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/ip_icmp.h> 80 #include <netinet/tcp.h> 81 #include <netinet/tcp_fsm.h> 82 #include <netinet/tcp_seq.h> 83 #include <netinet/tcp_timer.h> 84 #include <netinet/tcp_var.h> 85 #ifdef INET6 86 #include <netinet6/tcp6_var.h> 87 #endif 88 #include <netinet/tcpip.h> 89 #ifdef TCPDEBUG 90 #include <netinet/tcp_debug.h> 91 #endif 92 #include <netinet6/ip6protosw.h> 93 94 #ifdef IPSEC 95 #include <netinet6/ipsec.h> 96 #ifdef INET6 97 #include <netinet6/ipsec6.h> 98 #endif 99 #include <netkey/key.h> 100 #endif /*IPSEC*/ 101 102 #ifdef FAST_IPSEC 103 #include <netipsec/ipsec.h> 104 #include <netipsec/xform.h> 105 #ifdef INET6 106 #include <netipsec/ipsec6.h> 107 #endif 108 #include <netipsec/key.h> 109 #define IPSEC 110 #endif /*FAST_IPSEC*/ 111 112 #include <machine/in_cksum.h> 113 #include <sys/md5.h> 114 115 int tcp_mssdflt = TCP_MSS; 116 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 117 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); 118 119 #ifdef INET6 120 int tcp_v6mssdflt = TCP6_MSS; 121 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 122 CTLFLAG_RW, &tcp_v6mssdflt , 0, 123 "Default TCP Maximum Segment Size for IPv6"); 124 #endif 125 126 /* 127 * Minimum MSS we accept and use. This prevents DoS attacks where 128 * we are forced to a ridiculous low MSS like 20 and send hundreds 129 * of packets instead of one. The effect scales with the available 130 * bandwidth and quickly saturates the CPU and network interface 131 * with packet generation and sending. Set to zero to disable MINMSS 132 * checking. This setting prevents us from sending too small packets. 133 */ 134 int tcp_minmss = TCP_MINMSS; 135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 136 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 137 /* 138 * Number of TCP segments per second we accept from remote host 139 * before we start to calculate average segment size. If average 140 * segment size drops below the minimum TCP MSS we assume a DoS 141 * attack and reset+drop the connection. Care has to be taken not to 142 * set this value too small to not kill interactive type connections 143 * (telnet, SSH) which send many small packets. 144 */ 145 int tcp_minmssoverload = TCP_MINMSSOVERLOAD; 146 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW, 147 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to" 148 "be under the MINMSS Size"); 149 150 #if 0 151 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 152 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 153 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 154 #endif 155 156 int tcp_do_rfc1323 = 1; 157 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 158 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 159 160 static int tcp_tcbhashsize = 0; 161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 162 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 163 164 static int do_tcpdrain = 1; 165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 166 "Enable tcp_drain routine for extra help when low on mbufs"); 167 168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 169 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 170 171 static int icmp_may_rst = 1; 172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 173 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 174 175 static int tcp_isn_reseed_interval = 0; 176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 177 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 178 179 /* 180 * TCP bandwidth limiting sysctls. Note that the default lower bound of 181 * 1024 exists only for debugging. A good production default would be 182 * something like 6100. 183 */ 184 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 185 "TCP inflight data limiting"); 186 187 static int tcp_inflight_enable = 1; 188 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 189 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 190 191 static int tcp_inflight_debug = 0; 192 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 193 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 194 195 static int tcp_inflight_min = 6144; 196 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 197 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 198 199 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 200 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 201 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 202 203 static int tcp_inflight_stab = 20; 204 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 205 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 206 207 uma_zone_t sack_hole_zone; 208 209 static struct inpcb *tcp_notify(struct inpcb *, int); 210 static void tcp_discardcb(struct tcpcb *); 211 static void tcp_isn_tick(void *); 212 213 /* 214 * Target size of TCP PCB hash tables. Must be a power of two. 215 * 216 * Note that this can be overridden by the kernel environment 217 * variable net.inet.tcp.tcbhashsize 218 */ 219 #ifndef TCBHASHSIZE 220 #define TCBHASHSIZE 512 221 #endif 222 223 /* 224 * XXX 225 * Callouts should be moved into struct tcp directly. They are currently 226 * separate because the tcpcb structure is exported to userland for sysctl 227 * parsing purposes, which do not know about callouts. 228 */ 229 struct tcpcb_mem { 230 struct tcpcb tcb; 231 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 232 struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 233 }; 234 235 static uma_zone_t tcpcb_zone; 236 static uma_zone_t tcptw_zone; 237 struct callout isn_callout; 238 239 /* 240 * Tcp initialization 241 */ 242 void 243 tcp_init() 244 { 245 int hashsize = TCBHASHSIZE; 246 247 tcp_delacktime = TCPTV_DELACK; 248 tcp_keepinit = TCPTV_KEEP_INIT; 249 tcp_keepidle = TCPTV_KEEP_IDLE; 250 tcp_keepintvl = TCPTV_KEEPINTVL; 251 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 252 tcp_msl = TCPTV_MSL; 253 tcp_rexmit_min = TCPTV_MIN; 254 tcp_rexmit_slop = TCPTV_CPU_VAR; 255 256 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 257 LIST_INIT(&tcb); 258 tcbinfo.listhead = &tcb; 259 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 260 if (!powerof2(hashsize)) { 261 printf("WARNING: TCB hash size not a power of 2\n"); 262 hashsize = 512; /* safe default */ 263 } 264 tcp_tcbhashsize = hashsize; 265 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 266 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 267 &tcbinfo.porthashmask); 268 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 269 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 270 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 271 #ifdef INET6 272 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 273 #else /* INET6 */ 274 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 275 #endif /* INET6 */ 276 if (max_protohdr < TCP_MINPROTOHDR) 277 max_protohdr = TCP_MINPROTOHDR; 278 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 279 panic("tcp_init"); 280 #undef TCP_MINPROTOHDR 281 /* 282 * These have to be type stable for the benefit of the timers. 283 */ 284 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 285 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 286 uma_zone_set_max(tcpcb_zone, maxsockets); 287 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 288 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 289 uma_zone_set_max(tcptw_zone, maxsockets / 5); 290 tcp_timer_init(); 291 syncache_init(); 292 tcp_hc_init(); 293 tcp_reass_init(); 294 callout_init(&isn_callout, CALLOUT_MPSAFE); 295 tcp_isn_tick(NULL); 296 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 297 SHUTDOWN_PRI_DEFAULT); 298 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 299 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 300 } 301 302 void 303 tcp_fini(xtp) 304 void *xtp; 305 { 306 callout_stop(&isn_callout); 307 308 } 309 310 /* 311 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 312 * tcp_template used to store this data in mbufs, but we now recopy it out 313 * of the tcpcb each time to conserve mbufs. 314 */ 315 void 316 tcpip_fillheaders(inp, ip_ptr, tcp_ptr) 317 struct inpcb *inp; 318 void *ip_ptr; 319 void *tcp_ptr; 320 { 321 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 322 323 INP_LOCK_ASSERT(inp); 324 325 #ifdef INET6 326 if ((inp->inp_vflag & INP_IPV6) != 0) { 327 struct ip6_hdr *ip6; 328 329 ip6 = (struct ip6_hdr *)ip_ptr; 330 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 331 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 332 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 333 (IPV6_VERSION & IPV6_VERSION_MASK); 334 ip6->ip6_nxt = IPPROTO_TCP; 335 ip6->ip6_plen = sizeof(struct tcphdr); 336 ip6->ip6_src = inp->in6p_laddr; 337 ip6->ip6_dst = inp->in6p_faddr; 338 } else 339 #endif 340 { 341 struct ip *ip; 342 343 ip = (struct ip *)ip_ptr; 344 ip->ip_v = IPVERSION; 345 ip->ip_hl = 5; 346 ip->ip_tos = inp->inp_ip_tos; 347 ip->ip_len = 0; 348 ip->ip_id = 0; 349 ip->ip_off = 0; 350 ip->ip_ttl = inp->inp_ip_ttl; 351 ip->ip_sum = 0; 352 ip->ip_p = IPPROTO_TCP; 353 ip->ip_src = inp->inp_laddr; 354 ip->ip_dst = inp->inp_faddr; 355 } 356 th->th_sport = inp->inp_lport; 357 th->th_dport = inp->inp_fport; 358 th->th_seq = 0; 359 th->th_ack = 0; 360 th->th_x2 = 0; 361 th->th_off = 5; 362 th->th_flags = 0; 363 th->th_win = 0; 364 th->th_urp = 0; 365 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 366 } 367 368 /* 369 * Create template to be used to send tcp packets on a connection. 370 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 371 * use for this function is in keepalives, which use tcp_respond. 372 */ 373 struct tcptemp * 374 tcpip_maketemplate(inp) 375 struct inpcb *inp; 376 { 377 struct mbuf *m; 378 struct tcptemp *n; 379 380 m = m_get(M_DONTWAIT, MT_HEADER); 381 if (m == NULL) 382 return (0); 383 m->m_len = sizeof(struct tcptemp); 384 n = mtod(m, struct tcptemp *); 385 386 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 387 return (n); 388 } 389 390 /* 391 * Send a single message to the TCP at address specified by 392 * the given TCP/IP header. If m == NULL, then we make a copy 393 * of the tcpiphdr at ti and send directly to the addressed host. 394 * This is used to force keep alive messages out using the TCP 395 * template for a connection. If flags are given then we send 396 * a message back to the TCP which originated the * segment ti, 397 * and discard the mbuf containing it and any other attached mbufs. 398 * 399 * In any case the ack and sequence number of the transmitted 400 * segment are as specified by the parameters. 401 * 402 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 403 */ 404 void 405 tcp_respond(tp, ipgen, th, m, ack, seq, flags) 406 struct tcpcb *tp; 407 void *ipgen; 408 register struct tcphdr *th; 409 register struct mbuf *m; 410 tcp_seq ack, seq; 411 int flags; 412 { 413 register int tlen; 414 int win = 0; 415 struct ip *ip; 416 struct tcphdr *nth; 417 #ifdef INET6 418 struct ip6_hdr *ip6; 419 int isipv6; 420 #endif /* INET6 */ 421 int ipflags = 0; 422 struct inpcb *inp; 423 424 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 425 426 #ifdef INET6 427 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 428 ip6 = ipgen; 429 #endif /* INET6 */ 430 ip = ipgen; 431 432 if (tp != NULL) { 433 inp = tp->t_inpcb; 434 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 435 INP_INFO_WLOCK_ASSERT(&tcbinfo); 436 INP_LOCK_ASSERT(inp); 437 } else 438 inp = NULL; 439 440 if (tp != NULL) { 441 if (!(flags & TH_RST)) { 442 win = sbspace(&inp->inp_socket->so_rcv); 443 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 444 win = (long)TCP_MAXWIN << tp->rcv_scale; 445 } 446 } 447 if (m == NULL) { 448 m = m_gethdr(M_DONTWAIT, MT_HEADER); 449 if (m == NULL) 450 return; 451 tlen = 0; 452 m->m_data += max_linkhdr; 453 #ifdef INET6 454 if (isipv6) { 455 bcopy((caddr_t)ip6, mtod(m, caddr_t), 456 sizeof(struct ip6_hdr)); 457 ip6 = mtod(m, struct ip6_hdr *); 458 nth = (struct tcphdr *)(ip6 + 1); 459 } else 460 #endif /* INET6 */ 461 { 462 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 463 ip = mtod(m, struct ip *); 464 nth = (struct tcphdr *)(ip + 1); 465 } 466 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 467 flags = TH_ACK; 468 } else { 469 m_freem(m->m_next); 470 m->m_next = NULL; 471 m->m_data = (caddr_t)ipgen; 472 /* m_len is set later */ 473 tlen = 0; 474 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 475 #ifdef INET6 476 if (isipv6) { 477 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 478 nth = (struct tcphdr *)(ip6 + 1); 479 } else 480 #endif /* INET6 */ 481 { 482 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 483 nth = (struct tcphdr *)(ip + 1); 484 } 485 if (th != nth) { 486 /* 487 * this is usually a case when an extension header 488 * exists between the IPv6 header and the 489 * TCP header. 490 */ 491 nth->th_sport = th->th_sport; 492 nth->th_dport = th->th_dport; 493 } 494 xchg(nth->th_dport, nth->th_sport, n_short); 495 #undef xchg 496 } 497 #ifdef INET6 498 if (isipv6) { 499 ip6->ip6_flow = 0; 500 ip6->ip6_vfc = IPV6_VERSION; 501 ip6->ip6_nxt = IPPROTO_TCP; 502 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 503 tlen)); 504 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 505 } else 506 #endif 507 { 508 tlen += sizeof (struct tcpiphdr); 509 ip->ip_len = tlen; 510 ip->ip_ttl = ip_defttl; 511 if (path_mtu_discovery) 512 ip->ip_off |= IP_DF; 513 } 514 m->m_len = tlen; 515 m->m_pkthdr.len = tlen; 516 m->m_pkthdr.rcvif = NULL; 517 #ifdef MAC 518 if (inp != NULL) { 519 /* 520 * Packet is associated with a socket, so allow the 521 * label of the response to reflect the socket label. 522 */ 523 INP_LOCK_ASSERT(inp); 524 mac_create_mbuf_from_inpcb(inp, m); 525 } else { 526 /* 527 * Packet is not associated with a socket, so possibly 528 * update the label in place. 529 */ 530 mac_reflect_mbuf_tcp(m); 531 } 532 #endif 533 nth->th_seq = htonl(seq); 534 nth->th_ack = htonl(ack); 535 nth->th_x2 = 0; 536 nth->th_off = sizeof (struct tcphdr) >> 2; 537 nth->th_flags = flags; 538 if (tp != NULL) 539 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 540 else 541 nth->th_win = htons((u_short)win); 542 nth->th_urp = 0; 543 #ifdef INET6 544 if (isipv6) { 545 nth->th_sum = 0; 546 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 547 sizeof(struct ip6_hdr), 548 tlen - sizeof(struct ip6_hdr)); 549 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 550 NULL, NULL); 551 } else 552 #endif /* INET6 */ 553 { 554 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 555 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 556 m->m_pkthdr.csum_flags = CSUM_TCP; 557 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 558 } 559 #ifdef TCPDEBUG 560 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 561 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 562 #endif 563 #ifdef INET6 564 if (isipv6) 565 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 566 else 567 #endif /* INET6 */ 568 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 569 } 570 571 /* 572 * Create a new TCP control block, making an 573 * empty reassembly queue and hooking it to the argument 574 * protocol control block. The `inp' parameter must have 575 * come from the zone allocator set up in tcp_init(). 576 */ 577 struct tcpcb * 578 tcp_newtcpcb(inp) 579 struct inpcb *inp; 580 { 581 struct tcpcb_mem *tm; 582 struct tcpcb *tp; 583 #ifdef INET6 584 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 585 #endif /* INET6 */ 586 587 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 588 if (tm == NULL) 589 return (NULL); 590 tp = &tm->tcb; 591 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 592 tp->t_maxseg = tp->t_maxopd = 593 #ifdef INET6 594 isipv6 ? tcp_v6mssdflt : 595 #endif /* INET6 */ 596 tcp_mssdflt; 597 598 /* Set up our timeouts. */ 599 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE); 600 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE); 601 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE); 602 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE); 603 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE); 604 605 if (tcp_do_rfc1323) 606 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 607 tp->sack_enable = tcp_do_sack; 608 if (tp->sack_enable) 609 TAILQ_INIT(&tp->snd_holes); 610 tp->t_inpcb = inp; /* XXX */ 611 /* 612 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 613 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 614 * reasonable initial retransmit time. 615 */ 616 tp->t_srtt = TCPTV_SRTTBASE; 617 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 618 tp->t_rttmin = tcp_rexmit_min; 619 tp->t_rxtcur = TCPTV_RTOBASE; 620 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 621 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 622 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 623 tp->t_rcvtime = ticks; 624 tp->t_bw_rtttime = ticks; 625 /* 626 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 627 * because the socket may be bound to an IPv6 wildcard address, 628 * which may match an IPv4-mapped IPv6 address. 629 */ 630 inp->inp_ip_ttl = ip_defttl; 631 inp->inp_ppcb = (caddr_t)tp; 632 return (tp); /* XXX */ 633 } 634 635 /* 636 * Drop a TCP connection, reporting 637 * the specified error. If connection is synchronized, 638 * then send a RST to peer. 639 */ 640 struct tcpcb * 641 tcp_drop(tp, errno) 642 register struct tcpcb *tp; 643 int errno; 644 { 645 struct socket *so = tp->t_inpcb->inp_socket; 646 647 INP_INFO_WLOCK_ASSERT(&tcbinfo); 648 INP_LOCK_ASSERT(tp->t_inpcb); 649 650 if (TCPS_HAVERCVDSYN(tp->t_state)) { 651 tp->t_state = TCPS_CLOSED; 652 (void) tcp_output(tp); 653 tcpstat.tcps_drops++; 654 } else 655 tcpstat.tcps_conndrops++; 656 if (errno == ETIMEDOUT && tp->t_softerror) 657 errno = tp->t_softerror; 658 so->so_error = errno; 659 return (tcp_close(tp)); 660 } 661 662 static void 663 tcp_discardcb(tp) 664 struct tcpcb *tp; 665 { 666 struct tseg_qent *q; 667 struct inpcb *inp = tp->t_inpcb; 668 struct socket *so = inp->inp_socket; 669 #ifdef INET6 670 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 671 #endif /* INET6 */ 672 673 INP_LOCK_ASSERT(inp); 674 675 /* 676 * Make sure that all of our timers are stopped before we 677 * delete the PCB. 678 */ 679 callout_stop(tp->tt_rexmt); 680 callout_stop(tp->tt_persist); 681 callout_stop(tp->tt_keep); 682 callout_stop(tp->tt_2msl); 683 callout_stop(tp->tt_delack); 684 685 /* 686 * If we got enough samples through the srtt filter, 687 * save the rtt and rttvar in the routing entry. 688 * 'Enough' is arbitrarily defined as 4 rtt samples. 689 * 4 samples is enough for the srtt filter to converge 690 * to within enough % of the correct value; fewer samples 691 * and we could save a bogus rtt. The danger is not high 692 * as tcp quickly recovers from everything. 693 * XXX: Works very well but needs some more statistics! 694 */ 695 if (tp->t_rttupdated >= 4) { 696 struct hc_metrics_lite metrics; 697 u_long ssthresh; 698 699 bzero(&metrics, sizeof(metrics)); 700 /* 701 * Update the ssthresh always when the conditions below 702 * are satisfied. This gives us better new start value 703 * for the congestion avoidance for new connections. 704 * ssthresh is only set if packet loss occured on a session. 705 */ 706 ssthresh = tp->snd_ssthresh; 707 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 708 /* 709 * convert the limit from user data bytes to 710 * packets then to packet data bytes. 711 */ 712 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 713 if (ssthresh < 2) 714 ssthresh = 2; 715 ssthresh *= (u_long)(tp->t_maxseg + 716 #ifdef INET6 717 (isipv6 ? sizeof (struct ip6_hdr) + 718 sizeof (struct tcphdr) : 719 #endif 720 sizeof (struct tcpiphdr) 721 #ifdef INET6 722 ) 723 #endif 724 ); 725 } else 726 ssthresh = 0; 727 metrics.rmx_ssthresh = ssthresh; 728 729 metrics.rmx_rtt = tp->t_srtt; 730 metrics.rmx_rttvar = tp->t_rttvar; 731 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 732 metrics.rmx_bandwidth = tp->snd_bandwidth; 733 metrics.rmx_cwnd = tp->snd_cwnd; 734 metrics.rmx_sendpipe = 0; 735 metrics.rmx_recvpipe = 0; 736 737 tcp_hc_update(&inp->inp_inc, &metrics); 738 } 739 740 /* free the reassembly queue, if any */ 741 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 742 LIST_REMOVE(q, tqe_q); 743 m_freem(q->tqe_m); 744 uma_zfree(tcp_reass_zone, q); 745 tp->t_segqlen--; 746 tcp_reass_qsize--; 747 } 748 tcp_free_sackholes(tp); 749 inp->inp_ppcb = NULL; 750 tp->t_inpcb = NULL; 751 uma_zfree(tcpcb_zone, tp); 752 soisdisconnected(so); 753 } 754 755 /* 756 * Close a TCP control block: 757 * discard all space held by the tcp 758 * discard internet protocol block 759 * wake up any sleepers 760 */ 761 struct tcpcb * 762 tcp_close(tp) 763 struct tcpcb *tp; 764 { 765 struct inpcb *inp = tp->t_inpcb; 766 #ifdef INET6 767 struct socket *so = inp->inp_socket; 768 #endif 769 770 INP_INFO_WLOCK_ASSERT(&tcbinfo); 771 INP_LOCK_ASSERT(inp); 772 773 tcp_discardcb(tp); 774 #ifdef INET6 775 if (INP_CHECK_SOCKAF(so, AF_INET6)) 776 in6_pcbdetach(inp); 777 else 778 #endif 779 in_pcbdetach(inp); 780 tcpstat.tcps_closed++; 781 return (NULL); 782 } 783 784 void 785 tcp_drain() 786 { 787 if (do_tcpdrain) 788 { 789 struct inpcb *inpb; 790 struct tcpcb *tcpb; 791 struct tseg_qent *te; 792 793 /* 794 * Walk the tcpbs, if existing, and flush the reassembly queue, 795 * if there is one... 796 * XXX: The "Net/3" implementation doesn't imply that the TCP 797 * reassembly queue should be flushed, but in a situation 798 * where we're really low on mbufs, this is potentially 799 * usefull. 800 */ 801 INP_INFO_RLOCK(&tcbinfo); 802 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 803 if (inpb->inp_vflag & INP_TIMEWAIT) 804 continue; 805 INP_LOCK(inpb); 806 if ((tcpb = intotcpcb(inpb)) != NULL) { 807 while ((te = LIST_FIRST(&tcpb->t_segq)) 808 != NULL) { 809 LIST_REMOVE(te, tqe_q); 810 m_freem(te->tqe_m); 811 uma_zfree(tcp_reass_zone, te); 812 tcpb->t_segqlen--; 813 tcp_reass_qsize--; 814 } 815 tcp_clean_sackreport(tcpb); 816 } 817 INP_UNLOCK(inpb); 818 } 819 INP_INFO_RUNLOCK(&tcbinfo); 820 } 821 } 822 823 /* 824 * Notify a tcp user of an asynchronous error; 825 * store error as soft error, but wake up user 826 * (for now, won't do anything until can select for soft error). 827 * 828 * Do not wake up user since there currently is no mechanism for 829 * reporting soft errors (yet - a kqueue filter may be added). 830 */ 831 static struct inpcb * 832 tcp_notify(inp, error) 833 struct inpcb *inp; 834 int error; 835 { 836 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 837 838 INP_INFO_WLOCK_ASSERT(&tcbinfo); 839 INP_LOCK_ASSERT(inp); 840 841 /* 842 * Ignore some errors if we are hooked up. 843 * If connection hasn't completed, has retransmitted several times, 844 * and receives a second error, give up now. This is better 845 * than waiting a long time to establish a connection that 846 * can never complete. 847 */ 848 if (tp->t_state == TCPS_ESTABLISHED && 849 (error == EHOSTUNREACH || error == ENETUNREACH || 850 error == EHOSTDOWN)) { 851 return (inp); 852 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 853 tp->t_softerror) { 854 tcp_drop(tp, error); 855 return (struct inpcb *)0; 856 } else { 857 tp->t_softerror = error; 858 return (inp); 859 } 860 #if 0 861 wakeup( &so->so_timeo); 862 sorwakeup(so); 863 sowwakeup(so); 864 #endif 865 } 866 867 static int 868 tcp_pcblist(SYSCTL_HANDLER_ARGS) 869 { 870 int error, i, n, s; 871 struct inpcb *inp, **inp_list; 872 inp_gen_t gencnt; 873 struct xinpgen xig; 874 875 /* 876 * The process of preparing the TCB list is too time-consuming and 877 * resource-intensive to repeat twice on every request. 878 */ 879 if (req->oldptr == NULL) { 880 n = tcbinfo.ipi_count; 881 req->oldidx = 2 * (sizeof xig) 882 + (n + n/8) * sizeof(struct xtcpcb); 883 return (0); 884 } 885 886 if (req->newptr != NULL) 887 return (EPERM); 888 889 /* 890 * OK, now we're committed to doing something. 891 */ 892 s = splnet(); 893 INP_INFO_RLOCK(&tcbinfo); 894 gencnt = tcbinfo.ipi_gencnt; 895 n = tcbinfo.ipi_count; 896 INP_INFO_RUNLOCK(&tcbinfo); 897 splx(s); 898 899 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 900 + n * sizeof(struct xtcpcb)); 901 if (error != 0) 902 return (error); 903 904 xig.xig_len = sizeof xig; 905 xig.xig_count = n; 906 xig.xig_gen = gencnt; 907 xig.xig_sogen = so_gencnt; 908 error = SYSCTL_OUT(req, &xig, sizeof xig); 909 if (error) 910 return (error); 911 912 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 913 if (inp_list == NULL) 914 return (ENOMEM); 915 916 s = splnet(); 917 INP_INFO_RLOCK(&tcbinfo); 918 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 919 inp = LIST_NEXT(inp, inp_list)) { 920 INP_LOCK(inp); 921 if (inp->inp_gencnt <= gencnt) { 922 /* 923 * XXX: This use of cr_cansee(), introduced with 924 * TCP state changes, is not quite right, but for 925 * now, better than nothing. 926 */ 927 if (inp->inp_vflag & INP_TIMEWAIT) 928 error = cr_cansee(req->td->td_ucred, 929 intotw(inp)->tw_cred); 930 else 931 error = cr_canseesocket(req->td->td_ucred, 932 inp->inp_socket); 933 if (error == 0) 934 inp_list[i++] = inp; 935 } 936 INP_UNLOCK(inp); 937 } 938 INP_INFO_RUNLOCK(&tcbinfo); 939 splx(s); 940 n = i; 941 942 error = 0; 943 for (i = 0; i < n; i++) { 944 inp = inp_list[i]; 945 if (inp->inp_gencnt <= gencnt) { 946 struct xtcpcb xt; 947 caddr_t inp_ppcb; 948 949 bzero(&xt, sizeof(xt)); 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 in6_addr a6[2]; 1047 struct inpcb *inp; 1048 int error, s, mapped = 0; 1049 1050 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1051 if (error) 1052 return (error); 1053 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1054 if (error) 1055 return (error); 1056 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1057 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1058 mapped = 1; 1059 else 1060 return (EINVAL); 1061 } else { 1062 error = in6_embedscope(&a6[0], &addrs[0], NULL, NULL); 1063 if (error) 1064 return (EINVAL); 1065 error = in6_embedscope(&a6[1], &addrs[1], NULL, NULL); 1066 if (error) 1067 return (EINVAL); 1068 } 1069 s = splnet(); 1070 INP_INFO_RLOCK(&tcbinfo); 1071 if (mapped == 1) 1072 inp = in_pcblookup_hash(&tcbinfo, 1073 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1074 addrs[1].sin6_port, 1075 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1076 addrs[0].sin6_port, 1077 0, NULL); 1078 else 1079 inp = in6_pcblookup_hash(&tcbinfo, &a6[1], addrs[1].sin6_port, 1080 &a6[0], addrs[0].sin6_port, 0, NULL); 1081 if (inp == NULL) { 1082 error = ENOENT; 1083 goto outunlocked; 1084 } 1085 INP_LOCK(inp); 1086 if (inp->inp_socket == NULL) { 1087 error = ENOENT; 1088 goto out; 1089 } 1090 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1091 if (error) 1092 goto out; 1093 cru2x(inp->inp_socket->so_cred, &xuc); 1094 out: 1095 INP_UNLOCK(inp); 1096 outunlocked: 1097 INP_INFO_RUNLOCK(&tcbinfo); 1098 splx(s); 1099 if (error == 0) 1100 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1101 return (error); 1102 } 1103 1104 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1105 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1106 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1107 #endif 1108 1109 1110 void 1111 tcp_ctlinput(cmd, sa, vip) 1112 int cmd; 1113 struct sockaddr *sa; 1114 void *vip; 1115 { 1116 struct ip *ip = vip; 1117 struct tcphdr *th; 1118 struct in_addr faddr; 1119 struct inpcb *inp; 1120 struct tcpcb *tp; 1121 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1122 struct icmp *icp; 1123 struct in_conninfo inc; 1124 tcp_seq icmp_tcp_seq; 1125 int mtu, s; 1126 1127 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1128 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1129 return; 1130 1131 if (cmd == PRC_MSGSIZE) 1132 notify = tcp_mtudisc; 1133 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1134 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1135 notify = tcp_drop_syn_sent; 1136 /* 1137 * Redirects don't need to be handled up here. 1138 */ 1139 else if (PRC_IS_REDIRECT(cmd)) 1140 return; 1141 /* 1142 * Source quench is depreciated. 1143 */ 1144 else if (cmd == PRC_QUENCH) 1145 return; 1146 /* 1147 * Hostdead is ugly because it goes linearly through all PCBs. 1148 * XXX: We never get this from ICMP, otherwise it makes an 1149 * excellent DoS attack on machines with many connections. 1150 */ 1151 else if (cmd == PRC_HOSTDEAD) 1152 ip = NULL; 1153 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1154 return; 1155 if (ip != NULL) { 1156 s = splnet(); 1157 icp = (struct icmp *)((caddr_t)ip 1158 - offsetof(struct icmp, icmp_ip)); 1159 th = (struct tcphdr *)((caddr_t)ip 1160 + (ip->ip_hl << 2)); 1161 INP_INFO_WLOCK(&tcbinfo); 1162 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1163 ip->ip_src, th->th_sport, 0, NULL); 1164 if (inp != NULL) { 1165 INP_LOCK(inp); 1166 if (inp->inp_socket != NULL) { 1167 icmp_tcp_seq = htonl(th->th_seq); 1168 tp = intotcpcb(inp); 1169 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1170 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1171 if (cmd == PRC_MSGSIZE) { 1172 /* 1173 * MTU discovery: 1174 * If we got a needfrag set the MTU 1175 * in the route to the suggested new 1176 * value (if given) and then notify. 1177 */ 1178 bzero(&inc, sizeof(inc)); 1179 inc.inc_flags = 0; /* IPv4 */ 1180 inc.inc_faddr = faddr; 1181 1182 mtu = ntohs(icp->icmp_nextmtu); 1183 /* 1184 * If no alternative MTU was 1185 * proposed, try the next smaller 1186 * one. 1187 */ 1188 if (!mtu) 1189 mtu = ip_next_mtu(ntohs(ip->ip_len), 1190 1); 1191 if (mtu < max(296, (tcp_minmss) 1192 + sizeof(struct tcpiphdr))) 1193 mtu = 0; 1194 if (!mtu) 1195 mtu = tcp_mssdflt 1196 + sizeof(struct tcpiphdr); 1197 /* 1198 * Only cache the the MTU if it 1199 * is smaller than the interface 1200 * or route MTU. tcp_mtudisc() 1201 * will do right thing by itself. 1202 */ 1203 if (mtu <= tcp_maxmtu(&inc)) 1204 tcp_hc_updatemtu(&inc, mtu); 1205 } 1206 1207 inp = (*notify)(inp, inetctlerrmap[cmd]); 1208 } 1209 } 1210 if (inp != NULL) 1211 INP_UNLOCK(inp); 1212 } else { 1213 inc.inc_fport = th->th_dport; 1214 inc.inc_lport = th->th_sport; 1215 inc.inc_faddr = faddr; 1216 inc.inc_laddr = ip->ip_src; 1217 #ifdef INET6 1218 inc.inc_isipv6 = 0; 1219 #endif 1220 syncache_unreach(&inc, th); 1221 } 1222 INP_INFO_WUNLOCK(&tcbinfo); 1223 splx(s); 1224 } else 1225 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1226 } 1227 1228 #ifdef INET6 1229 void 1230 tcp6_ctlinput(cmd, sa, d) 1231 int cmd; 1232 struct sockaddr *sa; 1233 void *d; 1234 { 1235 struct tcphdr th; 1236 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1237 struct ip6_hdr *ip6; 1238 struct mbuf *m; 1239 struct ip6ctlparam *ip6cp = NULL; 1240 const struct sockaddr_in6 *sa6_src = NULL; 1241 int off; 1242 struct tcp_portonly { 1243 u_int16_t th_sport; 1244 u_int16_t th_dport; 1245 } *thp; 1246 1247 if (sa->sa_family != AF_INET6 || 1248 sa->sa_len != sizeof(struct sockaddr_in6)) 1249 return; 1250 1251 if (cmd == PRC_MSGSIZE) 1252 notify = tcp_mtudisc; 1253 else if (!PRC_IS_REDIRECT(cmd) && 1254 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1255 return; 1256 /* Source quench is depreciated. */ 1257 else if (cmd == PRC_QUENCH) 1258 return; 1259 1260 /* if the parameter is from icmp6, decode it. */ 1261 if (d != NULL) { 1262 ip6cp = (struct ip6ctlparam *)d; 1263 m = ip6cp->ip6c_m; 1264 ip6 = ip6cp->ip6c_ip6; 1265 off = ip6cp->ip6c_off; 1266 sa6_src = ip6cp->ip6c_src; 1267 } else { 1268 m = NULL; 1269 ip6 = NULL; 1270 off = 0; /* fool gcc */ 1271 sa6_src = &sa6_any; 1272 } 1273 1274 if (ip6 != NULL) { 1275 struct in_conninfo inc; 1276 /* 1277 * XXX: We assume that when IPV6 is non NULL, 1278 * M and OFF are valid. 1279 */ 1280 1281 /* check if we can safely examine src and dst ports */ 1282 if (m->m_pkthdr.len < off + sizeof(*thp)) 1283 return; 1284 1285 bzero(&th, sizeof(th)); 1286 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1287 1288 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1289 (struct sockaddr *)ip6cp->ip6c_src, 1290 th.th_sport, cmd, NULL, notify); 1291 1292 inc.inc_fport = th.th_dport; 1293 inc.inc_lport = th.th_sport; 1294 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1295 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1296 inc.inc_isipv6 = 1; 1297 INP_INFO_WLOCK(&tcbinfo); 1298 syncache_unreach(&inc, &th); 1299 INP_INFO_WUNLOCK(&tcbinfo); 1300 } else 1301 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1302 0, cmd, NULL, notify); 1303 } 1304 #endif /* INET6 */ 1305 1306 1307 /* 1308 * Following is where TCP initial sequence number generation occurs. 1309 * 1310 * There are two places where we must use initial sequence numbers: 1311 * 1. In SYN-ACK packets. 1312 * 2. In SYN packets. 1313 * 1314 * All ISNs for SYN-ACK packets are generated by the syncache. See 1315 * tcp_syncache.c for details. 1316 * 1317 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1318 * depends on this property. In addition, these ISNs should be 1319 * unguessable so as to prevent connection hijacking. To satisfy 1320 * the requirements of this situation, the algorithm outlined in 1321 * RFC 1948 is used, with only small modifications. 1322 * 1323 * Implementation details: 1324 * 1325 * Time is based off the system timer, and is corrected so that it 1326 * increases by one megabyte per second. This allows for proper 1327 * recycling on high speed LANs while still leaving over an hour 1328 * before rollover. 1329 * 1330 * As reading the *exact* system time is too expensive to be done 1331 * whenever setting up a TCP connection, we increment the time 1332 * offset in two ways. First, a small random positive increment 1333 * is added to isn_offset for each connection that is set up. 1334 * Second, the function tcp_isn_tick fires once per clock tick 1335 * and increments isn_offset as necessary so that sequence numbers 1336 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1337 * random positive increments serve only to ensure that the same 1338 * exact sequence number is never sent out twice (as could otherwise 1339 * happen when a port is recycled in less than the system tick 1340 * interval.) 1341 * 1342 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1343 * between seeding of isn_secret. This is normally set to zero, 1344 * as reseeding should not be necessary. 1345 * 1346 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1347 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1348 * general, this means holding an exclusive (write) lock. 1349 */ 1350 1351 #define ISN_BYTES_PER_SECOND 1048576 1352 #define ISN_STATIC_INCREMENT 4096 1353 #define ISN_RANDOM_INCREMENT (4096 - 1) 1354 1355 static u_char isn_secret[32]; 1356 static int isn_last_reseed; 1357 static u_int32_t isn_offset, isn_offset_old; 1358 static MD5_CTX isn_ctx; 1359 1360 tcp_seq 1361 tcp_new_isn(tp) 1362 struct tcpcb *tp; 1363 { 1364 u_int32_t md5_buffer[4]; 1365 tcp_seq new_isn; 1366 1367 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1368 INP_LOCK_ASSERT(tp->t_inpcb); 1369 1370 /* Seed if this is the first use, reseed if requested. */ 1371 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1372 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1373 < (u_int)ticks))) { 1374 read_random(&isn_secret, sizeof(isn_secret)); 1375 isn_last_reseed = ticks; 1376 } 1377 1378 /* Compute the md5 hash and return the ISN. */ 1379 MD5Init(&isn_ctx); 1380 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1381 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1382 #ifdef INET6 1383 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1384 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1385 sizeof(struct in6_addr)); 1386 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1387 sizeof(struct in6_addr)); 1388 } else 1389 #endif 1390 { 1391 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1392 sizeof(struct in_addr)); 1393 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1394 sizeof(struct in_addr)); 1395 } 1396 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1397 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1398 new_isn = (tcp_seq) md5_buffer[0]; 1399 isn_offset += ISN_STATIC_INCREMENT + 1400 (arc4random() & ISN_RANDOM_INCREMENT); 1401 new_isn += isn_offset; 1402 return (new_isn); 1403 } 1404 1405 /* 1406 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1407 * to keep time flowing at a relatively constant rate. If the random 1408 * increments have already pushed us past the projected offset, do nothing. 1409 */ 1410 static void 1411 tcp_isn_tick(xtp) 1412 void *xtp; 1413 { 1414 u_int32_t projected_offset; 1415 1416 INP_INFO_WLOCK(&tcbinfo); 1417 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1418 1419 if (projected_offset > isn_offset) 1420 isn_offset = projected_offset; 1421 1422 isn_offset_old = isn_offset; 1423 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1424 INP_INFO_WUNLOCK(&tcbinfo); 1425 } 1426 1427 /* 1428 * When a specific ICMP unreachable message is received and the 1429 * connection state is SYN-SENT, drop the connection. This behavior 1430 * is controlled by the icmp_may_rst sysctl. 1431 */ 1432 struct inpcb * 1433 tcp_drop_syn_sent(inp, errno) 1434 struct inpcb *inp; 1435 int errno; 1436 { 1437 struct tcpcb *tp = intotcpcb(inp); 1438 1439 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1440 INP_LOCK_ASSERT(inp); 1441 1442 if (tp != NULL && tp->t_state == TCPS_SYN_SENT) { 1443 tcp_drop(tp, errno); 1444 return (NULL); 1445 } 1446 return (inp); 1447 } 1448 1449 /* 1450 * When `need fragmentation' ICMP is received, update our idea of the MSS 1451 * based on the new value in the route. Also nudge TCP to send something, 1452 * since we know the packet we just sent was dropped. 1453 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1454 */ 1455 struct inpcb * 1456 tcp_mtudisc(inp, errno) 1457 struct inpcb *inp; 1458 int errno; 1459 { 1460 struct tcpcb *tp = intotcpcb(inp); 1461 struct socket *so = inp->inp_socket; 1462 u_int maxmtu; 1463 u_int romtu; 1464 int mss; 1465 #ifdef INET6 1466 int isipv6; 1467 #endif /* INET6 */ 1468 1469 INP_LOCK_ASSERT(inp); 1470 if (tp != NULL) { 1471 #ifdef INET6 1472 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1473 #endif 1474 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1475 romtu = 1476 #ifdef INET6 1477 isipv6 ? tcp_maxmtu6(&inp->inp_inc) : 1478 #endif /* INET6 */ 1479 tcp_maxmtu(&inp->inp_inc); 1480 if (!maxmtu) 1481 maxmtu = romtu; 1482 else 1483 maxmtu = min(maxmtu, romtu); 1484 if (!maxmtu) { 1485 tp->t_maxopd = tp->t_maxseg = 1486 #ifdef INET6 1487 isipv6 ? tcp_v6mssdflt : 1488 #endif /* INET6 */ 1489 tcp_mssdflt; 1490 return (inp); 1491 } 1492 mss = maxmtu - 1493 #ifdef INET6 1494 (isipv6 ? 1495 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1496 #endif /* INET6 */ 1497 sizeof(struct tcpiphdr) 1498 #ifdef INET6 1499 ) 1500 #endif /* INET6 */ 1501 ; 1502 1503 /* 1504 * XXX - The above conditional probably violates the TCP 1505 * spec. The problem is that, since we don't know the 1506 * other end's MSS, we are supposed to use a conservative 1507 * default. But, if we do that, then MTU discovery will 1508 * never actually take place, because the conservative 1509 * default is much less than the MTUs typically seen 1510 * on the Internet today. For the moment, we'll sweep 1511 * this under the carpet. 1512 * 1513 * The conservative default might not actually be a problem 1514 * if the only case this occurs is when sending an initial 1515 * SYN with options and data to a host we've never talked 1516 * to before. Then, they will reply with an MSS value which 1517 * will get recorded and the new parameters should get 1518 * recomputed. For Further Study. 1519 */ 1520 if (tp->t_maxopd <= mss) 1521 return (inp); 1522 tp->t_maxopd = mss; 1523 1524 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1525 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1526 mss -= TCPOLEN_TSTAMP_APPA; 1527 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1528 if (mss > MCLBYTES) 1529 mss &= ~(MCLBYTES-1); 1530 #else 1531 if (mss > MCLBYTES) 1532 mss = mss / MCLBYTES * MCLBYTES; 1533 #endif 1534 if (so->so_snd.sb_hiwat < mss) 1535 mss = so->so_snd.sb_hiwat; 1536 1537 tp->t_maxseg = mss; 1538 1539 tcpstat.tcps_mturesent++; 1540 tp->t_rtttime = 0; 1541 tp->snd_nxt = tp->snd_una; 1542 tcp_output(tp); 1543 } 1544 return (inp); 1545 } 1546 1547 /* 1548 * Look-up the routing entry to the peer of this inpcb. If no route 1549 * is found and it cannot be allocated, then return NULL. This routine 1550 * is called by TCP routines that access the rmx structure and by tcp_mss 1551 * to get the interface MTU. 1552 */ 1553 u_long 1554 tcp_maxmtu(inc) 1555 struct in_conninfo *inc; 1556 { 1557 struct route sro; 1558 struct sockaddr_in *dst; 1559 struct ifnet *ifp; 1560 u_long maxmtu = 0; 1561 1562 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1563 1564 bzero(&sro, sizeof(sro)); 1565 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1566 dst = (struct sockaddr_in *)&sro.ro_dst; 1567 dst->sin_family = AF_INET; 1568 dst->sin_len = sizeof(*dst); 1569 dst->sin_addr = inc->inc_faddr; 1570 rtalloc_ign(&sro, RTF_CLONING); 1571 } 1572 if (sro.ro_rt != NULL) { 1573 ifp = sro.ro_rt->rt_ifp; 1574 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1575 maxmtu = ifp->if_mtu; 1576 else 1577 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1578 RTFREE(sro.ro_rt); 1579 } 1580 return (maxmtu); 1581 } 1582 1583 #ifdef INET6 1584 u_long 1585 tcp_maxmtu6(inc) 1586 struct in_conninfo *inc; 1587 { 1588 struct route_in6 sro6; 1589 struct ifnet *ifp; 1590 u_long maxmtu = 0; 1591 1592 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1593 1594 bzero(&sro6, sizeof(sro6)); 1595 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1596 sro6.ro_dst.sin6_family = AF_INET6; 1597 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1598 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1599 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1600 } 1601 if (sro6.ro_rt != NULL) { 1602 ifp = sro6.ro_rt->rt_ifp; 1603 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1604 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1605 else 1606 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1607 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1608 RTFREE(sro6.ro_rt); 1609 } 1610 1611 return (maxmtu); 1612 } 1613 #endif /* INET6 */ 1614 1615 #ifdef IPSEC 1616 /* compute ESP/AH header size for TCP, including outer IP header. */ 1617 size_t 1618 ipsec_hdrsiz_tcp(tp) 1619 struct tcpcb *tp; 1620 { 1621 struct inpcb *inp; 1622 struct mbuf *m; 1623 size_t hdrsiz; 1624 struct ip *ip; 1625 #ifdef INET6 1626 struct ip6_hdr *ip6; 1627 #endif 1628 struct tcphdr *th; 1629 1630 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1631 return (0); 1632 MGETHDR(m, M_DONTWAIT, MT_DATA); 1633 if (!m) 1634 return (0); 1635 1636 #ifdef INET6 1637 if ((inp->inp_vflag & INP_IPV6) != 0) { 1638 ip6 = mtod(m, struct ip6_hdr *); 1639 th = (struct tcphdr *)(ip6 + 1); 1640 m->m_pkthdr.len = m->m_len = 1641 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1642 tcpip_fillheaders(inp, ip6, th); 1643 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1644 } else 1645 #endif /* INET6 */ 1646 { 1647 ip = mtod(m, struct ip *); 1648 th = (struct tcphdr *)(ip + 1); 1649 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1650 tcpip_fillheaders(inp, ip, th); 1651 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1652 } 1653 1654 m_free(m); 1655 return (hdrsiz); 1656 } 1657 #endif /*IPSEC*/ 1658 1659 /* 1660 * Move a TCP connection into TIME_WAIT state. 1661 * tcbinfo is locked. 1662 * inp is locked, and is unlocked before returning. 1663 */ 1664 void 1665 tcp_twstart(tp) 1666 struct tcpcb *tp; 1667 { 1668 struct tcptw *tw; 1669 struct inpcb *inp; 1670 int tw_time, acknow; 1671 struct socket *so; 1672 1673 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1674 INP_LOCK_ASSERT(tp->t_inpcb); 1675 1676 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1677 if (tw == NULL) { 1678 tw = tcp_timer_2msl_tw(1); 1679 if (tw == NULL) { 1680 tcp_close(tp); 1681 return; 1682 } 1683 } 1684 inp = tp->t_inpcb; 1685 tw->tw_inpcb = inp; 1686 1687 /* 1688 * Recover last window size sent. 1689 */ 1690 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1691 1692 /* 1693 * Set t_recent if timestamps are used on the connection. 1694 */ 1695 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1696 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1697 tw->t_recent = tp->ts_recent; 1698 else 1699 tw->t_recent = 0; 1700 1701 tw->snd_nxt = tp->snd_nxt; 1702 tw->rcv_nxt = tp->rcv_nxt; 1703 tw->iss = tp->iss; 1704 tw->irs = tp->irs; 1705 tw->t_starttime = tp->t_starttime; 1706 tw->tw_time = 0; 1707 1708 /* XXX 1709 * If this code will 1710 * be used for fin-wait-2 state also, then we may need 1711 * a ts_recent from the last segment. 1712 */ 1713 tw_time = 2 * tcp_msl; 1714 acknow = tp->t_flags & TF_ACKNOW; 1715 tcp_discardcb(tp); 1716 so = inp->inp_socket; 1717 ACCEPT_LOCK(); 1718 SOCK_LOCK(so); 1719 so->so_pcb = NULL; 1720 tw->tw_cred = crhold(so->so_cred); 1721 tw->tw_so_options = so->so_options; 1722 sotryfree(so); 1723 inp->inp_socket = NULL; 1724 if (acknow) 1725 tcp_twrespond(tw, TH_ACK); 1726 inp->inp_ppcb = (caddr_t)tw; 1727 inp->inp_vflag |= INP_TIMEWAIT; 1728 tcp_timer_2msl_reset(tw, tw_time); 1729 INP_UNLOCK(inp); 1730 } 1731 1732 /* 1733 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1734 * the actual rate is slightly higher due to the addition of 1735 * random positive increments. 1736 * 1737 * Most other new OSes use semi-randomized ISN values, so we 1738 * do not need to worry about them. 1739 */ 1740 #define MS_ISN_BYTES_PER_SECOND 250000 1741 1742 /* 1743 * Determine if the ISN we will generate has advanced beyond the last 1744 * sequence number used by the previous connection. If so, indicate 1745 * that it is safe to recycle this tw socket by returning 1. 1746 * 1747 * XXXRW: This function should assert the inpcb lock as it does multiple 1748 * non-atomic reads from the tcptw, but is currently called without it from 1749 * in_pcb.c:in_pcblookup_local(). 1750 */ 1751 int 1752 tcp_twrecycleable(struct tcptw *tw) 1753 { 1754 tcp_seq new_iss = tw->iss; 1755 tcp_seq new_irs = tw->irs; 1756 1757 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1758 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1759 1760 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1761 return (1); 1762 else 1763 return (0); 1764 } 1765 1766 struct tcptw * 1767 tcp_twclose(struct tcptw *tw, int reuse) 1768 { 1769 struct inpcb *inp; 1770 1771 inp = tw->tw_inpcb; 1772 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1773 INP_LOCK_ASSERT(inp); 1774 1775 tw->tw_inpcb = NULL; 1776 tcp_timer_2msl_stop(tw); 1777 inp->inp_ppcb = NULL; 1778 #ifdef INET6 1779 if (inp->inp_vflag & INP_IPV6PROTO) 1780 in6_pcbdetach(inp); 1781 else 1782 #endif 1783 in_pcbdetach(inp); 1784 tcpstat.tcps_closed++; 1785 crfree(tw->tw_cred); 1786 tw->tw_cred = NULL; 1787 if (reuse) 1788 return (tw); 1789 uma_zfree(tcptw_zone, tw); 1790 return (NULL); 1791 } 1792 1793 int 1794 tcp_twrespond(struct tcptw *tw, int flags) 1795 { 1796 struct inpcb *inp = tw->tw_inpcb; 1797 struct tcphdr *th; 1798 struct mbuf *m; 1799 struct ip *ip = NULL; 1800 u_int8_t *optp; 1801 u_int hdrlen, optlen; 1802 int error; 1803 #ifdef INET6 1804 struct ip6_hdr *ip6 = NULL; 1805 int isipv6 = inp->inp_inc.inc_isipv6; 1806 #endif 1807 1808 INP_LOCK_ASSERT(inp); 1809 1810 m = m_gethdr(M_DONTWAIT, MT_HEADER); 1811 if (m == NULL) 1812 return (ENOBUFS); 1813 m->m_data += max_linkhdr; 1814 1815 #ifdef MAC 1816 mac_create_mbuf_from_inpcb(inp, m); 1817 #endif 1818 1819 #ifdef INET6 1820 if (isipv6) { 1821 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1822 ip6 = mtod(m, struct ip6_hdr *); 1823 th = (struct tcphdr *)(ip6 + 1); 1824 tcpip_fillheaders(inp, ip6, th); 1825 } else 1826 #endif 1827 { 1828 hdrlen = sizeof(struct tcpiphdr); 1829 ip = mtod(m, struct ip *); 1830 th = (struct tcphdr *)(ip + 1); 1831 tcpip_fillheaders(inp, ip, th); 1832 } 1833 optp = (u_int8_t *)(th + 1); 1834 1835 /* 1836 * Send a timestamp and echo-reply if both our side and our peer 1837 * have sent timestamps in our SYN's and this is not a RST. 1838 */ 1839 if (tw->t_recent && flags == TH_ACK) { 1840 u_int32_t *lp = (u_int32_t *)optp; 1841 1842 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1843 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1844 *lp++ = htonl(ticks); 1845 *lp = htonl(tw->t_recent); 1846 optp += TCPOLEN_TSTAMP_APPA; 1847 } 1848 1849 optlen = optp - (u_int8_t *)(th + 1); 1850 1851 m->m_len = hdrlen + optlen; 1852 m->m_pkthdr.len = m->m_len; 1853 1854 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 1855 1856 th->th_seq = htonl(tw->snd_nxt); 1857 th->th_ack = htonl(tw->rcv_nxt); 1858 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1859 th->th_flags = flags; 1860 th->th_win = htons(tw->last_win); 1861 1862 #ifdef INET6 1863 if (isipv6) { 1864 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 1865 sizeof(struct tcphdr) + optlen); 1866 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 1867 error = ip6_output(m, inp->in6p_outputopts, NULL, 1868 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 1869 } else 1870 #endif 1871 { 1872 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1873 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 1874 m->m_pkthdr.csum_flags = CSUM_TCP; 1875 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1876 ip->ip_len = m->m_pkthdr.len; 1877 if (path_mtu_discovery) 1878 ip->ip_off |= IP_DF; 1879 error = ip_output(m, inp->inp_options, NULL, 1880 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 1881 NULL, inp); 1882 } 1883 if (flags & TH_ACK) 1884 tcpstat.tcps_sndacks++; 1885 else 1886 tcpstat.tcps_sndctrl++; 1887 tcpstat.tcps_sndtotal++; 1888 return (error); 1889 } 1890 1891 /* 1892 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1893 * 1894 * This code attempts to calculate the bandwidth-delay product as a 1895 * means of determining the optimal window size to maximize bandwidth, 1896 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1897 * routers. This code also does a fairly good job keeping RTTs in check 1898 * across slow links like modems. We implement an algorithm which is very 1899 * similar (but not meant to be) TCP/Vegas. The code operates on the 1900 * transmitter side of a TCP connection and so only effects the transmit 1901 * side of the connection. 1902 * 1903 * BACKGROUND: TCP makes no provision for the management of buffer space 1904 * at the end points or at the intermediate routers and switches. A TCP 1905 * stream, whether using NewReno or not, will eventually buffer as 1906 * many packets as it is able and the only reason this typically works is 1907 * due to the fairly small default buffers made available for a connection 1908 * (typicaly 16K or 32K). As machines use larger windows and/or window 1909 * scaling it is now fairly easy for even a single TCP connection to blow-out 1910 * all available buffer space not only on the local interface, but on 1911 * intermediate routers and switches as well. NewReno makes a misguided 1912 * attempt to 'solve' this problem by waiting for an actual failure to occur, 1913 * then backing off, then steadily increasing the window again until another 1914 * failure occurs, ad-infinitum. This results in terrible oscillation that 1915 * is only made worse as network loads increase and the idea of intentionally 1916 * blowing out network buffers is, frankly, a terrible way to manage network 1917 * resources. 1918 * 1919 * It is far better to limit the transmit window prior to the failure 1920 * condition being achieved. There are two general ways to do this: First 1921 * you can 'scan' through different transmit window sizes and locate the 1922 * point where the RTT stops increasing, indicating that you have filled the 1923 * pipe, then scan backwards until you note that RTT stops decreasing, then 1924 * repeat ad-infinitum. This method works in principle but has severe 1925 * implementation issues due to RTT variances, timer granularity, and 1926 * instability in the algorithm which can lead to many false positives and 1927 * create oscillations as well as interact badly with other TCP streams 1928 * implementing the same algorithm. 1929 * 1930 * The second method is to limit the window to the bandwidth delay product 1931 * of the link. This is the method we implement. RTT variances and our 1932 * own manipulation of the congestion window, bwnd, can potentially 1933 * destabilize the algorithm. For this reason we have to stabilize the 1934 * elements used to calculate the window. We do this by using the minimum 1935 * observed RTT, the long term average of the observed bandwidth, and 1936 * by adding two segments worth of slop. It isn't perfect but it is able 1937 * to react to changing conditions and gives us a very stable basis on 1938 * which to extend the algorithm. 1939 */ 1940 void 1941 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 1942 { 1943 u_long bw; 1944 u_long bwnd; 1945 int save_ticks; 1946 1947 INP_LOCK_ASSERT(tp->t_inpcb); 1948 1949 /* 1950 * If inflight_enable is disabled in the middle of a tcp connection, 1951 * make sure snd_bwnd is effectively disabled. 1952 */ 1953 if (tcp_inflight_enable == 0) { 1954 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1955 tp->snd_bandwidth = 0; 1956 return; 1957 } 1958 1959 /* 1960 * Figure out the bandwidth. Due to the tick granularity this 1961 * is a very rough number and it MUST be averaged over a fairly 1962 * long period of time. XXX we need to take into account a link 1963 * that is not using all available bandwidth, but for now our 1964 * slop will ramp us up if this case occurs and the bandwidth later 1965 * increases. 1966 * 1967 * Note: if ticks rollover 'bw' may wind up negative. We must 1968 * effectively reset t_bw_rtttime for this case. 1969 */ 1970 save_ticks = ticks; 1971 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 1972 return; 1973 1974 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 1975 (save_ticks - tp->t_bw_rtttime); 1976 tp->t_bw_rtttime = save_ticks; 1977 tp->t_bw_rtseq = ack_seq; 1978 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 1979 return; 1980 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 1981 1982 tp->snd_bandwidth = bw; 1983 1984 /* 1985 * Calculate the semi-static bandwidth delay product, plus two maximal 1986 * segments. The additional slop puts us squarely in the sweet 1987 * spot and also handles the bandwidth run-up case and stabilization. 1988 * Without the slop we could be locking ourselves into a lower 1989 * bandwidth. 1990 * 1991 * Situations Handled: 1992 * (1) Prevents over-queueing of packets on LANs, especially on 1993 * high speed LANs, allowing larger TCP buffers to be 1994 * specified, and also does a good job preventing 1995 * over-queueing of packets over choke points like modems 1996 * (at least for the transmit side). 1997 * 1998 * (2) Is able to handle changing network loads (bandwidth 1999 * drops so bwnd drops, bandwidth increases so bwnd 2000 * increases). 2001 * 2002 * (3) Theoretically should stabilize in the face of multiple 2003 * connections implementing the same algorithm (this may need 2004 * a little work). 2005 * 2006 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2007 * be adjusted with a sysctl but typically only needs to be 2008 * on very slow connections. A value no smaller then 5 2009 * should be used, but only reduce this default if you have 2010 * no other choice. 2011 */ 2012 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2013 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2014 #undef USERTT 2015 2016 if (tcp_inflight_debug > 0) { 2017 static int ltime; 2018 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2019 ltime = ticks; 2020 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2021 tp, 2022 bw, 2023 tp->t_rttbest, 2024 tp->t_srtt, 2025 bwnd 2026 ); 2027 } 2028 } 2029 if ((long)bwnd < tcp_inflight_min) 2030 bwnd = tcp_inflight_min; 2031 if (bwnd > tcp_inflight_max) 2032 bwnd = tcp_inflight_max; 2033 if ((long)bwnd < tp->t_maxseg * 2) 2034 bwnd = tp->t_maxseg * 2; 2035 tp->snd_bwnd = bwnd; 2036 } 2037 2038 #ifdef TCP_SIGNATURE 2039 /* 2040 * Callback function invoked by m_apply() to digest TCP segment data 2041 * contained within an mbuf chain. 2042 */ 2043 static int 2044 tcp_signature_apply(void *fstate, void *data, u_int len) 2045 { 2046 2047 MD5Update(fstate, (u_char *)data, len); 2048 return (0); 2049 } 2050 2051 /* 2052 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2053 * 2054 * Parameters: 2055 * m pointer to head of mbuf chain 2056 * off0 offset to TCP header within the mbuf chain 2057 * len length of TCP segment data, excluding options 2058 * optlen length of TCP segment options 2059 * buf pointer to storage for computed MD5 digest 2060 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2061 * 2062 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2063 * When called from tcp_input(), we can be sure that th_sum has been 2064 * zeroed out and verified already. 2065 * 2066 * This function is for IPv4 use only. Calling this function with an 2067 * IPv6 packet in the mbuf chain will yield undefined results. 2068 * 2069 * Return 0 if successful, otherwise return -1. 2070 * 2071 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2072 * search with the destination IP address, and a 'magic SPI' to be 2073 * determined by the application. This is hardcoded elsewhere to 1179 2074 * right now. Another branch of this code exists which uses the SPD to 2075 * specify per-application flows but it is unstable. 2076 */ 2077 int 2078 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2079 u_char *buf, u_int direction) 2080 { 2081 union sockaddr_union dst; 2082 struct ippseudo ippseudo; 2083 MD5_CTX ctx; 2084 int doff; 2085 struct ip *ip; 2086 struct ipovly *ipovly; 2087 struct secasvar *sav; 2088 struct tcphdr *th; 2089 u_short savecsum; 2090 2091 KASSERT(m != NULL, ("NULL mbuf chain")); 2092 KASSERT(buf != NULL, ("NULL signature pointer")); 2093 2094 /* Extract the destination from the IP header in the mbuf. */ 2095 ip = mtod(m, struct ip *); 2096 bzero(&dst, sizeof(union sockaddr_union)); 2097 dst.sa.sa_len = sizeof(struct sockaddr_in); 2098 dst.sa.sa_family = AF_INET; 2099 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2100 ip->ip_src : ip->ip_dst; 2101 2102 /* Look up an SADB entry which matches the address of the peer. */ 2103 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2104 if (sav == NULL) { 2105 printf("%s: SADB lookup failed for %s\n", __func__, 2106 inet_ntoa(dst.sin.sin_addr)); 2107 return (EINVAL); 2108 } 2109 2110 MD5Init(&ctx); 2111 ipovly = (struct ipovly *)ip; 2112 th = (struct tcphdr *)((u_char *)ip + off0); 2113 doff = off0 + sizeof(struct tcphdr) + optlen; 2114 2115 /* 2116 * Step 1: Update MD5 hash with IP pseudo-header. 2117 * 2118 * XXX The ippseudo header MUST be digested in network byte order, 2119 * or else we'll fail the regression test. Assume all fields we've 2120 * been doing arithmetic on have been in host byte order. 2121 * XXX One cannot depend on ipovly->ih_len here. When called from 2122 * tcp_output(), the underlying ip_len member has not yet been set. 2123 */ 2124 ippseudo.ippseudo_src = ipovly->ih_src; 2125 ippseudo.ippseudo_dst = ipovly->ih_dst; 2126 ippseudo.ippseudo_pad = 0; 2127 ippseudo.ippseudo_p = IPPROTO_TCP; 2128 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2129 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2130 2131 /* 2132 * Step 2: Update MD5 hash with TCP header, excluding options. 2133 * The TCP checksum must be set to zero. 2134 */ 2135 savecsum = th->th_sum; 2136 th->th_sum = 0; 2137 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2138 th->th_sum = savecsum; 2139 2140 /* 2141 * Step 3: Update MD5 hash with TCP segment data. 2142 * Use m_apply() to avoid an early m_pullup(). 2143 */ 2144 if (len > 0) 2145 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2146 2147 /* 2148 * Step 4: Update MD5 hash with shared secret. 2149 */ 2150 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2151 MD5Final(buf, &ctx); 2152 2153 key_sa_recordxfer(sav, m); 2154 KEY_FREESAV(&sav); 2155 return (0); 2156 } 2157 #endif /* TCP_SIGNATURE */ 2158 2159 static int 2160 sysctl_drop(SYSCTL_HANDLER_ARGS) 2161 { 2162 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2163 struct sockaddr_storage addrs[2]; 2164 struct inpcb *inp; 2165 struct tcpcb *tp; 2166 struct sockaddr_in *fin, *lin; 2167 #ifdef INET6 2168 struct sockaddr_in6 *fin6, *lin6; 2169 struct in6_addr f6, l6; 2170 #endif 2171 int error; 2172 2173 inp = NULL; 2174 fin = lin = NULL; 2175 #ifdef INET6 2176 fin6 = lin6 = NULL; 2177 #endif 2178 error = 0; 2179 2180 if (req->oldptr != NULL || req->oldlen != 0) 2181 return (EINVAL); 2182 if (req->newptr == NULL) 2183 return (EPERM); 2184 if (req->newlen < sizeof(addrs)) 2185 return (ENOMEM); 2186 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2187 if (error) 2188 return (error); 2189 2190 switch (addrs[0].ss_family) { 2191 #ifdef INET6 2192 case AF_INET6: 2193 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2194 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2195 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2196 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2197 return (EINVAL); 2198 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2199 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2200 return (EINVAL); 2201 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2202 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2203 fin = (struct sockaddr_in *)&addrs[0]; 2204 lin = (struct sockaddr_in *)&addrs[1]; 2205 break; 2206 } 2207 error = in6_embedscope(&f6, fin6, NULL, NULL); 2208 if (error) 2209 return (EINVAL); 2210 error = in6_embedscope(&l6, lin6, NULL, NULL); 2211 if (error) 2212 return (EINVAL); 2213 break; 2214 #endif 2215 case AF_INET: 2216 fin = (struct sockaddr_in *)&addrs[0]; 2217 lin = (struct sockaddr_in *)&addrs[1]; 2218 if (fin->sin_len != sizeof(struct sockaddr_in) || 2219 lin->sin_len != sizeof(struct sockaddr_in)) 2220 return (EINVAL); 2221 break; 2222 default: 2223 return (EINVAL); 2224 } 2225 INP_INFO_WLOCK(&tcbinfo); 2226 switch (addrs[0].ss_family) { 2227 #ifdef INET6 2228 case AF_INET6: 2229 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2230 &l6, lin6->sin6_port, 0, NULL); 2231 break; 2232 #endif 2233 case AF_INET: 2234 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2235 lin->sin_addr, lin->sin_port, 0, NULL); 2236 break; 2237 } 2238 if (inp != NULL) { 2239 INP_LOCK(inp); 2240 if ((tp = intotcpcb(inp)) && 2241 ((inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 2242 tp = tcp_drop(tp, ECONNABORTED); 2243 if (tp != NULL) 2244 INP_UNLOCK(inp); 2245 } else 2246 INP_UNLOCK(inp); 2247 } else 2248 error = ESRCH; 2249 INP_INFO_WUNLOCK(&tcbinfo); 2250 return (error); 2251 } 2252 2253 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2254 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2255 0, sysctl_drop, "", "Drop TCP connection"); 2256