1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 30 * $FreeBSD$ 31 */ 32 33 #include "opt_compat.h" 34 #include "opt_inet.h" 35 #include "opt_inet6.h" 36 #include "opt_ipsec.h" 37 #include "opt_mac.h" 38 #include "opt_tcpdebug.h" 39 #include "opt_tcp_sack.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/callout.h> 44 #include <sys/kernel.h> 45 #include <sys/sysctl.h> 46 #include <sys/malloc.h> 47 #include <sys/mbuf.h> 48 #ifdef INET6 49 #include <sys/domain.h> 50 #endif 51 #include <sys/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 * Number of TCP segments per second we accept from remote host 141 * before we start to calculate average segment size. If average 142 * segment size drops below the minimum TCP MSS we assume a DoS 143 * attack and reset+drop the connection. Care has to be taken not to 144 * set this value too small to not kill interactive type connections 145 * (telnet, SSH) which send many small packets. 146 */ 147 int tcp_minmssoverload = TCP_MINMSSOVERLOAD; 148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW, 149 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to" 150 "be under the MINMSS Size"); 151 152 #if 0 153 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 154 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 155 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 156 #endif 157 158 int tcp_do_rfc1323 = 1; 159 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 160 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 161 162 static int tcp_tcbhashsize = 0; 163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 164 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 165 166 static int do_tcpdrain = 1; 167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 168 "Enable tcp_drain routine for extra help when low on mbufs"); 169 170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 171 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 172 173 static int icmp_may_rst = 1; 174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 175 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 176 177 static int tcp_isn_reseed_interval = 0; 178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 179 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 180 181 static uma_zone_t tcptw_zone; 182 static int maxtcptw; 183 184 static int 185 tcptw_auto_size(void) 186 { 187 int halfrange; 188 189 /* 190 * Max out at half the ephemeral port range so that TIME_WAIT 191 * sockets don't tie up too many ephemeral ports. 192 */ 193 if (ipport_lastauto > ipport_firstauto) 194 halfrange = (ipport_lastauto - ipport_firstauto) / 2; 195 else 196 halfrange = (ipport_firstauto - ipport_lastauto) / 2; 197 /* Protect against goofy port ranges smaller than 32. */ 198 return (imin(imax(halfrange, 32), maxsockets / 5)); 199 } 200 201 static int 202 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 203 { 204 int error, new; 205 206 if (maxtcptw == 0) 207 new = tcptw_auto_size(); 208 else 209 new = maxtcptw; 210 error = sysctl_handle_int(oidp, &new, sizeof(int), req); 211 if (error == 0 && req->newptr) 212 if (new >= 32) { 213 maxtcptw = new; 214 uma_zone_set_max(tcptw_zone, maxtcptw); 215 } 216 return (error); 217 } 218 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 219 &maxtcptw, 0, sysctl_maxtcptw, "IU", 220 "Maximum number of compressed TCP TIME_WAIT entries"); 221 222 static int nolocaltimewait = 0; 223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW, 224 &nolocaltimewait, 0, "Do not create compressed TCP TIME_WAIT entries" 225 "for local connections"); 226 227 /* 228 * TCP bandwidth limiting sysctls. Note that the default lower bound of 229 * 1024 exists only for debugging. A good production default would be 230 * something like 6100. 231 */ 232 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 233 "TCP inflight data limiting"); 234 235 static int tcp_inflight_enable = 1; 236 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 237 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 238 239 static int tcp_inflight_debug = 0; 240 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 241 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 242 243 static int tcp_inflight_rttthresh; 244 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW, 245 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I", 246 "RTT threshold below which inflight will deactivate itself"); 247 248 static int tcp_inflight_min = 6144; 249 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 250 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 251 252 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 253 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 254 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 255 256 static int tcp_inflight_stab = 20; 257 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 258 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 259 260 uma_zone_t sack_hole_zone; 261 262 static struct inpcb *tcp_notify(struct inpcb *, int); 263 static void tcp_isn_tick(void *); 264 265 /* 266 * Target size of TCP PCB hash tables. Must be a power of two. 267 * 268 * Note that this can be overridden by the kernel environment 269 * variable net.inet.tcp.tcbhashsize 270 */ 271 #ifndef TCBHASHSIZE 272 #define TCBHASHSIZE 512 273 #endif 274 275 /* 276 * XXX 277 * Callouts should be moved into struct tcp directly. They are currently 278 * separate because the tcpcb structure is exported to userland for sysctl 279 * parsing purposes, which do not know about callouts. 280 */ 281 struct tcpcb_mem { 282 struct tcpcb tcb; 283 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 284 struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 285 }; 286 287 static uma_zone_t tcpcb_zone; 288 struct callout isn_callout; 289 static struct mtx isn_mtx; 290 291 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 292 #define ISN_LOCK() mtx_lock(&isn_mtx) 293 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 294 295 /* 296 * TCP initialization. 297 */ 298 static void 299 tcp_zone_change(void *tag) 300 { 301 302 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 303 uma_zone_set_max(tcpcb_zone, maxsockets); 304 if (maxtcptw == 0) 305 uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 306 } 307 308 static int 309 tcp_inpcb_init(void *mem, int size, int flags) 310 { 311 struct inpcb *inp = (struct inpcb *) mem; 312 INP_LOCK_INIT(inp, "inp", "tcpinp"); 313 return (0); 314 } 315 316 void 317 tcp_init(void) 318 { 319 int hashsize = TCBHASHSIZE; 320 321 tcp_delacktime = TCPTV_DELACK; 322 tcp_keepinit = TCPTV_KEEP_INIT; 323 tcp_keepidle = TCPTV_KEEP_IDLE; 324 tcp_keepintvl = TCPTV_KEEPINTVL; 325 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 326 tcp_msl = TCPTV_MSL; 327 tcp_rexmit_min = TCPTV_MIN; 328 tcp_rexmit_slop = TCPTV_CPU_VAR; 329 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH; 330 331 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 332 LIST_INIT(&tcb); 333 tcbinfo.listhead = &tcb; 334 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 335 if (!powerof2(hashsize)) { 336 printf("WARNING: TCB hash size not a power of 2\n"); 337 hashsize = 512; /* safe default */ 338 } 339 tcp_tcbhashsize = hashsize; 340 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 341 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 342 &tcbinfo.porthashmask); 343 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 344 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 345 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 346 #ifdef INET6 347 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 348 #else /* INET6 */ 349 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 350 #endif /* INET6 */ 351 if (max_protohdr < TCP_MINPROTOHDR) 352 max_protohdr = TCP_MINPROTOHDR; 353 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 354 panic("tcp_init"); 355 #undef TCP_MINPROTOHDR 356 /* 357 * These have to be type stable for the benefit of the timers. 358 */ 359 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 360 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 361 uma_zone_set_max(tcpcb_zone, maxsockets); 362 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 363 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 364 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 365 if (maxtcptw == 0) 366 uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 367 else 368 uma_zone_set_max(tcptw_zone, maxtcptw); 369 tcp_timer_init(); 370 syncache_init(); 371 tcp_hc_init(); 372 tcp_reass_init(); 373 ISN_LOCK_INIT(); 374 callout_init(&isn_callout, CALLOUT_MPSAFE); 375 tcp_isn_tick(NULL); 376 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 377 SHUTDOWN_PRI_DEFAULT); 378 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 379 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 380 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 381 EVENTHANDLER_PRI_ANY); 382 } 383 384 void 385 tcp_fini(void *xtp) 386 { 387 388 callout_stop(&isn_callout); 389 } 390 391 /* 392 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 393 * tcp_template used to store this data in mbufs, but we now recopy it out 394 * of the tcpcb each time to conserve mbufs. 395 */ 396 void 397 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 398 { 399 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 400 401 INP_LOCK_ASSERT(inp); 402 403 #ifdef INET6 404 if ((inp->inp_vflag & INP_IPV6) != 0) { 405 struct ip6_hdr *ip6; 406 407 ip6 = (struct ip6_hdr *)ip_ptr; 408 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 409 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 410 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 411 (IPV6_VERSION & IPV6_VERSION_MASK); 412 ip6->ip6_nxt = IPPROTO_TCP; 413 ip6->ip6_plen = sizeof(struct tcphdr); 414 ip6->ip6_src = inp->in6p_laddr; 415 ip6->ip6_dst = inp->in6p_faddr; 416 } else 417 #endif 418 { 419 struct ip *ip; 420 421 ip = (struct ip *)ip_ptr; 422 ip->ip_v = IPVERSION; 423 ip->ip_hl = 5; 424 ip->ip_tos = inp->inp_ip_tos; 425 ip->ip_len = 0; 426 ip->ip_id = 0; 427 ip->ip_off = 0; 428 ip->ip_ttl = inp->inp_ip_ttl; 429 ip->ip_sum = 0; 430 ip->ip_p = IPPROTO_TCP; 431 ip->ip_src = inp->inp_laddr; 432 ip->ip_dst = inp->inp_faddr; 433 } 434 th->th_sport = inp->inp_lport; 435 th->th_dport = inp->inp_fport; 436 th->th_seq = 0; 437 th->th_ack = 0; 438 th->th_x2 = 0; 439 th->th_off = 5; 440 th->th_flags = 0; 441 th->th_win = 0; 442 th->th_urp = 0; 443 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 444 } 445 446 /* 447 * Create template to be used to send tcp packets on a connection. 448 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 449 * use for this function is in keepalives, which use tcp_respond. 450 */ 451 struct tcptemp * 452 tcpip_maketemplate(struct inpcb *inp) 453 { 454 struct mbuf *m; 455 struct tcptemp *n; 456 457 m = m_get(M_DONTWAIT, MT_DATA); 458 if (m == NULL) 459 return (0); 460 m->m_len = sizeof(struct tcptemp); 461 n = mtod(m, struct tcptemp *); 462 463 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 464 return (n); 465 } 466 467 /* 468 * Send a single message to the TCP at address specified by 469 * the given TCP/IP header. If m == NULL, then we make a copy 470 * of the tcpiphdr at ti and send directly to the addressed host. 471 * This is used to force keep alive messages out using the TCP 472 * template for a connection. If flags are given then we send 473 * a message back to the TCP which originated the * segment ti, 474 * and discard the mbuf containing it and any other attached mbufs. 475 * 476 * In any case the ack and sequence number of the transmitted 477 * segment are as specified by the parameters. 478 * 479 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 480 */ 481 void 482 tcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th, 483 register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) 484 { 485 register int tlen; 486 int win = 0; 487 struct ip *ip; 488 struct tcphdr *nth; 489 #ifdef INET6 490 struct ip6_hdr *ip6; 491 int isipv6; 492 #endif /* INET6 */ 493 int ipflags = 0; 494 struct inpcb *inp; 495 496 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 497 498 #ifdef INET6 499 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 500 ip6 = ipgen; 501 #endif /* INET6 */ 502 ip = ipgen; 503 504 if (tp != NULL) { 505 inp = tp->t_inpcb; 506 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 507 INP_INFO_WLOCK_ASSERT(&tcbinfo); 508 INP_LOCK_ASSERT(inp); 509 } else 510 inp = NULL; 511 512 if (tp != NULL) { 513 if (!(flags & TH_RST)) { 514 win = sbspace(&inp->inp_socket->so_rcv); 515 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 516 win = (long)TCP_MAXWIN << tp->rcv_scale; 517 } 518 } 519 if (m == NULL) { 520 m = m_gethdr(M_DONTWAIT, MT_DATA); 521 if (m == NULL) 522 return; 523 tlen = 0; 524 m->m_data += max_linkhdr; 525 #ifdef INET6 526 if (isipv6) { 527 bcopy((caddr_t)ip6, mtod(m, caddr_t), 528 sizeof(struct ip6_hdr)); 529 ip6 = mtod(m, struct ip6_hdr *); 530 nth = (struct tcphdr *)(ip6 + 1); 531 } else 532 #endif /* INET6 */ 533 { 534 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 535 ip = mtod(m, struct ip *); 536 nth = (struct tcphdr *)(ip + 1); 537 } 538 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 539 flags = TH_ACK; 540 } else { 541 m_freem(m->m_next); 542 m->m_next = NULL; 543 m->m_data = (caddr_t)ipgen; 544 /* m_len is set later */ 545 tlen = 0; 546 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 547 #ifdef INET6 548 if (isipv6) { 549 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 550 nth = (struct tcphdr *)(ip6 + 1); 551 } else 552 #endif /* INET6 */ 553 { 554 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 555 nth = (struct tcphdr *)(ip + 1); 556 } 557 if (th != nth) { 558 /* 559 * this is usually a case when an extension header 560 * exists between the IPv6 header and the 561 * TCP header. 562 */ 563 nth->th_sport = th->th_sport; 564 nth->th_dport = th->th_dport; 565 } 566 xchg(nth->th_dport, nth->th_sport, n_short); 567 #undef xchg 568 } 569 #ifdef INET6 570 if (isipv6) { 571 ip6->ip6_flow = 0; 572 ip6->ip6_vfc = IPV6_VERSION; 573 ip6->ip6_nxt = IPPROTO_TCP; 574 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 575 tlen)); 576 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 577 } else 578 #endif 579 { 580 tlen += sizeof (struct tcpiphdr); 581 ip->ip_len = tlen; 582 ip->ip_ttl = ip_defttl; 583 if (path_mtu_discovery) 584 ip->ip_off |= IP_DF; 585 } 586 m->m_len = tlen; 587 m->m_pkthdr.len = tlen; 588 m->m_pkthdr.rcvif = NULL; 589 #ifdef MAC 590 if (inp != NULL) { 591 /* 592 * Packet is associated with a socket, so allow the 593 * label of the response to reflect the socket label. 594 */ 595 INP_LOCK_ASSERT(inp); 596 mac_create_mbuf_from_inpcb(inp, m); 597 } else { 598 /* 599 * Packet is not associated with a socket, so possibly 600 * update the label in place. 601 */ 602 mac_reflect_mbuf_tcp(m); 603 } 604 #endif 605 nth->th_seq = htonl(seq); 606 nth->th_ack = htonl(ack); 607 nth->th_x2 = 0; 608 nth->th_off = sizeof (struct tcphdr) >> 2; 609 nth->th_flags = flags; 610 if (tp != NULL) 611 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 612 else 613 nth->th_win = htons((u_short)win); 614 nth->th_urp = 0; 615 #ifdef INET6 616 if (isipv6) { 617 nth->th_sum = 0; 618 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 619 sizeof(struct ip6_hdr), 620 tlen - sizeof(struct ip6_hdr)); 621 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 622 NULL, NULL); 623 } else 624 #endif /* INET6 */ 625 { 626 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 627 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 628 m->m_pkthdr.csum_flags = CSUM_TCP; 629 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 630 } 631 #ifdef TCPDEBUG 632 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 633 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 634 #endif 635 #ifdef INET6 636 if (isipv6) 637 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 638 else 639 #endif /* INET6 */ 640 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 641 } 642 643 /* 644 * Create a new TCP control block, making an 645 * empty reassembly queue and hooking it to the argument 646 * protocol control block. The `inp' parameter must have 647 * come from the zone allocator set up in tcp_init(). 648 */ 649 struct tcpcb * 650 tcp_newtcpcb(struct inpcb *inp) 651 { 652 struct tcpcb_mem *tm; 653 struct tcpcb *tp; 654 #ifdef INET6 655 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 656 #endif /* INET6 */ 657 658 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 659 if (tm == NULL) 660 return (NULL); 661 tp = &tm->tcb; 662 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 663 tp->t_maxseg = tp->t_maxopd = 664 #ifdef INET6 665 isipv6 ? tcp_v6mssdflt : 666 #endif /* INET6 */ 667 tcp_mssdflt; 668 669 /* Set up our timeouts. */ 670 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE); 671 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE); 672 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE); 673 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE); 674 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE); 675 676 if (tcp_do_rfc1323) 677 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 678 tp->sack_enable = tcp_do_sack; 679 TAILQ_INIT(&tp->snd_holes); 680 tp->t_inpcb = inp; /* XXX */ 681 /* 682 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 683 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 684 * reasonable initial retransmit time. 685 */ 686 tp->t_srtt = TCPTV_SRTTBASE; 687 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 688 tp->t_rttmin = tcp_rexmit_min; 689 tp->t_rxtcur = TCPTV_RTOBASE; 690 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 691 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 692 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 693 tp->t_rcvtime = ticks; 694 tp->t_bw_rtttime = ticks; 695 /* 696 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 697 * because the socket may be bound to an IPv6 wildcard address, 698 * which may match an IPv4-mapped IPv6 address. 699 */ 700 inp->inp_ip_ttl = ip_defttl; 701 inp->inp_ppcb = tp; 702 return (tp); /* XXX */ 703 } 704 705 /* 706 * Drop a TCP connection, reporting 707 * the specified error. If connection is synchronized, 708 * then send a RST to peer. 709 */ 710 struct tcpcb * 711 tcp_drop(struct tcpcb *tp, int errno) 712 { 713 struct socket *so = tp->t_inpcb->inp_socket; 714 715 INP_INFO_WLOCK_ASSERT(&tcbinfo); 716 INP_LOCK_ASSERT(tp->t_inpcb); 717 718 if (TCPS_HAVERCVDSYN(tp->t_state)) { 719 tp->t_state = TCPS_CLOSED; 720 (void) tcp_output(tp); 721 tcpstat.tcps_drops++; 722 } else 723 tcpstat.tcps_conndrops++; 724 if (errno == ETIMEDOUT && tp->t_softerror) 725 errno = tp->t_softerror; 726 so->so_error = errno; 727 return (tcp_close(tp)); 728 } 729 730 void 731 tcp_discardcb(struct tcpcb *tp) 732 { 733 struct tseg_qent *q; 734 struct inpcb *inp = tp->t_inpcb; 735 struct socket *so = inp->inp_socket; 736 #ifdef INET6 737 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 738 #endif /* INET6 */ 739 740 INP_LOCK_ASSERT(inp); 741 742 /* 743 * Make sure that all of our timers are stopped before we 744 * delete the PCB. 745 */ 746 callout_stop(tp->tt_rexmt); 747 callout_stop(tp->tt_persist); 748 callout_stop(tp->tt_keep); 749 callout_stop(tp->tt_2msl); 750 callout_stop(tp->tt_delack); 751 752 /* 753 * If we got enough samples through the srtt filter, 754 * save the rtt and rttvar in the routing entry. 755 * 'Enough' is arbitrarily defined as 4 rtt samples. 756 * 4 samples is enough for the srtt filter to converge 757 * to within enough % of the correct value; fewer samples 758 * and we could save a bogus rtt. The danger is not high 759 * as tcp quickly recovers from everything. 760 * XXX: Works very well but needs some more statistics! 761 */ 762 if (tp->t_rttupdated >= 4) { 763 struct hc_metrics_lite metrics; 764 u_long ssthresh; 765 766 bzero(&metrics, sizeof(metrics)); 767 /* 768 * Update the ssthresh always when the conditions below 769 * are satisfied. This gives us better new start value 770 * for the congestion avoidance for new connections. 771 * ssthresh is only set if packet loss occured on a session. 772 * 773 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 774 * being torn down. Ideally this code would not use 'so'. 775 */ 776 ssthresh = tp->snd_ssthresh; 777 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 778 /* 779 * convert the limit from user data bytes to 780 * packets then to packet data bytes. 781 */ 782 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 783 if (ssthresh < 2) 784 ssthresh = 2; 785 ssthresh *= (u_long)(tp->t_maxseg + 786 #ifdef INET6 787 (isipv6 ? sizeof (struct ip6_hdr) + 788 sizeof (struct tcphdr) : 789 #endif 790 sizeof (struct tcpiphdr) 791 #ifdef INET6 792 ) 793 #endif 794 ); 795 } else 796 ssthresh = 0; 797 metrics.rmx_ssthresh = ssthresh; 798 799 metrics.rmx_rtt = tp->t_srtt; 800 metrics.rmx_rttvar = tp->t_rttvar; 801 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 802 metrics.rmx_bandwidth = tp->snd_bandwidth; 803 metrics.rmx_cwnd = tp->snd_cwnd; 804 metrics.rmx_sendpipe = 0; 805 metrics.rmx_recvpipe = 0; 806 807 tcp_hc_update(&inp->inp_inc, &metrics); 808 } 809 810 /* free the reassembly queue, if any */ 811 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 812 LIST_REMOVE(q, tqe_q); 813 m_freem(q->tqe_m); 814 uma_zfree(tcp_reass_zone, q); 815 tp->t_segqlen--; 816 tcp_reass_qsize--; 817 } 818 tcp_free_sackholes(tp); 819 inp->inp_ppcb = NULL; 820 tp->t_inpcb = NULL; 821 uma_zfree(tcpcb_zone, tp); 822 } 823 824 /* 825 * Attempt to close a TCP control block, marking it as dropped, and freeing 826 * the socket if we hold the only reference. 827 */ 828 struct tcpcb * 829 tcp_close(struct tcpcb *tp) 830 { 831 struct inpcb *inp = tp->t_inpcb; 832 struct socket *so; 833 834 INP_INFO_WLOCK_ASSERT(&tcbinfo); 835 INP_LOCK_ASSERT(inp); 836 837 in_pcbdrop(inp); 838 tcpstat.tcps_closed++; 839 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 840 so = inp->inp_socket; 841 soisdisconnected(so); 842 if (inp->inp_vflag & INP_SOCKREF) { 843 KASSERT(so->so_state & SS_PROTOREF, 844 ("tcp_close: !SS_PROTOREF")); 845 inp->inp_vflag &= ~INP_SOCKREF; 846 INP_UNLOCK(inp); 847 ACCEPT_LOCK(); 848 SOCK_LOCK(so); 849 so->so_state &= ~SS_PROTOREF; 850 sofree(so); 851 return (NULL); 852 } 853 return (tp); 854 } 855 856 void 857 tcp_drain(void) 858 { 859 860 if (do_tcpdrain) { 861 struct inpcb *inpb; 862 struct tcpcb *tcpb; 863 struct tseg_qent *te; 864 865 /* 866 * Walk the tcpbs, if existing, and flush the reassembly queue, 867 * if there is one... 868 * XXX: The "Net/3" implementation doesn't imply that the TCP 869 * reassembly queue should be flushed, but in a situation 870 * where we're really low on mbufs, this is potentially 871 * usefull. 872 */ 873 INP_INFO_RLOCK(&tcbinfo); 874 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 875 if (inpb->inp_vflag & INP_TIMEWAIT) 876 continue; 877 INP_LOCK(inpb); 878 if ((tcpb = intotcpcb(inpb)) != NULL) { 879 while ((te = LIST_FIRST(&tcpb->t_segq)) 880 != NULL) { 881 LIST_REMOVE(te, tqe_q); 882 m_freem(te->tqe_m); 883 uma_zfree(tcp_reass_zone, te); 884 tcpb->t_segqlen--; 885 tcp_reass_qsize--; 886 } 887 tcp_clean_sackreport(tcpb); 888 } 889 INP_UNLOCK(inpb); 890 } 891 INP_INFO_RUNLOCK(&tcbinfo); 892 } 893 } 894 895 /* 896 * Notify a tcp user of an asynchronous error; 897 * store error as soft error, but wake up user 898 * (for now, won't do anything until can select for soft error). 899 * 900 * Do not wake up user since there currently is no mechanism for 901 * reporting soft errors (yet - a kqueue filter may be added). 902 */ 903 static struct inpcb * 904 tcp_notify(struct inpcb *inp, int error) 905 { 906 struct tcpcb *tp; 907 908 INP_INFO_WLOCK_ASSERT(&tcbinfo); 909 INP_LOCK_ASSERT(inp); 910 911 if ((inp->inp_vflag & INP_TIMEWAIT) || 912 (inp->inp_vflag & INP_DROPPED)) 913 return (inp); 914 915 tp = intotcpcb(inp); 916 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 917 918 /* 919 * Ignore some errors if we are hooked up. 920 * If connection hasn't completed, has retransmitted several times, 921 * and receives a second error, give up now. This is better 922 * than waiting a long time to establish a connection that 923 * can never complete. 924 */ 925 if (tp->t_state == TCPS_ESTABLISHED && 926 (error == EHOSTUNREACH || error == ENETUNREACH || 927 error == EHOSTDOWN)) { 928 return (inp); 929 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 930 tp->t_softerror) { 931 tp = tcp_drop(tp, error); 932 if (tp != NULL) 933 return (inp); 934 else 935 return (NULL); 936 } else { 937 tp->t_softerror = error; 938 return (inp); 939 } 940 #if 0 941 wakeup( &so->so_timeo); 942 sorwakeup(so); 943 sowwakeup(so); 944 #endif 945 } 946 947 static int 948 tcp_pcblist(SYSCTL_HANDLER_ARGS) 949 { 950 int error, i, n; 951 struct inpcb *inp, **inp_list; 952 inp_gen_t gencnt; 953 struct xinpgen xig; 954 955 /* 956 * The process of preparing the TCB list is too time-consuming and 957 * resource-intensive to repeat twice on every request. 958 */ 959 if (req->oldptr == NULL) { 960 n = tcbinfo.ipi_count; 961 req->oldidx = 2 * (sizeof xig) 962 + (n + n/8) * sizeof(struct xtcpcb); 963 return (0); 964 } 965 966 if (req->newptr != NULL) 967 return (EPERM); 968 969 /* 970 * OK, now we're committed to doing something. 971 */ 972 INP_INFO_RLOCK(&tcbinfo); 973 gencnt = tcbinfo.ipi_gencnt; 974 n = tcbinfo.ipi_count; 975 INP_INFO_RUNLOCK(&tcbinfo); 976 977 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 978 + n * sizeof(struct xtcpcb)); 979 if (error != 0) 980 return (error); 981 982 xig.xig_len = sizeof xig; 983 xig.xig_count = n; 984 xig.xig_gen = gencnt; 985 xig.xig_sogen = so_gencnt; 986 error = SYSCTL_OUT(req, &xig, sizeof xig); 987 if (error) 988 return (error); 989 990 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 991 if (inp_list == NULL) 992 return (ENOMEM); 993 994 INP_INFO_RLOCK(&tcbinfo); 995 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 996 inp = LIST_NEXT(inp, inp_list)) { 997 INP_LOCK(inp); 998 if (inp->inp_gencnt <= gencnt) { 999 /* 1000 * XXX: This use of cr_cansee(), introduced with 1001 * TCP state changes, is not quite right, but for 1002 * now, better than nothing. 1003 */ 1004 if (inp->inp_vflag & INP_TIMEWAIT) { 1005 if (intotw(inp) != NULL) 1006 error = cr_cansee(req->td->td_ucred, 1007 intotw(inp)->tw_cred); 1008 else 1009 error = EINVAL; /* Skip this inp. */ 1010 } else 1011 error = cr_canseesocket(req->td->td_ucred, 1012 inp->inp_socket); 1013 if (error == 0) 1014 inp_list[i++] = inp; 1015 } 1016 INP_UNLOCK(inp); 1017 } 1018 INP_INFO_RUNLOCK(&tcbinfo); 1019 n = i; 1020 1021 error = 0; 1022 for (i = 0; i < n; i++) { 1023 inp = inp_list[i]; 1024 INP_LOCK(inp); 1025 if (inp->inp_gencnt <= gencnt) { 1026 struct xtcpcb xt; 1027 void *inp_ppcb; 1028 1029 bzero(&xt, sizeof(xt)); 1030 xt.xt_len = sizeof xt; 1031 /* XXX should avoid extra copy */ 1032 bcopy(inp, &xt.xt_inp, sizeof *inp); 1033 inp_ppcb = inp->inp_ppcb; 1034 if (inp_ppcb == NULL) 1035 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1036 else if (inp->inp_vflag & INP_TIMEWAIT) { 1037 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1038 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1039 } else 1040 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1041 if (inp->inp_socket != NULL) 1042 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1043 else { 1044 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1045 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1046 } 1047 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1048 INP_UNLOCK(inp); 1049 error = SYSCTL_OUT(req, &xt, sizeof xt); 1050 } else 1051 INP_UNLOCK(inp); 1052 1053 } 1054 if (!error) { 1055 /* 1056 * Give the user an updated idea of our state. 1057 * If the generation differs from what we told 1058 * her before, she knows that something happened 1059 * while we were processing this request, and it 1060 * might be necessary to retry. 1061 */ 1062 INP_INFO_RLOCK(&tcbinfo); 1063 xig.xig_gen = tcbinfo.ipi_gencnt; 1064 xig.xig_sogen = so_gencnt; 1065 xig.xig_count = tcbinfo.ipi_count; 1066 INP_INFO_RUNLOCK(&tcbinfo); 1067 error = SYSCTL_OUT(req, &xig, sizeof xig); 1068 } 1069 free(inp_list, M_TEMP); 1070 return (error); 1071 } 1072 1073 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1074 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1075 1076 static int 1077 tcp_getcred(SYSCTL_HANDLER_ARGS) 1078 { 1079 struct xucred xuc; 1080 struct sockaddr_in addrs[2]; 1081 struct inpcb *inp; 1082 int error; 1083 1084 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1085 if (error) 1086 return (error); 1087 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1088 if (error) 1089 return (error); 1090 INP_INFO_RLOCK(&tcbinfo); 1091 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1092 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1093 if (inp == NULL) { 1094 error = ENOENT; 1095 goto outunlocked; 1096 } 1097 INP_LOCK(inp); 1098 if (inp->inp_socket == NULL) { 1099 error = ENOENT; 1100 goto out; 1101 } 1102 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1103 if (error) 1104 goto out; 1105 cru2x(inp->inp_socket->so_cred, &xuc); 1106 out: 1107 INP_UNLOCK(inp); 1108 outunlocked: 1109 INP_INFO_RUNLOCK(&tcbinfo); 1110 if (error == 0) 1111 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1112 return (error); 1113 } 1114 1115 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1116 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1117 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1118 1119 #ifdef INET6 1120 static int 1121 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1122 { 1123 struct xucred xuc; 1124 struct sockaddr_in6 addrs[2]; 1125 struct inpcb *inp; 1126 int error, mapped = 0; 1127 1128 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1129 if (error) 1130 return (error); 1131 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1132 if (error) 1133 return (error); 1134 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1135 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1136 return (error); 1137 } 1138 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1139 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1140 mapped = 1; 1141 else 1142 return (EINVAL); 1143 } 1144 1145 INP_INFO_RLOCK(&tcbinfo); 1146 if (mapped == 1) 1147 inp = in_pcblookup_hash(&tcbinfo, 1148 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1149 addrs[1].sin6_port, 1150 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1151 addrs[0].sin6_port, 1152 0, NULL); 1153 else 1154 inp = in6_pcblookup_hash(&tcbinfo, 1155 &addrs[1].sin6_addr, addrs[1].sin6_port, 1156 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1157 if (inp == NULL) { 1158 error = ENOENT; 1159 goto outunlocked; 1160 } 1161 INP_LOCK(inp); 1162 if (inp->inp_socket == NULL) { 1163 error = ENOENT; 1164 goto out; 1165 } 1166 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1167 if (error) 1168 goto out; 1169 cru2x(inp->inp_socket->so_cred, &xuc); 1170 out: 1171 INP_UNLOCK(inp); 1172 outunlocked: 1173 INP_INFO_RUNLOCK(&tcbinfo); 1174 if (error == 0) 1175 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1176 return (error); 1177 } 1178 1179 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1180 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1181 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1182 #endif 1183 1184 1185 void 1186 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1187 { 1188 struct ip *ip = vip; 1189 struct tcphdr *th; 1190 struct in_addr faddr; 1191 struct inpcb *inp; 1192 struct tcpcb *tp; 1193 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1194 struct icmp *icp; 1195 struct in_conninfo inc; 1196 tcp_seq icmp_tcp_seq; 1197 int mtu; 1198 1199 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1200 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1201 return; 1202 1203 if (cmd == PRC_MSGSIZE) 1204 notify = tcp_mtudisc; 1205 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1206 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1207 notify = tcp_drop_syn_sent; 1208 /* 1209 * Redirects don't need to be handled up here. 1210 */ 1211 else if (PRC_IS_REDIRECT(cmd)) 1212 return; 1213 /* 1214 * Source quench is depreciated. 1215 */ 1216 else if (cmd == PRC_QUENCH) 1217 return; 1218 /* 1219 * Hostdead is ugly because it goes linearly through all PCBs. 1220 * XXX: We never get this from ICMP, otherwise it makes an 1221 * excellent DoS attack on machines with many connections. 1222 */ 1223 else if (cmd == PRC_HOSTDEAD) 1224 ip = NULL; 1225 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1226 return; 1227 if (ip != NULL) { 1228 icp = (struct icmp *)((caddr_t)ip 1229 - offsetof(struct icmp, icmp_ip)); 1230 th = (struct tcphdr *)((caddr_t)ip 1231 + (ip->ip_hl << 2)); 1232 INP_INFO_WLOCK(&tcbinfo); 1233 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1234 ip->ip_src, th->th_sport, 0, NULL); 1235 if (inp != NULL) { 1236 INP_LOCK(inp); 1237 if (!(inp->inp_vflag & INP_TIMEWAIT) && 1238 !(inp->inp_vflag & INP_DROPPED) && 1239 !(inp->inp_socket == NULL)) { 1240 icmp_tcp_seq = htonl(th->th_seq); 1241 tp = intotcpcb(inp); 1242 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1243 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1244 if (cmd == PRC_MSGSIZE) { 1245 /* 1246 * MTU discovery: 1247 * If we got a needfrag set the MTU 1248 * in the route to the suggested new 1249 * value (if given) and then notify. 1250 */ 1251 bzero(&inc, sizeof(inc)); 1252 inc.inc_flags = 0; /* IPv4 */ 1253 inc.inc_faddr = faddr; 1254 1255 mtu = ntohs(icp->icmp_nextmtu); 1256 /* 1257 * If no alternative MTU was 1258 * proposed, try the next smaller 1259 * one. ip->ip_len has already 1260 * been swapped in icmp_input(). 1261 */ 1262 if (!mtu) 1263 mtu = ip_next_mtu(ip->ip_len, 1264 1); 1265 if (mtu < max(296, (tcp_minmss) 1266 + sizeof(struct tcpiphdr))) 1267 mtu = 0; 1268 if (!mtu) 1269 mtu = tcp_mssdflt 1270 + sizeof(struct tcpiphdr); 1271 /* 1272 * Only cache the the MTU if it 1273 * is smaller than the interface 1274 * or route MTU. tcp_mtudisc() 1275 * will do right thing by itself. 1276 */ 1277 if (mtu <= tcp_maxmtu(&inc, NULL)) 1278 tcp_hc_updatemtu(&inc, mtu); 1279 } 1280 1281 inp = (*notify)(inp, inetctlerrmap[cmd]); 1282 } 1283 } 1284 if (inp != NULL) 1285 INP_UNLOCK(inp); 1286 } else { 1287 inc.inc_fport = th->th_dport; 1288 inc.inc_lport = th->th_sport; 1289 inc.inc_faddr = faddr; 1290 inc.inc_laddr = ip->ip_src; 1291 #ifdef INET6 1292 inc.inc_isipv6 = 0; 1293 #endif 1294 syncache_unreach(&inc, th); 1295 } 1296 INP_INFO_WUNLOCK(&tcbinfo); 1297 } else 1298 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1299 } 1300 1301 #ifdef INET6 1302 void 1303 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1304 { 1305 struct tcphdr th; 1306 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1307 struct ip6_hdr *ip6; 1308 struct mbuf *m; 1309 struct ip6ctlparam *ip6cp = NULL; 1310 const struct sockaddr_in6 *sa6_src = NULL; 1311 int off; 1312 struct tcp_portonly { 1313 u_int16_t th_sport; 1314 u_int16_t th_dport; 1315 } *thp; 1316 1317 if (sa->sa_family != AF_INET6 || 1318 sa->sa_len != sizeof(struct sockaddr_in6)) 1319 return; 1320 1321 if (cmd == PRC_MSGSIZE) 1322 notify = tcp_mtudisc; 1323 else if (!PRC_IS_REDIRECT(cmd) && 1324 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1325 return; 1326 /* Source quench is depreciated. */ 1327 else if (cmd == PRC_QUENCH) 1328 return; 1329 1330 /* if the parameter is from icmp6, decode it. */ 1331 if (d != NULL) { 1332 ip6cp = (struct ip6ctlparam *)d; 1333 m = ip6cp->ip6c_m; 1334 ip6 = ip6cp->ip6c_ip6; 1335 off = ip6cp->ip6c_off; 1336 sa6_src = ip6cp->ip6c_src; 1337 } else { 1338 m = NULL; 1339 ip6 = NULL; 1340 off = 0; /* fool gcc */ 1341 sa6_src = &sa6_any; 1342 } 1343 1344 if (ip6 != NULL) { 1345 struct in_conninfo inc; 1346 /* 1347 * XXX: We assume that when IPV6 is non NULL, 1348 * M and OFF are valid. 1349 */ 1350 1351 /* check if we can safely examine src and dst ports */ 1352 if (m->m_pkthdr.len < off + sizeof(*thp)) 1353 return; 1354 1355 bzero(&th, sizeof(th)); 1356 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1357 1358 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1359 (struct sockaddr *)ip6cp->ip6c_src, 1360 th.th_sport, cmd, NULL, notify); 1361 1362 inc.inc_fport = th.th_dport; 1363 inc.inc_lport = th.th_sport; 1364 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1365 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1366 inc.inc_isipv6 = 1; 1367 INP_INFO_WLOCK(&tcbinfo); 1368 syncache_unreach(&inc, &th); 1369 INP_INFO_WUNLOCK(&tcbinfo); 1370 } else 1371 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1372 0, cmd, NULL, notify); 1373 } 1374 #endif /* INET6 */ 1375 1376 1377 /* 1378 * Following is where TCP initial sequence number generation occurs. 1379 * 1380 * There are two places where we must use initial sequence numbers: 1381 * 1. In SYN-ACK packets. 1382 * 2. In SYN packets. 1383 * 1384 * All ISNs for SYN-ACK packets are generated by the syncache. See 1385 * tcp_syncache.c for details. 1386 * 1387 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1388 * depends on this property. In addition, these ISNs should be 1389 * unguessable so as to prevent connection hijacking. To satisfy 1390 * the requirements of this situation, the algorithm outlined in 1391 * RFC 1948 is used, with only small modifications. 1392 * 1393 * Implementation details: 1394 * 1395 * Time is based off the system timer, and is corrected so that it 1396 * increases by one megabyte per second. This allows for proper 1397 * recycling on high speed LANs while still leaving over an hour 1398 * before rollover. 1399 * 1400 * As reading the *exact* system time is too expensive to be done 1401 * whenever setting up a TCP connection, we increment the time 1402 * offset in two ways. First, a small random positive increment 1403 * is added to isn_offset for each connection that is set up. 1404 * Second, the function tcp_isn_tick fires once per clock tick 1405 * and increments isn_offset as necessary so that sequence numbers 1406 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1407 * random positive increments serve only to ensure that the same 1408 * exact sequence number is never sent out twice (as could otherwise 1409 * happen when a port is recycled in less than the system tick 1410 * interval.) 1411 * 1412 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1413 * between seeding of isn_secret. This is normally set to zero, 1414 * as reseeding should not be necessary. 1415 * 1416 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1417 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1418 * general, this means holding an exclusive (write) lock. 1419 */ 1420 1421 #define ISN_BYTES_PER_SECOND 1048576 1422 #define ISN_STATIC_INCREMENT 4096 1423 #define ISN_RANDOM_INCREMENT (4096 - 1) 1424 1425 static u_char isn_secret[32]; 1426 static int isn_last_reseed; 1427 static u_int32_t isn_offset, isn_offset_old; 1428 static MD5_CTX isn_ctx; 1429 1430 tcp_seq 1431 tcp_new_isn(struct tcpcb *tp) 1432 { 1433 u_int32_t md5_buffer[4]; 1434 tcp_seq new_isn; 1435 1436 INP_LOCK_ASSERT(tp->t_inpcb); 1437 1438 ISN_LOCK(); 1439 /* Seed if this is the first use, reseed if requested. */ 1440 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1441 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1442 < (u_int)ticks))) { 1443 read_random(&isn_secret, sizeof(isn_secret)); 1444 isn_last_reseed = ticks; 1445 } 1446 1447 /* Compute the md5 hash and return the ISN. */ 1448 MD5Init(&isn_ctx); 1449 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1450 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1451 #ifdef INET6 1452 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1453 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1454 sizeof(struct in6_addr)); 1455 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1456 sizeof(struct in6_addr)); 1457 } else 1458 #endif 1459 { 1460 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1461 sizeof(struct in_addr)); 1462 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1463 sizeof(struct in_addr)); 1464 } 1465 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1466 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1467 new_isn = (tcp_seq) md5_buffer[0]; 1468 isn_offset += ISN_STATIC_INCREMENT + 1469 (arc4random() & ISN_RANDOM_INCREMENT); 1470 new_isn += isn_offset; 1471 ISN_UNLOCK(); 1472 return (new_isn); 1473 } 1474 1475 /* 1476 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1477 * to keep time flowing at a relatively constant rate. If the random 1478 * increments have already pushed us past the projected offset, do nothing. 1479 */ 1480 static void 1481 tcp_isn_tick(void *xtp) 1482 { 1483 u_int32_t projected_offset; 1484 1485 ISN_LOCK(); 1486 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1487 1488 if (projected_offset > isn_offset) 1489 isn_offset = projected_offset; 1490 1491 isn_offset_old = isn_offset; 1492 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1493 ISN_UNLOCK(); 1494 } 1495 1496 /* 1497 * When a specific ICMP unreachable message is received and the 1498 * connection state is SYN-SENT, drop the connection. This behavior 1499 * is controlled by the icmp_may_rst sysctl. 1500 */ 1501 struct inpcb * 1502 tcp_drop_syn_sent(struct inpcb *inp, int errno) 1503 { 1504 struct tcpcb *tp; 1505 1506 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1507 INP_LOCK_ASSERT(inp); 1508 1509 if ((inp->inp_vflag & INP_TIMEWAIT) || 1510 (inp->inp_vflag & INP_DROPPED)) 1511 return (inp); 1512 1513 tp = intotcpcb(inp); 1514 if (tp->t_state != TCPS_SYN_SENT) 1515 return (inp); 1516 1517 tp = tcp_drop(tp, errno); 1518 if (tp != NULL) 1519 return (inp); 1520 else 1521 return (NULL); 1522 } 1523 1524 /* 1525 * When `need fragmentation' ICMP is received, update our idea of the MSS 1526 * based on the new value in the route. Also nudge TCP to send something, 1527 * since we know the packet we just sent was dropped. 1528 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1529 */ 1530 struct inpcb * 1531 tcp_mtudisc(struct inpcb *inp, int errno) 1532 { 1533 struct tcpcb *tp; 1534 struct socket *so = inp->inp_socket; 1535 u_int maxmtu; 1536 u_int romtu; 1537 int mss; 1538 #ifdef INET6 1539 int isipv6; 1540 #endif /* INET6 */ 1541 1542 INP_LOCK_ASSERT(inp); 1543 if ((inp->inp_vflag & INP_TIMEWAIT) || 1544 (inp->inp_vflag & INP_DROPPED)) 1545 return (inp); 1546 1547 tp = intotcpcb(inp); 1548 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1549 1550 #ifdef INET6 1551 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1552 #endif 1553 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1554 romtu = 1555 #ifdef INET6 1556 isipv6 ? tcp_maxmtu6(&inp->inp_inc, NULL) : 1557 #endif /* INET6 */ 1558 tcp_maxmtu(&inp->inp_inc, NULL); 1559 if (!maxmtu) 1560 maxmtu = romtu; 1561 else 1562 maxmtu = min(maxmtu, romtu); 1563 if (!maxmtu) { 1564 tp->t_maxopd = tp->t_maxseg = 1565 #ifdef INET6 1566 isipv6 ? tcp_v6mssdflt : 1567 #endif /* INET6 */ 1568 tcp_mssdflt; 1569 return (inp); 1570 } 1571 mss = maxmtu - 1572 #ifdef INET6 1573 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1574 #endif /* INET6 */ 1575 sizeof(struct tcpiphdr) 1576 #ifdef INET6 1577 ) 1578 #endif /* INET6 */ 1579 ; 1580 1581 /* 1582 * XXX - The above conditional probably violates the TCP 1583 * spec. The problem is that, since we don't know the 1584 * other end's MSS, we are supposed to use a conservative 1585 * default. But, if we do that, then MTU discovery will 1586 * never actually take place, because the conservative 1587 * default is much less than the MTUs typically seen 1588 * on the Internet today. For the moment, we'll sweep 1589 * this under the carpet. 1590 * 1591 * The conservative default might not actually be a problem 1592 * if the only case this occurs is when sending an initial 1593 * SYN with options and data to a host we've never talked 1594 * to before. Then, they will reply with an MSS value which 1595 * will get recorded and the new parameters should get 1596 * recomputed. For Further Study. 1597 */ 1598 if (tp->t_maxopd <= mss) 1599 return (inp); 1600 tp->t_maxopd = mss; 1601 1602 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1603 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1604 mss -= TCPOLEN_TSTAMP_APPA; 1605 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1606 if (mss > MCLBYTES) 1607 mss &= ~(MCLBYTES-1); 1608 #else 1609 if (mss > MCLBYTES) 1610 mss = mss / MCLBYTES * MCLBYTES; 1611 #endif 1612 if (so->so_snd.sb_hiwat < mss) 1613 mss = so->so_snd.sb_hiwat; 1614 1615 tp->t_maxseg = mss; 1616 1617 tcpstat.tcps_mturesent++; 1618 tp->t_rtttime = 0; 1619 tp->snd_nxt = tp->snd_una; 1620 tcp_free_sackholes(tp); 1621 tp->snd_recover = tp->snd_max; 1622 if (tp->sack_enable) 1623 EXIT_FASTRECOVERY(tp); 1624 tcp_output(tp); 1625 return (inp); 1626 } 1627 1628 /* 1629 * Look-up the routing entry to the peer of this inpcb. If no route 1630 * is found and it cannot be allocated, then return NULL. This routine 1631 * is called by TCP routines that access the rmx structure and by tcp_mss 1632 * to get the interface MTU. 1633 */ 1634 u_long 1635 tcp_maxmtu(struct in_conninfo *inc, int *flags) 1636 { 1637 struct route sro; 1638 struct sockaddr_in *dst; 1639 struct ifnet *ifp; 1640 u_long maxmtu = 0; 1641 1642 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1643 1644 bzero(&sro, sizeof(sro)); 1645 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1646 dst = (struct sockaddr_in *)&sro.ro_dst; 1647 dst->sin_family = AF_INET; 1648 dst->sin_len = sizeof(*dst); 1649 dst->sin_addr = inc->inc_faddr; 1650 rtalloc_ign(&sro, RTF_CLONING); 1651 } 1652 if (sro.ro_rt != NULL) { 1653 ifp = sro.ro_rt->rt_ifp; 1654 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1655 maxmtu = ifp->if_mtu; 1656 else 1657 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1658 1659 /* Report additional interface capabilities. */ 1660 if (flags != NULL) { 1661 if (ifp->if_capenable & IFCAP_TSO4 && 1662 ifp->if_hwassist & CSUM_TSO) 1663 *flags |= CSUM_TSO; 1664 } 1665 RTFREE(sro.ro_rt); 1666 } 1667 return (maxmtu); 1668 } 1669 1670 #ifdef INET6 1671 u_long 1672 tcp_maxmtu6(struct in_conninfo *inc, int *flags) 1673 { 1674 struct route_in6 sro6; 1675 struct ifnet *ifp; 1676 u_long maxmtu = 0; 1677 1678 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1679 1680 bzero(&sro6, sizeof(sro6)); 1681 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1682 sro6.ro_dst.sin6_family = AF_INET6; 1683 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1684 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1685 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1686 } 1687 if (sro6.ro_rt != NULL) { 1688 ifp = sro6.ro_rt->rt_ifp; 1689 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1690 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1691 else 1692 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1693 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1694 1695 /* Report additional interface capabilities. */ 1696 if (flags != NULL) { 1697 if (ifp->if_capenable & IFCAP_TSO6 && 1698 ifp->if_hwassist & CSUM_TSO) 1699 *flags |= CSUM_TSO; 1700 } 1701 RTFREE(sro6.ro_rt); 1702 } 1703 1704 return (maxmtu); 1705 } 1706 #endif /* INET6 */ 1707 1708 #ifdef IPSEC 1709 /* compute ESP/AH header size for TCP, including outer IP header. */ 1710 size_t 1711 ipsec_hdrsiz_tcp(struct tcpcb *tp) 1712 { 1713 struct inpcb *inp; 1714 struct mbuf *m; 1715 size_t hdrsiz; 1716 struct ip *ip; 1717 #ifdef INET6 1718 struct ip6_hdr *ip6; 1719 #endif 1720 struct tcphdr *th; 1721 1722 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1723 return (0); 1724 MGETHDR(m, M_DONTWAIT, MT_DATA); 1725 if (!m) 1726 return (0); 1727 1728 #ifdef INET6 1729 if ((inp->inp_vflag & INP_IPV6) != 0) { 1730 ip6 = mtod(m, struct ip6_hdr *); 1731 th = (struct tcphdr *)(ip6 + 1); 1732 m->m_pkthdr.len = m->m_len = 1733 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1734 tcpip_fillheaders(inp, ip6, th); 1735 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1736 } else 1737 #endif /* INET6 */ 1738 { 1739 ip = mtod(m, struct ip *); 1740 th = (struct tcphdr *)(ip + 1); 1741 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1742 tcpip_fillheaders(inp, ip, th); 1743 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1744 } 1745 1746 m_free(m); 1747 return (hdrsiz); 1748 } 1749 #endif /*IPSEC*/ 1750 1751 /* 1752 * Move a TCP connection into TIME_WAIT state. 1753 * tcbinfo is locked. 1754 * inp is locked, and is unlocked before returning. 1755 */ 1756 void 1757 tcp_twstart(struct tcpcb *tp) 1758 { 1759 struct tcptw *tw; 1760 struct inpcb *inp = tp->t_inpcb; 1761 int acknow; 1762 struct socket *so; 1763 1764 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1765 INP_LOCK_ASSERT(inp); 1766 1767 if (nolocaltimewait && in_localip(inp->inp_faddr)) { 1768 tp = tcp_close(tp); 1769 if (tp != NULL) 1770 INP_UNLOCK(inp); 1771 return; 1772 } 1773 1774 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1775 if (tw == NULL) { 1776 tw = tcp_timer_2msl_tw(1); 1777 if (tw == NULL) { 1778 tp = tcp_close(tp); 1779 if (tp != NULL) 1780 INP_UNLOCK(inp); 1781 return; 1782 } 1783 } 1784 tw->tw_inpcb = inp; 1785 1786 /* 1787 * Recover last window size sent. 1788 */ 1789 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1790 1791 /* 1792 * Set t_recent if timestamps are used on the connection. 1793 */ 1794 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1795 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1796 tw->t_recent = tp->ts_recent; 1797 else 1798 tw->t_recent = 0; 1799 1800 tw->snd_nxt = tp->snd_nxt; 1801 tw->rcv_nxt = tp->rcv_nxt; 1802 tw->iss = tp->iss; 1803 tw->irs = tp->irs; 1804 tw->t_starttime = tp->t_starttime; 1805 tw->tw_time = 0; 1806 1807 /* XXX 1808 * If this code will 1809 * be used for fin-wait-2 state also, then we may need 1810 * a ts_recent from the last segment. 1811 */ 1812 acknow = tp->t_flags & TF_ACKNOW; 1813 1814 /* 1815 * First, discard tcpcb state, which includes stopping its timers and 1816 * freeing it. tcp_discardcb() used to also release the inpcb, but 1817 * that work is now done in the caller. 1818 * 1819 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 1820 * and might not be needed here any longer. 1821 */ 1822 tcp_discardcb(tp); 1823 so = inp->inp_socket; 1824 soisdisconnected(so); 1825 SOCK_LOCK(so); 1826 tw->tw_cred = crhold(so->so_cred); 1827 tw->tw_so_options = so->so_options; 1828 SOCK_UNLOCK(so); 1829 if (acknow) 1830 tcp_twrespond(tw, TH_ACK); 1831 inp->inp_ppcb = tw; 1832 inp->inp_vflag |= INP_TIMEWAIT; 1833 tcp_timer_2msl_reset(tw, 0); 1834 1835 /* 1836 * If the inpcb owns the sole reference to the socket, then we can 1837 * detach and free the socket as it is not needed in time wait. 1838 */ 1839 if (inp->inp_vflag & INP_SOCKREF) { 1840 KASSERT(so->so_state & SS_PROTOREF, 1841 ("tcp_twstart: !SS_PROTOREF")); 1842 inp->inp_vflag &= ~INP_SOCKREF; 1843 INP_UNLOCK(inp); 1844 ACCEPT_LOCK(); 1845 SOCK_LOCK(so); 1846 so->so_state &= ~SS_PROTOREF; 1847 sofree(so); 1848 } else 1849 INP_UNLOCK(inp); 1850 } 1851 1852 #if 0 1853 /* 1854 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1855 * the actual rate is slightly higher due to the addition of 1856 * random positive increments. 1857 * 1858 * Most other new OSes use semi-randomized ISN values, so we 1859 * do not need to worry about them. 1860 */ 1861 #define MS_ISN_BYTES_PER_SECOND 250000 1862 1863 /* 1864 * Determine if the ISN we will generate has advanced beyond the last 1865 * sequence number used by the previous connection. If so, indicate 1866 * that it is safe to recycle this tw socket by returning 1. 1867 */ 1868 int 1869 tcp_twrecycleable(struct tcptw *tw) 1870 { 1871 tcp_seq new_iss = tw->iss; 1872 tcp_seq new_irs = tw->irs; 1873 1874 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1875 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1876 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1877 1878 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1879 return (1); 1880 else 1881 return (0); 1882 } 1883 #endif 1884 1885 void 1886 tcp_twclose(struct tcptw *tw, int reuse) 1887 { 1888 struct socket *so; 1889 struct inpcb *inp; 1890 1891 /* 1892 * At this point, we are in one of two situations: 1893 * 1894 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 1895 * all state. 1896 * 1897 * (2) We have a socket -- if we own a reference, release it and 1898 * notify the socket layer. 1899 */ 1900 inp = tw->tw_inpcb; 1901 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1902 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 1903 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1904 INP_LOCK_ASSERT(inp); 1905 1906 tw->tw_inpcb = NULL; 1907 tcp_timer_2msl_stop(tw); 1908 inp->inp_ppcb = NULL; 1909 in_pcbdrop(inp); 1910 1911 so = inp->inp_socket; 1912 if (so != NULL) { 1913 /* 1914 * If there's a socket, handle two cases: first, we own a 1915 * strong reference, which we will now release, or we don't 1916 * in which case another reference exists (XXXRW: think 1917 * about this more), and we don't need to take action. 1918 */ 1919 if (inp->inp_vflag & INP_SOCKREF) { 1920 inp->inp_vflag &= ~INP_SOCKREF; 1921 INP_UNLOCK(inp); 1922 ACCEPT_LOCK(); 1923 SOCK_LOCK(so); 1924 KASSERT(so->so_state & SS_PROTOREF, 1925 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1926 so->so_state &= ~SS_PROTOREF; 1927 sofree(so); 1928 } else { 1929 /* 1930 * If we don't own the only reference, the socket and 1931 * inpcb need to be left around to be handled by 1932 * tcp_usr_detach() later. 1933 */ 1934 INP_UNLOCK(inp); 1935 } 1936 } else { 1937 #ifdef INET6 1938 if (inp->inp_vflag & INP_IPV6PROTO) 1939 in6_pcbfree(inp); 1940 else 1941 #endif 1942 in_pcbfree(inp); 1943 } 1944 tcpstat.tcps_closed++; 1945 crfree(tw->tw_cred); 1946 tw->tw_cred = NULL; 1947 if (reuse) 1948 return; 1949 uma_zfree(tcptw_zone, tw); 1950 } 1951 1952 int 1953 tcp_twrespond(struct tcptw *tw, int flags) 1954 { 1955 struct inpcb *inp = tw->tw_inpcb; 1956 struct tcphdr *th; 1957 struct mbuf *m; 1958 struct ip *ip = NULL; 1959 u_int8_t *optp; 1960 u_int hdrlen, optlen; 1961 int error; 1962 #ifdef INET6 1963 struct ip6_hdr *ip6 = NULL; 1964 int isipv6 = inp->inp_inc.inc_isipv6; 1965 #endif 1966 1967 INP_LOCK_ASSERT(inp); 1968 1969 m = m_gethdr(M_DONTWAIT, MT_DATA); 1970 if (m == NULL) 1971 return (ENOBUFS); 1972 m->m_data += max_linkhdr; 1973 1974 #ifdef MAC 1975 mac_create_mbuf_from_inpcb(inp, m); 1976 #endif 1977 1978 #ifdef INET6 1979 if (isipv6) { 1980 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1981 ip6 = mtod(m, struct ip6_hdr *); 1982 th = (struct tcphdr *)(ip6 + 1); 1983 tcpip_fillheaders(inp, ip6, th); 1984 } else 1985 #endif 1986 { 1987 hdrlen = sizeof(struct tcpiphdr); 1988 ip = mtod(m, struct ip *); 1989 th = (struct tcphdr *)(ip + 1); 1990 tcpip_fillheaders(inp, ip, th); 1991 } 1992 optp = (u_int8_t *)(th + 1); 1993 1994 /* 1995 * Send a timestamp and echo-reply if both our side and our peer 1996 * have sent timestamps in our SYN's and this is not a RST. 1997 */ 1998 if (tw->t_recent && flags == TH_ACK) { 1999 u_int32_t *lp = (u_int32_t *)optp; 2000 2001 /* Form timestamp option as shown in appendix A of RFC 1323. */ 2002 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 2003 *lp++ = htonl(ticks); 2004 *lp = htonl(tw->t_recent); 2005 optp += TCPOLEN_TSTAMP_APPA; 2006 } 2007 2008 optlen = optp - (u_int8_t *)(th + 1); 2009 2010 m->m_len = hdrlen + optlen; 2011 m->m_pkthdr.len = m->m_len; 2012 2013 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 2014 2015 th->th_seq = htonl(tw->snd_nxt); 2016 th->th_ack = htonl(tw->rcv_nxt); 2017 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 2018 th->th_flags = flags; 2019 th->th_win = htons(tw->last_win); 2020 2021 #ifdef INET6 2022 if (isipv6) { 2023 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 2024 sizeof(struct tcphdr) + optlen); 2025 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2026 error = ip6_output(m, inp->in6p_outputopts, NULL, 2027 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 2028 } else 2029 #endif 2030 { 2031 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2032 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 2033 m->m_pkthdr.csum_flags = CSUM_TCP; 2034 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2035 ip->ip_len = m->m_pkthdr.len; 2036 if (path_mtu_discovery) 2037 ip->ip_off |= IP_DF; 2038 error = ip_output(m, inp->inp_options, NULL, 2039 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 2040 NULL, inp); 2041 } 2042 if (flags & TH_ACK) 2043 tcpstat.tcps_sndacks++; 2044 else 2045 tcpstat.tcps_sndctrl++; 2046 tcpstat.tcps_sndtotal++; 2047 return (error); 2048 } 2049 2050 /* 2051 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 2052 * 2053 * This code attempts to calculate the bandwidth-delay product as a 2054 * means of determining the optimal window size to maximize bandwidth, 2055 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 2056 * routers. This code also does a fairly good job keeping RTTs in check 2057 * across slow links like modems. We implement an algorithm which is very 2058 * similar (but not meant to be) TCP/Vegas. The code operates on the 2059 * transmitter side of a TCP connection and so only effects the transmit 2060 * side of the connection. 2061 * 2062 * BACKGROUND: TCP makes no provision for the management of buffer space 2063 * at the end points or at the intermediate routers and switches. A TCP 2064 * stream, whether using NewReno or not, will eventually buffer as 2065 * many packets as it is able and the only reason this typically works is 2066 * due to the fairly small default buffers made available for a connection 2067 * (typicaly 16K or 32K). As machines use larger windows and/or window 2068 * scaling it is now fairly easy for even a single TCP connection to blow-out 2069 * all available buffer space not only on the local interface, but on 2070 * intermediate routers and switches as well. NewReno makes a misguided 2071 * attempt to 'solve' this problem by waiting for an actual failure to occur, 2072 * then backing off, then steadily increasing the window again until another 2073 * failure occurs, ad-infinitum. This results in terrible oscillation that 2074 * is only made worse as network loads increase and the idea of intentionally 2075 * blowing out network buffers is, frankly, a terrible way to manage network 2076 * resources. 2077 * 2078 * It is far better to limit the transmit window prior to the failure 2079 * condition being achieved. There are two general ways to do this: First 2080 * you can 'scan' through different transmit window sizes and locate the 2081 * point where the RTT stops increasing, indicating that you have filled the 2082 * pipe, then scan backwards until you note that RTT stops decreasing, then 2083 * repeat ad-infinitum. This method works in principle but has severe 2084 * implementation issues due to RTT variances, timer granularity, and 2085 * instability in the algorithm which can lead to many false positives and 2086 * create oscillations as well as interact badly with other TCP streams 2087 * implementing the same algorithm. 2088 * 2089 * The second method is to limit the window to the bandwidth delay product 2090 * of the link. This is the method we implement. RTT variances and our 2091 * own manipulation of the congestion window, bwnd, can potentially 2092 * destabilize the algorithm. For this reason we have to stabilize the 2093 * elements used to calculate the window. We do this by using the minimum 2094 * observed RTT, the long term average of the observed bandwidth, and 2095 * by adding two segments worth of slop. It isn't perfect but it is able 2096 * to react to changing conditions and gives us a very stable basis on 2097 * which to extend the algorithm. 2098 */ 2099 void 2100 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2101 { 2102 u_long bw; 2103 u_long bwnd; 2104 int save_ticks; 2105 2106 INP_LOCK_ASSERT(tp->t_inpcb); 2107 2108 /* 2109 * If inflight_enable is disabled in the middle of a tcp connection, 2110 * make sure snd_bwnd is effectively disabled. 2111 */ 2112 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2113 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2114 tp->snd_bandwidth = 0; 2115 return; 2116 } 2117 2118 /* 2119 * Figure out the bandwidth. Due to the tick granularity this 2120 * is a very rough number and it MUST be averaged over a fairly 2121 * long period of time. XXX we need to take into account a link 2122 * that is not using all available bandwidth, but for now our 2123 * slop will ramp us up if this case occurs and the bandwidth later 2124 * increases. 2125 * 2126 * Note: if ticks rollover 'bw' may wind up negative. We must 2127 * effectively reset t_bw_rtttime for this case. 2128 */ 2129 save_ticks = ticks; 2130 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2131 return; 2132 2133 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2134 (save_ticks - tp->t_bw_rtttime); 2135 tp->t_bw_rtttime = save_ticks; 2136 tp->t_bw_rtseq = ack_seq; 2137 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2138 return; 2139 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2140 2141 tp->snd_bandwidth = bw; 2142 2143 /* 2144 * Calculate the semi-static bandwidth delay product, plus two maximal 2145 * segments. The additional slop puts us squarely in the sweet 2146 * spot and also handles the bandwidth run-up case and stabilization. 2147 * Without the slop we could be locking ourselves into a lower 2148 * bandwidth. 2149 * 2150 * Situations Handled: 2151 * (1) Prevents over-queueing of packets on LANs, especially on 2152 * high speed LANs, allowing larger TCP buffers to be 2153 * specified, and also does a good job preventing 2154 * over-queueing of packets over choke points like modems 2155 * (at least for the transmit side). 2156 * 2157 * (2) Is able to handle changing network loads (bandwidth 2158 * drops so bwnd drops, bandwidth increases so bwnd 2159 * increases). 2160 * 2161 * (3) Theoretically should stabilize in the face of multiple 2162 * connections implementing the same algorithm (this may need 2163 * a little work). 2164 * 2165 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2166 * be adjusted with a sysctl but typically only needs to be 2167 * on very slow connections. A value no smaller then 5 2168 * should be used, but only reduce this default if you have 2169 * no other choice. 2170 */ 2171 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2172 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2173 #undef USERTT 2174 2175 if (tcp_inflight_debug > 0) { 2176 static int ltime; 2177 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2178 ltime = ticks; 2179 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2180 tp, 2181 bw, 2182 tp->t_rttbest, 2183 tp->t_srtt, 2184 bwnd 2185 ); 2186 } 2187 } 2188 if ((long)bwnd < tcp_inflight_min) 2189 bwnd = tcp_inflight_min; 2190 if (bwnd > tcp_inflight_max) 2191 bwnd = tcp_inflight_max; 2192 if ((long)bwnd < tp->t_maxseg * 2) 2193 bwnd = tp->t_maxseg * 2; 2194 tp->snd_bwnd = bwnd; 2195 } 2196 2197 #ifdef TCP_SIGNATURE 2198 /* 2199 * Callback function invoked by m_apply() to digest TCP segment data 2200 * contained within an mbuf chain. 2201 */ 2202 static int 2203 tcp_signature_apply(void *fstate, void *data, u_int len) 2204 { 2205 2206 MD5Update(fstate, (u_char *)data, len); 2207 return (0); 2208 } 2209 2210 /* 2211 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2212 * 2213 * Parameters: 2214 * m pointer to head of mbuf chain 2215 * off0 offset to TCP header within the mbuf chain 2216 * len length of TCP segment data, excluding options 2217 * optlen length of TCP segment options 2218 * buf pointer to storage for computed MD5 digest 2219 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2220 * 2221 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2222 * When called from tcp_input(), we can be sure that th_sum has been 2223 * zeroed out and verified already. 2224 * 2225 * This function is for IPv4 use only. Calling this function with an 2226 * IPv6 packet in the mbuf chain will yield undefined results. 2227 * 2228 * Return 0 if successful, otherwise return -1. 2229 * 2230 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2231 * search with the destination IP address, and a 'magic SPI' to be 2232 * determined by the application. This is hardcoded elsewhere to 1179 2233 * right now. Another branch of this code exists which uses the SPD to 2234 * specify per-application flows but it is unstable. 2235 */ 2236 int 2237 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2238 u_char *buf, u_int direction) 2239 { 2240 union sockaddr_union dst; 2241 struct ippseudo ippseudo; 2242 MD5_CTX ctx; 2243 int doff; 2244 struct ip *ip; 2245 struct ipovly *ipovly; 2246 struct secasvar *sav; 2247 struct tcphdr *th; 2248 u_short savecsum; 2249 2250 KASSERT(m != NULL, ("NULL mbuf chain")); 2251 KASSERT(buf != NULL, ("NULL signature pointer")); 2252 2253 /* Extract the destination from the IP header in the mbuf. */ 2254 ip = mtod(m, struct ip *); 2255 bzero(&dst, sizeof(union sockaddr_union)); 2256 dst.sa.sa_len = sizeof(struct sockaddr_in); 2257 dst.sa.sa_family = AF_INET; 2258 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2259 ip->ip_src : ip->ip_dst; 2260 2261 /* Look up an SADB entry which matches the address of the peer. */ 2262 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2263 if (sav == NULL) { 2264 printf("%s: SADB lookup failed for %s\n", __func__, 2265 inet_ntoa(dst.sin.sin_addr)); 2266 return (EINVAL); 2267 } 2268 2269 MD5Init(&ctx); 2270 ipovly = (struct ipovly *)ip; 2271 th = (struct tcphdr *)((u_char *)ip + off0); 2272 doff = off0 + sizeof(struct tcphdr) + optlen; 2273 2274 /* 2275 * Step 1: Update MD5 hash with IP pseudo-header. 2276 * 2277 * XXX The ippseudo header MUST be digested in network byte order, 2278 * or else we'll fail the regression test. Assume all fields we've 2279 * been doing arithmetic on have been in host byte order. 2280 * XXX One cannot depend on ipovly->ih_len here. When called from 2281 * tcp_output(), the underlying ip_len member has not yet been set. 2282 */ 2283 ippseudo.ippseudo_src = ipovly->ih_src; 2284 ippseudo.ippseudo_dst = ipovly->ih_dst; 2285 ippseudo.ippseudo_pad = 0; 2286 ippseudo.ippseudo_p = IPPROTO_TCP; 2287 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2288 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2289 2290 /* 2291 * Step 2: Update MD5 hash with TCP header, excluding options. 2292 * The TCP checksum must be set to zero. 2293 */ 2294 savecsum = th->th_sum; 2295 th->th_sum = 0; 2296 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2297 th->th_sum = savecsum; 2298 2299 /* 2300 * Step 3: Update MD5 hash with TCP segment data. 2301 * Use m_apply() to avoid an early m_pullup(). 2302 */ 2303 if (len > 0) 2304 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2305 2306 /* 2307 * Step 4: Update MD5 hash with shared secret. 2308 */ 2309 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2310 MD5Final(buf, &ctx); 2311 2312 key_sa_recordxfer(sav, m); 2313 KEY_FREESAV(&sav); 2314 return (0); 2315 } 2316 #endif /* TCP_SIGNATURE */ 2317 2318 static int 2319 sysctl_drop(SYSCTL_HANDLER_ARGS) 2320 { 2321 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2322 struct sockaddr_storage addrs[2]; 2323 struct inpcb *inp; 2324 struct tcpcb *tp; 2325 struct tcptw *tw; 2326 struct sockaddr_in *fin, *lin; 2327 #ifdef INET6 2328 struct sockaddr_in6 *fin6, *lin6; 2329 struct in6_addr f6, l6; 2330 #endif 2331 int error; 2332 2333 inp = NULL; 2334 fin = lin = NULL; 2335 #ifdef INET6 2336 fin6 = lin6 = NULL; 2337 #endif 2338 error = 0; 2339 2340 if (req->oldptr != NULL || req->oldlen != 0) 2341 return (EINVAL); 2342 if (req->newptr == NULL) 2343 return (EPERM); 2344 if (req->newlen < sizeof(addrs)) 2345 return (ENOMEM); 2346 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2347 if (error) 2348 return (error); 2349 2350 switch (addrs[0].ss_family) { 2351 #ifdef INET6 2352 case AF_INET6: 2353 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2354 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2355 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2356 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2357 return (EINVAL); 2358 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2359 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2360 return (EINVAL); 2361 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2362 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2363 fin = (struct sockaddr_in *)&addrs[0]; 2364 lin = (struct sockaddr_in *)&addrs[1]; 2365 break; 2366 } 2367 error = sa6_embedscope(fin6, ip6_use_defzone); 2368 if (error) 2369 return (error); 2370 error = sa6_embedscope(lin6, ip6_use_defzone); 2371 if (error) 2372 return (error); 2373 break; 2374 #endif 2375 case AF_INET: 2376 fin = (struct sockaddr_in *)&addrs[0]; 2377 lin = (struct sockaddr_in *)&addrs[1]; 2378 if (fin->sin_len != sizeof(struct sockaddr_in) || 2379 lin->sin_len != sizeof(struct sockaddr_in)) 2380 return (EINVAL); 2381 break; 2382 default: 2383 return (EINVAL); 2384 } 2385 INP_INFO_WLOCK(&tcbinfo); 2386 switch (addrs[0].ss_family) { 2387 #ifdef INET6 2388 case AF_INET6: 2389 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2390 &l6, lin6->sin6_port, 0, NULL); 2391 break; 2392 #endif 2393 case AF_INET: 2394 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2395 lin->sin_addr, lin->sin_port, 0, NULL); 2396 break; 2397 } 2398 if (inp != NULL) { 2399 INP_LOCK(inp); 2400 if (inp->inp_vflag & INP_TIMEWAIT) { 2401 /* 2402 * XXXRW: There currently exists a state where an 2403 * inpcb is present, but its timewait state has been 2404 * discarded. For now, don't allow dropping of this 2405 * type of inpcb. 2406 */ 2407 tw = intotw(inp); 2408 if (tw != NULL) 2409 tcp_twclose(tw, 0); 2410 } else if (!(inp->inp_vflag & INP_DROPPED) && 2411 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2412 tp = intotcpcb(inp); 2413 tcp_drop(tp, ECONNABORTED); 2414 } 2415 INP_UNLOCK(inp); 2416 } else 2417 error = ESRCH; 2418 INP_INFO_WUNLOCK(&tcbinfo); 2419 return (error); 2420 } 2421 2422 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2423 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2424 0, sysctl_drop, "", "Drop TCP connection"); 2425