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 KASSERT(INP_ISIPV4(inp), ("not inet pcb")); 757 inp->inp_laddr = sc->sc_inc.inc_laddr; 758 } 759 inp->inp_lport = sc->sc_inc.inc_lport; 760 in_pcbinsporthash_lport(inp); 761 762 linp = lso->so_pcb; 763 #ifdef IPSEC 764 /* copy old policy into new socket's */ 765 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 766 kprintf("syncache_expand: could not copy policy\n"); 767 #endif 768 if (isipv6) { 769 struct in6_addr laddr6; 770 /* 771 * Inherit socket options from the listening socket. 772 * Note that in6p_inputopts are not (and should not be) 773 * copied, since it stores previously received options and is 774 * used to detect if each new option is different than the 775 * previous one and hence should be passed to a user. 776 * If we copied in6p_inputopts, a user would not be able to 777 * receive options just after calling the accept system call. 778 */ 779 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 780 if (linp->in6p_outputopts) 781 inp->in6p_outputopts = 782 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 783 inp->in6p_route = sc->sc_route6; 784 sc->sc_route6.ro_rt = NULL; 785 786 laddr6 = inp->in6p_laddr; 787 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 788 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 789 if (in6_pcbconnect(inp, faddr, &thread0)) { 790 inp->in6p_laddr = laddr6; 791 goto abort; 792 } 793 } else { 794 struct in_addr laddr; 795 796 inp->inp_options = ip_srcroute(m); 797 if (inp->inp_options == NULL) { 798 inp->inp_options = sc->sc_ipopts; 799 sc->sc_ipopts = NULL; 800 } 801 inp->inp_route = sc->sc_route; 802 sc->sc_route.ro_rt = NULL; 803 804 laddr = inp->inp_laddr; 805 if (inp->inp_laddr.s_addr == INADDR_ANY) 806 inp->inp_laddr = sc->sc_inc.inc_laddr; 807 if (in_pcbconnect(inp, faddr, &thread0)) { 808 inp->inp_laddr = laddr; 809 goto abort; 810 } 811 } 812 813 /* 814 * The current port should be in the context of the SYN+ACK and 815 * so should match the tcp address port. 816 */ 817 if (isipv6) { 818 port = tcp6_addrport(); 819 } else { 820 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport, 821 inp->inp_laddr.s_addr, inp->inp_lport); 822 } 823 KASSERT(port == &curthread->td_msgport, 824 ("TCP PORT MISMATCH %p vs %p\n", port, &curthread->td_msgport)); 825 826 tp = intotcpcb(inp); 827 tp->t_state = TCPS_SYN_RECEIVED; 828 tp->iss = sc->sc_iss; 829 tp->irs = sc->sc_irs; 830 tcp_rcvseqinit(tp); 831 tcp_sendseqinit(tp); 832 tp->snd_wnd = sc->sc_sndwnd; 833 tp->snd_wl1 = sc->sc_irs; 834 tp->rcv_up = sc->sc_irs + 1; 835 tp->rcv_wnd = sc->sc_wnd; 836 tp->rcv_adv += tp->rcv_wnd; 837 838 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 839 if (sc->sc_flags & SCF_NOOPT) 840 tp->t_flags |= TF_NOOPT; 841 if (sc->sc_flags & SCF_WINSCALE) { 842 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 843 tp->snd_scale = sc->sc_requested_s_scale; 844 tp->request_r_scale = sc->sc_request_r_scale; 845 } 846 if (sc->sc_flags & SCF_TIMESTAMP) { 847 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 848 tp->ts_recent = sc->sc_tsrecent; 849 tp->ts_recent_age = ticks; 850 } 851 if (sc->sc_flags & SCF_SACK_PERMITTED) 852 tp->t_flags |= TF_SACK_PERMITTED; 853 854 #ifdef TCP_SIGNATURE 855 if (sc->sc_flags & SCF_SIGNATURE) 856 tp->t_flags |= TF_SIGNATURE; 857 #endif /* TCP_SIGNATURE */ 858 859 tp->t_rxtsyn = sc->sc_rxtused; 860 tcp_mss(tp, sc->sc_peer_mss); 861 862 /* 863 * Inherit some properties from the listen socket 864 */ 865 ltp = intotcpcb(linp); 866 tp->t_keepinit = ltp->t_keepinit; 867 tp->t_keepidle = ltp->t_keepidle; 868 tp->t_keepintvl = ltp->t_keepintvl; 869 tp->t_keepcnt = ltp->t_keepcnt; 870 tp->t_maxidle = ltp->t_maxidle; 871 872 tcp_create_timermsg(tp, port); 873 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep); 874 875 tcpstat.tcps_accepts++; 876 return (so); 877 878 abort: 879 if (so != NULL) 880 soabort_oncpu(so); 881 return (NULL); 882 } 883 884 /* 885 * This function gets called when we receive an ACK for a 886 * socket in the LISTEN state. We look up the connection 887 * in the syncache, and if its there, we pull it out of 888 * the cache and turn it into a full-blown connection in 889 * the SYN-RECEIVED state. 890 */ 891 int 892 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop, 893 struct mbuf *m) 894 { 895 struct syncache *sc; 896 struct syncache_head *sch; 897 struct socket *so; 898 899 sc = syncache_lookup(inc, &sch); 900 if (sc == NULL) { 901 /* 902 * There is no syncache entry, so see if this ACK is 903 * a returning syncookie. To do this, first: 904 * A. See if this socket has had a syncache entry dropped in 905 * the past. We don't want to accept a bogus syncookie 906 * if we've never received a SYN. 907 * B. check that the syncookie is valid. If it is, then 908 * cobble up a fake syncache entry, and return. 909 */ 910 if (!tcp_syncookies) 911 return (0); 912 sc = syncookie_lookup(inc, th, *sop); 913 if (sc == NULL) 914 return (0); 915 sch = NULL; 916 tcpstat.tcps_sc_recvcookie++; 917 } 918 919 /* 920 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 921 */ 922 if (th->th_ack != sc->sc_iss + 1) 923 return (0); 924 925 so = syncache_socket(sc, *sop, m); 926 if (so == NULL) { 927 #if 0 928 resetandabort: 929 /* XXXjlemon check this - is this correct? */ 930 tcp_respond(NULL, m, m, th, 931 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 932 #endif 933 m_freem(m); /* XXX only needed for above */ 934 tcpstat.tcps_sc_aborted++; 935 } else { 936 tcpstat.tcps_sc_completed++; 937 } 938 if (sch == NULL) 939 syncache_free(sc); 940 else 941 syncache_drop(sc, sch); 942 *sop = so; 943 return (1); 944 } 945 946 /* 947 * Given a LISTEN socket and an inbound SYN request, add 948 * this to the syn cache, and send back a segment: 949 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 950 * to the source. 951 * 952 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 953 * Doing so would require that we hold onto the data and deliver it 954 * to the application. However, if we are the target of a SYN-flood 955 * DoS attack, an attacker could send data which would eventually 956 * consume all available buffer space if it were ACKed. By not ACKing 957 * the data, we avoid this DoS scenario. 958 */ 959 int 960 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 961 struct socket *so, struct mbuf *m) 962 { 963 struct tcp_syncache_percpu *syncache_percpu; 964 struct tcpcb *tp; 965 struct syncache *sc = NULL; 966 struct syncache_head *sch; 967 struct mbuf *ipopts = NULL; 968 int win; 969 970 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 971 tp = sototcpcb(so); 972 973 /* 974 * Remember the IP options, if any. 975 */ 976 #ifdef INET6 977 if (!inc->inc_isipv6) 978 #endif 979 ipopts = ip_srcroute(m); 980 981 /* 982 * See if we already have an entry for this connection. 983 * If we do, resend the SYN,ACK, and reset the retransmit timer. 984 * 985 * XXX 986 * The syncache should be re-initialized with the contents 987 * of the new SYN which may have different options. 988 */ 989 sc = syncache_lookup(inc, &sch); 990 if (sc != NULL) { 991 tcpstat.tcps_sc_dupsyn++; 992 if (ipopts) { 993 /* 994 * If we were remembering a previous source route, 995 * forget it and use the new one we've been given. 996 */ 997 if (sc->sc_ipopts) 998 m_free(sc->sc_ipopts); 999 sc->sc_ipopts = ipopts; 1000 } 1001 /* 1002 * Update timestamp if present. 1003 */ 1004 if (sc->sc_flags & SCF_TIMESTAMP) 1005 sc->sc_tsrecent = to->to_tsval; 1006 1007 /* Just update the TOF_SACK_PERMITTED for now. */ 1008 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1009 sc->sc_flags |= SCF_SACK_PERMITTED; 1010 else 1011 sc->sc_flags &= ~SCF_SACK_PERMITTED; 1012 1013 /* Update initial send window */ 1014 sc->sc_sndwnd = th->th_win; 1015 1016 /* 1017 * PCB may have changed, pick up new values. 1018 */ 1019 sc->sc_tp = tp; 1020 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1021 if (syncache_respond(sc, m) == 0) { 1022 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 1023 sc, sc_timerq); 1024 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 1025 tcpstat.tcps_sndacks++; 1026 tcpstat.tcps_sndtotal++; 1027 } 1028 return (1); 1029 } 1030 1031 /* 1032 * Fill in the syncache values. 1033 */ 1034 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1035 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1036 sc->sc_ipopts = ipopts; 1037 sc->sc_inc.inc_fport = inc->inc_fport; 1038 sc->sc_inc.inc_lport = inc->inc_lport; 1039 sc->sc_tp = tp; 1040 #ifdef INET6 1041 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1042 if (inc->inc_isipv6) { 1043 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1044 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1045 sc->sc_route6.ro_rt = NULL; 1046 } else 1047 #endif 1048 { 1049 sc->sc_inc.inc_faddr = inc->inc_faddr; 1050 sc->sc_inc.inc_laddr = inc->inc_laddr; 1051 sc->sc_route.ro_rt = NULL; 1052 } 1053 sc->sc_irs = th->th_seq; 1054 sc->sc_flags = 0; 1055 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 1056 if (tcp_syncookies) 1057 sc->sc_iss = syncookie_generate(sc); 1058 else 1059 sc->sc_iss = karc4random(); 1060 1061 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */ 1062 win = ssb_space(&so->so_rcv); 1063 win = imax(win, 0); 1064 win = imin(win, TCP_MAXWIN); 1065 sc->sc_wnd = win; 1066 1067 if (tcp_do_rfc1323) { 1068 /* 1069 * A timestamp received in a SYN makes 1070 * it ok to send timestamp requests and replies. 1071 */ 1072 if (to->to_flags & TOF_TS) { 1073 sc->sc_tsrecent = to->to_tsval; 1074 sc->sc_flags |= SCF_TIMESTAMP; 1075 } 1076 if (to->to_flags & TOF_SCALE) { 1077 int wscale = TCP_MIN_WINSHIFT; 1078 1079 /* Compute proper scaling value from buffer space */ 1080 while (wscale < TCP_MAX_WINSHIFT && 1081 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) { 1082 wscale++; 1083 } 1084 sc->sc_request_r_scale = wscale; 1085 sc->sc_requested_s_scale = to->to_requested_s_scale; 1086 sc->sc_flags |= SCF_WINSCALE; 1087 } 1088 } 1089 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1090 sc->sc_flags |= SCF_SACK_PERMITTED; 1091 if (tp->t_flags & TF_NOOPT) 1092 sc->sc_flags = SCF_NOOPT; 1093 #ifdef TCP_SIGNATURE 1094 /* 1095 * If listening socket requested TCP digests, and received SYN 1096 * contains the option, flag this in the syncache so that 1097 * syncache_respond() will do the right thing with the SYN+ACK. 1098 * XXX Currently we always record the option by default and will 1099 * attempt to use it in syncache_respond(). 1100 */ 1101 if (to->to_flags & TOF_SIGNATURE) 1102 sc->sc_flags = SCF_SIGNATURE; 1103 #endif /* TCP_SIGNATURE */ 1104 sc->sc_sndwnd = th->th_win; 1105 1106 if (syncache_respond(sc, m) == 0) { 1107 syncache_insert(sc, sch); 1108 tcpstat.tcps_sndacks++; 1109 tcpstat.tcps_sndtotal++; 1110 } else { 1111 syncache_free(sc); 1112 tcpstat.tcps_sc_dropped++; 1113 } 1114 return (1); 1115 } 1116 1117 static int 1118 syncache_respond(struct syncache *sc, struct mbuf *m) 1119 { 1120 u_int8_t *optp; 1121 int optlen, error; 1122 u_int16_t tlen, hlen, mssopt; 1123 struct ip *ip = NULL; 1124 struct rtentry *rt; 1125 struct tcphdr *th; 1126 struct ip6_hdr *ip6 = NULL; 1127 #ifdef INET6 1128 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1129 #else 1130 const boolean_t isipv6 = FALSE; 1131 #endif 1132 1133 if (isipv6) { 1134 rt = tcp_rtlookup6(&sc->sc_inc); 1135 if (rt != NULL) 1136 mssopt = rt->rt_ifp->if_mtu - 1137 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1138 else 1139 mssopt = tcp_v6mssdflt; 1140 hlen = sizeof(struct ip6_hdr); 1141 } else { 1142 rt = tcp_rtlookup(&sc->sc_inc); 1143 if (rt != NULL) 1144 mssopt = rt->rt_ifp->if_mtu - 1145 (sizeof(struct ip) + sizeof(struct tcphdr)); 1146 else 1147 mssopt = tcp_mssdflt; 1148 hlen = sizeof(struct ip); 1149 } 1150 1151 /* Compute the size of the TCP options. */ 1152 if (sc->sc_flags & SCF_NOOPT) { 1153 optlen = 0; 1154 } else { 1155 optlen = TCPOLEN_MAXSEG + 1156 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1157 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1158 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1159 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1160 #ifdef TCP_SIGNATURE 1161 optlen += ((sc->sc_flags & SCF_SIGNATURE) ? 1162 (TCPOLEN_SIGNATURE + 2) : 0); 1163 #endif /* TCP_SIGNATURE */ 1164 } 1165 tlen = hlen + sizeof(struct tcphdr) + optlen; 1166 1167 /* 1168 * XXX 1169 * assume that the entire packet will fit in a header mbuf 1170 */ 1171 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1172 1173 /* 1174 * XXX shouldn't this reuse the mbuf if possible ? 1175 * Create the IP+TCP header from scratch. 1176 */ 1177 if (m) 1178 m_freem(m); 1179 1180 m = m_gethdr(M_NOWAIT, MT_HEADER); 1181 if (m == NULL) 1182 return (ENOBUFS); 1183 m->m_data += max_linkhdr; 1184 m->m_len = tlen; 1185 m->m_pkthdr.len = tlen; 1186 m->m_pkthdr.rcvif = NULL; 1187 if (tcp_prio_synack) 1188 m->m_flags |= M_PRIO; 1189 1190 if (isipv6) { 1191 ip6 = mtod(m, struct ip6_hdr *); 1192 ip6->ip6_vfc = IPV6_VERSION; 1193 ip6->ip6_nxt = IPPROTO_TCP; 1194 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1195 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1196 ip6->ip6_plen = htons(tlen - hlen); 1197 /* ip6_hlim is set after checksum */ 1198 /* ip6_flow = ??? */ 1199 1200 th = (struct tcphdr *)(ip6 + 1); 1201 } else { 1202 ip = mtod(m, struct ip *); 1203 ip->ip_v = IPVERSION; 1204 ip->ip_hl = sizeof(struct ip) >> 2; 1205 ip->ip_len = tlen; 1206 ip->ip_id = 0; 1207 ip->ip_off = 0; 1208 ip->ip_sum = 0; 1209 ip->ip_p = IPPROTO_TCP; 1210 ip->ip_src = sc->sc_inc.inc_laddr; 1211 ip->ip_dst = sc->sc_inc.inc_faddr; 1212 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1213 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1214 1215 /* 1216 * See if we should do MTU discovery. Route lookups are 1217 * expensive, so we will only unset the DF bit if: 1218 * 1219 * 1) path_mtu_discovery is disabled 1220 * 2) the SCF_UNREACH flag has been set 1221 */ 1222 if (path_mtu_discovery 1223 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1224 ip->ip_off |= IP_DF; 1225 } 1226 1227 th = (struct tcphdr *)(ip + 1); 1228 } 1229 th->th_sport = sc->sc_inc.inc_lport; 1230 th->th_dport = sc->sc_inc.inc_fport; 1231 1232 th->th_seq = htonl(sc->sc_iss); 1233 th->th_ack = htonl(sc->sc_irs + 1); 1234 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1235 th->th_x2 = 0; 1236 th->th_flags = TH_SYN | TH_ACK; 1237 th->th_win = htons(sc->sc_wnd); 1238 th->th_urp = 0; 1239 1240 /* Tack on the TCP options. */ 1241 if (optlen == 0) 1242 goto no_options; 1243 optp = (u_int8_t *)(th + 1); 1244 *optp++ = TCPOPT_MAXSEG; 1245 *optp++ = TCPOLEN_MAXSEG; 1246 *optp++ = (mssopt >> 8) & 0xff; 1247 *optp++ = mssopt & 0xff; 1248 1249 if (sc->sc_flags & SCF_WINSCALE) { 1250 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1251 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1252 sc->sc_request_r_scale); 1253 optp += 4; 1254 } 1255 1256 if (sc->sc_flags & SCF_TIMESTAMP) { 1257 u_int32_t *lp = (u_int32_t *)(optp); 1258 1259 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1260 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1261 *lp++ = htonl(ticks); 1262 *lp = htonl(sc->sc_tsrecent); 1263 optp += TCPOLEN_TSTAMP_APPA; 1264 } 1265 1266 #ifdef TCP_SIGNATURE 1267 /* 1268 * Handle TCP-MD5 passive opener response. 1269 */ 1270 if (sc->sc_flags & SCF_SIGNATURE) { 1271 u_int8_t *bp = optp; 1272 int i; 1273 1274 *bp++ = TCPOPT_SIGNATURE; 1275 *bp++ = TCPOLEN_SIGNATURE; 1276 for (i = 0; i < TCP_SIGLEN; i++) 1277 *bp++ = 0; 1278 tcpsignature_compute(m, 0, optlen, 1279 optp + 2, IPSEC_DIR_OUTBOUND); 1280 *bp++ = TCPOPT_NOP; 1281 *bp++ = TCPOPT_EOL; 1282 optp += TCPOLEN_SIGNATURE + 2; 1283 } 1284 #endif /* TCP_SIGNATURE */ 1285 1286 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1287 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1288 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1289 } 1290 1291 no_options: 1292 if (isipv6) { 1293 struct route_in6 *ro6 = &sc->sc_route6; 1294 1295 th->th_sum = 0; 1296 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1297 ip6->ip6_hlim = in6_selecthlim(NULL, 1298 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1299 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1300 sc->sc_tp->t_inpcb); 1301 } else { 1302 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1303 htons(tlen - hlen + IPPROTO_TCP)); 1304 m->m_pkthdr.csum_flags = CSUM_TCP; 1305 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1306 m->m_pkthdr.csum_thlen = sizeof(struct tcphdr) + optlen; 1307 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 1308 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb); 1309 } 1310 return (error); 1311 } 1312 1313 /* 1314 * cookie layers: 1315 * 1316 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1317 * | peer iss | 1318 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1319 * | 0 |(A)| | 1320 * (A): peer mss index 1321 */ 1322 1323 /* 1324 * The values below are chosen to minimize the size of the tcp_secret 1325 * table, as well as providing roughly a 16 second lifetime for the cookie. 1326 */ 1327 1328 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1329 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1330 1331 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1332 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1333 #define SYNCOOKIE_TIMEOUT \ 1334 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1335 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1336 1337 static struct { 1338 u_int32_t ts_secbits[4]; 1339 u_int ts_expire; 1340 } tcp_secret[SYNCOOKIE_NSECRETS]; 1341 1342 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1343 1344 static MD5_CTX syn_ctx; 1345 1346 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1347 1348 struct md5_add { 1349 u_int32_t laddr, faddr; 1350 u_int32_t secbits[4]; 1351 u_int16_t lport, fport; 1352 }; 1353 1354 #ifdef CTASSERT 1355 CTASSERT(sizeof(struct md5_add) == 28); 1356 #endif 1357 1358 /* 1359 * Consider the problem of a recreated (and retransmitted) cookie. If the 1360 * original SYN was accepted, the connection is established. The second 1361 * SYN is inflight, and if it arrives with an ISN that falls within the 1362 * receive window, the connection is killed. 1363 * 1364 * However, since cookies have other problems, this may not be worth 1365 * worrying about. 1366 */ 1367 1368 static u_int32_t 1369 syncookie_generate(struct syncache *sc) 1370 { 1371 u_int32_t md5_buffer[4]; 1372 u_int32_t data; 1373 int idx, i; 1374 struct md5_add add; 1375 #ifdef INET6 1376 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1377 #else 1378 const boolean_t isipv6 = FALSE; 1379 #endif 1380 1381 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1382 if (tcp_secret[idx].ts_expire < ticks) { 1383 for (i = 0; i < 4; i++) 1384 tcp_secret[idx].ts_secbits[i] = karc4random(); 1385 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1386 } 1387 for (data = NELEM(tcp_msstab) - 1; data > 0; data--) 1388 if (tcp_msstab[data] <= sc->sc_peer_mss) 1389 break; 1390 data = (data << SYNCOOKIE_WNDBITS) | idx; 1391 data ^= sc->sc_irs; /* peer's iss */ 1392 MD5Init(&syn_ctx); 1393 if (isipv6) { 1394 MD5Add(sc->sc_inc.inc6_laddr); 1395 MD5Add(sc->sc_inc.inc6_faddr); 1396 add.laddr = 0; 1397 add.faddr = 0; 1398 } else { 1399 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1400 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1401 } 1402 add.lport = sc->sc_inc.inc_lport; 1403 add.fport = sc->sc_inc.inc_fport; 1404 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1405 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1406 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1407 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1408 MD5Add(add); 1409 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1410 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1411 return (data); 1412 } 1413 1414 static struct syncache * 1415 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so) 1416 { 1417 u_int32_t md5_buffer[4]; 1418 struct syncache *sc; 1419 u_int32_t data; 1420 int wnd, idx; 1421 struct md5_add add; 1422 1423 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1424 idx = data & SYNCOOKIE_WNDMASK; 1425 if (tcp_secret[idx].ts_expire < ticks || 1426 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1427 return (NULL); 1428 MD5Init(&syn_ctx); 1429 #ifdef INET6 1430 if (inc->inc_isipv6) { 1431 MD5Add(inc->inc6_laddr); 1432 MD5Add(inc->inc6_faddr); 1433 add.laddr = 0; 1434 add.faddr = 0; 1435 } else 1436 #endif 1437 { 1438 add.laddr = inc->inc_laddr.s_addr; 1439 add.faddr = inc->inc_faddr.s_addr; 1440 } 1441 add.lport = inc->inc_lport; 1442 add.fport = inc->inc_fport; 1443 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1444 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1445 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1446 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1447 MD5Add(add); 1448 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1449 data ^= md5_buffer[0]; 1450 if (data & ~SYNCOOKIE_DATAMASK) 1451 return (NULL); 1452 data = data >> SYNCOOKIE_WNDBITS; 1453 1454 /* 1455 * Fill in the syncache values. 1456 * XXX duplicate code from syncache_add 1457 */ 1458 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1459 sc->sc_ipopts = NULL; 1460 sc->sc_inc.inc_fport = inc->inc_fport; 1461 sc->sc_inc.inc_lport = inc->inc_lport; 1462 #ifdef INET6 1463 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1464 if (inc->inc_isipv6) { 1465 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1466 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1467 sc->sc_route6.ro_rt = NULL; 1468 } else 1469 #endif 1470 { 1471 sc->sc_inc.inc_faddr = inc->inc_faddr; 1472 sc->sc_inc.inc_laddr = inc->inc_laddr; 1473 sc->sc_route.ro_rt = NULL; 1474 } 1475 sc->sc_irs = th->th_seq - 1; 1476 sc->sc_iss = th->th_ack - 1; 1477 wnd = ssb_space(&so->so_rcv); 1478 wnd = imax(wnd, 0); 1479 wnd = imin(wnd, TCP_MAXWIN); 1480 sc->sc_wnd = wnd; 1481 sc->sc_flags = 0; 1482 sc->sc_rxtslot = 0; 1483 sc->sc_peer_mss = tcp_msstab[data]; 1484 return (sc); 1485 } 1486