLONED) && rt->rt_refcnt == 1) 300 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 301 rt_mask(rt), rt->rt_flags, NULL); 302 RTFREE(rt); 303 } 304 kfree(sc, M_SYNCACHE); 305 } 306 307 void 308 syncache_init(void) 309 { 310 int i, cpu; 311 312 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 313 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 314 tcp_syncache.cache_limit = 315 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 316 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 317 tcp_syncache.hash_secret = karc4random(); 318 319 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 320 &tcp_syncache.hashsize); 321 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 322 &tcp_syncache.cache_limit); 323 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 324 &tcp_syncache.bucket_limit); 325 if (!powerof2(tcp_syncache.hashsize)) { 326 kprintf("WARNING: syncache hash size is not a power of 2.\n"); 327 tcp_syncache.hashsize = 512; /* safe default */ 328 } 329 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 330 331 lwkt_initport_replyonly_null(&syncache_null_rport); 332 333 for (cpu = 0; cpu < ncpus2; cpu++) { 334 struct tcp_syncache_percpu *syncache_percpu; 335 336 syncache_percpu = &tcp_syncache_percpu[cpu]; 337 /* Allocate the hash table. */ 338 syncache_percpu->hashbase = kmalloc(tcp_syncache.hashsize * sizeof(struct syncache_head), 339 M_SYNCACHE, M_WAITOK); 340 341 /* Initialize the hash buckets. */ 342 for (i = 0; i < tcp_syncache.hashsize; i++) { 343 struct syncache_head *bucket; 344 345 bucket = &syncache_percpu->hashbase[i]; 346 TAILQ_INIT(&bucket->sch_bucket); 347 bucket->sch_length = 0; 348 } 349 350 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 351 /* Initialize the timer queues. */ 352 TAILQ_INIT(&syncache_percpu->timerq[i]); 353 callout_init_mp(&syncache_percpu->tt_timerq[i]); 354 355 syncache_percpu->mrec[i].slot = i; 356 syncache_percpu->mrec[i].port = netisr_cpuport(cpu); 357 syncache_percpu->mrec[i].msg.nm_mrec = 358 &syncache_percpu->mrec[i]; 359 netmsg_init(&syncache_percpu->mrec[i].msg.base, 360 NULL, &syncache_null_rport, 361 0, syncache_timer_handler); 362 } 363 } 364 } 365 366 static void 367 syncache_insert(struct syncache *sc, struct syncache_head *sch) 368 { 369 struct tcp_syncache_percpu *syncache_percpu; 370 struct syncache *sc2; 371 int i; 372 373 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 374 375 /* 376 * Make sure that we don't overflow the per-bucket 377 * limit or the total cache size limit. 378 */ 379 if (sch->sch_length >= tcp_syncache.bucket_limit) { 380 /* 381 * The bucket is full, toss the oldest element. 382 */ 383 sc2 = TAILQ_FIRST(&sch->sch_bucket); 384 sc2->sc_tp->ts_recent = ticks; 385 syncache_drop(sc2, sch); 386 tcpstat.tcps_sc_bucketoverflow++; 387 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) { 388 /* 389 * The cache is full. Toss the oldest entry in the 390 * entire cache. This is the front entry in the 391 * first non-empty timer queue with the largest 392 * timeout value. 393 */ 394 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 395 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]); 396 while (sc2 && (sc2->sc_flags & SCF_MARKER)) 397 sc2 = TAILQ_NEXT(sc2, sc_timerq); 398 if (sc2 != NULL) 399 break; 400 } 401 sc2->sc_tp->ts_recent = ticks; 402 syncache_drop(sc2, NULL); 403 tcpstat.tcps_sc_cacheoverflow++; 404 } 405 406 /* Initialize the entry's timer. */ 407 syncache_timeout(syncache_percpu, sc, 0); 408 409 /* Put it into the bucket. */ 410 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 411 sch->sch_length++; 412 syncache_percpu->cache_count++; 413 tcpstat.tcps_sc_added++; 414 } 415 416 void 417 syncache_destroy(struct tcpcb *tp) 418 { 419 struct tcp_syncache_percpu *syncache_percpu; 420 struct syncache_head *bucket; 421 struct syncache *sc; 422 int i; 423 424 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 425 sc = NULL; 426 427 for (i = 0; i < tcp_syncache.hashsize; i++) { 428 bucket = &syncache_percpu->hashbase[i]; 429 TAILQ_FOREACH(sc, &bucket->sch_bucket, sc_hash) { 430 if (sc->sc_tp == tp) 431 sc->sc_tp = NULL; 432 } 433 } 434 } 435 436 static void 437 syncache_drop(struct syncache *sc, struct syncache_head *sch) 438 { 439 struct tcp_syncache_percpu *syncache_percpu; 440 #ifdef INET6 441 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 442 #else 443 const boolean_t isipv6 = FALSE; 444 #endif 445 446 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 447 448 if (sch == NULL) { 449 if (isipv6) { 450 sch = &syncache_percpu->hashbase[ 451 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 452 } else { 453 sch = &syncache_percpu->hashbase[ 454 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 455 } 456 } 457 458 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 459 sch->sch_length--; 460 syncache_percpu->cache_count--; 461 462 /* 463 * Cleanup 464 */ 465 if (sc->sc_tp) 466 sc->sc_tp = NULL; 467 468 /* 469 * Remove the entry from the syncache timer/timeout queue. Note 470 * that we do not try to stop any running timer since we do not know 471 * whether the timer's message is in-transit or not. Since timeouts 472 * are fairly long, taking an unneeded callout does not detrimentally 473 * effect performance. 474 */ 475 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq); 476 477 syncache_free(sc); 478 } 479 480 /* 481 * Place a timeout message on the TCP thread's message queue. 482 * This routine runs in soft interrupt context. 483 * 484 * An invariant is for this routine to be called, the callout must 485 * have been active. Note that the callout is not deactivated until 486 * after the message has been processed in syncache_timer_handler() below. 487 */ 488 static void 489 syncache_timer(void *p) 490 { 491 struct netmsg_sc_timer *msg = p; 492 493 lwkt_sendmsg(msg->nm_mrec->port, &msg->base.lmsg); 494 } 495 496 /* 497 * Service a timer message queued by timer expiration. 498 * This routine runs in the TCP protocol thread. 499 * 500 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 501 * If we have retransmitted an entry the maximum number of times, expire it. 502 * 503 * When we finish processing timed-out entries, we restart the timer if there 504 * are any entries still on the queue and deactivate it otherwise. Only after 505 * a timer has been deactivated here can it be restarted by syncache_timeout(). 506 */ 507 static void 508 syncache_timer_handler(netmsg_t msg) 509 { 510 struct tcp_syncache_percpu *syncache_percpu; 511 struct syncache *sc; 512 struct syncache marker; 513 struct syncache_list *list; 514 struct inpcb *inp; 515 int slot; 516 517 slot = ((struct netmsg_sc_timer *)msg)->nm_mrec->slot; 518 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 519 520 list = &syncache_percpu->timerq[slot]; 521 522 /* 523 * Use a marker to keep our place in the scan. syncache_drop() 524 * can block and cause any next pointer we cache to become stale. 525 */ 526 marker.sc_flags = SCF_MARKER; 527 TAILQ_INSERT_HEAD(list, &marker, sc_timerq); 528 529 while ((sc = TAILQ_NEXT(&marker, sc_timerq)) != NULL) { 530 /* 531 * Move the marker. 532 */ 533 TAILQ_REMOVE(list, &marker, sc_timerq); 534 TAILQ_INSERT_AFTER(list, sc, &marker, sc_timerq); 535 536 if (sc->sc_flags & SCF_MARKER) 537 continue; 538 539 if (ticks < sc->sc_rxttime) 540 break; /* finished because timerq sorted by time */ 541 if (sc->sc_tp == NULL) { 542 syncache_drop(sc, NULL); 543 tcpstat.tcps_sc_stale++; 544 continue; 545 } 546 inp = sc->sc_tp->t_inpcb; 547 if (slot == SYNCACHE_MAXREXMTS || 548 slot >= tcp_syncache.rexmt_limit || 549 inp == NULL || 550 inp->inp_gencnt != sc->sc_inp_gencnt) { 551 syncache_drop(sc, NULL); 552 tcpstat.tcps_sc_stale++; 553 continue; 554 } 555 /* 556 * syncache_respond() may call back into the syncache to 557 * to modify another entry, so do not obtain the next 558 * entry on the timer chain until it has completed. 559 */ 560 syncache_respond(sc, NULL); 561 tcpstat.tcps_sc_retransmitted++; 562 TAILQ_REMOVE(list, sc, sc_timerq); 563 syncache_timeout(syncache_percpu, sc, slot + 1); 564 } 565 TAILQ_REMOVE(list, &marker, sc_timerq); 566 567 if (sc != NULL) { 568 callout_reset(&syncache_percpu->tt_timerq[slot], 569 sc->sc_rxttime - ticks, syncache_timer, 570 &syncache_percpu->mrec[slot]); 571 } else { 572 callout_deactivate(&syncache_percpu->tt_timerq[slot]); 573 } 574 lwkt_replymsg(&msg->base.lmsg, 0); 575 } 576 577 /* 578 * Find an entry in the syncache. 579 */ 580 struct syncache * 581 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 582 { 583 struct tcp_syncache_percpu *syncache_percpu; 584 struct syncache *sc; 585 struct syncache_head *sch; 586 587 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 588 #ifdef INET6 589 if (inc->inc_isipv6) { 590 sch = &syncache_percpu->hashbase[ 591 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 592 *schp = sch; 593 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) 594 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 595 return (sc); 596 } else 597 #endif 598 { 599 sch = &syncache_percpu->hashbase[ 600 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 601 *schp = sch; 602 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 603 #ifdef INET6 604 if (sc->sc_inc.inc_isipv6) 605 continue; 606 #endif 607 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 608 return (sc); 609 } 610 } 611 return (NULL); 612 } 613 614 /* 615 * This function is called when we get a RST for a 616 * non-existent connection, so that we can see if the 617 * connection is in the syn cache. If it is, zap it. 618 */ 619 void 620 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 621 { 622 struct syncache *sc; 623 struct syncache_head *sch; 624 625 sc = syncache_lookup(inc, &sch); 626 if (sc == NULL) { 627 return; 628 } 629 /* 630 * If the RST bit is set, check the sequence number to see 631 * if this is a valid reset segment. 632 * RFC 793 page 37: 633 * In all states except SYN-SENT, all reset (RST) segments 634 * are validated by checking their SEQ-fields. A reset is 635 * valid if its sequence number is in the window. 636 * 637 * The sequence number in the reset segment is normally an 638 * echo of our outgoing acknowlegement numbers, but some hosts 639 * send a reset with the sequence number at the rightmost edge 640 * of our receive window, and we have to handle this case. 641 */ 642 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 643 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 644 syncache_drop(sc, sch); 645 tcpstat.tcps_sc_reset++; 646 } 647 } 648 649 void 650 syncache_badack(struct in_conninfo *inc) 651 { 652 struct syncache *sc; 653 struct syncache_head *sch; 654 655 sc = syncache_lookup(inc, &sch); 656 if (sc != NULL) { 657 syncache_drop(sc, sch); 658 tcpstat.tcps_sc_badack++; 659 } 660 } 661 662 void 663 syncache_unreach(struct in_conninfo *inc, struct tcphdr *th) 664 { 665 struct syncache *sc; 666 struct syncache_head *sch; 667 668 /* we are called at splnet() here */ 669 sc = syncache_lookup(inc, &sch); 670 if (sc == NULL) 671 return; 672 673 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 674 if (ntohl(th->th_seq) != sc->sc_iss) 675 return; 676 677 /* 678 * If we've rertransmitted 3 times and this is our second error, 679 * we remove the entry. Otherwise, we allow it to continue on. 680 * This prevents us from incorrectly nuking an entry during a 681 * spurious network outage. 682 * 683 * See tcp_notify(). 684 */ 685 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 686 sc->sc_flags |= SCF_UNREACH; 687 return; 688 } 689 syncache_drop(sc, sch); 690 tcpstat.tcps_sc_unreach++; 691 } 692 693 /* 694 * Build a new TCP socket structure from a syncache entry. 695 * 696 * This is called from the context of the SYN+ACK 697 */ 698 static struct socket * 699 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 700 { 701 struct inpcb *inp = NULL, *linp; 702 struct socket *so; 703 struct tcpcb *tp, *ltp; 704 lwkt_port_t port; 705 #ifdef INET6 706 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 707 #else 708 const boolean_t isipv6 = FALSE; 709 #endif 710 struct sockaddr_in sin_faddr; 711 struct sockaddr_in6 sin6_faddr; 712 struct sockaddr *faddr; 713 714 if (isipv6) { 715 faddr = (struct sockaddr *)&sin6_faddr; 716 sin6_faddr.sin6_family = AF_INET6; 717 sin6_faddr.sin6_len = sizeof(sin6_faddr); 718 sin6_faddr.sin6_addr = sc->sc_inc.inc6_faddr; 719 sin6_faddr.sin6_port = sc->sc_inc.inc_fport; 720 sin6_faddr.sin6_flowinfo = sin6_faddr.sin6_scope_id = 0; 721 } else { 722 faddr = (struct sockaddr *)&sin_faddr; 723 sin_faddr.sin_family = AF_INET; 724 sin_faddr.sin_len = sizeof(sin_faddr); 725 sin_faddr.sin_addr = sc->sc_inc.inc_faddr; 726 sin_faddr.sin_port = sc->sc_inc.inc_fport; 727 bzero(sin_faddr.sin_zero, sizeof(sin_faddr.sin_zero)); 728 } 729 730 /* 731 * Ok, create the full blown connection, and set things up 732 * as they would have been set up if we had created the 733 * connection when the SYN arrived. If we can't create 734 * the connection, abort it. 735 * 736 * Set the protocol processing port for the socket to the current 737 * port (that the connection came in on). 738 */ 739 so = sonewconn_faddr(lso, SS_ISCONNECTED, faddr); 740 if (so == NULL) { 741 /* 742 * Drop the connection; we will send a RST if the peer 743 * retransmits the ACK, 744 */ 745 tcpstat.tcps_listendrop++; 746 goto abort; 747 } 748 749 /* 750 * Insert new socket into hash list. 751 */ 752 inp = so->so_pcb; 753 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 754 if (isipv6) { 755 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 756 } else { 757 #ifdef INET6 758 inp->inp_vflag &= ~INP_IPV6; 759 inp->inp_vflag |= INP_IPV4; 760 inp->inp_flags &= ~IN6P_IPV6_V6ONLY; 761 #endif 762 inp->inp_laddr = sc->sc_inc.inc_laddr; 763 } 764 inp->inp_lport = sc->sc_inc.inc_lport; 765 if (in_pcbinsporthash(inp) != 0) { 766 /* 767 * Undo the assignments above if we failed to 768 * put the PCB on the hash lists. 769 */ 770 if (isipv6) 771 inp->in6p_laddr = kin6addr_any; 772 else 773 inp->inp_laddr.s_addr = INADDR_ANY; 774 inp->inp_lport = 0; 775 goto abort; 776 } 777 linp = lso->so_pcb; 778 #ifdef IPSEC 779 /* copy old policy into new socket's */ 780 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 781 kprintf("syncache_expand: could not copy policy\n"); 782 #endif 783 if (isipv6) { 784 struct in6_addr laddr6; 785 /* 786 * Inherit socket options from the listening socket. 787 * Note that in6p_inputopts are not (and should not be) 788 * copied, since it stores previously received options and is 789 * used to detect if each new option is different than the 790 * previous one and hence should be passed to a user. 791 * If we copied in6p_inputopts, a user would not be able to 792 * receive options just after calling the accept system call. 793 */ 794 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 795 if (linp->in6p_outputopts) 796 inp->in6p_outputopts = 797 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 798 inp->in6p_route = sc->sc_route6; 799 sc->sc_route6.ro_rt = NULL; 800 801 laddr6 = inp->in6p_laddr; 802 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 803 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 804 if (in6_pcbconnect(inp, faddr, &thread0)) { 805 inp->in6p_laddr = laddr6; 806 goto abort; 807 } 808 } else { 809 struct in_addr laddr; 810 811 inp->inp_options = ip_srcroute(m); 812 if (inp->inp_options == NULL) { 813 inp->inp_options = sc->sc_ipopts; 814 sc->sc_ipopts = NULL; 815 } 816 inp->inp_route = sc->sc_route; 817 sc->sc_route.ro_rt = NULL; 818 819 laddr = inp->inp_laddr; 820 if (inp->inp_laddr.s_addr == INADDR_ANY) 821 inp->inp_laddr = sc->sc_inc.inc_laddr; 822 if (in_pcbconnect(inp, faddr, &thread0)) { 823 inp->inp_laddr = laddr; 824 goto abort; 825 } 826 } 827 828 /* 829 * The current port should be in the context of the SYN+ACK and 830 * so should match the tcp address port. 831 * 832 * XXX we may be running on the netisr thread instead of a tcp 833 * thread, in which case port will not match 834 * curthread->td_msgport. 835 */ 836 if (isipv6) { 837 port = tcp6_addrport(); 838 } else { 839 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport, 840 inp->inp_laddr.s_addr, inp->inp_lport); 841 } 842 if (port != &curthread->td_msgport) { 843 print_backtrace(-1); 844 kprintf("TCP PORT MISMATCH %p vs %p\n", 845 port, &curthread->td_msgport); 846 } 847 /*KKASSERT(port == &curthread->td_msgport);*/ 848 849 tp = intotcpcb(inp); 850 tp->t_state = TCPS_SYN_RECEIVED; 851 tp->iss = sc->sc_iss; 852 tp->irs = sc->sc_irs; 853 tcp_rcvseqinit(tp); 854 tcp_sendseqinit(tp); 855 tp->snd_wnd = sc->sc_sndwnd; 856 tp->snd_wl1 = sc->sc_irs; 857 tp->rcv_up = sc->sc_irs + 1; 858 tp->rcv_wnd = sc->sc_wnd; 859 tp->rcv_adv += tp->rcv_wnd; 860 861 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 862 if (sc->sc_flags & SCF_NOOPT) 863 tp->t_flags |= TF_NOOPT; 864 if (sc->sc_flags & SCF_WINSCALE) { 865 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 866 tp->snd_scale = sc->sc_requested_s_scale; 867 tp->request_r_scale = sc->sc_request_r_scale; 868 } 869 if (sc->sc_flags & SCF_TIMESTAMP) { 870 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 871 tp->ts_recent = sc->sc_tsrecent; 872 tp->ts_recent_age = ticks; 873 } 874 if (sc->sc_flags & SCF_SACK_PERMITTED) 875 tp->t_flags |= TF_SACK_PERMITTED; 876 877 #ifdef TCP_SIGNATURE 878 if (sc->sc_flags & SCF_SIGNATURE) 879 tp->t_flags |= TF_SIGNATURE; 880 #endif /* TCP_SIGNATURE */ 881 882 tp->t_rxtsyn = sc->sc_rxtused; 883 tcp_mss(tp, sc->sc_peer_mss); 884 885 /* 886 * Inherit some properties from the listen socket 887 */ 888 ltp = intotcpcb(linp); 889 tp->t_keepinit = ltp->t_keepinit; 890 tp->t_keepidle = ltp->t_keepidle; 891 tp->t_keepintvl = ltp->t_keepintvl; 892 tp->t_keepcnt = ltp->t_keepcnt; 893 tp->t_maxidle = ltp->t_maxidle; 894 895 tcp_create_timermsg(tp, port); 896 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep); 897 898 tcpstat.tcps_accepts++; 899 return (so); 900 901 abort: 902 if (so != NULL) 903 soabort_oncpu(so); 904 return (NULL); 905 } 906 907 /* 908 * This function gets called when we receive an ACK for a 909 * socket in the LISTEN state. We look up the connection 910 * in the syncache, and if its there, we pull it out of 911 * the cache and turn it into a full-blown connection in 912 * the SYN-RECEIVED state. 913 */ 914 int 915 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop, 916 struct mbuf *m) 917 { 918 struct syncache *sc; 919 struct syncache_head *sch; 920 struct socket *so; 921 922 sc = syncache_lookup(inc, &sch); 923 if (sc == NULL) { 924 /* 925 * There is no syncache entry, so see if this ACK is 926 * a returning syncookie. To do this, first: 927 * A. See if this socket has had a syncache entry dropped in 928 * the past. We don't want to accept a bogus syncookie 929 * if we've never received a SYN. 930 * B. check that the syncookie is valid. If it is, then 931 * cobble up a fake syncache entry, and return. 932 */ 933 if (!tcp_syncookies) 934 return (0); 935 sc = syncookie_lookup(inc, th, *sop); 936 if (sc == NULL) 937 return (0); 938 sch = NULL; 939 tcpstat.tcps_sc_recvcookie++; 940 } 941 942 /* 943 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 944 */ 945 if (th->th_ack != sc->sc_iss + 1) 946 return (0); 947 948 so = syncache_socket(sc, *sop, m); 949 if (so == NULL) { 950 #if 0 951 resetandabort: 952 /* XXXjlemon check this - is this correct? */ 953 tcp_respond(NULL, m, m, th, 954 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 955 #endif 956 m_freem(m); /* XXX only needed for above */ 957 tcpstat.tcps_sc_aborted++; 958 } else { 959 tcpstat.tcps_sc_completed++; 960 } 961 if (sch == NULL) 962 syncache_free(sc); 963 else 964 syncache_drop(sc, sch); 965 *sop = so; 966 return (1); 967 } 968 969 /* 970 * Given a LISTEN socket and an inbound SYN request, add 971 * this to the syn cache, and send back a segment: 972 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 973 * to the source. 974 * 975 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 976 * Doing so would require that we hold onto the data and deliver it 977 * to the application. However, if we are the target of a SYN-flood 978 * DoS attack, an attacker could send data which would eventually 979 * consume all available buffer space if it were ACKed. By not ACKing 980 * the data, we avoid this DoS scenario. 981 */ 982 int 983 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 984 struct socket *so, struct mbuf *m) 985 { 986 struct tcp_syncache_percpu *syncache_percpu; 987 struct tcpcb *tp; 988 struct syncache *sc = NULL; 989 struct syncache_head *sch; 990 struct mbuf *ipopts = NULL; 991 int win; 992 993 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 994 tp = sototcpcb(so); 995 996 /* 997 * Remember the IP options, if any. 998 */ 999 #ifdef INET6 1000 if (!inc->inc_isipv6) 1001 #endif 1002 ipopts = ip_srcroute(m); 1003 1004 /* 1005 * See if we already have an entry for this connection. 1006 * If we do, resend the SYN,ACK, and reset the retransmit timer. 1007 * 1008 * XXX 1009 * The syncache should be re-initialized with the contents 1010 * of the new SYN which may have different options. 1011 */ 1012 sc = syncache_lookup(inc, &sch); 1013 if (sc != NULL) { 1014 tcpstat.tcps_sc_dupsyn++; 1015 if (ipopts) { 1016 /* 1017 * If we were remembering a previous source route, 1018 * forget it and use the new one we've been given. 1019 */ 1020 if (sc->sc_ipopts) 1021 m_free(sc->sc_ipopts); 1022 sc->sc_ipopts = ipopts; 1023 } 1024 /* 1025 * Update timestamp if present. 1026 */ 1027 if (sc->sc_flags & SCF_TIMESTAMP) 1028 sc->sc_tsrecent = to->to_tsval; 1029 1030 /* Just update the TOF_SACK_PERMITTED for now. */ 1031 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1032 sc->sc_flags |= SCF_SACK_PERMITTED; 1033 else 1034 sc->sc_flags &= ~SCF_SACK_PERMITTED; 1035 1036 /* Update initial send window */ 1037 sc->sc_sndwnd = th->th_win; 1038 1039 /* 1040 * PCB may have changed, pick up new values. 1041 */ 1042 sc->sc_tp = tp; 1043 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1044 if (syncache_respond(sc, m) == 0) { 1045 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 1046 sc, sc_timerq); 1047 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 1048 tcpstat.tcps_sndacks++; 1049 tcpstat.tcps_sndtotal++; 1050 } 1051 return (1); 1052 } 1053 1054 /* 1055 * Fill in the syncache values. 1056 */ 1057 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1058 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1059 sc->sc_ipopts = ipopts; 1060 sc->sc_inc.inc_fport = inc->inc_fport; 1061 sc->sc_inc.inc_lport = inc->inc_lport; 1062 sc->sc_tp = tp; 1063 #ifdef INET6 1064 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1065 if (inc->inc_isipv6) { 1066 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1067 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1068 sc->sc_route6.ro_rt = NULL; 1069 } else 1070 #endif 1071 { 1072 sc->sc_inc.inc_faddr = inc->inc_faddr; 1073 sc->sc_inc.inc_laddr = inc->inc_laddr; 1074 sc->sc_route.ro_rt = NULL; 1075 } 1076 sc->sc_irs = th->th_seq; 1077 sc->sc_flags = 0; 1078 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 1079 if (tcp_syncookies) 1080 sc->sc_iss = syncookie_generate(sc); 1081 else 1082 sc->sc_iss = karc4random(); 1083 1084 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */ 1085 win = ssb_space(&so->so_rcv); 1086 win = imax(win, 0); 1087 win = imin(win, TCP_MAXWIN); 1088 sc->sc_wnd = win; 1089 1090 if (tcp_do_rfc1323) { 1091 /* 1092 * A timestamp received in a SYN makes 1093 * it ok to send timestamp requests and replies. 1094 */ 1095 if (to->to_flags & TOF_TS) { 1096 sc->sc_tsrecent = to->to_tsval; 1097 sc->sc_flags |= SCF_TIMESTAMP; 1098 } 1099 if (to->to_flags & TOF_SCALE) { 1100 int wscale = TCP_MIN_WINSHIFT; 1101 1102 /* Compute proper scaling value from buffer space */ 1103 while (wscale < TCP_MAX_WINSHIFT && 1104 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) { 1105 wscale++; 1106 } 1107 sc->sc_request_r_scale = wscale; 1108 sc->sc_requested_s_scale = to->to_requested_s_scale; 1109 sc->sc_flags |= SCF_WINSCALE; 1110 } 1111 } 1112 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1113 sc->sc_flags |= SCF_SACK_PERMITTED; 1114 if (tp->t_flags & TF_NOOPT) 1115 sc->sc_flags = SCF_NOOPT; 1116 #ifdef TCP_SIGNATURE 1117 /* 1118 * If listening socket requested TCP digests, and received SYN 1119 * contains the option, flag this in the syncache so that 1120 * syncache_respond() will do the right thing with the SYN+ACK. 1121 * XXX Currently we always record the option by default and will 1122 * attempt to use it in syncache_respond(). 1123 */ 1124 if (to->to_flags & TOF_SIGNATURE) 1125 sc->sc_flags = SCF_SIGNATURE; 1126 #endif /* TCP_SIGNATURE */ 1127 sc->sc_sndwnd = th->th_win; 1128 1129 if (syncache_respond(sc, m) == 0) { 1130 syncache_insert(sc, sch); 1131 tcpstat.tcps_sndacks++; 1132 tcpstat.tcps_sndtotal++; 1133 } else { 1134 syncache_free(sc); 1135 tcpstat.tcps_sc_dropped++; 1136 } 1137 return (1); 1138 } 1139 1140 static int 1141 syncache_respond(struct syncache *sc, struct mbuf *m) 1142 { 1143 u_int8_t *optp; 1144 int optlen, error; 1145 u_int16_t tlen, hlen, mssopt; 1146 struct ip *ip = NULL; 1147 struct rtentry *rt; 1148 struct tcphdr *th; 1149 struct ip6_hdr *ip6 = NULL; 1150 #ifdef INET6 1151 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1152 #else 1153 const boolean_t isipv6 = FALSE; 1154 #endif 1155 1156 if (isipv6) { 1157 rt = tcp_rtlookup6(&sc->sc_inc); 1158 if (rt != NULL) 1159 mssopt = rt->rt_ifp->if_mtu - 1160 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1161 else 1162 mssopt = tcp_v6mssdflt; 1163 hlen = sizeof(struct ip6_hdr); 1164 } else { 1165 rt = tcp_rtlookup(&sc->sc_inc); 1166 if (rt != NULL) 1167 mssopt = rt->rt_ifp->if_mtu - 1168 (sizeof(struct ip) + sizeof(struct tcphdr)); 1169 else 1170 mssopt = tcp_mssdflt; 1171 hlen = sizeof(struct ip); 1172 } 1173 1174 /* Compute the size of the TCP options. */ 1175 if (sc->sc_flags & SCF_NOOPT) { 1176 optlen = 0; 1177 } else { 1178 optlen = TCPOLEN_MAXSEG + 1179 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1180 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1181 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1182 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1183 #ifdef TCP_SIGNATURE 1184 optlen += ((sc->sc_flags & SCF_SIGNATURE) ? 1185 (TCPOLEN_SIGNATURE + 2) : 0); 1186 #endif /* TCP_SIGNATURE */ 1187 } 1188 tlen = hlen + sizeof(struct tcphdr) + optlen; 1189 1190 /* 1191 * XXX 1192 * assume that the entire packet will fit in a header mbuf 1193 */ 1194 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1195 1196 /* 1197 * XXX shouldn't this reuse the mbuf if possible ? 1198 * Create the IP+TCP header from scratch. 1199 */ 1200 if (m) 1201 m_freem(m); 1202 1203 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 1204 if (m == NULL) 1205 return (ENOBUFS); 1206 m->m_data += max_linkhdr; 1207 m->m_len = tlen; 1208 m->m_pkthdr.len = tlen; 1209 m->m_pkthdr.rcvif = NULL; 1210 1211 if (isipv6) { 1212 ip6 = mtod(m, struct ip6_hdr *); 1213 ip6->ip6_vfc = IPV6_VERSION; 1214 ip6->ip6_nxt = IPPROTO_TCP; 1215 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1216 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1217 ip6->ip6_plen = htons(tlen - hlen); 1218 /* ip6_hlim is set after checksum */ 1219 /* ip6_flow = ??? */ 1220 1221 th = (struct tcphdr *)(ip6 + 1); 1222 } else { 1223 ip = mtod(m, struct ip *); 1224 ip->ip_v = IPVERSION; 1225 ip->ip_hl = sizeof(struct ip) >> 2; 1226 ip->ip_len = tlen; 1227 ip->ip_id = 0; 1228 ip->ip_off = 0; 1229 ip->ip_sum = 0; 1230 ip->ip_p = IPPROTO_TCP; 1231 ip->ip_src = sc->sc_inc.inc_laddr; 1232 ip->ip_dst = sc->sc_inc.inc_faddr; 1233 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1234 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1235 1236 /* 1237 * See if we should do MTU discovery. Route lookups are 1238 * expensive, so we will only unset the DF bit if: 1239 * 1240 * 1) path_mtu_discovery is disabled 1241 * 2) the SCF_UNREACH flag has been set 1242 */ 1243 if (path_mtu_discovery 1244 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1245 ip->ip_off |= IP_DF; 1246 } 1247 1248 th = (struct tcphdr *)(ip + 1); 1249 } 1250 th->th_sport = sc->sc_inc.inc_lport; 1251 th->th_dport = sc->sc_inc.inc_fport; 1252 1253 th->th_seq = htonl(sc->sc_iss); 1254 th->th_ack = htonl(sc->sc_irs + 1); 1255 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1256 th->th_x2 = 0; 1257 th->th_flags = TH_SYN | TH_ACK; 1258 th->th_win = htons(sc->sc_wnd); 1259 th->th_urp = 0; 1260 1261 /* Tack on the TCP options. */ 1262 if (optlen == 0) 1263 goto no_options; 1264 optp = (u_int8_t *)(th + 1); 1265 *optp++ = TCPOPT_MAXSEG; 1266 *optp++ = TCPOLEN_MAXSEG; 1267 *optp++ = (mssopt >> 8) & 0xff; 1268 *optp++ = mssopt & 0xff; 1269 1270 if (sc->sc_flags & SCF_WINSCALE) { 1271 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1272 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1273 sc->sc_request_r_scale); 1274 optp += 4; 1275 } 1276 1277 if (sc->sc_flags & SCF_TIMESTAMP) { 1278 u_int32_t *lp = (u_int32_t *)(optp); 1279 1280 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1281 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1282 *lp++ = htonl(ticks); 1283 *lp = htonl(sc->sc_tsrecent); 1284 optp += TCPOLEN_TSTAMP_APPA; 1285 } 1286 1287 #ifdef TCP_SIGNATURE 1288 /* 1289 * Handle TCP-MD5 passive opener response. 1290 */ 1291 if (sc->sc_flags & SCF_SIGNATURE) { 1292 u_int8_t *bp = optp; 1293 int i; 1294 1295 *bp++ = TCPOPT_SIGNATURE; 1296 *bp++ = TCPOLEN_SIGNATURE; 1297 for (i = 0; i < TCP_SIGLEN; i++) 1298 *bp++ = 0; 1299 tcpsignature_compute(m, 0, optlen, 1300 optp + 2, IPSEC_DIR_OUTBOUND); 1301 *bp++ = TCPOPT_NOP; 1302 *bp++ = TCPOPT_EOL; 1303 optp += TCPOLEN_SIGNATURE + 2; 1304 } 1305 #endif /* TCP_SIGNATURE */ 1306 1307 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1308 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1309 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1310 } 1311 1312 no_options: 1313 if (isipv6) { 1314 struct route_in6 *ro6 = &sc->sc_route6; 1315 1316 th->th_sum = 0; 1317 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1318 ip6->ip6_hlim = in6_selecthlim(NULL, 1319 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1320 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1321 sc->sc_tp->t_inpcb); 1322 } else { 1323 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1324 htons(tlen - hlen + IPPROTO_TCP)); 1325 m->m_pkthdr.csum_flags = CSUM_TCP; 1326 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1327 m->m_pkthdr.csum_thlen = sizeof(struct tcphdr) + optlen; 1328 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 1329 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb); 1330 } 1331 return (error); 1332 } 1333 1334 /* 1335 * cookie layers: 1336 * 1337 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1338 * | peer iss | 1339 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1340 * | 0 |(A)| | 1341 * (A): peer mss index 1342 */ 1343 1344 /* 1345 * The values below are chosen to minimize the size of the tcp_secret 1346 * table, as well as providing roughly a 16 second lifetime for the cookie. 1347 */ 1348 1349 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1350 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1351 1352 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1353 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1354 #define SYNCOOKIE_TIMEOUT \ 1355 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1356 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1357 1358 static struct { 1359 u_int32_t ts_secbits[4]; 1360 u_int ts_expire; 1361 } tcp_secret[SYNCOOKIE_NSECRETS]; 1362 1363 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1364 1365 static MD5_CTX syn_ctx; 1366 1367 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1368 1369 struct md5_add { 1370 u_int32_t laddr, faddr; 1371 u_int32_t secbits[4]; 1372 u_int16_t lport, fport; 1373 }; 1374 1375 #ifdef CTASSERT 1376 CTASSERT(sizeof(struct md5_add) == 28); 1377 #endif 1378 1379 /* 1380 * Consider the problem of a recreated (and retransmitted) cookie. If the 1381 * original SYN was accepted, the connection is established. The second 1382 * SYN is inflight, and if it arrives with an ISN that falls within the 1383 * receive window, the connection is killed. 1384 * 1385 * However, since cookies have other problems, this may not be worth 1386 * worrying about. 1387 */ 1388 1389 static u_int32_t 1390 syncookie_generate(struct syncache *sc) 1391 { 1392 u_int32_t md5_buffer[4]; 1393 u_int32_t data; 1394 int idx, i; 1395 struct md5_add add; 1396 #ifdef INET6 1397 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1398 #else 1399 const boolean_t isipv6 = FALSE; 1400 #endif 1401 1402 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1403 if (tcp_secret[idx].ts_expire < ticks) { 1404 for (i = 0; i < 4; i++) 1405 tcp_secret[idx].ts_secbits[i] = karc4random(); 1406 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1407 } 1408 for (data = NELEM(tcp_msstab) - 1; data > 0; data--) 1409 if (tcp_msstab[data] <= sc->sc_peer_mss) 1410 break; 1411 data = (data << SYNCOOKIE_WNDBITS) | idx; 1412 data ^= sc->sc_irs; /* peer's iss */ 1413 MD5Init(&syn_ctx); 1414 if (isipv6) { 1415 MD5Add(sc->sc_inc.inc6_laddr); 1416 MD5Add(sc->sc_inc.inc6_faddr); 1417 add.laddr = 0; 1418 add.faddr = 0; 1419 } else { 1420 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1421 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1422 } 1423 add.lport = sc->sc_inc.inc_lport; 1424 add.fport = sc->sc_inc.inc_fport; 1425 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1426 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1427 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1428 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1429 MD5Add(add); 1430 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1431 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1432 return (data); 1433 } 1434 1435 static struct syncache * 1436 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so) 1437 { 1438 u_int32_t md5_buffer[4]; 1439 struct syncache *sc; 1440 u_int32_t data; 1441 int wnd, idx; 1442 struct md5_add add; 1443 1444 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1445 idx = data & SYNCOOKIE_WNDMASK; 1446 if (tcp_secret[idx].ts_expire < ticks || 1447 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1448 return (NULL); 1449 MD5Init(&syn_ctx); 1450 #ifdef INET6 1451 if (inc->inc_isipv6) { 1452 MD5Add(inc->inc6_laddr); 1453 MD5Add(inc->inc6_faddr); 1454 add.laddr = 0; 1455 add.faddr = 0; 1456 } else 1457 #endif 1458 { 1459 add.laddr = inc->inc_laddr.s_addr; 1460 add.faddr = inc->inc_faddr.s_addr; 1461 } 1462 add.lport = inc->inc_lport; 1463 add.fport = inc->inc_fport; 1464 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1465 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1466 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1467 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1468 MD5Add(add); 1469 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1470 data ^= md5_buffer[0]; 1471 if (data & ~SYNCOOKIE_DATAMASK) 1472 return (NULL); 1473 data = data >> SYNCOOKIE_WNDBITS; 1474 1475 /* 1476 * Fill in the syncache values. 1477 * XXX duplicate code from syncache_add 1478 */ 1479 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1480 sc->sc_ipopts = NULL; 1481 sc->sc_inc.inc_fport = inc->inc_fport; 1482 sc->sc_inc.inc_lport = inc->inc_lport; 1483 #ifdef INET6 1484 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1485 if (inc->inc_isipv6) { 1486 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1487 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1488 sc->sc_route6.ro_rt = NULL; 1489 } else 1490 #endif 1491 { 1492 sc->sc_inc.inc_faddr = inc->inc_faddr; 1493 sc->sc_inc.inc_laddr = inc->inc_laddr; 1494 sc->sc_route.ro_rt = NULL; 1495 } 1496 sc->sc_irs = th->th_seq - 1; 1497 sc->sc_iss = th->th_ack - 1; 1498 wnd = ssb_space(&so->so_rcv); 1499 wnd = imax(wnd, 0); 1500 wnd = imin(wnd, TCP_MAXWIN); 1501 sc->sc_wnd = wnd; 1502 sc->sc_flags = 0; 1503 sc->sc_rxtslot = 0; 1504 sc->sc_peer_mss = tcp_msstab[data]; 1505 return (sc); 1506 } 1507