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) || !SOLISTENING(inp->inp_socket))) {
954 inp->inp_flowid = m->m_pkthdr.flowid;
955 inp->inp_flowtype = M_HASHTYPE_GET(m);
957 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
959 if (isipv6 && IPSEC_ENABLED(ipv6) &&
960 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
968 if (IPSEC_ENABLED(ipv4) &&
969 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
976 * Check the minimum TTL for socket.
978 if (inp->inp_ip_minttl != 0) {
981 if (inp->inp_ip_minttl > ip6->ip6_hlim)
985 if (inp->inp_ip_minttl > ip->ip_ttl)
990 * A previous connection in TIMEWAIT state is supposed to catch stray
991 * or duplicate segments arriving late. If this segment was a
992 * legitimate new connection attempt, the old INPCB gets removed and
993 * we can try again to find a listening socket.
995 * At this point, due to earlier optimism, we may hold only an inpcb
996 * lock, and not the inpcbinfo write lock. If so, we need to try to
997 * acquire it, or if that fails, acquire a reference on the inpcb,
998 * drop all locks, acquire a global write lock, and then re-acquire
999 * the inpcb lock. We may at that point discover that another thread
1000 * has tried to free the inpcb, in which case we need to loop back
1001 * and try to find a new inpcb to deliver to.
1003 * XXXRW: It may be time to rethink timewait locking.
1005 if (inp->inp_flags & INP_TIMEWAIT) {
1006 tcp_dooptions(&to, optp, optlen,
1007 (thflags & TH_SYN) ? TO_SYN : 0);
1009 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
1011 if (tcp_twcheck(inp, &to, th, m, tlen))
1013 return (IPPROTO_DONE);
1016 * The TCPCB may no longer exist if the connection is winding
1017 * down or it is in the CLOSED state. Either way we drop the
1018 * segment and send an appropriate response.
1020 tp = intotcpcb(inp);
1021 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1022 rstreason = BANDLIM_RST_CLOSEDPORT;
1026 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) {
1027 rstreason = BANDLIM_RST_CLOSEDPORT;
1032 if (tp->t_flags & TF_TOE) {
1033 tcp_offload_input(tp, m);
1034 m = NULL; /* consumed by the TOE driver */
1040 INP_WLOCK_ASSERT(inp);
1041 if (mac_inpcb_check_deliver(inp, m))
1044 so = inp->inp_socket;
1045 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1047 if (so->so_options & SO_DEBUG) {
1048 ostate = tp->t_state;
1051 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1054 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1057 #endif /* TCPDEBUG */
1059 * When the socket is accepting connections (the INPCB is in LISTEN
1060 * state) we look into the SYN cache if this is a new connection
1061 * attempt or the completion of a previous one.
1063 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so),
1064 ("%s: so accepting but tp %p not listening", __func__, tp));
1065 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) {
1066 struct in_conninfo inc;
1068 bzero(&inc, sizeof(inc));
1071 inc.inc_flags |= INC_ISIPV6;
1072 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1073 inc.inc_flags |= INC_IPV6MINMTU;
1074 inc.inc6_faddr = ip6->ip6_src;
1075 inc.inc6_laddr = ip6->ip6_dst;
1079 inc.inc_faddr = ip->ip_src;
1080 inc.inc_laddr = ip->ip_dst;
1082 inc.inc_fport = th->th_sport;
1083 inc.inc_lport = th->th_dport;
1084 inc.inc_fibnum = so->so_fibnum;
1087 * Check for an existing connection attempt in syncache if
1088 * the flag is only ACK. A successful lookup creates a new
1089 * socket appended to the listen queue in SYN_RECEIVED state.
1091 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1093 * Parse the TCP options here because
1094 * syncookies need access to the reflected
1097 tcp_dooptions(&to, optp, optlen, 0);
1099 * NB: syncache_expand() doesn't unlock
1100 * inp and tcpinfo locks.
1102 rstreason = syncache_expand(&inc, &to, th, &so, m, port);
1103 if (rstreason < 0) {
1105 * A failing TCP MD5 signature comparison
1106 * must result in the segment being dropped
1107 * and must not produce any response back
1111 } else if (rstreason == 0) {
1113 * No syncache entry or ACK was not
1114 * for our SYN/ACK. Send a RST.
1115 * NB: syncache did its own logging
1116 * of the failure cause.
1118 rstreason = BANDLIM_RST_OPENPORT;
1124 * We completed the 3-way handshake
1125 * but could not allocate a socket
1126 * either due to memory shortage,
1127 * listen queue length limits or
1128 * global socket limits. Send RST
1129 * or wait and have the remote end
1130 * retransmit the ACK for another
1133 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1134 log(LOG_DEBUG, "%s; %s: Listen socket: "
1135 "Socket allocation failed due to "
1136 "limits or memory shortage, %s\n",
1138 V_tcp_sc_rst_sock_fail ?
1139 "sending RST" : "try again");
1140 if (V_tcp_sc_rst_sock_fail) {
1141 rstreason = BANDLIM_UNLIMITED;
1147 * Socket is created in state SYN_RECEIVED.
1148 * Unlock the listen socket, lock the newly
1149 * created socket and update the tp variable.
1151 INP_WUNLOCK(inp); /* listen socket */
1152 inp = sotoinpcb(so);
1154 * New connection inpcb is already locked by
1155 * syncache_expand().
1157 INP_WLOCK_ASSERT(inp);
1158 tp = intotcpcb(inp);
1159 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1160 ("%s: ", __func__));
1162 * Process the segment and the data it
1163 * contains. tcp_do_segment() consumes
1164 * the mbuf chain and unlocks the inpcb.
1166 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1167 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1169 return (IPPROTO_DONE);
1172 * Segment flag validation for new connection attempts:
1174 * Our (SYN|ACK) response was rejected.
1175 * Check with syncache and remove entry to prevent
1178 * NB: syncache_chkrst does its own logging of failure
1181 if (thflags & TH_RST) {
1182 syncache_chkrst(&inc, th, m, port);
1186 * We can't do anything without SYN.
1188 if ((thflags & TH_SYN) == 0) {
1189 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1190 log(LOG_DEBUG, "%s; %s: Listen socket: "
1191 "SYN is missing, segment ignored\n",
1193 TCPSTAT_INC(tcps_badsyn);
1197 * (SYN|ACK) is bogus on a listen socket.
1199 if (thflags & TH_ACK) {
1200 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1201 log(LOG_DEBUG, "%s; %s: Listen socket: "
1202 "SYN|ACK invalid, segment rejected\n",
1204 syncache_badack(&inc, port); /* XXX: Not needed! */
1205 TCPSTAT_INC(tcps_badsyn);
1206 rstreason = BANDLIM_RST_OPENPORT;
1210 * If the drop_synfin option is enabled, drop all
1211 * segments with both the SYN and FIN bits set.
1212 * This prevents e.g. nmap from identifying the
1214 * XXX: Poor reasoning. nmap has other methods
1215 * and is constantly refining its stack detection
1217 * XXX: This is a violation of the TCP specification
1218 * and was used by RFC1644.
1220 if ((thflags & TH_FIN) && V_drop_synfin) {
1221 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1222 log(LOG_DEBUG, "%s; %s: Listen socket: "
1223 "SYN|FIN segment ignored (based on "
1224 "sysctl setting)\n", s, __func__);
1225 TCPSTAT_INC(tcps_badsyn);
1229 * Segment's flags are (SYN) or (SYN|FIN).
1231 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1232 * as they do not affect the state of the TCP FSM.
1233 * The data pointed to by TH_URG and th_urp is ignored.
1235 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1236 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1237 KASSERT(thflags & (TH_SYN),
1238 ("%s: Listen socket: TH_SYN not set", __func__));
1241 * If deprecated address is forbidden,
1242 * we do not accept SYN to deprecated interface
1243 * address to prevent any new inbound connection from
1244 * getting established.
1245 * When we do not accept SYN, we send a TCP RST,
1246 * with deprecated source address (instead of dropping
1247 * it). We compromise it as it is much better for peer
1248 * to send a RST, and RST will be the final packet
1251 * If we do not forbid deprecated addresses, we accept
1252 * the SYN packet. RFC2462 does not suggest dropping
1254 * If we decipher RFC2462 5.5.4, it says like this:
1255 * 1. use of deprecated addr with existing
1256 * communication is okay - "SHOULD continue to be
1258 * 2. use of it with new communication:
1259 * (2a) "SHOULD NOT be used if alternate address
1260 * with sufficient scope is available"
1261 * (2b) nothing mentioned otherwise.
1262 * Here we fall into (2b) case as we have no choice in
1263 * our source address selection - we must obey the peer.
1265 * The wording in RFC2462 is confusing, and there are
1266 * multiple description text for deprecated address
1267 * handling - worse, they are not exactly the same.
1268 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1270 if (isipv6 && !V_ip6_use_deprecated) {
1271 struct in6_ifaddr *ia6;
1273 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1275 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1276 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1277 log(LOG_DEBUG, "%s; %s: Listen socket: "
1278 "Connection attempt to deprecated "
1279 "IPv6 address rejected\n",
1281 rstreason = BANDLIM_RST_OPENPORT;
1287 * Basic sanity checks on incoming SYN requests:
1288 * Don't respond if the destination is a link layer
1289 * broadcast according to RFC1122 4.2.3.10, p. 104.
1290 * If it is from this socket it must be forged.
1291 * Don't respond if the source or destination is a
1292 * global or subnet broad- or multicast address.
1293 * Note that it is quite possible to receive unicast
1294 * link-layer packets with a broadcast IP address. Use
1295 * in_broadcast() to find them.
1297 if (m->m_flags & (M_BCAST|M_MCAST)) {
1298 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1299 log(LOG_DEBUG, "%s; %s: Listen socket: "
1300 "Connection attempt from broad- or multicast "
1301 "link layer address ignored\n", s, __func__);
1306 if (th->th_dport == th->th_sport &&
1307 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1308 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1309 log(LOG_DEBUG, "%s; %s: Listen socket: "
1310 "Connection attempt to/from self "
1311 "ignored\n", s, __func__);
1314 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1315 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1316 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1317 log(LOG_DEBUG, "%s; %s: Listen socket: "
1318 "Connection attempt from/to multicast "
1319 "address ignored\n", s, __func__);
1324 #if defined(INET) && defined(INET6)
1329 if (th->th_dport == th->th_sport &&
1330 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1331 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1332 log(LOG_DEBUG, "%s; %s: Listen socket: "
1333 "Connection attempt from/to self "
1334 "ignored\n", s, __func__);
1337 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1338 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1339 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1340 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1341 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1342 log(LOG_DEBUG, "%s; %s: Listen socket: "
1343 "Connection attempt from/to broad- "
1344 "or multicast address ignored\n",
1351 * SYN appears to be valid. Create compressed TCP state
1355 if (so->so_options & SO_DEBUG)
1356 tcp_trace(TA_INPUT, ostate, tp,
1357 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1359 TCP_PROBE3(debug__input, tp, th, m);
1360 tcp_dooptions(&to, optp, optlen, TO_SYN);
1361 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos,
1363 goto tfo_socket_result;
1366 * Entry added to syncache and mbuf consumed.
1367 * Only the listen socket is unlocked by syncache_add().
1369 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1370 return (IPPROTO_DONE);
1371 } else if (tp->t_state == TCPS_LISTEN) {
1373 * When a listen socket is torn down the SO_ACCEPTCONN
1374 * flag is removed first while connections are drained
1375 * from the accept queue in a unlock/lock cycle of the
1376 * ACCEPT_LOCK, opening a race condition allowing a SYN
1377 * attempt go through unhandled.
1381 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1382 if (tp->t_flags & TF_SIGNATURE) {
1383 tcp_dooptions(&to, optp, optlen, thflags);
1384 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1385 TCPSTAT_INC(tcps_sig_err_nosigopt);
1388 if (!TCPMD5_ENABLED() ||
1389 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1393 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1396 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1397 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1398 * the inpcb, and unlocks pcbinfo.
1400 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1401 return (IPPROTO_DONE);
1404 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1407 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1410 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1411 m = NULL; /* mbuf chain got consumed. */
1416 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1422 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1427 return (IPPROTO_DONE);
1431 * Automatic sizing of receive socket buffer. Often the send
1432 * buffer size is not optimally adjusted to the actual network
1433 * conditions at hand (delay bandwidth product). Setting the
1434 * buffer size too small limits throughput on links with high
1435 * bandwidth and high delay (eg. trans-continental/oceanic links).
1437 * On the receive side the socket buffer memory is only rarely
1438 * used to any significant extent. This allows us to be much
1439 * more aggressive in scaling the receive socket buffer. For
1440 * the case that the buffer space is actually used to a large
1441 * extent and we run out of kernel memory we can simply drop
1442 * the new segments; TCP on the sender will just retransmit it
1443 * later. Setting the buffer size too big may only consume too
1444 * much kernel memory if the application doesn't read() from
1445 * the socket or packet loss or reordering makes use of the
1448 * The criteria to step up the receive buffer one notch are:
1449 * 1. Application has not set receive buffer size with
1450 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1451 * 2. the number of bytes received during 1/2 of an sRTT
1452 * is at least 3/8 of the current socket buffer size.
1453 * 3. receive buffer size has not hit maximal automatic size;
1455 * If all of the criteria are met we increaset the socket buffer
1456 * by a 1/2 (bounded by the max). This allows us to keep ahead
1457 * of slow-start but also makes it so our peer never gets limited
1458 * by our rwnd which we then open up causing a burst.
1460 * This algorithm does two steps per RTT at most and only if
1461 * we receive a bulk stream w/o packet losses or reorderings.
1462 * Shrinking the buffer during idle times is not necessary as
1463 * it doesn't consume any memory when idle.
1465 * TODO: Only step up if the application is actually serving
1466 * the buffer to better manage the socket buffer resources.
1469 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1470 struct tcpcb *tp, int tlen)
1474 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1475 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1476 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1477 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1478 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1479 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1480 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1482 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1484 /* Start over with next RTT. */
1488 tp->rfbuf_cnt += tlen; /* add up */
1494 tcp_input(struct mbuf **mp, int *offp, int proto)
1496 return(tcp_input_with_port(mp, offp, proto, 0));
1500 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1503 * Since tp might be gone if the session entered
1504 * the TIME_WAIT state before coming here, we need
1505 * to check if the socket is still connected.
1513 INP_LOCK_ASSERT(tp->t_inpcb);
1514 if (tp->t_flags & TF_WAKESOR) {
1515 tp->t_flags &= ~TF_WAKESOR;
1516 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1517 sorwakeup_locked(so);
1522 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1523 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1525 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1526 int rstreason, todrop, win, incforsyn = 0;
1530 struct in_conninfo *inc;
1538 * The size of tcp_saveipgen must be the size of the max ip header,
1541 u_char tcp_saveipgen[IP6_HDR_LEN];
1542 struct tcphdr tcp_savetcp;
1545 thflags = th->th_flags;
1546 inc = &tp->t_inpcb->inp_inc;
1547 tp->sackhint.last_sack_ack = 0;
1549 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1552 INP_WLOCK_ASSERT(tp->t_inpcb);
1553 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1555 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1559 /* Save segment, if requested. */
1560 tcp_pcap_add(th, m, &(tp->t_inpkts));
1562 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1565 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1566 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1567 log(LOG_DEBUG, "%s; %s: "
1568 "SYN|FIN segment ignored (based on "
1569 "sysctl setting)\n", s, __func__);
1576 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1577 * check SEQ.ACK first.
1579 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1580 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1581 rstreason = BANDLIM_UNLIMITED;
1586 * Segment received on connection.
1587 * Reset idle time and keep-alive timer.
1588 * XXX: This should be done after segment
1589 * validation to ignore broken/spoofed segs.
1591 tp->t_rcvtime = ticks;
1594 * Scale up the window into a 32-bit value.
1595 * For the SYN_SENT state the scale is zero.
1597 tiwin = th->th_win << tp->snd_scale;
1599 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1603 * TCP ECN processing.
1605 if (tp->t_flags2 & TF2_ECN_PERMIT) {
1606 if (thflags & TH_CWR) {
1607 tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1608 tp->t_flags |= TF_ACKNOW;
1610 switch (iptos & IPTOS_ECN_MASK) {
1612 tp->t_flags2 |= TF2_ECN_SND_ECE;
1613 TCPSTAT_INC(tcps_ecn_ce);
1615 case IPTOS_ECN_ECT0:
1616 TCPSTAT_INC(tcps_ecn_ect0);
1618 case IPTOS_ECN_ECT1:
1619 TCPSTAT_INC(tcps_ecn_ect1);
1623 /* Process a packet differently from RFC3168. */
1624 cc_ecnpkt_handler(tp, th, iptos);
1626 /* Congestion experienced. */
1627 if (thflags & TH_ECE) {
1628 cc_cong_signal(tp, th, CC_ECN);
1633 * Parse options on any incoming segment.
1635 tcp_dooptions(&to, (u_char *)(th + 1),
1636 (th->th_off << 2) - sizeof(struct tcphdr),
1637 (thflags & TH_SYN) ? TO_SYN : 0);
1639 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1640 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1641 (to.to_flags & TOF_SIGNATURE) == 0) {
1642 TCPSTAT_INC(tcps_sig_err_sigopt);
1643 /* XXX: should drop? */
1647 * If echoed timestamp is later than the current time,
1648 * fall back to non RFC1323 RTT calculation. Normalize
1649 * timestamp if syncookies were used when this connection
1652 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1653 to.to_tsecr -= tp->ts_offset;
1654 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1656 else if (tp->t_flags & TF_PREVVALID &&
1657 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1658 cc_cong_signal(tp, th, CC_RTO_ERR);
1661 * Process options only when we get SYN/ACK back. The SYN case
1662 * for incoming connections is handled in tcp_syncache.
1663 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1664 * or <SYN,ACK>) segment itself is never scaled.
1665 * XXX this is traditional behavior, may need to be cleaned up.
1667 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1668 /* Handle parallel SYN for ECN */
1669 if (!(thflags & TH_ACK) &&
1670 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1671 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1672 tp->t_flags2 |= TF2_ECN_PERMIT;
1673 tp->t_flags2 |= TF2_ECN_SND_ECE;
1674 TCPSTAT_INC(tcps_ecn_shs);
1676 if ((to.to_flags & TOF_SCALE) &&
1677 (tp->t_flags & TF_REQ_SCALE) &&
1678 !(tp->t_flags & TF_NOOPT)) {
1679 tp->t_flags |= TF_RCVD_SCALE;
1680 tp->snd_scale = to.to_wscale;
1682 tp->t_flags &= ~TF_REQ_SCALE;
1684 * Initial send window. It will be updated with
1685 * the next incoming segment to the scaled value.
1687 tp->snd_wnd = th->th_win;
1688 if ((to.to_flags & TOF_TS) &&
1689 (tp->t_flags & TF_REQ_TSTMP) &&
1690 !(tp->t_flags & TF_NOOPT)) {
1691 tp->t_flags |= TF_RCVD_TSTMP;
1692 tp->ts_recent = to.to_tsval;
1693 tp->ts_recent_age = tcp_ts_getticks();
1695 tp->t_flags &= ~TF_REQ_TSTMP;
1696 if (to.to_flags & TOF_MSS)
1697 tcp_mss(tp, to.to_mss);
1698 if ((tp->t_flags & TF_SACK_PERMIT) &&
1699 (!(to.to_flags & TOF_SACKPERM) ||
1700 (tp->t_flags & TF_NOOPT)))
1701 tp->t_flags &= ~TF_SACK_PERMIT;
1702 if (IS_FASTOPEN(tp->t_flags)) {
1703 if ((to.to_flags & TOF_FASTOPEN) &&
1704 !(tp->t_flags & TF_NOOPT)) {
1707 if (to.to_flags & TOF_MSS)
1710 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1714 tcp_fastopen_update_cache(tp, mss,
1715 to.to_tfo_len, to.to_tfo_cookie);
1717 tcp_fastopen_disable_path(tp);
1722 * If timestamps were negotiated during SYN/ACK and a
1723 * segment without a timestamp is received, silently drop
1724 * the segment, unless it is a RST segment or missing timestamps are
1726 * See section 3.2 of RFC 7323.
1728 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1729 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1730 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1731 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1732 "segment processed normally\n",
1737 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1738 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1739 "segment silently dropped\n", s, __func__);
1746 * If timestamps were not negotiated during SYN/ACK and a
1747 * segment with a timestamp is received, ignore the
1748 * timestamp and process the packet normally.
1749 * See section 3.2 of RFC 7323.
1751 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1752 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1753 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1754 "segment processed normally\n", s, __func__);
1760 * Header prediction: check for the two common cases
1761 * of a uni-directional data xfer. If the packet has
1762 * no control flags, is in-sequence, the window didn't
1763 * change and we're not retransmitting, it's a
1764 * candidate. If the length is zero and the ack moved
1765 * forward, we're the sender side of the xfer. Just
1766 * free the data acked & wake any higher level process
1767 * that was blocked waiting for space. If the length
1768 * is non-zero and the ack didn't move, we're the
1769 * receiver side. If we're getting packets in-order
1770 * (the reassembly queue is empty), add the data to
1771 * the socket buffer and note that we need a delayed ack.
1772 * Make sure that the hidden state-flags are also off.
1773 * Since we check for TCPS_ESTABLISHED first, it can only
1776 if (tp->t_state == TCPS_ESTABLISHED &&
1777 th->th_seq == tp->rcv_nxt &&
1778 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1779 tp->snd_nxt == tp->snd_max &&
1780 tiwin && tiwin == tp->snd_wnd &&
1781 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1783 ((to.to_flags & TOF_TS) == 0 ||
1784 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1786 * If last ACK falls within this segment's sequence numbers,
1787 * record the timestamp.
1788 * NOTE that the test is modified according to the latest
1789 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1791 if ((to.to_flags & TOF_TS) != 0 &&
1792 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1793 tp->ts_recent_age = tcp_ts_getticks();
1794 tp->ts_recent = to.to_tsval;
1798 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1799 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1800 !IN_RECOVERY(tp->t_flags) &&
1801 (to.to_flags & TOF_SACK) == 0 &&
1802 TAILQ_EMPTY(&tp->snd_holes)) {
1804 * This is a pure ack for outstanding data.
1806 TCPSTAT_INC(tcps_predack);
1809 * "bad retransmit" recovery without timestamps.
1811 if ((to.to_flags & TOF_TS) == 0 &&
1812 tp->t_rxtshift == 1 &&
1813 tp->t_flags & TF_PREVVALID &&
1814 (int)(ticks - tp->t_badrxtwin) < 0) {
1815 cc_cong_signal(tp, th, CC_RTO_ERR);
1819 * Recalculate the transmit timer / rtt.
1821 * Some boxes send broken timestamp replies
1822 * during the SYN+ACK phase, ignore
1823 * timestamps of 0 or we could calculate a
1824 * huge RTT and blow up the retransmit timer.
1826 if ((to.to_flags & TOF_TS) != 0 &&
1830 t = tcp_ts_getticks() - to.to_tsecr;
1831 if (!tp->t_rttlow || tp->t_rttlow > t)
1834 TCP_TS_TO_TICKS(t) + 1);
1835 } else if (tp->t_rtttime &&
1836 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1837 if (!tp->t_rttlow ||
1838 tp->t_rttlow > ticks - tp->t_rtttime)
1839 tp->t_rttlow = ticks - tp->t_rtttime;
1841 ticks - tp->t_rtttime);
1843 acked = BYTES_THIS_ACK(tp, th);
1846 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1847 hhook_run_tcp_est_in(tp, th, &to);
1850 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1851 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1852 sbdrop(&so->so_snd, acked);
1853 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1854 SEQ_LEQ(th->th_ack, tp->snd_recover))
1855 tp->snd_recover = th->th_ack - 1;
1858 * Let the congestion control algorithm update
1859 * congestion control related information. This
1860 * typically means increasing the congestion
1863 cc_ack_received(tp, th, nsegs, CC_ACK);
1865 tp->snd_una = th->th_ack;
1867 * Pull snd_wl2 up to prevent seq wrap relative
1870 tp->snd_wl2 = th->th_ack;
1875 * If all outstanding data are acked, stop
1876 * retransmit timer, otherwise restart timer
1877 * using current (possibly backed-off) value.
1878 * If process is waiting for space,
1879 * wakeup/selwakeup/signal. If data
1880 * are ready to send, let tcp_output
1881 * decide between more output or persist.
1884 if (so->so_options & SO_DEBUG)
1885 tcp_trace(TA_INPUT, ostate, tp,
1886 (void *)tcp_saveipgen,
1889 TCP_PROBE3(debug__input, tp, th, m);
1890 if (tp->snd_una == tp->snd_max)
1891 tcp_timer_activate(tp, TT_REXMT, 0);
1892 else if (!tcp_timer_active(tp, TT_PERSIST))
1893 tcp_timer_activate(tp, TT_REXMT,
1896 if (sbavail(&so->so_snd))
1897 (void) tp->t_fb->tfb_tcp_output(tp);
1900 } else if (th->th_ack == tp->snd_una &&
1901 tlen <= sbspace(&so->so_rcv)) {
1902 int newsize = 0; /* automatic sockbuf scaling */
1905 * This is a pure, in-sequence data packet with
1906 * nothing on the reassembly queue and we have enough
1907 * buffer space to take it.
1909 /* Clean receiver SACK report if present */
1910 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1911 tcp_clean_sackreport(tp);
1912 TCPSTAT_INC(tcps_preddat);
1913 tp->rcv_nxt += tlen;
1915 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1916 (tp->t_fbyte_in == 0)) {
1917 tp->t_fbyte_in = ticks;
1918 if (tp->t_fbyte_in == 0)
1920 if (tp->t_fbyte_out && tp->t_fbyte_in)
1921 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1924 * Pull snd_wl1 up to prevent seq wrap relative to
1927 tp->snd_wl1 = th->th_seq;
1929 * Pull rcv_up up to prevent seq wrap relative to
1932 tp->rcv_up = tp->rcv_nxt;
1933 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1934 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1936 if (so->so_options & SO_DEBUG)
1937 tcp_trace(TA_INPUT, ostate, tp,
1938 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1940 TCP_PROBE3(debug__input, tp, th, m);
1942 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1944 /* Add data to socket buffer. */
1945 SOCKBUF_LOCK(&so->so_rcv);
1946 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1950 * Set new socket buffer size.
1951 * Give up when limit is reached.
1954 if (!sbreserve_locked(&so->so_rcv,
1956 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1957 m_adj(m, drop_hdrlen); /* delayed header drop */
1958 sbappendstream_locked(&so->so_rcv, m, 0);
1960 /* NB: sorwakeup_locked() does an implicit unlock. */
1961 sorwakeup_locked(so);
1962 if (DELAY_ACK(tp, tlen)) {
1963 tp->t_flags |= TF_DELACK;
1965 tp->t_flags |= TF_ACKNOW;
1966 tp->t_fb->tfb_tcp_output(tp);
1973 * Calculate amount of space in receive window,
1974 * and then do TCP input processing.
1975 * Receive window is amount of space in rcv queue,
1976 * but not less than advertised window.
1978 win = sbspace(&so->so_rcv);
1981 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1983 switch (tp->t_state) {
1985 * If the state is SYN_RECEIVED:
1986 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1988 case TCPS_SYN_RECEIVED:
1989 if ((thflags & TH_ACK) &&
1990 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1991 SEQ_GT(th->th_ack, tp->snd_max))) {
1992 rstreason = BANDLIM_RST_OPENPORT;
1995 if (IS_FASTOPEN(tp->t_flags)) {
1997 * When a TFO connection is in SYN_RECEIVED, the
1998 * only valid packets are the initial SYN, a
1999 * retransmit/copy of the initial SYN (possibly with
2000 * a subset of the original data), a valid ACK, a
2003 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
2004 rstreason = BANDLIM_RST_OPENPORT;
2006 } else if (thflags & TH_SYN) {
2007 /* non-initial SYN is ignored */
2008 if ((tcp_timer_active(tp, TT_DELACK) ||
2009 tcp_timer_active(tp, TT_REXMT)))
2011 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2018 * If the state is SYN_SENT:
2019 * if seg contains a RST with valid ACK (SEQ.ACK has already
2020 * been verified), then drop the connection.
2021 * if seg contains a RST without an ACK, drop the seg.
2022 * if seg does not contain SYN, then drop the seg.
2023 * Otherwise this is an acceptable SYN segment
2024 * initialize tp->rcv_nxt and tp->irs
2025 * if seg contains ack then advance tp->snd_una
2026 * if seg contains an ECE and ECN support is enabled, the stream
2028 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2029 * arrange for segment to be acked (eventually)
2030 * continue processing rest of data/controls, beginning with URG
2033 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2034 TCP_PROBE5(connect__refused, NULL, tp,
2036 tp = tcp_drop(tp, ECONNREFUSED);
2038 if (thflags & TH_RST)
2040 if (!(thflags & TH_SYN))
2043 tp->irs = th->th_seq;
2045 if (thflags & TH_ACK) {
2046 int tfo_partial_ack = 0;
2048 TCPSTAT_INC(tcps_connects);
2051 mac_socketpeer_set_from_mbuf(m, so);
2053 /* Do window scaling on this connection? */
2054 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2055 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2056 tp->rcv_scale = tp->request_r_scale;
2058 tp->rcv_adv += min(tp->rcv_wnd,
2059 TCP_MAXWIN << tp->rcv_scale);
2060 tp->snd_una++; /* SYN is acked */
2062 * If not all the data that was sent in the TFO SYN
2063 * has been acked, resend the remainder right away.
2065 if (IS_FASTOPEN(tp->t_flags) &&
2066 (tp->snd_una != tp->snd_max)) {
2067 tp->snd_nxt = th->th_ack;
2068 tfo_partial_ack = 1;
2071 * If there's data, delay ACK; if there's also a FIN
2072 * ACKNOW will be turned on later.
2074 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2075 tcp_timer_activate(tp, TT_DELACK,
2078 tp->t_flags |= TF_ACKNOW;
2080 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2081 (V_tcp_do_ecn == 1)) {
2082 tp->t_flags2 |= TF2_ECN_PERMIT;
2083 TCPSTAT_INC(tcps_ecn_shs);
2087 * Received <SYN,ACK> in SYN_SENT[*] state.
2089 * SYN_SENT --> ESTABLISHED
2090 * SYN_SENT* --> FIN_WAIT_1
2092 tp->t_starttime = ticks;
2093 if (tp->t_flags & TF_NEEDFIN) {
2094 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2095 tp->t_flags &= ~TF_NEEDFIN;
2098 tcp_state_change(tp, TCPS_ESTABLISHED);
2099 TCP_PROBE5(connect__established, NULL, tp,
2102 tcp_timer_activate(tp, TT_KEEP,
2107 * Received initial SYN in SYN-SENT[*] state =>
2108 * simultaneous open.
2109 * If it succeeds, connection is * half-synchronized.
2110 * Otherwise, do 3-way handshake:
2111 * SYN-SENT -> SYN-RECEIVED
2112 * SYN-SENT* -> SYN-RECEIVED*
2114 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2115 tcp_timer_activate(tp, TT_REXMT, 0);
2116 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2119 INP_WLOCK_ASSERT(tp->t_inpcb);
2122 * Advance th->th_seq to correspond to first data byte.
2123 * If data, trim to stay within window,
2124 * dropping FIN if necessary.
2127 if (tlen > tp->rcv_wnd) {
2128 todrop = tlen - tp->rcv_wnd;
2132 TCPSTAT_INC(tcps_rcvpackafterwin);
2133 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2135 tp->snd_wl1 = th->th_seq - 1;
2136 tp->rcv_up = th->th_seq;
2138 * Client side of transaction: already sent SYN and data.
2139 * If the remote host used T/TCP to validate the SYN,
2140 * our data will be ACK'd; if so, enter normal data segment
2141 * processing in the middle of step 5, ack processing.
2142 * Otherwise, goto step 6.
2144 if (thflags & TH_ACK)
2150 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2151 * do normal processing.
2153 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2157 break; /* continue normal processing */
2161 * States other than LISTEN or SYN_SENT.
2162 * First check the RST flag and sequence number since reset segments
2163 * are exempt from the timestamp and connection count tests. This
2164 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2165 * below which allowed reset segments in half the sequence space
2166 * to fall though and be processed (which gives forged reset
2167 * segments with a random sequence number a 50 percent chance of
2168 * killing a connection).
2169 * Then check timestamp, if present.
2170 * Then check the connection count, if present.
2171 * Then check that at least some bytes of segment are within
2172 * receive window. If segment begins before rcv_nxt,
2173 * drop leading data (and SYN); if nothing left, just ack.
2175 if (thflags & TH_RST) {
2177 * RFC5961 Section 3.2
2179 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2180 * - If RST is in window, we send challenge ACK.
2182 * Note: to take into account delayed ACKs, we should
2183 * test against last_ack_sent instead of rcv_nxt.
2184 * Note 2: we handle special case of closed window, not
2185 * covered by the RFC.
2187 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2188 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2189 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2190 KASSERT(tp->t_state != TCPS_SYN_SENT,
2191 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2194 if (V_tcp_insecure_rst ||
2195 tp->last_ack_sent == th->th_seq) {
2196 TCPSTAT_INC(tcps_drops);
2197 /* Drop the connection. */
2198 switch (tp->t_state) {
2199 case TCPS_SYN_RECEIVED:
2200 so->so_error = ECONNREFUSED;
2202 case TCPS_ESTABLISHED:
2203 case TCPS_FIN_WAIT_1:
2204 case TCPS_FIN_WAIT_2:
2205 case TCPS_CLOSE_WAIT:
2208 so->so_error = ECONNRESET;
2215 TCPSTAT_INC(tcps_badrst);
2216 /* Send challenge ACK. */
2217 tcp_respond(tp, mtod(m, void *), th, m,
2218 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2219 tp->last_ack_sent = tp->rcv_nxt;
2227 * RFC5961 Section 4.2
2228 * Send challenge ACK for any SYN in synchronized state.
2230 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2231 tp->t_state != TCPS_SYN_RECEIVED) {
2232 TCPSTAT_INC(tcps_badsyn);
2233 if (V_tcp_insecure_syn &&
2234 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2235 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2236 tp = tcp_drop(tp, ECONNRESET);
2237 rstreason = BANDLIM_UNLIMITED;
2239 /* Send challenge ACK. */
2240 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2241 tp->snd_nxt, TH_ACK);
2242 tp->last_ack_sent = tp->rcv_nxt;
2249 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2250 * and it's less than ts_recent, drop it.
2252 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2253 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2254 /* Check to see if ts_recent is over 24 days old. */
2255 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2257 * Invalidate ts_recent. If this segment updates
2258 * ts_recent, the age will be reset later and ts_recent
2259 * will get a valid value. If it does not, setting
2260 * ts_recent to zero will at least satisfy the
2261 * requirement that zero be placed in the timestamp
2262 * echo reply when ts_recent isn't valid. The
2263 * age isn't reset until we get a valid ts_recent
2264 * because we don't want out-of-order segments to be
2265 * dropped when ts_recent is old.
2269 TCPSTAT_INC(tcps_rcvduppack);
2270 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2271 TCPSTAT_INC(tcps_pawsdrop);
2279 * In the SYN-RECEIVED state, validate that the packet belongs to
2280 * this connection before trimming the data to fit the receive
2281 * window. Check the sequence number versus IRS since we know
2282 * the sequence numbers haven't wrapped. This is a partial fix
2283 * for the "LAND" DoS attack.
2285 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2286 rstreason = BANDLIM_RST_OPENPORT;
2290 todrop = tp->rcv_nxt - th->th_seq;
2292 if (thflags & TH_SYN) {
2302 * Following if statement from Stevens, vol. 2, p. 960.
2305 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2307 * Any valid FIN must be to the left of the window.
2308 * At this point the FIN must be a duplicate or out
2309 * of sequence; drop it.
2314 * Send an ACK to resynchronize and drop any data.
2315 * But keep on processing for RST or ACK.
2317 tp->t_flags |= TF_ACKNOW;
2319 TCPSTAT_INC(tcps_rcvduppack);
2320 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2322 TCPSTAT_INC(tcps_rcvpartduppack);
2323 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2326 * DSACK - add SACK block for dropped range
2328 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2329 tcp_update_sack_list(tp, th->th_seq,
2330 th->th_seq + todrop);
2332 * ACK now, as the next in-sequence segment
2333 * will clear the DSACK block again
2335 tp->t_flags |= TF_ACKNOW;
2337 drop_hdrlen += todrop; /* drop from the top afterwards */
2338 th->th_seq += todrop;
2340 if (th->th_urp > todrop)
2341 th->th_urp -= todrop;
2349 * If new data are received on a connection after the
2350 * user processes are gone, then RST the other end.
2352 if ((so->so_state & SS_NOFDREF) &&
2353 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2354 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2355 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2356 "after socket was closed, "
2357 "sending RST and removing tcpcb\n",
2358 s, __func__, tcpstates[tp->t_state], tlen);
2362 TCPSTAT_INC(tcps_rcvafterclose);
2363 rstreason = BANDLIM_UNLIMITED;
2368 * If segment ends after window, drop trailing data
2369 * (and PUSH and FIN); if nothing left, just ACK.
2371 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2373 TCPSTAT_INC(tcps_rcvpackafterwin);
2374 if (todrop >= tlen) {
2375 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2377 * If window is closed can only take segments at
2378 * window edge, and have to drop data and PUSH from
2379 * incoming segments. Continue processing, but
2380 * remember to ack. Otherwise, drop segment
2383 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2384 tp->t_flags |= TF_ACKNOW;
2385 TCPSTAT_INC(tcps_rcvwinprobe);
2389 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2392 thflags &= ~(TH_PUSH|TH_FIN);
2396 * If last ACK falls within this segment's sequence numbers,
2397 * record its timestamp.
2399 * 1) That the test incorporates suggestions from the latest
2400 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2401 * 2) That updating only on newer timestamps interferes with
2402 * our earlier PAWS tests, so this check should be solely
2403 * predicated on the sequence space of this segment.
2404 * 3) That we modify the segment boundary check to be
2405 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2406 * instead of RFC1323's
2407 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2408 * This modified check allows us to overcome RFC1323's
2409 * limitations as described in Stevens TCP/IP Illustrated
2410 * Vol. 2 p.869. In such cases, we can still calculate the
2411 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2413 if ((to.to_flags & TOF_TS) != 0 &&
2414 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2415 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2416 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2417 tp->ts_recent_age = tcp_ts_getticks();
2418 tp->ts_recent = to.to_tsval;
2422 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2423 * flag is on (half-synchronized state), then queue data for
2424 * later processing; else drop segment and return.
2426 if ((thflags & TH_ACK) == 0) {
2427 if (tp->t_state == TCPS_SYN_RECEIVED ||
2428 (tp->t_flags & TF_NEEDSYN)) {
2429 if (tp->t_state == TCPS_SYN_RECEIVED &&
2430 IS_FASTOPEN(tp->t_flags)) {
2431 tp->snd_wnd = tiwin;
2435 } else if (tp->t_flags & TF_ACKNOW)
2444 switch (tp->t_state) {
2446 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2447 * ESTABLISHED state and continue processing.
2448 * The ACK was checked above.
2450 case TCPS_SYN_RECEIVED:
2452 TCPSTAT_INC(tcps_connects);
2454 /* Do window scaling? */
2455 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2456 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2457 tp->rcv_scale = tp->request_r_scale;
2459 tp->snd_wnd = tiwin;
2462 * SYN-RECEIVED -> ESTABLISHED
2463 * SYN-RECEIVED* -> FIN-WAIT-1
2465 tp->t_starttime = ticks;
2466 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2467 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2468 tp->t_tfo_pending = NULL;
2470 if (tp->t_flags & TF_NEEDFIN) {
2471 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2472 tp->t_flags &= ~TF_NEEDFIN;
2474 tcp_state_change(tp, TCPS_ESTABLISHED);
2475 TCP_PROBE5(accept__established, NULL, tp,
2478 * TFO connections call cc_conn_init() during SYN
2479 * processing. Calling it again here for such
2480 * connections is not harmless as it would undo the
2481 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2484 if (!IS_FASTOPEN(tp->t_flags))
2486 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2489 * Account for the ACK of our SYN prior to
2490 * regular ACK processing below, except for
2491 * simultaneous SYN, which is handled later.
2493 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2496 * If segment contains data or ACK, will call tcp_reass()
2497 * later; if not, do so now to pass queued data to user.
2499 if (tlen == 0 && (thflags & TH_FIN) == 0) {
2500 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2502 tcp_handle_wakeup(tp, so);
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 tcp_do_prr_ack(tp, th, &to);
2610 } else if ((tp->t_flags & TF_SACK_PERMIT) &&
2611 (to.to_flags & TOF_SACK) &&
2612 IN_FASTRECOVERY(tp->t_flags)) {
2616 * Compute the amount of data in flight first.
2617 * We can inject new data into the pipe iff
2618 * we have less than 1/2 the original window's
2619 * worth of data in flight.
2621 if (V_tcp_do_rfc6675_pipe)
2622 awnd = tcp_compute_pipe(tp);
2624 awnd = (tp->snd_nxt - tp->snd_fack) +
2625 tp->sackhint.sack_bytes_rexmit;
2627 if (awnd < tp->snd_ssthresh) {
2628 tp->snd_cwnd += maxseg;
2629 if (tp->snd_cwnd > tp->snd_ssthresh)
2630 tp->snd_cwnd = tp->snd_ssthresh;
2633 tp->snd_cwnd += maxseg;
2634 (void) tp->t_fb->tfb_tcp_output(tp);
2636 } else if (tp->t_dupacks == tcprexmtthresh ||
2637 (tp->t_flags & TF_SACK_PERMIT &&
2638 V_tcp_do_rfc6675_pipe &&
2639 tp->sackhint.sacked_bytes >
2640 (tcprexmtthresh - 1) * maxseg)) {
2643 * Above is the RFC6675 trigger condition of
2644 * more than (dupthresh-1)*maxseg sacked data.
2645 * If the count of holes in the
2646 * scoreboard is >= dupthresh, we could
2647 * also enter loss recovery, but don't
2648 * have that value readily available.
2650 tp->t_dupacks = tcprexmtthresh;
2651 tcp_seq onxt = tp->snd_nxt;
2654 * If we're doing sack, or prr, check
2655 * to see if we're already in sack
2656 * recovery. If we're not doing sack,
2657 * check to see if we're in newreno
2661 (tp->t_flags & TF_SACK_PERMIT)) {
2662 if (IN_FASTRECOVERY(tp->t_flags)) {
2667 if (SEQ_LEQ(th->th_ack,
2673 /* Congestion signal before ack. */
2674 cc_cong_signal(tp, th, CC_NDUPACK);
2675 cc_ack_received(tp, th, nsegs,
2677 tcp_timer_activate(tp, TT_REXMT, 0);
2681 * snd_ssthresh is already updated by
2684 if ((tp->t_flags & TF_SACK_PERMIT) &&
2685 (to.to_flags & TOF_SACK)) {
2686 tp->sackhint.prr_delivered =
2687 tp->sackhint.sacked_bytes;
2689 tp->sackhint.prr_delivered =
2690 imin(tp->snd_max - tp->snd_una,
2691 imin(INT_MAX / 65536,
2692 tp->t_dupacks) * maxseg);
2694 tp->sackhint.recover_fs = max(1,
2695 tp->snd_nxt - tp->snd_una);
2697 if ((tp->t_flags & TF_SACK_PERMIT) &&
2698 (to.to_flags & TOF_SACK)) {
2700 tcps_sack_recovery_episode);
2701 tp->snd_recover = tp->snd_nxt;
2702 tp->snd_cwnd = maxseg;
2703 (void) tp->t_fb->tfb_tcp_output(tp);
2704 if (SEQ_GT(th->th_ack, tp->snd_una))
2705 goto resume_partialack;
2708 tp->snd_nxt = th->th_ack;
2709 tp->snd_cwnd = maxseg;
2710 (void) tp->t_fb->tfb_tcp_output(tp);
2711 KASSERT(tp->snd_limited <= 2,
2712 ("%s: tp->snd_limited too big",
2714 tp->snd_cwnd = tp->snd_ssthresh +
2716 (tp->t_dupacks - tp->snd_limited);
2717 if (SEQ_GT(onxt, tp->snd_nxt))
2720 } else if (V_tcp_do_rfc3042) {
2722 * Process first and second duplicate
2723 * ACKs. Each indicates a segment
2724 * leaving the network, creating room
2725 * for more. Make sure we can send a
2726 * packet on reception of each duplicate
2727 * ACK by increasing snd_cwnd by one
2728 * segment. Restore the original
2729 * snd_cwnd after packet transmission.
2731 cc_ack_received(tp, th, nsegs,
2733 uint32_t oldcwnd = tp->snd_cwnd;
2734 tcp_seq oldsndmax = tp->snd_max;
2738 KASSERT(tp->t_dupacks == 1 ||
2740 ("%s: dupacks not 1 or 2",
2742 if (tp->t_dupacks == 1)
2743 tp->snd_limited = 0;
2745 (tp->snd_nxt - tp->snd_una) +
2746 (tp->t_dupacks - tp->snd_limited) *
2749 * Only call tcp_output when there
2750 * is new data available to be sent.
2751 * Otherwise we would send pure ACKs.
2753 SOCKBUF_LOCK(&so->so_snd);
2754 avail = sbavail(&so->so_snd) -
2755 (tp->snd_nxt - tp->snd_una);
2756 SOCKBUF_UNLOCK(&so->so_snd);
2758 (void) tp->t_fb->tfb_tcp_output(tp);
2759 sent = tp->snd_max - oldsndmax;
2760 if (sent > maxseg) {
2761 KASSERT((tp->t_dupacks == 2 &&
2762 tp->snd_limited == 0) ||
2763 (sent == maxseg + 1 &&
2764 tp->t_flags & TF_SENTFIN),
2765 ("%s: sent too much",
2767 tp->snd_limited = 2;
2768 } else if (sent > 0)
2770 tp->snd_cwnd = oldcwnd;
2777 * This ack is advancing the left edge, reset the
2782 * If this ack also has new SACK info, increment the
2783 * counter as per rfc6675. The variable
2784 * sack_changed tracks all changes to the SACK
2785 * scoreboard, including when partial ACKs without
2786 * SACK options are received, and clear the scoreboard
2787 * from the left side. Such partial ACKs should not be
2788 * counted as dupacks here.
2790 if ((tp->t_flags & TF_SACK_PERMIT) &&
2791 (to.to_flags & TOF_SACK) &&
2794 /* limit overhead by setting maxseg last */
2795 if (!IN_FASTRECOVERY(tp->t_flags) &&
2796 (tp->sackhint.sacked_bytes >
2797 ((tcprexmtthresh - 1) *
2798 (maxseg = tcp_maxseg(tp))))) {
2799 goto enter_recovery;
2805 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2806 ("%s: th_ack <= snd_una", __func__));
2809 * If the congestion window was inflated to account
2810 * for the other side's cached packets, retract it.
2812 if (IN_FASTRECOVERY(tp->t_flags)) {
2813 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2814 if (tp->t_flags & TF_SACK_PERMIT)
2815 if (V_tcp_do_prr && to.to_flags & TOF_SACK) {
2816 tcp_timer_activate(tp, TT_REXMT, 0);
2818 tcp_do_prr_ack(tp, th, &to);
2819 tp->t_flags |= TF_ACKNOW;
2820 (void) tcp_output(tp);
2822 tcp_sack_partialack(tp, th);
2824 tcp_newreno_partial_ack(tp, th);
2826 cc_post_recovery(tp, th);
2827 } else if (IN_CONGRECOVERY(tp->t_flags)) {
2828 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2830 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th);
2831 tp->snd_fack = th->th_ack;
2832 tcp_do_prr_ack(tp, th, &to);
2833 (void) tcp_output(tp);
2836 cc_post_recovery(tp, th);
2839 * If we reach this point, ACK is not a duplicate,
2840 * i.e., it ACKs something we sent.
2842 if (tp->t_flags & TF_NEEDSYN) {
2844 * T/TCP: Connection was half-synchronized, and our
2845 * SYN has been ACK'd (so connection is now fully
2846 * synchronized). Go to non-starred state,
2847 * increment snd_una for ACK of SYN, and check if
2848 * we can do window scaling.
2850 tp->t_flags &= ~TF_NEEDSYN;
2852 /* Do window scaling? */
2853 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2854 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2855 tp->rcv_scale = tp->request_r_scale;
2856 /* Send window already scaled. */
2861 INP_WLOCK_ASSERT(tp->t_inpcb);
2864 * Adjust for the SYN bit in sequence space,
2865 * but don't account for it in cwnd calculations.
2866 * This is for the SYN_RECEIVED, non-simultaneous
2867 * SYN case. SYN_SENT and simultaneous SYN are
2868 * treated elsewhere.
2872 acked = BYTES_THIS_ACK(tp, th);
2873 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2874 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2875 tp->snd_una, th->th_ack, tp, m));
2876 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2877 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2880 * If we just performed our first retransmit, and the ACK
2881 * arrives within our recovery window, then it was a mistake
2882 * to do the retransmit in the first place. Recover our
2883 * original cwnd and ssthresh, and proceed to transmit where
2886 if (tp->t_rxtshift == 1 &&
2887 tp->t_flags & TF_PREVVALID &&
2889 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2890 cc_cong_signal(tp, th, CC_RTO_ERR);
2893 * If we have a timestamp reply, update smoothed
2894 * round trip time. If no timestamp is present but
2895 * transmit timer is running and timed sequence
2896 * number was acked, update smoothed round trip time.
2897 * Since we now have an rtt measurement, cancel the
2898 * timer backoff (cf., Phil Karn's retransmit alg.).
2899 * Recompute the initial retransmit timer.
2901 * Some boxes send broken timestamp replies
2902 * during the SYN+ACK phase, ignore
2903 * timestamps of 0 or we could calculate a
2904 * huge RTT and blow up the retransmit timer.
2906 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2909 t = tcp_ts_getticks() - to.to_tsecr;
2910 if (!tp->t_rttlow || tp->t_rttlow > t)
2912 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2913 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2914 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2915 tp->t_rttlow = ticks - tp->t_rtttime;
2916 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2920 * If all outstanding data is acked, stop retransmit
2921 * timer and remember to restart (more output or persist).
2922 * If there is more data to be acked, restart retransmit
2923 * timer, using current (possibly backed-off) value.
2925 if (th->th_ack == tp->snd_max) {
2926 tcp_timer_activate(tp, TT_REXMT, 0);
2928 } else if (!tcp_timer_active(tp, TT_PERSIST))
2929 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2932 * If no data (only SYN) was ACK'd,
2933 * skip rest of ACK processing.
2939 * Let the congestion control algorithm update congestion
2940 * control related information. This typically means increasing
2941 * the congestion window.
2943 cc_ack_received(tp, th, nsegs, CC_ACK);
2945 SOCKBUF_LOCK(&so->so_snd);
2946 if (acked > sbavail(&so->so_snd)) {
2947 if (tp->snd_wnd >= sbavail(&so->so_snd))
2948 tp->snd_wnd -= sbavail(&so->so_snd);
2951 mfree = sbcut_locked(&so->so_snd,
2952 (int)sbavail(&so->so_snd));
2955 mfree = sbcut_locked(&so->so_snd, acked);
2956 if (tp->snd_wnd >= (uint32_t) acked)
2957 tp->snd_wnd -= acked;
2962 /* NB: sowwakeup_locked() does an implicit unlock. */
2963 sowwakeup_locked(so);
2965 /* Detect una wraparound. */
2966 if (!IN_RECOVERY(tp->t_flags) &&
2967 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2968 SEQ_LEQ(th->th_ack, tp->snd_recover))
2969 tp->snd_recover = th->th_ack - 1;
2970 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2971 if (IN_RECOVERY(tp->t_flags) &&
2972 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2973 EXIT_RECOVERY(tp->t_flags);
2975 tp->snd_una = th->th_ack;
2976 if (tp->t_flags & TF_SACK_PERMIT) {
2977 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2978 tp->snd_recover = tp->snd_una;
2980 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2981 tp->snd_nxt = tp->snd_una;
2983 switch (tp->t_state) {
2985 * In FIN_WAIT_1 STATE in addition to the processing
2986 * for the ESTABLISHED state if our FIN is now acknowledged
2987 * then enter FIN_WAIT_2.
2989 case TCPS_FIN_WAIT_1:
2990 if (ourfinisacked) {
2992 * If we can't receive any more
2993 * data, then closing user can proceed.
2994 * Starting the timer is contrary to the
2995 * specification, but if we don't get a FIN
2996 * we'll hang forever.
2999 * we should release the tp also, and use a
3002 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3003 soisdisconnected(so);
3004 tcp_timer_activate(tp, TT_2MSL,
3005 (tcp_fast_finwait2_recycle ?
3006 tcp_finwait2_timeout :
3009 tcp_state_change(tp, TCPS_FIN_WAIT_2);
3014 * In CLOSING STATE in addition to the processing for
3015 * the ESTABLISHED state if the ACK acknowledges our FIN
3016 * then enter the TIME-WAIT state, otherwise ignore
3020 if (ourfinisacked) {
3028 * In LAST_ACK, we may still be waiting for data to drain
3029 * and/or to be acked, as well as for the ack of our FIN.
3030 * If our FIN is now acknowledged, delete the TCB,
3031 * enter the closed state and return.
3034 if (ourfinisacked) {
3043 INP_WLOCK_ASSERT(tp->t_inpcb);
3046 * Update window information.
3047 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3049 if ((thflags & TH_ACK) &&
3050 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3051 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3052 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3053 /* keep track of pure window updates */
3055 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3056 TCPSTAT_INC(tcps_rcvwinupd);
3057 tp->snd_wnd = tiwin;
3058 tp->snd_wl1 = th->th_seq;
3059 tp->snd_wl2 = th->th_ack;
3060 if (tp->snd_wnd > tp->max_sndwnd)
3061 tp->max_sndwnd = tp->snd_wnd;
3066 * Process segments with URG.
3068 if ((thflags & TH_URG) && th->th_urp &&
3069 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3071 * This is a kludge, but if we receive and accept
3072 * random urgent pointers, we'll crash in
3073 * soreceive. It's hard to imagine someone
3074 * actually wanting to send this much urgent data.
3076 SOCKBUF_LOCK(&so->so_rcv);
3077 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3078 th->th_urp = 0; /* XXX */
3079 thflags &= ~TH_URG; /* XXX */
3080 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3081 goto dodata; /* XXX */
3084 * If this segment advances the known urgent pointer,
3085 * then mark the data stream. This should not happen
3086 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3087 * a FIN has been received from the remote side.
3088 * In these states we ignore the URG.
3090 * According to RFC961 (Assigned Protocols),
3091 * the urgent pointer points to the last octet
3092 * of urgent data. We continue, however,
3093 * to consider it to indicate the first octet
3094 * of data past the urgent section as the original
3095 * spec states (in one of two places).
3097 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3098 tp->rcv_up = th->th_seq + th->th_urp;
3099 so->so_oobmark = sbavail(&so->so_rcv) +
3100 (tp->rcv_up - tp->rcv_nxt) - 1;
3101 if (so->so_oobmark == 0)
3102 so->so_rcv.sb_state |= SBS_RCVATMARK;
3104 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3106 SOCKBUF_UNLOCK(&so->so_rcv);
3108 * Remove out of band data so doesn't get presented to user.
3109 * This can happen independent of advancing the URG pointer,
3110 * but if two URG's are pending at once, some out-of-band
3111 * data may creep in... ick.
3113 if (th->th_urp <= (uint32_t)tlen &&
3114 !(so->so_options & SO_OOBINLINE)) {
3115 /* hdr drop is delayed */
3116 tcp_pulloutofband(so, th, m, drop_hdrlen);
3120 * If no out of band data is expected,
3121 * pull receive urgent pointer along
3122 * with the receive window.
3124 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3125 tp->rcv_up = tp->rcv_nxt;
3128 INP_WLOCK_ASSERT(tp->t_inpcb);
3131 * Process the segment text, merging it into the TCP sequencing queue,
3132 * and arranging for acknowledgment of receipt if necessary.
3133 * This process logically involves adjusting tp->rcv_wnd as data
3134 * is presented to the user (this happens in tcp_usrreq.c,
3135 * case PRU_RCVD). If a FIN has already been received on this
3136 * connection then we just ignore the text.
3138 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3139 IS_FASTOPEN(tp->t_flags));
3140 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3141 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3142 tcp_seq save_start = th->th_seq;
3143 tcp_seq save_rnxt = tp->rcv_nxt;
3144 int save_tlen = tlen;
3145 m_adj(m, drop_hdrlen); /* delayed header drop */
3147 * Insert segment which includes th into TCP reassembly queue
3148 * with control block tp. Set thflags to whether reassembly now
3149 * includes a segment with FIN. This handles the common case
3150 * inline (segment is the next to be received on an established
3151 * connection, and the queue is empty), avoiding linkage into
3152 * and removal from the queue and repetition of various
3154 * Set DELACK for segments received in order, but ack
3155 * immediately when segments are out of order (so
3156 * fast retransmit can work).
3158 if (th->th_seq == tp->rcv_nxt &&
3160 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3162 if (DELAY_ACK(tp, tlen) || tfo_syn)
3163 tp->t_flags |= TF_DELACK;
3165 tp->t_flags |= TF_ACKNOW;
3166 tp->rcv_nxt += tlen;
3168 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3169 (tp->t_fbyte_in == 0)) {
3170 tp->t_fbyte_in = ticks;
3171 if (tp->t_fbyte_in == 0)
3173 if (tp->t_fbyte_out && tp->t_fbyte_in)
3174 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3176 thflags = th->th_flags & TH_FIN;
3177 TCPSTAT_INC(tcps_rcvpack);
3178 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3179 SOCKBUF_LOCK(&so->so_rcv);
3180 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3183 sbappendstream_locked(&so->so_rcv, m, 0);
3184 tp->t_flags |= TF_WAKESOR;
3187 * XXX: Due to the header drop above "th" is
3188 * theoretically invalid by now. Fortunately
3189 * m_adj() doesn't actually frees any mbufs
3190 * when trimming from the head.
3192 tcp_seq temp = save_start;
3194 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3195 tp->t_flags |= TF_ACKNOW;
3197 if ((tp->t_flags & TF_SACK_PERMIT) &&
3199 TCPS_HAVEESTABLISHED(tp->t_state)) {
3200 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3202 * DSACK actually handled in the fastpath
3205 tcp_update_sack_list(tp, save_start,
3206 save_start + save_tlen);
3207 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3208 if ((tp->rcv_numsacks >= 1) &&
3209 (tp->sackblks[0].end == save_start)) {
3211 * Partial overlap, recorded at todrop
3214 tcp_update_sack_list(tp,
3215 tp->sackblks[0].start,
3216 tp->sackblks[0].end);
3218 tcp_update_dsack_list(tp, save_start,
3219 save_start + save_tlen);
3221 } else if (tlen >= save_tlen) {
3222 /* Update of sackblks. */
3223 tcp_update_dsack_list(tp, save_start,
3224 save_start + save_tlen);
3225 } else if (tlen > 0) {
3226 tcp_update_dsack_list(tp, save_start,
3230 tcp_handle_wakeup(tp, so);
3233 * Note the amount of data that peer has sent into
3234 * our window, in order to estimate the sender's
3238 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3239 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3241 len = so->so_rcv.sb_hiwat;
3249 * If FIN is received ACK the FIN and let the user know
3250 * that the connection is closing.
3252 if (thflags & TH_FIN) {
3253 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3254 /* The socket upcall is handled by socantrcvmore. */
3257 * If connection is half-synchronized
3258 * (ie NEEDSYN flag on) then delay ACK,
3259 * so it may be piggybacked when SYN is sent.
3260 * Otherwise, since we received a FIN then no
3261 * more input can be expected, send ACK now.
3263 if (tp->t_flags & TF_NEEDSYN)
3264 tp->t_flags |= TF_DELACK;
3266 tp->t_flags |= TF_ACKNOW;
3269 switch (tp->t_state) {
3271 * In SYN_RECEIVED and ESTABLISHED STATES
3272 * enter the CLOSE_WAIT state.
3274 case TCPS_SYN_RECEIVED:
3275 tp->t_starttime = ticks;
3277 case TCPS_ESTABLISHED:
3278 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3282 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3283 * enter the CLOSING state.
3285 case TCPS_FIN_WAIT_1:
3286 tcp_state_change(tp, TCPS_CLOSING);
3290 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3291 * starting the time-wait timer, turning off the other
3294 case TCPS_FIN_WAIT_2:
3300 if (so->so_options & SO_DEBUG)
3301 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3304 TCP_PROBE3(debug__input, tp, th, m);
3307 * Return any desired output.
3309 if (needoutput || (tp->t_flags & TF_ACKNOW))
3310 (void) tp->t_fb->tfb_tcp_output(tp);
3313 INP_WLOCK_ASSERT(tp->t_inpcb);
3315 if (tp->t_flags & TF_DELACK) {
3316 tp->t_flags &= ~TF_DELACK;
3317 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3319 INP_WUNLOCK(tp->t_inpcb);
3324 * Generate an ACK dropping incoming segment if it occupies
3325 * sequence space, where the ACK reflects our state.
3327 * We can now skip the test for the RST flag since all
3328 * paths to this code happen after packets containing
3329 * RST have been dropped.
3331 * In the SYN-RECEIVED state, don't send an ACK unless the
3332 * segment we received passes the SYN-RECEIVED ACK test.
3333 * If it fails send a RST. This breaks the loop in the
3334 * "LAND" DoS attack, and also prevents an ACK storm
3335 * between two listening ports that have been sent forged
3336 * SYN segments, each with the source address of the other.
3338 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3339 (SEQ_GT(tp->snd_una, th->th_ack) ||
3340 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3341 rstreason = BANDLIM_RST_OPENPORT;
3345 if (so->so_options & SO_DEBUG)
3346 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3349 TCP_PROBE3(debug__input, tp, th, m);
3350 tp->t_flags |= TF_ACKNOW;
3351 (void) tp->t_fb->tfb_tcp_output(tp);
3352 INP_WUNLOCK(tp->t_inpcb);
3358 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3359 INP_WUNLOCK(tp->t_inpcb);
3361 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3366 * Drop space held by incoming segment and return.
3369 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3370 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3373 TCP_PROBE3(debug__input, tp, th, m);
3375 INP_WUNLOCK(tp->t_inpcb);
3381 * Issue RST and make ACK acceptable to originator of segment.
3382 * The mbuf must still include the original packet header.
3386 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3387 int tlen, int rstreason)
3393 struct ip6_hdr *ip6;
3397 INP_WLOCK_ASSERT(tp->t_inpcb);
3400 /* Don't bother if destination was broadcast/multicast. */
3401 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3404 if (mtod(m, struct ip *)->ip_v == 6) {
3405 ip6 = mtod(m, struct ip6_hdr *);
3406 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3407 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3409 /* IPv6 anycast check is done at tcp6_input() */
3412 #if defined(INET) && defined(INET6)
3417 ip = mtod(m, struct ip *);
3418 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3419 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3420 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3421 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3426 /* Perform bandwidth limiting. */
3427 if (badport_bandlim(rstreason) < 0)
3430 /* tcp_respond consumes the mbuf chain. */
3431 if (th->th_flags & TH_ACK) {
3432 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3433 th->th_ack, TH_RST);
3435 if (th->th_flags & TH_SYN)
3437 if (th->th_flags & TH_FIN)
3439 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3440 (tcp_seq)0, TH_RST|TH_ACK);
3448 * Parse TCP options and place in tcpopt.
3451 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3456 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3458 if (opt == TCPOPT_EOL)
3460 if (opt == TCPOPT_NOP)
3466 if (optlen < 2 || optlen > cnt)
3471 if (optlen != TCPOLEN_MAXSEG)
3473 if (!(flags & TO_SYN))
3475 to->to_flags |= TOF_MSS;
3476 bcopy((char *)cp + 2,
3477 (char *)&to->to_mss, sizeof(to->to_mss));
3478 to->to_mss = ntohs(to->to_mss);
3481 if (optlen != TCPOLEN_WINDOW)
3483 if (!(flags & TO_SYN))
3485 to->to_flags |= TOF_SCALE;
3486 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3488 case TCPOPT_TIMESTAMP:
3489 if (optlen != TCPOLEN_TIMESTAMP)
3491 to->to_flags |= TOF_TS;
3492 bcopy((char *)cp + 2,
3493 (char *)&to->to_tsval, sizeof(to->to_tsval));
3494 to->to_tsval = ntohl(to->to_tsval);
3495 bcopy((char *)cp + 6,
3496 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3497 to->to_tsecr = ntohl(to->to_tsecr);
3499 case TCPOPT_SIGNATURE:
3501 * In order to reply to a host which has set the
3502 * TCP_SIGNATURE option in its initial SYN, we have
3503 * to record the fact that the option was observed
3504 * here for the syncache code to perform the correct
3507 if (optlen != TCPOLEN_SIGNATURE)
3509 to->to_flags |= TOF_SIGNATURE;
3510 to->to_signature = cp + 2;
3512 case TCPOPT_SACK_PERMITTED:
3513 if (optlen != TCPOLEN_SACK_PERMITTED)
3515 if (!(flags & TO_SYN))
3519 to->to_flags |= TOF_SACKPERM;
3522 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3526 to->to_flags |= TOF_SACK;
3527 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3528 to->to_sacks = cp + 2;
3529 TCPSTAT_INC(tcps_sack_rcv_blocks);
3531 case TCPOPT_FAST_OPEN:
3533 * Cookie length validation is performed by the
3534 * server side cookie checking code or the client
3535 * side cookie cache update code.
3537 if (!(flags & TO_SYN))
3539 if (!V_tcp_fastopen_client_enable &&
3540 !V_tcp_fastopen_server_enable)
3542 to->to_flags |= TOF_FASTOPEN;
3543 to->to_tfo_len = optlen - 2;
3544 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3553 * Pull out of band byte out of a segment so
3554 * it doesn't appear in the user's data queue.
3555 * It is still reflected in the segment length for
3556 * sequencing purposes.
3559 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3562 int cnt = off + th->th_urp - 1;
3565 if (m->m_len > cnt) {
3566 char *cp = mtod(m, caddr_t) + cnt;
3567 struct tcpcb *tp = sototcpcb(so);
3569 INP_WLOCK_ASSERT(tp->t_inpcb);
3572 tp->t_oobflags |= TCPOOB_HAVEDATA;
3573 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3575 if (m->m_flags & M_PKTHDR)
3584 panic("tcp_pulloutofband");
3588 * Collect new round-trip time estimate
3589 * and update averages and current timeout.
3592 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3596 INP_WLOCK_ASSERT(tp->t_inpcb);
3598 TCPSTAT_INC(tcps_rttupdated);
3601 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3602 imax(0, rtt * 1000 / hz));
3604 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3606 * srtt is stored as fixed point with 5 bits after the
3607 * binary point (i.e., scaled by 8). The following magic
3608 * is equivalent to the smoothing algorithm in rfc793 with
3609 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3610 * point). Adjust rtt to origin 0.
3612 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3613 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3615 if ((tp->t_srtt += delta) <= 0)
3619 * We accumulate a smoothed rtt variance (actually, a
3620 * smoothed mean difference), then set the retransmit
3621 * timer to smoothed rtt + 4 times the smoothed variance.
3622 * rttvar is stored as fixed point with 4 bits after the
3623 * binary point (scaled by 16). The following is
3624 * equivalent to rfc793 smoothing with an alpha of .75
3625 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3626 * rfc793's wired-in beta.
3630 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3631 if ((tp->t_rttvar += delta) <= 0)
3633 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3634 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3637 * No rtt measurement yet - use the unsmoothed rtt.
3638 * Set the variance to half the rtt (so our first
3639 * retransmit happens at 3*rtt).
3641 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3642 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3643 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3649 * the retransmit should happen at rtt + 4 * rttvar.
3650 * Because of the way we do the smoothing, srtt and rttvar
3651 * will each average +1/2 tick of bias. When we compute
3652 * the retransmit timer, we want 1/2 tick of rounding and
3653 * 1 extra tick because of +-1/2 tick uncertainty in the
3654 * firing of the timer. The bias will give us exactly the
3655 * 1.5 tick we need. But, because the bias is
3656 * statistical, we have to test that we don't drop below
3657 * the minimum feasible timer (which is 2 ticks).
3659 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3660 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3663 * We received an ack for a packet that wasn't retransmitted;
3664 * it is probably safe to discard any error indications we've
3665 * received recently. This isn't quite right, but close enough
3666 * for now (a route might have failed after we sent a segment,
3667 * and the return path might not be symmetrical).
3669 tp->t_softerror = 0;
3673 * Determine a reasonable value for maxseg size.
3674 * If the route is known, check route for mtu.
3675 * If none, use an mss that can be handled on the outgoing interface
3676 * without forcing IP to fragment. If no route is found, route has no mtu,
3677 * or the destination isn't local, use a default, hopefully conservative
3678 * size (usually 512 or the default IP max size, but no more than the mtu
3679 * of the interface), as we can't discover anything about intervening
3680 * gateways or networks. We also initialize the congestion/slow start
3681 * window to be a single segment if the destination isn't local.
3682 * While looking at the routing entry, we also initialize other path-dependent
3683 * parameters from pre-set or cached values in the routing entry.
3685 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3686 * IP options, e.g. IPSEC data, since length of this data may vary, and
3687 * thus it is calculated for every segment separately in tcp_output().
3689 * NOTE that this routine is only called when we process an incoming
3690 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3691 * settings are handled in tcp_mssopt().
3694 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3695 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3698 uint32_t maxmtu = 0;
3699 struct inpcb *inp = tp->t_inpcb;
3700 struct hc_metrics_lite metrics;
3702 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3703 size_t min_protoh = isipv6 ?
3704 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3705 sizeof (struct tcpiphdr);
3707 size_t min_protoh = sizeof(struct tcpiphdr);
3710 INP_WLOCK_ASSERT(tp->t_inpcb);
3713 min_protoh += V_tcp_udp_tunneling_overhead;
3714 if (mtuoffer != -1) {
3715 KASSERT(offer == -1, ("%s: conflict", __func__));
3716 offer = mtuoffer - min_protoh;
3722 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3723 tp->t_maxseg = V_tcp_v6mssdflt;
3726 #if defined(INET) && defined(INET6)
3731 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3732 tp->t_maxseg = V_tcp_mssdflt;
3737 * No route to sender, stay with default mss and return.
3741 * In case we return early we need to initialize metrics
3742 * to a defined state as tcp_hc_get() would do for us
3743 * if there was no cache hit.
3745 if (metricptr != NULL)
3746 bzero(metricptr, sizeof(struct hc_metrics_lite));
3750 /* What have we got? */
3754 * Offer == 0 means that there was no MSS on the SYN
3755 * segment, in this case we use tcp_mssdflt as
3756 * already assigned to t_maxseg above.
3758 offer = tp->t_maxseg;
3763 * Offer == -1 means that we didn't receive SYN yet.
3769 * Prevent DoS attack with too small MSS. Round up
3770 * to at least minmss.
3772 offer = max(offer, V_tcp_minmss);
3776 * rmx information is now retrieved from tcp_hostcache.
3778 tcp_hc_get(&inp->inp_inc, &metrics);
3779 if (metricptr != NULL)
3780 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3783 * If there's a discovered mtu in tcp hostcache, use it.
3784 * Else, use the link mtu.
3786 if (metrics.rmx_mtu)
3787 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3791 mss = maxmtu - min_protoh;
3792 if (!V_path_mtu_discovery &&
3793 !in6_localaddr(&inp->in6p_faddr))
3794 mss = min(mss, V_tcp_v6mssdflt);
3797 #if defined(INET) && defined(INET6)
3802 mss = maxmtu - min_protoh;
3803 if (!V_path_mtu_discovery &&
3804 !in_localaddr(inp->inp_faddr))
3805 mss = min(mss, V_tcp_mssdflt);
3809 * XXX - The above conditional (mss = maxmtu - min_protoh)
3810 * probably violates the TCP spec.
3811 * The problem is that, since we don't know the
3812 * other end's MSS, we are supposed to use a conservative
3813 * default. But, if we do that, then MTU discovery will
3814 * never actually take place, because the conservative
3815 * default is much less than the MTUs typically seen
3816 * on the Internet today. For the moment, we'll sweep
3817 * this under the carpet.
3819 * The conservative default might not actually be a problem
3820 * if the only case this occurs is when sending an initial
3821 * SYN with options and data to a host we've never talked
3822 * to before. Then, they will reply with an MSS value which
3823 * will get recorded and the new parameters should get
3824 * recomputed. For Further Study.
3827 mss = min(mss, offer);
3830 * Sanity check: make sure that maxseg will be large
3831 * enough to allow some data on segments even if the
3832 * all the option space is used (40bytes). Otherwise
3833 * funny things may happen in tcp_output.
3835 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3843 tcp_mss(struct tcpcb *tp, int offer)
3849 struct hc_metrics_lite metrics;
3850 struct tcp_ifcap cap;
3852 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3854 bzero(&cap, sizeof(cap));
3855 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3861 * If there's a pipesize, change the socket buffer to that size,
3862 * don't change if sb_hiwat is different than default (then it
3863 * has been changed on purpose with setsockopt).
3864 * Make the socket buffers an integral number of mss units;
3865 * if the mss is larger than the socket buffer, decrease the mss.
3867 so = inp->inp_socket;
3868 SOCKBUF_LOCK(&so->so_snd);
3869 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3870 bufsize = metrics.rmx_sendpipe;
3872 bufsize = so->so_snd.sb_hiwat;
3876 bufsize = roundup(bufsize, mss);
3877 if (bufsize > sb_max)
3879 if (bufsize > so->so_snd.sb_hiwat)
3880 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3882 SOCKBUF_UNLOCK(&so->so_snd);
3884 * Sanity check: make sure that maxseg will be large
3885 * enough to allow some data on segments even if the
3886 * all the option space is used (40bytes). Otherwise
3887 * funny things may happen in tcp_output.
3889 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3891 tp->t_maxseg = max(mss, 64);
3893 SOCKBUF_LOCK(&so->so_rcv);
3894 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3895 bufsize = metrics.rmx_recvpipe;
3897 bufsize = so->so_rcv.sb_hiwat;
3898 if (bufsize > mss) {
3899 bufsize = roundup(bufsize, mss);
3900 if (bufsize > sb_max)
3902 if (bufsize > so->so_rcv.sb_hiwat)
3903 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3905 SOCKBUF_UNLOCK(&so->so_rcv);
3907 /* Check the interface for TSO capabilities. */
3908 if (cap.ifcap & CSUM_TSO) {
3909 tp->t_flags |= TF_TSO;
3910 tp->t_tsomax = cap.tsomax;
3911 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3912 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3917 * Determine the MSS option to send on an outgoing SYN.
3920 tcp_mssopt(struct in_conninfo *inc)
3923 uint32_t thcmtu = 0;
3924 uint32_t maxmtu = 0;
3927 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3930 if (inc->inc_flags & INC_ISIPV6) {
3931 mss = V_tcp_v6mssdflt;
3932 maxmtu = tcp_maxmtu6(inc, NULL);
3933 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3936 #if defined(INET) && defined(INET6)
3941 mss = V_tcp_mssdflt;
3942 maxmtu = tcp_maxmtu(inc, NULL);
3943 min_protoh = sizeof(struct tcpiphdr);
3946 #if defined(INET6) || defined(INET)
3947 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3950 if (maxmtu && thcmtu)
3951 mss = min(maxmtu, thcmtu) - min_protoh;
3952 else if (maxmtu || thcmtu)
3953 mss = max(maxmtu, thcmtu) - min_protoh;
3959 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
3961 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3962 int maxseg = tcp_maxseg(tp);
3964 INP_WLOCK_ASSERT(tp->t_inpcb);
3967 * Compute the amount of data that this ACK is indicating
3968 * (del_data) and an estimate of how many bytes are in the
3971 if (((tp->t_flags & TF_SACK_PERMIT) &&
3972 (to->to_flags & TOF_SACK)) ||
3973 (IN_CONGRECOVERY(tp->t_flags) &&
3974 !IN_FASTRECOVERY(tp->t_flags))) {
3975 del_data = tp->sackhint.delivered_data;
3976 if (V_tcp_do_rfc6675_pipe)
3977 pipe = tcp_compute_pipe(tp);
3979 pipe = (tp->snd_nxt - tp->snd_fack) +
3980 tp->sackhint.sack_bytes_rexmit;
3982 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg +
3983 tp->snd_recover - tp->snd_una))
3985 pipe = imax(0, tp->snd_max - tp->snd_una -
3986 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg);
3988 tp->sackhint.prr_delivered += del_data;
3990 * Proportional Rate Reduction
3992 if (pipe >= tp->snd_ssthresh) {
3993 if (tp->sackhint.recover_fs == 0)
3994 tp->sackhint.recover_fs =
3995 imax(1, tp->snd_nxt - tp->snd_una);
3996 snd_cnt = howmany((long)tp->sackhint.prr_delivered *
3997 tp->snd_ssthresh, tp->sackhint.recover_fs) -
3998 tp->sackhint.prr_out;
4000 if (V_tcp_do_prr_conservative || (del_data == 0))
4001 limit = tp->sackhint.prr_delivered -
4002 tp->sackhint.prr_out;
4004 limit = imax(tp->sackhint.prr_delivered -
4005 tp->sackhint.prr_out, del_data) +
4007 snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
4009 snd_cnt = imax(snd_cnt, 0) / maxseg;
4011 * Send snd_cnt new data into the network in response to this ack.
4012 * If there is going to be a SACK retransmission, adjust snd_cwnd
4015 if (IN_FASTRECOVERY(tp->t_flags)) {
4016 if ((tp->t_flags & TF_SACK_PERMIT) &&
4017 (to->to_flags & TOF_SACK)) {
4018 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
4019 tp->sackhint.sack_bytes_rexmit +
4022 tp->snd_cwnd = (tp->snd_max - tp->snd_una) +
4025 } else if (IN_CONGRECOVERY(tp->t_flags))
4026 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg);
4027 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd);
4031 * On a partial ack arrives, force the retransmission of the
4032 * next unacknowledged segment. Do not clear tp->t_dupacks.
4033 * By setting snd_nxt to ti_ack, this forces retransmission timer to
4037 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4039 tcp_seq onxt = tp->snd_nxt;
4040 uint32_t ocwnd = tp->snd_cwnd;
4041 u_int maxseg = tcp_maxseg(tp);
4043 INP_WLOCK_ASSERT(tp->t_inpcb);
4045 tcp_timer_activate(tp, TT_REXMT, 0);
4047 tp->snd_nxt = th->th_ack;
4049 * Set snd_cwnd to one segment beyond acknowledged offset.
4050 * (tp->snd_una has not yet been updated when this function is called.)
4052 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4053 tp->t_flags |= TF_ACKNOW;
4054 (void) tp->t_fb->tfb_tcp_output(tp);
4055 tp->snd_cwnd = ocwnd;
4056 if (SEQ_GT(onxt, tp->snd_nxt))
4059 * Partial window deflation. Relies on fact that tp->snd_una
4062 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4063 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4066 tp->snd_cwnd += maxseg;
4070 tcp_compute_pipe(struct tcpcb *tp)
4072 return (tp->snd_max - tp->snd_una +
4073 tp->sackhint.sack_bytes_rexmit -
4074 tp->sackhint.sacked_bytes);
4078 tcp_compute_initwnd(uint32_t maxseg)
4081 * Calculate the Initial Window, also used as Restart Window
4083 * RFC5681 Section 3.1 specifies the default conservative values.
4084 * RFC3390 specifies slightly more aggressive values.
4085 * RFC6928 increases it to ten segments.
4086 * Support for user specified value for initial flight size.
4088 if (V_tcp_initcwnd_segments)
4089 return min(V_tcp_initcwnd_segments * maxseg,
4090 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4091 else if (V_tcp_do_rfc3390)
4092 return min(4 * maxseg, max(2 * maxseg, 4380));
4094 /* Per RFC5681 Section 3.1 */
4096 return (2 * maxseg);
4097 else if (maxseg > 1095)
4098 return (3 * maxseg);
4100 return (4 * maxseg);