1 /* 2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 36 * The Regents of the University of California. All rights reserved. 37 * 38 * Redistribution and use in source and binary forms, with or without 39 * modification, are permitted provided that the following conditions 40 * are met: 41 * 1. Redistributions of source code must retain the above copyright 42 * notice, this list of conditions and the following disclaimer. 43 * 2. Redistributions in binary form must reproduce the above copyright 44 * notice, this list of conditions and the following disclaimer in the 45 * documentation and/or other materials provided with the distribution. 46 * 3. All advertising materials mentioning features or use of this software 47 * must display the following acknowledgement: 48 * This product includes software developed by the University of 49 * California, Berkeley and its contributors. 50 * 4. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 67 * $FreeBSD: src/sys/netinet/tcp_subr.c,v 1.73.2.31 2003/01/24 05:11:34 sam Exp $ 68 */ 69 70 #include "opt_compat.h" 71 #include "opt_inet.h" 72 #include "opt_inet6.h" 73 #include "opt_ipsec.h" 74 #include "opt_tcpdebug.h" 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/callout.h> 79 #include <sys/kernel.h> 80 #include <sys/sysctl.h> 81 #include <sys/malloc.h> 82 #include <sys/mpipe.h> 83 #include <sys/mbuf.h> 84 #ifdef INET6 85 #include <sys/domain.h> 86 #endif 87 #include <sys/proc.h> 88 #include <sys/priv.h> 89 #include <sys/socket.h> 90 #include <sys/socketvar.h> 91 #include <sys/protosw.h> 92 #include <sys/random.h> 93 #include <sys/in_cksum.h> 94 #include <sys/ktr.h> 95 96 #include <net/route.h> 97 #include <net/if.h> 98 #include <net/netisr.h> 99 100 #define _IP_VHL 101 #include <netinet/in.h> 102 #include <netinet/in_systm.h> 103 #include <netinet/ip.h> 104 #include <netinet/ip6.h> 105 #include <netinet/in_pcb.h> 106 #include <netinet6/in6_pcb.h> 107 #include <netinet/in_var.h> 108 #include <netinet/ip_var.h> 109 #include <netinet6/ip6_var.h> 110 #include <netinet/ip_icmp.h> 111 #ifdef INET6 112 #include <netinet/icmp6.h> 113 #endif 114 #include <netinet/tcp.h> 115 #include <netinet/tcp_fsm.h> 116 #include <netinet/tcp_seq.h> 117 #include <netinet/tcp_timer.h> 118 #include <netinet/tcp_timer2.h> 119 #include <netinet/tcp_var.h> 120 #include <netinet6/tcp6_var.h> 121 #include <netinet/tcpip.h> 122 #ifdef TCPDEBUG 123 #include <netinet/tcp_debug.h> 124 #endif 125 #include <netinet6/ip6protosw.h> 126 127 #ifdef IPSEC 128 #include <netinet6/ipsec.h> 129 #include <netproto/key/key.h> 130 #ifdef INET6 131 #include <netinet6/ipsec6.h> 132 #endif 133 #endif 134 135 #ifdef FAST_IPSEC 136 #include <netproto/ipsec/ipsec.h> 137 #ifdef INET6 138 #include <netproto/ipsec/ipsec6.h> 139 #endif 140 #define IPSEC 141 #endif 142 143 #include <sys/md5.h> 144 #include <machine/smp.h> 145 146 #include <sys/msgport2.h> 147 #include <sys/mplock2.h> 148 #include <net/netmsg2.h> 149 150 #if !defined(KTR_TCP) 151 #define KTR_TCP KTR_ALL 152 #endif 153 /* 154 KTR_INFO_MASTER(tcp); 155 KTR_INFO(KTR_TCP, tcp, rxmsg, 0, "tcp getmsg", 0); 156 KTR_INFO(KTR_TCP, tcp, wait, 1, "tcp waitmsg", 0); 157 KTR_INFO(KTR_TCP, tcp, delayed, 2, "tcp execute delayed ops", 0); 158 #define logtcp(name) KTR_LOG(tcp_ ## name) 159 */ 160 161 #define TCP_IW_MAXSEGS_DFLT 4 162 #define TCP_IW_CAPSEGS_DFLT 3 163 164 struct inpcbinfo tcbinfo[MAXCPU]; 165 struct tcpcbackqhead tcpcbackq[MAXCPU]; 166 167 static struct lwkt_token tcp_port_token = 168 LWKT_TOKEN_INITIALIZER(tcp_port_token); 169 170 int tcp_mssdflt = TCP_MSS; 171 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 172 &tcp_mssdflt, 0, "Default TCP Maximum Segment Size"); 173 174 #ifdef INET6 175 int tcp_v6mssdflt = TCP6_MSS; 176 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLFLAG_RW, 177 &tcp_v6mssdflt, 0, "Default TCP Maximum Segment Size for IPv6"); 178 #endif 179 180 /* 181 * Minimum MSS we accept and use. This prevents DoS attacks where 182 * we are forced to a ridiculous low MSS like 20 and send hundreds 183 * of packets instead of one. The effect scales with the available 184 * bandwidth and quickly saturates the CPU and network interface 185 * with packet generation and sending. Set to zero to disable MINMSS 186 * checking. This setting prevents us from sending too small packets. 187 */ 188 int tcp_minmss = TCP_MINMSS; 189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 190 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 191 192 #if 0 193 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 194 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 195 &tcp_rttdflt, 0, "Default maximum TCP Round Trip Time"); 196 #endif 197 198 int tcp_do_rfc1323 = 1; 199 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 200 &tcp_do_rfc1323, 0, "Enable rfc1323 (high performance TCP) extensions"); 201 202 static int tcp_tcbhashsize = 0; 203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD, 204 &tcp_tcbhashsize, 0, "Size of TCP control block hashtable"); 205 206 static int do_tcpdrain = 1; 207 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 208 "Enable tcp_drain routine for extra help when low on mbufs"); 209 210 static int icmp_may_rst = 1; 211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 212 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 213 214 static int tcp_isn_reseed_interval = 0; 215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 216 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 217 218 /* 219 * TCP bandwidth limiting sysctls. The inflight limiter is now turned on 220 * by default, but with generous values which should allow maximal 221 * bandwidth. In particular, the slop defaults to 50 (5 packets). 222 * 223 * The reason for doing this is that the limiter is the only mechanism we 224 * have which seems to do a really good job preventing receiver RX rings 225 * on network interfaces from getting blown out. Even though GigE/10GigE 226 * is supposed to flow control it looks like either it doesn't actually 227 * do it or Open Source drivers do not properly enable it. 228 * 229 * People using the limiter to reduce bottlenecks on slower WAN connections 230 * should set the slop to 20 (2 packets). 231 */ 232 static int tcp_inflight_enable = 1; 233 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_enable, CTLFLAG_RW, 234 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 235 236 static int tcp_inflight_debug = 0; 237 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_debug, CTLFLAG_RW, 238 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 239 240 static int tcp_inflight_min = 6144; 241 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_min, CTLFLAG_RW, 242 &tcp_inflight_min, 0, "Lower bound for TCP inflight window"); 243 244 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 245 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_max, CTLFLAG_RW, 246 &tcp_inflight_max, 0, "Upper bound for TCP inflight window"); 247 248 static int tcp_inflight_stab = 50; 249 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_stab, CTLFLAG_RW, 250 &tcp_inflight_stab, 0, "Slop in maximal packets / 10 (20 = 3 packets)"); 251 252 static int tcp_do_rfc3390 = 1; 253 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 254 &tcp_do_rfc3390, 0, 255 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 256 257 static u_long tcp_iw_maxsegs = TCP_IW_MAXSEGS_DFLT; 258 SYSCTL_ULONG(_net_inet_tcp, OID_AUTO, iwmaxsegs, CTLFLAG_RW, 259 &tcp_iw_maxsegs, 0, "TCP IW segments max"); 260 261 static u_long tcp_iw_capsegs = TCP_IW_CAPSEGS_DFLT; 262 SYSCTL_ULONG(_net_inet_tcp, OID_AUTO, iwcapsegs, CTLFLAG_RW, 263 &tcp_iw_capsegs, 0, "TCP IW segments"); 264 265 int tcp_low_rtobase = 1; 266 SYSCTL_INT(_net_inet_tcp, OID_AUTO, low_rtobase, CTLFLAG_RW, 267 &tcp_low_rtobase, 0, "Lowering the Initial RTO (RFC 6298)"); 268 269 static int tcp_do_ncr = 1; 270 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ncr, CTLFLAG_RW, 271 &tcp_do_ncr, 0, "Non-Congestion Robustness (RFC 4653)"); 272 273 static MALLOC_DEFINE(M_TCPTEMP, "tcptemp", "TCP Templates for Keepalives"); 274 static struct malloc_pipe tcptemp_mpipe; 275 276 static void tcp_willblock(void); 277 static void tcp_notify (struct inpcb *, int); 278 279 struct tcp_stats tcpstats_percpu[MAXCPU]; 280 #ifdef SMP 281 static int 282 sysctl_tcpstats(SYSCTL_HANDLER_ARGS) 283 { 284 int cpu, error = 0; 285 286 for (cpu = 0; cpu < ncpus; ++cpu) { 287 if ((error = SYSCTL_OUT(req, &tcpstats_percpu[cpu], 288 sizeof(struct tcp_stats)))) 289 break; 290 if ((error = SYSCTL_IN(req, &tcpstats_percpu[cpu], 291 sizeof(struct tcp_stats)))) 292 break; 293 } 294 295 return (error); 296 } 297 SYSCTL_PROC(_net_inet_tcp, TCPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW), 298 0, 0, sysctl_tcpstats, "S,tcp_stats", "TCP statistics"); 299 #else 300 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 301 &tcpstat, tcp_stats, "TCP statistics"); 302 #endif 303 304 /* 305 * Target size of TCP PCB hash tables. Must be a power of two. 306 * 307 * Note that this can be overridden by the kernel environment 308 * variable net.inet.tcp.tcbhashsize 309 */ 310 #ifndef TCBHASHSIZE 311 #define TCBHASHSIZE 512 312 #endif 313 314 /* 315 * This is the actual shape of what we allocate using the zone 316 * allocator. Doing it this way allows us to protect both structures 317 * using the same generation count, and also eliminates the overhead 318 * of allocating tcpcbs separately. By hiding the structure here, 319 * we avoid changing most of the rest of the code (although it needs 320 * to be changed, eventually, for greater efficiency). 321 */ 322 #define ALIGNMENT 32 323 #define ALIGNM1 (ALIGNMENT - 1) 324 struct inp_tp { 325 union { 326 struct inpcb inp; 327 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1]; 328 } inp_tp_u; 329 struct tcpcb tcb; 330 struct tcp_callout inp_tp_rexmt; 331 struct tcp_callout inp_tp_persist; 332 struct tcp_callout inp_tp_keep; 333 struct tcp_callout inp_tp_2msl; 334 struct tcp_callout inp_tp_delack; 335 struct netmsg_tcp_timer inp_tp_timermsg; 336 }; 337 #undef ALIGNMENT 338 #undef ALIGNM1 339 340 /* 341 * Tcp initialization 342 */ 343 void 344 tcp_init(void) 345 { 346 struct inpcbporthead *porthashbase; 347 struct inpcbinfo *ticb; 348 u_long porthashmask; 349 int hashsize = TCBHASHSIZE; 350 int cpu; 351 352 /* 353 * note: tcptemp is used for keepalives, and it is ok for an 354 * allocation to fail so do not specify MPF_INT. 355 */ 356 mpipe_init(&tcptemp_mpipe, M_TCPTEMP, sizeof(struct tcptemp), 357 25, -1, 0, NULL, NULL, NULL); 358 359 tcp_delacktime = TCPTV_DELACK; 360 tcp_keepinit = TCPTV_KEEP_INIT; 361 tcp_keepidle = TCPTV_KEEP_IDLE; 362 tcp_keepintvl = TCPTV_KEEPINTVL; 363 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 364 tcp_msl = TCPTV_MSL; 365 tcp_rexmit_min = TCPTV_MIN; 366 tcp_rexmit_slop = TCPTV_CPU_VAR; 367 368 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 369 if (!powerof2(hashsize)) { 370 kprintf("WARNING: TCB hash size not a power of 2\n"); 371 hashsize = 512; /* safe default */ 372 } 373 tcp_tcbhashsize = hashsize; 374 porthashbase = hashinit(hashsize, M_PCB, &porthashmask); 375 376 for (cpu = 0; cpu < ncpus2; cpu++) { 377 ticb = &tcbinfo[cpu]; 378 in_pcbinfo_init(ticb); 379 ticb->cpu = cpu; 380 ticb->hashbase = hashinit(hashsize, M_PCB, 381 &ticb->hashmask); 382 ticb->porthashbase = porthashbase; 383 ticb->porthashmask = porthashmask; 384 ticb->porttoken = &tcp_port_token; 385 #if 0 386 ticb->porthashbase = hashinit(hashsize, M_PCB, 387 &ticb->porthashmask); 388 #endif 389 ticb->wildcardhashbase = hashinit(hashsize, M_PCB, 390 &ticb->wildcardhashmask); 391 ticb->ipi_size = sizeof(struct inp_tp); 392 TAILQ_INIT(&tcpcbackq[cpu]); 393 } 394 395 tcp_reass_maxseg = nmbclusters / 16; 396 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", &tcp_reass_maxseg); 397 398 #ifdef INET6 399 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 400 #else 401 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 402 #endif 403 if (max_protohdr < TCP_MINPROTOHDR) 404 max_protohdr = TCP_MINPROTOHDR; 405 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 406 panic("tcp_init"); 407 #undef TCP_MINPROTOHDR 408 409 /* 410 * Initialize TCP statistics counters for each CPU. 411 */ 412 #ifdef SMP 413 for (cpu = 0; cpu < ncpus; ++cpu) { 414 bzero(&tcpstats_percpu[cpu], sizeof(struct tcp_stats)); 415 } 416 #else 417 bzero(&tcpstat, sizeof(struct tcp_stats)); 418 #endif 419 420 syncache_init(); 421 netisr_register_rollup(tcp_willblock); 422 } 423 424 static void 425 tcp_willblock(void) 426 { 427 struct tcpcb *tp; 428 int cpu = mycpu->gd_cpuid; 429 430 while ((tp = TAILQ_FIRST(&tcpcbackq[cpu])) != NULL) { 431 KKASSERT(tp->t_flags & TF_ONOUTPUTQ); 432 tp->t_flags &= ~TF_ONOUTPUTQ; 433 TAILQ_REMOVE(&tcpcbackq[cpu], tp, t_outputq); 434 tcp_output(tp); 435 } 436 } 437 438 /* 439 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 440 * tcp_template used to store this data in mbufs, but we now recopy it out 441 * of the tcpcb each time to conserve mbufs. 442 */ 443 void 444 tcp_fillheaders(struct tcpcb *tp, void *ip_ptr, void *tcp_ptr, boolean_t tso) 445 { 446 struct inpcb *inp = tp->t_inpcb; 447 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr; 448 449 #ifdef INET6 450 if (inp->inp_vflag & INP_IPV6) { 451 struct ip6_hdr *ip6; 452 453 ip6 = (struct ip6_hdr *)ip_ptr; 454 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 455 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 456 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 457 (IPV6_VERSION & IPV6_VERSION_MASK); 458 ip6->ip6_nxt = IPPROTO_TCP; 459 ip6->ip6_plen = sizeof(struct tcphdr); 460 ip6->ip6_src = inp->in6p_laddr; 461 ip6->ip6_dst = inp->in6p_faddr; 462 tcp_hdr->th_sum = 0; 463 } else 464 #endif 465 { 466 struct ip *ip = (struct ip *) ip_ptr; 467 u_int plen; 468 469 ip->ip_vhl = IP_VHL_BORING; 470 ip->ip_tos = 0; 471 ip->ip_len = 0; 472 ip->ip_id = 0; 473 ip->ip_off = 0; 474 ip->ip_ttl = 0; 475 ip->ip_sum = 0; 476 ip->ip_p = IPPROTO_TCP; 477 ip->ip_src = inp->inp_laddr; 478 ip->ip_dst = inp->inp_faddr; 479 480 if (tso) 481 plen = htons(IPPROTO_TCP); 482 else 483 plen = htons(sizeof(struct tcphdr) + IPPROTO_TCP); 484 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, 485 ip->ip_dst.s_addr, plen); 486 } 487 488 tcp_hdr->th_sport = inp->inp_lport; 489 tcp_hdr->th_dport = inp->inp_fport; 490 tcp_hdr->th_seq = 0; 491 tcp_hdr->th_ack = 0; 492 tcp_hdr->th_x2 = 0; 493 tcp_hdr->th_off = 5; 494 tcp_hdr->th_flags = 0; 495 tcp_hdr->th_win = 0; 496 tcp_hdr->th_urp = 0; 497 } 498 499 /* 500 * Create template to be used to send tcp packets on a connection. 501 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 502 * use for this function is in keepalives, which use tcp_respond. 503 */ 504 struct tcptemp * 505 tcp_maketemplate(struct tcpcb *tp) 506 { 507 struct tcptemp *tmp; 508 509 if ((tmp = mpipe_alloc_nowait(&tcptemp_mpipe)) == NULL) 510 return (NULL); 511 tcp_fillheaders(tp, &tmp->tt_ipgen, &tmp->tt_t, FALSE); 512 return (tmp); 513 } 514 515 void 516 tcp_freetemplate(struct tcptemp *tmp) 517 { 518 mpipe_free(&tcptemp_mpipe, tmp); 519 } 520 521 /* 522 * Send a single message to the TCP at address specified by 523 * the given TCP/IP header. If m == NULL, then we make a copy 524 * of the tcpiphdr at ti and send directly to the addressed host. 525 * This is used to force keep alive messages out using the TCP 526 * template for a connection. If flags are given then we send 527 * a message back to the TCP which originated the * segment ti, 528 * and discard the mbuf containing it and any other attached mbufs. 529 * 530 * In any case the ack and sequence number of the transmitted 531 * segment are as specified by the parameters. 532 * 533 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 534 */ 535 void 536 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 537 tcp_seq ack, tcp_seq seq, int flags) 538 { 539 int tlen; 540 int win = 0; 541 struct route *ro = NULL; 542 struct route sro; 543 struct ip *ip = ipgen; 544 struct tcphdr *nth; 545 int ipflags = 0; 546 struct route_in6 *ro6 = NULL; 547 struct route_in6 sro6; 548 struct ip6_hdr *ip6 = ipgen; 549 boolean_t use_tmpro = TRUE; 550 #ifdef INET6 551 boolean_t isipv6 = (IP_VHL_V(ip->ip_vhl) == 6); 552 #else 553 const boolean_t isipv6 = FALSE; 554 #endif 555 556 if (tp != NULL) { 557 if (!(flags & TH_RST)) { 558 win = ssb_space(&tp->t_inpcb->inp_socket->so_rcv); 559 if (win < 0) 560 win = 0; 561 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 562 win = (long)TCP_MAXWIN << tp->rcv_scale; 563 } 564 /* 565 * Don't use the route cache of a listen socket, 566 * it is not MPSAFE; use temporary route cache. 567 */ 568 if (tp->t_state != TCPS_LISTEN) { 569 if (isipv6) 570 ro6 = &tp->t_inpcb->in6p_route; 571 else 572 ro = &tp->t_inpcb->inp_route; 573 use_tmpro = FALSE; 574 } 575 } 576 if (use_tmpro) { 577 if (isipv6) { 578 ro6 = &sro6; 579 bzero(ro6, sizeof *ro6); 580 } else { 581 ro = &sro; 582 bzero(ro, sizeof *ro); 583 } 584 } 585 if (m == NULL) { 586 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 587 if (m == NULL) 588 return; 589 tlen = 0; 590 m->m_data += max_linkhdr; 591 if (isipv6) { 592 bcopy(ip6, mtod(m, caddr_t), sizeof(struct ip6_hdr)); 593 ip6 = mtod(m, struct ip6_hdr *); 594 nth = (struct tcphdr *)(ip6 + 1); 595 } else { 596 bcopy(ip, mtod(m, caddr_t), sizeof(struct ip)); 597 ip = mtod(m, struct ip *); 598 nth = (struct tcphdr *)(ip + 1); 599 } 600 bcopy(th, nth, sizeof(struct tcphdr)); 601 flags = TH_ACK; 602 } else { 603 m_freem(m->m_next); 604 m->m_next = NULL; 605 m->m_data = (caddr_t)ipgen; 606 /* m_len is set later */ 607 tlen = 0; 608 #define xchg(a, b, type) { type t; t = a; a = b; b = t; } 609 if (isipv6) { 610 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 611 nth = (struct tcphdr *)(ip6 + 1); 612 } else { 613 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 614 nth = (struct tcphdr *)(ip + 1); 615 } 616 if (th != nth) { 617 /* 618 * this is usually a case when an extension header 619 * exists between the IPv6 header and the 620 * TCP header. 621 */ 622 nth->th_sport = th->th_sport; 623 nth->th_dport = th->th_dport; 624 } 625 xchg(nth->th_dport, nth->th_sport, n_short); 626 #undef xchg 627 } 628 if (isipv6) { 629 ip6->ip6_flow = 0; 630 ip6->ip6_vfc = IPV6_VERSION; 631 ip6->ip6_nxt = IPPROTO_TCP; 632 ip6->ip6_plen = htons((u_short)(sizeof(struct tcphdr) + tlen)); 633 tlen += sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 634 } else { 635 tlen += sizeof(struct tcpiphdr); 636 ip->ip_len = tlen; 637 ip->ip_ttl = ip_defttl; 638 } 639 m->m_len = tlen; 640 m->m_pkthdr.len = tlen; 641 m->m_pkthdr.rcvif = NULL; 642 nth->th_seq = htonl(seq); 643 nth->th_ack = htonl(ack); 644 nth->th_x2 = 0; 645 nth->th_off = sizeof(struct tcphdr) >> 2; 646 nth->th_flags = flags; 647 if (tp != NULL) 648 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 649 else 650 nth->th_win = htons((u_short)win); 651 nth->th_urp = 0; 652 if (isipv6) { 653 nth->th_sum = 0; 654 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 655 sizeof(struct ip6_hdr), 656 tlen - sizeof(struct ip6_hdr)); 657 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL, 658 (ro6 && ro6->ro_rt) ? 659 ro6->ro_rt->rt_ifp : NULL); 660 } else { 661 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 662 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 663 m->m_pkthdr.csum_flags = CSUM_TCP; 664 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 665 m->m_pkthdr.csum_thlen = sizeof(struct tcphdr); 666 } 667 #ifdef TCPDEBUG 668 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 669 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 670 #endif 671 if (isipv6) { 672 ip6_output(m, NULL, ro6, ipflags, NULL, NULL, 673 tp ? tp->t_inpcb : NULL); 674 if ((ro6 == &sro6) && (ro6->ro_rt != NULL)) { 675 RTFREE(ro6->ro_rt); 676 ro6->ro_rt = NULL; 677 } 678 } else { 679 ipflags |= IP_DEBUGROUTE; 680 ip_output(m, NULL, ro, ipflags, NULL, tp ? tp->t_inpcb : NULL); 681 if ((ro == &sro) && (ro->ro_rt != NULL)) { 682 RTFREE(ro->ro_rt); 683 ro->ro_rt = NULL; 684 } 685 } 686 } 687 688 /* 689 * Create a new TCP control block, making an 690 * empty reassembly queue and hooking it to the argument 691 * protocol control block. The `inp' parameter must have 692 * come from the zone allocator set up in tcp_init(). 693 */ 694 struct tcpcb * 695 tcp_newtcpcb(struct inpcb *inp) 696 { 697 struct inp_tp *it; 698 struct tcpcb *tp; 699 #ifdef INET6 700 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0); 701 #else 702 const boolean_t isipv6 = FALSE; 703 #endif 704 705 it = (struct inp_tp *)inp; 706 tp = &it->tcb; 707 bzero(tp, sizeof(struct tcpcb)); 708 TAILQ_INIT(&tp->t_segq); 709 tp->t_maxseg = tp->t_maxopd = isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 710 tp->t_rxtthresh = tcprexmtthresh; 711 712 /* Set up our timeouts. */ 713 tp->tt_rexmt = &it->inp_tp_rexmt; 714 tp->tt_persist = &it->inp_tp_persist; 715 tp->tt_keep = &it->inp_tp_keep; 716 tp->tt_2msl = &it->inp_tp_2msl; 717 tp->tt_delack = &it->inp_tp_delack; 718 tcp_inittimers(tp); 719 720 /* 721 * Zero out timer message. We don't create it here, 722 * since the current CPU may not be the owner of this 723 * inpcb. 724 */ 725 tp->tt_msg = &it->inp_tp_timermsg; 726 bzero(tp->tt_msg, sizeof(*tp->tt_msg)); 727 728 tp->t_keepinit = tcp_keepinit; 729 tp->t_keepidle = tcp_keepidle; 730 tp->t_keepintvl = tcp_keepintvl; 731 tp->t_keepcnt = tcp_keepcnt; 732 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt; 733 734 if (tcp_do_ncr) 735 tp->t_flags |= TF_NCR; 736 if (tcp_do_rfc1323) 737 tp->t_flags |= (TF_REQ_SCALE | TF_REQ_TSTMP); 738 739 tp->t_inpcb = inp; /* XXX */ 740 tp->t_state = TCPS_CLOSED; 741 /* 742 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 743 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 744 * reasonable initial retransmit time. 745 */ 746 tp->t_srtt = TCPTV_SRTTBASE; 747 tp->t_rttvar = 748 ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 749 tp->t_rttmin = tcp_rexmit_min; 750 tp->t_rxtcur = TCPTV_RTOBASE; 751 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 752 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 753 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 754 tp->snd_last = ticks; 755 tp->t_rcvtime = ticks; 756 /* 757 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 758 * because the socket may be bound to an IPv6 wildcard address, 759 * which may match an IPv4-mapped IPv6 address. 760 */ 761 inp->inp_ip_ttl = ip_defttl; 762 inp->inp_ppcb = tp; 763 tcp_sack_tcpcb_init(tp); 764 return (tp); /* XXX */ 765 } 766 767 /* 768 * Drop a TCP connection, reporting the specified error. 769 * If connection is synchronized, then send a RST to peer. 770 */ 771 struct tcpcb * 772 tcp_drop(struct tcpcb *tp, int error) 773 { 774 struct socket *so = tp->t_inpcb->inp_socket; 775 776 if (TCPS_HAVERCVDSYN(tp->t_state)) { 777 tp->t_state = TCPS_CLOSED; 778 tcp_output(tp); 779 tcpstat.tcps_drops++; 780 } else 781 tcpstat.tcps_conndrops++; 782 if (error == ETIMEDOUT && tp->t_softerror) 783 error = tp->t_softerror; 784 so->so_error = error; 785 return (tcp_close(tp)); 786 } 787 788 #ifdef SMP 789 790 struct netmsg_listen_detach { 791 struct netmsg_base base; 792 struct tcpcb *nm_tp; 793 }; 794 795 static void 796 tcp_listen_detach_handler(netmsg_t msg) 797 { 798 struct netmsg_listen_detach *nmsg = (struct netmsg_listen_detach *)msg; 799 struct tcpcb *tp = nmsg->nm_tp; 800 int cpu = mycpuid, nextcpu; 801 802 if (tp->t_flags & TF_LISTEN) 803 syncache_destroy(tp); 804 805 in_pcbremwildcardhash_oncpu(tp->t_inpcb, &tcbinfo[cpu]); 806 807 nextcpu = cpu + 1; 808 if (nextcpu < ncpus2) 809 lwkt_forwardmsg(cpu_portfn(nextcpu), &nmsg->base.lmsg); 810 else 811 lwkt_replymsg(&nmsg->base.lmsg, 0); 812 } 813 814 #endif 815 816 /* 817 * Close a TCP control block: 818 * discard all space held by the tcp 819 * discard internet protocol block 820 * wake up any sleepers 821 */ 822 struct tcpcb * 823 tcp_close(struct tcpcb *tp) 824 { 825 struct tseg_qent *q; 826 struct inpcb *inp = tp->t_inpcb; 827 struct socket *so = inp->inp_socket; 828 struct rtentry *rt; 829 boolean_t dosavessthresh; 830 #ifdef INET6 831 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0); 832 boolean_t isafinet6 = (INP_CHECK_SOCKAF(so, AF_INET6) != 0); 833 #else 834 const boolean_t isipv6 = FALSE; 835 #endif 836 837 #ifdef SMP 838 /* 839 * INP_WILDCARD_MP indicates that listen(2) has been called on 840 * this socket. This implies: 841 * - A wildcard inp's hash is replicated for each protocol thread. 842 * - Syncache for this inp grows independently in each protocol 843 * thread. 844 * - There is more than one cpu 845 * 846 * We have to chain a message to the rest of the protocol threads 847 * to cleanup the wildcard hash and the syncache. The cleanup 848 * in the current protocol thread is defered till the end of this 849 * function. 850 * 851 * NOTE: 852 * After cleanup the inp's hash and syncache entries, this inp will 853 * no longer be available to the rest of the protocol threads, so we 854 * are safe to whack the inp in the following code. 855 */ 856 if (inp->inp_flags & INP_WILDCARD_MP) { 857 struct netmsg_listen_detach nmsg; 858 859 KKASSERT(so->so_port == cpu_portfn(0)); 860 KKASSERT(&curthread->td_msgport == cpu_portfn(0)); 861 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]); 862 863 netmsg_init(&nmsg.base, NULL, &curthread->td_msgport, 864 MSGF_PRIORITY, tcp_listen_detach_handler); 865 nmsg.nm_tp = tp; 866 lwkt_domsg(cpu_portfn(1), &nmsg.base.lmsg, 0); 867 868 inp->inp_flags &= ~INP_WILDCARD_MP; 869 } 870 #endif 871 872 KKASSERT(tp->t_state != TCPS_TERMINATING); 873 tp->t_state = TCPS_TERMINATING; 874 875 /* 876 * Make sure that all of our timers are stopped before we 877 * delete the PCB. For listen TCP socket (tp->tt_msg == NULL), 878 * timers are never used. If timer message is never created 879 * (tp->tt_msg->tt_tcb == NULL), timers are never used too. 880 */ 881 if (tp->tt_msg != NULL && tp->tt_msg->tt_tcb != NULL) { 882 tcp_callout_stop(tp, tp->tt_rexmt); 883 tcp_callout_stop(tp, tp->tt_persist); 884 tcp_callout_stop(tp, tp->tt_keep); 885 tcp_callout_stop(tp, tp->tt_2msl); 886 tcp_callout_stop(tp, tp->tt_delack); 887 } 888 889 if (tp->t_flags & TF_ONOUTPUTQ) { 890 KKASSERT(tp->tt_cpu == mycpu->gd_cpuid); 891 TAILQ_REMOVE(&tcpcbackq[tp->tt_cpu], tp, t_outputq); 892 tp->t_flags &= ~TF_ONOUTPUTQ; 893 } 894 895 /* 896 * If we got enough samples through the srtt filter, 897 * save the rtt and rttvar in the routing entry. 898 * 'Enough' is arbitrarily defined as the 16 samples. 899 * 16 samples is enough for the srtt filter to converge 900 * to within 5% of the correct value; fewer samples and 901 * we could save a very bogus rtt. 902 * 903 * Don't update the default route's characteristics and don't 904 * update anything that the user "locked". 905 */ 906 if (tp->t_rttupdated >= 16) { 907 u_long i = 0; 908 909 if (isipv6) { 910 struct sockaddr_in6 *sin6; 911 912 if ((rt = inp->in6p_route.ro_rt) == NULL) 913 goto no_valid_rt; 914 sin6 = (struct sockaddr_in6 *)rt_key(rt); 915 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 916 goto no_valid_rt; 917 } else 918 if ((rt = inp->inp_route.ro_rt) == NULL || 919 ((struct sockaddr_in *)rt_key(rt))-> 920 sin_addr.s_addr == INADDR_ANY) 921 goto no_valid_rt; 922 923 if (!(rt->rt_rmx.rmx_locks & RTV_RTT)) { 924 i = tp->t_srtt * (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 925 if (rt->rt_rmx.rmx_rtt && i) 926 /* 927 * filter this update to half the old & half 928 * the new values, converting scale. 929 * See route.h and tcp_var.h for a 930 * description of the scaling constants. 931 */ 932 rt->rt_rmx.rmx_rtt = 933 (rt->rt_rmx.rmx_rtt + i) / 2; 934 else 935 rt->rt_rmx.rmx_rtt = i; 936 tcpstat.tcps_cachedrtt++; 937 } 938 if (!(rt->rt_rmx.rmx_locks & RTV_RTTVAR)) { 939 i = tp->t_rttvar * 940 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 941 if (rt->rt_rmx.rmx_rttvar && i) 942 rt->rt_rmx.rmx_rttvar = 943 (rt->rt_rmx.rmx_rttvar + i) / 2; 944 else 945 rt->rt_rmx.rmx_rttvar = i; 946 tcpstat.tcps_cachedrttvar++; 947 } 948 /* 949 * The old comment here said: 950 * update the pipelimit (ssthresh) if it has been updated 951 * already or if a pipesize was specified & the threshhold 952 * got below half the pipesize. I.e., wait for bad news 953 * before we start updating, then update on both good 954 * and bad news. 955 * 956 * But we want to save the ssthresh even if no pipesize is 957 * specified explicitly in the route, because such 958 * connections still have an implicit pipesize specified 959 * by the global tcp_sendspace. In the absence of a reliable 960 * way to calculate the pipesize, it will have to do. 961 */ 962 i = tp->snd_ssthresh; 963 if (rt->rt_rmx.rmx_sendpipe != 0) 964 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe/2); 965 else 966 dosavessthresh = (i < so->so_snd.ssb_hiwat/2); 967 if (dosavessthresh || 968 (!(rt->rt_rmx.rmx_locks & RTV_SSTHRESH) && (i != 0) && 969 (rt->rt_rmx.rmx_ssthresh != 0))) { 970 /* 971 * convert the limit from user data bytes to 972 * packets then to packet data bytes. 973 */ 974 i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 975 if (i < 2) 976 i = 2; 977 i *= tp->t_maxseg + 978 (isipv6 ? 979 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 980 sizeof(struct tcpiphdr)); 981 if (rt->rt_rmx.rmx_ssthresh) 982 rt->rt_rmx.rmx_ssthresh = 983 (rt->rt_rmx.rmx_ssthresh + i) / 2; 984 else 985 rt->rt_rmx.rmx_ssthresh = i; 986 tcpstat.tcps_cachedssthresh++; 987 } 988 } 989 990 no_valid_rt: 991 /* free the reassembly queue, if any */ 992 while((q = TAILQ_FIRST(&tp->t_segq)) != NULL) { 993 TAILQ_REMOVE(&tp->t_segq, q, tqe_q); 994 m_freem(q->tqe_m); 995 kfree(q, M_TSEGQ); 996 atomic_add_int(&tcp_reass_qsize, -1); 997 } 998 /* throw away SACK blocks in scoreboard*/ 999 if (TCP_DO_SACK(tp)) 1000 tcp_sack_destroy(&tp->scb); 1001 1002 inp->inp_ppcb = NULL; 1003 soisdisconnected(so); 1004 /* note: pcb detached later on */ 1005 1006 tcp_destroy_timermsg(tp); 1007 1008 if (tp->t_flags & TF_LISTEN) 1009 syncache_destroy(tp); 1010 1011 /* 1012 * NOTE: 1013 * pcbdetach removes any wildcard hash entry on the current CPU. 1014 */ 1015 #ifdef INET6 1016 if (isafinet6) 1017 in6_pcbdetach(inp); 1018 else 1019 #endif 1020 in_pcbdetach(inp); 1021 1022 tcpstat.tcps_closed++; 1023 return (NULL); 1024 } 1025 1026 static __inline void 1027 tcp_drain_oncpu(struct inpcbhead *head) 1028 { 1029 struct inpcb *marker; 1030 struct inpcb *inpb; 1031 struct tcpcb *tcpb; 1032 struct tseg_qent *te; 1033 1034 /* 1035 * Allows us to block while running the list 1036 */ 1037 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO); 1038 marker->inp_flags |= INP_PLACEMARKER; 1039 LIST_INSERT_HEAD(head, marker, inp_list); 1040 1041 while ((inpb = LIST_NEXT(marker, inp_list)) != NULL) { 1042 if ((inpb->inp_flags & INP_PLACEMARKER) == 0 && 1043 (tcpb = intotcpcb(inpb)) != NULL && 1044 (te = TAILQ_FIRST(&tcpb->t_segq)) != NULL) { 1045 TAILQ_REMOVE(&tcpb->t_segq, te, tqe_q); 1046 if (te->tqe_th->th_flags & TH_FIN) 1047 tcpb->t_flags &= ~TF_QUEDFIN; 1048 m_freem(te->tqe_m); 1049 kfree(te, M_TSEGQ); 1050 atomic_add_int(&tcp_reass_qsize, -1); 1051 /* retry */ 1052 } else { 1053 LIST_REMOVE(marker, inp_list); 1054 LIST_INSERT_AFTER(inpb, marker, inp_list); 1055 } 1056 } 1057 LIST_REMOVE(marker, inp_list); 1058 kfree(marker, M_TEMP); 1059 } 1060 1061 #ifdef SMP 1062 struct netmsg_tcp_drain { 1063 struct netmsg_base base; 1064 struct inpcbhead *nm_head; 1065 }; 1066 1067 static void 1068 tcp_drain_handler(netmsg_t msg) 1069 { 1070 struct netmsg_tcp_drain *nm = (void *)msg; 1071 1072 tcp_drain_oncpu(nm->nm_head); 1073 lwkt_replymsg(&nm->base.lmsg, 0); 1074 } 1075 #endif 1076 1077 void 1078 tcp_drain(void) 1079 { 1080 #ifdef SMP 1081 int cpu; 1082 #endif 1083 1084 if (!do_tcpdrain) 1085 return; 1086 1087 /* 1088 * Walk the tcpbs, if existing, and flush the reassembly queue, 1089 * if there is one... 1090 * XXX: The "Net/3" implementation doesn't imply that the TCP 1091 * reassembly queue should be flushed, but in a situation 1092 * where we're really low on mbufs, this is potentially 1093 * useful. 1094 */ 1095 #ifdef SMP 1096 for (cpu = 0; cpu < ncpus2; cpu++) { 1097 struct netmsg_tcp_drain *nm; 1098 1099 if (cpu == mycpu->gd_cpuid) { 1100 tcp_drain_oncpu(&tcbinfo[cpu].pcblisthead); 1101 } else { 1102 nm = kmalloc(sizeof(struct netmsg_tcp_drain), 1103 M_LWKTMSG, M_NOWAIT); 1104 if (nm == NULL) 1105 continue; 1106 netmsg_init(&nm->base, NULL, &netisr_afree_rport, 1107 0, tcp_drain_handler); 1108 nm->nm_head = &tcbinfo[cpu].pcblisthead; 1109 lwkt_sendmsg(cpu_portfn(cpu), &nm->base.lmsg); 1110 } 1111 } 1112 #else 1113 tcp_drain_oncpu(&tcbinfo[0].pcblisthead); 1114 #endif 1115 } 1116 1117 /* 1118 * Notify a tcp user of an asynchronous error; 1119 * store error as soft error, but wake up user 1120 * (for now, won't do anything until can select for soft error). 1121 * 1122 * Do not wake up user since there currently is no mechanism for 1123 * reporting soft errors (yet - a kqueue filter may be added). 1124 */ 1125 static void 1126 tcp_notify(struct inpcb *inp, int error) 1127 { 1128 struct tcpcb *tp = intotcpcb(inp); 1129 1130 /* 1131 * Ignore some errors if we are hooked up. 1132 * If connection hasn't completed, has retransmitted several times, 1133 * and receives a second error, give up now. This is better 1134 * than waiting a long time to establish a connection that 1135 * can never complete. 1136 */ 1137 if (tp->t_state == TCPS_ESTABLISHED && 1138 (error == EHOSTUNREACH || error == ENETUNREACH || 1139 error == EHOSTDOWN)) { 1140 return; 1141 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 1142 tp->t_softerror) 1143 tcp_drop(tp, error); 1144 else 1145 tp->t_softerror = error; 1146 #if 0 1147 wakeup(&so->so_timeo); 1148 sorwakeup(so); 1149 sowwakeup(so); 1150 #endif 1151 } 1152 1153 static int 1154 tcp_pcblist(SYSCTL_HANDLER_ARGS) 1155 { 1156 int error, i, n; 1157 struct inpcb *marker; 1158 struct inpcb *inp; 1159 globaldata_t gd; 1160 int origcpu, ccpu; 1161 1162 error = 0; 1163 n = 0; 1164 1165 /* 1166 * The process of preparing the TCB list is too time-consuming and 1167 * resource-intensive to repeat twice on every request. 1168 */ 1169 if (req->oldptr == NULL) { 1170 for (ccpu = 0; ccpu < ncpus; ++ccpu) { 1171 gd = globaldata_find(ccpu); 1172 n += tcbinfo[gd->gd_cpuid].ipi_count; 1173 } 1174 req->oldidx = (n + n/8 + 10) * sizeof(struct xtcpcb); 1175 return (0); 1176 } 1177 1178 if (req->newptr != NULL) 1179 return (EPERM); 1180 1181 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO); 1182 marker->inp_flags |= INP_PLACEMARKER; 1183 1184 /* 1185 * OK, now we're committed to doing something. Run the inpcb list 1186 * for each cpu in the system and construct the output. Use a 1187 * list placemarker to deal with list changes occuring during 1188 * copyout blockages (but otherwise depend on being on the correct 1189 * cpu to avoid races). 1190 */ 1191 origcpu = mycpu->gd_cpuid; 1192 for (ccpu = 1; ccpu <= ncpus && error == 0; ++ccpu) { 1193 globaldata_t rgd; 1194 caddr_t inp_ppcb; 1195 struct xtcpcb xt; 1196 int cpu_id; 1197 1198 cpu_id = (origcpu + ccpu) % ncpus; 1199 if ((smp_active_mask & CPUMASK(cpu_id)) == 0) 1200 continue; 1201 rgd = globaldata_find(cpu_id); 1202 lwkt_setcpu_self(rgd); 1203 1204 n = tcbinfo[cpu_id].ipi_count; 1205 1206 LIST_INSERT_HEAD(&tcbinfo[cpu_id].pcblisthead, marker, inp_list); 1207 i = 0; 1208 while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) { 1209 /* 1210 * process a snapshot of pcbs, ignoring placemarkers 1211 * and using our own to allow SYSCTL_OUT to block. 1212 */ 1213 LIST_REMOVE(marker, inp_list); 1214 LIST_INSERT_AFTER(inp, marker, inp_list); 1215 1216 if (inp->inp_flags & INP_PLACEMARKER) 1217 continue; 1218 if (prison_xinpcb(req->td, inp)) 1219 continue; 1220 1221 xt.xt_len = sizeof xt; 1222 bcopy(inp, &xt.xt_inp, sizeof *inp); 1223 inp_ppcb = inp->inp_ppcb; 1224 if (inp_ppcb != NULL) 1225 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1226 else 1227 bzero(&xt.xt_tp, sizeof xt.xt_tp); 1228 if (inp->inp_socket) 1229 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1230 if ((error = SYSCTL_OUT(req, &xt, sizeof xt)) != 0) 1231 break; 1232 ++i; 1233 } 1234 LIST_REMOVE(marker, inp_list); 1235 if (error == 0 && i < n) { 1236 bzero(&xt, sizeof xt); 1237 xt.xt_len = sizeof xt; 1238 while (i < n) { 1239 error = SYSCTL_OUT(req, &xt, sizeof xt); 1240 if (error) 1241 break; 1242 ++i; 1243 } 1244 } 1245 } 1246 1247 /* 1248 * Make sure we are on the same cpu we were on originally, since 1249 * higher level callers expect this. Also don't pollute caches with 1250 * migrated userland data by (eventually) returning to userland 1251 * on a different cpu. 1252 */ 1253 lwkt_setcpu_self(globaldata_find(origcpu)); 1254 kfree(marker, M_TEMP); 1255 return (error); 1256 } 1257 1258 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1259 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1260 1261 static int 1262 tcp_getcred(SYSCTL_HANDLER_ARGS) 1263 { 1264 struct sockaddr_in addrs[2]; 1265 struct inpcb *inp; 1266 int cpu; 1267 int error; 1268 1269 error = priv_check(req->td, PRIV_ROOT); 1270 if (error != 0) 1271 return (error); 1272 error = SYSCTL_IN(req, addrs, sizeof addrs); 1273 if (error != 0) 1274 return (error); 1275 crit_enter(); 1276 cpu = tcp_addrcpu(addrs[1].sin_addr.s_addr, addrs[1].sin_port, 1277 addrs[0].sin_addr.s_addr, addrs[0].sin_port); 1278 inp = in_pcblookup_hash(&tcbinfo[cpu], addrs[1].sin_addr, 1279 addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1280 if (inp == NULL || inp->inp_socket == NULL) { 1281 error = ENOENT; 1282 goto out; 1283 } 1284 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred)); 1285 out: 1286 crit_exit(); 1287 return (error); 1288 } 1289 1290 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW), 1291 0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection"); 1292 1293 #ifdef INET6 1294 static int 1295 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1296 { 1297 struct sockaddr_in6 addrs[2]; 1298 struct inpcb *inp; 1299 int error; 1300 boolean_t mapped = FALSE; 1301 1302 error = priv_check(req->td, PRIV_ROOT); 1303 if (error != 0) 1304 return (error); 1305 error = SYSCTL_IN(req, addrs, sizeof addrs); 1306 if (error != 0) 1307 return (error); 1308 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1309 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1310 mapped = TRUE; 1311 else 1312 return (EINVAL); 1313 } 1314 crit_enter(); 1315 if (mapped) { 1316 inp = in_pcblookup_hash(&tcbinfo[0], 1317 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1318 addrs[1].sin6_port, 1319 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1320 addrs[0].sin6_port, 1321 0, NULL); 1322 } else { 1323 inp = in6_pcblookup_hash(&tcbinfo[0], 1324 &addrs[1].sin6_addr, addrs[1].sin6_port, 1325 &addrs[0].sin6_addr, addrs[0].sin6_port, 1326 0, NULL); 1327 } 1328 if (inp == NULL || inp->inp_socket == NULL) { 1329 error = ENOENT; 1330 goto out; 1331 } 1332 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred)); 1333 out: 1334 crit_exit(); 1335 return (error); 1336 } 1337 1338 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW), 1339 0, 0, 1340 tcp6_getcred, "S,ucred", "Get the ucred of a TCP6 connection"); 1341 #endif 1342 1343 struct netmsg_tcp_notify { 1344 struct netmsg_base base; 1345 void (*nm_notify)(struct inpcb *, int); 1346 struct in_addr nm_faddr; 1347 int nm_arg; 1348 }; 1349 1350 static void 1351 tcp_notifyall_oncpu(netmsg_t msg) 1352 { 1353 struct netmsg_tcp_notify *nm = (struct netmsg_tcp_notify *)msg; 1354 int nextcpu; 1355 1356 in_pcbnotifyall(&tcbinfo[mycpuid].pcblisthead, nm->nm_faddr, 1357 nm->nm_arg, nm->nm_notify); 1358 1359 nextcpu = mycpuid + 1; 1360 if (nextcpu < ncpus2) 1361 lwkt_forwardmsg(cpu_portfn(nextcpu), &nm->base.lmsg); 1362 else 1363 lwkt_replymsg(&nm->base.lmsg, 0); 1364 } 1365 1366 void 1367 tcp_ctlinput(netmsg_t msg) 1368 { 1369 int cmd = msg->ctlinput.nm_cmd; 1370 struct sockaddr *sa = msg->ctlinput.nm_arg; 1371 struct ip *ip = msg->ctlinput.nm_extra; 1372 struct tcphdr *th; 1373 struct in_addr faddr; 1374 struct inpcb *inp; 1375 struct tcpcb *tp; 1376 void (*notify)(struct inpcb *, int) = tcp_notify; 1377 tcp_seq icmpseq; 1378 int arg, cpu; 1379 1380 if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) { 1381 goto done; 1382 } 1383 1384 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1385 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1386 goto done; 1387 1388 arg = inetctlerrmap[cmd]; 1389 if (cmd == PRC_QUENCH) { 1390 notify = tcp_quench; 1391 } else if (icmp_may_rst && 1392 (cmd == PRC_UNREACH_ADMIN_PROHIB || 1393 cmd == PRC_UNREACH_PORT || 1394 cmd == PRC_TIMXCEED_INTRANS) && 1395 ip != NULL) { 1396 notify = tcp_drop_syn_sent; 1397 } else if (cmd == PRC_MSGSIZE) { 1398 struct icmp *icmp = (struct icmp *) 1399 ((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 1400 1401 arg = ntohs(icmp->icmp_nextmtu); 1402 notify = tcp_mtudisc; 1403 } else if (PRC_IS_REDIRECT(cmd)) { 1404 ip = NULL; 1405 notify = in_rtchange; 1406 } else if (cmd == PRC_HOSTDEAD) { 1407 ip = NULL; 1408 } 1409 1410 if (ip != NULL) { 1411 crit_enter(); 1412 th = (struct tcphdr *)((caddr_t)ip + 1413 (IP_VHL_HL(ip->ip_vhl) << 2)); 1414 cpu = tcp_addrcpu(faddr.s_addr, th->th_dport, 1415 ip->ip_src.s_addr, th->th_sport); 1416 inp = in_pcblookup_hash(&tcbinfo[cpu], faddr, th->th_dport, 1417 ip->ip_src, th->th_sport, 0, NULL); 1418 if ((inp != NULL) && (inp->inp_socket != NULL)) { 1419 icmpseq = htonl(th->th_seq); 1420 tp = intotcpcb(inp); 1421 if (SEQ_GEQ(icmpseq, tp->snd_una) && 1422 SEQ_LT(icmpseq, tp->snd_max)) 1423 (*notify)(inp, arg); 1424 } else { 1425 struct in_conninfo inc; 1426 1427 inc.inc_fport = th->th_dport; 1428 inc.inc_lport = th->th_sport; 1429 inc.inc_faddr = faddr; 1430 inc.inc_laddr = ip->ip_src; 1431 #ifdef INET6 1432 inc.inc_isipv6 = 0; 1433 #endif 1434 syncache_unreach(&inc, th); 1435 } 1436 crit_exit(); 1437 } else { 1438 struct netmsg_tcp_notify *nm; 1439 1440 KKASSERT(&curthread->td_msgport == cpu_portfn(0)); 1441 nm = kmalloc(sizeof(*nm), M_LWKTMSG, M_INTWAIT); 1442 netmsg_init(&nm->base, NULL, &netisr_afree_rport, 1443 0, tcp_notifyall_oncpu); 1444 nm->nm_faddr = faddr; 1445 nm->nm_arg = arg; 1446 nm->nm_notify = notify; 1447 1448 lwkt_sendmsg(cpu_portfn(0), &nm->base.lmsg); 1449 } 1450 done: 1451 lwkt_replymsg(&msg->lmsg, 0); 1452 } 1453 1454 #ifdef INET6 1455 1456 void 1457 tcp6_ctlinput(netmsg_t msg) 1458 { 1459 int cmd = msg->ctlinput.nm_cmd; 1460 struct sockaddr *sa = msg->ctlinput.nm_arg; 1461 void *d = msg->ctlinput.nm_extra; 1462 struct tcphdr th; 1463 void (*notify) (struct inpcb *, int) = tcp_notify; 1464 struct ip6_hdr *ip6; 1465 struct mbuf *m; 1466 struct ip6ctlparam *ip6cp = NULL; 1467 const struct sockaddr_in6 *sa6_src = NULL; 1468 int off; 1469 struct tcp_portonly { 1470 u_int16_t th_sport; 1471 u_int16_t th_dport; 1472 } *thp; 1473 int arg; 1474 1475 if (sa->sa_family != AF_INET6 || 1476 sa->sa_len != sizeof(struct sockaddr_in6)) { 1477 goto out; 1478 } 1479 1480 arg = 0; 1481 if (cmd == PRC_QUENCH) 1482 notify = tcp_quench; 1483 else if (cmd == PRC_MSGSIZE) { 1484 struct ip6ctlparam *ip6cp = d; 1485 struct icmp6_hdr *icmp6 = ip6cp->ip6c_icmp6; 1486 1487 arg = ntohl(icmp6->icmp6_mtu); 1488 notify = tcp_mtudisc; 1489 } else if (!PRC_IS_REDIRECT(cmd) && 1490 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) { 1491 goto out; 1492 } 1493 1494 /* if the parameter is from icmp6, decode it. */ 1495 if (d != NULL) { 1496 ip6cp = (struct ip6ctlparam *)d; 1497 m = ip6cp->ip6c_m; 1498 ip6 = ip6cp->ip6c_ip6; 1499 off = ip6cp->ip6c_off; 1500 sa6_src = ip6cp->ip6c_src; 1501 } else { 1502 m = NULL; 1503 ip6 = NULL; 1504 off = 0; /* fool gcc */ 1505 sa6_src = &sa6_any; 1506 } 1507 1508 if (ip6 != NULL) { 1509 struct in_conninfo inc; 1510 /* 1511 * XXX: We assume that when IPV6 is non NULL, 1512 * M and OFF are valid. 1513 */ 1514 1515 /* check if we can safely examine src and dst ports */ 1516 if (m->m_pkthdr.len < off + sizeof *thp) 1517 goto out; 1518 1519 bzero(&th, sizeof th); 1520 m_copydata(m, off, sizeof *thp, (caddr_t)&th); 1521 1522 in6_pcbnotify(&tcbinfo[0].pcblisthead, sa, th.th_dport, 1523 (struct sockaddr *)ip6cp->ip6c_src, 1524 th.th_sport, cmd, arg, notify); 1525 1526 inc.inc_fport = th.th_dport; 1527 inc.inc_lport = th.th_sport; 1528 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1529 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1530 inc.inc_isipv6 = 1; 1531 syncache_unreach(&inc, &th); 1532 } else { 1533 in6_pcbnotify(&tcbinfo[0].pcblisthead, sa, 0, 1534 (const struct sockaddr *)sa6_src, 0, cmd, arg, notify); 1535 } 1536 out: 1537 lwkt_replymsg(&msg->ctlinput.base.lmsg, 0); 1538 } 1539 1540 #endif 1541 1542 /* 1543 * Following is where TCP initial sequence number generation occurs. 1544 * 1545 * There are two places where we must use initial sequence numbers: 1546 * 1. In SYN-ACK packets. 1547 * 2. In SYN packets. 1548 * 1549 * All ISNs for SYN-ACK packets are generated by the syncache. See 1550 * tcp_syncache.c for details. 1551 * 1552 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1553 * depends on this property. In addition, these ISNs should be 1554 * unguessable so as to prevent connection hijacking. To satisfy 1555 * the requirements of this situation, the algorithm outlined in 1556 * RFC 1948 is used to generate sequence numbers. 1557 * 1558 * Implementation details: 1559 * 1560 * Time is based off the system timer, and is corrected so that it 1561 * increases by one megabyte per second. This allows for proper 1562 * recycling on high speed LANs while still leaving over an hour 1563 * before rollover. 1564 * 1565 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1566 * between seeding of isn_secret. This is normally set to zero, 1567 * as reseeding should not be necessary. 1568 * 1569 */ 1570 1571 #define ISN_BYTES_PER_SECOND 1048576 1572 1573 u_char isn_secret[32]; 1574 int isn_last_reseed; 1575 MD5_CTX isn_ctx; 1576 1577 tcp_seq 1578 tcp_new_isn(struct tcpcb *tp) 1579 { 1580 u_int32_t md5_buffer[4]; 1581 tcp_seq new_isn; 1582 1583 /* Seed if this is the first use, reseed if requested. */ 1584 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1585 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1586 < (u_int)ticks))) { 1587 read_random_unlimited(&isn_secret, sizeof isn_secret); 1588 isn_last_reseed = ticks; 1589 } 1590 1591 /* Compute the md5 hash and return the ISN. */ 1592 MD5Init(&isn_ctx); 1593 MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_fport, sizeof(u_short)); 1594 MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_lport, sizeof(u_short)); 1595 #ifdef INET6 1596 if (tp->t_inpcb->inp_vflag & INP_IPV6) { 1597 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1598 sizeof(struct in6_addr)); 1599 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1600 sizeof(struct in6_addr)); 1601 } else 1602 #endif 1603 { 1604 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1605 sizeof(struct in_addr)); 1606 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1607 sizeof(struct in_addr)); 1608 } 1609 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1610 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1611 new_isn = (tcp_seq) md5_buffer[0]; 1612 new_isn += ticks * (ISN_BYTES_PER_SECOND / hz); 1613 return (new_isn); 1614 } 1615 1616 /* 1617 * When a source quench is received, close congestion window 1618 * to one segment. We will gradually open it again as we proceed. 1619 */ 1620 void 1621 tcp_quench(struct inpcb *inp, int error) 1622 { 1623 struct tcpcb *tp = intotcpcb(inp); 1624 1625 if (tp != NULL) { 1626 tp->snd_cwnd = tp->t_maxseg; 1627 tp->snd_wacked = 0; 1628 } 1629 } 1630 1631 /* 1632 * When a specific ICMP unreachable message is received and the 1633 * connection state is SYN-SENT, drop the connection. This behavior 1634 * is controlled by the icmp_may_rst sysctl. 1635 */ 1636 void 1637 tcp_drop_syn_sent(struct inpcb *inp, int error) 1638 { 1639 struct tcpcb *tp = intotcpcb(inp); 1640 1641 if ((tp != NULL) && (tp->t_state == TCPS_SYN_SENT)) 1642 tcp_drop(tp, error); 1643 } 1644 1645 /* 1646 * When a `need fragmentation' ICMP is received, update our idea of the MSS 1647 * based on the new value in the route. Also nudge TCP to send something, 1648 * since we know the packet we just sent was dropped. 1649 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1650 */ 1651 void 1652 tcp_mtudisc(struct inpcb *inp, int mtu) 1653 { 1654 struct tcpcb *tp = intotcpcb(inp); 1655 struct rtentry *rt; 1656 struct socket *so = inp->inp_socket; 1657 int maxopd, mss; 1658 #ifdef INET6 1659 boolean_t isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0); 1660 #else 1661 const boolean_t isipv6 = FALSE; 1662 #endif 1663 1664 if (tp == NULL) 1665 return; 1666 1667 /* 1668 * If no MTU is provided in the ICMP message, use the 1669 * next lower likely value, as specified in RFC 1191. 1670 */ 1671 if (mtu == 0) { 1672 int oldmtu; 1673 1674 oldmtu = tp->t_maxopd + 1675 (isipv6 ? 1676 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1677 sizeof(struct tcpiphdr)); 1678 mtu = ip_next_mtu(oldmtu, 0); 1679 } 1680 1681 if (isipv6) 1682 rt = tcp_rtlookup6(&inp->inp_inc); 1683 else 1684 rt = tcp_rtlookup(&inp->inp_inc); 1685 if (rt != NULL) { 1686 if (rt->rt_rmx.rmx_mtu != 0 && rt->rt_rmx.rmx_mtu < mtu) 1687 mtu = rt->rt_rmx.rmx_mtu; 1688 1689 maxopd = mtu - 1690 (isipv6 ? 1691 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1692 sizeof(struct tcpiphdr)); 1693 1694 /* 1695 * XXX - The following conditional probably violates the TCP 1696 * spec. The problem is that, since we don't know the 1697 * other end's MSS, we are supposed to use a conservative 1698 * default. But, if we do that, then MTU discovery will 1699 * never actually take place, because the conservative 1700 * default is much less than the MTUs typically seen 1701 * on the Internet today. For the moment, we'll sweep 1702 * this under the carpet. 1703 * 1704 * The conservative default might not actually be a problem 1705 * if the only case this occurs is when sending an initial 1706 * SYN with options and data to a host we've never talked 1707 * to before. Then, they will reply with an MSS value which 1708 * will get recorded and the new parameters should get 1709 * recomputed. For Further Study. 1710 */ 1711 if (rt->rt_rmx.rmx_mssopt && rt->rt_rmx.rmx_mssopt < maxopd) 1712 maxopd = rt->rt_rmx.rmx_mssopt; 1713 } else 1714 maxopd = mtu - 1715 (isipv6 ? 1716 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1717 sizeof(struct tcpiphdr)); 1718 1719 if (tp->t_maxopd <= maxopd) 1720 return; 1721 tp->t_maxopd = maxopd; 1722 1723 mss = maxopd; 1724 if ((tp->t_flags & (TF_REQ_TSTMP | TF_RCVD_TSTMP | TF_NOOPT)) == 1725 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 1726 mss -= TCPOLEN_TSTAMP_APPA; 1727 1728 /* round down to multiple of MCLBYTES */ 1729 #if (MCLBYTES & (MCLBYTES - 1)) == 0 /* test if MCLBYTES power of 2 */ 1730 if (mss > MCLBYTES) 1731 mss &= ~(MCLBYTES - 1); 1732 #else 1733 if (mss > MCLBYTES) 1734 mss = (mss / MCLBYTES) * MCLBYTES; 1735 #endif 1736 1737 if (so->so_snd.ssb_hiwat < mss) 1738 mss = so->so_snd.ssb_hiwat; 1739 1740 tp->t_maxseg = mss; 1741 tp->t_rtttime = 0; 1742 tp->snd_nxt = tp->snd_una; 1743 tcp_output(tp); 1744 tcpstat.tcps_mturesent++; 1745 } 1746 1747 /* 1748 * Look-up the routing entry to the peer of this inpcb. If no route 1749 * is found and it cannot be allocated the return NULL. This routine 1750 * is called by TCP routines that access the rmx structure and by tcp_mss 1751 * to get the interface MTU. 1752 */ 1753 struct rtentry * 1754 tcp_rtlookup(struct in_conninfo *inc) 1755 { 1756 struct route *ro = &inc->inc_route; 1757 1758 if (ro->ro_rt == NULL || !(ro->ro_rt->rt_flags & RTF_UP)) { 1759 /* No route yet, so try to acquire one */ 1760 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1761 /* 1762 * unused portions of the structure MUST be zero'd 1763 * out because rtalloc() treats it as opaque data 1764 */ 1765 bzero(&ro->ro_dst, sizeof(struct sockaddr_in)); 1766 ro->ro_dst.sa_family = AF_INET; 1767 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 1768 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1769 inc->inc_faddr; 1770 rtalloc(ro); 1771 } 1772 } 1773 return (ro->ro_rt); 1774 } 1775 1776 #ifdef INET6 1777 struct rtentry * 1778 tcp_rtlookup6(struct in_conninfo *inc) 1779 { 1780 struct route_in6 *ro6 = &inc->inc6_route; 1781 1782 if (ro6->ro_rt == NULL || !(ro6->ro_rt->rt_flags & RTF_UP)) { 1783 /* No route yet, so try to acquire one */ 1784 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1785 /* 1786 * unused portions of the structure MUST be zero'd 1787 * out because rtalloc() treats it as opaque data 1788 */ 1789 bzero(&ro6->ro_dst, sizeof(struct sockaddr_in6)); 1790 ro6->ro_dst.sin6_family = AF_INET6; 1791 ro6->ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1792 ro6->ro_dst.sin6_addr = inc->inc6_faddr; 1793 rtalloc((struct route *)ro6); 1794 } 1795 } 1796 return (ro6->ro_rt); 1797 } 1798 #endif 1799 1800 #ifdef IPSEC 1801 /* compute ESP/AH header size for TCP, including outer IP header. */ 1802 size_t 1803 ipsec_hdrsiz_tcp(struct tcpcb *tp) 1804 { 1805 struct inpcb *inp; 1806 struct mbuf *m; 1807 size_t hdrsiz; 1808 struct ip *ip; 1809 struct tcphdr *th; 1810 1811 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1812 return (0); 1813 MGETHDR(m, MB_DONTWAIT, MT_DATA); 1814 if (!m) 1815 return (0); 1816 1817 #ifdef INET6 1818 if (inp->inp_vflag & INP_IPV6) { 1819 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 1820 1821 th = (struct tcphdr *)(ip6 + 1); 1822 m->m_pkthdr.len = m->m_len = 1823 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1824 tcp_fillheaders(tp, ip6, th, FALSE); 1825 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1826 } else 1827 #endif 1828 { 1829 ip = mtod(m, struct ip *); 1830 th = (struct tcphdr *)(ip + 1); 1831 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1832 tcp_fillheaders(tp, ip, th, FALSE); 1833 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1834 } 1835 1836 m_free(m); 1837 return (hdrsiz); 1838 } 1839 #endif 1840 1841 /* 1842 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1843 * 1844 * This code attempts to calculate the bandwidth-delay product as a 1845 * means of determining the optimal window size to maximize bandwidth, 1846 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1847 * routers. This code also does a fairly good job keeping RTTs in check 1848 * across slow links like modems. We implement an algorithm which is very 1849 * similar (but not meant to be) TCP/Vegas. The code operates on the 1850 * transmitter side of a TCP connection and so only effects the transmit 1851 * side of the connection. 1852 * 1853 * BACKGROUND: TCP makes no provision for the management of buffer space 1854 * at the end points or at the intermediate routers and switches. A TCP 1855 * stream, whether using NewReno or not, will eventually buffer as 1856 * many packets as it is able and the only reason this typically works is 1857 * due to the fairly small default buffers made available for a connection 1858 * (typicaly 16K or 32K). As machines use larger windows and/or window 1859 * scaling it is now fairly easy for even a single TCP connection to blow-out 1860 * all available buffer space not only on the local interface, but on 1861 * intermediate routers and switches as well. NewReno makes a misguided 1862 * attempt to 'solve' this problem by waiting for an actual failure to occur, 1863 * then backing off, then steadily increasing the window again until another 1864 * failure occurs, ad-infinitum. This results in terrible oscillation that 1865 * is only made worse as network loads increase and the idea of intentionally 1866 * blowing out network buffers is, frankly, a terrible way to manage network 1867 * resources. 1868 * 1869 * It is far better to limit the transmit window prior to the failure 1870 * condition being achieved. There are two general ways to do this: First 1871 * you can 'scan' through different transmit window sizes and locate the 1872 * point where the RTT stops increasing, indicating that you have filled the 1873 * pipe, then scan backwards until you note that RTT stops decreasing, then 1874 * repeat ad-infinitum. This method works in principle but has severe 1875 * implementation issues due to RTT variances, timer granularity, and 1876 * instability in the algorithm which can lead to many false positives and 1877 * create oscillations as well as interact badly with other TCP streams 1878 * implementing the same algorithm. 1879 * 1880 * The second method is to limit the window to the bandwidth delay product 1881 * of the link. This is the method we implement. RTT variances and our 1882 * own manipulation of the congestion window, bwnd, can potentially 1883 * destabilize the algorithm. For this reason we have to stabilize the 1884 * elements used to calculate the window. We do this by using the minimum 1885 * observed RTT, the long term average of the observed bandwidth, and 1886 * by adding two segments worth of slop. It isn't perfect but it is able 1887 * to react to changing conditions and gives us a very stable basis on 1888 * which to extend the algorithm. 1889 */ 1890 void 1891 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 1892 { 1893 u_long bw; 1894 u_long bwnd; 1895 int save_ticks; 1896 int delta_ticks; 1897 1898 /* 1899 * If inflight_enable is disabled in the middle of a tcp connection, 1900 * make sure snd_bwnd is effectively disabled. 1901 */ 1902 if (!tcp_inflight_enable) { 1903 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1904 tp->snd_bandwidth = 0; 1905 return; 1906 } 1907 1908 /* 1909 * Validate the delta time. If a connection is new or has been idle 1910 * a long time we have to reset the bandwidth calculator. 1911 */ 1912 save_ticks = ticks; 1913 delta_ticks = save_ticks - tp->t_bw_rtttime; 1914 if (tp->t_bw_rtttime == 0 || delta_ticks < 0 || delta_ticks > hz * 10) { 1915 tp->t_bw_rtttime = ticks; 1916 tp->t_bw_rtseq = ack_seq; 1917 if (tp->snd_bandwidth == 0) 1918 tp->snd_bandwidth = tcp_inflight_min; 1919 return; 1920 } 1921 if (delta_ticks == 0) 1922 return; 1923 1924 /* 1925 * Sanity check, plus ignore pure window update acks. 1926 */ 1927 if ((int)(ack_seq - tp->t_bw_rtseq) <= 0) 1928 return; 1929 1930 /* 1931 * Figure out the bandwidth. Due to the tick granularity this 1932 * is a very rough number and it MUST be averaged over a fairly 1933 * long period of time. XXX we need to take into account a link 1934 * that is not using all available bandwidth, but for now our 1935 * slop will ramp us up if this case occurs and the bandwidth later 1936 * increases. 1937 */ 1938 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / delta_ticks; 1939 tp->t_bw_rtttime = save_ticks; 1940 tp->t_bw_rtseq = ack_seq; 1941 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 1942 1943 tp->snd_bandwidth = bw; 1944 1945 /* 1946 * Calculate the semi-static bandwidth delay product, plus two maximal 1947 * segments. The additional slop puts us squarely in the sweet 1948 * spot and also handles the bandwidth run-up case. Without the 1949 * slop we could be locking ourselves into a lower bandwidth. 1950 * 1951 * Situations Handled: 1952 * (1) Prevents over-queueing of packets on LANs, especially on 1953 * high speed LANs, allowing larger TCP buffers to be 1954 * specified, and also does a good job preventing 1955 * over-queueing of packets over choke points like modems 1956 * (at least for the transmit side). 1957 * 1958 * (2) Is able to handle changing network loads (bandwidth 1959 * drops so bwnd drops, bandwidth increases so bwnd 1960 * increases). 1961 * 1962 * (3) Theoretically should stabilize in the face of multiple 1963 * connections implementing the same algorithm (this may need 1964 * a little work). 1965 * 1966 * (4) Stability value (defaults to 20 = 2 maximal packets) can 1967 * be adjusted with a sysctl but typically only needs to be on 1968 * very slow connections. A value no smaller then 5 should 1969 * be used, but only reduce this default if you have no other 1970 * choice. 1971 */ 1972 1973 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 1974 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + 1975 tcp_inflight_stab * (int)tp->t_maxseg / 10; 1976 #undef USERTT 1977 1978 if (tcp_inflight_debug > 0) { 1979 static int ltime; 1980 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 1981 ltime = ticks; 1982 kprintf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 1983 tp, bw, tp->t_rttbest, tp->t_srtt, bwnd); 1984 } 1985 } 1986 if ((long)bwnd < tcp_inflight_min) 1987 bwnd = tcp_inflight_min; 1988 if (bwnd > tcp_inflight_max) 1989 bwnd = tcp_inflight_max; 1990 if ((long)bwnd < tp->t_maxseg * 2) 1991 bwnd = tp->t_maxseg * 2; 1992 tp->snd_bwnd = bwnd; 1993 } 1994 1995 static void 1996 tcp_rmx_iwsegs(struct tcpcb *tp, u_long *maxsegs, u_long *capsegs) 1997 { 1998 struct rtentry *rt; 1999 struct inpcb *inp = tp->t_inpcb; 2000 #ifdef INET6 2001 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE); 2002 #else 2003 const boolean_t isipv6 = FALSE; 2004 #endif 2005 2006 /* XXX */ 2007 if (tcp_iw_maxsegs < TCP_IW_MAXSEGS_DFLT) 2008 tcp_iw_maxsegs = TCP_IW_MAXSEGS_DFLT; 2009 if (tcp_iw_capsegs < TCP_IW_CAPSEGS_DFLT) 2010 tcp_iw_capsegs = TCP_IW_CAPSEGS_DFLT; 2011 2012 if (isipv6) 2013 rt = tcp_rtlookup6(&inp->inp_inc); 2014 else 2015 rt = tcp_rtlookup(&inp->inp_inc); 2016 if (rt == NULL || 2017 rt->rt_rmx.rmx_iwmaxsegs < TCP_IW_MAXSEGS_DFLT || 2018 rt->rt_rmx.rmx_iwcapsegs < TCP_IW_CAPSEGS_DFLT) { 2019 *maxsegs = tcp_iw_maxsegs; 2020 *capsegs = tcp_iw_capsegs; 2021 return; 2022 } 2023 *maxsegs = rt->rt_rmx.rmx_iwmaxsegs; 2024 *capsegs = rt->rt_rmx.rmx_iwcapsegs; 2025 } 2026 2027 u_long 2028 tcp_initial_window(struct tcpcb *tp) 2029 { 2030 if (tcp_do_rfc3390) { 2031 /* 2032 * RFC3390: 2033 * "If the SYN or SYN/ACK is lost, the initial window 2034 * used by a sender after a correctly transmitted SYN 2035 * MUST be one segment consisting of MSS bytes." 2036 * 2037 * However, we do something a little bit more aggressive 2038 * then RFC3390 here: 2039 * - Only if time spent in the SYN or SYN|ACK retransmition 2040 * >= 3 seconds, the IW is reduced. We do this mainly 2041 * because when RFC3390 is published, the initial RTO is 2042 * still 3 seconds (the threshold we test here), while 2043 * after RFC6298, the initial RTO is 1 second. This 2044 * behaviour probably still falls within the spirit of 2045 * RFC3390. 2046 * - When IW is reduced, 2*MSS is used instead of 1*MSS. 2047 * Mainly to avoid sender and receiver deadlock until 2048 * delayed ACK timer expires. And even RFC2581 does not 2049 * try to reduce IW upon SYN or SYN|ACK retransmition 2050 * timeout. 2051 * 2052 * See also: 2053 * http://tools.ietf.org/html/draft-ietf-tcpm-initcwnd-03 2054 */ 2055 if (tp->t_rxtsyn >= TCPTV_RTOBASE3) { 2056 return (2 * tp->t_maxseg); 2057 } else { 2058 u_long maxsegs, capsegs; 2059 2060 tcp_rmx_iwsegs(tp, &maxsegs, &capsegs); 2061 return min(maxsegs * tp->t_maxseg, 2062 max(2 * tp->t_maxseg, capsegs * 1460)); 2063 } 2064 } else { 2065 /* 2066 * Even RFC2581 (back to 1999) allows 2*SMSS IW. 2067 * 2068 * Mainly to avoid sender and receiver deadlock 2069 * until delayed ACK timer expires. 2070 */ 2071 return (2 * tp->t_maxseg); 2072 } 2073 } 2074 2075 #ifdef TCP_SIGNATURE 2076 /* 2077 * Compute TCP-MD5 hash of a TCP segment. (RFC2385) 2078 * 2079 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2080 * When called from tcp_input(), we can be sure that th_sum has been 2081 * zeroed out and verified already. 2082 * 2083 * Return 0 if successful, otherwise return -1. 2084 * 2085 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2086 * search with the destination IP address, and a 'magic SPI' to be 2087 * determined by the application. This is hardcoded elsewhere to 1179 2088 * right now. Another branch of this code exists which uses the SPD to 2089 * specify per-application flows but it is unstable. 2090 */ 2091 int 2092 tcpsignature_compute( 2093 struct mbuf *m, /* mbuf chain */ 2094 int len, /* length of TCP data */ 2095 int optlen, /* length of TCP options */ 2096 u_char *buf, /* storage for MD5 digest */ 2097 u_int direction) /* direction of flow */ 2098 { 2099 struct ippseudo ippseudo; 2100 MD5_CTX ctx; 2101 int doff; 2102 struct ip *ip; 2103 struct ipovly *ipovly; 2104 struct secasvar *sav; 2105 struct tcphdr *th; 2106 #ifdef INET6 2107 struct ip6_hdr *ip6; 2108 struct in6_addr in6; 2109 uint32_t plen; 2110 uint16_t nhdr; 2111 #endif /* INET6 */ 2112 u_short savecsum; 2113 2114 KASSERT(m != NULL, ("passed NULL mbuf. Game over.")); 2115 KASSERT(buf != NULL, ("passed NULL storage pointer for MD5 signature")); 2116 /* 2117 * Extract the destination from the IP header in the mbuf. 2118 */ 2119 ip = mtod(m, struct ip *); 2120 #ifdef INET6 2121 ip6 = NULL; /* Make the compiler happy. */ 2122 #endif /* INET6 */ 2123 /* 2124 * Look up an SADB entry which matches the address found in 2125 * the segment. 2126 */ 2127 switch (IP_VHL_V(ip->ip_vhl)) { 2128 case IPVERSION: 2129 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, (caddr_t)&ip->ip_dst, 2130 IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2131 break; 2132 #ifdef INET6 2133 case (IPV6_VERSION >> 4): 2134 ip6 = mtod(m, struct ip6_hdr *); 2135 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, (caddr_t)&ip6->ip6_dst, 2136 IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2137 break; 2138 #endif /* INET6 */ 2139 default: 2140 return (EINVAL); 2141 /* NOTREACHED */ 2142 break; 2143 } 2144 if (sav == NULL) { 2145 kprintf("%s: SADB lookup failed\n", __func__); 2146 return (EINVAL); 2147 } 2148 MD5Init(&ctx); 2149 2150 /* 2151 * Step 1: Update MD5 hash with IP pseudo-header. 2152 * 2153 * XXX The ippseudo header MUST be digested in network byte order, 2154 * or else we'll fail the regression test. Assume all fields we've 2155 * been doing arithmetic on have been in host byte order. 2156 * XXX One cannot depend on ipovly->ih_len here. When called from 2157 * tcp_output(), the underlying ip_len member has not yet been set. 2158 */ 2159 switch (IP_VHL_V(ip->ip_vhl)) { 2160 case IPVERSION: 2161 ipovly = (struct ipovly *)ip; 2162 ippseudo.ippseudo_src = ipovly->ih_src; 2163 ippseudo.ippseudo_dst = ipovly->ih_dst; 2164 ippseudo.ippseudo_pad = 0; 2165 ippseudo.ippseudo_p = IPPROTO_TCP; 2166 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2167 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2168 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip)); 2169 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen; 2170 break; 2171 #ifdef INET6 2172 /* 2173 * RFC 2385, 2.0 Proposal 2174 * For IPv6, the pseudo-header is as described in RFC 2460, namely the 2175 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero- 2176 * extended next header value (to form 32 bits), and 32-bit segment 2177 * length. 2178 * Note: Upper-Layer Packet Length comes before Next Header. 2179 */ 2180 case (IPV6_VERSION >> 4): 2181 in6 = ip6->ip6_src; 2182 in6_clearscope(&in6); 2183 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2184 in6 = ip6->ip6_dst; 2185 in6_clearscope(&in6); 2186 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2187 plen = htonl(len + sizeof(struct tcphdr) + optlen); 2188 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t)); 2189 nhdr = 0; 2190 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2191 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2192 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2193 nhdr = IPPROTO_TCP; 2194 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2195 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr)); 2196 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen; 2197 break; 2198 #endif /* INET6 */ 2199 default: 2200 return (EINVAL); 2201 /* NOTREACHED */ 2202 break; 2203 } 2204 /* 2205 * Step 2: Update MD5 hash with TCP header, excluding options. 2206 * The TCP checksum must be set to zero. 2207 */ 2208 savecsum = th->th_sum; 2209 th->th_sum = 0; 2210 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2211 th->th_sum = savecsum; 2212 /* 2213 * Step 3: Update MD5 hash with TCP segment data. 2214 * Use m_apply() to avoid an early m_pullup(). 2215 */ 2216 if (len > 0) 2217 m_apply(m, doff, len, tcpsignature_apply, &ctx); 2218 /* 2219 * Step 4: Update MD5 hash with shared secret. 2220 */ 2221 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2222 MD5Final(buf, &ctx); 2223 key_sa_recordxfer(sav, m); 2224 key_freesav(sav); 2225 return (0); 2226 } 2227 2228 int 2229 tcpsignature_apply(void *fstate, void *data, unsigned int len) 2230 { 2231 2232 MD5Update((MD5_CTX *)fstate, (unsigned char *)data, len); 2233 return (0); 2234 } 2235 #endif /* TCP_SIGNATURE */ 2236 2237 boolean_t 2238 tcp_tso_pullup(struct mbuf **mp, int hoff, struct ip **ip0, int *iphlen0, 2239 struct tcphdr **th0, int *thoff0) 2240 { 2241 struct mbuf *m = *mp; 2242 struct ip *ip; 2243 int len, iphlen; 2244 struct tcphdr *th; 2245 int thoff; 2246 2247 len = hoff + sizeof(struct ip); 2248 2249 /* The fixed IP header must reside completely in the first mbuf. */ 2250 if (m->m_len < len) { 2251 m = m_pullup(m, len); 2252 if (m == NULL) 2253 goto fail; 2254 } 2255 2256 ip = mtodoff(m, struct ip *, hoff); 2257 iphlen = IP_VHL_HL(ip->ip_vhl) << 2; 2258 2259 /* The full IP header must reside completely in the one mbuf. */ 2260 if (m->m_len < hoff + iphlen) { 2261 m = m_pullup(m, hoff + iphlen); 2262 if (m == NULL) 2263 goto fail; 2264 ip = mtodoff(m, struct ip *, hoff); 2265 } 2266 2267 KASSERT(ip->ip_p == IPPROTO_TCP, ("not tcp %d", ip->ip_p)); 2268 2269 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr)) { 2270 m = m_pullup(m, hoff + iphlen + sizeof(struct tcphdr)); 2271 if (m == NULL) 2272 goto fail; 2273 ip = mtodoff(m, struct ip *, hoff); 2274 } 2275 2276 th = (struct tcphdr *)((caddr_t)ip + iphlen); 2277 thoff = th->th_off << 2; 2278 2279 if (m->m_len < hoff + iphlen + thoff) { 2280 m = m_pullup(m, hoff + iphlen + thoff); 2281 if (m == NULL) 2282 goto fail; 2283 ip = mtodoff(m, struct ip *, hoff); 2284 th = (struct tcphdr *)((caddr_t)ip + iphlen); 2285 } 2286 2287 *mp = m; 2288 *ip0 = ip; 2289 *iphlen0 = iphlen; 2290 *th0 = th; 2291 *thoff0 = thoff; 2292 return TRUE; 2293 2294 fail: 2295 if (m != NULL) 2296 m_freem(m); 2297 *mp = NULL; 2298 return FALSE; 2299 } 2300