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