1 /* 2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * All advertising materials mentioning features or use of this software 36 * must display the following acknowledgement: 37 * This product includes software developed by Jeffrey M. Hsu. 38 * 39 * Copyright (c) 2001 Networks Associates Technologies, Inc. 40 * All rights reserved. 41 * 42 * This software was developed for the FreeBSD Project by Jonathan Lemon 43 * and NAI Labs, the Security Research Division of Network Associates, Inc. 44 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 45 * DARPA CHATS research program. 46 * 47 * Redistribution and use in source and binary forms, with or without 48 * modification, are permitted provided that the following conditions 49 * are met: 50 * 1. Redistributions of source code must retain the above copyright 51 * notice, this list of conditions and the following disclaimer. 52 * 2. Redistributions in binary form must reproduce the above copyright 53 * notice, this list of conditions and the following disclaimer in the 54 * documentation and/or other materials provided with the distribution. 55 * 3. The name of the author may not be used to endorse or promote 56 * products derived from this software without specific prior written 57 * permission. 58 * 59 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 62 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 69 * SUCH DAMAGE. 70 * 71 * $FreeBSD: src/sys/netinet/tcp_syncache.c,v 1.5.2.14 2003/02/24 04:02:27 silby Exp $ 72 */ 73 74 #include "opt_inet.h" 75 #include "opt_inet6.h" 76 #include "opt_ipsec.h" 77 78 #include <sys/param.h> 79 #include <sys/systm.h> 80 #include <sys/kernel.h> 81 #include <sys/sysctl.h> 82 #include <sys/malloc.h> 83 #include <sys/mbuf.h> 84 #include <sys/md5.h> 85 #include <sys/proc.h> /* for proc0 declaration */ 86 #include <sys/random.h> 87 #include <sys/socket.h> 88 #include <sys/socketvar.h> 89 #include <sys/in_cksum.h> 90 91 #include <sys/msgport2.h> 92 #include <net/netmsg2.h> 93 94 #include <net/if.h> 95 #include <net/route.h> 96 97 #include <netinet/in.h> 98 #include <netinet/in_systm.h> 99 #include <netinet/ip.h> 100 #include <netinet/in_var.h> 101 #include <netinet/in_pcb.h> 102 #include <netinet/ip_var.h> 103 #include <netinet/ip6.h> 104 #ifdef INET6 105 #include <netinet/icmp6.h> 106 #include <netinet6/nd6.h> 107 #endif 108 #include <netinet6/ip6_var.h> 109 #include <netinet6/in6_pcb.h> 110 #include <netinet/tcp.h> 111 #include <netinet/tcp_fsm.h> 112 #include <netinet/tcp_seq.h> 113 #include <netinet/tcp_timer.h> 114 #include <netinet/tcp_timer2.h> 115 #include <netinet/tcp_var.h> 116 #include <netinet6/tcp6_var.h> 117 118 #ifdef IPSEC 119 #include <netinet6/ipsec.h> 120 #ifdef INET6 121 #include <netinet6/ipsec6.h> 122 #endif 123 #include <netproto/key/key.h> 124 #endif /*IPSEC*/ 125 126 #ifdef FAST_IPSEC 127 #include <netproto/ipsec/ipsec.h> 128 #ifdef INET6 129 #include <netproto/ipsec/ipsec6.h> 130 #endif 131 #include <netproto/ipsec/key.h> 132 #define IPSEC 133 #endif /*FAST_IPSEC*/ 134 135 static int tcp_syncookies = 1; 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 137 &tcp_syncookies, 0, 138 "Use TCP SYN cookies if the syncache overflows"); 139 140 static void syncache_drop(struct syncache *, struct syncache_head *); 141 static void syncache_free(struct syncache *); 142 static void syncache_insert(struct syncache *, struct syncache_head *); 143 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 144 static int syncache_respond(struct syncache *, struct mbuf *); 145 static struct socket *syncache_socket(struct syncache *, struct socket *, 146 struct mbuf *); 147 static void syncache_timer(void *); 148 static u_int32_t syncookie_generate(struct syncache *); 149 static struct syncache *syncookie_lookup(struct in_conninfo *, 150 struct tcphdr *, struct socket *); 151 152 /* 153 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 154 * 4 retransmits corresponds to a timeout of (3 + 3 + 3 + 3 + 3 == 15) seconds 155 * or (1 + 1 + 2 + 4 + 8 == 16) seconds if RFC6298 is used, the odds are that 156 * the user has given up attempting to connect by then. 157 */ 158 #define SYNCACHE_MAXREXMTS 4 159 160 /* Arbitrary values */ 161 #define TCP_SYNCACHE_HASHSIZE 512 162 #define TCP_SYNCACHE_BUCKETLIMIT 30 163 164 struct netmsg_sc_timer { 165 struct netmsg_base base; 166 struct msgrec *nm_mrec; /* back pointer to containing msgrec */ 167 }; 168 169 struct msgrec { 170 struct netmsg_sc_timer msg; 171 lwkt_port_t port; /* constant after init */ 172 int slot; /* constant after init */ 173 }; 174 175 static void syncache_timer_handler(netmsg_t); 176 177 struct tcp_syncache { 178 u_int hashsize; 179 u_int hashmask; 180 u_int bucket_limit; 181 u_int cache_limit; 182 u_int rexmt_limit; 183 u_int hash_secret; 184 }; 185 static struct tcp_syncache tcp_syncache; 186 187 TAILQ_HEAD(syncache_list, syncache); 188 189 struct tcp_syncache_percpu { 190 struct syncache_head *hashbase; 191 u_int cache_count; 192 struct syncache_list timerq[SYNCACHE_MAXREXMTS + 1]; 193 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1]; 194 struct msgrec mrec[SYNCACHE_MAXREXMTS + 1]; 195 }; 196 static struct tcp_syncache_percpu tcp_syncache_percpu[MAXCPU]; 197 198 static struct lwkt_port syncache_null_rport; 199 200 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 201 202 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD, 203 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 204 205 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD, 206 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 207 208 /* XXX JH */ 209 #if 0 210 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 211 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 212 #endif 213 214 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD, 215 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 216 217 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 218 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 219 220 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 221 222 #define SYNCACHE_HASH(inc, mask) \ 223 ((tcp_syncache.hash_secret ^ \ 224 (inc)->inc_faddr.s_addr ^ \ 225 ((inc)->inc_faddr.s_addr >> 16) ^ \ 226 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 227 228 #define SYNCACHE_HASH6(inc, mask) \ 229 ((tcp_syncache.hash_secret ^ \ 230 (inc)->inc6_faddr.s6_addr32[0] ^ \ 231 (inc)->inc6_faddr.s6_addr32[3] ^ \ 232 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 233 234 #define ENDPTS_EQ(a, b) ( \ 235 (a)->ie_fport == (b)->ie_fport && \ 236 (a)->ie_lport == (b)->ie_lport && \ 237 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 238 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 239 ) 240 241 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 242 243 static __inline int 244 syncache_rto(int slot) 245 { 246 if (tcp_low_rtobase) 247 return (TCPTV_RTOBASE * tcp_syn_backoff_low[slot]); 248 else 249 return (TCPTV_RTOBASE * tcp_syn_backoff[slot]); 250 } 251 252 static __inline void 253 syncache_timeout(struct tcp_syncache_percpu *syncache_percpu, 254 struct syncache *sc, int slot) 255 { 256 int rto; 257 258 if (slot > 0) { 259 /* 260 * Record the time that we spent in SYN|ACK 261 * retransmition. 262 * 263 * Needed by RFC3390 and RFC6298. 264 */ 265 sc->sc_rxtused += syncache_rto(slot - 1); 266 } 267 sc->sc_rxtslot = slot; 268 269 rto = syncache_rto(slot); 270 sc->sc_rxttime = ticks + rto; 271 272 TAILQ_INSERT_TAIL(&syncache_percpu->timerq[slot], sc, sc_timerq); 273 if (!callout_active(&syncache_percpu->tt_timerq[slot])) { 274 callout_reset(&syncache_percpu->tt_timerq[slot], rto, 275 syncache_timer, &syncache_percpu->mrec[slot]); 276 } 277 } 278 279 static void 280 syncache_free(struct syncache *sc) 281 { 282 struct rtentry *rt; 283 #ifdef INET6 284 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 285 #else 286 const boolean_t isipv6 = FALSE; 287 #endif 288 289 if (sc->sc_ipopts) 290 m_free(sc->sc_ipopts); 291 292 rt = isipv6 ? sc->sc_route6.ro_rt : sc->sc_route.ro_rt; 293 if (rt != NULL) { 294 /* 295 * If this is the only reference to a protocol-cloned 296 * route, remove it immediately. 297 */ 298 if ((rt->rt_flags & 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 RTFREE(rt); 302 } 303 kfree(sc, M_SYNCACHE); 304 } 305 306 void 307 syncache_init(void) 308 { 309 int i, cpu; 310 311 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 312 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 313 tcp_syncache.cache_limit = 314 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 315 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 316 tcp_syncache.hash_secret = karc4random(); 317 318 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 319 &tcp_syncache.hashsize); 320 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 321 &tcp_syncache.cache_limit); 322 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 323 &tcp_syncache.bucket_limit); 324 if (!powerof2(tcp_syncache.hashsize)) { 325 kprintf("WARNING: syncache hash size is not a power of 2.\n"); 326 tcp_syncache.hashsize = 512; /* safe default */ 327 } 328 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 329 330 lwkt_initport_replyonly_null(&syncache_null_rport); 331 332 for (cpu = 0; cpu < ncpus2; cpu++) { 333 struct tcp_syncache_percpu *syncache_percpu; 334 335 syncache_percpu = &tcp_syncache_percpu[cpu]; 336 /* Allocate the hash table. */ 337 syncache_percpu->hashbase = kmalloc(tcp_syncache.hashsize * sizeof(struct syncache_head), 338 M_SYNCACHE, M_WAITOK); 339 340 /* Initialize the hash buckets. */ 341 for (i = 0; i < tcp_syncache.hashsize; i++) { 342 struct syncache_head *bucket; 343 344 bucket = &syncache_percpu->hashbase[i]; 345 TAILQ_INIT(&bucket->sch_bucket); 346 bucket->sch_length = 0; 347 } 348 349 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 350 /* Initialize the timer queues. */ 351 TAILQ_INIT(&syncache_percpu->timerq[i]); 352 callout_init_mp(&syncache_percpu->tt_timerq[i]); 353 354 syncache_percpu->mrec[i].slot = i; 355 syncache_percpu->mrec[i].port = cpu_portfn(cpu); 356 syncache_percpu->mrec[i].msg.nm_mrec = 357 &syncache_percpu->mrec[i]; 358 netmsg_init(&syncache_percpu->mrec[i].msg.base, 359 NULL, &syncache_null_rport, 360 0, syncache_timer_handler); 361 } 362 } 363 } 364 365 static void 366 syncache_insert(struct syncache *sc, struct syncache_head *sch) 367 { 368 struct tcp_syncache_percpu *syncache_percpu; 369 struct syncache *sc2; 370 int i; 371 372 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 373 374 /* 375 * Make sure that we don't overflow the per-bucket 376 * limit or the total cache size limit. 377 */ 378 if (sch->sch_length >= tcp_syncache.bucket_limit) { 379 /* 380 * The bucket is full, toss the oldest element. 381 */ 382 sc2 = TAILQ_FIRST(&sch->sch_bucket); 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 sc2->sc_tp->ts_recent = ticks; 401 syncache_drop(sc2, NULL); 402 tcpstat.tcps_sc_cacheoverflow++; 403 } 404 405 /* Initialize the entry's timer. */ 406 syncache_timeout(syncache_percpu, sc, 0); 407 408 /* Put it into the bucket. */ 409 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 410 sch->sch_length++; 411 syncache_percpu->cache_count++; 412 tcpstat.tcps_sc_added++; 413 } 414 415 void 416 syncache_destroy(struct tcpcb *tp) 417 { 418 struct tcp_syncache_percpu *syncache_percpu; 419 struct syncache_head *bucket; 420 struct syncache *sc; 421 int i; 422 423 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 424 sc = NULL; 425 426 for (i = 0; i < tcp_syncache.hashsize; i++) { 427 bucket = &syncache_percpu->hashbase[i]; 428 TAILQ_FOREACH(sc, &bucket->sch_bucket, sc_hash) { 429 if (sc->sc_tp == tp) 430 sc->sc_tp = NULL; 431 } 432 } 433 } 434 435 static void 436 syncache_drop(struct syncache *sc, struct syncache_head *sch) 437 { 438 struct tcp_syncache_percpu *syncache_percpu; 439 #ifdef INET6 440 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 441 #else 442 const boolean_t isipv6 = FALSE; 443 #endif 444 445 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 446 447 if (sch == NULL) { 448 if (isipv6) { 449 sch = &syncache_percpu->hashbase[ 450 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 451 } else { 452 sch = &syncache_percpu->hashbase[ 453 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 454 } 455 } 456 457 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 458 sch->sch_length--; 459 syncache_percpu->cache_count--; 460 461 /* 462 * Cleanup 463 */ 464 if (sc->sc_tp) 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(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, 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 if (in_pcbinsporthash(inp) != 0) { 765 /* 766 * Undo the assignments above if we failed to 767 * put the PCB on the hash lists. 768 */ 769 if (isipv6) 770 inp->in6p_laddr = kin6addr_any; 771 else 772 inp->inp_laddr.s_addr = INADDR_ANY; 773 inp->inp_lport = 0; 774 goto abort; 775 } 776 linp = lso->so_pcb; 777 #ifdef IPSEC 778 /* copy old policy into new socket's */ 779 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 780 kprintf("syncache_expand: could not copy policy\n"); 781 #endif 782 if (isipv6) { 783 struct in6_addr laddr6; 784 /* 785 * Inherit socket options from the listening socket. 786 * Note that in6p_inputopts are not (and should not be) 787 * copied, since it stores previously received options and is 788 * used to detect if each new option is different than the 789 * previous one and hence should be passed to a user. 790 * If we copied in6p_inputopts, a user would not be able to 791 * receive options just after calling the accept system call. 792 */ 793 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 794 if (linp->in6p_outputopts) 795 inp->in6p_outputopts = 796 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 797 inp->in6p_route = sc->sc_route6; 798 sc->sc_route6.ro_rt = NULL; 799 800 laddr6 = inp->in6p_laddr; 801 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 802 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 803 if (in6_pcbconnect(inp, faddr, &thread0)) { 804 inp->in6p_laddr = laddr6; 805 goto abort; 806 } 807 } else { 808 struct in_addr laddr; 809 810 inp->inp_options = ip_srcroute(m); 811 if (inp->inp_options == NULL) { 812 inp->inp_options = sc->sc_ipopts; 813 sc->sc_ipopts = NULL; 814 } 815 inp->inp_route = sc->sc_route; 816 sc->sc_route.ro_rt = NULL; 817 818 laddr = inp->inp_laddr; 819 if (inp->inp_laddr.s_addr == INADDR_ANY) 820 inp->inp_laddr = sc->sc_inc.inc_laddr; 821 if (in_pcbconnect(inp, faddr, &thread0)) { 822 inp->inp_laddr = laddr; 823 goto abort; 824 } 825 } 826 827 /* 828 * The current port should be in the context of the SYN+ACK and 829 * so should match the tcp address port. 830 * 831 * XXX we may be running on the netisr thread instead of a tcp 832 * thread, in which case port will not match 833 * curthread->td_msgport. 834 */ 835 if (isipv6) { 836 port = tcp6_addrport(); 837 } else { 838 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport, 839 inp->inp_laddr.s_addr, inp->inp_lport); 840 } 841 if (port != &curthread->td_msgport) { 842 print_backtrace(-1); 843 kprintf("TCP PORT MISMATCH %p vs %p\n", 844 port, &curthread->td_msgport); 845 } 846 /*KKASSERT(port == &curthread->td_msgport);*/ 847 848 tp = intotcpcb(inp); 849 tp->t_state = TCPS_SYN_RECEIVED; 850 tp->iss = sc->sc_iss; 851 tp->irs = sc->sc_irs; 852 tcp_rcvseqinit(tp); 853 tcp_sendseqinit(tp); 854 tp->snd_wl1 = sc->sc_irs; 855 tp->rcv_up = sc->sc_irs + 1; 856 tp->rcv_wnd = sc->sc_wnd; 857 tp->rcv_adv += tp->rcv_wnd; 858 859 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 860 if (sc->sc_flags & SCF_NOOPT) 861 tp->t_flags |= TF_NOOPT; 862 if (sc->sc_flags & SCF_WINSCALE) { 863 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 864 tp->requested_s_scale = sc->sc_requested_s_scale; 865 tp->request_r_scale = sc->sc_request_r_scale; 866 } 867 if (sc->sc_flags & SCF_TIMESTAMP) { 868 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 869 tp->ts_recent = sc->sc_tsrecent; 870 tp->ts_recent_age = ticks; 871 } 872 if (sc->sc_flags & SCF_SACK_PERMITTED) 873 tp->t_flags |= TF_SACK_PERMITTED; 874 875 #ifdef TCP_SIGNATURE 876 if (sc->sc_flags & SCF_SIGNATURE) 877 tp->t_flags |= TF_SIGNATURE; 878 #endif /* TCP_SIGNATURE */ 879 880 tp->t_rxtsyn = sc->sc_rxtused; 881 tcp_mss(tp, sc->sc_peer_mss); 882 883 /* 884 * Inherit some properties from the listen socket 885 */ 886 ltp = intotcpcb(linp); 887 tp->t_keepinit = ltp->t_keepinit; 888 tp->t_keepidle = ltp->t_keepidle; 889 tp->t_keepintvl = ltp->t_keepintvl; 890 tp->t_keepcnt = ltp->t_keepcnt; 891 tp->t_maxidle = ltp->t_maxidle; 892 893 tcp_create_timermsg(tp, port); 894 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep); 895 896 tcpstat.tcps_accepts++; 897 return (so); 898 899 abort: 900 if (so != NULL) 901 soabort_oncpu(so); 902 return (NULL); 903 } 904 905 /* 906 * This function gets called when we receive an ACK for a 907 * socket in the LISTEN state. We look up the connection 908 * in the syncache, and if its there, we pull it out of 909 * the cache and turn it into a full-blown connection in 910 * the SYN-RECEIVED state. 911 */ 912 int 913 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop, 914 struct mbuf *m) 915 { 916 struct syncache *sc; 917 struct syncache_head *sch; 918 struct socket *so; 919 920 sc = syncache_lookup(inc, &sch); 921 if (sc == NULL) { 922 /* 923 * There is no syncache entry, so see if this ACK is 924 * a returning syncookie. To do this, first: 925 * A. See if this socket has had a syncache entry dropped in 926 * the past. We don't want to accept a bogus syncookie 927 * if we've never received a SYN. 928 * B. check that the syncookie is valid. If it is, then 929 * cobble up a fake syncache entry, and return. 930 */ 931 if (!tcp_syncookies) 932 return (0); 933 sc = syncookie_lookup(inc, th, *sop); 934 if (sc == NULL) 935 return (0); 936 sch = NULL; 937 tcpstat.tcps_sc_recvcookie++; 938 } 939 940 /* 941 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 942 */ 943 if (th->th_ack != sc->sc_iss + 1) 944 return (0); 945 946 so = syncache_socket(sc, *sop, m); 947 if (so == NULL) { 948 #if 0 949 resetandabort: 950 /* XXXjlemon check this - is this correct? */ 951 tcp_respond(NULL, m, m, th, 952 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 953 #endif 954 m_freem(m); /* XXX only needed for above */ 955 tcpstat.tcps_sc_aborted++; 956 } else { 957 tcpstat.tcps_sc_completed++; 958 } 959 if (sch == NULL) 960 syncache_free(sc); 961 else 962 syncache_drop(sc, sch); 963 *sop = so; 964 return (1); 965 } 966 967 /* 968 * Given a LISTEN socket and an inbound SYN request, add 969 * this to the syn cache, and send back a segment: 970 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 971 * to the source. 972 * 973 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 974 * Doing so would require that we hold onto the data and deliver it 975 * to the application. However, if we are the target of a SYN-flood 976 * DoS attack, an attacker could send data which would eventually 977 * consume all available buffer space if it were ACKed. By not ACKing 978 * the data, we avoid this DoS scenario. 979 */ 980 int 981 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 982 struct socket *so, struct mbuf *m) 983 { 984 struct tcp_syncache_percpu *syncache_percpu; 985 struct tcpcb *tp; 986 struct syncache *sc = NULL; 987 struct syncache_head *sch; 988 struct mbuf *ipopts = NULL; 989 int win; 990 991 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 992 tp = sototcpcb(so); 993 994 /* 995 * Remember the IP options, if any. 996 */ 997 #ifdef INET6 998 if (!inc->inc_isipv6) 999 #endif 1000 ipopts = ip_srcroute(m); 1001 1002 /* 1003 * See if we already have an entry for this connection. 1004 * If we do, resend the SYN,ACK, and reset the retransmit timer. 1005 * 1006 * XXX 1007 * The syncache should be re-initialized with the contents 1008 * of the new SYN which may have different options. 1009 */ 1010 sc = syncache_lookup(inc, &sch); 1011 if (sc != NULL) { 1012 tcpstat.tcps_sc_dupsyn++; 1013 if (ipopts) { 1014 /* 1015 * If we were remembering a previous source route, 1016 * forget it and use the new one we've been given. 1017 */ 1018 if (sc->sc_ipopts) 1019 m_free(sc->sc_ipopts); 1020 sc->sc_ipopts = ipopts; 1021 } 1022 /* 1023 * Update timestamp if present. 1024 */ 1025 if (sc->sc_flags & SCF_TIMESTAMP) 1026 sc->sc_tsrecent = to->to_tsval; 1027 1028 /* Just update the TOF_SACK_PERMITTED for now. */ 1029 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1030 sc->sc_flags |= SCF_SACK_PERMITTED; 1031 else 1032 sc->sc_flags &= ~SCF_SACK_PERMITTED; 1033 1034 /* 1035 * PCB may have changed, pick up new values. 1036 */ 1037 sc->sc_tp = tp; 1038 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1039 if (syncache_respond(sc, m) == 0) { 1040 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 1041 sc, sc_timerq); 1042 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 1043 tcpstat.tcps_sndacks++; 1044 tcpstat.tcps_sndtotal++; 1045 } 1046 return (1); 1047 } 1048 1049 /* 1050 * Fill in the syncache values. 1051 */ 1052 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1053 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1054 sc->sc_ipopts = ipopts; 1055 sc->sc_inc.inc_fport = inc->inc_fport; 1056 sc->sc_inc.inc_lport = inc->inc_lport; 1057 sc->sc_tp = tp; 1058 #ifdef INET6 1059 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1060 if (inc->inc_isipv6) { 1061 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1062 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1063 sc->sc_route6.ro_rt = NULL; 1064 } else 1065 #endif 1066 { 1067 sc->sc_inc.inc_faddr = inc->inc_faddr; 1068 sc->sc_inc.inc_laddr = inc->inc_laddr; 1069 sc->sc_route.ro_rt = NULL; 1070 } 1071 sc->sc_irs = th->th_seq; 1072 sc->sc_flags = 0; 1073 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 1074 if (tcp_syncookies) 1075 sc->sc_iss = syncookie_generate(sc); 1076 else 1077 sc->sc_iss = karc4random(); 1078 1079 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */ 1080 win = ssb_space(&so->so_rcv); 1081 win = imax(win, 0); 1082 win = imin(win, TCP_MAXWIN); 1083 sc->sc_wnd = win; 1084 1085 if (tcp_do_rfc1323) { 1086 /* 1087 * A timestamp received in a SYN makes 1088 * it ok to send timestamp requests and replies. 1089 */ 1090 if (to->to_flags & TOF_TS) { 1091 sc->sc_tsrecent = to->to_tsval; 1092 sc->sc_flags |= SCF_TIMESTAMP; 1093 } 1094 if (to->to_flags & TOF_SCALE) { 1095 int wscale = TCP_MIN_WINSHIFT; 1096 1097 /* Compute proper scaling value from buffer space */ 1098 while (wscale < TCP_MAX_WINSHIFT && 1099 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) { 1100 wscale++; 1101 } 1102 sc->sc_request_r_scale = wscale; 1103 sc->sc_requested_s_scale = to->to_requested_s_scale; 1104 sc->sc_flags |= SCF_WINSCALE; 1105 } 1106 } 1107 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1108 sc->sc_flags |= SCF_SACK_PERMITTED; 1109 if (tp->t_flags & TF_NOOPT) 1110 sc->sc_flags = SCF_NOOPT; 1111 #ifdef TCP_SIGNATURE 1112 /* 1113 * If listening socket requested TCP digests, and received SYN 1114 * contains the option, flag this in the syncache so that 1115 * syncache_respond() will do the right thing with the SYN+ACK. 1116 * XXX Currently we always record the option by default and will 1117 * attempt to use it in syncache_respond(). 1118 */ 1119 if (to->to_flags & TOF_SIGNATURE) 1120 sc->sc_flags = SCF_SIGNATURE; 1121 #endif /* TCP_SIGNATURE */ 1122 1123 if (syncache_respond(sc, m) == 0) { 1124 syncache_insert(sc, sch); 1125 tcpstat.tcps_sndacks++; 1126 tcpstat.tcps_sndtotal++; 1127 } else { 1128 syncache_free(sc); 1129 tcpstat.tcps_sc_dropped++; 1130 } 1131 return (1); 1132 } 1133 1134 static int 1135 syncache_respond(struct syncache *sc, struct mbuf *m) 1136 { 1137 u_int8_t *optp; 1138 int optlen, error; 1139 u_int16_t tlen, hlen, mssopt; 1140 struct ip *ip = NULL; 1141 struct rtentry *rt; 1142 struct tcphdr *th; 1143 struct ip6_hdr *ip6 = NULL; 1144 #ifdef INET6 1145 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1146 #else 1147 const boolean_t isipv6 = FALSE; 1148 #endif 1149 1150 if (isipv6) { 1151 rt = tcp_rtlookup6(&sc->sc_inc); 1152 if (rt != NULL) 1153 mssopt = rt->rt_ifp->if_mtu - 1154 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1155 else 1156 mssopt = tcp_v6mssdflt; 1157 hlen = sizeof(struct ip6_hdr); 1158 } else { 1159 rt = tcp_rtlookup(&sc->sc_inc); 1160 if (rt != NULL) 1161 mssopt = rt->rt_ifp->if_mtu - 1162 (sizeof(struct ip) + sizeof(struct tcphdr)); 1163 else 1164 mssopt = tcp_mssdflt; 1165 hlen = sizeof(struct ip); 1166 } 1167 1168 /* Compute the size of the TCP options. */ 1169 if (sc->sc_flags & SCF_NOOPT) { 1170 optlen = 0; 1171 } else { 1172 optlen = TCPOLEN_MAXSEG + 1173 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1174 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1175 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1176 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1177 #ifdef TCP_SIGNATURE 1178 optlen += ((sc->sc_flags & SCF_SIGNATURE) ? 1179 (TCPOLEN_SIGNATURE + 2) : 0); 1180 #endif /* TCP_SIGNATURE */ 1181 } 1182 tlen = hlen + sizeof(struct tcphdr) + optlen; 1183 1184 /* 1185 * XXX 1186 * assume that the entire packet will fit in a header mbuf 1187 */ 1188 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1189 1190 /* 1191 * XXX shouldn't this reuse the mbuf if possible ? 1192 * Create the IP+TCP header from scratch. 1193 */ 1194 if (m) 1195 m_freem(m); 1196 1197 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 1198 if (m == NULL) 1199 return (ENOBUFS); 1200 m->m_data += max_linkhdr; 1201 m->m_len = tlen; 1202 m->m_pkthdr.len = tlen; 1203 m->m_pkthdr.rcvif = NULL; 1204 1205 if (isipv6) { 1206 ip6 = mtod(m, struct ip6_hdr *); 1207 ip6->ip6_vfc = IPV6_VERSION; 1208 ip6->ip6_nxt = IPPROTO_TCP; 1209 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1210 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1211 ip6->ip6_plen = htons(tlen - hlen); 1212 /* ip6_hlim is set after checksum */ 1213 /* ip6_flow = ??? */ 1214 1215 th = (struct tcphdr *)(ip6 + 1); 1216 } else { 1217 ip = mtod(m, struct ip *); 1218 ip->ip_v = IPVERSION; 1219 ip->ip_hl = sizeof(struct ip) >> 2; 1220 ip->ip_len = tlen; 1221 ip->ip_id = 0; 1222 ip->ip_off = 0; 1223 ip->ip_sum = 0; 1224 ip->ip_p = IPPROTO_TCP; 1225 ip->ip_src = sc->sc_inc.inc_laddr; 1226 ip->ip_dst = sc->sc_inc.inc_faddr; 1227 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1228 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1229 1230 /* 1231 * See if we should do MTU discovery. Route lookups are 1232 * expensive, so we will only unset the DF bit if: 1233 * 1234 * 1) path_mtu_discovery is disabled 1235 * 2) the SCF_UNREACH flag has been set 1236 */ 1237 if (path_mtu_discovery 1238 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1239 ip->ip_off |= IP_DF; 1240 } 1241 1242 th = (struct tcphdr *)(ip + 1); 1243 } 1244 th->th_sport = sc->sc_inc.inc_lport; 1245 th->th_dport = sc->sc_inc.inc_fport; 1246 1247 th->th_seq = htonl(sc->sc_iss); 1248 th->th_ack = htonl(sc->sc_irs + 1); 1249 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1250 th->th_x2 = 0; 1251 th->th_flags = TH_SYN | TH_ACK; 1252 th->th_win = htons(sc->sc_wnd); 1253 th->th_urp = 0; 1254 1255 /* Tack on the TCP options. */ 1256 if (optlen == 0) 1257 goto no_options; 1258 optp = (u_int8_t *)(th + 1); 1259 *optp++ = TCPOPT_MAXSEG; 1260 *optp++ = TCPOLEN_MAXSEG; 1261 *optp++ = (mssopt >> 8) & 0xff; 1262 *optp++ = mssopt & 0xff; 1263 1264 if (sc->sc_flags & SCF_WINSCALE) { 1265 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1266 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1267 sc->sc_request_r_scale); 1268 optp += 4; 1269 } 1270 1271 if (sc->sc_flags & SCF_TIMESTAMP) { 1272 u_int32_t *lp = (u_int32_t *)(optp); 1273 1274 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1275 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1276 *lp++ = htonl(ticks); 1277 *lp = htonl(sc->sc_tsrecent); 1278 optp += TCPOLEN_TSTAMP_APPA; 1279 } 1280 1281 #ifdef TCP_SIGNATURE 1282 /* 1283 * Handle TCP-MD5 passive opener response. 1284 */ 1285 if (sc->sc_flags & SCF_SIGNATURE) { 1286 u_int8_t *bp = optp; 1287 int i; 1288 1289 *bp++ = TCPOPT_SIGNATURE; 1290 *bp++ = TCPOLEN_SIGNATURE; 1291 for (i = 0; i < TCP_SIGLEN; i++) 1292 *bp++ = 0; 1293 tcpsignature_compute(m, 0, optlen, 1294 optp + 2, IPSEC_DIR_OUTBOUND); 1295 *bp++ = TCPOPT_NOP; 1296 *bp++ = TCPOPT_EOL; 1297 optp += TCPOLEN_SIGNATURE + 2; 1298 } 1299 #endif /* TCP_SIGNATURE */ 1300 1301 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1302 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1303 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1304 } 1305 1306 no_options: 1307 if (isipv6) { 1308 struct route_in6 *ro6 = &sc->sc_route6; 1309 1310 th->th_sum = 0; 1311 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1312 ip6->ip6_hlim = in6_selecthlim(NULL, 1313 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1314 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1315 sc->sc_tp->t_inpcb); 1316 } else { 1317 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1318 htons(tlen - hlen + IPPROTO_TCP)); 1319 m->m_pkthdr.csum_flags = CSUM_TCP; 1320 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1321 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 1322 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb); 1323 } 1324 return (error); 1325 } 1326 1327 /* 1328 * cookie layers: 1329 * 1330 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1331 * | peer iss | 1332 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1333 * | 0 |(A)| | 1334 * (A): peer mss index 1335 */ 1336 1337 /* 1338 * The values below are chosen to minimize the size of the tcp_secret 1339 * table, as well as providing roughly a 16 second lifetime for the cookie. 1340 */ 1341 1342 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1343 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1344 1345 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1346 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1347 #define SYNCOOKIE_TIMEOUT \ 1348 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1349 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1350 1351 static struct { 1352 u_int32_t ts_secbits[4]; 1353 u_int ts_expire; 1354 } tcp_secret[SYNCOOKIE_NSECRETS]; 1355 1356 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1357 1358 static MD5_CTX syn_ctx; 1359 1360 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1361 1362 struct md5_add { 1363 u_int32_t laddr, faddr; 1364 u_int32_t secbits[4]; 1365 u_int16_t lport, fport; 1366 }; 1367 1368 #ifdef CTASSERT 1369 CTASSERT(sizeof(struct md5_add) == 28); 1370 #endif 1371 1372 /* 1373 * Consider the problem of a recreated (and retransmitted) cookie. If the 1374 * original SYN was accepted, the connection is established. The second 1375 * SYN is inflight, and if it arrives with an ISN that falls within the 1376 * receive window, the connection is killed. 1377 * 1378 * However, since cookies have other problems, this may not be worth 1379 * worrying about. 1380 */ 1381 1382 static u_int32_t 1383 syncookie_generate(struct syncache *sc) 1384 { 1385 u_int32_t md5_buffer[4]; 1386 u_int32_t data; 1387 int idx, i; 1388 struct md5_add add; 1389 #ifdef INET6 1390 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1391 #else 1392 const boolean_t isipv6 = FALSE; 1393 #endif 1394 1395 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1396 if (tcp_secret[idx].ts_expire < ticks) { 1397 for (i = 0; i < 4; i++) 1398 tcp_secret[idx].ts_secbits[i] = karc4random(); 1399 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1400 } 1401 for (data = NELEM(tcp_msstab) - 1; data > 0; data--) 1402 if (tcp_msstab[data] <= sc->sc_peer_mss) 1403 break; 1404 data = (data << SYNCOOKIE_WNDBITS) | idx; 1405 data ^= sc->sc_irs; /* peer's iss */ 1406 MD5Init(&syn_ctx); 1407 if (isipv6) { 1408 MD5Add(sc->sc_inc.inc6_laddr); 1409 MD5Add(sc->sc_inc.inc6_faddr); 1410 add.laddr = 0; 1411 add.faddr = 0; 1412 } else { 1413 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1414 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1415 } 1416 add.lport = sc->sc_inc.inc_lport; 1417 add.fport = sc->sc_inc.inc_fport; 1418 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1419 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1420 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1421 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1422 MD5Add(add); 1423 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1424 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1425 return (data); 1426 } 1427 1428 static struct syncache * 1429 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so) 1430 { 1431 u_int32_t md5_buffer[4]; 1432 struct syncache *sc; 1433 u_int32_t data; 1434 int wnd, idx; 1435 struct md5_add add; 1436 1437 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1438 idx = data & SYNCOOKIE_WNDMASK; 1439 if (tcp_secret[idx].ts_expire < ticks || 1440 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1441 return (NULL); 1442 MD5Init(&syn_ctx); 1443 #ifdef INET6 1444 if (inc->inc_isipv6) { 1445 MD5Add(inc->inc6_laddr); 1446 MD5Add(inc->inc6_faddr); 1447 add.laddr = 0; 1448 add.faddr = 0; 1449 } else 1450 #endif 1451 { 1452 add.laddr = inc->inc_laddr.s_addr; 1453 add.faddr = inc->inc_faddr.s_addr; 1454 } 1455 add.lport = inc->inc_lport; 1456 add.fport = inc->inc_fport; 1457 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1458 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1459 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1460 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1461 MD5Add(add); 1462 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1463 data ^= md5_buffer[0]; 1464 if (data & ~SYNCOOKIE_DATAMASK) 1465 return (NULL); 1466 data = data >> SYNCOOKIE_WNDBITS; 1467 1468 /* 1469 * Fill in the syncache values. 1470 * XXX duplicate code from syncache_add 1471 */ 1472 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1473 sc->sc_ipopts = NULL; 1474 sc->sc_inc.inc_fport = inc->inc_fport; 1475 sc->sc_inc.inc_lport = inc->inc_lport; 1476 #ifdef INET6 1477 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1478 if (inc->inc_isipv6) { 1479 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1480 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1481 sc->sc_route6.ro_rt = NULL; 1482 } else 1483 #endif 1484 { 1485 sc->sc_inc.inc_faddr = inc->inc_faddr; 1486 sc->sc_inc.inc_laddr = inc->inc_laddr; 1487 sc->sc_route.ro_rt = NULL; 1488 } 1489 sc->sc_irs = th->th_seq - 1; 1490 sc->sc_iss = th->th_ack - 1; 1491 wnd = ssb_space(&so->so_rcv); 1492 wnd = imax(wnd, 0); 1493 wnd = imin(wnd, TCP_MAXWIN); 1494 sc->sc_wnd = wnd; 1495 sc->sc_flags = 0; 1496 sc->sc_rxtslot = 0; 1497 sc->sc_peer_mss = tcp_msstab[data]; 1498 return (sc); 1499 } 1500