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