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 <[email protected]> 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 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 50 */ 51 52 #include <sys/cdefs.h> 53 #include "opt_inet.h" 54 #include "opt_inet6.h" 55 #include "opt_ipsec.h" 56 #include "opt_rss.h" 57 58 #include <sys/param.h> 59 #include <sys/arb.h> 60 #include <sys/kernel.h> 61 #ifdef TCP_HHOOK 62 #include <sys/hhook.h> 63 #endif 64 #include <sys/malloc.h> 65 #include <sys/mbuf.h> 66 #include <sys/proc.h> /* for proc0 declaration */ 67 #include <sys/protosw.h> 68 #include <sys/qmath.h> 69 #include <sys/sdt.h> 70 #include <sys/signalvar.h> 71 #include <sys/socket.h> 72 #include <sys/socketvar.h> 73 #include <sys/sysctl.h> 74 #include <sys/syslog.h> 75 #include <sys/systm.h> 76 #include <sys/stats.h> 77 78 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 79 80 #include <vm/uma.h> 81 82 #include <net/if.h> 83 #include <net/if_var.h> 84 #include <net/route.h> 85 #include <net/rss_config.h> 86 #include <net/vnet.h> 87 88 #define TCPSTATES /* for logging */ 89 90 #include <netinet/in.h> 91 #include <netinet/in_kdtrace.h> 92 #include <netinet/in_pcb.h> 93 #include <netinet/in_rss.h> 94 #include <netinet/in_systm.h> 95 #include <netinet/ip.h> 96 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 97 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 98 #include <netinet/ip_var.h> 99 #include <netinet/ip_options.h> 100 #include <netinet/ip6.h> 101 #include <netinet/icmp6.h> 102 #include <netinet6/in6_pcb.h> 103 #include <netinet6/in6_rss.h> 104 #include <netinet6/in6_var.h> 105 #include <netinet6/ip6_var.h> 106 #include <netinet6/nd6.h> 107 #include <netinet/tcp.h> 108 #include <netinet/tcp_fsm.h> 109 #include <netinet/tcp_seq.h> 110 #include <netinet/tcp_timer.h> 111 #include <netinet/tcp_var.h> 112 #include <netinet/tcp_log_buf.h> 113 #include <netinet6/tcp6_var.h> 114 #include <netinet/tcpip.h> 115 #include <netinet/cc/cc.h> 116 #include <netinet/tcp_fastopen.h> 117 #ifdef TCPPCAP 118 #include <netinet/tcp_pcap.h> 119 #endif 120 #include <netinet/tcp_syncache.h> 121 #ifdef TCP_OFFLOAD 122 #include <netinet/tcp_offload.h> 123 #endif 124 #include <netinet/tcp_ecn.h> 125 #include <netinet/udp.h> 126 127 #include <netipsec/ipsec_support.h> 128 129 #include <machine/in_cksum.h> 130 131 #include <security/mac/mac_framework.h> 132 133 const int tcprexmtthresh = 3; 134 135 VNET_DEFINE(int, tcp_log_in_vain) = 0; 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 137 &VNET_NAME(tcp_log_in_vain), 0, 138 "Log all incoming TCP segments to closed ports"); 139 140 VNET_DEFINE(int, blackhole) = 0; 141 #define V_blackhole VNET(blackhole) 142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 143 &VNET_NAME(blackhole), 0, 144 "Do not send RST on segments to closed ports"); 145 146 VNET_DEFINE(bool, blackhole_local) = false; 147 #define V_blackhole_local VNET(blackhole_local) 148 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET | 149 CTLFLAG_RW, &VNET_NAME(blackhole_local), false, 150 "Enforce net.inet.tcp.blackhole for locally originated packets"); 151 152 VNET_DEFINE(int, tcp_delack_enabled) = 1; 153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 154 &VNET_NAME(tcp_delack_enabled), 0, 155 "Delay ACK to try and piggyback it onto a data packet"); 156 157 VNET_DEFINE(int, drop_synfin) = 0; 158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 159 &VNET_NAME(drop_synfin), 0, 160 "Drop TCP packets with SYN+FIN set"); 161 162 VNET_DEFINE(int, tcp_do_prr) = 1; 163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 164 &VNET_NAME(tcp_do_prr), 1, 165 "Enable Proportional Rate Reduction per RFC 6937"); 166 167 VNET_DEFINE(int, tcp_do_lrd) = 0; 168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW, 169 &VNET_NAME(tcp_do_lrd), 1, 170 "Perform Lost Retransmission Detection"); 171 172 VNET_DEFINE(int, tcp_do_newcwv) = 0; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 174 &VNET_NAME(tcp_do_newcwv), 0, 175 "Enable New Congestion Window Validation per RFC7661"); 176 177 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 179 &VNET_NAME(tcp_do_rfc3042), 0, 180 "Enable RFC 3042 (Limited Transmit)"); 181 182 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 184 &VNET_NAME(tcp_do_rfc3390), 0, 185 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 186 187 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 189 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 190 "Slow-start flight size (initial congestion window) in number of segments"); 191 192 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 194 &VNET_NAME(tcp_do_rfc3465), 0, 195 "Enable RFC 3465 (Appropriate Byte Counting)"); 196 197 VNET_DEFINE(int, tcp_abc_l_var) = 2; 198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 199 &VNET_NAME(tcp_abc_l_var), 2, 200 "Cap the max cwnd increment during slow-start to this number of segments"); 201 202 VNET_DEFINE(int, tcp_insecure_syn) = 0; 203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 204 &VNET_NAME(tcp_insecure_syn), 0, 205 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 206 207 VNET_DEFINE(int, tcp_insecure_rst) = 0; 208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 209 &VNET_NAME(tcp_insecure_rst), 0, 210 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 211 212 VNET_DEFINE(int, tcp_insecure_ack) = 0; 213 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_ack, CTLFLAG_VNET | CTLFLAG_RW, 214 &VNET_NAME(tcp_insecure_ack), 0, 215 "Follow RFC793 criteria for validating SEG.ACK"); 216 217 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 218 #define V_tcp_recvspace VNET(tcp_recvspace) 219 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 220 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 221 222 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 224 &VNET_NAME(tcp_do_autorcvbuf), 0, 225 "Enable automatic receive buffer sizing"); 226 227 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 229 &VNET_NAME(tcp_autorcvbuf_max), 0, 230 "Max size of automatic receive buffer"); 231 232 VNET_DEFINE(struct inpcbinfo, tcbinfo); 233 234 /* 235 * TCP statistics are stored in an array of counter(9)s, which size matches 236 * size of struct tcpstat. TCP running connection count is a regular array. 237 */ 238 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 239 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 240 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 241 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 242 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 243 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 244 "TCP connection counts by TCP state"); 245 246 /* 247 * Kernel module interface for updating tcpstat. The first argument is an index 248 * into tcpstat treated as an array. 249 */ 250 void 251 kmod_tcpstat_add(int statnum, int val) 252 { 253 254 counter_u64_add(VNET(tcpstat)[statnum], val); 255 } 256 257 /* 258 * Make sure that we only start a SACK loss recovery when 259 * receiving a duplicate ACK with a SACK block, and also 260 * complete SACK loss recovery in case the other end 261 * reneges. 262 */ 263 static bool inline 264 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to) 265 { 266 return ((tp->t_flags & TF_SACK_PERMIT) && 267 ((to->to_flags & TOF_SACK) || 268 (!TAILQ_EMPTY(&tp->snd_holes)))); 269 } 270 271 #ifdef TCP_HHOOK 272 /* 273 * Wrapper for the TCP established input helper hook. 274 */ 275 void 276 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 277 { 278 struct tcp_hhook_data hhook_data; 279 280 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 281 hhook_data.tp = tp; 282 hhook_data.th = th; 283 hhook_data.to = to; 284 285 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 286 &tp->t_osd); 287 } 288 } 289 #endif 290 291 /* 292 * CC wrapper hook functions 293 */ 294 void 295 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 296 uint16_t type) 297 { 298 #ifdef STATS 299 int32_t gput; 300 #endif 301 302 INP_WLOCK_ASSERT(tptoinpcb(tp)); 303 304 tp->t_ccv.nsegs = nsegs; 305 tp->t_ccv.bytes_this_ack = BYTES_THIS_ACK(tp, th); 306 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 307 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 308 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 309 tp->t_ccv.flags |= CCF_CWND_LIMITED; 310 else 311 tp->t_ccv.flags &= ~CCF_CWND_LIMITED; 312 313 if (type == CC_ACK) { 314 #ifdef STATS 315 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 316 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 317 if (!IN_RECOVERY(tp->t_flags)) 318 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 319 tp->t_ccv.bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 320 if ((tp->t_flags & TF_GPUTINPROG) && 321 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 322 /* 323 * Compute goodput in bits per millisecond. 324 */ 325 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) / 326 max(1, tcp_ts_getticks() - tp->gput_ts); 327 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 328 gput); 329 /* 330 * XXXLAS: This is a temporary hack, and should be 331 * chained off VOI_TCP_GPUT when stats(9) grows an API 332 * to deal with chained VOIs. 333 */ 334 if (tp->t_stats_gput_prev > 0) 335 stats_voi_update_abs_s32(tp->t_stats, 336 VOI_TCP_GPUT_ND, 337 ((gput - tp->t_stats_gput_prev) * 100) / 338 tp->t_stats_gput_prev); 339 tp->t_flags &= ~TF_GPUTINPROG; 340 tp->t_stats_gput_prev = gput; 341 } 342 #endif /* STATS */ 343 if (tp->snd_cwnd > tp->snd_ssthresh) { 344 tp->t_bytes_acked += tp->t_ccv.bytes_this_ack; 345 if (tp->t_bytes_acked >= tp->snd_cwnd) { 346 tp->t_bytes_acked -= tp->snd_cwnd; 347 tp->t_ccv.flags |= CCF_ABC_SENTAWND; 348 } 349 } else { 350 tp->t_ccv.flags &= ~CCF_ABC_SENTAWND; 351 tp->t_bytes_acked = 0; 352 } 353 } 354 355 if (CC_ALGO(tp)->ack_received != NULL) { 356 /* XXXLAS: Find a way to live without this */ 357 tp->t_ccv.curack = th->th_ack; 358 CC_ALGO(tp)->ack_received(&tp->t_ccv, type); 359 } 360 #ifdef STATS 361 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 362 #endif 363 } 364 365 void 366 cc_conn_init(struct tcpcb *tp) 367 { 368 struct hc_metrics_lite metrics; 369 struct inpcb *inp = tptoinpcb(tp); 370 u_int maxseg; 371 int rtt; 372 373 INP_WLOCK_ASSERT(inp); 374 375 tcp_hc_get(&inp->inp_inc, &metrics); 376 maxseg = tcp_maxseg(tp); 377 378 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 379 tp->t_srtt = rtt; 380 TCPSTAT_INC(tcps_usedrtt); 381 if (metrics.rmx_rttvar) { 382 tp->t_rttvar = metrics.rmx_rttvar; 383 TCPSTAT_INC(tcps_usedrttvar); 384 } else { 385 /* default variation is +- 1 rtt */ 386 tp->t_rttvar = 387 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 388 } 389 TCPT_RANGESET(tp->t_rxtcur, 390 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 391 tp->t_rttmin, TCPTV_REXMTMAX); 392 } 393 if (metrics.rmx_ssthresh) { 394 /* 395 * There's some sort of gateway or interface 396 * buffer limit on the path. Use this to set 397 * the slow start threshold, but set the 398 * threshold to no less than 2*mss. 399 */ 400 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 401 TCPSTAT_INC(tcps_usedssthresh); 402 } 403 404 /* 405 * Set the initial slow-start flight size. 406 * 407 * If a SYN or SYN/ACK was lost and retransmitted, we have to 408 * reduce the initial CWND to one segment as congestion is likely 409 * requiring us to be cautious. 410 */ 411 if (tp->snd_cwnd == 1) 412 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 413 else 414 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 415 416 if (CC_ALGO(tp)->conn_init != NULL) 417 CC_ALGO(tp)->conn_init(&tp->t_ccv); 418 } 419 420 void inline 421 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 422 { 423 INP_WLOCK_ASSERT(tptoinpcb(tp)); 424 425 #ifdef STATS 426 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 427 #endif 428 429 switch(type) { 430 case CC_NDUPACK: 431 if (!IN_FASTRECOVERY(tp->t_flags)) { 432 tp->snd_recover = tp->snd_max; 433 if (tp->t_flags2 & TF2_ECN_PERMIT) 434 tp->t_flags2 |= TF2_ECN_SND_CWR; 435 } 436 break; 437 case CC_ECN: 438 if (!IN_CONGRECOVERY(tp->t_flags) || 439 /* 440 * Allow ECN reaction on ACK to CWR, if 441 * that data segment was also CE marked. 442 */ 443 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 444 EXIT_CONGRECOVERY(tp->t_flags); 445 TCPSTAT_INC(tcps_ecn_rcwnd); 446 tp->snd_recover = tp->snd_max + 1; 447 if (tp->t_flags2 & TF2_ECN_PERMIT) 448 tp->t_flags2 |= TF2_ECN_SND_CWR; 449 } 450 break; 451 case CC_RTO: 452 tp->t_dupacks = 0; 453 tp->t_bytes_acked = 0; 454 EXIT_RECOVERY(tp->t_flags); 455 if (tp->t_flags2 & TF2_ECN_PERMIT) 456 tp->t_flags2 |= TF2_ECN_SND_CWR; 457 break; 458 case CC_RTO_ERR: 459 TCPSTAT_INC(tcps_sndrexmitbad); 460 /* RTO was unnecessary, so reset everything. */ 461 tp->snd_cwnd = tp->snd_cwnd_prev; 462 tp->snd_ssthresh = tp->snd_ssthresh_prev; 463 tp->snd_recover = tp->snd_recover_prev; 464 if (tp->t_flags & TF_WASFRECOVERY) 465 ENTER_FASTRECOVERY(tp->t_flags); 466 if (tp->t_flags & TF_WASCRECOVERY) 467 ENTER_CONGRECOVERY(tp->t_flags); 468 tp->snd_nxt = tp->snd_max; 469 tp->t_flags &= ~TF_PREVVALID; 470 tp->t_badrxtwin = 0; 471 break; 472 } 473 474 if (CC_ALGO(tp)->cong_signal != NULL) { 475 if (th != NULL) 476 tp->t_ccv.curack = th->th_ack; 477 CC_ALGO(tp)->cong_signal(&tp->t_ccv, type); 478 } 479 } 480 481 void inline 482 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 483 { 484 INP_WLOCK_ASSERT(tptoinpcb(tp)); 485 486 if (CC_ALGO(tp)->post_recovery != NULL) { 487 tp->t_ccv.curack = th->th_ack; 488 CC_ALGO(tp)->post_recovery(&tp->t_ccv); 489 } 490 EXIT_RECOVERY(tp->t_flags); 491 492 tp->t_bytes_acked = 0; 493 tp->sackhint.delivered_data = 0; 494 tp->sackhint.prr_delivered = 0; 495 tp->sackhint.prr_out = 0; 496 } 497 498 /* 499 * Indicate whether this ack should be delayed. We can delay the ack if 500 * following conditions are met: 501 * - There is no delayed ack timer in progress. 502 * - Our last ack wasn't a 0-sized window. We never want to delay 503 * the ack that opens up a 0-sized window. 504 * - LRO wasn't used for this segment. We make sure by checking that the 505 * segment size is not larger than the MSS. 506 */ 507 #define DELAY_ACK(tp, tlen) \ 508 ((!tcp_timer_active(tp, TT_DELACK) && \ 509 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 510 (tlen <= tp->t_maxseg) && \ 511 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 512 513 void inline 514 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) 515 { 516 INP_WLOCK_ASSERT(tptoinpcb(tp)); 517 518 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 519 switch (iptos & IPTOS_ECN_MASK) { 520 case IPTOS_ECN_CE: 521 tp->t_ccv.flags |= CCF_IPHDR_CE; 522 break; 523 case IPTOS_ECN_ECT0: 524 /* FALLTHROUGH */ 525 case IPTOS_ECN_ECT1: 526 /* FALLTHROUGH */ 527 case IPTOS_ECN_NOTECT: 528 tp->t_ccv.flags &= ~CCF_IPHDR_CE; 529 break; 530 } 531 532 if (flags & TH_CWR) 533 tp->t_ccv.flags |= CCF_TCPHDR_CWR; 534 else 535 tp->t_ccv.flags &= ~CCF_TCPHDR_CWR; 536 537 CC_ALGO(tp)->ecnpkt_handler(&tp->t_ccv); 538 539 if (tp->t_ccv.flags & CCF_ACKNOW) { 540 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 541 tp->t_flags |= TF_ACKNOW; 542 } 543 } 544 } 545 546 void inline 547 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 548 { 549 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos); 550 } 551 552 /* 553 * TCP input handling is split into multiple parts: 554 * tcp6_input is a thin wrapper around tcp_input for the extended 555 * ip6_protox[] call format in ip6_input 556 * tcp_input handles primary segment validation, inpcb lookup and 557 * SYN processing on listen sockets 558 * tcp_do_segment processes the ACK and text of the segment for 559 * establishing, established and closing connections 560 */ 561 #ifdef INET6 562 int 563 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 564 { 565 struct mbuf *m; 566 struct in6_ifaddr *ia6; 567 struct ip6_hdr *ip6; 568 569 m = *mp; 570 if (m->m_len < *offp + sizeof(struct tcphdr)) { 571 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 572 if (m == NULL) { 573 *mp = m; 574 TCPSTAT_INC(tcps_rcvshort); 575 return (IPPROTO_DONE); 576 } 577 } 578 579 /* 580 * draft-itojun-ipv6-tcp-to-anycast 581 * better place to put this in? 582 */ 583 ip6 = mtod(m, struct ip6_hdr *); 584 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 585 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 586 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 587 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 588 *mp = NULL; 589 return (IPPROTO_DONE); 590 } 591 592 *mp = m; 593 return (tcp_input_with_port(mp, offp, proto, port)); 594 } 595 596 int 597 tcp6_input(struct mbuf **mp, int *offp, int proto) 598 { 599 600 return(tcp6_input_with_port(mp, offp, proto, 0)); 601 } 602 #endif /* INET6 */ 603 604 int 605 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 606 { 607 struct mbuf *m = *mp; 608 struct tcphdr *th = NULL; 609 struct ip *ip = NULL; 610 struct inpcb *inp = NULL; 611 struct tcpcb *tp = NULL; 612 struct socket *so = NULL; 613 u_char *optp = NULL; 614 int off0; 615 int optlen = 0; 616 #ifdef INET 617 int len; 618 uint8_t ipttl; 619 #endif 620 int tlen = 0, off; 621 int drop_hdrlen; 622 int thflags; 623 int rstreason = 0; /* For badport_bandlim accounting purposes */ 624 int lookupflag; 625 uint8_t iptos; 626 struct m_tag *fwd_tag = NULL; 627 #ifdef INET6 628 struct ip6_hdr *ip6 = NULL; 629 int isipv6; 630 #else 631 const void *ip6 = NULL; 632 #endif /* INET6 */ 633 struct tcpopt to; /* options in this segment */ 634 char *s = NULL; /* address and port logging */ 635 636 NET_EPOCH_ASSERT(); 637 638 #ifdef INET6 639 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 640 #endif 641 642 off0 = *offp; 643 m = *mp; 644 *mp = NULL; 645 to.to_flags = 0; 646 TCPSTAT_INC(tcps_rcvtotal); 647 648 m->m_pkthdr.tcp_tun_port = port; 649 #ifdef INET6 650 if (isipv6) { 651 ip6 = mtod(m, struct ip6_hdr *); 652 th = (struct tcphdr *)((caddr_t)ip6 + off0); 653 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 654 if (port) 655 goto skip6_csum; 656 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 657 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 658 th->th_sum = m->m_pkthdr.csum_data; 659 else 660 th->th_sum = in6_cksum_pseudo(ip6, tlen, 661 IPPROTO_TCP, m->m_pkthdr.csum_data); 662 th->th_sum ^= 0xffff; 663 } else 664 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 665 if (th->th_sum) { 666 TCPSTAT_INC(tcps_rcvbadsum); 667 goto drop; 668 } 669 skip6_csum: 670 /* 671 * Be proactive about unspecified IPv6 address in source. 672 * As we use all-zero to indicate unbounded/unconnected pcb, 673 * unspecified IPv6 address can be used to confuse us. 674 * 675 * Note that packets with unspecified IPv6 destination is 676 * already dropped in ip6_input. 677 */ 678 KASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_dst), 679 ("%s: unspecified destination v6 address", __func__)); 680 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 681 IP6STAT_INC(ip6s_badscope); /* XXX */ 682 goto drop; 683 } 684 iptos = IPV6_TRAFFIC_CLASS(ip6); 685 } 686 #endif 687 #if defined(INET) && defined(INET6) 688 else 689 #endif 690 #ifdef INET 691 { 692 /* 693 * Get IP and TCP header together in first mbuf. 694 * Note: IP leaves IP header in first mbuf. 695 */ 696 if (off0 > sizeof (struct ip)) { 697 ip_stripoptions(m); 698 off0 = sizeof(struct ip); 699 } 700 if (m->m_len < sizeof (struct tcpiphdr)) { 701 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 702 == NULL) { 703 TCPSTAT_INC(tcps_rcvshort); 704 return (IPPROTO_DONE); 705 } 706 } 707 ip = mtod(m, struct ip *); 708 th = (struct tcphdr *)((caddr_t)ip + off0); 709 tlen = ntohs(ip->ip_len) - off0; 710 711 iptos = ip->ip_tos; 712 if (port) 713 goto skip_csum; 714 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 715 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 716 th->th_sum = m->m_pkthdr.csum_data; 717 else 718 th->th_sum = in_pseudo(ip->ip_src.s_addr, 719 ip->ip_dst.s_addr, 720 htonl(m->m_pkthdr.csum_data + tlen + 721 IPPROTO_TCP)); 722 th->th_sum ^= 0xffff; 723 } else { 724 struct ipovly *ipov = (struct ipovly *)ip; 725 726 /* 727 * Checksum extended TCP header and data. 728 */ 729 len = off0 + tlen; 730 ipttl = ip->ip_ttl; 731 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 732 ipov->ih_len = htons(tlen); 733 th->th_sum = in_cksum(m, len); 734 /* Reset length for SDT probes. */ 735 ip->ip_len = htons(len); 736 /* Reset TOS bits */ 737 ip->ip_tos = iptos; 738 /* Re-initialization for later version check */ 739 ip->ip_ttl = ipttl; 740 ip->ip_v = IPVERSION; 741 ip->ip_hl = off0 >> 2; 742 } 743 skip_csum: 744 if (th->th_sum && (port == 0)) { 745 TCPSTAT_INC(tcps_rcvbadsum); 746 goto drop; 747 } 748 KASSERT(ip->ip_dst.s_addr != INADDR_ANY, 749 ("%s: unspecified destination v4 address", __func__)); 750 if (__predict_false(ip->ip_src.s_addr == INADDR_ANY)) { 751 IPSTAT_INC(ips_badaddr); 752 goto drop; 753 } 754 } 755 #endif /* INET */ 756 757 /* 758 * Check that TCP offset makes sense, 759 * pull out TCP options and adjust length. XXX 760 */ 761 off = th->th_off << 2; 762 if (off < sizeof (struct tcphdr) || off > tlen) { 763 TCPSTAT_INC(tcps_rcvbadoff); 764 goto drop; 765 } 766 tlen -= off; /* tlen is used instead of ti->ti_len */ 767 if (off > sizeof (struct tcphdr)) { 768 #ifdef INET6 769 if (isipv6) { 770 if (m->m_len < off0 + off) { 771 m = m_pullup(m, off0 + off); 772 if (m == NULL) { 773 TCPSTAT_INC(tcps_rcvshort); 774 return (IPPROTO_DONE); 775 } 776 } 777 ip6 = mtod(m, struct ip6_hdr *); 778 th = (struct tcphdr *)((caddr_t)ip6 + off0); 779 } 780 #endif 781 #if defined(INET) && defined(INET6) 782 else 783 #endif 784 #ifdef INET 785 { 786 if (m->m_len < sizeof(struct ip) + off) { 787 if ((m = m_pullup(m, sizeof (struct ip) + off)) 788 == NULL) { 789 TCPSTAT_INC(tcps_rcvshort); 790 return (IPPROTO_DONE); 791 } 792 ip = mtod(m, struct ip *); 793 th = (struct tcphdr *)((caddr_t)ip + off0); 794 } 795 } 796 #endif 797 optlen = off - sizeof (struct tcphdr); 798 optp = (u_char *)(th + 1); 799 } 800 thflags = tcp_get_flags(th); 801 802 /* 803 * Convert TCP protocol specific fields to host format. 804 */ 805 tcp_fields_to_host(th); 806 807 /* 808 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 809 */ 810 drop_hdrlen = off0 + off; 811 812 /* 813 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 814 */ 815 if ( 816 #ifdef INET6 817 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 818 #ifdef INET 819 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 820 #endif 821 #endif 822 #if defined(INET) && !defined(INET6) 823 (m->m_flags & M_IP_NEXTHOP) 824 #endif 825 ) 826 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 827 828 /* 829 * For initial SYN packets we don't need write lock on matching 830 * PCB, be it a listening one or a synchronized one. The packet 831 * shall not modify its state. 832 */ 833 lookupflag = INPLOOKUP_WILDCARD | 834 ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ? 835 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB); 836 findpcb: 837 tp = NULL; 838 #ifdef INET6 839 if (isipv6 && fwd_tag != NULL) { 840 struct sockaddr_in6 *next_hop6; 841 842 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 843 /* 844 * Transparently forwarded. Pretend to be the destination. 845 * Already got one like this? 846 */ 847 inp = in6_pcblookup_mbuf(&V_tcbinfo, 848 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 849 lookupflag & ~INPLOOKUP_WILDCARD, m->m_pkthdr.rcvif, m); 850 if (!inp) { 851 /* 852 * It's new. Try to find the ambushing socket. 853 * Because we've rewritten the destination address, 854 * any hardware-generated hash is ignored. 855 */ 856 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 857 th->th_sport, &next_hop6->sin6_addr, 858 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 859 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 860 } 861 } else if (isipv6) { 862 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 863 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag, 864 m->m_pkthdr.rcvif, m); 865 } 866 #endif /* INET6 */ 867 #if defined(INET6) && defined(INET) 868 else 869 #endif 870 #ifdef INET 871 if (fwd_tag != NULL) { 872 struct sockaddr_in *next_hop; 873 874 next_hop = (struct sockaddr_in *)(fwd_tag+1); 875 /* 876 * Transparently forwarded. Pretend to be the destination. 877 * already got one like this? 878 */ 879 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 880 ip->ip_dst, th->th_dport, lookupflag & ~INPLOOKUP_WILDCARD, 881 m->m_pkthdr.rcvif, m); 882 if (!inp) { 883 /* 884 * It's new. Try to find the ambushing socket. 885 * Because we've rewritten the destination address, 886 * any hardware-generated hash is ignored. 887 */ 888 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 889 th->th_sport, next_hop->sin_addr, 890 next_hop->sin_port ? ntohs(next_hop->sin_port) : 891 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 892 } 893 } else 894 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 895 th->th_sport, ip->ip_dst, th->th_dport, lookupflag, 896 m->m_pkthdr.rcvif, m); 897 #endif /* INET */ 898 899 /* 900 * If the INPCB does not exist then all data in the incoming 901 * segment is discarded and an appropriate RST is sent back. 902 * XXX MRT Send RST using which routing table? 903 */ 904 if (inp == NULL) { 905 if (rstreason != 0) { 906 /* We came here after second (safety) lookup. */ 907 MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0); 908 goto dropwithreset; 909 } 910 /* 911 * Log communication attempts to ports that are not 912 * in use. 913 */ 914 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 915 V_tcp_log_in_vain == 2) { 916 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 917 log(LOG_INFO, "%s; %s: Connection attempt " 918 "to closed port\n", s, __func__); 919 } 920 rstreason = BANDLIM_RST_CLOSEDPORT; 921 goto dropwithreset; 922 } 923 INP_LOCK_ASSERT(inp); 924 925 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 926 !SOLISTENING(inp->inp_socket)) { 927 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) { 928 inp->inp_flowid = m->m_pkthdr.flowid; 929 inp->inp_flowtype = M_HASHTYPE_GET(m); 930 #ifdef RSS 931 } else { 932 /* assign flowid by software RSS hash */ 933 #ifdef INET6 934 if (isipv6) { 935 rss_proto_software_hash_v6(&inp->in6p_faddr, 936 &inp->in6p_laddr, 937 inp->inp_fport, 938 inp->inp_lport, 939 IPPROTO_TCP, 940 &inp->inp_flowid, 941 &inp->inp_flowtype); 942 } else 943 #endif /* INET6 */ 944 { 945 rss_proto_software_hash_v4(inp->inp_faddr, 946 inp->inp_laddr, 947 inp->inp_fport, 948 inp->inp_lport, 949 IPPROTO_TCP, 950 &inp->inp_flowid, 951 &inp->inp_flowtype); 952 } 953 #endif /* RSS */ 954 } 955 } 956 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 957 #ifdef INET6 958 if (isipv6 && IPSEC_ENABLED(ipv6) && 959 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 960 goto dropunlock; 961 } 962 #ifdef INET 963 else 964 #endif 965 #endif /* INET6 */ 966 #ifdef INET 967 if (IPSEC_ENABLED(ipv4) && 968 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 969 goto dropunlock; 970 } 971 #endif /* INET */ 972 #endif /* IPSEC */ 973 974 /* 975 * Check the minimum TTL for socket. 976 */ 977 if (inp->inp_ip_minttl != 0) { 978 #ifdef INET6 979 if (isipv6) { 980 if (inp->inp_ip_minttl > ip6->ip6_hlim) 981 goto dropunlock; 982 } else 983 #endif 984 if (inp->inp_ip_minttl > ip->ip_ttl) 985 goto dropunlock; 986 } 987 988 tp = intotcpcb(inp); 989 switch (tp->t_state) { 990 case TCPS_TIME_WAIT: 991 /* 992 * A previous connection in TIMEWAIT state is supposed to catch 993 * stray or duplicate segments arriving late. If this segment 994 * was a legitimate new connection attempt, the old INPCB gets 995 * removed and we can try again to find a listening socket. 996 */ 997 tcp_dooptions(&to, optp, optlen, 998 (thflags & TH_SYN) ? TO_SYN : 0); 999 /* 1000 * tcp_twcheck unlocks the inp always, and frees the m if fails. 1001 */ 1002 if (tcp_twcheck(inp, &to, th, m, tlen)) 1003 goto findpcb; 1004 return (IPPROTO_DONE); 1005 case TCPS_CLOSED: 1006 /* 1007 * The TCPCB may no longer exist if the connection is winding 1008 * down or it is in the CLOSED state. Either way we drop the 1009 * segment and send an appropriate response. 1010 */ 1011 rstreason = BANDLIM_RST_CLOSEDPORT; 1012 goto dropwithreset; 1013 } 1014 1015 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 1016 rstreason = BANDLIM_RST_CLOSEDPORT; 1017 goto dropwithreset; 1018 } 1019 1020 #ifdef TCP_OFFLOAD 1021 if (tp->t_flags & TF_TOE) { 1022 tcp_offload_input(tp, m); 1023 m = NULL; /* consumed by the TOE driver */ 1024 goto dropunlock; 1025 } 1026 #endif 1027 1028 #ifdef MAC 1029 if (mac_inpcb_check_deliver(inp, m)) 1030 goto dropunlock; 1031 #endif 1032 so = inp->inp_socket; 1033 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1034 /* 1035 * When the socket is accepting connections (the INPCB is in LISTEN 1036 * state) we look into the SYN cache if this is a new connection 1037 * attempt or the completion of a previous one. 1038 */ 1039 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), 1040 ("%s: so accepting but tp %p not listening", __func__, tp)); 1041 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { 1042 struct in_conninfo inc; 1043 1044 bzero(&inc, sizeof(inc)); 1045 #ifdef INET6 1046 if (isipv6) { 1047 inc.inc_flags |= INC_ISIPV6; 1048 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1049 inc.inc_flags |= INC_IPV6MINMTU; 1050 inc.inc6_faddr = ip6->ip6_src; 1051 inc.inc6_laddr = ip6->ip6_dst; 1052 } else 1053 #endif 1054 { 1055 inc.inc_faddr = ip->ip_src; 1056 inc.inc_laddr = ip->ip_dst; 1057 } 1058 inc.inc_fport = th->th_sport; 1059 inc.inc_lport = th->th_dport; 1060 inc.inc_fibnum = so->so_fibnum; 1061 1062 /* 1063 * Check for an existing connection attempt in syncache if 1064 * the flag is only ACK. A successful lookup creates a new 1065 * socket appended to the listen queue in SYN_RECEIVED state. 1066 */ 1067 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1068 /* 1069 * Parse the TCP options here because 1070 * syncookies need access to the reflected 1071 * timestamp. 1072 */ 1073 tcp_dooptions(&to, optp, optlen, 0); 1074 /* 1075 * NB: syncache_expand() doesn't unlock inp. 1076 */ 1077 rstreason = syncache_expand(&inc, &to, th, &so, m, port); 1078 if (rstreason < 0) { 1079 /* 1080 * A failing TCP MD5 signature comparison 1081 * must result in the segment being dropped 1082 * and must not produce any response back 1083 * to the sender. 1084 */ 1085 goto dropunlock; 1086 } else if (rstreason == 0) { 1087 /* 1088 * No syncache entry, or ACK was not for our 1089 * SYN/ACK. Do our protection against double 1090 * ACK. If peer sent us 2 ACKs, then for the 1091 * first one syncache_expand() successfully 1092 * converted syncache entry into a socket, 1093 * while we were waiting on the inpcb lock. We 1094 * don't want to sent RST for the second ACK, 1095 * so we perform second lookup without wildcard 1096 * match, hoping to find the new socket. If 1097 * the ACK is stray indeed, rstreason would 1098 * hint the above code that the lookup was a 1099 * second attempt. 1100 * 1101 * NB: syncache did its own logging 1102 * of the failure cause. 1103 */ 1104 INP_WUNLOCK(inp); 1105 rstreason = BANDLIM_RST_OPENPORT; 1106 lookupflag &= ~INPLOOKUP_WILDCARD; 1107 goto findpcb; 1108 } 1109 tfo_socket_result: 1110 if (so == NULL) { 1111 /* 1112 * We completed the 3-way handshake 1113 * but could not allocate a socket 1114 * either due to memory shortage, 1115 * listen queue length limits or 1116 * global socket limits. Send RST 1117 * or wait and have the remote end 1118 * retransmit the ACK for another 1119 * try. 1120 */ 1121 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1122 log(LOG_DEBUG, "%s; %s: Listen socket: " 1123 "Socket allocation failed due to " 1124 "limits or memory shortage, %s\n", 1125 s, __func__, 1126 V_tcp_sc_rst_sock_fail ? 1127 "sending RST" : "try again"); 1128 if (V_tcp_sc_rst_sock_fail) { 1129 rstreason = BANDLIM_UNLIMITED; 1130 goto dropwithreset; 1131 } else 1132 goto dropunlock; 1133 } 1134 /* 1135 * Socket is created in state SYN_RECEIVED. 1136 * Unlock the listen socket, lock the newly 1137 * created socket and update the tp variable. 1138 * If we came here via jump to tfo_socket_result, 1139 * then listening socket is read-locked. 1140 */ 1141 INP_UNLOCK(inp); /* listen socket */ 1142 inp = sotoinpcb(so); 1143 /* 1144 * New connection inpcb is already locked by 1145 * syncache_expand(). 1146 */ 1147 INP_WLOCK_ASSERT(inp); 1148 tp = intotcpcb(inp); 1149 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1150 ("%s: ", __func__)); 1151 /* 1152 * Process the segment and the data it 1153 * contains. tcp_do_segment() consumes 1154 * the mbuf chain and unlocks the inpcb. 1155 */ 1156 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1157 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, 1158 tlen, iptos); 1159 return (IPPROTO_DONE); 1160 } 1161 /* 1162 * Segment flag validation for new connection attempts: 1163 * 1164 * Our (SYN|ACK) response was rejected. 1165 * Check with syncache and remove entry to prevent 1166 * retransmits. 1167 * 1168 * NB: syncache_chkrst does its own logging of failure 1169 * causes. 1170 */ 1171 if (thflags & TH_RST) { 1172 syncache_chkrst(&inc, th, m, port); 1173 goto dropunlock; 1174 } 1175 /* 1176 * We can't do anything without SYN. 1177 */ 1178 if ((thflags & TH_SYN) == 0) { 1179 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1180 log(LOG_DEBUG, "%s; %s: Listen socket: " 1181 "SYN is missing, segment ignored\n", 1182 s, __func__); 1183 TCPSTAT_INC(tcps_badsyn); 1184 goto dropunlock; 1185 } 1186 /* 1187 * (SYN|ACK) is bogus on a listen socket. 1188 */ 1189 if (thflags & TH_ACK) { 1190 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1191 log(LOG_DEBUG, "%s; %s: Listen socket: " 1192 "SYN|ACK invalid, segment rejected\n", 1193 s, __func__); 1194 syncache_badack(&inc, port); /* XXX: Not needed! */ 1195 TCPSTAT_INC(tcps_badsyn); 1196 rstreason = BANDLIM_RST_OPENPORT; 1197 goto dropwithreset; 1198 } 1199 /* 1200 * If the drop_synfin option is enabled, drop all 1201 * segments with both the SYN and FIN bits set. 1202 * This prevents e.g. nmap from identifying the 1203 * TCP/IP stack. 1204 * XXX: Poor reasoning. nmap has other methods 1205 * and is constantly refining its stack detection 1206 * strategies. 1207 * XXX: This is a violation of the TCP specification 1208 * and was used by RFC1644. 1209 */ 1210 if ((thflags & TH_FIN) && V_drop_synfin) { 1211 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1212 log(LOG_DEBUG, "%s; %s: Listen socket: " 1213 "SYN|FIN segment ignored (based on " 1214 "sysctl setting)\n", s, __func__); 1215 TCPSTAT_INC(tcps_badsyn); 1216 goto dropunlock; 1217 } 1218 /* 1219 * Segment's flags are (SYN) or (SYN|FIN). 1220 * 1221 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1222 * as they do not affect the state of the TCP FSM. 1223 * The data pointed to by TH_URG and th_urp is ignored. 1224 */ 1225 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1226 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1227 KASSERT(thflags & (TH_SYN), 1228 ("%s: Listen socket: TH_SYN not set", __func__)); 1229 INP_RLOCK_ASSERT(inp); 1230 #ifdef INET6 1231 /* 1232 * If deprecated address is forbidden, 1233 * we do not accept SYN to deprecated interface 1234 * address to prevent any new inbound connection from 1235 * getting established. 1236 * When we do not accept SYN, we send a TCP RST, 1237 * with deprecated source address (instead of dropping 1238 * it). We compromise it as it is much better for peer 1239 * to send a RST, and RST will be the final packet 1240 * for the exchange. 1241 * 1242 * If we do not forbid deprecated addresses, we accept 1243 * the SYN packet. RFC2462 does not suggest dropping 1244 * SYN in this case. 1245 * If we decipher RFC2462 5.5.4, it says like this: 1246 * 1. use of deprecated addr with existing 1247 * communication is okay - "SHOULD continue to be 1248 * used" 1249 * 2. use of it with new communication: 1250 * (2a) "SHOULD NOT be used if alternate address 1251 * with sufficient scope is available" 1252 * (2b) nothing mentioned otherwise. 1253 * Here we fall into (2b) case as we have no choice in 1254 * our source address selection - we must obey the peer. 1255 * 1256 * The wording in RFC2462 is confusing, and there are 1257 * multiple description text for deprecated address 1258 * handling - worse, they are not exactly the same. 1259 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1260 */ 1261 if (isipv6 && !V_ip6_use_deprecated) { 1262 struct in6_ifaddr *ia6; 1263 1264 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1265 if (ia6 != NULL && 1266 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1267 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1268 log(LOG_DEBUG, "%s; %s: Listen socket: " 1269 "Connection attempt to deprecated " 1270 "IPv6 address rejected\n", 1271 s, __func__); 1272 rstreason = BANDLIM_RST_OPENPORT; 1273 goto dropwithreset; 1274 } 1275 } 1276 #endif /* INET6 */ 1277 /* 1278 * Basic sanity checks on incoming SYN requests: 1279 * Don't respond if the destination is a link layer 1280 * broadcast according to RFC1122 4.2.3.10, p. 104. 1281 * If it is from this socket it must be forged. 1282 * Don't respond if the source or destination is a 1283 * global or subnet broad- or multicast address. 1284 * Note that it is quite possible to receive unicast 1285 * link-layer packets with a broadcast IP address. Use 1286 * in_broadcast() to find them. 1287 */ 1288 if (m->m_flags & (M_BCAST|M_MCAST)) { 1289 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1290 log(LOG_DEBUG, "%s; %s: Listen socket: " 1291 "Connection attempt from broad- or multicast " 1292 "link layer address ignored\n", s, __func__); 1293 goto dropunlock; 1294 } 1295 #ifdef INET6 1296 if (isipv6) { 1297 if (th->th_dport == th->th_sport && 1298 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1299 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1300 log(LOG_DEBUG, "%s; %s: Listen socket: " 1301 "Connection attempt to/from self " 1302 "ignored\n", s, __func__); 1303 goto dropunlock; 1304 } 1305 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1306 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1307 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1308 log(LOG_DEBUG, "%s; %s: Listen socket: " 1309 "Connection attempt from/to multicast " 1310 "address ignored\n", s, __func__); 1311 goto dropunlock; 1312 } 1313 } 1314 #endif 1315 #if defined(INET) && defined(INET6) 1316 else 1317 #endif 1318 #ifdef INET 1319 { 1320 if (th->th_dport == th->th_sport && 1321 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1322 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1323 log(LOG_DEBUG, "%s; %s: Listen socket: " 1324 "Connection attempt from/to self " 1325 "ignored\n", s, __func__); 1326 goto dropunlock; 1327 } 1328 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1329 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1330 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1331 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1332 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1333 log(LOG_DEBUG, "%s; %s: Listen socket: " 1334 "Connection attempt from/to broad- " 1335 "or multicast address ignored\n", 1336 s, __func__); 1337 goto dropunlock; 1338 } 1339 } 1340 #endif 1341 /* 1342 * SYN appears to be valid. Create compressed TCP state 1343 * for syncache. 1344 */ 1345 TCP_PROBE3(debug__input, tp, th, m); 1346 tcp_dooptions(&to, optp, optlen, TO_SYN); 1347 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, 1348 iptos, port)) != NULL) 1349 goto tfo_socket_result; 1350 1351 /* 1352 * Entry added to syncache and mbuf consumed. 1353 * Only the listen socket is unlocked by syncache_add(). 1354 */ 1355 return (IPPROTO_DONE); 1356 } else if (tp->t_state == TCPS_LISTEN) { 1357 /* 1358 * When a listen socket is torn down the SO_ACCEPTCONN 1359 * flag is removed first while connections are drained 1360 * from the accept queue in a unlock/lock cycle of the 1361 * ACCEPT_LOCK, opening a race condition allowing a SYN 1362 * attempt go through unhandled. 1363 */ 1364 goto dropunlock; 1365 } 1366 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1367 if (tp->t_flags & TF_SIGNATURE) { 1368 tcp_dooptions(&to, optp, optlen, thflags); 1369 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1370 TCPSTAT_INC(tcps_sig_err_nosigopt); 1371 goto dropunlock; 1372 } 1373 if (!TCPMD5_ENABLED() || 1374 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1375 goto dropunlock; 1376 } 1377 #endif 1378 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1379 1380 /* 1381 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1382 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1383 * the inpcb, and unlocks pcbinfo. 1384 * 1385 * XXXGL: in case of a pure SYN arriving on existing connection 1386 * TCP stacks won't need to modify the PCB, they would either drop 1387 * the segment silently, or send a challenge ACK. However, we try 1388 * to upgrade the lock, because calling convention for stacks is 1389 * write-lock on PCB. If upgrade fails, drop the SYN. 1390 */ 1391 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0) 1392 goto dropunlock; 1393 1394 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, tlen, iptos); 1395 return (IPPROTO_DONE); 1396 1397 dropwithreset: 1398 /* 1399 * When blackholing do not respond with a RST but 1400 * completely ignore the segment and drop it. 1401 */ 1402 if (((rstreason == BANDLIM_RST_OPENPORT && V_blackhole == 3) || 1403 (rstreason == BANDLIM_RST_CLOSEDPORT && 1404 ((V_blackhole == 1 && (thflags & TH_SYN)) || V_blackhole > 1))) && 1405 (V_blackhole_local || ( 1406 #ifdef INET6 1407 isipv6 ? !in6_localaddr(&ip6->ip6_src) : 1408 #endif 1409 #ifdef INET 1410 !in_localip(ip->ip_src) 1411 #else 1412 true 1413 #endif 1414 ))) 1415 goto dropunlock; 1416 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1417 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1418 m = NULL; /* mbuf chain got consumed. */ 1419 1420 dropunlock: 1421 if (m != NULL) 1422 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1423 1424 if (inp != NULL) 1425 INP_UNLOCK(inp); 1426 1427 drop: 1428 if (s != NULL) 1429 free(s, M_TCPLOG); 1430 if (m != NULL) 1431 m_freem(m); 1432 return (IPPROTO_DONE); 1433 } 1434 1435 /* 1436 * Automatic sizing of receive socket buffer. Often the send 1437 * buffer size is not optimally adjusted to the actual network 1438 * conditions at hand (delay bandwidth product). Setting the 1439 * buffer size too small limits throughput on links with high 1440 * bandwidth and high delay (eg. trans-continental/oceanic links). 1441 * 1442 * On the receive side the socket buffer memory is only rarely 1443 * used to any significant extent. This allows us to be much 1444 * more aggressive in scaling the receive socket buffer. For 1445 * the case that the buffer space is actually used to a large 1446 * extent and we run out of kernel memory we can simply drop 1447 * the new segments; TCP on the sender will just retransmit it 1448 * later. Setting the buffer size too big may only consume too 1449 * much kernel memory if the application doesn't read() from 1450 * the socket or packet loss or reordering makes use of the 1451 * reassembly queue. 1452 * 1453 * The criteria to step up the receive buffer one notch are: 1454 * 1. Application has not set receive buffer size with 1455 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1456 * 2. the number of bytes received during 1/2 of an sRTT 1457 * is at least 3/8 of the current socket buffer size. 1458 * 3. receive buffer size has not hit maximal automatic size; 1459 * 1460 * If all of the criteria are met we increaset the socket buffer 1461 * by a 1/2 (bounded by the max). This allows us to keep ahead 1462 * of slow-start but also makes it so our peer never gets limited 1463 * by our rwnd which we then open up causing a burst. 1464 * 1465 * This algorithm does two steps per RTT at most and only if 1466 * we receive a bulk stream w/o packet losses or reorderings. 1467 * Shrinking the buffer during idle times is not necessary as 1468 * it doesn't consume any memory when idle. 1469 * 1470 * TODO: Only step up if the application is actually serving 1471 * the buffer to better manage the socket buffer resources. 1472 */ 1473 int 1474 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1475 struct tcpcb *tp, int tlen) 1476 { 1477 int newsize = 0; 1478 1479 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1480 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1481 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1482 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1483 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1484 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1485 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1486 } 1487 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1488 1489 /* Start over with next RTT. */ 1490 tp->rfbuf_ts = 0; 1491 tp->rfbuf_cnt = 0; 1492 } else { 1493 tp->rfbuf_cnt += tlen; /* add up */ 1494 } 1495 return (newsize); 1496 } 1497 1498 int 1499 tcp_input(struct mbuf **mp, int *offp, int proto) 1500 { 1501 return(tcp_input_with_port(mp, offp, proto, 0)); 1502 } 1503 1504 static void 1505 tcp_handle_wakeup(struct tcpcb *tp) 1506 { 1507 1508 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1509 1510 if (tp->t_flags & TF_WAKESOR) { 1511 struct socket *so = tptosocket(tp); 1512 1513 tp->t_flags &= ~TF_WAKESOR; 1514 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1515 sorwakeup_locked(so); 1516 } 1517 } 1518 1519 void 1520 tcp_do_segment(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th, 1521 int drop_hdrlen, int tlen, uint8_t iptos) 1522 { 1523 uint16_t thflags; 1524 int acked, ourfinisacked, needoutput = 0; 1525 sackstatus_t sack_changed; 1526 int rstreason, todrop, win, incforsyn = 0; 1527 uint32_t tiwin; 1528 uint16_t nsegs; 1529 char *s; 1530 struct inpcb *inp = tptoinpcb(tp); 1531 struct socket *so = tptosocket(tp); 1532 struct in_conninfo *inc = &inp->inp_inc; 1533 struct mbuf *mfree; 1534 struct tcpopt to; 1535 int tfo_syn; 1536 u_int maxseg; 1537 1538 thflags = tcp_get_flags(th); 1539 tp->sackhint.last_sack_ack = 0; 1540 sack_changed = SACK_NOCHANGE; 1541 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1542 1543 NET_EPOCH_ASSERT(); 1544 INP_WLOCK_ASSERT(inp); 1545 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1546 __func__)); 1547 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1548 __func__)); 1549 1550 #ifdef TCPPCAP 1551 /* Save segment, if requested. */ 1552 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1553 #endif 1554 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1555 tlen, NULL, true); 1556 1557 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1558 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1559 log(LOG_DEBUG, "%s; %s: " 1560 "SYN|FIN segment ignored (based on " 1561 "sysctl setting)\n", s, __func__); 1562 free(s, M_TCPLOG); 1563 } 1564 goto drop; 1565 } 1566 1567 /* 1568 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1569 * check SEQ.ACK first. 1570 */ 1571 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1572 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1573 rstreason = BANDLIM_UNLIMITED; 1574 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1575 goto dropwithreset; 1576 } 1577 1578 /* 1579 * Segment received on connection. 1580 * Reset idle time and keep-alive timer. 1581 * XXX: This should be done after segment 1582 * validation to ignore broken/spoofed segs. 1583 */ 1584 if (tp->t_idle_reduce && 1585 (tp->snd_max == tp->snd_una) && 1586 ((ticks - tp->t_rcvtime) >= tp->t_rxtcur)) 1587 cc_after_idle(tp); 1588 tp->t_rcvtime = ticks; 1589 1590 if (thflags & TH_FIN) 1591 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN); 1592 /* 1593 * Scale up the window into a 32-bit value. 1594 * For the SYN_SENT state the scale is zero. 1595 */ 1596 tiwin = th->th_win << tp->snd_scale; 1597 #ifdef STATS 1598 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1599 #endif 1600 1601 /* 1602 * TCP ECN processing. 1603 */ 1604 if (tcp_ecn_input_segment(tp, thflags, tlen, 1605 tcp_packets_this_ack(tp, th->th_ack), 1606 iptos)) 1607 cc_cong_signal(tp, th, CC_ECN); 1608 1609 /* 1610 * Parse options on any incoming segment. 1611 */ 1612 tcp_dooptions(&to, (u_char *)(th + 1), 1613 (th->th_off << 2) - sizeof(struct tcphdr), 1614 (thflags & TH_SYN) ? TO_SYN : 0); 1615 1616 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1617 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1618 (to.to_flags & TOF_SIGNATURE) == 0) { 1619 TCPSTAT_INC(tcps_sig_err_sigopt); 1620 /* XXX: should drop? */ 1621 } 1622 #endif 1623 /* 1624 * If echoed timestamp is later than the current time, 1625 * fall back to non RFC1323 RTT calculation. Normalize 1626 * timestamp if syncookies were used when this connection 1627 * was established. 1628 */ 1629 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1630 to.to_tsecr -= tp->ts_offset; 1631 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1632 to.to_tsecr = 0; 1633 else if (tp->t_rxtshift == 1 && 1634 tp->t_flags & TF_PREVVALID && 1635 tp->t_badrxtwin != 0 && 1636 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 1637 cc_cong_signal(tp, th, CC_RTO_ERR); 1638 } 1639 /* 1640 * Process options only when we get SYN/ACK back. The SYN case 1641 * for incoming connections is handled in tcp_syncache. 1642 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1643 * or <SYN,ACK>) segment itself is never scaled. 1644 * XXX this is traditional behavior, may need to be cleaned up. 1645 */ 1646 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1647 /* Handle parallel SYN for ECN */ 1648 tcp_ecn_input_parallel_syn(tp, thflags, iptos); 1649 if ((to.to_flags & TOF_SCALE) && 1650 (tp->t_flags & TF_REQ_SCALE) && 1651 !(tp->t_flags & TF_NOOPT)) { 1652 tp->t_flags |= TF_RCVD_SCALE; 1653 tp->snd_scale = to.to_wscale; 1654 } else 1655 tp->t_flags &= ~TF_REQ_SCALE; 1656 /* 1657 * Initial send window. It will be updated with 1658 * the next incoming segment to the scaled value. 1659 */ 1660 tp->snd_wnd = th->th_win; 1661 if ((to.to_flags & TOF_TS) && 1662 (tp->t_flags & TF_REQ_TSTMP) && 1663 !(tp->t_flags & TF_NOOPT)) { 1664 tp->t_flags |= TF_RCVD_TSTMP; 1665 tp->ts_recent = to.to_tsval; 1666 tp->ts_recent_age = tcp_ts_getticks(); 1667 } else 1668 tp->t_flags &= ~TF_REQ_TSTMP; 1669 if (to.to_flags & TOF_MSS) 1670 tcp_mss(tp, to.to_mss); 1671 if ((tp->t_flags & TF_SACK_PERMIT) && 1672 (!(to.to_flags & TOF_SACKPERM) || 1673 (tp->t_flags & TF_NOOPT))) 1674 tp->t_flags &= ~TF_SACK_PERMIT; 1675 if (IS_FASTOPEN(tp->t_flags)) { 1676 if ((to.to_flags & TOF_FASTOPEN) && 1677 !(tp->t_flags & TF_NOOPT)) { 1678 uint16_t mss; 1679 1680 if (to.to_flags & TOF_MSS) 1681 mss = to.to_mss; 1682 else 1683 if ((inp->inp_vflag & INP_IPV6) != 0) 1684 mss = TCP6_MSS; 1685 else 1686 mss = TCP_MSS; 1687 tcp_fastopen_update_cache(tp, mss, 1688 to.to_tfo_len, to.to_tfo_cookie); 1689 } else 1690 tcp_fastopen_disable_path(tp); 1691 } 1692 } 1693 1694 /* 1695 * If timestamps were negotiated during SYN/ACK and a 1696 * segment without a timestamp is received, silently drop 1697 * the segment, unless it is a RST segment or missing timestamps are 1698 * tolerated. 1699 * See section 3.2 of RFC 7323. 1700 */ 1701 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1702 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1703 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1704 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1705 "segment processed normally\n", 1706 s, __func__); 1707 free(s, M_TCPLOG); 1708 } 1709 } else { 1710 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1711 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1712 "segment silently dropped\n", s, __func__); 1713 free(s, M_TCPLOG); 1714 } 1715 goto drop; 1716 } 1717 } 1718 /* 1719 * If timestamps were not negotiated during SYN/ACK and a 1720 * segment with a timestamp is received, ignore the 1721 * timestamp and process the packet normally. 1722 * See section 3.2 of RFC 7323. 1723 */ 1724 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1725 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1726 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1727 "segment processed normally\n", s, __func__); 1728 free(s, M_TCPLOG); 1729 } 1730 } 1731 1732 /* 1733 * Header prediction: check for the two common cases 1734 * of a uni-directional data xfer. If the packet has 1735 * no control flags, is in-sequence, the window didn't 1736 * change and we're not retransmitting, it's a 1737 * candidate. If the length is zero and the ack moved 1738 * forward, we're the sender side of the xfer. Just 1739 * free the data acked & wake any higher level process 1740 * that was blocked waiting for space. If the length 1741 * is non-zero and the ack didn't move, we're the 1742 * receiver side. If we're getting packets in-order 1743 * (the reassembly queue is empty), add the data to 1744 * the socket buffer and note that we need a delayed ack. 1745 * Make sure that the hidden state-flags are also off. 1746 * Since we check for TCPS_ESTABLISHED first, it can only 1747 * be TH_NEEDSYN. 1748 */ 1749 if (tp->t_state == TCPS_ESTABLISHED && 1750 th->th_seq == tp->rcv_nxt && 1751 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1752 tp->snd_nxt == tp->snd_max && 1753 tiwin && tiwin == tp->snd_wnd && 1754 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1755 SEGQ_EMPTY(tp) && 1756 ((to.to_flags & TOF_TS) == 0 || 1757 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1758 /* 1759 * If last ACK falls within this segment's sequence numbers, 1760 * record the timestamp. 1761 * NOTE that the test is modified according to the latest 1762 * proposal of the [email protected] list (Braden 1993/04/26). 1763 */ 1764 if ((to.to_flags & TOF_TS) != 0 && 1765 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1766 tp->ts_recent_age = tcp_ts_getticks(); 1767 tp->ts_recent = to.to_tsval; 1768 } 1769 1770 if (tlen == 0) { 1771 if (SEQ_GT(th->th_ack, tp->snd_una) && 1772 SEQ_LEQ(th->th_ack, tp->snd_max) && 1773 !IN_RECOVERY(tp->t_flags) && 1774 (to.to_flags & TOF_SACK) == 0 && 1775 TAILQ_EMPTY(&tp->snd_holes)) { 1776 /* 1777 * This is a pure ack for outstanding data. 1778 */ 1779 TCPSTAT_INC(tcps_predack); 1780 1781 /* 1782 * "bad retransmit" recovery without timestamps. 1783 */ 1784 if ((to.to_flags & TOF_TS) == 0 && 1785 tp->t_rxtshift == 1 && 1786 tp->t_flags & TF_PREVVALID && 1787 tp->t_badrxtwin != 0 && 1788 TSTMP_LT(ticks, tp->t_badrxtwin)) { 1789 cc_cong_signal(tp, th, CC_RTO_ERR); 1790 } 1791 1792 /* 1793 * Recalculate the transmit timer / rtt. 1794 * 1795 * Some boxes send broken timestamp replies 1796 * during the SYN+ACK phase, ignore 1797 * timestamps of 0 or we could calculate a 1798 * huge RTT and blow up the retransmit timer. 1799 */ 1800 if ((to.to_flags & TOF_TS) != 0 && 1801 to.to_tsecr) { 1802 uint32_t t; 1803 1804 t = tcp_ts_getticks() - to.to_tsecr; 1805 if (!tp->t_rttlow || tp->t_rttlow > t) 1806 tp->t_rttlow = t; 1807 tcp_xmit_timer(tp, 1808 TCP_TS_TO_TICKS(t) + 1); 1809 } else if (tp->t_rtttime && 1810 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1811 if (!tp->t_rttlow || 1812 tp->t_rttlow > ticks - tp->t_rtttime) 1813 tp->t_rttlow = ticks - tp->t_rtttime; 1814 tcp_xmit_timer(tp, 1815 ticks - tp->t_rtttime); 1816 } 1817 acked = BYTES_THIS_ACK(tp, th); 1818 1819 #ifdef TCP_HHOOK 1820 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1821 hhook_run_tcp_est_in(tp, th, &to); 1822 #endif 1823 1824 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1825 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1826 sbdrop(&so->so_snd, acked); 1827 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1828 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1829 tp->snd_recover = th->th_ack - 1; 1830 1831 /* 1832 * Let the congestion control algorithm update 1833 * congestion control related information. This 1834 * typically means increasing the congestion 1835 * window. 1836 */ 1837 cc_ack_received(tp, th, nsegs, CC_ACK); 1838 1839 tp->snd_una = th->th_ack; 1840 /* 1841 * Pull snd_wl2 up to prevent seq wrap relative 1842 * to th_ack. 1843 */ 1844 tp->snd_wl2 = th->th_ack; 1845 tp->t_dupacks = 0; 1846 m_freem(m); 1847 1848 /* 1849 * If all outstanding data are acked, stop 1850 * retransmit timer, otherwise restart timer 1851 * using current (possibly backed-off) value. 1852 * If process is waiting for space, 1853 * wakeup/selwakeup/signal. If data 1854 * are ready to send, let tcp_output 1855 * decide between more output or persist. 1856 */ 1857 TCP_PROBE3(debug__input, tp, th, m); 1858 /* 1859 * Clear t_acktime if remote side has ACKd 1860 * all data in the socket buffer. 1861 * Otherwise, update t_acktime if we received 1862 * a sufficiently large ACK. 1863 */ 1864 if (sbavail(&so->so_snd) == 0) 1865 tp->t_acktime = 0; 1866 else if (acked > 1) 1867 tp->t_acktime = ticks; 1868 if (tp->snd_una == tp->snd_max) 1869 tcp_timer_activate(tp, TT_REXMT, 0); 1870 else if (!tcp_timer_active(tp, TT_PERSIST)) 1871 tcp_timer_activate(tp, TT_REXMT, 1872 TP_RXTCUR(tp)); 1873 sowwakeup(so); 1874 /* 1875 * Only call tcp_output when there 1876 * is new data available to be sent 1877 * or we need to send an ACK. 1878 */ 1879 if (SEQ_GT(tp->snd_una + sbavail(&so->so_snd), 1880 tp->snd_max) || tp->t_flags & TF_ACKNOW) 1881 (void) tcp_output(tp); 1882 goto check_delack; 1883 } 1884 } else if (th->th_ack == tp->snd_una && 1885 tlen <= sbspace(&so->so_rcv)) { 1886 int newsize = 0; /* automatic sockbuf scaling */ 1887 1888 /* 1889 * This is a pure, in-sequence data packet with 1890 * nothing on the reassembly queue and we have enough 1891 * buffer space to take it. 1892 */ 1893 /* Clean receiver SACK report if present */ 1894 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1895 tcp_clean_sackreport(tp); 1896 TCPSTAT_INC(tcps_preddat); 1897 tp->rcv_nxt += tlen; 1898 if (tlen && 1899 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1900 (tp->t_fbyte_in == 0)) { 1901 tp->t_fbyte_in = ticks; 1902 if (tp->t_fbyte_in == 0) 1903 tp->t_fbyte_in = 1; 1904 if (tp->t_fbyte_out && tp->t_fbyte_in) 1905 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1906 } 1907 /* 1908 * Pull snd_wl1 up to prevent seq wrap relative to 1909 * th_seq. 1910 */ 1911 tp->snd_wl1 = th->th_seq; 1912 /* 1913 * Pull rcv_up up to prevent seq wrap relative to 1914 * rcv_nxt. 1915 */ 1916 tp->rcv_up = tp->rcv_nxt; 1917 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1918 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1919 TCP_PROBE3(debug__input, tp, th, m); 1920 1921 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1922 1923 /* Add data to socket buffer. */ 1924 SOCKBUF_LOCK(&so->so_rcv); 1925 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1926 m_freem(m); 1927 } else { 1928 /* 1929 * Set new socket buffer size. 1930 * Give up when limit is reached. 1931 */ 1932 if (newsize) 1933 if (!sbreserve_locked(so, SO_RCV, 1934 newsize, NULL)) 1935 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1936 m_adj(m, drop_hdrlen); /* delayed header drop */ 1937 sbappendstream_locked(&so->so_rcv, m, 0); 1938 } 1939 /* NB: sorwakeup_locked() does an implicit unlock. */ 1940 sorwakeup_locked(so); 1941 if (DELAY_ACK(tp, tlen)) { 1942 tp->t_flags |= TF_DELACK; 1943 } else { 1944 tp->t_flags |= TF_ACKNOW; 1945 (void) tcp_output(tp); 1946 } 1947 goto check_delack; 1948 } 1949 } 1950 1951 /* 1952 * Calculate amount of space in receive window, 1953 * and then do TCP input processing. 1954 * Receive window is amount of space in rcv queue, 1955 * but not less than advertised window. 1956 */ 1957 win = sbspace(&so->so_rcv); 1958 if (win < 0) 1959 win = 0; 1960 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1961 1962 switch (tp->t_state) { 1963 /* 1964 * If the state is SYN_RECEIVED: 1965 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1966 */ 1967 case TCPS_SYN_RECEIVED: 1968 if (thflags & TH_RST) { 1969 /* Handle RST segments later. */ 1970 break; 1971 } 1972 if ((thflags & TH_ACK) && 1973 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1974 SEQ_GT(th->th_ack, tp->snd_max))) { 1975 rstreason = BANDLIM_RST_OPENPORT; 1976 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1977 goto dropwithreset; 1978 } 1979 if (IS_FASTOPEN(tp->t_flags)) { 1980 /* 1981 * When a TFO connection is in SYN_RECEIVED, the 1982 * only valid packets are the initial SYN, a 1983 * retransmit/copy of the initial SYN (possibly with 1984 * a subset of the original data), a valid ACK, a 1985 * FIN, or a RST. 1986 */ 1987 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1988 rstreason = BANDLIM_RST_OPENPORT; 1989 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1990 goto dropwithreset; 1991 } else if (thflags & TH_SYN) { 1992 /* non-initial SYN is ignored */ 1993 if ((tcp_timer_active(tp, TT_DELACK) || 1994 tcp_timer_active(tp, TT_REXMT))) 1995 goto drop; 1996 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1997 goto drop; 1998 } 1999 } 2000 break; 2001 2002 /* 2003 * If the state is SYN_SENT: 2004 * if seg contains a RST with valid ACK (SEQ.ACK has already 2005 * been verified), then drop the connection. 2006 * if seg contains a RST without an ACK, drop the seg. 2007 * if seg does not contain SYN, then drop the seg. 2008 * Otherwise this is an acceptable SYN segment 2009 * initialize tp->rcv_nxt and tp->irs 2010 * if seg contains ack then advance tp->snd_una 2011 * if seg contains an ECE and ECN support is enabled, the stream 2012 * is ECN capable. 2013 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2014 * arrange for segment to be acked (eventually) 2015 * continue processing rest of data/controls, beginning with URG 2016 */ 2017 case TCPS_SYN_SENT: 2018 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2019 TCP_PROBE5(connect__refused, NULL, tp, 2020 m, tp, th); 2021 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2022 tp = tcp_drop(tp, ECONNREFUSED); 2023 } 2024 if (thflags & TH_RST) 2025 goto drop; 2026 if (!(thflags & TH_SYN)) 2027 goto drop; 2028 2029 tp->irs = th->th_seq; 2030 tcp_rcvseqinit(tp); 2031 if (thflags & TH_ACK) { 2032 int tfo_partial_ack = 0; 2033 2034 TCPSTAT_INC(tcps_connects); 2035 soisconnected(so); 2036 #ifdef MAC 2037 mac_socketpeer_set_from_mbuf(m, so); 2038 #endif 2039 /* Do window scaling on this connection? */ 2040 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2041 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2042 tp->rcv_scale = tp->request_r_scale; 2043 } 2044 tp->rcv_adv += min(tp->rcv_wnd, 2045 TCP_MAXWIN << tp->rcv_scale); 2046 tp->snd_una++; /* SYN is acked */ 2047 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2048 tp->snd_nxt = tp->snd_una; 2049 /* 2050 * If not all the data that was sent in the TFO SYN 2051 * has been acked, resend the remainder right away. 2052 */ 2053 if (IS_FASTOPEN(tp->t_flags) && 2054 (tp->snd_una != tp->snd_max)) { 2055 tp->snd_nxt = th->th_ack; 2056 tfo_partial_ack = 1; 2057 } 2058 /* 2059 * If there's data, delay ACK; if there's also a FIN 2060 * ACKNOW will be turned on later. 2061 */ 2062 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2063 tcp_timer_activate(tp, TT_DELACK, 2064 tcp_delacktime); 2065 else 2066 tp->t_flags |= TF_ACKNOW; 2067 2068 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2069 2070 /* 2071 * Received <SYN,ACK> in SYN_SENT[*] state. 2072 * Transitions: 2073 * SYN_SENT --> ESTABLISHED 2074 * SYN_SENT* --> FIN_WAIT_1 2075 */ 2076 tp->t_starttime = ticks; 2077 if (tp->t_flags & TF_NEEDFIN) { 2078 tp->t_acktime = ticks; 2079 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2080 tp->t_flags &= ~TF_NEEDFIN; 2081 thflags &= ~TH_SYN; 2082 } else { 2083 tcp_state_change(tp, TCPS_ESTABLISHED); 2084 TCP_PROBE5(connect__established, NULL, tp, 2085 m, tp, th); 2086 cc_conn_init(tp); 2087 tcp_timer_activate(tp, TT_KEEP, 2088 TP_KEEPIDLE(tp)); 2089 } 2090 } else { 2091 /* 2092 * Received initial SYN in SYN-SENT[*] state => 2093 * simultaneous open. 2094 * If it succeeds, connection is * half-synchronized. 2095 * Otherwise, do 3-way handshake: 2096 * SYN-SENT -> SYN-RECEIVED 2097 * SYN-SENT* -> SYN-RECEIVED* 2098 */ 2099 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN); 2100 tcp_timer_activate(tp, TT_REXMT, 0); 2101 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2102 } 2103 2104 /* 2105 * Advance th->th_seq to correspond to first data byte. 2106 * If data, trim to stay within window, 2107 * dropping FIN if necessary. 2108 */ 2109 th->th_seq++; 2110 if (tlen > tp->rcv_wnd) { 2111 todrop = tlen - tp->rcv_wnd; 2112 m_adj(m, -todrop); 2113 tlen = tp->rcv_wnd; 2114 thflags &= ~TH_FIN; 2115 TCPSTAT_INC(tcps_rcvpackafterwin); 2116 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2117 } 2118 tp->snd_wl1 = th->th_seq - 1; 2119 tp->rcv_up = th->th_seq; 2120 /* 2121 * Client side of transaction: already sent SYN and data. 2122 * If the remote host used T/TCP to validate the SYN, 2123 * our data will be ACK'd; if so, enter normal data segment 2124 * processing in the middle of step 5, ack processing. 2125 * Otherwise, goto step 6. 2126 */ 2127 if (thflags & TH_ACK) 2128 goto process_ACK; 2129 2130 goto step6; 2131 } 2132 2133 /* 2134 * States other than LISTEN or SYN_SENT. 2135 * First check the RST flag and sequence number since reset segments 2136 * are exempt from the timestamp and connection count tests. This 2137 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2138 * below which allowed reset segments in half the sequence space 2139 * to fall though and be processed (which gives forged reset 2140 * segments with a random sequence number a 50 percent chance of 2141 * killing a connection). 2142 * Then check timestamp, if present. 2143 * Then check the connection count, if present. 2144 * Then check that at least some bytes of segment are within 2145 * receive window. If segment begins before rcv_nxt, 2146 * drop leading data (and SYN); if nothing left, just ack. 2147 */ 2148 if (thflags & TH_RST) { 2149 /* 2150 * RFC5961 Section 3.2 2151 * 2152 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2153 * - If RST is in window, we send challenge ACK. 2154 * 2155 * Note: to take into account delayed ACKs, we should 2156 * test against last_ack_sent instead of rcv_nxt. 2157 * Note 2: we handle special case of closed window, not 2158 * covered by the RFC. 2159 */ 2160 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2161 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2162 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2163 KASSERT(tp->t_state != TCPS_SYN_SENT, 2164 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2165 __func__, th, tp)); 2166 2167 if (V_tcp_insecure_rst || 2168 tp->last_ack_sent == th->th_seq) { 2169 TCPSTAT_INC(tcps_drops); 2170 /* Drop the connection. */ 2171 switch (tp->t_state) { 2172 case TCPS_SYN_RECEIVED: 2173 so->so_error = ECONNREFUSED; 2174 goto close; 2175 case TCPS_ESTABLISHED: 2176 case TCPS_FIN_WAIT_1: 2177 case TCPS_FIN_WAIT_2: 2178 case TCPS_CLOSE_WAIT: 2179 case TCPS_CLOSING: 2180 case TCPS_LAST_ACK: 2181 so->so_error = ECONNRESET; 2182 close: 2183 /* FALLTHROUGH */ 2184 default: 2185 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST); 2186 tp = tcp_close(tp); 2187 } 2188 } else { 2189 TCPSTAT_INC(tcps_badrst); 2190 tcp_send_challenge_ack(tp, th, m); 2191 m = NULL; 2192 } 2193 } 2194 goto drop; 2195 } 2196 2197 /* 2198 * RFC5961 Section 4.2 2199 * Send challenge ACK for any SYN in synchronized state. 2200 */ 2201 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2202 tp->t_state != TCPS_SYN_RECEIVED) { 2203 TCPSTAT_INC(tcps_badsyn); 2204 if (V_tcp_insecure_syn && 2205 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2206 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2207 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2208 tp = tcp_drop(tp, ECONNRESET); 2209 rstreason = BANDLIM_UNLIMITED; 2210 } else { 2211 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2212 tcp_send_challenge_ack(tp, th, m); 2213 m = NULL; 2214 } 2215 goto drop; 2216 } 2217 2218 /* 2219 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2220 * and it's less than ts_recent, drop it. 2221 */ 2222 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2223 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2224 /* Check to see if ts_recent is over 24 days old. */ 2225 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2226 /* 2227 * Invalidate ts_recent. If this segment updates 2228 * ts_recent, the age will be reset later and ts_recent 2229 * will get a valid value. If it does not, setting 2230 * ts_recent to zero will at least satisfy the 2231 * requirement that zero be placed in the timestamp 2232 * echo reply when ts_recent isn't valid. The 2233 * age isn't reset until we get a valid ts_recent 2234 * because we don't want out-of-order segments to be 2235 * dropped when ts_recent is old. 2236 */ 2237 tp->ts_recent = 0; 2238 } else { 2239 TCPSTAT_INC(tcps_rcvduppack); 2240 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2241 TCPSTAT_INC(tcps_pawsdrop); 2242 if (tlen) 2243 goto dropafterack; 2244 goto drop; 2245 } 2246 } 2247 2248 /* 2249 * In the SYN-RECEIVED state, validate that the packet belongs to 2250 * this connection before trimming the data to fit the receive 2251 * window. Check the sequence number versus IRS since we know 2252 * the sequence numbers haven't wrapped. This is a partial fix 2253 * for the "LAND" DoS attack. 2254 */ 2255 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2256 rstreason = BANDLIM_RST_OPENPORT; 2257 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2258 goto dropwithreset; 2259 } 2260 2261 todrop = tp->rcv_nxt - th->th_seq; 2262 if (todrop > 0) { 2263 if (thflags & TH_SYN) { 2264 thflags &= ~TH_SYN; 2265 th->th_seq++; 2266 if (th->th_urp > 1) 2267 th->th_urp--; 2268 else 2269 thflags &= ~TH_URG; 2270 todrop--; 2271 } 2272 /* 2273 * Following if statement from Stevens, vol. 2, p. 960. 2274 */ 2275 if (todrop > tlen 2276 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2277 /* 2278 * Any valid FIN must be to the left of the window. 2279 * At this point the FIN must be a duplicate or out 2280 * of sequence; drop it. 2281 */ 2282 thflags &= ~TH_FIN; 2283 2284 /* 2285 * Send an ACK to resynchronize and drop any data. 2286 * But keep on processing for RST or ACK. 2287 */ 2288 tp->t_flags |= TF_ACKNOW; 2289 todrop = tlen; 2290 TCPSTAT_INC(tcps_rcvduppack); 2291 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2292 } else { 2293 TCPSTAT_INC(tcps_rcvpartduppack); 2294 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2295 } 2296 /* 2297 * DSACK - add SACK block for dropped range 2298 */ 2299 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2300 tcp_update_sack_list(tp, th->th_seq, 2301 th->th_seq + todrop); 2302 /* 2303 * ACK now, as the next in-sequence segment 2304 * will clear the DSACK block again 2305 */ 2306 tp->t_flags |= TF_ACKNOW; 2307 } 2308 drop_hdrlen += todrop; /* drop from the top afterwards */ 2309 th->th_seq += todrop; 2310 tlen -= todrop; 2311 if (th->th_urp > todrop) 2312 th->th_urp -= todrop; 2313 else { 2314 thflags &= ~TH_URG; 2315 th->th_urp = 0; 2316 } 2317 } 2318 2319 /* 2320 * If new data are received on a connection after the 2321 * user processes are gone, then RST the other end if 2322 * no FIN has been processed. 2323 */ 2324 if ((tp->t_flags & TF_CLOSED) && tlen > 0 && 2325 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2326 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2327 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2328 "after socket was closed, " 2329 "sending RST and removing tcpcb\n", 2330 s, __func__, tcpstates[tp->t_state], tlen); 2331 free(s, M_TCPLOG); 2332 } 2333 tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE); 2334 /* tcp_close will kill the inp pre-log the Reset */ 2335 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST); 2336 tp = tcp_close(tp); 2337 TCPSTAT_INC(tcps_rcvafterclose); 2338 rstreason = BANDLIM_UNLIMITED; 2339 goto dropwithreset; 2340 } 2341 2342 /* 2343 * If segment ends after window, drop trailing data 2344 * (and PUSH and FIN); if nothing left, just ACK. 2345 */ 2346 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2347 if (todrop > 0) { 2348 TCPSTAT_INC(tcps_rcvpackafterwin); 2349 if (todrop >= tlen) { 2350 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2351 /* 2352 * If window is closed can only take segments at 2353 * window edge, and have to drop data and PUSH from 2354 * incoming segments. Continue processing, but 2355 * remember to ack. Otherwise, drop segment 2356 * and ack. 2357 */ 2358 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2359 tp->t_flags |= TF_ACKNOW; 2360 TCPSTAT_INC(tcps_rcvwinprobe); 2361 } else 2362 goto dropafterack; 2363 } else 2364 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2365 m_adj(m, -todrop); 2366 tlen -= todrop; 2367 thflags &= ~(TH_PUSH|TH_FIN); 2368 } 2369 2370 /* 2371 * If last ACK falls within this segment's sequence numbers, 2372 * record its timestamp. 2373 * NOTE: 2374 * 1) That the test incorporates suggestions from the latest 2375 * proposal of the [email protected] list (Braden 1993/04/26). 2376 * 2) That updating only on newer timestamps interferes with 2377 * our earlier PAWS tests, so this check should be solely 2378 * predicated on the sequence space of this segment. 2379 * 3) That we modify the segment boundary check to be 2380 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2381 * instead of RFC1323's 2382 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2383 * This modified check allows us to overcome RFC1323's 2384 * limitations as described in Stevens TCP/IP Illustrated 2385 * Vol. 2 p.869. In such cases, we can still calculate the 2386 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2387 */ 2388 if ((to.to_flags & TOF_TS) != 0 && 2389 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2390 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2391 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2392 tp->ts_recent_age = tcp_ts_getticks(); 2393 tp->ts_recent = to.to_tsval; 2394 } 2395 2396 /* 2397 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2398 * flag is on (half-synchronized state), then queue data for 2399 * later processing; else drop segment and return. 2400 */ 2401 if ((thflags & TH_ACK) == 0) { 2402 if (tp->t_state == TCPS_SYN_RECEIVED || 2403 (tp->t_flags & TF_NEEDSYN)) { 2404 if (tp->t_state == TCPS_SYN_RECEIVED && 2405 IS_FASTOPEN(tp->t_flags)) { 2406 tp->snd_wnd = tiwin; 2407 cc_conn_init(tp); 2408 } 2409 goto step6; 2410 } else if (tp->t_flags & TF_ACKNOW) 2411 goto dropafterack; 2412 else 2413 goto drop; 2414 } 2415 2416 /* 2417 * Ack processing. 2418 */ 2419 if (SEQ_GEQ(tp->snd_una, tp->iss + (TCP_MAXWIN << tp->snd_scale))) { 2420 /* Checking SEG.ACK against ISS is definitely redundant. */ 2421 tp->t_flags2 |= TF2_NO_ISS_CHECK; 2422 } 2423 if (!V_tcp_insecure_ack) { 2424 tcp_seq seq_min; 2425 bool ghost_ack_check; 2426 2427 if (tp->t_flags2 & TF2_NO_ISS_CHECK) { 2428 /* Check for too old ACKs (RFC 5961, Section 5.2). */ 2429 seq_min = tp->snd_una - tp->max_sndwnd; 2430 ghost_ack_check = false; 2431 } else { 2432 if (SEQ_GT(tp->iss + 1, tp->snd_una - tp->max_sndwnd)) { 2433 /* Checking for ghost ACKs is stricter. */ 2434 seq_min = tp->iss + 1; 2435 ghost_ack_check = true; 2436 } else { 2437 /* 2438 * Checking for too old ACKs (RFC 5961, 2439 * Section 5.2) is stricter. 2440 */ 2441 seq_min = tp->snd_una - tp->max_sndwnd; 2442 ghost_ack_check = false; 2443 } 2444 } 2445 if (SEQ_LT(th->th_ack, seq_min)) { 2446 if (ghost_ack_check) 2447 TCPSTAT_INC(tcps_rcvghostack); 2448 else 2449 TCPSTAT_INC(tcps_rcvacktooold); 2450 tcp_send_challenge_ack(tp, th, m); 2451 m = NULL; 2452 goto drop; 2453 } 2454 } 2455 switch (tp->t_state) { 2456 /* 2457 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2458 * ESTABLISHED state and continue processing. 2459 * The ACK was checked above. 2460 */ 2461 case TCPS_SYN_RECEIVED: 2462 2463 TCPSTAT_INC(tcps_connects); 2464 if (tp->t_flags & TF_SONOTCONN) { 2465 /* 2466 * Usually SYN_RECEIVED had been created from a LISTEN, 2467 * and solisten_enqueue() has already marked the socket 2468 * layer as connected. If it didn't, which can happen 2469 * only with an accept_filter(9), then the tp is marked 2470 * with TF_SONOTCONN. The other reason for this mark 2471 * to be set is a simultaneous open, a SYN_RECEIVED 2472 * that had been created from SYN_SENT. 2473 */ 2474 tp->t_flags &= ~TF_SONOTCONN; 2475 soisconnected(so); 2476 } 2477 /* Do window scaling? */ 2478 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2479 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2480 tp->rcv_scale = tp->request_r_scale; 2481 } 2482 tp->snd_wnd = tiwin; 2483 /* 2484 * Make transitions: 2485 * SYN-RECEIVED -> ESTABLISHED 2486 * SYN-RECEIVED* -> FIN-WAIT-1 2487 */ 2488 tp->t_starttime = ticks; 2489 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2490 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2491 tp->t_tfo_pending = NULL; 2492 } 2493 if (tp->t_flags & TF_NEEDFIN) { 2494 tp->t_acktime = ticks; 2495 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2496 tp->t_flags &= ~TF_NEEDFIN; 2497 } else { 2498 tcp_state_change(tp, TCPS_ESTABLISHED); 2499 TCP_PROBE5(accept__established, NULL, tp, 2500 m, tp, th); 2501 /* 2502 * TFO connections call cc_conn_init() during SYN 2503 * processing. Calling it again here for such 2504 * connections is not harmless as it would undo the 2505 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2506 * is retransmitted. 2507 */ 2508 if (!IS_FASTOPEN(tp->t_flags)) 2509 cc_conn_init(tp); 2510 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2511 } 2512 /* 2513 * Account for the ACK of our SYN prior to 2514 * regular ACK processing below, except for 2515 * simultaneous SYN, which is handled later. 2516 */ 2517 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2518 incforsyn = 1; 2519 /* 2520 * If segment contains data or ACK, will call tcp_reass() 2521 * later; if not, do so now to pass queued data to user. 2522 */ 2523 if (tlen == 0 && (thflags & TH_FIN) == 0) { 2524 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2525 (struct mbuf *)0); 2526 tcp_handle_wakeup(tp); 2527 } 2528 tp->snd_wl1 = th->th_seq - 1; 2529 /* FALLTHROUGH */ 2530 2531 /* 2532 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2533 * ACKs. If the ack is in the range 2534 * tp->snd_una < th->th_ack <= tp->snd_max 2535 * then advance tp->snd_una to th->th_ack and drop 2536 * data from the retransmission queue. If this ACK reflects 2537 * more up to date window information we update our window information. 2538 */ 2539 case TCPS_ESTABLISHED: 2540 case TCPS_FIN_WAIT_1: 2541 case TCPS_FIN_WAIT_2: 2542 case TCPS_CLOSE_WAIT: 2543 case TCPS_CLOSING: 2544 case TCPS_LAST_ACK: 2545 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2546 TCPSTAT_INC(tcps_rcvacktoomuch); 2547 goto dropafterack; 2548 } 2549 if (tcp_is_sack_recovery(tp, &to)) { 2550 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2551 if ((sack_changed != SACK_NOCHANGE) && 2552 (tp->t_flags & TF_LRD)) { 2553 tcp_sack_lost_retransmission(tp, th); 2554 } 2555 } else 2556 /* 2557 * Reset the value so that previous (valid) value 2558 * from the last ack with SACK doesn't get used. 2559 */ 2560 tp->sackhint.sacked_bytes = 0; 2561 2562 #ifdef TCP_HHOOK 2563 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2564 hhook_run_tcp_est_in(tp, th, &to); 2565 #endif 2566 2567 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2568 maxseg = tcp_maxseg(tp); 2569 if (tlen == 0 && 2570 (tiwin == tp->snd_wnd || 2571 (tp->t_flags & TF_SACK_PERMIT))) { 2572 /* 2573 * If this is the first time we've seen a 2574 * FIN from the remote, this is not a 2575 * duplicate and it needs to be processed 2576 * normally. This happens during a 2577 * simultaneous close. 2578 */ 2579 if ((thflags & TH_FIN) && 2580 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2581 tp->t_dupacks = 0; 2582 break; 2583 } 2584 TCPSTAT_INC(tcps_rcvdupack); 2585 /* 2586 * If we have outstanding data (other than 2587 * a window probe), this is a completely 2588 * duplicate ack (ie, window info didn't 2589 * change and FIN isn't set), 2590 * the ack is the biggest we've 2591 * seen and we've seen exactly our rexmt 2592 * threshold of them, assume a packet 2593 * has been dropped and retransmit it. 2594 * Kludge snd_nxt & the congestion 2595 * window so we send only this one 2596 * packet. 2597 * 2598 * We know we're losing at the current 2599 * window size so do congestion avoidance 2600 * (set ssthresh to half the current window 2601 * and pull our congestion window back to 2602 * the new ssthresh). 2603 * 2604 * Dup acks mean that packets have left the 2605 * network (they're now cached at the receiver) 2606 * so bump cwnd by the amount in the receiver 2607 * to keep a constant cwnd packets in the 2608 * network. 2609 * 2610 * When using TCP ECN, notify the peer that 2611 * we reduced the cwnd. 2612 */ 2613 /* 2614 * Following 2 kinds of acks should not affect 2615 * dupack counting: 2616 * 1) Old acks 2617 * 2) Acks with SACK but without any new SACK 2618 * information in them. These could result from 2619 * any anomaly in the network like a switch 2620 * duplicating packets or a possible DoS attack. 2621 */ 2622 if (th->th_ack != tp->snd_una || 2623 (tcp_is_sack_recovery(tp, &to) && 2624 (sack_changed == SACK_NOCHANGE))) 2625 break; 2626 else if (!tcp_timer_active(tp, TT_REXMT)) 2627 tp->t_dupacks = 0; 2628 else if (++tp->t_dupacks > tcprexmtthresh || 2629 IN_FASTRECOVERY(tp->t_flags)) { 2630 cc_ack_received(tp, th, nsegs, 2631 CC_DUPACK); 2632 if (V_tcp_do_prr && 2633 IN_FASTRECOVERY(tp->t_flags) && 2634 (tp->t_flags & TF_SACK_PERMIT)) { 2635 tcp_do_prr_ack(tp, th, &to, sack_changed); 2636 } else if (tcp_is_sack_recovery(tp, &to) && 2637 IN_FASTRECOVERY(tp->t_flags)) { 2638 int awnd; 2639 2640 /* 2641 * Compute the amount of data in flight first. 2642 * We can inject new data into the pipe iff 2643 * we have less than 1/2 the original window's 2644 * worth of data in flight. 2645 */ 2646 if (V_tcp_do_newsack) 2647 awnd = tcp_compute_pipe(tp); 2648 else 2649 awnd = (tp->snd_nxt - tp->snd_fack) + 2650 tp->sackhint.sack_bytes_rexmit; 2651 2652 if (awnd < tp->snd_ssthresh) { 2653 tp->snd_cwnd += maxseg; 2654 if (tp->snd_cwnd > tp->snd_ssthresh) 2655 tp->snd_cwnd = tp->snd_ssthresh; 2656 } 2657 } else 2658 tp->snd_cwnd += maxseg; 2659 (void) tcp_output(tp); 2660 goto drop; 2661 } else if (tp->t_dupacks == tcprexmtthresh || 2662 (tp->t_flags & TF_SACK_PERMIT && 2663 V_tcp_do_newsack && 2664 tp->sackhint.sacked_bytes > 2665 (tcprexmtthresh - 1) * maxseg)) { 2666 enter_recovery: 2667 /* 2668 * Above is the RFC6675 trigger condition of 2669 * more than (dupthresh-1)*maxseg sacked data. 2670 * If the count of holes in the 2671 * scoreboard is >= dupthresh, we could 2672 * also enter loss recovery, but don't 2673 * have that value readily available. 2674 */ 2675 tp->t_dupacks = tcprexmtthresh; 2676 tcp_seq onxt = tp->snd_nxt; 2677 2678 /* 2679 * If we're doing sack, or prr, check 2680 * to see if we're already in sack 2681 * recovery. If we're not doing sack, 2682 * check to see if we're in newreno 2683 * recovery. 2684 */ 2685 if (V_tcp_do_prr || 2686 (tp->t_flags & TF_SACK_PERMIT)) { 2687 if (IN_FASTRECOVERY(tp->t_flags)) { 2688 tp->t_dupacks = 0; 2689 break; 2690 } 2691 } else { 2692 if (SEQ_LEQ(th->th_ack, 2693 tp->snd_recover)) { 2694 tp->t_dupacks = 0; 2695 break; 2696 } 2697 } 2698 /* Congestion signal before ack. */ 2699 cc_cong_signal(tp, th, CC_NDUPACK); 2700 cc_ack_received(tp, th, nsegs, 2701 CC_DUPACK); 2702 tcp_timer_activate(tp, TT_REXMT, 0); 2703 tp->t_rtttime = 0; 2704 if (V_tcp_do_prr) { 2705 /* 2706 * snd_ssthresh is already updated by 2707 * cc_cong_signal. 2708 */ 2709 if (tcp_is_sack_recovery(tp, &to)) { 2710 /* 2711 * Exclude Limited Transmit 2712 * segments here 2713 */ 2714 tp->sackhint.prr_delivered = 2715 maxseg; 2716 } else { 2717 tp->sackhint.prr_delivered = 2718 imin(tp->snd_max - tp->snd_una, 2719 imin(INT_MAX / 65536, 2720 tp->t_dupacks) * maxseg); 2721 } 2722 tp->sackhint.recover_fs = max(1, 2723 tp->snd_nxt - tp->snd_una); 2724 } 2725 if (tcp_is_sack_recovery(tp, &to)) { 2726 TCPSTAT_INC( 2727 tcps_sack_recovery_episode); 2728 tp->snd_recover = tp->snd_nxt; 2729 tp->snd_cwnd = maxseg; 2730 (void) tcp_output(tp); 2731 if (SEQ_GT(th->th_ack, tp->snd_una)) 2732 goto resume_partialack; 2733 goto drop; 2734 } 2735 tp->snd_nxt = th->th_ack; 2736 tp->snd_cwnd = maxseg; 2737 (void) tcp_output(tp); 2738 KASSERT(tp->snd_limited <= 2, 2739 ("%s: tp->snd_limited too big", 2740 __func__)); 2741 tp->snd_cwnd = tp->snd_ssthresh + 2742 maxseg * 2743 (tp->t_dupacks - tp->snd_limited); 2744 if (SEQ_GT(onxt, tp->snd_nxt)) 2745 tp->snd_nxt = onxt; 2746 goto drop; 2747 } else if (V_tcp_do_rfc3042) { 2748 /* 2749 * Process first and second duplicate 2750 * ACKs. Each indicates a segment 2751 * leaving the network, creating room 2752 * for more. Make sure we can send a 2753 * packet on reception of each duplicate 2754 * ACK by increasing snd_cwnd by one 2755 * segment. Restore the original 2756 * snd_cwnd after packet transmission. 2757 */ 2758 cc_ack_received(tp, th, nsegs, 2759 CC_DUPACK); 2760 uint32_t oldcwnd = tp->snd_cwnd; 2761 tcp_seq oldsndmax = tp->snd_max; 2762 u_int sent; 2763 int avail; 2764 2765 KASSERT(tp->t_dupacks == 1 || 2766 tp->t_dupacks == 2, 2767 ("%s: dupacks not 1 or 2", 2768 __func__)); 2769 if (tp->t_dupacks == 1) 2770 tp->snd_limited = 0; 2771 tp->snd_cwnd = 2772 (tp->snd_nxt - tp->snd_una) + 2773 (tp->t_dupacks - tp->snd_limited) * 2774 maxseg; 2775 /* 2776 * Only call tcp_output when there 2777 * is new data available to be sent 2778 * or we need to send an ACK. 2779 */ 2780 SOCKBUF_LOCK(&so->so_snd); 2781 avail = sbavail(&so->so_snd) - 2782 (tp->snd_nxt - tp->snd_una); 2783 SOCKBUF_UNLOCK(&so->so_snd); 2784 if (avail > 0 || tp->t_flags & TF_ACKNOW) 2785 (void) tcp_output(tp); 2786 sent = tp->snd_max - oldsndmax; 2787 if (sent > maxseg) { 2788 KASSERT((tp->t_dupacks == 2 && 2789 tp->snd_limited == 0) || 2790 (sent == maxseg + 1 && 2791 tp->t_flags & TF_SENTFIN), 2792 ("%s: sent too much", 2793 __func__)); 2794 tp->snd_limited = 2; 2795 } else if (sent > 0) 2796 ++tp->snd_limited; 2797 tp->snd_cwnd = oldcwnd; 2798 goto drop; 2799 } 2800 } 2801 break; 2802 } else { 2803 /* 2804 * This ack is advancing the left edge, reset the 2805 * counter. 2806 */ 2807 tp->t_dupacks = 0; 2808 /* 2809 * If this ack also has new SACK info, increment the 2810 * counter as per rfc6675. The variable 2811 * sack_changed tracks all changes to the SACK 2812 * scoreboard, including when partial ACKs without 2813 * SACK options are received, and clear the scoreboard 2814 * from the left side. Such partial ACKs should not be 2815 * counted as dupacks here. 2816 */ 2817 if (tcp_is_sack_recovery(tp, &to) && 2818 (sack_changed != SACK_NOCHANGE)) { 2819 tp->t_dupacks++; 2820 /* limit overhead by setting maxseg last */ 2821 if (!IN_FASTRECOVERY(tp->t_flags) && 2822 (tp->sackhint.sacked_bytes > 2823 ((tcprexmtthresh - 1) * 2824 (maxseg = tcp_maxseg(tp))))) { 2825 goto enter_recovery; 2826 } 2827 } 2828 } 2829 2830 resume_partialack: 2831 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2832 ("%s: th_ack <= snd_una", __func__)); 2833 2834 /* 2835 * If the congestion window was inflated to account 2836 * for the other side's cached packets, retract it. 2837 */ 2838 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2839 if (IN_FASTRECOVERY(tp->t_flags)) { 2840 if (tp->t_flags & TF_SACK_PERMIT) { 2841 if (V_tcp_do_prr && 2842 (to.to_flags & TOF_SACK)) { 2843 tcp_timer_activate(tp, 2844 TT_REXMT, 0); 2845 tp->t_rtttime = 0; 2846 tcp_do_prr_ack(tp, th, &to, 2847 sack_changed); 2848 tp->t_flags |= TF_ACKNOW; 2849 (void) tcp_output(tp); 2850 } else { 2851 tcp_sack_partialack(tp, th); 2852 } 2853 } else { 2854 tcp_newreno_partial_ack(tp, th); 2855 } 2856 } else if (IN_CONGRECOVERY(tp->t_flags) && 2857 (V_tcp_do_prr)) { 2858 tp->sackhint.delivered_data = 2859 BYTES_THIS_ACK(tp, th); 2860 tp->snd_fack = th->th_ack; 2861 /* 2862 * During ECN cwnd reduction 2863 * always use PRR-SSRB 2864 */ 2865 tcp_do_prr_ack(tp, th, &to, SACK_CHANGE); 2866 (void) tcp_output(tp); 2867 } 2868 } 2869 /* 2870 * If we reach this point, ACK is not a duplicate, 2871 * i.e., it ACKs something we sent. 2872 */ 2873 if (tp->t_flags & TF_NEEDSYN) { 2874 /* 2875 * T/TCP: Connection was half-synchronized, and our 2876 * SYN has been ACK'd (so connection is now fully 2877 * synchronized). Go to non-starred state, 2878 * increment snd_una for ACK of SYN, and check if 2879 * we can do window scaling. 2880 */ 2881 tp->t_flags &= ~TF_NEEDSYN; 2882 tp->snd_una++; 2883 /* Do window scaling? */ 2884 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2885 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2886 tp->rcv_scale = tp->request_r_scale; 2887 /* Send window already scaled. */ 2888 } 2889 } 2890 2891 process_ACK: 2892 INP_WLOCK_ASSERT(inp); 2893 2894 /* 2895 * Adjust for the SYN bit in sequence space, 2896 * but don't account for it in cwnd calculations. 2897 * This is for the SYN_RECEIVED, non-simultaneous 2898 * SYN case. SYN_SENT and simultaneous SYN are 2899 * treated elsewhere. 2900 */ 2901 if (incforsyn) 2902 tp->snd_una++; 2903 acked = BYTES_THIS_ACK(tp, th); 2904 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2905 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2906 tp->snd_una, th->th_ack, tp, m)); 2907 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2908 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2909 2910 /* 2911 * If we just performed our first retransmit, and the ACK 2912 * arrives within our recovery window, then it was a mistake 2913 * to do the retransmit in the first place. Recover our 2914 * original cwnd and ssthresh, and proceed to transmit where 2915 * we left off. 2916 */ 2917 if (tp->t_rxtshift == 1 && 2918 tp->t_flags & TF_PREVVALID && 2919 tp->t_badrxtwin != 0 && 2920 to.to_flags & TOF_TS && 2921 to.to_tsecr != 0 && 2922 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 2923 cc_cong_signal(tp, th, CC_RTO_ERR); 2924 2925 /* 2926 * If we have a timestamp reply, update smoothed 2927 * round trip time. If no timestamp is present but 2928 * transmit timer is running and timed sequence 2929 * number was acked, update smoothed round trip time. 2930 * Since we now have an rtt measurement, cancel the 2931 * timer backoff (cf., Phil Karn's retransmit alg.). 2932 * Recompute the initial retransmit timer. 2933 * 2934 * Some boxes send broken timestamp replies 2935 * during the SYN+ACK phase, ignore 2936 * timestamps of 0 or we could calculate a 2937 * huge RTT and blow up the retransmit timer. 2938 */ 2939 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2940 uint32_t t; 2941 2942 t = tcp_ts_getticks() - to.to_tsecr; 2943 if (!tp->t_rttlow || tp->t_rttlow > t) 2944 tp->t_rttlow = t; 2945 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2946 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2947 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2948 tp->t_rttlow = ticks - tp->t_rtttime; 2949 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2950 } 2951 2952 SOCKBUF_LOCK(&so->so_snd); 2953 /* 2954 * Clear t_acktime if remote side has ACKd all data in the 2955 * socket buffer and FIN (if applicable). 2956 * Otherwise, update t_acktime if we received a sufficiently 2957 * large ACK. 2958 */ 2959 if ((tp->t_state <= TCPS_CLOSE_WAIT && 2960 acked == sbavail(&so->so_snd)) || 2961 acked > sbavail(&so->so_snd)) 2962 tp->t_acktime = 0; 2963 else if (acked > 1) 2964 tp->t_acktime = ticks; 2965 2966 /* 2967 * If all outstanding data is acked, stop retransmit 2968 * timer and remember to restart (more output or persist). 2969 * If there is more data to be acked, restart retransmit 2970 * timer, using current (possibly backed-off) value. 2971 */ 2972 if (th->th_ack == tp->snd_max) { 2973 tcp_timer_activate(tp, TT_REXMT, 0); 2974 needoutput = 1; 2975 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2976 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); 2977 2978 /* 2979 * If no data (only SYN) was ACK'd, 2980 * skip rest of ACK processing. 2981 */ 2982 if (acked == 0) { 2983 SOCKBUF_UNLOCK(&so->so_snd); 2984 goto step6; 2985 } 2986 2987 /* 2988 * Let the congestion control algorithm update congestion 2989 * control related information. This typically means increasing 2990 * the congestion window. 2991 */ 2992 cc_ack_received(tp, th, nsegs, CC_ACK); 2993 2994 if (acked > sbavail(&so->so_snd)) { 2995 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2996 tp->snd_wnd -= sbavail(&so->so_snd); 2997 else 2998 tp->snd_wnd = 0; 2999 mfree = sbcut_locked(&so->so_snd, 3000 (int)sbavail(&so->so_snd)); 3001 ourfinisacked = 1; 3002 } else { 3003 mfree = sbcut_locked(&so->so_snd, acked); 3004 if (tp->snd_wnd >= (uint32_t) acked) 3005 tp->snd_wnd -= acked; 3006 else 3007 tp->snd_wnd = 0; 3008 ourfinisacked = 0; 3009 } 3010 /* NB: sowwakeup_locked() does an implicit unlock. */ 3011 sowwakeup_locked(so); 3012 m_freem(mfree); 3013 /* Detect una wraparound. */ 3014 if (!IN_RECOVERY(tp->t_flags) && 3015 SEQ_GT(tp->snd_una, tp->snd_recover) && 3016 SEQ_LEQ(th->th_ack, tp->snd_recover)) 3017 tp->snd_recover = th->th_ack - 1; 3018 tp->snd_una = th->th_ack; 3019 if (IN_RECOVERY(tp->t_flags) && 3020 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 3021 cc_post_recovery(tp, th); 3022 } 3023 if (tp->t_flags & TF_SACK_PERMIT) { 3024 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 3025 tp->snd_recover = tp->snd_una; 3026 } 3027 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 3028 tp->snd_nxt = tp->snd_una; 3029 3030 switch (tp->t_state) { 3031 /* 3032 * In FIN_WAIT_1 STATE in addition to the processing 3033 * for the ESTABLISHED state if our FIN is now acknowledged 3034 * then enter FIN_WAIT_2. 3035 */ 3036 case TCPS_FIN_WAIT_1: 3037 if (ourfinisacked) { 3038 /* 3039 * If we can't receive any more 3040 * data, then closing user can proceed. 3041 * Starting the timer is contrary to the 3042 * specification, but if we don't get a FIN 3043 * we'll hang forever. 3044 */ 3045 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3046 soisdisconnected(so); 3047 tcp_timer_activate(tp, TT_2MSL, 3048 (tcp_fast_finwait2_recycle ? 3049 tcp_finwait2_timeout : 3050 TP_MAXIDLE(tp))); 3051 } 3052 tcp_state_change(tp, TCPS_FIN_WAIT_2); 3053 } 3054 break; 3055 3056 /* 3057 * In CLOSING STATE in addition to the processing for 3058 * the ESTABLISHED state if the ACK acknowledges our FIN 3059 * then enter the TIME-WAIT state, otherwise ignore 3060 * the segment. 3061 */ 3062 case TCPS_CLOSING: 3063 if (ourfinisacked) { 3064 tcp_twstart(tp); 3065 m_freem(m); 3066 return; 3067 } 3068 break; 3069 3070 /* 3071 * In LAST_ACK, we may still be waiting for data to drain 3072 * and/or to be acked, as well as for the ack of our FIN. 3073 * If our FIN is now acknowledged, delete the TCB, 3074 * enter the closed state and return. 3075 */ 3076 case TCPS_LAST_ACK: 3077 if (ourfinisacked) { 3078 tp = tcp_close(tp); 3079 goto drop; 3080 } 3081 break; 3082 } 3083 } 3084 3085 step6: 3086 INP_WLOCK_ASSERT(inp); 3087 3088 /* 3089 * Update window information. 3090 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3091 */ 3092 if ((thflags & TH_ACK) && 3093 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3094 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3095 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3096 /* keep track of pure window updates */ 3097 if (tlen == 0 && 3098 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3099 TCPSTAT_INC(tcps_rcvwinupd); 3100 tp->snd_wnd = tiwin; 3101 tp->snd_wl1 = th->th_seq; 3102 tp->snd_wl2 = th->th_ack; 3103 if (tp->snd_wnd > tp->max_sndwnd) 3104 tp->max_sndwnd = tp->snd_wnd; 3105 needoutput = 1; 3106 } 3107 3108 /* 3109 * Process segments with URG. 3110 */ 3111 if ((thflags & TH_URG) && th->th_urp && 3112 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3113 /* 3114 * This is a kludge, but if we receive and accept 3115 * random urgent pointers, we'll crash in 3116 * soreceive. It's hard to imagine someone 3117 * actually wanting to send this much urgent data. 3118 */ 3119 SOCKBUF_LOCK(&so->so_rcv); 3120 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3121 th->th_urp = 0; /* XXX */ 3122 thflags &= ~TH_URG; /* XXX */ 3123 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3124 goto dodata; /* XXX */ 3125 } 3126 /* 3127 * If this segment advances the known urgent pointer, 3128 * then mark the data stream. This should not happen 3129 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3130 * a FIN has been received from the remote side. 3131 * In these states we ignore the URG. 3132 * 3133 * According to RFC961 (Assigned Protocols), 3134 * the urgent pointer points to the last octet 3135 * of urgent data. We continue, however, 3136 * to consider it to indicate the first octet 3137 * of data past the urgent section as the original 3138 * spec states (in one of two places). 3139 */ 3140 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3141 tp->rcv_up = th->th_seq + th->th_urp; 3142 so->so_oobmark = sbavail(&so->so_rcv) + 3143 (tp->rcv_up - tp->rcv_nxt) - 1; 3144 if (so->so_oobmark == 0) 3145 so->so_rcv.sb_state |= SBS_RCVATMARK; 3146 sohasoutofband(so); 3147 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3148 } 3149 SOCKBUF_UNLOCK(&so->so_rcv); 3150 /* 3151 * Remove out of band data so doesn't get presented to user. 3152 * This can happen independent of advancing the URG pointer, 3153 * but if two URG's are pending at once, some out-of-band 3154 * data may creep in... ick. 3155 */ 3156 if (th->th_urp <= (uint32_t)tlen && 3157 !(so->so_options & SO_OOBINLINE)) { 3158 /* hdr drop is delayed */ 3159 tcp_pulloutofband(so, th, m, drop_hdrlen); 3160 } 3161 } else { 3162 /* 3163 * If no out of band data is expected, 3164 * pull receive urgent pointer along 3165 * with the receive window. 3166 */ 3167 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3168 tp->rcv_up = tp->rcv_nxt; 3169 } 3170 dodata: /* XXX */ 3171 INP_WLOCK_ASSERT(inp); 3172 3173 /* 3174 * Process the segment text, merging it into the TCP sequencing queue, 3175 * and arranging for acknowledgment of receipt if necessary. 3176 * This process logically involves adjusting tp->rcv_wnd as data 3177 * is presented to the user (this happens in tcp_usrreq.c, 3178 * case PRU_RCVD). If a FIN has already been received on this 3179 * connection then we just ignore the text. 3180 */ 3181 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3182 IS_FASTOPEN(tp->t_flags)); 3183 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3184 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3185 tcp_seq save_start = th->th_seq; 3186 tcp_seq save_rnxt = tp->rcv_nxt; 3187 int save_tlen = tlen; 3188 m_adj(m, drop_hdrlen); /* delayed header drop */ 3189 /* 3190 * Insert segment which includes th into TCP reassembly queue 3191 * with control block tp. Set thflags to whether reassembly now 3192 * includes a segment with FIN. This handles the common case 3193 * inline (segment is the next to be received on an established 3194 * connection, and the queue is empty), avoiding linkage into 3195 * and removal from the queue and repetition of various 3196 * conversions. 3197 * Set DELACK for segments received in order, but ack 3198 * immediately when segments are out of order (so 3199 * fast retransmit can work). 3200 */ 3201 if (th->th_seq == tp->rcv_nxt && 3202 SEGQ_EMPTY(tp) && 3203 (TCPS_HAVEESTABLISHED(tp->t_state) || 3204 tfo_syn)) { 3205 if (DELAY_ACK(tp, tlen) || tfo_syn) 3206 tp->t_flags |= TF_DELACK; 3207 else 3208 tp->t_flags |= TF_ACKNOW; 3209 tp->rcv_nxt += tlen; 3210 if (tlen && 3211 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3212 (tp->t_fbyte_in == 0)) { 3213 tp->t_fbyte_in = ticks; 3214 if (tp->t_fbyte_in == 0) 3215 tp->t_fbyte_in = 1; 3216 if (tp->t_fbyte_out && tp->t_fbyte_in) 3217 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3218 } 3219 thflags = tcp_get_flags(th) & TH_FIN; 3220 TCPSTAT_INC(tcps_rcvpack); 3221 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3222 SOCKBUF_LOCK(&so->so_rcv); 3223 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3224 m_freem(m); 3225 else 3226 sbappendstream_locked(&so->so_rcv, m, 0); 3227 tp->t_flags |= TF_WAKESOR; 3228 } else { 3229 /* 3230 * XXX: Due to the header drop above "th" is 3231 * theoretically invalid by now. Fortunately 3232 * m_adj() doesn't actually frees any mbufs 3233 * when trimming from the head. 3234 */ 3235 tcp_seq temp = save_start; 3236 3237 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3238 tp->t_flags |= TF_ACKNOW; 3239 } 3240 if ((tp->t_flags & TF_SACK_PERMIT) && 3241 (save_tlen > 0) && 3242 TCPS_HAVEESTABLISHED(tp->t_state)) { 3243 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3244 /* 3245 * DSACK actually handled in the fastpath 3246 * above. 3247 */ 3248 tcp_update_sack_list(tp, save_start, 3249 save_start + save_tlen); 3250 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3251 if ((tp->rcv_numsacks >= 1) && 3252 (tp->sackblks[0].end == save_start)) { 3253 /* 3254 * Partial overlap, recorded at todrop 3255 * above. 3256 */ 3257 tcp_update_sack_list(tp, 3258 tp->sackblks[0].start, 3259 tp->sackblks[0].end); 3260 } else { 3261 tcp_update_dsack_list(tp, save_start, 3262 save_start + save_tlen); 3263 } 3264 } else if (tlen >= save_tlen) { 3265 /* Update of sackblks. */ 3266 tcp_update_dsack_list(tp, save_start, 3267 save_start + save_tlen); 3268 } else if (tlen > 0) { 3269 tcp_update_dsack_list(tp, save_start, 3270 save_start + tlen); 3271 } 3272 } 3273 tcp_handle_wakeup(tp); 3274 #if 0 3275 /* 3276 * Note the amount of data that peer has sent into 3277 * our window, in order to estimate the sender's 3278 * buffer size. 3279 * XXX: Unused. 3280 */ 3281 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3282 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3283 else 3284 len = so->so_rcv.sb_hiwat; 3285 #endif 3286 } else { 3287 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 3288 if (tlen > 0) { 3289 if ((thflags & TH_FIN) != 0) { 3290 log(LOG_DEBUG, "%s; %s: %s: " 3291 "Received %d bytes of data and FIN " 3292 "after having received a FIN, " 3293 "just dropping both\n", 3294 s, __func__, 3295 tcpstates[tp->t_state], tlen); 3296 } else { 3297 log(LOG_DEBUG, "%s; %s: %s: " 3298 "Received %d bytes of data " 3299 "after having received a FIN, " 3300 "just dropping it\n", 3301 s, __func__, 3302 tcpstates[tp->t_state], tlen); 3303 } 3304 } else { 3305 if ((thflags & TH_FIN) != 0) { 3306 log(LOG_DEBUG, "%s; %s: %s: " 3307 "Received FIN " 3308 "after having received a FIN, " 3309 "just dropping it\n", 3310 s, __func__, 3311 tcpstates[tp->t_state]); 3312 } 3313 } 3314 free(s, M_TCPLOG); 3315 } 3316 m_freem(m); 3317 thflags &= ~TH_FIN; 3318 } 3319 3320 /* 3321 * If FIN is received ACK the FIN and let the user know 3322 * that the connection is closing. 3323 */ 3324 if (thflags & TH_FIN) { 3325 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3326 /* The socket upcall is handled by socantrcvmore. */ 3327 socantrcvmore(so); 3328 /* 3329 * If connection is half-synchronized 3330 * (ie NEEDSYN flag on) then delay ACK, 3331 * so it may be piggybacked when SYN is sent. 3332 * Otherwise, since we received a FIN then no 3333 * more input can be expected, send ACK now. 3334 */ 3335 if (tp->t_flags & TF_NEEDSYN) 3336 tp->t_flags |= TF_DELACK; 3337 else 3338 tp->t_flags |= TF_ACKNOW; 3339 tp->rcv_nxt++; 3340 } 3341 switch (tp->t_state) { 3342 /* 3343 * In SYN_RECEIVED and ESTABLISHED STATES 3344 * enter the CLOSE_WAIT state. 3345 */ 3346 case TCPS_SYN_RECEIVED: 3347 tp->t_starttime = ticks; 3348 /* FALLTHROUGH */ 3349 case TCPS_ESTABLISHED: 3350 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3351 break; 3352 3353 /* 3354 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3355 * enter the CLOSING state. 3356 */ 3357 case TCPS_FIN_WAIT_1: 3358 tcp_state_change(tp, TCPS_CLOSING); 3359 break; 3360 3361 /* 3362 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3363 * starting the time-wait timer, turning off the other 3364 * standard timers. 3365 */ 3366 case TCPS_FIN_WAIT_2: 3367 tcp_twstart(tp); 3368 return; 3369 } 3370 } 3371 TCP_PROBE3(debug__input, tp, th, m); 3372 3373 /* 3374 * Return any desired output. 3375 */ 3376 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3377 (void) tcp_output(tp); 3378 3379 check_delack: 3380 INP_WLOCK_ASSERT(inp); 3381 3382 if (tp->t_flags & TF_DELACK) { 3383 tp->t_flags &= ~TF_DELACK; 3384 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3385 } 3386 INP_WUNLOCK(inp); 3387 return; 3388 3389 dropafterack: 3390 /* 3391 * Generate an ACK dropping incoming segment if it occupies 3392 * sequence space, where the ACK reflects our state. 3393 * 3394 * We can now skip the test for the RST flag since all 3395 * paths to this code happen after packets containing 3396 * RST have been dropped. 3397 * 3398 * In the SYN-RECEIVED state, don't send an ACK unless the 3399 * segment we received passes the SYN-RECEIVED ACK test. 3400 * If it fails send a RST. This breaks the loop in the 3401 * "LAND" DoS attack, and also prevents an ACK storm 3402 * between two listening ports that have been sent forged 3403 * SYN segments, each with the source address of the other. 3404 */ 3405 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3406 (SEQ_GT(tp->snd_una, th->th_ack) || 3407 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3408 rstreason = BANDLIM_RST_OPENPORT; 3409 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 3410 goto dropwithreset; 3411 } 3412 TCP_PROBE3(debug__input, tp, th, m); 3413 tp->t_flags |= TF_ACKNOW; 3414 (void) tcp_output(tp); 3415 INP_WUNLOCK(inp); 3416 m_freem(m); 3417 return; 3418 3419 dropwithreset: 3420 if (tp != NULL) { 3421 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3422 INP_WUNLOCK(inp); 3423 } else 3424 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3425 return; 3426 3427 drop: 3428 /* 3429 * Drop space held by incoming segment and return. 3430 */ 3431 TCP_PROBE3(debug__input, tp, th, m); 3432 if (tp != NULL) { 3433 INP_WUNLOCK(inp); 3434 } 3435 m_freem(m); 3436 } 3437 3438 /* 3439 * Issue RST and make ACK acceptable to originator of segment. 3440 * The mbuf must still include the original packet header. 3441 * tp may be NULL. 3442 */ 3443 void 3444 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3445 int tlen, int rstreason) 3446 { 3447 #ifdef INET 3448 struct ip *ip; 3449 #endif 3450 #ifdef INET6 3451 struct ip6_hdr *ip6; 3452 #endif 3453 3454 if (tp != NULL) { 3455 INP_LOCK_ASSERT(tptoinpcb(tp)); 3456 } 3457 3458 /* Don't bother if destination was broadcast/multicast. */ 3459 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3460 goto drop; 3461 #ifdef INET6 3462 if (mtod(m, struct ip *)->ip_v == 6) { 3463 ip6 = mtod(m, struct ip6_hdr *); 3464 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3465 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3466 goto drop; 3467 /* IPv6 anycast check is done at tcp6_input() */ 3468 } 3469 #endif 3470 #if defined(INET) && defined(INET6) 3471 else 3472 #endif 3473 #ifdef INET 3474 { 3475 ip = mtod(m, struct ip *); 3476 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3477 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3478 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3479 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3480 goto drop; 3481 } 3482 #endif 3483 3484 /* Perform bandwidth limiting. */ 3485 if (badport_bandlim(rstreason) < 0) 3486 goto drop; 3487 3488 /* tcp_respond consumes the mbuf chain. */ 3489 if (tcp_get_flags(th) & TH_ACK) { 3490 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3491 th->th_ack, TH_RST); 3492 } else { 3493 if (tcp_get_flags(th) & TH_SYN) 3494 tlen++; 3495 if (tcp_get_flags(th) & TH_FIN) 3496 tlen++; 3497 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3498 (tcp_seq)0, TH_RST|TH_ACK); 3499 } 3500 return; 3501 drop: 3502 m_freem(m); 3503 } 3504 3505 /* 3506 * Parse TCP options and place in tcpopt. 3507 */ 3508 void 3509 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3510 { 3511 int opt, optlen; 3512 3513 to->to_flags = 0; 3514 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3515 opt = cp[0]; 3516 if (opt == TCPOPT_EOL) 3517 break; 3518 if (opt == TCPOPT_NOP) 3519 optlen = 1; 3520 else { 3521 if (cnt < 2) 3522 break; 3523 optlen = cp[1]; 3524 if (optlen < 2 || optlen > cnt) 3525 break; 3526 } 3527 switch (opt) { 3528 case TCPOPT_MAXSEG: 3529 if (optlen != TCPOLEN_MAXSEG) 3530 continue; 3531 if (!(flags & TO_SYN)) 3532 continue; 3533 to->to_flags |= TOF_MSS; 3534 bcopy((char *)cp + 2, 3535 (char *)&to->to_mss, sizeof(to->to_mss)); 3536 to->to_mss = ntohs(to->to_mss); 3537 break; 3538 case TCPOPT_WINDOW: 3539 if (optlen != TCPOLEN_WINDOW) 3540 continue; 3541 if (!(flags & TO_SYN)) 3542 continue; 3543 to->to_flags |= TOF_SCALE; 3544 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3545 break; 3546 case TCPOPT_TIMESTAMP: 3547 if (optlen != TCPOLEN_TIMESTAMP) 3548 continue; 3549 to->to_flags |= TOF_TS; 3550 bcopy((char *)cp + 2, 3551 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3552 to->to_tsval = ntohl(to->to_tsval); 3553 bcopy((char *)cp + 6, 3554 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3555 to->to_tsecr = ntohl(to->to_tsecr); 3556 break; 3557 case TCPOPT_SIGNATURE: 3558 /* 3559 * In order to reply to a host which has set the 3560 * TCP_SIGNATURE option in its initial SYN, we have 3561 * to record the fact that the option was observed 3562 * here for the syncache code to perform the correct 3563 * response. 3564 */ 3565 if (optlen != TCPOLEN_SIGNATURE) 3566 continue; 3567 to->to_flags |= TOF_SIGNATURE; 3568 to->to_signature = cp + 2; 3569 break; 3570 case TCPOPT_SACK_PERMITTED: 3571 if (optlen != TCPOLEN_SACK_PERMITTED) 3572 continue; 3573 if (!(flags & TO_SYN)) 3574 continue; 3575 if (!V_tcp_do_sack) 3576 continue; 3577 to->to_flags |= TOF_SACKPERM; 3578 break; 3579 case TCPOPT_SACK: 3580 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3581 continue; 3582 if (flags & TO_SYN) 3583 continue; 3584 to->to_flags |= TOF_SACK; 3585 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3586 to->to_sacks = cp + 2; 3587 TCPSTAT_INC(tcps_sack_rcv_blocks); 3588 break; 3589 case TCPOPT_FAST_OPEN: 3590 /* 3591 * Cookie length validation is performed by the 3592 * server side cookie checking code or the client 3593 * side cookie cache update code. 3594 */ 3595 if (!(flags & TO_SYN)) 3596 continue; 3597 if (!V_tcp_fastopen_client_enable && 3598 !V_tcp_fastopen_server_enable) 3599 continue; 3600 to->to_flags |= TOF_FASTOPEN; 3601 to->to_tfo_len = optlen - 2; 3602 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3603 break; 3604 default: 3605 continue; 3606 } 3607 } 3608 } 3609 3610 /* 3611 * Pull out of band byte out of a segment so 3612 * it doesn't appear in the user's data queue. 3613 * It is still reflected in the segment length for 3614 * sequencing purposes. 3615 */ 3616 void 3617 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3618 int off) 3619 { 3620 int cnt = off + th->th_urp - 1; 3621 3622 while (cnt >= 0) { 3623 if (m->m_len > cnt) { 3624 char *cp = mtod(m, caddr_t) + cnt; 3625 struct tcpcb *tp = sototcpcb(so); 3626 3627 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3628 3629 tp->t_iobc = *cp; 3630 tp->t_oobflags |= TCPOOB_HAVEDATA; 3631 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3632 m->m_len--; 3633 if (m->m_flags & M_PKTHDR) 3634 m->m_pkthdr.len--; 3635 return; 3636 } 3637 cnt -= m->m_len; 3638 m = m->m_next; 3639 if (m == NULL) 3640 break; 3641 } 3642 panic("tcp_pulloutofband"); 3643 } 3644 3645 /* 3646 * Collect new round-trip time estimate 3647 * and update averages and current timeout. 3648 */ 3649 void 3650 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3651 { 3652 int delta; 3653 3654 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3655 3656 TCPSTAT_INC(tcps_rttupdated); 3657 if (tp->t_rttupdated < UCHAR_MAX) 3658 tp->t_rttupdated++; 3659 #ifdef STATS 3660 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3661 imax(0, rtt * 1000 / hz)); 3662 #endif 3663 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3664 /* 3665 * srtt is stored as fixed point with 5 bits after the 3666 * binary point (i.e., scaled by 8). The following magic 3667 * is equivalent to the smoothing algorithm in rfc793 with 3668 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3669 * point). Adjust rtt to origin 0. 3670 */ 3671 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3672 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3673 3674 if ((tp->t_srtt += delta) <= 0) 3675 tp->t_srtt = 1; 3676 3677 /* 3678 * We accumulate a smoothed rtt variance (actually, a 3679 * smoothed mean difference), then set the retransmit 3680 * timer to smoothed rtt + 4 times the smoothed variance. 3681 * rttvar is stored as fixed point with 4 bits after the 3682 * binary point (scaled by 16). The following is 3683 * equivalent to rfc793 smoothing with an alpha of .75 3684 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3685 * rfc793's wired-in beta. 3686 */ 3687 if (delta < 0) 3688 delta = -delta; 3689 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3690 if ((tp->t_rttvar += delta) <= 0) 3691 tp->t_rttvar = 1; 3692 } else { 3693 /* 3694 * No rtt measurement yet - use the unsmoothed rtt. 3695 * Set the variance to half the rtt (so our first 3696 * retransmit happens at 3*rtt). 3697 */ 3698 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3699 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3700 } 3701 tp->t_rtttime = 0; 3702 tp->t_rxtshift = 0; 3703 3704 /* 3705 * the retransmit should happen at rtt + 4 * rttvar. 3706 * Because of the way we do the smoothing, srtt and rttvar 3707 * will each average +1/2 tick of bias. When we compute 3708 * the retransmit timer, we want 1/2 tick of rounding and 3709 * 1 extra tick because of +-1/2 tick uncertainty in the 3710 * firing of the timer. The bias will give us exactly the 3711 * 1.5 tick we need. But, because the bias is 3712 * statistical, we have to test that we don't drop below 3713 * the minimum feasible timer (which is 2 ticks). 3714 */ 3715 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3716 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3717 3718 /* 3719 * We received an ack for a packet that wasn't retransmitted; 3720 * it is probably safe to discard any error indications we've 3721 * received recently. This isn't quite right, but close enough 3722 * for now (a route might have failed after we sent a segment, 3723 * and the return path might not be symmetrical). 3724 */ 3725 tp->t_softerror = 0; 3726 } 3727 3728 /* 3729 * Determine a reasonable value for maxseg size. 3730 * If the route is known, check route for mtu. 3731 * If none, use an mss that can be handled on the outgoing interface 3732 * without forcing IP to fragment. If no route is found, route has no mtu, 3733 * or the destination isn't local, use a default, hopefully conservative 3734 * size (usually 512 or the default IP max size, but no more than the mtu 3735 * of the interface), as we can't discover anything about intervening 3736 * gateways or networks. We also initialize the congestion/slow start 3737 * window to be a single segment if the destination isn't local. 3738 * While looking at the routing entry, we also initialize other path-dependent 3739 * parameters from pre-set or cached values in the routing entry. 3740 * 3741 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3742 * IP options, e.g. IPSEC data, since length of this data may vary, and 3743 * thus it is calculated for every segment separately in tcp_output(). 3744 * 3745 * NOTE that this routine is only called when we process an incoming 3746 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3747 * settings are handled in tcp_mssopt(). 3748 */ 3749 void 3750 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3751 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3752 { 3753 int mss = 0; 3754 uint32_t maxmtu = 0; 3755 struct inpcb *inp = tptoinpcb(tp); 3756 struct hc_metrics_lite metrics; 3757 #ifdef INET6 3758 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3759 size_t min_protoh = isipv6 ? 3760 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3761 sizeof (struct tcpiphdr); 3762 #else 3763 size_t min_protoh = sizeof(struct tcpiphdr); 3764 #endif 3765 3766 INP_WLOCK_ASSERT(inp); 3767 3768 if (tp->t_port) 3769 min_protoh += V_tcp_udp_tunneling_overhead; 3770 if (mtuoffer != -1) { 3771 KASSERT(offer == -1, ("%s: conflict", __func__)); 3772 offer = mtuoffer - min_protoh; 3773 } 3774 3775 /* Initialize. */ 3776 #ifdef INET6 3777 if (isipv6) { 3778 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3779 tp->t_maxseg = V_tcp_v6mssdflt; 3780 } 3781 #endif 3782 #if defined(INET) && defined(INET6) 3783 else 3784 #endif 3785 #ifdef INET 3786 { 3787 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3788 tp->t_maxseg = V_tcp_mssdflt; 3789 } 3790 #endif 3791 3792 /* 3793 * No route to sender, stay with default mss and return. 3794 */ 3795 if (maxmtu == 0) { 3796 /* 3797 * In case we return early we need to initialize metrics 3798 * to a defined state as tcp_hc_get() would do for us 3799 * if there was no cache hit. 3800 */ 3801 if (metricptr != NULL) 3802 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3803 return; 3804 } 3805 3806 /* What have we got? */ 3807 switch (offer) { 3808 case 0: 3809 /* 3810 * Offer == 0 means that there was no MSS on the SYN 3811 * segment, in this case we use tcp_mssdflt as 3812 * already assigned to t_maxseg above. 3813 */ 3814 offer = tp->t_maxseg; 3815 break; 3816 3817 case -1: 3818 /* 3819 * Offer == -1 means that we didn't receive SYN yet. 3820 */ 3821 /* FALLTHROUGH */ 3822 3823 default: 3824 /* 3825 * Prevent DoS attack with too small MSS. Round up 3826 * to at least minmss. 3827 */ 3828 offer = max(offer, V_tcp_minmss); 3829 } 3830 3831 /* 3832 * rmx information is now retrieved from tcp_hostcache. 3833 */ 3834 tcp_hc_get(&inp->inp_inc, &metrics); 3835 if (metricptr != NULL) 3836 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3837 3838 /* 3839 * If there's a discovered mtu in tcp hostcache, use it. 3840 * Else, use the link mtu. 3841 */ 3842 if (metrics.rmx_mtu) 3843 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3844 else { 3845 #ifdef INET6 3846 if (isipv6) { 3847 mss = maxmtu - min_protoh; 3848 if (!V_path_mtu_discovery && 3849 !in6_localaddr(&inp->in6p_faddr)) 3850 mss = min(mss, V_tcp_v6mssdflt); 3851 } 3852 #endif 3853 #if defined(INET) && defined(INET6) 3854 else 3855 #endif 3856 #ifdef INET 3857 { 3858 mss = maxmtu - min_protoh; 3859 if (!V_path_mtu_discovery && 3860 !in_localaddr(inp->inp_faddr)) 3861 mss = min(mss, V_tcp_mssdflt); 3862 } 3863 #endif 3864 /* 3865 * XXX - The above conditional (mss = maxmtu - min_protoh) 3866 * probably violates the TCP spec. 3867 * The problem is that, since we don't know the 3868 * other end's MSS, we are supposed to use a conservative 3869 * default. But, if we do that, then MTU discovery will 3870 * never actually take place, because the conservative 3871 * default is much less than the MTUs typically seen 3872 * on the Internet today. For the moment, we'll sweep 3873 * this under the carpet. 3874 * 3875 * The conservative default might not actually be a problem 3876 * if the only case this occurs is when sending an initial 3877 * SYN with options and data to a host we've never talked 3878 * to before. Then, they will reply with an MSS value which 3879 * will get recorded and the new parameters should get 3880 * recomputed. For Further Study. 3881 */ 3882 } 3883 mss = min(mss, offer); 3884 3885 /* 3886 * Sanity check: make sure that maxseg will be large 3887 * enough to allow some data on segments even if the 3888 * all the option space is used (40bytes). Otherwise 3889 * funny things may happen in tcp_output. 3890 * 3891 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3892 */ 3893 mss = max(mss, 64); 3894 3895 tp->t_maxseg = mss; 3896 } 3897 3898 void 3899 tcp_mss(struct tcpcb *tp, int offer) 3900 { 3901 int mss; 3902 uint32_t bufsize; 3903 struct inpcb *inp = tptoinpcb(tp); 3904 struct socket *so; 3905 struct hc_metrics_lite metrics; 3906 struct tcp_ifcap cap; 3907 3908 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3909 3910 bzero(&cap, sizeof(cap)); 3911 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3912 3913 mss = tp->t_maxseg; 3914 3915 /* 3916 * If there's a pipesize, change the socket buffer to that size, 3917 * don't change if sb_hiwat is different than default (then it 3918 * has been changed on purpose with setsockopt). 3919 * Make the socket buffers an integral number of mss units; 3920 * if the mss is larger than the socket buffer, decrease the mss. 3921 */ 3922 so = inp->inp_socket; 3923 SOCKBUF_LOCK(&so->so_snd); 3924 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3925 bufsize = metrics.rmx_sendpipe; 3926 else 3927 bufsize = so->so_snd.sb_hiwat; 3928 if (bufsize < mss) 3929 mss = bufsize; 3930 else { 3931 bufsize = roundup(bufsize, mss); 3932 if (bufsize > sb_max) 3933 bufsize = sb_max; 3934 if (bufsize > so->so_snd.sb_hiwat) 3935 (void)sbreserve_locked(so, SO_SND, bufsize, NULL); 3936 } 3937 SOCKBUF_UNLOCK(&so->so_snd); 3938 /* 3939 * Sanity check: make sure that maxseg will be large 3940 * enough to allow some data on segments even if the 3941 * all the option space is used (40bytes). Otherwise 3942 * funny things may happen in tcp_output. 3943 * 3944 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3945 */ 3946 tp->t_maxseg = max(mss, 64); 3947 3948 SOCKBUF_LOCK(&so->so_rcv); 3949 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3950 bufsize = metrics.rmx_recvpipe; 3951 else 3952 bufsize = so->so_rcv.sb_hiwat; 3953 if (bufsize > mss) { 3954 bufsize = roundup(bufsize, mss); 3955 if (bufsize > sb_max) 3956 bufsize = sb_max; 3957 if (bufsize > so->so_rcv.sb_hiwat) 3958 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL); 3959 } 3960 SOCKBUF_UNLOCK(&so->so_rcv); 3961 3962 /* Check the interface for TSO capabilities. */ 3963 if (cap.ifcap & CSUM_TSO) { 3964 tp->t_flags |= TF_TSO; 3965 tp->t_tsomax = cap.tsomax; 3966 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3967 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3968 } 3969 } 3970 3971 /* 3972 * Determine the MSS option to send on an outgoing SYN. 3973 */ 3974 int 3975 tcp_mssopt(struct in_conninfo *inc) 3976 { 3977 int mss = 0; 3978 uint32_t thcmtu = 0; 3979 uint32_t maxmtu = 0; 3980 size_t min_protoh; 3981 3982 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3983 3984 #ifdef INET6 3985 if (inc->inc_flags & INC_ISIPV6) { 3986 mss = V_tcp_v6mssdflt; 3987 maxmtu = tcp_maxmtu6(inc, NULL); 3988 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3989 } 3990 #endif 3991 #if defined(INET) && defined(INET6) 3992 else 3993 #endif 3994 #ifdef INET 3995 { 3996 mss = V_tcp_mssdflt; 3997 maxmtu = tcp_maxmtu(inc, NULL); 3998 min_protoh = sizeof(struct tcpiphdr); 3999 } 4000 #endif 4001 #if defined(INET6) || defined(INET) 4002 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 4003 #endif 4004 4005 if (maxmtu && thcmtu) 4006 mss = min(maxmtu, thcmtu) - min_protoh; 4007 else if (maxmtu || thcmtu) 4008 mss = max(maxmtu, thcmtu) - min_protoh; 4009 4010 return (mss); 4011 } 4012 4013 void 4014 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, sackstatus_t sack_changed) 4015 { 4016 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 4017 int maxseg = tcp_maxseg(tp); 4018 4019 INP_WLOCK_ASSERT(tptoinpcb(tp)); 4020 4021 /* 4022 * Compute the amount of data that this ACK is indicating 4023 * (del_data) and an estimate of how many bytes are in the 4024 * network. 4025 */ 4026 if (tcp_is_sack_recovery(tp, to) || 4027 (IN_CONGRECOVERY(tp->t_flags) && 4028 !IN_FASTRECOVERY(tp->t_flags))) { 4029 del_data = tp->sackhint.delivered_data; 4030 if (V_tcp_do_newsack) 4031 pipe = tcp_compute_pipe(tp); 4032 else 4033 pipe = (tp->snd_nxt - tp->snd_fack) + 4034 tp->sackhint.sack_bytes_rexmit; 4035 } else { 4036 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + 4037 tp->snd_recover - tp->snd_una)) 4038 del_data = maxseg; 4039 pipe = imax(0, tp->snd_max - tp->snd_una - 4040 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); 4041 } 4042 tp->sackhint.prr_delivered += del_data; 4043 /* 4044 * Proportional Rate Reduction 4045 */ 4046 if (pipe >= tp->snd_ssthresh) { 4047 if (tp->sackhint.recover_fs == 0) 4048 tp->sackhint.recover_fs = 4049 imax(1, tp->snd_nxt - tp->snd_una); 4050 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 4051 tp->snd_ssthresh, tp->sackhint.recover_fs) - 4052 tp->sackhint.prr_out + maxseg - 1; 4053 } else { 4054 /* 4055 * PRR 6937bis heuristic: 4056 * - A partial ack without SACK block beneath snd_recover 4057 * indicates further loss. 4058 * - An SACK scoreboard update adding a new hole indicates 4059 * further loss, so be conservative and send at most one 4060 * segment. 4061 * - Prevent ACK splitting attacks, by being conservative 4062 * when no new data is acked. 4063 */ 4064 if ((sack_changed == SACK_NEWLOSS) || (del_data == 0)) 4065 limit = tp->sackhint.prr_delivered - 4066 tp->sackhint.prr_out; 4067 else 4068 limit = imax(tp->sackhint.prr_delivered - 4069 tp->sackhint.prr_out, del_data) + 4070 maxseg; 4071 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 4072 } 4073 snd_cnt = imax(snd_cnt, 0) / maxseg; 4074 /* 4075 * Send snd_cnt new data into the network in response to this ack. 4076 * If there is going to be a SACK retransmission, adjust snd_cwnd 4077 * accordingly. 4078 */ 4079 if (IN_FASTRECOVERY(tp->t_flags)) { 4080 if (tcp_is_sack_recovery(tp, to)) { 4081 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + 4082 tp->sackhint.sack_bytes_rexmit + 4083 (snd_cnt * maxseg); 4084 } else { 4085 tp->snd_cwnd = (tp->snd_max - tp->snd_una) + 4086 (snd_cnt * maxseg); 4087 } 4088 } else if (IN_CONGRECOVERY(tp->t_flags)) 4089 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4090 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); 4091 } 4092 4093 /* 4094 * On a partial ack arrives, force the retransmission of the 4095 * next unacknowledged segment. Do not clear tp->t_dupacks. 4096 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4097 * be started again. 4098 */ 4099 void 4100 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4101 { 4102 tcp_seq onxt = tp->snd_nxt; 4103 uint32_t ocwnd = tp->snd_cwnd; 4104 u_int maxseg = tcp_maxseg(tp); 4105 4106 INP_WLOCK_ASSERT(tptoinpcb(tp)); 4107 4108 tcp_timer_activate(tp, TT_REXMT, 0); 4109 tp->t_rtttime = 0; 4110 tp->snd_nxt = th->th_ack; 4111 /* 4112 * Set snd_cwnd to one segment beyond acknowledged offset. 4113 * (tp->snd_una has not yet been updated when this function is called.) 4114 */ 4115 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4116 tp->t_flags |= TF_ACKNOW; 4117 (void) tcp_output(tp); 4118 tp->snd_cwnd = ocwnd; 4119 if (SEQ_GT(onxt, tp->snd_nxt)) 4120 tp->snd_nxt = onxt; 4121 /* 4122 * Partial window deflation. Relies on fact that tp->snd_una 4123 * not updated yet. 4124 */ 4125 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4126 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4127 else 4128 tp->snd_cwnd = 0; 4129 tp->snd_cwnd += maxseg; 4130 } 4131 4132 int 4133 tcp_compute_pipe(struct tcpcb *tp) 4134 { 4135 if (tp->t_fb->tfb_compute_pipe == NULL) { 4136 return (tp->snd_max - tp->snd_una + 4137 tp->sackhint.sack_bytes_rexmit - 4138 tp->sackhint.sacked_bytes); 4139 } else { 4140 return((*tp->t_fb->tfb_compute_pipe)(tp)); 4141 } 4142 } 4143 4144 uint32_t 4145 tcp_compute_initwnd(uint32_t maxseg) 4146 { 4147 /* 4148 * Calculate the Initial Window, also used as Restart Window 4149 * 4150 * RFC5681 Section 3.1 specifies the default conservative values. 4151 * RFC3390 specifies slightly more aggressive values. 4152 * RFC6928 increases it to ten segments. 4153 * Support for user specified value for initial flight size. 4154 */ 4155 if (V_tcp_initcwnd_segments) 4156 return min(V_tcp_initcwnd_segments * maxseg, 4157 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4158 else if (V_tcp_do_rfc3390) 4159 return min(4 * maxseg, max(2 * maxseg, 4380)); 4160 else { 4161 /* Per RFC5681 Section 3.1 */ 4162 if (maxseg > 2190) 4163 return (2 * maxseg); 4164 else if (maxseg > 1095) 4165 return (3 * maxseg); 4166 else 4167 return (4 * maxseg); 4168 } 4169 } 4170