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 "opt_inet.h" 51 #include "opt_inet6.h" 52 #include "opt_ipsec.h" 53 #include "opt_rss.h" 54 55 #include <sys/param.h> 56 #include <sys/arb.h> 57 #include <sys/kernel.h> 58 #ifdef TCP_HHOOK 59 #include <sys/hhook.h> 60 #endif 61 #include <sys/malloc.h> 62 #include <sys/mbuf.h> 63 #include <sys/proc.h> /* for proc0 declaration */ 64 #include <sys/protosw.h> 65 #include <sys/qmath.h> 66 #include <sys/sdt.h> 67 #include <sys/signalvar.h> 68 #include <sys/socket.h> 69 #include <sys/socketvar.h> 70 #include <sys/sysctl.h> 71 #include <sys/syslog.h> 72 #include <sys/systm.h> 73 #include <sys/stats.h> 74 75 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 76 77 #include <vm/uma.h> 78 79 #include <net/if.h> 80 #include <net/if_var.h> 81 #include <net/route.h> 82 #include <net/rss_config.h> 83 #include <net/vnet.h> 84 85 #define TCPSTATES /* for logging */ 86 87 #include <netinet/in.h> 88 #include <netinet/in_kdtrace.h> 89 #include <netinet/in_pcb.h> 90 #include <netinet/in_rss.h> 91 #include <netinet/in_systm.h> 92 #include <netinet/ip.h> 93 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 94 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 95 #include <netinet/ip_var.h> 96 #include <netinet/ip_options.h> 97 #include <netinet/ip6.h> 98 #include <netinet/icmp6.h> 99 #include <netinet6/in6_pcb.h> 100 #include <netinet6/in6_rss.h> 101 #include <netinet6/in6_var.h> 102 #include <netinet6/ip6_var.h> 103 #include <netinet6/nd6.h> 104 #include <netinet/tcp.h> 105 #include <netinet/tcp_fsm.h> 106 #include <netinet/tcp_seq.h> 107 #include <netinet/tcp_timer.h> 108 #include <netinet/tcp_var.h> 109 #include <netinet/tcp_log_buf.h> 110 #include <netinet6/tcp6_var.h> 111 #include <netinet/tcpip.h> 112 #include <netinet/cc/cc.h> 113 #include <netinet/tcp_fastopen.h> 114 #include <netinet/tcp_syncache.h> 115 #ifdef TCP_OFFLOAD 116 #include <netinet/tcp_offload.h> 117 #endif 118 #include <netinet/tcp_ecn.h> 119 #include <netinet/udp.h> 120 121 #include <netipsec/ipsec_support.h> 122 123 #include <machine/in_cksum.h> 124 125 #include <security/mac/mac_framework.h> 126 127 const int tcprexmtthresh = 3; 128 129 VNET_DEFINE(int, tcp_log_in_vain) = 0; 130 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 131 &VNET_NAME(tcp_log_in_vain), 0, 132 "Log all incoming TCP segments to closed ports"); 133 134 VNET_DEFINE(int, tcp_bind_all_fibs) = 1; 135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, bind_all_fibs, CTLFLAG_VNET | CTLFLAG_RDTUN, 136 &VNET_NAME(tcp_bind_all_fibs), 0, 137 "Bound sockets receive traffic from all FIBs"); 138 139 VNET_DEFINE(int, blackhole) = 0; 140 #define V_blackhole VNET(blackhole) 141 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 142 &VNET_NAME(blackhole), 0, 143 "Do not send RST on segments to closed ports"); 144 145 VNET_DEFINE(bool, blackhole_local) = false; 146 #define V_blackhole_local VNET(blackhole_local) 147 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET | 148 CTLFLAG_RW, &VNET_NAME(blackhole_local), false, 149 "Enforce net.inet.tcp.blackhole for locally originated packets"); 150 151 VNET_DEFINE(int, tcp_delack_enabled) = 1; 152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 153 &VNET_NAME(tcp_delack_enabled), 0, 154 "Delay ACK to try and piggyback it onto a data packet"); 155 156 VNET_DEFINE(int, drop_synfin) = 0; 157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 158 &VNET_NAME(drop_synfin), 0, 159 "Drop TCP packets with SYN+FIN set"); 160 161 VNET_DEFINE(int, tcp_do_prr) = 1; 162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 163 &VNET_NAME(tcp_do_prr), 1, 164 "Enable Proportional Rate Reduction per RFC 6937"); 165 166 VNET_DEFINE(int, tcp_do_newcwv) = 0; 167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 168 &VNET_NAME(tcp_do_newcwv), 0, 169 "Enable New Congestion Window Validation per RFC7661"); 170 171 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 173 &VNET_NAME(tcp_do_rfc3042), 0, 174 "Enable RFC 3042 (Limited Transmit)"); 175 176 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 178 &VNET_NAME(tcp_do_rfc3390), 0, 179 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 180 181 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 183 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 184 "Slow-start flight size (initial congestion window) in number of segments"); 185 186 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 188 &VNET_NAME(tcp_do_rfc3465), 0, 189 "Enable RFC 3465 (Appropriate Byte Counting)"); 190 191 VNET_DEFINE(int, tcp_abc_l_var) = 2; 192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 193 &VNET_NAME(tcp_abc_l_var), 2, 194 "Cap the max cwnd increment during slow-start to this number of segments"); 195 196 VNET_DEFINE(int, tcp_insecure_syn) = 0; 197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 198 &VNET_NAME(tcp_insecure_syn), 0, 199 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 200 201 VNET_DEFINE(int, tcp_insecure_rst) = 0; 202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 203 &VNET_NAME(tcp_insecure_rst), 0, 204 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 205 206 VNET_DEFINE(int, tcp_insecure_ack) = 0; 207 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_ack, CTLFLAG_VNET | CTLFLAG_RW, 208 &VNET_NAME(tcp_insecure_ack), 0, 209 "Follow RFC793 criteria for validating SEG.ACK"); 210 211 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 212 #define V_tcp_recvspace VNET(tcp_recvspace) 213 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 214 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 215 216 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 218 &VNET_NAME(tcp_do_autorcvbuf), 0, 219 "Enable automatic receive buffer sizing"); 220 221 VNET_DEFINE(int, tcp_autorcvbuf_max) = 8*1024*1024; 222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 223 &VNET_NAME(tcp_autorcvbuf_max), 0, 224 "Max size of automatic receive buffer"); 225 226 VNET_DEFINE(struct inpcbinfo, tcbinfo); 227 228 /* 229 * TCP statistics are stored in an array of counter(9)s, which size matches 230 * size of struct tcpstat. TCP running connection count is a regular array. 231 */ 232 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 233 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 234 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 235 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 236 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 237 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 238 "TCP connection counts by TCP state"); 239 240 /* 241 * Kernel module interface for updating tcpstat. The first argument is an index 242 * into tcpstat treated as an array. 243 */ 244 void 245 kmod_tcpstat_add(int statnum, int val) 246 { 247 248 counter_u64_add(VNET(tcpstat)[statnum], val); 249 } 250 251 /* 252 * Make sure that we only start a SACK loss recovery when 253 * receiving a duplicate ACK with a SACK block, and also 254 * complete SACK loss recovery in case the other end 255 * reneges. 256 */ 257 static bool inline 258 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to) 259 { 260 return ((tp->t_flags & TF_SACK_PERMIT) && 261 ((to->to_flags & TOF_SACK) || 262 (!TAILQ_EMPTY(&tp->snd_holes)))); 263 } 264 265 #ifdef TCP_HHOOK 266 /* 267 * Wrapper for the TCP established input helper hook. 268 */ 269 void 270 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 271 { 272 struct tcp_hhook_data hhook_data; 273 274 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 275 hhook_data.tp = tp; 276 hhook_data.th = th; 277 hhook_data.to = to; 278 279 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 280 &tp->t_osd); 281 } 282 } 283 #endif 284 285 /* 286 * CC wrapper hook functions 287 */ 288 void 289 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 290 uint16_t type) 291 { 292 #ifdef STATS 293 int32_t gput; 294 #endif 295 296 INP_WLOCK_ASSERT(tptoinpcb(tp)); 297 298 tp->t_ccv.nsegs = nsegs; 299 tp->t_ccv.bytes_this_ack = BYTES_THIS_ACK(tp, th); 300 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 301 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 302 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 303 tp->t_ccv.flags |= CCF_CWND_LIMITED; 304 else 305 tp->t_ccv.flags &= ~CCF_CWND_LIMITED; 306 307 if (type == CC_ACK) { 308 #ifdef STATS 309 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 310 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 311 if (!IN_RECOVERY(tp->t_flags)) 312 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 313 tp->t_ccv.bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 314 if ((tp->t_flags & TF_GPUTINPROG) && 315 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 316 /* 317 * Compute goodput in bits per millisecond. 318 */ 319 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) / 320 max(1, tcp_ts_getticks() - tp->gput_ts); 321 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 322 gput); 323 /* 324 * XXXLAS: This is a temporary hack, and should be 325 * chained off VOI_TCP_GPUT when stats(9) grows an API 326 * to deal with chained VOIs. 327 */ 328 if (tp->t_stats_gput_prev > 0) 329 stats_voi_update_abs_s32(tp->t_stats, 330 VOI_TCP_GPUT_ND, 331 ((gput - tp->t_stats_gput_prev) * 100) / 332 tp->t_stats_gput_prev); 333 tp->t_flags &= ~TF_GPUTINPROG; 334 tp->t_stats_gput_prev = gput; 335 } 336 #endif /* STATS */ 337 if (tp->snd_cwnd > tp->snd_ssthresh) { 338 tp->t_bytes_acked += tp->t_ccv.bytes_this_ack; 339 if (tp->t_bytes_acked >= tp->snd_cwnd) { 340 tp->t_bytes_acked -= tp->snd_cwnd; 341 tp->t_ccv.flags |= CCF_ABC_SENTAWND; 342 } 343 } else { 344 tp->t_ccv.flags &= ~CCF_ABC_SENTAWND; 345 tp->t_bytes_acked = 0; 346 } 347 } 348 349 if (CC_ALGO(tp)->ack_received != NULL) { 350 /* XXXLAS: Find a way to live without this */ 351 tp->t_ccv.curack = th->th_ack; 352 CC_ALGO(tp)->ack_received(&tp->t_ccv, type); 353 } 354 #ifdef STATS 355 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 356 #endif 357 } 358 359 void 360 cc_conn_init(struct tcpcb *tp) 361 { 362 struct hc_metrics_lite metrics; 363 struct inpcb *inp = tptoinpcb(tp); 364 u_int maxseg; 365 int rtt; 366 367 INP_WLOCK_ASSERT(inp); 368 369 tcp_hc_get(&inp->inp_inc, &metrics); 370 maxseg = tcp_maxseg(tp); 371 372 if (tp->t_srtt == 0 && (rtt = metrics.hc_rtt)) { 373 tp->t_srtt = rtt; 374 TCPSTAT_INC(tcps_usedrtt); 375 if (metrics.hc_rttvar) { 376 tp->t_rttvar = metrics.hc_rttvar; 377 TCPSTAT_INC(tcps_usedrttvar); 378 } else { 379 /* default variation is +- 1 rtt */ 380 tp->t_rttvar = 381 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 382 } 383 TCPT_RANGESET(tp->t_rxtcur, 384 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 385 tp->t_rttmin, tcp_rexmit_max); 386 } 387 if (metrics.hc_ssthresh) { 388 /* 389 * There's some sort of gateway or interface 390 * buffer limit on the path. Use this to set 391 * the slow start threshold, but set the 392 * threshold to no less than 2*mss. 393 */ 394 tp->snd_ssthresh = max(2 * maxseg, metrics.hc_ssthresh); 395 TCPSTAT_INC(tcps_usedssthresh); 396 } 397 398 /* 399 * Set the initial slow-start flight size. 400 * 401 * If a SYN or SYN/ACK was lost and retransmitted, we have to 402 * reduce the initial CWND to one segment as congestion is likely 403 * requiring us to be cautious. 404 */ 405 if (tp->snd_cwnd == 1) 406 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 407 else 408 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 409 410 if (CC_ALGO(tp)->conn_init != NULL) 411 CC_ALGO(tp)->conn_init(&tp->t_ccv); 412 } 413 414 void inline 415 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 416 { 417 INP_WLOCK_ASSERT(tptoinpcb(tp)); 418 419 #ifdef STATS 420 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 421 #endif 422 423 switch(type) { 424 case CC_NDUPACK: 425 if (!IN_FASTRECOVERY(tp->t_flags)) { 426 tp->snd_recover = tp->snd_max; 427 if (tp->t_flags2 & TF2_ECN_PERMIT) 428 tp->t_flags2 |= TF2_ECN_SND_CWR; 429 } 430 break; 431 case CC_ECN: 432 if (!IN_CONGRECOVERY(tp->t_flags) || 433 /* 434 * Allow ECN reaction on ACK to CWR, if 435 * that data segment was also CE marked. 436 */ 437 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 438 EXIT_CONGRECOVERY(tp->t_flags); 439 TCPSTAT_INC(tcps_ecn_rcwnd); 440 tp->snd_recover = tp->snd_max + 1; 441 if (tp->t_flags2 & TF2_ECN_PERMIT) 442 tp->t_flags2 |= TF2_ECN_SND_CWR; 443 } 444 break; 445 case CC_RTO: 446 tp->t_dupacks = 0; 447 tp->t_bytes_acked = 0; 448 EXIT_RECOVERY(tp->t_flags); 449 if (tp->t_flags2 & TF2_ECN_PERMIT) 450 tp->t_flags2 |= TF2_ECN_SND_CWR; 451 break; 452 case CC_RTO_ERR: 453 TCPSTAT_INC(tcps_sndrexmitbad); 454 /* RTO was unnecessary, so reset everything. */ 455 tp->snd_cwnd = tp->snd_cwnd_prev; 456 tp->snd_ssthresh = tp->snd_ssthresh_prev; 457 tp->snd_recover = tp->snd_recover_prev; 458 if (tp->t_flags & TF_WASFRECOVERY) 459 ENTER_FASTRECOVERY(tp->t_flags); 460 if (tp->t_flags & TF_WASCRECOVERY) 461 ENTER_CONGRECOVERY(tp->t_flags); 462 tp->snd_nxt = tp->snd_max; 463 tp->t_flags &= ~TF_PREVVALID; 464 tp->t_rxtshift = 0; 465 tp->t_badrxtwin = 0; 466 break; 467 } 468 if (SEQ_LT(tp->snd_fack, tp->snd_una) || 469 SEQ_GT(tp->snd_fack, tp->snd_max)) { 470 tp->snd_fack = tp->snd_una; 471 } 472 473 if (CC_ALGO(tp)->cong_signal != NULL) { 474 if (th != NULL) 475 tp->t_ccv.curack = th->th_ack; 476 CC_ALGO(tp)->cong_signal(&tp->t_ccv, type); 477 } 478 } 479 480 void inline 481 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 482 { 483 INP_WLOCK_ASSERT(tptoinpcb(tp)); 484 485 if (CC_ALGO(tp)->post_recovery != NULL) { 486 if (SEQ_LT(tp->snd_fack, th->th_ack) || 487 SEQ_GT(tp->snd_fack, tp->snd_max)) { 488 tp->snd_fack = th->th_ack; 489 } 490 tp->t_ccv.curack = th->th_ack; 491 CC_ALGO(tp)->post_recovery(&tp->t_ccv); 492 } 493 EXIT_RECOVERY(tp->t_flags); 494 495 tp->t_bytes_acked = 0; 496 tp->sackhint.delivered_data = 0; 497 tp->sackhint.prr_delivered = 0; 498 tp->sackhint.prr_out = 0; 499 } 500 501 /* 502 * Indicate whether this ack should be delayed. We can delay the ack if 503 * following conditions are met: 504 * - There is no delayed ack timer in progress. 505 * - Our last ack wasn't a 0-sized window. We never want to delay 506 * the ack that opens up a 0-sized window. 507 * - LRO wasn't used for this segment. We make sure by checking that the 508 * segment size is not larger than the MSS. 509 */ 510 #define DELAY_ACK(tp, tlen) \ 511 ((!tcp_timer_active(tp, TT_DELACK) && \ 512 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 513 (tlen <= tp->t_maxseg) && \ 514 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 515 516 void inline 517 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) 518 { 519 INP_WLOCK_ASSERT(tptoinpcb(tp)); 520 521 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 522 switch (iptos & IPTOS_ECN_MASK) { 523 case IPTOS_ECN_CE: 524 tp->t_ccv.flags |= CCF_IPHDR_CE; 525 break; 526 case IPTOS_ECN_ECT0: 527 /* FALLTHROUGH */ 528 case IPTOS_ECN_ECT1: 529 /* FALLTHROUGH */ 530 case IPTOS_ECN_NOTECT: 531 tp->t_ccv.flags &= ~CCF_IPHDR_CE; 532 break; 533 } 534 535 if (flags & TH_CWR) 536 tp->t_ccv.flags |= CCF_TCPHDR_CWR; 537 else 538 tp->t_ccv.flags &= ~CCF_TCPHDR_CWR; 539 540 CC_ALGO(tp)->ecnpkt_handler(&tp->t_ccv); 541 542 if (((tp->t_state == TCPS_ESTABLISHED) || 543 (tp->t_state == TCPS_FIN_WAIT_1) || 544 (tp->t_state == TCPS_FIN_WAIT_2)) && 545 (tp->t_ccv.flags & CCF_ACKNOW)) { 546 tp->t_flags |= TF_ACKNOW; 547 tp->t_ccv.flags &= ~CCF_ACKNOW; 548 } 549 } 550 } 551 552 void inline 553 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 554 { 555 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos); 556 } 557 558 /* 559 * TCP input handling is split into multiple parts: 560 * tcp6_input is a thin wrapper around tcp_input for the extended 561 * ip6_protox[] call format in ip6_input 562 * tcp_input handles primary segment validation, inpcb lookup and 563 * SYN processing on listen sockets 564 * tcp_do_segment processes the ACK and text of the segment for 565 * establishing, established and closing connections 566 */ 567 #ifdef INET6 568 int 569 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 570 { 571 struct mbuf *m; 572 573 m = *mp; 574 if (m->m_len < *offp + sizeof(struct tcphdr)) { 575 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 576 if (m == NULL) { 577 *mp = m; 578 TCPSTAT_INC(tcps_rcvshort); 579 return (IPPROTO_DONE); 580 } 581 } 582 583 *mp = m; 584 return (tcp_input_with_port(mp, offp, proto, port)); 585 } 586 587 int 588 tcp6_input(struct mbuf **mp, int *offp, int proto) 589 { 590 591 return(tcp6_input_with_port(mp, offp, proto, 0)); 592 } 593 #endif /* INET6 */ 594 595 int 596 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 597 { 598 struct mbuf *m = *mp; 599 struct tcphdr *th = NULL; 600 struct ip *ip = NULL; 601 struct inpcb *inp = NULL; 602 struct tcpcb *tp = NULL; 603 struct socket *so = NULL; 604 u_char *optp = NULL; 605 int off0; 606 int optlen = 0; 607 #ifdef INET 608 int len; 609 uint8_t ipttl; 610 #endif 611 int tlen = 0, off; 612 int drop_hdrlen; 613 int thflags; 614 int lookupflag; 615 uint8_t iptos; 616 struct m_tag *fwd_tag = NULL; 617 #ifdef INET6 618 struct ip6_hdr *ip6 = NULL; 619 int isipv6; 620 #else 621 const void *ip6 = NULL; 622 #endif /* INET6 */ 623 struct tcpopt to; /* options in this segment */ 624 char *s = NULL; /* address and port logging */ 625 bool closed_port = false; /* segment is hitting a closed port */ 626 627 NET_EPOCH_ASSERT(); 628 629 #ifdef INET6 630 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 631 #endif 632 633 off0 = *offp; 634 m = *mp; 635 *mp = NULL; 636 to.to_flags = 0; 637 TCPSTAT_INC(tcps_rcvtotal); 638 639 m->m_pkthdr.tcp_tun_port = port; 640 #ifdef INET6 641 if (isipv6) { 642 ip6 = mtod(m, struct ip6_hdr *); 643 th = (struct tcphdr *)((caddr_t)ip6 + off0); 644 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 645 if (port) 646 goto skip6_csum; 647 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 648 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 649 th->th_sum = m->m_pkthdr.csum_data; 650 else 651 th->th_sum = in6_cksum_pseudo(ip6, tlen, 652 IPPROTO_TCP, m->m_pkthdr.csum_data); 653 th->th_sum ^= 0xffff; 654 } else if (m->m_pkthdr.csum_flags & CSUM_IP6_TCP) { 655 /* 656 * Packet from local host (maybe from a VM). 657 * Checksum not required. 658 */ 659 th->th_sum = 0; 660 } else 661 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 662 if (th->th_sum) { 663 TCPSTAT_INC(tcps_rcvbadsum); 664 goto drop; 665 } 666 skip6_csum: 667 /* 668 * Be proactive about unspecified IPv6 address in source. 669 * As we use all-zero to indicate unbounded/unconnected pcb, 670 * unspecified IPv6 address can be used to confuse us. 671 * 672 * Note that packets with unspecified IPv6 destination is 673 * already dropped in ip6_input. 674 */ 675 KASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_dst), 676 ("%s: unspecified destination v6 address", __func__)); 677 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 678 IP6STAT_INC(ip6s_badscope); /* XXX */ 679 goto drop; 680 } 681 iptos = IPV6_TRAFFIC_CLASS(ip6); 682 } 683 #endif 684 #if defined(INET) && defined(INET6) 685 else 686 #endif 687 #ifdef INET 688 { 689 /* 690 * Get IP and TCP header together in first mbuf. 691 * Note: IP leaves IP header in first mbuf. 692 */ 693 if (off0 > sizeof (struct ip)) { 694 ip_stripoptions(m); 695 off0 = sizeof(struct ip); 696 } 697 if (m->m_len < sizeof (struct tcpiphdr)) { 698 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 699 == NULL) { 700 TCPSTAT_INC(tcps_rcvshort); 701 return (IPPROTO_DONE); 702 } 703 } 704 ip = mtod(m, struct ip *); 705 th = (struct tcphdr *)((caddr_t)ip + off0); 706 tlen = ntohs(ip->ip_len) - off0; 707 708 iptos = ip->ip_tos; 709 if (port) 710 goto skip_csum; 711 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 712 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 713 th->th_sum = m->m_pkthdr.csum_data; 714 else 715 th->th_sum = in_pseudo(ip->ip_src.s_addr, 716 ip->ip_dst.s_addr, 717 htonl(m->m_pkthdr.csum_data + tlen + 718 IPPROTO_TCP)); 719 th->th_sum ^= 0xffff; 720 } else if (m->m_pkthdr.csum_flags & CSUM_IP_TCP) { 721 /* 722 * Packet from local host (maybe from a VM). 723 * Checksum not required. 724 */ 725 th->th_sum = 0; 726 } else { 727 struct ipovly *ipov = (struct ipovly *)ip; 728 729 /* 730 * Checksum extended TCP header and data. 731 */ 732 len = off0 + tlen; 733 ipttl = ip->ip_ttl; 734 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 735 ipov->ih_len = htons(tlen); 736 th->th_sum = in_cksum(m, len); 737 /* Reset length for SDT probes. */ 738 ip->ip_len = htons(len); 739 /* Reset TOS bits */ 740 ip->ip_tos = iptos; 741 /* Re-initialization for later version check */ 742 ip->ip_ttl = ipttl; 743 ip->ip_v = IPVERSION; 744 ip->ip_hl = off0 >> 2; 745 } 746 skip_csum: 747 if (th->th_sum && (port == 0)) { 748 TCPSTAT_INC(tcps_rcvbadsum); 749 goto drop; 750 } 751 KASSERT(ip->ip_dst.s_addr != INADDR_ANY, 752 ("%s: unspecified destination v4 address", __func__)); 753 if (__predict_false(ip->ip_src.s_addr == INADDR_ANY)) { 754 IPSTAT_INC(ips_badaddr); 755 goto drop; 756 } 757 } 758 #endif /* INET */ 759 760 /* 761 * Check that TCP offset makes sense, 762 * pull out TCP options and adjust length. XXX 763 */ 764 off = th->th_off << 2; 765 if (off < sizeof (struct tcphdr) || off > tlen) { 766 TCPSTAT_INC(tcps_rcvbadoff); 767 goto drop; 768 } 769 tlen -= off; /* tlen is used instead of ti->ti_len */ 770 if (off > sizeof (struct tcphdr)) { 771 #ifdef INET6 772 if (isipv6) { 773 if (m->m_len < off0 + off) { 774 m = m_pullup(m, off0 + off); 775 if (m == NULL) { 776 TCPSTAT_INC(tcps_rcvshort); 777 return (IPPROTO_DONE); 778 } 779 } 780 ip6 = mtod(m, struct ip6_hdr *); 781 th = (struct tcphdr *)((caddr_t)ip6 + off0); 782 } 783 #endif 784 #if defined(INET) && defined(INET6) 785 else 786 #endif 787 #ifdef INET 788 { 789 if (m->m_len < sizeof(struct ip) + off) { 790 if ((m = m_pullup(m, sizeof (struct ip) + off)) 791 == NULL) { 792 TCPSTAT_INC(tcps_rcvshort); 793 return (IPPROTO_DONE); 794 } 795 ip = mtod(m, struct ip *); 796 th = (struct tcphdr *)((caddr_t)ip + off0); 797 } 798 } 799 #endif 800 optlen = off - sizeof (struct tcphdr); 801 optp = (u_char *)(th + 1); 802 } 803 thflags = tcp_get_flags(th); 804 805 /* 806 * Convert TCP protocol specific fields to host format. 807 */ 808 tcp_fields_to_host(th); 809 810 /* 811 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 812 */ 813 drop_hdrlen = off0 + off; 814 815 /* 816 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 817 */ 818 if ( 819 #ifdef INET6 820 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 821 #ifdef INET 822 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 823 #endif 824 #endif 825 #if defined(INET) && !defined(INET6) 826 (m->m_flags & M_IP_NEXTHOP) 827 #endif 828 ) 829 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 830 831 /* 832 * For initial SYN packets we don't need write lock on matching 833 * PCB, be it a listening one or a synchronized one. The packet 834 * shall not modify its state. 835 */ 836 lookupflag = INPLOOKUP_WILDCARD | 837 ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ? 838 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB) | 839 (V_tcp_bind_all_fibs ? 0 : INPLOOKUP_FIB); 840 findpcb: 841 tp = NULL; 842 #ifdef INET6 843 if (isipv6 && fwd_tag != NULL) { 844 struct sockaddr_in6 *next_hop6; 845 846 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 847 /* 848 * Transparently forwarded. Pretend to be the destination. 849 * Already got one like this? 850 */ 851 inp = in6_pcblookup_mbuf(&V_tcbinfo, 852 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 853 lookupflag & ~INPLOOKUP_WILDCARD, m->m_pkthdr.rcvif, m); 854 if (!inp) { 855 /* 856 * It's new. Try to find the ambushing socket. 857 * Because we've rewritten the destination address, 858 * any hardware-generated hash is ignored. 859 */ 860 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 861 th->th_sport, &next_hop6->sin6_addr, 862 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 863 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 864 } 865 } else if (isipv6) { 866 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 867 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag, 868 m->m_pkthdr.rcvif, m); 869 } 870 #endif /* INET6 */ 871 #if defined(INET6) && defined(INET) 872 else 873 #endif 874 #ifdef INET 875 if (fwd_tag != NULL) { 876 struct sockaddr_in *next_hop; 877 878 next_hop = (struct sockaddr_in *)(fwd_tag+1); 879 /* 880 * Transparently forwarded. Pretend to be the destination. 881 * already got one like this? 882 */ 883 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 884 ip->ip_dst, th->th_dport, lookupflag & ~INPLOOKUP_WILDCARD, 885 m->m_pkthdr.rcvif, m); 886 if (!inp) { 887 /* 888 * It's new. Try to find the ambushing socket. 889 * Because we've rewritten the destination address, 890 * any hardware-generated hash is ignored. 891 */ 892 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 893 th->th_sport, next_hop->sin_addr, 894 next_hop->sin_port ? ntohs(next_hop->sin_port) : 895 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 896 } 897 } else 898 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 899 th->th_sport, ip->ip_dst, th->th_dport, lookupflag, 900 m->m_pkthdr.rcvif, m); 901 #endif /* INET */ 902 903 /* 904 * If the INPCB does not exist then all data in the incoming 905 * segment is discarded and an appropriate RST is sent back. 906 * XXX MRT Send RST using which routing table? 907 */ 908 if (inp == NULL) { 909 if ((lookupflag & INPLOOKUP_WILDCARD) == 0) { 910 /* We came here after second (safety) lookup. */ 911 MPASS(!closed_port); 912 } else { 913 /* 914 * Log communication attempts to ports that are not 915 * in use. 916 */ 917 if (((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 918 V_tcp_log_in_vain == 2) && 919 (s = tcp_log_vain(NULL, th, (void *)ip, ip6))) { 920 log(LOG_INFO, "%s; %s: Connection attempt " 921 "to closed port\n", s, __func__); 922 } 923 closed_port = true; 924 } 925 goto dropwithreset; 926 } 927 INP_LOCK_ASSERT(inp); 928 929 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 930 #ifdef INET6 931 if (isipv6 && IPSEC_ENABLED(ipv6) && 932 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 933 goto dropunlock; 934 } 935 #ifdef INET 936 else 937 #endif 938 #endif /* INET6 */ 939 #ifdef INET 940 if (IPSEC_ENABLED(ipv4) && 941 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 942 goto dropunlock; 943 } 944 #endif /* INET */ 945 #endif /* IPSEC */ 946 947 /* 948 * Check the minimum TTL for socket. 949 */ 950 if (inp->inp_ip_minttl != 0) { 951 #ifdef INET6 952 if (isipv6) { 953 if (inp->inp_ip_minttl > ip6->ip6_hlim) 954 goto dropunlock; 955 } else 956 #endif 957 if (inp->inp_ip_minttl > ip->ip_ttl) 958 goto dropunlock; 959 } 960 961 tp = intotcpcb(inp); 962 switch (tp->t_state) { 963 case TCPS_TIME_WAIT: 964 /* 965 * A previous connection in TIMEWAIT state is supposed to catch 966 * stray or duplicate segments arriving late. If this segment 967 * was a legitimate new connection attempt, the old INPCB gets 968 * removed and we can try again to find a listening socket. 969 */ 970 tcp_dooptions(&to, optp, optlen, 971 (thflags & TH_SYN) ? TO_SYN : 0); 972 /* 973 * tcp_twcheck unlocks the inp always, and frees the m if fails. 974 */ 975 if (tcp_twcheck(inp, &to, th, m, tlen)) 976 goto findpcb; 977 return (IPPROTO_DONE); 978 case TCPS_CLOSED: 979 /* 980 * The TCPCB may no longer exist if the connection is winding 981 * down or it is in the CLOSED state. Either way we drop the 982 * segment and send an appropriate response. 983 */ 984 closed_port = true; 985 goto dropwithreset; 986 } 987 988 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 989 closed_port = true; 990 goto dropwithreset; 991 } 992 993 #ifdef TCP_OFFLOAD 994 if (tp->t_flags & TF_TOE) { 995 tcp_offload_input(tp, m); 996 m = NULL; /* consumed by the TOE driver */ 997 goto dropunlock; 998 } 999 #endif 1000 1001 #ifdef MAC 1002 if (mac_inpcb_check_deliver(inp, m)) 1003 goto dropunlock; 1004 #endif 1005 so = inp->inp_socket; 1006 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1007 /* 1008 * When the socket is accepting connections (the INPCB is in LISTEN 1009 * state) we look into the SYN cache if this is a new connection 1010 * attempt or the completion of a previous one. 1011 */ 1012 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), 1013 ("%s: so accepting but tp %p not listening", __func__, tp)); 1014 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { 1015 struct in_conninfo inc; 1016 1017 bzero(&inc, sizeof(inc)); 1018 #ifdef INET6 1019 if (isipv6) { 1020 inc.inc_flags |= INC_ISIPV6; 1021 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1022 inc.inc_flags |= INC_IPV6MINMTU; 1023 inc.inc6_faddr = ip6->ip6_src; 1024 inc.inc6_laddr = ip6->ip6_dst; 1025 } else 1026 #endif 1027 { 1028 inc.inc_faddr = ip->ip_src; 1029 inc.inc_laddr = ip->ip_dst; 1030 } 1031 inc.inc_fport = th->th_sport; 1032 inc.inc_lport = th->th_dport; 1033 inc.inc_fibnum = so->so_fibnum; 1034 1035 /* 1036 * Check for an existing connection attempt in syncache if 1037 * the flag is only ACK. A successful lookup creates a new 1038 * socket appended to the listen queue in SYN_RECEIVED state. 1039 */ 1040 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1041 int result; 1042 1043 /* 1044 * Parse the TCP options here because 1045 * syncookies need access to the reflected 1046 * timestamp. 1047 */ 1048 tcp_dooptions(&to, optp, optlen, 0); 1049 /* 1050 * NB: syncache_expand() doesn't unlock inp. 1051 */ 1052 result = syncache_expand(&inc, &to, th, &so, m, port); 1053 if (result < 0) { 1054 /* 1055 * A failing TCP MD5 signature comparison 1056 * must result in the segment being dropped 1057 * and must not produce any response back 1058 * to the sender. 1059 */ 1060 goto dropunlock; 1061 } else if (result == 0) { 1062 /* 1063 * No syncache entry, or ACK was not for our 1064 * SYN/ACK. Do our protection against double 1065 * ACK. If peer sent us 2 ACKs, then for the 1066 * first one syncache_expand() successfully 1067 * converted syncache entry into a socket, 1068 * while we were waiting on the inpcb lock. We 1069 * don't want to sent RST for the second ACK, 1070 * so we perform second lookup without wildcard 1071 * match, hoping to find the new socket. If 1072 * the ACK is stray indeed, the missing 1073 * INPLOOKUP_WILDCARD flag in lookupflag would 1074 * hint the above code that the lookup was a 1075 * second attempt. 1076 * 1077 * NB: syncache did its own logging 1078 * of the failure cause. 1079 */ 1080 INP_WUNLOCK(inp); 1081 lookupflag &= ~INPLOOKUP_WILDCARD; 1082 goto findpcb; 1083 } 1084 tfo_socket_result: 1085 if (so == NULL) { 1086 /* 1087 * We completed the 3-way handshake 1088 * but could not allocate a socket 1089 * either due to memory shortage, 1090 * listen queue length limits or 1091 * global socket limits. Send RST 1092 * or wait and have the remote end 1093 * retransmit the ACK for another 1094 * try. 1095 */ 1096 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1097 log(LOG_DEBUG, "%s; %s: Listen socket: " 1098 "Socket allocation failed due to " 1099 "limits or memory shortage, %s\n", 1100 s, __func__, 1101 V_tcp_sc_rst_sock_fail ? 1102 "sending RST" : "try again"); 1103 if (V_tcp_sc_rst_sock_fail) { 1104 goto dropwithreset; 1105 } else 1106 goto dropunlock; 1107 } 1108 /* 1109 * Socket is created in state SYN_RECEIVED. 1110 * Unlock the listen socket, lock the newly 1111 * created socket and update the tp variable. 1112 * If we came here via jump to tfo_socket_result, 1113 * then listening socket is read-locked. 1114 */ 1115 INP_UNLOCK(inp); /* listen socket */ 1116 inp = sotoinpcb(so); 1117 /* 1118 * New connection inpcb is already locked by 1119 * syncache_expand(). 1120 */ 1121 INP_WLOCK_ASSERT(inp); 1122 tp = intotcpcb(inp); 1123 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1124 ("%s: ", __func__)); 1125 /* 1126 * Process the segment and the data it 1127 * contains. tcp_do_segment() consumes 1128 * the mbuf chain and unlocks the inpcb. 1129 */ 1130 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1131 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, 1132 tlen, iptos); 1133 return (IPPROTO_DONE); 1134 } 1135 /* 1136 * Segment flag validation for new connection attempts: 1137 * 1138 * Our (SYN|ACK) response was rejected. 1139 * Check with syncache and remove entry to prevent 1140 * retransmits. 1141 * 1142 * NB: syncache_chkrst does its own logging of failure 1143 * causes. 1144 */ 1145 if (thflags & TH_RST) { 1146 syncache_chkrst(&inc, th, port); 1147 goto dropunlock; 1148 } 1149 /* 1150 * We can't do anything without SYN. 1151 */ 1152 if ((thflags & TH_SYN) == 0) { 1153 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1154 log(LOG_DEBUG, "%s; %s: Listen socket: " 1155 "SYN is missing, segment ignored\n", 1156 s, __func__); 1157 TCPSTAT_INC(tcps_badsyn); 1158 goto dropunlock; 1159 } 1160 /* 1161 * (SYN|ACK) is bogus on a listen socket. 1162 */ 1163 if (thflags & TH_ACK) { 1164 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1165 log(LOG_DEBUG, "%s; %s: Listen socket: " 1166 "SYN|ACK invalid, segment ignored\n", 1167 s, __func__); 1168 TCPSTAT_INC(tcps_badsyn); 1169 goto dropunlock; 1170 } 1171 /* 1172 * If the drop_synfin option is enabled, drop all 1173 * segments with both the SYN and FIN bits set. 1174 * This prevents e.g. nmap from identifying the 1175 * TCP/IP stack. 1176 * XXX: Poor reasoning. nmap has other methods 1177 * and is constantly refining its stack detection 1178 * strategies. 1179 * XXX: This is a violation of the TCP specification 1180 * and was used by RFC1644. 1181 */ 1182 if ((thflags & TH_FIN) && V_drop_synfin) { 1183 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1184 log(LOG_DEBUG, "%s; %s: Listen socket: " 1185 "SYN|FIN segment ignored (based on " 1186 "sysctl setting)\n", s, __func__); 1187 TCPSTAT_INC(tcps_badsyn); 1188 goto dropunlock; 1189 } 1190 /* 1191 * Segment's flags are (SYN) or (SYN|FIN). 1192 * 1193 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1194 * as they do not affect the state of the TCP FSM. 1195 * The data pointed to by TH_URG and th_urp is ignored. 1196 */ 1197 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1198 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1199 KASSERT(thflags & (TH_SYN), 1200 ("%s: Listen socket: TH_SYN not set", __func__)); 1201 INP_RLOCK_ASSERT(inp); 1202 #ifdef INET6 1203 /* 1204 * If deprecated address is forbidden, 1205 * we do not accept SYN to deprecated interface 1206 * address to prevent any new inbound connection from 1207 * getting established. 1208 * When we do not accept SYN, we send a TCP RST, 1209 * with deprecated source address (instead of dropping 1210 * it). We compromise it as it is much better for peer 1211 * to send a RST, and RST will be the final packet 1212 * for the exchange. 1213 * 1214 * If we do not forbid deprecated addresses, we accept 1215 * the SYN packet. RFC2462 does not suggest dropping 1216 * SYN in this case. 1217 * If we decipher RFC2462 5.5.4, it says like this: 1218 * 1. use of deprecated addr with existing 1219 * communication is okay - "SHOULD continue to be 1220 * used" 1221 * 2. use of it with new communication: 1222 * (2a) "SHOULD NOT be used if alternate address 1223 * with sufficient scope is available" 1224 * (2b) nothing mentioned otherwise. 1225 * Here we fall into (2b) case as we have no choice in 1226 * our source address selection - we must obey the peer. 1227 * 1228 * The wording in RFC2462 is confusing, and there are 1229 * multiple description text for deprecated address 1230 * handling - worse, they are not exactly the same. 1231 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1232 */ 1233 if (isipv6 && !V_ip6_use_deprecated) { 1234 struct in6_ifaddr *ia6; 1235 1236 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1237 if (ia6 != NULL && 1238 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1239 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1240 log(LOG_DEBUG, "%s; %s: Listen socket: " 1241 "Connection attempt to deprecated " 1242 "IPv6 address rejected\n", 1243 s, __func__); 1244 goto dropwithreset; 1245 } 1246 } 1247 #endif /* INET6 */ 1248 /* 1249 * Basic sanity checks on incoming SYN requests: 1250 * Don't respond if the destination is a link layer 1251 * broadcast according to RFC1122 4.2.3.10, p. 104. 1252 * If it is from this socket it must be forged. 1253 * Don't respond if the source or destination is a 1254 * global or subnet broad- or multicast address. 1255 * Note that it is quite possible to receive unicast 1256 * link-layer packets with a broadcast IP address. Use 1257 * in_ifnet_broadcast() to find them. 1258 */ 1259 if (m->m_flags & (M_BCAST|M_MCAST)) { 1260 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1261 log(LOG_DEBUG, "%s; %s: Listen socket: " 1262 "Connection attempt from broad- or multicast " 1263 "link layer address ignored\n", s, __func__); 1264 goto dropunlock; 1265 } 1266 #ifdef INET6 1267 if (isipv6) { 1268 if (th->th_dport == th->th_sport && 1269 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1270 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1271 log(LOG_DEBUG, "%s; %s: Listen socket: " 1272 "Connection attempt to/from self " 1273 "ignored\n", s, __func__); 1274 goto dropunlock; 1275 } 1276 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1277 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1278 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1279 log(LOG_DEBUG, "%s; %s: Listen socket: " 1280 "Connection attempt from/to multicast " 1281 "address ignored\n", s, __func__); 1282 goto dropunlock; 1283 } 1284 } 1285 #endif 1286 #if defined(INET) && defined(INET6) 1287 else 1288 #endif 1289 #ifdef INET 1290 { 1291 if (th->th_dport == th->th_sport && 1292 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1293 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1294 log(LOG_DEBUG, "%s; %s: Listen socket: " 1295 "Connection attempt from/to self " 1296 "ignored\n", s, __func__); 1297 goto dropunlock; 1298 } 1299 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1300 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1301 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1302 in_ifnet_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1303 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1304 log(LOG_DEBUG, "%s; %s: Listen socket: " 1305 "Connection attempt from/to broad- " 1306 "or multicast address ignored\n", 1307 s, __func__); 1308 goto dropunlock; 1309 } 1310 } 1311 #endif 1312 /* 1313 * SYN appears to be valid. Create compressed TCP state 1314 * for syncache. 1315 */ 1316 TCP_PROBE3(debug__input, tp, th, m); 1317 tcp_dooptions(&to, optp, optlen, TO_SYN); 1318 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, 1319 iptos, port)) != NULL) 1320 goto tfo_socket_result; 1321 1322 /* 1323 * Entry added to syncache and mbuf consumed. 1324 * Only the listen socket is unlocked by syncache_add(). 1325 */ 1326 return (IPPROTO_DONE); 1327 } 1328 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1329 if (tp->t_flags & TF_SIGNATURE) { 1330 tcp_dooptions(&to, optp, optlen, thflags); 1331 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1332 TCPSTAT_INC(tcps_sig_err_nosigopt); 1333 goto dropunlock; 1334 } 1335 if (!TCPMD5_ENABLED() || 1336 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1337 goto dropunlock; 1338 } 1339 #endif 1340 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1341 1342 /* 1343 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1344 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1345 * the inpcb, and unlocks pcbinfo. 1346 * 1347 * XXXGL: in case of a pure SYN arriving on existing connection 1348 * TCP stacks won't need to modify the PCB, they would either drop 1349 * the segment silently, or send a challenge ACK. However, we try 1350 * to upgrade the lock, because calling convention for stacks is 1351 * write-lock on PCB. If upgrade fails, drop the SYN. 1352 */ 1353 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0) 1354 goto dropunlock; 1355 1356 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, tlen, iptos); 1357 return (IPPROTO_DONE); 1358 1359 dropwithreset: 1360 /* 1361 * When blackholing do not respond with a RST but 1362 * completely ignore the segment and drop it. 1363 */ 1364 if (((!closed_port && V_blackhole == 3) || 1365 (closed_port && 1366 ((V_blackhole == 1 && (thflags & TH_SYN)) || V_blackhole > 1))) && 1367 (V_blackhole_local || ( 1368 #ifdef INET6 1369 isipv6 ? !in6_localip(&ip6->ip6_src) : 1370 #endif 1371 #ifdef INET 1372 !in_localip(ip->ip_src) 1373 #else 1374 true 1375 #endif 1376 ))) 1377 goto dropunlock; 1378 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1379 tcp_dropwithreset(m, th, tp, tlen); 1380 m = NULL; /* mbuf chain got consumed. */ 1381 1382 dropunlock: 1383 if (m != NULL) 1384 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1385 1386 if (inp != NULL) 1387 INP_UNLOCK(inp); 1388 1389 drop: 1390 if (s != NULL) 1391 free(s, M_TCPLOG); 1392 if (m != NULL) 1393 m_freem(m); 1394 return (IPPROTO_DONE); 1395 } 1396 1397 /* 1398 * Automatic sizing of receive socket buffer. Often the send 1399 * buffer size is not optimally adjusted to the actual network 1400 * conditions at hand (delay bandwidth product). Setting the 1401 * buffer size too small limits throughput on links with high 1402 * bandwidth and high delay (eg. trans-continental/oceanic links). 1403 * 1404 * On the receive side the socket buffer memory is only rarely 1405 * used to any significant extent. This allows us to be much 1406 * more aggressive in scaling the receive socket buffer. For 1407 * the case that the buffer space is actually used to a large 1408 * extent and we run out of kernel memory we can simply drop 1409 * the new segments; TCP on the sender will just retransmit it 1410 * later. Setting the buffer size too big may only consume too 1411 * much kernel memory if the application doesn't read() from 1412 * the socket or packet loss or reordering makes use of the 1413 * reassembly queue. 1414 * 1415 * The criteria to step up the receive buffer one notch are: 1416 * 1. Application has not set receive buffer size with 1417 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1418 * 2. the number of bytes received during 1/2 of an sRTT 1419 * is at least 3/8 of the current socket buffer size. 1420 * 3. receive buffer size has not hit maximal automatic size; 1421 * 1422 * If all of the criteria are met, we increase the socket buffer 1423 * by a 1/2 (bounded by the max). This allows us to keep ahead 1424 * of slow-start but also makes it so our peer never gets limited 1425 * by our rwnd which we then open up causing a burst. 1426 * 1427 * This algorithm does two steps per RTT at most and only if 1428 * we receive a bulk stream w/o packet losses or reorderings. 1429 * Shrinking the buffer during idle times is not necessary as 1430 * it doesn't consume any memory when idle. 1431 * 1432 * TODO: Only step up if the application is actually serving 1433 * the buffer to better manage the socket buffer resources. 1434 */ 1435 int 1436 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1437 struct tcpcb *tp, int tlen) 1438 { 1439 int newsize = 0; 1440 1441 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1442 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1443 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1444 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1445 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1446 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1447 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1448 } 1449 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1450 1451 /* Start over with next RTT. */ 1452 tp->rfbuf_ts = 0; 1453 tp->rfbuf_cnt = 0; 1454 } else { 1455 tp->rfbuf_cnt += tlen; /* add up */ 1456 } 1457 return (newsize); 1458 } 1459 1460 int 1461 tcp_input(struct mbuf **mp, int *offp, int proto) 1462 { 1463 return(tcp_input_with_port(mp, offp, proto, 0)); 1464 } 1465 1466 static void 1467 tcp_handle_wakeup(struct tcpcb *tp) 1468 { 1469 1470 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1471 1472 if (tp->t_flags & TF_WAKESOR) { 1473 struct socket *so = tptosocket(tp); 1474 1475 tp->t_flags &= ~TF_WAKESOR; 1476 SOCK_RECVBUF_LOCK_ASSERT(so); 1477 sorwakeup_locked(so); 1478 } 1479 } 1480 1481 void 1482 tcp_do_segment(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th, 1483 int drop_hdrlen, int tlen, uint8_t iptos) 1484 { 1485 uint16_t thflags; 1486 int acked, ourfinisacked, needoutput = 0; 1487 sackstatus_t sack_changed; 1488 int todrop, win, incforsyn = 0; 1489 uint32_t tiwin; 1490 uint16_t nsegs; 1491 char *s; 1492 struct inpcb *inp = tptoinpcb(tp); 1493 struct socket *so = tptosocket(tp); 1494 struct in_conninfo *inc = &inp->inp_inc; 1495 struct mbuf *mfree; 1496 struct tcpopt to; 1497 int tfo_syn; 1498 u_int maxseg = 0; 1499 bool no_data; 1500 1501 no_data = (tlen == 0); 1502 thflags = tcp_get_flags(th); 1503 tp->sackhint.last_sack_ack = 0; 1504 sack_changed = SACK_NOCHANGE; 1505 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1506 1507 NET_EPOCH_ASSERT(); 1508 INP_WLOCK_ASSERT(inp); 1509 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1510 __func__)); 1511 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1512 __func__)); 1513 1514 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1515 tlen, NULL, true); 1516 1517 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1518 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1519 log(LOG_DEBUG, "%s; %s: " 1520 "SYN|FIN segment ignored (based on " 1521 "sysctl setting)\n", s, __func__); 1522 free(s, M_TCPLOG); 1523 } 1524 goto drop; 1525 } 1526 1527 /* 1528 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1529 * check SEQ.ACK first. 1530 */ 1531 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1532 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1533 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1534 goto dropwithreset; 1535 } 1536 1537 /* 1538 * Segment received on connection. 1539 * Reset idle time and keep-alive timer. 1540 * XXX: This should be done after segment 1541 * validation to ignore broken/spoofed segs. 1542 */ 1543 if (tp->t_idle_reduce && 1544 (tp->snd_max == tp->snd_una) && 1545 ((ticks - tp->t_rcvtime) >= tp->t_rxtcur)) 1546 cc_after_idle(tp); 1547 tp->t_rcvtime = ticks; 1548 1549 if (thflags & TH_FIN) 1550 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN); 1551 /* 1552 * Scale up the window into a 32-bit value. 1553 * For the SYN_SENT state the scale is zero. 1554 */ 1555 tiwin = th->th_win << tp->snd_scale; 1556 #ifdef STATS 1557 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1558 #endif 1559 1560 /* 1561 * TCP ECN processing. 1562 */ 1563 if (tcp_ecn_input_segment(tp, thflags, tlen, 1564 tcp_packets_this_ack(tp, th->th_ack), 1565 iptos)) 1566 cc_cong_signal(tp, th, CC_ECN); 1567 1568 /* 1569 * Parse options on any incoming segment. 1570 */ 1571 tcp_dooptions(&to, (u_char *)(th + 1), 1572 (th->th_off << 2) - sizeof(struct tcphdr), 1573 (thflags & TH_SYN) ? TO_SYN : 0); 1574 if (tp->t_flags2 & TF2_PROC_SACK_PROHIBIT) { 1575 /* 1576 * We don't look at sack's from the 1577 * peer because the MSS is too small which 1578 * can subject us to an attack. 1579 */ 1580 to.to_flags &= ~TOF_SACK; 1581 } 1582 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1583 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1584 (to.to_flags & TOF_SIGNATURE) == 0) { 1585 TCPSTAT_INC(tcps_sig_err_sigopt); 1586 /* XXX: should drop? */ 1587 } 1588 #endif 1589 /* 1590 * If echoed timestamp is later than the current time, 1591 * fall back to non RFC1323 RTT calculation. Normalize 1592 * timestamp if syncookies were used when this connection 1593 * was established. 1594 */ 1595 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1596 to.to_tsecr -= tp->ts_offset; 1597 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) { 1598 to.to_tsecr = 0; 1599 } 1600 } 1601 /* 1602 * Process options only when we get SYN/ACK back. The SYN case 1603 * for incoming connections is handled in tcp_syncache. 1604 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1605 * or <SYN,ACK>) segment itself is never scaled. 1606 * XXX this is traditional behavior, may need to be cleaned up. 1607 */ 1608 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1609 /* Handle parallel SYN for ECN */ 1610 tcp_ecn_input_parallel_syn(tp, thflags, iptos); 1611 if ((to.to_flags & TOF_SCALE) && 1612 (tp->t_flags & TF_REQ_SCALE) && 1613 !(tp->t_flags & TF_NOOPT)) { 1614 tp->t_flags |= TF_RCVD_SCALE; 1615 tp->snd_scale = to.to_wscale; 1616 } else { 1617 tp->t_flags &= ~TF_REQ_SCALE; 1618 } 1619 /* 1620 * Initial send window. It will be updated with 1621 * the next incoming segment to the scaled value. 1622 */ 1623 tp->snd_wnd = th->th_win; 1624 if ((to.to_flags & TOF_TS) && 1625 (tp->t_flags & TF_REQ_TSTMP) && 1626 !(tp->t_flags & TF_NOOPT)) { 1627 tp->t_flags |= TF_RCVD_TSTMP; 1628 tp->ts_recent = to.to_tsval; 1629 tp->ts_recent_age = tcp_ts_getticks(); 1630 } else { 1631 tp->t_flags &= ~TF_REQ_TSTMP; 1632 } 1633 if (to.to_flags & TOF_MSS) { 1634 tcp_mss(tp, to.to_mss); 1635 } 1636 if ((tp->t_flags & TF_SACK_PERMIT) && 1637 (!(to.to_flags & TOF_SACKPERM) || 1638 (tp->t_flags & TF_NOOPT))) { 1639 tp->t_flags &= ~TF_SACK_PERMIT; 1640 } 1641 if (tp->t_flags & TF_FASTOPEN) { 1642 if ((to.to_flags & TOF_FASTOPEN) && 1643 !(tp->t_flags & TF_NOOPT)) { 1644 uint16_t mss; 1645 1646 if (to.to_flags & TOF_MSS) { 1647 mss = to.to_mss; 1648 } else { 1649 if ((inp->inp_vflag & INP_IPV6) != 0) { 1650 mss = TCP6_MSS; 1651 } else { 1652 mss = TCP_MSS; 1653 } 1654 } 1655 tcp_fastopen_update_cache(tp, mss, 1656 to.to_tfo_len, to.to_tfo_cookie); 1657 } else { 1658 tcp_fastopen_disable_path(tp); 1659 } 1660 } 1661 } 1662 1663 /* 1664 * If timestamps were negotiated during SYN/ACK and a 1665 * segment without a timestamp is received, silently drop 1666 * the segment, unless it is a RST segment or missing timestamps are 1667 * tolerated. 1668 * See section 3.2 of RFC 7323. 1669 */ 1670 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1671 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1672 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1673 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1674 "segment processed normally\n", 1675 s, __func__); 1676 free(s, M_TCPLOG); 1677 } 1678 } else { 1679 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1680 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1681 "segment silently dropped\n", s, __func__); 1682 free(s, M_TCPLOG); 1683 } 1684 goto drop; 1685 } 1686 } 1687 /* 1688 * If timestamps were not negotiated during SYN/ACK and a 1689 * segment with a timestamp is received, ignore the 1690 * timestamp and process the packet normally. 1691 * See section 3.2 of RFC 7323. 1692 */ 1693 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1694 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1695 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1696 "segment processed normally\n", s, __func__); 1697 free(s, M_TCPLOG); 1698 } 1699 } 1700 1701 /* 1702 * Header prediction: check for the two common cases 1703 * of a uni-directional data xfer. If the packet has 1704 * no control flags, is in-sequence, the window didn't 1705 * change and we're not retransmitting, it's a 1706 * candidate. If the length is zero and the ack moved 1707 * forward, we're the sender side of the xfer. Just 1708 * free the data acked & wake any higher level process 1709 * that was blocked waiting for space. If the length 1710 * is non-zero and the ack didn't move, we're the 1711 * receiver side. If we're getting packets in-order 1712 * (the reassembly queue is empty), add the data to 1713 * the socket buffer and note that we need a delayed ack. 1714 * Make sure that the hidden state-flags are also off. 1715 * Since we check for TCPS_ESTABLISHED first, it can only 1716 * be TH_NEEDSYN. 1717 */ 1718 if (tp->t_state == TCPS_ESTABLISHED && 1719 th->th_seq == tp->rcv_nxt && 1720 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1721 tp->snd_nxt == tp->snd_max && 1722 tiwin && tiwin == tp->snd_wnd && 1723 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1724 SEGQ_EMPTY(tp) && 1725 ((to.to_flags & TOF_TS) == 0 || 1726 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1727 /* 1728 * If last ACK falls within this segment's sequence numbers, 1729 * record the timestamp. 1730 * NOTE that the test is modified according to the latest 1731 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1732 */ 1733 if ((to.to_flags & TOF_TS) != 0 && 1734 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1735 tp->ts_recent_age = tcp_ts_getticks(); 1736 tp->ts_recent = to.to_tsval; 1737 } 1738 1739 if (no_data) { 1740 if (SEQ_GT(th->th_ack, tp->snd_una) && 1741 SEQ_LEQ(th->th_ack, tp->snd_max) && 1742 !IN_RECOVERY(tp->t_flags) && 1743 (to.to_flags & TOF_SACK) == 0 && 1744 TAILQ_EMPTY(&tp->snd_holes)) { 1745 /* 1746 * This is a pure ack for outstanding data. 1747 */ 1748 TCPSTAT_INC(tcps_predack); 1749 1750 /* 1751 * "bad retransmit" recovery. 1752 */ 1753 if (tp->t_rxtshift == 1 && 1754 tp->t_flags & TF_PREVVALID && 1755 tp->t_badrxtwin != 0 && 1756 (((to.to_flags & TOF_TS) != 0 && 1757 to.to_tsecr != 0 && 1758 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) || 1759 ((to.to_flags & TOF_TS) == 0 && 1760 TSTMP_LT(ticks, tp->t_badrxtwin)))) 1761 cc_cong_signal(tp, th, CC_RTO_ERR); 1762 1763 /* 1764 * Recalculate the transmit timer / rtt. 1765 * 1766 * Some boxes send broken timestamp replies 1767 * during the SYN+ACK phase, ignore 1768 * timestamps of 0 or we could calculate a 1769 * huge RTT and blow up the retransmit timer. 1770 */ 1771 if ((to.to_flags & TOF_TS) != 0 && 1772 to.to_tsecr) { 1773 uint32_t t; 1774 1775 t = tcp_ts_getticks() - to.to_tsecr; 1776 if (!tp->t_rttlow || tp->t_rttlow > t) 1777 tp->t_rttlow = t; 1778 tcp_xmit_timer(tp, 1779 TCP_TS_TO_TICKS(t) + 1); 1780 } else if (tp->t_rtttime && 1781 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1782 if (!tp->t_rttlow || 1783 tp->t_rttlow > ticks - tp->t_rtttime) 1784 tp->t_rttlow = ticks - tp->t_rtttime; 1785 tcp_xmit_timer(tp, 1786 ticks - tp->t_rtttime); 1787 } 1788 acked = BYTES_THIS_ACK(tp, th); 1789 1790 #ifdef TCP_HHOOK 1791 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1792 hhook_run_tcp_est_in(tp, th, &to); 1793 #endif 1794 1795 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1796 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1797 sbdrop(&so->so_snd, acked); 1798 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1799 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1800 tp->snd_recover = th->th_ack - 1; 1801 1802 /* 1803 * Let the congestion control algorithm update 1804 * congestion control related information. This 1805 * typically means increasing the congestion 1806 * window. 1807 */ 1808 cc_ack_received(tp, th, nsegs, CC_ACK); 1809 1810 tp->snd_una = th->th_ack; 1811 /* 1812 * Pull snd_wl2 up to prevent seq wrap relative 1813 * to th_ack. 1814 */ 1815 tp->snd_wl2 = th->th_ack; 1816 tp->t_dupacks = 0; 1817 m_freem(m); 1818 1819 /* 1820 * If all outstanding data are acked, stop 1821 * retransmit timer, otherwise restart timer 1822 * using current (possibly backed-off) value. 1823 * If process is waiting for space, 1824 * wakeup/selwakeup/signal. If data 1825 * are ready to send, let tcp_output 1826 * decide between more output or persist. 1827 */ 1828 TCP_PROBE3(debug__input, tp, th, m); 1829 /* 1830 * Clear t_acktime if remote side has ACKd 1831 * all data in the socket buffer. 1832 * Otherwise, update t_acktime if we received 1833 * a sufficiently large ACK. 1834 */ 1835 if (sbavail(&so->so_snd) == 0) 1836 tp->t_acktime = 0; 1837 else if (acked > 1) 1838 tp->t_acktime = ticks; 1839 if (tp->snd_una == tp->snd_max) 1840 tcp_timer_activate(tp, TT_REXMT, 0); 1841 else if (!tcp_timer_active(tp, TT_PERSIST)) 1842 tcp_timer_activate(tp, TT_REXMT, 1843 TP_RXTCUR(tp)); 1844 sowwakeup(so); 1845 /* 1846 * Only call tcp_output when there 1847 * is new data available to be sent 1848 * or we need to send an ACK. 1849 */ 1850 if ((tp->t_flags & TF_ACKNOW) || 1851 (sbavail(&so->so_snd) >= 1852 SEQ_SUB(tp->snd_max, tp->snd_una))) { 1853 (void) tcp_output(tp); 1854 } 1855 goto check_delack; 1856 } 1857 } else if (th->th_ack == tp->snd_una && 1858 tlen <= sbspace(&so->so_rcv)) { 1859 int newsize = 0; /* automatic sockbuf scaling */ 1860 1861 /* 1862 * This is a pure, in-sequence data packet with 1863 * nothing on the reassembly queue and we have enough 1864 * buffer space to take it. 1865 */ 1866 /* Clean receiver SACK report if present */ 1867 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1868 tcp_clean_sackreport(tp); 1869 TCPSTAT_INC(tcps_preddat); 1870 tp->rcv_nxt += tlen; 1871 if (tlen && 1872 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1873 (tp->t_fbyte_in == 0)) { 1874 tp->t_fbyte_in = ticks; 1875 if (tp->t_fbyte_in == 0) 1876 tp->t_fbyte_in = 1; 1877 if (tp->t_fbyte_out && tp->t_fbyte_in) 1878 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1879 } 1880 /* 1881 * Pull snd_wl1 up to prevent seq wrap relative to 1882 * th_seq. 1883 */ 1884 tp->snd_wl1 = th->th_seq; 1885 /* 1886 * Pull rcv_up up to prevent seq wrap relative to 1887 * rcv_nxt. 1888 */ 1889 tp->rcv_up = tp->rcv_nxt; 1890 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1891 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1892 TCP_PROBE3(debug__input, tp, th, m); 1893 1894 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1895 1896 /* Add data to socket buffer. */ 1897 SOCK_RECVBUF_LOCK(so); 1898 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1899 m_freem(m); 1900 } else { 1901 /* 1902 * Set new socket buffer size. 1903 * Give up when limit is reached. 1904 */ 1905 if (newsize) 1906 if (!sbreserve_locked(so, SO_RCV, 1907 newsize, NULL)) 1908 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1909 m_adj(m, drop_hdrlen); /* delayed header drop */ 1910 sbappendstream_locked(&so->so_rcv, m, 0); 1911 } 1912 /* NB: sorwakeup_locked() does an implicit unlock. */ 1913 sorwakeup_locked(so); 1914 if (DELAY_ACK(tp, tlen)) { 1915 tp->t_flags |= TF_DELACK; 1916 } else { 1917 tp->t_flags |= TF_ACKNOW; 1918 (void) tcp_output(tp); 1919 } 1920 goto check_delack; 1921 } 1922 } 1923 1924 /* 1925 * Calculate amount of space in receive window, 1926 * and then do TCP input processing. 1927 * Receive window is amount of space in rcv queue, 1928 * but not less than advertised window. 1929 */ 1930 win = sbspace(&so->so_rcv); 1931 if (win < 0) 1932 win = 0; 1933 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1934 1935 switch (tp->t_state) { 1936 /* 1937 * If the state is SYN_RECEIVED: 1938 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1939 */ 1940 case TCPS_SYN_RECEIVED: 1941 if (thflags & TH_RST) { 1942 /* Handle RST segments later. */ 1943 break; 1944 } 1945 if ((thflags & TH_ACK) && 1946 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1947 SEQ_GT(th->th_ack, tp->snd_max))) { 1948 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1949 goto dropwithreset; 1950 } 1951 if (tp->t_flags & TF_FASTOPEN) { 1952 /* 1953 * When a TFO connection is in SYN_RECEIVED, the 1954 * only valid packets are the initial SYN, a 1955 * retransmit/copy of the initial SYN (possibly with 1956 * a subset of the original data), a valid ACK, a 1957 * FIN, or a RST. 1958 */ 1959 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1960 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1961 goto dropwithreset; 1962 } else if (thflags & TH_SYN) { 1963 /* non-initial SYN is ignored */ 1964 if ((tcp_timer_active(tp, TT_DELACK) || 1965 tcp_timer_active(tp, TT_REXMT))) 1966 goto drop; 1967 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1968 goto drop; 1969 } 1970 } 1971 break; 1972 1973 /* 1974 * If the state is SYN_SENT: 1975 * if seg contains a RST with valid ACK (SEQ.ACK has already 1976 * been verified), then drop the connection. 1977 * if seg contains a RST without an ACK, drop the seg. 1978 * if seg does not contain SYN, then drop the seg. 1979 * Otherwise this is an acceptable SYN segment 1980 * initialize tp->rcv_nxt and tp->irs 1981 * if seg contains ack then advance tp->snd_una 1982 * if seg contains an ECE and ECN support is enabled, the stream 1983 * is ECN capable. 1984 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1985 * arrange for segment to be acked (eventually) 1986 * continue processing rest of data/controls, beginning with URG 1987 */ 1988 case TCPS_SYN_SENT: 1989 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1990 TCP_PROBE5(connect__refused, NULL, tp, 1991 m, tp, th); 1992 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1993 tp = tcp_drop(tp, ECONNREFUSED); 1994 } 1995 if (thflags & TH_RST) 1996 goto drop; 1997 if (!(thflags & TH_SYN)) 1998 goto drop; 1999 2000 tp->irs = th->th_seq; 2001 tcp_rcvseqinit(tp); 2002 if (thflags & TH_ACK) { 2003 int tfo_partial_ack = 0; 2004 2005 TCPSTAT_INC(tcps_connects); 2006 soisconnected(so); 2007 #ifdef MAC 2008 mac_socketpeer_set_from_mbuf(m, so); 2009 #endif 2010 /* Do window scaling on this connection? */ 2011 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2012 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2013 tp->rcv_scale = tp->request_r_scale; 2014 } 2015 tp->rcv_adv += min(tp->rcv_wnd, 2016 TCP_MAXWIN << tp->rcv_scale); 2017 tp->snd_una++; /* SYN is acked */ 2018 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2019 tp->snd_nxt = tp->snd_una; 2020 /* 2021 * If not all the data that was sent in the TFO SYN 2022 * has been acked, resend the remainder right away. 2023 */ 2024 if ((tp->t_flags & TF_FASTOPEN) && 2025 (tp->snd_una != tp->snd_max)) { 2026 tp->snd_nxt = th->th_ack; 2027 tfo_partial_ack = 1; 2028 } 2029 /* 2030 * If there's data, delay ACK; if there's also a FIN 2031 * ACKNOW will be turned on later. 2032 */ 2033 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2034 tcp_timer_activate(tp, TT_DELACK, 2035 tcp_delacktime); 2036 else 2037 tp->t_flags |= TF_ACKNOW; 2038 2039 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2040 2041 /* 2042 * Received <SYN,ACK> in SYN_SENT[*] state. 2043 * Transitions: 2044 * SYN_SENT --> ESTABLISHED 2045 * SYN_SENT* --> FIN_WAIT_1 2046 */ 2047 tp->t_starttime = ticks; 2048 if (tp->t_flags & TF_NEEDFIN) { 2049 tp->t_acktime = ticks; 2050 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2051 tp->t_flags &= ~TF_NEEDFIN; 2052 thflags &= ~TH_SYN; 2053 } else { 2054 tcp_state_change(tp, TCPS_ESTABLISHED); 2055 TCP_PROBE5(connect__established, NULL, tp, 2056 m, tp, th); 2057 cc_conn_init(tp); 2058 tcp_timer_activate(tp, TT_KEEP, 2059 TP_KEEPIDLE(tp)); 2060 } 2061 } else { 2062 /* 2063 * Received initial SYN in SYN-SENT[*] state => 2064 * simultaneous open. 2065 * If it succeeds, connection is * half-synchronized. 2066 * Otherwise, do 3-way handshake: 2067 * SYN-SENT -> SYN-RECEIVED 2068 * SYN-SENT* -> SYN-RECEIVED* 2069 */ 2070 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN); 2071 tcp_timer_activate(tp, TT_REXMT, 0); 2072 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2073 } 2074 2075 /* 2076 * Advance th->th_seq to correspond to first data byte. 2077 * If data, trim to stay within window, 2078 * dropping FIN if necessary. 2079 */ 2080 th->th_seq++; 2081 if (tlen > tp->rcv_wnd) { 2082 todrop = tlen - tp->rcv_wnd; 2083 m_adj(m, -todrop); 2084 tlen = tp->rcv_wnd; 2085 thflags &= ~TH_FIN; 2086 TCPSTAT_INC(tcps_rcvpackafterwin); 2087 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2088 } 2089 tp->snd_wl1 = th->th_seq - 1; 2090 tp->rcv_up = th->th_seq; 2091 /* 2092 * Client side of transaction: already sent SYN and data. 2093 * If the remote host used T/TCP to validate the SYN, 2094 * our data will be ACK'd; if so, enter normal data segment 2095 * processing in the middle of step 5, ack processing. 2096 * Otherwise, goto step 6. 2097 */ 2098 if (thflags & TH_ACK) 2099 goto process_ACK; 2100 2101 goto step6; 2102 } 2103 2104 /* 2105 * States other than LISTEN or SYN_SENT. 2106 * First check the RST flag and sequence number since reset segments 2107 * are exempt from the timestamp and connection count tests. This 2108 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2109 * below which allowed reset segments in half the sequence space 2110 * to fall though and be processed (which gives forged reset 2111 * segments with a random sequence number a 50 percent chance of 2112 * killing a connection). 2113 * Then check timestamp, if present. 2114 * Then check the connection count, if present. 2115 * Then check that at least some bytes of segment are within 2116 * receive window. If segment begins before rcv_nxt, 2117 * drop leading data (and SYN); if nothing left, just ack. 2118 */ 2119 if (thflags & TH_RST) { 2120 /* 2121 * RFC5961 Section 3.2 2122 * 2123 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2124 * - If RST is in window, we send challenge ACK. 2125 * 2126 * Note: to take into account delayed ACKs, we should 2127 * test against last_ack_sent instead of rcv_nxt. 2128 * Note 2: we handle special case of closed window, not 2129 * covered by the RFC. 2130 */ 2131 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2132 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2133 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2134 KASSERT(tp->t_state != TCPS_SYN_SENT, 2135 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2136 __func__, th, tp)); 2137 2138 if (V_tcp_insecure_rst || 2139 tp->last_ack_sent == th->th_seq) { 2140 TCPSTAT_INC(tcps_drops); 2141 /* Drop the connection. */ 2142 switch (tp->t_state) { 2143 case TCPS_SYN_RECEIVED: 2144 so->so_error = ECONNREFUSED; 2145 goto close; 2146 case TCPS_ESTABLISHED: 2147 case TCPS_FIN_WAIT_1: 2148 case TCPS_FIN_WAIT_2: 2149 case TCPS_CLOSE_WAIT: 2150 case TCPS_CLOSING: 2151 case TCPS_LAST_ACK: 2152 so->so_error = ECONNRESET; 2153 close: 2154 /* FALLTHROUGH */ 2155 default: 2156 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST); 2157 tp = tcp_close(tp); 2158 } 2159 } else { 2160 TCPSTAT_INC(tcps_badrst); 2161 tcp_send_challenge_ack(tp, th, m); 2162 m = NULL; 2163 } 2164 } 2165 goto drop; 2166 } 2167 2168 /* 2169 * RFC5961 Section 4.2 2170 * Send challenge ACK for any SYN in synchronized state. 2171 */ 2172 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2173 tp->t_state != TCPS_SYN_RECEIVED) { 2174 TCPSTAT_INC(tcps_badsyn); 2175 if (V_tcp_insecure_syn && 2176 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2177 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2178 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2179 tp = tcp_drop(tp, ECONNRESET); 2180 } else { 2181 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2182 tcp_send_challenge_ack(tp, th, m); 2183 m = NULL; 2184 } 2185 goto drop; 2186 } 2187 2188 /* 2189 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2190 * and it's less than ts_recent, drop it. 2191 */ 2192 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2193 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2194 /* Check to see if ts_recent is over 24 days old. */ 2195 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2196 /* 2197 * Invalidate ts_recent. If this segment updates 2198 * ts_recent, the age will be reset later and ts_recent 2199 * will get a valid value. If it does not, setting 2200 * ts_recent to zero will at least satisfy the 2201 * requirement that zero be placed in the timestamp 2202 * echo reply when ts_recent isn't valid. The 2203 * age isn't reset until we get a valid ts_recent 2204 * because we don't want out-of-order segments to be 2205 * dropped when ts_recent is old. 2206 */ 2207 tp->ts_recent = 0; 2208 } else { 2209 TCPSTAT_INC(tcps_rcvduppack); 2210 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2211 TCPSTAT_INC(tcps_pawsdrop); 2212 if (tlen) 2213 goto dropafterack; 2214 goto drop; 2215 } 2216 } 2217 2218 /* 2219 * In the SYN-RECEIVED state, validate that the packet belongs to 2220 * this connection before trimming the data to fit the receive 2221 * window. Check the sequence number versus IRS since we know 2222 * the sequence numbers haven't wrapped. This is a partial fix 2223 * for the "LAND" DoS attack. 2224 */ 2225 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2226 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2227 goto dropwithreset; 2228 } 2229 2230 todrop = tp->rcv_nxt - th->th_seq; 2231 if (todrop > 0) { 2232 if (thflags & TH_SYN) { 2233 thflags &= ~TH_SYN; 2234 th->th_seq++; 2235 if (th->th_urp > 1) 2236 th->th_urp--; 2237 else 2238 thflags &= ~TH_URG; 2239 todrop--; 2240 } 2241 /* 2242 * Following if statement from Stevens, vol. 2, p. 960. 2243 */ 2244 if (todrop > tlen 2245 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2246 /* 2247 * Any valid FIN must be to the left of the window. 2248 * At this point the FIN must be a duplicate or out 2249 * of sequence; drop it. 2250 */ 2251 thflags &= ~TH_FIN; 2252 2253 /* 2254 * Send an ACK to resynchronize and drop any data. 2255 * But keep on processing for RST or ACK. 2256 */ 2257 tp->t_flags |= TF_ACKNOW; 2258 todrop = tlen; 2259 TCPSTAT_INC(tcps_rcvduppack); 2260 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2261 } else { 2262 TCPSTAT_INC(tcps_rcvpartduppack); 2263 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2264 } 2265 /* 2266 * DSACK - add SACK block for dropped range 2267 */ 2268 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2269 tcp_update_sack_list(tp, th->th_seq, 2270 th->th_seq + todrop); 2271 /* 2272 * ACK now, as the next in-sequence segment 2273 * will clear the DSACK block again 2274 */ 2275 tp->t_flags |= TF_ACKNOW; 2276 } 2277 drop_hdrlen += todrop; /* drop from the top afterwards */ 2278 th->th_seq += todrop; 2279 tlen -= todrop; 2280 if (th->th_urp > todrop) 2281 th->th_urp -= todrop; 2282 else { 2283 thflags &= ~TH_URG; 2284 th->th_urp = 0; 2285 } 2286 } 2287 2288 /* 2289 * If new data are received on a connection after the 2290 * user processes are gone, then RST the other end if 2291 * no FIN has been processed. 2292 */ 2293 if ((tp->t_flags & TF_CLOSED) && tlen > 0 && 2294 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2295 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2296 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2297 "after socket was closed, " 2298 "sending RST and removing tcpcb\n", 2299 s, __func__, tcpstates[tp->t_state], tlen); 2300 free(s, M_TCPLOG); 2301 } 2302 tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE); 2303 /* tcp_close will kill the inp pre-log the Reset */ 2304 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST); 2305 tp = tcp_close(tp); 2306 TCPSTAT_INC(tcps_rcvafterclose); 2307 goto dropwithreset; 2308 } 2309 2310 /* 2311 * If segment ends after window, drop trailing data 2312 * (and PUSH and FIN); if nothing left, just ACK. 2313 */ 2314 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2315 if (todrop > 0) { 2316 TCPSTAT_INC(tcps_rcvpackafterwin); 2317 if (todrop >= tlen) { 2318 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2319 /* 2320 * If window is closed can only take segments at 2321 * window edge, and have to drop data and PUSH from 2322 * incoming segments. Continue processing, but 2323 * remember to ack. Otherwise, drop segment 2324 * and ack. 2325 */ 2326 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2327 tp->t_flags |= TF_ACKNOW; 2328 TCPSTAT_INC(tcps_rcvwinprobe); 2329 } else 2330 goto dropafterack; 2331 } else 2332 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2333 m_adj(m, -todrop); 2334 tlen -= todrop; 2335 thflags &= ~(TH_PUSH|TH_FIN); 2336 } 2337 2338 /* 2339 * If last ACK falls within this segment's sequence numbers, 2340 * record its timestamp. 2341 * NOTE: 2342 * 1) That the test incorporates suggestions from the latest 2343 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2344 * 2) That updating only on newer timestamps interferes with 2345 * our earlier PAWS tests, so this check should be solely 2346 * predicated on the sequence space of this segment. 2347 * 3) That we modify the segment boundary check to be 2348 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2349 * instead of RFC1323's 2350 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2351 * This modified check allows us to overcome RFC1323's 2352 * limitations as described in Stevens TCP/IP Illustrated 2353 * Vol. 2 p.869. In such cases, we can still calculate the 2354 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2355 */ 2356 if ((to.to_flags & TOF_TS) != 0 && 2357 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2358 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2359 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2360 tp->ts_recent_age = tcp_ts_getticks(); 2361 tp->ts_recent = to.to_tsval; 2362 } 2363 2364 /* 2365 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2366 * flag is on (half-synchronized state), then queue data for 2367 * later processing; else drop segment and return. 2368 */ 2369 if ((thflags & TH_ACK) == 0) { 2370 if (tp->t_state == TCPS_SYN_RECEIVED || 2371 (tp->t_flags & TF_NEEDSYN)) { 2372 if (tp->t_state == TCPS_SYN_RECEIVED && 2373 (tp->t_flags & TF_FASTOPEN)) { 2374 tp->snd_wnd = tiwin; 2375 cc_conn_init(tp); 2376 } 2377 goto step6; 2378 } else if (tp->t_flags & TF_ACKNOW) 2379 goto dropafterack; 2380 else 2381 goto drop; 2382 } 2383 2384 /* 2385 * Ack processing. 2386 */ 2387 if (SEQ_GEQ(tp->snd_una, tp->iss + (TCP_MAXWIN << tp->snd_scale))) { 2388 /* Checking SEG.ACK against ISS is definitely redundant. */ 2389 tp->t_flags2 |= TF2_NO_ISS_CHECK; 2390 } 2391 if (!V_tcp_insecure_ack) { 2392 tcp_seq seq_min; 2393 bool ghost_ack_check; 2394 2395 if (tp->t_flags2 & TF2_NO_ISS_CHECK) { 2396 /* Check for too old ACKs (RFC 5961, Section 5.2). */ 2397 seq_min = tp->snd_una - tp->max_sndwnd; 2398 ghost_ack_check = false; 2399 } else { 2400 if (SEQ_GT(tp->iss + 1, tp->snd_una - tp->max_sndwnd)) { 2401 /* Checking for ghost ACKs is stricter. */ 2402 seq_min = tp->iss + 1; 2403 ghost_ack_check = true; 2404 } else { 2405 /* 2406 * Checking for too old ACKs (RFC 5961, 2407 * Section 5.2) is stricter. 2408 */ 2409 seq_min = tp->snd_una - tp->max_sndwnd; 2410 ghost_ack_check = false; 2411 } 2412 } 2413 if (SEQ_LT(th->th_ack, seq_min)) { 2414 if (ghost_ack_check) 2415 TCPSTAT_INC(tcps_rcvghostack); 2416 else 2417 TCPSTAT_INC(tcps_rcvacktooold); 2418 tcp_send_challenge_ack(tp, th, m); 2419 m = NULL; 2420 goto drop; 2421 } 2422 } 2423 switch (tp->t_state) { 2424 /* 2425 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2426 * ESTABLISHED state and continue processing. 2427 * The ACK was checked above. 2428 */ 2429 case TCPS_SYN_RECEIVED: 2430 2431 TCPSTAT_INC(tcps_connects); 2432 if (tp->t_flags & TF_SONOTCONN) { 2433 /* 2434 * Usually SYN_RECEIVED had been created from a LISTEN, 2435 * and solisten_enqueue() has already marked the socket 2436 * layer as connected. If it didn't, which can happen 2437 * only with an accept_filter(9), then the tp is marked 2438 * with TF_SONOTCONN. The other reason for this mark 2439 * to be set is a simultaneous open, a SYN_RECEIVED 2440 * that had been created from SYN_SENT. 2441 */ 2442 tp->t_flags &= ~TF_SONOTCONN; 2443 soisconnected(so); 2444 } 2445 /* Do window scaling? */ 2446 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2447 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2448 tp->rcv_scale = tp->request_r_scale; 2449 } 2450 tp->snd_wnd = tiwin; 2451 /* 2452 * Make transitions: 2453 * SYN-RECEIVED -> ESTABLISHED 2454 * SYN-RECEIVED* -> FIN-WAIT-1 2455 */ 2456 tp->t_starttime = ticks; 2457 if ((tp->t_flags & TF_FASTOPEN) && tp->t_tfo_pending) { 2458 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2459 tp->t_tfo_pending = NULL; 2460 } 2461 if (tp->t_flags & TF_NEEDFIN) { 2462 tp->t_acktime = ticks; 2463 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2464 tp->t_flags &= ~TF_NEEDFIN; 2465 } else { 2466 tcp_state_change(tp, TCPS_ESTABLISHED); 2467 TCP_PROBE5(accept__established, NULL, tp, 2468 m, tp, th); 2469 /* 2470 * TFO connections call cc_conn_init() during SYN 2471 * processing. Calling it again here for such 2472 * connections is not harmless as it would undo the 2473 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2474 * is retransmitted. 2475 */ 2476 if (!(tp->t_flags & TF_FASTOPEN)) 2477 cc_conn_init(tp); 2478 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2479 } 2480 /* 2481 * Account for the ACK of our SYN prior to 2482 * regular ACK processing below, except for 2483 * simultaneous SYN, which is handled later. 2484 */ 2485 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2486 incforsyn = 1; 2487 /* 2488 * If segment contains data or ACK, will call tcp_reass() 2489 * later; if not, do so now to pass queued data to user. 2490 */ 2491 if (tlen == 0 && (thflags & TH_FIN) == 0) { 2492 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2493 (struct mbuf *)0); 2494 tcp_handle_wakeup(tp); 2495 } 2496 tp->snd_wl1 = th->th_seq - 1; 2497 /* FALLTHROUGH */ 2498 2499 /* 2500 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2501 * ACKs. If the ack is in the range 2502 * tp->snd_una < th->th_ack <= tp->snd_max 2503 * then advance tp->snd_una to th->th_ack and drop 2504 * data from the retransmission queue. If this ACK reflects 2505 * more up to date window information we update our window information. 2506 */ 2507 case TCPS_ESTABLISHED: 2508 case TCPS_FIN_WAIT_1: 2509 case TCPS_FIN_WAIT_2: 2510 case TCPS_CLOSE_WAIT: 2511 case TCPS_CLOSING: 2512 case TCPS_LAST_ACK: 2513 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2514 TCPSTAT_INC(tcps_rcvacktoomuch); 2515 goto dropafterack; 2516 } 2517 if (tcp_is_sack_recovery(tp, &to)) { 2518 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2519 if ((sack_changed != SACK_NOCHANGE) && 2520 (tp->t_flags & TF_LRD)) { 2521 tcp_sack_lost_retransmission(tp, th); 2522 } 2523 } else 2524 /* 2525 * Reset the value so that previous (valid) value 2526 * from the last ack with SACK doesn't get used. 2527 */ 2528 tp->sackhint.sacked_bytes = 0; 2529 2530 #ifdef TCP_HHOOK 2531 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2532 hhook_run_tcp_est_in(tp, th, &to); 2533 #endif 2534 2535 if (SEQ_LT(th->th_ack, tp->snd_una)) { 2536 /* This is old ACK information, don't process it. */ 2537 break; 2538 } 2539 if (th->th_ack == tp->snd_una) { 2540 /* Check if this is a duplicate ACK. */ 2541 if (tp->t_flags & TF_SACK_PERMIT) { 2542 /* 2543 * If SEG.ACK == SND.UNA, RFC 6675 requires a 2544 * duplicate ACK to selectively acknowledge 2545 * at least one byte, which was not selectively 2546 * acknowledged before. 2547 */ 2548 if (sack_changed == SACK_NOCHANGE) { 2549 break; 2550 } 2551 } else { 2552 /* 2553 * If SEG.ACK == SND.UNA, RFC 5681 requires a 2554 * duplicate ACK to have no data on it and to 2555 * not be a window update. 2556 */ 2557 if (!no_data || tiwin != tp->snd_wnd) { 2558 break; 2559 } 2560 } 2561 /* 2562 * If this is the first time we've seen a 2563 * FIN from the remote, this is not a 2564 * duplicate ACK and it needs to be processed 2565 * normally. 2566 * This happens during a simultaneous close. 2567 */ 2568 if ((thflags & TH_FIN) && 2569 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2570 tp->t_dupacks = 0; 2571 break; 2572 } 2573 /* Perform duplicate ACK processing. */ 2574 TCPSTAT_INC(tcps_rcvdupack); 2575 maxseg = tcp_maxseg(tp); 2576 if (!tcp_timer_active(tp, TT_REXMT)) { 2577 tp->t_dupacks = 0; 2578 } else if (++tp->t_dupacks > tcprexmtthresh || 2579 IN_FASTRECOVERY(tp->t_flags)) { 2580 cc_ack_received(tp, th, nsegs, CC_DUPACK); 2581 if (V_tcp_do_prr && 2582 IN_FASTRECOVERY(tp->t_flags) && 2583 (tp->t_flags & TF_SACK_PERMIT)) { 2584 tcp_do_prr_ack(tp, th, &to, 2585 sack_changed, &maxseg); 2586 } else if (tcp_is_sack_recovery(tp, &to) && 2587 IN_FASTRECOVERY(tp->t_flags) && 2588 (tp->snd_nxt == tp->snd_max)) { 2589 int awnd; 2590 2591 /* 2592 * Compute the amount of data in flight first. 2593 * We can inject new data into the pipe iff 2594 * we have less than ssthresh 2595 * worth of data in flight. 2596 */ 2597 awnd = tcp_compute_pipe(tp); 2598 if (awnd < tp->snd_ssthresh) { 2599 tp->snd_cwnd += imax(maxseg, 2600 imin(2 * maxseg, 2601 tp->sackhint.delivered_data)); 2602 if (tp->snd_cwnd > tp->snd_ssthresh) 2603 tp->snd_cwnd = tp->snd_ssthresh; 2604 } 2605 } else if (tcp_is_sack_recovery(tp, &to) && 2606 IN_FASTRECOVERY(tp->t_flags) && 2607 SEQ_LT(tp->snd_nxt, tp->snd_max)) { 2608 tp->snd_cwnd += imax(maxseg, 2609 imin(2 * maxseg, 2610 tp->sackhint.delivered_data)); 2611 } else { 2612 tp->snd_cwnd += maxseg; 2613 } 2614 (void) tcp_output(tp); 2615 goto drop; 2616 } else if (tp->t_dupacks == tcprexmtthresh || 2617 (tp->t_flags & TF_SACK_PERMIT && 2618 tp->sackhint.sacked_bytes > 2619 (tcprexmtthresh - 1) * maxseg)) { 2620 enter_recovery: 2621 /* 2622 * Above is the RFC6675 trigger condition of 2623 * more than (dupthresh-1)*maxseg sacked data. 2624 * If the count of holes in the 2625 * scoreboard is >= dupthresh, we could 2626 * also enter loss recovery, but don't 2627 * have that value readily available. 2628 */ 2629 tp->t_dupacks = tcprexmtthresh; 2630 tcp_seq onxt = tp->snd_nxt; 2631 2632 /* 2633 * If we're doing sack, check to 2634 * see if we're already in sack 2635 * recovery. If we're not doing sack, 2636 * check to see if we're in newreno 2637 * recovery. 2638 */ 2639 if (tcp_is_sack_recovery(tp, &to)) { 2640 if (IN_FASTRECOVERY(tp->t_flags)) { 2641 tp->t_dupacks = 0; 2642 break; 2643 } 2644 } else { 2645 if (SEQ_LEQ(th->th_ack, 2646 tp->snd_recover)) { 2647 tp->t_dupacks = 0; 2648 break; 2649 } 2650 } 2651 /* Congestion signal before ack. */ 2652 cc_cong_signal(tp, th, CC_NDUPACK); 2653 cc_ack_received(tp, th, nsegs, CC_DUPACK); 2654 tcp_timer_activate(tp, TT_REXMT, 0); 2655 tp->t_rtttime = 0; 2656 if (V_tcp_do_prr) { 2657 /* 2658 * snd_ssthresh and snd_recover are 2659 * already updated by cc_cong_signal. 2660 */ 2661 if (tcp_is_sack_recovery(tp, &to)) { 2662 /* 2663 * Include Limited Transmit 2664 * segments here 2665 */ 2666 tp->sackhint.prr_delivered = 2667 imin(tp->snd_max - th->th_ack, 2668 (tp->snd_limited + 1) * maxseg); 2669 } else { 2670 tp->sackhint.prr_delivered = 2671 maxseg; 2672 } 2673 tp->sackhint.recover_fs = max(1, 2674 tp->snd_nxt - tp->snd_una); 2675 } 2676 tp->snd_limited = 0; 2677 if (tcp_is_sack_recovery(tp, &to)) { 2678 TCPSTAT_INC(tcps_sack_recovery_episode); 2679 /* 2680 * When entering LR after RTO due to 2681 * Duplicate ACKs, retransmit existing 2682 * holes from the scoreboard. 2683 */ 2684 tcp_resend_sackholes(tp); 2685 /* Avoid inflating cwnd in tcp_output */ 2686 tp->snd_nxt = tp->snd_max; 2687 tp->snd_cwnd = tcp_compute_pipe(tp) + 2688 maxseg; 2689 (void) tcp_output(tp); 2690 /* Set cwnd to the expected flightsize */ 2691 tp->snd_cwnd = tp->snd_ssthresh; 2692 goto drop; 2693 } 2694 tp->snd_nxt = th->th_ack; 2695 tp->snd_cwnd = maxseg; 2696 (void) tcp_output(tp); 2697 KASSERT(tp->snd_limited <= 2, 2698 ("%s: tp->snd_limited too big", 2699 __func__)); 2700 tp->snd_cwnd = tp->snd_ssthresh + 2701 maxseg * 2702 (tp->t_dupacks - tp->snd_limited); 2703 if (SEQ_GT(onxt, tp->snd_nxt)) 2704 tp->snd_nxt = onxt; 2705 goto drop; 2706 } else if (V_tcp_do_rfc3042) { 2707 /* 2708 * Process first and second duplicate 2709 * ACKs. Each indicates a segment 2710 * leaving the network, creating room 2711 * for more. Make sure we can send a 2712 * packet on reception of each duplicate 2713 * ACK by increasing snd_cwnd by one 2714 * segment. Restore the original 2715 * snd_cwnd after packet transmission. 2716 */ 2717 cc_ack_received(tp, th, nsegs, CC_DUPACK); 2718 uint32_t oldcwnd = tp->snd_cwnd; 2719 tcp_seq oldsndmax = tp->snd_max; 2720 u_int sent; 2721 int avail; 2722 2723 KASSERT(tp->t_dupacks == 1 || 2724 tp->t_dupacks == 2, 2725 ("%s: dupacks not 1 or 2", 2726 __func__)); 2727 if (tp->t_dupacks == 1) 2728 tp->snd_limited = 0; 2729 if ((tp->snd_nxt == tp->snd_max) && 2730 (tp->t_rxtshift == 0)) 2731 tp->snd_cwnd = 2732 SEQ_SUB(tp->snd_nxt, tp->snd_una); 2733 tp->snd_cwnd += 2734 (tp->t_dupacks - tp->snd_limited) * maxseg; 2735 tp->snd_cwnd -= tcp_sack_adjust(tp); 2736 /* 2737 * Only call tcp_output when there 2738 * is new data available to be sent 2739 * or we need to send an ACK. 2740 */ 2741 SOCK_SENDBUF_LOCK(so); 2742 avail = sbavail(&so->so_snd); 2743 SOCK_SENDBUF_UNLOCK(so); 2744 if (tp->t_flags & TF_ACKNOW || 2745 (avail >= 2746 SEQ_SUB(tp->snd_nxt, tp->snd_una))) { 2747 (void) tcp_output(tp); 2748 } 2749 sent = SEQ_SUB(tp->snd_max, oldsndmax); 2750 if (sent > maxseg) { 2751 KASSERT((tp->t_dupacks == 2 && 2752 tp->snd_limited == 0) || 2753 (sent == maxseg + 1 && 2754 tp->t_flags & TF_SENTFIN) || 2755 (sent < 2 * maxseg && 2756 tp->t_flags & TF_NODELAY), 2757 ("%s: sent too much: %u>%u", 2758 __func__, sent, maxseg)); 2759 tp->snd_limited = 2; 2760 } else if (sent > 0) { 2761 ++tp->snd_limited; 2762 } 2763 tp->snd_cwnd = oldcwnd; 2764 goto drop; 2765 } 2766 break; 2767 } 2768 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2769 ("%s: SEQ_LEQ(th_ack, snd_una)", __func__)); 2770 /* 2771 * This ack is advancing the left edge, reset the 2772 * counter. 2773 */ 2774 tp->t_dupacks = 0; 2775 /* 2776 * If this ack also has new SACK info, increment the 2777 * t_dupacks as per RFC 6675. The variable 2778 * sack_changed tracks all changes to the SACK 2779 * scoreboard, including when partial ACKs without 2780 * SACK options are received, and clear the scoreboard 2781 * from the left side. Such partial ACKs should not be 2782 * counted as dupacks here. 2783 */ 2784 if (tcp_is_sack_recovery(tp, &to) && 2785 (((tp->t_rxtshift == 0) && (sack_changed != SACK_NOCHANGE)) || 2786 ((tp->t_rxtshift > 0) && (sack_changed == SACK_NEWLOSS))) && 2787 (tp->snd_nxt == tp->snd_max)) { 2788 tp->t_dupacks++; 2789 /* limit overhead by setting maxseg last */ 2790 if (!IN_FASTRECOVERY(tp->t_flags) && 2791 (tp->sackhint.sacked_bytes > 2792 (tcprexmtthresh - 1) * (maxseg = tcp_maxseg(tp)))) { 2793 goto enter_recovery; 2794 } 2795 } 2796 /* 2797 * If the congestion window was inflated to account 2798 * for the other side's cached packets, retract it. 2799 */ 2800 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2801 if (IN_FASTRECOVERY(tp->t_flags)) { 2802 if (tp->t_flags & TF_SACK_PERMIT) { 2803 if (V_tcp_do_prr && 2804 (to.to_flags & TOF_SACK)) { 2805 tcp_timer_activate(tp, 2806 TT_REXMT, 0); 2807 tp->t_rtttime = 0; 2808 tcp_do_prr_ack(tp, th, &to, 2809 sack_changed, &maxseg); 2810 tp->t_flags |= TF_ACKNOW; 2811 (void) tcp_output(tp); 2812 } else { 2813 tcp_sack_partialack(tp, th, 2814 &maxseg); 2815 } 2816 } else { 2817 tcp_newreno_partial_ack(tp, th); 2818 } 2819 } else if (IN_CONGRECOVERY(tp->t_flags) && 2820 (V_tcp_do_prr)) { 2821 tp->sackhint.delivered_data = 2822 BYTES_THIS_ACK(tp, th); 2823 tp->snd_fack = th->th_ack; 2824 /* 2825 * During ECN cwnd reduction 2826 * always use PRR-SSRB 2827 */ 2828 tcp_do_prr_ack(tp, th, &to, SACK_CHANGE, 2829 &maxseg); 2830 (void) tcp_output(tp); 2831 } 2832 } 2833 /* 2834 * If we reach this point, ACK is not a duplicate, 2835 * i.e., it ACKs something we sent. 2836 */ 2837 if (tp->t_flags & TF_NEEDSYN) { 2838 /* 2839 * T/TCP: Connection was half-synchronized, and our 2840 * SYN has been ACK'd (so connection is now fully 2841 * synchronized). Go to non-starred state, 2842 * increment snd_una for ACK of SYN, and check if 2843 * we can do window scaling. 2844 */ 2845 tp->t_flags &= ~TF_NEEDSYN; 2846 tp->snd_una++; 2847 /* Do window scaling? */ 2848 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2849 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2850 tp->rcv_scale = tp->request_r_scale; 2851 /* Send window already scaled. */ 2852 } 2853 } 2854 2855 process_ACK: 2856 INP_WLOCK_ASSERT(inp); 2857 2858 /* 2859 * Adjust for the SYN bit in sequence space, 2860 * but don't account for it in cwnd calculations. 2861 * This is for the SYN_RECEIVED, non-simultaneous 2862 * SYN case. SYN_SENT and simultaneous SYN are 2863 * treated elsewhere. 2864 */ 2865 if (incforsyn) 2866 tp->snd_una++; 2867 acked = BYTES_THIS_ACK(tp, th); 2868 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2869 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2870 tp->snd_una, th->th_ack, tp, m)); 2871 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2872 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2873 2874 /* 2875 * If we just performed our first retransmit, and the ACK 2876 * arrives within our recovery window, then it was a mistake 2877 * to do the retransmit in the first place. Recover our 2878 * original cwnd and ssthresh, and proceed to transmit where 2879 * we left off. 2880 */ 2881 if (tp->t_rxtshift == 1 && 2882 tp->t_flags & TF_PREVVALID && 2883 tp->t_badrxtwin != 0 && 2884 to.to_flags & TOF_TS && 2885 to.to_tsecr != 0 && 2886 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 2887 cc_cong_signal(tp, th, CC_RTO_ERR); 2888 2889 /* 2890 * If we have a timestamp reply, update smoothed 2891 * round trip time. If no timestamp is present but 2892 * transmit timer is running and timed sequence 2893 * number was acked, update smoothed round trip time. 2894 * Since we now have an rtt measurement, cancel the 2895 * timer backoff (cf., Phil Karn's retransmit alg.). 2896 * Recompute the initial retransmit timer. 2897 * 2898 * Some boxes send broken timestamp replies 2899 * during the SYN+ACK phase, ignore 2900 * timestamps of 0 or we could calculate a 2901 * huge RTT and blow up the retransmit timer. 2902 */ 2903 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2904 uint32_t t; 2905 2906 t = tcp_ts_getticks() - to.to_tsecr; 2907 if (!tp->t_rttlow || tp->t_rttlow > t) 2908 tp->t_rttlow = t; 2909 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2910 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2911 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2912 tp->t_rttlow = ticks - tp->t_rtttime; 2913 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2914 } 2915 2916 SOCK_SENDBUF_LOCK(so); 2917 /* 2918 * Clear t_acktime if remote side has ACKd all data in the 2919 * socket buffer and FIN (if applicable). 2920 * Otherwise, update t_acktime if we received a sufficiently 2921 * large ACK. 2922 */ 2923 if ((tp->t_state <= TCPS_CLOSE_WAIT && 2924 acked == sbavail(&so->so_snd)) || 2925 acked > sbavail(&so->so_snd)) 2926 tp->t_acktime = 0; 2927 else if (acked > 1) 2928 tp->t_acktime = ticks; 2929 2930 /* 2931 * If all outstanding data is acked, stop retransmit 2932 * timer and remember to restart (more output or persist). 2933 * If there is more data to be acked, restart retransmit 2934 * timer, using current (possibly backed-off) value. 2935 */ 2936 if (th->th_ack == tp->snd_max) { 2937 tcp_timer_activate(tp, TT_REXMT, 0); 2938 needoutput = 1; 2939 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2940 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); 2941 2942 /* 2943 * If no data (only SYN) was ACK'd, 2944 * skip rest of ACK processing. 2945 */ 2946 if (acked == 0) { 2947 SOCK_SENDBUF_UNLOCK(so); 2948 goto step6; 2949 } 2950 2951 /* 2952 * Let the congestion control algorithm update congestion 2953 * control related information. This typically means increasing 2954 * the congestion window. 2955 */ 2956 cc_ack_received(tp, th, nsegs, CC_ACK); 2957 2958 if (acked > sbavail(&so->so_snd)) { 2959 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2960 tp->snd_wnd -= sbavail(&so->so_snd); 2961 else 2962 tp->snd_wnd = 0; 2963 mfree = sbcut_locked(&so->so_snd, 2964 (int)sbavail(&so->so_snd)); 2965 ourfinisacked = 1; 2966 } else { 2967 mfree = sbcut_locked(&so->so_snd, acked); 2968 if (tp->snd_wnd >= (uint32_t) acked) 2969 tp->snd_wnd -= acked; 2970 else 2971 tp->snd_wnd = 0; 2972 ourfinisacked = 0; 2973 } 2974 /* NB: sowwakeup_locked() does an implicit unlock. */ 2975 sowwakeup_locked(so); 2976 m_freem(mfree); 2977 /* Detect una wraparound. */ 2978 if (!IN_RECOVERY(tp->t_flags) && 2979 SEQ_GT(tp->snd_una, tp->snd_recover) && 2980 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2981 tp->snd_recover = th->th_ack - 1; 2982 tp->snd_una = th->th_ack; 2983 if (IN_RECOVERY(tp->t_flags) && 2984 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2985 cc_post_recovery(tp, th); 2986 } 2987 if (SEQ_GT(tp->snd_una, tp->snd_recover)) { 2988 tp->snd_recover = tp->snd_una; 2989 } 2990 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2991 tp->snd_nxt = tp->snd_una; 2992 2993 switch (tp->t_state) { 2994 /* 2995 * In FIN_WAIT_1 STATE in addition to the processing 2996 * for the ESTABLISHED state if our FIN is now acknowledged 2997 * then enter FIN_WAIT_2. 2998 */ 2999 case TCPS_FIN_WAIT_1: 3000 if (ourfinisacked) { 3001 /* 3002 * If we can't receive any more 3003 * data, then closing user can proceed. 3004 * Starting the timer is contrary to the 3005 * specification, but if we don't get a FIN 3006 * we'll hang forever. 3007 */ 3008 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3009 tcp_free_sackholes(tp); 3010 soisdisconnected(so); 3011 tcp_timer_activate(tp, TT_2MSL, 3012 (tcp_fast_finwait2_recycle ? 3013 tcp_finwait2_timeout : 3014 TP_MAXIDLE(tp))); 3015 } 3016 tcp_state_change(tp, TCPS_FIN_WAIT_2); 3017 } 3018 break; 3019 3020 /* 3021 * In CLOSING STATE in addition to the processing for 3022 * the ESTABLISHED state if the ACK acknowledges our FIN 3023 * then enter the TIME-WAIT state, otherwise ignore 3024 * the segment. 3025 */ 3026 case TCPS_CLOSING: 3027 if (ourfinisacked) { 3028 tcp_twstart(tp); 3029 m_freem(m); 3030 return; 3031 } 3032 break; 3033 3034 /* 3035 * In LAST_ACK, we may still be waiting for data to drain 3036 * and/or to be acked, as well as for the ack of our FIN. 3037 * If our FIN is now acknowledged, delete the TCB, 3038 * enter the closed state and return. 3039 */ 3040 case TCPS_LAST_ACK: 3041 if (ourfinisacked) { 3042 tp = tcp_close(tp); 3043 goto drop; 3044 } 3045 break; 3046 } 3047 } 3048 3049 step6: 3050 INP_WLOCK_ASSERT(inp); 3051 3052 /* 3053 * Update window information. 3054 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3055 */ 3056 if ((thflags & TH_ACK) && 3057 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3058 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3059 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3060 /* keep track of pure window updates */ 3061 if (no_data && tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3062 TCPSTAT_INC(tcps_rcvwinupd); 3063 tp->snd_wnd = tiwin; 3064 tp->snd_wl1 = th->th_seq; 3065 tp->snd_wl2 = th->th_ack; 3066 if (tp->snd_wnd > tp->max_sndwnd) 3067 tp->max_sndwnd = tp->snd_wnd; 3068 needoutput = 1; 3069 } 3070 3071 /* 3072 * Process segments with URG. 3073 */ 3074 if ((thflags & TH_URG) && th->th_urp && 3075 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3076 /* 3077 * This is a kludge, but if we receive and accept 3078 * random urgent pointers, we'll crash in 3079 * soreceive. It's hard to imagine someone 3080 * actually wanting to send this much urgent data. 3081 */ 3082 SOCK_RECVBUF_LOCK(so); 3083 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3084 th->th_urp = 0; /* XXX */ 3085 thflags &= ~TH_URG; /* XXX */ 3086 SOCK_RECVBUF_UNLOCK(so); /* XXX */ 3087 goto dodata; /* XXX */ 3088 } 3089 /* 3090 * If this segment advances the known urgent pointer, 3091 * then mark the data stream. This should not happen 3092 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3093 * a FIN has been received from the remote side. 3094 * In these states we ignore the URG. 3095 * 3096 * According to RFC961 (Assigned Protocols), 3097 * the urgent pointer points to the last octet 3098 * of urgent data. We continue, however, 3099 * to consider it to indicate the first octet 3100 * of data past the urgent section as the original 3101 * spec states (in one of two places). 3102 */ 3103 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3104 tp->rcv_up = th->th_seq + th->th_urp; 3105 so->so_oobmark = sbavail(&so->so_rcv) + 3106 (tp->rcv_up - tp->rcv_nxt) - 1; 3107 if (so->so_oobmark == 0) 3108 so->so_rcv.sb_state |= SBS_RCVATMARK; 3109 sohasoutofband(so); 3110 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3111 } 3112 SOCK_RECVBUF_UNLOCK(so); 3113 /* 3114 * Remove out of band data so doesn't get presented to user. 3115 * This can happen independent of advancing the URG pointer, 3116 * but if two URG's are pending at once, some out-of-band 3117 * data may creep in... ick. 3118 */ 3119 if (th->th_urp <= (uint32_t)tlen && 3120 !(so->so_options & SO_OOBINLINE)) { 3121 /* hdr drop is delayed */ 3122 tcp_pulloutofband(so, th, m, drop_hdrlen); 3123 } 3124 } else { 3125 /* 3126 * If no out of band data is expected, 3127 * pull receive urgent pointer along 3128 * with the receive window. 3129 */ 3130 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3131 tp->rcv_up = tp->rcv_nxt; 3132 } 3133 dodata: /* XXX */ 3134 INP_WLOCK_ASSERT(inp); 3135 3136 /* 3137 * Process the segment text, merging it into the TCP sequencing queue, 3138 * and arranging for acknowledgment of receipt if necessary. 3139 * This process logically involves adjusting tp->rcv_wnd as data 3140 * is presented to the user (this happens in tcp_usrreq.c, 3141 * case PRU_RCVD). If a FIN has already been received on this 3142 * connection then we just ignore the text. 3143 */ 3144 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3145 (tp->t_flags & TF_FASTOPEN)); 3146 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3147 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3148 tcp_seq save_start = th->th_seq; 3149 tcp_seq save_rnxt = tp->rcv_nxt; 3150 int save_tlen = tlen; 3151 m_adj(m, drop_hdrlen); /* delayed header drop */ 3152 /* 3153 * Insert segment which includes th into TCP reassembly queue 3154 * with control block tp. Set thflags to whether reassembly now 3155 * includes a segment with FIN. This handles the common case 3156 * inline (segment is the next to be received on an established 3157 * connection, and the queue is empty), avoiding linkage into 3158 * and removal from the queue and repetition of various 3159 * conversions. 3160 * Set DELACK for segments received in order, but ack 3161 * immediately when segments are out of order (so 3162 * fast retransmit can work). 3163 */ 3164 if (th->th_seq == tp->rcv_nxt && 3165 SEGQ_EMPTY(tp) && 3166 (TCPS_HAVEESTABLISHED(tp->t_state) || 3167 tfo_syn)) { 3168 if (DELAY_ACK(tp, tlen) || tfo_syn) 3169 tp->t_flags |= TF_DELACK; 3170 else 3171 tp->t_flags |= TF_ACKNOW; 3172 tp->rcv_nxt += tlen; 3173 if (tlen && 3174 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3175 (tp->t_fbyte_in == 0)) { 3176 tp->t_fbyte_in = ticks; 3177 if (tp->t_fbyte_in == 0) 3178 tp->t_fbyte_in = 1; 3179 if (tp->t_fbyte_out && tp->t_fbyte_in) 3180 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3181 } 3182 thflags = tcp_get_flags(th) & TH_FIN; 3183 TCPSTAT_INC(tcps_rcvpack); 3184 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3185 SOCK_RECVBUF_LOCK(so); 3186 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3187 m_freem(m); 3188 else 3189 sbappendstream_locked(&so->so_rcv, m, 0); 3190 tp->t_flags |= TF_WAKESOR; 3191 } else { 3192 /* 3193 * XXX: Due to the header drop above "th" is 3194 * theoretically invalid by now. Fortunately 3195 * m_adj() doesn't actually frees any mbufs 3196 * when trimming from the head. 3197 */ 3198 tcp_seq temp = save_start; 3199 3200 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3201 tp->t_flags |= TF_ACKNOW; 3202 } 3203 if ((tp->t_flags & TF_SACK_PERMIT) && 3204 (save_tlen > 0) && 3205 TCPS_HAVEESTABLISHED(tp->t_state)) { 3206 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3207 /* 3208 * DSACK actually handled in the fastpath 3209 * above. 3210 */ 3211 tcp_update_sack_list(tp, save_start, 3212 save_start + save_tlen); 3213 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3214 if ((tp->rcv_numsacks >= 1) && 3215 (tp->sackblks[0].end == save_start)) { 3216 /* 3217 * Partial overlap, recorded at todrop 3218 * above. 3219 */ 3220 tcp_update_sack_list(tp, 3221 tp->sackblks[0].start, 3222 tp->sackblks[0].end); 3223 } else { 3224 tcp_update_dsack_list(tp, save_start, 3225 save_start + save_tlen); 3226 } 3227 } else if (tlen >= save_tlen) { 3228 /* Update of sackblks. */ 3229 tcp_update_dsack_list(tp, save_start, 3230 save_start + save_tlen); 3231 } else if (tlen > 0) { 3232 tcp_update_dsack_list(tp, save_start, 3233 save_start + tlen); 3234 } 3235 } 3236 tcp_handle_wakeup(tp); 3237 #if 0 3238 /* 3239 * Note the amount of data that peer has sent into 3240 * our window, in order to estimate the sender's 3241 * buffer size. 3242 * XXX: Unused. 3243 */ 3244 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3245 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3246 else 3247 len = so->so_rcv.sb_hiwat; 3248 #endif 3249 } else { 3250 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 3251 if (tlen > 0) { 3252 if ((thflags & TH_FIN) != 0) { 3253 log(LOG_DEBUG, "%s; %s: %s: " 3254 "Received %d bytes of data and FIN " 3255 "after having received a FIN, " 3256 "just dropping both\n", 3257 s, __func__, 3258 tcpstates[tp->t_state], tlen); 3259 } else { 3260 log(LOG_DEBUG, "%s; %s: %s: " 3261 "Received %d bytes of data " 3262 "after having received a FIN, " 3263 "just dropping it\n", 3264 s, __func__, 3265 tcpstates[tp->t_state], tlen); 3266 } 3267 } else { 3268 if ((thflags & TH_FIN) != 0) { 3269 log(LOG_DEBUG, "%s; %s: %s: " 3270 "Received FIN " 3271 "after having received a FIN, " 3272 "just dropping it\n", 3273 s, __func__, 3274 tcpstates[tp->t_state]); 3275 } 3276 } 3277 free(s, M_TCPLOG); 3278 } 3279 m_freem(m); 3280 thflags &= ~TH_FIN; 3281 } 3282 3283 /* 3284 * If FIN is received ACK the FIN and let the user know 3285 * that the connection is closing. 3286 */ 3287 if (thflags & TH_FIN) { 3288 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3289 /* The socket upcall is handled by socantrcvmore. */ 3290 socantrcvmore(so); 3291 /* 3292 * If connection is half-synchronized 3293 * (ie NEEDSYN flag on) then delay ACK, 3294 * so it may be piggybacked when SYN is sent. 3295 * Otherwise, since we received a FIN then no 3296 * more input can be expected, send ACK now. 3297 */ 3298 if (tp->t_flags & TF_NEEDSYN) 3299 tp->t_flags |= TF_DELACK; 3300 else 3301 tp->t_flags |= TF_ACKNOW; 3302 tp->rcv_nxt++; 3303 } 3304 switch (tp->t_state) { 3305 /* 3306 * In SYN_RECEIVED and ESTABLISHED STATES 3307 * enter the CLOSE_WAIT state. 3308 */ 3309 case TCPS_SYN_RECEIVED: 3310 tp->t_starttime = ticks; 3311 /* FALLTHROUGH */ 3312 case TCPS_ESTABLISHED: 3313 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3314 break; 3315 3316 /* 3317 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3318 * enter the CLOSING state. 3319 */ 3320 case TCPS_FIN_WAIT_1: 3321 tcp_state_change(tp, TCPS_CLOSING); 3322 break; 3323 3324 /* 3325 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3326 * starting the time-wait timer, turning off the other 3327 * standard timers. 3328 */ 3329 case TCPS_FIN_WAIT_2: 3330 tcp_twstart(tp); 3331 return; 3332 } 3333 } 3334 TCP_PROBE3(debug__input, tp, th, m); 3335 3336 /* 3337 * Return any desired output. 3338 */ 3339 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 3340 (void) tcp_output(tp); 3341 } 3342 check_delack: 3343 INP_WLOCK_ASSERT(inp); 3344 3345 if (tp->t_flags & TF_DELACK) { 3346 tp->t_flags &= ~TF_DELACK; 3347 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3348 } 3349 INP_WUNLOCK(inp); 3350 return; 3351 3352 dropafterack: 3353 /* 3354 * Generate an ACK dropping incoming segment if it occupies 3355 * sequence space, where the ACK reflects our state. 3356 * 3357 * We can now skip the test for the RST flag since all 3358 * paths to this code happen after packets containing 3359 * RST have been dropped. 3360 * 3361 * In the SYN-RECEIVED state, don't send an ACK unless the 3362 * segment we received passes the SYN-RECEIVED ACK test. 3363 * If it fails send a RST. This breaks the loop in the 3364 * "LAND" DoS attack, and also prevents an ACK storm 3365 * between two listening ports that have been sent forged 3366 * SYN segments, each with the source address of the other. 3367 */ 3368 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3369 (SEQ_GT(tp->snd_una, th->th_ack) || 3370 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3371 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 3372 goto dropwithreset; 3373 } 3374 TCP_PROBE3(debug__input, tp, th, m); 3375 tp->t_flags |= TF_ACKNOW; 3376 (void) tcp_output(tp); 3377 INP_WUNLOCK(inp); 3378 m_freem(m); 3379 return; 3380 3381 dropwithreset: 3382 tcp_dropwithreset(m, th, tp, tlen); 3383 if (tp != NULL) { 3384 INP_WUNLOCK(inp); 3385 } 3386 return; 3387 3388 drop: 3389 /* 3390 * Drop space held by incoming segment and return. 3391 */ 3392 TCP_PROBE3(debug__input, tp, th, m); 3393 if (tp != NULL) { 3394 INP_WUNLOCK(inp); 3395 } 3396 m_freem(m); 3397 } 3398 3399 /* 3400 * Issue RST and make ACK acceptable to originator of segment. 3401 * The mbuf must still include the original packet header. 3402 * tp may be NULL. 3403 */ 3404 void 3405 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, int tlen) 3406 { 3407 #ifdef INET 3408 struct ip *ip; 3409 #endif 3410 #ifdef INET6 3411 struct ip6_hdr *ip6; 3412 #endif 3413 3414 if (tp != NULL) { 3415 INP_LOCK_ASSERT(tptoinpcb(tp)); 3416 } 3417 3418 /* Don't bother if destination was broadcast/multicast. */ 3419 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3420 goto drop; 3421 #ifdef INET6 3422 if (mtod(m, struct ip *)->ip_v == 6) { 3423 ip6 = mtod(m, struct ip6_hdr *); 3424 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3425 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3426 goto drop; 3427 /* IPv6 anycast check is done at tcp6_input() */ 3428 } 3429 #endif 3430 #if defined(INET) && defined(INET6) 3431 else 3432 #endif 3433 #ifdef INET 3434 { 3435 ip = mtod(m, struct ip *); 3436 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3437 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3438 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3439 in_ifnet_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3440 goto drop; 3441 } 3442 #endif 3443 3444 /* Perform bandwidth limiting. */ 3445 if (badport_bandlim(BANDLIM_TCP_RST) < 0) 3446 goto drop; 3447 3448 /* tcp_respond consumes the mbuf chain. */ 3449 if (tcp_get_flags(th) & TH_ACK) { 3450 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3451 th->th_ack, TH_RST); 3452 } else { 3453 if (tcp_get_flags(th) & TH_SYN) 3454 tlen++; 3455 if (tcp_get_flags(th) & TH_FIN) 3456 tlen++; 3457 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3458 (tcp_seq)0, TH_RST|TH_ACK); 3459 } 3460 return; 3461 drop: 3462 m_freem(m); 3463 } 3464 3465 /* 3466 * Parse TCP options and place in tcpopt. 3467 */ 3468 void 3469 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3470 { 3471 int opt, optlen; 3472 3473 to->to_flags = 0; 3474 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3475 opt = cp[0]; 3476 if (opt == TCPOPT_EOL) 3477 break; 3478 if (opt == TCPOPT_NOP) 3479 optlen = 1; 3480 else { 3481 if (cnt < 2) 3482 break; 3483 optlen = cp[1]; 3484 if (optlen < 2 || optlen > cnt) 3485 break; 3486 } 3487 switch (opt) { 3488 case TCPOPT_MAXSEG: 3489 if (optlen != TCPOLEN_MAXSEG) 3490 continue; 3491 if (!(flags & TO_SYN)) 3492 continue; 3493 to->to_flags |= TOF_MSS; 3494 bcopy((char *)cp + 2, 3495 (char *)&to->to_mss, sizeof(to->to_mss)); 3496 to->to_mss = ntohs(to->to_mss); 3497 break; 3498 case TCPOPT_WINDOW: 3499 if (optlen != TCPOLEN_WINDOW) 3500 continue; 3501 if (!(flags & TO_SYN)) 3502 continue; 3503 to->to_flags |= TOF_SCALE; 3504 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3505 break; 3506 case TCPOPT_TIMESTAMP: 3507 if (optlen != TCPOLEN_TIMESTAMP) 3508 continue; 3509 to->to_flags |= TOF_TS; 3510 bcopy((char *)cp + 2, 3511 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3512 to->to_tsval = ntohl(to->to_tsval); 3513 bcopy((char *)cp + 6, 3514 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3515 to->to_tsecr = ntohl(to->to_tsecr); 3516 break; 3517 case TCPOPT_SIGNATURE: 3518 /* 3519 * In order to reply to a host which has set the 3520 * TCP_SIGNATURE option in its initial SYN, we have 3521 * to record the fact that the option was observed 3522 * here for the syncache code to perform the correct 3523 * response. 3524 */ 3525 if (optlen != TCPOLEN_SIGNATURE) 3526 continue; 3527 to->to_flags |= TOF_SIGNATURE; 3528 to->to_signature = cp + 2; 3529 break; 3530 case TCPOPT_SACK_PERMITTED: 3531 if (optlen != TCPOLEN_SACK_PERMITTED) 3532 continue; 3533 if (!(flags & TO_SYN)) 3534 continue; 3535 if (!V_tcp_do_sack) 3536 continue; 3537 to->to_flags |= TOF_SACKPERM; 3538 break; 3539 case TCPOPT_SACK: 3540 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3541 continue; 3542 if (flags & TO_SYN) 3543 continue; 3544 to->to_flags |= TOF_SACK; 3545 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3546 to->to_sacks = cp + 2; 3547 TCPSTAT_INC(tcps_sack_rcv_blocks); 3548 break; 3549 case TCPOPT_FAST_OPEN: 3550 /* 3551 * Cookie length validation is performed by the 3552 * server side cookie checking code or the client 3553 * side cookie cache update code. 3554 */ 3555 if (!(flags & TO_SYN)) 3556 continue; 3557 if (!V_tcp_fastopen_client_enable && 3558 !V_tcp_fastopen_server_enable) 3559 continue; 3560 to->to_flags |= TOF_FASTOPEN; 3561 to->to_tfo_len = optlen - 2; 3562 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3563 break; 3564 default: 3565 continue; 3566 } 3567 } 3568 } 3569 3570 /* 3571 * Pull out of band byte out of a segment so 3572 * it doesn't appear in the user's data queue. 3573 * It is still reflected in the segment length for 3574 * sequencing purposes. 3575 */ 3576 void 3577 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3578 int off) 3579 { 3580 int cnt = off + th->th_urp - 1; 3581 3582 while (cnt >= 0) { 3583 if (m->m_len > cnt) { 3584 char *cp = mtod(m, caddr_t) + cnt; 3585 struct tcpcb *tp = sototcpcb(so); 3586 3587 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3588 3589 tp->t_iobc = *cp; 3590 tp->t_oobflags |= TCPOOB_HAVEDATA; 3591 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3592 m->m_len--; 3593 if (m->m_flags & M_PKTHDR) 3594 m->m_pkthdr.len--; 3595 return; 3596 } 3597 cnt -= m->m_len; 3598 m = m->m_next; 3599 if (m == NULL) 3600 break; 3601 } 3602 panic("tcp_pulloutofband"); 3603 } 3604 3605 /* 3606 * Collect new round-trip time estimate 3607 * and update averages and current timeout. 3608 */ 3609 void 3610 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3611 { 3612 int delta; 3613 3614 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3615 3616 TCPSTAT_INC(tcps_rttupdated); 3617 if (tp->t_rttupdated < UCHAR_MAX) 3618 tp->t_rttupdated++; 3619 #ifdef STATS 3620 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3621 imax(0, rtt * 1000 / hz)); 3622 #endif 3623 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3624 /* 3625 * srtt is stored as fixed point with 5 bits after the 3626 * binary point (i.e., scaled by 8). The following magic 3627 * is equivalent to the smoothing algorithm in rfc793 with 3628 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3629 * point). Adjust rtt to origin 0. 3630 */ 3631 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3632 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3633 3634 if ((tp->t_srtt += delta) <= 0) 3635 tp->t_srtt = 1; 3636 3637 /* 3638 * We accumulate a smoothed rtt variance (actually, a 3639 * smoothed mean difference), then set the retransmit 3640 * timer to smoothed rtt + 4 times the smoothed variance. 3641 * rttvar is stored as fixed point with 4 bits after the 3642 * binary point (scaled by 16). The following is 3643 * equivalent to rfc793 smoothing with an alpha of .75 3644 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3645 * rfc793's wired-in beta. 3646 */ 3647 if (delta < 0) 3648 delta = -delta; 3649 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3650 if ((tp->t_rttvar += delta) <= 0) 3651 tp->t_rttvar = 1; 3652 } else { 3653 /* 3654 * No rtt measurement yet - use the unsmoothed rtt. 3655 * Set the variance to half the rtt (so our first 3656 * retransmit happens at 3*rtt). 3657 */ 3658 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3659 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3660 } 3661 tp->t_rtttime = 0; 3662 tp->t_rxtshift = 0; 3663 3664 /* 3665 * the retransmit should happen at rtt + 4 * rttvar. 3666 * Because of the way we do the smoothing, srtt and rttvar 3667 * will each average +1/2 tick of bias. When we compute 3668 * the retransmit timer, we want 1/2 tick of rounding and 3669 * 1 extra tick because of +-1/2 tick uncertainty in the 3670 * firing of the timer. The bias will give us exactly the 3671 * 1.5 tick we need. But, because the bias is 3672 * statistical, we have to test that we don't drop below 3673 * the minimum feasible timer (which is 2 ticks). 3674 */ 3675 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3676 max(tp->t_rttmin, rtt + 2), tcp_rexmit_max); 3677 3678 /* 3679 * We received an ack for a packet that wasn't retransmitted; 3680 * it is probably safe to discard any error indications we've 3681 * received recently. This isn't quite right, but close enough 3682 * for now (a route might have failed after we sent a segment, 3683 * and the return path might not be symmetrical). 3684 */ 3685 tp->t_softerror = 0; 3686 } 3687 3688 /* 3689 * Determine a reasonable value for maxseg size. 3690 * If the route is known, check route for mtu. 3691 * If none, use an mss that can be handled on the outgoing interface 3692 * without forcing IP to fragment. If no route is found, route has no mtu, 3693 * or the destination isn't local, use a default, hopefully conservative 3694 * size (usually 512 or the default IP max size, but no more than the mtu 3695 * of the interface), as we can't discover anything about intervening 3696 * gateways or networks. We also initialize the congestion/slow start 3697 * window to be a single segment if the destination isn't local. 3698 * While looking at the routing entry, we also initialize other path-dependent 3699 * parameters from pre-set or cached values in the routing entry. 3700 * 3701 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3702 * IP options, e.g. IPSEC data, since length of this data may vary, and 3703 * thus it is calculated for every segment separately in tcp_output(). 3704 * 3705 * NOTE that this routine is only called when we process an incoming 3706 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3707 * settings are handled in tcp_mssopt(). 3708 */ 3709 void 3710 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3711 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3712 { 3713 int mss = 0; 3714 uint32_t maxmtu = 0; 3715 struct inpcb *inp = tptoinpcb(tp); 3716 struct hc_metrics_lite metrics; 3717 #ifdef INET6 3718 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3719 size_t min_protoh = isipv6 ? 3720 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3721 sizeof (struct tcpiphdr); 3722 #else 3723 size_t min_protoh = sizeof(struct tcpiphdr); 3724 #endif 3725 3726 INP_WLOCK_ASSERT(inp); 3727 3728 if (tp->t_port) 3729 min_protoh += V_tcp_udp_tunneling_overhead; 3730 if (mtuoffer != -1) { 3731 KASSERT(offer == -1, ("%s: conflict", __func__)); 3732 offer = mtuoffer - min_protoh; 3733 } 3734 3735 /* Initialize. */ 3736 #ifdef INET6 3737 if (isipv6) { 3738 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3739 tp->t_maxseg = V_tcp_v6mssdflt; 3740 } 3741 #endif 3742 #if defined(INET) && defined(INET6) 3743 else 3744 #endif 3745 #ifdef INET 3746 { 3747 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3748 tp->t_maxseg = V_tcp_mssdflt; 3749 } 3750 #endif 3751 3752 /* 3753 * No route to sender, stay with default mss and return. 3754 */ 3755 if (maxmtu == 0) { 3756 /* 3757 * In case we return early we need to initialize metrics 3758 * to a defined state as tcp_hc_get() would do for us 3759 * if there was no cache hit. 3760 */ 3761 if (metricptr != NULL) 3762 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3763 return; 3764 } 3765 3766 /* What have we got? */ 3767 switch (offer) { 3768 case 0: 3769 /* 3770 * Offer == 0 means that there was no MSS on the SYN 3771 * segment, in this case we use tcp_mssdflt as 3772 * already assigned to t_maxseg above. 3773 */ 3774 offer = tp->t_maxseg; 3775 break; 3776 3777 case -1: 3778 /* 3779 * Offer == -1 means that we didn't receive SYN yet. 3780 */ 3781 /* FALLTHROUGH */ 3782 3783 default: 3784 /* 3785 * Prevent DoS attack with too small MSS. Round up 3786 * to at least minmss. 3787 */ 3788 offer = max(offer, V_tcp_minmss); 3789 } 3790 3791 if (metricptr == NULL) 3792 metricptr = &metrics; 3793 tcp_hc_get(&inp->inp_inc, metricptr); 3794 3795 /* 3796 * If there's a discovered mtu in tcp hostcache, use it. 3797 * Else, use the link mtu. 3798 */ 3799 if (metricptr->hc_mtu) 3800 mss = min(metricptr->hc_mtu, maxmtu) - min_protoh; 3801 else { 3802 #ifdef INET6 3803 if (isipv6) { 3804 mss = maxmtu - min_protoh; 3805 if (!V_path_mtu_discovery && 3806 !in6_localaddr(&inp->in6p_faddr)) 3807 mss = min(mss, V_tcp_v6mssdflt); 3808 } 3809 #endif 3810 #if defined(INET) && defined(INET6) 3811 else 3812 #endif 3813 #ifdef INET 3814 { 3815 mss = maxmtu - min_protoh; 3816 if (!V_path_mtu_discovery && 3817 !in_localaddr(inp->inp_faddr)) 3818 mss = min(mss, V_tcp_mssdflt); 3819 } 3820 #endif 3821 /* 3822 * XXX - The above conditional (mss = maxmtu - min_protoh) 3823 * probably violates the TCP spec. 3824 * The problem is that, since we don't know the 3825 * other end's MSS, we are supposed to use a conservative 3826 * default. But, if we do that, then MTU discovery will 3827 * never actually take place, because the conservative 3828 * default is much less than the MTUs typically seen 3829 * on the Internet today. For the moment, we'll sweep 3830 * this under the carpet. 3831 * 3832 * The conservative default might not actually be a problem 3833 * if the only case this occurs is when sending an initial 3834 * SYN with options and data to a host we've never talked 3835 * to before. Then, they will reply with an MSS value which 3836 * will get recorded and the new parameters should get 3837 * recomputed. For Further Study. 3838 */ 3839 } 3840 mss = min(mss, offer); 3841 3842 /* 3843 * Sanity check: make sure that maxseg will be large 3844 * enough to allow some data on segments even if the 3845 * all the option space is used (40bytes). Otherwise 3846 * funny things may happen in tcp_output. 3847 * 3848 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3849 */ 3850 mss = max(mss, 64); 3851 3852 tp->t_maxseg = mss; 3853 if (tp->t_maxseg < V_tcp_mssdflt) { 3854 /* 3855 * The MSS is so small we should not process incoming 3856 * SACK's since we are subject to attack in such a 3857 * case. 3858 */ 3859 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 3860 } else { 3861 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 3862 } 3863 3864 } 3865 3866 void 3867 tcp_mss(struct tcpcb *tp, int offer) 3868 { 3869 int mss; 3870 uint32_t bufsize; 3871 struct inpcb *inp = tptoinpcb(tp); 3872 struct socket *so; 3873 struct hc_metrics_lite metrics; 3874 struct tcp_ifcap cap; 3875 3876 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3877 3878 bzero(&cap, sizeof(cap)); 3879 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3880 3881 mss = tp->t_maxseg; 3882 3883 /* 3884 * If there's a pipesize, change the socket buffer to that size, 3885 * don't change if sb_hiwat is different than default (then it 3886 * has been changed on purpose with setsockopt). 3887 * Make the socket buffers an integral number of mss units; 3888 * if the mss is larger than the socket buffer, decrease the mss. 3889 */ 3890 so = inp->inp_socket; 3891 SOCK_SENDBUF_LOCK(so); 3892 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.hc_sendpipe) 3893 bufsize = metrics.hc_sendpipe; 3894 else 3895 bufsize = so->so_snd.sb_hiwat; 3896 if (bufsize < mss) 3897 mss = bufsize; 3898 else { 3899 bufsize = roundup(bufsize, mss); 3900 if (bufsize > sb_max) 3901 bufsize = sb_max; 3902 if (bufsize > so->so_snd.sb_hiwat) 3903 (void)sbreserve_locked(so, SO_SND, bufsize, NULL); 3904 } 3905 SOCK_SENDBUF_UNLOCK(so); 3906 /* 3907 * Sanity check: make sure that maxseg will be large 3908 * enough to allow some data on segments even if the 3909 * all the option space is used (40bytes). Otherwise 3910 * funny things may happen in tcp_output. 3911 * 3912 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3913 */ 3914 tp->t_maxseg = max(mss, 64); 3915 if (tp->t_maxseg < V_tcp_mssdflt) { 3916 /* 3917 * The MSS is so small we should not process incoming 3918 * SACK's since we are subject to attack in such a 3919 * case. 3920 */ 3921 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 3922 } else { 3923 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 3924 } 3925 3926 SOCK_RECVBUF_LOCK(so); 3927 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.hc_recvpipe) 3928 bufsize = metrics.hc_recvpipe; 3929 else 3930 bufsize = so->so_rcv.sb_hiwat; 3931 if (bufsize > mss) { 3932 bufsize = roundup(bufsize, mss); 3933 if (bufsize > sb_max) 3934 bufsize = sb_max; 3935 if (bufsize > so->so_rcv.sb_hiwat) 3936 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL); 3937 } 3938 SOCK_RECVBUF_UNLOCK(so); 3939 3940 /* Check the interface for TSO capabilities. */ 3941 if (cap.ifcap & CSUM_TSO) { 3942 tp->t_flags |= TF_TSO; 3943 tp->t_tsomax = cap.tsomax; 3944 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3945 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3946 if (cap.ipsec_tso) 3947 tp->t_flags2 |= TF2_IPSEC_TSO; 3948 } 3949 } 3950 3951 /* 3952 * Determine the MSS option to send on an outgoing SYN. 3953 */ 3954 int 3955 tcp_mssopt(struct in_conninfo *inc) 3956 { 3957 int mss = 0; 3958 uint32_t thcmtu = 0; 3959 uint32_t maxmtu = 0; 3960 size_t min_protoh; 3961 3962 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3963 3964 #ifdef INET6 3965 if (inc->inc_flags & INC_ISIPV6) { 3966 mss = V_tcp_v6mssdflt; 3967 maxmtu = tcp_maxmtu6(inc, NULL); 3968 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3969 } 3970 #endif 3971 #if defined(INET) && defined(INET6) 3972 else 3973 #endif 3974 #ifdef INET 3975 { 3976 mss = V_tcp_mssdflt; 3977 maxmtu = tcp_maxmtu(inc, NULL); 3978 min_protoh = sizeof(struct tcpiphdr); 3979 } 3980 #endif 3981 #if defined(INET6) || defined(INET) 3982 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3983 #endif 3984 3985 if (maxmtu && thcmtu) 3986 mss = min(maxmtu, thcmtu) - min_protoh; 3987 else if (maxmtu || thcmtu) 3988 mss = max(maxmtu, thcmtu) - min_protoh; 3989 3990 return (mss); 3991 } 3992 3993 void 3994 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, 3995 sackstatus_t sack_changed, u_int *maxsegp) 3996 { 3997 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3998 u_int maxseg; 3999 4000 INP_WLOCK_ASSERT(tptoinpcb(tp)); 4001 4002 if (*maxsegp == 0) { 4003 *maxsegp = tcp_maxseg(tp); 4004 } 4005 maxseg = *maxsegp; 4006 /* 4007 * Compute the amount of data that this ACK is indicating 4008 * (del_data) and an estimate of how many bytes are in the 4009 * network. 4010 */ 4011 if (tcp_is_sack_recovery(tp, to) || 4012 (IN_CONGRECOVERY(tp->t_flags) && 4013 !IN_FASTRECOVERY(tp->t_flags))) { 4014 del_data = tp->sackhint.delivered_data; 4015 pipe = tcp_compute_pipe(tp); 4016 } else { 4017 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + 4018 tp->snd_recover - tp->snd_una)) { 4019 del_data = maxseg; 4020 } 4021 pipe = imax(0, tp->snd_max - tp->snd_una - 4022 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); 4023 } 4024 tp->sackhint.prr_delivered += del_data; 4025 /* 4026 * Proportional Rate Reduction 4027 */ 4028 if (pipe >= tp->snd_ssthresh) { 4029 if (tp->sackhint.recover_fs == 0) 4030 tp->sackhint.recover_fs = 4031 imax(1, tp->snd_nxt - tp->snd_una); 4032 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 4033 tp->snd_ssthresh, tp->sackhint.recover_fs) - 4034 tp->sackhint.prr_out + maxseg - 1; 4035 } else { 4036 /* 4037 * PRR 6937bis heuristic: 4038 * - A partial ack without SACK block beneath snd_recover 4039 * indicates further loss. 4040 * - An SACK scoreboard update adding a new hole indicates 4041 * further loss, so be conservative and send at most one 4042 * segment. 4043 * - Prevent ACK splitting attacks, by being conservative 4044 * when no new data is acked. 4045 */ 4046 if ((sack_changed == SACK_NEWLOSS) || (del_data == 0)) { 4047 limit = tp->sackhint.prr_delivered - 4048 tp->sackhint.prr_out; 4049 } else { 4050 limit = imax(tp->sackhint.prr_delivered - 4051 tp->sackhint.prr_out, del_data) + 4052 maxseg; 4053 } 4054 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 4055 } 4056 snd_cnt = imax(snd_cnt, 0) / maxseg; 4057 /* 4058 * Send snd_cnt new data into the network in response to this ack. 4059 * If there is going to be a SACK retransmission, adjust snd_cwnd 4060 * accordingly. 4061 */ 4062 if (IN_FASTRECOVERY(tp->t_flags)) { 4063 if (tcp_is_sack_recovery(tp, to)) { 4064 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4065 } else { 4066 tp->snd_cwnd = (tp->snd_max - tp->snd_una) + 4067 (snd_cnt * maxseg); 4068 } 4069 } else if (IN_CONGRECOVERY(tp->t_flags)) { 4070 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4071 } 4072 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); 4073 } 4074 4075 /* 4076 * On a partial ack arrives, force the retransmission of the 4077 * next unacknowledged segment. Do not clear tp->t_dupacks. 4078 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4079 * be started again. 4080 */ 4081 void 4082 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4083 { 4084 tcp_seq onxt = tp->snd_nxt; 4085 uint32_t ocwnd = tp->snd_cwnd; 4086 u_int maxseg = tcp_maxseg(tp); 4087 4088 INP_WLOCK_ASSERT(tptoinpcb(tp)); 4089 4090 tcp_timer_activate(tp, TT_REXMT, 0); 4091 tp->t_rtttime = 0; 4092 if (IN_FASTRECOVERY(tp->t_flags)) { 4093 tp->snd_nxt = th->th_ack; 4094 /* 4095 * Set snd_cwnd to one segment beyond acknowledged offset. 4096 * (tp->snd_una has not yet been updated when this function is called.) 4097 */ 4098 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4099 tp->t_flags |= TF_ACKNOW; 4100 (void) tcp_output(tp); 4101 tp->snd_cwnd = ocwnd; 4102 if (SEQ_GT(onxt, tp->snd_nxt)) 4103 tp->snd_nxt = onxt; 4104 } 4105 /* 4106 * Partial window deflation. Relies on fact that tp->snd_una 4107 * not updated yet. 4108 */ 4109 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4110 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4111 else 4112 tp->snd_cwnd = 0; 4113 tp->snd_cwnd += maxseg; 4114 } 4115 4116 int 4117 tcp_compute_pipe(struct tcpcb *tp) 4118 { 4119 int pipe; 4120 4121 if (tp->t_fb->tfb_compute_pipe != NULL) { 4122 pipe = (*tp->t_fb->tfb_compute_pipe)(tp); 4123 } else { 4124 pipe = tp->snd_max - tp->snd_una + 4125 tp->sackhint.sack_bytes_rexmit - 4126 tp->sackhint.sacked_bytes - 4127 tp->sackhint.lost_bytes; 4128 } 4129 return (imax(pipe, 0)); 4130 } 4131 4132 uint32_t 4133 tcp_compute_initwnd(uint32_t maxseg) 4134 { 4135 /* 4136 * Calculate the Initial Window, also used as Restart Window 4137 * 4138 * RFC5681 Section 3.1 specifies the default conservative values. 4139 * RFC3390 specifies slightly more aggressive values. 4140 * RFC6928 increases it to ten segments. 4141 * Support for user specified value for initial flight size. 4142 */ 4143 if (V_tcp_initcwnd_segments) 4144 return min(V_tcp_initcwnd_segments * maxseg, 4145 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4146 else if (V_tcp_do_rfc3390) 4147 return min(4 * maxseg, max(2 * maxseg, 4380)); 4148 else { 4149 /* Per RFC5681 Section 3.1 */ 4150 if (maxseg > 2190) 4151 return (2 * maxseg); 4152 else if (maxseg > 1095) 4153 return (3 * maxseg); 4154 else 4155 return (4 * maxseg); 4156 } 4157 } 4158