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 1.24 2004/04/13 07:10:34 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 #ifdef INET6 71 #include <netinet/ip6.h> 72 #endif 73 #include <netinet/in_pcb.h> 74 #ifdef INET6 75 #include <netinet6/in6_pcb.h> 76 #endif 77 #include <netinet/in_var.h> 78 #include <netinet/ip_var.h> 79 #ifdef INET6 80 #include <netinet6/ip6_var.h> 81 #endif 82 #include <netinet/tcp.h> 83 #include <netinet/tcp_fsm.h> 84 #include <netinet/tcp_seq.h> 85 #include <netinet/tcp_timer.h> 86 #include <netinet/tcp_var.h> 87 #ifdef INET6 88 #include <netinet6/tcp6_var.h> 89 #endif 90 #include <netinet/tcpip.h> 91 #ifdef TCPDEBUG 92 #include <netinet/tcp_debug.h> 93 #endif 94 #include <netinet6/ip6protosw.h> 95 96 #ifdef IPSEC 97 #include <netinet6/ipsec.h> 98 #ifdef INET6 99 #include <netinet6/ipsec6.h> 100 #endif 101 #endif /*IPSEC*/ 102 103 #ifdef FAST_IPSEC 104 #include <netipsec/ipsec.h> 105 #ifdef INET6 106 #include <netipsec/ipsec6.h> 107 #endif 108 #define IPSEC 109 #endif /*FAST_IPSEC*/ 110 111 #include <sys/md5.h> 112 113 #include <sys/msgport2.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 #if 0 127 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 128 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 129 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 130 #endif 131 132 int tcp_do_rfc1323 = 1; 133 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 134 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 135 136 int tcp_do_rfc1644 = 0; 137 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW, 138 &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions"); 139 140 static int tcp_tcbhashsize = 0; 141 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD, 142 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 143 144 static int do_tcpdrain = 1; 145 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 146 "Enable tcp_drain routine for extra help when low on mbufs"); 147 148 /* XXX JH */ 149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 150 &tcbinfo[0].ipi_count, 0, "Number of active PCBs"); 151 152 static int icmp_may_rst = 1; 153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 154 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 155 156 static int tcp_isn_reseed_interval = 0; 157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 158 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 159 160 /* 161 * TCP bandwidth limiting sysctls. Note that the default lower bound of 162 * 1024 exists only for debugging. A good production default would be 163 * something like 6100. 164 */ 165 static int tcp_inflight_enable = 0; 166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_enable, CTLFLAG_RW, 167 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 168 169 static int tcp_inflight_debug = 0; 170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_debug, CTLFLAG_RW, 171 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 172 173 static int tcp_inflight_min = 6144; 174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_min, CTLFLAG_RW, 175 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 176 177 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_max, CTLFLAG_RW, 179 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 180 181 static int tcp_inflight_stab = 20; 182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_stab, CTLFLAG_RW, 183 &tcp_inflight_stab, 0, "Slop in maximal packets / 10 (20 = 2 packets)"); 184 185 static void tcp_cleartaocache (void); 186 static void tcp_notify (struct inpcb *, int); 187 188 struct tcp_stats tcpstats_ary[MAXCPU]; 189 #ifdef SMP 190 static int 191 sysctl_tcpstats(SYSCTL_HANDLER_ARGS) 192 { 193 int cpu, error; 194 195 for (cpu = error = 0; cpu < ncpus; ++cpu) { 196 if ((error = SYSCTL_OUT(req, (void *)&tcpstats_ary[cpu], 197 sizeof(struct tcp_stats)))) 198 break; 199 if ((error = SYSCTL_IN(req, (void *)&tcpstats_ary[cpu], 200 sizeof(struct tcp_stats)))) 201 break; 202 } 203 204 return (error); 205 } 206 SYSCTL_PROC(_net_inet_tcp, TCPCTL_STATS, stats, CTLTYPE_OPAQUE|CTLFLAG_RW, 207 0, 0, sysctl_tcpstats, "S,tcp_stats", 208 "TCP statistics (struct tcp_stats, netinet/tcp_stats.h)"); 209 #else /* !SMP */ 210 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 211 &tcpstat , tcp_stats, 212 "TCP statistics (struct tcp_stats, netinet/tcp_stats.h)"); 213 #endif 214 215 /* 216 * Target size of TCP PCB hash tables. Must be a power of two. 217 * 218 * Note that this can be overridden by the kernel environment 219 * variable net.inet.tcp.tcbhashsize 220 */ 221 #ifndef TCBHASHSIZE 222 #define TCBHASHSIZE 512 223 #endif 224 225 /* 226 * This is the actual shape of what we allocate using the zone 227 * allocator. Doing it this way allows us to protect both structures 228 * using the same generation count, and also eliminates the overhead 229 * of allocating tcpcbs separately. By hiding the structure here, 230 * we avoid changing most of the rest of the code (although it needs 231 * to be changed, eventually, for greater efficiency). 232 */ 233 #define ALIGNMENT 32 234 #define ALIGNM1 (ALIGNMENT - 1) 235 struct inp_tp { 236 union { 237 struct inpcb inp; 238 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1]; 239 } inp_tp_u; 240 struct tcpcb tcb; 241 struct callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl; 242 struct callout inp_tp_delack; 243 }; 244 #undef ALIGNMENT 245 #undef ALIGNM1 246 247 /* 248 * Tcp initialization 249 */ 250 void 251 tcp_init() 252 { 253 struct inpcbporthead *porthashbase; 254 u_long porthashmask; 255 struct inpcbhead *wildcardhashbase; 256 u_long wildcardhashmask; 257 struct vm_zone *ipi_zone; 258 int hashsize = TCBHASHSIZE; 259 int cpu; 260 261 tcp_ccgen = 1; 262 tcp_cleartaocache(); 263 264 tcp_delacktime = TCPTV_DELACK; 265 tcp_keepinit = TCPTV_KEEP_INIT; 266 tcp_keepidle = TCPTV_KEEP_IDLE; 267 tcp_keepintvl = TCPTV_KEEPINTVL; 268 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 269 tcp_msl = TCPTV_MSL; 270 tcp_rexmit_min = TCPTV_MIN; 271 tcp_rexmit_slop = TCPTV_CPU_VAR; 272 273 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 274 if (!powerof2(hashsize)) { 275 printf("WARNING: TCB hash size not a power of 2\n"); 276 hashsize = 512; /* safe default */ 277 } 278 tcp_tcbhashsize = hashsize; 279 porthashbase = hashinit(hashsize, M_PCB, &porthashmask); 280 wildcardhashbase = hashinit(hashsize, M_PCB, &wildcardhashmask); 281 ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets, 282 ZONE_INTERRUPT, 0); 283 284 for (cpu = 0; cpu < ncpus2; cpu++) { 285 LIST_INIT(&tcbinfo[cpu].listhead); 286 tcbinfo[cpu].hashbase = hashinit(hashsize, M_PCB, 287 &tcbinfo[cpu].hashmask); 288 tcbinfo[cpu].porthashbase = porthashbase; 289 tcbinfo[cpu].porthashmask = porthashmask; 290 tcbinfo[cpu].wildcardhashbase = wildcardhashbase; 291 tcbinfo[cpu].wildcardhashmask = wildcardhashmask; 292 tcbinfo[cpu].ipi_zone = ipi_zone; 293 } 294 295 tcp_reass_maxseg = nmbclusters / 16; 296 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 297 &tcp_reass_maxseg); 298 299 #ifdef INET6 300 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 301 #else /* INET6 */ 302 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 303 #endif /* INET6 */ 304 if (max_protohdr < TCP_MINPROTOHDR) 305 max_protohdr = TCP_MINPROTOHDR; 306 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 307 panic("tcp_init"); 308 #undef TCP_MINPROTOHDR 309 310 /* 311 * Initialize TCP statistics. 312 * 313 * It is layed out as an array which is has one element for UP, 314 * and SMP_MAXCPU elements for SMP. This allows us to retain 315 * the access mechanism from userland for both UP and SMP. 316 */ 317 #ifdef SMP 318 for (cpu = 0; cpu < ncpus; ++cpu) { 319 bzero(&tcpstats_ary[cpu], sizeof(struct tcp_stats)); 320 } 321 #else 322 bzero(&tcpstat, sizeof(struct tcp_stats)); 323 #endif 324 325 syncache_init(); 326 tcp_thread_init(); 327 } 328 329 /* 330 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 331 * tcp_template used to store this data in mbufs, but we now recopy it out 332 * of the tcpcb each time to conserve mbufs. 333 */ 334 void 335 tcp_fillheaders(tp, ip_ptr, tcp_ptr) 336 struct tcpcb *tp; 337 void *ip_ptr; 338 void *tcp_ptr; 339 { 340 struct inpcb *inp = tp->t_inpcb; 341 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr; 342 343 #ifdef INET6 344 if ((inp->inp_vflag & INP_IPV6) != 0) { 345 struct ip6_hdr *ip6; 346 347 ip6 = (struct ip6_hdr *)ip_ptr; 348 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 349 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 350 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 351 (IPV6_VERSION & IPV6_VERSION_MASK); 352 ip6->ip6_nxt = IPPROTO_TCP; 353 ip6->ip6_plen = sizeof(struct tcphdr); 354 ip6->ip6_src = inp->in6p_laddr; 355 ip6->ip6_dst = inp->in6p_faddr; 356 tcp_hdr->th_sum = 0; 357 } else 358 #endif 359 { 360 struct ip *ip = (struct ip *) ip_ptr; 361 362 ip->ip_vhl = IP_VHL_BORING; 363 ip->ip_tos = 0; 364 ip->ip_len = 0; 365 ip->ip_id = 0; 366 ip->ip_off = 0; 367 ip->ip_ttl = 0; 368 ip->ip_sum = 0; 369 ip->ip_p = IPPROTO_TCP; 370 ip->ip_src = inp->inp_laddr; 371 ip->ip_dst = inp->inp_faddr; 372 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 373 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 374 } 375 376 tcp_hdr->th_sport = inp->inp_lport; 377 tcp_hdr->th_dport = inp->inp_fport; 378 tcp_hdr->th_seq = 0; 379 tcp_hdr->th_ack = 0; 380 tcp_hdr->th_x2 = 0; 381 tcp_hdr->th_off = 5; 382 tcp_hdr->th_flags = 0; 383 tcp_hdr->th_win = 0; 384 tcp_hdr->th_urp = 0; 385 } 386 387 /* 388 * Create template to be used to send tcp packets on a connection. 389 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 390 * use for this function is in keepalives, which use tcp_respond. 391 */ 392 struct tcptemp * 393 tcp_maketemplate(tp) 394 struct tcpcb *tp; 395 { 396 struct mbuf *m; 397 struct tcptemp *n; 398 399 m = m_get(M_DONTWAIT, MT_HEADER); 400 if (m == NULL) 401 return (0); 402 m->m_len = sizeof(struct tcptemp); 403 n = mtod(m, struct tcptemp *); 404 405 tcp_fillheaders(tp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 406 return (n); 407 } 408 409 /* 410 * Send a single message to the TCP at address specified by 411 * the given TCP/IP header. If m == 0, then we make a copy 412 * of the tcpiphdr at ti and send directly to the addressed host. 413 * This is used to force keep alive messages out using the TCP 414 * template for a connection. If flags are given then we send 415 * a message back to the TCP which originated the * segment ti, 416 * and discard the mbuf containing it and any other attached mbufs. 417 * 418 * In any case the ack and sequence number of the transmitted 419 * segment are as specified by the parameters. 420 * 421 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 422 */ 423 void 424 tcp_respond(tp, ipgen, th, m, ack, seq, flags) 425 struct tcpcb *tp; 426 void *ipgen; 427 struct tcphdr *th; 428 struct mbuf *m; 429 tcp_seq ack, seq; 430 int flags; 431 { 432 int tlen; 433 int win = 0; 434 struct route *ro = 0; 435 struct route sro; 436 struct ip *ip; 437 struct tcphdr *nth; 438 #ifdef INET6 439 struct route_in6 *ro6 = 0; 440 struct route_in6 sro6; 441 struct ip6_hdr *ip6; 442 int isipv6; 443 #endif /* INET6 */ 444 int ipflags = 0; 445 446 #ifdef INET6 447 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6; 448 ip6 = ipgen; 449 #endif /* INET6 */ 450 ip = ipgen; 451 452 if (tp) { 453 if (!(flags & TH_RST)) { 454 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 455 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 456 win = (long)TCP_MAXWIN << tp->rcv_scale; 457 } 458 #ifdef INET6 459 if (isipv6) 460 ro6 = &tp->t_inpcb->in6p_route; 461 else 462 #endif /* INET6 */ 463 ro = &tp->t_inpcb->inp_route; 464 } else { 465 #ifdef INET6 466 if (isipv6) { 467 ro6 = &sro6; 468 bzero(ro6, sizeof *ro6); 469 } else 470 #endif /* INET6 */ 471 { 472 ro = &sro; 473 bzero(ro, sizeof *ro); 474 } 475 } 476 if (m == 0) { 477 m = m_gethdr(M_DONTWAIT, MT_HEADER); 478 if (m == NULL) 479 return; 480 tlen = 0; 481 m->m_data += max_linkhdr; 482 #ifdef INET6 483 if (isipv6) { 484 bcopy((caddr_t)ip6, mtod(m, caddr_t), 485 sizeof(struct ip6_hdr)); 486 ip6 = mtod(m, struct ip6_hdr *); 487 nth = (struct tcphdr *)(ip6 + 1); 488 } else 489 #endif /* INET6 */ 490 { 491 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 492 ip = mtod(m, struct ip *); 493 nth = (struct tcphdr *)(ip + 1); 494 } 495 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 496 flags = TH_ACK; 497 } else { 498 m_freem(m->m_next); 499 m->m_next = 0; 500 m->m_data = (caddr_t)ipgen; 501 /* m_len is set later */ 502 tlen = 0; 503 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 504 #ifdef INET6 505 if (isipv6) { 506 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 507 nth = (struct tcphdr *)(ip6 + 1); 508 } else 509 #endif /* INET6 */ 510 { 511 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 512 nth = (struct tcphdr *)(ip + 1); 513 } 514 if (th != nth) { 515 /* 516 * this is usually a case when an extension header 517 * exists between the IPv6 header and the 518 * TCP header. 519 */ 520 nth->th_sport = th->th_sport; 521 nth->th_dport = th->th_dport; 522 } 523 xchg(nth->th_dport, nth->th_sport, n_short); 524 #undef xchg 525 } 526 #ifdef INET6 527 if (isipv6) { 528 ip6->ip6_flow = 0; 529 ip6->ip6_vfc = IPV6_VERSION; 530 ip6->ip6_nxt = IPPROTO_TCP; 531 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 532 tlen)); 533 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 534 } else 535 #endif 536 { 537 tlen += sizeof (struct tcpiphdr); 538 ip->ip_len = tlen; 539 ip->ip_ttl = ip_defttl; 540 } 541 m->m_len = tlen; 542 m->m_pkthdr.len = tlen; 543 m->m_pkthdr.rcvif = (struct ifnet *) 0; 544 nth->th_seq = htonl(seq); 545 nth->th_ack = htonl(ack); 546 nth->th_x2 = 0; 547 nth->th_off = sizeof (struct tcphdr) >> 2; 548 nth->th_flags = flags; 549 if (tp) 550 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 551 else 552 nth->th_win = htons((u_short)win); 553 nth->th_urp = 0; 554 #ifdef INET6 555 if (isipv6) { 556 nth->th_sum = 0; 557 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 558 sizeof(struct ip6_hdr), 559 tlen - sizeof(struct ip6_hdr)); 560 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL, 561 ro6 && ro6->ro_rt ? 562 ro6->ro_rt->rt_ifp : 563 NULL); 564 } else 565 #endif /* INET6 */ 566 { 567 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 568 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 569 m->m_pkthdr.csum_flags = CSUM_TCP; 570 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 571 } 572 #ifdef TCPDEBUG 573 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 574 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 575 #endif 576 #ifdef INET6 577 if (isipv6) { 578 (void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL, 579 tp ? tp->t_inpcb : NULL); 580 if (ro6 == &sro6 && ro6->ro_rt) { 581 RTFREE(ro6->ro_rt); 582 ro6->ro_rt = NULL; 583 } 584 } else 585 #endif /* INET6 */ 586 { 587 (void) ip_output(m, NULL, ro, ipflags, NULL, tp ? tp->t_inpcb : NULL); 588 if (ro == &sro && ro->ro_rt) { 589 RTFREE(ro->ro_rt); 590 ro->ro_rt = NULL; 591 } 592 } 593 } 594 595 /* 596 * Create a new TCP control block, making an 597 * empty reassembly queue and hooking it to the argument 598 * protocol control block. The `inp' parameter must have 599 * come from the zone allocator set up in tcp_init(). 600 */ 601 struct tcpcb * 602 tcp_newtcpcb(inp) 603 struct inpcb *inp; 604 { 605 struct inp_tp *it; 606 struct tcpcb *tp; 607 #ifdef INET6 608 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 609 #endif /* INET6 */ 610 611 it = (struct inp_tp *)inp; 612 tp = &it->tcb; 613 bzero((char *) tp, sizeof(struct tcpcb)); 614 LIST_INIT(&tp->t_segq); 615 tp->t_maxseg = tp->t_maxopd = 616 #ifdef INET6 617 isipv6 ? tcp_v6mssdflt : 618 #endif /* INET6 */ 619 tcp_mssdflt; 620 621 /* Set up our timeouts. */ 622 callout_init(tp->tt_rexmt = &it->inp_tp_rexmt); 623 callout_init(tp->tt_persist = &it->inp_tp_persist); 624 callout_init(tp->tt_keep = &it->inp_tp_keep); 625 callout_init(tp->tt_2msl = &it->inp_tp_2msl); 626 callout_init(tp->tt_delack = &it->inp_tp_delack); 627 628 if (tcp_do_rfc1323) 629 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 630 if (tcp_do_rfc1644) 631 tp->t_flags |= TF_REQ_CC; 632 tp->t_inpcb = inp; /* XXX */ 633 /* 634 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 635 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 636 * reasonable initial retransmit time. 637 */ 638 tp->t_srtt = TCPTV_SRTTBASE; 639 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 640 tp->t_rttmin = tcp_rexmit_min; 641 tp->t_rxtcur = TCPTV_RTOBASE; 642 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 643 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 644 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 645 tp->t_rcvtime = ticks; 646 tp->t_bw_rtttime = ticks; 647 /* 648 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 649 * because the socket may be bound to an IPv6 wildcard address, 650 * which may match an IPv4-mapped IPv6 address. 651 */ 652 inp->inp_ip_ttl = ip_defttl; 653 inp->inp_ppcb = (caddr_t)tp; 654 return (tp); /* XXX */ 655 } 656 657 /* 658 * Drop a TCP connection, reporting 659 * the specified error. If connection is synchronized, 660 * then send a RST to peer. 661 */ 662 struct tcpcb * 663 tcp_drop(tp, errno) 664 struct tcpcb *tp; 665 int errno; 666 { 667 struct socket *so = tp->t_inpcb->inp_socket; 668 669 if (TCPS_HAVERCVDSYN(tp->t_state)) { 670 tp->t_state = TCPS_CLOSED; 671 (void) tcp_output(tp); 672 tcpstat.tcps_drops++; 673 } else 674 tcpstat.tcps_conndrops++; 675 if (errno == ETIMEDOUT && tp->t_softerror) 676 errno = tp->t_softerror; 677 so->so_error = errno; 678 return (tcp_close(tp)); 679 } 680 681 /* 682 * Close a TCP control block: 683 * discard all space held by the tcp 684 * discard internet protocol block 685 * wake up any sleepers 686 */ 687 struct tcpcb * 688 tcp_close(tp) 689 struct tcpcb *tp; 690 { 691 struct tseg_qent *q; 692 struct inpcb *inp = tp->t_inpcb; 693 struct socket *so = inp->inp_socket; 694 #ifdef INET6 695 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 696 #endif /* INET6 */ 697 struct rtentry *rt; 698 int dosavessthresh; 699 700 /* 701 * Make sure that all of our timers are stopped before we 702 * delete the PCB. 703 */ 704 callout_stop(tp->tt_rexmt); 705 callout_stop(tp->tt_persist); 706 callout_stop(tp->tt_keep); 707 callout_stop(tp->tt_2msl); 708 callout_stop(tp->tt_delack); 709 710 /* 711 * If we got enough samples through the srtt filter, 712 * save the rtt and rttvar in the routing entry. 713 * 'Enough' is arbitrarily defined as the 16 samples. 714 * 16 samples is enough for the srtt filter to converge 715 * to within 5% of the correct value; fewer samples and 716 * we could save a very bogus rtt. 717 * 718 * Don't update the default route's characteristics and don't 719 * update anything that the user "locked". 720 */ 721 if (tp->t_rttupdated >= 16) { 722 u_long i = 0; 723 #ifdef INET6 724 if (isipv6) { 725 struct sockaddr_in6 *sin6; 726 727 if ((rt = inp->in6p_route.ro_rt) == NULL) 728 goto no_valid_rt; 729 sin6 = (struct sockaddr_in6 *)rt_key(rt); 730 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 731 goto no_valid_rt; 732 } 733 else 734 #endif /* INET6 */ 735 if ((rt = inp->inp_route.ro_rt) == NULL || 736 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr 737 == INADDR_ANY) 738 goto no_valid_rt; 739 740 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 741 i = tp->t_srtt * 742 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 743 if (rt->rt_rmx.rmx_rtt && i) 744 /* 745 * filter this update to half the old & half 746 * the new values, converting scale. 747 * See route.h and tcp_var.h for a 748 * description of the scaling constants. 749 */ 750 rt->rt_rmx.rmx_rtt = 751 (rt->rt_rmx.rmx_rtt + i) / 2; 752 else 753 rt->rt_rmx.rmx_rtt = i; 754 tcpstat.tcps_cachedrtt++; 755 } 756 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 757 i = tp->t_rttvar * 758 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 759 if (rt->rt_rmx.rmx_rttvar && i) 760 rt->rt_rmx.rmx_rttvar = 761 (rt->rt_rmx.rmx_rttvar + i) / 2; 762 else 763 rt->rt_rmx.rmx_rttvar = i; 764 tcpstat.tcps_cachedrttvar++; 765 } 766 /* 767 * The old comment here said: 768 * update the pipelimit (ssthresh) if it has been updated 769 * already or if a pipesize was specified & the threshhold 770 * got below half the pipesize. I.e., wait for bad news 771 * before we start updating, then update on both good 772 * and bad news. 773 * 774 * But we want to save the ssthresh even if no pipesize is 775 * specified explicitly in the route, because such 776 * connections still have an implicit pipesize specified 777 * by the global tcp_sendspace. In the absence of a reliable 778 * way to calculate the pipesize, it will have to do. 779 */ 780 i = tp->snd_ssthresh; 781 if (rt->rt_rmx.rmx_sendpipe != 0) 782 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2); 783 else 784 dosavessthresh = (i < so->so_snd.sb_hiwat / 2); 785 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 786 i != 0 && rt->rt_rmx.rmx_ssthresh != 0) 787 || dosavessthresh) { 788 /* 789 * convert the limit from user data bytes to 790 * packets then to packet data bytes. 791 */ 792 i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 793 if (i < 2) 794 i = 2; 795 i *= (u_long)(tp->t_maxseg + 796 #ifdef INET6 797 (isipv6 ? sizeof (struct ip6_hdr) + 798 sizeof (struct tcphdr) : 799 #endif 800 sizeof (struct tcpiphdr) 801 #ifdef INET6 802 ) 803 #endif 804 ); 805 if (rt->rt_rmx.rmx_ssthresh) 806 rt->rt_rmx.rmx_ssthresh = 807 (rt->rt_rmx.rmx_ssthresh + i) / 2; 808 else 809 rt->rt_rmx.rmx_ssthresh = i; 810 tcpstat.tcps_cachedssthresh++; 811 } 812 } 813 no_valid_rt: 814 /* free the reassembly queue, if any */ 815 while((q = LIST_FIRST(&tp->t_segq)) != NULL) { 816 LIST_REMOVE(q, tqe_q); 817 m_freem(q->tqe_m); 818 FREE(q, M_TSEGQ); 819 tcp_reass_qsize--; 820 } 821 inp->inp_ppcb = NULL; 822 soisdisconnected(so); 823 #ifdef INET6 824 if (INP_CHECK_SOCKAF(so, AF_INET6)) 825 in6_pcbdetach(inp); 826 else 827 #endif /* INET6 */ 828 in_pcbdetach(inp); 829 tcpstat.tcps_closed++; 830 return ((struct tcpcb *)0); 831 } 832 833 static __inline void 834 tcp_drain_oncpu(struct inpcbhead *head) 835 { 836 struct inpcb *inpb; 837 struct tcpcb *tcpb; 838 struct tseg_qent *te; 839 840 LIST_FOREACH(inpb, head, inp_list) { 841 if ((tcpb = intotcpcb(inpb))) { 842 while ((te = LIST_FIRST(&tcpb->t_segq)) != NULL) { 843 LIST_REMOVE(te, tqe_q); 844 m_freem(te->tqe_m); 845 FREE(te, M_TSEGQ); 846 tcp_reass_qsize--; 847 } 848 } 849 } 850 } 851 852 #ifdef SMP 853 struct netmsg_tcp_drain { 854 struct lwkt_msg nm_lmsg; 855 netisr_fn_t nm_handler; 856 struct inpcbhead *nm_head; 857 }; 858 859 static void 860 tcp_drain_handler(struct netmsg *msg0) 861 { 862 struct netmsg_tcp_drain *nm = (struct netmsg_tcp_drain *)msg0; 863 864 tcp_drain_oncpu(nm->nm_head); 865 lwkt_replymsg(&msg0->nm_lmsg, 0); 866 } 867 #endif 868 869 void 870 tcp_drain() 871 { 872 #ifdef SMP 873 int cpu; 874 #endif 875 876 if (!do_tcpdrain) 877 return; 878 879 /* 880 * Walk the tcpbs, if existing, and flush the reassembly queue, 881 * if there is one... 882 * XXX: The "Net/3" implementation doesn't imply that the TCP 883 * reassembly queue should be flushed, but in a situation 884 * where we're really low on mbufs, this is potentially 885 * usefull. 886 */ 887 #ifdef SMP 888 for (cpu = 0; cpu < ncpus2; cpu++) { 889 struct netmsg_tcp_drain *msg; 890 891 if (cpu == mycpu->gd_cpuid) { 892 tcp_drain_oncpu(&tcbinfo[cpu].listhead); 893 } else { 894 msg = malloc(sizeof(struct netmsg_tcp_drain), 895 M_LWKTMSG, M_NOWAIT); 896 if (!msg) 897 continue; 898 lwkt_initmsg_rp(&msg->nm_lmsg, &netisr_afree_rport, 899 CMD_NETMSG_ONCPU); 900 msg->nm_handler = tcp_drain_handler; 901 msg->nm_head = &tcbinfo[cpu].listhead; 902 lwkt_sendmsg(tcp_cport(cpu), &msg->nm_lmsg); 903 } 904 } 905 #else 906 tcp_drain_oncpu(&tcbinfo[0].listhead); 907 #endif 908 } 909 910 /* 911 * Notify a tcp user of an asynchronous error; 912 * store error as soft error, but wake up user 913 * (for now, won't do anything until can select for soft error). 914 * 915 * Do not wake up user since there currently is no mechanism for 916 * reporting soft errors (yet - a kqueue filter may be added). 917 */ 918 static void 919 tcp_notify(inp, error) 920 struct inpcb *inp; 921 int error; 922 { 923 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 924 925 /* 926 * Ignore some errors if we are hooked up. 927 * If connection hasn't completed, has retransmitted several times, 928 * and receives a second error, give up now. This is better 929 * than waiting a long time to establish a connection that 930 * can never complete. 931 */ 932 if (tp->t_state == TCPS_ESTABLISHED && 933 (error == EHOSTUNREACH || error == ENETUNREACH || 934 error == EHOSTDOWN)) { 935 return; 936 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 937 tp->t_softerror) 938 tcp_drop(tp, error); 939 else 940 tp->t_softerror = error; 941 #if 0 942 wakeup((caddr_t) &so->so_timeo); 943 sorwakeup(so); 944 sowwakeup(so); 945 #endif 946 } 947 948 static int 949 tcp_pcblist(SYSCTL_HANDLER_ARGS) 950 { 951 int error, i, n, s; 952 struct inpcb *inp, **inp_list; 953 inp_gen_t gencnt; 954 struct xinpgen xig; 955 956 /* 957 * The process of preparing the TCB list is too time-consuming and 958 * resource-intensive to repeat twice on every request. 959 */ 960 if (req->oldptr == 0) { 961 n = tcbinfo[mycpu->gd_cpuid].ipi_count; 962 req->oldidx = 2 * (sizeof xig) 963 + (n + n/8) * sizeof(struct xtcpcb); 964 return 0; 965 } 966 967 if (req->newptr != 0) 968 return EPERM; 969 970 /* 971 * OK, now we're committed to doing something. 972 */ 973 s = splnet(); 974 gencnt = tcbinfo[mycpu->gd_cpuid].ipi_gencnt; 975 n = tcbinfo[mycpu->gd_cpuid].ipi_count; 976 splx(s); 977 978 xig.xig_len = sizeof xig; 979 xig.xig_count = n; 980 xig.xig_gen = gencnt; 981 xig.xig_sogen = so_gencnt; 982 error = SYSCTL_OUT(req, &xig, sizeof xig); 983 if (error) 984 return error; 985 986 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 987 if (inp_list == 0) 988 return ENOMEM; 989 990 s = splnet(); 991 for (inp = LIST_FIRST(&tcbinfo[mycpu->gd_cpuid].listhead), i = 0; 992 inp && i < n; inp = LIST_NEXT(inp, inp_list)) { 993 if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->td, inp)) 994 inp_list[i++] = inp; 995 } 996 splx(s); 997 n = i; 998 999 error = 0; 1000 for (i = 0; i < n; i++) { 1001 inp = inp_list[i]; 1002 if (inp->inp_gencnt <= gencnt) { 1003 struct xtcpcb xt; 1004 caddr_t inp_ppcb; 1005 xt.xt_len = sizeof xt; 1006 /* XXX should avoid extra copy */ 1007 bcopy(inp, &xt.xt_inp, sizeof *inp); 1008 inp_ppcb = inp->inp_ppcb; 1009 if (inp_ppcb != NULL) 1010 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1011 else 1012 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1013 if (inp->inp_socket) 1014 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1015 error = SYSCTL_OUT(req, &xt, sizeof xt); 1016 } 1017 } 1018 if (!error) { 1019 /* 1020 * Give the user an updated idea of our state. 1021 * If the generation differs from what we told 1022 * her before, she knows that something happened 1023 * while we were processing this request, and it 1024 * might be necessary to retry. 1025 */ 1026 s = splnet(); 1027 xig.xig_gen = tcbinfo[mycpu->gd_cpuid].ipi_gencnt; 1028 xig.xig_sogen = so_gencnt; 1029 xig.xig_count = tcbinfo[mycpu->gd_cpuid].ipi_count; 1030 splx(s); 1031 error = SYSCTL_OUT(req, &xig, sizeof xig); 1032 } 1033 free(inp_list, M_TEMP); 1034 return error; 1035 } 1036 1037 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1038 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1039 1040 static int 1041 tcp_getcred(SYSCTL_HANDLER_ARGS) 1042 { 1043 struct sockaddr_in addrs[2]; 1044 struct inpcb *inp; 1045 int cpu; 1046 int error, s; 1047 1048 error = suser(req->td); 1049 if (error) 1050 return (error); 1051 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1052 if (error) 1053 return (error); 1054 s = splnet(); 1055 1056 cpu = tcp_addrcpu(addrs[1].sin_addr.s_addr, addrs[1].sin_port, 1057 addrs[0].sin_addr.s_addr, addrs[0].sin_port); 1058 inp = in_pcblookup_hash(&tcbinfo[cpu], addrs[1].sin_addr, 1059 addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1060 if (inp == NULL || inp->inp_socket == NULL) { 1061 error = ENOENT; 1062 goto out; 1063 } 1064 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred)); 1065 out: 1066 splx(s); 1067 return (error); 1068 } 1069 1070 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW, 1071 0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection"); 1072 1073 #ifdef INET6 1074 static int 1075 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1076 { 1077 struct sockaddr_in6 addrs[2]; 1078 struct inpcb *inp; 1079 int error, s, mapped = 0; 1080 1081 error = suser(req->td); 1082 if (error) 1083 return (error); 1084 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1085 if (error) 1086 return (error); 1087 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1088 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1089 mapped = 1; 1090 else 1091 return (EINVAL); 1092 } 1093 s = splnet(); 1094 if (mapped == 1) { 1095 inp = in_pcblookup_hash(&tcbinfo[0], 1096 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1097 addrs[1].sin6_port, 1098 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1099 addrs[0].sin6_port, 1100 0, NULL); 1101 } else { 1102 inp = in6_pcblookup_hash(&tcbinfo[0], 1103 &addrs[1].sin6_addr, addrs[1].sin6_port, 1104 &addrs[0].sin6_addr, addrs[0].sin6_port, 1105 0, NULL); 1106 } 1107 if (inp == NULL || inp->inp_socket == NULL) { 1108 error = ENOENT; 1109 goto out; 1110 } 1111 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, 1112 sizeof(struct ucred)); 1113 out: 1114 splx(s); 1115 return (error); 1116 } 1117 1118 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW, 1119 0, 0, 1120 tcp6_getcred, "S,ucred", "Get the ucred of a TCP6 connection"); 1121 #endif 1122 1123 1124 void 1125 tcp_ctlinput(cmd, sa, vip) 1126 int cmd; 1127 struct sockaddr *sa; 1128 void *vip; 1129 { 1130 struct ip *ip = vip; 1131 struct tcphdr *th; 1132 struct in_addr faddr; 1133 struct inpcb *inp; 1134 struct tcpcb *tp; 1135 void (*notify) (struct inpcb *, int) = tcp_notify; 1136 tcp_seq icmp_seq; 1137 int cpu; 1138 int s; 1139 1140 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1141 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1142 return; 1143 1144 if (cmd == PRC_QUENCH) 1145 notify = tcp_quench; 1146 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1147 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1148 notify = tcp_drop_syn_sent; 1149 else if (cmd == PRC_MSGSIZE) 1150 notify = tcp_mtudisc; 1151 else if (PRC_IS_REDIRECT(cmd)) { 1152 ip = 0; 1153 notify = in_rtchange; 1154 } else if (cmd == PRC_HOSTDEAD) 1155 ip = 0; 1156 else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0) 1157 return; 1158 if (ip) { 1159 s = splnet(); 1160 th = (struct tcphdr *)((caddr_t)ip 1161 + (IP_VHL_HL(ip->ip_vhl) << 2)); 1162 cpu = tcp_addrcpu(faddr.s_addr, th->th_dport, 1163 ip->ip_src.s_addr, th->th_sport); 1164 inp = in_pcblookup_hash(&tcbinfo[cpu], faddr, th->th_dport, 1165 ip->ip_src, th->th_sport, 0, NULL); 1166 if (inp != NULL && inp->inp_socket != NULL) { 1167 icmp_seq = htonl(th->th_seq); 1168 tp = intotcpcb(inp); 1169 if (SEQ_GEQ(icmp_seq, tp->snd_una) && 1170 SEQ_LT(icmp_seq, tp->snd_max)) 1171 (*notify)(inp, inetctlerrmap[cmd]); 1172 } else { 1173 struct in_conninfo inc; 1174 1175 inc.inc_fport = th->th_dport; 1176 inc.inc_lport = th->th_sport; 1177 inc.inc_faddr = faddr; 1178 inc.inc_laddr = ip->ip_src; 1179 #ifdef INET6 1180 inc.inc_isipv6 = 0; 1181 #endif 1182 syncache_unreach(&inc, th); 1183 } 1184 splx(s); 1185 } else { 1186 for (cpu = 0; cpu < ncpus2; cpu++) { 1187 in_pcbnotifyall(&tcbinfo[cpu].listhead, faddr, 1188 inetctlerrmap[cmd], notify); 1189 } 1190 } 1191 } 1192 1193 #ifdef INET6 1194 void 1195 tcp6_ctlinput(cmd, sa, d) 1196 int cmd; 1197 struct sockaddr *sa; 1198 void *d; 1199 { 1200 struct tcphdr th; 1201 void (*notify) (struct inpcb *, int) = tcp_notify; 1202 struct ip6_hdr *ip6; 1203 struct mbuf *m; 1204 struct ip6ctlparam *ip6cp = NULL; 1205 const struct sockaddr_in6 *sa6_src = NULL; 1206 int off; 1207 struct tcp_portonly { 1208 u_int16_t th_sport; 1209 u_int16_t th_dport; 1210 } *thp; 1211 1212 if (sa->sa_family != AF_INET6 || 1213 sa->sa_len != sizeof(struct sockaddr_in6)) 1214 return; 1215 1216 if (cmd == PRC_QUENCH) 1217 notify = tcp_quench; 1218 else if (cmd == PRC_MSGSIZE) 1219 notify = tcp_mtudisc; 1220 else if (!PRC_IS_REDIRECT(cmd) && 1221 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1222 return; 1223 1224 /* if the parameter is from icmp6, decode it. */ 1225 if (d != NULL) { 1226 ip6cp = (struct ip6ctlparam *)d; 1227 m = ip6cp->ip6c_m; 1228 ip6 = ip6cp->ip6c_ip6; 1229 off = ip6cp->ip6c_off; 1230 sa6_src = ip6cp->ip6c_src; 1231 } else { 1232 m = NULL; 1233 ip6 = NULL; 1234 off = 0; /* fool gcc */ 1235 sa6_src = &sa6_any; 1236 } 1237 1238 if (ip6) { 1239 struct in_conninfo inc; 1240 /* 1241 * XXX: We assume that when IPV6 is non NULL, 1242 * M and OFF are valid. 1243 */ 1244 1245 /* check if we can safely examine src and dst ports */ 1246 if (m->m_pkthdr.len < off + sizeof(*thp)) 1247 return; 1248 1249 bzero(&th, sizeof(th)); 1250 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1251 1252 in6_pcbnotify(&tcbinfo[0].listhead, sa, th.th_dport, 1253 (struct sockaddr *)ip6cp->ip6c_src, 1254 th.th_sport, cmd, notify); 1255 1256 inc.inc_fport = th.th_dport; 1257 inc.inc_lport = th.th_sport; 1258 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1259 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1260 inc.inc_isipv6 = 1; 1261 syncache_unreach(&inc, &th); 1262 } else 1263 in6_pcbnotify(&tcbinfo[0].listhead, sa, 0, 1264 (const struct sockaddr *)sa6_src, 0, cmd, notify); 1265 } 1266 #endif /* INET6 */ 1267 1268 1269 /* 1270 * Following is where TCP initial sequence number generation occurs. 1271 * 1272 * There are two places where we must use initial sequence numbers: 1273 * 1. In SYN-ACK packets. 1274 * 2. In SYN packets. 1275 * 1276 * All ISNs for SYN-ACK packets are generated by the syncache. See 1277 * tcp_syncache.c for details. 1278 * 1279 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1280 * depends on this property. In addition, these ISNs should be 1281 * unguessable so as to prevent connection hijacking. To satisfy 1282 * the requirements of this situation, the algorithm outlined in 1283 * RFC 1948 is used to generate sequence numbers. 1284 * 1285 * Implementation details: 1286 * 1287 * Time is based off the system timer, and is corrected so that it 1288 * increases by one megabyte per second. This allows for proper 1289 * recycling on high speed LANs while still leaving over an hour 1290 * before rollover. 1291 * 1292 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1293 * between seeding of isn_secret. This is normally set to zero, 1294 * as reseeding should not be necessary. 1295 * 1296 */ 1297 1298 #define ISN_BYTES_PER_SECOND 1048576 1299 1300 u_char isn_secret[32]; 1301 int isn_last_reseed; 1302 MD5_CTX isn_ctx; 1303 1304 tcp_seq 1305 tcp_new_isn(tp) 1306 struct tcpcb *tp; 1307 { 1308 u_int32_t md5_buffer[4]; 1309 tcp_seq new_isn; 1310 1311 /* Seed if this is the first use, reseed if requested. */ 1312 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1313 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1314 < (u_int)ticks))) { 1315 read_random_unlimited(&isn_secret, sizeof(isn_secret)); 1316 isn_last_reseed = ticks; 1317 } 1318 1319 /* Compute the md5 hash and return the ISN. */ 1320 MD5Init(&isn_ctx); 1321 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1322 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1323 #ifdef INET6 1324 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1325 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1326 sizeof(struct in6_addr)); 1327 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1328 sizeof(struct in6_addr)); 1329 } else 1330 #endif 1331 { 1332 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1333 sizeof(struct in_addr)); 1334 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1335 sizeof(struct in_addr)); 1336 } 1337 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1338 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1339 new_isn = (tcp_seq) md5_buffer[0]; 1340 new_isn += ticks * (ISN_BYTES_PER_SECOND / hz); 1341 return new_isn; 1342 } 1343 1344 /* 1345 * When a source quench is received, close congestion window 1346 * to one segment. We will gradually open it again as we proceed. 1347 */ 1348 void 1349 tcp_quench(inp, errno) 1350 struct inpcb *inp; 1351 int errno; 1352 { 1353 struct tcpcb *tp = intotcpcb(inp); 1354 1355 if (tp) 1356 tp->snd_cwnd = tp->t_maxseg; 1357 } 1358 1359 /* 1360 * When a specific ICMP unreachable message is received and the 1361 * connection state is SYN-SENT, drop the connection. This behavior 1362 * is controlled by the icmp_may_rst sysctl. 1363 */ 1364 void 1365 tcp_drop_syn_sent(inp, errno) 1366 struct inpcb *inp; 1367 int errno; 1368 { 1369 struct tcpcb *tp = intotcpcb(inp); 1370 1371 if (tp && tp->t_state == TCPS_SYN_SENT) 1372 tcp_drop(tp, errno); 1373 } 1374 1375 /* 1376 * When `need fragmentation' ICMP is received, update our idea of the MSS 1377 * based on the new value in the route. Also nudge TCP to send something, 1378 * since we know the packet we just sent was dropped. 1379 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1380 */ 1381 void 1382 tcp_mtudisc(inp, errno) 1383 struct inpcb *inp; 1384 int errno; 1385 { 1386 struct tcpcb *tp = intotcpcb(inp); 1387 struct rtentry *rt; 1388 struct rmxp_tao *taop; 1389 struct socket *so = inp->inp_socket; 1390 int offered; 1391 int mss; 1392 #ifdef INET6 1393 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1394 #endif /* INET6 */ 1395 1396 if (tp) { 1397 #ifdef INET6 1398 if (isipv6) 1399 rt = tcp_rtlookup6(&inp->inp_inc); 1400 else 1401 #endif /* INET6 */ 1402 rt = tcp_rtlookup(&inp->inp_inc); 1403 if (!rt || !rt->rt_rmx.rmx_mtu) { 1404 tp->t_maxopd = tp->t_maxseg = 1405 #ifdef INET6 1406 isipv6 ? tcp_v6mssdflt : 1407 #endif /* INET6 */ 1408 tcp_mssdflt; 1409 return; 1410 } 1411 taop = rmx_taop(rt->rt_rmx); 1412 offered = taop->tao_mssopt; 1413 mss = rt->rt_rmx.rmx_mtu - 1414 #ifdef INET6 1415 (isipv6 ? 1416 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1417 #endif /* INET6 */ 1418 sizeof(struct tcpiphdr) 1419 #ifdef INET6 1420 ) 1421 #endif /* INET6 */ 1422 ; 1423 1424 if (offered) 1425 mss = min(mss, offered); 1426 /* 1427 * XXX - The above conditional probably violates the TCP 1428 * spec. The problem is that, since we don't know the 1429 * other end's MSS, we are supposed to use a conservative 1430 * default. But, if we do that, then MTU discovery will 1431 * never actually take place, because the conservative 1432 * default is much less than the MTUs typically seen 1433 * on the Internet today. For the moment, we'll sweep 1434 * this under the carpet. 1435 * 1436 * The conservative default might not actually be a problem 1437 * if the only case this occurs is when sending an initial 1438 * SYN with options and data to a host we've never talked 1439 * to before. Then, they will reply with an MSS value which 1440 * will get recorded and the new parameters should get 1441 * recomputed. For Further Study. 1442 */ 1443 if (tp->t_maxopd <= mss) 1444 return; 1445 tp->t_maxopd = mss; 1446 1447 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1448 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1449 mss -= TCPOLEN_TSTAMP_APPA; 1450 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 1451 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC) 1452 mss -= TCPOLEN_CC_APPA; 1453 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1454 if (mss > MCLBYTES) 1455 mss &= ~(MCLBYTES-1); 1456 #else 1457 if (mss > MCLBYTES) 1458 mss = mss / MCLBYTES * MCLBYTES; 1459 #endif 1460 if (so->so_snd.sb_hiwat < mss) 1461 mss = so->so_snd.sb_hiwat; 1462 1463 tp->t_maxseg = mss; 1464 1465 tcpstat.tcps_mturesent++; 1466 tp->t_rtttime = 0; 1467 tp->snd_nxt = tp->snd_una; 1468 tcp_output(tp); 1469 } 1470 } 1471 1472 /* 1473 * Look-up the routing entry to the peer of this inpcb. If no route 1474 * is found and it cannot be allocated the return NULL. This routine 1475 * is called by TCP routines that access the rmx structure and by tcp_mss 1476 * to get the interface MTU. 1477 */ 1478 struct rtentry * 1479 tcp_rtlookup(inc) 1480 struct in_conninfo *inc; 1481 { 1482 struct route *ro; 1483 struct rtentry *rt; 1484 1485 ro = &inc->inc_route; 1486 rt = ro->ro_rt; 1487 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 1488 /* No route yet, so try to acquire one */ 1489 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1490 ro->ro_dst.sa_family = AF_INET; 1491 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 1492 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1493 inc->inc_faddr; 1494 rtalloc(ro); 1495 rt = ro->ro_rt; 1496 } 1497 } 1498 return rt; 1499 } 1500 1501 #ifdef INET6 1502 struct rtentry * 1503 tcp_rtlookup6(inc) 1504 struct in_conninfo *inc; 1505 { 1506 struct route_in6 *ro6; 1507 struct rtentry *rt; 1508 1509 ro6 = &inc->inc6_route; 1510 rt = ro6->ro_rt; 1511 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 1512 /* No route yet, so try to acquire one */ 1513 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1514 ro6->ro_dst.sin6_family = AF_INET6; 1515 ro6->ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1516 ro6->ro_dst.sin6_addr = inc->inc6_faddr; 1517 rtalloc((struct route *)ro6); 1518 rt = ro6->ro_rt; 1519 } 1520 } 1521 return rt; 1522 } 1523 #endif /* INET6 */ 1524 1525 #ifdef IPSEC 1526 /* compute ESP/AH header size for TCP, including outer IP header. */ 1527 size_t 1528 ipsec_hdrsiz_tcp(tp) 1529 struct tcpcb *tp; 1530 { 1531 struct inpcb *inp; 1532 struct mbuf *m; 1533 size_t hdrsiz; 1534 struct ip *ip; 1535 #ifdef INET6 1536 struct ip6_hdr *ip6; 1537 #endif /* INET6 */ 1538 struct tcphdr *th; 1539 1540 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1541 return 0; 1542 MGETHDR(m, M_DONTWAIT, MT_DATA); 1543 if (!m) 1544 return 0; 1545 1546 #ifdef INET6 1547 if ((inp->inp_vflag & INP_IPV6) != 0) { 1548 ip6 = mtod(m, struct ip6_hdr *); 1549 th = (struct tcphdr *)(ip6 + 1); 1550 m->m_pkthdr.len = m->m_len = 1551 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1552 tcp_fillheaders(tp, ip6, th); 1553 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1554 } else 1555 #endif /* INET6 */ 1556 { 1557 ip = mtod(m, struct ip *); 1558 th = (struct tcphdr *)(ip + 1); 1559 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1560 tcp_fillheaders(tp, ip, th); 1561 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1562 } 1563 1564 m_free(m); 1565 return hdrsiz; 1566 } 1567 #endif /*IPSEC*/ 1568 1569 /* 1570 * Return a pointer to the cached information about the remote host. 1571 * The cached information is stored in the protocol specific part of 1572 * the route metrics. 1573 */ 1574 struct rmxp_tao * 1575 tcp_gettaocache(inc) 1576 struct in_conninfo *inc; 1577 { 1578 struct rtentry *rt; 1579 1580 #ifdef INET6 1581 if (inc->inc_isipv6) 1582 rt = tcp_rtlookup6(inc); 1583 else 1584 #endif /* INET6 */ 1585 rt = tcp_rtlookup(inc); 1586 1587 /* Make sure this is a host route and is up. */ 1588 if (rt == NULL || 1589 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) 1590 return NULL; 1591 1592 return rmx_taop(rt->rt_rmx); 1593 } 1594 1595 /* 1596 * Clear all the TAO cache entries, called from tcp_init. 1597 * 1598 * XXX 1599 * This routine is just an empty one, because we assume that the routing 1600 * routing tables are initialized at the same time when TCP, so there is 1601 * nothing in the cache left over. 1602 */ 1603 static void 1604 tcp_cleartaocache() 1605 { 1606 } 1607 1608 /* 1609 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1610 * 1611 * This code attempts to calculate the bandwidth-delay product as a 1612 * means of determining the optimal window size to maximize bandwidth, 1613 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1614 * routers. This code also does a fairly good job keeping RTTs in check 1615 * across slow links like modems. We implement an algorithm which is very 1616 * similar (but not meant to be) TCP/Vegas. The code operates on the 1617 * transmitter side of a TCP connection and so only effects the transmit 1618 * side of the connection. 1619 * 1620 * BACKGROUND: TCP makes no provision for the management of buffer space 1621 * at the end points or at the intermediate routers and switches. A TCP 1622 * stream, whether using NewReno or not, will eventually buffer as 1623 * many packets as it is able and the only reason this typically works is 1624 * due to the fairly small default buffers made available for a connection 1625 * (typicaly 16K or 32K). As machines use larger windows and/or window 1626 * scaling it is now fairly easy for even a single TCP connection to blow-out 1627 * all available buffer space not only on the local interface, but on 1628 * intermediate routers and switches as well. NewReno makes a misguided 1629 * attempt to 'solve' this problem by waiting for an actual failure to occur, 1630 * then backing off, then steadily increasing the window again until another 1631 * failure occurs, ad-infinitum. This results in terrible oscillation that 1632 * is only made worse as network loads increase and the idea of intentionally 1633 * blowing out network buffers is, frankly, a terrible way to manage network 1634 * resources. 1635 * 1636 * It is far better to limit the transmit window prior to the failure 1637 * condition being achieved. There are two general ways to do this: First 1638 * you can 'scan' through different transmit window sizes and locate the 1639 * point where the RTT stops increasing, indicating that you have filled the 1640 * pipe, then scan backwards until you note that RTT stops decreasing, then 1641 * repeat ad-infinitum. This method works in principle but has severe 1642 * implementation issues due to RTT variances, timer granularity, and 1643 * instability in the algorithm which can lead to many false positives and 1644 * create oscillations as well as interact badly with other TCP streams 1645 * implementing the same algorithm. 1646 * 1647 * The second method is to limit the window to the bandwidth delay product 1648 * of the link. This is the method we implement. RTT variances and our 1649 * own manipulation of the congestion window, bwnd, can potentially 1650 * destabilize the algorithm. For this reason we have to stabilize the 1651 * elements used to calculate the window. We do this by using the minimum 1652 * observed RTT, the long term average of the observed bandwidth, and 1653 * by adding two segments worth of slop. It isn't perfect but it is able 1654 * to react to changing conditions and gives us a very stable basis on 1655 * which to extend the algorithm. 1656 */ 1657 void 1658 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 1659 { 1660 u_long bw; 1661 u_long bwnd; 1662 int save_ticks; 1663 1664 /* 1665 * If inflight_enable is disabled in the middle of a tcp connection, 1666 * make sure snd_bwnd is effectively disabled. 1667 */ 1668 if (tcp_inflight_enable == 0) { 1669 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1670 tp->snd_bandwidth = 0; 1671 return; 1672 } 1673 1674 /* 1675 * Figure out the bandwidth. Due to the tick granularity this 1676 * is a very rough number and it MUST be averaged over a fairly 1677 * long period of time. XXX we need to take into account a link 1678 * that is not using all available bandwidth, but for now our 1679 * slop will ramp us up if this case occurs and the bandwidth later 1680 * increases. 1681 * 1682 * Note: if ticks rollover 'bw' may wind up negative. We must 1683 * effectively reset t_bw_rtttime for this case. 1684 */ 1685 save_ticks = ticks; 1686 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 1687 return; 1688 1689 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 1690 (save_ticks - tp->t_bw_rtttime); 1691 tp->t_bw_rtttime = save_ticks; 1692 tp->t_bw_rtseq = ack_seq; 1693 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 1694 return; 1695 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 1696 1697 tp->snd_bandwidth = bw; 1698 1699 /* 1700 * Calculate the semi-static bandwidth delay product, plus two maximal 1701 * segments. The additional slop puts us squarely in the sweet 1702 * spot and also handles the bandwidth run-up case. Without the 1703 * slop we could be locking ourselves into a lower bandwidth. 1704 * 1705 * Situations Handled: 1706 * (1) Prevents over-queueing of packets on LANs, especially on 1707 * high speed LANs, allowing larger TCP buffers to be 1708 * specified, and also does a good job preventing 1709 * over-queueing of packets over choke points like modems 1710 * (at least for the transmit side). 1711 * 1712 * (2) Is able to handle changing network loads (bandwidth 1713 * drops so bwnd drops, bandwidth increases so bwnd 1714 * increases). 1715 * 1716 * (3) Theoretically should stabilize in the face of multiple 1717 * connections implementing the same algorithm (this may need 1718 * a little work). 1719 * 1720 * (4) Stability value (defaults to 20 = 2 maximal packets) can 1721 * be adjusted with a sysctl but typically only needs to be on 1722 * very slow connections. A value no smaller then 5 should 1723 * be used, but only reduce this default if you have no other 1724 * choice. 1725 */ 1726 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 1727 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * (int)tp->t_maxseg / 10; 1728 #undef USERTT 1729 1730 if (tcp_inflight_debug > 0) { 1731 static int ltime; 1732 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 1733 ltime = ticks; 1734 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 1735 tp, 1736 bw, 1737 tp->t_rttbest, 1738 tp->t_srtt, 1739 bwnd 1740 ); 1741 } 1742 } 1743 if ((long)bwnd < tcp_inflight_min) 1744 bwnd = tcp_inflight_min; 1745 if (bwnd > tcp_inflight_max) 1746 bwnd = tcp_inflight_max; 1747 if ((long)bwnd < tp->t_maxseg * 2) 1748 bwnd = tp->t_maxseg * 2; 1749 tp->snd_bwnd = bwnd; 1750 } 1751