1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 30 * $FreeBSD$ 31 */ 32 33 #include "opt_compat.h" 34 #include "opt_inet.h" 35 #include "opt_inet6.h" 36 #include "opt_ipsec.h" 37 #include "opt_mac.h" 38 #include "opt_tcpdebug.h" 39 #include "opt_tcp_sack.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/callout.h> 44 #include <sys/kernel.h> 45 #include <sys/sysctl.h> 46 #include <sys/mac.h> 47 #include <sys/malloc.h> 48 #include <sys/mbuf.h> 49 #ifdef INET6 50 #include <sys/domain.h> 51 #endif 52 #include <sys/proc.h> 53 #include <sys/socket.h> 54 #include <sys/socketvar.h> 55 #include <sys/protosw.h> 56 #include <sys/random.h> 57 58 #include <vm/uma.h> 59 60 #include <net/route.h> 61 #include <net/if.h> 62 63 #include <netinet/in.h> 64 #include <netinet/in_systm.h> 65 #include <netinet/ip.h> 66 #ifdef INET6 67 #include <netinet/ip6.h> 68 #endif 69 #include <netinet/in_pcb.h> 70 #ifdef INET6 71 #include <netinet6/in6_pcb.h> 72 #endif 73 #include <netinet/in_var.h> 74 #include <netinet/ip_var.h> 75 #ifdef INET6 76 #include <netinet6/ip6_var.h> 77 #include <netinet6/scope6_var.h> 78 #include <netinet6/nd6.h> 79 #endif 80 #include <netinet/ip_icmp.h> 81 #include <netinet/tcp.h> 82 #include <netinet/tcp_fsm.h> 83 #include <netinet/tcp_seq.h> 84 #include <netinet/tcp_timer.h> 85 #include <netinet/tcp_var.h> 86 #ifdef INET6 87 #include <netinet6/tcp6_var.h> 88 #endif 89 #include <netinet/tcpip.h> 90 #ifdef TCPDEBUG 91 #include <netinet/tcp_debug.h> 92 #endif 93 #include <netinet6/ip6protosw.h> 94 95 #ifdef IPSEC 96 #include <netinet6/ipsec.h> 97 #ifdef INET6 98 #include <netinet6/ipsec6.h> 99 #endif 100 #include <netkey/key.h> 101 #endif /*IPSEC*/ 102 103 #ifdef FAST_IPSEC 104 #include <netipsec/ipsec.h> 105 #include <netipsec/xform.h> 106 #ifdef INET6 107 #include <netipsec/ipsec6.h> 108 #endif 109 #include <netipsec/key.h> 110 #define IPSEC 111 #endif /*FAST_IPSEC*/ 112 113 #include <machine/in_cksum.h> 114 #include <sys/md5.h> 115 116 int tcp_mssdflt = TCP_MSS; 117 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 118 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); 119 120 #ifdef INET6 121 int tcp_v6mssdflt = TCP6_MSS; 122 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 123 CTLFLAG_RW, &tcp_v6mssdflt , 0, 124 "Default TCP Maximum Segment Size for IPv6"); 125 #endif 126 127 /* 128 * Minimum MSS we accept and use. This prevents DoS attacks where 129 * we are forced to a ridiculous low MSS like 20 and send hundreds 130 * of packets instead of one. The effect scales with the available 131 * bandwidth and quickly saturates the CPU and network interface 132 * with packet generation and sending. Set to zero to disable MINMSS 133 * checking. This setting prevents us from sending too small packets. 134 */ 135 int tcp_minmss = TCP_MINMSS; 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 137 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 138 /* 139 * Number of TCP segments per second we accept from remote host 140 * before we start to calculate average segment size. If average 141 * segment size drops below the minimum TCP MSS we assume a DoS 142 * attack and reset+drop the connection. Care has to be taken not to 143 * set this value too small to not kill interactive type connections 144 * (telnet, SSH) which send many small packets. 145 */ 146 int tcp_minmssoverload = TCP_MINMSSOVERLOAD; 147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW, 148 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to" 149 "be under the MINMSS Size"); 150 151 #if 0 152 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 153 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 154 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 155 #endif 156 157 int tcp_do_rfc1323 = 1; 158 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 159 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 160 161 static int tcp_tcbhashsize = 0; 162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 163 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 164 165 static int do_tcpdrain = 1; 166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 167 "Enable tcp_drain routine for extra help when low on mbufs"); 168 169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 170 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 171 172 static int icmp_may_rst = 1; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 174 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 175 176 static int tcp_isn_reseed_interval = 0; 177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 178 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 179 180 static uma_zone_t tcptw_zone; 181 static int maxtcptw; 182 static int 183 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 184 { 185 int error, new; 186 187 if (maxtcptw == 0) 188 new = maxsockets / 5; 189 else 190 new = maxtcptw; 191 error = sysctl_handle_int(oidp, &new, sizeof(int), req); 192 if (error == 0 && req->newptr) { 193 if (new > maxtcptw) { 194 maxtcptw = new; 195 uma_zone_set_max(tcptw_zone, maxtcptw); 196 } else 197 error = EINVAL; 198 } 199 return (error); 200 } 201 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 202 &maxtcptw, 0, sysctl_maxtcptw, "IU", 203 "Maximum number of compressed TCP TIME_WAIT entries"); 204 205 static int nolocaltimewait = 0; 206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW, 207 &nolocaltimewait, 0, "Do not create compressed TCP TIME_WAIT entries" 208 "for local connections"); 209 210 /* 211 * TCP bandwidth limiting sysctls. Note that the default lower bound of 212 * 1024 exists only for debugging. A good production default would be 213 * something like 6100. 214 */ 215 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 216 "TCP inflight data limiting"); 217 218 static int tcp_inflight_enable = 1; 219 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 220 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 221 222 static int tcp_inflight_debug = 0; 223 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 224 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 225 226 static int tcp_inflight_rttthresh; 227 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW, 228 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I", 229 "RTT threshold below which inflight will deactivate itself"); 230 231 static int tcp_inflight_min = 6144; 232 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 233 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 234 235 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 236 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 237 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 238 239 static int tcp_inflight_stab = 20; 240 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 241 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 242 243 uma_zone_t sack_hole_zone; 244 245 static struct inpcb *tcp_notify(struct inpcb *, int); 246 static void tcp_isn_tick(void *); 247 248 /* 249 * Target size of TCP PCB hash tables. Must be a power of two. 250 * 251 * Note that this can be overridden by the kernel environment 252 * variable net.inet.tcp.tcbhashsize 253 */ 254 #ifndef TCBHASHSIZE 255 #define TCBHASHSIZE 512 256 #endif 257 258 /* 259 * XXX 260 * Callouts should be moved into struct tcp directly. They are currently 261 * separate because the tcpcb structure is exported to userland for sysctl 262 * parsing purposes, which do not know about callouts. 263 */ 264 struct tcpcb_mem { 265 struct tcpcb tcb; 266 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 267 struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 268 }; 269 270 static uma_zone_t tcpcb_zone; 271 struct callout isn_callout; 272 static struct mtx isn_mtx; 273 274 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 275 #define ISN_LOCK() mtx_lock(&isn_mtx) 276 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 277 278 /* 279 * TCP initialization. 280 */ 281 static void 282 tcp_zone_change(void *tag) 283 { 284 285 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 286 uma_zone_set_max(tcpcb_zone, maxsockets); 287 if (maxtcptw == 0) 288 uma_zone_set_max(tcptw_zone, maxsockets / 5); 289 } 290 291 static int 292 tcp_inpcb_init(void *mem, int size, int flags) 293 { 294 struct inpcb *inp = (struct inpcb *) mem; 295 INP_LOCK_INIT(inp, "inp", "tcpinp"); 296 return (0); 297 } 298 299 void 300 tcp_init(void) 301 { 302 int hashsize = TCBHASHSIZE; 303 304 tcp_delacktime = TCPTV_DELACK; 305 tcp_keepinit = TCPTV_KEEP_INIT; 306 tcp_keepidle = TCPTV_KEEP_IDLE; 307 tcp_keepintvl = TCPTV_KEEPINTVL; 308 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 309 tcp_msl = TCPTV_MSL; 310 tcp_rexmit_min = TCPTV_MIN; 311 tcp_rexmit_slop = TCPTV_CPU_VAR; 312 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH; 313 314 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 315 LIST_INIT(&tcb); 316 tcbinfo.listhead = &tcb; 317 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 318 if (!powerof2(hashsize)) { 319 printf("WARNING: TCB hash size not a power of 2\n"); 320 hashsize = 512; /* safe default */ 321 } 322 tcp_tcbhashsize = hashsize; 323 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 324 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 325 &tcbinfo.porthashmask); 326 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 327 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 328 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 329 #ifdef INET6 330 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 331 #else /* INET6 */ 332 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 333 #endif /* INET6 */ 334 if (max_protohdr < TCP_MINPROTOHDR) 335 max_protohdr = TCP_MINPROTOHDR; 336 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 337 panic("tcp_init"); 338 #undef TCP_MINPROTOHDR 339 /* 340 * These have to be type stable for the benefit of the timers. 341 */ 342 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 343 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 344 uma_zone_set_max(tcpcb_zone, maxsockets); 345 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 346 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 347 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 348 if (maxtcptw == 0) 349 uma_zone_set_max(tcptw_zone, maxsockets / 5); 350 else 351 uma_zone_set_max(tcptw_zone, maxtcptw); 352 tcp_timer_init(); 353 syncache_init(); 354 tcp_hc_init(); 355 tcp_reass_init(); 356 ISN_LOCK_INIT(); 357 callout_init(&isn_callout, CALLOUT_MPSAFE); 358 tcp_isn_tick(NULL); 359 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 360 SHUTDOWN_PRI_DEFAULT); 361 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 362 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 363 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 364 EVENTHANDLER_PRI_ANY); 365 } 366 367 void 368 tcp_fini(void *xtp) 369 { 370 371 callout_stop(&isn_callout); 372 } 373 374 /* 375 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 376 * tcp_template used to store this data in mbufs, but we now recopy it out 377 * of the tcpcb each time to conserve mbufs. 378 */ 379 void 380 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 381 { 382 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 383 384 INP_LOCK_ASSERT(inp); 385 386 #ifdef INET6 387 if ((inp->inp_vflag & INP_IPV6) != 0) { 388 struct ip6_hdr *ip6; 389 390 ip6 = (struct ip6_hdr *)ip_ptr; 391 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 392 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 393 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 394 (IPV6_VERSION & IPV6_VERSION_MASK); 395 ip6->ip6_nxt = IPPROTO_TCP; 396 ip6->ip6_plen = sizeof(struct tcphdr); 397 ip6->ip6_src = inp->in6p_laddr; 398 ip6->ip6_dst = inp->in6p_faddr; 399 } else 400 #endif 401 { 402 struct ip *ip; 403 404 ip = (struct ip *)ip_ptr; 405 ip->ip_v = IPVERSION; 406 ip->ip_hl = 5; 407 ip->ip_tos = inp->inp_ip_tos; 408 ip->ip_len = 0; 409 ip->ip_id = 0; 410 ip->ip_off = 0; 411 ip->ip_ttl = inp->inp_ip_ttl; 412 ip->ip_sum = 0; 413 ip->ip_p = IPPROTO_TCP; 414 ip->ip_src = inp->inp_laddr; 415 ip->ip_dst = inp->inp_faddr; 416 } 417 th->th_sport = inp->inp_lport; 418 th->th_dport = inp->inp_fport; 419 th->th_seq = 0; 420 th->th_ack = 0; 421 th->th_x2 = 0; 422 th->th_off = 5; 423 th->th_flags = 0; 424 th->th_win = 0; 425 th->th_urp = 0; 426 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 427 } 428 429 /* 430 * Create template to be used to send tcp packets on a connection. 431 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 432 * use for this function is in keepalives, which use tcp_respond. 433 */ 434 struct tcptemp * 435 tcpip_maketemplate(struct inpcb *inp) 436 { 437 struct mbuf *m; 438 struct tcptemp *n; 439 440 m = m_get(M_DONTWAIT, MT_DATA); 441 if (m == NULL) 442 return (0); 443 m->m_len = sizeof(struct tcptemp); 444 n = mtod(m, struct tcptemp *); 445 446 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 447 return (n); 448 } 449 450 /* 451 * Send a single message to the TCP at address specified by 452 * the given TCP/IP header. If m == NULL, then we make a copy 453 * of the tcpiphdr at ti and send directly to the addressed host. 454 * This is used to force keep alive messages out using the TCP 455 * template for a connection. If flags are given then we send 456 * a message back to the TCP which originated the * segment ti, 457 * and discard the mbuf containing it and any other attached mbufs. 458 * 459 * In any case the ack and sequence number of the transmitted 460 * segment are as specified by the parameters. 461 * 462 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 463 */ 464 void 465 tcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th, 466 register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) 467 { 468 register int tlen; 469 int win = 0; 470 struct ip *ip; 471 struct tcphdr *nth; 472 #ifdef INET6 473 struct ip6_hdr *ip6; 474 int isipv6; 475 #endif /* INET6 */ 476 int ipflags = 0; 477 struct inpcb *inp; 478 479 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 480 481 #ifdef INET6 482 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 483 ip6 = ipgen; 484 #endif /* INET6 */ 485 ip = ipgen; 486 487 if (tp != NULL) { 488 inp = tp->t_inpcb; 489 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 490 INP_INFO_WLOCK_ASSERT(&tcbinfo); 491 INP_LOCK_ASSERT(inp); 492 } else 493 inp = NULL; 494 495 if (tp != NULL) { 496 if (!(flags & TH_RST)) { 497 win = sbspace(&inp->inp_socket->so_rcv); 498 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 499 win = (long)TCP_MAXWIN << tp->rcv_scale; 500 } 501 } 502 if (m == NULL) { 503 m = m_gethdr(M_DONTWAIT, MT_DATA); 504 if (m == NULL) 505 return; 506 tlen = 0; 507 m->m_data += max_linkhdr; 508 #ifdef INET6 509 if (isipv6) { 510 bcopy((caddr_t)ip6, mtod(m, caddr_t), 511 sizeof(struct ip6_hdr)); 512 ip6 = mtod(m, struct ip6_hdr *); 513 nth = (struct tcphdr *)(ip6 + 1); 514 } else 515 #endif /* INET6 */ 516 { 517 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 518 ip = mtod(m, struct ip *); 519 nth = (struct tcphdr *)(ip + 1); 520 } 521 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 522 flags = TH_ACK; 523 } else { 524 m_freem(m->m_next); 525 m->m_next = NULL; 526 m->m_data = (caddr_t)ipgen; 527 /* m_len is set later */ 528 tlen = 0; 529 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 530 #ifdef INET6 531 if (isipv6) { 532 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 533 nth = (struct tcphdr *)(ip6 + 1); 534 } else 535 #endif /* INET6 */ 536 { 537 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 538 nth = (struct tcphdr *)(ip + 1); 539 } 540 if (th != nth) { 541 /* 542 * this is usually a case when an extension header 543 * exists between the IPv6 header and the 544 * TCP header. 545 */ 546 nth->th_sport = th->th_sport; 547 nth->th_dport = th->th_dport; 548 } 549 xchg(nth->th_dport, nth->th_sport, n_short); 550 #undef xchg 551 } 552 #ifdef INET6 553 if (isipv6) { 554 ip6->ip6_flow = 0; 555 ip6->ip6_vfc = IPV6_VERSION; 556 ip6->ip6_nxt = IPPROTO_TCP; 557 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 558 tlen)); 559 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 560 } else 561 #endif 562 { 563 tlen += sizeof (struct tcpiphdr); 564 ip->ip_len = tlen; 565 ip->ip_ttl = ip_defttl; 566 if (path_mtu_discovery) 567 ip->ip_off |= IP_DF; 568 } 569 m->m_len = tlen; 570 m->m_pkthdr.len = tlen; 571 m->m_pkthdr.rcvif = NULL; 572 #ifdef MAC 573 if (inp != NULL) { 574 /* 575 * Packet is associated with a socket, so allow the 576 * label of the response to reflect the socket label. 577 */ 578 INP_LOCK_ASSERT(inp); 579 mac_create_mbuf_from_inpcb(inp, m); 580 } else { 581 /* 582 * Packet is not associated with a socket, so possibly 583 * update the label in place. 584 */ 585 mac_reflect_mbuf_tcp(m); 586 } 587 #endif 588 nth->th_seq = htonl(seq); 589 nth->th_ack = htonl(ack); 590 nth->th_x2 = 0; 591 nth->th_off = sizeof (struct tcphdr) >> 2; 592 nth->th_flags = flags; 593 if (tp != NULL) 594 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 595 else 596 nth->th_win = htons((u_short)win); 597 nth->th_urp = 0; 598 #ifdef INET6 599 if (isipv6) { 600 nth->th_sum = 0; 601 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 602 sizeof(struct ip6_hdr), 603 tlen - sizeof(struct ip6_hdr)); 604 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 605 NULL, NULL); 606 } else 607 #endif /* INET6 */ 608 { 609 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 610 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 611 m->m_pkthdr.csum_flags = CSUM_TCP; 612 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 613 } 614 #ifdef TCPDEBUG 615 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 616 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 617 #endif 618 #ifdef INET6 619 if (isipv6) 620 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 621 else 622 #endif /* INET6 */ 623 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 624 } 625 626 /* 627 * Create a new TCP control block, making an 628 * empty reassembly queue and hooking it to the argument 629 * protocol control block. The `inp' parameter must have 630 * come from the zone allocator set up in tcp_init(). 631 */ 632 struct tcpcb * 633 tcp_newtcpcb(struct inpcb *inp) 634 { 635 struct tcpcb_mem *tm; 636 struct tcpcb *tp; 637 #ifdef INET6 638 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 639 #endif /* INET6 */ 640 641 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 642 if (tm == NULL) 643 return (NULL); 644 tp = &tm->tcb; 645 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 646 tp->t_maxseg = tp->t_maxopd = 647 #ifdef INET6 648 isipv6 ? tcp_v6mssdflt : 649 #endif /* INET6 */ 650 tcp_mssdflt; 651 652 /* Set up our timeouts. */ 653 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE); 654 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE); 655 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE); 656 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE); 657 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE); 658 659 if (tcp_do_rfc1323) 660 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 661 tp->sack_enable = tcp_do_sack; 662 TAILQ_INIT(&tp->snd_holes); 663 tp->t_inpcb = inp; /* XXX */ 664 /* 665 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 666 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 667 * reasonable initial retransmit time. 668 */ 669 tp->t_srtt = TCPTV_SRTTBASE; 670 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 671 tp->t_rttmin = tcp_rexmit_min; 672 tp->t_rxtcur = TCPTV_RTOBASE; 673 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 674 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 675 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 676 tp->t_rcvtime = ticks; 677 tp->t_bw_rtttime = ticks; 678 /* 679 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 680 * because the socket may be bound to an IPv6 wildcard address, 681 * which may match an IPv4-mapped IPv6 address. 682 */ 683 inp->inp_ip_ttl = ip_defttl; 684 inp->inp_ppcb = tp; 685 return (tp); /* XXX */ 686 } 687 688 /* 689 * Drop a TCP connection, reporting 690 * the specified error. If connection is synchronized, 691 * then send a RST to peer. 692 */ 693 struct tcpcb * 694 tcp_drop(struct tcpcb *tp, int errno) 695 { 696 struct socket *so = tp->t_inpcb->inp_socket; 697 698 INP_INFO_WLOCK_ASSERT(&tcbinfo); 699 INP_LOCK_ASSERT(tp->t_inpcb); 700 701 if (TCPS_HAVERCVDSYN(tp->t_state)) { 702 tp->t_state = TCPS_CLOSED; 703 (void) tcp_output(tp); 704 tcpstat.tcps_drops++; 705 } else 706 tcpstat.tcps_conndrops++; 707 if (errno == ETIMEDOUT && tp->t_softerror) 708 errno = tp->t_softerror; 709 so->so_error = errno; 710 return (tcp_close(tp)); 711 } 712 713 void 714 tcp_discardcb(struct tcpcb *tp) 715 { 716 struct tseg_qent *q; 717 struct inpcb *inp = tp->t_inpcb; 718 struct socket *so = inp->inp_socket; 719 #ifdef INET6 720 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 721 #endif /* INET6 */ 722 723 INP_LOCK_ASSERT(inp); 724 725 /* 726 * Make sure that all of our timers are stopped before we 727 * delete the PCB. 728 */ 729 callout_stop(tp->tt_rexmt); 730 callout_stop(tp->tt_persist); 731 callout_stop(tp->tt_keep); 732 callout_stop(tp->tt_2msl); 733 callout_stop(tp->tt_delack); 734 735 /* 736 * If we got enough samples through the srtt filter, 737 * save the rtt and rttvar in the routing entry. 738 * 'Enough' is arbitrarily defined as 4 rtt samples. 739 * 4 samples is enough for the srtt filter to converge 740 * to within enough % of the correct value; fewer samples 741 * and we could save a bogus rtt. The danger is not high 742 * as tcp quickly recovers from everything. 743 * XXX: Works very well but needs some more statistics! 744 */ 745 if (tp->t_rttupdated >= 4) { 746 struct hc_metrics_lite metrics; 747 u_long ssthresh; 748 749 bzero(&metrics, sizeof(metrics)); 750 /* 751 * Update the ssthresh always when the conditions below 752 * are satisfied. This gives us better new start value 753 * for the congestion avoidance for new connections. 754 * ssthresh is only set if packet loss occured on a session. 755 * 756 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 757 * being torn down. Ideally this code would not use 'so'. 758 */ 759 ssthresh = tp->snd_ssthresh; 760 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 761 /* 762 * convert the limit from user data bytes to 763 * packets then to packet data bytes. 764 */ 765 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 766 if (ssthresh < 2) 767 ssthresh = 2; 768 ssthresh *= (u_long)(tp->t_maxseg + 769 #ifdef INET6 770 (isipv6 ? sizeof (struct ip6_hdr) + 771 sizeof (struct tcphdr) : 772 #endif 773 sizeof (struct tcpiphdr) 774 #ifdef INET6 775 ) 776 #endif 777 ); 778 } else 779 ssthresh = 0; 780 metrics.rmx_ssthresh = ssthresh; 781 782 metrics.rmx_rtt = tp->t_srtt; 783 metrics.rmx_rttvar = tp->t_rttvar; 784 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 785 metrics.rmx_bandwidth = tp->snd_bandwidth; 786 metrics.rmx_cwnd = tp->snd_cwnd; 787 metrics.rmx_sendpipe = 0; 788 metrics.rmx_recvpipe = 0; 789 790 tcp_hc_update(&inp->inp_inc, &metrics); 791 } 792 793 /* free the reassembly queue, if any */ 794 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 795 LIST_REMOVE(q, tqe_q); 796 m_freem(q->tqe_m); 797 uma_zfree(tcp_reass_zone, q); 798 tp->t_segqlen--; 799 tcp_reass_qsize--; 800 } 801 tcp_free_sackholes(tp); 802 inp->inp_ppcb = NULL; 803 tp->t_inpcb = NULL; 804 uma_zfree(tcpcb_zone, tp); 805 } 806 807 /* 808 * Attempt to close a TCP control block, marking it as dropped, and freeing 809 * the socket if we hold the only reference. 810 */ 811 struct tcpcb * 812 tcp_close(struct tcpcb *tp) 813 { 814 struct inpcb *inp = tp->t_inpcb; 815 struct socket *so; 816 817 INP_INFO_WLOCK_ASSERT(&tcbinfo); 818 INP_LOCK_ASSERT(inp); 819 820 in_pcbdrop(inp); 821 tcpstat.tcps_closed++; 822 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 823 so = inp->inp_socket; 824 soisdisconnected(so); 825 if (inp->inp_vflag & INP_SOCKREF) { 826 KASSERT(so->so_state & SS_PROTOREF, 827 ("tcp_close: !SS_PROTOREF")); 828 inp->inp_vflag &= ~INP_SOCKREF; 829 INP_UNLOCK(inp); 830 ACCEPT_LOCK(); 831 SOCK_LOCK(so); 832 so->so_state &= ~SS_PROTOREF; 833 sofree(so); 834 return (NULL); 835 } 836 return (tp); 837 } 838 839 void 840 tcp_drain(void) 841 { 842 843 if (do_tcpdrain) { 844 struct inpcb *inpb; 845 struct tcpcb *tcpb; 846 struct tseg_qent *te; 847 848 /* 849 * Walk the tcpbs, if existing, and flush the reassembly queue, 850 * if there is one... 851 * XXX: The "Net/3" implementation doesn't imply that the TCP 852 * reassembly queue should be flushed, but in a situation 853 * where we're really low on mbufs, this is potentially 854 * usefull. 855 */ 856 INP_INFO_RLOCK(&tcbinfo); 857 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 858 if (inpb->inp_vflag & INP_TIMEWAIT) 859 continue; 860 INP_LOCK(inpb); 861 if ((tcpb = intotcpcb(inpb)) != NULL) { 862 while ((te = LIST_FIRST(&tcpb->t_segq)) 863 != NULL) { 864 LIST_REMOVE(te, tqe_q); 865 m_freem(te->tqe_m); 866 uma_zfree(tcp_reass_zone, te); 867 tcpb->t_segqlen--; 868 tcp_reass_qsize--; 869 } 870 tcp_clean_sackreport(tcpb); 871 } 872 INP_UNLOCK(inpb); 873 } 874 INP_INFO_RUNLOCK(&tcbinfo); 875 } 876 } 877 878 /* 879 * Notify a tcp user of an asynchronous error; 880 * store error as soft error, but wake up user 881 * (for now, won't do anything until can select for soft error). 882 * 883 * Do not wake up user since there currently is no mechanism for 884 * reporting soft errors (yet - a kqueue filter may be added). 885 */ 886 static struct inpcb * 887 tcp_notify(struct inpcb *inp, int error) 888 { 889 struct tcpcb *tp; 890 891 INP_INFO_WLOCK_ASSERT(&tcbinfo); 892 INP_LOCK_ASSERT(inp); 893 894 if ((inp->inp_vflag & INP_TIMEWAIT) || 895 (inp->inp_vflag & INP_DROPPED)) 896 return (inp); 897 898 tp = intotcpcb(inp); 899 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 900 901 /* 902 * Ignore some errors if we are hooked up. 903 * If connection hasn't completed, has retransmitted several times, 904 * and receives a second error, give up now. This is better 905 * than waiting a long time to establish a connection that 906 * can never complete. 907 */ 908 if (tp->t_state == TCPS_ESTABLISHED && 909 (error == EHOSTUNREACH || error == ENETUNREACH || 910 error == EHOSTDOWN)) { 911 return (inp); 912 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 913 tp->t_softerror) { 914 tp = tcp_drop(tp, error); 915 if (tp != NULL) 916 return (inp); 917 else 918 return (NULL); 919 } else { 920 tp->t_softerror = error; 921 return (inp); 922 } 923 #if 0 924 wakeup( &so->so_timeo); 925 sorwakeup(so); 926 sowwakeup(so); 927 #endif 928 } 929 930 static int 931 tcp_pcblist(SYSCTL_HANDLER_ARGS) 932 { 933 int error, i, n; 934 struct inpcb *inp, **inp_list; 935 inp_gen_t gencnt; 936 struct xinpgen xig; 937 938 /* 939 * The process of preparing the TCB list is too time-consuming and 940 * resource-intensive to repeat twice on every request. 941 */ 942 if (req->oldptr == NULL) { 943 n = tcbinfo.ipi_count; 944 req->oldidx = 2 * (sizeof xig) 945 + (n + n/8) * sizeof(struct xtcpcb); 946 return (0); 947 } 948 949 if (req->newptr != NULL) 950 return (EPERM); 951 952 /* 953 * OK, now we're committed to doing something. 954 */ 955 INP_INFO_RLOCK(&tcbinfo); 956 gencnt = tcbinfo.ipi_gencnt; 957 n = tcbinfo.ipi_count; 958 INP_INFO_RUNLOCK(&tcbinfo); 959 960 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 961 + n * sizeof(struct xtcpcb)); 962 if (error != 0) 963 return (error); 964 965 xig.xig_len = sizeof xig; 966 xig.xig_count = n; 967 xig.xig_gen = gencnt; 968 xig.xig_sogen = so_gencnt; 969 error = SYSCTL_OUT(req, &xig, sizeof xig); 970 if (error) 971 return (error); 972 973 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 974 if (inp_list == NULL) 975 return (ENOMEM); 976 977 INP_INFO_RLOCK(&tcbinfo); 978 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 979 inp = LIST_NEXT(inp, inp_list)) { 980 INP_LOCK(inp); 981 if (inp->inp_gencnt <= gencnt) { 982 /* 983 * XXX: This use of cr_cansee(), introduced with 984 * TCP state changes, is not quite right, but for 985 * now, better than nothing. 986 */ 987 if (inp->inp_vflag & INP_TIMEWAIT) { 988 if (intotw(inp) != NULL) 989 error = cr_cansee(req->td->td_ucred, 990 intotw(inp)->tw_cred); 991 else 992 error = EINVAL; /* Skip this inp. */ 993 } else 994 error = cr_canseesocket(req->td->td_ucred, 995 inp->inp_socket); 996 if (error == 0) 997 inp_list[i++] = inp; 998 } 999 INP_UNLOCK(inp); 1000 } 1001 INP_INFO_RUNLOCK(&tcbinfo); 1002 n = i; 1003 1004 error = 0; 1005 for (i = 0; i < n; i++) { 1006 inp = inp_list[i]; 1007 INP_LOCK(inp); 1008 if (inp->inp_gencnt <= gencnt) { 1009 struct xtcpcb xt; 1010 void *inp_ppcb; 1011 1012 bzero(&xt, sizeof(xt)); 1013 xt.xt_len = sizeof xt; 1014 /* XXX should avoid extra copy */ 1015 bcopy(inp, &xt.xt_inp, sizeof *inp); 1016 inp_ppcb = inp->inp_ppcb; 1017 if (inp_ppcb == NULL) 1018 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1019 else if (inp->inp_vflag & INP_TIMEWAIT) { 1020 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1021 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1022 } else 1023 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1024 if (inp->inp_socket != NULL) 1025 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1026 else { 1027 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1028 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1029 } 1030 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1031 INP_UNLOCK(inp); 1032 error = SYSCTL_OUT(req, &xt, sizeof xt); 1033 } else 1034 INP_UNLOCK(inp); 1035 1036 } 1037 if (!error) { 1038 /* 1039 * Give the user an updated idea of our state. 1040 * If the generation differs from what we told 1041 * her before, she knows that something happened 1042 * while we were processing this request, and it 1043 * might be necessary to retry. 1044 */ 1045 INP_INFO_RLOCK(&tcbinfo); 1046 xig.xig_gen = tcbinfo.ipi_gencnt; 1047 xig.xig_sogen = so_gencnt; 1048 xig.xig_count = tcbinfo.ipi_count; 1049 INP_INFO_RUNLOCK(&tcbinfo); 1050 error = SYSCTL_OUT(req, &xig, sizeof xig); 1051 } 1052 free(inp_list, M_TEMP); 1053 return (error); 1054 } 1055 1056 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1057 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1058 1059 static int 1060 tcp_getcred(SYSCTL_HANDLER_ARGS) 1061 { 1062 struct xucred xuc; 1063 struct sockaddr_in addrs[2]; 1064 struct inpcb *inp; 1065 int error; 1066 1067 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1068 if (error) 1069 return (error); 1070 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1071 if (error) 1072 return (error); 1073 INP_INFO_RLOCK(&tcbinfo); 1074 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1075 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1076 if (inp == NULL) { 1077 error = ENOENT; 1078 goto outunlocked; 1079 } 1080 INP_LOCK(inp); 1081 if (inp->inp_socket == NULL) { 1082 error = ENOENT; 1083 goto out; 1084 } 1085 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1086 if (error) 1087 goto out; 1088 cru2x(inp->inp_socket->so_cred, &xuc); 1089 out: 1090 INP_UNLOCK(inp); 1091 outunlocked: 1092 INP_INFO_RUNLOCK(&tcbinfo); 1093 if (error == 0) 1094 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1095 return (error); 1096 } 1097 1098 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1099 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1100 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1101 1102 #ifdef INET6 1103 static int 1104 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1105 { 1106 struct xucred xuc; 1107 struct sockaddr_in6 addrs[2]; 1108 struct inpcb *inp; 1109 int error, mapped = 0; 1110 1111 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1112 if (error) 1113 return (error); 1114 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1115 if (error) 1116 return (error); 1117 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1118 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1119 return (error); 1120 } 1121 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1122 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1123 mapped = 1; 1124 else 1125 return (EINVAL); 1126 } 1127 1128 INP_INFO_RLOCK(&tcbinfo); 1129 if (mapped == 1) 1130 inp = in_pcblookup_hash(&tcbinfo, 1131 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1132 addrs[1].sin6_port, 1133 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1134 addrs[0].sin6_port, 1135 0, NULL); 1136 else 1137 inp = in6_pcblookup_hash(&tcbinfo, 1138 &addrs[1].sin6_addr, addrs[1].sin6_port, 1139 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1140 if (inp == NULL) { 1141 error = ENOENT; 1142 goto outunlocked; 1143 } 1144 INP_LOCK(inp); 1145 if (inp->inp_socket == NULL) { 1146 error = ENOENT; 1147 goto out; 1148 } 1149 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1150 if (error) 1151 goto out; 1152 cru2x(inp->inp_socket->so_cred, &xuc); 1153 out: 1154 INP_UNLOCK(inp); 1155 outunlocked: 1156 INP_INFO_RUNLOCK(&tcbinfo); 1157 if (error == 0) 1158 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1159 return (error); 1160 } 1161 1162 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1163 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1164 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1165 #endif 1166 1167 1168 void 1169 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1170 { 1171 struct ip *ip = vip; 1172 struct tcphdr *th; 1173 struct in_addr faddr; 1174 struct inpcb *inp; 1175 struct tcpcb *tp; 1176 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1177 struct icmp *icp; 1178 struct in_conninfo inc; 1179 tcp_seq icmp_tcp_seq; 1180 int mtu; 1181 1182 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1183 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1184 return; 1185 1186 if (cmd == PRC_MSGSIZE) 1187 notify = tcp_mtudisc; 1188 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1189 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1190 notify = tcp_drop_syn_sent; 1191 /* 1192 * Redirects don't need to be handled up here. 1193 */ 1194 else if (PRC_IS_REDIRECT(cmd)) 1195 return; 1196 /* 1197 * Source quench is depreciated. 1198 */ 1199 else if (cmd == PRC_QUENCH) 1200 return; 1201 /* 1202 * Hostdead is ugly because it goes linearly through all PCBs. 1203 * XXX: We never get this from ICMP, otherwise it makes an 1204 * excellent DoS attack on machines with many connections. 1205 */ 1206 else if (cmd == PRC_HOSTDEAD) 1207 ip = NULL; 1208 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1209 return; 1210 if (ip != NULL) { 1211 icp = (struct icmp *)((caddr_t)ip 1212 - offsetof(struct icmp, icmp_ip)); 1213 th = (struct tcphdr *)((caddr_t)ip 1214 + (ip->ip_hl << 2)); 1215 INP_INFO_WLOCK(&tcbinfo); 1216 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1217 ip->ip_src, th->th_sport, 0, NULL); 1218 if (inp != NULL) { 1219 INP_LOCK(inp); 1220 if (!(inp->inp_vflag & INP_TIMEWAIT) && 1221 !(inp->inp_vflag & INP_DROPPED) && 1222 !(inp->inp_socket == NULL)) { 1223 icmp_tcp_seq = htonl(th->th_seq); 1224 tp = intotcpcb(inp); 1225 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1226 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1227 if (cmd == PRC_MSGSIZE) { 1228 /* 1229 * MTU discovery: 1230 * If we got a needfrag set the MTU 1231 * in the route to the suggested new 1232 * value (if given) and then notify. 1233 */ 1234 bzero(&inc, sizeof(inc)); 1235 inc.inc_flags = 0; /* IPv4 */ 1236 inc.inc_faddr = faddr; 1237 1238 mtu = ntohs(icp->icmp_nextmtu); 1239 /* 1240 * If no alternative MTU was 1241 * proposed, try the next smaller 1242 * one. ip->ip_len has already 1243 * been swapped in icmp_input(). 1244 */ 1245 if (!mtu) 1246 mtu = ip_next_mtu(ip->ip_len, 1247 1); 1248 if (mtu < max(296, (tcp_minmss) 1249 + sizeof(struct tcpiphdr))) 1250 mtu = 0; 1251 if (!mtu) 1252 mtu = tcp_mssdflt 1253 + sizeof(struct tcpiphdr); 1254 /* 1255 * Only cache the the MTU if it 1256 * is smaller than the interface 1257 * or route MTU. tcp_mtudisc() 1258 * will do right thing by itself. 1259 */ 1260 if (mtu <= tcp_maxmtu(&inc, NULL)) 1261 tcp_hc_updatemtu(&inc, mtu); 1262 } 1263 1264 inp = (*notify)(inp, inetctlerrmap[cmd]); 1265 } 1266 } 1267 if (inp != NULL) 1268 INP_UNLOCK(inp); 1269 } else { 1270 inc.inc_fport = th->th_dport; 1271 inc.inc_lport = th->th_sport; 1272 inc.inc_faddr = faddr; 1273 inc.inc_laddr = ip->ip_src; 1274 #ifdef INET6 1275 inc.inc_isipv6 = 0; 1276 #endif 1277 syncache_unreach(&inc, th); 1278 } 1279 INP_INFO_WUNLOCK(&tcbinfo); 1280 } else 1281 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1282 } 1283 1284 #ifdef INET6 1285 void 1286 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1287 { 1288 struct tcphdr th; 1289 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1290 struct ip6_hdr *ip6; 1291 struct mbuf *m; 1292 struct ip6ctlparam *ip6cp = NULL; 1293 const struct sockaddr_in6 *sa6_src = NULL; 1294 int off; 1295 struct tcp_portonly { 1296 u_int16_t th_sport; 1297 u_int16_t th_dport; 1298 } *thp; 1299 1300 if (sa->sa_family != AF_INET6 || 1301 sa->sa_len != sizeof(struct sockaddr_in6)) 1302 return; 1303 1304 if (cmd == PRC_MSGSIZE) 1305 notify = tcp_mtudisc; 1306 else if (!PRC_IS_REDIRECT(cmd) && 1307 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1308 return; 1309 /* Source quench is depreciated. */ 1310 else if (cmd == PRC_QUENCH) 1311 return; 1312 1313 /* if the parameter is from icmp6, decode it. */ 1314 if (d != NULL) { 1315 ip6cp = (struct ip6ctlparam *)d; 1316 m = ip6cp->ip6c_m; 1317 ip6 = ip6cp->ip6c_ip6; 1318 off = ip6cp->ip6c_off; 1319 sa6_src = ip6cp->ip6c_src; 1320 } else { 1321 m = NULL; 1322 ip6 = NULL; 1323 off = 0; /* fool gcc */ 1324 sa6_src = &sa6_any; 1325 } 1326 1327 if (ip6 != NULL) { 1328 struct in_conninfo inc; 1329 /* 1330 * XXX: We assume that when IPV6 is non NULL, 1331 * M and OFF are valid. 1332 */ 1333 1334 /* check if we can safely examine src and dst ports */ 1335 if (m->m_pkthdr.len < off + sizeof(*thp)) 1336 return; 1337 1338 bzero(&th, sizeof(th)); 1339 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1340 1341 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1342 (struct sockaddr *)ip6cp->ip6c_src, 1343 th.th_sport, cmd, NULL, notify); 1344 1345 inc.inc_fport = th.th_dport; 1346 inc.inc_lport = th.th_sport; 1347 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1348 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1349 inc.inc_isipv6 = 1; 1350 INP_INFO_WLOCK(&tcbinfo); 1351 syncache_unreach(&inc, &th); 1352 INP_INFO_WUNLOCK(&tcbinfo); 1353 } else 1354 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1355 0, cmd, NULL, notify); 1356 } 1357 #endif /* INET6 */ 1358 1359 1360 /* 1361 * Following is where TCP initial sequence number generation occurs. 1362 * 1363 * There are two places where we must use initial sequence numbers: 1364 * 1. In SYN-ACK packets. 1365 * 2. In SYN packets. 1366 * 1367 * All ISNs for SYN-ACK packets are generated by the syncache. See 1368 * tcp_syncache.c for details. 1369 * 1370 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1371 * depends on this property. In addition, these ISNs should be 1372 * unguessable so as to prevent connection hijacking. To satisfy 1373 * the requirements of this situation, the algorithm outlined in 1374 * RFC 1948 is used, with only small modifications. 1375 * 1376 * Implementation details: 1377 * 1378 * Time is based off the system timer, and is corrected so that it 1379 * increases by one megabyte per second. This allows for proper 1380 * recycling on high speed LANs while still leaving over an hour 1381 * before rollover. 1382 * 1383 * As reading the *exact* system time is too expensive to be done 1384 * whenever setting up a TCP connection, we increment the time 1385 * offset in two ways. First, a small random positive increment 1386 * is added to isn_offset for each connection that is set up. 1387 * Second, the function tcp_isn_tick fires once per clock tick 1388 * and increments isn_offset as necessary so that sequence numbers 1389 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1390 * random positive increments serve only to ensure that the same 1391 * exact sequence number is never sent out twice (as could otherwise 1392 * happen when a port is recycled in less than the system tick 1393 * interval.) 1394 * 1395 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1396 * between seeding of isn_secret. This is normally set to zero, 1397 * as reseeding should not be necessary. 1398 * 1399 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1400 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1401 * general, this means holding an exclusive (write) lock. 1402 */ 1403 1404 #define ISN_BYTES_PER_SECOND 1048576 1405 #define ISN_STATIC_INCREMENT 4096 1406 #define ISN_RANDOM_INCREMENT (4096 - 1) 1407 1408 static u_char isn_secret[32]; 1409 static int isn_last_reseed; 1410 static u_int32_t isn_offset, isn_offset_old; 1411 static MD5_CTX isn_ctx; 1412 1413 tcp_seq 1414 tcp_new_isn(struct tcpcb *tp) 1415 { 1416 u_int32_t md5_buffer[4]; 1417 tcp_seq new_isn; 1418 1419 INP_LOCK_ASSERT(tp->t_inpcb); 1420 1421 ISN_LOCK(); 1422 /* Seed if this is the first use, reseed if requested. */ 1423 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1424 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1425 < (u_int)ticks))) { 1426 read_random(&isn_secret, sizeof(isn_secret)); 1427 isn_last_reseed = ticks; 1428 } 1429 1430 /* Compute the md5 hash and return the ISN. */ 1431 MD5Init(&isn_ctx); 1432 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1433 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1434 #ifdef INET6 1435 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1436 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1437 sizeof(struct in6_addr)); 1438 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1439 sizeof(struct in6_addr)); 1440 } else 1441 #endif 1442 { 1443 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1444 sizeof(struct in_addr)); 1445 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1446 sizeof(struct in_addr)); 1447 } 1448 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1449 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1450 new_isn = (tcp_seq) md5_buffer[0]; 1451 isn_offset += ISN_STATIC_INCREMENT + 1452 (arc4random() & ISN_RANDOM_INCREMENT); 1453 new_isn += isn_offset; 1454 ISN_UNLOCK(); 1455 return (new_isn); 1456 } 1457 1458 /* 1459 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1460 * to keep time flowing at a relatively constant rate. If the random 1461 * increments have already pushed us past the projected offset, do nothing. 1462 */ 1463 static void 1464 tcp_isn_tick(void *xtp) 1465 { 1466 u_int32_t projected_offset; 1467 1468 ISN_LOCK(); 1469 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1470 1471 if (projected_offset > isn_offset) 1472 isn_offset = projected_offset; 1473 1474 isn_offset_old = isn_offset; 1475 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1476 ISN_UNLOCK(); 1477 } 1478 1479 /* 1480 * When a specific ICMP unreachable message is received and the 1481 * connection state is SYN-SENT, drop the connection. This behavior 1482 * is controlled by the icmp_may_rst sysctl. 1483 */ 1484 struct inpcb * 1485 tcp_drop_syn_sent(struct inpcb *inp, int errno) 1486 { 1487 struct tcpcb *tp; 1488 1489 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1490 INP_LOCK_ASSERT(inp); 1491 1492 if ((inp->inp_vflag & INP_TIMEWAIT) || 1493 (inp->inp_vflag & INP_DROPPED)) 1494 return (inp); 1495 1496 tp = intotcpcb(inp); 1497 if (tp->t_state != TCPS_SYN_SENT) 1498 return (inp); 1499 1500 tp = tcp_drop(tp, errno); 1501 if (tp != NULL) 1502 return (inp); 1503 else 1504 return (NULL); 1505 } 1506 1507 /* 1508 * When `need fragmentation' ICMP is received, update our idea of the MSS 1509 * based on the new value in the route. Also nudge TCP to send something, 1510 * since we know the packet we just sent was dropped. 1511 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1512 */ 1513 struct inpcb * 1514 tcp_mtudisc(struct inpcb *inp, int errno) 1515 { 1516 struct tcpcb *tp; 1517 struct socket *so = inp->inp_socket; 1518 u_int maxmtu; 1519 u_int romtu; 1520 int mss; 1521 #ifdef INET6 1522 int isipv6; 1523 #endif /* INET6 */ 1524 1525 INP_LOCK_ASSERT(inp); 1526 if ((inp->inp_vflag & INP_TIMEWAIT) || 1527 (inp->inp_vflag & INP_DROPPED)) 1528 return (inp); 1529 1530 tp = intotcpcb(inp); 1531 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1532 1533 #ifdef INET6 1534 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1535 #endif 1536 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1537 romtu = 1538 #ifdef INET6 1539 isipv6 ? tcp_maxmtu6(&inp->inp_inc, NULL) : 1540 #endif /* INET6 */ 1541 tcp_maxmtu(&inp->inp_inc, NULL); 1542 if (!maxmtu) 1543 maxmtu = romtu; 1544 else 1545 maxmtu = min(maxmtu, romtu); 1546 if (!maxmtu) { 1547 tp->t_maxopd = tp->t_maxseg = 1548 #ifdef INET6 1549 isipv6 ? tcp_v6mssdflt : 1550 #endif /* INET6 */ 1551 tcp_mssdflt; 1552 return (inp); 1553 } 1554 mss = maxmtu - 1555 #ifdef INET6 1556 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1557 #endif /* INET6 */ 1558 sizeof(struct tcpiphdr) 1559 #ifdef INET6 1560 ) 1561 #endif /* INET6 */ 1562 ; 1563 1564 /* 1565 * XXX - The above conditional probably violates the TCP 1566 * spec. The problem is that, since we don't know the 1567 * other end's MSS, we are supposed to use a conservative 1568 * default. But, if we do that, then MTU discovery will 1569 * never actually take place, because the conservative 1570 * default is much less than the MTUs typically seen 1571 * on the Internet today. For the moment, we'll sweep 1572 * this under the carpet. 1573 * 1574 * The conservative default might not actually be a problem 1575 * if the only case this occurs is when sending an initial 1576 * SYN with options and data to a host we've never talked 1577 * to before. Then, they will reply with an MSS value which 1578 * will get recorded and the new parameters should get 1579 * recomputed. For Further Study. 1580 */ 1581 if (tp->t_maxopd <= mss) 1582 return (inp); 1583 tp->t_maxopd = mss; 1584 1585 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1586 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1587 mss -= TCPOLEN_TSTAMP_APPA; 1588 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1589 if (mss > MCLBYTES) 1590 mss &= ~(MCLBYTES-1); 1591 #else 1592 if (mss > MCLBYTES) 1593 mss = mss / MCLBYTES * MCLBYTES; 1594 #endif 1595 if (so->so_snd.sb_hiwat < mss) 1596 mss = so->so_snd.sb_hiwat; 1597 1598 tp->t_maxseg = mss; 1599 1600 tcpstat.tcps_mturesent++; 1601 tp->t_rtttime = 0; 1602 tp->snd_nxt = tp->snd_una; 1603 tcp_free_sackholes(tp); 1604 tp->snd_recover = tp->snd_max; 1605 if (tp->sack_enable) 1606 EXIT_FASTRECOVERY(tp); 1607 tcp_output(tp); 1608 return (inp); 1609 } 1610 1611 /* 1612 * Look-up the routing entry to the peer of this inpcb. If no route 1613 * is found and it cannot be allocated, then return NULL. This routine 1614 * is called by TCP routines that access the rmx structure and by tcp_mss 1615 * to get the interface MTU. 1616 */ 1617 u_long 1618 tcp_maxmtu(struct in_conninfo *inc, int *flags) 1619 { 1620 struct route sro; 1621 struct sockaddr_in *dst; 1622 struct ifnet *ifp; 1623 u_long maxmtu = 0; 1624 1625 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1626 1627 bzero(&sro, sizeof(sro)); 1628 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1629 dst = (struct sockaddr_in *)&sro.ro_dst; 1630 dst->sin_family = AF_INET; 1631 dst->sin_len = sizeof(*dst); 1632 dst->sin_addr = inc->inc_faddr; 1633 rtalloc_ign(&sro, RTF_CLONING); 1634 } 1635 if (sro.ro_rt != NULL) { 1636 ifp = sro.ro_rt->rt_ifp; 1637 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1638 maxmtu = ifp->if_mtu; 1639 else 1640 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1641 1642 /* Report additional interface capabilities. */ 1643 if (flags != NULL) { 1644 if (ifp->if_capenable & IFCAP_TSO4 && 1645 ifp->if_hwassist & CSUM_TSO) 1646 *flags |= CSUM_TSO; 1647 } 1648 RTFREE(sro.ro_rt); 1649 } 1650 return (maxmtu); 1651 } 1652 1653 #ifdef INET6 1654 u_long 1655 tcp_maxmtu6(struct in_conninfo *inc, int *flags) 1656 { 1657 struct route_in6 sro6; 1658 struct ifnet *ifp; 1659 u_long maxmtu = 0; 1660 1661 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1662 1663 bzero(&sro6, sizeof(sro6)); 1664 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1665 sro6.ro_dst.sin6_family = AF_INET6; 1666 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1667 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1668 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1669 } 1670 if (sro6.ro_rt != NULL) { 1671 ifp = sro6.ro_rt->rt_ifp; 1672 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1673 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1674 else 1675 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1676 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1677 1678 /* Report additional interface capabilities. */ 1679 if (flags != NULL) { 1680 if (ifp->if_capenable & IFCAP_TSO6 && 1681 ifp->if_hwassist & CSUM_TSO) 1682 *flags |= CSUM_TSO; 1683 } 1684 RTFREE(sro6.ro_rt); 1685 } 1686 1687 return (maxmtu); 1688 } 1689 #endif /* INET6 */ 1690 1691 #ifdef IPSEC 1692 /* compute ESP/AH header size for TCP, including outer IP header. */ 1693 size_t 1694 ipsec_hdrsiz_tcp(struct tcpcb *tp) 1695 { 1696 struct inpcb *inp; 1697 struct mbuf *m; 1698 size_t hdrsiz; 1699 struct ip *ip; 1700 #ifdef INET6 1701 struct ip6_hdr *ip6; 1702 #endif 1703 struct tcphdr *th; 1704 1705 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1706 return (0); 1707 MGETHDR(m, M_DONTWAIT, MT_DATA); 1708 if (!m) 1709 return (0); 1710 1711 #ifdef INET6 1712 if ((inp->inp_vflag & INP_IPV6) != 0) { 1713 ip6 = mtod(m, struct ip6_hdr *); 1714 th = (struct tcphdr *)(ip6 + 1); 1715 m->m_pkthdr.len = m->m_len = 1716 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1717 tcpip_fillheaders(inp, ip6, th); 1718 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1719 } else 1720 #endif /* INET6 */ 1721 { 1722 ip = mtod(m, struct ip *); 1723 th = (struct tcphdr *)(ip + 1); 1724 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1725 tcpip_fillheaders(inp, ip, th); 1726 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1727 } 1728 1729 m_free(m); 1730 return (hdrsiz); 1731 } 1732 #endif /*IPSEC*/ 1733 1734 /* 1735 * Move a TCP connection into TIME_WAIT state. 1736 * tcbinfo is locked. 1737 * inp is locked, and is unlocked before returning. 1738 */ 1739 void 1740 tcp_twstart(struct tcpcb *tp) 1741 { 1742 struct tcptw *tw; 1743 struct inpcb *inp = tp->t_inpcb; 1744 int acknow; 1745 struct socket *so; 1746 1747 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1748 INP_LOCK_ASSERT(inp); 1749 1750 if (nolocaltimewait && in_localip(inp->inp_faddr)) { 1751 tp = tcp_close(tp); 1752 if (tp != NULL) 1753 INP_UNLOCK(inp); 1754 return; 1755 } 1756 1757 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1758 if (tw == NULL) { 1759 tw = tcp_timer_2msl_tw(1); 1760 if (tw == NULL) { 1761 tp = tcp_close(tp); 1762 if (tp != NULL) 1763 INP_UNLOCK(inp); 1764 return; 1765 } 1766 } 1767 tw->tw_inpcb = inp; 1768 1769 /* 1770 * Recover last window size sent. 1771 */ 1772 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1773 1774 /* 1775 * Set t_recent if timestamps are used on the connection. 1776 */ 1777 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1778 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1779 tw->t_recent = tp->ts_recent; 1780 else 1781 tw->t_recent = 0; 1782 1783 tw->snd_nxt = tp->snd_nxt; 1784 tw->rcv_nxt = tp->rcv_nxt; 1785 tw->iss = tp->iss; 1786 tw->irs = tp->irs; 1787 tw->t_starttime = tp->t_starttime; 1788 tw->tw_time = 0; 1789 1790 /* XXX 1791 * If this code will 1792 * be used for fin-wait-2 state also, then we may need 1793 * a ts_recent from the last segment. 1794 */ 1795 acknow = tp->t_flags & TF_ACKNOW; 1796 1797 /* 1798 * First, discard tcpcb state, which includes stopping its timers and 1799 * freeing it. tcp_discardcb() used to also release the inpcb, but 1800 * that work is now done in the caller. 1801 * 1802 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 1803 * and might not be needed here any longer. 1804 */ 1805 tcp_discardcb(tp); 1806 so = inp->inp_socket; 1807 soisdisconnected(so); 1808 SOCK_LOCK(so); 1809 tw->tw_cred = crhold(so->so_cred); 1810 tw->tw_so_options = so->so_options; 1811 SOCK_UNLOCK(so); 1812 if (acknow) 1813 tcp_twrespond(tw, TH_ACK); 1814 inp->inp_ppcb = tw; 1815 inp->inp_vflag |= INP_TIMEWAIT; 1816 tcp_timer_2msl_reset(tw, 0); 1817 1818 /* 1819 * If the inpcb owns the sole reference to the socket, then we can 1820 * detach and free the socket as it is not needed in time wait. 1821 */ 1822 if (inp->inp_vflag & INP_SOCKREF) { 1823 KASSERT(so->so_state & SS_PROTOREF, 1824 ("tcp_twstart: !SS_PROTOREF")); 1825 inp->inp_vflag &= ~INP_SOCKREF; 1826 INP_UNLOCK(inp); 1827 ACCEPT_LOCK(); 1828 SOCK_LOCK(so); 1829 so->so_state &= ~SS_PROTOREF; 1830 sofree(so); 1831 } else 1832 INP_UNLOCK(inp); 1833 } 1834 1835 #if 0 1836 /* 1837 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1838 * the actual rate is slightly higher due to the addition of 1839 * random positive increments. 1840 * 1841 * Most other new OSes use semi-randomized ISN values, so we 1842 * do not need to worry about them. 1843 */ 1844 #define MS_ISN_BYTES_PER_SECOND 250000 1845 1846 /* 1847 * Determine if the ISN we will generate has advanced beyond the last 1848 * sequence number used by the previous connection. If so, indicate 1849 * that it is safe to recycle this tw socket by returning 1. 1850 */ 1851 int 1852 tcp_twrecycleable(struct tcptw *tw) 1853 { 1854 tcp_seq new_iss = tw->iss; 1855 tcp_seq new_irs = tw->irs; 1856 1857 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1858 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1859 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1860 1861 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1862 return (1); 1863 else 1864 return (0); 1865 } 1866 #endif 1867 1868 void 1869 tcp_twclose(struct tcptw *tw, int reuse) 1870 { 1871 struct socket *so; 1872 struct inpcb *inp; 1873 1874 /* 1875 * At this point, we are in one of two situations: 1876 * 1877 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 1878 * all state. 1879 * 1880 * (2) We have a socket -- if we own a reference, release it and 1881 * notify the socket layer. 1882 */ 1883 inp = tw->tw_inpcb; 1884 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1885 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 1886 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1887 INP_LOCK_ASSERT(inp); 1888 1889 tw->tw_inpcb = NULL; 1890 tcp_timer_2msl_stop(tw); 1891 inp->inp_ppcb = NULL; 1892 in_pcbdrop(inp); 1893 1894 so = inp->inp_socket; 1895 if (so != NULL) { 1896 /* 1897 * If there's a socket, handle two cases: first, we own a 1898 * strong reference, which we will now release, or we don't 1899 * in which case another reference exists (XXXRW: think 1900 * about this more), and we don't need to take action. 1901 */ 1902 if (inp->inp_vflag & INP_SOCKREF) { 1903 inp->inp_vflag &= ~INP_SOCKREF; 1904 INP_UNLOCK(inp); 1905 ACCEPT_LOCK(); 1906 SOCK_LOCK(so); 1907 KASSERT(so->so_state & SS_PROTOREF, 1908 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1909 so->so_state &= ~SS_PROTOREF; 1910 sofree(so); 1911 } else { 1912 /* 1913 * If we don't own the only reference, the socket and 1914 * inpcb need to be left around to be handled by 1915 * tcp_usr_detach() later. 1916 */ 1917 INP_UNLOCK(inp); 1918 } 1919 } else { 1920 #ifdef INET6 1921 if (inp->inp_vflag & INP_IPV6PROTO) 1922 in6_pcbfree(inp); 1923 else 1924 #endif 1925 in_pcbfree(inp); 1926 } 1927 tcpstat.tcps_closed++; 1928 crfree(tw->tw_cred); 1929 tw->tw_cred = NULL; 1930 if (reuse) 1931 return; 1932 uma_zfree(tcptw_zone, tw); 1933 } 1934 1935 int 1936 tcp_twrespond(struct tcptw *tw, int flags) 1937 { 1938 struct inpcb *inp = tw->tw_inpcb; 1939 struct tcphdr *th; 1940 struct mbuf *m; 1941 struct ip *ip = NULL; 1942 u_int8_t *optp; 1943 u_int hdrlen, optlen; 1944 int error; 1945 #ifdef INET6 1946 struct ip6_hdr *ip6 = NULL; 1947 int isipv6 = inp->inp_inc.inc_isipv6; 1948 #endif 1949 1950 INP_LOCK_ASSERT(inp); 1951 1952 m = m_gethdr(M_DONTWAIT, MT_DATA); 1953 if (m == NULL) 1954 return (ENOBUFS); 1955 m->m_data += max_linkhdr; 1956 1957 #ifdef MAC 1958 mac_create_mbuf_from_inpcb(inp, m); 1959 #endif 1960 1961 #ifdef INET6 1962 if (isipv6) { 1963 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1964 ip6 = mtod(m, struct ip6_hdr *); 1965 th = (struct tcphdr *)(ip6 + 1); 1966 tcpip_fillheaders(inp, ip6, th); 1967 } else 1968 #endif 1969 { 1970 hdrlen = sizeof(struct tcpiphdr); 1971 ip = mtod(m, struct ip *); 1972 th = (struct tcphdr *)(ip + 1); 1973 tcpip_fillheaders(inp, ip, th); 1974 } 1975 optp = (u_int8_t *)(th + 1); 1976 1977 /* 1978 * Send a timestamp and echo-reply if both our side and our peer 1979 * have sent timestamps in our SYN's and this is not a RST. 1980 */ 1981 if (tw->t_recent && flags == TH_ACK) { 1982 u_int32_t *lp = (u_int32_t *)optp; 1983 1984 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1985 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1986 *lp++ = htonl(ticks); 1987 *lp = htonl(tw->t_recent); 1988 optp += TCPOLEN_TSTAMP_APPA; 1989 } 1990 1991 optlen = optp - (u_int8_t *)(th + 1); 1992 1993 m->m_len = hdrlen + optlen; 1994 m->m_pkthdr.len = m->m_len; 1995 1996 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 1997 1998 th->th_seq = htonl(tw->snd_nxt); 1999 th->th_ack = htonl(tw->rcv_nxt); 2000 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 2001 th->th_flags = flags; 2002 th->th_win = htons(tw->last_win); 2003 2004 #ifdef INET6 2005 if (isipv6) { 2006 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 2007 sizeof(struct tcphdr) + optlen); 2008 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2009 error = ip6_output(m, inp->in6p_outputopts, NULL, 2010 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 2011 } else 2012 #endif 2013 { 2014 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2015 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 2016 m->m_pkthdr.csum_flags = CSUM_TCP; 2017 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2018 ip->ip_len = m->m_pkthdr.len; 2019 if (path_mtu_discovery) 2020 ip->ip_off |= IP_DF; 2021 error = ip_output(m, inp->inp_options, NULL, 2022 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 2023 NULL, inp); 2024 } 2025 if (flags & TH_ACK) 2026 tcpstat.tcps_sndacks++; 2027 else 2028 tcpstat.tcps_sndctrl++; 2029 tcpstat.tcps_sndtotal++; 2030 return (error); 2031 } 2032 2033 /* 2034 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 2035 * 2036 * This code attempts to calculate the bandwidth-delay product as a 2037 * means of determining the optimal window size to maximize bandwidth, 2038 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 2039 * routers. This code also does a fairly good job keeping RTTs in check 2040 * across slow links like modems. We implement an algorithm which is very 2041 * similar (but not meant to be) TCP/Vegas. The code operates on the 2042 * transmitter side of a TCP connection and so only effects the transmit 2043 * side of the connection. 2044 * 2045 * BACKGROUND: TCP makes no provision for the management of buffer space 2046 * at the end points or at the intermediate routers and switches. A TCP 2047 * stream, whether using NewReno or not, will eventually buffer as 2048 * many packets as it is able and the only reason this typically works is 2049 * due to the fairly small default buffers made available for a connection 2050 * (typicaly 16K or 32K). As machines use larger windows and/or window 2051 * scaling it is now fairly easy for even a single TCP connection to blow-out 2052 * all available buffer space not only on the local interface, but on 2053 * intermediate routers and switches as well. NewReno makes a misguided 2054 * attempt to 'solve' this problem by waiting for an actual failure to occur, 2055 * then backing off, then steadily increasing the window again until another 2056 * failure occurs, ad-infinitum. This results in terrible oscillation that 2057 * is only made worse as network loads increase and the idea of intentionally 2058 * blowing out network buffers is, frankly, a terrible way to manage network 2059 * resources. 2060 * 2061 * It is far better to limit the transmit window prior to the failure 2062 * condition being achieved. There are two general ways to do this: First 2063 * you can 'scan' through different transmit window sizes and locate the 2064 * point where the RTT stops increasing, indicating that you have filled the 2065 * pipe, then scan backwards until you note that RTT stops decreasing, then 2066 * repeat ad-infinitum. This method works in principle but has severe 2067 * implementation issues due to RTT variances, timer granularity, and 2068 * instability in the algorithm which can lead to many false positives and 2069 * create oscillations as well as interact badly with other TCP streams 2070 * implementing the same algorithm. 2071 * 2072 * The second method is to limit the window to the bandwidth delay product 2073 * of the link. This is the method we implement. RTT variances and our 2074 * own manipulation of the congestion window, bwnd, can potentially 2075 * destabilize the algorithm. For this reason we have to stabilize the 2076 * elements used to calculate the window. We do this by using the minimum 2077 * observed RTT, the long term average of the observed bandwidth, and 2078 * by adding two segments worth of slop. It isn't perfect but it is able 2079 * to react to changing conditions and gives us a very stable basis on 2080 * which to extend the algorithm. 2081 */ 2082 void 2083 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2084 { 2085 u_long bw; 2086 u_long bwnd; 2087 int save_ticks; 2088 2089 INP_LOCK_ASSERT(tp->t_inpcb); 2090 2091 /* 2092 * If inflight_enable is disabled in the middle of a tcp connection, 2093 * make sure snd_bwnd is effectively disabled. 2094 */ 2095 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2096 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2097 tp->snd_bandwidth = 0; 2098 return; 2099 } 2100 2101 /* 2102 * Figure out the bandwidth. Due to the tick granularity this 2103 * is a very rough number and it MUST be averaged over a fairly 2104 * long period of time. XXX we need to take into account a link 2105 * that is not using all available bandwidth, but for now our 2106 * slop will ramp us up if this case occurs and the bandwidth later 2107 * increases. 2108 * 2109 * Note: if ticks rollover 'bw' may wind up negative. We must 2110 * effectively reset t_bw_rtttime for this case. 2111 */ 2112 save_ticks = ticks; 2113 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2114 return; 2115 2116 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2117 (save_ticks - tp->t_bw_rtttime); 2118 tp->t_bw_rtttime = save_ticks; 2119 tp->t_bw_rtseq = ack_seq; 2120 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2121 return; 2122 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2123 2124 tp->snd_bandwidth = bw; 2125 2126 /* 2127 * Calculate the semi-static bandwidth delay product, plus two maximal 2128 * segments. The additional slop puts us squarely in the sweet 2129 * spot and also handles the bandwidth run-up case and stabilization. 2130 * Without the slop we could be locking ourselves into a lower 2131 * bandwidth. 2132 * 2133 * Situations Handled: 2134 * (1) Prevents over-queueing of packets on LANs, especially on 2135 * high speed LANs, allowing larger TCP buffers to be 2136 * specified, and also does a good job preventing 2137 * over-queueing of packets over choke points like modems 2138 * (at least for the transmit side). 2139 * 2140 * (2) Is able to handle changing network loads (bandwidth 2141 * drops so bwnd drops, bandwidth increases so bwnd 2142 * increases). 2143 * 2144 * (3) Theoretically should stabilize in the face of multiple 2145 * connections implementing the same algorithm (this may need 2146 * a little work). 2147 * 2148 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2149 * be adjusted with a sysctl but typically only needs to be 2150 * on very slow connections. A value no smaller then 5 2151 * should be used, but only reduce this default if you have 2152 * no other choice. 2153 */ 2154 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2155 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2156 #undef USERTT 2157 2158 if (tcp_inflight_debug > 0) { 2159 static int ltime; 2160 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2161 ltime = ticks; 2162 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2163 tp, 2164 bw, 2165 tp->t_rttbest, 2166 tp->t_srtt, 2167 bwnd 2168 ); 2169 } 2170 } 2171 if ((long)bwnd < tcp_inflight_min) 2172 bwnd = tcp_inflight_min; 2173 if (bwnd > tcp_inflight_max) 2174 bwnd = tcp_inflight_max; 2175 if ((long)bwnd < tp->t_maxseg * 2) 2176 bwnd = tp->t_maxseg * 2; 2177 tp->snd_bwnd = bwnd; 2178 } 2179 2180 #ifdef TCP_SIGNATURE 2181 /* 2182 * Callback function invoked by m_apply() to digest TCP segment data 2183 * contained within an mbuf chain. 2184 */ 2185 static int 2186 tcp_signature_apply(void *fstate, void *data, u_int len) 2187 { 2188 2189 MD5Update(fstate, (u_char *)data, len); 2190 return (0); 2191 } 2192 2193 /* 2194 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2195 * 2196 * Parameters: 2197 * m pointer to head of mbuf chain 2198 * off0 offset to TCP header within the mbuf chain 2199 * len length of TCP segment data, excluding options 2200 * optlen length of TCP segment options 2201 * buf pointer to storage for computed MD5 digest 2202 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2203 * 2204 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2205 * When called from tcp_input(), we can be sure that th_sum has been 2206 * zeroed out and verified already. 2207 * 2208 * This function is for IPv4 use only. Calling this function with an 2209 * IPv6 packet in the mbuf chain will yield undefined results. 2210 * 2211 * Return 0 if successful, otherwise return -1. 2212 * 2213 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2214 * search with the destination IP address, and a 'magic SPI' to be 2215 * determined by the application. This is hardcoded elsewhere to 1179 2216 * right now. Another branch of this code exists which uses the SPD to 2217 * specify per-application flows but it is unstable. 2218 */ 2219 int 2220 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2221 u_char *buf, u_int direction) 2222 { 2223 union sockaddr_union dst; 2224 struct ippseudo ippseudo; 2225 MD5_CTX ctx; 2226 int doff; 2227 struct ip *ip; 2228 struct ipovly *ipovly; 2229 struct secasvar *sav; 2230 struct tcphdr *th; 2231 u_short savecsum; 2232 2233 KASSERT(m != NULL, ("NULL mbuf chain")); 2234 KASSERT(buf != NULL, ("NULL signature pointer")); 2235 2236 /* Extract the destination from the IP header in the mbuf. */ 2237 ip = mtod(m, struct ip *); 2238 bzero(&dst, sizeof(union sockaddr_union)); 2239 dst.sa.sa_len = sizeof(struct sockaddr_in); 2240 dst.sa.sa_family = AF_INET; 2241 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2242 ip->ip_src : ip->ip_dst; 2243 2244 /* Look up an SADB entry which matches the address of the peer. */ 2245 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2246 if (sav == NULL) { 2247 printf("%s: SADB lookup failed for %s\n", __func__, 2248 inet_ntoa(dst.sin.sin_addr)); 2249 return (EINVAL); 2250 } 2251 2252 MD5Init(&ctx); 2253 ipovly = (struct ipovly *)ip; 2254 th = (struct tcphdr *)((u_char *)ip + off0); 2255 doff = off0 + sizeof(struct tcphdr) + optlen; 2256 2257 /* 2258 * Step 1: Update MD5 hash with IP pseudo-header. 2259 * 2260 * XXX The ippseudo header MUST be digested in network byte order, 2261 * or else we'll fail the regression test. Assume all fields we've 2262 * been doing arithmetic on have been in host byte order. 2263 * XXX One cannot depend on ipovly->ih_len here. When called from 2264 * tcp_output(), the underlying ip_len member has not yet been set. 2265 */ 2266 ippseudo.ippseudo_src = ipovly->ih_src; 2267 ippseudo.ippseudo_dst = ipovly->ih_dst; 2268 ippseudo.ippseudo_pad = 0; 2269 ippseudo.ippseudo_p = IPPROTO_TCP; 2270 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2271 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2272 2273 /* 2274 * Step 2: Update MD5 hash with TCP header, excluding options. 2275 * The TCP checksum must be set to zero. 2276 */ 2277 savecsum = th->th_sum; 2278 th->th_sum = 0; 2279 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2280 th->th_sum = savecsum; 2281 2282 /* 2283 * Step 3: Update MD5 hash with TCP segment data. 2284 * Use m_apply() to avoid an early m_pullup(). 2285 */ 2286 if (len > 0) 2287 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2288 2289 /* 2290 * Step 4: Update MD5 hash with shared secret. 2291 */ 2292 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2293 MD5Final(buf, &ctx); 2294 2295 key_sa_recordxfer(sav, m); 2296 KEY_FREESAV(&sav); 2297 return (0); 2298 } 2299 #endif /* TCP_SIGNATURE */ 2300 2301 static int 2302 sysctl_drop(SYSCTL_HANDLER_ARGS) 2303 { 2304 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2305 struct sockaddr_storage addrs[2]; 2306 struct inpcb *inp; 2307 struct tcpcb *tp; 2308 struct tcptw *tw; 2309 struct sockaddr_in *fin, *lin; 2310 #ifdef INET6 2311 struct sockaddr_in6 *fin6, *lin6; 2312 struct in6_addr f6, l6; 2313 #endif 2314 int error; 2315 2316 inp = NULL; 2317 fin = lin = NULL; 2318 #ifdef INET6 2319 fin6 = lin6 = NULL; 2320 #endif 2321 error = 0; 2322 2323 if (req->oldptr != NULL || req->oldlen != 0) 2324 return (EINVAL); 2325 if (req->newptr == NULL) 2326 return (EPERM); 2327 if (req->newlen < sizeof(addrs)) 2328 return (ENOMEM); 2329 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2330 if (error) 2331 return (error); 2332 2333 switch (addrs[0].ss_family) { 2334 #ifdef INET6 2335 case AF_INET6: 2336 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2337 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2338 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2339 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2340 return (EINVAL); 2341 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2342 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2343 return (EINVAL); 2344 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2345 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2346 fin = (struct sockaddr_in *)&addrs[0]; 2347 lin = (struct sockaddr_in *)&addrs[1]; 2348 break; 2349 } 2350 error = sa6_embedscope(fin6, ip6_use_defzone); 2351 if (error) 2352 return (error); 2353 error = sa6_embedscope(lin6, ip6_use_defzone); 2354 if (error) 2355 return (error); 2356 break; 2357 #endif 2358 case AF_INET: 2359 fin = (struct sockaddr_in *)&addrs[0]; 2360 lin = (struct sockaddr_in *)&addrs[1]; 2361 if (fin->sin_len != sizeof(struct sockaddr_in) || 2362 lin->sin_len != sizeof(struct sockaddr_in)) 2363 return (EINVAL); 2364 break; 2365 default: 2366 return (EINVAL); 2367 } 2368 INP_INFO_WLOCK(&tcbinfo); 2369 switch (addrs[0].ss_family) { 2370 #ifdef INET6 2371 case AF_INET6: 2372 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2373 &l6, lin6->sin6_port, 0, NULL); 2374 break; 2375 #endif 2376 case AF_INET: 2377 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2378 lin->sin_addr, lin->sin_port, 0, NULL); 2379 break; 2380 } 2381 if (inp != NULL) { 2382 INP_LOCK(inp); 2383 if (inp->inp_vflag & INP_TIMEWAIT) { 2384 /* 2385 * XXXRW: There currently exists a state where an 2386 * inpcb is present, but its timewait state has been 2387 * discarded. For now, don't allow dropping of this 2388 * type of inpcb. 2389 */ 2390 tw = intotw(inp); 2391 if (tw != NULL) 2392 tcp_twclose(tw, 0); 2393 } else if (!(inp->inp_vflag & INP_DROPPED) && 2394 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2395 tp = intotcpcb(inp); 2396 tcp_drop(tp, ECONNABORTED); 2397 } 2398 INP_UNLOCK(inp); 2399 } else 2400 error = ESRCH; 2401 INP_INFO_WUNLOCK(&tcbinfo); 2402 return (error); 2403 } 2404 2405 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2406 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2407 0, sysctl_drop, "", "Drop TCP connection"); 2408