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 KASSERT(m->m_flags & M_HASH, ("mbuf has no hash")); 714 715 if (isipv6) { 716 faddr = (struct sockaddr *)&sin6_faddr; 717 sin6_faddr.sin6_family = AF_INET6; 718 sin6_faddr.sin6_len = sizeof(sin6_faddr); 719 sin6_faddr.sin6_addr = sc->sc_inc.inc6_faddr; 720 sin6_faddr.sin6_port = sc->sc_inc.inc_fport; 721 sin6_faddr.sin6_flowinfo = sin6_faddr.sin6_scope_id = 0; 722 } else { 723 faddr = (struct sockaddr *)&sin_faddr; 724 sin_faddr.sin_family = AF_INET; 725 sin_faddr.sin_len = sizeof(sin_faddr); 726 sin_faddr.sin_addr = sc->sc_inc.inc_faddr; 727 sin_faddr.sin_port = sc->sc_inc.inc_fport; 728 bzero(sin_faddr.sin_zero, sizeof(sin_faddr.sin_zero)); 729 } 730 731 /* 732 * Ok, create the full blown connection, and set things up 733 * as they would have been set up if we had created the 734 * connection when the SYN arrived. If we can't create 735 * the connection, abort it. 736 * 737 * Set the protocol processing port for the socket to the current 738 * port (that the connection came in on). 739 * 740 * NOTE: 741 * We don't keep a reference on the new socket, since its 742 * destruction will run in this thread (netisrN); there is no 743 * race here. 744 */ 745 so = sonewconn_faddr(lso, SS_ISCONNECTED, faddr, 746 FALSE /* don't ref */); 747 if (so == NULL) { 748 /* 749 * Drop the connection; we will send a RST if the peer 750 * retransmits the ACK, 751 */ 752 tcpstat.tcps_listendrop++; 753 goto abort; 754 } 755 756 /* 757 * Insert new socket into hash list. 758 */ 759 inp = so->so_pcb; 760 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 761 if (isipv6) { 762 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 763 } else { 764 KASSERT(INP_ISIPV4(inp), ("not inet pcb")); 765 inp->inp_laddr = sc->sc_inc.inc_laddr; 766 } 767 inp->inp_lport = sc->sc_inc.inc_lport; 768 769 linp = lso->so_pcb; 770 ltp = intotcpcb(linp); 771 772 tcp_pcbport_insert(ltp, inp); 773 774 #ifdef IPSEC 775 /* copy old policy into new socket's */ 776 if (ipsec_copy_policy(linp->inp_sp, inp->inp_sp)) 777 kprintf("syncache_expand: could not copy policy\n"); 778 #endif 779 if (isipv6) { 780 struct in6_addr laddr6; 781 /* 782 * Inherit socket options from the listening socket. 783 * Note that in6p_inputopts are not (and should not be) 784 * copied, since it stores previously received options and is 785 * used to detect if each new option is different than the 786 * previous one and hence should be passed to a user. 787 * If we copied in6p_inputopts, a user would not be able to 788 * receive options just after calling the accept system call. 789 */ 790 inp->inp_flags |= linp->inp_flags & INP_CONTROLOPTS; 791 if (linp->in6p_outputopts) 792 inp->in6p_outputopts = 793 ip6_copypktopts(linp->in6p_outputopts, M_INTWAIT); 794 inp->in6p_route = sc->sc_route6; 795 sc->sc_route6.ro_rt = NULL; 796 797 laddr6 = inp->in6p_laddr; 798 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 799 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 800 if (in6_pcbconnect(inp, faddr, &thread0)) { 801 inp->in6p_laddr = laddr6; 802 goto abort; 803 } 804 } else { 805 struct in_addr laddr; 806 807 inp->inp_options = ip_srcroute(m); 808 if (inp->inp_options == NULL) { 809 inp->inp_options = sc->sc_ipopts; 810 sc->sc_ipopts = NULL; 811 } 812 inp->inp_route = sc->sc_route; 813 sc->sc_route.ro_rt = NULL; 814 815 laddr = inp->inp_laddr; 816 if (inp->inp_laddr.s_addr == INADDR_ANY) 817 inp->inp_laddr = sc->sc_inc.inc_laddr; 818 if (in_pcbconnect(inp, faddr, &thread0)) { 819 inp->inp_laddr = laddr; 820 goto abort; 821 } 822 823 inp->inp_flags |= INP_HASH; 824 inp->inp_hashval = m->m_pkthdr.hash; 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 if (isipv6) { 832 port = tcp6_addrport(); 833 } else { 834 port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport, 835 inp->inp_laddr.s_addr, inp->inp_lport); 836 } 837 KASSERT(port == &curthread->td_msgport, 838 ("TCP PORT MISMATCH %p vs %p\n", port, &curthread->td_msgport)); 839 840 tp = intotcpcb(inp); 841 TCP_STATE_CHANGE(tp, TCPS_SYN_RECEIVED); 842 tp->iss = sc->sc_iss; 843 tp->irs = sc->sc_irs; 844 tcp_rcvseqinit(tp); 845 tcp_sendseqinit(tp); 846 tp->snd_wnd = sc->sc_sndwnd; 847 tp->snd_wl1 = sc->sc_irs; 848 tp->rcv_up = sc->sc_irs + 1; 849 tp->rcv_wnd = sc->sc_wnd; 850 tp->rcv_adv += tp->rcv_wnd; 851 852 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH | TF_NODELAY); 853 if (sc->sc_flags & SCF_NOOPT) 854 tp->t_flags |= TF_NOOPT; 855 if (sc->sc_flags & SCF_WINSCALE) { 856 tp->t_flags |= TF_REQ_SCALE | TF_RCVD_SCALE; 857 tp->snd_scale = sc->sc_requested_s_scale; 858 tp->request_r_scale = sc->sc_request_r_scale; 859 } 860 if (sc->sc_flags & SCF_TIMESTAMP) { 861 tp->t_flags |= TF_REQ_TSTMP | TF_RCVD_TSTMP; 862 tp->ts_recent = sc->sc_tsrecent; 863 tp->ts_recent_age = ticks; 864 } 865 if (sc->sc_flags & SCF_SACK_PERMITTED) 866 tp->t_flags |= TF_SACK_PERMITTED; 867 868 #ifdef TCP_SIGNATURE 869 if (sc->sc_flags & SCF_SIGNATURE) 870 tp->t_flags |= TF_SIGNATURE; 871 #endif /* TCP_SIGNATURE */ 872 873 tp->t_rxtsyn = sc->sc_rxtused; 874 tcp_mss(tp, sc->sc_peer_mss); 875 876 /* 877 * Inherit some properties from the listen socket 878 */ 879 tp->t_keepinit = ltp->t_keepinit; 880 tp->t_keepidle = ltp->t_keepidle; 881 tp->t_keepintvl = ltp->t_keepintvl; 882 tp->t_keepcnt = ltp->t_keepcnt; 883 tp->t_maxidle = ltp->t_maxidle; 884 885 tcp_create_timermsg(tp, port); 886 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep); 887 888 tcpstat.tcps_accepts++; 889 return (so); 890 891 abort: 892 if (so != NULL) 893 soabort_direct(so); 894 return (NULL); 895 } 896 897 /* 898 * This function gets called when we receive an ACK for a 899 * socket in the LISTEN state. We look up the connection 900 * in the syncache, and if its there, we pull it out of 901 * the cache and turn it into a full-blown connection in 902 * the SYN-RECEIVED state. 903 */ 904 int 905 syncache_expand(struct in_conninfo *inc, struct tcphdr *th, struct socket **sop, 906 struct mbuf *m) 907 { 908 struct syncache *sc; 909 struct syncache_head *sch; 910 struct socket *so; 911 912 sc = syncache_lookup(inc, &sch); 913 if (sc == NULL) { 914 /* 915 * There is no syncache entry, so see if this ACK is 916 * a returning syncookie. To do this, first: 917 * A. See if this socket has had a syncache entry dropped in 918 * the past. We don't want to accept a bogus syncookie 919 * if we've never received a SYN. 920 * B. check that the syncookie is valid. If it is, then 921 * cobble up a fake syncache entry, and return. 922 */ 923 if (!tcp_syncookies) 924 return (0); 925 sc = syncookie_lookup(inc, th, *sop); 926 if (sc == NULL) 927 return (0); 928 sch = NULL; 929 tcpstat.tcps_sc_recvcookie++; 930 } 931 932 /* 933 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 934 */ 935 if (th->th_ack != sc->sc_iss + 1) 936 return (0); 937 938 so = syncache_socket(sc, *sop, m); 939 if (so == NULL) { 940 #if 0 941 resetandabort: 942 /* XXXjlemon check this - is this correct? */ 943 tcp_respond(NULL, m, m, th, 944 th->th_seq + tlen, (tcp_seq)0, TH_RST | TH_ACK); 945 #endif 946 m_freem(m); /* XXX only needed for above */ 947 tcpstat.tcps_sc_aborted++; 948 } else { 949 tcpstat.tcps_sc_completed++; 950 } 951 if (sch == NULL) 952 syncache_free(sc); 953 else 954 syncache_drop(sc, sch); 955 *sop = so; 956 return (1); 957 } 958 959 /* 960 * Given a LISTEN socket and an inbound SYN request, add 961 * this to the syn cache, and send back a segment: 962 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 963 * to the source. 964 * 965 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 966 * Doing so would require that we hold onto the data and deliver it 967 * to the application. However, if we are the target of a SYN-flood 968 * DoS attack, an attacker could send data which would eventually 969 * consume all available buffer space if it were ACKed. By not ACKing 970 * the data, we avoid this DoS scenario. 971 */ 972 int 973 syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 974 struct socket *so, struct mbuf *m) 975 { 976 struct tcp_syncache_percpu *syncache_percpu; 977 struct tcpcb *tp; 978 struct syncache *sc = NULL; 979 struct syncache_head *sch; 980 struct mbuf *ipopts = NULL; 981 int win; 982 983 KASSERT(m->m_flags & M_HASH, ("mbuf has no hash")); 984 985 syncache_percpu = &tcp_syncache_percpu[mycpu->gd_cpuid]; 986 tp = sototcpcb(so); 987 988 /* 989 * Remember the IP options, if any. 990 */ 991 #ifdef INET6 992 if (!inc->inc_isipv6) 993 #endif 994 ipopts = ip_srcroute(m); 995 996 /* 997 * See if we already have an entry for this connection. 998 * If we do, resend the SYN,ACK, and reset the retransmit timer. 999 * 1000 * XXX 1001 * The syncache should be re-initialized with the contents 1002 * of the new SYN which may have different options. 1003 */ 1004 sc = syncache_lookup(inc, &sch); 1005 if (sc != NULL) { 1006 KASSERT(sc->sc_flags & SCF_HASH, ("syncache has no hash")); 1007 KASSERT(sc->sc_hashval == m->m_pkthdr.hash, 1008 ("syncache/mbuf hash mismatches")); 1009 1010 tcpstat.tcps_sc_dupsyn++; 1011 if (ipopts) { 1012 /* 1013 * If we were remembering a previous source route, 1014 * forget it and use the new one we've been given. 1015 */ 1016 if (sc->sc_ipopts) 1017 m_free(sc->sc_ipopts); 1018 sc->sc_ipopts = ipopts; 1019 } 1020 /* 1021 * Update timestamp if present. 1022 */ 1023 if (sc->sc_flags & SCF_TIMESTAMP) 1024 sc->sc_tsrecent = to->to_tsval; 1025 1026 /* Just update the TOF_SACK_PERMITTED for now. */ 1027 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1028 sc->sc_flags |= SCF_SACK_PERMITTED; 1029 else 1030 sc->sc_flags &= ~SCF_SACK_PERMITTED; 1031 1032 /* Update initial send window */ 1033 sc->sc_sndwnd = th->th_win; 1034 1035 /* 1036 * PCB may have changed, pick up new values. 1037 */ 1038 sc->sc_tp = tp; 1039 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1040 if (syncache_respond(sc, m) == 0) { 1041 TAILQ_REMOVE(&syncache_percpu->timerq[sc->sc_rxtslot], 1042 sc, sc_timerq); 1043 syncache_timeout(syncache_percpu, sc, sc->sc_rxtslot); 1044 tcpstat.tcps_sndacks++; 1045 tcpstat.tcps_sndtotal++; 1046 } 1047 return (1); 1048 } 1049 1050 /* 1051 * Fill in the syncache values. 1052 */ 1053 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1054 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 1055 sc->sc_ipopts = ipopts; 1056 sc->sc_inc.inc_fport = inc->inc_fport; 1057 sc->sc_inc.inc_lport = inc->inc_lport; 1058 sc->sc_tp = tp; 1059 #ifdef INET6 1060 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1061 if (inc->inc_isipv6) { 1062 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1063 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1064 sc->sc_route6.ro_rt = NULL; 1065 } else 1066 #endif 1067 { 1068 sc->sc_inc.inc_faddr = inc->inc_faddr; 1069 sc->sc_inc.inc_laddr = inc->inc_laddr; 1070 sc->sc_route.ro_rt = NULL; 1071 } 1072 sc->sc_irs = th->th_seq; 1073 sc->sc_flags = SCF_HASH; 1074 sc->sc_hashval = m->m_pkthdr.hash; 1075 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 1076 if (tcp_syncookies) 1077 sc->sc_iss = syncookie_generate(sc); 1078 else 1079 sc->sc_iss = karc4random(); 1080 1081 /* Initial receive window: clip ssb_space to [0 .. TCP_MAXWIN] */ 1082 win = ssb_space(&so->so_rcv); 1083 win = imax(win, 0); 1084 win = imin(win, TCP_MAXWIN); 1085 sc->sc_wnd = win; 1086 1087 if (tcp_do_rfc1323) { 1088 /* 1089 * A timestamp received in a SYN makes 1090 * it ok to send timestamp requests and replies. 1091 */ 1092 if (to->to_flags & TOF_TS) { 1093 sc->sc_tsrecent = to->to_tsval; 1094 sc->sc_flags |= SCF_TIMESTAMP; 1095 } 1096 if (to->to_flags & TOF_SCALE) { 1097 int wscale = TCP_MIN_WINSHIFT; 1098 1099 /* Compute proper scaling value from buffer space */ 1100 while (wscale < TCP_MAX_WINSHIFT && 1101 (TCP_MAXWIN << wscale) < so->so_rcv.ssb_hiwat) { 1102 wscale++; 1103 } 1104 sc->sc_request_r_scale = wscale; 1105 sc->sc_requested_s_scale = to->to_requested_s_scale; 1106 sc->sc_flags |= SCF_WINSCALE; 1107 } 1108 } 1109 if (tcp_do_sack && (to->to_flags & TOF_SACK_PERMITTED)) 1110 sc->sc_flags |= SCF_SACK_PERMITTED; 1111 if (tp->t_flags & TF_NOOPT) 1112 sc->sc_flags = SCF_NOOPT; 1113 #ifdef TCP_SIGNATURE 1114 /* 1115 * If listening socket requested TCP digests, and received SYN 1116 * contains the option, flag this in the syncache so that 1117 * syncache_respond() will do the right thing with the SYN+ACK. 1118 * XXX Currently we always record the option by default and will 1119 * attempt to use it in syncache_respond(). 1120 */ 1121 if (to->to_flags & TOF_SIGNATURE) 1122 sc->sc_flags = SCF_SIGNATURE; 1123 #endif /* TCP_SIGNATURE */ 1124 sc->sc_sndwnd = th->th_win; 1125 1126 if (syncache_respond(sc, m) == 0) { 1127 syncache_insert(sc, sch); 1128 tcpstat.tcps_sndacks++; 1129 tcpstat.tcps_sndtotal++; 1130 } else { 1131 syncache_free(sc); 1132 tcpstat.tcps_sc_dropped++; 1133 } 1134 return (1); 1135 } 1136 1137 static int 1138 syncache_respond(struct syncache *sc, struct mbuf *m) 1139 { 1140 u_int8_t *optp; 1141 int optlen, error; 1142 u_int16_t tlen, hlen, mssopt; 1143 struct ip *ip = NULL; 1144 struct rtentry *rt; 1145 struct tcphdr *th; 1146 struct ip6_hdr *ip6 = NULL; 1147 #ifdef INET6 1148 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1149 #else 1150 const boolean_t isipv6 = FALSE; 1151 #endif 1152 1153 if (isipv6) { 1154 rt = tcp_rtlookup6(&sc->sc_inc); 1155 if (rt != NULL) 1156 mssopt = rt->rt_ifp->if_mtu - 1157 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1158 else 1159 mssopt = tcp_v6mssdflt; 1160 hlen = sizeof(struct ip6_hdr); 1161 } else { 1162 rt = tcp_rtlookup(&sc->sc_inc); 1163 if (rt != NULL) 1164 mssopt = rt->rt_ifp->if_mtu - 1165 (sizeof(struct ip) + sizeof(struct tcphdr)); 1166 else 1167 mssopt = tcp_mssdflt; 1168 hlen = sizeof(struct ip); 1169 } 1170 1171 /* Compute the size of the TCP options. */ 1172 if (sc->sc_flags & SCF_NOOPT) { 1173 optlen = 0; 1174 } else { 1175 optlen = TCPOLEN_MAXSEG + 1176 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1177 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1178 ((sc->sc_flags & SCF_SACK_PERMITTED) ? 1179 TCPOLEN_SACK_PERMITTED_ALIGNED : 0); 1180 #ifdef TCP_SIGNATURE 1181 optlen += ((sc->sc_flags & SCF_SIGNATURE) ? 1182 (TCPOLEN_SIGNATURE + 2) : 0); 1183 #endif /* TCP_SIGNATURE */ 1184 } 1185 tlen = hlen + sizeof(struct tcphdr) + optlen; 1186 1187 /* 1188 * XXX 1189 * assume that the entire packet will fit in a header mbuf 1190 */ 1191 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1192 1193 /* 1194 * XXX shouldn't this reuse the mbuf if possible ? 1195 * Create the IP+TCP header from scratch. 1196 */ 1197 if (m) 1198 m_freem(m); 1199 1200 m = m_gethdr(M_NOWAIT, MT_HEADER); 1201 if (m == NULL) 1202 return (ENOBUFS); 1203 m->m_data += max_linkhdr; 1204 m->m_len = tlen; 1205 m->m_pkthdr.len = tlen; 1206 m->m_pkthdr.rcvif = NULL; 1207 if (tcp_prio_synack) 1208 m->m_flags |= M_PRIO; 1209 1210 if (isipv6) { 1211 ip6 = mtod(m, struct ip6_hdr *); 1212 ip6->ip6_vfc = IPV6_VERSION; 1213 ip6->ip6_nxt = IPPROTO_TCP; 1214 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1215 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1216 ip6->ip6_plen = htons(tlen - hlen); 1217 /* ip6_hlim is set after checksum */ 1218 /* ip6_flow = ??? */ 1219 1220 th = (struct tcphdr *)(ip6 + 1); 1221 } else { 1222 ip = mtod(m, struct ip *); 1223 ip->ip_v = IPVERSION; 1224 ip->ip_hl = sizeof(struct ip) >> 2; 1225 ip->ip_len = tlen; 1226 ip->ip_id = 0; 1227 ip->ip_off = 0; 1228 ip->ip_sum = 0; 1229 ip->ip_p = IPPROTO_TCP; 1230 ip->ip_src = sc->sc_inc.inc_laddr; 1231 ip->ip_dst = sc->sc_inc.inc_faddr; 1232 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1233 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1234 1235 /* 1236 * See if we should do MTU discovery. Route lookups are 1237 * expensive, so we will only unset the DF bit if: 1238 * 1239 * 1) path_mtu_discovery is disabled 1240 * 2) the SCF_UNREACH flag has been set 1241 */ 1242 if (path_mtu_discovery 1243 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1244 ip->ip_off |= IP_DF; 1245 } 1246 1247 th = (struct tcphdr *)(ip + 1); 1248 } 1249 th->th_sport = sc->sc_inc.inc_lport; 1250 th->th_dport = sc->sc_inc.inc_fport; 1251 1252 th->th_seq = htonl(sc->sc_iss); 1253 th->th_ack = htonl(sc->sc_irs + 1); 1254 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1255 th->th_x2 = 0; 1256 th->th_flags = TH_SYN | TH_ACK; 1257 th->th_win = htons(sc->sc_wnd); 1258 th->th_urp = 0; 1259 1260 /* Tack on the TCP options. */ 1261 if (optlen == 0) 1262 goto no_options; 1263 optp = (u_int8_t *)(th + 1); 1264 *optp++ = TCPOPT_MAXSEG; 1265 *optp++ = TCPOLEN_MAXSEG; 1266 *optp++ = (mssopt >> 8) & 0xff; 1267 *optp++ = mssopt & 0xff; 1268 1269 if (sc->sc_flags & SCF_WINSCALE) { 1270 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1271 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1272 sc->sc_request_r_scale); 1273 optp += 4; 1274 } 1275 1276 if (sc->sc_flags & SCF_TIMESTAMP) { 1277 u_int32_t *lp = (u_int32_t *)(optp); 1278 1279 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1280 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1281 *lp++ = htonl(ticks); 1282 *lp = htonl(sc->sc_tsrecent); 1283 optp += TCPOLEN_TSTAMP_APPA; 1284 } 1285 1286 #ifdef TCP_SIGNATURE 1287 /* 1288 * Handle TCP-MD5 passive opener response. 1289 */ 1290 if (sc->sc_flags & SCF_SIGNATURE) { 1291 u_int8_t *bp = optp; 1292 int i; 1293 1294 *bp++ = TCPOPT_SIGNATURE; 1295 *bp++ = TCPOLEN_SIGNATURE; 1296 for (i = 0; i < TCP_SIGLEN; i++) 1297 *bp++ = 0; 1298 tcpsignature_compute(m, 0, optlen, 1299 optp + 2, IPSEC_DIR_OUTBOUND); 1300 *bp++ = TCPOPT_NOP; 1301 *bp++ = TCPOPT_EOL; 1302 optp += TCPOLEN_SIGNATURE + 2; 1303 } 1304 #endif /* TCP_SIGNATURE */ 1305 1306 if (sc->sc_flags & SCF_SACK_PERMITTED) { 1307 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMITTED_ALIGNED); 1308 optp += TCPOLEN_SACK_PERMITTED_ALIGNED; 1309 } 1310 1311 no_options: 1312 if (isipv6) { 1313 struct route_in6 *ro6 = &sc->sc_route6; 1314 1315 th->th_sum = 0; 1316 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1317 ip6->ip6_hlim = in6_selecthlim(NULL, 1318 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1319 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1320 sc->sc_tp->t_inpcb); 1321 } else { 1322 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1323 htons(tlen - hlen + IPPROTO_TCP)); 1324 m->m_pkthdr.csum_flags = CSUM_TCP; 1325 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1326 m->m_pkthdr.csum_thlen = sizeof(struct tcphdr) + optlen; 1327 KASSERT(sc->sc_flags & SCF_HASH, ("syncache has no hash")); 1328 m_sethash(m, sc->sc_hashval); 1329 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 1330 IP_DEBUGROUTE, NULL, sc->sc_tp->t_inpcb); 1331 } 1332 return (error); 1333 } 1334 1335 /* 1336 * cookie layers: 1337 * 1338 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1339 * | peer iss | 1340 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1341 * | 0 |(A)| | 1342 * (A): peer mss index 1343 */ 1344 1345 /* 1346 * The values below are chosen to minimize the size of the tcp_secret 1347 * table, as well as providing roughly a 16 second lifetime for the cookie. 1348 */ 1349 1350 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1351 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1352 1353 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1354 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1355 #define SYNCOOKIE_TIMEOUT \ 1356 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1357 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1358 1359 static struct { 1360 u_int32_t ts_secbits[4]; 1361 u_int ts_expire; 1362 } tcp_secret[SYNCOOKIE_NSECRETS]; 1363 1364 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1365 1366 static MD5_CTX syn_ctx; 1367 1368 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1369 1370 struct md5_add { 1371 u_int32_t laddr, faddr; 1372 u_int32_t secbits[4]; 1373 u_int16_t lport, fport; 1374 }; 1375 1376 #ifdef CTASSERT 1377 CTASSERT(sizeof(struct md5_add) == 28); 1378 #endif 1379 1380 /* 1381 * Consider the problem of a recreated (and retransmitted) cookie. If the 1382 * original SYN was accepted, the connection is established. The second 1383 * SYN is inflight, and if it arrives with an ISN that falls within the 1384 * receive window, the connection is killed. 1385 * 1386 * However, since cookies have other problems, this may not be worth 1387 * worrying about. 1388 */ 1389 1390 static u_int32_t 1391 syncookie_generate(struct syncache *sc) 1392 { 1393 u_int32_t md5_buffer[4]; 1394 u_int32_t data; 1395 int idx, i; 1396 struct md5_add add; 1397 #ifdef INET6 1398 const boolean_t isipv6 = sc->sc_inc.inc_isipv6; 1399 #else 1400 const boolean_t isipv6 = FALSE; 1401 #endif 1402 1403 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1404 if (tcp_secret[idx].ts_expire < ticks) { 1405 for (i = 0; i < 4; i++) 1406 tcp_secret[idx].ts_secbits[i] = karc4random(); 1407 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1408 } 1409 for (data = NELEM(tcp_msstab) - 1; data > 0; data--) 1410 if (tcp_msstab[data] <= sc->sc_peer_mss) 1411 break; 1412 data = (data << SYNCOOKIE_WNDBITS) | idx; 1413 data ^= sc->sc_irs; /* peer's iss */ 1414 MD5Init(&syn_ctx); 1415 if (isipv6) { 1416 MD5Add(sc->sc_inc.inc6_laddr); 1417 MD5Add(sc->sc_inc.inc6_faddr); 1418 add.laddr = 0; 1419 add.faddr = 0; 1420 } else { 1421 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1422 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1423 } 1424 add.lport = sc->sc_inc.inc_lport; 1425 add.fport = sc->sc_inc.inc_fport; 1426 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1427 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1428 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1429 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1430 MD5Add(add); 1431 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1432 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1433 return (data); 1434 } 1435 1436 static struct syncache * 1437 syncookie_lookup(struct in_conninfo *inc, struct tcphdr *th, struct socket *so) 1438 { 1439 u_int32_t md5_buffer[4]; 1440 struct syncache *sc; 1441 u_int32_t data; 1442 int wnd, idx; 1443 struct md5_add add; 1444 1445 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1446 idx = data & SYNCOOKIE_WNDMASK; 1447 if (tcp_secret[idx].ts_expire < ticks || 1448 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1449 return (NULL); 1450 MD5Init(&syn_ctx); 1451 #ifdef INET6 1452 if (inc->inc_isipv6) { 1453 MD5Add(inc->inc6_laddr); 1454 MD5Add(inc->inc6_faddr); 1455 add.laddr = 0; 1456 add.faddr = 0; 1457 } else 1458 #endif 1459 { 1460 add.laddr = inc->inc_laddr.s_addr; 1461 add.faddr = inc->inc_faddr.s_addr; 1462 } 1463 add.lport = inc->inc_lport; 1464 add.fport = inc->inc_fport; 1465 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1466 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1467 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1468 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1469 MD5Add(add); 1470 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1471 data ^= md5_buffer[0]; 1472 if (data & ~SYNCOOKIE_DATAMASK) 1473 return (NULL); 1474 data = data >> SYNCOOKIE_WNDBITS; 1475 1476 /* 1477 * Fill in the syncache values. 1478 * XXX duplicate code from syncache_add 1479 */ 1480 sc = kmalloc(sizeof(struct syncache), M_SYNCACHE, M_WAITOK|M_ZERO); 1481 sc->sc_ipopts = NULL; 1482 sc->sc_inc.inc_fport = inc->inc_fport; 1483 sc->sc_inc.inc_lport = inc->inc_lport; 1484 #ifdef INET6 1485 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1486 if (inc->inc_isipv6) { 1487 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1488 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1489 sc->sc_route6.ro_rt = NULL; 1490 } else 1491 #endif 1492 { 1493 sc->sc_inc.inc_faddr = inc->inc_faddr; 1494 sc->sc_inc.inc_laddr = inc->inc_laddr; 1495 sc->sc_route.ro_rt = NULL; 1496 } 1497 sc->sc_irs = th->th_seq - 1; 1498 sc->sc_iss = th->th_ack - 1; 1499 wnd = ssb_space(&so->so_rcv); 1500 wnd = imax(wnd, 0); 1501 wnd = imin(wnd, TCP_MAXWIN); 1502 sc->sc_wnd = wnd; 1503 sc->sc_flags = 0; 1504 sc->sc_rxtslot = 0; 1505 sc->sc_peer_mss = tcp_msstab[data]; 1506 return (sc); 1507 } 1508