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 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 187 "TCP inflight data limiting"); 188 189 static int tcp_inflight_enable = 1; 190 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 191 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 192 193 static int tcp_inflight_debug = 0; 194 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 195 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 196 197 static int tcp_inflight_min = 6144; 198 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 199 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 200 201 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 202 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 203 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 204 205 static int tcp_inflight_stab = 20; 206 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 207 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 208 209 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW, 0, "TCP SACK"); 210 int tcp_do_sack = 1; 211 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, enable, CTLFLAG_RW, 212 &tcp_do_sack, 0, "Enable/Disable TCP SACK support"); 213 214 uma_zone_t sack_hole_zone; 215 216 static struct inpcb *tcp_notify(struct inpcb *, int); 217 static void tcp_discardcb(struct tcpcb *); 218 static void tcp_isn_tick(void *); 219 220 /* 221 * Target size of TCP PCB hash tables. Must be a power of two. 222 * 223 * Note that this can be overridden by the kernel environment 224 * variable net.inet.tcp.tcbhashsize 225 */ 226 #ifndef TCBHASHSIZE 227 #define TCBHASHSIZE 512 228 #endif 229 230 /* 231 * XXX 232 * Callouts should be moved into struct tcp directly. They are currently 233 * separate because the tcpcb structure is exported to userland for sysctl 234 * parsing purposes, which do not know about callouts. 235 */ 236 struct tcpcb_mem { 237 struct tcpcb tcb; 238 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 239 struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 240 }; 241 242 static uma_zone_t tcpcb_zone; 243 static uma_zone_t tcptw_zone; 244 struct callout isn_callout; 245 246 /* 247 * Tcp initialization 248 */ 249 void 250 tcp_init() 251 { 252 int hashsize = TCBHASHSIZE; 253 254 tcp_ccgen = 1; 255 256 tcp_delacktime = TCPTV_DELACK; 257 tcp_keepinit = TCPTV_KEEP_INIT; 258 tcp_keepidle = TCPTV_KEEP_IDLE; 259 tcp_keepintvl = TCPTV_KEEPINTVL; 260 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 261 tcp_msl = TCPTV_MSL; 262 tcp_rexmit_min = TCPTV_MIN; 263 tcp_rexmit_slop = TCPTV_CPU_VAR; 264 265 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 266 LIST_INIT(&tcb); 267 tcbinfo.listhead = &tcb; 268 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 269 if (!powerof2(hashsize)) { 270 printf("WARNING: TCB hash size not a power of 2\n"); 271 hashsize = 512; /* safe default */ 272 } 273 tcp_tcbhashsize = hashsize; 274 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 275 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 276 &tcbinfo.porthashmask); 277 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 278 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 279 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 280 #ifdef INET6 281 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 282 #else /* INET6 */ 283 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 284 #endif /* INET6 */ 285 if (max_protohdr < TCP_MINPROTOHDR) 286 max_protohdr = TCP_MINPROTOHDR; 287 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 288 panic("tcp_init"); 289 #undef TCP_MINPROTOHDR 290 /* 291 * These have to be type stable for the benefit of the timers. 292 */ 293 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 294 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 295 uma_zone_set_max(tcpcb_zone, maxsockets); 296 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 297 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 298 uma_zone_set_max(tcptw_zone, maxsockets / 5); 299 tcp_timer_init(); 300 syncache_init(); 301 tcp_hc_init(); 302 tcp_reass_init(); 303 callout_init(&isn_callout, CALLOUT_MPSAFE); 304 tcp_isn_tick(NULL); 305 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 306 SHUTDOWN_PRI_DEFAULT); 307 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 308 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 309 } 310 311 void 312 tcp_fini(xtp) 313 void *xtp; 314 { 315 callout_stop(&isn_callout); 316 317 } 318 319 /* 320 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 321 * tcp_template used to store this data in mbufs, but we now recopy it out 322 * of the tcpcb each time to conserve mbufs. 323 */ 324 void 325 tcpip_fillheaders(inp, ip_ptr, tcp_ptr) 326 struct inpcb *inp; 327 void *ip_ptr; 328 void *tcp_ptr; 329 { 330 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 331 332 #ifdef INET6 333 if ((inp->inp_vflag & INP_IPV6) != 0) { 334 struct ip6_hdr *ip6; 335 336 ip6 = (struct ip6_hdr *)ip_ptr; 337 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 338 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 339 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 340 (IPV6_VERSION & IPV6_VERSION_MASK); 341 ip6->ip6_nxt = IPPROTO_TCP; 342 ip6->ip6_plen = sizeof(struct tcphdr); 343 ip6->ip6_src = inp->in6p_laddr; 344 ip6->ip6_dst = inp->in6p_faddr; 345 } else 346 #endif 347 { 348 struct ip *ip; 349 350 ip = (struct ip *)ip_ptr; 351 ip->ip_v = IPVERSION; 352 ip->ip_hl = 5; 353 ip->ip_tos = inp->inp_ip_tos; 354 ip->ip_len = 0; 355 ip->ip_id = 0; 356 ip->ip_off = 0; 357 ip->ip_ttl = inp->inp_ip_ttl; 358 ip->ip_sum = 0; 359 ip->ip_p = IPPROTO_TCP; 360 ip->ip_src = inp->inp_laddr; 361 ip->ip_dst = inp->inp_faddr; 362 } 363 th->th_sport = inp->inp_lport; 364 th->th_dport = inp->inp_fport; 365 th->th_seq = 0; 366 th->th_ack = 0; 367 th->th_x2 = 0; 368 th->th_off = 5; 369 th->th_flags = 0; 370 th->th_win = 0; 371 th->th_urp = 0; 372 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 373 } 374 375 /* 376 * Create template to be used to send tcp packets on a connection. 377 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 378 * use for this function is in keepalives, which use tcp_respond. 379 */ 380 struct tcptemp * 381 tcpip_maketemplate(inp) 382 struct inpcb *inp; 383 { 384 struct mbuf *m; 385 struct tcptemp *n; 386 387 m = m_get(M_DONTWAIT, MT_HEADER); 388 if (m == NULL) 389 return (0); 390 m->m_len = sizeof(struct tcptemp); 391 n = mtod(m, struct tcptemp *); 392 393 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 394 return (n); 395 } 396 397 /* 398 * Send a single message to the TCP at address specified by 399 * the given TCP/IP header. If m == NULL, then we make a copy 400 * of the tcpiphdr at ti and send directly to the addressed host. 401 * This is used to force keep alive messages out using the TCP 402 * template for a connection. If flags are given then we send 403 * a message back to the TCP which originated the * segment ti, 404 * and discard the mbuf containing it and any other attached mbufs. 405 * 406 * In any case the ack and sequence number of the transmitted 407 * segment are as specified by the parameters. 408 * 409 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 410 */ 411 void 412 tcp_respond(tp, ipgen, th, m, ack, seq, flags) 413 struct tcpcb *tp; 414 void *ipgen; 415 register struct tcphdr *th; 416 register struct mbuf *m; 417 tcp_seq ack, seq; 418 int flags; 419 { 420 register int tlen; 421 int win = 0; 422 struct ip *ip; 423 struct tcphdr *nth; 424 #ifdef INET6 425 struct ip6_hdr *ip6; 426 int isipv6; 427 #endif /* INET6 */ 428 int ipflags = 0; 429 struct inpcb *inp; 430 431 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 432 433 #ifdef INET6 434 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 435 ip6 = ipgen; 436 #endif /* INET6 */ 437 ip = ipgen; 438 439 if (tp != NULL) { 440 inp = tp->t_inpcb; 441 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 442 INP_INFO_WLOCK_ASSERT(&tcbinfo); 443 INP_LOCK_ASSERT(inp); 444 } else 445 inp = NULL; 446 447 if (tp != NULL) { 448 if (!(flags & TH_RST)) { 449 win = sbspace(&inp->inp_socket->so_rcv); 450 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 451 win = (long)TCP_MAXWIN << tp->rcv_scale; 452 } 453 } 454 if (m == NULL) { 455 m = m_gethdr(M_DONTWAIT, MT_HEADER); 456 if (m == NULL) 457 return; 458 tlen = 0; 459 m->m_data += max_linkhdr; 460 #ifdef INET6 461 if (isipv6) { 462 bcopy((caddr_t)ip6, mtod(m, caddr_t), 463 sizeof(struct ip6_hdr)); 464 ip6 = mtod(m, struct ip6_hdr *); 465 nth = (struct tcphdr *)(ip6 + 1); 466 } else 467 #endif /* INET6 */ 468 { 469 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 470 ip = mtod(m, struct ip *); 471 nth = (struct tcphdr *)(ip + 1); 472 } 473 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 474 flags = TH_ACK; 475 } else { 476 m_freem(m->m_next); 477 m->m_next = NULL; 478 m->m_data = (caddr_t)ipgen; 479 /* m_len is set later */ 480 tlen = 0; 481 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 482 #ifdef INET6 483 if (isipv6) { 484 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 485 nth = (struct tcphdr *)(ip6 + 1); 486 } else 487 #endif /* INET6 */ 488 { 489 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 490 nth = (struct tcphdr *)(ip + 1); 491 } 492 if (th != nth) { 493 /* 494 * this is usually a case when an extension header 495 * exists between the IPv6 header and the 496 * TCP header. 497 */ 498 nth->th_sport = th->th_sport; 499 nth->th_dport = th->th_dport; 500 } 501 xchg(nth->th_dport, nth->th_sport, n_short); 502 #undef xchg 503 } 504 #ifdef INET6 505 if (isipv6) { 506 ip6->ip6_flow = 0; 507 ip6->ip6_vfc = IPV6_VERSION; 508 ip6->ip6_nxt = IPPROTO_TCP; 509 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 510 tlen)); 511 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 512 } else 513 #endif 514 { 515 tlen += sizeof (struct tcpiphdr); 516 ip->ip_len = tlen; 517 ip->ip_ttl = ip_defttl; 518 if (path_mtu_discovery) 519 ip->ip_off |= IP_DF; 520 } 521 m->m_len = tlen; 522 m->m_pkthdr.len = tlen; 523 m->m_pkthdr.rcvif = NULL; 524 #ifdef MAC 525 if (inp != NULL) { 526 /* 527 * Packet is associated with a socket, so allow the 528 * label of the response to reflect the socket label. 529 */ 530 INP_LOCK_ASSERT(inp); 531 mac_create_mbuf_from_inpcb(inp, m); 532 } else { 533 /* 534 * Packet is not associated with a socket, so possibly 535 * update the label in place. 536 */ 537 mac_reflect_mbuf_tcp(m); 538 } 539 #endif 540 nth->th_seq = htonl(seq); 541 nth->th_ack = htonl(ack); 542 nth->th_x2 = 0; 543 nth->th_off = sizeof (struct tcphdr) >> 2; 544 nth->th_flags = flags; 545 if (tp != NULL) 546 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 547 else 548 nth->th_win = htons((u_short)win); 549 nth->th_urp = 0; 550 #ifdef INET6 551 if (isipv6) { 552 nth->th_sum = 0; 553 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 554 sizeof(struct ip6_hdr), 555 tlen - sizeof(struct ip6_hdr)); 556 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 557 NULL, NULL); 558 } else 559 #endif /* INET6 */ 560 { 561 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 562 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 563 m->m_pkthdr.csum_flags = CSUM_TCP; 564 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 565 } 566 #ifdef TCPDEBUG 567 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 568 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 569 #endif 570 #ifdef INET6 571 if (isipv6) 572 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 573 else 574 #endif /* INET6 */ 575 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 576 } 577 578 /* 579 * Create a new TCP control block, making an 580 * empty reassembly queue and hooking it to the argument 581 * protocol control block. The `inp' parameter must have 582 * come from the zone allocator set up in tcp_init(). 583 */ 584 struct tcpcb * 585 tcp_newtcpcb(inp) 586 struct inpcb *inp; 587 { 588 struct tcpcb_mem *tm; 589 struct tcpcb *tp; 590 #ifdef INET6 591 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 592 #endif /* INET6 */ 593 int callout_flag; 594 595 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 596 if (tm == NULL) 597 return (NULL); 598 tp = &tm->tcb; 599 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 600 tp->t_maxseg = tp->t_maxopd = 601 #ifdef INET6 602 isipv6 ? tcp_v6mssdflt : 603 #endif /* INET6 */ 604 tcp_mssdflt; 605 606 /* Set up our timeouts. */ 607 /* 608 * XXXRW: Are these actually MPSAFE? I think so, but need to 609 * review the timed wait code, as it has some list variables, 610 * etc, that are global. 611 */ 612 callout_flag = debug_mpsafenet ? CALLOUT_MPSAFE : 0; 613 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, callout_flag); 614 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, callout_flag); 615 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, callout_flag); 616 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, callout_flag); 617 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, callout_flag); 618 619 if (tcp_do_rfc1323) 620 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 621 if (tcp_do_rfc1644) 622 tp->t_flags |= TF_REQ_CC; 623 tp->sack_enable = tcp_do_sack; 624 tp->t_inpcb = inp; /* XXX */ 625 /* 626 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 627 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 628 * reasonable initial retransmit time. 629 */ 630 tp->t_srtt = TCPTV_SRTTBASE; 631 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 632 tp->t_rttmin = tcp_rexmit_min; 633 tp->t_rxtcur = TCPTV_RTOBASE; 634 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 635 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 636 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 637 tp->t_rcvtime = ticks; 638 tp->t_bw_rtttime = ticks; 639 /* 640 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 641 * because the socket may be bound to an IPv6 wildcard address, 642 * which may match an IPv4-mapped IPv6 address. 643 */ 644 inp->inp_ip_ttl = ip_defttl; 645 inp->inp_ppcb = (caddr_t)tp; 646 return (tp); /* XXX */ 647 } 648 649 /* 650 * Drop a TCP connection, reporting 651 * the specified error. If connection is synchronized, 652 * then send a RST to peer. 653 */ 654 struct tcpcb * 655 tcp_drop(tp, errno) 656 register struct tcpcb *tp; 657 int errno; 658 { 659 struct socket *so = tp->t_inpcb->inp_socket; 660 661 if (TCPS_HAVERCVDSYN(tp->t_state)) { 662 tp->t_state = TCPS_CLOSED; 663 (void) tcp_output(tp); 664 tcpstat.tcps_drops++; 665 } else 666 tcpstat.tcps_conndrops++; 667 if (errno == ETIMEDOUT && tp->t_softerror) 668 errno = tp->t_softerror; 669 so->so_error = errno; 670 return (tcp_close(tp)); 671 } 672 673 static void 674 tcp_discardcb(tp) 675 struct tcpcb *tp; 676 { 677 struct tseg_qent *q; 678 struct inpcb *inp = tp->t_inpcb; 679 struct socket *so = inp->inp_socket; 680 #ifdef INET6 681 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 682 #endif /* INET6 */ 683 684 /* 685 * Make sure that all of our timers are stopped before we 686 * delete the PCB. 687 */ 688 callout_stop(tp->tt_rexmt); 689 callout_stop(tp->tt_persist); 690 callout_stop(tp->tt_keep); 691 callout_stop(tp->tt_2msl); 692 callout_stop(tp->tt_delack); 693 694 /* 695 * If we got enough samples through the srtt filter, 696 * save the rtt and rttvar in the routing entry. 697 * 'Enough' is arbitrarily defined as 4 rtt samples. 698 * 4 samples is enough for the srtt filter to converge 699 * to within enough % of the correct value; fewer samples 700 * and we could save a bogus rtt. The danger is not high 701 * as tcp quickly recovers from everything. 702 * XXX: Works very well but needs some more statistics! 703 */ 704 if (tp->t_rttupdated >= 4) { 705 struct hc_metrics_lite metrics; 706 u_long ssthresh; 707 708 bzero(&metrics, sizeof(metrics)); 709 /* 710 * Update the ssthresh always when the conditions below 711 * are satisfied. This gives us better new start value 712 * for the congestion avoidance for new connections. 713 * ssthresh is only set if packet loss occured on a session. 714 */ 715 ssthresh = tp->snd_ssthresh; 716 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 717 /* 718 * convert the limit from user data bytes to 719 * packets then to packet data bytes. 720 */ 721 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 722 if (ssthresh < 2) 723 ssthresh = 2; 724 ssthresh *= (u_long)(tp->t_maxseg + 725 #ifdef INET6 726 (isipv6 ? sizeof (struct ip6_hdr) + 727 sizeof (struct tcphdr) : 728 #endif 729 sizeof (struct tcpiphdr) 730 #ifdef INET6 731 ) 732 #endif 733 ); 734 } else 735 ssthresh = 0; 736 metrics.rmx_ssthresh = ssthresh; 737 738 metrics.rmx_rtt = tp->t_srtt; 739 metrics.rmx_rttvar = tp->t_rttvar; 740 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 741 metrics.rmx_bandwidth = tp->snd_bandwidth; 742 metrics.rmx_cwnd = tp->snd_cwnd; 743 metrics.rmx_sendpipe = 0; 744 metrics.rmx_recvpipe = 0; 745 746 tcp_hc_update(&inp->inp_inc, &metrics); 747 } 748 749 /* free the reassembly queue, if any */ 750 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 751 LIST_REMOVE(q, tqe_q); 752 m_freem(q->tqe_m); 753 uma_zfree(tcp_reass_zone, q); 754 tp->t_segqlen--; 755 tcp_reass_qsize--; 756 } 757 tcp_free_sackholes(tp); 758 inp->inp_ppcb = NULL; 759 tp->t_inpcb = NULL; 760 uma_zfree(tcpcb_zone, tp); 761 soisdisconnected(so); 762 } 763 764 /* 765 * Close a TCP control block: 766 * discard all space held by the tcp 767 * discard internet protocol block 768 * wake up any sleepers 769 */ 770 struct tcpcb * 771 tcp_close(tp) 772 struct tcpcb *tp; 773 { 774 struct inpcb *inp = tp->t_inpcb; 775 #ifdef INET6 776 struct socket *so = inp->inp_socket; 777 #endif 778 779 tcp_discardcb(tp); 780 #ifdef INET6 781 if (INP_CHECK_SOCKAF(so, AF_INET6)) 782 in6_pcbdetach(inp); 783 else 784 #endif 785 in_pcbdetach(inp); 786 tcpstat.tcps_closed++; 787 return (NULL); 788 } 789 790 void 791 tcp_drain() 792 { 793 if (do_tcpdrain) 794 { 795 struct inpcb *inpb; 796 struct tcpcb *tcpb; 797 struct tseg_qent *te; 798 799 /* 800 * Walk the tcpbs, if existing, and flush the reassembly queue, 801 * if there is one... 802 * XXX: The "Net/3" implementation doesn't imply that the TCP 803 * reassembly queue should be flushed, but in a situation 804 * where we're really low on mbufs, this is potentially 805 * usefull. 806 */ 807 INP_INFO_RLOCK(&tcbinfo); 808 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 809 if (inpb->inp_vflag & INP_TIMEWAIT) 810 continue; 811 INP_LOCK(inpb); 812 if ((tcpb = intotcpcb(inpb)) != NULL) { 813 while ((te = LIST_FIRST(&tcpb->t_segq)) 814 != NULL) { 815 LIST_REMOVE(te, tqe_q); 816 m_freem(te->tqe_m); 817 uma_zfree(tcp_reass_zone, te); 818 tcpb->t_segqlen--; 819 tcp_reass_qsize--; 820 } 821 } 822 INP_UNLOCK(inpb); 823 } 824 INP_INFO_RUNLOCK(&tcbinfo); 825 } 826 } 827 828 /* 829 * Notify a tcp user of an asynchronous error; 830 * store error as soft error, but wake up user 831 * (for now, won't do anything until can select for soft error). 832 * 833 * Do not wake up user since there currently is no mechanism for 834 * reporting soft errors (yet - a kqueue filter may be added). 835 */ 836 static struct inpcb * 837 tcp_notify(inp, error) 838 struct inpcb *inp; 839 int error; 840 { 841 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 842 843 /* 844 * Ignore some errors if we are hooked up. 845 * If connection hasn't completed, has retransmitted several times, 846 * and receives a second error, give up now. This is better 847 * than waiting a long time to establish a connection that 848 * can never complete. 849 */ 850 if (tp->t_state == TCPS_ESTABLISHED && 851 (error == EHOSTUNREACH || error == ENETUNREACH || 852 error == EHOSTDOWN)) { 853 return inp; 854 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 855 tp->t_softerror) { 856 tcp_drop(tp, error); 857 return (struct inpcb *)0; 858 } else { 859 tp->t_softerror = error; 860 return inp; 861 } 862 #if 0 863 wakeup( &so->so_timeo); 864 sorwakeup(so); 865 sowwakeup(so); 866 #endif 867 } 868 869 static int 870 tcp_pcblist(SYSCTL_HANDLER_ARGS) 871 { 872 int error, i, n, s; 873 struct inpcb *inp, **inp_list; 874 inp_gen_t gencnt; 875 struct xinpgen xig; 876 877 /* 878 * The process of preparing the TCB list is too time-consuming and 879 * resource-intensive to repeat twice on every request. 880 */ 881 if (req->oldptr == NULL) { 882 n = tcbinfo.ipi_count; 883 req->oldidx = 2 * (sizeof xig) 884 + (n + n/8) * sizeof(struct xtcpcb); 885 return 0; 886 } 887 888 if (req->newptr != NULL) 889 return EPERM; 890 891 /* 892 * OK, now we're committed to doing something. 893 */ 894 s = splnet(); 895 INP_INFO_RLOCK(&tcbinfo); 896 gencnt = tcbinfo.ipi_gencnt; 897 n = tcbinfo.ipi_count; 898 INP_INFO_RUNLOCK(&tcbinfo); 899 splx(s); 900 901 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 902 + n * sizeof(struct xtcpcb)); 903 if (error != 0) 904 return (error); 905 906 xig.xig_len = sizeof xig; 907 xig.xig_count = n; 908 xig.xig_gen = gencnt; 909 xig.xig_sogen = so_gencnt; 910 error = SYSCTL_OUT(req, &xig, sizeof xig); 911 if (error) 912 return error; 913 914 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 915 if (inp_list == NULL) 916 return ENOMEM; 917 918 s = splnet(); 919 INP_INFO_RLOCK(&tcbinfo); 920 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 921 inp = LIST_NEXT(inp, inp_list)) { 922 INP_LOCK(inp); 923 if (inp->inp_gencnt <= gencnt) { 924 /* 925 * XXX: This use of cr_cansee(), introduced with 926 * TCP state changes, is not quite right, but for 927 * now, better than nothing. 928 */ 929 if (inp->inp_vflag & INP_TIMEWAIT) 930 error = cr_cansee(req->td->td_ucred, 931 intotw(inp)->tw_cred); 932 else 933 error = cr_canseesocket(req->td->td_ucred, 934 inp->inp_socket); 935 if (error == 0) 936 inp_list[i++] = inp; 937 } 938 INP_UNLOCK(inp); 939 } 940 INP_INFO_RUNLOCK(&tcbinfo); 941 splx(s); 942 n = i; 943 944 error = 0; 945 for (i = 0; i < n; i++) { 946 inp = inp_list[i]; 947 if (inp->inp_gencnt <= gencnt) { 948 struct xtcpcb xt; 949 caddr_t inp_ppcb; 950 xt.xt_len = sizeof xt; 951 /* XXX should avoid extra copy */ 952 bcopy(inp, &xt.xt_inp, sizeof *inp); 953 inp_ppcb = inp->inp_ppcb; 954 if (inp_ppcb == NULL) 955 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 956 else if (inp->inp_vflag & INP_TIMEWAIT) { 957 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 958 xt.xt_tp.t_state = TCPS_TIME_WAIT; 959 } else 960 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 961 if (inp->inp_socket != NULL) 962 sotoxsocket(inp->inp_socket, &xt.xt_socket); 963 else { 964 bzero(&xt.xt_socket, sizeof xt.xt_socket); 965 xt.xt_socket.xso_protocol = IPPROTO_TCP; 966 } 967 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 968 error = SYSCTL_OUT(req, &xt, sizeof xt); 969 } 970 } 971 if (!error) { 972 /* 973 * Give the user an updated idea of our state. 974 * If the generation differs from what we told 975 * her before, she knows that something happened 976 * while we were processing this request, and it 977 * might be necessary to retry. 978 */ 979 s = splnet(); 980 INP_INFO_RLOCK(&tcbinfo); 981 xig.xig_gen = tcbinfo.ipi_gencnt; 982 xig.xig_sogen = so_gencnt; 983 xig.xig_count = tcbinfo.ipi_count; 984 INP_INFO_RUNLOCK(&tcbinfo); 985 splx(s); 986 error = SYSCTL_OUT(req, &xig, sizeof xig); 987 } 988 free(inp_list, M_TEMP); 989 return error; 990 } 991 992 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 993 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 994 995 static int 996 tcp_getcred(SYSCTL_HANDLER_ARGS) 997 { 998 struct xucred xuc; 999 struct sockaddr_in addrs[2]; 1000 struct inpcb *inp; 1001 int error, s; 1002 1003 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1004 if (error) 1005 return (error); 1006 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1007 if (error) 1008 return (error); 1009 s = splnet(); 1010 INP_INFO_RLOCK(&tcbinfo); 1011 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1012 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1013 if (inp == NULL) { 1014 error = ENOENT; 1015 goto outunlocked; 1016 } 1017 INP_LOCK(inp); 1018 if (inp->inp_socket == NULL) { 1019 error = ENOENT; 1020 goto out; 1021 } 1022 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1023 if (error) 1024 goto out; 1025 cru2x(inp->inp_socket->so_cred, &xuc); 1026 out: 1027 INP_UNLOCK(inp); 1028 outunlocked: 1029 INP_INFO_RUNLOCK(&tcbinfo); 1030 splx(s); 1031 if (error == 0) 1032 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1033 return (error); 1034 } 1035 1036 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1037 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1038 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1039 1040 #ifdef INET6 1041 static int 1042 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1043 { 1044 struct xucred xuc; 1045 struct sockaddr_in6 addrs[2]; 1046 struct inpcb *inp; 1047 int error, s, mapped = 0; 1048 1049 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1050 if (error) 1051 return (error); 1052 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1053 if (error) 1054 return (error); 1055 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1056 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1057 mapped = 1; 1058 else 1059 return (EINVAL); 1060 } 1061 s = splnet(); 1062 INP_INFO_RLOCK(&tcbinfo); 1063 if (mapped == 1) 1064 inp = in_pcblookup_hash(&tcbinfo, 1065 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1066 addrs[1].sin6_port, 1067 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1068 addrs[0].sin6_port, 1069 0, NULL); 1070 else 1071 inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr, 1072 addrs[1].sin6_port, 1073 &addrs[0].sin6_addr, addrs[0].sin6_port, 1074 0, NULL); 1075 if (inp == NULL) { 1076 error = ENOENT; 1077 goto outunlocked; 1078 } 1079 INP_LOCK(inp); 1080 if (inp->inp_socket == NULL) { 1081 error = ENOENT; 1082 goto out; 1083 } 1084 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1085 if (error) 1086 goto out; 1087 cru2x(inp->inp_socket->so_cred, &xuc); 1088 out: 1089 INP_UNLOCK(inp); 1090 outunlocked: 1091 INP_INFO_RUNLOCK(&tcbinfo); 1092 splx(s); 1093 if (error == 0) 1094 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1095 return (error); 1096 } 1097 1098 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1099 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1100 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1101 #endif 1102 1103 1104 void 1105 tcp_ctlinput(cmd, sa, vip) 1106 int cmd; 1107 struct sockaddr *sa; 1108 void *vip; 1109 { 1110 struct ip *ip = vip; 1111 struct tcphdr *th; 1112 struct in_addr faddr; 1113 struct inpcb *inp; 1114 struct tcpcb *tp; 1115 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1116 tcp_seq icmp_seq; 1117 int s; 1118 1119 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1120 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1121 return; 1122 1123 if (cmd == PRC_QUENCH) 1124 notify = tcp_quench; 1125 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1126 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1127 notify = tcp_drop_syn_sent; 1128 else if (cmd == PRC_MSGSIZE) 1129 notify = tcp_mtudisc; 1130 /* 1131 * Redirects don't need to be handled up here. 1132 */ 1133 else if (PRC_IS_REDIRECT(cmd)) 1134 return; 1135 /* 1136 * Hostdead is ugly because it goes linearly through all PCBs. 1137 * XXX: We never get this from ICMP, otherwise it makes an 1138 * excellent DoS attack on machines with many connections. 1139 */ 1140 else if (cmd == PRC_HOSTDEAD) 1141 ip = NULL; 1142 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1143 return; 1144 if (ip != NULL) { 1145 s = splnet(); 1146 th = (struct tcphdr *)((caddr_t)ip 1147 + (ip->ip_hl << 2)); 1148 INP_INFO_WLOCK(&tcbinfo); 1149 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1150 ip->ip_src, th->th_sport, 0, NULL); 1151 if (inp != NULL) { 1152 INP_LOCK(inp); 1153 if (inp->inp_socket != NULL) { 1154 icmp_seq = htonl(th->th_seq); 1155 tp = intotcpcb(inp); 1156 if (SEQ_GEQ(icmp_seq, tp->snd_una) && 1157 SEQ_LT(icmp_seq, tp->snd_max)) 1158 inp = (*notify)(inp, inetctlerrmap[cmd]); 1159 } 1160 if (inp != NULL) 1161 INP_UNLOCK(inp); 1162 } else { 1163 struct in_conninfo inc; 1164 1165 inc.inc_fport = th->th_dport; 1166 inc.inc_lport = th->th_sport; 1167 inc.inc_faddr = faddr; 1168 inc.inc_laddr = ip->ip_src; 1169 #ifdef INET6 1170 inc.inc_isipv6 = 0; 1171 #endif 1172 syncache_unreach(&inc, th); 1173 } 1174 INP_INFO_WUNLOCK(&tcbinfo); 1175 splx(s); 1176 } else 1177 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1178 } 1179 1180 #ifdef INET6 1181 void 1182 tcp6_ctlinput(cmd, sa, d) 1183 int cmd; 1184 struct sockaddr *sa; 1185 void *d; 1186 { 1187 struct tcphdr th; 1188 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1189 struct ip6_hdr *ip6; 1190 struct mbuf *m; 1191 struct ip6ctlparam *ip6cp = NULL; 1192 const struct sockaddr_in6 *sa6_src = NULL; 1193 int off; 1194 struct tcp_portonly { 1195 u_int16_t th_sport; 1196 u_int16_t th_dport; 1197 } *thp; 1198 1199 if (sa->sa_family != AF_INET6 || 1200 sa->sa_len != sizeof(struct sockaddr_in6)) 1201 return; 1202 1203 if (cmd == PRC_QUENCH) 1204 notify = tcp_quench; 1205 else if (cmd == PRC_MSGSIZE) 1206 notify = tcp_mtudisc; 1207 else if (!PRC_IS_REDIRECT(cmd) && 1208 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1209 return; 1210 1211 /* if the parameter is from icmp6, decode it. */ 1212 if (d != NULL) { 1213 ip6cp = (struct ip6ctlparam *)d; 1214 m = ip6cp->ip6c_m; 1215 ip6 = ip6cp->ip6c_ip6; 1216 off = ip6cp->ip6c_off; 1217 sa6_src = ip6cp->ip6c_src; 1218 } else { 1219 m = NULL; 1220 ip6 = NULL; 1221 off = 0; /* fool gcc */ 1222 sa6_src = &sa6_any; 1223 } 1224 1225 if (ip6 != NULL) { 1226 struct in_conninfo inc; 1227 /* 1228 * XXX: We assume that when IPV6 is non NULL, 1229 * M and OFF are valid. 1230 */ 1231 1232 /* check if we can safely examine src and dst ports */ 1233 if (m->m_pkthdr.len < off + sizeof(*thp)) 1234 return; 1235 1236 bzero(&th, sizeof(th)); 1237 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1238 1239 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1240 (struct sockaddr *)ip6cp->ip6c_src, 1241 th.th_sport, cmd, NULL, notify); 1242 1243 inc.inc_fport = th.th_dport; 1244 inc.inc_lport = th.th_sport; 1245 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1246 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1247 inc.inc_isipv6 = 1; 1248 INP_INFO_WLOCK(&tcbinfo); 1249 syncache_unreach(&inc, &th); 1250 INP_INFO_WUNLOCK(&tcbinfo); 1251 } else 1252 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1253 0, cmd, NULL, notify); 1254 } 1255 #endif /* INET6 */ 1256 1257 1258 /* 1259 * Following is where TCP initial sequence number generation occurs. 1260 * 1261 * There are two places where we must use initial sequence numbers: 1262 * 1. In SYN-ACK packets. 1263 * 2. In SYN packets. 1264 * 1265 * All ISNs for SYN-ACK packets are generated by the syncache. See 1266 * tcp_syncache.c for details. 1267 * 1268 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1269 * depends on this property. In addition, these ISNs should be 1270 * unguessable so as to prevent connection hijacking. To satisfy 1271 * the requirements of this situation, the algorithm outlined in 1272 * RFC 1948 is used, with only small modifications. 1273 * 1274 * Implementation details: 1275 * 1276 * Time is based off the system timer, and is corrected so that it 1277 * increases by one megabyte per second. This allows for proper 1278 * recycling on high speed LANs while still leaving over an hour 1279 * before rollover. 1280 * 1281 * As reading the *exact* system time is too expensive to be done 1282 * whenever setting up a TCP connection, we increment the time 1283 * offset in two ways. First, a small random positive increment 1284 * is added to isn_offset for each connection that is set up. 1285 * Second, the function tcp_isn_tick fires once per clock tick 1286 * and increments isn_offset as necessary so that sequence numbers 1287 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1288 * random positive increments serve only to ensure that the same 1289 * exact sequence number is never sent out twice (as could otherwise 1290 * happen when a port is recycled in less than the system tick 1291 * interval.) 1292 * 1293 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1294 * between seeding of isn_secret. This is normally set to zero, 1295 * as reseeding should not be necessary. 1296 * 1297 */ 1298 1299 #define ISN_BYTES_PER_SECOND 1048576 1300 #define ISN_STATIC_INCREMENT 4096 1301 #define ISN_RANDOM_INCREMENT (4096 - 1) 1302 1303 u_char isn_secret[32]; 1304 int isn_last_reseed; 1305 u_int32_t isn_offset, isn_offset_old; 1306 MD5_CTX isn_ctx; 1307 1308 tcp_seq 1309 tcp_new_isn(tp) 1310 struct tcpcb *tp; 1311 { 1312 u_int32_t md5_buffer[4]; 1313 tcp_seq new_isn; 1314 1315 /* Seed if this is the first use, reseed if requested. */ 1316 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1317 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1318 < (u_int)ticks))) { 1319 read_random(&isn_secret, sizeof(isn_secret)); 1320 isn_last_reseed = ticks; 1321 } 1322 1323 /* Compute the md5 hash and return the ISN. */ 1324 MD5Init(&isn_ctx); 1325 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1326 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1327 #ifdef INET6 1328 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1329 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1330 sizeof(struct in6_addr)); 1331 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1332 sizeof(struct in6_addr)); 1333 } else 1334 #endif 1335 { 1336 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1337 sizeof(struct in_addr)); 1338 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1339 sizeof(struct in_addr)); 1340 } 1341 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1342 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1343 new_isn = (tcp_seq) md5_buffer[0]; 1344 isn_offset += ISN_STATIC_INCREMENT + 1345 (arc4random() & ISN_RANDOM_INCREMENT); 1346 new_isn += isn_offset; 1347 return new_isn; 1348 } 1349 1350 /* 1351 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1352 * to keep time flowing at a relatively constant rate. If the random 1353 * increments have already pushed us past the projected offset, do nothing. 1354 */ 1355 static void 1356 tcp_isn_tick(xtp) 1357 void *xtp; 1358 { 1359 u_int32_t projected_offset; 1360 1361 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / hz; 1362 1363 if (projected_offset > isn_offset) 1364 isn_offset = projected_offset; 1365 1366 isn_offset_old = isn_offset; 1367 callout_reset(&isn_callout, 1, tcp_isn_tick, NULL); 1368 } 1369 1370 /* 1371 * When a source quench is received, close congestion window 1372 * to one segment. We will gradually open it again as we proceed. 1373 */ 1374 struct inpcb * 1375 tcp_quench(inp, errno) 1376 struct inpcb *inp; 1377 int errno; 1378 { 1379 struct tcpcb *tp = intotcpcb(inp); 1380 1381 if (tp != NULL) 1382 tp->snd_cwnd = tp->t_maxseg; 1383 return (inp); 1384 } 1385 1386 /* 1387 * When a specific ICMP unreachable message is received and the 1388 * connection state is SYN-SENT, drop the connection. This behavior 1389 * is controlled by the icmp_may_rst sysctl. 1390 */ 1391 struct inpcb * 1392 tcp_drop_syn_sent(inp, errno) 1393 struct inpcb *inp; 1394 int errno; 1395 { 1396 struct tcpcb *tp = intotcpcb(inp); 1397 1398 if (tp != NULL && tp->t_state == TCPS_SYN_SENT) { 1399 tcp_drop(tp, errno); 1400 return (struct inpcb *)0; 1401 } 1402 return inp; 1403 } 1404 1405 /* 1406 * When `need fragmentation' ICMP is received, update our idea of the MSS 1407 * based on the new value in the route. Also nudge TCP to send something, 1408 * since we know the packet we just sent was dropped. 1409 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1410 */ 1411 struct inpcb * 1412 tcp_mtudisc(inp, errno) 1413 struct inpcb *inp; 1414 int errno; 1415 { 1416 struct tcpcb *tp = intotcpcb(inp); 1417 struct rmxp_tao tao; 1418 struct socket *so = inp->inp_socket; 1419 u_int maxmtu; 1420 u_int romtu; 1421 int mss; 1422 #ifdef INET6 1423 int isipv6; 1424 #endif /* INET6 */ 1425 bzero(&tao, sizeof(tao)); 1426 1427 if (tp != NULL) { 1428 #ifdef INET6 1429 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1430 #endif 1431 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1432 romtu = 1433 #ifdef INET6 1434 isipv6 ? tcp_maxmtu6(&inp->inp_inc) : 1435 #endif /* INET6 */ 1436 tcp_maxmtu(&inp->inp_inc); 1437 if (!maxmtu) 1438 maxmtu = romtu; 1439 else 1440 maxmtu = min(maxmtu, romtu); 1441 if (!maxmtu) { 1442 tp->t_maxopd = tp->t_maxseg = 1443 #ifdef INET6 1444 isipv6 ? tcp_v6mssdflt : 1445 #endif /* INET6 */ 1446 tcp_mssdflt; 1447 return inp; 1448 } 1449 mss = maxmtu - 1450 #ifdef INET6 1451 (isipv6 ? 1452 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1453 #endif /* INET6 */ 1454 sizeof(struct tcpiphdr) 1455 #ifdef INET6 1456 ) 1457 #endif /* INET6 */ 1458 ; 1459 1460 if (tcp_do_rfc1644) { 1461 tcp_hc_gettao(&inp->inp_inc, &tao); 1462 if (tao.tao_mssopt) 1463 mss = min(mss, tao.tao_mssopt); 1464 } 1465 /* 1466 * XXX - The above conditional probably violates the TCP 1467 * spec. The problem is that, since we don't know the 1468 * other end's MSS, we are supposed to use a conservative 1469 * default. But, if we do that, then MTU discovery will 1470 * never actually take place, because the conservative 1471 * default is much less than the MTUs typically seen 1472 * on the Internet today. For the moment, we'll sweep 1473 * this under the carpet. 1474 * 1475 * The conservative default might not actually be a problem 1476 * if the only case this occurs is when sending an initial 1477 * SYN with options and data to a host we've never talked 1478 * to before. Then, they will reply with an MSS value which 1479 * will get recorded and the new parameters should get 1480 * recomputed. For Further Study. 1481 */ 1482 if (tp->t_maxopd <= mss) 1483 return inp; 1484 tp->t_maxopd = mss; 1485 1486 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1487 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1488 mss -= TCPOLEN_TSTAMP_APPA; 1489 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 1490 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC) 1491 mss -= TCPOLEN_CC_APPA; 1492 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1493 if (mss > MCLBYTES) 1494 mss &= ~(MCLBYTES-1); 1495 #else 1496 if (mss > MCLBYTES) 1497 mss = mss / MCLBYTES * MCLBYTES; 1498 #endif 1499 if (so->so_snd.sb_hiwat < mss) 1500 mss = so->so_snd.sb_hiwat; 1501 1502 tp->t_maxseg = mss; 1503 1504 tcpstat.tcps_mturesent++; 1505 tp->t_rtttime = 0; 1506 tp->snd_nxt = tp->snd_una; 1507 tcp_output(tp); 1508 } 1509 return inp; 1510 } 1511 1512 /* 1513 * Look-up the routing entry to the peer of this inpcb. If no route 1514 * is found and it cannot be allocated, then return NULL. This routine 1515 * is called by TCP routines that access the rmx structure and by tcp_mss 1516 * to get the interface MTU. 1517 */ 1518 u_long 1519 tcp_maxmtu(inc) 1520 struct in_conninfo *inc; 1521 { 1522 struct route sro; 1523 struct sockaddr_in *dst; 1524 struct ifnet *ifp; 1525 u_long maxmtu = 0; 1526 1527 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1528 1529 bzero(&sro, sizeof(sro)); 1530 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1531 dst = (struct sockaddr_in *)&sro.ro_dst; 1532 dst->sin_family = AF_INET; 1533 dst->sin_len = sizeof(*dst); 1534 dst->sin_addr = inc->inc_faddr; 1535 rtalloc_ign(&sro, RTF_CLONING); 1536 } 1537 if (sro.ro_rt != NULL) { 1538 ifp = sro.ro_rt->rt_ifp; 1539 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1540 maxmtu = ifp->if_mtu; 1541 else 1542 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1543 RTFREE(sro.ro_rt); 1544 } 1545 return (maxmtu); 1546 } 1547 1548 #ifdef INET6 1549 u_long 1550 tcp_maxmtu6(inc) 1551 struct in_conninfo *inc; 1552 { 1553 struct route_in6 sro6; 1554 struct ifnet *ifp; 1555 u_long maxmtu = 0; 1556 1557 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1558 1559 bzero(&sro6, sizeof(sro6)); 1560 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1561 sro6.ro_dst.sin6_family = AF_INET6; 1562 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1563 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1564 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1565 } 1566 if (sro6.ro_rt != NULL) { 1567 ifp = sro6.ro_rt->rt_ifp; 1568 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1569 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1570 else 1571 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1572 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1573 RTFREE(sro6.ro_rt); 1574 } 1575 1576 return (maxmtu); 1577 } 1578 #endif /* INET6 */ 1579 1580 #ifdef IPSEC 1581 /* compute ESP/AH header size for TCP, including outer IP header. */ 1582 size_t 1583 ipsec_hdrsiz_tcp(tp) 1584 struct tcpcb *tp; 1585 { 1586 struct inpcb *inp; 1587 struct mbuf *m; 1588 size_t hdrsiz; 1589 struct ip *ip; 1590 #ifdef INET6 1591 struct ip6_hdr *ip6; 1592 #endif 1593 struct tcphdr *th; 1594 1595 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1596 return 0; 1597 MGETHDR(m, M_DONTWAIT, MT_DATA); 1598 if (!m) 1599 return 0; 1600 1601 #ifdef INET6 1602 if ((inp->inp_vflag & INP_IPV6) != 0) { 1603 ip6 = mtod(m, struct ip6_hdr *); 1604 th = (struct tcphdr *)(ip6 + 1); 1605 m->m_pkthdr.len = m->m_len = 1606 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1607 tcpip_fillheaders(inp, ip6, th); 1608 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1609 } else 1610 #endif /* INET6 */ 1611 { 1612 ip = mtod(m, struct ip *); 1613 th = (struct tcphdr *)(ip + 1); 1614 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1615 tcpip_fillheaders(inp, ip, th); 1616 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1617 } 1618 1619 m_free(m); 1620 return hdrsiz; 1621 } 1622 #endif /*IPSEC*/ 1623 1624 /* 1625 * Move a TCP connection into TIME_WAIT state. 1626 * tcbinfo is unlocked. 1627 * inp is locked, and is unlocked before returning. 1628 */ 1629 void 1630 tcp_twstart(tp) 1631 struct tcpcb *tp; 1632 { 1633 struct tcptw *tw; 1634 struct inpcb *inp; 1635 int tw_time, acknow; 1636 struct socket *so; 1637 1638 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1639 if (tw == NULL) { 1640 tw = tcp_timer_2msl_tw(1); 1641 if (tw == NULL) { 1642 tcp_close(tp); 1643 return; 1644 } 1645 } 1646 inp = tp->t_inpcb; 1647 tw->tw_inpcb = inp; 1648 1649 /* 1650 * Recover last window size sent. 1651 */ 1652 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1653 1654 /* 1655 * Set t_recent if timestamps are used on the connection. 1656 */ 1657 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1658 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1659 tw->t_recent = tp->ts_recent; 1660 else 1661 tw->t_recent = 0; 1662 1663 tw->snd_nxt = tp->snd_nxt; 1664 tw->rcv_nxt = tp->rcv_nxt; 1665 tw->iss = tp->iss; 1666 tw->irs = tp->irs; 1667 tw->cc_recv = tp->cc_recv; 1668 tw->cc_send = tp->cc_send; 1669 tw->t_starttime = tp->t_starttime; 1670 tw->tw_time = 0; 1671 1672 /* XXX 1673 * If this code will 1674 * be used for fin-wait-2 state also, then we may need 1675 * a ts_recent from the last segment. 1676 */ 1677 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1678 if (tp->cc_recv != 0 && (ticks - tp->t_starttime) < tcp_msl) { 1679 tw_time = tp->t_rxtcur * TCPTV_TWTRUNC; 1680 /* For T/TCP client, force ACK now. */ 1681 acknow = 1; 1682 } else { 1683 tw_time = 2 * tcp_msl; 1684 acknow = tp->t_flags & TF_ACKNOW; 1685 } 1686 tcp_discardcb(tp); 1687 so = inp->inp_socket; 1688 SOCK_LOCK(so); 1689 so->so_pcb = NULL; 1690 tw->tw_cred = crhold(so->so_cred); 1691 tw->tw_so_options = so->so_options; 1692 sotryfree(so); 1693 inp->inp_socket = NULL; 1694 if (acknow) 1695 tcp_twrespond(tw, TH_ACK); 1696 inp->inp_ppcb = (caddr_t)tw; 1697 inp->inp_vflag |= INP_TIMEWAIT; 1698 tcp_timer_2msl_reset(tw, tw_time); 1699 INP_UNLOCK(inp); 1700 } 1701 1702 /* 1703 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1704 * the actual rate is slightly higher due to the addition of 1705 * random positive increments. 1706 * 1707 * Most other new OSes use semi-randomized ISN values, so we 1708 * do not need to worry about them. 1709 */ 1710 #define MS_ISN_BYTES_PER_SECOND 250000 1711 1712 /* 1713 * Determine if the ISN we will generate has advanced beyond the last 1714 * sequence number used by the previous connection. If so, indicate 1715 * that it is safe to recycle this tw socket by returning 1. 1716 */ 1717 int 1718 tcp_twrecycleable(struct tcptw *tw) 1719 { 1720 tcp_seq new_iss = tw->iss; 1721 tcp_seq new_irs = tw->irs; 1722 1723 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1724 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1725 1726 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1727 return 1; 1728 else 1729 return 0; 1730 } 1731 1732 struct tcptw * 1733 tcp_twclose(struct tcptw *tw, int reuse) 1734 { 1735 struct inpcb *inp; 1736 1737 inp = tw->tw_inpcb; 1738 tw->tw_inpcb = NULL; 1739 tcp_timer_2msl_stop(tw); 1740 inp->inp_ppcb = NULL; 1741 #ifdef INET6 1742 if (inp->inp_vflag & INP_IPV6PROTO) 1743 in6_pcbdetach(inp); 1744 else 1745 #endif 1746 in_pcbdetach(inp); 1747 tcpstat.tcps_closed++; 1748 crfree(tw->tw_cred); 1749 tw->tw_cred = NULL; 1750 if (reuse) 1751 return (tw); 1752 uma_zfree(tcptw_zone, tw); 1753 return (NULL); 1754 } 1755 1756 int 1757 tcp_twrespond(struct tcptw *tw, int flags) 1758 { 1759 struct inpcb *inp = tw->tw_inpcb; 1760 struct tcphdr *th; 1761 struct mbuf *m; 1762 struct ip *ip = NULL; 1763 u_int8_t *optp; 1764 u_int hdrlen, optlen; 1765 int error; 1766 #ifdef INET6 1767 struct ip6_hdr *ip6 = NULL; 1768 int isipv6 = inp->inp_inc.inc_isipv6; 1769 #endif 1770 1771 m = m_gethdr(M_DONTWAIT, MT_HEADER); 1772 if (m == NULL) 1773 return (ENOBUFS); 1774 m->m_data += max_linkhdr; 1775 1776 #ifdef MAC 1777 mac_create_mbuf_from_inpcb(inp, m); 1778 #endif 1779 1780 #ifdef INET6 1781 if (isipv6) { 1782 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1783 ip6 = mtod(m, struct ip6_hdr *); 1784 th = (struct tcphdr *)(ip6 + 1); 1785 tcpip_fillheaders(inp, ip6, th); 1786 } else 1787 #endif 1788 { 1789 hdrlen = sizeof(struct tcpiphdr); 1790 ip = mtod(m, struct ip *); 1791 th = (struct tcphdr *)(ip + 1); 1792 tcpip_fillheaders(inp, ip, th); 1793 } 1794 optp = (u_int8_t *)(th + 1); 1795 1796 /* 1797 * Send a timestamp and echo-reply if both our side and our peer 1798 * have sent timestamps in our SYN's and this is not a RST. 1799 */ 1800 if (tw->t_recent && flags == TH_ACK) { 1801 u_int32_t *lp = (u_int32_t *)optp; 1802 1803 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1804 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1805 *lp++ = htonl(ticks); 1806 *lp = htonl(tw->t_recent); 1807 optp += TCPOLEN_TSTAMP_APPA; 1808 } 1809 1810 /* 1811 * Send `CC-family' options if needed, and it's not a RST. 1812 */ 1813 if (tw->cc_recv != 0 && flags == TH_ACK) { 1814 u_int32_t *lp = (u_int32_t *)optp; 1815 1816 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC)); 1817 *lp = htonl(tw->cc_send); 1818 optp += TCPOLEN_CC_APPA; 1819 } 1820 optlen = optp - (u_int8_t *)(th + 1); 1821 1822 m->m_len = hdrlen + optlen; 1823 m->m_pkthdr.len = m->m_len; 1824 1825 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 1826 1827 th->th_seq = htonl(tw->snd_nxt); 1828 th->th_ack = htonl(tw->rcv_nxt); 1829 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1830 th->th_flags = flags; 1831 th->th_win = htons(tw->last_win); 1832 1833 #ifdef INET6 1834 if (isipv6) { 1835 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 1836 sizeof(struct tcphdr) + optlen); 1837 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 1838 error = ip6_output(m, inp->in6p_outputopts, NULL, 1839 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 1840 } else 1841 #endif 1842 { 1843 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1844 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 1845 m->m_pkthdr.csum_flags = CSUM_TCP; 1846 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1847 ip->ip_len = m->m_pkthdr.len; 1848 if (path_mtu_discovery) 1849 ip->ip_off |= IP_DF; 1850 error = ip_output(m, inp->inp_options, NULL, 1851 (tw->tw_so_options & SO_DONTROUTE), NULL, inp); 1852 } 1853 if (flags & TH_ACK) 1854 tcpstat.tcps_sndacks++; 1855 else 1856 tcpstat.tcps_sndctrl++; 1857 tcpstat.tcps_sndtotal++; 1858 return (error); 1859 } 1860 1861 /* 1862 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1863 * 1864 * This code attempts to calculate the bandwidth-delay product as a 1865 * means of determining the optimal window size to maximize bandwidth, 1866 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1867 * routers. This code also does a fairly good job keeping RTTs in check 1868 * across slow links like modems. We implement an algorithm which is very 1869 * similar (but not meant to be) TCP/Vegas. The code operates on the 1870 * transmitter side of a TCP connection and so only effects the transmit 1871 * side of the connection. 1872 * 1873 * BACKGROUND: TCP makes no provision for the management of buffer space 1874 * at the end points or at the intermediate routers and switches. A TCP 1875 * stream, whether using NewReno or not, will eventually buffer as 1876 * many packets as it is able and the only reason this typically works is 1877 * due to the fairly small default buffers made available for a connection 1878 * (typicaly 16K or 32K). As machines use larger windows and/or window 1879 * scaling it is now fairly easy for even a single TCP connection to blow-out 1880 * all available buffer space not only on the local interface, but on 1881 * intermediate routers and switches as well. NewReno makes a misguided 1882 * attempt to 'solve' this problem by waiting for an actual failure to occur, 1883 * then backing off, then steadily increasing the window again until another 1884 * failure occurs, ad-infinitum. This results in terrible oscillation that 1885 * is only made worse as network loads increase and the idea of intentionally 1886 * blowing out network buffers is, frankly, a terrible way to manage network 1887 * resources. 1888 * 1889 * It is far better to limit the transmit window prior to the failure 1890 * condition being achieved. There are two general ways to do this: First 1891 * you can 'scan' through different transmit window sizes and locate the 1892 * point where the RTT stops increasing, indicating that you have filled the 1893 * pipe, then scan backwards until you note that RTT stops decreasing, then 1894 * repeat ad-infinitum. This method works in principle but has severe 1895 * implementation issues due to RTT variances, timer granularity, and 1896 * instability in the algorithm which can lead to many false positives and 1897 * create oscillations as well as interact badly with other TCP streams 1898 * implementing the same algorithm. 1899 * 1900 * The second method is to limit the window to the bandwidth delay product 1901 * of the link. This is the method we implement. RTT variances and our 1902 * own manipulation of the congestion window, bwnd, can potentially 1903 * destabilize the algorithm. For this reason we have to stabilize the 1904 * elements used to calculate the window. We do this by using the minimum 1905 * observed RTT, the long term average of the observed bandwidth, and 1906 * by adding two segments worth of slop. It isn't perfect but it is able 1907 * to react to changing conditions and gives us a very stable basis on 1908 * which to extend the algorithm. 1909 */ 1910 void 1911 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 1912 { 1913 u_long bw; 1914 u_long bwnd; 1915 int save_ticks; 1916 1917 /* 1918 * If inflight_enable is disabled in the middle of a tcp connection, 1919 * make sure snd_bwnd is effectively disabled. 1920 */ 1921 if (tcp_inflight_enable == 0) { 1922 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1923 tp->snd_bandwidth = 0; 1924 return; 1925 } 1926 1927 /* 1928 * Figure out the bandwidth. Due to the tick granularity this 1929 * is a very rough number and it MUST be averaged over a fairly 1930 * long period of time. XXX we need to take into account a link 1931 * that is not using all available bandwidth, but for now our 1932 * slop will ramp us up if this case occurs and the bandwidth later 1933 * increases. 1934 * 1935 * Note: if ticks rollover 'bw' may wind up negative. We must 1936 * effectively reset t_bw_rtttime for this case. 1937 */ 1938 save_ticks = ticks; 1939 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 1940 return; 1941 1942 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 1943 (save_ticks - tp->t_bw_rtttime); 1944 tp->t_bw_rtttime = save_ticks; 1945 tp->t_bw_rtseq = ack_seq; 1946 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 1947 return; 1948 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 1949 1950 tp->snd_bandwidth = bw; 1951 1952 /* 1953 * Calculate the semi-static bandwidth delay product, plus two maximal 1954 * segments. The additional slop puts us squarely in the sweet 1955 * spot and also handles the bandwidth run-up case and stabilization. 1956 * Without the slop we could be locking ourselves into a lower 1957 * bandwidth. 1958 * 1959 * Situations Handled: 1960 * (1) Prevents over-queueing of packets on LANs, especially on 1961 * high speed LANs, allowing larger TCP buffers to be 1962 * specified, and also does a good job preventing 1963 * over-queueing of packets over choke points like modems 1964 * (at least for the transmit side). 1965 * 1966 * (2) Is able to handle changing network loads (bandwidth 1967 * drops so bwnd drops, bandwidth increases so bwnd 1968 * increases). 1969 * 1970 * (3) Theoretically should stabilize in the face of multiple 1971 * connections implementing the same algorithm (this may need 1972 * a little work). 1973 * 1974 * (4) Stability value (defaults to 20 = 2 maximal packets) can 1975 * be adjusted with a sysctl but typically only needs to be 1976 * on very slow connections. A value no smaller then 5 1977 * should be used, but only reduce this default if you have 1978 * no other choice. 1979 */ 1980 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 1981 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 1982 #undef USERTT 1983 1984 if (tcp_inflight_debug > 0) { 1985 static int ltime; 1986 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 1987 ltime = ticks; 1988 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 1989 tp, 1990 bw, 1991 tp->t_rttbest, 1992 tp->t_srtt, 1993 bwnd 1994 ); 1995 } 1996 } 1997 if ((long)bwnd < tcp_inflight_min) 1998 bwnd = tcp_inflight_min; 1999 if (bwnd > tcp_inflight_max) 2000 bwnd = tcp_inflight_max; 2001 if ((long)bwnd < tp->t_maxseg * 2) 2002 bwnd = tp->t_maxseg * 2; 2003 tp->snd_bwnd = bwnd; 2004 } 2005 2006 #ifdef TCP_SIGNATURE 2007 /* 2008 * Callback function invoked by m_apply() to digest TCP segment data 2009 * contained within an mbuf chain. 2010 */ 2011 static int 2012 tcp_signature_apply(void *fstate, void *data, u_int len) 2013 { 2014 2015 MD5Update(fstate, (u_char *)data, len); 2016 return (0); 2017 } 2018 2019 /* 2020 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2021 * 2022 * Parameters: 2023 * m pointer to head of mbuf chain 2024 * off0 offset to TCP header within the mbuf chain 2025 * len length of TCP segment data, excluding options 2026 * optlen length of TCP segment options 2027 * buf pointer to storage for computed MD5 digest 2028 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2029 * 2030 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2031 * When called from tcp_input(), we can be sure that th_sum has been 2032 * zeroed out and verified already. 2033 * 2034 * This function is for IPv4 use only. Calling this function with an 2035 * IPv6 packet in the mbuf chain will yield undefined results. 2036 * 2037 * Return 0 if successful, otherwise return -1. 2038 * 2039 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2040 * search with the destination IP address, and a 'magic SPI' to be 2041 * determined by the application. This is hardcoded elsewhere to 1179 2042 * right now. Another branch of this code exists which uses the SPD to 2043 * specify per-application flows but it is unstable. 2044 */ 2045 int 2046 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2047 u_char *buf, u_int direction) 2048 { 2049 union sockaddr_union dst; 2050 struct ippseudo ippseudo; 2051 MD5_CTX ctx; 2052 int doff; 2053 struct ip *ip; 2054 struct ipovly *ipovly; 2055 struct secasvar *sav; 2056 struct tcphdr *th; 2057 u_short savecsum; 2058 2059 KASSERT(m != NULL, ("NULL mbuf chain")); 2060 KASSERT(buf != NULL, ("NULL signature pointer")); 2061 2062 /* Extract the destination from the IP header in the mbuf. */ 2063 ip = mtod(m, struct ip *); 2064 bzero(&dst, sizeof(union sockaddr_union)); 2065 dst.sa.sa_len = sizeof(struct sockaddr_in); 2066 dst.sa.sa_family = AF_INET; 2067 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2068 ip->ip_src : ip->ip_dst; 2069 2070 /* Look up an SADB entry which matches the address of the peer. */ 2071 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2072 if (sav == NULL) { 2073 printf("%s: SADB lookup failed for %s\n", __func__, 2074 inet_ntoa(dst.sin.sin_addr)); 2075 return (EINVAL); 2076 } 2077 2078 MD5Init(&ctx); 2079 ipovly = (struct ipovly *)ip; 2080 th = (struct tcphdr *)((u_char *)ip + off0); 2081 doff = off0 + sizeof(struct tcphdr) + optlen; 2082 2083 /* 2084 * Step 1: Update MD5 hash with IP pseudo-header. 2085 * 2086 * XXX The ippseudo header MUST be digested in network byte order, 2087 * or else we'll fail the regression test. Assume all fields we've 2088 * been doing arithmetic on have been in host byte order. 2089 * XXX One cannot depend on ipovly->ih_len here. When called from 2090 * tcp_output(), the underlying ip_len member has not yet been set. 2091 */ 2092 ippseudo.ippseudo_src = ipovly->ih_src; 2093 ippseudo.ippseudo_dst = ipovly->ih_dst; 2094 ippseudo.ippseudo_pad = 0; 2095 ippseudo.ippseudo_p = IPPROTO_TCP; 2096 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2097 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2098 2099 /* 2100 * Step 2: Update MD5 hash with TCP header, excluding options. 2101 * The TCP checksum must be set to zero. 2102 */ 2103 savecsum = th->th_sum; 2104 th->th_sum = 0; 2105 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2106 th->th_sum = savecsum; 2107 2108 /* 2109 * Step 3: Update MD5 hash with TCP segment data. 2110 * Use m_apply() to avoid an early m_pullup(). 2111 */ 2112 if (len > 0) 2113 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2114 2115 /* 2116 * Step 4: Update MD5 hash with shared secret. 2117 */ 2118 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2119 MD5Final(buf, &ctx); 2120 2121 key_sa_recordxfer(sav, m); 2122 KEY_FREESAV(&sav); 2123 return (0); 2124 } 2125 #endif /* TCP_SIGNATURE */ 2126