2 * SPDX-License-Identifier: BSD-3-Clause
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.
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.
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.
22 * Portions of this software were developed by Robert N. M. Watson under
23 * contract to Juniper Networks, Inc.
25 * Redistribution and use in source and binary forms, with or without
26 * modification, are permitted provided that the following conditions
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.
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
49 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
52 #include <sys/cdefs.h>
53 __FBSDID("$FreeBSD$");
56 #include "opt_inet6.h"
57 #include "opt_ipsec.h"
58 #include "opt_tcpdebug.h"
60 #include <sys/param.h>
62 #include <sys/kernel.h>
64 #include <sys/hhook.h>
66 #include <sys/malloc.h>
68 #include <sys/proc.h> /* for proc0 declaration */
69 #include <sys/protosw.h>
70 #include <sys/qmath.h>
72 #include <sys/signalvar.h>
73 #include <sys/socket.h>
74 #include <sys/socketvar.h>
75 #include <sys/sysctl.h>
76 #include <sys/syslog.h>
77 #include <sys/systm.h>
78 #include <sys/stats.h>
80 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
85 #include <net/if_var.h>
86 #include <net/route.h>
89 #define TCPSTATES /* for logging */
91 #include <netinet/in.h>
92 #include <netinet/in_kdtrace.h>
93 #include <netinet/in_pcb.h>
94 #include <netinet/in_systm.h>
95 #include <netinet/ip.h>
96 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
97 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
98 #include <netinet/ip_var.h>
99 #include <netinet/ip_options.h>
100 #include <netinet/ip6.h>
101 #include <netinet/icmp6.h>
102 #include <netinet6/in6_pcb.h>
103 #include <netinet6/in6_var.h>
104 #include <netinet6/ip6_var.h>
105 #include <netinet6/nd6.h>
106 #include <netinet/tcp.h>
107 #include <netinet/tcp_fsm.h>
108 #include <netinet/tcp_log_buf.h>
109 #include <netinet/tcp_seq.h>
110 #include <netinet/tcp_timer.h>
111 #include <netinet/tcp_var.h>
112 #include <netinet6/tcp6_var.h>
113 #include <netinet/tcpip.h>
114 #include <netinet/cc/cc.h>
115 #include <netinet/tcp_fastopen.h>
117 #include <netinet/tcp_pcap.h>
119 #include <netinet/tcp_syncache.h>
121 #include <netinet/tcp_debug.h>
122 #endif /* TCPDEBUG */
124 #include <netinet/tcp_offload.h>
126 #include <netinet/tcp_ecn.h>
127 #include <netinet/udp.h>
129 #include <netipsec/ipsec_support.h>
131 #include <machine/in_cksum.h>
133 #include <security/mac/mac_framework.h>
135 const int tcprexmtthresh = 3;
137 VNET_DEFINE(int, tcp_log_in_vain) = 0;
138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
139 &VNET_NAME(tcp_log_in_vain), 0,
140 "Log all incoming TCP segments to closed ports");
142 VNET_DEFINE(int, blackhole) = 0;
143 #define V_blackhole VNET(blackhole)
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
145 &VNET_NAME(blackhole), 0,
146 "Do not send RST on segments to closed ports");
148 VNET_DEFINE(bool, blackhole_local) = false;
149 #define V_blackhole_local VNET(blackhole_local)
150 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET |
151 CTLFLAG_RW, &VNET_NAME(blackhole_local), false,
152 "Enforce net.inet.tcp.blackhole for locally originated packets");
154 VNET_DEFINE(int, tcp_delack_enabled) = 1;
155 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
156 &VNET_NAME(tcp_delack_enabled), 0,
157 "Delay ACK to try and piggyback it onto a data packet");
159 VNET_DEFINE(int, drop_synfin) = 0;
160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
161 &VNET_NAME(drop_synfin), 0,
162 "Drop TCP packets with SYN+FIN set");
164 VNET_DEFINE(int, tcp_do_prr_conservative) = 0;
165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW,
166 &VNET_NAME(tcp_do_prr_conservative), 0,
167 "Do conservative Proportional Rate Reduction");
169 VNET_DEFINE(int, tcp_do_prr) = 1;
170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW,
171 &VNET_NAME(tcp_do_prr), 1,
172 "Enable Proportional Rate Reduction per RFC 6937");
174 VNET_DEFINE(int, tcp_do_lrd) = 0;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW,
176 &VNET_NAME(tcp_do_lrd), 1,
177 "Perform Lost Retransmission Detection");
179 VNET_DEFINE(int, tcp_do_newcwv) = 0;
180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
181 &VNET_NAME(tcp_do_newcwv), 0,
182 "Enable New Congestion Window Validation per RFC7661");
184 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
185 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
186 &VNET_NAME(tcp_do_rfc3042), 0,
187 "Enable RFC 3042 (Limited Transmit)");
189 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
190 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
191 &VNET_NAME(tcp_do_rfc3390), 0,
192 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
194 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
195 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
196 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
197 "Slow-start flight size (initial congestion window) in number of segments");
199 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
200 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
201 &VNET_NAME(tcp_do_rfc3465), 0,
202 "Enable RFC 3465 (Appropriate Byte Counting)");
204 VNET_DEFINE(int, tcp_abc_l_var) = 2;
205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
206 &VNET_NAME(tcp_abc_l_var), 2,
207 "Cap the max cwnd increment during slow-start to this number of segments");
209 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn,
210 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
213 VNET_DEFINE(int, tcp_do_ecn) = 2;
214 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
215 &VNET_NAME(tcp_do_ecn), 0,
218 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
219 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
220 &VNET_NAME(tcp_ecn_maxretries), 0,
221 "Max retries before giving up on ECN");
223 VNET_DEFINE(int, tcp_insecure_syn) = 0;
224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
225 &VNET_NAME(tcp_insecure_syn), 0,
226 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
228 VNET_DEFINE(int, tcp_insecure_rst) = 0;
229 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
230 &VNET_NAME(tcp_insecure_rst), 0,
231 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
233 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
234 #define V_tcp_recvspace VNET(tcp_recvspace)
235 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
236 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
238 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
239 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
240 &VNET_NAME(tcp_do_autorcvbuf), 0,
241 "Enable automatic receive buffer sizing");
243 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
244 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
245 &VNET_NAME(tcp_autorcvbuf_max), 0,
246 "Max size of automatic receive buffer");
248 VNET_DEFINE(struct inpcbinfo, tcbinfo);
251 * TCP statistics are stored in an array of counter(9)s, which size matches
252 * size of struct tcpstat. TCP running connection count is a regular array.
254 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
255 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
256 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
257 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
258 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
259 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
260 "TCP connection counts by TCP state");
263 * Kernel module interface for updating tcpstat. The first argument is an index
264 * into tcpstat treated as an array.
267 kmod_tcpstat_add(int statnum, int val)
270 counter_u64_add(VNET(tcpstat)[statnum], val);
275 * Wrapper for the TCP established input helper hook.
278 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
280 struct tcp_hhook_data hhook_data;
282 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
287 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
294 * CC wrapper hook functions
297 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
304 INP_WLOCK_ASSERT(tp->t_inpcb);
306 tp->ccv->nsegs = nsegs;
307 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
308 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
309 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
310 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
311 tp->ccv->flags |= CCF_CWND_LIMITED;
313 tp->ccv->flags &= ~CCF_CWND_LIMITED;
315 if (type == CC_ACK) {
317 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
318 ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
319 if (!IN_RECOVERY(tp->t_flags))
320 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
321 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
322 if ((tp->t_flags & TF_GPUTINPROG) &&
323 SEQ_GEQ(th->th_ack, tp->gput_ack)) {
325 * Compute goodput in bits per millisecond.
327 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) /
328 max(1, tcp_ts_getticks() - tp->gput_ts);
329 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
332 * XXXLAS: This is a temporary hack, and should be
333 * chained off VOI_TCP_GPUT when stats(9) grows an API
334 * to deal with chained VOIs.
336 if (tp->t_stats_gput_prev > 0)
337 stats_voi_update_abs_s32(tp->t_stats,
339 ((gput - tp->t_stats_gput_prev) * 100) /
340 tp->t_stats_gput_prev);
341 tp->t_flags &= ~TF_GPUTINPROG;
342 tp->t_stats_gput_prev = gput;
345 if (tp->snd_cwnd > tp->snd_ssthresh) {
346 tp->t_bytes_acked += tp->ccv->bytes_this_ack;
347 if (tp->t_bytes_acked >= tp->snd_cwnd) {
348 tp->t_bytes_acked -= tp->snd_cwnd;
349 tp->ccv->flags |= CCF_ABC_SENTAWND;
352 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
353 tp->t_bytes_acked = 0;
357 if (CC_ALGO(tp)->ack_received != NULL) {
358 /* XXXLAS: Find a way to live without this */
359 tp->ccv->curack = th->th_ack;
360 CC_ALGO(tp)->ack_received(tp->ccv, type);
363 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
368 cc_conn_init(struct tcpcb *tp)
370 struct hc_metrics_lite metrics;
371 struct inpcb *inp = tp->t_inpcb;
375 INP_WLOCK_ASSERT(tp->t_inpcb);
377 tcp_hc_get(&inp->inp_inc, &metrics);
378 maxseg = tcp_maxseg(tp);
380 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
382 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
383 TCPSTAT_INC(tcps_usedrtt);
384 if (metrics.rmx_rttvar) {
385 tp->t_rttvar = metrics.rmx_rttvar;
386 TCPSTAT_INC(tcps_usedrttvar);
388 /* default variation is +- 1 rtt */
390 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
392 TCPT_RANGESET(tp->t_rxtcur,
393 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
394 tp->t_rttmin, TCPTV_REXMTMAX);
396 if (metrics.rmx_ssthresh) {
398 * There's some sort of gateway or interface
399 * buffer limit on the path. Use this to set
400 * the slow start threshold, but set the
401 * threshold to no less than 2*mss.
403 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
404 TCPSTAT_INC(tcps_usedssthresh);
408 * Set the initial slow-start flight size.
410 * If a SYN or SYN/ACK was lost and retransmitted, we have to
411 * reduce the initial CWND to one segment as congestion is likely
412 * requiring us to be cautious.
414 if (tp->snd_cwnd == 1)
415 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
417 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
419 if (CC_ALGO(tp)->conn_init != NULL)
420 CC_ALGO(tp)->conn_init(tp->ccv);
424 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
426 INP_WLOCK_ASSERT(tp->t_inpcb);
429 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
434 if (!IN_FASTRECOVERY(tp->t_flags)) {
435 tp->snd_recover = tp->snd_max;
436 if (tp->t_flags2 & TF2_ECN_PERMIT)
437 tp->t_flags2 |= TF2_ECN_SND_CWR;
441 if (!IN_CONGRECOVERY(tp->t_flags) ||
443 * Allow ECN reaction on ACK to CWR, if
444 * that data segment was also CE marked.
446 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
447 EXIT_CONGRECOVERY(tp->t_flags);
448 TCPSTAT_INC(tcps_ecn_rcwnd);
449 tp->snd_recover = tp->snd_max + 1;
450 if (tp->t_flags2 & TF2_ECN_PERMIT)
451 tp->t_flags2 |= TF2_ECN_SND_CWR;
456 tp->t_bytes_acked = 0;
457 EXIT_RECOVERY(tp->t_flags);
458 if (tp->t_flags2 & TF2_ECN_PERMIT)
459 tp->t_flags2 |= TF2_ECN_SND_CWR;
462 TCPSTAT_INC(tcps_sndrexmitbad);
463 /* RTO was unnecessary, so reset everything. */
464 tp->snd_cwnd = tp->snd_cwnd_prev;
465 tp->snd_ssthresh = tp->snd_ssthresh_prev;
466 tp->snd_recover = tp->snd_recover_prev;
467 if (tp->t_flags & TF_WASFRECOVERY)
468 ENTER_FASTRECOVERY(tp->t_flags);
469 if (tp->t_flags & TF_WASCRECOVERY)
470 ENTER_CONGRECOVERY(tp->t_flags);
471 tp->snd_nxt = tp->snd_max;
472 tp->t_flags &= ~TF_PREVVALID;
477 if (CC_ALGO(tp)->cong_signal != NULL) {
479 tp->ccv->curack = th->th_ack;
480 CC_ALGO(tp)->cong_signal(tp->ccv, type);
485 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
487 INP_WLOCK_ASSERT(tp->t_inpcb);
489 /* XXXLAS: KASSERT that we're in recovery? */
491 if (CC_ALGO(tp)->post_recovery != NULL) {
492 tp->ccv->curack = th->th_ack;
493 CC_ALGO(tp)->post_recovery(tp->ccv);
495 /* XXXLAS: EXIT_RECOVERY ? */
496 tp->t_bytes_acked = 0;
497 tp->sackhint.delivered_data = 0;
498 tp->sackhint.prr_out = 0;
502 * Indicate whether this ack should be delayed. We can delay the ack if
503 * following conditions are met:
504 * - There is no delayed ack timer in progress.
505 * - Our last ack wasn't a 0-sized window. We never want to delay
506 * the ack that opens up a 0-sized window.
507 * - LRO wasn't used for this segment. We make sure by checking that the
508 * segment size is not larger than the MSS.
510 #define DELAY_ACK(tp, tlen) \
511 ((!tcp_timer_active(tp, TT_DELACK) && \
512 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
513 (tlen <= tp->t_maxseg) && \
514 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
517 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos)
519 INP_WLOCK_ASSERT(tp->t_inpcb);
521 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
522 switch (iptos & IPTOS_ECN_MASK) {
524 tp->ccv->flags |= CCF_IPHDR_CE;
530 case IPTOS_ECN_NOTECT:
531 tp->ccv->flags &= ~CCF_IPHDR_CE;
536 tp->ccv->flags |= CCF_TCPHDR_CWR;
538 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
540 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
542 if (tp->ccv->flags & CCF_ACKNOW) {
543 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
544 tp->t_flags |= TF_ACKNOW;
550 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
552 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos);
556 * TCP input handling is split into multiple parts:
557 * tcp6_input is a thin wrapper around tcp_input for the extended
558 * ip6_protox[] call format in ip6_input
559 * tcp_input handles primary segment validation, inpcb lookup and
560 * SYN processing on listen sockets
561 * tcp_do_segment processes the ACK and text of the segment for
562 * establishing, established and closing connections
566 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
569 struct in6_ifaddr *ia6;
573 if (m->m_len < *offp + sizeof(struct tcphdr)) {
574 m = m_pullup(m, *offp + sizeof(struct tcphdr));
577 TCPSTAT_INC(tcps_rcvshort);
578 return (IPPROTO_DONE);
583 * draft-itojun-ipv6-tcp-to-anycast
584 * better place to put this in?
586 ip6 = mtod(m, struct ip6_hdr *);
587 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
588 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
589 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
590 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
592 return (IPPROTO_DONE);
596 return (tcp_input_with_port(mp, offp, proto, port));
600 tcp6_input(struct mbuf **mp, int *offp, int proto)
603 return(tcp6_input_with_port(mp, offp, proto, 0));
608 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
610 struct mbuf *m = *mp;
611 struct tcphdr *th = NULL;
612 struct ip *ip = NULL;
613 struct inpcb *inp = NULL;
614 struct tcpcb *tp = NULL;
615 struct socket *so = NULL;
626 int rstreason = 0; /* For badport_bandlim accounting purposes */
629 struct m_tag *fwd_tag = NULL;
631 struct ip6_hdr *ip6 = NULL;
634 const void *ip6 = NULL;
636 struct tcpopt to; /* options in this segment */
637 char *s = NULL; /* address and port logging */
640 * The size of tcp_saveipgen must be the size of the max ip header,
643 u_char tcp_saveipgen[IP6_HDR_LEN];
644 struct tcphdr tcp_savetcp;
651 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
658 TCPSTAT_INC(tcps_rcvtotal);
662 ip6 = mtod(m, struct ip6_hdr *);
663 th = (struct tcphdr *)((caddr_t)ip6 + off0);
664 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
667 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
668 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
669 th->th_sum = m->m_pkthdr.csum_data;
671 th->th_sum = in6_cksum_pseudo(ip6, tlen,
672 IPPROTO_TCP, m->m_pkthdr.csum_data);
673 th->th_sum ^= 0xffff;
675 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
677 TCPSTAT_INC(tcps_rcvbadsum);
682 * Be proactive about unspecified IPv6 address in source.
683 * As we use all-zero to indicate unbounded/unconnected pcb,
684 * unspecified IPv6 address can be used to confuse us.
686 * Note that packets with unspecified IPv6 destination is
687 * already dropped in ip6_input.
689 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
693 iptos = IPV6_TRAFFIC_CLASS(ip6);
696 #if defined(INET) && defined(INET6)
702 * Get IP and TCP header together in first mbuf.
703 * Note: IP leaves IP header in first mbuf.
705 if (off0 > sizeof (struct ip)) {
707 off0 = sizeof(struct ip);
709 if (m->m_len < sizeof (struct tcpiphdr)) {
710 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
712 TCPSTAT_INC(tcps_rcvshort);
713 return (IPPROTO_DONE);
716 ip = mtod(m, struct ip *);
717 th = (struct tcphdr *)((caddr_t)ip + off0);
718 tlen = ntohs(ip->ip_len) - off0;
723 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
724 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
725 th->th_sum = m->m_pkthdr.csum_data;
727 th->th_sum = in_pseudo(ip->ip_src.s_addr,
729 htonl(m->m_pkthdr.csum_data + tlen +
731 th->th_sum ^= 0xffff;
733 struct ipovly *ipov = (struct ipovly *)ip;
736 * Checksum extended TCP header and data.
740 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
741 ipov->ih_len = htons(tlen);
742 th->th_sum = in_cksum(m, len);
743 /* Reset length for SDT probes. */
744 ip->ip_len = htons(len);
747 /* Re-initialization for later version check */
749 ip->ip_v = IPVERSION;
750 ip->ip_hl = off0 >> 2;
753 if (th->th_sum && (port == 0)) {
754 TCPSTAT_INC(tcps_rcvbadsum);
761 * Check that TCP offset makes sense,
762 * pull out TCP options and adjust length. XXX
764 off = th->th_off << 2;
765 if (off < sizeof (struct tcphdr) || off > tlen) {
766 TCPSTAT_INC(tcps_rcvbadoff);
769 tlen -= off; /* tlen is used instead of ti->ti_len */
770 if (off > sizeof (struct tcphdr)) {
773 if (m->m_len < off0 + off) {
774 m = m_pullup(m, off0 + off);
776 TCPSTAT_INC(tcps_rcvshort);
777 return (IPPROTO_DONE);
780 ip6 = mtod(m, struct ip6_hdr *);
781 th = (struct tcphdr *)((caddr_t)ip6 + off0);
784 #if defined(INET) && defined(INET6)
789 if (m->m_len < sizeof(struct ip) + off) {
790 if ((m = m_pullup(m, sizeof (struct ip) + off))
792 TCPSTAT_INC(tcps_rcvshort);
793 return (IPPROTO_DONE);
795 ip = mtod(m, struct ip *);
796 th = (struct tcphdr *)((caddr_t)ip + off0);
800 optlen = off - sizeof (struct tcphdr);
801 optp = (u_char *)(th + 1);
803 thflags = tcp_get_flags(th);
806 * Convert TCP protocol specific fields to host format.
808 tcp_fields_to_host(th);
811 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
813 drop_hdrlen = off0 + off;
816 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
820 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
822 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
825 #if defined(INET) && !defined(INET6)
826 (m->m_flags & M_IP_NEXTHOP)
829 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
832 * For initial SYN packets we don't need write lock on matching
833 * PCB, be it a listening one or a synchronized one. The packet
834 * shall not modify its state.
836 lookupflag = (thflags & (TH_ACK|TH_SYN)) == TH_SYN ?
837 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB;
840 if (isipv6 && fwd_tag != NULL) {
841 struct sockaddr_in6 *next_hop6;
843 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
845 * Transparently forwarded. Pretend to be the destination.
846 * Already got one like this?
848 inp = in6_pcblookup_mbuf(&V_tcbinfo,
849 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
850 lookupflag, m->m_pkthdr.rcvif, m);
853 * It's new. Try to find the ambushing socket.
854 * Because we've rewritten the destination address,
855 * any hardware-generated hash is ignored.
857 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
858 th->th_sport, &next_hop6->sin6_addr,
859 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
860 th->th_dport, INPLOOKUP_WILDCARD | lookupflag,
864 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
865 th->th_sport, &ip6->ip6_dst, th->th_dport,
866 INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m);
869 #if defined(INET6) && defined(INET)
873 if (fwd_tag != NULL) {
874 struct sockaddr_in *next_hop;
876 next_hop = (struct sockaddr_in *)(fwd_tag+1);
878 * Transparently forwarded. Pretend to be the destination.
879 * already got one like this?
881 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
882 ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m);
885 * It's new. Try to find the ambushing socket.
886 * Because we've rewritten the destination address,
887 * any hardware-generated hash is ignored.
889 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
890 th->th_sport, next_hop->sin_addr,
891 next_hop->sin_port ? ntohs(next_hop->sin_port) :
892 th->th_dport, INPLOOKUP_WILDCARD | lookupflag,
896 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
897 th->th_sport, ip->ip_dst, th->th_dport,
898 INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m);
902 * If the INPCB does not exist then all data in the incoming
903 * segment is discarded and an appropriate RST is sent back.
904 * XXX MRT Send RST using which routing table?
908 * Log communication attempts to ports that are not
911 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
912 V_tcp_log_in_vain == 2) {
913 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
914 log(LOG_INFO, "%s; %s: Connection attempt "
915 "to closed port\n", s, __func__);
918 * When blackholing do not respond with a RST but
919 * completely ignore the segment and drop it.
921 if (((V_blackhole == 1 && (thflags & TH_SYN)) ||
922 V_blackhole == 2) && (V_blackhole_local || (
924 isipv6 ? !in6_localaddr(&ip6->ip6_src) :
927 !in_localip(ip->ip_src)
934 rstreason = BANDLIM_RST_CLOSEDPORT;
937 INP_LOCK_ASSERT(inp);
939 * While waiting for inp lock during the lookup, another thread
940 * can have dropped the inpcb, in which case we need to loop back
941 * and try to find a new inpcb to deliver to.
943 if (inp->inp_flags & INP_DROPPED) {
948 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
949 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
950 ((inp->inp_socket == NULL) || !SOLISTENING(inp->inp_socket))) {
951 inp->inp_flowid = m->m_pkthdr.flowid;
952 inp->inp_flowtype = M_HASHTYPE_GET(m);
954 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
956 if (isipv6 && IPSEC_ENABLED(ipv6) &&
957 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
965 if (IPSEC_ENABLED(ipv4) &&
966 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
973 * Check the minimum TTL for socket.
975 if (inp->inp_ip_minttl != 0) {
978 if (inp->inp_ip_minttl > ip6->ip6_hlim)
982 if (inp->inp_ip_minttl > ip->ip_ttl)
987 * A previous connection in TIMEWAIT state is supposed to catch stray
988 * or duplicate segments arriving late. If this segment was a
989 * legitimate new connection attempt, the old INPCB gets removed and
990 * we can try again to find a listening socket.
992 if (inp->inp_flags & INP_TIMEWAIT) {
993 tcp_dooptions(&to, optp, optlen,
994 (thflags & TH_SYN) ? TO_SYN : 0);
996 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
998 if (tcp_twcheck(inp, &to, th, m, tlen))
1000 return (IPPROTO_DONE);
1003 * The TCPCB may no longer exist if the connection is winding
1004 * down or it is in the CLOSED state. Either way we drop the
1005 * segment and send an appropriate response.
1007 tp = intotcpcb(inp);
1008 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1009 rstreason = BANDLIM_RST_CLOSEDPORT;
1013 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) {
1014 rstreason = BANDLIM_RST_CLOSEDPORT;
1019 if (tp->t_flags & TF_TOE) {
1020 tcp_offload_input(tp, m);
1021 m = NULL; /* consumed by the TOE driver */
1027 if (mac_inpcb_check_deliver(inp, m))
1030 so = inp->inp_socket;
1031 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1033 if (so->so_options & SO_DEBUG) {
1034 ostate = tp->t_state;
1037 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1040 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1043 #endif /* TCPDEBUG */
1045 * When the socket is accepting connections (the INPCB is in LISTEN
1046 * state) we look into the SYN cache if this is a new connection
1047 * attempt or the completion of a previous one.
1049 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so),
1050 ("%s: so accepting but tp %p not listening", __func__, tp));
1051 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) {
1052 struct in_conninfo inc;
1054 bzero(&inc, sizeof(inc));
1057 inc.inc_flags |= INC_ISIPV6;
1058 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1059 inc.inc_flags |= INC_IPV6MINMTU;
1060 inc.inc6_faddr = ip6->ip6_src;
1061 inc.inc6_laddr = ip6->ip6_dst;
1065 inc.inc_faddr = ip->ip_src;
1066 inc.inc_laddr = ip->ip_dst;
1068 inc.inc_fport = th->th_sport;
1069 inc.inc_lport = th->th_dport;
1070 inc.inc_fibnum = so->so_fibnum;
1073 * Check for an existing connection attempt in syncache if
1074 * the flag is only ACK. A successful lookup creates a new
1075 * socket appended to the listen queue in SYN_RECEIVED state.
1077 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1079 * Parse the TCP options here because
1080 * syncookies need access to the reflected
1083 tcp_dooptions(&to, optp, optlen, 0);
1085 * NB: syncache_expand() doesn't unlock inp.
1087 rstreason = syncache_expand(&inc, &to, th, &so, m, port);
1088 if (rstreason < 0) {
1090 * A failing TCP MD5 signature comparison
1091 * must result in the segment being dropped
1092 * and must not produce any response back
1096 } else if (rstreason == 0) {
1098 * No syncache entry or ACK was not
1099 * for our SYN/ACK. Send a RST.
1100 * NB: syncache did its own logging
1101 * of the failure cause.
1103 rstreason = BANDLIM_RST_OPENPORT;
1109 * We completed the 3-way handshake
1110 * but could not allocate a socket
1111 * either due to memory shortage,
1112 * listen queue length limits or
1113 * global socket limits. Send RST
1114 * or wait and have the remote end
1115 * retransmit the ACK for another
1118 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1119 log(LOG_DEBUG, "%s; %s: Listen socket: "
1120 "Socket allocation failed due to "
1121 "limits or memory shortage, %s\n",
1123 V_tcp_sc_rst_sock_fail ?
1124 "sending RST" : "try again");
1125 if (V_tcp_sc_rst_sock_fail) {
1126 rstreason = BANDLIM_UNLIMITED;
1132 * Socket is created in state SYN_RECEIVED.
1133 * Unlock the listen socket, lock the newly
1134 * created socket and update the tp variable.
1135 * If we came here via jump to tfo_socket_result,
1136 * then listening socket is read-locked.
1138 INP_UNLOCK(inp); /* listen socket */
1139 inp = sotoinpcb(so);
1141 * New connection inpcb is already locked by
1142 * syncache_expand().
1144 INP_WLOCK_ASSERT(inp);
1145 tp = intotcpcb(inp);
1146 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1147 ("%s: ", __func__));
1149 * Process the segment and the data it
1150 * contains. tcp_do_segment() consumes
1151 * the mbuf chain and unlocks the inpcb.
1153 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1154 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1156 return (IPPROTO_DONE);
1159 * Segment flag validation for new connection attempts:
1161 * Our (SYN|ACK) response was rejected.
1162 * Check with syncache and remove entry to prevent
1165 * NB: syncache_chkrst does its own logging of failure
1168 if (thflags & TH_RST) {
1169 syncache_chkrst(&inc, th, m, port);
1173 * We can't do anything without SYN.
1175 if ((thflags & TH_SYN) == 0) {
1176 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1177 log(LOG_DEBUG, "%s; %s: Listen socket: "
1178 "SYN is missing, segment ignored\n",
1180 TCPSTAT_INC(tcps_badsyn);
1184 * (SYN|ACK) is bogus on a listen socket.
1186 if (thflags & TH_ACK) {
1187 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1188 log(LOG_DEBUG, "%s; %s: Listen socket: "
1189 "SYN|ACK invalid, segment rejected\n",
1191 syncache_badack(&inc, port); /* XXX: Not needed! */
1192 TCPSTAT_INC(tcps_badsyn);
1193 rstreason = BANDLIM_RST_OPENPORT;
1197 * If the drop_synfin option is enabled, drop all
1198 * segments with both the SYN and FIN bits set.
1199 * This prevents e.g. nmap from identifying the
1201 * XXX: Poor reasoning. nmap has other methods
1202 * and is constantly refining its stack detection
1204 * XXX: This is a violation of the TCP specification
1205 * and was used by RFC1644.
1207 if ((thflags & TH_FIN) && V_drop_synfin) {
1208 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1209 log(LOG_DEBUG, "%s; %s: Listen socket: "
1210 "SYN|FIN segment ignored (based on "
1211 "sysctl setting)\n", s, __func__);
1212 TCPSTAT_INC(tcps_badsyn);
1216 * Segment's flags are (SYN) or (SYN|FIN).
1218 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1219 * as they do not affect the state of the TCP FSM.
1220 * The data pointed to by TH_URG and th_urp is ignored.
1222 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1223 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1224 KASSERT(thflags & (TH_SYN),
1225 ("%s: Listen socket: TH_SYN not set", __func__));
1226 INP_RLOCK_ASSERT(inp);
1229 * If deprecated address is forbidden,
1230 * we do not accept SYN to deprecated interface
1231 * address to prevent any new inbound connection from
1232 * getting established.
1233 * When we do not accept SYN, we send a TCP RST,
1234 * with deprecated source address (instead of dropping
1235 * it). We compromise it as it is much better for peer
1236 * to send a RST, and RST will be the final packet
1239 * If we do not forbid deprecated addresses, we accept
1240 * the SYN packet. RFC2462 does not suggest dropping
1242 * If we decipher RFC2462 5.5.4, it says like this:
1243 * 1. use of deprecated addr with existing
1244 * communication is okay - "SHOULD continue to be
1246 * 2. use of it with new communication:
1247 * (2a) "SHOULD NOT be used if alternate address
1248 * with sufficient scope is available"
1249 * (2b) nothing mentioned otherwise.
1250 * Here we fall into (2b) case as we have no choice in
1251 * our source address selection - we must obey the peer.
1253 * The wording in RFC2462 is confusing, and there are
1254 * multiple description text for deprecated address
1255 * handling - worse, they are not exactly the same.
1256 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1258 if (isipv6 && !V_ip6_use_deprecated) {
1259 struct in6_ifaddr *ia6;
1261 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1263 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1264 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1265 log(LOG_DEBUG, "%s; %s: Listen socket: "
1266 "Connection attempt to deprecated "
1267 "IPv6 address rejected\n",
1269 rstreason = BANDLIM_RST_OPENPORT;
1275 * Basic sanity checks on incoming SYN requests:
1276 * Don't respond if the destination is a link layer
1277 * broadcast according to RFC1122 4.2.3.10, p. 104.
1278 * If it is from this socket it must be forged.
1279 * Don't respond if the source or destination is a
1280 * global or subnet broad- or multicast address.
1281 * Note that it is quite possible to receive unicast
1282 * link-layer packets with a broadcast IP address. Use
1283 * in_broadcast() to find them.
1285 if (m->m_flags & (M_BCAST|M_MCAST)) {
1286 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1287 log(LOG_DEBUG, "%s; %s: Listen socket: "
1288 "Connection attempt from broad- or multicast "
1289 "link layer address ignored\n", s, __func__);
1294 if (th->th_dport == th->th_sport &&
1295 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1296 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1297 log(LOG_DEBUG, "%s; %s: Listen socket: "
1298 "Connection attempt to/from self "
1299 "ignored\n", s, __func__);
1302 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1303 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1304 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1305 log(LOG_DEBUG, "%s; %s: Listen socket: "
1306 "Connection attempt from/to multicast "
1307 "address ignored\n", s, __func__);
1312 #if defined(INET) && defined(INET6)
1317 if (th->th_dport == th->th_sport &&
1318 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1319 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1320 log(LOG_DEBUG, "%s; %s: Listen socket: "
1321 "Connection attempt from/to self "
1322 "ignored\n", s, __func__);
1325 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1326 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1327 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1328 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1329 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1330 log(LOG_DEBUG, "%s; %s: Listen socket: "
1331 "Connection attempt from/to broad- "
1332 "or multicast address ignored\n",
1339 * SYN appears to be valid. Create compressed TCP state
1343 if (so->so_options & SO_DEBUG)
1344 tcp_trace(TA_INPUT, ostate, tp,
1345 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1347 TCP_PROBE3(debug__input, tp, th, m);
1348 tcp_dooptions(&to, optp, optlen, TO_SYN);
1349 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL,
1350 iptos, port)) != NULL)
1351 goto tfo_socket_result;
1354 * Entry added to syncache and mbuf consumed.
1355 * Only the listen socket is unlocked by syncache_add().
1357 return (IPPROTO_DONE);
1358 } else if (tp->t_state == TCPS_LISTEN) {
1360 * When a listen socket is torn down the SO_ACCEPTCONN
1361 * flag is removed first while connections are drained
1362 * from the accept queue in a unlock/lock cycle of the
1363 * ACCEPT_LOCK, opening a race condition allowing a SYN
1364 * attempt go through unhandled.
1368 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1369 if (tp->t_flags & TF_SIGNATURE) {
1370 tcp_dooptions(&to, optp, optlen, thflags);
1371 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1372 TCPSTAT_INC(tcps_sig_err_nosigopt);
1375 if (!TCPMD5_ENABLED() ||
1376 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1380 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1383 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1384 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1385 * the inpcb, and unlocks pcbinfo.
1387 * XXXGL: in case of a pure SYN arriving on existing connection
1388 * TCP stacks won't need to modify the PCB, they would either drop
1389 * the segment silently, or send a challenge ACK. However, we try
1390 * to upgrade the lock, because calling convention for stacks is
1391 * write-lock on PCB. If upgrade fails, drop the SYN.
1393 if (lookupflag == INPLOOKUP_RLOCKPCB && INP_TRY_UPGRADE(inp) == 0)
1396 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1397 return (IPPROTO_DONE);
1400 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1403 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1406 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1407 m = NULL; /* mbuf chain got consumed. */
1412 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1422 return (IPPROTO_DONE);
1426 * Automatic sizing of receive socket buffer. Often the send
1427 * buffer size is not optimally adjusted to the actual network
1428 * conditions at hand (delay bandwidth product). Setting the
1429 * buffer size too small limits throughput on links with high
1430 * bandwidth and high delay (eg. trans-continental/oceanic links).
1432 * On the receive side the socket buffer memory is only rarely
1433 * used to any significant extent. This allows us to be much
1434 * more aggressive in scaling the receive socket buffer. For
1435 * the case that the buffer space is actually used to a large
1436 * extent and we run out of kernel memory we can simply drop
1437 * the new segments; TCP on the sender will just retransmit it
1438 * later. Setting the buffer size too big may only consume too
1439 * much kernel memory if the application doesn't read() from
1440 * the socket or packet loss or reordering makes use of the
1443 * The criteria to step up the receive buffer one notch are:
1444 * 1. Application has not set receive buffer size with
1445 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1446 * 2. the number of bytes received during 1/2 of an sRTT
1447 * is at least 3/8 of the current socket buffer size.
1448 * 3. receive buffer size has not hit maximal automatic size;
1450 * If all of the criteria are met we increaset the socket buffer
1451 * by a 1/2 (bounded by the max). This allows us to keep ahead
1452 * of slow-start but also makes it so our peer never gets limited
1453 * by our rwnd which we then open up causing a burst.
1455 * This algorithm does two steps per RTT at most and only if
1456 * we receive a bulk stream w/o packet losses or reorderings.
1457 * Shrinking the buffer during idle times is not necessary as
1458 * it doesn't consume any memory when idle.
1460 * TODO: Only step up if the application is actually serving
1461 * the buffer to better manage the socket buffer resources.
1464 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1465 struct tcpcb *tp, int tlen)
1469 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1470 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1471 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1472 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1473 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1474 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1475 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1477 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1479 /* Start over with next RTT. */
1483 tp->rfbuf_cnt += tlen; /* add up */
1489 tcp_input(struct mbuf **mp, int *offp, int proto)
1491 return(tcp_input_with_port(mp, offp, proto, 0));
1495 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1498 * Since tp might be gone if the session entered
1499 * the TIME_WAIT state before coming here, we need
1500 * to check if the socket is still connected.
1508 INP_LOCK_ASSERT(tp->t_inpcb);
1509 if (tp->t_flags & TF_WAKESOR) {
1510 tp->t_flags &= ~TF_WAKESOR;
1511 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1512 sorwakeup_locked(so);
1517 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1518 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1521 int acked, ourfinisacked, needoutput = 0, sack_changed;
1522 int rstreason, todrop, win, incforsyn = 0;
1526 struct in_conninfo *inc;
1534 * The size of tcp_saveipgen must be the size of the max ip header,
1537 u_char tcp_saveipgen[IP6_HDR_LEN];
1538 struct tcphdr tcp_savetcp;
1541 thflags = tcp_get_flags(th);
1542 inc = &tp->t_inpcb->inp_inc;
1543 tp->sackhint.last_sack_ack = 0;
1545 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1548 INP_WLOCK_ASSERT(tp->t_inpcb);
1549 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1551 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1555 /* Save segment, if requested. */
1556 tcp_pcap_add(th, m, &(tp->t_inpkts));
1558 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1561 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1562 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1563 log(LOG_DEBUG, "%s; %s: "
1564 "SYN|FIN segment ignored (based on "
1565 "sysctl setting)\n", s, __func__);
1572 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1573 * check SEQ.ACK first.
1575 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1576 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1577 rstreason = BANDLIM_UNLIMITED;
1582 * Segment received on connection.
1583 * Reset idle time and keep-alive timer.
1584 * XXX: This should be done after segment
1585 * validation to ignore broken/spoofed segs.
1587 tp->t_rcvtime = ticks;
1590 * Scale up the window into a 32-bit value.
1591 * For the SYN_SENT state the scale is zero.
1593 tiwin = th->th_win << tp->snd_scale;
1595 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1599 * TCP ECN processing.
1601 if (tcp_ecn_input_segment(tp, thflags, iptos))
1602 cc_cong_signal(tp, th, CC_ECN);
1605 * Parse options on any incoming segment.
1607 tcp_dooptions(&to, (u_char *)(th + 1),
1608 (th->th_off << 2) - sizeof(struct tcphdr),
1609 (thflags & TH_SYN) ? TO_SYN : 0);
1611 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1612 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1613 (to.to_flags & TOF_SIGNATURE) == 0) {
1614 TCPSTAT_INC(tcps_sig_err_sigopt);
1615 /* XXX: should drop? */
1619 * If echoed timestamp is later than the current time,
1620 * fall back to non RFC1323 RTT calculation. Normalize
1621 * timestamp if syncookies were used when this connection
1624 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1625 to.to_tsecr -= tp->ts_offset;
1626 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1628 else if (tp->t_rxtshift == 1 &&
1629 tp->t_flags & TF_PREVVALID &&
1630 tp->t_badrxtwin != 0 &&
1631 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
1632 cc_cong_signal(tp, th, CC_RTO_ERR);
1635 * Process options only when we get SYN/ACK back. The SYN case
1636 * for incoming connections is handled in tcp_syncache.
1637 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1638 * or <SYN,ACK>) segment itself is never scaled.
1639 * XXX this is traditional behavior, may need to be cleaned up.
1641 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1642 /* Handle parallel SYN for ECN */
1643 tcp_ecn_input_parallel_syn(tp, thflags, iptos);
1644 if ((to.to_flags & TOF_SCALE) &&
1645 (tp->t_flags & TF_REQ_SCALE) &&
1646 !(tp->t_flags & TF_NOOPT)) {
1647 tp->t_flags |= TF_RCVD_SCALE;
1648 tp->snd_scale = to.to_wscale;
1650 tp->t_flags &= ~TF_REQ_SCALE;
1652 * Initial send window. It will be updated with
1653 * the next incoming segment to the scaled value.
1655 tp->snd_wnd = th->th_win;
1656 if ((to.to_flags & TOF_TS) &&
1657 (tp->t_flags & TF_REQ_TSTMP) &&
1658 !(tp->t_flags & TF_NOOPT)) {
1659 tp->t_flags |= TF_RCVD_TSTMP;
1660 tp->ts_recent = to.to_tsval;
1661 tp->ts_recent_age = tcp_ts_getticks();
1663 tp->t_flags &= ~TF_REQ_TSTMP;
1664 if (to.to_flags & TOF_MSS)
1665 tcp_mss(tp, to.to_mss);
1666 if ((tp->t_flags & TF_SACK_PERMIT) &&
1667 (!(to.to_flags & TOF_SACKPERM) ||
1668 (tp->t_flags & TF_NOOPT)))
1669 tp->t_flags &= ~TF_SACK_PERMIT;
1670 if (IS_FASTOPEN(tp->t_flags)) {
1671 if ((to.to_flags & TOF_FASTOPEN) &&
1672 !(tp->t_flags & TF_NOOPT)) {
1675 if (to.to_flags & TOF_MSS)
1678 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1682 tcp_fastopen_update_cache(tp, mss,
1683 to.to_tfo_len, to.to_tfo_cookie);
1685 tcp_fastopen_disable_path(tp);
1690 * If timestamps were negotiated during SYN/ACK and a
1691 * segment without a timestamp is received, silently drop
1692 * the segment, unless it is a RST segment or missing timestamps are
1694 * See section 3.2 of RFC 7323.
1696 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1697 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1698 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1699 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1700 "segment processed normally\n",
1705 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1706 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1707 "segment silently dropped\n", s, __func__);
1714 * If timestamps were not negotiated during SYN/ACK and a
1715 * segment with a timestamp is received, ignore the
1716 * timestamp and process the packet normally.
1717 * See section 3.2 of RFC 7323.
1719 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1720 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1721 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1722 "segment processed normally\n", s, __func__);
1728 * Header prediction: check for the two common cases
1729 * of a uni-directional data xfer. If the packet has
1730 * no control flags, is in-sequence, the window didn't
1731 * change and we're not retransmitting, it's a
1732 * candidate. If the length is zero and the ack moved
1733 * forward, we're the sender side of the xfer. Just
1734 * free the data acked & wake any higher level process
1735 * that was blocked waiting for space. If the length
1736 * is non-zero and the ack didn't move, we're the
1737 * receiver side. If we're getting packets in-order
1738 * (the reassembly queue is empty), add the data to
1739 * the socket buffer and note that we need a delayed ack.
1740 * Make sure that the hidden state-flags are also off.
1741 * Since we check for TCPS_ESTABLISHED first, it can only
1744 if (tp->t_state == TCPS_ESTABLISHED &&
1745 th->th_seq == tp->rcv_nxt &&
1746 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1747 tp->snd_nxt == tp->snd_max &&
1748 tiwin && tiwin == tp->snd_wnd &&
1749 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1751 ((to.to_flags & TOF_TS) == 0 ||
1752 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1754 * If last ACK falls within this segment's sequence numbers,
1755 * record the timestamp.
1756 * NOTE that the test is modified according to the latest
1757 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1759 if ((to.to_flags & TOF_TS) != 0 &&
1760 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1761 tp->ts_recent_age = tcp_ts_getticks();
1762 tp->ts_recent = to.to_tsval;
1766 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1767 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1768 !IN_RECOVERY(tp->t_flags) &&
1769 (to.to_flags & TOF_SACK) == 0 &&
1770 TAILQ_EMPTY(&tp->snd_holes)) {
1772 * This is a pure ack for outstanding data.
1774 TCPSTAT_INC(tcps_predack);
1777 * "bad retransmit" recovery without timestamps.
1779 if ((to.to_flags & TOF_TS) == 0 &&
1780 tp->t_rxtshift == 1 &&
1781 tp->t_flags & TF_PREVVALID &&
1782 tp->t_badrxtwin != 0 &&
1783 TSTMP_LT(ticks, tp->t_badrxtwin)) {
1784 cc_cong_signal(tp, th, CC_RTO_ERR);
1788 * Recalculate the transmit timer / rtt.
1790 * Some boxes send broken timestamp replies
1791 * during the SYN+ACK phase, ignore
1792 * timestamps of 0 or we could calculate a
1793 * huge RTT and blow up the retransmit timer.
1795 if ((to.to_flags & TOF_TS) != 0 &&
1799 t = tcp_ts_getticks() - to.to_tsecr;
1800 if (!tp->t_rttlow || tp->t_rttlow > t)
1803 TCP_TS_TO_TICKS(t) + 1);
1804 } else if (tp->t_rtttime &&
1805 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1806 if (!tp->t_rttlow ||
1807 tp->t_rttlow > ticks - tp->t_rtttime)
1808 tp->t_rttlow = ticks - tp->t_rtttime;
1810 ticks - tp->t_rtttime);
1812 acked = BYTES_THIS_ACK(tp, th);
1815 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1816 hhook_run_tcp_est_in(tp, th, &to);
1819 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1820 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1821 sbdrop(&so->so_snd, acked);
1822 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1823 SEQ_LEQ(th->th_ack, tp->snd_recover))
1824 tp->snd_recover = th->th_ack - 1;
1827 * Let the congestion control algorithm update
1828 * congestion control related information. This
1829 * typically means increasing the congestion
1832 cc_ack_received(tp, th, nsegs, CC_ACK);
1834 tp->snd_una = th->th_ack;
1836 * Pull snd_wl2 up to prevent seq wrap relative
1839 tp->snd_wl2 = th->th_ack;
1844 * If all outstanding data are acked, stop
1845 * retransmit timer, otherwise restart timer
1846 * using current (possibly backed-off) value.
1847 * If process is waiting for space,
1848 * wakeup/selwakeup/signal. If data
1849 * are ready to send, let tcp_output
1850 * decide between more output or persist.
1853 if (so->so_options & SO_DEBUG)
1854 tcp_trace(TA_INPUT, ostate, tp,
1855 (void *)tcp_saveipgen,
1858 TCP_PROBE3(debug__input, tp, th, m);
1859 if (tp->snd_una == tp->snd_max)
1860 tcp_timer_activate(tp, TT_REXMT, 0);
1861 else if (!tcp_timer_active(tp, TT_PERSIST))
1862 tcp_timer_activate(tp, TT_REXMT,
1865 if (sbavail(&so->so_snd))
1866 (void) tcp_output(tp);
1869 } else if (th->th_ack == tp->snd_una &&
1870 tlen <= sbspace(&so->so_rcv)) {
1871 int newsize = 0; /* automatic sockbuf scaling */
1874 * This is a pure, in-sequence data packet with
1875 * nothing on the reassembly queue and we have enough
1876 * buffer space to take it.
1878 /* Clean receiver SACK report if present */
1879 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1880 tcp_clean_sackreport(tp);
1881 TCPSTAT_INC(tcps_preddat);
1882 tp->rcv_nxt += tlen;
1884 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1885 (tp->t_fbyte_in == 0)) {
1886 tp->t_fbyte_in = ticks;
1887 if (tp->t_fbyte_in == 0)
1889 if (tp->t_fbyte_out && tp->t_fbyte_in)
1890 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1893 * Pull snd_wl1 up to prevent seq wrap relative to
1896 tp->snd_wl1 = th->th_seq;
1898 * Pull rcv_up up to prevent seq wrap relative to
1901 tp->rcv_up = tp->rcv_nxt;
1902 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1903 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1905 if (so->so_options & SO_DEBUG)
1906 tcp_trace(TA_INPUT, ostate, tp,
1907 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1909 TCP_PROBE3(debug__input, tp, th, m);
1911 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1913 /* Add data to socket buffer. */
1914 SOCKBUF_LOCK(&so->so_rcv);
1915 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1919 * Set new socket buffer size.
1920 * Give up when limit is reached.
1923 if (!sbreserve_locked(so, SO_RCV,
1925 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1926 m_adj(m, drop_hdrlen); /* delayed header drop */
1927 sbappendstream_locked(&so->so_rcv, m, 0);
1929 /* NB: sorwakeup_locked() does an implicit unlock. */
1930 sorwakeup_locked(so);
1931 if (DELAY_ACK(tp, tlen)) {
1932 tp->t_flags |= TF_DELACK;
1934 tp->t_flags |= TF_ACKNOW;
1942 * Calculate amount of space in receive window,
1943 * and then do TCP input processing.
1944 * Receive window is amount of space in rcv queue,
1945 * but not less than advertised window.
1947 win = sbspace(&so->so_rcv);
1950 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1952 switch (tp->t_state) {
1954 * If the state is SYN_RECEIVED:
1955 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1957 case TCPS_SYN_RECEIVED:
1958 if ((thflags & TH_ACK) &&
1959 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1960 SEQ_GT(th->th_ack, tp->snd_max))) {
1961 rstreason = BANDLIM_RST_OPENPORT;
1964 if (IS_FASTOPEN(tp->t_flags)) {
1966 * When a TFO connection is in SYN_RECEIVED, the
1967 * only valid packets are the initial SYN, a
1968 * retransmit/copy of the initial SYN (possibly with
1969 * a subset of the original data), a valid ACK, a
1972 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1973 rstreason = BANDLIM_RST_OPENPORT;
1975 } else if (thflags & TH_SYN) {
1976 /* non-initial SYN is ignored */
1977 if ((tcp_timer_active(tp, TT_DELACK) ||
1978 tcp_timer_active(tp, TT_REXMT)))
1980 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1987 * If the state is SYN_SENT:
1988 * if seg contains a RST with valid ACK (SEQ.ACK has already
1989 * been verified), then drop the connection.
1990 * if seg contains a RST without an ACK, drop the seg.
1991 * if seg does not contain SYN, then drop the seg.
1992 * Otherwise this is an acceptable SYN segment
1993 * initialize tp->rcv_nxt and tp->irs
1994 * if seg contains ack then advance tp->snd_una
1995 * if seg contains an ECE and ECN support is enabled, the stream
1997 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1998 * arrange for segment to be acked (eventually)
1999 * continue processing rest of data/controls, beginning with URG
2002 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2003 TCP_PROBE5(connect__refused, NULL, tp,
2005 tp = tcp_drop(tp, ECONNREFUSED);
2007 if (thflags & TH_RST)
2009 if (!(thflags & TH_SYN))
2012 tp->irs = th->th_seq;
2014 if (thflags & TH_ACK) {
2015 int tfo_partial_ack = 0;
2017 TCPSTAT_INC(tcps_connects);
2020 mac_socketpeer_set_from_mbuf(m, so);
2022 /* Do window scaling on this connection? */
2023 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2024 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2025 tp->rcv_scale = tp->request_r_scale;
2027 tp->rcv_adv += min(tp->rcv_wnd,
2028 TCP_MAXWIN << tp->rcv_scale);
2029 tp->snd_una++; /* SYN is acked */
2031 * If not all the data that was sent in the TFO SYN
2032 * has been acked, resend the remainder right away.
2034 if (IS_FASTOPEN(tp->t_flags) &&
2035 (tp->snd_una != tp->snd_max)) {
2036 tp->snd_nxt = th->th_ack;
2037 tfo_partial_ack = 1;
2040 * If there's data, delay ACK; if there's also a FIN
2041 * ACKNOW will be turned on later.
2043 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2044 tcp_timer_activate(tp, TT_DELACK,
2047 tp->t_flags |= TF_ACKNOW;
2049 tcp_ecn_input_syn_sent(tp, thflags, iptos);
2052 * Received <SYN,ACK> in SYN_SENT[*] state.
2054 * SYN_SENT --> ESTABLISHED
2055 * SYN_SENT* --> FIN_WAIT_1
2057 tp->t_starttime = ticks;
2058 if (tp->t_flags & TF_NEEDFIN) {
2059 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2060 tp->t_flags &= ~TF_NEEDFIN;
2063 tcp_state_change(tp, TCPS_ESTABLISHED);
2064 TCP_PROBE5(connect__established, NULL, tp,
2067 tcp_timer_activate(tp, TT_KEEP,
2072 * Received initial SYN in SYN-SENT[*] state =>
2073 * simultaneous open.
2074 * If it succeeds, connection is * half-synchronized.
2075 * Otherwise, do 3-way handshake:
2076 * SYN-SENT -> SYN-RECEIVED
2077 * SYN-SENT* -> SYN-RECEIVED*
2079 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2080 tcp_timer_activate(tp, TT_REXMT, 0);
2081 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2084 INP_WLOCK_ASSERT(tp->t_inpcb);
2087 * Advance th->th_seq to correspond to first data byte.
2088 * If data, trim to stay within window,
2089 * dropping FIN if necessary.
2092 if (tlen > tp->rcv_wnd) {
2093 todrop = tlen - tp->rcv_wnd;
2097 TCPSTAT_INC(tcps_rcvpackafterwin);
2098 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2100 tp->snd_wl1 = th->th_seq - 1;
2101 tp->rcv_up = th->th_seq;
2103 * Client side of transaction: already sent SYN and data.
2104 * If the remote host used T/TCP to validate the SYN,
2105 * our data will be ACK'd; if so, enter normal data segment
2106 * processing in the middle of step 5, ack processing.
2107 * Otherwise, goto step 6.
2109 if (thflags & TH_ACK)
2115 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2116 * do normal processing.
2118 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2122 break; /* continue normal processing */
2126 * States other than LISTEN or SYN_SENT.
2127 * First check the RST flag and sequence number since reset segments
2128 * are exempt from the timestamp and connection count tests. This
2129 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2130 * below which allowed reset segments in half the sequence space
2131 * to fall though and be processed (which gives forged reset
2132 * segments with a random sequence number a 50 percent chance of
2133 * killing a connection).
2134 * Then check timestamp, if present.
2135 * Then check the connection count, if present.
2136 * Then check that at least some bytes of segment are within
2137 * receive window. If segment begins before rcv_nxt,
2138 * drop leading data (and SYN); if nothing left, just ack.
2140 if (thflags & TH_RST) {
2142 * RFC5961 Section 3.2
2144 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2145 * - If RST is in window, we send challenge ACK.
2147 * Note: to take into account delayed ACKs, we should
2148 * test against last_ack_sent instead of rcv_nxt.
2149 * Note 2: we handle special case of closed window, not
2150 * covered by the RFC.
2152 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2153 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2154 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2155 KASSERT(tp->t_state != TCPS_SYN_SENT,
2156 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2159 if (V_tcp_insecure_rst ||
2160 tp->last_ack_sent == th->th_seq) {
2161 TCPSTAT_INC(tcps_drops);
2162 /* Drop the connection. */
2163 switch (tp->t_state) {
2164 case TCPS_SYN_RECEIVED:
2165 so->so_error = ECONNREFUSED;
2167 case TCPS_ESTABLISHED:
2168 case TCPS_FIN_WAIT_1:
2169 case TCPS_FIN_WAIT_2:
2170 case TCPS_CLOSE_WAIT:
2173 so->so_error = ECONNRESET;
2180 TCPSTAT_INC(tcps_badrst);
2181 /* Send challenge ACK. */
2182 tcp_respond(tp, mtod(m, void *), th, m,
2183 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2184 tp->last_ack_sent = tp->rcv_nxt;
2192 * RFC5961 Section 4.2
2193 * Send challenge ACK for any SYN in synchronized state.
2195 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2196 tp->t_state != TCPS_SYN_RECEIVED) {
2197 TCPSTAT_INC(tcps_badsyn);
2198 if (V_tcp_insecure_syn &&
2199 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2200 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2201 tp = tcp_drop(tp, ECONNRESET);
2202 rstreason = BANDLIM_UNLIMITED;
2204 /* Send challenge ACK. */
2205 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2206 tp->snd_nxt, TH_ACK);
2207 tp->last_ack_sent = tp->rcv_nxt;
2214 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2215 * and it's less than ts_recent, drop it.
2217 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2218 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2219 /* Check to see if ts_recent is over 24 days old. */
2220 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2222 * Invalidate ts_recent. If this segment updates
2223 * ts_recent, the age will be reset later and ts_recent
2224 * will get a valid value. If it does not, setting
2225 * ts_recent to zero will at least satisfy the
2226 * requirement that zero be placed in the timestamp
2227 * echo reply when ts_recent isn't valid. The
2228 * age isn't reset until we get a valid ts_recent
2229 * because we don't want out-of-order segments to be
2230 * dropped when ts_recent is old.
2234 TCPSTAT_INC(tcps_rcvduppack);
2235 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2236 TCPSTAT_INC(tcps_pawsdrop);
2244 * In the SYN-RECEIVED state, validate that the packet belongs to
2245 * this connection before trimming the data to fit the receive
2246 * window. Check the sequence number versus IRS since we know
2247 * the sequence numbers haven't wrapped. This is a partial fix
2248 * for the "LAND" DoS attack.
2250 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2251 rstreason = BANDLIM_RST_OPENPORT;
2255 todrop = tp->rcv_nxt - th->th_seq;
2257 if (thflags & TH_SYN) {
2267 * Following if statement from Stevens, vol. 2, p. 960.
2270 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2272 * Any valid FIN must be to the left of the window.
2273 * At this point the FIN must be a duplicate or out
2274 * of sequence; drop it.
2279 * Send an ACK to resynchronize and drop any data.
2280 * But keep on processing for RST or ACK.
2282 tp->t_flags |= TF_ACKNOW;
2284 TCPSTAT_INC(tcps_rcvduppack);
2285 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2287 TCPSTAT_INC(tcps_rcvpartduppack);
2288 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2291 * DSACK - add SACK block for dropped range
2293 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2294 tcp_update_sack_list(tp, th->th_seq,
2295 th->th_seq + todrop);
2297 * ACK now, as the next in-sequence segment
2298 * will clear the DSACK block again
2300 tp->t_flags |= TF_ACKNOW;
2302 drop_hdrlen += todrop; /* drop from the top afterwards */
2303 th->th_seq += todrop;
2305 if (th->th_urp > todrop)
2306 th->th_urp -= todrop;
2314 * If new data are received on a connection after the
2315 * user processes are gone, then RST the other end.
2317 if ((tp->t_flags & TF_CLOSED) && tlen) {
2318 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2319 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2320 "after socket was closed, "
2321 "sending RST and removing tcpcb\n",
2322 s, __func__, tcpstates[tp->t_state], tlen);
2326 TCPSTAT_INC(tcps_rcvafterclose);
2327 rstreason = BANDLIM_UNLIMITED;
2332 * If segment ends after window, drop trailing data
2333 * (and PUSH and FIN); if nothing left, just ACK.
2335 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2337 TCPSTAT_INC(tcps_rcvpackafterwin);
2338 if (todrop >= tlen) {
2339 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2341 * If window is closed can only take segments at
2342 * window edge, and have to drop data and PUSH from
2343 * incoming segments. Continue processing, but
2344 * remember to ack. Otherwise, drop segment
2347 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2348 tp->t_flags |= TF_ACKNOW;
2349 TCPSTAT_INC(tcps_rcvwinprobe);
2353 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2356 thflags &= ~(TH_PUSH|TH_FIN);
2360 * If last ACK falls within this segment's sequence numbers,
2361 * record its timestamp.
2363 * 1) That the test incorporates suggestions from the latest
2364 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2365 * 2) That updating only on newer timestamps interferes with
2366 * our earlier PAWS tests, so this check should be solely
2367 * predicated on the sequence space of this segment.
2368 * 3) That we modify the segment boundary check to be
2369 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2370 * instead of RFC1323's
2371 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2372 * This modified check allows us to overcome RFC1323's
2373 * limitations as described in Stevens TCP/IP Illustrated
2374 * Vol. 2 p.869. In such cases, we can still calculate the
2375 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2377 if ((to.to_flags & TOF_TS) != 0 &&
2378 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2379 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2380 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2381 tp->ts_recent_age = tcp_ts_getticks();
2382 tp->ts_recent = to.to_tsval;
2386 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2387 * flag is on (half-synchronized state), then queue data for
2388 * later processing; else drop segment and return.
2390 if ((thflags & TH_ACK) == 0) {
2391 if (tp->t_state == TCPS_SYN_RECEIVED ||
2392 (tp->t_flags & TF_NEEDSYN)) {
2393 if (tp->t_state == TCPS_SYN_RECEIVED &&
2394 IS_FASTOPEN(tp->t_flags)) {
2395 tp->snd_wnd = tiwin;
2399 } else if (tp->t_flags & TF_ACKNOW)
2408 switch (tp->t_state) {
2410 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2411 * ESTABLISHED state and continue processing.
2412 * The ACK was checked above.
2414 case TCPS_SYN_RECEIVED:
2416 TCPSTAT_INC(tcps_connects);
2418 /* Do window scaling? */
2419 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2420 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2421 tp->rcv_scale = tp->request_r_scale;
2423 tp->snd_wnd = tiwin;
2426 * SYN-RECEIVED -> ESTABLISHED
2427 * SYN-RECEIVED* -> FIN-WAIT-1
2429 tp->t_starttime = ticks;
2430 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2431 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2432 tp->t_tfo_pending = NULL;
2434 if (tp->t_flags & TF_NEEDFIN) {
2435 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2436 tp->t_flags &= ~TF_NEEDFIN;
2438 tcp_state_change(tp, TCPS_ESTABLISHED);
2439 TCP_PROBE5(accept__established, NULL, tp,
2442 * TFO connections call cc_conn_init() during SYN
2443 * processing. Calling it again here for such
2444 * connections is not harmless as it would undo the
2445 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2448 if (!IS_FASTOPEN(tp->t_flags))
2450 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2453 * Account for the ACK of our SYN prior to
2454 * regular ACK processing below, except for
2455 * simultaneous SYN, which is handled later.
2457 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2460 * If segment contains data or ACK, will call tcp_reass()
2461 * later; if not, do so now to pass queued data to user.
2463 if (tlen == 0 && (thflags & TH_FIN) == 0) {
2464 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2466 tcp_handle_wakeup(tp, so);
2468 tp->snd_wl1 = th->th_seq - 1;
2472 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2473 * ACKs. If the ack is in the range
2474 * tp->snd_una < th->th_ack <= tp->snd_max
2475 * then advance tp->snd_una to th->th_ack and drop
2476 * data from the retransmission queue. If this ACK reflects
2477 * more up to date window information we update our window information.
2479 case TCPS_ESTABLISHED:
2480 case TCPS_FIN_WAIT_1:
2481 case TCPS_FIN_WAIT_2:
2482 case TCPS_CLOSE_WAIT:
2485 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2486 TCPSTAT_INC(tcps_rcvacktoomuch);
2489 if ((tp->t_flags & TF_SACK_PERMIT) &&
2490 ((to.to_flags & TOF_SACK) ||
2491 !TAILQ_EMPTY(&tp->snd_holes))) {
2492 if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) &&
2493 (tp->t_flags & TF_LRD)) {
2494 tcp_sack_lost_retransmission(tp, th);
2498 * Reset the value so that previous (valid) value
2499 * from the last ack with SACK doesn't get used.
2501 tp->sackhint.sacked_bytes = 0;
2504 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2505 hhook_run_tcp_est_in(tp, th, &to);
2508 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2509 maxseg = tcp_maxseg(tp);
2511 (tiwin == tp->snd_wnd ||
2512 (tp->t_flags & TF_SACK_PERMIT))) {
2514 * If this is the first time we've seen a
2515 * FIN from the remote, this is not a
2516 * duplicate and it needs to be processed
2517 * normally. This happens during a
2518 * simultaneous close.
2520 if ((thflags & TH_FIN) &&
2521 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2525 TCPSTAT_INC(tcps_rcvdupack);
2527 * If we have outstanding data (other than
2528 * a window probe), this is a completely
2529 * duplicate ack (ie, window info didn't
2530 * change and FIN isn't set),
2531 * the ack is the biggest we've
2532 * seen and we've seen exactly our rexmt
2533 * threshold of them, assume a packet
2534 * has been dropped and retransmit it.
2535 * Kludge snd_nxt & the congestion
2536 * window so we send only this one
2539 * We know we're losing at the current
2540 * window size so do congestion avoidance
2541 * (set ssthresh to half the current window
2542 * and pull our congestion window back to
2543 * the new ssthresh).
2545 * Dup acks mean that packets have left the
2546 * network (they're now cached at the receiver)
2547 * so bump cwnd by the amount in the receiver
2548 * to keep a constant cwnd packets in the
2551 * When using TCP ECN, notify the peer that
2552 * we reduced the cwnd.
2555 * Following 2 kinds of acks should not affect
2558 * 2) Acks with SACK but without any new SACK
2559 * information in them. These could result from
2560 * any anomaly in the network like a switch
2561 * duplicating packets or a possible DoS attack.
2563 if (th->th_ack != tp->snd_una ||
2564 ((tp->t_flags & TF_SACK_PERMIT) &&
2565 (to.to_flags & TOF_SACK) &&
2568 else if (!tcp_timer_active(tp, TT_REXMT))
2570 else if (++tp->t_dupacks > tcprexmtthresh ||
2571 IN_FASTRECOVERY(tp->t_flags)) {
2572 cc_ack_received(tp, th, nsegs,
2575 IN_FASTRECOVERY(tp->t_flags)) {
2576 tcp_do_prr_ack(tp, th, &to);
2577 } else if ((tp->t_flags & TF_SACK_PERMIT) &&
2578 (to.to_flags & TOF_SACK) &&
2579 IN_FASTRECOVERY(tp->t_flags)) {
2583 * Compute the amount of data in flight first.
2584 * We can inject new data into the pipe iff
2585 * we have less than 1/2 the original window's
2586 * worth of data in flight.
2588 if (V_tcp_do_newsack)
2589 awnd = tcp_compute_pipe(tp);
2591 awnd = (tp->snd_nxt - tp->snd_fack) +
2592 tp->sackhint.sack_bytes_rexmit;
2594 if (awnd < tp->snd_ssthresh) {
2595 tp->snd_cwnd += maxseg;
2596 if (tp->snd_cwnd > tp->snd_ssthresh)
2597 tp->snd_cwnd = tp->snd_ssthresh;
2600 tp->snd_cwnd += maxseg;
2601 (void) tcp_output(tp);
2603 } else if (tp->t_dupacks == tcprexmtthresh ||
2604 (tp->t_flags & TF_SACK_PERMIT &&
2606 tp->sackhint.sacked_bytes >
2607 (tcprexmtthresh - 1) * maxseg)) {
2610 * Above is the RFC6675 trigger condition of
2611 * more than (dupthresh-1)*maxseg sacked data.
2612 * If the count of holes in the
2613 * scoreboard is >= dupthresh, we could
2614 * also enter loss recovery, but don't
2615 * have that value readily available.
2617 tp->t_dupacks = tcprexmtthresh;
2618 tcp_seq onxt = tp->snd_nxt;
2621 * If we're doing sack, or prr, check
2622 * to see if we're already in sack
2623 * recovery. If we're not doing sack,
2624 * check to see if we're in newreno
2628 (tp->t_flags & TF_SACK_PERMIT)) {
2629 if (IN_FASTRECOVERY(tp->t_flags)) {
2634 if (SEQ_LEQ(th->th_ack,
2640 /* Congestion signal before ack. */
2641 cc_cong_signal(tp, th, CC_NDUPACK);
2642 cc_ack_received(tp, th, nsegs,
2644 tcp_timer_activate(tp, TT_REXMT, 0);
2648 * snd_ssthresh is already updated by
2651 if ((tp->t_flags & TF_SACK_PERMIT) &&
2652 (to.to_flags & TOF_SACK)) {
2653 tp->sackhint.prr_delivered =
2654 tp->sackhint.sacked_bytes;
2656 tp->sackhint.prr_delivered =
2657 imin(tp->snd_max - tp->snd_una,
2658 imin(INT_MAX / 65536,
2659 tp->t_dupacks) * maxseg);
2661 tp->sackhint.recover_fs = max(1,
2662 tp->snd_nxt - tp->snd_una);
2664 if ((tp->t_flags & TF_SACK_PERMIT) &&
2665 (to.to_flags & TOF_SACK)) {
2667 tcps_sack_recovery_episode);
2668 tp->snd_recover = tp->snd_nxt;
2669 tp->snd_cwnd = maxseg;
2670 (void) tcp_output(tp);
2671 if (SEQ_GT(th->th_ack, tp->snd_una))
2672 goto resume_partialack;
2675 tp->snd_nxt = th->th_ack;
2676 tp->snd_cwnd = maxseg;
2677 (void) tcp_output(tp);
2678 KASSERT(tp->snd_limited <= 2,
2679 ("%s: tp->snd_limited too big",
2681 tp->snd_cwnd = tp->snd_ssthresh +
2683 (tp->t_dupacks - tp->snd_limited);
2684 if (SEQ_GT(onxt, tp->snd_nxt))
2687 } else if (V_tcp_do_rfc3042) {
2689 * Process first and second duplicate
2690 * ACKs. Each indicates a segment
2691 * leaving the network, creating room
2692 * for more. Make sure we can send a
2693 * packet on reception of each duplicate
2694 * ACK by increasing snd_cwnd by one
2695 * segment. Restore the original
2696 * snd_cwnd after packet transmission.
2698 cc_ack_received(tp, th, nsegs,
2700 uint32_t oldcwnd = tp->snd_cwnd;
2701 tcp_seq oldsndmax = tp->snd_max;
2705 KASSERT(tp->t_dupacks == 1 ||
2707 ("%s: dupacks not 1 or 2",
2709 if (tp->t_dupacks == 1)
2710 tp->snd_limited = 0;
2712 (tp->snd_nxt - tp->snd_una) +
2713 (tp->t_dupacks - tp->snd_limited) *
2716 * Only call tcp_output when there
2717 * is new data available to be sent.
2718 * Otherwise we would send pure ACKs.
2720 SOCKBUF_LOCK(&so->so_snd);
2721 avail = sbavail(&so->so_snd) -
2722 (tp->snd_nxt - tp->snd_una);
2723 SOCKBUF_UNLOCK(&so->so_snd);
2725 (void) tcp_output(tp);
2726 sent = tp->snd_max - oldsndmax;
2727 if (sent > maxseg) {
2728 KASSERT((tp->t_dupacks == 2 &&
2729 tp->snd_limited == 0) ||
2730 (sent == maxseg + 1 &&
2731 tp->t_flags & TF_SENTFIN),
2732 ("%s: sent too much",
2734 tp->snd_limited = 2;
2735 } else if (sent > 0)
2737 tp->snd_cwnd = oldcwnd;
2744 * This ack is advancing the left edge, reset the
2749 * If this ack also has new SACK info, increment the
2750 * counter as per rfc6675. The variable
2751 * sack_changed tracks all changes to the SACK
2752 * scoreboard, including when partial ACKs without
2753 * SACK options are received, and clear the scoreboard
2754 * from the left side. Such partial ACKs should not be
2755 * counted as dupacks here.
2757 if ((tp->t_flags & TF_SACK_PERMIT) &&
2758 (to.to_flags & TOF_SACK) &&
2761 /* limit overhead by setting maxseg last */
2762 if (!IN_FASTRECOVERY(tp->t_flags) &&
2763 (tp->sackhint.sacked_bytes >
2764 ((tcprexmtthresh - 1) *
2765 (maxseg = tcp_maxseg(tp))))) {
2766 goto enter_recovery;
2772 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2773 ("%s: th_ack <= snd_una", __func__));
2776 * If the congestion window was inflated to account
2777 * for the other side's cached packets, retract it.
2779 if (IN_FASTRECOVERY(tp->t_flags)) {
2780 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2781 if (tp->t_flags & TF_SACK_PERMIT)
2782 if (V_tcp_do_prr && to.to_flags & TOF_SACK) {
2783 tcp_timer_activate(tp, TT_REXMT, 0);
2785 tcp_do_prr_ack(tp, th, &to);
2786 tp->t_flags |= TF_ACKNOW;
2787 (void) tcp_output(tp);
2789 tcp_sack_partialack(tp, th);
2791 tcp_newreno_partial_ack(tp, th);
2793 cc_post_recovery(tp, th);
2794 } else if (IN_CONGRECOVERY(tp->t_flags)) {
2795 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2797 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th);
2798 tp->snd_fack = th->th_ack;
2799 tcp_do_prr_ack(tp, th, &to);
2800 (void) tcp_output(tp);
2803 cc_post_recovery(tp, th);
2806 * If we reach this point, ACK is not a duplicate,
2807 * i.e., it ACKs something we sent.
2809 if (tp->t_flags & TF_NEEDSYN) {
2811 * T/TCP: Connection was half-synchronized, and our
2812 * SYN has been ACK'd (so connection is now fully
2813 * synchronized). Go to non-starred state,
2814 * increment snd_una for ACK of SYN, and check if
2815 * we can do window scaling.
2817 tp->t_flags &= ~TF_NEEDSYN;
2819 /* Do window scaling? */
2820 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2821 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2822 tp->rcv_scale = tp->request_r_scale;
2823 /* Send window already scaled. */
2828 INP_WLOCK_ASSERT(tp->t_inpcb);
2831 * Adjust for the SYN bit in sequence space,
2832 * but don't account for it in cwnd calculations.
2833 * This is for the SYN_RECEIVED, non-simultaneous
2834 * SYN case. SYN_SENT and simultaneous SYN are
2835 * treated elsewhere.
2839 acked = BYTES_THIS_ACK(tp, th);
2840 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2841 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2842 tp->snd_una, th->th_ack, tp, m));
2843 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2844 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2847 * If we just performed our first retransmit, and the ACK
2848 * arrives within our recovery window, then it was a mistake
2849 * to do the retransmit in the first place. Recover our
2850 * original cwnd and ssthresh, and proceed to transmit where
2853 if (tp->t_rxtshift == 1 &&
2854 tp->t_flags & TF_PREVVALID &&
2855 tp->t_badrxtwin != 0 &&
2856 to.to_flags & TOF_TS &&
2858 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
2859 cc_cong_signal(tp, th, CC_RTO_ERR);
2862 * If we have a timestamp reply, update smoothed
2863 * round trip time. If no timestamp is present but
2864 * transmit timer is running and timed sequence
2865 * number was acked, update smoothed round trip time.
2866 * Since we now have an rtt measurement, cancel the
2867 * timer backoff (cf., Phil Karn's retransmit alg.).
2868 * Recompute the initial retransmit timer.
2870 * Some boxes send broken timestamp replies
2871 * during the SYN+ACK phase, ignore
2872 * timestamps of 0 or we could calculate a
2873 * huge RTT and blow up the retransmit timer.
2875 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2878 t = tcp_ts_getticks() - to.to_tsecr;
2879 if (!tp->t_rttlow || tp->t_rttlow > t)
2881 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2882 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2883 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2884 tp->t_rttlow = ticks - tp->t_rtttime;
2885 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2889 * If all outstanding data is acked, stop retransmit
2890 * timer and remember to restart (more output or persist).
2891 * If there is more data to be acked, restart retransmit
2892 * timer, using current (possibly backed-off) value.
2894 if (th->th_ack == tp->snd_max) {
2895 tcp_timer_activate(tp, TT_REXMT, 0);
2897 } else if (!tcp_timer_active(tp, TT_PERSIST))
2898 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2901 * If no data (only SYN) was ACK'd,
2902 * skip rest of ACK processing.
2908 * Let the congestion control algorithm update congestion
2909 * control related information. This typically means increasing
2910 * the congestion window.
2912 cc_ack_received(tp, th, nsegs, CC_ACK);
2914 SOCKBUF_LOCK(&so->so_snd);
2915 if (acked > sbavail(&so->so_snd)) {
2916 if (tp->snd_wnd >= sbavail(&so->so_snd))
2917 tp->snd_wnd -= sbavail(&so->so_snd);
2920 mfree = sbcut_locked(&so->so_snd,
2921 (int)sbavail(&so->so_snd));
2924 mfree = sbcut_locked(&so->so_snd, acked);
2925 if (tp->snd_wnd >= (uint32_t) acked)
2926 tp->snd_wnd -= acked;
2931 /* NB: sowwakeup_locked() does an implicit unlock. */
2932 sowwakeup_locked(so);
2934 /* Detect una wraparound. */
2935 if (!IN_RECOVERY(tp->t_flags) &&
2936 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2937 SEQ_LEQ(th->th_ack, tp->snd_recover))
2938 tp->snd_recover = th->th_ack - 1;
2939 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2940 if (IN_RECOVERY(tp->t_flags) &&
2941 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2942 EXIT_RECOVERY(tp->t_flags);
2944 tp->snd_una = th->th_ack;
2945 if (tp->t_flags & TF_SACK_PERMIT) {
2946 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2947 tp->snd_recover = tp->snd_una;
2949 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2950 tp->snd_nxt = tp->snd_una;
2952 switch (tp->t_state) {
2954 * In FIN_WAIT_1 STATE in addition to the processing
2955 * for the ESTABLISHED state if our FIN is now acknowledged
2956 * then enter FIN_WAIT_2.
2958 case TCPS_FIN_WAIT_1:
2959 if (ourfinisacked) {
2961 * If we can't receive any more
2962 * data, then closing user can proceed.
2963 * Starting the timer is contrary to the
2964 * specification, but if we don't get a FIN
2965 * we'll hang forever.
2968 * we should release the tp also, and use a
2971 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2972 soisdisconnected(so);
2973 tcp_timer_activate(tp, TT_2MSL,
2974 (tcp_fast_finwait2_recycle ?
2975 tcp_finwait2_timeout :
2978 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2983 * In CLOSING STATE in addition to the processing for
2984 * the ESTABLISHED state if the ACK acknowledges our FIN
2985 * then enter the TIME-WAIT state, otherwise ignore
2989 if (ourfinisacked) {
2997 * In LAST_ACK, we may still be waiting for data to drain
2998 * and/or to be acked, as well as for the ack of our FIN.
2999 * If our FIN is now acknowledged, delete the TCB,
3000 * enter the closed state and return.
3003 if (ourfinisacked) {
3012 INP_WLOCK_ASSERT(tp->t_inpcb);
3015 * Update window information.
3016 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3018 if ((thflags & TH_ACK) &&
3019 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3020 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3021 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3022 /* keep track of pure window updates */
3024 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3025 TCPSTAT_INC(tcps_rcvwinupd);
3026 tp->snd_wnd = tiwin;
3027 tp->snd_wl1 = th->th_seq;
3028 tp->snd_wl2 = th->th_ack;
3029 if (tp->snd_wnd > tp->max_sndwnd)
3030 tp->max_sndwnd = tp->snd_wnd;
3035 * Process segments with URG.
3037 if ((thflags & TH_URG) && th->th_urp &&
3038 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3040 * This is a kludge, but if we receive and accept
3041 * random urgent pointers, we'll crash in
3042 * soreceive. It's hard to imagine someone
3043 * actually wanting to send this much urgent data.
3045 SOCKBUF_LOCK(&so->so_rcv);
3046 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3047 th->th_urp = 0; /* XXX */
3048 thflags &= ~TH_URG; /* XXX */
3049 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3050 goto dodata; /* XXX */
3053 * If this segment advances the known urgent pointer,
3054 * then mark the data stream. This should not happen
3055 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3056 * a FIN has been received from the remote side.
3057 * In these states we ignore the URG.
3059 * According to RFC961 (Assigned Protocols),
3060 * the urgent pointer points to the last octet
3061 * of urgent data. We continue, however,
3062 * to consider it to indicate the first octet
3063 * of data past the urgent section as the original
3064 * spec states (in one of two places).
3066 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3067 tp->rcv_up = th->th_seq + th->th_urp;
3068 so->so_oobmark = sbavail(&so->so_rcv) +
3069 (tp->rcv_up - tp->rcv_nxt) - 1;
3070 if (so->so_oobmark == 0)
3071 so->so_rcv.sb_state |= SBS_RCVATMARK;
3073 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3075 SOCKBUF_UNLOCK(&so->so_rcv);
3077 * Remove out of band data so doesn't get presented to user.
3078 * This can happen independent of advancing the URG pointer,
3079 * but if two URG's are pending at once, some out-of-band
3080 * data may creep in... ick.
3082 if (th->th_urp <= (uint32_t)tlen &&
3083 !(so->so_options & SO_OOBINLINE)) {
3084 /* hdr drop is delayed */
3085 tcp_pulloutofband(so, th, m, drop_hdrlen);
3089 * If no out of band data is expected,
3090 * pull receive urgent pointer along
3091 * with the receive window.
3093 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3094 tp->rcv_up = tp->rcv_nxt;
3097 INP_WLOCK_ASSERT(tp->t_inpcb);
3100 * Process the segment text, merging it into the TCP sequencing queue,
3101 * and arranging for acknowledgment of receipt if necessary.
3102 * This process logically involves adjusting tp->rcv_wnd as data
3103 * is presented to the user (this happens in tcp_usrreq.c,
3104 * case PRU_RCVD). If a FIN has already been received on this
3105 * connection then we just ignore the text.
3107 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3108 IS_FASTOPEN(tp->t_flags));
3109 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3110 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3111 tcp_seq save_start = th->th_seq;
3112 tcp_seq save_rnxt = tp->rcv_nxt;
3113 int save_tlen = tlen;
3114 m_adj(m, drop_hdrlen); /* delayed header drop */
3116 * Insert segment which includes th into TCP reassembly queue
3117 * with control block tp. Set thflags to whether reassembly now
3118 * includes a segment with FIN. This handles the common case
3119 * inline (segment is the next to be received on an established
3120 * connection, and the queue is empty), avoiding linkage into
3121 * and removal from the queue and repetition of various
3123 * Set DELACK for segments received in order, but ack
3124 * immediately when segments are out of order (so
3125 * fast retransmit can work).
3127 if (th->th_seq == tp->rcv_nxt &&
3129 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3131 if (DELAY_ACK(tp, tlen) || tfo_syn)
3132 tp->t_flags |= TF_DELACK;
3134 tp->t_flags |= TF_ACKNOW;
3135 tp->rcv_nxt += tlen;
3137 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3138 (tp->t_fbyte_in == 0)) {
3139 tp->t_fbyte_in = ticks;
3140 if (tp->t_fbyte_in == 0)
3142 if (tp->t_fbyte_out && tp->t_fbyte_in)
3143 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3145 thflags = tcp_get_flags(th) & TH_FIN;
3146 TCPSTAT_INC(tcps_rcvpack);
3147 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3148 SOCKBUF_LOCK(&so->so_rcv);
3149 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3152 sbappendstream_locked(&so->so_rcv, m, 0);
3153 tp->t_flags |= TF_WAKESOR;
3156 * XXX: Due to the header drop above "th" is
3157 * theoretically invalid by now. Fortunately
3158 * m_adj() doesn't actually frees any mbufs
3159 * when trimming from the head.
3161 tcp_seq temp = save_start;
3163 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3164 tp->t_flags |= TF_ACKNOW;
3166 if ((tp->t_flags & TF_SACK_PERMIT) &&
3168 TCPS_HAVEESTABLISHED(tp->t_state)) {
3169 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3171 * DSACK actually handled in the fastpath
3174 tcp_update_sack_list(tp, save_start,
3175 save_start + save_tlen);
3176 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3177 if ((tp->rcv_numsacks >= 1) &&
3178 (tp->sackblks[0].end == save_start)) {
3180 * Partial overlap, recorded at todrop
3183 tcp_update_sack_list(tp,
3184 tp->sackblks[0].start,
3185 tp->sackblks[0].end);
3187 tcp_update_dsack_list(tp, save_start,
3188 save_start + save_tlen);
3190 } else if (tlen >= save_tlen) {
3191 /* Update of sackblks. */
3192 tcp_update_dsack_list(tp, save_start,
3193 save_start + save_tlen);
3194 } else if (tlen > 0) {
3195 tcp_update_dsack_list(tp, save_start,
3199 tcp_handle_wakeup(tp, so);
3202 * Note the amount of data that peer has sent into
3203 * our window, in order to estimate the sender's
3207 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3208 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3210 len = so->so_rcv.sb_hiwat;
3218 * If FIN is received ACK the FIN and let the user know
3219 * that the connection is closing.
3221 if (thflags & TH_FIN) {
3222 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3223 /* The socket upcall is handled by socantrcvmore. */
3226 * If connection is half-synchronized
3227 * (ie NEEDSYN flag on) then delay ACK,
3228 * so it may be piggybacked when SYN is sent.
3229 * Otherwise, since we received a FIN then no
3230 * more input can be expected, send ACK now.
3232 if (tp->t_flags & TF_NEEDSYN)
3233 tp->t_flags |= TF_DELACK;
3235 tp->t_flags |= TF_ACKNOW;
3238 switch (tp->t_state) {
3240 * In SYN_RECEIVED and ESTABLISHED STATES
3241 * enter the CLOSE_WAIT state.
3243 case TCPS_SYN_RECEIVED:
3244 tp->t_starttime = ticks;
3246 case TCPS_ESTABLISHED:
3247 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3251 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3252 * enter the CLOSING state.
3254 case TCPS_FIN_WAIT_1:
3255 tcp_state_change(tp, TCPS_CLOSING);
3259 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3260 * starting the time-wait timer, turning off the other
3263 case TCPS_FIN_WAIT_2:
3269 if (so->so_options & SO_DEBUG)
3270 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3273 TCP_PROBE3(debug__input, tp, th, m);
3276 * Return any desired output.
3278 if (needoutput || (tp->t_flags & TF_ACKNOW))
3279 (void) tcp_output(tp);
3282 INP_WLOCK_ASSERT(tp->t_inpcb);
3284 if (tp->t_flags & TF_DELACK) {
3285 tp->t_flags &= ~TF_DELACK;
3286 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3288 INP_WUNLOCK(tp->t_inpcb);
3293 * Generate an ACK dropping incoming segment if it occupies
3294 * sequence space, where the ACK reflects our state.
3296 * We can now skip the test for the RST flag since all
3297 * paths to this code happen after packets containing
3298 * RST have been dropped.
3300 * In the SYN-RECEIVED state, don't send an ACK unless the
3301 * segment we received passes the SYN-RECEIVED ACK test.
3302 * If it fails send a RST. This breaks the loop in the
3303 * "LAND" DoS attack, and also prevents an ACK storm
3304 * between two listening ports that have been sent forged
3305 * SYN segments, each with the source address of the other.
3307 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3308 (SEQ_GT(tp->snd_una, th->th_ack) ||
3309 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3310 rstreason = BANDLIM_RST_OPENPORT;
3314 if (so->so_options & SO_DEBUG)
3315 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3318 TCP_PROBE3(debug__input, tp, th, m);
3319 tp->t_flags |= TF_ACKNOW;
3320 (void) tcp_output(tp);
3321 INP_WUNLOCK(tp->t_inpcb);
3327 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3328 INP_WUNLOCK(tp->t_inpcb);
3330 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3335 * Drop space held by incoming segment and return.
3338 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3339 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3342 TCP_PROBE3(debug__input, tp, th, m);
3344 INP_WUNLOCK(tp->t_inpcb);
3350 * Issue RST and make ACK acceptable to originator of segment.
3351 * The mbuf must still include the original packet header.
3355 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3356 int tlen, int rstreason)
3362 struct ip6_hdr *ip6;
3366 INP_LOCK_ASSERT(tp->t_inpcb);
3369 /* Don't bother if destination was broadcast/multicast. */
3370 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3373 if (mtod(m, struct ip *)->ip_v == 6) {
3374 ip6 = mtod(m, struct ip6_hdr *);
3375 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3376 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3378 /* IPv6 anycast check is done at tcp6_input() */
3381 #if defined(INET) && defined(INET6)
3386 ip = mtod(m, struct ip *);
3387 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3388 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3389 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3390 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3395 /* Perform bandwidth limiting. */
3396 if (badport_bandlim(rstreason) < 0)
3399 /* tcp_respond consumes the mbuf chain. */
3400 if (tcp_get_flags(th) & TH_ACK) {
3401 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3402 th->th_ack, TH_RST);
3404 if (tcp_get_flags(th) & TH_SYN)
3406 if (tcp_get_flags(th) & TH_FIN)
3408 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3409 (tcp_seq)0, TH_RST|TH_ACK);
3417 * Parse TCP options and place in tcpopt.
3420 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3425 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3427 if (opt == TCPOPT_EOL)
3429 if (opt == TCPOPT_NOP)
3435 if (optlen < 2 || optlen > cnt)
3440 if (optlen != TCPOLEN_MAXSEG)
3442 if (!(flags & TO_SYN))
3444 to->to_flags |= TOF_MSS;
3445 bcopy((char *)cp + 2,
3446 (char *)&to->to_mss, sizeof(to->to_mss));
3447 to->to_mss = ntohs(to->to_mss);
3450 if (optlen != TCPOLEN_WINDOW)
3452 if (!(flags & TO_SYN))
3454 to->to_flags |= TOF_SCALE;
3455 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3457 case TCPOPT_TIMESTAMP:
3458 if (optlen != TCPOLEN_TIMESTAMP)
3460 to->to_flags |= TOF_TS;
3461 bcopy((char *)cp + 2,
3462 (char *)&to->to_tsval, sizeof(to->to_tsval));
3463 to->to_tsval = ntohl(to->to_tsval);
3464 bcopy((char *)cp + 6,
3465 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3466 to->to_tsecr = ntohl(to->to_tsecr);
3468 case TCPOPT_SIGNATURE:
3470 * In order to reply to a host which has set the
3471 * TCP_SIGNATURE option in its initial SYN, we have
3472 * to record the fact that the option was observed
3473 * here for the syncache code to perform the correct
3476 if (optlen != TCPOLEN_SIGNATURE)
3478 to->to_flags |= TOF_SIGNATURE;
3479 to->to_signature = cp + 2;
3481 case TCPOPT_SACK_PERMITTED:
3482 if (optlen != TCPOLEN_SACK_PERMITTED)
3484 if (!(flags & TO_SYN))
3488 to->to_flags |= TOF_SACKPERM;
3491 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3495 to->to_flags |= TOF_SACK;
3496 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3497 to->to_sacks = cp + 2;
3498 TCPSTAT_INC(tcps_sack_rcv_blocks);
3500 case TCPOPT_FAST_OPEN:
3502 * Cookie length validation is performed by the
3503 * server side cookie checking code or the client
3504 * side cookie cache update code.
3506 if (!(flags & TO_SYN))
3508 if (!V_tcp_fastopen_client_enable &&
3509 !V_tcp_fastopen_server_enable)
3511 to->to_flags |= TOF_FASTOPEN;
3512 to->to_tfo_len = optlen - 2;
3513 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3522 * Pull out of band byte out of a segment so
3523 * it doesn't appear in the user's data queue.
3524 * It is still reflected in the segment length for
3525 * sequencing purposes.
3528 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3531 int cnt = off + th->th_urp - 1;
3534 if (m->m_len > cnt) {
3535 char *cp = mtod(m, caddr_t) + cnt;
3536 struct tcpcb *tp = sototcpcb(so);
3538 INP_WLOCK_ASSERT(tp->t_inpcb);
3541 tp->t_oobflags |= TCPOOB_HAVEDATA;
3542 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3544 if (m->m_flags & M_PKTHDR)
3553 panic("tcp_pulloutofband");
3557 * Collect new round-trip time estimate
3558 * and update averages and current timeout.
3561 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3565 INP_WLOCK_ASSERT(tp->t_inpcb);
3567 TCPSTAT_INC(tcps_rttupdated);
3570 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3571 imax(0, rtt * 1000 / hz));
3573 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3575 * srtt is stored as fixed point with 5 bits after the
3576 * binary point (i.e., scaled by 8). The following magic
3577 * is equivalent to the smoothing algorithm in rfc793 with
3578 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3579 * point). Adjust rtt to origin 0.
3581 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3582 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3584 if ((tp->t_srtt += delta) <= 0)
3588 * We accumulate a smoothed rtt variance (actually, a
3589 * smoothed mean difference), then set the retransmit
3590 * timer to smoothed rtt + 4 times the smoothed variance.
3591 * rttvar is stored as fixed point with 4 bits after the
3592 * binary point (scaled by 16). The following is
3593 * equivalent to rfc793 smoothing with an alpha of .75
3594 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3595 * rfc793's wired-in beta.
3599 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3600 if ((tp->t_rttvar += delta) <= 0)
3602 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3603 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3606 * No rtt measurement yet - use the unsmoothed rtt.
3607 * Set the variance to half the rtt (so our first
3608 * retransmit happens at 3*rtt).
3610 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3611 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3612 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3618 * the retransmit should happen at rtt + 4 * rttvar.
3619 * Because of the way we do the smoothing, srtt and rttvar
3620 * will each average +1/2 tick of bias. When we compute
3621 * the retransmit timer, we want 1/2 tick of rounding and
3622 * 1 extra tick because of +-1/2 tick uncertainty in the
3623 * firing of the timer. The bias will give us exactly the
3624 * 1.5 tick we need. But, because the bias is
3625 * statistical, we have to test that we don't drop below
3626 * the minimum feasible timer (which is 2 ticks).
3628 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3629 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3632 * We received an ack for a packet that wasn't retransmitted;
3633 * it is probably safe to discard any error indications we've
3634 * received recently. This isn't quite right, but close enough
3635 * for now (a route might have failed after we sent a segment,
3636 * and the return path might not be symmetrical).
3638 tp->t_softerror = 0;
3642 * Determine a reasonable value for maxseg size.
3643 * If the route is known, check route for mtu.
3644 * If none, use an mss that can be handled on the outgoing interface
3645 * without forcing IP to fragment. If no route is found, route has no mtu,
3646 * or the destination isn't local, use a default, hopefully conservative
3647 * size (usually 512 or the default IP max size, but no more than the mtu
3648 * of the interface), as we can't discover anything about intervening
3649 * gateways or networks. We also initialize the congestion/slow start
3650 * window to be a single segment if the destination isn't local.
3651 * While looking at the routing entry, we also initialize other path-dependent
3652 * parameters from pre-set or cached values in the routing entry.
3654 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3655 * IP options, e.g. IPSEC data, since length of this data may vary, and
3656 * thus it is calculated for every segment separately in tcp_output().
3658 * NOTE that this routine is only called when we process an incoming
3659 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3660 * settings are handled in tcp_mssopt().
3663 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3664 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3667 uint32_t maxmtu = 0;
3668 struct inpcb *inp = tp->t_inpcb;
3669 struct hc_metrics_lite metrics;
3671 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3672 size_t min_protoh = isipv6 ?
3673 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3674 sizeof (struct tcpiphdr);
3676 size_t min_protoh = sizeof(struct tcpiphdr);
3679 INP_WLOCK_ASSERT(tp->t_inpcb);
3682 min_protoh += V_tcp_udp_tunneling_overhead;
3683 if (mtuoffer != -1) {
3684 KASSERT(offer == -1, ("%s: conflict", __func__));
3685 offer = mtuoffer - min_protoh;
3691 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3692 tp->t_maxseg = V_tcp_v6mssdflt;
3695 #if defined(INET) && defined(INET6)
3700 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3701 tp->t_maxseg = V_tcp_mssdflt;
3706 * No route to sender, stay with default mss and return.
3710 * In case we return early we need to initialize metrics
3711 * to a defined state as tcp_hc_get() would do for us
3712 * if there was no cache hit.
3714 if (metricptr != NULL)
3715 bzero(metricptr, sizeof(struct hc_metrics_lite));
3719 /* What have we got? */
3723 * Offer == 0 means that there was no MSS on the SYN
3724 * segment, in this case we use tcp_mssdflt as
3725 * already assigned to t_maxseg above.
3727 offer = tp->t_maxseg;
3732 * Offer == -1 means that we didn't receive SYN yet.
3738 * Prevent DoS attack with too small MSS. Round up
3739 * to at least minmss.
3741 offer = max(offer, V_tcp_minmss);
3745 * rmx information is now retrieved from tcp_hostcache.
3747 tcp_hc_get(&inp->inp_inc, &metrics);
3748 if (metricptr != NULL)
3749 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3752 * If there's a discovered mtu in tcp hostcache, use it.
3753 * Else, use the link mtu.
3755 if (metrics.rmx_mtu)
3756 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3760 mss = maxmtu - min_protoh;
3761 if (!V_path_mtu_discovery &&
3762 !in6_localaddr(&inp->in6p_faddr))
3763 mss = min(mss, V_tcp_v6mssdflt);
3766 #if defined(INET) && defined(INET6)
3771 mss = maxmtu - min_protoh;
3772 if (!V_path_mtu_discovery &&
3773 !in_localaddr(inp->inp_faddr))
3774 mss = min(mss, V_tcp_mssdflt);
3778 * XXX - The above conditional (mss = maxmtu - min_protoh)
3779 * probably violates the TCP spec.
3780 * The problem is that, since we don't know the
3781 * other end's MSS, we are supposed to use a conservative
3782 * default. But, if we do that, then MTU discovery will
3783 * never actually take place, because the conservative
3784 * default is much less than the MTUs typically seen
3785 * on the Internet today. For the moment, we'll sweep
3786 * this under the carpet.
3788 * The conservative default might not actually be a problem
3789 * if the only case this occurs is when sending an initial
3790 * SYN with options and data to a host we've never talked
3791 * to before. Then, they will reply with an MSS value which
3792 * will get recorded and the new parameters should get
3793 * recomputed. For Further Study.
3796 mss = min(mss, offer);
3799 * Sanity check: make sure that maxseg will be large
3800 * enough to allow some data on segments even if the
3801 * all the option space is used (40bytes). Otherwise
3802 * funny things may happen in tcp_output.
3804 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3812 tcp_mss(struct tcpcb *tp, int offer)
3818 struct hc_metrics_lite metrics;
3819 struct tcp_ifcap cap;
3821 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3823 bzero(&cap, sizeof(cap));
3824 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3830 * If there's a pipesize, change the socket buffer to that size,
3831 * don't change if sb_hiwat is different than default (then it
3832 * has been changed on purpose with setsockopt).
3833 * Make the socket buffers an integral number of mss units;
3834 * if the mss is larger than the socket buffer, decrease the mss.
3836 so = inp->inp_socket;
3837 SOCKBUF_LOCK(&so->so_snd);
3838 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3839 bufsize = metrics.rmx_sendpipe;
3841 bufsize = so->so_snd.sb_hiwat;
3845 bufsize = roundup(bufsize, mss);
3846 if (bufsize > sb_max)
3848 if (bufsize > so->so_snd.sb_hiwat)
3849 (void)sbreserve_locked(so, SO_SND, bufsize, NULL);
3851 SOCKBUF_UNLOCK(&so->so_snd);
3853 * Sanity check: make sure that maxseg will be large
3854 * enough to allow some data on segments even if the
3855 * all the option space is used (40bytes). Otherwise
3856 * funny things may happen in tcp_output.
3858 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3860 tp->t_maxseg = max(mss, 64);
3862 SOCKBUF_LOCK(&so->so_rcv);
3863 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3864 bufsize = metrics.rmx_recvpipe;
3866 bufsize = so->so_rcv.sb_hiwat;
3867 if (bufsize > mss) {
3868 bufsize = roundup(bufsize, mss);
3869 if (bufsize > sb_max)
3871 if (bufsize > so->so_rcv.sb_hiwat)
3872 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL);
3874 SOCKBUF_UNLOCK(&so->so_rcv);
3876 /* Check the interface for TSO capabilities. */
3877 if (cap.ifcap & CSUM_TSO) {
3878 tp->t_flags |= TF_TSO;
3879 tp->t_tsomax = cap.tsomax;
3880 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3881 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3886 * Determine the MSS option to send on an outgoing SYN.
3889 tcp_mssopt(struct in_conninfo *inc)
3892 uint32_t thcmtu = 0;
3893 uint32_t maxmtu = 0;
3896 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3899 if (inc->inc_flags & INC_ISIPV6) {
3900 mss = V_tcp_v6mssdflt;
3901 maxmtu = tcp_maxmtu6(inc, NULL);
3902 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3905 #if defined(INET) && defined(INET6)
3910 mss = V_tcp_mssdflt;
3911 maxmtu = tcp_maxmtu(inc, NULL);
3912 min_protoh = sizeof(struct tcpiphdr);
3915 #if defined(INET6) || defined(INET)
3916 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3919 if (maxmtu && thcmtu)
3920 mss = min(maxmtu, thcmtu) - min_protoh;
3921 else if (maxmtu || thcmtu)
3922 mss = max(maxmtu, thcmtu) - min_protoh;
3928 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
3930 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3931 int maxseg = tcp_maxseg(tp);
3933 INP_WLOCK_ASSERT(tp->t_inpcb);
3936 * Compute the amount of data that this ACK is indicating
3937 * (del_data) and an estimate of how many bytes are in the
3940 if (((tp->t_flags & TF_SACK_PERMIT) &&
3941 (to->to_flags & TOF_SACK)) ||
3942 (IN_CONGRECOVERY(tp->t_flags) &&
3943 !IN_FASTRECOVERY(tp->t_flags))) {
3944 del_data = tp->sackhint.delivered_data;
3945 if (V_tcp_do_newsack)
3946 pipe = tcp_compute_pipe(tp);
3948 pipe = (tp->snd_nxt - tp->snd_fack) +
3949 tp->sackhint.sack_bytes_rexmit;
3951 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg +
3952 tp->snd_recover - tp->snd_una))
3954 pipe = imax(0, tp->snd_max - tp->snd_una -
3955 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg);
3957 tp->sackhint.prr_delivered += del_data;
3959 * Proportional Rate Reduction
3961 if (pipe >= tp->snd_ssthresh) {
3962 if (tp->sackhint.recover_fs == 0)
3963 tp->sackhint.recover_fs =
3964 imax(1, tp->snd_nxt - tp->snd_una);
3965 snd_cnt = howmany((long)tp->sackhint.prr_delivered *
3966 tp->snd_ssthresh, tp->sackhint.recover_fs) -
3967 tp->sackhint.prr_out;
3969 if (V_tcp_do_prr_conservative || (del_data == 0))
3970 limit = tp->sackhint.prr_delivered -
3971 tp->sackhint.prr_out;
3973 limit = imax(tp->sackhint.prr_delivered -
3974 tp->sackhint.prr_out, del_data) +
3976 snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
3978 snd_cnt = imax(snd_cnt, 0) / maxseg;
3980 * Send snd_cnt new data into the network in response to this ack.
3981 * If there is going to be a SACK retransmission, adjust snd_cwnd
3984 if (IN_FASTRECOVERY(tp->t_flags)) {
3985 if ((tp->t_flags & TF_SACK_PERMIT) &&
3986 (to->to_flags & TOF_SACK)) {
3987 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
3988 tp->sackhint.sack_bytes_rexmit +
3991 tp->snd_cwnd = (tp->snd_max - tp->snd_una) +
3994 } else if (IN_CONGRECOVERY(tp->t_flags))
3995 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg);
3996 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd);
4000 * On a partial ack arrives, force the retransmission of the
4001 * next unacknowledged segment. Do not clear tp->t_dupacks.
4002 * By setting snd_nxt to ti_ack, this forces retransmission timer to
4006 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4008 tcp_seq onxt = tp->snd_nxt;
4009 uint32_t ocwnd = tp->snd_cwnd;
4010 u_int maxseg = tcp_maxseg(tp);
4012 INP_WLOCK_ASSERT(tp->t_inpcb);
4014 tcp_timer_activate(tp, TT_REXMT, 0);
4016 tp->snd_nxt = th->th_ack;
4018 * Set snd_cwnd to one segment beyond acknowledged offset.
4019 * (tp->snd_una has not yet been updated when this function is called.)
4021 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4022 tp->t_flags |= TF_ACKNOW;
4023 (void) tcp_output(tp);
4024 tp->snd_cwnd = ocwnd;
4025 if (SEQ_GT(onxt, tp->snd_nxt))
4028 * Partial window deflation. Relies on fact that tp->snd_una
4031 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4032 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4035 tp->snd_cwnd += maxseg;
4039 tcp_compute_pipe(struct tcpcb *tp)
4041 return (tp->snd_max - tp->snd_una +
4042 tp->sackhint.sack_bytes_rexmit -
4043 tp->sackhint.sacked_bytes);
4047 tcp_compute_initwnd(uint32_t maxseg)
4050 * Calculate the Initial Window, also used as Restart Window
4052 * RFC5681 Section 3.1 specifies the default conservative values.
4053 * RFC3390 specifies slightly more aggressive values.
4054 * RFC6928 increases it to ten segments.
4055 * Support for user specified value for initial flight size.
4057 if (V_tcp_initcwnd_segments)
4058 return min(V_tcp_initcwnd_segments * maxseg,
4059 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4060 else if (V_tcp_do_rfc3390)
4061 return min(4 * maxseg, max(2 * maxseg, 4380));
4063 /* Per RFC5681 Section 3.1 */
4065 return (2 * maxseg);
4066 else if (maxseg > 1095)
4067 return (3 * maxseg);
4069 return (4 * maxseg);