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 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/callout.h> 43 #include <sys/kernel.h> 44 #include <sys/sysctl.h> 45 #include <sys/malloc.h> 46 #include <sys/mbuf.h> 47 #ifdef INET6 48 #include <sys/domain.h> 49 #endif 50 #include <sys/priv.h> 51 #include <sys/proc.h> 52 #include <sys/socket.h> 53 #include <sys/socketvar.h> 54 #include <sys/protosw.h> 55 #include <sys/random.h> 56 57 #include <vm/uma.h> 58 59 #include <net/route.h> 60 #include <net/if.h> 61 62 #include <netinet/in.h> 63 #include <netinet/in_systm.h> 64 #include <netinet/ip.h> 65 #ifdef INET6 66 #include <netinet/ip6.h> 67 #endif 68 #include <netinet/in_pcb.h> 69 #ifdef INET6 70 #include <netinet6/in6_pcb.h> 71 #endif 72 #include <netinet/in_var.h> 73 #include <netinet/ip_var.h> 74 #ifdef INET6 75 #include <netinet6/ip6_var.h> 76 #include <netinet6/scope6_var.h> 77 #include <netinet6/nd6.h> 78 #endif 79 #include <netinet/ip_icmp.h> 80 #include <netinet/tcp.h> 81 #include <netinet/tcp_fsm.h> 82 #include <netinet/tcp_seq.h> 83 #include <netinet/tcp_timer.h> 84 #include <netinet/tcp_var.h> 85 #ifdef INET6 86 #include <netinet6/tcp6_var.h> 87 #endif 88 #include <netinet/tcpip.h> 89 #ifdef TCPDEBUG 90 #include <netinet/tcp_debug.h> 91 #endif 92 #include <netinet6/ip6protosw.h> 93 94 #ifdef IPSEC 95 #include <netinet6/ipsec.h> 96 #ifdef INET6 97 #include <netinet6/ipsec6.h> 98 #endif 99 #include <netkey/key.h> 100 #endif /*IPSEC*/ 101 102 #ifdef FAST_IPSEC 103 #include <netipsec/ipsec.h> 104 #include <netipsec/xform.h> 105 #ifdef INET6 106 #include <netipsec/ipsec6.h> 107 #endif 108 #include <netipsec/key.h> 109 #define IPSEC 110 #endif /*FAST_IPSEC*/ 111 112 #include <machine/in_cksum.h> 113 #include <sys/md5.h> 114 115 #include <security/mac/mac_framework.h> 116 117 int tcp_mssdflt = TCP_MSS; 118 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 119 &tcp_mssdflt, 0, "Default TCP Maximum Segment Size"); 120 121 #ifdef INET6 122 int tcp_v6mssdflt = TCP6_MSS; 123 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 124 CTLFLAG_RW, &tcp_v6mssdflt , 0, 125 "Default TCP Maximum Segment Size for IPv6"); 126 #endif 127 128 /* 129 * Minimum MSS we accept and use. This prevents DoS attacks where 130 * we are forced to a ridiculous low MSS like 20 and send hundreds 131 * of packets instead of one. The effect scales with the available 132 * bandwidth and quickly saturates the CPU and network interface 133 * with packet generation and sending. Set to zero to disable MINMSS 134 * checking. This setting prevents us from sending too small packets. 135 */ 136 int tcp_minmss = TCP_MINMSS; 137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 138 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 139 140 int tcp_do_rfc1323 = 1; 141 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 142 &tcp_do_rfc1323, 0, "Enable rfc1323 (high performance TCP) extensions"); 143 144 static int tcp_tcbhashsize = 0; 145 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 146 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 147 148 static int do_tcpdrain = 1; 149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, 150 &do_tcpdrain, 0, 151 "Enable tcp_drain routine for extra help when low on mbufs"); 152 153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 154 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 155 156 static int icmp_may_rst = 1; 157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, 158 &icmp_may_rst, 0, 159 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 160 161 static int tcp_isn_reseed_interval = 0; 162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 163 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 164 165 static uma_zone_t tcptw_zone; 166 static int maxtcptw; 167 168 static int 169 tcptw_auto_size(void) 170 { 171 int halfrange; 172 173 /* 174 * Max out at half the ephemeral port range so that TIME_WAIT 175 * sockets don't tie up too many ephemeral ports. 176 */ 177 if (ipport_lastauto > ipport_firstauto) 178 halfrange = (ipport_lastauto - ipport_firstauto) / 2; 179 else 180 halfrange = (ipport_firstauto - ipport_lastauto) / 2; 181 /* Protect against goofy port ranges smaller than 32. */ 182 return (imin(imax(halfrange, 32), maxsockets / 5)); 183 } 184 185 static int 186 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 187 { 188 int error, new; 189 190 if (maxtcptw == 0) 191 new = tcptw_auto_size(); 192 else 193 new = maxtcptw; 194 error = sysctl_handle_int(oidp, &new, sizeof(int), req); 195 if (error == 0 && req->newptr) 196 if (new >= 32) { 197 maxtcptw = new; 198 uma_zone_set_max(tcptw_zone, maxtcptw); 199 } 200 return (error); 201 } 202 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 203 &maxtcptw, 0, sysctl_maxtcptw, "IU", 204 "Maximum number of compressed TCP TIME_WAIT entries"); 205 206 static int nolocaltimewait = 0; 207 SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW, 208 &nolocaltimewait, 0, 209 "Do not create compressed TCP TIME_WAIT entries for local connections"); 210 211 /* 212 * TCP bandwidth limiting sysctls. Note that the default lower bound of 213 * 1024 exists only for debugging. A good production default would be 214 * something like 6100. 215 */ 216 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 217 "TCP inflight data limiting"); 218 219 static int tcp_inflight_enable = 1; 220 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 221 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 222 223 static int tcp_inflight_debug = 0; 224 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 225 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 226 227 static int tcp_inflight_rttthresh; 228 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW, 229 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I", 230 "RTT threshold below which inflight will deactivate itself"); 231 232 static int tcp_inflight_min = 6144; 233 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 234 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 235 236 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 237 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 238 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 239 240 static int tcp_inflight_stab = 20; 241 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 242 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 243 244 uma_zone_t sack_hole_zone; 245 246 static struct inpcb *tcp_notify(struct inpcb *, int); 247 static void tcp_isn_tick(void *); 248 249 /* 250 * Target size of TCP PCB hash tables. Must be a power of two. 251 * 252 * Note that this can be overridden by the kernel environment 253 * variable net.inet.tcp.tcbhashsize 254 */ 255 #ifndef TCBHASHSIZE 256 #define TCBHASHSIZE 512 257 #endif 258 259 /* 260 * XXX 261 * Callouts should be moved into struct tcp directly. They are currently 262 * separate because the tcpcb structure is exported to userland for sysctl 263 * parsing purposes, which do not know about callouts. 264 */ 265 struct tcpcb_mem { 266 struct tcpcb tcb; 267 struct tcp_timer tt; 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, tcptw_auto_size()); 289 } 290 291 static int 292 tcp_inpcb_init(void *mem, int size, int flags) 293 { 294 struct inpcb *inp = mem; 295 296 INP_LOCK_INIT(inp, "inp", "tcpinp"); 297 return (0); 298 } 299 300 void 301 tcp_init(void) 302 { 303 int hashsize = TCBHASHSIZE; 304 305 tcp_delacktime = TCPTV_DELACK; 306 tcp_keepinit = TCPTV_KEEP_INIT; 307 tcp_keepidle = TCPTV_KEEP_IDLE; 308 tcp_keepintvl = TCPTV_KEEPINTVL; 309 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 310 tcp_msl = TCPTV_MSL; 311 tcp_rexmit_min = TCPTV_MIN; 312 tcp_rexmit_slop = TCPTV_CPU_VAR; 313 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH; 314 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 315 316 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 317 LIST_INIT(&tcb); 318 tcbinfo.listhead = &tcb; 319 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 320 if (!powerof2(hashsize)) { 321 printf("WARNING: TCB hash size not a power of 2\n"); 322 hashsize = 512; /* safe default */ 323 } 324 tcp_tcbhashsize = hashsize; 325 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 326 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 327 &tcbinfo.porthashmask); 328 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 329 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 330 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 331 #ifdef INET6 332 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 333 #else /* INET6 */ 334 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 335 #endif /* INET6 */ 336 if (max_protohdr < TCP_MINPROTOHDR) 337 max_protohdr = TCP_MINPROTOHDR; 338 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 339 panic("tcp_init"); 340 #undef TCP_MINPROTOHDR 341 /* 342 * These have to be type stable for the benefit of the timers. 343 */ 344 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 345 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 346 uma_zone_set_max(tcpcb_zone, maxsockets); 347 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 348 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 349 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 350 if (maxtcptw == 0) 351 uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 352 else 353 uma_zone_set_max(tcptw_zone, maxtcptw); 354 tcp_timer_init(); 355 syncache_init(); 356 tcp_hc_init(); 357 tcp_reass_init(); 358 ISN_LOCK_INIT(); 359 callout_init(&isn_callout, CALLOUT_MPSAFE); 360 tcp_isn_tick(NULL); 361 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 362 SHUTDOWN_PRI_DEFAULT); 363 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 364 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 365 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 366 EVENTHANDLER_PRI_ANY); 367 } 368 369 void 370 tcp_fini(void *xtp) 371 { 372 373 callout_stop(&isn_callout); 374 } 375 376 /* 377 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 378 * tcp_template used to store this data in mbufs, but we now recopy it out 379 * of the tcpcb each time to conserve mbufs. 380 */ 381 void 382 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 383 { 384 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 385 386 INP_LOCK_ASSERT(inp); 387 388 #ifdef INET6 389 if ((inp->inp_vflag & INP_IPV6) != 0) { 390 struct ip6_hdr *ip6; 391 392 ip6 = (struct ip6_hdr *)ip_ptr; 393 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 394 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 395 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 396 (IPV6_VERSION & IPV6_VERSION_MASK); 397 ip6->ip6_nxt = IPPROTO_TCP; 398 ip6->ip6_plen = sizeof(struct tcphdr); 399 ip6->ip6_src = inp->in6p_laddr; 400 ip6->ip6_dst = inp->in6p_faddr; 401 } else 402 #endif 403 { 404 struct ip *ip; 405 406 ip = (struct ip *)ip_ptr; 407 ip->ip_v = IPVERSION; 408 ip->ip_hl = 5; 409 ip->ip_tos = inp->inp_ip_tos; 410 ip->ip_len = 0; 411 ip->ip_id = 0; 412 ip->ip_off = 0; 413 ip->ip_ttl = inp->inp_ip_ttl; 414 ip->ip_sum = 0; 415 ip->ip_p = IPPROTO_TCP; 416 ip->ip_src = inp->inp_laddr; 417 ip->ip_dst = inp->inp_faddr; 418 } 419 th->th_sport = inp->inp_lport; 420 th->th_dport = inp->inp_fport; 421 th->th_seq = 0; 422 th->th_ack = 0; 423 th->th_x2 = 0; 424 th->th_off = 5; 425 th->th_flags = 0; 426 th->th_win = 0; 427 th->th_urp = 0; 428 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 429 } 430 431 /* 432 * Create template to be used to send tcp packets on a connection. 433 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 434 * use for this function is in keepalives, which use tcp_respond. 435 */ 436 struct tcptemp * 437 tcpip_maketemplate(struct inpcb *inp) 438 { 439 struct mbuf *m; 440 struct tcptemp *n; 441 442 m = m_get(M_DONTWAIT, MT_DATA); 443 if (m == NULL) 444 return (0); 445 m->m_len = sizeof(struct tcptemp); 446 n = mtod(m, struct tcptemp *); 447 448 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 449 return (n); 450 } 451 452 /* 453 * Send a single message to the TCP at address specified by 454 * the given TCP/IP header. If m == NULL, then we make a copy 455 * of the tcpiphdr at ti and send directly to the addressed host. 456 * This is used to force keep alive messages out using the TCP 457 * template for a connection. If flags are given then we send 458 * a message back to the TCP which originated the * segment ti, 459 * and discard the mbuf containing it and any other attached mbufs. 460 * 461 * In any case the ack and sequence number of the transmitted 462 * segment are as specified by the parameters. 463 * 464 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 465 */ 466 void 467 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, 468 struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) 469 { 470 int tlen; 471 int win = 0; 472 struct ip *ip; 473 struct tcphdr *nth; 474 #ifdef INET6 475 struct ip6_hdr *ip6; 476 int isipv6; 477 #endif /* INET6 */ 478 int ipflags = 0; 479 struct inpcb *inp; 480 481 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 482 483 #ifdef INET6 484 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 485 ip6 = ipgen; 486 #endif /* INET6 */ 487 ip = ipgen; 488 489 if (tp != NULL) { 490 inp = tp->t_inpcb; 491 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 492 INP_LOCK_ASSERT(inp); 493 } else 494 inp = NULL; 495 496 if (tp != NULL) { 497 if (!(flags & TH_RST)) { 498 win = sbspace(&inp->inp_socket->so_rcv); 499 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 500 win = (long)TCP_MAXWIN << tp->rcv_scale; 501 } 502 } 503 if (m == NULL) { 504 m = m_gethdr(M_DONTWAIT, MT_DATA); 505 if (m == NULL) 506 return; 507 tlen = 0; 508 m->m_data += max_linkhdr; 509 #ifdef INET6 510 if (isipv6) { 511 bcopy((caddr_t)ip6, mtod(m, caddr_t), 512 sizeof(struct ip6_hdr)); 513 ip6 = mtod(m, struct ip6_hdr *); 514 nth = (struct tcphdr *)(ip6 + 1); 515 } else 516 #endif /* INET6 */ 517 { 518 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 519 ip = mtod(m, struct ip *); 520 nth = (struct tcphdr *)(ip + 1); 521 } 522 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 523 flags = TH_ACK; 524 } else { 525 m_freem(m->m_next); 526 m->m_next = NULL; 527 m->m_data = (caddr_t)ipgen; 528 /* m_len is set later */ 529 tlen = 0; 530 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 531 #ifdef INET6 532 if (isipv6) { 533 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 534 nth = (struct tcphdr *)(ip6 + 1); 535 } else 536 #endif /* INET6 */ 537 { 538 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 539 nth = (struct tcphdr *)(ip + 1); 540 } 541 if (th != nth) { 542 /* 543 * this is usually a case when an extension header 544 * exists between the IPv6 header and the 545 * TCP header. 546 */ 547 nth->th_sport = th->th_sport; 548 nth->th_dport = th->th_dport; 549 } 550 xchg(nth->th_dport, nth->th_sport, n_short); 551 #undef xchg 552 } 553 #ifdef INET6 554 if (isipv6) { 555 ip6->ip6_flow = 0; 556 ip6->ip6_vfc = IPV6_VERSION; 557 ip6->ip6_nxt = IPPROTO_TCP; 558 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 559 tlen)); 560 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 561 } else 562 #endif 563 { 564 tlen += sizeof (struct tcpiphdr); 565 ip->ip_len = tlen; 566 ip->ip_ttl = ip_defttl; 567 if (path_mtu_discovery) 568 ip->ip_off |= IP_DF; 569 } 570 m->m_len = tlen; 571 m->m_pkthdr.len = tlen; 572 m->m_pkthdr.rcvif = NULL; 573 #ifdef MAC 574 if (inp != NULL) { 575 /* 576 * Packet is associated with a socket, so allow the 577 * label of the response to reflect the socket label. 578 */ 579 INP_LOCK_ASSERT(inp); 580 mac_create_mbuf_from_inpcb(inp, m); 581 } else { 582 /* 583 * Packet is not associated with a socket, so possibly 584 * update the label in place. 585 */ 586 mac_reflect_mbuf_tcp(m); 587 } 588 #endif 589 nth->th_seq = htonl(seq); 590 nth->th_ack = htonl(ack); 591 nth->th_x2 = 0; 592 nth->th_off = sizeof (struct tcphdr) >> 2; 593 nth->th_flags = flags; 594 if (tp != NULL) 595 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 596 else 597 nth->th_win = htons((u_short)win); 598 nth->th_urp = 0; 599 #ifdef INET6 600 if (isipv6) { 601 nth->th_sum = 0; 602 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 603 sizeof(struct ip6_hdr), 604 tlen - sizeof(struct ip6_hdr)); 605 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 606 NULL, NULL); 607 } else 608 #endif /* INET6 */ 609 { 610 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 611 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 612 m->m_pkthdr.csum_flags = CSUM_TCP; 613 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 614 } 615 #ifdef TCPDEBUG 616 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 617 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 618 #endif 619 #ifdef INET6 620 if (isipv6) 621 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 622 else 623 #endif /* INET6 */ 624 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 625 } 626 627 /* 628 * Create a new TCP control block, making an 629 * empty reassembly queue and hooking it to the argument 630 * protocol control block. The `inp' parameter must have 631 * come from the zone allocator set up in tcp_init(). 632 */ 633 struct tcpcb * 634 tcp_newtcpcb(struct inpcb *inp) 635 { 636 struct tcpcb_mem *tm; 637 struct tcpcb *tp; 638 #ifdef INET6 639 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 640 #endif /* INET6 */ 641 642 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 643 if (tm == NULL) 644 return (NULL); 645 tp = &tm->tcb; 646 tp->t_timers = &tm->tt; 647 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 648 tp->t_maxseg = tp->t_maxopd = 649 #ifdef INET6 650 isipv6 ? tcp_v6mssdflt : 651 #endif /* INET6 */ 652 tcp_mssdflt; 653 654 /* Set up our timeouts. */ 655 callout_init_mtx(&tp->t_timers->tt_timer, &inp->inp_mtx, 656 CALLOUT_RETURNUNLOCKED); 657 658 if (tcp_do_rfc1323) 659 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 660 tp->sack_enable = tcp_do_sack; 661 TAILQ_INIT(&tp->snd_holes); 662 tp->t_inpcb = inp; /* XXX */ 663 /* 664 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 665 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 666 * reasonable initial retransmit time. 667 */ 668 tp->t_srtt = TCPTV_SRTTBASE; 669 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 670 tp->t_rttmin = tcp_rexmit_min; 671 tp->t_rxtcur = TCPTV_RTOBASE; 672 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 673 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 674 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 675 tp->t_rcvtime = ticks; 676 tp->t_bw_rtttime = ticks; 677 /* 678 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 679 * because the socket may be bound to an IPv6 wildcard address, 680 * which may match an IPv4-mapped IPv6 address. 681 */ 682 inp->inp_ip_ttl = ip_defttl; 683 inp->inp_ppcb = tp; 684 return (tp); /* XXX */ 685 } 686 687 /* 688 * Drop a TCP connection, reporting 689 * the specified error. If connection is synchronized, 690 * then send a RST to peer. 691 */ 692 struct tcpcb * 693 tcp_drop(struct tcpcb *tp, int errno) 694 { 695 struct socket *so = tp->t_inpcb->inp_socket; 696 697 INP_INFO_WLOCK_ASSERT(&tcbinfo); 698 INP_LOCK_ASSERT(tp->t_inpcb); 699 700 if (TCPS_HAVERCVDSYN(tp->t_state)) { 701 tp->t_state = TCPS_CLOSED; 702 (void) tcp_output(tp); 703 tcpstat.tcps_drops++; 704 } else 705 tcpstat.tcps_conndrops++; 706 if (errno == ETIMEDOUT && tp->t_softerror) 707 errno = tp->t_softerror; 708 so->so_error = errno; 709 return (tcp_close(tp)); 710 } 711 712 void 713 tcp_discardcb(struct tcpcb *tp) 714 { 715 struct tseg_qent *q; 716 struct inpcb *inp = tp->t_inpcb; 717 struct socket *so = inp->inp_socket; 718 #ifdef INET6 719 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 720 #endif /* INET6 */ 721 722 INP_LOCK_ASSERT(inp); 723 724 /* 725 * Make sure that all of our timers are stopped before we 726 * delete the PCB. 727 * 728 * XXX: callout_stop() may race and a callout may already 729 * try to obtain the INP_LOCK. Only callout_drain() would 730 * stop this but it would cause a LOR thus we can't use it. 731 * The tcp_timer() function contains a lot of checks to 732 * handle this case rather gracefully. 733 */ 734 tp->t_timers->tt_active = 0; 735 callout_stop(&tp->t_timers->tt_timer); 736 737 /* 738 * If we got enough samples through the srtt filter, 739 * save the rtt and rttvar in the routing entry. 740 * 'Enough' is arbitrarily defined as 4 rtt samples. 741 * 4 samples is enough for the srtt filter to converge 742 * to within enough % of the correct value; fewer samples 743 * and we could save a bogus rtt. The danger is not high 744 * as tcp quickly recovers from everything. 745 * XXX: Works very well but needs some more statistics! 746 */ 747 if (tp->t_rttupdated >= 4) { 748 struct hc_metrics_lite metrics; 749 u_long ssthresh; 750 751 bzero(&metrics, sizeof(metrics)); 752 /* 753 * Update the ssthresh always when the conditions below 754 * are satisfied. This gives us better new start value 755 * for the congestion avoidance for new connections. 756 * ssthresh is only set if packet loss occured on a session. 757 * 758 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 759 * being torn down. Ideally this code would not use 'so'. 760 */ 761 ssthresh = tp->snd_ssthresh; 762 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 763 /* 764 * convert the limit from user data bytes to 765 * packets then to packet data bytes. 766 */ 767 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 768 if (ssthresh < 2) 769 ssthresh = 2; 770 ssthresh *= (u_long)(tp->t_maxseg + 771 #ifdef INET6 772 (isipv6 ? sizeof (struct ip6_hdr) + 773 sizeof (struct tcphdr) : 774 #endif 775 sizeof (struct tcpiphdr) 776 #ifdef INET6 777 ) 778 #endif 779 ); 780 } else 781 ssthresh = 0; 782 metrics.rmx_ssthresh = ssthresh; 783 784 metrics.rmx_rtt = tp->t_srtt; 785 metrics.rmx_rttvar = tp->t_rttvar; 786 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 787 metrics.rmx_bandwidth = tp->snd_bandwidth; 788 metrics.rmx_cwnd = tp->snd_cwnd; 789 metrics.rmx_sendpipe = 0; 790 metrics.rmx_recvpipe = 0; 791 792 tcp_hc_update(&inp->inp_inc, &metrics); 793 } 794 795 /* free the reassembly queue, if any */ 796 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 797 LIST_REMOVE(q, tqe_q); 798 m_freem(q->tqe_m); 799 uma_zfree(tcp_reass_zone, q); 800 tp->t_segqlen--; 801 tcp_reass_qsize--; 802 } 803 tcp_free_sackholes(tp); 804 inp->inp_ppcb = NULL; 805 tp->t_inpcb = NULL; 806 uma_zfree(tcpcb_zone, tp); 807 } 808 809 /* 810 * Attempt to close a TCP control block, marking it as dropped, and freeing 811 * the socket if we hold the only reference. 812 */ 813 struct tcpcb * 814 tcp_close(struct tcpcb *tp) 815 { 816 struct inpcb *inp = tp->t_inpcb; 817 struct socket *so; 818 819 INP_INFO_WLOCK_ASSERT(&tcbinfo); 820 INP_LOCK_ASSERT(inp); 821 822 in_pcbdrop(inp); 823 tcpstat.tcps_closed++; 824 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 825 so = inp->inp_socket; 826 soisdisconnected(so); 827 if (inp->inp_vflag & INP_SOCKREF) { 828 KASSERT(so->so_state & SS_PROTOREF, 829 ("tcp_close: !SS_PROTOREF")); 830 inp->inp_vflag &= ~INP_SOCKREF; 831 INP_UNLOCK(inp); 832 ACCEPT_LOCK(); 833 SOCK_LOCK(so); 834 so->so_state &= ~SS_PROTOREF; 835 sofree(so); 836 return (NULL); 837 } 838 return (tp); 839 } 840 841 void 842 tcp_drain(void) 843 { 844 845 if (do_tcpdrain) { 846 struct inpcb *inpb; 847 struct tcpcb *tcpb; 848 struct tseg_qent *te; 849 850 /* 851 * Walk the tcpbs, if existing, and flush the reassembly queue, 852 * if there is one... 853 * XXX: The "Net/3" implementation doesn't imply that the TCP 854 * reassembly queue should be flushed, but in a situation 855 * where we're really low on mbufs, this is potentially 856 * usefull. 857 */ 858 INP_INFO_RLOCK(&tcbinfo); 859 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 860 if (inpb->inp_vflag & INP_TIMEWAIT) 861 continue; 862 INP_LOCK(inpb); 863 if ((tcpb = intotcpcb(inpb)) != NULL) { 864 while ((te = LIST_FIRST(&tcpb->t_segq)) 865 != NULL) { 866 LIST_REMOVE(te, tqe_q); 867 m_freem(te->tqe_m); 868 uma_zfree(tcp_reass_zone, te); 869 tcpb->t_segqlen--; 870 tcp_reass_qsize--; 871 } 872 tcp_clean_sackreport(tcpb); 873 } 874 INP_UNLOCK(inpb); 875 } 876 INP_INFO_RUNLOCK(&tcbinfo); 877 } 878 } 879 880 /* 881 * Notify a tcp user of an asynchronous error; 882 * store error as soft error, but wake up user 883 * (for now, won't do anything until can select for soft error). 884 * 885 * Do not wake up user since there currently is no mechanism for 886 * reporting soft errors (yet - a kqueue filter may be added). 887 */ 888 static struct inpcb * 889 tcp_notify(struct inpcb *inp, int error) 890 { 891 struct tcpcb *tp; 892 893 INP_INFO_WLOCK_ASSERT(&tcbinfo); 894 INP_LOCK_ASSERT(inp); 895 896 if ((inp->inp_vflag & INP_TIMEWAIT) || 897 (inp->inp_vflag & INP_DROPPED)) 898 return (inp); 899 900 tp = intotcpcb(inp); 901 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 902 903 /* 904 * Ignore some errors if we are hooked up. 905 * If connection hasn't completed, has retransmitted several times, 906 * and receives a second error, give up now. This is better 907 * than waiting a long time to establish a connection that 908 * can never complete. 909 */ 910 if (tp->t_state == TCPS_ESTABLISHED && 911 (error == EHOSTUNREACH || error == ENETUNREACH || 912 error == EHOSTDOWN)) { 913 return (inp); 914 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 915 tp->t_softerror) { 916 tp = tcp_drop(tp, error); 917 if (tp != NULL) 918 return (inp); 919 else 920 return (NULL); 921 } else { 922 tp->t_softerror = error; 923 return (inp); 924 } 925 #if 0 926 wakeup( &so->so_timeo); 927 sorwakeup(so); 928 sowwakeup(so); 929 #endif 930 } 931 932 static int 933 tcp_pcblist(SYSCTL_HANDLER_ARGS) 934 { 935 int error, i, n; 936 struct inpcb *inp, **inp_list; 937 inp_gen_t gencnt; 938 struct xinpgen xig; 939 940 /* 941 * The process of preparing the TCB list is too time-consuming and 942 * resource-intensive to repeat twice on every request. 943 */ 944 if (req->oldptr == NULL) { 945 n = tcbinfo.ipi_count; 946 req->oldidx = 2 * (sizeof xig) 947 + (n + n/8) * sizeof(struct xtcpcb); 948 return (0); 949 } 950 951 if (req->newptr != NULL) 952 return (EPERM); 953 954 /* 955 * OK, now we're committed to doing something. 956 */ 957 INP_INFO_RLOCK(&tcbinfo); 958 gencnt = tcbinfo.ipi_gencnt; 959 n = tcbinfo.ipi_count; 960 INP_INFO_RUNLOCK(&tcbinfo); 961 962 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 963 + n * sizeof(struct xtcpcb)); 964 if (error != 0) 965 return (error); 966 967 xig.xig_len = sizeof xig; 968 xig.xig_count = n; 969 xig.xig_gen = gencnt; 970 xig.xig_sogen = so_gencnt; 971 error = SYSCTL_OUT(req, &xig, sizeof xig); 972 if (error) 973 return (error); 974 975 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 976 if (inp_list == NULL) 977 return (ENOMEM); 978 979 INP_INFO_RLOCK(&tcbinfo); 980 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 981 inp = LIST_NEXT(inp, inp_list)) { 982 INP_LOCK(inp); 983 if (inp->inp_gencnt <= gencnt) { 984 /* 985 * XXX: This use of cr_cansee(), introduced with 986 * TCP state changes, is not quite right, but for 987 * now, better than nothing. 988 */ 989 if (inp->inp_vflag & INP_TIMEWAIT) { 990 if (intotw(inp) != NULL) 991 error = cr_cansee(req->td->td_ucred, 992 intotw(inp)->tw_cred); 993 else 994 error = EINVAL; /* Skip this inp. */ 995 } else 996 error = cr_canseesocket(req->td->td_ucred, 997 inp->inp_socket); 998 if (error == 0) 999 inp_list[i++] = inp; 1000 } 1001 INP_UNLOCK(inp); 1002 } 1003 INP_INFO_RUNLOCK(&tcbinfo); 1004 n = i; 1005 1006 error = 0; 1007 for (i = 0; i < n; i++) { 1008 inp = inp_list[i]; 1009 INP_LOCK(inp); 1010 if (inp->inp_gencnt <= gencnt) { 1011 struct xtcpcb xt; 1012 void *inp_ppcb; 1013 1014 bzero(&xt, sizeof(xt)); 1015 xt.xt_len = sizeof xt; 1016 /* XXX should avoid extra copy */ 1017 bcopy(inp, &xt.xt_inp, sizeof *inp); 1018 inp_ppcb = inp->inp_ppcb; 1019 if (inp_ppcb == NULL) 1020 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1021 else if (inp->inp_vflag & INP_TIMEWAIT) { 1022 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1023 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1024 } else 1025 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1026 if (inp->inp_socket != NULL) 1027 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1028 else { 1029 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1030 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1031 } 1032 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1033 INP_UNLOCK(inp); 1034 error = SYSCTL_OUT(req, &xt, sizeof xt); 1035 } else 1036 INP_UNLOCK(inp); 1037 1038 } 1039 if (!error) { 1040 /* 1041 * Give the user an updated idea of our state. 1042 * If the generation differs from what we told 1043 * her before, she knows that something happened 1044 * while we were processing this request, and it 1045 * might be necessary to retry. 1046 */ 1047 INP_INFO_RLOCK(&tcbinfo); 1048 xig.xig_gen = tcbinfo.ipi_gencnt; 1049 xig.xig_sogen = so_gencnt; 1050 xig.xig_count = tcbinfo.ipi_count; 1051 INP_INFO_RUNLOCK(&tcbinfo); 1052 error = SYSCTL_OUT(req, &xig, sizeof xig); 1053 } 1054 free(inp_list, M_TEMP); 1055 return (error); 1056 } 1057 1058 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1059 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1060 1061 static int 1062 tcp_getcred(SYSCTL_HANDLER_ARGS) 1063 { 1064 struct xucred xuc; 1065 struct sockaddr_in addrs[2]; 1066 struct inpcb *inp; 1067 int error; 1068 1069 error = priv_check_cred(req->td->td_ucred, PRIV_NETINET_GETCRED, 1070 SUSER_ALLOWJAIL); 1071 if (error) 1072 return (error); 1073 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1074 if (error) 1075 return (error); 1076 INP_INFO_RLOCK(&tcbinfo); 1077 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1078 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1079 if (inp == NULL) { 1080 error = ENOENT; 1081 goto outunlocked; 1082 } 1083 INP_LOCK(inp); 1084 if (inp->inp_socket == NULL) { 1085 error = ENOENT; 1086 goto out; 1087 } 1088 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1089 if (error) 1090 goto out; 1091 cru2x(inp->inp_socket->so_cred, &xuc); 1092 out: 1093 INP_UNLOCK(inp); 1094 outunlocked: 1095 INP_INFO_RUNLOCK(&tcbinfo); 1096 if (error == 0) 1097 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1098 return (error); 1099 } 1100 1101 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1102 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1103 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1104 1105 #ifdef INET6 1106 static int 1107 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1108 { 1109 struct xucred xuc; 1110 struct sockaddr_in6 addrs[2]; 1111 struct inpcb *inp; 1112 int error, mapped = 0; 1113 1114 error = priv_check_cred(req->td->td_ucred, PRIV_NETINET_GETCRED, 1115 SUSER_ALLOWJAIL); 1116 if (error) 1117 return (error); 1118 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1119 if (error) 1120 return (error); 1121 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1122 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1123 return (error); 1124 } 1125 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1126 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1127 mapped = 1; 1128 else 1129 return (EINVAL); 1130 } 1131 1132 INP_INFO_RLOCK(&tcbinfo); 1133 if (mapped == 1) 1134 inp = in_pcblookup_hash(&tcbinfo, 1135 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1136 addrs[1].sin6_port, 1137 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1138 addrs[0].sin6_port, 1139 0, NULL); 1140 else 1141 inp = in6_pcblookup_hash(&tcbinfo, 1142 &addrs[1].sin6_addr, addrs[1].sin6_port, 1143 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1144 if (inp == NULL) { 1145 error = ENOENT; 1146 goto outunlocked; 1147 } 1148 INP_LOCK(inp); 1149 if (inp->inp_socket == NULL) { 1150 error = ENOENT; 1151 goto out; 1152 } 1153 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1154 if (error) 1155 goto out; 1156 cru2x(inp->inp_socket->so_cred, &xuc); 1157 out: 1158 INP_UNLOCK(inp); 1159 outunlocked: 1160 INP_INFO_RUNLOCK(&tcbinfo); 1161 if (error == 0) 1162 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1163 return (error); 1164 } 1165 1166 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1167 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1168 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1169 #endif 1170 1171 1172 void 1173 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1174 { 1175 struct ip *ip = vip; 1176 struct tcphdr *th; 1177 struct in_addr faddr; 1178 struct inpcb *inp; 1179 struct tcpcb *tp; 1180 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1181 struct icmp *icp; 1182 struct in_conninfo inc; 1183 tcp_seq icmp_tcp_seq; 1184 int mtu; 1185 1186 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1187 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1188 return; 1189 1190 if (cmd == PRC_MSGSIZE) 1191 notify = tcp_mtudisc; 1192 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1193 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1194 notify = tcp_drop_syn_sent; 1195 /* 1196 * Redirects don't need to be handled up here. 1197 */ 1198 else if (PRC_IS_REDIRECT(cmd)) 1199 return; 1200 /* 1201 * Source quench is depreciated. 1202 */ 1203 else if (cmd == PRC_QUENCH) 1204 return; 1205 /* 1206 * Hostdead is ugly because it goes linearly through all PCBs. 1207 * XXX: We never get this from ICMP, otherwise it makes an 1208 * excellent DoS attack on machines with many connections. 1209 */ 1210 else if (cmd == PRC_HOSTDEAD) 1211 ip = NULL; 1212 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1213 return; 1214 if (ip != NULL) { 1215 icp = (struct icmp *)((caddr_t)ip 1216 - offsetof(struct icmp, icmp_ip)); 1217 th = (struct tcphdr *)((caddr_t)ip 1218 + (ip->ip_hl << 2)); 1219 INP_INFO_WLOCK(&tcbinfo); 1220 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1221 ip->ip_src, th->th_sport, 0, NULL); 1222 if (inp != NULL) { 1223 INP_LOCK(inp); 1224 if (!(inp->inp_vflag & INP_TIMEWAIT) && 1225 !(inp->inp_vflag & INP_DROPPED) && 1226 !(inp->inp_socket == NULL)) { 1227 icmp_tcp_seq = htonl(th->th_seq); 1228 tp = intotcpcb(inp); 1229 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1230 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1231 if (cmd == PRC_MSGSIZE) { 1232 /* 1233 * MTU discovery: 1234 * If we got a needfrag set the MTU 1235 * in the route to the suggested new 1236 * value (if given) and then notify. 1237 */ 1238 bzero(&inc, sizeof(inc)); 1239 inc.inc_flags = 0; /* IPv4 */ 1240 inc.inc_faddr = faddr; 1241 1242 mtu = ntohs(icp->icmp_nextmtu); 1243 /* 1244 * If no alternative MTU was 1245 * proposed, try the next smaller 1246 * one. ip->ip_len has already 1247 * been swapped in icmp_input(). 1248 */ 1249 if (!mtu) 1250 mtu = ip_next_mtu(ip->ip_len, 1251 1); 1252 if (mtu < max(296, (tcp_minmss) 1253 + sizeof(struct tcpiphdr))) 1254 mtu = 0; 1255 if (!mtu) 1256 mtu = tcp_mssdflt 1257 + sizeof(struct tcpiphdr); 1258 /* 1259 * Only cache the the MTU if it 1260 * is smaller than the interface 1261 * or route MTU. tcp_mtudisc() 1262 * will do right thing by itself. 1263 */ 1264 if (mtu <= tcp_maxmtu(&inc, NULL)) 1265 tcp_hc_updatemtu(&inc, mtu); 1266 } 1267 1268 inp = (*notify)(inp, inetctlerrmap[cmd]); 1269 } 1270 } 1271 if (inp != NULL) 1272 INP_UNLOCK(inp); 1273 } else { 1274 inc.inc_fport = th->th_dport; 1275 inc.inc_lport = th->th_sport; 1276 inc.inc_faddr = faddr; 1277 inc.inc_laddr = ip->ip_src; 1278 #ifdef INET6 1279 inc.inc_isipv6 = 0; 1280 #endif 1281 syncache_unreach(&inc, th); 1282 } 1283 INP_INFO_WUNLOCK(&tcbinfo); 1284 } else 1285 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1286 } 1287 1288 #ifdef INET6 1289 void 1290 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1291 { 1292 struct tcphdr th; 1293 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1294 struct ip6_hdr *ip6; 1295 struct mbuf *m; 1296 struct ip6ctlparam *ip6cp = NULL; 1297 const struct sockaddr_in6 *sa6_src = NULL; 1298 int off; 1299 struct tcp_portonly { 1300 u_int16_t th_sport; 1301 u_int16_t th_dport; 1302 } *thp; 1303 1304 if (sa->sa_family != AF_INET6 || 1305 sa->sa_len != sizeof(struct sockaddr_in6)) 1306 return; 1307 1308 if (cmd == PRC_MSGSIZE) 1309 notify = tcp_mtudisc; 1310 else if (!PRC_IS_REDIRECT(cmd) && 1311 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1312 return; 1313 /* Source quench is depreciated. */ 1314 else if (cmd == PRC_QUENCH) 1315 return; 1316 1317 /* if the parameter is from icmp6, decode it. */ 1318 if (d != NULL) { 1319 ip6cp = (struct ip6ctlparam *)d; 1320 m = ip6cp->ip6c_m; 1321 ip6 = ip6cp->ip6c_ip6; 1322 off = ip6cp->ip6c_off; 1323 sa6_src = ip6cp->ip6c_src; 1324 } else { 1325 m = NULL; 1326 ip6 = NULL; 1327 off = 0; /* fool gcc */ 1328 sa6_src = &sa6_any; 1329 } 1330 1331 if (ip6 != NULL) { 1332 struct in_conninfo inc; 1333 /* 1334 * XXX: We assume that when IPV6 is non NULL, 1335 * M and OFF are valid. 1336 */ 1337 1338 /* check if we can safely examine src and dst ports */ 1339 if (m->m_pkthdr.len < off + sizeof(*thp)) 1340 return; 1341 1342 bzero(&th, sizeof(th)); 1343 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1344 1345 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1346 (struct sockaddr *)ip6cp->ip6c_src, 1347 th.th_sport, cmd, NULL, notify); 1348 1349 inc.inc_fport = th.th_dport; 1350 inc.inc_lport = th.th_sport; 1351 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1352 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1353 inc.inc_isipv6 = 1; 1354 INP_INFO_WLOCK(&tcbinfo); 1355 syncache_unreach(&inc, &th); 1356 INP_INFO_WUNLOCK(&tcbinfo); 1357 } else 1358 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1359 0, cmd, NULL, notify); 1360 } 1361 #endif /* INET6 */ 1362 1363 1364 /* 1365 * Following is where TCP initial sequence number generation occurs. 1366 * 1367 * There are two places where we must use initial sequence numbers: 1368 * 1. In SYN-ACK packets. 1369 * 2. In SYN packets. 1370 * 1371 * All ISNs for SYN-ACK packets are generated by the syncache. See 1372 * tcp_syncache.c for details. 1373 * 1374 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1375 * depends on this property. In addition, these ISNs should be 1376 * unguessable so as to prevent connection hijacking. To satisfy 1377 * the requirements of this situation, the algorithm outlined in 1378 * RFC 1948 is used, with only small modifications. 1379 * 1380 * Implementation details: 1381 * 1382 * Time is based off the system timer, and is corrected so that it 1383 * increases by one megabyte per second. This allows for proper 1384 * recycling on high speed LANs while still leaving over an hour 1385 * before rollover. 1386 * 1387 * As reading the *exact* system time is too expensive to be done 1388 * whenever setting up a TCP connection, we increment the time 1389 * offset in two ways. First, a small random positive increment 1390 * is added to isn_offset for each connection that is set up. 1391 * Second, the function tcp_isn_tick fires once per clock tick 1392 * and increments isn_offset as necessary so that sequence numbers 1393 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1394 * random positive increments serve only to ensure that the same 1395 * exact sequence number is never sent out twice (as could otherwise 1396 * happen when a port is recycled in less than the system tick 1397 * interval.) 1398 * 1399 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1400 * between seeding of isn_secret. This is normally set to zero, 1401 * as reseeding should not be necessary. 1402 * 1403 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1404 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1405 * general, this means holding an exclusive (write) lock. 1406 */ 1407 1408 #define ISN_BYTES_PER_SECOND 1048576 1409 #define ISN_STATIC_INCREMENT 4096 1410 #define ISN_RANDOM_INCREMENT (4096 - 1) 1411 1412 static u_char isn_secret[32]; 1413 static int isn_last_reseed; 1414 static u_int32_t isn_offset, isn_offset_old; 1415 static MD5_CTX isn_ctx; 1416 1417 tcp_seq 1418 tcp_new_isn(struct tcpcb *tp) 1419 { 1420 u_int32_t md5_buffer[4]; 1421 tcp_seq new_isn; 1422 1423 INP_LOCK_ASSERT(tp->t_inpcb); 1424 1425 ISN_LOCK(); 1426 /* Seed if this is the first use, reseed if requested. */ 1427 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1428 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1429 < (u_int)ticks))) { 1430 read_random(&isn_secret, sizeof(isn_secret)); 1431 isn_last_reseed = ticks; 1432 } 1433 1434 /* Compute the md5 hash and return the ISN. */ 1435 MD5Init(&isn_ctx); 1436 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1437 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1438 #ifdef INET6 1439 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1440 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1441 sizeof(struct in6_addr)); 1442 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1443 sizeof(struct in6_addr)); 1444 } else 1445 #endif 1446 { 1447 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1448 sizeof(struct in_addr)); 1449 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1450 sizeof(struct in_addr)); 1451 } 1452 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1453 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1454 new_isn = (tcp_seq) md5_buffer[0]; 1455 isn_offset += ISN_STATIC_INCREMENT + 1456 (arc4random() & ISN_RANDOM_INCREMENT); 1457 new_isn += isn_offset; 1458 ISN_UNLOCK(); 1459 return (new_isn); 1460 } 1461 1462 /* 1463 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1464 * to keep time flowing at a relatively constant rate. If the random 1465 * increments have already pushed us past the projected offset, do nothing. 1466 */ 1467 static void 1468 tcp_isn_tick(void *xtp) 1469 { 1470 u_int32_t projected_offset; 1471 1472 ISN_LOCK(); 1473 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1474 1475 if (projected_offset > isn_offset) 1476 isn_offset = projected_offset; 1477 1478 isn_offset_old = isn_offset; 1479 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1480 ISN_UNLOCK(); 1481 } 1482 1483 /* 1484 * When a specific ICMP unreachable message is received and the 1485 * connection state is SYN-SENT, drop the connection. This behavior 1486 * is controlled by the icmp_may_rst sysctl. 1487 */ 1488 struct inpcb * 1489 tcp_drop_syn_sent(struct inpcb *inp, int errno) 1490 { 1491 struct tcpcb *tp; 1492 1493 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1494 INP_LOCK_ASSERT(inp); 1495 1496 if ((inp->inp_vflag & INP_TIMEWAIT) || 1497 (inp->inp_vflag & INP_DROPPED)) 1498 return (inp); 1499 1500 tp = intotcpcb(inp); 1501 if (tp->t_state != TCPS_SYN_SENT) 1502 return (inp); 1503 1504 tp = tcp_drop(tp, errno); 1505 if (tp != NULL) 1506 return (inp); 1507 else 1508 return (NULL); 1509 } 1510 1511 /* 1512 * When `need fragmentation' ICMP is received, update our idea of the MSS 1513 * based on the new value in the route. Also nudge TCP to send something, 1514 * since we know the packet we just sent was dropped. 1515 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1516 */ 1517 struct inpcb * 1518 tcp_mtudisc(struct inpcb *inp, int errno) 1519 { 1520 struct tcpcb *tp; 1521 struct socket *so = inp->inp_socket; 1522 u_int maxmtu; 1523 u_int romtu; 1524 int mss; 1525 #ifdef INET6 1526 int isipv6; 1527 #endif /* INET6 */ 1528 1529 INP_LOCK_ASSERT(inp); 1530 if ((inp->inp_vflag & INP_TIMEWAIT) || 1531 (inp->inp_vflag & INP_DROPPED)) 1532 return (inp); 1533 1534 tp = intotcpcb(inp); 1535 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1536 1537 #ifdef INET6 1538 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1539 #endif 1540 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1541 romtu = 1542 #ifdef INET6 1543 isipv6 ? tcp_maxmtu6(&inp->inp_inc, NULL) : 1544 #endif /* INET6 */ 1545 tcp_maxmtu(&inp->inp_inc, NULL); 1546 if (!maxmtu) 1547 maxmtu = romtu; 1548 else 1549 maxmtu = min(maxmtu, romtu); 1550 if (!maxmtu) { 1551 tp->t_maxopd = tp->t_maxseg = 1552 #ifdef INET6 1553 isipv6 ? tcp_v6mssdflt : 1554 #endif /* INET6 */ 1555 tcp_mssdflt; 1556 return (inp); 1557 } 1558 mss = maxmtu - 1559 #ifdef INET6 1560 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1561 #endif /* INET6 */ 1562 sizeof(struct tcpiphdr) 1563 #ifdef INET6 1564 ) 1565 #endif /* INET6 */ 1566 ; 1567 1568 /* 1569 * XXX - The above conditional probably violates the TCP 1570 * spec. The problem is that, since we don't know the 1571 * other end's MSS, we are supposed to use a conservative 1572 * default. But, if we do that, then MTU discovery will 1573 * never actually take place, because the conservative 1574 * default is much less than the MTUs typically seen 1575 * on the Internet today. For the moment, we'll sweep 1576 * this under the carpet. 1577 * 1578 * The conservative default might not actually be a problem 1579 * if the only case this occurs is when sending an initial 1580 * SYN with options and data to a host we've never talked 1581 * to before. Then, they will reply with an MSS value which 1582 * will get recorded and the new parameters should get 1583 * recomputed. For Further Study. 1584 */ 1585 if (tp->t_maxopd <= mss) 1586 return (inp); 1587 tp->t_maxopd = mss; 1588 1589 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1590 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1591 mss -= TCPOLEN_TSTAMP_APPA; 1592 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1593 if (mss > MCLBYTES) 1594 mss &= ~(MCLBYTES-1); 1595 #else 1596 if (mss > MCLBYTES) 1597 mss = mss / MCLBYTES * MCLBYTES; 1598 #endif 1599 if (so->so_snd.sb_hiwat < mss) 1600 mss = so->so_snd.sb_hiwat; 1601 1602 tp->t_maxseg = mss; 1603 1604 tcpstat.tcps_mturesent++; 1605 tp->t_rtttime = 0; 1606 tp->snd_nxt = tp->snd_una; 1607 tcp_free_sackholes(tp); 1608 tp->snd_recover = tp->snd_max; 1609 if (tp->sack_enable) 1610 EXIT_FASTRECOVERY(tp); 1611 tcp_output(tp); 1612 return (inp); 1613 } 1614 1615 /* 1616 * Look-up the routing entry to the peer of this inpcb. If no route 1617 * is found and it cannot be allocated, then return NULL. This routine 1618 * is called by TCP routines that access the rmx structure and by tcp_mss 1619 * to get the interface MTU. 1620 */ 1621 u_long 1622 tcp_maxmtu(struct in_conninfo *inc, int *flags) 1623 { 1624 struct route sro; 1625 struct sockaddr_in *dst; 1626 struct ifnet *ifp; 1627 u_long maxmtu = 0; 1628 1629 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1630 1631 bzero(&sro, sizeof(sro)); 1632 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1633 dst = (struct sockaddr_in *)&sro.ro_dst; 1634 dst->sin_family = AF_INET; 1635 dst->sin_len = sizeof(*dst); 1636 dst->sin_addr = inc->inc_faddr; 1637 rtalloc_ign(&sro, RTF_CLONING); 1638 } 1639 if (sro.ro_rt != NULL) { 1640 ifp = sro.ro_rt->rt_ifp; 1641 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1642 maxmtu = ifp->if_mtu; 1643 else 1644 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1645 1646 /* Report additional interface capabilities. */ 1647 if (flags != NULL) { 1648 if (ifp->if_capenable & IFCAP_TSO4 && 1649 ifp->if_hwassist & CSUM_TSO) 1650 *flags |= CSUM_TSO; 1651 } 1652 RTFREE(sro.ro_rt); 1653 } 1654 return (maxmtu); 1655 } 1656 1657 #ifdef INET6 1658 u_long 1659 tcp_maxmtu6(struct in_conninfo *inc, int *flags) 1660 { 1661 struct route_in6 sro6; 1662 struct ifnet *ifp; 1663 u_long maxmtu = 0; 1664 1665 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1666 1667 bzero(&sro6, sizeof(sro6)); 1668 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1669 sro6.ro_dst.sin6_family = AF_INET6; 1670 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1671 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1672 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1673 } 1674 if (sro6.ro_rt != NULL) { 1675 ifp = sro6.ro_rt->rt_ifp; 1676 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1677 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1678 else 1679 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1680 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1681 1682 /* Report additional interface capabilities. */ 1683 if (flags != NULL) { 1684 if (ifp->if_capenable & IFCAP_TSO6 && 1685 ifp->if_hwassist & CSUM_TSO) 1686 *flags |= CSUM_TSO; 1687 } 1688 RTFREE(sro6.ro_rt); 1689 } 1690 1691 return (maxmtu); 1692 } 1693 #endif /* INET6 */ 1694 1695 #ifdef IPSEC 1696 /* compute ESP/AH header size for TCP, including outer IP header. */ 1697 size_t 1698 ipsec_hdrsiz_tcp(struct tcpcb *tp) 1699 { 1700 struct inpcb *inp; 1701 struct mbuf *m; 1702 size_t hdrsiz; 1703 struct ip *ip; 1704 #ifdef INET6 1705 struct ip6_hdr *ip6; 1706 #endif 1707 struct tcphdr *th; 1708 1709 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1710 return (0); 1711 MGETHDR(m, M_DONTWAIT, MT_DATA); 1712 if (!m) 1713 return (0); 1714 1715 #ifdef INET6 1716 if ((inp->inp_vflag & INP_IPV6) != 0) { 1717 ip6 = mtod(m, struct ip6_hdr *); 1718 th = (struct tcphdr *)(ip6 + 1); 1719 m->m_pkthdr.len = m->m_len = 1720 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1721 tcpip_fillheaders(inp, ip6, th); 1722 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1723 } else 1724 #endif /* INET6 */ 1725 { 1726 ip = mtod(m, struct ip *); 1727 th = (struct tcphdr *)(ip + 1); 1728 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1729 tcpip_fillheaders(inp, ip, th); 1730 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1731 } 1732 1733 m_free(m); 1734 return (hdrsiz); 1735 } 1736 #endif /*IPSEC*/ 1737 1738 /* 1739 * Move a TCP connection into TIME_WAIT state. 1740 * tcbinfo is locked. 1741 * inp is locked, and is unlocked before returning. 1742 */ 1743 void 1744 tcp_twstart(struct tcpcb *tp) 1745 { 1746 struct tcptw *tw; 1747 struct inpcb *inp = tp->t_inpcb; 1748 int acknow; 1749 struct socket *so; 1750 1751 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1752 INP_LOCK_ASSERT(inp); 1753 1754 if (nolocaltimewait && in_localip(inp->inp_faddr)) { 1755 tp = tcp_close(tp); 1756 if (tp != NULL) 1757 INP_UNLOCK(inp); 1758 return; 1759 } 1760 1761 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1762 if (tw == NULL) { 1763 tw = tcp_timer_2msl_tw(1); 1764 if (tw == NULL) { 1765 tp = tcp_close(tp); 1766 if (tp != NULL) 1767 INP_UNLOCK(inp); 1768 return; 1769 } 1770 } 1771 tw->tw_inpcb = inp; 1772 1773 /* 1774 * Recover last window size sent. 1775 */ 1776 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1777 1778 /* 1779 * Set t_recent if timestamps are used on the connection. 1780 */ 1781 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1782 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1783 tw->t_recent = tp->ts_recent; 1784 else 1785 tw->t_recent = 0; 1786 1787 tw->snd_nxt = tp->snd_nxt; 1788 tw->rcv_nxt = tp->rcv_nxt; 1789 tw->iss = tp->iss; 1790 tw->irs = tp->irs; 1791 tw->t_starttime = tp->t_starttime; 1792 tw->tw_time = 0; 1793 1794 /* XXX 1795 * If this code will 1796 * be used for fin-wait-2 state also, then we may need 1797 * a ts_recent from the last segment. 1798 */ 1799 acknow = tp->t_flags & TF_ACKNOW; 1800 1801 /* 1802 * First, discard tcpcb state, which includes stopping its timers and 1803 * freeing it. tcp_discardcb() used to also release the inpcb, but 1804 * that work is now done in the caller. 1805 * 1806 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 1807 * and might not be needed here any longer. 1808 */ 1809 tcp_discardcb(tp); 1810 so = inp->inp_socket; 1811 soisdisconnected(so); 1812 SOCK_LOCK(so); 1813 tw->tw_cred = crhold(so->so_cred); 1814 tw->tw_so_options = so->so_options; 1815 SOCK_UNLOCK(so); 1816 if (acknow) 1817 tcp_twrespond(tw, TH_ACK); 1818 inp->inp_ppcb = tw; 1819 inp->inp_vflag |= INP_TIMEWAIT; 1820 tcp_timer_2msl_reset(tw, 0); 1821 1822 /* 1823 * If the inpcb owns the sole reference to the socket, then we can 1824 * detach and free the socket as it is not needed in time wait. 1825 */ 1826 if (inp->inp_vflag & INP_SOCKREF) { 1827 KASSERT(so->so_state & SS_PROTOREF, 1828 ("tcp_twstart: !SS_PROTOREF")); 1829 inp->inp_vflag &= ~INP_SOCKREF; 1830 INP_UNLOCK(inp); 1831 ACCEPT_LOCK(); 1832 SOCK_LOCK(so); 1833 so->so_state &= ~SS_PROTOREF; 1834 sofree(so); 1835 } else 1836 INP_UNLOCK(inp); 1837 } 1838 1839 #if 0 1840 /* 1841 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1842 * the actual rate is slightly higher due to the addition of 1843 * random positive increments. 1844 * 1845 * Most other new OSes use semi-randomized ISN values, so we 1846 * do not need to worry about them. 1847 */ 1848 #define MS_ISN_BYTES_PER_SECOND 250000 1849 1850 /* 1851 * Determine if the ISN we will generate has advanced beyond the last 1852 * sequence number used by the previous connection. If so, indicate 1853 * that it is safe to recycle this tw socket by returning 1. 1854 */ 1855 int 1856 tcp_twrecycleable(struct tcptw *tw) 1857 { 1858 tcp_seq new_iss = tw->iss; 1859 tcp_seq new_irs = tw->irs; 1860 1861 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1862 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1863 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1864 1865 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1866 return (1); 1867 else 1868 return (0); 1869 } 1870 #endif 1871 1872 void 1873 tcp_twclose(struct tcptw *tw, int reuse) 1874 { 1875 struct socket *so; 1876 struct inpcb *inp; 1877 1878 /* 1879 * At this point, we are in one of two situations: 1880 * 1881 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 1882 * all state. 1883 * 1884 * (2) We have a socket -- if we own a reference, release it and 1885 * notify the socket layer. 1886 */ 1887 inp = tw->tw_inpcb; 1888 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1889 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 1890 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1891 INP_LOCK_ASSERT(inp); 1892 1893 tw->tw_inpcb = NULL; 1894 tcp_timer_2msl_stop(tw); 1895 inp->inp_ppcb = NULL; 1896 in_pcbdrop(inp); 1897 1898 so = inp->inp_socket; 1899 if (so != NULL) { 1900 /* 1901 * If there's a socket, handle two cases: first, we own a 1902 * strong reference, which we will now release, or we don't 1903 * in which case another reference exists (XXXRW: think 1904 * about this more), and we don't need to take action. 1905 */ 1906 if (inp->inp_vflag & INP_SOCKREF) { 1907 inp->inp_vflag &= ~INP_SOCKREF; 1908 INP_UNLOCK(inp); 1909 ACCEPT_LOCK(); 1910 SOCK_LOCK(so); 1911 KASSERT(so->so_state & SS_PROTOREF, 1912 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1913 so->so_state &= ~SS_PROTOREF; 1914 sofree(so); 1915 } else { 1916 /* 1917 * If we don't own the only reference, the socket and 1918 * inpcb need to be left around to be handled by 1919 * tcp_usr_detach() later. 1920 */ 1921 INP_UNLOCK(inp); 1922 } 1923 } else { 1924 #ifdef INET6 1925 if (inp->inp_vflag & INP_IPV6PROTO) 1926 in6_pcbfree(inp); 1927 else 1928 #endif 1929 in_pcbfree(inp); 1930 } 1931 tcpstat.tcps_closed++; 1932 crfree(tw->tw_cred); 1933 tw->tw_cred = NULL; 1934 if (reuse) 1935 return; 1936 uma_zfree(tcptw_zone, tw); 1937 } 1938 1939 int 1940 tcp_twrespond(struct tcptw *tw, int flags) 1941 { 1942 struct inpcb *inp = tw->tw_inpcb; 1943 struct tcphdr *th; 1944 struct mbuf *m; 1945 struct ip *ip = NULL; 1946 u_int hdrlen, optlen; 1947 int error; 1948 struct tcpopt to; 1949 #ifdef INET6 1950 struct ip6_hdr *ip6 = NULL; 1951 int isipv6 = inp->inp_inc.inc_isipv6; 1952 #endif 1953 1954 INP_LOCK_ASSERT(inp); 1955 1956 m = m_gethdr(M_DONTWAIT, MT_DATA); 1957 if (m == NULL) 1958 return (ENOBUFS); 1959 m->m_data += max_linkhdr; 1960 1961 #ifdef MAC 1962 mac_create_mbuf_from_inpcb(inp, m); 1963 #endif 1964 1965 #ifdef INET6 1966 if (isipv6) { 1967 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1968 ip6 = mtod(m, struct ip6_hdr *); 1969 th = (struct tcphdr *)(ip6 + 1); 1970 tcpip_fillheaders(inp, ip6, th); 1971 } else 1972 #endif 1973 { 1974 hdrlen = sizeof(struct tcpiphdr); 1975 ip = mtod(m, struct ip *); 1976 th = (struct tcphdr *)(ip + 1); 1977 tcpip_fillheaders(inp, ip, th); 1978 } 1979 to.to_flags = 0; 1980 1981 /* 1982 * Send a timestamp and echo-reply if both our side and our peer 1983 * have sent timestamps in our SYN's and this is not a RST. 1984 */ 1985 if (tw->t_recent && flags == TH_ACK) { 1986 to.to_flags |= TOF_TS; 1987 to.to_tsval = ticks; 1988 to.to_tsecr = tw->t_recent; 1989 } 1990 optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 1991 1992 m->m_len = hdrlen + optlen; 1993 m->m_pkthdr.len = m->m_len; 1994 1995 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 1996 1997 th->th_seq = htonl(tw->snd_nxt); 1998 th->th_ack = htonl(tw->rcv_nxt); 1999 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 2000 th->th_flags = flags; 2001 th->th_win = htons(tw->last_win); 2002 2003 #ifdef INET6 2004 if (isipv6) { 2005 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 2006 sizeof(struct tcphdr) + optlen); 2007 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2008 error = ip6_output(m, inp->in6p_outputopts, NULL, 2009 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 2010 } else 2011 #endif 2012 { 2013 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2014 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 2015 m->m_pkthdr.csum_flags = CSUM_TCP; 2016 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2017 ip->ip_len = m->m_pkthdr.len; 2018 if (path_mtu_discovery) 2019 ip->ip_off |= IP_DF; 2020 error = ip_output(m, inp->inp_options, NULL, 2021 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 2022 NULL, inp); 2023 } 2024 if (flags & TH_ACK) 2025 tcpstat.tcps_sndacks++; 2026 else 2027 tcpstat.tcps_sndctrl++; 2028 tcpstat.tcps_sndtotal++; 2029 return (error); 2030 } 2031 2032 /* 2033 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 2034 * 2035 * This code attempts to calculate the bandwidth-delay product as a 2036 * means of determining the optimal window size to maximize bandwidth, 2037 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 2038 * routers. This code also does a fairly good job keeping RTTs in check 2039 * across slow links like modems. We implement an algorithm which is very 2040 * similar (but not meant to be) TCP/Vegas. The code operates on the 2041 * transmitter side of a TCP connection and so only effects the transmit 2042 * side of the connection. 2043 * 2044 * BACKGROUND: TCP makes no provision for the management of buffer space 2045 * at the end points or at the intermediate routers and switches. A TCP 2046 * stream, whether using NewReno or not, will eventually buffer as 2047 * many packets as it is able and the only reason this typically works is 2048 * due to the fairly small default buffers made available for a connection 2049 * (typicaly 16K or 32K). As machines use larger windows and/or window 2050 * scaling it is now fairly easy for even a single TCP connection to blow-out 2051 * all available buffer space not only on the local interface, but on 2052 * intermediate routers and switches as well. NewReno makes a misguided 2053 * attempt to 'solve' this problem by waiting for an actual failure to occur, 2054 * then backing off, then steadily increasing the window again until another 2055 * failure occurs, ad-infinitum. This results in terrible oscillation that 2056 * is only made worse as network loads increase and the idea of intentionally 2057 * blowing out network buffers is, frankly, a terrible way to manage network 2058 * resources. 2059 * 2060 * It is far better to limit the transmit window prior to the failure 2061 * condition being achieved. There are two general ways to do this: First 2062 * you can 'scan' through different transmit window sizes and locate the 2063 * point where the RTT stops increasing, indicating that you have filled the 2064 * pipe, then scan backwards until you note that RTT stops decreasing, then 2065 * repeat ad-infinitum. This method works in principle but has severe 2066 * implementation issues due to RTT variances, timer granularity, and 2067 * instability in the algorithm which can lead to many false positives and 2068 * create oscillations as well as interact badly with other TCP streams 2069 * implementing the same algorithm. 2070 * 2071 * The second method is to limit the window to the bandwidth delay product 2072 * of the link. This is the method we implement. RTT variances and our 2073 * own manipulation of the congestion window, bwnd, can potentially 2074 * destabilize the algorithm. For this reason we have to stabilize the 2075 * elements used to calculate the window. We do this by using the minimum 2076 * observed RTT, the long term average of the observed bandwidth, and 2077 * by adding two segments worth of slop. It isn't perfect but it is able 2078 * to react to changing conditions and gives us a very stable basis on 2079 * which to extend the algorithm. 2080 */ 2081 void 2082 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2083 { 2084 u_long bw; 2085 u_long bwnd; 2086 int save_ticks; 2087 2088 INP_LOCK_ASSERT(tp->t_inpcb); 2089 2090 /* 2091 * If inflight_enable is disabled in the middle of a tcp connection, 2092 * make sure snd_bwnd is effectively disabled. 2093 */ 2094 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2095 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2096 tp->snd_bandwidth = 0; 2097 return; 2098 } 2099 2100 /* 2101 * Figure out the bandwidth. Due to the tick granularity this 2102 * is a very rough number and it MUST be averaged over a fairly 2103 * long period of time. XXX we need to take into account a link 2104 * that is not using all available bandwidth, but for now our 2105 * slop will ramp us up if this case occurs and the bandwidth later 2106 * increases. 2107 * 2108 * Note: if ticks rollover 'bw' may wind up negative. We must 2109 * effectively reset t_bw_rtttime for this case. 2110 */ 2111 save_ticks = ticks; 2112 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2113 return; 2114 2115 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2116 (save_ticks - tp->t_bw_rtttime); 2117 tp->t_bw_rtttime = save_ticks; 2118 tp->t_bw_rtseq = ack_seq; 2119 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2120 return; 2121 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2122 2123 tp->snd_bandwidth = bw; 2124 2125 /* 2126 * Calculate the semi-static bandwidth delay product, plus two maximal 2127 * segments. The additional slop puts us squarely in the sweet 2128 * spot and also handles the bandwidth run-up case and stabilization. 2129 * Without the slop we could be locking ourselves into a lower 2130 * bandwidth. 2131 * 2132 * Situations Handled: 2133 * (1) Prevents over-queueing of packets on LANs, especially on 2134 * high speed LANs, allowing larger TCP buffers to be 2135 * specified, and also does a good job preventing 2136 * over-queueing of packets over choke points like modems 2137 * (at least for the transmit side). 2138 * 2139 * (2) Is able to handle changing network loads (bandwidth 2140 * drops so bwnd drops, bandwidth increases so bwnd 2141 * increases). 2142 * 2143 * (3) Theoretically should stabilize in the face of multiple 2144 * connections implementing the same algorithm (this may need 2145 * a little work). 2146 * 2147 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2148 * be adjusted with a sysctl but typically only needs to be 2149 * on very slow connections. A value no smaller then 5 2150 * should be used, but only reduce this default if you have 2151 * no other choice. 2152 */ 2153 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2154 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2155 #undef USERTT 2156 2157 if (tcp_inflight_debug > 0) { 2158 static int ltime; 2159 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2160 ltime = ticks; 2161 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2162 tp, 2163 bw, 2164 tp->t_rttbest, 2165 tp->t_srtt, 2166 bwnd 2167 ); 2168 } 2169 } 2170 if ((long)bwnd < tcp_inflight_min) 2171 bwnd = tcp_inflight_min; 2172 if (bwnd > tcp_inflight_max) 2173 bwnd = tcp_inflight_max; 2174 if ((long)bwnd < tp->t_maxseg * 2) 2175 bwnd = tp->t_maxseg * 2; 2176 tp->snd_bwnd = bwnd; 2177 } 2178 2179 #ifdef TCP_SIGNATURE 2180 /* 2181 * Callback function invoked by m_apply() to digest TCP segment data 2182 * contained within an mbuf chain. 2183 */ 2184 static int 2185 tcp_signature_apply(void *fstate, void *data, u_int len) 2186 { 2187 2188 MD5Update(fstate, (u_char *)data, len); 2189 return (0); 2190 } 2191 2192 /* 2193 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2194 * 2195 * Parameters: 2196 * m pointer to head of mbuf chain 2197 * off0 offset to TCP header within the mbuf chain 2198 * len length of TCP segment data, excluding options 2199 * optlen length of TCP segment options 2200 * buf pointer to storage for computed MD5 digest 2201 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2202 * 2203 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2204 * When called from tcp_input(), we can be sure that th_sum has been 2205 * zeroed out and verified already. 2206 * 2207 * This function is for IPv4 use only. Calling this function with an 2208 * IPv6 packet in the mbuf chain will yield undefined results. 2209 * 2210 * Return 0 if successful, otherwise return -1. 2211 * 2212 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2213 * search with the destination IP address, and a 'magic SPI' to be 2214 * determined by the application. This is hardcoded elsewhere to 1179 2215 * right now. Another branch of this code exists which uses the SPD to 2216 * specify per-application flows but it is unstable. 2217 */ 2218 int 2219 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2220 u_char *buf, u_int direction) 2221 { 2222 union sockaddr_union dst; 2223 struct ippseudo ippseudo; 2224 MD5_CTX ctx; 2225 int doff; 2226 struct ip *ip; 2227 struct ipovly *ipovly; 2228 struct secasvar *sav; 2229 struct tcphdr *th; 2230 u_short savecsum; 2231 2232 KASSERT(m != NULL, ("NULL mbuf chain")); 2233 KASSERT(buf != NULL, ("NULL signature pointer")); 2234 2235 /* Extract the destination from the IP header in the mbuf. */ 2236 ip = mtod(m, struct ip *); 2237 bzero(&dst, sizeof(union sockaddr_union)); 2238 dst.sa.sa_len = sizeof(struct sockaddr_in); 2239 dst.sa.sa_family = AF_INET; 2240 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2241 ip->ip_src : ip->ip_dst; 2242 2243 /* Look up an SADB entry which matches the address of the peer. */ 2244 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2245 if (sav == NULL) { 2246 printf("%s: SADB lookup failed for %s\n", __func__, 2247 inet_ntoa(dst.sin.sin_addr)); 2248 return (EINVAL); 2249 } 2250 2251 MD5Init(&ctx); 2252 ipovly = (struct ipovly *)ip; 2253 th = (struct tcphdr *)((u_char *)ip + off0); 2254 doff = off0 + sizeof(struct tcphdr) + optlen; 2255 2256 /* 2257 * Step 1: Update MD5 hash with IP pseudo-header. 2258 * 2259 * XXX The ippseudo header MUST be digested in network byte order, 2260 * or else we'll fail the regression test. Assume all fields we've 2261 * been doing arithmetic on have been in host byte order. 2262 * XXX One cannot depend on ipovly->ih_len here. When called from 2263 * tcp_output(), the underlying ip_len member has not yet been set. 2264 */ 2265 ippseudo.ippseudo_src = ipovly->ih_src; 2266 ippseudo.ippseudo_dst = ipovly->ih_dst; 2267 ippseudo.ippseudo_pad = 0; 2268 ippseudo.ippseudo_p = IPPROTO_TCP; 2269 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2270 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2271 2272 /* 2273 * Step 2: Update MD5 hash with TCP header, excluding options. 2274 * The TCP checksum must be set to zero. 2275 */ 2276 savecsum = th->th_sum; 2277 th->th_sum = 0; 2278 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2279 th->th_sum = savecsum; 2280 2281 /* 2282 * Step 3: Update MD5 hash with TCP segment data. 2283 * Use m_apply() to avoid an early m_pullup(). 2284 */ 2285 if (len > 0) 2286 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2287 2288 /* 2289 * Step 4: Update MD5 hash with shared secret. 2290 */ 2291 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2292 MD5Final(buf, &ctx); 2293 2294 key_sa_recordxfer(sav, m); 2295 KEY_FREESAV(&sav); 2296 return (0); 2297 } 2298 #endif /* TCP_SIGNATURE */ 2299 2300 static int 2301 sysctl_drop(SYSCTL_HANDLER_ARGS) 2302 { 2303 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2304 struct sockaddr_storage addrs[2]; 2305 struct inpcb *inp; 2306 struct tcpcb *tp; 2307 struct tcptw *tw; 2308 struct sockaddr_in *fin, *lin; 2309 #ifdef INET6 2310 struct sockaddr_in6 *fin6, *lin6; 2311 struct in6_addr f6, l6; 2312 #endif 2313 int error; 2314 2315 inp = NULL; 2316 fin = lin = NULL; 2317 #ifdef INET6 2318 fin6 = lin6 = NULL; 2319 #endif 2320 error = 0; 2321 2322 if (req->oldptr != NULL || req->oldlen != 0) 2323 return (EINVAL); 2324 if (req->newptr == NULL) 2325 return (EPERM); 2326 if (req->newlen < sizeof(addrs)) 2327 return (ENOMEM); 2328 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2329 if (error) 2330 return (error); 2331 2332 switch (addrs[0].ss_family) { 2333 #ifdef INET6 2334 case AF_INET6: 2335 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2336 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2337 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2338 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2339 return (EINVAL); 2340 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2341 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2342 return (EINVAL); 2343 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2344 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2345 fin = (struct sockaddr_in *)&addrs[0]; 2346 lin = (struct sockaddr_in *)&addrs[1]; 2347 break; 2348 } 2349 error = sa6_embedscope(fin6, ip6_use_defzone); 2350 if (error) 2351 return (error); 2352 error = sa6_embedscope(lin6, ip6_use_defzone); 2353 if (error) 2354 return (error); 2355 break; 2356 #endif 2357 case AF_INET: 2358 fin = (struct sockaddr_in *)&addrs[0]; 2359 lin = (struct sockaddr_in *)&addrs[1]; 2360 if (fin->sin_len != sizeof(struct sockaddr_in) || 2361 lin->sin_len != sizeof(struct sockaddr_in)) 2362 return (EINVAL); 2363 break; 2364 default: 2365 return (EINVAL); 2366 } 2367 INP_INFO_WLOCK(&tcbinfo); 2368 switch (addrs[0].ss_family) { 2369 #ifdef INET6 2370 case AF_INET6: 2371 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2372 &l6, lin6->sin6_port, 0, NULL); 2373 break; 2374 #endif 2375 case AF_INET: 2376 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2377 lin->sin_addr, lin->sin_port, 0, NULL); 2378 break; 2379 } 2380 if (inp != NULL) { 2381 INP_LOCK(inp); 2382 if (inp->inp_vflag & INP_TIMEWAIT) { 2383 /* 2384 * XXXRW: There currently exists a state where an 2385 * inpcb is present, but its timewait state has been 2386 * discarded. For now, don't allow dropping of this 2387 * type of inpcb. 2388 */ 2389 tw = intotw(inp); 2390 if (tw != NULL) 2391 tcp_twclose(tw, 0); 2392 } else if (!(inp->inp_vflag & INP_DROPPED) && 2393 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2394 tp = intotcpcb(inp); 2395 tcp_drop(tp, ECONNABORTED); 2396 } 2397 INP_UNLOCK(inp); 2398 } else 2399 error = ESRCH; 2400 INP_INFO_WUNLOCK(&tcbinfo); 2401 return (error); 2402 } 2403 2404 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2405 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2406 0, sysctl_drop, "", "Drop TCP connection"); 2407