1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 5 * The Regents of the University of California. All rights reserved. 6 * Copyright (c) 2007-2008,2010 7 * Swinburne University of Technology, Melbourne, Australia. 8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 9 * Copyright (c) 2010 The FreeBSD Foundation 10 * Copyright (c) 2010-2011 Juniper Networks, Inc. 11 * All rights reserved. 12 * 13 * Portions of this software were developed at the Centre for Advanced Internet 14 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 15 * James Healy and David Hayes, made possible in part by a grant from the Cisco 16 * University Research Program Fund at Community Foundation Silicon Valley. 17 * 18 * Portions of this software were developed at the Centre for Advanced 19 * Internet Architectures, Swinburne University of Technology, Melbourne, 20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 21 * 22 * Portions of this software were developed by Robert N. M. Watson under 23 * contract to Juniper Networks, Inc. 24 * 25 * Redistribution and use in source and binary forms, with or without 26 * modification, are permitted provided that the following conditions 27 * are met: 28 * 1. Redistributions of source code must retain the above copyright 29 * notice, this list of conditions and the following disclaimer. 30 * 2. Redistributions in binary form must reproduce the above copyright 31 * notice, this list of conditions and the following disclaimer in the 32 * documentation and/or other materials provided with the distribution. 33 * 3. Neither the name of the University nor the names of its contributors 34 * may be used to endorse or promote products derived from this software 35 * without specific prior written permission. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 */ 49 50 #include <sys/cdefs.h> 51 #include "opt_inet.h" 52 #include "opt_inet6.h" 53 #include "opt_ipsec.h" 54 #include "opt_rss.h" 55 56 #include <sys/param.h> 57 #include <sys/arb.h> 58 #include <sys/kernel.h> 59 #ifdef TCP_HHOOK 60 #include <sys/hhook.h> 61 #endif 62 #include <sys/malloc.h> 63 #include <sys/mbuf.h> 64 #include <sys/proc.h> /* for proc0 declaration */ 65 #include <sys/protosw.h> 66 #include <sys/qmath.h> 67 #include <sys/sdt.h> 68 #include <sys/signalvar.h> 69 #include <sys/socket.h> 70 #include <sys/socketvar.h> 71 #include <sys/sysctl.h> 72 #include <sys/syslog.h> 73 #include <sys/systm.h> 74 #include <sys/stats.h> 75 76 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 77 78 #include <vm/uma.h> 79 80 #include <net/if.h> 81 #include <net/if_var.h> 82 #include <net/route.h> 83 #include <net/rss_config.h> 84 #include <net/vnet.h> 85 86 #define TCPSTATES /* for logging */ 87 88 #include <netinet/in.h> 89 #include <netinet/in_kdtrace.h> 90 #include <netinet/in_pcb.h> 91 #include <netinet/in_rss.h> 92 #include <netinet/in_systm.h> 93 #include <netinet/ip.h> 94 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 95 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 96 #include <netinet/ip_var.h> 97 #include <netinet/ip_options.h> 98 #include <netinet/ip6.h> 99 #include <netinet/icmp6.h> 100 #include <netinet6/in6_pcb.h> 101 #include <netinet6/in6_rss.h> 102 #include <netinet6/in6_var.h> 103 #include <netinet6/ip6_var.h> 104 #include <netinet6/nd6.h> 105 #include <netinet/tcp.h> 106 #include <netinet/tcp_fsm.h> 107 #include <netinet/tcp_seq.h> 108 #include <netinet/tcp_timer.h> 109 #include <netinet/tcp_var.h> 110 #include <netinet/tcp_log_buf.h> 111 #include <netinet6/tcp6_var.h> 112 #include <netinet/tcpip.h> 113 #include <netinet/cc/cc.h> 114 #include <netinet/tcp_fastopen.h> 115 #include <netinet/tcp_syncache.h> 116 #ifdef TCP_OFFLOAD 117 #include <netinet/tcp_offload.h> 118 #endif 119 #include <netinet/tcp_ecn.h> 120 #include <netinet/udp.h> 121 122 #include <netipsec/ipsec_support.h> 123 124 #include <machine/in_cksum.h> 125 126 #include <security/mac/mac_framework.h> 127 128 const int tcprexmtthresh = 3; 129 130 VNET_DEFINE(int, tcp_log_in_vain) = 0; 131 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 132 &VNET_NAME(tcp_log_in_vain), 0, 133 "Log all incoming TCP segments to closed ports"); 134 135 VNET_DEFINE(int, tcp_bind_all_fibs) = 1; 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, bind_all_fibs, CTLFLAG_VNET | CTLFLAG_RDTUN, 137 &VNET_NAME(tcp_bind_all_fibs), 0, 138 "Bound sockets receive traffic from all FIBs"); 139 140 VNET_DEFINE(int, blackhole) = 0; 141 #define V_blackhole VNET(blackhole) 142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 143 &VNET_NAME(blackhole), 0, 144 "Do not send RST on segments to closed ports"); 145 146 VNET_DEFINE(bool, blackhole_local) = false; 147 #define V_blackhole_local VNET(blackhole_local) 148 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET | 149 CTLFLAG_RW, &VNET_NAME(blackhole_local), false, 150 "Enforce net.inet.tcp.blackhole for locally originated packets"); 151 152 VNET_DEFINE(int, tcp_delack_enabled) = 1; 153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 154 &VNET_NAME(tcp_delack_enabled), 0, 155 "Delay ACK to try and piggyback it onto a data packet"); 156 157 VNET_DEFINE(int, drop_synfin) = 0; 158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 159 &VNET_NAME(drop_synfin), 0, 160 "Drop TCP packets with SYN+FIN set"); 161 162 VNET_DEFINE(int, tcp_do_prr) = 1; 163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 164 &VNET_NAME(tcp_do_prr), 1, 165 "Enable Proportional Rate Reduction per RFC 6937"); 166 167 VNET_DEFINE(int, tcp_do_newcwv) = 0; 168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 169 &VNET_NAME(tcp_do_newcwv), 0, 170 "Enable New Congestion Window Validation per RFC7661"); 171 172 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 174 &VNET_NAME(tcp_do_rfc3042), 0, 175 "Enable RFC 3042 (Limited Transmit)"); 176 177 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 179 &VNET_NAME(tcp_do_rfc3390), 0, 180 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 181 182 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 184 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 185 "Slow-start flight size (initial congestion window) in number of segments"); 186 187 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 189 &VNET_NAME(tcp_do_rfc3465), 0, 190 "Enable RFC 3465 (Appropriate Byte Counting)"); 191 192 VNET_DEFINE(int, tcp_abc_l_var) = 2; 193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 194 &VNET_NAME(tcp_abc_l_var), 2, 195 "Cap the max cwnd increment during slow-start to this number of segments"); 196 197 VNET_DEFINE(int, tcp_insecure_syn) = 0; 198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 199 &VNET_NAME(tcp_insecure_syn), 0, 200 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 201 202 VNET_DEFINE(int, tcp_insecure_rst) = 0; 203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 204 &VNET_NAME(tcp_insecure_rst), 0, 205 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 206 207 VNET_DEFINE(int, tcp_insecure_ack) = 0; 208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_ack, CTLFLAG_VNET | CTLFLAG_RW, 209 &VNET_NAME(tcp_insecure_ack), 0, 210 "Follow RFC793 criteria for validating SEG.ACK"); 211 212 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 213 #define V_tcp_recvspace VNET(tcp_recvspace) 214 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 215 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 216 217 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 218 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 219 &VNET_NAME(tcp_do_autorcvbuf), 0, 220 "Enable automatic receive buffer sizing"); 221 222 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 224 &VNET_NAME(tcp_autorcvbuf_max), 0, 225 "Max size of automatic receive buffer"); 226 227 VNET_DEFINE(struct inpcbinfo, tcbinfo); 228 229 /* 230 * TCP statistics are stored in an array of counter(9)s, which size matches 231 * size of struct tcpstat. TCP running connection count is a regular array. 232 */ 233 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 234 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 235 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 236 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 237 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 238 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 239 "TCP connection counts by TCP state"); 240 241 /* 242 * Kernel module interface for updating tcpstat. The first argument is an index 243 * into tcpstat treated as an array. 244 */ 245 void 246 kmod_tcpstat_add(int statnum, int val) 247 { 248 249 counter_u64_add(VNET(tcpstat)[statnum], val); 250 } 251 252 /* 253 * Make sure that we only start a SACK loss recovery when 254 * receiving a duplicate ACK with a SACK block, and also 255 * complete SACK loss recovery in case the other end 256 * reneges. 257 */ 258 static bool inline 259 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to) 260 { 261 return ((tp->t_flags & TF_SACK_PERMIT) && 262 ((to->to_flags & TOF_SACK) || 263 (!TAILQ_EMPTY(&tp->snd_holes)))); 264 } 265 266 #ifdef TCP_HHOOK 267 /* 268 * Wrapper for the TCP established input helper hook. 269 */ 270 void 271 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 272 { 273 struct tcp_hhook_data hhook_data; 274 275 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 276 hhook_data.tp = tp; 277 hhook_data.th = th; 278 hhook_data.to = to; 279 280 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 281 &tp->t_osd); 282 } 283 } 284 #endif 285 286 /* 287 * CC wrapper hook functions 288 */ 289 void 290 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 291 uint16_t type) 292 { 293 #ifdef STATS 294 int32_t gput; 295 #endif 296 297 INP_WLOCK_ASSERT(tptoinpcb(tp)); 298 299 tp->t_ccv.nsegs = nsegs; 300 tp->t_ccv.bytes_this_ack = BYTES_THIS_ACK(tp, th); 301 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 302 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 303 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 304 tp->t_ccv.flags |= CCF_CWND_LIMITED; 305 else 306 tp->t_ccv.flags &= ~CCF_CWND_LIMITED; 307 308 if (type == CC_ACK) { 309 #ifdef STATS 310 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 311 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 312 if (!IN_RECOVERY(tp->t_flags)) 313 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 314 tp->t_ccv.bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 315 if ((tp->t_flags & TF_GPUTINPROG) && 316 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 317 /* 318 * Compute goodput in bits per millisecond. 319 */ 320 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) / 321 max(1, tcp_ts_getticks() - tp->gput_ts); 322 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 323 gput); 324 /* 325 * XXXLAS: This is a temporary hack, and should be 326 * chained off VOI_TCP_GPUT when stats(9) grows an API 327 * to deal with chained VOIs. 328 */ 329 if (tp->t_stats_gput_prev > 0) 330 stats_voi_update_abs_s32(tp->t_stats, 331 VOI_TCP_GPUT_ND, 332 ((gput - tp->t_stats_gput_prev) * 100) / 333 tp->t_stats_gput_prev); 334 tp->t_flags &= ~TF_GPUTINPROG; 335 tp->t_stats_gput_prev = gput; 336 } 337 #endif /* STATS */ 338 if (tp->snd_cwnd > tp->snd_ssthresh) { 339 tp->t_bytes_acked += tp->t_ccv.bytes_this_ack; 340 if (tp->t_bytes_acked >= tp->snd_cwnd) { 341 tp->t_bytes_acked -= tp->snd_cwnd; 342 tp->t_ccv.flags |= CCF_ABC_SENTAWND; 343 } 344 } else { 345 tp->t_ccv.flags &= ~CCF_ABC_SENTAWND; 346 tp->t_bytes_acked = 0; 347 } 348 } 349 350 if (CC_ALGO(tp)->ack_received != NULL) { 351 /* XXXLAS: Find a way to live without this */ 352 tp->t_ccv.curack = th->th_ack; 353 CC_ALGO(tp)->ack_received(&tp->t_ccv, type); 354 } 355 #ifdef STATS 356 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 357 #endif 358 } 359 360 void 361 cc_conn_init(struct tcpcb *tp) 362 { 363 struct hc_metrics_lite metrics; 364 struct inpcb *inp = tptoinpcb(tp); 365 u_int maxseg; 366 int rtt; 367 368 INP_WLOCK_ASSERT(inp); 369 370 tcp_hc_get(&inp->inp_inc, &metrics); 371 maxseg = tcp_maxseg(tp); 372 373 if (tp->t_srtt == 0 && (rtt = metrics.hc_rtt)) { 374 tp->t_srtt = rtt; 375 TCPSTAT_INC(tcps_usedrtt); 376 if (metrics.hc_rttvar) { 377 tp->t_rttvar = metrics.hc_rttvar; 378 TCPSTAT_INC(tcps_usedrttvar); 379 } else { 380 /* default variation is +- 1 rtt */ 381 tp->t_rttvar = 382 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 383 } 384 TCPT_RANGESET(tp->t_rxtcur, 385 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 386 tp->t_rttmin, tcp_rexmit_max); 387 } 388 if (metrics.hc_ssthresh) { 389 /* 390 * There's some sort of gateway or interface 391 * buffer limit on the path. Use this to set 392 * the slow start threshold, but set the 393 * threshold to no less than 2*mss. 394 */ 395 tp->snd_ssthresh = max(2 * maxseg, metrics.hc_ssthresh); 396 TCPSTAT_INC(tcps_usedssthresh); 397 } 398 399 /* 400 * Set the initial slow-start flight size. 401 * 402 * If a SYN or SYN/ACK was lost and retransmitted, we have to 403 * reduce the initial CWND to one segment as congestion is likely 404 * requiring us to be cautious. 405 */ 406 if (tp->snd_cwnd == 1) 407 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 408 else 409 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 410 411 if (CC_ALGO(tp)->conn_init != NULL) 412 CC_ALGO(tp)->conn_init(&tp->t_ccv); 413 } 414 415 void inline 416 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 417 { 418 INP_WLOCK_ASSERT(tptoinpcb(tp)); 419 420 #ifdef STATS 421 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 422 #endif 423 424 switch(type) { 425 case CC_NDUPACK: 426 if (!IN_FASTRECOVERY(tp->t_flags)) { 427 tp->snd_recover = tp->snd_max; 428 if (tp->t_flags2 & TF2_ECN_PERMIT) 429 tp->t_flags2 |= TF2_ECN_SND_CWR; 430 } 431 break; 432 case CC_ECN: 433 if (!IN_CONGRECOVERY(tp->t_flags) || 434 /* 435 * Allow ECN reaction on ACK to CWR, if 436 * that data segment was also CE marked. 437 */ 438 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 439 EXIT_CONGRECOVERY(tp->t_flags); 440 TCPSTAT_INC(tcps_ecn_rcwnd); 441 tp->snd_recover = tp->snd_max + 1; 442 if (tp->t_flags2 & TF2_ECN_PERMIT) 443 tp->t_flags2 |= TF2_ECN_SND_CWR; 444 } 445 break; 446 case CC_RTO: 447 tp->t_dupacks = 0; 448 tp->t_bytes_acked = 0; 449 EXIT_RECOVERY(tp->t_flags); 450 if (tp->t_flags2 & TF2_ECN_PERMIT) 451 tp->t_flags2 |= TF2_ECN_SND_CWR; 452 break; 453 case CC_RTO_ERR: 454 TCPSTAT_INC(tcps_sndrexmitbad); 455 /* RTO was unnecessary, so reset everything. */ 456 tp->snd_cwnd = tp->snd_cwnd_prev; 457 tp->snd_ssthresh = tp->snd_ssthresh_prev; 458 tp->snd_recover = tp->snd_recover_prev; 459 if (tp->t_flags & TF_WASFRECOVERY) 460 ENTER_FASTRECOVERY(tp->t_flags); 461 if (tp->t_flags & TF_WASCRECOVERY) 462 ENTER_CONGRECOVERY(tp->t_flags); 463 tp->snd_nxt = tp->snd_max; 464 tp->t_flags &= ~TF_PREVVALID; 465 tp->t_rxtshift = 0; 466 tp->t_badrxtwin = 0; 467 break; 468 } 469 if (SEQ_LT(tp->snd_fack, tp->snd_una) || 470 SEQ_GT(tp->snd_fack, tp->snd_max)) { 471 tp->snd_fack = tp->snd_una; 472 } 473 474 if (CC_ALGO(tp)->cong_signal != NULL) { 475 if (th != NULL) 476 tp->t_ccv.curack = th->th_ack; 477 CC_ALGO(tp)->cong_signal(&tp->t_ccv, type); 478 } 479 } 480 481 void inline 482 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 483 { 484 INP_WLOCK_ASSERT(tptoinpcb(tp)); 485 486 if (CC_ALGO(tp)->post_recovery != NULL) { 487 if (SEQ_LT(tp->snd_fack, th->th_ack) || 488 SEQ_GT(tp->snd_fack, tp->snd_max)) { 489 tp->snd_fack = th->th_ack; 490 } 491 tp->t_ccv.curack = th->th_ack; 492 CC_ALGO(tp)->post_recovery(&tp->t_ccv); 493 } 494 EXIT_RECOVERY(tp->t_flags); 495 496 tp->t_bytes_acked = 0; 497 tp->sackhint.delivered_data = 0; 498 tp->sackhint.prr_delivered = 0; 499 tp->sackhint.prr_out = 0; 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_ccv.flags & CCF_ACKNOW) { 544 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 545 tp->t_flags |= TF_ACKNOW; 546 } 547 } 548 } 549 550 void inline 551 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 552 { 553 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos); 554 } 555 556 /* 557 * TCP input handling is split into multiple parts: 558 * tcp6_input is a thin wrapper around tcp_input for the extended 559 * ip6_protox[] call format in ip6_input 560 * tcp_input handles primary segment validation, inpcb lookup and 561 * SYN processing on listen sockets 562 * tcp_do_segment processes the ACK and text of the segment for 563 * establishing, established and closing connections 564 */ 565 #ifdef INET6 566 int 567 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 568 { 569 struct mbuf *m; 570 571 m = *mp; 572 if (m->m_len < *offp + sizeof(struct tcphdr)) { 573 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 574 if (m == NULL) { 575 *mp = m; 576 TCPSTAT_INC(tcps_rcvshort); 577 return (IPPROTO_DONE); 578 } 579 } 580 581 *mp = m; 582 return (tcp_input_with_port(mp, offp, proto, port)); 583 } 584 585 int 586 tcp6_input(struct mbuf **mp, int *offp, int proto) 587 { 588 589 return(tcp6_input_with_port(mp, offp, proto, 0)); 590 } 591 #endif /* INET6 */ 592 593 int 594 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 595 { 596 struct mbuf *m = *mp; 597 struct tcphdr *th = NULL; 598 struct ip *ip = NULL; 599 struct inpcb *inp = NULL; 600 struct tcpcb *tp = NULL; 601 struct socket *so = NULL; 602 u_char *optp = NULL; 603 int off0; 604 int optlen = 0; 605 #ifdef INET 606 int len; 607 uint8_t ipttl; 608 #endif 609 int tlen = 0, off; 610 int drop_hdrlen; 611 int thflags; 612 int rstreason = 0; /* For badport_bandlim accounting purposes */ 613 int lookupflag; 614 uint8_t iptos; 615 struct m_tag *fwd_tag = NULL; 616 #ifdef INET6 617 struct ip6_hdr *ip6 = NULL; 618 int isipv6; 619 #else 620 const void *ip6 = NULL; 621 #endif /* INET6 */ 622 struct tcpopt to; /* options in this segment */ 623 char *s = NULL; /* address and port logging */ 624 bool closed_port = false; /* segment is hitting a closed port */ 625 626 NET_EPOCH_ASSERT(); 627 628 #ifdef INET6 629 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 630 #endif 631 632 off0 = *offp; 633 m = *mp; 634 *mp = NULL; 635 to.to_flags = 0; 636 TCPSTAT_INC(tcps_rcvtotal); 637 638 m->m_pkthdr.tcp_tun_port = port; 639 #ifdef INET6 640 if (isipv6) { 641 ip6 = mtod(m, struct ip6_hdr *); 642 th = (struct tcphdr *)((caddr_t)ip6 + off0); 643 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 644 if (port) 645 goto skip6_csum; 646 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 647 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 648 th->th_sum = m->m_pkthdr.csum_data; 649 else 650 th->th_sum = in6_cksum_pseudo(ip6, tlen, 651 IPPROTO_TCP, m->m_pkthdr.csum_data); 652 th->th_sum ^= 0xffff; 653 } else if (m->m_pkthdr.csum_flags & CSUM_IP6_TCP) { 654 /* 655 * Packet from local host (maybe from a VM). 656 * Checksum not required. 657 */ 658 th->th_sum = 0; 659 } else 660 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 661 if (th->th_sum) { 662 TCPSTAT_INC(tcps_rcvbadsum); 663 goto drop; 664 } 665 skip6_csum: 666 /* 667 * Be proactive about unspecified IPv6 address in source. 668 * As we use all-zero to indicate unbounded/unconnected pcb, 669 * unspecified IPv6 address can be used to confuse us. 670 * 671 * Note that packets with unspecified IPv6 destination is 672 * already dropped in ip6_input. 673 */ 674 KASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_dst), 675 ("%s: unspecified destination v6 address", __func__)); 676 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 677 IP6STAT_INC(ip6s_badscope); /* XXX */ 678 goto drop; 679 } 680 iptos = IPV6_TRAFFIC_CLASS(ip6); 681 } 682 #endif 683 #if defined(INET) && defined(INET6) 684 else 685 #endif 686 #ifdef INET 687 { 688 /* 689 * Get IP and TCP header together in first mbuf. 690 * Note: IP leaves IP header in first mbuf. 691 */ 692 if (off0 > sizeof (struct ip)) { 693 ip_stripoptions(m); 694 off0 = sizeof(struct ip); 695 } 696 if (m->m_len < sizeof (struct tcpiphdr)) { 697 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 698 == NULL) { 699 TCPSTAT_INC(tcps_rcvshort); 700 return (IPPROTO_DONE); 701 } 702 } 703 ip = mtod(m, struct ip *); 704 th = (struct tcphdr *)((caddr_t)ip + off0); 705 tlen = ntohs(ip->ip_len) - off0; 706 707 iptos = ip->ip_tos; 708 if (port) 709 goto skip_csum; 710 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 711 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 712 th->th_sum = m->m_pkthdr.csum_data; 713 else 714 th->th_sum = in_pseudo(ip->ip_src.s_addr, 715 ip->ip_dst.s_addr, 716 htonl(m->m_pkthdr.csum_data + tlen + 717 IPPROTO_TCP)); 718 th->th_sum ^= 0xffff; 719 } else if (m->m_pkthdr.csum_flags & CSUM_IP_TCP) { 720 /* 721 * Packet from local host (maybe from a VM). 722 * Checksum not required. 723 */ 724 th->th_sum = 0; 725 } else { 726 struct ipovly *ipov = (struct ipovly *)ip; 727 728 /* 729 * Checksum extended TCP header and data. 730 */ 731 len = off0 + tlen; 732 ipttl = ip->ip_ttl; 733 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 734 ipov->ih_len = htons(tlen); 735 th->th_sum = in_cksum(m, len); 736 /* Reset length for SDT probes. */ 737 ip->ip_len = htons(len); 738 /* Reset TOS bits */ 739 ip->ip_tos = iptos; 740 /* Re-initialization for later version check */ 741 ip->ip_ttl = ipttl; 742 ip->ip_v = IPVERSION; 743 ip->ip_hl = off0 >> 2; 744 } 745 skip_csum: 746 if (th->th_sum && (port == 0)) { 747 TCPSTAT_INC(tcps_rcvbadsum); 748 goto drop; 749 } 750 KASSERT(ip->ip_dst.s_addr != INADDR_ANY, 751 ("%s: unspecified destination v4 address", __func__)); 752 if (__predict_false(ip->ip_src.s_addr == INADDR_ANY)) { 753 IPSTAT_INC(ips_badaddr); 754 goto drop; 755 } 756 } 757 #endif /* INET */ 758 759 /* 760 * Check that TCP offset makes sense, 761 * pull out TCP options and adjust length. XXX 762 */ 763 off = th->th_off << 2; 764 if (off < sizeof (struct tcphdr) || off > tlen) { 765 TCPSTAT_INC(tcps_rcvbadoff); 766 goto drop; 767 } 768 tlen -= off; /* tlen is used instead of ti->ti_len */ 769 if (off > sizeof (struct tcphdr)) { 770 #ifdef INET6 771 if (isipv6) { 772 if (m->m_len < off0 + off) { 773 m = m_pullup(m, off0 + off); 774 if (m == NULL) { 775 TCPSTAT_INC(tcps_rcvshort); 776 return (IPPROTO_DONE); 777 } 778 } 779 ip6 = mtod(m, struct ip6_hdr *); 780 th = (struct tcphdr *)((caddr_t)ip6 + off0); 781 } 782 #endif 783 #if defined(INET) && defined(INET6) 784 else 785 #endif 786 #ifdef INET 787 { 788 if (m->m_len < sizeof(struct ip) + off) { 789 if ((m = m_pullup(m, sizeof (struct ip) + off)) 790 == NULL) { 791 TCPSTAT_INC(tcps_rcvshort); 792 return (IPPROTO_DONE); 793 } 794 ip = mtod(m, struct ip *); 795 th = (struct tcphdr *)((caddr_t)ip + off0); 796 } 797 } 798 #endif 799 optlen = off - sizeof (struct tcphdr); 800 optp = (u_char *)(th + 1); 801 } 802 thflags = tcp_get_flags(th); 803 804 /* 805 * Convert TCP protocol specific fields to host format. 806 */ 807 tcp_fields_to_host(th); 808 809 /* 810 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 811 */ 812 drop_hdrlen = off0 + off; 813 814 /* 815 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 816 */ 817 if ( 818 #ifdef INET6 819 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 820 #ifdef INET 821 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 822 #endif 823 #endif 824 #if defined(INET) && !defined(INET6) 825 (m->m_flags & M_IP_NEXTHOP) 826 #endif 827 ) 828 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 829 830 /* 831 * For initial SYN packets we don't need write lock on matching 832 * PCB, be it a listening one or a synchronized one. The packet 833 * shall not modify its state. 834 */ 835 lookupflag = INPLOOKUP_WILDCARD | 836 ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ? 837 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB) | 838 (V_tcp_bind_all_fibs ? 0 : INPLOOKUP_FIB); 839 findpcb: 840 tp = NULL; 841 #ifdef INET6 842 if (isipv6 && fwd_tag != NULL) { 843 struct sockaddr_in6 *next_hop6; 844 845 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 846 /* 847 * Transparently forwarded. Pretend to be the destination. 848 * Already got one like this? 849 */ 850 inp = in6_pcblookup_mbuf(&V_tcbinfo, 851 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 852 lookupflag & ~INPLOOKUP_WILDCARD, m->m_pkthdr.rcvif, m); 853 if (!inp) { 854 /* 855 * It's new. Try to find the ambushing socket. 856 * Because we've rewritten the destination address, 857 * any hardware-generated hash is ignored. 858 */ 859 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 860 th->th_sport, &next_hop6->sin6_addr, 861 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 862 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 863 } 864 } else if (isipv6) { 865 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 866 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag, 867 m->m_pkthdr.rcvif, m); 868 } 869 #endif /* INET6 */ 870 #if defined(INET6) && defined(INET) 871 else 872 #endif 873 #ifdef INET 874 if (fwd_tag != NULL) { 875 struct sockaddr_in *next_hop; 876 877 next_hop = (struct sockaddr_in *)(fwd_tag+1); 878 /* 879 * Transparently forwarded. Pretend to be the destination. 880 * already got one like this? 881 */ 882 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 883 ip->ip_dst, th->th_dport, lookupflag & ~INPLOOKUP_WILDCARD, 884 m->m_pkthdr.rcvif, m); 885 if (!inp) { 886 /* 887 * It's new. Try to find the ambushing socket. 888 * Because we've rewritten the destination address, 889 * any hardware-generated hash is ignored. 890 */ 891 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 892 th->th_sport, next_hop->sin_addr, 893 next_hop->sin_port ? ntohs(next_hop->sin_port) : 894 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 895 } 896 } else 897 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 898 th->th_sport, ip->ip_dst, th->th_dport, lookupflag, 899 m->m_pkthdr.rcvif, m); 900 #endif /* INET */ 901 902 /* 903 * If the INPCB does not exist then all data in the incoming 904 * segment is discarded and an appropriate RST is sent back. 905 * XXX MRT Send RST using which routing table? 906 */ 907 if (inp == NULL) { 908 if (rstreason != 0) { 909 /* We came here after second (safety) lookup. */ 910 MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0); 911 goto dropwithreset; 912 } 913 /* 914 * Log communication attempts to ports that are not 915 * in use. 916 */ 917 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 918 V_tcp_log_in_vain == 2) { 919 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 920 log(LOG_INFO, "%s; %s: Connection attempt " 921 "to closed port\n", s, __func__); 922 } 923 rstreason = BANDLIM_TCP_RST; 924 closed_port = true; 925 goto dropwithreset; 926 } 927 INP_LOCK_ASSERT(inp); 928 929 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 930 !SOLISTENING(inp->inp_socket)) { 931 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) { 932 inp->inp_flowid = m->m_pkthdr.flowid; 933 inp->inp_flowtype = M_HASHTYPE_GET(m); 934 #ifdef RSS 935 } else { 936 /* assign flowid by software RSS hash */ 937 #ifdef INET6 938 if (isipv6) { 939 rss_proto_software_hash_v6(&inp->in6p_faddr, 940 &inp->in6p_laddr, 941 inp->inp_fport, 942 inp->inp_lport, 943 IPPROTO_TCP, 944 &inp->inp_flowid, 945 &inp->inp_flowtype); 946 } else 947 #endif /* INET6 */ 948 { 949 rss_proto_software_hash_v4(inp->inp_faddr, 950 inp->inp_laddr, 951 inp->inp_fport, 952 inp->inp_lport, 953 IPPROTO_TCP, 954 &inp->inp_flowid, 955 &inp->inp_flowtype); 956 } 957 #endif /* RSS */ 958 } 959 } 960 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 961 #ifdef INET6 962 if (isipv6 && IPSEC_ENABLED(ipv6) && 963 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 964 goto dropunlock; 965 } 966 #ifdef INET 967 else 968 #endif 969 #endif /* INET6 */ 970 #ifdef INET 971 if (IPSEC_ENABLED(ipv4) && 972 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 973 goto dropunlock; 974 } 975 #endif /* INET */ 976 #endif /* IPSEC */ 977 978 /* 979 * Check the minimum TTL for socket. 980 */ 981 if (inp->inp_ip_minttl != 0) { 982 #ifdef INET6 983 if (isipv6) { 984 if (inp->inp_ip_minttl > ip6->ip6_hlim) 985 goto dropunlock; 986 } else 987 #endif 988 if (inp->inp_ip_minttl > ip->ip_ttl) 989 goto dropunlock; 990 } 991 992 tp = intotcpcb(inp); 993 switch (tp->t_state) { 994 case TCPS_TIME_WAIT: 995 /* 996 * A previous connection in TIMEWAIT state is supposed to catch 997 * stray or duplicate segments arriving late. If this segment 998 * was a legitimate new connection attempt, the old INPCB gets 999 * removed and we can try again to find a listening socket. 1000 */ 1001 tcp_dooptions(&to, optp, optlen, 1002 (thflags & TH_SYN) ? TO_SYN : 0); 1003 /* 1004 * tcp_twcheck unlocks the inp always, and frees the m if fails. 1005 */ 1006 if (tcp_twcheck(inp, &to, th, m, tlen)) 1007 goto findpcb; 1008 return (IPPROTO_DONE); 1009 case TCPS_CLOSED: 1010 /* 1011 * The TCPCB may no longer exist if the connection is winding 1012 * down or it is in the CLOSED state. Either way we drop the 1013 * segment and send an appropriate response. 1014 */ 1015 rstreason = BANDLIM_TCP_RST; 1016 closed_port = true; 1017 goto dropwithreset; 1018 } 1019 1020 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 1021 rstreason = BANDLIM_TCP_RST; 1022 closed_port = true; 1023 goto dropwithreset; 1024 } 1025 1026 #ifdef TCP_OFFLOAD 1027 if (tp->t_flags & TF_TOE) { 1028 tcp_offload_input(tp, m); 1029 m = NULL; /* consumed by the TOE driver */ 1030 goto dropunlock; 1031 } 1032 #endif 1033 1034 #ifdef MAC 1035 if (mac_inpcb_check_deliver(inp, m)) 1036 goto dropunlock; 1037 #endif 1038 so = inp->inp_socket; 1039 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1040 /* 1041 * When the socket is accepting connections (the INPCB is in LISTEN 1042 * state) we look into the SYN cache if this is a new connection 1043 * attempt or the completion of a previous one. 1044 */ 1045 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), 1046 ("%s: so accepting but tp %p not listening", __func__, tp)); 1047 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { 1048 struct in_conninfo inc; 1049 1050 bzero(&inc, sizeof(inc)); 1051 #ifdef INET6 1052 if (isipv6) { 1053 inc.inc_flags |= INC_ISIPV6; 1054 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1055 inc.inc_flags |= INC_IPV6MINMTU; 1056 inc.inc6_faddr = ip6->ip6_src; 1057 inc.inc6_laddr = ip6->ip6_dst; 1058 } else 1059 #endif 1060 { 1061 inc.inc_faddr = ip->ip_src; 1062 inc.inc_laddr = ip->ip_dst; 1063 } 1064 inc.inc_fport = th->th_sport; 1065 inc.inc_lport = th->th_dport; 1066 inc.inc_fibnum = so->so_fibnum; 1067 1068 /* 1069 * Check for an existing connection attempt in syncache if 1070 * the flag is only ACK. A successful lookup creates a new 1071 * socket appended to the listen queue in SYN_RECEIVED state. 1072 */ 1073 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1074 int result; 1075 1076 /* 1077 * Parse the TCP options here because 1078 * syncookies need access to the reflected 1079 * timestamp. 1080 */ 1081 tcp_dooptions(&to, optp, optlen, 0); 1082 /* 1083 * NB: syncache_expand() doesn't unlock inp. 1084 */ 1085 result = syncache_expand(&inc, &to, th, &so, m, port); 1086 if (result < 0) { 1087 /* 1088 * A failing TCP MD5 signature comparison 1089 * must result in the segment being dropped 1090 * and must not produce any response back 1091 * to the sender. 1092 */ 1093 goto dropunlock; 1094 } else if (result == 0) { 1095 /* 1096 * No syncache entry, or ACK was not for our 1097 * SYN/ACK. Do our protection against double 1098 * ACK. If peer sent us 2 ACKs, then for the 1099 * first one syncache_expand() successfully 1100 * converted syncache entry into a socket, 1101 * while we were waiting on the inpcb lock. We 1102 * don't want to sent RST for the second ACK, 1103 * so we perform second lookup without wildcard 1104 * match, hoping to find the new socket. If 1105 * the ACK is stray indeed, rstreason would 1106 * hint the above code that the lookup was a 1107 * second attempt. 1108 * 1109 * NB: syncache did its own logging 1110 * of the failure cause. 1111 */ 1112 INP_WUNLOCK(inp); 1113 rstreason = BANDLIM_TCP_RST; 1114 lookupflag &= ~INPLOOKUP_WILDCARD; 1115 goto findpcb; 1116 } 1117 tfo_socket_result: 1118 if (so == NULL) { 1119 /* 1120 * We completed the 3-way handshake 1121 * but could not allocate a socket 1122 * either due to memory shortage, 1123 * listen queue length limits or 1124 * global socket limits. Send RST 1125 * or wait and have the remote end 1126 * retransmit the ACK for another 1127 * try. 1128 */ 1129 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1130 log(LOG_DEBUG, "%s; %s: Listen socket: " 1131 "Socket allocation failed due to " 1132 "limits or memory shortage, %s\n", 1133 s, __func__, 1134 V_tcp_sc_rst_sock_fail ? 1135 "sending RST" : "try again"); 1136 if (V_tcp_sc_rst_sock_fail) { 1137 rstreason = BANDLIM_UNLIMITED; 1138 goto dropwithreset; 1139 } else 1140 goto dropunlock; 1141 } 1142 /* 1143 * Socket is created in state SYN_RECEIVED. 1144 * Unlock the listen socket, lock the newly 1145 * created socket and update the tp variable. 1146 * If we came here via jump to tfo_socket_result, 1147 * then listening socket is read-locked. 1148 */ 1149 INP_UNLOCK(inp); /* listen socket */ 1150 inp = sotoinpcb(so); 1151 /* 1152 * New connection inpcb is already locked by 1153 * syncache_expand(). 1154 */ 1155 INP_WLOCK_ASSERT(inp); 1156 tp = intotcpcb(inp); 1157 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1158 ("%s: ", __func__)); 1159 /* 1160 * Process the segment and the data it 1161 * contains. tcp_do_segment() consumes 1162 * the mbuf chain and unlocks the inpcb. 1163 */ 1164 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1165 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, 1166 tlen, iptos); 1167 return (IPPROTO_DONE); 1168 } 1169 /* 1170 * Segment flag validation for new connection attempts: 1171 * 1172 * Our (SYN|ACK) response was rejected. 1173 * Check with syncache and remove entry to prevent 1174 * retransmits. 1175 * 1176 * NB: syncache_chkrst does its own logging of failure 1177 * causes. 1178 */ 1179 if (thflags & TH_RST) { 1180 syncache_chkrst(&inc, th, m, port); 1181 goto dropunlock; 1182 } 1183 /* 1184 * We can't do anything without SYN. 1185 */ 1186 if ((thflags & TH_SYN) == 0) { 1187 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1188 log(LOG_DEBUG, "%s; %s: Listen socket: " 1189 "SYN is missing, segment ignored\n", 1190 s, __func__); 1191 TCPSTAT_INC(tcps_badsyn); 1192 goto dropunlock; 1193 } 1194 /* 1195 * (SYN|ACK) is bogus on a listen socket. 1196 */ 1197 if (thflags & TH_ACK) { 1198 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1199 log(LOG_DEBUG, "%s; %s: Listen socket: " 1200 "SYN|ACK invalid, segment rejected\n", 1201 s, __func__); 1202 syncache_badack(&inc, port); /* XXX: Not needed! */ 1203 TCPSTAT_INC(tcps_badsyn); 1204 rstreason = BANDLIM_TCP_RST; 1205 goto dropwithreset; 1206 } 1207 /* 1208 * If the drop_synfin option is enabled, drop all 1209 * segments with both the SYN and FIN bits set. 1210 * This prevents e.g. nmap from identifying the 1211 * TCP/IP stack. 1212 * XXX: Poor reasoning. nmap has other methods 1213 * and is constantly refining its stack detection 1214 * strategies. 1215 * XXX: This is a violation of the TCP specification 1216 * and was used by RFC1644. 1217 */ 1218 if ((thflags & TH_FIN) && V_drop_synfin) { 1219 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1220 log(LOG_DEBUG, "%s; %s: Listen socket: " 1221 "SYN|FIN segment ignored (based on " 1222 "sysctl setting)\n", s, __func__); 1223 TCPSTAT_INC(tcps_badsyn); 1224 goto dropunlock; 1225 } 1226 /* 1227 * Segment's flags are (SYN) or (SYN|FIN). 1228 * 1229 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1230 * as they do not affect the state of the TCP FSM. 1231 * The data pointed to by TH_URG and th_urp is ignored. 1232 */ 1233 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1234 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1235 KASSERT(thflags & (TH_SYN), 1236 ("%s: Listen socket: TH_SYN not set", __func__)); 1237 INP_RLOCK_ASSERT(inp); 1238 #ifdef INET6 1239 /* 1240 * If deprecated address is forbidden, 1241 * we do not accept SYN to deprecated interface 1242 * address to prevent any new inbound connection from 1243 * getting established. 1244 * When we do not accept SYN, we send a TCP RST, 1245 * with deprecated source address (instead of dropping 1246 * it). We compromise it as it is much better for peer 1247 * to send a RST, and RST will be the final packet 1248 * for the exchange. 1249 * 1250 * If we do not forbid deprecated addresses, we accept 1251 * the SYN packet. RFC2462 does not suggest dropping 1252 * SYN in this case. 1253 * If we decipher RFC2462 5.5.4, it says like this: 1254 * 1. use of deprecated addr with existing 1255 * communication is okay - "SHOULD continue to be 1256 * used" 1257 * 2. use of it with new communication: 1258 * (2a) "SHOULD NOT be used if alternate address 1259 * with sufficient scope is available" 1260 * (2b) nothing mentioned otherwise. 1261 * Here we fall into (2b) case as we have no choice in 1262 * our source address selection - we must obey the peer. 1263 * 1264 * The wording in RFC2462 is confusing, and there are 1265 * multiple description text for deprecated address 1266 * handling - worse, they are not exactly the same. 1267 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1268 */ 1269 if (isipv6 && !V_ip6_use_deprecated) { 1270 struct in6_ifaddr *ia6; 1271 1272 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1273 if (ia6 != NULL && 1274 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1275 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1276 log(LOG_DEBUG, "%s; %s: Listen socket: " 1277 "Connection attempt to deprecated " 1278 "IPv6 address rejected\n", 1279 s, __func__); 1280 rstreason = BANDLIM_TCP_RST; 1281 goto dropwithreset; 1282 } 1283 } 1284 #endif /* INET6 */ 1285 /* 1286 * Basic sanity checks on incoming SYN requests: 1287 * Don't respond if the destination is a link layer 1288 * broadcast according to RFC1122 4.2.3.10, p. 104. 1289 * If it is from this socket it must be forged. 1290 * Don't respond if the source or destination is a 1291 * global or subnet broad- or multicast address. 1292 * Note that it is quite possible to receive unicast 1293 * link-layer packets with a broadcast IP address. Use 1294 * in_ifnet_broadcast() to find them. 1295 */ 1296 if (m->m_flags & (M_BCAST|M_MCAST)) { 1297 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1298 log(LOG_DEBUG, "%s; %s: Listen socket: " 1299 "Connection attempt from broad- or multicast " 1300 "link layer address ignored\n", s, __func__); 1301 goto dropunlock; 1302 } 1303 #ifdef INET6 1304 if (isipv6) { 1305 if (th->th_dport == th->th_sport && 1306 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1307 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1308 log(LOG_DEBUG, "%s; %s: Listen socket: " 1309 "Connection attempt to/from self " 1310 "ignored\n", s, __func__); 1311 goto dropunlock; 1312 } 1313 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1314 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1315 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1316 log(LOG_DEBUG, "%s; %s: Listen socket: " 1317 "Connection attempt from/to multicast " 1318 "address ignored\n", s, __func__); 1319 goto dropunlock; 1320 } 1321 } 1322 #endif 1323 #if defined(INET) && defined(INET6) 1324 else 1325 #endif 1326 #ifdef INET 1327 { 1328 if (th->th_dport == th->th_sport && 1329 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1330 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1331 log(LOG_DEBUG, "%s; %s: Listen socket: " 1332 "Connection attempt from/to self " 1333 "ignored\n", s, __func__); 1334 goto dropunlock; 1335 } 1336 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1337 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1338 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1339 in_ifnet_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1340 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1341 log(LOG_DEBUG, "%s; %s: Listen socket: " 1342 "Connection attempt from/to broad- " 1343 "or multicast address ignored\n", 1344 s, __func__); 1345 goto dropunlock; 1346 } 1347 } 1348 #endif 1349 /* 1350 * SYN appears to be valid. Create compressed TCP state 1351 * for syncache. 1352 */ 1353 TCP_PROBE3(debug__input, tp, th, m); 1354 tcp_dooptions(&to, optp, optlen, TO_SYN); 1355 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, 1356 iptos, port)) != NULL) 1357 goto tfo_socket_result; 1358 1359 /* 1360 * Entry added to syncache and mbuf consumed. 1361 * Only the listen socket is unlocked by syncache_add(). 1362 */ 1363 return (IPPROTO_DONE); 1364 } 1365 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1366 if (tp->t_flags & TF_SIGNATURE) { 1367 tcp_dooptions(&to, optp, optlen, thflags); 1368 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1369 TCPSTAT_INC(tcps_sig_err_nosigopt); 1370 goto dropunlock; 1371 } 1372 if (!TCPMD5_ENABLED() || 1373 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1374 goto dropunlock; 1375 } 1376 #endif 1377 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1378 1379 /* 1380 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1381 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1382 * the inpcb, and unlocks pcbinfo. 1383 * 1384 * XXXGL: in case of a pure SYN arriving on existing connection 1385 * TCP stacks won't need to modify the PCB, they would either drop 1386 * the segment silently, or send a challenge ACK. However, we try 1387 * to upgrade the lock, because calling convention for stacks is 1388 * write-lock on PCB. If upgrade fails, drop the SYN. 1389 */ 1390 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0) 1391 goto dropunlock; 1392 1393 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, tlen, iptos); 1394 return (IPPROTO_DONE); 1395 1396 dropwithreset: 1397 /* 1398 * When blackholing do not respond with a RST but 1399 * completely ignore the segment and drop it. 1400 */ 1401 if (rstreason == BANDLIM_TCP_RST && 1402 ((!closed_port && V_blackhole == 3) || 1403 (closed_port && 1404 ((V_blackhole == 1 && (thflags & TH_SYN)) || V_blackhole > 1))) && 1405 (V_blackhole_local || ( 1406 #ifdef INET6 1407 isipv6 ? !in6_localip(&ip6->ip6_src) : 1408 #endif 1409 #ifdef INET 1410 !in_localip(ip->ip_src) 1411 #else 1412 true 1413 #endif 1414 ))) 1415 goto dropunlock; 1416 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1417 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1418 m = NULL; /* mbuf chain got consumed. */ 1419 1420 dropunlock: 1421 if (m != NULL) 1422 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1423 1424 if (inp != NULL) 1425 INP_UNLOCK(inp); 1426 1427 drop: 1428 if (s != NULL) 1429 free(s, M_TCPLOG); 1430 if (m != NULL) 1431 m_freem(m); 1432 return (IPPROTO_DONE); 1433 } 1434 1435 /* 1436 * Automatic sizing of receive socket buffer. Often the send 1437 * buffer size is not optimally adjusted to the actual network 1438 * conditions at hand (delay bandwidth product). Setting the 1439 * buffer size too small limits throughput on links with high 1440 * bandwidth and high delay (eg. trans-continental/oceanic links). 1441 * 1442 * On the receive side the socket buffer memory is only rarely 1443 * used to any significant extent. This allows us to be much 1444 * more aggressive in scaling the receive socket buffer. For 1445 * the case that the buffer space is actually used to a large 1446 * extent and we run out of kernel memory we can simply drop 1447 * the new segments; TCP on the sender will just retransmit it 1448 * later. Setting the buffer size too big may only consume too 1449 * much kernel memory if the application doesn't read() from 1450 * the socket or packet loss or reordering makes use of the 1451 * reassembly queue. 1452 * 1453 * The criteria to step up the receive buffer one notch are: 1454 * 1. Application has not set receive buffer size with 1455 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1456 * 2. the number of bytes received during 1/2 of an sRTT 1457 * is at least 3/8 of the current socket buffer size. 1458 * 3. receive buffer size has not hit maximal automatic size; 1459 * 1460 * If all of the criteria are met, we increase the socket buffer 1461 * by a 1/2 (bounded by the max). This allows us to keep ahead 1462 * of slow-start but also makes it so our peer never gets limited 1463 * by our rwnd which we then open up causing a burst. 1464 * 1465 * This algorithm does two steps per RTT at most and only if 1466 * we receive a bulk stream w/o packet losses or reorderings. 1467 * Shrinking the buffer during idle times is not necessary as 1468 * it doesn't consume any memory when idle. 1469 * 1470 * TODO: Only step up if the application is actually serving 1471 * the buffer to better manage the socket buffer resources. 1472 */ 1473 int 1474 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1475 struct tcpcb *tp, int tlen) 1476 { 1477 int newsize = 0; 1478 1479 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1480 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1481 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1482 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1483 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1484 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1485 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1486 } 1487 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1488 1489 /* Start over with next RTT. */ 1490 tp->rfbuf_ts = 0; 1491 tp->rfbuf_cnt = 0; 1492 } else { 1493 tp->rfbuf_cnt += tlen; /* add up */ 1494 } 1495 return (newsize); 1496 } 1497 1498 int 1499 tcp_input(struct mbuf **mp, int *offp, int proto) 1500 { 1501 return(tcp_input_with_port(mp, offp, proto, 0)); 1502 } 1503 1504 static void 1505 tcp_handle_wakeup(struct tcpcb *tp) 1506 { 1507 1508 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1509 1510 if (tp->t_flags & TF_WAKESOR) { 1511 struct socket *so = tptosocket(tp); 1512 1513 tp->t_flags &= ~TF_WAKESOR; 1514 SOCK_RECVBUF_LOCK_ASSERT(so); 1515 sorwakeup_locked(so); 1516 } 1517 } 1518 1519 void 1520 tcp_do_segment(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th, 1521 int drop_hdrlen, int tlen, uint8_t iptos) 1522 { 1523 uint16_t thflags; 1524 int acked, ourfinisacked, needoutput = 0; 1525 sackstatus_t sack_changed; 1526 int rstreason, todrop, win, incforsyn = 0; 1527 uint32_t tiwin; 1528 uint16_t nsegs; 1529 char *s; 1530 struct inpcb *inp = tptoinpcb(tp); 1531 struct socket *so = tptosocket(tp); 1532 struct in_conninfo *inc = &inp->inp_inc; 1533 struct mbuf *mfree; 1534 struct tcpopt to; 1535 int tfo_syn; 1536 u_int maxseg = 0; 1537 bool no_data; 1538 1539 no_data = (tlen == 0); 1540 thflags = tcp_get_flags(th); 1541 tp->sackhint.last_sack_ack = 0; 1542 sack_changed = SACK_NOCHANGE; 1543 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1544 1545 NET_EPOCH_ASSERT(); 1546 INP_WLOCK_ASSERT(inp); 1547 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1548 __func__)); 1549 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1550 __func__)); 1551 1552 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1553 tlen, NULL, true); 1554 1555 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1556 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1557 log(LOG_DEBUG, "%s; %s: " 1558 "SYN|FIN segment ignored (based on " 1559 "sysctl setting)\n", s, __func__); 1560 free(s, M_TCPLOG); 1561 } 1562 goto drop; 1563 } 1564 1565 /* 1566 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1567 * check SEQ.ACK first. 1568 */ 1569 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1570 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1571 rstreason = BANDLIM_UNLIMITED; 1572 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1573 goto dropwithreset; 1574 } 1575 1576 /* 1577 * Segment received on connection. 1578 * Reset idle time and keep-alive timer. 1579 * XXX: This should be done after segment 1580 * validation to ignore broken/spoofed segs. 1581 */ 1582 if (tp->t_idle_reduce && 1583 (tp->snd_max == tp->snd_una) && 1584 ((ticks - tp->t_rcvtime) >= tp->t_rxtcur)) 1585 cc_after_idle(tp); 1586 tp->t_rcvtime = ticks; 1587 1588 if (thflags & TH_FIN) 1589 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN); 1590 /* 1591 * Scale up the window into a 32-bit value. 1592 * For the SYN_SENT state the scale is zero. 1593 */ 1594 tiwin = th->th_win << tp->snd_scale; 1595 #ifdef STATS 1596 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1597 #endif 1598 1599 /* 1600 * TCP ECN processing. 1601 */ 1602 if (tcp_ecn_input_segment(tp, thflags, tlen, 1603 tcp_packets_this_ack(tp, th->th_ack), 1604 iptos)) 1605 cc_cong_signal(tp, th, CC_ECN); 1606 1607 /* 1608 * Parse options on any incoming segment. 1609 */ 1610 tcp_dooptions(&to, (u_char *)(th + 1), 1611 (th->th_off << 2) - sizeof(struct tcphdr), 1612 (thflags & TH_SYN) ? TO_SYN : 0); 1613 if (tp->t_flags2 & TF2_PROC_SACK_PROHIBIT) { 1614 /* 1615 * We don't look at sack's from the 1616 * peer because the MSS is too small which 1617 * can subject us to an attack. 1618 */ 1619 to.to_flags &= ~TOF_SACK; 1620 } 1621 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1622 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1623 (to.to_flags & TOF_SIGNATURE) == 0) { 1624 TCPSTAT_INC(tcps_sig_err_sigopt); 1625 /* XXX: should drop? */ 1626 } 1627 #endif 1628 /* 1629 * If echoed timestamp is later than the current time, 1630 * fall back to non RFC1323 RTT calculation. Normalize 1631 * timestamp if syncookies were used when this connection 1632 * was established. 1633 */ 1634 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1635 to.to_tsecr -= tp->ts_offset; 1636 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) { 1637 to.to_tsecr = 0; 1638 } 1639 } 1640 /* 1641 * Process options only when we get SYN/ACK back. The SYN case 1642 * for incoming connections is handled in tcp_syncache. 1643 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1644 * or <SYN,ACK>) segment itself is never scaled. 1645 * XXX this is traditional behavior, may need to be cleaned up. 1646 */ 1647 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1648 /* Handle parallel SYN for ECN */ 1649 tcp_ecn_input_parallel_syn(tp, thflags, iptos); 1650 if ((to.to_flags & TOF_SCALE) && 1651 (tp->t_flags & TF_REQ_SCALE) && 1652 !(tp->t_flags & TF_NOOPT)) { 1653 tp->t_flags |= TF_RCVD_SCALE; 1654 tp->snd_scale = to.to_wscale; 1655 } else { 1656 tp->t_flags &= ~TF_REQ_SCALE; 1657 } 1658 /* 1659 * Initial send window. It will be updated with 1660 * the next incoming segment to the scaled value. 1661 */ 1662 tp->snd_wnd = th->th_win; 1663 if ((to.to_flags & TOF_TS) && 1664 (tp->t_flags & TF_REQ_TSTMP) && 1665 !(tp->t_flags & TF_NOOPT)) { 1666 tp->t_flags |= TF_RCVD_TSTMP; 1667 tp->ts_recent = to.to_tsval; 1668 tp->ts_recent_age = tcp_ts_getticks(); 1669 } else { 1670 tp->t_flags &= ~TF_REQ_TSTMP; 1671 } 1672 if (to.to_flags & TOF_MSS) { 1673 tcp_mss(tp, to.to_mss); 1674 } 1675 if ((tp->t_flags & TF_SACK_PERMIT) && 1676 (!(to.to_flags & TOF_SACKPERM) || 1677 (tp->t_flags & TF_NOOPT))) { 1678 tp->t_flags &= ~TF_SACK_PERMIT; 1679 } 1680 if (tp->t_flags & TF_FASTOPEN) { 1681 if ((to.to_flags & TOF_FASTOPEN) && 1682 !(tp->t_flags & TF_NOOPT)) { 1683 uint16_t mss; 1684 1685 if (to.to_flags & TOF_MSS) { 1686 mss = to.to_mss; 1687 } else { 1688 if ((inp->inp_vflag & INP_IPV6) != 0) { 1689 mss = TCP6_MSS; 1690 } else { 1691 mss = TCP_MSS; 1692 } 1693 } 1694 tcp_fastopen_update_cache(tp, mss, 1695 to.to_tfo_len, to.to_tfo_cookie); 1696 } else { 1697 tcp_fastopen_disable_path(tp); 1698 } 1699 } 1700 } 1701 1702 /* 1703 * If timestamps were negotiated during SYN/ACK and a 1704 * segment without a timestamp is received, silently drop 1705 * the segment, unless it is a RST segment or missing timestamps are 1706 * tolerated. 1707 * See section 3.2 of RFC 7323. 1708 */ 1709 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1710 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1711 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1712 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1713 "segment processed normally\n", 1714 s, __func__); 1715 free(s, M_TCPLOG); 1716 } 1717 } else { 1718 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1719 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1720 "segment silently dropped\n", s, __func__); 1721 free(s, M_TCPLOG); 1722 } 1723 goto drop; 1724 } 1725 } 1726 /* 1727 * If timestamps were not negotiated during SYN/ACK and a 1728 * segment with a timestamp is received, ignore the 1729 * timestamp and process the packet normally. 1730 * See section 3.2 of RFC 7323. 1731 */ 1732 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1733 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1734 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1735 "segment processed normally\n", s, __func__); 1736 free(s, M_TCPLOG); 1737 } 1738 } 1739 1740 /* 1741 * Header prediction: check for the two common cases 1742 * of a uni-directional data xfer. If the packet has 1743 * no control flags, is in-sequence, the window didn't 1744 * change and we're not retransmitting, it's a 1745 * candidate. If the length is zero and the ack moved 1746 * forward, we're the sender side of the xfer. Just 1747 * free the data acked & wake any higher level process 1748 * that was blocked waiting for space. If the length 1749 * is non-zero and the ack didn't move, we're the 1750 * receiver side. If we're getting packets in-order 1751 * (the reassembly queue is empty), add the data to 1752 * the socket buffer and note that we need a delayed ack. 1753 * Make sure that the hidden state-flags are also off. 1754 * Since we check for TCPS_ESTABLISHED first, it can only 1755 * be TH_NEEDSYN. 1756 */ 1757 if (tp->t_state == TCPS_ESTABLISHED && 1758 th->th_seq == tp->rcv_nxt && 1759 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1760 tp->snd_nxt == tp->snd_max && 1761 tiwin && tiwin == tp->snd_wnd && 1762 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1763 SEGQ_EMPTY(tp) && 1764 ((to.to_flags & TOF_TS) == 0 || 1765 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1766 /* 1767 * If last ACK falls within this segment's sequence numbers, 1768 * record the timestamp. 1769 * NOTE that the test is modified according to the latest 1770 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1771 */ 1772 if ((to.to_flags & TOF_TS) != 0 && 1773 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1774 tp->ts_recent_age = tcp_ts_getticks(); 1775 tp->ts_recent = to.to_tsval; 1776 } 1777 1778 if (no_data) { 1779 if (SEQ_GT(th->th_ack, tp->snd_una) && 1780 SEQ_LEQ(th->th_ack, tp->snd_max) && 1781 !IN_RECOVERY(tp->t_flags) && 1782 (to.to_flags & TOF_SACK) == 0 && 1783 TAILQ_EMPTY(&tp->snd_holes)) { 1784 /* 1785 * This is a pure ack for outstanding data. 1786 */ 1787 TCPSTAT_INC(tcps_predack); 1788 1789 /* 1790 * "bad retransmit" recovery. 1791 */ 1792 if (tp->t_rxtshift == 1 && 1793 tp->t_flags & TF_PREVVALID && 1794 tp->t_badrxtwin != 0 && 1795 (((to.to_flags & TOF_TS) != 0 && 1796 to.to_tsecr != 0 && 1797 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) || 1798 ((to.to_flags & TOF_TS) == 0 && 1799 TSTMP_LT(ticks, tp->t_badrxtwin)))) 1800 cc_cong_signal(tp, th, CC_RTO_ERR); 1801 1802 /* 1803 * Recalculate the transmit timer / rtt. 1804 * 1805 * Some boxes send broken timestamp replies 1806 * during the SYN+ACK phase, ignore 1807 * timestamps of 0 or we could calculate a 1808 * huge RTT and blow up the retransmit timer. 1809 */ 1810 if ((to.to_flags & TOF_TS) != 0 && 1811 to.to_tsecr) { 1812 uint32_t t; 1813 1814 t = tcp_ts_getticks() - to.to_tsecr; 1815 if (!tp->t_rttlow || tp->t_rttlow > t) 1816 tp->t_rttlow = t; 1817 tcp_xmit_timer(tp, 1818 TCP_TS_TO_TICKS(t) + 1); 1819 } else if (tp->t_rtttime && 1820 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1821 if (!tp->t_rttlow || 1822 tp->t_rttlow > ticks - tp->t_rtttime) 1823 tp->t_rttlow = ticks - tp->t_rtttime; 1824 tcp_xmit_timer(tp, 1825 ticks - tp->t_rtttime); 1826 } 1827 acked = BYTES_THIS_ACK(tp, th); 1828 1829 #ifdef TCP_HHOOK 1830 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1831 hhook_run_tcp_est_in(tp, th, &to); 1832 #endif 1833 1834 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1835 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1836 sbdrop(&so->so_snd, acked); 1837 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1838 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1839 tp->snd_recover = th->th_ack - 1; 1840 1841 /* 1842 * Let the congestion control algorithm update 1843 * congestion control related information. This 1844 * typically means increasing the congestion 1845 * window. 1846 */ 1847 cc_ack_received(tp, th, nsegs, CC_ACK); 1848 1849 tp->snd_una = th->th_ack; 1850 /* 1851 * Pull snd_wl2 up to prevent seq wrap relative 1852 * to th_ack. 1853 */ 1854 tp->snd_wl2 = th->th_ack; 1855 tp->t_dupacks = 0; 1856 m_freem(m); 1857 1858 /* 1859 * If all outstanding data are acked, stop 1860 * retransmit timer, otherwise restart timer 1861 * using current (possibly backed-off) value. 1862 * If process is waiting for space, 1863 * wakeup/selwakeup/signal. If data 1864 * are ready to send, let tcp_output 1865 * decide between more output or persist. 1866 */ 1867 TCP_PROBE3(debug__input, tp, th, m); 1868 /* 1869 * Clear t_acktime if remote side has ACKd 1870 * all data in the socket buffer. 1871 * Otherwise, update t_acktime if we received 1872 * a sufficiently large ACK. 1873 */ 1874 if (sbavail(&so->so_snd) == 0) 1875 tp->t_acktime = 0; 1876 else if (acked > 1) 1877 tp->t_acktime = ticks; 1878 if (tp->snd_una == tp->snd_max) 1879 tcp_timer_activate(tp, TT_REXMT, 0); 1880 else if (!tcp_timer_active(tp, TT_PERSIST)) 1881 tcp_timer_activate(tp, TT_REXMT, 1882 TP_RXTCUR(tp)); 1883 sowwakeup(so); 1884 /* 1885 * Only call tcp_output when there 1886 * is new data available to be sent 1887 * or we need to send an ACK. 1888 */ 1889 if ((tp->t_flags & TF_ACKNOW) || 1890 (sbavail(&so->so_snd) >= 1891 SEQ_SUB(tp->snd_max, tp->snd_una))) { 1892 (void) tcp_output(tp); 1893 } 1894 goto check_delack; 1895 } 1896 } else if (th->th_ack == tp->snd_una && 1897 tlen <= sbspace(&so->so_rcv)) { 1898 int newsize = 0; /* automatic sockbuf scaling */ 1899 1900 /* 1901 * This is a pure, in-sequence data packet with 1902 * nothing on the reassembly queue and we have enough 1903 * buffer space to take it. 1904 */ 1905 /* Clean receiver SACK report if present */ 1906 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1907 tcp_clean_sackreport(tp); 1908 TCPSTAT_INC(tcps_preddat); 1909 tp->rcv_nxt += tlen; 1910 if (tlen && 1911 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1912 (tp->t_fbyte_in == 0)) { 1913 tp->t_fbyte_in = ticks; 1914 if (tp->t_fbyte_in == 0) 1915 tp->t_fbyte_in = 1; 1916 if (tp->t_fbyte_out && tp->t_fbyte_in) 1917 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1918 } 1919 /* 1920 * Pull snd_wl1 up to prevent seq wrap relative to 1921 * th_seq. 1922 */ 1923 tp->snd_wl1 = th->th_seq; 1924 /* 1925 * Pull rcv_up up to prevent seq wrap relative to 1926 * rcv_nxt. 1927 */ 1928 tp->rcv_up = tp->rcv_nxt; 1929 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1930 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1931 TCP_PROBE3(debug__input, tp, th, m); 1932 1933 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1934 1935 /* Add data to socket buffer. */ 1936 SOCK_RECVBUF_LOCK(so); 1937 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1938 m_freem(m); 1939 } else { 1940 /* 1941 * Set new socket buffer size. 1942 * Give up when limit is reached. 1943 */ 1944 if (newsize) 1945 if (!sbreserve_locked(so, SO_RCV, 1946 newsize, NULL)) 1947 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1948 m_adj(m, drop_hdrlen); /* delayed header drop */ 1949 sbappendstream_locked(&so->so_rcv, m, 0); 1950 } 1951 /* NB: sorwakeup_locked() does an implicit unlock. */ 1952 sorwakeup_locked(so); 1953 if (DELAY_ACK(tp, tlen)) { 1954 tp->t_flags |= TF_DELACK; 1955 } else { 1956 tp->t_flags |= TF_ACKNOW; 1957 (void) tcp_output(tp); 1958 } 1959 goto check_delack; 1960 } 1961 } 1962 1963 /* 1964 * Calculate amount of space in receive window, 1965 * and then do TCP input processing. 1966 * Receive window is amount of space in rcv queue, 1967 * but not less than advertised window. 1968 */ 1969 win = sbspace(&so->so_rcv); 1970 if (win < 0) 1971 win = 0; 1972 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1973 1974 switch (tp->t_state) { 1975 /* 1976 * If the state is SYN_RECEIVED: 1977 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1978 */ 1979 case TCPS_SYN_RECEIVED: 1980 if (thflags & TH_RST) { 1981 /* Handle RST segments later. */ 1982 break; 1983 } 1984 if ((thflags & TH_ACK) && 1985 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1986 SEQ_GT(th->th_ack, tp->snd_max))) { 1987 rstreason = BANDLIM_TCP_RST; 1988 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1989 goto dropwithreset; 1990 } 1991 if (tp->t_flags & TF_FASTOPEN) { 1992 /* 1993 * When a TFO connection is in SYN_RECEIVED, the 1994 * only valid packets are the initial SYN, a 1995 * retransmit/copy of the initial SYN (possibly with 1996 * a subset of the original data), a valid ACK, a 1997 * FIN, or a RST. 1998 */ 1999 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 2000 rstreason = BANDLIM_TCP_RST; 2001 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2002 goto dropwithreset; 2003 } else if (thflags & TH_SYN) { 2004 /* non-initial SYN is ignored */ 2005 if ((tcp_timer_active(tp, TT_DELACK) || 2006 tcp_timer_active(tp, TT_REXMT))) 2007 goto drop; 2008 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 2009 goto drop; 2010 } 2011 } 2012 break; 2013 2014 /* 2015 * If the state is SYN_SENT: 2016 * if seg contains a RST with valid ACK (SEQ.ACK has already 2017 * been verified), then drop the connection. 2018 * if seg contains a RST without an ACK, drop the seg. 2019 * if seg does not contain SYN, then drop the seg. 2020 * Otherwise this is an acceptable SYN segment 2021 * initialize tp->rcv_nxt and tp->irs 2022 * if seg contains ack then advance tp->snd_una 2023 * if seg contains an ECE and ECN support is enabled, the stream 2024 * is ECN capable. 2025 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2026 * arrange for segment to be acked (eventually) 2027 * continue processing rest of data/controls, beginning with URG 2028 */ 2029 case TCPS_SYN_SENT: 2030 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2031 TCP_PROBE5(connect__refused, NULL, tp, 2032 m, tp, th); 2033 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2034 tp = tcp_drop(tp, ECONNREFUSED); 2035 } 2036 if (thflags & TH_RST) 2037 goto drop; 2038 if (!(thflags & TH_SYN)) 2039 goto drop; 2040 2041 tp->irs = th->th_seq; 2042 tcp_rcvseqinit(tp); 2043 if (thflags & TH_ACK) { 2044 int tfo_partial_ack = 0; 2045 2046 TCPSTAT_INC(tcps_connects); 2047 soisconnected(so); 2048 #ifdef MAC 2049 mac_socketpeer_set_from_mbuf(m, so); 2050 #endif 2051 /* Do window scaling on this connection? */ 2052 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2053 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2054 tp->rcv_scale = tp->request_r_scale; 2055 } 2056 tp->rcv_adv += min(tp->rcv_wnd, 2057 TCP_MAXWIN << tp->rcv_scale); 2058 tp->snd_una++; /* SYN is acked */ 2059 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2060 tp->snd_nxt = tp->snd_una; 2061 /* 2062 * If not all the data that was sent in the TFO SYN 2063 * has been acked, resend the remainder right away. 2064 */ 2065 if ((tp->t_flags & TF_FASTOPEN) && 2066 (tp->snd_una != tp->snd_max)) { 2067 tp->snd_nxt = th->th_ack; 2068 tfo_partial_ack = 1; 2069 } 2070 /* 2071 * If there's data, delay ACK; if there's also a FIN 2072 * ACKNOW will be turned on later. 2073 */ 2074 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2075 tcp_timer_activate(tp, TT_DELACK, 2076 tcp_delacktime); 2077 else 2078 tp->t_flags |= TF_ACKNOW; 2079 2080 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2081 2082 /* 2083 * Received <SYN,ACK> in SYN_SENT[*] state. 2084 * Transitions: 2085 * SYN_SENT --> ESTABLISHED 2086 * SYN_SENT* --> FIN_WAIT_1 2087 */ 2088 tp->t_starttime = ticks; 2089 if (tp->t_flags & TF_NEEDFIN) { 2090 tp->t_acktime = ticks; 2091 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2092 tp->t_flags &= ~TF_NEEDFIN; 2093 thflags &= ~TH_SYN; 2094 } else { 2095 tcp_state_change(tp, TCPS_ESTABLISHED); 2096 TCP_PROBE5(connect__established, NULL, tp, 2097 m, tp, th); 2098 cc_conn_init(tp); 2099 tcp_timer_activate(tp, TT_KEEP, 2100 TP_KEEPIDLE(tp)); 2101 } 2102 } else { 2103 /* 2104 * Received initial SYN in SYN-SENT[*] state => 2105 * simultaneous open. 2106 * If it succeeds, connection is * half-synchronized. 2107 * Otherwise, do 3-way handshake: 2108 * SYN-SENT -> SYN-RECEIVED 2109 * SYN-SENT* -> SYN-RECEIVED* 2110 */ 2111 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN); 2112 tcp_timer_activate(tp, TT_REXMT, 0); 2113 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2114 } 2115 2116 /* 2117 * Advance th->th_seq to correspond to first data byte. 2118 * If data, trim to stay within window, 2119 * dropping FIN if necessary. 2120 */ 2121 th->th_seq++; 2122 if (tlen > tp->rcv_wnd) { 2123 todrop = tlen - tp->rcv_wnd; 2124 m_adj(m, -todrop); 2125 tlen = tp->rcv_wnd; 2126 thflags &= ~TH_FIN; 2127 TCPSTAT_INC(tcps_rcvpackafterwin); 2128 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2129 } 2130 tp->snd_wl1 = th->th_seq - 1; 2131 tp->rcv_up = th->th_seq; 2132 /* 2133 * Client side of transaction: already sent SYN and data. 2134 * If the remote host used T/TCP to validate the SYN, 2135 * our data will be ACK'd; if so, enter normal data segment 2136 * processing in the middle of step 5, ack processing. 2137 * Otherwise, goto step 6. 2138 */ 2139 if (thflags & TH_ACK) 2140 goto process_ACK; 2141 2142 goto step6; 2143 } 2144 2145 /* 2146 * States other than LISTEN or SYN_SENT. 2147 * First check the RST flag and sequence number since reset segments 2148 * are exempt from the timestamp and connection count tests. This 2149 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2150 * below which allowed reset segments in half the sequence space 2151 * to fall though and be processed (which gives forged reset 2152 * segments with a random sequence number a 50 percent chance of 2153 * killing a connection). 2154 * Then check timestamp, if present. 2155 * Then check the connection count, if present. 2156 * Then check that at least some bytes of segment are within 2157 * receive window. If segment begins before rcv_nxt, 2158 * drop leading data (and SYN); if nothing left, just ack. 2159 */ 2160 if (thflags & TH_RST) { 2161 /* 2162 * RFC5961 Section 3.2 2163 * 2164 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2165 * - If RST is in window, we send challenge ACK. 2166 * 2167 * Note: to take into account delayed ACKs, we should 2168 * test against last_ack_sent instead of rcv_nxt. 2169 * Note 2: we handle special case of closed window, not 2170 * covered by the RFC. 2171 */ 2172 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2173 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2174 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2175 KASSERT(tp->t_state != TCPS_SYN_SENT, 2176 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2177 __func__, th, tp)); 2178 2179 if (V_tcp_insecure_rst || 2180 tp->last_ack_sent == th->th_seq) { 2181 TCPSTAT_INC(tcps_drops); 2182 /* Drop the connection. */ 2183 switch (tp->t_state) { 2184 case TCPS_SYN_RECEIVED: 2185 so->so_error = ECONNREFUSED; 2186 goto close; 2187 case TCPS_ESTABLISHED: 2188 case TCPS_FIN_WAIT_1: 2189 case TCPS_FIN_WAIT_2: 2190 case TCPS_CLOSE_WAIT: 2191 case TCPS_CLOSING: 2192 case TCPS_LAST_ACK: 2193 so->so_error = ECONNRESET; 2194 close: 2195 /* FALLTHROUGH */ 2196 default: 2197 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST); 2198 tp = tcp_close(tp); 2199 } 2200 } else { 2201 TCPSTAT_INC(tcps_badrst); 2202 tcp_send_challenge_ack(tp, th, m); 2203 m = NULL; 2204 } 2205 } 2206 goto drop; 2207 } 2208 2209 /* 2210 * RFC5961 Section 4.2 2211 * Send challenge ACK for any SYN in synchronized state. 2212 */ 2213 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2214 tp->t_state != TCPS_SYN_RECEIVED) { 2215 TCPSTAT_INC(tcps_badsyn); 2216 if (V_tcp_insecure_syn && 2217 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2218 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2219 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2220 tp = tcp_drop(tp, ECONNRESET); 2221 rstreason = BANDLIM_UNLIMITED; 2222 } else { 2223 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2224 tcp_send_challenge_ack(tp, th, m); 2225 m = NULL; 2226 } 2227 goto drop; 2228 } 2229 2230 /* 2231 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2232 * and it's less than ts_recent, drop it. 2233 */ 2234 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2235 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2236 /* Check to see if ts_recent is over 24 days old. */ 2237 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2238 /* 2239 * Invalidate ts_recent. If this segment updates 2240 * ts_recent, the age will be reset later and ts_recent 2241 * will get a valid value. If it does not, setting 2242 * ts_recent to zero will at least satisfy the 2243 * requirement that zero be placed in the timestamp 2244 * echo reply when ts_recent isn't valid. The 2245 * age isn't reset until we get a valid ts_recent 2246 * because we don't want out-of-order segments to be 2247 * dropped when ts_recent is old. 2248 */ 2249 tp->ts_recent = 0; 2250 } else { 2251 TCPSTAT_INC(tcps_rcvduppack); 2252 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2253 TCPSTAT_INC(tcps_pawsdrop); 2254 if (tlen) 2255 goto dropafterack; 2256 goto drop; 2257 } 2258 } 2259 2260 /* 2261 * In the SYN-RECEIVED state, validate that the packet belongs to 2262 * this connection before trimming the data to fit the receive 2263 * window. Check the sequence number versus IRS since we know 2264 * the sequence numbers haven't wrapped. This is a partial fix 2265 * for the "LAND" DoS attack. 2266 */ 2267 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2268 rstreason = BANDLIM_TCP_RST; 2269 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2270 goto dropwithreset; 2271 } 2272 2273 todrop = tp->rcv_nxt - th->th_seq; 2274 if (todrop > 0) { 2275 if (thflags & TH_SYN) { 2276 thflags &= ~TH_SYN; 2277 th->th_seq++; 2278 if (th->th_urp > 1) 2279 th->th_urp--; 2280 else 2281 thflags &= ~TH_URG; 2282 todrop--; 2283 } 2284 /* 2285 * Following if statement from Stevens, vol. 2, p. 960. 2286 */ 2287 if (todrop > tlen 2288 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2289 /* 2290 * Any valid FIN must be to the left of the window. 2291 * At this point the FIN must be a duplicate or out 2292 * of sequence; drop it. 2293 */ 2294 thflags &= ~TH_FIN; 2295 2296 /* 2297 * Send an ACK to resynchronize and drop any data. 2298 * But keep on processing for RST or ACK. 2299 */ 2300 tp->t_flags |= TF_ACKNOW; 2301 todrop = tlen; 2302 TCPSTAT_INC(tcps_rcvduppack); 2303 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2304 } else { 2305 TCPSTAT_INC(tcps_rcvpartduppack); 2306 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2307 } 2308 /* 2309 * DSACK - add SACK block for dropped range 2310 */ 2311 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2312 tcp_update_sack_list(tp, th->th_seq, 2313 th->th_seq + todrop); 2314 /* 2315 * ACK now, as the next in-sequence segment 2316 * will clear the DSACK block again 2317 */ 2318 tp->t_flags |= TF_ACKNOW; 2319 } 2320 drop_hdrlen += todrop; /* drop from the top afterwards */ 2321 th->th_seq += todrop; 2322 tlen -= todrop; 2323 if (th->th_urp > todrop) 2324 th->th_urp -= todrop; 2325 else { 2326 thflags &= ~TH_URG; 2327 th->th_urp = 0; 2328 } 2329 } 2330 2331 /* 2332 * If new data are received on a connection after the 2333 * user processes are gone, then RST the other end if 2334 * no FIN has been processed. 2335 */ 2336 if ((tp->t_flags & TF_CLOSED) && tlen > 0 && 2337 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2338 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2339 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2340 "after socket was closed, " 2341 "sending RST and removing tcpcb\n", 2342 s, __func__, tcpstates[tp->t_state], tlen); 2343 free(s, M_TCPLOG); 2344 } 2345 tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE); 2346 /* tcp_close will kill the inp pre-log the Reset */ 2347 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST); 2348 tp = tcp_close(tp); 2349 TCPSTAT_INC(tcps_rcvafterclose); 2350 rstreason = BANDLIM_UNLIMITED; 2351 goto dropwithreset; 2352 } 2353 2354 /* 2355 * If segment ends after window, drop trailing data 2356 * (and PUSH and FIN); if nothing left, just ACK. 2357 */ 2358 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2359 if (todrop > 0) { 2360 TCPSTAT_INC(tcps_rcvpackafterwin); 2361 if (todrop >= tlen) { 2362 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2363 /* 2364 * If window is closed can only take segments at 2365 * window edge, and have to drop data and PUSH from 2366 * incoming segments. Continue processing, but 2367 * remember to ack. Otherwise, drop segment 2368 * and ack. 2369 */ 2370 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2371 tp->t_flags |= TF_ACKNOW; 2372 TCPSTAT_INC(tcps_rcvwinprobe); 2373 } else 2374 goto dropafterack; 2375 } else 2376 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2377 m_adj(m, -todrop); 2378 tlen -= todrop; 2379 thflags &= ~(TH_PUSH|TH_FIN); 2380 } 2381 2382 /* 2383 * If last ACK falls within this segment's sequence numbers, 2384 * record its timestamp. 2385 * NOTE: 2386 * 1) That the test incorporates suggestions from the latest 2387 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2388 * 2) That updating only on newer timestamps interferes with 2389 * our earlier PAWS tests, so this check should be solely 2390 * predicated on the sequence space of this segment. 2391 * 3) That we modify the segment boundary check to be 2392 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2393 * instead of RFC1323's 2394 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2395 * This modified check allows us to overcome RFC1323's 2396 * limitations as described in Stevens TCP/IP Illustrated 2397 * Vol. 2 p.869. In such cases, we can still calculate the 2398 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2399 */ 2400 if ((to.to_flags & TOF_TS) != 0 && 2401 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2402 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2403 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2404 tp->ts_recent_age = tcp_ts_getticks(); 2405 tp->ts_recent = to.to_tsval; 2406 } 2407 2408 /* 2409 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2410 * flag is on (half-synchronized state), then queue data for 2411 * later processing; else drop segment and return. 2412 */ 2413 if ((thflags & TH_ACK) == 0) { 2414 if (tp->t_state == TCPS_SYN_RECEIVED || 2415 (tp->t_flags & TF_NEEDSYN)) { 2416 if (tp->t_state == TCPS_SYN_RECEIVED && 2417 (tp->t_flags & TF_FASTOPEN)) { 2418 tp->snd_wnd = tiwin; 2419 cc_conn_init(tp); 2420 } 2421 goto step6; 2422 } else if (tp->t_flags & TF_ACKNOW) 2423 goto dropafterack; 2424 else 2425 goto drop; 2426 } 2427 2428 /* 2429 * Ack processing. 2430 */ 2431 if (SEQ_GEQ(tp->snd_una, tp->iss + (TCP_MAXWIN << tp->snd_scale))) { 2432 /* Checking SEG.ACK against ISS is definitely redundant. */ 2433 tp->t_flags2 |= TF2_NO_ISS_CHECK; 2434 } 2435 if (!V_tcp_insecure_ack) { 2436 tcp_seq seq_min; 2437 bool ghost_ack_check; 2438 2439 if (tp->t_flags2 & TF2_NO_ISS_CHECK) { 2440 /* Check for too old ACKs (RFC 5961, Section 5.2). */ 2441 seq_min = tp->snd_una - tp->max_sndwnd; 2442 ghost_ack_check = false; 2443 } else { 2444 if (SEQ_GT(tp->iss + 1, tp->snd_una - tp->max_sndwnd)) { 2445 /* Checking for ghost ACKs is stricter. */ 2446 seq_min = tp->iss + 1; 2447 ghost_ack_check = true; 2448 } else { 2449 /* 2450 * Checking for too old ACKs (RFC 5961, 2451 * Section 5.2) is stricter. 2452 */ 2453 seq_min = tp->snd_una - tp->max_sndwnd; 2454 ghost_ack_check = false; 2455 } 2456 } 2457 if (SEQ_LT(th->th_ack, seq_min)) { 2458 if (ghost_ack_check) 2459 TCPSTAT_INC(tcps_rcvghostack); 2460 else 2461 TCPSTAT_INC(tcps_rcvacktooold); 2462 tcp_send_challenge_ack(tp, th, m); 2463 m = NULL; 2464 goto drop; 2465 } 2466 } 2467 switch (tp->t_state) { 2468 /* 2469 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2470 * ESTABLISHED state and continue processing. 2471 * The ACK was checked above. 2472 */ 2473 case TCPS_SYN_RECEIVED: 2474 2475 TCPSTAT_INC(tcps_connects); 2476 if (tp->t_flags & TF_SONOTCONN) { 2477 /* 2478 * Usually SYN_RECEIVED had been created from a LISTEN, 2479 * and solisten_enqueue() has already marked the socket 2480 * layer as connected. If it didn't, which can happen 2481 * only with an accept_filter(9), then the tp is marked 2482 * with TF_SONOTCONN. The other reason for this mark 2483 * to be set is a simultaneous open, a SYN_RECEIVED 2484 * that had been created from SYN_SENT. 2485 */ 2486 tp->t_flags &= ~TF_SONOTCONN; 2487 soisconnected(so); 2488 } 2489 /* Do window scaling? */ 2490 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2491 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2492 tp->rcv_scale = tp->request_r_scale; 2493 } 2494 tp->snd_wnd = tiwin; 2495 /* 2496 * Make transitions: 2497 * SYN-RECEIVED -> ESTABLISHED 2498 * SYN-RECEIVED* -> FIN-WAIT-1 2499 */ 2500 tp->t_starttime = ticks; 2501 if ((tp->t_flags & TF_FASTOPEN) && tp->t_tfo_pending) { 2502 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2503 tp->t_tfo_pending = NULL; 2504 } 2505 if (tp->t_flags & TF_NEEDFIN) { 2506 tp->t_acktime = ticks; 2507 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2508 tp->t_flags &= ~TF_NEEDFIN; 2509 } else { 2510 tcp_state_change(tp, TCPS_ESTABLISHED); 2511 TCP_PROBE5(accept__established, NULL, tp, 2512 m, tp, th); 2513 /* 2514 * TFO connections call cc_conn_init() during SYN 2515 * processing. Calling it again here for such 2516 * connections is not harmless as it would undo the 2517 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2518 * is retransmitted. 2519 */ 2520 if (!(tp->t_flags & TF_FASTOPEN)) 2521 cc_conn_init(tp); 2522 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2523 } 2524 /* 2525 * Account for the ACK of our SYN prior to 2526 * regular ACK processing below, except for 2527 * simultaneous SYN, which is handled later. 2528 */ 2529 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2530 incforsyn = 1; 2531 /* 2532 * If segment contains data or ACK, will call tcp_reass() 2533 * later; if not, do so now to pass queued data to user. 2534 */ 2535 if (tlen == 0 && (thflags & TH_FIN) == 0) { 2536 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2537 (struct mbuf *)0); 2538 tcp_handle_wakeup(tp); 2539 } 2540 tp->snd_wl1 = th->th_seq - 1; 2541 /* FALLTHROUGH */ 2542 2543 /* 2544 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2545 * ACKs. If the ack is in the range 2546 * tp->snd_una < th->th_ack <= tp->snd_max 2547 * then advance tp->snd_una to th->th_ack and drop 2548 * data from the retransmission queue. If this ACK reflects 2549 * more up to date window information we update our window information. 2550 */ 2551 case TCPS_ESTABLISHED: 2552 case TCPS_FIN_WAIT_1: 2553 case TCPS_FIN_WAIT_2: 2554 case TCPS_CLOSE_WAIT: 2555 case TCPS_CLOSING: 2556 case TCPS_LAST_ACK: 2557 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2558 TCPSTAT_INC(tcps_rcvacktoomuch); 2559 goto dropafterack; 2560 } 2561 if (tcp_is_sack_recovery(tp, &to)) { 2562 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2563 if ((sack_changed != SACK_NOCHANGE) && 2564 (tp->t_flags & TF_LRD)) { 2565 tcp_sack_lost_retransmission(tp, th); 2566 } 2567 } else 2568 /* 2569 * Reset the value so that previous (valid) value 2570 * from the last ack with SACK doesn't get used. 2571 */ 2572 tp->sackhint.sacked_bytes = 0; 2573 2574 #ifdef TCP_HHOOK 2575 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2576 hhook_run_tcp_est_in(tp, th, &to); 2577 #endif 2578 2579 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2580 maxseg = tcp_maxseg(tp); 2581 if (no_data && 2582 (tiwin == tp->snd_wnd || 2583 (tp->t_flags & TF_SACK_PERMIT))) { 2584 /* 2585 * If this is the first time we've seen a 2586 * FIN from the remote, this is not a 2587 * duplicate and it needs to be processed 2588 * normally. This happens during a 2589 * simultaneous close. 2590 */ 2591 if ((thflags & TH_FIN) && 2592 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2593 tp->t_dupacks = 0; 2594 break; 2595 } 2596 TCPSTAT_INC(tcps_rcvdupack); 2597 /* 2598 * If we have outstanding data (other than 2599 * a window probe), this is a completely 2600 * duplicate ack (ie, window info didn't 2601 * change and FIN isn't set), 2602 * the ack is the biggest we've 2603 * seen and we've seen exactly our rexmt 2604 * threshold of them, assume a packet 2605 * has been dropped and retransmit it. 2606 * Kludge snd_nxt & the congestion 2607 * window so we send only this one 2608 * packet. 2609 * 2610 * We know we're losing at the current 2611 * window size so do congestion avoidance 2612 * (set ssthresh to half the current window 2613 * and pull our congestion window back to 2614 * the new ssthresh). 2615 * 2616 * Dup acks mean that packets have left the 2617 * network (they're now cached at the receiver) 2618 * so bump cwnd by the amount in the receiver 2619 * to keep a constant cwnd packets in the 2620 * network. 2621 * 2622 * When using TCP ECN, notify the peer that 2623 * we reduced the cwnd. 2624 */ 2625 /* 2626 * Following 2 kinds of acks should not affect 2627 * dupack counting: 2628 * 1) Old acks 2629 * 2) Acks with SACK but without any new SACK 2630 * information in them. These could result from 2631 * any anomaly in the network like a switch 2632 * duplicating packets or a possible DoS attack. 2633 */ 2634 if (th->th_ack != tp->snd_una || 2635 (tcp_is_sack_recovery(tp, &to) && 2636 (sack_changed == SACK_NOCHANGE))) { 2637 break; 2638 } else if (!tcp_timer_active(tp, TT_REXMT)) { 2639 tp->t_dupacks = 0; 2640 } else if (++tp->t_dupacks > tcprexmtthresh || 2641 IN_FASTRECOVERY(tp->t_flags)) { 2642 cc_ack_received(tp, th, nsegs, 2643 CC_DUPACK); 2644 if (V_tcp_do_prr && 2645 IN_FASTRECOVERY(tp->t_flags) && 2646 (tp->t_flags & TF_SACK_PERMIT)) { 2647 tcp_do_prr_ack(tp, th, &to, 2648 sack_changed, &maxseg); 2649 } else if (tcp_is_sack_recovery(tp, &to) && 2650 IN_FASTRECOVERY(tp->t_flags) && 2651 (tp->snd_nxt == tp->snd_max)) { 2652 int awnd; 2653 2654 /* 2655 * Compute the amount of data in flight first. 2656 * We can inject new data into the pipe iff 2657 * we have less than ssthresh 2658 * worth of data in flight. 2659 */ 2660 awnd = tcp_compute_pipe(tp); 2661 if (awnd < tp->snd_ssthresh) { 2662 tp->snd_cwnd += imax(maxseg, 2663 imin(2 * maxseg, 2664 tp->sackhint.delivered_data)); 2665 if (tp->snd_cwnd > tp->snd_ssthresh) 2666 tp->snd_cwnd = tp->snd_ssthresh; 2667 } 2668 } else if (tcp_is_sack_recovery(tp, &to) && 2669 IN_FASTRECOVERY(tp->t_flags) && 2670 SEQ_LT(tp->snd_nxt, tp->snd_max)) { 2671 tp->snd_cwnd += imax(maxseg, 2672 imin(2 * maxseg, 2673 tp->sackhint.delivered_data)); 2674 } else { 2675 tp->snd_cwnd += maxseg; 2676 } 2677 (void) tcp_output(tp); 2678 goto drop; 2679 } else if (tp->t_dupacks == tcprexmtthresh || 2680 (tp->t_flags & TF_SACK_PERMIT && 2681 V_tcp_do_newsack && 2682 tp->sackhint.sacked_bytes > 2683 (tcprexmtthresh - 1) * maxseg)) { 2684 enter_recovery: 2685 /* 2686 * Above is the RFC6675 trigger condition of 2687 * more than (dupthresh-1)*maxseg sacked data. 2688 * If the count of holes in the 2689 * scoreboard is >= dupthresh, we could 2690 * also enter loss recovery, but don't 2691 * have that value readily available. 2692 */ 2693 tp->t_dupacks = tcprexmtthresh; 2694 tcp_seq onxt = tp->snd_nxt; 2695 2696 /* 2697 * If we're doing sack, check to 2698 * see if we're already in sack 2699 * recovery. If we're not doing sack, 2700 * check to see if we're in newreno 2701 * recovery. 2702 */ 2703 if (tcp_is_sack_recovery(tp, &to)) { 2704 if (IN_FASTRECOVERY(tp->t_flags)) { 2705 tp->t_dupacks = 0; 2706 break; 2707 } 2708 } else { 2709 if (SEQ_LEQ(th->th_ack, 2710 tp->snd_recover)) { 2711 tp->t_dupacks = 0; 2712 break; 2713 } 2714 } 2715 /* Congestion signal before ack. */ 2716 cc_cong_signal(tp, th, CC_NDUPACK); 2717 cc_ack_received(tp, th, nsegs, 2718 CC_DUPACK); 2719 tcp_timer_activate(tp, TT_REXMT, 0); 2720 tp->t_rtttime = 0; 2721 if (V_tcp_do_prr) { 2722 /* 2723 * snd_ssthresh and snd_recover are 2724 * already updated by cc_cong_signal. 2725 */ 2726 if (tcp_is_sack_recovery(tp, &to)) { 2727 /* 2728 * Include Limited Transmit 2729 * segments here 2730 */ 2731 tp->sackhint.prr_delivered = 2732 imin(tp->snd_max - th->th_ack, 2733 (tp->snd_limited + 1) * maxseg); 2734 } else { 2735 tp->sackhint.prr_delivered = 2736 maxseg; 2737 } 2738 tp->sackhint.recover_fs = max(1, 2739 tp->snd_nxt - tp->snd_una); 2740 } 2741 tp->snd_limited = 0; 2742 if (tcp_is_sack_recovery(tp, &to)) { 2743 TCPSTAT_INC(tcps_sack_recovery_episode); 2744 /* 2745 * When entering LR after RTO due to 2746 * Duplicate ACKs, retransmit existing 2747 * holes from the scoreboard. 2748 */ 2749 tcp_resend_sackholes(tp); 2750 /* Avoid inflating cwnd in tcp_output */ 2751 tp->snd_nxt = tp->snd_max; 2752 tp->snd_cwnd = tcp_compute_pipe(tp) + 2753 maxseg; 2754 (void) tcp_output(tp); 2755 /* Set cwnd to the expected flightsize */ 2756 tp->snd_cwnd = tp->snd_ssthresh; 2757 if (SEQ_GT(th->th_ack, tp->snd_una)) { 2758 goto resume_partialack; 2759 } 2760 goto drop; 2761 } 2762 tp->snd_nxt = th->th_ack; 2763 tp->snd_cwnd = maxseg; 2764 (void) tcp_output(tp); 2765 KASSERT(tp->snd_limited <= 2, 2766 ("%s: tp->snd_limited too big", 2767 __func__)); 2768 tp->snd_cwnd = tp->snd_ssthresh + 2769 maxseg * 2770 (tp->t_dupacks - tp->snd_limited); 2771 if (SEQ_GT(onxt, tp->snd_nxt)) 2772 tp->snd_nxt = onxt; 2773 goto drop; 2774 } else if (V_tcp_do_rfc3042) { 2775 /* 2776 * Process first and second duplicate 2777 * ACKs. Each indicates a segment 2778 * leaving the network, creating room 2779 * for more. Make sure we can send a 2780 * packet on reception of each duplicate 2781 * ACK by increasing snd_cwnd by one 2782 * segment. Restore the original 2783 * snd_cwnd after packet transmission. 2784 */ 2785 cc_ack_received(tp, th, nsegs, CC_DUPACK); 2786 uint32_t oldcwnd = tp->snd_cwnd; 2787 tcp_seq oldsndmax = tp->snd_max; 2788 u_int sent; 2789 int avail; 2790 2791 KASSERT(tp->t_dupacks == 1 || 2792 tp->t_dupacks == 2, 2793 ("%s: dupacks not 1 or 2", 2794 __func__)); 2795 if (tp->t_dupacks == 1) 2796 tp->snd_limited = 0; 2797 if ((tp->snd_nxt == tp->snd_max) && 2798 (tp->t_rxtshift == 0)) 2799 tp->snd_cwnd = 2800 SEQ_SUB(tp->snd_nxt, 2801 tp->snd_una) - 2802 tcp_sack_adjust(tp); 2803 tp->snd_cwnd += 2804 (tp->t_dupacks - tp->snd_limited) * 2805 maxseg - tcp_sack_adjust(tp); 2806 /* 2807 * Only call tcp_output when there 2808 * is new data available to be sent 2809 * or we need to send an ACK. 2810 */ 2811 SOCK_SENDBUF_LOCK(so); 2812 avail = sbavail(&so->so_snd); 2813 SOCK_SENDBUF_UNLOCK(so); 2814 if (tp->t_flags & TF_ACKNOW || 2815 (avail >= 2816 SEQ_SUB(tp->snd_nxt, tp->snd_una))) { 2817 (void) tcp_output(tp); 2818 } 2819 sent = SEQ_SUB(tp->snd_max, oldsndmax); 2820 if (sent > maxseg) { 2821 KASSERT((tp->t_dupacks == 2 && 2822 tp->snd_limited == 0) || 2823 (sent == maxseg + 1 && 2824 tp->t_flags & TF_SENTFIN) || 2825 (sent < 2 * maxseg && 2826 tp->t_flags & TF_NODELAY), 2827 ("%s: sent too much: %u>%u", 2828 __func__, sent, maxseg)); 2829 tp->snd_limited = 2; 2830 } else if (sent > 0) { 2831 ++tp->snd_limited; 2832 } 2833 tp->snd_cwnd = oldcwnd; 2834 goto drop; 2835 } 2836 } 2837 break; 2838 } else { 2839 /* 2840 * This ack is advancing the left edge, reset the 2841 * counter. 2842 */ 2843 tp->t_dupacks = 0; 2844 /* 2845 * If this ack also has new SACK info, increment the 2846 * counter as per rfc6675. The variable 2847 * sack_changed tracks all changes to the SACK 2848 * scoreboard, including when partial ACKs without 2849 * SACK options are received, and clear the scoreboard 2850 * from the left side. Such partial ACKs should not be 2851 * counted as dupacks here. 2852 */ 2853 if (tcp_is_sack_recovery(tp, &to) && 2854 (((tp->t_rxtshift == 0) && (sack_changed != SACK_NOCHANGE)) || 2855 ((tp->t_rxtshift > 0) && (sack_changed == SACK_NEWLOSS))) && 2856 (tp->snd_nxt == tp->snd_max)) { 2857 tp->t_dupacks++; 2858 /* limit overhead by setting maxseg last */ 2859 if (!IN_FASTRECOVERY(tp->t_flags) && 2860 (tp->sackhint.sacked_bytes > 2861 ((tcprexmtthresh - 1) * 2862 (maxseg = tcp_maxseg(tp))))) { 2863 goto enter_recovery; 2864 } 2865 } 2866 } 2867 2868 resume_partialack: 2869 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2870 ("%s: th_ack <= snd_una", __func__)); 2871 2872 /* 2873 * If the congestion window was inflated to account 2874 * for the other side's cached packets, retract it. 2875 */ 2876 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2877 if (IN_FASTRECOVERY(tp->t_flags)) { 2878 if (tp->t_flags & TF_SACK_PERMIT) { 2879 if (V_tcp_do_prr && 2880 (to.to_flags & TOF_SACK)) { 2881 tcp_timer_activate(tp, 2882 TT_REXMT, 0); 2883 tp->t_rtttime = 0; 2884 tcp_do_prr_ack(tp, th, &to, 2885 sack_changed, &maxseg); 2886 tp->t_flags |= TF_ACKNOW; 2887 (void) tcp_output(tp); 2888 } else { 2889 tcp_sack_partialack(tp, th, 2890 &maxseg); 2891 } 2892 } else { 2893 tcp_newreno_partial_ack(tp, th); 2894 } 2895 } else if (IN_CONGRECOVERY(tp->t_flags) && 2896 (V_tcp_do_prr)) { 2897 tp->sackhint.delivered_data = 2898 BYTES_THIS_ACK(tp, th); 2899 tp->snd_fack = th->th_ack; 2900 /* 2901 * During ECN cwnd reduction 2902 * always use PRR-SSRB 2903 */ 2904 tcp_do_prr_ack(tp, th, &to, SACK_CHANGE, 2905 &maxseg); 2906 (void) tcp_output(tp); 2907 } 2908 } 2909 /* 2910 * If we reach this point, ACK is not a duplicate, 2911 * i.e., it ACKs something we sent. 2912 */ 2913 if (tp->t_flags & TF_NEEDSYN) { 2914 /* 2915 * T/TCP: Connection was half-synchronized, and our 2916 * SYN has been ACK'd (so connection is now fully 2917 * synchronized). Go to non-starred state, 2918 * increment snd_una for ACK of SYN, and check if 2919 * we can do window scaling. 2920 */ 2921 tp->t_flags &= ~TF_NEEDSYN; 2922 tp->snd_una++; 2923 /* Do window scaling? */ 2924 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2925 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2926 tp->rcv_scale = tp->request_r_scale; 2927 /* Send window already scaled. */ 2928 } 2929 } 2930 2931 process_ACK: 2932 INP_WLOCK_ASSERT(inp); 2933 2934 /* 2935 * Adjust for the SYN bit in sequence space, 2936 * but don't account for it in cwnd calculations. 2937 * This is for the SYN_RECEIVED, non-simultaneous 2938 * SYN case. SYN_SENT and simultaneous SYN are 2939 * treated elsewhere. 2940 */ 2941 if (incforsyn) 2942 tp->snd_una++; 2943 acked = BYTES_THIS_ACK(tp, th); 2944 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2945 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2946 tp->snd_una, th->th_ack, tp, m)); 2947 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2948 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2949 2950 /* 2951 * If we just performed our first retransmit, and the ACK 2952 * arrives within our recovery window, then it was a mistake 2953 * to do the retransmit in the first place. Recover our 2954 * original cwnd and ssthresh, and proceed to transmit where 2955 * we left off. 2956 */ 2957 if (tp->t_rxtshift == 1 && 2958 tp->t_flags & TF_PREVVALID && 2959 tp->t_badrxtwin != 0 && 2960 to.to_flags & TOF_TS && 2961 to.to_tsecr != 0 && 2962 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 2963 cc_cong_signal(tp, th, CC_RTO_ERR); 2964 2965 /* 2966 * If we have a timestamp reply, update smoothed 2967 * round trip time. If no timestamp is present but 2968 * transmit timer is running and timed sequence 2969 * number was acked, update smoothed round trip time. 2970 * Since we now have an rtt measurement, cancel the 2971 * timer backoff (cf., Phil Karn's retransmit alg.). 2972 * Recompute the initial retransmit timer. 2973 * 2974 * Some boxes send broken timestamp replies 2975 * during the SYN+ACK phase, ignore 2976 * timestamps of 0 or we could calculate a 2977 * huge RTT and blow up the retransmit timer. 2978 */ 2979 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2980 uint32_t t; 2981 2982 t = tcp_ts_getticks() - to.to_tsecr; 2983 if (!tp->t_rttlow || tp->t_rttlow > t) 2984 tp->t_rttlow = t; 2985 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2986 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2987 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2988 tp->t_rttlow = ticks - tp->t_rtttime; 2989 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2990 } 2991 2992 SOCK_SENDBUF_LOCK(so); 2993 /* 2994 * Clear t_acktime if remote side has ACKd all data in the 2995 * socket buffer and FIN (if applicable). 2996 * Otherwise, update t_acktime if we received a sufficiently 2997 * large ACK. 2998 */ 2999 if ((tp->t_state <= TCPS_CLOSE_WAIT && 3000 acked == sbavail(&so->so_snd)) || 3001 acked > sbavail(&so->so_snd)) 3002 tp->t_acktime = 0; 3003 else if (acked > 1) 3004 tp->t_acktime = ticks; 3005 3006 /* 3007 * If all outstanding data is acked, stop retransmit 3008 * timer and remember to restart (more output or persist). 3009 * If there is more data to be acked, restart retransmit 3010 * timer, using current (possibly backed-off) value. 3011 */ 3012 if (th->th_ack == tp->snd_max) { 3013 tcp_timer_activate(tp, TT_REXMT, 0); 3014 needoutput = 1; 3015 } else if (!tcp_timer_active(tp, TT_PERSIST)) 3016 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); 3017 3018 /* 3019 * If no data (only SYN) was ACK'd, 3020 * skip rest of ACK processing. 3021 */ 3022 if (acked == 0) { 3023 SOCK_SENDBUF_UNLOCK(so); 3024 goto step6; 3025 } 3026 3027 /* 3028 * Let the congestion control algorithm update congestion 3029 * control related information. This typically means increasing 3030 * the congestion window. 3031 */ 3032 cc_ack_received(tp, th, nsegs, CC_ACK); 3033 3034 if (acked > sbavail(&so->so_snd)) { 3035 if (tp->snd_wnd >= sbavail(&so->so_snd)) 3036 tp->snd_wnd -= sbavail(&so->so_snd); 3037 else 3038 tp->snd_wnd = 0; 3039 mfree = sbcut_locked(&so->so_snd, 3040 (int)sbavail(&so->so_snd)); 3041 ourfinisacked = 1; 3042 } else { 3043 mfree = sbcut_locked(&so->so_snd, acked); 3044 if (tp->snd_wnd >= (uint32_t) acked) 3045 tp->snd_wnd -= acked; 3046 else 3047 tp->snd_wnd = 0; 3048 ourfinisacked = 0; 3049 } 3050 /* NB: sowwakeup_locked() does an implicit unlock. */ 3051 sowwakeup_locked(so); 3052 m_freem(mfree); 3053 /* Detect una wraparound. */ 3054 if (!IN_RECOVERY(tp->t_flags) && 3055 SEQ_GT(tp->snd_una, tp->snd_recover) && 3056 SEQ_LEQ(th->th_ack, tp->snd_recover)) 3057 tp->snd_recover = th->th_ack - 1; 3058 tp->snd_una = th->th_ack; 3059 if (IN_RECOVERY(tp->t_flags) && 3060 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 3061 cc_post_recovery(tp, th); 3062 } 3063 if (SEQ_GT(tp->snd_una, tp->snd_recover)) { 3064 tp->snd_recover = tp->snd_una; 3065 } 3066 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 3067 tp->snd_nxt = tp->snd_una; 3068 3069 switch (tp->t_state) { 3070 /* 3071 * In FIN_WAIT_1 STATE in addition to the processing 3072 * for the ESTABLISHED state if our FIN is now acknowledged 3073 * then enter FIN_WAIT_2. 3074 */ 3075 case TCPS_FIN_WAIT_1: 3076 if (ourfinisacked) { 3077 /* 3078 * If we can't receive any more 3079 * data, then closing user can proceed. 3080 * Starting the timer is contrary to the 3081 * specification, but if we don't get a FIN 3082 * we'll hang forever. 3083 */ 3084 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3085 tcp_free_sackholes(tp); 3086 soisdisconnected(so); 3087 tcp_timer_activate(tp, TT_2MSL, 3088 (tcp_fast_finwait2_recycle ? 3089 tcp_finwait2_timeout : 3090 TP_MAXIDLE(tp))); 3091 } 3092 tcp_state_change(tp, TCPS_FIN_WAIT_2); 3093 } 3094 break; 3095 3096 /* 3097 * In CLOSING STATE in addition to the processing for 3098 * the ESTABLISHED state if the ACK acknowledges our FIN 3099 * then enter the TIME-WAIT state, otherwise ignore 3100 * the segment. 3101 */ 3102 case TCPS_CLOSING: 3103 if (ourfinisacked) { 3104 tcp_twstart(tp); 3105 m_freem(m); 3106 return; 3107 } 3108 break; 3109 3110 /* 3111 * In LAST_ACK, we may still be waiting for data to drain 3112 * and/or to be acked, as well as for the ack of our FIN. 3113 * If our FIN is now acknowledged, delete the TCB, 3114 * enter the closed state and return. 3115 */ 3116 case TCPS_LAST_ACK: 3117 if (ourfinisacked) { 3118 tp = tcp_close(tp); 3119 goto drop; 3120 } 3121 break; 3122 } 3123 } 3124 3125 step6: 3126 INP_WLOCK_ASSERT(inp); 3127 3128 /* 3129 * Update window information. 3130 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3131 */ 3132 if ((thflags & TH_ACK) && 3133 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3134 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3135 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3136 /* keep track of pure window updates */ 3137 if (no_data && tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3138 TCPSTAT_INC(tcps_rcvwinupd); 3139 tp->snd_wnd = tiwin; 3140 tp->snd_wl1 = th->th_seq; 3141 tp->snd_wl2 = th->th_ack; 3142 if (tp->snd_wnd > tp->max_sndwnd) 3143 tp->max_sndwnd = tp->snd_wnd; 3144 needoutput = 1; 3145 } 3146 3147 /* 3148 * Process segments with URG. 3149 */ 3150 if ((thflags & TH_URG) && th->th_urp && 3151 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3152 /* 3153 * This is a kludge, but if we receive and accept 3154 * random urgent pointers, we'll crash in 3155 * soreceive. It's hard to imagine someone 3156 * actually wanting to send this much urgent data. 3157 */ 3158 SOCK_RECVBUF_LOCK(so); 3159 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3160 th->th_urp = 0; /* XXX */ 3161 thflags &= ~TH_URG; /* XXX */ 3162 SOCK_RECVBUF_UNLOCK(so); /* XXX */ 3163 goto dodata; /* XXX */ 3164 } 3165 /* 3166 * If this segment advances the known urgent pointer, 3167 * then mark the data stream. This should not happen 3168 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3169 * a FIN has been received from the remote side. 3170 * In these states we ignore the URG. 3171 * 3172 * According to RFC961 (Assigned Protocols), 3173 * the urgent pointer points to the last octet 3174 * of urgent data. We continue, however, 3175 * to consider it to indicate the first octet 3176 * of data past the urgent section as the original 3177 * spec states (in one of two places). 3178 */ 3179 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3180 tp->rcv_up = th->th_seq + th->th_urp; 3181 so->so_oobmark = sbavail(&so->so_rcv) + 3182 (tp->rcv_up - tp->rcv_nxt) - 1; 3183 if (so->so_oobmark == 0) 3184 so->so_rcv.sb_state |= SBS_RCVATMARK; 3185 sohasoutofband(so); 3186 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3187 } 3188 SOCK_RECVBUF_UNLOCK(so); 3189 /* 3190 * Remove out of band data so doesn't get presented to user. 3191 * This can happen independent of advancing the URG pointer, 3192 * but if two URG's are pending at once, some out-of-band 3193 * data may creep in... ick. 3194 */ 3195 if (th->th_urp <= (uint32_t)tlen && 3196 !(so->so_options & SO_OOBINLINE)) { 3197 /* hdr drop is delayed */ 3198 tcp_pulloutofband(so, th, m, drop_hdrlen); 3199 } 3200 } else { 3201 /* 3202 * If no out of band data is expected, 3203 * pull receive urgent pointer along 3204 * with the receive window. 3205 */ 3206 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3207 tp->rcv_up = tp->rcv_nxt; 3208 } 3209 dodata: /* XXX */ 3210 INP_WLOCK_ASSERT(inp); 3211 3212 /* 3213 * Process the segment text, merging it into the TCP sequencing queue, 3214 * and arranging for acknowledgment of receipt if necessary. 3215 * This process logically involves adjusting tp->rcv_wnd as data 3216 * is presented to the user (this happens in tcp_usrreq.c, 3217 * case PRU_RCVD). If a FIN has already been received on this 3218 * connection then we just ignore the text. 3219 */ 3220 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3221 (tp->t_flags & TF_FASTOPEN)); 3222 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3223 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3224 tcp_seq save_start = th->th_seq; 3225 tcp_seq save_rnxt = tp->rcv_nxt; 3226 int save_tlen = tlen; 3227 m_adj(m, drop_hdrlen); /* delayed header drop */ 3228 /* 3229 * Insert segment which includes th into TCP reassembly queue 3230 * with control block tp. Set thflags to whether reassembly now 3231 * includes a segment with FIN. This handles the common case 3232 * inline (segment is the next to be received on an established 3233 * connection, and the queue is empty), avoiding linkage into 3234 * and removal from the queue and repetition of various 3235 * conversions. 3236 * Set DELACK for segments received in order, but ack 3237 * immediately when segments are out of order (so 3238 * fast retransmit can work). 3239 */ 3240 if (th->th_seq == tp->rcv_nxt && 3241 SEGQ_EMPTY(tp) && 3242 (TCPS_HAVEESTABLISHED(tp->t_state) || 3243 tfo_syn)) { 3244 if (DELAY_ACK(tp, tlen) || tfo_syn) 3245 tp->t_flags |= TF_DELACK; 3246 else 3247 tp->t_flags |= TF_ACKNOW; 3248 tp->rcv_nxt += tlen; 3249 if (tlen && 3250 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3251 (tp->t_fbyte_in == 0)) { 3252 tp->t_fbyte_in = ticks; 3253 if (tp->t_fbyte_in == 0) 3254 tp->t_fbyte_in = 1; 3255 if (tp->t_fbyte_out && tp->t_fbyte_in) 3256 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3257 } 3258 thflags = tcp_get_flags(th) & TH_FIN; 3259 TCPSTAT_INC(tcps_rcvpack); 3260 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3261 SOCK_RECVBUF_LOCK(so); 3262 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3263 m_freem(m); 3264 else 3265 sbappendstream_locked(&so->so_rcv, m, 0); 3266 tp->t_flags |= TF_WAKESOR; 3267 } else { 3268 /* 3269 * XXX: Due to the header drop above "th" is 3270 * theoretically invalid by now. Fortunately 3271 * m_adj() doesn't actually frees any mbufs 3272 * when trimming from the head. 3273 */ 3274 tcp_seq temp = save_start; 3275 3276 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3277 tp->t_flags |= TF_ACKNOW; 3278 } 3279 if ((tp->t_flags & TF_SACK_PERMIT) && 3280 (save_tlen > 0) && 3281 TCPS_HAVEESTABLISHED(tp->t_state)) { 3282 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3283 /* 3284 * DSACK actually handled in the fastpath 3285 * above. 3286 */ 3287 tcp_update_sack_list(tp, save_start, 3288 save_start + save_tlen); 3289 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3290 if ((tp->rcv_numsacks >= 1) && 3291 (tp->sackblks[0].end == save_start)) { 3292 /* 3293 * Partial overlap, recorded at todrop 3294 * above. 3295 */ 3296 tcp_update_sack_list(tp, 3297 tp->sackblks[0].start, 3298 tp->sackblks[0].end); 3299 } else { 3300 tcp_update_dsack_list(tp, save_start, 3301 save_start + save_tlen); 3302 } 3303 } else if (tlen >= save_tlen) { 3304 /* Update of sackblks. */ 3305 tcp_update_dsack_list(tp, save_start, 3306 save_start + save_tlen); 3307 } else if (tlen > 0) { 3308 tcp_update_dsack_list(tp, save_start, 3309 save_start + tlen); 3310 } 3311 } 3312 tcp_handle_wakeup(tp); 3313 #if 0 3314 /* 3315 * Note the amount of data that peer has sent into 3316 * our window, in order to estimate the sender's 3317 * buffer size. 3318 * XXX: Unused. 3319 */ 3320 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3321 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3322 else 3323 len = so->so_rcv.sb_hiwat; 3324 #endif 3325 } else { 3326 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 3327 if (tlen > 0) { 3328 if ((thflags & TH_FIN) != 0) { 3329 log(LOG_DEBUG, "%s; %s: %s: " 3330 "Received %d bytes of data and FIN " 3331 "after having received a FIN, " 3332 "just dropping both\n", 3333 s, __func__, 3334 tcpstates[tp->t_state], tlen); 3335 } else { 3336 log(LOG_DEBUG, "%s; %s: %s: " 3337 "Received %d bytes of data " 3338 "after having received a FIN, " 3339 "just dropping it\n", 3340 s, __func__, 3341 tcpstates[tp->t_state], tlen); 3342 } 3343 } else { 3344 if ((thflags & TH_FIN) != 0) { 3345 log(LOG_DEBUG, "%s; %s: %s: " 3346 "Received FIN " 3347 "after having received a FIN, " 3348 "just dropping it\n", 3349 s, __func__, 3350 tcpstates[tp->t_state]); 3351 } 3352 } 3353 free(s, M_TCPLOG); 3354 } 3355 m_freem(m); 3356 thflags &= ~TH_FIN; 3357 } 3358 3359 /* 3360 * If FIN is received ACK the FIN and let the user know 3361 * that the connection is closing. 3362 */ 3363 if (thflags & TH_FIN) { 3364 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3365 /* The socket upcall is handled by socantrcvmore. */ 3366 socantrcvmore(so); 3367 /* 3368 * If connection is half-synchronized 3369 * (ie NEEDSYN flag on) then delay ACK, 3370 * so it may be piggybacked when SYN is sent. 3371 * Otherwise, since we received a FIN then no 3372 * more input can be expected, send ACK now. 3373 */ 3374 if (tp->t_flags & TF_NEEDSYN) 3375 tp->t_flags |= TF_DELACK; 3376 else 3377 tp->t_flags |= TF_ACKNOW; 3378 tp->rcv_nxt++; 3379 } 3380 switch (tp->t_state) { 3381 /* 3382 * In SYN_RECEIVED and ESTABLISHED STATES 3383 * enter the CLOSE_WAIT state. 3384 */ 3385 case TCPS_SYN_RECEIVED: 3386 tp->t_starttime = ticks; 3387 /* FALLTHROUGH */ 3388 case TCPS_ESTABLISHED: 3389 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3390 break; 3391 3392 /* 3393 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3394 * enter the CLOSING state. 3395 */ 3396 case TCPS_FIN_WAIT_1: 3397 tcp_state_change(tp, TCPS_CLOSING); 3398 break; 3399 3400 /* 3401 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3402 * starting the time-wait timer, turning off the other 3403 * standard timers. 3404 */ 3405 case TCPS_FIN_WAIT_2: 3406 tcp_twstart(tp); 3407 return; 3408 } 3409 } 3410 TCP_PROBE3(debug__input, tp, th, m); 3411 3412 /* 3413 * Return any desired output. 3414 */ 3415 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 3416 (void) tcp_output(tp); 3417 } 3418 check_delack: 3419 INP_WLOCK_ASSERT(inp); 3420 3421 if (tp->t_flags & TF_DELACK) { 3422 tp->t_flags &= ~TF_DELACK; 3423 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3424 } 3425 INP_WUNLOCK(inp); 3426 return; 3427 3428 dropafterack: 3429 /* 3430 * Generate an ACK dropping incoming segment if it occupies 3431 * sequence space, where the ACK reflects our state. 3432 * 3433 * We can now skip the test for the RST flag since all 3434 * paths to this code happen after packets containing 3435 * RST have been dropped. 3436 * 3437 * In the SYN-RECEIVED state, don't send an ACK unless the 3438 * segment we received passes the SYN-RECEIVED ACK test. 3439 * If it fails send a RST. This breaks the loop in the 3440 * "LAND" DoS attack, and also prevents an ACK storm 3441 * between two listening ports that have been sent forged 3442 * SYN segments, each with the source address of the other. 3443 */ 3444 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3445 (SEQ_GT(tp->snd_una, th->th_ack) || 3446 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3447 rstreason = BANDLIM_TCP_RST; 3448 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 3449 goto dropwithreset; 3450 } 3451 TCP_PROBE3(debug__input, tp, th, m); 3452 tp->t_flags |= TF_ACKNOW; 3453 (void) tcp_output(tp); 3454 INP_WUNLOCK(inp); 3455 m_freem(m); 3456 return; 3457 3458 dropwithreset: 3459 if (tp != NULL) { 3460 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3461 INP_WUNLOCK(inp); 3462 } else 3463 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3464 return; 3465 3466 drop: 3467 /* 3468 * Drop space held by incoming segment and return. 3469 */ 3470 TCP_PROBE3(debug__input, tp, th, m); 3471 if (tp != NULL) { 3472 INP_WUNLOCK(inp); 3473 } 3474 m_freem(m); 3475 } 3476 3477 /* 3478 * Issue RST and make ACK acceptable to originator of segment. 3479 * The mbuf must still include the original packet header. 3480 * tp may be NULL. 3481 */ 3482 void 3483 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3484 int tlen, int rstreason) 3485 { 3486 #ifdef INET 3487 struct ip *ip; 3488 #endif 3489 #ifdef INET6 3490 struct ip6_hdr *ip6; 3491 #endif 3492 3493 if (tp != NULL) { 3494 INP_LOCK_ASSERT(tptoinpcb(tp)); 3495 } 3496 3497 /* Don't bother if destination was broadcast/multicast. */ 3498 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3499 goto drop; 3500 #ifdef INET6 3501 if (mtod(m, struct ip *)->ip_v == 6) { 3502 ip6 = mtod(m, struct ip6_hdr *); 3503 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3504 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3505 goto drop; 3506 /* IPv6 anycast check is done at tcp6_input() */ 3507 } 3508 #endif 3509 #if defined(INET) && defined(INET6) 3510 else 3511 #endif 3512 #ifdef INET 3513 { 3514 ip = mtod(m, struct ip *); 3515 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3516 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3517 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3518 in_ifnet_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3519 goto drop; 3520 } 3521 #endif 3522 3523 /* Perform bandwidth limiting. */ 3524 if (badport_bandlim(rstreason) < 0) 3525 goto drop; 3526 3527 /* tcp_respond consumes the mbuf chain. */ 3528 if (tcp_get_flags(th) & TH_ACK) { 3529 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3530 th->th_ack, TH_RST); 3531 } else { 3532 if (tcp_get_flags(th) & TH_SYN) 3533 tlen++; 3534 if (tcp_get_flags(th) & TH_FIN) 3535 tlen++; 3536 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3537 (tcp_seq)0, TH_RST|TH_ACK); 3538 } 3539 return; 3540 drop: 3541 m_freem(m); 3542 } 3543 3544 /* 3545 * Parse TCP options and place in tcpopt. 3546 */ 3547 void 3548 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3549 { 3550 int opt, optlen; 3551 3552 to->to_flags = 0; 3553 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3554 opt = cp[0]; 3555 if (opt == TCPOPT_EOL) 3556 break; 3557 if (opt == TCPOPT_NOP) 3558 optlen = 1; 3559 else { 3560 if (cnt < 2) 3561 break; 3562 optlen = cp[1]; 3563 if (optlen < 2 || optlen > cnt) 3564 break; 3565 } 3566 switch (opt) { 3567 case TCPOPT_MAXSEG: 3568 if (optlen != TCPOLEN_MAXSEG) 3569 continue; 3570 if (!(flags & TO_SYN)) 3571 continue; 3572 to->to_flags |= TOF_MSS; 3573 bcopy((char *)cp + 2, 3574 (char *)&to->to_mss, sizeof(to->to_mss)); 3575 to->to_mss = ntohs(to->to_mss); 3576 break; 3577 case TCPOPT_WINDOW: 3578 if (optlen != TCPOLEN_WINDOW) 3579 continue; 3580 if (!(flags & TO_SYN)) 3581 continue; 3582 to->to_flags |= TOF_SCALE; 3583 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3584 break; 3585 case TCPOPT_TIMESTAMP: 3586 if (optlen != TCPOLEN_TIMESTAMP) 3587 continue; 3588 to->to_flags |= TOF_TS; 3589 bcopy((char *)cp + 2, 3590 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3591 to->to_tsval = ntohl(to->to_tsval); 3592 bcopy((char *)cp + 6, 3593 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3594 to->to_tsecr = ntohl(to->to_tsecr); 3595 break; 3596 case TCPOPT_SIGNATURE: 3597 /* 3598 * In order to reply to a host which has set the 3599 * TCP_SIGNATURE option in its initial SYN, we have 3600 * to record the fact that the option was observed 3601 * here for the syncache code to perform the correct 3602 * response. 3603 */ 3604 if (optlen != TCPOLEN_SIGNATURE) 3605 continue; 3606 to->to_flags |= TOF_SIGNATURE; 3607 to->to_signature = cp + 2; 3608 break; 3609 case TCPOPT_SACK_PERMITTED: 3610 if (optlen != TCPOLEN_SACK_PERMITTED) 3611 continue; 3612 if (!(flags & TO_SYN)) 3613 continue; 3614 if (!V_tcp_do_sack) 3615 continue; 3616 to->to_flags |= TOF_SACKPERM; 3617 break; 3618 case TCPOPT_SACK: 3619 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3620 continue; 3621 if (flags & TO_SYN) 3622 continue; 3623 to->to_flags |= TOF_SACK; 3624 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3625 to->to_sacks = cp + 2; 3626 TCPSTAT_INC(tcps_sack_rcv_blocks); 3627 break; 3628 case TCPOPT_FAST_OPEN: 3629 /* 3630 * Cookie length validation is performed by the 3631 * server side cookie checking code or the client 3632 * side cookie cache update code. 3633 */ 3634 if (!(flags & TO_SYN)) 3635 continue; 3636 if (!V_tcp_fastopen_client_enable && 3637 !V_tcp_fastopen_server_enable) 3638 continue; 3639 to->to_flags |= TOF_FASTOPEN; 3640 to->to_tfo_len = optlen - 2; 3641 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3642 break; 3643 default: 3644 continue; 3645 } 3646 } 3647 } 3648 3649 /* 3650 * Pull out of band byte out of a segment so 3651 * it doesn't appear in the user's data queue. 3652 * It is still reflected in the segment length for 3653 * sequencing purposes. 3654 */ 3655 void 3656 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3657 int off) 3658 { 3659 int cnt = off + th->th_urp - 1; 3660 3661 while (cnt >= 0) { 3662 if (m->m_len > cnt) { 3663 char *cp = mtod(m, caddr_t) + cnt; 3664 struct tcpcb *tp = sototcpcb(so); 3665 3666 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3667 3668 tp->t_iobc = *cp; 3669 tp->t_oobflags |= TCPOOB_HAVEDATA; 3670 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3671 m->m_len--; 3672 if (m->m_flags & M_PKTHDR) 3673 m->m_pkthdr.len--; 3674 return; 3675 } 3676 cnt -= m->m_len; 3677 m = m->m_next; 3678 if (m == NULL) 3679 break; 3680 } 3681 panic("tcp_pulloutofband"); 3682 } 3683 3684 /* 3685 * Collect new round-trip time estimate 3686 * and update averages and current timeout. 3687 */ 3688 void 3689 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3690 { 3691 int delta; 3692 3693 INP_WLOCK_ASSERT(tptoinpcb(tp)); 3694 3695 TCPSTAT_INC(tcps_rttupdated); 3696 if (tp->t_rttupdated < UCHAR_MAX) 3697 tp->t_rttupdated++; 3698 #ifdef STATS 3699 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3700 imax(0, rtt * 1000 / hz)); 3701 #endif 3702 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3703 /* 3704 * srtt is stored as fixed point with 5 bits after the 3705 * binary point (i.e., scaled by 8). The following magic 3706 * is equivalent to the smoothing algorithm in rfc793 with 3707 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3708 * point). Adjust rtt to origin 0. 3709 */ 3710 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3711 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3712 3713 if ((tp->t_srtt += delta) <= 0) 3714 tp->t_srtt = 1; 3715 3716 /* 3717 * We accumulate a smoothed rtt variance (actually, a 3718 * smoothed mean difference), then set the retransmit 3719 * timer to smoothed rtt + 4 times the smoothed variance. 3720 * rttvar is stored as fixed point with 4 bits after the 3721 * binary point (scaled by 16). The following is 3722 * equivalent to rfc793 smoothing with an alpha of .75 3723 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3724 * rfc793's wired-in beta. 3725 */ 3726 if (delta < 0) 3727 delta = -delta; 3728 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3729 if ((tp->t_rttvar += delta) <= 0) 3730 tp->t_rttvar = 1; 3731 } else { 3732 /* 3733 * No rtt measurement yet - use the unsmoothed rtt. 3734 * Set the variance to half the rtt (so our first 3735 * retransmit happens at 3*rtt). 3736 */ 3737 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3738 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3739 } 3740 tp->t_rtttime = 0; 3741 tp->t_rxtshift = 0; 3742 3743 /* 3744 * the retransmit should happen at rtt + 4 * rttvar. 3745 * Because of the way we do the smoothing, srtt and rttvar 3746 * will each average +1/2 tick of bias. When we compute 3747 * the retransmit timer, we want 1/2 tick of rounding and 3748 * 1 extra tick because of +-1/2 tick uncertainty in the 3749 * firing of the timer. The bias will give us exactly the 3750 * 1.5 tick we need. But, because the bias is 3751 * statistical, we have to test that we don't drop below 3752 * the minimum feasible timer (which is 2 ticks). 3753 */ 3754 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3755 max(tp->t_rttmin, rtt + 2), tcp_rexmit_max); 3756 3757 /* 3758 * We received an ack for a packet that wasn't retransmitted; 3759 * it is probably safe to discard any error indications we've 3760 * received recently. This isn't quite right, but close enough 3761 * for now (a route might have failed after we sent a segment, 3762 * and the return path might not be symmetrical). 3763 */ 3764 tp->t_softerror = 0; 3765 } 3766 3767 /* 3768 * Determine a reasonable value for maxseg size. 3769 * If the route is known, check route for mtu. 3770 * If none, use an mss that can be handled on the outgoing interface 3771 * without forcing IP to fragment. If no route is found, route has no mtu, 3772 * or the destination isn't local, use a default, hopefully conservative 3773 * size (usually 512 or the default IP max size, but no more than the mtu 3774 * of the interface), as we can't discover anything about intervening 3775 * gateways or networks. We also initialize the congestion/slow start 3776 * window to be a single segment if the destination isn't local. 3777 * While looking at the routing entry, we also initialize other path-dependent 3778 * parameters from pre-set or cached values in the routing entry. 3779 * 3780 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3781 * IP options, e.g. IPSEC data, since length of this data may vary, and 3782 * thus it is calculated for every segment separately in tcp_output(). 3783 * 3784 * NOTE that this routine is only called when we process an incoming 3785 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3786 * settings are handled in tcp_mssopt(). 3787 */ 3788 void 3789 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3790 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3791 { 3792 int mss = 0; 3793 uint32_t maxmtu = 0; 3794 struct inpcb *inp = tptoinpcb(tp); 3795 struct hc_metrics_lite metrics; 3796 #ifdef INET6 3797 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3798 size_t min_protoh = isipv6 ? 3799 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3800 sizeof (struct tcpiphdr); 3801 #else 3802 size_t min_protoh = sizeof(struct tcpiphdr); 3803 #endif 3804 3805 INP_WLOCK_ASSERT(inp); 3806 3807 if (tp->t_port) 3808 min_protoh += V_tcp_udp_tunneling_overhead; 3809 if (mtuoffer != -1) { 3810 KASSERT(offer == -1, ("%s: conflict", __func__)); 3811 offer = mtuoffer - min_protoh; 3812 } 3813 3814 /* Initialize. */ 3815 #ifdef INET6 3816 if (isipv6) { 3817 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3818 tp->t_maxseg = V_tcp_v6mssdflt; 3819 } 3820 #endif 3821 #if defined(INET) && defined(INET6) 3822 else 3823 #endif 3824 #ifdef INET 3825 { 3826 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3827 tp->t_maxseg = V_tcp_mssdflt; 3828 } 3829 #endif 3830 3831 /* 3832 * No route to sender, stay with default mss and return. 3833 */ 3834 if (maxmtu == 0) { 3835 /* 3836 * In case we return early we need to initialize metrics 3837 * to a defined state as tcp_hc_get() would do for us 3838 * if there was no cache hit. 3839 */ 3840 if (metricptr != NULL) 3841 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3842 return; 3843 } 3844 3845 /* What have we got? */ 3846 switch (offer) { 3847 case 0: 3848 /* 3849 * Offer == 0 means that there was no MSS on the SYN 3850 * segment, in this case we use tcp_mssdflt as 3851 * already assigned to t_maxseg above. 3852 */ 3853 offer = tp->t_maxseg; 3854 break; 3855 3856 case -1: 3857 /* 3858 * Offer == -1 means that we didn't receive SYN yet. 3859 */ 3860 /* FALLTHROUGH */ 3861 3862 default: 3863 /* 3864 * Prevent DoS attack with too small MSS. Round up 3865 * to at least minmss. 3866 */ 3867 offer = max(offer, V_tcp_minmss); 3868 } 3869 3870 if (metricptr == NULL) 3871 metricptr = &metrics; 3872 tcp_hc_get(&inp->inp_inc, metricptr); 3873 3874 /* 3875 * If there's a discovered mtu in tcp hostcache, use it. 3876 * Else, use the link mtu. 3877 */ 3878 if (metricptr->hc_mtu) 3879 mss = min(metricptr->hc_mtu, maxmtu) - min_protoh; 3880 else { 3881 #ifdef INET6 3882 if (isipv6) { 3883 mss = maxmtu - min_protoh; 3884 if (!V_path_mtu_discovery && 3885 !in6_localaddr(&inp->in6p_faddr)) 3886 mss = min(mss, V_tcp_v6mssdflt); 3887 } 3888 #endif 3889 #if defined(INET) && defined(INET6) 3890 else 3891 #endif 3892 #ifdef INET 3893 { 3894 mss = maxmtu - min_protoh; 3895 if (!V_path_mtu_discovery && 3896 !in_localaddr(inp->inp_faddr)) 3897 mss = min(mss, V_tcp_mssdflt); 3898 } 3899 #endif 3900 /* 3901 * XXX - The above conditional (mss = maxmtu - min_protoh) 3902 * probably violates the TCP spec. 3903 * The problem is that, since we don't know the 3904 * other end's MSS, we are supposed to use a conservative 3905 * default. But, if we do that, then MTU discovery will 3906 * never actually take place, because the conservative 3907 * default is much less than the MTUs typically seen 3908 * on the Internet today. For the moment, we'll sweep 3909 * this under the carpet. 3910 * 3911 * The conservative default might not actually be a problem 3912 * if the only case this occurs is when sending an initial 3913 * SYN with options and data to a host we've never talked 3914 * to before. Then, they will reply with an MSS value which 3915 * will get recorded and the new parameters should get 3916 * recomputed. For Further Study. 3917 */ 3918 } 3919 mss = min(mss, offer); 3920 3921 /* 3922 * Sanity check: make sure that maxseg will be large 3923 * enough to allow some data on segments even if the 3924 * all the option space is used (40bytes). Otherwise 3925 * funny things may happen in tcp_output. 3926 * 3927 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3928 */ 3929 mss = max(mss, 64); 3930 3931 tp->t_maxseg = mss; 3932 if (tp->t_maxseg < V_tcp_mssdflt) { 3933 /* 3934 * The MSS is so small we should not process incoming 3935 * SACK's since we are subject to attack in such a 3936 * case. 3937 */ 3938 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 3939 } else { 3940 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 3941 } 3942 3943 } 3944 3945 void 3946 tcp_mss(struct tcpcb *tp, int offer) 3947 { 3948 int mss; 3949 uint32_t bufsize; 3950 struct inpcb *inp = tptoinpcb(tp); 3951 struct socket *so; 3952 struct hc_metrics_lite metrics; 3953 struct tcp_ifcap cap; 3954 3955 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3956 3957 bzero(&cap, sizeof(cap)); 3958 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3959 3960 mss = tp->t_maxseg; 3961 3962 /* 3963 * If there's a pipesize, change the socket buffer to that size, 3964 * don't change if sb_hiwat is different than default (then it 3965 * has been changed on purpose with setsockopt). 3966 * Make the socket buffers an integral number of mss units; 3967 * if the mss is larger than the socket buffer, decrease the mss. 3968 */ 3969 so = inp->inp_socket; 3970 SOCK_SENDBUF_LOCK(so); 3971 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.hc_sendpipe) 3972 bufsize = metrics.hc_sendpipe; 3973 else 3974 bufsize = so->so_snd.sb_hiwat; 3975 if (bufsize < mss) 3976 mss = bufsize; 3977 else { 3978 bufsize = roundup(bufsize, mss); 3979 if (bufsize > sb_max) 3980 bufsize = sb_max; 3981 if (bufsize > so->so_snd.sb_hiwat) 3982 (void)sbreserve_locked(so, SO_SND, bufsize, NULL); 3983 } 3984 SOCK_SENDBUF_UNLOCK(so); 3985 /* 3986 * Sanity check: make sure that maxseg will be large 3987 * enough to allow some data on segments even if the 3988 * all the option space is used (40bytes). Otherwise 3989 * funny things may happen in tcp_output. 3990 * 3991 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3992 */ 3993 tp->t_maxseg = max(mss, 64); 3994 if (tp->t_maxseg < V_tcp_mssdflt) { 3995 /* 3996 * The MSS is so small we should not process incoming 3997 * SACK's since we are subject to attack in such a 3998 * case. 3999 */ 4000 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 4001 } else { 4002 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 4003 } 4004 4005 SOCK_RECVBUF_LOCK(so); 4006 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.hc_recvpipe) 4007 bufsize = metrics.hc_recvpipe; 4008 else 4009 bufsize = so->so_rcv.sb_hiwat; 4010 if (bufsize > mss) { 4011 bufsize = roundup(bufsize, mss); 4012 if (bufsize > sb_max) 4013 bufsize = sb_max; 4014 if (bufsize > so->so_rcv.sb_hiwat) 4015 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL); 4016 } 4017 SOCK_RECVBUF_UNLOCK(so); 4018 4019 /* Check the interface for TSO capabilities. */ 4020 if (cap.ifcap & CSUM_TSO) { 4021 tp->t_flags |= TF_TSO; 4022 tp->t_tsomax = cap.tsomax; 4023 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 4024 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 4025 if (cap.ipsec_tso) 4026 tp->t_flags2 |= TF2_IPSEC_TSO; 4027 } 4028 } 4029 4030 /* 4031 * Determine the MSS option to send on an outgoing SYN. 4032 */ 4033 int 4034 tcp_mssopt(struct in_conninfo *inc) 4035 { 4036 int mss = 0; 4037 uint32_t thcmtu = 0; 4038 uint32_t maxmtu = 0; 4039 size_t min_protoh; 4040 4041 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 4042 4043 #ifdef INET6 4044 if (inc->inc_flags & INC_ISIPV6) { 4045 mss = V_tcp_v6mssdflt; 4046 maxmtu = tcp_maxmtu6(inc, NULL); 4047 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 4048 } 4049 #endif 4050 #if defined(INET) && defined(INET6) 4051 else 4052 #endif 4053 #ifdef INET 4054 { 4055 mss = V_tcp_mssdflt; 4056 maxmtu = tcp_maxmtu(inc, NULL); 4057 min_protoh = sizeof(struct tcpiphdr); 4058 } 4059 #endif 4060 #if defined(INET6) || defined(INET) 4061 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 4062 #endif 4063 4064 if (maxmtu && thcmtu) 4065 mss = min(maxmtu, thcmtu) - min_protoh; 4066 else if (maxmtu || thcmtu) 4067 mss = max(maxmtu, thcmtu) - min_protoh; 4068 4069 return (mss); 4070 } 4071 4072 void 4073 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, 4074 sackstatus_t sack_changed, u_int *maxsegp) 4075 { 4076 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 4077 u_int maxseg; 4078 4079 INP_WLOCK_ASSERT(tptoinpcb(tp)); 4080 4081 if (*maxsegp == 0) { 4082 *maxsegp = tcp_maxseg(tp); 4083 } 4084 maxseg = *maxsegp; 4085 /* 4086 * Compute the amount of data that this ACK is indicating 4087 * (del_data) and an estimate of how many bytes are in the 4088 * network. 4089 */ 4090 if (tcp_is_sack_recovery(tp, to) || 4091 (IN_CONGRECOVERY(tp->t_flags) && 4092 !IN_FASTRECOVERY(tp->t_flags))) { 4093 del_data = tp->sackhint.delivered_data; 4094 pipe = tcp_compute_pipe(tp); 4095 } else { 4096 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + 4097 tp->snd_recover - tp->snd_una)) { 4098 del_data = maxseg; 4099 } 4100 pipe = imax(0, tp->snd_max - tp->snd_una - 4101 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); 4102 } 4103 tp->sackhint.prr_delivered += del_data; 4104 /* 4105 * Proportional Rate Reduction 4106 */ 4107 if (pipe >= tp->snd_ssthresh) { 4108 if (tp->sackhint.recover_fs == 0) 4109 tp->sackhint.recover_fs = 4110 imax(1, tp->snd_nxt - tp->snd_una); 4111 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 4112 tp->snd_ssthresh, tp->sackhint.recover_fs) - 4113 tp->sackhint.prr_out + maxseg - 1; 4114 } else { 4115 /* 4116 * PRR 6937bis heuristic: 4117 * - A partial ack without SACK block beneath snd_recover 4118 * indicates further loss. 4119 * - An SACK scoreboard update adding a new hole indicates 4120 * further loss, so be conservative and send at most one 4121 * segment. 4122 * - Prevent ACK splitting attacks, by being conservative 4123 * when no new data is acked. 4124 */ 4125 if ((sack_changed == SACK_NEWLOSS) || (del_data == 0)) { 4126 limit = tp->sackhint.prr_delivered - 4127 tp->sackhint.prr_out; 4128 } else { 4129 limit = imax(tp->sackhint.prr_delivered - 4130 tp->sackhint.prr_out, del_data) + 4131 maxseg; 4132 } 4133 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 4134 } 4135 snd_cnt = imax(snd_cnt, 0) / maxseg; 4136 /* 4137 * Send snd_cnt new data into the network in response to this ack. 4138 * If there is going to be a SACK retransmission, adjust snd_cwnd 4139 * accordingly. 4140 */ 4141 if (IN_FASTRECOVERY(tp->t_flags)) { 4142 if (tcp_is_sack_recovery(tp, to)) { 4143 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4144 } else { 4145 tp->snd_cwnd = (tp->snd_max - tp->snd_una) + 4146 (snd_cnt * maxseg); 4147 } 4148 } else if (IN_CONGRECOVERY(tp->t_flags)) { 4149 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4150 } 4151 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); 4152 } 4153 4154 /* 4155 * On a partial ack arrives, force the retransmission of the 4156 * next unacknowledged segment. Do not clear tp->t_dupacks. 4157 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4158 * be started again. 4159 */ 4160 void 4161 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4162 { 4163 tcp_seq onxt = tp->snd_nxt; 4164 uint32_t ocwnd = tp->snd_cwnd; 4165 u_int maxseg = tcp_maxseg(tp); 4166 4167 INP_WLOCK_ASSERT(tptoinpcb(tp)); 4168 4169 tcp_timer_activate(tp, TT_REXMT, 0); 4170 tp->t_rtttime = 0; 4171 if (IN_FASTRECOVERY(tp->t_flags)) { 4172 tp->snd_nxt = th->th_ack; 4173 /* 4174 * Set snd_cwnd to one segment beyond acknowledged offset. 4175 * (tp->snd_una has not yet been updated when this function is called.) 4176 */ 4177 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4178 tp->t_flags |= TF_ACKNOW; 4179 (void) tcp_output(tp); 4180 tp->snd_cwnd = ocwnd; 4181 if (SEQ_GT(onxt, tp->snd_nxt)) 4182 tp->snd_nxt = onxt; 4183 } 4184 /* 4185 * Partial window deflation. Relies on fact that tp->snd_una 4186 * not updated yet. 4187 */ 4188 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4189 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4190 else 4191 tp->snd_cwnd = 0; 4192 tp->snd_cwnd += maxseg; 4193 } 4194 4195 int 4196 tcp_compute_pipe(struct tcpcb *tp) 4197 { 4198 int pipe; 4199 4200 if (tp->t_fb->tfb_compute_pipe != NULL) { 4201 pipe = (*tp->t_fb->tfb_compute_pipe)(tp); 4202 } else if (V_tcp_do_newsack) { 4203 pipe = tp->snd_max - tp->snd_una + 4204 tp->sackhint.sack_bytes_rexmit - 4205 tp->sackhint.sacked_bytes - 4206 tp->sackhint.lost_bytes; 4207 } else { 4208 pipe = tp->snd_nxt - tp->snd_fack + tp->sackhint.sack_bytes_rexmit; 4209 } 4210 return (imax(pipe, 0)); 4211 } 4212 4213 uint32_t 4214 tcp_compute_initwnd(uint32_t maxseg) 4215 { 4216 /* 4217 * Calculate the Initial Window, also used as Restart Window 4218 * 4219 * RFC5681 Section 3.1 specifies the default conservative values. 4220 * RFC3390 specifies slightly more aggressive values. 4221 * RFC6928 increases it to ten segments. 4222 * Support for user specified value for initial flight size. 4223 */ 4224 if (V_tcp_initcwnd_segments) 4225 return min(V_tcp_initcwnd_segments * maxseg, 4226 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4227 else if (V_tcp_do_rfc3390) 4228 return min(4 * maxseg, max(2 * maxseg, 4380)); 4229 else { 4230 /* Per RFC5681 Section 3.1 */ 4231 if (maxseg > 2190) 4232 return (2 * maxseg); 4233 else if (maxseg > 1095) 4234 return (3 * maxseg); 4235 else 4236 return (4 * maxseg); 4237 } 4238 } 4239