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/udp.h>
128 #include <netipsec/ipsec_support.h>
130 #include <machine/in_cksum.h>
132 #include <security/mac/mac_framework.h>
134 const int tcprexmtthresh = 3;
136 VNET_DEFINE(int, tcp_log_in_vain) = 0;
137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
138 &VNET_NAME(tcp_log_in_vain), 0,
139 "Log all incoming TCP segments to closed ports");
141 VNET_DEFINE(int, blackhole) = 0;
142 #define V_blackhole VNET(blackhole)
143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
144 &VNET_NAME(blackhole), 0,
145 "Do not send RST on segments to closed ports");
147 VNET_DEFINE(int, tcp_delack_enabled) = 1;
148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
149 &VNET_NAME(tcp_delack_enabled), 0,
150 "Delay ACK to try and piggyback it onto a data packet");
152 VNET_DEFINE(int, drop_synfin) = 0;
153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
154 &VNET_NAME(drop_synfin), 0,
155 "Drop TCP packets with SYN+FIN set");
157 VNET_DEFINE(int, tcp_do_prr_conservative) = 0;
158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW,
159 &VNET_NAME(tcp_do_prr_conservative), 0,
160 "Do conservative Proportional Rate Reduction");
162 VNET_DEFINE(int, tcp_do_prr) = 1;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW,
164 &VNET_NAME(tcp_do_prr), 1,
165 "Enable Proportional Rate Reduction per RFC 6937");
167 VNET_DEFINE(int, tcp_do_newcwv) = 0;
168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
169 &VNET_NAME(tcp_do_newcwv), 0,
170 "Enable New Congestion Window Validation per RFC7661");
172 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
174 &VNET_NAME(tcp_do_rfc6675_pipe), 0,
175 "Use calculated pipe/in-flight bytes per RFC 6675");
177 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
179 &VNET_NAME(tcp_do_rfc3042), 0,
180 "Enable RFC 3042 (Limited Transmit)");
182 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
184 &VNET_NAME(tcp_do_rfc3390), 0,
185 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
187 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
189 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
190 "Slow-start flight size (initial congestion window) in number of segments");
192 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
194 &VNET_NAME(tcp_do_rfc3465), 0,
195 "Enable RFC 3465 (Appropriate Byte Counting)");
197 VNET_DEFINE(int, tcp_abc_l_var) = 2;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
199 &VNET_NAME(tcp_abc_l_var), 2,
200 "Cap the max cwnd increment during slow-start to this number of segments");
202 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn,
203 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
206 VNET_DEFINE(int, tcp_do_ecn) = 2;
207 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
208 &VNET_NAME(tcp_do_ecn), 0,
211 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
212 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
213 &VNET_NAME(tcp_ecn_maxretries), 0,
214 "Max retries before giving up on ECN");
216 VNET_DEFINE(int, tcp_insecure_syn) = 0;
217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
218 &VNET_NAME(tcp_insecure_syn), 0,
219 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
221 VNET_DEFINE(int, tcp_insecure_rst) = 0;
222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
223 &VNET_NAME(tcp_insecure_rst), 0,
224 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
226 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
227 #define V_tcp_recvspace VNET(tcp_recvspace)
228 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
229 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
231 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
232 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
233 &VNET_NAME(tcp_do_autorcvbuf), 0,
234 "Enable automatic receive buffer sizing");
236 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
237 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
238 &VNET_NAME(tcp_autorcvbuf_max), 0,
239 "Max size of automatic receive buffer");
241 VNET_DEFINE(struct inpcbhead, tcb);
242 #define tcb6 tcb /* for KAME src sync over BSD*'s */
243 VNET_DEFINE(struct inpcbinfo, tcbinfo);
246 * TCP statistics are stored in an array of counter(9)s, which size matches
247 * size of struct tcpstat. TCP running connection count is a regular array.
249 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
250 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
251 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
252 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
253 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
254 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
255 "TCP connection counts by TCP state");
258 tcp_vnet_init(const void *unused)
261 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
262 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
264 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
265 tcp_vnet_init, NULL);
269 tcp_vnet_uninit(const void *unused)
272 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
273 VNET_PCPUSTAT_FREE(tcpstat);
275 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
276 tcp_vnet_uninit, NULL);
280 * Kernel module interface for updating tcpstat. The first argument is an index
281 * into tcpstat treated as an array.
284 kmod_tcpstat_add(int statnum, int val)
287 counter_u64_add(VNET(tcpstat)[statnum], val);
291 * Make sure that we only start a SACK loss recovery when
292 * receiving a duplicate ACK with a SACK block, and also
293 * complete SACK loss recovery in case the other end
297 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to)
299 return ((tp->t_flags & TF_SACK_PERMIT) &&
300 ((to->to_flags & TOF_SACK) ||
301 (!TAILQ_EMPTY(&tp->snd_holes))));
306 * Wrapper for the TCP established input helper hook.
309 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
311 struct tcp_hhook_data hhook_data;
313 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
318 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
325 * CC wrapper hook functions
328 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
335 INP_WLOCK_ASSERT(tp->t_inpcb);
337 tp->ccv->nsegs = nsegs;
338 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
339 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
340 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
341 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
342 tp->ccv->flags |= CCF_CWND_LIMITED;
344 tp->ccv->flags &= ~CCF_CWND_LIMITED;
346 if (type == CC_ACK) {
348 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
349 ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
350 if (!IN_RECOVERY(tp->t_flags))
351 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
352 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
353 if ((tp->t_flags & TF_GPUTINPROG) &&
354 SEQ_GEQ(th->th_ack, tp->gput_ack)) {
356 * Compute goodput in bits per millisecond.
358 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) /
359 max(1, tcp_ts_getticks() - tp->gput_ts);
360 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
363 * XXXLAS: This is a temporary hack, and should be
364 * chained off VOI_TCP_GPUT when stats(9) grows an API
365 * to deal with chained VOIs.
367 if (tp->t_stats_gput_prev > 0)
368 stats_voi_update_abs_s32(tp->t_stats,
370 ((gput - tp->t_stats_gput_prev) * 100) /
371 tp->t_stats_gput_prev);
372 tp->t_flags &= ~TF_GPUTINPROG;
373 tp->t_stats_gput_prev = gput;
376 if (tp->snd_cwnd > tp->snd_ssthresh) {
377 tp->t_bytes_acked += tp->ccv->bytes_this_ack;
378 if (tp->t_bytes_acked >= tp->snd_cwnd) {
379 tp->t_bytes_acked -= tp->snd_cwnd;
380 tp->ccv->flags |= CCF_ABC_SENTAWND;
383 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
384 tp->t_bytes_acked = 0;
388 if (CC_ALGO(tp)->ack_received != NULL) {
389 /* XXXLAS: Find a way to live without this */
390 tp->ccv->curack = th->th_ack;
391 CC_ALGO(tp)->ack_received(tp->ccv, type);
394 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
399 cc_conn_init(struct tcpcb *tp)
401 struct hc_metrics_lite metrics;
402 struct inpcb *inp = tp->t_inpcb;
406 INP_WLOCK_ASSERT(tp->t_inpcb);
408 tcp_hc_get(&inp->inp_inc, &metrics);
409 maxseg = tcp_maxseg(tp);
411 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
413 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
414 TCPSTAT_INC(tcps_usedrtt);
415 if (metrics.rmx_rttvar) {
416 tp->t_rttvar = metrics.rmx_rttvar;
417 TCPSTAT_INC(tcps_usedrttvar);
419 /* default variation is +- 1 rtt */
421 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
423 TCPT_RANGESET(tp->t_rxtcur,
424 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
425 tp->t_rttmin, TCPTV_REXMTMAX);
427 if (metrics.rmx_ssthresh) {
429 * There's some sort of gateway or interface
430 * buffer limit on the path. Use this to set
431 * the slow start threshold, but set the
432 * threshold to no less than 2*mss.
434 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
435 TCPSTAT_INC(tcps_usedssthresh);
439 * Set the initial slow-start flight size.
441 * If a SYN or SYN/ACK was lost and retransmitted, we have to
442 * reduce the initial CWND to one segment as congestion is likely
443 * requiring us to be cautious.
445 if (tp->snd_cwnd == 1)
446 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
448 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
450 if (CC_ALGO(tp)->conn_init != NULL)
451 CC_ALGO(tp)->conn_init(tp->ccv);
455 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
457 INP_WLOCK_ASSERT(tp->t_inpcb);
460 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
465 if (!IN_FASTRECOVERY(tp->t_flags)) {
466 tp->snd_recover = tp->snd_max;
467 if (tp->t_flags2 & TF2_ECN_PERMIT)
468 tp->t_flags2 |= TF2_ECN_SND_CWR;
472 if (!IN_CONGRECOVERY(tp->t_flags) ||
474 * Allow ECN reaction on ACK to CWR, if
475 * that data segment was also CE marked.
477 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
478 EXIT_CONGRECOVERY(tp->t_flags);
479 TCPSTAT_INC(tcps_ecn_rcwnd);
480 tp->snd_recover = tp->snd_max + 1;
481 if (tp->t_flags2 & TF2_ECN_PERMIT)
482 tp->t_flags2 |= TF2_ECN_SND_CWR;
487 tp->t_bytes_acked = 0;
488 EXIT_RECOVERY(tp->t_flags);
489 if (tp->t_flags2 & TF2_ECN_PERMIT)
490 tp->t_flags2 |= TF2_ECN_SND_CWR;
493 TCPSTAT_INC(tcps_sndrexmitbad);
494 /* RTO was unnecessary, so reset everything. */
495 tp->snd_cwnd = tp->snd_cwnd_prev;
496 tp->snd_ssthresh = tp->snd_ssthresh_prev;
497 tp->snd_recover = tp->snd_recover_prev;
498 if (tp->t_flags & TF_WASFRECOVERY)
499 ENTER_FASTRECOVERY(tp->t_flags);
500 if (tp->t_flags & TF_WASCRECOVERY)
501 ENTER_CONGRECOVERY(tp->t_flags);
502 tp->snd_nxt = tp->snd_max;
503 tp->t_flags &= ~TF_PREVVALID;
508 if (CC_ALGO(tp)->cong_signal != NULL) {
510 tp->ccv->curack = th->th_ack;
511 CC_ALGO(tp)->cong_signal(tp->ccv, type);
516 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
518 INP_WLOCK_ASSERT(tp->t_inpcb);
520 /* XXXLAS: KASSERT that we're in recovery? */
522 if (CC_ALGO(tp)->post_recovery != NULL) {
523 tp->ccv->curack = th->th_ack;
524 CC_ALGO(tp)->post_recovery(tp->ccv);
526 /* XXXLAS: EXIT_RECOVERY ? */
527 tp->t_bytes_acked = 0;
528 tp->sackhint.delivered_data = 0;
529 tp->sackhint.prr_out = 0;
533 * Indicate whether this ack should be delayed. We can delay the ack if
534 * following conditions are met:
535 * - There is no delayed ack timer in progress.
536 * - Our last ack wasn't a 0-sized window. We never want to delay
537 * the ack that opens up a 0-sized window.
538 * - LRO wasn't used for this segment. We make sure by checking that the
539 * segment size is not larger than the MSS.
541 #define DELAY_ACK(tp, tlen) \
542 ((!tcp_timer_active(tp, TT_DELACK) && \
543 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
544 (tlen <= tp->t_maxseg) && \
545 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
548 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos)
550 INP_WLOCK_ASSERT(tp->t_inpcb);
552 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
553 switch (iptos & IPTOS_ECN_MASK) {
555 tp->ccv->flags |= CCF_IPHDR_CE;
561 case IPTOS_ECN_NOTECT:
562 tp->ccv->flags &= ~CCF_IPHDR_CE;
567 tp->ccv->flags |= CCF_TCPHDR_CWR;
569 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
571 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
573 if (tp->ccv->flags & CCF_ACKNOW) {
574 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
575 tp->t_flags |= TF_ACKNOW;
581 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
583 cc_ecnpkt_handler_flags(tp, th->th_flags, iptos);
587 * TCP input handling is split into multiple parts:
588 * tcp6_input is a thin wrapper around tcp_input for the extended
589 * ip6_protox[] call format in ip6_input
590 * tcp_input handles primary segment validation, inpcb lookup and
591 * SYN processing on listen sockets
592 * tcp_do_segment processes the ACK and text of the segment for
593 * establishing, established and closing connections
597 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
600 struct in6_ifaddr *ia6;
604 if (m->m_len < *offp + sizeof(struct tcphdr)) {
605 m = m_pullup(m, *offp + sizeof(struct tcphdr));
608 TCPSTAT_INC(tcps_rcvshort);
609 return (IPPROTO_DONE);
614 * draft-itojun-ipv6-tcp-to-anycast
615 * better place to put this in?
617 ip6 = mtod(m, struct ip6_hdr *);
618 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
619 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
620 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
621 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
623 return (IPPROTO_DONE);
627 return (tcp_input_with_port(mp, offp, proto, port));
631 tcp6_input(struct mbuf **mp, int *offp, int proto)
634 return(tcp6_input_with_port(mp, offp, proto, 0));
639 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
641 struct mbuf *m = *mp;
642 struct tcphdr *th = NULL;
643 struct ip *ip = NULL;
644 struct inpcb *inp = NULL;
645 struct tcpcb *tp = NULL;
646 struct socket *so = NULL;
657 int rstreason = 0; /* For badport_bandlim accounting purposes */
659 struct m_tag *fwd_tag = NULL;
661 struct ip6_hdr *ip6 = NULL;
664 const void *ip6 = NULL;
666 struct tcpopt to; /* options in this segment */
667 char *s = NULL; /* address and port logging */
670 * The size of tcp_saveipgen must be the size of the max ip header,
673 u_char tcp_saveipgen[IP6_HDR_LEN];
674 struct tcphdr tcp_savetcp;
681 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
688 TCPSTAT_INC(tcps_rcvtotal);
692 ip6 = mtod(m, struct ip6_hdr *);
693 th = (struct tcphdr *)((caddr_t)ip6 + off0);
694 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
697 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
698 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
699 th->th_sum = m->m_pkthdr.csum_data;
701 th->th_sum = in6_cksum_pseudo(ip6, tlen,
702 IPPROTO_TCP, m->m_pkthdr.csum_data);
703 th->th_sum ^= 0xffff;
705 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
707 TCPSTAT_INC(tcps_rcvbadsum);
712 * Be proactive about unspecified IPv6 address in source.
713 * As we use all-zero to indicate unbounded/unconnected pcb,
714 * unspecified IPv6 address can be used to confuse us.
716 * Note that packets with unspecified IPv6 destination is
717 * already dropped in ip6_input.
719 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
723 iptos = IPV6_TRAFFIC_CLASS(ip6);
726 #if defined(INET) && defined(INET6)
732 * Get IP and TCP header together in first mbuf.
733 * Note: IP leaves IP header in first mbuf.
735 if (off0 > sizeof (struct ip)) {
737 off0 = sizeof(struct ip);
739 if (m->m_len < sizeof (struct tcpiphdr)) {
740 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
742 TCPSTAT_INC(tcps_rcvshort);
743 return (IPPROTO_DONE);
746 ip = mtod(m, struct ip *);
747 th = (struct tcphdr *)((caddr_t)ip + off0);
748 tlen = ntohs(ip->ip_len) - off0;
753 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
754 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
755 th->th_sum = m->m_pkthdr.csum_data;
757 th->th_sum = in_pseudo(ip->ip_src.s_addr,
759 htonl(m->m_pkthdr.csum_data + tlen +
761 th->th_sum ^= 0xffff;
763 struct ipovly *ipov = (struct ipovly *)ip;
766 * Checksum extended TCP header and data.
770 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
771 ipov->ih_len = htons(tlen);
772 th->th_sum = in_cksum(m, len);
773 /* Reset length for SDT probes. */
774 ip->ip_len = htons(len);
777 /* Re-initialization for later version check */
779 ip->ip_v = IPVERSION;
780 ip->ip_hl = off0 >> 2;
783 if (th->th_sum && (port == 0)) {
784 TCPSTAT_INC(tcps_rcvbadsum);
791 * Check that TCP offset makes sense,
792 * pull out TCP options and adjust length. XXX
794 off = th->th_off << 2;
795 if (off < sizeof (struct tcphdr) || off > tlen) {
796 TCPSTAT_INC(tcps_rcvbadoff);
799 tlen -= off; /* tlen is used instead of ti->ti_len */
800 if (off > sizeof (struct tcphdr)) {
803 if (m->m_len < off0 + off) {
804 m = m_pullup(m, off0 + off);
806 TCPSTAT_INC(tcps_rcvshort);
807 return (IPPROTO_DONE);
810 ip6 = mtod(m, struct ip6_hdr *);
811 th = (struct tcphdr *)((caddr_t)ip6 + off0);
814 #if defined(INET) && defined(INET6)
819 if (m->m_len < sizeof(struct ip) + off) {
820 if ((m = m_pullup(m, sizeof (struct ip) + off))
822 TCPSTAT_INC(tcps_rcvshort);
823 return (IPPROTO_DONE);
825 ip = mtod(m, struct ip *);
826 th = (struct tcphdr *)((caddr_t)ip + off0);
830 optlen = off - sizeof (struct tcphdr);
831 optp = (u_char *)(th + 1);
833 thflags = th->th_flags;
836 * Convert TCP protocol specific fields to host format.
838 tcp_fields_to_host(th);
841 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
843 drop_hdrlen = off0 + off;
846 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
850 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
852 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
855 #if defined(INET) && !defined(INET6)
856 (m->m_flags & M_IP_NEXTHOP)
859 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
863 if (isipv6 && fwd_tag != NULL) {
864 struct sockaddr_in6 *next_hop6;
866 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
868 * Transparently forwarded. Pretend to be the destination.
869 * Already got one like this?
871 inp = in6_pcblookup_mbuf(&V_tcbinfo,
872 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
873 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
876 * It's new. Try to find the ambushing socket.
877 * Because we've rewritten the destination address,
878 * any hardware-generated hash is ignored.
880 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
881 th->th_sport, &next_hop6->sin6_addr,
882 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
883 th->th_dport, INPLOOKUP_WILDCARD |
884 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
887 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
888 th->th_sport, &ip6->ip6_dst, th->th_dport,
889 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
890 m->m_pkthdr.rcvif, m);
893 #if defined(INET6) && defined(INET)
897 if (fwd_tag != NULL) {
898 struct sockaddr_in *next_hop;
900 next_hop = (struct sockaddr_in *)(fwd_tag+1);
902 * Transparently forwarded. Pretend to be the destination.
903 * already got one like this?
905 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
906 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
907 m->m_pkthdr.rcvif, m);
910 * It's new. Try to find the ambushing socket.
911 * Because we've rewritten the destination address,
912 * any hardware-generated hash is ignored.
914 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
915 th->th_sport, next_hop->sin_addr,
916 next_hop->sin_port ? ntohs(next_hop->sin_port) :
917 th->th_dport, INPLOOKUP_WILDCARD |
918 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
921 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
922 th->th_sport, ip->ip_dst, th->th_dport,
923 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
924 m->m_pkthdr.rcvif, m);
928 * If the INPCB does not exist then all data in the incoming
929 * segment is discarded and an appropriate RST is sent back.
930 * XXX MRT Send RST using which routing table?
934 * Log communication attempts to ports that are not
937 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
938 V_tcp_log_in_vain == 2) {
939 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
940 log(LOG_INFO, "%s; %s: Connection attempt "
941 "to closed port\n", s, __func__);
944 * When blackholing do not respond with a RST but
945 * completely ignore the segment and drop it.
947 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
951 rstreason = BANDLIM_RST_CLOSEDPORT;
954 INP_WLOCK_ASSERT(inp);
956 * While waiting for inp lock during the lookup, another thread
957 * can have dropped the inpcb, in which case we need to loop back
958 * and try to find a new inpcb to deliver to.
960 if (inp->inp_flags & INP_DROPPED) {
965 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
966 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
967 ((inp->inp_socket == NULL) || !SOLISTENING(inp->inp_socket))) {
968 inp->inp_flowid = m->m_pkthdr.flowid;
969 inp->inp_flowtype = M_HASHTYPE_GET(m);
971 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
973 if (isipv6 && IPSEC_ENABLED(ipv6) &&
974 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
982 if (IPSEC_ENABLED(ipv4) &&
983 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
990 * Check the minimum TTL for socket.
992 if (inp->inp_ip_minttl != 0) {
995 if (inp->inp_ip_minttl > ip6->ip6_hlim)
999 if (inp->inp_ip_minttl > ip->ip_ttl)
1004 * A previous connection in TIMEWAIT state is supposed to catch stray
1005 * or duplicate segments arriving late. If this segment was a
1006 * legitimate new connection attempt, the old INPCB gets removed and
1007 * we can try again to find a listening socket.
1009 * At this point, due to earlier optimism, we may hold only an inpcb
1010 * lock, and not the inpcbinfo write lock. If so, we need to try to
1011 * acquire it, or if that fails, acquire a reference on the inpcb,
1012 * drop all locks, acquire a global write lock, and then re-acquire
1013 * the inpcb lock. We may at that point discover that another thread
1014 * has tried to free the inpcb, in which case we need to loop back
1015 * and try to find a new inpcb to deliver to.
1017 * XXXRW: It may be time to rethink timewait locking.
1019 if (inp->inp_flags & INP_TIMEWAIT) {
1020 tcp_dooptions(&to, optp, optlen,
1021 (thflags & TH_SYN) ? TO_SYN : 0);
1023 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
1025 if (tcp_twcheck(inp, &to, th, m, tlen))
1027 return (IPPROTO_DONE);
1030 * The TCPCB may no longer exist if the connection is winding
1031 * down or it is in the CLOSED state. Either way we drop the
1032 * segment and send an appropriate response.
1034 tp = intotcpcb(inp);
1035 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1036 rstreason = BANDLIM_RST_CLOSEDPORT;
1040 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) {
1041 rstreason = BANDLIM_RST_CLOSEDPORT;
1046 if (tp->t_flags & TF_TOE) {
1047 tcp_offload_input(tp, m);
1048 m = NULL; /* consumed by the TOE driver */
1054 INP_WLOCK_ASSERT(inp);
1055 if (mac_inpcb_check_deliver(inp, m))
1058 so = inp->inp_socket;
1059 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1061 if (so->so_options & SO_DEBUG) {
1062 ostate = tp->t_state;
1065 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1068 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1071 #endif /* TCPDEBUG */
1073 * When the socket is accepting connections (the INPCB is in LISTEN
1074 * state) we look into the SYN cache if this is a new connection
1075 * attempt or the completion of a previous one.
1077 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so),
1078 ("%s: so accepting but tp %p not listening", __func__, tp));
1079 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) {
1080 struct in_conninfo inc;
1082 bzero(&inc, sizeof(inc));
1085 inc.inc_flags |= INC_ISIPV6;
1086 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1087 inc.inc_flags |= INC_IPV6MINMTU;
1088 inc.inc6_faddr = ip6->ip6_src;
1089 inc.inc6_laddr = ip6->ip6_dst;
1093 inc.inc_faddr = ip->ip_src;
1094 inc.inc_laddr = ip->ip_dst;
1096 inc.inc_fport = th->th_sport;
1097 inc.inc_lport = th->th_dport;
1098 inc.inc_fibnum = so->so_fibnum;
1101 * Check for an existing connection attempt in syncache if
1102 * the flag is only ACK. A successful lookup creates a new
1103 * socket appended to the listen queue in SYN_RECEIVED state.
1105 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1107 * Parse the TCP options here because
1108 * syncookies need access to the reflected
1111 tcp_dooptions(&to, optp, optlen, 0);
1113 * NB: syncache_expand() doesn't unlock
1114 * inp and tcpinfo locks.
1116 rstreason = syncache_expand(&inc, &to, th, &so, m, port);
1117 if (rstreason < 0) {
1119 * A failing TCP MD5 signature comparison
1120 * must result in the segment being dropped
1121 * and must not produce any response back
1125 } else if (rstreason == 0) {
1127 * No syncache entry or ACK was not
1128 * for our SYN/ACK. Send a RST.
1129 * NB: syncache did its own logging
1130 * of the failure cause.
1132 rstreason = BANDLIM_RST_OPENPORT;
1138 * We completed the 3-way handshake
1139 * but could not allocate a socket
1140 * either due to memory shortage,
1141 * listen queue length limits or
1142 * global socket limits. Send RST
1143 * or wait and have the remote end
1144 * retransmit the ACK for another
1147 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1148 log(LOG_DEBUG, "%s; %s: Listen socket: "
1149 "Socket allocation failed due to "
1150 "limits or memory shortage, %s\n",
1152 V_tcp_sc_rst_sock_fail ?
1153 "sending RST" : "try again");
1154 if (V_tcp_sc_rst_sock_fail) {
1155 rstreason = BANDLIM_UNLIMITED;
1161 * Socket is created in state SYN_RECEIVED.
1162 * Unlock the listen socket, lock the newly
1163 * created socket and update the tp variable.
1165 INP_WUNLOCK(inp); /* listen socket */
1166 inp = sotoinpcb(so);
1168 * New connection inpcb is already locked by
1169 * syncache_expand().
1171 INP_WLOCK_ASSERT(inp);
1172 tp = intotcpcb(inp);
1173 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1174 ("%s: ", __func__));
1176 * Process the segment and the data it
1177 * contains. tcp_do_segment() consumes
1178 * the mbuf chain and unlocks the inpcb.
1180 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1181 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1183 return (IPPROTO_DONE);
1186 * Segment flag validation for new connection attempts:
1188 * Our (SYN|ACK) response was rejected.
1189 * Check with syncache and remove entry to prevent
1192 * NB: syncache_chkrst does its own logging of failure
1195 if (thflags & TH_RST) {
1196 syncache_chkrst(&inc, th, m, port);
1200 * We can't do anything without SYN.
1202 if ((thflags & TH_SYN) == 0) {
1203 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1204 log(LOG_DEBUG, "%s; %s: Listen socket: "
1205 "SYN is missing, segment ignored\n",
1207 TCPSTAT_INC(tcps_badsyn);
1211 * (SYN|ACK) is bogus on a listen socket.
1213 if (thflags & TH_ACK) {
1214 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1215 log(LOG_DEBUG, "%s; %s: Listen socket: "
1216 "SYN|ACK invalid, segment rejected\n",
1218 syncache_badack(&inc, port); /* XXX: Not needed! */
1219 TCPSTAT_INC(tcps_badsyn);
1220 rstreason = BANDLIM_RST_OPENPORT;
1224 * If the drop_synfin option is enabled, drop all
1225 * segments with both the SYN and FIN bits set.
1226 * This prevents e.g. nmap from identifying the
1228 * XXX: Poor reasoning. nmap has other methods
1229 * and is constantly refining its stack detection
1231 * XXX: This is a violation of the TCP specification
1232 * and was used by RFC1644.
1234 if ((thflags & TH_FIN) && V_drop_synfin) {
1235 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1236 log(LOG_DEBUG, "%s; %s: Listen socket: "
1237 "SYN|FIN segment ignored (based on "
1238 "sysctl setting)\n", s, __func__);
1239 TCPSTAT_INC(tcps_badsyn);
1243 * Segment's flags are (SYN) or (SYN|FIN).
1245 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1246 * as they do not affect the state of the TCP FSM.
1247 * The data pointed to by TH_URG and th_urp is ignored.
1249 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1250 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1251 KASSERT(thflags & (TH_SYN),
1252 ("%s: Listen socket: TH_SYN not set", __func__));
1255 * If deprecated address is forbidden,
1256 * we do not accept SYN to deprecated interface
1257 * address to prevent any new inbound connection from
1258 * getting established.
1259 * When we do not accept SYN, we send a TCP RST,
1260 * with deprecated source address (instead of dropping
1261 * it). We compromise it as it is much better for peer
1262 * to send a RST, and RST will be the final packet
1265 * If we do not forbid deprecated addresses, we accept
1266 * the SYN packet. RFC2462 does not suggest dropping
1268 * If we decipher RFC2462 5.5.4, it says like this:
1269 * 1. use of deprecated addr with existing
1270 * communication is okay - "SHOULD continue to be
1272 * 2. use of it with new communication:
1273 * (2a) "SHOULD NOT be used if alternate address
1274 * with sufficient scope is available"
1275 * (2b) nothing mentioned otherwise.
1276 * Here we fall into (2b) case as we have no choice in
1277 * our source address selection - we must obey the peer.
1279 * The wording in RFC2462 is confusing, and there are
1280 * multiple description text for deprecated address
1281 * handling - worse, they are not exactly the same.
1282 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1284 if (isipv6 && !V_ip6_use_deprecated) {
1285 struct in6_ifaddr *ia6;
1287 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1289 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1290 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1291 log(LOG_DEBUG, "%s; %s: Listen socket: "
1292 "Connection attempt to deprecated "
1293 "IPv6 address rejected\n",
1295 rstreason = BANDLIM_RST_OPENPORT;
1301 * Basic sanity checks on incoming SYN requests:
1302 * Don't respond if the destination is a link layer
1303 * broadcast according to RFC1122 4.2.3.10, p. 104.
1304 * If it is from this socket it must be forged.
1305 * Don't respond if the source or destination is a
1306 * global or subnet broad- or multicast address.
1307 * Note that it is quite possible to receive unicast
1308 * link-layer packets with a broadcast IP address. Use
1309 * in_broadcast() to find them.
1311 if (m->m_flags & (M_BCAST|M_MCAST)) {
1312 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1313 log(LOG_DEBUG, "%s; %s: Listen socket: "
1314 "Connection attempt from broad- or multicast "
1315 "link layer address ignored\n", s, __func__);
1320 if (th->th_dport == th->th_sport &&
1321 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1322 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1323 log(LOG_DEBUG, "%s; %s: Listen socket: "
1324 "Connection attempt to/from self "
1325 "ignored\n", s, __func__);
1328 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1329 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1330 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1331 log(LOG_DEBUG, "%s; %s: Listen socket: "
1332 "Connection attempt from/to multicast "
1333 "address ignored\n", s, __func__);
1338 #if defined(INET) && defined(INET6)
1343 if (th->th_dport == th->th_sport &&
1344 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1345 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1346 log(LOG_DEBUG, "%s; %s: Listen socket: "
1347 "Connection attempt from/to self "
1348 "ignored\n", s, __func__);
1351 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1352 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1353 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1354 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1355 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1356 log(LOG_DEBUG, "%s; %s: Listen socket: "
1357 "Connection attempt from/to broad- "
1358 "or multicast address ignored\n",
1365 * SYN appears to be valid. Create compressed TCP state
1369 if (so->so_options & SO_DEBUG)
1370 tcp_trace(TA_INPUT, ostate, tp,
1371 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1373 TCP_PROBE3(debug__input, tp, th, m);
1374 tcp_dooptions(&to, optp, optlen, TO_SYN);
1375 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos,
1377 goto tfo_socket_result;
1380 * Entry added to syncache and mbuf consumed.
1381 * Only the listen socket is unlocked by syncache_add().
1383 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1384 return (IPPROTO_DONE);
1385 } else if (tp->t_state == TCPS_LISTEN) {
1387 * When a listen socket is torn down the SO_ACCEPTCONN
1388 * flag is removed first while connections are drained
1389 * from the accept queue in a unlock/lock cycle of the
1390 * ACCEPT_LOCK, opening a race condition allowing a SYN
1391 * attempt go through unhandled.
1395 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1396 if (tp->t_flags & TF_SIGNATURE) {
1397 tcp_dooptions(&to, optp, optlen, thflags);
1398 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1399 TCPSTAT_INC(tcps_sig_err_nosigopt);
1402 if (!TCPMD5_ENABLED() ||
1403 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1407 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1410 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1411 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1412 * the inpcb, and unlocks pcbinfo.
1414 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1415 return (IPPROTO_DONE);
1418 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1421 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1424 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1425 m = NULL; /* mbuf chain got consumed. */
1430 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1436 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1441 return (IPPROTO_DONE);
1445 * Automatic sizing of receive socket buffer. Often the send
1446 * buffer size is not optimally adjusted to the actual network
1447 * conditions at hand (delay bandwidth product). Setting the
1448 * buffer size too small limits throughput on links with high
1449 * bandwidth and high delay (eg. trans-continental/oceanic links).
1451 * On the receive side the socket buffer memory is only rarely
1452 * used to any significant extent. This allows us to be much
1453 * more aggressive in scaling the receive socket buffer. For
1454 * the case that the buffer space is actually used to a large
1455 * extent and we run out of kernel memory we can simply drop
1456 * the new segments; TCP on the sender will just retransmit it
1457 * later. Setting the buffer size too big may only consume too
1458 * much kernel memory if the application doesn't read() from
1459 * the socket or packet loss or reordering makes use of the
1462 * The criteria to step up the receive buffer one notch are:
1463 * 1. Application has not set receive buffer size with
1464 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1465 * 2. the number of bytes received during 1/2 of an sRTT
1466 * is at least 3/8 of the current socket buffer size.
1467 * 3. receive buffer size has not hit maximal automatic size;
1469 * If all of the criteria are met we increaset the socket buffer
1470 * by a 1/2 (bounded by the max). This allows us to keep ahead
1471 * of slow-start but also makes it so our peer never gets limited
1472 * by our rwnd which we then open up causing a burst.
1474 * This algorithm does two steps per RTT at most and only if
1475 * we receive a bulk stream w/o packet losses or reorderings.
1476 * Shrinking the buffer during idle times is not necessary as
1477 * it doesn't consume any memory when idle.
1479 * TODO: Only step up if the application is actually serving
1480 * the buffer to better manage the socket buffer resources.
1483 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1484 struct tcpcb *tp, int tlen)
1488 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1489 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1490 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1491 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1492 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1493 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1494 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1496 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1498 /* Start over with next RTT. */
1502 tp->rfbuf_cnt += tlen; /* add up */
1508 tcp_input(struct mbuf **mp, int *offp, int proto)
1510 return(tcp_input_with_port(mp, offp, proto, 0));
1514 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1517 * Since tp might be gone if the session entered
1518 * the TIME_WAIT state before coming here, we need
1519 * to check if the socket is still connected.
1527 INP_LOCK_ASSERT(tp->t_inpcb);
1528 if (tp->t_flags & TF_WAKESOR) {
1529 tp->t_flags &= ~TF_WAKESOR;
1530 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1531 sorwakeup_locked(so);
1536 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1537 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1539 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1540 int rstreason, todrop, win, incforsyn = 0;
1544 struct in_conninfo *inc;
1552 * The size of tcp_saveipgen must be the size of the max ip header,
1555 u_char tcp_saveipgen[IP6_HDR_LEN];
1556 struct tcphdr tcp_savetcp;
1559 thflags = th->th_flags;
1560 inc = &tp->t_inpcb->inp_inc;
1561 tp->sackhint.last_sack_ack = 0;
1563 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1566 INP_WLOCK_ASSERT(tp->t_inpcb);
1567 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1569 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1573 /* Save segment, if requested. */
1574 tcp_pcap_add(th, m, &(tp->t_inpkts));
1576 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1579 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1580 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1581 log(LOG_DEBUG, "%s; %s: "
1582 "SYN|FIN segment ignored (based on "
1583 "sysctl setting)\n", s, __func__);
1590 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1591 * check SEQ.ACK first.
1593 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1594 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1595 rstreason = BANDLIM_UNLIMITED;
1600 * Segment received on connection.
1601 * Reset idle time and keep-alive timer.
1602 * XXX: This should be done after segment
1603 * validation to ignore broken/spoofed segs.
1605 tp->t_rcvtime = ticks;
1608 * Scale up the window into a 32-bit value.
1609 * For the SYN_SENT state the scale is zero.
1611 tiwin = th->th_win << tp->snd_scale;
1613 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1617 * TCP ECN processing.
1619 if (tp->t_flags2 & TF2_ECN_PERMIT) {
1620 if (thflags & TH_CWR) {
1621 tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1622 tp->t_flags |= TF_ACKNOW;
1624 switch (iptos & IPTOS_ECN_MASK) {
1626 tp->t_flags2 |= TF2_ECN_SND_ECE;
1627 TCPSTAT_INC(tcps_ecn_ce);
1629 case IPTOS_ECN_ECT0:
1630 TCPSTAT_INC(tcps_ecn_ect0);
1632 case IPTOS_ECN_ECT1:
1633 TCPSTAT_INC(tcps_ecn_ect1);
1637 /* Process a packet differently from RFC3168. */
1638 cc_ecnpkt_handler(tp, th, iptos);
1640 /* Congestion experienced. */
1641 if (thflags & TH_ECE) {
1642 cc_cong_signal(tp, th, CC_ECN);
1647 * Parse options on any incoming segment.
1649 tcp_dooptions(&to, (u_char *)(th + 1),
1650 (th->th_off << 2) - sizeof(struct tcphdr),
1651 (thflags & TH_SYN) ? TO_SYN : 0);
1653 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1654 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1655 (to.to_flags & TOF_SIGNATURE) == 0) {
1656 TCPSTAT_INC(tcps_sig_err_sigopt);
1657 /* XXX: should drop? */
1661 * If echoed timestamp is later than the current time,
1662 * fall back to non RFC1323 RTT calculation. Normalize
1663 * timestamp if syncookies were used when this connection
1666 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1667 to.to_tsecr -= tp->ts_offset;
1668 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1670 else if (tp->t_rxtshift == 1 &&
1671 tp->t_flags & TF_PREVVALID &&
1672 tp->t_badrxtwin != 0 &&
1673 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
1674 cc_cong_signal(tp, th, CC_RTO_ERR);
1677 * Process options only when we get SYN/ACK back. The SYN case
1678 * for incoming connections is handled in tcp_syncache.
1679 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1680 * or <SYN,ACK>) segment itself is never scaled.
1681 * XXX this is traditional behavior, may need to be cleaned up.
1683 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1684 /* Handle parallel SYN for ECN */
1685 if (!(thflags & TH_ACK) &&
1686 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1687 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1688 tp->t_flags2 |= TF2_ECN_PERMIT;
1689 tp->t_flags2 |= TF2_ECN_SND_ECE;
1690 TCPSTAT_INC(tcps_ecn_shs);
1692 if ((to.to_flags & TOF_SCALE) &&
1693 (tp->t_flags & TF_REQ_SCALE) &&
1694 !(tp->t_flags & TF_NOOPT)) {
1695 tp->t_flags |= TF_RCVD_SCALE;
1696 tp->snd_scale = to.to_wscale;
1698 tp->t_flags &= ~TF_REQ_SCALE;
1700 * Initial send window. It will be updated with
1701 * the next incoming segment to the scaled value.
1703 tp->snd_wnd = th->th_win;
1704 if ((to.to_flags & TOF_TS) &&
1705 (tp->t_flags & TF_REQ_TSTMP) &&
1706 !(tp->t_flags & TF_NOOPT)) {
1707 tp->t_flags |= TF_RCVD_TSTMP;
1708 tp->ts_recent = to.to_tsval;
1709 tp->ts_recent_age = tcp_ts_getticks();
1711 tp->t_flags &= ~TF_REQ_TSTMP;
1712 if (to.to_flags & TOF_MSS)
1713 tcp_mss(tp, to.to_mss);
1714 if ((tp->t_flags & TF_SACK_PERMIT) &&
1715 (!(to.to_flags & TOF_SACKPERM) ||
1716 (tp->t_flags & TF_NOOPT)))
1717 tp->t_flags &= ~TF_SACK_PERMIT;
1718 if (IS_FASTOPEN(tp->t_flags)) {
1719 if ((to.to_flags & TOF_FASTOPEN) &&
1720 !(tp->t_flags & TF_NOOPT)) {
1723 if (to.to_flags & TOF_MSS)
1726 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1730 tcp_fastopen_update_cache(tp, mss,
1731 to.to_tfo_len, to.to_tfo_cookie);
1733 tcp_fastopen_disable_path(tp);
1738 * If timestamps were negotiated during SYN/ACK and a
1739 * segment without a timestamp is received, silently drop
1740 * the segment, unless it is a RST segment or missing timestamps are
1742 * See section 3.2 of RFC 7323.
1744 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1745 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1746 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1747 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1748 "segment processed normally\n",
1753 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1754 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1755 "segment silently dropped\n", s, __func__);
1762 * If timestamps were not negotiated during SYN/ACK and a
1763 * segment with a timestamp is received, ignore the
1764 * timestamp and process the packet normally.
1765 * See section 3.2 of RFC 7323.
1767 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1768 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1769 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1770 "segment processed normally\n", s, __func__);
1776 * Header prediction: check for the two common cases
1777 * of a uni-directional data xfer. If the packet has
1778 * no control flags, is in-sequence, the window didn't
1779 * change and we're not retransmitting, it's a
1780 * candidate. If the length is zero and the ack moved
1781 * forward, we're the sender side of the xfer. Just
1782 * free the data acked & wake any higher level process
1783 * that was blocked waiting for space. If the length
1784 * is non-zero and the ack didn't move, we're the
1785 * receiver side. If we're getting packets in-order
1786 * (the reassembly queue is empty), add the data to
1787 * the socket buffer and note that we need a delayed ack.
1788 * Make sure that the hidden state-flags are also off.
1789 * Since we check for TCPS_ESTABLISHED first, it can only
1792 if (tp->t_state == TCPS_ESTABLISHED &&
1793 th->th_seq == tp->rcv_nxt &&
1794 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1795 tp->snd_nxt == tp->snd_max &&
1796 tiwin && tiwin == tp->snd_wnd &&
1797 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1799 ((to.to_flags & TOF_TS) == 0 ||
1800 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1802 * If last ACK falls within this segment's sequence numbers,
1803 * record the timestamp.
1804 * NOTE that the test is modified according to the latest
1805 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1807 if ((to.to_flags & TOF_TS) != 0 &&
1808 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1809 tp->ts_recent_age = tcp_ts_getticks();
1810 tp->ts_recent = to.to_tsval;
1814 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1815 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1816 !IN_RECOVERY(tp->t_flags) &&
1817 (to.to_flags & TOF_SACK) == 0 &&
1818 TAILQ_EMPTY(&tp->snd_holes)) {
1820 * This is a pure ack for outstanding data.
1822 TCPSTAT_INC(tcps_predack);
1825 * "bad retransmit" recovery without timestamps.
1827 if ((to.to_flags & TOF_TS) == 0 &&
1828 tp->t_rxtshift == 1 &&
1829 tp->t_flags & TF_PREVVALID &&
1830 tp->t_badrxtwin != 0 &&
1831 TSTMP_LT(ticks, tp->t_badrxtwin)) {
1832 cc_cong_signal(tp, th, CC_RTO_ERR);
1836 * Recalculate the transmit timer / rtt.
1838 * Some boxes send broken timestamp replies
1839 * during the SYN+ACK phase, ignore
1840 * timestamps of 0 or we could calculate a
1841 * huge RTT and blow up the retransmit timer.
1843 if ((to.to_flags & TOF_TS) != 0 &&
1847 t = tcp_ts_getticks() - to.to_tsecr;
1848 if (!tp->t_rttlow || tp->t_rttlow > t)
1851 TCP_TS_TO_TICKS(t) + 1);
1852 } else if (tp->t_rtttime &&
1853 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1854 if (!tp->t_rttlow ||
1855 tp->t_rttlow > ticks - tp->t_rtttime)
1856 tp->t_rttlow = ticks - tp->t_rtttime;
1858 ticks - tp->t_rtttime);
1860 acked = BYTES_THIS_ACK(tp, th);
1863 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1864 hhook_run_tcp_est_in(tp, th, &to);
1867 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1868 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1869 sbdrop(&so->so_snd, acked);
1870 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1871 SEQ_LEQ(th->th_ack, tp->snd_recover))
1872 tp->snd_recover = th->th_ack - 1;
1875 * Let the congestion control algorithm update
1876 * congestion control related information. This
1877 * typically means increasing the congestion
1880 cc_ack_received(tp, th, nsegs, CC_ACK);
1882 tp->snd_una = th->th_ack;
1884 * Pull snd_wl2 up to prevent seq wrap relative
1887 tp->snd_wl2 = th->th_ack;
1892 * If all outstanding data are acked, stop
1893 * retransmit timer, otherwise restart timer
1894 * using current (possibly backed-off) value.
1895 * If process is waiting for space,
1896 * wakeup/selwakeup/signal. If data
1897 * are ready to send, let tcp_output
1898 * decide between more output or persist.
1901 if (so->so_options & SO_DEBUG)
1902 tcp_trace(TA_INPUT, ostate, tp,
1903 (void *)tcp_saveipgen,
1906 TCP_PROBE3(debug__input, tp, th, m);
1907 if (tp->snd_una == tp->snd_max)
1908 tcp_timer_activate(tp, TT_REXMT, 0);
1909 else if (!tcp_timer_active(tp, TT_PERSIST))
1910 tcp_timer_activate(tp, TT_REXMT,
1913 if (sbavail(&so->so_snd))
1914 (void) tp->t_fb->tfb_tcp_output(tp);
1917 } else if (th->th_ack == tp->snd_una &&
1918 tlen <= sbspace(&so->so_rcv)) {
1919 int newsize = 0; /* automatic sockbuf scaling */
1922 * This is a pure, in-sequence data packet with
1923 * nothing on the reassembly queue and we have enough
1924 * buffer space to take it.
1926 /* Clean receiver SACK report if present */
1927 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1928 tcp_clean_sackreport(tp);
1929 TCPSTAT_INC(tcps_preddat);
1930 tp->rcv_nxt += tlen;
1932 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1933 (tp->t_fbyte_in == 0)) {
1934 tp->t_fbyte_in = ticks;
1935 if (tp->t_fbyte_in == 0)
1937 if (tp->t_fbyte_out && tp->t_fbyte_in)
1938 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1941 * Pull snd_wl1 up to prevent seq wrap relative to
1944 tp->snd_wl1 = th->th_seq;
1946 * Pull rcv_up up to prevent seq wrap relative to
1949 tp->rcv_up = tp->rcv_nxt;
1950 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1951 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1953 if (so->so_options & SO_DEBUG)
1954 tcp_trace(TA_INPUT, ostate, tp,
1955 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1957 TCP_PROBE3(debug__input, tp, th, m);
1959 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1961 /* Add data to socket buffer. */
1962 SOCKBUF_LOCK(&so->so_rcv);
1963 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1967 * Set new socket buffer size.
1968 * Give up when limit is reached.
1971 if (!sbreserve_locked(&so->so_rcv,
1973 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1974 m_adj(m, drop_hdrlen); /* delayed header drop */
1975 sbappendstream_locked(&so->so_rcv, m, 0);
1977 /* NB: sorwakeup_locked() does an implicit unlock. */
1978 sorwakeup_locked(so);
1979 if (DELAY_ACK(tp, tlen)) {
1980 tp->t_flags |= TF_DELACK;
1982 tp->t_flags |= TF_ACKNOW;
1983 tp->t_fb->tfb_tcp_output(tp);
1990 * Calculate amount of space in receive window,
1991 * and then do TCP input processing.
1992 * Receive window is amount of space in rcv queue,
1993 * but not less than advertised window.
1995 win = sbspace(&so->so_rcv);
1998 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
2000 switch (tp->t_state) {
2002 * If the state is SYN_RECEIVED:
2003 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
2005 case TCPS_SYN_RECEIVED:
2006 if ((thflags & TH_ACK) &&
2007 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
2008 SEQ_GT(th->th_ack, tp->snd_max))) {
2009 rstreason = BANDLIM_RST_OPENPORT;
2012 if (IS_FASTOPEN(tp->t_flags)) {
2014 * When a TFO connection is in SYN_RECEIVED, the
2015 * only valid packets are the initial SYN, a
2016 * retransmit/copy of the initial SYN (possibly with
2017 * a subset of the original data), a valid ACK, a
2020 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
2021 rstreason = BANDLIM_RST_OPENPORT;
2023 } else if (thflags & TH_SYN) {
2024 /* non-initial SYN is ignored */
2025 if ((tcp_timer_active(tp, TT_DELACK) ||
2026 tcp_timer_active(tp, TT_REXMT)))
2028 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2035 * If the state is SYN_SENT:
2036 * if seg contains a RST with valid ACK (SEQ.ACK has already
2037 * been verified), then drop the connection.
2038 * if seg contains a RST without an ACK, drop the seg.
2039 * if seg does not contain SYN, then drop the seg.
2040 * Otherwise this is an acceptable SYN segment
2041 * initialize tp->rcv_nxt and tp->irs
2042 * if seg contains ack then advance tp->snd_una
2043 * if seg contains an ECE and ECN support is enabled, the stream
2045 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2046 * arrange for segment to be acked (eventually)
2047 * continue processing rest of data/controls, beginning with URG
2050 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2051 TCP_PROBE5(connect__refused, NULL, tp,
2053 tp = tcp_drop(tp, ECONNREFUSED);
2055 if (thflags & TH_RST)
2057 if (!(thflags & TH_SYN))
2060 tp->irs = th->th_seq;
2062 if (thflags & TH_ACK) {
2063 int tfo_partial_ack = 0;
2065 TCPSTAT_INC(tcps_connects);
2068 mac_socketpeer_set_from_mbuf(m, so);
2070 /* Do window scaling on this connection? */
2071 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2072 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2073 tp->rcv_scale = tp->request_r_scale;
2075 tp->rcv_adv += min(tp->rcv_wnd,
2076 TCP_MAXWIN << tp->rcv_scale);
2077 tp->snd_una++; /* SYN is acked */
2079 * If not all the data that was sent in the TFO SYN
2080 * has been acked, resend the remainder right away.
2082 if (IS_FASTOPEN(tp->t_flags) &&
2083 (tp->snd_una != tp->snd_max)) {
2084 tp->snd_nxt = th->th_ack;
2085 tfo_partial_ack = 1;
2088 * If there's data, delay ACK; if there's also a FIN
2089 * ACKNOW will be turned on later.
2091 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2092 tcp_timer_activate(tp, TT_DELACK,
2095 tp->t_flags |= TF_ACKNOW;
2097 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2098 (V_tcp_do_ecn == 1)) {
2099 tp->t_flags2 |= TF2_ECN_PERMIT;
2100 TCPSTAT_INC(tcps_ecn_shs);
2104 * Received <SYN,ACK> in SYN_SENT[*] state.
2106 * SYN_SENT --> ESTABLISHED
2107 * SYN_SENT* --> FIN_WAIT_1
2109 tp->t_starttime = ticks;
2110 if (tp->t_flags & TF_NEEDFIN) {
2111 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2112 tp->t_flags &= ~TF_NEEDFIN;
2115 tcp_state_change(tp, TCPS_ESTABLISHED);
2116 TCP_PROBE5(connect__established, NULL, tp,
2119 tcp_timer_activate(tp, TT_KEEP,
2124 * Received initial SYN in SYN-SENT[*] state =>
2125 * simultaneous open.
2126 * If it succeeds, connection is * half-synchronized.
2127 * Otherwise, do 3-way handshake:
2128 * SYN-SENT -> SYN-RECEIVED
2129 * SYN-SENT* -> SYN-RECEIVED*
2131 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2132 tcp_timer_activate(tp, TT_REXMT, 0);
2133 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2136 INP_WLOCK_ASSERT(tp->t_inpcb);
2139 * Advance th->th_seq to correspond to first data byte.
2140 * If data, trim to stay within window,
2141 * dropping FIN if necessary.
2144 if (tlen > tp->rcv_wnd) {
2145 todrop = tlen - tp->rcv_wnd;
2149 TCPSTAT_INC(tcps_rcvpackafterwin);
2150 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2152 tp->snd_wl1 = th->th_seq - 1;
2153 tp->rcv_up = th->th_seq;
2155 * Client side of transaction: already sent SYN and data.
2156 * If the remote host used T/TCP to validate the SYN,
2157 * our data will be ACK'd; if so, enter normal data segment
2158 * processing in the middle of step 5, ack processing.
2159 * Otherwise, goto step 6.
2161 if (thflags & TH_ACK)
2167 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2168 * do normal processing.
2170 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2174 break; /* continue normal processing */
2178 * States other than LISTEN or SYN_SENT.
2179 * First check the RST flag and sequence number since reset segments
2180 * are exempt from the timestamp and connection count tests. This
2181 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2182 * below which allowed reset segments in half the sequence space
2183 * to fall though and be processed (which gives forged reset
2184 * segments with a random sequence number a 50 percent chance of
2185 * killing a connection).
2186 * Then check timestamp, if present.
2187 * Then check the connection count, if present.
2188 * Then check that at least some bytes of segment are within
2189 * receive window. If segment begins before rcv_nxt,
2190 * drop leading data (and SYN); if nothing left, just ack.
2192 if (thflags & TH_RST) {
2194 * RFC5961 Section 3.2
2196 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2197 * - If RST is in window, we send challenge ACK.
2199 * Note: to take into account delayed ACKs, we should
2200 * test against last_ack_sent instead of rcv_nxt.
2201 * Note 2: we handle special case of closed window, not
2202 * covered by the RFC.
2204 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2205 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2206 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2207 KASSERT(tp->t_state != TCPS_SYN_SENT,
2208 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2211 if (V_tcp_insecure_rst ||
2212 tp->last_ack_sent == th->th_seq) {
2213 TCPSTAT_INC(tcps_drops);
2214 /* Drop the connection. */
2215 switch (tp->t_state) {
2216 case TCPS_SYN_RECEIVED:
2217 so->so_error = ECONNREFUSED;
2219 case TCPS_ESTABLISHED:
2220 case TCPS_FIN_WAIT_1:
2221 case TCPS_FIN_WAIT_2:
2222 case TCPS_CLOSE_WAIT:
2225 so->so_error = ECONNRESET;
2232 TCPSTAT_INC(tcps_badrst);
2233 /* Send challenge ACK. */
2234 tcp_respond(tp, mtod(m, void *), th, m,
2235 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2236 tp->last_ack_sent = tp->rcv_nxt;
2244 * RFC5961 Section 4.2
2245 * Send challenge ACK for any SYN in synchronized state.
2247 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2248 tp->t_state != TCPS_SYN_RECEIVED) {
2249 TCPSTAT_INC(tcps_badsyn);
2250 if (V_tcp_insecure_syn &&
2251 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2252 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2253 tp = tcp_drop(tp, ECONNRESET);
2254 rstreason = BANDLIM_UNLIMITED;
2256 /* Send challenge ACK. */
2257 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2258 tp->snd_nxt, TH_ACK);
2259 tp->last_ack_sent = tp->rcv_nxt;
2266 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2267 * and it's less than ts_recent, drop it.
2269 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2270 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2271 /* Check to see if ts_recent is over 24 days old. */
2272 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2274 * Invalidate ts_recent. If this segment updates
2275 * ts_recent, the age will be reset later and ts_recent
2276 * will get a valid value. If it does not, setting
2277 * ts_recent to zero will at least satisfy the
2278 * requirement that zero be placed in the timestamp
2279 * echo reply when ts_recent isn't valid. The
2280 * age isn't reset until we get a valid ts_recent
2281 * because we don't want out-of-order segments to be
2282 * dropped when ts_recent is old.
2286 TCPSTAT_INC(tcps_rcvduppack);
2287 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2288 TCPSTAT_INC(tcps_pawsdrop);
2296 * In the SYN-RECEIVED state, validate that the packet belongs to
2297 * this connection before trimming the data to fit the receive
2298 * window. Check the sequence number versus IRS since we know
2299 * the sequence numbers haven't wrapped. This is a partial fix
2300 * for the "LAND" DoS attack.
2302 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2303 rstreason = BANDLIM_RST_OPENPORT;
2307 todrop = tp->rcv_nxt - th->th_seq;
2309 if (thflags & TH_SYN) {
2319 * Following if statement from Stevens, vol. 2, p. 960.
2322 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2324 * Any valid FIN must be to the left of the window.
2325 * At this point the FIN must be a duplicate or out
2326 * of sequence; drop it.
2331 * Send an ACK to resynchronize and drop any data.
2332 * But keep on processing for RST or ACK.
2334 tp->t_flags |= TF_ACKNOW;
2336 TCPSTAT_INC(tcps_rcvduppack);
2337 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2339 TCPSTAT_INC(tcps_rcvpartduppack);
2340 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2343 * DSACK - add SACK block for dropped range
2345 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2346 tcp_update_sack_list(tp, th->th_seq,
2347 th->th_seq + todrop);
2349 * ACK now, as the next in-sequence segment
2350 * will clear the DSACK block again
2352 tp->t_flags |= TF_ACKNOW;
2354 drop_hdrlen += todrop; /* drop from the top afterwards */
2355 th->th_seq += todrop;
2357 if (th->th_urp > todrop)
2358 th->th_urp -= todrop;
2366 * If new data are received on a connection after the
2367 * user processes are gone, then RST the other end.
2369 if ((so->so_state & SS_NOFDREF) &&
2370 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2371 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2372 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2373 "after socket was closed, "
2374 "sending RST and removing tcpcb\n",
2375 s, __func__, tcpstates[tp->t_state], tlen);
2379 TCPSTAT_INC(tcps_rcvafterclose);
2380 rstreason = BANDLIM_UNLIMITED;
2385 * If segment ends after window, drop trailing data
2386 * (and PUSH and FIN); if nothing left, just ACK.
2388 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2390 TCPSTAT_INC(tcps_rcvpackafterwin);
2391 if (todrop >= tlen) {
2392 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2394 * If window is closed can only take segments at
2395 * window edge, and have to drop data and PUSH from
2396 * incoming segments. Continue processing, but
2397 * remember to ack. Otherwise, drop segment
2400 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2401 tp->t_flags |= TF_ACKNOW;
2402 TCPSTAT_INC(tcps_rcvwinprobe);
2406 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2409 thflags &= ~(TH_PUSH|TH_FIN);
2413 * If last ACK falls within this segment's sequence numbers,
2414 * record its timestamp.
2416 * 1) That the test incorporates suggestions from the latest
2417 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2418 * 2) That updating only on newer timestamps interferes with
2419 * our earlier PAWS tests, so this check should be solely
2420 * predicated on the sequence space of this segment.
2421 * 3) That we modify the segment boundary check to be
2422 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2423 * instead of RFC1323's
2424 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2425 * This modified check allows us to overcome RFC1323's
2426 * limitations as described in Stevens TCP/IP Illustrated
2427 * Vol. 2 p.869. In such cases, we can still calculate the
2428 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2430 if ((to.to_flags & TOF_TS) != 0 &&
2431 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2432 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2433 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2434 tp->ts_recent_age = tcp_ts_getticks();
2435 tp->ts_recent = to.to_tsval;
2439 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2440 * flag is on (half-synchronized state), then queue data for
2441 * later processing; else drop segment and return.
2443 if ((thflags & TH_ACK) == 0) {
2444 if (tp->t_state == TCPS_SYN_RECEIVED ||
2445 (tp->t_flags & TF_NEEDSYN)) {
2446 if (tp->t_state == TCPS_SYN_RECEIVED &&
2447 IS_FASTOPEN(tp->t_flags)) {
2448 tp->snd_wnd = tiwin;
2452 } else if (tp->t_flags & TF_ACKNOW)
2461 switch (tp->t_state) {
2463 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2464 * ESTABLISHED state and continue processing.
2465 * The ACK was checked above.
2467 case TCPS_SYN_RECEIVED:
2469 TCPSTAT_INC(tcps_connects);
2471 /* Do window scaling? */
2472 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2473 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2474 tp->rcv_scale = tp->request_r_scale;
2476 tp->snd_wnd = tiwin;
2479 * SYN-RECEIVED -> ESTABLISHED
2480 * SYN-RECEIVED* -> FIN-WAIT-1
2482 tp->t_starttime = ticks;
2483 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2484 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2485 tp->t_tfo_pending = NULL;
2487 if (tp->t_flags & TF_NEEDFIN) {
2488 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2489 tp->t_flags &= ~TF_NEEDFIN;
2491 tcp_state_change(tp, TCPS_ESTABLISHED);
2492 TCP_PROBE5(accept__established, NULL, tp,
2495 * TFO connections call cc_conn_init() during SYN
2496 * processing. Calling it again here for such
2497 * connections is not harmless as it would undo the
2498 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2501 if (!IS_FASTOPEN(tp->t_flags))
2503 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2506 * Account for the ACK of our SYN prior to
2507 * regular ACK processing below, except for
2508 * simultaneous SYN, which is handled later.
2510 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2513 * If segment contains data or ACK, will call tcp_reass()
2514 * later; if not, do so now to pass queued data to user.
2516 if (tlen == 0 && (thflags & TH_FIN) == 0) {
2517 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2519 tcp_handle_wakeup(tp, so);
2521 tp->snd_wl1 = th->th_seq - 1;
2525 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2526 * ACKs. If the ack is in the range
2527 * tp->snd_una < th->th_ack <= tp->snd_max
2528 * then advance tp->snd_una to th->th_ack and drop
2529 * data from the retransmission queue. If this ACK reflects
2530 * more up to date window information we update our window information.
2532 case TCPS_ESTABLISHED:
2533 case TCPS_FIN_WAIT_1:
2534 case TCPS_FIN_WAIT_2:
2535 case TCPS_CLOSE_WAIT:
2538 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2539 TCPSTAT_INC(tcps_rcvacktoomuch);
2542 if (tcp_is_sack_recovery(tp, &to))
2543 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2546 * Reset the value so that previous (valid) value
2547 * from the last ack with SACK doesn't get used.
2549 tp->sackhint.sacked_bytes = 0;
2552 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2553 hhook_run_tcp_est_in(tp, th, &to);
2556 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2557 maxseg = tcp_maxseg(tp);
2559 (tiwin == tp->snd_wnd ||
2560 (tp->t_flags & TF_SACK_PERMIT))) {
2562 * If this is the first time we've seen a
2563 * FIN from the remote, this is not a
2564 * duplicate and it needs to be processed
2565 * normally. This happens during a
2566 * simultaneous close.
2568 if ((thflags & TH_FIN) &&
2569 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2573 TCPSTAT_INC(tcps_rcvdupack);
2575 * If we have outstanding data (other than
2576 * a window probe), this is a completely
2577 * duplicate ack (ie, window info didn't
2578 * change and FIN isn't set),
2579 * the ack is the biggest we've
2580 * seen and we've seen exactly our rexmt
2581 * threshold of them, assume a packet
2582 * has been dropped and retransmit it.
2583 * Kludge snd_nxt & the congestion
2584 * window so we send only this one
2587 * We know we're losing at the current
2588 * window size so do congestion avoidance
2589 * (set ssthresh to half the current window
2590 * and pull our congestion window back to
2591 * the new ssthresh).
2593 * Dup acks mean that packets have left the
2594 * network (they're now cached at the receiver)
2595 * so bump cwnd by the amount in the receiver
2596 * to keep a constant cwnd packets in the
2599 * When using TCP ECN, notify the peer that
2600 * we reduced the cwnd.
2603 * Following 2 kinds of acks should not affect
2606 * 2) Acks with SACK but without any new SACK
2607 * information in them. These could result from
2608 * any anomaly in the network like a switch
2609 * duplicating packets or a possible DoS attack.
2611 if (th->th_ack != tp->snd_una ||
2612 (tcp_is_sack_recovery(tp, &to) &&
2615 else if (!tcp_timer_active(tp, TT_REXMT))
2617 else if (++tp->t_dupacks > tcprexmtthresh ||
2618 IN_FASTRECOVERY(tp->t_flags)) {
2619 cc_ack_received(tp, th, nsegs,
2622 IN_FASTRECOVERY(tp->t_flags)) {
2623 tcp_do_prr_ack(tp, th, &to);
2624 } else if (tcp_is_sack_recovery(tp, &to) &&
2625 IN_FASTRECOVERY(tp->t_flags)) {
2629 * Compute the amount of data in flight first.
2630 * We can inject new data into the pipe iff
2631 * we have less than 1/2 the original window's
2632 * worth of data in flight.
2634 if (V_tcp_do_rfc6675_pipe)
2635 awnd = tcp_compute_pipe(tp);
2637 awnd = (tp->snd_nxt - tp->snd_fack) +
2638 tp->sackhint.sack_bytes_rexmit;
2640 if (awnd < tp->snd_ssthresh) {
2641 tp->snd_cwnd += maxseg;
2642 if (tp->snd_cwnd > tp->snd_ssthresh)
2643 tp->snd_cwnd = tp->snd_ssthresh;
2646 tp->snd_cwnd += maxseg;
2647 (void) tp->t_fb->tfb_tcp_output(tp);
2649 } else if (tp->t_dupacks == tcprexmtthresh ||
2650 (tp->t_flags & TF_SACK_PERMIT &&
2651 V_tcp_do_rfc6675_pipe &&
2652 tp->sackhint.sacked_bytes >
2653 (tcprexmtthresh - 1) * maxseg)) {
2656 * Above is the RFC6675 trigger condition of
2657 * more than (dupthresh-1)*maxseg sacked data.
2658 * If the count of holes in the
2659 * scoreboard is >= dupthresh, we could
2660 * also enter loss recovery, but don't
2661 * have that value readily available.
2663 tp->t_dupacks = tcprexmtthresh;
2664 tcp_seq onxt = tp->snd_nxt;
2667 * If we're doing sack, or prr, check
2668 * to see if we're already in sack
2669 * recovery. If we're not doing sack,
2670 * check to see if we're in newreno
2674 (tp->t_flags & TF_SACK_PERMIT)) {
2675 if (IN_FASTRECOVERY(tp->t_flags)) {
2680 if (SEQ_LEQ(th->th_ack,
2686 /* Congestion signal before ack. */
2687 cc_cong_signal(tp, th, CC_NDUPACK);
2688 cc_ack_received(tp, th, nsegs,
2690 tcp_timer_activate(tp, TT_REXMT, 0);
2694 * snd_ssthresh is already updated by
2697 if (tcp_is_sack_recovery(tp, &to)) {
2698 tp->sackhint.prr_delivered =
2699 tp->sackhint.sacked_bytes;
2701 tp->sackhint.prr_delivered =
2702 imin(tp->snd_max - tp->snd_una,
2703 imin(INT_MAX / 65536,
2704 tp->t_dupacks) * maxseg);
2706 tp->sackhint.recover_fs = max(1,
2707 tp->snd_nxt - tp->snd_una);
2709 if (tcp_is_sack_recovery(tp, &to)) {
2711 tcps_sack_recovery_episode);
2712 tp->snd_recover = tp->snd_nxt;
2713 tp->snd_cwnd = maxseg;
2714 (void) tp->t_fb->tfb_tcp_output(tp);
2715 if (SEQ_GT(th->th_ack, tp->snd_una))
2716 goto resume_partialack;
2719 tp->snd_nxt = th->th_ack;
2720 tp->snd_cwnd = maxseg;
2721 (void) tp->t_fb->tfb_tcp_output(tp);
2722 KASSERT(tp->snd_limited <= 2,
2723 ("%s: tp->snd_limited too big",
2725 tp->snd_cwnd = tp->snd_ssthresh +
2727 (tp->t_dupacks - tp->snd_limited);
2728 if (SEQ_GT(onxt, tp->snd_nxt))
2731 } else if (V_tcp_do_rfc3042) {
2733 * Process first and second duplicate
2734 * ACKs. Each indicates a segment
2735 * leaving the network, creating room
2736 * for more. Make sure we can send a
2737 * packet on reception of each duplicate
2738 * ACK by increasing snd_cwnd by one
2739 * segment. Restore the original
2740 * snd_cwnd after packet transmission.
2742 cc_ack_received(tp, th, nsegs,
2744 uint32_t oldcwnd = tp->snd_cwnd;
2745 tcp_seq oldsndmax = tp->snd_max;
2749 KASSERT(tp->t_dupacks == 1 ||
2751 ("%s: dupacks not 1 or 2",
2753 if (tp->t_dupacks == 1)
2754 tp->snd_limited = 0;
2756 (tp->snd_nxt - tp->snd_una) +
2757 (tp->t_dupacks - tp->snd_limited) *
2760 * Only call tcp_output when there
2761 * is new data available to be sent
2762 * or we need to send an ACK.
2764 SOCKBUF_LOCK(&so->so_snd);
2765 avail = sbavail(&so->so_snd) -
2766 (tp->snd_nxt - tp->snd_una);
2767 SOCKBUF_UNLOCK(&so->so_snd);
2768 if (avail > 0 || tp->t_flags & TF_ACKNOW)
2769 (void) tp->t_fb->tfb_tcp_output(tp);
2770 sent = tp->snd_max - oldsndmax;
2771 if (sent > maxseg) {
2772 KASSERT((tp->t_dupacks == 2 &&
2773 tp->snd_limited == 0) ||
2774 (sent == maxseg + 1 &&
2775 tp->t_flags & TF_SENTFIN),
2776 ("%s: sent too much",
2778 tp->snd_limited = 2;
2779 } else if (sent > 0)
2781 tp->snd_cwnd = oldcwnd;
2788 * This ack is advancing the left edge, reset the
2793 * If this ack also has new SACK info, increment the
2794 * counter as per rfc6675. The variable
2795 * sack_changed tracks all changes to the SACK
2796 * scoreboard, including when partial ACKs without
2797 * SACK options are received, and clear the scoreboard
2798 * from the left side. Such partial ACKs should not be
2799 * counted as dupacks here.
2801 if (tcp_is_sack_recovery(tp, &to) &&
2804 /* limit overhead by setting maxseg last */
2805 if (!IN_FASTRECOVERY(tp->t_flags) &&
2806 (tp->sackhint.sacked_bytes >
2807 ((tcprexmtthresh - 1) *
2808 (maxseg = tcp_maxseg(tp))))) {
2809 goto enter_recovery;
2815 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2816 ("%s: th_ack <= snd_una", __func__));
2819 * If the congestion window was inflated to account
2820 * for the other side's cached packets, retract it.
2822 if (IN_FASTRECOVERY(tp->t_flags)) {
2823 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2824 if (tp->t_flags & TF_SACK_PERMIT)
2825 if (V_tcp_do_prr && to.to_flags & TOF_SACK) {
2826 tcp_timer_activate(tp, TT_REXMT, 0);
2828 tcp_do_prr_ack(tp, th, &to);
2829 tp->t_flags |= TF_ACKNOW;
2830 (void) tcp_output(tp);
2832 tcp_sack_partialack(tp, th);
2834 tcp_newreno_partial_ack(tp, th);
2836 cc_post_recovery(tp, th);
2837 } else if (IN_CONGRECOVERY(tp->t_flags)) {
2838 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2840 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th);
2841 tp->snd_fack = th->th_ack;
2842 tcp_do_prr_ack(tp, th, &to);
2843 (void) tcp_output(tp);
2846 cc_post_recovery(tp, th);
2849 * If we reach this point, ACK is not a duplicate,
2850 * i.e., it ACKs something we sent.
2852 if (tp->t_flags & TF_NEEDSYN) {
2854 * T/TCP: Connection was half-synchronized, and our
2855 * SYN has been ACK'd (so connection is now fully
2856 * synchronized). Go to non-starred state,
2857 * increment snd_una for ACK of SYN, and check if
2858 * we can do window scaling.
2860 tp->t_flags &= ~TF_NEEDSYN;
2862 /* Do window scaling? */
2863 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2864 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2865 tp->rcv_scale = tp->request_r_scale;
2866 /* Send window already scaled. */
2871 INP_WLOCK_ASSERT(tp->t_inpcb);
2874 * Adjust for the SYN bit in sequence space,
2875 * but don't account for it in cwnd calculations.
2876 * This is for the SYN_RECEIVED, non-simultaneous
2877 * SYN case. SYN_SENT and simultaneous SYN are
2878 * treated elsewhere.
2882 acked = BYTES_THIS_ACK(tp, th);
2883 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2884 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2885 tp->snd_una, th->th_ack, tp, m));
2886 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2887 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2890 * If we just performed our first retransmit, and the ACK
2891 * arrives within our recovery window, then it was a mistake
2892 * to do the retransmit in the first place. Recover our
2893 * original cwnd and ssthresh, and proceed to transmit where
2896 if (tp->t_rxtshift == 1 &&
2897 tp->t_flags & TF_PREVVALID &&
2898 tp->t_badrxtwin != 0 &&
2899 to.to_flags & TOF_TS &&
2901 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
2902 cc_cong_signal(tp, th, CC_RTO_ERR);
2905 * If we have a timestamp reply, update smoothed
2906 * round trip time. If no timestamp is present but
2907 * transmit timer is running and timed sequence
2908 * number was acked, update smoothed round trip time.
2909 * Since we now have an rtt measurement, cancel the
2910 * timer backoff (cf., Phil Karn's retransmit alg.).
2911 * Recompute the initial retransmit timer.
2913 * Some boxes send broken timestamp replies
2914 * during the SYN+ACK phase, ignore
2915 * timestamps of 0 or we could calculate a
2916 * huge RTT and blow up the retransmit timer.
2918 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2921 t = tcp_ts_getticks() - to.to_tsecr;
2922 if (!tp->t_rttlow || tp->t_rttlow > t)
2924 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2925 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2926 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2927 tp->t_rttlow = ticks - tp->t_rtttime;
2928 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2932 * If all outstanding data is acked, stop retransmit
2933 * timer and remember to restart (more output or persist).
2934 * If there is more data to be acked, restart retransmit
2935 * timer, using current (possibly backed-off) value.
2937 if (th->th_ack == tp->snd_max) {
2938 tcp_timer_activate(tp, TT_REXMT, 0);
2940 } else if (!tcp_timer_active(tp, TT_PERSIST))
2941 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2944 * If no data (only SYN) was ACK'd,
2945 * skip rest of ACK processing.
2951 * Let the congestion control algorithm update congestion
2952 * control related information. This typically means increasing
2953 * the congestion window.
2955 cc_ack_received(tp, th, nsegs, CC_ACK);
2957 SOCKBUF_LOCK(&so->so_snd);
2958 if (acked > sbavail(&so->so_snd)) {
2959 if (tp->snd_wnd >= sbavail(&so->so_snd))
2960 tp->snd_wnd -= sbavail(&so->so_snd);
2963 mfree = sbcut_locked(&so->so_snd,
2964 (int)sbavail(&so->so_snd));
2967 mfree = sbcut_locked(&so->so_snd, acked);
2968 if (tp->snd_wnd >= (uint32_t) acked)
2969 tp->snd_wnd -= acked;
2974 /* NB: sowwakeup_locked() does an implicit unlock. */
2975 sowwakeup_locked(so);
2977 /* Detect una wraparound. */
2978 if (!IN_RECOVERY(tp->t_flags) &&
2979 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2980 SEQ_LEQ(th->th_ack, tp->snd_recover))
2981 tp->snd_recover = th->th_ack - 1;
2982 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2983 if (IN_RECOVERY(tp->t_flags) &&
2984 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2985 EXIT_RECOVERY(tp->t_flags);
2987 tp->snd_una = th->th_ack;
2988 if (tp->t_flags & TF_SACK_PERMIT) {
2989 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2990 tp->snd_recover = tp->snd_una;
2992 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2993 tp->snd_nxt = tp->snd_una;
2995 switch (tp->t_state) {
2997 * In FIN_WAIT_1 STATE in addition to the processing
2998 * for the ESTABLISHED state if our FIN is now acknowledged
2999 * then enter FIN_WAIT_2.
3001 case TCPS_FIN_WAIT_1:
3002 if (ourfinisacked) {
3004 * If we can't receive any more
3005 * data, then closing user can proceed.
3006 * Starting the timer is contrary to the
3007 * specification, but if we don't get a FIN
3008 * we'll hang forever.
3011 * we should release the tp also, and use a
3014 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3015 soisdisconnected(so);
3016 tcp_timer_activate(tp, TT_2MSL,
3017 (tcp_fast_finwait2_recycle ?
3018 tcp_finwait2_timeout :
3021 tcp_state_change(tp, TCPS_FIN_WAIT_2);
3026 * In CLOSING STATE in addition to the processing for
3027 * the ESTABLISHED state if the ACK acknowledges our FIN
3028 * then enter the TIME-WAIT state, otherwise ignore
3032 if (ourfinisacked) {
3040 * In LAST_ACK, we may still be waiting for data to drain
3041 * and/or to be acked, as well as for the ack of our FIN.
3042 * If our FIN is now acknowledged, delete the TCB,
3043 * enter the closed state and return.
3046 if (ourfinisacked) {
3055 INP_WLOCK_ASSERT(tp->t_inpcb);
3058 * Update window information.
3059 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3061 if ((thflags & TH_ACK) &&
3062 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3063 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3064 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3065 /* keep track of pure window updates */
3067 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3068 TCPSTAT_INC(tcps_rcvwinupd);
3069 tp->snd_wnd = tiwin;
3070 tp->snd_wl1 = th->th_seq;
3071 tp->snd_wl2 = th->th_ack;
3072 if (tp->snd_wnd > tp->max_sndwnd)
3073 tp->max_sndwnd = tp->snd_wnd;
3078 * Process segments with URG.
3080 if ((thflags & TH_URG) && th->th_urp &&
3081 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3083 * This is a kludge, but if we receive and accept
3084 * random urgent pointers, we'll crash in
3085 * soreceive. It's hard to imagine someone
3086 * actually wanting to send this much urgent data.
3088 SOCKBUF_LOCK(&so->so_rcv);
3089 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3090 th->th_urp = 0; /* XXX */
3091 thflags &= ~TH_URG; /* XXX */
3092 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3093 goto dodata; /* XXX */
3096 * If this segment advances the known urgent pointer,
3097 * then mark the data stream. This should not happen
3098 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3099 * a FIN has been received from the remote side.
3100 * In these states we ignore the URG.
3102 * According to RFC961 (Assigned Protocols),
3103 * the urgent pointer points to the last octet
3104 * of urgent data. We continue, however,
3105 * to consider it to indicate the first octet
3106 * of data past the urgent section as the original
3107 * spec states (in one of two places).
3109 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3110 tp->rcv_up = th->th_seq + th->th_urp;
3111 so->so_oobmark = sbavail(&so->so_rcv) +
3112 (tp->rcv_up - tp->rcv_nxt) - 1;
3113 if (so->so_oobmark == 0)
3114 so->so_rcv.sb_state |= SBS_RCVATMARK;
3116 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3118 SOCKBUF_UNLOCK(&so->so_rcv);
3120 * Remove out of band data so doesn't get presented to user.
3121 * This can happen independent of advancing the URG pointer,
3122 * but if two URG's are pending at once, some out-of-band
3123 * data may creep in... ick.
3125 if (th->th_urp <= (uint32_t)tlen &&
3126 !(so->so_options & SO_OOBINLINE)) {
3127 /* hdr drop is delayed */
3128 tcp_pulloutofband(so, th, m, drop_hdrlen);
3132 * If no out of band data is expected,
3133 * pull receive urgent pointer along
3134 * with the receive window.
3136 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3137 tp->rcv_up = tp->rcv_nxt;
3140 INP_WLOCK_ASSERT(tp->t_inpcb);
3143 * Process the segment text, merging it into the TCP sequencing queue,
3144 * and arranging for acknowledgment of receipt if necessary.
3145 * This process logically involves adjusting tp->rcv_wnd as data
3146 * is presented to the user (this happens in tcp_usrreq.c,
3147 * case PRU_RCVD). If a FIN has already been received on this
3148 * connection then we just ignore the text.
3150 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3151 IS_FASTOPEN(tp->t_flags));
3152 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3153 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3154 tcp_seq save_start = th->th_seq;
3155 tcp_seq save_rnxt = tp->rcv_nxt;
3156 int save_tlen = tlen;
3157 m_adj(m, drop_hdrlen); /* delayed header drop */
3159 * Insert segment which includes th into TCP reassembly queue
3160 * with control block tp. Set thflags to whether reassembly now
3161 * includes a segment with FIN. This handles the common case
3162 * inline (segment is the next to be received on an established
3163 * connection, and the queue is empty), avoiding linkage into
3164 * and removal from the queue and repetition of various
3166 * Set DELACK for segments received in order, but ack
3167 * immediately when segments are out of order (so
3168 * fast retransmit can work).
3170 if (th->th_seq == tp->rcv_nxt &&
3172 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3174 if (DELAY_ACK(tp, tlen) || tfo_syn)
3175 tp->t_flags |= TF_DELACK;
3177 tp->t_flags |= TF_ACKNOW;
3178 tp->rcv_nxt += tlen;
3180 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3181 (tp->t_fbyte_in == 0)) {
3182 tp->t_fbyte_in = ticks;
3183 if (tp->t_fbyte_in == 0)
3185 if (tp->t_fbyte_out && tp->t_fbyte_in)
3186 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3188 thflags = th->th_flags & TH_FIN;
3189 TCPSTAT_INC(tcps_rcvpack);
3190 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3191 SOCKBUF_LOCK(&so->so_rcv);
3192 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3195 sbappendstream_locked(&so->so_rcv, m, 0);
3196 tp->t_flags |= TF_WAKESOR;
3199 * XXX: Due to the header drop above "th" is
3200 * theoretically invalid by now. Fortunately
3201 * m_adj() doesn't actually frees any mbufs
3202 * when trimming from the head.
3204 tcp_seq temp = save_start;
3206 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3207 tp->t_flags |= TF_ACKNOW;
3209 if ((tp->t_flags & TF_SACK_PERMIT) &&
3211 TCPS_HAVEESTABLISHED(tp->t_state)) {
3212 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3214 * DSACK actually handled in the fastpath
3217 tcp_update_sack_list(tp, save_start,
3218 save_start + save_tlen);
3219 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3220 if ((tp->rcv_numsacks >= 1) &&
3221 (tp->sackblks[0].end == save_start)) {
3223 * Partial overlap, recorded at todrop
3226 tcp_update_sack_list(tp,
3227 tp->sackblks[0].start,
3228 tp->sackblks[0].end);
3230 tcp_update_dsack_list(tp, save_start,
3231 save_start + save_tlen);
3233 } else if (tlen >= save_tlen) {
3234 /* Update of sackblks. */
3235 tcp_update_dsack_list(tp, save_start,
3236 save_start + save_tlen);
3237 } else if (tlen > 0) {
3238 tcp_update_dsack_list(tp, save_start,
3242 tcp_handle_wakeup(tp, so);
3245 * Note the amount of data that peer has sent into
3246 * our window, in order to estimate the sender's
3250 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3251 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3253 len = so->so_rcv.sb_hiwat;
3261 * If FIN is received ACK the FIN and let the user know
3262 * that the connection is closing.
3264 if (thflags & TH_FIN) {
3265 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3266 /* The socket upcall is handled by socantrcvmore. */
3269 * If connection is half-synchronized
3270 * (ie NEEDSYN flag on) then delay ACK,
3271 * so it may be piggybacked when SYN is sent.
3272 * Otherwise, since we received a FIN then no
3273 * more input can be expected, send ACK now.
3275 if (tp->t_flags & TF_NEEDSYN)
3276 tp->t_flags |= TF_DELACK;
3278 tp->t_flags |= TF_ACKNOW;
3281 switch (tp->t_state) {
3283 * In SYN_RECEIVED and ESTABLISHED STATES
3284 * enter the CLOSE_WAIT state.
3286 case TCPS_SYN_RECEIVED:
3287 tp->t_starttime = ticks;
3289 case TCPS_ESTABLISHED:
3290 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3294 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3295 * enter the CLOSING state.
3297 case TCPS_FIN_WAIT_1:
3298 tcp_state_change(tp, TCPS_CLOSING);
3302 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3303 * starting the time-wait timer, turning off the other
3306 case TCPS_FIN_WAIT_2:
3312 if (so->so_options & SO_DEBUG)
3313 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3316 TCP_PROBE3(debug__input, tp, th, m);
3319 * Return any desired output.
3321 if (needoutput || (tp->t_flags & TF_ACKNOW))
3322 (void) tp->t_fb->tfb_tcp_output(tp);
3325 INP_WLOCK_ASSERT(tp->t_inpcb);
3327 if (tp->t_flags & TF_DELACK) {
3328 tp->t_flags &= ~TF_DELACK;
3329 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3331 INP_WUNLOCK(tp->t_inpcb);
3336 * Generate an ACK dropping incoming segment if it occupies
3337 * sequence space, where the ACK reflects our state.
3339 * We can now skip the test for the RST flag since all
3340 * paths to this code happen after packets containing
3341 * RST have been dropped.
3343 * In the SYN-RECEIVED state, don't send an ACK unless the
3344 * segment we received passes the SYN-RECEIVED ACK test.
3345 * If it fails send a RST. This breaks the loop in the
3346 * "LAND" DoS attack, and also prevents an ACK storm
3347 * between two listening ports that have been sent forged
3348 * SYN segments, each with the source address of the other.
3350 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3351 (SEQ_GT(tp->snd_una, th->th_ack) ||
3352 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3353 rstreason = BANDLIM_RST_OPENPORT;
3357 if (so->so_options & SO_DEBUG)
3358 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3361 TCP_PROBE3(debug__input, tp, th, m);
3362 tp->t_flags |= TF_ACKNOW;
3363 (void) tp->t_fb->tfb_tcp_output(tp);
3364 INP_WUNLOCK(tp->t_inpcb);
3370 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3371 INP_WUNLOCK(tp->t_inpcb);
3373 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3378 * Drop space held by incoming segment and return.
3381 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3382 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3385 TCP_PROBE3(debug__input, tp, th, m);
3387 INP_WUNLOCK(tp->t_inpcb);
3393 * Issue RST and make ACK acceptable to originator of segment.
3394 * The mbuf must still include the original packet header.
3398 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3399 int tlen, int rstreason)
3405 struct ip6_hdr *ip6;
3409 INP_WLOCK_ASSERT(tp->t_inpcb);
3412 /* Don't bother if destination was broadcast/multicast. */
3413 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3416 if (mtod(m, struct ip *)->ip_v == 6) {
3417 ip6 = mtod(m, struct ip6_hdr *);
3418 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3419 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3421 /* IPv6 anycast check is done at tcp6_input() */
3424 #if defined(INET) && defined(INET6)
3429 ip = mtod(m, struct ip *);
3430 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3431 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3432 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3433 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3438 /* Perform bandwidth limiting. */
3439 if (badport_bandlim(rstreason) < 0)
3442 /* tcp_respond consumes the mbuf chain. */
3443 if (th->th_flags & TH_ACK) {
3444 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3445 th->th_ack, TH_RST);
3447 if (th->th_flags & TH_SYN)
3449 if (th->th_flags & TH_FIN)
3451 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3452 (tcp_seq)0, TH_RST|TH_ACK);
3460 * Parse TCP options and place in tcpopt.
3463 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3468 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3470 if (opt == TCPOPT_EOL)
3472 if (opt == TCPOPT_NOP)
3478 if (optlen < 2 || optlen > cnt)
3483 if (optlen != TCPOLEN_MAXSEG)
3485 if (!(flags & TO_SYN))
3487 to->to_flags |= TOF_MSS;
3488 bcopy((char *)cp + 2,
3489 (char *)&to->to_mss, sizeof(to->to_mss));
3490 to->to_mss = ntohs(to->to_mss);
3493 if (optlen != TCPOLEN_WINDOW)
3495 if (!(flags & TO_SYN))
3497 to->to_flags |= TOF_SCALE;
3498 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3500 case TCPOPT_TIMESTAMP:
3501 if (optlen != TCPOLEN_TIMESTAMP)
3503 to->to_flags |= TOF_TS;
3504 bcopy((char *)cp + 2,
3505 (char *)&to->to_tsval, sizeof(to->to_tsval));
3506 to->to_tsval = ntohl(to->to_tsval);
3507 bcopy((char *)cp + 6,
3508 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3509 to->to_tsecr = ntohl(to->to_tsecr);
3511 case TCPOPT_SIGNATURE:
3513 * In order to reply to a host which has set the
3514 * TCP_SIGNATURE option in its initial SYN, we have
3515 * to record the fact that the option was observed
3516 * here for the syncache code to perform the correct
3519 if (optlen != TCPOLEN_SIGNATURE)
3521 to->to_flags |= TOF_SIGNATURE;
3522 to->to_signature = cp + 2;
3524 case TCPOPT_SACK_PERMITTED:
3525 if (optlen != TCPOLEN_SACK_PERMITTED)
3527 if (!(flags & TO_SYN))
3531 to->to_flags |= TOF_SACKPERM;
3534 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3538 to->to_flags |= TOF_SACK;
3539 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3540 to->to_sacks = cp + 2;
3541 TCPSTAT_INC(tcps_sack_rcv_blocks);
3543 case TCPOPT_FAST_OPEN:
3545 * Cookie length validation is performed by the
3546 * server side cookie checking code or the client
3547 * side cookie cache update code.
3549 if (!(flags & TO_SYN))
3551 if (!V_tcp_fastopen_client_enable &&
3552 !V_tcp_fastopen_server_enable)
3554 to->to_flags |= TOF_FASTOPEN;
3555 to->to_tfo_len = optlen - 2;
3556 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3565 * Pull out of band byte out of a segment so
3566 * it doesn't appear in the user's data queue.
3567 * It is still reflected in the segment length for
3568 * sequencing purposes.
3571 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3574 int cnt = off + th->th_urp - 1;
3577 if (m->m_len > cnt) {
3578 char *cp = mtod(m, caddr_t) + cnt;
3579 struct tcpcb *tp = sototcpcb(so);
3581 INP_WLOCK_ASSERT(tp->t_inpcb);
3584 tp->t_oobflags |= TCPOOB_HAVEDATA;
3585 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3587 if (m->m_flags & M_PKTHDR)
3596 panic("tcp_pulloutofband");
3600 * Collect new round-trip time estimate
3601 * and update averages and current timeout.
3604 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3608 INP_WLOCK_ASSERT(tp->t_inpcb);
3610 TCPSTAT_INC(tcps_rttupdated);
3613 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3614 imax(0, rtt * 1000 / hz));
3616 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3618 * srtt is stored as fixed point with 5 bits after the
3619 * binary point (i.e., scaled by 8). The following magic
3620 * is equivalent to the smoothing algorithm in rfc793 with
3621 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3622 * point). Adjust rtt to origin 0.
3624 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3625 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3627 if ((tp->t_srtt += delta) <= 0)
3631 * We accumulate a smoothed rtt variance (actually, a
3632 * smoothed mean difference), then set the retransmit
3633 * timer to smoothed rtt + 4 times the smoothed variance.
3634 * rttvar is stored as fixed point with 4 bits after the
3635 * binary point (scaled by 16). The following is
3636 * equivalent to rfc793 smoothing with an alpha of .75
3637 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3638 * rfc793's wired-in beta.
3642 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3643 if ((tp->t_rttvar += delta) <= 0)
3645 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3646 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3649 * No rtt measurement yet - use the unsmoothed rtt.
3650 * Set the variance to half the rtt (so our first
3651 * retransmit happens at 3*rtt).
3653 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3654 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3655 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3661 * the retransmit should happen at rtt + 4 * rttvar.
3662 * Because of the way we do the smoothing, srtt and rttvar
3663 * will each average +1/2 tick of bias. When we compute
3664 * the retransmit timer, we want 1/2 tick of rounding and
3665 * 1 extra tick because of +-1/2 tick uncertainty in the
3666 * firing of the timer. The bias will give us exactly the
3667 * 1.5 tick we need. But, because the bias is
3668 * statistical, we have to test that we don't drop below
3669 * the minimum feasible timer (which is 2 ticks).
3671 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3672 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3675 * We received an ack for a packet that wasn't retransmitted;
3676 * it is probably safe to discard any error indications we've
3677 * received recently. This isn't quite right, but close enough
3678 * for now (a route might have failed after we sent a segment,
3679 * and the return path might not be symmetrical).
3681 tp->t_softerror = 0;
3685 * Determine a reasonable value for maxseg size.
3686 * If the route is known, check route for mtu.
3687 * If none, use an mss that can be handled on the outgoing interface
3688 * without forcing IP to fragment. If no route is found, route has no mtu,
3689 * or the destination isn't local, use a default, hopefully conservative
3690 * size (usually 512 or the default IP max size, but no more than the mtu
3691 * of the interface), as we can't discover anything about intervening
3692 * gateways or networks. We also initialize the congestion/slow start
3693 * window to be a single segment if the destination isn't local.
3694 * While looking at the routing entry, we also initialize other path-dependent
3695 * parameters from pre-set or cached values in the routing entry.
3697 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3698 * IP options, e.g. IPSEC data, since length of this data may vary, and
3699 * thus it is calculated for every segment separately in tcp_output().
3701 * NOTE that this routine is only called when we process an incoming
3702 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3703 * settings are handled in tcp_mssopt().
3706 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3707 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3710 uint32_t maxmtu = 0;
3711 struct inpcb *inp = tp->t_inpcb;
3712 struct hc_metrics_lite metrics;
3714 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3715 size_t min_protoh = isipv6 ?
3716 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3717 sizeof (struct tcpiphdr);
3719 size_t min_protoh = sizeof(struct tcpiphdr);
3722 INP_WLOCK_ASSERT(tp->t_inpcb);
3725 min_protoh += V_tcp_udp_tunneling_overhead;
3726 if (mtuoffer != -1) {
3727 KASSERT(offer == -1, ("%s: conflict", __func__));
3728 offer = mtuoffer - min_protoh;
3734 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3735 tp->t_maxseg = V_tcp_v6mssdflt;
3738 #if defined(INET) && defined(INET6)
3743 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3744 tp->t_maxseg = V_tcp_mssdflt;
3749 * No route to sender, stay with default mss and return.
3753 * In case we return early we need to initialize metrics
3754 * to a defined state as tcp_hc_get() would do for us
3755 * if there was no cache hit.
3757 if (metricptr != NULL)
3758 bzero(metricptr, sizeof(struct hc_metrics_lite));
3762 /* What have we got? */
3766 * Offer == 0 means that there was no MSS on the SYN
3767 * segment, in this case we use tcp_mssdflt as
3768 * already assigned to t_maxseg above.
3770 offer = tp->t_maxseg;
3775 * Offer == -1 means that we didn't receive SYN yet.
3781 * Prevent DoS attack with too small MSS. Round up
3782 * to at least minmss.
3784 offer = max(offer, V_tcp_minmss);
3788 * rmx information is now retrieved from tcp_hostcache.
3790 tcp_hc_get(&inp->inp_inc, &metrics);
3791 if (metricptr != NULL)
3792 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3795 * If there's a discovered mtu in tcp hostcache, use it.
3796 * Else, use the link mtu.
3798 if (metrics.rmx_mtu)
3799 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3803 mss = maxmtu - min_protoh;
3804 if (!V_path_mtu_discovery &&
3805 !in6_localaddr(&inp->in6p_faddr))
3806 mss = min(mss, V_tcp_v6mssdflt);
3809 #if defined(INET) && defined(INET6)
3814 mss = maxmtu - min_protoh;
3815 if (!V_path_mtu_discovery &&
3816 !in_localaddr(inp->inp_faddr))
3817 mss = min(mss, V_tcp_mssdflt);
3821 * XXX - The above conditional (mss = maxmtu - min_protoh)
3822 * probably violates the TCP spec.
3823 * The problem is that, since we don't know the
3824 * other end's MSS, we are supposed to use a conservative
3825 * default. But, if we do that, then MTU discovery will
3826 * never actually take place, because the conservative
3827 * default is much less than the MTUs typically seen
3828 * on the Internet today. For the moment, we'll sweep
3829 * this under the carpet.
3831 * The conservative default might not actually be a problem
3832 * if the only case this occurs is when sending an initial
3833 * SYN with options and data to a host we've never talked
3834 * to before. Then, they will reply with an MSS value which
3835 * will get recorded and the new parameters should get
3836 * recomputed. For Further Study.
3839 mss = min(mss, offer);
3842 * Sanity check: make sure that maxseg will be large
3843 * enough to allow some data on segments even if the
3844 * all the option space is used (40bytes). Otherwise
3845 * funny things may happen in tcp_output.
3847 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3855 tcp_mss(struct tcpcb *tp, int offer)
3861 struct hc_metrics_lite metrics;
3862 struct tcp_ifcap cap;
3864 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3866 bzero(&cap, sizeof(cap));
3867 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3873 * If there's a pipesize, change the socket buffer to that size,
3874 * don't change if sb_hiwat is different than default (then it
3875 * has been changed on purpose with setsockopt).
3876 * Make the socket buffers an integral number of mss units;
3877 * if the mss is larger than the socket buffer, decrease the mss.
3879 so = inp->inp_socket;
3880 SOCKBUF_LOCK(&so->so_snd);
3881 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3882 bufsize = metrics.rmx_sendpipe;
3884 bufsize = so->so_snd.sb_hiwat;
3888 bufsize = roundup(bufsize, mss);
3889 if (bufsize > sb_max)
3891 if (bufsize > so->so_snd.sb_hiwat)
3892 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3894 SOCKBUF_UNLOCK(&so->so_snd);
3896 * Sanity check: make sure that maxseg will be large
3897 * enough to allow some data on segments even if the
3898 * all the option space is used (40bytes). Otherwise
3899 * funny things may happen in tcp_output.
3901 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3903 tp->t_maxseg = max(mss, 64);
3905 SOCKBUF_LOCK(&so->so_rcv);
3906 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3907 bufsize = metrics.rmx_recvpipe;
3909 bufsize = so->so_rcv.sb_hiwat;
3910 if (bufsize > mss) {
3911 bufsize = roundup(bufsize, mss);
3912 if (bufsize > sb_max)
3914 if (bufsize > so->so_rcv.sb_hiwat)
3915 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3917 SOCKBUF_UNLOCK(&so->so_rcv);
3919 /* Check the interface for TSO capabilities. */
3920 if (cap.ifcap & CSUM_TSO) {
3921 tp->t_flags |= TF_TSO;
3922 tp->t_tsomax = cap.tsomax;
3923 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3924 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3929 * Determine the MSS option to send on an outgoing SYN.
3932 tcp_mssopt(struct in_conninfo *inc)
3935 uint32_t thcmtu = 0;
3936 uint32_t maxmtu = 0;
3939 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3942 if (inc->inc_flags & INC_ISIPV6) {
3943 mss = V_tcp_v6mssdflt;
3944 maxmtu = tcp_maxmtu6(inc, NULL);
3945 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3948 #if defined(INET) && defined(INET6)
3953 mss = V_tcp_mssdflt;
3954 maxmtu = tcp_maxmtu(inc, NULL);
3955 min_protoh = sizeof(struct tcpiphdr);
3958 #if defined(INET6) || defined(INET)
3959 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3962 if (maxmtu && thcmtu)
3963 mss = min(maxmtu, thcmtu) - min_protoh;
3964 else if (maxmtu || thcmtu)
3965 mss = max(maxmtu, thcmtu) - min_protoh;
3971 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
3973 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3974 int maxseg = tcp_maxseg(tp);
3976 INP_WLOCK_ASSERT(tp->t_inpcb);
3979 * Compute the amount of data that this ACK is indicating
3980 * (del_data) and an estimate of how many bytes are in the
3983 if (tcp_is_sack_recovery(tp, to) ||
3984 (IN_CONGRECOVERY(tp->t_flags) &&
3985 !IN_FASTRECOVERY(tp->t_flags))) {
3986 del_data = tp->sackhint.delivered_data;
3987 if (V_tcp_do_rfc6675_pipe)
3988 pipe = tcp_compute_pipe(tp);
3990 pipe = (tp->snd_nxt - tp->snd_fack) +
3991 tp->sackhint.sack_bytes_rexmit;
3993 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg +
3994 tp->snd_recover - tp->snd_una))
3996 pipe = imax(0, tp->snd_max - tp->snd_una -
3997 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg);
3999 tp->sackhint.prr_delivered += del_data;
4001 * Proportional Rate Reduction
4003 if (pipe >= tp->snd_ssthresh) {
4004 if (tp->sackhint.recover_fs == 0)
4005 tp->sackhint.recover_fs =
4006 imax(1, tp->snd_nxt - tp->snd_una);
4007 snd_cnt = howmany((long)tp->sackhint.prr_delivered *
4008 tp->snd_ssthresh, tp->sackhint.recover_fs) -
4009 tp->sackhint.prr_out;
4011 if (V_tcp_do_prr_conservative || (del_data == 0))
4012 limit = tp->sackhint.prr_delivered -
4013 tp->sackhint.prr_out;
4015 limit = imax(tp->sackhint.prr_delivered -
4016 tp->sackhint.prr_out, del_data) +
4018 snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
4020 snd_cnt = imax(snd_cnt, 0) / maxseg;
4022 * Send snd_cnt new data into the network in response to this ack.
4023 * If there is going to be a SACK retransmission, adjust snd_cwnd
4026 if (IN_FASTRECOVERY(tp->t_flags)) {
4027 if (tcp_is_sack_recovery(tp, to)) {
4028 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
4029 tp->sackhint.sack_bytes_rexmit +
4032 tp->snd_cwnd = (tp->snd_max - tp->snd_una) +
4035 } else if (IN_CONGRECOVERY(tp->t_flags))
4036 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg);
4037 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd);
4041 * On a partial ack arrives, force the retransmission of the
4042 * next unacknowledged segment. Do not clear tp->t_dupacks.
4043 * By setting snd_nxt to ti_ack, this forces retransmission timer to
4047 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4049 tcp_seq onxt = tp->snd_nxt;
4050 uint32_t ocwnd = tp->snd_cwnd;
4051 u_int maxseg = tcp_maxseg(tp);
4053 INP_WLOCK_ASSERT(tp->t_inpcb);
4055 tcp_timer_activate(tp, TT_REXMT, 0);
4057 tp->snd_nxt = th->th_ack;
4059 * Set snd_cwnd to one segment beyond acknowledged offset.
4060 * (tp->snd_una has not yet been updated when this function is called.)
4062 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4063 tp->t_flags |= TF_ACKNOW;
4064 (void) tp->t_fb->tfb_tcp_output(tp);
4065 tp->snd_cwnd = ocwnd;
4066 if (SEQ_GT(onxt, tp->snd_nxt))
4069 * Partial window deflation. Relies on fact that tp->snd_una
4072 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4073 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4076 tp->snd_cwnd += maxseg;
4080 tcp_compute_pipe(struct tcpcb *tp)
4082 return (tp->snd_max - tp->snd_una +
4083 tp->sackhint.sack_bytes_rexmit -
4084 tp->sackhint.sacked_bytes);
4088 tcp_compute_initwnd(uint32_t maxseg)
4091 * Calculate the Initial Window, also used as Restart Window
4093 * RFC5681 Section 3.1 specifies the default conservative values.
4094 * RFC3390 specifies slightly more aggressive values.
4095 * RFC6928 increases it to ten segments.
4096 * Support for user specified value for initial flight size.
4098 if (V_tcp_initcwnd_segments)
4099 return min(V_tcp_initcwnd_segments * maxseg,
4100 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4101 else if (V_tcp_do_rfc3390)
4102 return min(4 * maxseg, max(2 * maxseg, 4380));
4104 /* Per RFC5681 Section 3.1 */
4106 return (2 * maxseg);
4107 else if (maxseg > 1095)
4108 return (3 * maxseg);
4110 return (4 * maxseg);