1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_compat.h" 36 #include "opt_inet.h" 37 #include "opt_inet6.h" 38 #include "opt_ipsec.h" 39 #include "opt_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 #include <netinet/tcp_offload.h> 89 #ifdef INET6 90 #include <netinet6/tcp6_var.h> 91 #endif 92 #include <netinet/tcpip.h> 93 #ifdef TCPDEBUG 94 #include <netinet/tcp_debug.h> 95 #endif 96 #ifdef INET6 97 #include <netinet6/ip6protosw.h> 98 #endif 99 100 #ifdef IPSEC 101 #include <netipsec/ipsec.h> 102 #include <netipsec/xform.h> 103 #ifdef INET6 104 #include <netipsec/ipsec6.h> 105 #endif 106 #include <netipsec/key.h> 107 #include <sys/syslog.h> 108 #endif /*IPSEC*/ 109 110 #include <machine/in_cksum.h> 111 #include <sys/md5.h> 112 113 #include <security/mac/mac_framework.h> 114 115 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 116 #ifdef INET6 117 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 118 #endif 119 120 static int 121 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 122 { 123 int error, new; 124 125 new = V_tcp_mssdflt; 126 error = sysctl_handle_int(oidp, &new, 0, req); 127 if (error == 0 && req->newptr) { 128 if (new < TCP_MINMSS) 129 error = EINVAL; 130 else 131 V_tcp_mssdflt = new; 132 } 133 return (error); 134 } 135 136 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 137 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0, 138 &sysctl_net_inet_tcp_mss_check, "I", 139 "Default TCP Maximum Segment Size"); 140 141 #ifdef INET6 142 static int 143 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 144 { 145 int error, new; 146 147 new = V_tcp_v6mssdflt; 148 error = sysctl_handle_int(oidp, &new, 0, req); 149 if (error == 0 && req->newptr) { 150 if (new < TCP_MINMSS) 151 error = EINVAL; 152 else 153 V_tcp_v6mssdflt = new; 154 } 155 return (error); 156 } 157 158 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 159 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0, 160 &sysctl_net_inet_tcp_mss_v6_check, "I", 161 "Default TCP Maximum Segment Size for IPv6"); 162 #endif /* INET6 */ 163 164 /* 165 * Minimum MSS we accept and use. This prevents DoS attacks where 166 * we are forced to a ridiculous low MSS like 20 and send hundreds 167 * of packets instead of one. The effect scales with the available 168 * bandwidth and quickly saturates the CPU and network interface 169 * with packet generation and sending. Set to zero to disable MINMSS 170 * checking. This setting prevents us from sending too small packets. 171 */ 172 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 173 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 174 &VNET_NAME(tcp_minmss), 0, 175 "Minmum TCP Maximum Segment Size"); 176 177 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 178 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 179 &VNET_NAME(tcp_do_rfc1323), 0, 180 "Enable rfc1323 (high performance TCP) extensions"); 181 182 static int tcp_log_debug = 0; 183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 184 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 185 186 static int tcp_tcbhashsize = 0; 187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 188 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 189 190 static int do_tcpdrain = 1; 191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 192 "Enable tcp_drain routine for extra help when low on mbufs"); 193 194 SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 195 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 196 197 static VNET_DEFINE(int, icmp_may_rst) = 1; 198 #define V_icmp_may_rst VNET(icmp_may_rst) 199 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, 200 &VNET_NAME(icmp_may_rst), 0, 201 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 202 203 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0; 204 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 205 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 206 &VNET_NAME(tcp_isn_reseed_interval), 0, 207 "Seconds between reseeding of ISN secret"); 208 209 #ifdef TCP_SORECEIVE_STREAM 210 static int tcp_soreceive_stream = 0; 211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 212 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 213 #endif 214 215 #ifdef TCP_SIGNATURE 216 static int tcp_sig_checksigs = 1; 217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW, 218 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic"); 219 #endif 220 221 VNET_DEFINE(uma_zone_t, sack_hole_zone); 222 #define V_sack_hole_zone VNET(sack_hole_zone) 223 224 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 225 226 static struct inpcb *tcp_notify(struct inpcb *, int); 227 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 228 void *ip4hdr, const void *ip6hdr); 229 230 /* 231 * Target size of TCP PCB hash tables. Must be a power of two. 232 * 233 * Note that this can be overridden by the kernel environment 234 * variable net.inet.tcp.tcbhashsize 235 */ 236 #ifndef TCBHASHSIZE 237 #define TCBHASHSIZE 512 238 #endif 239 240 /* 241 * XXX 242 * Callouts should be moved into struct tcp directly. They are currently 243 * separate because the tcpcb structure is exported to userland for sysctl 244 * parsing purposes, which do not know about callouts. 245 */ 246 struct tcpcb_mem { 247 struct tcpcb tcb; 248 struct tcp_timer tt; 249 struct cc_var ccv; 250 struct osd osd; 251 }; 252 253 static VNET_DEFINE(uma_zone_t, tcpcb_zone); 254 #define V_tcpcb_zone VNET(tcpcb_zone) 255 256 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 257 static struct mtx isn_mtx; 258 259 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 260 #define ISN_LOCK() mtx_lock(&isn_mtx) 261 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 262 263 /* 264 * TCP initialization. 265 */ 266 static void 267 tcp_zone_change(void *tag) 268 { 269 270 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 271 uma_zone_set_max(V_tcpcb_zone, maxsockets); 272 tcp_tw_zone_change(); 273 } 274 275 static int 276 tcp_inpcb_init(void *mem, int size, int flags) 277 { 278 struct inpcb *inp = mem; 279 280 INP_LOCK_INIT(inp, "inp", "tcpinp"); 281 return (0); 282 } 283 284 void 285 tcp_init(void) 286 { 287 int hashsize; 288 289 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 290 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 291 printf("%s: WARNING: unable to register helper hook\n", __func__); 292 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 293 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 294 printf("%s: WARNING: unable to register helper hook\n", __func__); 295 296 hashsize = TCBHASHSIZE; 297 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 298 if (!powerof2(hashsize)) { 299 printf("WARNING: TCB hash size not a power of 2\n"); 300 hashsize = 512; /* safe default */ 301 } 302 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 303 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE); 304 305 /* 306 * These have to be type stable for the benefit of the timers. 307 */ 308 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 309 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 310 uma_zone_set_max(V_tcpcb_zone, maxsockets); 311 312 tcp_tw_init(); 313 syncache_init(); 314 tcp_hc_init(); 315 tcp_reass_init(); 316 317 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 318 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 319 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 320 321 /* Skip initialization of globals for non-default instances. */ 322 if (!IS_DEFAULT_VNET(curvnet)) 323 return; 324 325 /* XXX virtualize those bellow? */ 326 tcp_delacktime = TCPTV_DELACK; 327 tcp_keepinit = TCPTV_KEEP_INIT; 328 tcp_keepidle = TCPTV_KEEP_IDLE; 329 tcp_keepintvl = TCPTV_KEEPINTVL; 330 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 331 tcp_msl = TCPTV_MSL; 332 tcp_rexmit_min = TCPTV_MIN; 333 if (tcp_rexmit_min < 1) 334 tcp_rexmit_min = 1; 335 tcp_rexmit_slop = TCPTV_CPU_VAR; 336 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 337 tcp_tcbhashsize = hashsize; 338 339 #ifdef TCP_SORECEIVE_STREAM 340 TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream); 341 if (tcp_soreceive_stream) { 342 tcp_usrreqs.pru_soreceive = soreceive_stream; 343 tcp6_usrreqs.pru_soreceive = soreceive_stream; 344 } 345 #endif 346 347 #ifdef INET6 348 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 349 #else /* INET6 */ 350 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 351 #endif /* INET6 */ 352 if (max_protohdr < TCP_MINPROTOHDR) 353 max_protohdr = TCP_MINPROTOHDR; 354 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 355 panic("tcp_init"); 356 #undef TCP_MINPROTOHDR 357 358 ISN_LOCK_INIT(); 359 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 360 SHUTDOWN_PRI_DEFAULT); 361 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 362 EVENTHANDLER_PRI_ANY); 363 } 364 365 #ifdef VIMAGE 366 void 367 tcp_destroy(void) 368 { 369 370 tcp_reass_destroy(); 371 tcp_hc_destroy(); 372 syncache_destroy(); 373 tcp_tw_destroy(); 374 in_pcbinfo_destroy(&V_tcbinfo); 375 uma_zdestroy(V_sack_hole_zone); 376 uma_zdestroy(V_tcpcb_zone); 377 } 378 #endif 379 380 void 381 tcp_fini(void *xtp) 382 { 383 384 } 385 386 /* 387 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 388 * tcp_template used to store this data in mbufs, but we now recopy it out 389 * of the tcpcb each time to conserve mbufs. 390 */ 391 void 392 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 393 { 394 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 395 396 INP_WLOCK_ASSERT(inp); 397 398 #ifdef INET6 399 if ((inp->inp_vflag & INP_IPV6) != 0) { 400 struct ip6_hdr *ip6; 401 402 ip6 = (struct ip6_hdr *)ip_ptr; 403 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 404 (inp->inp_flow & IPV6_FLOWINFO_MASK); 405 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 406 (IPV6_VERSION & IPV6_VERSION_MASK); 407 ip6->ip6_nxt = IPPROTO_TCP; 408 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 409 ip6->ip6_src = inp->in6p_laddr; 410 ip6->ip6_dst = inp->in6p_faddr; 411 } 412 #endif /* INET6 */ 413 #if defined(INET6) && defined(INET) 414 else 415 #endif 416 #ifdef INET 417 { 418 struct ip *ip; 419 420 ip = (struct ip *)ip_ptr; 421 ip->ip_v = IPVERSION; 422 ip->ip_hl = 5; 423 ip->ip_tos = inp->inp_ip_tos; 424 ip->ip_len = 0; 425 ip->ip_id = 0; 426 ip->ip_off = 0; 427 ip->ip_ttl = inp->inp_ip_ttl; 428 ip->ip_sum = 0; 429 ip->ip_p = IPPROTO_TCP; 430 ip->ip_src = inp->inp_laddr; 431 ip->ip_dst = inp->inp_faddr; 432 } 433 #endif /* INET */ 434 th->th_sport = inp->inp_lport; 435 th->th_dport = inp->inp_fport; 436 th->th_seq = 0; 437 th->th_ack = 0; 438 th->th_x2 = 0; 439 th->th_off = 5; 440 th->th_flags = 0; 441 th->th_win = 0; 442 th->th_urp = 0; 443 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 444 } 445 446 /* 447 * Create template to be used to send tcp packets on a connection. 448 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 449 * use for this function is in keepalives, which use tcp_respond. 450 */ 451 struct tcptemp * 452 tcpip_maketemplate(struct inpcb *inp) 453 { 454 struct tcptemp *t; 455 456 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 457 if (t == NULL) 458 return (NULL); 459 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 460 return (t); 461 } 462 463 /* 464 * Send a single message to the TCP at address specified by 465 * the given TCP/IP header. If m == NULL, then we make a copy 466 * of the tcpiphdr at ti and send directly to the addressed host. 467 * This is used to force keep alive messages out using the TCP 468 * template for a connection. If flags are given then we send 469 * a message back to the TCP which originated the * segment ti, 470 * and discard the mbuf containing it and any other attached mbufs. 471 * 472 * In any case the ack and sequence number of the transmitted 473 * segment are as specified by the parameters. 474 * 475 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 476 */ 477 void 478 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 479 tcp_seq ack, tcp_seq seq, int flags) 480 { 481 int tlen; 482 int win = 0; 483 struct ip *ip; 484 struct tcphdr *nth; 485 #ifdef INET6 486 struct ip6_hdr *ip6; 487 int isipv6; 488 #endif /* INET6 */ 489 int ipflags = 0; 490 struct inpcb *inp; 491 492 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 493 494 #ifdef INET6 495 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 496 ip6 = ipgen; 497 #endif /* INET6 */ 498 ip = ipgen; 499 500 if (tp != NULL) { 501 inp = tp->t_inpcb; 502 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 503 INP_WLOCK_ASSERT(inp); 504 } else 505 inp = NULL; 506 507 if (tp != NULL) { 508 if (!(flags & TH_RST)) { 509 win = sbspace(&inp->inp_socket->so_rcv); 510 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 511 win = (long)TCP_MAXWIN << tp->rcv_scale; 512 } 513 } 514 if (m == NULL) { 515 m = m_gethdr(M_DONTWAIT, MT_DATA); 516 if (m == NULL) 517 return; 518 tlen = 0; 519 m->m_data += max_linkhdr; 520 #ifdef INET6 521 if (isipv6) { 522 bcopy((caddr_t)ip6, mtod(m, caddr_t), 523 sizeof(struct ip6_hdr)); 524 ip6 = mtod(m, struct ip6_hdr *); 525 nth = (struct tcphdr *)(ip6 + 1); 526 } else 527 #endif /* INET6 */ 528 { 529 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 530 ip = mtod(m, struct ip *); 531 nth = (struct tcphdr *)(ip + 1); 532 } 533 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 534 flags = TH_ACK; 535 } else { 536 /* 537 * reuse the mbuf. 538 * XXX MRT We inherrit the FIB, which is lucky. 539 */ 540 m_freem(m->m_next); 541 m->m_next = NULL; 542 m->m_data = (caddr_t)ipgen; 543 /* m_len is set later */ 544 tlen = 0; 545 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 546 #ifdef INET6 547 if (isipv6) { 548 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 549 nth = (struct tcphdr *)(ip6 + 1); 550 } else 551 #endif /* INET6 */ 552 { 553 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 554 nth = (struct tcphdr *)(ip + 1); 555 } 556 if (th != nth) { 557 /* 558 * this is usually a case when an extension header 559 * exists between the IPv6 header and the 560 * TCP header. 561 */ 562 nth->th_sport = th->th_sport; 563 nth->th_dport = th->th_dport; 564 } 565 xchg(nth->th_dport, nth->th_sport, uint16_t); 566 #undef xchg 567 } 568 #ifdef INET6 569 if (isipv6) { 570 ip6->ip6_flow = 0; 571 ip6->ip6_vfc = IPV6_VERSION; 572 ip6->ip6_nxt = IPPROTO_TCP; 573 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 574 tlen)); 575 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 576 } 577 #endif 578 #if defined(INET) && defined(INET6) 579 else 580 #endif 581 #ifdef INET 582 { 583 tlen += sizeof (struct tcpiphdr); 584 ip->ip_len = tlen; 585 ip->ip_ttl = V_ip_defttl; 586 if (V_path_mtu_discovery) 587 ip->ip_off |= IP_DF; 588 } 589 #endif 590 m->m_len = tlen; 591 m->m_pkthdr.len = tlen; 592 m->m_pkthdr.rcvif = NULL; 593 #ifdef MAC 594 if (inp != NULL) { 595 /* 596 * Packet is associated with a socket, so allow the 597 * label of the response to reflect the socket label. 598 */ 599 INP_WLOCK_ASSERT(inp); 600 mac_inpcb_create_mbuf(inp, m); 601 } else { 602 /* 603 * Packet is not associated with a socket, so possibly 604 * update the label in place. 605 */ 606 mac_netinet_tcp_reply(m); 607 } 608 #endif 609 nth->th_seq = htonl(seq); 610 nth->th_ack = htonl(ack); 611 nth->th_x2 = 0; 612 nth->th_off = sizeof (struct tcphdr) >> 2; 613 nth->th_flags = flags; 614 if (tp != NULL) 615 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 616 else 617 nth->th_win = htons((u_short)win); 618 nth->th_urp = 0; 619 #ifdef INET6 620 if (isipv6) { 621 nth->th_sum = 0; 622 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 623 sizeof(struct ip6_hdr), 624 tlen - sizeof(struct ip6_hdr)); 625 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 626 NULL, NULL); 627 } 628 #endif /* INET6 */ 629 #if defined(INET6) && defined(INET) 630 else 631 #endif 632 #ifdef INET 633 { 634 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 635 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 636 m->m_pkthdr.csum_flags = CSUM_TCP; 637 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 638 } 639 #endif /* INET */ 640 #ifdef TCPDEBUG 641 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 642 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 643 #endif 644 #ifdef INET6 645 if (isipv6) 646 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 647 #endif /* INET6 */ 648 #if defined(INET) && defined(INET6) 649 else 650 #endif 651 #ifdef INET 652 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 653 #endif 654 } 655 656 /* 657 * Create a new TCP control block, making an 658 * empty reassembly queue and hooking it to the argument 659 * protocol control block. The `inp' parameter must have 660 * come from the zone allocator set up in tcp_init(). 661 */ 662 struct tcpcb * 663 tcp_newtcpcb(struct inpcb *inp) 664 { 665 struct tcpcb_mem *tm; 666 struct tcpcb *tp; 667 #ifdef INET6 668 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 669 #endif /* INET6 */ 670 671 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 672 if (tm == NULL) 673 return (NULL); 674 tp = &tm->tcb; 675 676 /* Initialise cc_var struct for this tcpcb. */ 677 tp->ccv = &tm->ccv; 678 tp->ccv->type = IPPROTO_TCP; 679 tp->ccv->ccvc.tcp = tp; 680 681 /* 682 * Use the current system default CC algorithm. 683 */ 684 CC_LIST_RLOCK(); 685 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 686 CC_ALGO(tp) = CC_DEFAULT(); 687 CC_LIST_RUNLOCK(); 688 689 if (CC_ALGO(tp)->cb_init != NULL) 690 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 691 uma_zfree(V_tcpcb_zone, tm); 692 return (NULL); 693 } 694 695 tp->osd = &tm->osd; 696 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 697 uma_zfree(V_tcpcb_zone, tm); 698 return (NULL); 699 } 700 701 #ifdef VIMAGE 702 tp->t_vnet = inp->inp_vnet; 703 #endif 704 tp->t_timers = &tm->tt; 705 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 706 tp->t_maxseg = tp->t_maxopd = 707 #ifdef INET6 708 isipv6 ? V_tcp_v6mssdflt : 709 #endif /* INET6 */ 710 V_tcp_mssdflt; 711 712 /* Set up our timeouts. */ 713 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE); 714 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE); 715 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE); 716 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE); 717 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE); 718 719 if (V_tcp_do_rfc1323) 720 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 721 if (V_tcp_do_sack) 722 tp->t_flags |= TF_SACK_PERMIT; 723 TAILQ_INIT(&tp->snd_holes); 724 tp->t_inpcb = inp; /* XXX */ 725 /* 726 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 727 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 728 * reasonable initial retransmit time. 729 */ 730 tp->t_srtt = TCPTV_SRTTBASE; 731 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 732 tp->t_rttmin = tcp_rexmit_min; 733 tp->t_rxtcur = TCPTV_RTOBASE; 734 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 735 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 736 tp->t_rcvtime = ticks; 737 /* 738 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 739 * because the socket may be bound to an IPv6 wildcard address, 740 * which may match an IPv4-mapped IPv6 address. 741 */ 742 inp->inp_ip_ttl = V_ip_defttl; 743 inp->inp_ppcb = tp; 744 return (tp); /* XXX */ 745 } 746 747 /* 748 * Switch the congestion control algorithm back to NewReno for any active 749 * control blocks using an algorithm which is about to go away. 750 * This ensures the CC framework can allow the unload to proceed without leaving 751 * any dangling pointers which would trigger a panic. 752 * Returning non-zero would inform the CC framework that something went wrong 753 * and it would be unsafe to allow the unload to proceed. However, there is no 754 * way for this to occur with this implementation so we always return zero. 755 */ 756 int 757 tcp_ccalgounload(struct cc_algo *unload_algo) 758 { 759 struct cc_algo *tmpalgo; 760 struct inpcb *inp; 761 struct tcpcb *tp; 762 VNET_ITERATOR_DECL(vnet_iter); 763 764 /* 765 * Check all active control blocks across all network stacks and change 766 * any that are using "unload_algo" back to NewReno. If "unload_algo" 767 * requires cleanup code to be run, call it. 768 */ 769 VNET_LIST_RLOCK(); 770 VNET_FOREACH(vnet_iter) { 771 CURVNET_SET(vnet_iter); 772 INP_INFO_RLOCK(&V_tcbinfo); 773 /* 774 * New connections already part way through being initialised 775 * with the CC algo we're removing will not race with this code 776 * because the INP_INFO_WLOCK is held during initialisation. We 777 * therefore don't enter the loop below until the connection 778 * list has stabilised. 779 */ 780 LIST_FOREACH(inp, &V_tcb, inp_list) { 781 INP_WLOCK(inp); 782 /* Important to skip tcptw structs. */ 783 if (!(inp->inp_flags & INP_TIMEWAIT) && 784 (tp = intotcpcb(inp)) != NULL) { 785 /* 786 * By holding INP_WLOCK here, we are assured 787 * that the connection is not currently 788 * executing inside the CC module's functions 789 * i.e. it is safe to make the switch back to 790 * NewReno. 791 */ 792 if (CC_ALGO(tp) == unload_algo) { 793 tmpalgo = CC_ALGO(tp); 794 /* NewReno does not require any init. */ 795 CC_ALGO(tp) = &newreno_cc_algo; 796 if (tmpalgo->cb_destroy != NULL) 797 tmpalgo->cb_destroy(tp->ccv); 798 } 799 } 800 INP_WUNLOCK(inp); 801 } 802 INP_INFO_RUNLOCK(&V_tcbinfo); 803 CURVNET_RESTORE(); 804 } 805 VNET_LIST_RUNLOCK(); 806 807 return (0); 808 } 809 810 /* 811 * Drop a TCP connection, reporting 812 * the specified error. If connection is synchronized, 813 * then send a RST to peer. 814 */ 815 struct tcpcb * 816 tcp_drop(struct tcpcb *tp, int errno) 817 { 818 struct socket *so = tp->t_inpcb->inp_socket; 819 820 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 821 INP_WLOCK_ASSERT(tp->t_inpcb); 822 823 if (TCPS_HAVERCVDSYN(tp->t_state)) { 824 tp->t_state = TCPS_CLOSED; 825 (void) tcp_output_reset(tp); 826 TCPSTAT_INC(tcps_drops); 827 } else 828 TCPSTAT_INC(tcps_conndrops); 829 if (errno == ETIMEDOUT && tp->t_softerror) 830 errno = tp->t_softerror; 831 so->so_error = errno; 832 return (tcp_close(tp)); 833 } 834 835 void 836 tcp_discardcb(struct tcpcb *tp) 837 { 838 struct inpcb *inp = tp->t_inpcb; 839 struct socket *so = inp->inp_socket; 840 #ifdef INET6 841 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 842 #endif /* INET6 */ 843 844 INP_WLOCK_ASSERT(inp); 845 846 /* 847 * Make sure that all of our timers are stopped before we delete the 848 * PCB. 849 * 850 * XXXRW: Really, we would like to use callout_drain() here in order 851 * to avoid races experienced in tcp_timer.c where a timer is already 852 * executing at this point. However, we can't, both because we're 853 * running in a context where we can't sleep, and also because we 854 * hold locks required by the timers. What we instead need to do is 855 * test to see if callout_drain() is required, and if so, defer some 856 * portion of the remainder of tcp_discardcb() to an asynchronous 857 * context that can callout_drain() and then continue. Some care 858 * will be required to ensure that no further processing takes place 859 * on the tcpcb, even though it hasn't been freed (a flag?). 860 */ 861 callout_stop(&tp->t_timers->tt_rexmt); 862 callout_stop(&tp->t_timers->tt_persist); 863 callout_stop(&tp->t_timers->tt_keep); 864 callout_stop(&tp->t_timers->tt_2msl); 865 callout_stop(&tp->t_timers->tt_delack); 866 867 /* 868 * If we got enough samples through the srtt filter, 869 * save the rtt and rttvar in the routing entry. 870 * 'Enough' is arbitrarily defined as 4 rtt samples. 871 * 4 samples is enough for the srtt filter to converge 872 * to within enough % of the correct value; fewer samples 873 * and we could save a bogus rtt. The danger is not high 874 * as tcp quickly recovers from everything. 875 * XXX: Works very well but needs some more statistics! 876 */ 877 if (tp->t_rttupdated >= 4) { 878 struct hc_metrics_lite metrics; 879 u_long ssthresh; 880 881 bzero(&metrics, sizeof(metrics)); 882 /* 883 * Update the ssthresh always when the conditions below 884 * are satisfied. This gives us better new start value 885 * for the congestion avoidance for new connections. 886 * ssthresh is only set if packet loss occured on a session. 887 * 888 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 889 * being torn down. Ideally this code would not use 'so'. 890 */ 891 ssthresh = tp->snd_ssthresh; 892 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 893 /* 894 * convert the limit from user data bytes to 895 * packets then to packet data bytes. 896 */ 897 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 898 if (ssthresh < 2) 899 ssthresh = 2; 900 ssthresh *= (u_long)(tp->t_maxseg + 901 #ifdef INET6 902 (isipv6 ? sizeof (struct ip6_hdr) + 903 sizeof (struct tcphdr) : 904 #endif 905 sizeof (struct tcpiphdr) 906 #ifdef INET6 907 ) 908 #endif 909 ); 910 } else 911 ssthresh = 0; 912 metrics.rmx_ssthresh = ssthresh; 913 914 metrics.rmx_rtt = tp->t_srtt; 915 metrics.rmx_rttvar = tp->t_rttvar; 916 metrics.rmx_cwnd = tp->snd_cwnd; 917 metrics.rmx_sendpipe = 0; 918 metrics.rmx_recvpipe = 0; 919 920 tcp_hc_update(&inp->inp_inc, &metrics); 921 } 922 923 /* free the reassembly queue, if any */ 924 tcp_reass_flush(tp); 925 /* Disconnect offload device, if any. */ 926 tcp_offload_detach(tp); 927 928 tcp_free_sackholes(tp); 929 930 /* Allow the CC algorithm to clean up after itself. */ 931 if (CC_ALGO(tp)->cb_destroy != NULL) 932 CC_ALGO(tp)->cb_destroy(tp->ccv); 933 934 khelp_destroy_osd(tp->osd); 935 936 CC_ALGO(tp) = NULL; 937 inp->inp_ppcb = NULL; 938 tp->t_inpcb = NULL; 939 uma_zfree(V_tcpcb_zone, tp); 940 } 941 942 /* 943 * Attempt to close a TCP control block, marking it as dropped, and freeing 944 * the socket if we hold the only reference. 945 */ 946 struct tcpcb * 947 tcp_close(struct tcpcb *tp) 948 { 949 struct inpcb *inp = tp->t_inpcb; 950 struct socket *so; 951 952 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 953 INP_WLOCK_ASSERT(inp); 954 955 /* Notify any offload devices of listener close */ 956 if (tp->t_state == TCPS_LISTEN) 957 tcp_offload_listen_close(tp); 958 in_pcbdrop(inp); 959 TCPSTAT_INC(tcps_closed); 960 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 961 so = inp->inp_socket; 962 soisdisconnected(so); 963 if (inp->inp_flags & INP_SOCKREF) { 964 KASSERT(so->so_state & SS_PROTOREF, 965 ("tcp_close: !SS_PROTOREF")); 966 inp->inp_flags &= ~INP_SOCKREF; 967 INP_WUNLOCK(inp); 968 ACCEPT_LOCK(); 969 SOCK_LOCK(so); 970 so->so_state &= ~SS_PROTOREF; 971 sofree(so); 972 return (NULL); 973 } 974 return (tp); 975 } 976 977 void 978 tcp_drain(void) 979 { 980 VNET_ITERATOR_DECL(vnet_iter); 981 982 if (!do_tcpdrain) 983 return; 984 985 VNET_LIST_RLOCK_NOSLEEP(); 986 VNET_FOREACH(vnet_iter) { 987 CURVNET_SET(vnet_iter); 988 struct inpcb *inpb; 989 struct tcpcb *tcpb; 990 991 /* 992 * Walk the tcpbs, if existing, and flush the reassembly queue, 993 * if there is one... 994 * XXX: The "Net/3" implementation doesn't imply that the TCP 995 * reassembly queue should be flushed, but in a situation 996 * where we're really low on mbufs, this is potentially 997 * usefull. 998 */ 999 INP_INFO_RLOCK(&V_tcbinfo); 1000 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 1001 if (inpb->inp_flags & INP_TIMEWAIT) 1002 continue; 1003 INP_WLOCK(inpb); 1004 if ((tcpb = intotcpcb(inpb)) != NULL) { 1005 tcp_reass_flush(tcpb); 1006 tcp_clean_sackreport(tcpb); 1007 } 1008 INP_WUNLOCK(inpb); 1009 } 1010 INP_INFO_RUNLOCK(&V_tcbinfo); 1011 CURVNET_RESTORE(); 1012 } 1013 VNET_LIST_RUNLOCK_NOSLEEP(); 1014 } 1015 1016 /* 1017 * Notify a tcp user of an asynchronous error; 1018 * store error as soft error, but wake up user 1019 * (for now, won't do anything until can select for soft error). 1020 * 1021 * Do not wake up user since there currently is no mechanism for 1022 * reporting soft errors (yet - a kqueue filter may be added). 1023 */ 1024 static struct inpcb * 1025 tcp_notify(struct inpcb *inp, int error) 1026 { 1027 struct tcpcb *tp; 1028 1029 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1030 INP_WLOCK_ASSERT(inp); 1031 1032 if ((inp->inp_flags & INP_TIMEWAIT) || 1033 (inp->inp_flags & INP_DROPPED)) 1034 return (inp); 1035 1036 tp = intotcpcb(inp); 1037 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 1038 1039 /* 1040 * Ignore some errors if we are hooked up. 1041 * If connection hasn't completed, has retransmitted several times, 1042 * and receives a second error, give up now. This is better 1043 * than waiting a long time to establish a connection that 1044 * can never complete. 1045 */ 1046 if (tp->t_state == TCPS_ESTABLISHED && 1047 (error == EHOSTUNREACH || error == ENETUNREACH || 1048 error == EHOSTDOWN)) { 1049 return (inp); 1050 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 1051 tp->t_softerror) { 1052 tp = tcp_drop(tp, error); 1053 if (tp != NULL) 1054 return (inp); 1055 else 1056 return (NULL); 1057 } else { 1058 tp->t_softerror = error; 1059 return (inp); 1060 } 1061 #if 0 1062 wakeup( &so->so_timeo); 1063 sorwakeup(so); 1064 sowwakeup(so); 1065 #endif 1066 } 1067 1068 static int 1069 tcp_pcblist(SYSCTL_HANDLER_ARGS) 1070 { 1071 int error, i, m, n, pcb_count; 1072 struct inpcb *inp, **inp_list; 1073 inp_gen_t gencnt; 1074 struct xinpgen xig; 1075 1076 /* 1077 * The process of preparing the TCB list is too time-consuming and 1078 * resource-intensive to repeat twice on every request. 1079 */ 1080 if (req->oldptr == NULL) { 1081 n = V_tcbinfo.ipi_count + syncache_pcbcount(); 1082 n += imax(n / 8, 10); 1083 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 1084 return (0); 1085 } 1086 1087 if (req->newptr != NULL) 1088 return (EPERM); 1089 1090 /* 1091 * OK, now we're committed to doing something. 1092 */ 1093 INP_INFO_RLOCK(&V_tcbinfo); 1094 gencnt = V_tcbinfo.ipi_gencnt; 1095 n = V_tcbinfo.ipi_count; 1096 INP_INFO_RUNLOCK(&V_tcbinfo); 1097 1098 m = syncache_pcbcount(); 1099 1100 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 1101 + (n + m) * sizeof(struct xtcpcb)); 1102 if (error != 0) 1103 return (error); 1104 1105 xig.xig_len = sizeof xig; 1106 xig.xig_count = n + m; 1107 xig.xig_gen = gencnt; 1108 xig.xig_sogen = so_gencnt; 1109 error = SYSCTL_OUT(req, &xig, sizeof xig); 1110 if (error) 1111 return (error); 1112 1113 error = syncache_pcblist(req, m, &pcb_count); 1114 if (error) 1115 return (error); 1116 1117 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 1118 if (inp_list == NULL) 1119 return (ENOMEM); 1120 1121 INP_INFO_RLOCK(&V_tcbinfo); 1122 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; 1123 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) { 1124 INP_WLOCK(inp); 1125 if (inp->inp_gencnt <= gencnt) { 1126 /* 1127 * XXX: This use of cr_cansee(), introduced with 1128 * TCP state changes, is not quite right, but for 1129 * now, better than nothing. 1130 */ 1131 if (inp->inp_flags & INP_TIMEWAIT) { 1132 if (intotw(inp) != NULL) 1133 error = cr_cansee(req->td->td_ucred, 1134 intotw(inp)->tw_cred); 1135 else 1136 error = EINVAL; /* Skip this inp. */ 1137 } else 1138 error = cr_canseeinpcb(req->td->td_ucred, inp); 1139 if (error == 0) { 1140 in_pcbref(inp); 1141 inp_list[i++] = inp; 1142 } 1143 } 1144 INP_WUNLOCK(inp); 1145 } 1146 INP_INFO_RUNLOCK(&V_tcbinfo); 1147 n = i; 1148 1149 error = 0; 1150 for (i = 0; i < n; i++) { 1151 inp = inp_list[i]; 1152 INP_RLOCK(inp); 1153 if (inp->inp_gencnt <= gencnt) { 1154 struct xtcpcb xt; 1155 void *inp_ppcb; 1156 1157 bzero(&xt, sizeof(xt)); 1158 xt.xt_len = sizeof xt; 1159 /* XXX should avoid extra copy */ 1160 bcopy(inp, &xt.xt_inp, sizeof *inp); 1161 inp_ppcb = inp->inp_ppcb; 1162 if (inp_ppcb == NULL) 1163 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1164 else if (inp->inp_flags & INP_TIMEWAIT) { 1165 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1166 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1167 } else { 1168 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1169 if (xt.xt_tp.t_timers) 1170 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer); 1171 } 1172 if (inp->inp_socket != NULL) 1173 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1174 else { 1175 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1176 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1177 } 1178 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1179 INP_RUNLOCK(inp); 1180 error = SYSCTL_OUT(req, &xt, sizeof xt); 1181 } else 1182 INP_RUNLOCK(inp); 1183 } 1184 INP_INFO_WLOCK(&V_tcbinfo); 1185 for (i = 0; i < n; i++) { 1186 inp = inp_list[i]; 1187 INP_WLOCK(inp); 1188 if (!in_pcbrele(inp)) 1189 INP_WUNLOCK(inp); 1190 } 1191 INP_INFO_WUNLOCK(&V_tcbinfo); 1192 1193 if (!error) { 1194 /* 1195 * Give the user an updated idea of our state. 1196 * If the generation differs from what we told 1197 * her before, she knows that something happened 1198 * while we were processing this request, and it 1199 * might be necessary to retry. 1200 */ 1201 INP_INFO_RLOCK(&V_tcbinfo); 1202 xig.xig_gen = V_tcbinfo.ipi_gencnt; 1203 xig.xig_sogen = so_gencnt; 1204 xig.xig_count = V_tcbinfo.ipi_count + pcb_count; 1205 INP_INFO_RUNLOCK(&V_tcbinfo); 1206 error = SYSCTL_OUT(req, &xig, sizeof xig); 1207 } 1208 free(inp_list, M_TEMP); 1209 return (error); 1210 } 1211 1212 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 1213 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 1214 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1215 1216 #ifdef INET 1217 static int 1218 tcp_getcred(SYSCTL_HANDLER_ARGS) 1219 { 1220 struct xucred xuc; 1221 struct sockaddr_in addrs[2]; 1222 struct inpcb *inp; 1223 int error; 1224 1225 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1226 if (error) 1227 return (error); 1228 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1229 if (error) 1230 return (error); 1231 INP_INFO_RLOCK(&V_tcbinfo); 1232 inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr, 1233 addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1234 if (inp != NULL) { 1235 INP_RLOCK(inp); 1236 INP_INFO_RUNLOCK(&V_tcbinfo); 1237 if (inp->inp_socket == NULL) 1238 error = ENOENT; 1239 if (error == 0) 1240 error = cr_canseeinpcb(req->td->td_ucred, inp); 1241 if (error == 0) 1242 cru2x(inp->inp_cred, &xuc); 1243 INP_RUNLOCK(inp); 1244 } else { 1245 INP_INFO_RUNLOCK(&V_tcbinfo); 1246 error = ENOENT; 1247 } 1248 if (error == 0) 1249 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1250 return (error); 1251 } 1252 1253 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1254 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1255 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1256 #endif /* INET */ 1257 1258 #ifdef INET6 1259 static int 1260 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1261 { 1262 struct xucred xuc; 1263 struct sockaddr_in6 addrs[2]; 1264 struct inpcb *inp; 1265 int error; 1266 #ifdef INET 1267 int mapped = 0; 1268 #endif 1269 1270 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1271 if (error) 1272 return (error); 1273 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1274 if (error) 1275 return (error); 1276 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 1277 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 1278 return (error); 1279 } 1280 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1281 #ifdef INET 1282 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1283 mapped = 1; 1284 else 1285 #endif 1286 return (EINVAL); 1287 } 1288 1289 INP_INFO_RLOCK(&V_tcbinfo); 1290 #ifdef INET 1291 if (mapped == 1) 1292 inp = in_pcblookup_hash(&V_tcbinfo, 1293 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1294 addrs[1].sin6_port, 1295 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1296 addrs[0].sin6_port, 1297 0, NULL); 1298 else 1299 #endif 1300 inp = in6_pcblookup_hash(&V_tcbinfo, 1301 &addrs[1].sin6_addr, addrs[1].sin6_port, 1302 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1303 if (inp != NULL) { 1304 INP_RLOCK(inp); 1305 INP_INFO_RUNLOCK(&V_tcbinfo); 1306 if (inp->inp_socket == NULL) 1307 error = ENOENT; 1308 if (error == 0) 1309 error = cr_canseeinpcb(req->td->td_ucred, inp); 1310 if (error == 0) 1311 cru2x(inp->inp_cred, &xuc); 1312 INP_RUNLOCK(inp); 1313 } else { 1314 INP_INFO_RUNLOCK(&V_tcbinfo); 1315 error = ENOENT; 1316 } 1317 if (error == 0) 1318 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1319 return (error); 1320 } 1321 1322 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1323 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1324 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1325 #endif /* INET6 */ 1326 1327 1328 #ifdef INET 1329 void 1330 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1331 { 1332 struct ip *ip = vip; 1333 struct tcphdr *th; 1334 struct in_addr faddr; 1335 struct inpcb *inp; 1336 struct tcpcb *tp; 1337 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1338 struct icmp *icp; 1339 struct in_conninfo inc; 1340 tcp_seq icmp_tcp_seq; 1341 int mtu; 1342 1343 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1344 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1345 return; 1346 1347 if (cmd == PRC_MSGSIZE) 1348 notify = tcp_mtudisc; 1349 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1350 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1351 notify = tcp_drop_syn_sent; 1352 /* 1353 * Redirects don't need to be handled up here. 1354 */ 1355 else if (PRC_IS_REDIRECT(cmd)) 1356 return; 1357 /* 1358 * Source quench is depreciated. 1359 */ 1360 else if (cmd == PRC_QUENCH) 1361 return; 1362 /* 1363 * Hostdead is ugly because it goes linearly through all PCBs. 1364 * XXX: We never get this from ICMP, otherwise it makes an 1365 * excellent DoS attack on machines with many connections. 1366 */ 1367 else if (cmd == PRC_HOSTDEAD) 1368 ip = NULL; 1369 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1370 return; 1371 if (ip != NULL) { 1372 icp = (struct icmp *)((caddr_t)ip 1373 - offsetof(struct icmp, icmp_ip)); 1374 th = (struct tcphdr *)((caddr_t)ip 1375 + (ip->ip_hl << 2)); 1376 INP_INFO_WLOCK(&V_tcbinfo); 1377 inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport, 1378 ip->ip_src, th->th_sport, 0, NULL); 1379 if (inp != NULL) { 1380 INP_WLOCK(inp); 1381 if (!(inp->inp_flags & INP_TIMEWAIT) && 1382 !(inp->inp_flags & INP_DROPPED) && 1383 !(inp->inp_socket == NULL)) { 1384 icmp_tcp_seq = htonl(th->th_seq); 1385 tp = intotcpcb(inp); 1386 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1387 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1388 if (cmd == PRC_MSGSIZE) { 1389 /* 1390 * MTU discovery: 1391 * If we got a needfrag set the MTU 1392 * in the route to the suggested new 1393 * value (if given) and then notify. 1394 */ 1395 bzero(&inc, sizeof(inc)); 1396 inc.inc_faddr = faddr; 1397 inc.inc_fibnum = 1398 inp->inp_inc.inc_fibnum; 1399 1400 mtu = ntohs(icp->icmp_nextmtu); 1401 /* 1402 * If no alternative MTU was 1403 * proposed, try the next smaller 1404 * one. ip->ip_len has already 1405 * been swapped in icmp_input(). 1406 */ 1407 if (!mtu) 1408 mtu = ip_next_mtu(ip->ip_len, 1409 1); 1410 if (mtu < V_tcp_minmss 1411 + sizeof(struct tcpiphdr)) 1412 mtu = V_tcp_minmss 1413 + sizeof(struct tcpiphdr); 1414 /* 1415 * Only cache the MTU if it 1416 * is smaller than the interface 1417 * or route MTU. tcp_mtudisc() 1418 * will do right thing by itself. 1419 */ 1420 if (mtu <= tcp_maxmtu(&inc, NULL)) 1421 tcp_hc_updatemtu(&inc, mtu); 1422 } 1423 1424 inp = (*notify)(inp, inetctlerrmap[cmd]); 1425 } 1426 } 1427 if (inp != NULL) 1428 INP_WUNLOCK(inp); 1429 } else { 1430 bzero(&inc, sizeof(inc)); 1431 inc.inc_fport = th->th_dport; 1432 inc.inc_lport = th->th_sport; 1433 inc.inc_faddr = faddr; 1434 inc.inc_laddr = ip->ip_src; 1435 syncache_unreach(&inc, th); 1436 } 1437 INP_INFO_WUNLOCK(&V_tcbinfo); 1438 } else 1439 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 1440 } 1441 #endif /* INET */ 1442 1443 #ifdef INET6 1444 void 1445 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1446 { 1447 struct tcphdr th; 1448 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1449 struct ip6_hdr *ip6; 1450 struct mbuf *m; 1451 struct ip6ctlparam *ip6cp = NULL; 1452 const struct sockaddr_in6 *sa6_src = NULL; 1453 int off; 1454 struct tcp_portonly { 1455 u_int16_t th_sport; 1456 u_int16_t th_dport; 1457 } *thp; 1458 1459 if (sa->sa_family != AF_INET6 || 1460 sa->sa_len != sizeof(struct sockaddr_in6)) 1461 return; 1462 1463 if (cmd == PRC_MSGSIZE) 1464 notify = tcp_mtudisc; 1465 else if (!PRC_IS_REDIRECT(cmd) && 1466 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1467 return; 1468 /* Source quench is depreciated. */ 1469 else if (cmd == PRC_QUENCH) 1470 return; 1471 1472 /* if the parameter is from icmp6, decode it. */ 1473 if (d != NULL) { 1474 ip6cp = (struct ip6ctlparam *)d; 1475 m = ip6cp->ip6c_m; 1476 ip6 = ip6cp->ip6c_ip6; 1477 off = ip6cp->ip6c_off; 1478 sa6_src = ip6cp->ip6c_src; 1479 } else { 1480 m = NULL; 1481 ip6 = NULL; 1482 off = 0; /* fool gcc */ 1483 sa6_src = &sa6_any; 1484 } 1485 1486 if (ip6 != NULL) { 1487 struct in_conninfo inc; 1488 /* 1489 * XXX: We assume that when IPV6 is non NULL, 1490 * M and OFF are valid. 1491 */ 1492 1493 /* check if we can safely examine src and dst ports */ 1494 if (m->m_pkthdr.len < off + sizeof(*thp)) 1495 return; 1496 1497 bzero(&th, sizeof(th)); 1498 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1499 1500 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport, 1501 (struct sockaddr *)ip6cp->ip6c_src, 1502 th.th_sport, cmd, NULL, notify); 1503 1504 bzero(&inc, sizeof(inc)); 1505 inc.inc_fport = th.th_dport; 1506 inc.inc_lport = th.th_sport; 1507 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1508 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1509 inc.inc_flags |= INC_ISIPV6; 1510 INP_INFO_WLOCK(&V_tcbinfo); 1511 syncache_unreach(&inc, &th); 1512 INP_INFO_WUNLOCK(&V_tcbinfo); 1513 } else 1514 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1515 0, cmd, NULL, notify); 1516 } 1517 #endif /* INET6 */ 1518 1519 1520 /* 1521 * Following is where TCP initial sequence number generation occurs. 1522 * 1523 * There are two places where we must use initial sequence numbers: 1524 * 1. In SYN-ACK packets. 1525 * 2. In SYN packets. 1526 * 1527 * All ISNs for SYN-ACK packets are generated by the syncache. See 1528 * tcp_syncache.c for details. 1529 * 1530 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1531 * depends on this property. In addition, these ISNs should be 1532 * unguessable so as to prevent connection hijacking. To satisfy 1533 * the requirements of this situation, the algorithm outlined in 1534 * RFC 1948 is used, with only small modifications. 1535 * 1536 * Implementation details: 1537 * 1538 * Time is based off the system timer, and is corrected so that it 1539 * increases by one megabyte per second. This allows for proper 1540 * recycling on high speed LANs while still leaving over an hour 1541 * before rollover. 1542 * 1543 * As reading the *exact* system time is too expensive to be done 1544 * whenever setting up a TCP connection, we increment the time 1545 * offset in two ways. First, a small random positive increment 1546 * is added to isn_offset for each connection that is set up. 1547 * Second, the function tcp_isn_tick fires once per clock tick 1548 * and increments isn_offset as necessary so that sequence numbers 1549 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1550 * random positive increments serve only to ensure that the same 1551 * exact sequence number is never sent out twice (as could otherwise 1552 * happen when a port is recycled in less than the system tick 1553 * interval.) 1554 * 1555 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1556 * between seeding of isn_secret. This is normally set to zero, 1557 * as reseeding should not be necessary. 1558 * 1559 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1560 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1561 * general, this means holding an exclusive (write) lock. 1562 */ 1563 1564 #define ISN_BYTES_PER_SECOND 1048576 1565 #define ISN_STATIC_INCREMENT 4096 1566 #define ISN_RANDOM_INCREMENT (4096 - 1) 1567 1568 static VNET_DEFINE(u_char, isn_secret[32]); 1569 static VNET_DEFINE(int, isn_last); 1570 static VNET_DEFINE(int, isn_last_reseed); 1571 static VNET_DEFINE(u_int32_t, isn_offset); 1572 static VNET_DEFINE(u_int32_t, isn_offset_old); 1573 1574 #define V_isn_secret VNET(isn_secret) 1575 #define V_isn_last VNET(isn_last) 1576 #define V_isn_last_reseed VNET(isn_last_reseed) 1577 #define V_isn_offset VNET(isn_offset) 1578 #define V_isn_offset_old VNET(isn_offset_old) 1579 1580 tcp_seq 1581 tcp_new_isn(struct tcpcb *tp) 1582 { 1583 MD5_CTX isn_ctx; 1584 u_int32_t md5_buffer[4]; 1585 tcp_seq new_isn; 1586 u_int32_t projected_offset; 1587 1588 INP_WLOCK_ASSERT(tp->t_inpcb); 1589 1590 ISN_LOCK(); 1591 /* Seed if this is the first use, reseed if requested. */ 1592 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 1593 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 1594 < (u_int)ticks))) { 1595 read_random(&V_isn_secret, sizeof(V_isn_secret)); 1596 V_isn_last_reseed = ticks; 1597 } 1598 1599 /* Compute the md5 hash and return the ISN. */ 1600 MD5Init(&isn_ctx); 1601 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1602 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1603 #ifdef INET6 1604 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1605 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1606 sizeof(struct in6_addr)); 1607 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1608 sizeof(struct in6_addr)); 1609 } else 1610 #endif 1611 { 1612 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1613 sizeof(struct in_addr)); 1614 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1615 sizeof(struct in_addr)); 1616 } 1617 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret)); 1618 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1619 new_isn = (tcp_seq) md5_buffer[0]; 1620 V_isn_offset += ISN_STATIC_INCREMENT + 1621 (arc4random() & ISN_RANDOM_INCREMENT); 1622 if (ticks != V_isn_last) { 1623 projected_offset = V_isn_offset_old + 1624 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 1625 if (SEQ_GT(projected_offset, V_isn_offset)) 1626 V_isn_offset = projected_offset; 1627 V_isn_offset_old = V_isn_offset; 1628 V_isn_last = ticks; 1629 } 1630 new_isn += V_isn_offset; 1631 ISN_UNLOCK(); 1632 return (new_isn); 1633 } 1634 1635 /* 1636 * When a specific ICMP unreachable message is received and the 1637 * connection state is SYN-SENT, drop the connection. This behavior 1638 * is controlled by the icmp_may_rst sysctl. 1639 */ 1640 struct inpcb * 1641 tcp_drop_syn_sent(struct inpcb *inp, int errno) 1642 { 1643 struct tcpcb *tp; 1644 1645 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1646 INP_WLOCK_ASSERT(inp); 1647 1648 if ((inp->inp_flags & INP_TIMEWAIT) || 1649 (inp->inp_flags & INP_DROPPED)) 1650 return (inp); 1651 1652 tp = intotcpcb(inp); 1653 if (tp->t_state != TCPS_SYN_SENT) 1654 return (inp); 1655 1656 tp = tcp_drop(tp, errno); 1657 if (tp != NULL) 1658 return (inp); 1659 else 1660 return (NULL); 1661 } 1662 1663 /* 1664 * When `need fragmentation' ICMP is received, update our idea of the MSS 1665 * based on the new value in the route. Also nudge TCP to send something, 1666 * since we know the packet we just sent was dropped. 1667 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1668 */ 1669 struct inpcb * 1670 tcp_mtudisc(struct inpcb *inp, int errno) 1671 { 1672 struct tcpcb *tp; 1673 struct socket *so; 1674 1675 INP_WLOCK_ASSERT(inp); 1676 if ((inp->inp_flags & INP_TIMEWAIT) || 1677 (inp->inp_flags & INP_DROPPED)) 1678 return (inp); 1679 1680 tp = intotcpcb(inp); 1681 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1682 1683 tcp_mss_update(tp, -1, NULL, NULL); 1684 1685 so = inp->inp_socket; 1686 SOCKBUF_LOCK(&so->so_snd); 1687 /* If the mss is larger than the socket buffer, decrease the mss. */ 1688 if (so->so_snd.sb_hiwat < tp->t_maxseg) 1689 tp->t_maxseg = so->so_snd.sb_hiwat; 1690 SOCKBUF_UNLOCK(&so->so_snd); 1691 1692 TCPSTAT_INC(tcps_mturesent); 1693 tp->t_rtttime = 0; 1694 tp->snd_nxt = tp->snd_una; 1695 tcp_free_sackholes(tp); 1696 tp->snd_recover = tp->snd_max; 1697 if (tp->t_flags & TF_SACK_PERMIT) 1698 EXIT_FASTRECOVERY(tp->t_flags); 1699 tcp_output_send(tp); 1700 return (inp); 1701 } 1702 1703 #ifdef INET 1704 /* 1705 * Look-up the routing entry to the peer of this inpcb. If no route 1706 * is found and it cannot be allocated, then return 0. This routine 1707 * is called by TCP routines that access the rmx structure and by 1708 * tcp_mss_update to get the peer/interface MTU. 1709 */ 1710 u_long 1711 tcp_maxmtu(struct in_conninfo *inc, int *flags) 1712 { 1713 struct route sro; 1714 struct sockaddr_in *dst; 1715 struct ifnet *ifp; 1716 u_long maxmtu = 0; 1717 1718 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1719 1720 bzero(&sro, sizeof(sro)); 1721 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1722 dst = (struct sockaddr_in *)&sro.ro_dst; 1723 dst->sin_family = AF_INET; 1724 dst->sin_len = sizeof(*dst); 1725 dst->sin_addr = inc->inc_faddr; 1726 in_rtalloc_ign(&sro, 0, inc->inc_fibnum); 1727 } 1728 if (sro.ro_rt != NULL) { 1729 ifp = sro.ro_rt->rt_ifp; 1730 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1731 maxmtu = ifp->if_mtu; 1732 else 1733 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1734 1735 /* Report additional interface capabilities. */ 1736 if (flags != NULL) { 1737 if (ifp->if_capenable & IFCAP_TSO4 && 1738 ifp->if_hwassist & CSUM_TSO) 1739 *flags |= CSUM_TSO; 1740 } 1741 RTFREE(sro.ro_rt); 1742 } 1743 return (maxmtu); 1744 } 1745 #endif /* INET */ 1746 1747 #ifdef INET6 1748 u_long 1749 tcp_maxmtu6(struct in_conninfo *inc, int *flags) 1750 { 1751 struct route_in6 sro6; 1752 struct ifnet *ifp; 1753 u_long maxmtu = 0; 1754 1755 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1756 1757 bzero(&sro6, sizeof(sro6)); 1758 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1759 sro6.ro_dst.sin6_family = AF_INET6; 1760 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1761 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1762 rtalloc_ign((struct route *)&sro6, 0); 1763 } 1764 if (sro6.ro_rt != NULL) { 1765 ifp = sro6.ro_rt->rt_ifp; 1766 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1767 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1768 else 1769 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1770 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1771 1772 /* Report additional interface capabilities. */ 1773 if (flags != NULL) { 1774 if (ifp->if_capenable & IFCAP_TSO6 && 1775 ifp->if_hwassist & CSUM_TSO) 1776 *flags |= CSUM_TSO; 1777 } 1778 RTFREE(sro6.ro_rt); 1779 } 1780 1781 return (maxmtu); 1782 } 1783 #endif /* INET6 */ 1784 1785 #ifdef IPSEC 1786 /* compute ESP/AH header size for TCP, including outer IP header. */ 1787 size_t 1788 ipsec_hdrsiz_tcp(struct tcpcb *tp) 1789 { 1790 struct inpcb *inp; 1791 struct mbuf *m; 1792 size_t hdrsiz; 1793 struct ip *ip; 1794 #ifdef INET6 1795 struct ip6_hdr *ip6; 1796 #endif 1797 struct tcphdr *th; 1798 1799 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1800 return (0); 1801 MGETHDR(m, M_DONTWAIT, MT_DATA); 1802 if (!m) 1803 return (0); 1804 1805 #ifdef INET6 1806 if ((inp->inp_vflag & INP_IPV6) != 0) { 1807 ip6 = mtod(m, struct ip6_hdr *); 1808 th = (struct tcphdr *)(ip6 + 1); 1809 m->m_pkthdr.len = m->m_len = 1810 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1811 tcpip_fillheaders(inp, ip6, th); 1812 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1813 } else 1814 #endif /* INET6 */ 1815 { 1816 ip = mtod(m, struct ip *); 1817 th = (struct tcphdr *)(ip + 1); 1818 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1819 tcpip_fillheaders(inp, ip, th); 1820 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1821 } 1822 1823 m_free(m); 1824 return (hdrsiz); 1825 } 1826 #endif /* IPSEC */ 1827 1828 #ifdef TCP_SIGNATURE 1829 /* 1830 * Callback function invoked by m_apply() to digest TCP segment data 1831 * contained within an mbuf chain. 1832 */ 1833 static int 1834 tcp_signature_apply(void *fstate, void *data, u_int len) 1835 { 1836 1837 MD5Update(fstate, (u_char *)data, len); 1838 return (0); 1839 } 1840 1841 /* 1842 * Compute TCP-MD5 hash of a TCP segment. (RFC2385) 1843 * 1844 * Parameters: 1845 * m pointer to head of mbuf chain 1846 * _unused 1847 * len length of TCP segment data, excluding options 1848 * optlen length of TCP segment options 1849 * buf pointer to storage for computed MD5 digest 1850 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 1851 * 1852 * We do this over ip, tcphdr, segment data, and the key in the SADB. 1853 * When called from tcp_input(), we can be sure that th_sum has been 1854 * zeroed out and verified already. 1855 * 1856 * Return 0 if successful, otherwise return -1. 1857 * 1858 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 1859 * search with the destination IP address, and a 'magic SPI' to be 1860 * determined by the application. This is hardcoded elsewhere to 1179 1861 * right now. Another branch of this code exists which uses the SPD to 1862 * specify per-application flows but it is unstable. 1863 */ 1864 int 1865 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen, 1866 u_char *buf, u_int direction) 1867 { 1868 union sockaddr_union dst; 1869 #ifdef INET 1870 struct ippseudo ippseudo; 1871 #endif 1872 MD5_CTX ctx; 1873 int doff; 1874 struct ip *ip; 1875 #ifdef INET 1876 struct ipovly *ipovly; 1877 #endif 1878 struct secasvar *sav; 1879 struct tcphdr *th; 1880 #ifdef INET6 1881 struct ip6_hdr *ip6; 1882 struct in6_addr in6; 1883 char ip6buf[INET6_ADDRSTRLEN]; 1884 uint32_t plen; 1885 uint16_t nhdr; 1886 #endif 1887 u_short savecsum; 1888 1889 KASSERT(m != NULL, ("NULL mbuf chain")); 1890 KASSERT(buf != NULL, ("NULL signature pointer")); 1891 1892 /* Extract the destination from the IP header in the mbuf. */ 1893 bzero(&dst, sizeof(union sockaddr_union)); 1894 ip = mtod(m, struct ip *); 1895 #ifdef INET6 1896 ip6 = NULL; /* Make the compiler happy. */ 1897 #endif 1898 switch (ip->ip_v) { 1899 #ifdef INET 1900 case IPVERSION: 1901 dst.sa.sa_len = sizeof(struct sockaddr_in); 1902 dst.sa.sa_family = AF_INET; 1903 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 1904 ip->ip_src : ip->ip_dst; 1905 break; 1906 #endif 1907 #ifdef INET6 1908 case (IPV6_VERSION >> 4): 1909 ip6 = mtod(m, struct ip6_hdr *); 1910 dst.sa.sa_len = sizeof(struct sockaddr_in6); 1911 dst.sa.sa_family = AF_INET6; 1912 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ? 1913 ip6->ip6_src : ip6->ip6_dst; 1914 break; 1915 #endif 1916 default: 1917 return (EINVAL); 1918 /* NOTREACHED */ 1919 break; 1920 } 1921 1922 /* Look up an SADB entry which matches the address of the peer. */ 1923 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 1924 if (sav == NULL) { 1925 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__, 1926 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) : 1927 #ifdef INET6 1928 (ip->ip_v == (IPV6_VERSION >> 4)) ? 1929 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) : 1930 #endif 1931 "(unsupported)")); 1932 return (EINVAL); 1933 } 1934 1935 MD5Init(&ctx); 1936 /* 1937 * Step 1: Update MD5 hash with IP(v6) pseudo-header. 1938 * 1939 * XXX The ippseudo header MUST be digested in network byte order, 1940 * or else we'll fail the regression test. Assume all fields we've 1941 * been doing arithmetic on have been in host byte order. 1942 * XXX One cannot depend on ipovly->ih_len here. When called from 1943 * tcp_output(), the underlying ip_len member has not yet been set. 1944 */ 1945 switch (ip->ip_v) { 1946 #ifdef INET 1947 case IPVERSION: 1948 ipovly = (struct ipovly *)ip; 1949 ippseudo.ippseudo_src = ipovly->ih_src; 1950 ippseudo.ippseudo_dst = ipovly->ih_dst; 1951 ippseudo.ippseudo_pad = 0; 1952 ippseudo.ippseudo_p = IPPROTO_TCP; 1953 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + 1954 optlen); 1955 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 1956 1957 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip)); 1958 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen; 1959 break; 1960 #endif 1961 #ifdef INET6 1962 /* 1963 * RFC 2385, 2.0 Proposal 1964 * For IPv6, the pseudo-header is as described in RFC 2460, namely the 1965 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero- 1966 * extended next header value (to form 32 bits), and 32-bit segment 1967 * length. 1968 * Note: Upper-Layer Packet Length comes before Next Header. 1969 */ 1970 case (IPV6_VERSION >> 4): 1971 in6 = ip6->ip6_src; 1972 in6_clearscope(&in6); 1973 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 1974 in6 = ip6->ip6_dst; 1975 in6_clearscope(&in6); 1976 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 1977 plen = htonl(len + sizeof(struct tcphdr) + optlen); 1978 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t)); 1979 nhdr = 0; 1980 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 1981 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 1982 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 1983 nhdr = IPPROTO_TCP; 1984 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 1985 1986 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr)); 1987 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen; 1988 break; 1989 #endif 1990 default: 1991 return (EINVAL); 1992 /* NOTREACHED */ 1993 break; 1994 } 1995 1996 1997 /* 1998 * Step 2: Update MD5 hash with TCP header, excluding options. 1999 * The TCP checksum must be set to zero. 2000 */ 2001 savecsum = th->th_sum; 2002 th->th_sum = 0; 2003 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2004 th->th_sum = savecsum; 2005 2006 /* 2007 * Step 3: Update MD5 hash with TCP segment data. 2008 * Use m_apply() to avoid an early m_pullup(). 2009 */ 2010 if (len > 0) 2011 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2012 2013 /* 2014 * Step 4: Update MD5 hash with shared secret. 2015 */ 2016 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth)); 2017 MD5Final(buf, &ctx); 2018 2019 key_sa_recordxfer(sav, m); 2020 KEY_FREESAV(&sav); 2021 return (0); 2022 } 2023 2024 /* 2025 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385) 2026 * 2027 * Parameters: 2028 * m pointer to head of mbuf chain 2029 * len length of TCP segment data, excluding options 2030 * optlen length of TCP segment options 2031 * buf pointer to storage for computed MD5 digest 2032 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2033 * 2034 * Return 1 if successful, otherwise return 0. 2035 */ 2036 int 2037 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen, 2038 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 2039 { 2040 char tmpdigest[TCP_SIGLEN]; 2041 2042 if (tcp_sig_checksigs == 0) 2043 return (1); 2044 if ((tcpbflag & TF_SIGNATURE) == 0) { 2045 if ((to->to_flags & TOF_SIGNATURE) != 0) { 2046 2047 /* 2048 * If this socket is not expecting signature but 2049 * the segment contains signature just fail. 2050 */ 2051 TCPSTAT_INC(tcps_sig_err_sigopt); 2052 TCPSTAT_INC(tcps_sig_rcvbadsig); 2053 return (0); 2054 } 2055 2056 /* Signature is not expected, and not present in segment. */ 2057 return (1); 2058 } 2059 2060 /* 2061 * If this socket is expecting signature but the segment does not 2062 * contain any just fail. 2063 */ 2064 if ((to->to_flags & TOF_SIGNATURE) == 0) { 2065 TCPSTAT_INC(tcps_sig_err_nosigopt); 2066 TCPSTAT_INC(tcps_sig_rcvbadsig); 2067 return (0); 2068 } 2069 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0], 2070 IPSEC_DIR_INBOUND) == -1) { 2071 TCPSTAT_INC(tcps_sig_err_buildsig); 2072 TCPSTAT_INC(tcps_sig_rcvbadsig); 2073 return (0); 2074 } 2075 2076 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) { 2077 TCPSTAT_INC(tcps_sig_rcvbadsig); 2078 return (0); 2079 } 2080 TCPSTAT_INC(tcps_sig_rcvgoodsig); 2081 return (1); 2082 } 2083 #endif /* TCP_SIGNATURE */ 2084 2085 static int 2086 sysctl_drop(SYSCTL_HANDLER_ARGS) 2087 { 2088 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2089 struct sockaddr_storage addrs[2]; 2090 struct inpcb *inp; 2091 struct tcpcb *tp; 2092 struct tcptw *tw; 2093 struct sockaddr_in *fin, *lin; 2094 #ifdef INET6 2095 struct sockaddr_in6 *fin6, *lin6; 2096 #endif 2097 int error; 2098 2099 inp = NULL; 2100 fin = lin = NULL; 2101 #ifdef INET6 2102 fin6 = lin6 = NULL; 2103 #endif 2104 error = 0; 2105 2106 if (req->oldptr != NULL || req->oldlen != 0) 2107 return (EINVAL); 2108 if (req->newptr == NULL) 2109 return (EPERM); 2110 if (req->newlen < sizeof(addrs)) 2111 return (ENOMEM); 2112 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2113 if (error) 2114 return (error); 2115 2116 switch (addrs[0].ss_family) { 2117 #ifdef INET6 2118 case AF_INET6: 2119 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2120 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2121 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2122 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2123 return (EINVAL); 2124 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2125 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2126 return (EINVAL); 2127 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2128 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2129 fin = (struct sockaddr_in *)&addrs[0]; 2130 lin = (struct sockaddr_in *)&addrs[1]; 2131 break; 2132 } 2133 error = sa6_embedscope(fin6, V_ip6_use_defzone); 2134 if (error) 2135 return (error); 2136 error = sa6_embedscope(lin6, V_ip6_use_defzone); 2137 if (error) 2138 return (error); 2139 break; 2140 #endif 2141 #ifdef INET 2142 case AF_INET: 2143 fin = (struct sockaddr_in *)&addrs[0]; 2144 lin = (struct sockaddr_in *)&addrs[1]; 2145 if (fin->sin_len != sizeof(struct sockaddr_in) || 2146 lin->sin_len != sizeof(struct sockaddr_in)) 2147 return (EINVAL); 2148 break; 2149 #endif 2150 default: 2151 return (EINVAL); 2152 } 2153 INP_INFO_WLOCK(&V_tcbinfo); 2154 switch (addrs[0].ss_family) { 2155 #ifdef INET6 2156 case AF_INET6: 2157 inp = in6_pcblookup_hash(&V_tcbinfo, &fin6->sin6_addr, 2158 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 0, 2159 NULL); 2160 break; 2161 #endif 2162 #ifdef INET 2163 case AF_INET: 2164 inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr, 2165 fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL); 2166 break; 2167 #endif 2168 } 2169 if (inp != NULL) { 2170 INP_WLOCK(inp); 2171 if (inp->inp_flags & INP_TIMEWAIT) { 2172 /* 2173 * XXXRW: There currently exists a state where an 2174 * inpcb is present, but its timewait state has been 2175 * discarded. For now, don't allow dropping of this 2176 * type of inpcb. 2177 */ 2178 tw = intotw(inp); 2179 if (tw != NULL) 2180 tcp_twclose(tw, 0); 2181 else 2182 INP_WUNLOCK(inp); 2183 } else if (!(inp->inp_flags & INP_DROPPED) && 2184 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2185 tp = intotcpcb(inp); 2186 tp = tcp_drop(tp, ECONNABORTED); 2187 if (tp != NULL) 2188 INP_WUNLOCK(inp); 2189 } else 2190 INP_WUNLOCK(inp); 2191 } else 2192 error = ESRCH; 2193 INP_INFO_WUNLOCK(&V_tcbinfo); 2194 return (error); 2195 } 2196 2197 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2198 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2199 0, sysctl_drop, "", "Drop TCP connection"); 2200 2201 /* 2202 * Generate a standardized TCP log line for use throughout the 2203 * tcp subsystem. Memory allocation is done with M_NOWAIT to 2204 * allow use in the interrupt context. 2205 * 2206 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 2207 * NB: The function may return NULL if memory allocation failed. 2208 * 2209 * Due to header inclusion and ordering limitations the struct ip 2210 * and ip6_hdr pointers have to be passed as void pointers. 2211 */ 2212 char * 2213 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2214 const void *ip6hdr) 2215 { 2216 2217 /* Is logging enabled? */ 2218 if (tcp_log_in_vain == 0) 2219 return (NULL); 2220 2221 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2222 } 2223 2224 char * 2225 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2226 const void *ip6hdr) 2227 { 2228 2229 /* Is logging enabled? */ 2230 if (tcp_log_debug == 0) 2231 return (NULL); 2232 2233 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2234 } 2235 2236 static char * 2237 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2238 const void *ip6hdr) 2239 { 2240 char *s, *sp; 2241 size_t size; 2242 struct ip *ip; 2243 #ifdef INET6 2244 const struct ip6_hdr *ip6; 2245 2246 ip6 = (const struct ip6_hdr *)ip6hdr; 2247 #endif /* INET6 */ 2248 ip = (struct ip *)ip4hdr; 2249 2250 /* 2251 * The log line looks like this: 2252 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 2253 */ 2254 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 2255 sizeof(PRINT_TH_FLAGS) + 1 + 2256 #ifdef INET6 2257 2 * INET6_ADDRSTRLEN; 2258 #else 2259 2 * INET_ADDRSTRLEN; 2260 #endif /* INET6 */ 2261 2262 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 2263 if (s == NULL) 2264 return (NULL); 2265 2266 strcat(s, "TCP: ["); 2267 sp = s + strlen(s); 2268 2269 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 2270 inet_ntoa_r(inc->inc_faddr, sp); 2271 sp = s + strlen(s); 2272 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2273 sp = s + strlen(s); 2274 inet_ntoa_r(inc->inc_laddr, sp); 2275 sp = s + strlen(s); 2276 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2277 #ifdef INET6 2278 } else if (inc) { 2279 ip6_sprintf(sp, &inc->inc6_faddr); 2280 sp = s + strlen(s); 2281 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2282 sp = s + strlen(s); 2283 ip6_sprintf(sp, &inc->inc6_laddr); 2284 sp = s + strlen(s); 2285 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2286 } else if (ip6 && th) { 2287 ip6_sprintf(sp, &ip6->ip6_src); 2288 sp = s + strlen(s); 2289 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2290 sp = s + strlen(s); 2291 ip6_sprintf(sp, &ip6->ip6_dst); 2292 sp = s + strlen(s); 2293 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2294 #endif /* INET6 */ 2295 #ifdef INET 2296 } else if (ip && th) { 2297 inet_ntoa_r(ip->ip_src, sp); 2298 sp = s + strlen(s); 2299 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2300 sp = s + strlen(s); 2301 inet_ntoa_r(ip->ip_dst, sp); 2302 sp = s + strlen(s); 2303 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2304 #endif /* INET */ 2305 } else { 2306 free(s, M_TCPLOG); 2307 return (NULL); 2308 } 2309 sp = s + strlen(s); 2310 if (th) 2311 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 2312 if (*(s + size - 1) != '\0') 2313 panic("%s: string too long", __func__); 2314 return (s); 2315 } 2316