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)(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
1086 * inp and tcpinfo locks.
1088 rstreason = syncache_expand(&inc, &to, th, &so, m, port);
1089 if (rstreason < 0) {
1091 * A failing TCP MD5 signature comparison
1092 * must result in the segment being dropped
1093 * and must not produce any response back
1097 } else if (rstreason == 0) {
1099 * No syncache entry or ACK was not
1100 * for our SYN/ACK. Send a RST.
1101 * NB: syncache did its own logging
1102 * of the failure cause.
1104 rstreason = BANDLIM_RST_OPENPORT;
1110 * We completed the 3-way handshake
1111 * but could not allocate a socket
1112 * either due to memory shortage,
1113 * listen queue length limits or
1114 * global socket limits. Send RST
1115 * or wait and have the remote end
1116 * retransmit the ACK for another
1119 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1120 log(LOG_DEBUG, "%s; %s: Listen socket: "
1121 "Socket allocation failed due to "
1122 "limits or memory shortage, %s\n",
1124 V_tcp_sc_rst_sock_fail ?
1125 "sending RST" : "try again");
1126 if (V_tcp_sc_rst_sock_fail) {
1127 rstreason = BANDLIM_UNLIMITED;
1133 * Socket is created in state SYN_RECEIVED.
1134 * Unlock the listen socket, lock the newly
1135 * created socket and update the tp variable.
1136 * If we came here via jump to tfo_socket_result,
1137 * then listening socket is read-locked.
1139 INP_UNLOCK(inp); /* listen socket */
1140 inp = sotoinpcb(so);
1142 * New connection inpcb is already locked by
1143 * syncache_expand().
1145 INP_WLOCK_ASSERT(inp);
1146 tp = intotcpcb(inp);
1147 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1148 ("%s: ", __func__));
1150 * Process the segment and the data it
1151 * contains. tcp_do_segment() consumes
1152 * the mbuf chain and unlocks the inpcb.
1154 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1155 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1157 return (IPPROTO_DONE);
1160 * Segment flag validation for new connection attempts:
1162 * Our (SYN|ACK) response was rejected.
1163 * Check with syncache and remove entry to prevent
1166 * NB: syncache_chkrst does its own logging of failure
1169 if (thflags & TH_RST) {
1170 syncache_chkrst(&inc, th, m, port);
1174 * We can't do anything without SYN.
1176 if ((thflags & TH_SYN) == 0) {
1177 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1178 log(LOG_DEBUG, "%s; %s: Listen socket: "
1179 "SYN is missing, segment ignored\n",
1181 TCPSTAT_INC(tcps_badsyn);
1185 * (SYN|ACK) is bogus on a listen socket.
1187 if (thflags & TH_ACK) {
1188 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1189 log(LOG_DEBUG, "%s; %s: Listen socket: "
1190 "SYN|ACK invalid, segment rejected\n",
1192 syncache_badack(&inc, port); /* XXX: Not needed! */
1193 TCPSTAT_INC(tcps_badsyn);
1194 rstreason = BANDLIM_RST_OPENPORT;
1198 * If the drop_synfin option is enabled, drop all
1199 * segments with both the SYN and FIN bits set.
1200 * This prevents e.g. nmap from identifying the
1202 * XXX: Poor reasoning. nmap has other methods
1203 * and is constantly refining its stack detection
1205 * XXX: This is a violation of the TCP specification
1206 * and was used by RFC1644.
1208 if ((thflags & TH_FIN) && V_drop_synfin) {
1209 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1210 log(LOG_DEBUG, "%s; %s: Listen socket: "
1211 "SYN|FIN segment ignored (based on "
1212 "sysctl setting)\n", s, __func__);
1213 TCPSTAT_INC(tcps_badsyn);
1217 * Segment's flags are (SYN) or (SYN|FIN).
1219 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1220 * as they do not affect the state of the TCP FSM.
1221 * The data pointed to by TH_URG and th_urp is ignored.
1223 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1224 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1225 KASSERT(thflags & (TH_SYN),
1226 ("%s: Listen socket: TH_SYN not set", __func__));
1227 INP_RLOCK_ASSERT(inp);
1230 * If deprecated address is forbidden,
1231 * we do not accept SYN to deprecated interface
1232 * address to prevent any new inbound connection from
1233 * getting established.
1234 * When we do not accept SYN, we send a TCP RST,
1235 * with deprecated source address (instead of dropping
1236 * it). We compromise it as it is much better for peer
1237 * to send a RST, and RST will be the final packet
1240 * If we do not forbid deprecated addresses, we accept
1241 * the SYN packet. RFC2462 does not suggest dropping
1243 * If we decipher RFC2462 5.5.4, it says like this:
1244 * 1. use of deprecated addr with existing
1245 * communication is okay - "SHOULD continue to be
1247 * 2. use of it with new communication:
1248 * (2a) "SHOULD NOT be used if alternate address
1249 * with sufficient scope is available"
1250 * (2b) nothing mentioned otherwise.
1251 * Here we fall into (2b) case as we have no choice in
1252 * our source address selection - we must obey the peer.
1254 * The wording in RFC2462 is confusing, and there are
1255 * multiple description text for deprecated address
1256 * handling - worse, they are not exactly the same.
1257 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1259 if (isipv6 && !V_ip6_use_deprecated) {
1260 struct in6_ifaddr *ia6;
1262 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1264 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1265 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1266 log(LOG_DEBUG, "%s; %s: Listen socket: "
1267 "Connection attempt to deprecated "
1268 "IPv6 address rejected\n",
1270 rstreason = BANDLIM_RST_OPENPORT;
1276 * Basic sanity checks on incoming SYN requests:
1277 * Don't respond if the destination is a link layer
1278 * broadcast according to RFC1122 4.2.3.10, p. 104.
1279 * If it is from this socket it must be forged.
1280 * Don't respond if the source or destination is a
1281 * global or subnet broad- or multicast address.
1282 * Note that it is quite possible to receive unicast
1283 * link-layer packets with a broadcast IP address. Use
1284 * in_broadcast() to find them.
1286 if (m->m_flags & (M_BCAST|M_MCAST)) {
1287 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1288 log(LOG_DEBUG, "%s; %s: Listen socket: "
1289 "Connection attempt from broad- or multicast "
1290 "link layer address ignored\n", s, __func__);
1295 if (th->th_dport == th->th_sport &&
1296 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1297 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1298 log(LOG_DEBUG, "%s; %s: Listen socket: "
1299 "Connection attempt to/from self "
1300 "ignored\n", s, __func__);
1303 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1304 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1305 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1306 log(LOG_DEBUG, "%s; %s: Listen socket: "
1307 "Connection attempt from/to multicast "
1308 "address ignored\n", s, __func__);
1313 #if defined(INET) && defined(INET6)
1318 if (th->th_dport == th->th_sport &&
1319 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1320 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1321 log(LOG_DEBUG, "%s; %s: Listen socket: "
1322 "Connection attempt from/to self "
1323 "ignored\n", s, __func__);
1326 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1327 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1328 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1329 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1330 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1331 log(LOG_DEBUG, "%s; %s: Listen socket: "
1332 "Connection attempt from/to broad- "
1333 "or multicast address ignored\n",
1340 * SYN appears to be valid. Create compressed TCP state
1344 if (so->so_options & SO_DEBUG)
1345 tcp_trace(TA_INPUT, ostate, tp,
1346 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1348 TCP_PROBE3(debug__input, tp, th, m);
1349 tcp_dooptions(&to, optp, optlen, TO_SYN);
1350 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL,
1351 iptos, port)) != NULL)
1352 goto tfo_socket_result;
1355 * Entry added to syncache and mbuf consumed.
1356 * Only the listen socket is unlocked by syncache_add().
1358 return (IPPROTO_DONE);
1359 } else if (tp->t_state == TCPS_LISTEN) {
1361 * When a listen socket is torn down the SO_ACCEPTCONN
1362 * flag is removed first while connections are drained
1363 * from the accept queue in a unlock/lock cycle of the
1364 * ACCEPT_LOCK, opening a race condition allowing a SYN
1365 * attempt go through unhandled.
1369 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1370 if (tp->t_flags & TF_SIGNATURE) {
1371 tcp_dooptions(&to, optp, optlen, thflags);
1372 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1373 TCPSTAT_INC(tcps_sig_err_nosigopt);
1376 if (!TCPMD5_ENABLED() ||
1377 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1381 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1384 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1385 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1386 * the inpcb, and unlocks pcbinfo.
1388 * XXXGL: in case of a pure SYN arriving on existing connection
1389 * TCP stacks won't need to modify the PCB, they would either drop
1390 * the segment silently, or send a challenge ACK. However, we try
1391 * to upgrade the lock, because calling convention for stacks is
1392 * write-lock on PCB. If upgrade fails, drop the SYN.
1394 if (lookupflag == INPLOOKUP_RLOCKPCB && INP_TRY_UPGRADE(inp) == 0)
1397 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1398 return (IPPROTO_DONE);
1401 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1404 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1407 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1408 m = NULL; /* mbuf chain got consumed. */
1413 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1423 return (IPPROTO_DONE);
1427 * Automatic sizing of receive socket buffer. Often the send
1428 * buffer size is not optimally adjusted to the actual network
1429 * conditions at hand (delay bandwidth product). Setting the
1430 * buffer size too small limits throughput on links with high
1431 * bandwidth and high delay (eg. trans-continental/oceanic links).
1433 * On the receive side the socket buffer memory is only rarely
1434 * used to any significant extent. This allows us to be much
1435 * more aggressive in scaling the receive socket buffer. For
1436 * the case that the buffer space is actually used to a large
1437 * extent and we run out of kernel memory we can simply drop
1438 * the new segments; TCP on the sender will just retransmit it
1439 * later. Setting the buffer size too big may only consume too
1440 * much kernel memory if the application doesn't read() from
1441 * the socket or packet loss or reordering makes use of the
1444 * The criteria to step up the receive buffer one notch are:
1445 * 1. Application has not set receive buffer size with
1446 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1447 * 2. the number of bytes received during 1/2 of an sRTT
1448 * is at least 3/8 of the current socket buffer size.
1449 * 3. receive buffer size has not hit maximal automatic size;
1451 * If all of the criteria are met we increaset the socket buffer
1452 * by a 1/2 (bounded by the max). This allows us to keep ahead
1453 * of slow-start but also makes it so our peer never gets limited
1454 * by our rwnd which we then open up causing a burst.
1456 * This algorithm does two steps per RTT at most and only if
1457 * we receive a bulk stream w/o packet losses or reorderings.
1458 * Shrinking the buffer during idle times is not necessary as
1459 * it doesn't consume any memory when idle.
1461 * TODO: Only step up if the application is actually serving
1462 * the buffer to better manage the socket buffer resources.
1465 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1466 struct tcpcb *tp, int tlen)
1470 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1471 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1472 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1473 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1474 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1475 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1476 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1478 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1480 /* Start over with next RTT. */
1484 tp->rfbuf_cnt += tlen; /* add up */
1490 tcp_input(struct mbuf **mp, int *offp, int proto)
1492 return(tcp_input_with_port(mp, offp, proto, 0));
1496 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1499 * Since tp might be gone if the session entered
1500 * the TIME_WAIT state before coming here, we need
1501 * to check if the socket is still connected.
1509 INP_LOCK_ASSERT(tp->t_inpcb);
1510 if (tp->t_flags & TF_WAKESOR) {
1511 tp->t_flags &= ~TF_WAKESOR;
1512 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1513 sorwakeup_locked(so);
1518 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1519 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1522 int acked, ourfinisacked, needoutput = 0, sack_changed;
1523 int rstreason, todrop, win, incforsyn = 0;
1527 struct in_conninfo *inc;
1535 * The size of tcp_saveipgen must be the size of the max ip header,
1538 u_char tcp_saveipgen[IP6_HDR_LEN];
1539 struct tcphdr tcp_savetcp;
1542 thflags = tcp_get_flags(th);
1543 inc = &tp->t_inpcb->inp_inc;
1544 tp->sackhint.last_sack_ack = 0;
1546 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1549 INP_WLOCK_ASSERT(tp->t_inpcb);
1550 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1552 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1556 /* Save segment, if requested. */
1557 tcp_pcap_add(th, m, &(tp->t_inpkts));
1559 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1562 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1563 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1564 log(LOG_DEBUG, "%s; %s: "
1565 "SYN|FIN segment ignored (based on "
1566 "sysctl setting)\n", s, __func__);
1573 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1574 * check SEQ.ACK first.
1576 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1577 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1578 rstreason = BANDLIM_UNLIMITED;
1583 * Segment received on connection.
1584 * Reset idle time and keep-alive timer.
1585 * XXX: This should be done after segment
1586 * validation to ignore broken/spoofed segs.
1588 tp->t_rcvtime = ticks;
1591 * Scale up the window into a 32-bit value.
1592 * For the SYN_SENT state the scale is zero.
1594 tiwin = th->th_win << tp->snd_scale;
1596 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1600 * TCP ECN processing.
1602 if (tcp_ecn_input_segment(tp, thflags, iptos))
1603 cc_cong_signal(tp, th, CC_ECN);
1606 * Parse options on any incoming segment.
1608 tcp_dooptions(&to, (u_char *)(th + 1),
1609 (th->th_off << 2) - sizeof(struct tcphdr),
1610 (thflags & TH_SYN) ? TO_SYN : 0);
1612 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1613 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1614 (to.to_flags & TOF_SIGNATURE) == 0) {
1615 TCPSTAT_INC(tcps_sig_err_sigopt);
1616 /* XXX: should drop? */
1620 * If echoed timestamp is later than the current time,
1621 * fall back to non RFC1323 RTT calculation. Normalize
1622 * timestamp if syncookies were used when this connection
1625 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1626 to.to_tsecr -= tp->ts_offset;
1627 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1629 else if (tp->t_rxtshift == 1 &&
1630 tp->t_flags & TF_PREVVALID &&
1631 tp->t_badrxtwin != 0 &&
1632 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
1633 cc_cong_signal(tp, th, CC_RTO_ERR);
1636 * Process options only when we get SYN/ACK back. The SYN case
1637 * for incoming connections is handled in tcp_syncache.
1638 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1639 * or <SYN,ACK>) segment itself is never scaled.
1640 * XXX this is traditional behavior, may need to be cleaned up.
1642 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1643 /* Handle parallel SYN for ECN */
1644 tcp_ecn_input_parallel_syn(tp, thflags, iptos);
1645 if ((to.to_flags & TOF_SCALE) &&
1646 (tp->t_flags & TF_REQ_SCALE) &&
1647 !(tp->t_flags & TF_NOOPT)) {
1648 tp->t_flags |= TF_RCVD_SCALE;
1649 tp->snd_scale = to.to_wscale;
1651 tp->t_flags &= ~TF_REQ_SCALE;
1653 * Initial send window. It will be updated with
1654 * the next incoming segment to the scaled value.
1656 tp->snd_wnd = th->th_win;
1657 if ((to.to_flags & TOF_TS) &&
1658 (tp->t_flags & TF_REQ_TSTMP) &&
1659 !(tp->t_flags & TF_NOOPT)) {
1660 tp->t_flags |= TF_RCVD_TSTMP;
1661 tp->ts_recent = to.to_tsval;
1662 tp->ts_recent_age = tcp_ts_getticks();
1664 tp->t_flags &= ~TF_REQ_TSTMP;
1665 if (to.to_flags & TOF_MSS)
1666 tcp_mss(tp, to.to_mss);
1667 if ((tp->t_flags & TF_SACK_PERMIT) &&
1668 (!(to.to_flags & TOF_SACKPERM) ||
1669 (tp->t_flags & TF_NOOPT)))
1670 tp->t_flags &= ~TF_SACK_PERMIT;
1671 if (IS_FASTOPEN(tp->t_flags)) {
1672 if ((to.to_flags & TOF_FASTOPEN) &&
1673 !(tp->t_flags & TF_NOOPT)) {
1676 if (to.to_flags & TOF_MSS)
1679 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1683 tcp_fastopen_update_cache(tp, mss,
1684 to.to_tfo_len, to.to_tfo_cookie);
1686 tcp_fastopen_disable_path(tp);
1691 * If timestamps were negotiated during SYN/ACK and a
1692 * segment without a timestamp is received, silently drop
1693 * the segment, unless it is a RST segment or missing timestamps are
1695 * See section 3.2 of RFC 7323.
1697 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1698 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1699 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1700 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1701 "segment processed normally\n",
1706 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1707 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1708 "segment silently dropped\n", s, __func__);
1715 * If timestamps were not negotiated during SYN/ACK and a
1716 * segment with a timestamp is received, ignore the
1717 * timestamp and process the packet normally.
1718 * See section 3.2 of RFC 7323.
1720 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1721 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1722 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1723 "segment processed normally\n", s, __func__);
1729 * Header prediction: check for the two common cases
1730 * of a uni-directional data xfer. If the packet has
1731 * no control flags, is in-sequence, the window didn't
1732 * change and we're not retransmitting, it's a
1733 * candidate. If the length is zero and the ack moved
1734 * forward, we're the sender side of the xfer. Just
1735 * free the data acked & wake any higher level process
1736 * that was blocked waiting for space. If the length
1737 * is non-zero and the ack didn't move, we're the
1738 * receiver side. If we're getting packets in-order
1739 * (the reassembly queue is empty), add the data to
1740 * the socket buffer and note that we need a delayed ack.
1741 * Make sure that the hidden state-flags are also off.
1742 * Since we check for TCPS_ESTABLISHED first, it can only
1745 if (tp->t_state == TCPS_ESTABLISHED &&
1746 th->th_seq == tp->rcv_nxt &&
1747 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1748 tp->snd_nxt == tp->snd_max &&
1749 tiwin && tiwin == tp->snd_wnd &&
1750 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1752 ((to.to_flags & TOF_TS) == 0 ||
1753 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1755 * If last ACK falls within this segment's sequence numbers,
1756 * record the timestamp.
1757 * NOTE that the test is modified according to the latest
1758 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1760 if ((to.to_flags & TOF_TS) != 0 &&
1761 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1762 tp->ts_recent_age = tcp_ts_getticks();
1763 tp->ts_recent = to.to_tsval;
1767 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1768 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1769 !IN_RECOVERY(tp->t_flags) &&
1770 (to.to_flags & TOF_SACK) == 0 &&
1771 TAILQ_EMPTY(&tp->snd_holes)) {
1773 * This is a pure ack for outstanding data.
1775 TCPSTAT_INC(tcps_predack);
1778 * "bad retransmit" recovery without timestamps.
1780 if ((to.to_flags & TOF_TS) == 0 &&
1781 tp->t_rxtshift == 1 &&
1782 tp->t_flags & TF_PREVVALID &&
1783 tp->t_badrxtwin != 0 &&
1784 TSTMP_LT(ticks, tp->t_badrxtwin)) {
1785 cc_cong_signal(tp, th, CC_RTO_ERR);
1789 * Recalculate the transmit timer / rtt.
1791 * Some boxes send broken timestamp replies
1792 * during the SYN+ACK phase, ignore
1793 * timestamps of 0 or we could calculate a
1794 * huge RTT and blow up the retransmit timer.
1796 if ((to.to_flags & TOF_TS) != 0 &&
1800 t = tcp_ts_getticks() - to.to_tsecr;
1801 if (!tp->t_rttlow || tp->t_rttlow > t)
1804 TCP_TS_TO_TICKS(t) + 1);
1805 } else if (tp->t_rtttime &&
1806 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1807 if (!tp->t_rttlow ||
1808 tp->t_rttlow > ticks - tp->t_rtttime)
1809 tp->t_rttlow = ticks - tp->t_rtttime;
1811 ticks - tp->t_rtttime);
1813 acked = BYTES_THIS_ACK(tp, th);
1816 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1817 hhook_run_tcp_est_in(tp, th, &to);
1820 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1821 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1822 sbdrop(&so->so_snd, acked);
1823 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1824 SEQ_LEQ(th->th_ack, tp->snd_recover))
1825 tp->snd_recover = th->th_ack - 1;
1828 * Let the congestion control algorithm update
1829 * congestion control related information. This
1830 * typically means increasing the congestion
1833 cc_ack_received(tp, th, nsegs, CC_ACK);
1835 tp->snd_una = th->th_ack;
1837 * Pull snd_wl2 up to prevent seq wrap relative
1840 tp->snd_wl2 = th->th_ack;
1845 * If all outstanding data are acked, stop
1846 * retransmit timer, otherwise restart timer
1847 * using current (possibly backed-off) value.
1848 * If process is waiting for space,
1849 * wakeup/selwakeup/signal. If data
1850 * are ready to send, let tcp_output
1851 * decide between more output or persist.
1854 if (so->so_options & SO_DEBUG)
1855 tcp_trace(TA_INPUT, ostate, tp,
1856 (void *)tcp_saveipgen,
1859 TCP_PROBE3(debug__input, tp, th, m);
1860 if (tp->snd_una == tp->snd_max)
1861 tcp_timer_activate(tp, TT_REXMT, 0);
1862 else if (!tcp_timer_active(tp, TT_PERSIST))
1863 tcp_timer_activate(tp, TT_REXMT,
1866 if (sbavail(&so->so_snd))
1867 (void) tcp_output(tp);
1870 } else if (th->th_ack == tp->snd_una &&
1871 tlen <= sbspace(&so->so_rcv)) {
1872 int newsize = 0; /* automatic sockbuf scaling */
1875 * This is a pure, in-sequence data packet with
1876 * nothing on the reassembly queue and we have enough
1877 * buffer space to take it.
1879 /* Clean receiver SACK report if present */
1880 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1881 tcp_clean_sackreport(tp);
1882 TCPSTAT_INC(tcps_preddat);
1883 tp->rcv_nxt += tlen;
1885 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1886 (tp->t_fbyte_in == 0)) {
1887 tp->t_fbyte_in = ticks;
1888 if (tp->t_fbyte_in == 0)
1890 if (tp->t_fbyte_out && tp->t_fbyte_in)
1891 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1894 * Pull snd_wl1 up to prevent seq wrap relative to
1897 tp->snd_wl1 = th->th_seq;
1899 * Pull rcv_up up to prevent seq wrap relative to
1902 tp->rcv_up = tp->rcv_nxt;
1903 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1904 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1906 if (so->so_options & SO_DEBUG)
1907 tcp_trace(TA_INPUT, ostate, tp,
1908 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1910 TCP_PROBE3(debug__input, tp, th, m);
1912 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1914 /* Add data to socket buffer. */
1915 SOCKBUF_LOCK(&so->so_rcv);
1916 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1920 * Set new socket buffer size.
1921 * Give up when limit is reached.
1924 if (!sbreserve_locked(&so->so_rcv,
1926 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1927 m_adj(m, drop_hdrlen); /* delayed header drop */
1928 sbappendstream_locked(&so->so_rcv, m, 0);
1930 /* NB: sorwakeup_locked() does an implicit unlock. */
1931 sorwakeup_locked(so);
1932 if (DELAY_ACK(tp, tlen)) {
1933 tp->t_flags |= TF_DELACK;
1935 tp->t_flags |= TF_ACKNOW;
1943 * Calculate amount of space in receive window,
1944 * and then do TCP input processing.
1945 * Receive window is amount of space in rcv queue,
1946 * but not less than advertised window.
1948 win = sbspace(&so->so_rcv);
1951 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1953 switch (tp->t_state) {
1955 * If the state is SYN_RECEIVED:
1956 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1958 case TCPS_SYN_RECEIVED:
1959 if ((thflags & TH_ACK) &&
1960 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1961 SEQ_GT(th->th_ack, tp->snd_max))) {
1962 rstreason = BANDLIM_RST_OPENPORT;
1965 if (IS_FASTOPEN(tp->t_flags)) {
1967 * When a TFO connection is in SYN_RECEIVED, the
1968 * only valid packets are the initial SYN, a
1969 * retransmit/copy of the initial SYN (possibly with
1970 * a subset of the original data), a valid ACK, a
1973 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1974 rstreason = BANDLIM_RST_OPENPORT;
1976 } else if (thflags & TH_SYN) {
1977 /* non-initial SYN is ignored */
1978 if ((tcp_timer_active(tp, TT_DELACK) ||
1979 tcp_timer_active(tp, TT_REXMT)))
1981 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1988 * If the state is SYN_SENT:
1989 * if seg contains a RST with valid ACK (SEQ.ACK has already
1990 * been verified), then drop the connection.
1991 * if seg contains a RST without an ACK, drop the seg.
1992 * if seg does not contain SYN, then drop the seg.
1993 * Otherwise this is an acceptable SYN segment
1994 * initialize tp->rcv_nxt and tp->irs
1995 * if seg contains ack then advance tp->snd_una
1996 * if seg contains an ECE and ECN support is enabled, the stream
1998 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1999 * arrange for segment to be acked (eventually)
2000 * continue processing rest of data/controls, beginning with URG
2003 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2004 TCP_PROBE5(connect__refused, NULL, tp,
2006 tp = tcp_drop(tp, ECONNREFUSED);
2008 if (thflags & TH_RST)
2010 if (!(thflags & TH_SYN))
2013 tp->irs = th->th_seq;
2015 if (thflags & TH_ACK) {
2016 int tfo_partial_ack = 0;
2018 TCPSTAT_INC(tcps_connects);
2021 mac_socketpeer_set_from_mbuf(m, so);
2023 /* Do window scaling on this connection? */
2024 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2025 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2026 tp->rcv_scale = tp->request_r_scale;
2028 tp->rcv_adv += min(tp->rcv_wnd,
2029 TCP_MAXWIN << tp->rcv_scale);
2030 tp->snd_una++; /* SYN is acked */
2032 * If not all the data that was sent in the TFO SYN
2033 * has been acked, resend the remainder right away.
2035 if (IS_FASTOPEN(tp->t_flags) &&
2036 (tp->snd_una != tp->snd_max)) {
2037 tp->snd_nxt = th->th_ack;
2038 tfo_partial_ack = 1;
2041 * If there's data, delay ACK; if there's also a FIN
2042 * ACKNOW will be turned on later.
2044 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2045 tcp_timer_activate(tp, TT_DELACK,
2048 tp->t_flags |= TF_ACKNOW;
2050 tcp_ecn_input_syn_sent(tp, thflags, iptos);
2053 * Received <SYN,ACK> in SYN_SENT[*] state.
2055 * SYN_SENT --> ESTABLISHED
2056 * SYN_SENT* --> FIN_WAIT_1
2058 tp->t_starttime = ticks;
2059 if (tp->t_flags & TF_NEEDFIN) {
2060 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2061 tp->t_flags &= ~TF_NEEDFIN;
2064 tcp_state_change(tp, TCPS_ESTABLISHED);
2065 TCP_PROBE5(connect__established, NULL, tp,
2068 tcp_timer_activate(tp, TT_KEEP,
2073 * Received initial SYN in SYN-SENT[*] state =>
2074 * simultaneous open.
2075 * If it succeeds, connection is * half-synchronized.
2076 * Otherwise, do 3-way handshake:
2077 * SYN-SENT -> SYN-RECEIVED
2078 * SYN-SENT* -> SYN-RECEIVED*
2080 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2081 tcp_timer_activate(tp, TT_REXMT, 0);
2082 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2085 INP_WLOCK_ASSERT(tp->t_inpcb);
2088 * Advance th->th_seq to correspond to first data byte.
2089 * If data, trim to stay within window,
2090 * dropping FIN if necessary.
2093 if (tlen > tp->rcv_wnd) {
2094 todrop = tlen - tp->rcv_wnd;
2098 TCPSTAT_INC(tcps_rcvpackafterwin);
2099 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2101 tp->snd_wl1 = th->th_seq - 1;
2102 tp->rcv_up = th->th_seq;
2104 * Client side of transaction: already sent SYN and data.
2105 * If the remote host used T/TCP to validate the SYN,
2106 * our data will be ACK'd; if so, enter normal data segment
2107 * processing in the middle of step 5, ack processing.
2108 * Otherwise, goto step 6.
2110 if (thflags & TH_ACK)
2116 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2117 * do normal processing.
2119 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2123 break; /* continue normal processing */
2127 * States other than LISTEN or SYN_SENT.
2128 * First check the RST flag and sequence number since reset segments
2129 * are exempt from the timestamp and connection count tests. This
2130 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2131 * below which allowed reset segments in half the sequence space
2132 * to fall though and be processed (which gives forged reset
2133 * segments with a random sequence number a 50 percent chance of
2134 * killing a connection).
2135 * Then check timestamp, if present.
2136 * Then check the connection count, if present.
2137 * Then check that at least some bytes of segment are within
2138 * receive window. If segment begins before rcv_nxt,
2139 * drop leading data (and SYN); if nothing left, just ack.
2141 if (thflags & TH_RST) {
2143 * RFC5961 Section 3.2
2145 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2146 * - If RST is in window, we send challenge ACK.
2148 * Note: to take into account delayed ACKs, we should
2149 * test against last_ack_sent instead of rcv_nxt.
2150 * Note 2: we handle special case of closed window, not
2151 * covered by the RFC.
2153 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2154 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2155 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2156 KASSERT(tp->t_state != TCPS_SYN_SENT,
2157 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2160 if (V_tcp_insecure_rst ||
2161 tp->last_ack_sent == th->th_seq) {
2162 TCPSTAT_INC(tcps_drops);
2163 /* Drop the connection. */
2164 switch (tp->t_state) {
2165 case TCPS_SYN_RECEIVED:
2166 so->so_error = ECONNREFUSED;
2168 case TCPS_ESTABLISHED:
2169 case TCPS_FIN_WAIT_1:
2170 case TCPS_FIN_WAIT_2:
2171 case TCPS_CLOSE_WAIT:
2174 so->so_error = ECONNRESET;
2181 TCPSTAT_INC(tcps_badrst);
2182 /* Send challenge ACK. */
2183 tcp_respond(tp, mtod(m, void *), th, m,
2184 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2185 tp->last_ack_sent = tp->rcv_nxt;
2193 * RFC5961 Section 4.2
2194 * Send challenge ACK for any SYN in synchronized state.
2196 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2197 tp->t_state != TCPS_SYN_RECEIVED) {
2198 TCPSTAT_INC(tcps_badsyn);
2199 if (V_tcp_insecure_syn &&
2200 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2201 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2202 tp = tcp_drop(tp, ECONNRESET);
2203 rstreason = BANDLIM_UNLIMITED;
2205 /* Send challenge ACK. */
2206 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2207 tp->snd_nxt, TH_ACK);
2208 tp->last_ack_sent = tp->rcv_nxt;
2215 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2216 * and it's less than ts_recent, drop it.
2218 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2219 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2220 /* Check to see if ts_recent is over 24 days old. */
2221 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2223 * Invalidate ts_recent. If this segment updates
2224 * ts_recent, the age will be reset later and ts_recent
2225 * will get a valid value. If it does not, setting
2226 * ts_recent to zero will at least satisfy the
2227 * requirement that zero be placed in the timestamp
2228 * echo reply when ts_recent isn't valid. The
2229 * age isn't reset until we get a valid ts_recent
2230 * because we don't want out-of-order segments to be
2231 * dropped when ts_recent is old.
2235 TCPSTAT_INC(tcps_rcvduppack);
2236 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2237 TCPSTAT_INC(tcps_pawsdrop);
2245 * In the SYN-RECEIVED state, validate that the packet belongs to
2246 * this connection before trimming the data to fit the receive
2247 * window. Check the sequence number versus IRS since we know
2248 * the sequence numbers haven't wrapped. This is a partial fix
2249 * for the "LAND" DoS attack.
2251 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2252 rstreason = BANDLIM_RST_OPENPORT;
2256 todrop = tp->rcv_nxt - th->th_seq;
2258 if (thflags & TH_SYN) {
2268 * Following if statement from Stevens, vol. 2, p. 960.
2271 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2273 * Any valid FIN must be to the left of the window.
2274 * At this point the FIN must be a duplicate or out
2275 * of sequence; drop it.
2280 * Send an ACK to resynchronize and drop any data.
2281 * But keep on processing for RST or ACK.
2283 tp->t_flags |= TF_ACKNOW;
2285 TCPSTAT_INC(tcps_rcvduppack);
2286 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2288 TCPSTAT_INC(tcps_rcvpartduppack);
2289 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2292 * DSACK - add SACK block for dropped range
2294 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2295 tcp_update_sack_list(tp, th->th_seq,
2296 th->th_seq + todrop);
2298 * ACK now, as the next in-sequence segment
2299 * will clear the DSACK block again
2301 tp->t_flags |= TF_ACKNOW;
2303 drop_hdrlen += todrop; /* drop from the top afterwards */
2304 th->th_seq += todrop;
2306 if (th->th_urp > todrop)
2307 th->th_urp -= todrop;
2315 * If new data are received on a connection after the
2316 * user processes are gone, then RST the other end.
2318 if ((so->so_state & SS_NOFDREF) &&
2319 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2320 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2321 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2322 "after socket was closed, "
2323 "sending RST and removing tcpcb\n",
2324 s, __func__, tcpstates[tp->t_state], tlen);
2328 TCPSTAT_INC(tcps_rcvafterclose);
2329 rstreason = BANDLIM_UNLIMITED;
2334 * If segment ends after window, drop trailing data
2335 * (and PUSH and FIN); if nothing left, just ACK.
2337 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2339 TCPSTAT_INC(tcps_rcvpackafterwin);
2340 if (todrop >= tlen) {
2341 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2343 * If window is closed can only take segments at
2344 * window edge, and have to drop data and PUSH from
2345 * incoming segments. Continue processing, but
2346 * remember to ack. Otherwise, drop segment
2349 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2350 tp->t_flags |= TF_ACKNOW;
2351 TCPSTAT_INC(tcps_rcvwinprobe);
2355 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2358 thflags &= ~(TH_PUSH|TH_FIN);
2362 * If last ACK falls within this segment's sequence numbers,
2363 * record its timestamp.
2365 * 1) That the test incorporates suggestions from the latest
2366 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2367 * 2) That updating only on newer timestamps interferes with
2368 * our earlier PAWS tests, so this check should be solely
2369 * predicated on the sequence space of this segment.
2370 * 3) That we modify the segment boundary check to be
2371 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2372 * instead of RFC1323's
2373 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2374 * This modified check allows us to overcome RFC1323's
2375 * limitations as described in Stevens TCP/IP Illustrated
2376 * Vol. 2 p.869. In such cases, we can still calculate the
2377 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2379 if ((to.to_flags & TOF_TS) != 0 &&
2380 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2381 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2382 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2383 tp->ts_recent_age = tcp_ts_getticks();
2384 tp->ts_recent = to.to_tsval;
2388 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2389 * flag is on (half-synchronized state), then queue data for
2390 * later processing; else drop segment and return.
2392 if ((thflags & TH_ACK) == 0) {
2393 if (tp->t_state == TCPS_SYN_RECEIVED ||
2394 (tp->t_flags & TF_NEEDSYN)) {
2395 if (tp->t_state == TCPS_SYN_RECEIVED &&
2396 IS_FASTOPEN(tp->t_flags)) {
2397 tp->snd_wnd = tiwin;
2401 } else if (tp->t_flags & TF_ACKNOW)
2410 switch (tp->t_state) {
2412 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2413 * ESTABLISHED state and continue processing.
2414 * The ACK was checked above.
2416 case TCPS_SYN_RECEIVED:
2418 TCPSTAT_INC(tcps_connects);
2420 /* Do window scaling? */
2421 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2422 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2423 tp->rcv_scale = tp->request_r_scale;
2425 tp->snd_wnd = tiwin;
2428 * SYN-RECEIVED -> ESTABLISHED
2429 * SYN-RECEIVED* -> FIN-WAIT-1
2431 tp->t_starttime = ticks;
2432 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2433 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2434 tp->t_tfo_pending = NULL;
2436 if (tp->t_flags & TF_NEEDFIN) {
2437 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2438 tp->t_flags &= ~TF_NEEDFIN;
2440 tcp_state_change(tp, TCPS_ESTABLISHED);
2441 TCP_PROBE5(accept__established, NULL, tp,
2444 * TFO connections call cc_conn_init() during SYN
2445 * processing. Calling it again here for such
2446 * connections is not harmless as it would undo the
2447 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2450 if (!IS_FASTOPEN(tp->t_flags))
2452 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2455 * Account for the ACK of our SYN prior to
2456 * regular ACK processing below, except for
2457 * simultaneous SYN, which is handled later.
2459 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2462 * If segment contains data or ACK, will call tcp_reass()
2463 * later; if not, do so now to pass queued data to user.
2465 if (tlen == 0 && (thflags & TH_FIN) == 0) {
2466 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2468 tcp_handle_wakeup(tp, so);
2470 tp->snd_wl1 = th->th_seq - 1;
2474 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2475 * ACKs. If the ack is in the range
2476 * tp->snd_una < th->th_ack <= tp->snd_max
2477 * then advance tp->snd_una to th->th_ack and drop
2478 * data from the retransmission queue. If this ACK reflects
2479 * more up to date window information we update our window information.
2481 case TCPS_ESTABLISHED:
2482 case TCPS_FIN_WAIT_1:
2483 case TCPS_FIN_WAIT_2:
2484 case TCPS_CLOSE_WAIT:
2487 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2488 TCPSTAT_INC(tcps_rcvacktoomuch);
2491 if ((tp->t_flags & TF_SACK_PERMIT) &&
2492 ((to.to_flags & TOF_SACK) ||
2493 !TAILQ_EMPTY(&tp->snd_holes))) {
2494 if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) &&
2495 (tp->t_flags & TF_LRD)) {
2496 tcp_sack_lost_retransmission(tp, th);
2500 * Reset the value so that previous (valid) value
2501 * from the last ack with SACK doesn't get used.
2503 tp->sackhint.sacked_bytes = 0;
2506 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2507 hhook_run_tcp_est_in(tp, th, &to);
2510 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2511 maxseg = tcp_maxseg(tp);
2513 (tiwin == tp->snd_wnd ||
2514 (tp->t_flags & TF_SACK_PERMIT))) {
2516 * If this is the first time we've seen a
2517 * FIN from the remote, this is not a
2518 * duplicate and it needs to be processed
2519 * normally. This happens during a
2520 * simultaneous close.
2522 if ((thflags & TH_FIN) &&
2523 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2527 TCPSTAT_INC(tcps_rcvdupack);
2529 * If we have outstanding data (other than
2530 * a window probe), this is a completely
2531 * duplicate ack (ie, window info didn't
2532 * change and FIN isn't set),
2533 * the ack is the biggest we've
2534 * seen and we've seen exactly our rexmt
2535 * threshold of them, assume a packet
2536 * has been dropped and retransmit it.
2537 * Kludge snd_nxt & the congestion
2538 * window so we send only this one
2541 * We know we're losing at the current
2542 * window size so do congestion avoidance
2543 * (set ssthresh to half the current window
2544 * and pull our congestion window back to
2545 * the new ssthresh).
2547 * Dup acks mean that packets have left the
2548 * network (they're now cached at the receiver)
2549 * so bump cwnd by the amount in the receiver
2550 * to keep a constant cwnd packets in the
2553 * When using TCP ECN, notify the peer that
2554 * we reduced the cwnd.
2557 * Following 2 kinds of acks should not affect
2560 * 2) Acks with SACK but without any new SACK
2561 * information in them. These could result from
2562 * any anomaly in the network like a switch
2563 * duplicating packets or a possible DoS attack.
2565 if (th->th_ack != tp->snd_una ||
2566 ((tp->t_flags & TF_SACK_PERMIT) &&
2567 (to.to_flags & TOF_SACK) &&
2570 else if (!tcp_timer_active(tp, TT_REXMT))
2572 else if (++tp->t_dupacks > tcprexmtthresh ||
2573 IN_FASTRECOVERY(tp->t_flags)) {
2574 cc_ack_received(tp, th, nsegs,
2577 IN_FASTRECOVERY(tp->t_flags)) {
2578 tcp_do_prr_ack(tp, th, &to);
2579 } else if ((tp->t_flags & TF_SACK_PERMIT) &&
2580 (to.to_flags & TOF_SACK) &&
2581 IN_FASTRECOVERY(tp->t_flags)) {
2585 * Compute the amount of data in flight first.
2586 * We can inject new data into the pipe iff
2587 * we have less than 1/2 the original window's
2588 * worth of data in flight.
2590 if (V_tcp_do_newsack)
2591 awnd = tcp_compute_pipe(tp);
2593 awnd = (tp->snd_nxt - tp->snd_fack) +
2594 tp->sackhint.sack_bytes_rexmit;
2596 if (awnd < tp->snd_ssthresh) {
2597 tp->snd_cwnd += maxseg;
2598 if (tp->snd_cwnd > tp->snd_ssthresh)
2599 tp->snd_cwnd = tp->snd_ssthresh;
2602 tp->snd_cwnd += maxseg;
2603 (void) tcp_output(tp);
2605 } else if (tp->t_dupacks == tcprexmtthresh ||
2606 (tp->t_flags & TF_SACK_PERMIT &&
2608 tp->sackhint.sacked_bytes >
2609 (tcprexmtthresh - 1) * maxseg)) {
2612 * Above is the RFC6675 trigger condition of
2613 * more than (dupthresh-1)*maxseg sacked data.
2614 * If the count of holes in the
2615 * scoreboard is >= dupthresh, we could
2616 * also enter loss recovery, but don't
2617 * have that value readily available.
2619 tp->t_dupacks = tcprexmtthresh;
2620 tcp_seq onxt = tp->snd_nxt;
2623 * If we're doing sack, or prr, check
2624 * to see if we're already in sack
2625 * recovery. If we're not doing sack,
2626 * check to see if we're in newreno
2630 (tp->t_flags & TF_SACK_PERMIT)) {
2631 if (IN_FASTRECOVERY(tp->t_flags)) {
2636 if (SEQ_LEQ(th->th_ack,
2642 /* Congestion signal before ack. */
2643 cc_cong_signal(tp, th, CC_NDUPACK);
2644 cc_ack_received(tp, th, nsegs,
2646 tcp_timer_activate(tp, TT_REXMT, 0);
2650 * snd_ssthresh is already updated by
2653 if ((tp->t_flags & TF_SACK_PERMIT) &&
2654 (to.to_flags & TOF_SACK)) {
2655 tp->sackhint.prr_delivered =
2656 tp->sackhint.sacked_bytes;
2658 tp->sackhint.prr_delivered =
2659 imin(tp->snd_max - tp->snd_una,
2660 imin(INT_MAX / 65536,
2661 tp->t_dupacks) * maxseg);
2663 tp->sackhint.recover_fs = max(1,
2664 tp->snd_nxt - tp->snd_una);
2666 if ((tp->t_flags & TF_SACK_PERMIT) &&
2667 (to.to_flags & TOF_SACK)) {
2669 tcps_sack_recovery_episode);
2670 tp->snd_recover = tp->snd_nxt;
2671 tp->snd_cwnd = maxseg;
2672 (void) tcp_output(tp);
2673 if (SEQ_GT(th->th_ack, tp->snd_una))
2674 goto resume_partialack;
2677 tp->snd_nxt = th->th_ack;
2678 tp->snd_cwnd = maxseg;
2679 (void) tcp_output(tp);
2680 KASSERT(tp->snd_limited <= 2,
2681 ("%s: tp->snd_limited too big",
2683 tp->snd_cwnd = tp->snd_ssthresh +
2685 (tp->t_dupacks - tp->snd_limited);
2686 if (SEQ_GT(onxt, tp->snd_nxt))
2689 } else if (V_tcp_do_rfc3042) {
2691 * Process first and second duplicate
2692 * ACKs. Each indicates a segment
2693 * leaving the network, creating room
2694 * for more. Make sure we can send a
2695 * packet on reception of each duplicate
2696 * ACK by increasing snd_cwnd by one
2697 * segment. Restore the original
2698 * snd_cwnd after packet transmission.
2700 cc_ack_received(tp, th, nsegs,
2702 uint32_t oldcwnd = tp->snd_cwnd;
2703 tcp_seq oldsndmax = tp->snd_max;
2707 KASSERT(tp->t_dupacks == 1 ||
2709 ("%s: dupacks not 1 or 2",
2711 if (tp->t_dupacks == 1)
2712 tp->snd_limited = 0;
2714 (tp->snd_nxt - tp->snd_una) +
2715 (tp->t_dupacks - tp->snd_limited) *
2718 * Only call tcp_output when there
2719 * is new data available to be sent.
2720 * Otherwise we would send pure ACKs.
2722 SOCKBUF_LOCK(&so->so_snd);
2723 avail = sbavail(&so->so_snd) -
2724 (tp->snd_nxt - tp->snd_una);
2725 SOCKBUF_UNLOCK(&so->so_snd);
2727 (void) tcp_output(tp);
2728 sent = tp->snd_max - oldsndmax;
2729 if (sent > maxseg) {
2730 KASSERT((tp->t_dupacks == 2 &&
2731 tp->snd_limited == 0) ||
2732 (sent == maxseg + 1 &&
2733 tp->t_flags & TF_SENTFIN),
2734 ("%s: sent too much",
2736 tp->snd_limited = 2;
2737 } else if (sent > 0)
2739 tp->snd_cwnd = oldcwnd;
2746 * This ack is advancing the left edge, reset the
2751 * If this ack also has new SACK info, increment the
2752 * counter as per rfc6675. The variable
2753 * sack_changed tracks all changes to the SACK
2754 * scoreboard, including when partial ACKs without
2755 * SACK options are received, and clear the scoreboard
2756 * from the left side. Such partial ACKs should not be
2757 * counted as dupacks here.
2759 if ((tp->t_flags & TF_SACK_PERMIT) &&
2760 (to.to_flags & TOF_SACK) &&
2763 /* limit overhead by setting maxseg last */
2764 if (!IN_FASTRECOVERY(tp->t_flags) &&
2765 (tp->sackhint.sacked_bytes >
2766 ((tcprexmtthresh - 1) *
2767 (maxseg = tcp_maxseg(tp))))) {
2768 goto enter_recovery;
2774 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2775 ("%s: th_ack <= snd_una", __func__));
2778 * If the congestion window was inflated to account
2779 * for the other side's cached packets, retract it.
2781 if (IN_FASTRECOVERY(tp->t_flags)) {
2782 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2783 if (tp->t_flags & TF_SACK_PERMIT)
2784 if (V_tcp_do_prr && to.to_flags & TOF_SACK) {
2785 tcp_timer_activate(tp, TT_REXMT, 0);
2787 tcp_do_prr_ack(tp, th, &to);
2788 tp->t_flags |= TF_ACKNOW;
2789 (void) tcp_output(tp);
2791 tcp_sack_partialack(tp, th);
2793 tcp_newreno_partial_ack(tp, th);
2795 cc_post_recovery(tp, th);
2796 } else if (IN_CONGRECOVERY(tp->t_flags)) {
2797 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2799 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th);
2800 tp->snd_fack = th->th_ack;
2801 tcp_do_prr_ack(tp, th, &to);
2802 (void) tcp_output(tp);
2805 cc_post_recovery(tp, th);
2808 * If we reach this point, ACK is not a duplicate,
2809 * i.e., it ACKs something we sent.
2811 if (tp->t_flags & TF_NEEDSYN) {
2813 * T/TCP: Connection was half-synchronized, and our
2814 * SYN has been ACK'd (so connection is now fully
2815 * synchronized). Go to non-starred state,
2816 * increment snd_una for ACK of SYN, and check if
2817 * we can do window scaling.
2819 tp->t_flags &= ~TF_NEEDSYN;
2821 /* Do window scaling? */
2822 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2823 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2824 tp->rcv_scale = tp->request_r_scale;
2825 /* Send window already scaled. */
2830 INP_WLOCK_ASSERT(tp->t_inpcb);
2833 * Adjust for the SYN bit in sequence space,
2834 * but don't account for it in cwnd calculations.
2835 * This is for the SYN_RECEIVED, non-simultaneous
2836 * SYN case. SYN_SENT and simultaneous SYN are
2837 * treated elsewhere.
2841 acked = BYTES_THIS_ACK(tp, th);
2842 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2843 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2844 tp->snd_una, th->th_ack, tp, m));
2845 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2846 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2849 * If we just performed our first retransmit, and the ACK
2850 * arrives within our recovery window, then it was a mistake
2851 * to do the retransmit in the first place. Recover our
2852 * original cwnd and ssthresh, and proceed to transmit where
2855 if (tp->t_rxtshift == 1 &&
2856 tp->t_flags & TF_PREVVALID &&
2857 tp->t_badrxtwin != 0 &&
2858 to.to_flags & TOF_TS &&
2860 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
2861 cc_cong_signal(tp, th, CC_RTO_ERR);
2864 * If we have a timestamp reply, update smoothed
2865 * round trip time. If no timestamp is present but
2866 * transmit timer is running and timed sequence
2867 * number was acked, update smoothed round trip time.
2868 * Since we now have an rtt measurement, cancel the
2869 * timer backoff (cf., Phil Karn's retransmit alg.).
2870 * Recompute the initial retransmit timer.
2872 * Some boxes send broken timestamp replies
2873 * during the SYN+ACK phase, ignore
2874 * timestamps of 0 or we could calculate a
2875 * huge RTT and blow up the retransmit timer.
2877 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2880 t = tcp_ts_getticks() - to.to_tsecr;
2881 if (!tp->t_rttlow || tp->t_rttlow > t)
2883 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2884 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2885 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2886 tp->t_rttlow = ticks - tp->t_rtttime;
2887 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2891 * If all outstanding data is acked, stop retransmit
2892 * timer and remember to restart (more output or persist).
2893 * If there is more data to be acked, restart retransmit
2894 * timer, using current (possibly backed-off) value.
2896 if (th->th_ack == tp->snd_max) {
2897 tcp_timer_activate(tp, TT_REXMT, 0);
2899 } else if (!tcp_timer_active(tp, TT_PERSIST))
2900 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2903 * If no data (only SYN) was ACK'd,
2904 * skip rest of ACK processing.
2910 * Let the congestion control algorithm update congestion
2911 * control related information. This typically means increasing
2912 * the congestion window.
2914 cc_ack_received(tp, th, nsegs, CC_ACK);
2916 SOCKBUF_LOCK(&so->so_snd);
2917 if (acked > sbavail(&so->so_snd)) {
2918 if (tp->snd_wnd >= sbavail(&so->so_snd))
2919 tp->snd_wnd -= sbavail(&so->so_snd);
2922 mfree = sbcut_locked(&so->so_snd,
2923 (int)sbavail(&so->so_snd));
2926 mfree = sbcut_locked(&so->so_snd, acked);
2927 if (tp->snd_wnd >= (uint32_t) acked)
2928 tp->snd_wnd -= acked;
2933 /* NB: sowwakeup_locked() does an implicit unlock. */
2934 sowwakeup_locked(so);
2936 /* Detect una wraparound. */
2937 if (!IN_RECOVERY(tp->t_flags) &&
2938 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2939 SEQ_LEQ(th->th_ack, tp->snd_recover))
2940 tp->snd_recover = th->th_ack - 1;
2941 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2942 if (IN_RECOVERY(tp->t_flags) &&
2943 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2944 EXIT_RECOVERY(tp->t_flags);
2946 tp->snd_una = th->th_ack;
2947 if (tp->t_flags & TF_SACK_PERMIT) {
2948 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2949 tp->snd_recover = tp->snd_una;
2951 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2952 tp->snd_nxt = tp->snd_una;
2954 switch (tp->t_state) {
2956 * In FIN_WAIT_1 STATE in addition to the processing
2957 * for the ESTABLISHED state if our FIN is now acknowledged
2958 * then enter FIN_WAIT_2.
2960 case TCPS_FIN_WAIT_1:
2961 if (ourfinisacked) {
2963 * If we can't receive any more
2964 * data, then closing user can proceed.
2965 * Starting the timer is contrary to the
2966 * specification, but if we don't get a FIN
2967 * we'll hang forever.
2970 * we should release the tp also, and use a
2973 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2974 soisdisconnected(so);
2975 tcp_timer_activate(tp, TT_2MSL,
2976 (tcp_fast_finwait2_recycle ?
2977 tcp_finwait2_timeout :
2980 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2985 * In CLOSING STATE in addition to the processing for
2986 * the ESTABLISHED state if the ACK acknowledges our FIN
2987 * then enter the TIME-WAIT state, otherwise ignore
2991 if (ourfinisacked) {
2999 * In LAST_ACK, we may still be waiting for data to drain
3000 * and/or to be acked, as well as for the ack of our FIN.
3001 * If our FIN is now acknowledged, delete the TCB,
3002 * enter the closed state and return.
3005 if (ourfinisacked) {
3014 INP_WLOCK_ASSERT(tp->t_inpcb);
3017 * Update window information.
3018 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3020 if ((thflags & TH_ACK) &&
3021 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3022 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3023 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3024 /* keep track of pure window updates */
3026 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3027 TCPSTAT_INC(tcps_rcvwinupd);
3028 tp->snd_wnd = tiwin;
3029 tp->snd_wl1 = th->th_seq;
3030 tp->snd_wl2 = th->th_ack;
3031 if (tp->snd_wnd > tp->max_sndwnd)
3032 tp->max_sndwnd = tp->snd_wnd;
3037 * Process segments with URG.
3039 if ((thflags & TH_URG) && th->th_urp &&
3040 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3042 * This is a kludge, but if we receive and accept
3043 * random urgent pointers, we'll crash in
3044 * soreceive. It's hard to imagine someone
3045 * actually wanting to send this much urgent data.
3047 SOCKBUF_LOCK(&so->so_rcv);
3048 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3049 th->th_urp = 0; /* XXX */
3050 thflags &= ~TH_URG; /* XXX */
3051 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3052 goto dodata; /* XXX */
3055 * If this segment advances the known urgent pointer,
3056 * then mark the data stream. This should not happen
3057 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3058 * a FIN has been received from the remote side.
3059 * In these states we ignore the URG.
3061 * According to RFC961 (Assigned Protocols),
3062 * the urgent pointer points to the last octet
3063 * of urgent data. We continue, however,
3064 * to consider it to indicate the first octet
3065 * of data past the urgent section as the original
3066 * spec states (in one of two places).
3068 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3069 tp->rcv_up = th->th_seq + th->th_urp;
3070 so->so_oobmark = sbavail(&so->so_rcv) +
3071 (tp->rcv_up - tp->rcv_nxt) - 1;
3072 if (so->so_oobmark == 0)
3073 so->so_rcv.sb_state |= SBS_RCVATMARK;
3075 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3077 SOCKBUF_UNLOCK(&so->so_rcv);
3079 * Remove out of band data so doesn't get presented to user.
3080 * This can happen independent of advancing the URG pointer,
3081 * but if two URG's are pending at once, some out-of-band
3082 * data may creep in... ick.
3084 if (th->th_urp <= (uint32_t)tlen &&
3085 !(so->so_options & SO_OOBINLINE)) {
3086 /* hdr drop is delayed */
3087 tcp_pulloutofband(so, th, m, drop_hdrlen);
3091 * If no out of band data is expected,
3092 * pull receive urgent pointer along
3093 * with the receive window.
3095 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3096 tp->rcv_up = tp->rcv_nxt;
3099 INP_WLOCK_ASSERT(tp->t_inpcb);
3102 * Process the segment text, merging it into the TCP sequencing queue,
3103 * and arranging for acknowledgment of receipt if necessary.
3104 * This process logically involves adjusting tp->rcv_wnd as data
3105 * is presented to the user (this happens in tcp_usrreq.c,
3106 * case PRU_RCVD). If a FIN has already been received on this
3107 * connection then we just ignore the text.
3109 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3110 IS_FASTOPEN(tp->t_flags));
3111 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3112 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3113 tcp_seq save_start = th->th_seq;
3114 tcp_seq save_rnxt = tp->rcv_nxt;
3115 int save_tlen = tlen;
3116 m_adj(m, drop_hdrlen); /* delayed header drop */
3118 * Insert segment which includes th into TCP reassembly queue
3119 * with control block tp. Set thflags to whether reassembly now
3120 * includes a segment with FIN. This handles the common case
3121 * inline (segment is the next to be received on an established
3122 * connection, and the queue is empty), avoiding linkage into
3123 * and removal from the queue and repetition of various
3125 * Set DELACK for segments received in order, but ack
3126 * immediately when segments are out of order (so
3127 * fast retransmit can work).
3129 if (th->th_seq == tp->rcv_nxt &&
3131 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3133 if (DELAY_ACK(tp, tlen) || tfo_syn)
3134 tp->t_flags |= TF_DELACK;
3136 tp->t_flags |= TF_ACKNOW;
3137 tp->rcv_nxt += tlen;
3139 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3140 (tp->t_fbyte_in == 0)) {
3141 tp->t_fbyte_in = ticks;
3142 if (tp->t_fbyte_in == 0)
3144 if (tp->t_fbyte_out && tp->t_fbyte_in)
3145 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3147 thflags = tcp_get_flags(th) & TH_FIN;
3148 TCPSTAT_INC(tcps_rcvpack);
3149 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3150 SOCKBUF_LOCK(&so->so_rcv);
3151 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3154 sbappendstream_locked(&so->so_rcv, m, 0);
3155 tp->t_flags |= TF_WAKESOR;
3158 * XXX: Due to the header drop above "th" is
3159 * theoretically invalid by now. Fortunately
3160 * m_adj() doesn't actually frees any mbufs
3161 * when trimming from the head.
3163 tcp_seq temp = save_start;
3165 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3166 tp->t_flags |= TF_ACKNOW;
3168 if ((tp->t_flags & TF_SACK_PERMIT) &&
3170 TCPS_HAVEESTABLISHED(tp->t_state)) {
3171 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3173 * DSACK actually handled in the fastpath
3176 tcp_update_sack_list(tp, save_start,
3177 save_start + save_tlen);
3178 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3179 if ((tp->rcv_numsacks >= 1) &&
3180 (tp->sackblks[0].end == save_start)) {
3182 * Partial overlap, recorded at todrop
3185 tcp_update_sack_list(tp,
3186 tp->sackblks[0].start,
3187 tp->sackblks[0].end);
3189 tcp_update_dsack_list(tp, save_start,
3190 save_start + save_tlen);
3192 } else if (tlen >= save_tlen) {
3193 /* Update of sackblks. */
3194 tcp_update_dsack_list(tp, save_start,
3195 save_start + save_tlen);
3196 } else if (tlen > 0) {
3197 tcp_update_dsack_list(tp, save_start,
3201 tcp_handle_wakeup(tp, so);
3204 * Note the amount of data that peer has sent into
3205 * our window, in order to estimate the sender's
3209 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3210 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3212 len = so->so_rcv.sb_hiwat;
3220 * If FIN is received ACK the FIN and let the user know
3221 * that the connection is closing.
3223 if (thflags & TH_FIN) {
3224 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3225 /* The socket upcall is handled by socantrcvmore. */
3228 * If connection is half-synchronized
3229 * (ie NEEDSYN flag on) then delay ACK,
3230 * so it may be piggybacked when SYN is sent.
3231 * Otherwise, since we received a FIN then no
3232 * more input can be expected, send ACK now.
3234 if (tp->t_flags & TF_NEEDSYN)
3235 tp->t_flags |= TF_DELACK;
3237 tp->t_flags |= TF_ACKNOW;
3240 switch (tp->t_state) {
3242 * In SYN_RECEIVED and ESTABLISHED STATES
3243 * enter the CLOSE_WAIT state.
3245 case TCPS_SYN_RECEIVED:
3246 tp->t_starttime = ticks;
3248 case TCPS_ESTABLISHED:
3249 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3253 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3254 * enter the CLOSING state.
3256 case TCPS_FIN_WAIT_1:
3257 tcp_state_change(tp, TCPS_CLOSING);
3261 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3262 * starting the time-wait timer, turning off the other
3265 case TCPS_FIN_WAIT_2:
3271 if (so->so_options & SO_DEBUG)
3272 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3275 TCP_PROBE3(debug__input, tp, th, m);
3278 * Return any desired output.
3280 if (needoutput || (tp->t_flags & TF_ACKNOW))
3281 (void) tcp_output(tp);
3284 INP_WLOCK_ASSERT(tp->t_inpcb);
3286 if (tp->t_flags & TF_DELACK) {
3287 tp->t_flags &= ~TF_DELACK;
3288 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3290 INP_WUNLOCK(tp->t_inpcb);
3295 * Generate an ACK dropping incoming segment if it occupies
3296 * sequence space, where the ACK reflects our state.
3298 * We can now skip the test for the RST flag since all
3299 * paths to this code happen after packets containing
3300 * RST have been dropped.
3302 * In the SYN-RECEIVED state, don't send an ACK unless the
3303 * segment we received passes the SYN-RECEIVED ACK test.
3304 * If it fails send a RST. This breaks the loop in the
3305 * "LAND" DoS attack, and also prevents an ACK storm
3306 * between two listening ports that have been sent forged
3307 * SYN segments, each with the source address of the other.
3309 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3310 (SEQ_GT(tp->snd_una, th->th_ack) ||
3311 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3312 rstreason = BANDLIM_RST_OPENPORT;
3316 if (so->so_options & SO_DEBUG)
3317 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3320 TCP_PROBE3(debug__input, tp, th, m);
3321 tp->t_flags |= TF_ACKNOW;
3322 (void) tcp_output(tp);
3323 INP_WUNLOCK(tp->t_inpcb);
3329 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3330 INP_WUNLOCK(tp->t_inpcb);
3332 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3337 * Drop space held by incoming segment and return.
3340 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3341 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3344 TCP_PROBE3(debug__input, tp, th, m);
3346 INP_WUNLOCK(tp->t_inpcb);
3352 * Issue RST and make ACK acceptable to originator of segment.
3353 * The mbuf must still include the original packet header.
3357 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3358 int tlen, int rstreason)
3364 struct ip6_hdr *ip6;
3368 INP_LOCK_ASSERT(tp->t_inpcb);
3371 /* Don't bother if destination was broadcast/multicast. */
3372 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3375 if (mtod(m, struct ip *)->ip_v == 6) {
3376 ip6 = mtod(m, struct ip6_hdr *);
3377 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3378 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3380 /* IPv6 anycast check is done at tcp6_input() */
3383 #if defined(INET) && defined(INET6)
3388 ip = mtod(m, struct ip *);
3389 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3390 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3391 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3392 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3397 /* Perform bandwidth limiting. */
3398 if (badport_bandlim(rstreason) < 0)
3401 /* tcp_respond consumes the mbuf chain. */
3402 if (tcp_get_flags(th) & TH_ACK) {
3403 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3404 th->th_ack, TH_RST);
3406 if (tcp_get_flags(th) & TH_SYN)
3408 if (tcp_get_flags(th) & TH_FIN)
3410 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3411 (tcp_seq)0, TH_RST|TH_ACK);
3419 * Parse TCP options and place in tcpopt.
3422 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3427 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3429 if (opt == TCPOPT_EOL)
3431 if (opt == TCPOPT_NOP)
3437 if (optlen < 2 || optlen > cnt)
3442 if (optlen != TCPOLEN_MAXSEG)
3444 if (!(flags & TO_SYN))
3446 to->to_flags |= TOF_MSS;
3447 bcopy((char *)cp + 2,
3448 (char *)&to->to_mss, sizeof(to->to_mss));
3449 to->to_mss = ntohs(to->to_mss);
3452 if (optlen != TCPOLEN_WINDOW)
3454 if (!(flags & TO_SYN))
3456 to->to_flags |= TOF_SCALE;
3457 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3459 case TCPOPT_TIMESTAMP:
3460 if (optlen != TCPOLEN_TIMESTAMP)
3462 to->to_flags |= TOF_TS;
3463 bcopy((char *)cp + 2,
3464 (char *)&to->to_tsval, sizeof(to->to_tsval));
3465 to->to_tsval = ntohl(to->to_tsval);
3466 bcopy((char *)cp + 6,
3467 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3468 to->to_tsecr = ntohl(to->to_tsecr);
3470 case TCPOPT_SIGNATURE:
3472 * In order to reply to a host which has set the
3473 * TCP_SIGNATURE option in its initial SYN, we have
3474 * to record the fact that the option was observed
3475 * here for the syncache code to perform the correct
3478 if (optlen != TCPOLEN_SIGNATURE)
3480 to->to_flags |= TOF_SIGNATURE;
3481 to->to_signature = cp + 2;
3483 case TCPOPT_SACK_PERMITTED:
3484 if (optlen != TCPOLEN_SACK_PERMITTED)
3486 if (!(flags & TO_SYN))
3490 to->to_flags |= TOF_SACKPERM;
3493 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3497 to->to_flags |= TOF_SACK;
3498 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3499 to->to_sacks = cp + 2;
3500 TCPSTAT_INC(tcps_sack_rcv_blocks);
3502 case TCPOPT_FAST_OPEN:
3504 * Cookie length validation is performed by the
3505 * server side cookie checking code or the client
3506 * side cookie cache update code.
3508 if (!(flags & TO_SYN))
3510 if (!V_tcp_fastopen_client_enable &&
3511 !V_tcp_fastopen_server_enable)
3513 to->to_flags |= TOF_FASTOPEN;
3514 to->to_tfo_len = optlen - 2;
3515 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3524 * Pull out of band byte out of a segment so
3525 * it doesn't appear in the user's data queue.
3526 * It is still reflected in the segment length for
3527 * sequencing purposes.
3530 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3533 int cnt = off + th->th_urp - 1;
3536 if (m->m_len > cnt) {
3537 char *cp = mtod(m, caddr_t) + cnt;
3538 struct tcpcb *tp = sototcpcb(so);
3540 INP_WLOCK_ASSERT(tp->t_inpcb);
3543 tp->t_oobflags |= TCPOOB_HAVEDATA;
3544 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3546 if (m->m_flags & M_PKTHDR)
3555 panic("tcp_pulloutofband");
3559 * Collect new round-trip time estimate
3560 * and update averages and current timeout.
3563 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3567 INP_WLOCK_ASSERT(tp->t_inpcb);
3569 TCPSTAT_INC(tcps_rttupdated);
3572 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3573 imax(0, rtt * 1000 / hz));
3575 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3577 * srtt is stored as fixed point with 5 bits after the
3578 * binary point (i.e., scaled by 8). The following magic
3579 * is equivalent to the smoothing algorithm in rfc793 with
3580 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3581 * point). Adjust rtt to origin 0.
3583 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3584 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3586 if ((tp->t_srtt += delta) <= 0)
3590 * We accumulate a smoothed rtt variance (actually, a
3591 * smoothed mean difference), then set the retransmit
3592 * timer to smoothed rtt + 4 times the smoothed variance.
3593 * rttvar is stored as fixed point with 4 bits after the
3594 * binary point (scaled by 16). The following is
3595 * equivalent to rfc793 smoothing with an alpha of .75
3596 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3597 * rfc793's wired-in beta.
3601 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3602 if ((tp->t_rttvar += delta) <= 0)
3604 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3605 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3608 * No rtt measurement yet - use the unsmoothed rtt.
3609 * Set the variance to half the rtt (so our first
3610 * retransmit happens at 3*rtt).
3612 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3613 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3614 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3620 * the retransmit should happen at rtt + 4 * rttvar.
3621 * Because of the way we do the smoothing, srtt and rttvar
3622 * will each average +1/2 tick of bias. When we compute
3623 * the retransmit timer, we want 1/2 tick of rounding and
3624 * 1 extra tick because of +-1/2 tick uncertainty in the
3625 * firing of the timer. The bias will give us exactly the
3626 * 1.5 tick we need. But, because the bias is
3627 * statistical, we have to test that we don't drop below
3628 * the minimum feasible timer (which is 2 ticks).
3630 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3631 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3634 * We received an ack for a packet that wasn't retransmitted;
3635 * it is probably safe to discard any error indications we've
3636 * received recently. This isn't quite right, but close enough
3637 * for now (a route might have failed after we sent a segment,
3638 * and the return path might not be symmetrical).
3640 tp->t_softerror = 0;
3644 * Determine a reasonable value for maxseg size.
3645 * If the route is known, check route for mtu.
3646 * If none, use an mss that can be handled on the outgoing interface
3647 * without forcing IP to fragment. If no route is found, route has no mtu,
3648 * or the destination isn't local, use a default, hopefully conservative
3649 * size (usually 512 or the default IP max size, but no more than the mtu
3650 * of the interface), as we can't discover anything about intervening
3651 * gateways or networks. We also initialize the congestion/slow start
3652 * window to be a single segment if the destination isn't local.
3653 * While looking at the routing entry, we also initialize other path-dependent
3654 * parameters from pre-set or cached values in the routing entry.
3656 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3657 * IP options, e.g. IPSEC data, since length of this data may vary, and
3658 * thus it is calculated for every segment separately in tcp_output().
3660 * NOTE that this routine is only called when we process an incoming
3661 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3662 * settings are handled in tcp_mssopt().
3665 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3666 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3669 uint32_t maxmtu = 0;
3670 struct inpcb *inp = tp->t_inpcb;
3671 struct hc_metrics_lite metrics;
3673 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3674 size_t min_protoh = isipv6 ?
3675 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3676 sizeof (struct tcpiphdr);
3678 size_t min_protoh = sizeof(struct tcpiphdr);
3681 INP_WLOCK_ASSERT(tp->t_inpcb);
3684 min_protoh += V_tcp_udp_tunneling_overhead;
3685 if (mtuoffer != -1) {
3686 KASSERT(offer == -1, ("%s: conflict", __func__));
3687 offer = mtuoffer - min_protoh;
3693 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3694 tp->t_maxseg = V_tcp_v6mssdflt;
3697 #if defined(INET) && defined(INET6)
3702 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3703 tp->t_maxseg = V_tcp_mssdflt;
3708 * No route to sender, stay with default mss and return.
3712 * In case we return early we need to initialize metrics
3713 * to a defined state as tcp_hc_get() would do for us
3714 * if there was no cache hit.
3716 if (metricptr != NULL)
3717 bzero(metricptr, sizeof(struct hc_metrics_lite));
3721 /* What have we got? */
3725 * Offer == 0 means that there was no MSS on the SYN
3726 * segment, in this case we use tcp_mssdflt as
3727 * already assigned to t_maxseg above.
3729 offer = tp->t_maxseg;
3734 * Offer == -1 means that we didn't receive SYN yet.
3740 * Prevent DoS attack with too small MSS. Round up
3741 * to at least minmss.
3743 offer = max(offer, V_tcp_minmss);
3747 * rmx information is now retrieved from tcp_hostcache.
3749 tcp_hc_get(&inp->inp_inc, &metrics);
3750 if (metricptr != NULL)
3751 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3754 * If there's a discovered mtu in tcp hostcache, use it.
3755 * Else, use the link mtu.
3757 if (metrics.rmx_mtu)
3758 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3762 mss = maxmtu - min_protoh;
3763 if (!V_path_mtu_discovery &&
3764 !in6_localaddr(&inp->in6p_faddr))
3765 mss = min(mss, V_tcp_v6mssdflt);
3768 #if defined(INET) && defined(INET6)
3773 mss = maxmtu - min_protoh;
3774 if (!V_path_mtu_discovery &&
3775 !in_localaddr(inp->inp_faddr))
3776 mss = min(mss, V_tcp_mssdflt);
3780 * XXX - The above conditional (mss = maxmtu - min_protoh)
3781 * probably violates the TCP spec.
3782 * The problem is that, since we don't know the
3783 * other end's MSS, we are supposed to use a conservative
3784 * default. But, if we do that, then MTU discovery will
3785 * never actually take place, because the conservative
3786 * default is much less than the MTUs typically seen
3787 * on the Internet today. For the moment, we'll sweep
3788 * this under the carpet.
3790 * The conservative default might not actually be a problem
3791 * if the only case this occurs is when sending an initial
3792 * SYN with options and data to a host we've never talked
3793 * to before. Then, they will reply with an MSS value which
3794 * will get recorded and the new parameters should get
3795 * recomputed. For Further Study.
3798 mss = min(mss, offer);
3801 * Sanity check: make sure that maxseg will be large
3802 * enough to allow some data on segments even if the
3803 * all the option space is used (40bytes). Otherwise
3804 * funny things may happen in tcp_output.
3806 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3814 tcp_mss(struct tcpcb *tp, int offer)
3820 struct hc_metrics_lite metrics;
3821 struct tcp_ifcap cap;
3823 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3825 bzero(&cap, sizeof(cap));
3826 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3832 * If there's a pipesize, change the socket buffer to that size,
3833 * don't change if sb_hiwat is different than default (then it
3834 * has been changed on purpose with setsockopt).
3835 * Make the socket buffers an integral number of mss units;
3836 * if the mss is larger than the socket buffer, decrease the mss.
3838 so = inp->inp_socket;
3839 SOCKBUF_LOCK(&so->so_snd);
3840 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3841 bufsize = metrics.rmx_sendpipe;
3843 bufsize = so->so_snd.sb_hiwat;
3847 bufsize = roundup(bufsize, mss);
3848 if (bufsize > sb_max)
3850 if (bufsize > so->so_snd.sb_hiwat)
3851 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3853 SOCKBUF_UNLOCK(&so->so_snd);
3855 * Sanity check: make sure that maxseg will be large
3856 * enough to allow some data on segments even if the
3857 * all the option space is used (40bytes). Otherwise
3858 * funny things may happen in tcp_output.
3860 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3862 tp->t_maxseg = max(mss, 64);
3864 SOCKBUF_LOCK(&so->so_rcv);
3865 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3866 bufsize = metrics.rmx_recvpipe;
3868 bufsize = so->so_rcv.sb_hiwat;
3869 if (bufsize > mss) {
3870 bufsize = roundup(bufsize, mss);
3871 if (bufsize > sb_max)
3873 if (bufsize > so->so_rcv.sb_hiwat)
3874 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3876 SOCKBUF_UNLOCK(&so->so_rcv);
3878 /* Check the interface for TSO capabilities. */
3879 if (cap.ifcap & CSUM_TSO) {
3880 tp->t_flags |= TF_TSO;
3881 tp->t_tsomax = cap.tsomax;
3882 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3883 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3888 * Determine the MSS option to send on an outgoing SYN.
3891 tcp_mssopt(struct in_conninfo *inc)
3894 uint32_t thcmtu = 0;
3895 uint32_t maxmtu = 0;
3898 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3901 if (inc->inc_flags & INC_ISIPV6) {
3902 mss = V_tcp_v6mssdflt;
3903 maxmtu = tcp_maxmtu6(inc, NULL);
3904 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3907 #if defined(INET) && defined(INET6)
3912 mss = V_tcp_mssdflt;
3913 maxmtu = tcp_maxmtu(inc, NULL);
3914 min_protoh = sizeof(struct tcpiphdr);
3917 #if defined(INET6) || defined(INET)
3918 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3921 if (maxmtu && thcmtu)
3922 mss = min(maxmtu, thcmtu) - min_protoh;
3923 else if (maxmtu || thcmtu)
3924 mss = max(maxmtu, thcmtu) - min_protoh;
3930 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
3932 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3933 int maxseg = tcp_maxseg(tp);
3935 INP_WLOCK_ASSERT(tp->t_inpcb);
3938 * Compute the amount of data that this ACK is indicating
3939 * (del_data) and an estimate of how many bytes are in the
3942 if (((tp->t_flags & TF_SACK_PERMIT) &&
3943 (to->to_flags & TOF_SACK)) ||
3944 (IN_CONGRECOVERY(tp->t_flags) &&
3945 !IN_FASTRECOVERY(tp->t_flags))) {
3946 del_data = tp->sackhint.delivered_data;
3947 if (V_tcp_do_newsack)
3948 pipe = tcp_compute_pipe(tp);
3950 pipe = (tp->snd_nxt - tp->snd_fack) +
3951 tp->sackhint.sack_bytes_rexmit;
3953 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg +
3954 tp->snd_recover - tp->snd_una))
3956 pipe = imax(0, tp->snd_max - tp->snd_una -
3957 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg);
3959 tp->sackhint.prr_delivered += del_data;
3961 * Proportional Rate Reduction
3963 if (pipe >= tp->snd_ssthresh) {
3964 if (tp->sackhint.recover_fs == 0)
3965 tp->sackhint.recover_fs =
3966 imax(1, tp->snd_nxt - tp->snd_una);
3967 snd_cnt = howmany((long)tp->sackhint.prr_delivered *
3968 tp->snd_ssthresh, tp->sackhint.recover_fs) -
3969 tp->sackhint.prr_out;
3971 if (V_tcp_do_prr_conservative || (del_data == 0))
3972 limit = tp->sackhint.prr_delivered -
3973 tp->sackhint.prr_out;
3975 limit = imax(tp->sackhint.prr_delivered -
3976 tp->sackhint.prr_out, del_data) +
3978 snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
3980 snd_cnt = imax(snd_cnt, 0) / maxseg;
3982 * Send snd_cnt new data into the network in response to this ack.
3983 * If there is going to be a SACK retransmission, adjust snd_cwnd
3986 if (IN_FASTRECOVERY(tp->t_flags)) {
3987 if ((tp->t_flags & TF_SACK_PERMIT) &&
3988 (to->to_flags & TOF_SACK)) {
3989 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
3990 tp->sackhint.sack_bytes_rexmit +
3993 tp->snd_cwnd = (tp->snd_max - tp->snd_una) +
3996 } else if (IN_CONGRECOVERY(tp->t_flags))
3997 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg);
3998 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd);
4002 * On a partial ack arrives, force the retransmission of the
4003 * next unacknowledged segment. Do not clear tp->t_dupacks.
4004 * By setting snd_nxt to ti_ack, this forces retransmission timer to
4008 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4010 tcp_seq onxt = tp->snd_nxt;
4011 uint32_t ocwnd = tp->snd_cwnd;
4012 u_int maxseg = tcp_maxseg(tp);
4014 INP_WLOCK_ASSERT(tp->t_inpcb);
4016 tcp_timer_activate(tp, TT_REXMT, 0);
4018 tp->snd_nxt = th->th_ack;
4020 * Set snd_cwnd to one segment beyond acknowledged offset.
4021 * (tp->snd_una has not yet been updated when this function is called.)
4023 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4024 tp->t_flags |= TF_ACKNOW;
4025 (void) tcp_output(tp);
4026 tp->snd_cwnd = ocwnd;
4027 if (SEQ_GT(onxt, tp->snd_nxt))
4030 * Partial window deflation. Relies on fact that tp->snd_una
4033 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4034 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4037 tp->snd_cwnd += maxseg;
4041 tcp_compute_pipe(struct tcpcb *tp)
4043 return (tp->snd_max - tp->snd_una +
4044 tp->sackhint.sack_bytes_rexmit -
4045 tp->sackhint.sacked_bytes);
4049 tcp_compute_initwnd(uint32_t maxseg)
4052 * Calculate the Initial Window, also used as Restart Window
4054 * RFC5681 Section 3.1 specifies the default conservative values.
4055 * RFC3390 specifies slightly more aggressive values.
4056 * RFC6928 increases it to ten segments.
4057 * Support for user specified value for initial flight size.
4059 if (V_tcp_initcwnd_segments)
4060 return min(V_tcp_initcwnd_segments * maxseg,
4061 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4062 else if (V_tcp_do_rfc3390)
4063 return min(4 * maxseg, max(2 * maxseg, 4380));
4065 /* Per RFC5681 Section 3.1 */
4067 return (2 * maxseg);
4068 else if (maxseg > 1095)
4069 return (3 * maxseg);
4071 return (4 * maxseg);