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. Neither the name of the University nor the names of its contributors 47 * may be used to endorse or promote products derived from this software 48 * without specific prior written permission. 49 * 50 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 53 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 60 * SUCH DAMAGE. 61 * 62 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 63 * $FreeBSD: src/sys/netinet/tcp_subr.c,v 1.73.2.31 2003/01/24 05:11:34 sam Exp $ 64 */ 65 66 #include "opt_compat.h" 67 #include "opt_inet.h" 68 #include "opt_inet6.h" 69 #include "opt_ipsec.h" 70 #include "opt_tcpdebug.h" 71 72 #include <sys/param.h> 73 #include <sys/systm.h> 74 #include <sys/callout.h> 75 #include <sys/kernel.h> 76 #include <sys/sysctl.h> 77 #include <sys/malloc.h> 78 #include <sys/mpipe.h> 79 #include <sys/mbuf.h> 80 #ifdef INET6 81 #include <sys/domain.h> 82 #endif 83 #include <sys/proc.h> 84 #include <sys/priv.h> 85 #include <sys/socket.h> 86 #include <sys/socketops.h> 87 #include <sys/socketvar.h> 88 #include <sys/protosw.h> 89 #include <sys/random.h> 90 #include <sys/in_cksum.h> 91 #include <sys/ktr.h> 92 93 #include <net/route.h> 94 #include <net/if.h> 95 #include <net/netisr2.h> 96 97 #define _IP_VHL 98 #include <netinet/in.h> 99 #include <netinet/in_systm.h> 100 #include <netinet/ip.h> 101 #include <netinet/ip6.h> 102 #include <netinet/in_pcb.h> 103 #include <netinet6/in6_pcb.h> 104 #include <netinet/in_var.h> 105 #include <netinet/ip_var.h> 106 #include <netinet6/ip6_var.h> 107 #include <netinet/ip_icmp.h> 108 #ifdef INET6 109 #include <netinet/icmp6.h> 110 #endif 111 #include <netinet/tcp.h> 112 #include <netinet/tcp_fsm.h> 113 #include <netinet/tcp_seq.h> 114 #include <netinet/tcp_timer.h> 115 #include <netinet/tcp_timer2.h> 116 #include <netinet/tcp_var.h> 117 #include <netinet6/tcp6_var.h> 118 #include <netinet/tcpip.h> 119 #ifdef TCPDEBUG 120 #include <netinet/tcp_debug.h> 121 #endif 122 #include <netinet6/ip6protosw.h> 123 124 #ifdef IPSEC 125 #include <netinet6/ipsec.h> 126 #include <netproto/key/key.h> 127 #ifdef INET6 128 #include <netinet6/ipsec6.h> 129 #endif 130 #endif 131 132 #ifdef FAST_IPSEC 133 #include <netproto/ipsec/ipsec.h> 134 #ifdef INET6 135 #include <netproto/ipsec/ipsec6.h> 136 #endif 137 #define IPSEC 138 #endif 139 140 #include <sys/md5.h> 141 #include <machine/smp.h> 142 143 #include <sys/msgport2.h> 144 #include <sys/mplock2.h> 145 #include <net/netmsg2.h> 146 147 #if !defined(KTR_TCP) 148 #define KTR_TCP KTR_ALL 149 #endif 150 /* 151 KTR_INFO_MASTER(tcp); 152 KTR_INFO(KTR_TCP, tcp, rxmsg, 0, "tcp getmsg", 0); 153 KTR_INFO(KTR_TCP, tcp, wait, 1, "tcp waitmsg", 0); 154 KTR_INFO(KTR_TCP, tcp, delayed, 2, "tcp execute delayed ops", 0); 155 #define logtcp(name) KTR_LOG(tcp_ ## name) 156 */ 157 158 #define TCP_IW_MAXSEGS_DFLT 4 159 #define TCP_IW_CAPSEGS_DFLT 3 160 161 struct inpcbinfo tcbinfo[MAXCPU]; 162 struct tcpcbackqhead tcpcbackq[MAXCPU]; 163 164 int tcp_mssdflt = TCP_MSS; 165 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 166 &tcp_mssdflt, 0, "Default TCP Maximum Segment Size"); 167 168 #ifdef INET6 169 int tcp_v6mssdflt = TCP6_MSS; 170 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLFLAG_RW, 171 &tcp_v6mssdflt, 0, "Default TCP Maximum Segment Size for IPv6"); 172 #endif 173 174 /* 175 * Minimum MSS we accept and use. This prevents DoS attacks where 176 * we are forced to a ridiculous low MSS like 20 and send hundreds 177 * of packets instead of one. The effect scales with the available 178 * bandwidth and quickly saturates the CPU and network interface 179 * with packet generation and sending. Set to zero to disable MINMSS 180 * checking. This setting prevents us from sending too small packets. 181 */ 182 int tcp_minmss = TCP_MINMSS; 183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 184 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 185 186 #if 0 187 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 188 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 189 &tcp_rttdflt, 0, "Default maximum TCP Round Trip Time"); 190 #endif 191 192 int tcp_do_rfc1323 = 1; 193 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 194 &tcp_do_rfc1323, 0, "Enable rfc1323 (high performance TCP) extensions"); 195 196 static int tcp_tcbhashsize = 0; 197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD, 198 &tcp_tcbhashsize, 0, "Size of TCP control block hashtable"); 199 200 static int do_tcpdrain = 1; 201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 202 "Enable tcp_drain routine for extra help when low on mbufs"); 203 204 static int icmp_may_rst = 1; 205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 206 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 207 208 static int tcp_isn_reseed_interval = 0; 209 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 210 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 211 212 /* 213 * TCP bandwidth limiting sysctls. The inflight limiter is now turned on 214 * by default, but with generous values which should allow maximal 215 * bandwidth. In particular, the slop defaults to 50 (5 packets). 216 * 217 * The reason for doing this is that the limiter is the only mechanism we 218 * have which seems to do a really good job preventing receiver RX rings 219 * on network interfaces from getting blown out. Even though GigE/10GigE 220 * is supposed to flow control it looks like either it doesn't actually 221 * do it or Open Source drivers do not properly enable it. 222 * 223 * People using the limiter to reduce bottlenecks on slower WAN connections 224 * should set the slop to 20 (2 packets). 225 */ 226 static int tcp_inflight_enable = 1; 227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_enable, CTLFLAG_RW, 228 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 229 230 static int tcp_inflight_debug = 0; 231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_debug, CTLFLAG_RW, 232 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 233 234 static int tcp_inflight_min = 6144; 235 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_min, CTLFLAG_RW, 236 &tcp_inflight_min, 0, "Lower bound for TCP inflight window"); 237 238 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 239 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_max, CTLFLAG_RW, 240 &tcp_inflight_max, 0, "Upper bound for TCP inflight window"); 241 242 static int tcp_inflight_stab = 50; 243 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_stab, CTLFLAG_RW, 244 &tcp_inflight_stab, 0, "Fudge bw 1/10% (50=5%)"); 245 246 static int tcp_inflight_adjrtt = 2; 247 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_adjrtt, CTLFLAG_RW, 248 &tcp_inflight_adjrtt, 0, "Slop for rtt 1/(hz*32)"); 249 250 static int tcp_do_rfc3390 = 1; 251 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 252 &tcp_do_rfc3390, 0, 253 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 254 255 static u_long tcp_iw_maxsegs = TCP_IW_MAXSEGS_DFLT; 256 SYSCTL_ULONG(_net_inet_tcp, OID_AUTO, iwmaxsegs, CTLFLAG_RW, 257 &tcp_iw_maxsegs, 0, "TCP IW segments max"); 258 259 static u_long tcp_iw_capsegs = TCP_IW_CAPSEGS_DFLT; 260 SYSCTL_ULONG(_net_inet_tcp, OID_AUTO, iwcapsegs, CTLFLAG_RW, 261 &tcp_iw_capsegs, 0, "TCP IW segments"); 262 263 int tcp_low_rtobase = 1; 264 SYSCTL_INT(_net_inet_tcp, OID_AUTO, low_rtobase, CTLFLAG_RW, 265 &tcp_low_rtobase, 0, "Lowering the Initial RTO (RFC 6298)"); 266 267 static int tcp_do_ncr = 1; 268 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ncr, CTLFLAG_RW, 269 &tcp_do_ncr, 0, "Non-Congestion Robustness (RFC 4653)"); 270 271 int tcp_ncr_rxtthresh_max = 16; 272 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ncr_rxtthresh_max, CTLFLAG_RW, 273 &tcp_ncr_rxtthresh_max, 0, 274 "Non-Congestion Robustness (RFC 4653), DupThresh upper limit"); 275 276 static MALLOC_DEFINE(M_TCPTEMP, "tcptemp", "TCP Templates for Keepalives"); 277 static struct malloc_pipe tcptemp_mpipe; 278 279 static void tcp_willblock(void); 280 static void tcp_notify (struct inpcb *, int); 281 282 struct tcp_stats tcpstats_percpu[MAXCPU] __cachealign; 283 284 static int 285 sysctl_tcpstats(SYSCTL_HANDLER_ARGS) 286 { 287 int cpu, error = 0; 288 289 for (cpu = 0; cpu < ncpus2; ++cpu) { 290 if ((error = SYSCTL_OUT(req, &tcpstats_percpu[cpu], 291 sizeof(struct tcp_stats)))) 292 break; 293 if ((error = SYSCTL_IN(req, &tcpstats_percpu[cpu], 294 sizeof(struct tcp_stats)))) 295 break; 296 } 297 298 return (error); 299 } 300 SYSCTL_PROC(_net_inet_tcp, TCPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW), 301 0, 0, sysctl_tcpstats, "S,tcp_stats", "TCP statistics"); 302 303 /* 304 * Target size of TCP PCB hash tables. Must be a power of two. 305 * 306 * Note that this can be overridden by the kernel environment 307 * variable net.inet.tcp.tcbhashsize 308 */ 309 #ifndef TCBHASHSIZE 310 #define TCBHASHSIZE 512 311 #endif 312 313 /* 314 * This is the actual shape of what we allocate using the zone 315 * allocator. Doing it this way allows us to protect both structures 316 * using the same generation count, and also eliminates the overhead 317 * of allocating tcpcbs separately. By hiding the structure here, 318 * we avoid changing most of the rest of the code (although it needs 319 * to be changed, eventually, for greater efficiency). 320 */ 321 #define ALIGNMENT 32 322 #define ALIGNM1 (ALIGNMENT - 1) 323 struct inp_tp { 324 union { 325 struct inpcb inp; 326 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1]; 327 } inp_tp_u; 328 struct tcpcb tcb; 329 struct tcp_callout inp_tp_rexmt; 330 struct tcp_callout inp_tp_persist; 331 struct tcp_callout inp_tp_keep; 332 struct tcp_callout inp_tp_2msl; 333 struct tcp_callout inp_tp_delack; 334 struct netmsg_tcp_timer inp_tp_timermsg; 335 struct netmsg_base inp_tp_sndmore; 336 }; 337 #undef ALIGNMENT 338 #undef ALIGNM1 339 340 /* 341 * Tcp initialization 342 */ 343 void 344 tcp_init(void) 345 { 346 struct inpcbportinfo *portinfo; 347 struct inpcbinfo *ticb; 348 int hashsize = TCBHASHSIZE; 349 int cpu; 350 351 /* 352 * note: tcptemp is used for keepalives, and it is ok for an 353 * allocation to fail so do not specify MPF_INT. 354 */ 355 mpipe_init(&tcptemp_mpipe, M_TCPTEMP, sizeof(struct tcptemp), 356 25, -1, 0, NULL, NULL, NULL); 357 358 tcp_delacktime = TCPTV_DELACK; 359 tcp_keepinit = TCPTV_KEEP_INIT; 360 tcp_keepidle = TCPTV_KEEP_IDLE; 361 tcp_keepintvl = TCPTV_KEEPINTVL; 362 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 363 tcp_msl = TCPTV_MSL; 364 tcp_rexmit_min = TCPTV_MIN; 365 tcp_rexmit_slop = TCPTV_CPU_VAR; 366 367 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 368 if (!powerof2(hashsize)) { 369 kprintf("WARNING: TCB hash size not a power of 2\n"); 370 hashsize = 512; /* safe default */ 371 } 372 tcp_tcbhashsize = hashsize; 373 374 portinfo = kmalloc_cachealign(sizeof(*portinfo) * ncpus2, M_PCB, 375 M_WAITOK); 376 377 for (cpu = 0; cpu < ncpus2; cpu++) { 378 ticb = &tcbinfo[cpu]; 379 in_pcbinfo_init(ticb, cpu, FALSE); 380 ticb->hashbase = hashinit(hashsize, M_PCB, 381 &ticb->hashmask); 382 in_pcbportinfo_init(&portinfo[cpu], hashsize, TRUE, cpu); 383 ticb->portinfo = portinfo; 384 ticb->portinfo_mask = ncpus2_mask; 385 ticb->wildcardhashbase = hashinit(hashsize, M_PCB, 386 &ticb->wildcardhashmask); 387 ticb->localgrphashbase = hashinit(hashsize, M_PCB, 388 &ticb->localgrphashmask); 389 ticb->ipi_size = sizeof(struct inp_tp); 390 TAILQ_INIT(&tcpcbackq[cpu]); 391 } 392 393 tcp_reass_maxseg = nmbclusters / 16; 394 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", &tcp_reass_maxseg); 395 396 #ifdef INET6 397 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 398 #else 399 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 400 #endif 401 if (max_protohdr < TCP_MINPROTOHDR) 402 max_protohdr = TCP_MINPROTOHDR; 403 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 404 panic("tcp_init"); 405 #undef TCP_MINPROTOHDR 406 407 /* 408 * Initialize TCP statistics counters for each CPU. 409 */ 410 for (cpu = 0; cpu < ncpus2; ++cpu) 411 bzero(&tcpstats_percpu[cpu], sizeof(struct tcp_stats)); 412 413 syncache_init(); 414 netisr_register_rollup(tcp_willblock, NETISR_ROLLUP_PRIO_TCP); 415 } 416 417 static void 418 tcp_willblock(void) 419 { 420 struct tcpcb *tp; 421 int cpu = mycpu->gd_cpuid; 422 423 while ((tp = TAILQ_FIRST(&tcpcbackq[cpu])) != NULL) { 424 KKASSERT(tp->t_flags & TF_ONOUTPUTQ); 425 tp->t_flags &= ~TF_ONOUTPUTQ; 426 TAILQ_REMOVE(&tcpcbackq[cpu], tp, t_outputq); 427 tcp_output(tp); 428 } 429 } 430 431 /* 432 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 433 * tcp_template used to store this data in mbufs, but we now recopy it out 434 * of the tcpcb each time to conserve mbufs. 435 */ 436 void 437 tcp_fillheaders(struct tcpcb *tp, void *ip_ptr, void *tcp_ptr, boolean_t tso) 438 { 439 struct inpcb *inp = tp->t_inpcb; 440 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr; 441 442 #ifdef INET6 443 if (inp->inp_vflag & INP_IPV6) { 444 struct ip6_hdr *ip6; 445 446 ip6 = (struct ip6_hdr *)ip_ptr; 447 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 448 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 449 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 450 (IPV6_VERSION & IPV6_VERSION_MASK); 451 ip6->ip6_nxt = IPPROTO_TCP; 452 ip6->ip6_plen = sizeof(struct tcphdr); 453 ip6->ip6_src = inp->in6p_laddr; 454 ip6->ip6_dst = inp->in6p_faddr; 455 tcp_hdr->th_sum = 0; 456 } else 457 #endif 458 { 459 struct ip *ip = (struct ip *) ip_ptr; 460 u_int plen; 461 462 ip->ip_vhl = IP_VHL_BORING; 463 ip->ip_tos = 0; 464 ip->ip_len = 0; 465 ip->ip_id = 0; 466 ip->ip_off = 0; 467 ip->ip_ttl = 0; 468 ip->ip_sum = 0; 469 ip->ip_p = IPPROTO_TCP; 470 ip->ip_src = inp->inp_laddr; 471 ip->ip_dst = inp->inp_faddr; 472 473 if (tso) 474 plen = htons(IPPROTO_TCP); 475 else 476 plen = htons(sizeof(struct tcphdr) + IPPROTO_TCP); 477 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, 478 ip->ip_dst.s_addr, plen); 479 } 480 481 tcp_hdr->th_sport = inp->inp_lport; 482 tcp_hdr->th_dport = inp->inp_fport; 483 tcp_hdr->th_seq = 0; 484 tcp_hdr->th_ack = 0; 485 tcp_hdr->th_x2 = 0; 486 tcp_hdr->th_off = 5; 487 tcp_hdr->th_flags = 0; 488 tcp_hdr->th_win = 0; 489 tcp_hdr->th_urp = 0; 490 } 491 492 /* 493 * Create template to be used to send tcp packets on a connection. 494 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 495 * use for this function is in keepalives, which use tcp_respond. 496 */ 497 struct tcptemp * 498 tcp_maketemplate(struct tcpcb *tp) 499 { 500 struct tcptemp *tmp; 501 502 if ((tmp = mpipe_alloc_nowait(&tcptemp_mpipe)) == NULL) 503 return (NULL); 504 tcp_fillheaders(tp, &tmp->tt_ipgen, &tmp->tt_t, FALSE); 505 return (tmp); 506 } 507 508 void 509 tcp_freetemplate(struct tcptemp *tmp) 510 { 511 mpipe_free(&tcptemp_mpipe, tmp); 512 } 513 514 /* 515 * Send a single message to the TCP at address specified by 516 * the given TCP/IP header. If m == NULL, then we make a copy 517 * of the tcpiphdr at ti and send directly to the addressed host. 518 * This is used to force keep alive messages out using the TCP 519 * template for a connection. If flags are given then we send 520 * a message back to the TCP which originated the * segment ti, 521 * and discard the mbuf containing it and any other attached mbufs. 522 * 523 * In any case the ack and sequence number of the transmitted 524 * segment are as specified by the parameters. 525 * 526 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 527 */ 528 void 529 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 530 tcp_seq ack, tcp_seq seq, int flags) 531 { 532 int tlen; 533 long win = 0; 534 struct route *ro = NULL; 535 struct route sro; 536 struct ip *ip = ipgen; 537 struct tcphdr *nth; 538 int ipflags = 0; 539 struct route_in6 *ro6 = NULL; 540 struct route_in6 sro6; 541 struct ip6_hdr *ip6 = ipgen; 542 boolean_t use_tmpro = TRUE; 543 #ifdef INET6 544 boolean_t isipv6 = (IP_VHL_V(ip->ip_vhl) == 6); 545 #else 546 const boolean_t isipv6 = FALSE; 547 #endif 548 549 if (tp != NULL) { 550 if (!(flags & TH_RST)) { 551 win = ssb_space(&tp->t_inpcb->inp_socket->so_rcv); 552 if (win < 0) 553 win = 0; 554 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 555 win = (long)TCP_MAXWIN << tp->rcv_scale; 556 } 557 /* 558 * Don't use the route cache of a listen socket, 559 * it is not MPSAFE; use temporary route cache. 560 */ 561 if (tp->t_state != TCPS_LISTEN) { 562 if (isipv6) 563 ro6 = &tp->t_inpcb->in6p_route; 564 else 565 ro = &tp->t_inpcb->inp_route; 566 use_tmpro = FALSE; 567 } 568 } 569 if (use_tmpro) { 570 if (isipv6) { 571 ro6 = &sro6; 572 bzero(ro6, sizeof *ro6); 573 } else { 574 ro = &sro; 575 bzero(ro, sizeof *ro); 576 } 577 } 578 if (m == NULL) { 579 m = m_gethdr(MB_DONTWAIT, MT_HEADER); 580 if (m == NULL) 581 return; 582 tlen = 0; 583 m->m_data += max_linkhdr; 584 if (isipv6) { 585 bcopy(ip6, mtod(m, caddr_t), sizeof(struct ip6_hdr)); 586 ip6 = mtod(m, struct ip6_hdr *); 587 nth = (struct tcphdr *)(ip6 + 1); 588 } else { 589 bcopy(ip, mtod(m, caddr_t), sizeof(struct ip)); 590 ip = mtod(m, struct ip *); 591 nth = (struct tcphdr *)(ip + 1); 592 } 593 bcopy(th, nth, sizeof(struct tcphdr)); 594 flags = TH_ACK; 595 } else { 596 m_freem(m->m_next); 597 m->m_next = NULL; 598 m->m_data = (caddr_t)ipgen; 599 /* m_len is set later */ 600 tlen = 0; 601 #define xchg(a, b, type) { type t; t = a; a = b; b = t; } 602 if (isipv6) { 603 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 604 nth = (struct tcphdr *)(ip6 + 1); 605 } else { 606 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 607 nth = (struct tcphdr *)(ip + 1); 608 } 609 if (th != nth) { 610 /* 611 * this is usually a case when an extension header 612 * exists between the IPv6 header and the 613 * TCP header. 614 */ 615 nth->th_sport = th->th_sport; 616 nth->th_dport = th->th_dport; 617 } 618 xchg(nth->th_dport, nth->th_sport, n_short); 619 #undef xchg 620 } 621 if (isipv6) { 622 ip6->ip6_flow = 0; 623 ip6->ip6_vfc = IPV6_VERSION; 624 ip6->ip6_nxt = IPPROTO_TCP; 625 ip6->ip6_plen = htons((u_short)(sizeof(struct tcphdr) + tlen)); 626 tlen += sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 627 } else { 628 tlen += sizeof(struct tcpiphdr); 629 ip->ip_len = tlen; 630 ip->ip_ttl = ip_defttl; 631 } 632 m->m_len = tlen; 633 m->m_pkthdr.len = tlen; 634 m->m_pkthdr.rcvif = NULL; 635 nth->th_seq = htonl(seq); 636 nth->th_ack = htonl(ack); 637 nth->th_x2 = 0; 638 nth->th_off = sizeof(struct tcphdr) >> 2; 639 nth->th_flags = flags; 640 if (tp != NULL) 641 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 642 else 643 nth->th_win = htons((u_short)win); 644 nth->th_urp = 0; 645 if (isipv6) { 646 nth->th_sum = 0; 647 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 648 sizeof(struct ip6_hdr), 649 tlen - sizeof(struct ip6_hdr)); 650 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL, 651 (ro6 && ro6->ro_rt) ? 652 ro6->ro_rt->rt_ifp : NULL); 653 } else { 654 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 655 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 656 m->m_pkthdr.csum_flags = CSUM_TCP; 657 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 658 m->m_pkthdr.csum_thlen = sizeof(struct tcphdr); 659 } 660 #ifdef TCPDEBUG 661 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 662 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 663 #endif 664 if (isipv6) { 665 ip6_output(m, NULL, ro6, ipflags, NULL, NULL, 666 tp ? tp->t_inpcb : NULL); 667 if ((ro6 == &sro6) && (ro6->ro_rt != NULL)) { 668 RTFREE(ro6->ro_rt); 669 ro6->ro_rt = NULL; 670 } 671 } else { 672 ipflags |= IP_DEBUGROUTE; 673 ip_output(m, NULL, ro, ipflags, NULL, tp ? tp->t_inpcb : NULL); 674 if ((ro == &sro) && (ro->ro_rt != NULL)) { 675 RTFREE(ro->ro_rt); 676 ro->ro_rt = NULL; 677 } 678 } 679 } 680 681 /* 682 * Create a new TCP control block, making an 683 * empty reassembly queue and hooking it to the argument 684 * protocol control block. The `inp' parameter must have 685 * come from the zone allocator set up in tcp_init(). 686 */ 687 struct tcpcb * 688 tcp_newtcpcb(struct inpcb *inp) 689 { 690 struct inp_tp *it; 691 struct tcpcb *tp; 692 #ifdef INET6 693 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0); 694 #else 695 const boolean_t isipv6 = FALSE; 696 #endif 697 698 it = (struct inp_tp *)inp; 699 tp = &it->tcb; 700 bzero(tp, sizeof(struct tcpcb)); 701 TAILQ_INIT(&tp->t_segq); 702 tp->t_maxseg = tp->t_maxopd = isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 703 tp->t_rxtthresh = tcprexmtthresh; 704 705 /* Set up our timeouts. */ 706 tp->tt_rexmt = &it->inp_tp_rexmt; 707 tp->tt_persist = &it->inp_tp_persist; 708 tp->tt_keep = &it->inp_tp_keep; 709 tp->tt_2msl = &it->inp_tp_2msl; 710 tp->tt_delack = &it->inp_tp_delack; 711 tcp_inittimers(tp); 712 713 /* 714 * Zero out timer message. We don't create it here, 715 * since the current CPU may not be the owner of this 716 * inpcb. 717 */ 718 tp->tt_msg = &it->inp_tp_timermsg; 719 bzero(tp->tt_msg, sizeof(*tp->tt_msg)); 720 721 tp->t_keepinit = tcp_keepinit; 722 tp->t_keepidle = tcp_keepidle; 723 tp->t_keepintvl = tcp_keepintvl; 724 tp->t_keepcnt = tcp_keepcnt; 725 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt; 726 727 if (tcp_do_ncr) 728 tp->t_flags |= TF_NCR; 729 if (tcp_do_rfc1323) 730 tp->t_flags |= (TF_REQ_SCALE | TF_REQ_TSTMP); 731 732 tp->t_inpcb = inp; /* XXX */ 733 tp->t_state = TCPS_CLOSED; 734 /* 735 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 736 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 737 * reasonable initial retransmit time. 738 */ 739 tp->t_srtt = TCPTV_SRTTBASE; 740 tp->t_rttvar = 741 ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 742 tp->t_rttmin = tcp_rexmit_min; 743 tp->t_rxtcur = TCPTV_RTOBASE; 744 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 745 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 746 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 747 tp->snd_last = ticks; 748 tp->t_rcvtime = ticks; 749 /* 750 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 751 * because the socket may be bound to an IPv6 wildcard address, 752 * which may match an IPv4-mapped IPv6 address. 753 */ 754 inp->inp_ip_ttl = ip_defttl; 755 inp->inp_ppcb = tp; 756 tcp_sack_tcpcb_init(tp); 757 758 tp->tt_sndmore = &it->inp_tp_sndmore; 759 tcp_output_init(tp); 760 761 return (tp); /* XXX */ 762 } 763 764 /* 765 * Drop a TCP connection, reporting the specified error. 766 * If connection is synchronized, then send a RST to peer. 767 */ 768 struct tcpcb * 769 tcp_drop(struct tcpcb *tp, int error) 770 { 771 struct socket *so = tp->t_inpcb->inp_socket; 772 773 if (TCPS_HAVERCVDSYN(tp->t_state)) { 774 tp->t_state = TCPS_CLOSED; 775 tcp_output(tp); 776 tcpstat.tcps_drops++; 777 } else 778 tcpstat.tcps_conndrops++; 779 if (error == ETIMEDOUT && tp->t_softerror) 780 error = tp->t_softerror; 781 so->so_error = error; 782 return (tcp_close(tp)); 783 } 784 785 struct netmsg_listen_detach { 786 struct netmsg_base base; 787 struct tcpcb *nm_tp; 788 struct tcpcb *nm_tp_inh; 789 }; 790 791 static void 792 tcp_listen_detach_handler(netmsg_t msg) 793 { 794 struct netmsg_listen_detach *nmsg = (struct netmsg_listen_detach *)msg; 795 struct tcpcb *tp = nmsg->nm_tp; 796 int cpu = mycpuid, nextcpu; 797 798 if (tp->t_flags & TF_LISTEN) 799 syncache_destroy(tp, nmsg->nm_tp_inh); 800 801 in_pcbremwildcardhash_oncpu(tp->t_inpcb, &tcbinfo[cpu]); 802 803 nextcpu = cpu + 1; 804 if (nextcpu < ncpus2) 805 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nmsg->base.lmsg); 806 else 807 lwkt_replymsg(&nmsg->base.lmsg, 0); 808 } 809 810 /* 811 * Close a TCP control block: 812 * discard all space held by the tcp 813 * discard internet protocol block 814 * wake up any sleepers 815 */ 816 struct tcpcb * 817 tcp_close(struct tcpcb *tp) 818 { 819 struct tseg_qent *q; 820 struct inpcb *inp = tp->t_inpcb; 821 struct inpcb *inp_inh = NULL; 822 struct tcpcb *tp_inh = NULL; 823 struct socket *so = inp->inp_socket; 824 struct rtentry *rt; 825 boolean_t dosavessthresh; 826 #ifdef INET6 827 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0); 828 boolean_t isafinet6 = (INP_CHECK_SOCKAF(so, AF_INET6) != 0); 829 #else 830 const boolean_t isipv6 = FALSE; 831 #endif 832 833 if (tp->t_flags & TF_LISTEN) { 834 /* 835 * Pending socket/syncache inheritance 836 * 837 * If this is a listen(2) socket, find another listen(2) 838 * socket in the same local group, which could inherit 839 * the syncache and sockets pending on the completion 840 * and incompletion queues. 841 * 842 * NOTE: 843 * Currently the inheritance could only happen on the 844 * listen(2) sockets w/ SO_REUSEPORT set. 845 */ 846 KASSERT(&curthread->td_msgport == netisr_cpuport(0), 847 ("listen socket close not in netisr0")); 848 inp_inh = in_pcblocalgroup_last(&tcbinfo[0], inp); 849 if (inp_inh != NULL) 850 tp_inh = intotcpcb(inp_inh); 851 } 852 853 /* 854 * INP_WILDCARD_MP indicates that listen(2) has been called on 855 * this socket. This implies: 856 * - A wildcard inp's hash is replicated for each protocol thread. 857 * - Syncache for this inp grows independently in each protocol 858 * thread. 859 * - There is more than one cpu 860 * 861 * We have to chain a message to the rest of the protocol threads 862 * to cleanup the wildcard hash and the syncache. The cleanup 863 * in the current protocol thread is defered till the end of this 864 * function. 865 * 866 * NOTE: 867 * After cleanup the inp's hash and syncache entries, this inp will 868 * no longer be available to the rest of the protocol threads, so we 869 * are safe to whack the inp in the following code. 870 */ 871 if (inp->inp_flags & INP_WILDCARD_MP) { 872 struct netmsg_listen_detach nmsg; 873 874 KKASSERT(so->so_port == netisr_cpuport(0)); 875 KKASSERT(&curthread->td_msgport == netisr_cpuport(0)); 876 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]); 877 878 netmsg_init(&nmsg.base, NULL, &curthread->td_msgport, 879 MSGF_PRIORITY, tcp_listen_detach_handler); 880 nmsg.nm_tp = tp; 881 nmsg.nm_tp_inh = tp_inh; 882 lwkt_domsg(netisr_cpuport(1), &nmsg.base.lmsg, 0); 883 884 inp->inp_flags &= ~INP_WILDCARD_MP; 885 } 886 887 KKASSERT(tp->t_state != TCPS_TERMINATING); 888 tp->t_state = TCPS_TERMINATING; 889 890 /* 891 * Make sure that all of our timers are stopped before we 892 * delete the PCB. For listen TCP socket (tp->tt_msg == NULL), 893 * timers are never used. If timer message is never created 894 * (tp->tt_msg->tt_tcb == NULL), timers are never used too. 895 */ 896 if (tp->tt_msg != NULL && tp->tt_msg->tt_tcb != NULL) { 897 tcp_callout_stop(tp, tp->tt_rexmt); 898 tcp_callout_stop(tp, tp->tt_persist); 899 tcp_callout_stop(tp, tp->tt_keep); 900 tcp_callout_stop(tp, tp->tt_2msl); 901 tcp_callout_stop(tp, tp->tt_delack); 902 } 903 904 if (tp->t_flags & TF_ONOUTPUTQ) { 905 KKASSERT(tp->tt_cpu == mycpu->gd_cpuid); 906 TAILQ_REMOVE(&tcpcbackq[tp->tt_cpu], tp, t_outputq); 907 tp->t_flags &= ~TF_ONOUTPUTQ; 908 } 909 910 /* 911 * If we got enough samples through the srtt filter, 912 * save the rtt and rttvar in the routing entry. 913 * 'Enough' is arbitrarily defined as the 16 samples. 914 * 16 samples is enough for the srtt filter to converge 915 * to within 5% of the correct value; fewer samples and 916 * we could save a very bogus rtt. 917 * 918 * Don't update the default route's characteristics and don't 919 * update anything that the user "locked". 920 */ 921 if (tp->t_rttupdated >= 16) { 922 u_long i = 0; 923 924 if (isipv6) { 925 struct sockaddr_in6 *sin6; 926 927 if ((rt = inp->in6p_route.ro_rt) == NULL) 928 goto no_valid_rt; 929 sin6 = (struct sockaddr_in6 *)rt_key(rt); 930 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 931 goto no_valid_rt; 932 } else 933 if ((rt = inp->inp_route.ro_rt) == NULL || 934 ((struct sockaddr_in *)rt_key(rt))-> 935 sin_addr.s_addr == INADDR_ANY) 936 goto no_valid_rt; 937 938 if (!(rt->rt_rmx.rmx_locks & RTV_RTT)) { 939 i = tp->t_srtt * (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 940 if (rt->rt_rmx.rmx_rtt && i) 941 /* 942 * filter this update to half the old & half 943 * the new values, converting scale. 944 * See route.h and tcp_var.h for a 945 * description of the scaling constants. 946 */ 947 rt->rt_rmx.rmx_rtt = 948 (rt->rt_rmx.rmx_rtt + i) / 2; 949 else 950 rt->rt_rmx.rmx_rtt = i; 951 tcpstat.tcps_cachedrtt++; 952 } 953 if (!(rt->rt_rmx.rmx_locks & RTV_RTTVAR)) { 954 i = tp->t_rttvar * 955 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 956 if (rt->rt_rmx.rmx_rttvar && i) 957 rt->rt_rmx.rmx_rttvar = 958 (rt->rt_rmx.rmx_rttvar + i) / 2; 959 else 960 rt->rt_rmx.rmx_rttvar = i; 961 tcpstat.tcps_cachedrttvar++; 962 } 963 /* 964 * The old comment here said: 965 * update the pipelimit (ssthresh) if it has been updated 966 * already or if a pipesize was specified & the threshhold 967 * got below half the pipesize. I.e., wait for bad news 968 * before we start updating, then update on both good 969 * and bad news. 970 * 971 * But we want to save the ssthresh even if no pipesize is 972 * specified explicitly in the route, because such 973 * connections still have an implicit pipesize specified 974 * by the global tcp_sendspace. In the absence of a reliable 975 * way to calculate the pipesize, it will have to do. 976 */ 977 i = tp->snd_ssthresh; 978 if (rt->rt_rmx.rmx_sendpipe != 0) 979 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe/2); 980 else 981 dosavessthresh = (i < so->so_snd.ssb_hiwat/2); 982 if (dosavessthresh || 983 (!(rt->rt_rmx.rmx_locks & RTV_SSTHRESH) && (i != 0) && 984 (rt->rt_rmx.rmx_ssthresh != 0))) { 985 /* 986 * convert the limit from user data bytes to 987 * packets then to packet data bytes. 988 */ 989 i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 990 if (i < 2) 991 i = 2; 992 i *= tp->t_maxseg + 993 (isipv6 ? 994 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 995 sizeof(struct tcpiphdr)); 996 if (rt->rt_rmx.rmx_ssthresh) 997 rt->rt_rmx.rmx_ssthresh = 998 (rt->rt_rmx.rmx_ssthresh + i) / 2; 999 else 1000 rt->rt_rmx.rmx_ssthresh = i; 1001 tcpstat.tcps_cachedssthresh++; 1002 } 1003 } 1004 1005 no_valid_rt: 1006 /* free the reassembly queue, if any */ 1007 while((q = TAILQ_FIRST(&tp->t_segq)) != NULL) { 1008 TAILQ_REMOVE(&tp->t_segq, q, tqe_q); 1009 m_freem(q->tqe_m); 1010 kfree(q, M_TSEGQ); 1011 atomic_add_int(&tcp_reass_qsize, -1); 1012 } 1013 /* throw away SACK blocks in scoreboard*/ 1014 if (TCP_DO_SACK(tp)) 1015 tcp_sack_destroy(&tp->scb); 1016 1017 inp->inp_ppcb = NULL; 1018 soisdisconnected(so); 1019 /* note: pcb detached later on */ 1020 1021 tcp_destroy_timermsg(tp); 1022 tcp_output_cancel(tp); 1023 1024 if (tp->t_flags & TF_LISTEN) { 1025 syncache_destroy(tp, tp_inh); 1026 if (inp_inh != NULL && inp_inh->inp_socket != NULL) { 1027 /* 1028 * Pending sockets inheritance only needs 1029 * to be done once in the current thread, 1030 * i.e. netisr0. 1031 */ 1032 soinherit(so, inp_inh->inp_socket); 1033 } 1034 } 1035 1036 so_async_rcvd_drop(so); 1037 /* Drop the reference for the asynchronized pru_rcvd */ 1038 sofree(so); 1039 1040 /* 1041 * NOTE: 1042 * pcbdetach removes any wildcard hash entry on the current CPU. 1043 */ 1044 #ifdef INET6 1045 if (isafinet6) 1046 in6_pcbdetach(inp); 1047 else 1048 #endif 1049 in_pcbdetach(inp); 1050 1051 tcpstat.tcps_closed++; 1052 return (NULL); 1053 } 1054 1055 static __inline void 1056 tcp_drain_oncpu(struct inpcbhead *head) 1057 { 1058 struct inpcb *marker; 1059 struct inpcb *inpb; 1060 struct tcpcb *tcpb; 1061 struct tseg_qent *te; 1062 1063 /* 1064 * Allows us to block while running the list 1065 */ 1066 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO); 1067 marker->inp_flags |= INP_PLACEMARKER; 1068 LIST_INSERT_HEAD(head, marker, inp_list); 1069 1070 while ((inpb = LIST_NEXT(marker, inp_list)) != NULL) { 1071 LIST_REMOVE(marker, inp_list); 1072 LIST_INSERT_AFTER(inpb, marker, inp_list); 1073 1074 if ((inpb->inp_flags & INP_PLACEMARKER) == 0 && 1075 (tcpb = intotcpcb(inpb)) != NULL && 1076 (te = TAILQ_FIRST(&tcpb->t_segq)) != NULL) { 1077 TAILQ_REMOVE(&tcpb->t_segq, te, tqe_q); 1078 if (te->tqe_th->th_flags & TH_FIN) 1079 tcpb->t_flags &= ~TF_QUEDFIN; 1080 m_freem(te->tqe_m); 1081 kfree(te, M_TSEGQ); 1082 atomic_add_int(&tcp_reass_qsize, -1); 1083 /* retry */ 1084 } 1085 } 1086 LIST_REMOVE(marker, inp_list); 1087 kfree(marker, M_TEMP); 1088 } 1089 1090 struct netmsg_tcp_drain { 1091 struct netmsg_base base; 1092 struct inpcbhead *nm_head; 1093 }; 1094 1095 static void 1096 tcp_drain_handler(netmsg_t msg) 1097 { 1098 struct netmsg_tcp_drain *nm = (void *)msg; 1099 1100 tcp_drain_oncpu(nm->nm_head); 1101 lwkt_replymsg(&nm->base.lmsg, 0); 1102 } 1103 1104 void 1105 tcp_drain(void) 1106 { 1107 int cpu; 1108 1109 if (!do_tcpdrain) 1110 return; 1111 1112 /* 1113 * Walk the tcpbs, if existing, and flush the reassembly queue, 1114 * if there is one... 1115 * XXX: The "Net/3" implementation doesn't imply that the TCP 1116 * reassembly queue should be flushed, but in a situation 1117 * where we're really low on mbufs, this is potentially 1118 * useful. 1119 */ 1120 for (cpu = 0; cpu < ncpus2; cpu++) { 1121 struct netmsg_tcp_drain *nm; 1122 1123 if (cpu == mycpu->gd_cpuid) { 1124 tcp_drain_oncpu(&tcbinfo[cpu].pcblisthead); 1125 } else { 1126 nm = kmalloc(sizeof(struct netmsg_tcp_drain), 1127 M_LWKTMSG, M_NOWAIT); 1128 if (nm == NULL) 1129 continue; 1130 netmsg_init(&nm->base, NULL, &netisr_afree_rport, 1131 0, tcp_drain_handler); 1132 nm->nm_head = &tcbinfo[cpu].pcblisthead; 1133 lwkt_sendmsg(netisr_cpuport(cpu), &nm->base.lmsg); 1134 } 1135 } 1136 } 1137 1138 /* 1139 * Notify a tcp user of an asynchronous error; 1140 * store error as soft error, but wake up user 1141 * (for now, won't do anything until can select for soft error). 1142 * 1143 * Do not wake up user since there currently is no mechanism for 1144 * reporting soft errors (yet - a kqueue filter may be added). 1145 */ 1146 static void 1147 tcp_notify(struct inpcb *inp, int error) 1148 { 1149 struct tcpcb *tp = intotcpcb(inp); 1150 1151 /* 1152 * Ignore some errors if we are hooked up. 1153 * If connection hasn't completed, has retransmitted several times, 1154 * and receives a second error, give up now. This is better 1155 * than waiting a long time to establish a connection that 1156 * can never complete. 1157 */ 1158 if (tp->t_state == TCPS_ESTABLISHED && 1159 (error == EHOSTUNREACH || error == ENETUNREACH || 1160 error == EHOSTDOWN)) { 1161 return; 1162 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 1163 tp->t_softerror) 1164 tcp_drop(tp, error); 1165 else 1166 tp->t_softerror = error; 1167 #if 0 1168 wakeup(&so->so_timeo); 1169 sorwakeup(so); 1170 sowwakeup(so); 1171 #endif 1172 } 1173 1174 static int 1175 tcp_pcblist(SYSCTL_HANDLER_ARGS) 1176 { 1177 int error, i, n; 1178 struct inpcb *marker; 1179 struct inpcb *inp; 1180 int origcpu, ccpu; 1181 1182 error = 0; 1183 n = 0; 1184 1185 /* 1186 * The process of preparing the TCB list is too time-consuming and 1187 * resource-intensive to repeat twice on every request. 1188 */ 1189 if (req->oldptr == NULL) { 1190 for (ccpu = 0; ccpu < ncpus2; ++ccpu) 1191 n += tcbinfo[ccpu].ipi_count; 1192 req->oldidx = (n + n/8 + 10) * sizeof(struct xtcpcb); 1193 return (0); 1194 } 1195 1196 if (req->newptr != NULL) 1197 return (EPERM); 1198 1199 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO); 1200 marker->inp_flags |= INP_PLACEMARKER; 1201 1202 /* 1203 * OK, now we're committed to doing something. Run the inpcb list 1204 * for each cpu in the system and construct the output. Use a 1205 * list placemarker to deal with list changes occuring during 1206 * copyout blockages (but otherwise depend on being on the correct 1207 * cpu to avoid races). 1208 */ 1209 origcpu = mycpu->gd_cpuid; 1210 for (ccpu = 0; ccpu < ncpus2 && error == 0; ++ccpu) { 1211 caddr_t inp_ppcb; 1212 struct xtcpcb xt; 1213 1214 lwkt_migratecpu(ccpu); 1215 1216 n = tcbinfo[ccpu].ipi_count; 1217 1218 LIST_INSERT_HEAD(&tcbinfo[ccpu].pcblisthead, marker, inp_list); 1219 i = 0; 1220 while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) { 1221 /* 1222 * process a snapshot of pcbs, ignoring placemarkers 1223 * and using our own to allow SYSCTL_OUT to block. 1224 */ 1225 LIST_REMOVE(marker, inp_list); 1226 LIST_INSERT_AFTER(inp, marker, inp_list); 1227 1228 if (inp->inp_flags & INP_PLACEMARKER) 1229 continue; 1230 if (prison_xinpcb(req->td, inp)) 1231 continue; 1232 1233 xt.xt_len = sizeof xt; 1234 bcopy(inp, &xt.xt_inp, sizeof *inp); 1235 inp_ppcb = inp->inp_ppcb; 1236 if (inp_ppcb != NULL) 1237 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1238 else 1239 bzero(&xt.xt_tp, sizeof xt.xt_tp); 1240 if (inp->inp_socket) 1241 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1242 if ((error = SYSCTL_OUT(req, &xt, sizeof xt)) != 0) 1243 break; 1244 ++i; 1245 } 1246 LIST_REMOVE(marker, inp_list); 1247 if (error == 0 && i < n) { 1248 bzero(&xt, sizeof xt); 1249 xt.xt_len = sizeof xt; 1250 while (i < n) { 1251 error = SYSCTL_OUT(req, &xt, sizeof xt); 1252 if (error) 1253 break; 1254 ++i; 1255 } 1256 } 1257 } 1258 1259 /* 1260 * Make sure we are on the same cpu we were on originally, since 1261 * higher level callers expect this. Also don't pollute caches with 1262 * migrated userland data by (eventually) returning to userland 1263 * on a different cpu. 1264 */ 1265 lwkt_migratecpu(origcpu); 1266 kfree(marker, M_TEMP); 1267 return (error); 1268 } 1269 1270 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1271 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1272 1273 static int 1274 tcp_getcred(SYSCTL_HANDLER_ARGS) 1275 { 1276 struct sockaddr_in addrs[2]; 1277 struct ucred cred0, *cred = NULL; 1278 struct inpcb *inp; 1279 int cpu, origcpu, error; 1280 1281 error = priv_check(req->td, PRIV_ROOT); 1282 if (error != 0) 1283 return (error); 1284 error = SYSCTL_IN(req, addrs, sizeof addrs); 1285 if (error != 0) 1286 return (error); 1287 1288 origcpu = mycpuid; 1289 cpu = tcp_addrcpu(addrs[1].sin_addr.s_addr, addrs[1].sin_port, 1290 addrs[0].sin_addr.s_addr, addrs[0].sin_port); 1291 1292 lwkt_migratecpu(cpu); 1293 1294 inp = in_pcblookup_hash(&tcbinfo[cpu], addrs[1].sin_addr, 1295 addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1296 if (inp == NULL || inp->inp_socket == NULL) { 1297 error = ENOENT; 1298 } else if (inp->inp_socket->so_cred != NULL) { 1299 cred0 = *(inp->inp_socket->so_cred); 1300 cred = &cred0; 1301 } 1302 1303 lwkt_migratecpu(origcpu); 1304 1305 if (error) 1306 return (error); 1307 1308 return SYSCTL_OUT(req, cred, sizeof(struct ucred)); 1309 } 1310 1311 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW), 1312 0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection"); 1313 1314 #ifdef INET6 1315 static int 1316 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1317 { 1318 struct sockaddr_in6 addrs[2]; 1319 struct inpcb *inp; 1320 int error; 1321 boolean_t mapped = FALSE; 1322 1323 error = priv_check(req->td, PRIV_ROOT); 1324 if (error != 0) 1325 return (error); 1326 error = SYSCTL_IN(req, addrs, sizeof addrs); 1327 if (error != 0) 1328 return (error); 1329 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1330 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1331 mapped = TRUE; 1332 else 1333 return (EINVAL); 1334 } 1335 crit_enter(); 1336 if (mapped) { 1337 inp = in_pcblookup_hash(&tcbinfo[0], 1338 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1339 addrs[1].sin6_port, 1340 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1341 addrs[0].sin6_port, 1342 0, NULL); 1343 } else { 1344 inp = in6_pcblookup_hash(&tcbinfo[0], 1345 &addrs[1].sin6_addr, addrs[1].sin6_port, 1346 &addrs[0].sin6_addr, addrs[0].sin6_port, 1347 0, NULL); 1348 } 1349 if (inp == NULL || inp->inp_socket == NULL) { 1350 error = ENOENT; 1351 goto out; 1352 } 1353 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred)); 1354 out: 1355 crit_exit(); 1356 return (error); 1357 } 1358 1359 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW), 1360 0, 0, 1361 tcp6_getcred, "S,ucred", "Get the ucred of a TCP6 connection"); 1362 #endif 1363 1364 struct netmsg_tcp_notify { 1365 struct netmsg_base base; 1366 void (*nm_notify)(struct inpcb *, int); 1367 struct in_addr nm_faddr; 1368 int nm_arg; 1369 }; 1370 1371 static void 1372 tcp_notifyall_oncpu(netmsg_t msg) 1373 { 1374 struct netmsg_tcp_notify *nm = (struct netmsg_tcp_notify *)msg; 1375 int nextcpu; 1376 1377 in_pcbnotifyall(&tcbinfo[mycpuid], nm->nm_faddr, 1378 nm->nm_arg, nm->nm_notify); 1379 1380 nextcpu = mycpuid + 1; 1381 if (nextcpu < ncpus2) 1382 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg); 1383 else 1384 lwkt_replymsg(&nm->base.lmsg, 0); 1385 } 1386 1387 void 1388 tcp_ctlinput(netmsg_t msg) 1389 { 1390 int cmd = msg->ctlinput.nm_cmd; 1391 struct sockaddr *sa = msg->ctlinput.nm_arg; 1392 struct ip *ip = msg->ctlinput.nm_extra; 1393 struct tcphdr *th; 1394 struct in_addr faddr; 1395 struct inpcb *inp; 1396 struct tcpcb *tp; 1397 void (*notify)(struct inpcb *, int) = tcp_notify; 1398 tcp_seq icmpseq; 1399 int arg, cpu; 1400 1401 if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) { 1402 goto done; 1403 } 1404 1405 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1406 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1407 goto done; 1408 1409 arg = inetctlerrmap[cmd]; 1410 if (cmd == PRC_QUENCH) { 1411 notify = tcp_quench; 1412 } else if (icmp_may_rst && 1413 (cmd == PRC_UNREACH_ADMIN_PROHIB || 1414 cmd == PRC_UNREACH_PORT || 1415 cmd == PRC_TIMXCEED_INTRANS) && 1416 ip != NULL) { 1417 notify = tcp_drop_syn_sent; 1418 } else if (cmd == PRC_MSGSIZE) { 1419 struct icmp *icmp = (struct icmp *) 1420 ((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 1421 1422 arg = ntohs(icmp->icmp_nextmtu); 1423 notify = tcp_mtudisc; 1424 } else if (PRC_IS_REDIRECT(cmd)) { 1425 ip = NULL; 1426 notify = in_rtchange; 1427 } else if (cmd == PRC_HOSTDEAD) { 1428 ip = NULL; 1429 } 1430 1431 if (ip != NULL) { 1432 crit_enter(); 1433 th = (struct tcphdr *)((caddr_t)ip + 1434 (IP_VHL_HL(ip->ip_vhl) << 2)); 1435 cpu = tcp_addrcpu(faddr.s_addr, th->th_dport, 1436 ip->ip_src.s_addr, th->th_sport); 1437 inp = in_pcblookup_hash(&tcbinfo[cpu], faddr, th->th_dport, 1438 ip->ip_src, th->th_sport, 0, NULL); 1439 if ((inp != NULL) && (inp->inp_socket != NULL)) { 1440 icmpseq = htonl(th->th_seq); 1441 tp = intotcpcb(inp); 1442 if (SEQ_GEQ(icmpseq, tp->snd_una) && 1443 SEQ_LT(icmpseq, tp->snd_max)) 1444 (*notify)(inp, arg); 1445 } else { 1446 struct in_conninfo inc; 1447 1448 inc.inc_fport = th->th_dport; 1449 inc.inc_lport = th->th_sport; 1450 inc.inc_faddr = faddr; 1451 inc.inc_laddr = ip->ip_src; 1452 #ifdef INET6 1453 inc.inc_isipv6 = 0; 1454 #endif 1455 syncache_unreach(&inc, th); 1456 } 1457 crit_exit(); 1458 } else { 1459 struct netmsg_tcp_notify *nm; 1460 1461 KKASSERT(&curthread->td_msgport == netisr_cpuport(0)); 1462 nm = kmalloc(sizeof(*nm), M_LWKTMSG, M_INTWAIT); 1463 netmsg_init(&nm->base, NULL, &netisr_afree_rport, 1464 0, tcp_notifyall_oncpu); 1465 nm->nm_faddr = faddr; 1466 nm->nm_arg = arg; 1467 nm->nm_notify = notify; 1468 1469 lwkt_sendmsg(netisr_cpuport(0), &nm->base.lmsg); 1470 } 1471 done: 1472 lwkt_replymsg(&msg->lmsg, 0); 1473 } 1474 1475 #ifdef INET6 1476 1477 void 1478 tcp6_ctlinput(netmsg_t msg) 1479 { 1480 int cmd = msg->ctlinput.nm_cmd; 1481 struct sockaddr *sa = msg->ctlinput.nm_arg; 1482 void *d = msg->ctlinput.nm_extra; 1483 struct tcphdr th; 1484 void (*notify) (struct inpcb *, int) = tcp_notify; 1485 struct ip6_hdr *ip6; 1486 struct mbuf *m; 1487 struct ip6ctlparam *ip6cp = NULL; 1488 const struct sockaddr_in6 *sa6_src = NULL; 1489 int off; 1490 struct tcp_portonly { 1491 u_int16_t th_sport; 1492 u_int16_t th_dport; 1493 } *thp; 1494 int arg; 1495 1496 if (sa->sa_family != AF_INET6 || 1497 sa->sa_len != sizeof(struct sockaddr_in6)) { 1498 goto out; 1499 } 1500 1501 arg = 0; 1502 if (cmd == PRC_QUENCH) 1503 notify = tcp_quench; 1504 else if (cmd == PRC_MSGSIZE) { 1505 struct ip6ctlparam *ip6cp = d; 1506 struct icmp6_hdr *icmp6 = ip6cp->ip6c_icmp6; 1507 1508 arg = ntohl(icmp6->icmp6_mtu); 1509 notify = tcp_mtudisc; 1510 } else if (!PRC_IS_REDIRECT(cmd) && 1511 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) { 1512 goto out; 1513 } 1514 1515 /* if the parameter is from icmp6, decode it. */ 1516 if (d != NULL) { 1517 ip6cp = (struct ip6ctlparam *)d; 1518 m = ip6cp->ip6c_m; 1519 ip6 = ip6cp->ip6c_ip6; 1520 off = ip6cp->ip6c_off; 1521 sa6_src = ip6cp->ip6c_src; 1522 } else { 1523 m = NULL; 1524 ip6 = NULL; 1525 off = 0; /* fool gcc */ 1526 sa6_src = &sa6_any; 1527 } 1528 1529 if (ip6 != NULL) { 1530 struct in_conninfo inc; 1531 /* 1532 * XXX: We assume that when IPV6 is non NULL, 1533 * M and OFF are valid. 1534 */ 1535 1536 /* check if we can safely examine src and dst ports */ 1537 if (m->m_pkthdr.len < off + sizeof *thp) 1538 goto out; 1539 1540 bzero(&th, sizeof th); 1541 m_copydata(m, off, sizeof *thp, (caddr_t)&th); 1542 1543 in6_pcbnotify(&tcbinfo[0], sa, th.th_dport, 1544 (struct sockaddr *)ip6cp->ip6c_src, 1545 th.th_sport, cmd, arg, notify); 1546 1547 inc.inc_fport = th.th_dport; 1548 inc.inc_lport = th.th_sport; 1549 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1550 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1551 inc.inc_isipv6 = 1; 1552 syncache_unreach(&inc, &th); 1553 } else { 1554 in6_pcbnotify(&tcbinfo[0], sa, 0, 1555 (const struct sockaddr *)sa6_src, 0, cmd, arg, notify); 1556 } 1557 out: 1558 lwkt_replymsg(&msg->ctlinput.base.lmsg, 0); 1559 } 1560 1561 #endif 1562 1563 /* 1564 * Following is where TCP initial sequence number generation occurs. 1565 * 1566 * There are two places where we must use initial sequence numbers: 1567 * 1. In SYN-ACK packets. 1568 * 2. In SYN packets. 1569 * 1570 * All ISNs for SYN-ACK packets are generated by the syncache. See 1571 * tcp_syncache.c for details. 1572 * 1573 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1574 * depends on this property. In addition, these ISNs should be 1575 * unguessable so as to prevent connection hijacking. To satisfy 1576 * the requirements of this situation, the algorithm outlined in 1577 * RFC 1948 is used to generate sequence numbers. 1578 * 1579 * Implementation details: 1580 * 1581 * Time is based off the system timer, and is corrected so that it 1582 * increases by one megabyte per second. This allows for proper 1583 * recycling on high speed LANs while still leaving over an hour 1584 * before rollover. 1585 * 1586 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1587 * between seeding of isn_secret. This is normally set to zero, 1588 * as reseeding should not be necessary. 1589 * 1590 */ 1591 1592 #define ISN_BYTES_PER_SECOND 1048576 1593 1594 u_char isn_secret[32]; 1595 int isn_last_reseed; 1596 MD5_CTX isn_ctx; 1597 1598 tcp_seq 1599 tcp_new_isn(struct tcpcb *tp) 1600 { 1601 u_int32_t md5_buffer[4]; 1602 tcp_seq new_isn; 1603 1604 /* Seed if this is the first use, reseed if requested. */ 1605 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1606 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1607 < (u_int)ticks))) { 1608 read_random_unlimited(&isn_secret, sizeof isn_secret); 1609 isn_last_reseed = ticks; 1610 } 1611 1612 /* Compute the md5 hash and return the ISN. */ 1613 MD5Init(&isn_ctx); 1614 MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_fport, sizeof(u_short)); 1615 MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_lport, sizeof(u_short)); 1616 #ifdef INET6 1617 if (tp->t_inpcb->inp_vflag & INP_IPV6) { 1618 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1619 sizeof(struct in6_addr)); 1620 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1621 sizeof(struct in6_addr)); 1622 } else 1623 #endif 1624 { 1625 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1626 sizeof(struct in_addr)); 1627 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1628 sizeof(struct in_addr)); 1629 } 1630 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1631 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1632 new_isn = (tcp_seq) md5_buffer[0]; 1633 new_isn += ticks * (ISN_BYTES_PER_SECOND / hz); 1634 return (new_isn); 1635 } 1636 1637 /* 1638 * When a source quench is received, close congestion window 1639 * to one segment. We will gradually open it again as we proceed. 1640 */ 1641 void 1642 tcp_quench(struct inpcb *inp, int error) 1643 { 1644 struct tcpcb *tp = intotcpcb(inp); 1645 1646 if (tp != NULL) { 1647 tp->snd_cwnd = tp->t_maxseg; 1648 tp->snd_wacked = 0; 1649 } 1650 } 1651 1652 /* 1653 * When a specific ICMP unreachable message is received and the 1654 * connection state is SYN-SENT, drop the connection. This behavior 1655 * is controlled by the icmp_may_rst sysctl. 1656 */ 1657 void 1658 tcp_drop_syn_sent(struct inpcb *inp, int error) 1659 { 1660 struct tcpcb *tp = intotcpcb(inp); 1661 1662 if ((tp != NULL) && (tp->t_state == TCPS_SYN_SENT)) 1663 tcp_drop(tp, error); 1664 } 1665 1666 /* 1667 * When a `need fragmentation' ICMP is received, update our idea of the MSS 1668 * based on the new value in the route. Also nudge TCP to send something, 1669 * since we know the packet we just sent was dropped. 1670 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1671 */ 1672 void 1673 tcp_mtudisc(struct inpcb *inp, int mtu) 1674 { 1675 struct tcpcb *tp = intotcpcb(inp); 1676 struct rtentry *rt; 1677 struct socket *so = inp->inp_socket; 1678 int maxopd, mss; 1679 #ifdef INET6 1680 boolean_t isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0); 1681 #else 1682 const boolean_t isipv6 = FALSE; 1683 #endif 1684 1685 if (tp == NULL) 1686 return; 1687 1688 /* 1689 * If no MTU is provided in the ICMP message, use the 1690 * next lower likely value, as specified in RFC 1191. 1691 */ 1692 if (mtu == 0) { 1693 int oldmtu; 1694 1695 oldmtu = tp->t_maxopd + 1696 (isipv6 ? 1697 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1698 sizeof(struct tcpiphdr)); 1699 mtu = ip_next_mtu(oldmtu, 0); 1700 } 1701 1702 if (isipv6) 1703 rt = tcp_rtlookup6(&inp->inp_inc); 1704 else 1705 rt = tcp_rtlookup(&inp->inp_inc); 1706 if (rt != NULL) { 1707 if (rt->rt_rmx.rmx_mtu != 0 && rt->rt_rmx.rmx_mtu < mtu) 1708 mtu = rt->rt_rmx.rmx_mtu; 1709 1710 maxopd = mtu - 1711 (isipv6 ? 1712 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1713 sizeof(struct tcpiphdr)); 1714 1715 /* 1716 * XXX - The following conditional probably violates the TCP 1717 * spec. The problem is that, since we don't know the 1718 * other end's MSS, we are supposed to use a conservative 1719 * default. But, if we do that, then MTU discovery will 1720 * never actually take place, because the conservative 1721 * default is much less than the MTUs typically seen 1722 * on the Internet today. For the moment, we'll sweep 1723 * this under the carpet. 1724 * 1725 * The conservative default might not actually be a problem 1726 * if the only case this occurs is when sending an initial 1727 * SYN with options and data to a host we've never talked 1728 * to before. Then, they will reply with an MSS value which 1729 * will get recorded and the new parameters should get 1730 * recomputed. For Further Study. 1731 */ 1732 if (rt->rt_rmx.rmx_mssopt && rt->rt_rmx.rmx_mssopt < maxopd) 1733 maxopd = rt->rt_rmx.rmx_mssopt; 1734 } else 1735 maxopd = mtu - 1736 (isipv6 ? 1737 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1738 sizeof(struct tcpiphdr)); 1739 1740 if (tp->t_maxopd <= maxopd) 1741 return; 1742 tp->t_maxopd = maxopd; 1743 1744 mss = maxopd; 1745 if ((tp->t_flags & (TF_REQ_TSTMP | TF_RCVD_TSTMP | TF_NOOPT)) == 1746 (TF_REQ_TSTMP | TF_RCVD_TSTMP)) 1747 mss -= TCPOLEN_TSTAMP_APPA; 1748 1749 /* round down to multiple of MCLBYTES */ 1750 #if (MCLBYTES & (MCLBYTES - 1)) == 0 /* test if MCLBYTES power of 2 */ 1751 if (mss > MCLBYTES) 1752 mss &= ~(MCLBYTES - 1); 1753 #else 1754 if (mss > MCLBYTES) 1755 mss = (mss / MCLBYTES) * MCLBYTES; 1756 #endif 1757 1758 if (so->so_snd.ssb_hiwat < mss) 1759 mss = so->so_snd.ssb_hiwat; 1760 1761 tp->t_maxseg = mss; 1762 tp->t_rtttime = 0; 1763 tp->snd_nxt = tp->snd_una; 1764 tcp_output(tp); 1765 tcpstat.tcps_mturesent++; 1766 } 1767 1768 /* 1769 * Look-up the routing entry to the peer of this inpcb. If no route 1770 * is found and it cannot be allocated the return NULL. This routine 1771 * is called by TCP routines that access the rmx structure and by tcp_mss 1772 * to get the interface MTU. 1773 */ 1774 struct rtentry * 1775 tcp_rtlookup(struct in_conninfo *inc) 1776 { 1777 struct route *ro = &inc->inc_route; 1778 1779 if (ro->ro_rt == NULL || !(ro->ro_rt->rt_flags & RTF_UP)) { 1780 /* No route yet, so try to acquire one */ 1781 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1782 /* 1783 * unused portions of the structure MUST be zero'd 1784 * out because rtalloc() treats it as opaque data 1785 */ 1786 bzero(&ro->ro_dst, sizeof(struct sockaddr_in)); 1787 ro->ro_dst.sa_family = AF_INET; 1788 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 1789 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1790 inc->inc_faddr; 1791 rtalloc(ro); 1792 } 1793 } 1794 return (ro->ro_rt); 1795 } 1796 1797 #ifdef INET6 1798 struct rtentry * 1799 tcp_rtlookup6(struct in_conninfo *inc) 1800 { 1801 struct route_in6 *ro6 = &inc->inc6_route; 1802 1803 if (ro6->ro_rt == NULL || !(ro6->ro_rt->rt_flags & RTF_UP)) { 1804 /* No route yet, so try to acquire one */ 1805 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1806 /* 1807 * unused portions of the structure MUST be zero'd 1808 * out because rtalloc() treats it as opaque data 1809 */ 1810 bzero(&ro6->ro_dst, sizeof(struct sockaddr_in6)); 1811 ro6->ro_dst.sin6_family = AF_INET6; 1812 ro6->ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1813 ro6->ro_dst.sin6_addr = inc->inc6_faddr; 1814 rtalloc((struct route *)ro6); 1815 } 1816 } 1817 return (ro6->ro_rt); 1818 } 1819 #endif 1820 1821 #ifdef IPSEC 1822 /* compute ESP/AH header size for TCP, including outer IP header. */ 1823 size_t 1824 ipsec_hdrsiz_tcp(struct tcpcb *tp) 1825 { 1826 struct inpcb *inp; 1827 struct mbuf *m; 1828 size_t hdrsiz; 1829 struct ip *ip; 1830 struct tcphdr *th; 1831 1832 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1833 return (0); 1834 MGETHDR(m, MB_DONTWAIT, MT_DATA); 1835 if (!m) 1836 return (0); 1837 1838 #ifdef INET6 1839 if (inp->inp_vflag & INP_IPV6) { 1840 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 1841 1842 th = (struct tcphdr *)(ip6 + 1); 1843 m->m_pkthdr.len = m->m_len = 1844 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1845 tcp_fillheaders(tp, ip6, th, FALSE); 1846 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1847 } else 1848 #endif 1849 { 1850 ip = mtod(m, struct ip *); 1851 th = (struct tcphdr *)(ip + 1); 1852 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1853 tcp_fillheaders(tp, ip, th, FALSE); 1854 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1855 } 1856 1857 m_free(m); 1858 return (hdrsiz); 1859 } 1860 #endif 1861 1862 /* 1863 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1864 * 1865 * This code attempts to calculate the bandwidth-delay product as a 1866 * means of determining the optimal window size to maximize bandwidth, 1867 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1868 * routers. This code also does a fairly good job keeping RTTs in check 1869 * across slow links like modems. We implement an algorithm which is very 1870 * similar (but not meant to be) TCP/Vegas. The code operates on the 1871 * transmitter side of a TCP connection and so only effects the transmit 1872 * side of the connection. 1873 * 1874 * BACKGROUND: TCP makes no provision for the management of buffer space 1875 * at the end points or at the intermediate routers and switches. A TCP 1876 * stream, whether using NewReno or not, will eventually buffer as 1877 * many packets as it is able and the only reason this typically works is 1878 * due to the fairly small default buffers made available for a connection 1879 * (typicaly 16K or 32K). As machines use larger windows and/or window 1880 * scaling it is now fairly easy for even a single TCP connection to blow-out 1881 * all available buffer space not only on the local interface, but on 1882 * intermediate routers and switches as well. NewReno makes a misguided 1883 * attempt to 'solve' this problem by waiting for an actual failure to occur, 1884 * then backing off, then steadily increasing the window again until another 1885 * failure occurs, ad-infinitum. This results in terrible oscillation that 1886 * is only made worse as network loads increase and the idea of intentionally 1887 * blowing out network buffers is, frankly, a terrible way to manage network 1888 * resources. 1889 * 1890 * It is far better to limit the transmit window prior to the failure 1891 * condition being achieved. There are two general ways to do this: First 1892 * you can 'scan' through different transmit window sizes and locate the 1893 * point where the RTT stops increasing, indicating that you have filled the 1894 * pipe, then scan backwards until you note that RTT stops decreasing, then 1895 * repeat ad-infinitum. This method works in principle but has severe 1896 * implementation issues due to RTT variances, timer granularity, and 1897 * instability in the algorithm which can lead to many false positives and 1898 * create oscillations as well as interact badly with other TCP streams 1899 * implementing the same algorithm. 1900 * 1901 * The second method is to limit the window to the bandwidth delay product 1902 * of the link. This is the method we implement. RTT variances and our 1903 * own manipulation of the congestion window, bwnd, can potentially 1904 * destabilize the algorithm. For this reason we have to stabilize the 1905 * elements used to calculate the window. We do this by using the minimum 1906 * observed RTT, the long term average of the observed bandwidth, and 1907 * by adding two segments worth of slop. It isn't perfect but it is able 1908 * to react to changing conditions and gives us a very stable basis on 1909 * which to extend the algorithm. 1910 */ 1911 void 1912 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 1913 { 1914 u_long bw; 1915 u_long ibw; 1916 u_long bwnd; 1917 int save_ticks; 1918 int delta_ticks; 1919 1920 /* 1921 * If inflight_enable is disabled in the middle of a tcp connection, 1922 * make sure snd_bwnd is effectively disabled. 1923 */ 1924 if (!tcp_inflight_enable) { 1925 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1926 tp->snd_bandwidth = 0; 1927 return; 1928 } 1929 1930 /* 1931 * Validate the delta time. If a connection is new or has been idle 1932 * a long time we have to reset the bandwidth calculator. 1933 */ 1934 save_ticks = ticks; 1935 cpu_ccfence(); 1936 delta_ticks = save_ticks - tp->t_bw_rtttime; 1937 if (tp->t_bw_rtttime == 0 || delta_ticks < 0 || delta_ticks > hz * 10) { 1938 tp->t_bw_rtttime = save_ticks; 1939 tp->t_bw_rtseq = ack_seq; 1940 if (tp->snd_bandwidth == 0) 1941 tp->snd_bandwidth = tcp_inflight_min; 1942 return; 1943 } 1944 1945 /* 1946 * A delta of at least 1 tick is required. Waiting 2 ticks will 1947 * result in better (bw) accuracy. More than that and the ramp-up 1948 * will be too slow. 1949 */ 1950 if (delta_ticks == 0 || delta_ticks == 1) 1951 return; 1952 1953 /* 1954 * Sanity check, plus ignore pure window update acks. 1955 */ 1956 if ((int)(ack_seq - tp->t_bw_rtseq) <= 0) 1957 return; 1958 1959 /* 1960 * Figure out the bandwidth. Due to the tick granularity this 1961 * is a very rough number and it MUST be averaged over a fairly 1962 * long period of time. XXX we need to take into account a link 1963 * that is not using all available bandwidth, but for now our 1964 * slop will ramp us up if this case occurs and the bandwidth later 1965 * increases. 1966 */ 1967 ibw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / delta_ticks; 1968 tp->t_bw_rtttime = save_ticks; 1969 tp->t_bw_rtseq = ack_seq; 1970 bw = ((int64_t)tp->snd_bandwidth * 15 + ibw) >> 4; 1971 1972 tp->snd_bandwidth = bw; 1973 1974 /* 1975 * Calculate the semi-static bandwidth delay product, plus two maximal 1976 * segments. The additional slop puts us squarely in the sweet 1977 * spot and also handles the bandwidth run-up case. Without the 1978 * slop we could be locking ourselves into a lower bandwidth. 1979 * 1980 * At very high speeds the bw calculation can become overly sensitive 1981 * and error prone when delta_ticks is low (e.g. usually 1). To deal 1982 * with the problem the stab must be scaled to the bw. A stab of 50 1983 * (the default) increases the bw for the purposes of the bwnd 1984 * calculation by 5%. 1985 * 1986 * Situations Handled: 1987 * (1) Prevents over-queueing of packets on LANs, especially on 1988 * high speed LANs, allowing larger TCP buffers to be 1989 * specified, and also does a good job preventing 1990 * over-queueing of packets over choke points like modems 1991 * (at least for the transmit side). 1992 * 1993 * (2) Is able to handle changing network loads (bandwidth 1994 * drops so bwnd drops, bandwidth increases so bwnd 1995 * increases). 1996 * 1997 * (3) Theoretically should stabilize in the face of multiple 1998 * connections implementing the same algorithm (this may need 1999 * a little work). 2000 * 2001 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2002 * be adjusted with a sysctl but typically only needs to be on 2003 * very slow connections. A value no smaller then 5 should 2004 * be used, but only reduce this default if you have no other 2005 * choice. 2006 */ 2007 2008 #define USERTT ((tp->t_srtt + tp->t_rttvar) + tcp_inflight_adjrtt) 2009 bw += bw * tcp_inflight_stab / 1000; 2010 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + 2011 (int)tp->t_maxseg * 2; 2012 #undef USERTT 2013 2014 if (tcp_inflight_debug > 0) { 2015 static int ltime; 2016 if ((u_int)(save_ticks - ltime) >= hz / tcp_inflight_debug) { 2017 ltime = save_ticks; 2018 kprintf("%p ibw %ld bw %ld rttvar %d srtt %d " 2019 "bwnd %ld delta %d snd_win %ld\n", 2020 tp, ibw, bw, tp->t_rttvar, tp->t_srtt, 2021 bwnd, delta_ticks, tp->snd_wnd); 2022 } 2023 } 2024 if ((long)bwnd < tcp_inflight_min) 2025 bwnd = tcp_inflight_min; 2026 if (bwnd > tcp_inflight_max) 2027 bwnd = tcp_inflight_max; 2028 if ((long)bwnd < tp->t_maxseg * 2) 2029 bwnd = tp->t_maxseg * 2; 2030 tp->snd_bwnd = bwnd; 2031 } 2032 2033 static void 2034 tcp_rmx_iwsegs(struct tcpcb *tp, u_long *maxsegs, u_long *capsegs) 2035 { 2036 struct rtentry *rt; 2037 struct inpcb *inp = tp->t_inpcb; 2038 #ifdef INET6 2039 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE); 2040 #else 2041 const boolean_t isipv6 = FALSE; 2042 #endif 2043 2044 /* XXX */ 2045 if (tcp_iw_maxsegs < TCP_IW_MAXSEGS_DFLT) 2046 tcp_iw_maxsegs = TCP_IW_MAXSEGS_DFLT; 2047 if (tcp_iw_capsegs < TCP_IW_CAPSEGS_DFLT) 2048 tcp_iw_capsegs = TCP_IW_CAPSEGS_DFLT; 2049 2050 if (isipv6) 2051 rt = tcp_rtlookup6(&inp->inp_inc); 2052 else 2053 rt = tcp_rtlookup(&inp->inp_inc); 2054 if (rt == NULL || 2055 rt->rt_rmx.rmx_iwmaxsegs < TCP_IW_MAXSEGS_DFLT || 2056 rt->rt_rmx.rmx_iwcapsegs < TCP_IW_CAPSEGS_DFLT) { 2057 *maxsegs = tcp_iw_maxsegs; 2058 *capsegs = tcp_iw_capsegs; 2059 return; 2060 } 2061 *maxsegs = rt->rt_rmx.rmx_iwmaxsegs; 2062 *capsegs = rt->rt_rmx.rmx_iwcapsegs; 2063 } 2064 2065 u_long 2066 tcp_initial_window(struct tcpcb *tp) 2067 { 2068 if (tcp_do_rfc3390) { 2069 /* 2070 * RFC3390: 2071 * "If the SYN or SYN/ACK is lost, the initial window 2072 * used by a sender after a correctly transmitted SYN 2073 * MUST be one segment consisting of MSS bytes." 2074 * 2075 * However, we do something a little bit more aggressive 2076 * then RFC3390 here: 2077 * - Only if time spent in the SYN or SYN|ACK retransmition 2078 * >= 3 seconds, the IW is reduced. We do this mainly 2079 * because when RFC3390 is published, the initial RTO is 2080 * still 3 seconds (the threshold we test here), while 2081 * after RFC6298, the initial RTO is 1 second. This 2082 * behaviour probably still falls within the spirit of 2083 * RFC3390. 2084 * - When IW is reduced, 2*MSS is used instead of 1*MSS. 2085 * Mainly to avoid sender and receiver deadlock until 2086 * delayed ACK timer expires. And even RFC2581 does not 2087 * try to reduce IW upon SYN or SYN|ACK retransmition 2088 * timeout. 2089 * 2090 * See also: 2091 * http://tools.ietf.org/html/draft-ietf-tcpm-initcwnd-03 2092 */ 2093 if (tp->t_rxtsyn >= TCPTV_RTOBASE3) { 2094 return (2 * tp->t_maxseg); 2095 } else { 2096 u_long maxsegs, capsegs; 2097 2098 tcp_rmx_iwsegs(tp, &maxsegs, &capsegs); 2099 return min(maxsegs * tp->t_maxseg, 2100 max(2 * tp->t_maxseg, capsegs * 1460)); 2101 } 2102 } else { 2103 /* 2104 * Even RFC2581 (back to 1999) allows 2*SMSS IW. 2105 * 2106 * Mainly to avoid sender and receiver deadlock 2107 * until delayed ACK timer expires. 2108 */ 2109 return (2 * tp->t_maxseg); 2110 } 2111 } 2112 2113 #ifdef TCP_SIGNATURE 2114 /* 2115 * Compute TCP-MD5 hash of a TCP segment. (RFC2385) 2116 * 2117 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2118 * When called from tcp_input(), we can be sure that th_sum has been 2119 * zeroed out and verified already. 2120 * 2121 * Return 0 if successful, otherwise return -1. 2122 * 2123 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2124 * search with the destination IP address, and a 'magic SPI' to be 2125 * determined by the application. This is hardcoded elsewhere to 1179 2126 * right now. Another branch of this code exists which uses the SPD to 2127 * specify per-application flows but it is unstable. 2128 */ 2129 int 2130 tcpsignature_compute( 2131 struct mbuf *m, /* mbuf chain */ 2132 int len, /* length of TCP data */ 2133 int optlen, /* length of TCP options */ 2134 u_char *buf, /* storage for MD5 digest */ 2135 u_int direction) /* direction of flow */ 2136 { 2137 struct ippseudo ippseudo; 2138 MD5_CTX ctx; 2139 int doff; 2140 struct ip *ip; 2141 struct ipovly *ipovly; 2142 struct secasvar *sav; 2143 struct tcphdr *th; 2144 #ifdef INET6 2145 struct ip6_hdr *ip6; 2146 struct in6_addr in6; 2147 uint32_t plen; 2148 uint16_t nhdr; 2149 #endif /* INET6 */ 2150 u_short savecsum; 2151 2152 KASSERT(m != NULL, ("passed NULL mbuf. Game over.")); 2153 KASSERT(buf != NULL, ("passed NULL storage pointer for MD5 signature")); 2154 /* 2155 * Extract the destination from the IP header in the mbuf. 2156 */ 2157 ip = mtod(m, struct ip *); 2158 #ifdef INET6 2159 ip6 = NULL; /* Make the compiler happy. */ 2160 #endif /* INET6 */ 2161 /* 2162 * Look up an SADB entry which matches the address found in 2163 * the segment. 2164 */ 2165 switch (IP_VHL_V(ip->ip_vhl)) { 2166 case IPVERSION: 2167 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, (caddr_t)&ip->ip_dst, 2168 IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2169 break; 2170 #ifdef INET6 2171 case (IPV6_VERSION >> 4): 2172 ip6 = mtod(m, struct ip6_hdr *); 2173 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, (caddr_t)&ip6->ip6_dst, 2174 IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2175 break; 2176 #endif /* INET6 */ 2177 default: 2178 return (EINVAL); 2179 /* NOTREACHED */ 2180 break; 2181 } 2182 if (sav == NULL) { 2183 kprintf("%s: SADB lookup failed\n", __func__); 2184 return (EINVAL); 2185 } 2186 MD5Init(&ctx); 2187 2188 /* 2189 * Step 1: Update MD5 hash with IP pseudo-header. 2190 * 2191 * XXX The ippseudo header MUST be digested in network byte order, 2192 * or else we'll fail the regression test. Assume all fields we've 2193 * been doing arithmetic on have been in host byte order. 2194 * XXX One cannot depend on ipovly->ih_len here. When called from 2195 * tcp_output(), the underlying ip_len member has not yet been set. 2196 */ 2197 switch (IP_VHL_V(ip->ip_vhl)) { 2198 case IPVERSION: 2199 ipovly = (struct ipovly *)ip; 2200 ippseudo.ippseudo_src = ipovly->ih_src; 2201 ippseudo.ippseudo_dst = ipovly->ih_dst; 2202 ippseudo.ippseudo_pad = 0; 2203 ippseudo.ippseudo_p = IPPROTO_TCP; 2204 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2205 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2206 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip)); 2207 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen; 2208 break; 2209 #ifdef INET6 2210 /* 2211 * RFC 2385, 2.0 Proposal 2212 * For IPv6, the pseudo-header is as described in RFC 2460, namely the 2213 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero- 2214 * extended next header value (to form 32 bits), and 32-bit segment 2215 * length. 2216 * Note: Upper-Layer Packet Length comes before Next Header. 2217 */ 2218 case (IPV6_VERSION >> 4): 2219 in6 = ip6->ip6_src; 2220 in6_clearscope(&in6); 2221 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2222 in6 = ip6->ip6_dst; 2223 in6_clearscope(&in6); 2224 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2225 plen = htonl(len + sizeof(struct tcphdr) + optlen); 2226 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t)); 2227 nhdr = 0; 2228 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2229 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2230 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2231 nhdr = IPPROTO_TCP; 2232 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2233 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr)); 2234 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen; 2235 break; 2236 #endif /* INET6 */ 2237 default: 2238 return (EINVAL); 2239 /* NOTREACHED */ 2240 break; 2241 } 2242 /* 2243 * Step 2: Update MD5 hash with TCP header, excluding options. 2244 * The TCP checksum must be set to zero. 2245 */ 2246 savecsum = th->th_sum; 2247 th->th_sum = 0; 2248 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2249 th->th_sum = savecsum; 2250 /* 2251 * Step 3: Update MD5 hash with TCP segment data. 2252 * Use m_apply() to avoid an early m_pullup(). 2253 */ 2254 if (len > 0) 2255 m_apply(m, doff, len, tcpsignature_apply, &ctx); 2256 /* 2257 * Step 4: Update MD5 hash with shared secret. 2258 */ 2259 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2260 MD5Final(buf, &ctx); 2261 key_sa_recordxfer(sav, m); 2262 key_freesav(sav); 2263 return (0); 2264 } 2265 2266 int 2267 tcpsignature_apply(void *fstate, void *data, unsigned int len) 2268 { 2269 2270 MD5Update((MD5_CTX *)fstate, (unsigned char *)data, len); 2271 return (0); 2272 } 2273 #endif /* TCP_SIGNATURE */ 2274