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_lrd) = 0; 166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW, 167 &VNET_NAME(tcp_do_lrd), 1, 168 "Perform Lost Retransmission Detection"); 169 170 VNET_DEFINE(int, tcp_do_newcwv) = 0; 171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 172 &VNET_NAME(tcp_do_newcwv), 0, 173 "Enable New Congestion Window Validation per RFC7661"); 174 175 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 177 &VNET_NAME(tcp_do_rfc3042), 0, 178 "Enable RFC 3042 (Limited Transmit)"); 179 180 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 181 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 182 &VNET_NAME(tcp_do_rfc3390), 0, 183 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 184 185 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 186 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 187 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 188 "Slow-start flight size (initial congestion window) in number of segments"); 189 190 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 192 &VNET_NAME(tcp_do_rfc3465), 0, 193 "Enable RFC 3465 (Appropriate Byte Counting)"); 194 195 VNET_DEFINE(int, tcp_abc_l_var) = 2; 196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 197 &VNET_NAME(tcp_abc_l_var), 2, 198 "Cap the max cwnd increment during slow-start to this number of segments"); 199 200 VNET_DEFINE(int, tcp_insecure_syn) = 0; 201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 202 &VNET_NAME(tcp_insecure_syn), 0, 203 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 204 205 VNET_DEFINE(int, tcp_insecure_rst) = 0; 206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 207 &VNET_NAME(tcp_insecure_rst), 0, 208 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 209 210 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 211 #define V_tcp_recvspace VNET(tcp_recvspace) 212 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 213 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 214 215 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 217 &VNET_NAME(tcp_do_autorcvbuf), 0, 218 "Enable automatic receive buffer sizing"); 219 220 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(tcp_autorcvbuf_max), 0, 223 "Max size of automatic receive buffer"); 224 225 VNET_DEFINE(struct inpcbinfo, tcbinfo); 226 227 /* 228 * TCP statistics are stored in an array of counter(9)s, which size matches 229 * size of struct tcpstat. TCP running connection count is a regular array. 230 */ 231 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 232 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 233 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 234 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 235 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 236 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 237 "TCP connection counts by TCP state"); 238 239 /* 240 * Kernel module interface for updating tcpstat. The first argument is an index 241 * into tcpstat treated as an array. 242 */ 243 void 244 kmod_tcpstat_add(int statnum, int val) 245 { 246 247 counter_u64_add(VNET(tcpstat)[statnum], val); 248 } 249 250 /* 251 * Make sure that we only start a SACK loss recovery when 252 * receiving a duplicate ACK with a SACK block, and also 253 * complete SACK loss recovery in case the other end 254 * reneges. 255 */ 256 static bool inline 257 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to) 258 { 259 return ((tp->t_flags & TF_SACK_PERMIT) && 260 ((to->to_flags & TOF_SACK) || 261 (!TAILQ_EMPTY(&tp->snd_holes)))); 262 } 263 264 #ifdef TCP_HHOOK 265 /* 266 * Wrapper for the TCP established input helper hook. 267 */ 268 void 269 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 270 { 271 struct tcp_hhook_data hhook_data; 272 273 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 274 hhook_data.tp = tp; 275 hhook_data.th = th; 276 hhook_data.to = to; 277 278 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 279 &tp->t_osd); 280 } 281 } 282 #endif 283 284 /* 285 * CC wrapper hook functions 286 */ 287 void 288 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 289 uint16_t type) 290 { 291 #ifdef STATS 292 int32_t gput; 293 #endif 294 295 INP_WLOCK_ASSERT(tptoinpcb(tp)); 296 297 tp->t_ccv.nsegs = nsegs; 298 tp->t_ccv.bytes_this_ack = BYTES_THIS_ACK(tp, th); 299 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 300 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 301 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 302 tp->t_ccv.flags |= CCF_CWND_LIMITED; 303 else 304 tp->t_ccv.flags &= ~CCF_CWND_LIMITED; 305 306 if (type == CC_ACK) { 307 #ifdef STATS 308 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 309 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 310 if (!IN_RECOVERY(tp->t_flags)) 311 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 312 tp->t_ccv.bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 313 if ((tp->t_flags & TF_GPUTINPROG) && 314 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 315 /* 316 * Compute goodput in bits per millisecond. 317 */ 318 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) / 319 max(1, tcp_ts_getticks() - tp->gput_ts); 320 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 321 gput); 322 /* 323 * XXXLAS: This is a temporary hack, and should be 324 * chained off VOI_TCP_GPUT when stats(9) grows an API 325 * to deal with chained VOIs. 326 */ 327 if (tp->t_stats_gput_prev > 0) 328 stats_voi_update_abs_s32(tp->t_stats, 329 VOI_TCP_GPUT_ND, 330 ((gput - tp->t_stats_gput_prev) * 100) / 331 tp->t_stats_gput_prev); 332 tp->t_flags &= ~TF_GPUTINPROG; 333 tp->t_stats_gput_prev = gput; 334 } 335 #endif /* STATS */ 336 if (tp->snd_cwnd > tp->snd_ssthresh) { 337 tp->t_bytes_acked += tp->t_ccv.bytes_this_ack; 338 if (tp->t_bytes_acked >= tp->snd_cwnd) { 339 tp->t_bytes_acked -= tp->snd_cwnd; 340 tp->t_ccv.flags |= CCF_ABC_SENTAWND; 341 } 342 } else { 343 tp->t_ccv.flags &= ~CCF_ABC_SENTAWND; 344 tp->t_bytes_acked = 0; 345 } 346 } 347 348 if (CC_ALGO(tp)->ack_received != NULL) { 349 /* XXXLAS: Find a way to live without this */ 350 tp->t_ccv.curack = th->th_ack; 351 CC_ALGO(tp)->ack_received(&tp->t_ccv, type); 352 } 353 #ifdef STATS 354 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 355 #endif 356 } 357 358 void 359 cc_conn_init(struct tcpcb *tp) 360 { 361 struct hc_metrics_lite metrics; 362 struct inpcb *inp = tptoinpcb(tp); 363 u_int maxseg; 364 int rtt; 365 366 INP_WLOCK_ASSERT(inp); 367 368 tcp_hc_get(&inp->inp_inc, &metrics); 369 maxseg = tcp_maxseg(tp); 370 371 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 372 tp->t_srtt = rtt; 373 TCPSTAT_INC(tcps_usedrtt); 374 if (metrics.rmx_rttvar) { 375 tp->t_rttvar = metrics.rmx_rttvar; 376 TCPSTAT_INC(tcps_usedrttvar); 377 } else { 378 /* default variation is +- 1 rtt */ 379 tp->t_rttvar = 380 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 381 } 382 TCPT_RANGESET(tp->t_rxtcur, 383 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 384 tp->t_rttmin, TCPTV_REXMTMAX); 385 } 386 if (metrics.rmx_ssthresh) { 387 /* 388 * There's some sort of gateway or interface 389 * buffer limit on the path. Use this to set 390 * the slow start threshold, but set the 391 * threshold to no less than 2*mss. 392 */ 393 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 394 TCPSTAT_INC(tcps_usedssthresh); 395 } 396 397 /* 398 * Set the initial slow-start flight size. 399 * 400 * If a SYN or SYN/ACK was lost and retransmitted, we have to 401 * reduce the initial CWND to one segment as congestion is likely 402 * requiring us to be cautious. 403 */ 404 if (tp->snd_cwnd == 1) 405 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 406 else 407 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 408 409 if (CC_ALGO(tp)->conn_init != NULL) 410 CC_ALGO(tp)->conn_init(&tp->t_ccv); 411 } 412 413 void inline 414 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 415 { 416 INP_WLOCK_ASSERT(tptoinpcb(tp)); 417 418 #ifdef STATS 419 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 420 #endif 421 422 switch(type) { 423 case CC_NDUPACK: 424 if (!IN_FASTRECOVERY(tp->t_flags)) { 425 tp->snd_recover = tp->snd_max; 426 if (tp->t_flags2 & TF2_ECN_PERMIT) 427 tp->t_flags2 |= TF2_ECN_SND_CWR; 428 } 429 break; 430 case CC_ECN: 431 if (!IN_CONGRECOVERY(tp->t_flags) || 432 /* 433 * Allow ECN reaction on ACK to CWR, if 434 * that data segment was also CE marked. 435 */ 436 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 437 EXIT_CONGRECOVERY(tp->t_flags); 438 TCPSTAT_INC(tcps_ecn_rcwnd); 439 tp->snd_recover = tp->snd_max + 1; 440 if (tp->t_flags2 & TF2_ECN_PERMIT) 441 tp->t_flags2 |= TF2_ECN_SND_CWR; 442 } 443 break; 444 case CC_RTO: 445 tp->t_dupacks = 0; 446 tp->t_bytes_acked = 0; 447 EXIT_RECOVERY(tp->t_flags); 448 if (tp->t_flags2 & TF2_ECN_PERMIT) 449 tp->t_flags2 |= TF2_ECN_SND_CWR; 450 break; 451 case CC_RTO_ERR: 452 TCPSTAT_INC(tcps_sndrexmitbad); 453 /* RTO was unnecessary, so reset everything. */ 454 tp->snd_cwnd = tp->snd_cwnd_prev; 455 tp->snd_ssthresh = tp->snd_ssthresh_prev; 456 tp->snd_recover = tp->snd_recover_prev; 457 if (tp->t_flags & TF_WASFRECOVERY) 458 ENTER_FASTRECOVERY(tp->t_flags); 459 if (tp->t_flags & TF_WASCRECOVERY) 460 ENTER_CONGRECOVERY(tp->t_flags); 461 tp->snd_nxt = tp->snd_max; 462 tp->t_flags &= ~TF_PREVVALID; 463 tp->t_badrxtwin = 0; 464 break; 465 } 466 467 if (CC_ALGO(tp)->cong_signal != NULL) { 468 if (th != NULL) 469 tp->t_ccv.curack = th->th_ack; 470 CC_ALGO(tp)->cong_signal(&tp->t_ccv, type); 471 } 472 } 473 474 void inline 475 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 476 { 477 INP_WLOCK_ASSERT(tptoinpcb(tp)); 478 479 /* XXXLAS: KASSERT that we're in recovery? */ 480 481 if (CC_ALGO(tp)->post_recovery != NULL) { 482 tp->t_ccv.curack = th->th_ack; 483 CC_ALGO(tp)->post_recovery(&tp->t_ccv); 484 } 485 /* XXXLAS: EXIT_RECOVERY ? */ 486 tp->t_bytes_acked = 0; 487 tp->sackhint.delivered_data = 0; 488 tp->sackhint.prr_out = 0; 489 } 490 491 /* 492 * Indicate whether this ack should be delayed. We can delay the ack if 493 * following conditions are met: 494 * - There is no delayed ack timer in progress. 495 * - Our last ack wasn't a 0-sized window. We never want to delay 496 * the ack that opens up a 0-sized window. 497 * - LRO wasn't used for this segment. We make sure by checking that the 498 * segment size is not larger than the MSS. 499 */ 500 #define DELAY_ACK(tp, tlen) \ 501 ((!tcp_timer_active(tp, TT_DELACK) && \ 502 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 503 (tlen <= tp->t_maxseg) && \ 504 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 505 506 void inline 507 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) 508 { 509 INP_WLOCK_ASSERT(tptoinpcb(tp)); 510 511 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 512 switch (iptos & IPTOS_ECN_MASK) { 513 case IPTOS_ECN_CE: 514 tp->t_ccv.flags |= CCF_IPHDR_CE; 515 break; 516 case IPTOS_ECN_ECT0: 517 /* FALLTHROUGH */ 518 case IPTOS_ECN_ECT1: 519 /* FALLTHROUGH */ 520 case IPTOS_ECN_NOTECT: 521 tp->t_ccv.flags &= ~CCF_IPHDR_CE; 522 break; 523 } 524 525 if (flags & TH_CWR) 526 tp->t_ccv.flags |= CCF_TCPHDR_CWR; 527 else 528 tp->t_ccv.flags &= ~CCF_TCPHDR_CWR; 529 530 CC_ALGO(tp)->ecnpkt_handler(&tp->t_ccv); 531 532 if (tp->t_ccv.flags & CCF_ACKNOW) { 533 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 534 tp->t_flags |= TF_ACKNOW; 535 } 536 } 537 } 538 539 void inline 540 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 541 { 542 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos); 543 } 544 545 /* 546 * TCP input handling is split into multiple parts: 547 * tcp6_input is a thin wrapper around tcp_input for the extended 548 * ip6_protox[] call format in ip6_input 549 * tcp_input handles primary segment validation, inpcb lookup and 550 * SYN processing on listen sockets 551 * tcp_do_segment processes the ACK and text of the segment for 552 * establishing, established and closing connections 553 */ 554 #ifdef INET6 555 int 556 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 557 { 558 struct mbuf *m; 559 struct in6_ifaddr *ia6; 560 struct ip6_hdr *ip6; 561 562 m = *mp; 563 if (m->m_len < *offp + sizeof(struct tcphdr)) { 564 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 565 if (m == NULL) { 566 *mp = m; 567 TCPSTAT_INC(tcps_rcvshort); 568 return (IPPROTO_DONE); 569 } 570 } 571 572 /* 573 * draft-itojun-ipv6-tcp-to-anycast 574 * better place to put this in? 575 */ 576 ip6 = mtod(m, struct ip6_hdr *); 577 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 578 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 579 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 580 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 581 *mp = NULL; 582 return (IPPROTO_DONE); 583 } 584 585 *mp = m; 586 return (tcp_input_with_port(mp, offp, proto, port)); 587 } 588 589 int 590 tcp6_input(struct mbuf **mp, int *offp, int proto) 591 { 592 593 return(tcp6_input_with_port(mp, offp, proto, 0)); 594 } 595 #endif /* INET6 */ 596 597 int 598 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 599 { 600 struct mbuf *m = *mp; 601 struct tcphdr *th = NULL; 602 struct ip *ip = NULL; 603 struct inpcb *inp = NULL; 604 struct tcpcb *tp = NULL; 605 struct socket *so = NULL; 606 u_char *optp = NULL; 607 int off0; 608 int optlen = 0; 609 #ifdef INET 610 int len; 611 uint8_t ipttl; 612 #endif 613 int tlen = 0, off; 614 int drop_hdrlen; 615 int thflags; 616 int rstreason = 0; /* For badport_bandlim accounting purposes */ 617 int lookupflag; 618 uint8_t iptos; 619 struct m_tag *fwd_tag = NULL; 620 #ifdef INET6 621 struct ip6_hdr *ip6 = NULL; 622 int isipv6; 623 #else 624 const void *ip6 = NULL; 625 #endif /* INET6 */ 626 struct tcpopt to; /* options in this segment */ 627 char *s = NULL; /* address and port logging */ 628 629 NET_EPOCH_ASSERT(); 630 631 #ifdef INET6 632 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 633 #endif 634 635 off0 = *offp; 636 m = *mp; 637 *mp = NULL; 638 to.to_flags = 0; 639 TCPSTAT_INC(tcps_rcvtotal); 640 641 #ifdef INET6 642 if (isipv6) { 643 ip6 = mtod(m, struct ip6_hdr *); 644 th = (struct tcphdr *)((caddr_t)ip6 + off0); 645 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 646 if (port) 647 goto skip6_csum; 648 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 649 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 650 th->th_sum = m->m_pkthdr.csum_data; 651 else 652 th->th_sum = in6_cksum_pseudo(ip6, tlen, 653 IPPROTO_TCP, m->m_pkthdr.csum_data); 654 th->th_sum ^= 0xffff; 655 } else 656 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 657 if (th->th_sum) { 658 TCPSTAT_INC(tcps_rcvbadsum); 659 goto drop; 660 } 661 skip6_csum: 662 /* 663 * Be proactive about unspecified IPv6 address in source. 664 * As we use all-zero to indicate unbounded/unconnected pcb, 665 * unspecified IPv6 address can be used to confuse us. 666 * 667 * Note that packets with unspecified IPv6 destination is 668 * already dropped in ip6_input. 669 */ 670 KASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_dst), 671 ("%s: unspecified destination v6 address", __func__)); 672 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 673 IP6STAT_INC(ip6s_badscope); /* XXX */ 674 goto drop; 675 } 676 iptos = IPV6_TRAFFIC_CLASS(ip6); 677 } 678 #endif 679 #if defined(INET) && defined(INET6) 680 else 681 #endif 682 #ifdef INET 683 { 684 /* 685 * Get IP and TCP header together in first mbuf. 686 * Note: IP leaves IP header in first mbuf. 687 */ 688 if (off0 > sizeof (struct ip)) { 689 ip_stripoptions(m); 690 off0 = sizeof(struct ip); 691 } 692 if (m->m_len < sizeof (struct tcpiphdr)) { 693 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 694 == NULL) { 695 TCPSTAT_INC(tcps_rcvshort); 696 return (IPPROTO_DONE); 697 } 698 } 699 ip = mtod(m, struct ip *); 700 th = (struct tcphdr *)((caddr_t)ip + off0); 701 tlen = ntohs(ip->ip_len) - off0; 702 703 iptos = ip->ip_tos; 704 if (port) 705 goto skip_csum; 706 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 707 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 708 th->th_sum = m->m_pkthdr.csum_data; 709 else 710 th->th_sum = in_pseudo(ip->ip_src.s_addr, 711 ip->ip_dst.s_addr, 712 htonl(m->m_pkthdr.csum_data + tlen + 713 IPPROTO_TCP)); 714 th->th_sum ^= 0xffff; 715 } else { 716 struct ipovly *ipov = (struct ipovly *)ip; 717 718 /* 719 * Checksum extended TCP header and data. 720 */ 721 len = off0 + tlen; 722 ipttl = ip->ip_ttl; 723 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 724 ipov->ih_len = htons(tlen); 725 th->th_sum = in_cksum(m, len); 726 /* Reset length for SDT probes. */ 727 ip->ip_len = htons(len); 728 /* Reset TOS bits */ 729 ip->ip_tos = iptos; 730 /* Re-initialization for later version check */ 731 ip->ip_ttl = ipttl; 732 ip->ip_v = IPVERSION; 733 ip->ip_hl = off0 >> 2; 734 } 735 skip_csum: 736 if (th->th_sum && (port == 0)) { 737 TCPSTAT_INC(tcps_rcvbadsum); 738 goto drop; 739 } 740 KASSERT(ip->ip_dst.s_addr != INADDR_ANY, 741 ("%s: unspecified destination v4 address", __func__)); 742 if (__predict_false(ip->ip_src.s_addr == INADDR_ANY)) { 743 IPSTAT_INC(ips_badaddr); 744 goto drop; 745 } 746 } 747 #endif /* INET */ 748 749 /* 750 * Check that TCP offset makes sense, 751 * pull out TCP options and adjust length. XXX 752 */ 753 off = th->th_off << 2; 754 if (off < sizeof (struct tcphdr) || off > tlen) { 755 TCPSTAT_INC(tcps_rcvbadoff); 756 goto drop; 757 } 758 tlen -= off; /* tlen is used instead of ti->ti_len */ 759 if (off > sizeof (struct tcphdr)) { 760 #ifdef INET6 761 if (isipv6) { 762 if (m->m_len < off0 + off) { 763 m = m_pullup(m, off0 + off); 764 if (m == NULL) { 765 TCPSTAT_INC(tcps_rcvshort); 766 return (IPPROTO_DONE); 767 } 768 } 769 ip6 = mtod(m, struct ip6_hdr *); 770 th = (struct tcphdr *)((caddr_t)ip6 + off0); 771 } 772 #endif 773 #if defined(INET) && defined(INET6) 774 else 775 #endif 776 #ifdef INET 777 { 778 if (m->m_len < sizeof(struct ip) + off) { 779 if ((m = m_pullup(m, sizeof (struct ip) + off)) 780 == NULL) { 781 TCPSTAT_INC(tcps_rcvshort); 782 return (IPPROTO_DONE); 783 } 784 ip = mtod(m, struct ip *); 785 th = (struct tcphdr *)((caddr_t)ip + off0); 786 } 787 } 788 #endif 789 optlen = off - sizeof (struct tcphdr); 790 optp = (u_char *)(th + 1); 791 } 792 thflags = tcp_get_flags(th); 793 794 /* 795 * Convert TCP protocol specific fields to host format. 796 */ 797 tcp_fields_to_host(th); 798 799 /* 800 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 801 */ 802 drop_hdrlen = off0 + off; 803 804 /* 805 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 806 */ 807 if ( 808 #ifdef INET6 809 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 810 #ifdef INET 811 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 812 #endif 813 #endif 814 #if defined(INET) && !defined(INET6) 815 (m->m_flags & M_IP_NEXTHOP) 816 #endif 817 ) 818 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 819 820 /* 821 * For initial SYN packets we don't need write lock on matching 822 * PCB, be it a listening one or a synchronized one. The packet 823 * shall not modify its state. 824 */ 825 lookupflag = INPLOOKUP_WILDCARD | 826 ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ? 827 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB); 828 findpcb: 829 #ifdef INET6 830 if (isipv6 && fwd_tag != NULL) { 831 struct sockaddr_in6 *next_hop6; 832 833 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 834 /* 835 * Transparently forwarded. Pretend to be the destination. 836 * Already got one like this? 837 */ 838 inp = in6_pcblookup_mbuf(&V_tcbinfo, 839 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 840 lookupflag & ~INPLOOKUP_WILDCARD, m->m_pkthdr.rcvif, m); 841 if (!inp) { 842 /* 843 * It's new. Try to find the ambushing socket. 844 * Because we've rewritten the destination address, 845 * any hardware-generated hash is ignored. 846 */ 847 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 848 th->th_sport, &next_hop6->sin6_addr, 849 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 850 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 851 } 852 } else if (isipv6) { 853 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 854 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag, 855 m->m_pkthdr.rcvif, m); 856 } 857 #endif /* INET6 */ 858 #if defined(INET6) && defined(INET) 859 else 860 #endif 861 #ifdef INET 862 if (fwd_tag != NULL) { 863 struct sockaddr_in *next_hop; 864 865 next_hop = (struct sockaddr_in *)(fwd_tag+1); 866 /* 867 * Transparently forwarded. Pretend to be the destination. 868 * already got one like this? 869 */ 870 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 871 ip->ip_dst, th->th_dport, lookupflag & ~INPLOOKUP_WILDCARD, 872 m->m_pkthdr.rcvif, m); 873 if (!inp) { 874 /* 875 * It's new. Try to find the ambushing socket. 876 * Because we've rewritten the destination address, 877 * any hardware-generated hash is ignored. 878 */ 879 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 880 th->th_sport, next_hop->sin_addr, 881 next_hop->sin_port ? ntohs(next_hop->sin_port) : 882 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 883 } 884 } else 885 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 886 th->th_sport, ip->ip_dst, th->th_dport, lookupflag, 887 m->m_pkthdr.rcvif, m); 888 #endif /* INET */ 889 890 /* 891 * If the INPCB does not exist then all data in the incoming 892 * segment is discarded and an appropriate RST is sent back. 893 * XXX MRT Send RST using which routing table? 894 */ 895 if (inp == NULL) { 896 if (rstreason != 0) { 897 /* We came here after second (safety) lookup. */ 898 MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0); 899 goto dropwithreset; 900 } 901 /* 902 * Log communication attempts to ports that are not 903 * in use. 904 */ 905 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 906 V_tcp_log_in_vain == 2) { 907 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 908 log(LOG_INFO, "%s; %s: Connection attempt " 909 "to closed port\n", s, __func__); 910 } 911 /* 912 * When blackholing do not respond with a RST but 913 * completely ignore the segment and drop it. 914 */ 915 if (((V_blackhole == 1 && (thflags & TH_SYN)) || 916 V_blackhole == 2) && (V_blackhole_local || ( 917 #ifdef INET6 918 isipv6 ? !in6_localaddr(&ip6->ip6_src) : 919 #endif 920 #ifdef INET 921 !in_localip(ip->ip_src) 922 #else 923 true 924 #endif 925 ))) 926 goto dropunlock; 927 928 rstreason = BANDLIM_RST_CLOSEDPORT; 929 goto dropwithreset; 930 } 931 INP_LOCK_ASSERT(inp); 932 933 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 934 !SOLISTENING(inp->inp_socket)) { 935 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) { 936 inp->inp_flowid = m->m_pkthdr.flowid; 937 inp->inp_flowtype = M_HASHTYPE_GET(m); 938 #ifdef RSS 939 } else { 940 /* assign flowid by software RSS hash */ 941 #ifdef INET6 942 if (isipv6) { 943 rss_proto_software_hash_v6(&inp->in6p_faddr, 944 &inp->in6p_laddr, 945 inp->inp_fport, 946 inp->inp_lport, 947 IPPROTO_TCP, 948 &inp->inp_flowid, 949 &inp->inp_flowtype); 950 } else 951 #endif /* INET6 */ 952 { 953 rss_proto_software_hash_v4(inp->inp_faddr, 954 inp->inp_laddr, 955 inp->inp_fport, 956 inp->inp_lport, 957 IPPROTO_TCP, 958 &inp->inp_flowid, 959 &inp->inp_flowtype); 960 } 961 #endif /* RSS */ 962 } 963 } 964 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 965 #ifdef INET6 966 if (isipv6 && IPSEC_ENABLED(ipv6) && 967 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 968 goto dropunlock; 969 } 970 #ifdef INET 971 else 972 #endif 973 #endif /* INET6 */ 974 #ifdef INET 975 if (IPSEC_ENABLED(ipv4) && 976 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 977 goto dropunlock; 978 } 979 #endif /* INET */ 980 #endif /* IPSEC */ 981 982 /* 983 * Check the minimum TTL for socket. 984 */ 985 if (inp->inp_ip_minttl != 0) { 986 #ifdef INET6 987 if (isipv6) { 988 if (inp->inp_ip_minttl > ip6->ip6_hlim) 989 goto dropunlock; 990 } else 991 #endif 992 if (inp->inp_ip_minttl > ip->ip_ttl) 993 goto dropunlock; 994 } 995 996 tp = intotcpcb(inp); 997 switch (tp->t_state) { 998 case TCPS_TIME_WAIT: 999 /* 1000 * A previous connection in TIMEWAIT state is supposed to catch 1001 * stray or duplicate segments arriving late. If this segment 1002 * was a legitimate new connection attempt, the old INPCB gets 1003 * removed and we can try again to find a listening socket. 1004 */ 1005 tcp_dooptions(&to, optp, optlen, 1006 (thflags & TH_SYN) ? TO_SYN : 0); 1007 /* 1008 * tcp_twcheck unlocks the inp always, and frees the m if fails. 1009 */ 1010 if (tcp_twcheck(inp, &to, th, m, tlen)) 1011 goto findpcb; 1012 return (IPPROTO_DONE); 1013 case TCPS_CLOSED: 1014 /* 1015 * The TCPCB may no longer exist if the connection is winding 1016 * down or it is in the CLOSED state. Either way we drop the 1017 * segment and send an appropriate response. 1018 */ 1019 rstreason = BANDLIM_RST_CLOSEDPORT; 1020 goto dropwithreset; 1021 } 1022 1023 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 1024 rstreason = BANDLIM_RST_CLOSEDPORT; 1025 goto dropwithreset; 1026 } 1027 1028 #ifdef TCP_OFFLOAD 1029 if (tp->t_flags & TF_TOE) { 1030 tcp_offload_input(tp, m); 1031 m = NULL; /* consumed by the TOE driver */ 1032 goto dropunlock; 1033 } 1034 #endif 1035 1036 #ifdef MAC 1037 if (mac_inpcb_check_deliver(inp, m)) 1038 goto dropunlock; 1039 #endif 1040 so = inp->inp_socket; 1041 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1042 /* 1043 * When the socket is accepting connections (the INPCB is in LISTEN 1044 * state) we look into the SYN cache if this is a new connection 1045 * attempt or the completion of a previous one. 1046 */ 1047 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), 1048 ("%s: so accepting but tp %p not listening", __func__, tp)); 1049 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { 1050 struct in_conninfo inc; 1051 1052 bzero(&inc, sizeof(inc)); 1053 #ifdef INET6 1054 if (isipv6) { 1055 inc.inc_flags |= INC_ISIPV6; 1056 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1057 inc.inc_flags |= INC_IPV6MINMTU; 1058 inc.inc6_faddr = ip6->ip6_src; 1059 inc.inc6_laddr = ip6->ip6_dst; 1060 } else 1061 #endif 1062 { 1063 inc.inc_faddr = ip->ip_src; 1064 inc.inc_laddr = ip->ip_dst; 1065 } 1066 inc.inc_fport = th->th_sport; 1067 inc.inc_lport = th->th_dport; 1068 inc.inc_fibnum = so->so_fibnum; 1069 1070 /* 1071 * Check for an existing connection attempt in syncache if 1072 * the flag is only ACK. A successful lookup creates a new 1073 * socket appended to the listen queue in SYN_RECEIVED state. 1074 */ 1075 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1076 /* 1077 * Parse the TCP options here because 1078 * syncookies need access to the reflected 1079 * timestamp. 1080 */ 1081 tcp_dooptions(&to, optp, optlen, 0); 1082 /* 1083 * NB: syncache_expand() doesn't unlock inp. 1084 */ 1085 rstreason = syncache_expand(&inc, &to, th, &so, m, port); 1086 if (rstreason < 0) { 1087 /* 1088 * A failing TCP MD5 signature comparison 1089 * must result in the segment being dropped 1090 * and must not produce any response back 1091 * to the sender. 1092 */ 1093 goto dropunlock; 1094 } else if (rstreason == 0) { 1095 /* 1096 * No syncache entry, or ACK was not for our 1097 * SYN/ACK. Do our protection against double 1098 * ACK. If peer sent us 2 ACKs, then for the 1099 * first one syncache_expand() successfully 1100 * converted syncache entry into a socket, 1101 * while we were waiting on the inpcb lock. We 1102 * don't want to sent RST for the second ACK, 1103 * so we perform second lookup without wildcard 1104 * match, hoping to find the new socket. If 1105 * the ACK is stray indeed, rstreason would 1106 * hint the above code that the lookup was a 1107 * second attempt. 1108 * 1109 * NB: syncache did its own logging 1110 * of the failure cause. 1111 */ 1112 INP_WUNLOCK(inp); 1113 rstreason = BANDLIM_RST_OPENPORT; 1114 lookupflag &= ~INPLOOKUP_WILDCARD; 1115 goto findpcb; 1116 } 1117 tfo_socket_result: 1118 if (so == NULL) { 1119 /* 1120 * We completed the 3-way handshake 1121 * but could not allocate a socket 1122 * either due to memory shortage, 1123 * listen queue length limits or 1124 * global socket limits. Send RST 1125 * or wait and have the remote end 1126 * retransmit the ACK for another 1127 * try. 1128 */ 1129 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1130 log(LOG_DEBUG, "%s; %s: Listen socket: " 1131 "Socket allocation failed due to " 1132 "limits or memory shortage, %s\n", 1133 s, __func__, 1134 V_tcp_sc_rst_sock_fail ? 1135 "sending RST" : "try again"); 1136 if (V_tcp_sc_rst_sock_fail) { 1137 rstreason = BANDLIM_UNLIMITED; 1138 goto dropwithreset; 1139 } else 1140 goto dropunlock; 1141 } 1142 /* 1143 * Socket is created in state SYN_RECEIVED. 1144 * Unlock the listen socket, lock the newly 1145 * created socket and update the tp variable. 1146 * If we came here via jump to tfo_socket_result, 1147 * then listening socket is read-locked. 1148 */ 1149 INP_UNLOCK(inp); /* listen socket */ 1150 inp = sotoinpcb(so); 1151 /* 1152 * New connection inpcb is already locked by 1153 * syncache_expand(). 1154 */ 1155 INP_WLOCK_ASSERT(inp); 1156 tp = intotcpcb(inp); 1157 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1158 ("%s: ", __func__)); 1159 /* 1160 * Process the segment and the data it 1161 * contains. tcp_do_segment() consumes 1162 * the mbuf chain and unlocks the inpcb. 1163 */ 1164 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1165 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, 1166 tlen, iptos); 1167 return (IPPROTO_DONE); 1168 } 1169 /* 1170 * Segment flag validation for new connection attempts: 1171 * 1172 * Our (SYN|ACK) response was rejected. 1173 * Check with syncache and remove entry to prevent 1174 * retransmits. 1175 * 1176 * NB: syncache_chkrst does its own logging of failure 1177 * causes. 1178 */ 1179 if (thflags & TH_RST) { 1180 syncache_chkrst(&inc, th, m, port); 1181 goto dropunlock; 1182 } 1183 /* 1184 * We can't do anything without SYN. 1185 */ 1186 if ((thflags & TH_SYN) == 0) { 1187 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1188 log(LOG_DEBUG, "%s; %s: Listen socket: " 1189 "SYN is missing, segment ignored\n", 1190 s, __func__); 1191 TCPSTAT_INC(tcps_badsyn); 1192 goto dropunlock; 1193 } 1194 /* 1195 * (SYN|ACK) is bogus on a listen socket. 1196 */ 1197 if (thflags & TH_ACK) { 1198 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1199 log(LOG_DEBUG, "%s; %s: Listen socket: " 1200 "SYN|ACK invalid, segment rejected\n", 1201 s, __func__); 1202 syncache_badack(&inc, port); /* XXX: Not needed! */ 1203 TCPSTAT_INC(tcps_badsyn); 1204 rstreason = BANDLIM_RST_OPENPORT; 1205 goto dropwithreset; 1206 } 1207 /* 1208 * If the drop_synfin option is enabled, drop all 1209 * segments with both the SYN and FIN bits set. 1210 * This prevents e.g. nmap from identifying the 1211 * TCP/IP stack. 1212 * XXX: Poor reasoning. nmap has other methods 1213 * and is constantly refining its stack detection 1214 * strategies. 1215 * XXX: This is a violation of the TCP specification 1216 * and was used by RFC1644. 1217 */ 1218 if ((thflags & TH_FIN) && V_drop_synfin) { 1219 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1220 log(LOG_DEBUG, "%s; %s: Listen socket: " 1221 "SYN|FIN segment ignored (based on " 1222 "sysctl setting)\n", s, __func__); 1223 TCPSTAT_INC(tcps_badsyn); 1224 goto dropunlock; 1225 } 1226 /* 1227 * Segment's flags are (SYN) or (SYN|FIN). 1228 * 1229 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1230 * as they do not affect the state of the TCP FSM. 1231 * The data pointed to by TH_URG and th_urp is ignored. 1232 */ 1233 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1234 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1235 KASSERT(thflags & (TH_SYN), 1236 ("%s: Listen socket: TH_SYN not set", __func__)); 1237 INP_RLOCK_ASSERT(inp); 1238 #ifdef INET6 1239 /* 1240 * If deprecated address is forbidden, 1241 * we do not accept SYN to deprecated interface 1242 * address to prevent any new inbound connection from 1243 * getting established. 1244 * When we do not accept SYN, we send a TCP RST, 1245 * with deprecated source address (instead of dropping 1246 * it). We compromise it as it is much better for peer 1247 * to send a RST, and RST will be the final packet 1248 * for the exchange. 1249 * 1250 * If we do not forbid deprecated addresses, we accept 1251 * the SYN packet. RFC2462 does not suggest dropping 1252 * SYN in this case. 1253 * If we decipher RFC2462 5.5.4, it says like this: 1254 * 1. use of deprecated addr with existing 1255 * communication is okay - "SHOULD continue to be 1256 * used" 1257 * 2. use of it with new communication: 1258 * (2a) "SHOULD NOT be used if alternate address 1259 * with sufficient scope is available" 1260 * (2b) nothing mentioned otherwise. 1261 * Here we fall into (2b) case as we have no choice in 1262 * our source address selection - we must obey the peer. 1263 * 1264 * The wording in RFC2462 is confusing, and there are 1265 * multiple description text for deprecated address 1266 * handling - worse, they are not exactly the same. 1267 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1268 */ 1269 if (isipv6 && !V_ip6_use_deprecated) { 1270 struct in6_ifaddr *ia6; 1271 1272 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1273 if (ia6 != NULL && 1274 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1275 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1276 log(LOG_DEBUG, "%s; %s: Listen socket: " 1277 "Connection attempt to deprecated " 1278 "IPv6 address rejected\n", 1279 s, __func__); 1280 rstreason = BANDLIM_RST_OPENPORT; 1281 goto dropwithreset; 1282 } 1283 } 1284 #endif /* INET6 */ 1285 /* 1286 * Basic sanity checks on incoming SYN requests: 1287 * Don't respond if the destination is a link layer 1288 * broadcast according to RFC1122 4.2.3.10, p. 104. 1289 * If it is from this socket it must be forged. 1290 * Don't respond if the source or destination is a 1291 * global or subnet broad- or multicast address. 1292 * Note that it is quite possible to receive unicast 1293 * link-layer packets with a broadcast IP address. Use 1294 * in_broadcast() to find them. 1295 */ 1296 if (m->m_flags & (M_BCAST|M_MCAST)) { 1297 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1298 log(LOG_DEBUG, "%s; %s: Listen socket: " 1299 "Connection attempt from broad- or multicast " 1300 "link layer address ignored\n", s, __func__); 1301 goto dropunlock; 1302 } 1303 #ifdef INET6 1304 if (isipv6) { 1305 if (th->th_dport == th->th_sport && 1306 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1307 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1308 log(LOG_DEBUG, "%s; %s: Listen socket: " 1309 "Connection attempt to/from self " 1310 "ignored\n", s, __func__); 1311 goto dropunlock; 1312 } 1313 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1314 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1315 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1316 log(LOG_DEBUG, "%s; %s: Listen socket: " 1317 "Connection attempt from/to multicast " 1318 "address ignored\n", s, __func__); 1319 goto dropunlock; 1320 } 1321 } 1322 #endif 1323 #if defined(INET) && defined(INET6) 1324 else 1325 #endif 1326 #ifdef INET 1327 { 1328 if (th->th_dport == th->th_sport && 1329 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1330 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1331 log(LOG_DEBUG, "%s; %s: Listen socket: " 1332 "Connection attempt from/to self " 1333 "ignored\n", s, __func__); 1334 goto dropunlock; 1335 } 1336 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1337 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1338 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1339 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1340 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1341 log(LOG_DEBUG, "%s; %s: Listen socket: " 1342 "Connection attempt from/to broad- " 1343 "or multicast address ignored\n", 1344 s, __func__); 1345 goto dropunlock; 1346 } 1347 } 1348 #endif 1349 /* 1350 * SYN appears to be valid. Create compressed TCP state 1351 * for syncache. 1352 */ 1353 TCP_PROBE3(debug__input, tp, th, m); 1354 tcp_dooptions(&to, optp, optlen, TO_SYN); 1355 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, 1356 iptos, port)) != NULL) 1357 goto tfo_socket_result; 1358 1359 /* 1360 * Entry added to syncache and mbuf consumed. 1361 * Only the listen socket is unlocked by syncache_add(). 1362 */ 1363 return (IPPROTO_DONE); 1364 } else if (tp->t_state == TCPS_LISTEN) { 1365 /* 1366 * When a listen socket is torn down the SO_ACCEPTCONN 1367 * flag is removed first while connections are drained 1368 * from the accept queue in a unlock/lock cycle of the 1369 * ACCEPT_LOCK, opening a race condition allowing a SYN 1370 * attempt go through unhandled. 1371 */ 1372 goto dropunlock; 1373 } 1374 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1375 if (tp->t_flags & TF_SIGNATURE) { 1376 tcp_dooptions(&to, optp, optlen, thflags); 1377 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1378 TCPSTAT_INC(tcps_sig_err_nosigopt); 1379 goto dropunlock; 1380 } 1381 if (!TCPMD5_ENABLED() || 1382 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1383 goto dropunlock; 1384 } 1385 #endif 1386 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1387 1388 /* 1389 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1390 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1391 * the inpcb, and unlocks pcbinfo. 1392 * 1393 * XXXGL: in case of a pure SYN arriving on existing connection 1394 * TCP stacks won't need to modify the PCB, they would either drop 1395 * the segment silently, or send a challenge ACK. However, we try 1396 * to upgrade the lock, because calling convention for stacks is 1397 * write-lock on PCB. If upgrade fails, drop the SYN. 1398 */ 1399 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0) 1400 goto dropunlock; 1401 1402 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, tlen, iptos); 1403 return (IPPROTO_DONE); 1404 1405 dropwithreset: 1406 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1407 1408 if (inp != NULL) { 1409 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1410 INP_UNLOCK(inp); 1411 } else 1412 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1413 m = NULL; /* mbuf chain got consumed. */ 1414 goto drop; 1415 1416 dropunlock: 1417 if (m != NULL) 1418 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1419 1420 if (inp != NULL) 1421 INP_UNLOCK(inp); 1422 1423 drop: 1424 if (s != NULL) 1425 free(s, M_TCPLOG); 1426 if (m != NULL) 1427 m_freem(m); 1428 return (IPPROTO_DONE); 1429 } 1430 1431 /* 1432 * Automatic sizing of receive socket buffer. Often the send 1433 * buffer size is not optimally adjusted to the actual network 1434 * conditions at hand (delay bandwidth product). Setting the 1435 * buffer size too small limits throughput on links with high 1436 * bandwidth and high delay (eg. trans-continental/oceanic links). 1437 * 1438 * On the receive side the socket buffer memory is only rarely 1439 * used to any significant extent. This allows us to be much 1440 * more aggressive in scaling the receive socket buffer. For 1441 * the case that the buffer space is actually used to a large 1442 * extent and we run out of kernel memory we can simply drop 1443 * the new segments; TCP on the sender will just retransmit it 1444 * later. Setting the buffer size too big may only consume too 1445 * much kernel memory if the application doesn't read() from 1446 * the socket or packet loss or reordering makes use of the 1447 * reassembly queue. 1448 * 1449 * The criteria to step up the receive buffer one notch are: 1450 * 1. Application has not set receive buffer size with 1451 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1452 * 2. the number of bytes received during 1/2 of an sRTT 1453 * is at least 3/8 of the current socket buffer size. 1454 * 3. receive buffer size has not hit maximal automatic size; 1455 * 1456 * If all of the criteria are met we increaset the socket buffer 1457 * by a 1/2 (bounded by the max). This allows us to keep ahead 1458 * of slow-start but also makes it so our peer never gets limited 1459 * by our rwnd which we then open up causing a burst. 1460 * 1461 * This algorithm does two steps per RTT at most and only if 1462 * we receive a bulk stream w/o packet losses or reorderings. 1463 * Shrinking the buffer during idle times is not necessary as 1464 * it doesn't consume any memory when idle. 1465 * 1466 * TODO: Only step up if the application is actually serving 1467 * the buffer to better manage the socket buffer resources. 1468 */ 1469 int 1470 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1471 struct tcpcb *tp, int tlen) 1472 { 1473 int newsize = 0; 1474 1475 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1476 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1477 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1478 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1479 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1480 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1481 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1482 } 1483 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1484 1485 /* Start over with next RTT. */ 1486 tp->rfbuf_ts = 0; 1487 tp->rfbuf_cnt = 0; 1488 } else { 1489 tp->rfbuf_cnt += tlen; /* add up */ 1490 } 1491 return (newsize); 1492 } 1493 1494 int 1495 tcp_input(struct mbuf **mp, int *offp, int proto) 1496 { 1497 return(tcp_input_with_port(mp, offp, proto, 0)); 1498 } 1499 1500 static void 1501 tcp_handle_wakeup(struct tcpcb *tp) 1502 { 1503 1504 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1505 1506 if (tp->t_flags & TF_WAKESOR) { 1507 struct socket *so = tptosocket(tp); 1508 1509 tp->t_flags &= ~TF_WAKESOR; 1510 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1511 sorwakeup_locked(so); 1512 } 1513 } 1514 1515 void 1516 tcp_do_segment(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th, 1517 int drop_hdrlen, int tlen, uint8_t iptos) 1518 { 1519 uint16_t thflags; 1520 int acked, ourfinisacked, needoutput = 0; 1521 sackstatus_t sack_changed; 1522 int rstreason, todrop, win, incforsyn = 0; 1523 uint32_t tiwin; 1524 uint16_t nsegs; 1525 char *s; 1526 struct inpcb *inp = tptoinpcb(tp); 1527 struct socket *so = tptosocket(tp); 1528 struct in_conninfo *inc = &inp->inp_inc; 1529 struct mbuf *mfree; 1530 struct tcpopt to; 1531 int tfo_syn; 1532 u_int maxseg; 1533 1534 thflags = tcp_get_flags(th); 1535 tp->sackhint.last_sack_ack = 0; 1536 sack_changed = SACK_NOCHANGE; 1537 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1538 1539 NET_EPOCH_ASSERT(); 1540 INP_WLOCK_ASSERT(inp); 1541 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1542 __func__)); 1543 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1544 __func__)); 1545 1546 #ifdef TCPPCAP 1547 /* Save segment, if requested. */ 1548 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1549 #endif 1550 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1551 tlen, NULL, true); 1552 1553 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1554 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1555 log(LOG_DEBUG, "%s; %s: " 1556 "SYN|FIN segment ignored (based on " 1557 "sysctl setting)\n", s, __func__); 1558 free(s, M_TCPLOG); 1559 } 1560 goto drop; 1561 } 1562 1563 /* 1564 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1565 * check SEQ.ACK first. 1566 */ 1567 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1568 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1569 rstreason = BANDLIM_UNLIMITED; 1570 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1571 goto dropwithreset; 1572 } 1573 1574 /* 1575 * Segment received on connection. 1576 * Reset idle time and keep-alive timer. 1577 * XXX: This should be done after segment 1578 * validation to ignore broken/spoofed segs. 1579 */ 1580 if (tp->t_idle_reduce && 1581 (tp->snd_max == tp->snd_una) && 1582 ((ticks - tp->t_rcvtime) >= tp->t_rxtcur)) 1583 cc_after_idle(tp); 1584 tp->t_rcvtime = ticks; 1585 1586 if (thflags & TH_FIN) 1587 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN); 1588 /* 1589 * Scale up the window into a 32-bit value. 1590 * For the SYN_SENT state the scale is zero. 1591 */ 1592 tiwin = th->th_win << tp->snd_scale; 1593 #ifdef STATS 1594 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1595 #endif 1596 1597 /* 1598 * TCP ECN processing. 1599 */ 1600 if (tcp_ecn_input_segment(tp, thflags, tlen, 1601 tcp_packets_this_ack(tp, th->th_ack), 1602 iptos)) 1603 cc_cong_signal(tp, th, CC_ECN); 1604 1605 /* 1606 * Parse options on any incoming segment. 1607 */ 1608 tcp_dooptions(&to, (u_char *)(th + 1), 1609 (th->th_off << 2) - sizeof(struct tcphdr), 1610 (thflags & TH_SYN) ? TO_SYN : 0); 1611 1612 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1613 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1614 (to.to_flags & TOF_SIGNATURE) == 0) { 1615 TCPSTAT_INC(tcps_sig_err_sigopt); 1616 /* XXX: should drop? */ 1617 } 1618 #endif 1619 /* 1620 * If echoed timestamp is later than the current time, 1621 * fall back to non RFC1323 RTT calculation. Normalize 1622 * timestamp if syncookies were used when this connection 1623 * was established. 1624 */ 1625 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1626 to.to_tsecr -= tp->ts_offset; 1627 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1628 to.to_tsecr = 0; 1629 else if (tp->t_rxtshift == 1 && 1630 tp->t_flags & TF_PREVVALID && 1631 tp->t_badrxtwin != 0 && 1632 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 1633 cc_cong_signal(tp, th, CC_RTO_ERR); 1634 } 1635 /* 1636 * Process options only when we get SYN/ACK back. The SYN case 1637 * for incoming connections is handled in tcp_syncache. 1638 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1639 * or <SYN,ACK>) segment itself is never scaled. 1640 * XXX this is traditional behavior, may need to be cleaned up. 1641 */ 1642 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1643 /* Handle parallel SYN for ECN */ 1644 tcp_ecn_input_parallel_syn(tp, thflags, iptos); 1645 if ((to.to_flags & TOF_SCALE) && 1646 (tp->t_flags & TF_REQ_SCALE) && 1647 !(tp->t_flags & TF_NOOPT)) { 1648 tp->t_flags |= TF_RCVD_SCALE; 1649 tp->snd_scale = to.to_wscale; 1650 } else 1651 tp->t_flags &= ~TF_REQ_SCALE; 1652 /* 1653 * Initial send window. It will be updated with 1654 * the next incoming segment to the scaled value. 1655 */ 1656 tp->snd_wnd = th->th_win; 1657 if ((to.to_flags & TOF_TS) && 1658 (tp->t_flags & TF_REQ_TSTMP) && 1659 !(tp->t_flags & TF_NOOPT)) { 1660 tp->t_flags |= TF_RCVD_TSTMP; 1661 tp->ts_recent = to.to_tsval; 1662 tp->ts_recent_age = tcp_ts_getticks(); 1663 } else 1664 tp->t_flags &= ~TF_REQ_TSTMP; 1665 if (to.to_flags & TOF_MSS) 1666 tcp_mss(tp, to.to_mss); 1667 if ((tp->t_flags & TF_SACK_PERMIT) && 1668 (!(to.to_flags & TOF_SACKPERM) || 1669 (tp->t_flags & TF_NOOPT))) 1670 tp->t_flags &= ~TF_SACK_PERMIT; 1671 if (IS_FASTOPEN(tp->t_flags)) { 1672 if ((to.to_flags & TOF_FASTOPEN) && 1673 !(tp->t_flags & TF_NOOPT)) { 1674 uint16_t mss; 1675 1676 if (to.to_flags & TOF_MSS) 1677 mss = to.to_mss; 1678 else 1679 if ((inp->inp_vflag & INP_IPV6) != 0) 1680 mss = TCP6_MSS; 1681 else 1682 mss = TCP_MSS; 1683 tcp_fastopen_update_cache(tp, mss, 1684 to.to_tfo_len, to.to_tfo_cookie); 1685 } else 1686 tcp_fastopen_disable_path(tp); 1687 } 1688 } 1689 1690 /* 1691 * If timestamps were negotiated during SYN/ACK and a 1692 * segment without a timestamp is received, silently drop 1693 * the segment, unless it is a RST segment or missing timestamps are 1694 * tolerated. 1695 * See section 3.2 of RFC 7323. 1696 */ 1697 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1698 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1699 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1700 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1701 "segment processed normally\n", 1702 s, __func__); 1703 free(s, M_TCPLOG); 1704 } 1705 } else { 1706 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1707 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1708 "segment silently dropped\n", s, __func__); 1709 free(s, M_TCPLOG); 1710 } 1711 goto drop; 1712 } 1713 } 1714 /* 1715 * If timestamps were not negotiated during SYN/ACK and a 1716 * segment with a timestamp is received, ignore the 1717 * timestamp and process the packet normally. 1718 * See section 3.2 of RFC 7323. 1719 */ 1720 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1721 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1722 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1723 "segment processed normally\n", s, __func__); 1724 free(s, M_TCPLOG); 1725 } 1726 } 1727 1728 /* 1729 * Header prediction: check for the two common cases 1730 * of a uni-directional data xfer. If the packet has 1731 * no control flags, is in-sequence, the window didn't 1732 * change and we're not retransmitting, it's a 1733 * candidate. If the length is zero and the ack moved 1734 * forward, we're the sender side of the xfer. Just 1735 * free the data acked & wake any higher level process 1736 * that was blocked waiting for space. If the length 1737 * is non-zero and the ack didn't move, we're the 1738 * receiver side. If we're getting packets in-order 1739 * (the reassembly queue is empty), add the data to 1740 * the socket buffer and note that we need a delayed ack. 1741 * Make sure that the hidden state-flags are also off. 1742 * Since we check for TCPS_ESTABLISHED first, it can only 1743 * be TH_NEEDSYN. 1744 */ 1745 if (tp->t_state == TCPS_ESTABLISHED && 1746 th->th_seq == tp->rcv_nxt && 1747 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1748 tp->snd_nxt == tp->snd_max && 1749 tiwin && tiwin == tp->snd_wnd && 1750 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1751 SEGQ_EMPTY(tp) && 1752 ((to.to_flags & TOF_TS) == 0 || 1753 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1754 /* 1755 * If last ACK falls within this segment's sequence numbers, 1756 * record the timestamp. 1757 * NOTE that the test is modified according to the latest 1758 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1759 */ 1760 if ((to.to_flags & TOF_TS) != 0 && 1761 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1762 tp->ts_recent_age = tcp_ts_getticks(); 1763 tp->ts_recent = to.to_tsval; 1764 } 1765 1766 if (tlen == 0) { 1767 if (SEQ_GT(th->th_ack, tp->snd_una) && 1768 SEQ_LEQ(th->th_ack, tp->snd_max) && 1769 !IN_RECOVERY(tp->t_flags) && 1770 (to.to_flags & TOF_SACK) == 0 && 1771 TAILQ_EMPTY(&tp->snd_holes)) { 1772 /* 1773 * This is a pure ack for outstanding data. 1774 */ 1775 TCPSTAT_INC(tcps_predack); 1776 1777 /* 1778 * "bad retransmit" recovery without timestamps. 1779 */ 1780 if ((to.to_flags & TOF_TS) == 0 && 1781 tp->t_rxtshift == 1 && 1782 tp->t_flags & TF_PREVVALID && 1783 tp->t_badrxtwin != 0 && 1784 TSTMP_LT(ticks, tp->t_badrxtwin)) { 1785 cc_cong_signal(tp, th, CC_RTO_ERR); 1786 } 1787 1788 /* 1789 * Recalculate the transmit timer / rtt. 1790 * 1791 * Some boxes send broken timestamp replies 1792 * during the SYN+ACK phase, ignore 1793 * timestamps of 0 or we could calculate a 1794 * huge RTT and blow up the retransmit timer. 1795 */ 1796 if ((to.to_flags & TOF_TS) != 0 && 1797 to.to_tsecr) { 1798 uint32_t t; 1799 1800 t = tcp_ts_getticks() - to.to_tsecr; 1801 if (!tp->t_rttlow || tp->t_rttlow > t) 1802 tp->t_rttlow = t; 1803 tcp_xmit_timer(tp, 1804 TCP_TS_TO_TICKS(t) + 1); 1805 } else if (tp->t_rtttime && 1806 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1807 if (!tp->t_rttlow || 1808 tp->t_rttlow > ticks - tp->t_rtttime) 1809 tp->t_rttlow = ticks - tp->t_rtttime; 1810 tcp_xmit_timer(tp, 1811 ticks - tp->t_rtttime); 1812 } 1813 acked = BYTES_THIS_ACK(tp, th); 1814 1815 #ifdef TCP_HHOOK 1816 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1817 hhook_run_tcp_est_in(tp, th, &to); 1818 #endif 1819 1820 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1821 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1822 sbdrop(&so->so_snd, acked); 1823 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1824 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1825 tp->snd_recover = th->th_ack - 1; 1826 1827 /* 1828 * Let the congestion control algorithm update 1829 * congestion control related information. This 1830 * typically means increasing the congestion 1831 * window. 1832 */ 1833 cc_ack_received(tp, th, nsegs, CC_ACK); 1834 1835 tp->snd_una = th->th_ack; 1836 /* 1837 * Pull snd_wl2 up to prevent seq wrap relative 1838 * to th_ack. 1839 */ 1840 tp->snd_wl2 = th->th_ack; 1841 tp->t_dupacks = 0; 1842 m_freem(m); 1843 1844 /* 1845 * If all outstanding data are acked, stop 1846 * retransmit timer, otherwise restart timer 1847 * using current (possibly backed-off) value. 1848 * If process is waiting for space, 1849 * wakeup/selwakeup/signal. If data 1850 * are ready to send, let tcp_output 1851 * decide between more output or persist. 1852 */ 1853 TCP_PROBE3(debug__input, tp, th, m); 1854 /* 1855 * Clear t_acktime if remote side has ACKd 1856 * all data in the socket buffer. 1857 * Otherwise, update t_acktime if we received 1858 * a sufficiently large ACK. 1859 */ 1860 if (sbavail(&so->so_snd) == 0) 1861 tp->t_acktime = 0; 1862 else if (acked > 1) 1863 tp->t_acktime = ticks; 1864 if (tp->snd_una == tp->snd_max) 1865 tcp_timer_activate(tp, TT_REXMT, 0); 1866 else if (!tcp_timer_active(tp, TT_PERSIST)) 1867 tcp_timer_activate(tp, TT_REXMT, 1868 TP_RXTCUR(tp)); 1869 sowwakeup(so); 1870 if (sbavail(&so->so_snd)) 1871 (void) tcp_output(tp); 1872 goto check_delack; 1873 } 1874 } else if (th->th_ack == tp->snd_una && 1875 tlen <= sbspace(&so->so_rcv)) { 1876 int newsize = 0; /* automatic sockbuf scaling */ 1877 1878 /* 1879 * This is a pure, in-sequence data packet with 1880 * nothing on the reassembly queue and we have enough 1881 * buffer space to take it. 1882 */ 1883 /* Clean receiver SACK report if present */ 1884 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1885 tcp_clean_sackreport(tp); 1886 TCPSTAT_INC(tcps_preddat); 1887 tp->rcv_nxt += tlen; 1888 if (tlen && 1889 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1890 (tp->t_fbyte_in == 0)) { 1891 tp->t_fbyte_in = ticks; 1892 if (tp->t_fbyte_in == 0) 1893 tp->t_fbyte_in = 1; 1894 if (tp->t_fbyte_out && tp->t_fbyte_in) 1895 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1896 } 1897 /* 1898 * Pull snd_wl1 up to prevent seq wrap relative to 1899 * th_seq. 1900 */ 1901 tp->snd_wl1 = th->th_seq; 1902 /* 1903 * Pull rcv_up up to prevent seq wrap relative to 1904 * rcv_nxt. 1905 */ 1906 tp->rcv_up = tp->rcv_nxt; 1907 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1908 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1909 TCP_PROBE3(debug__input, tp, th, m); 1910 1911 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1912 1913 /* Add data to socket buffer. */ 1914 SOCKBUF_LOCK(&so->so_rcv); 1915 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1916 m_freem(m); 1917 } else { 1918 /* 1919 * Set new socket buffer size. 1920 * Give up when limit is reached. 1921 */ 1922 if (newsize) 1923 if (!sbreserve_locked(so, SO_RCV, 1924 newsize, NULL)) 1925 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1926 m_adj(m, drop_hdrlen); /* delayed header drop */ 1927 sbappendstream_locked(&so->so_rcv, m, 0); 1928 } 1929 /* NB: sorwakeup_locked() does an implicit unlock. */ 1930 sorwakeup_locked(so); 1931 if (DELAY_ACK(tp, tlen)) { 1932 tp->t_flags |= TF_DELACK; 1933 } else { 1934 tp->t_flags |= TF_ACKNOW; 1935 tcp_output(tp); 1936 } 1937 goto check_delack; 1938 } 1939 } 1940 1941 /* 1942 * Calculate amount of space in receive window, 1943 * and then do TCP input processing. 1944 * Receive window is amount of space in rcv queue, 1945 * but not less than advertised window. 1946 */ 1947 win = sbspace(&so->so_rcv); 1948 if (win < 0) 1949 win = 0; 1950 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1951 1952 switch (tp->t_state) { 1953 /* 1954 * If the state is SYN_RECEIVED: 1955 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1956 */ 1957 case TCPS_SYN_RECEIVED: 1958 if (thflags & TH_RST) { 1959 /* Handle RST segments later. */ 1960 break; 1961 } 1962 if ((thflags & TH_ACK) && 1963 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1964 SEQ_GT(th->th_ack, tp->snd_max))) { 1965 rstreason = BANDLIM_RST_OPENPORT; 1966 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1967 goto dropwithreset; 1968 } 1969 if (IS_FASTOPEN(tp->t_flags)) { 1970 /* 1971 * When a TFO connection is in SYN_RECEIVED, the 1972 * only valid packets are the initial SYN, a 1973 * retransmit/copy of the initial SYN (possibly with 1974 * a subset of the original data), a valid ACK, a 1975 * FIN, or a RST. 1976 */ 1977 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1978 rstreason = BANDLIM_RST_OPENPORT; 1979 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1980 goto dropwithreset; 1981 } else if (thflags & TH_SYN) { 1982 /* non-initial SYN is ignored */ 1983 if ((tcp_timer_active(tp, TT_DELACK) || 1984 tcp_timer_active(tp, TT_REXMT))) 1985 goto drop; 1986 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1987 goto drop; 1988 } 1989 } 1990 break; 1991 1992 /* 1993 * If the state is SYN_SENT: 1994 * if seg contains a RST with valid ACK (SEQ.ACK has already 1995 * been verified), then drop the connection. 1996 * if seg contains a RST without an ACK, drop the seg. 1997 * if seg does not contain SYN, then drop the seg. 1998 * Otherwise this is an acceptable SYN segment 1999 * initialize tp->rcv_nxt and tp->irs 2000 * if seg contains ack then advance tp->snd_una 2001 * if seg contains an ECE and ECN support is enabled, the stream 2002 * is ECN capable. 2003 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2004 * arrange for segment to be acked (eventually) 2005 * continue processing rest of data/controls, beginning with URG 2006 */ 2007 case TCPS_SYN_SENT: 2008 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2009 TCP_PROBE5(connect__refused, NULL, tp, 2010 m, tp, th); 2011 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2012 tp = tcp_drop(tp, ECONNREFUSED); 2013 } 2014 if (thflags & TH_RST) 2015 goto drop; 2016 if (!(thflags & TH_SYN)) 2017 goto drop; 2018 2019 tp->irs = th->th_seq; 2020 tcp_rcvseqinit(tp); 2021 if (thflags & TH_ACK) { 2022 int tfo_partial_ack = 0; 2023 2024 TCPSTAT_INC(tcps_connects); 2025 soisconnected(so); 2026 #ifdef MAC 2027 mac_socketpeer_set_from_mbuf(m, so); 2028 #endif 2029 /* Do window scaling on this connection? */ 2030 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2031 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2032 tp->rcv_scale = tp->request_r_scale; 2033 } 2034 tp->rcv_adv += min(tp->rcv_wnd, 2035 TCP_MAXWIN << tp->rcv_scale); 2036 tp->snd_una++; /* SYN is acked */ 2037 /* 2038 * If not all the data that was sent in the TFO SYN 2039 * has been acked, resend the remainder right away. 2040 */ 2041 if (IS_FASTOPEN(tp->t_flags) && 2042 (tp->snd_una != tp->snd_max)) { 2043 tp->snd_nxt = th->th_ack; 2044 tfo_partial_ack = 1; 2045 } 2046 /* 2047 * If there's data, delay ACK; if there's also a FIN 2048 * ACKNOW will be turned on later. 2049 */ 2050 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2051 tcp_timer_activate(tp, TT_DELACK, 2052 tcp_delacktime); 2053 else 2054 tp->t_flags |= TF_ACKNOW; 2055 2056 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2057 2058 /* 2059 * Received <SYN,ACK> in SYN_SENT[*] state. 2060 * Transitions: 2061 * SYN_SENT --> ESTABLISHED 2062 * SYN_SENT* --> FIN_WAIT_1 2063 */ 2064 tp->t_starttime = ticks; 2065 if (tp->t_flags & TF_NEEDFIN) { 2066 tp->t_acktime = ticks; 2067 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2068 tp->t_flags &= ~TF_NEEDFIN; 2069 thflags &= ~TH_SYN; 2070 } else { 2071 tcp_state_change(tp, TCPS_ESTABLISHED); 2072 TCP_PROBE5(connect__established, NULL, tp, 2073 m, tp, th); 2074 cc_conn_init(tp); 2075 tcp_timer_activate(tp, TT_KEEP, 2076 TP_KEEPIDLE(tp)); 2077 } 2078 } else { 2079 /* 2080 * Received initial SYN in SYN-SENT[*] state => 2081 * simultaneous open. 2082 * If it succeeds, connection is * half-synchronized. 2083 * Otherwise, do 3-way handshake: 2084 * SYN-SENT -> SYN-RECEIVED 2085 * SYN-SENT* -> SYN-RECEIVED* 2086 */ 2087 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN); 2088 tcp_timer_activate(tp, TT_REXMT, 0); 2089 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2090 } 2091 2092 /* 2093 * Advance th->th_seq to correspond to first data byte. 2094 * If data, trim to stay within window, 2095 * dropping FIN if necessary. 2096 */ 2097 th->th_seq++; 2098 if (tlen > tp->rcv_wnd) { 2099 todrop = tlen - tp->rcv_wnd; 2100 m_adj(m, -todrop); 2101 tlen = tp->rcv_wnd; 2102 thflags &= ~TH_FIN; 2103 TCPSTAT_INC(tcps_rcvpackafterwin); 2104 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2105 } 2106 tp->snd_wl1 = th->th_seq - 1; 2107 tp->rcv_up = th->th_seq; 2108 /* 2109 * Client side of transaction: already sent SYN and data. 2110 * If the remote host used T/TCP to validate the SYN, 2111 * our data will be ACK'd; if so, enter normal data segment 2112 * processing in the middle of step 5, ack processing. 2113 * Otherwise, goto step 6. 2114 */ 2115 if (thflags & TH_ACK) 2116 goto process_ACK; 2117 2118 goto step6; 2119 } 2120 2121 /* 2122 * States other than LISTEN or SYN_SENT. 2123 * First check the RST flag and sequence number since reset segments 2124 * are exempt from the timestamp and connection count tests. This 2125 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2126 * below which allowed reset segments in half the sequence space 2127 * to fall though and be processed (which gives forged reset 2128 * segments with a random sequence number a 50 percent chance of 2129 * killing a connection). 2130 * Then check timestamp, if present. 2131 * Then check the connection count, if present. 2132 * Then check that at least some bytes of segment are within 2133 * receive window. If segment begins before rcv_nxt, 2134 * drop leading data (and SYN); if nothing left, just ack. 2135 */ 2136 if (thflags & TH_RST) { 2137 /* 2138 * RFC5961 Section 3.2 2139 * 2140 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2141 * - If RST is in window, we send challenge ACK. 2142 * 2143 * Note: to take into account delayed ACKs, we should 2144 * test against last_ack_sent instead of rcv_nxt. 2145 * Note 2: we handle special case of closed window, not 2146 * covered by the RFC. 2147 */ 2148 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2149 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2150 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2151 KASSERT(tp->t_state != TCPS_SYN_SENT, 2152 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2153 __func__, th, tp)); 2154 2155 if (V_tcp_insecure_rst || 2156 tp->last_ack_sent == th->th_seq) { 2157 TCPSTAT_INC(tcps_drops); 2158 /* Drop the connection. */ 2159 switch (tp->t_state) { 2160 case TCPS_SYN_RECEIVED: 2161 so->so_error = ECONNREFUSED; 2162 goto close; 2163 case TCPS_ESTABLISHED: 2164 case TCPS_FIN_WAIT_1: 2165 case TCPS_FIN_WAIT_2: 2166 case TCPS_CLOSE_WAIT: 2167 case TCPS_CLOSING: 2168 case TCPS_LAST_ACK: 2169 so->so_error = ECONNRESET; 2170 close: 2171 /* FALLTHROUGH */ 2172 default: 2173 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST); 2174 tp = tcp_close(tp); 2175 } 2176 } else { 2177 TCPSTAT_INC(tcps_badrst); 2178 /* Send challenge ACK. */ 2179 tcp_respond(tp, mtod(m, void *), th, m, 2180 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2181 tp->last_ack_sent = tp->rcv_nxt; 2182 m = NULL; 2183 } 2184 } 2185 goto drop; 2186 } 2187 2188 /* 2189 * RFC5961 Section 4.2 2190 * Send challenge ACK for any SYN in synchronized state. 2191 */ 2192 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2193 tp->t_state != TCPS_SYN_RECEIVED) { 2194 TCPSTAT_INC(tcps_badsyn); 2195 if (V_tcp_insecure_syn && 2196 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2197 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2198 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2199 tp = tcp_drop(tp, ECONNRESET); 2200 rstreason = BANDLIM_UNLIMITED; 2201 } else { 2202 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2203 /* Send challenge ACK. */ 2204 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2205 tp->snd_nxt, TH_ACK); 2206 tp->last_ack_sent = tp->rcv_nxt; 2207 m = NULL; 2208 } 2209 goto drop; 2210 } 2211 2212 /* 2213 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2214 * and it's less than ts_recent, drop it. 2215 */ 2216 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2217 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2218 /* Check to see if ts_recent is over 24 days old. */ 2219 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2220 /* 2221 * Invalidate ts_recent. If this segment updates 2222 * ts_recent, the age will be reset later and ts_recent 2223 * will get a valid value. If it does not, setting 2224 * ts_recent to zero will at least satisfy the 2225 * requirement that zero be placed in the timestamp 2226 * echo reply when ts_recent isn't valid. The 2227 * age isn't reset until we get a valid ts_recent 2228 * because we don't want out-of-order segments to be 2229 * dropped when ts_recent is old. 2230 */ 2231 tp->ts_recent = 0; 2232 } else { 2233 TCPSTAT_INC(tcps_rcvduppack); 2234 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2235 TCPSTAT_INC(tcps_pawsdrop); 2236 if (tlen) 2237 goto dropafterack; 2238 goto drop; 2239 } 2240 } 2241 2242 /* 2243 * In the SYN-RECEIVED state, validate that the packet belongs to 2244 * this connection before trimming the data to fit the receive 2245 * window. Check the sequence number versus IRS since we know 2246 * the sequence numbers haven't wrapped. This is a partial fix 2247 * for the "LAND" DoS attack. 2248 */ 2249 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2250 rstreason = BANDLIM_RST_OPENPORT; 2251 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2252 goto dropwithreset; 2253 } 2254 2255 todrop = tp->rcv_nxt - th->th_seq; 2256 if (todrop > 0) { 2257 if (thflags & TH_SYN) { 2258 thflags &= ~TH_SYN; 2259 th->th_seq++; 2260 if (th->th_urp > 1) 2261 th->th_urp--; 2262 else 2263 thflags &= ~TH_URG; 2264 todrop--; 2265 } 2266 /* 2267 * Following if statement from Stevens, vol. 2, p. 960. 2268 */ 2269 if (todrop > tlen 2270 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2271 /* 2272 * Any valid FIN must be to the left of the window. 2273 * At this point the FIN must be a duplicate or out 2274 * of sequence; drop it. 2275 */ 2276 thflags &= ~TH_FIN; 2277 2278 /* 2279 * Send an ACK to resynchronize and drop any data. 2280 * But keep on processing for RST or ACK. 2281 */ 2282 tp->t_flags |= TF_ACKNOW; 2283 todrop = tlen; 2284 TCPSTAT_INC(tcps_rcvduppack); 2285 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2286 } else { 2287 TCPSTAT_INC(tcps_rcvpartduppack); 2288 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2289 } 2290 /* 2291 * DSACK - add SACK block for dropped range 2292 */ 2293 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2294 tcp_update_sack_list(tp, th->th_seq, 2295 th->th_seq + todrop); 2296 /* 2297 * ACK now, as the next in-sequence segment 2298 * will clear the DSACK block again 2299 */ 2300 tp->t_flags |= TF_ACKNOW; 2301 } 2302 drop_hdrlen += todrop; /* drop from the top afterwards */ 2303 th->th_seq += todrop; 2304 tlen -= todrop; 2305 if (th->th_urp > todrop) 2306 th->th_urp -= todrop; 2307 else { 2308 thflags &= ~TH_URG; 2309 th->th_urp = 0; 2310 } 2311 } 2312 2313 /* 2314 * If new data are received on a connection after the 2315 * user processes are gone, then RST the other end. 2316 */ 2317 if ((tp->t_flags & TF_CLOSED) && tlen) { 2318 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2319 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2320 "after socket was closed, " 2321 "sending RST and removing tcpcb\n", 2322 s, __func__, tcpstates[tp->t_state], tlen); 2323 free(s, M_TCPLOG); 2324 } 2325 tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE); 2326 /* tcp_close will kill the inp pre-log the Reset */ 2327 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST); 2328 tp = tcp_close(tp); 2329 TCPSTAT_INC(tcps_rcvafterclose); 2330 rstreason = BANDLIM_UNLIMITED; 2331 goto dropwithreset; 2332 } 2333 2334 /* 2335 * If segment ends after window, drop trailing data 2336 * (and PUSH and FIN); if nothing left, just ACK. 2337 */ 2338 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2339 if (todrop > 0) { 2340 TCPSTAT_INC(tcps_rcvpackafterwin); 2341 if (todrop >= tlen) { 2342 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2343 /* 2344 * If window is closed can only take segments at 2345 * window edge, and have to drop data and PUSH from 2346 * incoming segments. Continue processing, but 2347 * remember to ack. Otherwise, drop segment 2348 * and ack. 2349 */ 2350 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2351 tp->t_flags |= TF_ACKNOW; 2352 TCPSTAT_INC(tcps_rcvwinprobe); 2353 } else 2354 goto dropafterack; 2355 } else 2356 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2357 m_adj(m, -todrop); 2358 tlen -= todrop; 2359 thflags &= ~(TH_PUSH|TH_FIN); 2360 } 2361 2362 /* 2363 * If last ACK falls within this segment's sequence numbers, 2364 * record its timestamp. 2365 * NOTE: 2366 * 1) That the test incorporates suggestions from the latest 2367 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2368 * 2) That updating only on newer timestamps interferes with 2369 * our earlier PAWS tests, so this check should be solely 2370 * predicated on the sequence space of this segment. 2371 * 3) That we modify the segment boundary check to be 2372 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2373 * instead of RFC1323's 2374 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2375 * This modified check allows us to overcome RFC1323's 2376 * limitations as described in Stevens TCP/IP Illustrated 2377 * Vol. 2 p.869. In such cases, we can still calculate the 2378 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2379 */ 2380 if ((to.to_flags & TOF_TS) != 0 && 2381 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2382 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2383 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2384 tp->ts_recent_age = tcp_ts_getticks(); 2385 tp->ts_recent = to.to_tsval; 2386 } 2387 2388 /* 2389 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2390 * flag is on (half-synchronized state), then queue data for 2391 * later processing; else drop segment and return. 2392 */ 2393 if ((thflags & TH_ACK) == 0) { 2394 if (tp->t_state == TCPS_SYN_RECEIVED || 2395 (tp->t_flags & TF_NEEDSYN)) { 2396 if (tp->t_state == TCPS_SYN_RECEIVED && 2397 IS_FASTOPEN(tp->t_flags)) { 2398 tp->snd_wnd = tiwin; 2399 cc_conn_init(tp); 2400 } 2401 goto step6; 2402 } else if (tp->t_flags & TF_ACKNOW) 2403 goto dropafterack; 2404 else 2405 goto drop; 2406 } 2407 2408 /* 2409 * Ack processing. 2410 */ 2411 switch (tp->t_state) { 2412 /* 2413 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2414 * ESTABLISHED state and continue processing. 2415 * The ACK was checked above. 2416 */ 2417 case TCPS_SYN_RECEIVED: 2418 2419 TCPSTAT_INC(tcps_connects); 2420 if (tp->t_flags & TF_SONOTCONN) { 2421 /* 2422 * Usually SYN_RECEIVED had been created from a LISTEN, 2423 * and solisten_enqueue() has already marked the socket 2424 * layer as connected. If it didn't, which can happen 2425 * only with an accept_filter(9), then the tp is marked 2426 * with TF_SONOTCONN. The other reason for this mark 2427 * to be set is a simultaneous open, a SYN_RECEIVED 2428 * that had been created from SYN_SENT. 2429 */ 2430 tp->t_flags &= ~TF_SONOTCONN; 2431 soisconnected(so); 2432 } 2433 /* Do window scaling? */ 2434 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2435 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2436 tp->rcv_scale = tp->request_r_scale; 2437 } 2438 tp->snd_wnd = tiwin; 2439 /* 2440 * Make transitions: 2441 * SYN-RECEIVED -> ESTABLISHED 2442 * SYN-RECEIVED* -> FIN-WAIT-1 2443 */ 2444 tp->t_starttime = ticks; 2445 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2446 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2447 tp->t_tfo_pending = NULL; 2448 } 2449 if (tp->t_flags & TF_NEEDFIN) { 2450 tp->t_acktime = ticks; 2451 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2452 tp->t_flags &= ~TF_NEEDFIN; 2453 } else { 2454 tcp_state_change(tp, TCPS_ESTABLISHED); 2455 TCP_PROBE5(accept__established, NULL, tp, 2456 m, tp, th); 2457 /* 2458 * TFO connections call cc_conn_init() during SYN 2459 * processing. Calling it again here for such 2460 * connections is not harmless as it would undo the 2461 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2462 * is retransmitted. 2463 */ 2464 if (!IS_FASTOPEN(tp->t_flags)) 2465 cc_conn_init(tp); 2466 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2467 } 2468 /* 2469 * Account for the ACK of our SYN prior to 2470 * regular ACK processing below, except for 2471 * simultaneous SYN, which is handled later. 2472 */ 2473 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2474 incforsyn = 1; 2475 /* 2476 * If segment contains data or ACK, will call tcp_reass() 2477 * later; if not, do so now to pass queued data to user. 2478 */ 2479 if (tlen == 0 && (thflags & TH_FIN) == 0) { 2480 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2481 (struct mbuf *)0); 2482 tcp_handle_wakeup(tp); 2483 } 2484 tp->snd_wl1 = th->th_seq - 1; 2485 /* FALLTHROUGH */ 2486 2487 /* 2488 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2489 * ACKs. If the ack is in the range 2490 * tp->snd_una < th->th_ack <= tp->snd_max 2491 * then advance tp->snd_una to th->th_ack and drop 2492 * data from the retransmission queue. If this ACK reflects 2493 * more up to date window information we update our window information. 2494 */ 2495 case TCPS_ESTABLISHED: 2496 case TCPS_FIN_WAIT_1: 2497 case TCPS_FIN_WAIT_2: 2498 case TCPS_CLOSE_WAIT: 2499 case TCPS_CLOSING: 2500 case TCPS_LAST_ACK: 2501 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2502 TCPSTAT_INC(tcps_rcvacktoomuch); 2503 goto dropafterack; 2504 } 2505 if (tcp_is_sack_recovery(tp, &to)) { 2506 if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) && 2507 (tp->t_flags & TF_LRD)) { 2508 tcp_sack_lost_retransmission(tp, th); 2509 } 2510 } else 2511 /* 2512 * Reset the value so that previous (valid) value 2513 * from the last ack with SACK doesn't get used. 2514 */ 2515 tp->sackhint.sacked_bytes = 0; 2516 2517 #ifdef TCP_HHOOK 2518 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2519 hhook_run_tcp_est_in(tp, th, &to); 2520 #endif 2521 2522 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2523 maxseg = tcp_maxseg(tp); 2524 if (tlen == 0 && 2525 (tiwin == tp->snd_wnd || 2526 (tp->t_flags & TF_SACK_PERMIT))) { 2527 /* 2528 * If this is the first time we've seen a 2529 * FIN from the remote, this is not a 2530 * duplicate and it needs to be processed 2531 * normally. This happens during a 2532 * simultaneous close. 2533 */ 2534 if ((thflags & TH_FIN) && 2535 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2536 tp->t_dupacks = 0; 2537 break; 2538 } 2539 TCPSTAT_INC(tcps_rcvdupack); 2540 /* 2541 * If we have outstanding data (other than 2542 * a window probe), this is a completely 2543 * duplicate ack (ie, window info didn't 2544 * change and FIN isn't set), 2545 * the ack is the biggest we've 2546 * seen and we've seen exactly our rexmt 2547 * threshold of them, assume a packet 2548 * has been dropped and retransmit it. 2549 * Kludge snd_nxt & the congestion 2550 * window so we send only this one 2551 * packet. 2552 * 2553 * We know we're losing at the current 2554 * window size so do congestion avoidance 2555 * (set ssthresh to half the current window 2556 * and pull our congestion window back to 2557 * the new ssthresh). 2558 * 2559 * Dup acks mean that packets have left the 2560 * network (they're now cached at the receiver) 2561 * so bump cwnd by the amount in the receiver 2562 * to keep a constant cwnd packets in the 2563 * network. 2564 * 2565 * When using TCP ECN, notify the peer that 2566 * we reduced the cwnd. 2567 */ 2568 /* 2569 * Following 2 kinds of acks should not affect 2570 * dupack counting: 2571 * 1) Old acks 2572 * 2) Acks with SACK but without any new SACK 2573 * information in them. These could result from 2574 * any anomaly in the network like a switch 2575 * duplicating packets or a possible DoS attack. 2576 */ 2577 if (th->th_ack != tp->snd_una || 2578 (tcp_is_sack_recovery(tp, &to) && 2579 (sack_changed == SACK_NOCHANGE))) 2580 break; 2581 else if (!tcp_timer_active(tp, TT_REXMT)) 2582 tp->t_dupacks = 0; 2583 else if (++tp->t_dupacks > tcprexmtthresh || 2584 IN_FASTRECOVERY(tp->t_flags)) { 2585 cc_ack_received(tp, th, nsegs, 2586 CC_DUPACK); 2587 if (V_tcp_do_prr && 2588 IN_FASTRECOVERY(tp->t_flags) && 2589 (tp->t_flags & TF_SACK_PERMIT)) { 2590 tcp_do_prr_ack(tp, th, &to, sack_changed); 2591 } else if (tcp_is_sack_recovery(tp, &to) && 2592 IN_FASTRECOVERY(tp->t_flags)) { 2593 int awnd; 2594 2595 /* 2596 * Compute the amount of data in flight first. 2597 * We can inject new data into the pipe iff 2598 * we have less than 1/2 the original window's 2599 * worth of data in flight. 2600 */ 2601 if (V_tcp_do_newsack) 2602 awnd = tcp_compute_pipe(tp); 2603 else 2604 awnd = (tp->snd_nxt - tp->snd_fack) + 2605 tp->sackhint.sack_bytes_rexmit; 2606 2607 if (awnd < tp->snd_ssthresh) { 2608 tp->snd_cwnd += maxseg; 2609 if (tp->snd_cwnd > tp->snd_ssthresh) 2610 tp->snd_cwnd = tp->snd_ssthresh; 2611 } 2612 } else 2613 tp->snd_cwnd += maxseg; 2614 (void) tcp_output(tp); 2615 goto drop; 2616 } else if (tp->t_dupacks == tcprexmtthresh || 2617 (tp->t_flags & TF_SACK_PERMIT && 2618 V_tcp_do_newsack && 2619 tp->sackhint.sacked_bytes > 2620 (tcprexmtthresh - 1) * maxseg)) { 2621 enter_recovery: 2622 /* 2623 * Above is the RFC6675 trigger condition of 2624 * more than (dupthresh-1)*maxseg sacked data. 2625 * If the count of holes in the 2626 * scoreboard is >= dupthresh, we could 2627 * also enter loss recovery, but don't 2628 * have that value readily available. 2629 */ 2630 tp->t_dupacks = tcprexmtthresh; 2631 tcp_seq onxt = tp->snd_nxt; 2632 2633 /* 2634 * If we're doing sack, or prr, check 2635 * to see if we're already in sack 2636 * recovery. If we're not doing sack, 2637 * check to see if we're in newreno 2638 * recovery. 2639 */ 2640 if (V_tcp_do_prr || 2641 (tp->t_flags & TF_SACK_PERMIT)) { 2642 if (IN_FASTRECOVERY(tp->t_flags)) { 2643 tp->t_dupacks = 0; 2644 break; 2645 } 2646 } else { 2647 if (SEQ_LEQ(th->th_ack, 2648 tp->snd_recover)) { 2649 tp->t_dupacks = 0; 2650 break; 2651 } 2652 } 2653 /* Congestion signal before ack. */ 2654 cc_cong_signal(tp, th, CC_NDUPACK); 2655 cc_ack_received(tp, th, nsegs, 2656 CC_DUPACK); 2657 tcp_timer_activate(tp, TT_REXMT, 0); 2658 tp->t_rtttime = 0; 2659 if (V_tcp_do_prr) { 2660 /* 2661 * snd_ssthresh is already updated by 2662 * cc_cong_signal. 2663 */ 2664 if (tcp_is_sack_recovery(tp, &to)) { 2665 /* 2666 * Exclude Limited Transmit 2667 * segments here 2668 */ 2669 tp->sackhint.prr_delivered = 2670 maxseg; 2671 } else { 2672 tp->sackhint.prr_delivered = 2673 imin(tp->snd_max - tp->snd_una, 2674 imin(INT_MAX / 65536, 2675 tp->t_dupacks) * maxseg); 2676 } 2677 tp->sackhint.recover_fs = max(1, 2678 tp->snd_nxt - tp->snd_una); 2679 } 2680 if (tcp_is_sack_recovery(tp, &to)) { 2681 TCPSTAT_INC( 2682 tcps_sack_recovery_episode); 2683 tp->snd_recover = tp->snd_nxt; 2684 tp->snd_cwnd = maxseg; 2685 (void) tcp_output(tp); 2686 if (SEQ_GT(th->th_ack, tp->snd_una)) 2687 goto resume_partialack; 2688 goto drop; 2689 } 2690 tp->snd_nxt = th->th_ack; 2691 tp->snd_cwnd = maxseg; 2692 (void) tcp_output(tp); 2693 KASSERT(tp->snd_limited <= 2, 2694 ("%s: tp->snd_limited too big", 2695 __func__)); 2696 tp->snd_cwnd = tp->snd_ssthresh + 2697 maxseg * 2698 (tp->t_dupacks - tp->snd_limited); 2699 if (SEQ_GT(onxt, tp->snd_nxt)) 2700 tp->snd_nxt = onxt; 2701 goto drop; 2702 } else if (V_tcp_do_rfc3042) { 2703 /* 2704 * Process first and second duplicate 2705 * ACKs. Each indicates a segment 2706 * leaving the network, creating room 2707 * for more. Make sure we can send a 2708 * packet on reception of each duplicate 2709 * ACK by increasing snd_cwnd by one 2710 * segment. Restore the original 2711 * snd_cwnd after packet transmission. 2712 */ 2713 cc_ack_received(tp, th, nsegs, 2714 CC_DUPACK); 2715 uint32_t oldcwnd = tp->snd_cwnd; 2716 tcp_seq oldsndmax = tp->snd_max; 2717 u_int sent; 2718 int avail; 2719 2720 KASSERT(tp->t_dupacks == 1 || 2721 tp->t_dupacks == 2, 2722 ("%s: dupacks not 1 or 2", 2723 __func__)); 2724 if (tp->t_dupacks == 1) 2725 tp->snd_limited = 0; 2726 tp->snd_cwnd = 2727 (tp->snd_nxt - tp->snd_una) + 2728 (tp->t_dupacks - tp->snd_limited) * 2729 maxseg; 2730 /* 2731 * Only call tcp_output when there 2732 * is new data available to be sent 2733 * or we need to send an ACK. 2734 */ 2735 SOCKBUF_LOCK(&so->so_snd); 2736 avail = sbavail(&so->so_snd) - 2737 (tp->snd_nxt - tp->snd_una); 2738 SOCKBUF_UNLOCK(&so->so_snd); 2739 if (avail > 0 || tp->t_flags & TF_ACKNOW) 2740 (void) tcp_output(tp); 2741 sent = tp->snd_max - oldsndmax; 2742 if (sent > maxseg) { 2743 KASSERT((tp->t_dupacks == 2 && 2744 tp->snd_limited == 0) || 2745 (sent == maxseg + 1 && 2746 tp->t_flags & TF_SENTFIN), 2747 ("%s: sent too much", 2748 __func__)); 2749 tp->snd_limited = 2; 2750 } else if (sent > 0) 2751 ++tp->snd_limited; 2752 tp->snd_cwnd = oldcwnd; 2753 goto drop; 2754 } 2755 } 2756 break; 2757 } else { 2758 /* 2759 * This ack is advancing the left edge, reset the 2760 * counter. 2761 */ 2762 tp->t_dupacks = 0; 2763 /* 2764 * If this ack also has new SACK info, increment the 2765 * counter as per rfc6675. The variable 2766 * sack_changed tracks all changes to the SACK 2767 * scoreboard, including when partial ACKs without 2768 * SACK options are received, and clear the scoreboard 2769 * from the left side. Such partial ACKs should not be 2770 * counted as dupacks here. 2771 */ 2772 if (tcp_is_sack_recovery(tp, &to) && 2773 (sack_changed != SACK_NOCHANGE)) { 2774 tp->t_dupacks++; 2775 /* limit overhead by setting maxseg last */ 2776 if (!IN_FASTRECOVERY(tp->t_flags) && 2777 (tp->sackhint.sacked_bytes > 2778 ((tcprexmtthresh - 1) * 2779 (maxseg = tcp_maxseg(tp))))) { 2780 goto enter_recovery; 2781 } 2782 } 2783 } 2784 2785 resume_partialack: 2786 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2787 ("%s: th_ack <= snd_una", __func__)); 2788 2789 /* 2790 * If the congestion window was inflated to account 2791 * for the other side's cached packets, retract it. 2792 */ 2793 if (IN_FASTRECOVERY(tp->t_flags)) { 2794 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2795 if (tp->t_flags & TF_SACK_PERMIT) 2796 if (V_tcp_do_prr && to.to_flags & TOF_SACK) { 2797 tcp_timer_activate(tp, TT_REXMT, 0); 2798 tp->t_rtttime = 0; 2799 tcp_do_prr_ack(tp, th, &to, sack_changed); 2800 tp->t_flags |= TF_ACKNOW; 2801 (void) tcp_output(tp); 2802 } else 2803 tcp_sack_partialack(tp, th); 2804 else 2805 tcp_newreno_partial_ack(tp, th); 2806 } else 2807 cc_post_recovery(tp, th); 2808 } else if (IN_CONGRECOVERY(tp->t_flags)) { 2809 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2810 if (V_tcp_do_prr) { 2811 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th); 2812 tp->snd_fack = th->th_ack; 2813 /* 2814 * During ECN cwnd reduction 2815 * always use PRR-SSRB 2816 */ 2817 tcp_do_prr_ack(tp, th, &to, SACK_CHANGE); 2818 (void) tcp_output(tp); 2819 } 2820 } else 2821 cc_post_recovery(tp, th); 2822 } 2823 /* 2824 * If we reach this point, ACK is not a duplicate, 2825 * i.e., it ACKs something we sent. 2826 */ 2827 if (tp->t_flags & TF_NEEDSYN) { 2828 /* 2829 * T/TCP: Connection was half-synchronized, and our 2830 * SYN has been ACK'd (so connection is now fully 2831 * synchronized). Go to non-starred state, 2832 * increment snd_una for ACK of SYN, and check if 2833 * we can do window scaling. 2834 */ 2835 tp->t_flags &= ~TF_NEEDSYN; 2836 tp->snd_una++; 2837 /* Do window scaling? */ 2838 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2839 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2840 tp->rcv_scale = tp->request_r_scale; 2841 /* Send window already scaled. */ 2842 } 2843 } 2844 2845 process_ACK: 2846 INP_WLOCK_ASSERT(inp); 2847 2848 /* 2849 * Adjust for the SYN bit in sequence space, 2850 * but don't account for it in cwnd calculations. 2851 * This is for the SYN_RECEIVED, non-simultaneous 2852 * SYN case. SYN_SENT and simultaneous SYN are 2853 * treated elsewhere. 2854 */ 2855 if (incforsyn) 2856 tp->snd_una++; 2857 acked = BYTES_THIS_ACK(tp, th); 2858 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2859 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2860 tp->snd_una, th->th_ack, tp, m)); 2861 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2862 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2863 2864 /* 2865 * If we just performed our first retransmit, and the ACK 2866 * arrives within our recovery window, then it was a mistake 2867 * to do the retransmit in the first place. Recover our 2868 * original cwnd and ssthresh, and proceed to transmit where 2869 * we left off. 2870 */ 2871 if (tp->t_rxtshift == 1 && 2872 tp->t_flags & TF_PREVVALID && 2873 tp->t_badrxtwin != 0 && 2874 to.to_flags & TOF_TS && 2875 to.to_tsecr != 0 && 2876 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 2877 cc_cong_signal(tp, th, CC_RTO_ERR); 2878 2879 /* 2880 * If we have a timestamp reply, update smoothed 2881 * round trip time. If no timestamp is present but 2882 * transmit timer is running and timed sequence 2883 * number was acked, update smoothed round trip time. 2884 * Since we now have an rtt measurement, cancel the 2885 * timer backoff (cf., Phil Karn's retransmit alg.). 2886 * Recompute the initial retransmit timer. 2887 * 2888 * Some boxes send broken timestamp replies 2889 * during the SYN+ACK phase, ignore 2890 * timestamps of 0 or we could calculate a 2891 * huge RTT and blow up the retransmit timer. 2892 */ 2893 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2894 uint32_t t; 2895 2896 t = tcp_ts_getticks() - to.to_tsecr; 2897 if (!tp->t_rttlow || tp->t_rttlow > t) 2898 tp->t_rttlow = t; 2899 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2900 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2901 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2902 tp->t_rttlow = ticks - tp->t_rtttime; 2903 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2904 } 2905 2906 SOCKBUF_LOCK(&so->so_snd); 2907 /* 2908 * Clear t_acktime if remote side has ACKd all data in the 2909 * socket buffer and FIN (if applicable). 2910 * Otherwise, update t_acktime if we received a sufficiently 2911 * large ACK. 2912 */ 2913 if ((tp->t_state <= TCPS_CLOSE_WAIT && 2914 acked == sbavail(&so->so_snd)) || 2915 acked > sbavail(&so->so_snd)) 2916 tp->t_acktime = 0; 2917 else if (acked > 1) 2918 tp->t_acktime = ticks; 2919 2920 /* 2921 * If all outstanding data is acked, stop retransmit 2922 * timer and remember to restart (more output or persist). 2923 * If there is more data to be acked, restart retransmit 2924 * timer, using current (possibly backed-off) value. 2925 */ 2926 if (th->th_ack == tp->snd_max) { 2927 tcp_timer_activate(tp, TT_REXMT, 0); 2928 needoutput = 1; 2929 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2930 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); 2931 2932 /* 2933 * If no data (only SYN) was ACK'd, 2934 * skip rest of ACK processing. 2935 */ 2936 if (acked == 0) { 2937 SOCKBUF_UNLOCK(&so->so_snd); 2938 goto step6; 2939 } 2940 2941 /* 2942 * Let the congestion control algorithm update congestion 2943 * control related information. This typically means increasing 2944 * the congestion window. 2945 */ 2946 cc_ack_received(tp, th, nsegs, CC_ACK); 2947 2948 if (acked > sbavail(&so->so_snd)) { 2949 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2950 tp->snd_wnd -= sbavail(&so->so_snd); 2951 else 2952 tp->snd_wnd = 0; 2953 mfree = sbcut_locked(&so->so_snd, 2954 (int)sbavail(&so->so_snd)); 2955 ourfinisacked = 1; 2956 } else { 2957 mfree = sbcut_locked(&so->so_snd, acked); 2958 if (tp->snd_wnd >= (uint32_t) acked) 2959 tp->snd_wnd -= acked; 2960 else 2961 tp->snd_wnd = 0; 2962 ourfinisacked = 0; 2963 } 2964 /* NB: sowwakeup_locked() does an implicit unlock. */ 2965 sowwakeup_locked(so); 2966 m_freem(mfree); 2967 /* Detect una wraparound. */ 2968 if (!IN_RECOVERY(tp->t_flags) && 2969 SEQ_GT(tp->snd_una, tp->snd_recover) && 2970 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2971 tp->snd_recover = th->th_ack - 1; 2972 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2973 if (IN_RECOVERY(tp->t_flags) && 2974 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2975 EXIT_RECOVERY(tp->t_flags); 2976 } 2977 tp->snd_una = th->th_ack; 2978 if (tp->t_flags & TF_SACK_PERMIT) { 2979 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2980 tp->snd_recover = tp->snd_una; 2981 } 2982 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2983 tp->snd_nxt = tp->snd_una; 2984 2985 switch (tp->t_state) { 2986 /* 2987 * In FIN_WAIT_1 STATE in addition to the processing 2988 * for the ESTABLISHED state if our FIN is now acknowledged 2989 * then enter FIN_WAIT_2. 2990 */ 2991 case TCPS_FIN_WAIT_1: 2992 if (ourfinisacked) { 2993 /* 2994 * If we can't receive any more 2995 * data, then closing user can proceed. 2996 * Starting the timer is contrary to the 2997 * specification, but if we don't get a FIN 2998 * we'll hang forever. 2999 */ 3000 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3001 soisdisconnected(so); 3002 tcp_timer_activate(tp, TT_2MSL, 3003 (tcp_fast_finwait2_recycle ? 3004 tcp_finwait2_timeout : 3005 TP_MAXIDLE(tp))); 3006 } 3007 tcp_state_change(tp, TCPS_FIN_WAIT_2); 3008 } 3009 break; 3010 3011 /* 3012 * In CLOSING STATE in addition to the processing for 3013 * the ESTABLISHED state if the ACK acknowledges our FIN 3014 * then enter the TIME-WAIT state, otherwise ignore 3015 * the segment. 3016 */ 3017 case TCPS_CLOSING: 3018 if (ourfinisacked) { 3019 tcp_twstart(tp); 3020 m_freem(m); 3021 return; 3022 } 3023 break; 3024 3025 /* 3026 * In LAST_ACK, we may still be waiting for data to drain 3027 * and/or to be acked, as well as for the ack of our FIN. 3028 * If our FIN is now acknowledged, delete the TCB, 3029 * enter the closed state and return. 3030 */ 3031 case TCPS_LAST_ACK: 3032 if (ourfinisacked) { 3033 tp = tcp_close(tp); 3034 goto drop; 3035 } 3036 break; 3037 } 3038 } 3039 3040 step6: 3041 INP_WLOCK_ASSERT(inp); 3042 3043 /* 3044 * Update window information. 3045 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3046 */ 3047 if ((thflags & TH_ACK) && 3048 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3049 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3050 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3051 /* keep track of pure window updates */ 3052 if (tlen == 0 && 3053 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3054 TCPSTAT_INC(tcps_rcvwinupd); 3055 tp->snd_wnd = tiwin; 3056 tp->snd_wl1 = th->th_seq; 3057 tp->snd_wl2 = th->th_ack; 3058 if (tp->snd_wnd > tp->max_sndwnd) 3059 tp->max_sndwnd = tp->snd_wnd; 3060 needoutput = 1; 3061 } 3062 3063 /* 3064 * Process segments with URG. 3065 */ 3066 if ((thflags & TH_URG) && th->th_urp && 3067 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3068 /* 3069 * This is a kludge, but if we receive and accept 3070 * random urgent pointers, we'll crash in 3071 * soreceive. It's hard to imagine someone 3072 * actually wanting to send this much urgent data. 3073 */ 3074 SOCKBUF_LOCK(&so->so_rcv); 3075 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3076 th->th_urp = 0; /* XXX */ 3077 thflags &= ~TH_URG; /* XXX */ 3078 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3079 goto dodata; /* XXX */ 3080 } 3081 /* 3082 * If this segment advances the known urgent pointer, 3083 * then mark the data stream. This should not happen 3084 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3085 * a FIN has been received from the remote side. 3086 * In these states we ignore the URG. 3087 * 3088 * According to RFC961 (Assigned Protocols), 3089 * the urgent pointer points to the last octet 3090 * of urgent data. We continue, however, 3091 * to consider it to indicate the first octet 3092 * of data past the urgent section as the original 3093 * spec states (in one of two places). 3094 */ 3095 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3096 tp->rcv_up = th->th_seq + th->th_urp; 3097 so->so_oobmark = sbavail(&so->so_rcv) + 3098 (tp->rcv_up - tp->rcv_nxt) - 1; 3099 if (so->so_oobmark == 0) 3100 so->so_rcv.sb_state |= SBS_RCVATMARK; 3101 sohasoutofband(so); 3102 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3103 } 3104 SOCKBUF_UNLOCK(&so->so_rcv); 3105 /* 3106 * Remove out of band data so doesn't get presented to user. 3107 * This can happen independent of advancing the URG pointer, 3108 * but if two URG's are pending at once, some out-of-band 3109 * data may creep in... ick. 3110 */ 3111 if (th->th_urp <= (uint32_t)tlen && 3112 !(so->so_options & SO_OOBINLINE)) { 3113 /* hdr drop is delayed */ 3114 tcp_pulloutofband(so, th, m, drop_hdrlen); 3115 } 3116 } else { 3117 /* 3118 * If no out of band data is expected, 3119 * pull receive urgent pointer along 3120 * with the receive window. 3121 */ 3122 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3123 tp->rcv_up = tp->rcv_nxt; 3124 } 3125 dodata: /* XXX */ 3126 INP_WLOCK_ASSERT(inp); 3127 3128 /* 3129 * Process the segment text, merging it into the TCP sequencing queue, 3130 * and arranging for acknowledgment of receipt if necessary. 3131 * This process logically involves adjusting tp->rcv_wnd as data 3132 * is presented to the user (this happens in tcp_usrreq.c, 3133 * case PRU_RCVD). If a FIN has already been received on this 3134 * connection then we just ignore the text. 3135 */ 3136 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3137 IS_FASTOPEN(tp->t_flags)); 3138 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3139 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3140 tcp_seq save_start = th->th_seq; 3141 tcp_seq save_rnxt = tp->rcv_nxt; 3142 int save_tlen = tlen; 3143 m_adj(m, drop_hdrlen); /* delayed header drop */ 3144 /* 3145 * Insert segment which includes th into TCP reassembly queue 3146 * with control block tp. Set thflags to whether reassembly now 3147 * includes a segment with FIN. This handles the common case 3148 * inline (segment is the next to be received on an established 3149 * connection, and the queue is empty), avoiding linkage into 3150 * and removal from the queue and repetition of various 3151 * conversions. 3152 * Set DELACK for segments received in order, but ack 3153 * immediately when segments are out of order (so 3154 * fast retransmit can work). 3155 */ 3156 if (th->th_seq == tp->rcv_nxt && 3157 SEGQ_EMPTY(tp) && 3158 (TCPS_HAVEESTABLISHED(tp->t_state) || 3159 tfo_syn)) { 3160 if (DELAY_ACK(tp, tlen) || tfo_syn) 3161 tp->t_flags |= TF_DELACK; 3162 else 3163 tp->t_flags |= TF_ACKNOW; 3164 tp->rcv_nxt += tlen; 3165 if (tlen && 3166 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3167 (tp->t_fbyte_in == 0)) { 3168 tp->t_fbyte_in = ticks; 3169 if (tp->t_fbyte_in == 0) 3170 tp->t_fbyte_in = 1; 3171 if (tp->t_fbyte_out && tp->t_fbyte_in) 3172 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3173 } 3174 thflags = tcp_get_flags(th) & TH_FIN; 3175 TCPSTAT_INC(tcps_rcvpack); 3176 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3177 SOCKBUF_LOCK(&so->so_rcv); 3178 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3179 m_freem(m); 3180 else 3181 sbappendstream_locked(&so->so_rcv, m, 0); 3182 tp->t_flags |= TF_WAKESOR; 3183 } else { 3184 /* 3185 * XXX: Due to the header drop above "th" is 3186 * theoretically invalid by now. Fortunately 3187 * m_adj() doesn't actually frees any mbufs 3188 * when trimming from the head. 3189 */ 3190 tcp_seq temp = save_start; 3191 3192 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3193 tp->t_flags |= TF_ACKNOW; 3194 } 3195 if ((tp->t_flags & TF_SACK_PERMIT) && 3196 (save_tlen > 0) && 3197 TCPS_HAVEESTABLISHED(tp->t_state)) { 3198 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3199 /* 3200 * DSACK actually handled in the fastpath 3201 * above. 3202 */ 3203 tcp_update_sack_list(tp, save_start, 3204 save_start + save_tlen); 3205 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3206 if ((tp->rcv_numsacks >= 1) && 3207 (tp->sackblks[0].end == save_start)) { 3208 /* 3209 * Partial overlap, recorded at todrop 3210 * above. 3211 */ 3212 tcp_update_sack_list(tp, 3213 tp->sackblks[0].start, 3214 tp->sackblks[0].end); 3215 } else { 3216 tcp_update_dsack_list(tp, save_start, 3217 save_start + save_tlen); 3218 } 3219 } else if (tlen >= save_tlen) { 3220 /* Update of sackblks. */ 3221 tcp_update_dsack_list(tp, save_start, 3222 save_start + save_tlen); 3223 } else if (tlen > 0) { 3224 tcp_update_dsack_list(tp, save_start, 3225 save_start + tlen); 3226 } 3227 } 3228 tcp_handle_wakeup(tp); 3229 #if 0 3230 /* 3231 * Note the amount of data that peer has sent into 3232 * our window, in order to estimate the sender's 3233 * buffer size. 3234 * XXX: Unused. 3235 */ 3236 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3237 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3238 else 3239 len = so->so_rcv.sb_hiwat; 3240 #endif 3241 } else { 3242 m_freem(m); 3243 thflags &= ~TH_FIN; 3244 } 3245 3246 /* 3247 * If FIN is received ACK the FIN and let the user know 3248 * that the connection is closing. 3249 */ 3250 if (thflags & TH_FIN) { 3251 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3252 /* The socket upcall is handled by socantrcvmore. */ 3253 socantrcvmore(so); 3254 /* 3255 * If connection is half-synchronized 3256 * (ie NEEDSYN flag on) then delay ACK, 3257 * so it may be piggybacked when SYN is sent. 3258 * Otherwise, since we received a FIN then no 3259 * more input can be expected, send ACK now. 3260 */ 3261 if (tp->t_flags & TF_NEEDSYN) 3262 tp->t_flags |= TF_DELACK; 3263 else 3264 tp->t_flags |= TF_ACKNOW; 3265 tp->rcv_nxt++; 3266 } 3267 switch (tp->t_state) { 3268 /* 3269 * In SYN_RECEIVED and ESTABLISHED STATES 3270 * enter the CLOSE_WAIT state. 3271 */ 3272 case TCPS_SYN_RECEIVED: 3273 tp->t_starttime = ticks; 3274 /* FALLTHROUGH */ 3275 case TCPS_ESTABLISHED: 3276 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3277 break; 3278 3279 /* 3280 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3281 * enter the CLOSING state. 3282 */ 3283 case TCPS_FIN_WAIT_1: 3284 tcp_state_change(tp, TCPS_CLOSING); 3285 break; 3286 3287 /* 3288 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3289 * starting the time-wait timer, turning off the other 3290 * standard timers. 3291 */ 3292 case TCPS_FIN_WAIT_2: 3293 tcp_twstart(tp); 3294 return; 3295 } 3296 } 3297 TCP_PROBE3(debug__input, tp, th, m); 3298 3299 /* 3300 * Return any desired output. 3301 */ 3302 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3303 (void) tcp_output(tp); 3304 3305 check_delack: 3306 INP_WLOCK_ASSERT(inp); 3307 3308 if (tp->t_flags & TF_DELACK) { 3309 tp->t_flags &= ~TF_DELACK; 3310 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3311 } 3312 INP_WUNLOCK(inp); 3313 return; 3314 3315 dropafterack: 3316 /* 3317 * Generate an ACK dropping incoming segment if it occupies 3318 * sequence space, where the ACK reflects our state. 3319 * 3320 * We can now skip the test for the RST flag since all 3321 * paths to this code happen after packets containing 3322 * RST have been dropped. 3323 * 3324 * In the SYN-RECEIVED state, don't send an ACK unless the 3325 * segment we received passes the SYN-RECEIVED ACK test. 3326 * If it fails send a RST. This breaks the loop in the 3327 * "LAND" DoS attack, and also prevents an ACK storm 3328 * between two listening ports that have been sent forged 3329 * SYN segments, each with the source address of the other. 3330 */ 3331 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3332 (SEQ_GT(tp->snd_una, th->th_ack) || 3333 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3334 rstreason = BANDLIM_RST_OPENPORT; 3335 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 3336 goto dropwithreset; 3337 } 3338 TCP_PROBE3(debug__input, tp, th, m); 3339 tp->t_flags |= TF_ACKNOW; 3340 (void) tcp_output(tp); 3341 INP_WUNLOCK(inp); 3342 m_freem(m); 3343 return; 3344 3345 dropwithreset: 3346 if (tp != NULL) { 3347 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3348 INP_WUNLOCK(inp); 3349 } else 3350 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3351 return; 3352 3353 drop: 3354 /* 3355 * Drop space held by incoming segment and return. 3356 */ 3357 TCP_PROBE3(debug__input, tp, th, m); 3358 if (tp != NULL) { 3359 INP_WUNLOCK(inp); 3360 } 3361 m_freem(m); 3362 } 3363 3364 /* 3365 * Issue RST and make ACK acceptable to originator of segment. 3366 * The mbuf must still include the original packet header. 3367 * tp may be NULL. 3368 */ 3369 void 3370 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3371 int tlen, int rstreason) 3372 { 3373 #ifdef INET 3374 struct ip *ip; 3375 #endif 3376 #ifdef INET6 3377 struct ip6_hdr *ip6; 3378 #endif 3379 3380 if (tp != NULL) { 3381 INP_LOCK_ASSERT(tptoinpcb(tp)); 3382 } 3383 3384 /* Don't bother if destination was broadcast/multicast. */ 3385 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3386 goto drop; 3387 #ifdef INET6 3388 if (mtod(m, struct ip *)->ip_v == 6) { 3389 ip6 = mtod(m, struct ip6_hdr *); 3390 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3391 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3392 goto drop; 3393 /* IPv6 anycast check is done at tcp6_input() */ 3394 } 3395 #endif 3396 #if defined(INET) && defined(INET6) 3397 else 3398 #endif 3399 #ifdef INET 3400 { 3401 ip = mtod(m, struct ip *); 3402 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3403 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3404 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3405 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3406 goto drop; 3407 } 3408 #endif 3409 3410 /* Perform bandwidth limiting. */ 3411 if (badport_bandlim(rstreason) < 0) 3412 goto drop; 3413 3414 /* tcp_respond consumes the mbuf chain. */ 3415 if (tcp_get_flags(th) & TH_ACK) { 3416 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3417 th->th_ack, TH_RST); 3418 } else { 3419 if (tcp_get_flags(th) & TH_SYN) 3420 tlen++; 3421 if (tcp_get_flags(th) & TH_FIN) 3422 tlen++; 3423 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3424 (tcp_seq)0, TH_RST|TH_ACK); 3425 } 3426 return; 3427 drop: 3428 m_freem(m); 3429 } 3430 3431 /* 3432 * Parse TCP options and place in tcpopt. 3433 */ 3434 void 3435 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3436 { 3437 int opt, optlen; 3438 3439 to->to_flags = 0; 3440 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3441 opt = cp[0]; 3442 if (opt == TCPOPT_EOL) 3443 break; 3444 if (opt == TCPOPT_NOP) 3445 optlen = 1; 3446 else { 3447 if (cnt < 2) 3448 break; 3449 optlen = cp[1]; 3450 if (optlen < 2 || optlen > cnt) 3451 break; 3452 } 3453 switch (opt) { 3454 case TCPOPT_MAXSEG: 3455 if (optlen != TCPOLEN_MAXSEG) 3456 continue; 3457 if (!(flags & TO_SYN)) 3458 continue; 3459 to->to_flags |= TOF_MSS; 3460 bcopy((char *)cp + 2, 3461 (char *)&to->to_mss, sizeof(to->to_mss)); 3462 to->to_mss = ntohs(to->to_mss); 3463 break; 3464 case TCPOPT_WINDOW: 3465 if (optlen != TCPOLEN_WINDOW) 3466 continue; 3467 if (!(flags & TO_SYN)) 3468 continue; 3469 to->to_flags |= TOF_SCALE; 3470 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3471 break; 3472 case TCPOPT_TIMESTAMP: 3473 if (optlen != TCPOLEN_TIMESTAMP) 3474 continue; 3475 to->to_flags |= TOF_TS; 3476 bcopy((char *)cp + 2, 3477 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3478 to->to_tsval = ntohl(to->to_tsval); 3479 bcopy((char *)cp + 6, 3480 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3481 to->to_tsecr = ntohl(to->to_tsecr); 3482 break; 3483 case TCPOPT_SIGNATURE: 3484 /* 3485 * In order to reply to a host which has set the 3486 * TCP_SIGNATURE option in its initial SYN, we have 3487 * to record the fact that the option was observed 3488 * here for the syncache code to perform the correct 3489 * response. 3490 */ 3491 if (optlen != TCPOLEN_SIGNATURE) 3492 continue; 3493 to->to_flags |= TOF_SIGNATURE; 3494 to->to_signature = cp + 2; 3495 break; 3496 case TCPOPT_SACK_PERMITTED: 3497 if (optlen != TCPOLEN_SACK_PERMITTED) 3498 continue; 3499 if (!(flags & TO_SYN)) 3500 continue; 3501 if (!V_tcp_do_sack) 3502 continue; 3503 to->to_flags |= TOF_SACKPERM; 3504 break; 3505 case TCPOPT_SACK: 3506 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3507 continue; 3508 if (flags & TO_SYN) 3509 continue; 3510 to->to_flags |= TOF_SACK; 3511 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3512 to->to_sacks = cp + 2; 3513 TCPSTAT_INC(tcps_sack_rcv_blocks); 3514 break; 3515 case TCPOPT_FAST_OPEN: 3516 /* 3517 * Cookie length validation is performed by the 3518 * server side cookie checking code or the client 3519 * side cookie cache update code. 3520 */ 3521 if (!(flags & TO_SYN)) 3522 continue; 3523 if (!V_tcp_fastopen_client_enable && 3524 !V_tcp_fastopen_server_enable) 3525 continue; 3526 to->to_flags |= TOF_FASTOPEN; 3527 to->to_tfo_len = optlen - 2; 3528 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3529 break; 3530 default: 3531 continue; 3532 } 3533 } 3534 } 3535 3536 /* 3537 * Pull out of band byte out of a segment so 3538 * it doesn't appear in the user's data queue. 3539 * It is still reflected in the segment length for 3540 * sequencing purposes. 3541 */ 3542 void 3543 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3544 int off) 3545 { 3546 int cnt = off + th->th_urp - 1; 3547 3548 while (cnt >= 0) { 3549 if (m->m_len > cnt) { 3550 char *cp = mtod(m, caddr_t) + cnt; 3551 struct tcpcb *tp = sototcpcb(so); 3552 3553 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3554 3555 tp->t_iobc = *cp; 3556 tp->t_oobflags |= TCPOOB_HAVEDATA; 3557 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3558 m->m_len--; 3559 if (m->m_flags & M_PKTHDR) 3560 m->m_pkthdr.len--; 3561 return; 3562 } 3563 cnt -= m->m_len; 3564 m = m->m_next; 3565 if (m == NULL) 3566 break; 3567 } 3568 panic("tcp_pulloutofband"); 3569 } 3570 3571 /* 3572 * Collect new round-trip time estimate 3573 * and update averages and current timeout. 3574 */ 3575 void 3576 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3577 { 3578 int delta; 3579 3580 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3581 3582 TCPSTAT_INC(tcps_rttupdated); 3583 if (tp->t_rttupdated < UCHAR_MAX) 3584 tp->t_rttupdated++; 3585 #ifdef STATS 3586 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3587 imax(0, rtt * 1000 / hz)); 3588 #endif 3589 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3590 /* 3591 * srtt is stored as fixed point with 5 bits after the 3592 * binary point (i.e., scaled by 8). The following magic 3593 * is equivalent to the smoothing algorithm in rfc793 with 3594 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3595 * point). Adjust rtt to origin 0. 3596 */ 3597 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3598 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3599 3600 if ((tp->t_srtt += delta) <= 0) 3601 tp->t_srtt = 1; 3602 3603 /* 3604 * We accumulate a smoothed rtt variance (actually, a 3605 * smoothed mean difference), then set the retransmit 3606 * timer to smoothed rtt + 4 times the smoothed variance. 3607 * rttvar is stored as fixed point with 4 bits after the 3608 * binary point (scaled by 16). The following is 3609 * equivalent to rfc793 smoothing with an alpha of .75 3610 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3611 * rfc793's wired-in beta. 3612 */ 3613 if (delta < 0) 3614 delta = -delta; 3615 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3616 if ((tp->t_rttvar += delta) <= 0) 3617 tp->t_rttvar = 1; 3618 } else { 3619 /* 3620 * No rtt measurement yet - use the unsmoothed rtt. 3621 * Set the variance to half the rtt (so our first 3622 * retransmit happens at 3*rtt). 3623 */ 3624 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3625 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3626 } 3627 tp->t_rtttime = 0; 3628 tp->t_rxtshift = 0; 3629 3630 /* 3631 * the retransmit should happen at rtt + 4 * rttvar. 3632 * Because of the way we do the smoothing, srtt and rttvar 3633 * will each average +1/2 tick of bias. When we compute 3634 * the retransmit timer, we want 1/2 tick of rounding and 3635 * 1 extra tick because of +-1/2 tick uncertainty in the 3636 * firing of the timer. The bias will give us exactly the 3637 * 1.5 tick we need. But, because the bias is 3638 * statistical, we have to test that we don't drop below 3639 * the minimum feasible timer (which is 2 ticks). 3640 */ 3641 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3642 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3643 3644 /* 3645 * We received an ack for a packet that wasn't retransmitted; 3646 * it is probably safe to discard any error indications we've 3647 * received recently. This isn't quite right, but close enough 3648 * for now (a route might have failed after we sent a segment, 3649 * and the return path might not be symmetrical). 3650 */ 3651 tp->t_softerror = 0; 3652 } 3653 3654 /* 3655 * Determine a reasonable value for maxseg size. 3656 * If the route is known, check route for mtu. 3657 * If none, use an mss that can be handled on the outgoing interface 3658 * without forcing IP to fragment. If no route is found, route has no mtu, 3659 * or the destination isn't local, use a default, hopefully conservative 3660 * size (usually 512 or the default IP max size, but no more than the mtu 3661 * of the interface), as we can't discover anything about intervening 3662 * gateways or networks. We also initialize the congestion/slow start 3663 * window to be a single segment if the destination isn't local. 3664 * While looking at the routing entry, we also initialize other path-dependent 3665 * parameters from pre-set or cached values in the routing entry. 3666 * 3667 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3668 * IP options, e.g. IPSEC data, since length of this data may vary, and 3669 * thus it is calculated for every segment separately in tcp_output(). 3670 * 3671 * NOTE that this routine is only called when we process an incoming 3672 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3673 * settings are handled in tcp_mssopt(). 3674 */ 3675 void 3676 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3677 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3678 { 3679 int mss = 0; 3680 uint32_t maxmtu = 0; 3681 struct inpcb *inp = tptoinpcb(tp); 3682 struct hc_metrics_lite metrics; 3683 #ifdef INET6 3684 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3685 size_t min_protoh = isipv6 ? 3686 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3687 sizeof (struct tcpiphdr); 3688 #else 3689 size_t min_protoh = sizeof(struct tcpiphdr); 3690 #endif 3691 3692 INP_WLOCK_ASSERT(inp); 3693 3694 if (tp->t_port) 3695 min_protoh += V_tcp_udp_tunneling_overhead; 3696 if (mtuoffer != -1) { 3697 KASSERT(offer == -1, ("%s: conflict", __func__)); 3698 offer = mtuoffer - min_protoh; 3699 } 3700 3701 /* Initialize. */ 3702 #ifdef INET6 3703 if (isipv6) { 3704 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3705 tp->t_maxseg = V_tcp_v6mssdflt; 3706 } 3707 #endif 3708 #if defined(INET) && defined(INET6) 3709 else 3710 #endif 3711 #ifdef INET 3712 { 3713 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3714 tp->t_maxseg = V_tcp_mssdflt; 3715 } 3716 #endif 3717 3718 /* 3719 * No route to sender, stay with default mss and return. 3720 */ 3721 if (maxmtu == 0) { 3722 /* 3723 * In case we return early we need to initialize metrics 3724 * to a defined state as tcp_hc_get() would do for us 3725 * if there was no cache hit. 3726 */ 3727 if (metricptr != NULL) 3728 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3729 return; 3730 } 3731 3732 /* What have we got? */ 3733 switch (offer) { 3734 case 0: 3735 /* 3736 * Offer == 0 means that there was no MSS on the SYN 3737 * segment, in this case we use tcp_mssdflt as 3738 * already assigned to t_maxseg above. 3739 */ 3740 offer = tp->t_maxseg; 3741 break; 3742 3743 case -1: 3744 /* 3745 * Offer == -1 means that we didn't receive SYN yet. 3746 */ 3747 /* FALLTHROUGH */ 3748 3749 default: 3750 /* 3751 * Prevent DoS attack with too small MSS. Round up 3752 * to at least minmss. 3753 */ 3754 offer = max(offer, V_tcp_minmss); 3755 } 3756 3757 /* 3758 * rmx information is now retrieved from tcp_hostcache. 3759 */ 3760 tcp_hc_get(&inp->inp_inc, &metrics); 3761 if (metricptr != NULL) 3762 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3763 3764 /* 3765 * If there's a discovered mtu in tcp hostcache, use it. 3766 * Else, use the link mtu. 3767 */ 3768 if (metrics.rmx_mtu) 3769 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3770 else { 3771 #ifdef INET6 3772 if (isipv6) { 3773 mss = maxmtu - min_protoh; 3774 if (!V_path_mtu_discovery && 3775 !in6_localaddr(&inp->in6p_faddr)) 3776 mss = min(mss, V_tcp_v6mssdflt); 3777 } 3778 #endif 3779 #if defined(INET) && defined(INET6) 3780 else 3781 #endif 3782 #ifdef INET 3783 { 3784 mss = maxmtu - min_protoh; 3785 if (!V_path_mtu_discovery && 3786 !in_localaddr(inp->inp_faddr)) 3787 mss = min(mss, V_tcp_mssdflt); 3788 } 3789 #endif 3790 /* 3791 * XXX - The above conditional (mss = maxmtu - min_protoh) 3792 * probably violates the TCP spec. 3793 * The problem is that, since we don't know the 3794 * other end's MSS, we are supposed to use a conservative 3795 * default. But, if we do that, then MTU discovery will 3796 * never actually take place, because the conservative 3797 * default is much less than the MTUs typically seen 3798 * on the Internet today. For the moment, we'll sweep 3799 * this under the carpet. 3800 * 3801 * The conservative default might not actually be a problem 3802 * if the only case this occurs is when sending an initial 3803 * SYN with options and data to a host we've never talked 3804 * to before. Then, they will reply with an MSS value which 3805 * will get recorded and the new parameters should get 3806 * recomputed. For Further Study. 3807 */ 3808 } 3809 mss = min(mss, offer); 3810 3811 /* 3812 * Sanity check: make sure that maxseg will be large 3813 * enough to allow some data on segments even if the 3814 * all the option space is used (40bytes). Otherwise 3815 * funny things may happen in tcp_output. 3816 * 3817 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3818 */ 3819 mss = max(mss, 64); 3820 3821 tp->t_maxseg = mss; 3822 } 3823 3824 void 3825 tcp_mss(struct tcpcb *tp, int offer) 3826 { 3827 int mss; 3828 uint32_t bufsize; 3829 struct inpcb *inp = tptoinpcb(tp); 3830 struct socket *so; 3831 struct hc_metrics_lite metrics; 3832 struct tcp_ifcap cap; 3833 3834 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3835 3836 bzero(&cap, sizeof(cap)); 3837 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3838 3839 mss = tp->t_maxseg; 3840 3841 /* 3842 * If there's a pipesize, change the socket buffer to that size, 3843 * don't change if sb_hiwat is different than default (then it 3844 * has been changed on purpose with setsockopt). 3845 * Make the socket buffers an integral number of mss units; 3846 * if the mss is larger than the socket buffer, decrease the mss. 3847 */ 3848 so = inp->inp_socket; 3849 SOCKBUF_LOCK(&so->so_snd); 3850 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3851 bufsize = metrics.rmx_sendpipe; 3852 else 3853 bufsize = so->so_snd.sb_hiwat; 3854 if (bufsize < mss) 3855 mss = bufsize; 3856 else { 3857 bufsize = roundup(bufsize, mss); 3858 if (bufsize > sb_max) 3859 bufsize = sb_max; 3860 if (bufsize > so->so_snd.sb_hiwat) 3861 (void)sbreserve_locked(so, SO_SND, bufsize, NULL); 3862 } 3863 SOCKBUF_UNLOCK(&so->so_snd); 3864 /* 3865 * Sanity check: make sure that maxseg will be large 3866 * enough to allow some data on segments even if the 3867 * all the option space is used (40bytes). Otherwise 3868 * funny things may happen in tcp_output. 3869 * 3870 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3871 */ 3872 tp->t_maxseg = max(mss, 64); 3873 3874 SOCKBUF_LOCK(&so->so_rcv); 3875 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3876 bufsize = metrics.rmx_recvpipe; 3877 else 3878 bufsize = so->so_rcv.sb_hiwat; 3879 if (bufsize > mss) { 3880 bufsize = roundup(bufsize, mss); 3881 if (bufsize > sb_max) 3882 bufsize = sb_max; 3883 if (bufsize > so->so_rcv.sb_hiwat) 3884 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL); 3885 } 3886 SOCKBUF_UNLOCK(&so->so_rcv); 3887 3888 /* Check the interface for TSO capabilities. */ 3889 if (cap.ifcap & CSUM_TSO) { 3890 tp->t_flags |= TF_TSO; 3891 tp->t_tsomax = cap.tsomax; 3892 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3893 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3894 } 3895 } 3896 3897 /* 3898 * Determine the MSS option to send on an outgoing SYN. 3899 */ 3900 int 3901 tcp_mssopt(struct in_conninfo *inc) 3902 { 3903 int mss = 0; 3904 uint32_t thcmtu = 0; 3905 uint32_t maxmtu = 0; 3906 size_t min_protoh; 3907 3908 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3909 3910 #ifdef INET6 3911 if (inc->inc_flags & INC_ISIPV6) { 3912 mss = V_tcp_v6mssdflt; 3913 maxmtu = tcp_maxmtu6(inc, NULL); 3914 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3915 } 3916 #endif 3917 #if defined(INET) && defined(INET6) 3918 else 3919 #endif 3920 #ifdef INET 3921 { 3922 mss = V_tcp_mssdflt; 3923 maxmtu = tcp_maxmtu(inc, NULL); 3924 min_protoh = sizeof(struct tcpiphdr); 3925 } 3926 #endif 3927 #if defined(INET6) || defined(INET) 3928 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3929 #endif 3930 3931 if (maxmtu && thcmtu) 3932 mss = min(maxmtu, thcmtu) - min_protoh; 3933 else if (maxmtu || thcmtu) 3934 mss = max(maxmtu, thcmtu) - min_protoh; 3935 3936 return (mss); 3937 } 3938 3939 void 3940 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, sackstatus_t sack_changed) 3941 { 3942 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3943 int maxseg = tcp_maxseg(tp); 3944 3945 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3946 3947 /* 3948 * Compute the amount of data that this ACK is indicating 3949 * (del_data) and an estimate of how many bytes are in the 3950 * network. 3951 */ 3952 if (tcp_is_sack_recovery(tp, to) || 3953 (IN_CONGRECOVERY(tp->t_flags) && 3954 !IN_FASTRECOVERY(tp->t_flags))) { 3955 del_data = tp->sackhint.delivered_data; 3956 if (V_tcp_do_newsack) 3957 pipe = tcp_compute_pipe(tp); 3958 else 3959 pipe = (tp->snd_nxt - tp->snd_fack) + 3960 tp->sackhint.sack_bytes_rexmit; 3961 } else { 3962 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + 3963 tp->snd_recover - tp->snd_una)) 3964 del_data = maxseg; 3965 pipe = imax(0, tp->snd_max - tp->snd_una - 3966 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); 3967 } 3968 tp->sackhint.prr_delivered += del_data; 3969 /* 3970 * Proportional Rate Reduction 3971 */ 3972 if (pipe >= tp->snd_ssthresh) { 3973 if (tp->sackhint.recover_fs == 0) 3974 tp->sackhint.recover_fs = 3975 imax(1, tp->snd_nxt - tp->snd_una); 3976 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 3977 tp->snd_ssthresh, tp->sackhint.recover_fs) - 3978 tp->sackhint.prr_out; 3979 } else { 3980 /* 3981 * PRR 6937bis heuristic: 3982 * - A partial ack without SACK block beneath snd_recover 3983 * indicates further loss. 3984 * - An SACK scoreboard update adding a new hole indicates 3985 * further loss, so be conservative and send at most one 3986 * segment. 3987 * - Prevent ACK splitting attacks, by being conservative 3988 * when no new data is acked. 3989 */ 3990 if ((sack_changed == SACK_NEWLOSS) || (del_data == 0)) 3991 limit = tp->sackhint.prr_delivered - 3992 tp->sackhint.prr_out; 3993 else 3994 limit = imax(tp->sackhint.prr_delivered - 3995 tp->sackhint.prr_out, del_data) + 3996 maxseg; 3997 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 3998 } 3999 snd_cnt = imax(snd_cnt, 0) / maxseg; 4000 /* 4001 * Send snd_cnt new data into the network in response to this ack. 4002 * If there is going to be a SACK retransmission, adjust snd_cwnd 4003 * accordingly. 4004 */ 4005 if (IN_FASTRECOVERY(tp->t_flags)) { 4006 if (tcp_is_sack_recovery(tp, to)) { 4007 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + 4008 tp->sackhint.sack_bytes_rexmit + 4009 (snd_cnt * maxseg); 4010 } else { 4011 tp->snd_cwnd = (tp->snd_max - tp->snd_una) + 4012 (snd_cnt * maxseg); 4013 } 4014 } else if (IN_CONGRECOVERY(tp->t_flags)) 4015 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4016 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); 4017 } 4018 4019 /* 4020 * On a partial ack arrives, force the retransmission of the 4021 * next unacknowledged segment. Do not clear tp->t_dupacks. 4022 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4023 * be started again. 4024 */ 4025 void 4026 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4027 { 4028 tcp_seq onxt = tp->snd_nxt; 4029 uint32_t ocwnd = tp->snd_cwnd; 4030 u_int maxseg = tcp_maxseg(tp); 4031 4032 INP_WLOCK_ASSERT(tptoinpcb(tp)); 4033 4034 tcp_timer_activate(tp, TT_REXMT, 0); 4035 tp->t_rtttime = 0; 4036 tp->snd_nxt = th->th_ack; 4037 /* 4038 * Set snd_cwnd to one segment beyond acknowledged offset. 4039 * (tp->snd_una has not yet been updated when this function is called.) 4040 */ 4041 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4042 tp->t_flags |= TF_ACKNOW; 4043 (void) tcp_output(tp); 4044 tp->snd_cwnd = ocwnd; 4045 if (SEQ_GT(onxt, tp->snd_nxt)) 4046 tp->snd_nxt = onxt; 4047 /* 4048 * Partial window deflation. Relies on fact that tp->snd_una 4049 * not updated yet. 4050 */ 4051 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4052 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4053 else 4054 tp->snd_cwnd = 0; 4055 tp->snd_cwnd += maxseg; 4056 } 4057 4058 int 4059 tcp_compute_pipe(struct tcpcb *tp) 4060 { 4061 if (tp->t_fb->tfb_compute_pipe == NULL) { 4062 return (tp->snd_max - tp->snd_una + 4063 tp->sackhint.sack_bytes_rexmit - 4064 tp->sackhint.sacked_bytes - 4065 tp->sackhint.lost_bytes); 4066 } else { 4067 return((*tp->t_fb->tfb_compute_pipe)(tp)); 4068 } 4069 } 4070 4071 uint32_t 4072 tcp_compute_initwnd(uint32_t maxseg) 4073 { 4074 /* 4075 * Calculate the Initial Window, also used as Restart Window 4076 * 4077 * RFC5681 Section 3.1 specifies the default conservative values. 4078 * RFC3390 specifies slightly more aggressive values. 4079 * RFC6928 increases it to ten segments. 4080 * Support for user specified value for initial flight size. 4081 */ 4082 if (V_tcp_initcwnd_segments) 4083 return min(V_tcp_initcwnd_segments * maxseg, 4084 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4085 else if (V_tcp_do_rfc3390) 4086 return min(4 * maxseg, max(2 * maxseg, 4380)); 4087 else { 4088 /* Per RFC5681 Section 3.1 */ 4089 if (maxseg > 2190) 4090 return (2 * maxseg); 4091 else if (maxseg > 1095) 4092 return (3 * maxseg); 4093 else 4094 return (4 * maxseg); 4095 } 4096 } 4097