1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_compat.h" 36 #include "opt_inet.h" 37 #include "opt_inet6.h" 38 #include "opt_ipsec.h" 39 #include "opt_mac.h" 40 #include "opt_tcpdebug.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/callout.h> 45 #include <sys/kernel.h> 46 #include <sys/sysctl.h> 47 #include <sys/malloc.h> 48 #include <sys/mbuf.h> 49 #ifdef INET6 50 #include <sys/domain.h> 51 #endif 52 #include <sys/priv.h> 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/vimage.h> 59 60 #include <vm/uma.h> 61 62 #include <net/route.h> 63 #include <net/if.h> 64 65 #include <netinet/in.h> 66 #include <netinet/in_systm.h> 67 #include <netinet/ip.h> 68 #ifdef INET6 69 #include <netinet/ip6.h> 70 #endif 71 #include <netinet/in_pcb.h> 72 #ifdef INET6 73 #include <netinet6/in6_pcb.h> 74 #endif 75 #include <netinet/in_var.h> 76 #include <netinet/ip_var.h> 77 #ifdef INET6 78 #include <netinet6/ip6_var.h> 79 #include <netinet6/scope6_var.h> 80 #include <netinet6/nd6.h> 81 #endif 82 #include <netinet/ip_icmp.h> 83 #include <netinet/tcp.h> 84 #include <netinet/tcp_fsm.h> 85 #include <netinet/tcp_seq.h> 86 #include <netinet/tcp_timer.h> 87 #include <netinet/tcp_var.h> 88 #include <netinet/tcp_syncache.h> 89 #include <netinet/tcp_offload.h> 90 #ifdef INET6 91 #include <netinet6/tcp6_var.h> 92 #endif 93 #include <netinet/tcpip.h> 94 #ifdef TCPDEBUG 95 #include <netinet/tcp_debug.h> 96 #endif 97 #include <netinet6/ip6protosw.h> 98 99 #ifdef IPSEC 100 #include <netipsec/ipsec.h> 101 #include <netipsec/xform.h> 102 #ifdef INET6 103 #include <netipsec/ipsec6.h> 104 #endif 105 #include <netipsec/key.h> 106 #include <sys/syslog.h> 107 #endif /*IPSEC*/ 108 109 #include <machine/in_cksum.h> 110 #include <sys/md5.h> 111 112 #include <security/mac/mac_framework.h> 113 114 int tcp_mssdflt = TCP_MSS; 115 #ifdef INET6 116 int tcp_v6mssdflt = TCP6_MSS; 117 #endif 118 119 static int 120 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 121 { 122 INIT_VNET_INET(TD_TO_VNET(curthread)); 123 int error, new; 124 125 new = V_tcp_mssdflt; 126 error = sysctl_handle_int(oidp, &new, 0, req); 127 if (error == 0 && req->newptr) { 128 if (new < TCP_MINMSS) 129 error = EINVAL; 130 else 131 V_tcp_mssdflt = new; 132 } 133 return (error); 134 } 135 136 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLTYPE_INT|CTLFLAG_RW, 137 &tcp_mssdflt, 0, &sysctl_net_inet_tcp_mss_check, "I", 138 "Default TCP Maximum Segment Size"); 139 140 #ifdef INET6 141 static int 142 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 143 { 144 INIT_VNET_INET6(TD_TO_VNET(curthread)); 145 int error, new; 146 147 new = V_tcp_v6mssdflt; 148 error = sysctl_handle_int(oidp, &new, 0, req); 149 if (error == 0 && req->newptr) { 150 if (new < TCP_MINMSS) 151 error = EINVAL; 152 else 153 V_tcp_v6mssdflt = new; 154 } 155 return (error); 156 } 157 158 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLTYPE_INT|CTLFLAG_RW, 159 &tcp_v6mssdflt, 0, &sysctl_net_inet_tcp_mss_v6_check, "I", 160 "Default TCP Maximum Segment Size for IPv6"); 161 #endif 162 163 /* 164 * Minimum MSS we accept and use. This prevents DoS attacks where 165 * we are forced to a ridiculous low MSS like 20 and send hundreds 166 * of packets instead of one. The effect scales with the available 167 * bandwidth and quickly saturates the CPU and network interface 168 * with packet generation and sending. Set to zero to disable MINMSS 169 * checking. This setting prevents us from sending too small packets. 170 */ 171 int tcp_minmss = TCP_MINMSS; 172 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, minmss, 173 CTLFLAG_RW, tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 174 175 int tcp_do_rfc1323 = 1; 176 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, 177 CTLFLAG_RW, tcp_do_rfc1323, 0, 178 "Enable rfc1323 (high performance TCP) extensions"); 179 180 static int tcp_log_debug = 0; 181 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 182 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 183 184 static int tcp_tcbhashsize = 0; 185 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 186 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 187 188 static int do_tcpdrain = 1; 189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 190 "Enable tcp_drain routine for extra help when low on mbufs"); 191 192 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, pcbcount, 193 CTLFLAG_RD, tcbinfo.ipi_count, 0, "Number of active PCBs"); 194 195 static int icmp_may_rst = 1; 196 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, icmp_may_rst, 197 CTLFLAG_RW, icmp_may_rst, 0, 198 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 199 200 static int tcp_isn_reseed_interval = 0; 201 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, isn_reseed_interval, 202 CTLFLAG_RW, tcp_isn_reseed_interval, 0, 203 "Seconds between reseeding of ISN secret"); 204 205 /* 206 * TCP bandwidth limiting sysctls. Note that the default lower bound of 207 * 1024 exists only for debugging. A good production default would be 208 * something like 6100. 209 */ 210 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 211 "TCP inflight data limiting"); 212 213 static int tcp_inflight_enable = 1; 214 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, enable, 215 CTLFLAG_RW, tcp_inflight_enable, 0, 216 "Enable automatic TCP inflight data limiting"); 217 218 static int tcp_inflight_debug = 0; 219 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 220 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 221 222 static int tcp_inflight_rttthresh; 223 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW, 224 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I", 225 "RTT threshold below which inflight will deactivate itself"); 226 227 static int tcp_inflight_min = 6144; 228 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, min, 229 CTLFLAG_RW, tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 230 231 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 232 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, max, 233 CTLFLAG_RW, tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 234 235 static int tcp_inflight_stab = 20; 236 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_inflight, OID_AUTO, stab, 237 CTLFLAG_RW, tcp_inflight_stab, 0, 238 "Inflight Algorithm Stabilization 20 = 2 packets"); 239 240 uma_zone_t sack_hole_zone; 241 242 static struct inpcb *tcp_notify(struct inpcb *, int); 243 static void tcp_isn_tick(void *); 244 245 /* 246 * Target size of TCP PCB hash tables. Must be a power of two. 247 * 248 * Note that this can be overridden by the kernel environment 249 * variable net.inet.tcp.tcbhashsize 250 */ 251 #ifndef TCBHASHSIZE 252 #define TCBHASHSIZE 512 253 #endif 254 255 /* 256 * XXX 257 * Callouts should be moved into struct tcp directly. They are currently 258 * separate because the tcpcb structure is exported to userland for sysctl 259 * parsing purposes, which do not know about callouts. 260 */ 261 struct tcpcb_mem { 262 struct tcpcb tcb; 263 struct tcp_timer tt; 264 }; 265 266 static uma_zone_t tcpcb_zone; 267 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 268 struct callout isn_callout; 269 static struct mtx isn_mtx; 270 271 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 272 #define ISN_LOCK() mtx_lock(&isn_mtx) 273 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 274 275 /* 276 * TCP initialization. 277 */ 278 static void 279 tcp_zone_change(void *tag) 280 { 281 282 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 283 uma_zone_set_max(tcpcb_zone, maxsockets); 284 tcp_tw_zone_change(); 285 } 286 287 static int 288 tcp_inpcb_init(void *mem, int size, int flags) 289 { 290 struct inpcb *inp = mem; 291 292 INP_LOCK_INIT(inp, "inp", "tcpinp"); 293 return (0); 294 } 295 296 void 297 tcp_init(void) 298 { 299 INIT_VNET_INET(curvnet); 300 301 int hashsize = TCBHASHSIZE; 302 tcp_delacktime = TCPTV_DELACK; 303 tcp_keepinit = TCPTV_KEEP_INIT; 304 tcp_keepidle = TCPTV_KEEP_IDLE; 305 tcp_keepintvl = TCPTV_KEEPINTVL; 306 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 307 tcp_msl = TCPTV_MSL; 308 tcp_rexmit_min = TCPTV_MIN; 309 if (tcp_rexmit_min < 1) 310 tcp_rexmit_min = 1; 311 tcp_rexmit_slop = TCPTV_CPU_VAR; 312 V_tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH; 313 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 314 315 INP_INFO_LOCK_INIT(&V_tcbinfo, "tcp"); 316 LIST_INIT(&V_tcb); 317 V_tcbinfo.ipi_listhead = &V_tcb; 318 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 319 if (!powerof2(hashsize)) { 320 printf("WARNING: TCB hash size not a power of 2\n"); 321 hashsize = 512; /* safe default */ 322 } 323 tcp_tcbhashsize = hashsize; 324 V_tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB, 325 &V_tcbinfo.ipi_hashmask); 326 V_tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB, 327 &V_tcbinfo.ipi_porthashmask); 328 V_tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 329 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 330 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 331 #ifdef INET6 332 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 333 #else /* INET6 */ 334 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 335 #endif /* INET6 */ 336 if (max_protohdr < TCP_MINPROTOHDR) 337 max_protohdr = TCP_MINPROTOHDR; 338 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 339 panic("tcp_init"); 340 #undef TCP_MINPROTOHDR 341 /* 342 * These have to be type stable for the benefit of the timers. 343 */ 344 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 345 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 346 uma_zone_set_max(tcpcb_zone, maxsockets); 347 tcp_tw_init(); 348 syncache_init(); 349 tcp_hc_init(); 350 tcp_reass_init(); 351 ISN_LOCK_INIT(); 352 callout_init(&isn_callout, CALLOUT_MPSAFE); 353 tcp_isn_tick(NULL); 354 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 355 SHUTDOWN_PRI_DEFAULT); 356 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 357 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 358 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 359 EVENTHANDLER_PRI_ANY); 360 } 361 362 void 363 tcp_fini(void *xtp) 364 { 365 366 callout_stop(&isn_callout); 367 } 368 369 /* 370 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 371 * tcp_template used to store this data in mbufs, but we now recopy it out 372 * of the tcpcb each time to conserve mbufs. 373 */ 374 void 375 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 376 { 377 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 378 379 INP_WLOCK_ASSERT(inp); 380 381 #ifdef INET6 382 if ((inp->inp_vflag & INP_IPV6) != 0) { 383 struct ip6_hdr *ip6; 384 385 ip6 = (struct ip6_hdr *)ip_ptr; 386 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 387 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 388 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 389 (IPV6_VERSION & IPV6_VERSION_MASK); 390 ip6->ip6_nxt = IPPROTO_TCP; 391 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 392 ip6->ip6_src = inp->in6p_laddr; 393 ip6->ip6_dst = inp->in6p_faddr; 394 } else 395 #endif 396 { 397 struct ip *ip; 398 399 ip = (struct ip *)ip_ptr; 400 ip->ip_v = IPVERSION; 401 ip->ip_hl = 5; 402 ip->ip_tos = inp->inp_ip_tos; 403 ip->ip_len = 0; 404 ip->ip_id = 0; 405 ip->ip_off = 0; 406 ip->ip_ttl = inp->inp_ip_ttl; 407 ip->ip_sum = 0; 408 ip->ip_p = IPPROTO_TCP; 409 ip->ip_src = inp->inp_laddr; 410 ip->ip_dst = inp->inp_faddr; 411 } 412 th->th_sport = inp->inp_lport; 413 th->th_dport = inp->inp_fport; 414 th->th_seq = 0; 415 th->th_ack = 0; 416 th->th_x2 = 0; 417 th->th_off = 5; 418 th->th_flags = 0; 419 th->th_win = 0; 420 th->th_urp = 0; 421 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 422 } 423 424 /* 425 * Create template to be used to send tcp packets on a connection. 426 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 427 * use for this function is in keepalives, which use tcp_respond. 428 */ 429 struct tcptemp * 430 tcpip_maketemplate(struct inpcb *inp) 431 { 432 struct tcptemp *t; 433 434 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 435 if (t == NULL) 436 return (NULL); 437 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 438 return (t); 439 } 440 441 /* 442 * Send a single message to the TCP at address specified by 443 * the given TCP/IP header. If m == NULL, then we make a copy 444 * of the tcpiphdr at ti and send directly to the addressed host. 445 * This is used to force keep alive messages out using the TCP 446 * template for a connection. If flags are given then we send 447 * a message back to the TCP which originated the * segment ti, 448 * and discard the mbuf containing it and any other attached mbufs. 449 * 450 * In any case the ack and sequence number of the transmitted 451 * segment are as specified by the parameters. 452 * 453 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 454 */ 455 void 456 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 457 tcp_seq ack, tcp_seq seq, int flags) 458 { 459 INIT_VNET_INET(curvnet); 460 int tlen; 461 int win = 0; 462 struct ip *ip; 463 struct tcphdr *nth; 464 #ifdef INET6 465 struct ip6_hdr *ip6; 466 int isipv6; 467 #endif /* INET6 */ 468 int ipflags = 0; 469 struct inpcb *inp; 470 471 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 472 473 #ifdef INET6 474 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 475 ip6 = ipgen; 476 #endif /* INET6 */ 477 ip = ipgen; 478 479 if (tp != NULL) { 480 inp = tp->t_inpcb; 481 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 482 INP_WLOCK_ASSERT(inp); 483 } else 484 inp = NULL; 485 486 if (tp != NULL) { 487 if (!(flags & TH_RST)) { 488 win = sbspace(&inp->inp_socket->so_rcv); 489 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 490 win = (long)TCP_MAXWIN << tp->rcv_scale; 491 } 492 } 493 if (m == NULL) { 494 m = m_gethdr(M_DONTWAIT, MT_DATA); 495 if (m == NULL) 496 return; 497 tlen = 0; 498 m->m_data += max_linkhdr; 499 #ifdef INET6 500 if (isipv6) { 501 bcopy((caddr_t)ip6, mtod(m, caddr_t), 502 sizeof(struct ip6_hdr)); 503 ip6 = mtod(m, struct ip6_hdr *); 504 nth = (struct tcphdr *)(ip6 + 1); 505 } else 506 #endif /* INET6 */ 507 { 508 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 509 ip = mtod(m, struct ip *); 510 nth = (struct tcphdr *)(ip + 1); 511 } 512 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 513 flags = TH_ACK; 514 } else { 515 /* 516 * reuse the mbuf. 517 * XXX MRT We inherrit the FIB, which is lucky. 518 */ 519 m_freem(m->m_next); 520 m->m_next = NULL; 521 m->m_data = (caddr_t)ipgen; 522 /* m_len is set later */ 523 tlen = 0; 524 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 525 #ifdef INET6 526 if (isipv6) { 527 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 528 nth = (struct tcphdr *)(ip6 + 1); 529 } else 530 #endif /* INET6 */ 531 { 532 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 533 nth = (struct tcphdr *)(ip + 1); 534 } 535 if (th != nth) { 536 /* 537 * this is usually a case when an extension header 538 * exists between the IPv6 header and the 539 * TCP header. 540 */ 541 nth->th_sport = th->th_sport; 542 nth->th_dport = th->th_dport; 543 } 544 xchg(nth->th_dport, nth->th_sport, n_short); 545 #undef xchg 546 } 547 #ifdef INET6 548 if (isipv6) { 549 ip6->ip6_flow = 0; 550 ip6->ip6_vfc = IPV6_VERSION; 551 ip6->ip6_nxt = IPPROTO_TCP; 552 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 553 tlen)); 554 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 555 } else 556 #endif 557 { 558 tlen += sizeof (struct tcpiphdr); 559 ip->ip_len = tlen; 560 ip->ip_ttl = V_ip_defttl; 561 if (V_path_mtu_discovery) 562 ip->ip_off |= IP_DF; 563 } 564 m->m_len = tlen; 565 m->m_pkthdr.len = tlen; 566 m->m_pkthdr.rcvif = NULL; 567 #ifdef MAC 568 if (inp != NULL) { 569 /* 570 * Packet is associated with a socket, so allow the 571 * label of the response to reflect the socket label. 572 */ 573 INP_WLOCK_ASSERT(inp); 574 mac_inpcb_create_mbuf(inp, m); 575 } else { 576 /* 577 * Packet is not associated with a socket, so possibly 578 * update the label in place. 579 */ 580 mac_netinet_tcp_reply(m); 581 } 582 #endif 583 nth->th_seq = htonl(seq); 584 nth->th_ack = htonl(ack); 585 nth->th_x2 = 0; 586 nth->th_off = sizeof (struct tcphdr) >> 2; 587 nth->th_flags = flags; 588 if (tp != NULL) 589 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 590 else 591 nth->th_win = htons((u_short)win); 592 nth->th_urp = 0; 593 #ifdef INET6 594 if (isipv6) { 595 nth->th_sum = 0; 596 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 597 sizeof(struct ip6_hdr), 598 tlen - sizeof(struct ip6_hdr)); 599 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 600 NULL, NULL); 601 } else 602 #endif /* INET6 */ 603 { 604 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 605 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 606 m->m_pkthdr.csum_flags = CSUM_TCP; 607 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 608 } 609 #ifdef TCPDEBUG 610 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 611 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 612 #endif 613 #ifdef INET6 614 if (isipv6) 615 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 616 else 617 #endif /* INET6 */ 618 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 619 } 620 621 /* 622 * Create a new TCP control block, making an 623 * empty reassembly queue and hooking it to the argument 624 * protocol control block. The `inp' parameter must have 625 * come from the zone allocator set up in tcp_init(). 626 */ 627 struct tcpcb * 628 tcp_newtcpcb(struct inpcb *inp) 629 { 630 INIT_VNET_INET(inp->inp_vnet); 631 struct tcpcb_mem *tm; 632 struct tcpcb *tp; 633 #ifdef INET6 634 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 635 #endif /* INET6 */ 636 637 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 638 if (tm == NULL) 639 return (NULL); 640 tp = &tm->tcb; 641 tp->t_timers = &tm->tt; 642 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 643 tp->t_maxseg = tp->t_maxopd = 644 #ifdef INET6 645 isipv6 ? V_tcp_v6mssdflt : 646 #endif /* INET6 */ 647 V_tcp_mssdflt; 648 649 /* Set up our timeouts. */ 650 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE); 651 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE); 652 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE); 653 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE); 654 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE); 655 656 if (V_tcp_do_rfc1323) 657 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 658 if (V_tcp_do_sack) 659 tp->t_flags |= TF_SACK_PERMIT; 660 TAILQ_INIT(&tp->snd_holes); 661 tp->t_inpcb = inp; /* XXX */ 662 /* 663 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 664 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 665 * reasonable initial retransmit time. 666 */ 667 tp->t_srtt = TCPTV_SRTTBASE; 668 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 669 tp->t_rttmin = tcp_rexmit_min; 670 tp->t_rxtcur = TCPTV_RTOBASE; 671 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 672 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 673 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 674 tp->t_rcvtime = ticks; 675 tp->t_bw_rtttime = ticks; 676 /* 677 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 678 * because the socket may be bound to an IPv6 wildcard address, 679 * which may match an IPv4-mapped IPv6 address. 680 */ 681 inp->inp_ip_ttl = V_ip_defttl; 682 inp->inp_ppcb = tp; 683 return (tp); /* XXX */ 684 } 685 686 /* 687 * Drop a TCP connection, reporting 688 * the specified error. If connection is synchronized, 689 * then send a RST to peer. 690 */ 691 struct tcpcb * 692 tcp_drop(struct tcpcb *tp, int errno) 693 { 694 INIT_VNET_INET(tp->t_inpcb->inp_vnet); 695 struct socket *so = tp->t_inpcb->inp_socket; 696 697 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 698 INP_WLOCK_ASSERT(tp->t_inpcb); 699 700 if (TCPS_HAVERCVDSYN(tp->t_state)) { 701 tp->t_state = TCPS_CLOSED; 702 (void) tcp_output_reset(tp); 703 V_tcpstat.tcps_drops++; 704 } else 705 V_tcpstat.tcps_conndrops++; 706 if (errno == ETIMEDOUT && tp->t_softerror) 707 errno = tp->t_softerror; 708 so->so_error = errno; 709 return (tcp_close(tp)); 710 } 711 712 void 713 tcp_discardcb(struct tcpcb *tp) 714 { 715 INIT_VNET_INET(tp->t_vnet); 716 struct tseg_qent *q; 717 struct inpcb *inp = tp->t_inpcb; 718 struct socket *so = inp->inp_socket; 719 #ifdef INET6 720 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 721 #endif /* INET6 */ 722 723 INP_WLOCK_ASSERT(inp); 724 725 /* 726 * Make sure that all of our timers are stopped before we 727 * delete the PCB. 728 */ 729 callout_stop(&tp->t_timers->tt_rexmt); 730 callout_stop(&tp->t_timers->tt_persist); 731 callout_stop(&tp->t_timers->tt_keep); 732 callout_stop(&tp->t_timers->tt_2msl); 733 callout_stop(&tp->t_timers->tt_delack); 734 735 /* 736 * If we got enough samples through the srtt filter, 737 * save the rtt and rttvar in the routing entry. 738 * 'Enough' is arbitrarily defined as 4 rtt samples. 739 * 4 samples is enough for the srtt filter to converge 740 * to within enough % of the correct value; fewer samples 741 * and we could save a bogus rtt. The danger is not high 742 * as tcp quickly recovers from everything. 743 * XXX: Works very well but needs some more statistics! 744 */ 745 if (tp->t_rttupdated >= 4) { 746 struct hc_metrics_lite metrics; 747 u_long ssthresh; 748 749 bzero(&metrics, sizeof(metrics)); 750 /* 751 * Update the ssthresh always when the conditions below 752 * are satisfied. This gives us better new start value 753 * for the congestion avoidance for new connections. 754 * ssthresh is only set if packet loss occured on a session. 755 * 756 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 757 * being torn down. Ideally this code would not use 'so'. 758 */ 759 ssthresh = tp->snd_ssthresh; 760 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 761 /* 762 * convert the limit from user data bytes to 763 * packets then to packet data bytes. 764 */ 765 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 766 if (ssthresh < 2) 767 ssthresh = 2; 768 ssthresh *= (u_long)(tp->t_maxseg + 769 #ifdef INET6 770 (isipv6 ? sizeof (struct ip6_hdr) + 771 sizeof (struct tcphdr) : 772 #endif 773 sizeof (struct tcpiphdr) 774 #ifdef INET6 775 ) 776 #endif 777 ); 778 } else 779 ssthresh = 0; 780 metrics.rmx_ssthresh = ssthresh; 781 782 metrics.rmx_rtt = tp->t_srtt; 783 metrics.rmx_rttvar = tp->t_rttvar; 784 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 785 metrics.rmx_bandwidth = tp->snd_bandwidth; 786 metrics.rmx_cwnd = tp->snd_cwnd; 787 metrics.rmx_sendpipe = 0; 788 metrics.rmx_recvpipe = 0; 789 790 tcp_hc_update(&inp->inp_inc, &metrics); 791 } 792 793 /* free the reassembly queue, if any */ 794 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 795 LIST_REMOVE(q, tqe_q); 796 m_freem(q->tqe_m); 797 uma_zfree(tcp_reass_zone, q); 798 tp->t_segqlen--; 799 V_tcp_reass_qsize--; 800 } 801 /* Disconnect offload device, if any. */ 802 tcp_offload_detach(tp); 803 804 tcp_free_sackholes(tp); 805 inp->inp_ppcb = NULL; 806 tp->t_inpcb = NULL; 807 uma_zfree(tcpcb_zone, tp); 808 } 809 810 /* 811 * Attempt to close a TCP control block, marking it as dropped, and freeing 812 * the socket if we hold the only reference. 813 */ 814 struct tcpcb * 815 tcp_close(struct tcpcb *tp) 816 { 817 INIT_VNET_INET(tp->t_inpcb->inp_vnet); 818 struct inpcb *inp = tp->t_inpcb; 819 struct socket *so; 820 821 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 822 INP_WLOCK_ASSERT(inp); 823 824 /* Notify any offload devices of listener close */ 825 if (tp->t_state == TCPS_LISTEN) 826 tcp_offload_listen_close(tp); 827 in_pcbdrop(inp); 828 V_tcpstat.tcps_closed++; 829 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 830 so = inp->inp_socket; 831 soisdisconnected(so); 832 if (inp->inp_vflag & INP_SOCKREF) { 833 KASSERT(so->so_state & SS_PROTOREF, 834 ("tcp_close: !SS_PROTOREF")); 835 inp->inp_vflag &= ~INP_SOCKREF; 836 INP_WUNLOCK(inp); 837 ACCEPT_LOCK(); 838 SOCK_LOCK(so); 839 so->so_state &= ~SS_PROTOREF; 840 sofree(so); 841 return (NULL); 842 } 843 return (tp); 844 } 845 846 void 847 tcp_drain(void) 848 { 849 VNET_ITERATOR_DECL(vnet_iter); 850 851 if (!do_tcpdrain) 852 return; 853 854 VNET_LIST_RLOCK(); 855 VNET_FOREACH(vnet_iter) { 856 CURVNET_SET(vnet_iter); 857 INIT_VNET_INET(vnet_iter); 858 struct inpcb *inpb; 859 struct tcpcb *tcpb; 860 struct tseg_qent *te; 861 862 /* 863 * Walk the tcpbs, if existing, and flush the reassembly queue, 864 * if there is one... 865 * XXX: The "Net/3" implementation doesn't imply that the TCP 866 * reassembly queue should be flushed, but in a situation 867 * where we're really low on mbufs, this is potentially 868 * usefull. 869 */ 870 INP_INFO_RLOCK(&V_tcbinfo); 871 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 872 if (inpb->inp_vflag & INP_TIMEWAIT) 873 continue; 874 INP_WLOCK(inpb); 875 if ((tcpb = intotcpcb(inpb)) != NULL) { 876 while ((te = LIST_FIRST(&tcpb->t_segq)) 877 != NULL) { 878 LIST_REMOVE(te, tqe_q); 879 m_freem(te->tqe_m); 880 uma_zfree(tcp_reass_zone, te); 881 tcpb->t_segqlen--; 882 V_tcp_reass_qsize--; 883 } 884 tcp_clean_sackreport(tcpb); 885 } 886 INP_WUNLOCK(inpb); 887 } 888 INP_INFO_RUNLOCK(&V_tcbinfo); 889 CURVNET_RESTORE(); 890 } 891 VNET_LIST_RUNLOCK(); 892 } 893 894 /* 895 * Notify a tcp user of an asynchronous error; 896 * store error as soft error, but wake up user 897 * (for now, won't do anything until can select for soft error). 898 * 899 * Do not wake up user since there currently is no mechanism for 900 * reporting soft errors (yet - a kqueue filter may be added). 901 */ 902 static struct inpcb * 903 tcp_notify(struct inpcb *inp, int error) 904 { 905 struct tcpcb *tp; 906 907 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 908 INP_WLOCK_ASSERT(inp); 909 910 if ((inp->inp_vflag & INP_TIMEWAIT) || 911 (inp->inp_vflag & INP_DROPPED)) 912 return (inp); 913 914 tp = intotcpcb(inp); 915 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 916 917 /* 918 * Ignore some errors if we are hooked up. 919 * If connection hasn't completed, has retransmitted several times, 920 * and receives a second error, give up now. This is better 921 * than waiting a long time to establish a connection that 922 * can never complete. 923 */ 924 if (tp->t_state == TCPS_ESTABLISHED && 925 (error == EHOSTUNREACH || error == ENETUNREACH || 926 error == EHOSTDOWN)) { 927 return (inp); 928 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 929 tp->t_softerror) { 930 tp = tcp_drop(tp, error); 931 if (tp != NULL) 932 return (inp); 933 else 934 return (NULL); 935 } else { 936 tp->t_softerror = error; 937 return (inp); 938 } 939 #if 0 940 wakeup( &so->so_timeo); 941 sorwakeup(so); 942 sowwakeup(so); 943 #endif 944 } 945 946 static int 947 tcp_pcblist(SYSCTL_HANDLER_ARGS) 948 { 949 INIT_VNET_INET(curvnet); 950 int error, i, m, n, pcb_count; 951 struct inpcb *inp, **inp_list; 952 inp_gen_t gencnt; 953 struct xinpgen xig; 954 955 /* 956 * The process of preparing the TCB list is too time-consuming and 957 * resource-intensive to repeat twice on every request. 958 */ 959 if (req->oldptr == NULL) { 960 m = syncache_pcbcount(); 961 n = V_tcbinfo.ipi_count; 962 req->oldidx = 2 * (sizeof xig) 963 + ((m + n) + n/8) * sizeof(struct xtcpcb); 964 return (0); 965 } 966 967 if (req->newptr != NULL) 968 return (EPERM); 969 970 /* 971 * OK, now we're committed to doing something. 972 */ 973 INP_INFO_RLOCK(&V_tcbinfo); 974 gencnt = V_tcbinfo.ipi_gencnt; 975 n = V_tcbinfo.ipi_count; 976 INP_INFO_RUNLOCK(&V_tcbinfo); 977 978 m = syncache_pcbcount(); 979 980 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 981 + (n + m) * sizeof(struct xtcpcb)); 982 if (error != 0) 983 return (error); 984 985 xig.xig_len = sizeof xig; 986 xig.xig_count = n + m; 987 xig.xig_gen = gencnt; 988 xig.xig_sogen = so_gencnt; 989 error = SYSCTL_OUT(req, &xig, sizeof xig); 990 if (error) 991 return (error); 992 993 error = syncache_pcblist(req, m, &pcb_count); 994 if (error) 995 return (error); 996 997 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 998 if (inp_list == NULL) 999 return (ENOMEM); 1000 1001 INP_INFO_RLOCK(&V_tcbinfo); 1002 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; 1003 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) { 1004 INP_RLOCK(inp); 1005 if (inp->inp_gencnt <= gencnt) { 1006 /* 1007 * XXX: This use of cr_cansee(), introduced with 1008 * TCP state changes, is not quite right, but for 1009 * now, better than nothing. 1010 */ 1011 if (inp->inp_vflag & INP_TIMEWAIT) { 1012 if (intotw(inp) != NULL) 1013 error = cr_cansee(req->td->td_ucred, 1014 intotw(inp)->tw_cred); 1015 else 1016 error = EINVAL; /* Skip this inp. */ 1017 } else 1018 error = cr_canseesocket(req->td->td_ucred, 1019 inp->inp_socket); 1020 if (error == 0) 1021 inp_list[i++] = inp; 1022 } 1023 INP_RUNLOCK(inp); 1024 } 1025 INP_INFO_RUNLOCK(&V_tcbinfo); 1026 n = i; 1027 1028 error = 0; 1029 for (i = 0; i < n; i++) { 1030 inp = inp_list[i]; 1031 INP_RLOCK(inp); 1032 if (inp->inp_gencnt <= gencnt) { 1033 struct xtcpcb xt; 1034 void *inp_ppcb; 1035 1036 bzero(&xt, sizeof(xt)); 1037 xt.xt_len = sizeof xt; 1038 /* XXX should avoid extra copy */ 1039 bcopy(inp, &xt.xt_inp, sizeof *inp); 1040 inp_ppcb = inp->inp_ppcb; 1041 if (inp_ppcb == NULL) 1042 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1043 else if (inp->inp_vflag & INP_TIMEWAIT) { 1044 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1045 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1046 } else 1047 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1048 if (inp->inp_socket != NULL) 1049 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1050 else { 1051 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1052 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1053 } 1054 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1055 INP_RUNLOCK(inp); 1056 error = SYSCTL_OUT(req, &xt, sizeof xt); 1057 } else 1058 INP_RUNLOCK(inp); 1059 1060 } 1061 if (!error) { 1062 /* 1063 * Give the user an updated idea of our state. 1064 * If the generation differs from what we told 1065 * her before, she knows that something happened 1066 * while we were processing this request, and it 1067 * might be necessary to retry. 1068 */ 1069 INP_INFO_RLOCK(&V_tcbinfo); 1070 xig.xig_gen = V_tcbinfo.ipi_gencnt; 1071 xig.xig_sogen = so_gencnt; 1072 xig.xig_count = V_tcbinfo.ipi_count + pcb_count; 1073 INP_INFO_RUNLOCK(&V_tcbinfo); 1074 error = SYSCTL_OUT(req, &xig, sizeof xig); 1075 } 1076 free(inp_list, M_TEMP); 1077 return (error); 1078 } 1079 1080 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1081 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1082 1083 static int 1084 tcp_getcred(SYSCTL_HANDLER_ARGS) 1085 { 1086 INIT_VNET_INET(curvnet); 1087 struct xucred xuc; 1088 struct sockaddr_in addrs[2]; 1089 struct inpcb *inp; 1090 int error; 1091 1092 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1093 if (error) 1094 return (error); 1095 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1096 if (error) 1097 return (error); 1098 INP_INFO_RLOCK(&V_tcbinfo); 1099 inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr, 1100 addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1101 if (inp != NULL) { 1102 INP_RLOCK(inp); 1103 INP_INFO_RUNLOCK(&V_tcbinfo); 1104 if (inp->inp_socket == NULL) 1105 error = ENOENT; 1106 if (error == 0) 1107 error = cr_canseesocket(req->td->td_ucred, 1108 inp->inp_socket); 1109 if (error == 0) 1110 cru2x(inp->inp_cred, &xuc); 1111 INP_RUNLOCK(inp); 1112 } else { 1113 INP_INFO_RUNLOCK(&V_tcbinfo); 1114 error = ENOENT; 1115 } 1116 if (error == 0) 1117 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1118 return (error); 1119 } 1120 1121 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1122 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1123 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1124 1125 #ifdef INET6 1126 static int 1127 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1128 { 1129 INIT_VNET_INET(curvnet); 1130 INIT_VNET_INET6(curvnet); 1131 struct xucred xuc; 1132 struct sockaddr_in6 addrs[2]; 1133 struct inpcb *inp; 1134 int error, mapped = 0; 1135 1136 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1137 if (error) 1138 return (error); 1139 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1140 if (error) 1141 return (error); 1142 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 1143 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 1144 return (error); 1145 } 1146 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1147 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1148 mapped = 1; 1149 else 1150 return (EINVAL); 1151 } 1152 1153 INP_INFO_RLOCK(&V_tcbinfo); 1154 if (mapped == 1) 1155 inp = in_pcblookup_hash(&V_tcbinfo, 1156 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1157 addrs[1].sin6_port, 1158 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1159 addrs[0].sin6_port, 1160 0, NULL); 1161 else 1162 inp = in6_pcblookup_hash(&V_tcbinfo, 1163 &addrs[1].sin6_addr, addrs[1].sin6_port, 1164 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1165 if (inp != NULL) { 1166 INP_RLOCK(inp); 1167 INP_INFO_RUNLOCK(&V_tcbinfo); 1168 if (inp->inp_socket == NULL) 1169 error = ENOENT; 1170 if (error == 0) 1171 error = cr_canseesocket(req->td->td_ucred, 1172 inp->inp_socket); 1173 if (error == 0) 1174 cru2x(inp->inp_cred, &xuc); 1175 INP_RUNLOCK(inp); 1176 } else { 1177 INP_INFO_RUNLOCK(&V_tcbinfo); 1178 error = ENOENT; 1179 } 1180 if (error == 0) 1181 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1182 return (error); 1183 } 1184 1185 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1186 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1187 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1188 #endif 1189 1190 1191 void 1192 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1193 { 1194 INIT_VNET_INET(curvnet); 1195 struct ip *ip = vip; 1196 struct tcphdr *th; 1197 struct in_addr faddr; 1198 struct inpcb *inp; 1199 struct tcpcb *tp; 1200 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1201 struct icmp *icp; 1202 struct in_conninfo inc; 1203 tcp_seq icmp_tcp_seq; 1204 int mtu; 1205 1206 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1207 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1208 return; 1209 1210 if (cmd == PRC_MSGSIZE) 1211 notify = tcp_mtudisc; 1212 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1213 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1214 notify = tcp_drop_syn_sent; 1215 /* 1216 * Redirects don't need to be handled up here. 1217 */ 1218 else if (PRC_IS_REDIRECT(cmd)) 1219 return; 1220 /* 1221 * Source quench is depreciated. 1222 */ 1223 else if (cmd == PRC_QUENCH) 1224 return; 1225 /* 1226 * Hostdead is ugly because it goes linearly through all PCBs. 1227 * XXX: We never get this from ICMP, otherwise it makes an 1228 * excellent DoS attack on machines with many connections. 1229 */ 1230 else if (cmd == PRC_HOSTDEAD) 1231 ip = NULL; 1232 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1233 return; 1234 if (ip != NULL) { 1235 icp = (struct icmp *)((caddr_t)ip 1236 - offsetof(struct icmp, icmp_ip)); 1237 th = (struct tcphdr *)((caddr_t)ip 1238 + (ip->ip_hl << 2)); 1239 INP_INFO_WLOCK(&V_tcbinfo); 1240 inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport, 1241 ip->ip_src, th->th_sport, 0, NULL); 1242 if (inp != NULL) { 1243 INP_WLOCK(inp); 1244 if (!(inp->inp_vflag & INP_TIMEWAIT) && 1245 !(inp->inp_vflag & INP_DROPPED) && 1246 !(inp->inp_socket == NULL)) { 1247 icmp_tcp_seq = htonl(th->th_seq); 1248 tp = intotcpcb(inp); 1249 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1250 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1251 if (cmd == PRC_MSGSIZE) { 1252 /* 1253 * MTU discovery: 1254 * If we got a needfrag set the MTU 1255 * in the route to the suggested new 1256 * value (if given) and then notify. 1257 */ 1258 bzero(&inc, sizeof(inc)); 1259 inc.inc_flags = 0; /* IPv4 */ 1260 inc.inc_faddr = faddr; 1261 inc.inc_fibnum = 1262 inp->inp_inc.inc_fibnum; 1263 1264 mtu = ntohs(icp->icmp_nextmtu); 1265 /* 1266 * If no alternative MTU was 1267 * proposed, try the next smaller 1268 * one. ip->ip_len has already 1269 * been swapped in icmp_input(). 1270 */ 1271 if (!mtu) 1272 mtu = ip_next_mtu(ip->ip_len, 1273 1); 1274 if (mtu < max(296, V_tcp_minmss 1275 + sizeof(struct tcpiphdr))) 1276 mtu = 0; 1277 if (!mtu) 1278 mtu = V_tcp_mssdflt 1279 + sizeof(struct tcpiphdr); 1280 /* 1281 * Only cache the the MTU if it 1282 * is smaller than the interface 1283 * or route MTU. tcp_mtudisc() 1284 * will do right thing by itself. 1285 */ 1286 if (mtu <= tcp_maxmtu(&inc, NULL)) 1287 tcp_hc_updatemtu(&inc, mtu); 1288 } 1289 1290 inp = (*notify)(inp, inetctlerrmap[cmd]); 1291 } 1292 } 1293 if (inp != NULL) 1294 INP_WUNLOCK(inp); 1295 } else { 1296 inc.inc_fport = th->th_dport; 1297 inc.inc_lport = th->th_sport; 1298 inc.inc_faddr = faddr; 1299 inc.inc_laddr = ip->ip_src; 1300 #ifdef INET6 1301 inc.inc_isipv6 = 0; 1302 #endif 1303 syncache_unreach(&inc, th); 1304 } 1305 INP_INFO_WUNLOCK(&V_tcbinfo); 1306 } else 1307 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 1308 } 1309 1310 #ifdef INET6 1311 void 1312 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1313 { 1314 INIT_VNET_INET(curvnet); 1315 struct tcphdr th; 1316 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1317 struct ip6_hdr *ip6; 1318 struct mbuf *m; 1319 struct ip6ctlparam *ip6cp = NULL; 1320 const struct sockaddr_in6 *sa6_src = NULL; 1321 int off; 1322 struct tcp_portonly { 1323 u_int16_t th_sport; 1324 u_int16_t th_dport; 1325 } *thp; 1326 1327 if (sa->sa_family != AF_INET6 || 1328 sa->sa_len != sizeof(struct sockaddr_in6)) 1329 return; 1330 1331 if (cmd == PRC_MSGSIZE) 1332 notify = tcp_mtudisc; 1333 else if (!PRC_IS_REDIRECT(cmd) && 1334 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1335 return; 1336 /* Source quench is depreciated. */ 1337 else if (cmd == PRC_QUENCH) 1338 return; 1339 1340 /* if the parameter is from icmp6, decode it. */ 1341 if (d != NULL) { 1342 ip6cp = (struct ip6ctlparam *)d; 1343 m = ip6cp->ip6c_m; 1344 ip6 = ip6cp->ip6c_ip6; 1345 off = ip6cp->ip6c_off; 1346 sa6_src = ip6cp->ip6c_src; 1347 } else { 1348 m = NULL; 1349 ip6 = NULL; 1350 off = 0; /* fool gcc */ 1351 sa6_src = &sa6_any; 1352 } 1353 1354 if (ip6 != NULL) { 1355 struct in_conninfo inc; 1356 /* 1357 * XXX: We assume that when IPV6 is non NULL, 1358 * M and OFF are valid. 1359 */ 1360 1361 /* check if we can safely examine src and dst ports */ 1362 if (m->m_pkthdr.len < off + sizeof(*thp)) 1363 return; 1364 1365 bzero(&th, sizeof(th)); 1366 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1367 1368 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport, 1369 (struct sockaddr *)ip6cp->ip6c_src, 1370 th.th_sport, cmd, NULL, notify); 1371 1372 inc.inc_fport = th.th_dport; 1373 inc.inc_lport = th.th_sport; 1374 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1375 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1376 inc.inc_isipv6 = 1; 1377 INP_INFO_WLOCK(&V_tcbinfo); 1378 syncache_unreach(&inc, &th); 1379 INP_INFO_WUNLOCK(&V_tcbinfo); 1380 } else 1381 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1382 0, cmd, NULL, notify); 1383 } 1384 #endif /* INET6 */ 1385 1386 1387 /* 1388 * Following is where TCP initial sequence number generation occurs. 1389 * 1390 * There are two places where we must use initial sequence numbers: 1391 * 1. In SYN-ACK packets. 1392 * 2. In SYN packets. 1393 * 1394 * All ISNs for SYN-ACK packets are generated by the syncache. See 1395 * tcp_syncache.c for details. 1396 * 1397 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1398 * depends on this property. In addition, these ISNs should be 1399 * unguessable so as to prevent connection hijacking. To satisfy 1400 * the requirements of this situation, the algorithm outlined in 1401 * RFC 1948 is used, with only small modifications. 1402 * 1403 * Implementation details: 1404 * 1405 * Time is based off the system timer, and is corrected so that it 1406 * increases by one megabyte per second. This allows for proper 1407 * recycling on high speed LANs while still leaving over an hour 1408 * before rollover. 1409 * 1410 * As reading the *exact* system time is too expensive to be done 1411 * whenever setting up a TCP connection, we increment the time 1412 * offset in two ways. First, a small random positive increment 1413 * is added to isn_offset for each connection that is set up. 1414 * Second, the function tcp_isn_tick fires once per clock tick 1415 * and increments isn_offset as necessary so that sequence numbers 1416 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1417 * random positive increments serve only to ensure that the same 1418 * exact sequence number is never sent out twice (as could otherwise 1419 * happen when a port is recycled in less than the system tick 1420 * interval.) 1421 * 1422 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1423 * between seeding of isn_secret. This is normally set to zero, 1424 * as reseeding should not be necessary. 1425 * 1426 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1427 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1428 * general, this means holding an exclusive (write) lock. 1429 */ 1430 1431 #define ISN_BYTES_PER_SECOND 1048576 1432 #define ISN_STATIC_INCREMENT 4096 1433 #define ISN_RANDOM_INCREMENT (4096 - 1) 1434 1435 static u_char isn_secret[32]; 1436 static int isn_last_reseed; 1437 static u_int32_t isn_offset, isn_offset_old; 1438 static MD5_CTX isn_ctx; 1439 1440 tcp_seq 1441 tcp_new_isn(struct tcpcb *tp) 1442 { 1443 INIT_VNET_INET(tp->t_vnet); 1444 u_int32_t md5_buffer[4]; 1445 tcp_seq new_isn; 1446 1447 INP_WLOCK_ASSERT(tp->t_inpcb); 1448 1449 ISN_LOCK(); 1450 /* Seed if this is the first use, reseed if requested. */ 1451 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 1452 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 1453 < (u_int)ticks))) { 1454 read_random(&V_isn_secret, sizeof(V_isn_secret)); 1455 V_isn_last_reseed = ticks; 1456 } 1457 1458 /* Compute the md5 hash and return the ISN. */ 1459 MD5Init(&V_isn_ctx); 1460 MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1461 MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1462 #ifdef INET6 1463 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1464 MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1465 sizeof(struct in6_addr)); 1466 MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1467 sizeof(struct in6_addr)); 1468 } else 1469 #endif 1470 { 1471 MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1472 sizeof(struct in_addr)); 1473 MD5Update(&V_isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1474 sizeof(struct in_addr)); 1475 } 1476 MD5Update(&V_isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret)); 1477 MD5Final((u_char *) &md5_buffer, &V_isn_ctx); 1478 new_isn = (tcp_seq) md5_buffer[0]; 1479 V_isn_offset += ISN_STATIC_INCREMENT + 1480 (arc4random() & ISN_RANDOM_INCREMENT); 1481 new_isn += V_isn_offset; 1482 ISN_UNLOCK(); 1483 return (new_isn); 1484 } 1485 1486 /* 1487 * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary 1488 * to keep time flowing at a relatively constant rate. If the random 1489 * increments have already pushed us past the projected offset, do nothing. 1490 */ 1491 static void 1492 tcp_isn_tick(void *xtp) 1493 { 1494 VNET_ITERATOR_DECL(vnet_iter); 1495 u_int32_t projected_offset; 1496 1497 ISN_LOCK(); 1498 VNET_LIST_RLOCK(); 1499 VNET_FOREACH(vnet_iter) { 1500 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */ 1501 INIT_VNET_INET(curvnet); 1502 projected_offset = 1503 V_isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1504 1505 if (SEQ_GT(projected_offset, V_isn_offset)) 1506 V_isn_offset = projected_offset; 1507 1508 V_isn_offset_old = V_isn_offset; 1509 CURVNET_RESTORE(); 1510 } 1511 VNET_LIST_RUNLOCK(); 1512 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1513 ISN_UNLOCK(); 1514 } 1515 1516 /* 1517 * When a specific ICMP unreachable message is received and the 1518 * connection state is SYN-SENT, drop the connection. This behavior 1519 * is controlled by the icmp_may_rst sysctl. 1520 */ 1521 struct inpcb * 1522 tcp_drop_syn_sent(struct inpcb *inp, int errno) 1523 { 1524 #ifdef INVARIANTS 1525 INIT_VNET_INET(inp->inp_vnet); 1526 #endif 1527 struct tcpcb *tp; 1528 1529 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1530 INP_WLOCK_ASSERT(inp); 1531 1532 if ((inp->inp_vflag & INP_TIMEWAIT) || 1533 (inp->inp_vflag & INP_DROPPED)) 1534 return (inp); 1535 1536 tp = intotcpcb(inp); 1537 if (tp->t_state != TCPS_SYN_SENT) 1538 return (inp); 1539 1540 tp = tcp_drop(tp, errno); 1541 if (tp != NULL) 1542 return (inp); 1543 else 1544 return (NULL); 1545 } 1546 1547 /* 1548 * When `need fragmentation' ICMP is received, update our idea of the MSS 1549 * based on the new value in the route. Also nudge TCP to send something, 1550 * since we know the packet we just sent was dropped. 1551 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1552 */ 1553 struct inpcb * 1554 tcp_mtudisc(struct inpcb *inp, int errno) 1555 { 1556 INIT_VNET_INET(inp->inp_vnet); 1557 struct tcpcb *tp; 1558 struct socket *so; 1559 1560 INP_WLOCK_ASSERT(inp); 1561 if ((inp->inp_vflag & INP_TIMEWAIT) || 1562 (inp->inp_vflag & INP_DROPPED)) 1563 return (inp); 1564 1565 tp = intotcpcb(inp); 1566 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1567 1568 tcp_mss_update(tp, -1, NULL); 1569 1570 so = inp->inp_socket; 1571 SOCKBUF_LOCK(&so->so_snd); 1572 /* If the mss is larger than the socket buffer, decrease the mss. */ 1573 if (so->so_snd.sb_hiwat < tp->t_maxseg) 1574 tp->t_maxseg = so->so_snd.sb_hiwat; 1575 SOCKBUF_UNLOCK(&so->so_snd); 1576 1577 V_tcpstat.tcps_mturesent++; 1578 tp->t_rtttime = 0; 1579 tp->snd_nxt = tp->snd_una; 1580 tcp_free_sackholes(tp); 1581 tp->snd_recover = tp->snd_max; 1582 if (tp->t_flags & TF_SACK_PERMIT) 1583 EXIT_FASTRECOVERY(tp); 1584 tcp_output_send(tp); 1585 return (inp); 1586 } 1587 1588 /* 1589 * Look-up the routing entry to the peer of this inpcb. If no route 1590 * is found and it cannot be allocated, then return NULL. This routine 1591 * is called by TCP routines that access the rmx structure and by tcp_mss 1592 * to get the interface MTU. 1593 */ 1594 u_long 1595 tcp_maxmtu(struct in_conninfo *inc, int *flags) 1596 { 1597 struct route sro; 1598 struct sockaddr_in *dst; 1599 struct ifnet *ifp; 1600 u_long maxmtu = 0; 1601 1602 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1603 1604 bzero(&sro, sizeof(sro)); 1605 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1606 dst = (struct sockaddr_in *)&sro.ro_dst; 1607 dst->sin_family = AF_INET; 1608 dst->sin_len = sizeof(*dst); 1609 dst->sin_addr = inc->inc_faddr; 1610 in_rtalloc_ign(&sro, RTF_CLONING, inc->inc_fibnum); 1611 } 1612 if (sro.ro_rt != NULL) { 1613 ifp = sro.ro_rt->rt_ifp; 1614 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1615 maxmtu = ifp->if_mtu; 1616 else 1617 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1618 1619 /* Report additional interface capabilities. */ 1620 if (flags != NULL) { 1621 if (ifp->if_capenable & IFCAP_TSO4 && 1622 ifp->if_hwassist & CSUM_TSO) 1623 *flags |= CSUM_TSO; 1624 } 1625 RTFREE(sro.ro_rt); 1626 } 1627 return (maxmtu); 1628 } 1629 1630 #ifdef INET6 1631 u_long 1632 tcp_maxmtu6(struct in_conninfo *inc, int *flags) 1633 { 1634 struct route_in6 sro6; 1635 struct ifnet *ifp; 1636 u_long maxmtu = 0; 1637 1638 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1639 1640 bzero(&sro6, sizeof(sro6)); 1641 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1642 sro6.ro_dst.sin6_family = AF_INET6; 1643 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1644 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1645 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1646 } 1647 if (sro6.ro_rt != NULL) { 1648 ifp = sro6.ro_rt->rt_ifp; 1649 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1650 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1651 else 1652 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1653 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1654 1655 /* Report additional interface capabilities. */ 1656 if (flags != NULL) { 1657 if (ifp->if_capenable & IFCAP_TSO6 && 1658 ifp->if_hwassist & CSUM_TSO) 1659 *flags |= CSUM_TSO; 1660 } 1661 RTFREE(sro6.ro_rt); 1662 } 1663 1664 return (maxmtu); 1665 } 1666 #endif /* INET6 */ 1667 1668 #ifdef IPSEC 1669 /* compute ESP/AH header size for TCP, including outer IP header. */ 1670 size_t 1671 ipsec_hdrsiz_tcp(struct tcpcb *tp) 1672 { 1673 struct inpcb *inp; 1674 struct mbuf *m; 1675 size_t hdrsiz; 1676 struct ip *ip; 1677 #ifdef INET6 1678 struct ip6_hdr *ip6; 1679 #endif 1680 struct tcphdr *th; 1681 1682 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1683 return (0); 1684 MGETHDR(m, M_DONTWAIT, MT_DATA); 1685 if (!m) 1686 return (0); 1687 1688 #ifdef INET6 1689 if ((inp->inp_vflag & INP_IPV6) != 0) { 1690 ip6 = mtod(m, struct ip6_hdr *); 1691 th = (struct tcphdr *)(ip6 + 1); 1692 m->m_pkthdr.len = m->m_len = 1693 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1694 tcpip_fillheaders(inp, ip6, th); 1695 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1696 } else 1697 #endif /* INET6 */ 1698 { 1699 ip = mtod(m, struct ip *); 1700 th = (struct tcphdr *)(ip + 1); 1701 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1702 tcpip_fillheaders(inp, ip, th); 1703 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1704 } 1705 1706 m_free(m); 1707 return (hdrsiz); 1708 } 1709 #endif /* IPSEC */ 1710 1711 /* 1712 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1713 * 1714 * This code attempts to calculate the bandwidth-delay product as a 1715 * means of determining the optimal window size to maximize bandwidth, 1716 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1717 * routers. This code also does a fairly good job keeping RTTs in check 1718 * across slow links like modems. We implement an algorithm which is very 1719 * similar (but not meant to be) TCP/Vegas. The code operates on the 1720 * transmitter side of a TCP connection and so only effects the transmit 1721 * side of the connection. 1722 * 1723 * BACKGROUND: TCP makes no provision for the management of buffer space 1724 * at the end points or at the intermediate routers and switches. A TCP 1725 * stream, whether using NewReno or not, will eventually buffer as 1726 * many packets as it is able and the only reason this typically works is 1727 * due to the fairly small default buffers made available for a connection 1728 * (typicaly 16K or 32K). As machines use larger windows and/or window 1729 * scaling it is now fairly easy for even a single TCP connection to blow-out 1730 * all available buffer space not only on the local interface, but on 1731 * intermediate routers and switches as well. NewReno makes a misguided 1732 * attempt to 'solve' this problem by waiting for an actual failure to occur, 1733 * then backing off, then steadily increasing the window again until another 1734 * failure occurs, ad-infinitum. This results in terrible oscillation that 1735 * is only made worse as network loads increase and the idea of intentionally 1736 * blowing out network buffers is, frankly, a terrible way to manage network 1737 * resources. 1738 * 1739 * It is far better to limit the transmit window prior to the failure 1740 * condition being achieved. There are two general ways to do this: First 1741 * you can 'scan' through different transmit window sizes and locate the 1742 * point where the RTT stops increasing, indicating that you have filled the 1743 * pipe, then scan backwards until you note that RTT stops decreasing, then 1744 * repeat ad-infinitum. This method works in principle but has severe 1745 * implementation issues due to RTT variances, timer granularity, and 1746 * instability in the algorithm which can lead to many false positives and 1747 * create oscillations as well as interact badly with other TCP streams 1748 * implementing the same algorithm. 1749 * 1750 * The second method is to limit the window to the bandwidth delay product 1751 * of the link. This is the method we implement. RTT variances and our 1752 * own manipulation of the congestion window, bwnd, can potentially 1753 * destabilize the algorithm. For this reason we have to stabilize the 1754 * elements used to calculate the window. We do this by using the minimum 1755 * observed RTT, the long term average of the observed bandwidth, and 1756 * by adding two segments worth of slop. It isn't perfect but it is able 1757 * to react to changing conditions and gives us a very stable basis on 1758 * which to extend the algorithm. 1759 */ 1760 void 1761 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 1762 { 1763 INIT_VNET_INET(tp->t_vnet); 1764 u_long bw; 1765 u_long bwnd; 1766 int save_ticks; 1767 1768 INP_WLOCK_ASSERT(tp->t_inpcb); 1769 1770 /* 1771 * If inflight_enable is disabled in the middle of a tcp connection, 1772 * make sure snd_bwnd is effectively disabled. 1773 */ 1774 if (V_tcp_inflight_enable == 0 || 1775 tp->t_rttlow < V_tcp_inflight_rttthresh) { 1776 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1777 tp->snd_bandwidth = 0; 1778 return; 1779 } 1780 1781 /* 1782 * Figure out the bandwidth. Due to the tick granularity this 1783 * is a very rough number and it MUST be averaged over a fairly 1784 * long period of time. XXX we need to take into account a link 1785 * that is not using all available bandwidth, but for now our 1786 * slop will ramp us up if this case occurs and the bandwidth later 1787 * increases. 1788 * 1789 * Note: if ticks rollover 'bw' may wind up negative. We must 1790 * effectively reset t_bw_rtttime for this case. 1791 */ 1792 save_ticks = ticks; 1793 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 1794 return; 1795 1796 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 1797 (save_ticks - tp->t_bw_rtttime); 1798 tp->t_bw_rtttime = save_ticks; 1799 tp->t_bw_rtseq = ack_seq; 1800 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 1801 return; 1802 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 1803 1804 tp->snd_bandwidth = bw; 1805 1806 /* 1807 * Calculate the semi-static bandwidth delay product, plus two maximal 1808 * segments. The additional slop puts us squarely in the sweet 1809 * spot and also handles the bandwidth run-up case and stabilization. 1810 * Without the slop we could be locking ourselves into a lower 1811 * bandwidth. 1812 * 1813 * Situations Handled: 1814 * (1) Prevents over-queueing of packets on LANs, especially on 1815 * high speed LANs, allowing larger TCP buffers to be 1816 * specified, and also does a good job preventing 1817 * over-queueing of packets over choke points like modems 1818 * (at least for the transmit side). 1819 * 1820 * (2) Is able to handle changing network loads (bandwidth 1821 * drops so bwnd drops, bandwidth increases so bwnd 1822 * increases). 1823 * 1824 * (3) Theoretically should stabilize in the face of multiple 1825 * connections implementing the same algorithm (this may need 1826 * a little work). 1827 * 1828 * (4) Stability value (defaults to 20 = 2 maximal packets) can 1829 * be adjusted with a sysctl but typically only needs to be 1830 * on very slow connections. A value no smaller then 5 1831 * should be used, but only reduce this default if you have 1832 * no other choice. 1833 */ 1834 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 1835 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10; 1836 #undef USERTT 1837 1838 if (tcp_inflight_debug > 0) { 1839 static int ltime; 1840 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 1841 ltime = ticks; 1842 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 1843 tp, 1844 bw, 1845 tp->t_rttbest, 1846 tp->t_srtt, 1847 bwnd 1848 ); 1849 } 1850 } 1851 if ((long)bwnd < V_tcp_inflight_min) 1852 bwnd = V_tcp_inflight_min; 1853 if (bwnd > V_tcp_inflight_max) 1854 bwnd = V_tcp_inflight_max; 1855 if ((long)bwnd < tp->t_maxseg * 2) 1856 bwnd = tp->t_maxseg * 2; 1857 tp->snd_bwnd = bwnd; 1858 } 1859 1860 #ifdef TCP_SIGNATURE 1861 /* 1862 * Callback function invoked by m_apply() to digest TCP segment data 1863 * contained within an mbuf chain. 1864 */ 1865 static int 1866 tcp_signature_apply(void *fstate, void *data, u_int len) 1867 { 1868 1869 MD5Update(fstate, (u_char *)data, len); 1870 return (0); 1871 } 1872 1873 /* 1874 * Compute TCP-MD5 hash of a TCP segment. (RFC2385) 1875 * 1876 * Parameters: 1877 * m pointer to head of mbuf chain 1878 * _unused 1879 * len length of TCP segment data, excluding options 1880 * optlen length of TCP segment options 1881 * buf pointer to storage for computed MD5 digest 1882 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 1883 * 1884 * We do this over ip, tcphdr, segment data, and the key in the SADB. 1885 * When called from tcp_input(), we can be sure that th_sum has been 1886 * zeroed out and verified already. 1887 * 1888 * Return 0 if successful, otherwise return -1. 1889 * 1890 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 1891 * search with the destination IP address, and a 'magic SPI' to be 1892 * determined by the application. This is hardcoded elsewhere to 1179 1893 * right now. Another branch of this code exists which uses the SPD to 1894 * specify per-application flows but it is unstable. 1895 */ 1896 int 1897 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen, 1898 u_char *buf, u_int direction) 1899 { 1900 union sockaddr_union dst; 1901 struct ippseudo ippseudo; 1902 MD5_CTX ctx; 1903 int doff; 1904 struct ip *ip; 1905 struct ipovly *ipovly; 1906 struct secasvar *sav; 1907 struct tcphdr *th; 1908 #ifdef INET6 1909 struct ip6_hdr *ip6; 1910 struct in6_addr in6; 1911 char ip6buf[INET6_ADDRSTRLEN]; 1912 uint32_t plen; 1913 uint16_t nhdr; 1914 #endif 1915 u_short savecsum; 1916 1917 KASSERT(m != NULL, ("NULL mbuf chain")); 1918 KASSERT(buf != NULL, ("NULL signature pointer")); 1919 1920 /* Extract the destination from the IP header in the mbuf. */ 1921 bzero(&dst, sizeof(union sockaddr_union)); 1922 ip = mtod(m, struct ip *); 1923 #ifdef INET6 1924 ip6 = NULL; /* Make the compiler happy. */ 1925 #endif 1926 switch (ip->ip_v) { 1927 case IPVERSION: 1928 dst.sa.sa_len = sizeof(struct sockaddr_in); 1929 dst.sa.sa_family = AF_INET; 1930 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 1931 ip->ip_src : ip->ip_dst; 1932 break; 1933 #ifdef INET6 1934 case (IPV6_VERSION >> 4): 1935 ip6 = mtod(m, struct ip6_hdr *); 1936 dst.sa.sa_len = sizeof(struct sockaddr_in6); 1937 dst.sa.sa_family = AF_INET6; 1938 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ? 1939 ip6->ip6_src : ip6->ip6_dst; 1940 break; 1941 #endif 1942 default: 1943 return (EINVAL); 1944 /* NOTREACHED */ 1945 break; 1946 } 1947 1948 /* Look up an SADB entry which matches the address of the peer. */ 1949 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 1950 if (sav == NULL) { 1951 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__, 1952 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) : 1953 #ifdef INET6 1954 (ip->ip_v == (IPV6_VERSION >> 4)) ? 1955 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) : 1956 #endif 1957 "(unsupported)")); 1958 return (EINVAL); 1959 } 1960 1961 MD5Init(&ctx); 1962 /* 1963 * Step 1: Update MD5 hash with IP(v6) pseudo-header. 1964 * 1965 * XXX The ippseudo header MUST be digested in network byte order, 1966 * or else we'll fail the regression test. Assume all fields we've 1967 * been doing arithmetic on have been in host byte order. 1968 * XXX One cannot depend on ipovly->ih_len here. When called from 1969 * tcp_output(), the underlying ip_len member has not yet been set. 1970 */ 1971 switch (ip->ip_v) { 1972 case IPVERSION: 1973 ipovly = (struct ipovly *)ip; 1974 ippseudo.ippseudo_src = ipovly->ih_src; 1975 ippseudo.ippseudo_dst = ipovly->ih_dst; 1976 ippseudo.ippseudo_pad = 0; 1977 ippseudo.ippseudo_p = IPPROTO_TCP; 1978 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + 1979 optlen); 1980 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 1981 1982 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip)); 1983 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen; 1984 break; 1985 #ifdef INET6 1986 /* 1987 * RFC 2385, 2.0 Proposal 1988 * For IPv6, the pseudo-header is as described in RFC 2460, namely the 1989 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero- 1990 * extended next header value (to form 32 bits), and 32-bit segment 1991 * length. 1992 * Note: Upper-Layer Packet Length comes before Next Header. 1993 */ 1994 case (IPV6_VERSION >> 4): 1995 in6 = ip6->ip6_src; 1996 in6_clearscope(&in6); 1997 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 1998 in6 = ip6->ip6_dst; 1999 in6_clearscope(&in6); 2000 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2001 plen = htonl(len + sizeof(struct tcphdr) + optlen); 2002 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t)); 2003 nhdr = 0; 2004 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2005 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2006 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2007 nhdr = IPPROTO_TCP; 2008 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2009 2010 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr)); 2011 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen; 2012 break; 2013 #endif 2014 default: 2015 return (EINVAL); 2016 /* NOTREACHED */ 2017 break; 2018 } 2019 2020 2021 /* 2022 * Step 2: Update MD5 hash with TCP header, excluding options. 2023 * The TCP checksum must be set to zero. 2024 */ 2025 savecsum = th->th_sum; 2026 th->th_sum = 0; 2027 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2028 th->th_sum = savecsum; 2029 2030 /* 2031 * Step 3: Update MD5 hash with TCP segment data. 2032 * Use m_apply() to avoid an early m_pullup(). 2033 */ 2034 if (len > 0) 2035 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2036 2037 /* 2038 * Step 4: Update MD5 hash with shared secret. 2039 */ 2040 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth)); 2041 MD5Final(buf, &ctx); 2042 2043 key_sa_recordxfer(sav, m); 2044 KEY_FREESAV(&sav); 2045 return (0); 2046 } 2047 #endif /* TCP_SIGNATURE */ 2048 2049 static int 2050 sysctl_drop(SYSCTL_HANDLER_ARGS) 2051 { 2052 INIT_VNET_INET(curvnet); 2053 #ifdef INET6 2054 INIT_VNET_INET6(curvnet); 2055 #endif 2056 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2057 struct sockaddr_storage addrs[2]; 2058 struct inpcb *inp; 2059 struct tcpcb *tp; 2060 struct tcptw *tw; 2061 struct sockaddr_in *fin, *lin; 2062 #ifdef INET6 2063 struct sockaddr_in6 *fin6, *lin6; 2064 struct in6_addr f6, l6; 2065 #endif 2066 int error; 2067 2068 inp = NULL; 2069 fin = lin = NULL; 2070 #ifdef INET6 2071 fin6 = lin6 = NULL; 2072 #endif 2073 error = 0; 2074 2075 if (req->oldptr != NULL || req->oldlen != 0) 2076 return (EINVAL); 2077 if (req->newptr == NULL) 2078 return (EPERM); 2079 if (req->newlen < sizeof(addrs)) 2080 return (ENOMEM); 2081 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2082 if (error) 2083 return (error); 2084 2085 switch (addrs[0].ss_family) { 2086 #ifdef INET6 2087 case AF_INET6: 2088 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2089 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2090 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2091 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2092 return (EINVAL); 2093 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2094 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2095 return (EINVAL); 2096 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2097 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2098 fin = (struct sockaddr_in *)&addrs[0]; 2099 lin = (struct sockaddr_in *)&addrs[1]; 2100 break; 2101 } 2102 error = sa6_embedscope(fin6, V_ip6_use_defzone); 2103 if (error) 2104 return (error); 2105 error = sa6_embedscope(lin6, V_ip6_use_defzone); 2106 if (error) 2107 return (error); 2108 break; 2109 #endif 2110 case AF_INET: 2111 fin = (struct sockaddr_in *)&addrs[0]; 2112 lin = (struct sockaddr_in *)&addrs[1]; 2113 if (fin->sin_len != sizeof(struct sockaddr_in) || 2114 lin->sin_len != sizeof(struct sockaddr_in)) 2115 return (EINVAL); 2116 break; 2117 default: 2118 return (EINVAL); 2119 } 2120 INP_INFO_WLOCK(&V_tcbinfo); 2121 switch (addrs[0].ss_family) { 2122 #ifdef INET6 2123 case AF_INET6: 2124 inp = in6_pcblookup_hash(&V_tcbinfo, &f6, fin6->sin6_port, 2125 &l6, lin6->sin6_port, 0, NULL); 2126 break; 2127 #endif 2128 case AF_INET: 2129 inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr, 2130 fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL); 2131 break; 2132 } 2133 if (inp != NULL) { 2134 INP_WLOCK(inp); 2135 if (inp->inp_vflag & INP_TIMEWAIT) { 2136 /* 2137 * XXXRW: There currently exists a state where an 2138 * inpcb is present, but its timewait state has been 2139 * discarded. For now, don't allow dropping of this 2140 * type of inpcb. 2141 */ 2142 tw = intotw(inp); 2143 if (tw != NULL) 2144 tcp_twclose(tw, 0); 2145 else 2146 INP_WUNLOCK(inp); 2147 } else if (!(inp->inp_vflag & INP_DROPPED) && 2148 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2149 tp = intotcpcb(inp); 2150 tp = tcp_drop(tp, ECONNABORTED); 2151 if (tp != NULL) 2152 INP_WUNLOCK(inp); 2153 } else 2154 INP_WUNLOCK(inp); 2155 } else 2156 error = ESRCH; 2157 INP_INFO_WUNLOCK(&V_tcbinfo); 2158 return (error); 2159 } 2160 2161 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2162 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2163 0, sysctl_drop, "", "Drop TCP connection"); 2164 2165 /* 2166 * Generate a standardized TCP log line for use throughout the 2167 * tcp subsystem. Memory allocation is done with M_NOWAIT to 2168 * allow use in the interrupt context. 2169 * 2170 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 2171 * NB: The function may return NULL if memory allocation failed. 2172 * 2173 * Due to header inclusion and ordering limitations the struct ip 2174 * and ip6_hdr pointers have to be passed as void pointers. 2175 */ 2176 char * 2177 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2178 const void *ip6hdr) 2179 { 2180 char *s, *sp; 2181 size_t size; 2182 struct ip *ip; 2183 #ifdef INET6 2184 const struct ip6_hdr *ip6; 2185 2186 ip6 = (const struct ip6_hdr *)ip6hdr; 2187 #endif /* INET6 */ 2188 ip = (struct ip *)ip4hdr; 2189 2190 /* 2191 * The log line looks like this: 2192 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 2193 */ 2194 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 2195 sizeof(PRINT_TH_FLAGS) + 1 + 2196 #ifdef INET6 2197 2 * INET6_ADDRSTRLEN; 2198 #else 2199 2 * INET_ADDRSTRLEN; 2200 #endif /* INET6 */ 2201 2202 /* Is logging enabled? */ 2203 if (tcp_log_debug == 0 && tcp_log_in_vain == 0) 2204 return (NULL); 2205 2206 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 2207 if (s == NULL) 2208 return (NULL); 2209 2210 strcat(s, "TCP: ["); 2211 sp = s + strlen(s); 2212 2213 if (inc && inc->inc_isipv6 == 0) { 2214 inet_ntoa_r(inc->inc_faddr, sp); 2215 sp = s + strlen(s); 2216 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2217 sp = s + strlen(s); 2218 inet_ntoa_r(inc->inc_laddr, sp); 2219 sp = s + strlen(s); 2220 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2221 #ifdef INET6 2222 } else if (inc) { 2223 ip6_sprintf(sp, &inc->inc6_faddr); 2224 sp = s + strlen(s); 2225 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2226 sp = s + strlen(s); 2227 ip6_sprintf(sp, &inc->inc6_laddr); 2228 sp = s + strlen(s); 2229 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2230 } else if (ip6 && th) { 2231 ip6_sprintf(sp, &ip6->ip6_src); 2232 sp = s + strlen(s); 2233 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2234 sp = s + strlen(s); 2235 ip6_sprintf(sp, &ip6->ip6_dst); 2236 sp = s + strlen(s); 2237 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2238 #endif /* INET6 */ 2239 } else if (ip && th) { 2240 inet_ntoa_r(ip->ip_src, sp); 2241 sp = s + strlen(s); 2242 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2243 sp = s + strlen(s); 2244 inet_ntoa_r(ip->ip_dst, sp); 2245 sp = s + strlen(s); 2246 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2247 } else { 2248 free(s, M_TCPLOG); 2249 return (NULL); 2250 } 2251 sp = s + strlen(s); 2252 if (th) 2253 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 2254 if (*(s + size - 1) != '\0') 2255 panic("%s: string too long", __func__); 2256 return (s); 2257 } 2258