fastpath.c revision 310212
1/*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (c) 2007-2008,2010 5 * Swinburne University of Technology, Melbourne, Australia. 6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 7 * Copyright (c) 2010 The FreeBSD Foundation 8 * Copyright (c) 2010-2011 Juniper Networks, Inc. 9 * Copyright (c) 2015 Netflix Inc. 10 * All rights reserved. 11 * 12 * Portions of this software were developed at the Centre for Advanced Internet 13 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 14 * James Healy and David Hayes, made possible in part by a grant from the Cisco 15 * University Research Program Fund at Community Foundation Silicon Valley. 16 * 17 * Portions of this software were developed at the Centre for Advanced 18 * Internet Architectures, Swinburne University of Technology, Melbourne, 19 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 20 * 21 * Portions of this software were developed by Robert N. M. Watson under 22 * contract to Juniper Networks, Inc. 23 * 24 * Portions of this software were developed by Randall R. Stewart while 25 * working for Netflix Inc. 26 * 27 * Redistribution and use in source and binary forms, with or without 28 * modification, are permitted provided that the following conditions 29 * are met: 30 * 1. Redistributions of source code must retain the above copyright 31 * notice, this list of conditions and the following disclaimer. 32 * 2. Redistributions in binary form must reproduce the above copyright 33 * notice, this list of conditions and the following disclaimer in the 34 * documentation and/or other materials provided with the distribution. 35 * 4. Neither the name of the University nor the names of its contributors 36 * may be used to endorse or promote products derived from this software 37 * without specific prior written permission. 38 * 39 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 40 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 41 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 42 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 43 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 44 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 45 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 46 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 47 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 48 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 49 * SUCH DAMAGE. 50 * 51 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 52 */ 53 54#include <sys/cdefs.h> 55__FBSDID("$FreeBSD: stable/11/sys/netinet/tcp_stacks/fastpath.c 310212 2016-12-18 12:23:48Z tuexen $"); 56 57#include "opt_inet.h" 58#include "opt_inet6.h" 59#include "opt_ipsec.h" 60#include "opt_tcpdebug.h" 61 62#include <sys/param.h> 63#include <sys/module.h> 64#include <sys/kernel.h> 65#include <sys/hhook.h> 66#include <sys/malloc.h> 67#include <sys/mbuf.h> 68#include <sys/proc.h> /* for proc0 declaration */ 69#include <sys/protosw.h> 70#include <sys/sdt.h> 71#include <sys/signalvar.h> 72#include <sys/socket.h> 73#include <sys/socketvar.h> 74#include <sys/sysctl.h> 75#include <sys/syslog.h> 76#include <sys/systm.h> 77 78#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 79 80#include <vm/uma.h> 81 82#include <net/route.h> 83#include <net/vnet.h> 84 85#define TCPSTATES /* for logging */ 86 87#include <netinet/in.h> 88#include <netinet/in_kdtrace.h> 89#include <netinet/in_pcb.h> 90#include <netinet/in_systm.h> 91#include <netinet/ip.h> 92#include <netinet/ip_icmp.h> /* required for icmp_var.h */ 93#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 94#include <netinet/ip_var.h> 95#include <netinet/ip_options.h> 96#include <netinet/ip6.h> 97#include <netinet/icmp6.h> 98#include <netinet6/in6_pcb.h> 99#include <netinet6/ip6_var.h> 100#include <netinet/tcp.h> 101#include <netinet/tcp_fsm.h> 102#include <netinet/tcp_seq.h> 103#include <netinet/tcp_timer.h> 104#include <netinet/tcp_var.h> 105#include <netinet6/tcp6_var.h> 106#include <netinet/tcpip.h> 107#include <netinet/tcp_syncache.h> 108#include <netinet/cc/cc.h> 109#ifdef TCPDEBUG 110#include <netinet/tcp_debug.h> 111#endif /* TCPDEBUG */ 112#ifdef TCP_OFFLOAD 113#include <netinet/tcp_offload.h> 114#endif 115 116#ifdef IPSEC 117#include <netipsec/ipsec.h> 118#include <netipsec/ipsec6.h> 119#endif /*IPSEC*/ 120 121#include <machine/in_cksum.h> 122 123#include <security/mac/mac_framework.h> 124 125VNET_DECLARE(int, tcp_autorcvbuf_inc); 126#define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 127VNET_DECLARE(int, tcp_autorcvbuf_max); 128#define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 129VNET_DECLARE(int, tcp_do_rfc3042); 130#define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 131VNET_DECLARE(int, tcp_do_autorcvbuf); 132#define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 133VNET_DECLARE(int, tcp_insecure_rst); 134#define V_tcp_insecure_rst VNET(tcp_insecure_rst) 135VNET_DECLARE(int, tcp_insecure_syn); 136#define V_tcp_insecure_syn VNET(tcp_insecure_syn) 137 138static void tcp_do_segment_fastslow(struct mbuf *, struct tcphdr *, 139 struct socket *, struct tcpcb *, int, int, uint8_t, 140 int); 141 142static void tcp_do_segment_fastack(struct mbuf *, struct tcphdr *, 143 struct socket *, struct tcpcb *, int, int, uint8_t, 144 int); 145 146/* 147 * Indicate whether this ack should be delayed. We can delay the ack if 148 * following conditions are met: 149 * - There is no delayed ack timer in progress. 150 * - Our last ack wasn't a 0-sized window. We never want to delay 151 * the ack that opens up a 0-sized window. 152 * - LRO wasn't used for this segment. We make sure by checking that the 153 * segment size is not larger than the MSS. 154 */ 155#define DELAY_ACK(tp, tlen) \ 156 ((!tcp_timer_active(tp, TT_DELACK) && \ 157 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 158 (tlen <= tp->t_maxseg) && \ 159 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 160 161/* 162 * So how is this faster than the normal fast ack? 163 * It basically allows us to also stay in the fastpath 164 * when a window-update ack also arrives. In testing 165 * we saw only 25-30% of connections doing fastpath 166 * due to the fact that along with moving forward 167 * in sequence the window was also updated. 168 */ 169static void 170tcp_do_fastack(struct mbuf *m, struct tcphdr *th, struct socket *so, 171 struct tcpcb *tp, struct tcpopt *to, int drop_hdrlen, int tlen, 172 int ti_locked, u_long tiwin) 173{ 174 int acked; 175 int winup_only=0; 176#ifdef TCPDEBUG 177 /* 178 * The size of tcp_saveipgen must be the size of the max ip header, 179 * now IPv6. 180 */ 181 u_char tcp_saveipgen[IP6_HDR_LEN]; 182 struct tcphdr tcp_savetcp; 183 short ostate = 0; 184#endif 185 /* 186 * The following if statement will be true if 187 * we are doing the win_up_in_fp <and> 188 * - We have more new data (SEQ_LT(tp->snd_wl1, th->th_seq)) <or> 189 * - No more new data, but we have an ack for new data 190 * (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) 191 * - No more new data, the same ack point but the window grew 192 * (tp->snd_wl1 == th->th_seq && tp->snd_wl2 == th->th_ack && twin > tp->snd_wnd) 193 */ 194 if ((SEQ_LT(tp->snd_wl1, th->th_seq) || 195 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 196 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 197 /* keep track of pure window updates */ 198 if (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) { 199 winup_only = 1; 200 TCPSTAT_INC(tcps_rcvwinupd); 201 } 202 tp->snd_wnd = tiwin; 203 tp->snd_wl1 = th->th_seq; 204 tp->snd_wl2 = th->th_ack; 205 if (tp->snd_wnd > tp->max_sndwnd) 206 tp->max_sndwnd = tp->snd_wnd; 207 } 208 /* 209 * If last ACK falls within this segment's sequence numbers, 210 * record the timestamp. 211 * NOTE that the test is modified according to the latest 212 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 213 */ 214 if ((to->to_flags & TOF_TS) != 0 && 215 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 216 tp->ts_recent_age = tcp_ts_getticks(); 217 tp->ts_recent = to->to_tsval; 218 } 219 /* 220 * This is a pure ack for outstanding data. 221 */ 222 if (ti_locked == TI_RLOCKED) { 223 INP_INFO_RUNLOCK(&V_tcbinfo); 224 } 225 ti_locked = TI_UNLOCKED; 226 227 TCPSTAT_INC(tcps_predack); 228 229 /* 230 * "bad retransmit" recovery. 231 */ 232 if (tp->t_rxtshift == 1 && 233 tp->t_flags & TF_PREVVALID && 234 (int)(ticks - tp->t_badrxtwin) < 0) { 235 cc_cong_signal(tp, th, CC_RTO_ERR); 236 } 237 238 /* 239 * Recalculate the transmit timer / rtt. 240 * 241 * Some boxes send broken timestamp replies 242 * during the SYN+ACK phase, ignore 243 * timestamps of 0 or we could calculate a 244 * huge RTT and blow up the retransmit timer. 245 */ 246 if ((to->to_flags & TOF_TS) != 0 && 247 to->to_tsecr) { 248 u_int t; 249 250 t = tcp_ts_getticks() - to->to_tsecr; 251 if (!tp->t_rttlow || tp->t_rttlow > t) 252 tp->t_rttlow = t; 253 tcp_xmit_timer(tp, 254 TCP_TS_TO_TICKS(t) + 1); 255 } else if (tp->t_rtttime && 256 SEQ_GT(th->th_ack, tp->t_rtseq)) { 257 if (!tp->t_rttlow || 258 tp->t_rttlow > ticks - tp->t_rtttime) 259 tp->t_rttlow = ticks - tp->t_rtttime; 260 tcp_xmit_timer(tp, 261 ticks - tp->t_rtttime); 262 } 263 if (winup_only == 0) { 264 acked = BYTES_THIS_ACK(tp, th); 265 266 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 267 hhook_run_tcp_est_in(tp, th, to); 268 269 TCPSTAT_ADD(tcps_rcvackbyte, acked); 270 sbdrop(&so->so_snd, acked); 271 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 272 SEQ_LEQ(th->th_ack, tp->snd_recover)) 273 tp->snd_recover = th->th_ack - 1; 274 275 /* 276 * Let the congestion control algorithm update 277 * congestion control related information. This 278 * typically means increasing the congestion 279 * window. 280 */ 281 cc_ack_received(tp, th, CC_ACK); 282 283 tp->snd_una = th->th_ack; 284 /* 285 * Pull snd_wl2 up to prevent seq wrap relative 286 * to th_ack. 287 */ 288 tp->snd_wl2 = th->th_ack; 289 tp->t_dupacks = 0; 290 291 /* 292 * If all outstanding data are acked, stop 293 * retransmit timer, otherwise restart timer 294 * using current (possibly backed-off) value. 295 * If process is waiting for space, 296 * wakeup/selwakeup/signal. If data 297 * are ready to send, let tcp_output 298 * decide between more output or persist. 299 */ 300#ifdef TCPDEBUG 301 if (so->so_options & SO_DEBUG) 302 tcp_trace(TA_INPUT, ostate, tp, 303 (void *)tcp_saveipgen, 304 &tcp_savetcp, 0); 305#endif 306 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 307 m_freem(m); 308 if (tp->snd_una == tp->snd_max) 309 tcp_timer_activate(tp, TT_REXMT, 0); 310 else if (!tcp_timer_active(tp, TT_PERSIST)) 311 tcp_timer_activate(tp, TT_REXMT, 312 tp->t_rxtcur); 313 } else { 314 /* 315 * Window update only, just free the mbufs and 316 * send out whatever we can. 317 */ 318 m_freem(m); 319 } 320 sowwakeup(so); 321 if (sbavail(&so->so_snd)) 322 (void) tcp_output(tp); 323 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 324 __func__, ti_locked)); 325 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 326 INP_WLOCK_ASSERT(tp->t_inpcb); 327 328 if (tp->t_flags & TF_DELACK) { 329 tp->t_flags &= ~TF_DELACK; 330 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 331 } 332 INP_WUNLOCK(tp->t_inpcb); 333} 334 335/* 336 * Here nothing is really faster, its just that we 337 * have broken out the fast-data path also just like 338 * the fast-ack. 339 */ 340static void 341tcp_do_fastnewdata(struct mbuf *m, struct tcphdr *th, struct socket *so, 342 struct tcpcb *tp, struct tcpopt *to, int drop_hdrlen, int tlen, 343 int ti_locked, u_long tiwin) 344{ 345 int newsize = 0; /* automatic sockbuf scaling */ 346#ifdef TCPDEBUG 347 /* 348 * The size of tcp_saveipgen must be the size of the max ip header, 349 * now IPv6. 350 */ 351 u_char tcp_saveipgen[IP6_HDR_LEN]; 352 struct tcphdr tcp_savetcp; 353 short ostate = 0; 354#endif 355 /* 356 * If last ACK falls within this segment's sequence numbers, 357 * record the timestamp. 358 * NOTE that the test is modified according to the latest 359 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 360 */ 361 if ((to->to_flags & TOF_TS) != 0 && 362 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 363 tp->ts_recent_age = tcp_ts_getticks(); 364 tp->ts_recent = to->to_tsval; 365 } 366 367 /* 368 * This is a pure, in-sequence data packet with 369 * nothing on the reassembly queue and we have enough 370 * buffer space to take it. 371 */ 372 if (ti_locked == TI_RLOCKED) { 373 INP_INFO_RUNLOCK(&V_tcbinfo); 374 } 375 ti_locked = TI_UNLOCKED; 376 377 /* Clean receiver SACK report if present */ 378 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 379 tcp_clean_sackreport(tp); 380 TCPSTAT_INC(tcps_preddat); 381 tp->rcv_nxt += tlen; 382 /* 383 * Pull snd_wl1 up to prevent seq wrap relative to 384 * th_seq. 385 */ 386 tp->snd_wl1 = th->th_seq; 387 /* 388 * Pull rcv_up up to prevent seq wrap relative to 389 * rcv_nxt. 390 */ 391 tp->rcv_up = tp->rcv_nxt; 392 TCPSTAT_ADD(tcps_rcvbyte, tlen); 393#ifdef TCPDEBUG 394 if (so->so_options & SO_DEBUG) 395 tcp_trace(TA_INPUT, ostate, tp, 396 (void *)tcp_saveipgen, &tcp_savetcp, 0); 397#endif 398 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 399 /* 400 * Automatic sizing of receive socket buffer. Often the send 401 * buffer size is not optimally adjusted to the actual network 402 * conditions at hand (delay bandwidth product). Setting the 403 * buffer size too small limits throughput on links with high 404 * bandwidth and high delay (eg. trans-continental/oceanic links). 405 * 406 * On the receive side the socket buffer memory is only rarely 407 * used to any significant extent. This allows us to be much 408 * more aggressive in scaling the receive socket buffer. For 409 * the case that the buffer space is actually used to a large 410 * extent and we run out of kernel memory we can simply drop 411 * the new segments; TCP on the sender will just retransmit it 412 * later. Setting the buffer size too big may only consume too 413 * much kernel memory if the application doesn't read() from 414 * the socket or packet loss or reordering makes use of the 415 * reassembly queue. 416 * 417 * The criteria to step up the receive buffer one notch are: 418 * 1. Application has not set receive buffer size with 419 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 420 * 2. the number of bytes received during the time it takes 421 * one timestamp to be reflected back to us (the RTT); 422 * 3. received bytes per RTT is within seven eighth of the 423 * current socket buffer size; 424 * 4. receive buffer size has not hit maximal automatic size; 425 * 426 * This algorithm does one step per RTT at most and only if 427 * we receive a bulk stream w/o packet losses or reorderings. 428 * Shrinking the buffer during idle times is not necessary as 429 * it doesn't consume any memory when idle. 430 * 431 * TODO: Only step up if the application is actually serving 432 * the buffer to better manage the socket buffer resources. 433 */ 434 if (V_tcp_do_autorcvbuf && 435 (to->to_flags & TOF_TS) && 436 to->to_tsecr && 437 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 438 if (TSTMP_GT(to->to_tsecr, tp->rfbuf_ts) && 439 to->to_tsecr - tp->rfbuf_ts < hz) { 440 if (tp->rfbuf_cnt > 441 (so->so_rcv.sb_hiwat / 8 * 7) && 442 so->so_rcv.sb_hiwat < 443 V_tcp_autorcvbuf_max) { 444 newsize = 445 min(so->so_rcv.sb_hiwat + 446 V_tcp_autorcvbuf_inc, 447 V_tcp_autorcvbuf_max); 448 } 449 /* Start over with next RTT. */ 450 tp->rfbuf_ts = 0; 451 tp->rfbuf_cnt = 0; 452 } else 453 tp->rfbuf_cnt += tlen; /* add up */ 454 } 455 456 /* Add data to socket buffer. */ 457 SOCKBUF_LOCK(&so->so_rcv); 458 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 459 m_freem(m); 460 } else { 461 /* 462 * Set new socket buffer size. 463 * Give up when limit is reached. 464 */ 465 if (newsize) 466 if (!sbreserve_locked(&so->so_rcv, 467 newsize, so, NULL)) 468 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 469 m_adj(m, drop_hdrlen); /* delayed header drop */ 470 sbappendstream_locked(&so->so_rcv, m, 0); 471 } 472 /* NB: sorwakeup_locked() does an implicit unlock. */ 473 sorwakeup_locked(so); 474 if (DELAY_ACK(tp, tlen)) { 475 tp->t_flags |= TF_DELACK; 476 } else { 477 tp->t_flags |= TF_ACKNOW; 478 tcp_output(tp); 479 } 480 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 481 __func__, ti_locked)); 482 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 483 INP_WLOCK_ASSERT(tp->t_inpcb); 484 485 if (tp->t_flags & TF_DELACK) { 486 tp->t_flags &= ~TF_DELACK; 487 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 488 } 489 INP_WUNLOCK(tp->t_inpcb); 490} 491 492/* 493 * The slow-path is the clone of the long long part 494 * of tcp_do_segment past all the fast-path stuff. We 495 * use it here by two different callers, the fast/slow and 496 * the fastack only. 497 */ 498static void 499tcp_do_slowpath(struct mbuf *m, struct tcphdr *th, struct socket *so, 500 struct tcpcb *tp, struct tcpopt *to, int drop_hdrlen, int tlen, 501 int ti_locked, u_long tiwin, int thflags) 502{ 503 int acked, ourfinisacked, needoutput = 0; 504 int rstreason, todrop, win; 505 char *s; 506 struct in_conninfo *inc; 507 struct mbuf *mfree = NULL; 508#ifdef TCPDEBUG 509 /* 510 * The size of tcp_saveipgen must be the size of the max ip header, 511 * now IPv6. 512 */ 513 u_char tcp_saveipgen[IP6_HDR_LEN]; 514 struct tcphdr tcp_savetcp; 515 short ostate = 0; 516#endif 517 /* 518 * Calculate amount of space in receive window, 519 * and then do TCP input processing. 520 * Receive window is amount of space in rcv queue, 521 * but not less than advertised window. 522 */ 523 inc = &tp->t_inpcb->inp_inc; 524 win = sbspace(&so->so_rcv); 525 if (win < 0) 526 win = 0; 527 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 528 529 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 530 tp->rfbuf_ts = 0; 531 tp->rfbuf_cnt = 0; 532 533 switch (tp->t_state) { 534 535 /* 536 * If the state is SYN_RECEIVED: 537 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 538 */ 539 case TCPS_SYN_RECEIVED: 540 if ((thflags & TH_ACK) && 541 (SEQ_LEQ(th->th_ack, tp->snd_una) || 542 SEQ_GT(th->th_ack, tp->snd_max))) { 543 rstreason = BANDLIM_RST_OPENPORT; 544 goto dropwithreset; 545 } 546 break; 547 548 /* 549 * If the state is SYN_SENT: 550 * if seg contains an ACK, but not for our SYN, drop the input. 551 * if seg contains a RST, then drop the connection. 552 * if seg does not contain SYN, then drop it. 553 * Otherwise this is an acceptable SYN segment 554 * initialize tp->rcv_nxt and tp->irs 555 * if seg contains ack then advance tp->snd_una 556 * if seg contains an ECE and ECN support is enabled, the stream 557 * is ECN capable. 558 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 559 * arrange for segment to be acked (eventually) 560 * continue processing rest of data/controls, beginning with URG 561 */ 562 case TCPS_SYN_SENT: 563 if ((thflags & TH_ACK) && 564 (SEQ_LEQ(th->th_ack, tp->iss) || 565 SEQ_GT(th->th_ack, tp->snd_max))) { 566 rstreason = BANDLIM_UNLIMITED; 567 goto dropwithreset; 568 } 569 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 570 TCP_PROBE5(connect__refused, NULL, tp, 571 mtod(m, const char *), tp, th); 572 tp = tcp_drop(tp, ECONNREFUSED); 573 } 574 if (thflags & TH_RST) 575 goto drop; 576 if (!(thflags & TH_SYN)) 577 goto drop; 578 579 tp->irs = th->th_seq; 580 tcp_rcvseqinit(tp); 581 if (thflags & TH_ACK) { 582 TCPSTAT_INC(tcps_connects); 583 soisconnected(so); 584#ifdef MAC 585 mac_socketpeer_set_from_mbuf(m, so); 586#endif 587 /* Do window scaling on this connection? */ 588 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 589 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 590 tp->rcv_scale = tp->request_r_scale; 591 } 592 tp->rcv_adv += imin(tp->rcv_wnd, 593 TCP_MAXWIN << tp->rcv_scale); 594 tp->snd_una++; /* SYN is acked */ 595 /* 596 * If there's data, delay ACK; if there's also a FIN 597 * ACKNOW will be turned on later. 598 */ 599 if (DELAY_ACK(tp, tlen) && tlen != 0) 600 tcp_timer_activate(tp, TT_DELACK, 601 tcp_delacktime); 602 else 603 tp->t_flags |= TF_ACKNOW; 604 605 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 606 tp->t_flags |= TF_ECN_PERMIT; 607 TCPSTAT_INC(tcps_ecn_shs); 608 } 609 610 /* 611 * Received <SYN,ACK> in SYN_SENT[*] state. 612 * Transitions: 613 * SYN_SENT --> ESTABLISHED 614 * SYN_SENT* --> FIN_WAIT_1 615 */ 616 tp->t_starttime = ticks; 617 if (tp->t_flags & TF_NEEDFIN) { 618 tcp_state_change(tp, TCPS_FIN_WAIT_1); 619 tp->t_flags &= ~TF_NEEDFIN; 620 thflags &= ~TH_SYN; 621 } else { 622 tcp_state_change(tp, TCPS_ESTABLISHED); 623 TCP_PROBE5(connect__established, NULL, tp, 624 mtod(m, const char *), tp, th); 625 cc_conn_init(tp); 626 tcp_timer_activate(tp, TT_KEEP, 627 TP_KEEPIDLE(tp)); 628 } 629 } else { 630 /* 631 * Received initial SYN in SYN-SENT[*] state => 632 * simultaneous open. 633 * If it succeeds, connection is * half-synchronized. 634 * Otherwise, do 3-way handshake: 635 * SYN-SENT -> SYN-RECEIVED 636 * SYN-SENT* -> SYN-RECEIVED* 637 */ 638 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 639 tcp_timer_activate(tp, TT_REXMT, 0); 640 tcp_state_change(tp, TCPS_SYN_RECEIVED); 641 } 642 643 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 644 "ti_locked %d", __func__, ti_locked)); 645 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 646 INP_WLOCK_ASSERT(tp->t_inpcb); 647 648 /* 649 * Advance th->th_seq to correspond to first data byte. 650 * If data, trim to stay within window, 651 * dropping FIN if necessary. 652 */ 653 th->th_seq++; 654 if (tlen > tp->rcv_wnd) { 655 todrop = tlen - tp->rcv_wnd; 656 m_adj(m, -todrop); 657 tlen = tp->rcv_wnd; 658 thflags &= ~TH_FIN; 659 TCPSTAT_INC(tcps_rcvpackafterwin); 660 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 661 } 662 tp->snd_wl1 = th->th_seq - 1; 663 tp->rcv_up = th->th_seq; 664 /* 665 * Client side of transaction: already sent SYN and data. 666 * If the remote host used T/TCP to validate the SYN, 667 * our data will be ACK'd; if so, enter normal data segment 668 * processing in the middle of step 5, ack processing. 669 * Otherwise, goto step 6. 670 */ 671 if (thflags & TH_ACK) 672 goto process_ACK; 673 674 goto step6; 675 676 /* 677 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 678 * do normal processing. 679 * 680 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 681 */ 682 case TCPS_LAST_ACK: 683 case TCPS_CLOSING: 684 break; /* continue normal processing */ 685 } 686 687 /* 688 * States other than LISTEN or SYN_SENT. 689 * First check the RST flag and sequence number since reset segments 690 * are exempt from the timestamp and connection count tests. This 691 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 692 * below which allowed reset segments in half the sequence space 693 * to fall though and be processed (which gives forged reset 694 * segments with a random sequence number a 50 percent chance of 695 * killing a connection). 696 * Then check timestamp, if present. 697 * Then check the connection count, if present. 698 * Then check that at least some bytes of segment are within 699 * receive window. If segment begins before rcv_nxt, 700 * drop leading data (and SYN); if nothing left, just ack. 701 */ 702 if (thflags & TH_RST) { 703 /* 704 * RFC5961 Section 3.2 705 * 706 * - RST drops connection only if SEG.SEQ == RCV.NXT. 707 * - If RST is in window, we send challenge ACK. 708 * 709 * Note: to take into account delayed ACKs, we should 710 * test against last_ack_sent instead of rcv_nxt. 711 * Note 2: we handle special case of closed window, not 712 * covered by the RFC. 713 */ 714 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 715 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 716 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 717 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 718 KASSERT(ti_locked == TI_RLOCKED, 719 ("%s: TH_RST ti_locked %d, th %p tp %p", 720 __func__, ti_locked, th, tp)); 721 KASSERT(tp->t_state != TCPS_SYN_SENT, 722 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 723 __func__, th, tp)); 724 725 if (V_tcp_insecure_rst || 726 tp->last_ack_sent == th->th_seq) { 727 TCPSTAT_INC(tcps_drops); 728 /* Drop the connection. */ 729 switch (tp->t_state) { 730 case TCPS_SYN_RECEIVED: 731 so->so_error = ECONNREFUSED; 732 goto close; 733 case TCPS_ESTABLISHED: 734 case TCPS_FIN_WAIT_1: 735 case TCPS_FIN_WAIT_2: 736 case TCPS_CLOSE_WAIT: 737 case TCPS_CLOSING: 738 case TCPS_LAST_ACK: 739 so->so_error = ECONNRESET; 740 close: 741 /* FALLTHROUGH */ 742 default: 743 tp = tcp_close(tp); 744 } 745 } else { 746 TCPSTAT_INC(tcps_badrst); 747 /* Send challenge ACK. */ 748 tcp_respond(tp, mtod(m, void *), th, m, 749 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 750 tp->last_ack_sent = tp->rcv_nxt; 751 m = NULL; 752 } 753 } 754 goto drop; 755 } 756 757 /* 758 * RFC5961 Section 4.2 759 * Send challenge ACK for any SYN in synchronized state. 760 */ 761 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT) { 762 KASSERT(ti_locked == TI_RLOCKED, 763 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 764 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 765 766 TCPSTAT_INC(tcps_badsyn); 767 if (V_tcp_insecure_syn && 768 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 769 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 770 tp = tcp_drop(tp, ECONNRESET); 771 rstreason = BANDLIM_UNLIMITED; 772 } else { 773 /* Send challenge ACK. */ 774 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 775 tp->snd_nxt, TH_ACK); 776 tp->last_ack_sent = tp->rcv_nxt; 777 m = NULL; 778 } 779 goto drop; 780 } 781 782 /* 783 * RFC 1323 PAWS: If we have a timestamp reply on this segment 784 * and it's less than ts_recent, drop it. 785 */ 786 if ((to->to_flags & TOF_TS) != 0 && tp->ts_recent && 787 TSTMP_LT(to->to_tsval, tp->ts_recent)) { 788 789 /* Check to see if ts_recent is over 24 days old. */ 790 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 791 /* 792 * Invalidate ts_recent. If this segment updates 793 * ts_recent, the age will be reset later and ts_recent 794 * will get a valid value. If it does not, setting 795 * ts_recent to zero will at least satisfy the 796 * requirement that zero be placed in the timestamp 797 * echo reply when ts_recent isn't valid. The 798 * age isn't reset until we get a valid ts_recent 799 * because we don't want out-of-order segments to be 800 * dropped when ts_recent is old. 801 */ 802 tp->ts_recent = 0; 803 } else { 804 TCPSTAT_INC(tcps_rcvduppack); 805 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 806 TCPSTAT_INC(tcps_pawsdrop); 807 if (tlen) 808 goto dropafterack; 809 goto drop; 810 } 811 } 812 813 /* 814 * In the SYN-RECEIVED state, validate that the packet belongs to 815 * this connection before trimming the data to fit the receive 816 * window. Check the sequence number versus IRS since we know 817 * the sequence numbers haven't wrapped. This is a partial fix 818 * for the "LAND" DoS attack. 819 */ 820 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 821 rstreason = BANDLIM_RST_OPENPORT; 822 goto dropwithreset; 823 } 824 825 todrop = tp->rcv_nxt - th->th_seq; 826 if (todrop > 0) { 827 if (thflags & TH_SYN) { 828 thflags &= ~TH_SYN; 829 th->th_seq++; 830 if (th->th_urp > 1) 831 th->th_urp--; 832 else 833 thflags &= ~TH_URG; 834 todrop--; 835 } 836 /* 837 * Following if statement from Stevens, vol. 2, p. 960. 838 */ 839 if (todrop > tlen 840 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 841 /* 842 * Any valid FIN must be to the left of the window. 843 * At this point the FIN must be a duplicate or out 844 * of sequence; drop it. 845 */ 846 thflags &= ~TH_FIN; 847 848 /* 849 * Send an ACK to resynchronize and drop any data. 850 * But keep on processing for RST or ACK. 851 */ 852 tp->t_flags |= TF_ACKNOW; 853 todrop = tlen; 854 TCPSTAT_INC(tcps_rcvduppack); 855 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 856 } else { 857 TCPSTAT_INC(tcps_rcvpartduppack); 858 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 859 } 860 drop_hdrlen += todrop; /* drop from the top afterwards */ 861 th->th_seq += todrop; 862 tlen -= todrop; 863 if (th->th_urp > todrop) 864 th->th_urp -= todrop; 865 else { 866 thflags &= ~TH_URG; 867 th->th_urp = 0; 868 } 869 } 870 871 /* 872 * If new data are received on a connection after the 873 * user processes are gone, then RST the other end. 874 */ 875 if ((so->so_state & SS_NOFDREF) && 876 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 877 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 878 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 879 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 880 881 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 882 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 883 "after socket was closed, " 884 "sending RST and removing tcpcb\n", 885 s, __func__, tcpstates[tp->t_state], tlen); 886 free(s, M_TCPLOG); 887 } 888 tp = tcp_close(tp); 889 TCPSTAT_INC(tcps_rcvafterclose); 890 rstreason = BANDLIM_UNLIMITED; 891 goto dropwithreset; 892 } 893 894 /* 895 * If segment ends after window, drop trailing data 896 * (and PUSH and FIN); if nothing left, just ACK. 897 */ 898 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 899 if (todrop > 0) { 900 TCPSTAT_INC(tcps_rcvpackafterwin); 901 if (todrop >= tlen) { 902 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 903 /* 904 * If window is closed can only take segments at 905 * window edge, and have to drop data and PUSH from 906 * incoming segments. Continue processing, but 907 * remember to ack. Otherwise, drop segment 908 * and ack. 909 */ 910 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 911 tp->t_flags |= TF_ACKNOW; 912 TCPSTAT_INC(tcps_rcvwinprobe); 913 } else 914 goto dropafterack; 915 } else 916 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 917 m_adj(m, -todrop); 918 tlen -= todrop; 919 thflags &= ~(TH_PUSH|TH_FIN); 920 } 921 922 /* 923 * If last ACK falls within this segment's sequence numbers, 924 * record its timestamp. 925 * NOTE: 926 * 1) That the test incorporates suggestions from the latest 927 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 928 * 2) That updating only on newer timestamps interferes with 929 * our earlier PAWS tests, so this check should be solely 930 * predicated on the sequence space of this segment. 931 * 3) That we modify the segment boundary check to be 932 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 933 * instead of RFC1323's 934 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 935 * This modified check allows us to overcome RFC1323's 936 * limitations as described in Stevens TCP/IP Illustrated 937 * Vol. 2 p.869. In such cases, we can still calculate the 938 * RTT correctly when RCV.NXT == Last.ACK.Sent. 939 */ 940 if ((to->to_flags & TOF_TS) != 0 && 941 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 942 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 943 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 944 tp->ts_recent_age = tcp_ts_getticks(); 945 tp->ts_recent = to->to_tsval; 946 } 947 948 /* 949 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 950 * flag is on (half-synchronized state), then queue data for 951 * later processing; else drop segment and return. 952 */ 953 if ((thflags & TH_ACK) == 0) { 954 if (tp->t_state == TCPS_SYN_RECEIVED || 955 (tp->t_flags & TF_NEEDSYN)) 956 goto step6; 957 else if (tp->t_flags & TF_ACKNOW) 958 goto dropafterack; 959 else 960 goto drop; 961 } 962 963 /* 964 * Ack processing. 965 */ 966 switch (tp->t_state) { 967 968 /* 969 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 970 * ESTABLISHED state and continue processing. 971 * The ACK was checked above. 972 */ 973 case TCPS_SYN_RECEIVED: 974 975 TCPSTAT_INC(tcps_connects); 976 soisconnected(so); 977 /* Do window scaling? */ 978 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 979 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 980 tp->rcv_scale = tp->request_r_scale; 981 tp->snd_wnd = tiwin; 982 } 983 /* 984 * Make transitions: 985 * SYN-RECEIVED -> ESTABLISHED 986 * SYN-RECEIVED* -> FIN-WAIT-1 987 */ 988 tp->t_starttime = ticks; 989 if (tp->t_flags & TF_NEEDFIN) { 990 tcp_state_change(tp, TCPS_FIN_WAIT_1); 991 tp->t_flags &= ~TF_NEEDFIN; 992 } else { 993 tcp_state_change(tp, TCPS_ESTABLISHED); 994 TCP_PROBE5(accept__established, NULL, tp, 995 mtod(m, const char *), tp, th); 996 cc_conn_init(tp); 997 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 998 } 999 /* 1000 * If segment contains data or ACK, will call tcp_reass() 1001 * later; if not, do so now to pass queued data to user. 1002 */ 1003 if (tlen == 0 && (thflags & TH_FIN) == 0) 1004 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1005 (struct mbuf *)0); 1006 tp->snd_wl1 = th->th_seq - 1; 1007 /* FALLTHROUGH */ 1008 1009 /* 1010 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1011 * ACKs. If the ack is in the range 1012 * tp->snd_una < th->th_ack <= tp->snd_max 1013 * then advance tp->snd_una to th->th_ack and drop 1014 * data from the retransmission queue. If this ACK reflects 1015 * more up to date window information we update our window information. 1016 */ 1017 case TCPS_ESTABLISHED: 1018 case TCPS_FIN_WAIT_1: 1019 case TCPS_FIN_WAIT_2: 1020 case TCPS_CLOSE_WAIT: 1021 case TCPS_CLOSING: 1022 case TCPS_LAST_ACK: 1023 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1024 TCPSTAT_INC(tcps_rcvacktoomuch); 1025 goto dropafterack; 1026 } 1027 if ((tp->t_flags & TF_SACK_PERMIT) && 1028 ((to->to_flags & TOF_SACK) || 1029 !TAILQ_EMPTY(&tp->snd_holes))) 1030 tcp_sack_doack(tp, to, th->th_ack); 1031 else 1032 /* 1033 * Reset the value so that previous (valid) value 1034 * from the last ack with SACK doesn't get used. 1035 */ 1036 tp->sackhint.sacked_bytes = 0; 1037 1038 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1039 hhook_run_tcp_est_in(tp, th, to); 1040 1041 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1042 if (tlen == 0 && tiwin == tp->snd_wnd) { 1043 /* 1044 * If this is the first time we've seen a 1045 * FIN from the remote, this is not a 1046 * duplicate and it needs to be processed 1047 * normally. This happens during a 1048 * simultaneous close. 1049 */ 1050 if ((thflags & TH_FIN) && 1051 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 1052 tp->t_dupacks = 0; 1053 break; 1054 } 1055 TCPSTAT_INC(tcps_rcvdupack); 1056 /* 1057 * If we have outstanding data (other than 1058 * a window probe), this is a completely 1059 * duplicate ack (ie, window info didn't 1060 * change and FIN isn't set), 1061 * the ack is the biggest we've 1062 * seen and we've seen exactly our rexmt 1063 * threshold of them, assume a packet 1064 * has been dropped and retransmit it. 1065 * Kludge snd_nxt & the congestion 1066 * window so we send only this one 1067 * packet. 1068 * 1069 * We know we're losing at the current 1070 * window size so do congestion avoidance 1071 * (set ssthresh to half the current window 1072 * and pull our congestion window back to 1073 * the new ssthresh). 1074 * 1075 * Dup acks mean that packets have left the 1076 * network (they're now cached at the receiver) 1077 * so bump cwnd by the amount in the receiver 1078 * to keep a constant cwnd packets in the 1079 * network. 1080 * 1081 * When using TCP ECN, notify the peer that 1082 * we reduced the cwnd. 1083 */ 1084 if (!tcp_timer_active(tp, TT_REXMT) || 1085 th->th_ack != tp->snd_una) 1086 tp->t_dupacks = 0; 1087 else if (++tp->t_dupacks > tcprexmtthresh || 1088 IN_FASTRECOVERY(tp->t_flags)) { 1089 cc_ack_received(tp, th, CC_DUPACK); 1090 if ((tp->t_flags & TF_SACK_PERMIT) && 1091 IN_FASTRECOVERY(tp->t_flags)) { 1092 int awnd; 1093 1094 /* 1095 * Compute the amount of data in flight first. 1096 * We can inject new data into the pipe iff 1097 * we have less than 1/2 the original window's 1098 * worth of data in flight. 1099 */ 1100 if (V_tcp_do_rfc6675_pipe) 1101 awnd = tcp_compute_pipe(tp); 1102 else 1103 awnd = (tp->snd_nxt - tp->snd_fack) + 1104 tp->sackhint.sack_bytes_rexmit; 1105 1106 if (awnd < tp->snd_ssthresh) { 1107 tp->snd_cwnd += tp->t_maxseg; 1108 if (tp->snd_cwnd > tp->snd_ssthresh) 1109 tp->snd_cwnd = tp->snd_ssthresh; 1110 } 1111 } else 1112 tp->snd_cwnd += tp->t_maxseg; 1113 (void) tp->t_fb->tfb_tcp_output(tp); 1114 goto drop; 1115 } else if (tp->t_dupacks == tcprexmtthresh) { 1116 tcp_seq onxt = tp->snd_nxt; 1117 1118 /* 1119 * If we're doing sack, check to 1120 * see if we're already in sack 1121 * recovery. If we're not doing sack, 1122 * check to see if we're in newreno 1123 * recovery. 1124 */ 1125 if (tp->t_flags & TF_SACK_PERMIT) { 1126 if (IN_FASTRECOVERY(tp->t_flags)) { 1127 tp->t_dupacks = 0; 1128 break; 1129 } 1130 } else { 1131 if (SEQ_LEQ(th->th_ack, 1132 tp->snd_recover)) { 1133 tp->t_dupacks = 0; 1134 break; 1135 } 1136 } 1137 /* Congestion signal before ack. */ 1138 cc_cong_signal(tp, th, CC_NDUPACK); 1139 cc_ack_received(tp, th, CC_DUPACK); 1140 tcp_timer_activate(tp, TT_REXMT, 0); 1141 tp->t_rtttime = 0; 1142 if (tp->t_flags & TF_SACK_PERMIT) { 1143 TCPSTAT_INC( 1144 tcps_sack_recovery_episode); 1145 tp->sack_newdata = tp->snd_nxt; 1146 tp->snd_cwnd = tp->t_maxseg; 1147 (void) tp->t_fb->tfb_tcp_output(tp); 1148 goto drop; 1149 } 1150 tp->snd_nxt = th->th_ack; 1151 tp->snd_cwnd = tp->t_maxseg; 1152 (void) tp->t_fb->tfb_tcp_output(tp); 1153 KASSERT(tp->snd_limited <= 2, 1154 ("%s: tp->snd_limited too big", 1155 __func__)); 1156 tp->snd_cwnd = tp->snd_ssthresh + 1157 tp->t_maxseg * 1158 (tp->t_dupacks - tp->snd_limited); 1159 if (SEQ_GT(onxt, tp->snd_nxt)) 1160 tp->snd_nxt = onxt; 1161 goto drop; 1162 } else if (V_tcp_do_rfc3042) { 1163 /* 1164 * Process first and second duplicate 1165 * ACKs. Each indicates a segment 1166 * leaving the network, creating room 1167 * for more. Make sure we can send a 1168 * packet on reception of each duplicate 1169 * ACK by increasing snd_cwnd by one 1170 * segment. Restore the original 1171 * snd_cwnd after packet transmission. 1172 */ 1173 cc_ack_received(tp, th, CC_DUPACK); 1174 u_long oldcwnd = tp->snd_cwnd; 1175 tcp_seq oldsndmax = tp->snd_max; 1176 u_int sent; 1177 int avail; 1178 1179 KASSERT(tp->t_dupacks == 1 || 1180 tp->t_dupacks == 2, 1181 ("%s: dupacks not 1 or 2", 1182 __func__)); 1183 if (tp->t_dupacks == 1) 1184 tp->snd_limited = 0; 1185 tp->snd_cwnd = 1186 (tp->snd_nxt - tp->snd_una) + 1187 (tp->t_dupacks - tp->snd_limited) * 1188 tp->t_maxseg; 1189 /* 1190 * Only call tcp_output when there 1191 * is new data available to be sent. 1192 * Otherwise we would send pure ACKs. 1193 */ 1194 SOCKBUF_LOCK(&so->so_snd); 1195 avail = sbavail(&so->so_snd) - 1196 (tp->snd_nxt - tp->snd_una); 1197 SOCKBUF_UNLOCK(&so->so_snd); 1198 if (avail > 0) 1199 (void) tp->t_fb->tfb_tcp_output(tp); 1200 sent = tp->snd_max - oldsndmax; 1201 if (sent > tp->t_maxseg) { 1202 KASSERT((tp->t_dupacks == 2 && 1203 tp->snd_limited == 0) || 1204 (sent == tp->t_maxseg + 1 && 1205 tp->t_flags & TF_SENTFIN), 1206 ("%s: sent too much", 1207 __func__)); 1208 tp->snd_limited = 2; 1209 } else if (sent > 0) 1210 ++tp->snd_limited; 1211 tp->snd_cwnd = oldcwnd; 1212 goto drop; 1213 } 1214 } else 1215 tp->t_dupacks = 0; 1216 break; 1217 } 1218 1219 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 1220 ("%s: th_ack <= snd_una", __func__)); 1221 1222 /* 1223 * If the congestion window was inflated to account 1224 * for the other side's cached packets, retract it. 1225 */ 1226 if (IN_FASTRECOVERY(tp->t_flags)) { 1227 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1228 if (tp->t_flags & TF_SACK_PERMIT) 1229 tcp_sack_partialack(tp, th); 1230 else 1231 tcp_newreno_partial_ack(tp, th); 1232 } else 1233 cc_post_recovery(tp, th); 1234 } 1235 tp->t_dupacks = 0; 1236 /* 1237 * If we reach this point, ACK is not a duplicate, 1238 * i.e., it ACKs something we sent. 1239 */ 1240 if (tp->t_flags & TF_NEEDSYN) { 1241 /* 1242 * T/TCP: Connection was half-synchronized, and our 1243 * SYN has been ACK'd (so connection is now fully 1244 * synchronized). Go to non-starred state, 1245 * increment snd_una for ACK of SYN, and check if 1246 * we can do window scaling. 1247 */ 1248 tp->t_flags &= ~TF_NEEDSYN; 1249 tp->snd_una++; 1250 /* Do window scaling? */ 1251 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1252 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1253 tp->rcv_scale = tp->request_r_scale; 1254 /* Send window already scaled. */ 1255 } 1256 } 1257 1258process_ACK: 1259 INP_WLOCK_ASSERT(tp->t_inpcb); 1260 1261 acked = BYTES_THIS_ACK(tp, th); 1262 TCPSTAT_INC(tcps_rcvackpack); 1263 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1264 1265 /* 1266 * If we just performed our first retransmit, and the ACK 1267 * arrives within our recovery window, then it was a mistake 1268 * to do the retransmit in the first place. Recover our 1269 * original cwnd and ssthresh, and proceed to transmit where 1270 * we left off. 1271 */ 1272 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 1273 (int)(ticks - tp->t_badrxtwin) < 0) 1274 cc_cong_signal(tp, th, CC_RTO_ERR); 1275 1276 /* 1277 * If we have a timestamp reply, update smoothed 1278 * round trip time. If no timestamp is present but 1279 * transmit timer is running and timed sequence 1280 * number was acked, update smoothed round trip time. 1281 * Since we now have an rtt measurement, cancel the 1282 * timer backoff (cf., Phil Karn's retransmit alg.). 1283 * Recompute the initial retransmit timer. 1284 * 1285 * Some boxes send broken timestamp replies 1286 * during the SYN+ACK phase, ignore 1287 * timestamps of 0 or we could calculate a 1288 * huge RTT and blow up the retransmit timer. 1289 */ 1290 if ((to->to_flags & TOF_TS) != 0 && to->to_tsecr) { 1291 u_int t; 1292 1293 t = tcp_ts_getticks() - to->to_tsecr; 1294 if (!tp->t_rttlow || tp->t_rttlow > t) 1295 tp->t_rttlow = t; 1296 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 1297 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 1298 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 1299 tp->t_rttlow = ticks - tp->t_rtttime; 1300 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1301 } 1302 1303 /* 1304 * If all outstanding data is acked, stop retransmit 1305 * timer and remember to restart (more output or persist). 1306 * If there is more data to be acked, restart retransmit 1307 * timer, using current (possibly backed-off) value. 1308 */ 1309 if (th->th_ack == tp->snd_max) { 1310 tcp_timer_activate(tp, TT_REXMT, 0); 1311 needoutput = 1; 1312 } else if (!tcp_timer_active(tp, TT_PERSIST)) 1313 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 1314 1315 /* 1316 * If no data (only SYN) was ACK'd, 1317 * skip rest of ACK processing. 1318 */ 1319 if (acked == 0) 1320 goto step6; 1321 1322 /* 1323 * Let the congestion control algorithm update congestion 1324 * control related information. This typically means increasing 1325 * the congestion window. 1326 */ 1327 cc_ack_received(tp, th, CC_ACK); 1328 1329 SOCKBUF_LOCK(&so->so_snd); 1330 if (acked > sbavail(&so->so_snd)) { 1331 tp->snd_wnd -= sbavail(&so->so_snd); 1332 mfree = sbcut_locked(&so->so_snd, 1333 (int)sbavail(&so->so_snd)); 1334 ourfinisacked = 1; 1335 } else { 1336 mfree = sbcut_locked(&so->so_snd, acked); 1337 tp->snd_wnd -= acked; 1338 ourfinisacked = 0; 1339 } 1340 /* NB: sowwakeup_locked() does an implicit unlock. */ 1341 sowwakeup_locked(so); 1342 m_freem(mfree); 1343 /* Detect una wraparound. */ 1344 if (!IN_RECOVERY(tp->t_flags) && 1345 SEQ_GT(tp->snd_una, tp->snd_recover) && 1346 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1347 tp->snd_recover = th->th_ack - 1; 1348 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 1349 if (IN_RECOVERY(tp->t_flags) && 1350 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 1351 EXIT_RECOVERY(tp->t_flags); 1352 } 1353 tp->snd_una = th->th_ack; 1354 if (tp->t_flags & TF_SACK_PERMIT) { 1355 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 1356 tp->snd_recover = tp->snd_una; 1357 } 1358 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1359 tp->snd_nxt = tp->snd_una; 1360 1361 switch (tp->t_state) { 1362 1363 /* 1364 * In FIN_WAIT_1 STATE in addition to the processing 1365 * for the ESTABLISHED state if our FIN is now acknowledged 1366 * then enter FIN_WAIT_2. 1367 */ 1368 case TCPS_FIN_WAIT_1: 1369 if (ourfinisacked) { 1370 /* 1371 * If we can't receive any more 1372 * data, then closing user can proceed. 1373 * Starting the timer is contrary to the 1374 * specification, but if we don't get a FIN 1375 * we'll hang forever. 1376 * 1377 * XXXjl: 1378 * we should release the tp also, and use a 1379 * compressed state. 1380 */ 1381 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1382 soisdisconnected(so); 1383 tcp_timer_activate(tp, TT_2MSL, 1384 (tcp_fast_finwait2_recycle ? 1385 tcp_finwait2_timeout : 1386 TP_MAXIDLE(tp))); 1387 } 1388 tcp_state_change(tp, TCPS_FIN_WAIT_2); 1389 } 1390 break; 1391 1392 /* 1393 * In CLOSING STATE in addition to the processing for 1394 * the ESTABLISHED state if the ACK acknowledges our FIN 1395 * then enter the TIME-WAIT state, otherwise ignore 1396 * the segment. 1397 */ 1398 case TCPS_CLOSING: 1399 if (ourfinisacked) { 1400 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1401 tcp_twstart(tp); 1402 INP_INFO_RUNLOCK(&V_tcbinfo); 1403 m_freem(m); 1404 return; 1405 } 1406 break; 1407 1408 /* 1409 * In LAST_ACK, we may still be waiting for data to drain 1410 * and/or to be acked, as well as for the ack of our FIN. 1411 * If our FIN is now acknowledged, delete the TCB, 1412 * enter the closed state and return. 1413 */ 1414 case TCPS_LAST_ACK: 1415 if (ourfinisacked) { 1416 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1417 tp = tcp_close(tp); 1418 goto drop; 1419 } 1420 break; 1421 } 1422 } 1423 1424step6: 1425 INP_WLOCK_ASSERT(tp->t_inpcb); 1426 1427 /* 1428 * Update window information. 1429 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1430 */ 1431 if ((thflags & TH_ACK) && 1432 (SEQ_LT(tp->snd_wl1, th->th_seq) || 1433 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 1434 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 1435 /* keep track of pure window updates */ 1436 if (tlen == 0 && 1437 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1438 TCPSTAT_INC(tcps_rcvwinupd); 1439 tp->snd_wnd = tiwin; 1440 tp->snd_wl1 = th->th_seq; 1441 tp->snd_wl2 = th->th_ack; 1442 if (tp->snd_wnd > tp->max_sndwnd) 1443 tp->max_sndwnd = tp->snd_wnd; 1444 needoutput = 1; 1445 } 1446 1447 /* 1448 * Process segments with URG. 1449 */ 1450 if ((thflags & TH_URG) && th->th_urp && 1451 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1452 /* 1453 * This is a kludge, but if we receive and accept 1454 * random urgent pointers, we'll crash in 1455 * soreceive. It's hard to imagine someone 1456 * actually wanting to send this much urgent data. 1457 */ 1458 SOCKBUF_LOCK(&so->so_rcv); 1459 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 1460 th->th_urp = 0; /* XXX */ 1461 thflags &= ~TH_URG; /* XXX */ 1462 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 1463 goto dodata; /* XXX */ 1464 } 1465 /* 1466 * If this segment advances the known urgent pointer, 1467 * then mark the data stream. This should not happen 1468 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1469 * a FIN has been received from the remote side. 1470 * In these states we ignore the URG. 1471 * 1472 * According to RFC961 (Assigned Protocols), 1473 * the urgent pointer points to the last octet 1474 * of urgent data. We continue, however, 1475 * to consider it to indicate the first octet 1476 * of data past the urgent section as the original 1477 * spec states (in one of two places). 1478 */ 1479 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 1480 tp->rcv_up = th->th_seq + th->th_urp; 1481 so->so_oobmark = sbavail(&so->so_rcv) + 1482 (tp->rcv_up - tp->rcv_nxt) - 1; 1483 if (so->so_oobmark == 0) 1484 so->so_rcv.sb_state |= SBS_RCVATMARK; 1485 sohasoutofband(so); 1486 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1487 } 1488 SOCKBUF_UNLOCK(&so->so_rcv); 1489 /* 1490 * Remove out of band data so doesn't get presented to user. 1491 * This can happen independent of advancing the URG pointer, 1492 * but if two URG's are pending at once, some out-of-band 1493 * data may creep in... ick. 1494 */ 1495 if (th->th_urp <= (u_long)tlen && 1496 !(so->so_options & SO_OOBINLINE)) { 1497 /* hdr drop is delayed */ 1498 tcp_pulloutofband(so, th, m, drop_hdrlen); 1499 } 1500 } else { 1501 /* 1502 * If no out of band data is expected, 1503 * pull receive urgent pointer along 1504 * with the receive window. 1505 */ 1506 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1507 tp->rcv_up = tp->rcv_nxt; 1508 } 1509dodata: /* XXX */ 1510 INP_WLOCK_ASSERT(tp->t_inpcb); 1511 1512 /* 1513 * Process the segment text, merging it into the TCP sequencing queue, 1514 * and arranging for acknowledgment of receipt if necessary. 1515 * This process logically involves adjusting tp->rcv_wnd as data 1516 * is presented to the user (this happens in tcp_usrreq.c, 1517 * case PRU_RCVD). If a FIN has already been received on this 1518 * connection then we just ignore the text. 1519 */ 1520 if ((tlen || (thflags & TH_FIN)) && 1521 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1522 tcp_seq save_start = th->th_seq; 1523 m_adj(m, drop_hdrlen); /* delayed header drop */ 1524 /* 1525 * Insert segment which includes th into TCP reassembly queue 1526 * with control block tp. Set thflags to whether reassembly now 1527 * includes a segment with FIN. This handles the common case 1528 * inline (segment is the next to be received on an established 1529 * connection, and the queue is empty), avoiding linkage into 1530 * and removal from the queue and repetition of various 1531 * conversions. 1532 * Set DELACK for segments received in order, but ack 1533 * immediately when segments are out of order (so 1534 * fast retransmit can work). 1535 */ 1536 if (th->th_seq == tp->rcv_nxt && 1537 LIST_EMPTY(&tp->t_segq) && 1538 TCPS_HAVEESTABLISHED(tp->t_state)) { 1539 if (DELAY_ACK(tp, tlen)) 1540 tp->t_flags |= TF_DELACK; 1541 else 1542 tp->t_flags |= TF_ACKNOW; 1543 tp->rcv_nxt += tlen; 1544 thflags = th->th_flags & TH_FIN; 1545 TCPSTAT_INC(tcps_rcvpack); 1546 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1547 SOCKBUF_LOCK(&so->so_rcv); 1548 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 1549 m_freem(m); 1550 else 1551 sbappendstream_locked(&so->so_rcv, m, 0); 1552 /* NB: sorwakeup_locked() does an implicit unlock. */ 1553 sorwakeup_locked(so); 1554 } else { 1555 /* 1556 * XXX: Due to the header drop above "th" is 1557 * theoretically invalid by now. Fortunately 1558 * m_adj() doesn't actually frees any mbufs 1559 * when trimming from the head. 1560 */ 1561 thflags = tcp_reass(tp, th, &tlen, m); 1562 tp->t_flags |= TF_ACKNOW; 1563 } 1564 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 1565 tcp_update_sack_list(tp, save_start, save_start + tlen); 1566#if 0 1567 /* 1568 * Note the amount of data that peer has sent into 1569 * our window, in order to estimate the sender's 1570 * buffer size. 1571 * XXX: Unused. 1572 */ 1573 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 1574 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1575 else 1576 len = so->so_rcv.sb_hiwat; 1577#endif 1578 } else { 1579 m_freem(m); 1580 thflags &= ~TH_FIN; 1581 } 1582 1583 /* 1584 * If FIN is received ACK the FIN and let the user know 1585 * that the connection is closing. 1586 */ 1587 if (thflags & TH_FIN) { 1588 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1589 socantrcvmore(so); 1590 /* 1591 * If connection is half-synchronized 1592 * (ie NEEDSYN flag on) then delay ACK, 1593 * so it may be piggybacked when SYN is sent. 1594 * Otherwise, since we received a FIN then no 1595 * more input can be expected, send ACK now. 1596 */ 1597 if (tp->t_flags & TF_NEEDSYN) 1598 tp->t_flags |= TF_DELACK; 1599 else 1600 tp->t_flags |= TF_ACKNOW; 1601 tp->rcv_nxt++; 1602 } 1603 switch (tp->t_state) { 1604 1605 /* 1606 * In SYN_RECEIVED and ESTABLISHED STATES 1607 * enter the CLOSE_WAIT state. 1608 */ 1609 case TCPS_SYN_RECEIVED: 1610 tp->t_starttime = ticks; 1611 /* FALLTHROUGH */ 1612 case TCPS_ESTABLISHED: 1613 tcp_state_change(tp, TCPS_CLOSE_WAIT); 1614 break; 1615 1616 /* 1617 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1618 * enter the CLOSING state. 1619 */ 1620 case TCPS_FIN_WAIT_1: 1621 tcp_state_change(tp, TCPS_CLOSING); 1622 break; 1623 1624 /* 1625 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1626 * starting the time-wait timer, turning off the other 1627 * standard timers. 1628 */ 1629 case TCPS_FIN_WAIT_2: 1630 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1631 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 1632 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 1633 ti_locked)); 1634 1635 tcp_twstart(tp); 1636 INP_INFO_RUNLOCK(&V_tcbinfo); 1637 return; 1638 } 1639 } 1640 if (ti_locked == TI_RLOCKED) { 1641 INP_INFO_RUNLOCK(&V_tcbinfo); 1642 } 1643 ti_locked = TI_UNLOCKED; 1644 1645#ifdef TCPDEBUG 1646 if (so->so_options & SO_DEBUG) 1647 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 1648 &tcp_savetcp, 0); 1649#endif 1650 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 1651 1652 /* 1653 * Return any desired output. 1654 */ 1655 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1656 (void) tp->t_fb->tfb_tcp_output(tp); 1657 1658 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 1659 __func__, ti_locked)); 1660 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1661 INP_WLOCK_ASSERT(tp->t_inpcb); 1662 1663 if (tp->t_flags & TF_DELACK) { 1664 tp->t_flags &= ~TF_DELACK; 1665 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 1666 } 1667 INP_WUNLOCK(tp->t_inpcb); 1668 return; 1669 1670dropafterack: 1671 /* 1672 * Generate an ACK dropping incoming segment if it occupies 1673 * sequence space, where the ACK reflects our state. 1674 * 1675 * We can now skip the test for the RST flag since all 1676 * paths to this code happen after packets containing 1677 * RST have been dropped. 1678 * 1679 * In the SYN-RECEIVED state, don't send an ACK unless the 1680 * segment we received passes the SYN-RECEIVED ACK test. 1681 * If it fails send a RST. This breaks the loop in the 1682 * "LAND" DoS attack, and also prevents an ACK storm 1683 * between two listening ports that have been sent forged 1684 * SYN segments, each with the source address of the other. 1685 */ 1686 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 1687 (SEQ_GT(tp->snd_una, th->th_ack) || 1688 SEQ_GT(th->th_ack, tp->snd_max)) ) { 1689 rstreason = BANDLIM_RST_OPENPORT; 1690 goto dropwithreset; 1691 } 1692#ifdef TCPDEBUG 1693 if (so->so_options & SO_DEBUG) 1694 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 1695 &tcp_savetcp, 0); 1696#endif 1697 TCP_PROBE3(debug__drop, tp, th, mtod(m, const char *)); 1698 if (ti_locked == TI_RLOCKED) { 1699 INP_INFO_RUNLOCK(&V_tcbinfo); 1700 } 1701 ti_locked = TI_UNLOCKED; 1702 1703 tp->t_flags |= TF_ACKNOW; 1704 (void) tp->t_fb->tfb_tcp_output(tp); 1705 INP_WUNLOCK(tp->t_inpcb); 1706 m_freem(m); 1707 return; 1708 1709dropwithreset: 1710 if (ti_locked == TI_RLOCKED) { 1711 INP_INFO_RUNLOCK(&V_tcbinfo); 1712 } 1713 ti_locked = TI_UNLOCKED; 1714 1715 if (tp != NULL) { 1716 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1717 INP_WUNLOCK(tp->t_inpcb); 1718 } else 1719 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1720 return; 1721 1722drop: 1723 if (ti_locked == TI_RLOCKED) { 1724 INP_INFO_RUNLOCK(&V_tcbinfo); 1725 ti_locked = TI_UNLOCKED; 1726 } 1727#ifdef INVARIANTS 1728 else 1729 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1730#endif 1731 1732 /* 1733 * Drop space held by incoming segment and return. 1734 */ 1735#ifdef TCPDEBUG 1736 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1737 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 1738 &tcp_savetcp, 0); 1739#endif 1740 TCP_PROBE3(debug__drop, tp, th, mtod(m, const char *)); 1741 if (tp != NULL) 1742 INP_WUNLOCK(tp->t_inpcb); 1743 m_freem(m); 1744} 1745 1746 1747/* 1748 * Do fast slow is a combination of the original 1749 * tcp_dosegment and a split fastpath, one function 1750 * for the fast-ack which also includes allowing fastpath 1751 * for window advanced in sequence acks. And also a 1752 * sub-function that handles the insequence data. 1753 */ 1754void 1755tcp_do_segment_fastslow(struct mbuf *m, struct tcphdr *th, struct socket *so, 1756 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1757 int ti_locked) 1758{ 1759 int thflags; 1760 u_long tiwin; 1761 char *s; 1762 int can_enter; 1763 struct in_conninfo *inc; 1764 struct tcpopt to; 1765 1766 thflags = th->th_flags; 1767 tp->sackhint.last_sack_ack = 0; 1768 inc = &tp->t_inpcb->inp_inc; 1769 /* 1770 * If this is either a state-changing packet or current state isn't 1771 * established, we require a write lock on tcbinfo. Otherwise, we 1772 * allow the tcbinfo to be in either alocked or unlocked, as the 1773 * caller may have unnecessarily acquired a write lock due to a race. 1774 */ 1775 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1776 tp->t_state != TCPS_ESTABLISHED) { 1777 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1778 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1779 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1780 } else { 1781#ifdef INVARIANTS 1782 if (ti_locked == TI_RLOCKED) { 1783 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1784 } else { 1785 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1786 "ti_locked: %d", __func__, ti_locked)); 1787 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1788 } 1789#endif 1790 } 1791 INP_WLOCK_ASSERT(tp->t_inpcb); 1792 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1793 __func__)); 1794 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1795 __func__)); 1796 1797 /* 1798 * Segment received on connection. 1799 * Reset idle time and keep-alive timer. 1800 * XXX: This should be done after segment 1801 * validation to ignore broken/spoofed segs. 1802 */ 1803 tp->t_rcvtime = ticks; 1804 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1805 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1806 1807 /* 1808 * Unscale the window into a 32-bit value. 1809 * For the SYN_SENT state the scale is zero. 1810 */ 1811 tiwin = th->th_win << tp->snd_scale; 1812 1813 /* 1814 * TCP ECN processing. 1815 */ 1816 if (tp->t_flags & TF_ECN_PERMIT) { 1817 if (thflags & TH_CWR) 1818 tp->t_flags &= ~TF_ECN_SND_ECE; 1819 switch (iptos & IPTOS_ECN_MASK) { 1820 case IPTOS_ECN_CE: 1821 tp->t_flags |= TF_ECN_SND_ECE; 1822 TCPSTAT_INC(tcps_ecn_ce); 1823 break; 1824 case IPTOS_ECN_ECT0: 1825 TCPSTAT_INC(tcps_ecn_ect0); 1826 break; 1827 case IPTOS_ECN_ECT1: 1828 TCPSTAT_INC(tcps_ecn_ect1); 1829 break; 1830 } 1831 /* Congestion experienced. */ 1832 if (thflags & TH_ECE) { 1833 cc_cong_signal(tp, th, CC_ECN); 1834 } 1835 } 1836 1837 /* 1838 * Parse options on any incoming segment. 1839 */ 1840 tcp_dooptions(&to, (u_char *)(th + 1), 1841 (th->th_off << 2) - sizeof(struct tcphdr), 1842 (thflags & TH_SYN) ? TO_SYN : 0); 1843 1844 /* 1845 * If echoed timestamp is later than the current time, 1846 * fall back to non RFC1323 RTT calculation. Normalize 1847 * timestamp if syncookies were used when this connection 1848 * was established. 1849 */ 1850 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1851 to.to_tsecr -= tp->ts_offset; 1852 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1853 to.to_tsecr = 0; 1854 } 1855 /* 1856 * If timestamps were negotiated during SYN/ACK they should 1857 * appear on every segment during this session and vice versa. 1858 */ 1859 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1860 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1861 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1862 "no action\n", s, __func__); 1863 free(s, M_TCPLOG); 1864 } 1865 } 1866 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1867 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1868 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1869 "no action\n", s, __func__); 1870 free(s, M_TCPLOG); 1871 } 1872 } 1873 1874 /* 1875 * Process options only when we get SYN/ACK back. The SYN case 1876 * for incoming connections is handled in tcp_syncache. 1877 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1878 * or <SYN,ACK>) segment itself is never scaled. 1879 * XXX this is traditional behavior, may need to be cleaned up. 1880 */ 1881 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1882 if ((to.to_flags & TOF_SCALE) && 1883 (tp->t_flags & TF_REQ_SCALE)) { 1884 tp->t_flags |= TF_RCVD_SCALE; 1885 tp->snd_scale = to.to_wscale; 1886 } 1887 /* 1888 * Initial send window. It will be updated with 1889 * the next incoming segment to the scaled value. 1890 */ 1891 tp->snd_wnd = th->th_win; 1892 if (to.to_flags & TOF_TS) { 1893 tp->t_flags |= TF_RCVD_TSTMP; 1894 tp->ts_recent = to.to_tsval; 1895 tp->ts_recent_age = tcp_ts_getticks(); 1896 } 1897 if (to.to_flags & TOF_MSS) 1898 tcp_mss(tp, to.to_mss); 1899 if ((tp->t_flags & TF_SACK_PERMIT) && 1900 (to.to_flags & TOF_SACKPERM) == 0) 1901 tp->t_flags &= ~TF_SACK_PERMIT; 1902 } 1903 can_enter = 0; 1904 if (__predict_true((tlen == 0))) { 1905 /* 1906 * The ack moved forward and we have a window (non-zero) 1907 * <or> 1908 * The ack did not move forward, but the window increased. 1909 */ 1910 if (__predict_true((SEQ_GT(th->th_ack, tp->snd_una) && tiwin) || 1911 ((th->th_ack == tp->snd_una) && tiwin && (tiwin > tp->snd_wnd)))) { 1912 can_enter = 1; 1913 } 1914 } else { 1915 /* 1916 * Data incoming, use the old entry criteria 1917 * for fast-path with data. 1918 */ 1919 if ((tiwin && tiwin == tp->snd_wnd)) { 1920 can_enter = 1; 1921 } 1922 } 1923 /* 1924 * Header prediction: check for the two common cases 1925 * of a uni-directional data xfer. If the packet has 1926 * no control flags, is in-sequence, the window didn't 1927 * change and we're not retransmitting, it's a 1928 * candidate. If the length is zero and the ack moved 1929 * forward, we're the sender side of the xfer. Just 1930 * free the data acked & wake any higher level process 1931 * that was blocked waiting for space. If the length 1932 * is non-zero and the ack didn't move, we're the 1933 * receiver side. If we're getting packets in-order 1934 * (the reassembly queue is empty), add the data to 1935 * the socket buffer and note that we need a delayed ack. 1936 * Make sure that the hidden state-flags are also off. 1937 * Since we check for TCPS_ESTABLISHED first, it can only 1938 * be TH_NEEDSYN. 1939 */ 1940 if (__predict_true(tp->t_state == TCPS_ESTABLISHED && 1941 th->th_seq == tp->rcv_nxt && 1942 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1943 tp->snd_nxt == tp->snd_max && 1944 can_enter && 1945 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1946 LIST_EMPTY(&tp->t_segq) && 1947 ((to.to_flags & TOF_TS) == 0 || 1948 TSTMP_GEQ(to.to_tsval, tp->ts_recent)))) { 1949 if (__predict_true((tlen == 0) && 1950 (SEQ_LEQ(th->th_ack, tp->snd_max) && 1951 !IN_RECOVERY(tp->t_flags) && 1952 (to.to_flags & TOF_SACK) == 0 && 1953 TAILQ_EMPTY(&tp->snd_holes)))) { 1954 /* We are done */ 1955 tcp_do_fastack(m, th, so, tp, &to, drop_hdrlen, tlen, 1956 ti_locked, tiwin); 1957 return; 1958 } else if ((tlen) && 1959 (th->th_ack == tp->snd_una && 1960 tlen <= sbspace(&so->so_rcv))) { 1961 tcp_do_fastnewdata(m, th, so, tp, &to, drop_hdrlen, tlen, 1962 ti_locked, tiwin); 1963 /* We are done */ 1964 return; 1965 } 1966 } 1967 tcp_do_slowpath(m, th, so, tp, &to, drop_hdrlen, tlen, 1968 ti_locked, tiwin, thflags); 1969} 1970 1971 1972/* 1973 * This subfunction is used to try to highly optimize the 1974 * fast path. We again allow window updates that are 1975 * in sequence to remain in the fast-path. We also add 1976 * in the __predict's to attempt to help the compiler. 1977 * Note that if we return a 0, then we can *not* process 1978 * it and the caller should push the packet into the 1979 * slow-path. 1980 */ 1981static int 1982tcp_fastack(struct mbuf *m, struct tcphdr *th, struct socket *so, 1983 struct tcpcb *tp, struct tcpopt *to, int drop_hdrlen, int tlen, 1984 int ti_locked, u_long tiwin) 1985{ 1986 int acked; 1987 int winup_only=0; 1988#ifdef TCPDEBUG 1989 /* 1990 * The size of tcp_saveipgen must be the size of the max ip header, 1991 * now IPv6. 1992 */ 1993 u_char tcp_saveipgen[IP6_HDR_LEN]; 1994 struct tcphdr tcp_savetcp; 1995 short ostate = 0; 1996#endif 1997 1998 1999 if (__predict_false(SEQ_LEQ(th->th_ack, tp->snd_una))) { 2000 /* Old ack, behind (or duplicate to) the last one rcv'd */ 2001 return (0); 2002 } 2003 if (__predict_false(th->th_ack == tp->snd_una) && 2004 __predict_false(tiwin <= tp->snd_wnd)) { 2005 /* duplicate ack <or> a shrinking dup ack with shrinking window */ 2006 return (0); 2007 } 2008 if (__predict_false(tiwin == 0)) { 2009 /* zero window */ 2010 return (0); 2011 } 2012 if (__predict_false(SEQ_GT(th->th_ack, tp->snd_max))) { 2013 /* Above what we have sent? */ 2014 return (0); 2015 } 2016 if (__predict_false(tp->snd_nxt != tp->snd_max)) { 2017 /* We are retransmitting */ 2018 return (0); 2019 } 2020 if (__predict_false(tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN))) { 2021 /* We need a SYN or a FIN, unlikely.. */ 2022 return (0); 2023 } 2024 if((to->to_flags & TOF_TS) && __predict_false(TSTMP_LT(to->to_tsval, tp->ts_recent))) { 2025 /* Timestamp is behind .. old ack with seq wrap? */ 2026 return (0); 2027 } 2028 if (__predict_false(IN_RECOVERY(tp->t_flags))) { 2029 /* Still recovering */ 2030 return (0); 2031 } 2032 if (__predict_false(to->to_flags & TOF_SACK)) { 2033 /* Sack included in the ack.. */ 2034 return (0); 2035 } 2036 if (!TAILQ_EMPTY(&tp->snd_holes)) { 2037 /* We have sack holes on our scoreboard */ 2038 return (0); 2039 } 2040 /* Ok if we reach here, we can process a fast-ack */ 2041 2042 /* Did the window get updated? */ 2043 if (tiwin != tp->snd_wnd) { 2044 /* keep track of pure window updates */ 2045 if (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) { 2046 winup_only = 1; 2047 TCPSTAT_INC(tcps_rcvwinupd); 2048 } 2049 tp->snd_wnd = tiwin; 2050 tp->snd_wl1 = th->th_seq; 2051 if (tp->snd_wnd > tp->max_sndwnd) 2052 tp->max_sndwnd = tp->snd_wnd; 2053 } 2054 /* 2055 * Pull snd_wl2 up to prevent seq wrap relative 2056 * to th_ack. 2057 */ 2058 tp->snd_wl2 = th->th_ack; 2059 /* 2060 * If last ACK falls within this segment's sequence numbers, 2061 * record the timestamp. 2062 * NOTE that the test is modified according to the latest 2063 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2064 */ 2065 if ((to->to_flags & TOF_TS) != 0 && 2066 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 2067 tp->ts_recent_age = tcp_ts_getticks(); 2068 tp->ts_recent = to->to_tsval; 2069 } 2070 /* 2071 * This is a pure ack for outstanding data. 2072 */ 2073 if (ti_locked == TI_RLOCKED) { 2074 INP_INFO_RUNLOCK(&V_tcbinfo); 2075 } 2076 ti_locked = TI_UNLOCKED; 2077 2078 TCPSTAT_INC(tcps_predack); 2079 2080 /* 2081 * "bad retransmit" recovery. 2082 */ 2083 if (tp->t_rxtshift == 1 && 2084 tp->t_flags & TF_PREVVALID && 2085 (int)(ticks - tp->t_badrxtwin) < 0) { 2086 cc_cong_signal(tp, th, CC_RTO_ERR); 2087 } 2088 2089 /* 2090 * Recalculate the transmit timer / rtt. 2091 * 2092 * Some boxes send broken timestamp replies 2093 * during the SYN+ACK phase, ignore 2094 * timestamps of 0 or we could calculate a 2095 * huge RTT and blow up the retransmit timer. 2096 */ 2097 if ((to->to_flags & TOF_TS) != 0 && 2098 to->to_tsecr) { 2099 u_int t; 2100 2101 t = tcp_ts_getticks() - to->to_tsecr; 2102 if (!tp->t_rttlow || tp->t_rttlow > t) 2103 tp->t_rttlow = t; 2104 tcp_xmit_timer(tp, 2105 TCP_TS_TO_TICKS(t) + 1); 2106 } else if (tp->t_rtttime && 2107 SEQ_GT(th->th_ack, tp->t_rtseq)) { 2108 if (!tp->t_rttlow || 2109 tp->t_rttlow > ticks - tp->t_rtttime) 2110 tp->t_rttlow = ticks - tp->t_rtttime; 2111 tcp_xmit_timer(tp, 2112 ticks - tp->t_rtttime); 2113 } 2114 if (winup_only == 0) { 2115 acked = BYTES_THIS_ACK(tp, th); 2116 2117 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2118 hhook_run_tcp_est_in(tp, th, to); 2119 2120 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2121 sbdrop(&so->so_snd, acked); 2122 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 2123 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2124 tp->snd_recover = th->th_ack - 1; 2125 2126 /* 2127 * Let the congestion control algorithm update 2128 * congestion control related information. This 2129 * typically means increasing the congestion 2130 * window. 2131 */ 2132 cc_ack_received(tp, th, CC_ACK); 2133 2134 tp->snd_una = th->th_ack; 2135 tp->t_dupacks = 0; 2136 2137 /* 2138 * If all outstanding data are acked, stop 2139 * retransmit timer, otherwise restart timer 2140 * using current (possibly backed-off) value. 2141 * If process is waiting for space, 2142 * wakeup/selwakeup/signal. If data 2143 * are ready to send, let tcp_output 2144 * decide between more output or persist. 2145 */ 2146#ifdef TCPDEBUG 2147 if (so->so_options & SO_DEBUG) 2148 tcp_trace(TA_INPUT, ostate, tp, 2149 (void *)tcp_saveipgen, 2150 &tcp_savetcp, 0); 2151#endif 2152 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 2153 m_freem(m); 2154 if (tp->snd_una == tp->snd_max) 2155 tcp_timer_activate(tp, TT_REXMT, 0); 2156 else if (!tcp_timer_active(tp, TT_PERSIST)) 2157 tcp_timer_activate(tp, TT_REXMT, 2158 tp->t_rxtcur); 2159 /* Wake up the socket if we have room to write more */ 2160 sowwakeup(so); 2161 } else { 2162 /* 2163 * Window update only, just free the mbufs and 2164 * send out whatever we can. 2165 */ 2166 m_freem(m); 2167 } 2168 if (sbavail(&so->so_snd)) 2169 (void) tcp_output(tp); 2170 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 2171 __func__, ti_locked)); 2172 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 2173 INP_WLOCK_ASSERT(tp->t_inpcb); 2174 2175 if (tp->t_flags & TF_DELACK) { 2176 tp->t_flags &= ~TF_DELACK; 2177 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2178 } 2179 INP_WUNLOCK(tp->t_inpcb); 2180 return (1); 2181} 2182 2183/* 2184 * This tcp-do-segment concentrates on making the fastest 2185 * ack processing path. It does not have a fast-path for 2186 * data (it possibly could which would then eliminate the 2187 * need for fast-slow above). For a content distributor having 2188 * large outgoing elephants and very very little coming in 2189 * having no fastpath for data does not really help (since you 2190 * don't get much data in). The most important thing is 2191 * processing ack's quickly and getting the rest of the data 2192 * output to the peer as quickly as possible. This routine 2193 * seems to be about an overall 3% faster then the old 2194 * tcp_do_segment and keeps us in the fast-path for packets 2195 * much more (by allowing window updates to also stay in the fastpath). 2196 */ 2197void 2198tcp_do_segment_fastack(struct mbuf *m, struct tcphdr *th, struct socket *so, 2199 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 2200 int ti_locked) 2201{ 2202 int thflags; 2203 u_long tiwin; 2204 char *s; 2205 struct in_conninfo *inc; 2206 struct tcpopt to; 2207 2208 thflags = th->th_flags; 2209 tp->sackhint.last_sack_ack = 0; 2210 inc = &tp->t_inpcb->inp_inc; 2211 /* 2212 * If this is either a state-changing packet or current state isn't 2213 * established, we require a write lock on tcbinfo. Otherwise, we 2214 * allow the tcbinfo to be in either alocked or unlocked, as the 2215 * caller may have unnecessarily acquired a write lock due to a race. 2216 */ 2217 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 2218 tp->t_state != TCPS_ESTABLISHED) { 2219 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 2220 "SYN/FIN/RST/!EST", __func__, ti_locked)); 2221 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2222 } else { 2223#ifdef INVARIANTS 2224 if (ti_locked == TI_RLOCKED) { 2225 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2226 } else { 2227 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 2228 "ti_locked: %d", __func__, ti_locked)); 2229 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 2230 } 2231#endif 2232 } 2233 INP_WLOCK_ASSERT(tp->t_inpcb); 2234 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 2235 __func__)); 2236 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 2237 __func__)); 2238 2239 /* 2240 * Segment received on connection. 2241 * Reset idle time and keep-alive timer. 2242 * XXX: This should be done after segment 2243 * validation to ignore broken/spoofed segs. 2244 */ 2245 tp->t_rcvtime = ticks; 2246 if (TCPS_HAVEESTABLISHED(tp->t_state)) 2247 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2248 2249 /* 2250 * Unscale the window into a 32-bit value. 2251 * For the SYN_SENT state the scale is zero. 2252 */ 2253 tiwin = th->th_win << tp->snd_scale; 2254 2255 /* 2256 * TCP ECN processing. 2257 */ 2258 if (tp->t_flags & TF_ECN_PERMIT) { 2259 if (thflags & TH_CWR) 2260 tp->t_flags &= ~TF_ECN_SND_ECE; 2261 switch (iptos & IPTOS_ECN_MASK) { 2262 case IPTOS_ECN_CE: 2263 tp->t_flags |= TF_ECN_SND_ECE; 2264 TCPSTAT_INC(tcps_ecn_ce); 2265 break; 2266 case IPTOS_ECN_ECT0: 2267 TCPSTAT_INC(tcps_ecn_ect0); 2268 break; 2269 case IPTOS_ECN_ECT1: 2270 TCPSTAT_INC(tcps_ecn_ect1); 2271 break; 2272 } 2273 /* Congestion experienced. */ 2274 if (thflags & TH_ECE) { 2275 cc_cong_signal(tp, th, CC_ECN); 2276 } 2277 } 2278 2279 /* 2280 * Parse options on any incoming segment. 2281 */ 2282 tcp_dooptions(&to, (u_char *)(th + 1), 2283 (th->th_off << 2) - sizeof(struct tcphdr), 2284 (thflags & TH_SYN) ? TO_SYN : 0); 2285 2286 /* 2287 * If echoed timestamp is later than the current time, 2288 * fall back to non RFC1323 RTT calculation. Normalize 2289 * timestamp if syncookies were used when this connection 2290 * was established. 2291 */ 2292 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 2293 to.to_tsecr -= tp->ts_offset; 2294 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 2295 to.to_tsecr = 0; 2296 } 2297 /* 2298 * If timestamps were negotiated during SYN/ACK they should 2299 * appear on every segment during this session and vice versa. 2300 */ 2301 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 2302 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2303 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 2304 "no action\n", s, __func__); 2305 free(s, M_TCPLOG); 2306 } 2307 } 2308 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 2309 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2310 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 2311 "no action\n", s, __func__); 2312 free(s, M_TCPLOG); 2313 } 2314 } 2315 2316 /* 2317 * Process options only when we get SYN/ACK back. The SYN case 2318 * for incoming connections is handled in tcp_syncache. 2319 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 2320 * or <SYN,ACK>) segment itself is never scaled. 2321 * XXX this is traditional behavior, may need to be cleaned up. 2322 */ 2323 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 2324 if ((to.to_flags & TOF_SCALE) && 2325 (tp->t_flags & TF_REQ_SCALE)) { 2326 tp->t_flags |= TF_RCVD_SCALE; 2327 tp->snd_scale = to.to_wscale; 2328 } 2329 /* 2330 * Initial send window. It will be updated with 2331 * the next incoming segment to the scaled value. 2332 */ 2333 tp->snd_wnd = th->th_win; 2334 if (to.to_flags & TOF_TS) { 2335 tp->t_flags |= TF_RCVD_TSTMP; 2336 tp->ts_recent = to.to_tsval; 2337 tp->ts_recent_age = tcp_ts_getticks(); 2338 } 2339 if (to.to_flags & TOF_MSS) 2340 tcp_mss(tp, to.to_mss); 2341 if ((tp->t_flags & TF_SACK_PERMIT) && 2342 (to.to_flags & TOF_SACKPERM) == 0) 2343 tp->t_flags &= ~TF_SACK_PERMIT; 2344 } 2345 /* 2346 * Header prediction: check for the two common cases 2347 * of a uni-directional data xfer. If the packet has 2348 * no control flags, is in-sequence, the window didn't 2349 * change and we're not retransmitting, it's a 2350 * candidate. If the length is zero and the ack moved 2351 * forward, we're the sender side of the xfer. Just 2352 * free the data acked & wake any higher level process 2353 * that was blocked waiting for space. If the length 2354 * is non-zero and the ack didn't move, we're the 2355 * receiver side. If we're getting packets in-order 2356 * (the reassembly queue is empty), add the data to 2357 * the socket buffer and note that we need a delayed ack. 2358 * Make sure that the hidden state-flags are also off. 2359 * Since we check for TCPS_ESTABLISHED first, it can only 2360 * be TH_NEEDSYN. 2361 */ 2362 if (__predict_true(tp->t_state == TCPS_ESTABLISHED) && 2363 __predict_true(((to.to_flags & TOF_SACK) == 0)) && 2364 __predict_true(tlen == 0) && 2365 __predict_true((thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK) && 2366 __predict_true(LIST_EMPTY(&tp->t_segq)) && 2367 __predict_true(th->th_seq == tp->rcv_nxt)) { 2368 if (tcp_fastack(m, th, so, tp, &to, drop_hdrlen, tlen, 2369 ti_locked, tiwin)) { 2370 return; 2371 } 2372 } 2373 tcp_do_slowpath(m, th, so, tp, &to, drop_hdrlen, tlen, 2374 ti_locked, tiwin, thflags); 2375} 2376 2377struct tcp_function_block __tcp_fastslow = { 2378 .tfb_tcp_block_name = "fastslow", 2379 .tfb_tcp_output = tcp_output, 2380 .tfb_tcp_do_segment = tcp_do_segment_fastslow, 2381 .tfb_tcp_ctloutput = tcp_default_ctloutput, 2382}; 2383 2384struct tcp_function_block __tcp_fastack = { 2385 .tfb_tcp_block_name = "fastack", 2386 .tfb_tcp_output = tcp_output, 2387 .tfb_tcp_do_segment = tcp_do_segment_fastack, 2388 .tfb_tcp_ctloutput = tcp_default_ctloutput 2389}; 2390 2391static int 2392tcp_addfastpaths(module_t mod, int type, void *data) 2393{ 2394 int err=0; 2395 2396 switch (type) { 2397 case MOD_LOAD: 2398 err = register_tcp_functions(&__tcp_fastack, M_WAITOK); 2399 if (err) { 2400 printf("Failed to register fastack module -- err:%d\n", err); 2401 return(err); 2402 } 2403 err = register_tcp_functions(&__tcp_fastslow, M_WAITOK); 2404 if (err) { 2405 printf("Failed to register fastslow module -- err:%d\n", err); 2406 deregister_tcp_functions(&__tcp_fastack); 2407 return(err); 2408 } 2409 break; 2410 case MOD_QUIESCE: 2411 if ((__tcp_fastslow.tfb_refcnt) ||( __tcp_fastack.tfb_refcnt)) { 2412 return(EBUSY); 2413 } 2414 break; 2415 case MOD_UNLOAD: 2416 err = deregister_tcp_functions(&__tcp_fastack); 2417 if (err == EBUSY) 2418 break; 2419 err = deregister_tcp_functions(&__tcp_fastslow); 2420 if (err == EBUSY) 2421 break; 2422 err = 0; 2423 break; 2424 default: 2425 return (EOPNOTSUPP); 2426 } 2427 return (err); 2428} 2429 2430static moduledata_t new_tcp_fastpaths = { 2431 .name = "tcp_fastpaths", 2432 .evhand = tcp_addfastpaths, 2433 .priv = 0 2434}; 2435 2436MODULE_VERSION(kern_tcpfastpaths, 1); 2437DECLARE_MODULE(kern_tcpfastpaths, new_tcp_fastpaths, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY); 2438