1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 5 * The Regents of the University of California. All rights reserved. 6 * Copyright (c) 2007-2008,2010 7 * Swinburne University of Technology, Melbourne, Australia. 8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 9 * Copyright (c) 2010 The FreeBSD Foundation 10 * Copyright (c) 2010-2011 Juniper Networks, Inc. 11 * All rights reserved. 12 * 13 * Portions of this software were developed at the Centre for Advanced Internet 14 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 15 * James Healy and David Hayes, made possible in part by a grant from the Cisco 16 * University Research Program Fund at Community Foundation Silicon Valley. 17 * 18 * Portions of this software were developed at the Centre for Advanced 19 * Internet Architectures, Swinburne University of Technology, Melbourne, 20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 21 * 22 * Portions of this software were developed by Robert N. M. Watson under 23 * contract to Juniper Networks, Inc. 24 * 25 * Redistribution and use in source and binary forms, with or without 26 * modification, are permitted provided that the following conditions 27 * are met: 28 * 1. Redistributions of source code must retain the above copyright 29 * notice, this list of conditions and the following disclaimer. 30 * 2. Redistributions in binary form must reproduce the above copyright 31 * notice, this list of conditions and the following disclaimer in the 32 * documentation and/or other materials provided with the distribution. 33 * 3. Neither the name of the University nor the names of its contributors 34 * may be used to endorse or promote products derived from this software 35 * without specific prior written permission. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 */ 49 50 #include <sys/cdefs.h> 51 #include "opt_inet.h" 52 #include "opt_inet6.h" 53 #include "opt_ipsec.h" 54 #include "opt_rss.h" 55 56 #include <sys/param.h> 57 #include <sys/arb.h> 58 #include <sys/kernel.h> 59 #ifdef TCP_HHOOK 60 #include <sys/hhook.h> 61 #endif 62 #include <sys/malloc.h> 63 #include <sys/mbuf.h> 64 #include <sys/proc.h> /* for proc0 declaration */ 65 #include <sys/protosw.h> 66 #include <sys/qmath.h> 67 #include <sys/sdt.h> 68 #include <sys/signalvar.h> 69 #include <sys/socket.h> 70 #include <sys/socketvar.h> 71 #include <sys/sysctl.h> 72 #include <sys/syslog.h> 73 #include <sys/systm.h> 74 #include <sys/stats.h> 75 76 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 77 78 #include <vm/uma.h> 79 80 #include <net/if.h> 81 #include <net/if_var.h> 82 #include <net/route.h> 83 #include <net/rss_config.h> 84 #include <net/vnet.h> 85 86 #define TCPSTATES /* for logging */ 87 88 #include <netinet/in.h> 89 #include <netinet/in_kdtrace.h> 90 #include <netinet/in_pcb.h> 91 #include <netinet/in_rss.h> 92 #include <netinet/in_systm.h> 93 #include <netinet/ip.h> 94 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 95 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 96 #include <netinet/ip_var.h> 97 #include <netinet/ip_options.h> 98 #include <netinet/ip6.h> 99 #include <netinet/icmp6.h> 100 #include <netinet6/in6_pcb.h> 101 #include <netinet6/in6_rss.h> 102 #include <netinet6/in6_var.h> 103 #include <netinet6/ip6_var.h> 104 #include <netinet6/nd6.h> 105 #include <netinet/tcp.h> 106 #include <netinet/tcp_fsm.h> 107 #include <netinet/tcp_seq.h> 108 #include <netinet/tcp_timer.h> 109 #include <netinet/tcp_var.h> 110 #include <netinet/tcp_log_buf.h> 111 #include <netinet6/tcp6_var.h> 112 #include <netinet/tcpip.h> 113 #include <netinet/cc/cc.h> 114 #include <netinet/tcp_fastopen.h> 115 #ifdef TCPPCAP 116 #include <netinet/tcp_pcap.h> 117 #endif 118 #include <netinet/tcp_syncache.h> 119 #ifdef TCP_OFFLOAD 120 #include <netinet/tcp_offload.h> 121 #endif 122 #include <netinet/tcp_ecn.h> 123 #include <netinet/udp.h> 124 125 #include <netipsec/ipsec_support.h> 126 127 #include <machine/in_cksum.h> 128 129 #include <security/mac/mac_framework.h> 130 131 const int tcprexmtthresh = 3; 132 133 VNET_DEFINE(int, tcp_log_in_vain) = 0; 134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 135 &VNET_NAME(tcp_log_in_vain), 0, 136 "Log all incoming TCP segments to closed ports"); 137 138 VNET_DEFINE(int, blackhole) = 0; 139 #define V_blackhole VNET(blackhole) 140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 141 &VNET_NAME(blackhole), 0, 142 "Do not send RST on segments to closed ports"); 143 144 VNET_DEFINE(bool, blackhole_local) = false; 145 #define V_blackhole_local VNET(blackhole_local) 146 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET | 147 CTLFLAG_RW, &VNET_NAME(blackhole_local), false, 148 "Enforce net.inet.tcp.blackhole for locally originated packets"); 149 150 VNET_DEFINE(int, tcp_delack_enabled) = 1; 151 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 152 &VNET_NAME(tcp_delack_enabled), 0, 153 "Delay ACK to try and piggyback it onto a data packet"); 154 155 VNET_DEFINE(int, drop_synfin) = 0; 156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 157 &VNET_NAME(drop_synfin), 0, 158 "Drop TCP packets with SYN+FIN set"); 159 160 VNET_DEFINE(int, tcp_do_prr) = 1; 161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 162 &VNET_NAME(tcp_do_prr), 1, 163 "Enable Proportional Rate Reduction per RFC 6937"); 164 165 VNET_DEFINE(int, tcp_do_newcwv) = 0; 166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 167 &VNET_NAME(tcp_do_newcwv), 0, 168 "Enable New Congestion Window Validation per RFC7661"); 169 170 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 172 &VNET_NAME(tcp_do_rfc3042), 0, 173 "Enable RFC 3042 (Limited Transmit)"); 174 175 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 177 &VNET_NAME(tcp_do_rfc3390), 0, 178 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 179 180 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 181 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 182 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 183 "Slow-start flight size (initial congestion window) in number of segments"); 184 185 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 186 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 187 &VNET_NAME(tcp_do_rfc3465), 0, 188 "Enable RFC 3465 (Appropriate Byte Counting)"); 189 190 VNET_DEFINE(int, tcp_abc_l_var) = 2; 191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 192 &VNET_NAME(tcp_abc_l_var), 2, 193 "Cap the max cwnd increment during slow-start to this number of segments"); 194 195 VNET_DEFINE(int, tcp_insecure_syn) = 0; 196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 197 &VNET_NAME(tcp_insecure_syn), 0, 198 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 199 200 VNET_DEFINE(int, tcp_insecure_rst) = 0; 201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 202 &VNET_NAME(tcp_insecure_rst), 0, 203 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 204 205 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 206 #define V_tcp_recvspace VNET(tcp_recvspace) 207 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 208 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 209 210 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 212 &VNET_NAME(tcp_do_autorcvbuf), 0, 213 "Enable automatic receive buffer sizing"); 214 215 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 217 &VNET_NAME(tcp_autorcvbuf_max), 0, 218 "Max size of automatic receive buffer"); 219 220 VNET_DEFINE(struct inpcbinfo, tcbinfo); 221 222 /* 223 * TCP statistics are stored in an array of counter(9)s, which size matches 224 * size of struct tcpstat. TCP running connection count is a regular array. 225 */ 226 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 227 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 228 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 229 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 230 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 231 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 232 "TCP connection counts by TCP state"); 233 234 /* 235 * Kernel module interface for updating tcpstat. The first argument is an index 236 * into tcpstat treated as an array. 237 */ 238 void 239 kmod_tcpstat_add(int statnum, int val) 240 { 241 242 counter_u64_add(VNET(tcpstat)[statnum], val); 243 } 244 245 /* 246 * Make sure that we only start a SACK loss recovery when 247 * receiving a duplicate ACK with a SACK block, and also 248 * complete SACK loss recovery in case the other end 249 * reneges. 250 */ 251 static bool inline 252 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to) 253 { 254 return ((tp->t_flags & TF_SACK_PERMIT) && 255 ((to->to_flags & TOF_SACK) || 256 (!TAILQ_EMPTY(&tp->snd_holes)))); 257 } 258 259 #ifdef TCP_HHOOK 260 /* 261 * Wrapper for the TCP established input helper hook. 262 */ 263 void 264 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 265 { 266 struct tcp_hhook_data hhook_data; 267 268 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 269 hhook_data.tp = tp; 270 hhook_data.th = th; 271 hhook_data.to = to; 272 273 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 274 &tp->t_osd); 275 } 276 } 277 #endif 278 279 /* 280 * CC wrapper hook functions 281 */ 282 void 283 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 284 uint16_t type) 285 { 286 #ifdef STATS 287 int32_t gput; 288 #endif 289 290 INP_WLOCK_ASSERT(tptoinpcb(tp)); 291 292 tp->t_ccv.nsegs = nsegs; 293 tp->t_ccv.bytes_this_ack = BYTES_THIS_ACK(tp, th); 294 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 295 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 296 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 297 tp->t_ccv.flags |= CCF_CWND_LIMITED; 298 else 299 tp->t_ccv.flags &= ~CCF_CWND_LIMITED; 300 301 if (type == CC_ACK) { 302 #ifdef STATS 303 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 304 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 305 if (!IN_RECOVERY(tp->t_flags)) 306 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 307 tp->t_ccv.bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 308 if ((tp->t_flags & TF_GPUTINPROG) && 309 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 310 /* 311 * Compute goodput in bits per millisecond. 312 */ 313 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) / 314 max(1, tcp_ts_getticks() - tp->gput_ts); 315 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 316 gput); 317 /* 318 * XXXLAS: This is a temporary hack, and should be 319 * chained off VOI_TCP_GPUT when stats(9) grows an API 320 * to deal with chained VOIs. 321 */ 322 if (tp->t_stats_gput_prev > 0) 323 stats_voi_update_abs_s32(tp->t_stats, 324 VOI_TCP_GPUT_ND, 325 ((gput - tp->t_stats_gput_prev) * 100) / 326 tp->t_stats_gput_prev); 327 tp->t_flags &= ~TF_GPUTINPROG; 328 tp->t_stats_gput_prev = gput; 329 } 330 #endif /* STATS */ 331 if (tp->snd_cwnd > tp->snd_ssthresh) { 332 tp->t_bytes_acked += tp->t_ccv.bytes_this_ack; 333 if (tp->t_bytes_acked >= tp->snd_cwnd) { 334 tp->t_bytes_acked -= tp->snd_cwnd; 335 tp->t_ccv.flags |= CCF_ABC_SENTAWND; 336 } 337 } else { 338 tp->t_ccv.flags &= ~CCF_ABC_SENTAWND; 339 tp->t_bytes_acked = 0; 340 } 341 } 342 343 if (CC_ALGO(tp)->ack_received != NULL) { 344 /* XXXLAS: Find a way to live without this */ 345 tp->t_ccv.curack = th->th_ack; 346 CC_ALGO(tp)->ack_received(&tp->t_ccv, type); 347 } 348 #ifdef STATS 349 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 350 #endif 351 } 352 353 void 354 cc_conn_init(struct tcpcb *tp) 355 { 356 struct hc_metrics_lite metrics; 357 struct inpcb *inp = tptoinpcb(tp); 358 u_int maxseg; 359 int rtt; 360 361 INP_WLOCK_ASSERT(inp); 362 363 tcp_hc_get(&inp->inp_inc, &metrics); 364 maxseg = tcp_maxseg(tp); 365 366 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 367 tp->t_srtt = rtt; 368 TCPSTAT_INC(tcps_usedrtt); 369 if (metrics.rmx_rttvar) { 370 tp->t_rttvar = metrics.rmx_rttvar; 371 TCPSTAT_INC(tcps_usedrttvar); 372 } else { 373 /* default variation is +- 1 rtt */ 374 tp->t_rttvar = 375 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 376 } 377 TCPT_RANGESET(tp->t_rxtcur, 378 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 379 tp->t_rttmin, TCPTV_REXMTMAX); 380 } 381 if (metrics.rmx_ssthresh) { 382 /* 383 * There's some sort of gateway or interface 384 * buffer limit on the path. Use this to set 385 * the slow start threshold, but set the 386 * threshold to no less than 2*mss. 387 */ 388 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 389 TCPSTAT_INC(tcps_usedssthresh); 390 } 391 392 /* 393 * Set the initial slow-start flight size. 394 * 395 * If a SYN or SYN/ACK was lost and retransmitted, we have to 396 * reduce the initial CWND to one segment as congestion is likely 397 * requiring us to be cautious. 398 */ 399 if (tp->snd_cwnd == 1) 400 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 401 else 402 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 403 404 if (CC_ALGO(tp)->conn_init != NULL) 405 CC_ALGO(tp)->conn_init(&tp->t_ccv); 406 } 407 408 void inline 409 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 410 { 411 INP_WLOCK_ASSERT(tptoinpcb(tp)); 412 413 #ifdef STATS 414 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 415 #endif 416 417 switch(type) { 418 case CC_NDUPACK: 419 if (!IN_FASTRECOVERY(tp->t_flags)) { 420 tp->snd_recover = tp->snd_max; 421 if (tp->t_flags2 & TF2_ECN_PERMIT) 422 tp->t_flags2 |= TF2_ECN_SND_CWR; 423 } 424 break; 425 case CC_ECN: 426 if (!IN_CONGRECOVERY(tp->t_flags) || 427 /* 428 * Allow ECN reaction on ACK to CWR, if 429 * that data segment was also CE marked. 430 */ 431 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 432 EXIT_CONGRECOVERY(tp->t_flags); 433 TCPSTAT_INC(tcps_ecn_rcwnd); 434 tp->snd_recover = tp->snd_max + 1; 435 if (tp->t_flags2 & TF2_ECN_PERMIT) 436 tp->t_flags2 |= TF2_ECN_SND_CWR; 437 } 438 break; 439 case CC_RTO: 440 tp->t_dupacks = 0; 441 tp->t_bytes_acked = 0; 442 EXIT_RECOVERY(tp->t_flags); 443 if (tp->t_flags2 & TF2_ECN_PERMIT) 444 tp->t_flags2 |= TF2_ECN_SND_CWR; 445 break; 446 case CC_RTO_ERR: 447 TCPSTAT_INC(tcps_sndrexmitbad); 448 /* RTO was unnecessary, so reset everything. */ 449 tp->snd_cwnd = tp->snd_cwnd_prev; 450 tp->snd_ssthresh = tp->snd_ssthresh_prev; 451 tp->snd_recover = tp->snd_recover_prev; 452 if (tp->t_flags & TF_WASFRECOVERY) 453 ENTER_FASTRECOVERY(tp->t_flags); 454 if (tp->t_flags & TF_WASCRECOVERY) 455 ENTER_CONGRECOVERY(tp->t_flags); 456 tp->snd_nxt = tp->snd_max; 457 tp->t_flags &= ~TF_PREVVALID; 458 tp->t_badrxtwin = 0; 459 break; 460 } 461 462 if (CC_ALGO(tp)->cong_signal != NULL) { 463 if (th != NULL) 464 tp->t_ccv.curack = th->th_ack; 465 CC_ALGO(tp)->cong_signal(&tp->t_ccv, type); 466 } 467 } 468 469 void inline 470 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 471 { 472 INP_WLOCK_ASSERT(tptoinpcb(tp)); 473 474 /* XXXLAS: KASSERT that we're in recovery? */ 475 476 if (CC_ALGO(tp)->post_recovery != NULL) { 477 tp->t_ccv.curack = th->th_ack; 478 CC_ALGO(tp)->post_recovery(&tp->t_ccv); 479 } 480 /* XXXLAS: EXIT_RECOVERY ? */ 481 tp->t_bytes_acked = 0; 482 tp->sackhint.delivered_data = 0; 483 tp->sackhint.prr_out = 0; 484 } 485 486 /* 487 * Indicate whether this ack should be delayed. We can delay the ack if 488 * following conditions are met: 489 * - There is no delayed ack timer in progress. 490 * - Our last ack wasn't a 0-sized window. We never want to delay 491 * the ack that opens up a 0-sized window. 492 * - LRO wasn't used for this segment. We make sure by checking that the 493 * segment size is not larger than the MSS. 494 */ 495 #define DELAY_ACK(tp, tlen) \ 496 ((!tcp_timer_active(tp, TT_DELACK) && \ 497 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 498 (tlen <= tp->t_maxseg) && \ 499 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 500 501 void inline 502 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) 503 { 504 INP_WLOCK_ASSERT(tptoinpcb(tp)); 505 506 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 507 switch (iptos & IPTOS_ECN_MASK) { 508 case IPTOS_ECN_CE: 509 tp->t_ccv.flags |= CCF_IPHDR_CE; 510 break; 511 case IPTOS_ECN_ECT0: 512 /* FALLTHROUGH */ 513 case IPTOS_ECN_ECT1: 514 /* FALLTHROUGH */ 515 case IPTOS_ECN_NOTECT: 516 tp->t_ccv.flags &= ~CCF_IPHDR_CE; 517 break; 518 } 519 520 if (flags & TH_CWR) 521 tp->t_ccv.flags |= CCF_TCPHDR_CWR; 522 else 523 tp->t_ccv.flags &= ~CCF_TCPHDR_CWR; 524 525 CC_ALGO(tp)->ecnpkt_handler(&tp->t_ccv); 526 527 if (tp->t_ccv.flags & CCF_ACKNOW) { 528 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 529 tp->t_flags |= TF_ACKNOW; 530 } 531 } 532 } 533 534 void inline 535 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 536 { 537 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos); 538 } 539 540 /* 541 * TCP input handling is split into multiple parts: 542 * tcp6_input is a thin wrapper around tcp_input for the extended 543 * ip6_protox[] call format in ip6_input 544 * tcp_input handles primary segment validation, inpcb lookup and 545 * SYN processing on listen sockets 546 * tcp_do_segment processes the ACK and text of the segment for 547 * establishing, established and closing connections 548 */ 549 #ifdef INET6 550 int 551 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 552 { 553 struct mbuf *m; 554 struct in6_ifaddr *ia6; 555 struct ip6_hdr *ip6; 556 557 m = *mp; 558 if (m->m_len < *offp + sizeof(struct tcphdr)) { 559 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 560 if (m == NULL) { 561 *mp = m; 562 TCPSTAT_INC(tcps_rcvshort); 563 return (IPPROTO_DONE); 564 } 565 } 566 567 /* 568 * draft-itojun-ipv6-tcp-to-anycast 569 * better place to put this in? 570 */ 571 ip6 = mtod(m, struct ip6_hdr *); 572 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 573 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 574 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 575 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 576 *mp = NULL; 577 return (IPPROTO_DONE); 578 } 579 580 *mp = m; 581 return (tcp_input_with_port(mp, offp, proto, port)); 582 } 583 584 int 585 tcp6_input(struct mbuf **mp, int *offp, int proto) 586 { 587 588 return(tcp6_input_with_port(mp, offp, proto, 0)); 589 } 590 #endif /* INET6 */ 591 592 int 593 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 594 { 595 struct mbuf *m = *mp; 596 struct tcphdr *th = NULL; 597 struct ip *ip = NULL; 598 struct inpcb *inp = NULL; 599 struct tcpcb *tp = NULL; 600 struct socket *so = NULL; 601 u_char *optp = NULL; 602 int off0; 603 int optlen = 0; 604 #ifdef INET 605 int len; 606 uint8_t ipttl; 607 #endif 608 int tlen = 0, off; 609 int drop_hdrlen; 610 int thflags; 611 int rstreason = 0; /* For badport_bandlim accounting purposes */ 612 int lookupflag; 613 uint8_t iptos; 614 struct m_tag *fwd_tag = NULL; 615 #ifdef INET6 616 struct ip6_hdr *ip6 = NULL; 617 int isipv6; 618 #else 619 const void *ip6 = NULL; 620 #endif /* INET6 */ 621 struct tcpopt to; /* options in this segment */ 622 char *s = NULL; /* address and port logging */ 623 624 NET_EPOCH_ASSERT(); 625 626 #ifdef INET6 627 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 628 #endif 629 630 off0 = *offp; 631 m = *mp; 632 *mp = NULL; 633 to.to_flags = 0; 634 TCPSTAT_INC(tcps_rcvtotal); 635 636 m->m_pkthdr.tcp_tun_port = port; 637 #ifdef INET6 638 if (isipv6) { 639 ip6 = mtod(m, struct ip6_hdr *); 640 th = (struct tcphdr *)((caddr_t)ip6 + off0); 641 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 642 if (port) 643 goto skip6_csum; 644 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 645 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 646 th->th_sum = m->m_pkthdr.csum_data; 647 else 648 th->th_sum = in6_cksum_pseudo(ip6, tlen, 649 IPPROTO_TCP, m->m_pkthdr.csum_data); 650 th->th_sum ^= 0xffff; 651 } else 652 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 653 if (th->th_sum) { 654 TCPSTAT_INC(tcps_rcvbadsum); 655 goto drop; 656 } 657 skip6_csum: 658 /* 659 * Be proactive about unspecified IPv6 address in source. 660 * As we use all-zero to indicate unbounded/unconnected pcb, 661 * unspecified IPv6 address can be used to confuse us. 662 * 663 * Note that packets with unspecified IPv6 destination is 664 * already dropped in ip6_input. 665 */ 666 KASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_dst), 667 ("%s: unspecified destination v6 address", __func__)); 668 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 669 IP6STAT_INC(ip6s_badscope); /* XXX */ 670 goto drop; 671 } 672 iptos = IPV6_TRAFFIC_CLASS(ip6); 673 } 674 #endif 675 #if defined(INET) && defined(INET6) 676 else 677 #endif 678 #ifdef INET 679 { 680 /* 681 * Get IP and TCP header together in first mbuf. 682 * Note: IP leaves IP header in first mbuf. 683 */ 684 if (off0 > sizeof (struct ip)) { 685 ip_stripoptions(m); 686 off0 = sizeof(struct ip); 687 } 688 if (m->m_len < sizeof (struct tcpiphdr)) { 689 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 690 == NULL) { 691 TCPSTAT_INC(tcps_rcvshort); 692 return (IPPROTO_DONE); 693 } 694 } 695 ip = mtod(m, struct ip *); 696 th = (struct tcphdr *)((caddr_t)ip + off0); 697 tlen = ntohs(ip->ip_len) - off0; 698 699 iptos = ip->ip_tos; 700 if (port) 701 goto skip_csum; 702 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 703 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 704 th->th_sum = m->m_pkthdr.csum_data; 705 else 706 th->th_sum = in_pseudo(ip->ip_src.s_addr, 707 ip->ip_dst.s_addr, 708 htonl(m->m_pkthdr.csum_data + tlen + 709 IPPROTO_TCP)); 710 th->th_sum ^= 0xffff; 711 } else { 712 struct ipovly *ipov = (struct ipovly *)ip; 713 714 /* 715 * Checksum extended TCP header and data. 716 */ 717 len = off0 + tlen; 718 ipttl = ip->ip_ttl; 719 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 720 ipov->ih_len = htons(tlen); 721 th->th_sum = in_cksum(m, len); 722 /* Reset length for SDT probes. */ 723 ip->ip_len = htons(len); 724 /* Reset TOS bits */ 725 ip->ip_tos = iptos; 726 /* Re-initialization for later version check */ 727 ip->ip_ttl = ipttl; 728 ip->ip_v = IPVERSION; 729 ip->ip_hl = off0 >> 2; 730 } 731 skip_csum: 732 if (th->th_sum && (port == 0)) { 733 TCPSTAT_INC(tcps_rcvbadsum); 734 goto drop; 735 } 736 KASSERT(ip->ip_dst.s_addr != INADDR_ANY, 737 ("%s: unspecified destination v4 address", __func__)); 738 if (__predict_false(ip->ip_src.s_addr == INADDR_ANY)) { 739 IPSTAT_INC(ips_badaddr); 740 goto drop; 741 } 742 } 743 #endif /* INET */ 744 745 /* 746 * Check that TCP offset makes sense, 747 * pull out TCP options and adjust length. XXX 748 */ 749 off = th->th_off << 2; 750 if (off < sizeof (struct tcphdr) || off > tlen) { 751 TCPSTAT_INC(tcps_rcvbadoff); 752 goto drop; 753 } 754 tlen -= off; /* tlen is used instead of ti->ti_len */ 755 if (off > sizeof (struct tcphdr)) { 756 #ifdef INET6 757 if (isipv6) { 758 if (m->m_len < off0 + off) { 759 m = m_pullup(m, off0 + off); 760 if (m == NULL) { 761 TCPSTAT_INC(tcps_rcvshort); 762 return (IPPROTO_DONE); 763 } 764 } 765 ip6 = mtod(m, struct ip6_hdr *); 766 th = (struct tcphdr *)((caddr_t)ip6 + off0); 767 } 768 #endif 769 #if defined(INET) && defined(INET6) 770 else 771 #endif 772 #ifdef INET 773 { 774 if (m->m_len < sizeof(struct ip) + off) { 775 if ((m = m_pullup(m, sizeof (struct ip) + off)) 776 == NULL) { 777 TCPSTAT_INC(tcps_rcvshort); 778 return (IPPROTO_DONE); 779 } 780 ip = mtod(m, struct ip *); 781 th = (struct tcphdr *)((caddr_t)ip + off0); 782 } 783 } 784 #endif 785 optlen = off - sizeof (struct tcphdr); 786 optp = (u_char *)(th + 1); 787 } 788 thflags = tcp_get_flags(th); 789 790 /* 791 * Convert TCP protocol specific fields to host format. 792 */ 793 tcp_fields_to_host(th); 794 795 /* 796 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 797 */ 798 drop_hdrlen = off0 + off; 799 800 /* 801 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 802 */ 803 if ( 804 #ifdef INET6 805 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 806 #ifdef INET 807 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 808 #endif 809 #endif 810 #if defined(INET) && !defined(INET6) 811 (m->m_flags & M_IP_NEXTHOP) 812 #endif 813 ) 814 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 815 816 /* 817 * For initial SYN packets we don't need write lock on matching 818 * PCB, be it a listening one or a synchronized one. The packet 819 * shall not modify its state. 820 */ 821 lookupflag = INPLOOKUP_WILDCARD | 822 ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ? 823 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB); 824 findpcb: 825 #ifdef INET6 826 if (isipv6 && fwd_tag != NULL) { 827 struct sockaddr_in6 *next_hop6; 828 829 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 830 /* 831 * Transparently forwarded. Pretend to be the destination. 832 * Already got one like this? 833 */ 834 inp = in6_pcblookup_mbuf(&V_tcbinfo, 835 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 836 lookupflag & ~INPLOOKUP_WILDCARD, m->m_pkthdr.rcvif, m); 837 if (!inp) { 838 /* 839 * It's new. Try to find the ambushing socket. 840 * Because we've rewritten the destination address, 841 * any hardware-generated hash is ignored. 842 */ 843 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 844 th->th_sport, &next_hop6->sin6_addr, 845 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 846 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 847 } 848 } else if (isipv6) { 849 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 850 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag, 851 m->m_pkthdr.rcvif, m); 852 } 853 #endif /* INET6 */ 854 #if defined(INET6) && defined(INET) 855 else 856 #endif 857 #ifdef INET 858 if (fwd_tag != NULL) { 859 struct sockaddr_in *next_hop; 860 861 next_hop = (struct sockaddr_in *)(fwd_tag+1); 862 /* 863 * Transparently forwarded. Pretend to be the destination. 864 * already got one like this? 865 */ 866 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 867 ip->ip_dst, th->th_dport, lookupflag & ~INPLOOKUP_WILDCARD, 868 m->m_pkthdr.rcvif, m); 869 if (!inp) { 870 /* 871 * It's new. Try to find the ambushing socket. 872 * Because we've rewritten the destination address, 873 * any hardware-generated hash is ignored. 874 */ 875 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 876 th->th_sport, next_hop->sin_addr, 877 next_hop->sin_port ? ntohs(next_hop->sin_port) : 878 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 879 } 880 } else 881 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 882 th->th_sport, ip->ip_dst, th->th_dport, lookupflag, 883 m->m_pkthdr.rcvif, m); 884 #endif /* INET */ 885 886 /* 887 * If the INPCB does not exist then all data in the incoming 888 * segment is discarded and an appropriate RST is sent back. 889 * XXX MRT Send RST using which routing table? 890 */ 891 if (inp == NULL) { 892 if (rstreason != 0) { 893 /* We came here after second (safety) lookup. */ 894 MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0); 895 goto dropwithreset; 896 } 897 /* 898 * Log communication attempts to ports that are not 899 * in use. 900 */ 901 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 902 V_tcp_log_in_vain == 2) { 903 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 904 log(LOG_INFO, "%s; %s: Connection attempt " 905 "to closed port\n", s, __func__); 906 } 907 /* 908 * When blackholing do not respond with a RST but 909 * completely ignore the segment and drop it. 910 */ 911 if (((V_blackhole == 1 && (thflags & TH_SYN)) || 912 V_blackhole == 2) && (V_blackhole_local || ( 913 #ifdef INET6 914 isipv6 ? !in6_localaddr(&ip6->ip6_src) : 915 #endif 916 #ifdef INET 917 !in_localip(ip->ip_src) 918 #else 919 true 920 #endif 921 ))) 922 goto dropunlock; 923 924 rstreason = BANDLIM_RST_CLOSEDPORT; 925 goto dropwithreset; 926 } 927 INP_LOCK_ASSERT(inp); 928 929 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 930 !SOLISTENING(inp->inp_socket)) { 931 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) { 932 inp->inp_flowid = m->m_pkthdr.flowid; 933 inp->inp_flowtype = M_HASHTYPE_GET(m); 934 #ifdef RSS 935 } else { 936 /* assign flowid by software RSS hash */ 937 #ifdef INET6 938 if (isipv6) { 939 rss_proto_software_hash_v6(&inp->in6p_faddr, 940 &inp->in6p_laddr, 941 inp->inp_fport, 942 inp->inp_lport, 943 IPPROTO_TCP, 944 &inp->inp_flowid, 945 &inp->inp_flowtype); 946 } else 947 #endif /* INET6 */ 948 { 949 rss_proto_software_hash_v4(inp->inp_faddr, 950 inp->inp_laddr, 951 inp->inp_fport, 952 inp->inp_lport, 953 IPPROTO_TCP, 954 &inp->inp_flowid, 955 &inp->inp_flowtype); 956 } 957 #endif /* RSS */ 958 } 959 } 960 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 961 #ifdef INET6 962 if (isipv6 && IPSEC_ENABLED(ipv6) && 963 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 964 goto dropunlock; 965 } 966 #ifdef INET 967 else 968 #endif 969 #endif /* INET6 */ 970 #ifdef INET 971 if (IPSEC_ENABLED(ipv4) && 972 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 973 goto dropunlock; 974 } 975 #endif /* INET */ 976 #endif /* IPSEC */ 977 978 /* 979 * Check the minimum TTL for socket. 980 */ 981 if (inp->inp_ip_minttl != 0) { 982 #ifdef INET6 983 if (isipv6) { 984 if (inp->inp_ip_minttl > ip6->ip6_hlim) 985 goto dropunlock; 986 } else 987 #endif 988 if (inp->inp_ip_minttl > ip->ip_ttl) 989 goto dropunlock; 990 } 991 992 tp = intotcpcb(inp); 993 switch (tp->t_state) { 994 case TCPS_TIME_WAIT: 995 /* 996 * A previous connection in TIMEWAIT state is supposed to catch 997 * stray or duplicate segments arriving late. If this segment 998 * was a legitimate new connection attempt, the old INPCB gets 999 * removed and we can try again to find a listening socket. 1000 */ 1001 tcp_dooptions(&to, optp, optlen, 1002 (thflags & TH_SYN) ? TO_SYN : 0); 1003 /* 1004 * tcp_twcheck unlocks the inp always, and frees the m if fails. 1005 */ 1006 if (tcp_twcheck(inp, &to, th, m, tlen)) 1007 goto findpcb; 1008 return (IPPROTO_DONE); 1009 case TCPS_CLOSED: 1010 /* 1011 * The TCPCB may no longer exist if the connection is winding 1012 * down or it is in the CLOSED state. Either way we drop the 1013 * segment and send an appropriate response. 1014 */ 1015 rstreason = BANDLIM_RST_CLOSEDPORT; 1016 goto dropwithreset; 1017 } 1018 1019 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 1020 rstreason = BANDLIM_RST_CLOSEDPORT; 1021 goto dropwithreset; 1022 } 1023 1024 #ifdef TCP_OFFLOAD 1025 if (tp->t_flags & TF_TOE) { 1026 tcp_offload_input(tp, m); 1027 m = NULL; /* consumed by the TOE driver */ 1028 goto dropunlock; 1029 } 1030 #endif 1031 1032 #ifdef MAC 1033 if (mac_inpcb_check_deliver(inp, m)) 1034 goto dropunlock; 1035 #endif 1036 so = inp->inp_socket; 1037 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1038 /* 1039 * When the socket is accepting connections (the INPCB is in LISTEN 1040 * state) we look into the SYN cache if this is a new connection 1041 * attempt or the completion of a previous one. 1042 */ 1043 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), 1044 ("%s: so accepting but tp %p not listening", __func__, tp)); 1045 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { 1046 struct in_conninfo inc; 1047 1048 bzero(&inc, sizeof(inc)); 1049 #ifdef INET6 1050 if (isipv6) { 1051 inc.inc_flags |= INC_ISIPV6; 1052 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1053 inc.inc_flags |= INC_IPV6MINMTU; 1054 inc.inc6_faddr = ip6->ip6_src; 1055 inc.inc6_laddr = ip6->ip6_dst; 1056 } else 1057 #endif 1058 { 1059 inc.inc_faddr = ip->ip_src; 1060 inc.inc_laddr = ip->ip_dst; 1061 } 1062 inc.inc_fport = th->th_sport; 1063 inc.inc_lport = th->th_dport; 1064 inc.inc_fibnum = so->so_fibnum; 1065 1066 /* 1067 * Check for an existing connection attempt in syncache if 1068 * the flag is only ACK. A successful lookup creates a new 1069 * socket appended to the listen queue in SYN_RECEIVED state. 1070 */ 1071 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1072 /* 1073 * Parse the TCP options here because 1074 * syncookies need access to the reflected 1075 * timestamp. 1076 */ 1077 tcp_dooptions(&to, optp, optlen, 0); 1078 /* 1079 * NB: syncache_expand() doesn't unlock inp. 1080 */ 1081 rstreason = syncache_expand(&inc, &to, th, &so, m, port); 1082 if (rstreason < 0) { 1083 /* 1084 * A failing TCP MD5 signature comparison 1085 * must result in the segment being dropped 1086 * and must not produce any response back 1087 * to the sender. 1088 */ 1089 goto dropunlock; 1090 } else if (rstreason == 0) { 1091 /* 1092 * No syncache entry, or ACK was not for our 1093 * SYN/ACK. Do our protection against double 1094 * ACK. If peer sent us 2 ACKs, then for the 1095 * first one syncache_expand() successfully 1096 * converted syncache entry into a socket, 1097 * while we were waiting on the inpcb lock. We 1098 * don't want to sent RST for the second ACK, 1099 * so we perform second lookup without wildcard 1100 * match, hoping to find the new socket. If 1101 * the ACK is stray indeed, rstreason would 1102 * hint the above code that the lookup was a 1103 * second attempt. 1104 * 1105 * NB: syncache did its own logging 1106 * of the failure cause. 1107 */ 1108 INP_WUNLOCK(inp); 1109 rstreason = BANDLIM_RST_OPENPORT; 1110 lookupflag &= ~INPLOOKUP_WILDCARD; 1111 goto findpcb; 1112 } 1113 tfo_socket_result: 1114 if (so == NULL) { 1115 /* 1116 * We completed the 3-way handshake 1117 * but could not allocate a socket 1118 * either due to memory shortage, 1119 * listen queue length limits or 1120 * global socket limits. Send RST 1121 * or wait and have the remote end 1122 * retransmit the ACK for another 1123 * try. 1124 */ 1125 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1126 log(LOG_DEBUG, "%s; %s: Listen socket: " 1127 "Socket allocation failed due to " 1128 "limits or memory shortage, %s\n", 1129 s, __func__, 1130 V_tcp_sc_rst_sock_fail ? 1131 "sending RST" : "try again"); 1132 if (V_tcp_sc_rst_sock_fail) { 1133 rstreason = BANDLIM_UNLIMITED; 1134 goto dropwithreset; 1135 } else 1136 goto dropunlock; 1137 } 1138 /* 1139 * Socket is created in state SYN_RECEIVED. 1140 * Unlock the listen socket, lock the newly 1141 * created socket and update the tp variable. 1142 * If we came here via jump to tfo_socket_result, 1143 * then listening socket is read-locked. 1144 */ 1145 INP_UNLOCK(inp); /* listen socket */ 1146 inp = sotoinpcb(so); 1147 /* 1148 * New connection inpcb is already locked by 1149 * syncache_expand(). 1150 */ 1151 INP_WLOCK_ASSERT(inp); 1152 tp = intotcpcb(inp); 1153 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1154 ("%s: ", __func__)); 1155 /* 1156 * Process the segment and the data it 1157 * contains. tcp_do_segment() consumes 1158 * the mbuf chain and unlocks the inpcb. 1159 */ 1160 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1161 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, 1162 tlen, iptos); 1163 return (IPPROTO_DONE); 1164 } 1165 /* 1166 * Segment flag validation for new connection attempts: 1167 * 1168 * Our (SYN|ACK) response was rejected. 1169 * Check with syncache and remove entry to prevent 1170 * retransmits. 1171 * 1172 * NB: syncache_chkrst does its own logging of failure 1173 * causes. 1174 */ 1175 if (thflags & TH_RST) { 1176 syncache_chkrst(&inc, th, m, port); 1177 goto dropunlock; 1178 } 1179 /* 1180 * We can't do anything without SYN. 1181 */ 1182 if ((thflags & TH_SYN) == 0) { 1183 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1184 log(LOG_DEBUG, "%s; %s: Listen socket: " 1185 "SYN is missing, segment ignored\n", 1186 s, __func__); 1187 TCPSTAT_INC(tcps_badsyn); 1188 goto dropunlock; 1189 } 1190 /* 1191 * (SYN|ACK) is bogus on a listen socket. 1192 */ 1193 if (thflags & TH_ACK) { 1194 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1195 log(LOG_DEBUG, "%s; %s: Listen socket: " 1196 "SYN|ACK invalid, segment rejected\n", 1197 s, __func__); 1198 syncache_badack(&inc, port); /* XXX: Not needed! */ 1199 TCPSTAT_INC(tcps_badsyn); 1200 rstreason = BANDLIM_RST_OPENPORT; 1201 goto dropwithreset; 1202 } 1203 /* 1204 * If the drop_synfin option is enabled, drop all 1205 * segments with both the SYN and FIN bits set. 1206 * This prevents e.g. nmap from identifying the 1207 * TCP/IP stack. 1208 * XXX: Poor reasoning. nmap has other methods 1209 * and is constantly refining its stack detection 1210 * strategies. 1211 * XXX: This is a violation of the TCP specification 1212 * and was used by RFC1644. 1213 */ 1214 if ((thflags & TH_FIN) && V_drop_synfin) { 1215 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1216 log(LOG_DEBUG, "%s; %s: Listen socket: " 1217 "SYN|FIN segment ignored (based on " 1218 "sysctl setting)\n", s, __func__); 1219 TCPSTAT_INC(tcps_badsyn); 1220 goto dropunlock; 1221 } 1222 /* 1223 * Segment's flags are (SYN) or (SYN|FIN). 1224 * 1225 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1226 * as they do not affect the state of the TCP FSM. 1227 * The data pointed to by TH_URG and th_urp is ignored. 1228 */ 1229 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1230 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1231 KASSERT(thflags & (TH_SYN), 1232 ("%s: Listen socket: TH_SYN not set", __func__)); 1233 INP_RLOCK_ASSERT(inp); 1234 #ifdef INET6 1235 /* 1236 * If deprecated address is forbidden, 1237 * we do not accept SYN to deprecated interface 1238 * address to prevent any new inbound connection from 1239 * getting established. 1240 * When we do not accept SYN, we send a TCP RST, 1241 * with deprecated source address (instead of dropping 1242 * it). We compromise it as it is much better for peer 1243 * to send a RST, and RST will be the final packet 1244 * for the exchange. 1245 * 1246 * If we do not forbid deprecated addresses, we accept 1247 * the SYN packet. RFC2462 does not suggest dropping 1248 * SYN in this case. 1249 * If we decipher RFC2462 5.5.4, it says like this: 1250 * 1. use of deprecated addr with existing 1251 * communication is okay - "SHOULD continue to be 1252 * used" 1253 * 2. use of it with new communication: 1254 * (2a) "SHOULD NOT be used if alternate address 1255 * with sufficient scope is available" 1256 * (2b) nothing mentioned otherwise. 1257 * Here we fall into (2b) case as we have no choice in 1258 * our source address selection - we must obey the peer. 1259 * 1260 * The wording in RFC2462 is confusing, and there are 1261 * multiple description text for deprecated address 1262 * handling - worse, they are not exactly the same. 1263 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1264 */ 1265 if (isipv6 && !V_ip6_use_deprecated) { 1266 struct in6_ifaddr *ia6; 1267 1268 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1269 if (ia6 != NULL && 1270 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1271 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1272 log(LOG_DEBUG, "%s; %s: Listen socket: " 1273 "Connection attempt to deprecated " 1274 "IPv6 address rejected\n", 1275 s, __func__); 1276 rstreason = BANDLIM_RST_OPENPORT; 1277 goto dropwithreset; 1278 } 1279 } 1280 #endif /* INET6 */ 1281 /* 1282 * Basic sanity checks on incoming SYN requests: 1283 * Don't respond if the destination is a link layer 1284 * broadcast according to RFC1122 4.2.3.10, p. 104. 1285 * If it is from this socket it must be forged. 1286 * Don't respond if the source or destination is a 1287 * global or subnet broad- or multicast address. 1288 * Note that it is quite possible to receive unicast 1289 * link-layer packets with a broadcast IP address. Use 1290 * in_broadcast() to find them. 1291 */ 1292 if (m->m_flags & (M_BCAST|M_MCAST)) { 1293 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1294 log(LOG_DEBUG, "%s; %s: Listen socket: " 1295 "Connection attempt from broad- or multicast " 1296 "link layer address ignored\n", s, __func__); 1297 goto dropunlock; 1298 } 1299 #ifdef INET6 1300 if (isipv6) { 1301 if (th->th_dport == th->th_sport && 1302 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1303 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1304 log(LOG_DEBUG, "%s; %s: Listen socket: " 1305 "Connection attempt to/from self " 1306 "ignored\n", s, __func__); 1307 goto dropunlock; 1308 } 1309 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1310 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1311 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1312 log(LOG_DEBUG, "%s; %s: Listen socket: " 1313 "Connection attempt from/to multicast " 1314 "address ignored\n", s, __func__); 1315 goto dropunlock; 1316 } 1317 } 1318 #endif 1319 #if defined(INET) && defined(INET6) 1320 else 1321 #endif 1322 #ifdef INET 1323 { 1324 if (th->th_dport == th->th_sport && 1325 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1326 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1327 log(LOG_DEBUG, "%s; %s: Listen socket: " 1328 "Connection attempt from/to self " 1329 "ignored\n", s, __func__); 1330 goto dropunlock; 1331 } 1332 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1333 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1334 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1335 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1336 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1337 log(LOG_DEBUG, "%s; %s: Listen socket: " 1338 "Connection attempt from/to broad- " 1339 "or multicast address ignored\n", 1340 s, __func__); 1341 goto dropunlock; 1342 } 1343 } 1344 #endif 1345 /* 1346 * SYN appears to be valid. Create compressed TCP state 1347 * for syncache. 1348 */ 1349 TCP_PROBE3(debug__input, tp, th, m); 1350 tcp_dooptions(&to, optp, optlen, TO_SYN); 1351 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, 1352 iptos, port)) != NULL) 1353 goto tfo_socket_result; 1354 1355 /* 1356 * Entry added to syncache and mbuf consumed. 1357 * Only the listen socket is unlocked by syncache_add(). 1358 */ 1359 return (IPPROTO_DONE); 1360 } else if (tp->t_state == TCPS_LISTEN) { 1361 /* 1362 * When a listen socket is torn down the SO_ACCEPTCONN 1363 * flag is removed first while connections are drained 1364 * from the accept queue in a unlock/lock cycle of the 1365 * ACCEPT_LOCK, opening a race condition allowing a SYN 1366 * attempt go through unhandled. 1367 */ 1368 goto dropunlock; 1369 } 1370 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1371 if (tp->t_flags & TF_SIGNATURE) { 1372 tcp_dooptions(&to, optp, optlen, thflags); 1373 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1374 TCPSTAT_INC(tcps_sig_err_nosigopt); 1375 goto dropunlock; 1376 } 1377 if (!TCPMD5_ENABLED() || 1378 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1379 goto dropunlock; 1380 } 1381 #endif 1382 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1383 1384 /* 1385 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1386 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1387 * the inpcb, and unlocks pcbinfo. 1388 * 1389 * XXXGL: in case of a pure SYN arriving on existing connection 1390 * TCP stacks won't need to modify the PCB, they would either drop 1391 * the segment silently, or send a challenge ACK. However, we try 1392 * to upgrade the lock, because calling convention for stacks is 1393 * write-lock on PCB. If upgrade fails, drop the SYN. 1394 */ 1395 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0) 1396 goto dropunlock; 1397 1398 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, tlen, iptos); 1399 return (IPPROTO_DONE); 1400 1401 dropwithreset: 1402 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1403 1404 if (inp != NULL) { 1405 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1406 INP_UNLOCK(inp); 1407 } else 1408 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1409 m = NULL; /* mbuf chain got consumed. */ 1410 goto drop; 1411 1412 dropunlock: 1413 if (m != NULL) 1414 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1415 1416 if (inp != NULL) 1417 INP_UNLOCK(inp); 1418 1419 drop: 1420 if (s != NULL) 1421 free(s, M_TCPLOG); 1422 if (m != NULL) 1423 m_freem(m); 1424 return (IPPROTO_DONE); 1425 } 1426 1427 /* 1428 * Automatic sizing of receive socket buffer. Often the send 1429 * buffer size is not optimally adjusted to the actual network 1430 * conditions at hand (delay bandwidth product). Setting the 1431 * buffer size too small limits throughput on links with high 1432 * bandwidth and high delay (eg. trans-continental/oceanic links). 1433 * 1434 * On the receive side the socket buffer memory is only rarely 1435 * used to any significant extent. This allows us to be much 1436 * more aggressive in scaling the receive socket buffer. For 1437 * the case that the buffer space is actually used to a large 1438 * extent and we run out of kernel memory we can simply drop 1439 * the new segments; TCP on the sender will just retransmit it 1440 * later. Setting the buffer size too big may only consume too 1441 * much kernel memory if the application doesn't read() from 1442 * the socket or packet loss or reordering makes use of the 1443 * reassembly queue. 1444 * 1445 * The criteria to step up the receive buffer one notch are: 1446 * 1. Application has not set receive buffer size with 1447 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1448 * 2. the number of bytes received during 1/2 of an sRTT 1449 * is at least 3/8 of the current socket buffer size. 1450 * 3. receive buffer size has not hit maximal automatic size; 1451 * 1452 * If all of the criteria are met we increaset the socket buffer 1453 * by a 1/2 (bounded by the max). This allows us to keep ahead 1454 * of slow-start but also makes it so our peer never gets limited 1455 * by our rwnd which we then open up causing a burst. 1456 * 1457 * This algorithm does two steps per RTT at most and only if 1458 * we receive a bulk stream w/o packet losses or reorderings. 1459 * Shrinking the buffer during idle times is not necessary as 1460 * it doesn't consume any memory when idle. 1461 * 1462 * TODO: Only step up if the application is actually serving 1463 * the buffer to better manage the socket buffer resources. 1464 */ 1465 int 1466 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1467 struct tcpcb *tp, int tlen) 1468 { 1469 int newsize = 0; 1470 1471 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1472 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1473 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1474 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1475 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1476 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1477 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1478 } 1479 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1480 1481 /* Start over with next RTT. */ 1482 tp->rfbuf_ts = 0; 1483 tp->rfbuf_cnt = 0; 1484 } else { 1485 tp->rfbuf_cnt += tlen; /* add up */ 1486 } 1487 return (newsize); 1488 } 1489 1490 int 1491 tcp_input(struct mbuf **mp, int *offp, int proto) 1492 { 1493 return(tcp_input_with_port(mp, offp, proto, 0)); 1494 } 1495 1496 static void 1497 tcp_handle_wakeup(struct tcpcb *tp) 1498 { 1499 1500 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1501 1502 if (tp->t_flags & TF_WAKESOR) { 1503 struct socket *so = tptosocket(tp); 1504 1505 tp->t_flags &= ~TF_WAKESOR; 1506 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1507 sorwakeup_locked(so); 1508 } 1509 } 1510 1511 void 1512 tcp_do_segment(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th, 1513 int drop_hdrlen, int tlen, uint8_t iptos) 1514 { 1515 uint16_t thflags; 1516 int acked, ourfinisacked, needoutput = 0; 1517 sackstatus_t sack_changed; 1518 int rstreason, todrop, win, incforsyn = 0; 1519 uint32_t tiwin; 1520 uint16_t nsegs; 1521 char *s; 1522 struct inpcb *inp = tptoinpcb(tp); 1523 struct socket *so = tptosocket(tp); 1524 struct in_conninfo *inc = &inp->inp_inc; 1525 struct mbuf *mfree; 1526 struct tcpopt to; 1527 int tfo_syn; 1528 u_int maxseg; 1529 1530 thflags = tcp_get_flags(th); 1531 tp->sackhint.last_sack_ack = 0; 1532 sack_changed = SACK_NOCHANGE; 1533 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1534 1535 NET_EPOCH_ASSERT(); 1536 INP_WLOCK_ASSERT(inp); 1537 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1538 __func__)); 1539 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1540 __func__)); 1541 1542 #ifdef TCPPCAP 1543 /* Save segment, if requested. */ 1544 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1545 #endif 1546 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1547 tlen, NULL, true); 1548 1549 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1550 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1551 log(LOG_DEBUG, "%s; %s: " 1552 "SYN|FIN segment ignored (based on " 1553 "sysctl setting)\n", s, __func__); 1554 free(s, M_TCPLOG); 1555 } 1556 goto drop; 1557 } 1558 1559 /* 1560 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1561 * check SEQ.ACK first. 1562 */ 1563 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1564 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1565 rstreason = BANDLIM_UNLIMITED; 1566 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1567 goto dropwithreset; 1568 } 1569 1570 /* 1571 * Segment received on connection. 1572 * Reset idle time and keep-alive timer. 1573 * XXX: This should be done after segment 1574 * validation to ignore broken/spoofed segs. 1575 */ 1576 if (tp->t_idle_reduce && 1577 (tp->snd_max == tp->snd_una) && 1578 ((ticks - tp->t_rcvtime) >= tp->t_rxtcur)) 1579 cc_after_idle(tp); 1580 tp->t_rcvtime = ticks; 1581 1582 if (thflags & TH_FIN) 1583 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN); 1584 /* 1585 * Scale up the window into a 32-bit value. 1586 * For the SYN_SENT state the scale is zero. 1587 */ 1588 tiwin = th->th_win << tp->snd_scale; 1589 #ifdef STATS 1590 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1591 #endif 1592 1593 /* 1594 * TCP ECN processing. 1595 */ 1596 if (tcp_ecn_input_segment(tp, thflags, tlen, 1597 tcp_packets_this_ack(tp, th->th_ack), 1598 iptos)) 1599 cc_cong_signal(tp, th, CC_ECN); 1600 1601 /* 1602 * Parse options on any incoming segment. 1603 */ 1604 tcp_dooptions(&to, (u_char *)(th + 1), 1605 (th->th_off << 2) - sizeof(struct tcphdr), 1606 (thflags & TH_SYN) ? TO_SYN : 0); 1607 1608 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1609 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1610 (to.to_flags & TOF_SIGNATURE) == 0) { 1611 TCPSTAT_INC(tcps_sig_err_sigopt); 1612 /* XXX: should drop? */ 1613 } 1614 #endif 1615 /* 1616 * If echoed timestamp is later than the current time, 1617 * fall back to non RFC1323 RTT calculation. Normalize 1618 * timestamp if syncookies were used when this connection 1619 * was established. 1620 */ 1621 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1622 to.to_tsecr -= tp->ts_offset; 1623 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1624 to.to_tsecr = 0; 1625 else if (tp->t_rxtshift == 1 && 1626 tp->t_flags & TF_PREVVALID && 1627 tp->t_badrxtwin != 0 && 1628 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 1629 cc_cong_signal(tp, th, CC_RTO_ERR); 1630 } 1631 /* 1632 * Process options only when we get SYN/ACK back. The SYN case 1633 * for incoming connections is handled in tcp_syncache. 1634 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1635 * or <SYN,ACK>) segment itself is never scaled. 1636 * XXX this is traditional behavior, may need to be cleaned up. 1637 */ 1638 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1639 /* Handle parallel SYN for ECN */ 1640 tcp_ecn_input_parallel_syn(tp, thflags, iptos); 1641 if ((to.to_flags & TOF_SCALE) && 1642 (tp->t_flags & TF_REQ_SCALE) && 1643 !(tp->t_flags & TF_NOOPT)) { 1644 tp->t_flags |= TF_RCVD_SCALE; 1645 tp->snd_scale = to.to_wscale; 1646 } else 1647 tp->t_flags &= ~TF_REQ_SCALE; 1648 /* 1649 * Initial send window. It will be updated with 1650 * the next incoming segment to the scaled value. 1651 */ 1652 tp->snd_wnd = th->th_win; 1653 if ((to.to_flags & TOF_TS) && 1654 (tp->t_flags & TF_REQ_TSTMP) && 1655 !(tp->t_flags & TF_NOOPT)) { 1656 tp->t_flags |= TF_RCVD_TSTMP; 1657 tp->ts_recent = to.to_tsval; 1658 tp->ts_recent_age = tcp_ts_getticks(); 1659 } else 1660 tp->t_flags &= ~TF_REQ_TSTMP; 1661 if (to.to_flags & TOF_MSS) 1662 tcp_mss(tp, to.to_mss); 1663 if ((tp->t_flags & TF_SACK_PERMIT) && 1664 (!(to.to_flags & TOF_SACKPERM) || 1665 (tp->t_flags & TF_NOOPT))) 1666 tp->t_flags &= ~TF_SACK_PERMIT; 1667 if (IS_FASTOPEN(tp->t_flags)) { 1668 if ((to.to_flags & TOF_FASTOPEN) && 1669 !(tp->t_flags & TF_NOOPT)) { 1670 uint16_t mss; 1671 1672 if (to.to_flags & TOF_MSS) 1673 mss = to.to_mss; 1674 else 1675 if ((inp->inp_vflag & INP_IPV6) != 0) 1676 mss = TCP6_MSS; 1677 else 1678 mss = TCP_MSS; 1679 tcp_fastopen_update_cache(tp, mss, 1680 to.to_tfo_len, to.to_tfo_cookie); 1681 } else 1682 tcp_fastopen_disable_path(tp); 1683 } 1684 } 1685 1686 /* 1687 * If timestamps were negotiated during SYN/ACK and a 1688 * segment without a timestamp is received, silently drop 1689 * the segment, unless it is a RST segment or missing timestamps are 1690 * tolerated. 1691 * See section 3.2 of RFC 7323. 1692 */ 1693 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1694 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1695 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1696 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1697 "segment processed normally\n", 1698 s, __func__); 1699 free(s, M_TCPLOG); 1700 } 1701 } else { 1702 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1703 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1704 "segment silently dropped\n", s, __func__); 1705 free(s, M_TCPLOG); 1706 } 1707 goto drop; 1708 } 1709 } 1710 /* 1711 * If timestamps were not negotiated during SYN/ACK and a 1712 * segment with a timestamp is received, ignore the 1713 * timestamp and process the packet normally. 1714 * See section 3.2 of RFC 7323. 1715 */ 1716 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1717 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1718 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1719 "segment processed normally\n", s, __func__); 1720 free(s, M_TCPLOG); 1721 } 1722 } 1723 1724 /* 1725 * Header prediction: check for the two common cases 1726 * of a uni-directional data xfer. If the packet has 1727 * no control flags, is in-sequence, the window didn't 1728 * change and we're not retransmitting, it's a 1729 * candidate. If the length is zero and the ack moved 1730 * forward, we're the sender side of the xfer. Just 1731 * free the data acked & wake any higher level process 1732 * that was blocked waiting for space. If the length 1733 * is non-zero and the ack didn't move, we're the 1734 * receiver side. If we're getting packets in-order 1735 * (the reassembly queue is empty), add the data to 1736 * the socket buffer and note that we need a delayed ack. 1737 * Make sure that the hidden state-flags are also off. 1738 * Since we check for TCPS_ESTABLISHED first, it can only 1739 * be TH_NEEDSYN. 1740 */ 1741 if (tp->t_state == TCPS_ESTABLISHED && 1742 th->th_seq == tp->rcv_nxt && 1743 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1744 tp->snd_nxt == tp->snd_max && 1745 tiwin && tiwin == tp->snd_wnd && 1746 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1747 SEGQ_EMPTY(tp) && 1748 ((to.to_flags & TOF_TS) == 0 || 1749 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1750 /* 1751 * If last ACK falls within this segment's sequence numbers, 1752 * record the timestamp. 1753 * NOTE that the test is modified according to the latest 1754 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1755 */ 1756 if ((to.to_flags & TOF_TS) != 0 && 1757 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1758 tp->ts_recent_age = tcp_ts_getticks(); 1759 tp->ts_recent = to.to_tsval; 1760 } 1761 1762 if (tlen == 0) { 1763 if (SEQ_GT(th->th_ack, tp->snd_una) && 1764 SEQ_LEQ(th->th_ack, tp->snd_max) && 1765 !IN_RECOVERY(tp->t_flags) && 1766 (to.to_flags & TOF_SACK) == 0 && 1767 TAILQ_EMPTY(&tp->snd_holes)) { 1768 /* 1769 * This is a pure ack for outstanding data. 1770 */ 1771 TCPSTAT_INC(tcps_predack); 1772 1773 /* 1774 * "bad retransmit" recovery without timestamps. 1775 */ 1776 if ((to.to_flags & TOF_TS) == 0 && 1777 tp->t_rxtshift == 1 && 1778 tp->t_flags & TF_PREVVALID && 1779 tp->t_badrxtwin != 0 && 1780 TSTMP_LT(ticks, tp->t_badrxtwin)) { 1781 cc_cong_signal(tp, th, CC_RTO_ERR); 1782 } 1783 1784 /* 1785 * Recalculate the transmit timer / rtt. 1786 * 1787 * Some boxes send broken timestamp replies 1788 * during the SYN+ACK phase, ignore 1789 * timestamps of 0 or we could calculate a 1790 * huge RTT and blow up the retransmit timer. 1791 */ 1792 if ((to.to_flags & TOF_TS) != 0 && 1793 to.to_tsecr) { 1794 uint32_t t; 1795 1796 t = tcp_ts_getticks() - to.to_tsecr; 1797 if (!tp->t_rttlow || tp->t_rttlow > t) 1798 tp->t_rttlow = t; 1799 tcp_xmit_timer(tp, 1800 TCP_TS_TO_TICKS(t) + 1); 1801 } else if (tp->t_rtttime && 1802 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1803 if (!tp->t_rttlow || 1804 tp->t_rttlow > ticks - tp->t_rtttime) 1805 tp->t_rttlow = ticks - tp->t_rtttime; 1806 tcp_xmit_timer(tp, 1807 ticks - tp->t_rtttime); 1808 } 1809 acked = BYTES_THIS_ACK(tp, th); 1810 1811 #ifdef TCP_HHOOK 1812 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1813 hhook_run_tcp_est_in(tp, th, &to); 1814 #endif 1815 1816 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1817 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1818 sbdrop(&so->so_snd, acked); 1819 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1820 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1821 tp->snd_recover = th->th_ack - 1; 1822 1823 /* 1824 * Let the congestion control algorithm update 1825 * congestion control related information. This 1826 * typically means increasing the congestion 1827 * window. 1828 */ 1829 cc_ack_received(tp, th, nsegs, CC_ACK); 1830 1831 tp->snd_una = th->th_ack; 1832 /* 1833 * Pull snd_wl2 up to prevent seq wrap relative 1834 * to th_ack. 1835 */ 1836 tp->snd_wl2 = th->th_ack; 1837 tp->t_dupacks = 0; 1838 m_freem(m); 1839 1840 /* 1841 * If all outstanding data are acked, stop 1842 * retransmit timer, otherwise restart timer 1843 * using current (possibly backed-off) value. 1844 * If process is waiting for space, 1845 * wakeup/selwakeup/signal. If data 1846 * are ready to send, let tcp_output 1847 * decide between more output or persist. 1848 */ 1849 TCP_PROBE3(debug__input, tp, th, m); 1850 /* 1851 * Clear t_acktime if remote side has ACKd 1852 * all data in the socket buffer. 1853 * Otherwise, update t_acktime if we received 1854 * a sufficiently large ACK. 1855 */ 1856 if (sbavail(&so->so_snd) == 0) 1857 tp->t_acktime = 0; 1858 else if (acked > 1) 1859 tp->t_acktime = ticks; 1860 if (tp->snd_una == tp->snd_max) 1861 tcp_timer_activate(tp, TT_REXMT, 0); 1862 else if (!tcp_timer_active(tp, TT_PERSIST)) 1863 tcp_timer_activate(tp, TT_REXMT, 1864 TP_RXTCUR(tp)); 1865 sowwakeup(so); 1866 if (sbavail(&so->so_snd)) 1867 (void) tcp_output(tp); 1868 goto check_delack; 1869 } 1870 } else if (th->th_ack == tp->snd_una && 1871 tlen <= sbspace(&so->so_rcv)) { 1872 int newsize = 0; /* automatic sockbuf scaling */ 1873 1874 /* 1875 * This is a pure, in-sequence data packet with 1876 * nothing on the reassembly queue and we have enough 1877 * buffer space to take it. 1878 */ 1879 /* Clean receiver SACK report if present */ 1880 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1881 tcp_clean_sackreport(tp); 1882 TCPSTAT_INC(tcps_preddat); 1883 tp->rcv_nxt += tlen; 1884 if (tlen && 1885 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1886 (tp->t_fbyte_in == 0)) { 1887 tp->t_fbyte_in = ticks; 1888 if (tp->t_fbyte_in == 0) 1889 tp->t_fbyte_in = 1; 1890 if (tp->t_fbyte_out && tp->t_fbyte_in) 1891 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1892 } 1893 /* 1894 * Pull snd_wl1 up to prevent seq wrap relative to 1895 * th_seq. 1896 */ 1897 tp->snd_wl1 = th->th_seq; 1898 /* 1899 * Pull rcv_up up to prevent seq wrap relative to 1900 * rcv_nxt. 1901 */ 1902 tp->rcv_up = tp->rcv_nxt; 1903 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1904 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1905 TCP_PROBE3(debug__input, tp, th, m); 1906 1907 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1908 1909 /* Add data to socket buffer. */ 1910 SOCKBUF_LOCK(&so->so_rcv); 1911 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1912 m_freem(m); 1913 } else { 1914 /* 1915 * Set new socket buffer size. 1916 * Give up when limit is reached. 1917 */ 1918 if (newsize) 1919 if (!sbreserve_locked(so, SO_RCV, 1920 newsize, NULL)) 1921 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1922 m_adj(m, drop_hdrlen); /* delayed header drop */ 1923 sbappendstream_locked(&so->so_rcv, m, 0); 1924 } 1925 /* NB: sorwakeup_locked() does an implicit unlock. */ 1926 sorwakeup_locked(so); 1927 if (DELAY_ACK(tp, tlen)) { 1928 tp->t_flags |= TF_DELACK; 1929 } else { 1930 tp->t_flags |= TF_ACKNOW; 1931 tcp_output(tp); 1932 } 1933 goto check_delack; 1934 } 1935 } 1936 1937 /* 1938 * Calculate amount of space in receive window, 1939 * and then do TCP input processing. 1940 * Receive window is amount of space in rcv queue, 1941 * but not less than advertised window. 1942 */ 1943 win = sbspace(&so->so_rcv); 1944 if (win < 0) 1945 win = 0; 1946 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1947 1948 switch (tp->t_state) { 1949 /* 1950 * If the state is SYN_RECEIVED: 1951 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1952 */ 1953 case TCPS_SYN_RECEIVED: 1954 if (thflags & TH_RST) { 1955 /* Handle RST segments later. */ 1956 break; 1957 } 1958 if ((thflags & TH_ACK) && 1959 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1960 SEQ_GT(th->th_ack, tp->snd_max))) { 1961 rstreason = BANDLIM_RST_OPENPORT; 1962 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1963 goto dropwithreset; 1964 } 1965 if (IS_FASTOPEN(tp->t_flags)) { 1966 /* 1967 * When a TFO connection is in SYN_RECEIVED, the 1968 * only valid packets are the initial SYN, a 1969 * retransmit/copy of the initial SYN (possibly with 1970 * a subset of the original data), a valid ACK, a 1971 * FIN, or a RST. 1972 */ 1973 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1974 rstreason = BANDLIM_RST_OPENPORT; 1975 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1976 goto dropwithreset; 1977 } else if (thflags & TH_SYN) { 1978 /* non-initial SYN is ignored */ 1979 if ((tcp_timer_active(tp, TT_DELACK) || 1980 tcp_timer_active(tp, TT_REXMT))) 1981 goto drop; 1982 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1983 goto drop; 1984 } 1985 } 1986 break; 1987 1988 /* 1989 * If the state is SYN_SENT: 1990 * if seg contains a RST with valid ACK (SEQ.ACK has already 1991 * been verified), then drop the connection. 1992 * if seg contains a RST without an ACK, drop the seg. 1993 * if seg does not contain SYN, then drop the seg. 1994 * Otherwise this is an acceptable SYN segment 1995 * initialize tp->rcv_nxt and tp->irs 1996 * if seg contains ack then advance tp->snd_una 1997 * if seg contains an ECE and ECN support is enabled, the stream 1998 * is ECN capable. 1999 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2000 * arrange for segment to be acked (eventually) 2001 * continue processing rest of data/controls, beginning with URG 2002 */ 2003 case TCPS_SYN_SENT: 2004 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2005 TCP_PROBE5(connect__refused, NULL, tp, 2006 m, tp, th); 2007 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2008 tp = tcp_drop(tp, ECONNREFUSED); 2009 } 2010 if (thflags & TH_RST) 2011 goto drop; 2012 if (!(thflags & TH_SYN)) 2013 goto drop; 2014 2015 tp->irs = th->th_seq; 2016 tcp_rcvseqinit(tp); 2017 if (thflags & TH_ACK) { 2018 int tfo_partial_ack = 0; 2019 2020 TCPSTAT_INC(tcps_connects); 2021 soisconnected(so); 2022 #ifdef MAC 2023 mac_socketpeer_set_from_mbuf(m, so); 2024 #endif 2025 /* Do window scaling on this connection? */ 2026 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2027 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2028 tp->rcv_scale = tp->request_r_scale; 2029 } 2030 tp->rcv_adv += min(tp->rcv_wnd, 2031 TCP_MAXWIN << tp->rcv_scale); 2032 tp->snd_una++; /* SYN is acked */ 2033 /* 2034 * If not all the data that was sent in the TFO SYN 2035 * has been acked, resend the remainder right away. 2036 */ 2037 if (IS_FASTOPEN(tp->t_flags) && 2038 (tp->snd_una != tp->snd_max)) { 2039 tp->snd_nxt = th->th_ack; 2040 tfo_partial_ack = 1; 2041 } 2042 /* 2043 * If there's data, delay ACK; if there's also a FIN 2044 * ACKNOW will be turned on later. 2045 */ 2046 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2047 tcp_timer_activate(tp, TT_DELACK, 2048 tcp_delacktime); 2049 else 2050 tp->t_flags |= TF_ACKNOW; 2051 2052 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2053 2054 /* 2055 * Received <SYN,ACK> in SYN_SENT[*] state. 2056 * Transitions: 2057 * SYN_SENT --> ESTABLISHED 2058 * SYN_SENT* --> FIN_WAIT_1 2059 */ 2060 tp->t_starttime = ticks; 2061 if (tp->t_flags & TF_NEEDFIN) { 2062 tp->t_acktime = ticks; 2063 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2064 tp->t_flags &= ~TF_NEEDFIN; 2065 thflags &= ~TH_SYN; 2066 } else { 2067 tcp_state_change(tp, TCPS_ESTABLISHED); 2068 TCP_PROBE5(connect__established, NULL, tp, 2069 m, tp, th); 2070 cc_conn_init(tp); 2071 tcp_timer_activate(tp, TT_KEEP, 2072 TP_KEEPIDLE(tp)); 2073 } 2074 } else { 2075 /* 2076 * Received initial SYN in SYN-SENT[*] state => 2077 * simultaneous open. 2078 * If it succeeds, connection is * half-synchronized. 2079 * Otherwise, do 3-way handshake: 2080 * SYN-SENT -> SYN-RECEIVED 2081 * SYN-SENT* -> SYN-RECEIVED* 2082 */ 2083 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN); 2084 tcp_timer_activate(tp, TT_REXMT, 0); 2085 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2086 } 2087 2088 /* 2089 * Advance th->th_seq to correspond to first data byte. 2090 * If data, trim to stay within window, 2091 * dropping FIN if necessary. 2092 */ 2093 th->th_seq++; 2094 if (tlen > tp->rcv_wnd) { 2095 todrop = tlen - tp->rcv_wnd; 2096 m_adj(m, -todrop); 2097 tlen = tp->rcv_wnd; 2098 thflags &= ~TH_FIN; 2099 TCPSTAT_INC(tcps_rcvpackafterwin); 2100 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2101 } 2102 tp->snd_wl1 = th->th_seq - 1; 2103 tp->rcv_up = th->th_seq; 2104 /* 2105 * Client side of transaction: already sent SYN and data. 2106 * If the remote host used T/TCP to validate the SYN, 2107 * our data will be ACK'd; if so, enter normal data segment 2108 * processing in the middle of step 5, ack processing. 2109 * Otherwise, goto step 6. 2110 */ 2111 if (thflags & TH_ACK) 2112 goto process_ACK; 2113 2114 goto step6; 2115 } 2116 2117 /* 2118 * States other than LISTEN or SYN_SENT. 2119 * First check the RST flag and sequence number since reset segments 2120 * are exempt from the timestamp and connection count tests. This 2121 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2122 * below which allowed reset segments in half the sequence space 2123 * to fall though and be processed (which gives forged reset 2124 * segments with a random sequence number a 50 percent chance of 2125 * killing a connection). 2126 * Then check timestamp, if present. 2127 * Then check the connection count, if present. 2128 * Then check that at least some bytes of segment are within 2129 * receive window. If segment begins before rcv_nxt, 2130 * drop leading data (and SYN); if nothing left, just ack. 2131 */ 2132 if (thflags & TH_RST) { 2133 /* 2134 * RFC5961 Section 3.2 2135 * 2136 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2137 * - If RST is in window, we send challenge ACK. 2138 * 2139 * Note: to take into account delayed ACKs, we should 2140 * test against last_ack_sent instead of rcv_nxt. 2141 * Note 2: we handle special case of closed window, not 2142 * covered by the RFC. 2143 */ 2144 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2145 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2146 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2147 KASSERT(tp->t_state != TCPS_SYN_SENT, 2148 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2149 __func__, th, tp)); 2150 2151 if (V_tcp_insecure_rst || 2152 tp->last_ack_sent == th->th_seq) { 2153 TCPSTAT_INC(tcps_drops); 2154 /* Drop the connection. */ 2155 switch (tp->t_state) { 2156 case TCPS_SYN_RECEIVED: 2157 so->so_error = ECONNREFUSED; 2158 goto close; 2159 case TCPS_ESTABLISHED: 2160 case TCPS_FIN_WAIT_1: 2161 case TCPS_FIN_WAIT_2: 2162 case TCPS_CLOSE_WAIT: 2163 case TCPS_CLOSING: 2164 case TCPS_LAST_ACK: 2165 so->so_error = ECONNRESET; 2166 close: 2167 /* FALLTHROUGH */ 2168 default: 2169 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST); 2170 tp = tcp_close(tp); 2171 } 2172 } else { 2173 TCPSTAT_INC(tcps_badrst); 2174 /* Send challenge ACK. */ 2175 tcp_respond(tp, mtod(m, void *), th, m, 2176 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2177 tp->last_ack_sent = tp->rcv_nxt; 2178 m = NULL; 2179 } 2180 } 2181 goto drop; 2182 } 2183 2184 /* 2185 * RFC5961 Section 4.2 2186 * Send challenge ACK for any SYN in synchronized state. 2187 */ 2188 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2189 tp->t_state != TCPS_SYN_RECEIVED) { 2190 TCPSTAT_INC(tcps_badsyn); 2191 if (V_tcp_insecure_syn && 2192 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2193 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2194 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2195 tp = tcp_drop(tp, ECONNRESET); 2196 rstreason = BANDLIM_UNLIMITED; 2197 } else { 2198 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2199 /* Send challenge ACK. */ 2200 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2201 tp->snd_nxt, TH_ACK); 2202 tp->last_ack_sent = tp->rcv_nxt; 2203 m = NULL; 2204 } 2205 goto drop; 2206 } 2207 2208 /* 2209 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2210 * and it's less than ts_recent, drop it. 2211 */ 2212 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2213 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2214 /* Check to see if ts_recent is over 24 days old. */ 2215 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2216 /* 2217 * Invalidate ts_recent. If this segment updates 2218 * ts_recent, the age will be reset later and ts_recent 2219 * will get a valid value. If it does not, setting 2220 * ts_recent to zero will at least satisfy the 2221 * requirement that zero be placed in the timestamp 2222 * echo reply when ts_recent isn't valid. The 2223 * age isn't reset until we get a valid ts_recent 2224 * because we don't want out-of-order segments to be 2225 * dropped when ts_recent is old. 2226 */ 2227 tp->ts_recent = 0; 2228 } else { 2229 TCPSTAT_INC(tcps_rcvduppack); 2230 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2231 TCPSTAT_INC(tcps_pawsdrop); 2232 if (tlen) 2233 goto dropafterack; 2234 goto drop; 2235 } 2236 } 2237 2238 /* 2239 * In the SYN-RECEIVED state, validate that the packet belongs to 2240 * this connection before trimming the data to fit the receive 2241 * window. Check the sequence number versus IRS since we know 2242 * the sequence numbers haven't wrapped. This is a partial fix 2243 * for the "LAND" DoS attack. 2244 */ 2245 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2246 rstreason = BANDLIM_RST_OPENPORT; 2247 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2248 goto dropwithreset; 2249 } 2250 2251 todrop = tp->rcv_nxt - th->th_seq; 2252 if (todrop > 0) { 2253 if (thflags & TH_SYN) { 2254 thflags &= ~TH_SYN; 2255 th->th_seq++; 2256 if (th->th_urp > 1) 2257 th->th_urp--; 2258 else 2259 thflags &= ~TH_URG; 2260 todrop--; 2261 } 2262 /* 2263 * Following if statement from Stevens, vol. 2, p. 960. 2264 */ 2265 if (todrop > tlen 2266 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2267 /* 2268 * Any valid FIN must be to the left of the window. 2269 * At this point the FIN must be a duplicate or out 2270 * of sequence; drop it. 2271 */ 2272 thflags &= ~TH_FIN; 2273 2274 /* 2275 * Send an ACK to resynchronize and drop any data. 2276 * But keep on processing for RST or ACK. 2277 */ 2278 tp->t_flags |= TF_ACKNOW; 2279 todrop = tlen; 2280 TCPSTAT_INC(tcps_rcvduppack); 2281 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2282 } else { 2283 TCPSTAT_INC(tcps_rcvpartduppack); 2284 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2285 } 2286 /* 2287 * DSACK - add SACK block for dropped range 2288 */ 2289 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2290 tcp_update_sack_list(tp, th->th_seq, 2291 th->th_seq + todrop); 2292 /* 2293 * ACK now, as the next in-sequence segment 2294 * will clear the DSACK block again 2295 */ 2296 tp->t_flags |= TF_ACKNOW; 2297 } 2298 drop_hdrlen += todrop; /* drop from the top afterwards */ 2299 th->th_seq += todrop; 2300 tlen -= todrop; 2301 if (th->th_urp > todrop) 2302 th->th_urp -= todrop; 2303 else { 2304 thflags &= ~TH_URG; 2305 th->th_urp = 0; 2306 } 2307 } 2308 2309 /* 2310 * If new data are received on a connection after the 2311 * user processes are gone, then RST the other end. 2312 */ 2313 if ((tp->t_flags & TF_CLOSED) && tlen) { 2314 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2315 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2316 "after socket was closed, " 2317 "sending RST and removing tcpcb\n", 2318 s, __func__, tcpstates[tp->t_state], tlen); 2319 free(s, M_TCPLOG); 2320 } 2321 tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE); 2322 /* tcp_close will kill the inp pre-log the Reset */ 2323 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST); 2324 tp = tcp_close(tp); 2325 TCPSTAT_INC(tcps_rcvafterclose); 2326 rstreason = BANDLIM_UNLIMITED; 2327 goto dropwithreset; 2328 } 2329 2330 /* 2331 * If segment ends after window, drop trailing data 2332 * (and PUSH and FIN); if nothing left, just ACK. 2333 */ 2334 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2335 if (todrop > 0) { 2336 TCPSTAT_INC(tcps_rcvpackafterwin); 2337 if (todrop >= tlen) { 2338 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2339 /* 2340 * If window is closed can only take segments at 2341 * window edge, and have to drop data and PUSH from 2342 * incoming segments. Continue processing, but 2343 * remember to ack. Otherwise, drop segment 2344 * and ack. 2345 */ 2346 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2347 tp->t_flags |= TF_ACKNOW; 2348 TCPSTAT_INC(tcps_rcvwinprobe); 2349 } else 2350 goto dropafterack; 2351 } else 2352 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2353 m_adj(m, -todrop); 2354 tlen -= todrop; 2355 thflags &= ~(TH_PUSH|TH_FIN); 2356 } 2357 2358 /* 2359 * If last ACK falls within this segment's sequence numbers, 2360 * record its timestamp. 2361 * NOTE: 2362 * 1) That the test incorporates suggestions from the latest 2363 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2364 * 2) That updating only on newer timestamps interferes with 2365 * our earlier PAWS tests, so this check should be solely 2366 * predicated on the sequence space of this segment. 2367 * 3) That we modify the segment boundary check to be 2368 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2369 * instead of RFC1323's 2370 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2371 * This modified check allows us to overcome RFC1323's 2372 * limitations as described in Stevens TCP/IP Illustrated 2373 * Vol. 2 p.869. In such cases, we can still calculate the 2374 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2375 */ 2376 if ((to.to_flags & TOF_TS) != 0 && 2377 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2378 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2379 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2380 tp->ts_recent_age = tcp_ts_getticks(); 2381 tp->ts_recent = to.to_tsval; 2382 } 2383 2384 /* 2385 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2386 * flag is on (half-synchronized state), then queue data for 2387 * later processing; else drop segment and return. 2388 */ 2389 if ((thflags & TH_ACK) == 0) { 2390 if (tp->t_state == TCPS_SYN_RECEIVED || 2391 (tp->t_flags & TF_NEEDSYN)) { 2392 if (tp->t_state == TCPS_SYN_RECEIVED && 2393 IS_FASTOPEN(tp->t_flags)) { 2394 tp->snd_wnd = tiwin; 2395 cc_conn_init(tp); 2396 } 2397 goto step6; 2398 } else if (tp->t_flags & TF_ACKNOW) 2399 goto dropafterack; 2400 else 2401 goto drop; 2402 } 2403 2404 /* 2405 * Ack processing. 2406 */ 2407 switch (tp->t_state) { 2408 /* 2409 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2410 * ESTABLISHED state and continue processing. 2411 * The ACK was checked above. 2412 */ 2413 case TCPS_SYN_RECEIVED: 2414 2415 TCPSTAT_INC(tcps_connects); 2416 if (tp->t_flags & TF_SONOTCONN) { 2417 /* 2418 * Usually SYN_RECEIVED had been created from a LISTEN, 2419 * and solisten_enqueue() has already marked the socket 2420 * layer as connected. If it didn't, which can happen 2421 * only with an accept_filter(9), then the tp is marked 2422 * with TF_SONOTCONN. The other reason for this mark 2423 * to be set is a simultaneous open, a SYN_RECEIVED 2424 * that had been created from SYN_SENT. 2425 */ 2426 tp->t_flags &= ~TF_SONOTCONN; 2427 soisconnected(so); 2428 } 2429 /* Do window scaling? */ 2430 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2431 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2432 tp->rcv_scale = tp->request_r_scale; 2433 } 2434 tp->snd_wnd = tiwin; 2435 /* 2436 * Make transitions: 2437 * SYN-RECEIVED -> ESTABLISHED 2438 * SYN-RECEIVED* -> FIN-WAIT-1 2439 */ 2440 tp->t_starttime = ticks; 2441 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2442 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2443 tp->t_tfo_pending = NULL; 2444 } 2445 if (tp->t_flags & TF_NEEDFIN) { 2446 tp->t_acktime = ticks; 2447 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2448 tp->t_flags &= ~TF_NEEDFIN; 2449 } else { 2450 tcp_state_change(tp, TCPS_ESTABLISHED); 2451 TCP_PROBE5(accept__established, NULL, tp, 2452 m, tp, th); 2453 /* 2454 * TFO connections call cc_conn_init() during SYN 2455 * processing. Calling it again here for such 2456 * connections is not harmless as it would undo the 2457 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2458 * is retransmitted. 2459 */ 2460 if (!IS_FASTOPEN(tp->t_flags)) 2461 cc_conn_init(tp); 2462 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2463 } 2464 /* 2465 * Account for the ACK of our SYN prior to 2466 * regular ACK processing below, except for 2467 * simultaneous SYN, which is handled later. 2468 */ 2469 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2470 incforsyn = 1; 2471 /* 2472 * If segment contains data or ACK, will call tcp_reass() 2473 * later; if not, do so now to pass queued data to user. 2474 */ 2475 if (tlen == 0 && (thflags & TH_FIN) == 0) { 2476 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2477 (struct mbuf *)0); 2478 tcp_handle_wakeup(tp); 2479 } 2480 tp->snd_wl1 = th->th_seq - 1; 2481 /* FALLTHROUGH */ 2482 2483 /* 2484 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2485 * ACKs. If the ack is in the range 2486 * tp->snd_una < th->th_ack <= tp->snd_max 2487 * then advance tp->snd_una to th->th_ack and drop 2488 * data from the retransmission queue. If this ACK reflects 2489 * more up to date window information we update our window information. 2490 */ 2491 case TCPS_ESTABLISHED: 2492 case TCPS_FIN_WAIT_1: 2493 case TCPS_FIN_WAIT_2: 2494 case TCPS_CLOSE_WAIT: 2495 case TCPS_CLOSING: 2496 case TCPS_LAST_ACK: 2497 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2498 TCPSTAT_INC(tcps_rcvacktoomuch); 2499 goto dropafterack; 2500 } 2501 if (tcp_is_sack_recovery(tp, &to)) { 2502 if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) && 2503 (tp->t_flags & TF_LRD)) { 2504 tcp_sack_lost_retransmission(tp, th); 2505 } 2506 } else 2507 /* 2508 * Reset the value so that previous (valid) value 2509 * from the last ack with SACK doesn't get used. 2510 */ 2511 tp->sackhint.sacked_bytes = 0; 2512 2513 #ifdef TCP_HHOOK 2514 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2515 hhook_run_tcp_est_in(tp, th, &to); 2516 #endif 2517 2518 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2519 maxseg = tcp_maxseg(tp); 2520 if (tlen == 0 && 2521 (tiwin == tp->snd_wnd || 2522 (tp->t_flags & TF_SACK_PERMIT))) { 2523 /* 2524 * If this is the first time we've seen a 2525 * FIN from the remote, this is not a 2526 * duplicate and it needs to be processed 2527 * normally. This happens during a 2528 * simultaneous close. 2529 */ 2530 if ((thflags & TH_FIN) && 2531 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2532 tp->t_dupacks = 0; 2533 break; 2534 } 2535 TCPSTAT_INC(tcps_rcvdupack); 2536 /* 2537 * If we have outstanding data (other than 2538 * a window probe), this is a completely 2539 * duplicate ack (ie, window info didn't 2540 * change and FIN isn't set), 2541 * the ack is the biggest we've 2542 * seen and we've seen exactly our rexmt 2543 * threshold of them, assume a packet 2544 * has been dropped and retransmit it. 2545 * Kludge snd_nxt & the congestion 2546 * window so we send only this one 2547 * packet. 2548 * 2549 * We know we're losing at the current 2550 * window size so do congestion avoidance 2551 * (set ssthresh to half the current window 2552 * and pull our congestion window back to 2553 * the new ssthresh). 2554 * 2555 * Dup acks mean that packets have left the 2556 * network (they're now cached at the receiver) 2557 * so bump cwnd by the amount in the receiver 2558 * to keep a constant cwnd packets in the 2559 * network. 2560 * 2561 * When using TCP ECN, notify the peer that 2562 * we reduced the cwnd. 2563 */ 2564 /* 2565 * Following 2 kinds of acks should not affect 2566 * dupack counting: 2567 * 1) Old acks 2568 * 2) Acks with SACK but without any new SACK 2569 * information in them. These could result from 2570 * any anomaly in the network like a switch 2571 * duplicating packets or a possible DoS attack. 2572 */ 2573 if (th->th_ack != tp->snd_una || 2574 (tcp_is_sack_recovery(tp, &to) && 2575 (sack_changed == SACK_NOCHANGE))) 2576 break; 2577 else if (!tcp_timer_active(tp, TT_REXMT)) 2578 tp->t_dupacks = 0; 2579 else if (++tp->t_dupacks > tcprexmtthresh || 2580 IN_FASTRECOVERY(tp->t_flags)) { 2581 cc_ack_received(tp, th, nsegs, 2582 CC_DUPACK); 2583 if (V_tcp_do_prr && 2584 IN_FASTRECOVERY(tp->t_flags) && 2585 (tp->t_flags & TF_SACK_PERMIT)) { 2586 tcp_do_prr_ack(tp, th, &to, sack_changed); 2587 } else if (tcp_is_sack_recovery(tp, &to) && 2588 IN_FASTRECOVERY(tp->t_flags)) { 2589 int awnd; 2590 2591 /* 2592 * Compute the amount of data in flight first. 2593 * We can inject new data into the pipe iff 2594 * we have less than 1/2 the original window's 2595 * worth of data in flight. 2596 */ 2597 if (V_tcp_do_newsack) 2598 awnd = tcp_compute_pipe(tp); 2599 else 2600 awnd = (tp->snd_nxt - tp->snd_fack) + 2601 tp->sackhint.sack_bytes_rexmit; 2602 2603 if (awnd < tp->snd_ssthresh) { 2604 tp->snd_cwnd += maxseg; 2605 if (tp->snd_cwnd > tp->snd_ssthresh) 2606 tp->snd_cwnd = tp->snd_ssthresh; 2607 } 2608 } else 2609 tp->snd_cwnd += maxseg; 2610 (void) tcp_output(tp); 2611 goto drop; 2612 } else if (tp->t_dupacks == tcprexmtthresh || 2613 (tp->t_flags & TF_SACK_PERMIT && 2614 V_tcp_do_newsack && 2615 tp->sackhint.sacked_bytes > 2616 (tcprexmtthresh - 1) * maxseg)) { 2617 enter_recovery: 2618 /* 2619 * Above is the RFC6675 trigger condition of 2620 * more than (dupthresh-1)*maxseg sacked data. 2621 * If the count of holes in the 2622 * scoreboard is >= dupthresh, we could 2623 * also enter loss recovery, but don't 2624 * have that value readily available. 2625 */ 2626 tp->t_dupacks = tcprexmtthresh; 2627 tcp_seq onxt = tp->snd_nxt; 2628 2629 /* 2630 * If we're doing sack, or prr, check 2631 * to see if we're already in sack 2632 * recovery. If we're not doing sack, 2633 * check to see if we're in newreno 2634 * recovery. 2635 */ 2636 if (V_tcp_do_prr || 2637 (tp->t_flags & TF_SACK_PERMIT)) { 2638 if (IN_FASTRECOVERY(tp->t_flags)) { 2639 tp->t_dupacks = 0; 2640 break; 2641 } 2642 } else { 2643 if (SEQ_LEQ(th->th_ack, 2644 tp->snd_recover)) { 2645 tp->t_dupacks = 0; 2646 break; 2647 } 2648 } 2649 /* Congestion signal before ack. */ 2650 cc_cong_signal(tp, th, CC_NDUPACK); 2651 cc_ack_received(tp, th, nsegs, 2652 CC_DUPACK); 2653 tcp_timer_activate(tp, TT_REXMT, 0); 2654 tp->t_rtttime = 0; 2655 if (V_tcp_do_prr) { 2656 /* 2657 * snd_ssthresh is already updated by 2658 * cc_cong_signal. 2659 */ 2660 if (tcp_is_sack_recovery(tp, &to)) { 2661 /* 2662 * Exclude Limited Transmit 2663 * segments here 2664 */ 2665 tp->sackhint.prr_delivered = 2666 maxseg; 2667 } else { 2668 tp->sackhint.prr_delivered = 2669 imin(tp->snd_max - tp->snd_una, 2670 imin(INT_MAX / 65536, 2671 tp->t_dupacks) * maxseg); 2672 } 2673 tp->sackhint.recover_fs = max(1, 2674 tp->snd_nxt - tp->snd_una); 2675 } 2676 if (tcp_is_sack_recovery(tp, &to)) { 2677 TCPSTAT_INC( 2678 tcps_sack_recovery_episode); 2679 tp->snd_recover = tp->snd_nxt; 2680 tp->snd_cwnd = maxseg; 2681 (void) tcp_output(tp); 2682 if (SEQ_GT(th->th_ack, tp->snd_una)) 2683 goto resume_partialack; 2684 goto drop; 2685 } 2686 tp->snd_nxt = th->th_ack; 2687 tp->snd_cwnd = maxseg; 2688 (void) tcp_output(tp); 2689 KASSERT(tp->snd_limited <= 2, 2690 ("%s: tp->snd_limited too big", 2691 __func__)); 2692 tp->snd_cwnd = tp->snd_ssthresh + 2693 maxseg * 2694 (tp->t_dupacks - tp->snd_limited); 2695 if (SEQ_GT(onxt, tp->snd_nxt)) 2696 tp->snd_nxt = onxt; 2697 goto drop; 2698 } else if (V_tcp_do_rfc3042) { 2699 /* 2700 * Process first and second duplicate 2701 * ACKs. Each indicates a segment 2702 * leaving the network, creating room 2703 * for more. Make sure we can send a 2704 * packet on reception of each duplicate 2705 * ACK by increasing snd_cwnd by one 2706 * segment. Restore the original 2707 * snd_cwnd after packet transmission. 2708 */ 2709 cc_ack_received(tp, th, nsegs, 2710 CC_DUPACK); 2711 uint32_t oldcwnd = tp->snd_cwnd; 2712 tcp_seq oldsndmax = tp->snd_max; 2713 u_int sent; 2714 int avail; 2715 2716 KASSERT(tp->t_dupacks == 1 || 2717 tp->t_dupacks == 2, 2718 ("%s: dupacks not 1 or 2", 2719 __func__)); 2720 if (tp->t_dupacks == 1) 2721 tp->snd_limited = 0; 2722 tp->snd_cwnd = 2723 (tp->snd_nxt - tp->snd_una) + 2724 (tp->t_dupacks - tp->snd_limited) * 2725 maxseg; 2726 /* 2727 * Only call tcp_output when there 2728 * is new data available to be sent 2729 * or we need to send an ACK. 2730 */ 2731 SOCKBUF_LOCK(&so->so_snd); 2732 avail = sbavail(&so->so_snd) - 2733 (tp->snd_nxt - tp->snd_una); 2734 SOCKBUF_UNLOCK(&so->so_snd); 2735 if (avail > 0 || tp->t_flags & TF_ACKNOW) 2736 (void) tcp_output(tp); 2737 sent = tp->snd_max - oldsndmax; 2738 if (sent > maxseg) { 2739 KASSERT((tp->t_dupacks == 2 && 2740 tp->snd_limited == 0) || 2741 (sent == maxseg + 1 && 2742 tp->t_flags & TF_SENTFIN), 2743 ("%s: sent too much", 2744 __func__)); 2745 tp->snd_limited = 2; 2746 } else if (sent > 0) 2747 ++tp->snd_limited; 2748 tp->snd_cwnd = oldcwnd; 2749 goto drop; 2750 } 2751 } 2752 break; 2753 } else { 2754 /* 2755 * This ack is advancing the left edge, reset the 2756 * counter. 2757 */ 2758 tp->t_dupacks = 0; 2759 /* 2760 * If this ack also has new SACK info, increment the 2761 * counter as per rfc6675. The variable 2762 * sack_changed tracks all changes to the SACK 2763 * scoreboard, including when partial ACKs without 2764 * SACK options are received, and clear the scoreboard 2765 * from the left side. Such partial ACKs should not be 2766 * counted as dupacks here. 2767 */ 2768 if (tcp_is_sack_recovery(tp, &to) && 2769 (sack_changed != SACK_NOCHANGE)) { 2770 tp->t_dupacks++; 2771 /* limit overhead by setting maxseg last */ 2772 if (!IN_FASTRECOVERY(tp->t_flags) && 2773 (tp->sackhint.sacked_bytes > 2774 ((tcprexmtthresh - 1) * 2775 (maxseg = tcp_maxseg(tp))))) { 2776 goto enter_recovery; 2777 } 2778 } 2779 } 2780 2781 resume_partialack: 2782 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2783 ("%s: th_ack <= snd_una", __func__)); 2784 2785 /* 2786 * If the congestion window was inflated to account 2787 * for the other side's cached packets, retract it. 2788 */ 2789 if (IN_FASTRECOVERY(tp->t_flags)) { 2790 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2791 if (tp->t_flags & TF_SACK_PERMIT) 2792 if (V_tcp_do_prr && to.to_flags & TOF_SACK) { 2793 tcp_timer_activate(tp, TT_REXMT, 0); 2794 tp->t_rtttime = 0; 2795 tcp_do_prr_ack(tp, th, &to, sack_changed); 2796 tp->t_flags |= TF_ACKNOW; 2797 (void) tcp_output(tp); 2798 } else 2799 tcp_sack_partialack(tp, th); 2800 else 2801 tcp_newreno_partial_ack(tp, th); 2802 } else 2803 cc_post_recovery(tp, th); 2804 } else if (IN_CONGRECOVERY(tp->t_flags)) { 2805 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2806 if (V_tcp_do_prr) { 2807 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th); 2808 tp->snd_fack = th->th_ack; 2809 /* 2810 * During ECN cwnd reduction 2811 * always use PRR-SSRB 2812 */ 2813 tcp_do_prr_ack(tp, th, &to, SACK_CHANGE); 2814 (void) tcp_output(tp); 2815 } 2816 } else 2817 cc_post_recovery(tp, th); 2818 } 2819 /* 2820 * If we reach this point, ACK is not a duplicate, 2821 * i.e., it ACKs something we sent. 2822 */ 2823 if (tp->t_flags & TF_NEEDSYN) { 2824 /* 2825 * T/TCP: Connection was half-synchronized, and our 2826 * SYN has been ACK'd (so connection is now fully 2827 * synchronized). Go to non-starred state, 2828 * increment snd_una for ACK of SYN, and check if 2829 * we can do window scaling. 2830 */ 2831 tp->t_flags &= ~TF_NEEDSYN; 2832 tp->snd_una++; 2833 /* Do window scaling? */ 2834 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2835 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2836 tp->rcv_scale = tp->request_r_scale; 2837 /* Send window already scaled. */ 2838 } 2839 } 2840 2841 process_ACK: 2842 INP_WLOCK_ASSERT(inp); 2843 2844 /* 2845 * Adjust for the SYN bit in sequence space, 2846 * but don't account for it in cwnd calculations. 2847 * This is for the SYN_RECEIVED, non-simultaneous 2848 * SYN case. SYN_SENT and simultaneous SYN are 2849 * treated elsewhere. 2850 */ 2851 if (incforsyn) 2852 tp->snd_una++; 2853 acked = BYTES_THIS_ACK(tp, th); 2854 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2855 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2856 tp->snd_una, th->th_ack, tp, m)); 2857 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2858 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2859 2860 /* 2861 * If we just performed our first retransmit, and the ACK 2862 * arrives within our recovery window, then it was a mistake 2863 * to do the retransmit in the first place. Recover our 2864 * original cwnd and ssthresh, and proceed to transmit where 2865 * we left off. 2866 */ 2867 if (tp->t_rxtshift == 1 && 2868 tp->t_flags & TF_PREVVALID && 2869 tp->t_badrxtwin != 0 && 2870 to.to_flags & TOF_TS && 2871 to.to_tsecr != 0 && 2872 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 2873 cc_cong_signal(tp, th, CC_RTO_ERR); 2874 2875 /* 2876 * If we have a timestamp reply, update smoothed 2877 * round trip time. If no timestamp is present but 2878 * transmit timer is running and timed sequence 2879 * number was acked, update smoothed round trip time. 2880 * Since we now have an rtt measurement, cancel the 2881 * timer backoff (cf., Phil Karn's retransmit alg.). 2882 * Recompute the initial retransmit timer. 2883 * 2884 * Some boxes send broken timestamp replies 2885 * during the SYN+ACK phase, ignore 2886 * timestamps of 0 or we could calculate a 2887 * huge RTT and blow up the retransmit timer. 2888 */ 2889 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2890 uint32_t t; 2891 2892 t = tcp_ts_getticks() - to.to_tsecr; 2893 if (!tp->t_rttlow || tp->t_rttlow > t) 2894 tp->t_rttlow = t; 2895 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2896 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2897 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2898 tp->t_rttlow = ticks - tp->t_rtttime; 2899 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2900 } 2901 2902 SOCKBUF_LOCK(&so->so_snd); 2903 /* 2904 * Clear t_acktime if remote side has ACKd all data in the 2905 * socket buffer and FIN (if applicable). 2906 * Otherwise, update t_acktime if we received a sufficiently 2907 * large ACK. 2908 */ 2909 if ((tp->t_state <= TCPS_CLOSE_WAIT && 2910 acked == sbavail(&so->so_snd)) || 2911 acked > sbavail(&so->so_snd)) 2912 tp->t_acktime = 0; 2913 else if (acked > 1) 2914 tp->t_acktime = ticks; 2915 2916 /* 2917 * If all outstanding data is acked, stop retransmit 2918 * timer and remember to restart (more output or persist). 2919 * If there is more data to be acked, restart retransmit 2920 * timer, using current (possibly backed-off) value. 2921 */ 2922 if (th->th_ack == tp->snd_max) { 2923 tcp_timer_activate(tp, TT_REXMT, 0); 2924 needoutput = 1; 2925 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2926 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); 2927 2928 /* 2929 * If no data (only SYN) was ACK'd, 2930 * skip rest of ACK processing. 2931 */ 2932 if (acked == 0) { 2933 SOCKBUF_UNLOCK(&so->so_snd); 2934 goto step6; 2935 } 2936 2937 /* 2938 * Let the congestion control algorithm update congestion 2939 * control related information. This typically means increasing 2940 * the congestion window. 2941 */ 2942 cc_ack_received(tp, th, nsegs, CC_ACK); 2943 2944 if (acked > sbavail(&so->so_snd)) { 2945 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2946 tp->snd_wnd -= sbavail(&so->so_snd); 2947 else 2948 tp->snd_wnd = 0; 2949 mfree = sbcut_locked(&so->so_snd, 2950 (int)sbavail(&so->so_snd)); 2951 ourfinisacked = 1; 2952 } else { 2953 mfree = sbcut_locked(&so->so_snd, acked); 2954 if (tp->snd_wnd >= (uint32_t) acked) 2955 tp->snd_wnd -= acked; 2956 else 2957 tp->snd_wnd = 0; 2958 ourfinisacked = 0; 2959 } 2960 /* NB: sowwakeup_locked() does an implicit unlock. */ 2961 sowwakeup_locked(so); 2962 m_freem(mfree); 2963 /* Detect una wraparound. */ 2964 if (!IN_RECOVERY(tp->t_flags) && 2965 SEQ_GT(tp->snd_una, tp->snd_recover) && 2966 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2967 tp->snd_recover = th->th_ack - 1; 2968 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2969 if (IN_RECOVERY(tp->t_flags) && 2970 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2971 EXIT_RECOVERY(tp->t_flags); 2972 } 2973 tp->snd_una = th->th_ack; 2974 if (tp->t_flags & TF_SACK_PERMIT) { 2975 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2976 tp->snd_recover = tp->snd_una; 2977 } 2978 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2979 tp->snd_nxt = tp->snd_una; 2980 2981 switch (tp->t_state) { 2982 /* 2983 * In FIN_WAIT_1 STATE in addition to the processing 2984 * for the ESTABLISHED state if our FIN is now acknowledged 2985 * then enter FIN_WAIT_2. 2986 */ 2987 case TCPS_FIN_WAIT_1: 2988 if (ourfinisacked) { 2989 /* 2990 * If we can't receive any more 2991 * data, then closing user can proceed. 2992 * Starting the timer is contrary to the 2993 * specification, but if we don't get a FIN 2994 * we'll hang forever. 2995 */ 2996 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2997 soisdisconnected(so); 2998 tcp_timer_activate(tp, TT_2MSL, 2999 (tcp_fast_finwait2_recycle ? 3000 tcp_finwait2_timeout : 3001 TP_MAXIDLE(tp))); 3002 } 3003 tcp_state_change(tp, TCPS_FIN_WAIT_2); 3004 } 3005 break; 3006 3007 /* 3008 * In CLOSING STATE in addition to the processing for 3009 * the ESTABLISHED state if the ACK acknowledges our FIN 3010 * then enter the TIME-WAIT state, otherwise ignore 3011 * the segment. 3012 */ 3013 case TCPS_CLOSING: 3014 if (ourfinisacked) { 3015 tcp_twstart(tp); 3016 m_freem(m); 3017 return; 3018 } 3019 break; 3020 3021 /* 3022 * In LAST_ACK, we may still be waiting for data to drain 3023 * and/or to be acked, as well as for the ack of our FIN. 3024 * If our FIN is now acknowledged, delete the TCB, 3025 * enter the closed state and return. 3026 */ 3027 case TCPS_LAST_ACK: 3028 if (ourfinisacked) { 3029 tp = tcp_close(tp); 3030 goto drop; 3031 } 3032 break; 3033 } 3034 } 3035 3036 step6: 3037 INP_WLOCK_ASSERT(inp); 3038 3039 /* 3040 * Update window information. 3041 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3042 */ 3043 if ((thflags & TH_ACK) && 3044 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3045 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3046 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3047 /* keep track of pure window updates */ 3048 if (tlen == 0 && 3049 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3050 TCPSTAT_INC(tcps_rcvwinupd); 3051 tp->snd_wnd = tiwin; 3052 tp->snd_wl1 = th->th_seq; 3053 tp->snd_wl2 = th->th_ack; 3054 if (tp->snd_wnd > tp->max_sndwnd) 3055 tp->max_sndwnd = tp->snd_wnd; 3056 needoutput = 1; 3057 } 3058 3059 /* 3060 * Process segments with URG. 3061 */ 3062 if ((thflags & TH_URG) && th->th_urp && 3063 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3064 /* 3065 * This is a kludge, but if we receive and accept 3066 * random urgent pointers, we'll crash in 3067 * soreceive. It's hard to imagine someone 3068 * actually wanting to send this much urgent data. 3069 */ 3070 SOCKBUF_LOCK(&so->so_rcv); 3071 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3072 th->th_urp = 0; /* XXX */ 3073 thflags &= ~TH_URG; /* XXX */ 3074 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3075 goto dodata; /* XXX */ 3076 } 3077 /* 3078 * If this segment advances the known urgent pointer, 3079 * then mark the data stream. This should not happen 3080 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3081 * a FIN has been received from the remote side. 3082 * In these states we ignore the URG. 3083 * 3084 * According to RFC961 (Assigned Protocols), 3085 * the urgent pointer points to the last octet 3086 * of urgent data. We continue, however, 3087 * to consider it to indicate the first octet 3088 * of data past the urgent section as the original 3089 * spec states (in one of two places). 3090 */ 3091 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3092 tp->rcv_up = th->th_seq + th->th_urp; 3093 so->so_oobmark = sbavail(&so->so_rcv) + 3094 (tp->rcv_up - tp->rcv_nxt) - 1; 3095 if (so->so_oobmark == 0) 3096 so->so_rcv.sb_state |= SBS_RCVATMARK; 3097 sohasoutofband(so); 3098 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3099 } 3100 SOCKBUF_UNLOCK(&so->so_rcv); 3101 /* 3102 * Remove out of band data so doesn't get presented to user. 3103 * This can happen independent of advancing the URG pointer, 3104 * but if two URG's are pending at once, some out-of-band 3105 * data may creep in... ick. 3106 */ 3107 if (th->th_urp <= (uint32_t)tlen && 3108 !(so->so_options & SO_OOBINLINE)) { 3109 /* hdr drop is delayed */ 3110 tcp_pulloutofband(so, th, m, drop_hdrlen); 3111 } 3112 } else { 3113 /* 3114 * If no out of band data is expected, 3115 * pull receive urgent pointer along 3116 * with the receive window. 3117 */ 3118 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3119 tp->rcv_up = tp->rcv_nxt; 3120 } 3121 dodata: /* XXX */ 3122 INP_WLOCK_ASSERT(inp); 3123 3124 /* 3125 * Process the segment text, merging it into the TCP sequencing queue, 3126 * and arranging for acknowledgment of receipt if necessary. 3127 * This process logically involves adjusting tp->rcv_wnd as data 3128 * is presented to the user (this happens in tcp_usrreq.c, 3129 * case PRU_RCVD). If a FIN has already been received on this 3130 * connection then we just ignore the text. 3131 */ 3132 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3133 IS_FASTOPEN(tp->t_flags)); 3134 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3135 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3136 tcp_seq save_start = th->th_seq; 3137 tcp_seq save_rnxt = tp->rcv_nxt; 3138 int save_tlen = tlen; 3139 m_adj(m, drop_hdrlen); /* delayed header drop */ 3140 /* 3141 * Insert segment which includes th into TCP reassembly queue 3142 * with control block tp. Set thflags to whether reassembly now 3143 * includes a segment with FIN. This handles the common case 3144 * inline (segment is the next to be received on an established 3145 * connection, and the queue is empty), avoiding linkage into 3146 * and removal from the queue and repetition of various 3147 * conversions. 3148 * Set DELACK for segments received in order, but ack 3149 * immediately when segments are out of order (so 3150 * fast retransmit can work). 3151 */ 3152 if (th->th_seq == tp->rcv_nxt && 3153 SEGQ_EMPTY(tp) && 3154 (TCPS_HAVEESTABLISHED(tp->t_state) || 3155 tfo_syn)) { 3156 if (DELAY_ACK(tp, tlen) || tfo_syn) 3157 tp->t_flags |= TF_DELACK; 3158 else 3159 tp->t_flags |= TF_ACKNOW; 3160 tp->rcv_nxt += tlen; 3161 if (tlen && 3162 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3163 (tp->t_fbyte_in == 0)) { 3164 tp->t_fbyte_in = ticks; 3165 if (tp->t_fbyte_in == 0) 3166 tp->t_fbyte_in = 1; 3167 if (tp->t_fbyte_out && tp->t_fbyte_in) 3168 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3169 } 3170 thflags = tcp_get_flags(th) & TH_FIN; 3171 TCPSTAT_INC(tcps_rcvpack); 3172 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3173 SOCKBUF_LOCK(&so->so_rcv); 3174 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3175 m_freem(m); 3176 else 3177 sbappendstream_locked(&so->so_rcv, m, 0); 3178 tp->t_flags |= TF_WAKESOR; 3179 } else { 3180 /* 3181 * XXX: Due to the header drop above "th" is 3182 * theoretically invalid by now. Fortunately 3183 * m_adj() doesn't actually frees any mbufs 3184 * when trimming from the head. 3185 */ 3186 tcp_seq temp = save_start; 3187 3188 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3189 tp->t_flags |= TF_ACKNOW; 3190 } 3191 if ((tp->t_flags & TF_SACK_PERMIT) && 3192 (save_tlen > 0) && 3193 TCPS_HAVEESTABLISHED(tp->t_state)) { 3194 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3195 /* 3196 * DSACK actually handled in the fastpath 3197 * above. 3198 */ 3199 tcp_update_sack_list(tp, save_start, 3200 save_start + save_tlen); 3201 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3202 if ((tp->rcv_numsacks >= 1) && 3203 (tp->sackblks[0].end == save_start)) { 3204 /* 3205 * Partial overlap, recorded at todrop 3206 * above. 3207 */ 3208 tcp_update_sack_list(tp, 3209 tp->sackblks[0].start, 3210 tp->sackblks[0].end); 3211 } else { 3212 tcp_update_dsack_list(tp, save_start, 3213 save_start + save_tlen); 3214 } 3215 } else if (tlen >= save_tlen) { 3216 /* Update of sackblks. */ 3217 tcp_update_dsack_list(tp, save_start, 3218 save_start + save_tlen); 3219 } else if (tlen > 0) { 3220 tcp_update_dsack_list(tp, save_start, 3221 save_start + tlen); 3222 } 3223 } 3224 tcp_handle_wakeup(tp); 3225 #if 0 3226 /* 3227 * Note the amount of data that peer has sent into 3228 * our window, in order to estimate the sender's 3229 * buffer size. 3230 * XXX: Unused. 3231 */ 3232 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3233 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3234 else 3235 len = so->so_rcv.sb_hiwat; 3236 #endif 3237 } else { 3238 m_freem(m); 3239 thflags &= ~TH_FIN; 3240 } 3241 3242 /* 3243 * If FIN is received ACK the FIN and let the user know 3244 * that the connection is closing. 3245 */ 3246 if (thflags & TH_FIN) { 3247 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3248 /* The socket upcall is handled by socantrcvmore. */ 3249 socantrcvmore(so); 3250 /* 3251 * If connection is half-synchronized 3252 * (ie NEEDSYN flag on) then delay ACK, 3253 * so it may be piggybacked when SYN is sent. 3254 * Otherwise, since we received a FIN then no 3255 * more input can be expected, send ACK now. 3256 */ 3257 if (tp->t_flags & TF_NEEDSYN) 3258 tp->t_flags |= TF_DELACK; 3259 else 3260 tp->t_flags |= TF_ACKNOW; 3261 tp->rcv_nxt++; 3262 } 3263 switch (tp->t_state) { 3264 /* 3265 * In SYN_RECEIVED and ESTABLISHED STATES 3266 * enter the CLOSE_WAIT state. 3267 */ 3268 case TCPS_SYN_RECEIVED: 3269 tp->t_starttime = ticks; 3270 /* FALLTHROUGH */ 3271 case TCPS_ESTABLISHED: 3272 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3273 break; 3274 3275 /* 3276 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3277 * enter the CLOSING state. 3278 */ 3279 case TCPS_FIN_WAIT_1: 3280 tcp_state_change(tp, TCPS_CLOSING); 3281 break; 3282 3283 /* 3284 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3285 * starting the time-wait timer, turning off the other 3286 * standard timers. 3287 */ 3288 case TCPS_FIN_WAIT_2: 3289 tcp_twstart(tp); 3290 return; 3291 } 3292 } 3293 TCP_PROBE3(debug__input, tp, th, m); 3294 3295 /* 3296 * Return any desired output. 3297 */ 3298 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3299 (void) tcp_output(tp); 3300 3301 check_delack: 3302 INP_WLOCK_ASSERT(inp); 3303 3304 if (tp->t_flags & TF_DELACK) { 3305 tp->t_flags &= ~TF_DELACK; 3306 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3307 } 3308 INP_WUNLOCK(inp); 3309 return; 3310 3311 dropafterack: 3312 /* 3313 * Generate an ACK dropping incoming segment if it occupies 3314 * sequence space, where the ACK reflects our state. 3315 * 3316 * We can now skip the test for the RST flag since all 3317 * paths to this code happen after packets containing 3318 * RST have been dropped. 3319 * 3320 * In the SYN-RECEIVED state, don't send an ACK unless the 3321 * segment we received passes the SYN-RECEIVED ACK test. 3322 * If it fails send a RST. This breaks the loop in the 3323 * "LAND" DoS attack, and also prevents an ACK storm 3324 * between two listening ports that have been sent forged 3325 * SYN segments, each with the source address of the other. 3326 */ 3327 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3328 (SEQ_GT(tp->snd_una, th->th_ack) || 3329 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3330 rstreason = BANDLIM_RST_OPENPORT; 3331 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 3332 goto dropwithreset; 3333 } 3334 TCP_PROBE3(debug__input, tp, th, m); 3335 tp->t_flags |= TF_ACKNOW; 3336 (void) tcp_output(tp); 3337 INP_WUNLOCK(inp); 3338 m_freem(m); 3339 return; 3340 3341 dropwithreset: 3342 if (tp != NULL) { 3343 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3344 INP_WUNLOCK(inp); 3345 } else 3346 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3347 return; 3348 3349 drop: 3350 /* 3351 * Drop space held by incoming segment and return. 3352 */ 3353 TCP_PROBE3(debug__input, tp, th, m); 3354 if (tp != NULL) { 3355 INP_WUNLOCK(inp); 3356 } 3357 m_freem(m); 3358 } 3359 3360 /* 3361 * Issue RST and make ACK acceptable to originator of segment. 3362 * The mbuf must still include the original packet header. 3363 * tp may be NULL. 3364 */ 3365 void 3366 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3367 int tlen, int rstreason) 3368 { 3369 #ifdef INET 3370 struct ip *ip; 3371 #endif 3372 #ifdef INET6 3373 struct ip6_hdr *ip6; 3374 #endif 3375 3376 if (tp != NULL) { 3377 INP_LOCK_ASSERT(tptoinpcb(tp)); 3378 } 3379 3380 /* Don't bother if destination was broadcast/multicast. */ 3381 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3382 goto drop; 3383 #ifdef INET6 3384 if (mtod(m, struct ip *)->ip_v == 6) { 3385 ip6 = mtod(m, struct ip6_hdr *); 3386 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3387 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3388 goto drop; 3389 /* IPv6 anycast check is done at tcp6_input() */ 3390 } 3391 #endif 3392 #if defined(INET) && defined(INET6) 3393 else 3394 #endif 3395 #ifdef INET 3396 { 3397 ip = mtod(m, struct ip *); 3398 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3399 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3400 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3401 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3402 goto drop; 3403 } 3404 #endif 3405 3406 /* Perform bandwidth limiting. */ 3407 if (badport_bandlim(rstreason) < 0) 3408 goto drop; 3409 3410 /* tcp_respond consumes the mbuf chain. */ 3411 if (tcp_get_flags(th) & TH_ACK) { 3412 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3413 th->th_ack, TH_RST); 3414 } else { 3415 if (tcp_get_flags(th) & TH_SYN) 3416 tlen++; 3417 if (tcp_get_flags(th) & TH_FIN) 3418 tlen++; 3419 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3420 (tcp_seq)0, TH_RST|TH_ACK); 3421 } 3422 return; 3423 drop: 3424 m_freem(m); 3425 } 3426 3427 /* 3428 * Parse TCP options and place in tcpopt. 3429 */ 3430 void 3431 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3432 { 3433 int opt, optlen; 3434 3435 to->to_flags = 0; 3436 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3437 opt = cp[0]; 3438 if (opt == TCPOPT_EOL) 3439 break; 3440 if (opt == TCPOPT_NOP) 3441 optlen = 1; 3442 else { 3443 if (cnt < 2) 3444 break; 3445 optlen = cp[1]; 3446 if (optlen < 2 || optlen > cnt) 3447 break; 3448 } 3449 switch (opt) { 3450 case TCPOPT_MAXSEG: 3451 if (optlen != TCPOLEN_MAXSEG) 3452 continue; 3453 if (!(flags & TO_SYN)) 3454 continue; 3455 to->to_flags |= TOF_MSS; 3456 bcopy((char *)cp + 2, 3457 (char *)&to->to_mss, sizeof(to->to_mss)); 3458 to->to_mss = ntohs(to->to_mss); 3459 break; 3460 case TCPOPT_WINDOW: 3461 if (optlen != TCPOLEN_WINDOW) 3462 continue; 3463 if (!(flags & TO_SYN)) 3464 continue; 3465 to->to_flags |= TOF_SCALE; 3466 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3467 break; 3468 case TCPOPT_TIMESTAMP: 3469 if (optlen != TCPOLEN_TIMESTAMP) 3470 continue; 3471 to->to_flags |= TOF_TS; 3472 bcopy((char *)cp + 2, 3473 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3474 to->to_tsval = ntohl(to->to_tsval); 3475 bcopy((char *)cp + 6, 3476 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3477 to->to_tsecr = ntohl(to->to_tsecr); 3478 break; 3479 case TCPOPT_SIGNATURE: 3480 /* 3481 * In order to reply to a host which has set the 3482 * TCP_SIGNATURE option in its initial SYN, we have 3483 * to record the fact that the option was observed 3484 * here for the syncache code to perform the correct 3485 * response. 3486 */ 3487 if (optlen != TCPOLEN_SIGNATURE) 3488 continue; 3489 to->to_flags |= TOF_SIGNATURE; 3490 to->to_signature = cp + 2; 3491 break; 3492 case TCPOPT_SACK_PERMITTED: 3493 if (optlen != TCPOLEN_SACK_PERMITTED) 3494 continue; 3495 if (!(flags & TO_SYN)) 3496 continue; 3497 if (!V_tcp_do_sack) 3498 continue; 3499 to->to_flags |= TOF_SACKPERM; 3500 break; 3501 case TCPOPT_SACK: 3502 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3503 continue; 3504 if (flags & TO_SYN) 3505 continue; 3506 to->to_flags |= TOF_SACK; 3507 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3508 to->to_sacks = cp + 2; 3509 TCPSTAT_INC(tcps_sack_rcv_blocks); 3510 break; 3511 case TCPOPT_FAST_OPEN: 3512 /* 3513 * Cookie length validation is performed by the 3514 * server side cookie checking code or the client 3515 * side cookie cache update code. 3516 */ 3517 if (!(flags & TO_SYN)) 3518 continue; 3519 if (!V_tcp_fastopen_client_enable && 3520 !V_tcp_fastopen_server_enable) 3521 continue; 3522 to->to_flags |= TOF_FASTOPEN; 3523 to->to_tfo_len = optlen - 2; 3524 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3525 break; 3526 default: 3527 continue; 3528 } 3529 } 3530 } 3531 3532 /* 3533 * Pull out of band byte out of a segment so 3534 * it doesn't appear in the user's data queue. 3535 * It is still reflected in the segment length for 3536 * sequencing purposes. 3537 */ 3538 void 3539 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3540 int off) 3541 { 3542 int cnt = off + th->th_urp - 1; 3543 3544 while (cnt >= 0) { 3545 if (m->m_len > cnt) { 3546 char *cp = mtod(m, caddr_t) + cnt; 3547 struct tcpcb *tp = sototcpcb(so); 3548 3549 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3550 3551 tp->t_iobc = *cp; 3552 tp->t_oobflags |= TCPOOB_HAVEDATA; 3553 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3554 m->m_len--; 3555 if (m->m_flags & M_PKTHDR) 3556 m->m_pkthdr.len--; 3557 return; 3558 } 3559 cnt -= m->m_len; 3560 m = m->m_next; 3561 if (m == NULL) 3562 break; 3563 } 3564 panic("tcp_pulloutofband"); 3565 } 3566 3567 /* 3568 * Collect new round-trip time estimate 3569 * and update averages and current timeout. 3570 */ 3571 void 3572 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3573 { 3574 int delta; 3575 3576 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3577 3578 TCPSTAT_INC(tcps_rttupdated); 3579 if (tp->t_rttupdated < UCHAR_MAX) 3580 tp->t_rttupdated++; 3581 #ifdef STATS 3582 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3583 imax(0, rtt * 1000 / hz)); 3584 #endif 3585 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3586 /* 3587 * srtt is stored as fixed point with 5 bits after the 3588 * binary point (i.e., scaled by 8). The following magic 3589 * is equivalent to the smoothing algorithm in rfc793 with 3590 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3591 * point). Adjust rtt to origin 0. 3592 */ 3593 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3594 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3595 3596 if ((tp->t_srtt += delta) <= 0) 3597 tp->t_srtt = 1; 3598 3599 /* 3600 * We accumulate a smoothed rtt variance (actually, a 3601 * smoothed mean difference), then set the retransmit 3602 * timer to smoothed rtt + 4 times the smoothed variance. 3603 * rttvar is stored as fixed point with 4 bits after the 3604 * binary point (scaled by 16). The following is 3605 * equivalent to rfc793 smoothing with an alpha of .75 3606 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3607 * rfc793's wired-in beta. 3608 */ 3609 if (delta < 0) 3610 delta = -delta; 3611 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3612 if ((tp->t_rttvar += delta) <= 0) 3613 tp->t_rttvar = 1; 3614 } else { 3615 /* 3616 * No rtt measurement yet - use the unsmoothed rtt. 3617 * Set the variance to half the rtt (so our first 3618 * retransmit happens at 3*rtt). 3619 */ 3620 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3621 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3622 } 3623 tp->t_rtttime = 0; 3624 tp->t_rxtshift = 0; 3625 3626 /* 3627 * the retransmit should happen at rtt + 4 * rttvar. 3628 * Because of the way we do the smoothing, srtt and rttvar 3629 * will each average +1/2 tick of bias. When we compute 3630 * the retransmit timer, we want 1/2 tick of rounding and 3631 * 1 extra tick because of +-1/2 tick uncertainty in the 3632 * firing of the timer. The bias will give us exactly the 3633 * 1.5 tick we need. But, because the bias is 3634 * statistical, we have to test that we don't drop below 3635 * the minimum feasible timer (which is 2 ticks). 3636 */ 3637 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3638 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3639 3640 /* 3641 * We received an ack for a packet that wasn't retransmitted; 3642 * it is probably safe to discard any error indications we've 3643 * received recently. This isn't quite right, but close enough 3644 * for now (a route might have failed after we sent a segment, 3645 * and the return path might not be symmetrical). 3646 */ 3647 tp->t_softerror = 0; 3648 } 3649 3650 /* 3651 * Determine a reasonable value for maxseg size. 3652 * If the route is known, check route for mtu. 3653 * If none, use an mss that can be handled on the outgoing interface 3654 * without forcing IP to fragment. If no route is found, route has no mtu, 3655 * or the destination isn't local, use a default, hopefully conservative 3656 * size (usually 512 or the default IP max size, but no more than the mtu 3657 * of the interface), as we can't discover anything about intervening 3658 * gateways or networks. We also initialize the congestion/slow start 3659 * window to be a single segment if the destination isn't local. 3660 * While looking at the routing entry, we also initialize other path-dependent 3661 * parameters from pre-set or cached values in the routing entry. 3662 * 3663 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3664 * IP options, e.g. IPSEC data, since length of this data may vary, and 3665 * thus it is calculated for every segment separately in tcp_output(). 3666 * 3667 * NOTE that this routine is only called when we process an incoming 3668 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3669 * settings are handled in tcp_mssopt(). 3670 */ 3671 void 3672 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3673 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3674 { 3675 int mss = 0; 3676 uint32_t maxmtu = 0; 3677 struct inpcb *inp = tptoinpcb(tp); 3678 struct hc_metrics_lite metrics; 3679 #ifdef INET6 3680 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3681 size_t min_protoh = isipv6 ? 3682 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3683 sizeof (struct tcpiphdr); 3684 #else 3685 size_t min_protoh = sizeof(struct tcpiphdr); 3686 #endif 3687 3688 INP_WLOCK_ASSERT(inp); 3689 3690 if (tp->t_port) 3691 min_protoh += V_tcp_udp_tunneling_overhead; 3692 if (mtuoffer != -1) { 3693 KASSERT(offer == -1, ("%s: conflict", __func__)); 3694 offer = mtuoffer - min_protoh; 3695 } 3696 3697 /* Initialize. */ 3698 #ifdef INET6 3699 if (isipv6) { 3700 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3701 tp->t_maxseg = V_tcp_v6mssdflt; 3702 } 3703 #endif 3704 #if defined(INET) && defined(INET6) 3705 else 3706 #endif 3707 #ifdef INET 3708 { 3709 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3710 tp->t_maxseg = V_tcp_mssdflt; 3711 } 3712 #endif 3713 3714 /* 3715 * No route to sender, stay with default mss and return. 3716 */ 3717 if (maxmtu == 0) { 3718 /* 3719 * In case we return early we need to initialize metrics 3720 * to a defined state as tcp_hc_get() would do for us 3721 * if there was no cache hit. 3722 */ 3723 if (metricptr != NULL) 3724 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3725 return; 3726 } 3727 3728 /* What have we got? */ 3729 switch (offer) { 3730 case 0: 3731 /* 3732 * Offer == 0 means that there was no MSS on the SYN 3733 * segment, in this case we use tcp_mssdflt as 3734 * already assigned to t_maxseg above. 3735 */ 3736 offer = tp->t_maxseg; 3737 break; 3738 3739 case -1: 3740 /* 3741 * Offer == -1 means that we didn't receive SYN yet. 3742 */ 3743 /* FALLTHROUGH */ 3744 3745 default: 3746 /* 3747 * Prevent DoS attack with too small MSS. Round up 3748 * to at least minmss. 3749 */ 3750 offer = max(offer, V_tcp_minmss); 3751 } 3752 3753 /* 3754 * rmx information is now retrieved from tcp_hostcache. 3755 */ 3756 tcp_hc_get(&inp->inp_inc, &metrics); 3757 if (metricptr != NULL) 3758 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3759 3760 /* 3761 * If there's a discovered mtu in tcp hostcache, use it. 3762 * Else, use the link mtu. 3763 */ 3764 if (metrics.rmx_mtu) 3765 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3766 else { 3767 #ifdef INET6 3768 if (isipv6) { 3769 mss = maxmtu - min_protoh; 3770 if (!V_path_mtu_discovery && 3771 !in6_localaddr(&inp->in6p_faddr)) 3772 mss = min(mss, V_tcp_v6mssdflt); 3773 } 3774 #endif 3775 #if defined(INET) && defined(INET6) 3776 else 3777 #endif 3778 #ifdef INET 3779 { 3780 mss = maxmtu - min_protoh; 3781 if (!V_path_mtu_discovery && 3782 !in_localaddr(inp->inp_faddr)) 3783 mss = min(mss, V_tcp_mssdflt); 3784 } 3785 #endif 3786 /* 3787 * XXX - The above conditional (mss = maxmtu - min_protoh) 3788 * probably violates the TCP spec. 3789 * The problem is that, since we don't know the 3790 * other end's MSS, we are supposed to use a conservative 3791 * default. But, if we do that, then MTU discovery will 3792 * never actually take place, because the conservative 3793 * default is much less than the MTUs typically seen 3794 * on the Internet today. For the moment, we'll sweep 3795 * this under the carpet. 3796 * 3797 * The conservative default might not actually be a problem 3798 * if the only case this occurs is when sending an initial 3799 * SYN with options and data to a host we've never talked 3800 * to before. Then, they will reply with an MSS value which 3801 * will get recorded and the new parameters should get 3802 * recomputed. For Further Study. 3803 */ 3804 } 3805 mss = min(mss, offer); 3806 3807 /* 3808 * Sanity check: make sure that maxseg will be large 3809 * enough to allow some data on segments even if the 3810 * all the option space is used (40bytes). Otherwise 3811 * funny things may happen in tcp_output. 3812 * 3813 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3814 */ 3815 mss = max(mss, 64); 3816 3817 tp->t_maxseg = mss; 3818 } 3819 3820 void 3821 tcp_mss(struct tcpcb *tp, int offer) 3822 { 3823 int mss; 3824 uint32_t bufsize; 3825 struct inpcb *inp = tptoinpcb(tp); 3826 struct socket *so; 3827 struct hc_metrics_lite metrics; 3828 struct tcp_ifcap cap; 3829 3830 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3831 3832 bzero(&cap, sizeof(cap)); 3833 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3834 3835 mss = tp->t_maxseg; 3836 3837 /* 3838 * If there's a pipesize, change the socket buffer to that size, 3839 * don't change if sb_hiwat is different than default (then it 3840 * has been changed on purpose with setsockopt). 3841 * Make the socket buffers an integral number of mss units; 3842 * if the mss is larger than the socket buffer, decrease the mss. 3843 */ 3844 so = inp->inp_socket; 3845 SOCKBUF_LOCK(&so->so_snd); 3846 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3847 bufsize = metrics.rmx_sendpipe; 3848 else 3849 bufsize = so->so_snd.sb_hiwat; 3850 if (bufsize < mss) 3851 mss = bufsize; 3852 else { 3853 bufsize = roundup(bufsize, mss); 3854 if (bufsize > sb_max) 3855 bufsize = sb_max; 3856 if (bufsize > so->so_snd.sb_hiwat) 3857 (void)sbreserve_locked(so, SO_SND, bufsize, NULL); 3858 } 3859 SOCKBUF_UNLOCK(&so->so_snd); 3860 /* 3861 * Sanity check: make sure that maxseg will be large 3862 * enough to allow some data on segments even if the 3863 * all the option space is used (40bytes). Otherwise 3864 * funny things may happen in tcp_output. 3865 * 3866 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3867 */ 3868 tp->t_maxseg = max(mss, 64); 3869 3870 SOCKBUF_LOCK(&so->so_rcv); 3871 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3872 bufsize = metrics.rmx_recvpipe; 3873 else 3874 bufsize = so->so_rcv.sb_hiwat; 3875 if (bufsize > mss) { 3876 bufsize = roundup(bufsize, mss); 3877 if (bufsize > sb_max) 3878 bufsize = sb_max; 3879 if (bufsize > so->so_rcv.sb_hiwat) 3880 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL); 3881 } 3882 SOCKBUF_UNLOCK(&so->so_rcv); 3883 3884 /* Check the interface for TSO capabilities. */ 3885 if (cap.ifcap & CSUM_TSO) { 3886 tp->t_flags |= TF_TSO; 3887 tp->t_tsomax = cap.tsomax; 3888 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3889 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3890 } 3891 } 3892 3893 /* 3894 * Determine the MSS option to send on an outgoing SYN. 3895 */ 3896 int 3897 tcp_mssopt(struct in_conninfo *inc) 3898 { 3899 int mss = 0; 3900 uint32_t thcmtu = 0; 3901 uint32_t maxmtu = 0; 3902 size_t min_protoh; 3903 3904 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3905 3906 #ifdef INET6 3907 if (inc->inc_flags & INC_ISIPV6) { 3908 mss = V_tcp_v6mssdflt; 3909 maxmtu = tcp_maxmtu6(inc, NULL); 3910 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3911 } 3912 #endif 3913 #if defined(INET) && defined(INET6) 3914 else 3915 #endif 3916 #ifdef INET 3917 { 3918 mss = V_tcp_mssdflt; 3919 maxmtu = tcp_maxmtu(inc, NULL); 3920 min_protoh = sizeof(struct tcpiphdr); 3921 } 3922 #endif 3923 #if defined(INET6) || defined(INET) 3924 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3925 #endif 3926 3927 if (maxmtu && thcmtu) 3928 mss = min(maxmtu, thcmtu) - min_protoh; 3929 else if (maxmtu || thcmtu) 3930 mss = max(maxmtu, thcmtu) - min_protoh; 3931 3932 return (mss); 3933 } 3934 3935 void 3936 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, sackstatus_t sack_changed) 3937 { 3938 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3939 int maxseg = tcp_maxseg(tp); 3940 3941 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3942 3943 /* 3944 * Compute the amount of data that this ACK is indicating 3945 * (del_data) and an estimate of how many bytes are in the 3946 * network. 3947 */ 3948 if (tcp_is_sack_recovery(tp, to) || 3949 (IN_CONGRECOVERY(tp->t_flags) && 3950 !IN_FASTRECOVERY(tp->t_flags))) { 3951 del_data = tp->sackhint.delivered_data; 3952 if (V_tcp_do_newsack) 3953 pipe = tcp_compute_pipe(tp); 3954 else 3955 pipe = (tp->snd_nxt - tp->snd_fack) + 3956 tp->sackhint.sack_bytes_rexmit; 3957 } else { 3958 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + 3959 tp->snd_recover - tp->snd_una)) 3960 del_data = maxseg; 3961 pipe = imax(0, tp->snd_max - tp->snd_una - 3962 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); 3963 } 3964 tp->sackhint.prr_delivered += del_data; 3965 /* 3966 * Proportional Rate Reduction 3967 */ 3968 if (pipe >= tp->snd_ssthresh) { 3969 if (tp->sackhint.recover_fs == 0) 3970 tp->sackhint.recover_fs = 3971 imax(1, tp->snd_nxt - tp->snd_una); 3972 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 3973 tp->snd_ssthresh, tp->sackhint.recover_fs) - 3974 tp->sackhint.prr_out + maxseg - 1; 3975 } else { 3976 /* 3977 * PRR 6937bis heuristic: 3978 * - A partial ack without SACK block beneath snd_recover 3979 * indicates further loss. 3980 * - An SACK scoreboard update adding a new hole indicates 3981 * further loss, so be conservative and send at most one 3982 * segment. 3983 * - Prevent ACK splitting attacks, by being conservative 3984 * when no new data is acked. 3985 */ 3986 if ((sack_changed == SACK_NEWLOSS) || (del_data == 0)) 3987 limit = tp->sackhint.prr_delivered - 3988 tp->sackhint.prr_out; 3989 else 3990 limit = imax(tp->sackhint.prr_delivered - 3991 tp->sackhint.prr_out, del_data) + 3992 maxseg; 3993 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 3994 } 3995 snd_cnt = imax(snd_cnt, 0) / maxseg; 3996 /* 3997 * Send snd_cnt new data into the network in response to this ack. 3998 * If there is going to be a SACK retransmission, adjust snd_cwnd 3999 * accordingly. 4000 */ 4001 if (IN_FASTRECOVERY(tp->t_flags)) { 4002 if (tcp_is_sack_recovery(tp, to)) { 4003 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + 4004 tp->sackhint.sack_bytes_rexmit + 4005 (snd_cnt * maxseg); 4006 } else { 4007 tp->snd_cwnd = (tp->snd_max - tp->snd_una) + 4008 (snd_cnt * maxseg); 4009 } 4010 } else if (IN_CONGRECOVERY(tp->t_flags)) 4011 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4012 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); 4013 } 4014 4015 /* 4016 * On a partial ack arrives, force the retransmission of the 4017 * next unacknowledged segment. Do not clear tp->t_dupacks. 4018 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4019 * be started again. 4020 */ 4021 void 4022 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4023 { 4024 tcp_seq onxt = tp->snd_nxt; 4025 uint32_t ocwnd = tp->snd_cwnd; 4026 u_int maxseg = tcp_maxseg(tp); 4027 4028 INP_WLOCK_ASSERT(tptoinpcb(tp)); 4029 4030 tcp_timer_activate(tp, TT_REXMT, 0); 4031 tp->t_rtttime = 0; 4032 tp->snd_nxt = th->th_ack; 4033 /* 4034 * Set snd_cwnd to one segment beyond acknowledged offset. 4035 * (tp->snd_una has not yet been updated when this function is called.) 4036 */ 4037 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4038 tp->t_flags |= TF_ACKNOW; 4039 (void) tcp_output(tp); 4040 tp->snd_cwnd = ocwnd; 4041 if (SEQ_GT(onxt, tp->snd_nxt)) 4042 tp->snd_nxt = onxt; 4043 /* 4044 * Partial window deflation. Relies on fact that tp->snd_una 4045 * not updated yet. 4046 */ 4047 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4048 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4049 else 4050 tp->snd_cwnd = 0; 4051 tp->snd_cwnd += maxseg; 4052 } 4053 4054 int 4055 tcp_compute_pipe(struct tcpcb *tp) 4056 { 4057 if (tp->t_fb->tfb_compute_pipe == NULL) { 4058 return (tp->snd_max - tp->snd_una + 4059 tp->sackhint.sack_bytes_rexmit - 4060 tp->sackhint.sacked_bytes - 4061 tp->sackhint.lost_bytes); 4062 } else { 4063 return((*tp->t_fb->tfb_compute_pipe)(tp)); 4064 } 4065 } 4066 4067 uint32_t 4068 tcp_compute_initwnd(uint32_t maxseg) 4069 { 4070 /* 4071 * Calculate the Initial Window, also used as Restart Window 4072 * 4073 * RFC5681 Section 3.1 specifies the default conservative values. 4074 * RFC3390 specifies slightly more aggressive values. 4075 * RFC6928 increases it to ten segments. 4076 * Support for user specified value for initial flight size. 4077 */ 4078 if (V_tcp_initcwnd_segments) 4079 return min(V_tcp_initcwnd_segments * maxseg, 4080 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4081 else if (V_tcp_do_rfc3390) 4082 return min(4 * maxseg, max(2 * maxseg, 4380)); 4083 else { 4084 /* Per RFC5681 Section 3.1 */ 4085 if (maxseg > 2190) 4086 return (2 * maxseg); 4087 else if (maxseg > 1095) 4088 return (3 * maxseg); 4089 else 4090 return (4 * maxseg); 4091 } 4092 } 4093