tcp_timewait.c revision 157474
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: head/sys/netinet/tcp_timewait.c 157474 2006-04-04 12:26:07Z rwatson $ 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 116int tcp_mssdflt = TCP_MSS; 117SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 118 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); 119 120#ifdef INET6 121int tcp_v6mssdflt = TCP6_MSS; 122SYSCTL_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 */ 135int tcp_minmss = TCP_MINMSS; 136SYSCTL_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 */ 146int tcp_minmssoverload = TCP_MINMSSOVERLOAD; 147SYSCTL_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 152static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 153SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 154 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 155#endif 156 157int tcp_do_rfc1323 = 1; 158SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 159 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 160 161static int tcp_tcbhashsize = 0; 162SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 163 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 164 165static int do_tcpdrain = 1; 166SYSCTL_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 169SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 170 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 171 172static int icmp_may_rst = 1; 173SYSCTL_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 176static int tcp_isn_reseed_interval = 0; 177SYSCTL_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/* 181 * TCP bandwidth limiting sysctls. Note that the default lower bound of 182 * 1024 exists only for debugging. A good production default would be 183 * something like 6100. 184 */ 185SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 186 "TCP inflight data limiting"); 187 188static int tcp_inflight_enable = 1; 189SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 190 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 191 192static int tcp_inflight_debug = 0; 193SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 194 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 195 196static int tcp_inflight_rttthresh; 197SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW, 198 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I", 199 "RTT threshold below which inflight will deactivate itself"); 200 201static int tcp_inflight_min = 6144; 202SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 203 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 204 205static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 206SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 207 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 208 209static int tcp_inflight_stab = 20; 210SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 211 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 212 213uma_zone_t sack_hole_zone; 214 215static struct inpcb *tcp_notify(struct inpcb *, int); 216static void tcp_isn_tick(void *); 217 218/* 219 * Target size of TCP PCB hash tables. Must be a power of two. 220 * 221 * Note that this can be overridden by the kernel environment 222 * variable net.inet.tcp.tcbhashsize 223 */ 224#ifndef TCBHASHSIZE 225#define TCBHASHSIZE 512 226#endif 227 228/* 229 * XXX 230 * Callouts should be moved into struct tcp directly. They are currently 231 * separate because the tcpcb structure is exported to userland for sysctl 232 * parsing purposes, which do not know about callouts. 233 */ 234struct tcpcb_mem { 235 struct tcpcb tcb; 236 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 237 struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 238}; 239 240static uma_zone_t tcpcb_zone; 241static uma_zone_t tcptw_zone; 242struct callout isn_callout; 243 244/* 245 * TCP initialization. 246 */ 247void 248tcp_init(void) 249{ 250 int hashsize = TCBHASHSIZE; 251 252 tcp_delacktime = TCPTV_DELACK; 253 tcp_keepinit = TCPTV_KEEP_INIT; 254 tcp_keepidle = TCPTV_KEEP_IDLE; 255 tcp_keepintvl = TCPTV_KEEPINTVL; 256 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 257 tcp_msl = TCPTV_MSL; 258 tcp_rexmit_min = TCPTV_MIN; 259 tcp_rexmit_slop = TCPTV_CPU_VAR; 260 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH; 261 262 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 263 LIST_INIT(&tcb); 264 tcbinfo.listhead = &tcb; 265 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 266 if (!powerof2(hashsize)) { 267 printf("WARNING: TCB hash size not a power of 2\n"); 268 hashsize = 512; /* safe default */ 269 } 270 tcp_tcbhashsize = hashsize; 271 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 272 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 273 &tcbinfo.porthashmask); 274 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 275 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 276 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 277#ifdef INET6 278#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 279#else /* INET6 */ 280#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 281#endif /* INET6 */ 282 if (max_protohdr < TCP_MINPROTOHDR) 283 max_protohdr = TCP_MINPROTOHDR; 284 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 285 panic("tcp_init"); 286#undef TCP_MINPROTOHDR 287 /* 288 * These have to be type stable for the benefit of the timers. 289 */ 290 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 291 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 292 uma_zone_set_max(tcpcb_zone, maxsockets); 293 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 294 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 295 uma_zone_set_max(tcptw_zone, maxsockets / 5); 296 tcp_timer_init(); 297 syncache_init(); 298 tcp_hc_init(); 299 tcp_reass_init(); 300 callout_init(&isn_callout, CALLOUT_MPSAFE); 301 tcp_isn_tick(NULL); 302 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 303 SHUTDOWN_PRI_DEFAULT); 304 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 305 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 306} 307 308void 309tcp_fini(void *xtp) 310{ 311 312 callout_stop(&isn_callout); 313} 314 315/* 316 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 317 * tcp_template used to store this data in mbufs, but we now recopy it out 318 * of the tcpcb each time to conserve mbufs. 319 */ 320void 321tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 322{ 323 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 324 325 INP_LOCK_ASSERT(inp); 326 327#ifdef INET6 328 if ((inp->inp_vflag & INP_IPV6) != 0) { 329 struct ip6_hdr *ip6; 330 331 ip6 = (struct ip6_hdr *)ip_ptr; 332 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 333 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 334 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 335 (IPV6_VERSION & IPV6_VERSION_MASK); 336 ip6->ip6_nxt = IPPROTO_TCP; 337 ip6->ip6_plen = sizeof(struct tcphdr); 338 ip6->ip6_src = inp->in6p_laddr; 339 ip6->ip6_dst = inp->in6p_faddr; 340 } else 341#endif 342 { 343 struct ip *ip; 344 345 ip = (struct ip *)ip_ptr; 346 ip->ip_v = IPVERSION; 347 ip->ip_hl = 5; 348 ip->ip_tos = inp->inp_ip_tos; 349 ip->ip_len = 0; 350 ip->ip_id = 0; 351 ip->ip_off = 0; 352 ip->ip_ttl = inp->inp_ip_ttl; 353 ip->ip_sum = 0; 354 ip->ip_p = IPPROTO_TCP; 355 ip->ip_src = inp->inp_laddr; 356 ip->ip_dst = inp->inp_faddr; 357 } 358 th->th_sport = inp->inp_lport; 359 th->th_dport = inp->inp_fport; 360 th->th_seq = 0; 361 th->th_ack = 0; 362 th->th_x2 = 0; 363 th->th_off = 5; 364 th->th_flags = 0; 365 th->th_win = 0; 366 th->th_urp = 0; 367 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 368} 369 370/* 371 * Create template to be used to send tcp packets on a connection. 372 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 373 * use for this function is in keepalives, which use tcp_respond. 374 */ 375struct tcptemp * 376tcpip_maketemplate(struct inpcb *inp) 377{ 378 struct mbuf *m; 379 struct tcptemp *n; 380 381 m = m_get(M_DONTWAIT, MT_DATA); 382 if (m == NULL) 383 return (0); 384 m->m_len = sizeof(struct tcptemp); 385 n = mtod(m, struct tcptemp *); 386 387 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 388 return (n); 389} 390 391/* 392 * Send a single message to the TCP at address specified by 393 * the given TCP/IP header. If m == NULL, then we make a copy 394 * of the tcpiphdr at ti and send directly to the addressed host. 395 * This is used to force keep alive messages out using the TCP 396 * template for a connection. If flags are given then we send 397 * a message back to the TCP which originated the * segment ti, 398 * and discard the mbuf containing it and any other attached mbufs. 399 * 400 * In any case the ack and sequence number of the transmitted 401 * segment are as specified by the parameters. 402 * 403 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 404 */ 405void 406tcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th, 407 register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) 408{ 409 register int tlen; 410 int win = 0; 411 struct ip *ip; 412 struct tcphdr *nth; 413#ifdef INET6 414 struct ip6_hdr *ip6; 415 int isipv6; 416#endif /* INET6 */ 417 int ipflags = 0; 418 struct inpcb *inp; 419 420 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 421 422#ifdef INET6 423 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 424 ip6 = ipgen; 425#endif /* INET6 */ 426 ip = ipgen; 427 428 if (tp != NULL) { 429 inp = tp->t_inpcb; 430 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 431 INP_INFO_WLOCK_ASSERT(&tcbinfo); 432 INP_LOCK_ASSERT(inp); 433 } else 434 inp = NULL; 435 436 if (tp != NULL) { 437 if (!(flags & TH_RST)) { 438 win = sbspace(&inp->inp_socket->so_rcv); 439 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 440 win = (long)TCP_MAXWIN << tp->rcv_scale; 441 } 442 } 443 if (m == NULL) { 444 m = m_gethdr(M_DONTWAIT, MT_DATA); 445 if (m == NULL) 446 return; 447 tlen = 0; 448 m->m_data += max_linkhdr; 449#ifdef INET6 450 if (isipv6) { 451 bcopy((caddr_t)ip6, mtod(m, caddr_t), 452 sizeof(struct ip6_hdr)); 453 ip6 = mtod(m, struct ip6_hdr *); 454 nth = (struct tcphdr *)(ip6 + 1); 455 } else 456#endif /* INET6 */ 457 { 458 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 459 ip = mtod(m, struct ip *); 460 nth = (struct tcphdr *)(ip + 1); 461 } 462 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 463 flags = TH_ACK; 464 } else { 465 m_freem(m->m_next); 466 m->m_next = NULL; 467 m->m_data = (caddr_t)ipgen; 468 /* m_len is set later */ 469 tlen = 0; 470#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 471#ifdef INET6 472 if (isipv6) { 473 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 474 nth = (struct tcphdr *)(ip6 + 1); 475 } else 476#endif /* INET6 */ 477 { 478 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 479 nth = (struct tcphdr *)(ip + 1); 480 } 481 if (th != nth) { 482 /* 483 * this is usually a case when an extension header 484 * exists between the IPv6 header and the 485 * TCP header. 486 */ 487 nth->th_sport = th->th_sport; 488 nth->th_dport = th->th_dport; 489 } 490 xchg(nth->th_dport, nth->th_sport, n_short); 491#undef xchg 492 } 493#ifdef INET6 494 if (isipv6) { 495 ip6->ip6_flow = 0; 496 ip6->ip6_vfc = IPV6_VERSION; 497 ip6->ip6_nxt = IPPROTO_TCP; 498 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 499 tlen)); 500 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 501 } else 502#endif 503 { 504 tlen += sizeof (struct tcpiphdr); 505 ip->ip_len = tlen; 506 ip->ip_ttl = ip_defttl; 507 if (path_mtu_discovery) 508 ip->ip_off |= IP_DF; 509 } 510 m->m_len = tlen; 511 m->m_pkthdr.len = tlen; 512 m->m_pkthdr.rcvif = NULL; 513#ifdef MAC 514 if (inp != NULL) { 515 /* 516 * Packet is associated with a socket, so allow the 517 * label of the response to reflect the socket label. 518 */ 519 INP_LOCK_ASSERT(inp); 520 mac_create_mbuf_from_inpcb(inp, m); 521 } else { 522 /* 523 * Packet is not associated with a socket, so possibly 524 * update the label in place. 525 */ 526 mac_reflect_mbuf_tcp(m); 527 } 528#endif 529 nth->th_seq = htonl(seq); 530 nth->th_ack = htonl(ack); 531 nth->th_x2 = 0; 532 nth->th_off = sizeof (struct tcphdr) >> 2; 533 nth->th_flags = flags; 534 if (tp != NULL) 535 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 536 else 537 nth->th_win = htons((u_short)win); 538 nth->th_urp = 0; 539#ifdef INET6 540 if (isipv6) { 541 nth->th_sum = 0; 542 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 543 sizeof(struct ip6_hdr), 544 tlen - sizeof(struct ip6_hdr)); 545 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 546 NULL, NULL); 547 } else 548#endif /* INET6 */ 549 { 550 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 551 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 552 m->m_pkthdr.csum_flags = CSUM_TCP; 553 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 554 } 555#ifdef TCPDEBUG 556 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 557 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 558#endif 559#ifdef INET6 560 if (isipv6) 561 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 562 else 563#endif /* INET6 */ 564 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 565} 566 567/* 568 * Create a new TCP control block, making an 569 * empty reassembly queue and hooking it to the argument 570 * protocol control block. The `inp' parameter must have 571 * come from the zone allocator set up in tcp_init(). 572 */ 573struct tcpcb * 574tcp_newtcpcb(struct inpcb *inp) 575{ 576 struct tcpcb_mem *tm; 577 struct tcpcb *tp; 578#ifdef INET6 579 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 580#endif /* INET6 */ 581 582 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 583 if (tm == NULL) 584 return (NULL); 585 tp = &tm->tcb; 586 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 587 tp->t_maxseg = tp->t_maxopd = 588#ifdef INET6 589 isipv6 ? tcp_v6mssdflt : 590#endif /* INET6 */ 591 tcp_mssdflt; 592 593 /* Set up our timeouts. */ 594 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE); 595 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE); 596 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE); 597 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE); 598 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE); 599 600 if (tcp_do_rfc1323) 601 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 602 tp->sack_enable = tcp_do_sack; 603 TAILQ_INIT(&tp->snd_holes); 604 tp->t_inpcb = inp; /* XXX */ 605 /* 606 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 607 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 608 * reasonable initial retransmit time. 609 */ 610 tp->t_srtt = TCPTV_SRTTBASE; 611 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 612 tp->t_rttmin = tcp_rexmit_min; 613 tp->t_rxtcur = TCPTV_RTOBASE; 614 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 615 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 616 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 617 tp->t_rcvtime = ticks; 618 tp->t_bw_rtttime = ticks; 619 /* 620 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 621 * because the socket may be bound to an IPv6 wildcard address, 622 * which may match an IPv4-mapped IPv6 address. 623 */ 624 inp->inp_ip_ttl = ip_defttl; 625 inp->inp_ppcb = tp; 626 return (tp); /* XXX */ 627} 628 629/* 630 * Drop a TCP connection, reporting 631 * the specified error. If connection is synchronized, 632 * then send a RST to peer. 633 */ 634struct tcpcb * 635tcp_drop(struct tcpcb *tp, int errno) 636{ 637 struct socket *so = tp->t_inpcb->inp_socket; 638 639 INP_INFO_WLOCK_ASSERT(&tcbinfo); 640 INP_LOCK_ASSERT(tp->t_inpcb); 641 642 if (TCPS_HAVERCVDSYN(tp->t_state)) { 643 tp->t_state = TCPS_CLOSED; 644 (void) tcp_output(tp); 645 tcpstat.tcps_drops++; 646 } else 647 tcpstat.tcps_conndrops++; 648 if (errno == ETIMEDOUT && tp->t_softerror) 649 errno = tp->t_softerror; 650 so->so_error = errno; 651 return (tcp_close(tp)); 652} 653 654void 655tcp_discardcb(struct tcpcb *tp) 656{ 657 struct tseg_qent *q; 658 struct inpcb *inp = tp->t_inpcb; 659 struct socket *so = inp->inp_socket; 660#ifdef INET6 661 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 662#endif /* INET6 */ 663 664 /* 665 * XXXRW: This is all very well and good, but actually, we might be 666 * discarding the tcpcb after the socket is gone, so we can't do 667 * this: 668 KASSERT(so != NULL, ("tcp_discardcb: so == NULL")); 669 */ 670 INP_LOCK_ASSERT(inp); 671 672 /* 673 * Make sure that all of our timers are stopped before we 674 * delete the PCB. 675 */ 676 callout_stop(tp->tt_rexmt); 677 callout_stop(tp->tt_persist); 678 callout_stop(tp->tt_keep); 679 callout_stop(tp->tt_2msl); 680 callout_stop(tp->tt_delack); 681 682 /* 683 * If we got enough samples through the srtt filter, 684 * save the rtt and rttvar in the routing entry. 685 * 'Enough' is arbitrarily defined as 4 rtt samples. 686 * 4 samples is enough for the srtt filter to converge 687 * to within enough % of the correct value; fewer samples 688 * and we could save a bogus rtt. The danger is not high 689 * as tcp quickly recovers from everything. 690 * XXX: Works very well but needs some more statistics! 691 */ 692 if (tp->t_rttupdated >= 4) { 693 struct hc_metrics_lite metrics; 694 u_long ssthresh; 695 696 bzero(&metrics, sizeof(metrics)); 697 /* 698 * Update the ssthresh always when the conditions below 699 * are satisfied. This gives us better new start value 700 * for the congestion avoidance for new connections. 701 * ssthresh is only set if packet loss occured on a session. 702 */ 703 ssthresh = tp->snd_ssthresh; 704 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 705 /* 706 * convert the limit from user data bytes to 707 * packets then to packet data bytes. 708 */ 709 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 710 if (ssthresh < 2) 711 ssthresh = 2; 712 ssthresh *= (u_long)(tp->t_maxseg + 713#ifdef INET6 714 (isipv6 ? sizeof (struct ip6_hdr) + 715 sizeof (struct tcphdr) : 716#endif 717 sizeof (struct tcpiphdr) 718#ifdef INET6 719 ) 720#endif 721 ); 722 } else 723 ssthresh = 0; 724 metrics.rmx_ssthresh = ssthresh; 725 726 metrics.rmx_rtt = tp->t_srtt; 727 metrics.rmx_rttvar = tp->t_rttvar; 728 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 729 metrics.rmx_bandwidth = tp->snd_bandwidth; 730 metrics.rmx_cwnd = tp->snd_cwnd; 731 metrics.rmx_sendpipe = 0; 732 metrics.rmx_recvpipe = 0; 733 734 tcp_hc_update(&inp->inp_inc, &metrics); 735 } 736 737 /* free the reassembly queue, if any */ 738 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 739 LIST_REMOVE(q, tqe_q); 740 m_freem(q->tqe_m); 741 uma_zfree(tcp_reass_zone, q); 742 tp->t_segqlen--; 743 tcp_reass_qsize--; 744 } 745 tcp_free_sackholes(tp); 746 inp->inp_ppcb = NULL; 747 tp->t_inpcb = NULL; 748 uma_zfree(tcpcb_zone, tp); 749 750 /* 751 * XXXRW: This seems a bit unclean. 752 */ 753 if (so != NULL) 754 soisdisconnected(so); 755} 756 757/* 758 * Attempt to close a TCP control block, marking it as dropped, and freeing 759 * the socket if we hold the only reference. 760 */ 761struct tcpcb * 762tcp_close(struct tcpcb *tp) 763{ 764 struct inpcb *inp = tp->t_inpcb; 765 struct socket *so; 766 767 INP_INFO_WLOCK_ASSERT(&tcbinfo); 768 INP_LOCK_ASSERT(inp); 769 770 inp->inp_vflag |= INP_DROPPED; 771 772 tcpstat.tcps_closed++; 773 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 774 so = inp->inp_socket; 775 soisdisconnected(so); 776 if (inp->inp_vflag & INP_SOCKREF) { 777 KASSERT(so->so_state & SS_PROTOREF, 778 ("tcp_close: !SS_PROTOREF")); 779 inp->inp_vflag &= ~INP_SOCKREF; 780 tcp_discardcb(tp); 781#ifdef INET6 782 if (inp->inp_vflag & INP_IPV6PROTO) { 783 in6_pcbdetach(inp); 784 in6_pcbfree(inp); 785 } else { 786#endif 787 in_pcbdetach(inp); 788 in_pcbfree(inp); 789#ifdef INET6 790 } 791#endif 792 ACCEPT_LOCK(); 793 SOCK_LOCK(so); 794 so->so_state &= ~SS_PROTOREF; 795 sofree(so); 796 return (NULL); 797 } 798 return (tp); 799} 800 801void 802tcp_drain(void) 803{ 804 805 if (do_tcpdrain) { 806 struct inpcb *inpb; 807 struct tcpcb *tcpb; 808 struct tseg_qent *te; 809 810 /* 811 * Walk the tcpbs, if existing, and flush the reassembly queue, 812 * if there is one... 813 * XXX: The "Net/3" implementation doesn't imply that the TCP 814 * reassembly queue should be flushed, but in a situation 815 * where we're really low on mbufs, this is potentially 816 * usefull. 817 */ 818 INP_INFO_RLOCK(&tcbinfo); 819 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 820 if (inpb->inp_vflag & INP_TIMEWAIT) 821 continue; 822 INP_LOCK(inpb); 823 if ((tcpb = intotcpcb(inpb)) != NULL) { 824 while ((te = LIST_FIRST(&tcpb->t_segq)) 825 != NULL) { 826 LIST_REMOVE(te, tqe_q); 827 m_freem(te->tqe_m); 828 uma_zfree(tcp_reass_zone, te); 829 tcpb->t_segqlen--; 830 tcp_reass_qsize--; 831 } 832 tcp_clean_sackreport(tcpb); 833 } 834 INP_UNLOCK(inpb); 835 } 836 INP_INFO_RUNLOCK(&tcbinfo); 837 } 838} 839 840/* 841 * Notify a tcp user of an asynchronous error; 842 * store error as soft error, but wake up user 843 * (for now, won't do anything until can select for soft error). 844 * 845 * Do not wake up user since there currently is no mechanism for 846 * reporting soft errors (yet - a kqueue filter may be added). 847 */ 848static struct inpcb * 849tcp_notify(struct inpcb *inp, int error) 850{ 851 struct tcpcb *tp; 852 853 INP_INFO_WLOCK_ASSERT(&tcbinfo); 854 INP_LOCK_ASSERT(inp); 855 856 if ((inp->inp_vflag & INP_TIMEWAIT) || 857 (inp->inp_vflag & INP_DROPPED)) 858 return (inp); 859 860 tp = intotcpcb(inp); 861 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 862 863 /* 864 * Ignore some errors if we are hooked up. 865 * If connection hasn't completed, has retransmitted several times, 866 * and receives a second error, give up now. This is better 867 * than waiting a long time to establish a connection that 868 * can never complete. 869 */ 870 if (tp->t_state == TCPS_ESTABLISHED && 871 (error == EHOSTUNREACH || error == ENETUNREACH || 872 error == EHOSTDOWN)) { 873 return (inp); 874 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 875 tp->t_softerror) { 876 tp = tcp_drop(tp, error); 877 if (tp != NULL) 878 return (inp); 879 else 880 return (NULL); 881 } else { 882 tp->t_softerror = error; 883 return (inp); 884 } 885#if 0 886 wakeup( &so->so_timeo); 887 sorwakeup(so); 888 sowwakeup(so); 889#endif 890} 891 892static int 893tcp_pcblist(SYSCTL_HANDLER_ARGS) 894{ 895 int error, i, n; 896 struct inpcb *inp, **inp_list; 897 inp_gen_t gencnt; 898 struct xinpgen xig; 899 900 /* 901 * The process of preparing the TCB list is too time-consuming and 902 * resource-intensive to repeat twice on every request. 903 */ 904 if (req->oldptr == NULL) { 905 n = tcbinfo.ipi_count; 906 req->oldidx = 2 * (sizeof xig) 907 + (n + n/8) * sizeof(struct xtcpcb); 908 return (0); 909 } 910 911 if (req->newptr != NULL) 912 return (EPERM); 913 914 /* 915 * OK, now we're committed to doing something. 916 */ 917 INP_INFO_RLOCK(&tcbinfo); 918 gencnt = tcbinfo.ipi_gencnt; 919 n = tcbinfo.ipi_count; 920 INP_INFO_RUNLOCK(&tcbinfo); 921 922 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 923 + n * sizeof(struct xtcpcb)); 924 if (error != 0) 925 return (error); 926 927 xig.xig_len = sizeof xig; 928 xig.xig_count = n; 929 xig.xig_gen = gencnt; 930 xig.xig_sogen = so_gencnt; 931 error = SYSCTL_OUT(req, &xig, sizeof xig); 932 if (error) 933 return (error); 934 935 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 936 if (inp_list == NULL) 937 return (ENOMEM); 938 939 INP_INFO_RLOCK(&tcbinfo); 940 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 941 inp = LIST_NEXT(inp, inp_list)) { 942 INP_LOCK(inp); 943 if (inp->inp_gencnt <= gencnt) { 944 /* 945 * XXX: This use of cr_cansee(), introduced with 946 * TCP state changes, is not quite right, but for 947 * now, better than nothing. 948 */ 949 if (inp->inp_vflag & INP_TIMEWAIT) { 950 if (intotw(inp) != NULL) 951 error = cr_cansee(req->td->td_ucred, 952 intotw(inp)->tw_cred); 953 else 954 error = EINVAL; /* Skip this inp. */ 955 } else 956 error = cr_canseesocket(req->td->td_ucred, 957 inp->inp_socket); 958 if (error == 0) 959 inp_list[i++] = inp; 960 } 961 INP_UNLOCK(inp); 962 } 963 INP_INFO_RUNLOCK(&tcbinfo); 964 n = i; 965 966 error = 0; 967 for (i = 0; i < n; i++) { 968 inp = inp_list[i]; 969 if (inp->inp_gencnt <= gencnt) { 970 struct xtcpcb xt; 971 void *inp_ppcb; 972 973 bzero(&xt, sizeof(xt)); 974 xt.xt_len = sizeof xt; 975 /* XXX should avoid extra copy */ 976 bcopy(inp, &xt.xt_inp, sizeof *inp); 977 inp_ppcb = inp->inp_ppcb; 978 if (inp_ppcb == NULL) 979 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 980 else if (inp->inp_vflag & INP_TIMEWAIT) { 981 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 982 xt.xt_tp.t_state = TCPS_TIME_WAIT; 983 } else 984 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 985 if (inp->inp_socket != NULL) 986 sotoxsocket(inp->inp_socket, &xt.xt_socket); 987 else { 988 bzero(&xt.xt_socket, sizeof xt.xt_socket); 989 xt.xt_socket.xso_protocol = IPPROTO_TCP; 990 } 991 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 992 error = SYSCTL_OUT(req, &xt, sizeof xt); 993 } 994 } 995 if (!error) { 996 /* 997 * Give the user an updated idea of our state. 998 * If the generation differs from what we told 999 * her before, she knows that something happened 1000 * while we were processing this request, and it 1001 * might be necessary to retry. 1002 */ 1003 INP_INFO_RLOCK(&tcbinfo); 1004 xig.xig_gen = tcbinfo.ipi_gencnt; 1005 xig.xig_sogen = so_gencnt; 1006 xig.xig_count = tcbinfo.ipi_count; 1007 INP_INFO_RUNLOCK(&tcbinfo); 1008 error = SYSCTL_OUT(req, &xig, sizeof xig); 1009 } 1010 free(inp_list, M_TEMP); 1011 return (error); 1012} 1013 1014SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1015 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1016 1017static int 1018tcp_getcred(SYSCTL_HANDLER_ARGS) 1019{ 1020 struct xucred xuc; 1021 struct sockaddr_in addrs[2]; 1022 struct inpcb *inp; 1023 int error; 1024 1025 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1026 if (error) 1027 return (error); 1028 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1029 if (error) 1030 return (error); 1031 INP_INFO_RLOCK(&tcbinfo); 1032 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1033 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1034 if (inp == NULL) { 1035 error = ENOENT; 1036 goto outunlocked; 1037 } 1038 INP_LOCK(inp); 1039 if (inp->inp_socket == NULL) { 1040 error = ENOENT; 1041 goto out; 1042 } 1043 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1044 if (error) 1045 goto out; 1046 cru2x(inp->inp_socket->so_cred, &xuc); 1047out: 1048 INP_UNLOCK(inp); 1049outunlocked: 1050 INP_INFO_RUNLOCK(&tcbinfo); 1051 if (error == 0) 1052 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1053 return (error); 1054} 1055 1056SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1057 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1058 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1059 1060#ifdef INET6 1061static int 1062tcp6_getcred(SYSCTL_HANDLER_ARGS) 1063{ 1064 struct xucred xuc; 1065 struct sockaddr_in6 addrs[2]; 1066 struct inpcb *inp; 1067 int error, mapped = 0; 1068 1069 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1070 if (error) 1071 return (error); 1072 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1073 if (error) 1074 return (error); 1075 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1076 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1077 return (error); 1078 } 1079 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1080 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1081 mapped = 1; 1082 else 1083 return (EINVAL); 1084 } 1085 1086 INP_INFO_RLOCK(&tcbinfo); 1087 if (mapped == 1) 1088 inp = in_pcblookup_hash(&tcbinfo, 1089 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1090 addrs[1].sin6_port, 1091 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1092 addrs[0].sin6_port, 1093 0, NULL); 1094 else 1095 inp = in6_pcblookup_hash(&tcbinfo, 1096 &addrs[1].sin6_addr, addrs[1].sin6_port, 1097 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1098 if (inp == NULL) { 1099 error = ENOENT; 1100 goto outunlocked; 1101 } 1102 INP_LOCK(inp); 1103 if (inp->inp_socket == NULL) { 1104 error = ENOENT; 1105 goto out; 1106 } 1107 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1108 if (error) 1109 goto out; 1110 cru2x(inp->inp_socket->so_cred, &xuc); 1111out: 1112 INP_UNLOCK(inp); 1113outunlocked: 1114 INP_INFO_RUNLOCK(&tcbinfo); 1115 if (error == 0) 1116 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1117 return (error); 1118} 1119 1120SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1121 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1122 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1123#endif 1124 1125 1126void 1127tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1128{ 1129 struct ip *ip = vip; 1130 struct tcphdr *th; 1131 struct in_addr faddr; 1132 struct inpcb *inp; 1133 struct tcpcb *tp; 1134 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1135 struct icmp *icp; 1136 struct in_conninfo inc; 1137 tcp_seq icmp_tcp_seq; 1138 int mtu; 1139 1140 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1141 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1142 return; 1143 1144 if (cmd == PRC_MSGSIZE) 1145 notify = tcp_mtudisc; 1146 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1147 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1148 notify = tcp_drop_syn_sent; 1149 /* 1150 * Redirects don't need to be handled up here. 1151 */ 1152 else if (PRC_IS_REDIRECT(cmd)) 1153 return; 1154 /* 1155 * Source quench is depreciated. 1156 */ 1157 else if (cmd == PRC_QUENCH) 1158 return; 1159 /* 1160 * Hostdead is ugly because it goes linearly through all PCBs. 1161 * XXX: We never get this from ICMP, otherwise it makes an 1162 * excellent DoS attack on machines with many connections. 1163 */ 1164 else if (cmd == PRC_HOSTDEAD) 1165 ip = NULL; 1166 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1167 return; 1168 if (ip != NULL) { 1169 icp = (struct icmp *)((caddr_t)ip 1170 - offsetof(struct icmp, icmp_ip)); 1171 th = (struct tcphdr *)((caddr_t)ip 1172 + (ip->ip_hl << 2)); 1173 INP_INFO_WLOCK(&tcbinfo); 1174 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1175 ip->ip_src, th->th_sport, 0, NULL); 1176 if (inp != NULL) { 1177 INP_LOCK(inp); 1178 if (!(inp->inp_vflag & INP_TIMEWAIT) && 1179 !(inp->inp_vflag & INP_DROPPED) && 1180 !(inp->inp_socket == NULL)) { 1181 icmp_tcp_seq = htonl(th->th_seq); 1182 tp = intotcpcb(inp); 1183 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1184 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1185 if (cmd == PRC_MSGSIZE) { 1186 /* 1187 * MTU discovery: 1188 * If we got a needfrag set the MTU 1189 * in the route to the suggested new 1190 * value (if given) and then notify. 1191 */ 1192 bzero(&inc, sizeof(inc)); 1193 inc.inc_flags = 0; /* IPv4 */ 1194 inc.inc_faddr = faddr; 1195 1196 mtu = ntohs(icp->icmp_nextmtu); 1197 /* 1198 * If no alternative MTU was 1199 * proposed, try the next smaller 1200 * one. ip->ip_len has already 1201 * been swapped in icmp_input(). 1202 */ 1203 if (!mtu) 1204 mtu = ip_next_mtu(ip->ip_len, 1205 1); 1206 if (mtu < max(296, (tcp_minmss) 1207 + sizeof(struct tcpiphdr))) 1208 mtu = 0; 1209 if (!mtu) 1210 mtu = tcp_mssdflt 1211 + sizeof(struct tcpiphdr); 1212 /* 1213 * Only cache the the MTU if it 1214 * is smaller than the interface 1215 * or route MTU. tcp_mtudisc() 1216 * will do right thing by itself. 1217 */ 1218 if (mtu <= tcp_maxmtu(&inc)) 1219 tcp_hc_updatemtu(&inc, mtu); 1220 } 1221 1222 inp = (*notify)(inp, inetctlerrmap[cmd]); 1223 } 1224 } 1225 if (inp != NULL) 1226 INP_UNLOCK(inp); 1227 } else { 1228 inc.inc_fport = th->th_dport; 1229 inc.inc_lport = th->th_sport; 1230 inc.inc_faddr = faddr; 1231 inc.inc_laddr = ip->ip_src; 1232#ifdef INET6 1233 inc.inc_isipv6 = 0; 1234#endif 1235 syncache_unreach(&inc, th); 1236 } 1237 INP_INFO_WUNLOCK(&tcbinfo); 1238 } else 1239 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1240} 1241 1242#ifdef INET6 1243void 1244tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1245{ 1246 struct tcphdr th; 1247 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1248 struct ip6_hdr *ip6; 1249 struct mbuf *m; 1250 struct ip6ctlparam *ip6cp = NULL; 1251 const struct sockaddr_in6 *sa6_src = NULL; 1252 int off; 1253 struct tcp_portonly { 1254 u_int16_t th_sport; 1255 u_int16_t th_dport; 1256 } *thp; 1257 1258 if (sa->sa_family != AF_INET6 || 1259 sa->sa_len != sizeof(struct sockaddr_in6)) 1260 return; 1261 1262 if (cmd == PRC_MSGSIZE) 1263 notify = tcp_mtudisc; 1264 else if (!PRC_IS_REDIRECT(cmd) && 1265 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1266 return; 1267 /* Source quench is depreciated. */ 1268 else if (cmd == PRC_QUENCH) 1269 return; 1270 1271 /* if the parameter is from icmp6, decode it. */ 1272 if (d != NULL) { 1273 ip6cp = (struct ip6ctlparam *)d; 1274 m = ip6cp->ip6c_m; 1275 ip6 = ip6cp->ip6c_ip6; 1276 off = ip6cp->ip6c_off; 1277 sa6_src = ip6cp->ip6c_src; 1278 } else { 1279 m = NULL; 1280 ip6 = NULL; 1281 off = 0; /* fool gcc */ 1282 sa6_src = &sa6_any; 1283 } 1284 1285 if (ip6 != NULL) { 1286 struct in_conninfo inc; 1287 /* 1288 * XXX: We assume that when IPV6 is non NULL, 1289 * M and OFF are valid. 1290 */ 1291 1292 /* check if we can safely examine src and dst ports */ 1293 if (m->m_pkthdr.len < off + sizeof(*thp)) 1294 return; 1295 1296 bzero(&th, sizeof(th)); 1297 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1298 1299 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1300 (struct sockaddr *)ip6cp->ip6c_src, 1301 th.th_sport, cmd, NULL, notify); 1302 1303 inc.inc_fport = th.th_dport; 1304 inc.inc_lport = th.th_sport; 1305 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1306 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1307 inc.inc_isipv6 = 1; 1308 INP_INFO_WLOCK(&tcbinfo); 1309 syncache_unreach(&inc, &th); 1310 INP_INFO_WUNLOCK(&tcbinfo); 1311 } else 1312 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1313 0, cmd, NULL, notify); 1314} 1315#endif /* INET6 */ 1316 1317 1318/* 1319 * Following is where TCP initial sequence number generation occurs. 1320 * 1321 * There are two places where we must use initial sequence numbers: 1322 * 1. In SYN-ACK packets. 1323 * 2. In SYN packets. 1324 * 1325 * All ISNs for SYN-ACK packets are generated by the syncache. See 1326 * tcp_syncache.c for details. 1327 * 1328 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1329 * depends on this property. In addition, these ISNs should be 1330 * unguessable so as to prevent connection hijacking. To satisfy 1331 * the requirements of this situation, the algorithm outlined in 1332 * RFC 1948 is used, with only small modifications. 1333 * 1334 * Implementation details: 1335 * 1336 * Time is based off the system timer, and is corrected so that it 1337 * increases by one megabyte per second. This allows for proper 1338 * recycling on high speed LANs while still leaving over an hour 1339 * before rollover. 1340 * 1341 * As reading the *exact* system time is too expensive to be done 1342 * whenever setting up a TCP connection, we increment the time 1343 * offset in two ways. First, a small random positive increment 1344 * is added to isn_offset for each connection that is set up. 1345 * Second, the function tcp_isn_tick fires once per clock tick 1346 * and increments isn_offset as necessary so that sequence numbers 1347 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1348 * random positive increments serve only to ensure that the same 1349 * exact sequence number is never sent out twice (as could otherwise 1350 * happen when a port is recycled in less than the system tick 1351 * interval.) 1352 * 1353 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1354 * between seeding of isn_secret. This is normally set to zero, 1355 * as reseeding should not be necessary. 1356 * 1357 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1358 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1359 * general, this means holding an exclusive (write) lock. 1360 */ 1361 1362#define ISN_BYTES_PER_SECOND 1048576 1363#define ISN_STATIC_INCREMENT 4096 1364#define ISN_RANDOM_INCREMENT (4096 - 1) 1365 1366static u_char isn_secret[32]; 1367static int isn_last_reseed; 1368static u_int32_t isn_offset, isn_offset_old; 1369static MD5_CTX isn_ctx; 1370 1371tcp_seq 1372tcp_new_isn(struct tcpcb *tp) 1373{ 1374 u_int32_t md5_buffer[4]; 1375 tcp_seq new_isn; 1376 1377 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1378 INP_LOCK_ASSERT(tp->t_inpcb); 1379 1380 /* Seed if this is the first use, reseed if requested. */ 1381 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1382 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1383 < (u_int)ticks))) { 1384 read_random(&isn_secret, sizeof(isn_secret)); 1385 isn_last_reseed = ticks; 1386 } 1387 1388 /* Compute the md5 hash and return the ISN. */ 1389 MD5Init(&isn_ctx); 1390 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1391 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1392#ifdef INET6 1393 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1394 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1395 sizeof(struct in6_addr)); 1396 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1397 sizeof(struct in6_addr)); 1398 } else 1399#endif 1400 { 1401 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1402 sizeof(struct in_addr)); 1403 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1404 sizeof(struct in_addr)); 1405 } 1406 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1407 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1408 new_isn = (tcp_seq) md5_buffer[0]; 1409 isn_offset += ISN_STATIC_INCREMENT + 1410 (arc4random() & ISN_RANDOM_INCREMENT); 1411 new_isn += isn_offset; 1412 return (new_isn); 1413} 1414 1415/* 1416 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1417 * to keep time flowing at a relatively constant rate. If the random 1418 * increments have already pushed us past the projected offset, do nothing. 1419 */ 1420static void 1421tcp_isn_tick(void *xtp) 1422{ 1423 u_int32_t projected_offset; 1424 1425 INP_INFO_WLOCK(&tcbinfo); 1426 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1427 1428 if (projected_offset > isn_offset) 1429 isn_offset = projected_offset; 1430 1431 isn_offset_old = isn_offset; 1432 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1433 INP_INFO_WUNLOCK(&tcbinfo); 1434} 1435 1436/* 1437 * When a specific ICMP unreachable message is received and the 1438 * connection state is SYN-SENT, drop the connection. This behavior 1439 * is controlled by the icmp_may_rst sysctl. 1440 */ 1441struct inpcb * 1442tcp_drop_syn_sent(struct inpcb *inp, int errno) 1443{ 1444 struct tcpcb *tp; 1445 1446 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1447 INP_LOCK_ASSERT(inp); 1448 1449 if ((inp->inp_vflag & INP_TIMEWAIT) || 1450 (inp->inp_vflag & INP_DROPPED)) 1451 return (inp); 1452 1453 tp = intotcpcb(inp); 1454 if (tp->t_state != TCPS_SYN_SENT) 1455 return (inp); 1456 1457 tp = tcp_drop(tp, errno); 1458 if (tp != NULL) 1459 return (inp); 1460 else 1461 return (NULL); 1462} 1463 1464/* 1465 * When `need fragmentation' ICMP is received, update our idea of the MSS 1466 * based on the new value in the route. Also nudge TCP to send something, 1467 * since we know the packet we just sent was dropped. 1468 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1469 */ 1470struct inpcb * 1471tcp_mtudisc(struct inpcb *inp, int errno) 1472{ 1473 struct tcpcb *tp; 1474 struct socket *so = inp->inp_socket; 1475 u_int maxmtu; 1476 u_int romtu; 1477 int mss; 1478#ifdef INET6 1479 int isipv6; 1480#endif /* INET6 */ 1481 1482 INP_LOCK_ASSERT(inp); 1483 if ((inp->inp_vflag & INP_TIMEWAIT) || 1484 (inp->inp_vflag & INP_DROPPED)) 1485 return (inp); 1486 1487 tp = intotcpcb(inp); 1488 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1489 1490#ifdef INET6 1491 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1492#endif 1493 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1494 romtu = 1495#ifdef INET6 1496 isipv6 ? tcp_maxmtu6(&inp->inp_inc) : 1497#endif /* INET6 */ 1498 tcp_maxmtu(&inp->inp_inc); 1499 if (!maxmtu) 1500 maxmtu = romtu; 1501 else 1502 maxmtu = min(maxmtu, romtu); 1503 if (!maxmtu) { 1504 tp->t_maxopd = tp->t_maxseg = 1505#ifdef INET6 1506 isipv6 ? tcp_v6mssdflt : 1507#endif /* INET6 */ 1508 tcp_mssdflt; 1509 return (inp); 1510 } 1511 mss = maxmtu - 1512#ifdef INET6 1513 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1514#endif /* INET6 */ 1515 sizeof(struct tcpiphdr) 1516#ifdef INET6 1517 ) 1518#endif /* INET6 */ 1519 ; 1520 1521 /* 1522 * XXX - The above conditional probably violates the TCP 1523 * spec. The problem is that, since we don't know the 1524 * other end's MSS, we are supposed to use a conservative 1525 * default. But, if we do that, then MTU discovery will 1526 * never actually take place, because the conservative 1527 * default is much less than the MTUs typically seen 1528 * on the Internet today. For the moment, we'll sweep 1529 * this under the carpet. 1530 * 1531 * The conservative default might not actually be a problem 1532 * if the only case this occurs is when sending an initial 1533 * SYN with options and data to a host we've never talked 1534 * to before. Then, they will reply with an MSS value which 1535 * will get recorded and the new parameters should get 1536 * recomputed. For Further Study. 1537 */ 1538 if (tp->t_maxopd <= mss) 1539 return (inp); 1540 tp->t_maxopd = mss; 1541 1542 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1543 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1544 mss -= TCPOLEN_TSTAMP_APPA; 1545#if (MCLBYTES & (MCLBYTES - 1)) == 0 1546 if (mss > MCLBYTES) 1547 mss &= ~(MCLBYTES-1); 1548#else 1549 if (mss > MCLBYTES) 1550 mss = mss / MCLBYTES * MCLBYTES; 1551#endif 1552 if (so->so_snd.sb_hiwat < mss) 1553 mss = so->so_snd.sb_hiwat; 1554 1555 tp->t_maxseg = mss; 1556 1557 tcpstat.tcps_mturesent++; 1558 tp->t_rtttime = 0; 1559 tp->snd_nxt = tp->snd_una; 1560 tcp_output(tp); 1561 return (inp); 1562} 1563 1564/* 1565 * Look-up the routing entry to the peer of this inpcb. If no route 1566 * is found and it cannot be allocated, then return NULL. This routine 1567 * is called by TCP routines that access the rmx structure and by tcp_mss 1568 * to get the interface MTU. 1569 */ 1570u_long 1571tcp_maxmtu(struct in_conninfo *inc) 1572{ 1573 struct route sro; 1574 struct sockaddr_in *dst; 1575 struct ifnet *ifp; 1576 u_long maxmtu = 0; 1577 1578 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1579 1580 bzero(&sro, sizeof(sro)); 1581 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1582 dst = (struct sockaddr_in *)&sro.ro_dst; 1583 dst->sin_family = AF_INET; 1584 dst->sin_len = sizeof(*dst); 1585 dst->sin_addr = inc->inc_faddr; 1586 rtalloc_ign(&sro, RTF_CLONING); 1587 } 1588 if (sro.ro_rt != NULL) { 1589 ifp = sro.ro_rt->rt_ifp; 1590 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1591 maxmtu = ifp->if_mtu; 1592 else 1593 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1594 RTFREE(sro.ro_rt); 1595 } 1596 return (maxmtu); 1597} 1598 1599#ifdef INET6 1600u_long 1601tcp_maxmtu6(struct in_conninfo *inc) 1602{ 1603 struct route_in6 sro6; 1604 struct ifnet *ifp; 1605 u_long maxmtu = 0; 1606 1607 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1608 1609 bzero(&sro6, sizeof(sro6)); 1610 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1611 sro6.ro_dst.sin6_family = AF_INET6; 1612 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1613 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1614 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1615 } 1616 if (sro6.ro_rt != NULL) { 1617 ifp = sro6.ro_rt->rt_ifp; 1618 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1619 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1620 else 1621 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1622 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1623 RTFREE(sro6.ro_rt); 1624 } 1625 1626 return (maxmtu); 1627} 1628#endif /* INET6 */ 1629 1630#ifdef IPSEC 1631/* compute ESP/AH header size for TCP, including outer IP header. */ 1632size_t 1633ipsec_hdrsiz_tcp(struct tcpcb *tp) 1634{ 1635 struct inpcb *inp; 1636 struct mbuf *m; 1637 size_t hdrsiz; 1638 struct ip *ip; 1639#ifdef INET6 1640 struct ip6_hdr *ip6; 1641#endif 1642 struct tcphdr *th; 1643 1644 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1645 return (0); 1646 MGETHDR(m, M_DONTWAIT, MT_DATA); 1647 if (!m) 1648 return (0); 1649 1650#ifdef INET6 1651 if ((inp->inp_vflag & INP_IPV6) != 0) { 1652 ip6 = mtod(m, struct ip6_hdr *); 1653 th = (struct tcphdr *)(ip6 + 1); 1654 m->m_pkthdr.len = m->m_len = 1655 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1656 tcpip_fillheaders(inp, ip6, th); 1657 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1658 } else 1659#endif /* INET6 */ 1660 { 1661 ip = mtod(m, struct ip *); 1662 th = (struct tcphdr *)(ip + 1); 1663 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1664 tcpip_fillheaders(inp, ip, th); 1665 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1666 } 1667 1668 m_free(m); 1669 return (hdrsiz); 1670} 1671#endif /*IPSEC*/ 1672 1673/* 1674 * Move a TCP connection into TIME_WAIT state. 1675 * tcbinfo is locked. 1676 * inp is locked, and is unlocked before returning. 1677 */ 1678void 1679tcp_twstart(struct tcpcb *tp) 1680{ 1681 struct tcptw *tw; 1682 struct inpcb *inp; 1683 int tw_time, acknow; 1684 struct socket *so; 1685 1686 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1687 INP_LOCK_ASSERT(tp->t_inpcb); 1688 1689 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1690 if (tw == NULL) { 1691 tw = tcp_timer_2msl_tw(1); 1692 if (tw == NULL) { 1693 tp = tcp_close(tp); 1694 if (tp != NULL) 1695 INP_UNLOCK(tp->t_inpcb); 1696 return; 1697 } 1698 } 1699 inp = tp->t_inpcb; 1700 tw->tw_inpcb = inp; 1701 1702 /* 1703 * Recover last window size sent. 1704 */ 1705 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1706 1707 /* 1708 * Set t_recent if timestamps are used on the connection. 1709 */ 1710 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1711 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1712 tw->t_recent = tp->ts_recent; 1713 else 1714 tw->t_recent = 0; 1715 1716 tw->snd_nxt = tp->snd_nxt; 1717 tw->rcv_nxt = tp->rcv_nxt; 1718 tw->iss = tp->iss; 1719 tw->irs = tp->irs; 1720 tw->t_starttime = tp->t_starttime; 1721 tw->tw_time = 0; 1722 1723/* XXX 1724 * If this code will 1725 * be used for fin-wait-2 state also, then we may need 1726 * a ts_recent from the last segment. 1727 */ 1728 tw_time = 2 * tcp_msl; 1729 acknow = tp->t_flags & TF_ACKNOW; 1730 1731 /* 1732 * First, discard tcpcb state, which includes stopping its timers and 1733 * freeing it. tcp_discardcb() used to also release the inpcb, but 1734 * that work is now done in the caller. 1735 */ 1736 tcp_discardcb(tp); 1737 so = inp->inp_socket; 1738 SOCK_LOCK(so); 1739 tw->tw_cred = crhold(so->so_cred); 1740 tw->tw_so_options = so->so_options; 1741 SOCK_UNLOCK(so); 1742 if (acknow) 1743 tcp_twrespond(tw, TH_ACK); 1744 inp->inp_ppcb = tw; 1745 inp->inp_vflag |= INP_TIMEWAIT; 1746 tcp_timer_2msl_reset(tw, tw_time); 1747 1748 /* 1749 * If the inpcb owns the sole reference to the socket, then we can 1750 * detach and free the socket as it is not needed in time wait. 1751 */ 1752 if (inp->inp_vflag & INP_SOCKREF) { 1753 KASSERT(so->so_state & SS_PROTOREF, 1754 ("tcp_twstart: !SS_PROTOREF")); 1755 inp->inp_vflag &= ~INP_SOCKREF; 1756#ifdef INET6 1757 if (inp->inp_vflag & INP_IPV6PROTO) 1758 in6_pcbdetach(inp); 1759 else 1760#endif 1761 in_pcbdetach(inp); 1762 INP_UNLOCK(inp); 1763 ACCEPT_LOCK(); 1764 SOCK_LOCK(so); 1765 so->so_state &= ~SS_PROTOREF; 1766 sofree(so); 1767 } else 1768 INP_UNLOCK(inp); 1769} 1770 1771/* 1772 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1773 * the actual rate is slightly higher due to the addition of 1774 * random positive increments. 1775 * 1776 * Most other new OSes use semi-randomized ISN values, so we 1777 * do not need to worry about them. 1778 */ 1779#define MS_ISN_BYTES_PER_SECOND 250000 1780 1781/* 1782 * Determine if the ISN we will generate has advanced beyond the last 1783 * sequence number used by the previous connection. If so, indicate 1784 * that it is safe to recycle this tw socket by returning 1. 1785 * 1786 * XXXRW: This function should assert the inpcb lock as it does multiple 1787 * non-atomic reads from the tcptw, but is currently called without it from 1788 * in_pcb.c:in_pcblookup_local(). 1789 */ 1790int 1791tcp_twrecycleable(struct tcptw *tw) 1792{ 1793 tcp_seq new_iss = tw->iss; 1794 tcp_seq new_irs = tw->irs; 1795 1796 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1797 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1798 1799 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1800 return (1); 1801 else 1802 return (0); 1803} 1804 1805void 1806tcp_twclose(struct tcptw *tw, int reuse) 1807{ 1808 struct socket *so; 1809 struct inpcb *inp; 1810 1811 /* 1812 * At this point, we are in one of two situations: 1813 * 1814 * (1) We have no socket, just an inpcb<->twtcp pair. Release it all 1815 * after validating. 1816 * 1817 * (2) We have a socket, which we may or may now own the reference 1818 * for. If we own the reference, release all the state after 1819 * validating. If not, leave it for the socket close to clean up. 1820 */ 1821 inp = tw->tw_inpcb; 1822 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1823 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 1824 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1825 INP_LOCK_ASSERT(inp); 1826 1827 tw->tw_inpcb = NULL; 1828 tcp_timer_2msl_stop(tw); 1829 inp->inp_ppcb = NULL; 1830 inp->inp_vflag |= INP_DROPPED; 1831 1832 so = inp->inp_socket; 1833 if (so != NULL) { 1834 if (inp->inp_vflag & INP_SOCKREF) { 1835 /* 1836 * If a socket is present, and we own the only 1837 * reference, we need to tear down the socket and the 1838 * inpcb. 1839 */ 1840 inp->inp_vflag &= ~INP_SOCKREF; 1841#ifdef INET6 1842 if (inp->inp_vflag & INP_IPV6PROTO) { 1843 in6_pcbdetach(inp); 1844 in6_pcbfree(inp); 1845 } else { 1846 in_pcbdetach(inp); 1847 in_pcbfree(inp); 1848 } 1849#endif 1850 ACCEPT_LOCK(); 1851 SOCK_LOCK(so); 1852 KASSERT(so->so_state & SS_PROTOREF, 1853 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1854 so->so_state &= ~SS_PROTOREF; 1855 sofree(so); 1856 } else { 1857 /* 1858 * If we don't own the only reference, the socket and 1859 * inpcb need to be left around to be handled by 1860 * tcp_usr_detach() later. 1861 */ 1862 INP_UNLOCK(inp); 1863 } 1864 } else { 1865#ifdef INET6 1866 if (inp->inp_vflag & INP_IPV6PROTO) 1867 in6_pcbfree(inp); 1868 else 1869#endif 1870 in_pcbfree(inp); 1871 } 1872 tcpstat.tcps_closed++; 1873 crfree(tw->tw_cred); 1874 tw->tw_cred = NULL; 1875 if (reuse) 1876 return; 1877 uma_zfree(tcptw_zone, tw); 1878} 1879 1880int 1881tcp_twrespond(struct tcptw *tw, int flags) 1882{ 1883 struct inpcb *inp = tw->tw_inpcb; 1884 struct tcphdr *th; 1885 struct mbuf *m; 1886 struct ip *ip = NULL; 1887 u_int8_t *optp; 1888 u_int hdrlen, optlen; 1889 int error; 1890#ifdef INET6 1891 struct ip6_hdr *ip6 = NULL; 1892 int isipv6 = inp->inp_inc.inc_isipv6; 1893#endif 1894 1895 INP_LOCK_ASSERT(inp); 1896 1897 m = m_gethdr(M_DONTWAIT, MT_DATA); 1898 if (m == NULL) 1899 return (ENOBUFS); 1900 m->m_data += max_linkhdr; 1901 1902#ifdef MAC 1903 mac_create_mbuf_from_inpcb(inp, m); 1904#endif 1905 1906#ifdef INET6 1907 if (isipv6) { 1908 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1909 ip6 = mtod(m, struct ip6_hdr *); 1910 th = (struct tcphdr *)(ip6 + 1); 1911 tcpip_fillheaders(inp, ip6, th); 1912 } else 1913#endif 1914 { 1915 hdrlen = sizeof(struct tcpiphdr); 1916 ip = mtod(m, struct ip *); 1917 th = (struct tcphdr *)(ip + 1); 1918 tcpip_fillheaders(inp, ip, th); 1919 } 1920 optp = (u_int8_t *)(th + 1); 1921 1922 /* 1923 * Send a timestamp and echo-reply if both our side and our peer 1924 * have sent timestamps in our SYN's and this is not a RST. 1925 */ 1926 if (tw->t_recent && flags == TH_ACK) { 1927 u_int32_t *lp = (u_int32_t *)optp; 1928 1929 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1930 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1931 *lp++ = htonl(ticks); 1932 *lp = htonl(tw->t_recent); 1933 optp += TCPOLEN_TSTAMP_APPA; 1934 } 1935 1936 optlen = optp - (u_int8_t *)(th + 1); 1937 1938 m->m_len = hdrlen + optlen; 1939 m->m_pkthdr.len = m->m_len; 1940 1941 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 1942 1943 th->th_seq = htonl(tw->snd_nxt); 1944 th->th_ack = htonl(tw->rcv_nxt); 1945 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1946 th->th_flags = flags; 1947 th->th_win = htons(tw->last_win); 1948 1949#ifdef INET6 1950 if (isipv6) { 1951 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 1952 sizeof(struct tcphdr) + optlen); 1953 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 1954 error = ip6_output(m, inp->in6p_outputopts, NULL, 1955 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 1956 } else 1957#endif 1958 { 1959 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1960 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 1961 m->m_pkthdr.csum_flags = CSUM_TCP; 1962 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1963 ip->ip_len = m->m_pkthdr.len; 1964 if (path_mtu_discovery) 1965 ip->ip_off |= IP_DF; 1966 error = ip_output(m, inp->inp_options, NULL, 1967 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 1968 NULL, inp); 1969 } 1970 if (flags & TH_ACK) 1971 tcpstat.tcps_sndacks++; 1972 else 1973 tcpstat.tcps_sndctrl++; 1974 tcpstat.tcps_sndtotal++; 1975 return (error); 1976} 1977 1978/* 1979 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1980 * 1981 * This code attempts to calculate the bandwidth-delay product as a 1982 * means of determining the optimal window size to maximize bandwidth, 1983 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1984 * routers. This code also does a fairly good job keeping RTTs in check 1985 * across slow links like modems. We implement an algorithm which is very 1986 * similar (but not meant to be) TCP/Vegas. The code operates on the 1987 * transmitter side of a TCP connection and so only effects the transmit 1988 * side of the connection. 1989 * 1990 * BACKGROUND: TCP makes no provision for the management of buffer space 1991 * at the end points or at the intermediate routers and switches. A TCP 1992 * stream, whether using NewReno or not, will eventually buffer as 1993 * many packets as it is able and the only reason this typically works is 1994 * due to the fairly small default buffers made available for a connection 1995 * (typicaly 16K or 32K). As machines use larger windows and/or window 1996 * scaling it is now fairly easy for even a single TCP connection to blow-out 1997 * all available buffer space not only on the local interface, but on 1998 * intermediate routers and switches as well. NewReno makes a misguided 1999 * attempt to 'solve' this problem by waiting for an actual failure to occur, 2000 * then backing off, then steadily increasing the window again until another 2001 * failure occurs, ad-infinitum. This results in terrible oscillation that 2002 * is only made worse as network loads increase and the idea of intentionally 2003 * blowing out network buffers is, frankly, a terrible way to manage network 2004 * resources. 2005 * 2006 * It is far better to limit the transmit window prior to the failure 2007 * condition being achieved. There are two general ways to do this: First 2008 * you can 'scan' through different transmit window sizes and locate the 2009 * point where the RTT stops increasing, indicating that you have filled the 2010 * pipe, then scan backwards until you note that RTT stops decreasing, then 2011 * repeat ad-infinitum. This method works in principle but has severe 2012 * implementation issues due to RTT variances, timer granularity, and 2013 * instability in the algorithm which can lead to many false positives and 2014 * create oscillations as well as interact badly with other TCP streams 2015 * implementing the same algorithm. 2016 * 2017 * The second method is to limit the window to the bandwidth delay product 2018 * of the link. This is the method we implement. RTT variances and our 2019 * own manipulation of the congestion window, bwnd, can potentially 2020 * destabilize the algorithm. For this reason we have to stabilize the 2021 * elements used to calculate the window. We do this by using the minimum 2022 * observed RTT, the long term average of the observed bandwidth, and 2023 * by adding two segments worth of slop. It isn't perfect but it is able 2024 * to react to changing conditions and gives us a very stable basis on 2025 * which to extend the algorithm. 2026 */ 2027void 2028tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2029{ 2030 u_long bw; 2031 u_long bwnd; 2032 int save_ticks; 2033 2034 INP_LOCK_ASSERT(tp->t_inpcb); 2035 2036 /* 2037 * If inflight_enable is disabled in the middle of a tcp connection, 2038 * make sure snd_bwnd is effectively disabled. 2039 */ 2040 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2041 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2042 tp->snd_bandwidth = 0; 2043 return; 2044 } 2045 2046 /* 2047 * Figure out the bandwidth. Due to the tick granularity this 2048 * is a very rough number and it MUST be averaged over a fairly 2049 * long period of time. XXX we need to take into account a link 2050 * that is not using all available bandwidth, but for now our 2051 * slop will ramp us up if this case occurs and the bandwidth later 2052 * increases. 2053 * 2054 * Note: if ticks rollover 'bw' may wind up negative. We must 2055 * effectively reset t_bw_rtttime for this case. 2056 */ 2057 save_ticks = ticks; 2058 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2059 return; 2060 2061 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2062 (save_ticks - tp->t_bw_rtttime); 2063 tp->t_bw_rtttime = save_ticks; 2064 tp->t_bw_rtseq = ack_seq; 2065 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2066 return; 2067 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2068 2069 tp->snd_bandwidth = bw; 2070 2071 /* 2072 * Calculate the semi-static bandwidth delay product, plus two maximal 2073 * segments. The additional slop puts us squarely in the sweet 2074 * spot and also handles the bandwidth run-up case and stabilization. 2075 * Without the slop we could be locking ourselves into a lower 2076 * bandwidth. 2077 * 2078 * Situations Handled: 2079 * (1) Prevents over-queueing of packets on LANs, especially on 2080 * high speed LANs, allowing larger TCP buffers to be 2081 * specified, and also does a good job preventing 2082 * over-queueing of packets over choke points like modems 2083 * (at least for the transmit side). 2084 * 2085 * (2) Is able to handle changing network loads (bandwidth 2086 * drops so bwnd drops, bandwidth increases so bwnd 2087 * increases). 2088 * 2089 * (3) Theoretically should stabilize in the face of multiple 2090 * connections implementing the same algorithm (this may need 2091 * a little work). 2092 * 2093 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2094 * be adjusted with a sysctl but typically only needs to be 2095 * on very slow connections. A value no smaller then 5 2096 * should be used, but only reduce this default if you have 2097 * no other choice. 2098 */ 2099#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2100 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2101#undef USERTT 2102 2103 if (tcp_inflight_debug > 0) { 2104 static int ltime; 2105 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2106 ltime = ticks; 2107 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2108 tp, 2109 bw, 2110 tp->t_rttbest, 2111 tp->t_srtt, 2112 bwnd 2113 ); 2114 } 2115 } 2116 if ((long)bwnd < tcp_inflight_min) 2117 bwnd = tcp_inflight_min; 2118 if (bwnd > tcp_inflight_max) 2119 bwnd = tcp_inflight_max; 2120 if ((long)bwnd < tp->t_maxseg * 2) 2121 bwnd = tp->t_maxseg * 2; 2122 tp->snd_bwnd = bwnd; 2123} 2124 2125#ifdef TCP_SIGNATURE 2126/* 2127 * Callback function invoked by m_apply() to digest TCP segment data 2128 * contained within an mbuf chain. 2129 */ 2130static int 2131tcp_signature_apply(void *fstate, void *data, u_int len) 2132{ 2133 2134 MD5Update(fstate, (u_char *)data, len); 2135 return (0); 2136} 2137 2138/* 2139 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2140 * 2141 * Parameters: 2142 * m pointer to head of mbuf chain 2143 * off0 offset to TCP header within the mbuf chain 2144 * len length of TCP segment data, excluding options 2145 * optlen length of TCP segment options 2146 * buf pointer to storage for computed MD5 digest 2147 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2148 * 2149 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2150 * When called from tcp_input(), we can be sure that th_sum has been 2151 * zeroed out and verified already. 2152 * 2153 * This function is for IPv4 use only. Calling this function with an 2154 * IPv6 packet in the mbuf chain will yield undefined results. 2155 * 2156 * Return 0 if successful, otherwise return -1. 2157 * 2158 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2159 * search with the destination IP address, and a 'magic SPI' to be 2160 * determined by the application. This is hardcoded elsewhere to 1179 2161 * right now. Another branch of this code exists which uses the SPD to 2162 * specify per-application flows but it is unstable. 2163 */ 2164int 2165tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2166 u_char *buf, u_int direction) 2167{ 2168 union sockaddr_union dst; 2169 struct ippseudo ippseudo; 2170 MD5_CTX ctx; 2171 int doff; 2172 struct ip *ip; 2173 struct ipovly *ipovly; 2174 struct secasvar *sav; 2175 struct tcphdr *th; 2176 u_short savecsum; 2177 2178 KASSERT(m != NULL, ("NULL mbuf chain")); 2179 KASSERT(buf != NULL, ("NULL signature pointer")); 2180 2181 /* Extract the destination from the IP header in the mbuf. */ 2182 ip = mtod(m, struct ip *); 2183 bzero(&dst, sizeof(union sockaddr_union)); 2184 dst.sa.sa_len = sizeof(struct sockaddr_in); 2185 dst.sa.sa_family = AF_INET; 2186 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2187 ip->ip_src : ip->ip_dst; 2188 2189 /* Look up an SADB entry which matches the address of the peer. */ 2190 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2191 if (sav == NULL) { 2192 printf("%s: SADB lookup failed for %s\n", __func__, 2193 inet_ntoa(dst.sin.sin_addr)); 2194 return (EINVAL); 2195 } 2196 2197 MD5Init(&ctx); 2198 ipovly = (struct ipovly *)ip; 2199 th = (struct tcphdr *)((u_char *)ip + off0); 2200 doff = off0 + sizeof(struct tcphdr) + optlen; 2201 2202 /* 2203 * Step 1: Update MD5 hash with IP pseudo-header. 2204 * 2205 * XXX The ippseudo header MUST be digested in network byte order, 2206 * or else we'll fail the regression test. Assume all fields we've 2207 * been doing arithmetic on have been in host byte order. 2208 * XXX One cannot depend on ipovly->ih_len here. When called from 2209 * tcp_output(), the underlying ip_len member has not yet been set. 2210 */ 2211 ippseudo.ippseudo_src = ipovly->ih_src; 2212 ippseudo.ippseudo_dst = ipovly->ih_dst; 2213 ippseudo.ippseudo_pad = 0; 2214 ippseudo.ippseudo_p = IPPROTO_TCP; 2215 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2216 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2217 2218 /* 2219 * Step 2: Update MD5 hash with TCP header, excluding options. 2220 * The TCP checksum must be set to zero. 2221 */ 2222 savecsum = th->th_sum; 2223 th->th_sum = 0; 2224 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2225 th->th_sum = savecsum; 2226 2227 /* 2228 * Step 3: Update MD5 hash with TCP segment data. 2229 * Use m_apply() to avoid an early m_pullup(). 2230 */ 2231 if (len > 0) 2232 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2233 2234 /* 2235 * Step 4: Update MD5 hash with shared secret. 2236 */ 2237 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2238 MD5Final(buf, &ctx); 2239 2240 key_sa_recordxfer(sav, m); 2241 KEY_FREESAV(&sav); 2242 return (0); 2243} 2244#endif /* TCP_SIGNATURE */ 2245 2246static int 2247sysctl_drop(SYSCTL_HANDLER_ARGS) 2248{ 2249 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2250 struct sockaddr_storage addrs[2]; 2251 struct inpcb *inp; 2252 struct tcpcb *tp; 2253 struct tcptw *tw; 2254 struct sockaddr_in *fin, *lin; 2255#ifdef INET6 2256 struct sockaddr_in6 *fin6, *lin6; 2257 struct in6_addr f6, l6; 2258#endif 2259 int error; 2260 2261 inp = NULL; 2262 fin = lin = NULL; 2263#ifdef INET6 2264 fin6 = lin6 = NULL; 2265#endif 2266 error = 0; 2267 2268 if (req->oldptr != NULL || req->oldlen != 0) 2269 return (EINVAL); 2270 if (req->newptr == NULL) 2271 return (EPERM); 2272 if (req->newlen < sizeof(addrs)) 2273 return (ENOMEM); 2274 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2275 if (error) 2276 return (error); 2277 2278 switch (addrs[0].ss_family) { 2279#ifdef INET6 2280 case AF_INET6: 2281 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2282 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2283 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2284 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2285 return (EINVAL); 2286 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2287 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2288 return (EINVAL); 2289 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2290 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2291 fin = (struct sockaddr_in *)&addrs[0]; 2292 lin = (struct sockaddr_in *)&addrs[1]; 2293 break; 2294 } 2295 error = sa6_embedscope(fin6, ip6_use_defzone); 2296 if (error) 2297 return (error); 2298 error = sa6_embedscope(lin6, ip6_use_defzone); 2299 if (error) 2300 return (error); 2301 break; 2302#endif 2303 case AF_INET: 2304 fin = (struct sockaddr_in *)&addrs[0]; 2305 lin = (struct sockaddr_in *)&addrs[1]; 2306 if (fin->sin_len != sizeof(struct sockaddr_in) || 2307 lin->sin_len != sizeof(struct sockaddr_in)) 2308 return (EINVAL); 2309 break; 2310 default: 2311 return (EINVAL); 2312 } 2313 INP_INFO_WLOCK(&tcbinfo); 2314 switch (addrs[0].ss_family) { 2315#ifdef INET6 2316 case AF_INET6: 2317 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2318 &l6, lin6->sin6_port, 0, NULL); 2319 break; 2320#endif 2321 case AF_INET: 2322 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2323 lin->sin_addr, lin->sin_port, 0, NULL); 2324 break; 2325 } 2326 if (inp != NULL) { 2327 INP_LOCK(inp); 2328 if (inp->inp_vflag & INP_TIMEWAIT) { 2329 /* 2330 * XXXRW: There currently exists a state where an 2331 * inpcb is present, but its timewait state has been 2332 * discarded. For now, don't allow dropping of this 2333 * type of inpcb. 2334 */ 2335 tw = intotw(inp); 2336 if (tw != NULL) 2337 tcp_twclose(tw, 0); 2338 } else if (!(inp->inp_vflag & INP_DROPPED) && 2339 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2340 tp = intotcpcb(inp); 2341 tcp_drop(tp, ECONNABORTED); 2342 } 2343 INP_UNLOCK(inp); 2344 } else 2345 error = ESRCH; 2346 INP_INFO_WUNLOCK(&tcbinfo); 2347 return (error); 2348} 2349 2350SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2351 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2352 0, sysctl_drop, "", "Drop TCP connection"); 2353