1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 30 * $FreeBSD$ 31 */ 32 33 #include "opt_compat.h" 34 #include "opt_inet.h" 35 #include "opt_inet6.h" 36 #include "opt_ipsec.h" 37 #include "opt_mac.h" 38 #include "opt_tcpdebug.h" 39 #include "opt_tcp_sack.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/callout.h> 44 #include <sys/kernel.h> 45 #include <sys/sysctl.h> 46 #include <sys/malloc.h> 47 #include <sys/mbuf.h> 48 #ifdef INET6 49 #include <sys/domain.h> 50 #endif 51 #include <sys/priv.h> 52 #include <sys/proc.h> 53 #include <sys/socket.h> 54 #include <sys/socketvar.h> 55 #include <sys/protosw.h> 56 #include <sys/random.h> 57 58 #include <vm/uma.h> 59 60 #include <net/route.h> 61 #include <net/if.h> 62 63 #include <netinet/in.h> 64 #include <netinet/in_systm.h> 65 #include <netinet/ip.h> 66 #ifdef INET6 67 #include <netinet/ip6.h> 68 #endif 69 #include <netinet/in_pcb.h> 70 #ifdef INET6 71 #include <netinet6/in6_pcb.h> 72 #endif 73 #include <netinet/in_var.h> 74 #include <netinet/ip_var.h> 75 #ifdef INET6 76 #include <netinet6/ip6_var.h> 77 #include <netinet6/scope6_var.h> 78 #include <netinet6/nd6.h> 79 #endif 80 #include <netinet/ip_icmp.h> 81 #include <netinet/tcp.h> 82 #include <netinet/tcp_fsm.h> 83 #include <netinet/tcp_seq.h> 84 #include <netinet/tcp_timer.h> 85 #include <netinet/tcp_var.h> 86 #ifdef INET6 87 #include <netinet6/tcp6_var.h> 88 #endif 89 #include <netinet/tcpip.h> 90 #ifdef TCPDEBUG 91 #include <netinet/tcp_debug.h> 92 #endif 93 #include <netinet6/ip6protosw.h> 94 95 #ifdef IPSEC 96 #include <netinet6/ipsec.h> 97 #ifdef INET6 98 #include <netinet6/ipsec6.h> 99 #endif 100 #include <netkey/key.h> 101 #endif /*IPSEC*/ 102 103 #ifdef FAST_IPSEC 104 #include <netipsec/ipsec.h> 105 #include <netipsec/xform.h> 106 #ifdef INET6 107 #include <netipsec/ipsec6.h> 108 #endif 109 #include <netipsec/key.h> 110 #define IPSEC 111 #endif /*FAST_IPSEC*/ 112 113 #include <machine/in_cksum.h> 114 #include <sys/md5.h> 115 116 #include <security/mac/mac_framework.h> 117 118 int tcp_mssdflt = TCP_MSS; 119 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 120 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); 121 122 #ifdef INET6 123 int tcp_v6mssdflt = TCP6_MSS; 124 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 125 CTLFLAG_RW, &tcp_v6mssdflt , 0, 126 "Default TCP Maximum Segment Size for IPv6"); 127 #endif 128 129 /* 130 * Minimum MSS we accept and use. This prevents DoS attacks where 131 * we are forced to a ridiculous low MSS like 20 and send hundreds 132 * of packets instead of one. The effect scales with the available 133 * bandwidth and quickly saturates the CPU and network interface 134 * with packet generation and sending. Set to zero to disable MINMSS 135 * checking. This setting prevents us from sending too small packets. 136 */ 137 int tcp_minmss = TCP_MINMSS; 138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 139 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 140 /* 141 * Number of TCP segments per second we accept from remote host 142 * before we start to calculate average segment size. If average 143 * segment size drops below the minimum TCP MSS we assume a DoS 144 * attack and reset+drop the connection. Care has to be taken not to 145 * set this value too small to not kill interactive type connections 146 * (telnet, SSH) which send many small packets. 147 */ 148 int tcp_minmssoverload = TCP_MINMSSOVERLOAD; 149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW, 150 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to" 151 "be under the MINMSS Size"); 152 153 #if 0 154 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 155 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 156 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 157 #endif 158 159 int tcp_do_rfc1323 = 1; 160 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 161 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 162 163 static int tcp_tcbhashsize = 0; 164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 165 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 166 167 static int do_tcpdrain = 1; 168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 169 "Enable tcp_drain routine for extra help when low on mbufs"); 170 171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 172 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 173 174 static int icmp_may_rst = 1; 175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 176 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 177 178 static int tcp_isn_reseed_interval = 0; 179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 180 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 181 182 static uma_zone_t tcptw_zone; 183 static int maxtcptw; 184 185 static int 186 tcptw_auto_size(void) 187 { 188 int halfrange; 189 190 /* 191 * Max out at half the ephemeral port range so that TIME_WAIT 192 * sockets don't tie up too many ephemeral ports. 193 */ 194 if (ipport_lastauto > ipport_firstauto) 195 halfrange = (ipport_lastauto - ipport_firstauto) / 2; 196 else 197 halfrange = (ipport_firstauto - ipport_lastauto) / 2; 198 /* Protect against goofy port ranges smaller than 32. */ 199 return (imin(imax(halfrange, 32), maxsockets / 5)); 200 } 201 202 static int 203 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 204 { 205 int error, new; 206 207 if (maxtcptw == 0) 208 new = tcptw_auto_size(); 209 else 210 new = maxtcptw; 211 error = sysctl_handle_int(oidp, &new, sizeof(int), req); 212 if (error == 0 && req->newptr) 213 if (new >= 32) { 214 maxtcptw = new; 215 uma_zone_set_max(tcptw_zone, maxtcptw); 216 } 217 return (error); 218 } 219 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 220 &maxtcptw, 0, sysctl_maxtcptw, "IU", 221 "Maximum number of compressed TCP TIME_WAIT entries"); 222 223 static int nolocaltimewait = 0; 224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW, 225 &nolocaltimewait, 0, "Do not create compressed TCP TIME_WAIT entries" 226 "for local connections"); 227 228 /* 229 * TCP bandwidth limiting sysctls. Note that the default lower bound of 230 * 1024 exists only for debugging. A good production default would be 231 * something like 6100. 232 */ 233 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 234 "TCP inflight data limiting"); 235 236 static int tcp_inflight_enable = 1; 237 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 238 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 239 240 static int tcp_inflight_debug = 0; 241 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 242 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 243 244 static int tcp_inflight_rttthresh; 245 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW, 246 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I", 247 "RTT threshold below which inflight will deactivate itself"); 248 249 static int tcp_inflight_min = 6144; 250 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 251 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 252 253 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 254 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 255 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 256 257 static int tcp_inflight_stab = 20; 258 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 259 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 260 261 uma_zone_t sack_hole_zone; 262 263 static struct inpcb *tcp_notify(struct inpcb *, int); 264 static void tcp_isn_tick(void *); 265 266 /* 267 * Target size of TCP PCB hash tables. Must be a power of two. 268 * 269 * Note that this can be overridden by the kernel environment 270 * variable net.inet.tcp.tcbhashsize 271 */ 272 #ifndef TCBHASHSIZE 273 #define TCBHASHSIZE 512 274 #endif 275 276 /* 277 * XXX 278 * Callouts should be moved into struct tcp directly. They are currently 279 * separate because the tcpcb structure is exported to userland for sysctl 280 * parsing purposes, which do not know about callouts. 281 */ 282 struct tcpcb_mem { 283 struct tcpcb tcb; 284 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 285 struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 286 }; 287 288 static uma_zone_t tcpcb_zone; 289 struct callout isn_callout; 290 static struct mtx isn_mtx; 291 292 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 293 #define ISN_LOCK() mtx_lock(&isn_mtx) 294 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 295 296 /* 297 * TCP initialization. 298 */ 299 static void 300 tcp_zone_change(void *tag) 301 { 302 303 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 304 uma_zone_set_max(tcpcb_zone, maxsockets); 305 if (maxtcptw == 0) 306 uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 307 } 308 309 static int 310 tcp_inpcb_init(void *mem, int size, int flags) 311 { 312 struct inpcb *inp = (struct inpcb *) mem; 313 INP_LOCK_INIT(inp, "inp", "tcpinp"); 314 return (0); 315 } 316 317 void 318 tcp_init(void) 319 { 320 int hashsize = TCBHASHSIZE; 321 322 tcp_delacktime = TCPTV_DELACK; 323 tcp_keepinit = TCPTV_KEEP_INIT; 324 tcp_keepidle = TCPTV_KEEP_IDLE; 325 tcp_keepintvl = TCPTV_KEEPINTVL; 326 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 327 tcp_msl = TCPTV_MSL; 328 tcp_rexmit_min = TCPTV_MIN; 329 tcp_rexmit_slop = TCPTV_CPU_VAR; 330 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH; 331 332 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 333 LIST_INIT(&tcb); 334 tcbinfo.listhead = &tcb; 335 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 336 if (!powerof2(hashsize)) { 337 printf("WARNING: TCB hash size not a power of 2\n"); 338 hashsize = 512; /* safe default */ 339 } 340 tcp_tcbhashsize = hashsize; 341 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 342 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 343 &tcbinfo.porthashmask); 344 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 345 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 346 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 347 #ifdef INET6 348 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 349 #else /* INET6 */ 350 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 351 #endif /* INET6 */ 352 if (max_protohdr < TCP_MINPROTOHDR) 353 max_protohdr = TCP_MINPROTOHDR; 354 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 355 panic("tcp_init"); 356 #undef TCP_MINPROTOHDR 357 /* 358 * These have to be type stable for the benefit of the timers. 359 */ 360 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 361 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 362 uma_zone_set_max(tcpcb_zone, maxsockets); 363 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 364 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 365 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 366 if (maxtcptw == 0) 367 uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 368 else 369 uma_zone_set_max(tcptw_zone, maxtcptw); 370 tcp_timer_init(); 371 syncache_init(); 372 tcp_hc_init(); 373 tcp_reass_init(); 374 ISN_LOCK_INIT(); 375 callout_init(&isn_callout, CALLOUT_MPSAFE); 376 tcp_isn_tick(NULL); 377 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 378 SHUTDOWN_PRI_DEFAULT); 379 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 380 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 381 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 382 EVENTHANDLER_PRI_ANY); 383 } 384 385 void 386 tcp_fini(void *xtp) 387 { 388 389 callout_stop(&isn_callout); 390 } 391 392 /* 393 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 394 * tcp_template used to store this data in mbufs, but we now recopy it out 395 * of the tcpcb each time to conserve mbufs. 396 */ 397 void 398 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 399 { 400 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 401 402 INP_LOCK_ASSERT(inp); 403 404 #ifdef INET6 405 if ((inp->inp_vflag & INP_IPV6) != 0) { 406 struct ip6_hdr *ip6; 407 408 ip6 = (struct ip6_hdr *)ip_ptr; 409 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 410 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 411 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 412 (IPV6_VERSION & IPV6_VERSION_MASK); 413 ip6->ip6_nxt = IPPROTO_TCP; 414 ip6->ip6_plen = sizeof(struct tcphdr); 415 ip6->ip6_src = inp->in6p_laddr; 416 ip6->ip6_dst = inp->in6p_faddr; 417 } else 418 #endif 419 { 420 struct ip *ip; 421 422 ip = (struct ip *)ip_ptr; 423 ip->ip_v = IPVERSION; 424 ip->ip_hl = 5; 425 ip->ip_tos = inp->inp_ip_tos; 426 ip->ip_len = 0; 427 ip->ip_id = 0; 428 ip->ip_off = 0; 429 ip->ip_ttl = inp->inp_ip_ttl; 430 ip->ip_sum = 0; 431 ip->ip_p = IPPROTO_TCP; 432 ip->ip_src = inp->inp_laddr; 433 ip->ip_dst = inp->inp_faddr; 434 } 435 th->th_sport = inp->inp_lport; 436 th->th_dport = inp->inp_fport; 437 th->th_seq = 0; 438 th->th_ack = 0; 439 th->th_x2 = 0; 440 th->th_off = 5; 441 th->th_flags = 0; 442 th->th_win = 0; 443 th->th_urp = 0; 444 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 445 } 446 447 /* 448 * Create template to be used to send tcp packets on a connection. 449 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 450 * use for this function is in keepalives, which use tcp_respond. 451 */ 452 struct tcptemp * 453 tcpip_maketemplate(struct inpcb *inp) 454 { 455 struct mbuf *m; 456 struct tcptemp *n; 457 458 m = m_get(M_DONTWAIT, MT_DATA); 459 if (m == NULL) 460 return (0); 461 m->m_len = sizeof(struct tcptemp); 462 n = mtod(m, struct tcptemp *); 463 464 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 465 return (n); 466 } 467 468 /* 469 * Send a single message to the TCP at address specified by 470 * the given TCP/IP header. If m == NULL, then we make a copy 471 * of the tcpiphdr at ti and send directly to the addressed host. 472 * This is used to force keep alive messages out using the TCP 473 * template for a connection. If flags are given then we send 474 * a message back to the TCP which originated the * segment ti, 475 * and discard the mbuf containing it and any other attached mbufs. 476 * 477 * In any case the ack and sequence number of the transmitted 478 * segment are as specified by the parameters. 479 * 480 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 481 */ 482 void 483 tcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th, 484 register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) 485 { 486 register int tlen; 487 int win = 0; 488 struct ip *ip; 489 struct tcphdr *nth; 490 #ifdef INET6 491 struct ip6_hdr *ip6; 492 int isipv6; 493 #endif /* INET6 */ 494 int ipflags = 0; 495 struct inpcb *inp; 496 497 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 498 499 #ifdef INET6 500 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 501 ip6 = ipgen; 502 #endif /* INET6 */ 503 ip = ipgen; 504 505 if (tp != NULL) { 506 inp = tp->t_inpcb; 507 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 508 INP_INFO_WLOCK_ASSERT(&tcbinfo); 509 INP_LOCK_ASSERT(inp); 510 } else 511 inp = NULL; 512 513 if (tp != NULL) { 514 if (!(flags & TH_RST)) { 515 win = sbspace(&inp->inp_socket->so_rcv); 516 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 517 win = (long)TCP_MAXWIN << tp->rcv_scale; 518 } 519 } 520 if (m == NULL) { 521 m = m_gethdr(M_DONTWAIT, MT_DATA); 522 if (m == NULL) 523 return; 524 tlen = 0; 525 m->m_data += max_linkhdr; 526 #ifdef INET6 527 if (isipv6) { 528 bcopy((caddr_t)ip6, mtod(m, caddr_t), 529 sizeof(struct ip6_hdr)); 530 ip6 = mtod(m, struct ip6_hdr *); 531 nth = (struct tcphdr *)(ip6 + 1); 532 } else 533 #endif /* INET6 */ 534 { 535 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 536 ip = mtod(m, struct ip *); 537 nth = (struct tcphdr *)(ip + 1); 538 } 539 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 540 flags = TH_ACK; 541 } else { 542 m_freem(m->m_next); 543 m->m_next = NULL; 544 m->m_data = (caddr_t)ipgen; 545 /* m_len is set later */ 546 tlen = 0; 547 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 548 #ifdef INET6 549 if (isipv6) { 550 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 551 nth = (struct tcphdr *)(ip6 + 1); 552 } else 553 #endif /* INET6 */ 554 { 555 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 556 nth = (struct tcphdr *)(ip + 1); 557 } 558 if (th != nth) { 559 /* 560 * this is usually a case when an extension header 561 * exists between the IPv6 header and the 562 * TCP header. 563 */ 564 nth->th_sport = th->th_sport; 565 nth->th_dport = th->th_dport; 566 } 567 xchg(nth->th_dport, nth->th_sport, n_short); 568 #undef xchg 569 } 570 #ifdef INET6 571 if (isipv6) { 572 ip6->ip6_flow = 0; 573 ip6->ip6_vfc = IPV6_VERSION; 574 ip6->ip6_nxt = IPPROTO_TCP; 575 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 576 tlen)); 577 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 578 } else 579 #endif 580 { 581 tlen += sizeof (struct tcpiphdr); 582 ip->ip_len = tlen; 583 ip->ip_ttl = ip_defttl; 584 if (path_mtu_discovery) 585 ip->ip_off |= IP_DF; 586 } 587 m->m_len = tlen; 588 m->m_pkthdr.len = tlen; 589 m->m_pkthdr.rcvif = NULL; 590 #ifdef MAC 591 if (inp != NULL) { 592 /* 593 * Packet is associated with a socket, so allow the 594 * label of the response to reflect the socket label. 595 */ 596 INP_LOCK_ASSERT(inp); 597 mac_create_mbuf_from_inpcb(inp, m); 598 } else { 599 /* 600 * Packet is not associated with a socket, so possibly 601 * update the label in place. 602 */ 603 mac_reflect_mbuf_tcp(m); 604 } 605 #endif 606 nth->th_seq = htonl(seq); 607 nth->th_ack = htonl(ack); 608 nth->th_x2 = 0; 609 nth->th_off = sizeof (struct tcphdr) >> 2; 610 nth->th_flags = flags; 611 if (tp != NULL) 612 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 613 else 614 nth->th_win = htons((u_short)win); 615 nth->th_urp = 0; 616 #ifdef INET6 617 if (isipv6) { 618 nth->th_sum = 0; 619 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 620 sizeof(struct ip6_hdr), 621 tlen - sizeof(struct ip6_hdr)); 622 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 623 NULL, NULL); 624 } else 625 #endif /* INET6 */ 626 { 627 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 628 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 629 m->m_pkthdr.csum_flags = CSUM_TCP; 630 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 631 } 632 #ifdef TCPDEBUG 633 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 634 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 635 #endif 636 #ifdef INET6 637 if (isipv6) 638 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 639 else 640 #endif /* INET6 */ 641 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 642 } 643 644 /* 645 * Create a new TCP control block, making an 646 * empty reassembly queue and hooking it to the argument 647 * protocol control block. The `inp' parameter must have 648 * come from the zone allocator set up in tcp_init(). 649 */ 650 struct tcpcb * 651 tcp_newtcpcb(struct inpcb *inp) 652 { 653 struct tcpcb_mem *tm; 654 struct tcpcb *tp; 655 #ifdef INET6 656 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 657 #endif /* INET6 */ 658 659 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 660 if (tm == NULL) 661 return (NULL); 662 tp = &tm->tcb; 663 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 664 tp->t_maxseg = tp->t_maxopd = 665 #ifdef INET6 666 isipv6 ? tcp_v6mssdflt : 667 #endif /* INET6 */ 668 tcp_mssdflt; 669 670 /* Set up our timeouts. */ 671 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE); 672 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE); 673 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE); 674 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE); 675 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE); 676 677 if (tcp_do_rfc1323) 678 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 679 tp->sack_enable = tcp_do_sack; 680 TAILQ_INIT(&tp->snd_holes); 681 tp->t_inpcb = inp; /* XXX */ 682 /* 683 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 684 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 685 * reasonable initial retransmit time. 686 */ 687 tp->t_srtt = TCPTV_SRTTBASE; 688 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 689 tp->t_rttmin = tcp_rexmit_min; 690 tp->t_rxtcur = TCPTV_RTOBASE; 691 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 692 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 693 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 694 tp->t_rcvtime = ticks; 695 tp->t_bw_rtttime = ticks; 696 /* 697 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 698 * because the socket may be bound to an IPv6 wildcard address, 699 * which may match an IPv4-mapped IPv6 address. 700 */ 701 inp->inp_ip_ttl = ip_defttl; 702 inp->inp_ppcb = tp; 703 return (tp); /* XXX */ 704 } 705 706 /* 707 * Drop a TCP connection, reporting 708 * the specified error. If connection is synchronized, 709 * then send a RST to peer. 710 */ 711 struct tcpcb * 712 tcp_drop(struct tcpcb *tp, int errno) 713 { 714 struct socket *so = tp->t_inpcb->inp_socket; 715 716 INP_INFO_WLOCK_ASSERT(&tcbinfo); 717 INP_LOCK_ASSERT(tp->t_inpcb); 718 719 if (TCPS_HAVERCVDSYN(tp->t_state)) { 720 tp->t_state = TCPS_CLOSED; 721 (void) tcp_output(tp); 722 tcpstat.tcps_drops++; 723 } else 724 tcpstat.tcps_conndrops++; 725 if (errno == ETIMEDOUT && tp->t_softerror) 726 errno = tp->t_softerror; 727 so->so_error = errno; 728 return (tcp_close(tp)); 729 } 730 731 void 732 tcp_discardcb(struct tcpcb *tp) 733 { 734 struct tseg_qent *q; 735 struct inpcb *inp = tp->t_inpcb; 736 struct socket *so = inp->inp_socket; 737 #ifdef INET6 738 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 739 #endif /* INET6 */ 740 741 INP_LOCK_ASSERT(inp); 742 743 /* 744 * Make sure that all of our timers are stopped before we 745 * delete the PCB. 746 */ 747 callout_stop(tp->tt_rexmt); 748 callout_stop(tp->tt_persist); 749 callout_stop(tp->tt_keep); 750 callout_stop(tp->tt_2msl); 751 callout_stop(tp->tt_delack); 752 753 /* 754 * If we got enough samples through the srtt filter, 755 * save the rtt and rttvar in the routing entry. 756 * 'Enough' is arbitrarily defined as 4 rtt samples. 757 * 4 samples is enough for the srtt filter to converge 758 * to within enough % of the correct value; fewer samples 759 * and we could save a bogus rtt. The danger is not high 760 * as tcp quickly recovers from everything. 761 * XXX: Works very well but needs some more statistics! 762 */ 763 if (tp->t_rttupdated >= 4) { 764 struct hc_metrics_lite metrics; 765 u_long ssthresh; 766 767 bzero(&metrics, sizeof(metrics)); 768 /* 769 * Update the ssthresh always when the conditions below 770 * are satisfied. This gives us better new start value 771 * for the congestion avoidance for new connections. 772 * ssthresh is only set if packet loss occured on a session. 773 * 774 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 775 * being torn down. Ideally this code would not use 'so'. 776 */ 777 ssthresh = tp->snd_ssthresh; 778 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 779 /* 780 * convert the limit from user data bytes to 781 * packets then to packet data bytes. 782 */ 783 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 784 if (ssthresh < 2) 785 ssthresh = 2; 786 ssthresh *= (u_long)(tp->t_maxseg + 787 #ifdef INET6 788 (isipv6 ? sizeof (struct ip6_hdr) + 789 sizeof (struct tcphdr) : 790 #endif 791 sizeof (struct tcpiphdr) 792 #ifdef INET6 793 ) 794 #endif 795 ); 796 } else 797 ssthresh = 0; 798 metrics.rmx_ssthresh = ssthresh; 799 800 metrics.rmx_rtt = tp->t_srtt; 801 metrics.rmx_rttvar = tp->t_rttvar; 802 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 803 metrics.rmx_bandwidth = tp->snd_bandwidth; 804 metrics.rmx_cwnd = tp->snd_cwnd; 805 metrics.rmx_sendpipe = 0; 806 metrics.rmx_recvpipe = 0; 807 808 tcp_hc_update(&inp->inp_inc, &metrics); 809 } 810 811 /* free the reassembly queue, if any */ 812 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 813 LIST_REMOVE(q, tqe_q); 814 m_freem(q->tqe_m); 815 uma_zfree(tcp_reass_zone, q); 816 tp->t_segqlen--; 817 tcp_reass_qsize--; 818 } 819 tcp_free_sackholes(tp); 820 inp->inp_ppcb = NULL; 821 tp->t_inpcb = NULL; 822 uma_zfree(tcpcb_zone, tp); 823 } 824 825 /* 826 * Attempt to close a TCP control block, marking it as dropped, and freeing 827 * the socket if we hold the only reference. 828 */ 829 struct tcpcb * 830 tcp_close(struct tcpcb *tp) 831 { 832 struct inpcb *inp = tp->t_inpcb; 833 struct socket *so; 834 835 INP_INFO_WLOCK_ASSERT(&tcbinfo); 836 INP_LOCK_ASSERT(inp); 837 838 in_pcbdrop(inp); 839 tcpstat.tcps_closed++; 840 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 841 so = inp->inp_socket; 842 soisdisconnected(so); 843 if (inp->inp_vflag & INP_SOCKREF) { 844 KASSERT(so->so_state & SS_PROTOREF, 845 ("tcp_close: !SS_PROTOREF")); 846 inp->inp_vflag &= ~INP_SOCKREF; 847 INP_UNLOCK(inp); 848 ACCEPT_LOCK(); 849 SOCK_LOCK(so); 850 so->so_state &= ~SS_PROTOREF; 851 sofree(so); 852 return (NULL); 853 } 854 return (tp); 855 } 856 857 void 858 tcp_drain(void) 859 { 860 861 if (do_tcpdrain) { 862 struct inpcb *inpb; 863 struct tcpcb *tcpb; 864 struct tseg_qent *te; 865 866 /* 867 * Walk the tcpbs, if existing, and flush the reassembly queue, 868 * if there is one... 869 * XXX: The "Net/3" implementation doesn't imply that the TCP 870 * reassembly queue should be flushed, but in a situation 871 * where we're really low on mbufs, this is potentially 872 * usefull. 873 */ 874 INP_INFO_RLOCK(&tcbinfo); 875 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 876 if (inpb->inp_vflag & INP_TIMEWAIT) 877 continue; 878 INP_LOCK(inpb); 879 if ((tcpb = intotcpcb(inpb)) != NULL) { 880 while ((te = LIST_FIRST(&tcpb->t_segq)) 881 != NULL) { 882 LIST_REMOVE(te, tqe_q); 883 m_freem(te->tqe_m); 884 uma_zfree(tcp_reass_zone, te); 885 tcpb->t_segqlen--; 886 tcp_reass_qsize--; 887 } 888 tcp_clean_sackreport(tcpb); 889 } 890 INP_UNLOCK(inpb); 891 } 892 INP_INFO_RUNLOCK(&tcbinfo); 893 } 894 } 895 896 /* 897 * Notify a tcp user of an asynchronous error; 898 * store error as soft error, but wake up user 899 * (for now, won't do anything until can select for soft error). 900 * 901 * Do not wake up user since there currently is no mechanism for 902 * reporting soft errors (yet - a kqueue filter may be added). 903 */ 904 static struct inpcb * 905 tcp_notify(struct inpcb *inp, int error) 906 { 907 struct tcpcb *tp; 908 909 INP_INFO_WLOCK_ASSERT(&tcbinfo); 910 INP_LOCK_ASSERT(inp); 911 912 if ((inp->inp_vflag & INP_TIMEWAIT) || 913 (inp->inp_vflag & INP_DROPPED)) 914 return (inp); 915 916 tp = intotcpcb(inp); 917 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 918 919 /* 920 * Ignore some errors if we are hooked up. 921 * If connection hasn't completed, has retransmitted several times, 922 * and receives a second error, give up now. This is better 923 * than waiting a long time to establish a connection that 924 * can never complete. 925 */ 926 if (tp->t_state == TCPS_ESTABLISHED && 927 (error == EHOSTUNREACH || error == ENETUNREACH || 928 error == EHOSTDOWN)) { 929 return (inp); 930 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 931 tp->t_softerror) { 932 tp = tcp_drop(tp, error); 933 if (tp != NULL) 934 return (inp); 935 else 936 return (NULL); 937 } else { 938 tp->t_softerror = error; 939 return (inp); 940 } 941 #if 0 942 wakeup( &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; 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 == NULL) { 961 n = tcbinfo.ipi_count; 962 req->oldidx = 2 * (sizeof xig) 963 + (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(&tcbinfo); 974 gencnt = tcbinfo.ipi_gencnt; 975 n = tcbinfo.ipi_count; 976 INP_INFO_RUNLOCK(&tcbinfo); 977 978 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 979 + n * sizeof(struct xtcpcb)); 980 if (error != 0) 981 return (error); 982 983 xig.xig_len = sizeof xig; 984 xig.xig_count = n; 985 xig.xig_gen = gencnt; 986 xig.xig_sogen = so_gencnt; 987 error = SYSCTL_OUT(req, &xig, sizeof xig); 988 if (error) 989 return (error); 990 991 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 992 if (inp_list == NULL) 993 return (ENOMEM); 994 995 INP_INFO_RLOCK(&tcbinfo); 996 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 997 inp = LIST_NEXT(inp, inp_list)) { 998 INP_LOCK(inp); 999 if (inp->inp_gencnt <= gencnt) { 1000 /* 1001 * XXX: This use of cr_cansee(), introduced with 1002 * TCP state changes, is not quite right, but for 1003 * now, better than nothing. 1004 */ 1005 if (inp->inp_vflag & INP_TIMEWAIT) { 1006 if (intotw(inp) != NULL) 1007 error = cr_cansee(req->td->td_ucred, 1008 intotw(inp)->tw_cred); 1009 else 1010 error = EINVAL; /* Skip this inp. */ 1011 } else 1012 error = cr_canseesocket(req->td->td_ucred, 1013 inp->inp_socket); 1014 if (error == 0) 1015 inp_list[i++] = inp; 1016 } 1017 INP_UNLOCK(inp); 1018 } 1019 INP_INFO_RUNLOCK(&tcbinfo); 1020 n = i; 1021 1022 error = 0; 1023 for (i = 0; i < n; i++) { 1024 inp = inp_list[i]; 1025 INP_LOCK(inp); 1026 if (inp->inp_gencnt <= gencnt) { 1027 struct xtcpcb xt; 1028 void *inp_ppcb; 1029 1030 bzero(&xt, sizeof(xt)); 1031 xt.xt_len = sizeof xt; 1032 /* XXX should avoid extra copy */ 1033 bcopy(inp, &xt.xt_inp, sizeof *inp); 1034 inp_ppcb = inp->inp_ppcb; 1035 if (inp_ppcb == NULL) 1036 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1037 else if (inp->inp_vflag & INP_TIMEWAIT) { 1038 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1039 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1040 } else 1041 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1042 if (inp->inp_socket != NULL) 1043 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1044 else { 1045 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1046 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1047 } 1048 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1049 INP_UNLOCK(inp); 1050 error = SYSCTL_OUT(req, &xt, sizeof xt); 1051 } else 1052 INP_UNLOCK(inp); 1053 1054 } 1055 if (!error) { 1056 /* 1057 * Give the user an updated idea of our state. 1058 * If the generation differs from what we told 1059 * her before, she knows that something happened 1060 * while we were processing this request, and it 1061 * might be necessary to retry. 1062 */ 1063 INP_INFO_RLOCK(&tcbinfo); 1064 xig.xig_gen = tcbinfo.ipi_gencnt; 1065 xig.xig_sogen = so_gencnt; 1066 xig.xig_count = tcbinfo.ipi_count; 1067 INP_INFO_RUNLOCK(&tcbinfo); 1068 error = SYSCTL_OUT(req, &xig, sizeof xig); 1069 } 1070 free(inp_list, M_TEMP); 1071 return (error); 1072 } 1073 1074 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1075 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1076 1077 static int 1078 tcp_getcred(SYSCTL_HANDLER_ARGS) 1079 { 1080 struct xucred xuc; 1081 struct sockaddr_in addrs[2]; 1082 struct inpcb *inp; 1083 int error; 1084 1085 error = priv_check_cred(req->td->td_ucred, PRIV_NETINET_GETCRED, 1086 SUSER_ALLOWJAIL); 1087 if (error) 1088 return (error); 1089 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1090 if (error) 1091 return (error); 1092 INP_INFO_RLOCK(&tcbinfo); 1093 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1094 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1095 if (inp == NULL) { 1096 error = ENOENT; 1097 goto outunlocked; 1098 } 1099 INP_LOCK(inp); 1100 if (inp->inp_socket == NULL) { 1101 error = ENOENT; 1102 goto out; 1103 } 1104 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1105 if (error) 1106 goto out; 1107 cru2x(inp->inp_socket->so_cred, &xuc); 1108 out: 1109 INP_UNLOCK(inp); 1110 outunlocked: 1111 INP_INFO_RUNLOCK(&tcbinfo); 1112 if (error == 0) 1113 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1114 return (error); 1115 } 1116 1117 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1118 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1119 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1120 1121 #ifdef INET6 1122 static int 1123 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1124 { 1125 struct xucred xuc; 1126 struct sockaddr_in6 addrs[2]; 1127 struct inpcb *inp; 1128 int error, mapped = 0; 1129 1130 error = priv_check_cred(req->td->td_ucred, PRIV_NETINET_GETCRED, 1131 SUSER_ALLOWJAIL); 1132 if (error) 1133 return (error); 1134 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1135 if (error) 1136 return (error); 1137 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1138 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1139 return (error); 1140 } 1141 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1142 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1143 mapped = 1; 1144 else 1145 return (EINVAL); 1146 } 1147 1148 INP_INFO_RLOCK(&tcbinfo); 1149 if (mapped == 1) 1150 inp = in_pcblookup_hash(&tcbinfo, 1151 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1152 addrs[1].sin6_port, 1153 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1154 addrs[0].sin6_port, 1155 0, NULL); 1156 else 1157 inp = in6_pcblookup_hash(&tcbinfo, 1158 &addrs[1].sin6_addr, addrs[1].sin6_port, 1159 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1160 if (inp == NULL) { 1161 error = ENOENT; 1162 goto outunlocked; 1163 } 1164 INP_LOCK(inp); 1165 if (inp->inp_socket == NULL) { 1166 error = ENOENT; 1167 goto out; 1168 } 1169 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1170 if (error) 1171 goto out; 1172 cru2x(inp->inp_socket->so_cred, &xuc); 1173 out: 1174 INP_UNLOCK(inp); 1175 outunlocked: 1176 INP_INFO_RUNLOCK(&tcbinfo); 1177 if (error == 0) 1178 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1179 return (error); 1180 } 1181 1182 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1183 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1184 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1185 #endif 1186 1187 1188 void 1189 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1190 { 1191 struct ip *ip = vip; 1192 struct tcphdr *th; 1193 struct in_addr faddr; 1194 struct inpcb *inp; 1195 struct tcpcb *tp; 1196 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1197 struct icmp *icp; 1198 struct in_conninfo inc; 1199 tcp_seq icmp_tcp_seq; 1200 int mtu; 1201 1202 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1203 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1204 return; 1205 1206 if (cmd == PRC_MSGSIZE) 1207 notify = tcp_mtudisc; 1208 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1209 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1210 notify = tcp_drop_syn_sent; 1211 /* 1212 * Redirects don't need to be handled up here. 1213 */ 1214 else if (PRC_IS_REDIRECT(cmd)) 1215 return; 1216 /* 1217 * Source quench is depreciated. 1218 */ 1219 else if (cmd == PRC_QUENCH) 1220 return; 1221 /* 1222 * Hostdead is ugly because it goes linearly through all PCBs. 1223 * XXX: We never get this from ICMP, otherwise it makes an 1224 * excellent DoS attack on machines with many connections. 1225 */ 1226 else if (cmd == PRC_HOSTDEAD) 1227 ip = NULL; 1228 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1229 return; 1230 if (ip != NULL) { 1231 icp = (struct icmp *)((caddr_t)ip 1232 - offsetof(struct icmp, icmp_ip)); 1233 th = (struct tcphdr *)((caddr_t)ip 1234 + (ip->ip_hl << 2)); 1235 INP_INFO_WLOCK(&tcbinfo); 1236 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1237 ip->ip_src, th->th_sport, 0, NULL); 1238 if (inp != NULL) { 1239 INP_LOCK(inp); 1240 if (!(inp->inp_vflag & INP_TIMEWAIT) && 1241 !(inp->inp_vflag & INP_DROPPED) && 1242 !(inp->inp_socket == NULL)) { 1243 icmp_tcp_seq = htonl(th->th_seq); 1244 tp = intotcpcb(inp); 1245 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1246 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1247 if (cmd == PRC_MSGSIZE) { 1248 /* 1249 * MTU discovery: 1250 * If we got a needfrag set the MTU 1251 * in the route to the suggested new 1252 * value (if given) and then notify. 1253 */ 1254 bzero(&inc, sizeof(inc)); 1255 inc.inc_flags = 0; /* IPv4 */ 1256 inc.inc_faddr = faddr; 1257 1258 mtu = ntohs(icp->icmp_nextmtu); 1259 /* 1260 * If no alternative MTU was 1261 * proposed, try the next smaller 1262 * one. ip->ip_len has already 1263 * been swapped in icmp_input(). 1264 */ 1265 if (!mtu) 1266 mtu = ip_next_mtu(ip->ip_len, 1267 1); 1268 if (mtu < max(296, (tcp_minmss) 1269 + sizeof(struct tcpiphdr))) 1270 mtu = 0; 1271 if (!mtu) 1272 mtu = tcp_mssdflt 1273 + sizeof(struct tcpiphdr); 1274 /* 1275 * Only cache the the MTU if it 1276 * is smaller than the interface 1277 * or route MTU. tcp_mtudisc() 1278 * will do right thing by itself. 1279 */ 1280 if (mtu <= tcp_maxmtu(&inc, NULL)) 1281 tcp_hc_updatemtu(&inc, mtu); 1282 } 1283 1284 inp = (*notify)(inp, inetctlerrmap[cmd]); 1285 } 1286 } 1287 if (inp != NULL) 1288 INP_UNLOCK(inp); 1289 } else { 1290 inc.inc_fport = th->th_dport; 1291 inc.inc_lport = th->th_sport; 1292 inc.inc_faddr = faddr; 1293 inc.inc_laddr = ip->ip_src; 1294 #ifdef INET6 1295 inc.inc_isipv6 = 0; 1296 #endif 1297 syncache_unreach(&inc, th); 1298 } 1299 INP_INFO_WUNLOCK(&tcbinfo); 1300 } else 1301 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1302 } 1303 1304 #ifdef INET6 1305 void 1306 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1307 { 1308 struct tcphdr th; 1309 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1310 struct ip6_hdr *ip6; 1311 struct mbuf *m; 1312 struct ip6ctlparam *ip6cp = NULL; 1313 const struct sockaddr_in6 *sa6_src = NULL; 1314 int off; 1315 struct tcp_portonly { 1316 u_int16_t th_sport; 1317 u_int16_t th_dport; 1318 } *thp; 1319 1320 if (sa->sa_family != AF_INET6 || 1321 sa->sa_len != sizeof(struct sockaddr_in6)) 1322 return; 1323 1324 if (cmd == PRC_MSGSIZE) 1325 notify = tcp_mtudisc; 1326 else if (!PRC_IS_REDIRECT(cmd) && 1327 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1328 return; 1329 /* Source quench is depreciated. */ 1330 else if (cmd == PRC_QUENCH) 1331 return; 1332 1333 /* if the parameter is from icmp6, decode it. */ 1334 if (d != NULL) { 1335 ip6cp = (struct ip6ctlparam *)d; 1336 m = ip6cp->ip6c_m; 1337 ip6 = ip6cp->ip6c_ip6; 1338 off = ip6cp->ip6c_off; 1339 sa6_src = ip6cp->ip6c_src; 1340 } else { 1341 m = NULL; 1342 ip6 = NULL; 1343 off = 0; /* fool gcc */ 1344 sa6_src = &sa6_any; 1345 } 1346 1347 if (ip6 != NULL) { 1348 struct in_conninfo inc; 1349 /* 1350 * XXX: We assume that when IPV6 is non NULL, 1351 * M and OFF are valid. 1352 */ 1353 1354 /* check if we can safely examine src and dst ports */ 1355 if (m->m_pkthdr.len < off + sizeof(*thp)) 1356 return; 1357 1358 bzero(&th, sizeof(th)); 1359 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1360 1361 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1362 (struct sockaddr *)ip6cp->ip6c_src, 1363 th.th_sport, cmd, NULL, notify); 1364 1365 inc.inc_fport = th.th_dport; 1366 inc.inc_lport = th.th_sport; 1367 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1368 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1369 inc.inc_isipv6 = 1; 1370 INP_INFO_WLOCK(&tcbinfo); 1371 syncache_unreach(&inc, &th); 1372 INP_INFO_WUNLOCK(&tcbinfo); 1373 } else 1374 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1375 0, cmd, NULL, notify); 1376 } 1377 #endif /* INET6 */ 1378 1379 1380 /* 1381 * Following is where TCP initial sequence number generation occurs. 1382 * 1383 * There are two places where we must use initial sequence numbers: 1384 * 1. In SYN-ACK packets. 1385 * 2. In SYN packets. 1386 * 1387 * All ISNs for SYN-ACK packets are generated by the syncache. See 1388 * tcp_syncache.c for details. 1389 * 1390 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1391 * depends on this property. In addition, these ISNs should be 1392 * unguessable so as to prevent connection hijacking. To satisfy 1393 * the requirements of this situation, the algorithm outlined in 1394 * RFC 1948 is used, with only small modifications. 1395 * 1396 * Implementation details: 1397 * 1398 * Time is based off the system timer, and is corrected so that it 1399 * increases by one megabyte per second. This allows for proper 1400 * recycling on high speed LANs while still leaving over an hour 1401 * before rollover. 1402 * 1403 * As reading the *exact* system time is too expensive to be done 1404 * whenever setting up a TCP connection, we increment the time 1405 * offset in two ways. First, a small random positive increment 1406 * is added to isn_offset for each connection that is set up. 1407 * Second, the function tcp_isn_tick fires once per clock tick 1408 * and increments isn_offset as necessary so that sequence numbers 1409 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1410 * random positive increments serve only to ensure that the same 1411 * exact sequence number is never sent out twice (as could otherwise 1412 * happen when a port is recycled in less than the system tick 1413 * interval.) 1414 * 1415 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1416 * between seeding of isn_secret. This is normally set to zero, 1417 * as reseeding should not be necessary. 1418 * 1419 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1420 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1421 * general, this means holding an exclusive (write) lock. 1422 */ 1423 1424 #define ISN_BYTES_PER_SECOND 1048576 1425 #define ISN_STATIC_INCREMENT 4096 1426 #define ISN_RANDOM_INCREMENT (4096 - 1) 1427 1428 static u_char isn_secret[32]; 1429 static int isn_last_reseed; 1430 static u_int32_t isn_offset, isn_offset_old; 1431 static MD5_CTX isn_ctx; 1432 1433 tcp_seq 1434 tcp_new_isn(struct tcpcb *tp) 1435 { 1436 u_int32_t md5_buffer[4]; 1437 tcp_seq new_isn; 1438 1439 INP_LOCK_ASSERT(tp->t_inpcb); 1440 1441 ISN_LOCK(); 1442 /* Seed if this is the first use, reseed if requested. */ 1443 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1444 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1445 < (u_int)ticks))) { 1446 read_random(&isn_secret, sizeof(isn_secret)); 1447 isn_last_reseed = ticks; 1448 } 1449 1450 /* Compute the md5 hash and return the ISN. */ 1451 MD5Init(&isn_ctx); 1452 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1453 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1454 #ifdef INET6 1455 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1456 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1457 sizeof(struct in6_addr)); 1458 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1459 sizeof(struct in6_addr)); 1460 } else 1461 #endif 1462 { 1463 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1464 sizeof(struct in_addr)); 1465 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1466 sizeof(struct in_addr)); 1467 } 1468 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1469 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1470 new_isn = (tcp_seq) md5_buffer[0]; 1471 isn_offset += ISN_STATIC_INCREMENT + 1472 (arc4random() & ISN_RANDOM_INCREMENT); 1473 new_isn += isn_offset; 1474 ISN_UNLOCK(); 1475 return (new_isn); 1476 } 1477 1478 /* 1479 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1480 * to keep time flowing at a relatively constant rate. If the random 1481 * increments have already pushed us past the projected offset, do nothing. 1482 */ 1483 static void 1484 tcp_isn_tick(void *xtp) 1485 { 1486 u_int32_t projected_offset; 1487 1488 ISN_LOCK(); 1489 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1490 1491 if (projected_offset > isn_offset) 1492 isn_offset = projected_offset; 1493 1494 isn_offset_old = isn_offset; 1495 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1496 ISN_UNLOCK(); 1497 } 1498 1499 /* 1500 * When a specific ICMP unreachable message is received and the 1501 * connection state is SYN-SENT, drop the connection. This behavior 1502 * is controlled by the icmp_may_rst sysctl. 1503 */ 1504 struct inpcb * 1505 tcp_drop_syn_sent(struct inpcb *inp, int errno) 1506 { 1507 struct tcpcb *tp; 1508 1509 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1510 INP_LOCK_ASSERT(inp); 1511 1512 if ((inp->inp_vflag & INP_TIMEWAIT) || 1513 (inp->inp_vflag & INP_DROPPED)) 1514 return (inp); 1515 1516 tp = intotcpcb(inp); 1517 if (tp->t_state != TCPS_SYN_SENT) 1518 return (inp); 1519 1520 tp = tcp_drop(tp, errno); 1521 if (tp != NULL) 1522 return (inp); 1523 else 1524 return (NULL); 1525 } 1526 1527 /* 1528 * When `need fragmentation' ICMP is received, update our idea of the MSS 1529 * based on the new value in the route. Also nudge TCP to send something, 1530 * since we know the packet we just sent was dropped. 1531 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1532 */ 1533 struct inpcb * 1534 tcp_mtudisc(struct inpcb *inp, int errno) 1535 { 1536 struct tcpcb *tp; 1537 struct socket *so = inp->inp_socket; 1538 u_int maxmtu; 1539 u_int romtu; 1540 int mss; 1541 #ifdef INET6 1542 int isipv6; 1543 #endif /* INET6 */ 1544 1545 INP_LOCK_ASSERT(inp); 1546 if ((inp->inp_vflag & INP_TIMEWAIT) || 1547 (inp->inp_vflag & INP_DROPPED)) 1548 return (inp); 1549 1550 tp = intotcpcb(inp); 1551 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1552 1553 #ifdef INET6 1554 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1555 #endif 1556 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1557 romtu = 1558 #ifdef INET6 1559 isipv6 ? tcp_maxmtu6(&inp->inp_inc, NULL) : 1560 #endif /* INET6 */ 1561 tcp_maxmtu(&inp->inp_inc, NULL); 1562 if (!maxmtu) 1563 maxmtu = romtu; 1564 else 1565 maxmtu = min(maxmtu, romtu); 1566 if (!maxmtu) { 1567 tp->t_maxopd = tp->t_maxseg = 1568 #ifdef INET6 1569 isipv6 ? tcp_v6mssdflt : 1570 #endif /* INET6 */ 1571 tcp_mssdflt; 1572 return (inp); 1573 } 1574 mss = maxmtu - 1575 #ifdef INET6 1576 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1577 #endif /* INET6 */ 1578 sizeof(struct tcpiphdr) 1579 #ifdef INET6 1580 ) 1581 #endif /* INET6 */ 1582 ; 1583 1584 /* 1585 * XXX - The above conditional probably violates the TCP 1586 * spec. The problem is that, since we don't know the 1587 * other end's MSS, we are supposed to use a conservative 1588 * default. But, if we do that, then MTU discovery will 1589 * never actually take place, because the conservative 1590 * default is much less than the MTUs typically seen 1591 * on the Internet today. For the moment, we'll sweep 1592 * this under the carpet. 1593 * 1594 * The conservative default might not actually be a problem 1595 * if the only case this occurs is when sending an initial 1596 * SYN with options and data to a host we've never talked 1597 * to before. Then, they will reply with an MSS value which 1598 * will get recorded and the new parameters should get 1599 * recomputed. For Further Study. 1600 */ 1601 if (tp->t_maxopd <= mss) 1602 return (inp); 1603 tp->t_maxopd = mss; 1604 1605 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1606 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1607 mss -= TCPOLEN_TSTAMP_APPA; 1608 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1609 if (mss > MCLBYTES) 1610 mss &= ~(MCLBYTES-1); 1611 #else 1612 if (mss > MCLBYTES) 1613 mss = mss / MCLBYTES * MCLBYTES; 1614 #endif 1615 if (so->so_snd.sb_hiwat < mss) 1616 mss = so->so_snd.sb_hiwat; 1617 1618 tp->t_maxseg = mss; 1619 1620 tcpstat.tcps_mturesent++; 1621 tp->t_rtttime = 0; 1622 tp->snd_nxt = tp->snd_una; 1623 tcp_free_sackholes(tp); 1624 tp->snd_recover = tp->snd_max; 1625 if (tp->sack_enable) 1626 EXIT_FASTRECOVERY(tp); 1627 tcp_output(tp); 1628 return (inp); 1629 } 1630 1631 /* 1632 * Look-up the routing entry to the peer of this inpcb. If no route 1633 * is found and it cannot be allocated, then return NULL. This routine 1634 * is called by TCP routines that access the rmx structure and by tcp_mss 1635 * to get the interface MTU. 1636 */ 1637 u_long 1638 tcp_maxmtu(struct in_conninfo *inc, int *flags) 1639 { 1640 struct route sro; 1641 struct sockaddr_in *dst; 1642 struct ifnet *ifp; 1643 u_long maxmtu = 0; 1644 1645 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1646 1647 bzero(&sro, sizeof(sro)); 1648 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1649 dst = (struct sockaddr_in *)&sro.ro_dst; 1650 dst->sin_family = AF_INET; 1651 dst->sin_len = sizeof(*dst); 1652 dst->sin_addr = inc->inc_faddr; 1653 rtalloc_ign(&sro, RTF_CLONING); 1654 } 1655 if (sro.ro_rt != NULL) { 1656 ifp = sro.ro_rt->rt_ifp; 1657 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1658 maxmtu = ifp->if_mtu; 1659 else 1660 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1661 1662 /* Report additional interface capabilities. */ 1663 if (flags != NULL) { 1664 if (ifp->if_capenable & IFCAP_TSO4 && 1665 ifp->if_hwassist & CSUM_TSO) 1666 *flags |= CSUM_TSO; 1667 } 1668 RTFREE(sro.ro_rt); 1669 } 1670 return (maxmtu); 1671 } 1672 1673 #ifdef INET6 1674 u_long 1675 tcp_maxmtu6(struct in_conninfo *inc, int *flags) 1676 { 1677 struct route_in6 sro6; 1678 struct ifnet *ifp; 1679 u_long maxmtu = 0; 1680 1681 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1682 1683 bzero(&sro6, sizeof(sro6)); 1684 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1685 sro6.ro_dst.sin6_family = AF_INET6; 1686 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1687 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1688 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1689 } 1690 if (sro6.ro_rt != NULL) { 1691 ifp = sro6.ro_rt->rt_ifp; 1692 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1693 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1694 else 1695 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1696 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1697 1698 /* Report additional interface capabilities. */ 1699 if (flags != NULL) { 1700 if (ifp->if_capenable & IFCAP_TSO6 && 1701 ifp->if_hwassist & CSUM_TSO) 1702 *flags |= CSUM_TSO; 1703 } 1704 RTFREE(sro6.ro_rt); 1705 } 1706 1707 return (maxmtu); 1708 } 1709 #endif /* INET6 */ 1710 1711 #ifdef IPSEC 1712 /* compute ESP/AH header size for TCP, including outer IP header. */ 1713 size_t 1714 ipsec_hdrsiz_tcp(struct tcpcb *tp) 1715 { 1716 struct inpcb *inp; 1717 struct mbuf *m; 1718 size_t hdrsiz; 1719 struct ip *ip; 1720 #ifdef INET6 1721 struct ip6_hdr *ip6; 1722 #endif 1723 struct tcphdr *th; 1724 1725 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1726 return (0); 1727 MGETHDR(m, M_DONTWAIT, MT_DATA); 1728 if (!m) 1729 return (0); 1730 1731 #ifdef INET6 1732 if ((inp->inp_vflag & INP_IPV6) != 0) { 1733 ip6 = mtod(m, struct ip6_hdr *); 1734 th = (struct tcphdr *)(ip6 + 1); 1735 m->m_pkthdr.len = m->m_len = 1736 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1737 tcpip_fillheaders(inp, ip6, th); 1738 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1739 } else 1740 #endif /* INET6 */ 1741 { 1742 ip = mtod(m, struct ip *); 1743 th = (struct tcphdr *)(ip + 1); 1744 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1745 tcpip_fillheaders(inp, ip, th); 1746 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1747 } 1748 1749 m_free(m); 1750 return (hdrsiz); 1751 } 1752 #endif /*IPSEC*/ 1753 1754 /* 1755 * Move a TCP connection into TIME_WAIT state. 1756 * tcbinfo is locked. 1757 * inp is locked, and is unlocked before returning. 1758 */ 1759 void 1760 tcp_twstart(struct tcpcb *tp) 1761 { 1762 struct tcptw *tw; 1763 struct inpcb *inp = tp->t_inpcb; 1764 int acknow; 1765 struct socket *so; 1766 1767 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1768 INP_LOCK_ASSERT(inp); 1769 1770 if (nolocaltimewait && in_localip(inp->inp_faddr)) { 1771 tp = tcp_close(tp); 1772 if (tp != NULL) 1773 INP_UNLOCK(inp); 1774 return; 1775 } 1776 1777 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1778 if (tw == NULL) { 1779 tw = tcp_timer_2msl_tw(1); 1780 if (tw == NULL) { 1781 tp = tcp_close(tp); 1782 if (tp != NULL) 1783 INP_UNLOCK(inp); 1784 return; 1785 } 1786 } 1787 tw->tw_inpcb = inp; 1788 1789 /* 1790 * Recover last window size sent. 1791 */ 1792 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1793 1794 /* 1795 * Set t_recent if timestamps are used on the connection. 1796 */ 1797 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1798 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1799 tw->t_recent = tp->ts_recent; 1800 else 1801 tw->t_recent = 0; 1802 1803 tw->snd_nxt = tp->snd_nxt; 1804 tw->rcv_nxt = tp->rcv_nxt; 1805 tw->iss = tp->iss; 1806 tw->irs = tp->irs; 1807 tw->t_starttime = tp->t_starttime; 1808 tw->tw_time = 0; 1809 1810 /* XXX 1811 * If this code will 1812 * be used for fin-wait-2 state also, then we may need 1813 * a ts_recent from the last segment. 1814 */ 1815 acknow = tp->t_flags & TF_ACKNOW; 1816 1817 /* 1818 * First, discard tcpcb state, which includes stopping its timers and 1819 * freeing it. tcp_discardcb() used to also release the inpcb, but 1820 * that work is now done in the caller. 1821 * 1822 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 1823 * and might not be needed here any longer. 1824 */ 1825 tcp_discardcb(tp); 1826 so = inp->inp_socket; 1827 soisdisconnected(so); 1828 SOCK_LOCK(so); 1829 tw->tw_cred = crhold(so->so_cred); 1830 tw->tw_so_options = so->so_options; 1831 SOCK_UNLOCK(so); 1832 if (acknow) 1833 tcp_twrespond(tw, TH_ACK); 1834 inp->inp_ppcb = tw; 1835 inp->inp_vflag |= INP_TIMEWAIT; 1836 tcp_timer_2msl_reset(tw, 0); 1837 1838 /* 1839 * If the inpcb owns the sole reference to the socket, then we can 1840 * detach and free the socket as it is not needed in time wait. 1841 */ 1842 if (inp->inp_vflag & INP_SOCKREF) { 1843 KASSERT(so->so_state & SS_PROTOREF, 1844 ("tcp_twstart: !SS_PROTOREF")); 1845 inp->inp_vflag &= ~INP_SOCKREF; 1846 INP_UNLOCK(inp); 1847 ACCEPT_LOCK(); 1848 SOCK_LOCK(so); 1849 so->so_state &= ~SS_PROTOREF; 1850 sofree(so); 1851 } else 1852 INP_UNLOCK(inp); 1853 } 1854 1855 #if 0 1856 /* 1857 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1858 * the actual rate is slightly higher due to the addition of 1859 * random positive increments. 1860 * 1861 * Most other new OSes use semi-randomized ISN values, so we 1862 * do not need to worry about them. 1863 */ 1864 #define MS_ISN_BYTES_PER_SECOND 250000 1865 1866 /* 1867 * Determine if the ISN we will generate has advanced beyond the last 1868 * sequence number used by the previous connection. If so, indicate 1869 * that it is safe to recycle this tw socket by returning 1. 1870 */ 1871 int 1872 tcp_twrecycleable(struct tcptw *tw) 1873 { 1874 tcp_seq new_iss = tw->iss; 1875 tcp_seq new_irs = tw->irs; 1876 1877 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1878 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1879 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1880 1881 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1882 return (1); 1883 else 1884 return (0); 1885 } 1886 #endif 1887 1888 void 1889 tcp_twclose(struct tcptw *tw, int reuse) 1890 { 1891 struct socket *so; 1892 struct inpcb *inp; 1893 1894 /* 1895 * At this point, we are in one of two situations: 1896 * 1897 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 1898 * all state. 1899 * 1900 * (2) We have a socket -- if we own a reference, release it and 1901 * notify the socket layer. 1902 */ 1903 inp = tw->tw_inpcb; 1904 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1905 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 1906 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1907 INP_LOCK_ASSERT(inp); 1908 1909 tw->tw_inpcb = NULL; 1910 tcp_timer_2msl_stop(tw); 1911 inp->inp_ppcb = NULL; 1912 in_pcbdrop(inp); 1913 1914 so = inp->inp_socket; 1915 if (so != NULL) { 1916 /* 1917 * If there's a socket, handle two cases: first, we own a 1918 * strong reference, which we will now release, or we don't 1919 * in which case another reference exists (XXXRW: think 1920 * about this more), and we don't need to take action. 1921 */ 1922 if (inp->inp_vflag & INP_SOCKREF) { 1923 inp->inp_vflag &= ~INP_SOCKREF; 1924 INP_UNLOCK(inp); 1925 ACCEPT_LOCK(); 1926 SOCK_LOCK(so); 1927 KASSERT(so->so_state & SS_PROTOREF, 1928 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1929 so->so_state &= ~SS_PROTOREF; 1930 sofree(so); 1931 } else { 1932 /* 1933 * If we don't own the only reference, the socket and 1934 * inpcb need to be left around to be handled by 1935 * tcp_usr_detach() later. 1936 */ 1937 INP_UNLOCK(inp); 1938 } 1939 } else { 1940 #ifdef INET6 1941 if (inp->inp_vflag & INP_IPV6PROTO) 1942 in6_pcbfree(inp); 1943 else 1944 #endif 1945 in_pcbfree(inp); 1946 } 1947 tcpstat.tcps_closed++; 1948 crfree(tw->tw_cred); 1949 tw->tw_cred = NULL; 1950 if (reuse) 1951 return; 1952 uma_zfree(tcptw_zone, tw); 1953 } 1954 1955 int 1956 tcp_twrespond(struct tcptw *tw, int flags) 1957 { 1958 struct inpcb *inp = tw->tw_inpcb; 1959 struct tcphdr *th; 1960 struct mbuf *m; 1961 struct ip *ip = NULL; 1962 u_int8_t *optp; 1963 u_int hdrlen, optlen; 1964 int error; 1965 #ifdef INET6 1966 struct ip6_hdr *ip6 = NULL; 1967 int isipv6 = inp->inp_inc.inc_isipv6; 1968 #endif 1969 1970 INP_LOCK_ASSERT(inp); 1971 1972 m = m_gethdr(M_DONTWAIT, MT_DATA); 1973 if (m == NULL) 1974 return (ENOBUFS); 1975 m->m_data += max_linkhdr; 1976 1977 #ifdef MAC 1978 mac_create_mbuf_from_inpcb(inp, m); 1979 #endif 1980 1981 #ifdef INET6 1982 if (isipv6) { 1983 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1984 ip6 = mtod(m, struct ip6_hdr *); 1985 th = (struct tcphdr *)(ip6 + 1); 1986 tcpip_fillheaders(inp, ip6, th); 1987 } else 1988 #endif 1989 { 1990 hdrlen = sizeof(struct tcpiphdr); 1991 ip = mtod(m, struct ip *); 1992 th = (struct tcphdr *)(ip + 1); 1993 tcpip_fillheaders(inp, ip, th); 1994 } 1995 optp = (u_int8_t *)(th + 1); 1996 1997 /* 1998 * Send a timestamp and echo-reply if both our side and our peer 1999 * have sent timestamps in our SYN's and this is not a RST. 2000 */ 2001 if (tw->t_recent && flags == TH_ACK) { 2002 u_int32_t *lp = (u_int32_t *)optp; 2003 2004 /* Form timestamp option as shown in appendix A of RFC 1323. */ 2005 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 2006 *lp++ = htonl(ticks); 2007 *lp = htonl(tw->t_recent); 2008 optp += TCPOLEN_TSTAMP_APPA; 2009 } 2010 2011 optlen = optp - (u_int8_t *)(th + 1); 2012 2013 m->m_len = hdrlen + optlen; 2014 m->m_pkthdr.len = m->m_len; 2015 2016 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 2017 2018 th->th_seq = htonl(tw->snd_nxt); 2019 th->th_ack = htonl(tw->rcv_nxt); 2020 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 2021 th->th_flags = flags; 2022 th->th_win = htons(tw->last_win); 2023 2024 #ifdef INET6 2025 if (isipv6) { 2026 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 2027 sizeof(struct tcphdr) + optlen); 2028 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2029 error = ip6_output(m, inp->in6p_outputopts, NULL, 2030 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 2031 } else 2032 #endif 2033 { 2034 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2035 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 2036 m->m_pkthdr.csum_flags = CSUM_TCP; 2037 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2038 ip->ip_len = m->m_pkthdr.len; 2039 if (path_mtu_discovery) 2040 ip->ip_off |= IP_DF; 2041 error = ip_output(m, inp->inp_options, NULL, 2042 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 2043 NULL, inp); 2044 } 2045 if (flags & TH_ACK) 2046 tcpstat.tcps_sndacks++; 2047 else 2048 tcpstat.tcps_sndctrl++; 2049 tcpstat.tcps_sndtotal++; 2050 return (error); 2051 } 2052 2053 /* 2054 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 2055 * 2056 * This code attempts to calculate the bandwidth-delay product as a 2057 * means of determining the optimal window size to maximize bandwidth, 2058 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 2059 * routers. This code also does a fairly good job keeping RTTs in check 2060 * across slow links like modems. We implement an algorithm which is very 2061 * similar (but not meant to be) TCP/Vegas. The code operates on the 2062 * transmitter side of a TCP connection and so only effects the transmit 2063 * side of the connection. 2064 * 2065 * BACKGROUND: TCP makes no provision for the management of buffer space 2066 * at the end points or at the intermediate routers and switches. A TCP 2067 * stream, whether using NewReno or not, will eventually buffer as 2068 * many packets as it is able and the only reason this typically works is 2069 * due to the fairly small default buffers made available for a connection 2070 * (typicaly 16K or 32K). As machines use larger windows and/or window 2071 * scaling it is now fairly easy for even a single TCP connection to blow-out 2072 * all available buffer space not only on the local interface, but on 2073 * intermediate routers and switches as well. NewReno makes a misguided 2074 * attempt to 'solve' this problem by waiting for an actual failure to occur, 2075 * then backing off, then steadily increasing the window again until another 2076 * failure occurs, ad-infinitum. This results in terrible oscillation that 2077 * is only made worse as network loads increase and the idea of intentionally 2078 * blowing out network buffers is, frankly, a terrible way to manage network 2079 * resources. 2080 * 2081 * It is far better to limit the transmit window prior to the failure 2082 * condition being achieved. There are two general ways to do this: First 2083 * you can 'scan' through different transmit window sizes and locate the 2084 * point where the RTT stops increasing, indicating that you have filled the 2085 * pipe, then scan backwards until you note that RTT stops decreasing, then 2086 * repeat ad-infinitum. This method works in principle but has severe 2087 * implementation issues due to RTT variances, timer granularity, and 2088 * instability in the algorithm which can lead to many false positives and 2089 * create oscillations as well as interact badly with other TCP streams 2090 * implementing the same algorithm. 2091 * 2092 * The second method is to limit the window to the bandwidth delay product 2093 * of the link. This is the method we implement. RTT variances and our 2094 * own manipulation of the congestion window, bwnd, can potentially 2095 * destabilize the algorithm. For this reason we have to stabilize the 2096 * elements used to calculate the window. We do this by using the minimum 2097 * observed RTT, the long term average of the observed bandwidth, and 2098 * by adding two segments worth of slop. It isn't perfect but it is able 2099 * to react to changing conditions and gives us a very stable basis on 2100 * which to extend the algorithm. 2101 */ 2102 void 2103 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2104 { 2105 u_long bw; 2106 u_long bwnd; 2107 int save_ticks; 2108 2109 INP_LOCK_ASSERT(tp->t_inpcb); 2110 2111 /* 2112 * If inflight_enable is disabled in the middle of a tcp connection, 2113 * make sure snd_bwnd is effectively disabled. 2114 */ 2115 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2116 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2117 tp->snd_bandwidth = 0; 2118 return; 2119 } 2120 2121 /* 2122 * Figure out the bandwidth. Due to the tick granularity this 2123 * is a very rough number and it MUST be averaged over a fairly 2124 * long period of time. XXX we need to take into account a link 2125 * that is not using all available bandwidth, but for now our 2126 * slop will ramp us up if this case occurs and the bandwidth later 2127 * increases. 2128 * 2129 * Note: if ticks rollover 'bw' may wind up negative. We must 2130 * effectively reset t_bw_rtttime for this case. 2131 */ 2132 save_ticks = ticks; 2133 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2134 return; 2135 2136 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2137 (save_ticks - tp->t_bw_rtttime); 2138 tp->t_bw_rtttime = save_ticks; 2139 tp->t_bw_rtseq = ack_seq; 2140 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2141 return; 2142 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2143 2144 tp->snd_bandwidth = bw; 2145 2146 /* 2147 * Calculate the semi-static bandwidth delay product, plus two maximal 2148 * segments. The additional slop puts us squarely in the sweet 2149 * spot and also handles the bandwidth run-up case and stabilization. 2150 * Without the slop we could be locking ourselves into a lower 2151 * bandwidth. 2152 * 2153 * Situations Handled: 2154 * (1) Prevents over-queueing of packets on LANs, especially on 2155 * high speed LANs, allowing larger TCP buffers to be 2156 * specified, and also does a good job preventing 2157 * over-queueing of packets over choke points like modems 2158 * (at least for the transmit side). 2159 * 2160 * (2) Is able to handle changing network loads (bandwidth 2161 * drops so bwnd drops, bandwidth increases so bwnd 2162 * increases). 2163 * 2164 * (3) Theoretically should stabilize in the face of multiple 2165 * connections implementing the same algorithm (this may need 2166 * a little work). 2167 * 2168 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2169 * be adjusted with a sysctl but typically only needs to be 2170 * on very slow connections. A value no smaller then 5 2171 * should be used, but only reduce this default if you have 2172 * no other choice. 2173 */ 2174 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2175 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2176 #undef USERTT 2177 2178 if (tcp_inflight_debug > 0) { 2179 static int ltime; 2180 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2181 ltime = ticks; 2182 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2183 tp, 2184 bw, 2185 tp->t_rttbest, 2186 tp->t_srtt, 2187 bwnd 2188 ); 2189 } 2190 } 2191 if ((long)bwnd < tcp_inflight_min) 2192 bwnd = tcp_inflight_min; 2193 if (bwnd > tcp_inflight_max) 2194 bwnd = tcp_inflight_max; 2195 if ((long)bwnd < tp->t_maxseg * 2) 2196 bwnd = tp->t_maxseg * 2; 2197 tp->snd_bwnd = bwnd; 2198 } 2199 2200 #ifdef TCP_SIGNATURE 2201 /* 2202 * Callback function invoked by m_apply() to digest TCP segment data 2203 * contained within an mbuf chain. 2204 */ 2205 static int 2206 tcp_signature_apply(void *fstate, void *data, u_int len) 2207 { 2208 2209 MD5Update(fstate, (u_char *)data, len); 2210 return (0); 2211 } 2212 2213 /* 2214 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2215 * 2216 * Parameters: 2217 * m pointer to head of mbuf chain 2218 * off0 offset to TCP header within the mbuf chain 2219 * len length of TCP segment data, excluding options 2220 * optlen length of TCP segment options 2221 * buf pointer to storage for computed MD5 digest 2222 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2223 * 2224 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2225 * When called from tcp_input(), we can be sure that th_sum has been 2226 * zeroed out and verified already. 2227 * 2228 * This function is for IPv4 use only. Calling this function with an 2229 * IPv6 packet in the mbuf chain will yield undefined results. 2230 * 2231 * Return 0 if successful, otherwise return -1. 2232 * 2233 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2234 * search with the destination IP address, and a 'magic SPI' to be 2235 * determined by the application. This is hardcoded elsewhere to 1179 2236 * right now. Another branch of this code exists which uses the SPD to 2237 * specify per-application flows but it is unstable. 2238 */ 2239 int 2240 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2241 u_char *buf, u_int direction) 2242 { 2243 union sockaddr_union dst; 2244 struct ippseudo ippseudo; 2245 MD5_CTX ctx; 2246 int doff; 2247 struct ip *ip; 2248 struct ipovly *ipovly; 2249 struct secasvar *sav; 2250 struct tcphdr *th; 2251 u_short savecsum; 2252 2253 KASSERT(m != NULL, ("NULL mbuf chain")); 2254 KASSERT(buf != NULL, ("NULL signature pointer")); 2255 2256 /* Extract the destination from the IP header in the mbuf. */ 2257 ip = mtod(m, struct ip *); 2258 bzero(&dst, sizeof(union sockaddr_union)); 2259 dst.sa.sa_len = sizeof(struct sockaddr_in); 2260 dst.sa.sa_family = AF_INET; 2261 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2262 ip->ip_src : ip->ip_dst; 2263 2264 /* Look up an SADB entry which matches the address of the peer. */ 2265 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2266 if (sav == NULL) { 2267 printf("%s: SADB lookup failed for %s\n", __func__, 2268 inet_ntoa(dst.sin.sin_addr)); 2269 return (EINVAL); 2270 } 2271 2272 MD5Init(&ctx); 2273 ipovly = (struct ipovly *)ip; 2274 th = (struct tcphdr *)((u_char *)ip + off0); 2275 doff = off0 + sizeof(struct tcphdr) + optlen; 2276 2277 /* 2278 * Step 1: Update MD5 hash with IP pseudo-header. 2279 * 2280 * XXX The ippseudo header MUST be digested in network byte order, 2281 * or else we'll fail the regression test. Assume all fields we've 2282 * been doing arithmetic on have been in host byte order. 2283 * XXX One cannot depend on ipovly->ih_len here. When called from 2284 * tcp_output(), the underlying ip_len member has not yet been set. 2285 */ 2286 ippseudo.ippseudo_src = ipovly->ih_src; 2287 ippseudo.ippseudo_dst = ipovly->ih_dst; 2288 ippseudo.ippseudo_pad = 0; 2289 ippseudo.ippseudo_p = IPPROTO_TCP; 2290 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2291 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2292 2293 /* 2294 * Step 2: Update MD5 hash with TCP header, excluding options. 2295 * The TCP checksum must be set to zero. 2296 */ 2297 savecsum = th->th_sum; 2298 th->th_sum = 0; 2299 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2300 th->th_sum = savecsum; 2301 2302 /* 2303 * Step 3: Update MD5 hash with TCP segment data. 2304 * Use m_apply() to avoid an early m_pullup(). 2305 */ 2306 if (len > 0) 2307 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2308 2309 /* 2310 * Step 4: Update MD5 hash with shared secret. 2311 */ 2312 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2313 MD5Final(buf, &ctx); 2314 2315 key_sa_recordxfer(sav, m); 2316 KEY_FREESAV(&sav); 2317 return (0); 2318 } 2319 #endif /* TCP_SIGNATURE */ 2320 2321 static int 2322 sysctl_drop(SYSCTL_HANDLER_ARGS) 2323 { 2324 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2325 struct sockaddr_storage addrs[2]; 2326 struct inpcb *inp; 2327 struct tcpcb *tp; 2328 struct tcptw *tw; 2329 struct sockaddr_in *fin, *lin; 2330 #ifdef INET6 2331 struct sockaddr_in6 *fin6, *lin6; 2332 struct in6_addr f6, l6; 2333 #endif 2334 int error; 2335 2336 inp = NULL; 2337 fin = lin = NULL; 2338 #ifdef INET6 2339 fin6 = lin6 = NULL; 2340 #endif 2341 error = 0; 2342 2343 if (req->oldptr != NULL || req->oldlen != 0) 2344 return (EINVAL); 2345 if (req->newptr == NULL) 2346 return (EPERM); 2347 if (req->newlen < sizeof(addrs)) 2348 return (ENOMEM); 2349 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2350 if (error) 2351 return (error); 2352 2353 switch (addrs[0].ss_family) { 2354 #ifdef INET6 2355 case AF_INET6: 2356 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2357 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2358 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2359 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2360 return (EINVAL); 2361 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2362 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2363 return (EINVAL); 2364 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2365 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2366 fin = (struct sockaddr_in *)&addrs[0]; 2367 lin = (struct sockaddr_in *)&addrs[1]; 2368 break; 2369 } 2370 error = sa6_embedscope(fin6, ip6_use_defzone); 2371 if (error) 2372 return (error); 2373 error = sa6_embedscope(lin6, ip6_use_defzone); 2374 if (error) 2375 return (error); 2376 break; 2377 #endif 2378 case AF_INET: 2379 fin = (struct sockaddr_in *)&addrs[0]; 2380 lin = (struct sockaddr_in *)&addrs[1]; 2381 if (fin->sin_len != sizeof(struct sockaddr_in) || 2382 lin->sin_len != sizeof(struct sockaddr_in)) 2383 return (EINVAL); 2384 break; 2385 default: 2386 return (EINVAL); 2387 } 2388 INP_INFO_WLOCK(&tcbinfo); 2389 switch (addrs[0].ss_family) { 2390 #ifdef INET6 2391 case AF_INET6: 2392 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2393 &l6, lin6->sin6_port, 0, NULL); 2394 break; 2395 #endif 2396 case AF_INET: 2397 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2398 lin->sin_addr, lin->sin_port, 0, NULL); 2399 break; 2400 } 2401 if (inp != NULL) { 2402 INP_LOCK(inp); 2403 if (inp->inp_vflag & INP_TIMEWAIT) { 2404 /* 2405 * XXXRW: There currently exists a state where an 2406 * inpcb is present, but its timewait state has been 2407 * discarded. For now, don't allow dropping of this 2408 * type of inpcb. 2409 */ 2410 tw = intotw(inp); 2411 if (tw != NULL) 2412 tcp_twclose(tw, 0); 2413 } else if (!(inp->inp_vflag & INP_DROPPED) && 2414 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2415 tp = intotcpcb(inp); 2416 tcp_drop(tp, ECONNABORTED); 2417 } 2418 INP_UNLOCK(inp); 2419 } else 2420 error = ESRCH; 2421 INP_INFO_WUNLOCK(&tcbinfo); 2422 return (error); 2423 } 2424 2425 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2426 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2427 0, sysctl_drop, "", "Drop TCP connection"); 2428