1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 5 * The Regents of the University of California. All rights reserved. 6 * Copyright (c) 2007-2008,2010 7 * Swinburne University of Technology, Melbourne, Australia. 8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 9 * Copyright (c) 2010 The FreeBSD Foundation 10 * Copyright (c) 2010-2011 Juniper Networks, Inc. 11 * All rights reserved. 12 * 13 * Portions of this software were developed at the Centre for Advanced Internet 14 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 15 * James Healy and David Hayes, made possible in part by a grant from the Cisco 16 * University Research Program Fund at Community Foundation Silicon Valley. 17 * 18 * Portions of this software were developed at the Centre for Advanced 19 * Internet Architectures, Swinburne University of Technology, Melbourne, 20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 21 * 22 * Portions of this software were developed by Robert N. M. Watson under 23 * contract to Juniper Networks, Inc. 24 * 25 * Redistribution and use in source and binary forms, with or without 26 * modification, are permitted provided that the following conditions 27 * are met: 28 * 1. Redistributions of source code must retain the above copyright 29 * notice, this list of conditions and the following disclaimer. 30 * 2. Redistributions in binary form must reproduce the above copyright 31 * notice, this list of conditions and the following disclaimer in the 32 * documentation and/or other materials provided with the distribution. 33 * 3. Neither the name of the University nor the names of its contributors 34 * may be used to endorse or promote products derived from this software 35 * without specific prior written permission. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 */ 49 50 #include "opt_inet.h" 51 #include "opt_inet6.h" 52 #include "opt_ipsec.h" 53 #include "opt_rss.h" 54 55 #include <sys/param.h> 56 #include <sys/arb.h> 57 #include <sys/kernel.h> 58 #ifdef TCP_HHOOK 59 #include <sys/hhook.h> 60 #endif 61 #include <sys/malloc.h> 62 #include <sys/mbuf.h> 63 #include <sys/proc.h> /* for proc0 declaration */ 64 #include <sys/protosw.h> 65 #include <sys/qmath.h> 66 #include <sys/sdt.h> 67 #include <sys/signalvar.h> 68 #include <sys/socket.h> 69 #include <sys/socketvar.h> 70 #include <sys/sysctl.h> 71 #include <sys/syslog.h> 72 #include <sys/systm.h> 73 #include <sys/stats.h> 74 75 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 76 77 #include <vm/uma.h> 78 79 #include <net/if.h> 80 #include <net/if_var.h> 81 #include <net/route.h> 82 #include <net/rss_config.h> 83 #include <net/vnet.h> 84 85 #define TCPSTATES /* for logging */ 86 87 #include <netinet/in.h> 88 #include <netinet/in_kdtrace.h> 89 #include <netinet/in_pcb.h> 90 #include <netinet/in_rss.h> 91 #include <netinet/in_systm.h> 92 #include <netinet/ip.h> 93 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 94 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 95 #include <netinet/ip_var.h> 96 #include <netinet/ip_options.h> 97 #include <netinet/ip6.h> 98 #include <netinet/icmp6.h> 99 #include <netinet6/in6_pcb.h> 100 #include <netinet6/in6_rss.h> 101 #include <netinet6/in6_var.h> 102 #include <netinet6/ip6_var.h> 103 #include <netinet6/nd6.h> 104 #include <netinet/tcp.h> 105 #include <netinet/tcp_fsm.h> 106 #include <netinet/tcp_seq.h> 107 #include <netinet/tcp_timer.h> 108 #include <netinet/tcp_var.h> 109 #include <netinet/tcp_log_buf.h> 110 #include <netinet6/tcp6_var.h> 111 #include <netinet/tcpip.h> 112 #include <netinet/cc/cc.h> 113 #include <netinet/tcp_fastopen.h> 114 #include <netinet/tcp_syncache.h> 115 #ifdef TCP_OFFLOAD 116 #include <netinet/tcp_offload.h> 117 #endif 118 #include <netinet/tcp_ecn.h> 119 #include <netinet/udp.h> 120 121 #include <netipsec/ipsec_support.h> 122 123 #include <machine/in_cksum.h> 124 125 #include <security/mac/mac_framework.h> 126 127 const int tcprexmtthresh = 3; 128 129 VNET_DEFINE(int, tcp_log_in_vain) = 0; 130 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 131 &VNET_NAME(tcp_log_in_vain), 0, 132 "Log all incoming TCP segments to closed ports"); 133 134 VNET_DEFINE(int, tcp_bind_all_fibs) = 1; 135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, bind_all_fibs, CTLFLAG_VNET | CTLFLAG_RDTUN, 136 &VNET_NAME(tcp_bind_all_fibs), 0, 137 "Bound sockets receive traffic from all FIBs"); 138 139 VNET_DEFINE(int, blackhole) = 0; 140 #define V_blackhole VNET(blackhole) 141 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 142 &VNET_NAME(blackhole), 0, 143 "Do not send RST on segments to closed ports"); 144 145 VNET_DEFINE(bool, blackhole_local) = false; 146 #define V_blackhole_local VNET(blackhole_local) 147 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET | 148 CTLFLAG_RW, &VNET_NAME(blackhole_local), false, 149 "Enforce net.inet.tcp.blackhole for locally originated packets"); 150 151 VNET_DEFINE(int, tcp_delack_enabled) = 1; 152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 153 &VNET_NAME(tcp_delack_enabled), 0, 154 "Delay ACK to try and piggyback it onto a data packet"); 155 156 VNET_DEFINE(int, drop_synfin) = 0; 157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 158 &VNET_NAME(drop_synfin), 0, 159 "Drop TCP packets with SYN+FIN set"); 160 161 VNET_DEFINE(int, tcp_do_prr) = 1; 162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 163 &VNET_NAME(tcp_do_prr), 1, 164 "Enable Proportional Rate Reduction per RFC 6937"); 165 166 VNET_DEFINE(int, tcp_do_newcwv) = 0; 167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 168 &VNET_NAME(tcp_do_newcwv), 0, 169 "Enable New Congestion Window Validation per RFC7661"); 170 171 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 173 &VNET_NAME(tcp_do_rfc3042), 0, 174 "Enable RFC 3042 (Limited Transmit)"); 175 176 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 178 &VNET_NAME(tcp_do_rfc3390), 0, 179 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 180 181 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 183 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 184 "Slow-start flight size (initial congestion window) in number of segments"); 185 186 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 188 &VNET_NAME(tcp_do_rfc3465), 0, 189 "Enable RFC 3465 (Appropriate Byte Counting)"); 190 191 VNET_DEFINE(int, tcp_abc_l_var) = 2; 192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 193 &VNET_NAME(tcp_abc_l_var), 2, 194 "Cap the max cwnd increment during slow-start to this number of segments"); 195 196 VNET_DEFINE(int, tcp_insecure_syn) = 0; 197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 198 &VNET_NAME(tcp_insecure_syn), 0, 199 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 200 201 VNET_DEFINE(int, tcp_insecure_rst) = 0; 202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 203 &VNET_NAME(tcp_insecure_rst), 0, 204 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 205 206 VNET_DEFINE(int, tcp_insecure_ack) = 0; 207 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_ack, CTLFLAG_VNET | CTLFLAG_RW, 208 &VNET_NAME(tcp_insecure_ack), 0, 209 "Follow RFC793 criteria for validating SEG.ACK"); 210 211 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 212 #define V_tcp_recvspace VNET(tcp_recvspace) 213 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 214 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 215 216 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 218 &VNET_NAME(tcp_do_autorcvbuf), 0, 219 "Enable automatic receive buffer sizing"); 220 221 VNET_DEFINE(int, tcp_autorcvbuf_max) = 8*1024*1024; 222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 223 &VNET_NAME(tcp_autorcvbuf_max), 0, 224 "Max size of automatic receive buffer"); 225 226 VNET_DEFINE(struct inpcbinfo, tcbinfo); 227 228 /* 229 * TCP statistics are stored in an array of counter(9)s, which size matches 230 * size of struct tcpstat. TCP running connection count is a regular array. 231 */ 232 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 233 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 234 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 235 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 236 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 237 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 238 "TCP connection counts by TCP state"); 239 240 /* 241 * Kernel module interface for updating tcpstat. The first argument is an index 242 * into tcpstat treated as an array. 243 */ 244 void 245 kmod_tcpstat_add(int statnum, int val) 246 { 247 248 counter_u64_add(VNET(tcpstat)[statnum], val); 249 } 250 251 /* 252 * Make sure that we only start a SACK loss recovery when 253 * receiving a duplicate ACK with a SACK block, and also 254 * complete SACK loss recovery in case the other end 255 * reneges. 256 */ 257 static bool inline 258 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to) 259 { 260 return ((tp->t_flags & TF_SACK_PERMIT) && 261 ((to->to_flags & TOF_SACK) || 262 (!TAILQ_EMPTY(&tp->snd_holes)))); 263 } 264 265 #ifdef TCP_HHOOK 266 /* 267 * Wrapper for the TCP established input helper hook. 268 */ 269 void 270 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 271 { 272 struct tcp_hhook_data hhook_data; 273 274 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 275 hhook_data.tp = tp; 276 hhook_data.th = th; 277 hhook_data.to = to; 278 279 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 280 &tp->t_osd); 281 } 282 } 283 #endif 284 285 /* 286 * CC wrapper hook functions 287 */ 288 void 289 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 290 uint16_t type) 291 { 292 #ifdef STATS 293 int32_t gput; 294 #endif 295 296 INP_WLOCK_ASSERT(tptoinpcb(tp)); 297 298 tp->t_ccv.nsegs = nsegs; 299 tp->t_ccv.bytes_this_ack = BYTES_THIS_ACK(tp, th); 300 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 301 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 302 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 303 tp->t_ccv.flags |= CCF_CWND_LIMITED; 304 else 305 tp->t_ccv.flags &= ~CCF_CWND_LIMITED; 306 307 if (type == CC_ACK) { 308 #ifdef STATS 309 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 310 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 311 if (!IN_RECOVERY(tp->t_flags)) 312 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 313 tp->t_ccv.bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 314 if ((tp->t_flags & TF_GPUTINPROG) && 315 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 316 /* 317 * Compute goodput in bits per millisecond. 318 */ 319 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) / 320 max(1, tcp_ts_getticks() - tp->gput_ts); 321 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 322 gput); 323 /* 324 * XXXLAS: This is a temporary hack, and should be 325 * chained off VOI_TCP_GPUT when stats(9) grows an API 326 * to deal with chained VOIs. 327 */ 328 if (tp->t_stats_gput_prev > 0) 329 stats_voi_update_abs_s32(tp->t_stats, 330 VOI_TCP_GPUT_ND, 331 ((gput - tp->t_stats_gput_prev) * 100) / 332 tp->t_stats_gput_prev); 333 tp->t_flags &= ~TF_GPUTINPROG; 334 tp->t_stats_gput_prev = gput; 335 } 336 #endif /* STATS */ 337 if (tp->snd_cwnd > tp->snd_ssthresh) { 338 tp->t_bytes_acked += tp->t_ccv.bytes_this_ack; 339 if (tp->t_bytes_acked >= tp->snd_cwnd) { 340 tp->t_bytes_acked -= tp->snd_cwnd; 341 tp->t_ccv.flags |= CCF_ABC_SENTAWND; 342 } 343 } else { 344 tp->t_ccv.flags &= ~CCF_ABC_SENTAWND; 345 tp->t_bytes_acked = 0; 346 } 347 } 348 349 if (CC_ALGO(tp)->ack_received != NULL) { 350 /* XXXLAS: Find a way to live without this */ 351 tp->t_ccv.curack = th->th_ack; 352 CC_ALGO(tp)->ack_received(&tp->t_ccv, type); 353 } 354 #ifdef STATS 355 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 356 #endif 357 } 358 359 void 360 cc_conn_init(struct tcpcb *tp) 361 { 362 struct tcp_hc_metrics metrics; 363 struct inpcb *inp = tptoinpcb(tp); 364 u_int maxseg; 365 int rtt; 366 367 INP_WLOCK_ASSERT(inp); 368 369 tcp_hc_get(&inp->inp_inc, &metrics); 370 maxseg = tcp_maxseg(tp); 371 372 if (tp->t_srtt == 0 && (rtt = metrics.hc_rtt)) { 373 tp->t_srtt = rtt; 374 TCPSTAT_INC(tcps_usedrtt); 375 if (metrics.hc_rttvar) { 376 tp->t_rttvar = metrics.hc_rttvar; 377 TCPSTAT_INC(tcps_usedrttvar); 378 } else { 379 /* default variation is +- 1 rtt */ 380 tp->t_rttvar = 381 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 382 } 383 TCPT_RANGESET(tp->t_rxtcur, 384 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 385 tp->t_rttmin, tcp_rexmit_max); 386 } 387 if (metrics.hc_ssthresh) { 388 /* 389 * There's some sort of gateway or interface 390 * buffer limit on the path. Use this to set 391 * the slow start threshold, but set the 392 * threshold to no less than 2*mss. 393 */ 394 tp->snd_ssthresh = max(2 * maxseg, metrics.hc_ssthresh); 395 TCPSTAT_INC(tcps_usedssthresh); 396 } 397 398 /* 399 * Set the initial slow-start flight size. 400 * 401 * If a SYN or SYN/ACK was lost and retransmitted, we have to 402 * reduce the initial CWND to one segment as congestion is likely 403 * requiring us to be cautious. 404 */ 405 if (tp->snd_cwnd == 1) 406 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 407 else 408 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 409 410 if (CC_ALGO(tp)->conn_init != NULL) 411 CC_ALGO(tp)->conn_init(&tp->t_ccv); 412 } 413 414 void inline 415 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 416 { 417 INP_WLOCK_ASSERT(tptoinpcb(tp)); 418 419 #ifdef STATS 420 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 421 #endif 422 423 switch(type) { 424 case CC_NDUPACK: 425 if (!IN_FASTRECOVERY(tp->t_flags)) { 426 tp->snd_recover = tp->snd_max; 427 if (tp->t_flags2 & TF2_ECN_PERMIT) 428 tp->t_flags2 |= TF2_ECN_SND_CWR; 429 } 430 break; 431 case CC_ECN: 432 if (!IN_CONGRECOVERY(tp->t_flags) || 433 /* 434 * Allow ECN reaction on ACK to CWR, if 435 * that data segment was also CE marked. 436 */ 437 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 438 EXIT_CONGRECOVERY(tp->t_flags); 439 TCPSTAT_INC(tcps_ecn_rcwnd); 440 tp->snd_recover = tp->snd_max + 1; 441 if (tp->t_flags2 & TF2_ECN_PERMIT) 442 tp->t_flags2 |= TF2_ECN_SND_CWR; 443 } 444 break; 445 case CC_RTO: 446 tp->t_dupacks = 0; 447 tp->t_bytes_acked = 0; 448 EXIT_RECOVERY(tp->t_flags); 449 if (tp->t_flags2 & TF2_ECN_PERMIT) 450 tp->t_flags2 |= TF2_ECN_SND_CWR; 451 break; 452 case CC_RTO_ERR: 453 TCPSTAT_INC(tcps_sndrexmitbad); 454 /* RTO was unnecessary, so reset everything. */ 455 tp->snd_cwnd = tp->snd_cwnd_prev; 456 tp->snd_ssthresh = tp->snd_ssthresh_prev; 457 tp->snd_recover = tp->snd_recover_prev; 458 if (tp->t_flags & TF_WASFRECOVERY) 459 ENTER_FASTRECOVERY(tp->t_flags); 460 if (tp->t_flags & TF_WASCRECOVERY) 461 ENTER_CONGRECOVERY(tp->t_flags); 462 tp->snd_nxt = tp->snd_max; 463 tp->t_flags &= ~TF_PREVVALID; 464 tp->t_rxtshift = 0; 465 tp->t_badrxtwin = 0; 466 break; 467 } 468 if (SEQ_LT(tp->snd_fack, tp->snd_una) || 469 SEQ_GT(tp->snd_fack, tp->snd_max)) { 470 tp->snd_fack = tp->snd_una; 471 } 472 473 if (CC_ALGO(tp)->cong_signal != NULL) { 474 if (th != NULL) 475 tp->t_ccv.curack = th->th_ack; 476 CC_ALGO(tp)->cong_signal(&tp->t_ccv, type); 477 } 478 } 479 480 void inline 481 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 482 { 483 INP_WLOCK_ASSERT(tptoinpcb(tp)); 484 485 if (CC_ALGO(tp)->post_recovery != NULL) { 486 if (SEQ_LT(tp->snd_fack, th->th_ack) || 487 SEQ_GT(tp->snd_fack, tp->snd_max)) { 488 tp->snd_fack = th->th_ack; 489 } 490 tp->t_ccv.curack = th->th_ack; 491 CC_ALGO(tp)->post_recovery(&tp->t_ccv); 492 } 493 EXIT_RECOVERY(tp->t_flags); 494 495 tp->t_bytes_acked = 0; 496 tp->sackhint.delivered_data = 0; 497 tp->sackhint.prr_delivered = 0; 498 tp->sackhint.prr_out = 0; 499 tp->snd_cwnd = tp->snd_ssthresh; 500 } 501 502 /* 503 * Indicate whether this ack should be delayed. We can delay the ack if 504 * following conditions are met: 505 * - There is no delayed ack timer in progress. 506 * - Our last ack wasn't a 0-sized window. We never want to delay 507 * the ack that opens up a 0-sized window. 508 * - LRO wasn't used for this segment. We make sure by checking that the 509 * segment size is not larger than the MSS. 510 */ 511 #define DELAY_ACK(tp, tlen) \ 512 ((!tcp_timer_active(tp, TT_DELACK) && \ 513 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 514 (tlen <= tp->t_maxseg) && \ 515 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 516 517 void inline 518 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) 519 { 520 INP_WLOCK_ASSERT(tptoinpcb(tp)); 521 522 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 523 switch (iptos & IPTOS_ECN_MASK) { 524 case IPTOS_ECN_CE: 525 tp->t_ccv.flags |= CCF_IPHDR_CE; 526 break; 527 case IPTOS_ECN_ECT0: 528 /* FALLTHROUGH */ 529 case IPTOS_ECN_ECT1: 530 /* FALLTHROUGH */ 531 case IPTOS_ECN_NOTECT: 532 tp->t_ccv.flags &= ~CCF_IPHDR_CE; 533 break; 534 } 535 536 if (flags & TH_CWR) 537 tp->t_ccv.flags |= CCF_TCPHDR_CWR; 538 else 539 tp->t_ccv.flags &= ~CCF_TCPHDR_CWR; 540 541 CC_ALGO(tp)->ecnpkt_handler(&tp->t_ccv); 542 543 if (((tp->t_state == TCPS_ESTABLISHED) || 544 (tp->t_state == TCPS_FIN_WAIT_1) || 545 (tp->t_state == TCPS_FIN_WAIT_2)) && 546 (tp->t_ccv.flags & CCF_ACKNOW)) { 547 tp->t_flags |= TF_ACKNOW; 548 tp->t_ccv.flags &= ~CCF_ACKNOW; 549 } 550 } 551 } 552 553 void inline 554 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 555 { 556 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos); 557 } 558 559 /* 560 * TCP input handling is split into multiple parts: 561 * tcp6_input is a thin wrapper around tcp_input for the extended 562 * ip6_protox[] call format in ip6_input 563 * tcp_input handles primary segment validation, inpcb lookup and 564 * SYN processing on listen sockets 565 * tcp_do_segment processes the ACK and text of the segment for 566 * establishing, established and closing connections 567 */ 568 #ifdef INET6 569 int 570 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 571 { 572 struct mbuf *m; 573 574 m = *mp; 575 if (m->m_len < *offp + sizeof(struct tcphdr)) { 576 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 577 if (m == NULL) { 578 *mp = m; 579 TCPSTAT_INC(tcps_rcvshort); 580 return (IPPROTO_DONE); 581 } 582 } 583 584 *mp = m; 585 return (tcp_input_with_port(mp, offp, proto, port)); 586 } 587 588 int 589 tcp6_input(struct mbuf **mp, int *offp, int proto) 590 { 591 592 return(tcp6_input_with_port(mp, offp, proto, 0)); 593 } 594 #endif /* INET6 */ 595 596 int 597 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 598 { 599 struct mbuf *m; 600 struct tcphdr *th = NULL; 601 struct ip *ip = NULL; 602 struct inpcb *inp = NULL; 603 struct tcpcb *tp = NULL; 604 struct socket *so = NULL; 605 u_char *optp = NULL; 606 int off0; 607 int optlen = 0; 608 #ifdef INET 609 int len; 610 uint8_t ipttl; 611 #endif 612 int tlen = 0, off; 613 int drop_hdrlen; 614 int thflags; 615 int lookupflag; 616 uint8_t iptos; 617 struct m_tag *fwd_tag = NULL; 618 #ifdef INET6 619 struct ip6_hdr *ip6 = NULL; 620 bool isipv6; 621 #else 622 const void *ip6 = NULL; 623 #endif /* INET6 */ 624 struct tcpopt to; /* options in this segment */ 625 char *s = NULL; /* address and port logging */ 626 bool closed_port = false; /* segment is hitting a closed port */ 627 628 NET_EPOCH_ASSERT(); 629 630 off0 = *offp; 631 m = *mp; 632 *mp = NULL; 633 to.to_flags = 0; 634 TCPSTAT_INC(tcps_rcvtotal); 635 636 m->m_pkthdr.tcp_tun_port = port; 637 #ifdef INET6 638 isipv6 = mtod(m, struct ip *)->ip_v == 6; 639 if (isipv6) { 640 ip6 = mtod(m, struct ip6_hdr *); 641 th = (struct tcphdr *)((caddr_t)ip6 + off0); 642 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 643 if (port) 644 goto skip6_csum; 645 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 646 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 647 th->th_sum = m->m_pkthdr.csum_data; 648 else 649 th->th_sum = in6_cksum_pseudo(ip6, tlen, 650 IPPROTO_TCP, m->m_pkthdr.csum_data); 651 th->th_sum ^= 0xffff; 652 } else if (m->m_pkthdr.csum_flags & CSUM_IP6_TCP) { 653 /* 654 * Packet from local host (maybe from a VM). 655 * Checksum not required. 656 */ 657 th->th_sum = 0; 658 } else 659 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 660 if (th->th_sum) { 661 TCPSTAT_INC(tcps_rcvbadsum); 662 goto drop; 663 } 664 skip6_csum: 665 /* 666 * Be proactive about unspecified IPv6 address in source. 667 * As we use all-zero to indicate unbounded/unconnected pcb, 668 * unspecified IPv6 address can be used to confuse us. 669 * 670 * Note that packets with unspecified IPv6 destination is 671 * already dropped in ip6_input. 672 */ 673 KASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_dst), 674 ("%s: unspecified destination v6 address", __func__)); 675 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 676 IP6STAT_INC(ip6s_badscope); /* XXX */ 677 goto drop; 678 } 679 iptos = IPV6_TRAFFIC_CLASS(ip6); 680 } 681 #endif 682 #if defined(INET) && defined(INET6) 683 else 684 #endif 685 #ifdef INET 686 { 687 /* 688 * Get IP and TCP header together in first mbuf. 689 * Note: IP leaves IP header in first mbuf. 690 */ 691 if (off0 > sizeof (struct ip)) { 692 ip_stripoptions(m); 693 off0 = sizeof(struct ip); 694 } 695 if (m->m_len < sizeof (struct tcpiphdr)) { 696 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 697 == NULL) { 698 TCPSTAT_INC(tcps_rcvshort); 699 return (IPPROTO_DONE); 700 } 701 } 702 ip = mtod(m, struct ip *); 703 th = (struct tcphdr *)((caddr_t)ip + off0); 704 tlen = ntohs(ip->ip_len) - off0; 705 706 iptos = ip->ip_tos; 707 if (port) 708 goto skip_csum; 709 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 710 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 711 th->th_sum = m->m_pkthdr.csum_data; 712 else 713 th->th_sum = in_pseudo(ip->ip_src.s_addr, 714 ip->ip_dst.s_addr, 715 htonl(m->m_pkthdr.csum_data + tlen + 716 IPPROTO_TCP)); 717 th->th_sum ^= 0xffff; 718 } else if (m->m_pkthdr.csum_flags & CSUM_IP_TCP) { 719 /* 720 * Packet from local host (maybe from a VM). 721 * Checksum not required. 722 */ 723 th->th_sum = 0; 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 ip6 = mtod(m, struct ip6_hdr *); 778 th = (struct tcphdr *)((caddr_t)ip6 + off0); 779 } 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 (V_tcp_bind_all_fibs ? 0 : INPLOOKUP_FIB); 838 findpcb: 839 tp = NULL; 840 #ifdef INET6 841 if (isipv6 && fwd_tag != NULL) { 842 struct sockaddr_in6 *next_hop6; 843 844 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 845 /* 846 * Transparently forwarded. Pretend to be the destination. 847 * Already got one like this? 848 */ 849 inp = in6_pcblookup_mbuf(&V_tcbinfo, 850 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 851 lookupflag & ~INPLOOKUP_WILDCARD, m->m_pkthdr.rcvif, m); 852 if (!inp) { 853 /* 854 * It's new. Try to find the ambushing socket. 855 * Because we've rewritten the destination address, 856 * any hardware-generated hash is ignored. 857 */ 858 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 859 th->th_sport, &next_hop6->sin6_addr, 860 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 861 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 862 } 863 } else if (isipv6) { 864 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 865 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag, 866 m->m_pkthdr.rcvif, m); 867 } 868 #endif /* INET6 */ 869 #if defined(INET6) && defined(INET) 870 else 871 #endif 872 #ifdef INET 873 if (fwd_tag != NULL) { 874 struct sockaddr_in *next_hop; 875 876 next_hop = (struct sockaddr_in *)(fwd_tag+1); 877 /* 878 * Transparently forwarded. Pretend to be the destination. 879 * already got one like this? 880 */ 881 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 882 ip->ip_dst, th->th_dport, lookupflag & ~INPLOOKUP_WILDCARD, 883 m->m_pkthdr.rcvif, m); 884 if (!inp) { 885 /* 886 * It's new. Try to find the ambushing socket. 887 * Because we've rewritten the destination address, 888 * any hardware-generated hash is ignored. 889 */ 890 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 891 th->th_sport, next_hop->sin_addr, 892 next_hop->sin_port ? ntohs(next_hop->sin_port) : 893 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 894 } 895 } else 896 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 897 th->th_sport, ip->ip_dst, th->th_dport, lookupflag, 898 m->m_pkthdr.rcvif, m); 899 #endif /* INET */ 900 901 /* 902 * If the INPCB does not exist then all data in the incoming 903 * segment is discarded and an appropriate RST is sent back. 904 * XXX MRT Send RST using which routing table? 905 */ 906 if (inp == NULL) { 907 if ((lookupflag & INPLOOKUP_WILDCARD) == 0) { 908 /* We came here after second (safety) lookup. */ 909 MPASS(!closed_port); 910 } else { 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 (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 closed_port = true; 922 } 923 goto dropwithreset; 924 } 925 INP_LOCK_ASSERT(inp); 926 927 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 928 #ifdef INET6 929 if (isipv6 && IPSEC_ENABLED(ipv6) && 930 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 931 goto dropunlock; 932 } 933 #ifdef INET 934 else 935 #endif 936 #endif /* INET6 */ 937 #ifdef INET 938 if (IPSEC_ENABLED(ipv4) && 939 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 940 goto dropunlock; 941 } 942 #endif /* INET */ 943 #endif /* IPSEC */ 944 945 /* 946 * Check the minimum TTL for socket. 947 */ 948 if (inp->inp_ip_minttl != 0) { 949 #ifdef INET6 950 if (isipv6) { 951 if (inp->inp_ip_minttl > ip6->ip6_hlim) 952 goto dropunlock; 953 } else 954 #endif 955 if (inp->inp_ip_minttl > ip->ip_ttl) 956 goto dropunlock; 957 } 958 959 tp = intotcpcb(inp); 960 switch (tp->t_state) { 961 case TCPS_TIME_WAIT: 962 /* 963 * A previous connection in TIMEWAIT state is supposed to catch 964 * stray or duplicate segments arriving late. If this segment 965 * was a legitimate new connection attempt, the old INPCB gets 966 * removed and we can try again to find a listening socket. 967 */ 968 tcp_dooptions(&to, optp, optlen, 969 (thflags & TH_SYN) ? TO_SYN : 0); 970 /* 971 * tcp_twcheck unlocks the inp always, and frees the m if fails. 972 */ 973 if (tcp_twcheck(inp, &to, th, m, tlen)) 974 goto findpcb; 975 return (IPPROTO_DONE); 976 case TCPS_CLOSED: 977 /* 978 * The TCPCB may no longer exist if the connection is winding 979 * down or it is in the CLOSED state. Either way we drop the 980 * segment and send an appropriate response. 981 */ 982 closed_port = true; 983 goto dropwithreset; 984 } 985 986 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 987 closed_port = true; 988 goto dropwithreset; 989 } 990 991 #ifdef TCP_OFFLOAD 992 if (tp->t_flags & TF_TOE) { 993 tcp_offload_input(tp, m); 994 m = NULL; /* consumed by the TOE driver */ 995 goto dropunlock; 996 } 997 #endif 998 999 #ifdef MAC 1000 if (mac_inpcb_check_deliver(inp, m)) 1001 goto dropunlock; 1002 #endif 1003 so = inp->inp_socket; 1004 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1005 /* 1006 * When the socket is accepting connections (the INPCB is in LISTEN 1007 * state) we look into the SYN cache if this is a new connection 1008 * attempt or the completion of a previous one. 1009 */ 1010 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), 1011 ("%s: so accepting but tp %p not listening", __func__, tp)); 1012 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { 1013 struct in_conninfo inc; 1014 1015 bzero(&inc, sizeof(inc)); 1016 #ifdef INET6 1017 if (isipv6) { 1018 inc.inc_flags |= INC_ISIPV6; 1019 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1020 inc.inc_flags |= INC_IPV6MINMTU; 1021 inc.inc6_faddr = ip6->ip6_src; 1022 inc.inc6_laddr = ip6->ip6_dst; 1023 } else 1024 #endif 1025 { 1026 inc.inc_faddr = ip->ip_src; 1027 inc.inc_laddr = ip->ip_dst; 1028 } 1029 inc.inc_fport = th->th_sport; 1030 inc.inc_lport = th->th_dport; 1031 inc.inc_fibnum = so->so_fibnum; 1032 1033 /* 1034 * Check for an existing connection attempt in syncache if 1035 * the flag is only ACK. A successful lookup creates a new 1036 * socket appended to the listen queue in SYN_RECEIVED state. 1037 */ 1038 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1039 int result; 1040 1041 /* 1042 * Parse the TCP options here because 1043 * syncookies need access to the reflected 1044 * timestamp. 1045 */ 1046 tcp_dooptions(&to, optp, optlen, 0); 1047 /* 1048 * NB: syncache_expand() doesn't unlock inp. 1049 */ 1050 result = syncache_expand(&inc, &to, th, &so, m, port); 1051 if (result < 0) { 1052 /* 1053 * A failing TCP MD5 signature comparison 1054 * must result in the segment being dropped 1055 * and must not produce any response back 1056 * to the sender. 1057 */ 1058 goto dropunlock; 1059 } else if (result == 0) { 1060 /* 1061 * No syncache entry, or ACK was not for our 1062 * SYN/ACK. Do our protection against double 1063 * ACK. If peer sent us 2 ACKs, then for the 1064 * first one syncache_expand() successfully 1065 * converted syncache entry into a socket, 1066 * while we were waiting on the inpcb lock. We 1067 * don't want to sent RST for the second ACK, 1068 * so we perform second lookup without wildcard 1069 * match, hoping to find the new socket. If 1070 * the ACK is stray indeed, the missing 1071 * INPLOOKUP_WILDCARD flag in lookupflag would 1072 * hint the above code that the lookup was a 1073 * second attempt. 1074 * 1075 * NB: syncache did its own logging 1076 * of the failure cause. 1077 */ 1078 INP_WUNLOCK(inp); 1079 lookupflag &= ~INPLOOKUP_WILDCARD; 1080 goto findpcb; 1081 } 1082 tfo_socket_result: 1083 if (so == NULL) { 1084 /* 1085 * We completed the 3-way handshake 1086 * but could not allocate a socket 1087 * either due to memory shortage, 1088 * listen queue length limits or 1089 * global socket limits. Send RST 1090 * or wait and have the remote end 1091 * retransmit the ACK for another 1092 * try. 1093 */ 1094 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1095 log(LOG_DEBUG, "%s; %s: Listen socket: " 1096 "Socket allocation failed due to " 1097 "limits or memory shortage, %s\n", 1098 s, __func__, 1099 V_tcp_sc_rst_sock_fail ? 1100 "sending RST" : "try again"); 1101 if (V_tcp_sc_rst_sock_fail) { 1102 goto dropwithreset; 1103 } else 1104 goto dropunlock; 1105 } 1106 /* 1107 * Socket is created in state SYN_RECEIVED. 1108 * Unlock the listen socket, lock the newly 1109 * created socket and update the tp variable. 1110 * If we came here via jump to tfo_socket_result, 1111 * then listening socket is read-locked. 1112 */ 1113 INP_UNLOCK(inp); /* listen socket */ 1114 inp = sotoinpcb(so); 1115 /* 1116 * New connection inpcb is already locked by 1117 * syncache_expand(). 1118 */ 1119 INP_WLOCK_ASSERT(inp); 1120 tp = intotcpcb(inp); 1121 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1122 ("%s: ", __func__)); 1123 /* 1124 * Process the segment and the data it 1125 * contains. tcp_do_segment() consumes 1126 * the mbuf chain and unlocks the inpcb. 1127 */ 1128 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1129 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, 1130 tlen, iptos); 1131 return (IPPROTO_DONE); 1132 } 1133 /* 1134 * Segment flag validation for new connection attempts: 1135 * 1136 * Our (SYN|ACK) response was rejected. 1137 * Check with syncache and remove entry to prevent 1138 * retransmits. 1139 * 1140 * NB: syncache_chkrst does its own logging of failure 1141 * causes. 1142 */ 1143 if (thflags & TH_RST) { 1144 syncache_chkrst(&inc, th, port); 1145 goto dropunlock; 1146 } 1147 /* 1148 * We can't do anything without SYN. 1149 */ 1150 if ((thflags & TH_SYN) == 0) { 1151 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1152 log(LOG_DEBUG, "%s; %s: Listen socket: " 1153 "SYN is missing, segment ignored\n", 1154 s, __func__); 1155 TCPSTAT_INC(tcps_badsyn); 1156 goto dropunlock; 1157 } 1158 /* 1159 * (SYN|ACK) is bogus on a listen socket. 1160 */ 1161 if (thflags & TH_ACK) { 1162 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1163 log(LOG_DEBUG, "%s; %s: Listen socket: " 1164 "SYN|ACK invalid, segment ignored\n", 1165 s, __func__); 1166 TCPSTAT_INC(tcps_badsyn); 1167 goto dropunlock; 1168 } 1169 /* 1170 * If the drop_synfin option is enabled, drop all 1171 * segments with both the SYN and FIN bits set. 1172 * This prevents e.g. nmap from identifying the 1173 * TCP/IP stack. 1174 * XXX: Poor reasoning. nmap has other methods 1175 * and is constantly refining its stack detection 1176 * strategies. 1177 * XXX: This is a violation of the TCP specification 1178 * and was used by RFC1644. 1179 */ 1180 if ((thflags & TH_FIN) && V_drop_synfin) { 1181 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1182 log(LOG_DEBUG, "%s; %s: Listen socket: " 1183 "SYN|FIN segment ignored (based on " 1184 "sysctl setting)\n", s, __func__); 1185 TCPSTAT_INC(tcps_badsyn); 1186 goto dropunlock; 1187 } 1188 /* 1189 * Segment's flags are (SYN) or (SYN|FIN). 1190 * 1191 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1192 * as they do not affect the state of the TCP FSM. 1193 * The data pointed to by TH_URG and th_urp is ignored. 1194 */ 1195 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1196 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1197 KASSERT(thflags & (TH_SYN), 1198 ("%s: Listen socket: TH_SYN not set", __func__)); 1199 INP_RLOCK_ASSERT(inp); 1200 #ifdef INET6 1201 /* 1202 * If deprecated address is forbidden, 1203 * we do not accept SYN to deprecated interface 1204 * address to prevent any new inbound connection from 1205 * getting established. 1206 * When we do not accept SYN, we send a TCP RST, 1207 * with deprecated source address (instead of dropping 1208 * it). We compromise it as it is much better for peer 1209 * to send a RST, and RST will be the final packet 1210 * for the exchange. 1211 * 1212 * If we do not forbid deprecated addresses, we accept 1213 * the SYN packet. RFC2462 does not suggest dropping 1214 * SYN in this case. 1215 * If we decipher RFC2462 5.5.4, it says like this: 1216 * 1. use of deprecated addr with existing 1217 * communication is okay - "SHOULD continue to be 1218 * used" 1219 * 2. use of it with new communication: 1220 * (2a) "SHOULD NOT be used if alternate address 1221 * with sufficient scope is available" 1222 * (2b) nothing mentioned otherwise. 1223 * Here we fall into (2b) case as we have no choice in 1224 * our source address selection - we must obey the peer. 1225 * 1226 * The wording in RFC2462 is confusing, and there are 1227 * multiple description text for deprecated address 1228 * handling - worse, they are not exactly the same. 1229 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1230 */ 1231 if (isipv6 && !V_ip6_use_deprecated) { 1232 struct in6_ifaddr *ia6; 1233 1234 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1235 if (ia6 != NULL && 1236 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1237 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1238 log(LOG_DEBUG, "%s; %s: Listen socket: " 1239 "Connection attempt to deprecated " 1240 "IPv6 address rejected\n", 1241 s, __func__); 1242 goto dropwithreset; 1243 } 1244 } 1245 #endif /* INET6 */ 1246 /* 1247 * Basic sanity checks on incoming SYN requests: 1248 * Don't respond if the destination is a link layer 1249 * broadcast according to RFC1122 4.2.3.10, p. 104. 1250 * If it is from this socket it must be forged. 1251 * Don't respond if the source or destination is a 1252 * global or subnet broad- or multicast address. 1253 * Note that it is quite possible to receive unicast 1254 * link-layer packets with a broadcast IP address. Use 1255 * in_ifnet_broadcast() to find them. 1256 */ 1257 if (m->m_flags & (M_BCAST|M_MCAST)) { 1258 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1259 log(LOG_DEBUG, "%s; %s: Listen socket: " 1260 "Connection attempt from broad- or multicast " 1261 "link layer address ignored\n", s, __func__); 1262 goto dropunlock; 1263 } 1264 #ifdef INET6 1265 if (isipv6) { 1266 if (th->th_dport == th->th_sport && 1267 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1268 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1269 log(LOG_DEBUG, "%s; %s: Listen socket: " 1270 "Connection attempt to/from self " 1271 "ignored\n", s, __func__); 1272 goto dropunlock; 1273 } 1274 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1275 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1276 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1277 log(LOG_DEBUG, "%s; %s: Listen socket: " 1278 "Connection attempt from/to multicast " 1279 "address ignored\n", s, __func__); 1280 goto dropunlock; 1281 } 1282 } 1283 #endif 1284 #if defined(INET) && defined(INET6) 1285 else 1286 #endif 1287 #ifdef INET 1288 { 1289 if (th->th_dport == th->th_sport && 1290 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1291 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1292 log(LOG_DEBUG, "%s; %s: Listen socket: " 1293 "Connection attempt from/to self " 1294 "ignored\n", s, __func__); 1295 goto dropunlock; 1296 } 1297 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1298 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1299 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1300 in_ifnet_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1301 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1302 log(LOG_DEBUG, "%s; %s: Listen socket: " 1303 "Connection attempt from/to broad- " 1304 "or multicast address ignored\n", 1305 s, __func__); 1306 goto dropunlock; 1307 } 1308 } 1309 #endif 1310 /* 1311 * SYN appears to be valid. Create compressed TCP state 1312 * for syncache. 1313 */ 1314 TCP_PROBE3(debug__input, tp, th, m); 1315 tcp_dooptions(&to, optp, optlen, TO_SYN); 1316 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, 1317 iptos, port)) != NULL) 1318 goto tfo_socket_result; 1319 1320 /* 1321 * Entry added to syncache and mbuf consumed. 1322 * Only the listen socket is unlocked by syncache_add(). 1323 */ 1324 return (IPPROTO_DONE); 1325 } 1326 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1327 if (tp->t_flags & TF_SIGNATURE) { 1328 tcp_dooptions(&to, optp, optlen, thflags); 1329 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1330 TCPSTAT_INC(tcps_sig_err_nosigopt); 1331 goto dropunlock; 1332 } 1333 if (!TCPMD5_ENABLED() || 1334 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1335 goto dropunlock; 1336 } 1337 #endif 1338 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1339 1340 /* 1341 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1342 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1343 * the inpcb, and unlocks pcbinfo. 1344 * 1345 * XXXGL: in case of a pure SYN arriving on existing connection 1346 * TCP stacks won't need to modify the PCB, they would either drop 1347 * the segment silently, or send a challenge ACK. However, we try 1348 * to upgrade the lock, because calling convention for stacks is 1349 * write-lock on PCB. If upgrade fails, drop the SYN. 1350 */ 1351 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0) 1352 goto dropunlock; 1353 1354 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, tlen, iptos); 1355 return (IPPROTO_DONE); 1356 1357 dropwithreset: 1358 /* 1359 * When blackholing do not respond with a RST but 1360 * completely ignore the segment and drop it. 1361 */ 1362 if (((!closed_port && V_blackhole == 3) || 1363 (closed_port && 1364 ((V_blackhole == 1 && (thflags & TH_SYN)) || V_blackhole > 1))) && 1365 (V_blackhole_local || ( 1366 #ifdef INET6 1367 isipv6 ? !in6_localip(&ip6->ip6_src) : 1368 #endif 1369 #ifdef INET 1370 !in_localip(ip->ip_src) 1371 #else 1372 true 1373 #endif 1374 ))) 1375 goto dropunlock; 1376 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1377 tcp_dropwithreset(m, th, tp, tlen); 1378 m = NULL; /* mbuf chain got consumed. */ 1379 1380 dropunlock: 1381 if (m != NULL) 1382 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1383 1384 if (inp != NULL) 1385 INP_UNLOCK(inp); 1386 1387 drop: 1388 if (s != NULL) 1389 free(s, M_TCPLOG); 1390 if (m != NULL) 1391 m_freem(m); 1392 return (IPPROTO_DONE); 1393 } 1394 1395 /* 1396 * Automatic sizing of receive socket buffer. Often the send 1397 * buffer size is not optimally adjusted to the actual network 1398 * conditions at hand (delay bandwidth product). Setting the 1399 * buffer size too small limits throughput on links with high 1400 * bandwidth and high delay (eg. trans-continental/oceanic links). 1401 * 1402 * On the receive side the socket buffer memory is only rarely 1403 * used to any significant extent. This allows us to be much 1404 * more aggressive in scaling the receive socket buffer. For 1405 * the case that the buffer space is actually used to a large 1406 * extent and we run out of kernel memory we can simply drop 1407 * the new segments; TCP on the sender will just retransmit it 1408 * later. Setting the buffer size too big may only consume too 1409 * much kernel memory if the application doesn't read() from 1410 * the socket or packet loss or reordering makes use of the 1411 * reassembly queue. 1412 * 1413 * The criteria to step up the receive buffer one notch are: 1414 * 1. Application has not set receive buffer size with 1415 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1416 * 2. the number of bytes received during 1/2 of an sRTT 1417 * is at least 3/8 of the current socket buffer size. 1418 * 3. receive buffer size has not hit maximal automatic size; 1419 * 1420 * If all of the criteria are met, we increase the socket buffer 1421 * by a 1/2 (bounded by the max). This allows us to keep ahead 1422 * of slow-start but also makes it so our peer never gets limited 1423 * by our rwnd which we then open up causing a burst. 1424 * 1425 * This algorithm does two steps per RTT at most and only if 1426 * we receive a bulk stream w/o packet losses or reorderings. 1427 * Shrinking the buffer during idle times is not necessary as 1428 * it doesn't consume any memory when idle. 1429 * 1430 * TODO: Only step up if the application is actually serving 1431 * the buffer to better manage the socket buffer resources. 1432 */ 1433 int 1434 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1435 struct tcpcb *tp, int tlen) 1436 { 1437 int newsize = 0; 1438 1439 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1440 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1441 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1442 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1443 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1444 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1445 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1446 } 1447 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1448 1449 /* Start over with next RTT. */ 1450 tp->rfbuf_ts = 0; 1451 tp->rfbuf_cnt = 0; 1452 } else { 1453 tp->rfbuf_cnt += tlen; /* add up */ 1454 } 1455 return (newsize); 1456 } 1457 1458 int 1459 tcp_input(struct mbuf **mp, int *offp, int proto) 1460 { 1461 return(tcp_input_with_port(mp, offp, proto, 0)); 1462 } 1463 1464 static void 1465 tcp_handle_wakeup(struct tcpcb *tp) 1466 { 1467 1468 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1469 1470 if (tp->t_flags & TF_WAKESOR) { 1471 struct socket *so = tptosocket(tp); 1472 1473 tp->t_flags &= ~TF_WAKESOR; 1474 SOCK_RECVBUF_LOCK_ASSERT(so); 1475 sorwakeup_locked(so); 1476 } 1477 } 1478 1479 void 1480 tcp_do_segment(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th, 1481 int drop_hdrlen, int tlen, uint8_t iptos) 1482 { 1483 uint16_t thflags; 1484 int acked, ourfinisacked, needoutput = 0; 1485 sackstatus_t sack_changed; 1486 int todrop, win, incforsyn = 0; 1487 uint32_t tiwin; 1488 uint16_t nsegs; 1489 char *s; 1490 struct inpcb *inp = tptoinpcb(tp); 1491 struct socket *so = tptosocket(tp); 1492 struct in_conninfo *inc = &inp->inp_inc; 1493 struct mbuf *mfree; 1494 struct tcpopt to; 1495 int tfo_syn; 1496 u_int maxseg = 0; 1497 uint32_t prev_sacked_bytes = 0; 1498 bool no_data; 1499 1500 no_data = (tlen == 0); 1501 thflags = tcp_get_flags(th); 1502 tp->sackhint.last_sack_ack = 0; 1503 sack_changed = SACK_NOCHANGE; 1504 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1505 1506 NET_EPOCH_ASSERT(); 1507 INP_WLOCK_ASSERT(inp); 1508 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1509 __func__)); 1510 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1511 __func__)); 1512 1513 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1514 tlen, NULL, true); 1515 1516 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1517 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1518 log(LOG_DEBUG, "%s; %s: " 1519 "SYN|FIN segment ignored (based on " 1520 "sysctl setting)\n", s, __func__); 1521 free(s, M_TCPLOG); 1522 } 1523 goto drop; 1524 } 1525 1526 /* 1527 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1528 * check SEQ.ACK first. 1529 */ 1530 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1531 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1532 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1533 goto dropwithreset; 1534 } 1535 1536 /* 1537 * Segment received on connection. 1538 * Reset idle time and keep-alive timer. 1539 * XXX: This should be done after segment 1540 * validation to ignore broken/spoofed segs. 1541 */ 1542 if (tp->t_idle_reduce && 1543 (tp->snd_max == tp->snd_una) && 1544 ((ticks - tp->t_rcvtime) >= tp->t_rxtcur)) 1545 cc_after_idle(tp); 1546 tp->t_rcvtime = ticks; 1547 1548 if (thflags & TH_FIN) 1549 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN); 1550 /* 1551 * Scale up the window into a 32-bit value. 1552 * For the SYN_SENT state the scale is zero. 1553 */ 1554 tiwin = th->th_win << tp->snd_scale; 1555 #ifdef STATS 1556 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1557 #endif 1558 1559 /* 1560 * TCP ECN processing. 1561 */ 1562 if (tcp_ecn_input_segment(tp, thflags, tlen, 1563 tcp_packets_this_ack(tp, th->th_ack), 1564 iptos)) 1565 cc_cong_signal(tp, th, CC_ECN); 1566 1567 /* 1568 * Parse options on any incoming segment. 1569 */ 1570 tcp_dooptions(&to, (u_char *)(th + 1), 1571 (th->th_off << 2) - sizeof(struct tcphdr), 1572 (thflags & TH_SYN) ? TO_SYN : 0); 1573 if (tp->t_flags2 & TF2_PROC_SACK_PROHIBIT) { 1574 /* 1575 * We don't look at sack's from the 1576 * peer because the MSS is too small which 1577 * can subject us to an attack. 1578 */ 1579 to.to_flags &= ~TOF_SACK; 1580 } 1581 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1582 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1583 (to.to_flags & TOF_SIGNATURE) == 0) { 1584 TCPSTAT_INC(tcps_sig_err_sigopt); 1585 /* XXX: should drop? */ 1586 } 1587 #endif 1588 /* 1589 * If echoed timestamp is later than the current time, 1590 * fall back to non RFC1323 RTT calculation. Normalize 1591 * timestamp if syncookies were used when this connection 1592 * was established. 1593 */ 1594 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1595 to.to_tsecr -= tp->ts_offset; 1596 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) { 1597 to.to_tsecr = 0; 1598 } 1599 } 1600 /* 1601 * Process options only when we get SYN/ACK back. The SYN case 1602 * for incoming connections is handled in tcp_syncache. 1603 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1604 * or <SYN,ACK>) segment itself is never scaled. 1605 * XXX this is traditional behavior, may need to be cleaned up. 1606 */ 1607 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1608 /* Handle parallel SYN for ECN */ 1609 tcp_ecn_input_parallel_syn(tp, thflags, iptos); 1610 if ((to.to_flags & TOF_SCALE) && 1611 (tp->t_flags & TF_REQ_SCALE) && 1612 !(tp->t_flags & TF_NOOPT)) { 1613 tp->t_flags |= TF_RCVD_SCALE; 1614 tp->snd_scale = to.to_wscale; 1615 } else { 1616 tp->t_flags &= ~TF_REQ_SCALE; 1617 } 1618 /* 1619 * Initial send window. It will be updated with 1620 * the next incoming segment to the scaled value. 1621 */ 1622 tp->snd_wnd = th->th_win; 1623 if ((to.to_flags & TOF_TS) && 1624 (tp->t_flags & TF_REQ_TSTMP) && 1625 !(tp->t_flags & TF_NOOPT)) { 1626 tp->t_flags |= TF_RCVD_TSTMP; 1627 tp->ts_recent = to.to_tsval; 1628 tp->ts_recent_age = tcp_ts_getticks(); 1629 } else { 1630 tp->t_flags &= ~TF_REQ_TSTMP; 1631 } 1632 if (to.to_flags & TOF_MSS) { 1633 tcp_mss(tp, to.to_mss); 1634 } 1635 if ((tp->t_flags & TF_SACK_PERMIT) && 1636 (!(to.to_flags & TOF_SACKPERM) || 1637 (tp->t_flags & TF_NOOPT))) { 1638 tp->t_flags &= ~TF_SACK_PERMIT; 1639 } 1640 if (tp->t_flags & TF_FASTOPEN) { 1641 if ((to.to_flags & TOF_FASTOPEN) && 1642 !(tp->t_flags & TF_NOOPT)) { 1643 uint16_t mss; 1644 1645 if (to.to_flags & TOF_MSS) { 1646 mss = to.to_mss; 1647 } else { 1648 if ((inp->inp_vflag & INP_IPV6) != 0) { 1649 mss = TCP6_MSS; 1650 } else { 1651 mss = TCP_MSS; 1652 } 1653 } 1654 tcp_fastopen_update_cache(tp, mss, 1655 to.to_tfo_len, to.to_tfo_cookie); 1656 } else { 1657 tcp_fastopen_disable_path(tp); 1658 } 1659 } 1660 } 1661 1662 /* 1663 * If timestamps were negotiated during SYN/ACK and a 1664 * segment without a timestamp is received, silently drop 1665 * the segment, unless it is a RST segment or missing timestamps are 1666 * tolerated. 1667 * See section 3.2 of RFC 7323. 1668 */ 1669 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1670 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1671 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1672 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1673 "segment processed normally\n", 1674 s, __func__); 1675 free(s, M_TCPLOG); 1676 } 1677 } else { 1678 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1679 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1680 "segment silently dropped\n", s, __func__); 1681 free(s, M_TCPLOG); 1682 } 1683 goto drop; 1684 } 1685 } 1686 /* 1687 * If timestamps were not negotiated during SYN/ACK and a 1688 * segment with a timestamp is received, ignore the 1689 * timestamp and process the packet normally. 1690 * See section 3.2 of RFC 7323. 1691 */ 1692 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1693 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1694 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1695 "segment processed normally\n", s, __func__); 1696 free(s, M_TCPLOG); 1697 } 1698 } 1699 1700 /* 1701 * Header prediction: check for the two common cases 1702 * of a uni-directional data xfer. If the packet has 1703 * no control flags, is in-sequence, the window didn't 1704 * change and we're not retransmitting, it's a 1705 * candidate. If the length is zero and the ack moved 1706 * forward, we're the sender side of the xfer. Just 1707 * free the data acked & wake any higher level process 1708 * that was blocked waiting for space. If the length 1709 * is non-zero and the ack didn't move, we're the 1710 * receiver side. If we're getting packets in-order 1711 * (the reassembly queue is empty), add the data to 1712 * the socket buffer and note that we need a delayed ack. 1713 * Make sure that the hidden state-flags are also off. 1714 * Since we check for TCPS_ESTABLISHED first, it can only 1715 * be TH_NEEDSYN. 1716 */ 1717 if (tp->t_state == TCPS_ESTABLISHED && 1718 th->th_seq == tp->rcv_nxt && 1719 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1720 tp->snd_nxt == tp->snd_max && 1721 tiwin && tiwin == tp->snd_wnd && 1722 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1723 SEGQ_EMPTY(tp) && 1724 ((to.to_flags & TOF_TS) == 0 || 1725 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1726 /* 1727 * If last ACK falls within this segment's sequence numbers, 1728 * record the timestamp. 1729 * NOTE that the test is modified according to the latest 1730 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1731 */ 1732 if ((to.to_flags & TOF_TS) != 0 && 1733 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1734 tp->ts_recent_age = tcp_ts_getticks(); 1735 tp->ts_recent = to.to_tsval; 1736 } 1737 1738 if (no_data) { 1739 if (SEQ_GT(th->th_ack, tp->snd_una) && 1740 SEQ_LEQ(th->th_ack, tp->snd_max) && 1741 !IN_RECOVERY(tp->t_flags) && 1742 (to.to_flags & TOF_SACK) == 0 && 1743 TAILQ_EMPTY(&tp->snd_holes)) { 1744 /* 1745 * This is a pure ack for outstanding data. 1746 */ 1747 TCPSTAT_INC(tcps_predack); 1748 1749 /* 1750 * "bad retransmit" recovery. 1751 */ 1752 if (tp->t_rxtshift == 1 && 1753 tp->t_flags & TF_PREVVALID && 1754 tp->t_badrxtwin != 0 && 1755 (((to.to_flags & TOF_TS) != 0 && 1756 to.to_tsecr != 0 && 1757 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) || 1758 ((to.to_flags & TOF_TS) == 0 && 1759 TSTMP_LT(ticks, tp->t_badrxtwin)))) 1760 cc_cong_signal(tp, th, CC_RTO_ERR); 1761 1762 /* 1763 * Recalculate the transmit timer / rtt. 1764 * 1765 * Some boxes send broken timestamp replies 1766 * during the SYN+ACK phase, ignore 1767 * timestamps of 0 or we could calculate a 1768 * huge RTT and blow up the retransmit timer. 1769 */ 1770 if ((to.to_flags & TOF_TS) != 0 && 1771 to.to_tsecr) { 1772 uint32_t t; 1773 1774 t = tcp_ts_getticks() - to.to_tsecr; 1775 if (!tp->t_rttlow || tp->t_rttlow > t) 1776 tp->t_rttlow = t; 1777 tcp_xmit_timer(tp, 1778 TCP_TS_TO_TICKS(t) + 1); 1779 } else if (tp->t_rtttime && 1780 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1781 if (!tp->t_rttlow || 1782 tp->t_rttlow > ticks - tp->t_rtttime) 1783 tp->t_rttlow = ticks - tp->t_rtttime; 1784 tcp_xmit_timer(tp, 1785 ticks - tp->t_rtttime); 1786 } 1787 acked = BYTES_THIS_ACK(tp, th); 1788 1789 #ifdef TCP_HHOOK 1790 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1791 hhook_run_tcp_est_in(tp, th, &to); 1792 #endif 1793 1794 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1795 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1796 sbdrop(&so->so_snd, acked); 1797 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1798 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1799 tp->snd_recover = th->th_ack - 1; 1800 1801 /* 1802 * Let the congestion control algorithm update 1803 * congestion control related information. This 1804 * typically means increasing the congestion 1805 * window. 1806 */ 1807 cc_ack_received(tp, th, nsegs, CC_ACK); 1808 1809 tp->snd_una = th->th_ack; 1810 /* 1811 * Pull snd_wl2 up to prevent seq wrap relative 1812 * to th_ack. 1813 */ 1814 tp->snd_wl2 = th->th_ack; 1815 tp->t_dupacks = 0; 1816 m_freem(m); 1817 1818 /* 1819 * If all outstanding data are acked, stop 1820 * retransmit timer, otherwise restart timer 1821 * using current (possibly backed-off) value. 1822 * If process is waiting for space, 1823 * wakeup/selwakeup/signal. If data 1824 * are ready to send, let tcp_output 1825 * decide between more output or persist. 1826 */ 1827 TCP_PROBE3(debug__input, tp, th, m); 1828 /* 1829 * Clear t_acktime if remote side has ACKd 1830 * all data in the socket buffer. 1831 * Otherwise, update t_acktime if we received 1832 * a sufficiently large ACK. 1833 */ 1834 if (sbavail(&so->so_snd) == 0) 1835 tp->t_acktime = 0; 1836 else if (acked > 1) 1837 tp->t_acktime = ticks; 1838 if (tp->snd_una == tp->snd_max) 1839 tcp_timer_activate(tp, TT_REXMT, 0); 1840 else if (!tcp_timer_active(tp, TT_PERSIST)) 1841 tcp_timer_activate(tp, TT_REXMT, 1842 TP_RXTCUR(tp)); 1843 sowwakeup(so); 1844 /* 1845 * Only call tcp_output when there 1846 * is new data available to be sent 1847 * or we need to send an ACK. 1848 */ 1849 if ((tp->t_flags & TF_ACKNOW) || 1850 (sbavail(&so->so_snd) >= 1851 SEQ_SUB(tp->snd_max, tp->snd_una))) { 1852 (void) tcp_output(tp); 1853 } 1854 goto check_delack; 1855 } 1856 } else if (th->th_ack == tp->snd_una && 1857 tlen <= sbspace(&so->so_rcv)) { 1858 int newsize = 0; /* automatic sockbuf scaling */ 1859 1860 /* 1861 * This is a pure, in-sequence data packet with 1862 * nothing on the reassembly queue and we have enough 1863 * buffer space to take it. 1864 */ 1865 /* Clean receiver SACK report if present */ 1866 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1867 tcp_clean_sackreport(tp); 1868 TCPSTAT_INC(tcps_preddat); 1869 tp->rcv_nxt += tlen; 1870 if (tlen && 1871 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1872 (tp->t_fbyte_in == 0)) { 1873 tp->t_fbyte_in = ticks; 1874 if (tp->t_fbyte_in == 0) 1875 tp->t_fbyte_in = 1; 1876 if (tp->t_fbyte_out && tp->t_fbyte_in) 1877 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1878 } 1879 /* 1880 * Pull snd_wl1 up to prevent seq wrap relative to 1881 * th_seq. 1882 */ 1883 tp->snd_wl1 = th->th_seq; 1884 /* 1885 * Pull rcv_up up to prevent seq wrap relative to 1886 * rcv_nxt. 1887 */ 1888 tp->rcv_up = tp->rcv_nxt; 1889 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1890 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1891 TCP_PROBE3(debug__input, tp, th, m); 1892 1893 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1894 1895 /* Add data to socket buffer. */ 1896 SOCK_RECVBUF_LOCK(so); 1897 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1898 m_freem(m); 1899 } else { 1900 /* 1901 * Set new socket buffer size. 1902 * Give up when limit is reached. 1903 */ 1904 if (newsize) 1905 if (!sbreserve_locked(so, SO_RCV, 1906 newsize, NULL)) 1907 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1908 m_adj(m, drop_hdrlen); /* delayed header drop */ 1909 sbappendstream_locked(&so->so_rcv, m, 0); 1910 } 1911 /* NB: sorwakeup_locked() does an implicit unlock. */ 1912 sorwakeup_locked(so); 1913 if (DELAY_ACK(tp, tlen)) { 1914 tp->t_flags |= TF_DELACK; 1915 } else { 1916 tp->t_flags |= TF_ACKNOW; 1917 (void) tcp_output(tp); 1918 } 1919 goto check_delack; 1920 } 1921 } 1922 1923 /* 1924 * Calculate amount of space in receive window, 1925 * and then do TCP input processing. 1926 * Receive window is amount of space in rcv queue, 1927 * but not less than advertised window. 1928 */ 1929 win = sbspace(&so->so_rcv); 1930 if (win < 0) 1931 win = 0; 1932 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1933 1934 switch (tp->t_state) { 1935 /* 1936 * If the state is SYN_RECEIVED: 1937 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1938 */ 1939 case TCPS_SYN_RECEIVED: 1940 if (thflags & TH_RST) { 1941 /* Handle RST segments later. */ 1942 break; 1943 } 1944 if ((thflags & TH_ACK) && 1945 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1946 SEQ_GT(th->th_ack, tp->snd_max))) { 1947 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1948 goto dropwithreset; 1949 } 1950 if (tp->t_flags & TF_FASTOPEN) { 1951 /* 1952 * When a TFO connection is in SYN_RECEIVED, the 1953 * only valid packets are the initial SYN, a 1954 * retransmit/copy of the initial SYN (possibly with 1955 * a subset of the original data), a valid ACK, a 1956 * FIN, or a RST. 1957 */ 1958 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1959 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1960 goto dropwithreset; 1961 } else if (thflags & TH_SYN) { 1962 /* non-initial SYN is ignored */ 1963 if ((tcp_timer_active(tp, TT_DELACK) || 1964 tcp_timer_active(tp, TT_REXMT))) 1965 goto drop; 1966 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1967 goto drop; 1968 } 1969 } 1970 break; 1971 1972 /* 1973 * If the state is SYN_SENT: 1974 * if seg contains a RST with valid ACK (SEQ.ACK has already 1975 * been verified), then drop the connection. 1976 * if seg contains a RST without an ACK, drop the seg. 1977 * if seg does not contain SYN, then drop the seg. 1978 * Otherwise this is an acceptable SYN segment 1979 * initialize tp->rcv_nxt and tp->irs 1980 * if seg contains ack then advance tp->snd_una 1981 * if seg contains an ECE and ECN support is enabled, the stream 1982 * is ECN capable. 1983 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1984 * arrange for segment to be acked (eventually) 1985 * continue processing rest of data/controls, beginning with URG 1986 */ 1987 case TCPS_SYN_SENT: 1988 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1989 TCP_PROBE5(connect__refused, NULL, tp, 1990 m, tp, th); 1991 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1992 tp = tcp_drop(tp, ECONNREFUSED); 1993 } 1994 if (thflags & TH_RST) 1995 goto drop; 1996 if (!(thflags & TH_SYN)) 1997 goto drop; 1998 1999 tp->irs = th->th_seq; 2000 tcp_rcvseqinit(tp); 2001 if (thflags & TH_ACK) { 2002 int tfo_partial_ack = 0; 2003 2004 TCPSTAT_INC(tcps_connects); 2005 soisconnected(so); 2006 #ifdef MAC 2007 mac_socketpeer_set_from_mbuf(m, so); 2008 #endif 2009 /* Do window scaling on this connection? */ 2010 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2011 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2012 tp->rcv_scale = tp->request_r_scale; 2013 } 2014 tp->rcv_adv += min(tp->rcv_wnd, 2015 TCP_MAXWIN << tp->rcv_scale); 2016 tp->snd_una++; /* SYN is acked */ 2017 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2018 tp->snd_nxt = tp->snd_una; 2019 /* 2020 * If not all the data that was sent in the TFO SYN 2021 * has been acked, resend the remainder right away. 2022 */ 2023 if ((tp->t_flags & TF_FASTOPEN) && 2024 (tp->snd_una != tp->snd_max)) { 2025 tp->snd_nxt = th->th_ack; 2026 tfo_partial_ack = 1; 2027 } 2028 /* 2029 * If there's data, delay ACK; if there's also a FIN 2030 * ACKNOW will be turned on later. 2031 */ 2032 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2033 tcp_timer_activate(tp, TT_DELACK, 2034 tcp_delacktime); 2035 else 2036 tp->t_flags |= TF_ACKNOW; 2037 2038 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2039 2040 /* 2041 * Received <SYN,ACK> in SYN_SENT[*] state. 2042 * Transitions: 2043 * SYN_SENT --> ESTABLISHED 2044 * SYN_SENT* --> FIN_WAIT_1 2045 */ 2046 tp->t_starttime = ticks; 2047 if (tp->t_flags & TF_NEEDFIN) { 2048 tp->t_acktime = ticks; 2049 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2050 tp->t_flags &= ~TF_NEEDFIN; 2051 thflags &= ~TH_SYN; 2052 } else { 2053 tcp_state_change(tp, TCPS_ESTABLISHED); 2054 TCP_PROBE5(connect__established, NULL, tp, 2055 m, tp, th); 2056 cc_conn_init(tp); 2057 tcp_timer_activate(tp, TT_KEEP, 2058 TP_KEEPIDLE(tp)); 2059 } 2060 } else { 2061 /* 2062 * Received initial SYN in SYN-SENT[*] state => 2063 * simultaneous open. 2064 * If it succeeds, connection is * half-synchronized. 2065 * Otherwise, do 3-way handshake: 2066 * SYN-SENT -> SYN-RECEIVED 2067 * SYN-SENT* -> SYN-RECEIVED* 2068 */ 2069 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN); 2070 tcp_timer_activate(tp, TT_REXMT, 0); 2071 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2072 } 2073 2074 /* 2075 * Advance th->th_seq to correspond to first data byte. 2076 * If data, trim to stay within window, 2077 * dropping FIN if necessary. 2078 */ 2079 th->th_seq++; 2080 if (tlen > tp->rcv_wnd) { 2081 todrop = tlen - tp->rcv_wnd; 2082 m_adj(m, -todrop); 2083 tlen = tp->rcv_wnd; 2084 thflags &= ~TH_FIN; 2085 TCPSTAT_INC(tcps_rcvpackafterwin); 2086 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2087 } 2088 tp->snd_wl1 = th->th_seq - 1; 2089 tp->rcv_up = th->th_seq; 2090 /* 2091 * Client side of transaction: already sent SYN and data. 2092 * If the remote host used T/TCP to validate the SYN, 2093 * our data will be ACK'd; if so, enter normal data segment 2094 * processing in the middle of step 5, ack processing. 2095 * Otherwise, goto step 6. 2096 */ 2097 if (thflags & TH_ACK) 2098 goto process_ACK; 2099 2100 goto step6; 2101 } 2102 2103 /* 2104 * States other than LISTEN or SYN_SENT. 2105 * First check the RST flag and sequence number since reset segments 2106 * are exempt from the timestamp and connection count tests. This 2107 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2108 * below which allowed reset segments in half the sequence space 2109 * to fall though and be processed (which gives forged reset 2110 * segments with a random sequence number a 50 percent chance of 2111 * killing a connection). 2112 * Then check timestamp, if present. 2113 * Then check the connection count, if present. 2114 * Then check that at least some bytes of segment are within 2115 * receive window. If segment begins before rcv_nxt, 2116 * drop leading data (and SYN); if nothing left, just ack. 2117 */ 2118 if (thflags & TH_RST) { 2119 /* 2120 * RFC5961 Section 3.2 2121 * 2122 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2123 * - If RST is in window, we send challenge ACK. 2124 * 2125 * Note 1: to take into account delayed ACKs, we should 2126 * test against last_ack_sent in addition to rcv_nxt. 2127 * Note 2: we handle special case of closed window, not 2128 * covered by the RFC. 2129 * Note 3 (XXXMT): check against rcv_adv instead of 2130 * tp->rcv_nxt + tp->rcv_wnd. 2131 */ 2132 if ((tp->rcv_wnd > 0 && 2133 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2134 SEQ_LT(th->th_seq, tp->rcv_nxt + tp->rcv_wnd)) || 2135 (tp->rcv_wnd == 0 && 2136 (tp->last_ack_sent == th->th_seq || 2137 tp->rcv_nxt == th->th_seq))) { 2138 KASSERT(tp->t_state != TCPS_SYN_SENT, 2139 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2140 __func__, th, tp)); 2141 2142 if (V_tcp_insecure_rst || 2143 tp->last_ack_sent == th->th_seq || 2144 tp->rcv_nxt == th->th_seq) { 2145 TCPSTAT_INC(tcps_drops); 2146 /* Drop the connection. */ 2147 switch (tp->t_state) { 2148 case TCPS_SYN_RECEIVED: 2149 so->so_error = ECONNREFUSED; 2150 goto close; 2151 case TCPS_ESTABLISHED: 2152 case TCPS_FIN_WAIT_1: 2153 case TCPS_FIN_WAIT_2: 2154 case TCPS_CLOSE_WAIT: 2155 case TCPS_CLOSING: 2156 case TCPS_LAST_ACK: 2157 so->so_error = ECONNRESET; 2158 close: 2159 /* FALLTHROUGH */ 2160 default: 2161 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST); 2162 tp = tcp_close(tp); 2163 } 2164 } else { 2165 TCPSTAT_INC(tcps_badrst); 2166 tcp_send_challenge_ack(tp, th, m); 2167 m = NULL; 2168 } 2169 } 2170 goto drop; 2171 } 2172 2173 /* 2174 * RFC5961 Section 4.2 2175 * Send challenge ACK for any SYN in synchronized state. 2176 */ 2177 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2178 tp->t_state != TCPS_SYN_RECEIVED) { 2179 TCPSTAT_INC(tcps_badsyn); 2180 if (V_tcp_insecure_syn && 2181 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2182 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2183 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2184 tp = tcp_drop(tp, ECONNRESET); 2185 } else { 2186 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2187 tcp_send_challenge_ack(tp, th, m); 2188 m = NULL; 2189 } 2190 goto drop; 2191 } 2192 2193 /* 2194 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2195 * and it's less than ts_recent, drop it. 2196 */ 2197 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2198 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2199 /* Check to see if ts_recent is over 24 days old. */ 2200 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2201 /* 2202 * Invalidate ts_recent. If this segment updates 2203 * ts_recent, the age will be reset later and ts_recent 2204 * will get a valid value. If it does not, setting 2205 * ts_recent to zero will at least satisfy the 2206 * requirement that zero be placed in the timestamp 2207 * echo reply when ts_recent isn't valid. The 2208 * age isn't reset until we get a valid ts_recent 2209 * because we don't want out-of-order segments to be 2210 * dropped when ts_recent is old. 2211 */ 2212 tp->ts_recent = 0; 2213 } else { 2214 TCPSTAT_INC(tcps_rcvduppack); 2215 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2216 TCPSTAT_INC(tcps_pawsdrop); 2217 if (tlen) 2218 goto dropafterack; 2219 goto drop; 2220 } 2221 } 2222 2223 /* 2224 * In the SYN-RECEIVED state, validate that the packet belongs to 2225 * this connection before trimming the data to fit the receive 2226 * window. Check the sequence number versus IRS since we know 2227 * the sequence numbers haven't wrapped. This is a partial fix 2228 * for the "LAND" DoS attack. 2229 */ 2230 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2231 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2232 goto dropwithreset; 2233 } 2234 2235 todrop = tp->rcv_nxt - th->th_seq; 2236 if (todrop > 0) { 2237 if (thflags & TH_SYN) { 2238 thflags &= ~TH_SYN; 2239 th->th_seq++; 2240 if (th->th_urp > 1) 2241 th->th_urp--; 2242 else 2243 thflags &= ~TH_URG; 2244 todrop--; 2245 } 2246 /* 2247 * Following if statement from Stevens, vol. 2, p. 960. 2248 */ 2249 if (todrop > tlen 2250 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2251 /* 2252 * Any valid FIN must be to the left of the window. 2253 * At this point the FIN must be a duplicate or out 2254 * of sequence; drop it. 2255 */ 2256 thflags &= ~TH_FIN; 2257 2258 /* 2259 * Send an ACK to resynchronize and drop any data. 2260 * But keep on processing for RST or ACK. 2261 */ 2262 tp->t_flags |= TF_ACKNOW; 2263 todrop = tlen; 2264 TCPSTAT_INC(tcps_rcvduppack); 2265 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2266 } else { 2267 TCPSTAT_INC(tcps_rcvpartduppack); 2268 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2269 } 2270 /* 2271 * DSACK - add SACK block for dropped range 2272 */ 2273 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2274 tcp_update_sack_list(tp, th->th_seq, 2275 th->th_seq + todrop); 2276 /* 2277 * ACK now, as the next in-sequence segment 2278 * will clear the DSACK block again 2279 */ 2280 tp->t_flags |= TF_ACKNOW; 2281 } 2282 drop_hdrlen += todrop; /* drop from the top afterwards */ 2283 th->th_seq += todrop; 2284 tlen -= todrop; 2285 if (th->th_urp > todrop) 2286 th->th_urp -= todrop; 2287 else { 2288 thflags &= ~TH_URG; 2289 th->th_urp = 0; 2290 } 2291 } 2292 2293 /* 2294 * If new data are received on a connection after the 2295 * user processes are gone, then RST the other end if 2296 * no FIN has been processed. 2297 */ 2298 if ((tp->t_flags & TF_CLOSED) && tlen > 0 && 2299 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2300 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2301 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2302 "after socket was closed, " 2303 "sending RST and removing tcpcb\n", 2304 s, __func__, tcpstates[tp->t_state], tlen); 2305 free(s, M_TCPLOG); 2306 } 2307 tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE); 2308 /* tcp_close will kill the inp pre-log the Reset */ 2309 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST); 2310 tp = tcp_close(tp); 2311 TCPSTAT_INC(tcps_rcvafterclose); 2312 goto dropwithreset; 2313 } 2314 2315 /* 2316 * If segment ends after window, drop trailing data 2317 * (and PUSH and FIN); if nothing left, just ACK. 2318 */ 2319 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2320 if (todrop > 0) { 2321 TCPSTAT_INC(tcps_rcvpackafterwin); 2322 if (todrop >= tlen) { 2323 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2324 /* 2325 * If window is closed can only take segments at 2326 * window edge, and have to drop data and PUSH from 2327 * incoming segments. Continue processing, but 2328 * remember to ack. Otherwise, drop segment 2329 * and ack. 2330 */ 2331 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2332 tp->t_flags |= TF_ACKNOW; 2333 TCPSTAT_INC(tcps_rcvwinprobe); 2334 } else 2335 goto dropafterack; 2336 } else 2337 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2338 m_adj(m, -todrop); 2339 tlen -= todrop; 2340 thflags &= ~(TH_PUSH|TH_FIN); 2341 } 2342 2343 /* 2344 * If last ACK falls within this segment's sequence numbers, 2345 * record its timestamp. 2346 * NOTE: 2347 * 1) That the test incorporates suggestions from the latest 2348 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2349 * 2) That updating only on newer timestamps interferes with 2350 * our earlier PAWS tests, so this check should be solely 2351 * predicated on the sequence space of this segment. 2352 * 3) That we modify the segment boundary check to be 2353 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2354 * instead of RFC1323's 2355 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2356 * This modified check allows us to overcome RFC1323's 2357 * limitations as described in Stevens TCP/IP Illustrated 2358 * Vol. 2 p.869. In such cases, we can still calculate the 2359 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2360 */ 2361 if ((to.to_flags & TOF_TS) != 0 && 2362 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2363 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2364 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2365 tp->ts_recent_age = tcp_ts_getticks(); 2366 tp->ts_recent = to.to_tsval; 2367 } 2368 2369 /* 2370 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2371 * flag is on (half-synchronized state), then queue data for 2372 * later processing; else drop segment and return. 2373 */ 2374 if ((thflags & TH_ACK) == 0) { 2375 if (tp->t_state == TCPS_SYN_RECEIVED || 2376 (tp->t_flags & TF_NEEDSYN)) { 2377 if (tp->t_state == TCPS_SYN_RECEIVED && 2378 (tp->t_flags & TF_FASTOPEN)) { 2379 tp->snd_wnd = tiwin; 2380 cc_conn_init(tp); 2381 } 2382 goto step6; 2383 } else if (tp->t_flags & TF_ACKNOW) 2384 goto dropafterack; 2385 else 2386 goto drop; 2387 } 2388 2389 /* 2390 * Ack processing. 2391 */ 2392 if (SEQ_GT(tp->snd_una, tp->iss + (TCP_MAXWIN << tp->snd_scale))) { 2393 /* Checking SEG.ACK against ISS is definitely redundant. */ 2394 tp->t_flags2 |= TF2_NO_ISS_CHECK; 2395 } 2396 if (!V_tcp_insecure_ack) { 2397 tcp_seq seq_min; 2398 bool ghost_ack_check; 2399 2400 if (tp->t_flags2 & TF2_NO_ISS_CHECK) { 2401 /* Check for too old ACKs (RFC 5961, Section 5.2). */ 2402 seq_min = tp->snd_una - tp->max_sndwnd; 2403 ghost_ack_check = false; 2404 } else { 2405 if (SEQ_GT(tp->iss + 1, tp->snd_una - tp->max_sndwnd)) { 2406 /* Checking for ghost ACKs is stricter. */ 2407 seq_min = tp->iss + 1; 2408 ghost_ack_check = true; 2409 } else { 2410 /* 2411 * Checking for too old ACKs (RFC 5961, 2412 * Section 5.2) is stricter. 2413 */ 2414 seq_min = tp->snd_una - tp->max_sndwnd; 2415 ghost_ack_check = false; 2416 } 2417 } 2418 if (SEQ_LT(th->th_ack, seq_min)) { 2419 if (ghost_ack_check) 2420 TCPSTAT_INC(tcps_rcvghostack); 2421 else 2422 TCPSTAT_INC(tcps_rcvacktooold); 2423 tcp_send_challenge_ack(tp, th, m); 2424 m = NULL; 2425 goto drop; 2426 } 2427 } 2428 switch (tp->t_state) { 2429 /* 2430 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2431 * ESTABLISHED state and continue processing. 2432 * The ACK was checked above. 2433 */ 2434 case TCPS_SYN_RECEIVED: 2435 2436 TCPSTAT_INC(tcps_connects); 2437 if (tp->t_flags & TF_SONOTCONN) { 2438 /* 2439 * Usually SYN_RECEIVED had been created from a LISTEN, 2440 * and solisten_enqueue() has already marked the socket 2441 * layer as connected. If it didn't, which can happen 2442 * only with an accept_filter(9), then the tp is marked 2443 * with TF_SONOTCONN. The other reason for this mark 2444 * to be set is a simultaneous open, a SYN_RECEIVED 2445 * that had been created from SYN_SENT. 2446 */ 2447 tp->t_flags &= ~TF_SONOTCONN; 2448 soisconnected(so); 2449 } 2450 /* Do window scaling? */ 2451 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2452 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2453 tp->rcv_scale = tp->request_r_scale; 2454 } 2455 tp->snd_wnd = tiwin; 2456 /* 2457 * Make transitions: 2458 * SYN-RECEIVED -> ESTABLISHED 2459 * SYN-RECEIVED* -> FIN-WAIT-1 2460 */ 2461 tp->t_starttime = ticks; 2462 if ((tp->t_flags & TF_FASTOPEN) && tp->t_tfo_pending) { 2463 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2464 tp->t_tfo_pending = NULL; 2465 } 2466 if (tp->t_flags & TF_NEEDFIN) { 2467 tp->t_acktime = ticks; 2468 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2469 tp->t_flags &= ~TF_NEEDFIN; 2470 } else { 2471 tcp_state_change(tp, TCPS_ESTABLISHED); 2472 TCP_PROBE5(accept__established, NULL, tp, 2473 m, tp, th); 2474 /* 2475 * TFO connections call cc_conn_init() during SYN 2476 * processing. Calling it again here for such 2477 * connections is not harmless as it would undo the 2478 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2479 * is retransmitted. 2480 */ 2481 if (!(tp->t_flags & TF_FASTOPEN)) 2482 cc_conn_init(tp); 2483 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2484 } 2485 /* 2486 * Account for the ACK of our SYN prior to 2487 * regular ACK processing below, except for 2488 * simultaneous SYN, which is handled later. 2489 */ 2490 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2491 incforsyn = 1; 2492 /* 2493 * If segment contains data or ACK, will call tcp_reass() 2494 * later; if not, do so now to pass queued data to user. 2495 */ 2496 if (tlen == 0 && (thflags & TH_FIN) == 0) { 2497 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2498 (struct mbuf *)0); 2499 tcp_handle_wakeup(tp); 2500 } 2501 tp->snd_wl1 = th->th_seq - 1; 2502 /* FALLTHROUGH */ 2503 2504 /* 2505 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2506 * ACKs. If the ack is in the range 2507 * tp->snd_una < th->th_ack <= tp->snd_max 2508 * then advance tp->snd_una to th->th_ack and drop 2509 * data from the retransmission queue. If this ACK reflects 2510 * more up to date window information we update our window information. 2511 */ 2512 case TCPS_ESTABLISHED: 2513 case TCPS_FIN_WAIT_1: 2514 case TCPS_FIN_WAIT_2: 2515 case TCPS_CLOSE_WAIT: 2516 case TCPS_CLOSING: 2517 case TCPS_LAST_ACK: 2518 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2519 TCPSTAT_INC(tcps_rcvacktoomuch); 2520 goto dropafterack; 2521 } 2522 if (tcp_is_sack_recovery(tp, &to)) { 2523 prev_sacked_bytes = tp->sackhint.sacked_bytes; 2524 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2525 if ((sack_changed != SACK_NOCHANGE) && 2526 (tp->t_flags & TF_LRD)) { 2527 tcp_sack_lost_retransmission(tp, th); 2528 } 2529 } else 2530 /* 2531 * Reset the value so that previous (valid) value 2532 * from the last ack with SACK doesn't get used. 2533 */ 2534 tp->sackhint.sacked_bytes = 0; 2535 2536 #ifdef TCP_HHOOK 2537 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2538 hhook_run_tcp_est_in(tp, th, &to); 2539 #endif 2540 2541 if (SEQ_LT(th->th_ack, tp->snd_una)) { 2542 /* This is old ACK information, don't process it. */ 2543 break; 2544 } 2545 if (th->th_ack == tp->snd_una) { 2546 /* Check if this is a duplicate ACK. */ 2547 if (tp->t_flags & TF_SACK_PERMIT) { 2548 /* 2549 * If SEG.ACK == SND.UNA, RFC 6675 requires a 2550 * duplicate ACK to selectively acknowledge 2551 * at least one byte, which was not selectively 2552 * acknowledged before. 2553 */ 2554 if (sack_changed == SACK_NOCHANGE) { 2555 break; 2556 } 2557 } else { 2558 /* 2559 * If SEG.ACK == SND.UNA, RFC 5681 requires a 2560 * duplicate ACK to have no data on it and to 2561 * not be a window update. 2562 */ 2563 if (!no_data || tiwin != tp->snd_wnd) { 2564 break; 2565 } 2566 } 2567 /* 2568 * If this is the first time we've seen a 2569 * FIN from the remote, this is not a 2570 * duplicate ACK and it needs to be processed 2571 * normally. 2572 * This happens during a simultaneous close. 2573 */ 2574 if ((thflags & TH_FIN) && 2575 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2576 tp->t_dupacks = 0; 2577 break; 2578 } 2579 /* Perform duplicate ACK processing. */ 2580 TCPSTAT_INC(tcps_rcvdupack); 2581 maxseg = tcp_maxseg(tp); 2582 if (!tcp_timer_active(tp, TT_REXMT)) { 2583 tp->t_dupacks = 0; 2584 } else if (++tp->t_dupacks > tcprexmtthresh || 2585 IN_FASTRECOVERY(tp->t_flags)) { 2586 cc_ack_received(tp, th, nsegs, CC_DUPACK); 2587 if (V_tcp_do_prr && 2588 IN_FASTRECOVERY(tp->t_flags) && 2589 (tp->t_flags & TF_SACK_PERMIT)) { 2590 tcp_do_prr_ack(tp, th, &to, 2591 sack_changed, &maxseg); 2592 } else if (tcp_is_sack_recovery(tp, &to) && 2593 IN_FASTRECOVERY(tp->t_flags) && 2594 (tp->snd_nxt == tp->snd_max)) { 2595 int awnd; 2596 2597 /* 2598 * Compute the amount of data in flight first. 2599 * We can inject new data into the pipe iff 2600 * we have less than ssthresh 2601 * worth of data in flight. 2602 */ 2603 awnd = tcp_compute_pipe(tp); 2604 if (awnd < tp->snd_ssthresh) { 2605 tp->snd_cwnd += imax(maxseg, 2606 imin(2 * maxseg, 2607 tp->sackhint.delivered_data)); 2608 if (tp->snd_cwnd > tp->snd_ssthresh) 2609 tp->snd_cwnd = tp->snd_ssthresh; 2610 } 2611 } else if (tcp_is_sack_recovery(tp, &to) && 2612 IN_FASTRECOVERY(tp->t_flags) && 2613 SEQ_LT(tp->snd_nxt, tp->snd_max)) { 2614 tp->snd_cwnd += imax(maxseg, 2615 imin(2 * maxseg, 2616 tp->sackhint.delivered_data)); 2617 } else { 2618 tp->snd_cwnd += maxseg; 2619 } 2620 (void) tcp_output(tp); 2621 goto drop; 2622 } else if (tp->t_dupacks == tcprexmtthresh || 2623 (tp->t_flags & TF_SACK_PERMIT && 2624 tp->sackhint.sacked_bytes > 2625 (tcprexmtthresh - 1) * maxseg)) { 2626 enter_recovery: 2627 /* 2628 * Above is the RFC6675 trigger condition of 2629 * more than (dupthresh-1)*maxseg sacked data. 2630 * If the count of holes in the 2631 * scoreboard is >= dupthresh, we could 2632 * also enter loss recovery, but don't 2633 * have that value readily available. 2634 */ 2635 tp->t_dupacks = tcprexmtthresh; 2636 tcp_seq onxt = tp->snd_nxt; 2637 2638 /* 2639 * If we're doing sack, check to 2640 * see if we're already in sack 2641 * recovery. If we're not doing sack, 2642 * check to see if we're in newreno 2643 * recovery. 2644 */ 2645 if (tcp_is_sack_recovery(tp, &to)) { 2646 if (IN_FASTRECOVERY(tp->t_flags)) { 2647 tp->t_dupacks = 0; 2648 break; 2649 } 2650 } else { 2651 if (SEQ_LEQ(th->th_ack, 2652 tp->snd_recover)) { 2653 tp->t_dupacks = 0; 2654 break; 2655 } 2656 } 2657 /* Congestion signal before ack. */ 2658 cc_cong_signal(tp, th, CC_NDUPACK); 2659 cc_ack_received(tp, th, nsegs, CC_DUPACK); 2660 tcp_timer_activate(tp, TT_REXMT, 0); 2661 tp->t_rtttime = 0; 2662 if (V_tcp_do_prr) { 2663 /* 2664 * snd_ssthresh and snd_recover are 2665 * already updated by cc_cong_signal. 2666 */ 2667 if (tcp_is_sack_recovery(tp, &to)) { 2668 /* 2669 * Include Limited Transmit 2670 * segments here 2671 */ 2672 tp->sackhint.prr_delivered = 2673 imin(tp->snd_max - th->th_ack, 2674 (tp->snd_limited + 1) * maxseg); 2675 tp->sackhint.recover_fs = imax(1, 2676 (tp->snd_nxt - tp->snd_una) - prev_sacked_bytes 2677 + tp->sackhint.delivered_data); 2678 } else { 2679 tp->sackhint.prr_delivered = 2680 maxseg; 2681 tp->sackhint.recover_fs = max(1, 2682 tp->snd_nxt - tp->snd_una); 2683 } 2684 } 2685 tp->snd_limited = 0; 2686 if (tcp_is_sack_recovery(tp, &to)) { 2687 TCPSTAT_INC(tcps_sack_recovery_episode); 2688 /* 2689 * When entering LR after RTO due to 2690 * Duplicate ACKs, retransmit existing 2691 * holes from the scoreboard. 2692 */ 2693 tcp_resend_sackholes(tp); 2694 /* Avoid inflating cwnd in tcp_output */ 2695 tp->snd_nxt = tp->snd_max; 2696 tp->snd_cwnd = tcp_compute_pipe(tp) + 2697 maxseg; 2698 (void) tcp_output(tp); 2699 /* Set cwnd to the expected flightsize */ 2700 tp->snd_cwnd = tp->snd_ssthresh; 2701 goto drop; 2702 } 2703 tp->snd_nxt = th->th_ack; 2704 tp->snd_cwnd = maxseg; 2705 (void) tcp_output(tp); 2706 KASSERT(tp->snd_limited <= 2, 2707 ("%s: tp->snd_limited too big", 2708 __func__)); 2709 tp->snd_cwnd = tp->snd_ssthresh + 2710 maxseg * 2711 (tp->t_dupacks - tp->snd_limited); 2712 if (SEQ_GT(onxt, tp->snd_nxt)) 2713 tp->snd_nxt = onxt; 2714 goto drop; 2715 } else if (V_tcp_do_rfc3042) { 2716 /* 2717 * Process first and second duplicate 2718 * ACKs. Each indicates a segment 2719 * leaving the network, creating room 2720 * for more. Make sure we can send a 2721 * packet on reception of each duplicate 2722 * ACK by increasing snd_cwnd by one 2723 * segment. Restore the original 2724 * snd_cwnd after packet transmission. 2725 */ 2726 cc_ack_received(tp, th, nsegs, CC_DUPACK); 2727 uint32_t oldcwnd = tp->snd_cwnd; 2728 tcp_seq oldsndmax = tp->snd_max; 2729 u_int sent; 2730 int avail; 2731 2732 KASSERT(tp->t_dupacks == 1 || 2733 tp->t_dupacks == 2, 2734 ("%s: dupacks not 1 or 2", 2735 __func__)); 2736 if (tp->t_dupacks == 1) 2737 tp->snd_limited = 0; 2738 if ((tp->snd_nxt == tp->snd_max) && 2739 (tp->t_rxtshift == 0)) 2740 tp->snd_cwnd = 2741 SEQ_SUB(tp->snd_nxt, tp->snd_una); 2742 tp->snd_cwnd += 2743 (tp->t_dupacks - tp->snd_limited) * maxseg; 2744 tp->snd_cwnd -= tcp_sack_adjust(tp); 2745 /* 2746 * Only call tcp_output when there 2747 * is new data available to be sent 2748 * or we need to send an ACK. 2749 */ 2750 SOCK_SENDBUF_LOCK(so); 2751 avail = sbavail(&so->so_snd); 2752 SOCK_SENDBUF_UNLOCK(so); 2753 if (tp->t_flags & TF_ACKNOW || 2754 (avail >= 2755 SEQ_SUB(tp->snd_nxt, tp->snd_una))) { 2756 (void) tcp_output(tp); 2757 } 2758 sent = SEQ_SUB(tp->snd_max, oldsndmax); 2759 if (sent > maxseg) { 2760 KASSERT((tp->t_dupacks == 2 && 2761 tp->snd_limited == 0) || 2762 (sent == maxseg + 1 && 2763 tp->t_flags & TF_SENTFIN) || 2764 (sent < 2 * maxseg && 2765 tp->t_flags & TF_NODELAY), 2766 ("%s: sent too much: %u>%u", 2767 __func__, sent, maxseg)); 2768 tp->snd_limited = 2; 2769 } else if (sent > 0) { 2770 ++tp->snd_limited; 2771 } 2772 tp->snd_cwnd = oldcwnd; 2773 goto drop; 2774 } 2775 break; 2776 } 2777 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2778 ("%s: SEQ_LEQ(th_ack, snd_una)", __func__)); 2779 /* 2780 * This ack is advancing the left edge, reset the 2781 * counter. 2782 */ 2783 tp->t_dupacks = 0; 2784 /* 2785 * If this ack also has new SACK info, increment the 2786 * t_dupacks as per RFC 6675. The variable 2787 * sack_changed tracks all changes to the SACK 2788 * scoreboard, including when partial ACKs without 2789 * SACK options are received, and clear the scoreboard 2790 * from the left side. Such partial ACKs should not be 2791 * counted as dupacks here. 2792 */ 2793 if (tcp_is_sack_recovery(tp, &to) && 2794 (((tp->t_rxtshift == 0) && (sack_changed != SACK_NOCHANGE)) || 2795 ((tp->t_rxtshift > 0) && (sack_changed == SACK_NEWLOSS))) && 2796 (tp->snd_nxt == tp->snd_max)) { 2797 tp->t_dupacks++; 2798 /* limit overhead by setting maxseg last */ 2799 if (!IN_FASTRECOVERY(tp->t_flags) && 2800 (tp->sackhint.sacked_bytes > 2801 (tcprexmtthresh - 1) * (maxseg = tcp_maxseg(tp)))) { 2802 goto enter_recovery; 2803 } 2804 } 2805 /* 2806 * If the congestion window was inflated to account 2807 * for the other side's cached packets, retract it. 2808 */ 2809 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2810 if (IN_FASTRECOVERY(tp->t_flags)) { 2811 if (tp->t_flags & TF_SACK_PERMIT) { 2812 if (V_tcp_do_prr && 2813 (to.to_flags & TOF_SACK)) { 2814 tcp_timer_activate(tp, 2815 TT_REXMT, 0); 2816 tp->t_rtttime = 0; 2817 tcp_do_prr_ack(tp, th, &to, 2818 sack_changed, &maxseg); 2819 tp->t_flags |= TF_ACKNOW; 2820 (void) tcp_output(tp); 2821 } else { 2822 tcp_sack_partialack(tp, th, 2823 &maxseg); 2824 } 2825 } else { 2826 tcp_newreno_partial_ack(tp, th); 2827 } 2828 } else if (IN_CONGRECOVERY(tp->t_flags) && 2829 (V_tcp_do_prr)) { 2830 tp->sackhint.delivered_data = 2831 BYTES_THIS_ACK(tp, th); 2832 tp->snd_fack = th->th_ack; 2833 /* 2834 * During ECN cwnd reduction 2835 * always use PRR-SSRB 2836 */ 2837 tcp_do_prr_ack(tp, th, &to, SACK_CHANGE, 2838 &maxseg); 2839 (void) tcp_output(tp); 2840 } 2841 } 2842 /* 2843 * If we reach this point, ACK is not a duplicate, 2844 * i.e., it ACKs something we sent. 2845 */ 2846 if (tp->t_flags & TF_NEEDSYN) { 2847 /* 2848 * T/TCP: Connection was half-synchronized, and our 2849 * SYN has been ACK'd (so connection is now fully 2850 * synchronized). Go to non-starred state, 2851 * increment snd_una for ACK of SYN, and check if 2852 * we can do window scaling. 2853 */ 2854 tp->t_flags &= ~TF_NEEDSYN; 2855 tp->snd_una++; 2856 /* Do window scaling? */ 2857 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2858 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2859 tp->rcv_scale = tp->request_r_scale; 2860 /* Send window already scaled. */ 2861 } 2862 } 2863 2864 process_ACK: 2865 INP_WLOCK_ASSERT(inp); 2866 2867 /* 2868 * Adjust for the SYN bit in sequence space, 2869 * but don't account for it in cwnd calculations. 2870 * This is for the SYN_RECEIVED, non-simultaneous 2871 * SYN case. SYN_SENT and simultaneous SYN are 2872 * treated elsewhere. 2873 */ 2874 if (incforsyn) 2875 tp->snd_una++; 2876 acked = BYTES_THIS_ACK(tp, th); 2877 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2878 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2879 tp->snd_una, th->th_ack, tp, m)); 2880 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2881 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2882 2883 /* 2884 * If we just performed our first retransmit, and the ACK 2885 * arrives within our recovery window, then it was a mistake 2886 * to do the retransmit in the first place. Recover our 2887 * original cwnd and ssthresh, and proceed to transmit where 2888 * we left off. 2889 */ 2890 if (tp->t_rxtshift == 1 && 2891 tp->t_flags & TF_PREVVALID && 2892 tp->t_badrxtwin != 0 && 2893 to.to_flags & TOF_TS && 2894 to.to_tsecr != 0 && 2895 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 2896 cc_cong_signal(tp, th, CC_RTO_ERR); 2897 2898 /* 2899 * If we have a timestamp reply, update smoothed 2900 * round trip time. If no timestamp is present but 2901 * transmit timer is running and timed sequence 2902 * number was acked, update smoothed round trip time. 2903 * Since we now have an rtt measurement, cancel the 2904 * timer backoff (cf., Phil Karn's retransmit alg.). 2905 * Recompute the initial retransmit timer. 2906 * 2907 * Some boxes send broken timestamp replies 2908 * during the SYN+ACK phase, ignore 2909 * timestamps of 0 or we could calculate a 2910 * huge RTT and blow up the retransmit timer. 2911 */ 2912 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2913 uint32_t t; 2914 2915 t = tcp_ts_getticks() - to.to_tsecr; 2916 if (!tp->t_rttlow || tp->t_rttlow > t) 2917 tp->t_rttlow = t; 2918 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2919 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2920 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2921 tp->t_rttlow = ticks - tp->t_rtttime; 2922 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2923 } 2924 2925 SOCK_SENDBUF_LOCK(so); 2926 /* 2927 * Clear t_acktime if remote side has ACKd all data in the 2928 * socket buffer and FIN (if applicable). 2929 * Otherwise, update t_acktime if we received a sufficiently 2930 * large ACK. 2931 */ 2932 if ((tp->t_state <= TCPS_CLOSE_WAIT && 2933 acked == sbavail(&so->so_snd)) || 2934 acked > sbavail(&so->so_snd)) 2935 tp->t_acktime = 0; 2936 else if (acked > 1) 2937 tp->t_acktime = ticks; 2938 2939 /* 2940 * If all outstanding data is acked, stop retransmit 2941 * timer and remember to restart (more output or persist). 2942 * If there is more data to be acked, restart retransmit 2943 * timer, using current (possibly backed-off) value. 2944 */ 2945 if (th->th_ack == tp->snd_max) { 2946 tcp_timer_activate(tp, TT_REXMT, 0); 2947 needoutput = 1; 2948 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2949 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); 2950 2951 /* 2952 * If no data (only SYN) was ACK'd, 2953 * skip rest of ACK processing. 2954 */ 2955 if (acked == 0) { 2956 SOCK_SENDBUF_UNLOCK(so); 2957 goto step6; 2958 } 2959 2960 /* 2961 * Let the congestion control algorithm update congestion 2962 * control related information. This typically means increasing 2963 * the congestion window. 2964 */ 2965 cc_ack_received(tp, th, nsegs, CC_ACK); 2966 2967 if (acked > sbavail(&so->so_snd)) { 2968 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2969 tp->snd_wnd -= sbavail(&so->so_snd); 2970 else 2971 tp->snd_wnd = 0; 2972 mfree = sbcut_locked(&so->so_snd, 2973 (int)sbavail(&so->so_snd)); 2974 ourfinisacked = 1; 2975 } else { 2976 mfree = sbcut_locked(&so->so_snd, acked); 2977 if (tp->snd_wnd >= (uint32_t) acked) 2978 tp->snd_wnd -= acked; 2979 else 2980 tp->snd_wnd = 0; 2981 ourfinisacked = 0; 2982 } 2983 /* NB: sowwakeup_locked() does an implicit unlock. */ 2984 sowwakeup_locked(so); 2985 m_freem(mfree); 2986 /* Detect una wraparound. */ 2987 if (!IN_RECOVERY(tp->t_flags) && 2988 SEQ_GT(tp->snd_una, tp->snd_recover) && 2989 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2990 tp->snd_recover = th->th_ack - 1; 2991 tp->snd_una = th->th_ack; 2992 if (IN_RECOVERY(tp->t_flags) && 2993 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2994 cc_post_recovery(tp, th); 2995 } 2996 if (SEQ_GT(tp->snd_una, tp->snd_recover)) { 2997 tp->snd_recover = tp->snd_una; 2998 } 2999 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 3000 tp->snd_nxt = tp->snd_una; 3001 3002 switch (tp->t_state) { 3003 /* 3004 * In FIN_WAIT_1 STATE in addition to the processing 3005 * for the ESTABLISHED state if our FIN is now acknowledged 3006 * then enter FIN_WAIT_2. 3007 */ 3008 case TCPS_FIN_WAIT_1: 3009 if (ourfinisacked) { 3010 /* 3011 * If we can't receive any more 3012 * data, then closing user can proceed. 3013 * Starting the timer is contrary to the 3014 * specification, but if we don't get a FIN 3015 * we'll hang forever. 3016 */ 3017 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3018 tcp_free_sackholes(tp); 3019 soisdisconnected(so); 3020 tcp_timer_activate(tp, TT_2MSL, 3021 (tcp_fast_finwait2_recycle ? 3022 tcp_finwait2_timeout : 3023 TP_MAXIDLE(tp))); 3024 } 3025 tcp_state_change(tp, TCPS_FIN_WAIT_2); 3026 } 3027 break; 3028 3029 /* 3030 * In CLOSING STATE in addition to the processing for 3031 * the ESTABLISHED state if the ACK acknowledges our FIN 3032 * then enter the TIME-WAIT state, otherwise ignore 3033 * the segment. 3034 */ 3035 case TCPS_CLOSING: 3036 if (ourfinisacked) { 3037 tcp_twstart(tp); 3038 m_freem(m); 3039 return; 3040 } 3041 break; 3042 3043 /* 3044 * In LAST_ACK, we may still be waiting for data to drain 3045 * and/or to be acked, as well as for the ack of our FIN. 3046 * If our FIN is now acknowledged, delete the TCB, 3047 * enter the closed state and return. 3048 */ 3049 case TCPS_LAST_ACK: 3050 if (ourfinisacked) { 3051 tp = tcp_close(tp); 3052 goto drop; 3053 } 3054 break; 3055 } 3056 } 3057 3058 step6: 3059 INP_WLOCK_ASSERT(inp); 3060 3061 /* 3062 * Update window information. 3063 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3064 */ 3065 if ((thflags & TH_ACK) && 3066 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3067 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3068 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3069 /* keep track of pure window updates */ 3070 if (no_data && tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3071 TCPSTAT_INC(tcps_rcvwinupd); 3072 tp->snd_wnd = tiwin; 3073 tp->snd_wl1 = th->th_seq; 3074 tp->snd_wl2 = th->th_ack; 3075 if (tp->snd_wnd > tp->max_sndwnd) 3076 tp->max_sndwnd = tp->snd_wnd; 3077 needoutput = 1; 3078 } 3079 3080 /* 3081 * Process segments with URG. 3082 */ 3083 if ((thflags & TH_URG) && th->th_urp && 3084 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3085 /* 3086 * This is a kludge, but if we receive and accept 3087 * random urgent pointers, we'll crash in 3088 * soreceive. It's hard to imagine someone 3089 * actually wanting to send this much urgent data. 3090 */ 3091 SOCK_RECVBUF_LOCK(so); 3092 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3093 th->th_urp = 0; /* XXX */ 3094 thflags &= ~TH_URG; /* XXX */ 3095 SOCK_RECVBUF_UNLOCK(so); /* XXX */ 3096 goto dodata; /* XXX */ 3097 } 3098 /* 3099 * If this segment advances the known urgent pointer, 3100 * then mark the data stream. This should not happen 3101 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3102 * a FIN has been received from the remote side. 3103 * In these states we ignore the URG. 3104 * 3105 * According to RFC961 (Assigned Protocols), 3106 * the urgent pointer points to the last octet 3107 * of urgent data. We continue, however, 3108 * to consider it to indicate the first octet 3109 * of data past the urgent section as the original 3110 * spec states (in one of two places). 3111 */ 3112 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3113 tp->rcv_up = th->th_seq + th->th_urp; 3114 so->so_oobmark = sbavail(&so->so_rcv) + 3115 (tp->rcv_up - tp->rcv_nxt) - 1; 3116 if (so->so_oobmark == 0) 3117 so->so_rcv.sb_state |= SBS_RCVATMARK; 3118 sohasoutofband(so); 3119 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3120 } 3121 SOCK_RECVBUF_UNLOCK(so); 3122 /* 3123 * Remove out of band data so doesn't get presented to user. 3124 * This can happen independent of advancing the URG pointer, 3125 * but if two URG's are pending at once, some out-of-band 3126 * data may creep in... ick. 3127 */ 3128 if (th->th_urp <= (uint32_t)tlen && 3129 !(so->so_options & SO_OOBINLINE)) { 3130 /* hdr drop is delayed */ 3131 tcp_pulloutofband(so, th, m, drop_hdrlen); 3132 } 3133 } else { 3134 /* 3135 * If no out of band data is expected, 3136 * pull receive urgent pointer along 3137 * with the receive window. 3138 */ 3139 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3140 tp->rcv_up = tp->rcv_nxt; 3141 } 3142 dodata: /* XXX */ 3143 INP_WLOCK_ASSERT(inp); 3144 3145 /* 3146 * Process the segment text, merging it into the TCP sequencing queue, 3147 * and arranging for acknowledgment of receipt if necessary. 3148 * This process logically involves adjusting tp->rcv_wnd as data 3149 * is presented to the user (this happens in tcp_usrreq.c, 3150 * case PRU_RCVD). If a FIN has already been received on this 3151 * connection then we just ignore the text. 3152 */ 3153 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3154 (tp->t_flags & TF_FASTOPEN)); 3155 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3156 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3157 tcp_seq save_start = th->th_seq; 3158 tcp_seq save_rnxt = tp->rcv_nxt; 3159 int save_tlen = tlen; 3160 m_adj(m, drop_hdrlen); /* delayed header drop */ 3161 /* 3162 * Insert segment which includes th into TCP reassembly queue 3163 * with control block tp. Set thflags to whether reassembly now 3164 * includes a segment with FIN. This handles the common case 3165 * inline (segment is the next to be received on an established 3166 * connection, and the queue is empty), avoiding linkage into 3167 * and removal from the queue and repetition of various 3168 * conversions. 3169 * Set DELACK for segments received in order, but ack 3170 * immediately when segments are out of order (so 3171 * fast retransmit can work). 3172 */ 3173 if (th->th_seq == tp->rcv_nxt && 3174 SEGQ_EMPTY(tp) && 3175 (TCPS_HAVEESTABLISHED(tp->t_state) || 3176 tfo_syn)) { 3177 if (DELAY_ACK(tp, tlen) || tfo_syn) 3178 tp->t_flags |= TF_DELACK; 3179 else 3180 tp->t_flags |= TF_ACKNOW; 3181 tp->rcv_nxt += tlen; 3182 if (tlen && 3183 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3184 (tp->t_fbyte_in == 0)) { 3185 tp->t_fbyte_in = ticks; 3186 if (tp->t_fbyte_in == 0) 3187 tp->t_fbyte_in = 1; 3188 if (tp->t_fbyte_out && tp->t_fbyte_in) 3189 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3190 } 3191 thflags = tcp_get_flags(th) & TH_FIN; 3192 TCPSTAT_INC(tcps_rcvpack); 3193 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3194 SOCK_RECVBUF_LOCK(so); 3195 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3196 m_freem(m); 3197 else 3198 sbappendstream_locked(&so->so_rcv, m, 0); 3199 tp->t_flags |= TF_WAKESOR; 3200 } else { 3201 /* 3202 * XXX: Due to the header drop above "th" is 3203 * theoretically invalid by now. Fortunately 3204 * m_adj() doesn't actually frees any mbufs 3205 * when trimming from the head. 3206 */ 3207 tcp_seq temp = save_start; 3208 3209 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3210 tp->t_flags |= TF_ACKNOW; 3211 } 3212 if ((tp->t_flags & TF_SACK_PERMIT) && 3213 (save_tlen > 0) && 3214 TCPS_HAVEESTABLISHED(tp->t_state)) { 3215 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3216 /* 3217 * DSACK actually handled in the fastpath 3218 * above. 3219 */ 3220 tcp_update_sack_list(tp, save_start, 3221 save_start + save_tlen); 3222 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3223 if ((tp->rcv_numsacks >= 1) && 3224 (tp->sackblks[0].end == save_start)) { 3225 /* 3226 * Partial overlap, recorded at todrop 3227 * above. 3228 */ 3229 tcp_update_sack_list(tp, 3230 tp->sackblks[0].start, 3231 tp->sackblks[0].end); 3232 } else { 3233 tcp_update_dsack_list(tp, save_start, 3234 save_start + save_tlen); 3235 } 3236 } else if (tlen >= save_tlen) { 3237 /* Update of sackblks. */ 3238 tcp_update_dsack_list(tp, save_start, 3239 save_start + save_tlen); 3240 } else if (tlen > 0) { 3241 tcp_update_dsack_list(tp, save_start, 3242 save_start + tlen); 3243 } 3244 } 3245 tcp_handle_wakeup(tp); 3246 #if 0 3247 /* 3248 * Note the amount of data that peer has sent into 3249 * our window, in order to estimate the sender's 3250 * buffer size. 3251 * XXX: Unused. 3252 */ 3253 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3254 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3255 else 3256 len = so->so_rcv.sb_hiwat; 3257 #endif 3258 } else { 3259 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 3260 if (tlen > 0) { 3261 if ((thflags & TH_FIN) != 0) { 3262 log(LOG_DEBUG, "%s; %s: %s: " 3263 "Received %d bytes of data and FIN " 3264 "after having received a FIN, " 3265 "just dropping both\n", 3266 s, __func__, 3267 tcpstates[tp->t_state], tlen); 3268 } else { 3269 log(LOG_DEBUG, "%s; %s: %s: " 3270 "Received %d bytes of data " 3271 "after having received a FIN, " 3272 "just dropping it\n", 3273 s, __func__, 3274 tcpstates[tp->t_state], tlen); 3275 } 3276 } else { 3277 if ((thflags & TH_FIN) != 0) { 3278 log(LOG_DEBUG, "%s; %s: %s: " 3279 "Received FIN " 3280 "after having received a FIN, " 3281 "just dropping it\n", 3282 s, __func__, 3283 tcpstates[tp->t_state]); 3284 } 3285 } 3286 free(s, M_TCPLOG); 3287 } 3288 m_freem(m); 3289 thflags &= ~TH_FIN; 3290 } 3291 3292 /* 3293 * If FIN is received ACK the FIN and let the user know 3294 * that the connection is closing. 3295 */ 3296 if (thflags & TH_FIN) { 3297 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3298 /* The socket upcall is handled by socantrcvmore. */ 3299 socantrcvmore(so); 3300 /* 3301 * If connection is half-synchronized 3302 * (ie NEEDSYN flag on) then delay ACK, 3303 * so it may be piggybacked when SYN is sent. 3304 * Otherwise, since we received a FIN then no 3305 * more input can be expected, send ACK now. 3306 */ 3307 if (tp->t_flags & TF_NEEDSYN) 3308 tp->t_flags |= TF_DELACK; 3309 else 3310 tp->t_flags |= TF_ACKNOW; 3311 tp->rcv_nxt++; 3312 } 3313 switch (tp->t_state) { 3314 /* 3315 * In SYN_RECEIVED and ESTABLISHED STATES 3316 * enter the CLOSE_WAIT state. 3317 */ 3318 case TCPS_SYN_RECEIVED: 3319 tp->t_starttime = ticks; 3320 /* FALLTHROUGH */ 3321 case TCPS_ESTABLISHED: 3322 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3323 break; 3324 3325 /* 3326 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3327 * enter the CLOSING state. 3328 */ 3329 case TCPS_FIN_WAIT_1: 3330 tcp_state_change(tp, TCPS_CLOSING); 3331 break; 3332 3333 /* 3334 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3335 * starting the time-wait timer, turning off the other 3336 * standard timers. 3337 */ 3338 case TCPS_FIN_WAIT_2: 3339 tcp_twstart(tp); 3340 return; 3341 } 3342 } 3343 TCP_PROBE3(debug__input, tp, th, m); 3344 3345 /* 3346 * Return any desired output. 3347 */ 3348 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 3349 (void) tcp_output(tp); 3350 } 3351 check_delack: 3352 INP_WLOCK_ASSERT(inp); 3353 3354 if (tp->t_flags & TF_DELACK) { 3355 tp->t_flags &= ~TF_DELACK; 3356 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3357 } 3358 INP_WUNLOCK(inp); 3359 return; 3360 3361 dropafterack: 3362 /* 3363 * Generate an ACK dropping incoming segment if it occupies 3364 * sequence space, where the ACK reflects our state. 3365 * 3366 * We can now skip the test for the RST flag since all 3367 * paths to this code happen after packets containing 3368 * RST have been dropped. 3369 * 3370 * In the SYN-RECEIVED state, don't send an ACK unless the 3371 * segment we received passes the SYN-RECEIVED ACK test. 3372 * If it fails send a RST. This breaks the loop in the 3373 * "LAND" DoS attack, and also prevents an ACK storm 3374 * between two listening ports that have been sent forged 3375 * SYN segments, each with the source address of the other. 3376 */ 3377 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3378 (SEQ_GT(tp->snd_una, th->th_ack) || 3379 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3380 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 3381 goto dropwithreset; 3382 } 3383 TCP_PROBE3(debug__input, tp, th, m); 3384 tp->t_flags |= TF_ACKNOW; 3385 (void) tcp_output(tp); 3386 INP_WUNLOCK(inp); 3387 m_freem(m); 3388 return; 3389 3390 dropwithreset: 3391 tcp_dropwithreset(m, th, tp, tlen); 3392 if (tp != NULL) { 3393 INP_WUNLOCK(inp); 3394 } 3395 return; 3396 3397 drop: 3398 /* 3399 * Drop space held by incoming segment and return. 3400 */ 3401 TCP_PROBE3(debug__input, tp, th, m); 3402 if (tp != NULL) { 3403 INP_WUNLOCK(inp); 3404 } 3405 m_freem(m); 3406 } 3407 3408 /* 3409 * Issue RST and make ACK acceptable to originator of segment. 3410 * The mbuf must still include the original packet header. 3411 * tp may be NULL. 3412 */ 3413 void 3414 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, int tlen) 3415 { 3416 #ifdef INET 3417 struct ip *ip; 3418 #endif 3419 #ifdef INET6 3420 struct ip6_hdr *ip6; 3421 #endif 3422 3423 if (tp != NULL) { 3424 INP_LOCK_ASSERT(tptoinpcb(tp)); 3425 } 3426 3427 /* Don't bother if destination was broadcast/multicast. */ 3428 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3429 goto drop; 3430 #ifdef INET6 3431 if (mtod(m, struct ip *)->ip_v == 6) { 3432 ip6 = mtod(m, struct ip6_hdr *); 3433 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3434 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3435 goto drop; 3436 /* IPv6 anycast check is done at tcp6_input() */ 3437 } 3438 #endif 3439 #if defined(INET) && defined(INET6) 3440 else 3441 #endif 3442 #ifdef INET 3443 { 3444 ip = mtod(m, struct ip *); 3445 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3446 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3447 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3448 in_ifnet_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3449 goto drop; 3450 } 3451 #endif 3452 3453 /* Perform bandwidth limiting. */ 3454 if (badport_bandlim(BANDLIM_TCP_RST) < 0) 3455 goto drop; 3456 3457 /* tcp_respond consumes the mbuf chain. */ 3458 if (tcp_get_flags(th) & TH_ACK) { 3459 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3460 th->th_ack, TH_RST); 3461 } else { 3462 if (tcp_get_flags(th) & TH_SYN) 3463 tlen++; 3464 if (tcp_get_flags(th) & TH_FIN) 3465 tlen++; 3466 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3467 (tcp_seq)0, TH_RST|TH_ACK); 3468 } 3469 return; 3470 drop: 3471 m_freem(m); 3472 } 3473 3474 /* 3475 * Parse TCP options and place in tcpopt. 3476 */ 3477 void 3478 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3479 { 3480 int opt, optlen; 3481 3482 to->to_flags = 0; 3483 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3484 opt = cp[0]; 3485 if (opt == TCPOPT_EOL) 3486 break; 3487 if (opt == TCPOPT_NOP) 3488 optlen = 1; 3489 else { 3490 if (cnt < 2) 3491 break; 3492 optlen = cp[1]; 3493 if (optlen < 2 || optlen > cnt) 3494 break; 3495 } 3496 switch (opt) { 3497 case TCPOPT_MAXSEG: 3498 if (optlen != TCPOLEN_MAXSEG) 3499 continue; 3500 if (!(flags & TO_SYN)) 3501 continue; 3502 to->to_flags |= TOF_MSS; 3503 bcopy((char *)cp + 2, 3504 (char *)&to->to_mss, sizeof(to->to_mss)); 3505 to->to_mss = ntohs(to->to_mss); 3506 break; 3507 case TCPOPT_WINDOW: 3508 if (optlen != TCPOLEN_WINDOW) 3509 continue; 3510 if (!(flags & TO_SYN)) 3511 continue; 3512 to->to_flags |= TOF_SCALE; 3513 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3514 break; 3515 case TCPOPT_TIMESTAMP: 3516 if (optlen != TCPOLEN_TIMESTAMP) 3517 continue; 3518 to->to_flags |= TOF_TS; 3519 bcopy((char *)cp + 2, 3520 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3521 to->to_tsval = ntohl(to->to_tsval); 3522 bcopy((char *)cp + 6, 3523 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3524 to->to_tsecr = ntohl(to->to_tsecr); 3525 break; 3526 case TCPOPT_SIGNATURE: 3527 /* 3528 * In order to reply to a host which has set the 3529 * TCP_SIGNATURE option in its initial SYN, we have 3530 * to record the fact that the option was observed 3531 * here for the syncache code to perform the correct 3532 * response. 3533 */ 3534 if (optlen != TCPOLEN_SIGNATURE) 3535 continue; 3536 to->to_flags |= TOF_SIGNATURE; 3537 to->to_signature = cp + 2; 3538 break; 3539 case TCPOPT_SACK_PERMITTED: 3540 if (optlen != TCPOLEN_SACK_PERMITTED) 3541 continue; 3542 if (!(flags & TO_SYN)) 3543 continue; 3544 if (!V_tcp_do_sack) 3545 continue; 3546 to->to_flags |= TOF_SACKPERM; 3547 break; 3548 case TCPOPT_SACK: 3549 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3550 continue; 3551 if (flags & TO_SYN) 3552 continue; 3553 to->to_flags |= TOF_SACK; 3554 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3555 to->to_sacks = cp + 2; 3556 TCPSTAT_INC(tcps_sack_rcv_blocks); 3557 break; 3558 case TCPOPT_FAST_OPEN: 3559 /* 3560 * Cookie length validation is performed by the 3561 * server side cookie checking code or the client 3562 * side cookie cache update code. 3563 */ 3564 if (!(flags & TO_SYN)) 3565 continue; 3566 if (!V_tcp_fastopen_client_enable && 3567 !V_tcp_fastopen_server_enable) 3568 continue; 3569 to->to_flags |= TOF_FASTOPEN; 3570 to->to_tfo_len = optlen - 2; 3571 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3572 break; 3573 default: 3574 continue; 3575 } 3576 } 3577 } 3578 3579 /* 3580 * Pull out of band byte out of a segment so 3581 * it doesn't appear in the user's data queue. 3582 * It is still reflected in the segment length for 3583 * sequencing purposes. 3584 */ 3585 void 3586 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3587 int off) 3588 { 3589 int cnt = off + th->th_urp - 1; 3590 3591 while (cnt >= 0) { 3592 if (m->m_len > cnt) { 3593 char *cp = mtod(m, caddr_t) + cnt; 3594 struct tcpcb *tp = sototcpcb(so); 3595 3596 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3597 3598 tp->t_iobc = *cp; 3599 tp->t_oobflags |= TCPOOB_HAVEDATA; 3600 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3601 m->m_len--; 3602 if (m->m_flags & M_PKTHDR) 3603 m->m_pkthdr.len--; 3604 return; 3605 } 3606 cnt -= m->m_len; 3607 m = m->m_next; 3608 if (m == NULL) 3609 break; 3610 } 3611 panic("tcp_pulloutofband"); 3612 } 3613 3614 /* 3615 * Collect new round-trip time estimate 3616 * and update averages and current timeout. 3617 */ 3618 void 3619 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3620 { 3621 int delta; 3622 3623 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3624 3625 TCPSTAT_INC(tcps_rttupdated); 3626 if (tp->t_rttupdated < UCHAR_MAX) 3627 tp->t_rttupdated++; 3628 #ifdef STATS 3629 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3630 imax(0, rtt * 1000 / hz)); 3631 #endif 3632 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3633 /* 3634 * srtt is stored as fixed point with 5 bits after the 3635 * binary point (i.e., scaled by 8). The following magic 3636 * is equivalent to the smoothing algorithm in rfc793 with 3637 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3638 * point). Adjust rtt to origin 0. 3639 */ 3640 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3641 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3642 3643 if ((tp->t_srtt += delta) <= 0) 3644 tp->t_srtt = 1; 3645 3646 /* 3647 * We accumulate a smoothed rtt variance (actually, a 3648 * smoothed mean difference), then set the retransmit 3649 * timer to smoothed rtt + 4 times the smoothed variance. 3650 * rttvar is stored as fixed point with 4 bits after the 3651 * binary point (scaled by 16). The following is 3652 * equivalent to rfc793 smoothing with an alpha of .75 3653 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3654 * rfc793's wired-in beta. 3655 */ 3656 if (delta < 0) 3657 delta = -delta; 3658 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3659 if ((tp->t_rttvar += delta) <= 0) 3660 tp->t_rttvar = 1; 3661 } else { 3662 /* 3663 * No rtt measurement yet - use the unsmoothed rtt. 3664 * Set the variance to half the rtt (so our first 3665 * retransmit happens at 3*rtt). 3666 */ 3667 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3668 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3669 } 3670 tp->t_rtttime = 0; 3671 tp->t_rxtshift = 0; 3672 3673 /* 3674 * the retransmit should happen at rtt + 4 * rttvar. 3675 * Because of the way we do the smoothing, srtt and rttvar 3676 * will each average +1/2 tick of bias. When we compute 3677 * the retransmit timer, we want 1/2 tick of rounding and 3678 * 1 extra tick because of +-1/2 tick uncertainty in the 3679 * firing of the timer. The bias will give us exactly the 3680 * 1.5 tick we need. But, because the bias is 3681 * statistical, we have to test that we don't drop below 3682 * the minimum feasible timer (which is 2 ticks). 3683 */ 3684 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3685 max(tp->t_rttmin, rtt + 2), tcp_rexmit_max); 3686 3687 /* 3688 * We received an ack for a packet that wasn't retransmitted; 3689 * it is probably safe to discard any error indications we've 3690 * received recently. This isn't quite right, but close enough 3691 * for now (a route might have failed after we sent a segment, 3692 * and the return path might not be symmetrical). 3693 */ 3694 tp->t_softerror = 0; 3695 } 3696 3697 /* 3698 * Determine a reasonable value for maxseg size. 3699 * If the route is known, check route for mtu. 3700 * If none, use an mss that can be handled on the outgoing interface 3701 * without forcing IP to fragment. If no route is found, route has no mtu, 3702 * or the destination isn't local, use a default, hopefully conservative 3703 * size (usually 512 or the default IP max size, but no more than the mtu 3704 * of the interface), as we can't discover anything about intervening 3705 * gateways or networks. We also initialize the congestion/slow start 3706 * window to be a single segment if the destination isn't local. 3707 * While looking at the routing entry, we also initialize other path-dependent 3708 * parameters from pre-set or cached values in the routing entry. 3709 * 3710 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3711 * IP options, e.g. IPSEC data, since length of this data may vary, and 3712 * thus it is calculated for every segment separately in tcp_output(). 3713 * 3714 * NOTE that this routine is only called when we process an incoming 3715 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3716 * settings are handled in tcp_mssopt(). 3717 */ 3718 void 3719 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3720 struct tcp_hc_metrics *metricptr, struct tcp_ifcap *cap) 3721 { 3722 int mss = 0; 3723 uint32_t maxmtu = 0; 3724 struct inpcb *inp = tptoinpcb(tp); 3725 struct tcp_hc_metrics metrics; 3726 #ifdef INET6 3727 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3728 size_t min_protoh = isipv6 ? 3729 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3730 sizeof (struct tcpiphdr); 3731 #else 3732 size_t min_protoh = sizeof(struct tcpiphdr); 3733 #endif 3734 3735 INP_WLOCK_ASSERT(inp); 3736 3737 if (tp->t_port) 3738 min_protoh += V_tcp_udp_tunneling_overhead; 3739 if (mtuoffer != -1) { 3740 KASSERT(offer == -1, ("%s: conflict", __func__)); 3741 offer = mtuoffer - min_protoh; 3742 } 3743 3744 /* Initialize. */ 3745 #ifdef INET6 3746 if (isipv6) { 3747 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3748 tp->t_maxseg = V_tcp_v6mssdflt; 3749 } 3750 #endif 3751 #if defined(INET) && defined(INET6) 3752 else 3753 #endif 3754 #ifdef INET 3755 { 3756 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3757 tp->t_maxseg = V_tcp_mssdflt; 3758 } 3759 #endif 3760 3761 /* 3762 * No route to sender, stay with default mss and return. 3763 */ 3764 if (maxmtu == 0) { 3765 /* 3766 * In case we return early we need to initialize metrics 3767 * to a defined state as tcp_hc_get() would do for us 3768 * if there was no cache hit. 3769 */ 3770 if (metricptr != NULL) 3771 bzero(metricptr, sizeof(struct tcp_hc_metrics)); 3772 return; 3773 } 3774 3775 /* What have we got? */ 3776 switch (offer) { 3777 case 0: 3778 /* 3779 * Offer == 0 means that there was no MSS on the SYN 3780 * segment, in this case we use tcp_mssdflt as 3781 * already assigned to t_maxseg above. 3782 */ 3783 offer = tp->t_maxseg; 3784 break; 3785 3786 case -1: 3787 /* 3788 * Offer == -1 means that we didn't receive SYN yet. 3789 */ 3790 /* FALLTHROUGH */ 3791 3792 default: 3793 /* 3794 * Prevent DoS attack with too small MSS. Round up 3795 * to at least minmss. 3796 */ 3797 offer = max(offer, V_tcp_minmss); 3798 } 3799 3800 if (metricptr == NULL) 3801 metricptr = &metrics; 3802 tcp_hc_get(&inp->inp_inc, metricptr); 3803 3804 /* 3805 * If there's a discovered mtu in tcp hostcache, use it. 3806 * Else, use the link mtu. 3807 */ 3808 if (metricptr->hc_mtu) 3809 mss = min(metricptr->hc_mtu, maxmtu) - min_protoh; 3810 else { 3811 #ifdef INET6 3812 if (isipv6) { 3813 mss = maxmtu - min_protoh; 3814 if (!V_path_mtu_discovery && 3815 !in6_localaddr(&inp->in6p_faddr)) 3816 mss = min(mss, V_tcp_v6mssdflt); 3817 } 3818 #endif 3819 #if defined(INET) && defined(INET6) 3820 else 3821 #endif 3822 #ifdef INET 3823 { 3824 mss = maxmtu - min_protoh; 3825 if (!V_path_mtu_discovery && 3826 !in_localaddr(inp->inp_faddr)) 3827 mss = min(mss, V_tcp_mssdflt); 3828 } 3829 #endif 3830 /* 3831 * XXX - The above conditional (mss = maxmtu - min_protoh) 3832 * probably violates the TCP spec. 3833 * The problem is that, since we don't know the 3834 * other end's MSS, we are supposed to use a conservative 3835 * default. But, if we do that, then MTU discovery will 3836 * never actually take place, because the conservative 3837 * default is much less than the MTUs typically seen 3838 * on the Internet today. For the moment, we'll sweep 3839 * this under the carpet. 3840 * 3841 * The conservative default might not actually be a problem 3842 * if the only case this occurs is when sending an initial 3843 * SYN with options and data to a host we've never talked 3844 * to before. Then, they will reply with an MSS value which 3845 * will get recorded and the new parameters should get 3846 * recomputed. For Further Study. 3847 */ 3848 } 3849 mss = min(mss, offer); 3850 3851 /* 3852 * Sanity check: make sure that maxseg will be large 3853 * enough to allow some data on segments even if the 3854 * all the option space is used (40bytes). Otherwise 3855 * funny things may happen in tcp_output. 3856 * 3857 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3858 */ 3859 mss = max(mss, 64); 3860 3861 tp->t_maxseg = mss; 3862 if (tp->t_maxseg < V_tcp_mssdflt) { 3863 /* 3864 * The MSS is so small we should not process incoming 3865 * SACK's since we are subject to attack in such a 3866 * case. 3867 */ 3868 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 3869 } else { 3870 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 3871 } 3872 3873 } 3874 3875 void 3876 tcp_mss(struct tcpcb *tp, int offer) 3877 { 3878 int mss; 3879 uint32_t bufsize; 3880 struct inpcb *inp = tptoinpcb(tp); 3881 struct socket *so; 3882 struct tcp_hc_metrics metrics; 3883 struct tcp_ifcap cap; 3884 3885 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3886 3887 bzero(&cap, sizeof(cap)); 3888 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3889 3890 mss = tp->t_maxseg; 3891 3892 /* 3893 * If there's a pipesize, change the socket buffer to that size, 3894 * don't change if sb_hiwat is different than default (then it 3895 * has been changed on purpose with setsockopt). 3896 * Make the socket buffers an integral number of mss units; 3897 * if the mss is larger than the socket buffer, decrease the mss. 3898 */ 3899 so = inp->inp_socket; 3900 SOCK_SENDBUF_LOCK(so); 3901 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.hc_sendpipe) 3902 bufsize = metrics.hc_sendpipe; 3903 else 3904 bufsize = so->so_snd.sb_hiwat; 3905 if (bufsize < mss) 3906 mss = bufsize; 3907 else { 3908 bufsize = roundup(bufsize, mss); 3909 if (bufsize > sb_max) 3910 bufsize = sb_max; 3911 if (bufsize > so->so_snd.sb_hiwat) 3912 (void)sbreserve_locked(so, SO_SND, bufsize, NULL); 3913 } 3914 SOCK_SENDBUF_UNLOCK(so); 3915 /* 3916 * Sanity check: make sure that maxseg will be large 3917 * enough to allow some data on segments even if the 3918 * all the option space is used (40bytes). Otherwise 3919 * funny things may happen in tcp_output. 3920 * 3921 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3922 */ 3923 tp->t_maxseg = max(mss, 64); 3924 if (tp->t_maxseg < V_tcp_mssdflt) { 3925 /* 3926 * The MSS is so small we should not process incoming 3927 * SACK's since we are subject to attack in such a 3928 * case. 3929 */ 3930 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 3931 } else { 3932 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 3933 } 3934 3935 SOCK_RECVBUF_LOCK(so); 3936 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.hc_recvpipe) 3937 bufsize = metrics.hc_recvpipe; 3938 else 3939 bufsize = so->so_rcv.sb_hiwat; 3940 if (bufsize > mss) { 3941 bufsize = roundup(bufsize, mss); 3942 if (bufsize > sb_max) 3943 bufsize = sb_max; 3944 if (bufsize > so->so_rcv.sb_hiwat) 3945 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL); 3946 } 3947 SOCK_RECVBUF_UNLOCK(so); 3948 3949 /* Check the interface for TSO capabilities. */ 3950 if (cap.ifcap & CSUM_TSO) { 3951 tp->t_flags |= TF_TSO; 3952 tp->t_tsomax = cap.tsomax; 3953 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3954 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3955 if (cap.ipsec_tso) 3956 tp->t_flags2 |= TF2_IPSEC_TSO; 3957 } 3958 } 3959 3960 /* 3961 * Determine the MSS option to send on an outgoing SYN. 3962 */ 3963 int 3964 tcp_mssopt(struct in_conninfo *inc) 3965 { 3966 int mss = 0; 3967 uint32_t thcmtu = 0; 3968 uint32_t maxmtu = 0; 3969 size_t min_protoh; 3970 3971 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3972 3973 #ifdef INET6 3974 if (inc->inc_flags & INC_ISIPV6) { 3975 mss = V_tcp_v6mssdflt; 3976 maxmtu = tcp_maxmtu6(inc, NULL); 3977 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3978 } 3979 #endif 3980 #if defined(INET) && defined(INET6) 3981 else 3982 #endif 3983 #ifdef INET 3984 { 3985 mss = V_tcp_mssdflt; 3986 maxmtu = tcp_maxmtu(inc, NULL); 3987 min_protoh = sizeof(struct tcpiphdr); 3988 } 3989 #endif 3990 #if defined(INET6) || defined(INET) 3991 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3992 #endif 3993 3994 if (maxmtu && thcmtu) 3995 mss = min(maxmtu, thcmtu) - min_protoh; 3996 else if (maxmtu || thcmtu) 3997 mss = max(maxmtu, thcmtu) - min_protoh; 3998 3999 return (mss); 4000 } 4001 4002 void 4003 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, 4004 sackstatus_t sack_changed, u_int *maxsegp) 4005 { 4006 int snd_cnt = 0, del_data = 0, pipe = 0; 4007 u_int maxseg; 4008 bool safe_ack; 4009 4010 INP_WLOCK_ASSERT(tptoinpcb(tp)); 4011 4012 if (*maxsegp == 0) { 4013 *maxsegp = tcp_maxseg(tp); 4014 } 4015 maxseg = *maxsegp; 4016 /* 4017 * Compute the amount of data that this ACK is indicating 4018 * (del_data) and an estimate of how many bytes are in the 4019 * network. 4020 */ 4021 if (tcp_is_sack_recovery(tp, to) || 4022 (IN_CONGRECOVERY(tp->t_flags) && 4023 !IN_FASTRECOVERY(tp->t_flags))) { 4024 del_data = tp->sackhint.delivered_data; 4025 pipe = tcp_compute_pipe(tp); 4026 } else { 4027 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + 4028 tp->snd_recover - tp->snd_una)) { 4029 del_data = maxseg; 4030 } 4031 pipe = imax(0, tp->snd_max - tp->snd_una - 4032 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); 4033 } 4034 4035 if (del_data == 0) 4036 return; 4037 4038 tp->sackhint.prr_delivered += del_data; 4039 /* 4040 * Proportional Rate Reduction 4041 */ 4042 if (pipe >= tp->snd_ssthresh) { 4043 if (tp->sackhint.recover_fs == 0) 4044 tp->sackhint.recover_fs = 4045 imax(1, tp->snd_nxt - tp->snd_una); 4046 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 4047 tp->snd_ssthresh, tp->sackhint.recover_fs) - 4048 tp->sackhint.prr_out + maxseg - 1; 4049 } else { 4050 safe_ack = SEQ_GT(th->th_ack, tp->snd_una) && (sack_changed != SACK_NEWLOSS); 4051 snd_cnt = imax(tp->sackhint.prr_delivered - tp->sackhint.prr_out, del_data); 4052 if (safe_ack) { 4053 snd_cnt += maxseg; 4054 } 4055 snd_cnt = imin(tp->snd_ssthresh - pipe, snd_cnt); 4056 } 4057 4058 if (tp->sackhint.prr_out == 0 && snd_cnt == 0) { 4059 snd_cnt = maxseg; 4060 } 4061 snd_cnt = imax(snd_cnt, 0) / maxseg; 4062 /* 4063 * Send snd_cnt new data into the network in response to this ack. 4064 * If there is going to be a SACK retransmission, adjust snd_cwnd 4065 * accordingly. 4066 */ 4067 if (IN_FASTRECOVERY(tp->t_flags)) { 4068 if (tcp_is_sack_recovery(tp, to)) { 4069 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4070 } else { 4071 tp->snd_cwnd = (tp->snd_max - tp->snd_una) + 4072 (snd_cnt * maxseg); 4073 } 4074 } else if (IN_CONGRECOVERY(tp->t_flags)) { 4075 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4076 } 4077 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); 4078 } 4079 4080 /* 4081 * On a partial ack arrives, force the retransmission of the 4082 * next unacknowledged segment. Do not clear tp->t_dupacks. 4083 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4084 * be started again. 4085 */ 4086 void 4087 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4088 { 4089 tcp_seq onxt = tp->snd_nxt; 4090 uint32_t ocwnd = tp->snd_cwnd; 4091 u_int maxseg = tcp_maxseg(tp); 4092 4093 INP_WLOCK_ASSERT(tptoinpcb(tp)); 4094 4095 tcp_timer_activate(tp, TT_REXMT, 0); 4096 tp->t_rtttime = 0; 4097 if (IN_FASTRECOVERY(tp->t_flags)) { 4098 tp->snd_nxt = th->th_ack; 4099 /* 4100 * Set snd_cwnd to one segment beyond acknowledged offset. 4101 * (tp->snd_una has not yet been updated when this function is called.) 4102 */ 4103 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4104 tp->t_flags |= TF_ACKNOW; 4105 (void) tcp_output(tp); 4106 tp->snd_cwnd = ocwnd; 4107 if (SEQ_GT(onxt, tp->snd_nxt)) 4108 tp->snd_nxt = onxt; 4109 } 4110 /* 4111 * Partial window deflation. Relies on fact that tp->snd_una 4112 * not updated yet. 4113 */ 4114 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4115 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4116 else 4117 tp->snd_cwnd = 0; 4118 tp->snd_cwnd += maxseg; 4119 } 4120 4121 int 4122 tcp_compute_pipe(struct tcpcb *tp) 4123 { 4124 int pipe; 4125 4126 if (tp->t_fb->tfb_compute_pipe != NULL) { 4127 pipe = (*tp->t_fb->tfb_compute_pipe)(tp); 4128 } else { 4129 pipe = tp->snd_max - tp->snd_una + 4130 tp->sackhint.sack_bytes_rexmit - 4131 tp->sackhint.sacked_bytes - 4132 tp->sackhint.lost_bytes; 4133 } 4134 return (imax(pipe, 0)); 4135 } 4136 4137 uint32_t 4138 tcp_compute_initwnd(uint32_t maxseg) 4139 { 4140 /* 4141 * Calculate the Initial Window, also used as Restart Window 4142 * 4143 * RFC5681 Section 3.1 specifies the default conservative values. 4144 * RFC3390 specifies slightly more aggressive values. 4145 * RFC6928 increases it to ten segments. 4146 * Support for user specified value for initial flight size. 4147 */ 4148 if (V_tcp_initcwnd_segments) 4149 return min(V_tcp_initcwnd_segments * maxseg, 4150 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4151 else if (V_tcp_do_rfc3390) 4152 return min(4 * maxseg, max(2 * maxseg, 4380)); 4153 else { 4154 /* Per RFC5681 Section 3.1 */ 4155 if (maxseg > 2190) 4156 return (2 * maxseg); 4157 else if (maxseg > 1095) 4158 return (3 * maxseg); 4159 else 4160 return (4 * maxseg); 4161 } 4162 } 4163