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/malloc.h> 47 #include <sys/mbuf.h> 48 #ifdef INET6 49 #include <sys/domain.h> 50 #endif 51 #include <sys/priv.h> 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 #include <security/mac/mac_framework.h> 117 118 int tcp_mssdflt = TCP_MSS; 119 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 120 &tcp_mssdflt, 0, "Default TCP Maximum Segment Size"); 121 122 #ifdef INET6 123 int tcp_v6mssdflt = TCP6_MSS; 124 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 125 CTLFLAG_RW, &tcp_v6mssdflt , 0, 126 "Default TCP Maximum Segment Size for IPv6"); 127 #endif 128 129 /* 130 * Minimum MSS we accept and use. This prevents DoS attacks where 131 * we are forced to a ridiculous low MSS like 20 and send hundreds 132 * of packets instead of one. The effect scales with the available 133 * bandwidth and quickly saturates the CPU and network interface 134 * with packet generation and sending. Set to zero to disable MINMSS 135 * checking. This setting prevents us from sending too small packets. 136 */ 137 int tcp_minmss = TCP_MINMSS; 138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 139 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 140 141 int tcp_do_rfc1323 = 1; 142 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 143 &tcp_do_rfc1323, 0, "Enable rfc1323 (high performance TCP) extensions"); 144 145 static int tcp_tcbhashsize = 0; 146 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 147 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 148 149 static int do_tcpdrain = 1; 150 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, 151 &do_tcpdrain, 0, 152 "Enable tcp_drain routine for extra help when low on mbufs"); 153 154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 155 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 156 157 static int icmp_may_rst = 1; 158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, 159 &icmp_may_rst, 0, 160 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 161 162 static int tcp_isn_reseed_interval = 0; 163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 164 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 165 166 static uma_zone_t tcptw_zone; 167 static int maxtcptw; 168 169 static int 170 tcptw_auto_size(void) 171 { 172 int halfrange; 173 174 /* 175 * Max out at half the ephemeral port range so that TIME_WAIT 176 * sockets don't tie up too many ephemeral ports. 177 */ 178 if (ipport_lastauto > ipport_firstauto) 179 halfrange = (ipport_lastauto - ipport_firstauto) / 2; 180 else 181 halfrange = (ipport_firstauto - ipport_lastauto) / 2; 182 /* Protect against goofy port ranges smaller than 32. */ 183 return (imin(imax(halfrange, 32), maxsockets / 5)); 184 } 185 186 static int 187 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 188 { 189 int error, new; 190 191 if (maxtcptw == 0) 192 new = tcptw_auto_size(); 193 else 194 new = maxtcptw; 195 error = sysctl_handle_int(oidp, &new, sizeof(int), req); 196 if (error == 0 && req->newptr) 197 if (new >= 32) { 198 maxtcptw = new; 199 uma_zone_set_max(tcptw_zone, maxtcptw); 200 } 201 return (error); 202 } 203 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 204 &maxtcptw, 0, sysctl_maxtcptw, "IU", 205 "Maximum number of compressed TCP TIME_WAIT entries"); 206 207 static int nolocaltimewait = 0; 208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW, 209 &nolocaltimewait, 0, 210 "Do not create compressed TCP TIME_WAIT entries for local connections"); 211 212 /* 213 * TCP bandwidth limiting sysctls. Note that the default lower bound of 214 * 1024 exists only for debugging. A good production default would be 215 * something like 6100. 216 */ 217 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 218 "TCP inflight data limiting"); 219 220 static int tcp_inflight_enable = 1; 221 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 222 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 223 224 static int tcp_inflight_debug = 0; 225 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 226 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 227 228 static int tcp_inflight_rttthresh; 229 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW, 230 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I", 231 "RTT threshold below which inflight will deactivate itself"); 232 233 static int tcp_inflight_min = 6144; 234 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 235 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 236 237 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 238 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 239 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 240 241 static int tcp_inflight_stab = 20; 242 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 243 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 244 245 uma_zone_t sack_hole_zone; 246 247 static struct inpcb *tcp_notify(struct inpcb *, int); 248 static void tcp_isn_tick(void *); 249 250 /* 251 * Target size of TCP PCB hash tables. Must be a power of two. 252 * 253 * Note that this can be overridden by the kernel environment 254 * variable net.inet.tcp.tcbhashsize 255 */ 256 #ifndef TCBHASHSIZE 257 #define TCBHASHSIZE 512 258 #endif 259 260 /* 261 * XXX 262 * Callouts should be moved into struct tcp directly. They are currently 263 * separate because the tcpcb structure is exported to userland for sysctl 264 * parsing purposes, which do not know about callouts. 265 */ 266 struct tcpcb_mem { 267 struct tcpcb tcb; 268 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 269 struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 270 }; 271 272 static uma_zone_t tcpcb_zone; 273 struct callout isn_callout; 274 static struct mtx isn_mtx; 275 276 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 277 #define ISN_LOCK() mtx_lock(&isn_mtx) 278 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 279 280 /* 281 * TCP initialization. 282 */ 283 static void 284 tcp_zone_change(void *tag) 285 { 286 287 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 288 uma_zone_set_max(tcpcb_zone, maxsockets); 289 if (maxtcptw == 0) 290 uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 291 } 292 293 static int 294 tcp_inpcb_init(void *mem, int size, int flags) 295 { 296 struct inpcb *inp = mem; 297 298 INP_LOCK_INIT(inp, "inp", "tcpinp"); 299 return (0); 300 } 301 302 void 303 tcp_init(void) 304 { 305 int hashsize = TCBHASHSIZE; 306 307 tcp_delacktime = TCPTV_DELACK; 308 tcp_keepinit = TCPTV_KEEP_INIT; 309 tcp_keepidle = TCPTV_KEEP_IDLE; 310 tcp_keepintvl = TCPTV_KEEPINTVL; 311 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 312 tcp_msl = TCPTV_MSL; 313 tcp_rexmit_min = TCPTV_MIN; 314 tcp_rexmit_slop = TCPTV_CPU_VAR; 315 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH; 316 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 317 318 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 319 LIST_INIT(&tcb); 320 tcbinfo.listhead = &tcb; 321 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 322 if (!powerof2(hashsize)) { 323 printf("WARNING: TCB hash size not a power of 2\n"); 324 hashsize = 512; /* safe default */ 325 } 326 tcp_tcbhashsize = hashsize; 327 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 328 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 329 &tcbinfo.porthashmask); 330 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 331 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 332 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 333 #ifdef INET6 334 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 335 #else /* INET6 */ 336 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 337 #endif /* INET6 */ 338 if (max_protohdr < TCP_MINPROTOHDR) 339 max_protohdr = TCP_MINPROTOHDR; 340 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 341 panic("tcp_init"); 342 #undef TCP_MINPROTOHDR 343 /* 344 * These have to be type stable for the benefit of the timers. 345 */ 346 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 347 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 348 uma_zone_set_max(tcpcb_zone, maxsockets); 349 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 350 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 351 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 352 if (maxtcptw == 0) 353 uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 354 else 355 uma_zone_set_max(tcptw_zone, maxtcptw); 356 tcp_timer_init(); 357 syncache_init(); 358 tcp_hc_init(); 359 tcp_reass_init(); 360 ISN_LOCK_INIT(); 361 callout_init(&isn_callout, CALLOUT_MPSAFE); 362 tcp_isn_tick(NULL); 363 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 364 SHUTDOWN_PRI_DEFAULT); 365 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 366 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 367 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 368 EVENTHANDLER_PRI_ANY); 369 } 370 371 void 372 tcp_fini(void *xtp) 373 { 374 375 callout_stop(&isn_callout); 376 } 377 378 /* 379 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 380 * tcp_template used to store this data in mbufs, but we now recopy it out 381 * of the tcpcb each time to conserve mbufs. 382 */ 383 void 384 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 385 { 386 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 387 388 INP_LOCK_ASSERT(inp); 389 390 #ifdef INET6 391 if ((inp->inp_vflag & INP_IPV6) != 0) { 392 struct ip6_hdr *ip6; 393 394 ip6 = (struct ip6_hdr *)ip_ptr; 395 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 396 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 397 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 398 (IPV6_VERSION & IPV6_VERSION_MASK); 399 ip6->ip6_nxt = IPPROTO_TCP; 400 ip6->ip6_plen = sizeof(struct tcphdr); 401 ip6->ip6_src = inp->in6p_laddr; 402 ip6->ip6_dst = inp->in6p_faddr; 403 } else 404 #endif 405 { 406 struct ip *ip; 407 408 ip = (struct ip *)ip_ptr; 409 ip->ip_v = IPVERSION; 410 ip->ip_hl = 5; 411 ip->ip_tos = inp->inp_ip_tos; 412 ip->ip_len = 0; 413 ip->ip_id = 0; 414 ip->ip_off = 0; 415 ip->ip_ttl = inp->inp_ip_ttl; 416 ip->ip_sum = 0; 417 ip->ip_p = IPPROTO_TCP; 418 ip->ip_src = inp->inp_laddr; 419 ip->ip_dst = inp->inp_faddr; 420 } 421 th->th_sport = inp->inp_lport; 422 th->th_dport = inp->inp_fport; 423 th->th_seq = 0; 424 th->th_ack = 0; 425 th->th_x2 = 0; 426 th->th_off = 5; 427 th->th_flags = 0; 428 th->th_win = 0; 429 th->th_urp = 0; 430 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 431 } 432 433 /* 434 * Create template to be used to send tcp packets on a connection. 435 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 436 * use for this function is in keepalives, which use tcp_respond. 437 */ 438 struct tcptemp * 439 tcpip_maketemplate(struct inpcb *inp) 440 { 441 struct mbuf *m; 442 struct tcptemp *n; 443 444 m = m_get(M_DONTWAIT, MT_DATA); 445 if (m == NULL) 446 return (0); 447 m->m_len = sizeof(struct tcptemp); 448 n = mtod(m, struct tcptemp *); 449 450 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 451 return (n); 452 } 453 454 /* 455 * Send a single message to the TCP at address specified by 456 * the given TCP/IP header. If m == NULL, then we make a copy 457 * of the tcpiphdr at ti and send directly to the addressed host. 458 * This is used to force keep alive messages out using the TCP 459 * template for a connection. If flags are given then we send 460 * a message back to the TCP which originated the * segment ti, 461 * and discard the mbuf containing it and any other attached mbufs. 462 * 463 * In any case the ack and sequence number of the transmitted 464 * segment are as specified by the parameters. 465 * 466 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 467 */ 468 void 469 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, 470 struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) 471 { 472 int tlen; 473 int win = 0; 474 struct ip *ip; 475 struct tcphdr *nth; 476 #ifdef INET6 477 struct ip6_hdr *ip6; 478 int isipv6; 479 #endif /* INET6 */ 480 int ipflags = 0; 481 struct inpcb *inp; 482 483 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 484 485 #ifdef INET6 486 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 487 ip6 = ipgen; 488 #endif /* INET6 */ 489 ip = ipgen; 490 491 if (tp != NULL) { 492 inp = tp->t_inpcb; 493 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 494 INP_INFO_WLOCK_ASSERT(&tcbinfo); 495 INP_LOCK_ASSERT(inp); 496 } else 497 inp = NULL; 498 499 if (tp != NULL) { 500 if (!(flags & TH_RST)) { 501 win = sbspace(&inp->inp_socket->so_rcv); 502 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 503 win = (long)TCP_MAXWIN << tp->rcv_scale; 504 } 505 } 506 if (m == NULL) { 507 m = m_gethdr(M_DONTWAIT, MT_DATA); 508 if (m == NULL) 509 return; 510 tlen = 0; 511 m->m_data += max_linkhdr; 512 #ifdef INET6 513 if (isipv6) { 514 bcopy((caddr_t)ip6, mtod(m, caddr_t), 515 sizeof(struct ip6_hdr)); 516 ip6 = mtod(m, struct ip6_hdr *); 517 nth = (struct tcphdr *)(ip6 + 1); 518 } else 519 #endif /* INET6 */ 520 { 521 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 522 ip = mtod(m, struct ip *); 523 nth = (struct tcphdr *)(ip + 1); 524 } 525 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 526 flags = TH_ACK; 527 } else { 528 m_freem(m->m_next); 529 m->m_next = NULL; 530 m->m_data = (caddr_t)ipgen; 531 /* m_len is set later */ 532 tlen = 0; 533 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 534 #ifdef INET6 535 if (isipv6) { 536 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 537 nth = (struct tcphdr *)(ip6 + 1); 538 } else 539 #endif /* INET6 */ 540 { 541 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 542 nth = (struct tcphdr *)(ip + 1); 543 } 544 if (th != nth) { 545 /* 546 * this is usually a case when an extension header 547 * exists between the IPv6 header and the 548 * TCP header. 549 */ 550 nth->th_sport = th->th_sport; 551 nth->th_dport = th->th_dport; 552 } 553 xchg(nth->th_dport, nth->th_sport, n_short); 554 #undef xchg 555 } 556 #ifdef INET6 557 if (isipv6) { 558 ip6->ip6_flow = 0; 559 ip6->ip6_vfc = IPV6_VERSION; 560 ip6->ip6_nxt = IPPROTO_TCP; 561 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 562 tlen)); 563 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 564 } else 565 #endif 566 { 567 tlen += sizeof (struct tcpiphdr); 568 ip->ip_len = tlen; 569 ip->ip_ttl = ip_defttl; 570 if (path_mtu_discovery) 571 ip->ip_off |= IP_DF; 572 } 573 m->m_len = tlen; 574 m->m_pkthdr.len = tlen; 575 m->m_pkthdr.rcvif = NULL; 576 #ifdef MAC 577 if (inp != NULL) { 578 /* 579 * Packet is associated with a socket, so allow the 580 * label of the response to reflect the socket label. 581 */ 582 INP_LOCK_ASSERT(inp); 583 mac_create_mbuf_from_inpcb(inp, m); 584 } else { 585 /* 586 * Packet is not associated with a socket, so possibly 587 * update the label in place. 588 */ 589 mac_reflect_mbuf_tcp(m); 590 } 591 #endif 592 nth->th_seq = htonl(seq); 593 nth->th_ack = htonl(ack); 594 nth->th_x2 = 0; 595 nth->th_off = sizeof (struct tcphdr) >> 2; 596 nth->th_flags = flags; 597 if (tp != NULL) 598 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 599 else 600 nth->th_win = htons((u_short)win); 601 nth->th_urp = 0; 602 #ifdef INET6 603 if (isipv6) { 604 nth->th_sum = 0; 605 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 606 sizeof(struct ip6_hdr), 607 tlen - sizeof(struct ip6_hdr)); 608 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 609 NULL, NULL); 610 } else 611 #endif /* INET6 */ 612 { 613 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 614 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 615 m->m_pkthdr.csum_flags = CSUM_TCP; 616 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 617 } 618 #ifdef TCPDEBUG 619 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 620 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 621 #endif 622 #ifdef INET6 623 if (isipv6) 624 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 625 else 626 #endif /* INET6 */ 627 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 628 } 629 630 /* 631 * Create a new TCP control block, making an 632 * empty reassembly queue and hooking it to the argument 633 * protocol control block. The `inp' parameter must have 634 * come from the zone allocator set up in tcp_init(). 635 */ 636 struct tcpcb * 637 tcp_newtcpcb(struct inpcb *inp) 638 { 639 struct tcpcb_mem *tm; 640 struct tcpcb *tp; 641 #ifdef INET6 642 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 643 #endif /* INET6 */ 644 645 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 646 if (tm == NULL) 647 return (NULL); 648 tp = &tm->tcb; 649 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 650 tp->t_maxseg = tp->t_maxopd = 651 #ifdef INET6 652 isipv6 ? tcp_v6mssdflt : 653 #endif /* INET6 */ 654 tcp_mssdflt; 655 656 /* Set up our timeouts. */ 657 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE); 658 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE); 659 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE); 660 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE); 661 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE); 662 663 if (tcp_do_rfc1323) 664 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 665 tp->sack_enable = tcp_do_sack; 666 TAILQ_INIT(&tp->snd_holes); 667 tp->t_inpcb = inp; /* XXX */ 668 /* 669 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 670 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 671 * reasonable initial retransmit time. 672 */ 673 tp->t_srtt = TCPTV_SRTTBASE; 674 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 675 tp->t_rttmin = tcp_rexmit_min; 676 tp->t_rxtcur = TCPTV_RTOBASE; 677 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 678 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 679 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 680 tp->t_rcvtime = ticks; 681 tp->t_bw_rtttime = ticks; 682 /* 683 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 684 * because the socket may be bound to an IPv6 wildcard address, 685 * which may match an IPv4-mapped IPv6 address. 686 */ 687 inp->inp_ip_ttl = ip_defttl; 688 inp->inp_ppcb = tp; 689 return (tp); /* XXX */ 690 } 691 692 /* 693 * Drop a TCP connection, reporting 694 * the specified error. If connection is synchronized, 695 * then send a RST to peer. 696 */ 697 struct tcpcb * 698 tcp_drop(struct tcpcb *tp, int errno) 699 { 700 struct socket *so = tp->t_inpcb->inp_socket; 701 702 INP_INFO_WLOCK_ASSERT(&tcbinfo); 703 INP_LOCK_ASSERT(tp->t_inpcb); 704 705 if (TCPS_HAVERCVDSYN(tp->t_state)) { 706 tp->t_state = TCPS_CLOSED; 707 (void) tcp_output(tp); 708 tcpstat.tcps_drops++; 709 } else 710 tcpstat.tcps_conndrops++; 711 if (errno == ETIMEDOUT && tp->t_softerror) 712 errno = tp->t_softerror; 713 so->so_error = errno; 714 return (tcp_close(tp)); 715 } 716 717 void 718 tcp_discardcb(struct tcpcb *tp) 719 { 720 struct tseg_qent *q; 721 struct inpcb *inp = tp->t_inpcb; 722 struct socket *so = inp->inp_socket; 723 #ifdef INET6 724 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 725 #endif /* INET6 */ 726 727 INP_LOCK_ASSERT(inp); 728 729 /* 730 * Make sure that all of our timers are stopped before we 731 * delete the PCB. 732 */ 733 callout_stop(tp->tt_rexmt); 734 callout_stop(tp->tt_persist); 735 callout_stop(tp->tt_keep); 736 callout_stop(tp->tt_2msl); 737 callout_stop(tp->tt_delack); 738 739 /* 740 * If we got enough samples through the srtt filter, 741 * save the rtt and rttvar in the routing entry. 742 * 'Enough' is arbitrarily defined as 4 rtt samples. 743 * 4 samples is enough for the srtt filter to converge 744 * to within enough % of the correct value; fewer samples 745 * and we could save a bogus rtt. The danger is not high 746 * as tcp quickly recovers from everything. 747 * XXX: Works very well but needs some more statistics! 748 */ 749 if (tp->t_rttupdated >= 4) { 750 struct hc_metrics_lite metrics; 751 u_long ssthresh; 752 753 bzero(&metrics, sizeof(metrics)); 754 /* 755 * Update the ssthresh always when the conditions below 756 * are satisfied. This gives us better new start value 757 * for the congestion avoidance for new connections. 758 * ssthresh is only set if packet loss occured on a session. 759 * 760 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 761 * being torn down. Ideally this code would not use 'so'. 762 */ 763 ssthresh = tp->snd_ssthresh; 764 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 765 /* 766 * convert the limit from user data bytes to 767 * packets then to packet data bytes. 768 */ 769 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 770 if (ssthresh < 2) 771 ssthresh = 2; 772 ssthresh *= (u_long)(tp->t_maxseg + 773 #ifdef INET6 774 (isipv6 ? sizeof (struct ip6_hdr) + 775 sizeof (struct tcphdr) : 776 #endif 777 sizeof (struct tcpiphdr) 778 #ifdef INET6 779 ) 780 #endif 781 ); 782 } else 783 ssthresh = 0; 784 metrics.rmx_ssthresh = ssthresh; 785 786 metrics.rmx_rtt = tp->t_srtt; 787 metrics.rmx_rttvar = tp->t_rttvar; 788 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 789 metrics.rmx_bandwidth = tp->snd_bandwidth; 790 metrics.rmx_cwnd = tp->snd_cwnd; 791 metrics.rmx_sendpipe = 0; 792 metrics.rmx_recvpipe = 0; 793 794 tcp_hc_update(&inp->inp_inc, &metrics); 795 } 796 797 /* free the reassembly queue, if any */ 798 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 799 LIST_REMOVE(q, tqe_q); 800 m_freem(q->tqe_m); 801 uma_zfree(tcp_reass_zone, q); 802 tp->t_segqlen--; 803 tcp_reass_qsize--; 804 } 805 tcp_free_sackholes(tp); 806 inp->inp_ppcb = NULL; 807 tp->t_inpcb = NULL; 808 uma_zfree(tcpcb_zone, tp); 809 } 810 811 /* 812 * Attempt to close a TCP control block, marking it as dropped, and freeing 813 * the socket if we hold the only reference. 814 */ 815 struct tcpcb * 816 tcp_close(struct tcpcb *tp) 817 { 818 struct inpcb *inp = tp->t_inpcb; 819 struct socket *so; 820 821 INP_INFO_WLOCK_ASSERT(&tcbinfo); 822 INP_LOCK_ASSERT(inp); 823 824 in_pcbdrop(inp); 825 tcpstat.tcps_closed++; 826 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 827 so = inp->inp_socket; 828 soisdisconnected(so); 829 if (inp->inp_vflag & INP_SOCKREF) { 830 KASSERT(so->so_state & SS_PROTOREF, 831 ("tcp_close: !SS_PROTOREF")); 832 inp->inp_vflag &= ~INP_SOCKREF; 833 INP_UNLOCK(inp); 834 ACCEPT_LOCK(); 835 SOCK_LOCK(so); 836 so->so_state &= ~SS_PROTOREF; 837 sofree(so); 838 return (NULL); 839 } 840 return (tp); 841 } 842 843 void 844 tcp_drain(void) 845 { 846 847 if (do_tcpdrain) { 848 struct inpcb *inpb; 849 struct tcpcb *tcpb; 850 struct tseg_qent *te; 851 852 /* 853 * Walk the tcpbs, if existing, and flush the reassembly queue, 854 * if there is one... 855 * XXX: The "Net/3" implementation doesn't imply that the TCP 856 * reassembly queue should be flushed, but in a situation 857 * where we're really low on mbufs, this is potentially 858 * usefull. 859 */ 860 INP_INFO_RLOCK(&tcbinfo); 861 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 862 if (inpb->inp_vflag & INP_TIMEWAIT) 863 continue; 864 INP_LOCK(inpb); 865 if ((tcpb = intotcpcb(inpb)) != NULL) { 866 while ((te = LIST_FIRST(&tcpb->t_segq)) 867 != NULL) { 868 LIST_REMOVE(te, tqe_q); 869 m_freem(te->tqe_m); 870 uma_zfree(tcp_reass_zone, te); 871 tcpb->t_segqlen--; 872 tcp_reass_qsize--; 873 } 874 tcp_clean_sackreport(tcpb); 875 } 876 INP_UNLOCK(inpb); 877 } 878 INP_INFO_RUNLOCK(&tcbinfo); 879 } 880 } 881 882 /* 883 * Notify a tcp user of an asynchronous error; 884 * store error as soft error, but wake up user 885 * (for now, won't do anything until can select for soft error). 886 * 887 * Do not wake up user since there currently is no mechanism for 888 * reporting soft errors (yet - a kqueue filter may be added). 889 */ 890 static struct inpcb * 891 tcp_notify(struct inpcb *inp, int error) 892 { 893 struct tcpcb *tp; 894 895 INP_INFO_WLOCK_ASSERT(&tcbinfo); 896 INP_LOCK_ASSERT(inp); 897 898 if ((inp->inp_vflag & INP_TIMEWAIT) || 899 (inp->inp_vflag & INP_DROPPED)) 900 return (inp); 901 902 tp = intotcpcb(inp); 903 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 904 905 /* 906 * Ignore some errors if we are hooked up. 907 * If connection hasn't completed, has retransmitted several times, 908 * and receives a second error, give up now. This is better 909 * than waiting a long time to establish a connection that 910 * can never complete. 911 */ 912 if (tp->t_state == TCPS_ESTABLISHED && 913 (error == EHOSTUNREACH || error == ENETUNREACH || 914 error == EHOSTDOWN)) { 915 return (inp); 916 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 917 tp->t_softerror) { 918 tp = tcp_drop(tp, error); 919 if (tp != NULL) 920 return (inp); 921 else 922 return (NULL); 923 } else { 924 tp->t_softerror = error; 925 return (inp); 926 } 927 #if 0 928 wakeup( &so->so_timeo); 929 sorwakeup(so); 930 sowwakeup(so); 931 #endif 932 } 933 934 static int 935 tcp_pcblist(SYSCTL_HANDLER_ARGS) 936 { 937 int error, i, n; 938 struct inpcb *inp, **inp_list; 939 inp_gen_t gencnt; 940 struct xinpgen xig; 941 942 /* 943 * The process of preparing the TCB list is too time-consuming and 944 * resource-intensive to repeat twice on every request. 945 */ 946 if (req->oldptr == NULL) { 947 n = tcbinfo.ipi_count; 948 req->oldidx = 2 * (sizeof xig) 949 + (n + n/8) * sizeof(struct xtcpcb); 950 return (0); 951 } 952 953 if (req->newptr != NULL) 954 return (EPERM); 955 956 /* 957 * OK, now we're committed to doing something. 958 */ 959 INP_INFO_RLOCK(&tcbinfo); 960 gencnt = tcbinfo.ipi_gencnt; 961 n = tcbinfo.ipi_count; 962 INP_INFO_RUNLOCK(&tcbinfo); 963 964 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 965 + n * sizeof(struct xtcpcb)); 966 if (error != 0) 967 return (error); 968 969 xig.xig_len = sizeof xig; 970 xig.xig_count = n; 971 xig.xig_gen = gencnt; 972 xig.xig_sogen = so_gencnt; 973 error = SYSCTL_OUT(req, &xig, sizeof xig); 974 if (error) 975 return (error); 976 977 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 978 if (inp_list == NULL) 979 return (ENOMEM); 980 981 INP_INFO_RLOCK(&tcbinfo); 982 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 983 inp = LIST_NEXT(inp, inp_list)) { 984 INP_LOCK(inp); 985 if (inp->inp_gencnt <= gencnt) { 986 /* 987 * XXX: This use of cr_cansee(), introduced with 988 * TCP state changes, is not quite right, but for 989 * now, better than nothing. 990 */ 991 if (inp->inp_vflag & INP_TIMEWAIT) { 992 if (intotw(inp) != NULL) 993 error = cr_cansee(req->td->td_ucred, 994 intotw(inp)->tw_cred); 995 else 996 error = EINVAL; /* Skip this inp. */ 997 } else 998 error = cr_canseesocket(req->td->td_ucred, 999 inp->inp_socket); 1000 if (error == 0) 1001 inp_list[i++] = inp; 1002 } 1003 INP_UNLOCK(inp); 1004 } 1005 INP_INFO_RUNLOCK(&tcbinfo); 1006 n = i; 1007 1008 error = 0; 1009 for (i = 0; i < n; i++) { 1010 inp = inp_list[i]; 1011 INP_LOCK(inp); 1012 if (inp->inp_gencnt <= gencnt) { 1013 struct xtcpcb xt; 1014 void *inp_ppcb; 1015 1016 bzero(&xt, sizeof(xt)); 1017 xt.xt_len = sizeof xt; 1018 /* XXX should avoid extra copy */ 1019 bcopy(inp, &xt.xt_inp, sizeof *inp); 1020 inp_ppcb = inp->inp_ppcb; 1021 if (inp_ppcb == NULL) 1022 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1023 else if (inp->inp_vflag & INP_TIMEWAIT) { 1024 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1025 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1026 } else 1027 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1028 if (inp->inp_socket != NULL) 1029 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1030 else { 1031 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1032 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1033 } 1034 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1035 INP_UNLOCK(inp); 1036 error = SYSCTL_OUT(req, &xt, sizeof xt); 1037 } else 1038 INP_UNLOCK(inp); 1039 1040 } 1041 if (!error) { 1042 /* 1043 * Give the user an updated idea of our state. 1044 * If the generation differs from what we told 1045 * her before, she knows that something happened 1046 * while we were processing this request, and it 1047 * might be necessary to retry. 1048 */ 1049 INP_INFO_RLOCK(&tcbinfo); 1050 xig.xig_gen = tcbinfo.ipi_gencnt; 1051 xig.xig_sogen = so_gencnt; 1052 xig.xig_count = tcbinfo.ipi_count; 1053 INP_INFO_RUNLOCK(&tcbinfo); 1054 error = SYSCTL_OUT(req, &xig, sizeof xig); 1055 } 1056 free(inp_list, M_TEMP); 1057 return (error); 1058 } 1059 1060 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1061 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1062 1063 static int 1064 tcp_getcred(SYSCTL_HANDLER_ARGS) 1065 { 1066 struct xucred xuc; 1067 struct sockaddr_in addrs[2]; 1068 struct inpcb *inp; 1069 int error; 1070 1071 error = priv_check_cred(req->td->td_ucred, PRIV_NETINET_GETCRED, 1072 SUSER_ALLOWJAIL); 1073 if (error) 1074 return (error); 1075 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1076 if (error) 1077 return (error); 1078 INP_INFO_RLOCK(&tcbinfo); 1079 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1080 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1081 if (inp == NULL) { 1082 error = ENOENT; 1083 goto outunlocked; 1084 } 1085 INP_LOCK(inp); 1086 if (inp->inp_socket == NULL) { 1087 error = ENOENT; 1088 goto out; 1089 } 1090 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1091 if (error) 1092 goto out; 1093 cru2x(inp->inp_socket->so_cred, &xuc); 1094 out: 1095 INP_UNLOCK(inp); 1096 outunlocked: 1097 INP_INFO_RUNLOCK(&tcbinfo); 1098 if (error == 0) 1099 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1100 return (error); 1101 } 1102 1103 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1104 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1105 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1106 1107 #ifdef INET6 1108 static int 1109 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1110 { 1111 struct xucred xuc; 1112 struct sockaddr_in6 addrs[2]; 1113 struct inpcb *inp; 1114 int error, mapped = 0; 1115 1116 error = priv_check_cred(req->td->td_ucred, PRIV_NETINET_GETCRED, 1117 SUSER_ALLOWJAIL); 1118 if (error) 1119 return (error); 1120 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1121 if (error) 1122 return (error); 1123 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1124 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1125 return (error); 1126 } 1127 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1128 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1129 mapped = 1; 1130 else 1131 return (EINVAL); 1132 } 1133 1134 INP_INFO_RLOCK(&tcbinfo); 1135 if (mapped == 1) 1136 inp = in_pcblookup_hash(&tcbinfo, 1137 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1138 addrs[1].sin6_port, 1139 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1140 addrs[0].sin6_port, 1141 0, NULL); 1142 else 1143 inp = in6_pcblookup_hash(&tcbinfo, 1144 &addrs[1].sin6_addr, addrs[1].sin6_port, 1145 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1146 if (inp == NULL) { 1147 error = ENOENT; 1148 goto outunlocked; 1149 } 1150 INP_LOCK(inp); 1151 if (inp->inp_socket == NULL) { 1152 error = ENOENT; 1153 goto out; 1154 } 1155 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1156 if (error) 1157 goto out; 1158 cru2x(inp->inp_socket->so_cred, &xuc); 1159 out: 1160 INP_UNLOCK(inp); 1161 outunlocked: 1162 INP_INFO_RUNLOCK(&tcbinfo); 1163 if (error == 0) 1164 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1165 return (error); 1166 } 1167 1168 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1169 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1170 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1171 #endif 1172 1173 1174 void 1175 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1176 { 1177 struct ip *ip = vip; 1178 struct tcphdr *th; 1179 struct in_addr faddr; 1180 struct inpcb *inp; 1181 struct tcpcb *tp; 1182 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1183 struct icmp *icp; 1184 struct in_conninfo inc; 1185 tcp_seq icmp_tcp_seq; 1186 int mtu; 1187 1188 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1189 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1190 return; 1191 1192 if (cmd == PRC_MSGSIZE) 1193 notify = tcp_mtudisc; 1194 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1195 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1196 notify = tcp_drop_syn_sent; 1197 /* 1198 * Redirects don't need to be handled up here. 1199 */ 1200 else if (PRC_IS_REDIRECT(cmd)) 1201 return; 1202 /* 1203 * Source quench is depreciated. 1204 */ 1205 else if (cmd == PRC_QUENCH) 1206 return; 1207 /* 1208 * Hostdead is ugly because it goes linearly through all PCBs. 1209 * XXX: We never get this from ICMP, otherwise it makes an 1210 * excellent DoS attack on machines with many connections. 1211 */ 1212 else if (cmd == PRC_HOSTDEAD) 1213 ip = NULL; 1214 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1215 return; 1216 if (ip != NULL) { 1217 icp = (struct icmp *)((caddr_t)ip 1218 - offsetof(struct icmp, icmp_ip)); 1219 th = (struct tcphdr *)((caddr_t)ip 1220 + (ip->ip_hl << 2)); 1221 INP_INFO_WLOCK(&tcbinfo); 1222 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1223 ip->ip_src, th->th_sport, 0, NULL); 1224 if (inp != NULL) { 1225 INP_LOCK(inp); 1226 if (!(inp->inp_vflag & INP_TIMEWAIT) && 1227 !(inp->inp_vflag & INP_DROPPED) && 1228 !(inp->inp_socket == NULL)) { 1229 icmp_tcp_seq = htonl(th->th_seq); 1230 tp = intotcpcb(inp); 1231 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1232 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1233 if (cmd == PRC_MSGSIZE) { 1234 /* 1235 * MTU discovery: 1236 * If we got a needfrag set the MTU 1237 * in the route to the suggested new 1238 * value (if given) and then notify. 1239 */ 1240 bzero(&inc, sizeof(inc)); 1241 inc.inc_flags = 0; /* IPv4 */ 1242 inc.inc_faddr = faddr; 1243 1244 mtu = ntohs(icp->icmp_nextmtu); 1245 /* 1246 * If no alternative MTU was 1247 * proposed, try the next smaller 1248 * one. ip->ip_len has already 1249 * been swapped in icmp_input(). 1250 */ 1251 if (!mtu) 1252 mtu = ip_next_mtu(ip->ip_len, 1253 1); 1254 if (mtu < max(296, (tcp_minmss) 1255 + sizeof(struct tcpiphdr))) 1256 mtu = 0; 1257 if (!mtu) 1258 mtu = tcp_mssdflt 1259 + sizeof(struct tcpiphdr); 1260 /* 1261 * Only cache the the MTU if it 1262 * is smaller than the interface 1263 * or route MTU. tcp_mtudisc() 1264 * will do right thing by itself. 1265 */ 1266 if (mtu <= tcp_maxmtu(&inc, NULL)) 1267 tcp_hc_updatemtu(&inc, mtu); 1268 } 1269 1270 inp = (*notify)(inp, inetctlerrmap[cmd]); 1271 } 1272 } 1273 if (inp != NULL) 1274 INP_UNLOCK(inp); 1275 } else { 1276 inc.inc_fport = th->th_dport; 1277 inc.inc_lport = th->th_sport; 1278 inc.inc_faddr = faddr; 1279 inc.inc_laddr = ip->ip_src; 1280 #ifdef INET6 1281 inc.inc_isipv6 = 0; 1282 #endif 1283 syncache_unreach(&inc, th); 1284 } 1285 INP_INFO_WUNLOCK(&tcbinfo); 1286 } else 1287 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1288 } 1289 1290 #ifdef INET6 1291 void 1292 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1293 { 1294 struct tcphdr th; 1295 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1296 struct ip6_hdr *ip6; 1297 struct mbuf *m; 1298 struct ip6ctlparam *ip6cp = NULL; 1299 const struct sockaddr_in6 *sa6_src = NULL; 1300 int off; 1301 struct tcp_portonly { 1302 u_int16_t th_sport; 1303 u_int16_t th_dport; 1304 } *thp; 1305 1306 if (sa->sa_family != AF_INET6 || 1307 sa->sa_len != sizeof(struct sockaddr_in6)) 1308 return; 1309 1310 if (cmd == PRC_MSGSIZE) 1311 notify = tcp_mtudisc; 1312 else if (!PRC_IS_REDIRECT(cmd) && 1313 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1314 return; 1315 /* Source quench is depreciated. */ 1316 else if (cmd == PRC_QUENCH) 1317 return; 1318 1319 /* if the parameter is from icmp6, decode it. */ 1320 if (d != NULL) { 1321 ip6cp = (struct ip6ctlparam *)d; 1322 m = ip6cp->ip6c_m; 1323 ip6 = ip6cp->ip6c_ip6; 1324 off = ip6cp->ip6c_off; 1325 sa6_src = ip6cp->ip6c_src; 1326 } else { 1327 m = NULL; 1328 ip6 = NULL; 1329 off = 0; /* fool gcc */ 1330 sa6_src = &sa6_any; 1331 } 1332 1333 if (ip6 != NULL) { 1334 struct in_conninfo inc; 1335 /* 1336 * XXX: We assume that when IPV6 is non NULL, 1337 * M and OFF are valid. 1338 */ 1339 1340 /* check if we can safely examine src and dst ports */ 1341 if (m->m_pkthdr.len < off + sizeof(*thp)) 1342 return; 1343 1344 bzero(&th, sizeof(th)); 1345 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1346 1347 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1348 (struct sockaddr *)ip6cp->ip6c_src, 1349 th.th_sport, cmd, NULL, notify); 1350 1351 inc.inc_fport = th.th_dport; 1352 inc.inc_lport = th.th_sport; 1353 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1354 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1355 inc.inc_isipv6 = 1; 1356 INP_INFO_WLOCK(&tcbinfo); 1357 syncache_unreach(&inc, &th); 1358 INP_INFO_WUNLOCK(&tcbinfo); 1359 } else 1360 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1361 0, cmd, NULL, notify); 1362 } 1363 #endif /* INET6 */ 1364 1365 1366 /* 1367 * Following is where TCP initial sequence number generation occurs. 1368 * 1369 * There are two places where we must use initial sequence numbers: 1370 * 1. In SYN-ACK packets. 1371 * 2. In SYN packets. 1372 * 1373 * All ISNs for SYN-ACK packets are generated by the syncache. See 1374 * tcp_syncache.c for details. 1375 * 1376 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1377 * depends on this property. In addition, these ISNs should be 1378 * unguessable so as to prevent connection hijacking. To satisfy 1379 * the requirements of this situation, the algorithm outlined in 1380 * RFC 1948 is used, with only small modifications. 1381 * 1382 * Implementation details: 1383 * 1384 * Time is based off the system timer, and is corrected so that it 1385 * increases by one megabyte per second. This allows for proper 1386 * recycling on high speed LANs while still leaving over an hour 1387 * before rollover. 1388 * 1389 * As reading the *exact* system time is too expensive to be done 1390 * whenever setting up a TCP connection, we increment the time 1391 * offset in two ways. First, a small random positive increment 1392 * is added to isn_offset for each connection that is set up. 1393 * Second, the function tcp_isn_tick fires once per clock tick 1394 * and increments isn_offset as necessary so that sequence numbers 1395 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1396 * random positive increments serve only to ensure that the same 1397 * exact sequence number is never sent out twice (as could otherwise 1398 * happen when a port is recycled in less than the system tick 1399 * interval.) 1400 * 1401 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1402 * between seeding of isn_secret. This is normally set to zero, 1403 * as reseeding should not be necessary. 1404 * 1405 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1406 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1407 * general, this means holding an exclusive (write) lock. 1408 */ 1409 1410 #define ISN_BYTES_PER_SECOND 1048576 1411 #define ISN_STATIC_INCREMENT 4096 1412 #define ISN_RANDOM_INCREMENT (4096 - 1) 1413 1414 static u_char isn_secret[32]; 1415 static int isn_last_reseed; 1416 static u_int32_t isn_offset, isn_offset_old; 1417 static MD5_CTX isn_ctx; 1418 1419 tcp_seq 1420 tcp_new_isn(struct tcpcb *tp) 1421 { 1422 u_int32_t md5_buffer[4]; 1423 tcp_seq new_isn; 1424 1425 INP_LOCK_ASSERT(tp->t_inpcb); 1426 1427 ISN_LOCK(); 1428 /* Seed if this is the first use, reseed if requested. */ 1429 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1430 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1431 < (u_int)ticks))) { 1432 read_random(&isn_secret, sizeof(isn_secret)); 1433 isn_last_reseed = ticks; 1434 } 1435 1436 /* Compute the md5 hash and return the ISN. */ 1437 MD5Init(&isn_ctx); 1438 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1439 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1440 #ifdef INET6 1441 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1442 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1443 sizeof(struct in6_addr)); 1444 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1445 sizeof(struct in6_addr)); 1446 } else 1447 #endif 1448 { 1449 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1450 sizeof(struct in_addr)); 1451 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1452 sizeof(struct in_addr)); 1453 } 1454 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1455 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1456 new_isn = (tcp_seq) md5_buffer[0]; 1457 isn_offset += ISN_STATIC_INCREMENT + 1458 (arc4random() & ISN_RANDOM_INCREMENT); 1459 new_isn += isn_offset; 1460 ISN_UNLOCK(); 1461 return (new_isn); 1462 } 1463 1464 /* 1465 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1466 * to keep time flowing at a relatively constant rate. If the random 1467 * increments have already pushed us past the projected offset, do nothing. 1468 */ 1469 static void 1470 tcp_isn_tick(void *xtp) 1471 { 1472 u_int32_t projected_offset; 1473 1474 ISN_LOCK(); 1475 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1476 1477 if (projected_offset > isn_offset) 1478 isn_offset = projected_offset; 1479 1480 isn_offset_old = isn_offset; 1481 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1482 ISN_UNLOCK(); 1483 } 1484 1485 /* 1486 * When a specific ICMP unreachable message is received and the 1487 * connection state is SYN-SENT, drop the connection. This behavior 1488 * is controlled by the icmp_may_rst sysctl. 1489 */ 1490 struct inpcb * 1491 tcp_drop_syn_sent(struct inpcb *inp, int errno) 1492 { 1493 struct tcpcb *tp; 1494 1495 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1496 INP_LOCK_ASSERT(inp); 1497 1498 if ((inp->inp_vflag & INP_TIMEWAIT) || 1499 (inp->inp_vflag & INP_DROPPED)) 1500 return (inp); 1501 1502 tp = intotcpcb(inp); 1503 if (tp->t_state != TCPS_SYN_SENT) 1504 return (inp); 1505 1506 tp = tcp_drop(tp, errno); 1507 if (tp != NULL) 1508 return (inp); 1509 else 1510 return (NULL); 1511 } 1512 1513 /* 1514 * When `need fragmentation' ICMP is received, update our idea of the MSS 1515 * based on the new value in the route. Also nudge TCP to send something, 1516 * since we know the packet we just sent was dropped. 1517 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1518 */ 1519 struct inpcb * 1520 tcp_mtudisc(struct inpcb *inp, int errno) 1521 { 1522 struct tcpcb *tp; 1523 struct socket *so = inp->inp_socket; 1524 u_int maxmtu; 1525 u_int romtu; 1526 int mss; 1527 #ifdef INET6 1528 int isipv6; 1529 #endif /* INET6 */ 1530 1531 INP_LOCK_ASSERT(inp); 1532 if ((inp->inp_vflag & INP_TIMEWAIT) || 1533 (inp->inp_vflag & INP_DROPPED)) 1534 return (inp); 1535 1536 tp = intotcpcb(inp); 1537 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1538 1539 #ifdef INET6 1540 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1541 #endif 1542 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1543 romtu = 1544 #ifdef INET6 1545 isipv6 ? tcp_maxmtu6(&inp->inp_inc, NULL) : 1546 #endif /* INET6 */ 1547 tcp_maxmtu(&inp->inp_inc, NULL); 1548 if (!maxmtu) 1549 maxmtu = romtu; 1550 else 1551 maxmtu = min(maxmtu, romtu); 1552 if (!maxmtu) { 1553 tp->t_maxopd = tp->t_maxseg = 1554 #ifdef INET6 1555 isipv6 ? tcp_v6mssdflt : 1556 #endif /* INET6 */ 1557 tcp_mssdflt; 1558 return (inp); 1559 } 1560 mss = maxmtu - 1561 #ifdef INET6 1562 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1563 #endif /* INET6 */ 1564 sizeof(struct tcpiphdr) 1565 #ifdef INET6 1566 ) 1567 #endif /* INET6 */ 1568 ; 1569 1570 /* 1571 * XXX - The above conditional probably violates the TCP 1572 * spec. The problem is that, since we don't know the 1573 * other end's MSS, we are supposed to use a conservative 1574 * default. But, if we do that, then MTU discovery will 1575 * never actually take place, because the conservative 1576 * default is much less than the MTUs typically seen 1577 * on the Internet today. For the moment, we'll sweep 1578 * this under the carpet. 1579 * 1580 * The conservative default might not actually be a problem 1581 * if the only case this occurs is when sending an initial 1582 * SYN with options and data to a host we've never talked 1583 * to before. Then, they will reply with an MSS value which 1584 * will get recorded and the new parameters should get 1585 * recomputed. For Further Study. 1586 */ 1587 if (tp->t_maxopd <= mss) 1588 return (inp); 1589 tp->t_maxopd = mss; 1590 1591 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1592 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1593 mss -= TCPOLEN_TSTAMP_APPA; 1594 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1595 if (mss > MCLBYTES) 1596 mss &= ~(MCLBYTES-1); 1597 #else 1598 if (mss > MCLBYTES) 1599 mss = mss / MCLBYTES * MCLBYTES; 1600 #endif 1601 if (so->so_snd.sb_hiwat < mss) 1602 mss = so->so_snd.sb_hiwat; 1603 1604 tp->t_maxseg = mss; 1605 1606 tcpstat.tcps_mturesent++; 1607 tp->t_rtttime = 0; 1608 tp->snd_nxt = tp->snd_una; 1609 tcp_free_sackholes(tp); 1610 tp->snd_recover = tp->snd_max; 1611 if (tp->sack_enable) 1612 EXIT_FASTRECOVERY(tp); 1613 tcp_output(tp); 1614 return (inp); 1615 } 1616 1617 /* 1618 * Look-up the routing entry to the peer of this inpcb. If no route 1619 * is found and it cannot be allocated, then return NULL. This routine 1620 * is called by TCP routines that access the rmx structure and by tcp_mss 1621 * to get the interface MTU. 1622 */ 1623 u_long 1624 tcp_maxmtu(struct in_conninfo *inc, int *flags) 1625 { 1626 struct route sro; 1627 struct sockaddr_in *dst; 1628 struct ifnet *ifp; 1629 u_long maxmtu = 0; 1630 1631 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1632 1633 bzero(&sro, sizeof(sro)); 1634 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1635 dst = (struct sockaddr_in *)&sro.ro_dst; 1636 dst->sin_family = AF_INET; 1637 dst->sin_len = sizeof(*dst); 1638 dst->sin_addr = inc->inc_faddr; 1639 rtalloc_ign(&sro, RTF_CLONING); 1640 } 1641 if (sro.ro_rt != NULL) { 1642 ifp = sro.ro_rt->rt_ifp; 1643 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1644 maxmtu = ifp->if_mtu; 1645 else 1646 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1647 1648 /* Report additional interface capabilities. */ 1649 if (flags != NULL) { 1650 if (ifp->if_capenable & IFCAP_TSO4 && 1651 ifp->if_hwassist & CSUM_TSO) 1652 *flags |= CSUM_TSO; 1653 } 1654 RTFREE(sro.ro_rt); 1655 } 1656 return (maxmtu); 1657 } 1658 1659 #ifdef INET6 1660 u_long 1661 tcp_maxmtu6(struct in_conninfo *inc, int *flags) 1662 { 1663 struct route_in6 sro6; 1664 struct ifnet *ifp; 1665 u_long maxmtu = 0; 1666 1667 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1668 1669 bzero(&sro6, sizeof(sro6)); 1670 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1671 sro6.ro_dst.sin6_family = AF_INET6; 1672 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1673 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1674 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1675 } 1676 if (sro6.ro_rt != NULL) { 1677 ifp = sro6.ro_rt->rt_ifp; 1678 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1679 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1680 else 1681 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1682 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1683 1684 /* Report additional interface capabilities. */ 1685 if (flags != NULL) { 1686 if (ifp->if_capenable & IFCAP_TSO6 && 1687 ifp->if_hwassist & CSUM_TSO) 1688 *flags |= CSUM_TSO; 1689 } 1690 RTFREE(sro6.ro_rt); 1691 } 1692 1693 return (maxmtu); 1694 } 1695 #endif /* INET6 */ 1696 1697 #ifdef IPSEC 1698 /* compute ESP/AH header size for TCP, including outer IP header. */ 1699 size_t 1700 ipsec_hdrsiz_tcp(struct tcpcb *tp) 1701 { 1702 struct inpcb *inp; 1703 struct mbuf *m; 1704 size_t hdrsiz; 1705 struct ip *ip; 1706 #ifdef INET6 1707 struct ip6_hdr *ip6; 1708 #endif 1709 struct tcphdr *th; 1710 1711 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1712 return (0); 1713 MGETHDR(m, M_DONTWAIT, MT_DATA); 1714 if (!m) 1715 return (0); 1716 1717 #ifdef INET6 1718 if ((inp->inp_vflag & INP_IPV6) != 0) { 1719 ip6 = mtod(m, struct ip6_hdr *); 1720 th = (struct tcphdr *)(ip6 + 1); 1721 m->m_pkthdr.len = m->m_len = 1722 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1723 tcpip_fillheaders(inp, ip6, th); 1724 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1725 } else 1726 #endif /* INET6 */ 1727 { 1728 ip = mtod(m, struct ip *); 1729 th = (struct tcphdr *)(ip + 1); 1730 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1731 tcpip_fillheaders(inp, ip, th); 1732 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1733 } 1734 1735 m_free(m); 1736 return (hdrsiz); 1737 } 1738 #endif /*IPSEC*/ 1739 1740 /* 1741 * Move a TCP connection into TIME_WAIT state. 1742 * tcbinfo is locked. 1743 * inp is locked, and is unlocked before returning. 1744 */ 1745 void 1746 tcp_twstart(struct tcpcb *tp) 1747 { 1748 struct tcptw *tw; 1749 struct inpcb *inp = tp->t_inpcb; 1750 int acknow; 1751 struct socket *so; 1752 1753 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1754 INP_LOCK_ASSERT(inp); 1755 1756 if (nolocaltimewait && in_localip(inp->inp_faddr)) { 1757 tp = tcp_close(tp); 1758 if (tp != NULL) 1759 INP_UNLOCK(inp); 1760 return; 1761 } 1762 1763 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1764 if (tw == NULL) { 1765 tw = tcp_timer_2msl_tw(1); 1766 if (tw == NULL) { 1767 tp = tcp_close(tp); 1768 if (tp != NULL) 1769 INP_UNLOCK(inp); 1770 return; 1771 } 1772 } 1773 tw->tw_inpcb = inp; 1774 1775 /* 1776 * Recover last window size sent. 1777 */ 1778 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1779 1780 /* 1781 * Set t_recent if timestamps are used on the connection. 1782 */ 1783 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1784 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1785 tw->t_recent = tp->ts_recent; 1786 else 1787 tw->t_recent = 0; 1788 1789 tw->snd_nxt = tp->snd_nxt; 1790 tw->rcv_nxt = tp->rcv_nxt; 1791 tw->iss = tp->iss; 1792 tw->irs = tp->irs; 1793 tw->t_starttime = tp->t_starttime; 1794 tw->tw_time = 0; 1795 1796 /* XXX 1797 * If this code will 1798 * be used for fin-wait-2 state also, then we may need 1799 * a ts_recent from the last segment. 1800 */ 1801 acknow = tp->t_flags & TF_ACKNOW; 1802 1803 /* 1804 * First, discard tcpcb state, which includes stopping its timers and 1805 * freeing it. tcp_discardcb() used to also release the inpcb, but 1806 * that work is now done in the caller. 1807 * 1808 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 1809 * and might not be needed here any longer. 1810 */ 1811 tcp_discardcb(tp); 1812 so = inp->inp_socket; 1813 soisdisconnected(so); 1814 SOCK_LOCK(so); 1815 tw->tw_cred = crhold(so->so_cred); 1816 tw->tw_so_options = so->so_options; 1817 SOCK_UNLOCK(so); 1818 if (acknow) 1819 tcp_twrespond(tw, TH_ACK); 1820 inp->inp_ppcb = tw; 1821 inp->inp_vflag |= INP_TIMEWAIT; 1822 tcp_timer_2msl_reset(tw, 0); 1823 1824 /* 1825 * If the inpcb owns the sole reference to the socket, then we can 1826 * detach and free the socket as it is not needed in time wait. 1827 */ 1828 if (inp->inp_vflag & INP_SOCKREF) { 1829 KASSERT(so->so_state & SS_PROTOREF, 1830 ("tcp_twstart: !SS_PROTOREF")); 1831 inp->inp_vflag &= ~INP_SOCKREF; 1832 INP_UNLOCK(inp); 1833 ACCEPT_LOCK(); 1834 SOCK_LOCK(so); 1835 so->so_state &= ~SS_PROTOREF; 1836 sofree(so); 1837 } else 1838 INP_UNLOCK(inp); 1839 } 1840 1841 #if 0 1842 /* 1843 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1844 * the actual rate is slightly higher due to the addition of 1845 * random positive increments. 1846 * 1847 * Most other new OSes use semi-randomized ISN values, so we 1848 * do not need to worry about them. 1849 */ 1850 #define MS_ISN_BYTES_PER_SECOND 250000 1851 1852 /* 1853 * Determine if the ISN we will generate has advanced beyond the last 1854 * sequence number used by the previous connection. If so, indicate 1855 * that it is safe to recycle this tw socket by returning 1. 1856 */ 1857 int 1858 tcp_twrecycleable(struct tcptw *tw) 1859 { 1860 tcp_seq new_iss = tw->iss; 1861 tcp_seq new_irs = tw->irs; 1862 1863 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1864 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1865 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1866 1867 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1868 return (1); 1869 else 1870 return (0); 1871 } 1872 #endif 1873 1874 void 1875 tcp_twclose(struct tcptw *tw, int reuse) 1876 { 1877 struct socket *so; 1878 struct inpcb *inp; 1879 1880 /* 1881 * At this point, we are in one of two situations: 1882 * 1883 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 1884 * all state. 1885 * 1886 * (2) We have a socket -- if we own a reference, release it and 1887 * notify the socket layer. 1888 */ 1889 inp = tw->tw_inpcb; 1890 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1891 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 1892 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1893 INP_LOCK_ASSERT(inp); 1894 1895 tw->tw_inpcb = NULL; 1896 tcp_timer_2msl_stop(tw); 1897 inp->inp_ppcb = NULL; 1898 in_pcbdrop(inp); 1899 1900 so = inp->inp_socket; 1901 if (so != NULL) { 1902 /* 1903 * If there's a socket, handle two cases: first, we own a 1904 * strong reference, which we will now release, or we don't 1905 * in which case another reference exists (XXXRW: think 1906 * about this more), and we don't need to take action. 1907 */ 1908 if (inp->inp_vflag & INP_SOCKREF) { 1909 inp->inp_vflag &= ~INP_SOCKREF; 1910 INP_UNLOCK(inp); 1911 ACCEPT_LOCK(); 1912 SOCK_LOCK(so); 1913 KASSERT(so->so_state & SS_PROTOREF, 1914 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1915 so->so_state &= ~SS_PROTOREF; 1916 sofree(so); 1917 } else { 1918 /* 1919 * If we don't own the only reference, the socket and 1920 * inpcb need to be left around to be handled by 1921 * tcp_usr_detach() later. 1922 */ 1923 INP_UNLOCK(inp); 1924 } 1925 } else { 1926 #ifdef INET6 1927 if (inp->inp_vflag & INP_IPV6PROTO) 1928 in6_pcbfree(inp); 1929 else 1930 #endif 1931 in_pcbfree(inp); 1932 } 1933 tcpstat.tcps_closed++; 1934 crfree(tw->tw_cred); 1935 tw->tw_cred = NULL; 1936 if (reuse) 1937 return; 1938 uma_zfree(tcptw_zone, tw); 1939 } 1940 1941 int 1942 tcp_twrespond(struct tcptw *tw, int flags) 1943 { 1944 struct inpcb *inp = tw->tw_inpcb; 1945 struct tcphdr *th; 1946 struct mbuf *m; 1947 struct ip *ip = NULL; 1948 u_int8_t *optp; 1949 u_int hdrlen, optlen; 1950 int error; 1951 #ifdef INET6 1952 struct ip6_hdr *ip6 = NULL; 1953 int isipv6 = inp->inp_inc.inc_isipv6; 1954 #endif 1955 1956 INP_LOCK_ASSERT(inp); 1957 1958 m = m_gethdr(M_DONTWAIT, MT_DATA); 1959 if (m == NULL) 1960 return (ENOBUFS); 1961 m->m_data += max_linkhdr; 1962 1963 #ifdef MAC 1964 mac_create_mbuf_from_inpcb(inp, m); 1965 #endif 1966 1967 #ifdef INET6 1968 if (isipv6) { 1969 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1970 ip6 = mtod(m, struct ip6_hdr *); 1971 th = (struct tcphdr *)(ip6 + 1); 1972 tcpip_fillheaders(inp, ip6, th); 1973 } else 1974 #endif 1975 { 1976 hdrlen = sizeof(struct tcpiphdr); 1977 ip = mtod(m, struct ip *); 1978 th = (struct tcphdr *)(ip + 1); 1979 tcpip_fillheaders(inp, ip, th); 1980 } 1981 optp = (u_int8_t *)(th + 1); 1982 1983 /* 1984 * Send a timestamp and echo-reply if both our side and our peer 1985 * have sent timestamps in our SYN's and this is not a RST. 1986 */ 1987 if (tw->t_recent && flags == TH_ACK) { 1988 u_int32_t *lp = (u_int32_t *)optp; 1989 1990 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1991 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1992 *lp++ = htonl(ticks); 1993 *lp = htonl(tw->t_recent); 1994 optp += TCPOLEN_TSTAMP_APPA; 1995 } 1996 1997 optlen = optp - (u_int8_t *)(th + 1); 1998 1999 m->m_len = hdrlen + optlen; 2000 m->m_pkthdr.len = m->m_len; 2001 2002 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 2003 2004 th->th_seq = htonl(tw->snd_nxt); 2005 th->th_ack = htonl(tw->rcv_nxt); 2006 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 2007 th->th_flags = flags; 2008 th->th_win = htons(tw->last_win); 2009 2010 #ifdef INET6 2011 if (isipv6) { 2012 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 2013 sizeof(struct tcphdr) + optlen); 2014 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2015 error = ip6_output(m, inp->in6p_outputopts, NULL, 2016 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 2017 } else 2018 #endif 2019 { 2020 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2021 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 2022 m->m_pkthdr.csum_flags = CSUM_TCP; 2023 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2024 ip->ip_len = m->m_pkthdr.len; 2025 if (path_mtu_discovery) 2026 ip->ip_off |= IP_DF; 2027 error = ip_output(m, inp->inp_options, NULL, 2028 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 2029 NULL, inp); 2030 } 2031 if (flags & TH_ACK) 2032 tcpstat.tcps_sndacks++; 2033 else 2034 tcpstat.tcps_sndctrl++; 2035 tcpstat.tcps_sndtotal++; 2036 return (error); 2037 } 2038 2039 /* 2040 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 2041 * 2042 * This code attempts to calculate the bandwidth-delay product as a 2043 * means of determining the optimal window size to maximize bandwidth, 2044 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 2045 * routers. This code also does a fairly good job keeping RTTs in check 2046 * across slow links like modems. We implement an algorithm which is very 2047 * similar (but not meant to be) TCP/Vegas. The code operates on the 2048 * transmitter side of a TCP connection and so only effects the transmit 2049 * side of the connection. 2050 * 2051 * BACKGROUND: TCP makes no provision for the management of buffer space 2052 * at the end points or at the intermediate routers and switches. A TCP 2053 * stream, whether using NewReno or not, will eventually buffer as 2054 * many packets as it is able and the only reason this typically works is 2055 * due to the fairly small default buffers made available for a connection 2056 * (typicaly 16K or 32K). As machines use larger windows and/or window 2057 * scaling it is now fairly easy for even a single TCP connection to blow-out 2058 * all available buffer space not only on the local interface, but on 2059 * intermediate routers and switches as well. NewReno makes a misguided 2060 * attempt to 'solve' this problem by waiting for an actual failure to occur, 2061 * then backing off, then steadily increasing the window again until another 2062 * failure occurs, ad-infinitum. This results in terrible oscillation that 2063 * is only made worse as network loads increase and the idea of intentionally 2064 * blowing out network buffers is, frankly, a terrible way to manage network 2065 * resources. 2066 * 2067 * It is far better to limit the transmit window prior to the failure 2068 * condition being achieved. There are two general ways to do this: First 2069 * you can 'scan' through different transmit window sizes and locate the 2070 * point where the RTT stops increasing, indicating that you have filled the 2071 * pipe, then scan backwards until you note that RTT stops decreasing, then 2072 * repeat ad-infinitum. This method works in principle but has severe 2073 * implementation issues due to RTT variances, timer granularity, and 2074 * instability in the algorithm which can lead to many false positives and 2075 * create oscillations as well as interact badly with other TCP streams 2076 * implementing the same algorithm. 2077 * 2078 * The second method is to limit the window to the bandwidth delay product 2079 * of the link. This is the method we implement. RTT variances and our 2080 * own manipulation of the congestion window, bwnd, can potentially 2081 * destabilize the algorithm. For this reason we have to stabilize the 2082 * elements used to calculate the window. We do this by using the minimum 2083 * observed RTT, the long term average of the observed bandwidth, and 2084 * by adding two segments worth of slop. It isn't perfect but it is able 2085 * to react to changing conditions and gives us a very stable basis on 2086 * which to extend the algorithm. 2087 */ 2088 void 2089 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2090 { 2091 u_long bw; 2092 u_long bwnd; 2093 int save_ticks; 2094 2095 INP_LOCK_ASSERT(tp->t_inpcb); 2096 2097 /* 2098 * If inflight_enable is disabled in the middle of a tcp connection, 2099 * make sure snd_bwnd is effectively disabled. 2100 */ 2101 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2102 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2103 tp->snd_bandwidth = 0; 2104 return; 2105 } 2106 2107 /* 2108 * Figure out the bandwidth. Due to the tick granularity this 2109 * is a very rough number and it MUST be averaged over a fairly 2110 * long period of time. XXX we need to take into account a link 2111 * that is not using all available bandwidth, but for now our 2112 * slop will ramp us up if this case occurs and the bandwidth later 2113 * increases. 2114 * 2115 * Note: if ticks rollover 'bw' may wind up negative. We must 2116 * effectively reset t_bw_rtttime for this case. 2117 */ 2118 save_ticks = ticks; 2119 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2120 return; 2121 2122 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2123 (save_ticks - tp->t_bw_rtttime); 2124 tp->t_bw_rtttime = save_ticks; 2125 tp->t_bw_rtseq = ack_seq; 2126 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2127 return; 2128 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2129 2130 tp->snd_bandwidth = bw; 2131 2132 /* 2133 * Calculate the semi-static bandwidth delay product, plus two maximal 2134 * segments. The additional slop puts us squarely in the sweet 2135 * spot and also handles the bandwidth run-up case and stabilization. 2136 * Without the slop we could be locking ourselves into a lower 2137 * bandwidth. 2138 * 2139 * Situations Handled: 2140 * (1) Prevents over-queueing of packets on LANs, especially on 2141 * high speed LANs, allowing larger TCP buffers to be 2142 * specified, and also does a good job preventing 2143 * over-queueing of packets over choke points like modems 2144 * (at least for the transmit side). 2145 * 2146 * (2) Is able to handle changing network loads (bandwidth 2147 * drops so bwnd drops, bandwidth increases so bwnd 2148 * increases). 2149 * 2150 * (3) Theoretically should stabilize in the face of multiple 2151 * connections implementing the same algorithm (this may need 2152 * a little work). 2153 * 2154 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2155 * be adjusted with a sysctl but typically only needs to be 2156 * on very slow connections. A value no smaller then 5 2157 * should be used, but only reduce this default if you have 2158 * no other choice. 2159 */ 2160 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2161 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2162 #undef USERTT 2163 2164 if (tcp_inflight_debug > 0) { 2165 static int ltime; 2166 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2167 ltime = ticks; 2168 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2169 tp, 2170 bw, 2171 tp->t_rttbest, 2172 tp->t_srtt, 2173 bwnd 2174 ); 2175 } 2176 } 2177 if ((long)bwnd < tcp_inflight_min) 2178 bwnd = tcp_inflight_min; 2179 if (bwnd > tcp_inflight_max) 2180 bwnd = tcp_inflight_max; 2181 if ((long)bwnd < tp->t_maxseg * 2) 2182 bwnd = tp->t_maxseg * 2; 2183 tp->snd_bwnd = bwnd; 2184 } 2185 2186 #ifdef TCP_SIGNATURE 2187 /* 2188 * Callback function invoked by m_apply() to digest TCP segment data 2189 * contained within an mbuf chain. 2190 */ 2191 static int 2192 tcp_signature_apply(void *fstate, void *data, u_int len) 2193 { 2194 2195 MD5Update(fstate, (u_char *)data, len); 2196 return (0); 2197 } 2198 2199 /* 2200 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2201 * 2202 * Parameters: 2203 * m pointer to head of mbuf chain 2204 * off0 offset to TCP header within the mbuf chain 2205 * len length of TCP segment data, excluding options 2206 * optlen length of TCP segment options 2207 * buf pointer to storage for computed MD5 digest 2208 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2209 * 2210 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2211 * When called from tcp_input(), we can be sure that th_sum has been 2212 * zeroed out and verified already. 2213 * 2214 * This function is for IPv4 use only. Calling this function with an 2215 * IPv6 packet in the mbuf chain will yield undefined results. 2216 * 2217 * Return 0 if successful, otherwise return -1. 2218 * 2219 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2220 * search with the destination IP address, and a 'magic SPI' to be 2221 * determined by the application. This is hardcoded elsewhere to 1179 2222 * right now. Another branch of this code exists which uses the SPD to 2223 * specify per-application flows but it is unstable. 2224 */ 2225 int 2226 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2227 u_char *buf, u_int direction) 2228 { 2229 union sockaddr_union dst; 2230 struct ippseudo ippseudo; 2231 MD5_CTX ctx; 2232 int doff; 2233 struct ip *ip; 2234 struct ipovly *ipovly; 2235 struct secasvar *sav; 2236 struct tcphdr *th; 2237 u_short savecsum; 2238 2239 KASSERT(m != NULL, ("NULL mbuf chain")); 2240 KASSERT(buf != NULL, ("NULL signature pointer")); 2241 2242 /* Extract the destination from the IP header in the mbuf. */ 2243 ip = mtod(m, struct ip *); 2244 bzero(&dst, sizeof(union sockaddr_union)); 2245 dst.sa.sa_len = sizeof(struct sockaddr_in); 2246 dst.sa.sa_family = AF_INET; 2247 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2248 ip->ip_src : ip->ip_dst; 2249 2250 /* Look up an SADB entry which matches the address of the peer. */ 2251 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2252 if (sav == NULL) { 2253 printf("%s: SADB lookup failed for %s\n", __func__, 2254 inet_ntoa(dst.sin.sin_addr)); 2255 return (EINVAL); 2256 } 2257 2258 MD5Init(&ctx); 2259 ipovly = (struct ipovly *)ip; 2260 th = (struct tcphdr *)((u_char *)ip + off0); 2261 doff = off0 + sizeof(struct tcphdr) + optlen; 2262 2263 /* 2264 * Step 1: Update MD5 hash with IP pseudo-header. 2265 * 2266 * XXX The ippseudo header MUST be digested in network byte order, 2267 * or else we'll fail the regression test. Assume all fields we've 2268 * been doing arithmetic on have been in host byte order. 2269 * XXX One cannot depend on ipovly->ih_len here. When called from 2270 * tcp_output(), the underlying ip_len member has not yet been set. 2271 */ 2272 ippseudo.ippseudo_src = ipovly->ih_src; 2273 ippseudo.ippseudo_dst = ipovly->ih_dst; 2274 ippseudo.ippseudo_pad = 0; 2275 ippseudo.ippseudo_p = IPPROTO_TCP; 2276 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2277 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2278 2279 /* 2280 * Step 2: Update MD5 hash with TCP header, excluding options. 2281 * The TCP checksum must be set to zero. 2282 */ 2283 savecsum = th->th_sum; 2284 th->th_sum = 0; 2285 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2286 th->th_sum = savecsum; 2287 2288 /* 2289 * Step 3: Update MD5 hash with TCP segment data. 2290 * Use m_apply() to avoid an early m_pullup(). 2291 */ 2292 if (len > 0) 2293 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2294 2295 /* 2296 * Step 4: Update MD5 hash with shared secret. 2297 */ 2298 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2299 MD5Final(buf, &ctx); 2300 2301 key_sa_recordxfer(sav, m); 2302 KEY_FREESAV(&sav); 2303 return (0); 2304 } 2305 #endif /* TCP_SIGNATURE */ 2306 2307 static int 2308 sysctl_drop(SYSCTL_HANDLER_ARGS) 2309 { 2310 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2311 struct sockaddr_storage addrs[2]; 2312 struct inpcb *inp; 2313 struct tcpcb *tp; 2314 struct tcptw *tw; 2315 struct sockaddr_in *fin, *lin; 2316 #ifdef INET6 2317 struct sockaddr_in6 *fin6, *lin6; 2318 struct in6_addr f6, l6; 2319 #endif 2320 int error; 2321 2322 inp = NULL; 2323 fin = lin = NULL; 2324 #ifdef INET6 2325 fin6 = lin6 = NULL; 2326 #endif 2327 error = 0; 2328 2329 if (req->oldptr != NULL || req->oldlen != 0) 2330 return (EINVAL); 2331 if (req->newptr == NULL) 2332 return (EPERM); 2333 if (req->newlen < sizeof(addrs)) 2334 return (ENOMEM); 2335 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2336 if (error) 2337 return (error); 2338 2339 switch (addrs[0].ss_family) { 2340 #ifdef INET6 2341 case AF_INET6: 2342 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2343 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2344 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2345 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2346 return (EINVAL); 2347 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2348 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2349 return (EINVAL); 2350 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2351 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2352 fin = (struct sockaddr_in *)&addrs[0]; 2353 lin = (struct sockaddr_in *)&addrs[1]; 2354 break; 2355 } 2356 error = sa6_embedscope(fin6, ip6_use_defzone); 2357 if (error) 2358 return (error); 2359 error = sa6_embedscope(lin6, ip6_use_defzone); 2360 if (error) 2361 return (error); 2362 break; 2363 #endif 2364 case AF_INET: 2365 fin = (struct sockaddr_in *)&addrs[0]; 2366 lin = (struct sockaddr_in *)&addrs[1]; 2367 if (fin->sin_len != sizeof(struct sockaddr_in) || 2368 lin->sin_len != sizeof(struct sockaddr_in)) 2369 return (EINVAL); 2370 break; 2371 default: 2372 return (EINVAL); 2373 } 2374 INP_INFO_WLOCK(&tcbinfo); 2375 switch (addrs[0].ss_family) { 2376 #ifdef INET6 2377 case AF_INET6: 2378 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2379 &l6, lin6->sin6_port, 0, NULL); 2380 break; 2381 #endif 2382 case AF_INET: 2383 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2384 lin->sin_addr, lin->sin_port, 0, NULL); 2385 break; 2386 } 2387 if (inp != NULL) { 2388 INP_LOCK(inp); 2389 if (inp->inp_vflag & INP_TIMEWAIT) { 2390 /* 2391 * XXXRW: There currently exists a state where an 2392 * inpcb is present, but its timewait state has been 2393 * discarded. For now, don't allow dropping of this 2394 * type of inpcb. 2395 */ 2396 tw = intotw(inp); 2397 if (tw != NULL) 2398 tcp_twclose(tw, 0); 2399 } else if (!(inp->inp_vflag & INP_DROPPED) && 2400 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2401 tp = intotcpcb(inp); 2402 tcp_drop(tp, ECONNABORTED); 2403 } 2404 INP_UNLOCK(inp); 2405 } else 2406 error = ESRCH; 2407 INP_INFO_WUNLOCK(&tcbinfo); 2408 return (error); 2409 } 2410 2411 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2412 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2413 0, sysctl_drop, "", "Drop TCP connection"); 2414