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 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 int 243 syncache_rto(int slot) 244 { 245 if (tcp_low_rtobase) 246 return (TCPTV_RTOBASE * tcp_syn_backoff_low[slot]); 247 else 248 return (TCPTV_RTOBASE * tcp_syn_backoff[slot]); 249 } 250 251 static __inline void 252 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu, 253 struct syncache *sc, int slot) 254 { 255 int rto; 256 257 if (slot > 0) { 258 /* 259 * Record the time that we spent in SYN|ACK 260 * retransmition. 261 * 262 * Needed by RFC3390 and RFC6298. 263 */ 264 sc->sc_rxtused += syncache_rto(slot - 1); 265 } 266 sc->sc_rxtslot = slot; 267 268 rto = syncache_rto(slot); 269 sc->sc_rxttime = ticks + rto; 270 271 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq); 272 if (!callout_active(&syncache_percpu->tt_timerq[slot])) { 273 callout_reset(&syncache_percpu->tt_timerq[slot], rto, 274 syncache_timer, &syncache_percpu->mrec[slot]); 275 } 276 } 277 278 static void 279 syncache_free(struct syncache *sc) 280 { 281 struct rtentry *rt; 282 #ifdef INET6 283 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 284 #else 285 const boolean_t isipv6 = FALSE; 286 #endif 287 288 if (sc->sc_ipopts) 289 m_free(sc->sc_ipopts); 290 291 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt; 292 if (rt != NULL) { 293 /* 294 * If this is the only reference to a protocol-cloned 295 * route, remove it immediately. 296 */ 297 if ((rt->rt_flags & (RTF_WASCLONED | RTF_LLINFO)) == 298 RTF_WASCLONED && rt->rt_refcnt == 1) { 299 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 300 rt_mask(rt), rt->rt_flags, NULL); 301 } 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 for (cpu = 0; cpu < ncpus2; cpu++) { 332 struct tcp_syncache_percpu *syncache_percpu; 333 334 syncache_percpu = &tcp_syncache_percpu[cpu]; 335 /* Allocate the hash table. */ 336 syncache_percpu->hashbase = kmalloc(tcp_syncache.hashsize * sizeof(struct syncache_head), 337 M_SYNCACHE, M_WAITOK); 338 339 /* Initialize the hash buckets. */ 340 for (i = 0; i < tcp_syncache.hashsize; i++) { 341 struct syncache_head *bucket; 342 343 bucket = &syncache_percpu->hashbase[i]; 344 TAILQ_INIT(&bucket->sch_bucket); 345 bucket->sch_length = 0; 346 } 347 348 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 349 /* Initialize the timer queues. */ 350 TAILQ_INIT(&syncache_percpu->timerq[i]); 351 callout_init_mp(&syncache_percpu->tt_timerq[i]); 352 353 syncache_percpu->mrec[i].slot = i; 354 syncache_percpu->mrec[i].port = netisr_cpuport(cpu); 355 syncache_percpu->mrec[i].msg.nm_mrec = 356 &syncache_percpu->mrec[i]; 357 netmsg_init(&syncache_percpu->mrec[i].msg.base, 358 NULL, &netisr_adone_rport, 359 MSGF_PRIORITY, syncache_timer_handler); 360 } 361 } 362 } 363 364 static void 365 syncache_insert(struct syncache *sc, struct syncache_head *sch) 366 { 367 struct tcp_syncache_percpu *syncache_percpu; 368 struct syncache *sc2; 369 int i; 370 371 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 372 373 /* 374 * Make sure that we don't overflow the per-bucket 375 * limit or the total cache size limit. 376 */ 377 if (sch->sch_length >= tcp_syncache.bucket_limit) { 378 /* 379 * The bucket is full, toss the oldest element. 380 */ 381 sc2 = TAILQ_FIRST(&sch->sch_bucket); 382 if (sc2->sc_tp != NULL) 383 sc2->sc_tp->ts_recent = ticks; 384 syncache_drop(sc2, sch); 385 tcpstat.tcps_sc_bucketoverflow++; 386 } else if (syncache_percpu->cache_count >= tcp_syncache.cache_limit) { 387 /* 388 * The cache is full. Toss the oldest entry in the 389 * entire cache. This is the front entry in the 390 * first non-empty timer queue with the largest 391 * timeout value. 392 */ 393 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 394 sc2 = TAILQ_FIRST(&syncache_percpu->timerq[i]); 395 while (sc2 && (sc2->sc_flags & SCF_MARKER)) 396 sc2 = TAILQ_NEXT(sc2, sc_timerq); 397 if (sc2 != NULL) 398 break; 399 } 400 if (sc2->sc_tp != NULL) 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, struct tcpcb *tp_inh) 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 = tp_inh; 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 sc->sc_tp = NULL; 466 467 /* 468 * Remove the entry from the syncache timer/timeout queue. Note 469 * that we do not try to stop any running timer since we do not know 470 * whether the timer's message is in-transit or not. Since timeouts 471 * are fairly long, taking an unneeded callout does not detrimentally 472 * effect performance. 473 */ 474 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], sc, sc_timerq); 475 476 syncache_free(sc); 477 } 478 479 /* 480 * Place a timeout message on the TCP thread's message queue. 481 * This routine runs in soft interrupt context. 482 * 483 * An invariant is for this routine to be called, the callout must 484 * have been active. Note that the callout is not deactivated until 485 * after the message has been processed in syncache_timer_handler() below. 486 */ 487 static void 488 syncache_timer(void *p) 489 { 490 struct netmsg_sc_timer *msg = p; 491 492 lwkt_sendmsg_oncpu(msg->nm_mrec->port, &msg->base.lmsg); 493 } 494 495 /* 496 * Service a timer message queued by timer expiration. 497 * This routine runs in the TCP protocol thread. 498 * 499 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 500 * If we have retransmitted an entry the maximum number of times, expire it. 501 * 502 * When we finish processing timed-out entries, we restart the timer if there 503 * are any entries still on the queue and deactivate it otherwise. Only after 504 * a timer has been deactivated here can it be restarted by syncache_timeout(). 505 */ 506 static void 507 syncache_timer_handler(netmsg_t msg) 508 { 509 struct tcp_syncache_percpu *syncache_percpu; 510 struct syncache *sc; 511 struct syncache marker; 512 struct syncache_list *list; 513 struct inpcb *inp; 514 int slot; 515 516 slot = ((struct netmsg_sc_timer *)msg)->nm_mrec->slot; 517 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 518 519 list = &syncache_percpu->timerq[slot]; 520 521 /* 522 * Use a marker to keep our place in the scan. syncache_drop() 523 * can block and cause any next pointer we cache to become stale. 524 */ 525 marker.sc_flags = SCF_MARKER; 526 TAILQ_INSERT_HEAD(list, &marker, sc_timerq); 527 528 while ((sc = TAILQ_NEXT(&marker, sc_timerq)) != NULL) { 529 /* 530 * Move the marker. 531 */ 532 TAILQ_REMOVE(list, &marker, sc_timerq); 533 TAILQ_INSERT_AFTER(list, sc, &marker, sc_timerq); 534 535 if (sc->sc_flags & SCF_MARKER) 536 continue; 537 538 if (ticks < sc->sc_rxttime) 539 break; /* finished because timerq sorted by time */ 540 if (sc->sc_tp == NULL) { 541 syncache_drop(sc, NULL); 542 tcpstat.tcps_sc_stale++; 543 continue; 544 } 545 inp = sc->sc_tp->t_inpcb; 546 if (slot == SYNCACHE_MAXREXMTS || 547 slot >= tcp_syncache.rexmt_limit || 548 inp == NULL || 549 inp->inp_gencnt != sc->sc_inp_gencnt) { 550 syncache_drop(sc, NULL); 551 tcpstat.tcps_sc_stale++; 552 continue; 553 } 554 /* 555 * syncache_respond() may call back into the syncache to 556 * to modify another entry, so do not obtain the next 557 * entry on the timer chain until it has completed. 558 */ 559 syncache_respond(sc, NULL); 560 tcpstat.tcps_sc_retransmitted++; 561 TAILQ_REMOVE(list, sc, sc_timerq); 562 syncache_timeout(syncache_percpu, sc, slot + 1); 563 } 564 TAILQ_REMOVE(list, &marker, sc_timerq); 565 566 if (sc != NULL) { 567 callout_reset(&syncache_percpu->tt_timerq[slot], 568 sc->sc_rxttime - ticks, syncache_timer, 569 &syncache_percpu->mrec[slot]); 570 } else { 571 callout_deactivate(&syncache_percpu->tt_timerq[slot]); 572 } 573 lwkt_replymsg(&msg->base.lmsg, 0); 574 } 575 576 /* 577 * Find an entry in the syncache. 578 */ 579 struct syncache * 580 syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) 581 { 582 struct tcp_syncache_percpu *syncache_percpu; 583 struct syncache *sc; 584 struct syncache_head *sch; 585 586 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 587 #ifdef INET6 588 if (inc->inc_isipv6) { 589 sch = &syncache_percpu->hashbase[ 590 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 591 *schp = sch; 592 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) 593 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 594 return (sc); 595 } else 596 #endif 597 { 598 sch = &syncache_percpu->hashbase[ 599 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 600 *schp = sch; 601 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 602 #ifdef INET6 603 if (sc->sc_inc.inc_isipv6) 604 continue; 605 #endif 606 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) 607 return (sc); 608 } 609 } 610 return (NULL); 611 } 612 613 /* 614 * This function is called when we get a RST for a 615 * non-existent connection, so that we can see if the 616 * connection is in the syn cache. If it is, zap it. 617 */ 618 void 619 syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) 620 { 621 struct syncache *sc; 622 struct syncache_head *sch; 623 624 sc = syncache_lookup(inc, &sch); 625 if (sc == NULL) { 626 return; 627 } 628 /* 629 * If the RST bit is set, check the sequence number to see 630 * if this is a valid reset segment. 631 * RFC 793 page 37: 632 * In all states except SYN-SENT, all reset (RST) segments 633 * are validated by checking their SEQ-fields. A reset is 634 * valid if its sequence number is in the window. 635 * 636 * The sequence number in the reset segment is normally an 637 * echo of our outgoing acknowlegement numbers, but some hosts 638 * send a reset with the sequence number at the rightmost edge 639 * of our receive window, and we have to handle this case. 640 */ 641 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 642 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 643 syncache_drop(sc, sch); 644 tcpstat.tcps_sc_reset++; 645 } 646 } 647 648 void 649 syncache_badack(struct in_conninfo *inc) 650 { 651 struct syncache *sc; 652 struct syncache_head *sch; 653 654 sc = syncache_lookup(inc, &sch); 655 if (sc != NULL) { 656 syncache_drop(sc, sch); 657 tcpstat.tcps_sc_badack++; 658 } 659 } 660 661 void 662 syncache_unreach(struct in_conninfo *inc, const struct tcphdr *th) 663 { 664 struct syncache *sc; 665 struct syncache_head *sch; 666 667 /* we are called at splnet() here */ 668 sc = syncache_lookup(inc, &sch); 669 if (sc == NULL) 670 return; 671 672 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 673 if (ntohl(th->th_seq) != sc->sc_iss) 674 return; 675 676 /* 677 * If we've rertransmitted 3 times and this is our second error, 678 * we remove the entry. Otherwise, we allow it to continue on. 679 * This prevents us from incorrectly nuking an entry during a 680 * spurious network outage. 681 * 682 * See tcp_notify(). 683 */ 684 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 685 sc->sc_flags |= SCF_UNREACH; 686 return; 687 } 688 syncache_drop(sc, sch); 689 tcpstat.tcps_sc_unreach++; 690 } 691 692 /* 693 * Build a new TCP socket structure from a syncache entry. 694 * 695 * This is called from the context of the SYN+ACK 696 */ 697 static struct socket * 698 syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) 699 { 700 struct inpcb *inp = NULL, *linp; 701 struct socket *so; 702 struct tcpcb *tp, *ltp; 703 lwkt_port_t port; 704 #ifdef INET6 705 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 706 #else 707 const boolean_t isipv6 = FALSE; 708 #endif 709 struct sockaddr_in sin_faddr; 710 struct sockaddr_in6 sin6_faddr; 711 struct sockaddr *faddr; 712 713 if (isipv6) { 714 faddr = (struct sockaddr *)&sin6_faddr; 715 sin6_faddr.sin6_family = AF_INET6; 716 sin6_faddr.sin6_len = sizeof(sin6_faddr); 717 sin6_faddr.sin6_addr = sc->sc_inc.inc6_faddr; 718 sin6_faddr.sin6_port = sc->sc_inc.inc_fport; 719 sin6_faddr.sin6_flowinfo = sin6_faddr.sin6_scope_id = 0; 720 } else { 721 faddr = (struct sockaddr *)&sin_faddr; 722 sin_faddr.sin_family = AF_INET; 723 sin_faddr.sin_len = sizeof(sin_faddr); 724 sin_faddr.sin_addr = sc->sc_inc.inc_faddr; 725 sin_faddr.sin_port = sc->sc_inc.inc_fport; 726 bzero(sin_faddr.sin_zero, sizeof(sin_faddr.sin_zero)); 727 } 728 729 /* 730 * Ok, create the full blown connection, and set things up 731 * as they would have been set up if we had created the 732 * connection when the SYN arrived. If we can't create 733 * the connection, abort it. 734 * 735 * Set the protocol processing port for the socket to the current 736 * port (that the connection came in on). 737 */ 738 so = sonewconn_faddr(lso, SS_ISCONNECTED, faddr); 739 if (so == NULL) { 740 /* 741 * Drop the connection; we will send a RST if the peer 742 * retransmits the ACK, 743 */ 744 tcpstat.tcps_listendrop++; 745 goto abort; 746 } 747 748 /* 749 * Insert new socket into hash list. 750 */ 751 inp = so->so_pcb; 752 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 753 if (isipv6) { 754 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 755 } else { 756 #ifdef INET6 757 inp->inp_vflag &= ~INP_IPV6; 758 inp->inp_vflag |= INP_IPV4; 759 inp->inp_flags &= ~IN6P_IPV6_V6ONLY; 760 #endif 761 inp->inp_laddr = sc->sc_inc.inc_laddr; 762 } 763 inp->inp_lport = sc->sc_inc.inc_lport; 764 in_pcbinsporthash_lport(inp); 765 766 linp = lso->so_pcb; 767 #ifdef IPSEC 768 /* copy old policy into new socket's */ 769 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 770 kprintf("syncache_expand: could not copy policy\n"); 771 #endif 772 if (isipv6) { 773 struct in6_addr laddr6; 774 /* 775 * Inherit socket options from the listening socket. 776 * Note that in6p_inputopts are not (and should not be) 777 * copied, since it stores previously received options and is 778 * used to detect if each new option is different than the 779 * previous one and hence should be passed to a user. 780 * If we copied in6p_inputopts, a user would not be able to 781 * receive options just after calling the accept system call. 782 */ 783 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 784 if (linp->in6p_outputopts) 785 inp->in6p_outputopts = 786 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 787 inp->in6p_route = sc->sc_route6; 788 sc->sc_route6.ro_rt = NULL; 789 790 laddr6 = inp->in6p_laddr; 791 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 792 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 793 if (in6_pcbconnect(inp, faddr, &thread0)) { 794 inp->in6p_laddr = laddr6; 795 goto abort; 796 } 797 } else { 798 struct in_addr laddr; 799 800 inp->inp_options = ip_srcroute(m); 801 if (inp->inp_options == NULL) { 802 inp->inp_options = sc->sc_ipopts; 803 sc->sc_ipopts = NULL; 804 } 805 inp->inp_route = sc->sc_route; 806 sc->sc_route.ro_rt = NULL; 807 808 laddr = inp->inp_laddr; 809 if (inp->inp_laddr.s_addr == INADDR_ANY) 810 inp->inp_laddr = sc->sc_inc.inc_laddr; 811 if (in_pcbconnect(inp, faddr, &thread0)) { 812 inp->inp_laddr = laddr; 813 goto abort; 814 } 815 } 816 817 /* 818 * The current port should be in the context of the SYN+ACK and 819 * so should match the tcp address port. 820 */ 821 if (isipv6) { 822 port = tcp6_addrport(); 823 } else { 824 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport, 825 inp->inp_laddr.s_addr, inp->inp_lport); 826 } 827 KASSERT(port == &curthread->td_msgport, 828 ("TCP PORT MISMATCH %p vs %p\n", port, &curthread->td_msgport)); 829 830 tp = intotcpcb(inp); 831 tp->t_state = TCPS_SYN_RECEIVED; 832 tp->iss = sc->sc_iss; 833 tp->irs = sc->sc_irs; 834 tcp_rcvseqinit(tp); 835 tcp_sendseqinit(tp); 836 tp->snd_wnd = sc->sc_sndwnd; 837 tp->snd_wl1 = sc->sc_irs; 838 tp->rcv_up = sc->sc_irs + 1; 839 tp->rcv_wnd = sc->sc_wnd; 840 tp->rcv_adv += tp->rcv_wnd; 841 842 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 843 if (sc->sc_flags & SCF_NOOPT) 844 tp->t_flags |= TF_NOOPT; 845 if (sc->sc_flags & SCF_WINSCALE) { 846 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 847 tp->snd_scale = sc->sc_requested_s_scale; 848 tp->request_r_scale = sc->sc_request_r_scale; 849 } 850 if (sc->sc_flags & SCF_TIMESTAMP) { 851 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 852 tp->ts_recent = sc->sc_tsrecent; 853 tp->ts_recent_age = ticks; 854 } 855 if (sc->sc_flags & SCF_SACK_PERMITTED) 856 tp->t_flags |= TF_SACK_PERMITTED; 857 858 #ifdef TCP_SIGNATURE 859 if (sc->sc_flags & SCF_SIGNATURE) 860 tp->t_flags |= TF_SIGNATURE; 861 #endif /* TCP_SIGNATURE */ 862 863 tp->t_rxtsyn = sc->sc_rxtused; 864 tcp_mss(tp, sc->sc_peer_mss); 865 866 /* 867 * Inherit some properties from the listen socket 868 */ 869 ltp = intotcpcb(linp); 870 tp->t_keepinit = ltp->t_keepinit; 871 tp->t_keepidle = ltp->t_keepidle; 872 tp->t_keepintvl = ltp->t_keepintvl; 873 tp->t_keepcnt = ltp->t_keepcnt; 874 tp->t_maxidle = ltp->t_maxidle; 875 876 tcp_create_timermsg(tp, port); 877 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep); 878 879 tcpstat.tcps_accepts++; 880 return (so); 881 882 abort: 883 if (so != NULL) 884 soabort_oncpu(so); 885 return (NULL); 886 } 887 888 /* 889 * This function gets called when we receive an ACK for a 890 * socket in the LISTEN state. We look up the connection 891 * in the syncache, and if its there, we pull it out of 892 * the cache and turn it into a full-blown connection in 893 * the SYN-RECEIVED state. 894 */ 895 int 896 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop, 897 struct mbuf *m) 898 { 899 struct syncache *sc; 900 struct syncache_head *sch; 901 struct socket *so; 902 903 sc = syncache_lookup(inc, &sch); 904 if (sc == NULL) { 905 /* 906 * There is no syncache entry, so see if this ACK is 907 * a returning syncookie. To do this, first: 908 * A. See if this socket has had a syncache entry dropped in 909 * the past. We don't want to accept a bogus syncookie 910 * if we've never received a SYN. 911 * B. check that the syncookie is valid. If it is, then 912 * cobble up a fake syncache entry, and return. 913 */ 914 if (!tcp_syncookies) 915 return (0); 916 sc = syncookie_lookup(inc, th, *sop); 917 if (sc == NULL) 918 return (0); 919 sch = NULL; 920 tcpstat.tcps_sc_recvcookie++; 921 } 922 923 /* 924 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 925 */ 926 if (th->th_ack != sc->sc_iss + 1) 927 return (0); 928 929 so = syncache_socket(sc, *sop, m); 930 if (so == NULL) { 931 #if 0 932 resetandabort: 933 /* XXXjlemon check this - is this correct? */ 934 tcp_respond(NULL, m, m, th, 935 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 936 #endif 937 m_freem(m); /* XXX only needed for above */ 938 tcpstat.tcps_sc_aborted++; 939 } else { 940 tcpstat.tcps_sc_completed++; 941 } 942 if (sch == NULL) 943 syncache_free(sc); 944 else 945 syncache_drop(sc, sch); 946 *sop = so; 947 return (1); 948 } 949 950 /* 951 * Given a LISTEN socket and an inbound SYN request, add 952 * this to the syn cache, and send back a segment: 953 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 954 * to the source. 955 * 956 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 957 * Doing so would require that we hold onto the data and deliver it 958 * to the application. However, if we are the target of a SYN-flood 959 * DoS attack, an attacker could send data which would eventually 960 * consume all available buffer space if it were ACKed. By not ACKing 961 * the data, we avoid this DoS scenario. 962 */ 963 int 964 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 965 struct socket *so, struct mbuf *m) 966 { 967 struct tcp_syncache_percpu *syncache_percpu; 968 struct tcpcb *tp; 969 struct syncache *sc = NULL; 970 struct syncache_head *sch; 971 struct mbuf *ipopts = NULL; 972 int win; 973 974 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 975 tp = sototcpcb(so); 976 977 /* 978 * Remember the IP options, if any. 979 */ 980 #ifdef INET6 981 if (!inc->inc_isipv6) 982 #endif 983 ipopts = ip_srcroute(m); 984 985 /* 986 * See if we already have an entry for this connection. 987 * If we do, resend the SYN,ACK, and reset the retransmit timer. 988 * 989 * XXX 990 * The syncache should be re-initialized with the contents 991 * of the new SYN which may have different options. 992 */ 993 sc = syncache_lookup(inc, &sch); 994 if (sc != NULL) { 995 tcpstat.tcps_sc_dupsyn++; 996 if (ipopts) { 997 /* 998 * If we were remembering a previous source route, 999 * forget it and use the new one we've been given. 1000 */ 1001 if (sc->sc_ipopts) 1002 m_free(sc->sc_ipopts); 1003 sc->sc_ipopts = ipopts; 1004 } 1005 /* 1006 * Update timestamp if present. 1007 */ 1008 if (sc->sc_flags & SCF_TIMESTAMP) 1009 sc->sc_tsrecent = to->to_tsval; 1010 1011 /* Just update the TOF_SACK_PERMITTED for now. */ 1012 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1013 sc->sc_flags |= SCF_SACK_PERMITTED; 1014 else 1015 sc->sc_flags &= ~SCF_SACK_PERMITTED; 1016 1017 /* Update initial send window */ 1018 sc->sc_sndwnd = th->th_win; 1019 1020 /* 1021 * PCB may have changed, pick up new values. 1022 */ 1023 sc->sc_tp = tp; 1024 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1025 if (syncache_respond(sc, m) == 0) { 1026 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 1027 sc, sc_timerq); 1028 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 1029 tcpstat.tcps_sndacks++; 1030 tcpstat.tcps_sndtotal++; 1031 } 1032 return (1); 1033 } 1034 1035 /* 1036 * Fill in the syncache values. 1037 */ 1038 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1039 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1040 sc->sc_ipopts = ipopts; 1041 sc->sc_inc.inc_fport = inc->inc_fport; 1042 sc->sc_inc.inc_lport = inc->inc_lport; 1043 sc->sc_tp = tp; 1044 #ifdef INET6 1045 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1046 if (inc->inc_isipv6) { 1047 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1048 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1049 sc->sc_route6.ro_rt = NULL; 1050 } else 1051 #endif 1052 { 1053 sc->sc_inc.inc_faddr = inc->inc_faddr; 1054 sc->sc_inc.inc_laddr = inc->inc_laddr; 1055 sc->sc_route.ro_rt = NULL; 1056 } 1057 sc->sc_irs = th->th_seq; 1058 sc->sc_flags = 0; 1059 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 1060 if (tcp_syncookies) 1061 sc->sc_iss = syncookie_generate(sc); 1062 else 1063 sc->sc_iss = karc4random(); 1064 1065 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */ 1066 win = ssb_space(&so->so_rcv); 1067 win = imax(win, 0); 1068 win = imin(win, TCP_MAXWIN); 1069 sc->sc_wnd = win; 1070 1071 if (tcp_do_rfc1323) { 1072 /* 1073 * A timestamp received in a SYN makes 1074 * it ok to send timestamp requests and replies. 1075 */ 1076 if (to->to_flags & TOF_TS) { 1077 sc->sc_tsrecent = to->to_tsval; 1078 sc->sc_flags |= SCF_TIMESTAMP; 1079 } 1080 if (to->to_flags & TOF_SCALE) { 1081 int wscale = TCP_MIN_WINSHIFT; 1082 1083 /* Compute proper scaling value from buffer space */ 1084 while (wscale < TCP_MAX_WINSHIFT && 1085 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) { 1086 wscale++; 1087 } 1088 sc->sc_request_r_scale = wscale; 1089 sc->sc_requested_s_scale = to->to_requested_s_scale; 1090 sc->sc_flags |= SCF_WINSCALE; 1091 } 1092 } 1093 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1094 sc->sc_flags |= SCF_SACK_PERMITTED; 1095 if (tp->t_flags & TF_NOOPT) 1096 sc->sc_flags = SCF_NOOPT; 1097 #ifdef TCP_SIGNATURE 1098 /* 1099 * If listening socket requested TCP digests, and received SYN 1100 * contains the option, flag this in the syncache so that 1101 * syncache_respond() will do the right thing with the SYN+ACK. 1102 * XXX Currently we always record the option by default and will 1103 * attempt to use it in syncache_respond(). 1104 */ 1105 if (to->to_flags & TOF_SIGNATURE) 1106 sc->sc_flags = SCF_SIGNATURE; 1107 #endif /* TCP_SIGNATURE */ 1108 sc->sc_sndwnd = th->th_win; 1109 1110 if (syncache_respond(sc, m) == 0) { 1111 syncache_insert(sc, sch); 1112 tcpstat.tcps_sndacks++; 1113 tcpstat.tcps_sndtotal++; 1114 } else { 1115 syncache_free(sc); 1116 tcpstat.tcps_sc_dropped++; 1117 } 1118 return (1); 1119 } 1120 1121 static int 1122 syncache_respond(struct syncache *sc, struct mbuf *m) 1123 { 1124 u_int8_t *optp; 1125 int optlen, error; 1126 u_int16_t tlen, hlen, mssopt; 1127 struct ip *ip = NULL; 1128 struct rtentry *rt; 1129 struct tcphdr *th; 1130 struct ip6_hdr *ip6 = NULL; 1131 #ifdef INET6 1132 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1133 #else 1134 const boolean_t isipv6 = FALSE; 1135 #endif 1136 1137 if (isipv6) { 1138 rt = tcp_rtlookup6(&sc->sc_inc); 1139 if (rt != NULL) 1140 mssopt = rt->rt_ifp->if_mtu - 1141 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1142 else 1143 mssopt = tcp_v6mssdflt; 1144 hlen = sizeof(struct ip6_hdr); 1145 } else { 1146 rt = tcp_rtlookup(&sc->sc_inc); 1147 if (rt != NULL) 1148 mssopt = rt->rt_ifp->if_mtu - 1149 (sizeof(struct ip) + sizeof(struct tcphdr)); 1150 else 1151 mssopt = tcp_mssdflt; 1152 hlen = sizeof(struct ip); 1153 } 1154 1155 /* Compute the size of the TCP options. */ 1156 if (sc->sc_flags & SCF_NOOPT) { 1157 optlen = 0; 1158 } else { 1159 optlen = TCPOLEN_MAXSEG + 1160 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1161 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1162 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1163 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1164 #ifdef TCP_SIGNATURE 1165 optlen += ((sc->sc_flags & SCF_SIGNATURE) ? 1166 (TCPOLEN_SIGNATURE + 2) : 0); 1167 #endif /* TCP_SIGNATURE */ 1168 } 1169 tlen = hlen + sizeof(struct tcphdr) + optlen; 1170 1171 /* 1172 * XXX 1173 * assume that the entire packet will fit in a header mbuf 1174 */ 1175 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1176 1177 /* 1178 * XXX shouldn't this reuse the mbuf if possible ? 1179 * Create the IP+TCP header from scratch. 1180 */ 1181 if (m) 1182 m_freem(m); 1183 1184 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 1185 if (m == NULL) 1186 return (ENOBUFS); 1187 m->m_data += max_linkhdr; 1188 m->m_len = tlen; 1189 m->m_pkthdr.len = tlen; 1190 m->m_pkthdr.rcvif = NULL; 1191 if (tcp_prio_synack) 1192 m->m_flags |= M_PRIO; 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 m->m_pkthdr.csum_thlen = sizeof(struct tcphdr) + optlen; 1311 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 1312 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb); 1313 } 1314 return (error); 1315 } 1316 1317 /* 1318 * cookie layers: 1319 * 1320 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1321 * | peer iss | 1322 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1323 * | 0 |(A)| | 1324 * (A): peer mss index 1325 */ 1326 1327 /* 1328 * The values below are chosen to minimize the size of the tcp_secret 1329 * table, as well as providing roughly a 16 second lifetime for the cookie. 1330 */ 1331 1332 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1333 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1334 1335 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1336 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1337 #define SYNCOOKIE_TIMEOUT \ 1338 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1339 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1340 1341 static struct { 1342 u_int32_t ts_secbits[4]; 1343 u_int ts_expire; 1344 } tcp_secret[SYNCOOKIE_NSECRETS]; 1345 1346 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1347 1348 static MD5_CTX syn_ctx; 1349 1350 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1351 1352 struct md5_add { 1353 u_int32_t laddr, faddr; 1354 u_int32_t secbits[4]; 1355 u_int16_t lport, fport; 1356 }; 1357 1358 #ifdef CTASSERT 1359 CTASSERT(sizeof(struct md5_add) == 28); 1360 #endif 1361 1362 /* 1363 * Consider the problem of a recreated (and retransmitted) cookie. If the 1364 * original SYN was accepted, the connection is established. The second 1365 * SYN is inflight, and if it arrives with an ISN that falls within the 1366 * receive window, the connection is killed. 1367 * 1368 * However, since cookies have other problems, this may not be worth 1369 * worrying about. 1370 */ 1371 1372 static u_int32_t 1373 syncookie_generate(struct syncache *sc) 1374 { 1375 u_int32_t md5_buffer[4]; 1376 u_int32_t data; 1377 int idx, i; 1378 struct md5_add add; 1379 #ifdef INET6 1380 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1381 #else 1382 const boolean_t isipv6 = FALSE; 1383 #endif 1384 1385 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1386 if (tcp_secret[idx].ts_expire < ticks) { 1387 for (i = 0; i < 4; i++) 1388 tcp_secret[idx].ts_secbits[i] = karc4random(); 1389 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1390 } 1391 for (data = NELEM(tcp_msstab) - 1; data > 0; data--) 1392 if (tcp_msstab[data] <= sc->sc_peer_mss) 1393 break; 1394 data = (data << SYNCOOKIE_WNDBITS) | idx; 1395 data ^= sc->sc_irs; /* peer's iss */ 1396 MD5Init(&syn_ctx); 1397 if (isipv6) { 1398 MD5Add(sc->sc_inc.inc6_laddr); 1399 MD5Add(sc->sc_inc.inc6_faddr); 1400 add.laddr = 0; 1401 add.faddr = 0; 1402 } else { 1403 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1404 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1405 } 1406 add.lport = sc->sc_inc.inc_lport; 1407 add.fport = sc->sc_inc.inc_fport; 1408 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1409 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1410 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1411 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1412 MD5Add(add); 1413 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1414 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1415 return (data); 1416 } 1417 1418 static struct syncache * 1419 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so) 1420 { 1421 u_int32_t md5_buffer[4]; 1422 struct syncache *sc; 1423 u_int32_t data; 1424 int wnd, idx; 1425 struct md5_add add; 1426 1427 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1428 idx = data & SYNCOOKIE_WNDMASK; 1429 if (tcp_secret[idx].ts_expire < ticks || 1430 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1431 return (NULL); 1432 MD5Init(&syn_ctx); 1433 #ifdef INET6 1434 if (inc->inc_isipv6) { 1435 MD5Add(inc->inc6_laddr); 1436 MD5Add(inc->inc6_faddr); 1437 add.laddr = 0; 1438 add.faddr = 0; 1439 } else 1440 #endif 1441 { 1442 add.laddr = inc->inc_laddr.s_addr; 1443 add.faddr = inc->inc_faddr.s_addr; 1444 } 1445 add.lport = inc->inc_lport; 1446 add.fport = inc->inc_fport; 1447 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1448 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1449 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1450 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1451 MD5Add(add); 1452 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1453 data ^= md5_buffer[0]; 1454 if (data & ~SYNCOOKIE_DATAMASK) 1455 return (NULL); 1456 data = data >> SYNCOOKIE_WNDBITS; 1457 1458 /* 1459 * Fill in the syncache values. 1460 * XXX duplicate code from syncache_add 1461 */ 1462 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1463 sc->sc_ipopts = NULL; 1464 sc->sc_inc.inc_fport = inc->inc_fport; 1465 sc->sc_inc.inc_lport = inc->inc_lport; 1466 #ifdef INET6 1467 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1468 if (inc->inc_isipv6) { 1469 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1470 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1471 sc->sc_route6.ro_rt = NULL; 1472 } else 1473 #endif 1474 { 1475 sc->sc_inc.inc_faddr = inc->inc_faddr; 1476 sc->sc_inc.inc_laddr = inc->inc_laddr; 1477 sc->sc_route.ro_rt = NULL; 1478 } 1479 sc->sc_irs = th->th_seq - 1; 1480 sc->sc_iss = th->th_ack - 1; 1481 wnd = ssb_space(&so->so_rcv); 1482 wnd = imax(wnd, 0); 1483 wnd = imin(wnd, TCP_MAXWIN); 1484 sc->sc_wnd = wnd; 1485 sc->sc_flags = 0; 1486 sc->sc_rxtslot = 0; 1487 sc->sc_peer_mss = tcp_msstab[data]; 1488 return (sc); 1489 } 1490