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 = mem; 313 314 INP_LOCK_INIT(inp, "inp", "tcpinp"); 315 return (0); 316 } 317 318 void 319 tcp_init(void) 320 { 321 int hashsize = TCBHASHSIZE; 322 323 tcp_delacktime = TCPTV_DELACK; 324 tcp_keepinit = TCPTV_KEEP_INIT; 325 tcp_keepidle = TCPTV_KEEP_IDLE; 326 tcp_keepintvl = TCPTV_KEEPINTVL; 327 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 328 tcp_msl = TCPTV_MSL; 329 tcp_rexmit_min = TCPTV_MIN; 330 tcp_rexmit_slop = TCPTV_CPU_VAR; 331 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH; 332 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 333 334 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 335 LIST_INIT(&tcb); 336 tcbinfo.listhead = &tcb; 337 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 338 if (!powerof2(hashsize)) { 339 printf("WARNING: TCB hash size not a power of 2\n"); 340 hashsize = 512; /* safe default */ 341 } 342 tcp_tcbhashsize = hashsize; 343 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 344 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 345 &tcbinfo.porthashmask); 346 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 347 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 348 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 349 #ifdef INET6 350 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 351 #else /* INET6 */ 352 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 353 #endif /* INET6 */ 354 if (max_protohdr < TCP_MINPROTOHDR) 355 max_protohdr = TCP_MINPROTOHDR; 356 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 357 panic("tcp_init"); 358 #undef TCP_MINPROTOHDR 359 /* 360 * These have to be type stable for the benefit of the timers. 361 */ 362 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 363 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 364 uma_zone_set_max(tcpcb_zone, maxsockets); 365 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 366 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 367 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 368 if (maxtcptw == 0) 369 uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 370 else 371 uma_zone_set_max(tcptw_zone, maxtcptw); 372 tcp_timer_init(); 373 syncache_init(); 374 tcp_hc_init(); 375 tcp_reass_init(); 376 ISN_LOCK_INIT(); 377 callout_init(&isn_callout, CALLOUT_MPSAFE); 378 tcp_isn_tick(NULL); 379 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 380 SHUTDOWN_PRI_DEFAULT); 381 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 382 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 383 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 384 EVENTHANDLER_PRI_ANY); 385 } 386 387 void 388 tcp_fini(void *xtp) 389 { 390 391 callout_stop(&isn_callout); 392 } 393 394 /* 395 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 396 * tcp_template used to store this data in mbufs, but we now recopy it out 397 * of the tcpcb each time to conserve mbufs. 398 */ 399 void 400 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 401 { 402 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 403 404 INP_LOCK_ASSERT(inp); 405 406 #ifdef INET6 407 if ((inp->inp_vflag & INP_IPV6) != 0) { 408 struct ip6_hdr *ip6; 409 410 ip6 = (struct ip6_hdr *)ip_ptr; 411 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 412 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 413 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 414 (IPV6_VERSION & IPV6_VERSION_MASK); 415 ip6->ip6_nxt = IPPROTO_TCP; 416 ip6->ip6_plen = sizeof(struct tcphdr); 417 ip6->ip6_src = inp->in6p_laddr; 418 ip6->ip6_dst = inp->in6p_faddr; 419 } else 420 #endif 421 { 422 struct ip *ip; 423 424 ip = (struct ip *)ip_ptr; 425 ip->ip_v = IPVERSION; 426 ip->ip_hl = 5; 427 ip->ip_tos = inp->inp_ip_tos; 428 ip->ip_len = 0; 429 ip->ip_id = 0; 430 ip->ip_off = 0; 431 ip->ip_ttl = inp->inp_ip_ttl; 432 ip->ip_sum = 0; 433 ip->ip_p = IPPROTO_TCP; 434 ip->ip_src = inp->inp_laddr; 435 ip->ip_dst = inp->inp_faddr; 436 } 437 th->th_sport = inp->inp_lport; 438 th->th_dport = inp->inp_fport; 439 th->th_seq = 0; 440 th->th_ack = 0; 441 th->th_x2 = 0; 442 th->th_off = 5; 443 th->th_flags = 0; 444 th->th_win = 0; 445 th->th_urp = 0; 446 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 447 } 448 449 /* 450 * Create template to be used to send tcp packets on a connection. 451 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 452 * use for this function is in keepalives, which use tcp_respond. 453 */ 454 struct tcptemp * 455 tcpip_maketemplate(struct inpcb *inp) 456 { 457 struct mbuf *m; 458 struct tcptemp *n; 459 460 m = m_get(M_DONTWAIT, MT_DATA); 461 if (m == NULL) 462 return (0); 463 m->m_len = sizeof(struct tcptemp); 464 n = mtod(m, struct tcptemp *); 465 466 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 467 return (n); 468 } 469 470 /* 471 * Send a single message to the TCP at address specified by 472 * the given TCP/IP header. If m == NULL, then we make a copy 473 * of the tcpiphdr at ti and send directly to the addressed host. 474 * This is used to force keep alive messages out using the TCP 475 * template for a connection. If flags are given then we send 476 * a message back to the TCP which originated the * segment ti, 477 * and discard the mbuf containing it and any other attached mbufs. 478 * 479 * In any case the ack and sequence number of the transmitted 480 * segment are as specified by the parameters. 481 * 482 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 483 */ 484 void 485 tcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th, 486 register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) 487 { 488 register int tlen; 489 int win = 0; 490 struct ip *ip; 491 struct tcphdr *nth; 492 #ifdef INET6 493 struct ip6_hdr *ip6; 494 int isipv6; 495 #endif /* INET6 */ 496 int ipflags = 0; 497 struct inpcb *inp; 498 499 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 500 501 #ifdef INET6 502 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 503 ip6 = ipgen; 504 #endif /* INET6 */ 505 ip = ipgen; 506 507 if (tp != NULL) { 508 inp = tp->t_inpcb; 509 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 510 INP_INFO_WLOCK_ASSERT(&tcbinfo); 511 INP_LOCK_ASSERT(inp); 512 } else 513 inp = NULL; 514 515 if (tp != NULL) { 516 if (!(flags & TH_RST)) { 517 win = sbspace(&inp->inp_socket->so_rcv); 518 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 519 win = (long)TCP_MAXWIN << tp->rcv_scale; 520 } 521 } 522 if (m == NULL) { 523 m = m_gethdr(M_DONTWAIT, MT_DATA); 524 if (m == NULL) 525 return; 526 tlen = 0; 527 m->m_data += max_linkhdr; 528 #ifdef INET6 529 if (isipv6) { 530 bcopy((caddr_t)ip6, mtod(m, caddr_t), 531 sizeof(struct ip6_hdr)); 532 ip6 = mtod(m, struct ip6_hdr *); 533 nth = (struct tcphdr *)(ip6 + 1); 534 } else 535 #endif /* INET6 */ 536 { 537 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 538 ip = mtod(m, struct ip *); 539 nth = (struct tcphdr *)(ip + 1); 540 } 541 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 542 flags = TH_ACK; 543 } else { 544 m_freem(m->m_next); 545 m->m_next = NULL; 546 m->m_data = (caddr_t)ipgen; 547 /* m_len is set later */ 548 tlen = 0; 549 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 550 #ifdef INET6 551 if (isipv6) { 552 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 553 nth = (struct tcphdr *)(ip6 + 1); 554 } else 555 #endif /* INET6 */ 556 { 557 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 558 nth = (struct tcphdr *)(ip + 1); 559 } 560 if (th != nth) { 561 /* 562 * this is usually a case when an extension header 563 * exists between the IPv6 header and the 564 * TCP header. 565 */ 566 nth->th_sport = th->th_sport; 567 nth->th_dport = th->th_dport; 568 } 569 xchg(nth->th_dport, nth->th_sport, n_short); 570 #undef xchg 571 } 572 #ifdef INET6 573 if (isipv6) { 574 ip6->ip6_flow = 0; 575 ip6->ip6_vfc = IPV6_VERSION; 576 ip6->ip6_nxt = IPPROTO_TCP; 577 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 578 tlen)); 579 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 580 } else 581 #endif 582 { 583 tlen += sizeof (struct tcpiphdr); 584 ip->ip_len = tlen; 585 ip->ip_ttl = ip_defttl; 586 if (path_mtu_discovery) 587 ip->ip_off |= IP_DF; 588 } 589 m->m_len = tlen; 590 m->m_pkthdr.len = tlen; 591 m->m_pkthdr.rcvif = NULL; 592 #ifdef MAC 593 if (inp != NULL) { 594 /* 595 * Packet is associated with a socket, so allow the 596 * label of the response to reflect the socket label. 597 */ 598 INP_LOCK_ASSERT(inp); 599 mac_create_mbuf_from_inpcb(inp, m); 600 } else { 601 /* 602 * Packet is not associated with a socket, so possibly 603 * update the label in place. 604 */ 605 mac_reflect_mbuf_tcp(m); 606 } 607 #endif 608 nth->th_seq = htonl(seq); 609 nth->th_ack = htonl(ack); 610 nth->th_x2 = 0; 611 nth->th_off = sizeof (struct tcphdr) >> 2; 612 nth->th_flags = flags; 613 if (tp != NULL) 614 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 615 else 616 nth->th_win = htons((u_short)win); 617 nth->th_urp = 0; 618 #ifdef INET6 619 if (isipv6) { 620 nth->th_sum = 0; 621 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 622 sizeof(struct ip6_hdr), 623 tlen - sizeof(struct ip6_hdr)); 624 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 625 NULL, NULL); 626 } else 627 #endif /* INET6 */ 628 { 629 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 630 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 631 m->m_pkthdr.csum_flags = CSUM_TCP; 632 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 633 } 634 #ifdef TCPDEBUG 635 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 636 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 637 #endif 638 #ifdef INET6 639 if (isipv6) 640 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 641 else 642 #endif /* INET6 */ 643 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 644 } 645 646 /* 647 * Create a new TCP control block, making an 648 * empty reassembly queue and hooking it to the argument 649 * protocol control block. The `inp' parameter must have 650 * come from the zone allocator set up in tcp_init(). 651 */ 652 struct tcpcb * 653 tcp_newtcpcb(struct inpcb *inp) 654 { 655 struct tcpcb_mem *tm; 656 struct tcpcb *tp; 657 #ifdef INET6 658 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 659 #endif /* INET6 */ 660 661 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 662 if (tm == NULL) 663 return (NULL); 664 tp = &tm->tcb; 665 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 666 tp->t_maxseg = tp->t_maxopd = 667 #ifdef INET6 668 isipv6 ? tcp_v6mssdflt : 669 #endif /* INET6 */ 670 tcp_mssdflt; 671 672 /* Set up our timeouts. */ 673 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE); 674 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE); 675 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE); 676 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE); 677 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE); 678 679 if (tcp_do_rfc1323) 680 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 681 tp->sack_enable = tcp_do_sack; 682 TAILQ_INIT(&tp->snd_holes); 683 tp->t_inpcb = inp; /* XXX */ 684 /* 685 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 686 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 687 * reasonable initial retransmit time. 688 */ 689 tp->t_srtt = TCPTV_SRTTBASE; 690 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 691 tp->t_rttmin = tcp_rexmit_min; 692 tp->t_rxtcur = TCPTV_RTOBASE; 693 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 694 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 695 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 696 tp->t_rcvtime = ticks; 697 tp->t_bw_rtttime = ticks; 698 /* 699 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 700 * because the socket may be bound to an IPv6 wildcard address, 701 * which may match an IPv4-mapped IPv6 address. 702 */ 703 inp->inp_ip_ttl = ip_defttl; 704 inp->inp_ppcb = tp; 705 return (tp); /* XXX */ 706 } 707 708 /* 709 * Drop a TCP connection, reporting 710 * the specified error. If connection is synchronized, 711 * then send a RST to peer. 712 */ 713 struct tcpcb * 714 tcp_drop(struct tcpcb *tp, int errno) 715 { 716 struct socket *so = tp->t_inpcb->inp_socket; 717 718 INP_INFO_WLOCK_ASSERT(&tcbinfo); 719 INP_LOCK_ASSERT(tp->t_inpcb); 720 721 if (TCPS_HAVERCVDSYN(tp->t_state)) { 722 tp->t_state = TCPS_CLOSED; 723 (void) tcp_output(tp); 724 tcpstat.tcps_drops++; 725 } else 726 tcpstat.tcps_conndrops++; 727 if (errno == ETIMEDOUT && tp->t_softerror) 728 errno = tp->t_softerror; 729 so->so_error = errno; 730 return (tcp_close(tp)); 731 } 732 733 void 734 tcp_discardcb(struct tcpcb *tp) 735 { 736 struct tseg_qent *q; 737 struct inpcb *inp = tp->t_inpcb; 738 struct socket *so = inp->inp_socket; 739 #ifdef INET6 740 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 741 #endif /* INET6 */ 742 743 INP_LOCK_ASSERT(inp); 744 745 /* 746 * Make sure that all of our timers are stopped before we 747 * delete the PCB. 748 */ 749 callout_stop(tp->tt_rexmt); 750 callout_stop(tp->tt_persist); 751 callout_stop(tp->tt_keep); 752 callout_stop(tp->tt_2msl); 753 callout_stop(tp->tt_delack); 754 755 /* 756 * If we got enough samples through the srtt filter, 757 * save the rtt and rttvar in the routing entry. 758 * 'Enough' is arbitrarily defined as 4 rtt samples. 759 * 4 samples is enough for the srtt filter to converge 760 * to within enough % of the correct value; fewer samples 761 * and we could save a bogus rtt. The danger is not high 762 * as tcp quickly recovers from everything. 763 * XXX: Works very well but needs some more statistics! 764 */ 765 if (tp->t_rttupdated >= 4) { 766 struct hc_metrics_lite metrics; 767 u_long ssthresh; 768 769 bzero(&metrics, sizeof(metrics)); 770 /* 771 * Update the ssthresh always when the conditions below 772 * are satisfied. This gives us better new start value 773 * for the congestion avoidance for new connections. 774 * ssthresh is only set if packet loss occured on a session. 775 * 776 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 777 * being torn down. Ideally this code would not use 'so'. 778 */ 779 ssthresh = tp->snd_ssthresh; 780 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 781 /* 782 * convert the limit from user data bytes to 783 * packets then to packet data bytes. 784 */ 785 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 786 if (ssthresh < 2) 787 ssthresh = 2; 788 ssthresh *= (u_long)(tp->t_maxseg + 789 #ifdef INET6 790 (isipv6 ? sizeof (struct ip6_hdr) + 791 sizeof (struct tcphdr) : 792 #endif 793 sizeof (struct tcpiphdr) 794 #ifdef INET6 795 ) 796 #endif 797 ); 798 } else 799 ssthresh = 0; 800 metrics.rmx_ssthresh = ssthresh; 801 802 metrics.rmx_rtt = tp->t_srtt; 803 metrics.rmx_rttvar = tp->t_rttvar; 804 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 805 metrics.rmx_bandwidth = tp->snd_bandwidth; 806 metrics.rmx_cwnd = tp->snd_cwnd; 807 metrics.rmx_sendpipe = 0; 808 metrics.rmx_recvpipe = 0; 809 810 tcp_hc_update(&inp->inp_inc, &metrics); 811 } 812 813 /* free the reassembly queue, if any */ 814 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 815 LIST_REMOVE(q, tqe_q); 816 m_freem(q->tqe_m); 817 uma_zfree(tcp_reass_zone, q); 818 tp->t_segqlen--; 819 tcp_reass_qsize--; 820 } 821 tcp_free_sackholes(tp); 822 inp->inp_ppcb = NULL; 823 tp->t_inpcb = NULL; 824 uma_zfree(tcpcb_zone, tp); 825 } 826 827 /* 828 * Attempt to close a TCP control block, marking it as dropped, and freeing 829 * the socket if we hold the only reference. 830 */ 831 struct tcpcb * 832 tcp_close(struct tcpcb *tp) 833 { 834 struct inpcb *inp = tp->t_inpcb; 835 struct socket *so; 836 837 INP_INFO_WLOCK_ASSERT(&tcbinfo); 838 INP_LOCK_ASSERT(inp); 839 840 in_pcbdrop(inp); 841 tcpstat.tcps_closed++; 842 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 843 so = inp->inp_socket; 844 soisdisconnected(so); 845 if (inp->inp_vflag & INP_SOCKREF) { 846 KASSERT(so->so_state & SS_PROTOREF, 847 ("tcp_close: !SS_PROTOREF")); 848 inp->inp_vflag &= ~INP_SOCKREF; 849 INP_UNLOCK(inp); 850 ACCEPT_LOCK(); 851 SOCK_LOCK(so); 852 so->so_state &= ~SS_PROTOREF; 853 sofree(so); 854 return (NULL); 855 } 856 return (tp); 857 } 858 859 void 860 tcp_drain(void) 861 { 862 863 if (do_tcpdrain) { 864 struct inpcb *inpb; 865 struct tcpcb *tcpb; 866 struct tseg_qent *te; 867 868 /* 869 * Walk the tcpbs, if existing, and flush the reassembly queue, 870 * if there is one... 871 * XXX: The "Net/3" implementation doesn't imply that the TCP 872 * reassembly queue should be flushed, but in a situation 873 * where we're really low on mbufs, this is potentially 874 * usefull. 875 */ 876 INP_INFO_RLOCK(&tcbinfo); 877 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 878 if (inpb->inp_vflag & INP_TIMEWAIT) 879 continue; 880 INP_LOCK(inpb); 881 if ((tcpb = intotcpcb(inpb)) != NULL) { 882 while ((te = LIST_FIRST(&tcpb->t_segq)) 883 != NULL) { 884 LIST_REMOVE(te, tqe_q); 885 m_freem(te->tqe_m); 886 uma_zfree(tcp_reass_zone, te); 887 tcpb->t_segqlen--; 888 tcp_reass_qsize--; 889 } 890 tcp_clean_sackreport(tcpb); 891 } 892 INP_UNLOCK(inpb); 893 } 894 INP_INFO_RUNLOCK(&tcbinfo); 895 } 896 } 897 898 /* 899 * Notify a tcp user of an asynchronous error; 900 * store error as soft error, but wake up user 901 * (for now, won't do anything until can select for soft error). 902 * 903 * Do not wake up user since there currently is no mechanism for 904 * reporting soft errors (yet - a kqueue filter may be added). 905 */ 906 static struct inpcb * 907 tcp_notify(struct inpcb *inp, int error) 908 { 909 struct tcpcb *tp; 910 911 INP_INFO_WLOCK_ASSERT(&tcbinfo); 912 INP_LOCK_ASSERT(inp); 913 914 if ((inp->inp_vflag & INP_TIMEWAIT) || 915 (inp->inp_vflag & INP_DROPPED)) 916 return (inp); 917 918 tp = intotcpcb(inp); 919 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 920 921 /* 922 * Ignore some errors if we are hooked up. 923 * If connection hasn't completed, has retransmitted several times, 924 * and receives a second error, give up now. This is better 925 * than waiting a long time to establish a connection that 926 * can never complete. 927 */ 928 if (tp->t_state == TCPS_ESTABLISHED && 929 (error == EHOSTUNREACH || error == ENETUNREACH || 930 error == EHOSTDOWN)) { 931 return (inp); 932 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 933 tp->t_softerror) { 934 tp = tcp_drop(tp, error); 935 if (tp != NULL) 936 return (inp); 937 else 938 return (NULL); 939 } else { 940 tp->t_softerror = error; 941 return (inp); 942 } 943 #if 0 944 wakeup( &so->so_timeo); 945 sorwakeup(so); 946 sowwakeup(so); 947 #endif 948 } 949 950 static int 951 tcp_pcblist(SYSCTL_HANDLER_ARGS) 952 { 953 int error, i, n; 954 struct inpcb *inp, **inp_list; 955 inp_gen_t gencnt; 956 struct xinpgen xig; 957 958 /* 959 * The process of preparing the TCB list is too time-consuming and 960 * resource-intensive to repeat twice on every request. 961 */ 962 if (req->oldptr == NULL) { 963 n = tcbinfo.ipi_count; 964 req->oldidx = 2 * (sizeof xig) 965 + (n + n/8) * sizeof(struct xtcpcb); 966 return (0); 967 } 968 969 if (req->newptr != NULL) 970 return (EPERM); 971 972 /* 973 * OK, now we're committed to doing something. 974 */ 975 INP_INFO_RLOCK(&tcbinfo); 976 gencnt = tcbinfo.ipi_gencnt; 977 n = tcbinfo.ipi_count; 978 INP_INFO_RUNLOCK(&tcbinfo); 979 980 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 981 + n * sizeof(struct xtcpcb)); 982 if (error != 0) 983 return (error); 984 985 xig.xig_len = sizeof xig; 986 xig.xig_count = n; 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 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 994 if (inp_list == NULL) 995 return (ENOMEM); 996 997 INP_INFO_RLOCK(&tcbinfo); 998 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 999 inp = LIST_NEXT(inp, inp_list)) { 1000 INP_LOCK(inp); 1001 if (inp->inp_gencnt <= gencnt) { 1002 /* 1003 * XXX: This use of cr_cansee(), introduced with 1004 * TCP state changes, is not quite right, but for 1005 * now, better than nothing. 1006 */ 1007 if (inp->inp_vflag & INP_TIMEWAIT) { 1008 if (intotw(inp) != NULL) 1009 error = cr_cansee(req->td->td_ucred, 1010 intotw(inp)->tw_cred); 1011 else 1012 error = EINVAL; /* Skip this inp. */ 1013 } else 1014 error = cr_canseesocket(req->td->td_ucred, 1015 inp->inp_socket); 1016 if (error == 0) 1017 inp_list[i++] = inp; 1018 } 1019 INP_UNLOCK(inp); 1020 } 1021 INP_INFO_RUNLOCK(&tcbinfo); 1022 n = i; 1023 1024 error = 0; 1025 for (i = 0; i < n; i++) { 1026 inp = inp_list[i]; 1027 INP_LOCK(inp); 1028 if (inp->inp_gencnt <= gencnt) { 1029 struct xtcpcb xt; 1030 void *inp_ppcb; 1031 1032 bzero(&xt, sizeof(xt)); 1033 xt.xt_len = sizeof xt; 1034 /* XXX should avoid extra copy */ 1035 bcopy(inp, &xt.xt_inp, sizeof *inp); 1036 inp_ppcb = inp->inp_ppcb; 1037 if (inp_ppcb == NULL) 1038 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1039 else if (inp->inp_vflag & INP_TIMEWAIT) { 1040 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1041 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1042 } else 1043 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1044 if (inp->inp_socket != NULL) 1045 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1046 else { 1047 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1048 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1049 } 1050 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1051 INP_UNLOCK(inp); 1052 error = SYSCTL_OUT(req, &xt, sizeof xt); 1053 } else 1054 INP_UNLOCK(inp); 1055 1056 } 1057 if (!error) { 1058 /* 1059 * Give the user an updated idea of our state. 1060 * If the generation differs from what we told 1061 * her before, she knows that something happened 1062 * while we were processing this request, and it 1063 * might be necessary to retry. 1064 */ 1065 INP_INFO_RLOCK(&tcbinfo); 1066 xig.xig_gen = tcbinfo.ipi_gencnt; 1067 xig.xig_sogen = so_gencnt; 1068 xig.xig_count = tcbinfo.ipi_count; 1069 INP_INFO_RUNLOCK(&tcbinfo); 1070 error = SYSCTL_OUT(req, &xig, sizeof xig); 1071 } 1072 free(inp_list, M_TEMP); 1073 return (error); 1074 } 1075 1076 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1077 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1078 1079 static int 1080 tcp_getcred(SYSCTL_HANDLER_ARGS) 1081 { 1082 struct xucred xuc; 1083 struct sockaddr_in addrs[2]; 1084 struct inpcb *inp; 1085 int error; 1086 1087 error = priv_check_cred(req->td->td_ucred, PRIV_NETINET_GETCRED, 1088 SUSER_ALLOWJAIL); 1089 if (error) 1090 return (error); 1091 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1092 if (error) 1093 return (error); 1094 INP_INFO_RLOCK(&tcbinfo); 1095 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1096 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1097 if (inp == NULL) { 1098 error = ENOENT; 1099 goto outunlocked; 1100 } 1101 INP_LOCK(inp); 1102 if (inp->inp_socket == NULL) { 1103 error = ENOENT; 1104 goto out; 1105 } 1106 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1107 if (error) 1108 goto out; 1109 cru2x(inp->inp_socket->so_cred, &xuc); 1110 out: 1111 INP_UNLOCK(inp); 1112 outunlocked: 1113 INP_INFO_RUNLOCK(&tcbinfo); 1114 if (error == 0) 1115 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1116 return (error); 1117 } 1118 1119 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1120 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1121 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1122 1123 #ifdef INET6 1124 static int 1125 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1126 { 1127 struct xucred xuc; 1128 struct sockaddr_in6 addrs[2]; 1129 struct inpcb *inp; 1130 int error, mapped = 0; 1131 1132 error = priv_check_cred(req->td->td_ucred, PRIV_NETINET_GETCRED, 1133 SUSER_ALLOWJAIL); 1134 if (error) 1135 return (error); 1136 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1137 if (error) 1138 return (error); 1139 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1140 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1141 return (error); 1142 } 1143 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1144 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1145 mapped = 1; 1146 else 1147 return (EINVAL); 1148 } 1149 1150 INP_INFO_RLOCK(&tcbinfo); 1151 if (mapped == 1) 1152 inp = in_pcblookup_hash(&tcbinfo, 1153 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1154 addrs[1].sin6_port, 1155 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1156 addrs[0].sin6_port, 1157 0, NULL); 1158 else 1159 inp = in6_pcblookup_hash(&tcbinfo, 1160 &addrs[1].sin6_addr, addrs[1].sin6_port, 1161 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1162 if (inp == NULL) { 1163 error = ENOENT; 1164 goto outunlocked; 1165 } 1166 INP_LOCK(inp); 1167 if (inp->inp_socket == NULL) { 1168 error = ENOENT; 1169 goto out; 1170 } 1171 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1172 if (error) 1173 goto out; 1174 cru2x(inp->inp_socket->so_cred, &xuc); 1175 out: 1176 INP_UNLOCK(inp); 1177 outunlocked: 1178 INP_INFO_RUNLOCK(&tcbinfo); 1179 if (error == 0) 1180 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1181 return (error); 1182 } 1183 1184 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1185 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1186 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1187 #endif 1188 1189 1190 void 1191 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1192 { 1193 struct ip *ip = vip; 1194 struct tcphdr *th; 1195 struct in_addr faddr; 1196 struct inpcb *inp; 1197 struct tcpcb *tp; 1198 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1199 struct icmp *icp; 1200 struct in_conninfo inc; 1201 tcp_seq icmp_tcp_seq; 1202 int mtu; 1203 1204 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1205 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1206 return; 1207 1208 if (cmd == PRC_MSGSIZE) 1209 notify = tcp_mtudisc; 1210 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1211 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1212 notify = tcp_drop_syn_sent; 1213 /* 1214 * Redirects don't need to be handled up here. 1215 */ 1216 else if (PRC_IS_REDIRECT(cmd)) 1217 return; 1218 /* 1219 * Source quench is depreciated. 1220 */ 1221 else if (cmd == PRC_QUENCH) 1222 return; 1223 /* 1224 * Hostdead is ugly because it goes linearly through all PCBs. 1225 * XXX: We never get this from ICMP, otherwise it makes an 1226 * excellent DoS attack on machines with many connections. 1227 */ 1228 else if (cmd == PRC_HOSTDEAD) 1229 ip = NULL; 1230 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1231 return; 1232 if (ip != NULL) { 1233 icp = (struct icmp *)((caddr_t)ip 1234 - offsetof(struct icmp, icmp_ip)); 1235 th = (struct tcphdr *)((caddr_t)ip 1236 + (ip->ip_hl << 2)); 1237 INP_INFO_WLOCK(&tcbinfo); 1238 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1239 ip->ip_src, th->th_sport, 0, NULL); 1240 if (inp != NULL) { 1241 INP_LOCK(inp); 1242 if (!(inp->inp_vflag & INP_TIMEWAIT) && 1243 !(inp->inp_vflag & INP_DROPPED) && 1244 !(inp->inp_socket == NULL)) { 1245 icmp_tcp_seq = htonl(th->th_seq); 1246 tp = intotcpcb(inp); 1247 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1248 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1249 if (cmd == PRC_MSGSIZE) { 1250 /* 1251 * MTU discovery: 1252 * If we got a needfrag set the MTU 1253 * in the route to the suggested new 1254 * value (if given) and then notify. 1255 */ 1256 bzero(&inc, sizeof(inc)); 1257 inc.inc_flags = 0; /* IPv4 */ 1258 inc.inc_faddr = faddr; 1259 1260 mtu = ntohs(icp->icmp_nextmtu); 1261 /* 1262 * If no alternative MTU was 1263 * proposed, try the next smaller 1264 * one. ip->ip_len has already 1265 * been swapped in icmp_input(). 1266 */ 1267 if (!mtu) 1268 mtu = ip_next_mtu(ip->ip_len, 1269 1); 1270 if (mtu < max(296, (tcp_minmss) 1271 + sizeof(struct tcpiphdr))) 1272 mtu = 0; 1273 if (!mtu) 1274 mtu = tcp_mssdflt 1275 + sizeof(struct tcpiphdr); 1276 /* 1277 * Only cache the the MTU if it 1278 * is smaller than the interface 1279 * or route MTU. tcp_mtudisc() 1280 * will do right thing by itself. 1281 */ 1282 if (mtu <= tcp_maxmtu(&inc, NULL)) 1283 tcp_hc_updatemtu(&inc, mtu); 1284 } 1285 1286 inp = (*notify)(inp, inetctlerrmap[cmd]); 1287 } 1288 } 1289 if (inp != NULL) 1290 INP_UNLOCK(inp); 1291 } else { 1292 inc.inc_fport = th->th_dport; 1293 inc.inc_lport = th->th_sport; 1294 inc.inc_faddr = faddr; 1295 inc.inc_laddr = ip->ip_src; 1296 #ifdef INET6 1297 inc.inc_isipv6 = 0; 1298 #endif 1299 syncache_unreach(&inc, th); 1300 } 1301 INP_INFO_WUNLOCK(&tcbinfo); 1302 } else 1303 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1304 } 1305 1306 #ifdef INET6 1307 void 1308 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1309 { 1310 struct tcphdr th; 1311 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1312 struct ip6_hdr *ip6; 1313 struct mbuf *m; 1314 struct ip6ctlparam *ip6cp = NULL; 1315 const struct sockaddr_in6 *sa6_src = NULL; 1316 int off; 1317 struct tcp_portonly { 1318 u_int16_t th_sport; 1319 u_int16_t th_dport; 1320 } *thp; 1321 1322 if (sa->sa_family != AF_INET6 || 1323 sa->sa_len != sizeof(struct sockaddr_in6)) 1324 return; 1325 1326 if (cmd == PRC_MSGSIZE) 1327 notify = tcp_mtudisc; 1328 else if (!PRC_IS_REDIRECT(cmd) && 1329 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1330 return; 1331 /* Source quench is depreciated. */ 1332 else if (cmd == PRC_QUENCH) 1333 return; 1334 1335 /* if the parameter is from icmp6, decode it. */ 1336 if (d != NULL) { 1337 ip6cp = (struct ip6ctlparam *)d; 1338 m = ip6cp->ip6c_m; 1339 ip6 = ip6cp->ip6c_ip6; 1340 off = ip6cp->ip6c_off; 1341 sa6_src = ip6cp->ip6c_src; 1342 } else { 1343 m = NULL; 1344 ip6 = NULL; 1345 off = 0; /* fool gcc */ 1346 sa6_src = &sa6_any; 1347 } 1348 1349 if (ip6 != NULL) { 1350 struct in_conninfo inc; 1351 /* 1352 * XXX: We assume that when IPV6 is non NULL, 1353 * M and OFF are valid. 1354 */ 1355 1356 /* check if we can safely examine src and dst ports */ 1357 if (m->m_pkthdr.len < off + sizeof(*thp)) 1358 return; 1359 1360 bzero(&th, sizeof(th)); 1361 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1362 1363 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1364 (struct sockaddr *)ip6cp->ip6c_src, 1365 th.th_sport, cmd, NULL, notify); 1366 1367 inc.inc_fport = th.th_dport; 1368 inc.inc_lport = th.th_sport; 1369 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1370 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1371 inc.inc_isipv6 = 1; 1372 INP_INFO_WLOCK(&tcbinfo); 1373 syncache_unreach(&inc, &th); 1374 INP_INFO_WUNLOCK(&tcbinfo); 1375 } else 1376 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1377 0, cmd, NULL, notify); 1378 } 1379 #endif /* INET6 */ 1380 1381 1382 /* 1383 * Following is where TCP initial sequence number generation occurs. 1384 * 1385 * There are two places where we must use initial sequence numbers: 1386 * 1. In SYN-ACK packets. 1387 * 2. In SYN packets. 1388 * 1389 * All ISNs for SYN-ACK packets are generated by the syncache. See 1390 * tcp_syncache.c for details. 1391 * 1392 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1393 * depends on this property. In addition, these ISNs should be 1394 * unguessable so as to prevent connection hijacking. To satisfy 1395 * the requirements of this situation, the algorithm outlined in 1396 * RFC 1948 is used, with only small modifications. 1397 * 1398 * Implementation details: 1399 * 1400 * Time is based off the system timer, and is corrected so that it 1401 * increases by one megabyte per second. This allows for proper 1402 * recycling on high speed LANs while still leaving over an hour 1403 * before rollover. 1404 * 1405 * As reading the *exact* system time is too expensive to be done 1406 * whenever setting up a TCP connection, we increment the time 1407 * offset in two ways. First, a small random positive increment 1408 * is added to isn_offset for each connection that is set up. 1409 * Second, the function tcp_isn_tick fires once per clock tick 1410 * and increments isn_offset as necessary so that sequence numbers 1411 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1412 * random positive increments serve only to ensure that the same 1413 * exact sequence number is never sent out twice (as could otherwise 1414 * happen when a port is recycled in less than the system tick 1415 * interval.) 1416 * 1417 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1418 * between seeding of isn_secret. This is normally set to zero, 1419 * as reseeding should not be necessary. 1420 * 1421 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1422 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1423 * general, this means holding an exclusive (write) lock. 1424 */ 1425 1426 #define ISN_BYTES_PER_SECOND 1048576 1427 #define ISN_STATIC_INCREMENT 4096 1428 #define ISN_RANDOM_INCREMENT (4096 - 1) 1429 1430 static u_char isn_secret[32]; 1431 static int isn_last_reseed; 1432 static u_int32_t isn_offset, isn_offset_old; 1433 static MD5_CTX isn_ctx; 1434 1435 tcp_seq 1436 tcp_new_isn(struct tcpcb *tp) 1437 { 1438 u_int32_t md5_buffer[4]; 1439 tcp_seq new_isn; 1440 1441 INP_LOCK_ASSERT(tp->t_inpcb); 1442 1443 ISN_LOCK(); 1444 /* Seed if this is the first use, reseed if requested. */ 1445 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1446 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1447 < (u_int)ticks))) { 1448 read_random(&isn_secret, sizeof(isn_secret)); 1449 isn_last_reseed = ticks; 1450 } 1451 1452 /* Compute the md5 hash and return the ISN. */ 1453 MD5Init(&isn_ctx); 1454 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1455 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1456 #ifdef INET6 1457 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1458 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1459 sizeof(struct in6_addr)); 1460 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1461 sizeof(struct in6_addr)); 1462 } else 1463 #endif 1464 { 1465 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1466 sizeof(struct in_addr)); 1467 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1468 sizeof(struct in_addr)); 1469 } 1470 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1471 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1472 new_isn = (tcp_seq) md5_buffer[0]; 1473 isn_offset += ISN_STATIC_INCREMENT + 1474 (arc4random() & ISN_RANDOM_INCREMENT); 1475 new_isn += isn_offset; 1476 ISN_UNLOCK(); 1477 return (new_isn); 1478 } 1479 1480 /* 1481 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1482 * to keep time flowing at a relatively constant rate. If the random 1483 * increments have already pushed us past the projected offset, do nothing. 1484 */ 1485 static void 1486 tcp_isn_tick(void *xtp) 1487 { 1488 u_int32_t projected_offset; 1489 1490 ISN_LOCK(); 1491 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1492 1493 if (projected_offset > isn_offset) 1494 isn_offset = projected_offset; 1495 1496 isn_offset_old = isn_offset; 1497 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1498 ISN_UNLOCK(); 1499 } 1500 1501 /* 1502 * When a specific ICMP unreachable message is received and the 1503 * connection state is SYN-SENT, drop the connection. This behavior 1504 * is controlled by the icmp_may_rst sysctl. 1505 */ 1506 struct inpcb * 1507 tcp_drop_syn_sent(struct inpcb *inp, int errno) 1508 { 1509 struct tcpcb *tp; 1510 1511 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1512 INP_LOCK_ASSERT(inp); 1513 1514 if ((inp->inp_vflag & INP_TIMEWAIT) || 1515 (inp->inp_vflag & INP_DROPPED)) 1516 return (inp); 1517 1518 tp = intotcpcb(inp); 1519 if (tp->t_state != TCPS_SYN_SENT) 1520 return (inp); 1521 1522 tp = tcp_drop(tp, errno); 1523 if (tp != NULL) 1524 return (inp); 1525 else 1526 return (NULL); 1527 } 1528 1529 /* 1530 * When `need fragmentation' ICMP is received, update our idea of the MSS 1531 * based on the new value in the route. Also nudge TCP to send something, 1532 * since we know the packet we just sent was dropped. 1533 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1534 */ 1535 struct inpcb * 1536 tcp_mtudisc(struct inpcb *inp, int errno) 1537 { 1538 struct tcpcb *tp; 1539 struct socket *so = inp->inp_socket; 1540 u_int maxmtu; 1541 u_int romtu; 1542 int mss; 1543 #ifdef INET6 1544 int isipv6; 1545 #endif /* INET6 */ 1546 1547 INP_LOCK_ASSERT(inp); 1548 if ((inp->inp_vflag & INP_TIMEWAIT) || 1549 (inp->inp_vflag & INP_DROPPED)) 1550 return (inp); 1551 1552 tp = intotcpcb(inp); 1553 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1554 1555 #ifdef INET6 1556 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1557 #endif 1558 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1559 romtu = 1560 #ifdef INET6 1561 isipv6 ? tcp_maxmtu6(&inp->inp_inc, NULL) : 1562 #endif /* INET6 */ 1563 tcp_maxmtu(&inp->inp_inc, NULL); 1564 if (!maxmtu) 1565 maxmtu = romtu; 1566 else 1567 maxmtu = min(maxmtu, romtu); 1568 if (!maxmtu) { 1569 tp->t_maxopd = tp->t_maxseg = 1570 #ifdef INET6 1571 isipv6 ? tcp_v6mssdflt : 1572 #endif /* INET6 */ 1573 tcp_mssdflt; 1574 return (inp); 1575 } 1576 mss = maxmtu - 1577 #ifdef INET6 1578 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1579 #endif /* INET6 */ 1580 sizeof(struct tcpiphdr) 1581 #ifdef INET6 1582 ) 1583 #endif /* INET6 */ 1584 ; 1585 1586 /* 1587 * XXX - The above conditional probably violates the TCP 1588 * spec. The problem is that, since we don't know the 1589 * other end's MSS, we are supposed to use a conservative 1590 * default. But, if we do that, then MTU discovery will 1591 * never actually take place, because the conservative 1592 * default is much less than the MTUs typically seen 1593 * on the Internet today. For the moment, we'll sweep 1594 * this under the carpet. 1595 * 1596 * The conservative default might not actually be a problem 1597 * if the only case this occurs is when sending an initial 1598 * SYN with options and data to a host we've never talked 1599 * to before. Then, they will reply with an MSS value which 1600 * will get recorded and the new parameters should get 1601 * recomputed. For Further Study. 1602 */ 1603 if (tp->t_maxopd <= mss) 1604 return (inp); 1605 tp->t_maxopd = mss; 1606 1607 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1608 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1609 mss -= TCPOLEN_TSTAMP_APPA; 1610 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1611 if (mss > MCLBYTES) 1612 mss &= ~(MCLBYTES-1); 1613 #else 1614 if (mss > MCLBYTES) 1615 mss = mss / MCLBYTES * MCLBYTES; 1616 #endif 1617 if (so->so_snd.sb_hiwat < mss) 1618 mss = so->so_snd.sb_hiwat; 1619 1620 tp->t_maxseg = mss; 1621 1622 tcpstat.tcps_mturesent++; 1623 tp->t_rtttime = 0; 1624 tp->snd_nxt = tp->snd_una; 1625 tcp_free_sackholes(tp); 1626 tp->snd_recover = tp->snd_max; 1627 if (tp->sack_enable) 1628 EXIT_FASTRECOVERY(tp); 1629 tcp_output(tp); 1630 return (inp); 1631 } 1632 1633 /* 1634 * Look-up the routing entry to the peer of this inpcb. If no route 1635 * is found and it cannot be allocated, then return NULL. This routine 1636 * is called by TCP routines that access the rmx structure and by tcp_mss 1637 * to get the interface MTU. 1638 */ 1639 u_long 1640 tcp_maxmtu(struct in_conninfo *inc, int *flags) 1641 { 1642 struct route sro; 1643 struct sockaddr_in *dst; 1644 struct ifnet *ifp; 1645 u_long maxmtu = 0; 1646 1647 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1648 1649 bzero(&sro, sizeof(sro)); 1650 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1651 dst = (struct sockaddr_in *)&sro.ro_dst; 1652 dst->sin_family = AF_INET; 1653 dst->sin_len = sizeof(*dst); 1654 dst->sin_addr = inc->inc_faddr; 1655 rtalloc_ign(&sro, RTF_CLONING); 1656 } 1657 if (sro.ro_rt != NULL) { 1658 ifp = sro.ro_rt->rt_ifp; 1659 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1660 maxmtu = ifp->if_mtu; 1661 else 1662 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1663 1664 /* Report additional interface capabilities. */ 1665 if (flags != NULL) { 1666 if (ifp->if_capenable & IFCAP_TSO4 && 1667 ifp->if_hwassist & CSUM_TSO) 1668 *flags |= CSUM_TSO; 1669 } 1670 RTFREE(sro.ro_rt); 1671 } 1672 return (maxmtu); 1673 } 1674 1675 #ifdef INET6 1676 u_long 1677 tcp_maxmtu6(struct in_conninfo *inc, int *flags) 1678 { 1679 struct route_in6 sro6; 1680 struct ifnet *ifp; 1681 u_long maxmtu = 0; 1682 1683 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1684 1685 bzero(&sro6, sizeof(sro6)); 1686 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1687 sro6.ro_dst.sin6_family = AF_INET6; 1688 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1689 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1690 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1691 } 1692 if (sro6.ro_rt != NULL) { 1693 ifp = sro6.ro_rt->rt_ifp; 1694 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1695 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1696 else 1697 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1698 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1699 1700 /* Report additional interface capabilities. */ 1701 if (flags != NULL) { 1702 if (ifp->if_capenable & IFCAP_TSO6 && 1703 ifp->if_hwassist & CSUM_TSO) 1704 *flags |= CSUM_TSO; 1705 } 1706 RTFREE(sro6.ro_rt); 1707 } 1708 1709 return (maxmtu); 1710 } 1711 #endif /* INET6 */ 1712 1713 #ifdef IPSEC 1714 /* compute ESP/AH header size for TCP, including outer IP header. */ 1715 size_t 1716 ipsec_hdrsiz_tcp(struct tcpcb *tp) 1717 { 1718 struct inpcb *inp; 1719 struct mbuf *m; 1720 size_t hdrsiz; 1721 struct ip *ip; 1722 #ifdef INET6 1723 struct ip6_hdr *ip6; 1724 #endif 1725 struct tcphdr *th; 1726 1727 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1728 return (0); 1729 MGETHDR(m, M_DONTWAIT, MT_DATA); 1730 if (!m) 1731 return (0); 1732 1733 #ifdef INET6 1734 if ((inp->inp_vflag & INP_IPV6) != 0) { 1735 ip6 = mtod(m, struct ip6_hdr *); 1736 th = (struct tcphdr *)(ip6 + 1); 1737 m->m_pkthdr.len = m->m_len = 1738 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1739 tcpip_fillheaders(inp, ip6, th); 1740 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1741 } else 1742 #endif /* INET6 */ 1743 { 1744 ip = mtod(m, struct ip *); 1745 th = (struct tcphdr *)(ip + 1); 1746 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1747 tcpip_fillheaders(inp, ip, th); 1748 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1749 } 1750 1751 m_free(m); 1752 return (hdrsiz); 1753 } 1754 #endif /*IPSEC*/ 1755 1756 /* 1757 * Move a TCP connection into TIME_WAIT state. 1758 * tcbinfo is locked. 1759 * inp is locked, and is unlocked before returning. 1760 */ 1761 void 1762 tcp_twstart(struct tcpcb *tp) 1763 { 1764 struct tcptw *tw; 1765 struct inpcb *inp = tp->t_inpcb; 1766 int acknow; 1767 struct socket *so; 1768 1769 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1770 INP_LOCK_ASSERT(inp); 1771 1772 if (nolocaltimewait && in_localip(inp->inp_faddr)) { 1773 tp = tcp_close(tp); 1774 if (tp != NULL) 1775 INP_UNLOCK(inp); 1776 return; 1777 } 1778 1779 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1780 if (tw == NULL) { 1781 tw = tcp_timer_2msl_tw(1); 1782 if (tw == NULL) { 1783 tp = tcp_close(tp); 1784 if (tp != NULL) 1785 INP_UNLOCK(inp); 1786 return; 1787 } 1788 } 1789 tw->tw_inpcb = inp; 1790 1791 /* 1792 * Recover last window size sent. 1793 */ 1794 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1795 1796 /* 1797 * Set t_recent if timestamps are used on the connection. 1798 */ 1799 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1800 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1801 tw->t_recent = tp->ts_recent; 1802 else 1803 tw->t_recent = 0; 1804 1805 tw->snd_nxt = tp->snd_nxt; 1806 tw->rcv_nxt = tp->rcv_nxt; 1807 tw->iss = tp->iss; 1808 tw->irs = tp->irs; 1809 tw->t_starttime = tp->t_starttime; 1810 tw->tw_time = 0; 1811 1812 /* XXX 1813 * If this code will 1814 * be used for fin-wait-2 state also, then we may need 1815 * a ts_recent from the last segment. 1816 */ 1817 acknow = tp->t_flags & TF_ACKNOW; 1818 1819 /* 1820 * First, discard tcpcb state, which includes stopping its timers and 1821 * freeing it. tcp_discardcb() used to also release the inpcb, but 1822 * that work is now done in the caller. 1823 * 1824 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 1825 * and might not be needed here any longer. 1826 */ 1827 tcp_discardcb(tp); 1828 so = inp->inp_socket; 1829 soisdisconnected(so); 1830 SOCK_LOCK(so); 1831 tw->tw_cred = crhold(so->so_cred); 1832 tw->tw_so_options = so->so_options; 1833 SOCK_UNLOCK(so); 1834 if (acknow) 1835 tcp_twrespond(tw, TH_ACK); 1836 inp->inp_ppcb = tw; 1837 inp->inp_vflag |= INP_TIMEWAIT; 1838 tcp_timer_2msl_reset(tw, 0); 1839 1840 /* 1841 * If the inpcb owns the sole reference to the socket, then we can 1842 * detach and free the socket as it is not needed in time wait. 1843 */ 1844 if (inp->inp_vflag & INP_SOCKREF) { 1845 KASSERT(so->so_state & SS_PROTOREF, 1846 ("tcp_twstart: !SS_PROTOREF")); 1847 inp->inp_vflag &= ~INP_SOCKREF; 1848 INP_UNLOCK(inp); 1849 ACCEPT_LOCK(); 1850 SOCK_LOCK(so); 1851 so->so_state &= ~SS_PROTOREF; 1852 sofree(so); 1853 } else 1854 INP_UNLOCK(inp); 1855 } 1856 1857 #if 0 1858 /* 1859 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1860 * the actual rate is slightly higher due to the addition of 1861 * random positive increments. 1862 * 1863 * Most other new OSes use semi-randomized ISN values, so we 1864 * do not need to worry about them. 1865 */ 1866 #define MS_ISN_BYTES_PER_SECOND 250000 1867 1868 /* 1869 * Determine if the ISN we will generate has advanced beyond the last 1870 * sequence number used by the previous connection. If so, indicate 1871 * that it is safe to recycle this tw socket by returning 1. 1872 */ 1873 int 1874 tcp_twrecycleable(struct tcptw *tw) 1875 { 1876 tcp_seq new_iss = tw->iss; 1877 tcp_seq new_irs = tw->irs; 1878 1879 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1880 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1881 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1882 1883 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1884 return (1); 1885 else 1886 return (0); 1887 } 1888 #endif 1889 1890 void 1891 tcp_twclose(struct tcptw *tw, int reuse) 1892 { 1893 struct socket *so; 1894 struct inpcb *inp; 1895 1896 /* 1897 * At this point, we are in one of two situations: 1898 * 1899 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 1900 * all state. 1901 * 1902 * (2) We have a socket -- if we own a reference, release it and 1903 * notify the socket layer. 1904 */ 1905 inp = tw->tw_inpcb; 1906 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1907 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 1908 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1909 INP_LOCK_ASSERT(inp); 1910 1911 tw->tw_inpcb = NULL; 1912 tcp_timer_2msl_stop(tw); 1913 inp->inp_ppcb = NULL; 1914 in_pcbdrop(inp); 1915 1916 so = inp->inp_socket; 1917 if (so != NULL) { 1918 /* 1919 * If there's a socket, handle two cases: first, we own a 1920 * strong reference, which we will now release, or we don't 1921 * in which case another reference exists (XXXRW: think 1922 * about this more), and we don't need to take action. 1923 */ 1924 if (inp->inp_vflag & INP_SOCKREF) { 1925 inp->inp_vflag &= ~INP_SOCKREF; 1926 INP_UNLOCK(inp); 1927 ACCEPT_LOCK(); 1928 SOCK_LOCK(so); 1929 KASSERT(so->so_state & SS_PROTOREF, 1930 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1931 so->so_state &= ~SS_PROTOREF; 1932 sofree(so); 1933 } else { 1934 /* 1935 * If we don't own the only reference, the socket and 1936 * inpcb need to be left around to be handled by 1937 * tcp_usr_detach() later. 1938 */ 1939 INP_UNLOCK(inp); 1940 } 1941 } else { 1942 #ifdef INET6 1943 if (inp->inp_vflag & INP_IPV6PROTO) 1944 in6_pcbfree(inp); 1945 else 1946 #endif 1947 in_pcbfree(inp); 1948 } 1949 tcpstat.tcps_closed++; 1950 crfree(tw->tw_cred); 1951 tw->tw_cred = NULL; 1952 if (reuse) 1953 return; 1954 uma_zfree(tcptw_zone, tw); 1955 } 1956 1957 int 1958 tcp_twrespond(struct tcptw *tw, int flags) 1959 { 1960 struct inpcb *inp = tw->tw_inpcb; 1961 struct tcphdr *th; 1962 struct mbuf *m; 1963 struct ip *ip = NULL; 1964 u_int8_t *optp; 1965 u_int hdrlen, optlen; 1966 int error; 1967 #ifdef INET6 1968 struct ip6_hdr *ip6 = NULL; 1969 int isipv6 = inp->inp_inc.inc_isipv6; 1970 #endif 1971 1972 INP_LOCK_ASSERT(inp); 1973 1974 m = m_gethdr(M_DONTWAIT, MT_DATA); 1975 if (m == NULL) 1976 return (ENOBUFS); 1977 m->m_data += max_linkhdr; 1978 1979 #ifdef MAC 1980 mac_create_mbuf_from_inpcb(inp, m); 1981 #endif 1982 1983 #ifdef INET6 1984 if (isipv6) { 1985 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1986 ip6 = mtod(m, struct ip6_hdr *); 1987 th = (struct tcphdr *)(ip6 + 1); 1988 tcpip_fillheaders(inp, ip6, th); 1989 } else 1990 #endif 1991 { 1992 hdrlen = sizeof(struct tcpiphdr); 1993 ip = mtod(m, struct ip *); 1994 th = (struct tcphdr *)(ip + 1); 1995 tcpip_fillheaders(inp, ip, th); 1996 } 1997 optp = (u_int8_t *)(th + 1); 1998 1999 /* 2000 * Send a timestamp and echo-reply if both our side and our peer 2001 * have sent timestamps in our SYN's and this is not a RST. 2002 */ 2003 if (tw->t_recent && flags == TH_ACK) { 2004 u_int32_t *lp = (u_int32_t *)optp; 2005 2006 /* Form timestamp option as shown in appendix A of RFC 1323. */ 2007 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 2008 *lp++ = htonl(ticks); 2009 *lp = htonl(tw->t_recent); 2010 optp += TCPOLEN_TSTAMP_APPA; 2011 } 2012 2013 optlen = optp - (u_int8_t *)(th + 1); 2014 2015 m->m_len = hdrlen + optlen; 2016 m->m_pkthdr.len = m->m_len; 2017 2018 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 2019 2020 th->th_seq = htonl(tw->snd_nxt); 2021 th->th_ack = htonl(tw->rcv_nxt); 2022 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 2023 th->th_flags = flags; 2024 th->th_win = htons(tw->last_win); 2025 2026 #ifdef INET6 2027 if (isipv6) { 2028 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 2029 sizeof(struct tcphdr) + optlen); 2030 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2031 error = ip6_output(m, inp->in6p_outputopts, NULL, 2032 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 2033 } else 2034 #endif 2035 { 2036 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2037 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 2038 m->m_pkthdr.csum_flags = CSUM_TCP; 2039 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2040 ip->ip_len = m->m_pkthdr.len; 2041 if (path_mtu_discovery) 2042 ip->ip_off |= IP_DF; 2043 error = ip_output(m, inp->inp_options, NULL, 2044 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 2045 NULL, inp); 2046 } 2047 if (flags & TH_ACK) 2048 tcpstat.tcps_sndacks++; 2049 else 2050 tcpstat.tcps_sndctrl++; 2051 tcpstat.tcps_sndtotal++; 2052 return (error); 2053 } 2054 2055 /* 2056 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 2057 * 2058 * This code attempts to calculate the bandwidth-delay product as a 2059 * means of determining the optimal window size to maximize bandwidth, 2060 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 2061 * routers. This code also does a fairly good job keeping RTTs in check 2062 * across slow links like modems. We implement an algorithm which is very 2063 * similar (but not meant to be) TCP/Vegas. The code operates on the 2064 * transmitter side of a TCP connection and so only effects the transmit 2065 * side of the connection. 2066 * 2067 * BACKGROUND: TCP makes no provision for the management of buffer space 2068 * at the end points or at the intermediate routers and switches. A TCP 2069 * stream, whether using NewReno or not, will eventually buffer as 2070 * many packets as it is able and the only reason this typically works is 2071 * due to the fairly small default buffers made available for a connection 2072 * (typicaly 16K or 32K). As machines use larger windows and/or window 2073 * scaling it is now fairly easy for even a single TCP connection to blow-out 2074 * all available buffer space not only on the local interface, but on 2075 * intermediate routers and switches as well. NewReno makes a misguided 2076 * attempt to 'solve' this problem by waiting for an actual failure to occur, 2077 * then backing off, then steadily increasing the window again until another 2078 * failure occurs, ad-infinitum. This results in terrible oscillation that 2079 * is only made worse as network loads increase and the idea of intentionally 2080 * blowing out network buffers is, frankly, a terrible way to manage network 2081 * resources. 2082 * 2083 * It is far better to limit the transmit window prior to the failure 2084 * condition being achieved. There are two general ways to do this: First 2085 * you can 'scan' through different transmit window sizes and locate the 2086 * point where the RTT stops increasing, indicating that you have filled the 2087 * pipe, then scan backwards until you note that RTT stops decreasing, then 2088 * repeat ad-infinitum. This method works in principle but has severe 2089 * implementation issues due to RTT variances, timer granularity, and 2090 * instability in the algorithm which can lead to many false positives and 2091 * create oscillations as well as interact badly with other TCP streams 2092 * implementing the same algorithm. 2093 * 2094 * The second method is to limit the window to the bandwidth delay product 2095 * of the link. This is the method we implement. RTT variances and our 2096 * own manipulation of the congestion window, bwnd, can potentially 2097 * destabilize the algorithm. For this reason we have to stabilize the 2098 * elements used to calculate the window. We do this by using the minimum 2099 * observed RTT, the long term average of the observed bandwidth, and 2100 * by adding two segments worth of slop. It isn't perfect but it is able 2101 * to react to changing conditions and gives us a very stable basis on 2102 * which to extend the algorithm. 2103 */ 2104 void 2105 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2106 { 2107 u_long bw; 2108 u_long bwnd; 2109 int save_ticks; 2110 2111 INP_LOCK_ASSERT(tp->t_inpcb); 2112 2113 /* 2114 * If inflight_enable is disabled in the middle of a tcp connection, 2115 * make sure snd_bwnd is effectively disabled. 2116 */ 2117 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2118 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2119 tp->snd_bandwidth = 0; 2120 return; 2121 } 2122 2123 /* 2124 * Figure out the bandwidth. Due to the tick granularity this 2125 * is a very rough number and it MUST be averaged over a fairly 2126 * long period of time. XXX we need to take into account a link 2127 * that is not using all available bandwidth, but for now our 2128 * slop will ramp us up if this case occurs and the bandwidth later 2129 * increases. 2130 * 2131 * Note: if ticks rollover 'bw' may wind up negative. We must 2132 * effectively reset t_bw_rtttime for this case. 2133 */ 2134 save_ticks = ticks; 2135 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2136 return; 2137 2138 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2139 (save_ticks - tp->t_bw_rtttime); 2140 tp->t_bw_rtttime = save_ticks; 2141 tp->t_bw_rtseq = ack_seq; 2142 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2143 return; 2144 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2145 2146 tp->snd_bandwidth = bw; 2147 2148 /* 2149 * Calculate the semi-static bandwidth delay product, plus two maximal 2150 * segments. The additional slop puts us squarely in the sweet 2151 * spot and also handles the bandwidth run-up case and stabilization. 2152 * Without the slop we could be locking ourselves into a lower 2153 * bandwidth. 2154 * 2155 * Situations Handled: 2156 * (1) Prevents over-queueing of packets on LANs, especially on 2157 * high speed LANs, allowing larger TCP buffers to be 2158 * specified, and also does a good job preventing 2159 * over-queueing of packets over choke points like modems 2160 * (at least for the transmit side). 2161 * 2162 * (2) Is able to handle changing network loads (bandwidth 2163 * drops so bwnd drops, bandwidth increases so bwnd 2164 * increases). 2165 * 2166 * (3) Theoretically should stabilize in the face of multiple 2167 * connections implementing the same algorithm (this may need 2168 * a little work). 2169 * 2170 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2171 * be adjusted with a sysctl but typically only needs to be 2172 * on very slow connections. A value no smaller then 5 2173 * should be used, but only reduce this default if you have 2174 * no other choice. 2175 */ 2176 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2177 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2178 #undef USERTT 2179 2180 if (tcp_inflight_debug > 0) { 2181 static int ltime; 2182 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2183 ltime = ticks; 2184 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2185 tp, 2186 bw, 2187 tp->t_rttbest, 2188 tp->t_srtt, 2189 bwnd 2190 ); 2191 } 2192 } 2193 if ((long)bwnd < tcp_inflight_min) 2194 bwnd = tcp_inflight_min; 2195 if (bwnd > tcp_inflight_max) 2196 bwnd = tcp_inflight_max; 2197 if ((long)bwnd < tp->t_maxseg * 2) 2198 bwnd = tp->t_maxseg * 2; 2199 tp->snd_bwnd = bwnd; 2200 } 2201 2202 #ifdef TCP_SIGNATURE 2203 /* 2204 * Callback function invoked by m_apply() to digest TCP segment data 2205 * contained within an mbuf chain. 2206 */ 2207 static int 2208 tcp_signature_apply(void *fstate, void *data, u_int len) 2209 { 2210 2211 MD5Update(fstate, (u_char *)data, len); 2212 return (0); 2213 } 2214 2215 /* 2216 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2217 * 2218 * Parameters: 2219 * m pointer to head of mbuf chain 2220 * off0 offset to TCP header within the mbuf chain 2221 * len length of TCP segment data, excluding options 2222 * optlen length of TCP segment options 2223 * buf pointer to storage for computed MD5 digest 2224 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2225 * 2226 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2227 * When called from tcp_input(), we can be sure that th_sum has been 2228 * zeroed out and verified already. 2229 * 2230 * This function is for IPv4 use only. Calling this function with an 2231 * IPv6 packet in the mbuf chain will yield undefined results. 2232 * 2233 * Return 0 if successful, otherwise return -1. 2234 * 2235 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2236 * search with the destination IP address, and a 'magic SPI' to be 2237 * determined by the application. This is hardcoded elsewhere to 1179 2238 * right now. Another branch of this code exists which uses the SPD to 2239 * specify per-application flows but it is unstable. 2240 */ 2241 int 2242 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2243 u_char *buf, u_int direction) 2244 { 2245 union sockaddr_union dst; 2246 struct ippseudo ippseudo; 2247 MD5_CTX ctx; 2248 int doff; 2249 struct ip *ip; 2250 struct ipovly *ipovly; 2251 struct secasvar *sav; 2252 struct tcphdr *th; 2253 u_short savecsum; 2254 2255 KASSERT(m != NULL, ("NULL mbuf chain")); 2256 KASSERT(buf != NULL, ("NULL signature pointer")); 2257 2258 /* Extract the destination from the IP header in the mbuf. */ 2259 ip = mtod(m, struct ip *); 2260 bzero(&dst, sizeof(union sockaddr_union)); 2261 dst.sa.sa_len = sizeof(struct sockaddr_in); 2262 dst.sa.sa_family = AF_INET; 2263 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2264 ip->ip_src : ip->ip_dst; 2265 2266 /* Look up an SADB entry which matches the address of the peer. */ 2267 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2268 if (sav == NULL) { 2269 printf("%s: SADB lookup failed for %s\n", __func__, 2270 inet_ntoa(dst.sin.sin_addr)); 2271 return (EINVAL); 2272 } 2273 2274 MD5Init(&ctx); 2275 ipovly = (struct ipovly *)ip; 2276 th = (struct tcphdr *)((u_char *)ip + off0); 2277 doff = off0 + sizeof(struct tcphdr) + optlen; 2278 2279 /* 2280 * Step 1: Update MD5 hash with IP pseudo-header. 2281 * 2282 * XXX The ippseudo header MUST be digested in network byte order, 2283 * or else we'll fail the regression test. Assume all fields we've 2284 * been doing arithmetic on have been in host byte order. 2285 * XXX One cannot depend on ipovly->ih_len here. When called from 2286 * tcp_output(), the underlying ip_len member has not yet been set. 2287 */ 2288 ippseudo.ippseudo_src = ipovly->ih_src; 2289 ippseudo.ippseudo_dst = ipovly->ih_dst; 2290 ippseudo.ippseudo_pad = 0; 2291 ippseudo.ippseudo_p = IPPROTO_TCP; 2292 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2293 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2294 2295 /* 2296 * Step 2: Update MD5 hash with TCP header, excluding options. 2297 * The TCP checksum must be set to zero. 2298 */ 2299 savecsum = th->th_sum; 2300 th->th_sum = 0; 2301 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2302 th->th_sum = savecsum; 2303 2304 /* 2305 * Step 3: Update MD5 hash with TCP segment data. 2306 * Use m_apply() to avoid an early m_pullup(). 2307 */ 2308 if (len > 0) 2309 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2310 2311 /* 2312 * Step 4: Update MD5 hash with shared secret. 2313 */ 2314 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2315 MD5Final(buf, &ctx); 2316 2317 key_sa_recordxfer(sav, m); 2318 KEY_FREESAV(&sav); 2319 return (0); 2320 } 2321 #endif /* TCP_SIGNATURE */ 2322 2323 static int 2324 sysctl_drop(SYSCTL_HANDLER_ARGS) 2325 { 2326 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2327 struct sockaddr_storage addrs[2]; 2328 struct inpcb *inp; 2329 struct tcpcb *tp; 2330 struct tcptw *tw; 2331 struct sockaddr_in *fin, *lin; 2332 #ifdef INET6 2333 struct sockaddr_in6 *fin6, *lin6; 2334 struct in6_addr f6, l6; 2335 #endif 2336 int error; 2337 2338 inp = NULL; 2339 fin = lin = NULL; 2340 #ifdef INET6 2341 fin6 = lin6 = NULL; 2342 #endif 2343 error = 0; 2344 2345 if (req->oldptr != NULL || req->oldlen != 0) 2346 return (EINVAL); 2347 if (req->newptr == NULL) 2348 return (EPERM); 2349 if (req->newlen < sizeof(addrs)) 2350 return (ENOMEM); 2351 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2352 if (error) 2353 return (error); 2354 2355 switch (addrs[0].ss_family) { 2356 #ifdef INET6 2357 case AF_INET6: 2358 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2359 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2360 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2361 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2362 return (EINVAL); 2363 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2364 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2365 return (EINVAL); 2366 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2367 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2368 fin = (struct sockaddr_in *)&addrs[0]; 2369 lin = (struct sockaddr_in *)&addrs[1]; 2370 break; 2371 } 2372 error = sa6_embedscope(fin6, ip6_use_defzone); 2373 if (error) 2374 return (error); 2375 error = sa6_embedscope(lin6, ip6_use_defzone); 2376 if (error) 2377 return (error); 2378 break; 2379 #endif 2380 case AF_INET: 2381 fin = (struct sockaddr_in *)&addrs[0]; 2382 lin = (struct sockaddr_in *)&addrs[1]; 2383 if (fin->sin_len != sizeof(struct sockaddr_in) || 2384 lin->sin_len != sizeof(struct sockaddr_in)) 2385 return (EINVAL); 2386 break; 2387 default: 2388 return (EINVAL); 2389 } 2390 INP_INFO_WLOCK(&tcbinfo); 2391 switch (addrs[0].ss_family) { 2392 #ifdef INET6 2393 case AF_INET6: 2394 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2395 &l6, lin6->sin6_port, 0, NULL); 2396 break; 2397 #endif 2398 case AF_INET: 2399 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2400 lin->sin_addr, lin->sin_port, 0, NULL); 2401 break; 2402 } 2403 if (inp != NULL) { 2404 INP_LOCK(inp); 2405 if (inp->inp_vflag & INP_TIMEWAIT) { 2406 /* 2407 * XXXRW: There currently exists a state where an 2408 * inpcb is present, but its timewait state has been 2409 * discarded. For now, don't allow dropping of this 2410 * type of inpcb. 2411 */ 2412 tw = intotw(inp); 2413 if (tw != NULL) 2414 tcp_twclose(tw, 0); 2415 } else if (!(inp->inp_vflag & INP_DROPPED) && 2416 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2417 tp = intotcpcb(inp); 2418 tcp_drop(tp, ECONNABORTED); 2419 } 2420 INP_UNLOCK(inp); 2421 } else 2422 error = ESRCH; 2423 INP_INFO_WUNLOCK(&tcbinfo); 2424 return (error); 2425 } 2426 2427 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2428 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2429 0, sysctl_drop, "", "Drop TCP connection"); 2430