1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 5 * The Regents of the University of California. All rights reserved. 6 * Copyright (c) 2007-2008,2010 7 * Swinburne University of Technology, Melbourne, Australia. 8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 9 * Copyright (c) 2010 The FreeBSD Foundation 10 * Copyright (c) 2010-2011 Juniper Networks, Inc. 11 * All rights reserved. 12 * 13 * Portions of this software were developed at the Centre for Advanced Internet 14 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 15 * James Healy and David Hayes, made possible in part by a grant from the Cisco 16 * University Research Program Fund at Community Foundation Silicon Valley. 17 * 18 * Portions of this software were developed at the Centre for Advanced 19 * Internet Architectures, Swinburne University of Technology, Melbourne, 20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 21 * 22 * Portions of this software were developed by Robert N. M. Watson under 23 * contract to Juniper Networks, Inc. 24 * 25 * Redistribution and use in source and binary forms, with or without 26 * modification, are permitted provided that the following conditions 27 * are met: 28 * 1. Redistributions of source code must retain the above copyright 29 * notice, this list of conditions and the following disclaimer. 30 * 2. Redistributions in binary form must reproduce the above copyright 31 * notice, this list of conditions and the following disclaimer in the 32 * documentation and/or other materials provided with the distribution. 33 * 3. Neither the name of the University nor the names of its contributors 34 * may be used to endorse or promote products derived from this software 35 * without specific prior written permission. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 */ 49 50 #include <sys/cdefs.h> 51 #include "opt_inet.h" 52 #include "opt_inet6.h" 53 #include "opt_ipsec.h" 54 #include "opt_rss.h" 55 56 #include <sys/param.h> 57 #include <sys/arb.h> 58 #include <sys/kernel.h> 59 #ifdef TCP_HHOOK 60 #include <sys/hhook.h> 61 #endif 62 #include <sys/malloc.h> 63 #include <sys/mbuf.h> 64 #include <sys/proc.h> /* for proc0 declaration */ 65 #include <sys/protosw.h> 66 #include <sys/qmath.h> 67 #include <sys/sdt.h> 68 #include <sys/signalvar.h> 69 #include <sys/socket.h> 70 #include <sys/socketvar.h> 71 #include <sys/sysctl.h> 72 #include <sys/syslog.h> 73 #include <sys/systm.h> 74 #include <sys/stats.h> 75 76 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 77 78 #include <vm/uma.h> 79 80 #include <net/if.h> 81 #include <net/if_var.h> 82 #include <net/route.h> 83 #include <net/rss_config.h> 84 #include <net/vnet.h> 85 86 #define TCPSTATES /* for logging */ 87 88 #include <netinet/in.h> 89 #include <netinet/in_kdtrace.h> 90 #include <netinet/in_pcb.h> 91 #include <netinet/in_rss.h> 92 #include <netinet/in_systm.h> 93 #include <netinet/ip.h> 94 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 95 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 96 #include <netinet/ip_var.h> 97 #include <netinet/ip_options.h> 98 #include <netinet/ip6.h> 99 #include <netinet/icmp6.h> 100 #include <netinet6/in6_pcb.h> 101 #include <netinet6/in6_rss.h> 102 #include <netinet6/in6_var.h> 103 #include <netinet6/ip6_var.h> 104 #include <netinet6/nd6.h> 105 #include <netinet/tcp.h> 106 #include <netinet/tcp_fsm.h> 107 #include <netinet/tcp_seq.h> 108 #include <netinet/tcp_timer.h> 109 #include <netinet/tcp_var.h> 110 #include <netinet/tcp_log_buf.h> 111 #include <netinet6/tcp6_var.h> 112 #include <netinet/tcpip.h> 113 #include <netinet/cc/cc.h> 114 #include <netinet/tcp_fastopen.h> 115 #ifdef TCPPCAP 116 #include <netinet/tcp_pcap.h> 117 #endif 118 #include <netinet/tcp_syncache.h> 119 #ifdef TCP_OFFLOAD 120 #include <netinet/tcp_offload.h> 121 #endif 122 #include <netinet/tcp_ecn.h> 123 #include <netinet/udp.h> 124 125 #include <netipsec/ipsec_support.h> 126 127 #include <machine/in_cksum.h> 128 129 #include <security/mac/mac_framework.h> 130 131 const int tcprexmtthresh = 3; 132 133 VNET_DEFINE(int, tcp_log_in_vain) = 0; 134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 135 &VNET_NAME(tcp_log_in_vain), 0, 136 "Log all incoming TCP segments to closed ports"); 137 138 VNET_DEFINE(int, blackhole) = 0; 139 #define V_blackhole VNET(blackhole) 140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 141 &VNET_NAME(blackhole), 0, 142 "Do not send RST on segments to closed ports"); 143 144 VNET_DEFINE(bool, blackhole_local) = false; 145 #define V_blackhole_local VNET(blackhole_local) 146 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET | 147 CTLFLAG_RW, &VNET_NAME(blackhole_local), false, 148 "Enforce net.inet.tcp.blackhole for locally originated packets"); 149 150 VNET_DEFINE(int, tcp_delack_enabled) = 1; 151 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 152 &VNET_NAME(tcp_delack_enabled), 0, 153 "Delay ACK to try and piggyback it onto a data packet"); 154 155 VNET_DEFINE(int, drop_synfin) = 0; 156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 157 &VNET_NAME(drop_synfin), 0, 158 "Drop TCP packets with SYN+FIN set"); 159 160 VNET_DEFINE(int, tcp_do_prr) = 1; 161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 162 &VNET_NAME(tcp_do_prr), 1, 163 "Enable Proportional Rate Reduction per RFC 6937"); 164 165 VNET_DEFINE(int, tcp_do_newcwv) = 0; 166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 167 &VNET_NAME(tcp_do_newcwv), 0, 168 "Enable New Congestion Window Validation per RFC7661"); 169 170 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 172 &VNET_NAME(tcp_do_rfc3042), 0, 173 "Enable RFC 3042 (Limited Transmit)"); 174 175 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 177 &VNET_NAME(tcp_do_rfc3390), 0, 178 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 179 180 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 181 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 182 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 183 "Slow-start flight size (initial congestion window) in number of segments"); 184 185 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 186 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 187 &VNET_NAME(tcp_do_rfc3465), 0, 188 "Enable RFC 3465 (Appropriate Byte Counting)"); 189 190 VNET_DEFINE(int, tcp_abc_l_var) = 2; 191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 192 &VNET_NAME(tcp_abc_l_var), 2, 193 "Cap the max cwnd increment during slow-start to this number of segments"); 194 195 VNET_DEFINE(int, tcp_insecure_syn) = 0; 196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 197 &VNET_NAME(tcp_insecure_syn), 0, 198 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 199 200 VNET_DEFINE(int, tcp_insecure_rst) = 0; 201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 202 &VNET_NAME(tcp_insecure_rst), 0, 203 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 204 205 VNET_DEFINE(int, tcp_insecure_ack) = 0; 206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_ack, CTLFLAG_VNET | CTLFLAG_RW, 207 &VNET_NAME(tcp_insecure_ack), 0, 208 "Follow RFC793 criteria for validating SEG.ACK"); 209 210 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 211 #define V_tcp_recvspace VNET(tcp_recvspace) 212 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 213 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 214 215 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 217 &VNET_NAME(tcp_do_autorcvbuf), 0, 218 "Enable automatic receive buffer sizing"); 219 220 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(tcp_autorcvbuf_max), 0, 223 "Max size of automatic receive buffer"); 224 225 VNET_DEFINE(struct inpcbinfo, tcbinfo); 226 227 /* 228 * TCP statistics are stored in an array of counter(9)s, which size matches 229 * size of struct tcpstat. TCP running connection count is a regular array. 230 */ 231 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 232 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 233 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 234 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 235 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 236 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 237 "TCP connection counts by TCP state"); 238 239 /* 240 * Kernel module interface for updating tcpstat. The first argument is an index 241 * into tcpstat treated as an array. 242 */ 243 void 244 kmod_tcpstat_add(int statnum, int val) 245 { 246 247 counter_u64_add(VNET(tcpstat)[statnum], val); 248 } 249 250 /* 251 * Make sure that we only start a SACK loss recovery when 252 * receiving a duplicate ACK with a SACK block, and also 253 * complete SACK loss recovery in case the other end 254 * reneges. 255 */ 256 static bool inline 257 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to) 258 { 259 return ((tp->t_flags & TF_SACK_PERMIT) && 260 ((to->to_flags & TOF_SACK) || 261 (!TAILQ_EMPTY(&tp->snd_holes)))); 262 } 263 264 #ifdef TCP_HHOOK 265 /* 266 * Wrapper for the TCP established input helper hook. 267 */ 268 void 269 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 270 { 271 struct tcp_hhook_data hhook_data; 272 273 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 274 hhook_data.tp = tp; 275 hhook_data.th = th; 276 hhook_data.to = to; 277 278 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 279 &tp->t_osd); 280 } 281 } 282 #endif 283 284 /* 285 * CC wrapper hook functions 286 */ 287 void 288 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 289 uint16_t type) 290 { 291 #ifdef STATS 292 int32_t gput; 293 #endif 294 295 INP_WLOCK_ASSERT(tptoinpcb(tp)); 296 297 tp->t_ccv.nsegs = nsegs; 298 tp->t_ccv.bytes_this_ack = BYTES_THIS_ACK(tp, th); 299 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 300 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 301 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 302 tp->t_ccv.flags |= CCF_CWND_LIMITED; 303 else 304 tp->t_ccv.flags &= ~CCF_CWND_LIMITED; 305 306 if (type == CC_ACK) { 307 #ifdef STATS 308 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 309 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 310 if (!IN_RECOVERY(tp->t_flags)) 311 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 312 tp->t_ccv.bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 313 if ((tp->t_flags & TF_GPUTINPROG) && 314 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 315 /* 316 * Compute goodput in bits per millisecond. 317 */ 318 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) / 319 max(1, tcp_ts_getticks() - tp->gput_ts); 320 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 321 gput); 322 /* 323 * XXXLAS: This is a temporary hack, and should be 324 * chained off VOI_TCP_GPUT when stats(9) grows an API 325 * to deal with chained VOIs. 326 */ 327 if (tp->t_stats_gput_prev > 0) 328 stats_voi_update_abs_s32(tp->t_stats, 329 VOI_TCP_GPUT_ND, 330 ((gput - tp->t_stats_gput_prev) * 100) / 331 tp->t_stats_gput_prev); 332 tp->t_flags &= ~TF_GPUTINPROG; 333 tp->t_stats_gput_prev = gput; 334 } 335 #endif /* STATS */ 336 if (tp->snd_cwnd > tp->snd_ssthresh) { 337 tp->t_bytes_acked += tp->t_ccv.bytes_this_ack; 338 if (tp->t_bytes_acked >= tp->snd_cwnd) { 339 tp->t_bytes_acked -= tp->snd_cwnd; 340 tp->t_ccv.flags |= CCF_ABC_SENTAWND; 341 } 342 } else { 343 tp->t_ccv.flags &= ~CCF_ABC_SENTAWND; 344 tp->t_bytes_acked = 0; 345 } 346 } 347 348 if (CC_ALGO(tp)->ack_received != NULL) { 349 /* XXXLAS: Find a way to live without this */ 350 tp->t_ccv.curack = th->th_ack; 351 CC_ALGO(tp)->ack_received(&tp->t_ccv, type); 352 } 353 #ifdef STATS 354 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 355 #endif 356 } 357 358 void 359 cc_conn_init(struct tcpcb *tp) 360 { 361 struct hc_metrics_lite metrics; 362 struct inpcb *inp = tptoinpcb(tp); 363 u_int maxseg; 364 int rtt; 365 366 INP_WLOCK_ASSERT(inp); 367 368 tcp_hc_get(&inp->inp_inc, &metrics); 369 maxseg = tcp_maxseg(tp); 370 371 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 372 tp->t_srtt = rtt; 373 TCPSTAT_INC(tcps_usedrtt); 374 if (metrics.rmx_rttvar) { 375 tp->t_rttvar = metrics.rmx_rttvar; 376 TCPSTAT_INC(tcps_usedrttvar); 377 } else { 378 /* default variation is +- 1 rtt */ 379 tp->t_rttvar = 380 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 381 } 382 TCPT_RANGESET(tp->t_rxtcur, 383 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 384 tp->t_rttmin, TCPTV_REXMTMAX); 385 } 386 if (metrics.rmx_ssthresh) { 387 /* 388 * There's some sort of gateway or interface 389 * buffer limit on the path. Use this to set 390 * the slow start threshold, but set the 391 * threshold to no less than 2*mss. 392 */ 393 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 394 TCPSTAT_INC(tcps_usedssthresh); 395 } 396 397 /* 398 * Set the initial slow-start flight size. 399 * 400 * If a SYN or SYN/ACK was lost and retransmitted, we have to 401 * reduce the initial CWND to one segment as congestion is likely 402 * requiring us to be cautious. 403 */ 404 if (tp->snd_cwnd == 1) 405 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 406 else 407 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 408 409 if (CC_ALGO(tp)->conn_init != NULL) 410 CC_ALGO(tp)->conn_init(&tp->t_ccv); 411 } 412 413 void inline 414 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 415 { 416 INP_WLOCK_ASSERT(tptoinpcb(tp)); 417 418 #ifdef STATS 419 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 420 #endif 421 422 switch(type) { 423 case CC_NDUPACK: 424 if (!IN_FASTRECOVERY(tp->t_flags)) { 425 tp->snd_recover = tp->snd_max; 426 if (tp->t_flags2 & TF2_ECN_PERMIT) 427 tp->t_flags2 |= TF2_ECN_SND_CWR; 428 } 429 break; 430 case CC_ECN: 431 if (!IN_CONGRECOVERY(tp->t_flags) || 432 /* 433 * Allow ECN reaction on ACK to CWR, if 434 * that data segment was also CE marked. 435 */ 436 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 437 EXIT_CONGRECOVERY(tp->t_flags); 438 TCPSTAT_INC(tcps_ecn_rcwnd); 439 tp->snd_recover = tp->snd_max + 1; 440 if (tp->t_flags2 & TF2_ECN_PERMIT) 441 tp->t_flags2 |= TF2_ECN_SND_CWR; 442 } 443 break; 444 case CC_RTO: 445 tp->t_dupacks = 0; 446 tp->t_bytes_acked = 0; 447 EXIT_RECOVERY(tp->t_flags); 448 if (tp->t_flags2 & TF2_ECN_PERMIT) 449 tp->t_flags2 |= TF2_ECN_SND_CWR; 450 break; 451 case CC_RTO_ERR: 452 TCPSTAT_INC(tcps_sndrexmitbad); 453 /* RTO was unnecessary, so reset everything. */ 454 tp->snd_cwnd = tp->snd_cwnd_prev; 455 tp->snd_ssthresh = tp->snd_ssthresh_prev; 456 tp->snd_recover = tp->snd_recover_prev; 457 if (tp->t_flags & TF_WASFRECOVERY) 458 ENTER_FASTRECOVERY(tp->t_flags); 459 if (tp->t_flags & TF_WASCRECOVERY) 460 ENTER_CONGRECOVERY(tp->t_flags); 461 tp->snd_nxt = tp->snd_max; 462 tp->t_flags &= ~TF_PREVVALID; 463 tp->t_badrxtwin = 0; 464 break; 465 } 466 if (SEQ_LT(tp->snd_fack, tp->snd_una) || 467 SEQ_GT(tp->snd_fack, tp->snd_max)) { 468 tp->snd_fack = tp->snd_una; 469 } 470 471 if (CC_ALGO(tp)->cong_signal != NULL) { 472 if (th != NULL) 473 tp->t_ccv.curack = th->th_ack; 474 CC_ALGO(tp)->cong_signal(&tp->t_ccv, type); 475 } 476 } 477 478 void inline 479 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 480 { 481 INP_WLOCK_ASSERT(tptoinpcb(tp)); 482 483 if (CC_ALGO(tp)->post_recovery != NULL) { 484 if (SEQ_LT(tp->snd_fack, th->th_ack) || 485 SEQ_GT(tp->snd_fack, tp->snd_max)) { 486 tp->snd_fack = th->th_ack; 487 } 488 tp->t_ccv.curack = th->th_ack; 489 CC_ALGO(tp)->post_recovery(&tp->t_ccv); 490 } 491 EXIT_RECOVERY(tp->t_flags); 492 493 tp->t_bytes_acked = 0; 494 tp->sackhint.delivered_data = 0; 495 tp->sackhint.prr_delivered = 0; 496 tp->sackhint.prr_out = 0; 497 } 498 499 /* 500 * Indicate whether this ack should be delayed. We can delay the ack if 501 * following conditions are met: 502 * - There is no delayed ack timer in progress. 503 * - Our last ack wasn't a 0-sized window. We never want to delay 504 * the ack that opens up a 0-sized window. 505 * - LRO wasn't used for this segment. We make sure by checking that the 506 * segment size is not larger than the MSS. 507 */ 508 #define DELAY_ACK(tp, tlen) \ 509 ((!tcp_timer_active(tp, TT_DELACK) && \ 510 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 511 (tlen <= tp->t_maxseg) && \ 512 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 513 514 void inline 515 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) 516 { 517 INP_WLOCK_ASSERT(tptoinpcb(tp)); 518 519 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 520 switch (iptos & IPTOS_ECN_MASK) { 521 case IPTOS_ECN_CE: 522 tp->t_ccv.flags |= CCF_IPHDR_CE; 523 break; 524 case IPTOS_ECN_ECT0: 525 /* FALLTHROUGH */ 526 case IPTOS_ECN_ECT1: 527 /* FALLTHROUGH */ 528 case IPTOS_ECN_NOTECT: 529 tp->t_ccv.flags &= ~CCF_IPHDR_CE; 530 break; 531 } 532 533 if (flags & TH_CWR) 534 tp->t_ccv.flags |= CCF_TCPHDR_CWR; 535 else 536 tp->t_ccv.flags &= ~CCF_TCPHDR_CWR; 537 538 CC_ALGO(tp)->ecnpkt_handler(&tp->t_ccv); 539 540 if (tp->t_ccv.flags & CCF_ACKNOW) { 541 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 542 tp->t_flags |= TF_ACKNOW; 543 } 544 } 545 } 546 547 void inline 548 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 549 { 550 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos); 551 } 552 553 /* 554 * TCP input handling is split into multiple parts: 555 * tcp6_input is a thin wrapper around tcp_input for the extended 556 * ip6_protox[] call format in ip6_input 557 * tcp_input handles primary segment validation, inpcb lookup and 558 * SYN processing on listen sockets 559 * tcp_do_segment processes the ACK and text of the segment for 560 * establishing, established and closing connections 561 */ 562 #ifdef INET6 563 int 564 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 565 { 566 struct mbuf *m; 567 struct in6_ifaddr *ia6; 568 struct ip6_hdr *ip6; 569 570 m = *mp; 571 if (m->m_len < *offp + sizeof(struct tcphdr)) { 572 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 573 if (m == NULL) { 574 *mp = m; 575 TCPSTAT_INC(tcps_rcvshort); 576 return (IPPROTO_DONE); 577 } 578 } 579 580 /* 581 * draft-itojun-ipv6-tcp-to-anycast 582 * better place to put this in? 583 */ 584 ip6 = mtod(m, struct ip6_hdr *); 585 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 586 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 587 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 588 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 589 *mp = NULL; 590 return (IPPROTO_DONE); 591 } 592 593 *mp = m; 594 return (tcp_input_with_port(mp, offp, proto, port)); 595 } 596 597 int 598 tcp6_input(struct mbuf **mp, int *offp, int proto) 599 { 600 601 return(tcp6_input_with_port(mp, offp, proto, 0)); 602 } 603 #endif /* INET6 */ 604 605 int 606 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 607 { 608 struct mbuf *m = *mp; 609 struct tcphdr *th = NULL; 610 struct ip *ip = NULL; 611 struct inpcb *inp = NULL; 612 struct tcpcb *tp = NULL; 613 struct socket *so = NULL; 614 u_char *optp = NULL; 615 int off0; 616 int optlen = 0; 617 #ifdef INET 618 int len; 619 uint8_t ipttl; 620 #endif 621 int tlen = 0, off; 622 int drop_hdrlen; 623 int thflags; 624 int rstreason = 0; /* For badport_bandlim accounting purposes */ 625 int lookupflag; 626 uint8_t iptos; 627 struct m_tag *fwd_tag = NULL; 628 #ifdef INET6 629 struct ip6_hdr *ip6 = NULL; 630 int isipv6; 631 #else 632 const void *ip6 = NULL; 633 #endif /* INET6 */ 634 struct tcpopt to; /* options in this segment */ 635 char *s = NULL; /* address and port logging */ 636 637 NET_EPOCH_ASSERT(); 638 639 #ifdef INET6 640 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 641 #endif 642 643 off0 = *offp; 644 m = *mp; 645 *mp = NULL; 646 to.to_flags = 0; 647 TCPSTAT_INC(tcps_rcvtotal); 648 649 m->m_pkthdr.tcp_tun_port = port; 650 #ifdef INET6 651 if (isipv6) { 652 ip6 = mtod(m, struct ip6_hdr *); 653 th = (struct tcphdr *)((caddr_t)ip6 + off0); 654 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 655 if (port) 656 goto skip6_csum; 657 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 658 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 659 th->th_sum = m->m_pkthdr.csum_data; 660 else 661 th->th_sum = in6_cksum_pseudo(ip6, tlen, 662 IPPROTO_TCP, m->m_pkthdr.csum_data); 663 th->th_sum ^= 0xffff; 664 } else 665 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 666 if (th->th_sum) { 667 TCPSTAT_INC(tcps_rcvbadsum); 668 goto drop; 669 } 670 skip6_csum: 671 /* 672 * Be proactive about unspecified IPv6 address in source. 673 * As we use all-zero to indicate unbounded/unconnected pcb, 674 * unspecified IPv6 address can be used to confuse us. 675 * 676 * Note that packets with unspecified IPv6 destination is 677 * already dropped in ip6_input. 678 */ 679 KASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_dst), 680 ("%s: unspecified destination v6 address", __func__)); 681 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 682 IP6STAT_INC(ip6s_badscope); /* XXX */ 683 goto drop; 684 } 685 iptos = IPV6_TRAFFIC_CLASS(ip6); 686 } 687 #endif 688 #if defined(INET) && defined(INET6) 689 else 690 #endif 691 #ifdef INET 692 { 693 /* 694 * Get IP and TCP header together in first mbuf. 695 * Note: IP leaves IP header in first mbuf. 696 */ 697 if (off0 > sizeof (struct ip)) { 698 ip_stripoptions(m); 699 off0 = sizeof(struct ip); 700 } 701 if (m->m_len < sizeof (struct tcpiphdr)) { 702 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 703 == NULL) { 704 TCPSTAT_INC(tcps_rcvshort); 705 return (IPPROTO_DONE); 706 } 707 } 708 ip = mtod(m, struct ip *); 709 th = (struct tcphdr *)((caddr_t)ip + off0); 710 tlen = ntohs(ip->ip_len) - off0; 711 712 iptos = ip->ip_tos; 713 if (port) 714 goto skip_csum; 715 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 716 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 717 th->th_sum = m->m_pkthdr.csum_data; 718 else 719 th->th_sum = in_pseudo(ip->ip_src.s_addr, 720 ip->ip_dst.s_addr, 721 htonl(m->m_pkthdr.csum_data + tlen + 722 IPPROTO_TCP)); 723 th->th_sum ^= 0xffff; 724 } else { 725 struct ipovly *ipov = (struct ipovly *)ip; 726 727 /* 728 * Checksum extended TCP header and data. 729 */ 730 len = off0 + tlen; 731 ipttl = ip->ip_ttl; 732 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 733 ipov->ih_len = htons(tlen); 734 th->th_sum = in_cksum(m, len); 735 /* Reset length for SDT probes. */ 736 ip->ip_len = htons(len); 737 /* Reset TOS bits */ 738 ip->ip_tos = iptos; 739 /* Re-initialization for later version check */ 740 ip->ip_ttl = ipttl; 741 ip->ip_v = IPVERSION; 742 ip->ip_hl = off0 >> 2; 743 } 744 skip_csum: 745 if (th->th_sum && (port == 0)) { 746 TCPSTAT_INC(tcps_rcvbadsum); 747 goto drop; 748 } 749 KASSERT(ip->ip_dst.s_addr != INADDR_ANY, 750 ("%s: unspecified destination v4 address", __func__)); 751 if (__predict_false(ip->ip_src.s_addr == INADDR_ANY)) { 752 IPSTAT_INC(ips_badaddr); 753 goto drop; 754 } 755 } 756 #endif /* INET */ 757 758 /* 759 * Check that TCP offset makes sense, 760 * pull out TCP options and adjust length. XXX 761 */ 762 off = th->th_off << 2; 763 if (off < sizeof (struct tcphdr) || off > tlen) { 764 TCPSTAT_INC(tcps_rcvbadoff); 765 goto drop; 766 } 767 tlen -= off; /* tlen is used instead of ti->ti_len */ 768 if (off > sizeof (struct tcphdr)) { 769 #ifdef INET6 770 if (isipv6) { 771 if (m->m_len < off0 + off) { 772 m = m_pullup(m, off0 + off); 773 if (m == NULL) { 774 TCPSTAT_INC(tcps_rcvshort); 775 return (IPPROTO_DONE); 776 } 777 } 778 ip6 = mtod(m, struct ip6_hdr *); 779 th = (struct tcphdr *)((caddr_t)ip6 + off0); 780 } 781 #endif 782 #if defined(INET) && defined(INET6) 783 else 784 #endif 785 #ifdef INET 786 { 787 if (m->m_len < sizeof(struct ip) + off) { 788 if ((m = m_pullup(m, sizeof (struct ip) + off)) 789 == NULL) { 790 TCPSTAT_INC(tcps_rcvshort); 791 return (IPPROTO_DONE); 792 } 793 ip = mtod(m, struct ip *); 794 th = (struct tcphdr *)((caddr_t)ip + off0); 795 } 796 } 797 #endif 798 optlen = off - sizeof (struct tcphdr); 799 optp = (u_char *)(th + 1); 800 } 801 thflags = tcp_get_flags(th); 802 803 /* 804 * Convert TCP protocol specific fields to host format. 805 */ 806 tcp_fields_to_host(th); 807 808 /* 809 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 810 */ 811 drop_hdrlen = off0 + off; 812 813 /* 814 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 815 */ 816 if ( 817 #ifdef INET6 818 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 819 #ifdef INET 820 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 821 #endif 822 #endif 823 #if defined(INET) && !defined(INET6) 824 (m->m_flags & M_IP_NEXTHOP) 825 #endif 826 ) 827 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 828 829 /* 830 * For initial SYN packets we don't need write lock on matching 831 * PCB, be it a listening one or a synchronized one. The packet 832 * shall not modify its state. 833 */ 834 lookupflag = INPLOOKUP_WILDCARD | 835 ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ? 836 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB); 837 findpcb: 838 tp = NULL; 839 #ifdef INET6 840 if (isipv6 && fwd_tag != NULL) { 841 struct sockaddr_in6 *next_hop6; 842 843 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 844 /* 845 * Transparently forwarded. Pretend to be the destination. 846 * Already got one like this? 847 */ 848 inp = in6_pcblookup_mbuf(&V_tcbinfo, 849 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 850 lookupflag & ~INPLOOKUP_WILDCARD, m->m_pkthdr.rcvif, m); 851 if (!inp) { 852 /* 853 * It's new. Try to find the ambushing socket. 854 * Because we've rewritten the destination address, 855 * any hardware-generated hash is ignored. 856 */ 857 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 858 th->th_sport, &next_hop6->sin6_addr, 859 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 860 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 861 } 862 } else if (isipv6) { 863 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 864 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag, 865 m->m_pkthdr.rcvif, m); 866 } 867 #endif /* INET6 */ 868 #if defined(INET6) && defined(INET) 869 else 870 #endif 871 #ifdef INET 872 if (fwd_tag != NULL) { 873 struct sockaddr_in *next_hop; 874 875 next_hop = (struct sockaddr_in *)(fwd_tag+1); 876 /* 877 * Transparently forwarded. Pretend to be the destination. 878 * already got one like this? 879 */ 880 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 881 ip->ip_dst, th->th_dport, lookupflag & ~INPLOOKUP_WILDCARD, 882 m->m_pkthdr.rcvif, m); 883 if (!inp) { 884 /* 885 * It's new. Try to find the ambushing socket. 886 * Because we've rewritten the destination address, 887 * any hardware-generated hash is ignored. 888 */ 889 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 890 th->th_sport, next_hop->sin_addr, 891 next_hop->sin_port ? ntohs(next_hop->sin_port) : 892 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 893 } 894 } else 895 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 896 th->th_sport, ip->ip_dst, th->th_dport, lookupflag, 897 m->m_pkthdr.rcvif, m); 898 #endif /* INET */ 899 900 /* 901 * If the INPCB does not exist then all data in the incoming 902 * segment is discarded and an appropriate RST is sent back. 903 * XXX MRT Send RST using which routing table? 904 */ 905 if (inp == NULL) { 906 if (rstreason != 0) { 907 /* We came here after second (safety) lookup. */ 908 MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0); 909 goto dropwithreset; 910 } 911 /* 912 * Log communication attempts to ports that are not 913 * in use. 914 */ 915 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 916 V_tcp_log_in_vain == 2) { 917 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 918 log(LOG_INFO, "%s; %s: Connection attempt " 919 "to closed port\n", s, __func__); 920 } 921 rstreason = BANDLIM_RST_CLOSEDPORT; 922 goto dropwithreset; 923 } 924 INP_LOCK_ASSERT(inp); 925 926 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 927 !SOLISTENING(inp->inp_socket)) { 928 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) { 929 inp->inp_flowid = m->m_pkthdr.flowid; 930 inp->inp_flowtype = M_HASHTYPE_GET(m); 931 #ifdef RSS 932 } else { 933 /* assign flowid by software RSS hash */ 934 #ifdef INET6 935 if (isipv6) { 936 rss_proto_software_hash_v6(&inp->in6p_faddr, 937 &inp->in6p_laddr, 938 inp->inp_fport, 939 inp->inp_lport, 940 IPPROTO_TCP, 941 &inp->inp_flowid, 942 &inp->inp_flowtype); 943 } else 944 #endif /* INET6 */ 945 { 946 rss_proto_software_hash_v4(inp->inp_faddr, 947 inp->inp_laddr, 948 inp->inp_fport, 949 inp->inp_lport, 950 IPPROTO_TCP, 951 &inp->inp_flowid, 952 &inp->inp_flowtype); 953 } 954 #endif /* RSS */ 955 } 956 } 957 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 958 #ifdef INET6 959 if (isipv6 && IPSEC_ENABLED(ipv6) && 960 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 961 goto dropunlock; 962 } 963 #ifdef INET 964 else 965 #endif 966 #endif /* INET6 */ 967 #ifdef INET 968 if (IPSEC_ENABLED(ipv4) && 969 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 970 goto dropunlock; 971 } 972 #endif /* INET */ 973 #endif /* IPSEC */ 974 975 /* 976 * Check the minimum TTL for socket. 977 */ 978 if (inp->inp_ip_minttl != 0) { 979 #ifdef INET6 980 if (isipv6) { 981 if (inp->inp_ip_minttl > ip6->ip6_hlim) 982 goto dropunlock; 983 } else 984 #endif 985 if (inp->inp_ip_minttl > ip->ip_ttl) 986 goto dropunlock; 987 } 988 989 tp = intotcpcb(inp); 990 switch (tp->t_state) { 991 case TCPS_TIME_WAIT: 992 /* 993 * A previous connection in TIMEWAIT state is supposed to catch 994 * stray or duplicate segments arriving late. If this segment 995 * was a legitimate new connection attempt, the old INPCB gets 996 * removed and we can try again to find a listening socket. 997 */ 998 tcp_dooptions(&to, optp, optlen, 999 (thflags & TH_SYN) ? TO_SYN : 0); 1000 /* 1001 * tcp_twcheck unlocks the inp always, and frees the m if fails. 1002 */ 1003 if (tcp_twcheck(inp, &to, th, m, tlen)) 1004 goto findpcb; 1005 return (IPPROTO_DONE); 1006 case TCPS_CLOSED: 1007 /* 1008 * The TCPCB may no longer exist if the connection is winding 1009 * down or it is in the CLOSED state. Either way we drop the 1010 * segment and send an appropriate response. 1011 */ 1012 rstreason = BANDLIM_RST_CLOSEDPORT; 1013 goto dropwithreset; 1014 } 1015 1016 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 1017 rstreason = BANDLIM_RST_CLOSEDPORT; 1018 goto dropwithreset; 1019 } 1020 1021 #ifdef TCP_OFFLOAD 1022 if (tp->t_flags & TF_TOE) { 1023 tcp_offload_input(tp, m); 1024 m = NULL; /* consumed by the TOE driver */ 1025 goto dropunlock; 1026 } 1027 #endif 1028 1029 #ifdef MAC 1030 if (mac_inpcb_check_deliver(inp, m)) 1031 goto dropunlock; 1032 #endif 1033 so = inp->inp_socket; 1034 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1035 /* 1036 * When the socket is accepting connections (the INPCB is in LISTEN 1037 * state) we look into the SYN cache if this is a new connection 1038 * attempt or the completion of a previous one. 1039 */ 1040 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), 1041 ("%s: so accepting but tp %p not listening", __func__, tp)); 1042 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { 1043 struct in_conninfo inc; 1044 1045 bzero(&inc, sizeof(inc)); 1046 #ifdef INET6 1047 if (isipv6) { 1048 inc.inc_flags |= INC_ISIPV6; 1049 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1050 inc.inc_flags |= INC_IPV6MINMTU; 1051 inc.inc6_faddr = ip6->ip6_src; 1052 inc.inc6_laddr = ip6->ip6_dst; 1053 } else 1054 #endif 1055 { 1056 inc.inc_faddr = ip->ip_src; 1057 inc.inc_laddr = ip->ip_dst; 1058 } 1059 inc.inc_fport = th->th_sport; 1060 inc.inc_lport = th->th_dport; 1061 inc.inc_fibnum = so->so_fibnum; 1062 1063 /* 1064 * Check for an existing connection attempt in syncache if 1065 * the flag is only ACK. A successful lookup creates a new 1066 * socket appended to the listen queue in SYN_RECEIVED state. 1067 */ 1068 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1069 /* 1070 * Parse the TCP options here because 1071 * syncookies need access to the reflected 1072 * timestamp. 1073 */ 1074 tcp_dooptions(&to, optp, optlen, 0); 1075 /* 1076 * NB: syncache_expand() doesn't unlock inp. 1077 */ 1078 rstreason = syncache_expand(&inc, &to, th, &so, m, port); 1079 if (rstreason < 0) { 1080 /* 1081 * A failing TCP MD5 signature comparison 1082 * must result in the segment being dropped 1083 * and must not produce any response back 1084 * to the sender. 1085 */ 1086 goto dropunlock; 1087 } else if (rstreason == 0) { 1088 /* 1089 * No syncache entry, or ACK was not for our 1090 * SYN/ACK. Do our protection against double 1091 * ACK. If peer sent us 2 ACKs, then for the 1092 * first one syncache_expand() successfully 1093 * converted syncache entry into a socket, 1094 * while we were waiting on the inpcb lock. We 1095 * don't want to sent RST for the second ACK, 1096 * so we perform second lookup without wildcard 1097 * match, hoping to find the new socket. If 1098 * the ACK is stray indeed, rstreason would 1099 * hint the above code that the lookup was a 1100 * second attempt. 1101 * 1102 * NB: syncache did its own logging 1103 * of the failure cause. 1104 */ 1105 INP_WUNLOCK(inp); 1106 rstreason = BANDLIM_RST_OPENPORT; 1107 lookupflag &= ~INPLOOKUP_WILDCARD; 1108 goto findpcb; 1109 } 1110 tfo_socket_result: 1111 if (so == NULL) { 1112 /* 1113 * We completed the 3-way handshake 1114 * but could not allocate a socket 1115 * either due to memory shortage, 1116 * listen queue length limits or 1117 * global socket limits. Send RST 1118 * or wait and have the remote end 1119 * retransmit the ACK for another 1120 * try. 1121 */ 1122 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1123 log(LOG_DEBUG, "%s; %s: Listen socket: " 1124 "Socket allocation failed due to " 1125 "limits or memory shortage, %s\n", 1126 s, __func__, 1127 V_tcp_sc_rst_sock_fail ? 1128 "sending RST" : "try again"); 1129 if (V_tcp_sc_rst_sock_fail) { 1130 rstreason = BANDLIM_UNLIMITED; 1131 goto dropwithreset; 1132 } else 1133 goto dropunlock; 1134 } 1135 /* 1136 * Socket is created in state SYN_RECEIVED. 1137 * Unlock the listen socket, lock the newly 1138 * created socket and update the tp variable. 1139 * If we came here via jump to tfo_socket_result, 1140 * then listening socket is read-locked. 1141 */ 1142 INP_UNLOCK(inp); /* listen socket */ 1143 inp = sotoinpcb(so); 1144 /* 1145 * New connection inpcb is already locked by 1146 * syncache_expand(). 1147 */ 1148 INP_WLOCK_ASSERT(inp); 1149 tp = intotcpcb(inp); 1150 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1151 ("%s: ", __func__)); 1152 /* 1153 * Process the segment and the data it 1154 * contains. tcp_do_segment() consumes 1155 * the mbuf chain and unlocks the inpcb. 1156 */ 1157 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1158 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, 1159 tlen, iptos); 1160 return (IPPROTO_DONE); 1161 } 1162 /* 1163 * Segment flag validation for new connection attempts: 1164 * 1165 * Our (SYN|ACK) response was rejected. 1166 * Check with syncache and remove entry to prevent 1167 * retransmits. 1168 * 1169 * NB: syncache_chkrst does its own logging of failure 1170 * causes. 1171 */ 1172 if (thflags & TH_RST) { 1173 syncache_chkrst(&inc, th, m, port); 1174 goto dropunlock; 1175 } 1176 /* 1177 * We can't do anything without SYN. 1178 */ 1179 if ((thflags & TH_SYN) == 0) { 1180 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1181 log(LOG_DEBUG, "%s; %s: Listen socket: " 1182 "SYN is missing, segment ignored\n", 1183 s, __func__); 1184 TCPSTAT_INC(tcps_badsyn); 1185 goto dropunlock; 1186 } 1187 /* 1188 * (SYN|ACK) is bogus on a listen socket. 1189 */ 1190 if (thflags & TH_ACK) { 1191 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1192 log(LOG_DEBUG, "%s; %s: Listen socket: " 1193 "SYN|ACK invalid, segment rejected\n", 1194 s, __func__); 1195 syncache_badack(&inc, port); /* XXX: Not needed! */ 1196 TCPSTAT_INC(tcps_badsyn); 1197 rstreason = BANDLIM_RST_OPENPORT; 1198 goto dropwithreset; 1199 } 1200 /* 1201 * If the drop_synfin option is enabled, drop all 1202 * segments with both the SYN and FIN bits set. 1203 * This prevents e.g. nmap from identifying the 1204 * TCP/IP stack. 1205 * XXX: Poor reasoning. nmap has other methods 1206 * and is constantly refining its stack detection 1207 * strategies. 1208 * XXX: This is a violation of the TCP specification 1209 * and was used by RFC1644. 1210 */ 1211 if ((thflags & TH_FIN) && V_drop_synfin) { 1212 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1213 log(LOG_DEBUG, "%s; %s: Listen socket: " 1214 "SYN|FIN segment ignored (based on " 1215 "sysctl setting)\n", s, __func__); 1216 TCPSTAT_INC(tcps_badsyn); 1217 goto dropunlock; 1218 } 1219 /* 1220 * Segment's flags are (SYN) or (SYN|FIN). 1221 * 1222 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1223 * as they do not affect the state of the TCP FSM. 1224 * The data pointed to by TH_URG and th_urp is ignored. 1225 */ 1226 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1227 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1228 KASSERT(thflags & (TH_SYN), 1229 ("%s: Listen socket: TH_SYN not set", __func__)); 1230 INP_RLOCK_ASSERT(inp); 1231 #ifdef INET6 1232 /* 1233 * If deprecated address is forbidden, 1234 * we do not accept SYN to deprecated interface 1235 * address to prevent any new inbound connection from 1236 * getting established. 1237 * When we do not accept SYN, we send a TCP RST, 1238 * with deprecated source address (instead of dropping 1239 * it). We compromise it as it is much better for peer 1240 * to send a RST, and RST will be the final packet 1241 * for the exchange. 1242 * 1243 * If we do not forbid deprecated addresses, we accept 1244 * the SYN packet. RFC2462 does not suggest dropping 1245 * SYN in this case. 1246 * If we decipher RFC2462 5.5.4, it says like this: 1247 * 1. use of deprecated addr with existing 1248 * communication is okay - "SHOULD continue to be 1249 * used" 1250 * 2. use of it with new communication: 1251 * (2a) "SHOULD NOT be used if alternate address 1252 * with sufficient scope is available" 1253 * (2b) nothing mentioned otherwise. 1254 * Here we fall into (2b) case as we have no choice in 1255 * our source address selection - we must obey the peer. 1256 * 1257 * The wording in RFC2462 is confusing, and there are 1258 * multiple description text for deprecated address 1259 * handling - worse, they are not exactly the same. 1260 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1261 */ 1262 if (isipv6 && !V_ip6_use_deprecated) { 1263 struct in6_ifaddr *ia6; 1264 1265 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1266 if (ia6 != NULL && 1267 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1268 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1269 log(LOG_DEBUG, "%s; %s: Listen socket: " 1270 "Connection attempt to deprecated " 1271 "IPv6 address rejected\n", 1272 s, __func__); 1273 rstreason = BANDLIM_RST_OPENPORT; 1274 goto dropwithreset; 1275 } 1276 } 1277 #endif /* INET6 */ 1278 /* 1279 * Basic sanity checks on incoming SYN requests: 1280 * Don't respond if the destination is a link layer 1281 * broadcast according to RFC1122 4.2.3.10, p. 104. 1282 * If it is from this socket it must be forged. 1283 * Don't respond if the source or destination is a 1284 * global or subnet broad- or multicast address. 1285 * Note that it is quite possible to receive unicast 1286 * link-layer packets with a broadcast IP address. Use 1287 * in_broadcast() to find them. 1288 */ 1289 if (m->m_flags & (M_BCAST|M_MCAST)) { 1290 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1291 log(LOG_DEBUG, "%s; %s: Listen socket: " 1292 "Connection attempt from broad- or multicast " 1293 "link layer address ignored\n", s, __func__); 1294 goto dropunlock; 1295 } 1296 #ifdef INET6 1297 if (isipv6) { 1298 if (th->th_dport == th->th_sport && 1299 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1300 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1301 log(LOG_DEBUG, "%s; %s: Listen socket: " 1302 "Connection attempt to/from self " 1303 "ignored\n", s, __func__); 1304 goto dropunlock; 1305 } 1306 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1307 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1308 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1309 log(LOG_DEBUG, "%s; %s: Listen socket: " 1310 "Connection attempt from/to multicast " 1311 "address ignored\n", s, __func__); 1312 goto dropunlock; 1313 } 1314 } 1315 #endif 1316 #if defined(INET) && defined(INET6) 1317 else 1318 #endif 1319 #ifdef INET 1320 { 1321 if (th->th_dport == th->th_sport && 1322 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1323 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1324 log(LOG_DEBUG, "%s; %s: Listen socket: " 1325 "Connection attempt from/to self " 1326 "ignored\n", s, __func__); 1327 goto dropunlock; 1328 } 1329 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1330 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1331 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1332 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1333 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1334 log(LOG_DEBUG, "%s; %s: Listen socket: " 1335 "Connection attempt from/to broad- " 1336 "or multicast address ignored\n", 1337 s, __func__); 1338 goto dropunlock; 1339 } 1340 } 1341 #endif 1342 /* 1343 * SYN appears to be valid. Create compressed TCP state 1344 * for syncache. 1345 */ 1346 TCP_PROBE3(debug__input, tp, th, m); 1347 tcp_dooptions(&to, optp, optlen, TO_SYN); 1348 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, 1349 iptos, port)) != NULL) 1350 goto tfo_socket_result; 1351 1352 /* 1353 * Entry added to syncache and mbuf consumed. 1354 * Only the listen socket is unlocked by syncache_add(). 1355 */ 1356 return (IPPROTO_DONE); 1357 } else if (tp->t_state == TCPS_LISTEN) { 1358 /* 1359 * When a listen socket is torn down the SO_ACCEPTCONN 1360 * flag is removed first while connections are drained 1361 * from the accept queue in a unlock/lock cycle of the 1362 * ACCEPT_LOCK, opening a race condition allowing a SYN 1363 * attempt go through unhandled. 1364 */ 1365 goto dropunlock; 1366 } 1367 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1368 if (tp->t_flags & TF_SIGNATURE) { 1369 tcp_dooptions(&to, optp, optlen, thflags); 1370 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1371 TCPSTAT_INC(tcps_sig_err_nosigopt); 1372 goto dropunlock; 1373 } 1374 if (!TCPMD5_ENABLED() || 1375 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1376 goto dropunlock; 1377 } 1378 #endif 1379 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1380 1381 /* 1382 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1383 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1384 * the inpcb, and unlocks pcbinfo. 1385 * 1386 * XXXGL: in case of a pure SYN arriving on existing connection 1387 * TCP stacks won't need to modify the PCB, they would either drop 1388 * the segment silently, or send a challenge ACK. However, we try 1389 * to upgrade the lock, because calling convention for stacks is 1390 * write-lock on PCB. If upgrade fails, drop the SYN. 1391 */ 1392 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0) 1393 goto dropunlock; 1394 1395 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, tlen, iptos); 1396 return (IPPROTO_DONE); 1397 1398 dropwithreset: 1399 /* 1400 * When blackholing do not respond with a RST but 1401 * completely ignore the segment and drop it. 1402 */ 1403 if (((rstreason == BANDLIM_RST_OPENPORT && V_blackhole == 3) || 1404 (rstreason == BANDLIM_RST_CLOSEDPORT && 1405 ((V_blackhole == 1 && (thflags & TH_SYN)) || V_blackhole > 1))) && 1406 (V_blackhole_local || ( 1407 #ifdef INET6 1408 isipv6 ? !in6_localaddr(&ip6->ip6_src) : 1409 #endif 1410 #ifdef INET 1411 !in_localip(ip->ip_src) 1412 #else 1413 true 1414 #endif 1415 ))) 1416 goto dropunlock; 1417 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1418 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1419 m = NULL; /* mbuf chain got consumed. */ 1420 1421 dropunlock: 1422 if (m != NULL) 1423 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1424 1425 if (inp != NULL) 1426 INP_UNLOCK(inp); 1427 1428 drop: 1429 if (s != NULL) 1430 free(s, M_TCPLOG); 1431 if (m != NULL) 1432 m_freem(m); 1433 return (IPPROTO_DONE); 1434 } 1435 1436 /* 1437 * Automatic sizing of receive socket buffer. Often the send 1438 * buffer size is not optimally adjusted to the actual network 1439 * conditions at hand (delay bandwidth product). Setting the 1440 * buffer size too small limits throughput on links with high 1441 * bandwidth and high delay (eg. trans-continental/oceanic links). 1442 * 1443 * On the receive side the socket buffer memory is only rarely 1444 * used to any significant extent. This allows us to be much 1445 * more aggressive in scaling the receive socket buffer. For 1446 * the case that the buffer space is actually used to a large 1447 * extent and we run out of kernel memory we can simply drop 1448 * the new segments; TCP on the sender will just retransmit it 1449 * later. Setting the buffer size too big may only consume too 1450 * much kernel memory if the application doesn't read() from 1451 * the socket or packet loss or reordering makes use of the 1452 * reassembly queue. 1453 * 1454 * The criteria to step up the receive buffer one notch are: 1455 * 1. Application has not set receive buffer size with 1456 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1457 * 2. the number of bytes received during 1/2 of an sRTT 1458 * is at least 3/8 of the current socket buffer size. 1459 * 3. receive buffer size has not hit maximal automatic size; 1460 * 1461 * If all of the criteria are met we increaset the socket buffer 1462 * by a 1/2 (bounded by the max). This allows us to keep ahead 1463 * of slow-start but also makes it so our peer never gets limited 1464 * by our rwnd which we then open up causing a burst. 1465 * 1466 * This algorithm does two steps per RTT at most and only if 1467 * we receive a bulk stream w/o packet losses or reorderings. 1468 * Shrinking the buffer during idle times is not necessary as 1469 * it doesn't consume any memory when idle. 1470 * 1471 * TODO: Only step up if the application is actually serving 1472 * the buffer to better manage the socket buffer resources. 1473 */ 1474 int 1475 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1476 struct tcpcb *tp, int tlen) 1477 { 1478 int newsize = 0; 1479 1480 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1481 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1482 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1483 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1484 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1485 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1486 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1487 } 1488 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1489 1490 /* Start over with next RTT. */ 1491 tp->rfbuf_ts = 0; 1492 tp->rfbuf_cnt = 0; 1493 } else { 1494 tp->rfbuf_cnt += tlen; /* add up */ 1495 } 1496 return (newsize); 1497 } 1498 1499 int 1500 tcp_input(struct mbuf **mp, int *offp, int proto) 1501 { 1502 return(tcp_input_with_port(mp, offp, proto, 0)); 1503 } 1504 1505 static void 1506 tcp_handle_wakeup(struct tcpcb *tp) 1507 { 1508 1509 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1510 1511 if (tp->t_flags & TF_WAKESOR) { 1512 struct socket *so = tptosocket(tp); 1513 1514 tp->t_flags &= ~TF_WAKESOR; 1515 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1516 sorwakeup_locked(so); 1517 } 1518 } 1519 1520 void 1521 tcp_do_segment(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th, 1522 int drop_hdrlen, int tlen, uint8_t iptos) 1523 { 1524 uint16_t thflags; 1525 int acked, ourfinisacked, needoutput = 0; 1526 sackstatus_t sack_changed; 1527 int rstreason, todrop, win, incforsyn = 0; 1528 uint32_t tiwin; 1529 uint16_t nsegs; 1530 char *s; 1531 struct inpcb *inp = tptoinpcb(tp); 1532 struct socket *so = tptosocket(tp); 1533 struct in_conninfo *inc = &inp->inp_inc; 1534 struct mbuf *mfree; 1535 struct tcpopt to; 1536 int tfo_syn; 1537 u_int maxseg = 0; 1538 1539 thflags = tcp_get_flags(th); 1540 tp->sackhint.last_sack_ack = 0; 1541 sack_changed = SACK_NOCHANGE; 1542 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1543 1544 NET_EPOCH_ASSERT(); 1545 INP_WLOCK_ASSERT(inp); 1546 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1547 __func__)); 1548 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1549 __func__)); 1550 1551 #ifdef TCPPCAP 1552 /* Save segment, if requested. */ 1553 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1554 #endif 1555 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1556 tlen, NULL, true); 1557 1558 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1559 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1560 log(LOG_DEBUG, "%s; %s: " 1561 "SYN|FIN segment ignored (based on " 1562 "sysctl setting)\n", s, __func__); 1563 free(s, M_TCPLOG); 1564 } 1565 goto drop; 1566 } 1567 1568 /* 1569 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1570 * check SEQ.ACK first. 1571 */ 1572 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1573 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1574 rstreason = BANDLIM_UNLIMITED; 1575 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1576 goto dropwithreset; 1577 } 1578 1579 /* 1580 * Segment received on connection. 1581 * Reset idle time and keep-alive timer. 1582 * XXX: This should be done after segment 1583 * validation to ignore broken/spoofed segs. 1584 */ 1585 if (tp->t_idle_reduce && 1586 (tp->snd_max == tp->snd_una) && 1587 ((ticks - tp->t_rcvtime) >= tp->t_rxtcur)) 1588 cc_after_idle(tp); 1589 tp->t_rcvtime = ticks; 1590 1591 if (thflags & TH_FIN) 1592 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN); 1593 /* 1594 * Scale up the window into a 32-bit value. 1595 * For the SYN_SENT state the scale is zero. 1596 */ 1597 tiwin = th->th_win << tp->snd_scale; 1598 #ifdef STATS 1599 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1600 #endif 1601 1602 /* 1603 * TCP ECN processing. 1604 */ 1605 if (tcp_ecn_input_segment(tp, thflags, tlen, 1606 tcp_packets_this_ack(tp, th->th_ack), 1607 iptos)) 1608 cc_cong_signal(tp, th, CC_ECN); 1609 1610 /* 1611 * Parse options on any incoming segment. 1612 */ 1613 tcp_dooptions(&to, (u_char *)(th + 1), 1614 (th->th_off << 2) - sizeof(struct tcphdr), 1615 (thflags & TH_SYN) ? TO_SYN : 0); 1616 if (tp->t_flags2 & TF2_PROC_SACK_PROHIBIT) { 1617 /* 1618 * We don't look at sack's from the 1619 * peer because the MSS is too small which 1620 * can subject us to an attack. 1621 */ 1622 to.to_flags &= ~TOF_SACK; 1623 } 1624 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1625 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1626 (to.to_flags & TOF_SIGNATURE) == 0) { 1627 TCPSTAT_INC(tcps_sig_err_sigopt); 1628 /* XXX: should drop? */ 1629 } 1630 #endif 1631 /* 1632 * If echoed timestamp is later than the current time, 1633 * fall back to non RFC1323 RTT calculation. Normalize 1634 * timestamp if syncookies were used when this connection 1635 * was established. 1636 */ 1637 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1638 to.to_tsecr -= tp->ts_offset; 1639 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) { 1640 to.to_tsecr = 0; 1641 } else if (tp->t_rxtshift == 1 && 1642 tp->t_flags & TF_PREVVALID && 1643 tp->t_badrxtwin != 0 && 1644 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) { 1645 cc_cong_signal(tp, th, CC_RTO_ERR); 1646 } 1647 } 1648 /* 1649 * Process options only when we get SYN/ACK back. The SYN case 1650 * for incoming connections is handled in tcp_syncache. 1651 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1652 * or <SYN,ACK>) segment itself is never scaled. 1653 * XXX this is traditional behavior, may need to be cleaned up. 1654 */ 1655 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1656 /* Handle parallel SYN for ECN */ 1657 tcp_ecn_input_parallel_syn(tp, thflags, iptos); 1658 if ((to.to_flags & TOF_SCALE) && 1659 (tp->t_flags & TF_REQ_SCALE) && 1660 !(tp->t_flags & TF_NOOPT)) { 1661 tp->t_flags |= TF_RCVD_SCALE; 1662 tp->snd_scale = to.to_wscale; 1663 } else { 1664 tp->t_flags &= ~TF_REQ_SCALE; 1665 } 1666 /* 1667 * Initial send window. It will be updated with 1668 * the next incoming segment to the scaled value. 1669 */ 1670 tp->snd_wnd = th->th_win; 1671 if ((to.to_flags & TOF_TS) && 1672 (tp->t_flags & TF_REQ_TSTMP) && 1673 !(tp->t_flags & TF_NOOPT)) { 1674 tp->t_flags |= TF_RCVD_TSTMP; 1675 tp->ts_recent = to.to_tsval; 1676 tp->ts_recent_age = tcp_ts_getticks(); 1677 } else { 1678 tp->t_flags &= ~TF_REQ_TSTMP; 1679 } 1680 if (to.to_flags & TOF_MSS) { 1681 tcp_mss(tp, to.to_mss); 1682 } 1683 if ((tp->t_flags & TF_SACK_PERMIT) && 1684 (!(to.to_flags & TOF_SACKPERM) || 1685 (tp->t_flags & TF_NOOPT))) { 1686 tp->t_flags &= ~TF_SACK_PERMIT; 1687 } 1688 if (tp->t_flags & TF_FASTOPEN) { 1689 if ((to.to_flags & TOF_FASTOPEN) && 1690 !(tp->t_flags & TF_NOOPT)) { 1691 uint16_t mss; 1692 1693 if (to.to_flags & TOF_MSS) { 1694 mss = to.to_mss; 1695 } else { 1696 if ((inp->inp_vflag & INP_IPV6) != 0) { 1697 mss = TCP6_MSS; 1698 } else { 1699 mss = TCP_MSS; 1700 } 1701 } 1702 tcp_fastopen_update_cache(tp, mss, 1703 to.to_tfo_len, to.to_tfo_cookie); 1704 } else { 1705 tcp_fastopen_disable_path(tp); 1706 } 1707 } 1708 } 1709 1710 /* 1711 * If timestamps were negotiated during SYN/ACK and a 1712 * segment without a timestamp is received, silently drop 1713 * the segment, unless it is a RST segment or missing timestamps are 1714 * tolerated. 1715 * See section 3.2 of RFC 7323. 1716 */ 1717 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1718 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1719 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1720 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1721 "segment processed normally\n", 1722 s, __func__); 1723 free(s, M_TCPLOG); 1724 } 1725 } else { 1726 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1727 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1728 "segment silently dropped\n", s, __func__); 1729 free(s, M_TCPLOG); 1730 } 1731 goto drop; 1732 } 1733 } 1734 /* 1735 * If timestamps were not negotiated during SYN/ACK and a 1736 * segment with a timestamp is received, ignore the 1737 * timestamp and process the packet normally. 1738 * See section 3.2 of RFC 7323. 1739 */ 1740 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1741 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1742 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1743 "segment processed normally\n", s, __func__); 1744 free(s, M_TCPLOG); 1745 } 1746 } 1747 1748 /* 1749 * Header prediction: check for the two common cases 1750 * of a uni-directional data xfer. If the packet has 1751 * no control flags, is in-sequence, the window didn't 1752 * change and we're not retransmitting, it's a 1753 * candidate. If the length is zero and the ack moved 1754 * forward, we're the sender side of the xfer. Just 1755 * free the data acked & wake any higher level process 1756 * that was blocked waiting for space. If the length 1757 * is non-zero and the ack didn't move, we're the 1758 * receiver side. If we're getting packets in-order 1759 * (the reassembly queue is empty), add the data to 1760 * the socket buffer and note that we need a delayed ack. 1761 * Make sure that the hidden state-flags are also off. 1762 * Since we check for TCPS_ESTABLISHED first, it can only 1763 * be TH_NEEDSYN. 1764 */ 1765 if (tp->t_state == TCPS_ESTABLISHED && 1766 th->th_seq == tp->rcv_nxt && 1767 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1768 tp->snd_nxt == tp->snd_max && 1769 tiwin && tiwin == tp->snd_wnd && 1770 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1771 SEGQ_EMPTY(tp) && 1772 ((to.to_flags & TOF_TS) == 0 || 1773 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1774 /* 1775 * If last ACK falls within this segment's sequence numbers, 1776 * record the timestamp. 1777 * NOTE that the test is modified according to the latest 1778 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1779 */ 1780 if ((to.to_flags & TOF_TS) != 0 && 1781 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1782 tp->ts_recent_age = tcp_ts_getticks(); 1783 tp->ts_recent = to.to_tsval; 1784 } 1785 1786 if (tlen == 0) { 1787 if (SEQ_GT(th->th_ack, tp->snd_una) && 1788 SEQ_LEQ(th->th_ack, tp->snd_max) && 1789 !IN_RECOVERY(tp->t_flags) && 1790 (to.to_flags & TOF_SACK) == 0 && 1791 TAILQ_EMPTY(&tp->snd_holes)) { 1792 /* 1793 * This is a pure ack for outstanding data. 1794 */ 1795 TCPSTAT_INC(tcps_predack); 1796 1797 /* 1798 * "bad retransmit" recovery without timestamps. 1799 */ 1800 if ((to.to_flags & TOF_TS) == 0 && 1801 tp->t_rxtshift == 1 && 1802 tp->t_flags & TF_PREVVALID && 1803 tp->t_badrxtwin != 0 && 1804 TSTMP_LT(ticks, tp->t_badrxtwin)) { 1805 cc_cong_signal(tp, th, CC_RTO_ERR); 1806 } 1807 1808 /* 1809 * Recalculate the transmit timer / rtt. 1810 * 1811 * Some boxes send broken timestamp replies 1812 * during the SYN+ACK phase, ignore 1813 * timestamps of 0 or we could calculate a 1814 * huge RTT and blow up the retransmit timer. 1815 */ 1816 if ((to.to_flags & TOF_TS) != 0 && 1817 to.to_tsecr) { 1818 uint32_t t; 1819 1820 t = tcp_ts_getticks() - to.to_tsecr; 1821 if (!tp->t_rttlow || tp->t_rttlow > t) 1822 tp->t_rttlow = t; 1823 tcp_xmit_timer(tp, 1824 TCP_TS_TO_TICKS(t) + 1); 1825 } else if (tp->t_rtttime && 1826 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1827 if (!tp->t_rttlow || 1828 tp->t_rttlow > ticks - tp->t_rtttime) 1829 tp->t_rttlow = ticks - tp->t_rtttime; 1830 tcp_xmit_timer(tp, 1831 ticks - tp->t_rtttime); 1832 } 1833 acked = BYTES_THIS_ACK(tp, th); 1834 1835 #ifdef TCP_HHOOK 1836 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1837 hhook_run_tcp_est_in(tp, th, &to); 1838 #endif 1839 1840 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1841 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1842 sbdrop(&so->so_snd, acked); 1843 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1844 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1845 tp->snd_recover = th->th_ack - 1; 1846 1847 /* 1848 * Let the congestion control algorithm update 1849 * congestion control related information. This 1850 * typically means increasing the congestion 1851 * window. 1852 */ 1853 cc_ack_received(tp, th, nsegs, CC_ACK); 1854 1855 tp->snd_una = th->th_ack; 1856 /* 1857 * Pull snd_wl2 up to prevent seq wrap relative 1858 * to th_ack. 1859 */ 1860 tp->snd_wl2 = th->th_ack; 1861 tp->t_dupacks = 0; 1862 m_freem(m); 1863 1864 /* 1865 * If all outstanding data are acked, stop 1866 * retransmit timer, otherwise restart timer 1867 * using current (possibly backed-off) value. 1868 * If process is waiting for space, 1869 * wakeup/selwakeup/signal. If data 1870 * are ready to send, let tcp_output 1871 * decide between more output or persist. 1872 */ 1873 TCP_PROBE3(debug__input, tp, th, m); 1874 /* 1875 * Clear t_acktime if remote side has ACKd 1876 * all data in the socket buffer. 1877 * Otherwise, update t_acktime if we received 1878 * a sufficiently large ACK. 1879 */ 1880 if (sbavail(&so->so_snd) == 0) 1881 tp->t_acktime = 0; 1882 else if (acked > 1) 1883 tp->t_acktime = ticks; 1884 if (tp->snd_una == tp->snd_max) 1885 tcp_timer_activate(tp, TT_REXMT, 0); 1886 else if (!tcp_timer_active(tp, TT_PERSIST)) 1887 tcp_timer_activate(tp, TT_REXMT, 1888 TP_RXTCUR(tp)); 1889 sowwakeup(so); 1890 /* 1891 * Only call tcp_output when there 1892 * is new data available to be sent 1893 * or we need to send an ACK. 1894 */ 1895 if ((tp->t_flags & TF_ACKNOW) || 1896 (sbavail(&so->so_snd) >= 1897 SEQ_SUB(tp->snd_max, tp->snd_una))) { 1898 (void) tcp_output(tp); 1899 } 1900 goto check_delack; 1901 } 1902 } else if (th->th_ack == tp->snd_una && 1903 tlen <= sbspace(&so->so_rcv)) { 1904 int newsize = 0; /* automatic sockbuf scaling */ 1905 1906 /* 1907 * This is a pure, in-sequence data packet with 1908 * nothing on the reassembly queue and we have enough 1909 * buffer space to take it. 1910 */ 1911 /* Clean receiver SACK report if present */ 1912 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1913 tcp_clean_sackreport(tp); 1914 TCPSTAT_INC(tcps_preddat); 1915 tp->rcv_nxt += tlen; 1916 if (tlen && 1917 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1918 (tp->t_fbyte_in == 0)) { 1919 tp->t_fbyte_in = ticks; 1920 if (tp->t_fbyte_in == 0) 1921 tp->t_fbyte_in = 1; 1922 if (tp->t_fbyte_out && tp->t_fbyte_in) 1923 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1924 } 1925 /* 1926 * Pull snd_wl1 up to prevent seq wrap relative to 1927 * th_seq. 1928 */ 1929 tp->snd_wl1 = th->th_seq; 1930 /* 1931 * Pull rcv_up up to prevent seq wrap relative to 1932 * rcv_nxt. 1933 */ 1934 tp->rcv_up = tp->rcv_nxt; 1935 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1936 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1937 TCP_PROBE3(debug__input, tp, th, m); 1938 1939 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1940 1941 /* Add data to socket buffer. */ 1942 SOCKBUF_LOCK(&so->so_rcv); 1943 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1944 m_freem(m); 1945 } else { 1946 /* 1947 * Set new socket buffer size. 1948 * Give up when limit is reached. 1949 */ 1950 if (newsize) 1951 if (!sbreserve_locked(so, SO_RCV, 1952 newsize, NULL)) 1953 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1954 m_adj(m, drop_hdrlen); /* delayed header drop */ 1955 sbappendstream_locked(&so->so_rcv, m, 0); 1956 } 1957 /* NB: sorwakeup_locked() does an implicit unlock. */ 1958 sorwakeup_locked(so); 1959 if (DELAY_ACK(tp, tlen)) { 1960 tp->t_flags |= TF_DELACK; 1961 } else { 1962 tp->t_flags |= TF_ACKNOW; 1963 (void) tcp_output(tp); 1964 } 1965 goto check_delack; 1966 } 1967 } 1968 1969 /* 1970 * Calculate amount of space in receive window, 1971 * and then do TCP input processing. 1972 * Receive window is amount of space in rcv queue, 1973 * but not less than advertised window. 1974 */ 1975 win = sbspace(&so->so_rcv); 1976 if (win < 0) 1977 win = 0; 1978 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1979 1980 switch (tp->t_state) { 1981 /* 1982 * If the state is SYN_RECEIVED: 1983 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1984 */ 1985 case TCPS_SYN_RECEIVED: 1986 if (thflags & TH_RST) { 1987 /* Handle RST segments later. */ 1988 break; 1989 } 1990 if ((thflags & TH_ACK) && 1991 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1992 SEQ_GT(th->th_ack, tp->snd_max))) { 1993 rstreason = BANDLIM_RST_OPENPORT; 1994 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1995 goto dropwithreset; 1996 } 1997 if (tp->t_flags & TF_FASTOPEN) { 1998 /* 1999 * When a TFO connection is in SYN_RECEIVED, the 2000 * only valid packets are the initial SYN, a 2001 * retransmit/copy of the initial SYN (possibly with 2002 * a subset of the original data), a valid ACK, a 2003 * FIN, or a RST. 2004 */ 2005 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 2006 rstreason = BANDLIM_RST_OPENPORT; 2007 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2008 goto dropwithreset; 2009 } else if (thflags & TH_SYN) { 2010 /* non-initial SYN is ignored */ 2011 if ((tcp_timer_active(tp, TT_DELACK) || 2012 tcp_timer_active(tp, TT_REXMT))) 2013 goto drop; 2014 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 2015 goto drop; 2016 } 2017 } 2018 break; 2019 2020 /* 2021 * If the state is SYN_SENT: 2022 * if seg contains a RST with valid ACK (SEQ.ACK has already 2023 * been verified), then drop the connection. 2024 * if seg contains a RST without an ACK, drop the seg. 2025 * if seg does not contain SYN, then drop the seg. 2026 * Otherwise this is an acceptable SYN segment 2027 * initialize tp->rcv_nxt and tp->irs 2028 * if seg contains ack then advance tp->snd_una 2029 * if seg contains an ECE and ECN support is enabled, the stream 2030 * is ECN capable. 2031 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2032 * arrange for segment to be acked (eventually) 2033 * continue processing rest of data/controls, beginning with URG 2034 */ 2035 case TCPS_SYN_SENT: 2036 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2037 TCP_PROBE5(connect__refused, NULL, tp, 2038 m, tp, th); 2039 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2040 tp = tcp_drop(tp, ECONNREFUSED); 2041 } 2042 if (thflags & TH_RST) 2043 goto drop; 2044 if (!(thflags & TH_SYN)) 2045 goto drop; 2046 2047 tp->irs = th->th_seq; 2048 tcp_rcvseqinit(tp); 2049 if (thflags & TH_ACK) { 2050 int tfo_partial_ack = 0; 2051 2052 TCPSTAT_INC(tcps_connects); 2053 soisconnected(so); 2054 #ifdef MAC 2055 mac_socketpeer_set_from_mbuf(m, so); 2056 #endif 2057 /* Do window scaling on this connection? */ 2058 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2059 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2060 tp->rcv_scale = tp->request_r_scale; 2061 } 2062 tp->rcv_adv += min(tp->rcv_wnd, 2063 TCP_MAXWIN << tp->rcv_scale); 2064 tp->snd_una++; /* SYN is acked */ 2065 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2066 tp->snd_nxt = tp->snd_una; 2067 /* 2068 * If not all the data that was sent in the TFO SYN 2069 * has been acked, resend the remainder right away. 2070 */ 2071 if ((tp->t_flags & TF_FASTOPEN) && 2072 (tp->snd_una != tp->snd_max)) { 2073 tp->snd_nxt = th->th_ack; 2074 tfo_partial_ack = 1; 2075 } 2076 /* 2077 * If there's data, delay ACK; if there's also a FIN 2078 * ACKNOW will be turned on later. 2079 */ 2080 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2081 tcp_timer_activate(tp, TT_DELACK, 2082 tcp_delacktime); 2083 else 2084 tp->t_flags |= TF_ACKNOW; 2085 2086 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2087 2088 /* 2089 * Received <SYN,ACK> in SYN_SENT[*] state. 2090 * Transitions: 2091 * SYN_SENT --> ESTABLISHED 2092 * SYN_SENT* --> FIN_WAIT_1 2093 */ 2094 tp->t_starttime = ticks; 2095 if (tp->t_flags & TF_NEEDFIN) { 2096 tp->t_acktime = ticks; 2097 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2098 tp->t_flags &= ~TF_NEEDFIN; 2099 thflags &= ~TH_SYN; 2100 } else { 2101 tcp_state_change(tp, TCPS_ESTABLISHED); 2102 TCP_PROBE5(connect__established, NULL, tp, 2103 m, tp, th); 2104 cc_conn_init(tp); 2105 tcp_timer_activate(tp, TT_KEEP, 2106 TP_KEEPIDLE(tp)); 2107 } 2108 } else { 2109 /* 2110 * Received initial SYN in SYN-SENT[*] state => 2111 * simultaneous open. 2112 * If it succeeds, connection is * half-synchronized. 2113 * Otherwise, do 3-way handshake: 2114 * SYN-SENT -> SYN-RECEIVED 2115 * SYN-SENT* -> SYN-RECEIVED* 2116 */ 2117 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN); 2118 tcp_timer_activate(tp, TT_REXMT, 0); 2119 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2120 } 2121 2122 /* 2123 * Advance th->th_seq to correspond to first data byte. 2124 * If data, trim to stay within window, 2125 * dropping FIN if necessary. 2126 */ 2127 th->th_seq++; 2128 if (tlen > tp->rcv_wnd) { 2129 todrop = tlen - tp->rcv_wnd; 2130 m_adj(m, -todrop); 2131 tlen = tp->rcv_wnd; 2132 thflags &= ~TH_FIN; 2133 TCPSTAT_INC(tcps_rcvpackafterwin); 2134 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2135 } 2136 tp->snd_wl1 = th->th_seq - 1; 2137 tp->rcv_up = th->th_seq; 2138 /* 2139 * Client side of transaction: already sent SYN and data. 2140 * If the remote host used T/TCP to validate the SYN, 2141 * our data will be ACK'd; if so, enter normal data segment 2142 * processing in the middle of step 5, ack processing. 2143 * Otherwise, goto step 6. 2144 */ 2145 if (thflags & TH_ACK) 2146 goto process_ACK; 2147 2148 goto step6; 2149 } 2150 2151 /* 2152 * States other than LISTEN or SYN_SENT. 2153 * First check the RST flag and sequence number since reset segments 2154 * are exempt from the timestamp and connection count tests. This 2155 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2156 * below which allowed reset segments in half the sequence space 2157 * to fall though and be processed (which gives forged reset 2158 * segments with a random sequence number a 50 percent chance of 2159 * killing a connection). 2160 * Then check timestamp, if present. 2161 * Then check the connection count, if present. 2162 * Then check that at least some bytes of segment are within 2163 * receive window. If segment begins before rcv_nxt, 2164 * drop leading data (and SYN); if nothing left, just ack. 2165 */ 2166 if (thflags & TH_RST) { 2167 /* 2168 * RFC5961 Section 3.2 2169 * 2170 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2171 * - If RST is in window, we send challenge ACK. 2172 * 2173 * Note: to take into account delayed ACKs, we should 2174 * test against last_ack_sent instead of rcv_nxt. 2175 * Note 2: we handle special case of closed window, not 2176 * covered by the RFC. 2177 */ 2178 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2179 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2180 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2181 KASSERT(tp->t_state != TCPS_SYN_SENT, 2182 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2183 __func__, th, tp)); 2184 2185 if (V_tcp_insecure_rst || 2186 tp->last_ack_sent == th->th_seq) { 2187 TCPSTAT_INC(tcps_drops); 2188 /* Drop the connection. */ 2189 switch (tp->t_state) { 2190 case TCPS_SYN_RECEIVED: 2191 so->so_error = ECONNREFUSED; 2192 goto close; 2193 case TCPS_ESTABLISHED: 2194 case TCPS_FIN_WAIT_1: 2195 case TCPS_FIN_WAIT_2: 2196 case TCPS_CLOSE_WAIT: 2197 case TCPS_CLOSING: 2198 case TCPS_LAST_ACK: 2199 so->so_error = ECONNRESET; 2200 close: 2201 /* FALLTHROUGH */ 2202 default: 2203 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST); 2204 tp = tcp_close(tp); 2205 } 2206 } else { 2207 TCPSTAT_INC(tcps_badrst); 2208 tcp_send_challenge_ack(tp, th, m); 2209 m = NULL; 2210 } 2211 } 2212 goto drop; 2213 } 2214 2215 /* 2216 * RFC5961 Section 4.2 2217 * Send challenge ACK for any SYN in synchronized state. 2218 */ 2219 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2220 tp->t_state != TCPS_SYN_RECEIVED) { 2221 TCPSTAT_INC(tcps_badsyn); 2222 if (V_tcp_insecure_syn && 2223 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2224 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2225 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2226 tp = tcp_drop(tp, ECONNRESET); 2227 rstreason = BANDLIM_UNLIMITED; 2228 } else { 2229 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2230 tcp_send_challenge_ack(tp, th, m); 2231 m = NULL; 2232 } 2233 goto drop; 2234 } 2235 2236 /* 2237 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2238 * and it's less than ts_recent, drop it. 2239 */ 2240 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2241 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2242 /* Check to see if ts_recent is over 24 days old. */ 2243 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2244 /* 2245 * Invalidate ts_recent. If this segment updates 2246 * ts_recent, the age will be reset later and ts_recent 2247 * will get a valid value. If it does not, setting 2248 * ts_recent to zero will at least satisfy the 2249 * requirement that zero be placed in the timestamp 2250 * echo reply when ts_recent isn't valid. The 2251 * age isn't reset until we get a valid ts_recent 2252 * because we don't want out-of-order segments to be 2253 * dropped when ts_recent is old. 2254 */ 2255 tp->ts_recent = 0; 2256 } else { 2257 TCPSTAT_INC(tcps_rcvduppack); 2258 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2259 TCPSTAT_INC(tcps_pawsdrop); 2260 if (tlen) 2261 goto dropafterack; 2262 goto drop; 2263 } 2264 } 2265 2266 /* 2267 * In the SYN-RECEIVED state, validate that the packet belongs to 2268 * this connection before trimming the data to fit the receive 2269 * window. Check the sequence number versus IRS since we know 2270 * the sequence numbers haven't wrapped. This is a partial fix 2271 * for the "LAND" DoS attack. 2272 */ 2273 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2274 rstreason = BANDLIM_RST_OPENPORT; 2275 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2276 goto dropwithreset; 2277 } 2278 2279 todrop = tp->rcv_nxt - th->th_seq; 2280 if (todrop > 0) { 2281 if (thflags & TH_SYN) { 2282 thflags &= ~TH_SYN; 2283 th->th_seq++; 2284 if (th->th_urp > 1) 2285 th->th_urp--; 2286 else 2287 thflags &= ~TH_URG; 2288 todrop--; 2289 } 2290 /* 2291 * Following if statement from Stevens, vol. 2, p. 960. 2292 */ 2293 if (todrop > tlen 2294 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2295 /* 2296 * Any valid FIN must be to the left of the window. 2297 * At this point the FIN must be a duplicate or out 2298 * of sequence; drop it. 2299 */ 2300 thflags &= ~TH_FIN; 2301 2302 /* 2303 * Send an ACK to resynchronize and drop any data. 2304 * But keep on processing for RST or ACK. 2305 */ 2306 tp->t_flags |= TF_ACKNOW; 2307 todrop = tlen; 2308 TCPSTAT_INC(tcps_rcvduppack); 2309 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2310 } else { 2311 TCPSTAT_INC(tcps_rcvpartduppack); 2312 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2313 } 2314 /* 2315 * DSACK - add SACK block for dropped range 2316 */ 2317 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2318 tcp_update_sack_list(tp, th->th_seq, 2319 th->th_seq + todrop); 2320 /* 2321 * ACK now, as the next in-sequence segment 2322 * will clear the DSACK block again 2323 */ 2324 tp->t_flags |= TF_ACKNOW; 2325 } 2326 drop_hdrlen += todrop; /* drop from the top afterwards */ 2327 th->th_seq += todrop; 2328 tlen -= todrop; 2329 if (th->th_urp > todrop) 2330 th->th_urp -= todrop; 2331 else { 2332 thflags &= ~TH_URG; 2333 th->th_urp = 0; 2334 } 2335 } 2336 2337 /* 2338 * If new data are received on a connection after the 2339 * user processes are gone, then RST the other end if 2340 * no FIN has been processed. 2341 */ 2342 if ((tp->t_flags & TF_CLOSED) && tlen > 0 && 2343 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2344 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2345 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2346 "after socket was closed, " 2347 "sending RST and removing tcpcb\n", 2348 s, __func__, tcpstates[tp->t_state], tlen); 2349 free(s, M_TCPLOG); 2350 } 2351 tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE); 2352 /* tcp_close will kill the inp pre-log the Reset */ 2353 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST); 2354 tp = tcp_close(tp); 2355 TCPSTAT_INC(tcps_rcvafterclose); 2356 rstreason = BANDLIM_UNLIMITED; 2357 goto dropwithreset; 2358 } 2359 2360 /* 2361 * If segment ends after window, drop trailing data 2362 * (and PUSH and FIN); if nothing left, just ACK. 2363 */ 2364 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2365 if (todrop > 0) { 2366 TCPSTAT_INC(tcps_rcvpackafterwin); 2367 if (todrop >= tlen) { 2368 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2369 /* 2370 * If window is closed can only take segments at 2371 * window edge, and have to drop data and PUSH from 2372 * incoming segments. Continue processing, but 2373 * remember to ack. Otherwise, drop segment 2374 * and ack. 2375 */ 2376 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2377 tp->t_flags |= TF_ACKNOW; 2378 TCPSTAT_INC(tcps_rcvwinprobe); 2379 } else 2380 goto dropafterack; 2381 } else 2382 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2383 m_adj(m, -todrop); 2384 tlen -= todrop; 2385 thflags &= ~(TH_PUSH|TH_FIN); 2386 } 2387 2388 /* 2389 * If last ACK falls within this segment's sequence numbers, 2390 * record its timestamp. 2391 * NOTE: 2392 * 1) That the test incorporates suggestions from the latest 2393 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2394 * 2) That updating only on newer timestamps interferes with 2395 * our earlier PAWS tests, so this check should be solely 2396 * predicated on the sequence space of this segment. 2397 * 3) That we modify the segment boundary check to be 2398 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2399 * instead of RFC1323's 2400 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2401 * This modified check allows us to overcome RFC1323's 2402 * limitations as described in Stevens TCP/IP Illustrated 2403 * Vol. 2 p.869. In such cases, we can still calculate the 2404 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2405 */ 2406 if ((to.to_flags & TOF_TS) != 0 && 2407 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2408 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2409 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2410 tp->ts_recent_age = tcp_ts_getticks(); 2411 tp->ts_recent = to.to_tsval; 2412 } 2413 2414 /* 2415 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2416 * flag is on (half-synchronized state), then queue data for 2417 * later processing; else drop segment and return. 2418 */ 2419 if ((thflags & TH_ACK) == 0) { 2420 if (tp->t_state == TCPS_SYN_RECEIVED || 2421 (tp->t_flags & TF_NEEDSYN)) { 2422 if (tp->t_state == TCPS_SYN_RECEIVED && 2423 (tp->t_flags & TF_FASTOPEN)) { 2424 tp->snd_wnd = tiwin; 2425 cc_conn_init(tp); 2426 } 2427 goto step6; 2428 } else if (tp->t_flags & TF_ACKNOW) 2429 goto dropafterack; 2430 else 2431 goto drop; 2432 } 2433 2434 /* 2435 * Ack processing. 2436 */ 2437 if (SEQ_GEQ(tp->snd_una, tp->iss + (TCP_MAXWIN << tp->snd_scale))) { 2438 /* Checking SEG.ACK against ISS is definitely redundant. */ 2439 tp->t_flags2 |= TF2_NO_ISS_CHECK; 2440 } 2441 if (!V_tcp_insecure_ack) { 2442 tcp_seq seq_min; 2443 bool ghost_ack_check; 2444 2445 if (tp->t_flags2 & TF2_NO_ISS_CHECK) { 2446 /* Check for too old ACKs (RFC 5961, Section 5.2). */ 2447 seq_min = tp->snd_una - tp->max_sndwnd; 2448 ghost_ack_check = false; 2449 } else { 2450 if (SEQ_GT(tp->iss + 1, tp->snd_una - tp->max_sndwnd)) { 2451 /* Checking for ghost ACKs is stricter. */ 2452 seq_min = tp->iss + 1; 2453 ghost_ack_check = true; 2454 } else { 2455 /* 2456 * Checking for too old ACKs (RFC 5961, 2457 * Section 5.2) is stricter. 2458 */ 2459 seq_min = tp->snd_una - tp->max_sndwnd; 2460 ghost_ack_check = false; 2461 } 2462 } 2463 if (SEQ_LT(th->th_ack, seq_min)) { 2464 if (ghost_ack_check) 2465 TCPSTAT_INC(tcps_rcvghostack); 2466 else 2467 TCPSTAT_INC(tcps_rcvacktooold); 2468 tcp_send_challenge_ack(tp, th, m); 2469 m = NULL; 2470 goto drop; 2471 } 2472 } 2473 switch (tp->t_state) { 2474 /* 2475 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2476 * ESTABLISHED state and continue processing. 2477 * The ACK was checked above. 2478 */ 2479 case TCPS_SYN_RECEIVED: 2480 2481 TCPSTAT_INC(tcps_connects); 2482 if (tp->t_flags & TF_SONOTCONN) { 2483 /* 2484 * Usually SYN_RECEIVED had been created from a LISTEN, 2485 * and solisten_enqueue() has already marked the socket 2486 * layer as connected. If it didn't, which can happen 2487 * only with an accept_filter(9), then the tp is marked 2488 * with TF_SONOTCONN. The other reason for this mark 2489 * to be set is a simultaneous open, a SYN_RECEIVED 2490 * that had been created from SYN_SENT. 2491 */ 2492 tp->t_flags &= ~TF_SONOTCONN; 2493 soisconnected(so); 2494 } 2495 /* Do window scaling? */ 2496 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2497 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2498 tp->rcv_scale = tp->request_r_scale; 2499 } 2500 tp->snd_wnd = tiwin; 2501 /* 2502 * Make transitions: 2503 * SYN-RECEIVED -> ESTABLISHED 2504 * SYN-RECEIVED* -> FIN-WAIT-1 2505 */ 2506 tp->t_starttime = ticks; 2507 if ((tp->t_flags & TF_FASTOPEN) && tp->t_tfo_pending) { 2508 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2509 tp->t_tfo_pending = NULL; 2510 } 2511 if (tp->t_flags & TF_NEEDFIN) { 2512 tp->t_acktime = ticks; 2513 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2514 tp->t_flags &= ~TF_NEEDFIN; 2515 } else { 2516 tcp_state_change(tp, TCPS_ESTABLISHED); 2517 TCP_PROBE5(accept__established, NULL, tp, 2518 m, tp, th); 2519 /* 2520 * TFO connections call cc_conn_init() during SYN 2521 * processing. Calling it again here for such 2522 * connections is not harmless as it would undo the 2523 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2524 * is retransmitted. 2525 */ 2526 if (!(tp->t_flags & TF_FASTOPEN)) 2527 cc_conn_init(tp); 2528 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2529 } 2530 /* 2531 * Account for the ACK of our SYN prior to 2532 * regular ACK processing below, except for 2533 * simultaneous SYN, which is handled later. 2534 */ 2535 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2536 incforsyn = 1; 2537 /* 2538 * If segment contains data or ACK, will call tcp_reass() 2539 * later; if not, do so now to pass queued data to user. 2540 */ 2541 if (tlen == 0 && (thflags & TH_FIN) == 0) { 2542 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2543 (struct mbuf *)0); 2544 tcp_handle_wakeup(tp); 2545 } 2546 tp->snd_wl1 = th->th_seq - 1; 2547 /* FALLTHROUGH */ 2548 2549 /* 2550 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2551 * ACKs. If the ack is in the range 2552 * tp->snd_una < th->th_ack <= tp->snd_max 2553 * then advance tp->snd_una to th->th_ack and drop 2554 * data from the retransmission queue. If this ACK reflects 2555 * more up to date window information we update our window information. 2556 */ 2557 case TCPS_ESTABLISHED: 2558 case TCPS_FIN_WAIT_1: 2559 case TCPS_FIN_WAIT_2: 2560 case TCPS_CLOSE_WAIT: 2561 case TCPS_CLOSING: 2562 case TCPS_LAST_ACK: 2563 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2564 TCPSTAT_INC(tcps_rcvacktoomuch); 2565 goto dropafterack; 2566 } 2567 if (tcp_is_sack_recovery(tp, &to)) { 2568 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2569 if ((sack_changed != SACK_NOCHANGE) && 2570 (tp->t_flags & TF_LRD)) { 2571 tcp_sack_lost_retransmission(tp, th); 2572 } 2573 } else 2574 /* 2575 * Reset the value so that previous (valid) value 2576 * from the last ack with SACK doesn't get used. 2577 */ 2578 tp->sackhint.sacked_bytes = 0; 2579 2580 #ifdef TCP_HHOOK 2581 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2582 hhook_run_tcp_est_in(tp, th, &to); 2583 #endif 2584 2585 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2586 maxseg = tcp_maxseg(tp); 2587 if (tlen == 0 && 2588 (tiwin == tp->snd_wnd || 2589 (tp->t_flags & TF_SACK_PERMIT))) { 2590 /* 2591 * If this is the first time we've seen a 2592 * FIN from the remote, this is not a 2593 * duplicate and it needs to be processed 2594 * normally. This happens during a 2595 * simultaneous close. 2596 */ 2597 if ((thflags & TH_FIN) && 2598 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2599 tp->t_dupacks = 0; 2600 break; 2601 } 2602 TCPSTAT_INC(tcps_rcvdupack); 2603 /* 2604 * If we have outstanding data (other than 2605 * a window probe), this is a completely 2606 * duplicate ack (ie, window info didn't 2607 * change and FIN isn't set), 2608 * the ack is the biggest we've 2609 * seen and we've seen exactly our rexmt 2610 * threshold of them, assume a packet 2611 * has been dropped and retransmit it. 2612 * Kludge snd_nxt & the congestion 2613 * window so we send only this one 2614 * packet. 2615 * 2616 * We know we're losing at the current 2617 * window size so do congestion avoidance 2618 * (set ssthresh to half the current window 2619 * and pull our congestion window back to 2620 * the new ssthresh). 2621 * 2622 * Dup acks mean that packets have left the 2623 * network (they're now cached at the receiver) 2624 * so bump cwnd by the amount in the receiver 2625 * to keep a constant cwnd packets in the 2626 * network. 2627 * 2628 * When using TCP ECN, notify the peer that 2629 * we reduced the cwnd. 2630 */ 2631 /* 2632 * Following 2 kinds of acks should not affect 2633 * dupack counting: 2634 * 1) Old acks 2635 * 2) Acks with SACK but without any new SACK 2636 * information in them. These could result from 2637 * any anomaly in the network like a switch 2638 * duplicating packets or a possible DoS attack. 2639 */ 2640 if (th->th_ack != tp->snd_una || 2641 (tcp_is_sack_recovery(tp, &to) && 2642 (sack_changed == SACK_NOCHANGE))) { 2643 break; 2644 } else if (!tcp_timer_active(tp, TT_REXMT)) { 2645 tp->t_dupacks = 0; 2646 } else if (++tp->t_dupacks > tcprexmtthresh || 2647 IN_FASTRECOVERY(tp->t_flags)) { 2648 cc_ack_received(tp, th, nsegs, 2649 CC_DUPACK); 2650 if (V_tcp_do_prr && 2651 IN_FASTRECOVERY(tp->t_flags) && 2652 (tp->t_flags & TF_SACK_PERMIT)) { 2653 tcp_do_prr_ack(tp, th, &to, 2654 sack_changed, &maxseg); 2655 } else if (tcp_is_sack_recovery(tp, &to) && 2656 IN_FASTRECOVERY(tp->t_flags) && 2657 (tp->snd_nxt == tp->snd_max)) { 2658 int awnd; 2659 2660 /* 2661 * Compute the amount of data in flight first. 2662 * We can inject new data into the pipe iff 2663 * we have less than ssthresh 2664 * worth of data in flight. 2665 */ 2666 if (V_tcp_do_newsack) { 2667 awnd = tcp_compute_pipe(tp); 2668 } else { 2669 awnd = (tp->snd_nxt - tp->snd_fack) + 2670 tp->sackhint.sack_bytes_rexmit; 2671 } 2672 if (awnd < tp->snd_ssthresh) { 2673 tp->snd_cwnd += imax(maxseg, 2674 imin(2 * maxseg, 2675 tp->sackhint.delivered_data)); 2676 if (tp->snd_cwnd > tp->snd_ssthresh) 2677 tp->snd_cwnd = tp->snd_ssthresh; 2678 } 2679 } else if (tcp_is_sack_recovery(tp, &to) && 2680 IN_FASTRECOVERY(tp->t_flags) && 2681 SEQ_LT(tp->snd_nxt, tp->snd_max)) { 2682 tp->snd_cwnd += imax(maxseg, 2683 imin(2 * maxseg, 2684 tp->sackhint.delivered_data)); 2685 } else { 2686 tp->snd_cwnd += maxseg; 2687 } 2688 (void) tcp_output(tp); 2689 goto drop; 2690 } else if (tp->t_dupacks == tcprexmtthresh || 2691 (tp->t_flags & TF_SACK_PERMIT && 2692 V_tcp_do_newsack && 2693 tp->sackhint.sacked_bytes > 2694 (tcprexmtthresh - 1) * maxseg)) { 2695 enter_recovery: 2696 /* 2697 * Above is the RFC6675 trigger condition of 2698 * more than (dupthresh-1)*maxseg sacked data. 2699 * If the count of holes in the 2700 * scoreboard is >= dupthresh, we could 2701 * also enter loss recovery, but don't 2702 * have that value readily available. 2703 */ 2704 tp->t_dupacks = tcprexmtthresh; 2705 tcp_seq onxt = tp->snd_nxt; 2706 2707 /* 2708 * If we're doing sack, check to 2709 * see if we're already in sack 2710 * recovery. If we're not doing sack, 2711 * check to see if we're in newreno 2712 * recovery. 2713 */ 2714 if (tcp_is_sack_recovery(tp, &to)) { 2715 if (IN_FASTRECOVERY(tp->t_flags)) { 2716 tp->t_dupacks = 0; 2717 break; 2718 } 2719 } else { 2720 if (SEQ_LEQ(th->th_ack, 2721 tp->snd_recover)) { 2722 tp->t_dupacks = 0; 2723 break; 2724 } 2725 } 2726 /* Congestion signal before ack. */ 2727 cc_cong_signal(tp, th, CC_NDUPACK); 2728 cc_ack_received(tp, th, nsegs, 2729 CC_DUPACK); 2730 tcp_timer_activate(tp, TT_REXMT, 0); 2731 tp->t_rtttime = 0; 2732 if (V_tcp_do_prr) { 2733 /* 2734 * snd_ssthresh and snd_recover are 2735 * already updated by cc_cong_signal. 2736 */ 2737 if (tcp_is_sack_recovery(tp, &to)) { 2738 /* 2739 * Include Limited Transmit 2740 * segments here 2741 */ 2742 tp->sackhint.prr_delivered = 2743 imin(tp->snd_max - th->th_ack, 2744 (tp->snd_limited + 1) * maxseg); 2745 } else { 2746 tp->sackhint.prr_delivered = 2747 maxseg; 2748 } 2749 tp->sackhint.recover_fs = max(1, 2750 tp->snd_nxt - tp->snd_una); 2751 } 2752 tp->snd_limited = 0; 2753 if (tcp_is_sack_recovery(tp, &to)) { 2754 TCPSTAT_INC(tcps_sack_recovery_episode); 2755 /* 2756 * When entering LR after RTO due to 2757 * Duplicate ACKs, retransmit existing 2758 * holes from the scoreboard. 2759 */ 2760 tcp_resend_sackholes(tp); 2761 /* Avoid inflating cwnd in tcp_output */ 2762 tp->snd_nxt = tp->snd_max; 2763 tp->snd_cwnd = tcp_compute_pipe(tp) + 2764 maxseg; 2765 (void) tcp_output(tp); 2766 /* Set cwnd to the expected flightsize */ 2767 tp->snd_cwnd = tp->snd_ssthresh; 2768 if (SEQ_GT(th->th_ack, tp->snd_una)) { 2769 goto resume_partialack; 2770 } 2771 goto drop; 2772 } 2773 tp->snd_nxt = th->th_ack; 2774 tp->snd_cwnd = maxseg; 2775 (void) tcp_output(tp); 2776 KASSERT(tp->snd_limited <= 2, 2777 ("%s: tp->snd_limited too big", 2778 __func__)); 2779 tp->snd_cwnd = tp->snd_ssthresh + 2780 maxseg * 2781 (tp->t_dupacks - tp->snd_limited); 2782 if (SEQ_GT(onxt, tp->snd_nxt)) 2783 tp->snd_nxt = onxt; 2784 goto drop; 2785 } else if (V_tcp_do_rfc3042) { 2786 /* 2787 * Process first and second duplicate 2788 * ACKs. Each indicates a segment 2789 * leaving the network, creating room 2790 * for more. Make sure we can send a 2791 * packet on reception of each duplicate 2792 * ACK by increasing snd_cwnd by one 2793 * segment. Restore the original 2794 * snd_cwnd after packet transmission. 2795 */ 2796 cc_ack_received(tp, th, nsegs, CC_DUPACK); 2797 uint32_t oldcwnd = tp->snd_cwnd; 2798 tcp_seq oldsndmax = tp->snd_max; 2799 u_int sent; 2800 int avail; 2801 2802 KASSERT(tp->t_dupacks == 1 || 2803 tp->t_dupacks == 2, 2804 ("%s: dupacks not 1 or 2", 2805 __func__)); 2806 if (tp->t_dupacks == 1) 2807 tp->snd_limited = 0; 2808 if ((tp->snd_nxt == tp->snd_max) && 2809 (tp->t_rxtshift == 0)) 2810 tp->snd_cwnd = 2811 SEQ_SUB(tp->snd_nxt, 2812 tp->snd_una) - 2813 tcp_sack_adjust(tp); 2814 tp->snd_cwnd += 2815 (tp->t_dupacks - tp->snd_limited) * 2816 maxseg; 2817 /* 2818 * Only call tcp_output when there 2819 * is new data available to be sent 2820 * or we need to send an ACK. 2821 */ 2822 SOCKBUF_LOCK(&so->so_snd); 2823 avail = sbavail(&so->so_snd); 2824 SOCKBUF_UNLOCK(&so->so_snd); 2825 if (tp->t_flags & TF_ACKNOW || 2826 (avail >= 2827 SEQ_SUB(tp->snd_nxt, tp->snd_una))) { 2828 (void) tcp_output(tp); 2829 } 2830 sent = SEQ_SUB(tp->snd_max, oldsndmax); 2831 if (sent > maxseg) { 2832 KASSERT((tp->t_dupacks == 2 && 2833 tp->snd_limited == 0) || 2834 (sent == maxseg + 1 && 2835 tp->t_flags & TF_SENTFIN), 2836 ("%s: sent too much", 2837 __func__)); 2838 tp->snd_limited = 2; 2839 } else if (sent > 0) { 2840 ++tp->snd_limited; 2841 } 2842 tp->snd_cwnd = oldcwnd; 2843 goto drop; 2844 } 2845 } 2846 break; 2847 } else { 2848 /* 2849 * This ack is advancing the left edge, reset the 2850 * counter. 2851 */ 2852 tp->t_dupacks = 0; 2853 /* 2854 * If this ack also has new SACK info, increment the 2855 * counter as per rfc6675. The variable 2856 * sack_changed tracks all changes to the SACK 2857 * scoreboard, including when partial ACKs without 2858 * SACK options are received, and clear the scoreboard 2859 * from the left side. Such partial ACKs should not be 2860 * counted as dupacks here. 2861 */ 2862 if (tcp_is_sack_recovery(tp, &to) && 2863 (((tp->t_rxtshift == 0) && (sack_changed != SACK_NOCHANGE)) || 2864 ((tp->t_rxtshift > 0) && (sack_changed == SACK_NEWLOSS))) && 2865 (tp->snd_nxt == tp->snd_max)) { 2866 tp->t_dupacks++; 2867 /* limit overhead by setting maxseg last */ 2868 if (!IN_FASTRECOVERY(tp->t_flags) && 2869 (tp->sackhint.sacked_bytes > 2870 ((tcprexmtthresh - 1) * 2871 (maxseg = tcp_maxseg(tp))))) { 2872 goto enter_recovery; 2873 } 2874 } 2875 } 2876 2877 resume_partialack: 2878 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2879 ("%s: th_ack <= snd_una", __func__)); 2880 2881 /* 2882 * If the congestion window was inflated to account 2883 * for the other side's cached packets, retract it. 2884 */ 2885 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2886 if (IN_FASTRECOVERY(tp->t_flags)) { 2887 if (tp->t_flags & TF_SACK_PERMIT) { 2888 if (V_tcp_do_prr && 2889 (to.to_flags & TOF_SACK)) { 2890 tcp_timer_activate(tp, 2891 TT_REXMT, 0); 2892 tp->t_rtttime = 0; 2893 tcp_do_prr_ack(tp, th, &to, 2894 sack_changed, &maxseg); 2895 tp->t_flags |= TF_ACKNOW; 2896 (void) tcp_output(tp); 2897 } else { 2898 tcp_sack_partialack(tp, th, 2899 &maxseg); 2900 } 2901 } else { 2902 tcp_newreno_partial_ack(tp, th); 2903 } 2904 } else if (IN_CONGRECOVERY(tp->t_flags) && 2905 (V_tcp_do_prr)) { 2906 tp->sackhint.delivered_data = 2907 BYTES_THIS_ACK(tp, th); 2908 tp->snd_fack = th->th_ack; 2909 /* 2910 * During ECN cwnd reduction 2911 * always use PRR-SSRB 2912 */ 2913 tcp_do_prr_ack(tp, th, &to, SACK_CHANGE, 2914 &maxseg); 2915 (void) tcp_output(tp); 2916 } 2917 } 2918 /* 2919 * If we reach this point, ACK is not a duplicate, 2920 * i.e., it ACKs something we sent. 2921 */ 2922 if (tp->t_flags & TF_NEEDSYN) { 2923 /* 2924 * T/TCP: Connection was half-synchronized, and our 2925 * SYN has been ACK'd (so connection is now fully 2926 * synchronized). Go to non-starred state, 2927 * increment snd_una for ACK of SYN, and check if 2928 * we can do window scaling. 2929 */ 2930 tp->t_flags &= ~TF_NEEDSYN; 2931 tp->snd_una++; 2932 /* Do window scaling? */ 2933 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2934 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2935 tp->rcv_scale = tp->request_r_scale; 2936 /* Send window already scaled. */ 2937 } 2938 } 2939 2940 process_ACK: 2941 INP_WLOCK_ASSERT(inp); 2942 2943 /* 2944 * Adjust for the SYN bit in sequence space, 2945 * but don't account for it in cwnd calculations. 2946 * This is for the SYN_RECEIVED, non-simultaneous 2947 * SYN case. SYN_SENT and simultaneous SYN are 2948 * treated elsewhere. 2949 */ 2950 if (incforsyn) 2951 tp->snd_una++; 2952 acked = BYTES_THIS_ACK(tp, th); 2953 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2954 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2955 tp->snd_una, th->th_ack, tp, m)); 2956 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2957 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2958 2959 /* 2960 * If we just performed our first retransmit, and the ACK 2961 * arrives within our recovery window, then it was a mistake 2962 * to do the retransmit in the first place. Recover our 2963 * original cwnd and ssthresh, and proceed to transmit where 2964 * we left off. 2965 */ 2966 if (tp->t_rxtshift == 1 && 2967 tp->t_flags & TF_PREVVALID && 2968 tp->t_badrxtwin != 0 && 2969 to.to_flags & TOF_TS && 2970 to.to_tsecr != 0 && 2971 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 2972 cc_cong_signal(tp, th, CC_RTO_ERR); 2973 2974 /* 2975 * If we have a timestamp reply, update smoothed 2976 * round trip time. If no timestamp is present but 2977 * transmit timer is running and timed sequence 2978 * number was acked, update smoothed round trip time. 2979 * Since we now have an rtt measurement, cancel the 2980 * timer backoff (cf., Phil Karn's retransmit alg.). 2981 * Recompute the initial retransmit timer. 2982 * 2983 * Some boxes send broken timestamp replies 2984 * during the SYN+ACK phase, ignore 2985 * timestamps of 0 or we could calculate a 2986 * huge RTT and blow up the retransmit timer. 2987 */ 2988 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2989 uint32_t t; 2990 2991 t = tcp_ts_getticks() - to.to_tsecr; 2992 if (!tp->t_rttlow || tp->t_rttlow > t) 2993 tp->t_rttlow = t; 2994 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2995 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2996 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2997 tp->t_rttlow = ticks - tp->t_rtttime; 2998 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2999 } 3000 3001 SOCKBUF_LOCK(&so->so_snd); 3002 /* 3003 * Clear t_acktime if remote side has ACKd all data in the 3004 * socket buffer and FIN (if applicable). 3005 * Otherwise, update t_acktime if we received a sufficiently 3006 * large ACK. 3007 */ 3008 if ((tp->t_state <= TCPS_CLOSE_WAIT && 3009 acked == sbavail(&so->so_snd)) || 3010 acked > sbavail(&so->so_snd)) 3011 tp->t_acktime = 0; 3012 else if (acked > 1) 3013 tp->t_acktime = ticks; 3014 3015 /* 3016 * If all outstanding data is acked, stop retransmit 3017 * timer and remember to restart (more output or persist). 3018 * If there is more data to be acked, restart retransmit 3019 * timer, using current (possibly backed-off) value. 3020 */ 3021 if (th->th_ack == tp->snd_max) { 3022 tcp_timer_activate(tp, TT_REXMT, 0); 3023 needoutput = 1; 3024 } else if (!tcp_timer_active(tp, TT_PERSIST)) 3025 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); 3026 3027 /* 3028 * If no data (only SYN) was ACK'd, 3029 * skip rest of ACK processing. 3030 */ 3031 if (acked == 0) { 3032 SOCKBUF_UNLOCK(&so->so_snd); 3033 goto step6; 3034 } 3035 3036 /* 3037 * Let the congestion control algorithm update congestion 3038 * control related information. This typically means increasing 3039 * the congestion window. 3040 */ 3041 cc_ack_received(tp, th, nsegs, CC_ACK); 3042 3043 if (acked > sbavail(&so->so_snd)) { 3044 if (tp->snd_wnd >= sbavail(&so->so_snd)) 3045 tp->snd_wnd -= sbavail(&so->so_snd); 3046 else 3047 tp->snd_wnd = 0; 3048 mfree = sbcut_locked(&so->so_snd, 3049 (int)sbavail(&so->so_snd)); 3050 ourfinisacked = 1; 3051 } else { 3052 mfree = sbcut_locked(&so->so_snd, acked); 3053 if (tp->snd_wnd >= (uint32_t) acked) 3054 tp->snd_wnd -= acked; 3055 else 3056 tp->snd_wnd = 0; 3057 ourfinisacked = 0; 3058 } 3059 /* NB: sowwakeup_locked() does an implicit unlock. */ 3060 sowwakeup_locked(so); 3061 m_freem(mfree); 3062 /* Detect una wraparound. */ 3063 if (!IN_RECOVERY(tp->t_flags) && 3064 SEQ_GT(tp->snd_una, tp->snd_recover) && 3065 SEQ_LEQ(th->th_ack, tp->snd_recover)) 3066 tp->snd_recover = th->th_ack - 1; 3067 tp->snd_una = th->th_ack; 3068 if (IN_RECOVERY(tp->t_flags) && 3069 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 3070 cc_post_recovery(tp, th); 3071 } 3072 if (SEQ_GT(tp->snd_una, tp->snd_recover)) { 3073 tp->snd_recover = tp->snd_una; 3074 } 3075 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 3076 tp->snd_nxt = tp->snd_una; 3077 3078 switch (tp->t_state) { 3079 /* 3080 * In FIN_WAIT_1 STATE in addition to the processing 3081 * for the ESTABLISHED state if our FIN is now acknowledged 3082 * then enter FIN_WAIT_2. 3083 */ 3084 case TCPS_FIN_WAIT_1: 3085 if (ourfinisacked) { 3086 /* 3087 * If we can't receive any more 3088 * data, then closing user can proceed. 3089 * Starting the timer is contrary to the 3090 * specification, but if we don't get a FIN 3091 * we'll hang forever. 3092 */ 3093 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3094 tcp_free_sackholes(tp); 3095 soisdisconnected(so); 3096 tcp_timer_activate(tp, TT_2MSL, 3097 (tcp_fast_finwait2_recycle ? 3098 tcp_finwait2_timeout : 3099 TP_MAXIDLE(tp))); 3100 } 3101 tcp_state_change(tp, TCPS_FIN_WAIT_2); 3102 } 3103 break; 3104 3105 /* 3106 * In CLOSING STATE in addition to the processing for 3107 * the ESTABLISHED state if the ACK acknowledges our FIN 3108 * then enter the TIME-WAIT state, otherwise ignore 3109 * the segment. 3110 */ 3111 case TCPS_CLOSING: 3112 if (ourfinisacked) { 3113 tcp_twstart(tp); 3114 m_freem(m); 3115 return; 3116 } 3117 break; 3118 3119 /* 3120 * In LAST_ACK, we may still be waiting for data to drain 3121 * and/or to be acked, as well as for the ack of our FIN. 3122 * If our FIN is now acknowledged, delete the TCB, 3123 * enter the closed state and return. 3124 */ 3125 case TCPS_LAST_ACK: 3126 if (ourfinisacked) { 3127 tp = tcp_close(tp); 3128 goto drop; 3129 } 3130 break; 3131 } 3132 } 3133 3134 step6: 3135 INP_WLOCK_ASSERT(inp); 3136 3137 /* 3138 * Update window information. 3139 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3140 */ 3141 if ((thflags & TH_ACK) && 3142 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3143 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3144 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3145 /* keep track of pure window updates */ 3146 if (tlen == 0 && 3147 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3148 TCPSTAT_INC(tcps_rcvwinupd); 3149 tp->snd_wnd = tiwin; 3150 tp->snd_wl1 = th->th_seq; 3151 tp->snd_wl2 = th->th_ack; 3152 if (tp->snd_wnd > tp->max_sndwnd) 3153 tp->max_sndwnd = tp->snd_wnd; 3154 needoutput = 1; 3155 } 3156 3157 /* 3158 * Process segments with URG. 3159 */ 3160 if ((thflags & TH_URG) && th->th_urp && 3161 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3162 /* 3163 * This is a kludge, but if we receive and accept 3164 * random urgent pointers, we'll crash in 3165 * soreceive. It's hard to imagine someone 3166 * actually wanting to send this much urgent data. 3167 */ 3168 SOCKBUF_LOCK(&so->so_rcv); 3169 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3170 th->th_urp = 0; /* XXX */ 3171 thflags &= ~TH_URG; /* XXX */ 3172 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3173 goto dodata; /* XXX */ 3174 } 3175 /* 3176 * If this segment advances the known urgent pointer, 3177 * then mark the data stream. This should not happen 3178 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3179 * a FIN has been received from the remote side. 3180 * In these states we ignore the URG. 3181 * 3182 * According to RFC961 (Assigned Protocols), 3183 * the urgent pointer points to the last octet 3184 * of urgent data. We continue, however, 3185 * to consider it to indicate the first octet 3186 * of data past the urgent section as the original 3187 * spec states (in one of two places). 3188 */ 3189 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3190 tp->rcv_up = th->th_seq + th->th_urp; 3191 so->so_oobmark = sbavail(&so->so_rcv) + 3192 (tp->rcv_up - tp->rcv_nxt) - 1; 3193 if (so->so_oobmark == 0) 3194 so->so_rcv.sb_state |= SBS_RCVATMARK; 3195 sohasoutofband(so); 3196 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3197 } 3198 SOCKBUF_UNLOCK(&so->so_rcv); 3199 /* 3200 * Remove out of band data so doesn't get presented to user. 3201 * This can happen independent of advancing the URG pointer, 3202 * but if two URG's are pending at once, some out-of-band 3203 * data may creep in... ick. 3204 */ 3205 if (th->th_urp <= (uint32_t)tlen && 3206 !(so->so_options & SO_OOBINLINE)) { 3207 /* hdr drop is delayed */ 3208 tcp_pulloutofband(so, th, m, drop_hdrlen); 3209 } 3210 } else { 3211 /* 3212 * If no out of band data is expected, 3213 * pull receive urgent pointer along 3214 * with the receive window. 3215 */ 3216 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3217 tp->rcv_up = tp->rcv_nxt; 3218 } 3219 dodata: /* XXX */ 3220 INP_WLOCK_ASSERT(inp); 3221 3222 /* 3223 * Process the segment text, merging it into the TCP sequencing queue, 3224 * and arranging for acknowledgment of receipt if necessary. 3225 * This process logically involves adjusting tp->rcv_wnd as data 3226 * is presented to the user (this happens in tcp_usrreq.c, 3227 * case PRU_RCVD). If a FIN has already been received on this 3228 * connection then we just ignore the text. 3229 */ 3230 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3231 (tp->t_flags & TF_FASTOPEN)); 3232 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3233 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3234 tcp_seq save_start = th->th_seq; 3235 tcp_seq save_rnxt = tp->rcv_nxt; 3236 int save_tlen = tlen; 3237 m_adj(m, drop_hdrlen); /* delayed header drop */ 3238 /* 3239 * Insert segment which includes th into TCP reassembly queue 3240 * with control block tp. Set thflags to whether reassembly now 3241 * includes a segment with FIN. This handles the common case 3242 * inline (segment is the next to be received on an established 3243 * connection, and the queue is empty), avoiding linkage into 3244 * and removal from the queue and repetition of various 3245 * conversions. 3246 * Set DELACK for segments received in order, but ack 3247 * immediately when segments are out of order (so 3248 * fast retransmit can work). 3249 */ 3250 if (th->th_seq == tp->rcv_nxt && 3251 SEGQ_EMPTY(tp) && 3252 (TCPS_HAVEESTABLISHED(tp->t_state) || 3253 tfo_syn)) { 3254 if (DELAY_ACK(tp, tlen) || tfo_syn) 3255 tp->t_flags |= TF_DELACK; 3256 else 3257 tp->t_flags |= TF_ACKNOW; 3258 tp->rcv_nxt += tlen; 3259 if (tlen && 3260 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3261 (tp->t_fbyte_in == 0)) { 3262 tp->t_fbyte_in = ticks; 3263 if (tp->t_fbyte_in == 0) 3264 tp->t_fbyte_in = 1; 3265 if (tp->t_fbyte_out && tp->t_fbyte_in) 3266 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3267 } 3268 thflags = tcp_get_flags(th) & TH_FIN; 3269 TCPSTAT_INC(tcps_rcvpack); 3270 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3271 SOCKBUF_LOCK(&so->so_rcv); 3272 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3273 m_freem(m); 3274 else 3275 sbappendstream_locked(&so->so_rcv, m, 0); 3276 tp->t_flags |= TF_WAKESOR; 3277 } else { 3278 /* 3279 * XXX: Due to the header drop above "th" is 3280 * theoretically invalid by now. Fortunately 3281 * m_adj() doesn't actually frees any mbufs 3282 * when trimming from the head. 3283 */ 3284 tcp_seq temp = save_start; 3285 3286 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3287 tp->t_flags |= TF_ACKNOW; 3288 } 3289 if ((tp->t_flags & TF_SACK_PERMIT) && 3290 (save_tlen > 0) && 3291 TCPS_HAVEESTABLISHED(tp->t_state)) { 3292 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3293 /* 3294 * DSACK actually handled in the fastpath 3295 * above. 3296 */ 3297 tcp_update_sack_list(tp, save_start, 3298 save_start + save_tlen); 3299 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3300 if ((tp->rcv_numsacks >= 1) && 3301 (tp->sackblks[0].end == save_start)) { 3302 /* 3303 * Partial overlap, recorded at todrop 3304 * above. 3305 */ 3306 tcp_update_sack_list(tp, 3307 tp->sackblks[0].start, 3308 tp->sackblks[0].end); 3309 } else { 3310 tcp_update_dsack_list(tp, save_start, 3311 save_start + save_tlen); 3312 } 3313 } else if (tlen >= save_tlen) { 3314 /* Update of sackblks. */ 3315 tcp_update_dsack_list(tp, save_start, 3316 save_start + save_tlen); 3317 } else if (tlen > 0) { 3318 tcp_update_dsack_list(tp, save_start, 3319 save_start + tlen); 3320 } 3321 } 3322 tcp_handle_wakeup(tp); 3323 #if 0 3324 /* 3325 * Note the amount of data that peer has sent into 3326 * our window, in order to estimate the sender's 3327 * buffer size. 3328 * XXX: Unused. 3329 */ 3330 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3331 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3332 else 3333 len = so->so_rcv.sb_hiwat; 3334 #endif 3335 } else { 3336 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 3337 if (tlen > 0) { 3338 if ((thflags & TH_FIN) != 0) { 3339 log(LOG_DEBUG, "%s; %s: %s: " 3340 "Received %d bytes of data and FIN " 3341 "after having received a FIN, " 3342 "just dropping both\n", 3343 s, __func__, 3344 tcpstates[tp->t_state], tlen); 3345 } else { 3346 log(LOG_DEBUG, "%s; %s: %s: " 3347 "Received %d bytes of data " 3348 "after having received a FIN, " 3349 "just dropping it\n", 3350 s, __func__, 3351 tcpstates[tp->t_state], tlen); 3352 } 3353 } else { 3354 if ((thflags & TH_FIN) != 0) { 3355 log(LOG_DEBUG, "%s; %s: %s: " 3356 "Received FIN " 3357 "after having received a FIN, " 3358 "just dropping it\n", 3359 s, __func__, 3360 tcpstates[tp->t_state]); 3361 } 3362 } 3363 free(s, M_TCPLOG); 3364 } 3365 m_freem(m); 3366 thflags &= ~TH_FIN; 3367 } 3368 3369 /* 3370 * If FIN is received ACK the FIN and let the user know 3371 * that the connection is closing. 3372 */ 3373 if (thflags & TH_FIN) { 3374 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3375 /* The socket upcall is handled by socantrcvmore. */ 3376 socantrcvmore(so); 3377 /* 3378 * If connection is half-synchronized 3379 * (ie NEEDSYN flag on) then delay ACK, 3380 * so it may be piggybacked when SYN is sent. 3381 * Otherwise, since we received a FIN then no 3382 * more input can be expected, send ACK now. 3383 */ 3384 if (tp->t_flags & TF_NEEDSYN) 3385 tp->t_flags |= TF_DELACK; 3386 else 3387 tp->t_flags |= TF_ACKNOW; 3388 tp->rcv_nxt++; 3389 } 3390 switch (tp->t_state) { 3391 /* 3392 * In SYN_RECEIVED and ESTABLISHED STATES 3393 * enter the CLOSE_WAIT state. 3394 */ 3395 case TCPS_SYN_RECEIVED: 3396 tp->t_starttime = ticks; 3397 /* FALLTHROUGH */ 3398 case TCPS_ESTABLISHED: 3399 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3400 break; 3401 3402 /* 3403 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3404 * enter the CLOSING state. 3405 */ 3406 case TCPS_FIN_WAIT_1: 3407 tcp_state_change(tp, TCPS_CLOSING); 3408 break; 3409 3410 /* 3411 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3412 * starting the time-wait timer, turning off the other 3413 * standard timers. 3414 */ 3415 case TCPS_FIN_WAIT_2: 3416 tcp_twstart(tp); 3417 return; 3418 } 3419 } 3420 TCP_PROBE3(debug__input, tp, th, m); 3421 3422 /* 3423 * Return any desired output. 3424 */ 3425 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 3426 (void) tcp_output(tp); 3427 } 3428 check_delack: 3429 INP_WLOCK_ASSERT(inp); 3430 3431 if (tp->t_flags & TF_DELACK) { 3432 tp->t_flags &= ~TF_DELACK; 3433 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3434 } 3435 INP_WUNLOCK(inp); 3436 return; 3437 3438 dropafterack: 3439 /* 3440 * Generate an ACK dropping incoming segment if it occupies 3441 * sequence space, where the ACK reflects our state. 3442 * 3443 * We can now skip the test for the RST flag since all 3444 * paths to this code happen after packets containing 3445 * RST have been dropped. 3446 * 3447 * In the SYN-RECEIVED state, don't send an ACK unless the 3448 * segment we received passes the SYN-RECEIVED ACK test. 3449 * If it fails send a RST. This breaks the loop in the 3450 * "LAND" DoS attack, and also prevents an ACK storm 3451 * between two listening ports that have been sent forged 3452 * SYN segments, each with the source address of the other. 3453 */ 3454 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3455 (SEQ_GT(tp->snd_una, th->th_ack) || 3456 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3457 rstreason = BANDLIM_RST_OPENPORT; 3458 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 3459 goto dropwithreset; 3460 } 3461 TCP_PROBE3(debug__input, tp, th, m); 3462 tp->t_flags |= TF_ACKNOW; 3463 (void) tcp_output(tp); 3464 INP_WUNLOCK(inp); 3465 m_freem(m); 3466 return; 3467 3468 dropwithreset: 3469 if (tp != NULL) { 3470 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3471 INP_WUNLOCK(inp); 3472 } else 3473 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3474 return; 3475 3476 drop: 3477 /* 3478 * Drop space held by incoming segment and return. 3479 */ 3480 TCP_PROBE3(debug__input, tp, th, m); 3481 if (tp != NULL) { 3482 INP_WUNLOCK(inp); 3483 } 3484 m_freem(m); 3485 } 3486 3487 /* 3488 * Issue RST and make ACK acceptable to originator of segment. 3489 * The mbuf must still include the original packet header. 3490 * tp may be NULL. 3491 */ 3492 void 3493 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3494 int tlen, int rstreason) 3495 { 3496 #ifdef INET 3497 struct ip *ip; 3498 #endif 3499 #ifdef INET6 3500 struct ip6_hdr *ip6; 3501 #endif 3502 3503 if (tp != NULL) { 3504 INP_LOCK_ASSERT(tptoinpcb(tp)); 3505 } 3506 3507 /* Don't bother if destination was broadcast/multicast. */ 3508 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3509 goto drop; 3510 #ifdef INET6 3511 if (mtod(m, struct ip *)->ip_v == 6) { 3512 ip6 = mtod(m, struct ip6_hdr *); 3513 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3514 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3515 goto drop; 3516 /* IPv6 anycast check is done at tcp6_input() */ 3517 } 3518 #endif 3519 #if defined(INET) && defined(INET6) 3520 else 3521 #endif 3522 #ifdef INET 3523 { 3524 ip = mtod(m, struct ip *); 3525 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3526 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3527 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3528 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3529 goto drop; 3530 } 3531 #endif 3532 3533 /* Perform bandwidth limiting. */ 3534 if (badport_bandlim(rstreason) < 0) 3535 goto drop; 3536 3537 /* tcp_respond consumes the mbuf chain. */ 3538 if (tcp_get_flags(th) & TH_ACK) { 3539 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3540 th->th_ack, TH_RST); 3541 } else { 3542 if (tcp_get_flags(th) & TH_SYN) 3543 tlen++; 3544 if (tcp_get_flags(th) & TH_FIN) 3545 tlen++; 3546 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3547 (tcp_seq)0, TH_RST|TH_ACK); 3548 } 3549 return; 3550 drop: 3551 m_freem(m); 3552 } 3553 3554 /* 3555 * Parse TCP options and place in tcpopt. 3556 */ 3557 void 3558 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3559 { 3560 int opt, optlen; 3561 3562 to->to_flags = 0; 3563 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3564 opt = cp[0]; 3565 if (opt == TCPOPT_EOL) 3566 break; 3567 if (opt == TCPOPT_NOP) 3568 optlen = 1; 3569 else { 3570 if (cnt < 2) 3571 break; 3572 optlen = cp[1]; 3573 if (optlen < 2 || optlen > cnt) 3574 break; 3575 } 3576 switch (opt) { 3577 case TCPOPT_MAXSEG: 3578 if (optlen != TCPOLEN_MAXSEG) 3579 continue; 3580 if (!(flags & TO_SYN)) 3581 continue; 3582 to->to_flags |= TOF_MSS; 3583 bcopy((char *)cp + 2, 3584 (char *)&to->to_mss, sizeof(to->to_mss)); 3585 to->to_mss = ntohs(to->to_mss); 3586 break; 3587 case TCPOPT_WINDOW: 3588 if (optlen != TCPOLEN_WINDOW) 3589 continue; 3590 if (!(flags & TO_SYN)) 3591 continue; 3592 to->to_flags |= TOF_SCALE; 3593 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3594 break; 3595 case TCPOPT_TIMESTAMP: 3596 if (optlen != TCPOLEN_TIMESTAMP) 3597 continue; 3598 to->to_flags |= TOF_TS; 3599 bcopy((char *)cp + 2, 3600 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3601 to->to_tsval = ntohl(to->to_tsval); 3602 bcopy((char *)cp + 6, 3603 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3604 to->to_tsecr = ntohl(to->to_tsecr); 3605 break; 3606 case TCPOPT_SIGNATURE: 3607 /* 3608 * In order to reply to a host which has set the 3609 * TCP_SIGNATURE option in its initial SYN, we have 3610 * to record the fact that the option was observed 3611 * here for the syncache code to perform the correct 3612 * response. 3613 */ 3614 if (optlen != TCPOLEN_SIGNATURE) 3615 continue; 3616 to->to_flags |= TOF_SIGNATURE; 3617 to->to_signature = cp + 2; 3618 break; 3619 case TCPOPT_SACK_PERMITTED: 3620 if (optlen != TCPOLEN_SACK_PERMITTED) 3621 continue; 3622 if (!(flags & TO_SYN)) 3623 continue; 3624 if (!V_tcp_do_sack) 3625 continue; 3626 to->to_flags |= TOF_SACKPERM; 3627 break; 3628 case TCPOPT_SACK: 3629 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3630 continue; 3631 if (flags & TO_SYN) 3632 continue; 3633 to->to_flags |= TOF_SACK; 3634 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3635 to->to_sacks = cp + 2; 3636 TCPSTAT_INC(tcps_sack_rcv_blocks); 3637 break; 3638 case TCPOPT_FAST_OPEN: 3639 /* 3640 * Cookie length validation is performed by the 3641 * server side cookie checking code or the client 3642 * side cookie cache update code. 3643 */ 3644 if (!(flags & TO_SYN)) 3645 continue; 3646 if (!V_tcp_fastopen_client_enable && 3647 !V_tcp_fastopen_server_enable) 3648 continue; 3649 to->to_flags |= TOF_FASTOPEN; 3650 to->to_tfo_len = optlen - 2; 3651 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3652 break; 3653 default: 3654 continue; 3655 } 3656 } 3657 } 3658 3659 /* 3660 * Pull out of band byte out of a segment so 3661 * it doesn't appear in the user's data queue. 3662 * It is still reflected in the segment length for 3663 * sequencing purposes. 3664 */ 3665 void 3666 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3667 int off) 3668 { 3669 int cnt = off + th->th_urp - 1; 3670 3671 while (cnt >= 0) { 3672 if (m->m_len > cnt) { 3673 char *cp = mtod(m, caddr_t) + cnt; 3674 struct tcpcb *tp = sototcpcb(so); 3675 3676 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3677 3678 tp->t_iobc = *cp; 3679 tp->t_oobflags |= TCPOOB_HAVEDATA; 3680 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3681 m->m_len--; 3682 if (m->m_flags & M_PKTHDR) 3683 m->m_pkthdr.len--; 3684 return; 3685 } 3686 cnt -= m->m_len; 3687 m = m->m_next; 3688 if (m == NULL) 3689 break; 3690 } 3691 panic("tcp_pulloutofband"); 3692 } 3693 3694 /* 3695 * Collect new round-trip time estimate 3696 * and update averages and current timeout. 3697 */ 3698 void 3699 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3700 { 3701 int delta; 3702 3703 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3704 3705 TCPSTAT_INC(tcps_rttupdated); 3706 if (tp->t_rttupdated < UCHAR_MAX) 3707 tp->t_rttupdated++; 3708 #ifdef STATS 3709 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3710 imax(0, rtt * 1000 / hz)); 3711 #endif 3712 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3713 /* 3714 * srtt is stored as fixed point with 5 bits after the 3715 * binary point (i.e., scaled by 8). The following magic 3716 * is equivalent to the smoothing algorithm in rfc793 with 3717 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3718 * point). Adjust rtt to origin 0. 3719 */ 3720 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3721 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3722 3723 if ((tp->t_srtt += delta) <= 0) 3724 tp->t_srtt = 1; 3725 3726 /* 3727 * We accumulate a smoothed rtt variance (actually, a 3728 * smoothed mean difference), then set the retransmit 3729 * timer to smoothed rtt + 4 times the smoothed variance. 3730 * rttvar is stored as fixed point with 4 bits after the 3731 * binary point (scaled by 16). The following is 3732 * equivalent to rfc793 smoothing with an alpha of .75 3733 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3734 * rfc793's wired-in beta. 3735 */ 3736 if (delta < 0) 3737 delta = -delta; 3738 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3739 if ((tp->t_rttvar += delta) <= 0) 3740 tp->t_rttvar = 1; 3741 } else { 3742 /* 3743 * No rtt measurement yet - use the unsmoothed rtt. 3744 * Set the variance to half the rtt (so our first 3745 * retransmit happens at 3*rtt). 3746 */ 3747 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3748 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3749 } 3750 tp->t_rtttime = 0; 3751 tp->t_rxtshift = 0; 3752 3753 /* 3754 * the retransmit should happen at rtt + 4 * rttvar. 3755 * Because of the way we do the smoothing, srtt and rttvar 3756 * will each average +1/2 tick of bias. When we compute 3757 * the retransmit timer, we want 1/2 tick of rounding and 3758 * 1 extra tick because of +-1/2 tick uncertainty in the 3759 * firing of the timer. The bias will give us exactly the 3760 * 1.5 tick we need. But, because the bias is 3761 * statistical, we have to test that we don't drop below 3762 * the minimum feasible timer (which is 2 ticks). 3763 */ 3764 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3765 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3766 3767 /* 3768 * We received an ack for a packet that wasn't retransmitted; 3769 * it is probably safe to discard any error indications we've 3770 * received recently. This isn't quite right, but close enough 3771 * for now (a route might have failed after we sent a segment, 3772 * and the return path might not be symmetrical). 3773 */ 3774 tp->t_softerror = 0; 3775 } 3776 3777 /* 3778 * Determine a reasonable value for maxseg size. 3779 * If the route is known, check route for mtu. 3780 * If none, use an mss that can be handled on the outgoing interface 3781 * without forcing IP to fragment. If no route is found, route has no mtu, 3782 * or the destination isn't local, use a default, hopefully conservative 3783 * size (usually 512 or the default IP max size, but no more than the mtu 3784 * of the interface), as we can't discover anything about intervening 3785 * gateways or networks. We also initialize the congestion/slow start 3786 * window to be a single segment if the destination isn't local. 3787 * While looking at the routing entry, we also initialize other path-dependent 3788 * parameters from pre-set or cached values in the routing entry. 3789 * 3790 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3791 * IP options, e.g. IPSEC data, since length of this data may vary, and 3792 * thus it is calculated for every segment separately in tcp_output(). 3793 * 3794 * NOTE that this routine is only called when we process an incoming 3795 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3796 * settings are handled in tcp_mssopt(). 3797 */ 3798 void 3799 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3800 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3801 { 3802 int mss = 0; 3803 uint32_t maxmtu = 0; 3804 struct inpcb *inp = tptoinpcb(tp); 3805 struct hc_metrics_lite metrics; 3806 #ifdef INET6 3807 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3808 size_t min_protoh = isipv6 ? 3809 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3810 sizeof (struct tcpiphdr); 3811 #else 3812 size_t min_protoh = sizeof(struct tcpiphdr); 3813 #endif 3814 3815 INP_WLOCK_ASSERT(inp); 3816 3817 if (tp->t_port) 3818 min_protoh += V_tcp_udp_tunneling_overhead; 3819 if (mtuoffer != -1) { 3820 KASSERT(offer == -1, ("%s: conflict", __func__)); 3821 offer = mtuoffer - min_protoh; 3822 } 3823 3824 /* Initialize. */ 3825 #ifdef INET6 3826 if (isipv6) { 3827 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3828 tp->t_maxseg = V_tcp_v6mssdflt; 3829 } 3830 #endif 3831 #if defined(INET) && defined(INET6) 3832 else 3833 #endif 3834 #ifdef INET 3835 { 3836 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3837 tp->t_maxseg = V_tcp_mssdflt; 3838 } 3839 #endif 3840 3841 /* 3842 * No route to sender, stay with default mss and return. 3843 */ 3844 if (maxmtu == 0) { 3845 /* 3846 * In case we return early we need to initialize metrics 3847 * to a defined state as tcp_hc_get() would do for us 3848 * if there was no cache hit. 3849 */ 3850 if (metricptr != NULL) 3851 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3852 return; 3853 } 3854 3855 /* What have we got? */ 3856 switch (offer) { 3857 case 0: 3858 /* 3859 * Offer == 0 means that there was no MSS on the SYN 3860 * segment, in this case we use tcp_mssdflt as 3861 * already assigned to t_maxseg above. 3862 */ 3863 offer = tp->t_maxseg; 3864 break; 3865 3866 case -1: 3867 /* 3868 * Offer == -1 means that we didn't receive SYN yet. 3869 */ 3870 /* FALLTHROUGH */ 3871 3872 default: 3873 /* 3874 * Prevent DoS attack with too small MSS. Round up 3875 * to at least minmss. 3876 */ 3877 offer = max(offer, V_tcp_minmss); 3878 } 3879 3880 /* 3881 * rmx information is now retrieved from tcp_hostcache. 3882 */ 3883 tcp_hc_get(&inp->inp_inc, &metrics); 3884 if (metricptr != NULL) 3885 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3886 3887 /* 3888 * If there's a discovered mtu in tcp hostcache, use it. 3889 * Else, use the link mtu. 3890 */ 3891 if (metrics.rmx_mtu) 3892 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3893 else { 3894 #ifdef INET6 3895 if (isipv6) { 3896 mss = maxmtu - min_protoh; 3897 if (!V_path_mtu_discovery && 3898 !in6_localaddr(&inp->in6p_faddr)) 3899 mss = min(mss, V_tcp_v6mssdflt); 3900 } 3901 #endif 3902 #if defined(INET) && defined(INET6) 3903 else 3904 #endif 3905 #ifdef INET 3906 { 3907 mss = maxmtu - min_protoh; 3908 if (!V_path_mtu_discovery && 3909 !in_localaddr(inp->inp_faddr)) 3910 mss = min(mss, V_tcp_mssdflt); 3911 } 3912 #endif 3913 /* 3914 * XXX - The above conditional (mss = maxmtu - min_protoh) 3915 * probably violates the TCP spec. 3916 * The problem is that, since we don't know the 3917 * other end's MSS, we are supposed to use a conservative 3918 * default. But, if we do that, then MTU discovery will 3919 * never actually take place, because the conservative 3920 * default is much less than the MTUs typically seen 3921 * on the Internet today. For the moment, we'll sweep 3922 * this under the carpet. 3923 * 3924 * The conservative default might not actually be a problem 3925 * if the only case this occurs is when sending an initial 3926 * SYN with options and data to a host we've never talked 3927 * to before. Then, they will reply with an MSS value which 3928 * will get recorded and the new parameters should get 3929 * recomputed. For Further Study. 3930 */ 3931 } 3932 mss = min(mss, offer); 3933 3934 /* 3935 * Sanity check: make sure that maxseg will be large 3936 * enough to allow some data on segments even if the 3937 * all the option space is used (40bytes). Otherwise 3938 * funny things may happen in tcp_output. 3939 * 3940 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3941 */ 3942 mss = max(mss, 64); 3943 3944 tp->t_maxseg = mss; 3945 if (tp->t_maxseg < V_tcp_mssdflt) { 3946 /* 3947 * The MSS is so small we should not process incoming 3948 * SACK's since we are subject to attack in such a 3949 * case. 3950 */ 3951 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 3952 } else { 3953 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 3954 } 3955 3956 } 3957 3958 void 3959 tcp_mss(struct tcpcb *tp, int offer) 3960 { 3961 int mss; 3962 uint32_t bufsize; 3963 struct inpcb *inp = tptoinpcb(tp); 3964 struct socket *so; 3965 struct hc_metrics_lite metrics; 3966 struct tcp_ifcap cap; 3967 3968 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3969 3970 bzero(&cap, sizeof(cap)); 3971 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3972 3973 mss = tp->t_maxseg; 3974 3975 /* 3976 * If there's a pipesize, change the socket buffer to that size, 3977 * don't change if sb_hiwat is different than default (then it 3978 * has been changed on purpose with setsockopt). 3979 * Make the socket buffers an integral number of mss units; 3980 * if the mss is larger than the socket buffer, decrease the mss. 3981 */ 3982 so = inp->inp_socket; 3983 SOCKBUF_LOCK(&so->so_snd); 3984 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3985 bufsize = metrics.rmx_sendpipe; 3986 else 3987 bufsize = so->so_snd.sb_hiwat; 3988 if (bufsize < mss) 3989 mss = bufsize; 3990 else { 3991 bufsize = roundup(bufsize, mss); 3992 if (bufsize > sb_max) 3993 bufsize = sb_max; 3994 if (bufsize > so->so_snd.sb_hiwat) 3995 (void)sbreserve_locked(so, SO_SND, bufsize, NULL); 3996 } 3997 SOCKBUF_UNLOCK(&so->so_snd); 3998 /* 3999 * Sanity check: make sure that maxseg will be large 4000 * enough to allow some data on segments even if the 4001 * all the option space is used (40bytes). Otherwise 4002 * funny things may happen in tcp_output. 4003 * 4004 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 4005 */ 4006 tp->t_maxseg = max(mss, 64); 4007 if (tp->t_maxseg < V_tcp_mssdflt) { 4008 /* 4009 * The MSS is so small we should not process incoming 4010 * SACK's since we are subject to attack in such a 4011 * case. 4012 */ 4013 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 4014 } else { 4015 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 4016 } 4017 4018 SOCKBUF_LOCK(&so->so_rcv); 4019 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 4020 bufsize = metrics.rmx_recvpipe; 4021 else 4022 bufsize = so->so_rcv.sb_hiwat; 4023 if (bufsize > mss) { 4024 bufsize = roundup(bufsize, mss); 4025 if (bufsize > sb_max) 4026 bufsize = sb_max; 4027 if (bufsize > so->so_rcv.sb_hiwat) 4028 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL); 4029 } 4030 SOCKBUF_UNLOCK(&so->so_rcv); 4031 4032 /* Check the interface for TSO capabilities. */ 4033 if (cap.ifcap & CSUM_TSO) { 4034 tp->t_flags |= TF_TSO; 4035 tp->t_tsomax = cap.tsomax; 4036 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 4037 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 4038 if (cap.ipsec_tso) 4039 tp->t_flags2 |= TF2_IPSEC_TSO; 4040 } 4041 } 4042 4043 /* 4044 * Determine the MSS option to send on an outgoing SYN. 4045 */ 4046 int 4047 tcp_mssopt(struct in_conninfo *inc) 4048 { 4049 int mss = 0; 4050 uint32_t thcmtu = 0; 4051 uint32_t maxmtu = 0; 4052 size_t min_protoh; 4053 4054 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 4055 4056 #ifdef INET6 4057 if (inc->inc_flags & INC_ISIPV6) { 4058 mss = V_tcp_v6mssdflt; 4059 maxmtu = tcp_maxmtu6(inc, NULL); 4060 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 4061 } 4062 #endif 4063 #if defined(INET) && defined(INET6) 4064 else 4065 #endif 4066 #ifdef INET 4067 { 4068 mss = V_tcp_mssdflt; 4069 maxmtu = tcp_maxmtu(inc, NULL); 4070 min_protoh = sizeof(struct tcpiphdr); 4071 } 4072 #endif 4073 #if defined(INET6) || defined(INET) 4074 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 4075 #endif 4076 4077 if (maxmtu && thcmtu) 4078 mss = min(maxmtu, thcmtu) - min_protoh; 4079 else if (maxmtu || thcmtu) 4080 mss = max(maxmtu, thcmtu) - min_protoh; 4081 4082 return (mss); 4083 } 4084 4085 void 4086 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, 4087 sackstatus_t sack_changed, u_int *maxsegp) 4088 { 4089 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 4090 u_int maxseg; 4091 4092 INP_WLOCK_ASSERT(tptoinpcb(tp)); 4093 4094 if (*maxsegp == 0) { 4095 *maxsegp = tcp_maxseg(tp); 4096 } 4097 maxseg = *maxsegp; 4098 /* 4099 * Compute the amount of data that this ACK is indicating 4100 * (del_data) and an estimate of how many bytes are in the 4101 * network. 4102 */ 4103 if (tcp_is_sack_recovery(tp, to) || 4104 (IN_CONGRECOVERY(tp->t_flags) && 4105 !IN_FASTRECOVERY(tp->t_flags))) { 4106 del_data = tp->sackhint.delivered_data; 4107 if (V_tcp_do_newsack) 4108 pipe = tcp_compute_pipe(tp); 4109 else 4110 pipe = (tp->snd_nxt - tp->snd_fack) + 4111 tp->sackhint.sack_bytes_rexmit; 4112 } else { 4113 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + 4114 tp->snd_recover - tp->snd_una)) { 4115 del_data = maxseg; 4116 } 4117 pipe = imax(0, tp->snd_max - tp->snd_una - 4118 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); 4119 } 4120 tp->sackhint.prr_delivered += del_data; 4121 /* 4122 * Proportional Rate Reduction 4123 */ 4124 if (pipe >= tp->snd_ssthresh) { 4125 if (tp->sackhint.recover_fs == 0) 4126 tp->sackhint.recover_fs = 4127 imax(1, tp->snd_nxt - tp->snd_una); 4128 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 4129 tp->snd_ssthresh, tp->sackhint.recover_fs) - 4130 tp->sackhint.prr_out + maxseg - 1; 4131 } else { 4132 /* 4133 * PRR 6937bis heuristic: 4134 * - A partial ack without SACK block beneath snd_recover 4135 * indicates further loss. 4136 * - An SACK scoreboard update adding a new hole indicates 4137 * further loss, so be conservative and send at most one 4138 * segment. 4139 * - Prevent ACK splitting attacks, by being conservative 4140 * when no new data is acked. 4141 */ 4142 if ((sack_changed == SACK_NEWLOSS) || (del_data == 0)) { 4143 limit = tp->sackhint.prr_delivered - 4144 tp->sackhint.prr_out; 4145 } else { 4146 limit = imax(tp->sackhint.prr_delivered - 4147 tp->sackhint.prr_out, del_data) + 4148 maxseg; 4149 } 4150 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 4151 } 4152 snd_cnt = imax(snd_cnt, 0) / maxseg; 4153 /* 4154 * Send snd_cnt new data into the network in response to this ack. 4155 * If there is going to be a SACK retransmission, adjust snd_cwnd 4156 * accordingly. 4157 */ 4158 if (IN_FASTRECOVERY(tp->t_flags)) { 4159 if (tcp_is_sack_recovery(tp, to)) { 4160 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4161 } else { 4162 tp->snd_cwnd = (tp->snd_max - tp->snd_una) + 4163 (snd_cnt * maxseg); 4164 } 4165 } else if (IN_CONGRECOVERY(tp->t_flags)) { 4166 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4167 } 4168 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); 4169 } 4170 4171 /* 4172 * On a partial ack arrives, force the retransmission of the 4173 * next unacknowledged segment. Do not clear tp->t_dupacks. 4174 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4175 * be started again. 4176 */ 4177 void 4178 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4179 { 4180 tcp_seq onxt = tp->snd_nxt; 4181 uint32_t ocwnd = tp->snd_cwnd; 4182 u_int maxseg = tcp_maxseg(tp); 4183 4184 INP_WLOCK_ASSERT(tptoinpcb(tp)); 4185 4186 tcp_timer_activate(tp, TT_REXMT, 0); 4187 tp->t_rtttime = 0; 4188 if (IN_FASTRECOVERY(tp->t_flags)) { 4189 tp->snd_nxt = th->th_ack; 4190 /* 4191 * Set snd_cwnd to one segment beyond acknowledged offset. 4192 * (tp->snd_una has not yet been updated when this function is called.) 4193 */ 4194 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4195 tp->t_flags |= TF_ACKNOW; 4196 (void) tcp_output(tp); 4197 tp->snd_cwnd = ocwnd; 4198 if (SEQ_GT(onxt, tp->snd_nxt)) 4199 tp->snd_nxt = onxt; 4200 } 4201 /* 4202 * Partial window deflation. Relies on fact that tp->snd_una 4203 * not updated yet. 4204 */ 4205 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4206 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4207 else 4208 tp->snd_cwnd = 0; 4209 tp->snd_cwnd += maxseg; 4210 } 4211 4212 int 4213 tcp_compute_pipe(struct tcpcb *tp) 4214 { 4215 if (tp->t_fb->tfb_compute_pipe == NULL) { 4216 return (tp->snd_max - tp->snd_una + 4217 tp->sackhint.sack_bytes_rexmit - 4218 tp->sackhint.sacked_bytes - 4219 tp->sackhint.lost_bytes); 4220 } else { 4221 return((*tp->t_fb->tfb_compute_pipe)(tp)); 4222 } 4223 } 4224 4225 uint32_t 4226 tcp_compute_initwnd(uint32_t maxseg) 4227 { 4228 /* 4229 * Calculate the Initial Window, also used as Restart Window 4230 * 4231 * RFC5681 Section 3.1 specifies the default conservative values. 4232 * RFC3390 specifies slightly more aggressive values. 4233 * RFC6928 increases it to ten segments. 4234 * Support for user specified value for initial flight size. 4235 */ 4236 if (V_tcp_initcwnd_segments) 4237 return min(V_tcp_initcwnd_segments * maxseg, 4238 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4239 else if (V_tcp_do_rfc3390) 4240 return min(4 * maxseg, max(2 * maxseg, 4380)); 4241 else { 4242 /* Per RFC5681 Section 3.1 */ 4243 if (maxseg > 2190) 4244 return (2 * maxseg); 4245 else if (maxseg > 1095) 4246 return (3 * maxseg); 4247 else 4248 return (4 * maxseg); 4249 } 4250 } 4251