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