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 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 34 * $FreeBSD$ 35 */ 36 37 #include "opt_compat.h" 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 #include "opt_mac.h" 41 #include "opt_tcpdebug.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/callout.h> 46 #include <sys/kernel.h> 47 #include <sys/sysctl.h> 48 #include <sys/mac.h> 49 #include <sys/malloc.h> 50 #include <sys/mbuf.h> 51 #ifdef INET6 52 #include <sys/domain.h> 53 #endif 54 #include <sys/proc.h> 55 #include <sys/socket.h> 56 #include <sys/socketvar.h> 57 #include <sys/protosw.h> 58 #include <sys/random.h> 59 60 #include <vm/uma.h> 61 62 #include <net/route.h> 63 #include <net/if.h> 64 65 #include <netinet/in.h> 66 #include <netinet/in_systm.h> 67 #include <netinet/ip.h> 68 #ifdef INET6 69 #include <netinet/ip6.h> 70 #endif 71 #include <netinet/in_pcb.h> 72 #ifdef INET6 73 #include <netinet6/in6_pcb.h> 74 #endif 75 #include <netinet/in_var.h> 76 #include <netinet/ip_var.h> 77 #ifdef INET6 78 #include <netinet6/ip6_var.h> 79 #endif 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 #endif /*IPSEC*/ 100 101 #ifdef FAST_IPSEC 102 #include <netipsec/ipsec.h> 103 #ifdef INET6 104 #include <netipsec/ipsec6.h> 105 #endif 106 #define IPSEC 107 #endif /*FAST_IPSEC*/ 108 109 #include <machine/in_cksum.h> 110 #include <sys/md5.h> 111 112 int tcp_mssdflt = TCP_MSS; 113 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 114 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); 115 116 #ifdef INET6 117 int tcp_v6mssdflt = TCP6_MSS; 118 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 119 CTLFLAG_RW, &tcp_v6mssdflt , 0, 120 "Default TCP Maximum Segment Size for IPv6"); 121 #endif 122 123 #if 0 124 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 125 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 126 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 127 #endif 128 129 int tcp_do_rfc1323 = 1; 130 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 131 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 132 133 int tcp_do_rfc1644 = 0; 134 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW, 135 &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions"); 136 137 static int tcp_tcbhashsize = 0; 138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD, 139 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 140 141 static int do_tcpdrain = 1; 142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 143 "Enable tcp_drain routine for extra help when low on mbufs"); 144 145 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 146 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 147 148 static int icmp_may_rst = 1; 149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 150 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 151 152 static int tcp_isn_reseed_interval = 0; 153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 154 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 155 156 /* 157 * TCP bandwidth limiting sysctls. Note that the default lower bound of 158 * 1024 exists only for debugging. A good production default would be 159 * something like 6100. 160 */ 161 static int tcp_inflight_enable = 0; 162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_enable, CTLFLAG_RW, 163 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 164 165 static int tcp_inflight_debug = 0; 166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_debug, CTLFLAG_RW, 167 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 168 169 static int tcp_inflight_min = 1024; 170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_min, CTLFLAG_RW, 171 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 172 173 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_max, CTLFLAG_RW, 175 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 176 177 static void tcp_cleartaocache(void); 178 static struct inpcb *tcp_notify(struct inpcb *, int); 179 180 /* 181 * Target size of TCP PCB hash tables. Must be a power of two. 182 * 183 * Note that this can be overridden by the kernel environment 184 * variable net.inet.tcp.tcbhashsize 185 */ 186 #ifndef TCBHASHSIZE 187 #define TCBHASHSIZE 512 188 #endif 189 190 /* 191 * This is the actual shape of what we allocate using the zone 192 * allocator. Doing it this way allows us to protect both structures 193 * using the same generation count, and also eliminates the overhead 194 * of allocating tcpcbs separately. By hiding the structure here, 195 * we avoid changing most of the rest of the code (although it needs 196 * to be changed, eventually, for greater efficiency). 197 */ 198 #define ALIGNMENT 32 199 #define ALIGNM1 (ALIGNMENT - 1) 200 struct inp_tp { 201 union { 202 struct inpcb inp; 203 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1]; 204 } inp_tp_u; 205 struct tcpcb tcb; 206 struct callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl; 207 struct callout inp_tp_delack; 208 }; 209 #undef ALIGNMENT 210 #undef ALIGNM1 211 212 /* 213 * Tcp initialization 214 */ 215 void 216 tcp_init() 217 { 218 int hashsize = TCBHASHSIZE; 219 220 tcp_ccgen = 1; 221 tcp_cleartaocache(); 222 223 tcp_delacktime = TCPTV_DELACK; 224 tcp_keepinit = TCPTV_KEEP_INIT; 225 tcp_keepidle = TCPTV_KEEP_IDLE; 226 tcp_keepintvl = TCPTV_KEEPINTVL; 227 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 228 tcp_msl = TCPTV_MSL; 229 tcp_rexmit_min = TCPTV_MIN; 230 tcp_rexmit_slop = TCPTV_CPU_VAR; 231 232 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 233 LIST_INIT(&tcb); 234 tcbinfo.listhead = &tcb; 235 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 236 if (!powerof2(hashsize)) { 237 printf("WARNING: TCB hash size not a power of 2\n"); 238 hashsize = 512; /* safe default */ 239 } 240 tcp_tcbhashsize = hashsize; 241 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 242 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 243 &tcbinfo.porthashmask); 244 tcbinfo.ipi_zone = uma_zcreate("tcpcb", sizeof(struct inp_tp), 245 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 246 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 247 #ifdef INET6 248 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 249 #else /* INET6 */ 250 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 251 #endif /* INET6 */ 252 if (max_protohdr < TCP_MINPROTOHDR) 253 max_protohdr = TCP_MINPROTOHDR; 254 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 255 panic("tcp_init"); 256 #undef TCP_MINPROTOHDR 257 258 syncache_init(); 259 } 260 261 /* 262 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 263 * tcp_template used to store this data in mbufs, but we now recopy it out 264 * of the tcpcb each time to conserve mbufs. 265 */ 266 void 267 tcp_fillheaders(tp, ip_ptr, tcp_ptr) 268 struct tcpcb *tp; 269 void *ip_ptr; 270 void *tcp_ptr; 271 { 272 struct inpcb *inp = tp->t_inpcb; 273 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr; 274 275 #ifdef INET6 276 if ((inp->inp_vflag & INP_IPV6) != 0) { 277 struct ip6_hdr *ip6; 278 279 ip6 = (struct ip6_hdr *)ip_ptr; 280 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 281 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 282 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 283 (IPV6_VERSION & IPV6_VERSION_MASK); 284 ip6->ip6_nxt = IPPROTO_TCP; 285 ip6->ip6_plen = sizeof(struct tcphdr); 286 ip6->ip6_src = inp->in6p_laddr; 287 ip6->ip6_dst = inp->in6p_faddr; 288 tcp_hdr->th_sum = 0; 289 } else 290 #endif 291 { 292 struct ip *ip = (struct ip *) ip_ptr; 293 294 ip->ip_v = IPVERSION; 295 ip->ip_hl = 5; 296 ip->ip_tos = 0; 297 ip->ip_len = 0; 298 ip->ip_id = 0; 299 ip->ip_off = 0; 300 ip->ip_ttl = 0; 301 ip->ip_sum = 0; 302 ip->ip_p = IPPROTO_TCP; 303 ip->ip_src = inp->inp_laddr; 304 ip->ip_dst = inp->inp_faddr; 305 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 306 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 307 } 308 309 tcp_hdr->th_sport = inp->inp_lport; 310 tcp_hdr->th_dport = inp->inp_fport; 311 tcp_hdr->th_seq = 0; 312 tcp_hdr->th_ack = 0; 313 tcp_hdr->th_x2 = 0; 314 tcp_hdr->th_off = 5; 315 tcp_hdr->th_flags = 0; 316 tcp_hdr->th_win = 0; 317 tcp_hdr->th_urp = 0; 318 } 319 320 /* 321 * Create template to be used to send tcp packets on a connection. 322 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 323 * use for this function is in keepalives, which use tcp_respond. 324 */ 325 struct tcptemp * 326 tcp_maketemplate(tp) 327 struct tcpcb *tp; 328 { 329 struct mbuf *m; 330 struct tcptemp *n; 331 332 m = m_get(M_DONTWAIT, MT_HEADER); 333 if (m == NULL) 334 return (0); 335 m->m_len = sizeof(struct tcptemp); 336 n = mtod(m, struct tcptemp *); 337 338 tcp_fillheaders(tp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 339 return (n); 340 } 341 342 /* 343 * Send a single message to the TCP at address specified by 344 * the given TCP/IP header. If m == 0, then we make a copy 345 * of the tcpiphdr at ti and send directly to the addressed host. 346 * This is used to force keep alive messages out using the TCP 347 * template for a connection. If flags are given then we send 348 * a message back to the TCP which originated the * segment ti, 349 * and discard the mbuf containing it and any other attached mbufs. 350 * 351 * In any case the ack and sequence number of the transmitted 352 * segment are as specified by the parameters. 353 * 354 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 355 */ 356 void 357 tcp_respond(tp, ipgen, th, m, ack, seq, flags) 358 struct tcpcb *tp; 359 void *ipgen; 360 register struct tcphdr *th; 361 register struct mbuf *m; 362 tcp_seq ack, seq; 363 int flags; 364 { 365 register int tlen; 366 int win = 0; 367 struct route *ro = 0; 368 struct route sro; 369 struct ip *ip; 370 struct tcphdr *nth; 371 #ifdef INET6 372 struct route_in6 *ro6 = 0; 373 struct route_in6 sro6; 374 struct ip6_hdr *ip6; 375 int isipv6; 376 #endif /* INET6 */ 377 int ipflags = 0; 378 379 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 380 381 #ifdef INET6 382 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 383 ip6 = ipgen; 384 #endif /* INET6 */ 385 ip = ipgen; 386 387 if (tp) { 388 if (!(flags & TH_RST)) { 389 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 390 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 391 win = (long)TCP_MAXWIN << tp->rcv_scale; 392 } 393 #ifdef INET6 394 if (isipv6) 395 ro6 = &tp->t_inpcb->in6p_route; 396 else 397 #endif /* INET6 */ 398 ro = &tp->t_inpcb->inp_route; 399 } else { 400 #ifdef INET6 401 if (isipv6) { 402 ro6 = &sro6; 403 bzero(ro6, sizeof *ro6); 404 } else 405 #endif /* INET6 */ 406 { 407 ro = &sro; 408 bzero(ro, sizeof *ro); 409 } 410 } 411 if (m == 0) { 412 m = m_gethdr(M_DONTWAIT, MT_HEADER); 413 if (m == NULL) 414 return; 415 tlen = 0; 416 m->m_data += max_linkhdr; 417 #ifdef INET6 418 if (isipv6) { 419 bcopy((caddr_t)ip6, mtod(m, caddr_t), 420 sizeof(struct ip6_hdr)); 421 ip6 = mtod(m, struct ip6_hdr *); 422 nth = (struct tcphdr *)(ip6 + 1); 423 } else 424 #endif /* INET6 */ 425 { 426 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 427 ip = mtod(m, struct ip *); 428 nth = (struct tcphdr *)(ip + 1); 429 } 430 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 431 flags = TH_ACK; 432 } else { 433 m_freem(m->m_next); 434 m->m_next = 0; 435 m->m_data = (caddr_t)ipgen; 436 /* m_len is set later */ 437 tlen = 0; 438 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 439 #ifdef INET6 440 if (isipv6) { 441 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 442 nth = (struct tcphdr *)(ip6 + 1); 443 } else 444 #endif /* INET6 */ 445 { 446 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 447 nth = (struct tcphdr *)(ip + 1); 448 } 449 if (th != nth) { 450 /* 451 * this is usually a case when an extension header 452 * exists between the IPv6 header and the 453 * TCP header. 454 */ 455 nth->th_sport = th->th_sport; 456 nth->th_dport = th->th_dport; 457 } 458 xchg(nth->th_dport, nth->th_sport, n_short); 459 #undef xchg 460 } 461 #ifdef INET6 462 if (isipv6) { 463 ip6->ip6_flow = 0; 464 ip6->ip6_vfc = IPV6_VERSION; 465 ip6->ip6_nxt = IPPROTO_TCP; 466 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 467 tlen)); 468 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 469 } else 470 #endif 471 { 472 tlen += sizeof (struct tcpiphdr); 473 ip->ip_len = tlen; 474 ip->ip_ttl = ip_defttl; 475 } 476 m->m_len = tlen; 477 m->m_pkthdr.len = tlen; 478 m->m_pkthdr.rcvif = (struct ifnet *) 0; 479 #ifdef MAC 480 if (tp != NULL) { 481 /* 482 * Packet is associated with a socket, so allow the 483 * label of the response to reflect the socket label. 484 */ 485 mac_create_mbuf_from_socket(tp->t_inpcb->inp_socket, m); 486 } else { 487 /* 488 * XXXMAC: This will need to call a mac function that 489 * modifies the mbuf label in place for TCP datagrams 490 * not associated with a PCB. 491 */ 492 } 493 #endif 494 nth->th_seq = htonl(seq); 495 nth->th_ack = htonl(ack); 496 nth->th_x2 = 0; 497 nth->th_off = sizeof (struct tcphdr) >> 2; 498 nth->th_flags = flags; 499 if (tp) 500 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 501 else 502 nth->th_win = htons((u_short)win); 503 nth->th_urp = 0; 504 #ifdef INET6 505 if (isipv6) { 506 nth->th_sum = 0; 507 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 508 sizeof(struct ip6_hdr), 509 tlen - sizeof(struct ip6_hdr)); 510 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL, 511 ro6 && ro6->ro_rt ? 512 ro6->ro_rt->rt_ifp : 513 NULL); 514 } else 515 #endif /* INET6 */ 516 { 517 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 518 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 519 m->m_pkthdr.csum_flags = CSUM_TCP; 520 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 521 } 522 #ifdef TCPDEBUG 523 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 524 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 525 #endif 526 #ifdef INET6 527 if (isipv6) { 528 (void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL, 529 tp ? tp->t_inpcb : NULL); 530 if (ro6 == &sro6 && ro6->ro_rt) { 531 RTFREE(ro6->ro_rt); 532 ro6->ro_rt = NULL; 533 } 534 } else 535 #endif /* INET6 */ 536 { 537 (void) ip_output(m, NULL, ro, ipflags, NULL, tp ? tp->t_inpcb : NULL); 538 if (ro == &sro && ro->ro_rt) { 539 RTFREE(ro->ro_rt); 540 ro->ro_rt = NULL; 541 } 542 } 543 } 544 545 /* 546 * Create a new TCP control block, making an 547 * empty reassembly queue and hooking it to the argument 548 * protocol control block. The `inp' parameter must have 549 * come from the zone allocator set up in tcp_init(). 550 */ 551 struct tcpcb * 552 tcp_newtcpcb(inp) 553 struct inpcb *inp; 554 { 555 struct inp_tp *it; 556 register struct tcpcb *tp; 557 #ifdef INET6 558 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 559 #endif /* INET6 */ 560 561 it = (struct inp_tp *)inp; 562 tp = &it->tcb; 563 bzero((char *) tp, sizeof(struct tcpcb)); 564 LIST_INIT(&tp->t_segq); 565 tp->t_maxseg = tp->t_maxopd = 566 #ifdef INET6 567 isipv6 ? tcp_v6mssdflt : 568 #endif /* INET6 */ 569 tcp_mssdflt; 570 571 /* Set up our timeouts. */ 572 callout_init(tp->tt_rexmt = &it->inp_tp_rexmt, 0); 573 callout_init(tp->tt_persist = &it->inp_tp_persist, 0); 574 callout_init(tp->tt_keep = &it->inp_tp_keep, 0); 575 callout_init(tp->tt_2msl = &it->inp_tp_2msl, 0); 576 callout_init(tp->tt_delack = &it->inp_tp_delack, 0); 577 578 if (tcp_do_rfc1323) 579 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 580 if (tcp_do_rfc1644) 581 tp->t_flags |= TF_REQ_CC; 582 tp->t_inpcb = inp; /* XXX */ 583 /* 584 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 585 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 586 * reasonable initial retransmit time. 587 */ 588 tp->t_srtt = TCPTV_SRTTBASE; 589 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 590 tp->t_rttmin = tcp_rexmit_min; 591 tp->t_rxtcur = TCPTV_RTOBASE; 592 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 593 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 594 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 595 tp->t_rcvtime = ticks; 596 tp->t_bw_rtttime = ticks; 597 /* 598 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 599 * because the socket may be bound to an IPv6 wildcard address, 600 * which may match an IPv4-mapped IPv6 address. 601 */ 602 inp->inp_ip_ttl = ip_defttl; 603 inp->inp_ppcb = (caddr_t)tp; 604 return (tp); /* XXX */ 605 } 606 607 /* 608 * Drop a TCP connection, reporting 609 * the specified error. If connection is synchronized, 610 * then send a RST to peer. 611 */ 612 struct tcpcb * 613 tcp_drop(tp, errno) 614 register struct tcpcb *tp; 615 int errno; 616 { 617 struct socket *so = tp->t_inpcb->inp_socket; 618 619 if (TCPS_HAVERCVDSYN(tp->t_state)) { 620 tp->t_state = TCPS_CLOSED; 621 (void) tcp_output(tp); 622 tcpstat.tcps_drops++; 623 } else 624 tcpstat.tcps_conndrops++; 625 if (errno == ETIMEDOUT && tp->t_softerror) 626 errno = tp->t_softerror; 627 so->so_error = errno; 628 return (tcp_close(tp)); 629 } 630 631 /* 632 * Close a TCP control block: 633 * discard all space held by the tcp 634 * discard internet protocol block 635 * wake up any sleepers 636 */ 637 struct tcpcb * 638 tcp_close(tp) 639 register struct tcpcb *tp; 640 { 641 register struct tseg_qent *q; 642 struct inpcb *inp = tp->t_inpcb; 643 struct socket *so = inp->inp_socket; 644 #ifdef INET6 645 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 646 #endif /* INET6 */ 647 register struct rtentry *rt; 648 int dosavessthresh; 649 650 /* 651 * Make sure that all of our timers are stopped before we 652 * delete the PCB. 653 */ 654 callout_stop(tp->tt_rexmt); 655 callout_stop(tp->tt_persist); 656 callout_stop(tp->tt_keep); 657 callout_stop(tp->tt_2msl); 658 callout_stop(tp->tt_delack); 659 660 /* 661 * If we got enough samples through the srtt filter, 662 * save the rtt and rttvar in the routing entry. 663 * 'Enough' is arbitrarily defined as the 16 samples. 664 * 16 samples is enough for the srtt filter to converge 665 * to within 5% of the correct value; fewer samples and 666 * we could save a very bogus rtt. 667 * 668 * Don't update the default route's characteristics and don't 669 * update anything that the user "locked". 670 */ 671 if (tp->t_rttupdated >= 16) { 672 register u_long i = 0; 673 #ifdef INET6 674 if (isipv6) { 675 struct sockaddr_in6 *sin6; 676 677 if ((rt = inp->in6p_route.ro_rt) == NULL) 678 goto no_valid_rt; 679 sin6 = (struct sockaddr_in6 *)rt_key(rt); 680 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 681 goto no_valid_rt; 682 } 683 else 684 #endif /* INET6 */ 685 if ((rt = inp->inp_route.ro_rt) == NULL || 686 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr 687 == INADDR_ANY) 688 goto no_valid_rt; 689 690 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 691 i = tp->t_srtt * 692 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 693 if (rt->rt_rmx.rmx_rtt && i) 694 /* 695 * filter this update to half the old & half 696 * the new values, converting scale. 697 * See route.h and tcp_var.h for a 698 * description of the scaling constants. 699 */ 700 rt->rt_rmx.rmx_rtt = 701 (rt->rt_rmx.rmx_rtt + i) / 2; 702 else 703 rt->rt_rmx.rmx_rtt = i; 704 tcpstat.tcps_cachedrtt++; 705 } 706 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 707 i = tp->t_rttvar * 708 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 709 if (rt->rt_rmx.rmx_rttvar && i) 710 rt->rt_rmx.rmx_rttvar = 711 (rt->rt_rmx.rmx_rttvar + i) / 2; 712 else 713 rt->rt_rmx.rmx_rttvar = i; 714 tcpstat.tcps_cachedrttvar++; 715 } 716 /* 717 * The old comment here said: 718 * update the pipelimit (ssthresh) if it has been updated 719 * already or if a pipesize was specified & the threshhold 720 * got below half the pipesize. I.e., wait for bad news 721 * before we start updating, then update on both good 722 * and bad news. 723 * 724 * But we want to save the ssthresh even if no pipesize is 725 * specified explicitly in the route, because such 726 * connections still have an implicit pipesize specified 727 * by the global tcp_sendspace. In the absence of a reliable 728 * way to calculate the pipesize, it will have to do. 729 */ 730 i = tp->snd_ssthresh; 731 if (rt->rt_rmx.rmx_sendpipe != 0) 732 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2); 733 else 734 dosavessthresh = (i < so->so_snd.sb_hiwat / 2); 735 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 736 i != 0 && rt->rt_rmx.rmx_ssthresh != 0) 737 || dosavessthresh) { 738 /* 739 * convert the limit from user data bytes to 740 * packets then to packet data bytes. 741 */ 742 i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 743 if (i < 2) 744 i = 2; 745 i *= (u_long)(tp->t_maxseg + 746 #ifdef INET6 747 (isipv6 ? sizeof (struct ip6_hdr) + 748 sizeof (struct tcphdr) : 749 #endif 750 sizeof (struct tcpiphdr) 751 #ifdef INET6 752 ) 753 #endif 754 ); 755 if (rt->rt_rmx.rmx_ssthresh) 756 rt->rt_rmx.rmx_ssthresh = 757 (rt->rt_rmx.rmx_ssthresh + i) / 2; 758 else 759 rt->rt_rmx.rmx_ssthresh = i; 760 tcpstat.tcps_cachedssthresh++; 761 } 762 } 763 no_valid_rt: 764 /* free the reassembly queue, if any */ 765 while((q = LIST_FIRST(&tp->t_segq)) != NULL) { 766 LIST_REMOVE(q, tqe_q); 767 m_freem(q->tqe_m); 768 FREE(q, M_TSEGQ); 769 } 770 inp->inp_ppcb = NULL; 771 soisdisconnected(so); 772 #ifdef INET6 773 if (INP_CHECK_SOCKAF(so, AF_INET6)) 774 in6_pcbdetach(inp); 775 else 776 #endif /* INET6 */ 777 in_pcbdetach(inp); 778 tcpstat.tcps_closed++; 779 return ((struct tcpcb *)0); 780 } 781 782 void 783 tcp_drain() 784 { 785 if (do_tcpdrain) 786 { 787 struct inpcb *inpb; 788 struct tcpcb *tcpb; 789 struct tseg_qent *te; 790 791 /* 792 * Walk the tcpbs, if existing, and flush the reassembly queue, 793 * if there is one... 794 * XXX: The "Net/3" implementation doesn't imply that the TCP 795 * reassembly queue should be flushed, but in a situation 796 * where we're really low on mbufs, this is potentially 797 * usefull. 798 */ 799 INP_INFO_RLOCK(&tcbinfo); 800 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 801 INP_LOCK(inpb); 802 if ((tcpb = intotcpcb(inpb))) { 803 while ((te = LIST_FIRST(&tcpb->t_segq)) 804 != NULL) { 805 LIST_REMOVE(te, tqe_q); 806 m_freem(te->tqe_m); 807 FREE(te, M_TSEGQ); 808 } 809 } 810 INP_UNLOCK(inpb); 811 } 812 INP_INFO_RUNLOCK(&tcbinfo); 813 } 814 } 815 816 /* 817 * Notify a tcp user of an asynchronous error; 818 * store error as soft error, but wake up user 819 * (for now, won't do anything until can select for soft error). 820 * 821 * Do not wake up user since there currently is no mechanism for 822 * reporting soft errors (yet - a kqueue filter may be added). 823 */ 824 static struct inpcb * 825 tcp_notify(inp, error) 826 struct inpcb *inp; 827 int error; 828 { 829 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 830 831 /* 832 * Ignore some errors if we are hooked up. 833 * If connection hasn't completed, has retransmitted several times, 834 * and receives a second error, give up now. This is better 835 * than waiting a long time to establish a connection that 836 * can never complete. 837 */ 838 if (tp->t_state == TCPS_ESTABLISHED && 839 (error == EHOSTUNREACH || error == ENETUNREACH || 840 error == EHOSTDOWN)) { 841 return inp; 842 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 843 tp->t_softerror) { 844 tcp_drop(tp, error); 845 return (struct inpcb *)0; 846 } else { 847 tp->t_softerror = error; 848 return inp; 849 } 850 #if 0 851 wakeup((caddr_t) &so->so_timeo); 852 sorwakeup(so); 853 sowwakeup(so); 854 #endif 855 } 856 857 static int 858 tcp_pcblist(SYSCTL_HANDLER_ARGS) 859 { 860 int error, i, n, s; 861 struct inpcb *inp, **inp_list; 862 inp_gen_t gencnt; 863 struct xinpgen xig; 864 865 /* 866 * The process of preparing the TCB list is too time-consuming and 867 * resource-intensive to repeat twice on every request. 868 */ 869 if (req->oldptr == 0) { 870 n = tcbinfo.ipi_count; 871 req->oldidx = 2 * (sizeof xig) 872 + (n + n/8) * sizeof(struct xtcpcb); 873 return 0; 874 } 875 876 if (req->newptr != 0) 877 return EPERM; 878 879 /* 880 * OK, now we're committed to doing something. 881 */ 882 s = splnet(); 883 INP_INFO_RLOCK(&tcbinfo); 884 gencnt = tcbinfo.ipi_gencnt; 885 n = tcbinfo.ipi_count; 886 INP_INFO_RUNLOCK(&tcbinfo); 887 splx(s); 888 889 sysctl_wire_old_buffer(req, 2 * (sizeof xig) 890 + n * sizeof(struct xtcpcb)); 891 892 xig.xig_len = sizeof xig; 893 xig.xig_count = n; 894 xig.xig_gen = gencnt; 895 xig.xig_sogen = so_gencnt; 896 error = SYSCTL_OUT(req, &xig, sizeof xig); 897 if (error) 898 return error; 899 900 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 901 if (inp_list == 0) 902 return ENOMEM; 903 904 s = splnet(); 905 INP_INFO_RLOCK(&tcbinfo); 906 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp && i < n; 907 inp = LIST_NEXT(inp, inp_list)) { 908 INP_LOCK(inp); 909 if (inp->inp_gencnt <= gencnt && 910 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0) 911 inp_list[i++] = inp; 912 INP_UNLOCK(inp); 913 } 914 INP_INFO_RUNLOCK(&tcbinfo); 915 splx(s); 916 n = i; 917 918 error = 0; 919 for (i = 0; i < n; i++) { 920 inp = inp_list[i]; 921 INP_LOCK(inp); 922 if (inp->inp_gencnt <= gencnt) { 923 struct xtcpcb xt; 924 caddr_t inp_ppcb; 925 xt.xt_len = sizeof xt; 926 /* XXX should avoid extra copy */ 927 bcopy(inp, &xt.xt_inp, sizeof *inp); 928 inp_ppcb = inp->inp_ppcb; 929 if (inp_ppcb != NULL) 930 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 931 else 932 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 933 if (inp->inp_socket) 934 sotoxsocket(inp->inp_socket, &xt.xt_socket); 935 error = SYSCTL_OUT(req, &xt, sizeof xt); 936 } 937 INP_UNLOCK(inp); 938 } 939 if (!error) { 940 /* 941 * Give the user an updated idea of our state. 942 * If the generation differs from what we told 943 * her before, she knows that something happened 944 * while we were processing this request, and it 945 * might be necessary to retry. 946 */ 947 s = splnet(); 948 INP_INFO_RLOCK(&tcbinfo); 949 xig.xig_gen = tcbinfo.ipi_gencnt; 950 xig.xig_sogen = so_gencnt; 951 xig.xig_count = tcbinfo.ipi_count; 952 INP_INFO_RUNLOCK(&tcbinfo); 953 splx(s); 954 error = SYSCTL_OUT(req, &xig, sizeof xig); 955 } 956 free(inp_list, M_TEMP); 957 return error; 958 } 959 960 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 961 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 962 963 static int 964 tcp_getcred(SYSCTL_HANDLER_ARGS) 965 { 966 struct xucred xuc; 967 struct sockaddr_in addrs[2]; 968 struct inpcb *inp; 969 int error, s; 970 971 error = suser_cred(req->td->td_ucred, PRISON_ROOT); 972 if (error) 973 return (error); 974 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 975 if (error) 976 return (error); 977 s = splnet(); 978 INP_INFO_RLOCK(&tcbinfo); 979 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 980 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 981 if (inp == NULL) { 982 error = ENOENT; 983 goto outunlocked; 984 } 985 INP_LOCK(inp); 986 if (inp->inp_socket == NULL) { 987 error = ENOENT; 988 goto out; 989 } 990 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 991 if (error) 992 goto out; 993 cru2x(inp->inp_socket->so_cred, &xuc); 994 out: 995 INP_UNLOCK(inp); 996 outunlocked: 997 INP_INFO_RUNLOCK(&tcbinfo); 998 splx(s); 999 if (error == 0) 1000 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1001 return (error); 1002 } 1003 1004 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1005 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1006 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1007 1008 #ifdef INET6 1009 static int 1010 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1011 { 1012 struct xucred xuc; 1013 struct sockaddr_in6 addrs[2]; 1014 struct inpcb *inp; 1015 int error, s, mapped = 0; 1016 1017 error = suser_cred(req->td->td_ucred, PRISON_ROOT); 1018 if (error) 1019 return (error); 1020 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1021 if (error) 1022 return (error); 1023 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1024 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1025 mapped = 1; 1026 else 1027 return (EINVAL); 1028 } 1029 s = splnet(); 1030 INP_INFO_RLOCK(&tcbinfo); 1031 if (mapped == 1) 1032 inp = in_pcblookup_hash(&tcbinfo, 1033 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1034 addrs[1].sin6_port, 1035 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1036 addrs[0].sin6_port, 1037 0, NULL); 1038 else 1039 inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr, 1040 addrs[1].sin6_port, 1041 &addrs[0].sin6_addr, addrs[0].sin6_port, 1042 0, NULL); 1043 if (inp == NULL) { 1044 error = ENOENT; 1045 goto outunlocked; 1046 } 1047 INP_LOCK(inp); 1048 if (inp->inp_socket == NULL) { 1049 error = ENOENT; 1050 goto out; 1051 } 1052 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1053 if (error) 1054 goto out; 1055 cru2x(inp->inp_socket->so_cred, &xuc); 1056 out: 1057 INP_UNLOCK(inp); 1058 outunlocked: 1059 INP_INFO_RUNLOCK(&tcbinfo); 1060 splx(s); 1061 if (error == 0) 1062 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1063 return (error); 1064 } 1065 1066 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1067 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1068 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1069 #endif 1070 1071 1072 void 1073 tcp_ctlinput(cmd, sa, vip) 1074 int cmd; 1075 struct sockaddr *sa; 1076 void *vip; 1077 { 1078 struct ip *ip = vip; 1079 struct tcphdr *th; 1080 struct in_addr faddr; 1081 struct inpcb *inp; 1082 struct tcpcb *tp; 1083 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1084 tcp_seq icmp_seq; 1085 int s; 1086 1087 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1088 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1089 return; 1090 1091 if (cmd == PRC_QUENCH) 1092 notify = tcp_quench; 1093 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1094 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1095 notify = tcp_drop_syn_sent; 1096 else if (cmd == PRC_MSGSIZE) 1097 notify = tcp_mtudisc; 1098 else if (PRC_IS_REDIRECT(cmd)) { 1099 ip = 0; 1100 notify = in_rtchange; 1101 } else if (cmd == PRC_HOSTDEAD) 1102 ip = 0; 1103 else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0) 1104 return; 1105 if (ip) { 1106 s = splnet(); 1107 th = (struct tcphdr *)((caddr_t)ip 1108 + (ip->ip_hl << 2)); 1109 INP_INFO_WLOCK(&tcbinfo); 1110 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1111 ip->ip_src, th->th_sport, 0, NULL); 1112 if (inp != NULL) { 1113 INP_LOCK(inp); 1114 if (inp->inp_socket != NULL) { 1115 icmp_seq = htonl(th->th_seq); 1116 tp = intotcpcb(inp); 1117 if (SEQ_GEQ(icmp_seq, tp->snd_una) && 1118 SEQ_LT(icmp_seq, tp->snd_max)) 1119 inp = (*notify)(inp, inetctlerrmap[cmd]); 1120 } 1121 if (inp) 1122 INP_UNLOCK(inp); 1123 } else { 1124 struct in_conninfo inc; 1125 1126 inc.inc_fport = th->th_dport; 1127 inc.inc_lport = th->th_sport; 1128 inc.inc_faddr = faddr; 1129 inc.inc_laddr = ip->ip_src; 1130 #ifdef INET6 1131 inc.inc_isipv6 = 0; 1132 #endif 1133 syncache_unreach(&inc, th); 1134 } 1135 INP_INFO_WUNLOCK(&tcbinfo); 1136 splx(s); 1137 } else 1138 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1139 } 1140 1141 #ifdef INET6 1142 void 1143 tcp6_ctlinput(cmd, sa, d) 1144 int cmd; 1145 struct sockaddr *sa; 1146 void *d; 1147 { 1148 struct tcphdr th; 1149 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1150 struct ip6_hdr *ip6; 1151 struct mbuf *m; 1152 struct ip6ctlparam *ip6cp = NULL; 1153 const struct sockaddr_in6 *sa6_src = NULL; 1154 int off; 1155 struct tcp_portonly { 1156 u_int16_t th_sport; 1157 u_int16_t th_dport; 1158 } *thp; 1159 1160 if (sa->sa_family != AF_INET6 || 1161 sa->sa_len != sizeof(struct sockaddr_in6)) 1162 return; 1163 1164 if (cmd == PRC_QUENCH) 1165 notify = tcp_quench; 1166 else if (cmd == PRC_MSGSIZE) 1167 notify = tcp_mtudisc; 1168 else if (!PRC_IS_REDIRECT(cmd) && 1169 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1170 return; 1171 1172 /* if the parameter is from icmp6, decode it. */ 1173 if (d != NULL) { 1174 ip6cp = (struct ip6ctlparam *)d; 1175 m = ip6cp->ip6c_m; 1176 ip6 = ip6cp->ip6c_ip6; 1177 off = ip6cp->ip6c_off; 1178 sa6_src = ip6cp->ip6c_src; 1179 } else { 1180 m = NULL; 1181 ip6 = NULL; 1182 off = 0; /* fool gcc */ 1183 sa6_src = &sa6_any; 1184 } 1185 1186 if (ip6) { 1187 struct in_conninfo inc; 1188 /* 1189 * XXX: We assume that when IPV6 is non NULL, 1190 * M and OFF are valid. 1191 */ 1192 1193 /* check if we can safely examine src and dst ports */ 1194 if (m->m_pkthdr.len < off + sizeof(*thp)) 1195 return; 1196 1197 bzero(&th, sizeof(th)); 1198 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1199 1200 in6_pcbnotify(&tcb, sa, th.th_dport, 1201 (struct sockaddr *)ip6cp->ip6c_src, 1202 th.th_sport, cmd, notify); 1203 1204 inc.inc_fport = th.th_dport; 1205 inc.inc_lport = th.th_sport; 1206 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1207 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1208 inc.inc_isipv6 = 1; 1209 syncache_unreach(&inc, &th); 1210 } else 1211 in6_pcbnotify(&tcb, sa, 0, (const struct sockaddr *)sa6_src, 1212 0, cmd, notify); 1213 } 1214 #endif /* INET6 */ 1215 1216 1217 /* 1218 * Following is where TCP initial sequence number generation occurs. 1219 * 1220 * There are two places where we must use initial sequence numbers: 1221 * 1. In SYN-ACK packets. 1222 * 2. In SYN packets. 1223 * 1224 * All ISNs for SYN-ACK packets are generated by the syncache. See 1225 * tcp_syncache.c for details. 1226 * 1227 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1228 * depends on this property. In addition, these ISNs should be 1229 * unguessable so as to prevent connection hijacking. To satisfy 1230 * the requirements of this situation, the algorithm outlined in 1231 * RFC 1948 is used to generate sequence numbers. 1232 * 1233 * Implementation details: 1234 * 1235 * Time is based off the system timer, and is corrected so that it 1236 * increases by one megabyte per second. This allows for proper 1237 * recycling on high speed LANs while still leaving over an hour 1238 * before rollover. 1239 * 1240 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1241 * between seeding of isn_secret. This is normally set to zero, 1242 * as reseeding should not be necessary. 1243 * 1244 */ 1245 1246 #define ISN_BYTES_PER_SECOND 1048576 1247 1248 u_char isn_secret[32]; 1249 int isn_last_reseed; 1250 MD5_CTX isn_ctx; 1251 1252 tcp_seq 1253 tcp_new_isn(tp) 1254 struct tcpcb *tp; 1255 { 1256 u_int32_t md5_buffer[4]; 1257 tcp_seq new_isn; 1258 1259 /* Seed if this is the first use, reseed if requested. */ 1260 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1261 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1262 < (u_int)ticks))) { 1263 read_random(&isn_secret, sizeof(isn_secret)); 1264 isn_last_reseed = ticks; 1265 } 1266 1267 /* Compute the md5 hash and return the ISN. */ 1268 MD5Init(&isn_ctx); 1269 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1270 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1271 #ifdef INET6 1272 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1273 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1274 sizeof(struct in6_addr)); 1275 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1276 sizeof(struct in6_addr)); 1277 } else 1278 #endif 1279 { 1280 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1281 sizeof(struct in_addr)); 1282 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1283 sizeof(struct in_addr)); 1284 } 1285 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1286 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1287 new_isn = (tcp_seq) md5_buffer[0]; 1288 new_isn += ticks * (ISN_BYTES_PER_SECOND / hz); 1289 return new_isn; 1290 } 1291 1292 /* 1293 * When a source quench is received, close congestion window 1294 * to one segment. We will gradually open it again as we proceed. 1295 */ 1296 struct inpcb * 1297 tcp_quench(inp, errno) 1298 struct inpcb *inp; 1299 int errno; 1300 { 1301 struct tcpcb *tp = intotcpcb(inp); 1302 1303 if (tp) 1304 tp->snd_cwnd = tp->t_maxseg; 1305 return (inp); 1306 } 1307 1308 /* 1309 * When a specific ICMP unreachable message is received and the 1310 * connection state is SYN-SENT, drop the connection. This behavior 1311 * is controlled by the icmp_may_rst sysctl. 1312 */ 1313 struct inpcb * 1314 tcp_drop_syn_sent(inp, errno) 1315 struct inpcb *inp; 1316 int errno; 1317 { 1318 struct tcpcb *tp = intotcpcb(inp); 1319 1320 if (tp && tp->t_state == TCPS_SYN_SENT) { 1321 tcp_drop(tp, errno); 1322 return (struct inpcb *)0; 1323 } 1324 return inp; 1325 } 1326 1327 /* 1328 * When `need fragmentation' ICMP is received, update our idea of the MSS 1329 * based on the new value in the route. Also nudge TCP to send something, 1330 * since we know the packet we just sent was dropped. 1331 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1332 */ 1333 struct inpcb * 1334 tcp_mtudisc(inp, errno) 1335 struct inpcb *inp; 1336 int errno; 1337 { 1338 struct tcpcb *tp = intotcpcb(inp); 1339 struct rtentry *rt; 1340 struct rmxp_tao *taop; 1341 struct socket *so = inp->inp_socket; 1342 int offered; 1343 int mss; 1344 #ifdef INET6 1345 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1346 #endif /* INET6 */ 1347 1348 if (tp) { 1349 #ifdef INET6 1350 if (isipv6) 1351 rt = tcp_rtlookup6(&inp->inp_inc); 1352 else 1353 #endif /* INET6 */ 1354 rt = tcp_rtlookup(&inp->inp_inc); 1355 if (!rt || !rt->rt_rmx.rmx_mtu) { 1356 tp->t_maxopd = tp->t_maxseg = 1357 #ifdef INET6 1358 isipv6 ? tcp_v6mssdflt : 1359 #endif /* INET6 */ 1360 tcp_mssdflt; 1361 return inp; 1362 } 1363 taop = rmx_taop(rt->rt_rmx); 1364 offered = taop->tao_mssopt; 1365 mss = rt->rt_rmx.rmx_mtu - 1366 #ifdef INET6 1367 (isipv6 ? 1368 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1369 #endif /* INET6 */ 1370 sizeof(struct tcpiphdr) 1371 #ifdef INET6 1372 ) 1373 #endif /* INET6 */ 1374 ; 1375 1376 if (offered) 1377 mss = min(mss, offered); 1378 /* 1379 * XXX - The above conditional probably violates the TCP 1380 * spec. The problem is that, since we don't know the 1381 * other end's MSS, we are supposed to use a conservative 1382 * default. But, if we do that, then MTU discovery will 1383 * never actually take place, because the conservative 1384 * default is much less than the MTUs typically seen 1385 * on the Internet today. For the moment, we'll sweep 1386 * this under the carpet. 1387 * 1388 * The conservative default might not actually be a problem 1389 * if the only case this occurs is when sending an initial 1390 * SYN with options and data to a host we've never talked 1391 * to before. Then, they will reply with an MSS value which 1392 * will get recorded and the new parameters should get 1393 * recomputed. For Further Study. 1394 */ 1395 if (tp->t_maxopd <= mss) 1396 return inp; 1397 tp->t_maxopd = mss; 1398 1399 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1400 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1401 mss -= TCPOLEN_TSTAMP_APPA; 1402 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 1403 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC) 1404 mss -= TCPOLEN_CC_APPA; 1405 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1406 if (mss > MCLBYTES) 1407 mss &= ~(MCLBYTES-1); 1408 #else 1409 if (mss > MCLBYTES) 1410 mss = mss / MCLBYTES * MCLBYTES; 1411 #endif 1412 if (so->so_snd.sb_hiwat < mss) 1413 mss = so->so_snd.sb_hiwat; 1414 1415 tp->t_maxseg = mss; 1416 1417 tcpstat.tcps_mturesent++; 1418 tp->t_rtttime = 0; 1419 tp->snd_nxt = tp->snd_una; 1420 tcp_output(tp); 1421 } 1422 return inp; 1423 } 1424 1425 /* 1426 * Look-up the routing entry to the peer of this inpcb. If no route 1427 * is found and it cannot be allocated the return NULL. This routine 1428 * is called by TCP routines that access the rmx structure and by tcp_mss 1429 * to get the interface MTU. 1430 */ 1431 struct rtentry * 1432 tcp_rtlookup(inc) 1433 struct in_conninfo *inc; 1434 { 1435 struct route *ro; 1436 struct rtentry *rt; 1437 1438 ro = &inc->inc_route; 1439 rt = ro->ro_rt; 1440 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 1441 /* No route yet, so try to acquire one */ 1442 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1443 ro->ro_dst.sa_family = AF_INET; 1444 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 1445 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1446 inc->inc_faddr; 1447 rtalloc(ro); 1448 rt = ro->ro_rt; 1449 } 1450 } 1451 return rt; 1452 } 1453 1454 #ifdef INET6 1455 struct rtentry * 1456 tcp_rtlookup6(inc) 1457 struct in_conninfo *inc; 1458 { 1459 struct route_in6 *ro6; 1460 struct rtentry *rt; 1461 1462 ro6 = &inc->inc6_route; 1463 rt = ro6->ro_rt; 1464 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 1465 /* No route yet, so try to acquire one */ 1466 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1467 ro6->ro_dst.sin6_family = AF_INET6; 1468 ro6->ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1469 ro6->ro_dst.sin6_addr = inc->inc6_faddr; 1470 rtalloc((struct route *)ro6); 1471 rt = ro6->ro_rt; 1472 } 1473 } 1474 return rt; 1475 } 1476 #endif /* INET6 */ 1477 1478 #ifdef IPSEC 1479 /* compute ESP/AH header size for TCP, including outer IP header. */ 1480 size_t 1481 ipsec_hdrsiz_tcp(tp) 1482 struct tcpcb *tp; 1483 { 1484 struct inpcb *inp; 1485 struct mbuf *m; 1486 size_t hdrsiz; 1487 struct ip *ip; 1488 #ifdef INET6 1489 struct ip6_hdr *ip6; 1490 #endif /* INET6 */ 1491 struct tcphdr *th; 1492 1493 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1494 return 0; 1495 MGETHDR(m, M_DONTWAIT, MT_DATA); 1496 if (!m) 1497 return 0; 1498 1499 #ifdef INET6 1500 if ((inp->inp_vflag & INP_IPV6) != 0) { 1501 ip6 = mtod(m, struct ip6_hdr *); 1502 th = (struct tcphdr *)(ip6 + 1); 1503 m->m_pkthdr.len = m->m_len = 1504 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1505 tcp_fillheaders(tp, ip6, th); 1506 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1507 } else 1508 #endif /* INET6 */ 1509 { 1510 ip = mtod(m, struct ip *); 1511 th = (struct tcphdr *)(ip + 1); 1512 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1513 tcp_fillheaders(tp, ip, th); 1514 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1515 } 1516 1517 m_free(m); 1518 return hdrsiz; 1519 } 1520 #endif /*IPSEC*/ 1521 1522 /* 1523 * Return a pointer to the cached information about the remote host. 1524 * The cached information is stored in the protocol specific part of 1525 * the route metrics. 1526 */ 1527 struct rmxp_tao * 1528 tcp_gettaocache(inc) 1529 struct in_conninfo *inc; 1530 { 1531 struct rtentry *rt; 1532 1533 #ifdef INET6 1534 if (inc->inc_isipv6) 1535 rt = tcp_rtlookup6(inc); 1536 else 1537 #endif /* INET6 */ 1538 rt = tcp_rtlookup(inc); 1539 1540 /* Make sure this is a host route and is up. */ 1541 if (rt == NULL || 1542 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) 1543 return NULL; 1544 1545 return rmx_taop(rt->rt_rmx); 1546 } 1547 1548 /* 1549 * Clear all the TAO cache entries, called from tcp_init. 1550 * 1551 * XXX 1552 * This routine is just an empty one, because we assume that the routing 1553 * routing tables are initialized at the same time when TCP, so there is 1554 * nothing in the cache left over. 1555 */ 1556 static void 1557 tcp_cleartaocache() 1558 { 1559 } 1560 1561 /* 1562 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1563 * 1564 * This code attempts to calculate the bandwidth-delay product as a 1565 * means of determining the optimal window size to maximize bandwidth, 1566 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1567 * routers. This code also does a fairly good job keeping RTTs in check 1568 * across slow links like modems. We implement an algorithm which is very 1569 * similar (but not meant to be) TCP/Vegas. The code operates on the 1570 * transmitter side of a TCP connection and so only effects the transmit 1571 * side of the connection. 1572 * 1573 * BACKGROUND: TCP makes no provision for the management of buffer space 1574 * at the end points or at the intermediate routers and switches. A TCP 1575 * stream, whether using NewReno or not, will eventually buffer as 1576 * many packets as it is able and the only reason this typically works is 1577 * due to the fairly small default buffers made available for a connection 1578 * (typicaly 16K or 32K). As machines use larger windows and/or window 1579 * scaling it is now fairly easy for even a single TCP connection to blow-out 1580 * all available buffer space not only on the local interface, but on 1581 * intermediate routers and switches as well. NewReno makes a misguided 1582 * attempt to 'solve' this problem by waiting for an actual failure to occur, 1583 * then backing off, then steadily increasing the window again until another 1584 * failure occurs, ad-infinitum. This results in terrible oscillation that 1585 * is only made worse as network loads increase and the idea of intentionally 1586 * blowing out network buffers is, frankly, a terrible way to manage network 1587 * resources. 1588 * 1589 * It is far better to limit the transmit window prior to the failure 1590 * condition being achieved. There are two general ways to do this: First 1591 * you can 'scan' through different transmit window sizes and locate the 1592 * point where the RTT stops increasing, indicating that you have filled the 1593 * pipe, then scan backwards until you note that RTT stops decreasing, then 1594 * repeat ad-infinitum. This method works in principle but has severe 1595 * implementation issues due to RTT variances, timer granularity, and 1596 * instability in the algorithm which can lead to many false positives and 1597 * create oscillations as well as interact badly with other TCP streams 1598 * implementing the same algorithm. 1599 * 1600 * The second method is to limit the window to the bandwidth delay product 1601 * of the link. This is the method we implement. RTT variances and our 1602 * own manipulation of the congestion window, bwnd, can potentially 1603 * destabilize the algorithm. For this reason we have to stabilize the 1604 * elements used to calculate the window. We do this by using the minimum 1605 * observed RTT, the long term average of the observed bandwidth, and 1606 * by adding two segments worth of slop. It isn't perfect but it is able 1607 * to react to changing conditions and gives us a very stable basis on 1608 * which to extend the algorithm. 1609 */ 1610 void 1611 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 1612 { 1613 u_long bw; 1614 u_long bwnd; 1615 int save_ticks; 1616 1617 /* 1618 * If inflight_enable is disabled in the middle of a tcp connection, 1619 * make sure snd_bwnd is effectively disabled. 1620 */ 1621 if (tcp_inflight_enable == 0) { 1622 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1623 tp->snd_bandwidth = 0; 1624 return; 1625 } 1626 1627 /* 1628 * Figure out the bandwidth. Due to the tick granularity this 1629 * is a very rough number and it MUST be averaged over a fairly 1630 * long period of time. XXX we need to take into account a link 1631 * that is not using all available bandwidth, but for now our 1632 * slop will ramp us up if this case occurs and the bandwidth later 1633 * increases. 1634 * 1635 * Note: if ticks rollover 'bw' may wind up negative. We must 1636 * effectively reset t_bw_rtttime for this case. 1637 */ 1638 save_ticks = ticks; 1639 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 1640 return; 1641 1642 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 1643 (save_ticks - tp->t_bw_rtttime); 1644 tp->t_bw_rtttime = save_ticks; 1645 tp->t_bw_rtseq = ack_seq; 1646 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 1647 return; 1648 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 1649 1650 tp->snd_bandwidth = bw; 1651 1652 /* 1653 * Calculate the semi-static bandwidth delay product, plus two maximal 1654 * segments. The additional slop puts us squarely in the sweet 1655 * spot and also handles the bandwidth run-up case. Without the 1656 * slop we could be locking ourselves into a lower bandwidth. 1657 * 1658 * Situations Handled: 1659 * (1) Prevents over-queueing of packets on LANs, especially on 1660 * high speed LANs, allowing larger TCP buffers to be 1661 * specified, and also does a good job preventing 1662 * over-queueing of packets over choke points like modems 1663 * (at least for the transmit side). 1664 * 1665 * (2) Is able to handle changing network loads (bandwidth 1666 * drops so bwnd drops, bandwidth increases so bwnd 1667 * increases). 1668 * 1669 * (3) Theoretically should stabilize in the face of multiple 1670 * connections implementing the same algorithm (this may need 1671 * a little work). 1672 */ 1673 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 1674 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + 2 * tp->t_maxseg; 1675 #undef USERTT 1676 1677 if (tcp_inflight_debug > 0) { 1678 static int ltime; 1679 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 1680 ltime = ticks; 1681 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 1682 tp, 1683 bw, 1684 tp->t_rttbest, 1685 tp->t_srtt, 1686 bwnd 1687 ); 1688 } 1689 } 1690 if ((long)bwnd < tcp_inflight_min) 1691 bwnd = tcp_inflight_min; 1692 if (bwnd > tcp_inflight_max) 1693 bwnd = tcp_inflight_max; 1694 if ((long)bwnd < tp->t_maxseg * 2) 1695 bwnd = tp->t_maxseg * 2; 1696 tp->snd_bwnd = bwnd; 1697 } 1698 1699