1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 34 * $FreeBSD: src/sys/netinet/tcp_subr.c,v 1.73.2.31 2003/01/24 05:11:34 sam Exp $ 35 * $DragonFly: src/sys/netinet/tcp_subr.c,v 1.30 2004/04/28 08:00:35 hsu Exp $ 36 */ 37 38 #include "opt_compat.h" 39 #include "opt_inet6.h" 40 #include "opt_ipsec.h" 41 #include "opt_tcpdebug.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/callout.h> 46 #include <sys/kernel.h> 47 #include <sys/sysctl.h> 48 #include <sys/malloc.h> 49 #include <sys/mbuf.h> 50 #ifdef INET6 51 #include <sys/domain.h> 52 #endif 53 #include <sys/proc.h> 54 #include <sys/socket.h> 55 #include <sys/socketvar.h> 56 #include <sys/protosw.h> 57 #include <sys/random.h> 58 #include <sys/in_cksum.h> 59 60 #include <vm/vm_zone.h> 61 62 #include <net/route.h> 63 #include <net/if.h> 64 #include <net/netisr.h> 65 66 #define _IP_VHL 67 #include <netinet/in.h> 68 #include <netinet/in_systm.h> 69 #include <netinet/ip.h> 70 #include <netinet/ip6.h> 71 #include <netinet/in_pcb.h> 72 #include <netinet6/in6_pcb.h> 73 #include <netinet/in_var.h> 74 #include <netinet/ip_var.h> 75 #include <netinet6/ip6_var.h> 76 #include <netinet/tcp.h> 77 #include <netinet/tcp_fsm.h> 78 #include <netinet/tcp_seq.h> 79 #include <netinet/tcp_timer.h> 80 #include <netinet/tcp_var.h> 81 #include <netinet6/tcp6_var.h> 82 #include <netinet/tcpip.h> 83 #ifdef TCPDEBUG 84 #include <netinet/tcp_debug.h> 85 #endif 86 #include <netinet6/ip6protosw.h> 87 88 #ifdef IPSEC 89 #include <netinet6/ipsec.h> 90 #ifdef INET6 91 #include <netinet6/ipsec6.h> 92 #endif 93 #endif 94 95 #ifdef FAST_IPSEC 96 #include <netipsec/ipsec.h> 97 #ifdef INET6 98 #include <netipsec/ipsec6.h> 99 #endif 100 #define IPSEC 101 #endif 102 103 #include <sys/md5.h> 104 105 #include <sys/msgport2.h> 106 107 int tcp_mssdflt = TCP_MSS; 108 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 109 &tcp_mssdflt, 0, "Default TCP Maximum Segment Size"); 110 111 #ifdef INET6 112 int tcp_v6mssdflt = TCP6_MSS; 113 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLFLAG_RW, 114 &tcp_v6mssdflt, 0, "Default TCP Maximum Segment Size for IPv6"); 115 #endif 116 117 #if 0 118 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 119 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 120 &tcp_rttdflt, 0, "Default maximum TCP Round Trip Time"); 121 #endif 122 123 int tcp_do_rfc1323 = 1; 124 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 125 &tcp_do_rfc1323, 0, "Enable rfc1323 (high performance TCP) extensions"); 126 127 int tcp_do_rfc1644 = 0; 128 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW, 129 &tcp_do_rfc1644, 0, "Enable rfc1644 (TTCP) extensions"); 130 131 static int tcp_tcbhashsize = 0; 132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD, 133 &tcp_tcbhashsize, 0, "Size of TCP control block hashtable"); 134 135 static int do_tcpdrain = 1; 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 137 "Enable tcp_drain routine for extra help when low on mbufs"); 138 139 /* XXX JH */ 140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 141 &tcbinfo[0].ipi_count, 0, "Number of active PCBs"); 142 143 static int icmp_may_rst = 1; 144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 145 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 146 147 static int tcp_isn_reseed_interval = 0; 148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 149 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 150 151 /* 152 * TCP bandwidth limiting sysctls. Note that the default lower bound of 153 * 1024 exists only for debugging. A good production default would be 154 * something like 6100. 155 */ 156 static int tcp_inflight_enable = 0; 157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_enable, CTLFLAG_RW, 158 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 159 160 static int tcp_inflight_debug = 0; 161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_debug, CTLFLAG_RW, 162 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 163 164 static int tcp_inflight_min = 6144; 165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_min, CTLFLAG_RW, 166 &tcp_inflight_min, 0, "Lower bound for TCP inflight window"); 167 168 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_max, CTLFLAG_RW, 170 &tcp_inflight_max, 0, "Upper bound for TCP inflight window"); 171 172 static int tcp_inflight_stab = 20; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_stab, CTLFLAG_RW, 174 &tcp_inflight_stab, 0, "Slop in maximal packets / 10 (20 = 2 packets)"); 175 176 static void tcp_cleartaocache (void); 177 static void tcp_notify (struct inpcb *, int); 178 179 struct tcp_stats tcpstats_ary[MAXCPU]; 180 #ifdef SMP 181 static int 182 sysctl_tcpstats(SYSCTL_HANDLER_ARGS) 183 { 184 int cpu, error = 0; 185 186 for (cpu = 0; cpu < ncpus; ++cpu) { 187 if ((error = SYSCTL_OUT(req, (void *)&tcpstats_ary[cpu], 188 sizeof(struct tcp_stats)))) 189 break; 190 if ((error = SYSCTL_IN(req, (void *)&tcpstats_ary[cpu], 191 sizeof(struct tcp_stats)))) 192 break; 193 } 194 195 return (error); 196 } 197 SYSCTL_PROC(_net_inet_tcp, TCPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW), 198 0, 0, sysctl_tcpstats, "S,tcp_stats", "TCP statistics"); 199 #else 200 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 201 &tcpstat, tcp_stats, "TCP statistics"); 202 #endif 203 204 /* 205 * Target size of TCP PCB hash tables. Must be a power of two. 206 * 207 * Note that this can be overridden by the kernel environment 208 * variable net.inet.tcp.tcbhashsize 209 */ 210 #ifndef TCBHASHSIZE 211 #define TCBHASHSIZE 512 212 #endif 213 214 /* 215 * This is the actual shape of what we allocate using the zone 216 * allocator. Doing it this way allows us to protect both structures 217 * using the same generation count, and also eliminates the overhead 218 * of allocating tcpcbs separately. By hiding the structure here, 219 * we avoid changing most of the rest of the code (although it needs 220 * to be changed, eventually, for greater efficiency). 221 */ 222 #define ALIGNMENT 32 223 #define ALIGNM1 (ALIGNMENT - 1) 224 struct inp_tp { 225 union { 226 struct inpcb inp; 227 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1]; 228 } inp_tp_u; 229 struct tcpcb tcb; 230 struct callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl; 231 struct callout inp_tp_delack; 232 }; 233 #undef ALIGNMENT 234 #undef ALIGNM1 235 236 /* 237 * Tcp initialization 238 */ 239 void 240 tcp_init() 241 { 242 struct inpcbporthead *porthashbase; 243 u_long porthashmask; 244 struct vm_zone *ipi_zone; 245 int hashsize = TCBHASHSIZE; 246 int cpu; 247 248 tcp_ccgen = 1; 249 tcp_cleartaocache(); 250 251 tcp_delacktime = TCPTV_DELACK; 252 tcp_keepinit = TCPTV_KEEP_INIT; 253 tcp_keepidle = TCPTV_KEEP_IDLE; 254 tcp_keepintvl = TCPTV_KEEPINTVL; 255 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 256 tcp_msl = TCPTV_MSL; 257 tcp_rexmit_min = TCPTV_MIN; 258 tcp_rexmit_slop = TCPTV_CPU_VAR; 259 260 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 261 if (!powerof2(hashsize)) { 262 printf("WARNING: TCB hash size not a power of 2\n"); 263 hashsize = 512; /* safe default */ 264 } 265 tcp_tcbhashsize = hashsize; 266 porthashbase = hashinit(hashsize, M_PCB, &porthashmask); 267 ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets, 268 ZONE_INTERRUPT, 0); 269 270 for (cpu = 0; cpu < ncpus2; cpu++) { 271 LIST_INIT(&tcbinfo[cpu].listhead); 272 tcbinfo[cpu].hashbase = hashinit(hashsize, M_PCB, 273 &tcbinfo[cpu].hashmask); 274 tcbinfo[cpu].porthashbase = porthashbase; 275 tcbinfo[cpu].porthashmask = porthashmask; 276 tcbinfo[cpu].wildcardhashbase = hashinit(hashsize, M_PCB, 277 &tcbinfo[cpu].wildcardhashmask); 278 tcbinfo[cpu].ipi_zone = ipi_zone; 279 } 280 281 tcp_reass_maxseg = nmbclusters / 16; 282 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", &tcp_reass_maxseg); 283 284 #ifdef INET6 285 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 286 #else 287 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 288 #endif 289 if (max_protohdr < TCP_MINPROTOHDR) 290 max_protohdr = TCP_MINPROTOHDR; 291 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 292 panic("tcp_init"); 293 #undef TCP_MINPROTOHDR 294 295 /* 296 * Initialize TCP statistics. 297 * 298 * It is layed out as an array which is has one element for UP, 299 * and SMP_MAXCPU elements for SMP. This allows us to retain 300 * the access mechanism from userland for both UP and SMP. 301 */ 302 #ifdef SMP 303 for (cpu = 0; cpu < ncpus; ++cpu) { 304 bzero(&tcpstats_ary[cpu], sizeof(struct tcp_stats)); 305 } 306 #else 307 bzero(&tcpstat, sizeof(struct tcp_stats)); 308 #endif 309 310 syncache_init(); 311 tcp_thread_init(); 312 } 313 314 /* 315 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 316 * tcp_template used to store this data in mbufs, but we now recopy it out 317 * of the tcpcb each time to conserve mbufs. 318 */ 319 void 320 tcp_fillheaders(struct tcpcb *tp, void *ip_ptr, void *tcp_ptr) 321 { 322 struct inpcb *inp = tp->t_inpcb; 323 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr; 324 325 #ifdef INET6 326 if (inp->inp_vflag & INP_IPV6) { 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 tcp_hdr->th_sum = 0; 339 } else 340 #endif 341 { 342 struct ip *ip = (struct ip *) ip_ptr; 343 344 ip->ip_vhl = IP_VHL_BORING; 345 ip->ip_tos = 0; 346 ip->ip_len = 0; 347 ip->ip_id = 0; 348 ip->ip_off = 0; 349 ip->ip_ttl = 0; 350 ip->ip_sum = 0; 351 ip->ip_p = IPPROTO_TCP; 352 ip->ip_src = inp->inp_laddr; 353 ip->ip_dst = inp->inp_faddr; 354 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, 355 ip->ip_dst.s_addr, 356 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 357 } 358 359 tcp_hdr->th_sport = inp->inp_lport; 360 tcp_hdr->th_dport = inp->inp_fport; 361 tcp_hdr->th_seq = 0; 362 tcp_hdr->th_ack = 0; 363 tcp_hdr->th_x2 = 0; 364 tcp_hdr->th_off = 5; 365 tcp_hdr->th_flags = 0; 366 tcp_hdr->th_win = 0; 367 tcp_hdr->th_urp = 0; 368 } 369 370 /* 371 * Create template to be used to send tcp packets on a connection. 372 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 373 * use for this function is in keepalives, which use tcp_respond. 374 */ 375 struct tcptemp * 376 tcp_maketemplate(struct tcpcb *tp) 377 { 378 struct mbuf *m; 379 struct tcptemp *n; 380 381 m = m_get(M_DONTWAIT, MT_HEADER); 382 if (m == NULL) 383 return (NULL); 384 m->m_len = sizeof(struct tcptemp); 385 n = mtod(m, struct tcptemp *); 386 387 tcp_fillheaders(tp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 388 return (n); 389 } 390 391 /* 392 * Send a single message to the TCP at address specified by 393 * the given TCP/IP header. If m == NULL, then we make a copy 394 * of the tcpiphdr at ti and send directly to the addressed host. 395 * This is used to force keep alive messages out using the TCP 396 * template for a connection. If flags are given then we send 397 * a message back to the TCP which originated the * segment ti, 398 * and discard the mbuf containing it and any other attached mbufs. 399 * 400 * In any case the ack and sequence number of the transmitted 401 * segment are as specified by the parameters. 402 * 403 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 404 */ 405 void 406 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 407 tcp_seq ack, tcp_seq seq, int flags) 408 { 409 int tlen; 410 int win = 0; 411 struct route *ro = NULL; 412 struct route sro; 413 struct ip *ip = ipgen; 414 struct tcphdr *nth; 415 int ipflags = 0; 416 struct route_in6 *ro6 = NULL; 417 struct route_in6 sro6; 418 struct ip6_hdr *ip6 = ipgen; 419 #ifdef INET6 420 boolean_t isipv6 = (IP_VHL_V(ip->ip_vhl) == 6); 421 #else 422 const boolean_t isipv6 = FALSE; 423 #endif 424 425 if (tp != NULL) { 426 if (!(flags & TH_RST)) { 427 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 428 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 429 win = (long)TCP_MAXWIN << tp->rcv_scale; 430 } 431 if (isipv6) 432 ro6 = &tp->t_inpcb->in6p_route; 433 else 434 ro = &tp->t_inpcb->inp_route; 435 } else { 436 if (isipv6) { 437 ro6 = &sro6; 438 bzero(ro6, sizeof *ro6); 439 } else { 440 ro = &sro; 441 bzero(ro, sizeof *ro); 442 } 443 } 444 if (m == NULL) { 445 m = m_gethdr(M_DONTWAIT, MT_HEADER); 446 if (m == NULL) 447 return; 448 tlen = 0; 449 m->m_data += max_linkhdr; 450 if (isipv6) { 451 bcopy(ip6, mtod(m, caddr_t), sizeof(struct ip6_hdr)); 452 ip6 = mtod(m, struct ip6_hdr *); 453 nth = (struct tcphdr *)(ip6 + 1); 454 } else { 455 bcopy(ip, mtod(m, caddr_t), sizeof(struct ip)); 456 ip = mtod(m, struct ip *); 457 nth = (struct tcphdr *)(ip + 1); 458 } 459 bcopy(th, nth, sizeof(struct tcphdr)); 460 flags = TH_ACK; 461 } else { 462 m_freem(m->m_next); 463 m->m_next = NULL; 464 m->m_data = (caddr_t)ipgen; 465 /* m_len is set later */ 466 tlen = 0; 467 #define xchg(a, b, type) { type t; t = a; a = b; b = t; } 468 if (isipv6) { 469 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 470 nth = (struct tcphdr *)(ip6 + 1); 471 } else { 472 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 473 nth = (struct tcphdr *)(ip + 1); 474 } 475 if (th != nth) { 476 /* 477 * this is usually a case when an extension header 478 * exists between the IPv6 header and the 479 * TCP header. 480 */ 481 nth->th_sport = th->th_sport; 482 nth->th_dport = th->th_dport; 483 } 484 xchg(nth->th_dport, nth->th_sport, n_short); 485 #undef xchg 486 } 487 if (isipv6) { 488 ip6->ip6_flow = 0; 489 ip6->ip6_vfc = IPV6_VERSION; 490 ip6->ip6_nxt = IPPROTO_TCP; 491 ip6->ip6_plen = htons((u_short)(sizeof(struct tcphdr) + tlen)); 492 tlen += sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 493 } else { 494 tlen += sizeof(struct tcpiphdr); 495 ip->ip_len = tlen; 496 ip->ip_ttl = ip_defttl; 497 } 498 m->m_len = tlen; 499 m->m_pkthdr.len = tlen; 500 m->m_pkthdr.rcvif = (struct ifnet *) NULL; 501 nth->th_seq = htonl(seq); 502 nth->th_ack = htonl(ack); 503 nth->th_x2 = 0; 504 nth->th_off = sizeof(struct tcphdr) >> 2; 505 nth->th_flags = flags; 506 if (tp != NULL) 507 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 508 else 509 nth->th_win = htons((u_short)win); 510 nth->th_urp = 0; 511 if (isipv6) { 512 nth->th_sum = 0; 513 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 514 sizeof(struct ip6_hdr), 515 tlen - sizeof(struct ip6_hdr)); 516 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL, 517 (ro6 && ro6->ro_rt) ? 518 ro6->ro_rt->rt_ifp : NULL); 519 } else { 520 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 521 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 522 m->m_pkthdr.csum_flags = CSUM_TCP; 523 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 524 } 525 #ifdef TCPDEBUG 526 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 527 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 528 #endif 529 if (isipv6) { 530 (void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL, 531 tp ? tp->t_inpcb : NULL); 532 if ((ro6 == &sro6) && (ro6->ro_rt != NULL)) { 533 RTFREE(ro6->ro_rt); 534 ro6->ro_rt = NULL; 535 } 536 } else { 537 (void)ip_output(m, NULL, ro, ipflags, NULL, 538 tp ? tp->t_inpcb : NULL); 539 if ((ro == &sro) && (ro->ro_rt != NULL)) { 540 RTFREE(ro->ro_rt); 541 ro->ro_rt = NULL; 542 } 543 } 544 } 545 546 /* 547 * Create a new TCP control block, making an 548 * empty reassembly queue and hooking it to the argument 549 * protocol control block. The `inp' parameter must have 550 * come from the zone allocator set up in tcp_init(). 551 */ 552 struct tcpcb * 553 tcp_newtcpcb(struct inpcb *inp) 554 { 555 struct inp_tp *it; 556 struct tcpcb *tp; 557 #ifdef INET6 558 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0); 559 #else 560 const boolean_t isipv6 = FALSE; 561 #endif 562 563 it = (struct inp_tp *)inp; 564 tp = &it->tcb; 565 bzero(tp, sizeof(struct tcpcb)); 566 LIST_INIT(&tp->t_segq); 567 tp->t_maxseg = tp->t_maxopd = isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 568 569 /* Set up our timeouts. */ 570 callout_init(tp->tt_rexmt = &it->inp_tp_rexmt); 571 callout_init(tp->tt_persist = &it->inp_tp_persist); 572 callout_init(tp->tt_keep = &it->inp_tp_keep); 573 callout_init(tp->tt_2msl = &it->inp_tp_2msl); 574 callout_init(tp->tt_delack = &it->inp_tp_delack); 575 576 if (tcp_do_rfc1323) 577 tp->t_flags = (TF_REQ_SCALE | TF_REQ_TSTMP); 578 if (tcp_do_rfc1644) 579 tp->t_flags |= TF_REQ_CC; 580 tp->t_inpcb = inp; /* XXX */ 581 /* 582 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 583 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 584 * reasonable initial retransmit time. 585 */ 586 tp->t_srtt = TCPTV_SRTTBASE; 587 tp->t_rttvar = 588 ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 589 tp->t_rttmin = tcp_rexmit_min; 590 tp->t_rxtcur = TCPTV_RTOBASE; 591 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 592 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 593 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 594 tp->t_rcvtime = ticks; 595 tp->t_bw_rtttime = ticks; 596 /* 597 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 598 * because the socket may be bound to an IPv6 wildcard address, 599 * which may match an IPv4-mapped IPv6 address. 600 */ 601 inp->inp_ip_ttl = ip_defttl; 602 inp->inp_ppcb = (caddr_t)tp; 603 return (tp); /* XXX */ 604 } 605 606 /* 607 * Drop a TCP connection, reporting the specified error. 608 * If connection is synchronized, then send a RST to peer. 609 */ 610 struct tcpcb * 611 tcp_drop(struct tcpcb *tp, int errno) 612 { 613 struct socket *so = tp->t_inpcb->inp_socket; 614 615 if (TCPS_HAVERCVDSYN(tp->t_state)) { 616 tp->t_state = TCPS_CLOSED; 617 (void) tcp_output(tp); 618 tcpstat.tcps_drops++; 619 } else 620 tcpstat.tcps_conndrops++; 621 if (errno == ETIMEDOUT && tp->t_softerror) 622 errno = tp->t_softerror; 623 so->so_error = errno; 624 return (tcp_close(tp)); 625 } 626 627 #ifdef SMP 628 struct netmsg_remwildcard { 629 struct lwkt_msg nm_lmsg; 630 struct inpcb *nm_inp; 631 struct inpcbinfo *nm_pcbinfo; 632 }; 633 634 static int 635 in_pcbremwildcardhash_handler(struct lwkt_msg *msg0) 636 { 637 struct netmsg_remwildcard *msg = (struct netmsg_remwildcard *)msg0; 638 639 in_pcbremwildcardhash_oncpu(msg->nm_inp, msg->nm_pcbinfo); 640 lwkt_replymsg(&msg->nm_lmsg, 0); 641 return (EASYNC); 642 } 643 #endif 644 645 /* 646 * Close a TCP control block: 647 * discard all space held by the tcp 648 * discard internet protocol block 649 * wake up any sleepers 650 */ 651 struct tcpcb * 652 tcp_close(struct tcpcb *tp) 653 { 654 struct tseg_qent *q; 655 struct inpcb *inp = tp->t_inpcb; 656 struct socket *so = inp->inp_socket; 657 struct rtentry *rt; 658 boolean_t dosavessthresh; 659 #ifdef SMP 660 int cpu; 661 #endif 662 #ifdef INET6 663 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0); 664 #else 665 const boolean_t isipv6 = FALSE; 666 #endif 667 668 /* 669 * Make sure that all of our timers are stopped before we 670 * delete the PCB. 671 */ 672 callout_stop(tp->tt_rexmt); 673 callout_stop(tp->tt_persist); 674 callout_stop(tp->tt_keep); 675 callout_stop(tp->tt_2msl); 676 callout_stop(tp->tt_delack); 677 678 /* 679 * If we got enough samples through the srtt filter, 680 * save the rtt and rttvar in the routing entry. 681 * 'Enough' is arbitrarily defined as the 16 samples. 682 * 16 samples is enough for the srtt filter to converge 683 * to within 5% of the correct value; fewer samples and 684 * we could save a very bogus rtt. 685 * 686 * Don't update the default route's characteristics and don't 687 * update anything that the user "locked". 688 */ 689 if (tp->t_rttupdated >= 16) { 690 u_long i = 0; 691 692 if (isipv6) { 693 struct sockaddr_in6 *sin6; 694 695 if ((rt = inp->in6p_route.ro_rt) == NULL) 696 goto no_valid_rt; 697 sin6 = (struct sockaddr_in6 *)rt_key(rt); 698 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 699 goto no_valid_rt; 700 } else 701 if ((rt = inp->inp_route.ro_rt) == NULL || 702 ((struct sockaddr_in *)rt_key(rt))-> 703 sin_addr.s_addr == INADDR_ANY) 704 goto no_valid_rt; 705 706 if (!(rt->rt_rmx.rmx_locks & RTV_RTT)) { 707 i = tp->t_srtt * (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 708 if (rt->rt_rmx.rmx_rtt && i) 709 /* 710 * filter this update to half the old & half 711 * the new values, converting scale. 712 * See route.h and tcp_var.h for a 713 * description of the scaling constants. 714 */ 715 rt->rt_rmx.rmx_rtt = 716 (rt->rt_rmx.rmx_rtt + i) / 2; 717 else 718 rt->rt_rmx.rmx_rtt = i; 719 tcpstat.tcps_cachedrtt++; 720 } 721 if (!(rt->rt_rmx.rmx_locks & RTV_RTTVAR)) { 722 i = tp->t_rttvar * 723 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 724 if (rt->rt_rmx.rmx_rttvar && i) 725 rt->rt_rmx.rmx_rttvar = 726 (rt->rt_rmx.rmx_rttvar + i) / 2; 727 else 728 rt->rt_rmx.rmx_rttvar = i; 729 tcpstat.tcps_cachedrttvar++; 730 } 731 /* 732 * The old comment here said: 733 * update the pipelimit (ssthresh) if it has been updated 734 * already or if a pipesize was specified & the threshhold 735 * got below half the pipesize. I.e., wait for bad news 736 * before we start updating, then update on both good 737 * and bad news. 738 * 739 * But we want to save the ssthresh even if no pipesize is 740 * specified explicitly in the route, because such 741 * connections still have an implicit pipesize specified 742 * by the global tcp_sendspace. In the absence of a reliable 743 * way to calculate the pipesize, it will have to do. 744 */ 745 i = tp->snd_ssthresh; 746 if (rt->rt_rmx.rmx_sendpipe != 0) 747 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe/2); 748 else 749 dosavessthresh = (i < so->so_snd.sb_hiwat/2); 750 if (dosavessthresh || 751 (!(rt->rt_rmx.rmx_locks & RTV_SSTHRESH) && (i != 0) && 752 (rt->rt_rmx.rmx_ssthresh != 0))) { 753 /* 754 * convert the limit from user data bytes to 755 * packets then to packet data bytes. 756 */ 757 i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 758 if (i < 2) 759 i = 2; 760 i *= tp->t_maxseg + 761 (isipv6 ? 762 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 763 sizeof(struct tcpiphdr)); 764 if (rt->rt_rmx.rmx_ssthresh) 765 rt->rt_rmx.rmx_ssthresh = 766 (rt->rt_rmx.rmx_ssthresh + i) / 2; 767 else 768 rt->rt_rmx.rmx_ssthresh = i; 769 tcpstat.tcps_cachedssthresh++; 770 } 771 } 772 773 no_valid_rt: 774 /* free the reassembly queue, if any */ 775 while((q = LIST_FIRST(&tp->t_segq)) != NULL) { 776 LIST_REMOVE(q, tqe_q); 777 m_freem(q->tqe_m); 778 FREE(q, M_TSEGQ); 779 tcp_reass_qsize--; 780 } 781 inp->inp_ppcb = NULL; 782 soisdisconnected(so); 783 784 #ifdef SMP 785 if (inp->inp_flags & INP_WILDCARD_MP) { 786 for (cpu = 0; cpu < ncpus2; cpu ++) { 787 struct netmsg_remwildcard *msg; 788 789 msg = malloc(sizeof(struct netmsg_remwildcard), 790 M_LWKTMSG, M_INTWAIT); 791 lwkt_initmsg(&msg->nm_lmsg, &netisr_afree_rport, 0, 792 lwkt_cmd_func(in_pcbremwildcardhash_handler), 793 lwkt_cmd_op_none); 794 msg->nm_inp = inp; 795 msg->nm_pcbinfo = &tcbinfo[cpu]; 796 lwkt_sendmsg(tcp_cport(cpu), &msg->nm_lmsg); 797 } 798 } 799 #endif 800 801 #ifdef INET6 802 if (INP_CHECK_SOCKAF(so, AF_INET6)) 803 in6_pcbdetach(inp); 804 else 805 #endif 806 in_pcbdetach(inp); 807 tcpstat.tcps_closed++; 808 return (NULL); 809 } 810 811 static __inline void 812 tcp_drain_oncpu(struct inpcbhead *head) 813 { 814 struct inpcb *inpb; 815 struct tcpcb *tcpb; 816 struct tseg_qent *te; 817 818 LIST_FOREACH(inpb, head, inp_list) { 819 if ((tcpb = intotcpcb(inpb))) { 820 while ((te = LIST_FIRST(&tcpb->t_segq)) != NULL) { 821 LIST_REMOVE(te, tqe_q); 822 m_freem(te->tqe_m); 823 FREE(te, M_TSEGQ); 824 tcp_reass_qsize--; 825 } 826 } 827 } 828 } 829 830 #ifdef SMP 831 struct netmsg_tcp_drain { 832 struct lwkt_msg nm_lmsg; 833 struct inpcbhead *nm_head; 834 }; 835 836 static int 837 tcp_drain_handler(lwkt_msg_t lmsg) 838 { 839 struct netmsg_tcp_drain *nm = (void *)lmsg; 840 841 tcp_drain_oncpu(nm->nm_head); 842 lwkt_replymsg(lmsg, 0); 843 return(EASYNC); 844 } 845 #endif 846 847 void 848 tcp_drain() 849 { 850 #ifdef SMP 851 int cpu; 852 #endif 853 854 if (!do_tcpdrain) 855 return; 856 857 /* 858 * Walk the tcpbs, if existing, and flush the reassembly queue, 859 * if there is one... 860 * XXX: The "Net/3" implementation doesn't imply that the TCP 861 * reassembly queue should be flushed, but in a situation 862 * where we're really low on mbufs, this is potentially 863 * useful. 864 */ 865 #ifdef SMP 866 for (cpu = 0; cpu < ncpus2; cpu++) { 867 struct netmsg_tcp_drain *msg; 868 869 if (cpu == mycpu->gd_cpuid) { 870 tcp_drain_oncpu(&tcbinfo[cpu].listhead); 871 } else { 872 msg = malloc(sizeof(struct netmsg_tcp_drain), 873 M_LWKTMSG, M_NOWAIT); 874 if (msg == NULL) 875 continue; 876 lwkt_initmsg(&msg->nm_lmsg, &netisr_afree_rport, 0, 877 lwkt_cmd_func(tcp_drain_handler), 878 lwkt_cmd_op_none); 879 msg->nm_head = &tcbinfo[cpu].listhead; 880 lwkt_sendmsg(tcp_cport(cpu), &msg->nm_lmsg); 881 } 882 } 883 #else 884 tcp_drain_oncpu(&tcbinfo[0].listhead); 885 #endif 886 } 887 888 /* 889 * Notify a tcp user of an asynchronous error; 890 * store error as soft error, but wake up user 891 * (for now, won't do anything until can select for soft error). 892 * 893 * Do not wake up user since there currently is no mechanism for 894 * reporting soft errors (yet - a kqueue filter may be added). 895 */ 896 static void 897 tcp_notify(struct inpcb *inp, int error) 898 { 899 struct tcpcb *tp = intotcpcb(inp); 900 901 /* 902 * Ignore some errors if we are hooked up. 903 * If connection hasn't completed, has retransmitted several times, 904 * and receives a second error, give up now. This is better 905 * than waiting a long time to establish a connection that 906 * can never complete. 907 */ 908 if (tp->t_state == TCPS_ESTABLISHED && 909 (error == EHOSTUNREACH || error == ENETUNREACH || 910 error == EHOSTDOWN)) { 911 return; 912 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 913 tp->t_softerror) 914 tcp_drop(tp, error); 915 else 916 tp->t_softerror = error; 917 #if 0 918 wakeup((caddr_t) &so->so_timeo); 919 sorwakeup(so); 920 sowwakeup(so); 921 #endif 922 } 923 924 static int 925 tcp_pcblist(SYSCTL_HANDLER_ARGS) 926 { 927 int error, i, n, s; 928 struct inpcb *inp, **inp_list; 929 inp_gen_t gencnt; 930 struct xinpgen xig; 931 932 /* 933 * The process of preparing the TCB list is too time-consuming and 934 * resource-intensive to repeat twice on every request. 935 */ 936 if (req->oldptr == NULL) { 937 n = tcbinfo[mycpu->gd_cpuid].ipi_count; 938 req->oldidx = 2 * (sizeof xig) + 939 (n + n/8) * sizeof(struct xtcpcb); 940 return (0); 941 } 942 943 if (req->newptr != NULL) 944 return (EPERM); 945 946 /* 947 * OK, now we're committed to doing something. 948 */ 949 s = splnet(); 950 gencnt = tcbinfo[mycpu->gd_cpuid].ipi_gencnt; 951 n = tcbinfo[mycpu->gd_cpuid].ipi_count; 952 splx(s); 953 954 xig.xig_len = sizeof xig; 955 xig.xig_count = n; 956 xig.xig_gen = gencnt; 957 xig.xig_sogen = so_gencnt; 958 error = SYSCTL_OUT(req, &xig, sizeof xig); 959 if (error != 0) 960 return (error); 961 962 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 963 if (inp_list == NULL) 964 return (ENOMEM); 965 966 s = splnet(); 967 for (inp = LIST_FIRST(&tcbinfo[mycpu->gd_cpuid].listhead), i = 0; 968 inp && i < n; inp = LIST_NEXT(inp, inp_list)) { 969 if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->td, inp)) 970 inp_list[i++] = inp; 971 } 972 splx(s); 973 n = i; 974 975 error = 0; 976 for (i = 0; i < n; i++) { 977 inp = inp_list[i]; 978 if (inp->inp_gencnt <= gencnt) { 979 struct xtcpcb xt; 980 caddr_t inp_ppcb; 981 xt.xt_len = sizeof xt; 982 /* XXX should avoid extra copy */ 983 bcopy(inp, &xt.xt_inp, sizeof *inp); 984 inp_ppcb = inp->inp_ppcb; 985 if (inp_ppcb != NULL) 986 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 987 else 988 bzero(&xt.xt_tp, sizeof xt.xt_tp); 989 if (inp->inp_socket) 990 sotoxsocket(inp->inp_socket, &xt.xt_socket); 991 error = SYSCTL_OUT(req, &xt, sizeof xt); 992 } 993 } 994 if (!error) { 995 /* 996 * Give the user an updated idea of our state. 997 * If the generation differs from what we told 998 * her before, she knows that something happened 999 * while we were processing this request, and it 1000 * might be necessary to retry. 1001 */ 1002 s = splnet(); 1003 xig.xig_gen = tcbinfo[mycpu->gd_cpuid].ipi_gencnt; 1004 xig.xig_sogen = so_gencnt; 1005 xig.xig_count = tcbinfo[mycpu->gd_cpuid].ipi_count; 1006 splx(s); 1007 error = SYSCTL_OUT(req, &xig, sizeof xig); 1008 } 1009 free(inp_list, M_TEMP); 1010 return (error); 1011 } 1012 1013 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1014 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1015 1016 static int 1017 tcp_getcred(SYSCTL_HANDLER_ARGS) 1018 { 1019 struct sockaddr_in addrs[2]; 1020 struct inpcb *inp; 1021 int cpu; 1022 int error, s; 1023 1024 error = suser(req->td); 1025 if (error != 0) 1026 return (error); 1027 error = SYSCTL_IN(req, addrs, sizeof addrs); 1028 if (error != 0) 1029 return (error); 1030 s = splnet(); 1031 1032 cpu = tcp_addrcpu(addrs[1].sin_addr.s_addr, addrs[1].sin_port, 1033 addrs[0].sin_addr.s_addr, addrs[0].sin_port); 1034 inp = in_pcblookup_hash(&tcbinfo[cpu], addrs[1].sin_addr, 1035 addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1036 if (inp == NULL || inp->inp_socket == NULL) { 1037 error = ENOENT; 1038 goto out; 1039 } 1040 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred)); 1041 out: 1042 splx(s); 1043 return (error); 1044 } 1045 1046 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW), 1047 0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection"); 1048 1049 #ifdef INET6 1050 static int 1051 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1052 { 1053 struct sockaddr_in6 addrs[2]; 1054 struct inpcb *inp; 1055 int error, s; 1056 boolean_t mapped = FALSE; 1057 1058 error = suser(req->td); 1059 if (error != 0) 1060 return (error); 1061 error = SYSCTL_IN(req, addrs, sizeof addrs); 1062 if (error != 0) 1063 return (error); 1064 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1065 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1066 mapped = TRUE; 1067 else 1068 return (EINVAL); 1069 } 1070 s = splnet(); 1071 if (mapped) { 1072 inp = in_pcblookup_hash(&tcbinfo[0], 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[0], 1080 &addrs[1].sin6_addr, addrs[1].sin6_port, 1081 &addrs[0].sin6_addr, addrs[0].sin6_port, 1082 0, NULL); 1083 } 1084 if (inp == NULL || inp->inp_socket == NULL) { 1085 error = ENOENT; 1086 goto out; 1087 } 1088 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred)); 1089 out: 1090 splx(s); 1091 return (error); 1092 } 1093 1094 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW), 1095 0, 0, 1096 tcp6_getcred, "S,ucred", "Get the ucred of a TCP6 connection"); 1097 #endif 1098 1099 void 1100 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1101 { 1102 struct ip *ip = vip; 1103 struct tcphdr *th; 1104 struct in_addr faddr; 1105 struct inpcb *inp; 1106 struct tcpcb *tp; 1107 void (*notify)(struct inpcb *, int) = tcp_notify; 1108 tcp_seq icmp_seq; 1109 int cpu; 1110 int s; 1111 1112 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1113 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1114 return; 1115 1116 if (cmd == PRC_QUENCH) 1117 notify = tcp_quench; 1118 else if (icmp_may_rst && 1119 (cmd == PRC_UNREACH_ADMIN_PROHIB || cmd == PRC_UNREACH_PORT || 1120 cmd == PRC_TIMXCEED_INTRANS) && 1121 ip != NULL) 1122 notify = tcp_drop_syn_sent; 1123 else if (cmd == PRC_MSGSIZE) 1124 notify = tcp_mtudisc; 1125 else if (PRC_IS_REDIRECT(cmd)) { 1126 ip = NULL; 1127 notify = in_rtchange; 1128 } else if (cmd == PRC_HOSTDEAD) 1129 ip = NULL; 1130 else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0) 1131 return; 1132 if (ip != NULL) { 1133 s = splnet(); 1134 th = (struct tcphdr *)((caddr_t)ip + 1135 (IP_VHL_HL(ip->ip_vhl) << 2)); 1136 cpu = tcp_addrcpu(faddr.s_addr, th->th_dport, 1137 ip->ip_src.s_addr, th->th_sport); 1138 inp = in_pcblookup_hash(&tcbinfo[cpu], faddr, th->th_dport, 1139 ip->ip_src, th->th_sport, 0, NULL); 1140 if ((inp != NULL) && (inp->inp_socket != NULL)) { 1141 icmp_seq = htonl(th->th_seq); 1142 tp = intotcpcb(inp); 1143 if (SEQ_GEQ(icmp_seq, tp->snd_una) && 1144 SEQ_LT(icmp_seq, tp->snd_max)) 1145 (*notify)(inp, inetctlerrmap[cmd]); 1146 } else { 1147 struct in_conninfo inc; 1148 1149 inc.inc_fport = th->th_dport; 1150 inc.inc_lport = th->th_sport; 1151 inc.inc_faddr = faddr; 1152 inc.inc_laddr = ip->ip_src; 1153 #ifdef INET6 1154 inc.inc_isipv6 = 0; 1155 #endif 1156 syncache_unreach(&inc, th); 1157 } 1158 splx(s); 1159 } else { 1160 for (cpu = 0; cpu < ncpus2; cpu++) { 1161 in_pcbnotifyall(&tcbinfo[cpu].listhead, faddr, 1162 inetctlerrmap[cmd], notify); 1163 } 1164 } 1165 } 1166 1167 #ifdef INET6 1168 void 1169 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1170 { 1171 struct tcphdr th; 1172 void (*notify) (struct inpcb *, int) = tcp_notify; 1173 struct ip6_hdr *ip6; 1174 struct mbuf *m; 1175 struct ip6ctlparam *ip6cp = NULL; 1176 const struct sockaddr_in6 *sa6_src = NULL; 1177 int off; 1178 struct tcp_portonly { 1179 u_int16_t th_sport; 1180 u_int16_t th_dport; 1181 } *thp; 1182 1183 if (sa->sa_family != AF_INET6 || 1184 sa->sa_len != sizeof(struct sockaddr_in6)) 1185 return; 1186 1187 if (cmd == PRC_QUENCH) 1188 notify = tcp_quench; 1189 else if (cmd == PRC_MSGSIZE) 1190 notify = tcp_mtudisc; 1191 else if (!PRC_IS_REDIRECT(cmd) && 1192 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1193 return; 1194 1195 /* if the parameter is from icmp6, decode it. */ 1196 if (d != NULL) { 1197 ip6cp = (struct ip6ctlparam *)d; 1198 m = ip6cp->ip6c_m; 1199 ip6 = ip6cp->ip6c_ip6; 1200 off = ip6cp->ip6c_off; 1201 sa6_src = ip6cp->ip6c_src; 1202 } else { 1203 m = NULL; 1204 ip6 = NULL; 1205 off = 0; /* fool gcc */ 1206 sa6_src = &sa6_any; 1207 } 1208 1209 if (ip6 != NULL) { 1210 struct in_conninfo inc; 1211 /* 1212 * XXX: We assume that when IPV6 is non NULL, 1213 * M and OFF are valid. 1214 */ 1215 1216 /* check if we can safely examine src and dst ports */ 1217 if (m->m_pkthdr.len < off + sizeof *thp) 1218 return; 1219 1220 bzero(&th, sizeof th); 1221 m_copydata(m, off, sizeof *thp, (caddr_t)&th); 1222 1223 in6_pcbnotify(&tcbinfo[0].listhead, sa, th.th_dport, 1224 (struct sockaddr *)ip6cp->ip6c_src, 1225 th.th_sport, cmd, notify); 1226 1227 inc.inc_fport = th.th_dport; 1228 inc.inc_lport = th.th_sport; 1229 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1230 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1231 inc.inc_isipv6 = 1; 1232 syncache_unreach(&inc, &th); 1233 } else 1234 in6_pcbnotify(&tcbinfo[0].listhead, sa, 0, 1235 (const struct sockaddr *)sa6_src, 0, cmd, notify); 1236 } 1237 #endif 1238 1239 /* 1240 * Following is where TCP initial sequence number generation occurs. 1241 * 1242 * There are two places where we must use initial sequence numbers: 1243 * 1. In SYN-ACK packets. 1244 * 2. In SYN packets. 1245 * 1246 * All ISNs for SYN-ACK packets are generated by the syncache. See 1247 * tcp_syncache.c for details. 1248 * 1249 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1250 * depends on this property. In addition, these ISNs should be 1251 * unguessable so as to prevent connection hijacking. To satisfy 1252 * the requirements of this situation, the algorithm outlined in 1253 * RFC 1948 is used to generate sequence numbers. 1254 * 1255 * Implementation details: 1256 * 1257 * Time is based off the system timer, and is corrected so that it 1258 * increases by one megabyte per second. This allows for proper 1259 * recycling on high speed LANs while still leaving over an hour 1260 * before rollover. 1261 * 1262 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1263 * between seeding of isn_secret. This is normally set to zero, 1264 * as reseeding should not be necessary. 1265 * 1266 */ 1267 1268 #define ISN_BYTES_PER_SECOND 1048576 1269 1270 u_char isn_secret[32]; 1271 int isn_last_reseed; 1272 MD5_CTX isn_ctx; 1273 1274 tcp_seq 1275 tcp_new_isn(struct tcpcb *tp) 1276 { 1277 u_int32_t md5_buffer[4]; 1278 tcp_seq new_isn; 1279 1280 /* Seed if this is the first use, reseed if requested. */ 1281 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1282 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1283 < (u_int)ticks))) { 1284 read_random_unlimited(&isn_secret, sizeof isn_secret); 1285 isn_last_reseed = ticks; 1286 } 1287 1288 /* Compute the md5 hash and return the ISN. */ 1289 MD5Init(&isn_ctx); 1290 MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_fport, sizeof(u_short)); 1291 MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_lport, sizeof(u_short)); 1292 #ifdef INET6 1293 if (tp->t_inpcb->inp_vflag & INP_IPV6) { 1294 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1295 sizeof(struct in6_addr)); 1296 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1297 sizeof(struct in6_addr)); 1298 } else 1299 #endif 1300 { 1301 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1302 sizeof(struct in_addr)); 1303 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1304 sizeof(struct in_addr)); 1305 } 1306 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1307 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1308 new_isn = (tcp_seq) md5_buffer[0]; 1309 new_isn += ticks * (ISN_BYTES_PER_SECOND / hz); 1310 return (new_isn); 1311 } 1312 1313 /* 1314 * When a source quench is received, close congestion window 1315 * to one segment. We will gradually open it again as we proceed. 1316 */ 1317 void 1318 tcp_quench(struct inpcb *inp, int errno) 1319 { 1320 struct tcpcb *tp = intotcpcb(inp); 1321 1322 if (tp != NULL) 1323 tp->snd_cwnd = tp->t_maxseg; 1324 } 1325 1326 /* 1327 * When a specific ICMP unreachable message is received and the 1328 * connection state is SYN-SENT, drop the connection. This behavior 1329 * is controlled by the icmp_may_rst sysctl. 1330 */ 1331 void 1332 tcp_drop_syn_sent(struct inpcb *inp, int errno) 1333 { 1334 struct tcpcb *tp = intotcpcb(inp); 1335 1336 if ((tp != NULL) && (tp->t_state == TCPS_SYN_SENT)) 1337 tcp_drop(tp, errno); 1338 } 1339 1340 /* 1341 * When `need fragmentation' ICMP is received, update our idea of the MSS 1342 * based on the new value in the route. Also nudge TCP to send something, 1343 * since we know the packet we just sent was dropped. 1344 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1345 */ 1346 void 1347 tcp_mtudisc(struct inpcb *inp, int errno) 1348 { 1349 struct tcpcb *tp = intotcpcb(inp); 1350 struct rtentry *rt; 1351 struct rmxp_tao *taop; 1352 struct socket *so = inp->inp_socket; 1353 int offered; 1354 int mss; 1355 #ifdef INET6 1356 boolean_t isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0); 1357 #else 1358 const boolean_t isipv6 = FALSE; 1359 #endif 1360 1361 if (tp != NULL) { 1362 if (isipv6) 1363 rt = tcp_rtlookup6(&inp->inp_inc); 1364 else 1365 rt = tcp_rtlookup(&inp->inp_inc); 1366 if (rt == NULL || rt->rt_rmx.rmx_mtu == 0) { 1367 tp->t_maxopd = tp->t_maxseg = 1368 isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 1369 return; 1370 } 1371 taop = rmx_taop(rt->rt_rmx); 1372 offered = taop->tao_mssopt; 1373 mss = rt->rt_rmx.rmx_mtu - 1374 (isipv6 ? 1375 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1376 sizeof(struct tcpiphdr)); 1377 1378 if (offered != 0) 1379 mss = min(mss, offered); 1380 /* 1381 * XXX - The above conditional probably violates the TCP 1382 * spec. The problem is that, since we don't know the 1383 * other end's MSS, we are supposed to use a conservative 1384 * default. But, if we do that, then MTU discovery will 1385 * never actually take place, because the conservative 1386 * default is much less than the MTUs typically seen 1387 * on the Internet today. For the moment, we'll sweep 1388 * this under the carpet. 1389 * 1390 * The conservative default might not actually be a problem 1391 * if the only case this occurs is when sending an initial 1392 * SYN with options and data to a host we've never talked 1393 * to before. Then, they will reply with an MSS value which 1394 * will get recorded and the new parameters should get 1395 * recomputed. For Further Study. 1396 */ 1397 if (tp->t_maxopd <= mss) 1398 return; 1399 tp->t_maxopd = mss; 1400 1401 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP && 1402 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1403 mss -= TCPOLEN_TSTAMP_APPA; 1404 if ((tp->t_flags & (TF_REQ_CC | TF_NOOPT)) == TF_REQ_CC && 1405 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC) 1406 mss -= TCPOLEN_CC_APPA; 1407 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1408 if (mss > MCLBYTES) 1409 mss &= ~(MCLBYTES - 1); 1410 #else 1411 if (mss > MCLBYTES) 1412 mss = mss / MCLBYTES * MCLBYTES; 1413 #endif 1414 if (so->so_snd.sb_hiwat < mss) 1415 mss = so->so_snd.sb_hiwat; 1416 1417 tp->t_maxseg = mss; 1418 1419 tcpstat.tcps_mturesent++; 1420 tp->t_rtttime = 0; 1421 tp->snd_nxt = tp->snd_una; 1422 tcp_output(tp); 1423 } 1424 } 1425 1426 /* 1427 * Look-up the routing entry to the peer of this inpcb. If no route 1428 * is found and it cannot be allocated the return NULL. This routine 1429 * is called by TCP routines that access the rmx structure and by tcp_mss 1430 * to get the interface MTU. 1431 */ 1432 struct rtentry * 1433 tcp_rtlookup(struct in_conninfo *inc) 1434 { 1435 struct route *ro; 1436 struct rtentry *rt; 1437 1438 ro = &inc->inc_route; 1439 rt = ro->ro_rt; 1440 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 1441 /* No route yet, so try to acquire one */ 1442 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1443 ro->ro_dst.sa_family = AF_INET; 1444 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 1445 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1446 inc->inc_faddr; 1447 rtalloc(ro); 1448 rt = ro->ro_rt; 1449 } 1450 } 1451 return (rt); 1452 } 1453 1454 #ifdef INET6 1455 struct rtentry * 1456 tcp_rtlookup6(struct in_conninfo *inc) 1457 { 1458 struct route_in6 *ro6; 1459 struct rtentry *rt; 1460 1461 ro6 = &inc->inc6_route; 1462 rt = ro6->ro_rt; 1463 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 1464 /* No route yet, so try to acquire one */ 1465 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1466 ro6->ro_dst.sin6_family = AF_INET6; 1467 ro6->ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1468 ro6->ro_dst.sin6_addr = inc->inc6_faddr; 1469 rtalloc((struct route *)ro6); 1470 rt = ro6->ro_rt; 1471 } 1472 } 1473 return (rt); 1474 } 1475 #endif 1476 1477 #ifdef IPSEC 1478 /* compute ESP/AH header size for TCP, including outer IP header. */ 1479 size_t 1480 ipsec_hdrsiz_tcp(struct tcpcb *tp) 1481 { 1482 struct inpcb *inp; 1483 struct mbuf *m; 1484 size_t hdrsiz; 1485 struct ip *ip; 1486 struct tcphdr *th; 1487 1488 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1489 return (0); 1490 MGETHDR(m, M_DONTWAIT, MT_DATA); 1491 if (!m) 1492 return (0); 1493 1494 #ifdef INET6 1495 if (inp->inp_vflag & INP_IPV6) { 1496 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 1497 1498 th = (struct tcphdr *)(ip6 + 1); 1499 m->m_pkthdr.len = m->m_len = 1500 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1501 tcp_fillheaders(tp, ip6, th); 1502 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1503 } else 1504 #endif 1505 { 1506 ip = mtod(m, struct ip *); 1507 th = (struct tcphdr *)(ip + 1); 1508 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1509 tcp_fillheaders(tp, ip, th); 1510 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1511 } 1512 1513 m_free(m); 1514 return (hdrsiz); 1515 } 1516 #endif 1517 1518 /* 1519 * Return a pointer to the cached information about the remote host. 1520 * The cached information is stored in the protocol specific part of 1521 * the route metrics. 1522 */ 1523 struct rmxp_tao * 1524 tcp_gettaocache(struct in_conninfo *inc) 1525 { 1526 struct rtentry *rt; 1527 1528 #ifdef INET6 1529 if (inc->inc_isipv6) 1530 rt = tcp_rtlookup6(inc); 1531 else 1532 #endif 1533 rt = tcp_rtlookup(inc); 1534 1535 /* Make sure this is a host route and is up. */ 1536 if (rt == NULL || 1537 (rt->rt_flags & (RTF_UP | RTF_HOST)) != (RTF_UP | RTF_HOST)) 1538 return (NULL); 1539 1540 return (rmx_taop(rt->rt_rmx)); 1541 } 1542 1543 /* 1544 * Clear all the TAO cache entries, called from tcp_init. 1545 * 1546 * XXX 1547 * This routine is just an empty one, because we assume that the routing 1548 * routing tables are initialized at the same time when TCP, so there is 1549 * nothing in the cache left over. 1550 */ 1551 static void 1552 tcp_cleartaocache() 1553 { 1554 } 1555 1556 /* 1557 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1558 * 1559 * This code attempts to calculate the bandwidth-delay product as a 1560 * means of determining the optimal window size to maximize bandwidth, 1561 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1562 * routers. This code also does a fairly good job keeping RTTs in check 1563 * across slow links like modems. We implement an algorithm which is very 1564 * similar (but not meant to be) TCP/Vegas. The code operates on the 1565 * transmitter side of a TCP connection and so only effects the transmit 1566 * side of the connection. 1567 * 1568 * BACKGROUND: TCP makes no provision for the management of buffer space 1569 * at the end points or at the intermediate routers and switches. A TCP 1570 * stream, whether using NewReno or not, will eventually buffer as 1571 * many packets as it is able and the only reason this typically works is 1572 * due to the fairly small default buffers made available for a connection 1573 * (typicaly 16K or 32K). As machines use larger windows and/or window 1574 * scaling it is now fairly easy for even a single TCP connection to blow-out 1575 * all available buffer space not only on the local interface, but on 1576 * intermediate routers and switches as well. NewReno makes a misguided 1577 * attempt to 'solve' this problem by waiting for an actual failure to occur, 1578 * then backing off, then steadily increasing the window again until another 1579 * failure occurs, ad-infinitum. This results in terrible oscillation that 1580 * is only made worse as network loads increase and the idea of intentionally 1581 * blowing out network buffers is, frankly, a terrible way to manage network 1582 * resources. 1583 * 1584 * It is far better to limit the transmit window prior to the failure 1585 * condition being achieved. There are two general ways to do this: First 1586 * you can 'scan' through different transmit window sizes and locate the 1587 * point where the RTT stops increasing, indicating that you have filled the 1588 * pipe, then scan backwards until you note that RTT stops decreasing, then 1589 * repeat ad-infinitum. This method works in principle but has severe 1590 * implementation issues due to RTT variances, timer granularity, and 1591 * instability in the algorithm which can lead to many false positives and 1592 * create oscillations as well as interact badly with other TCP streams 1593 * implementing the same algorithm. 1594 * 1595 * The second method is to limit the window to the bandwidth delay product 1596 * of the link. This is the method we implement. RTT variances and our 1597 * own manipulation of the congestion window, bwnd, can potentially 1598 * destabilize the algorithm. For this reason we have to stabilize the 1599 * elements used to calculate the window. We do this by using the minimum 1600 * observed RTT, the long term average of the observed bandwidth, and 1601 * by adding two segments worth of slop. It isn't perfect but it is able 1602 * to react to changing conditions and gives us a very stable basis on 1603 * which to extend the algorithm. 1604 */ 1605 void 1606 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 1607 { 1608 u_long bw; 1609 u_long bwnd; 1610 int save_ticks; 1611 1612 /* 1613 * If inflight_enable is disabled in the middle of a tcp connection, 1614 * make sure snd_bwnd is effectively disabled. 1615 */ 1616 if (!tcp_inflight_enable) { 1617 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1618 tp->snd_bandwidth = 0; 1619 return; 1620 } 1621 1622 /* 1623 * Figure out the bandwidth. Due to the tick granularity this 1624 * is a very rough number and it MUST be averaged over a fairly 1625 * long period of time. XXX we need to take into account a link 1626 * that is not using all available bandwidth, but for now our 1627 * slop will ramp us up if this case occurs and the bandwidth later 1628 * increases. 1629 * 1630 * Note: if ticks rollover 'bw' may wind up negative. We must 1631 * effectively reset t_bw_rtttime for this case. 1632 */ 1633 save_ticks = ticks; 1634 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 1635 return; 1636 1637 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 1638 (save_ticks - tp->t_bw_rtttime); 1639 tp->t_bw_rtttime = save_ticks; 1640 tp->t_bw_rtseq = ack_seq; 1641 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 1642 return; 1643 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 1644 1645 tp->snd_bandwidth = bw; 1646 1647 /* 1648 * Calculate the semi-static bandwidth delay product, plus two maximal 1649 * segments. The additional slop puts us squarely in the sweet 1650 * spot and also handles the bandwidth run-up case. Without the 1651 * slop we could be locking ourselves into a lower bandwidth. 1652 * 1653 * Situations Handled: 1654 * (1) Prevents over-queueing of packets on LANs, especially on 1655 * high speed LANs, allowing larger TCP buffers to be 1656 * specified, and also does a good job preventing 1657 * over-queueing of packets over choke points like modems 1658 * (at least for the transmit side). 1659 * 1660 * (2) Is able to handle changing network loads (bandwidth 1661 * drops so bwnd drops, bandwidth increases so bwnd 1662 * increases). 1663 * 1664 * (3) Theoretically should stabilize in the face of multiple 1665 * connections implementing the same algorithm (this may need 1666 * a little work). 1667 * 1668 * (4) Stability value (defaults to 20 = 2 maximal packets) can 1669 * be adjusted with a sysctl but typically only needs to be on 1670 * very slow connections. A value no smaller then 5 should 1671 * be used, but only reduce this default if you have no other 1672 * choice. 1673 */ 1674 1675 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 1676 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + 1677 tcp_inflight_stab * (int)tp->t_maxseg / 10; 1678 #undef USERTT 1679 1680 if (tcp_inflight_debug > 0) { 1681 static int ltime; 1682 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 1683 ltime = ticks; 1684 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 1685 tp, bw, tp->t_rttbest, tp->t_srtt, bwnd); 1686 } 1687 } 1688 if ((long)bwnd < tcp_inflight_min) 1689 bwnd = tcp_inflight_min; 1690 if (bwnd > tcp_inflight_max) 1691 bwnd = tcp_inflight_max; 1692 if ((long)bwnd < tp->t_maxseg * 2) 1693 bwnd = tp->t_maxseg * 2; 1694 tp->snd_bwnd = bwnd; 1695 } 1696