1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 30 * $FreeBSD$ 31 */ 32 33 #include "opt_compat.h" 34 #include "opt_inet.h" 35 #include "opt_inet6.h" 36 #include "opt_ipsec.h" 37 #include "opt_mac.h" 38 #include "opt_tcpdebug.h" 39 #include "opt_tcp_sack.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/callout.h> 44 #include <sys/kernel.h> 45 #include <sys/sysctl.h> 46 #include <sys/mac.h> 47 #include <sys/malloc.h> 48 #include <sys/mbuf.h> 49 #ifdef INET6 50 #include <sys/domain.h> 51 #endif 52 #include <sys/proc.h> 53 #include <sys/socket.h> 54 #include <sys/socketvar.h> 55 #include <sys/protosw.h> 56 #include <sys/random.h> 57 58 #include <vm/uma.h> 59 60 #include <net/route.h> 61 #include <net/if.h> 62 63 #include <netinet/in.h> 64 #include <netinet/in_systm.h> 65 #include <netinet/ip.h> 66 #ifdef INET6 67 #include <netinet/ip6.h> 68 #endif 69 #include <netinet/in_pcb.h> 70 #ifdef INET6 71 #include <netinet6/in6_pcb.h> 72 #endif 73 #include <netinet/in_var.h> 74 #include <netinet/ip_var.h> 75 #ifdef INET6 76 #include <netinet6/ip6_var.h> 77 #include <netinet6/nd6.h> 78 #endif 79 #include <netinet/tcp.h> 80 #include <netinet/tcp_fsm.h> 81 #include <netinet/tcp_seq.h> 82 #include <netinet/tcp_timer.h> 83 #include <netinet/tcp_var.h> 84 #ifdef INET6 85 #include <netinet6/tcp6_var.h> 86 #endif 87 #include <netinet/tcpip.h> 88 #ifdef TCPDEBUG 89 #include <netinet/tcp_debug.h> 90 #endif 91 #include <netinet6/ip6protosw.h> 92 93 #ifdef IPSEC 94 #include <netinet6/ipsec.h> 95 #ifdef INET6 96 #include <netinet6/ipsec6.h> 97 #endif 98 #endif /*IPSEC*/ 99 100 #ifdef FAST_IPSEC 101 #include <netipsec/ipsec.h> 102 #include <netipsec/xform.h> 103 #ifdef INET6 104 #include <netipsec/ipsec6.h> 105 #endif 106 #include <netipsec/key.h> 107 #define IPSEC 108 #endif /*FAST_IPSEC*/ 109 110 #include <machine/in_cksum.h> 111 #include <sys/md5.h> 112 113 int tcp_mssdflt = TCP_MSS; 114 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 115 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); 116 117 #ifdef INET6 118 int tcp_v6mssdflt = TCP6_MSS; 119 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 120 CTLFLAG_RW, &tcp_v6mssdflt , 0, 121 "Default TCP Maximum Segment Size for IPv6"); 122 #endif 123 124 /* 125 * Minimum MSS we accept and use. This prevents DoS attacks where 126 * we are forced to a ridiculous low MSS like 20 and send hundreds 127 * of packets instead of one. The effect scales with the available 128 * bandwidth and quickly saturates the CPU and network interface 129 * with packet generation and sending. Set to zero to disable MINMSS 130 * checking. This setting prevents us from sending too small packets. 131 */ 132 int tcp_minmss = TCP_MINMSS; 133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 134 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 135 /* 136 * Number of TCP segments per second we accept from remote host 137 * before we start to calculate average segment size. If average 138 * segment size drops below the minimum TCP MSS we assume a DoS 139 * attack and reset+drop the connection. Care has to be taken not to 140 * set this value too small to not kill interactive type connections 141 * (telnet, SSH) which send many small packets. 142 */ 143 int tcp_minmssoverload = TCP_MINMSSOVERLOAD; 144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW, 145 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to" 146 "be under the MINMSS Size"); 147 148 #if 0 149 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 150 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 151 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 152 #endif 153 154 int tcp_do_rfc1323 = 1; 155 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 156 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 157 158 int tcp_do_rfc1644 = 0; 159 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW, 160 &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions"); 161 162 static int tcp_tcbhashsize = 0; 163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 164 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 165 166 static int do_tcpdrain = 1; 167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 168 "Enable tcp_drain routine for extra help when low on mbufs"); 169 170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 171 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 172 173 static int icmp_may_rst = 1; 174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 175 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 176 177 static int tcp_isn_reseed_interval = 0; 178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 179 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 180 181 /* 182 * TCP bandwidth limiting sysctls. Note that the default lower bound of 183 * 1024 exists only for debugging. A good production default would be 184 * something like 6100. 185 */ 186 static int tcp_inflight_enable = 1; 187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_enable, CTLFLAG_RW, 188 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 189 190 static int tcp_inflight_debug = 0; 191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_debug, CTLFLAG_RW, 192 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 193 194 static int tcp_inflight_min = 6144; 195 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_min, CTLFLAG_RW, 196 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 197 198 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_max, CTLFLAG_RW, 200 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 201 static int tcp_inflight_stab = 20; 202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_stab, CTLFLAG_RW, 203 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 204 205 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW, 0, "TCP SACK"); 206 int tcp_do_sack = 1; 207 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, enable, CTLFLAG_RW, 208 &tcp_do_sack, 0, "Enable/Disable TCP SACK support"); 209 210 uma_zone_t sack_hole_zone; 211 212 static struct inpcb *tcp_notify(struct inpcb *, int); 213 static void tcp_discardcb(struct tcpcb *); 214 static void tcp_isn_tick(void *); 215 216 /* 217 * Target size of TCP PCB hash tables. Must be a power of two. 218 * 219 * Note that this can be overridden by the kernel environment 220 * variable net.inet.tcp.tcbhashsize 221 */ 222 #ifndef TCBHASHSIZE 223 #define TCBHASHSIZE 512 224 #endif 225 226 /* 227 * XXX 228 * Callouts should be moved into struct tcp directly. They are currently 229 * separate because the tcpcb structure is exported to userland for sysctl 230 * parsing purposes, which do not know about callouts. 231 */ 232 struct tcpcb_mem { 233 struct tcpcb tcb; 234 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 235 struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 236 }; 237 238 static uma_zone_t tcpcb_zone; 239 static uma_zone_t tcptw_zone; 240 struct callout isn_callout; 241 242 /* 243 * Tcp initialization 244 */ 245 void 246 tcp_init() 247 { 248 int hashsize = TCBHASHSIZE; 249 250 tcp_ccgen = 1; 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; 260 261 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 262 LIST_INIT(&tcb); 263 tcbinfo.listhead = &tcb; 264 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 265 if (!powerof2(hashsize)) { 266 printf("WARNING: TCB hash size not a power of 2\n"); 267 hashsize = 512; /* safe default */ 268 } 269 tcp_tcbhashsize = hashsize; 270 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 271 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 272 &tcbinfo.porthashmask); 273 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 274 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 275 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 276 #ifdef INET6 277 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 278 #else /* INET6 */ 279 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 280 #endif /* INET6 */ 281 if (max_protohdr < TCP_MINPROTOHDR) 282 max_protohdr = TCP_MINPROTOHDR; 283 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 284 panic("tcp_init"); 285 #undef TCP_MINPROTOHDR 286 /* 287 * These have to be type stable for the benefit of the timers. 288 */ 289 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 290 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 291 uma_zone_set_max(tcpcb_zone, maxsockets); 292 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 293 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 294 uma_zone_set_max(tcptw_zone, maxsockets / 5); 295 tcp_timer_init(); 296 syncache_init(); 297 tcp_hc_init(); 298 tcp_reass_init(); 299 callout_init(&isn_callout, CALLOUT_MPSAFE); 300 tcp_isn_tick(NULL); 301 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 302 SHUTDOWN_PRI_DEFAULT); 303 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 304 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 305 } 306 307 void 308 tcp_fini(xtp) 309 void *xtp; 310 { 311 callout_stop(&isn_callout); 312 313 } 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 */ 320 void 321 tcpip_fillheaders(inp, ip_ptr, tcp_ptr) 322 struct inpcb *inp; 323 void *ip_ptr; 324 void *tcp_ptr; 325 { 326 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 327 328 #ifdef INET6 329 if ((inp->inp_vflag & INP_IPV6) != 0) { 330 struct ip6_hdr *ip6; 331 332 ip6 = (struct ip6_hdr *)ip_ptr; 333 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 334 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 335 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 336 (IPV6_VERSION & IPV6_VERSION_MASK); 337 ip6->ip6_nxt = IPPROTO_TCP; 338 ip6->ip6_plen = sizeof(struct tcphdr); 339 ip6->ip6_src = inp->in6p_laddr; 340 ip6->ip6_dst = inp->in6p_faddr; 341 } else 342 #endif 343 { 344 struct ip *ip; 345 346 ip = (struct ip *)ip_ptr; 347 ip->ip_v = IPVERSION; 348 ip->ip_hl = 5; 349 ip->ip_tos = inp->inp_ip_tos; 350 ip->ip_len = 0; 351 ip->ip_id = 0; 352 ip->ip_off = 0; 353 ip->ip_ttl = inp->inp_ip_ttl; 354 ip->ip_sum = 0; 355 ip->ip_p = IPPROTO_TCP; 356 ip->ip_src = inp->inp_laddr; 357 ip->ip_dst = inp->inp_faddr; 358 } 359 th->th_sport = inp->inp_lport; 360 th->th_dport = inp->inp_fport; 361 th->th_seq = 0; 362 th->th_ack = 0; 363 th->th_x2 = 0; 364 th->th_off = 5; 365 th->th_flags = 0; 366 th->th_win = 0; 367 th->th_urp = 0; 368 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 369 } 370 371 /* 372 * Create template to be used to send tcp packets on a connection. 373 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 374 * use for this function is in keepalives, which use tcp_respond. 375 */ 376 struct tcptemp * 377 tcpip_maketemplate(inp) 378 struct inpcb *inp; 379 { 380 struct mbuf *m; 381 struct tcptemp *n; 382 383 m = m_get(M_DONTWAIT, MT_HEADER); 384 if (m == NULL) 385 return (0); 386 m->m_len = sizeof(struct tcptemp); 387 n = mtod(m, struct tcptemp *); 388 389 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 390 return (n); 391 } 392 393 /* 394 * Send a single message to the TCP at address specified by 395 * the given TCP/IP header. If m == NULL, then we make a copy 396 * of the tcpiphdr at ti and send directly to the addressed host. 397 * This is used to force keep alive messages out using the TCP 398 * template for a connection. If flags are given then we send 399 * a message back to the TCP which originated the * segment ti, 400 * and discard the mbuf containing it and any other attached mbufs. 401 * 402 * In any case the ack and sequence number of the transmitted 403 * segment are as specified by the parameters. 404 * 405 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 406 */ 407 void 408 tcp_respond(tp, ipgen, th, m, ack, seq, flags) 409 struct tcpcb *tp; 410 void *ipgen; 411 register struct tcphdr *th; 412 register struct mbuf *m; 413 tcp_seq ack, seq; 414 int flags; 415 { 416 register int tlen; 417 int win = 0; 418 struct ip *ip; 419 struct tcphdr *nth; 420 #ifdef INET6 421 struct ip6_hdr *ip6; 422 int isipv6; 423 #endif /* INET6 */ 424 int ipflags = 0; 425 struct inpcb *inp; 426 427 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 428 429 #ifdef INET6 430 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 431 ip6 = ipgen; 432 #endif /* INET6 */ 433 ip = ipgen; 434 435 if (tp != NULL) { 436 inp = tp->t_inpcb; 437 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 438 INP_INFO_WLOCK_ASSERT(&tcbinfo); 439 INP_LOCK_ASSERT(inp); 440 } else 441 inp = NULL; 442 443 if (tp != NULL) { 444 if (!(flags & TH_RST)) { 445 win = sbspace(&inp->inp_socket->so_rcv); 446 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 447 win = (long)TCP_MAXWIN << tp->rcv_scale; 448 } 449 } 450 if (m == NULL) { 451 m = m_gethdr(M_DONTWAIT, MT_HEADER); 452 if (m == NULL) 453 return; 454 tlen = 0; 455 m->m_data += max_linkhdr; 456 #ifdef INET6 457 if (isipv6) { 458 bcopy((caddr_t)ip6, mtod(m, caddr_t), 459 sizeof(struct ip6_hdr)); 460 ip6 = mtod(m, struct ip6_hdr *); 461 nth = (struct tcphdr *)(ip6 + 1); 462 } else 463 #endif /* INET6 */ 464 { 465 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 466 ip = mtod(m, struct ip *); 467 nth = (struct tcphdr *)(ip + 1); 468 } 469 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 470 flags = TH_ACK; 471 } else { 472 m_freem(m->m_next); 473 m->m_next = NULL; 474 m->m_data = (caddr_t)ipgen; 475 /* m_len is set later */ 476 tlen = 0; 477 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 478 #ifdef INET6 479 if (isipv6) { 480 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 481 nth = (struct tcphdr *)(ip6 + 1); 482 } else 483 #endif /* INET6 */ 484 { 485 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 486 nth = (struct tcphdr *)(ip + 1); 487 } 488 if (th != nth) { 489 /* 490 * this is usually a case when an extension header 491 * exists between the IPv6 header and the 492 * TCP header. 493 */ 494 nth->th_sport = th->th_sport; 495 nth->th_dport = th->th_dport; 496 } 497 xchg(nth->th_dport, nth->th_sport, n_short); 498 #undef xchg 499 } 500 #ifdef INET6 501 if (isipv6) { 502 ip6->ip6_flow = 0; 503 ip6->ip6_vfc = IPV6_VERSION; 504 ip6->ip6_nxt = IPPROTO_TCP; 505 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 506 tlen)); 507 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 508 } else 509 #endif 510 { 511 tlen += sizeof (struct tcpiphdr); 512 ip->ip_len = tlen; 513 ip->ip_ttl = ip_defttl; 514 if (path_mtu_discovery) 515 ip->ip_off |= IP_DF; 516 } 517 m->m_len = tlen; 518 m->m_pkthdr.len = tlen; 519 m->m_pkthdr.rcvif = NULL; 520 #ifdef MAC 521 if (inp != NULL) { 522 /* 523 * Packet is associated with a socket, so allow the 524 * label of the response to reflect the socket label. 525 */ 526 INP_LOCK_ASSERT(inp); 527 mac_create_mbuf_from_inpcb(inp, m); 528 } else { 529 /* 530 * Packet is not associated with a socket, so possibly 531 * update the label in place. 532 */ 533 mac_reflect_mbuf_tcp(m); 534 } 535 #endif 536 nth->th_seq = htonl(seq); 537 nth->th_ack = htonl(ack); 538 nth->th_x2 = 0; 539 nth->th_off = sizeof (struct tcphdr) >> 2; 540 nth->th_flags = flags; 541 if (tp != NULL) 542 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 543 else 544 nth->th_win = htons((u_short)win); 545 nth->th_urp = 0; 546 #ifdef INET6 547 if (isipv6) { 548 nth->th_sum = 0; 549 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 550 sizeof(struct ip6_hdr), 551 tlen - sizeof(struct ip6_hdr)); 552 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 553 NULL, NULL); 554 } else 555 #endif /* INET6 */ 556 { 557 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 558 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 559 m->m_pkthdr.csum_flags = CSUM_TCP; 560 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 561 } 562 #ifdef TCPDEBUG 563 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 564 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 565 #endif 566 #ifdef INET6 567 if (isipv6) 568 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 569 else 570 #endif /* INET6 */ 571 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 572 } 573 574 /* 575 * Create a new TCP control block, making an 576 * empty reassembly queue and hooking it to the argument 577 * protocol control block. The `inp' parameter must have 578 * come from the zone allocator set up in tcp_init(). 579 */ 580 struct tcpcb * 581 tcp_newtcpcb(inp) 582 struct inpcb *inp; 583 { 584 struct tcpcb_mem *tm; 585 struct tcpcb *tp; 586 #ifdef INET6 587 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 588 #endif /* INET6 */ 589 int callout_flag; 590 591 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 592 if (tm == NULL) 593 return (NULL); 594 tp = &tm->tcb; 595 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 596 tp->t_maxseg = tp->t_maxopd = 597 #ifdef INET6 598 isipv6 ? tcp_v6mssdflt : 599 #endif /* INET6 */ 600 tcp_mssdflt; 601 602 /* Set up our timeouts. */ 603 /* 604 * XXXRW: Are these actually MPSAFE? I think so, but need to 605 * review the timed wait code, as it has some list variables, 606 * etc, that are global. 607 */ 608 callout_flag = debug_mpsafenet ? CALLOUT_MPSAFE : 0; 609 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, callout_flag); 610 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, callout_flag); 611 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, callout_flag); 612 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, callout_flag); 613 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, callout_flag); 614 615 if (tcp_do_rfc1323) 616 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 617 if (tcp_do_rfc1644) 618 tp->t_flags |= TF_REQ_CC; 619 tp->sack_enable = tcp_do_sack; 620 tp->t_inpcb = inp; /* XXX */ 621 /* 622 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 623 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 624 * reasonable initial retransmit time. 625 */ 626 tp->t_srtt = TCPTV_SRTTBASE; 627 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 628 tp->t_rttmin = tcp_rexmit_min; 629 tp->t_rxtcur = TCPTV_RTOBASE; 630 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 631 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 632 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 633 tp->t_rcvtime = ticks; 634 tp->t_bw_rtttime = ticks; 635 /* 636 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 637 * because the socket may be bound to an IPv6 wildcard address, 638 * which may match an IPv4-mapped IPv6 address. 639 */ 640 inp->inp_ip_ttl = ip_defttl; 641 inp->inp_ppcb = (caddr_t)tp; 642 return (tp); /* XXX */ 643 } 644 645 /* 646 * Drop a TCP connection, reporting 647 * the specified error. If connection is synchronized, 648 * then send a RST to peer. 649 */ 650 struct tcpcb * 651 tcp_drop(tp, errno) 652 register struct tcpcb *tp; 653 int errno; 654 { 655 struct socket *so = tp->t_inpcb->inp_socket; 656 657 if (TCPS_HAVERCVDSYN(tp->t_state)) { 658 tp->t_state = TCPS_CLOSED; 659 (void) tcp_output(tp); 660 tcpstat.tcps_drops++; 661 } else 662 tcpstat.tcps_conndrops++; 663 if (errno == ETIMEDOUT && tp->t_softerror) 664 errno = tp->t_softerror; 665 so->so_error = errno; 666 return (tcp_close(tp)); 667 } 668 669 static void 670 tcp_discardcb(tp) 671 struct tcpcb *tp; 672 { 673 struct tseg_qent *q; 674 struct inpcb *inp = tp->t_inpcb; 675 struct socket *so = inp->inp_socket; 676 #ifdef INET6 677 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 678 #endif /* INET6 */ 679 680 /* 681 * Make sure that all of our timers are stopped before we 682 * delete the PCB. 683 */ 684 callout_stop(tp->tt_rexmt); 685 callout_stop(tp->tt_persist); 686 callout_stop(tp->tt_keep); 687 callout_stop(tp->tt_2msl); 688 callout_stop(tp->tt_delack); 689 690 /* 691 * If we got enough samples through the srtt filter, 692 * save the rtt and rttvar in the routing entry. 693 * 'Enough' is arbitrarily defined as 4 rtt samples. 694 * 4 samples is enough for the srtt filter to converge 695 * to within enough % of the correct value; fewer samples 696 * and we could save a bogus rtt. The danger is not high 697 * as tcp quickly recovers from everything. 698 * XXX: Works very well but needs some more statistics! 699 */ 700 if (tp->t_rttupdated >= 4) { 701 struct hc_metrics_lite metrics; 702 u_long ssthresh; 703 704 bzero(&metrics, sizeof(metrics)); 705 /* 706 * Update the ssthresh always when the conditions below 707 * are satisfied. This gives us better new start value 708 * for the congestion avoidance for new connections. 709 * ssthresh is only set if packet loss occured on a session. 710 */ 711 ssthresh = tp->snd_ssthresh; 712 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 713 /* 714 * convert the limit from user data bytes to 715 * packets then to packet data bytes. 716 */ 717 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 718 if (ssthresh < 2) 719 ssthresh = 2; 720 ssthresh *= (u_long)(tp->t_maxseg + 721 #ifdef INET6 722 (isipv6 ? sizeof (struct ip6_hdr) + 723 sizeof (struct tcphdr) : 724 #endif 725 sizeof (struct tcpiphdr) 726 #ifdef INET6 727 ) 728 #endif 729 ); 730 } else 731 ssthresh = 0; 732 metrics.rmx_ssthresh = ssthresh; 733 734 metrics.rmx_rtt = tp->t_srtt; 735 metrics.rmx_rttvar = tp->t_rttvar; 736 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 737 metrics.rmx_bandwidth = tp->snd_bandwidth; 738 metrics.rmx_cwnd = tp->snd_cwnd; 739 metrics.rmx_sendpipe = 0; 740 metrics.rmx_recvpipe = 0; 741 742 tcp_hc_update(&inp->inp_inc, &metrics); 743 } 744 745 /* free the reassembly queue, if any */ 746 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 747 LIST_REMOVE(q, tqe_q); 748 m_freem(q->tqe_m); 749 uma_zfree(tcp_reass_zone, q); 750 tp->t_segqlen--; 751 tcp_reass_qsize--; 752 } 753 tcp_free_sackholes(tp); 754 inp->inp_ppcb = NULL; 755 tp->t_inpcb = NULL; 756 uma_zfree(tcpcb_zone, tp); 757 soisdisconnected(so); 758 } 759 760 /* 761 * Close a TCP control block: 762 * discard all space held by the tcp 763 * discard internet protocol block 764 * wake up any sleepers 765 */ 766 struct tcpcb * 767 tcp_close(tp) 768 struct tcpcb *tp; 769 { 770 struct inpcb *inp = tp->t_inpcb; 771 #ifdef INET6 772 struct socket *so = inp->inp_socket; 773 #endif 774 775 tcp_discardcb(tp); 776 #ifdef INET6 777 if (INP_CHECK_SOCKAF(so, AF_INET6)) 778 in6_pcbdetach(inp); 779 else 780 #endif 781 in_pcbdetach(inp); 782 tcpstat.tcps_closed++; 783 return (NULL); 784 } 785 786 void 787 tcp_drain() 788 { 789 if (do_tcpdrain) 790 { 791 struct inpcb *inpb; 792 struct tcpcb *tcpb; 793 struct tseg_qent *te; 794 795 /* 796 * Walk the tcpbs, if existing, and flush the reassembly queue, 797 * if there is one... 798 * XXX: The "Net/3" implementation doesn't imply that the TCP 799 * reassembly queue should be flushed, but in a situation 800 * where we're really low on mbufs, this is potentially 801 * usefull. 802 */ 803 INP_INFO_RLOCK(&tcbinfo); 804 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 805 if (inpb->inp_vflag & INP_TIMEWAIT) 806 continue; 807 INP_LOCK(inpb); 808 if ((tcpb = intotcpcb(inpb)) != NULL) { 809 while ((te = LIST_FIRST(&tcpb->t_segq)) 810 != NULL) { 811 LIST_REMOVE(te, tqe_q); 812 m_freem(te->tqe_m); 813 uma_zfree(tcp_reass_zone, te); 814 tcpb->t_segqlen--; 815 tcp_reass_qsize--; 816 } 817 } 818 INP_UNLOCK(inpb); 819 } 820 INP_INFO_RUNLOCK(&tcbinfo); 821 } 822 } 823 824 /* 825 * Notify a tcp user of an asynchronous error; 826 * store error as soft error, but wake up user 827 * (for now, won't do anything until can select for soft error). 828 * 829 * Do not wake up user since there currently is no mechanism for 830 * reporting soft errors (yet - a kqueue filter may be added). 831 */ 832 static struct inpcb * 833 tcp_notify(inp, error) 834 struct inpcb *inp; 835 int error; 836 { 837 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 838 839 /* 840 * Ignore some errors if we are hooked up. 841 * If connection hasn't completed, has retransmitted several times, 842 * and receives a second error, give up now. This is better 843 * than waiting a long time to establish a connection that 844 * can never complete. 845 */ 846 if (tp->t_state == TCPS_ESTABLISHED && 847 (error == EHOSTUNREACH || error == ENETUNREACH || 848 error == EHOSTDOWN)) { 849 return inp; 850 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 851 tp->t_softerror) { 852 tcp_drop(tp, error); 853 return (struct inpcb *)0; 854 } else { 855 tp->t_softerror = error; 856 return inp; 857 } 858 #if 0 859 wakeup( &so->so_timeo); 860 sorwakeup(so); 861 sowwakeup(so); 862 #endif 863 } 864 865 static int 866 tcp_pcblist(SYSCTL_HANDLER_ARGS) 867 { 868 int error, i, n, s; 869 struct inpcb *inp, **inp_list; 870 inp_gen_t gencnt; 871 struct xinpgen xig; 872 873 /* 874 * The process of preparing the TCB list is too time-consuming and 875 * resource-intensive to repeat twice on every request. 876 */ 877 if (req->oldptr == NULL) { 878 n = tcbinfo.ipi_count; 879 req->oldidx = 2 * (sizeof xig) 880 + (n + n/8) * sizeof(struct xtcpcb); 881 return 0; 882 } 883 884 if (req->newptr != NULL) 885 return EPERM; 886 887 /* 888 * OK, now we're committed to doing something. 889 */ 890 s = splnet(); 891 INP_INFO_RLOCK(&tcbinfo); 892 gencnt = tcbinfo.ipi_gencnt; 893 n = tcbinfo.ipi_count; 894 INP_INFO_RUNLOCK(&tcbinfo); 895 splx(s); 896 897 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 898 + n * sizeof(struct xtcpcb)); 899 if (error != 0) 900 return (error); 901 902 xig.xig_len = sizeof xig; 903 xig.xig_count = n; 904 xig.xig_gen = gencnt; 905 xig.xig_sogen = so_gencnt; 906 error = SYSCTL_OUT(req, &xig, sizeof xig); 907 if (error) 908 return error; 909 910 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 911 if (inp_list == NULL) 912 return ENOMEM; 913 914 s = splnet(); 915 INP_INFO_RLOCK(&tcbinfo); 916 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 917 inp = LIST_NEXT(inp, inp_list)) { 918 INP_LOCK(inp); 919 if (inp->inp_gencnt <= gencnt) { 920 /* 921 * XXX: This use of cr_cansee(), introduced with 922 * TCP state changes, is not quite right, but for 923 * now, better than nothing. 924 */ 925 if (inp->inp_vflag & INP_TIMEWAIT) 926 error = cr_cansee(req->td->td_ucred, 927 intotw(inp)->tw_cred); 928 else 929 error = cr_canseesocket(req->td->td_ucred, 930 inp->inp_socket); 931 if (error == 0) 932 inp_list[i++] = inp; 933 } 934 INP_UNLOCK(inp); 935 } 936 INP_INFO_RUNLOCK(&tcbinfo); 937 splx(s); 938 n = i; 939 940 error = 0; 941 for (i = 0; i < n; i++) { 942 inp = inp_list[i]; 943 if (inp->inp_gencnt <= gencnt) { 944 struct xtcpcb xt; 945 caddr_t inp_ppcb; 946 xt.xt_len = sizeof xt; 947 /* XXX should avoid extra copy */ 948 bcopy(inp, &xt.xt_inp, sizeof *inp); 949 inp_ppcb = inp->inp_ppcb; 950 if (inp_ppcb == NULL) 951 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 952 else if (inp->inp_vflag & INP_TIMEWAIT) { 953 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 954 xt.xt_tp.t_state = TCPS_TIME_WAIT; 955 } else 956 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 957 if (inp->inp_socket != NULL) 958 sotoxsocket(inp->inp_socket, &xt.xt_socket); 959 else { 960 bzero(&xt.xt_socket, sizeof xt.xt_socket); 961 xt.xt_socket.xso_protocol = IPPROTO_TCP; 962 } 963 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 964 error = SYSCTL_OUT(req, &xt, sizeof xt); 965 } 966 } 967 if (!error) { 968 /* 969 * Give the user an updated idea of our state. 970 * If the generation differs from what we told 971 * her before, she knows that something happened 972 * while we were processing this request, and it 973 * might be necessary to retry. 974 */ 975 s = splnet(); 976 INP_INFO_RLOCK(&tcbinfo); 977 xig.xig_gen = tcbinfo.ipi_gencnt; 978 xig.xig_sogen = so_gencnt; 979 xig.xig_count = tcbinfo.ipi_count; 980 INP_INFO_RUNLOCK(&tcbinfo); 981 splx(s); 982 error = SYSCTL_OUT(req, &xig, sizeof xig); 983 } 984 free(inp_list, M_TEMP); 985 return error; 986 } 987 988 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 989 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 990 991 static int 992 tcp_getcred(SYSCTL_HANDLER_ARGS) 993 { 994 struct xucred xuc; 995 struct sockaddr_in addrs[2]; 996 struct inpcb *inp; 997 int error, s; 998 999 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1000 if (error) 1001 return (error); 1002 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1003 if (error) 1004 return (error); 1005 s = splnet(); 1006 INP_INFO_RLOCK(&tcbinfo); 1007 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1008 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1009 if (inp == NULL) { 1010 error = ENOENT; 1011 goto outunlocked; 1012 } 1013 INP_LOCK(inp); 1014 if (inp->inp_socket == NULL) { 1015 error = ENOENT; 1016 goto out; 1017 } 1018 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1019 if (error) 1020 goto out; 1021 cru2x(inp->inp_socket->so_cred, &xuc); 1022 out: 1023 INP_UNLOCK(inp); 1024 outunlocked: 1025 INP_INFO_RUNLOCK(&tcbinfo); 1026 splx(s); 1027 if (error == 0) 1028 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1029 return (error); 1030 } 1031 1032 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1033 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1034 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1035 1036 #ifdef INET6 1037 static int 1038 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1039 { 1040 struct xucred xuc; 1041 struct sockaddr_in6 addrs[2]; 1042 struct inpcb *inp; 1043 int error, s, mapped = 0; 1044 1045 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1046 if (error) 1047 return (error); 1048 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1049 if (error) 1050 return (error); 1051 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1052 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1053 mapped = 1; 1054 else 1055 return (EINVAL); 1056 } 1057 s = splnet(); 1058 INP_INFO_RLOCK(&tcbinfo); 1059 if (mapped == 1) 1060 inp = in_pcblookup_hash(&tcbinfo, 1061 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1062 addrs[1].sin6_port, 1063 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1064 addrs[0].sin6_port, 1065 0, NULL); 1066 else 1067 inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr, 1068 addrs[1].sin6_port, 1069 &addrs[0].sin6_addr, addrs[0].sin6_port, 1070 0, NULL); 1071 if (inp == NULL) { 1072 error = ENOENT; 1073 goto outunlocked; 1074 } 1075 INP_LOCK(inp); 1076 if (inp->inp_socket == NULL) { 1077 error = ENOENT; 1078 goto out; 1079 } 1080 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1081 if (error) 1082 goto out; 1083 cru2x(inp->inp_socket->so_cred, &xuc); 1084 out: 1085 INP_UNLOCK(inp); 1086 outunlocked: 1087 INP_INFO_RUNLOCK(&tcbinfo); 1088 splx(s); 1089 if (error == 0) 1090 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1091 return (error); 1092 } 1093 1094 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1095 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1096 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1097 #endif 1098 1099 1100 void 1101 tcp_ctlinput(cmd, sa, vip) 1102 int cmd; 1103 struct sockaddr *sa; 1104 void *vip; 1105 { 1106 struct ip *ip = vip; 1107 struct tcphdr *th; 1108 struct in_addr faddr; 1109 struct inpcb *inp; 1110 struct tcpcb *tp; 1111 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1112 tcp_seq icmp_seq; 1113 int s; 1114 1115 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1116 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1117 return; 1118 1119 if (cmd == PRC_QUENCH) 1120 notify = tcp_quench; 1121 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1122 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1123 notify = tcp_drop_syn_sent; 1124 else if (cmd == PRC_MSGSIZE) 1125 notify = tcp_mtudisc; 1126 /* 1127 * Redirects don't need to be handled up here. 1128 */ 1129 else if (PRC_IS_REDIRECT(cmd)) 1130 return; 1131 /* 1132 * Hostdead is ugly because it goes linearly through all PCBs. 1133 * XXX: We never get this from ICMP, otherwise it makes an 1134 * excellent DoS attack on machines with many connections. 1135 */ 1136 else if (cmd == PRC_HOSTDEAD) 1137 ip = NULL; 1138 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1139 return; 1140 if (ip != NULL) { 1141 s = splnet(); 1142 th = (struct tcphdr *)((caddr_t)ip 1143 + (ip->ip_hl << 2)); 1144 INP_INFO_WLOCK(&tcbinfo); 1145 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1146 ip->ip_src, th->th_sport, 0, NULL); 1147 if (inp != NULL) { 1148 INP_LOCK(inp); 1149 if (inp->inp_socket != NULL) { 1150 icmp_seq = htonl(th->th_seq); 1151 tp = intotcpcb(inp); 1152 if (SEQ_GEQ(icmp_seq, tp->snd_una) && 1153 SEQ_LT(icmp_seq, tp->snd_max)) 1154 inp = (*notify)(inp, inetctlerrmap[cmd]); 1155 } 1156 if (inp != NULL) 1157 INP_UNLOCK(inp); 1158 } else { 1159 struct in_conninfo inc; 1160 1161 inc.inc_fport = th->th_dport; 1162 inc.inc_lport = th->th_sport; 1163 inc.inc_faddr = faddr; 1164 inc.inc_laddr = ip->ip_src; 1165 #ifdef INET6 1166 inc.inc_isipv6 = 0; 1167 #endif 1168 syncache_unreach(&inc, th); 1169 } 1170 INP_INFO_WUNLOCK(&tcbinfo); 1171 splx(s); 1172 } else 1173 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1174 } 1175 1176 #ifdef INET6 1177 void 1178 tcp6_ctlinput(cmd, sa, d) 1179 int cmd; 1180 struct sockaddr *sa; 1181 void *d; 1182 { 1183 struct tcphdr th; 1184 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1185 struct ip6_hdr *ip6; 1186 struct mbuf *m; 1187 struct ip6ctlparam *ip6cp = NULL; 1188 const struct sockaddr_in6 *sa6_src = NULL; 1189 int off; 1190 struct tcp_portonly { 1191 u_int16_t th_sport; 1192 u_int16_t th_dport; 1193 } *thp; 1194 1195 if (sa->sa_family != AF_INET6 || 1196 sa->sa_len != sizeof(struct sockaddr_in6)) 1197 return; 1198 1199 if (cmd == PRC_QUENCH) 1200 notify = tcp_quench; 1201 else if (cmd == PRC_MSGSIZE) 1202 notify = tcp_mtudisc; 1203 else if (!PRC_IS_REDIRECT(cmd) && 1204 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1205 return; 1206 1207 /* if the parameter is from icmp6, decode it. */ 1208 if (d != NULL) { 1209 ip6cp = (struct ip6ctlparam *)d; 1210 m = ip6cp->ip6c_m; 1211 ip6 = ip6cp->ip6c_ip6; 1212 off = ip6cp->ip6c_off; 1213 sa6_src = ip6cp->ip6c_src; 1214 } else { 1215 m = NULL; 1216 ip6 = NULL; 1217 off = 0; /* fool gcc */ 1218 sa6_src = &sa6_any; 1219 } 1220 1221 if (ip6 != NULL) { 1222 struct in_conninfo inc; 1223 /* 1224 * XXX: We assume that when IPV6 is non NULL, 1225 * M and OFF are valid. 1226 */ 1227 1228 /* check if we can safely examine src and dst ports */ 1229 if (m->m_pkthdr.len < off + sizeof(*thp)) 1230 return; 1231 1232 bzero(&th, sizeof(th)); 1233 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1234 1235 in6_pcbnotify(&tcb, sa, th.th_dport, 1236 (struct sockaddr *)ip6cp->ip6c_src, 1237 th.th_sport, cmd, NULL, notify); 1238 1239 inc.inc_fport = th.th_dport; 1240 inc.inc_lport = th.th_sport; 1241 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1242 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1243 inc.inc_isipv6 = 1; 1244 syncache_unreach(&inc, &th); 1245 } else 1246 in6_pcbnotify(&tcb, sa, 0, (const struct sockaddr *)sa6_src, 1247 0, cmd, NULL, notify); 1248 } 1249 #endif /* INET6 */ 1250 1251 1252 /* 1253 * Following is where TCP initial sequence number generation occurs. 1254 * 1255 * There are two places where we must use initial sequence numbers: 1256 * 1. In SYN-ACK packets. 1257 * 2. In SYN packets. 1258 * 1259 * All ISNs for SYN-ACK packets are generated by the syncache. See 1260 * tcp_syncache.c for details. 1261 * 1262 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1263 * depends on this property. In addition, these ISNs should be 1264 * unguessable so as to prevent connection hijacking. To satisfy 1265 * the requirements of this situation, the algorithm outlined in 1266 * RFC 1948 is used, with only small modifications. 1267 * 1268 * Implementation details: 1269 * 1270 * Time is based off the system timer, and is corrected so that it 1271 * increases by one megabyte per second. This allows for proper 1272 * recycling on high speed LANs while still leaving over an hour 1273 * before rollover. 1274 * 1275 * As reading the *exact* system time is too expensive to be done 1276 * whenever setting up a TCP connection, we increment the time 1277 * offset in two ways. First, a small random positive increment 1278 * is added to isn_offset for each connection that is set up. 1279 * Second, the function tcp_isn_tick fires once per clock tick 1280 * and increments isn_offset as necessary so that sequence numbers 1281 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1282 * random positive increments serve only to ensure that the same 1283 * exact sequence number is never sent out twice (as could otherwise 1284 * happen when a port is recycled in less than the system tick 1285 * interval.) 1286 * 1287 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1288 * between seeding of isn_secret. This is normally set to zero, 1289 * as reseeding should not be necessary. 1290 * 1291 */ 1292 1293 #define ISN_BYTES_PER_SECOND 1048576 1294 #define ISN_STATIC_INCREMENT 4096 1295 #define ISN_RANDOM_INCREMENT (4096 - 1) 1296 1297 u_char isn_secret[32]; 1298 int isn_last_reseed; 1299 u_int32_t isn_offset, isn_offset_old; 1300 MD5_CTX isn_ctx; 1301 1302 tcp_seq 1303 tcp_new_isn(tp) 1304 struct tcpcb *tp; 1305 { 1306 u_int32_t md5_buffer[4]; 1307 tcp_seq new_isn; 1308 1309 /* Seed if this is the first use, reseed if requested. */ 1310 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1311 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1312 < (u_int)ticks))) { 1313 read_random(&isn_secret, sizeof(isn_secret)); 1314 isn_last_reseed = ticks; 1315 } 1316 1317 /* Compute the md5 hash and return the ISN. */ 1318 MD5Init(&isn_ctx); 1319 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1320 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1321 #ifdef INET6 1322 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1323 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1324 sizeof(struct in6_addr)); 1325 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1326 sizeof(struct in6_addr)); 1327 } else 1328 #endif 1329 { 1330 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1331 sizeof(struct in_addr)); 1332 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1333 sizeof(struct in_addr)); 1334 } 1335 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1336 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1337 new_isn = (tcp_seq) md5_buffer[0]; 1338 isn_offset += ISN_STATIC_INCREMENT + 1339 (arc4random() & ISN_RANDOM_INCREMENT); 1340 new_isn += isn_offset; 1341 return new_isn; 1342 } 1343 1344 /* 1345 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1346 * to keep time flowing at a relatively constant rate. If the random 1347 * increments have already pushed us past the projected offset, do nothing. 1348 */ 1349 static void 1350 tcp_isn_tick(xtp) 1351 void *xtp; 1352 { 1353 u_int32_t projected_offset; 1354 1355 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / hz; 1356 1357 if (projected_offset > isn_offset) 1358 isn_offset = projected_offset; 1359 1360 isn_offset_old = isn_offset; 1361 callout_reset(&isn_callout, 1, tcp_isn_tick, NULL); 1362 } 1363 1364 /* 1365 * When a source quench is received, close congestion window 1366 * to one segment. We will gradually open it again as we proceed. 1367 */ 1368 struct inpcb * 1369 tcp_quench(inp, errno) 1370 struct inpcb *inp; 1371 int errno; 1372 { 1373 struct tcpcb *tp = intotcpcb(inp); 1374 1375 if (tp != NULL) 1376 tp->snd_cwnd = tp->t_maxseg; 1377 return (inp); 1378 } 1379 1380 /* 1381 * When a specific ICMP unreachable message is received and the 1382 * connection state is SYN-SENT, drop the connection. This behavior 1383 * is controlled by the icmp_may_rst sysctl. 1384 */ 1385 struct inpcb * 1386 tcp_drop_syn_sent(inp, errno) 1387 struct inpcb *inp; 1388 int errno; 1389 { 1390 struct tcpcb *tp = intotcpcb(inp); 1391 1392 if (tp != NULL && tp->t_state == TCPS_SYN_SENT) { 1393 tcp_drop(tp, errno); 1394 return (struct inpcb *)0; 1395 } 1396 return inp; 1397 } 1398 1399 /* 1400 * When `need fragmentation' ICMP is received, update our idea of the MSS 1401 * based on the new value in the route. Also nudge TCP to send something, 1402 * since we know the packet we just sent was dropped. 1403 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1404 */ 1405 struct inpcb * 1406 tcp_mtudisc(inp, errno) 1407 struct inpcb *inp; 1408 int errno; 1409 { 1410 struct tcpcb *tp = intotcpcb(inp); 1411 struct rmxp_tao tao; 1412 struct socket *so = inp->inp_socket; 1413 u_int maxmtu; 1414 u_int romtu; 1415 int mss; 1416 #ifdef INET6 1417 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1418 #endif /* INET6 */ 1419 bzero(&tao, sizeof(tao)); 1420 1421 if (tp != NULL) { 1422 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1423 romtu = 1424 #ifdef INET6 1425 isipv6 ? tcp_maxmtu6(&inp->inp_inc) : 1426 #endif /* INET6 */ 1427 tcp_maxmtu(&inp->inp_inc); 1428 if (!maxmtu) 1429 maxmtu = romtu; 1430 else 1431 maxmtu = min(maxmtu, romtu); 1432 if (!maxmtu) { 1433 tp->t_maxopd = tp->t_maxseg = 1434 #ifdef INET6 1435 isipv6 ? tcp_v6mssdflt : 1436 #endif /* INET6 */ 1437 tcp_mssdflt; 1438 return inp; 1439 } 1440 mss = maxmtu - 1441 #ifdef INET6 1442 (isipv6 ? 1443 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1444 #endif /* INET6 */ 1445 sizeof(struct tcpiphdr) 1446 #ifdef INET6 1447 ) 1448 #endif /* INET6 */ 1449 ; 1450 1451 if (tcp_do_rfc1644) { 1452 tcp_hc_gettao(&inp->inp_inc, &tao); 1453 if (tao.tao_mssopt) 1454 mss = min(mss, tao.tao_mssopt); 1455 } 1456 /* 1457 * XXX - The above conditional probably violates the TCP 1458 * spec. The problem is that, since we don't know the 1459 * other end's MSS, we are supposed to use a conservative 1460 * default. But, if we do that, then MTU discovery will 1461 * never actually take place, because the conservative 1462 * default is much less than the MTUs typically seen 1463 * on the Internet today. For the moment, we'll sweep 1464 * this under the carpet. 1465 * 1466 * The conservative default might not actually be a problem 1467 * if the only case this occurs is when sending an initial 1468 * SYN with options and data to a host we've never talked 1469 * to before. Then, they will reply with an MSS value which 1470 * will get recorded and the new parameters should get 1471 * recomputed. For Further Study. 1472 */ 1473 if (tp->t_maxopd <= mss) 1474 return inp; 1475 tp->t_maxopd = mss; 1476 1477 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1478 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1479 mss -= TCPOLEN_TSTAMP_APPA; 1480 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 1481 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC) 1482 mss -= TCPOLEN_CC_APPA; 1483 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1484 if (mss > MCLBYTES) 1485 mss &= ~(MCLBYTES-1); 1486 #else 1487 if (mss > MCLBYTES) 1488 mss = mss / MCLBYTES * MCLBYTES; 1489 #endif 1490 if (so->so_snd.sb_hiwat < mss) 1491 mss = so->so_snd.sb_hiwat; 1492 1493 tp->t_maxseg = mss; 1494 1495 tcpstat.tcps_mturesent++; 1496 tp->t_rtttime = 0; 1497 tp->snd_nxt = tp->snd_una; 1498 tcp_output(tp); 1499 } 1500 return inp; 1501 } 1502 1503 /* 1504 * Look-up the routing entry to the peer of this inpcb. If no route 1505 * is found and it cannot be allocated, then return NULL. This routine 1506 * is called by TCP routines that access the rmx structure and by tcp_mss 1507 * to get the interface MTU. 1508 */ 1509 u_long 1510 tcp_maxmtu(inc) 1511 struct in_conninfo *inc; 1512 { 1513 struct route sro; 1514 struct sockaddr_in *dst; 1515 struct ifnet *ifp; 1516 u_long maxmtu = 0; 1517 1518 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1519 1520 bzero(&sro, sizeof(sro)); 1521 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1522 dst = (struct sockaddr_in *)&sro.ro_dst; 1523 dst->sin_family = AF_INET; 1524 dst->sin_len = sizeof(*dst); 1525 dst->sin_addr = inc->inc_faddr; 1526 rtalloc_ign(&sro, RTF_CLONING); 1527 } 1528 if (sro.ro_rt != NULL) { 1529 ifp = sro.ro_rt->rt_ifp; 1530 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1531 maxmtu = ifp->if_mtu; 1532 else 1533 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1534 RTFREE(sro.ro_rt); 1535 } 1536 return (maxmtu); 1537 } 1538 1539 #ifdef INET6 1540 u_long 1541 tcp_maxmtu6(inc) 1542 struct in_conninfo *inc; 1543 { 1544 struct route_in6 sro6; 1545 struct ifnet *ifp; 1546 u_long maxmtu = 0; 1547 1548 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1549 1550 bzero(&sro6, sizeof(sro6)); 1551 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1552 sro6.ro_dst.sin6_family = AF_INET6; 1553 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1554 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1555 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1556 } 1557 if (sro6.ro_rt != NULL) { 1558 ifp = sro6.ro_rt->rt_ifp; 1559 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1560 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1561 else 1562 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1563 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1564 RTFREE(sro6.ro_rt); 1565 } 1566 1567 return (maxmtu); 1568 } 1569 #endif /* INET6 */ 1570 1571 #ifdef IPSEC 1572 /* compute ESP/AH header size for TCP, including outer IP header. */ 1573 size_t 1574 ipsec_hdrsiz_tcp(tp) 1575 struct tcpcb *tp; 1576 { 1577 struct inpcb *inp; 1578 struct mbuf *m; 1579 size_t hdrsiz; 1580 struct ip *ip; 1581 #ifdef INET6 1582 struct ip6_hdr *ip6; 1583 #endif 1584 struct tcphdr *th; 1585 1586 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1587 return 0; 1588 MGETHDR(m, M_DONTWAIT, MT_DATA); 1589 if (!m) 1590 return 0; 1591 1592 #ifdef INET6 1593 if ((inp->inp_vflag & INP_IPV6) != 0) { 1594 ip6 = mtod(m, struct ip6_hdr *); 1595 th = (struct tcphdr *)(ip6 + 1); 1596 m->m_pkthdr.len = m->m_len = 1597 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1598 tcpip_fillheaders(inp, ip6, th); 1599 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1600 } else 1601 #endif /* INET6 */ 1602 { 1603 ip = mtod(m, struct ip *); 1604 th = (struct tcphdr *)(ip + 1); 1605 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1606 tcpip_fillheaders(inp, ip, th); 1607 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1608 } 1609 1610 m_free(m); 1611 return hdrsiz; 1612 } 1613 #endif /*IPSEC*/ 1614 1615 /* 1616 * Move a TCP connection into TIME_WAIT state. 1617 * tcbinfo is unlocked. 1618 * inp is locked, and is unlocked before returning. 1619 */ 1620 void 1621 tcp_twstart(tp) 1622 struct tcpcb *tp; 1623 { 1624 struct tcptw *tw; 1625 struct inpcb *inp; 1626 int tw_time, acknow; 1627 struct socket *so; 1628 1629 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1630 if (tw == NULL) { 1631 tw = tcp_timer_2msl_tw(1); 1632 if (tw == NULL) { 1633 tcp_close(tp); 1634 return; 1635 } 1636 } 1637 inp = tp->t_inpcb; 1638 tw->tw_inpcb = inp; 1639 1640 /* 1641 * Recover last window size sent. 1642 */ 1643 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1644 1645 /* 1646 * Set t_recent if timestamps are used on the connection. 1647 */ 1648 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1649 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1650 tw->t_recent = tp->ts_recent; 1651 else 1652 tw->t_recent = 0; 1653 1654 tw->snd_nxt = tp->snd_nxt; 1655 tw->rcv_nxt = tp->rcv_nxt; 1656 tw->iss = tp->iss; 1657 tw->irs = tp->irs; 1658 tw->cc_recv = tp->cc_recv; 1659 tw->cc_send = tp->cc_send; 1660 tw->t_starttime = tp->t_starttime; 1661 tw->tw_time = 0; 1662 1663 /* XXX 1664 * If this code will 1665 * be used for fin-wait-2 state also, then we may need 1666 * a ts_recent from the last segment. 1667 */ 1668 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1669 if (tp->cc_recv != 0 && (ticks - tp->t_starttime) < tcp_msl) { 1670 tw_time = tp->t_rxtcur * TCPTV_TWTRUNC; 1671 /* For T/TCP client, force ACK now. */ 1672 acknow = 1; 1673 } else { 1674 tw_time = 2 * tcp_msl; 1675 acknow = tp->t_flags & TF_ACKNOW; 1676 } 1677 tcp_discardcb(tp); 1678 so = inp->inp_socket; 1679 SOCK_LOCK(so); 1680 so->so_pcb = NULL; 1681 tw->tw_cred = crhold(so->so_cred); 1682 tw->tw_so_options = so->so_options; 1683 sotryfree(so); 1684 inp->inp_socket = NULL; 1685 if (acknow) 1686 tcp_twrespond(tw, TH_ACK); 1687 inp->inp_ppcb = (caddr_t)tw; 1688 inp->inp_vflag |= INP_TIMEWAIT; 1689 tcp_timer_2msl_reset(tw, tw_time); 1690 INP_UNLOCK(inp); 1691 } 1692 1693 /* 1694 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1695 * the actual rate is slightly higher due to the addition of 1696 * random positive increments. 1697 * 1698 * Most other new OSes use semi-randomized ISN values, so we 1699 * do not need to worry about them. 1700 */ 1701 #define MS_ISN_BYTES_PER_SECOND 250000 1702 1703 /* 1704 * Determine if the ISN we will generate has advanced beyond the last 1705 * sequence number used by the previous connection. If so, indicate 1706 * that it is safe to recycle this tw socket by returning 1. 1707 */ 1708 int 1709 tcp_twrecycleable(struct tcptw *tw) 1710 { 1711 tcp_seq new_iss = tw->iss; 1712 tcp_seq new_irs = tw->irs; 1713 1714 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1715 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1716 1717 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1718 return 1; 1719 else 1720 return 0; 1721 } 1722 1723 struct tcptw * 1724 tcp_twclose(struct tcptw *tw, int reuse) 1725 { 1726 struct inpcb *inp; 1727 1728 inp = tw->tw_inpcb; 1729 tw->tw_inpcb = NULL; 1730 tcp_timer_2msl_stop(tw); 1731 inp->inp_ppcb = NULL; 1732 #ifdef INET6 1733 if (inp->inp_vflag & INP_IPV6PROTO) 1734 in6_pcbdetach(inp); 1735 else 1736 #endif 1737 in_pcbdetach(inp); 1738 tcpstat.tcps_closed++; 1739 crfree(tw->tw_cred); 1740 tw->tw_cred = NULL; 1741 if (reuse) 1742 return (tw); 1743 uma_zfree(tcptw_zone, tw); 1744 return (NULL); 1745 } 1746 1747 int 1748 tcp_twrespond(struct tcptw *tw, int flags) 1749 { 1750 struct inpcb *inp = tw->tw_inpcb; 1751 struct tcphdr *th; 1752 struct mbuf *m; 1753 struct ip *ip = NULL; 1754 u_int8_t *optp; 1755 u_int hdrlen, optlen; 1756 int error; 1757 #ifdef INET6 1758 struct ip6_hdr *ip6 = NULL; 1759 int isipv6 = inp->inp_inc.inc_isipv6; 1760 #endif 1761 1762 m = m_gethdr(M_DONTWAIT, MT_HEADER); 1763 if (m == NULL) 1764 return (ENOBUFS); 1765 m->m_data += max_linkhdr; 1766 1767 #ifdef MAC 1768 mac_create_mbuf_from_inpcb(inp, m); 1769 #endif 1770 1771 #ifdef INET6 1772 if (isipv6) { 1773 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1774 ip6 = mtod(m, struct ip6_hdr *); 1775 th = (struct tcphdr *)(ip6 + 1); 1776 tcpip_fillheaders(inp, ip6, th); 1777 } else 1778 #endif 1779 { 1780 hdrlen = sizeof(struct tcpiphdr); 1781 ip = mtod(m, struct ip *); 1782 th = (struct tcphdr *)(ip + 1); 1783 tcpip_fillheaders(inp, ip, th); 1784 } 1785 optp = (u_int8_t *)(th + 1); 1786 1787 /* 1788 * Send a timestamp and echo-reply if both our side and our peer 1789 * have sent timestamps in our SYN's and this is not a RST. 1790 */ 1791 if (tw->t_recent && flags == TH_ACK) { 1792 u_int32_t *lp = (u_int32_t *)optp; 1793 1794 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1795 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1796 *lp++ = htonl(ticks); 1797 *lp = htonl(tw->t_recent); 1798 optp += TCPOLEN_TSTAMP_APPA; 1799 } 1800 1801 /* 1802 * Send `CC-family' options if needed, and it's not a RST. 1803 */ 1804 if (tw->cc_recv != 0 && flags == TH_ACK) { 1805 u_int32_t *lp = (u_int32_t *)optp; 1806 1807 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC)); 1808 *lp = htonl(tw->cc_send); 1809 optp += TCPOLEN_CC_APPA; 1810 } 1811 optlen = optp - (u_int8_t *)(th + 1); 1812 1813 m->m_len = hdrlen + optlen; 1814 m->m_pkthdr.len = m->m_len; 1815 1816 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 1817 1818 th->th_seq = htonl(tw->snd_nxt); 1819 th->th_ack = htonl(tw->rcv_nxt); 1820 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1821 th->th_flags = flags; 1822 th->th_win = htons(tw->last_win); 1823 1824 #ifdef INET6 1825 if (isipv6) { 1826 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 1827 sizeof(struct tcphdr) + optlen); 1828 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 1829 error = ip6_output(m, inp->in6p_outputopts, NULL, 1830 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 1831 } else 1832 #endif 1833 { 1834 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1835 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 1836 m->m_pkthdr.csum_flags = CSUM_TCP; 1837 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1838 ip->ip_len = m->m_pkthdr.len; 1839 if (path_mtu_discovery) 1840 ip->ip_off |= IP_DF; 1841 error = ip_output(m, inp->inp_options, NULL, 1842 (tw->tw_so_options & SO_DONTROUTE), NULL, inp); 1843 } 1844 if (flags & TH_ACK) 1845 tcpstat.tcps_sndacks++; 1846 else 1847 tcpstat.tcps_sndctrl++; 1848 tcpstat.tcps_sndtotal++; 1849 return (error); 1850 } 1851 1852 /* 1853 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1854 * 1855 * This code attempts to calculate the bandwidth-delay product as a 1856 * means of determining the optimal window size to maximize bandwidth, 1857 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1858 * routers. This code also does a fairly good job keeping RTTs in check 1859 * across slow links like modems. We implement an algorithm which is very 1860 * similar (but not meant to be) TCP/Vegas. The code operates on the 1861 * transmitter side of a TCP connection and so only effects the transmit 1862 * side of the connection. 1863 * 1864 * BACKGROUND: TCP makes no provision for the management of buffer space 1865 * at the end points or at the intermediate routers and switches. A TCP 1866 * stream, whether using NewReno or not, will eventually buffer as 1867 * many packets as it is able and the only reason this typically works is 1868 * due to the fairly small default buffers made available for a connection 1869 * (typicaly 16K or 32K). As machines use larger windows and/or window 1870 * scaling it is now fairly easy for even a single TCP connection to blow-out 1871 * all available buffer space not only on the local interface, but on 1872 * intermediate routers and switches as well. NewReno makes a misguided 1873 * attempt to 'solve' this problem by waiting for an actual failure to occur, 1874 * then backing off, then steadily increasing the window again until another 1875 * failure occurs, ad-infinitum. This results in terrible oscillation that 1876 * is only made worse as network loads increase and the idea of intentionally 1877 * blowing out network buffers is, frankly, a terrible way to manage network 1878 * resources. 1879 * 1880 * It is far better to limit the transmit window prior to the failure 1881 * condition being achieved. There are two general ways to do this: First 1882 * you can 'scan' through different transmit window sizes and locate the 1883 * point where the RTT stops increasing, indicating that you have filled the 1884 * pipe, then scan backwards until you note that RTT stops decreasing, then 1885 * repeat ad-infinitum. This method works in principle but has severe 1886 * implementation issues due to RTT variances, timer granularity, and 1887 * instability in the algorithm which can lead to many false positives and 1888 * create oscillations as well as interact badly with other TCP streams 1889 * implementing the same algorithm. 1890 * 1891 * The second method is to limit the window to the bandwidth delay product 1892 * of the link. This is the method we implement. RTT variances and our 1893 * own manipulation of the congestion window, bwnd, can potentially 1894 * destabilize the algorithm. For this reason we have to stabilize the 1895 * elements used to calculate the window. We do this by using the minimum 1896 * observed RTT, the long term average of the observed bandwidth, and 1897 * by adding two segments worth of slop. It isn't perfect but it is able 1898 * to react to changing conditions and gives us a very stable basis on 1899 * which to extend the algorithm. 1900 */ 1901 void 1902 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 1903 { 1904 u_long bw; 1905 u_long bwnd; 1906 int save_ticks; 1907 1908 /* 1909 * If inflight_enable is disabled in the middle of a tcp connection, 1910 * make sure snd_bwnd is effectively disabled. 1911 */ 1912 if (tcp_inflight_enable == 0) { 1913 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1914 tp->snd_bandwidth = 0; 1915 return; 1916 } 1917 1918 /* 1919 * Figure out the bandwidth. Due to the tick granularity this 1920 * is a very rough number and it MUST be averaged over a fairly 1921 * long period of time. XXX we need to take into account a link 1922 * that is not using all available bandwidth, but for now our 1923 * slop will ramp us up if this case occurs and the bandwidth later 1924 * increases. 1925 * 1926 * Note: if ticks rollover 'bw' may wind up negative. We must 1927 * effectively reset t_bw_rtttime for this case. 1928 */ 1929 save_ticks = ticks; 1930 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 1931 return; 1932 1933 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 1934 (save_ticks - tp->t_bw_rtttime); 1935 tp->t_bw_rtttime = save_ticks; 1936 tp->t_bw_rtseq = ack_seq; 1937 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 1938 return; 1939 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 1940 1941 tp->snd_bandwidth = bw; 1942 1943 /* 1944 * Calculate the semi-static bandwidth delay product, plus two maximal 1945 * segments. The additional slop puts us squarely in the sweet 1946 * spot and also handles the bandwidth run-up case and stabilization. 1947 * Without the slop we could be locking ourselves into a lower 1948 * bandwidth. 1949 * 1950 * Situations Handled: 1951 * (1) Prevents over-queueing of packets on LANs, especially on 1952 * high speed LANs, allowing larger TCP buffers to be 1953 * specified, and also does a good job preventing 1954 * over-queueing of packets over choke points like modems 1955 * (at least for the transmit side). 1956 * 1957 * (2) Is able to handle changing network loads (bandwidth 1958 * drops so bwnd drops, bandwidth increases so bwnd 1959 * increases). 1960 * 1961 * (3) Theoretically should stabilize in the face of multiple 1962 * connections implementing the same algorithm (this may need 1963 * a little work). 1964 * 1965 * (4) Stability value (defaults to 20 = 2 maximal packets) can 1966 * be adjusted with a sysctl but typically only needs to be 1967 * on very slow connections. A value no smaller then 5 1968 * should be used, but only reduce this default if you have 1969 * no other choice. 1970 */ 1971 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 1972 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 1973 #undef USERTT 1974 1975 if (tcp_inflight_debug > 0) { 1976 static int ltime; 1977 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 1978 ltime = ticks; 1979 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 1980 tp, 1981 bw, 1982 tp->t_rttbest, 1983 tp->t_srtt, 1984 bwnd 1985 ); 1986 } 1987 } 1988 if ((long)bwnd < tcp_inflight_min) 1989 bwnd = tcp_inflight_min; 1990 if (bwnd > tcp_inflight_max) 1991 bwnd = tcp_inflight_max; 1992 if ((long)bwnd < tp->t_maxseg * 2) 1993 bwnd = tp->t_maxseg * 2; 1994 tp->snd_bwnd = bwnd; 1995 } 1996 1997 #ifdef TCP_SIGNATURE 1998 /* 1999 * Callback function invoked by m_apply() to digest TCP segment data 2000 * contained within an mbuf chain. 2001 */ 2002 static int 2003 tcp_signature_apply(void *fstate, void *data, u_int len) 2004 { 2005 2006 MD5Update(fstate, (u_char *)data, len); 2007 return (0); 2008 } 2009 2010 /* 2011 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2012 * 2013 * Parameters: 2014 * m pointer to head of mbuf chain 2015 * off0 offset to TCP header within the mbuf chain 2016 * len length of TCP segment data, excluding options 2017 * optlen length of TCP segment options 2018 * buf pointer to storage for computed MD5 digest 2019 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2020 * 2021 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2022 * When called from tcp_input(), we can be sure that th_sum has been 2023 * zeroed out and verified already. 2024 * 2025 * This function is for IPv4 use only. Calling this function with an 2026 * IPv6 packet in the mbuf chain will yield undefined results. 2027 * 2028 * Return 0 if successful, otherwise return -1. 2029 * 2030 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2031 * search with the destination IP address, and a 'magic SPI' to be 2032 * determined by the application. This is hardcoded elsewhere to 1179 2033 * right now. Another branch of this code exists which uses the SPD to 2034 * specify per-application flows but it is unstable. 2035 */ 2036 int 2037 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2038 u_char *buf, u_int direction) 2039 { 2040 union sockaddr_union dst; 2041 struct ippseudo ippseudo; 2042 MD5_CTX ctx; 2043 int doff; 2044 struct ip *ip; 2045 struct ipovly *ipovly; 2046 struct secasvar *sav; 2047 struct tcphdr *th; 2048 u_short savecsum; 2049 2050 KASSERT(m != NULL, ("NULL mbuf chain")); 2051 KASSERT(buf != NULL, ("NULL signature pointer")); 2052 2053 /* Extract the destination from the IP header in the mbuf. */ 2054 ip = mtod(m, struct ip *); 2055 bzero(&dst, sizeof(union sockaddr_union)); 2056 dst.sa.sa_len = sizeof(struct sockaddr_in); 2057 dst.sa.sa_family = AF_INET; 2058 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2059 ip->ip_src : ip->ip_dst; 2060 2061 /* Look up an SADB entry which matches the address of the peer. */ 2062 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2063 if (sav == NULL) { 2064 printf("%s: SADB lookup failed for %s\n", __func__, 2065 inet_ntoa(dst.sin.sin_addr)); 2066 return (EINVAL); 2067 } 2068 2069 MD5Init(&ctx); 2070 ipovly = (struct ipovly *)ip; 2071 th = (struct tcphdr *)((u_char *)ip + off0); 2072 doff = off0 + sizeof(struct tcphdr) + optlen; 2073 2074 /* 2075 * Step 1: Update MD5 hash with IP pseudo-header. 2076 * 2077 * XXX The ippseudo header MUST be digested in network byte order, 2078 * or else we'll fail the regression test. Assume all fields we've 2079 * been doing arithmetic on have been in host byte order. 2080 * XXX One cannot depend on ipovly->ih_len here. When called from 2081 * tcp_output(), the underlying ip_len member has not yet been set. 2082 */ 2083 ippseudo.ippseudo_src = ipovly->ih_src; 2084 ippseudo.ippseudo_dst = ipovly->ih_dst; 2085 ippseudo.ippseudo_pad = 0; 2086 ippseudo.ippseudo_p = IPPROTO_TCP; 2087 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2088 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2089 2090 /* 2091 * Step 2: Update MD5 hash with TCP header, excluding options. 2092 * The TCP checksum must be set to zero. 2093 */ 2094 savecsum = th->th_sum; 2095 th->th_sum = 0; 2096 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2097 th->th_sum = savecsum; 2098 2099 /* 2100 * Step 3: Update MD5 hash with TCP segment data. 2101 * Use m_apply() to avoid an early m_pullup(). 2102 */ 2103 if (len > 0) 2104 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2105 2106 /* 2107 * Step 4: Update MD5 hash with shared secret. 2108 */ 2109 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2110 MD5Final(buf, &ctx); 2111 2112 key_sa_recordxfer(sav, m); 2113 KEY_FREESAV(&sav); 2114 return (0); 2115 } 2116 #endif /* TCP_SIGNATURE */ 2117