tcp_subr.c revision 285980
1/*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 30 */ 31 32#include <sys/cdefs.h> 33__FBSDID("$FreeBSD: releng/9.3/sys/netinet/tcp_subr.c 285980 2015-07-28 19:59:22Z delphij $"); 34 35#include "opt_compat.h" 36#include "opt_inet.h" 37#include "opt_inet6.h" 38#include "opt_ipsec.h" 39#include "opt_tcpdebug.h" 40 41#include <sys/param.h> 42#include <sys/systm.h> 43#include <sys/callout.h> 44#include <sys/hhook.h> 45#include <sys/kernel.h> 46#include <sys/khelp.h> 47#include <sys/sysctl.h> 48#include <sys/jail.h> 49#include <sys/malloc.h> 50#include <sys/mbuf.h> 51#ifdef INET6 52#include <sys/domain.h> 53#endif 54#include <sys/priv.h> 55#include <sys/proc.h> 56#include <sys/socket.h> 57#include <sys/socketvar.h> 58#include <sys/protosw.h> 59#include <sys/random.h> 60 61#include <vm/uma.h> 62 63#include <net/route.h> 64#include <net/if.h> 65#include <net/vnet.h> 66 67#include <netinet/cc.h> 68#include <netinet/in.h> 69#include <netinet/in_pcb.h> 70#include <netinet/in_systm.h> 71#include <netinet/in_var.h> 72#include <netinet/ip.h> 73#include <netinet/ip_icmp.h> 74#include <netinet/ip_var.h> 75#ifdef INET6 76#include <netinet/ip6.h> 77#include <netinet6/in6_pcb.h> 78#include <netinet6/ip6_var.h> 79#include <netinet6/scope6_var.h> 80#include <netinet6/nd6.h> 81#endif 82 83#include <netinet/tcp_fsm.h> 84#include <netinet/tcp_seq.h> 85#include <netinet/tcp_timer.h> 86#include <netinet/tcp_var.h> 87#include <netinet/tcp_syncache.h> 88#ifdef INET6 89#include <netinet6/tcp6_var.h> 90#endif 91#include <netinet/tcpip.h> 92#ifdef TCPDEBUG 93#include <netinet/tcp_debug.h> 94#endif 95#ifdef INET6 96#include <netinet6/ip6protosw.h> 97#endif 98#ifdef TCP_OFFLOAD 99#include <netinet/tcp_offload.h> 100#endif 101 102#ifdef IPSEC 103#include <netipsec/ipsec.h> 104#include <netipsec/xform.h> 105#ifdef INET6 106#include <netipsec/ipsec6.h> 107#endif 108#include <netipsec/key.h> 109#include <sys/syslog.h> 110#endif /*IPSEC*/ 111 112#include <machine/in_cksum.h> 113#include <sys/md5.h> 114 115#include <security/mac/mac_framework.h> 116 117VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 118#ifdef INET6 119VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 120#endif 121 122static int 123sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 124{ 125 int error, new; 126 127 new = V_tcp_mssdflt; 128 error = sysctl_handle_int(oidp, &new, 0, req); 129 if (error == 0 && req->newptr) { 130 if (new < TCP_MINMSS) 131 error = EINVAL; 132 else 133 V_tcp_mssdflt = new; 134 } 135 return (error); 136} 137 138SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 139 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0, 140 &sysctl_net_inet_tcp_mss_check, "I", 141 "Default TCP Maximum Segment Size"); 142 143#ifdef INET6 144static int 145sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 146{ 147 int error, new; 148 149 new = V_tcp_v6mssdflt; 150 error = sysctl_handle_int(oidp, &new, 0, req); 151 if (error == 0 && req->newptr) { 152 if (new < TCP_MINMSS) 153 error = EINVAL; 154 else 155 V_tcp_v6mssdflt = new; 156 } 157 return (error); 158} 159 160SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 161 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0, 162 &sysctl_net_inet_tcp_mss_v6_check, "I", 163 "Default TCP Maximum Segment Size for IPv6"); 164#endif /* INET6 */ 165 166/* 167 * Minimum MSS we accept and use. This prevents DoS attacks where 168 * we are forced to a ridiculous low MSS like 20 and send hundreds 169 * of packets instead of one. The effect scales with the available 170 * bandwidth and quickly saturates the CPU and network interface 171 * with packet generation and sending. Set to zero to disable MINMSS 172 * checking. This setting prevents us from sending too small packets. 173 */ 174VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 175SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 176 &VNET_NAME(tcp_minmss), 0, 177 "Minimum TCP Maximum Segment Size"); 178 179VNET_DEFINE(int, tcp_do_rfc1323) = 1; 180SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 181 &VNET_NAME(tcp_do_rfc1323), 0, 182 "Enable rfc1323 (high performance TCP) extensions"); 183 184static int tcp_log_debug = 0; 185SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 186 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 187 188static int tcp_tcbhashsize = 0; 189SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 190 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 191 192static int do_tcpdrain = 1; 193SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 194 "Enable tcp_drain routine for extra help when low on mbufs"); 195 196SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 197 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 198 199static VNET_DEFINE(int, icmp_may_rst) = 1; 200#define V_icmp_may_rst VNET(icmp_may_rst) 201SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, 202 &VNET_NAME(icmp_may_rst), 0, 203 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 204 205static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0; 206#define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 207SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 208 &VNET_NAME(tcp_isn_reseed_interval), 0, 209 "Seconds between reseeding of ISN secret"); 210 211static int tcp_soreceive_stream = 0; 212SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 213 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 214 215#ifdef TCP_SIGNATURE 216static int tcp_sig_checksigs = 1; 217SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW, 218 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic"); 219#endif 220 221VNET_DEFINE(uma_zone_t, sack_hole_zone); 222#define V_sack_hole_zone VNET(sack_hole_zone) 223 224VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 225 226static struct inpcb *tcp_notify(struct inpcb *, int); 227static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 228static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 229 void *ip4hdr, const void *ip6hdr); 230 231/* 232 * Target size of TCP PCB hash tables. Must be a power of two. 233 * 234 * Note that this can be overridden by the kernel environment 235 * variable net.inet.tcp.tcbhashsize 236 */ 237#ifndef TCBHASHSIZE 238#define TCBHASHSIZE 512 239#endif 240 241/* 242 * XXX 243 * Callouts should be moved into struct tcp directly. They are currently 244 * separate because the tcpcb structure is exported to userland for sysctl 245 * parsing purposes, which do not know about callouts. 246 */ 247struct tcpcb_mem { 248 struct tcpcb tcb; 249 struct tcp_timer tt; 250 struct cc_var ccv; 251 struct osd osd; 252}; 253 254static VNET_DEFINE(uma_zone_t, tcpcb_zone); 255#define V_tcpcb_zone VNET(tcpcb_zone) 256 257MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 258static struct mtx isn_mtx; 259 260#define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 261#define ISN_LOCK() mtx_lock(&isn_mtx) 262#define ISN_UNLOCK() mtx_unlock(&isn_mtx) 263 264/* 265 * TCP initialization. 266 */ 267static void 268tcp_zone_change(void *tag) 269{ 270 271 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 272 uma_zone_set_max(V_tcpcb_zone, maxsockets); 273 tcp_tw_zone_change(); 274} 275 276static int 277tcp_inpcb_init(void *mem, int size, int flags) 278{ 279 struct inpcb *inp = mem; 280 281 INP_LOCK_INIT(inp, "inp", "tcpinp"); 282 return (0); 283} 284 285void 286tcp_init(void) 287{ 288 int hashsize; 289 290 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 291 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 292 printf("%s: WARNING: unable to register helper hook\n", __func__); 293 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 294 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 295 printf("%s: WARNING: unable to register helper hook\n", __func__); 296 297 hashsize = TCBHASHSIZE; 298 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 299 if (!powerof2(hashsize)) { 300 printf("WARNING: TCB hash size not a power of 2\n"); 301 hashsize = 512; /* safe default */ 302 } 303 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 304 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE, 305 IPI_HASHFIELDS_4TUPLE); 306 307 /* 308 * These have to be type stable for the benefit of the timers. 309 */ 310 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 311 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 312 uma_zone_set_max(V_tcpcb_zone, maxsockets); 313 314 tcp_tw_init(); 315 syncache_init(); 316 tcp_hc_init(); 317 318 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 319 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 320 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 321 322 /* Skip initialization of globals for non-default instances. */ 323 if (!IS_DEFAULT_VNET(curvnet)) 324 return; 325 326 tcp_reass_global_init(); 327 328 /* XXX virtualize those bellow? */ 329 tcp_delacktime = TCPTV_DELACK; 330 tcp_keepinit = TCPTV_KEEP_INIT; 331 tcp_keepidle = TCPTV_KEEP_IDLE; 332 tcp_keepintvl = TCPTV_KEEPINTVL; 333 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 334 tcp_msl = TCPTV_MSL; 335 tcp_rexmit_min = TCPTV_MIN; 336 if (tcp_rexmit_min < 1) 337 tcp_rexmit_min = 1; 338 tcp_rexmit_slop = TCPTV_CPU_VAR; 339 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 340 tcp_tcbhashsize = hashsize; 341 342 TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream); 343 if (tcp_soreceive_stream) { 344#ifdef INET 345 tcp_usrreqs.pru_soreceive = soreceive_stream; 346#endif 347#ifdef INET6 348 tcp6_usrreqs.pru_soreceive = soreceive_stream; 349#endif /* INET6 */ 350 } 351 352#ifdef INET6 353#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 354#else /* INET6 */ 355#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 356#endif /* INET6 */ 357 if (max_protohdr < TCP_MINPROTOHDR) 358 max_protohdr = TCP_MINPROTOHDR; 359 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 360 panic("tcp_init"); 361#undef TCP_MINPROTOHDR 362 363 ISN_LOCK_INIT(); 364 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 365 SHUTDOWN_PRI_DEFAULT); 366 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 367 EVENTHANDLER_PRI_ANY); 368} 369 370#ifdef VIMAGE 371void 372tcp_destroy(void) 373{ 374 375 tcp_hc_destroy(); 376 syncache_destroy(); 377 tcp_tw_destroy(); 378 in_pcbinfo_destroy(&V_tcbinfo); 379 uma_zdestroy(V_sack_hole_zone); 380 uma_zdestroy(V_tcpcb_zone); 381} 382#endif 383 384void 385tcp_fini(void *xtp) 386{ 387 388} 389 390/* 391 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 392 * tcp_template used to store this data in mbufs, but we now recopy it out 393 * of the tcpcb each time to conserve mbufs. 394 */ 395void 396tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 397{ 398 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 399 400 INP_WLOCK_ASSERT(inp); 401 402#ifdef INET6 403 if ((inp->inp_vflag & INP_IPV6) != 0) { 404 struct ip6_hdr *ip6; 405 406 ip6 = (struct ip6_hdr *)ip_ptr; 407 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 408 (inp->inp_flow & IPV6_FLOWINFO_MASK); 409 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 410 (IPV6_VERSION & IPV6_VERSION_MASK); 411 ip6->ip6_nxt = IPPROTO_TCP; 412 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 413 ip6->ip6_src = inp->in6p_laddr; 414 ip6->ip6_dst = inp->in6p_faddr; 415 } 416#endif /* INET6 */ 417#if defined(INET6) && defined(INET) 418 else 419#endif 420#ifdef INET 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#endif /* INET */ 438 th->th_sport = inp->inp_lport; 439 th->th_dport = inp->inp_fport; 440 th->th_seq = 0; 441 th->th_ack = 0; 442 th->th_x2 = 0; 443 th->th_off = 5; 444 th->th_flags = 0; 445 th->th_win = 0; 446 th->th_urp = 0; 447 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 448} 449 450/* 451 * Create template to be used to send tcp packets on a connection. 452 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 453 * use for this function is in keepalives, which use tcp_respond. 454 */ 455struct tcptemp * 456tcpip_maketemplate(struct inpcb *inp) 457{ 458 struct tcptemp *t; 459 460 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 461 if (t == NULL) 462 return (NULL); 463 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 464 return (t); 465} 466 467/* 468 * Send a single message to the TCP at address specified by 469 * the given TCP/IP header. If m == NULL, then we make a copy 470 * of the tcpiphdr at ti and send directly to the addressed host. 471 * This is used to force keep alive messages out using the TCP 472 * template for a connection. If flags are given then we send 473 * a message back to the TCP which originated the * segment ti, 474 * and discard the mbuf containing it and any other attached mbufs. 475 * 476 * In any case the ack and sequence number of the transmitted 477 * segment are as specified by the parameters. 478 * 479 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 480 */ 481void 482tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 483 tcp_seq ack, tcp_seq seq, int flags) 484{ 485 int tlen; 486 int win = 0; 487 struct ip *ip; 488 struct tcphdr *nth; 489#ifdef INET6 490 struct ip6_hdr *ip6; 491 int isipv6; 492#endif /* INET6 */ 493 int ipflags = 0; 494 struct inpcb *inp; 495 496 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 497 498#ifdef INET6 499 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 500 ip6 = ipgen; 501#endif /* INET6 */ 502 ip = ipgen; 503 504 if (tp != NULL) { 505 inp = tp->t_inpcb; 506 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 507 INP_WLOCK_ASSERT(inp); 508 } else 509 inp = NULL; 510 511 if (tp != NULL) { 512 if (!(flags & TH_RST)) { 513 win = sbspace(&inp->inp_socket->so_rcv); 514 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 515 win = (long)TCP_MAXWIN << tp->rcv_scale; 516 } 517 } 518 if (m == NULL) { 519 m = m_gethdr(M_DONTWAIT, MT_DATA); 520 if (m == NULL) 521 return; 522 tlen = 0; 523 m->m_data += max_linkhdr; 524#ifdef INET6 525 if (isipv6) { 526 bcopy((caddr_t)ip6, mtod(m, caddr_t), 527 sizeof(struct ip6_hdr)); 528 ip6 = mtod(m, struct ip6_hdr *); 529 nth = (struct tcphdr *)(ip6 + 1); 530 } else 531#endif /* INET6 */ 532 { 533 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 534 ip = mtod(m, struct ip *); 535 nth = (struct tcphdr *)(ip + 1); 536 } 537 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 538 flags = TH_ACK; 539 } else { 540 /* 541 * reuse the mbuf. 542 * XXX MRT We inherrit the FIB, which is lucky. 543 */ 544 m_freem(m->m_next); 545 m->m_next = NULL; 546 m->m_data = (caddr_t)ipgen; 547 m_addr_changed(m); 548 /* m_len is set later */ 549 tlen = 0; 550#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 551#ifdef INET6 552 if (isipv6) { 553 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 554 nth = (struct tcphdr *)(ip6 + 1); 555 } else 556#endif /* INET6 */ 557 { 558 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 559 nth = (struct tcphdr *)(ip + 1); 560 } 561 if (th != nth) { 562 /* 563 * this is usually a case when an extension header 564 * exists between the IPv6 header and the 565 * TCP header. 566 */ 567 nth->th_sport = th->th_sport; 568 nth->th_dport = th->th_dport; 569 } 570 xchg(nth->th_dport, nth->th_sport, uint16_t); 571#undef xchg 572 } 573#ifdef INET6 574 if (isipv6) { 575 ip6->ip6_flow = 0; 576 ip6->ip6_vfc = IPV6_VERSION; 577 ip6->ip6_nxt = IPPROTO_TCP; 578 ip6->ip6_plen = 0; /* Set in ip6_output(). */ 579 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 580 } 581#endif 582#if defined(INET) && defined(INET6) 583 else 584#endif 585#ifdef INET 586 { 587 tlen += sizeof (struct tcpiphdr); 588 ip->ip_len = tlen; 589 ip->ip_ttl = V_ip_defttl; 590 if (V_path_mtu_discovery) 591 ip->ip_off |= IP_DF; 592 } 593#endif 594 m->m_len = tlen; 595 m->m_pkthdr.len = tlen; 596 m->m_pkthdr.rcvif = NULL; 597#ifdef MAC 598 if (inp != NULL) { 599 /* 600 * Packet is associated with a socket, so allow the 601 * label of the response to reflect the socket label. 602 */ 603 INP_WLOCK_ASSERT(inp); 604 mac_inpcb_create_mbuf(inp, m); 605 } else { 606 /* 607 * Packet is not associated with a socket, so possibly 608 * update the label in place. 609 */ 610 mac_netinet_tcp_reply(m); 611 } 612#endif 613 nth->th_seq = htonl(seq); 614 nth->th_ack = htonl(ack); 615 nth->th_x2 = 0; 616 nth->th_off = sizeof (struct tcphdr) >> 2; 617 nth->th_flags = flags; 618 if (tp != NULL) 619 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 620 else 621 nth->th_win = htons((u_short)win); 622 nth->th_urp = 0; 623 624 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 625#ifdef INET6 626 if (isipv6) { 627 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 628 nth->th_sum = in6_cksum_pseudo(ip6, 629 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 630 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 631 NULL, NULL); 632 } 633#endif /* INET6 */ 634#if defined(INET6) && defined(INET) 635 else 636#endif 637#ifdef INET 638 { 639 m->m_pkthdr.csum_flags = CSUM_TCP; 640 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 641 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 642 } 643#endif /* INET */ 644#ifdef TCPDEBUG 645 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 646 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 647#endif 648#ifdef INET6 649 if (isipv6) 650 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 651#endif /* INET6 */ 652#if defined(INET) && defined(INET6) 653 else 654#endif 655#ifdef INET 656 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 657#endif 658} 659 660/* 661 * Create a new TCP control block, making an 662 * empty reassembly queue and hooking it to the argument 663 * protocol control block. The `inp' parameter must have 664 * come from the zone allocator set up in tcp_init(). 665 */ 666struct tcpcb * 667tcp_newtcpcb(struct inpcb *inp) 668{ 669 struct tcpcb_mem *tm; 670 struct tcpcb *tp; 671#ifdef INET6 672 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 673#endif /* INET6 */ 674 675 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 676 if (tm == NULL) 677 return (NULL); 678 tp = &tm->tcb; 679 680 /* Initialise cc_var struct for this tcpcb. */ 681 tp->ccv = &tm->ccv; 682 tp->ccv->type = IPPROTO_TCP; 683 tp->ccv->ccvc.tcp = tp; 684 685 /* 686 * Use the current system default CC algorithm. 687 */ 688 CC_LIST_RLOCK(); 689 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 690 CC_ALGO(tp) = CC_DEFAULT(); 691 CC_LIST_RUNLOCK(); 692 693 if (CC_ALGO(tp)->cb_init != NULL) 694 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 695 uma_zfree(V_tcpcb_zone, tm); 696 return (NULL); 697 } 698 699 tp->osd = &tm->osd; 700 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 701 uma_zfree(V_tcpcb_zone, tm); 702 return (NULL); 703 } 704 705#ifdef VIMAGE 706 tp->t_vnet = inp->inp_vnet; 707#endif 708 tp->t_timers = &tm->tt; 709 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 710 tp->t_maxseg = tp->t_maxopd = 711#ifdef INET6 712 isipv6 ? V_tcp_v6mssdflt : 713#endif /* INET6 */ 714 V_tcp_mssdflt; 715 716 /* Set up our timeouts. */ 717 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE); 718 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE); 719 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE); 720 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE); 721 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE); 722 723 if (V_tcp_do_rfc1323) 724 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 725 if (V_tcp_do_sack) 726 tp->t_flags |= TF_SACK_PERMIT; 727 TAILQ_INIT(&tp->snd_holes); 728 tp->t_inpcb = inp; /* XXX */ 729 /* 730 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 731 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 732 * reasonable initial retransmit time. 733 */ 734 tp->t_srtt = TCPTV_SRTTBASE; 735 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 736 tp->t_rttmin = tcp_rexmit_min; 737 tp->t_rxtcur = TCPTV_RTOBASE; 738 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 739 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 740 tp->t_rcvtime = ticks; 741 /* 742 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 743 * because the socket may be bound to an IPv6 wildcard address, 744 * which may match an IPv4-mapped IPv6 address. 745 */ 746 inp->inp_ip_ttl = V_ip_defttl; 747 inp->inp_ppcb = tp; 748 return (tp); /* XXX */ 749} 750 751/* 752 * Switch the congestion control algorithm back to NewReno for any active 753 * control blocks using an algorithm which is about to go away. 754 * This ensures the CC framework can allow the unload to proceed without leaving 755 * any dangling pointers which would trigger a panic. 756 * Returning non-zero would inform the CC framework that something went wrong 757 * and it would be unsafe to allow the unload to proceed. However, there is no 758 * way for this to occur with this implementation so we always return zero. 759 */ 760int 761tcp_ccalgounload(struct cc_algo *unload_algo) 762{ 763 struct cc_algo *tmpalgo; 764 struct inpcb *inp; 765 struct tcpcb *tp; 766 VNET_ITERATOR_DECL(vnet_iter); 767 768 /* 769 * Check all active control blocks across all network stacks and change 770 * any that are using "unload_algo" back to NewReno. If "unload_algo" 771 * requires cleanup code to be run, call it. 772 */ 773 VNET_LIST_RLOCK(); 774 VNET_FOREACH(vnet_iter) { 775 CURVNET_SET(vnet_iter); 776 INP_INFO_RLOCK(&V_tcbinfo); 777 /* 778 * New connections already part way through being initialised 779 * with the CC algo we're removing will not race with this code 780 * because the INP_INFO_WLOCK is held during initialisation. We 781 * therefore don't enter the loop below until the connection 782 * list has stabilised. 783 */ 784 LIST_FOREACH(inp, &V_tcb, inp_list) { 785 INP_WLOCK(inp); 786 /* Important to skip tcptw structs. */ 787 if (!(inp->inp_flags & INP_TIMEWAIT) && 788 (tp = intotcpcb(inp)) != NULL) { 789 /* 790 * By holding INP_WLOCK here, we are assured 791 * that the connection is not currently 792 * executing inside the CC module's functions 793 * i.e. it is safe to make the switch back to 794 * NewReno. 795 */ 796 if (CC_ALGO(tp) == unload_algo) { 797 tmpalgo = CC_ALGO(tp); 798 /* NewReno does not require any init. */ 799 CC_ALGO(tp) = &newreno_cc_algo; 800 if (tmpalgo->cb_destroy != NULL) 801 tmpalgo->cb_destroy(tp->ccv); 802 } 803 } 804 INP_WUNLOCK(inp); 805 } 806 INP_INFO_RUNLOCK(&V_tcbinfo); 807 CURVNET_RESTORE(); 808 } 809 VNET_LIST_RUNLOCK(); 810 811 return (0); 812} 813 814/* 815 * Drop a TCP connection, reporting 816 * the specified error. If connection is synchronized, 817 * then send a RST to peer. 818 */ 819struct tcpcb * 820tcp_drop(struct tcpcb *tp, int errno) 821{ 822 struct socket *so = tp->t_inpcb->inp_socket; 823 824 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 825 INP_WLOCK_ASSERT(tp->t_inpcb); 826 827 if (TCPS_HAVERCVDSYN(tp->t_state)) { 828 tp->t_state = TCPS_CLOSED; 829 (void) tcp_output(tp); 830 TCPSTAT_INC(tcps_drops); 831 } else 832 TCPSTAT_INC(tcps_conndrops); 833 if (errno == ETIMEDOUT && tp->t_softerror) 834 errno = tp->t_softerror; 835 so->so_error = errno; 836 return (tcp_close(tp)); 837} 838 839void 840tcp_discardcb(struct tcpcb *tp) 841{ 842 struct inpcb *inp = tp->t_inpcb; 843 struct socket *so = inp->inp_socket; 844#ifdef INET6 845 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 846#endif /* INET6 */ 847 848 INP_WLOCK_ASSERT(inp); 849 850 /* 851 * Make sure that all of our timers are stopped before we delete the 852 * PCB. 853 * 854 * XXXRW: Really, we would like to use callout_drain() here in order 855 * to avoid races experienced in tcp_timer.c where a timer is already 856 * executing at this point. However, we can't, both because we're 857 * running in a context where we can't sleep, and also because we 858 * hold locks required by the timers. What we instead need to do is 859 * test to see if callout_drain() is required, and if so, defer some 860 * portion of the remainder of tcp_discardcb() to an asynchronous 861 * context that can callout_drain() and then continue. Some care 862 * will be required to ensure that no further processing takes place 863 * on the tcpcb, even though it hasn't been freed (a flag?). 864 */ 865 callout_stop(&tp->t_timers->tt_rexmt); 866 callout_stop(&tp->t_timers->tt_persist); 867 callout_stop(&tp->t_timers->tt_keep); 868 callout_stop(&tp->t_timers->tt_2msl); 869 callout_stop(&tp->t_timers->tt_delack); 870 871 /* 872 * If we got enough samples through the srtt filter, 873 * save the rtt and rttvar in the routing entry. 874 * 'Enough' is arbitrarily defined as 4 rtt samples. 875 * 4 samples is enough for the srtt filter to converge 876 * to within enough % of the correct value; fewer samples 877 * and we could save a bogus rtt. The danger is not high 878 * as tcp quickly recovers from everything. 879 * XXX: Works very well but needs some more statistics! 880 */ 881 if (tp->t_rttupdated >= 4) { 882 struct hc_metrics_lite metrics; 883 u_long ssthresh; 884 885 bzero(&metrics, sizeof(metrics)); 886 /* 887 * Update the ssthresh always when the conditions below 888 * are satisfied. This gives us better new start value 889 * for the congestion avoidance for new connections. 890 * ssthresh is only set if packet loss occured on a session. 891 * 892 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 893 * being torn down. Ideally this code would not use 'so'. 894 */ 895 ssthresh = tp->snd_ssthresh; 896 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 897 /* 898 * convert the limit from user data bytes to 899 * packets then to packet data bytes. 900 */ 901 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 902 if (ssthresh < 2) 903 ssthresh = 2; 904 ssthresh *= (u_long)(tp->t_maxseg + 905#ifdef INET6 906 (isipv6 ? sizeof (struct ip6_hdr) + 907 sizeof (struct tcphdr) : 908#endif 909 sizeof (struct tcpiphdr) 910#ifdef INET6 911 ) 912#endif 913 ); 914 } else 915 ssthresh = 0; 916 metrics.rmx_ssthresh = ssthresh; 917 918 metrics.rmx_rtt = tp->t_srtt; 919 metrics.rmx_rttvar = tp->t_rttvar; 920 metrics.rmx_cwnd = tp->snd_cwnd; 921 metrics.rmx_sendpipe = 0; 922 metrics.rmx_recvpipe = 0; 923 924 tcp_hc_update(&inp->inp_inc, &metrics); 925 } 926 927 /* free the reassembly queue, if any */ 928 tcp_reass_flush(tp); 929 930#ifdef TCP_OFFLOAD 931 /* Disconnect offload device, if any. */ 932 if (tp->t_flags & TF_TOE) 933 tcp_offload_detach(tp); 934#endif 935 936 tcp_free_sackholes(tp); 937 938 /* Allow the CC algorithm to clean up after itself. */ 939 if (CC_ALGO(tp)->cb_destroy != NULL) 940 CC_ALGO(tp)->cb_destroy(tp->ccv); 941 942 khelp_destroy_osd(tp->osd); 943 944 CC_ALGO(tp) = NULL; 945 inp->inp_ppcb = NULL; 946 tp->t_inpcb = NULL; 947 uma_zfree(V_tcpcb_zone, tp); 948} 949 950/* 951 * Attempt to close a TCP control block, marking it as dropped, and freeing 952 * the socket if we hold the only reference. 953 */ 954struct tcpcb * 955tcp_close(struct tcpcb *tp) 956{ 957 struct inpcb *inp = tp->t_inpcb; 958 struct socket *so; 959 960 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 961 INP_WLOCK_ASSERT(inp); 962 963#ifdef TCP_OFFLOAD 964 if (tp->t_state == TCPS_LISTEN) 965 tcp_offload_listen_stop(tp); 966#endif 967 in_pcbdrop(inp); 968 TCPSTAT_INC(tcps_closed); 969 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 970 so = inp->inp_socket; 971 soisdisconnected(so); 972 if (inp->inp_flags & INP_SOCKREF) { 973 KASSERT(so->so_state & SS_PROTOREF, 974 ("tcp_close: !SS_PROTOREF")); 975 inp->inp_flags &= ~INP_SOCKREF; 976 INP_WUNLOCK(inp); 977 ACCEPT_LOCK(); 978 SOCK_LOCK(so); 979 so->so_state &= ~SS_PROTOREF; 980 sofree(so); 981 return (NULL); 982 } 983 return (tp); 984} 985 986void 987tcp_drain(void) 988{ 989 VNET_ITERATOR_DECL(vnet_iter); 990 991 if (!do_tcpdrain) 992 return; 993 994 VNET_LIST_RLOCK_NOSLEEP(); 995 VNET_FOREACH(vnet_iter) { 996 CURVNET_SET(vnet_iter); 997 struct inpcb *inpb; 998 struct tcpcb *tcpb; 999 1000 /* 1001 * Walk the tcpbs, if existing, and flush the reassembly queue, 1002 * if there is one... 1003 * XXX: The "Net/3" implementation doesn't imply that the TCP 1004 * reassembly queue should be flushed, but in a situation 1005 * where we're really low on mbufs, this is potentially 1006 * usefull. 1007 */ 1008 INP_INFO_RLOCK(&V_tcbinfo); 1009 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 1010 if (inpb->inp_flags & INP_TIMEWAIT) 1011 continue; 1012 INP_WLOCK(inpb); 1013 if ((tcpb = intotcpcb(inpb)) != NULL) { 1014 tcp_reass_flush(tcpb); 1015 tcp_clean_sackreport(tcpb); 1016 } 1017 INP_WUNLOCK(inpb); 1018 } 1019 INP_INFO_RUNLOCK(&V_tcbinfo); 1020 CURVNET_RESTORE(); 1021 } 1022 VNET_LIST_RUNLOCK_NOSLEEP(); 1023} 1024 1025/* 1026 * Notify a tcp user of an asynchronous error; 1027 * store error as soft error, but wake up user 1028 * (for now, won't do anything until can select for soft error). 1029 * 1030 * Do not wake up user since there currently is no mechanism for 1031 * reporting soft errors (yet - a kqueue filter may be added). 1032 */ 1033static struct inpcb * 1034tcp_notify(struct inpcb *inp, int error) 1035{ 1036 struct tcpcb *tp; 1037 1038 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1039 INP_WLOCK_ASSERT(inp); 1040 1041 if ((inp->inp_flags & INP_TIMEWAIT) || 1042 (inp->inp_flags & INP_DROPPED)) 1043 return (inp); 1044 1045 tp = intotcpcb(inp); 1046 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 1047 1048 /* 1049 * Ignore some errors if we are hooked up. 1050 * If connection hasn't completed, has retransmitted several times, 1051 * and receives a second error, give up now. This is better 1052 * than waiting a long time to establish a connection that 1053 * can never complete. 1054 */ 1055 if (tp->t_state == TCPS_ESTABLISHED && 1056 (error == EHOSTUNREACH || error == ENETUNREACH || 1057 error == EHOSTDOWN)) { 1058 return (inp); 1059 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 1060 tp->t_softerror) { 1061 tp = tcp_drop(tp, error); 1062 if (tp != NULL) 1063 return (inp); 1064 else 1065 return (NULL); 1066 } else { 1067 tp->t_softerror = error; 1068 return (inp); 1069 } 1070#if 0 1071 wakeup( &so->so_timeo); 1072 sorwakeup(so); 1073 sowwakeup(so); 1074#endif 1075} 1076 1077static int 1078tcp_pcblist(SYSCTL_HANDLER_ARGS) 1079{ 1080 int error, i, m, n, pcb_count; 1081 struct inpcb *inp, **inp_list; 1082 inp_gen_t gencnt; 1083 struct xinpgen xig; 1084 1085 /* 1086 * The process of preparing the TCB list is too time-consuming and 1087 * resource-intensive to repeat twice on every request. 1088 */ 1089 if (req->oldptr == NULL) { 1090 n = V_tcbinfo.ipi_count + syncache_pcbcount(); 1091 n += imax(n / 8, 10); 1092 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 1093 return (0); 1094 } 1095 1096 if (req->newptr != NULL) 1097 return (EPERM); 1098 1099 /* 1100 * OK, now we're committed to doing something. 1101 */ 1102 INP_INFO_RLOCK(&V_tcbinfo); 1103 gencnt = V_tcbinfo.ipi_gencnt; 1104 n = V_tcbinfo.ipi_count; 1105 INP_INFO_RUNLOCK(&V_tcbinfo); 1106 1107 m = syncache_pcbcount(); 1108 1109 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 1110 + (n + m) * sizeof(struct xtcpcb)); 1111 if (error != 0) 1112 return (error); 1113 1114 xig.xig_len = sizeof xig; 1115 xig.xig_count = n + m; 1116 xig.xig_gen = gencnt; 1117 xig.xig_sogen = so_gencnt; 1118 error = SYSCTL_OUT(req, &xig, sizeof xig); 1119 if (error) 1120 return (error); 1121 1122 error = syncache_pcblist(req, m, &pcb_count); 1123 if (error) 1124 return (error); 1125 1126 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 1127 if (inp_list == NULL) 1128 return (ENOMEM); 1129 1130 INP_INFO_RLOCK(&V_tcbinfo); 1131 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; 1132 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) { 1133 INP_WLOCK(inp); 1134 if (inp->inp_gencnt <= gencnt) { 1135 /* 1136 * XXX: This use of cr_cansee(), introduced with 1137 * TCP state changes, is not quite right, but for 1138 * now, better than nothing. 1139 */ 1140 if (inp->inp_flags & INP_TIMEWAIT) { 1141 if (intotw(inp) != NULL) 1142 error = cr_cansee(req->td->td_ucred, 1143 intotw(inp)->tw_cred); 1144 else 1145 error = EINVAL; /* Skip this inp. */ 1146 } else 1147 error = cr_canseeinpcb(req->td->td_ucred, inp); 1148 if (error == 0) { 1149 in_pcbref(inp); 1150 inp_list[i++] = inp; 1151 } 1152 } 1153 INP_WUNLOCK(inp); 1154 } 1155 INP_INFO_RUNLOCK(&V_tcbinfo); 1156 n = i; 1157 1158 error = 0; 1159 for (i = 0; i < n; i++) { 1160 inp = inp_list[i]; 1161 INP_RLOCK(inp); 1162 if (inp->inp_gencnt <= gencnt) { 1163 struct xtcpcb xt; 1164 void *inp_ppcb; 1165 1166 bzero(&xt, sizeof(xt)); 1167 xt.xt_len = sizeof xt; 1168 /* XXX should avoid extra copy */ 1169 bcopy(inp, &xt.xt_inp, sizeof *inp); 1170 inp_ppcb = inp->inp_ppcb; 1171 if (inp_ppcb == NULL) 1172 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1173 else if (inp->inp_flags & INP_TIMEWAIT) { 1174 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1175 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1176 } else { 1177 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1178 if (xt.xt_tp.t_timers) 1179 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer); 1180 } 1181 if (inp->inp_socket != NULL) 1182 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1183 else { 1184 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1185 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1186 } 1187 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1188 INP_RUNLOCK(inp); 1189 error = SYSCTL_OUT(req, &xt, sizeof xt); 1190 } else 1191 INP_RUNLOCK(inp); 1192 } 1193 INP_INFO_WLOCK(&V_tcbinfo); 1194 for (i = 0; i < n; i++) { 1195 inp = inp_list[i]; 1196 INP_RLOCK(inp); 1197 if (!in_pcbrele_rlocked(inp)) 1198 INP_RUNLOCK(inp); 1199 } 1200 INP_INFO_WUNLOCK(&V_tcbinfo); 1201 1202 if (!error) { 1203 /* 1204 * Give the user an updated idea of our state. 1205 * If the generation differs from what we told 1206 * her before, she knows that something happened 1207 * while we were processing this request, and it 1208 * might be necessary to retry. 1209 */ 1210 INP_INFO_RLOCK(&V_tcbinfo); 1211 xig.xig_gen = V_tcbinfo.ipi_gencnt; 1212 xig.xig_sogen = so_gencnt; 1213 xig.xig_count = V_tcbinfo.ipi_count + pcb_count; 1214 INP_INFO_RUNLOCK(&V_tcbinfo); 1215 error = SYSCTL_OUT(req, &xig, sizeof xig); 1216 } 1217 free(inp_list, M_TEMP); 1218 return (error); 1219} 1220 1221SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 1222 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 1223 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1224 1225#ifdef INET 1226static int 1227tcp_getcred(SYSCTL_HANDLER_ARGS) 1228{ 1229 struct xucred xuc; 1230 struct sockaddr_in addrs[2]; 1231 struct inpcb *inp; 1232 int error; 1233 1234 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1235 if (error) 1236 return (error); 1237 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1238 if (error) 1239 return (error); 1240 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1241 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 1242 if (inp != NULL) { 1243 if (inp->inp_socket == NULL) 1244 error = ENOENT; 1245 if (error == 0) 1246 error = cr_canseeinpcb(req->td->td_ucred, inp); 1247 if (error == 0) 1248 cru2x(inp->inp_cred, &xuc); 1249 INP_RUNLOCK(inp); 1250 } else 1251 error = ENOENT; 1252 if (error == 0) 1253 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1254 return (error); 1255} 1256 1257SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1258 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1259 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1260#endif /* INET */ 1261 1262#ifdef INET6 1263static int 1264tcp6_getcred(SYSCTL_HANDLER_ARGS) 1265{ 1266 struct xucred xuc; 1267 struct sockaddr_in6 addrs[2]; 1268 struct inpcb *inp; 1269 int error; 1270#ifdef INET 1271 int mapped = 0; 1272#endif 1273 1274 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1275 if (error) 1276 return (error); 1277 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1278 if (error) 1279 return (error); 1280 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 1281 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 1282 return (error); 1283 } 1284 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1285#ifdef INET 1286 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1287 mapped = 1; 1288 else 1289#endif 1290 return (EINVAL); 1291 } 1292 1293#ifdef INET 1294 if (mapped == 1) 1295 inp = in_pcblookup(&V_tcbinfo, 1296 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1297 addrs[1].sin6_port, 1298 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1299 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 1300 else 1301#endif 1302 inp = in6_pcblookup(&V_tcbinfo, 1303 &addrs[1].sin6_addr, addrs[1].sin6_port, 1304 &addrs[0].sin6_addr, addrs[0].sin6_port, 1305 INPLOOKUP_RLOCKPCB, NULL); 1306 if (inp != NULL) { 1307 if (inp->inp_socket == NULL) 1308 error = ENOENT; 1309 if (error == 0) 1310 error = cr_canseeinpcb(req->td->td_ucred, inp); 1311 if (error == 0) 1312 cru2x(inp->inp_cred, &xuc); 1313 INP_RUNLOCK(inp); 1314 } else 1315 error = ENOENT; 1316 if (error == 0) 1317 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1318 return (error); 1319} 1320 1321SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1322 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1323 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1324#endif /* INET6 */ 1325 1326 1327#ifdef INET 1328void 1329tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1330{ 1331 struct ip *ip = vip; 1332 struct tcphdr *th; 1333 struct in_addr faddr; 1334 struct inpcb *inp; 1335 struct tcpcb *tp; 1336 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1337 struct icmp *icp; 1338 struct in_conninfo inc; 1339 tcp_seq icmp_tcp_seq; 1340 int mtu; 1341 1342 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1343 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1344 return; 1345 1346 if (cmd == PRC_MSGSIZE) 1347 notify = tcp_mtudisc_notify; 1348 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1349 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1350 notify = tcp_drop_syn_sent; 1351 /* 1352 * Redirects don't need to be handled up here. 1353 */ 1354 else if (PRC_IS_REDIRECT(cmd)) 1355 return; 1356 /* 1357 * Source quench is depreciated. 1358 */ 1359 else if (cmd == PRC_QUENCH) 1360 return; 1361 /* 1362 * Hostdead is ugly because it goes linearly through all PCBs. 1363 * XXX: We never get this from ICMP, otherwise it makes an 1364 * excellent DoS attack on machines with many connections. 1365 */ 1366 else if (cmd == PRC_HOSTDEAD) 1367 ip = NULL; 1368 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1369 return; 1370 if (ip != NULL) { 1371 icp = (struct icmp *)((caddr_t)ip 1372 - offsetof(struct icmp, icmp_ip)); 1373 th = (struct tcphdr *)((caddr_t)ip 1374 + (ip->ip_hl << 2)); 1375 INP_INFO_WLOCK(&V_tcbinfo); 1376 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, 1377 ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 1378 if (inp != NULL) { 1379 if (!(inp->inp_flags & INP_TIMEWAIT) && 1380 !(inp->inp_flags & INP_DROPPED) && 1381 !(inp->inp_socket == NULL)) { 1382 icmp_tcp_seq = htonl(th->th_seq); 1383 tp = intotcpcb(inp); 1384 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1385 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1386 if (cmd == PRC_MSGSIZE) { 1387 /* 1388 * MTU discovery: 1389 * If we got a needfrag set the MTU 1390 * in the route to the suggested new 1391 * value (if given) and then notify. 1392 */ 1393 bzero(&inc, sizeof(inc)); 1394 inc.inc_faddr = faddr; 1395 inc.inc_fibnum = 1396 inp->inp_inc.inc_fibnum; 1397 1398 mtu = ntohs(icp->icmp_nextmtu); 1399 /* 1400 * If no alternative MTU was 1401 * proposed, try the next smaller 1402 * one. ip->ip_len has already 1403 * been swapped in icmp_input(). 1404 */ 1405 if (!mtu) 1406 mtu = ip_next_mtu(ip->ip_len, 1407 1); 1408 if (mtu < V_tcp_minmss 1409 + sizeof(struct tcpiphdr)) 1410 mtu = V_tcp_minmss 1411 + sizeof(struct tcpiphdr); 1412 /* 1413 * Only cache the MTU if it 1414 * is smaller than the interface 1415 * or route MTU. tcp_mtudisc() 1416 * will do right thing by itself. 1417 */ 1418 if (mtu <= tcp_maxmtu(&inc, NULL)) 1419 tcp_hc_updatemtu(&inc, mtu); 1420 tcp_mtudisc(inp, mtu); 1421 } else 1422 inp = (*notify)(inp, 1423 inetctlerrmap[cmd]); 1424 } 1425 } 1426 if (inp != NULL) 1427 INP_WUNLOCK(inp); 1428 } else { 1429 bzero(&inc, sizeof(inc)); 1430 inc.inc_fport = th->th_dport; 1431 inc.inc_lport = th->th_sport; 1432 inc.inc_faddr = faddr; 1433 inc.inc_laddr = ip->ip_src; 1434 syncache_unreach(&inc, th); 1435 } 1436 INP_INFO_WUNLOCK(&V_tcbinfo); 1437 } else 1438 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 1439} 1440#endif /* INET */ 1441 1442#ifdef INET6 1443void 1444tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1445{ 1446 struct tcphdr th; 1447 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1448 struct ip6_hdr *ip6; 1449 struct mbuf *m; 1450 struct ip6ctlparam *ip6cp = NULL; 1451 const struct sockaddr_in6 *sa6_src = NULL; 1452 int off; 1453 struct tcp_portonly { 1454 u_int16_t th_sport; 1455 u_int16_t th_dport; 1456 } *thp; 1457 1458 if (sa->sa_family != AF_INET6 || 1459 sa->sa_len != sizeof(struct sockaddr_in6)) 1460 return; 1461 1462 if (cmd == PRC_MSGSIZE) 1463 notify = tcp_mtudisc_notify; 1464 else if (!PRC_IS_REDIRECT(cmd) && 1465 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1466 return; 1467 /* Source quench is depreciated. */ 1468 else if (cmd == PRC_QUENCH) 1469 return; 1470 1471 /* if the parameter is from icmp6, decode it. */ 1472 if (d != NULL) { 1473 ip6cp = (struct ip6ctlparam *)d; 1474 m = ip6cp->ip6c_m; 1475 ip6 = ip6cp->ip6c_ip6; 1476 off = ip6cp->ip6c_off; 1477 sa6_src = ip6cp->ip6c_src; 1478 } else { 1479 m = NULL; 1480 ip6 = NULL; 1481 off = 0; /* fool gcc */ 1482 sa6_src = &sa6_any; 1483 } 1484 1485 if (ip6 != NULL) { 1486 struct in_conninfo inc; 1487 /* 1488 * XXX: We assume that when IPV6 is non NULL, 1489 * M and OFF are valid. 1490 */ 1491 1492 /* check if we can safely examine src and dst ports */ 1493 if (m->m_pkthdr.len < off + sizeof(*thp)) 1494 return; 1495 1496 bzero(&th, sizeof(th)); 1497 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1498 1499 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport, 1500 (struct sockaddr *)ip6cp->ip6c_src, 1501 th.th_sport, cmd, NULL, notify); 1502 1503 bzero(&inc, sizeof(inc)); 1504 inc.inc_fport = th.th_dport; 1505 inc.inc_lport = th.th_sport; 1506 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1507 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1508 inc.inc_flags |= INC_ISIPV6; 1509 INP_INFO_WLOCK(&V_tcbinfo); 1510 syncache_unreach(&inc, &th); 1511 INP_INFO_WUNLOCK(&V_tcbinfo); 1512 } else 1513 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1514 0, cmd, NULL, notify); 1515} 1516#endif /* INET6 */ 1517 1518 1519/* 1520 * Following is where TCP initial sequence number generation occurs. 1521 * 1522 * There are two places where we must use initial sequence numbers: 1523 * 1. In SYN-ACK packets. 1524 * 2. In SYN packets. 1525 * 1526 * All ISNs for SYN-ACK packets are generated by the syncache. See 1527 * tcp_syncache.c for details. 1528 * 1529 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1530 * depends on this property. In addition, these ISNs should be 1531 * unguessable so as to prevent connection hijacking. To satisfy 1532 * the requirements of this situation, the algorithm outlined in 1533 * RFC 1948 is used, with only small modifications. 1534 * 1535 * Implementation details: 1536 * 1537 * Time is based off the system timer, and is corrected so that it 1538 * increases by one megabyte per second. This allows for proper 1539 * recycling on high speed LANs while still leaving over an hour 1540 * before rollover. 1541 * 1542 * As reading the *exact* system time is too expensive to be done 1543 * whenever setting up a TCP connection, we increment the time 1544 * offset in two ways. First, a small random positive increment 1545 * is added to isn_offset for each connection that is set up. 1546 * Second, the function tcp_isn_tick fires once per clock tick 1547 * and increments isn_offset as necessary so that sequence numbers 1548 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1549 * random positive increments serve only to ensure that the same 1550 * exact sequence number is never sent out twice (as could otherwise 1551 * happen when a port is recycled in less than the system tick 1552 * interval.) 1553 * 1554 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1555 * between seeding of isn_secret. This is normally set to zero, 1556 * as reseeding should not be necessary. 1557 * 1558 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1559 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1560 * general, this means holding an exclusive (write) lock. 1561 */ 1562 1563#define ISN_BYTES_PER_SECOND 1048576 1564#define ISN_STATIC_INCREMENT 4096 1565#define ISN_RANDOM_INCREMENT (4096 - 1) 1566 1567static VNET_DEFINE(u_char, isn_secret[32]); 1568static VNET_DEFINE(int, isn_last); 1569static VNET_DEFINE(int, isn_last_reseed); 1570static VNET_DEFINE(u_int32_t, isn_offset); 1571static VNET_DEFINE(u_int32_t, isn_offset_old); 1572 1573#define V_isn_secret VNET(isn_secret) 1574#define V_isn_last VNET(isn_last) 1575#define V_isn_last_reseed VNET(isn_last_reseed) 1576#define V_isn_offset VNET(isn_offset) 1577#define V_isn_offset_old VNET(isn_offset_old) 1578 1579tcp_seq 1580tcp_new_isn(struct tcpcb *tp) 1581{ 1582 MD5_CTX isn_ctx; 1583 u_int32_t md5_buffer[4]; 1584 tcp_seq new_isn; 1585 u_int32_t projected_offset; 1586 1587 INP_WLOCK_ASSERT(tp->t_inpcb); 1588 1589 ISN_LOCK(); 1590 /* Seed if this is the first use, reseed if requested. */ 1591 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 1592 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 1593 < (u_int)ticks))) { 1594 read_random(&V_isn_secret, sizeof(V_isn_secret)); 1595 V_isn_last_reseed = ticks; 1596 } 1597 1598 /* Compute the md5 hash and return the ISN. */ 1599 MD5Init(&isn_ctx); 1600 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1601 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1602#ifdef INET6 1603 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1604 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1605 sizeof(struct in6_addr)); 1606 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1607 sizeof(struct in6_addr)); 1608 } else 1609#endif 1610 { 1611 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1612 sizeof(struct in_addr)); 1613 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1614 sizeof(struct in_addr)); 1615 } 1616 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret)); 1617 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1618 new_isn = (tcp_seq) md5_buffer[0]; 1619 V_isn_offset += ISN_STATIC_INCREMENT + 1620 (arc4random() & ISN_RANDOM_INCREMENT); 1621 if (ticks != V_isn_last) { 1622 projected_offset = V_isn_offset_old + 1623 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 1624 if (SEQ_GT(projected_offset, V_isn_offset)) 1625 V_isn_offset = projected_offset; 1626 V_isn_offset_old = V_isn_offset; 1627 V_isn_last = ticks; 1628 } 1629 new_isn += V_isn_offset; 1630 ISN_UNLOCK(); 1631 return (new_isn); 1632} 1633 1634/* 1635 * When a specific ICMP unreachable message is received and the 1636 * connection state is SYN-SENT, drop the connection. This behavior 1637 * is controlled by the icmp_may_rst sysctl. 1638 */ 1639struct inpcb * 1640tcp_drop_syn_sent(struct inpcb *inp, int errno) 1641{ 1642 struct tcpcb *tp; 1643 1644 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1645 INP_WLOCK_ASSERT(inp); 1646 1647 if ((inp->inp_flags & INP_TIMEWAIT) || 1648 (inp->inp_flags & INP_DROPPED)) 1649 return (inp); 1650 1651 tp = intotcpcb(inp); 1652 if (tp->t_state != TCPS_SYN_SENT) 1653 return (inp); 1654 1655 tp = tcp_drop(tp, errno); 1656 if (tp != NULL) 1657 return (inp); 1658 else 1659 return (NULL); 1660} 1661 1662/* 1663 * When `need fragmentation' ICMP is received, update our idea of the MSS 1664 * based on the new value. Also nudge TCP to send something, since we 1665 * know the packet we just sent was dropped. 1666 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1667 */ 1668static struct inpcb * 1669tcp_mtudisc_notify(struct inpcb *inp, int error) 1670{ 1671 1672 return (tcp_mtudisc(inp, -1)); 1673} 1674 1675struct inpcb * 1676tcp_mtudisc(struct inpcb *inp, int mtuoffer) 1677{ 1678 struct tcpcb *tp; 1679 struct socket *so; 1680 1681 INP_WLOCK_ASSERT(inp); 1682 if ((inp->inp_flags & INP_TIMEWAIT) || 1683 (inp->inp_flags & INP_DROPPED)) 1684 return (inp); 1685 1686 tp = intotcpcb(inp); 1687 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1688 1689 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 1690 1691 so = inp->inp_socket; 1692 SOCKBUF_LOCK(&so->so_snd); 1693 /* If the mss is larger than the socket buffer, decrease the mss. */ 1694 if (so->so_snd.sb_hiwat < tp->t_maxseg) 1695 tp->t_maxseg = so->so_snd.sb_hiwat; 1696 SOCKBUF_UNLOCK(&so->so_snd); 1697 1698 TCPSTAT_INC(tcps_mturesent); 1699 tp->t_rtttime = 0; 1700 tp->snd_nxt = tp->snd_una; 1701 tcp_free_sackholes(tp); 1702 tp->snd_recover = tp->snd_max; 1703 if (tp->t_flags & TF_SACK_PERMIT) 1704 EXIT_FASTRECOVERY(tp->t_flags); 1705 tcp_output(tp); 1706 return (inp); 1707} 1708 1709#ifdef INET 1710/* 1711 * Look-up the routing entry to the peer of this inpcb. If no route 1712 * is found and it cannot be allocated, then return 0. This routine 1713 * is called by TCP routines that access the rmx structure and by 1714 * tcp_mss_update to get the peer/interface MTU. 1715 */ 1716u_long 1717tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 1718{ 1719 struct route sro; 1720 struct sockaddr_in *dst; 1721 struct ifnet *ifp; 1722 u_long maxmtu = 0; 1723 1724 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1725 1726 bzero(&sro, sizeof(sro)); 1727 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1728 dst = (struct sockaddr_in *)&sro.ro_dst; 1729 dst->sin_family = AF_INET; 1730 dst->sin_len = sizeof(*dst); 1731 dst->sin_addr = inc->inc_faddr; 1732 in_rtalloc_ign(&sro, 0, inc->inc_fibnum); 1733 } 1734 if (sro.ro_rt != NULL) { 1735 ifp = sro.ro_rt->rt_ifp; 1736 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1737 maxmtu = ifp->if_mtu; 1738 else 1739 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1740 1741 /* Report additional interface capabilities. */ 1742 if (cap != NULL) { 1743 if (ifp->if_capenable & IFCAP_TSO4 && 1744 ifp->if_hwassist & CSUM_TSO) { 1745 cap->ifcap |= CSUM_TSO; 1746 cap->tsomax = ifp->if_hw_tsomax; 1747 } 1748 } 1749 RTFREE(sro.ro_rt); 1750 } 1751 return (maxmtu); 1752} 1753#endif /* INET */ 1754 1755#ifdef INET6 1756u_long 1757tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 1758{ 1759 struct route_in6 sro6; 1760 struct ifnet *ifp; 1761 u_long maxmtu = 0; 1762 1763 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1764 1765 bzero(&sro6, sizeof(sro6)); 1766 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1767 sro6.ro_dst.sin6_family = AF_INET6; 1768 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1769 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1770 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum); 1771 } 1772 if (sro6.ro_rt != NULL) { 1773 ifp = sro6.ro_rt->rt_ifp; 1774 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1775 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1776 else 1777 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1778 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1779 1780 /* Report additional interface capabilities. */ 1781 if (cap != NULL) { 1782 if (ifp->if_capenable & IFCAP_TSO6 && 1783 ifp->if_hwassist & CSUM_TSO) { 1784 cap->ifcap |= CSUM_TSO; 1785 cap->tsomax = ifp->if_hw_tsomax; 1786 } 1787 } 1788 RTFREE(sro6.ro_rt); 1789 } 1790 1791 return (maxmtu); 1792} 1793#endif /* INET6 */ 1794 1795#ifdef IPSEC 1796/* compute ESP/AH header size for TCP, including outer IP header. */ 1797size_t 1798ipsec_hdrsiz_tcp(struct tcpcb *tp) 1799{ 1800 struct inpcb *inp; 1801 struct mbuf *m; 1802 size_t hdrsiz; 1803 struct ip *ip; 1804#ifdef INET6 1805 struct ip6_hdr *ip6; 1806#endif 1807 struct tcphdr *th; 1808 1809 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1810 return (0); 1811 MGETHDR(m, M_DONTWAIT, MT_DATA); 1812 if (!m) 1813 return (0); 1814 1815#ifdef INET6 1816 if ((inp->inp_vflag & INP_IPV6) != 0) { 1817 ip6 = mtod(m, struct ip6_hdr *); 1818 th = (struct tcphdr *)(ip6 + 1); 1819 m->m_pkthdr.len = m->m_len = 1820 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1821 tcpip_fillheaders(inp, ip6, th); 1822 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1823 } else 1824#endif /* INET6 */ 1825 { 1826 ip = mtod(m, struct ip *); 1827 th = (struct tcphdr *)(ip + 1); 1828 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1829 tcpip_fillheaders(inp, ip, th); 1830 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1831 } 1832 1833 m_free(m); 1834 return (hdrsiz); 1835} 1836#endif /* IPSEC */ 1837 1838#ifdef TCP_SIGNATURE 1839/* 1840 * Callback function invoked by m_apply() to digest TCP segment data 1841 * contained within an mbuf chain. 1842 */ 1843static int 1844tcp_signature_apply(void *fstate, void *data, u_int len) 1845{ 1846 1847 MD5Update(fstate, (u_char *)data, len); 1848 return (0); 1849} 1850 1851/* 1852 * Compute TCP-MD5 hash of a TCP segment. (RFC2385) 1853 * 1854 * Parameters: 1855 * m pointer to head of mbuf chain 1856 * _unused 1857 * len length of TCP segment data, excluding options 1858 * optlen length of TCP segment options 1859 * buf pointer to storage for computed MD5 digest 1860 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 1861 * 1862 * We do this over ip, tcphdr, segment data, and the key in the SADB. 1863 * When called from tcp_input(), we can be sure that th_sum has been 1864 * zeroed out and verified already. 1865 * 1866 * Return 0 if successful, otherwise return -1. 1867 * 1868 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 1869 * search with the destination IP address, and a 'magic SPI' to be 1870 * determined by the application. This is hardcoded elsewhere to 1179 1871 * right now. Another branch of this code exists which uses the SPD to 1872 * specify per-application flows but it is unstable. 1873 */ 1874int 1875tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen, 1876 u_char *buf, u_int direction) 1877{ 1878 union sockaddr_union dst; 1879#ifdef INET 1880 struct ippseudo ippseudo; 1881#endif 1882 MD5_CTX ctx; 1883 int doff; 1884 struct ip *ip; 1885#ifdef INET 1886 struct ipovly *ipovly; 1887#endif 1888 struct secasvar *sav; 1889 struct tcphdr *th; 1890#ifdef INET6 1891 struct ip6_hdr *ip6; 1892 struct in6_addr in6; 1893 char ip6buf[INET6_ADDRSTRLEN]; 1894 uint32_t plen; 1895 uint16_t nhdr; 1896#endif 1897 u_short savecsum; 1898 1899 KASSERT(m != NULL, ("NULL mbuf chain")); 1900 KASSERT(buf != NULL, ("NULL signature pointer")); 1901 1902 /* Extract the destination from the IP header in the mbuf. */ 1903 bzero(&dst, sizeof(union sockaddr_union)); 1904 ip = mtod(m, struct ip *); 1905#ifdef INET6 1906 ip6 = NULL; /* Make the compiler happy. */ 1907#endif 1908 switch (ip->ip_v) { 1909#ifdef INET 1910 case IPVERSION: 1911 dst.sa.sa_len = sizeof(struct sockaddr_in); 1912 dst.sa.sa_family = AF_INET; 1913 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 1914 ip->ip_src : ip->ip_dst; 1915 break; 1916#endif 1917#ifdef INET6 1918 case (IPV6_VERSION >> 4): 1919 ip6 = mtod(m, struct ip6_hdr *); 1920 dst.sa.sa_len = sizeof(struct sockaddr_in6); 1921 dst.sa.sa_family = AF_INET6; 1922 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ? 1923 ip6->ip6_src : ip6->ip6_dst; 1924 break; 1925#endif 1926 default: 1927 return (EINVAL); 1928 /* NOTREACHED */ 1929 break; 1930 } 1931 1932 /* Look up an SADB entry which matches the address of the peer. */ 1933 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 1934 if (sav == NULL) { 1935 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__, 1936 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) : 1937#ifdef INET6 1938 (ip->ip_v == (IPV6_VERSION >> 4)) ? 1939 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) : 1940#endif 1941 "(unsupported)")); 1942 return (EINVAL); 1943 } 1944 1945 MD5Init(&ctx); 1946 /* 1947 * Step 1: Update MD5 hash with IP(v6) pseudo-header. 1948 * 1949 * XXX The ippseudo header MUST be digested in network byte order, 1950 * or else we'll fail the regression test. Assume all fields we've 1951 * been doing arithmetic on have been in host byte order. 1952 * XXX One cannot depend on ipovly->ih_len here. When called from 1953 * tcp_output(), the underlying ip_len member has not yet been set. 1954 */ 1955 switch (ip->ip_v) { 1956#ifdef INET 1957 case IPVERSION: 1958 ipovly = (struct ipovly *)ip; 1959 ippseudo.ippseudo_src = ipovly->ih_src; 1960 ippseudo.ippseudo_dst = ipovly->ih_dst; 1961 ippseudo.ippseudo_pad = 0; 1962 ippseudo.ippseudo_p = IPPROTO_TCP; 1963 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + 1964 optlen); 1965 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 1966 1967 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip)); 1968 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen; 1969 break; 1970#endif 1971#ifdef INET6 1972 /* 1973 * RFC 2385, 2.0 Proposal 1974 * For IPv6, the pseudo-header is as described in RFC 2460, namely the 1975 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero- 1976 * extended next header value (to form 32 bits), and 32-bit segment 1977 * length. 1978 * Note: Upper-Layer Packet Length comes before Next Header. 1979 */ 1980 case (IPV6_VERSION >> 4): 1981 in6 = ip6->ip6_src; 1982 in6_clearscope(&in6); 1983 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 1984 in6 = ip6->ip6_dst; 1985 in6_clearscope(&in6); 1986 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 1987 plen = htonl(len + sizeof(struct tcphdr) + optlen); 1988 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t)); 1989 nhdr = 0; 1990 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 1991 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 1992 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 1993 nhdr = IPPROTO_TCP; 1994 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 1995 1996 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr)); 1997 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen; 1998 break; 1999#endif 2000 default: 2001 return (EINVAL); 2002 /* NOTREACHED */ 2003 break; 2004 } 2005 2006 2007 /* 2008 * Step 2: Update MD5 hash with TCP header, excluding options. 2009 * The TCP checksum must be set to zero. 2010 */ 2011 savecsum = th->th_sum; 2012 th->th_sum = 0; 2013 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2014 th->th_sum = savecsum; 2015 2016 /* 2017 * Step 3: Update MD5 hash with TCP segment data. 2018 * Use m_apply() to avoid an early m_pullup(). 2019 */ 2020 if (len > 0) 2021 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2022 2023 /* 2024 * Step 4: Update MD5 hash with shared secret. 2025 */ 2026 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth)); 2027 MD5Final(buf, &ctx); 2028 2029 key_sa_recordxfer(sav, m); 2030 KEY_FREESAV(&sav); 2031 return (0); 2032} 2033 2034/* 2035 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385) 2036 * 2037 * Parameters: 2038 * m pointer to head of mbuf chain 2039 * len length of TCP segment data, excluding options 2040 * optlen length of TCP segment options 2041 * buf pointer to storage for computed MD5 digest 2042 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2043 * 2044 * Return 1 if successful, otherwise return 0. 2045 */ 2046int 2047tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen, 2048 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 2049{ 2050 char tmpdigest[TCP_SIGLEN]; 2051 2052 if (tcp_sig_checksigs == 0) 2053 return (1); 2054 if ((tcpbflag & TF_SIGNATURE) == 0) { 2055 if ((to->to_flags & TOF_SIGNATURE) != 0) { 2056 2057 /* 2058 * If this socket is not expecting signature but 2059 * the segment contains signature just fail. 2060 */ 2061 TCPSTAT_INC(tcps_sig_err_sigopt); 2062 TCPSTAT_INC(tcps_sig_rcvbadsig); 2063 return (0); 2064 } 2065 2066 /* Signature is not expected, and not present in segment. */ 2067 return (1); 2068 } 2069 2070 /* 2071 * If this socket is expecting signature but the segment does not 2072 * contain any just fail. 2073 */ 2074 if ((to->to_flags & TOF_SIGNATURE) == 0) { 2075 TCPSTAT_INC(tcps_sig_err_nosigopt); 2076 TCPSTAT_INC(tcps_sig_rcvbadsig); 2077 return (0); 2078 } 2079 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0], 2080 IPSEC_DIR_INBOUND) == -1) { 2081 TCPSTAT_INC(tcps_sig_err_buildsig); 2082 TCPSTAT_INC(tcps_sig_rcvbadsig); 2083 return (0); 2084 } 2085 2086 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) { 2087 TCPSTAT_INC(tcps_sig_rcvbadsig); 2088 return (0); 2089 } 2090 TCPSTAT_INC(tcps_sig_rcvgoodsig); 2091 return (1); 2092} 2093#endif /* TCP_SIGNATURE */ 2094 2095static int 2096sysctl_drop(SYSCTL_HANDLER_ARGS) 2097{ 2098 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2099 struct sockaddr_storage addrs[2]; 2100 struct inpcb *inp; 2101 struct tcpcb *tp; 2102 struct tcptw *tw; 2103 struct sockaddr_in *fin, *lin; 2104#ifdef INET6 2105 struct sockaddr_in6 *fin6, *lin6; 2106#endif 2107 int error; 2108 2109 inp = NULL; 2110 fin = lin = NULL; 2111#ifdef INET6 2112 fin6 = lin6 = NULL; 2113#endif 2114 error = 0; 2115 2116 if (req->oldptr != NULL || req->oldlen != 0) 2117 return (EINVAL); 2118 if (req->newptr == NULL) 2119 return (EPERM); 2120 if (req->newlen < sizeof(addrs)) 2121 return (ENOMEM); 2122 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2123 if (error) 2124 return (error); 2125 2126 switch (addrs[0].ss_family) { 2127#ifdef INET6 2128 case AF_INET6: 2129 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2130 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2131 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2132 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2133 return (EINVAL); 2134 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2135 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2136 return (EINVAL); 2137 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2138 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2139 fin = (struct sockaddr_in *)&addrs[0]; 2140 lin = (struct sockaddr_in *)&addrs[1]; 2141 break; 2142 } 2143 error = sa6_embedscope(fin6, V_ip6_use_defzone); 2144 if (error) 2145 return (error); 2146 error = sa6_embedscope(lin6, V_ip6_use_defzone); 2147 if (error) 2148 return (error); 2149 break; 2150#endif 2151#ifdef INET 2152 case AF_INET: 2153 fin = (struct sockaddr_in *)&addrs[0]; 2154 lin = (struct sockaddr_in *)&addrs[1]; 2155 if (fin->sin_len != sizeof(struct sockaddr_in) || 2156 lin->sin_len != sizeof(struct sockaddr_in)) 2157 return (EINVAL); 2158 break; 2159#endif 2160 default: 2161 return (EINVAL); 2162 } 2163 INP_INFO_WLOCK(&V_tcbinfo); 2164 switch (addrs[0].ss_family) { 2165#ifdef INET6 2166 case AF_INET6: 2167 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 2168 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 2169 INPLOOKUP_WLOCKPCB, NULL); 2170 break; 2171#endif 2172#ifdef INET 2173 case AF_INET: 2174 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 2175 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 2176 break; 2177#endif 2178 } 2179 if (inp != NULL) { 2180 if (inp->inp_flags & INP_TIMEWAIT) { 2181 /* 2182 * XXXRW: There currently exists a state where an 2183 * inpcb is present, but its timewait state has been 2184 * discarded. For now, don't allow dropping of this 2185 * type of inpcb. 2186 */ 2187 tw = intotw(inp); 2188 if (tw != NULL) 2189 tcp_twclose(tw, 0); 2190 else 2191 INP_WUNLOCK(inp); 2192 } else if (!(inp->inp_flags & INP_DROPPED) && 2193 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2194 tp = intotcpcb(inp); 2195 tp = tcp_drop(tp, ECONNABORTED); 2196 if (tp != NULL) 2197 INP_WUNLOCK(inp); 2198 } else 2199 INP_WUNLOCK(inp); 2200 } else 2201 error = ESRCH; 2202 INP_INFO_WUNLOCK(&V_tcbinfo); 2203 return (error); 2204} 2205 2206SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2207 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2208 0, sysctl_drop, "", "Drop TCP connection"); 2209 2210/* 2211 * Generate a standardized TCP log line for use throughout the 2212 * tcp subsystem. Memory allocation is done with M_NOWAIT to 2213 * allow use in the interrupt context. 2214 * 2215 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 2216 * NB: The function may return NULL if memory allocation failed. 2217 * 2218 * Due to header inclusion and ordering limitations the struct ip 2219 * and ip6_hdr pointers have to be passed as void pointers. 2220 */ 2221char * 2222tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2223 const void *ip6hdr) 2224{ 2225 2226 /* Is logging enabled? */ 2227 if (tcp_log_in_vain == 0) 2228 return (NULL); 2229 2230 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2231} 2232 2233char * 2234tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2235 const void *ip6hdr) 2236{ 2237 2238 /* Is logging enabled? */ 2239 if (tcp_log_debug == 0) 2240 return (NULL); 2241 2242 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2243} 2244 2245static char * 2246tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2247 const void *ip6hdr) 2248{ 2249 char *s, *sp; 2250 size_t size; 2251 struct ip *ip; 2252#ifdef INET6 2253 const struct ip6_hdr *ip6; 2254 2255 ip6 = (const struct ip6_hdr *)ip6hdr; 2256#endif /* INET6 */ 2257 ip = (struct ip *)ip4hdr; 2258 2259 /* 2260 * The log line looks like this: 2261 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 2262 */ 2263 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 2264 sizeof(PRINT_TH_FLAGS) + 1 + 2265#ifdef INET6 2266 2 * INET6_ADDRSTRLEN; 2267#else 2268 2 * INET_ADDRSTRLEN; 2269#endif /* INET6 */ 2270 2271 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 2272 if (s == NULL) 2273 return (NULL); 2274 2275 strcat(s, "TCP: ["); 2276 sp = s + strlen(s); 2277 2278 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 2279 inet_ntoa_r(inc->inc_faddr, sp); 2280 sp = s + strlen(s); 2281 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2282 sp = s + strlen(s); 2283 inet_ntoa_r(inc->inc_laddr, sp); 2284 sp = s + strlen(s); 2285 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2286#ifdef INET6 2287 } else if (inc) { 2288 ip6_sprintf(sp, &inc->inc6_faddr); 2289 sp = s + strlen(s); 2290 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2291 sp = s + strlen(s); 2292 ip6_sprintf(sp, &inc->inc6_laddr); 2293 sp = s + strlen(s); 2294 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2295 } else if (ip6 && th) { 2296 ip6_sprintf(sp, &ip6->ip6_src); 2297 sp = s + strlen(s); 2298 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2299 sp = s + strlen(s); 2300 ip6_sprintf(sp, &ip6->ip6_dst); 2301 sp = s + strlen(s); 2302 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2303#endif /* INET6 */ 2304#ifdef INET 2305 } else if (ip && th) { 2306 inet_ntoa_r(ip->ip_src, sp); 2307 sp = s + strlen(s); 2308 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2309 sp = s + strlen(s); 2310 inet_ntoa_r(ip->ip_dst, sp); 2311 sp = s + strlen(s); 2312 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2313#endif /* INET */ 2314 } else { 2315 free(s, M_TCPLOG); 2316 return (NULL); 2317 } 2318 sp = s + strlen(s); 2319 if (th) 2320 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 2321 if (*(s + size - 1) != '\0') 2322 panic("%s: string too long", __func__); 2323 return (s); 2324} 2325