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);
292 * Wrapper for the TCP established input helper hook.
295 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
297 struct tcp_hhook_data hhook_data;
299 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
304 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
311 * CC wrapper hook functions
314 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
321 INP_WLOCK_ASSERT(tp->t_inpcb);
323 tp->ccv->nsegs = nsegs;
324 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
325 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
326 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
327 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
328 tp->ccv->flags |= CCF_CWND_LIMITED;
330 tp->ccv->flags &= ~CCF_CWND_LIMITED;
332 if (type == CC_ACK) {
334 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
335 ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
336 if (!IN_RECOVERY(tp->t_flags))
337 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
338 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
339 if ((tp->t_flags & TF_GPUTINPROG) &&
340 SEQ_GEQ(th->th_ack, tp->gput_ack)) {
342 * Compute goodput in bits per millisecond.
344 gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) /
345 max(1, tcp_ts_getticks() - tp->gput_ts);
346 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
349 * XXXLAS: This is a temporary hack, and should be
350 * chained off VOI_TCP_GPUT when stats(9) grows an API
351 * to deal with chained VOIs.
353 if (tp->t_stats_gput_prev > 0)
354 stats_voi_update_abs_s32(tp->t_stats,
356 ((gput - tp->t_stats_gput_prev) * 100) /
357 tp->t_stats_gput_prev);
358 tp->t_flags &= ~TF_GPUTINPROG;
359 tp->t_stats_gput_prev = gput;
362 if (tp->snd_cwnd > tp->snd_ssthresh) {
363 tp->t_bytes_acked += tp->ccv->bytes_this_ack;
364 if (tp->t_bytes_acked >= tp->snd_cwnd) {
365 tp->t_bytes_acked -= tp->snd_cwnd;
366 tp->ccv->flags |= CCF_ABC_SENTAWND;
369 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
370 tp->t_bytes_acked = 0;
374 if (CC_ALGO(tp)->ack_received != NULL) {
375 /* XXXLAS: Find a way to live without this */
376 tp->ccv->curack = th->th_ack;
377 CC_ALGO(tp)->ack_received(tp->ccv, type);
380 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
385 cc_conn_init(struct tcpcb *tp)
387 struct hc_metrics_lite metrics;
388 struct inpcb *inp = tp->t_inpcb;
392 INP_WLOCK_ASSERT(tp->t_inpcb);
394 tcp_hc_get(&inp->inp_inc, &metrics);
395 maxseg = tcp_maxseg(tp);
397 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
399 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
400 TCPSTAT_INC(tcps_usedrtt);
401 if (metrics.rmx_rttvar) {
402 tp->t_rttvar = metrics.rmx_rttvar;
403 TCPSTAT_INC(tcps_usedrttvar);
405 /* default variation is +- 1 rtt */
407 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
409 TCPT_RANGESET(tp->t_rxtcur,
410 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
411 tp->t_rttmin, TCPTV_REXMTMAX);
413 if (metrics.rmx_ssthresh) {
415 * There's some sort of gateway or interface
416 * buffer limit on the path. Use this to set
417 * the slow start threshold, but set the
418 * threshold to no less than 2*mss.
420 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
421 TCPSTAT_INC(tcps_usedssthresh);
425 * Set the initial slow-start flight size.
427 * If a SYN or SYN/ACK was lost and retransmitted, we have to
428 * reduce the initial CWND to one segment as congestion is likely
429 * requiring us to be cautious.
431 if (tp->snd_cwnd == 1)
432 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
434 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
436 if (CC_ALGO(tp)->conn_init != NULL)
437 CC_ALGO(tp)->conn_init(tp->ccv);
441 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
443 INP_WLOCK_ASSERT(tp->t_inpcb);
446 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
451 if (!IN_FASTRECOVERY(tp->t_flags)) {
452 tp->snd_recover = tp->snd_max;
453 if (tp->t_flags2 & TF2_ECN_PERMIT)
454 tp->t_flags2 |= TF2_ECN_SND_CWR;
458 if (!IN_CONGRECOVERY(tp->t_flags) ||
460 * Allow ECN reaction on ACK to CWR, if
461 * that data segment was also CE marked.
463 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
464 EXIT_CONGRECOVERY(tp->t_flags);
465 TCPSTAT_INC(tcps_ecn_rcwnd);
466 tp->snd_recover = tp->snd_max + 1;
467 if (tp->t_flags2 & TF2_ECN_PERMIT)
468 tp->t_flags2 |= TF2_ECN_SND_CWR;
473 tp->t_bytes_acked = 0;
474 EXIT_RECOVERY(tp->t_flags);
475 if (tp->t_flags2 & TF2_ECN_PERMIT)
476 tp->t_flags2 |= TF2_ECN_SND_CWR;
479 TCPSTAT_INC(tcps_sndrexmitbad);
480 /* RTO was unnecessary, so reset everything. */
481 tp->snd_cwnd = tp->snd_cwnd_prev;
482 tp->snd_ssthresh = tp->snd_ssthresh_prev;
483 tp->snd_recover = tp->snd_recover_prev;
484 if (tp->t_flags & TF_WASFRECOVERY)
485 ENTER_FASTRECOVERY(tp->t_flags);
486 if (tp->t_flags & TF_WASCRECOVERY)
487 ENTER_CONGRECOVERY(tp->t_flags);
488 tp->snd_nxt = tp->snd_max;
489 tp->t_flags &= ~TF_PREVVALID;
494 if (CC_ALGO(tp)->cong_signal != NULL) {
496 tp->ccv->curack = th->th_ack;
497 CC_ALGO(tp)->cong_signal(tp->ccv, type);
502 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
504 INP_WLOCK_ASSERT(tp->t_inpcb);
506 /* XXXLAS: KASSERT that we're in recovery? */
508 if (CC_ALGO(tp)->post_recovery != NULL) {
509 tp->ccv->curack = th->th_ack;
510 CC_ALGO(tp)->post_recovery(tp->ccv);
512 /* XXXLAS: EXIT_RECOVERY ? */
513 tp->t_bytes_acked = 0;
514 tp->sackhint.delivered_data = 0;
515 tp->sackhint.prr_out = 0;
519 * Indicate whether this ack should be delayed. We can delay the ack if
520 * following conditions are met:
521 * - There is no delayed ack timer in progress.
522 * - Our last ack wasn't a 0-sized window. We never want to delay
523 * the ack that opens up a 0-sized window.
524 * - LRO wasn't used for this segment. We make sure by checking that the
525 * segment size is not larger than the MSS.
527 #define DELAY_ACK(tp, tlen) \
528 ((!tcp_timer_active(tp, TT_DELACK) && \
529 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
530 (tlen <= tp->t_maxseg) && \
531 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
534 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos)
536 INP_WLOCK_ASSERT(tp->t_inpcb);
538 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
539 switch (iptos & IPTOS_ECN_MASK) {
541 tp->ccv->flags |= CCF_IPHDR_CE;
547 case IPTOS_ECN_NOTECT:
548 tp->ccv->flags &= ~CCF_IPHDR_CE;
553 tp->ccv->flags |= CCF_TCPHDR_CWR;
555 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
557 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
559 if (tp->ccv->flags & CCF_ACKNOW) {
560 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
561 tp->t_flags |= TF_ACKNOW;
567 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
569 cc_ecnpkt_handler_flags(tp, th->th_flags, iptos);
573 * TCP input handling is split into multiple parts:
574 * tcp6_input is a thin wrapper around tcp_input for the extended
575 * ip6_protox[] call format in ip6_input
576 * tcp_input handles primary segment validation, inpcb lookup and
577 * SYN processing on listen sockets
578 * tcp_do_segment processes the ACK and text of the segment for
579 * establishing, established and closing connections
583 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
586 struct in6_ifaddr *ia6;
590 if (m->m_len < *offp + sizeof(struct tcphdr)) {
591 m = m_pullup(m, *offp + sizeof(struct tcphdr));
594 TCPSTAT_INC(tcps_rcvshort);
595 return (IPPROTO_DONE);
600 * draft-itojun-ipv6-tcp-to-anycast
601 * better place to put this in?
603 ip6 = mtod(m, struct ip6_hdr *);
604 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
605 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
606 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
607 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
609 return (IPPROTO_DONE);
613 return (tcp_input_with_port(mp, offp, proto, port));
617 tcp6_input(struct mbuf **mp, int *offp, int proto)
620 return(tcp6_input_with_port(mp, offp, proto, 0));
625 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
627 struct mbuf *m = *mp;
628 struct tcphdr *th = NULL;
629 struct ip *ip = NULL;
630 struct inpcb *inp = NULL;
631 struct tcpcb *tp = NULL;
632 struct socket *so = NULL;
643 int rstreason = 0; /* For badport_bandlim accounting purposes */
645 struct m_tag *fwd_tag = NULL;
647 struct ip6_hdr *ip6 = NULL;
650 const void *ip6 = NULL;
652 struct tcpopt to; /* options in this segment */
653 char *s = NULL; /* address and port logging */
656 * The size of tcp_saveipgen must be the size of the max ip header,
659 u_char tcp_saveipgen[IP6_HDR_LEN];
660 struct tcphdr tcp_savetcp;
667 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
674 TCPSTAT_INC(tcps_rcvtotal);
678 ip6 = mtod(m, struct ip6_hdr *);
679 th = (struct tcphdr *)((caddr_t)ip6 + off0);
680 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
683 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
684 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
685 th->th_sum = m->m_pkthdr.csum_data;
687 th->th_sum = in6_cksum_pseudo(ip6, tlen,
688 IPPROTO_TCP, m->m_pkthdr.csum_data);
689 th->th_sum ^= 0xffff;
691 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
693 TCPSTAT_INC(tcps_rcvbadsum);
698 * Be proactive about unspecified IPv6 address in source.
699 * As we use all-zero to indicate unbounded/unconnected pcb,
700 * unspecified IPv6 address can be used to confuse us.
702 * Note that packets with unspecified IPv6 destination is
703 * already dropped in ip6_input.
705 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
709 iptos = IPV6_TRAFFIC_CLASS(ip6);
712 #if defined(INET) && defined(INET6)
718 * Get IP and TCP header together in first mbuf.
719 * Note: IP leaves IP header in first mbuf.
721 if (off0 > sizeof (struct ip)) {
723 off0 = sizeof(struct ip);
725 if (m->m_len < sizeof (struct tcpiphdr)) {
726 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
728 TCPSTAT_INC(tcps_rcvshort);
729 return (IPPROTO_DONE);
732 ip = mtod(m, struct ip *);
733 th = (struct tcphdr *)((caddr_t)ip + off0);
734 tlen = ntohs(ip->ip_len) - off0;
739 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
740 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
741 th->th_sum = m->m_pkthdr.csum_data;
743 th->th_sum = in_pseudo(ip->ip_src.s_addr,
745 htonl(m->m_pkthdr.csum_data + tlen +
747 th->th_sum ^= 0xffff;
749 struct ipovly *ipov = (struct ipovly *)ip;
752 * Checksum extended TCP header and data.
756 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
757 ipov->ih_len = htons(tlen);
758 th->th_sum = in_cksum(m, len);
759 /* Reset length for SDT probes. */
760 ip->ip_len = htons(len);
763 /* Re-initialization for later version check */
765 ip->ip_v = IPVERSION;
766 ip->ip_hl = off0 >> 2;
769 if (th->th_sum && (port == 0)) {
770 TCPSTAT_INC(tcps_rcvbadsum);
777 * Check that TCP offset makes sense,
778 * pull out TCP options and adjust length. XXX
780 off = th->th_off << 2;
781 if (off < sizeof (struct tcphdr) || off > tlen) {
782 TCPSTAT_INC(tcps_rcvbadoff);
785 tlen -= off; /* tlen is used instead of ti->ti_len */
786 if (off > sizeof (struct tcphdr)) {
789 if (m->m_len < off0 + off) {
790 m = m_pullup(m, off0 + off);
792 TCPSTAT_INC(tcps_rcvshort);
793 return (IPPROTO_DONE);
796 ip6 = mtod(m, struct ip6_hdr *);
797 th = (struct tcphdr *)((caddr_t)ip6 + off0);
800 #if defined(INET) && defined(INET6)
805 if (m->m_len < sizeof(struct ip) + off) {
806 if ((m = m_pullup(m, sizeof (struct ip) + off))
808 TCPSTAT_INC(tcps_rcvshort);
809 return (IPPROTO_DONE);
811 ip = mtod(m, struct ip *);
812 th = (struct tcphdr *)((caddr_t)ip + off0);
816 optlen = off - sizeof (struct tcphdr);
817 optp = (u_char *)(th + 1);
819 thflags = th->th_flags;
822 * Convert TCP protocol specific fields to host format.
824 tcp_fields_to_host(th);
827 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
829 drop_hdrlen = off0 + off;
832 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
836 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
838 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
841 #if defined(INET) && !defined(INET6)
842 (m->m_flags & M_IP_NEXTHOP)
845 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
849 if (isipv6 && fwd_tag != NULL) {
850 struct sockaddr_in6 *next_hop6;
852 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
854 * Transparently forwarded. Pretend to be the destination.
855 * Already got one like this?
857 inp = in6_pcblookup_mbuf(&V_tcbinfo,
858 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
859 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
862 * It's new. Try to find the ambushing socket.
863 * Because we've rewritten the destination address,
864 * any hardware-generated hash is ignored.
866 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
867 th->th_sport, &next_hop6->sin6_addr,
868 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
869 th->th_dport, INPLOOKUP_WILDCARD |
870 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
873 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
874 th->th_sport, &ip6->ip6_dst, th->th_dport,
875 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
876 m->m_pkthdr.rcvif, m);
879 #if defined(INET6) && defined(INET)
883 if (fwd_tag != NULL) {
884 struct sockaddr_in *next_hop;
886 next_hop = (struct sockaddr_in *)(fwd_tag+1);
888 * Transparently forwarded. Pretend to be the destination.
889 * already got one like this?
891 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
892 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
893 m->m_pkthdr.rcvif, m);
896 * It's new. Try to find the ambushing socket.
897 * Because we've rewritten the destination address,
898 * any hardware-generated hash is ignored.
900 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
901 th->th_sport, next_hop->sin_addr,
902 next_hop->sin_port ? ntohs(next_hop->sin_port) :
903 th->th_dport, INPLOOKUP_WILDCARD |
904 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
907 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
908 th->th_sport, ip->ip_dst, th->th_dport,
909 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
910 m->m_pkthdr.rcvif, m);
914 * If the INPCB does not exist then all data in the incoming
915 * segment is discarded and an appropriate RST is sent back.
916 * XXX MRT Send RST using which routing table?
920 * Log communication attempts to ports that are not
923 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
924 V_tcp_log_in_vain == 2) {
925 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
926 log(LOG_INFO, "%s; %s: Connection attempt "
927 "to closed port\n", s, __func__);
930 * When blackholing do not respond with a RST but
931 * completely ignore the segment and drop it.
933 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
937 rstreason = BANDLIM_RST_CLOSEDPORT;
940 INP_WLOCK_ASSERT(inp);
942 * While waiting for inp lock during the lookup, another thread
943 * can have dropped the inpcb, in which case we need to loop back
944 * and try to find a new inpcb to deliver to.
946 if (inp->inp_flags & INP_DROPPED) {
951 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
952 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
953 ((inp->inp_socket == NULL) ||
954 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
955 inp->inp_flowid = m->m_pkthdr.flowid;
956 inp->inp_flowtype = M_HASHTYPE_GET(m);
958 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
960 if (isipv6 && IPSEC_ENABLED(ipv6) &&
961 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
969 if (IPSEC_ENABLED(ipv4) &&
970 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
977 * Check the minimum TTL for socket.
979 if (inp->inp_ip_minttl != 0) {
982 if (inp->inp_ip_minttl > ip6->ip6_hlim)
986 if (inp->inp_ip_minttl > ip->ip_ttl)
991 * A previous connection in TIMEWAIT state is supposed to catch stray
992 * or duplicate segments arriving late. If this segment was a
993 * legitimate new connection attempt, the old INPCB gets removed and
994 * we can try again to find a listening socket.
996 * At this point, due to earlier optimism, we may hold only an inpcb
997 * lock, and not the inpcbinfo write lock. If so, we need to try to
998 * acquire it, or if that fails, acquire a reference on the inpcb,
999 * drop all locks, acquire a global write lock, and then re-acquire
1000 * the inpcb lock. We may at that point discover that another thread
1001 * has tried to free the inpcb, in which case we need to loop back
1002 * and try to find a new inpcb to deliver to.
1004 * XXXRW: It may be time to rethink timewait locking.
1006 if (inp->inp_flags & INP_TIMEWAIT) {
1007 tcp_dooptions(&to, optp, optlen,
1008 (thflags & TH_SYN) ? TO_SYN : 0);
1010 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
1012 if (tcp_twcheck(inp, &to, th, m, tlen))
1014 return (IPPROTO_DONE);
1017 * The TCPCB may no longer exist if the connection is winding
1018 * down or it is in the CLOSED state. Either way we drop the
1019 * segment and send an appropriate response.
1021 tp = intotcpcb(inp);
1022 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1023 rstreason = BANDLIM_RST_CLOSEDPORT;
1027 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) {
1028 rstreason = BANDLIM_RST_CLOSEDPORT;
1033 if (tp->t_flags & TF_TOE) {
1034 tcp_offload_input(tp, m);
1035 m = NULL; /* consumed by the TOE driver */
1041 INP_WLOCK_ASSERT(inp);
1042 if (mac_inpcb_check_deliver(inp, m))
1045 so = inp->inp_socket;
1046 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1048 if (so->so_options & SO_DEBUG) {
1049 ostate = tp->t_state;
1052 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1055 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1058 #endif /* TCPDEBUG */
1060 * When the socket is accepting connections (the INPCB is in LISTEN
1061 * state) we look into the SYN cache if this is a new connection
1062 * attempt or the completion of a previous one.
1064 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1065 ("%s: so accepting but tp %p not listening", __func__, tp));
1066 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1067 struct in_conninfo inc;
1069 bzero(&inc, sizeof(inc));
1072 inc.inc_flags |= INC_ISIPV6;
1073 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1074 inc.inc_flags |= INC_IPV6MINMTU;
1075 inc.inc6_faddr = ip6->ip6_src;
1076 inc.inc6_laddr = ip6->ip6_dst;
1080 inc.inc_faddr = ip->ip_src;
1081 inc.inc_laddr = ip->ip_dst;
1083 inc.inc_fport = th->th_sport;
1084 inc.inc_lport = th->th_dport;
1085 inc.inc_fibnum = so->so_fibnum;
1088 * Check for an existing connection attempt in syncache if
1089 * the flag is only ACK. A successful lookup creates a new
1090 * socket appended to the listen queue in SYN_RECEIVED state.
1092 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1094 * Parse the TCP options here because
1095 * syncookies need access to the reflected
1098 tcp_dooptions(&to, optp, optlen, 0);
1100 * NB: syncache_expand() doesn't unlock
1101 * inp and tcpinfo locks.
1103 rstreason = syncache_expand(&inc, &to, th, &so, m, port);
1104 if (rstreason < 0) {
1106 * A failing TCP MD5 signature comparison
1107 * must result in the segment being dropped
1108 * and must not produce any response back
1112 } else if (rstreason == 0) {
1114 * No syncache entry or ACK was not
1115 * for our SYN/ACK. Send a RST.
1116 * NB: syncache did its own logging
1117 * of the failure cause.
1119 rstreason = BANDLIM_RST_OPENPORT;
1125 * We completed the 3-way handshake
1126 * but could not allocate a socket
1127 * either due to memory shortage,
1128 * listen queue length limits or
1129 * global socket limits. Send RST
1130 * or wait and have the remote end
1131 * retransmit the ACK for another
1134 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1135 log(LOG_DEBUG, "%s; %s: Listen socket: "
1136 "Socket allocation failed due to "
1137 "limits or memory shortage, %s\n",
1139 V_tcp_sc_rst_sock_fail ?
1140 "sending RST" : "try again");
1141 if (V_tcp_sc_rst_sock_fail) {
1142 rstreason = BANDLIM_UNLIMITED;
1148 * Socket is created in state SYN_RECEIVED.
1149 * Unlock the listen socket, lock the newly
1150 * created socket and update the tp variable.
1152 INP_WUNLOCK(inp); /* listen socket */
1153 inp = sotoinpcb(so);
1155 * New connection inpcb is already locked by
1156 * syncache_expand().
1158 INP_WLOCK_ASSERT(inp);
1159 tp = intotcpcb(inp);
1160 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1161 ("%s: ", __func__));
1163 * Process the segment and the data it
1164 * contains. tcp_do_segment() consumes
1165 * the mbuf chain and unlocks the inpcb.
1167 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1168 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1170 return (IPPROTO_DONE);
1173 * Segment flag validation for new connection attempts:
1175 * Our (SYN|ACK) response was rejected.
1176 * Check with syncache and remove entry to prevent
1179 * NB: syncache_chkrst does its own logging of failure
1182 if (thflags & TH_RST) {
1183 syncache_chkrst(&inc, th, m, port);
1187 * We can't do anything without SYN.
1189 if ((thflags & TH_SYN) == 0) {
1190 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1191 log(LOG_DEBUG, "%s; %s: Listen socket: "
1192 "SYN is missing, segment ignored\n",
1194 TCPSTAT_INC(tcps_badsyn);
1198 * (SYN|ACK) is bogus on a listen socket.
1200 if (thflags & TH_ACK) {
1201 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1202 log(LOG_DEBUG, "%s; %s: Listen socket: "
1203 "SYN|ACK invalid, segment rejected\n",
1205 syncache_badack(&inc, port); /* XXX: Not needed! */
1206 TCPSTAT_INC(tcps_badsyn);
1207 rstreason = BANDLIM_RST_OPENPORT;
1211 * If the drop_synfin option is enabled, drop all
1212 * segments with both the SYN and FIN bits set.
1213 * This prevents e.g. nmap from identifying the
1215 * XXX: Poor reasoning. nmap has other methods
1216 * and is constantly refining its stack detection
1218 * XXX: This is a violation of the TCP specification
1219 * and was used by RFC1644.
1221 if ((thflags & TH_FIN) && V_drop_synfin) {
1222 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1223 log(LOG_DEBUG, "%s; %s: Listen socket: "
1224 "SYN|FIN segment ignored (based on "
1225 "sysctl setting)\n", s, __func__);
1226 TCPSTAT_INC(tcps_badsyn);
1230 * Segment's flags are (SYN) or (SYN|FIN).
1232 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1233 * as they do not affect the state of the TCP FSM.
1234 * The data pointed to by TH_URG and th_urp is ignored.
1236 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1237 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1238 KASSERT(thflags & (TH_SYN),
1239 ("%s: Listen socket: TH_SYN not set", __func__));
1242 * If deprecated address is forbidden,
1243 * we do not accept SYN to deprecated interface
1244 * address to prevent any new inbound connection from
1245 * getting established.
1246 * When we do not accept SYN, we send a TCP RST,
1247 * with deprecated source address (instead of dropping
1248 * it). We compromise it as it is much better for peer
1249 * to send a RST, and RST will be the final packet
1252 * If we do not forbid deprecated addresses, we accept
1253 * the SYN packet. RFC2462 does not suggest dropping
1255 * If we decipher RFC2462 5.5.4, it says like this:
1256 * 1. use of deprecated addr with existing
1257 * communication is okay - "SHOULD continue to be
1259 * 2. use of it with new communication:
1260 * (2a) "SHOULD NOT be used if alternate address
1261 * with sufficient scope is available"
1262 * (2b) nothing mentioned otherwise.
1263 * Here we fall into (2b) case as we have no choice in
1264 * our source address selection - we must obey the peer.
1266 * The wording in RFC2462 is confusing, and there are
1267 * multiple description text for deprecated address
1268 * handling - worse, they are not exactly the same.
1269 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1271 if (isipv6 && !V_ip6_use_deprecated) {
1272 struct in6_ifaddr *ia6;
1274 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1276 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1277 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1278 log(LOG_DEBUG, "%s; %s: Listen socket: "
1279 "Connection attempt to deprecated "
1280 "IPv6 address rejected\n",
1282 rstreason = BANDLIM_RST_OPENPORT;
1288 * Basic sanity checks on incoming SYN requests:
1289 * Don't respond if the destination is a link layer
1290 * broadcast according to RFC1122 4.2.3.10, p. 104.
1291 * If it is from this socket it must be forged.
1292 * Don't respond if the source or destination is a
1293 * global or subnet broad- or multicast address.
1294 * Note that it is quite possible to receive unicast
1295 * link-layer packets with a broadcast IP address. Use
1296 * in_broadcast() to find them.
1298 if (m->m_flags & (M_BCAST|M_MCAST)) {
1299 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1300 log(LOG_DEBUG, "%s; %s: Listen socket: "
1301 "Connection attempt from broad- or multicast "
1302 "link layer address ignored\n", s, __func__);
1307 if (th->th_dport == th->th_sport &&
1308 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1309 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1310 log(LOG_DEBUG, "%s; %s: Listen socket: "
1311 "Connection attempt to/from self "
1312 "ignored\n", s, __func__);
1315 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1316 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1317 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1318 log(LOG_DEBUG, "%s; %s: Listen socket: "
1319 "Connection attempt from/to multicast "
1320 "address ignored\n", s, __func__);
1325 #if defined(INET) && defined(INET6)
1330 if (th->th_dport == th->th_sport &&
1331 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1332 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1333 log(LOG_DEBUG, "%s; %s: Listen socket: "
1334 "Connection attempt from/to self "
1335 "ignored\n", s, __func__);
1338 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1339 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1340 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1341 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1342 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1343 log(LOG_DEBUG, "%s; %s: Listen socket: "
1344 "Connection attempt from/to broad- "
1345 "or multicast address ignored\n",
1352 * SYN appears to be valid. Create compressed TCP state
1356 if (so->so_options & SO_DEBUG)
1357 tcp_trace(TA_INPUT, ostate, tp,
1358 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1360 TCP_PROBE3(debug__input, tp, th, m);
1361 tcp_dooptions(&to, optp, optlen, TO_SYN);
1362 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos,
1364 goto tfo_socket_result;
1367 * Entry added to syncache and mbuf consumed.
1368 * Only the listen socket is unlocked by syncache_add().
1370 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1371 return (IPPROTO_DONE);
1372 } else if (tp->t_state == TCPS_LISTEN) {
1374 * When a listen socket is torn down the SO_ACCEPTCONN
1375 * flag is removed first while connections are drained
1376 * from the accept queue in a unlock/lock cycle of the
1377 * ACCEPT_LOCK, opening a race condition allowing a SYN
1378 * attempt go through unhandled.
1382 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1383 if (tp->t_flags & TF_SIGNATURE) {
1384 tcp_dooptions(&to, optp, optlen, thflags);
1385 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1386 TCPSTAT_INC(tcps_sig_err_nosigopt);
1389 if (!TCPMD5_ENABLED() ||
1390 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1394 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1397 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1398 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1399 * the inpcb, and unlocks pcbinfo.
1401 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1402 return (IPPROTO_DONE);
1405 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1408 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1411 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1412 m = NULL; /* mbuf chain got consumed. */
1417 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1423 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1428 return (IPPROTO_DONE);
1432 * Automatic sizing of receive socket buffer. Often the send
1433 * buffer size is not optimally adjusted to the actual network
1434 * conditions at hand (delay bandwidth product). Setting the
1435 * buffer size too small limits throughput on links with high
1436 * bandwidth and high delay (eg. trans-continental/oceanic links).
1438 * On the receive side the socket buffer memory is only rarely
1439 * used to any significant extent. This allows us to be much
1440 * more aggressive in scaling the receive socket buffer. For
1441 * the case that the buffer space is actually used to a large
1442 * extent and we run out of kernel memory we can simply drop
1443 * the new segments; TCP on the sender will just retransmit it
1444 * later. Setting the buffer size too big may only consume too
1445 * much kernel memory if the application doesn't read() from
1446 * the socket or packet loss or reordering makes use of the
1449 * The criteria to step up the receive buffer one notch are:
1450 * 1. Application has not set receive buffer size with
1451 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1452 * 2. the number of bytes received during 1/2 of an sRTT
1453 * is at least 3/8 of the current socket buffer size.
1454 * 3. receive buffer size has not hit maximal automatic size;
1456 * If all of the criteria are met we increaset the socket buffer
1457 * by a 1/2 (bounded by the max). This allows us to keep ahead
1458 * of slow-start but also makes it so our peer never gets limited
1459 * by our rwnd which we then open up causing a burst.
1461 * This algorithm does two steps per RTT at most and only if
1462 * we receive a bulk stream w/o packet losses or reorderings.
1463 * Shrinking the buffer during idle times is not necessary as
1464 * it doesn't consume any memory when idle.
1466 * TODO: Only step up if the application is actually serving
1467 * the buffer to better manage the socket buffer resources.
1470 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1471 struct tcpcb *tp, int tlen)
1475 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1476 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1477 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1478 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1479 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1480 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1481 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1483 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1485 /* Start over with next RTT. */
1489 tp->rfbuf_cnt += tlen; /* add up */
1495 tcp_input(struct mbuf **mp, int *offp, int proto)
1497 return(tcp_input_with_port(mp, offp, proto, 0));
1501 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1504 * Since tp might be gone if the session entered
1505 * the TIME_WAIT state before coming here, we need
1506 * to check if the socket is still connected.
1508 if ((so->so_state & SS_ISCONNECTED) == 0)
1510 INP_LOCK_ASSERT(tp->t_inpcb);
1511 if (tp->t_flags & TF_WAKESOR) {
1512 tp->t_flags &= ~TF_WAKESOR;
1513 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
1516 if (tp->t_flags & TF_WAKESOW) {
1517 tp->t_flags &= ~TF_WAKESOW;
1518 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
1524 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1525 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1527 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1528 int rstreason, todrop, win, incforsyn = 0;
1532 struct in_conninfo *inc;
1540 * The size of tcp_saveipgen must be the size of the max ip header,
1543 u_char tcp_saveipgen[IP6_HDR_LEN];
1544 struct tcphdr tcp_savetcp;
1547 thflags = th->th_flags;
1548 inc = &tp->t_inpcb->inp_inc;
1549 tp->sackhint.last_sack_ack = 0;
1551 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1554 INP_WLOCK_ASSERT(tp->t_inpcb);
1555 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1557 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1561 /* Save segment, if requested. */
1562 tcp_pcap_add(th, m, &(tp->t_inpkts));
1564 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1567 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1568 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1569 log(LOG_DEBUG, "%s; %s: "
1570 "SYN|FIN segment ignored (based on "
1571 "sysctl setting)\n", s, __func__);
1578 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1579 * check SEQ.ACK first.
1581 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1582 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1583 rstreason = BANDLIM_UNLIMITED;
1588 * Segment received on connection.
1589 * Reset idle time and keep-alive timer.
1590 * XXX: This should be done after segment
1591 * validation to ignore broken/spoofed segs.
1593 tp->t_rcvtime = ticks;
1596 * Scale up the window into a 32-bit value.
1597 * For the SYN_SENT state the scale is zero.
1599 tiwin = th->th_win << tp->snd_scale;
1601 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1605 * TCP ECN processing.
1607 if (tp->t_flags2 & TF2_ECN_PERMIT) {
1608 if (thflags & TH_CWR) {
1609 tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1610 tp->t_flags |= TF_ACKNOW;
1612 switch (iptos & IPTOS_ECN_MASK) {
1614 tp->t_flags2 |= TF2_ECN_SND_ECE;
1615 TCPSTAT_INC(tcps_ecn_ce);
1617 case IPTOS_ECN_ECT0:
1618 TCPSTAT_INC(tcps_ecn_ect0);
1620 case IPTOS_ECN_ECT1:
1621 TCPSTAT_INC(tcps_ecn_ect1);
1625 /* Process a packet differently from RFC3168. */
1626 cc_ecnpkt_handler(tp, th, iptos);
1628 /* Congestion experienced. */
1629 if (thflags & TH_ECE) {
1630 cc_cong_signal(tp, th, CC_ECN);
1635 * Parse options on any incoming segment.
1637 tcp_dooptions(&to, (u_char *)(th + 1),
1638 (th->th_off << 2) - sizeof(struct tcphdr),
1639 (thflags & TH_SYN) ? TO_SYN : 0);
1641 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1642 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1643 (to.to_flags & TOF_SIGNATURE) == 0) {
1644 TCPSTAT_INC(tcps_sig_err_sigopt);
1645 /* XXX: should drop? */
1649 * If echoed timestamp is later than the current time,
1650 * fall back to non RFC1323 RTT calculation. Normalize
1651 * timestamp if syncookies were used when this connection
1654 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1655 to.to_tsecr -= tp->ts_offset;
1656 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1658 else if (tp->t_flags & TF_PREVVALID &&
1659 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1660 cc_cong_signal(tp, th, CC_RTO_ERR);
1663 * Process options only when we get SYN/ACK back. The SYN case
1664 * for incoming connections is handled in tcp_syncache.
1665 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1666 * or <SYN,ACK>) segment itself is never scaled.
1667 * XXX this is traditional behavior, may need to be cleaned up.
1669 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1670 /* Handle parallel SYN for ECN */
1671 if (!(thflags & TH_ACK) &&
1672 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1673 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1674 tp->t_flags2 |= TF2_ECN_PERMIT;
1675 tp->t_flags2 |= TF2_ECN_SND_ECE;
1676 TCPSTAT_INC(tcps_ecn_shs);
1678 if ((to.to_flags & TOF_SCALE) &&
1679 (tp->t_flags & TF_REQ_SCALE) &&
1680 !(tp->t_flags & TF_NOOPT)) {
1681 tp->t_flags |= TF_RCVD_SCALE;
1682 tp->snd_scale = to.to_wscale;
1684 tp->t_flags &= ~TF_REQ_SCALE;
1686 * Initial send window. It will be updated with
1687 * the next incoming segment to the scaled value.
1689 tp->snd_wnd = th->th_win;
1690 if ((to.to_flags & TOF_TS) &&
1691 (tp->t_flags & TF_REQ_TSTMP) &&
1692 !(tp->t_flags & TF_NOOPT)) {
1693 tp->t_flags |= TF_RCVD_TSTMP;
1694 tp->ts_recent = to.to_tsval;
1695 tp->ts_recent_age = tcp_ts_getticks();
1697 tp->t_flags &= ~TF_REQ_TSTMP;
1698 if (to.to_flags & TOF_MSS)
1699 tcp_mss(tp, to.to_mss);
1700 if ((tp->t_flags & TF_SACK_PERMIT) &&
1701 (!(to.to_flags & TOF_SACKPERM) ||
1702 (tp->t_flags & TF_NOOPT)))
1703 tp->t_flags &= ~TF_SACK_PERMIT;
1704 if (IS_FASTOPEN(tp->t_flags)) {
1705 if ((to.to_flags & TOF_FASTOPEN) &&
1706 !(tp->t_flags & TF_NOOPT)) {
1709 if (to.to_flags & TOF_MSS)
1712 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1716 tcp_fastopen_update_cache(tp, mss,
1717 to.to_tfo_len, to.to_tfo_cookie);
1719 tcp_fastopen_disable_path(tp);
1724 * If timestamps were negotiated during SYN/ACK and a
1725 * segment without a timestamp is received, silently drop
1726 * the segment, unless it is a RST segment or missing timestamps are
1728 * See section 3.2 of RFC 7323.
1730 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1731 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1732 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1733 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1734 "segment processed normally\n",
1739 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1740 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1741 "segment silently dropped\n", s, __func__);
1748 * If timestamps were not negotiated during SYN/ACK and a
1749 * segment with a timestamp is received, ignore the
1750 * timestamp and process the packet normally.
1751 * See section 3.2 of RFC 7323.
1753 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1754 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1755 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1756 "segment processed normally\n", s, __func__);
1762 * Header prediction: check for the two common cases
1763 * of a uni-directional data xfer. If the packet has
1764 * no control flags, is in-sequence, the window didn't
1765 * change and we're not retransmitting, it's a
1766 * candidate. If the length is zero and the ack moved
1767 * forward, we're the sender side of the xfer. Just
1768 * free the data acked & wake any higher level process
1769 * that was blocked waiting for space. If the length
1770 * is non-zero and the ack didn't move, we're the
1771 * receiver side. If we're getting packets in-order
1772 * (the reassembly queue is empty), add the data to
1773 * the socket buffer and note that we need a delayed ack.
1774 * Make sure that the hidden state-flags are also off.
1775 * Since we check for TCPS_ESTABLISHED first, it can only
1778 if (tp->t_state == TCPS_ESTABLISHED &&
1779 th->th_seq == tp->rcv_nxt &&
1780 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1781 tp->snd_nxt == tp->snd_max &&
1782 tiwin && tiwin == tp->snd_wnd &&
1783 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1785 ((to.to_flags & TOF_TS) == 0 ||
1786 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1788 * If last ACK falls within this segment's sequence numbers,
1789 * record the timestamp.
1790 * NOTE that the test is modified according to the latest
1791 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1793 if ((to.to_flags & TOF_TS) != 0 &&
1794 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1795 tp->ts_recent_age = tcp_ts_getticks();
1796 tp->ts_recent = to.to_tsval;
1800 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1801 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1802 !IN_RECOVERY(tp->t_flags) &&
1803 (to.to_flags & TOF_SACK) == 0 &&
1804 TAILQ_EMPTY(&tp->snd_holes)) {
1806 * This is a pure ack for outstanding data.
1808 TCPSTAT_INC(tcps_predack);
1811 * "bad retransmit" recovery without timestamps.
1813 if ((to.to_flags & TOF_TS) == 0 &&
1814 tp->t_rxtshift == 1 &&
1815 tp->t_flags & TF_PREVVALID &&
1816 (int)(ticks - tp->t_badrxtwin) < 0) {
1817 cc_cong_signal(tp, th, CC_RTO_ERR);
1821 * Recalculate the transmit timer / rtt.
1823 * Some boxes send broken timestamp replies
1824 * during the SYN+ACK phase, ignore
1825 * timestamps of 0 or we could calculate a
1826 * huge RTT and blow up the retransmit timer.
1828 if ((to.to_flags & TOF_TS) != 0 &&
1832 t = tcp_ts_getticks() - to.to_tsecr;
1833 if (!tp->t_rttlow || tp->t_rttlow > t)
1836 TCP_TS_TO_TICKS(t) + 1);
1837 } else if (tp->t_rtttime &&
1838 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1839 if (!tp->t_rttlow ||
1840 tp->t_rttlow > ticks - tp->t_rtttime)
1841 tp->t_rttlow = ticks - tp->t_rtttime;
1843 ticks - tp->t_rtttime);
1845 acked = BYTES_THIS_ACK(tp, th);
1848 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1849 hhook_run_tcp_est_in(tp, th, &to);
1852 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1853 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1854 sbdrop(&so->so_snd, acked);
1855 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1856 SEQ_LEQ(th->th_ack, tp->snd_recover))
1857 tp->snd_recover = th->th_ack - 1;
1860 * Let the congestion control algorithm update
1861 * congestion control related information. This
1862 * typically means increasing the congestion
1865 cc_ack_received(tp, th, nsegs, CC_ACK);
1867 tp->snd_una = th->th_ack;
1869 * Pull snd_wl2 up to prevent seq wrap relative
1872 tp->snd_wl2 = th->th_ack;
1877 * If all outstanding data are acked, stop
1878 * retransmit timer, otherwise restart timer
1879 * using current (possibly backed-off) value.
1880 * If process is waiting for space,
1881 * wakeup/selwakeup/signal. If data
1882 * are ready to send, let tcp_output
1883 * decide between more output or persist.
1886 if (so->so_options & SO_DEBUG)
1887 tcp_trace(TA_INPUT, ostate, tp,
1888 (void *)tcp_saveipgen,
1891 TCP_PROBE3(debug__input, tp, th, m);
1892 if (tp->snd_una == tp->snd_max)
1893 tcp_timer_activate(tp, TT_REXMT, 0);
1894 else if (!tcp_timer_active(tp, TT_PERSIST))
1895 tcp_timer_activate(tp, TT_REXMT,
1897 tp->t_flags |= TF_WAKESOW;
1898 if (sbavail(&so->so_snd))
1899 (void) tp->t_fb->tfb_tcp_output(tp);
1902 } else if (th->th_ack == tp->snd_una &&
1903 tlen <= sbspace(&so->so_rcv)) {
1904 int newsize = 0; /* automatic sockbuf scaling */
1907 * This is a pure, in-sequence data packet with
1908 * nothing on the reassembly queue and we have enough
1909 * buffer space to take it.
1911 /* Clean receiver SACK report if present */
1912 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1913 tcp_clean_sackreport(tp);
1914 TCPSTAT_INC(tcps_preddat);
1915 tp->rcv_nxt += tlen;
1917 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1918 (tp->t_fbyte_in == 0)) {
1919 tp->t_fbyte_in = ticks;
1920 if (tp->t_fbyte_in == 0)
1922 if (tp->t_fbyte_out && tp->t_fbyte_in)
1923 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1926 * Pull snd_wl1 up to prevent seq wrap relative to
1929 tp->snd_wl1 = th->th_seq;
1931 * Pull rcv_up up to prevent seq wrap relative to
1934 tp->rcv_up = tp->rcv_nxt;
1935 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1936 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1938 if (so->so_options & SO_DEBUG)
1939 tcp_trace(TA_INPUT, ostate, tp,
1940 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1942 TCP_PROBE3(debug__input, tp, th, m);
1944 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1946 /* Add data to socket buffer. */
1947 SOCKBUF_LOCK(&so->so_rcv);
1948 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1952 * Set new socket buffer size.
1953 * Give up when limit is reached.
1956 if (!sbreserve_locked(&so->so_rcv,
1958 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1959 m_adj(m, drop_hdrlen); /* delayed header drop */
1960 sbappendstream_locked(&so->so_rcv, m, 0);
1962 SOCKBUF_UNLOCK(&so->so_rcv);
1963 tp->t_flags |= TF_WAKESOR;
1964 if (DELAY_ACK(tp, tlen)) {
1965 tp->t_flags |= TF_DELACK;
1967 tp->t_flags |= TF_ACKNOW;
1968 tp->t_fb->tfb_tcp_output(tp);
1975 * Calculate amount of space in receive window,
1976 * and then do TCP input processing.
1977 * Receive window is amount of space in rcv queue,
1978 * but not less than advertised window.
1980 win = sbspace(&so->so_rcv);
1983 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1985 switch (tp->t_state) {
1987 * If the state is SYN_RECEIVED:
1988 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1990 case TCPS_SYN_RECEIVED:
1991 if ((thflags & TH_ACK) &&
1992 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1993 SEQ_GT(th->th_ack, tp->snd_max))) {
1994 rstreason = BANDLIM_RST_OPENPORT;
1997 if (IS_FASTOPEN(tp->t_flags)) {
1999 * When a TFO connection is in SYN_RECEIVED, the
2000 * only valid packets are the initial SYN, a
2001 * retransmit/copy of the initial SYN (possibly with
2002 * a subset of the original data), a valid ACK, a
2005 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
2006 rstreason = BANDLIM_RST_OPENPORT;
2008 } else if (thflags & TH_SYN) {
2009 /* non-initial SYN is ignored */
2010 if ((tcp_timer_active(tp, TT_DELACK) ||
2011 tcp_timer_active(tp, TT_REXMT)))
2013 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2020 * If the state is SYN_SENT:
2021 * if seg contains a RST with valid ACK (SEQ.ACK has already
2022 * been verified), then drop the connection.
2023 * if seg contains a RST without an ACK, drop the seg.
2024 * if seg does not contain SYN, then drop the seg.
2025 * Otherwise this is an acceptable SYN segment
2026 * initialize tp->rcv_nxt and tp->irs
2027 * if seg contains ack then advance tp->snd_una
2028 * if seg contains an ECE and ECN support is enabled, the stream
2030 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2031 * arrange for segment to be acked (eventually)
2032 * continue processing rest of data/controls, beginning with URG
2035 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2036 TCP_PROBE5(connect__refused, NULL, tp,
2038 tp = tcp_drop(tp, ECONNREFUSED);
2040 if (thflags & TH_RST)
2042 if (!(thflags & TH_SYN))
2045 tp->irs = th->th_seq;
2047 if (thflags & TH_ACK) {
2048 int tfo_partial_ack = 0;
2050 TCPSTAT_INC(tcps_connects);
2053 mac_socketpeer_set_from_mbuf(m, so);
2055 /* Do window scaling on this connection? */
2056 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2057 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2058 tp->rcv_scale = tp->request_r_scale;
2060 tp->rcv_adv += min(tp->rcv_wnd,
2061 TCP_MAXWIN << tp->rcv_scale);
2062 tp->snd_una++; /* SYN is acked */
2064 * If not all the data that was sent in the TFO SYN
2065 * has been acked, resend the remainder right away.
2067 if (IS_FASTOPEN(tp->t_flags) &&
2068 (tp->snd_una != tp->snd_max)) {
2069 tp->snd_nxt = th->th_ack;
2070 tfo_partial_ack = 1;
2073 * If there's data, delay ACK; if there's also a FIN
2074 * ACKNOW will be turned on later.
2076 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2077 tcp_timer_activate(tp, TT_DELACK,
2080 tp->t_flags |= TF_ACKNOW;
2082 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2083 (V_tcp_do_ecn == 1)) {
2084 tp->t_flags2 |= TF2_ECN_PERMIT;
2085 TCPSTAT_INC(tcps_ecn_shs);
2089 * Received <SYN,ACK> in SYN_SENT[*] state.
2091 * SYN_SENT --> ESTABLISHED
2092 * SYN_SENT* --> FIN_WAIT_1
2094 tp->t_starttime = ticks;
2095 if (tp->t_flags & TF_NEEDFIN) {
2096 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2097 tp->t_flags &= ~TF_NEEDFIN;
2100 tcp_state_change(tp, TCPS_ESTABLISHED);
2101 TCP_PROBE5(connect__established, NULL, tp,
2104 tcp_timer_activate(tp, TT_KEEP,
2109 * Received initial SYN in SYN-SENT[*] state =>
2110 * simultaneous open.
2111 * If it succeeds, connection is * half-synchronized.
2112 * Otherwise, do 3-way handshake:
2113 * SYN-SENT -> SYN-RECEIVED
2114 * SYN-SENT* -> SYN-RECEIVED*
2116 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2117 tcp_timer_activate(tp, TT_REXMT, 0);
2118 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2121 INP_WLOCK_ASSERT(tp->t_inpcb);
2124 * Advance th->th_seq to correspond to first data byte.
2125 * If data, trim to stay within window,
2126 * dropping FIN if necessary.
2129 if (tlen > tp->rcv_wnd) {
2130 todrop = tlen - tp->rcv_wnd;
2134 TCPSTAT_INC(tcps_rcvpackafterwin);
2135 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2137 tp->snd_wl1 = th->th_seq - 1;
2138 tp->rcv_up = th->th_seq;
2140 * Client side of transaction: already sent SYN and data.
2141 * If the remote host used T/TCP to validate the SYN,
2142 * our data will be ACK'd; if so, enter normal data segment
2143 * processing in the middle of step 5, ack processing.
2144 * Otherwise, goto step 6.
2146 if (thflags & TH_ACK)
2152 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2153 * do normal processing.
2155 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2159 break; /* continue normal processing */
2163 * States other than LISTEN or SYN_SENT.
2164 * First check the RST flag and sequence number since reset segments
2165 * are exempt from the timestamp and connection count tests. This
2166 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2167 * below which allowed reset segments in half the sequence space
2168 * to fall though and be processed (which gives forged reset
2169 * segments with a random sequence number a 50 percent chance of
2170 * killing a connection).
2171 * Then check timestamp, if present.
2172 * Then check the connection count, if present.
2173 * Then check that at least some bytes of segment are within
2174 * receive window. If segment begins before rcv_nxt,
2175 * drop leading data (and SYN); if nothing left, just ack.
2177 if (thflags & TH_RST) {
2179 * RFC5961 Section 3.2
2181 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2182 * - If RST is in window, we send challenge ACK.
2184 * Note: to take into account delayed ACKs, we should
2185 * test against last_ack_sent instead of rcv_nxt.
2186 * Note 2: we handle special case of closed window, not
2187 * covered by the RFC.
2189 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2190 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2191 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2192 KASSERT(tp->t_state != TCPS_SYN_SENT,
2193 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2196 if (V_tcp_insecure_rst ||
2197 tp->last_ack_sent == th->th_seq) {
2198 TCPSTAT_INC(tcps_drops);
2199 /* Drop the connection. */
2200 switch (tp->t_state) {
2201 case TCPS_SYN_RECEIVED:
2202 so->so_error = ECONNREFUSED;
2204 case TCPS_ESTABLISHED:
2205 case TCPS_FIN_WAIT_1:
2206 case TCPS_FIN_WAIT_2:
2207 case TCPS_CLOSE_WAIT:
2210 so->so_error = ECONNRESET;
2217 TCPSTAT_INC(tcps_badrst);
2218 /* Send challenge ACK. */
2219 tcp_respond(tp, mtod(m, void *), th, m,
2220 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2221 tp->last_ack_sent = tp->rcv_nxt;
2229 * RFC5961 Section 4.2
2230 * Send challenge ACK for any SYN in synchronized state.
2232 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2233 tp->t_state != TCPS_SYN_RECEIVED) {
2234 TCPSTAT_INC(tcps_badsyn);
2235 if (V_tcp_insecure_syn &&
2236 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2237 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2238 tp = tcp_drop(tp, ECONNRESET);
2239 rstreason = BANDLIM_UNLIMITED;
2241 /* Send challenge ACK. */
2242 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2243 tp->snd_nxt, TH_ACK);
2244 tp->last_ack_sent = tp->rcv_nxt;
2251 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2252 * and it's less than ts_recent, drop it.
2254 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2255 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2256 /* Check to see if ts_recent is over 24 days old. */
2257 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2259 * Invalidate ts_recent. If this segment updates
2260 * ts_recent, the age will be reset later and ts_recent
2261 * will get a valid value. If it does not, setting
2262 * ts_recent to zero will at least satisfy the
2263 * requirement that zero be placed in the timestamp
2264 * echo reply when ts_recent isn't valid. The
2265 * age isn't reset until we get a valid ts_recent
2266 * because we don't want out-of-order segments to be
2267 * dropped when ts_recent is old.
2271 TCPSTAT_INC(tcps_rcvduppack);
2272 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2273 TCPSTAT_INC(tcps_pawsdrop);
2281 * In the SYN-RECEIVED state, validate that the packet belongs to
2282 * this connection before trimming the data to fit the receive
2283 * window. Check the sequence number versus IRS since we know
2284 * the sequence numbers haven't wrapped. This is a partial fix
2285 * for the "LAND" DoS attack.
2287 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2288 rstreason = BANDLIM_RST_OPENPORT;
2292 todrop = tp->rcv_nxt - th->th_seq;
2294 if (thflags & TH_SYN) {
2304 * Following if statement from Stevens, vol. 2, p. 960.
2307 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2309 * Any valid FIN must be to the left of the window.
2310 * At this point the FIN must be a duplicate or out
2311 * of sequence; drop it.
2316 * Send an ACK to resynchronize and drop any data.
2317 * But keep on processing for RST or ACK.
2319 tp->t_flags |= TF_ACKNOW;
2321 TCPSTAT_INC(tcps_rcvduppack);
2322 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2324 TCPSTAT_INC(tcps_rcvpartduppack);
2325 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2328 * DSACK - add SACK block for dropped range
2330 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2331 tcp_update_sack_list(tp, th->th_seq,
2332 th->th_seq + todrop);
2334 * ACK now, as the next in-sequence segment
2335 * will clear the DSACK block again
2337 tp->t_flags |= TF_ACKNOW;
2339 drop_hdrlen += todrop; /* drop from the top afterwards */
2340 th->th_seq += todrop;
2342 if (th->th_urp > todrop)
2343 th->th_urp -= todrop;
2351 * If new data are received on a connection after the
2352 * user processes are gone, then RST the other end.
2354 if ((so->so_state & SS_NOFDREF) &&
2355 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2356 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2357 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2358 "after socket was closed, "
2359 "sending RST and removing tcpcb\n",
2360 s, __func__, tcpstates[tp->t_state], tlen);
2364 TCPSTAT_INC(tcps_rcvafterclose);
2365 rstreason = BANDLIM_UNLIMITED;
2370 * If segment ends after window, drop trailing data
2371 * (and PUSH and FIN); if nothing left, just ACK.
2373 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2375 TCPSTAT_INC(tcps_rcvpackafterwin);
2376 if (todrop >= tlen) {
2377 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2379 * If window is closed can only take segments at
2380 * window edge, and have to drop data and PUSH from
2381 * incoming segments. Continue processing, but
2382 * remember to ack. Otherwise, drop segment
2385 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2386 tp->t_flags |= TF_ACKNOW;
2387 TCPSTAT_INC(tcps_rcvwinprobe);
2391 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2394 thflags &= ~(TH_PUSH|TH_FIN);
2398 * If last ACK falls within this segment's sequence numbers,
2399 * record its timestamp.
2401 * 1) That the test incorporates suggestions from the latest
2402 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2403 * 2) That updating only on newer timestamps interferes with
2404 * our earlier PAWS tests, so this check should be solely
2405 * predicated on the sequence space of this segment.
2406 * 3) That we modify the segment boundary check to be
2407 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2408 * instead of RFC1323's
2409 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2410 * This modified check allows us to overcome RFC1323's
2411 * limitations as described in Stevens TCP/IP Illustrated
2412 * Vol. 2 p.869. In such cases, we can still calculate the
2413 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2415 if ((to.to_flags & TOF_TS) != 0 &&
2416 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2417 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2418 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2419 tp->ts_recent_age = tcp_ts_getticks();
2420 tp->ts_recent = to.to_tsval;
2424 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2425 * flag is on (half-synchronized state), then queue data for
2426 * later processing; else drop segment and return.
2428 if ((thflags & TH_ACK) == 0) {
2429 if (tp->t_state == TCPS_SYN_RECEIVED ||
2430 (tp->t_flags & TF_NEEDSYN)) {
2431 if (tp->t_state == TCPS_SYN_RECEIVED &&
2432 IS_FASTOPEN(tp->t_flags)) {
2433 tp->snd_wnd = tiwin;
2437 } else if (tp->t_flags & TF_ACKNOW)
2446 switch (tp->t_state) {
2448 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2449 * ESTABLISHED state and continue processing.
2450 * The ACK was checked above.
2452 case TCPS_SYN_RECEIVED:
2454 TCPSTAT_INC(tcps_connects);
2456 /* Do window scaling? */
2457 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2458 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2459 tp->rcv_scale = tp->request_r_scale;
2461 tp->snd_wnd = tiwin;
2464 * SYN-RECEIVED -> ESTABLISHED
2465 * SYN-RECEIVED* -> FIN-WAIT-1
2467 tp->t_starttime = ticks;
2468 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2469 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2470 tp->t_tfo_pending = NULL;
2472 if (tp->t_flags & TF_NEEDFIN) {
2473 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2474 tp->t_flags &= ~TF_NEEDFIN;
2476 tcp_state_change(tp, TCPS_ESTABLISHED);
2477 TCP_PROBE5(accept__established, NULL, tp,
2480 * TFO connections call cc_conn_init() during SYN
2481 * processing. Calling it again here for such
2482 * connections is not harmless as it would undo the
2483 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2486 if (!IS_FASTOPEN(tp->t_flags))
2488 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2491 * Account for the ACK of our SYN prior to
2492 * regular ACK processing below, except for
2493 * simultaneous SYN, which is handled later.
2495 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2498 * If segment contains data or ACK, will call tcp_reass()
2499 * later; if not, do so now to pass queued data to user.
2501 if (tlen == 0 && (thflags & TH_FIN) == 0)
2502 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2504 tp->snd_wl1 = th->th_seq - 1;
2508 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2509 * ACKs. If the ack is in the range
2510 * tp->snd_una < th->th_ack <= tp->snd_max
2511 * then advance tp->snd_una to th->th_ack and drop
2512 * data from the retransmission queue. If this ACK reflects
2513 * more up to date window information we update our window information.
2515 case TCPS_ESTABLISHED:
2516 case TCPS_FIN_WAIT_1:
2517 case TCPS_FIN_WAIT_2:
2518 case TCPS_CLOSE_WAIT:
2521 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2522 TCPSTAT_INC(tcps_rcvacktoomuch);
2525 if ((tp->t_flags & TF_SACK_PERMIT) &&
2526 ((to.to_flags & TOF_SACK) ||
2527 !TAILQ_EMPTY(&tp->snd_holes)))
2528 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2531 * Reset the value so that previous (valid) value
2532 * from the last ack with SACK doesn't get used.
2534 tp->sackhint.sacked_bytes = 0;
2537 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2538 hhook_run_tcp_est_in(tp, th, &to);
2541 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2542 maxseg = tcp_maxseg(tp);
2544 (tiwin == tp->snd_wnd ||
2545 (tp->t_flags & TF_SACK_PERMIT))) {
2547 * If this is the first time we've seen a
2548 * FIN from the remote, this is not a
2549 * duplicate and it needs to be processed
2550 * normally. This happens during a
2551 * simultaneous close.
2553 if ((thflags & TH_FIN) &&
2554 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2558 TCPSTAT_INC(tcps_rcvdupack);
2560 * If we have outstanding data (other than
2561 * a window probe), this is a completely
2562 * duplicate ack (ie, window info didn't
2563 * change and FIN isn't set),
2564 * the ack is the biggest we've
2565 * seen and we've seen exactly our rexmt
2566 * threshold of them, assume a packet
2567 * has been dropped and retransmit it.
2568 * Kludge snd_nxt & the congestion
2569 * window so we send only this one
2572 * We know we're losing at the current
2573 * window size so do congestion avoidance
2574 * (set ssthresh to half the current window
2575 * and pull our congestion window back to
2576 * the new ssthresh).
2578 * Dup acks mean that packets have left the
2579 * network (they're now cached at the receiver)
2580 * so bump cwnd by the amount in the receiver
2581 * to keep a constant cwnd packets in the
2584 * When using TCP ECN, notify the peer that
2585 * we reduced the cwnd.
2588 * Following 2 kinds of acks should not affect
2591 * 2) Acks with SACK but without any new SACK
2592 * information in them. These could result from
2593 * any anomaly in the network like a switch
2594 * duplicating packets or a possible DoS attack.
2596 if (th->th_ack != tp->snd_una ||
2597 ((tp->t_flags & TF_SACK_PERMIT) &&
2598 (to.to_flags & TOF_SACK) &&
2601 else if (!tcp_timer_active(tp, TT_REXMT))
2603 else if (++tp->t_dupacks > tcprexmtthresh ||
2604 IN_FASTRECOVERY(tp->t_flags)) {
2605 cc_ack_received(tp, th, nsegs,
2608 IN_FASTRECOVERY(tp->t_flags) &&
2609 (tp->t_flags & TF_SACK_PERMIT)) {
2610 tcp_do_prr_ack(tp, th, &to);
2611 } else if ((tp->t_flags & TF_SACK_PERMIT) &&
2612 (to.to_flags & TOF_SACK) &&
2613 IN_FASTRECOVERY(tp->t_flags)) {
2617 * Compute the amount of data in flight first.
2618 * We can inject new data into the pipe iff
2619 * we have less than 1/2 the original window's
2620 * worth of data in flight.
2622 if (V_tcp_do_rfc6675_pipe)
2623 awnd = tcp_compute_pipe(tp);
2625 awnd = (tp->snd_nxt - tp->snd_fack) +
2626 tp->sackhint.sack_bytes_rexmit;
2628 if (awnd < tp->snd_ssthresh) {
2629 tp->snd_cwnd += maxseg;
2630 if (tp->snd_cwnd > tp->snd_ssthresh)
2631 tp->snd_cwnd = tp->snd_ssthresh;
2634 tp->snd_cwnd += maxseg;
2635 (void) tp->t_fb->tfb_tcp_output(tp);
2637 } else if (tp->t_dupacks == tcprexmtthresh ||
2638 (tp->t_flags & TF_SACK_PERMIT &&
2639 V_tcp_do_rfc6675_pipe &&
2640 tp->sackhint.sacked_bytes >
2641 (tcprexmtthresh - 1) * maxseg)) {
2644 * Above is the RFC6675 trigger condition of
2645 * more than (dupthresh-1)*maxseg sacked data.
2646 * If the count of holes in the
2647 * scoreboard is >= dupthresh, we could
2648 * also enter loss recovery, but don't
2649 * have that value readily available.
2651 tp->t_dupacks = tcprexmtthresh;
2652 tcp_seq onxt = tp->snd_nxt;
2655 * If we're doing sack, or prr, check
2656 * to see if we're already in sack
2657 * recovery. If we're not doing sack,
2658 * check to see if we're in newreno
2662 (tp->t_flags & TF_SACK_PERMIT)) {
2663 if (IN_FASTRECOVERY(tp->t_flags)) {
2668 if (SEQ_LEQ(th->th_ack,
2674 /* Congestion signal before ack. */
2675 cc_cong_signal(tp, th, CC_NDUPACK);
2676 cc_ack_received(tp, th, nsegs,
2678 tcp_timer_activate(tp, TT_REXMT, 0);
2682 * snd_ssthresh is already updated by
2685 tp->sackhint.prr_delivered =
2686 tp->sackhint.sacked_bytes;
2687 tp->sackhint.recover_fs = max(1,
2688 tp->snd_nxt - tp->snd_una);
2690 if ((tp->t_flags & TF_SACK_PERMIT) &&
2691 (to.to_flags & TOF_SACK)) {
2693 tcps_sack_recovery_episode);
2694 tp->snd_recover = tp->snd_nxt;
2695 tp->snd_cwnd = maxseg;
2696 (void) tp->t_fb->tfb_tcp_output(tp);
2697 if (SEQ_GT(th->th_ack, tp->snd_una))
2698 goto resume_partialack;
2701 tp->snd_nxt = th->th_ack;
2702 tp->snd_cwnd = maxseg;
2703 (void) tp->t_fb->tfb_tcp_output(tp);
2704 KASSERT(tp->snd_limited <= 2,
2705 ("%s: tp->snd_limited too big",
2707 tp->snd_cwnd = tp->snd_ssthresh +
2709 (tp->t_dupacks - tp->snd_limited);
2710 if (SEQ_GT(onxt, tp->snd_nxt))
2713 } else if (V_tcp_do_rfc3042) {
2715 * Process first and second duplicate
2716 * ACKs. Each indicates a segment
2717 * leaving the network, creating room
2718 * for more. Make sure we can send a
2719 * packet on reception of each duplicate
2720 * ACK by increasing snd_cwnd by one
2721 * segment. Restore the original
2722 * snd_cwnd after packet transmission.
2724 cc_ack_received(tp, th, nsegs,
2726 uint32_t oldcwnd = tp->snd_cwnd;
2727 tcp_seq oldsndmax = tp->snd_max;
2731 KASSERT(tp->t_dupacks == 1 ||
2733 ("%s: dupacks not 1 or 2",
2735 if (tp->t_dupacks == 1)
2736 tp->snd_limited = 0;
2738 (tp->snd_nxt - tp->snd_una) +
2739 (tp->t_dupacks - tp->snd_limited) *
2742 * Only call tcp_output when there
2743 * is new data available to be sent.
2744 * Otherwise we would send pure ACKs.
2746 SOCKBUF_LOCK(&so->so_snd);
2747 avail = sbavail(&so->so_snd) -
2748 (tp->snd_nxt - tp->snd_una);
2749 SOCKBUF_UNLOCK(&so->so_snd);
2751 (void) tp->t_fb->tfb_tcp_output(tp);
2752 sent = tp->snd_max - oldsndmax;
2753 if (sent > maxseg) {
2754 KASSERT((tp->t_dupacks == 2 &&
2755 tp->snd_limited == 0) ||
2756 (sent == maxseg + 1 &&
2757 tp->t_flags & TF_SENTFIN),
2758 ("%s: sent too much",
2760 tp->snd_limited = 2;
2761 } else if (sent > 0)
2763 tp->snd_cwnd = oldcwnd;
2770 * This ack is advancing the left edge, reset the
2775 * If this ack also has new SACK info, increment the
2776 * counter as per rfc6675. The variable
2777 * sack_changed tracks all changes to the SACK
2778 * scoreboard, including when partial ACKs without
2779 * SACK options are received, and clear the scoreboard
2780 * from the left side. Such partial ACKs should not be
2781 * counted as dupacks here.
2783 if ((tp->t_flags & TF_SACK_PERMIT) &&
2784 (to.to_flags & TOF_SACK) &&
2787 /* limit overhead by setting maxseg last */
2788 if (!IN_FASTRECOVERY(tp->t_flags) &&
2789 (tp->sackhint.sacked_bytes >
2790 ((tcprexmtthresh - 1) *
2791 (maxseg = tcp_maxseg(tp))))) {
2792 goto enter_recovery;
2798 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2799 ("%s: th_ack <= snd_una", __func__));
2802 * If the congestion window was inflated to account
2803 * for the other side's cached packets, retract it.
2805 if (IN_FASTRECOVERY(tp->t_flags)) {
2806 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2807 if (tp->t_flags & TF_SACK_PERMIT)
2808 if (V_tcp_do_prr && to.to_flags & TOF_SACK) {
2809 tcp_timer_activate(tp, TT_REXMT, 0);
2811 tcp_do_prr_ack(tp, th, &to);
2812 tp->t_flags |= TF_ACKNOW;
2813 (void) tcp_output(tp);
2815 tcp_sack_partialack(tp, th);
2817 tcp_newreno_partial_ack(tp, th);
2819 cc_post_recovery(tp, th);
2820 } else if (IN_CONGRECOVERY(tp->t_flags)) {
2821 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2823 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th);
2824 tp->snd_fack = th->th_ack;
2825 tcp_do_prr_ack(tp, th, &to);
2826 (void) tcp_output(tp);
2829 cc_post_recovery(tp, th);
2832 * If we reach this point, ACK is not a duplicate,
2833 * i.e., it ACKs something we sent.
2835 if (tp->t_flags & TF_NEEDSYN) {
2837 * T/TCP: Connection was half-synchronized, and our
2838 * SYN has been ACK'd (so connection is now fully
2839 * synchronized). Go to non-starred state,
2840 * increment snd_una for ACK of SYN, and check if
2841 * we can do window scaling.
2843 tp->t_flags &= ~TF_NEEDSYN;
2845 /* Do window scaling? */
2846 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2847 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2848 tp->rcv_scale = tp->request_r_scale;
2849 /* Send window already scaled. */
2854 INP_WLOCK_ASSERT(tp->t_inpcb);
2857 * Adjust for the SYN bit in sequence space,
2858 * but don't account for it in cwnd calculations.
2859 * This is for the SYN_RECEIVED, non-simultaneous
2860 * SYN case. SYN_SENT and simultaneous SYN are
2861 * treated elsewhere.
2865 acked = BYTES_THIS_ACK(tp, th);
2866 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2867 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2868 tp->snd_una, th->th_ack, tp, m));
2869 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2870 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2873 * If we just performed our first retransmit, and the ACK
2874 * arrives within our recovery window, then it was a mistake
2875 * to do the retransmit in the first place. Recover our
2876 * original cwnd and ssthresh, and proceed to transmit where
2879 if (tp->t_rxtshift == 1 &&
2880 tp->t_flags & TF_PREVVALID &&
2882 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2883 cc_cong_signal(tp, th, CC_RTO_ERR);
2886 * If we have a timestamp reply, update smoothed
2887 * round trip time. If no timestamp is present but
2888 * transmit timer is running and timed sequence
2889 * number was acked, update smoothed round trip time.
2890 * Since we now have an rtt measurement, cancel the
2891 * timer backoff (cf., Phil Karn's retransmit alg.).
2892 * Recompute the initial retransmit timer.
2894 * Some boxes send broken timestamp replies
2895 * during the SYN+ACK phase, ignore
2896 * timestamps of 0 or we could calculate a
2897 * huge RTT and blow up the retransmit timer.
2899 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2902 t = tcp_ts_getticks() - to.to_tsecr;
2903 if (!tp->t_rttlow || tp->t_rttlow > t)
2905 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2906 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2907 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2908 tp->t_rttlow = ticks - tp->t_rtttime;
2909 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2913 * If all outstanding data is acked, stop retransmit
2914 * timer and remember to restart (more output or persist).
2915 * If there is more data to be acked, restart retransmit
2916 * timer, using current (possibly backed-off) value.
2918 if (th->th_ack == tp->snd_max) {
2919 tcp_timer_activate(tp, TT_REXMT, 0);
2921 } else if (!tcp_timer_active(tp, TT_PERSIST))
2922 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2925 * If no data (only SYN) was ACK'd,
2926 * skip rest of ACK processing.
2932 * Let the congestion control algorithm update congestion
2933 * control related information. This typically means increasing
2934 * the congestion window.
2936 cc_ack_received(tp, th, nsegs, CC_ACK);
2938 SOCKBUF_LOCK(&so->so_snd);
2939 if (acked > sbavail(&so->so_snd)) {
2940 if (tp->snd_wnd >= sbavail(&so->so_snd))
2941 tp->snd_wnd -= sbavail(&so->so_snd);
2944 mfree = sbcut_locked(&so->so_snd,
2945 (int)sbavail(&so->so_snd));
2948 mfree = sbcut_locked(&so->so_snd, acked);
2949 if (tp->snd_wnd >= (uint32_t) acked)
2950 tp->snd_wnd -= acked;
2955 SOCKBUF_UNLOCK(&so->so_snd);
2956 tp->t_flags |= TF_WAKESOW;
2958 /* Detect una wraparound. */
2959 if (!IN_RECOVERY(tp->t_flags) &&
2960 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2961 SEQ_LEQ(th->th_ack, tp->snd_recover))
2962 tp->snd_recover = th->th_ack - 1;
2963 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2964 if (IN_RECOVERY(tp->t_flags) &&
2965 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2966 EXIT_RECOVERY(tp->t_flags);
2968 tp->snd_una = th->th_ack;
2969 if (tp->t_flags & TF_SACK_PERMIT) {
2970 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2971 tp->snd_recover = tp->snd_una;
2973 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2974 tp->snd_nxt = tp->snd_una;
2976 switch (tp->t_state) {
2978 * In FIN_WAIT_1 STATE in addition to the processing
2979 * for the ESTABLISHED state if our FIN is now acknowledged
2980 * then enter FIN_WAIT_2.
2982 case TCPS_FIN_WAIT_1:
2983 if (ourfinisacked) {
2985 * If we can't receive any more
2986 * data, then closing user can proceed.
2987 * Starting the timer is contrary to the
2988 * specification, but if we don't get a FIN
2989 * we'll hang forever.
2992 * we should release the tp also, and use a
2995 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2996 soisdisconnected(so);
2997 tcp_timer_activate(tp, TT_2MSL,
2998 (tcp_fast_finwait2_recycle ?
2999 tcp_finwait2_timeout :
3002 tcp_state_change(tp, TCPS_FIN_WAIT_2);
3007 * In CLOSING STATE in addition to the processing for
3008 * the ESTABLISHED state if the ACK acknowledges our FIN
3009 * then enter the TIME-WAIT state, otherwise ignore
3013 if (ourfinisacked) {
3021 * In LAST_ACK, we may still be waiting for data to drain
3022 * and/or to be acked, as well as for the ack of our FIN.
3023 * If our FIN is now acknowledged, delete the TCB,
3024 * enter the closed state and return.
3027 if (ourfinisacked) {
3036 INP_WLOCK_ASSERT(tp->t_inpcb);
3039 * Update window information.
3040 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3042 if ((thflags & TH_ACK) &&
3043 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3044 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3045 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3046 /* keep track of pure window updates */
3048 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3049 TCPSTAT_INC(tcps_rcvwinupd);
3050 tp->snd_wnd = tiwin;
3051 tp->snd_wl1 = th->th_seq;
3052 tp->snd_wl2 = th->th_ack;
3053 if (tp->snd_wnd > tp->max_sndwnd)
3054 tp->max_sndwnd = tp->snd_wnd;
3059 * Process segments with URG.
3061 if ((thflags & TH_URG) && th->th_urp &&
3062 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3064 * This is a kludge, but if we receive and accept
3065 * random urgent pointers, we'll crash in
3066 * soreceive. It's hard to imagine someone
3067 * actually wanting to send this much urgent data.
3069 SOCKBUF_LOCK(&so->so_rcv);
3070 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3071 th->th_urp = 0; /* XXX */
3072 thflags &= ~TH_URG; /* XXX */
3073 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3074 goto dodata; /* XXX */
3077 * If this segment advances the known urgent pointer,
3078 * then mark the data stream. This should not happen
3079 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3080 * a FIN has been received from the remote side.
3081 * In these states we ignore the URG.
3083 * According to RFC961 (Assigned Protocols),
3084 * the urgent pointer points to the last octet
3085 * of urgent data. We continue, however,
3086 * to consider it to indicate the first octet
3087 * of data past the urgent section as the original
3088 * spec states (in one of two places).
3090 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3091 tp->rcv_up = th->th_seq + th->th_urp;
3092 so->so_oobmark = sbavail(&so->so_rcv) +
3093 (tp->rcv_up - tp->rcv_nxt) - 1;
3094 if (so->so_oobmark == 0)
3095 so->so_rcv.sb_state |= SBS_RCVATMARK;
3097 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3099 SOCKBUF_UNLOCK(&so->so_rcv);
3101 * Remove out of band data so doesn't get presented to user.
3102 * This can happen independent of advancing the URG pointer,
3103 * but if two URG's are pending at once, some out-of-band
3104 * data may creep in... ick.
3106 if (th->th_urp <= (uint32_t)tlen &&
3107 !(so->so_options & SO_OOBINLINE)) {
3108 /* hdr drop is delayed */
3109 tcp_pulloutofband(so, th, m, drop_hdrlen);
3113 * If no out of band data is expected,
3114 * pull receive urgent pointer along
3115 * with the receive window.
3117 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3118 tp->rcv_up = tp->rcv_nxt;
3121 INP_WLOCK_ASSERT(tp->t_inpcb);
3124 * Process the segment text, merging it into the TCP sequencing queue,
3125 * and arranging for acknowledgment of receipt if necessary.
3126 * This process logically involves adjusting tp->rcv_wnd as data
3127 * is presented to the user (this happens in tcp_usrreq.c,
3128 * case PRU_RCVD). If a FIN has already been received on this
3129 * connection then we just ignore the text.
3131 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3132 IS_FASTOPEN(tp->t_flags));
3133 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3134 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3135 tcp_seq save_start = th->th_seq;
3136 tcp_seq save_rnxt = tp->rcv_nxt;
3137 int save_tlen = tlen;
3138 m_adj(m, drop_hdrlen); /* delayed header drop */
3140 * Insert segment which includes th into TCP reassembly queue
3141 * with control block tp. Set thflags to whether reassembly now
3142 * includes a segment with FIN. This handles the common case
3143 * inline (segment is the next to be received on an established
3144 * connection, and the queue is empty), avoiding linkage into
3145 * and removal from the queue and repetition of various
3147 * Set DELACK for segments received in order, but ack
3148 * immediately when segments are out of order (so
3149 * fast retransmit can work).
3151 if (th->th_seq == tp->rcv_nxt &&
3153 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3155 if (DELAY_ACK(tp, tlen) || tfo_syn)
3156 tp->t_flags |= TF_DELACK;
3158 tp->t_flags |= TF_ACKNOW;
3159 tp->rcv_nxt += tlen;
3161 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3162 (tp->t_fbyte_in == 0)) {
3163 tp->t_fbyte_in = ticks;
3164 if (tp->t_fbyte_in == 0)
3166 if (tp->t_fbyte_out && tp->t_fbyte_in)
3167 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3169 thflags = th->th_flags & TH_FIN;
3170 TCPSTAT_INC(tcps_rcvpack);
3171 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3172 SOCKBUF_LOCK(&so->so_rcv);
3173 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3176 sbappendstream_locked(&so->so_rcv, m, 0);
3177 SOCKBUF_UNLOCK(&so->so_rcv);
3178 tp->t_flags |= TF_WAKESOR;
3181 * XXX: Due to the header drop above "th" is
3182 * theoretically invalid by now. Fortunately
3183 * m_adj() doesn't actually frees any mbufs
3184 * when trimming from the head.
3186 tcp_seq temp = save_start;
3188 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3189 tp->t_flags |= TF_ACKNOW;
3191 if ((tp->t_flags & TF_SACK_PERMIT) &&
3193 TCPS_HAVEESTABLISHED(tp->t_state)) {
3194 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3196 * DSACK actually handled in the fastpath
3199 tcp_update_sack_list(tp, save_start,
3200 save_start + save_tlen);
3201 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3202 if ((tp->rcv_numsacks >= 1) &&
3203 (tp->sackblks[0].end == save_start)) {
3205 * Partial overlap, recorded at todrop
3208 tcp_update_sack_list(tp,
3209 tp->sackblks[0].start,
3210 tp->sackblks[0].end);
3212 tcp_update_dsack_list(tp, save_start,
3213 save_start + save_tlen);
3215 } else if (tlen >= save_tlen) {
3216 /* Update of sackblks. */
3217 tcp_update_dsack_list(tp, save_start,
3218 save_start + save_tlen);
3219 } else if (tlen > 0) {
3220 tcp_update_dsack_list(tp, save_start,
3226 * Note the amount of data that peer has sent into
3227 * our window, in order to estimate the sender's
3231 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3232 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3234 len = so->so_rcv.sb_hiwat;
3242 * If FIN is received ACK the FIN and let the user know
3243 * that the connection is closing.
3245 if (thflags & TH_FIN) {
3246 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3248 /* The socket upcall is handled by socantrcvmore. */
3249 tp->t_flags &= ~TF_WAKESOR;
3251 * If connection is half-synchronized
3252 * (ie NEEDSYN flag on) then delay ACK,
3253 * so it may be piggybacked when SYN is sent.
3254 * Otherwise, since we received a FIN then no
3255 * more input can be expected, send ACK now.
3257 if (tp->t_flags & TF_NEEDSYN)
3258 tp->t_flags |= TF_DELACK;
3260 tp->t_flags |= TF_ACKNOW;
3263 switch (tp->t_state) {
3265 * In SYN_RECEIVED and ESTABLISHED STATES
3266 * enter the CLOSE_WAIT state.
3268 case TCPS_SYN_RECEIVED:
3269 tp->t_starttime = ticks;
3271 case TCPS_ESTABLISHED:
3272 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3276 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3277 * enter the CLOSING state.
3279 case TCPS_FIN_WAIT_1:
3280 tcp_state_change(tp, TCPS_CLOSING);
3284 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3285 * starting the time-wait timer, turning off the other
3288 case TCPS_FIN_WAIT_2:
3294 if (so->so_options & SO_DEBUG)
3295 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3298 TCP_PROBE3(debug__input, tp, th, m);
3301 * Return any desired output.
3303 if (needoutput || (tp->t_flags & TF_ACKNOW))
3304 (void) tp->t_fb->tfb_tcp_output(tp);
3307 INP_WLOCK_ASSERT(tp->t_inpcb);
3309 if (tp->t_flags & TF_DELACK) {
3310 tp->t_flags &= ~TF_DELACK;
3311 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3313 tcp_handle_wakeup(tp, so);
3314 INP_WUNLOCK(tp->t_inpcb);
3319 * Generate an ACK dropping incoming segment if it occupies
3320 * sequence space, where the ACK reflects our state.
3322 * We can now skip the test for the RST flag since all
3323 * paths to this code happen after packets containing
3324 * RST have been dropped.
3326 * In the SYN-RECEIVED state, don't send an ACK unless the
3327 * segment we received passes the SYN-RECEIVED ACK test.
3328 * If it fails send a RST. This breaks the loop in the
3329 * "LAND" DoS attack, and also prevents an ACK storm
3330 * between two listening ports that have been sent forged
3331 * SYN segments, each with the source address of the other.
3333 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3334 (SEQ_GT(tp->snd_una, th->th_ack) ||
3335 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3336 rstreason = BANDLIM_RST_OPENPORT;
3340 if (so->so_options & SO_DEBUG)
3341 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3344 TCP_PROBE3(debug__input, tp, th, m);
3345 tp->t_flags |= TF_ACKNOW;
3346 (void) tp->t_fb->tfb_tcp_output(tp);
3347 tcp_handle_wakeup(tp, so);
3348 INP_WUNLOCK(tp->t_inpcb);
3354 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3355 tcp_handle_wakeup(tp, so);
3356 INP_WUNLOCK(tp->t_inpcb);
3358 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3363 * Drop space held by incoming segment and return.
3366 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3367 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3370 TCP_PROBE3(debug__input, tp, th, m);
3372 tcp_handle_wakeup(tp, so);
3373 INP_WUNLOCK(tp->t_inpcb);
3379 * Issue RST and make ACK acceptable to originator of segment.
3380 * The mbuf must still include the original packet header.
3384 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3385 int tlen, int rstreason)
3391 struct ip6_hdr *ip6;
3395 INP_WLOCK_ASSERT(tp->t_inpcb);
3398 /* Don't bother if destination was broadcast/multicast. */
3399 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3402 if (mtod(m, struct ip *)->ip_v == 6) {
3403 ip6 = mtod(m, struct ip6_hdr *);
3404 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3405 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3407 /* IPv6 anycast check is done at tcp6_input() */
3410 #if defined(INET) && defined(INET6)
3415 ip = mtod(m, struct ip *);
3416 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3417 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3418 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3419 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3424 /* Perform bandwidth limiting. */
3425 if (badport_bandlim(rstreason) < 0)
3428 /* tcp_respond consumes the mbuf chain. */
3429 if (th->th_flags & TH_ACK) {
3430 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3431 th->th_ack, TH_RST);
3433 if (th->th_flags & TH_SYN)
3435 if (th->th_flags & TH_FIN)
3437 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3438 (tcp_seq)0, TH_RST|TH_ACK);
3446 * Parse TCP options and place in tcpopt.
3449 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3454 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3456 if (opt == TCPOPT_EOL)
3458 if (opt == TCPOPT_NOP)
3464 if (optlen < 2 || optlen > cnt)
3469 if (optlen != TCPOLEN_MAXSEG)
3471 if (!(flags & TO_SYN))
3473 to->to_flags |= TOF_MSS;
3474 bcopy((char *)cp + 2,
3475 (char *)&to->to_mss, sizeof(to->to_mss));
3476 to->to_mss = ntohs(to->to_mss);
3479 if (optlen != TCPOLEN_WINDOW)
3481 if (!(flags & TO_SYN))
3483 to->to_flags |= TOF_SCALE;
3484 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3486 case TCPOPT_TIMESTAMP:
3487 if (optlen != TCPOLEN_TIMESTAMP)
3489 to->to_flags |= TOF_TS;
3490 bcopy((char *)cp + 2,
3491 (char *)&to->to_tsval, sizeof(to->to_tsval));
3492 to->to_tsval = ntohl(to->to_tsval);
3493 bcopy((char *)cp + 6,
3494 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3495 to->to_tsecr = ntohl(to->to_tsecr);
3497 case TCPOPT_SIGNATURE:
3499 * In order to reply to a host which has set the
3500 * TCP_SIGNATURE option in its initial SYN, we have
3501 * to record the fact that the option was observed
3502 * here for the syncache code to perform the correct
3505 if (optlen != TCPOLEN_SIGNATURE)
3507 to->to_flags |= TOF_SIGNATURE;
3508 to->to_signature = cp + 2;
3510 case TCPOPT_SACK_PERMITTED:
3511 if (optlen != TCPOLEN_SACK_PERMITTED)
3513 if (!(flags & TO_SYN))
3517 to->to_flags |= TOF_SACKPERM;
3520 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3524 to->to_flags |= TOF_SACK;
3525 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3526 to->to_sacks = cp + 2;
3527 TCPSTAT_INC(tcps_sack_rcv_blocks);
3529 case TCPOPT_FAST_OPEN:
3531 * Cookie length validation is performed by the
3532 * server side cookie checking code or the client
3533 * side cookie cache update code.
3535 if (!(flags & TO_SYN))
3537 if (!V_tcp_fastopen_client_enable &&
3538 !V_tcp_fastopen_server_enable)
3540 to->to_flags |= TOF_FASTOPEN;
3541 to->to_tfo_len = optlen - 2;
3542 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3551 * Pull out of band byte out of a segment so
3552 * it doesn't appear in the user's data queue.
3553 * It is still reflected in the segment length for
3554 * sequencing purposes.
3557 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3560 int cnt = off + th->th_urp - 1;
3563 if (m->m_len > cnt) {
3564 char *cp = mtod(m, caddr_t) + cnt;
3565 struct tcpcb *tp = sototcpcb(so);
3567 INP_WLOCK_ASSERT(tp->t_inpcb);
3570 tp->t_oobflags |= TCPOOB_HAVEDATA;
3571 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3573 if (m->m_flags & M_PKTHDR)
3582 panic("tcp_pulloutofband");
3586 * Collect new round-trip time estimate
3587 * and update averages and current timeout.
3590 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3594 INP_WLOCK_ASSERT(tp->t_inpcb);
3596 TCPSTAT_INC(tcps_rttupdated);
3599 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3600 imax(0, rtt * 1000 / hz));
3602 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3604 * srtt is stored as fixed point with 5 bits after the
3605 * binary point (i.e., scaled by 8). The following magic
3606 * is equivalent to the smoothing algorithm in rfc793 with
3607 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3608 * point). Adjust rtt to origin 0.
3610 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3611 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3613 if ((tp->t_srtt += delta) <= 0)
3617 * We accumulate a smoothed rtt variance (actually, a
3618 * smoothed mean difference), then set the retransmit
3619 * timer to smoothed rtt + 4 times the smoothed variance.
3620 * rttvar is stored as fixed point with 4 bits after the
3621 * binary point (scaled by 16). The following is
3622 * equivalent to rfc793 smoothing with an alpha of .75
3623 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3624 * rfc793's wired-in beta.
3628 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3629 if ((tp->t_rttvar += delta) <= 0)
3631 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3632 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3635 * No rtt measurement yet - use the unsmoothed rtt.
3636 * Set the variance to half the rtt (so our first
3637 * retransmit happens at 3*rtt).
3639 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3640 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3641 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3647 * the retransmit should happen at rtt + 4 * rttvar.
3648 * Because of the way we do the smoothing, srtt and rttvar
3649 * will each average +1/2 tick of bias. When we compute
3650 * the retransmit timer, we want 1/2 tick of rounding and
3651 * 1 extra tick because of +-1/2 tick uncertainty in the
3652 * firing of the timer. The bias will give us exactly the
3653 * 1.5 tick we need. But, because the bias is
3654 * statistical, we have to test that we don't drop below
3655 * the minimum feasible timer (which is 2 ticks).
3657 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3658 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3661 * We received an ack for a packet that wasn't retransmitted;
3662 * it is probably safe to discard any error indications we've
3663 * received recently. This isn't quite right, but close enough
3664 * for now (a route might have failed after we sent a segment,
3665 * and the return path might not be symmetrical).
3667 tp->t_softerror = 0;
3671 * Determine a reasonable value for maxseg size.
3672 * If the route is known, check route for mtu.
3673 * If none, use an mss that can be handled on the outgoing interface
3674 * without forcing IP to fragment. If no route is found, route has no mtu,
3675 * or the destination isn't local, use a default, hopefully conservative
3676 * size (usually 512 or the default IP max size, but no more than the mtu
3677 * of the interface), as we can't discover anything about intervening
3678 * gateways or networks. We also initialize the congestion/slow start
3679 * window to be a single segment if the destination isn't local.
3680 * While looking at the routing entry, we also initialize other path-dependent
3681 * parameters from pre-set or cached values in the routing entry.
3683 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3684 * IP options, e.g. IPSEC data, since length of this data may vary, and
3685 * thus it is calculated for every segment separately in tcp_output().
3687 * NOTE that this routine is only called when we process an incoming
3688 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3689 * settings are handled in tcp_mssopt().
3692 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3693 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3696 uint32_t maxmtu = 0;
3697 struct inpcb *inp = tp->t_inpcb;
3698 struct hc_metrics_lite metrics;
3700 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3701 size_t min_protoh = isipv6 ?
3702 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3703 sizeof (struct tcpiphdr);
3705 size_t min_protoh = sizeof(struct tcpiphdr);
3708 INP_WLOCK_ASSERT(tp->t_inpcb);
3711 min_protoh += V_tcp_udp_tunneling_overhead;
3712 if (mtuoffer != -1) {
3713 KASSERT(offer == -1, ("%s: conflict", __func__));
3714 offer = mtuoffer - min_protoh;
3720 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3721 tp->t_maxseg = V_tcp_v6mssdflt;
3724 #if defined(INET) && defined(INET6)
3729 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3730 tp->t_maxseg = V_tcp_mssdflt;
3735 * No route to sender, stay with default mss and return.
3739 * In case we return early we need to initialize metrics
3740 * to a defined state as tcp_hc_get() would do for us
3741 * if there was no cache hit.
3743 if (metricptr != NULL)
3744 bzero(metricptr, sizeof(struct hc_metrics_lite));
3748 /* What have we got? */
3752 * Offer == 0 means that there was no MSS on the SYN
3753 * segment, in this case we use tcp_mssdflt as
3754 * already assigned to t_maxseg above.
3756 offer = tp->t_maxseg;
3761 * Offer == -1 means that we didn't receive SYN yet.
3767 * Prevent DoS attack with too small MSS. Round up
3768 * to at least minmss.
3770 offer = max(offer, V_tcp_minmss);
3774 * rmx information is now retrieved from tcp_hostcache.
3776 tcp_hc_get(&inp->inp_inc, &metrics);
3777 if (metricptr != NULL)
3778 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3781 * If there's a discovered mtu in tcp hostcache, use it.
3782 * Else, use the link mtu.
3784 if (metrics.rmx_mtu)
3785 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3789 mss = maxmtu - min_protoh;
3790 if (!V_path_mtu_discovery &&
3791 !in6_localaddr(&inp->in6p_faddr))
3792 mss = min(mss, V_tcp_v6mssdflt);
3795 #if defined(INET) && defined(INET6)
3800 mss = maxmtu - min_protoh;
3801 if (!V_path_mtu_discovery &&
3802 !in_localaddr(inp->inp_faddr))
3803 mss = min(mss, V_tcp_mssdflt);
3807 * XXX - The above conditional (mss = maxmtu - min_protoh)
3808 * probably violates the TCP spec.
3809 * The problem is that, since we don't know the
3810 * other end's MSS, we are supposed to use a conservative
3811 * default. But, if we do that, then MTU discovery will
3812 * never actually take place, because the conservative
3813 * default is much less than the MTUs typically seen
3814 * on the Internet today. For the moment, we'll sweep
3815 * this under the carpet.
3817 * The conservative default might not actually be a problem
3818 * if the only case this occurs is when sending an initial
3819 * SYN with options and data to a host we've never talked
3820 * to before. Then, they will reply with an MSS value which
3821 * will get recorded and the new parameters should get
3822 * recomputed. For Further Study.
3825 mss = min(mss, offer);
3828 * Sanity check: make sure that maxseg will be large
3829 * enough to allow some data on segments even if the
3830 * all the option space is used (40bytes). Otherwise
3831 * funny things may happen in tcp_output.
3833 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3841 tcp_mss(struct tcpcb *tp, int offer)
3847 struct hc_metrics_lite metrics;
3848 struct tcp_ifcap cap;
3850 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3852 bzero(&cap, sizeof(cap));
3853 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3859 * If there's a pipesize, change the socket buffer to that size,
3860 * don't change if sb_hiwat is different than default (then it
3861 * has been changed on purpose with setsockopt).
3862 * Make the socket buffers an integral number of mss units;
3863 * if the mss is larger than the socket buffer, decrease the mss.
3865 so = inp->inp_socket;
3866 SOCKBUF_LOCK(&so->so_snd);
3867 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3868 bufsize = metrics.rmx_sendpipe;
3870 bufsize = so->so_snd.sb_hiwat;
3874 bufsize = roundup(bufsize, mss);
3875 if (bufsize > sb_max)
3877 if (bufsize > so->so_snd.sb_hiwat)
3878 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3880 SOCKBUF_UNLOCK(&so->so_snd);
3882 * Sanity check: make sure that maxseg will be large
3883 * enough to allow some data on segments even if the
3884 * all the option space is used (40bytes). Otherwise
3885 * funny things may happen in tcp_output.
3887 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3889 tp->t_maxseg = max(mss, 64);
3891 SOCKBUF_LOCK(&so->so_rcv);
3892 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3893 bufsize = metrics.rmx_recvpipe;
3895 bufsize = so->so_rcv.sb_hiwat;
3896 if (bufsize > mss) {
3897 bufsize = roundup(bufsize, mss);
3898 if (bufsize > sb_max)
3900 if (bufsize > so->so_rcv.sb_hiwat)
3901 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3903 SOCKBUF_UNLOCK(&so->so_rcv);
3905 /* Check the interface for TSO capabilities. */
3906 if (cap.ifcap & CSUM_TSO) {
3907 tp->t_flags |= TF_TSO;
3908 tp->t_tsomax = cap.tsomax;
3909 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3910 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3915 * Determine the MSS option to send on an outgoing SYN.
3918 tcp_mssopt(struct in_conninfo *inc)
3921 uint32_t thcmtu = 0;
3922 uint32_t maxmtu = 0;
3925 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3928 if (inc->inc_flags & INC_ISIPV6) {
3929 mss = V_tcp_v6mssdflt;
3930 maxmtu = tcp_maxmtu6(inc, NULL);
3931 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3934 #if defined(INET) && defined(INET6)
3939 mss = V_tcp_mssdflt;
3940 maxmtu = tcp_maxmtu(inc, NULL);
3941 min_protoh = sizeof(struct tcpiphdr);
3944 #if defined(INET6) || defined(INET)
3945 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3948 if (maxmtu && thcmtu)
3949 mss = min(maxmtu, thcmtu) - min_protoh;
3950 else if (maxmtu || thcmtu)
3951 mss = max(maxmtu, thcmtu) - min_protoh;
3957 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
3959 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3960 int maxseg = tcp_maxseg(tp);
3962 INP_WLOCK_ASSERT(tp->t_inpcb);
3965 * Compute the amount of data that this ACK is indicating
3966 * (del_data) and an estimate of how many bytes are in the
3969 del_data = tp->sackhint.delivered_data;
3970 if (V_tcp_do_rfc6675_pipe)
3971 pipe = tcp_compute_pipe(tp);
3973 pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit;
3974 tp->sackhint.prr_delivered += del_data;
3976 * Proportional Rate Reduction
3978 if (pipe >= tp->snd_ssthresh) {
3979 if (tp->sackhint.recover_fs == 0)
3980 tp->sackhint.recover_fs =
3981 imax(1, tp->snd_nxt - tp->snd_una);
3982 snd_cnt = howmany((long)tp->sackhint.prr_delivered *
3983 tp->snd_ssthresh, tp->sackhint.recover_fs) -
3984 tp->sackhint.prr_out;
3986 if (V_tcp_do_prr_conservative)
3987 limit = tp->sackhint.prr_delivered -
3988 tp->sackhint.prr_out;
3990 limit = imax(tp->sackhint.prr_delivered -
3991 tp->sackhint.prr_out, del_data) +
3993 snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
3995 snd_cnt = imax(snd_cnt, 0) / maxseg;
3997 * Send snd_cnt new data into the network in response to this ack.
3998 * If there is going to be a SACK retransmission, adjust snd_cwnd
4001 if (IN_FASTRECOVERY(tp->t_flags)) {
4002 tp->snd_cwnd = imax(maxseg, tp->snd_nxt - tp->snd_recover +
4003 tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg));
4004 } else if (IN_CONGRECOVERY(tp->t_flags))
4005 tp->snd_cwnd = imax(maxseg, pipe - del_data +
4006 (snd_cnt * maxseg));
4010 * On a partial ack arrives, force the retransmission of the
4011 * next unacknowledged segment. Do not clear tp->t_dupacks.
4012 * By setting snd_nxt to ti_ack, this forces retransmission timer to
4016 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4018 tcp_seq onxt = tp->snd_nxt;
4019 uint32_t ocwnd = tp->snd_cwnd;
4020 u_int maxseg = tcp_maxseg(tp);
4022 INP_WLOCK_ASSERT(tp->t_inpcb);
4024 tcp_timer_activate(tp, TT_REXMT, 0);
4026 tp->snd_nxt = th->th_ack;
4028 * Set snd_cwnd to one segment beyond acknowledged offset.
4029 * (tp->snd_una has not yet been updated when this function is called.)
4031 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4032 tp->t_flags |= TF_ACKNOW;
4033 (void) tp->t_fb->tfb_tcp_output(tp);
4034 tp->snd_cwnd = ocwnd;
4035 if (SEQ_GT(onxt, tp->snd_nxt))
4038 * Partial window deflation. Relies on fact that tp->snd_una
4041 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4042 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4045 tp->snd_cwnd += maxseg;
4049 tcp_compute_pipe(struct tcpcb *tp)
4051 return (tp->snd_max - tp->snd_una +
4052 tp->sackhint.sack_bytes_rexmit -
4053 tp->sackhint.sacked_bytes);
4057 tcp_compute_initwnd(uint32_t maxseg)
4060 * Calculate the Initial Window, also used as Restart Window
4062 * RFC5681 Section 3.1 specifies the default conservative values.
4063 * RFC3390 specifies slightly more aggressive values.
4064 * RFC6928 increases it to ten segments.
4065 * Support for user specified value for initial flight size.
4067 if (V_tcp_initcwnd_segments)
4068 return min(V_tcp_initcwnd_segments * maxseg,
4069 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4070 else if (V_tcp_do_rfc3390)
4071 return min(4 * maxseg, max(2 * maxseg, 4380));
4073 /* Per RFC5681 Section 3.1 */
4075 return (2 * maxseg);
4076 else if (maxseg > 1095)
4077 return (3 * maxseg);
4079 return (4 * maxseg);