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_rfc3042) = 1;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
174 &VNET_NAME(tcp_do_rfc3042), 0,
175 "Enable RFC 3042 (Limited Transmit)");
177 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
179 &VNET_NAME(tcp_do_rfc3390), 0,
180 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
182 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
184 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
185 "Slow-start flight size (initial congestion window) in number of segments");
187 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
189 &VNET_NAME(tcp_do_rfc3465), 0,
190 "Enable RFC 3465 (Appropriate Byte Counting)");
192 VNET_DEFINE(int, tcp_abc_l_var) = 2;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
194 &VNET_NAME(tcp_abc_l_var), 2,
195 "Cap the max cwnd increment during slow-start to this number of segments");
197 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn,
198 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
201 VNET_DEFINE(int, tcp_do_ecn) = 2;
202 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
203 &VNET_NAME(tcp_do_ecn), 0,
206 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
207 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
208 &VNET_NAME(tcp_ecn_maxretries), 0,
209 "Max retries before giving up on ECN");
211 VNET_DEFINE(int, tcp_insecure_syn) = 0;
212 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
213 &VNET_NAME(tcp_insecure_syn), 0,
214 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
216 VNET_DEFINE(int, tcp_insecure_rst) = 0;
217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
218 &VNET_NAME(tcp_insecure_rst), 0,
219 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
221 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
222 #define V_tcp_recvspace VNET(tcp_recvspace)
223 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
224 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
226 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
228 &VNET_NAME(tcp_do_autorcvbuf), 0,
229 "Enable automatic receive buffer sizing");
231 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
232 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
233 &VNET_NAME(tcp_autorcvbuf_max), 0,
234 "Max size of automatic receive buffer");
236 VNET_DEFINE(struct inpcbhead, tcb);
237 #define tcb6 tcb /* for KAME src sync over BSD*'s */
238 VNET_DEFINE(struct inpcbinfo, tcbinfo);
241 * TCP statistics are stored in an array of counter(9)s, which size matches
242 * size of struct tcpstat. TCP running connection count is a regular array.
244 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
245 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
246 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
247 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
248 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
249 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
250 "TCP connection counts by TCP state");
253 tcp_vnet_init(const void *unused)
256 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
257 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
259 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
260 tcp_vnet_init, NULL);
264 tcp_vnet_uninit(const void *unused)
267 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
268 VNET_PCPUSTAT_FREE(tcpstat);
270 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
271 tcp_vnet_uninit, NULL);
275 * Kernel module interface for updating tcpstat. The first argument is an index
276 * into tcpstat treated as an array.
279 kmod_tcpstat_add(int statnum, int val)
282 counter_u64_add(VNET(tcpstat)[statnum], val);
287 * Wrapper for the TCP established input helper hook.
290 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
292 struct tcp_hhook_data hhook_data;
294 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
299 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
306 * CC wrapper hook functions
309 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
316 INP_WLOCK_ASSERT(tp->t_inpcb);
318 tp->ccv->nsegs = nsegs;
319 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
320 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
321 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
322 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
323 tp->ccv->flags |= CCF_CWND_LIMITED;
325 tp->ccv->flags &= ~CCF_CWND_LIMITED;
327 if (type == CC_ACK) {
329 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
330 ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
331 if (!IN_RECOVERY(tp->t_flags))
332 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
333 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
334 if ((tp->t_flags & TF_GPUTINPROG) &&
335 SEQ_GEQ(th->th_ack, tp->gput_ack)) {
337 * Compute goodput in bits per millisecond.
339 gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) /
340 max(1, tcp_ts_getticks() - tp->gput_ts);
341 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
344 * XXXLAS: This is a temporary hack, and should be
345 * chained off VOI_TCP_GPUT when stats(9) grows an API
346 * to deal with chained VOIs.
348 if (tp->t_stats_gput_prev > 0)
349 stats_voi_update_abs_s32(tp->t_stats,
351 ((gput - tp->t_stats_gput_prev) * 100) /
352 tp->t_stats_gput_prev);
353 tp->t_flags &= ~TF_GPUTINPROG;
354 tp->t_stats_gput_prev = gput;
357 if (tp->snd_cwnd > tp->snd_ssthresh) {
358 tp->t_bytes_acked += tp->ccv->bytes_this_ack;
359 if (tp->t_bytes_acked >= tp->snd_cwnd) {
360 tp->t_bytes_acked -= tp->snd_cwnd;
361 tp->ccv->flags |= CCF_ABC_SENTAWND;
364 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
365 tp->t_bytes_acked = 0;
369 if (CC_ALGO(tp)->ack_received != NULL) {
370 /* XXXLAS: Find a way to live without this */
371 tp->ccv->curack = th->th_ack;
372 CC_ALGO(tp)->ack_received(tp->ccv, type);
375 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
380 cc_conn_init(struct tcpcb *tp)
382 struct hc_metrics_lite metrics;
383 struct inpcb *inp = tp->t_inpcb;
387 INP_WLOCK_ASSERT(tp->t_inpcb);
389 tcp_hc_get(&inp->inp_inc, &metrics);
390 maxseg = tcp_maxseg(tp);
392 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
394 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
395 TCPSTAT_INC(tcps_usedrtt);
396 if (metrics.rmx_rttvar) {
397 tp->t_rttvar = metrics.rmx_rttvar;
398 TCPSTAT_INC(tcps_usedrttvar);
400 /* default variation is +- 1 rtt */
402 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
404 TCPT_RANGESET(tp->t_rxtcur,
405 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
406 tp->t_rttmin, TCPTV_REXMTMAX);
408 if (metrics.rmx_ssthresh) {
410 * There's some sort of gateway or interface
411 * buffer limit on the path. Use this to set
412 * the slow start threshold, but set the
413 * threshold to no less than 2*mss.
415 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
416 TCPSTAT_INC(tcps_usedssthresh);
420 * Set the initial slow-start flight size.
422 * If a SYN or SYN/ACK was lost and retransmitted, we have to
423 * reduce the initial CWND to one segment as congestion is likely
424 * requiring us to be cautious.
426 if (tp->snd_cwnd == 1)
427 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
429 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
431 if (CC_ALGO(tp)->conn_init != NULL)
432 CC_ALGO(tp)->conn_init(tp->ccv);
436 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
438 INP_WLOCK_ASSERT(tp->t_inpcb);
441 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
446 if (!IN_FASTRECOVERY(tp->t_flags)) {
447 tp->snd_recover = tp->snd_max;
448 if (tp->t_flags2 & TF2_ECN_PERMIT)
449 tp->t_flags2 |= TF2_ECN_SND_CWR;
453 if (!IN_CONGRECOVERY(tp->t_flags) ||
455 * Allow ECN reaction on ACK to CWR, if
456 * that data segment was also CE marked.
458 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
459 EXIT_CONGRECOVERY(tp->t_flags);
460 TCPSTAT_INC(tcps_ecn_rcwnd);
461 tp->snd_recover = tp->snd_max + 1;
462 if (tp->t_flags2 & TF2_ECN_PERMIT)
463 tp->t_flags2 |= TF2_ECN_SND_CWR;
468 tp->t_bytes_acked = 0;
469 EXIT_RECOVERY(tp->t_flags);
470 if (tp->t_flags2 & TF2_ECN_PERMIT)
471 tp->t_flags2 |= TF2_ECN_SND_CWR;
474 TCPSTAT_INC(tcps_sndrexmitbad);
475 /* RTO was unnecessary, so reset everything. */
476 tp->snd_cwnd = tp->snd_cwnd_prev;
477 tp->snd_ssthresh = tp->snd_ssthresh_prev;
478 tp->snd_recover = tp->snd_recover_prev;
479 if (tp->t_flags & TF_WASFRECOVERY)
480 ENTER_FASTRECOVERY(tp->t_flags);
481 if (tp->t_flags & TF_WASCRECOVERY)
482 ENTER_CONGRECOVERY(tp->t_flags);
483 tp->snd_nxt = tp->snd_max;
484 tp->t_flags &= ~TF_PREVVALID;
489 if (CC_ALGO(tp)->cong_signal != NULL) {
491 tp->ccv->curack = th->th_ack;
492 CC_ALGO(tp)->cong_signal(tp->ccv, type);
497 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
499 INP_WLOCK_ASSERT(tp->t_inpcb);
501 /* XXXLAS: KASSERT that we're in recovery? */
503 if (CC_ALGO(tp)->post_recovery != NULL) {
504 tp->ccv->curack = th->th_ack;
505 CC_ALGO(tp)->post_recovery(tp->ccv);
507 /* XXXLAS: EXIT_RECOVERY ? */
508 tp->t_bytes_acked = 0;
509 tp->sackhint.prr_out = 0;
513 * Indicate whether this ack should be delayed. We can delay the ack if
514 * following conditions are met:
515 * - There is no delayed ack timer in progress.
516 * - Our last ack wasn't a 0-sized window. We never want to delay
517 * the ack that opens up a 0-sized window.
518 * - LRO wasn't used for this segment. We make sure by checking that the
519 * segment size is not larger than the MSS.
521 #define DELAY_ACK(tp, tlen) \
522 ((!tcp_timer_active(tp, TT_DELACK) && \
523 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
524 (tlen <= tp->t_maxseg) && \
525 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
528 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
530 INP_WLOCK_ASSERT(tp->t_inpcb);
532 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
533 switch (iptos & IPTOS_ECN_MASK) {
535 tp->ccv->flags |= CCF_IPHDR_CE;
541 case IPTOS_ECN_NOTECT:
542 tp->ccv->flags &= ~CCF_IPHDR_CE;
546 if (th->th_flags & TH_CWR)
547 tp->ccv->flags |= CCF_TCPHDR_CWR;
549 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
551 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
553 if (tp->ccv->flags & CCF_ACKNOW) {
554 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
555 tp->t_flags |= TF_ACKNOW;
561 * TCP input handling is split into multiple parts:
562 * tcp6_input is a thin wrapper around tcp_input for the extended
563 * ip6_protox[] call format in ip6_input
564 * tcp_input handles primary segment validation, inpcb lookup and
565 * SYN processing on listen sockets
566 * tcp_do_segment processes the ACK and text of the segment for
567 * establishing, established and closing connections
571 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
574 struct in6_ifaddr *ia6;
578 if (m->m_len < *offp + sizeof(struct tcphdr)) {
579 m = m_pullup(m, *offp + sizeof(struct tcphdr));
582 TCPSTAT_INC(tcps_rcvshort);
583 return (IPPROTO_DONE);
588 * draft-itojun-ipv6-tcp-to-anycast
589 * better place to put this in?
591 ip6 = mtod(m, struct ip6_hdr *);
592 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
593 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
594 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
595 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
597 return (IPPROTO_DONE);
601 return (tcp_input_with_port(mp, offp, proto, port));
605 tcp6_input(struct mbuf **mp, int *offp, int proto)
608 return(tcp6_input_with_port(mp, offp, proto, 0));
613 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
615 struct mbuf *m = *mp;
616 struct tcphdr *th = NULL;
617 struct ip *ip = NULL;
618 struct inpcb *inp = NULL;
619 struct tcpcb *tp = NULL;
620 struct socket *so = NULL;
631 int rstreason = 0; /* For badport_bandlim accounting purposes */
634 struct m_tag *fwd_tag = NULL;
636 struct ip6_hdr *ip6 = NULL;
639 const void *ip6 = NULL;
641 struct tcpopt to; /* options in this segment */
642 char *s = NULL; /* address and port logging */
645 * The size of tcp_saveipgen must be the size of the max ip header,
648 u_char tcp_saveipgen[IP6_HDR_LEN];
649 struct tcphdr tcp_savetcp;
656 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
663 TCPSTAT_INC(tcps_rcvtotal);
667 ip6 = mtod(m, struct ip6_hdr *);
668 th = (struct tcphdr *)((caddr_t)ip6 + off0);
669 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
672 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
673 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
674 th->th_sum = m->m_pkthdr.csum_data;
676 th->th_sum = in6_cksum_pseudo(ip6, tlen,
677 IPPROTO_TCP, m->m_pkthdr.csum_data);
678 th->th_sum ^= 0xffff;
680 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
682 TCPSTAT_INC(tcps_rcvbadsum);
687 * Be proactive about unspecified IPv6 address in source.
688 * As we use all-zero to indicate unbounded/unconnected pcb,
689 * unspecified IPv6 address can be used to confuse us.
691 * Note that packets with unspecified IPv6 destination is
692 * already dropped in ip6_input.
694 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
698 iptos = IPV6_TRAFFIC_CLASS(ip6);
701 #if defined(INET) && defined(INET6)
707 * Get IP and TCP header together in first mbuf.
708 * Note: IP leaves IP header in first mbuf.
710 if (off0 > sizeof (struct ip)) {
712 off0 = sizeof(struct ip);
714 if (m->m_len < sizeof (struct tcpiphdr)) {
715 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
717 TCPSTAT_INC(tcps_rcvshort);
718 return (IPPROTO_DONE);
721 ip = mtod(m, struct ip *);
722 th = (struct tcphdr *)((caddr_t)ip + off0);
723 tlen = ntohs(ip->ip_len) - off0;
728 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
729 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
730 th->th_sum = m->m_pkthdr.csum_data;
732 th->th_sum = in_pseudo(ip->ip_src.s_addr,
734 htonl(m->m_pkthdr.csum_data + tlen +
736 th->th_sum ^= 0xffff;
738 struct ipovly *ipov = (struct ipovly *)ip;
741 * Checksum extended TCP header and data.
745 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
746 ipov->ih_len = htons(tlen);
747 th->th_sum = in_cksum(m, len);
748 /* Reset length for SDT probes. */
749 ip->ip_len = htons(len);
752 /* Re-initialization for later version check */
754 ip->ip_v = IPVERSION;
755 ip->ip_hl = off0 >> 2;
758 if (th->th_sum && (port == 0)) {
759 TCPSTAT_INC(tcps_rcvbadsum);
766 * Check that TCP offset makes sense,
767 * pull out TCP options and adjust length. XXX
769 off = th->th_off << 2;
770 if (off < sizeof (struct tcphdr) || off > tlen) {
771 TCPSTAT_INC(tcps_rcvbadoff);
774 tlen -= off; /* tlen is used instead of ti->ti_len */
775 if (off > sizeof (struct tcphdr)) {
778 if (m->m_len < off0 + off) {
779 m = m_pullup(m, off0 + off);
781 TCPSTAT_INC(tcps_rcvshort);
782 return (IPPROTO_DONE);
785 ip6 = mtod(m, struct ip6_hdr *);
786 th = (struct tcphdr *)((caddr_t)ip6 + off0);
789 #if defined(INET) && defined(INET6)
794 if (m->m_len < sizeof(struct ip) + off) {
795 if ((m = m_pullup(m, sizeof (struct ip) + off))
797 TCPSTAT_INC(tcps_rcvshort);
798 return (IPPROTO_DONE);
800 ip = mtod(m, struct ip *);
801 th = (struct tcphdr *)((caddr_t)ip + off0);
805 optlen = off - sizeof (struct tcphdr);
806 optp = (u_char *)(th + 1);
808 thflags = th->th_flags;
811 * Convert TCP protocol specific fields to host format.
813 tcp_fields_to_host(th);
816 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
818 drop_hdrlen = off0 + off;
821 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
825 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
827 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
830 #if defined(INET) && !defined(INET6)
831 (m->m_flags & M_IP_NEXTHOP)
834 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
837 * For initial SYN packets arriving on listening socket,
838 * we don't need write lock.
840 lookupflag = (thflags & (TH_ACK|TH_SYN)) == TH_SYN ?
841 INPLOOKUP_RLOCKLISTEN : INPLOOKUP_WLOCKPCB;
844 if (isipv6 && fwd_tag != NULL) {
845 struct sockaddr_in6 *next_hop6;
847 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
849 * Transparently forwarded. Pretend to be the destination.
850 * Already got one like this?
852 inp = in6_pcblookup_mbuf(&V_tcbinfo,
853 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
854 lookupflag, m->m_pkthdr.rcvif, m);
857 * It's new. Try to find the ambushing socket.
858 * Because we've rewritten the destination address,
859 * any hardware-generated hash is ignored.
861 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
862 th->th_sport, &next_hop6->sin6_addr,
863 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
864 th->th_dport, INPLOOKUP_WILDCARD | lookupflag,
868 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
869 th->th_sport, &ip6->ip6_dst, th->th_dport,
870 INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m);
873 #if defined(INET6) && defined(INET)
877 if (fwd_tag != NULL) {
878 struct sockaddr_in *next_hop;
880 next_hop = (struct sockaddr_in *)(fwd_tag+1);
882 * Transparently forwarded. Pretend to be the destination.
883 * already got one like this?
885 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
886 ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m);
889 * It's new. Try to find the ambushing socket.
890 * Because we've rewritten the destination address,
891 * any hardware-generated hash is ignored.
893 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
894 th->th_sport, next_hop->sin_addr,
895 next_hop->sin_port ? ntohs(next_hop->sin_port) :
896 th->th_dport, INPLOOKUP_WILDCARD | lookupflag,
900 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
901 th->th_sport, ip->ip_dst, th->th_dport,
902 INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m);
906 * If the INPCB does not exist then all data in the incoming
907 * segment is discarded and an appropriate RST is sent back.
908 * XXX MRT Send RST using which routing table?
912 * Log communication attempts to ports that are not
915 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
916 V_tcp_log_in_vain == 2) {
917 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
918 log(LOG_INFO, "%s; %s: Connection attempt "
919 "to closed port\n", s, __func__);
922 * When blackholing do not respond with a RST but
923 * completely ignore the segment and drop it.
925 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
929 rstreason = BANDLIM_RST_CLOSEDPORT;
932 INP_LOCK_ASSERT(inp);
934 * While waiting for inp lock during the lookup, another thread
935 * can have dropped the inpcb, in which case we need to loop back
936 * and try to find a new inpcb to deliver to.
938 if (inp->inp_flags & INP_DROPPED) {
943 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
944 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
945 ((inp->inp_socket == NULL) ||
946 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
947 inp->inp_flowid = m->m_pkthdr.flowid;
948 inp->inp_flowtype = M_HASHTYPE_GET(m);
950 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
952 if (isipv6 && IPSEC_ENABLED(ipv6) &&
953 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
961 if (IPSEC_ENABLED(ipv4) &&
962 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
969 * Check the minimum TTL for socket.
971 if (inp->inp_ip_minttl != 0) {
974 if (inp->inp_ip_minttl > ip6->ip6_hlim)
978 if (inp->inp_ip_minttl > ip->ip_ttl)
983 * A previous connection in TIMEWAIT state is supposed to catch stray
984 * or duplicate segments arriving late. If this segment was a
985 * legitimate new connection attempt, the old INPCB gets removed and
986 * we can try again to find a listening socket.
988 * At this point, due to earlier optimism, we may hold only an inpcb
989 * lock, and not the inpcbinfo write lock. If so, we need to try to
990 * acquire it, or if that fails, acquire a reference on the inpcb,
991 * drop all locks, acquire a global write lock, and then re-acquire
992 * the inpcb lock. We may at that point discover that another thread
993 * has tried to free the inpcb, in which case we need to loop back
994 * and try to find a new inpcb to deliver to.
996 * XXXRW: It may be time to rethink timewait locking.
998 if (inp->inp_flags & INP_TIMEWAIT) {
999 tcp_dooptions(&to, optp, optlen,
1000 (thflags & TH_SYN) ? TO_SYN : 0);
1002 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
1004 if (tcp_twcheck(inp, &to, th, m, tlen))
1006 return (IPPROTO_DONE);
1009 * The TCPCB may no longer exist if the connection is winding
1010 * down or it is in the CLOSED state. Either way we drop the
1011 * segment and send an appropriate response.
1013 tp = intotcpcb(inp);
1014 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1015 rstreason = BANDLIM_RST_CLOSEDPORT;
1019 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) {
1020 rstreason = BANDLIM_RST_CLOSEDPORT;
1025 if (tp->t_flags & TF_TOE) {
1026 tcp_offload_input(tp, m);
1027 m = NULL; /* consumed by the TOE driver */
1033 if (mac_inpcb_check_deliver(inp, m))
1036 so = inp->inp_socket;
1037 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1039 if (so->so_options & SO_DEBUG) {
1040 ostate = tp->t_state;
1043 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1046 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1049 #endif /* TCPDEBUG */
1051 * When the socket is accepting connections (the INPCB is in LISTEN
1052 * state) we look into the SYN cache if this is a new connection
1053 * attempt or the completion of a previous one.
1055 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1056 ("%s: so accepting but tp %p not listening", __func__, tp));
1057 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1058 struct in_conninfo inc;
1060 bzero(&inc, sizeof(inc));
1063 inc.inc_flags |= INC_ISIPV6;
1064 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1065 inc.inc_flags |= INC_IPV6MINMTU;
1066 inc.inc6_faddr = ip6->ip6_src;
1067 inc.inc6_laddr = ip6->ip6_dst;
1071 inc.inc_faddr = ip->ip_src;
1072 inc.inc_laddr = ip->ip_dst;
1074 inc.inc_fport = th->th_sport;
1075 inc.inc_lport = th->th_dport;
1076 inc.inc_fibnum = so->so_fibnum;
1079 * Check for an existing connection attempt in syncache if
1080 * the flag is only ACK. A successful lookup creates a new
1081 * socket appended to the listen queue in SYN_RECEIVED state.
1083 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1085 * Parse the TCP options here because
1086 * syncookies need access to the reflected
1089 tcp_dooptions(&to, optp, optlen, 0);
1091 * NB: syncache_expand() doesn't unlock
1092 * inp and tcpinfo locks.
1094 rstreason = syncache_expand(&inc, &to, th, &so, m, port);
1095 if (rstreason < 0) {
1097 * A failing TCP MD5 signature comparison
1098 * must result in the segment being dropped
1099 * and must not produce any response back
1103 } else if (rstreason == 0) {
1105 * No syncache entry or ACK was not
1106 * for our SYN/ACK. Send a RST.
1107 * NB: syncache did its own logging
1108 * of the failure cause.
1110 rstreason = BANDLIM_RST_OPENPORT;
1116 * We completed the 3-way handshake
1117 * but could not allocate a socket
1118 * either due to memory shortage,
1119 * listen queue length limits or
1120 * global socket limits. Send RST
1121 * or wait and have the remote end
1122 * retransmit the ACK for another
1125 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1126 log(LOG_DEBUG, "%s; %s: Listen socket: "
1127 "Socket allocation failed due to "
1128 "limits or memory shortage, %s\n",
1130 V_tcp_sc_rst_sock_fail ?
1131 "sending RST" : "try again");
1132 if (V_tcp_sc_rst_sock_fail) {
1133 rstreason = BANDLIM_UNLIMITED;
1139 * Socket is created in state SYN_RECEIVED.
1140 * Unlock the listen socket, lock the newly
1141 * created socket and update the tp variable.
1142 * If we came here via jump to tfo_socket_result,
1143 * then listening socket is read-locked.
1145 INP_UNLOCK(inp); /* listen socket */
1146 inp = sotoinpcb(so);
1148 * New connection inpcb is already locked by
1149 * syncache_expand().
1151 INP_WLOCK_ASSERT(inp);
1152 tp = intotcpcb(inp);
1153 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1154 ("%s: ", __func__));
1156 * Process the segment and the data it
1157 * contains. tcp_do_segment() consumes
1158 * the mbuf chain and unlocks the inpcb.
1160 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1161 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1163 return (IPPROTO_DONE);
1166 * Segment flag validation for new connection attempts:
1168 * Our (SYN|ACK) response was rejected.
1169 * Check with syncache and remove entry to prevent
1172 * NB: syncache_chkrst does its own logging of failure
1175 if (thflags & TH_RST) {
1176 syncache_chkrst(&inc, th, m, port);
1180 * We can't do anything without SYN.
1182 if ((thflags & TH_SYN) == 0) {
1183 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1184 log(LOG_DEBUG, "%s; %s: Listen socket: "
1185 "SYN is missing, segment ignored\n",
1187 TCPSTAT_INC(tcps_badsyn);
1191 * (SYN|ACK) is bogus on a listen socket.
1193 if (thflags & TH_ACK) {
1194 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1195 log(LOG_DEBUG, "%s; %s: Listen socket: "
1196 "SYN|ACK invalid, segment rejected\n",
1198 syncache_badack(&inc, port); /* XXX: Not needed! */
1199 TCPSTAT_INC(tcps_badsyn);
1200 rstreason = BANDLIM_RST_OPENPORT;
1204 * If the drop_synfin option is enabled, drop all
1205 * segments with both the SYN and FIN bits set.
1206 * This prevents e.g. nmap from identifying the
1208 * XXX: Poor reasoning. nmap has other methods
1209 * and is constantly refining its stack detection
1211 * XXX: This is a violation of the TCP specification
1212 * and was used by RFC1644.
1214 if ((thflags & TH_FIN) && V_drop_synfin) {
1215 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1216 log(LOG_DEBUG, "%s; %s: Listen socket: "
1217 "SYN|FIN segment ignored (based on "
1218 "sysctl setting)\n", s, __func__);
1219 TCPSTAT_INC(tcps_badsyn);
1223 * Segment's flags are (SYN) or (SYN|FIN).
1225 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1226 * as they do not affect the state of the TCP FSM.
1227 * The data pointed to by TH_URG and th_urp is ignored.
1229 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1230 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1231 KASSERT(thflags & (TH_SYN),
1232 ("%s: Listen socket: TH_SYN not set", __func__));
1233 INP_RLOCK_ASSERT(inp);
1236 * If deprecated address is forbidden,
1237 * we do not accept SYN to deprecated interface
1238 * address to prevent any new inbound connection from
1239 * getting established.
1240 * When we do not accept SYN, we send a TCP RST,
1241 * with deprecated source address (instead of dropping
1242 * it). We compromise it as it is much better for peer
1243 * to send a RST, and RST will be the final packet
1246 * If we do not forbid deprecated addresses, we accept
1247 * the SYN packet. RFC2462 does not suggest dropping
1249 * If we decipher RFC2462 5.5.4, it says like this:
1250 * 1. use of deprecated addr with existing
1251 * communication is okay - "SHOULD continue to be
1253 * 2. use of it with new communication:
1254 * (2a) "SHOULD NOT be used if alternate address
1255 * with sufficient scope is available"
1256 * (2b) nothing mentioned otherwise.
1257 * Here we fall into (2b) case as we have no choice in
1258 * our source address selection - we must obey the peer.
1260 * The wording in RFC2462 is confusing, and there are
1261 * multiple description text for deprecated address
1262 * handling - worse, they are not exactly the same.
1263 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1265 if (isipv6 && !V_ip6_use_deprecated) {
1266 struct in6_ifaddr *ia6;
1268 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1270 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1271 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1272 log(LOG_DEBUG, "%s; %s: Listen socket: "
1273 "Connection attempt to deprecated "
1274 "IPv6 address rejected\n",
1276 rstreason = BANDLIM_RST_OPENPORT;
1282 * Basic sanity checks on incoming SYN requests:
1283 * Don't respond if the destination is a link layer
1284 * broadcast according to RFC1122 4.2.3.10, p. 104.
1285 * If it is from this socket it must be forged.
1286 * Don't respond if the source or destination is a
1287 * global or subnet broad- or multicast address.
1288 * Note that it is quite possible to receive unicast
1289 * link-layer packets with a broadcast IP address. Use
1290 * in_broadcast() to find them.
1292 if (m->m_flags & (M_BCAST|M_MCAST)) {
1293 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1294 log(LOG_DEBUG, "%s; %s: Listen socket: "
1295 "Connection attempt from broad- or multicast "
1296 "link layer address ignored\n", s, __func__);
1301 if (th->th_dport == th->th_sport &&
1302 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1303 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1304 log(LOG_DEBUG, "%s; %s: Listen socket: "
1305 "Connection attempt to/from self "
1306 "ignored\n", s, __func__);
1309 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1310 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1311 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1312 log(LOG_DEBUG, "%s; %s: Listen socket: "
1313 "Connection attempt from/to multicast "
1314 "address ignored\n", s, __func__);
1319 #if defined(INET) && defined(INET6)
1324 if (th->th_dport == th->th_sport &&
1325 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1326 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1327 log(LOG_DEBUG, "%s; %s: Listen socket: "
1328 "Connection attempt from/to self "
1329 "ignored\n", s, __func__);
1332 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1333 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1334 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1335 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1336 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1337 log(LOG_DEBUG, "%s; %s: Listen socket: "
1338 "Connection attempt from/to broad- "
1339 "or multicast address ignored\n",
1346 * SYN appears to be valid. Create compressed TCP state
1350 if (so->so_options & SO_DEBUG)
1351 tcp_trace(TA_INPUT, ostate, tp,
1352 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1354 TCP_PROBE3(debug__input, tp, th, m);
1355 tcp_dooptions(&to, optp, optlen, TO_SYN);
1356 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL,
1357 iptos, port)) != NULL)
1358 goto tfo_socket_result;
1361 * Entry added to syncache and mbuf consumed.
1362 * Only the listen socket is unlocked by syncache_add().
1364 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1365 return (IPPROTO_DONE);
1366 } else if (tp->t_state == TCPS_LISTEN) {
1368 * When a listen socket is torn down the SO_ACCEPTCONN
1369 * flag is removed first while connections are drained
1370 * from the accept queue in a unlock/lock cycle of the
1371 * ACCEPT_LOCK, opening a race condition allowing a SYN
1372 * attempt go through unhandled.
1376 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1377 if (tp->t_flags & TF_SIGNATURE) {
1378 tcp_dooptions(&to, optp, optlen, thflags);
1379 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1380 TCPSTAT_INC(tcps_sig_err_nosigopt);
1383 if (!TCPMD5_ENABLED() ||
1384 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1388 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1391 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1392 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1393 * the inpcb, and unlocks pcbinfo.
1395 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1396 return (IPPROTO_DONE);
1399 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1402 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1405 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1406 m = NULL; /* mbuf chain got consumed. */
1411 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1417 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1422 return (IPPROTO_DONE);
1426 * Automatic sizing of receive socket buffer. Often the send
1427 * buffer size is not optimally adjusted to the actual network
1428 * conditions at hand (delay bandwidth product). Setting the
1429 * buffer size too small limits throughput on links with high
1430 * bandwidth and high delay (eg. trans-continental/oceanic links).
1432 * On the receive side the socket buffer memory is only rarely
1433 * used to any significant extent. This allows us to be much
1434 * more aggressive in scaling the receive socket buffer. For
1435 * the case that the buffer space is actually used to a large
1436 * extent and we run out of kernel memory we can simply drop
1437 * the new segments; TCP on the sender will just retransmit it
1438 * later. Setting the buffer size too big may only consume too
1439 * much kernel memory if the application doesn't read() from
1440 * the socket or packet loss or reordering makes use of the
1443 * The criteria to step up the receive buffer one notch are:
1444 * 1. Application has not set receive buffer size with
1445 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1446 * 2. the number of bytes received during 1/2 of an sRTT
1447 * is at least 3/8 of the current socket buffer size.
1448 * 3. receive buffer size has not hit maximal automatic size;
1450 * If all of the criteria are met we increaset the socket buffer
1451 * by a 1/2 (bounded by the max). This allows us to keep ahead
1452 * of slow-start but also makes it so our peer never gets limited
1453 * by our rwnd which we then open up causing a burst.
1455 * This algorithm does two steps per RTT at most and only if
1456 * we receive a bulk stream w/o packet losses or reorderings.
1457 * Shrinking the buffer during idle times is not necessary as
1458 * it doesn't consume any memory when idle.
1460 * TODO: Only step up if the application is actually serving
1461 * the buffer to better manage the socket buffer resources.
1464 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1465 struct tcpcb *tp, int tlen)
1469 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1470 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1471 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1472 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1473 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1474 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1475 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1477 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1479 /* Start over with next RTT. */
1483 tp->rfbuf_cnt += tlen; /* add up */
1489 tcp_input(struct mbuf **mp, int *offp, int proto)
1491 return(tcp_input_with_port(mp, offp, proto, 0));
1495 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1498 * Since tp might be gone if the session entered
1499 * the TIME_WAIT state before coming here, we need
1500 * to check if the socket is still connected.
1502 if ((so->so_state & SS_ISCONNECTED) == 0)
1504 INP_LOCK_ASSERT(tp->t_inpcb);
1505 if (tp->t_flags & TF_WAKESOR) {
1506 tp->t_flags &= ~TF_WAKESOR;
1507 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
1510 if (tp->t_flags & TF_WAKESOW) {
1511 tp->t_flags &= ~TF_WAKESOW;
1512 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
1518 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1519 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1521 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1522 int rstreason, todrop, win, incforsyn = 0;
1526 struct in_conninfo *inc;
1534 * The size of tcp_saveipgen must be the size of the max ip header,
1537 u_char tcp_saveipgen[IP6_HDR_LEN];
1538 struct tcphdr tcp_savetcp;
1541 thflags = th->th_flags;
1542 inc = &tp->t_inpcb->inp_inc;
1543 tp->sackhint.last_sack_ack = 0;
1545 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1548 INP_WLOCK_ASSERT(tp->t_inpcb);
1549 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1551 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1555 /* Save segment, if requested. */
1556 tcp_pcap_add(th, m, &(tp->t_inpkts));
1558 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1561 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1562 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1563 log(LOG_DEBUG, "%s; %s: "
1564 "SYN|FIN segment ignored (based on "
1565 "sysctl setting)\n", s, __func__);
1572 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1573 * check SEQ.ACK first.
1575 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1576 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1577 rstreason = BANDLIM_UNLIMITED;
1582 * Segment received on connection.
1583 * Reset idle time and keep-alive timer.
1584 * XXX: This should be done after segment
1585 * validation to ignore broken/spoofed segs.
1587 tp->t_rcvtime = ticks;
1590 * Scale up the window into a 32-bit value.
1591 * For the SYN_SENT state the scale is zero.
1593 tiwin = th->th_win << tp->snd_scale;
1595 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1599 * TCP ECN processing.
1601 if (tp->t_flags2 & TF2_ECN_PERMIT) {
1602 if (thflags & TH_CWR) {
1603 tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1604 tp->t_flags |= TF_ACKNOW;
1606 switch (iptos & IPTOS_ECN_MASK) {
1608 tp->t_flags2 |= TF2_ECN_SND_ECE;
1609 TCPSTAT_INC(tcps_ecn_ce);
1611 case IPTOS_ECN_ECT0:
1612 TCPSTAT_INC(tcps_ecn_ect0);
1614 case IPTOS_ECN_ECT1:
1615 TCPSTAT_INC(tcps_ecn_ect1);
1619 /* Process a packet differently from RFC3168. */
1620 cc_ecnpkt_handler(tp, th, iptos);
1622 /* Congestion experienced. */
1623 if (thflags & TH_ECE) {
1624 cc_cong_signal(tp, th, CC_ECN);
1629 * Parse options on any incoming segment.
1631 tcp_dooptions(&to, (u_char *)(th + 1),
1632 (th->th_off << 2) - sizeof(struct tcphdr),
1633 (thflags & TH_SYN) ? TO_SYN : 0);
1635 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1636 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1637 (to.to_flags & TOF_SIGNATURE) == 0) {
1638 TCPSTAT_INC(tcps_sig_err_sigopt);
1639 /* XXX: should drop? */
1643 * If echoed timestamp is later than the current time,
1644 * fall back to non RFC1323 RTT calculation. Normalize
1645 * timestamp if syncookies were used when this connection
1648 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1649 to.to_tsecr -= tp->ts_offset;
1650 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1652 else if (tp->t_flags & TF_PREVVALID &&
1653 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1654 cc_cong_signal(tp, th, CC_RTO_ERR);
1657 * Process options only when we get SYN/ACK back. The SYN case
1658 * for incoming connections is handled in tcp_syncache.
1659 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1660 * or <SYN,ACK>) segment itself is never scaled.
1661 * XXX this is traditional behavior, may need to be cleaned up.
1663 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1664 /* Handle parallel SYN for ECN */
1665 if (!(thflags & TH_ACK) &&
1666 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1667 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1668 tp->t_flags2 |= TF2_ECN_PERMIT;
1669 tp->t_flags2 |= TF2_ECN_SND_ECE;
1670 TCPSTAT_INC(tcps_ecn_shs);
1672 if ((to.to_flags & TOF_SCALE) &&
1673 (tp->t_flags & TF_REQ_SCALE) &&
1674 !(tp->t_flags & TF_NOOPT)) {
1675 tp->t_flags |= TF_RCVD_SCALE;
1676 tp->snd_scale = to.to_wscale;
1678 tp->t_flags &= ~TF_REQ_SCALE;
1680 * Initial send window. It will be updated with
1681 * the next incoming segment to the scaled value.
1683 tp->snd_wnd = th->th_win;
1684 if ((to.to_flags & TOF_TS) &&
1685 (tp->t_flags & TF_REQ_TSTMP) &&
1686 !(tp->t_flags & TF_NOOPT)) {
1687 tp->t_flags |= TF_RCVD_TSTMP;
1688 tp->ts_recent = to.to_tsval;
1689 tp->ts_recent_age = tcp_ts_getticks();
1691 tp->t_flags &= ~TF_REQ_TSTMP;
1692 if (to.to_flags & TOF_MSS)
1693 tcp_mss(tp, to.to_mss);
1694 if ((tp->t_flags & TF_SACK_PERMIT) &&
1695 (!(to.to_flags & TOF_SACKPERM) ||
1696 (tp->t_flags & TF_NOOPT)))
1697 tp->t_flags &= ~TF_SACK_PERMIT;
1698 if (IS_FASTOPEN(tp->t_flags)) {
1699 if ((to.to_flags & TOF_FASTOPEN) &&
1700 !(tp->t_flags & TF_NOOPT)) {
1703 if (to.to_flags & TOF_MSS)
1706 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1710 tcp_fastopen_update_cache(tp, mss,
1711 to.to_tfo_len, to.to_tfo_cookie);
1713 tcp_fastopen_disable_path(tp);
1718 * If timestamps were negotiated during SYN/ACK and a
1719 * segment without a timestamp is received, silently drop
1720 * the segment, unless it is a RST segment or missing timestamps are
1722 * See section 3.2 of RFC 7323.
1724 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1725 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1726 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1727 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1728 "segment processed normally\n",
1733 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1734 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1735 "segment silently dropped\n", s, __func__);
1742 * If timestamps were not negotiated during SYN/ACK and a
1743 * segment with a timestamp is received, ignore the
1744 * timestamp and process the packet normally.
1745 * See section 3.2 of RFC 7323.
1747 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1748 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1749 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1750 "segment processed normally\n", s, __func__);
1756 * Header prediction: check for the two common cases
1757 * of a uni-directional data xfer. If the packet has
1758 * no control flags, is in-sequence, the window didn't
1759 * change and we're not retransmitting, it's a
1760 * candidate. If the length is zero and the ack moved
1761 * forward, we're the sender side of the xfer. Just
1762 * free the data acked & wake any higher level process
1763 * that was blocked waiting for space. If the length
1764 * is non-zero and the ack didn't move, we're the
1765 * receiver side. If we're getting packets in-order
1766 * (the reassembly queue is empty), add the data to
1767 * the socket buffer and note that we need a delayed ack.
1768 * Make sure that the hidden state-flags are also off.
1769 * Since we check for TCPS_ESTABLISHED first, it can only
1772 if (tp->t_state == TCPS_ESTABLISHED &&
1773 th->th_seq == tp->rcv_nxt &&
1774 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1775 tp->snd_nxt == tp->snd_max &&
1776 tiwin && tiwin == tp->snd_wnd &&
1777 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1779 ((to.to_flags & TOF_TS) == 0 ||
1780 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1782 * If last ACK falls within this segment's sequence numbers,
1783 * record the timestamp.
1784 * NOTE that the test is modified according to the latest
1785 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1787 if ((to.to_flags & TOF_TS) != 0 &&
1788 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1789 tp->ts_recent_age = tcp_ts_getticks();
1790 tp->ts_recent = to.to_tsval;
1794 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1795 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1796 !IN_RECOVERY(tp->t_flags) &&
1797 (to.to_flags & TOF_SACK) == 0 &&
1798 TAILQ_EMPTY(&tp->snd_holes)) {
1800 * This is a pure ack for outstanding data.
1802 TCPSTAT_INC(tcps_predack);
1805 * "bad retransmit" recovery without timestamps.
1807 if ((to.to_flags & TOF_TS) == 0 &&
1808 tp->t_rxtshift == 1 &&
1809 tp->t_flags & TF_PREVVALID &&
1810 (int)(ticks - tp->t_badrxtwin) < 0) {
1811 cc_cong_signal(tp, th, CC_RTO_ERR);
1815 * Recalculate the transmit timer / rtt.
1817 * Some boxes send broken timestamp replies
1818 * during the SYN+ACK phase, ignore
1819 * timestamps of 0 or we could calculate a
1820 * huge RTT and blow up the retransmit timer.
1822 if ((to.to_flags & TOF_TS) != 0 &&
1826 t = tcp_ts_getticks() - to.to_tsecr;
1827 if (!tp->t_rttlow || tp->t_rttlow > t)
1830 TCP_TS_TO_TICKS(t) + 1);
1831 } else if (tp->t_rtttime &&
1832 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1833 if (!tp->t_rttlow ||
1834 tp->t_rttlow > ticks - tp->t_rtttime)
1835 tp->t_rttlow = ticks - tp->t_rtttime;
1837 ticks - tp->t_rtttime);
1839 acked = BYTES_THIS_ACK(tp, th);
1842 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1843 hhook_run_tcp_est_in(tp, th, &to);
1846 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1847 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1848 sbdrop(&so->so_snd, acked);
1849 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1850 SEQ_LEQ(th->th_ack, tp->snd_recover))
1851 tp->snd_recover = th->th_ack - 1;
1854 * Let the congestion control algorithm update
1855 * congestion control related information. This
1856 * typically means increasing the congestion
1859 cc_ack_received(tp, th, nsegs, CC_ACK);
1861 tp->snd_una = th->th_ack;
1863 * Pull snd_wl2 up to prevent seq wrap relative
1866 tp->snd_wl2 = th->th_ack;
1871 * If all outstanding data are acked, stop
1872 * retransmit timer, otherwise restart timer
1873 * using current (possibly backed-off) value.
1874 * If process is waiting for space,
1875 * wakeup/selwakeup/signal. If data
1876 * are ready to send, let tcp_output
1877 * decide between more output or persist.
1880 if (so->so_options & SO_DEBUG)
1881 tcp_trace(TA_INPUT, ostate, tp,
1882 (void *)tcp_saveipgen,
1885 TCP_PROBE3(debug__input, tp, th, m);
1886 if (tp->snd_una == tp->snd_max)
1887 tcp_timer_activate(tp, TT_REXMT, 0);
1888 else if (!tcp_timer_active(tp, TT_PERSIST))
1889 tcp_timer_activate(tp, TT_REXMT,
1891 tp->t_flags |= TF_WAKESOW;
1892 if (sbavail(&so->so_snd))
1893 (void) tp->t_fb->tfb_tcp_output(tp);
1896 } else if (th->th_ack == tp->snd_una &&
1897 tlen <= sbspace(&so->so_rcv)) {
1898 int newsize = 0; /* automatic sockbuf scaling */
1901 * This is a pure, in-sequence data packet with
1902 * nothing on the reassembly queue and we have enough
1903 * buffer space to take it.
1905 /* Clean receiver SACK report if present */
1906 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1907 tcp_clean_sackreport(tp);
1908 TCPSTAT_INC(tcps_preddat);
1909 tp->rcv_nxt += tlen;
1911 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1912 (tp->t_fbyte_in == 0)) {
1913 tp->t_fbyte_in = ticks;
1914 if (tp->t_fbyte_in == 0)
1916 if (tp->t_fbyte_out && tp->t_fbyte_in)
1917 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1920 * Pull snd_wl1 up to prevent seq wrap relative to
1923 tp->snd_wl1 = th->th_seq;
1925 * Pull rcv_up up to prevent seq wrap relative to
1928 tp->rcv_up = tp->rcv_nxt;
1929 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1930 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1932 if (so->so_options & SO_DEBUG)
1933 tcp_trace(TA_INPUT, ostate, tp,
1934 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1936 TCP_PROBE3(debug__input, tp, th, m);
1938 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1940 /* Add data to socket buffer. */
1941 SOCKBUF_LOCK(&so->so_rcv);
1942 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1946 * Set new socket buffer size.
1947 * Give up when limit is reached.
1950 if (!sbreserve_locked(&so->so_rcv,
1952 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1953 m_adj(m, drop_hdrlen); /* delayed header drop */
1954 sbappendstream_locked(&so->so_rcv, m, 0);
1956 SOCKBUF_UNLOCK(&so->so_rcv);
1957 tp->t_flags |= TF_WAKESOR;
1958 if (DELAY_ACK(tp, tlen)) {
1959 tp->t_flags |= TF_DELACK;
1961 tp->t_flags |= TF_ACKNOW;
1962 tp->t_fb->tfb_tcp_output(tp);
1969 * Calculate amount of space in receive window,
1970 * and then do TCP input processing.
1971 * Receive window is amount of space in rcv queue,
1972 * but not less than advertised window.
1974 win = sbspace(&so->so_rcv);
1977 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1979 switch (tp->t_state) {
1981 * If the state is SYN_RECEIVED:
1982 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1984 case TCPS_SYN_RECEIVED:
1985 if ((thflags & TH_ACK) &&
1986 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1987 SEQ_GT(th->th_ack, tp->snd_max))) {
1988 rstreason = BANDLIM_RST_OPENPORT;
1991 if (IS_FASTOPEN(tp->t_flags)) {
1993 * When a TFO connection is in SYN_RECEIVED, the
1994 * only valid packets are the initial SYN, a
1995 * retransmit/copy of the initial SYN (possibly with
1996 * a subset of the original data), a valid ACK, a
1999 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
2000 rstreason = BANDLIM_RST_OPENPORT;
2002 } else if (thflags & TH_SYN) {
2003 /* non-initial SYN is ignored */
2004 if ((tcp_timer_active(tp, TT_DELACK) ||
2005 tcp_timer_active(tp, TT_REXMT)))
2007 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2014 * If the state is SYN_SENT:
2015 * if seg contains a RST with valid ACK (SEQ.ACK has already
2016 * been verified), then drop the connection.
2017 * if seg contains a RST without an ACK, drop the seg.
2018 * if seg does not contain SYN, then drop the seg.
2019 * Otherwise this is an acceptable SYN segment
2020 * initialize tp->rcv_nxt and tp->irs
2021 * if seg contains ack then advance tp->snd_una
2022 * if seg contains an ECE and ECN support is enabled, the stream
2024 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2025 * arrange for segment to be acked (eventually)
2026 * continue processing rest of data/controls, beginning with URG
2029 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2030 TCP_PROBE5(connect__refused, NULL, tp,
2032 tp = tcp_drop(tp, ECONNREFUSED);
2034 if (thflags & TH_RST)
2036 if (!(thflags & TH_SYN))
2039 tp->irs = th->th_seq;
2041 if (thflags & TH_ACK) {
2042 int tfo_partial_ack = 0;
2044 TCPSTAT_INC(tcps_connects);
2047 mac_socketpeer_set_from_mbuf(m, so);
2049 /* Do window scaling on this connection? */
2050 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2051 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2052 tp->rcv_scale = tp->request_r_scale;
2054 tp->rcv_adv += min(tp->rcv_wnd,
2055 TCP_MAXWIN << tp->rcv_scale);
2056 tp->snd_una++; /* SYN is acked */
2058 * If not all the data that was sent in the TFO SYN
2059 * has been acked, resend the remainder right away.
2061 if (IS_FASTOPEN(tp->t_flags) &&
2062 (tp->snd_una != tp->snd_max)) {
2063 tp->snd_nxt = th->th_ack;
2064 tfo_partial_ack = 1;
2067 * If there's data, delay ACK; if there's also a FIN
2068 * ACKNOW will be turned on later.
2070 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2071 tcp_timer_activate(tp, TT_DELACK,
2074 tp->t_flags |= TF_ACKNOW;
2076 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2077 (V_tcp_do_ecn == 1)) {
2078 tp->t_flags2 |= TF2_ECN_PERMIT;
2079 TCPSTAT_INC(tcps_ecn_shs);
2083 * Received <SYN,ACK> in SYN_SENT[*] state.
2085 * SYN_SENT --> ESTABLISHED
2086 * SYN_SENT* --> FIN_WAIT_1
2088 tp->t_starttime = ticks;
2089 if (tp->t_flags & TF_NEEDFIN) {
2090 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2091 tp->t_flags &= ~TF_NEEDFIN;
2094 tcp_state_change(tp, TCPS_ESTABLISHED);
2095 TCP_PROBE5(connect__established, NULL, tp,
2098 tcp_timer_activate(tp, TT_KEEP,
2103 * Received initial SYN in SYN-SENT[*] state =>
2104 * simultaneous open.
2105 * If it succeeds, connection is * half-synchronized.
2106 * Otherwise, do 3-way handshake:
2107 * SYN-SENT -> SYN-RECEIVED
2108 * SYN-SENT* -> SYN-RECEIVED*
2110 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2111 tcp_timer_activate(tp, TT_REXMT, 0);
2112 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2115 INP_WLOCK_ASSERT(tp->t_inpcb);
2118 * Advance th->th_seq to correspond to first data byte.
2119 * If data, trim to stay within window,
2120 * dropping FIN if necessary.
2123 if (tlen > tp->rcv_wnd) {
2124 todrop = tlen - tp->rcv_wnd;
2128 TCPSTAT_INC(tcps_rcvpackafterwin);
2129 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2131 tp->snd_wl1 = th->th_seq - 1;
2132 tp->rcv_up = th->th_seq;
2134 * Client side of transaction: already sent SYN and data.
2135 * If the remote host used T/TCP to validate the SYN,
2136 * our data will be ACK'd; if so, enter normal data segment
2137 * processing in the middle of step 5, ack processing.
2138 * Otherwise, goto step 6.
2140 if (thflags & TH_ACK)
2146 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2147 * do normal processing.
2149 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2153 break; /* continue normal processing */
2157 * States other than LISTEN or SYN_SENT.
2158 * First check the RST flag and sequence number since reset segments
2159 * are exempt from the timestamp and connection count tests. This
2160 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2161 * below which allowed reset segments in half the sequence space
2162 * to fall though and be processed (which gives forged reset
2163 * segments with a random sequence number a 50 percent chance of
2164 * killing a connection).
2165 * Then check timestamp, if present.
2166 * Then check the connection count, if present.
2167 * Then check that at least some bytes of segment are within
2168 * receive window. If segment begins before rcv_nxt,
2169 * drop leading data (and SYN); if nothing left, just ack.
2171 if (thflags & TH_RST) {
2173 * RFC5961 Section 3.2
2175 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2176 * - If RST is in window, we send challenge ACK.
2178 * Note: to take into account delayed ACKs, we should
2179 * test against last_ack_sent instead of rcv_nxt.
2180 * Note 2: we handle special case of closed window, not
2181 * covered by the RFC.
2183 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2184 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2185 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2186 KASSERT(tp->t_state != TCPS_SYN_SENT,
2187 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2190 if (V_tcp_insecure_rst ||
2191 tp->last_ack_sent == th->th_seq) {
2192 TCPSTAT_INC(tcps_drops);
2193 /* Drop the connection. */
2194 switch (tp->t_state) {
2195 case TCPS_SYN_RECEIVED:
2196 so->so_error = ECONNREFUSED;
2198 case TCPS_ESTABLISHED:
2199 case TCPS_FIN_WAIT_1:
2200 case TCPS_FIN_WAIT_2:
2201 case TCPS_CLOSE_WAIT:
2204 so->so_error = ECONNRESET;
2211 TCPSTAT_INC(tcps_badrst);
2212 /* Send challenge ACK. */
2213 tcp_respond(tp, mtod(m, void *), th, m,
2214 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2215 tp->last_ack_sent = tp->rcv_nxt;
2223 * RFC5961 Section 4.2
2224 * Send challenge ACK for any SYN in synchronized state.
2226 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2227 tp->t_state != TCPS_SYN_RECEIVED) {
2228 TCPSTAT_INC(tcps_badsyn);
2229 if (V_tcp_insecure_syn &&
2230 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2231 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2232 tp = tcp_drop(tp, ECONNRESET);
2233 rstreason = BANDLIM_UNLIMITED;
2235 /* Send challenge ACK. */
2236 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2237 tp->snd_nxt, TH_ACK);
2238 tp->last_ack_sent = tp->rcv_nxt;
2245 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2246 * and it's less than ts_recent, drop it.
2248 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2249 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2250 /* Check to see if ts_recent is over 24 days old. */
2251 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2253 * Invalidate ts_recent. If this segment updates
2254 * ts_recent, the age will be reset later and ts_recent
2255 * will get a valid value. If it does not, setting
2256 * ts_recent to zero will at least satisfy the
2257 * requirement that zero be placed in the timestamp
2258 * echo reply when ts_recent isn't valid. The
2259 * age isn't reset until we get a valid ts_recent
2260 * because we don't want out-of-order segments to be
2261 * dropped when ts_recent is old.
2265 TCPSTAT_INC(tcps_rcvduppack);
2266 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2267 TCPSTAT_INC(tcps_pawsdrop);
2275 * In the SYN-RECEIVED state, validate that the packet belongs to
2276 * this connection before trimming the data to fit the receive
2277 * window. Check the sequence number versus IRS since we know
2278 * the sequence numbers haven't wrapped. This is a partial fix
2279 * for the "LAND" DoS attack.
2281 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2282 rstreason = BANDLIM_RST_OPENPORT;
2286 todrop = tp->rcv_nxt - th->th_seq;
2288 if (thflags & TH_SYN) {
2298 * Following if statement from Stevens, vol. 2, p. 960.
2301 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2303 * Any valid FIN must be to the left of the window.
2304 * At this point the FIN must be a duplicate or out
2305 * of sequence; drop it.
2310 * Send an ACK to resynchronize and drop any data.
2311 * But keep on processing for RST or ACK.
2313 tp->t_flags |= TF_ACKNOW;
2315 TCPSTAT_INC(tcps_rcvduppack);
2316 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2318 TCPSTAT_INC(tcps_rcvpartduppack);
2319 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2322 * DSACK - add SACK block for dropped range
2324 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2325 tcp_update_sack_list(tp, th->th_seq,
2326 th->th_seq + todrop);
2328 * ACK now, as the next in-sequence segment
2329 * will clear the DSACK block again
2331 tp->t_flags |= TF_ACKNOW;
2333 drop_hdrlen += todrop; /* drop from the top afterwards */
2334 th->th_seq += todrop;
2336 if (th->th_urp > todrop)
2337 th->th_urp -= todrop;
2345 * If new data are received on a connection after the
2346 * user processes are gone, then RST the other end.
2348 if ((so->so_state & SS_NOFDREF) &&
2349 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2350 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2351 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2352 "after socket was closed, "
2353 "sending RST and removing tcpcb\n",
2354 s, __func__, tcpstates[tp->t_state], tlen);
2358 TCPSTAT_INC(tcps_rcvafterclose);
2359 rstreason = BANDLIM_UNLIMITED;
2364 * If segment ends after window, drop trailing data
2365 * (and PUSH and FIN); if nothing left, just ACK.
2367 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2369 TCPSTAT_INC(tcps_rcvpackafterwin);
2370 if (todrop >= tlen) {
2371 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2373 * If window is closed can only take segments at
2374 * window edge, and have to drop data and PUSH from
2375 * incoming segments. Continue processing, but
2376 * remember to ack. Otherwise, drop segment
2379 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2380 tp->t_flags |= TF_ACKNOW;
2381 TCPSTAT_INC(tcps_rcvwinprobe);
2385 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2388 thflags &= ~(TH_PUSH|TH_FIN);
2392 * If last ACK falls within this segment's sequence numbers,
2393 * record its timestamp.
2395 * 1) That the test incorporates suggestions from the latest
2396 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2397 * 2) That updating only on newer timestamps interferes with
2398 * our earlier PAWS tests, so this check should be solely
2399 * predicated on the sequence space of this segment.
2400 * 3) That we modify the segment boundary check to be
2401 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2402 * instead of RFC1323's
2403 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2404 * This modified check allows us to overcome RFC1323's
2405 * limitations as described in Stevens TCP/IP Illustrated
2406 * Vol. 2 p.869. In such cases, we can still calculate the
2407 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2409 if ((to.to_flags & TOF_TS) != 0 &&
2410 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2411 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2412 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2413 tp->ts_recent_age = tcp_ts_getticks();
2414 tp->ts_recent = to.to_tsval;
2418 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2419 * flag is on (half-synchronized state), then queue data for
2420 * later processing; else drop segment and return.
2422 if ((thflags & TH_ACK) == 0) {
2423 if (tp->t_state == TCPS_SYN_RECEIVED ||
2424 (tp->t_flags & TF_NEEDSYN)) {
2425 if (tp->t_state == TCPS_SYN_RECEIVED &&
2426 IS_FASTOPEN(tp->t_flags)) {
2427 tp->snd_wnd = tiwin;
2431 } else if (tp->t_flags & TF_ACKNOW)
2440 switch (tp->t_state) {
2442 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2443 * ESTABLISHED state and continue processing.
2444 * The ACK was checked above.
2446 case TCPS_SYN_RECEIVED:
2448 TCPSTAT_INC(tcps_connects);
2450 /* Do window scaling? */
2451 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2452 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2453 tp->rcv_scale = tp->request_r_scale;
2455 tp->snd_wnd = tiwin;
2458 * SYN-RECEIVED -> ESTABLISHED
2459 * SYN-RECEIVED* -> FIN-WAIT-1
2461 tp->t_starttime = ticks;
2462 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2463 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2464 tp->t_tfo_pending = NULL;
2466 if (tp->t_flags & TF_NEEDFIN) {
2467 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2468 tp->t_flags &= ~TF_NEEDFIN;
2470 tcp_state_change(tp, TCPS_ESTABLISHED);
2471 TCP_PROBE5(accept__established, NULL, tp,
2474 * TFO connections call cc_conn_init() during SYN
2475 * processing. Calling it again here for such
2476 * connections is not harmless as it would undo the
2477 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2480 if (!IS_FASTOPEN(tp->t_flags))
2482 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2485 * Account for the ACK of our SYN prior to
2486 * regular ACK processing below, except for
2487 * simultaneous SYN, which is handled later.
2489 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2492 * If segment contains data or ACK, will call tcp_reass()
2493 * later; if not, do so now to pass queued data to user.
2495 if (tlen == 0 && (thflags & TH_FIN) == 0)
2496 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2498 tp->snd_wl1 = th->th_seq - 1;
2502 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2503 * ACKs. If the ack is in the range
2504 * tp->snd_una < th->th_ack <= tp->snd_max
2505 * then advance tp->snd_una to th->th_ack and drop
2506 * data from the retransmission queue. If this ACK reflects
2507 * more up to date window information we update our window information.
2509 case TCPS_ESTABLISHED:
2510 case TCPS_FIN_WAIT_1:
2511 case TCPS_FIN_WAIT_2:
2512 case TCPS_CLOSE_WAIT:
2515 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2516 TCPSTAT_INC(tcps_rcvacktoomuch);
2519 if ((tp->t_flags & TF_SACK_PERMIT) &&
2520 ((to.to_flags & TOF_SACK) ||
2521 !TAILQ_EMPTY(&tp->snd_holes)))
2522 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2525 * Reset the value so that previous (valid) value
2526 * from the last ack with SACK doesn't get used.
2528 tp->sackhint.sacked_bytes = 0;
2531 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2532 hhook_run_tcp_est_in(tp, th, &to);
2535 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2536 maxseg = tcp_maxseg(tp);
2538 (tiwin == tp->snd_wnd ||
2539 (tp->t_flags & TF_SACK_PERMIT))) {
2541 * If this is the first time we've seen a
2542 * FIN from the remote, this is not a
2543 * duplicate and it needs to be processed
2544 * normally. This happens during a
2545 * simultaneous close.
2547 if ((thflags & TH_FIN) &&
2548 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2552 TCPSTAT_INC(tcps_rcvdupack);
2554 * If we have outstanding data (other than
2555 * a window probe), this is a completely
2556 * duplicate ack (ie, window info didn't
2557 * change and FIN isn't set),
2558 * the ack is the biggest we've
2559 * seen and we've seen exactly our rexmt
2560 * threshold of them, assume a packet
2561 * has been dropped and retransmit it.
2562 * Kludge snd_nxt & the congestion
2563 * window so we send only this one
2566 * We know we're losing at the current
2567 * window size so do congestion avoidance
2568 * (set ssthresh to half the current window
2569 * and pull our congestion window back to
2570 * the new ssthresh).
2572 * Dup acks mean that packets have left the
2573 * network (they're now cached at the receiver)
2574 * so bump cwnd by the amount in the receiver
2575 * to keep a constant cwnd packets in the
2578 * When using TCP ECN, notify the peer that
2579 * we reduced the cwnd.
2582 * Following 2 kinds of acks should not affect
2585 * 2) Acks with SACK but without any new SACK
2586 * information in them. These could result from
2587 * any anomaly in the network like a switch
2588 * duplicating packets or a possible DoS attack.
2590 if (th->th_ack != tp->snd_una ||
2591 ((tp->t_flags & TF_SACK_PERMIT) &&
2592 (to.to_flags & TOF_SACK) &&
2595 else if (!tcp_timer_active(tp, TT_REXMT))
2597 else if (++tp->t_dupacks > tcprexmtthresh ||
2598 IN_FASTRECOVERY(tp->t_flags)) {
2599 cc_ack_received(tp, th, nsegs,
2602 IN_FASTRECOVERY(tp->t_flags) &&
2603 (tp->t_flags & TF_SACK_PERMIT)) {
2604 tcp_do_prr_ack(tp, th, &to);
2605 } else if ((tp->t_flags & TF_SACK_PERMIT) &&
2606 (to.to_flags & TOF_SACK) &&
2607 IN_FASTRECOVERY(tp->t_flags)) {
2611 * Compute the amount of data in flight first.
2612 * We can inject new data into the pipe iff
2613 * we have less than 1/2 the original window's
2614 * worth of data in flight.
2616 if (V_tcp_do_newsack)
2617 awnd = tcp_compute_pipe(tp);
2619 awnd = (tp->snd_nxt - tp->snd_fack) +
2620 tp->sackhint.sack_bytes_rexmit;
2622 if (awnd < tp->snd_ssthresh) {
2623 tp->snd_cwnd += maxseg;
2624 if (tp->snd_cwnd > tp->snd_ssthresh)
2625 tp->snd_cwnd = tp->snd_ssthresh;
2628 tp->snd_cwnd += maxseg;
2629 (void) tp->t_fb->tfb_tcp_output(tp);
2631 } else if (tp->t_dupacks == tcprexmtthresh ||
2632 (tp->t_flags & TF_SACK_PERMIT &&
2634 tp->sackhint.sacked_bytes >
2635 (tcprexmtthresh - 1) * maxseg)) {
2638 * Above is the RFC6675 trigger condition of
2639 * more than (dupthresh-1)*maxseg sacked data.
2640 * If the count of holes in the
2641 * scoreboard is >= dupthresh, we could
2642 * also enter loss recovery, but don't
2643 * have that value readily available.
2645 tp->t_dupacks = tcprexmtthresh;
2646 tcp_seq onxt = tp->snd_nxt;
2649 * If we're doing sack, or prr, check
2650 * to see if we're already in sack
2651 * recovery. If we're not doing sack,
2652 * check to see if we're in newreno
2656 (tp->t_flags & TF_SACK_PERMIT)) {
2657 if (IN_FASTRECOVERY(tp->t_flags)) {
2662 if (SEQ_LEQ(th->th_ack,
2668 /* Congestion signal before ack. */
2669 cc_cong_signal(tp, th, CC_NDUPACK);
2670 cc_ack_received(tp, th, nsegs,
2672 tcp_timer_activate(tp, TT_REXMT, 0);
2676 * snd_ssthresh is already updated by
2679 tp->sackhint.prr_delivered =
2680 tp->sackhint.sacked_bytes;
2681 tp->sackhint.recover_fs = max(1,
2682 tp->snd_nxt - tp->snd_una);
2684 if ((tp->t_flags & TF_SACK_PERMIT) &&
2685 (to.to_flags & TOF_SACK)) {
2687 tcps_sack_recovery_episode);
2688 tp->snd_recover = tp->snd_nxt;
2689 tp->snd_cwnd = maxseg;
2690 (void) tp->t_fb->tfb_tcp_output(tp);
2691 if (SEQ_GT(th->th_ack, tp->snd_una))
2692 goto resume_partialack;
2695 tp->snd_nxt = th->th_ack;
2696 tp->snd_cwnd = maxseg;
2697 (void) tp->t_fb->tfb_tcp_output(tp);
2698 KASSERT(tp->snd_limited <= 2,
2699 ("%s: tp->snd_limited too big",
2701 tp->snd_cwnd = tp->snd_ssthresh +
2703 (tp->t_dupacks - tp->snd_limited);
2704 if (SEQ_GT(onxt, tp->snd_nxt))
2707 } else if (V_tcp_do_rfc3042) {
2709 * Process first and second duplicate
2710 * ACKs. Each indicates a segment
2711 * leaving the network, creating room
2712 * for more. Make sure we can send a
2713 * packet on reception of each duplicate
2714 * ACK by increasing snd_cwnd by one
2715 * segment. Restore the original
2716 * snd_cwnd after packet transmission.
2718 cc_ack_received(tp, th, nsegs,
2720 uint32_t oldcwnd = tp->snd_cwnd;
2721 tcp_seq oldsndmax = tp->snd_max;
2725 KASSERT(tp->t_dupacks == 1 ||
2727 ("%s: dupacks not 1 or 2",
2729 if (tp->t_dupacks == 1)
2730 tp->snd_limited = 0;
2732 (tp->snd_nxt - tp->snd_una) +
2733 (tp->t_dupacks - tp->snd_limited) *
2736 * Only call tcp_output when there
2737 * is new data available to be sent.
2738 * Otherwise we would send pure ACKs.
2740 SOCKBUF_LOCK(&so->so_snd);
2741 avail = sbavail(&so->so_snd) -
2742 (tp->snd_nxt - tp->snd_una);
2743 SOCKBUF_UNLOCK(&so->so_snd);
2745 (void) tp->t_fb->tfb_tcp_output(tp);
2746 sent = tp->snd_max - oldsndmax;
2747 if (sent > maxseg) {
2748 KASSERT((tp->t_dupacks == 2 &&
2749 tp->snd_limited == 0) ||
2750 (sent == maxseg + 1 &&
2751 tp->t_flags & TF_SENTFIN),
2752 ("%s: sent too much",
2754 tp->snd_limited = 2;
2755 } else if (sent > 0)
2757 tp->snd_cwnd = oldcwnd;
2764 * This ack is advancing the left edge, reset the
2769 * If this ack also has new SACK info, increment the
2770 * counter as per rfc6675. The variable
2771 * sack_changed tracks all changes to the SACK
2772 * scoreboard, including when partial ACKs without
2773 * SACK options are received, and clear the scoreboard
2774 * from the left side. Such partial ACKs should not be
2775 * counted as dupacks here.
2777 if ((tp->t_flags & TF_SACK_PERMIT) &&
2778 (to.to_flags & TOF_SACK) &&
2781 /* limit overhead by setting maxseg last */
2782 if (!IN_FASTRECOVERY(tp->t_flags) &&
2783 (tp->sackhint.sacked_bytes >
2784 ((tcprexmtthresh - 1) *
2785 (maxseg = tcp_maxseg(tp))))) {
2786 goto enter_recovery;
2792 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2793 ("%s: th_ack <= snd_una", __func__));
2796 * If the congestion window was inflated to account
2797 * for the other side's cached packets, retract it.
2799 if (IN_FASTRECOVERY(tp->t_flags)) {
2800 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2801 if (tp->t_flags & TF_SACK_PERMIT)
2802 if (V_tcp_do_prr && to.to_flags & TOF_SACK) {
2803 tcp_timer_activate(tp, TT_REXMT, 0);
2805 tcp_do_prr_ack(tp, th, &to);
2806 tp->t_flags |= TF_ACKNOW;
2807 (void) tcp_output(tp);
2809 tcp_sack_partialack(tp, th);
2811 tcp_newreno_partial_ack(tp, th);
2813 cc_post_recovery(tp, th);
2814 } else if (IN_CONGRECOVERY(tp->t_flags)) {
2815 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2817 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th);
2818 tp->snd_fack = th->th_ack;
2819 tcp_do_prr_ack(tp, th, &to);
2820 (void) tcp_output(tp);
2823 cc_post_recovery(tp, th);
2826 * If we reach this point, ACK is not a duplicate,
2827 * i.e., it ACKs something we sent.
2829 if (tp->t_flags & TF_NEEDSYN) {
2831 * T/TCP: Connection was half-synchronized, and our
2832 * SYN has been ACK'd (so connection is now fully
2833 * synchronized). Go to non-starred state,
2834 * increment snd_una for ACK of SYN, and check if
2835 * we can do window scaling.
2837 tp->t_flags &= ~TF_NEEDSYN;
2839 /* Do window scaling? */
2840 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2841 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2842 tp->rcv_scale = tp->request_r_scale;
2843 /* Send window already scaled. */
2848 INP_WLOCK_ASSERT(tp->t_inpcb);
2851 * Adjust for the SYN bit in sequence space,
2852 * but don't account for it in cwnd calculations.
2853 * This is for the SYN_RECEIVED, non-simultaneous
2854 * SYN case. SYN_SENT and simultaneous SYN are
2855 * treated elsewhere.
2859 acked = BYTES_THIS_ACK(tp, th);
2860 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2861 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2862 tp->snd_una, th->th_ack, tp, m));
2863 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2864 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2867 * If we just performed our first retransmit, and the ACK
2868 * arrives within our recovery window, then it was a mistake
2869 * to do the retransmit in the first place. Recover our
2870 * original cwnd and ssthresh, and proceed to transmit where
2873 if (tp->t_rxtshift == 1 &&
2874 tp->t_flags & TF_PREVVALID &&
2876 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2877 cc_cong_signal(tp, th, CC_RTO_ERR);
2880 * If we have a timestamp reply, update smoothed
2881 * round trip time. If no timestamp is present but
2882 * transmit timer is running and timed sequence
2883 * number was acked, update smoothed round trip time.
2884 * Since we now have an rtt measurement, cancel the
2885 * timer backoff (cf., Phil Karn's retransmit alg.).
2886 * Recompute the initial retransmit timer.
2888 * Some boxes send broken timestamp replies
2889 * during the SYN+ACK phase, ignore
2890 * timestamps of 0 or we could calculate a
2891 * huge RTT and blow up the retransmit timer.
2893 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2896 t = tcp_ts_getticks() - to.to_tsecr;
2897 if (!tp->t_rttlow || tp->t_rttlow > t)
2899 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2900 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2901 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2902 tp->t_rttlow = ticks - tp->t_rtttime;
2903 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2907 * If all outstanding data is acked, stop retransmit
2908 * timer and remember to restart (more output or persist).
2909 * If there is more data to be acked, restart retransmit
2910 * timer, using current (possibly backed-off) value.
2912 if (th->th_ack == tp->snd_max) {
2913 tcp_timer_activate(tp, TT_REXMT, 0);
2915 } else if (!tcp_timer_active(tp, TT_PERSIST))
2916 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2919 * If no data (only SYN) was ACK'd,
2920 * skip rest of ACK processing.
2926 * Let the congestion control algorithm update congestion
2927 * control related information. This typically means increasing
2928 * the congestion window.
2930 cc_ack_received(tp, th, nsegs, CC_ACK);
2932 SOCKBUF_LOCK(&so->so_snd);
2933 if (acked > sbavail(&so->so_snd)) {
2934 if (tp->snd_wnd >= sbavail(&so->so_snd))
2935 tp->snd_wnd -= sbavail(&so->so_snd);
2938 mfree = sbcut_locked(&so->so_snd,
2939 (int)sbavail(&so->so_snd));
2942 mfree = sbcut_locked(&so->so_snd, acked);
2943 if (tp->snd_wnd >= (uint32_t) acked)
2944 tp->snd_wnd -= acked;
2949 SOCKBUF_UNLOCK(&so->so_snd);
2950 tp->t_flags |= TF_WAKESOW;
2952 /* Detect una wraparound. */
2953 if (!IN_RECOVERY(tp->t_flags) &&
2954 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2955 SEQ_LEQ(th->th_ack, tp->snd_recover))
2956 tp->snd_recover = th->th_ack - 1;
2957 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2958 if (IN_RECOVERY(tp->t_flags) &&
2959 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2960 EXIT_RECOVERY(tp->t_flags);
2962 tp->snd_una = th->th_ack;
2963 if (tp->t_flags & TF_SACK_PERMIT) {
2964 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2965 tp->snd_recover = tp->snd_una;
2967 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2968 tp->snd_nxt = tp->snd_una;
2970 switch (tp->t_state) {
2972 * In FIN_WAIT_1 STATE in addition to the processing
2973 * for the ESTABLISHED state if our FIN is now acknowledged
2974 * then enter FIN_WAIT_2.
2976 case TCPS_FIN_WAIT_1:
2977 if (ourfinisacked) {
2979 * If we can't receive any more
2980 * data, then closing user can proceed.
2981 * Starting the timer is contrary to the
2982 * specification, but if we don't get a FIN
2983 * we'll hang forever.
2986 * we should release the tp also, and use a
2989 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2990 soisdisconnected(so);
2991 tcp_timer_activate(tp, TT_2MSL,
2992 (tcp_fast_finwait2_recycle ?
2993 tcp_finwait2_timeout :
2996 tcp_state_change(tp, TCPS_FIN_WAIT_2);
3001 * In CLOSING STATE in addition to the processing for
3002 * the ESTABLISHED state if the ACK acknowledges our FIN
3003 * then enter the TIME-WAIT state, otherwise ignore
3007 if (ourfinisacked) {
3015 * In LAST_ACK, we may still be waiting for data to drain
3016 * and/or to be acked, as well as for the ack of our FIN.
3017 * If our FIN is now acknowledged, delete the TCB,
3018 * enter the closed state and return.
3021 if (ourfinisacked) {
3030 INP_WLOCK_ASSERT(tp->t_inpcb);
3033 * Update window information.
3034 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3036 if ((thflags & TH_ACK) &&
3037 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3038 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3039 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3040 /* keep track of pure window updates */
3042 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3043 TCPSTAT_INC(tcps_rcvwinupd);
3044 tp->snd_wnd = tiwin;
3045 tp->snd_wl1 = th->th_seq;
3046 tp->snd_wl2 = th->th_ack;
3047 if (tp->snd_wnd > tp->max_sndwnd)
3048 tp->max_sndwnd = tp->snd_wnd;
3053 * Process segments with URG.
3055 if ((thflags & TH_URG) && th->th_urp &&
3056 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3058 * This is a kludge, but if we receive and accept
3059 * random urgent pointers, we'll crash in
3060 * soreceive. It's hard to imagine someone
3061 * actually wanting to send this much urgent data.
3063 SOCKBUF_LOCK(&so->so_rcv);
3064 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3065 th->th_urp = 0; /* XXX */
3066 thflags &= ~TH_URG; /* XXX */
3067 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3068 goto dodata; /* XXX */
3071 * If this segment advances the known urgent pointer,
3072 * then mark the data stream. This should not happen
3073 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3074 * a FIN has been received from the remote side.
3075 * In these states we ignore the URG.
3077 * According to RFC961 (Assigned Protocols),
3078 * the urgent pointer points to the last octet
3079 * of urgent data. We continue, however,
3080 * to consider it to indicate the first octet
3081 * of data past the urgent section as the original
3082 * spec states (in one of two places).
3084 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3085 tp->rcv_up = th->th_seq + th->th_urp;
3086 so->so_oobmark = sbavail(&so->so_rcv) +
3087 (tp->rcv_up - tp->rcv_nxt) - 1;
3088 if (so->so_oobmark == 0)
3089 so->so_rcv.sb_state |= SBS_RCVATMARK;
3091 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3093 SOCKBUF_UNLOCK(&so->so_rcv);
3095 * Remove out of band data so doesn't get presented to user.
3096 * This can happen independent of advancing the URG pointer,
3097 * but if two URG's are pending at once, some out-of-band
3098 * data may creep in... ick.
3100 if (th->th_urp <= (uint32_t)tlen &&
3101 !(so->so_options & SO_OOBINLINE)) {
3102 /* hdr drop is delayed */
3103 tcp_pulloutofband(so, th, m, drop_hdrlen);
3107 * If no out of band data is expected,
3108 * pull receive urgent pointer along
3109 * with the receive window.
3111 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3112 tp->rcv_up = tp->rcv_nxt;
3115 INP_WLOCK_ASSERT(tp->t_inpcb);
3118 * Process the segment text, merging it into the TCP sequencing queue,
3119 * and arranging for acknowledgment of receipt if necessary.
3120 * This process logically involves adjusting tp->rcv_wnd as data
3121 * is presented to the user (this happens in tcp_usrreq.c,
3122 * case PRU_RCVD). If a FIN has already been received on this
3123 * connection then we just ignore the text.
3125 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3126 IS_FASTOPEN(tp->t_flags));
3127 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3128 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3129 tcp_seq save_start = th->th_seq;
3130 tcp_seq save_rnxt = tp->rcv_nxt;
3131 int save_tlen = tlen;
3132 m_adj(m, drop_hdrlen); /* delayed header drop */
3134 * Insert segment which includes th into TCP reassembly queue
3135 * with control block tp. Set thflags to whether reassembly now
3136 * includes a segment with FIN. This handles the common case
3137 * inline (segment is the next to be received on an established
3138 * connection, and the queue is empty), avoiding linkage into
3139 * and removal from the queue and repetition of various
3141 * Set DELACK for segments received in order, but ack
3142 * immediately when segments are out of order (so
3143 * fast retransmit can work).
3145 if (th->th_seq == tp->rcv_nxt &&
3147 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3149 if (DELAY_ACK(tp, tlen) || tfo_syn)
3150 tp->t_flags |= TF_DELACK;
3152 tp->t_flags |= TF_ACKNOW;
3153 tp->rcv_nxt += tlen;
3155 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3156 (tp->t_fbyte_in == 0)) {
3157 tp->t_fbyte_in = ticks;
3158 if (tp->t_fbyte_in == 0)
3160 if (tp->t_fbyte_out && tp->t_fbyte_in)
3161 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3163 thflags = th->th_flags & TH_FIN;
3164 TCPSTAT_INC(tcps_rcvpack);
3165 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3166 SOCKBUF_LOCK(&so->so_rcv);
3167 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3170 sbappendstream_locked(&so->so_rcv, m, 0);
3171 SOCKBUF_UNLOCK(&so->so_rcv);
3172 tp->t_flags |= TF_WAKESOR;
3175 * XXX: Due to the header drop above "th" is
3176 * theoretically invalid by now. Fortunately
3177 * m_adj() doesn't actually frees any mbufs
3178 * when trimming from the head.
3180 tcp_seq temp = save_start;
3181 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3182 tp->t_flags |= TF_ACKNOW;
3184 if ((tp->t_flags & TF_SACK_PERMIT) &&
3186 TCPS_HAVEESTABLISHED(tp->t_state)) {
3187 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3189 * DSACK actually handled in the fastpath
3192 tcp_update_sack_list(tp, save_start,
3193 save_start + save_tlen);
3194 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3195 if ((tp->rcv_numsacks >= 1) &&
3196 (tp->sackblks[0].end == save_start)) {
3198 * Partial overlap, recorded at todrop
3201 tcp_update_sack_list(tp,
3202 tp->sackblks[0].start,
3203 tp->sackblks[0].end);
3205 tcp_update_dsack_list(tp, save_start,
3206 save_start + save_tlen);
3208 } else if (tlen >= save_tlen) {
3209 /* Update of sackblks. */
3210 tcp_update_dsack_list(tp, save_start,
3211 save_start + save_tlen);
3212 } else if (tlen > 0) {
3213 tcp_update_dsack_list(tp, save_start,
3219 * Note the amount of data that peer has sent into
3220 * our window, in order to estimate the sender's
3224 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3225 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3227 len = so->so_rcv.sb_hiwat;
3235 * If FIN is received ACK the FIN and let the user know
3236 * that the connection is closing.
3238 if (thflags & TH_FIN) {
3239 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3241 /* The socket upcall is handled by socantrcvmore. */
3242 tp->t_flags &= ~TF_WAKESOR;
3244 * If connection is half-synchronized
3245 * (ie NEEDSYN flag on) then delay ACK,
3246 * so it may be piggybacked when SYN is sent.
3247 * Otherwise, since we received a FIN then no
3248 * more input can be expected, send ACK now.
3250 if (tp->t_flags & TF_NEEDSYN)
3251 tp->t_flags |= TF_DELACK;
3253 tp->t_flags |= TF_ACKNOW;
3256 switch (tp->t_state) {
3258 * In SYN_RECEIVED and ESTABLISHED STATES
3259 * enter the CLOSE_WAIT state.
3261 case TCPS_SYN_RECEIVED:
3262 tp->t_starttime = ticks;
3264 case TCPS_ESTABLISHED:
3265 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3269 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3270 * enter the CLOSING state.
3272 case TCPS_FIN_WAIT_1:
3273 tcp_state_change(tp, TCPS_CLOSING);
3277 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3278 * starting the time-wait timer, turning off the other
3281 case TCPS_FIN_WAIT_2:
3287 if (so->so_options & SO_DEBUG)
3288 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3291 TCP_PROBE3(debug__input, tp, th, m);
3294 * Return any desired output.
3296 if (needoutput || (tp->t_flags & TF_ACKNOW))
3297 (void) tp->t_fb->tfb_tcp_output(tp);
3300 INP_WLOCK_ASSERT(tp->t_inpcb);
3302 if (tp->t_flags & TF_DELACK) {
3303 tp->t_flags &= ~TF_DELACK;
3304 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3306 tcp_handle_wakeup(tp, so);
3307 INP_WUNLOCK(tp->t_inpcb);
3312 * Generate an ACK dropping incoming segment if it occupies
3313 * sequence space, where the ACK reflects our state.
3315 * We can now skip the test for the RST flag since all
3316 * paths to this code happen after packets containing
3317 * RST have been dropped.
3319 * In the SYN-RECEIVED state, don't send an ACK unless the
3320 * segment we received passes the SYN-RECEIVED ACK test.
3321 * If it fails send a RST. This breaks the loop in the
3322 * "LAND" DoS attack, and also prevents an ACK storm
3323 * between two listening ports that have been sent forged
3324 * SYN segments, each with the source address of the other.
3326 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3327 (SEQ_GT(tp->snd_una, th->th_ack) ||
3328 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3329 rstreason = BANDLIM_RST_OPENPORT;
3333 if (so->so_options & SO_DEBUG)
3334 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3337 TCP_PROBE3(debug__input, tp, th, m);
3338 tp->t_flags |= TF_ACKNOW;
3339 (void) tp->t_fb->tfb_tcp_output(tp);
3340 tcp_handle_wakeup(tp, so);
3341 INP_WUNLOCK(tp->t_inpcb);
3347 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3348 tcp_handle_wakeup(tp, so);
3349 INP_WUNLOCK(tp->t_inpcb);
3351 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3356 * Drop space held by incoming segment and return.
3359 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3360 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3363 TCP_PROBE3(debug__input, tp, th, m);
3365 tcp_handle_wakeup(tp, so);
3366 INP_WUNLOCK(tp->t_inpcb);
3372 * Issue RST and make ACK acceptable to originator of segment.
3373 * The mbuf must still include the original packet header.
3377 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3378 int tlen, int rstreason)
3384 struct ip6_hdr *ip6;
3388 INP_LOCK_ASSERT(tp->t_inpcb);
3391 /* Don't bother if destination was broadcast/multicast. */
3392 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3395 if (mtod(m, struct ip *)->ip_v == 6) {
3396 ip6 = mtod(m, struct ip6_hdr *);
3397 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3398 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3400 /* IPv6 anycast check is done at tcp6_input() */
3403 #if defined(INET) && defined(INET6)
3408 ip = mtod(m, struct ip *);
3409 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3410 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3411 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3412 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3417 /* Perform bandwidth limiting. */
3418 if (badport_bandlim(rstreason) < 0)
3421 /* tcp_respond consumes the mbuf chain. */
3422 if (th->th_flags & TH_ACK) {
3423 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3424 th->th_ack, TH_RST);
3426 if (th->th_flags & TH_SYN)
3428 if (th->th_flags & TH_FIN)
3430 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3431 (tcp_seq)0, TH_RST|TH_ACK);
3439 * Parse TCP options and place in tcpopt.
3442 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3447 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3449 if (opt == TCPOPT_EOL)
3451 if (opt == TCPOPT_NOP)
3457 if (optlen < 2 || optlen > cnt)
3462 if (optlen != TCPOLEN_MAXSEG)
3464 if (!(flags & TO_SYN))
3466 to->to_flags |= TOF_MSS;
3467 bcopy((char *)cp + 2,
3468 (char *)&to->to_mss, sizeof(to->to_mss));
3469 to->to_mss = ntohs(to->to_mss);
3472 if (optlen != TCPOLEN_WINDOW)
3474 if (!(flags & TO_SYN))
3476 to->to_flags |= TOF_SCALE;
3477 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3479 case TCPOPT_TIMESTAMP:
3480 if (optlen != TCPOLEN_TIMESTAMP)
3482 to->to_flags |= TOF_TS;
3483 bcopy((char *)cp + 2,
3484 (char *)&to->to_tsval, sizeof(to->to_tsval));
3485 to->to_tsval = ntohl(to->to_tsval);
3486 bcopy((char *)cp + 6,
3487 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3488 to->to_tsecr = ntohl(to->to_tsecr);
3490 case TCPOPT_SIGNATURE:
3492 * In order to reply to a host which has set the
3493 * TCP_SIGNATURE option in its initial SYN, we have
3494 * to record the fact that the option was observed
3495 * here for the syncache code to perform the correct
3498 if (optlen != TCPOLEN_SIGNATURE)
3500 to->to_flags |= TOF_SIGNATURE;
3501 to->to_signature = cp + 2;
3503 case TCPOPT_SACK_PERMITTED:
3504 if (optlen != TCPOLEN_SACK_PERMITTED)
3506 if (!(flags & TO_SYN))
3510 to->to_flags |= TOF_SACKPERM;
3513 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3517 to->to_flags |= TOF_SACK;
3518 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3519 to->to_sacks = cp + 2;
3520 TCPSTAT_INC(tcps_sack_rcv_blocks);
3522 case TCPOPT_FAST_OPEN:
3524 * Cookie length validation is performed by the
3525 * server side cookie checking code or the client
3526 * side cookie cache update code.
3528 if (!(flags & TO_SYN))
3530 if (!V_tcp_fastopen_client_enable &&
3531 !V_tcp_fastopen_server_enable)
3533 to->to_flags |= TOF_FASTOPEN;
3534 to->to_tfo_len = optlen - 2;
3535 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3544 * Pull out of band byte out of a segment so
3545 * it doesn't appear in the user's data queue.
3546 * It is still reflected in the segment length for
3547 * sequencing purposes.
3550 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3553 int cnt = off + th->th_urp - 1;
3556 if (m->m_len > cnt) {
3557 char *cp = mtod(m, caddr_t) + cnt;
3558 struct tcpcb *tp = sototcpcb(so);
3560 INP_WLOCK_ASSERT(tp->t_inpcb);
3563 tp->t_oobflags |= TCPOOB_HAVEDATA;
3564 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3566 if (m->m_flags & M_PKTHDR)
3575 panic("tcp_pulloutofband");
3579 * Collect new round-trip time estimate
3580 * and update averages and current timeout.
3583 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3587 INP_WLOCK_ASSERT(tp->t_inpcb);
3589 TCPSTAT_INC(tcps_rttupdated);
3592 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3593 imax(0, rtt * 1000 / hz));
3595 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3597 * srtt is stored as fixed point with 5 bits after the
3598 * binary point (i.e., scaled by 8). The following magic
3599 * is equivalent to the smoothing algorithm in rfc793 with
3600 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3601 * point). Adjust rtt to origin 0.
3603 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3604 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3606 if ((tp->t_srtt += delta) <= 0)
3610 * We accumulate a smoothed rtt variance (actually, a
3611 * smoothed mean difference), then set the retransmit
3612 * timer to smoothed rtt + 4 times the smoothed variance.
3613 * rttvar is stored as fixed point with 4 bits after the
3614 * binary point (scaled by 16). The following is
3615 * equivalent to rfc793 smoothing with an alpha of .75
3616 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3617 * rfc793's wired-in beta.
3621 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3622 if ((tp->t_rttvar += delta) <= 0)
3624 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3625 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3628 * No rtt measurement yet - use the unsmoothed rtt.
3629 * Set the variance to half the rtt (so our first
3630 * retransmit happens at 3*rtt).
3632 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3633 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3634 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3640 * the retransmit should happen at rtt + 4 * rttvar.
3641 * Because of the way we do the smoothing, srtt and rttvar
3642 * will each average +1/2 tick of bias. When we compute
3643 * the retransmit timer, we want 1/2 tick of rounding and
3644 * 1 extra tick because of +-1/2 tick uncertainty in the
3645 * firing of the timer. The bias will give us exactly the
3646 * 1.5 tick we need. But, because the bias is
3647 * statistical, we have to test that we don't drop below
3648 * the minimum feasible timer (which is 2 ticks).
3650 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3651 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3654 * We received an ack for a packet that wasn't retransmitted;
3655 * it is probably safe to discard any error indications we've
3656 * received recently. This isn't quite right, but close enough
3657 * for now (a route might have failed after we sent a segment,
3658 * and the return path might not be symmetrical).
3660 tp->t_softerror = 0;
3664 * Determine a reasonable value for maxseg size.
3665 * If the route is known, check route for mtu.
3666 * If none, use an mss that can be handled on the outgoing interface
3667 * without forcing IP to fragment. If no route is found, route has no mtu,
3668 * or the destination isn't local, use a default, hopefully conservative
3669 * size (usually 512 or the default IP max size, but no more than the mtu
3670 * of the interface), as we can't discover anything about intervening
3671 * gateways or networks. We also initialize the congestion/slow start
3672 * window to be a single segment if the destination isn't local.
3673 * While looking at the routing entry, we also initialize other path-dependent
3674 * parameters from pre-set or cached values in the routing entry.
3676 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3677 * IP options, e.g. IPSEC data, since length of this data may vary, and
3678 * thus it is calculated for every segment separately in tcp_output().
3680 * NOTE that this routine is only called when we process an incoming
3681 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3682 * settings are handled in tcp_mssopt().
3685 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3686 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3689 uint32_t maxmtu = 0;
3690 struct inpcb *inp = tp->t_inpcb;
3691 struct hc_metrics_lite metrics;
3693 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3694 size_t min_protoh = isipv6 ?
3695 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3696 sizeof (struct tcpiphdr);
3698 size_t min_protoh = sizeof(struct tcpiphdr);
3701 INP_WLOCK_ASSERT(tp->t_inpcb);
3704 min_protoh += V_tcp_udp_tunneling_overhead;
3705 if (mtuoffer != -1) {
3706 KASSERT(offer == -1, ("%s: conflict", __func__));
3707 offer = mtuoffer - min_protoh;
3713 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3714 tp->t_maxseg = V_tcp_v6mssdflt;
3717 #if defined(INET) && defined(INET6)
3722 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3723 tp->t_maxseg = V_tcp_mssdflt;
3728 * No route to sender, stay with default mss and return.
3732 * In case we return early we need to initialize metrics
3733 * to a defined state as tcp_hc_get() would do for us
3734 * if there was no cache hit.
3736 if (metricptr != NULL)
3737 bzero(metricptr, sizeof(struct hc_metrics_lite));
3741 /* What have we got? */
3745 * Offer == 0 means that there was no MSS on the SYN
3746 * segment, in this case we use tcp_mssdflt as
3747 * already assigned to t_maxseg above.
3749 offer = tp->t_maxseg;
3754 * Offer == -1 means that we didn't receive SYN yet.
3760 * Prevent DoS attack with too small MSS. Round up
3761 * to at least minmss.
3763 offer = max(offer, V_tcp_minmss);
3767 * rmx information is now retrieved from tcp_hostcache.
3769 tcp_hc_get(&inp->inp_inc, &metrics);
3770 if (metricptr != NULL)
3771 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3774 * If there's a discovered mtu in tcp hostcache, use it.
3775 * Else, use the link mtu.
3777 if (metrics.rmx_mtu)
3778 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3782 mss = maxmtu - min_protoh;
3783 if (!V_path_mtu_discovery &&
3784 !in6_localaddr(&inp->in6p_faddr))
3785 mss = min(mss, V_tcp_v6mssdflt);
3788 #if defined(INET) && defined(INET6)
3793 mss = maxmtu - min_protoh;
3794 if (!V_path_mtu_discovery &&
3795 !in_localaddr(inp->inp_faddr))
3796 mss = min(mss, V_tcp_mssdflt);
3800 * XXX - The above conditional (mss = maxmtu - min_protoh)
3801 * probably violates the TCP spec.
3802 * The problem is that, since we don't know the
3803 * other end's MSS, we are supposed to use a conservative
3804 * default. But, if we do that, then MTU discovery will
3805 * never actually take place, because the conservative
3806 * default is much less than the MTUs typically seen
3807 * on the Internet today. For the moment, we'll sweep
3808 * this under the carpet.
3810 * The conservative default might not actually be a problem
3811 * if the only case this occurs is when sending an initial
3812 * SYN with options and data to a host we've never talked
3813 * to before. Then, they will reply with an MSS value which
3814 * will get recorded and the new parameters should get
3815 * recomputed. For Further Study.
3818 mss = min(mss, offer);
3821 * Sanity check: make sure that maxseg will be large
3822 * enough to allow some data on segments even if the
3823 * all the option space is used (40bytes). Otherwise
3824 * funny things may happen in tcp_output.
3826 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3834 tcp_mss(struct tcpcb *tp, int offer)
3840 struct hc_metrics_lite metrics;
3841 struct tcp_ifcap cap;
3843 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3845 bzero(&cap, sizeof(cap));
3846 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3852 * If there's a pipesize, change the socket buffer to that size,
3853 * don't change if sb_hiwat is different than default (then it
3854 * has been changed on purpose with setsockopt).
3855 * Make the socket buffers an integral number of mss units;
3856 * if the mss is larger than the socket buffer, decrease the mss.
3858 so = inp->inp_socket;
3859 SOCKBUF_LOCK(&so->so_snd);
3860 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3861 bufsize = metrics.rmx_sendpipe;
3863 bufsize = so->so_snd.sb_hiwat;
3867 bufsize = roundup(bufsize, mss);
3868 if (bufsize > sb_max)
3870 if (bufsize > so->so_snd.sb_hiwat)
3871 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3873 SOCKBUF_UNLOCK(&so->so_snd);
3875 * Sanity check: make sure that maxseg will be large
3876 * enough to allow some data on segments even if the
3877 * all the option space is used (40bytes). Otherwise
3878 * funny things may happen in tcp_output.
3880 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3882 tp->t_maxseg = max(mss, 64);
3884 SOCKBUF_LOCK(&so->so_rcv);
3885 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3886 bufsize = metrics.rmx_recvpipe;
3888 bufsize = so->so_rcv.sb_hiwat;
3889 if (bufsize > mss) {
3890 bufsize = roundup(bufsize, mss);
3891 if (bufsize > sb_max)
3893 if (bufsize > so->so_rcv.sb_hiwat)
3894 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3896 SOCKBUF_UNLOCK(&so->so_rcv);
3898 /* Check the interface for TSO capabilities. */
3899 if (cap.ifcap & CSUM_TSO) {
3900 tp->t_flags |= TF_TSO;
3901 tp->t_tsomax = cap.tsomax;
3902 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3903 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3908 * Determine the MSS option to send on an outgoing SYN.
3911 tcp_mssopt(struct in_conninfo *inc)
3914 uint32_t thcmtu = 0;
3915 uint32_t maxmtu = 0;
3918 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3921 if (inc->inc_flags & INC_ISIPV6) {
3922 mss = V_tcp_v6mssdflt;
3923 maxmtu = tcp_maxmtu6(inc, NULL);
3924 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3927 #if defined(INET) && defined(INET6)
3932 mss = V_tcp_mssdflt;
3933 maxmtu = tcp_maxmtu(inc, NULL);
3934 min_protoh = sizeof(struct tcpiphdr);
3937 #if defined(INET6) || defined(INET)
3938 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3941 if (maxmtu && thcmtu)
3942 mss = min(maxmtu, thcmtu) - min_protoh;
3943 else if (maxmtu || thcmtu)
3944 mss = max(maxmtu, thcmtu) - min_protoh;
3950 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
3952 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3953 int maxseg = tcp_maxseg(tp);
3955 INP_WLOCK_ASSERT(tp->t_inpcb);
3958 * Compute the amount of data that this ACK is indicating
3959 * (del_data) and an estimate of how many bytes are in the
3962 del_data = tp->sackhint.delivered_data;
3963 if (V_tcp_do_newsack)
3964 pipe = tcp_compute_pipe(tp);
3966 pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit;
3967 tp->sackhint.prr_delivered += del_data;
3969 * Proportional Rate Reduction
3971 if (pipe >= tp->snd_ssthresh) {
3972 if (tp->sackhint.recover_fs == 0)
3973 tp->sackhint.recover_fs =
3974 imax(1, tp->snd_nxt - tp->snd_una);
3975 snd_cnt = howmany((long)tp->sackhint.prr_delivered *
3976 tp->snd_ssthresh, tp->sackhint.recover_fs) -
3977 tp->sackhint.prr_out;
3979 if (V_tcp_do_prr_conservative)
3980 limit = tp->sackhint.prr_delivered -
3981 tp->sackhint.prr_out;
3983 limit = imax(tp->sackhint.prr_delivered -
3984 tp->sackhint.prr_out, del_data) +
3986 snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
3988 snd_cnt = imax(snd_cnt, 0) / maxseg;
3990 * Send snd_cnt new data into the network in response to this ack.
3991 * If there is going to be a SACK retransmission, adjust snd_cwnd
3994 if (IN_FASTRECOVERY(tp->t_flags)) {
3995 tp->snd_cwnd = imax(maxseg, tp->snd_nxt - tp->snd_recover +
3996 tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg));
3997 } else if (IN_CONGRECOVERY(tp->t_flags))
3998 tp->snd_cwnd = imax(maxseg, pipe - del_data +
3999 (snd_cnt * maxseg));
4003 * On a partial ack arrives, force the retransmission of the
4004 * next unacknowledged segment. Do not clear tp->t_dupacks.
4005 * By setting snd_nxt to ti_ack, this forces retransmission timer to
4009 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4011 tcp_seq onxt = tp->snd_nxt;
4012 uint32_t ocwnd = tp->snd_cwnd;
4013 u_int maxseg = tcp_maxseg(tp);
4015 INP_WLOCK_ASSERT(tp->t_inpcb);
4017 tcp_timer_activate(tp, TT_REXMT, 0);
4019 tp->snd_nxt = th->th_ack;
4021 * Set snd_cwnd to one segment beyond acknowledged offset.
4022 * (tp->snd_una has not yet been updated when this function is called.)
4024 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4025 tp->t_flags |= TF_ACKNOW;
4026 (void) tp->t_fb->tfb_tcp_output(tp);
4027 tp->snd_cwnd = ocwnd;
4028 if (SEQ_GT(onxt, tp->snd_nxt))
4031 * Partial window deflation. Relies on fact that tp->snd_una
4034 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4035 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4038 tp->snd_cwnd += maxseg;
4042 tcp_compute_pipe(struct tcpcb *tp)
4044 return (tp->snd_max - tp->snd_una +
4045 tp->sackhint.sack_bytes_rexmit -
4046 tp->sackhint.sacked_bytes);
4050 tcp_compute_initwnd(uint32_t maxseg)
4053 * Calculate the Initial Window, also used as Restart Window
4055 * RFC5681 Section 3.1 specifies the default conservative values.
4056 * RFC3390 specifies slightly more aggressive values.
4057 * RFC6928 increases it to ten segments.
4058 * Support for user specified value for initial flight size.
4060 if (V_tcp_initcwnd_segments)
4061 return min(V_tcp_initcwnd_segments * maxseg,
4062 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4063 else if (V_tcp_do_rfc3390)
4064 return min(4 * maxseg, max(2 * maxseg, 4380));
4066 /* Per RFC5681 Section 3.1 */
4068 return (2 * maxseg);
4069 else if (maxseg > 1095)
4070 return (3 * maxseg);
4072 return (4 * maxseg);