1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 30 * $FreeBSD$ 31 */ 32 33 #include "opt_compat.h" 34 #include "opt_inet.h" 35 #include "opt_inet6.h" 36 #include "opt_ipsec.h" 37 #include "opt_mac.h" 38 #include "opt_tcpdebug.h" 39 #include "opt_tcp_sack.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/callout.h> 44 #include <sys/kernel.h> 45 #include <sys/sysctl.h> 46 #include <sys/mac.h> 47 #include <sys/malloc.h> 48 #include <sys/mbuf.h> 49 #ifdef INET6 50 #include <sys/domain.h> 51 #endif 52 #include <sys/proc.h> 53 #include <sys/socket.h> 54 #include <sys/socketvar.h> 55 #include <sys/protosw.h> 56 #include <sys/random.h> 57 58 #include <vm/uma.h> 59 60 #include <net/route.h> 61 #include <net/if.h> 62 63 #include <netinet/in.h> 64 #include <netinet/in_systm.h> 65 #include <netinet/ip.h> 66 #ifdef INET6 67 #include <netinet/ip6.h> 68 #endif 69 #include <netinet/in_pcb.h> 70 #ifdef INET6 71 #include <netinet6/in6_pcb.h> 72 #endif 73 #include <netinet/in_var.h> 74 #include <netinet/ip_var.h> 75 #ifdef INET6 76 #include <netinet6/ip6_var.h> 77 #include <netinet6/scope6_var.h> 78 #include <netinet6/nd6.h> 79 #endif 80 #include <netinet/ip_icmp.h> 81 #include <netinet/tcp.h> 82 #include <netinet/tcp_fsm.h> 83 #include <netinet/tcp_seq.h> 84 #include <netinet/tcp_timer.h> 85 #include <netinet/tcp_var.h> 86 #ifdef INET6 87 #include <netinet6/tcp6_var.h> 88 #endif 89 #include <netinet/tcpip.h> 90 #ifdef TCPDEBUG 91 #include <netinet/tcp_debug.h> 92 #endif 93 #include <netinet6/ip6protosw.h> 94 95 #ifdef IPSEC 96 #include <netinet6/ipsec.h> 97 #ifdef INET6 98 #include <netinet6/ipsec6.h> 99 #endif 100 #include <netkey/key.h> 101 #endif /*IPSEC*/ 102 103 #ifdef FAST_IPSEC 104 #include <netipsec/ipsec.h> 105 #include <netipsec/xform.h> 106 #ifdef INET6 107 #include <netipsec/ipsec6.h> 108 #endif 109 #include <netipsec/key.h> 110 #define IPSEC 111 #endif /*FAST_IPSEC*/ 112 113 #include <machine/in_cksum.h> 114 #include <sys/md5.h> 115 116 int tcp_mssdflt = TCP_MSS; 117 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 118 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); 119 120 #ifdef INET6 121 int tcp_v6mssdflt = TCP6_MSS; 122 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 123 CTLFLAG_RW, &tcp_v6mssdflt , 0, 124 "Default TCP Maximum Segment Size for IPv6"); 125 #endif 126 127 /* 128 * Minimum MSS we accept and use. This prevents DoS attacks where 129 * we are forced to a ridiculous low MSS like 20 and send hundreds 130 * of packets instead of one. The effect scales with the available 131 * bandwidth and quickly saturates the CPU and network interface 132 * with packet generation and sending. Set to zero to disable MINMSS 133 * checking. This setting prevents us from sending too small packets. 134 */ 135 int tcp_minmss = TCP_MINMSS; 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 137 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 138 /* 139 * Number of TCP segments per second we accept from remote host 140 * before we start to calculate average segment size. If average 141 * segment size drops below the minimum TCP MSS we assume a DoS 142 * attack and reset+drop the connection. Care has to be taken not to 143 * set this value too small to not kill interactive type connections 144 * (telnet, SSH) which send many small packets. 145 */ 146 int tcp_minmssoverload = TCP_MINMSSOVERLOAD; 147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW, 148 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to" 149 "be under the MINMSS Size"); 150 151 #if 0 152 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 153 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 154 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 155 #endif 156 157 int tcp_do_rfc1323 = 1; 158 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 159 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 160 161 static int tcp_tcbhashsize = 0; 162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 163 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 164 165 static int do_tcpdrain = 1; 166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 167 "Enable tcp_drain routine for extra help when low on mbufs"); 168 169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 170 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 171 172 static int icmp_may_rst = 1; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 174 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 175 176 static int tcp_isn_reseed_interval = 0; 177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 178 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 179 180 static uma_zone_t tcptw_zone; 181 static int maxtcptw; 182 183 static int 184 tcptw_auto_size(void) 185 { 186 int halfrange; 187 188 /* 189 * Max out at half the ephemeral port range so that TIME_WAIT 190 * sockets don't tie up too many ephemeral ports. 191 */ 192 if (ipport_lastauto > ipport_firstauto) 193 halfrange = (ipport_lastauto - ipport_firstauto) / 2; 194 else 195 halfrange = (ipport_firstauto - ipport_lastauto) / 2; 196 /* Protect against goofy port ranges smaller than 32. */ 197 return (imin(imax(halfrange, 32), maxsockets / 5)); 198 } 199 200 static int 201 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 202 { 203 int error, new; 204 205 if (maxtcptw == 0) 206 new = tcptw_auto_size(); 207 else 208 new = maxtcptw; 209 error = sysctl_handle_int(oidp, &new, sizeof(int), req); 210 if (error == 0 && req->newptr) 211 if (new >= 32) { 212 maxtcptw = new; 213 uma_zone_set_max(tcptw_zone, maxtcptw); 214 } 215 return (error); 216 } 217 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 218 &maxtcptw, 0, sysctl_maxtcptw, "IU", 219 "Maximum number of compressed TCP TIME_WAIT entries"); 220 221 static int nolocaltimewait = 0; 222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW, 223 &nolocaltimewait, 0, "Do not create compressed TCP TIME_WAIT entries" 224 "for local connections"); 225 226 /* 227 * TCP bandwidth limiting sysctls. Note that the default lower bound of 228 * 1024 exists only for debugging. A good production default would be 229 * something like 6100. 230 */ 231 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 232 "TCP inflight data limiting"); 233 234 static int tcp_inflight_enable = 1; 235 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 236 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 237 238 static int tcp_inflight_debug = 0; 239 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 240 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 241 242 static int tcp_inflight_rttthresh; 243 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW, 244 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I", 245 "RTT threshold below which inflight will deactivate itself"); 246 247 static int tcp_inflight_min = 6144; 248 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 249 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 250 251 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 252 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 253 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 254 255 static int tcp_inflight_stab = 20; 256 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 257 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 258 259 uma_zone_t sack_hole_zone; 260 261 static struct inpcb *tcp_notify(struct inpcb *, int); 262 static void tcp_isn_tick(void *); 263 264 /* 265 * Target size of TCP PCB hash tables. Must be a power of two. 266 * 267 * Note that this can be overridden by the kernel environment 268 * variable net.inet.tcp.tcbhashsize 269 */ 270 #ifndef TCBHASHSIZE 271 #define TCBHASHSIZE 512 272 #endif 273 274 /* 275 * XXX 276 * Callouts should be moved into struct tcp directly. They are currently 277 * separate because the tcpcb structure is exported to userland for sysctl 278 * parsing purposes, which do not know about callouts. 279 */ 280 struct tcpcb_mem { 281 struct tcpcb tcb; 282 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 283 struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 284 }; 285 286 static uma_zone_t tcpcb_zone; 287 struct callout isn_callout; 288 static struct mtx isn_mtx; 289 290 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 291 #define ISN_LOCK() mtx_lock(&isn_mtx) 292 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 293 294 /* 295 * TCP initialization. 296 */ 297 static void 298 tcp_zone_change(void *tag) 299 { 300 301 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 302 uma_zone_set_max(tcpcb_zone, maxsockets); 303 if (maxtcptw == 0) 304 uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 305 } 306 307 static int 308 tcp_inpcb_init(void *mem, int size, int flags) 309 { 310 struct inpcb *inp = (struct inpcb *) mem; 311 INP_LOCK_INIT(inp, "inp", "tcpinp"); 312 return (0); 313 } 314 315 void 316 tcp_init(void) 317 { 318 int hashsize = TCBHASHSIZE; 319 320 tcp_delacktime = TCPTV_DELACK; 321 tcp_keepinit = TCPTV_KEEP_INIT; 322 tcp_keepidle = TCPTV_KEEP_IDLE; 323 tcp_keepintvl = TCPTV_KEEPINTVL; 324 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 325 tcp_msl = TCPTV_MSL; 326 tcp_rexmit_min = TCPTV_MIN; 327 tcp_rexmit_slop = TCPTV_CPU_VAR; 328 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH; 329 330 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 331 LIST_INIT(&tcb); 332 tcbinfo.listhead = &tcb; 333 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 334 if (!powerof2(hashsize)) { 335 printf("WARNING: TCB hash size not a power of 2\n"); 336 hashsize = 512; /* safe default */ 337 } 338 tcp_tcbhashsize = hashsize; 339 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 340 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 341 &tcbinfo.porthashmask); 342 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 343 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 344 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 345 #ifdef INET6 346 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 347 #else /* INET6 */ 348 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 349 #endif /* INET6 */ 350 if (max_protohdr < TCP_MINPROTOHDR) 351 max_protohdr = TCP_MINPROTOHDR; 352 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 353 panic("tcp_init"); 354 #undef TCP_MINPROTOHDR 355 /* 356 * These have to be type stable for the benefit of the timers. 357 */ 358 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 359 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 360 uma_zone_set_max(tcpcb_zone, maxsockets); 361 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 362 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 363 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 364 if (maxtcptw == 0) 365 uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 366 else 367 uma_zone_set_max(tcptw_zone, maxtcptw); 368 tcp_timer_init(); 369 syncache_init(); 370 tcp_hc_init(); 371 tcp_reass_init(); 372 ISN_LOCK_INIT(); 373 callout_init(&isn_callout, CALLOUT_MPSAFE); 374 tcp_isn_tick(NULL); 375 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 376 SHUTDOWN_PRI_DEFAULT); 377 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 378 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 379 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 380 EVENTHANDLER_PRI_ANY); 381 } 382 383 void 384 tcp_fini(void *xtp) 385 { 386 387 callout_stop(&isn_callout); 388 } 389 390 /* 391 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 392 * tcp_template used to store this data in mbufs, but we now recopy it out 393 * of the tcpcb each time to conserve mbufs. 394 */ 395 void 396 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 397 { 398 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 399 400 INP_LOCK_ASSERT(inp); 401 402 #ifdef INET6 403 if ((inp->inp_vflag & INP_IPV6) != 0) { 404 struct ip6_hdr *ip6; 405 406 ip6 = (struct ip6_hdr *)ip_ptr; 407 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 408 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 409 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 410 (IPV6_VERSION & IPV6_VERSION_MASK); 411 ip6->ip6_nxt = IPPROTO_TCP; 412 ip6->ip6_plen = sizeof(struct tcphdr); 413 ip6->ip6_src = inp->in6p_laddr; 414 ip6->ip6_dst = inp->in6p_faddr; 415 } else 416 #endif 417 { 418 struct ip *ip; 419 420 ip = (struct ip *)ip_ptr; 421 ip->ip_v = IPVERSION; 422 ip->ip_hl = 5; 423 ip->ip_tos = inp->inp_ip_tos; 424 ip->ip_len = 0; 425 ip->ip_id = 0; 426 ip->ip_off = 0; 427 ip->ip_ttl = inp->inp_ip_ttl; 428 ip->ip_sum = 0; 429 ip->ip_p = IPPROTO_TCP; 430 ip->ip_src = inp->inp_laddr; 431 ip->ip_dst = inp->inp_faddr; 432 } 433 th->th_sport = inp->inp_lport; 434 th->th_dport = inp->inp_fport; 435 th->th_seq = 0; 436 th->th_ack = 0; 437 th->th_x2 = 0; 438 th->th_off = 5; 439 th->th_flags = 0; 440 th->th_win = 0; 441 th->th_urp = 0; 442 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 443 } 444 445 /* 446 * Create template to be used to send tcp packets on a connection. 447 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 448 * use for this function is in keepalives, which use tcp_respond. 449 */ 450 struct tcptemp * 451 tcpip_maketemplate(struct inpcb *inp) 452 { 453 struct mbuf *m; 454 struct tcptemp *n; 455 456 m = m_get(M_DONTWAIT, MT_DATA); 457 if (m == NULL) 458 return (0); 459 m->m_len = sizeof(struct tcptemp); 460 n = mtod(m, struct tcptemp *); 461 462 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 463 return (n); 464 } 465 466 /* 467 * Send a single message to the TCP at address specified by 468 * the given TCP/IP header. If m == NULL, then we make a copy 469 * of the tcpiphdr at ti and send directly to the addressed host. 470 * This is used to force keep alive messages out using the TCP 471 * template for a connection. If flags are given then we send 472 * a message back to the TCP which originated the * segment ti, 473 * and discard the mbuf containing it and any other attached mbufs. 474 * 475 * In any case the ack and sequence number of the transmitted 476 * segment are as specified by the parameters. 477 * 478 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 479 */ 480 void 481 tcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th, 482 register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) 483 { 484 register int tlen; 485 int win = 0; 486 struct ip *ip; 487 struct tcphdr *nth; 488 #ifdef INET6 489 struct ip6_hdr *ip6; 490 int isipv6; 491 #endif /* INET6 */ 492 int ipflags = 0; 493 struct inpcb *inp; 494 495 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 496 497 #ifdef INET6 498 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 499 ip6 = ipgen; 500 #endif /* INET6 */ 501 ip = ipgen; 502 503 if (tp != NULL) { 504 inp = tp->t_inpcb; 505 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 506 INP_INFO_WLOCK_ASSERT(&tcbinfo); 507 INP_LOCK_ASSERT(inp); 508 } else 509 inp = NULL; 510 511 if (tp != NULL) { 512 if (!(flags & TH_RST)) { 513 win = sbspace(&inp->inp_socket->so_rcv); 514 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 515 win = (long)TCP_MAXWIN << tp->rcv_scale; 516 } 517 } 518 if (m == NULL) { 519 m = m_gethdr(M_DONTWAIT, MT_DATA); 520 if (m == NULL) 521 return; 522 tlen = 0; 523 m->m_data += max_linkhdr; 524 #ifdef INET6 525 if (isipv6) { 526 bcopy((caddr_t)ip6, mtod(m, caddr_t), 527 sizeof(struct ip6_hdr)); 528 ip6 = mtod(m, struct ip6_hdr *); 529 nth = (struct tcphdr *)(ip6 + 1); 530 } else 531 #endif /* INET6 */ 532 { 533 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 534 ip = mtod(m, struct ip *); 535 nth = (struct tcphdr *)(ip + 1); 536 } 537 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 538 flags = TH_ACK; 539 } else { 540 m_freem(m->m_next); 541 m->m_next = NULL; 542 m->m_data = (caddr_t)ipgen; 543 /* m_len is set later */ 544 tlen = 0; 545 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 546 #ifdef INET6 547 if (isipv6) { 548 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 549 nth = (struct tcphdr *)(ip6 + 1); 550 } else 551 #endif /* INET6 */ 552 { 553 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 554 nth = (struct tcphdr *)(ip + 1); 555 } 556 if (th != nth) { 557 /* 558 * this is usually a case when an extension header 559 * exists between the IPv6 header and the 560 * TCP header. 561 */ 562 nth->th_sport = th->th_sport; 563 nth->th_dport = th->th_dport; 564 } 565 xchg(nth->th_dport, nth->th_sport, n_short); 566 #undef xchg 567 } 568 #ifdef INET6 569 if (isipv6) { 570 ip6->ip6_flow = 0; 571 ip6->ip6_vfc = IPV6_VERSION; 572 ip6->ip6_nxt = IPPROTO_TCP; 573 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 574 tlen)); 575 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 576 } else 577 #endif 578 { 579 tlen += sizeof (struct tcpiphdr); 580 ip->ip_len = tlen; 581 ip->ip_ttl = ip_defttl; 582 if (path_mtu_discovery) 583 ip->ip_off |= IP_DF; 584 } 585 m->m_len = tlen; 586 m->m_pkthdr.len = tlen; 587 m->m_pkthdr.rcvif = NULL; 588 #ifdef MAC 589 if (inp != NULL) { 590 /* 591 * Packet is associated with a socket, so allow the 592 * label of the response to reflect the socket label. 593 */ 594 INP_LOCK_ASSERT(inp); 595 mac_create_mbuf_from_inpcb(inp, m); 596 } else { 597 /* 598 * Packet is not associated with a socket, so possibly 599 * update the label in place. 600 */ 601 mac_reflect_mbuf_tcp(m); 602 } 603 #endif 604 nth->th_seq = htonl(seq); 605 nth->th_ack = htonl(ack); 606 nth->th_x2 = 0; 607 nth->th_off = sizeof (struct tcphdr) >> 2; 608 nth->th_flags = flags; 609 if (tp != NULL) 610 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 611 else 612 nth->th_win = htons((u_short)win); 613 nth->th_urp = 0; 614 #ifdef INET6 615 if (isipv6) { 616 nth->th_sum = 0; 617 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 618 sizeof(struct ip6_hdr), 619 tlen - sizeof(struct ip6_hdr)); 620 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 621 NULL, NULL); 622 } else 623 #endif /* INET6 */ 624 { 625 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 626 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 627 m->m_pkthdr.csum_flags = CSUM_TCP; 628 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 629 } 630 #ifdef TCPDEBUG 631 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 632 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 633 #endif 634 #ifdef INET6 635 if (isipv6) 636 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 637 else 638 #endif /* INET6 */ 639 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 640 } 641 642 /* 643 * Create a new TCP control block, making an 644 * empty reassembly queue and hooking it to the argument 645 * protocol control block. The `inp' parameter must have 646 * come from the zone allocator set up in tcp_init(). 647 */ 648 struct tcpcb * 649 tcp_newtcpcb(struct inpcb *inp) 650 { 651 struct tcpcb_mem *tm; 652 struct tcpcb *tp; 653 #ifdef INET6 654 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 655 #endif /* INET6 */ 656 657 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 658 if (tm == NULL) 659 return (NULL); 660 tp = &tm->tcb; 661 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 662 tp->t_maxseg = tp->t_maxopd = 663 #ifdef INET6 664 isipv6 ? tcp_v6mssdflt : 665 #endif /* INET6 */ 666 tcp_mssdflt; 667 668 /* Set up our timeouts. */ 669 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE); 670 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE); 671 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE); 672 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE); 673 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE); 674 675 if (tcp_do_rfc1323) 676 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 677 tp->sack_enable = tcp_do_sack; 678 TAILQ_INIT(&tp->snd_holes); 679 tp->t_inpcb = inp; /* XXX */ 680 /* 681 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 682 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 683 * reasonable initial retransmit time. 684 */ 685 tp->t_srtt = TCPTV_SRTTBASE; 686 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 687 tp->t_rttmin = tcp_rexmit_min; 688 tp->t_rxtcur = TCPTV_RTOBASE; 689 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 690 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 691 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 692 tp->t_rcvtime = ticks; 693 tp->t_bw_rtttime = ticks; 694 /* 695 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 696 * because the socket may be bound to an IPv6 wildcard address, 697 * which may match an IPv4-mapped IPv6 address. 698 */ 699 inp->inp_ip_ttl = ip_defttl; 700 inp->inp_ppcb = tp; 701 return (tp); /* XXX */ 702 } 703 704 /* 705 * Drop a TCP connection, reporting 706 * the specified error. If connection is synchronized, 707 * then send a RST to peer. 708 */ 709 struct tcpcb * 710 tcp_drop(struct tcpcb *tp, int errno) 711 { 712 struct socket *so = tp->t_inpcb->inp_socket; 713 714 INP_INFO_WLOCK_ASSERT(&tcbinfo); 715 INP_LOCK_ASSERT(tp->t_inpcb); 716 717 if (TCPS_HAVERCVDSYN(tp->t_state)) { 718 tp->t_state = TCPS_CLOSED; 719 (void) tcp_output(tp); 720 tcpstat.tcps_drops++; 721 } else 722 tcpstat.tcps_conndrops++; 723 if (errno == ETIMEDOUT && tp->t_softerror) 724 errno = tp->t_softerror; 725 so->so_error = errno; 726 return (tcp_close(tp)); 727 } 728 729 void 730 tcp_discardcb(struct tcpcb *tp) 731 { 732 struct tseg_qent *q; 733 struct inpcb *inp = tp->t_inpcb; 734 struct socket *so = inp->inp_socket; 735 #ifdef INET6 736 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 737 #endif /* INET6 */ 738 739 INP_LOCK_ASSERT(inp); 740 741 /* 742 * Make sure that all of our timers are stopped before we 743 * delete the PCB. 744 */ 745 callout_stop(tp->tt_rexmt); 746 callout_stop(tp->tt_persist); 747 callout_stop(tp->tt_keep); 748 callout_stop(tp->tt_2msl); 749 callout_stop(tp->tt_delack); 750 751 /* 752 * If we got enough samples through the srtt filter, 753 * save the rtt and rttvar in the routing entry. 754 * 'Enough' is arbitrarily defined as 4 rtt samples. 755 * 4 samples is enough for the srtt filter to converge 756 * to within enough % of the correct value; fewer samples 757 * and we could save a bogus rtt. The danger is not high 758 * as tcp quickly recovers from everything. 759 * XXX: Works very well but needs some more statistics! 760 */ 761 if (tp->t_rttupdated >= 4) { 762 struct hc_metrics_lite metrics; 763 u_long ssthresh; 764 765 bzero(&metrics, sizeof(metrics)); 766 /* 767 * Update the ssthresh always when the conditions below 768 * are satisfied. This gives us better new start value 769 * for the congestion avoidance for new connections. 770 * ssthresh is only set if packet loss occured on a session. 771 * 772 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 773 * being torn down. Ideally this code would not use 'so'. 774 */ 775 ssthresh = tp->snd_ssthresh; 776 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 777 /* 778 * convert the limit from user data bytes to 779 * packets then to packet data bytes. 780 */ 781 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 782 if (ssthresh < 2) 783 ssthresh = 2; 784 ssthresh *= (u_long)(tp->t_maxseg + 785 #ifdef INET6 786 (isipv6 ? sizeof (struct ip6_hdr) + 787 sizeof (struct tcphdr) : 788 #endif 789 sizeof (struct tcpiphdr) 790 #ifdef INET6 791 ) 792 #endif 793 ); 794 } else 795 ssthresh = 0; 796 metrics.rmx_ssthresh = ssthresh; 797 798 metrics.rmx_rtt = tp->t_srtt; 799 metrics.rmx_rttvar = tp->t_rttvar; 800 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 801 metrics.rmx_bandwidth = tp->snd_bandwidth; 802 metrics.rmx_cwnd = tp->snd_cwnd; 803 metrics.rmx_sendpipe = 0; 804 metrics.rmx_recvpipe = 0; 805 806 tcp_hc_update(&inp->inp_inc, &metrics); 807 } 808 809 /* free the reassembly queue, if any */ 810 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 811 LIST_REMOVE(q, tqe_q); 812 m_freem(q->tqe_m); 813 uma_zfree(tcp_reass_zone, q); 814 tp->t_segqlen--; 815 tcp_reass_qsize--; 816 } 817 tcp_free_sackholes(tp); 818 inp->inp_ppcb = NULL; 819 tp->t_inpcb = NULL; 820 uma_zfree(tcpcb_zone, tp); 821 } 822 823 /* 824 * Attempt to close a TCP control block, marking it as dropped, and freeing 825 * the socket if we hold the only reference. 826 */ 827 struct tcpcb * 828 tcp_close(struct tcpcb *tp) 829 { 830 struct inpcb *inp = tp->t_inpcb; 831 struct socket *so; 832 833 INP_INFO_WLOCK_ASSERT(&tcbinfo); 834 INP_LOCK_ASSERT(inp); 835 836 in_pcbdrop(inp); 837 tcpstat.tcps_closed++; 838 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 839 so = inp->inp_socket; 840 soisdisconnected(so); 841 if (inp->inp_vflag & INP_SOCKREF) { 842 KASSERT(so->so_state & SS_PROTOREF, 843 ("tcp_close: !SS_PROTOREF")); 844 inp->inp_vflag &= ~INP_SOCKREF; 845 INP_UNLOCK(inp); 846 ACCEPT_LOCK(); 847 SOCK_LOCK(so); 848 so->so_state &= ~SS_PROTOREF; 849 sofree(so); 850 return (NULL); 851 } 852 return (tp); 853 } 854 855 void 856 tcp_drain(void) 857 { 858 859 if (do_tcpdrain) { 860 struct inpcb *inpb; 861 struct tcpcb *tcpb; 862 struct tseg_qent *te; 863 864 /* 865 * Walk the tcpbs, if existing, and flush the reassembly queue, 866 * if there is one... 867 * XXX: The "Net/3" implementation doesn't imply that the TCP 868 * reassembly queue should be flushed, but in a situation 869 * where we're really low on mbufs, this is potentially 870 * usefull. 871 */ 872 INP_INFO_RLOCK(&tcbinfo); 873 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 874 if (inpb->inp_vflag & INP_TIMEWAIT) 875 continue; 876 INP_LOCK(inpb); 877 if ((tcpb = intotcpcb(inpb)) != NULL) { 878 while ((te = LIST_FIRST(&tcpb->t_segq)) 879 != NULL) { 880 LIST_REMOVE(te, tqe_q); 881 m_freem(te->tqe_m); 882 uma_zfree(tcp_reass_zone, te); 883 tcpb->t_segqlen--; 884 tcp_reass_qsize--; 885 } 886 tcp_clean_sackreport(tcpb); 887 } 888 INP_UNLOCK(inpb); 889 } 890 INP_INFO_RUNLOCK(&tcbinfo); 891 } 892 } 893 894 /* 895 * Notify a tcp user of an asynchronous error; 896 * store error as soft error, but wake up user 897 * (for now, won't do anything until can select for soft error). 898 * 899 * Do not wake up user since there currently is no mechanism for 900 * reporting soft errors (yet - a kqueue filter may be added). 901 */ 902 static struct inpcb * 903 tcp_notify(struct inpcb *inp, int error) 904 { 905 struct tcpcb *tp; 906 907 INP_INFO_WLOCK_ASSERT(&tcbinfo); 908 INP_LOCK_ASSERT(inp); 909 910 if ((inp->inp_vflag & INP_TIMEWAIT) || 911 (inp->inp_vflag & INP_DROPPED)) 912 return (inp); 913 914 tp = intotcpcb(inp); 915 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 916 917 /* 918 * Ignore some errors if we are hooked up. 919 * If connection hasn't completed, has retransmitted several times, 920 * and receives a second error, give up now. This is better 921 * than waiting a long time to establish a connection that 922 * can never complete. 923 */ 924 if (tp->t_state == TCPS_ESTABLISHED && 925 (error == EHOSTUNREACH || error == ENETUNREACH || 926 error == EHOSTDOWN)) { 927 return (inp); 928 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 929 tp->t_softerror) { 930 tp = tcp_drop(tp, error); 931 if (tp != NULL) 932 return (inp); 933 else 934 return (NULL); 935 } else { 936 tp->t_softerror = error; 937 return (inp); 938 } 939 #if 0 940 wakeup( &so->so_timeo); 941 sorwakeup(so); 942 sowwakeup(so); 943 #endif 944 } 945 946 static int 947 tcp_pcblist(SYSCTL_HANDLER_ARGS) 948 { 949 int error, i, n; 950 struct inpcb *inp, **inp_list; 951 inp_gen_t gencnt; 952 struct xinpgen xig; 953 954 /* 955 * The process of preparing the TCB list is too time-consuming and 956 * resource-intensive to repeat twice on every request. 957 */ 958 if (req->oldptr == NULL) { 959 n = tcbinfo.ipi_count; 960 req->oldidx = 2 * (sizeof xig) 961 + (n + n/8) * sizeof(struct xtcpcb); 962 return (0); 963 } 964 965 if (req->newptr != NULL) 966 return (EPERM); 967 968 /* 969 * OK, now we're committed to doing something. 970 */ 971 INP_INFO_RLOCK(&tcbinfo); 972 gencnt = tcbinfo.ipi_gencnt; 973 n = tcbinfo.ipi_count; 974 INP_INFO_RUNLOCK(&tcbinfo); 975 976 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 977 + n * sizeof(struct xtcpcb)); 978 if (error != 0) 979 return (error); 980 981 xig.xig_len = sizeof xig; 982 xig.xig_count = n; 983 xig.xig_gen = gencnt; 984 xig.xig_sogen = so_gencnt; 985 error = SYSCTL_OUT(req, &xig, sizeof xig); 986 if (error) 987 return (error); 988 989 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 990 if (inp_list == NULL) 991 return (ENOMEM); 992 993 INP_INFO_RLOCK(&tcbinfo); 994 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 995 inp = LIST_NEXT(inp, inp_list)) { 996 INP_LOCK(inp); 997 if (inp->inp_gencnt <= gencnt) { 998 /* 999 * XXX: This use of cr_cansee(), introduced with 1000 * TCP state changes, is not quite right, but for 1001 * now, better than nothing. 1002 */ 1003 if (inp->inp_vflag & INP_TIMEWAIT) { 1004 if (intotw(inp) != NULL) 1005 error = cr_cansee(req->td->td_ucred, 1006 intotw(inp)->tw_cred); 1007 else 1008 error = EINVAL; /* Skip this inp. */ 1009 } else 1010 error = cr_canseesocket(req->td->td_ucred, 1011 inp->inp_socket); 1012 if (error == 0) 1013 inp_list[i++] = inp; 1014 } 1015 INP_UNLOCK(inp); 1016 } 1017 INP_INFO_RUNLOCK(&tcbinfo); 1018 n = i; 1019 1020 error = 0; 1021 for (i = 0; i < n; i++) { 1022 inp = inp_list[i]; 1023 INP_LOCK(inp); 1024 if (inp->inp_gencnt <= gencnt) { 1025 struct xtcpcb xt; 1026 void *inp_ppcb; 1027 1028 bzero(&xt, sizeof(xt)); 1029 xt.xt_len = sizeof xt; 1030 /* XXX should avoid extra copy */ 1031 bcopy(inp, &xt.xt_inp, sizeof *inp); 1032 inp_ppcb = inp->inp_ppcb; 1033 if (inp_ppcb == NULL) 1034 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1035 else if (inp->inp_vflag & INP_TIMEWAIT) { 1036 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1037 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1038 } else 1039 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1040 if (inp->inp_socket != NULL) 1041 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1042 else { 1043 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1044 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1045 } 1046 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1047 INP_UNLOCK(inp); 1048 error = SYSCTL_OUT(req, &xt, sizeof xt); 1049 } else 1050 INP_UNLOCK(inp); 1051 1052 } 1053 if (!error) { 1054 /* 1055 * Give the user an updated idea of our state. 1056 * If the generation differs from what we told 1057 * her before, she knows that something happened 1058 * while we were processing this request, and it 1059 * might be necessary to retry. 1060 */ 1061 INP_INFO_RLOCK(&tcbinfo); 1062 xig.xig_gen = tcbinfo.ipi_gencnt; 1063 xig.xig_sogen = so_gencnt; 1064 xig.xig_count = tcbinfo.ipi_count; 1065 INP_INFO_RUNLOCK(&tcbinfo); 1066 error = SYSCTL_OUT(req, &xig, sizeof xig); 1067 } 1068 free(inp_list, M_TEMP); 1069 return (error); 1070 } 1071 1072 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1073 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1074 1075 static int 1076 tcp_getcred(SYSCTL_HANDLER_ARGS) 1077 { 1078 struct xucred xuc; 1079 struct sockaddr_in addrs[2]; 1080 struct inpcb *inp; 1081 int error; 1082 1083 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1084 if (error) 1085 return (error); 1086 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1087 if (error) 1088 return (error); 1089 INP_INFO_RLOCK(&tcbinfo); 1090 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1091 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1092 if (inp == NULL) { 1093 error = ENOENT; 1094 goto outunlocked; 1095 } 1096 INP_LOCK(inp); 1097 if (inp->inp_socket == NULL) { 1098 error = ENOENT; 1099 goto out; 1100 } 1101 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1102 if (error) 1103 goto out; 1104 cru2x(inp->inp_socket->so_cred, &xuc); 1105 out: 1106 INP_UNLOCK(inp); 1107 outunlocked: 1108 INP_INFO_RUNLOCK(&tcbinfo); 1109 if (error == 0) 1110 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1111 return (error); 1112 } 1113 1114 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1115 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1116 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1117 1118 #ifdef INET6 1119 static int 1120 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1121 { 1122 struct xucred xuc; 1123 struct sockaddr_in6 addrs[2]; 1124 struct inpcb *inp; 1125 int error, mapped = 0; 1126 1127 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1128 if (error) 1129 return (error); 1130 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1131 if (error) 1132 return (error); 1133 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1134 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1135 return (error); 1136 } 1137 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1138 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1139 mapped = 1; 1140 else 1141 return (EINVAL); 1142 } 1143 1144 INP_INFO_RLOCK(&tcbinfo); 1145 if (mapped == 1) 1146 inp = in_pcblookup_hash(&tcbinfo, 1147 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1148 addrs[1].sin6_port, 1149 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1150 addrs[0].sin6_port, 1151 0, NULL); 1152 else 1153 inp = in6_pcblookup_hash(&tcbinfo, 1154 &addrs[1].sin6_addr, addrs[1].sin6_port, 1155 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1156 if (inp == NULL) { 1157 error = ENOENT; 1158 goto outunlocked; 1159 } 1160 INP_LOCK(inp); 1161 if (inp->inp_socket == NULL) { 1162 error = ENOENT; 1163 goto out; 1164 } 1165 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1166 if (error) 1167 goto out; 1168 cru2x(inp->inp_socket->so_cred, &xuc); 1169 out: 1170 INP_UNLOCK(inp); 1171 outunlocked: 1172 INP_INFO_RUNLOCK(&tcbinfo); 1173 if (error == 0) 1174 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1175 return (error); 1176 } 1177 1178 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1179 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1180 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1181 #endif 1182 1183 1184 void 1185 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1186 { 1187 struct ip *ip = vip; 1188 struct tcphdr *th; 1189 struct in_addr faddr; 1190 struct inpcb *inp; 1191 struct tcpcb *tp; 1192 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1193 struct icmp *icp; 1194 struct in_conninfo inc; 1195 tcp_seq icmp_tcp_seq; 1196 int mtu; 1197 1198 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1199 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1200 return; 1201 1202 if (cmd == PRC_MSGSIZE) 1203 notify = tcp_mtudisc; 1204 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1205 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1206 notify = tcp_drop_syn_sent; 1207 /* 1208 * Redirects don't need to be handled up here. 1209 */ 1210 else if (PRC_IS_REDIRECT(cmd)) 1211 return; 1212 /* 1213 * Source quench is depreciated. 1214 */ 1215 else if (cmd == PRC_QUENCH) 1216 return; 1217 /* 1218 * Hostdead is ugly because it goes linearly through all PCBs. 1219 * XXX: We never get this from ICMP, otherwise it makes an 1220 * excellent DoS attack on machines with many connections. 1221 */ 1222 else if (cmd == PRC_HOSTDEAD) 1223 ip = NULL; 1224 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1225 return; 1226 if (ip != NULL) { 1227 icp = (struct icmp *)((caddr_t)ip 1228 - offsetof(struct icmp, icmp_ip)); 1229 th = (struct tcphdr *)((caddr_t)ip 1230 + (ip->ip_hl << 2)); 1231 INP_INFO_WLOCK(&tcbinfo); 1232 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1233 ip->ip_src, th->th_sport, 0, NULL); 1234 if (inp != NULL) { 1235 INP_LOCK(inp); 1236 if (!(inp->inp_vflag & INP_TIMEWAIT) && 1237 !(inp->inp_vflag & INP_DROPPED) && 1238 !(inp->inp_socket == NULL)) { 1239 icmp_tcp_seq = htonl(th->th_seq); 1240 tp = intotcpcb(inp); 1241 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1242 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1243 if (cmd == PRC_MSGSIZE) { 1244 /* 1245 * MTU discovery: 1246 * If we got a needfrag set the MTU 1247 * in the route to the suggested new 1248 * value (if given) and then notify. 1249 */ 1250 bzero(&inc, sizeof(inc)); 1251 inc.inc_flags = 0; /* IPv4 */ 1252 inc.inc_faddr = faddr; 1253 1254 mtu = ntohs(icp->icmp_nextmtu); 1255 /* 1256 * If no alternative MTU was 1257 * proposed, try the next smaller 1258 * one. ip->ip_len has already 1259 * been swapped in icmp_input(). 1260 */ 1261 if (!mtu) 1262 mtu = ip_next_mtu(ip->ip_len, 1263 1); 1264 if (mtu < max(296, (tcp_minmss) 1265 + sizeof(struct tcpiphdr))) 1266 mtu = 0; 1267 if (!mtu) 1268 mtu = tcp_mssdflt 1269 + sizeof(struct tcpiphdr); 1270 /* 1271 * Only cache the the MTU if it 1272 * is smaller than the interface 1273 * or route MTU. tcp_mtudisc() 1274 * will do right thing by itself. 1275 */ 1276 if (mtu <= tcp_maxmtu(&inc, NULL)) 1277 tcp_hc_updatemtu(&inc, mtu); 1278 } 1279 1280 inp = (*notify)(inp, inetctlerrmap[cmd]); 1281 } 1282 } 1283 if (inp != NULL) 1284 INP_UNLOCK(inp); 1285 } else { 1286 inc.inc_fport = th->th_dport; 1287 inc.inc_lport = th->th_sport; 1288 inc.inc_faddr = faddr; 1289 inc.inc_laddr = ip->ip_src; 1290 #ifdef INET6 1291 inc.inc_isipv6 = 0; 1292 #endif 1293 syncache_unreach(&inc, th); 1294 } 1295 INP_INFO_WUNLOCK(&tcbinfo); 1296 } else 1297 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1298 } 1299 1300 #ifdef INET6 1301 void 1302 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1303 { 1304 struct tcphdr th; 1305 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1306 struct ip6_hdr *ip6; 1307 struct mbuf *m; 1308 struct ip6ctlparam *ip6cp = NULL; 1309 const struct sockaddr_in6 *sa6_src = NULL; 1310 int off; 1311 struct tcp_portonly { 1312 u_int16_t th_sport; 1313 u_int16_t th_dport; 1314 } *thp; 1315 1316 if (sa->sa_family != AF_INET6 || 1317 sa->sa_len != sizeof(struct sockaddr_in6)) 1318 return; 1319 1320 if (cmd == PRC_MSGSIZE) 1321 notify = tcp_mtudisc; 1322 else if (!PRC_IS_REDIRECT(cmd) && 1323 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1324 return; 1325 /* Source quench is depreciated. */ 1326 else if (cmd == PRC_QUENCH) 1327 return; 1328 1329 /* if the parameter is from icmp6, decode it. */ 1330 if (d != NULL) { 1331 ip6cp = (struct ip6ctlparam *)d; 1332 m = ip6cp->ip6c_m; 1333 ip6 = ip6cp->ip6c_ip6; 1334 off = ip6cp->ip6c_off; 1335 sa6_src = ip6cp->ip6c_src; 1336 } else { 1337 m = NULL; 1338 ip6 = NULL; 1339 off = 0; /* fool gcc */ 1340 sa6_src = &sa6_any; 1341 } 1342 1343 if (ip6 != NULL) { 1344 struct in_conninfo inc; 1345 /* 1346 * XXX: We assume that when IPV6 is non NULL, 1347 * M and OFF are valid. 1348 */ 1349 1350 /* check if we can safely examine src and dst ports */ 1351 if (m->m_pkthdr.len < off + sizeof(*thp)) 1352 return; 1353 1354 bzero(&th, sizeof(th)); 1355 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1356 1357 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1358 (struct sockaddr *)ip6cp->ip6c_src, 1359 th.th_sport, cmd, NULL, notify); 1360 1361 inc.inc_fport = th.th_dport; 1362 inc.inc_lport = th.th_sport; 1363 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1364 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1365 inc.inc_isipv6 = 1; 1366 INP_INFO_WLOCK(&tcbinfo); 1367 syncache_unreach(&inc, &th); 1368 INP_INFO_WUNLOCK(&tcbinfo); 1369 } else 1370 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1371 0, cmd, NULL, notify); 1372 } 1373 #endif /* INET6 */ 1374 1375 1376 /* 1377 * Following is where TCP initial sequence number generation occurs. 1378 * 1379 * There are two places where we must use initial sequence numbers: 1380 * 1. In SYN-ACK packets. 1381 * 2. In SYN packets. 1382 * 1383 * All ISNs for SYN-ACK packets are generated by the syncache. See 1384 * tcp_syncache.c for details. 1385 * 1386 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1387 * depends on this property. In addition, these ISNs should be 1388 * unguessable so as to prevent connection hijacking. To satisfy 1389 * the requirements of this situation, the algorithm outlined in 1390 * RFC 1948 is used, with only small modifications. 1391 * 1392 * Implementation details: 1393 * 1394 * Time is based off the system timer, and is corrected so that it 1395 * increases by one megabyte per second. This allows for proper 1396 * recycling on high speed LANs while still leaving over an hour 1397 * before rollover. 1398 * 1399 * As reading the *exact* system time is too expensive to be done 1400 * whenever setting up a TCP connection, we increment the time 1401 * offset in two ways. First, a small random positive increment 1402 * is added to isn_offset for each connection that is set up. 1403 * Second, the function tcp_isn_tick fires once per clock tick 1404 * and increments isn_offset as necessary so that sequence numbers 1405 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1406 * random positive increments serve only to ensure that the same 1407 * exact sequence number is never sent out twice (as could otherwise 1408 * happen when a port is recycled in less than the system tick 1409 * interval.) 1410 * 1411 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1412 * between seeding of isn_secret. This is normally set to zero, 1413 * as reseeding should not be necessary. 1414 * 1415 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1416 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1417 * general, this means holding an exclusive (write) lock. 1418 */ 1419 1420 #define ISN_BYTES_PER_SECOND 1048576 1421 #define ISN_STATIC_INCREMENT 4096 1422 #define ISN_RANDOM_INCREMENT (4096 - 1) 1423 1424 static u_char isn_secret[32]; 1425 static int isn_last_reseed; 1426 static u_int32_t isn_offset, isn_offset_old; 1427 static MD5_CTX isn_ctx; 1428 1429 tcp_seq 1430 tcp_new_isn(struct tcpcb *tp) 1431 { 1432 u_int32_t md5_buffer[4]; 1433 tcp_seq new_isn; 1434 1435 INP_LOCK_ASSERT(tp->t_inpcb); 1436 1437 ISN_LOCK(); 1438 /* Seed if this is the first use, reseed if requested. */ 1439 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1440 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1441 < (u_int)ticks))) { 1442 read_random(&isn_secret, sizeof(isn_secret)); 1443 isn_last_reseed = ticks; 1444 } 1445 1446 /* Compute the md5 hash and return the ISN. */ 1447 MD5Init(&isn_ctx); 1448 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1449 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1450 #ifdef INET6 1451 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1452 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1453 sizeof(struct in6_addr)); 1454 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1455 sizeof(struct in6_addr)); 1456 } else 1457 #endif 1458 { 1459 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1460 sizeof(struct in_addr)); 1461 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1462 sizeof(struct in_addr)); 1463 } 1464 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1465 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1466 new_isn = (tcp_seq) md5_buffer[0]; 1467 isn_offset += ISN_STATIC_INCREMENT + 1468 (arc4random() & ISN_RANDOM_INCREMENT); 1469 new_isn += isn_offset; 1470 ISN_UNLOCK(); 1471 return (new_isn); 1472 } 1473 1474 /* 1475 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1476 * to keep time flowing at a relatively constant rate. If the random 1477 * increments have already pushed us past the projected offset, do nothing. 1478 */ 1479 static void 1480 tcp_isn_tick(void *xtp) 1481 { 1482 u_int32_t projected_offset; 1483 1484 ISN_LOCK(); 1485 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1486 1487 if (projected_offset > isn_offset) 1488 isn_offset = projected_offset; 1489 1490 isn_offset_old = isn_offset; 1491 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1492 ISN_UNLOCK(); 1493 } 1494 1495 /* 1496 * When a specific ICMP unreachable message is received and the 1497 * connection state is SYN-SENT, drop the connection. This behavior 1498 * is controlled by the icmp_may_rst sysctl. 1499 */ 1500 struct inpcb * 1501 tcp_drop_syn_sent(struct inpcb *inp, int errno) 1502 { 1503 struct tcpcb *tp; 1504 1505 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1506 INP_LOCK_ASSERT(inp); 1507 1508 if ((inp->inp_vflag & INP_TIMEWAIT) || 1509 (inp->inp_vflag & INP_DROPPED)) 1510 return (inp); 1511 1512 tp = intotcpcb(inp); 1513 if (tp->t_state != TCPS_SYN_SENT) 1514 return (inp); 1515 1516 tp = tcp_drop(tp, errno); 1517 if (tp != NULL) 1518 return (inp); 1519 else 1520 return (NULL); 1521 } 1522 1523 /* 1524 * When `need fragmentation' ICMP is received, update our idea of the MSS 1525 * based on the new value in the route. Also nudge TCP to send something, 1526 * since we know the packet we just sent was dropped. 1527 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1528 */ 1529 struct inpcb * 1530 tcp_mtudisc(struct inpcb *inp, int errno) 1531 { 1532 struct tcpcb *tp; 1533 struct socket *so = inp->inp_socket; 1534 u_int maxmtu; 1535 u_int romtu; 1536 int mss; 1537 #ifdef INET6 1538 int isipv6; 1539 #endif /* INET6 */ 1540 1541 INP_LOCK_ASSERT(inp); 1542 if ((inp->inp_vflag & INP_TIMEWAIT) || 1543 (inp->inp_vflag & INP_DROPPED)) 1544 return (inp); 1545 1546 tp = intotcpcb(inp); 1547 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1548 1549 #ifdef INET6 1550 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1551 #endif 1552 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1553 romtu = 1554 #ifdef INET6 1555 isipv6 ? tcp_maxmtu6(&inp->inp_inc, NULL) : 1556 #endif /* INET6 */ 1557 tcp_maxmtu(&inp->inp_inc, NULL); 1558 if (!maxmtu) 1559 maxmtu = romtu; 1560 else 1561 maxmtu = min(maxmtu, romtu); 1562 if (!maxmtu) { 1563 tp->t_maxopd = tp->t_maxseg = 1564 #ifdef INET6 1565 isipv6 ? tcp_v6mssdflt : 1566 #endif /* INET6 */ 1567 tcp_mssdflt; 1568 return (inp); 1569 } 1570 mss = maxmtu - 1571 #ifdef INET6 1572 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1573 #endif /* INET6 */ 1574 sizeof(struct tcpiphdr) 1575 #ifdef INET6 1576 ) 1577 #endif /* INET6 */ 1578 ; 1579 1580 /* 1581 * XXX - The above conditional probably violates the TCP 1582 * spec. The problem is that, since we don't know the 1583 * other end's MSS, we are supposed to use a conservative 1584 * default. But, if we do that, then MTU discovery will 1585 * never actually take place, because the conservative 1586 * default is much less than the MTUs typically seen 1587 * on the Internet today. For the moment, we'll sweep 1588 * this under the carpet. 1589 * 1590 * The conservative default might not actually be a problem 1591 * if the only case this occurs is when sending an initial 1592 * SYN with options and data to a host we've never talked 1593 * to before. Then, they will reply with an MSS value which 1594 * will get recorded and the new parameters should get 1595 * recomputed. For Further Study. 1596 */ 1597 if (tp->t_maxopd <= mss) 1598 return (inp); 1599 tp->t_maxopd = mss; 1600 1601 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1602 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1603 mss -= TCPOLEN_TSTAMP_APPA; 1604 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1605 if (mss > MCLBYTES) 1606 mss &= ~(MCLBYTES-1); 1607 #else 1608 if (mss > MCLBYTES) 1609 mss = mss / MCLBYTES * MCLBYTES; 1610 #endif 1611 if (so->so_snd.sb_hiwat < mss) 1612 mss = so->so_snd.sb_hiwat; 1613 1614 tp->t_maxseg = mss; 1615 1616 tcpstat.tcps_mturesent++; 1617 tp->t_rtttime = 0; 1618 tp->snd_nxt = tp->snd_una; 1619 tcp_free_sackholes(tp); 1620 tp->snd_recover = tp->snd_max; 1621 if (tp->sack_enable) 1622 EXIT_FASTRECOVERY(tp); 1623 tcp_output(tp); 1624 return (inp); 1625 } 1626 1627 /* 1628 * Look-up the routing entry to the peer of this inpcb. If no route 1629 * is found and it cannot be allocated, then return NULL. This routine 1630 * is called by TCP routines that access the rmx structure and by tcp_mss 1631 * to get the interface MTU. 1632 */ 1633 u_long 1634 tcp_maxmtu(struct in_conninfo *inc, int *flags) 1635 { 1636 struct route sro; 1637 struct sockaddr_in *dst; 1638 struct ifnet *ifp; 1639 u_long maxmtu = 0; 1640 1641 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1642 1643 bzero(&sro, sizeof(sro)); 1644 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1645 dst = (struct sockaddr_in *)&sro.ro_dst; 1646 dst->sin_family = AF_INET; 1647 dst->sin_len = sizeof(*dst); 1648 dst->sin_addr = inc->inc_faddr; 1649 rtalloc_ign(&sro, RTF_CLONING); 1650 } 1651 if (sro.ro_rt != NULL) { 1652 ifp = sro.ro_rt->rt_ifp; 1653 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1654 maxmtu = ifp->if_mtu; 1655 else 1656 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1657 1658 /* Report additional interface capabilities. */ 1659 if (flags != NULL) { 1660 if (ifp->if_capenable & IFCAP_TSO4 && 1661 ifp->if_hwassist & CSUM_TSO) 1662 *flags |= CSUM_TSO; 1663 } 1664 RTFREE(sro.ro_rt); 1665 } 1666 return (maxmtu); 1667 } 1668 1669 #ifdef INET6 1670 u_long 1671 tcp_maxmtu6(struct in_conninfo *inc, int *flags) 1672 { 1673 struct route_in6 sro6; 1674 struct ifnet *ifp; 1675 u_long maxmtu = 0; 1676 1677 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1678 1679 bzero(&sro6, sizeof(sro6)); 1680 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1681 sro6.ro_dst.sin6_family = AF_INET6; 1682 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1683 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1684 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1685 } 1686 if (sro6.ro_rt != NULL) { 1687 ifp = sro6.ro_rt->rt_ifp; 1688 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1689 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1690 else 1691 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1692 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1693 1694 /* Report additional interface capabilities. */ 1695 if (flags != NULL) { 1696 if (ifp->if_capenable & IFCAP_TSO6 && 1697 ifp->if_hwassist & CSUM_TSO) 1698 *flags |= CSUM_TSO; 1699 } 1700 RTFREE(sro6.ro_rt); 1701 } 1702 1703 return (maxmtu); 1704 } 1705 #endif /* INET6 */ 1706 1707 #ifdef IPSEC 1708 /* compute ESP/AH header size for TCP, including outer IP header. */ 1709 size_t 1710 ipsec_hdrsiz_tcp(struct tcpcb *tp) 1711 { 1712 struct inpcb *inp; 1713 struct mbuf *m; 1714 size_t hdrsiz; 1715 struct ip *ip; 1716 #ifdef INET6 1717 struct ip6_hdr *ip6; 1718 #endif 1719 struct tcphdr *th; 1720 1721 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1722 return (0); 1723 MGETHDR(m, M_DONTWAIT, MT_DATA); 1724 if (!m) 1725 return (0); 1726 1727 #ifdef INET6 1728 if ((inp->inp_vflag & INP_IPV6) != 0) { 1729 ip6 = mtod(m, struct ip6_hdr *); 1730 th = (struct tcphdr *)(ip6 + 1); 1731 m->m_pkthdr.len = m->m_len = 1732 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1733 tcpip_fillheaders(inp, ip6, th); 1734 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1735 } else 1736 #endif /* INET6 */ 1737 { 1738 ip = mtod(m, struct ip *); 1739 th = (struct tcphdr *)(ip + 1); 1740 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1741 tcpip_fillheaders(inp, ip, th); 1742 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1743 } 1744 1745 m_free(m); 1746 return (hdrsiz); 1747 } 1748 #endif /*IPSEC*/ 1749 1750 /* 1751 * Move a TCP connection into TIME_WAIT state. 1752 * tcbinfo is locked. 1753 * inp is locked, and is unlocked before returning. 1754 */ 1755 void 1756 tcp_twstart(struct tcpcb *tp) 1757 { 1758 struct tcptw *tw; 1759 struct inpcb *inp = tp->t_inpcb; 1760 int acknow; 1761 struct socket *so; 1762 1763 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1764 INP_LOCK_ASSERT(inp); 1765 1766 if (nolocaltimewait && in_localip(inp->inp_faddr)) { 1767 tp = tcp_close(tp); 1768 if (tp != NULL) 1769 INP_UNLOCK(inp); 1770 return; 1771 } 1772 1773 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1774 if (tw == NULL) { 1775 tw = tcp_timer_2msl_tw(1); 1776 if (tw == NULL) { 1777 tp = tcp_close(tp); 1778 if (tp != NULL) 1779 INP_UNLOCK(inp); 1780 return; 1781 } 1782 } 1783 tw->tw_inpcb = inp; 1784 1785 /* 1786 * Recover last window size sent. 1787 */ 1788 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1789 1790 /* 1791 * Set t_recent if timestamps are used on the connection. 1792 */ 1793 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1794 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1795 tw->t_recent = tp->ts_recent; 1796 else 1797 tw->t_recent = 0; 1798 1799 tw->snd_nxt = tp->snd_nxt; 1800 tw->rcv_nxt = tp->rcv_nxt; 1801 tw->iss = tp->iss; 1802 tw->irs = tp->irs; 1803 tw->t_starttime = tp->t_starttime; 1804 tw->tw_time = 0; 1805 1806 /* XXX 1807 * If this code will 1808 * be used for fin-wait-2 state also, then we may need 1809 * a ts_recent from the last segment. 1810 */ 1811 acknow = tp->t_flags & TF_ACKNOW; 1812 1813 /* 1814 * First, discard tcpcb state, which includes stopping its timers and 1815 * freeing it. tcp_discardcb() used to also release the inpcb, but 1816 * that work is now done in the caller. 1817 * 1818 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 1819 * and might not be needed here any longer. 1820 */ 1821 tcp_discardcb(tp); 1822 so = inp->inp_socket; 1823 soisdisconnected(so); 1824 SOCK_LOCK(so); 1825 tw->tw_cred = crhold(so->so_cred); 1826 tw->tw_so_options = so->so_options; 1827 SOCK_UNLOCK(so); 1828 if (acknow) 1829 tcp_twrespond(tw, TH_ACK); 1830 inp->inp_ppcb = tw; 1831 inp->inp_vflag |= INP_TIMEWAIT; 1832 tcp_timer_2msl_reset(tw, 0); 1833 1834 /* 1835 * If the inpcb owns the sole reference to the socket, then we can 1836 * detach and free the socket as it is not needed in time wait. 1837 */ 1838 if (inp->inp_vflag & INP_SOCKREF) { 1839 KASSERT(so->so_state & SS_PROTOREF, 1840 ("tcp_twstart: !SS_PROTOREF")); 1841 inp->inp_vflag &= ~INP_SOCKREF; 1842 INP_UNLOCK(inp); 1843 ACCEPT_LOCK(); 1844 SOCK_LOCK(so); 1845 so->so_state &= ~SS_PROTOREF; 1846 sofree(so); 1847 } else 1848 INP_UNLOCK(inp); 1849 } 1850 1851 #if 0 1852 /* 1853 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1854 * the actual rate is slightly higher due to the addition of 1855 * random positive increments. 1856 * 1857 * Most other new OSes use semi-randomized ISN values, so we 1858 * do not need to worry about them. 1859 */ 1860 #define MS_ISN_BYTES_PER_SECOND 250000 1861 1862 /* 1863 * Determine if the ISN we will generate has advanced beyond the last 1864 * sequence number used by the previous connection. If so, indicate 1865 * that it is safe to recycle this tw socket by returning 1. 1866 */ 1867 int 1868 tcp_twrecycleable(struct tcptw *tw) 1869 { 1870 tcp_seq new_iss = tw->iss; 1871 tcp_seq new_irs = tw->irs; 1872 1873 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1874 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1875 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1876 1877 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1878 return (1); 1879 else 1880 return (0); 1881 } 1882 #endif 1883 1884 void 1885 tcp_twclose(struct tcptw *tw, int reuse) 1886 { 1887 struct socket *so; 1888 struct inpcb *inp; 1889 1890 /* 1891 * At this point, we are in one of two situations: 1892 * 1893 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 1894 * all state. 1895 * 1896 * (2) We have a socket -- if we own a reference, release it and 1897 * notify the socket layer. 1898 */ 1899 inp = tw->tw_inpcb; 1900 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1901 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 1902 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1903 INP_LOCK_ASSERT(inp); 1904 1905 tw->tw_inpcb = NULL; 1906 tcp_timer_2msl_stop(tw); 1907 inp->inp_ppcb = NULL; 1908 in_pcbdrop(inp); 1909 1910 so = inp->inp_socket; 1911 if (so != NULL) { 1912 /* 1913 * If there's a socket, handle two cases: first, we own a 1914 * strong reference, which we will now release, or we don't 1915 * in which case another reference exists (XXXRW: think 1916 * about this more), and we don't need to take action. 1917 */ 1918 if (inp->inp_vflag & INP_SOCKREF) { 1919 inp->inp_vflag &= ~INP_SOCKREF; 1920 INP_UNLOCK(inp); 1921 ACCEPT_LOCK(); 1922 SOCK_LOCK(so); 1923 KASSERT(so->so_state & SS_PROTOREF, 1924 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1925 so->so_state &= ~SS_PROTOREF; 1926 sofree(so); 1927 } else { 1928 /* 1929 * If we don't own the only reference, the socket and 1930 * inpcb need to be left around to be handled by 1931 * tcp_usr_detach() later. 1932 */ 1933 INP_UNLOCK(inp); 1934 } 1935 } else { 1936 #ifdef INET6 1937 if (inp->inp_vflag & INP_IPV6PROTO) 1938 in6_pcbfree(inp); 1939 else 1940 #endif 1941 in_pcbfree(inp); 1942 } 1943 tcpstat.tcps_closed++; 1944 crfree(tw->tw_cred); 1945 tw->tw_cred = NULL; 1946 if (reuse) 1947 return; 1948 uma_zfree(tcptw_zone, tw); 1949 } 1950 1951 int 1952 tcp_twrespond(struct tcptw *tw, int flags) 1953 { 1954 struct inpcb *inp = tw->tw_inpcb; 1955 struct tcphdr *th; 1956 struct mbuf *m; 1957 struct ip *ip = NULL; 1958 u_int8_t *optp; 1959 u_int hdrlen, optlen; 1960 int error; 1961 #ifdef INET6 1962 struct ip6_hdr *ip6 = NULL; 1963 int isipv6 = inp->inp_inc.inc_isipv6; 1964 #endif 1965 1966 INP_LOCK_ASSERT(inp); 1967 1968 m = m_gethdr(M_DONTWAIT, MT_DATA); 1969 if (m == NULL) 1970 return (ENOBUFS); 1971 m->m_data += max_linkhdr; 1972 1973 #ifdef MAC 1974 mac_create_mbuf_from_inpcb(inp, m); 1975 #endif 1976 1977 #ifdef INET6 1978 if (isipv6) { 1979 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1980 ip6 = mtod(m, struct ip6_hdr *); 1981 th = (struct tcphdr *)(ip6 + 1); 1982 tcpip_fillheaders(inp, ip6, th); 1983 } else 1984 #endif 1985 { 1986 hdrlen = sizeof(struct tcpiphdr); 1987 ip = mtod(m, struct ip *); 1988 th = (struct tcphdr *)(ip + 1); 1989 tcpip_fillheaders(inp, ip, th); 1990 } 1991 optp = (u_int8_t *)(th + 1); 1992 1993 /* 1994 * Send a timestamp and echo-reply if both our side and our peer 1995 * have sent timestamps in our SYN's and this is not a RST. 1996 */ 1997 if (tw->t_recent && flags == TH_ACK) { 1998 u_int32_t *lp = (u_int32_t *)optp; 1999 2000 /* Form timestamp option as shown in appendix A of RFC 1323. */ 2001 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 2002 *lp++ = htonl(ticks); 2003 *lp = htonl(tw->t_recent); 2004 optp += TCPOLEN_TSTAMP_APPA; 2005 } 2006 2007 optlen = optp - (u_int8_t *)(th + 1); 2008 2009 m->m_len = hdrlen + optlen; 2010 m->m_pkthdr.len = m->m_len; 2011 2012 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 2013 2014 th->th_seq = htonl(tw->snd_nxt); 2015 th->th_ack = htonl(tw->rcv_nxt); 2016 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 2017 th->th_flags = flags; 2018 th->th_win = htons(tw->last_win); 2019 2020 #ifdef INET6 2021 if (isipv6) { 2022 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 2023 sizeof(struct tcphdr) + optlen); 2024 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2025 error = ip6_output(m, inp->in6p_outputopts, NULL, 2026 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 2027 } else 2028 #endif 2029 { 2030 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2031 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 2032 m->m_pkthdr.csum_flags = CSUM_TCP; 2033 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2034 ip->ip_len = m->m_pkthdr.len; 2035 if (path_mtu_discovery) 2036 ip->ip_off |= IP_DF; 2037 error = ip_output(m, inp->inp_options, NULL, 2038 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 2039 NULL, inp); 2040 } 2041 if (flags & TH_ACK) 2042 tcpstat.tcps_sndacks++; 2043 else 2044 tcpstat.tcps_sndctrl++; 2045 tcpstat.tcps_sndtotal++; 2046 return (error); 2047 } 2048 2049 /* 2050 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 2051 * 2052 * This code attempts to calculate the bandwidth-delay product as a 2053 * means of determining the optimal window size to maximize bandwidth, 2054 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 2055 * routers. This code also does a fairly good job keeping RTTs in check 2056 * across slow links like modems. We implement an algorithm which is very 2057 * similar (but not meant to be) TCP/Vegas. The code operates on the 2058 * transmitter side of a TCP connection and so only effects the transmit 2059 * side of the connection. 2060 * 2061 * BACKGROUND: TCP makes no provision for the management of buffer space 2062 * at the end points or at the intermediate routers and switches. A TCP 2063 * stream, whether using NewReno or not, will eventually buffer as 2064 * many packets as it is able and the only reason this typically works is 2065 * due to the fairly small default buffers made available for a connection 2066 * (typicaly 16K or 32K). As machines use larger windows and/or window 2067 * scaling it is now fairly easy for even a single TCP connection to blow-out 2068 * all available buffer space not only on the local interface, but on 2069 * intermediate routers and switches as well. NewReno makes a misguided 2070 * attempt to 'solve' this problem by waiting for an actual failure to occur, 2071 * then backing off, then steadily increasing the window again until another 2072 * failure occurs, ad-infinitum. This results in terrible oscillation that 2073 * is only made worse as network loads increase and the idea of intentionally 2074 * blowing out network buffers is, frankly, a terrible way to manage network 2075 * resources. 2076 * 2077 * It is far better to limit the transmit window prior to the failure 2078 * condition being achieved. There are two general ways to do this: First 2079 * you can 'scan' through different transmit window sizes and locate the 2080 * point where the RTT stops increasing, indicating that you have filled the 2081 * pipe, then scan backwards until you note that RTT stops decreasing, then 2082 * repeat ad-infinitum. This method works in principle but has severe 2083 * implementation issues due to RTT variances, timer granularity, and 2084 * instability in the algorithm which can lead to many false positives and 2085 * create oscillations as well as interact badly with other TCP streams 2086 * implementing the same algorithm. 2087 * 2088 * The second method is to limit the window to the bandwidth delay product 2089 * of the link. This is the method we implement. RTT variances and our 2090 * own manipulation of the congestion window, bwnd, can potentially 2091 * destabilize the algorithm. For this reason we have to stabilize the 2092 * elements used to calculate the window. We do this by using the minimum 2093 * observed RTT, the long term average of the observed bandwidth, and 2094 * by adding two segments worth of slop. It isn't perfect but it is able 2095 * to react to changing conditions and gives us a very stable basis on 2096 * which to extend the algorithm. 2097 */ 2098 void 2099 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2100 { 2101 u_long bw; 2102 u_long bwnd; 2103 int save_ticks; 2104 2105 INP_LOCK_ASSERT(tp->t_inpcb); 2106 2107 /* 2108 * If inflight_enable is disabled in the middle of a tcp connection, 2109 * make sure snd_bwnd is effectively disabled. 2110 */ 2111 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2112 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2113 tp->snd_bandwidth = 0; 2114 return; 2115 } 2116 2117 /* 2118 * Figure out the bandwidth. Due to the tick granularity this 2119 * is a very rough number and it MUST be averaged over a fairly 2120 * long period of time. XXX we need to take into account a link 2121 * that is not using all available bandwidth, but for now our 2122 * slop will ramp us up if this case occurs and the bandwidth later 2123 * increases. 2124 * 2125 * Note: if ticks rollover 'bw' may wind up negative. We must 2126 * effectively reset t_bw_rtttime for this case. 2127 */ 2128 save_ticks = ticks; 2129 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2130 return; 2131 2132 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2133 (save_ticks - tp->t_bw_rtttime); 2134 tp->t_bw_rtttime = save_ticks; 2135 tp->t_bw_rtseq = ack_seq; 2136 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2137 return; 2138 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2139 2140 tp->snd_bandwidth = bw; 2141 2142 /* 2143 * Calculate the semi-static bandwidth delay product, plus two maximal 2144 * segments. The additional slop puts us squarely in the sweet 2145 * spot and also handles the bandwidth run-up case and stabilization. 2146 * Without the slop we could be locking ourselves into a lower 2147 * bandwidth. 2148 * 2149 * Situations Handled: 2150 * (1) Prevents over-queueing of packets on LANs, especially on 2151 * high speed LANs, allowing larger TCP buffers to be 2152 * specified, and also does a good job preventing 2153 * over-queueing of packets over choke points like modems 2154 * (at least for the transmit side). 2155 * 2156 * (2) Is able to handle changing network loads (bandwidth 2157 * drops so bwnd drops, bandwidth increases so bwnd 2158 * increases). 2159 * 2160 * (3) Theoretically should stabilize in the face of multiple 2161 * connections implementing the same algorithm (this may need 2162 * a little work). 2163 * 2164 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2165 * be adjusted with a sysctl but typically only needs to be 2166 * on very slow connections. A value no smaller then 5 2167 * should be used, but only reduce this default if you have 2168 * no other choice. 2169 */ 2170 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2171 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2172 #undef USERTT 2173 2174 if (tcp_inflight_debug > 0) { 2175 static int ltime; 2176 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2177 ltime = ticks; 2178 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2179 tp, 2180 bw, 2181 tp->t_rttbest, 2182 tp->t_srtt, 2183 bwnd 2184 ); 2185 } 2186 } 2187 if ((long)bwnd < tcp_inflight_min) 2188 bwnd = tcp_inflight_min; 2189 if (bwnd > tcp_inflight_max) 2190 bwnd = tcp_inflight_max; 2191 if ((long)bwnd < tp->t_maxseg * 2) 2192 bwnd = tp->t_maxseg * 2; 2193 tp->snd_bwnd = bwnd; 2194 } 2195 2196 #ifdef TCP_SIGNATURE 2197 /* 2198 * Callback function invoked by m_apply() to digest TCP segment data 2199 * contained within an mbuf chain. 2200 */ 2201 static int 2202 tcp_signature_apply(void *fstate, void *data, u_int len) 2203 { 2204 2205 MD5Update(fstate, (u_char *)data, len); 2206 return (0); 2207 } 2208 2209 /* 2210 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2211 * 2212 * Parameters: 2213 * m pointer to head of mbuf chain 2214 * off0 offset to TCP header within the mbuf chain 2215 * len length of TCP segment data, excluding options 2216 * optlen length of TCP segment options 2217 * buf pointer to storage for computed MD5 digest 2218 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2219 * 2220 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2221 * When called from tcp_input(), we can be sure that th_sum has been 2222 * zeroed out and verified already. 2223 * 2224 * This function is for IPv4 use only. Calling this function with an 2225 * IPv6 packet in the mbuf chain will yield undefined results. 2226 * 2227 * Return 0 if successful, otherwise return -1. 2228 * 2229 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2230 * search with the destination IP address, and a 'magic SPI' to be 2231 * determined by the application. This is hardcoded elsewhere to 1179 2232 * right now. Another branch of this code exists which uses the SPD to 2233 * specify per-application flows but it is unstable. 2234 */ 2235 int 2236 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2237 u_char *buf, u_int direction) 2238 { 2239 union sockaddr_union dst; 2240 struct ippseudo ippseudo; 2241 MD5_CTX ctx; 2242 int doff; 2243 struct ip *ip; 2244 struct ipovly *ipovly; 2245 struct secasvar *sav; 2246 struct tcphdr *th; 2247 u_short savecsum; 2248 2249 KASSERT(m != NULL, ("NULL mbuf chain")); 2250 KASSERT(buf != NULL, ("NULL signature pointer")); 2251 2252 /* Extract the destination from the IP header in the mbuf. */ 2253 ip = mtod(m, struct ip *); 2254 bzero(&dst, sizeof(union sockaddr_union)); 2255 dst.sa.sa_len = sizeof(struct sockaddr_in); 2256 dst.sa.sa_family = AF_INET; 2257 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2258 ip->ip_src : ip->ip_dst; 2259 2260 /* Look up an SADB entry which matches the address of the peer. */ 2261 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2262 if (sav == NULL) { 2263 printf("%s: SADB lookup failed for %s\n", __func__, 2264 inet_ntoa(dst.sin.sin_addr)); 2265 return (EINVAL); 2266 } 2267 2268 MD5Init(&ctx); 2269 ipovly = (struct ipovly *)ip; 2270 th = (struct tcphdr *)((u_char *)ip + off0); 2271 doff = off0 + sizeof(struct tcphdr) + optlen; 2272 2273 /* 2274 * Step 1: Update MD5 hash with IP pseudo-header. 2275 * 2276 * XXX The ippseudo header MUST be digested in network byte order, 2277 * or else we'll fail the regression test. Assume all fields we've 2278 * been doing arithmetic on have been in host byte order. 2279 * XXX One cannot depend on ipovly->ih_len here. When called from 2280 * tcp_output(), the underlying ip_len member has not yet been set. 2281 */ 2282 ippseudo.ippseudo_src = ipovly->ih_src; 2283 ippseudo.ippseudo_dst = ipovly->ih_dst; 2284 ippseudo.ippseudo_pad = 0; 2285 ippseudo.ippseudo_p = IPPROTO_TCP; 2286 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2287 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2288 2289 /* 2290 * Step 2: Update MD5 hash with TCP header, excluding options. 2291 * The TCP checksum must be set to zero. 2292 */ 2293 savecsum = th->th_sum; 2294 th->th_sum = 0; 2295 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2296 th->th_sum = savecsum; 2297 2298 /* 2299 * Step 3: Update MD5 hash with TCP segment data. 2300 * Use m_apply() to avoid an early m_pullup(). 2301 */ 2302 if (len > 0) 2303 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2304 2305 /* 2306 * Step 4: Update MD5 hash with shared secret. 2307 */ 2308 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2309 MD5Final(buf, &ctx); 2310 2311 key_sa_recordxfer(sav, m); 2312 KEY_FREESAV(&sav); 2313 return (0); 2314 } 2315 #endif /* TCP_SIGNATURE */ 2316 2317 static int 2318 sysctl_drop(SYSCTL_HANDLER_ARGS) 2319 { 2320 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2321 struct sockaddr_storage addrs[2]; 2322 struct inpcb *inp; 2323 struct tcpcb *tp; 2324 struct tcptw *tw; 2325 struct sockaddr_in *fin, *lin; 2326 #ifdef INET6 2327 struct sockaddr_in6 *fin6, *lin6; 2328 struct in6_addr f6, l6; 2329 #endif 2330 int error; 2331 2332 inp = NULL; 2333 fin = lin = NULL; 2334 #ifdef INET6 2335 fin6 = lin6 = NULL; 2336 #endif 2337 error = 0; 2338 2339 if (req->oldptr != NULL || req->oldlen != 0) 2340 return (EINVAL); 2341 if (req->newptr == NULL) 2342 return (EPERM); 2343 if (req->newlen < sizeof(addrs)) 2344 return (ENOMEM); 2345 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2346 if (error) 2347 return (error); 2348 2349 switch (addrs[0].ss_family) { 2350 #ifdef INET6 2351 case AF_INET6: 2352 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2353 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2354 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2355 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2356 return (EINVAL); 2357 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2358 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2359 return (EINVAL); 2360 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2361 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2362 fin = (struct sockaddr_in *)&addrs[0]; 2363 lin = (struct sockaddr_in *)&addrs[1]; 2364 break; 2365 } 2366 error = sa6_embedscope(fin6, ip6_use_defzone); 2367 if (error) 2368 return (error); 2369 error = sa6_embedscope(lin6, ip6_use_defzone); 2370 if (error) 2371 return (error); 2372 break; 2373 #endif 2374 case AF_INET: 2375 fin = (struct sockaddr_in *)&addrs[0]; 2376 lin = (struct sockaddr_in *)&addrs[1]; 2377 if (fin->sin_len != sizeof(struct sockaddr_in) || 2378 lin->sin_len != sizeof(struct sockaddr_in)) 2379 return (EINVAL); 2380 break; 2381 default: 2382 return (EINVAL); 2383 } 2384 INP_INFO_WLOCK(&tcbinfo); 2385 switch (addrs[0].ss_family) { 2386 #ifdef INET6 2387 case AF_INET6: 2388 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2389 &l6, lin6->sin6_port, 0, NULL); 2390 break; 2391 #endif 2392 case AF_INET: 2393 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2394 lin->sin_addr, lin->sin_port, 0, NULL); 2395 break; 2396 } 2397 if (inp != NULL) { 2398 INP_LOCK(inp); 2399 if (inp->inp_vflag & INP_TIMEWAIT) { 2400 /* 2401 * XXXRW: There currently exists a state where an 2402 * inpcb is present, but its timewait state has been 2403 * discarded. For now, don't allow dropping of this 2404 * type of inpcb. 2405 */ 2406 tw = intotw(inp); 2407 if (tw != NULL) 2408 tcp_twclose(tw, 0); 2409 } else if (!(inp->inp_vflag & INP_DROPPED) && 2410 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2411 tp = intotcpcb(inp); 2412 tcp_drop(tp, ECONNABORTED); 2413 } 2414 INP_UNLOCK(inp); 2415 } else 2416 error = ESRCH; 2417 INP_INFO_WUNLOCK(&tcbinfo); 2418 return (error); 2419 } 2420 2421 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2422 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2423 0, sysctl_drop, "", "Drop TCP connection"); 2424