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/socketops.h> 91 #include <sys/socketvar.h> 92 #include <sys/protosw.h> 93 #include <sys/random.h> 94 #include <sys/in_cksum.h> 95 #include <sys/ktr.h> 96 97 #include <net/route.h> 98 #include <net/if.h> 99 #include <net/netisr.h> 100 101 #define _IP_VHL 102 #include <netinet/in.h> 103 #include <netinet/in_systm.h> 104 #include <netinet/ip.h> 105 #include <netinet/ip6.h> 106 #include <netinet/in_pcb.h> 107 #include <netinet6/in6_pcb.h> 108 #include <netinet/in_var.h> 109 #include <netinet/ip_var.h> 110 #include <netinet6/ip6_var.h> 111 #include <netinet/ip_icmp.h> 112 #ifdef INET6 113 #include <netinet/icmp6.h> 114 #endif 115 #include <netinet/tcp.h> 116 #include <netinet/tcp_fsm.h> 117 #include <netinet/tcp_seq.h> 118 #include <netinet/tcp_timer.h> 119 #include <netinet/tcp_timer2.h> 120 #include <netinet/tcp_var.h> 121 #include <netinet6/tcp6_var.h> 122 #include <netinet/tcpip.h> 123 #ifdef TCPDEBUG 124 #include <netinet/tcp_debug.h> 125 #endif 126 #include <netinet6/ip6protosw.h> 127 128 #ifdef IPSEC 129 #include <netinet6/ipsec.h> 130 #include <netproto/key/key.h> 131 #ifdef INET6 132 #include <netinet6/ipsec6.h> 133 #endif 134 #endif 135 136 #ifdef FAST_IPSEC 137 #include <netproto/ipsec/ipsec.h> 138 #ifdef INET6 139 #include <netproto/ipsec/ipsec6.h> 140 #endif 141 #define IPSEC 142 #endif 143 144 #include <sys/md5.h> 145 #include <machine/smp.h> 146 147 #include <sys/msgport2.h> 148 #include <sys/mplock2.h> 149 #include <net/netmsg2.h> 150 151 #if !defined(KTR_TCP) 152 #define KTR_TCP KTR_ALL 153 #endif 154 /* 155 KTR_INFO_MASTER(tcp); 156 KTR_INFO(KTR_TCP, tcp, rxmsg, 0, "tcp getmsg", 0); 157 KTR_INFO(KTR_TCP, tcp, wait, 1, "tcp waitmsg", 0); 158 KTR_INFO(KTR_TCP, tcp, delayed, 2, "tcp execute delayed ops", 0); 159 #define logtcp(name) KTR_LOG(tcp_ ## name) 160 */ 161 162 #define TCP_IW_MAXSEGS_DFLT 4 163 #define TCP_IW_CAPSEGS_DFLT 3 164 165 struct inpcbinfo tcbinfo[MAXCPU]; 166 struct tcpcbackqhead tcpcbackq[MAXCPU]; 167 168 static struct lwkt_token tcp_port_token = 169 LWKT_TOKEN_INITIALIZER(tcp_port_token); 170 171 int tcp_mssdflt = TCP_MSS; 172 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 173 &tcp_mssdflt, 0, "Default TCP Maximum Segment Size"); 174 175 #ifdef INET6 176 int tcp_v6mssdflt = TCP6_MSS; 177 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLFLAG_RW, 178 &tcp_v6mssdflt, 0, "Default TCP Maximum Segment Size for IPv6"); 179 #endif 180 181 /* 182 * Minimum MSS we accept and use. This prevents DoS attacks where 183 * we are forced to a ridiculous low MSS like 20 and send hundreds 184 * of packets instead of one. The effect scales with the available 185 * bandwidth and quickly saturates the CPU and network interface 186 * with packet generation and sending. Set to zero to disable MINMSS 187 * checking. This setting prevents us from sending too small packets. 188 */ 189 int tcp_minmss = TCP_MINMSS; 190 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 191 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 192 193 #if 0 194 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 195 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 196 &tcp_rttdflt, 0, "Default maximum TCP Round Trip Time"); 197 #endif 198 199 int tcp_do_rfc1323 = 1; 200 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 201 &tcp_do_rfc1323, 0, "Enable rfc1323 (high performance TCP) extensions"); 202 203 static int tcp_tcbhashsize = 0; 204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD, 205 &tcp_tcbhashsize, 0, "Size of TCP control block hashtable"); 206 207 static int do_tcpdrain = 1; 208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 209 "Enable tcp_drain routine for extra help when low on mbufs"); 210 211 static int icmp_may_rst = 1; 212 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 213 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 214 215 static int tcp_isn_reseed_interval = 0; 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 217 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 218 219 /* 220 * TCP bandwidth limiting sysctls. The inflight limiter is now turned on 221 * by default, but with generous values which should allow maximal 222 * bandwidth. In particular, the slop defaults to 50 (5 packets). 223 * 224 * The reason for doing this is that the limiter is the only mechanism we 225 * have which seems to do a really good job preventing receiver RX rings 226 * on network interfaces from getting blown out. Even though GigE/10GigE 227 * is supposed to flow control it looks like either it doesn't actually 228 * do it or Open Source drivers do not properly enable it. 229 * 230 * People using the limiter to reduce bottlenecks on slower WAN connections 231 * should set the slop to 20 (2 packets). 232 */ 233 static int tcp_inflight_enable = 1; 234 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_enable, CTLFLAG_RW, 235 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 236 237 static int tcp_inflight_debug = 0; 238 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_debug, CTLFLAG_RW, 239 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 240 241 static int tcp_inflight_min = 6144; 242 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_min, CTLFLAG_RW, 243 &tcp_inflight_min, 0, "Lower bound for TCP inflight window"); 244 245 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 246 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_max, CTLFLAG_RW, 247 &tcp_inflight_max, 0, "Upper bound for TCP inflight window"); 248 249 static int tcp_inflight_stab = 50; 250 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_stab, CTLFLAG_RW, 251 &tcp_inflight_stab, 0, "Slop in maximal packets / 10 (20 = 3 packets)"); 252 253 static int tcp_do_rfc3390 = 1; 254 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 255 &tcp_do_rfc3390, 0, 256 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 257 258 static u_long tcp_iw_maxsegs = TCP_IW_MAXSEGS_DFLT; 259 SYSCTL_ULONG(_net_inet_tcp, OID_AUTO, iwmaxsegs, CTLFLAG_RW, 260 &tcp_iw_maxsegs, 0, "TCP IW segments max"); 261 262 static u_long tcp_iw_capsegs = TCP_IW_CAPSEGS_DFLT; 263 SYSCTL_ULONG(_net_inet_tcp, OID_AUTO, iwcapsegs, CTLFLAG_RW, 264 &tcp_iw_capsegs, 0, "TCP IW segments"); 265 266 int tcp_low_rtobase = 1; 267 SYSCTL_INT(_net_inet_tcp, OID_AUTO, low_rtobase, CTLFLAG_RW, 268 &tcp_low_rtobase, 0, "Lowering the Initial RTO (RFC 6298)"); 269 270 static int tcp_do_ncr = 1; 271 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ncr, CTLFLAG_RW, 272 &tcp_do_ncr, 0, "Non-Congestion Robustness (RFC 4653)"); 273 274 static MALLOC_DEFINE(M_TCPTEMP, "tcptemp", "TCP Templates for Keepalives"); 275 static struct malloc_pipe tcptemp_mpipe; 276 277 static void tcp_willblock(void); 278 static void tcp_notify (struct inpcb *, int); 279 280 struct tcp_stats tcpstats_percpu[MAXCPU]; 281 #ifdef SMP 282 static int 283 sysctl_tcpstats(SYSCTL_HANDLER_ARGS) 284 { 285 int cpu, error = 0; 286 287 for (cpu = 0; cpu < ncpus; ++cpu) { 288 if ((error = SYSCTL_OUT(req, &tcpstats_percpu[cpu], 289 sizeof(struct tcp_stats)))) 290 break; 291 if ((error = SYSCTL_IN(req, &tcpstats_percpu[cpu], 292 sizeof(struct tcp_stats)))) 293 break; 294 } 295 296 return (error); 297 } 298 SYSCTL_PROC(_net_inet_tcp, TCPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW), 299 0, 0, sysctl_tcpstats, "S,tcp_stats", "TCP statistics"); 300 #else 301 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 302 &tcpstat, tcp_stats, "TCP statistics"); 303 #endif 304 305 /* 306 * Target size of TCP PCB hash tables. Must be a power of two. 307 * 308 * Note that this can be overridden by the kernel environment 309 * variable net.inet.tcp.tcbhashsize 310 */ 311 #ifndef TCBHASHSIZE 312 #define TCBHASHSIZE 512 313 #endif 314 315 /* 316 * This is the actual shape of what we allocate using the zone 317 * allocator. Doing it this way allows us to protect both structures 318 * using the same generation count, and also eliminates the overhead 319 * of allocating tcpcbs separately. By hiding the structure here, 320 * we avoid changing most of the rest of the code (although it needs 321 * to be changed, eventually, for greater efficiency). 322 */ 323 #define ALIGNMENT 32 324 #define ALIGNM1 (ALIGNMENT - 1) 325 struct inp_tp { 326 union { 327 struct inpcb inp; 328 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1]; 329 } inp_tp_u; 330 struct tcpcb tcb; 331 struct tcp_callout inp_tp_rexmt; 332 struct tcp_callout inp_tp_persist; 333 struct tcp_callout inp_tp_keep; 334 struct tcp_callout inp_tp_2msl; 335 struct tcp_callout inp_tp_delack; 336 struct netmsg_tcp_timer inp_tp_timermsg; 337 }; 338 #undef ALIGNMENT 339 #undef ALIGNM1 340 341 /* 342 * Tcp initialization 343 */ 344 void 345 tcp_init(void) 346 { 347 struct inpcbporthead *porthashbase; 348 struct inpcbinfo *ticb; 349 u_long porthashmask; 350 int hashsize = TCBHASHSIZE; 351 int cpu; 352 353 /* 354 * note: tcptemp is used for keepalives, and it is ok for an 355 * allocation to fail so do not specify MPF_INT. 356 */ 357 mpipe_init(&tcptemp_mpipe, M_TCPTEMP, sizeof(struct tcptemp), 358 25, -1, 0, NULL, NULL, NULL); 359 360 tcp_delacktime = TCPTV_DELACK; 361 tcp_keepinit = TCPTV_KEEP_INIT; 362 tcp_keepidle = TCPTV_KEEP_IDLE; 363 tcp_keepintvl = TCPTV_KEEPINTVL; 364 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 365 tcp_msl = TCPTV_MSL; 366 tcp_rexmit_min = TCPTV_MIN; 367 tcp_rexmit_slop = TCPTV_CPU_VAR; 368 369 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 370 if (!powerof2(hashsize)) { 371 kprintf("WARNING: TCB hash size not a power of 2\n"); 372 hashsize = 512; /* safe default */ 373 } 374 tcp_tcbhashsize = hashsize; 375 porthashbase = hashinit(hashsize, M_PCB, &porthashmask); 376 377 for (cpu = 0; cpu < ncpus2; cpu++) { 378 ticb = &tcbinfo[cpu]; 379 in_pcbinfo_init(ticb); 380 ticb->cpu = cpu; 381 ticb->hashbase = hashinit(hashsize, M_PCB, 382 &ticb->hashmask); 383 ticb->porthashbase = porthashbase; 384 ticb->porthashmask = porthashmask; 385 ticb->porttoken = &tcp_port_token; 386 #if 0 387 ticb->porthashbase = hashinit(hashsize, M_PCB, 388 &ticb->porthashmask); 389 #endif 390 ticb->wildcardhashbase = hashinit(hashsize, M_PCB, 391 &ticb->wildcardhashmask); 392 ticb->ipi_size = sizeof(struct inp_tp); 393 TAILQ_INIT(&tcpcbackq[cpu]); 394 } 395 396 tcp_reass_maxseg = nmbclusters / 16; 397 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", &tcp_reass_maxseg); 398 399 #ifdef INET6 400 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 401 #else 402 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 403 #endif 404 if (max_protohdr < TCP_MINPROTOHDR) 405 max_protohdr = TCP_MINPROTOHDR; 406 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 407 panic("tcp_init"); 408 #undef TCP_MINPROTOHDR 409 410 /* 411 * Initialize TCP statistics counters for each CPU. 412 */ 413 #ifdef SMP 414 for (cpu = 0; cpu < ncpus; ++cpu) { 415 bzero(&tcpstats_percpu[cpu], sizeof(struct tcp_stats)); 416 } 417 #else 418 bzero(&tcpstat, sizeof(struct tcp_stats)); 419 #endif 420 421 syncache_init(); 422 netisr_register_rollup(tcp_willblock); 423 } 424 425 static void 426 tcp_willblock(void) 427 { 428 struct tcpcb *tp; 429 int cpu = mycpu->gd_cpuid; 430 431 while ((tp = TAILQ_FIRST(&tcpcbackq[cpu])) != NULL) { 432 KKASSERT(tp->t_flags & TF_ONOUTPUTQ); 433 tp->t_flags &= ~TF_ONOUTPUTQ; 434 TAILQ_REMOVE(&tcpcbackq[cpu], tp, t_outputq); 435 tcp_output(tp); 436 } 437 } 438 439 /* 440 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 441 * tcp_template used to store this data in mbufs, but we now recopy it out 442 * of the tcpcb each time to conserve mbufs. 443 */ 444 void 445 tcp_fillheaders(struct tcpcb *tp, void *ip_ptr, void *tcp_ptr, boolean_t tso) 446 { 447 struct inpcb *inp = tp->t_inpcb; 448 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr; 449 450 #ifdef INET6 451 if (inp->inp_vflag & INP_IPV6) { 452 struct ip6_hdr *ip6; 453 454 ip6 = (struct ip6_hdr *)ip_ptr; 455 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 456 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 457 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 458 (IPV6_VERSION & IPV6_VERSION_MASK); 459 ip6->ip6_nxt = IPPROTO_TCP; 460 ip6->ip6_plen = sizeof(struct tcphdr); 461 ip6->ip6_src = inp->in6p_laddr; 462 ip6->ip6_dst = inp->in6p_faddr; 463 tcp_hdr->th_sum = 0; 464 } else 465 #endif 466 { 467 struct ip *ip = (struct ip *) ip_ptr; 468 u_int plen; 469 470 ip->ip_vhl = IP_VHL_BORING; 471 ip->ip_tos = 0; 472 ip->ip_len = 0; 473 ip->ip_id = 0; 474 ip->ip_off = 0; 475 ip->ip_ttl = 0; 476 ip->ip_sum = 0; 477 ip->ip_p = IPPROTO_TCP; 478 ip->ip_src = inp->inp_laddr; 479 ip->ip_dst = inp->inp_faddr; 480 481 if (tso) 482 plen = htons(IPPROTO_TCP); 483 else 484 plen = htons(sizeof(struct tcphdr) + IPPROTO_TCP); 485 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, 486 ip->ip_dst.s_addr, plen); 487 } 488 489 tcp_hdr->th_sport = inp->inp_lport; 490 tcp_hdr->th_dport = inp->inp_fport; 491 tcp_hdr->th_seq = 0; 492 tcp_hdr->th_ack = 0; 493 tcp_hdr->th_x2 = 0; 494 tcp_hdr->th_off = 5; 495 tcp_hdr->th_flags = 0; 496 tcp_hdr->th_win = 0; 497 tcp_hdr->th_urp = 0; 498 } 499 500 /* 501 * Create template to be used to send tcp packets on a connection. 502 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 503 * use for this function is in keepalives, which use tcp_respond. 504 */ 505 struct tcptemp * 506 tcp_maketemplate(struct tcpcb *tp) 507 { 508 struct tcptemp *tmp; 509 510 if ((tmp = mpipe_alloc_nowait(&tcptemp_mpipe)) == NULL) 511 return (NULL); 512 tcp_fillheaders(tp, &tmp->tt_ipgen, &tmp->tt_t, FALSE); 513 return (tmp); 514 } 515 516 void 517 tcp_freetemplate(struct tcptemp *tmp) 518 { 519 mpipe_free(&tcptemp_mpipe, tmp); 520 } 521 522 /* 523 * Send a single message to the TCP at address specified by 524 * the given TCP/IP header. If m == NULL, then we make a copy 525 * of the tcpiphdr at ti and send directly to the addressed host. 526 * This is used to force keep alive messages out using the TCP 527 * template for a connection. If flags are given then we send 528 * a message back to the TCP which originated the * segment ti, 529 * and discard the mbuf containing it and any other attached mbufs. 530 * 531 * In any case the ack and sequence number of the transmitted 532 * segment are as specified by the parameters. 533 * 534 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 535 */ 536 void 537 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 538 tcp_seq ack, tcp_seq seq, int flags) 539 { 540 int tlen; 541 int win = 0; 542 struct route *ro = NULL; 543 struct route sro; 544 struct ip *ip = ipgen; 545 struct tcphdr *nth; 546 int ipflags = 0; 547 struct route_in6 *ro6 = NULL; 548 struct route_in6 sro6; 549 struct ip6_hdr *ip6 = ipgen; 550 boolean_t use_tmpro = TRUE; 551 #ifdef INET6 552 boolean_t isipv6 = (IP_VHL_V(ip->ip_vhl) == 6); 553 #else 554 const boolean_t isipv6 = FALSE; 555 #endif 556 557 if (tp != NULL) { 558 if (!(flags & TH_RST)) { 559 win = ssb_space(&tp->t_inpcb->inp_socket->so_rcv); 560 if (win < 0) 561 win = 0; 562 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 563 win = (long)TCP_MAXWIN << tp->rcv_scale; 564 } 565 /* 566 * Don't use the route cache of a listen socket, 567 * it is not MPSAFE; use temporary route cache. 568 */ 569 if (tp->t_state != TCPS_LISTEN) { 570 if (isipv6) 571 ro6 = &tp->t_inpcb->in6p_route; 572 else 573 ro = &tp->t_inpcb->inp_route; 574 use_tmpro = FALSE; 575 } 576 } 577 if (use_tmpro) { 578 if (isipv6) { 579 ro6 = &sro6; 580 bzero(ro6, sizeof *ro6); 581 } else { 582 ro = &sro; 583 bzero(ro, sizeof *ro); 584 } 585 } 586 if (m == NULL) { 587 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 588 if (m == NULL) 589 return; 590 tlen = 0; 591 m->m_data += max_linkhdr; 592 if (isipv6) { 593 bcopy(ip6, mtod(m, caddr_t), sizeof(struct ip6_hdr)); 594 ip6 = mtod(m, struct ip6_hdr *); 595 nth = (struct tcphdr *)(ip6 + 1); 596 } else { 597 bcopy(ip, mtod(m, caddr_t), sizeof(struct ip)); 598 ip = mtod(m, struct ip *); 599 nth = (struct tcphdr *)(ip + 1); 600 } 601 bcopy(th, nth, sizeof(struct tcphdr)); 602 flags = TH_ACK; 603 } else { 604 m_freem(m->m_next); 605 m->m_next = NULL; 606 m->m_data = (caddr_t)ipgen; 607 /* m_len is set later */ 608 tlen = 0; 609 #define xchg(a, b, type) { type t; t = a; a = b; b = t; } 610 if (isipv6) { 611 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 612 nth = (struct tcphdr *)(ip6 + 1); 613 } else { 614 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 615 nth = (struct tcphdr *)(ip + 1); 616 } 617 if (th != nth) { 618 /* 619 * this is usually a case when an extension header 620 * exists between the IPv6 header and the 621 * TCP header. 622 */ 623 nth->th_sport = th->th_sport; 624 nth->th_dport = th->th_dport; 625 } 626 xchg(nth->th_dport, nth->th_sport, n_short); 627 #undef xchg 628 } 629 if (isipv6) { 630 ip6->ip6_flow = 0; 631 ip6->ip6_vfc = IPV6_VERSION; 632 ip6->ip6_nxt = IPPROTO_TCP; 633 ip6->ip6_plen = htons((u_short)(sizeof(struct tcphdr) + tlen)); 634 tlen += sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 635 } else { 636 tlen += sizeof(struct tcpiphdr); 637 ip->ip_len = tlen; 638 ip->ip_ttl = ip_defttl; 639 } 640 m->m_len = tlen; 641 m->m_pkthdr.len = tlen; 642 m->m_pkthdr.rcvif = NULL; 643 nth->th_seq = htonl(seq); 644 nth->th_ack = htonl(ack); 645 nth->th_x2 = 0; 646 nth->th_off = sizeof(struct tcphdr) >> 2; 647 nth->th_flags = flags; 648 if (tp != NULL) 649 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 650 else 651 nth->th_win = htons((u_short)win); 652 nth->th_urp = 0; 653 if (isipv6) { 654 nth->th_sum = 0; 655 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 656 sizeof(struct ip6_hdr), 657 tlen - sizeof(struct ip6_hdr)); 658 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL, 659 (ro6 && ro6->ro_rt) ? 660 ro6->ro_rt->rt_ifp : NULL); 661 } else { 662 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 663 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 664 m->m_pkthdr.csum_flags = CSUM_TCP; 665 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 666 m->m_pkthdr.csum_thlen = sizeof(struct tcphdr); 667 } 668 #ifdef TCPDEBUG 669 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 670 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 671 #endif 672 if (isipv6) { 673 ip6_output(m, NULL, ro6, ipflags, NULL, NULL, 674 tp ? tp->t_inpcb : NULL); 675 if ((ro6 == &sro6) && (ro6->ro_rt != NULL)) { 676 RTFREE(ro6->ro_rt); 677 ro6->ro_rt = NULL; 678 } 679 } else { 680 ipflags |= IP_DEBUGROUTE; 681 ip_output(m, NULL, ro, ipflags, NULL, tp ? tp->t_inpcb : NULL); 682 if ((ro == &sro) && (ro->ro_rt != NULL)) { 683 RTFREE(ro->ro_rt); 684 ro->ro_rt = NULL; 685 } 686 } 687 } 688 689 /* 690 * Create a new TCP control block, making an 691 * empty reassembly queue and hooking it to the argument 692 * protocol control block. The `inp' parameter must have 693 * come from the zone allocator set up in tcp_init(). 694 */ 695 struct tcpcb * 696 tcp_newtcpcb(struct inpcb *inp) 697 { 698 struct inp_tp *it; 699 struct tcpcb *tp; 700 #ifdef INET6 701 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0); 702 #else 703 const boolean_t isipv6 = FALSE; 704 #endif 705 706 it = (struct inp_tp *)inp; 707 tp = &it->tcb; 708 bzero(tp, sizeof(struct tcpcb)); 709 TAILQ_INIT(&tp->t_segq); 710 tp->t_maxseg = tp->t_maxopd = isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 711 tp->t_rxtthresh = tcprexmtthresh; 712 713 /* Set up our timeouts. */ 714 tp->tt_rexmt = &it->inp_tp_rexmt; 715 tp->tt_persist = &it->inp_tp_persist; 716 tp->tt_keep = &it->inp_tp_keep; 717 tp->tt_2msl = &it->inp_tp_2msl; 718 tp->tt_delack = &it->inp_tp_delack; 719 tcp_inittimers(tp); 720 721 /* 722 * Zero out timer message. We don't create it here, 723 * since the current CPU may not be the owner of this 724 * inpcb. 725 */ 726 tp->tt_msg = &it->inp_tp_timermsg; 727 bzero(tp->tt_msg, sizeof(*tp->tt_msg)); 728 729 tp->t_keepinit = tcp_keepinit; 730 tp->t_keepidle = tcp_keepidle; 731 tp->t_keepintvl = tcp_keepintvl; 732 tp->t_keepcnt = tcp_keepcnt; 733 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt; 734 735 if (tcp_do_ncr) 736 tp->t_flags |= TF_NCR; 737 if (tcp_do_rfc1323) 738 tp->t_flags |= (TF_REQ_SCALE | TF_REQ_TSTMP); 739 740 tp->t_inpcb = inp; /* XXX */ 741 tp->t_state = TCPS_CLOSED; 742 /* 743 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 744 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 745 * reasonable initial retransmit time. 746 */ 747 tp->t_srtt = TCPTV_SRTTBASE; 748 tp->t_rttvar = 749 ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 750 tp->t_rttmin = tcp_rexmit_min; 751 tp->t_rxtcur = TCPTV_RTOBASE; 752 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 753 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 754 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 755 tp->snd_last = ticks; 756 tp->t_rcvtime = ticks; 757 /* 758 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 759 * because the socket may be bound to an IPv6 wildcard address, 760 * which may match an IPv4-mapped IPv6 address. 761 */ 762 inp->inp_ip_ttl = ip_defttl; 763 inp->inp_ppcb = tp; 764 tcp_sack_tcpcb_init(tp); 765 return (tp); /* XXX */ 766 } 767 768 /* 769 * Drop a TCP connection, reporting the specified error. 770 * If connection is synchronized, then send a RST to peer. 771 */ 772 struct tcpcb * 773 tcp_drop(struct tcpcb *tp, int error) 774 { 775 struct socket *so = tp->t_inpcb->inp_socket; 776 777 if (TCPS_HAVERCVDSYN(tp->t_state)) { 778 tp->t_state = TCPS_CLOSED; 779 tcp_output(tp); 780 tcpstat.tcps_drops++; 781 } else 782 tcpstat.tcps_conndrops++; 783 if (error == ETIMEDOUT && tp->t_softerror) 784 error = tp->t_softerror; 785 so->so_error = error; 786 return (tcp_close(tp)); 787 } 788 789 #ifdef SMP 790 791 struct netmsg_listen_detach { 792 struct netmsg_base base; 793 struct tcpcb *nm_tp; 794 }; 795 796 static void 797 tcp_listen_detach_handler(netmsg_t msg) 798 { 799 struct netmsg_listen_detach *nmsg = (struct netmsg_listen_detach *)msg; 800 struct tcpcb *tp = nmsg->nm_tp; 801 int cpu = mycpuid, nextcpu; 802 803 if (tp->t_flags & TF_LISTEN) 804 syncache_destroy(tp); 805 806 in_pcbremwildcardhash_oncpu(tp->t_inpcb, &tcbinfo[cpu]); 807 808 nextcpu = cpu + 1; 809 if (nextcpu < ncpus2) 810 lwkt_forwardmsg(netisr_portfn(nextcpu), &nmsg->base.lmsg); 811 else 812 lwkt_replymsg(&nmsg->base.lmsg, 0); 813 } 814 815 #endif 816 817 /* 818 * Close a TCP control block: 819 * discard all space held by the tcp 820 * discard internet protocol block 821 * wake up any sleepers 822 */ 823 struct tcpcb * 824 tcp_close(struct tcpcb *tp) 825 { 826 struct tseg_qent *q; 827 struct inpcb *inp = tp->t_inpcb; 828 struct socket *so = inp->inp_socket; 829 struct rtentry *rt; 830 boolean_t dosavessthresh; 831 #ifdef INET6 832 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0); 833 boolean_t isafinet6 = (INP_CHECK_SOCKAF(so, AF_INET6) != 0); 834 #else 835 const boolean_t isipv6 = FALSE; 836 #endif 837 838 #ifdef SMP 839 /* 840 * INP_WILDCARD_MP indicates that listen(2) has been called on 841 * this socket. This implies: 842 * - A wildcard inp's hash is replicated for each protocol thread. 843 * - Syncache for this inp grows independently in each protocol 844 * thread. 845 * - There is more than one cpu 846 * 847 * We have to chain a message to the rest of the protocol threads 848 * to cleanup the wildcard hash and the syncache. The cleanup 849 * in the current protocol thread is defered till the end of this 850 * function. 851 * 852 * NOTE: 853 * After cleanup the inp's hash and syncache entries, this inp will 854 * no longer be available to the rest of the protocol threads, so we 855 * are safe to whack the inp in the following code. 856 */ 857 if (inp->inp_flags & INP_WILDCARD_MP) { 858 struct netmsg_listen_detach nmsg; 859 860 KKASSERT(so->so_port == netisr_portfn(0)); 861 KKASSERT(&curthread->td_msgport == netisr_portfn(0)); 862 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]); 863 864 netmsg_init(&nmsg.base, NULL, &curthread->td_msgport, 865 MSGF_PRIORITY, tcp_listen_detach_handler); 866 nmsg.nm_tp = tp; 867 lwkt_domsg(netisr_portfn(1), &nmsg.base.lmsg, 0); 868 869 inp->inp_flags &= ~INP_WILDCARD_MP; 870 } 871 #endif 872 873 KKASSERT(tp->t_state != TCPS_TERMINATING); 874 tp->t_state = TCPS_TERMINATING; 875 876 /* 877 * Make sure that all of our timers are stopped before we 878 * delete the PCB. For listen TCP socket (tp->tt_msg == NULL), 879 * timers are never used. If timer message is never created 880 * (tp->tt_msg->tt_tcb == NULL), timers are never used too. 881 */ 882 if (tp->tt_msg != NULL && tp->tt_msg->tt_tcb != NULL) { 883 tcp_callout_stop(tp, tp->tt_rexmt); 884 tcp_callout_stop(tp, tp->tt_persist); 885 tcp_callout_stop(tp, tp->tt_keep); 886 tcp_callout_stop(tp, tp->tt_2msl); 887 tcp_callout_stop(tp, tp->tt_delack); 888 } 889 890 if (tp->t_flags & TF_ONOUTPUTQ) { 891 KKASSERT(tp->tt_cpu == mycpu->gd_cpuid); 892 TAILQ_REMOVE(&tcpcbackq[tp->tt_cpu], tp, t_outputq); 893 tp->t_flags &= ~TF_ONOUTPUTQ; 894 } 895 896 /* 897 * If we got enough samples through the srtt filter, 898 * save the rtt and rttvar in the routing entry. 899 * 'Enough' is arbitrarily defined as the 16 samples. 900 * 16 samples is enough for the srtt filter to converge 901 * to within 5% of the correct value; fewer samples and 902 * we could save a very bogus rtt. 903 * 904 * Don't update the default route's characteristics and don't 905 * update anything that the user "locked". 906 */ 907 if (tp->t_rttupdated >= 16) { 908 u_long i = 0; 909 910 if (isipv6) { 911 struct sockaddr_in6 *sin6; 912 913 if ((rt = inp->in6p_route.ro_rt) == NULL) 914 goto no_valid_rt; 915 sin6 = (struct sockaddr_in6 *)rt_key(rt); 916 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 917 goto no_valid_rt; 918 } else 919 if ((rt = inp->inp_route.ro_rt) == NULL || 920 ((struct sockaddr_in *)rt_key(rt))-> 921 sin_addr.s_addr == INADDR_ANY) 922 goto no_valid_rt; 923 924 if (!(rt->rt_rmx.rmx_locks & RTV_RTT)) { 925 i = tp->t_srtt * (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 926 if (rt->rt_rmx.rmx_rtt && i) 927 /* 928 * filter this update to half the old & half 929 * the new values, converting scale. 930 * See route.h and tcp_var.h for a 931 * description of the scaling constants. 932 */ 933 rt->rt_rmx.rmx_rtt = 934 (rt->rt_rmx.rmx_rtt + i) / 2; 935 else 936 rt->rt_rmx.rmx_rtt = i; 937 tcpstat.tcps_cachedrtt++; 938 } 939 if (!(rt->rt_rmx.rmx_locks & RTV_RTTVAR)) { 940 i = tp->t_rttvar * 941 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 942 if (rt->rt_rmx.rmx_rttvar && i) 943 rt->rt_rmx.rmx_rttvar = 944 (rt->rt_rmx.rmx_rttvar + i) / 2; 945 else 946 rt->rt_rmx.rmx_rttvar = i; 947 tcpstat.tcps_cachedrttvar++; 948 } 949 /* 950 * The old comment here said: 951 * update the pipelimit (ssthresh) if it has been updated 952 * already or if a pipesize was specified & the threshhold 953 * got below half the pipesize. I.e., wait for bad news 954 * before we start updating, then update on both good 955 * and bad news. 956 * 957 * But we want to save the ssthresh even if no pipesize is 958 * specified explicitly in the route, because such 959 * connections still have an implicit pipesize specified 960 * by the global tcp_sendspace. In the absence of a reliable 961 * way to calculate the pipesize, it will have to do. 962 */ 963 i = tp->snd_ssthresh; 964 if (rt->rt_rmx.rmx_sendpipe != 0) 965 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe/2); 966 else 967 dosavessthresh = (i < so->so_snd.ssb_hiwat/2); 968 if (dosavessthresh || 969 (!(rt->rt_rmx.rmx_locks & RTV_SSTHRESH) && (i != 0) && 970 (rt->rt_rmx.rmx_ssthresh != 0))) { 971 /* 972 * convert the limit from user data bytes to 973 * packets then to packet data bytes. 974 */ 975 i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 976 if (i < 2) 977 i = 2; 978 i *= tp->t_maxseg + 979 (isipv6 ? 980 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 981 sizeof(struct tcpiphdr)); 982 if (rt->rt_rmx.rmx_ssthresh) 983 rt->rt_rmx.rmx_ssthresh = 984 (rt->rt_rmx.rmx_ssthresh + i) / 2; 985 else 986 rt->rt_rmx.rmx_ssthresh = i; 987 tcpstat.tcps_cachedssthresh++; 988 } 989 } 990 991 no_valid_rt: 992 /* free the reassembly queue, if any */ 993 while((q = TAILQ_FIRST(&tp->t_segq)) != NULL) { 994 TAILQ_REMOVE(&tp->t_segq, q, tqe_q); 995 m_freem(q->tqe_m); 996 kfree(q, M_TSEGQ); 997 atomic_add_int(&tcp_reass_qsize, -1); 998 } 999 /* throw away SACK blocks in scoreboard*/ 1000 if (TCP_DO_SACK(tp)) 1001 tcp_sack_destroy(&tp->scb); 1002 1003 inp->inp_ppcb = NULL; 1004 soisdisconnected(so); 1005 /* note: pcb detached later on */ 1006 1007 tcp_destroy_timermsg(tp); 1008 1009 if (tp->t_flags & TF_LISTEN) 1010 syncache_destroy(tp); 1011 1012 so_async_rcvd_drop(so); 1013 1014 /* 1015 * NOTE: 1016 * pcbdetach removes any wildcard hash entry on the current CPU. 1017 */ 1018 #ifdef INET6 1019 if (isafinet6) 1020 in6_pcbdetach(inp); 1021 else 1022 #endif 1023 in_pcbdetach(inp); 1024 1025 tcpstat.tcps_closed++; 1026 return (NULL); 1027 } 1028 1029 static __inline void 1030 tcp_drain_oncpu(struct inpcbhead *head) 1031 { 1032 struct inpcb *marker; 1033 struct inpcb *inpb; 1034 struct tcpcb *tcpb; 1035 struct tseg_qent *te; 1036 1037 /* 1038 * Allows us to block while running the list 1039 */ 1040 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO); 1041 marker->inp_flags |= INP_PLACEMARKER; 1042 LIST_INSERT_HEAD(head, marker, inp_list); 1043 1044 while ((inpb = LIST_NEXT(marker, inp_list)) != NULL) { 1045 if ((inpb->inp_flags & INP_PLACEMARKER) == 0 && 1046 (tcpb = intotcpcb(inpb)) != NULL && 1047 (te = TAILQ_FIRST(&tcpb->t_segq)) != NULL) { 1048 TAILQ_REMOVE(&tcpb->t_segq, te, tqe_q); 1049 if (te->tqe_th->th_flags & TH_FIN) 1050 tcpb->t_flags &= ~TF_QUEDFIN; 1051 m_freem(te->tqe_m); 1052 kfree(te, M_TSEGQ); 1053 atomic_add_int(&tcp_reass_qsize, -1); 1054 /* retry */ 1055 } else { 1056 LIST_REMOVE(marker, inp_list); 1057 LIST_INSERT_AFTER(inpb, marker, inp_list); 1058 } 1059 } 1060 LIST_REMOVE(marker, inp_list); 1061 kfree(marker, M_TEMP); 1062 } 1063 1064 #ifdef SMP 1065 struct netmsg_tcp_drain { 1066 struct netmsg_base base; 1067 struct inpcbhead *nm_head; 1068 }; 1069 1070 static void 1071 tcp_drain_handler(netmsg_t msg) 1072 { 1073 struct netmsg_tcp_drain *nm = (void *)msg; 1074 1075 tcp_drain_oncpu(nm->nm_head); 1076 lwkt_replymsg(&nm->base.lmsg, 0); 1077 } 1078 #endif 1079 1080 void 1081 tcp_drain(void) 1082 { 1083 #ifdef SMP 1084 int cpu; 1085 #endif 1086 1087 if (!do_tcpdrain) 1088 return; 1089 1090 /* 1091 * Walk the tcpbs, if existing, and flush the reassembly queue, 1092 * if there is one... 1093 * XXX: The "Net/3" implementation doesn't imply that the TCP 1094 * reassembly queue should be flushed, but in a situation 1095 * where we're really low on mbufs, this is potentially 1096 * useful. 1097 */ 1098 #ifdef SMP 1099 for (cpu = 0; cpu < ncpus2; cpu++) { 1100 struct netmsg_tcp_drain *nm; 1101 1102 if (cpu == mycpu->gd_cpuid) { 1103 tcp_drain_oncpu(&tcbinfo[cpu].pcblisthead); 1104 } else { 1105 nm = kmalloc(sizeof(struct netmsg_tcp_drain), 1106 M_LWKTMSG, M_NOWAIT); 1107 if (nm == NULL) 1108 continue; 1109 netmsg_init(&nm->base, NULL, &netisr_afree_rport, 1110 0, tcp_drain_handler); 1111 nm->nm_head = &tcbinfo[cpu].pcblisthead; 1112 lwkt_sendmsg(netisr_portfn(cpu), &nm->base.lmsg); 1113 } 1114 } 1115 #else 1116 tcp_drain_oncpu(&tcbinfo[0].pcblisthead); 1117 #endif 1118 } 1119 1120 /* 1121 * Notify a tcp user of an asynchronous error; 1122 * store error as soft error, but wake up user 1123 * (for now, won't do anything until can select for soft error). 1124 * 1125 * Do not wake up user since there currently is no mechanism for 1126 * reporting soft errors (yet - a kqueue filter may be added). 1127 */ 1128 static void 1129 tcp_notify(struct inpcb *inp, int error) 1130 { 1131 struct tcpcb *tp = intotcpcb(inp); 1132 1133 /* 1134 * Ignore some errors if we are hooked up. 1135 * If connection hasn't completed, has retransmitted several times, 1136 * and receives a second error, give up now. This is better 1137 * than waiting a long time to establish a connection that 1138 * can never complete. 1139 */ 1140 if (tp->t_state == TCPS_ESTABLISHED && 1141 (error == EHOSTUNREACH || error == ENETUNREACH || 1142 error == EHOSTDOWN)) { 1143 return; 1144 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 1145 tp->t_softerror) 1146 tcp_drop(tp, error); 1147 else 1148 tp->t_softerror = error; 1149 #if 0 1150 wakeup(&so->so_timeo); 1151 sorwakeup(so); 1152 sowwakeup(so); 1153 #endif 1154 } 1155 1156 static int 1157 tcp_pcblist(SYSCTL_HANDLER_ARGS) 1158 { 1159 int error, i, n; 1160 struct inpcb *marker; 1161 struct inpcb *inp; 1162 globaldata_t gd; 1163 int origcpu, ccpu; 1164 1165 error = 0; 1166 n = 0; 1167 1168 /* 1169 * The process of preparing the TCB list is too time-consuming and 1170 * resource-intensive to repeat twice on every request. 1171 */ 1172 if (req->oldptr == NULL) { 1173 for (ccpu = 0; ccpu < ncpus; ++ccpu) { 1174 gd = globaldata_find(ccpu); 1175 n += tcbinfo[gd->gd_cpuid].ipi_count; 1176 } 1177 req->oldidx = (n + n/8 + 10) * sizeof(struct xtcpcb); 1178 return (0); 1179 } 1180 1181 if (req->newptr != NULL) 1182 return (EPERM); 1183 1184 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO); 1185 marker->inp_flags |= INP_PLACEMARKER; 1186 1187 /* 1188 * OK, now we're committed to doing something. Run the inpcb list 1189 * for each cpu in the system and construct the output. Use a 1190 * list placemarker to deal with list changes occuring during 1191 * copyout blockages (but otherwise depend on being on the correct 1192 * cpu to avoid races). 1193 */ 1194 origcpu = mycpu->gd_cpuid; 1195 for (ccpu = 1; ccpu <= ncpus && error == 0; ++ccpu) { 1196 globaldata_t rgd; 1197 caddr_t inp_ppcb; 1198 struct xtcpcb xt; 1199 int cpu_id; 1200 1201 cpu_id = (origcpu + ccpu) % ncpus; 1202 if ((smp_active_mask & CPUMASK(cpu_id)) == 0) 1203 continue; 1204 rgd = globaldata_find(cpu_id); 1205 lwkt_setcpu_self(rgd); 1206 1207 n = tcbinfo[cpu_id].ipi_count; 1208 1209 LIST_INSERT_HEAD(&tcbinfo[cpu_id].pcblisthead, marker, inp_list); 1210 i = 0; 1211 while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) { 1212 /* 1213 * process a snapshot of pcbs, ignoring placemarkers 1214 * and using our own to allow SYSCTL_OUT to block. 1215 */ 1216 LIST_REMOVE(marker, inp_list); 1217 LIST_INSERT_AFTER(inp, marker, inp_list); 1218 1219 if (inp->inp_flags & INP_PLACEMARKER) 1220 continue; 1221 if (prison_xinpcb(req->td, inp)) 1222 continue; 1223 1224 xt.xt_len = sizeof xt; 1225 bcopy(inp, &xt.xt_inp, sizeof *inp); 1226 inp_ppcb = inp->inp_ppcb; 1227 if (inp_ppcb != NULL) 1228 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1229 else 1230 bzero(&xt.xt_tp, sizeof xt.xt_tp); 1231 if (inp->inp_socket) 1232 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1233 if ((error = SYSCTL_OUT(req, &xt, sizeof xt)) != 0) 1234 break; 1235 ++i; 1236 } 1237 LIST_REMOVE(marker, inp_list); 1238 if (error == 0 && i < n) { 1239 bzero(&xt, sizeof xt); 1240 xt.xt_len = sizeof xt; 1241 while (i < n) { 1242 error = SYSCTL_OUT(req, &xt, sizeof xt); 1243 if (error) 1244 break; 1245 ++i; 1246 } 1247 } 1248 } 1249 1250 /* 1251 * Make sure we are on the same cpu we were on originally, since 1252 * higher level callers expect this. Also don't pollute caches with 1253 * migrated userland data by (eventually) returning to userland 1254 * on a different cpu. 1255 */ 1256 lwkt_setcpu_self(globaldata_find(origcpu)); 1257 kfree(marker, M_TEMP); 1258 return (error); 1259 } 1260 1261 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1262 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1263 1264 static int 1265 tcp_getcred(SYSCTL_HANDLER_ARGS) 1266 { 1267 struct sockaddr_in addrs[2]; 1268 struct inpcb *inp; 1269 int cpu; 1270 int error; 1271 1272 error = priv_check(req->td, PRIV_ROOT); 1273 if (error != 0) 1274 return (error); 1275 error = SYSCTL_IN(req, addrs, sizeof addrs); 1276 if (error != 0) 1277 return (error); 1278 crit_enter(); 1279 cpu = tcp_addrcpu(addrs[1].sin_addr.s_addr, addrs[1].sin_port, 1280 addrs[0].sin_addr.s_addr, addrs[0].sin_port); 1281 inp = in_pcblookup_hash(&tcbinfo[cpu], addrs[1].sin_addr, 1282 addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1283 if (inp == NULL || inp->inp_socket == NULL) { 1284 error = ENOENT; 1285 goto out; 1286 } 1287 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred)); 1288 out: 1289 crit_exit(); 1290 return (error); 1291 } 1292 1293 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW), 1294 0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection"); 1295 1296 #ifdef INET6 1297 static int 1298 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1299 { 1300 struct sockaddr_in6 addrs[2]; 1301 struct inpcb *inp; 1302 int error; 1303 boolean_t mapped = FALSE; 1304 1305 error = priv_check(req->td, PRIV_ROOT); 1306 if (error != 0) 1307 return (error); 1308 error = SYSCTL_IN(req, addrs, sizeof addrs); 1309 if (error != 0) 1310 return (error); 1311 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1312 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1313 mapped = TRUE; 1314 else 1315 return (EINVAL); 1316 } 1317 crit_enter(); 1318 if (mapped) { 1319 inp = in_pcblookup_hash(&tcbinfo[0], 1320 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1321 addrs[1].sin6_port, 1322 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1323 addrs[0].sin6_port, 1324 0, NULL); 1325 } else { 1326 inp = in6_pcblookup_hash(&tcbinfo[0], 1327 &addrs[1].sin6_addr, addrs[1].sin6_port, 1328 &addrs[0].sin6_addr, addrs[0].sin6_port, 1329 0, NULL); 1330 } 1331 if (inp == NULL || inp->inp_socket == NULL) { 1332 error = ENOENT; 1333 goto out; 1334 } 1335 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred)); 1336 out: 1337 crit_exit(); 1338 return (error); 1339 } 1340 1341 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW), 1342 0, 0, 1343 tcp6_getcred, "S,ucred", "Get the ucred of a TCP6 connection"); 1344 #endif 1345 1346 struct netmsg_tcp_notify { 1347 struct netmsg_base base; 1348 void (*nm_notify)(struct inpcb *, int); 1349 struct in_addr nm_faddr; 1350 int nm_arg; 1351 }; 1352 1353 static void 1354 tcp_notifyall_oncpu(netmsg_t msg) 1355 { 1356 struct netmsg_tcp_notify *nm = (struct netmsg_tcp_notify *)msg; 1357 int nextcpu; 1358 1359 in_pcbnotifyall(&tcbinfo[mycpuid].pcblisthead, nm->nm_faddr, 1360 nm->nm_arg, nm->nm_notify); 1361 1362 nextcpu = mycpuid + 1; 1363 if (nextcpu < ncpus2) 1364 lwkt_forwardmsg(netisr_portfn(nextcpu), &nm->base.lmsg); 1365 else 1366 lwkt_replymsg(&nm->base.lmsg, 0); 1367 } 1368 1369 void 1370 tcp_ctlinput(netmsg_t msg) 1371 { 1372 int cmd = msg->ctlinput.nm_cmd; 1373 struct sockaddr *sa = msg->ctlinput.nm_arg; 1374 struct ip *ip = msg->ctlinput.nm_extra; 1375 struct tcphdr *th; 1376 struct in_addr faddr; 1377 struct inpcb *inp; 1378 struct tcpcb *tp; 1379 void (*notify)(struct inpcb *, int) = tcp_notify; 1380 tcp_seq icmpseq; 1381 int arg, cpu; 1382 1383 if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) { 1384 goto done; 1385 } 1386 1387 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1388 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1389 goto done; 1390 1391 arg = inetctlerrmap[cmd]; 1392 if (cmd == PRC_QUENCH) { 1393 notify = tcp_quench; 1394 } else if (icmp_may_rst && 1395 (cmd == PRC_UNREACH_ADMIN_PROHIB || 1396 cmd == PRC_UNREACH_PORT || 1397 cmd == PRC_TIMXCEED_INTRANS) && 1398 ip != NULL) { 1399 notify = tcp_drop_syn_sent; 1400 } else if (cmd == PRC_MSGSIZE) { 1401 struct icmp *icmp = (struct icmp *) 1402 ((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 1403 1404 arg = ntohs(icmp->icmp_nextmtu); 1405 notify = tcp_mtudisc; 1406 } else if (PRC_IS_REDIRECT(cmd)) { 1407 ip = NULL; 1408 notify = in_rtchange; 1409 } else if (cmd == PRC_HOSTDEAD) { 1410 ip = NULL; 1411 } 1412 1413 if (ip != NULL) { 1414 crit_enter(); 1415 th = (struct tcphdr *)((caddr_t)ip + 1416 (IP_VHL_HL(ip->ip_vhl) << 2)); 1417 cpu = tcp_addrcpu(faddr.s_addr, th->th_dport, 1418 ip->ip_src.s_addr, th->th_sport); 1419 inp = in_pcblookup_hash(&tcbinfo[cpu], faddr, th->th_dport, 1420 ip->ip_src, th->th_sport, 0, NULL); 1421 if ((inp != NULL) && (inp->inp_socket != NULL)) { 1422 icmpseq = htonl(th->th_seq); 1423 tp = intotcpcb(inp); 1424 if (SEQ_GEQ(icmpseq, tp->snd_una) && 1425 SEQ_LT(icmpseq, tp->snd_max)) 1426 (*notify)(inp, arg); 1427 } else { 1428 struct in_conninfo inc; 1429 1430 inc.inc_fport = th->th_dport; 1431 inc.inc_lport = th->th_sport; 1432 inc.inc_faddr = faddr; 1433 inc.inc_laddr = ip->ip_src; 1434 #ifdef INET6 1435 inc.inc_isipv6 = 0; 1436 #endif 1437 syncache_unreach(&inc, th); 1438 } 1439 crit_exit(); 1440 } else { 1441 struct netmsg_tcp_notify *nm; 1442 1443 KKASSERT(&curthread->td_msgport == netisr_portfn(0)); 1444 nm = kmalloc(sizeof(*nm), M_LWKTMSG, M_INTWAIT); 1445 netmsg_init(&nm->base, NULL, &netisr_afree_rport, 1446 0, tcp_notifyall_oncpu); 1447 nm->nm_faddr = faddr; 1448 nm->nm_arg = arg; 1449 nm->nm_notify = notify; 1450 1451 lwkt_sendmsg(netisr_portfn(0), &nm->base.lmsg); 1452 } 1453 done: 1454 lwkt_replymsg(&msg->lmsg, 0); 1455 } 1456 1457 #ifdef INET6 1458 1459 void 1460 tcp6_ctlinput(netmsg_t msg) 1461 { 1462 int cmd = msg->ctlinput.nm_cmd; 1463 struct sockaddr *sa = msg->ctlinput.nm_arg; 1464 void *d = msg->ctlinput.nm_extra; 1465 struct tcphdr th; 1466 void (*notify) (struct inpcb *, int) = tcp_notify; 1467 struct ip6_hdr *ip6; 1468 struct mbuf *m; 1469 struct ip6ctlparam *ip6cp = NULL; 1470 const struct sockaddr_in6 *sa6_src = NULL; 1471 int off; 1472 struct tcp_portonly { 1473 u_int16_t th_sport; 1474 u_int16_t th_dport; 1475 } *thp; 1476 int arg; 1477 1478 if (sa->sa_family != AF_INET6 || 1479 sa->sa_len != sizeof(struct sockaddr_in6)) { 1480 goto out; 1481 } 1482 1483 arg = 0; 1484 if (cmd == PRC_QUENCH) 1485 notify = tcp_quench; 1486 else if (cmd == PRC_MSGSIZE) { 1487 struct ip6ctlparam *ip6cp = d; 1488 struct icmp6_hdr *icmp6 = ip6cp->ip6c_icmp6; 1489 1490 arg = ntohl(icmp6->icmp6_mtu); 1491 notify = tcp_mtudisc; 1492 } else if (!PRC_IS_REDIRECT(cmd) && 1493 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) { 1494 goto out; 1495 } 1496 1497 /* if the parameter is from icmp6, decode it. */ 1498 if (d != NULL) { 1499 ip6cp = (struct ip6ctlparam *)d; 1500 m = ip6cp->ip6c_m; 1501 ip6 = ip6cp->ip6c_ip6; 1502 off = ip6cp->ip6c_off; 1503 sa6_src = ip6cp->ip6c_src; 1504 } else { 1505 m = NULL; 1506 ip6 = NULL; 1507 off = 0; /* fool gcc */ 1508 sa6_src = &sa6_any; 1509 } 1510 1511 if (ip6 != NULL) { 1512 struct in_conninfo inc; 1513 /* 1514 * XXX: We assume that when IPV6 is non NULL, 1515 * M and OFF are valid. 1516 */ 1517 1518 /* check if we can safely examine src and dst ports */ 1519 if (m->m_pkthdr.len < off + sizeof *thp) 1520 goto out; 1521 1522 bzero(&th, sizeof th); 1523 m_copydata(m, off, sizeof *thp, (caddr_t)&th); 1524 1525 in6_pcbnotify(&tcbinfo[0].pcblisthead, sa, th.th_dport, 1526 (struct sockaddr *)ip6cp->ip6c_src, 1527 th.th_sport, cmd, arg, notify); 1528 1529 inc.inc_fport = th.th_dport; 1530 inc.inc_lport = th.th_sport; 1531 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1532 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1533 inc.inc_isipv6 = 1; 1534 syncache_unreach(&inc, &th); 1535 } else { 1536 in6_pcbnotify(&tcbinfo[0].pcblisthead, sa, 0, 1537 (const struct sockaddr *)sa6_src, 0, cmd, arg, notify); 1538 } 1539 out: 1540 lwkt_replymsg(&msg->ctlinput.base.lmsg, 0); 1541 } 1542 1543 #endif 1544 1545 /* 1546 * Following is where TCP initial sequence number generation occurs. 1547 * 1548 * There are two places where we must use initial sequence numbers: 1549 * 1. In SYN-ACK packets. 1550 * 2. In SYN packets. 1551 * 1552 * All ISNs for SYN-ACK packets are generated by the syncache. See 1553 * tcp_syncache.c for details. 1554 * 1555 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1556 * depends on this property. In addition, these ISNs should be 1557 * unguessable so as to prevent connection hijacking. To satisfy 1558 * the requirements of this situation, the algorithm outlined in 1559 * RFC 1948 is used to generate sequence numbers. 1560 * 1561 * Implementation details: 1562 * 1563 * Time is based off the system timer, and is corrected so that it 1564 * increases by one megabyte per second. This allows for proper 1565 * recycling on high speed LANs while still leaving over an hour 1566 * before rollover. 1567 * 1568 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1569 * between seeding of isn_secret. This is normally set to zero, 1570 * as reseeding should not be necessary. 1571 * 1572 */ 1573 1574 #define ISN_BYTES_PER_SECOND 1048576 1575 1576 u_char isn_secret[32]; 1577 int isn_last_reseed; 1578 MD5_CTX isn_ctx; 1579 1580 tcp_seq 1581 tcp_new_isn(struct tcpcb *tp) 1582 { 1583 u_int32_t md5_buffer[4]; 1584 tcp_seq new_isn; 1585 1586 /* Seed if this is the first use, reseed if requested. */ 1587 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1588 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1589 < (u_int)ticks))) { 1590 read_random_unlimited(&isn_secret, sizeof isn_secret); 1591 isn_last_reseed = ticks; 1592 } 1593 1594 /* Compute the md5 hash and return the ISN. */ 1595 MD5Init(&isn_ctx); 1596 MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_fport, sizeof(u_short)); 1597 MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_lport, sizeof(u_short)); 1598 #ifdef INET6 1599 if (tp->t_inpcb->inp_vflag & INP_IPV6) { 1600 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1601 sizeof(struct in6_addr)); 1602 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1603 sizeof(struct in6_addr)); 1604 } else 1605 #endif 1606 { 1607 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1608 sizeof(struct in_addr)); 1609 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1610 sizeof(struct in_addr)); 1611 } 1612 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1613 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1614 new_isn = (tcp_seq) md5_buffer[0]; 1615 new_isn += ticks * (ISN_BYTES_PER_SECOND / hz); 1616 return (new_isn); 1617 } 1618 1619 /* 1620 * When a source quench is received, close congestion window 1621 * to one segment. We will gradually open it again as we proceed. 1622 */ 1623 void 1624 tcp_quench(struct inpcb *inp, int error) 1625 { 1626 struct tcpcb *tp = intotcpcb(inp); 1627 1628 if (tp != NULL) { 1629 tp->snd_cwnd = tp->t_maxseg; 1630 tp->snd_wacked = 0; 1631 } 1632 } 1633 1634 /* 1635 * When a specific ICMP unreachable message is received and the 1636 * connection state is SYN-SENT, drop the connection. This behavior 1637 * is controlled by the icmp_may_rst sysctl. 1638 */ 1639 void 1640 tcp_drop_syn_sent(struct inpcb *inp, int error) 1641 { 1642 struct tcpcb *tp = intotcpcb(inp); 1643 1644 if ((tp != NULL) && (tp->t_state == TCPS_SYN_SENT)) 1645 tcp_drop(tp, error); 1646 } 1647 1648 /* 1649 * When a `need fragmentation' ICMP is received, update our idea of the MSS 1650 * based on the new value in the route. Also nudge TCP to send something, 1651 * since we know the packet we just sent was dropped. 1652 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1653 */ 1654 void 1655 tcp_mtudisc(struct inpcb *inp, int mtu) 1656 { 1657 struct tcpcb *tp = intotcpcb(inp); 1658 struct rtentry *rt; 1659 struct socket *so = inp->inp_socket; 1660 int maxopd, mss; 1661 #ifdef INET6 1662 boolean_t isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0); 1663 #else 1664 const boolean_t isipv6 = FALSE; 1665 #endif 1666 1667 if (tp == NULL) 1668 return; 1669 1670 /* 1671 * If no MTU is provided in the ICMP message, use the 1672 * next lower likely value, as specified in RFC 1191. 1673 */ 1674 if (mtu == 0) { 1675 int oldmtu; 1676 1677 oldmtu = tp->t_maxopd + 1678 (isipv6 ? 1679 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1680 sizeof(struct tcpiphdr)); 1681 mtu = ip_next_mtu(oldmtu, 0); 1682 } 1683 1684 if (isipv6) 1685 rt = tcp_rtlookup6(&inp->inp_inc); 1686 else 1687 rt = tcp_rtlookup(&inp->inp_inc); 1688 if (rt != NULL) { 1689 if (rt->rt_rmx.rmx_mtu != 0 && rt->rt_rmx.rmx_mtu < mtu) 1690 mtu = rt->rt_rmx.rmx_mtu; 1691 1692 maxopd = mtu - 1693 (isipv6 ? 1694 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1695 sizeof(struct tcpiphdr)); 1696 1697 /* 1698 * XXX - The following conditional probably violates the TCP 1699 * spec. The problem is that, since we don't know the 1700 * other end's MSS, we are supposed to use a conservative 1701 * default. But, if we do that, then MTU discovery will 1702 * never actually take place, because the conservative 1703 * default is much less than the MTUs typically seen 1704 * on the Internet today. For the moment, we'll sweep 1705 * this under the carpet. 1706 * 1707 * The conservative default might not actually be a problem 1708 * if the only case this occurs is when sending an initial 1709 * SYN with options and data to a host we've never talked 1710 * to before. Then, they will reply with an MSS value which 1711 * will get recorded and the new parameters should get 1712 * recomputed. For Further Study. 1713 */ 1714 if (rt->rt_rmx.rmx_mssopt && rt->rt_rmx.rmx_mssopt < maxopd) 1715 maxopd = rt->rt_rmx.rmx_mssopt; 1716 } else 1717 maxopd = mtu - 1718 (isipv6 ? 1719 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1720 sizeof(struct tcpiphdr)); 1721 1722 if (tp->t_maxopd <= maxopd) 1723 return; 1724 tp->t_maxopd = maxopd; 1725 1726 mss = maxopd; 1727 if ((tp->t_flags & (TF_REQ_TSTMP | TF_RCVD_TSTMP | TF_NOOPT)) == 1728 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 1729 mss -= TCPOLEN_TSTAMP_APPA; 1730 1731 /* round down to multiple of MCLBYTES */ 1732 #if (MCLBYTES & (MCLBYTES - 1)) == 0 /* test if MCLBYTES power of 2 */ 1733 if (mss > MCLBYTES) 1734 mss &= ~(MCLBYTES - 1); 1735 #else 1736 if (mss > MCLBYTES) 1737 mss = (mss / MCLBYTES) * MCLBYTES; 1738 #endif 1739 1740 if (so->so_snd.ssb_hiwat < mss) 1741 mss = so->so_snd.ssb_hiwat; 1742 1743 tp->t_maxseg = mss; 1744 tp->t_rtttime = 0; 1745 tp->snd_nxt = tp->snd_una; 1746 tcp_output(tp); 1747 tcpstat.tcps_mturesent++; 1748 } 1749 1750 /* 1751 * Look-up the routing entry to the peer of this inpcb. If no route 1752 * is found and it cannot be allocated the return NULL. This routine 1753 * is called by TCP routines that access the rmx structure and by tcp_mss 1754 * to get the interface MTU. 1755 */ 1756 struct rtentry * 1757 tcp_rtlookup(struct in_conninfo *inc) 1758 { 1759 struct route *ro = &inc->inc_route; 1760 1761 if (ro->ro_rt == NULL || !(ro->ro_rt->rt_flags & RTF_UP)) { 1762 /* No route yet, so try to acquire one */ 1763 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1764 /* 1765 * unused portions of the structure MUST be zero'd 1766 * out because rtalloc() treats it as opaque data 1767 */ 1768 bzero(&ro->ro_dst, sizeof(struct sockaddr_in)); 1769 ro->ro_dst.sa_family = AF_INET; 1770 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 1771 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1772 inc->inc_faddr; 1773 rtalloc(ro); 1774 } 1775 } 1776 return (ro->ro_rt); 1777 } 1778 1779 #ifdef INET6 1780 struct rtentry * 1781 tcp_rtlookup6(struct in_conninfo *inc) 1782 { 1783 struct route_in6 *ro6 = &inc->inc6_route; 1784 1785 if (ro6->ro_rt == NULL || !(ro6->ro_rt->rt_flags & RTF_UP)) { 1786 /* No route yet, so try to acquire one */ 1787 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1788 /* 1789 * unused portions of the structure MUST be zero'd 1790 * out because rtalloc() treats it as opaque data 1791 */ 1792 bzero(&ro6->ro_dst, sizeof(struct sockaddr_in6)); 1793 ro6->ro_dst.sin6_family = AF_INET6; 1794 ro6->ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1795 ro6->ro_dst.sin6_addr = inc->inc6_faddr; 1796 rtalloc((struct route *)ro6); 1797 } 1798 } 1799 return (ro6->ro_rt); 1800 } 1801 #endif 1802 1803 #ifdef IPSEC 1804 /* compute ESP/AH header size for TCP, including outer IP header. */ 1805 size_t 1806 ipsec_hdrsiz_tcp(struct tcpcb *tp) 1807 { 1808 struct inpcb *inp; 1809 struct mbuf *m; 1810 size_t hdrsiz; 1811 struct ip *ip; 1812 struct tcphdr *th; 1813 1814 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1815 return (0); 1816 MGETHDR(m, MB_DONTWAIT, MT_DATA); 1817 if (!m) 1818 return (0); 1819 1820 #ifdef INET6 1821 if (inp->inp_vflag & INP_IPV6) { 1822 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 1823 1824 th = (struct tcphdr *)(ip6 + 1); 1825 m->m_pkthdr.len = m->m_len = 1826 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1827 tcp_fillheaders(tp, ip6, th, FALSE); 1828 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1829 } else 1830 #endif 1831 { 1832 ip = mtod(m, struct ip *); 1833 th = (struct tcphdr *)(ip + 1); 1834 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1835 tcp_fillheaders(tp, ip, th, FALSE); 1836 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1837 } 1838 1839 m_free(m); 1840 return (hdrsiz); 1841 } 1842 #endif 1843 1844 /* 1845 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1846 * 1847 * This code attempts to calculate the bandwidth-delay product as a 1848 * means of determining the optimal window size to maximize bandwidth, 1849 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1850 * routers. This code also does a fairly good job keeping RTTs in check 1851 * across slow links like modems. We implement an algorithm which is very 1852 * similar (but not meant to be) TCP/Vegas. The code operates on the 1853 * transmitter side of a TCP connection and so only effects the transmit 1854 * side of the connection. 1855 * 1856 * BACKGROUND: TCP makes no provision for the management of buffer space 1857 * at the end points or at the intermediate routers and switches. A TCP 1858 * stream, whether using NewReno or not, will eventually buffer as 1859 * many packets as it is able and the only reason this typically works is 1860 * due to the fairly small default buffers made available for a connection 1861 * (typicaly 16K or 32K). As machines use larger windows and/or window 1862 * scaling it is now fairly easy for even a single TCP connection to blow-out 1863 * all available buffer space not only on the local interface, but on 1864 * intermediate routers and switches as well. NewReno makes a misguided 1865 * attempt to 'solve' this problem by waiting for an actual failure to occur, 1866 * then backing off, then steadily increasing the window again until another 1867 * failure occurs, ad-infinitum. This results in terrible oscillation that 1868 * is only made worse as network loads increase and the idea of intentionally 1869 * blowing out network buffers is, frankly, a terrible way to manage network 1870 * resources. 1871 * 1872 * It is far better to limit the transmit window prior to the failure 1873 * condition being achieved. There are two general ways to do this: First 1874 * you can 'scan' through different transmit window sizes and locate the 1875 * point where the RTT stops increasing, indicating that you have filled the 1876 * pipe, then scan backwards until you note that RTT stops decreasing, then 1877 * repeat ad-infinitum. This method works in principle but has severe 1878 * implementation issues due to RTT variances, timer granularity, and 1879 * instability in the algorithm which can lead to many false positives and 1880 * create oscillations as well as interact badly with other TCP streams 1881 * implementing the same algorithm. 1882 * 1883 * The second method is to limit the window to the bandwidth delay product 1884 * of the link. This is the method we implement. RTT variances and our 1885 * own manipulation of the congestion window, bwnd, can potentially 1886 * destabilize the algorithm. For this reason we have to stabilize the 1887 * elements used to calculate the window. We do this by using the minimum 1888 * observed RTT, the long term average of the observed bandwidth, and 1889 * by adding two segments worth of slop. It isn't perfect but it is able 1890 * to react to changing conditions and gives us a very stable basis on 1891 * which to extend the algorithm. 1892 */ 1893 void 1894 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 1895 { 1896 u_long bw; 1897 u_long bwnd; 1898 int save_ticks; 1899 int delta_ticks; 1900 1901 /* 1902 * If inflight_enable is disabled in the middle of a tcp connection, 1903 * make sure snd_bwnd is effectively disabled. 1904 */ 1905 if (!tcp_inflight_enable) { 1906 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1907 tp->snd_bandwidth = 0; 1908 return; 1909 } 1910 1911 /* 1912 * Validate the delta time. If a connection is new or has been idle 1913 * a long time we have to reset the bandwidth calculator. 1914 */ 1915 save_ticks = ticks; 1916 delta_ticks = save_ticks - tp->t_bw_rtttime; 1917 if (tp->t_bw_rtttime == 0 || delta_ticks < 0 || delta_ticks > hz * 10) { 1918 tp->t_bw_rtttime = ticks; 1919 tp->t_bw_rtseq = ack_seq; 1920 if (tp->snd_bandwidth == 0) 1921 tp->snd_bandwidth = tcp_inflight_min; 1922 return; 1923 } 1924 if (delta_ticks == 0) 1925 return; 1926 1927 /* 1928 * Sanity check, plus ignore pure window update acks. 1929 */ 1930 if ((int)(ack_seq - tp->t_bw_rtseq) <= 0) 1931 return; 1932 1933 /* 1934 * Figure out the bandwidth. Due to the tick granularity this 1935 * is a very rough number and it MUST be averaged over a fairly 1936 * long period of time. XXX we need to take into account a link 1937 * that is not using all available bandwidth, but for now our 1938 * slop will ramp us up if this case occurs and the bandwidth later 1939 * increases. 1940 */ 1941 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / delta_ticks; 1942 tp->t_bw_rtttime = save_ticks; 1943 tp->t_bw_rtseq = ack_seq; 1944 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 1945 1946 tp->snd_bandwidth = bw; 1947 1948 /* 1949 * Calculate the semi-static bandwidth delay product, plus two maximal 1950 * segments. The additional slop puts us squarely in the sweet 1951 * spot and also handles the bandwidth run-up case. Without the 1952 * slop we could be locking ourselves into a lower bandwidth. 1953 * 1954 * Situations Handled: 1955 * (1) Prevents over-queueing of packets on LANs, especially on 1956 * high speed LANs, allowing larger TCP buffers to be 1957 * specified, and also does a good job preventing 1958 * over-queueing of packets over choke points like modems 1959 * (at least for the transmit side). 1960 * 1961 * (2) Is able to handle changing network loads (bandwidth 1962 * drops so bwnd drops, bandwidth increases so bwnd 1963 * increases). 1964 * 1965 * (3) Theoretically should stabilize in the face of multiple 1966 * connections implementing the same algorithm (this may need 1967 * a little work). 1968 * 1969 * (4) Stability value (defaults to 20 = 2 maximal packets) can 1970 * be adjusted with a sysctl but typically only needs to be on 1971 * very slow connections. A value no smaller then 5 should 1972 * be used, but only reduce this default if you have no other 1973 * choice. 1974 */ 1975 1976 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 1977 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + 1978 tcp_inflight_stab * (int)tp->t_maxseg / 10; 1979 #undef USERTT 1980 1981 if (tcp_inflight_debug > 0) { 1982 static int ltime; 1983 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 1984 ltime = ticks; 1985 kprintf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 1986 tp, bw, tp->t_rttbest, tp->t_srtt, bwnd); 1987 } 1988 } 1989 if ((long)bwnd < tcp_inflight_min) 1990 bwnd = tcp_inflight_min; 1991 if (bwnd > tcp_inflight_max) 1992 bwnd = tcp_inflight_max; 1993 if ((long)bwnd < tp->t_maxseg * 2) 1994 bwnd = tp->t_maxseg * 2; 1995 tp->snd_bwnd = bwnd; 1996 } 1997 1998 static void 1999 tcp_rmx_iwsegs(struct tcpcb *tp, u_long *maxsegs, u_long *capsegs) 2000 { 2001 struct rtentry *rt; 2002 struct inpcb *inp = tp->t_inpcb; 2003 #ifdef INET6 2004 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE); 2005 #else 2006 const boolean_t isipv6 = FALSE; 2007 #endif 2008 2009 /* XXX */ 2010 if (tcp_iw_maxsegs < TCP_IW_MAXSEGS_DFLT) 2011 tcp_iw_maxsegs = TCP_IW_MAXSEGS_DFLT; 2012 if (tcp_iw_capsegs < TCP_IW_CAPSEGS_DFLT) 2013 tcp_iw_capsegs = TCP_IW_CAPSEGS_DFLT; 2014 2015 if (isipv6) 2016 rt = tcp_rtlookup6(&inp->inp_inc); 2017 else 2018 rt = tcp_rtlookup(&inp->inp_inc); 2019 if (rt == NULL || 2020 rt->rt_rmx.rmx_iwmaxsegs < TCP_IW_MAXSEGS_DFLT || 2021 rt->rt_rmx.rmx_iwcapsegs < TCP_IW_CAPSEGS_DFLT) { 2022 *maxsegs = tcp_iw_maxsegs; 2023 *capsegs = tcp_iw_capsegs; 2024 return; 2025 } 2026 *maxsegs = rt->rt_rmx.rmx_iwmaxsegs; 2027 *capsegs = rt->rt_rmx.rmx_iwcapsegs; 2028 } 2029 2030 u_long 2031 tcp_initial_window(struct tcpcb *tp) 2032 { 2033 if (tcp_do_rfc3390) { 2034 /* 2035 * RFC3390: 2036 * "If the SYN or SYN/ACK is lost, the initial window 2037 * used by a sender after a correctly transmitted SYN 2038 * MUST be one segment consisting of MSS bytes." 2039 * 2040 * However, we do something a little bit more aggressive 2041 * then RFC3390 here: 2042 * - Only if time spent in the SYN or SYN|ACK retransmition 2043 * >= 3 seconds, the IW is reduced. We do this mainly 2044 * because when RFC3390 is published, the initial RTO is 2045 * still 3 seconds (the threshold we test here), while 2046 * after RFC6298, the initial RTO is 1 second. This 2047 * behaviour probably still falls within the spirit of 2048 * RFC3390. 2049 * - When IW is reduced, 2*MSS is used instead of 1*MSS. 2050 * Mainly to avoid sender and receiver deadlock until 2051 * delayed ACK timer expires. And even RFC2581 does not 2052 * try to reduce IW upon SYN or SYN|ACK retransmition 2053 * timeout. 2054 * 2055 * See also: 2056 * http://tools.ietf.org/html/draft-ietf-tcpm-initcwnd-03 2057 */ 2058 if (tp->t_rxtsyn >= TCPTV_RTOBASE3) { 2059 return (2 * tp->t_maxseg); 2060 } else { 2061 u_long maxsegs, capsegs; 2062 2063 tcp_rmx_iwsegs(tp, &maxsegs, &capsegs); 2064 return min(maxsegs * tp->t_maxseg, 2065 max(2 * tp->t_maxseg, capsegs * 1460)); 2066 } 2067 } else { 2068 /* 2069 * Even RFC2581 (back to 1999) allows 2*SMSS IW. 2070 * 2071 * Mainly to avoid sender and receiver deadlock 2072 * until delayed ACK timer expires. 2073 */ 2074 return (2 * tp->t_maxseg); 2075 } 2076 } 2077 2078 #ifdef TCP_SIGNATURE 2079 /* 2080 * Compute TCP-MD5 hash of a TCP segment. (RFC2385) 2081 * 2082 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2083 * When called from tcp_input(), we can be sure that th_sum has been 2084 * zeroed out and verified already. 2085 * 2086 * Return 0 if successful, otherwise return -1. 2087 * 2088 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2089 * search with the destination IP address, and a 'magic SPI' to be 2090 * determined by the application. This is hardcoded elsewhere to 1179 2091 * right now. Another branch of this code exists which uses the SPD to 2092 * specify per-application flows but it is unstable. 2093 */ 2094 int 2095 tcpsignature_compute( 2096 struct mbuf *m, /* mbuf chain */ 2097 int len, /* length of TCP data */ 2098 int optlen, /* length of TCP options */ 2099 u_char *buf, /* storage for MD5 digest */ 2100 u_int direction) /* direction of flow */ 2101 { 2102 struct ippseudo ippseudo; 2103 MD5_CTX ctx; 2104 int doff; 2105 struct ip *ip; 2106 struct ipovly *ipovly; 2107 struct secasvar *sav; 2108 struct tcphdr *th; 2109 #ifdef INET6 2110 struct ip6_hdr *ip6; 2111 struct in6_addr in6; 2112 uint32_t plen; 2113 uint16_t nhdr; 2114 #endif /* INET6 */ 2115 u_short savecsum; 2116 2117 KASSERT(m != NULL, ("passed NULL mbuf. Game over.")); 2118 KASSERT(buf != NULL, ("passed NULL storage pointer for MD5 signature")); 2119 /* 2120 * Extract the destination from the IP header in the mbuf. 2121 */ 2122 ip = mtod(m, struct ip *); 2123 #ifdef INET6 2124 ip6 = NULL; /* Make the compiler happy. */ 2125 #endif /* INET6 */ 2126 /* 2127 * Look up an SADB entry which matches the address found in 2128 * the segment. 2129 */ 2130 switch (IP_VHL_V(ip->ip_vhl)) { 2131 case IPVERSION: 2132 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, (caddr_t)&ip->ip_dst, 2133 IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2134 break; 2135 #ifdef INET6 2136 case (IPV6_VERSION >> 4): 2137 ip6 = mtod(m, struct ip6_hdr *); 2138 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, (caddr_t)&ip6->ip6_dst, 2139 IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2140 break; 2141 #endif /* INET6 */ 2142 default: 2143 return (EINVAL); 2144 /* NOTREACHED */ 2145 break; 2146 } 2147 if (sav == NULL) { 2148 kprintf("%s: SADB lookup failed\n", __func__); 2149 return (EINVAL); 2150 } 2151 MD5Init(&ctx); 2152 2153 /* 2154 * Step 1: Update MD5 hash with IP pseudo-header. 2155 * 2156 * XXX The ippseudo header MUST be digested in network byte order, 2157 * or else we'll fail the regression test. Assume all fields we've 2158 * been doing arithmetic on have been in host byte order. 2159 * XXX One cannot depend on ipovly->ih_len here. When called from 2160 * tcp_output(), the underlying ip_len member has not yet been set. 2161 */ 2162 switch (IP_VHL_V(ip->ip_vhl)) { 2163 case IPVERSION: 2164 ipovly = (struct ipovly *)ip; 2165 ippseudo.ippseudo_src = ipovly->ih_src; 2166 ippseudo.ippseudo_dst = ipovly->ih_dst; 2167 ippseudo.ippseudo_pad = 0; 2168 ippseudo.ippseudo_p = IPPROTO_TCP; 2169 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2170 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2171 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip)); 2172 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen; 2173 break; 2174 #ifdef INET6 2175 /* 2176 * RFC 2385, 2.0 Proposal 2177 * For IPv6, the pseudo-header is as described in RFC 2460, namely the 2178 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero- 2179 * extended next header value (to form 32 bits), and 32-bit segment 2180 * length. 2181 * Note: Upper-Layer Packet Length comes before Next Header. 2182 */ 2183 case (IPV6_VERSION >> 4): 2184 in6 = ip6->ip6_src; 2185 in6_clearscope(&in6); 2186 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2187 in6 = ip6->ip6_dst; 2188 in6_clearscope(&in6); 2189 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2190 plen = htonl(len + sizeof(struct tcphdr) + optlen); 2191 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t)); 2192 nhdr = 0; 2193 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2194 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2195 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2196 nhdr = IPPROTO_TCP; 2197 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2198 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr)); 2199 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen; 2200 break; 2201 #endif /* INET6 */ 2202 default: 2203 return (EINVAL); 2204 /* NOTREACHED */ 2205 break; 2206 } 2207 /* 2208 * Step 2: Update MD5 hash with TCP header, excluding options. 2209 * The TCP checksum must be set to zero. 2210 */ 2211 savecsum = th->th_sum; 2212 th->th_sum = 0; 2213 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2214 th->th_sum = savecsum; 2215 /* 2216 * Step 3: Update MD5 hash with TCP segment data. 2217 * Use m_apply() to avoid an early m_pullup(). 2218 */ 2219 if (len > 0) 2220 m_apply(m, doff, len, tcpsignature_apply, &ctx); 2221 /* 2222 * Step 4: Update MD5 hash with shared secret. 2223 */ 2224 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2225 MD5Final(buf, &ctx); 2226 key_sa_recordxfer(sav, m); 2227 key_freesav(sav); 2228 return (0); 2229 } 2230 2231 int 2232 tcpsignature_apply(void *fstate, void *data, unsigned int len) 2233 { 2234 2235 MD5Update((MD5_CTX *)fstate, (unsigned char *)data, len); 2236 return (0); 2237 } 2238 #endif /* TCP_SIGNATURE */ 2239 2240 boolean_t 2241 tcp_tso_pullup(struct mbuf **mp, int hoff, struct ip **ip0, int *iphlen0, 2242 struct tcphdr **th0, int *thoff0) 2243 { 2244 struct mbuf *m = *mp; 2245 struct ip *ip; 2246 int len, iphlen; 2247 struct tcphdr *th; 2248 int thoff; 2249 2250 len = hoff + sizeof(struct ip); 2251 2252 /* The fixed IP header must reside completely in the first mbuf. */ 2253 if (m->m_len < len) { 2254 m = m_pullup(m, len); 2255 if (m == NULL) 2256 goto fail; 2257 } 2258 2259 ip = mtodoff(m, struct ip *, hoff); 2260 iphlen = IP_VHL_HL(ip->ip_vhl) << 2; 2261 2262 /* The full IP header must reside completely in the one mbuf. */ 2263 if (m->m_len < hoff + iphlen) { 2264 m = m_pullup(m, hoff + iphlen); 2265 if (m == NULL) 2266 goto fail; 2267 ip = mtodoff(m, struct ip *, hoff); 2268 } 2269 2270 KASSERT(ip->ip_p == IPPROTO_TCP, ("not tcp %d", ip->ip_p)); 2271 2272 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr)) { 2273 m = m_pullup(m, hoff + iphlen + sizeof(struct tcphdr)); 2274 if (m == NULL) 2275 goto fail; 2276 ip = mtodoff(m, struct ip *, hoff); 2277 } 2278 2279 th = (struct tcphdr *)((caddr_t)ip + iphlen); 2280 thoff = th->th_off << 2; 2281 2282 if (m->m_len < hoff + iphlen + thoff) { 2283 m = m_pullup(m, hoff + iphlen + thoff); 2284 if (m == NULL) 2285 goto fail; 2286 ip = mtodoff(m, struct ip *, hoff); 2287 th = (struct tcphdr *)((caddr_t)ip + iphlen); 2288 } 2289 2290 *mp = m; 2291 *ip0 = ip; 2292 *iphlen0 = iphlen; 2293 *th0 = th; 2294 *thoff0 = thoff; 2295 return TRUE; 2296 2297 fail: 2298 if (m != NULL) 2299 m_freem(m); 2300 *mp = NULL; 2301 return FALSE; 2302 } 2303