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>
61 #include <sys/kernel.h>
63 #include <sys/hhook.h>
65 #include <sys/malloc.h>
67 #include <sys/proc.h> /* for proc0 declaration */
68 #include <sys/protosw.h>
70 #include <sys/signalvar.h>
71 #include <sys/socket.h>
72 #include <sys/socketvar.h>
73 #include <sys/sysctl.h>
74 #include <sys/syslog.h>
75 #include <sys/systm.h>
77 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
82 #include <net/if_var.h>
83 #include <net/route.h>
86 #define TCPSTATES /* for logging */
88 #include <netinet/in.h>
89 #include <netinet/in_kdtrace.h>
90 #include <netinet/in_pcb.h>
91 #include <netinet/in_systm.h>
92 #include <netinet/ip.h>
93 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
94 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
95 #include <netinet/ip_var.h>
96 #include <netinet/ip_options.h>
97 #include <netinet/ip6.h>
98 #include <netinet/icmp6.h>
99 #include <netinet6/in6_pcb.h>
100 #include <netinet6/in6_var.h>
101 #include <netinet6/ip6_var.h>
102 #include <netinet6/nd6.h>
103 #include <netinet/tcp.h>
104 #include <netinet/tcp_fsm.h>
105 #include <netinet/tcp_log_buf.h>
106 #include <netinet/tcp_seq.h>
107 #include <netinet/tcp_timer.h>
108 #include <netinet/tcp_var.h>
109 #include <netinet6/tcp6_var.h>
110 #include <netinet/tcpip.h>
111 #include <netinet/cc/cc.h>
112 #include <netinet/tcp_fastopen.h>
114 #include <netinet/tcp_pcap.h>
116 #include <netinet/tcp_syncache.h>
118 #include <netinet/tcp_debug.h>
119 #endif /* TCPDEBUG */
121 #include <netinet/tcp_offload.h>
124 #include <netipsec/ipsec_support.h>
126 #include <machine/in_cksum.h>
128 #include <security/mac/mac_framework.h>
130 const int tcprexmtthresh = 3;
132 int tcp_log_in_vain = 0;
133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
135 "Log all incoming TCP segments to closed ports");
137 VNET_DEFINE(int, blackhole) = 0;
138 #define V_blackhole VNET(blackhole)
139 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
140 &VNET_NAME(blackhole), 0,
141 "Do not send RST on segments to closed ports");
143 VNET_DEFINE(int, tcp_delack_enabled) = 1;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
145 &VNET_NAME(tcp_delack_enabled), 0,
146 "Delay ACK to try and piggyback it onto a data packet");
148 VNET_DEFINE(int, drop_synfin) = 0;
149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
150 &VNET_NAME(drop_synfin), 0,
151 "Drop TCP packets with SYN+FIN set");
153 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
155 &VNET_NAME(tcp_do_rfc6675_pipe), 0,
156 "Use calculated pipe/in-flight bytes per RFC 6675");
158 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
160 &VNET_NAME(tcp_do_rfc3042), 0,
161 "Enable RFC 3042 (Limited Transmit)");
163 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
165 &VNET_NAME(tcp_do_rfc3390), 0,
166 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
168 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
170 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
171 "Slow-start flight size (initial congestion window) in number of segments");
173 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
175 &VNET_NAME(tcp_do_rfc3465), 0,
176 "Enable RFC 3465 (Appropriate Byte Counting)");
178 VNET_DEFINE(int, tcp_abc_l_var) = 2;
179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
180 &VNET_NAME(tcp_abc_l_var), 2,
181 "Cap the max cwnd increment during slow-start to this number of segments");
183 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
185 VNET_DEFINE(int, tcp_do_ecn) = 2;
186 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
187 &VNET_NAME(tcp_do_ecn), 0,
190 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
191 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
192 &VNET_NAME(tcp_ecn_maxretries), 0,
193 "Max retries before giving up on ECN");
195 VNET_DEFINE(int, tcp_insecure_syn) = 0;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
197 &VNET_NAME(tcp_insecure_syn), 0,
198 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
200 VNET_DEFINE(int, tcp_insecure_rst) = 0;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
202 &VNET_NAME(tcp_insecure_rst), 0,
203 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
205 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
206 #define V_tcp_recvspace VNET(tcp_recvspace)
207 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
208 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
210 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
212 &VNET_NAME(tcp_do_autorcvbuf), 0,
213 "Enable automatic receive buffer sizing");
215 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW,
217 &VNET_NAME(tcp_autorcvbuf_inc), 0,
218 "Incrementor step size of automatic receive buffer");
220 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
222 &VNET_NAME(tcp_autorcvbuf_max), 0,
223 "Max size of automatic receive buffer");
225 VNET_DEFINE(struct inpcbhead, tcb);
226 #define tcb6 tcb /* for KAME src sync over BSD*'s */
227 VNET_DEFINE(struct inpcbinfo, tcbinfo);
230 * TCP statistics are stored in an array of counter(9)s, which size matches
231 * size of struct tcpstat. TCP running connection count is a regular array.
233 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
234 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
235 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
236 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
237 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
238 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
239 "TCP connection counts by TCP state");
242 tcp_vnet_init(const void *unused)
245 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
246 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
248 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
249 tcp_vnet_init, NULL);
253 tcp_vnet_uninit(const void *unused)
256 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
257 VNET_PCPUSTAT_FREE(tcpstat);
259 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
260 tcp_vnet_uninit, NULL);
264 * Kernel module interface for updating tcpstat. The argument is an index
265 * into tcpstat treated as an array.
268 kmod_tcpstat_inc(int statnum)
271 counter_u64_add(VNET(tcpstat)[statnum], 1);
276 * Wrapper for the TCP established input helper hook.
279 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
281 struct tcp_hhook_data hhook_data;
283 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
288 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
295 * CC wrapper hook functions
298 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
301 INP_WLOCK_ASSERT(tp->t_inpcb);
303 tp->ccv->nsegs = nsegs;
304 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
305 if (tp->snd_cwnd <= tp->snd_wnd)
306 tp->ccv->flags |= CCF_CWND_LIMITED;
308 tp->ccv->flags &= ~CCF_CWND_LIMITED;
310 if (type == CC_ACK) {
311 if (tp->snd_cwnd > tp->snd_ssthresh) {
312 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
313 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp));
314 if (tp->t_bytes_acked >= tp->snd_cwnd) {
315 tp->t_bytes_acked -= tp->snd_cwnd;
316 tp->ccv->flags |= CCF_ABC_SENTAWND;
319 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
320 tp->t_bytes_acked = 0;
324 if (CC_ALGO(tp)->ack_received != NULL) {
325 /* XXXLAS: Find a way to live without this */
326 tp->ccv->curack = th->th_ack;
327 CC_ALGO(tp)->ack_received(tp->ccv, type);
332 cc_conn_init(struct tcpcb *tp)
334 struct hc_metrics_lite metrics;
335 struct inpcb *inp = tp->t_inpcb;
339 INP_WLOCK_ASSERT(tp->t_inpcb);
341 tcp_hc_get(&inp->inp_inc, &metrics);
342 maxseg = tcp_maxseg(tp);
344 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
346 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
347 TCPSTAT_INC(tcps_usedrtt);
348 if (metrics.rmx_rttvar) {
349 tp->t_rttvar = metrics.rmx_rttvar;
350 TCPSTAT_INC(tcps_usedrttvar);
352 /* default variation is +- 1 rtt */
354 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
356 TCPT_RANGESET(tp->t_rxtcur,
357 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
358 tp->t_rttmin, TCPTV_REXMTMAX);
360 if (metrics.rmx_ssthresh) {
362 * There's some sort of gateway or interface
363 * buffer limit on the path. Use this to set
364 * the slow start threshold, but set the
365 * threshold to no less than 2*mss.
367 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
368 TCPSTAT_INC(tcps_usedssthresh);
372 * Set the initial slow-start flight size.
374 * If a SYN or SYN/ACK was lost and retransmitted, we have to
375 * reduce the initial CWND to one segment as congestion is likely
376 * requiring us to be cautious.
378 if (tp->snd_cwnd == 1)
379 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
381 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
383 if (CC_ALGO(tp)->conn_init != NULL)
384 CC_ALGO(tp)->conn_init(tp->ccv);
388 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
392 INP_WLOCK_ASSERT(tp->t_inpcb);
396 if (!IN_FASTRECOVERY(tp->t_flags)) {
397 tp->snd_recover = tp->snd_max;
398 if (tp->t_flags & TF_ECN_PERMIT)
399 tp->t_flags |= TF_ECN_SND_CWR;
403 if (!IN_CONGRECOVERY(tp->t_flags)) {
404 TCPSTAT_INC(tcps_ecn_rcwnd);
405 tp->snd_recover = tp->snd_max;
406 if (tp->t_flags & TF_ECN_PERMIT)
407 tp->t_flags |= TF_ECN_SND_CWR;
411 maxseg = tcp_maxseg(tp);
413 tp->t_bytes_acked = 0;
414 EXIT_RECOVERY(tp->t_flags);
415 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
417 tp->snd_cwnd = maxseg;
420 TCPSTAT_INC(tcps_sndrexmitbad);
421 /* RTO was unnecessary, so reset everything. */
422 tp->snd_cwnd = tp->snd_cwnd_prev;
423 tp->snd_ssthresh = tp->snd_ssthresh_prev;
424 tp->snd_recover = tp->snd_recover_prev;
425 if (tp->t_flags & TF_WASFRECOVERY)
426 ENTER_FASTRECOVERY(tp->t_flags);
427 if (tp->t_flags & TF_WASCRECOVERY)
428 ENTER_CONGRECOVERY(tp->t_flags);
429 tp->snd_nxt = tp->snd_max;
430 tp->t_flags &= ~TF_PREVVALID;
435 if (CC_ALGO(tp)->cong_signal != NULL) {
437 tp->ccv->curack = th->th_ack;
438 CC_ALGO(tp)->cong_signal(tp->ccv, type);
443 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
445 INP_WLOCK_ASSERT(tp->t_inpcb);
447 /* XXXLAS: KASSERT that we're in recovery? */
449 if (CC_ALGO(tp)->post_recovery != NULL) {
450 tp->ccv->curack = th->th_ack;
451 CC_ALGO(tp)->post_recovery(tp->ccv);
453 /* XXXLAS: EXIT_RECOVERY ? */
454 tp->t_bytes_acked = 0;
458 * Indicate whether this ack should be delayed. We can delay the ack if
459 * following conditions are met:
460 * - There is no delayed ack timer in progress.
461 * - Our last ack wasn't a 0-sized window. We never want to delay
462 * the ack that opens up a 0-sized window.
463 * - LRO wasn't used for this segment. We make sure by checking that the
464 * segment size is not larger than the MSS.
466 #define DELAY_ACK(tp, tlen) \
467 ((!tcp_timer_active(tp, TT_DELACK) && \
468 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
469 (tlen <= tp->t_maxseg) && \
470 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
473 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
475 INP_WLOCK_ASSERT(tp->t_inpcb);
477 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
478 switch (iptos & IPTOS_ECN_MASK) {
480 tp->ccv->flags |= CCF_IPHDR_CE;
483 tp->ccv->flags &= ~CCF_IPHDR_CE;
486 tp->ccv->flags &= ~CCF_IPHDR_CE;
490 if (th->th_flags & TH_CWR)
491 tp->ccv->flags |= CCF_TCPHDR_CWR;
493 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
495 if (tp->t_flags & TF_DELACK)
496 tp->ccv->flags |= CCF_DELACK;
498 tp->ccv->flags &= ~CCF_DELACK;
500 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
502 if (tp->ccv->flags & CCF_ACKNOW)
503 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
508 * TCP input handling is split into multiple parts:
509 * tcp6_input is a thin wrapper around tcp_input for the extended
510 * ip6_protox[] call format in ip6_input
511 * tcp_input handles primary segment validation, inpcb lookup and
512 * SYN processing on listen sockets
513 * tcp_do_segment processes the ACK and text of the segment for
514 * establishing, established and closing connections
518 tcp6_input(struct mbuf **mp, int *offp, int proto)
520 struct mbuf *m = *mp;
521 struct in6_ifaddr *ia6;
524 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
527 * draft-itojun-ipv6-tcp-to-anycast
528 * better place to put this in?
530 ip6 = mtod(m, struct ip6_hdr *);
531 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
532 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
535 ifa_free(&ia6->ia_ifa);
536 ip6 = mtod(m, struct ip6_hdr *);
537 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
538 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
539 return (IPPROTO_DONE);
542 ifa_free(&ia6->ia_ifa);
544 return (tcp_input(mp, offp, proto));
549 tcp_input(struct mbuf **mp, int *offp, int proto)
551 struct mbuf *m = *mp;
552 struct tcphdr *th = NULL;
553 struct ip *ip = NULL;
554 struct inpcb *inp = NULL;
555 struct tcpcb *tp = NULL;
556 struct socket *so = NULL;
566 int rstreason = 0; /* For badport_bandlim accounting purposes */
568 struct m_tag *fwd_tag = NULL;
569 struct epoch_tracker et;
571 struct ip6_hdr *ip6 = NULL;
574 const void *ip6 = NULL;
576 struct tcpopt to; /* options in this segment */
577 char *s = NULL; /* address and port logging */
581 * The size of tcp_saveipgen must be the size of the max ip header,
584 u_char tcp_saveipgen[IP6_HDR_LEN];
585 struct tcphdr tcp_savetcp;
590 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
597 TCPSTAT_INC(tcps_rcvtotal);
601 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
603 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
604 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
606 TCPSTAT_INC(tcps_rcvshort);
607 return (IPPROTO_DONE);
611 ip6 = mtod(m, struct ip6_hdr *);
612 th = (struct tcphdr *)((caddr_t)ip6 + off0);
613 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
614 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
615 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
616 th->th_sum = m->m_pkthdr.csum_data;
618 th->th_sum = in6_cksum_pseudo(ip6, tlen,
619 IPPROTO_TCP, m->m_pkthdr.csum_data);
620 th->th_sum ^= 0xffff;
622 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
624 TCPSTAT_INC(tcps_rcvbadsum);
629 * Be proactive about unspecified IPv6 address in source.
630 * As we use all-zero to indicate unbounded/unconnected pcb,
631 * unspecified IPv6 address can be used to confuse us.
633 * Note that packets with unspecified IPv6 destination is
634 * already dropped in ip6_input.
636 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
640 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
643 #if defined(INET) && defined(INET6)
649 * Get IP and TCP header together in first mbuf.
650 * Note: IP leaves IP header in first mbuf.
652 if (off0 > sizeof (struct ip)) {
654 off0 = sizeof(struct ip);
656 if (m->m_len < sizeof (struct tcpiphdr)) {
657 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
659 TCPSTAT_INC(tcps_rcvshort);
660 return (IPPROTO_DONE);
663 ip = mtod(m, struct ip *);
664 th = (struct tcphdr *)((caddr_t)ip + off0);
665 tlen = ntohs(ip->ip_len) - off0;
668 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
669 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
670 th->th_sum = m->m_pkthdr.csum_data;
672 th->th_sum = in_pseudo(ip->ip_src.s_addr,
674 htonl(m->m_pkthdr.csum_data + tlen +
676 th->th_sum ^= 0xffff;
678 struct ipovly *ipov = (struct ipovly *)ip;
681 * Checksum extended TCP header and data.
684 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
685 ipov->ih_len = htons(tlen);
686 th->th_sum = in_cksum(m, len);
687 /* Reset length for SDT probes. */
688 ip->ip_len = htons(len);
691 /* Re-initialization for later version check */
692 ip->ip_v = IPVERSION;
693 ip->ip_hl = off0 >> 2;
697 TCPSTAT_INC(tcps_rcvbadsum);
704 * Check that TCP offset makes sense,
705 * pull out TCP options and adjust length. XXX
707 off = th->th_off << 2;
708 if (off < sizeof (struct tcphdr) || off > tlen) {
709 TCPSTAT_INC(tcps_rcvbadoff);
712 tlen -= off; /* tlen is used instead of ti->ti_len */
713 if (off > sizeof (struct tcphdr)) {
716 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
717 ip6 = mtod(m, struct ip6_hdr *);
718 th = (struct tcphdr *)((caddr_t)ip6 + off0);
721 #if defined(INET) && defined(INET6)
726 if (m->m_len < sizeof(struct ip) + off) {
727 if ((m = m_pullup(m, sizeof (struct ip) + off))
729 TCPSTAT_INC(tcps_rcvshort);
730 return (IPPROTO_DONE);
732 ip = mtod(m, struct ip *);
733 th = (struct tcphdr *)((caddr_t)ip + off0);
737 optlen = off - sizeof (struct tcphdr);
738 optp = (u_char *)(th + 1);
740 thflags = th->th_flags;
743 * Convert TCP protocol specific fields to host format.
745 tcp_fields_to_host(th);
748 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
750 drop_hdrlen = off0 + off;
753 * Locate pcb for segment; if we're likely to add or remove a
754 * connection then first acquire pcbinfo lock. There are three cases
755 * where we might discover later we need a write lock despite the
756 * flags: ACKs moving a connection out of the syncache, ACKs for a
757 * connection in TIMEWAIT and SYNs not targeting a listening socket.
759 if ((thflags & (TH_FIN | TH_RST)) != 0) {
760 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
761 ti_locked = TI_RLOCKED;
763 ti_locked = TI_UNLOCKED;
766 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
770 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
772 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
775 #if defined(INET) && !defined(INET6)
776 (m->m_flags & M_IP_NEXTHOP)
779 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
783 if (ti_locked == TI_RLOCKED) {
784 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
786 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
790 if (isipv6 && fwd_tag != NULL) {
791 struct sockaddr_in6 *next_hop6;
793 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
795 * Transparently forwarded. Pretend to be the destination.
796 * Already got one like this?
798 inp = in6_pcblookup_mbuf(&V_tcbinfo,
799 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
800 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
803 * It's new. Try to find the ambushing socket.
804 * Because we've rewritten the destination address,
805 * any hardware-generated hash is ignored.
807 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
808 th->th_sport, &next_hop6->sin6_addr,
809 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
810 th->th_dport, INPLOOKUP_WILDCARD |
811 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
814 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
815 th->th_sport, &ip6->ip6_dst, th->th_dport,
816 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
817 m->m_pkthdr.rcvif, m);
820 #if defined(INET6) && defined(INET)
824 if (fwd_tag != NULL) {
825 struct sockaddr_in *next_hop;
827 next_hop = (struct sockaddr_in *)(fwd_tag+1);
829 * Transparently forwarded. Pretend to be the destination.
830 * already got one like this?
832 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
833 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
834 m->m_pkthdr.rcvif, m);
837 * It's new. Try to find the ambushing socket.
838 * Because we've rewritten the destination address,
839 * any hardware-generated hash is ignored.
841 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
842 th->th_sport, next_hop->sin_addr,
843 next_hop->sin_port ? ntohs(next_hop->sin_port) :
844 th->th_dport, INPLOOKUP_WILDCARD |
845 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
848 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
849 th->th_sport, ip->ip_dst, th->th_dport,
850 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
851 m->m_pkthdr.rcvif, m);
855 * If the INPCB does not exist then all data in the incoming
856 * segment is discarded and an appropriate RST is sent back.
857 * XXX MRT Send RST using which routing table?
861 * Log communication attempts to ports that are not
864 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
865 tcp_log_in_vain == 2) {
866 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
867 log(LOG_INFO, "%s; %s: Connection attempt "
868 "to closed port\n", s, __func__);
871 * When blackholing do not respond with a RST but
872 * completely ignore the segment and drop it.
874 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
878 rstreason = BANDLIM_RST_CLOSEDPORT;
881 INP_WLOCK_ASSERT(inp);
883 * While waiting for inp lock during the lookup, another thread
884 * can have dropped the inpcb, in which case we need to loop back
885 * and try to find a new inpcb to deliver to.
887 if (inp->inp_flags & INP_DROPPED) {
892 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
893 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
894 ((inp->inp_socket == NULL) ||
895 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
896 inp->inp_flowid = m->m_pkthdr.flowid;
897 inp->inp_flowtype = M_HASHTYPE_GET(m);
899 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
901 if (isipv6 && IPSEC_ENABLED(ipv6) &&
902 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
910 if (IPSEC_ENABLED(ipv4) &&
911 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
918 * Check the minimum TTL for socket.
920 if (inp->inp_ip_minttl != 0) {
923 if (inp->inp_ip_minttl > ip6->ip6_hlim)
927 if (inp->inp_ip_minttl > ip->ip_ttl)
932 * A previous connection in TIMEWAIT state is supposed to catch stray
933 * or duplicate segments arriving late. If this segment was a
934 * legitimate new connection attempt, the old INPCB gets removed and
935 * we can try again to find a listening socket.
937 * At this point, due to earlier optimism, we may hold only an inpcb
938 * lock, and not the inpcbinfo write lock. If so, we need to try to
939 * acquire it, or if that fails, acquire a reference on the inpcb,
940 * drop all locks, acquire a global write lock, and then re-acquire
941 * the inpcb lock. We may at that point discover that another thread
942 * has tried to free the inpcb, in which case we need to loop back
943 * and try to find a new inpcb to deliver to.
945 * XXXRW: It may be time to rethink timewait locking.
947 if (inp->inp_flags & INP_TIMEWAIT) {
948 if (ti_locked == TI_UNLOCKED) {
949 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
950 ti_locked = TI_RLOCKED;
952 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
954 if (thflags & TH_SYN)
955 tcp_dooptions(&to, optp, optlen, TO_SYN);
957 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
959 if (tcp_twcheck(inp, &to, th, m, tlen))
961 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
962 return (IPPROTO_DONE);
965 * The TCPCB may no longer exist if the connection is winding
966 * down or it is in the CLOSED state. Either way we drop the
967 * segment and send an appropriate response.
970 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
971 rstreason = BANDLIM_RST_CLOSEDPORT;
976 if (tp->t_flags & TF_TOE) {
977 tcp_offload_input(tp, m);
978 m = NULL; /* consumed by the TOE driver */
984 * We've identified a valid inpcb, but it could be that we need an
985 * inpcbinfo write lock but don't hold it. In this case, attempt to
986 * acquire using the same strategy as the TIMEWAIT case above. If we
987 * relock, we have to jump back to 'relocked' as the connection might
988 * now be in TIMEWAIT.
991 if ((thflags & (TH_FIN | TH_RST)) != 0)
992 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
994 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) ||
995 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) &&
996 !IS_FASTOPEN(tp->t_flags)))) {
997 if (ti_locked == TI_UNLOCKED) {
998 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
999 ti_locked = TI_RLOCKED;
1001 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1005 INP_WLOCK_ASSERT(inp);
1006 if (mac_inpcb_check_deliver(inp, m))
1009 so = inp->inp_socket;
1010 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1012 if (so->so_options & SO_DEBUG) {
1013 ostate = tp->t_state;
1016 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1019 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1022 #endif /* TCPDEBUG */
1024 * When the socket is accepting connections (the INPCB is in LISTEN
1025 * state) we look into the SYN cache if this is a new connection
1026 * attempt or the completion of a previous one.
1028 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1029 ("%s: so accepting but tp %p not listening", __func__, tp));
1030 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1031 struct in_conninfo inc;
1033 bzero(&inc, sizeof(inc));
1036 inc.inc_flags |= INC_ISIPV6;
1037 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1038 inc.inc_flags |= INC_IPV6MINMTU;
1039 inc.inc6_faddr = ip6->ip6_src;
1040 inc.inc6_laddr = ip6->ip6_dst;
1044 inc.inc_faddr = ip->ip_src;
1045 inc.inc_laddr = ip->ip_dst;
1047 inc.inc_fport = th->th_sport;
1048 inc.inc_lport = th->th_dport;
1049 inc.inc_fibnum = so->so_fibnum;
1052 * Check for an existing connection attempt in syncache if
1053 * the flag is only ACK. A successful lookup creates a new
1054 * socket appended to the listen queue in SYN_RECEIVED state.
1056 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1058 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1060 * Parse the TCP options here because
1061 * syncookies need access to the reflected
1064 tcp_dooptions(&to, optp, optlen, 0);
1066 * NB: syncache_expand() doesn't unlock
1067 * inp and tcpinfo locks.
1069 rstreason = syncache_expand(&inc, &to, th, &so, m);
1070 if (rstreason < 0) {
1072 * A failing TCP MD5 signature comparison
1073 * must result in the segment being dropped
1074 * and must not produce any response back
1078 } else if (rstreason == 0) {
1080 * No syncache entry or ACK was not
1081 * for our SYN/ACK. Send a RST.
1082 * NB: syncache did its own logging
1083 * of the failure cause.
1085 rstreason = BANDLIM_RST_OPENPORT;
1091 * We completed the 3-way handshake
1092 * but could not allocate a socket
1093 * either due to memory shortage,
1094 * listen queue length limits or
1095 * global socket limits. Send RST
1096 * or wait and have the remote end
1097 * retransmit the ACK for another
1100 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1101 log(LOG_DEBUG, "%s; %s: Listen socket: "
1102 "Socket allocation failed due to "
1103 "limits or memory shortage, %s\n",
1105 V_tcp_sc_rst_sock_fail ?
1106 "sending RST" : "try again");
1107 if (V_tcp_sc_rst_sock_fail) {
1108 rstreason = BANDLIM_UNLIMITED;
1114 * Socket is created in state SYN_RECEIVED.
1115 * Unlock the listen socket, lock the newly
1116 * created socket and update the tp variable.
1118 INP_WUNLOCK(inp); /* listen socket */
1119 inp = sotoinpcb(so);
1121 * New connection inpcb is already locked by
1122 * syncache_expand().
1124 INP_WLOCK_ASSERT(inp);
1125 tp = intotcpcb(inp);
1126 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1127 ("%s: ", __func__));
1129 * Process the segment and the data it
1130 * contains. tcp_do_segment() consumes
1131 * the mbuf chain and unlocks the inpcb.
1133 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1134 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1136 if (ti_locked == TI_RLOCKED)
1137 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1138 return (IPPROTO_DONE);
1141 * Segment flag validation for new connection attempts:
1143 * Our (SYN|ACK) response was rejected.
1144 * Check with syncache and remove entry to prevent
1147 * NB: syncache_chkrst does its own logging of failure
1150 if (thflags & TH_RST) {
1151 syncache_chkrst(&inc, th, m);
1155 * We can't do anything without SYN.
1157 if ((thflags & TH_SYN) == 0) {
1158 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1159 log(LOG_DEBUG, "%s; %s: Listen socket: "
1160 "SYN is missing, segment ignored\n",
1162 TCPSTAT_INC(tcps_badsyn);
1166 * (SYN|ACK) is bogus on a listen socket.
1168 if (thflags & TH_ACK) {
1169 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1170 log(LOG_DEBUG, "%s; %s: Listen socket: "
1171 "SYN|ACK invalid, segment rejected\n",
1173 syncache_badack(&inc); /* XXX: Not needed! */
1174 TCPSTAT_INC(tcps_badsyn);
1175 rstreason = BANDLIM_RST_OPENPORT;
1179 * If the drop_synfin option is enabled, drop all
1180 * segments with both the SYN and FIN bits set.
1181 * This prevents e.g. nmap from identifying the
1183 * XXX: Poor reasoning. nmap has other methods
1184 * and is constantly refining its stack detection
1186 * XXX: This is a violation of the TCP specification
1187 * and was used by RFC1644.
1189 if ((thflags & TH_FIN) && V_drop_synfin) {
1190 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1191 log(LOG_DEBUG, "%s; %s: Listen socket: "
1192 "SYN|FIN segment ignored (based on "
1193 "sysctl setting)\n", s, __func__);
1194 TCPSTAT_INC(tcps_badsyn);
1198 * Segment's flags are (SYN) or (SYN|FIN).
1200 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1201 * as they do not affect the state of the TCP FSM.
1202 * The data pointed to by TH_URG and th_urp is ignored.
1204 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1205 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1206 KASSERT(thflags & (TH_SYN),
1207 ("%s: Listen socket: TH_SYN not set", __func__));
1210 * If deprecated address is forbidden,
1211 * we do not accept SYN to deprecated interface
1212 * address to prevent any new inbound connection from
1213 * getting established.
1214 * When we do not accept SYN, we send a TCP RST,
1215 * with deprecated source address (instead of dropping
1216 * it). We compromise it as it is much better for peer
1217 * to send a RST, and RST will be the final packet
1220 * If we do not forbid deprecated addresses, we accept
1221 * the SYN packet. RFC2462 does not suggest dropping
1223 * If we decipher RFC2462 5.5.4, it says like this:
1224 * 1. use of deprecated addr with existing
1225 * communication is okay - "SHOULD continue to be
1227 * 2. use of it with new communication:
1228 * (2a) "SHOULD NOT be used if alternate address
1229 * with sufficient scope is available"
1230 * (2b) nothing mentioned otherwise.
1231 * Here we fall into (2b) case as we have no choice in
1232 * our source address selection - we must obey the peer.
1234 * The wording in RFC2462 is confusing, and there are
1235 * multiple description text for deprecated address
1236 * handling - worse, they are not exactly the same.
1237 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1239 if (isipv6 && !V_ip6_use_deprecated) {
1240 struct in6_ifaddr *ia6;
1242 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1244 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1245 ifa_free(&ia6->ia_ifa);
1246 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1247 log(LOG_DEBUG, "%s; %s: Listen socket: "
1248 "Connection attempt to deprecated "
1249 "IPv6 address rejected\n",
1251 rstreason = BANDLIM_RST_OPENPORT;
1255 ifa_free(&ia6->ia_ifa);
1259 * Basic sanity checks on incoming SYN requests:
1260 * Don't respond if the destination is a link layer
1261 * broadcast according to RFC1122 4.2.3.10, p. 104.
1262 * If it is from this socket it must be forged.
1263 * Don't respond if the source or destination is a
1264 * global or subnet broad- or multicast address.
1265 * Note that it is quite possible to receive unicast
1266 * link-layer packets with a broadcast IP address. Use
1267 * in_broadcast() to find them.
1269 if (m->m_flags & (M_BCAST|M_MCAST)) {
1270 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1271 log(LOG_DEBUG, "%s; %s: Listen socket: "
1272 "Connection attempt from broad- or multicast "
1273 "link layer address ignored\n", s, __func__);
1278 if (th->th_dport == th->th_sport &&
1279 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1280 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1281 log(LOG_DEBUG, "%s; %s: Listen socket: "
1282 "Connection attempt to/from self "
1283 "ignored\n", s, __func__);
1286 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1287 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1288 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1289 log(LOG_DEBUG, "%s; %s: Listen socket: "
1290 "Connection attempt from/to multicast "
1291 "address ignored\n", s, __func__);
1296 #if defined(INET) && defined(INET6)
1301 if (th->th_dport == th->th_sport &&
1302 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1303 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1304 log(LOG_DEBUG, "%s; %s: Listen socket: "
1305 "Connection attempt from/to self "
1306 "ignored\n", s, __func__);
1309 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1310 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1311 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1312 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1313 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1314 log(LOG_DEBUG, "%s; %s: Listen socket: "
1315 "Connection attempt from/to broad- "
1316 "or multicast address ignored\n",
1323 * SYN appears to be valid. Create compressed TCP state
1327 if (so->so_options & SO_DEBUG)
1328 tcp_trace(TA_INPUT, ostate, tp,
1329 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1331 TCP_PROBE3(debug__input, tp, th, m);
1332 tcp_dooptions(&to, optp, optlen, TO_SYN);
1333 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL))
1334 goto tfo_socket_result;
1337 * Entry added to syncache and mbuf consumed.
1338 * Only the listen socket is unlocked by syncache_add().
1340 if (ti_locked == TI_RLOCKED) {
1341 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1342 ti_locked = TI_UNLOCKED;
1344 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1345 return (IPPROTO_DONE);
1346 } else if (tp->t_state == TCPS_LISTEN) {
1348 * When a listen socket is torn down the SO_ACCEPTCONN
1349 * flag is removed first while connections are drained
1350 * from the accept queue in a unlock/lock cycle of the
1351 * ACCEPT_LOCK, opening a race condition allowing a SYN
1352 * attempt go through unhandled.
1356 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1357 if (tp->t_flags & TF_SIGNATURE) {
1358 tcp_dooptions(&to, optp, optlen, thflags);
1359 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1360 TCPSTAT_INC(tcps_sig_err_nosigopt);
1363 if (!TCPMD5_ENABLED() ||
1364 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1368 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1371 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1372 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1373 * the inpcb, and unlocks pcbinfo.
1375 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1376 if (ti_locked == TI_RLOCKED)
1377 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1378 return (IPPROTO_DONE);
1381 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1383 if (ti_locked == TI_RLOCKED) {
1384 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1385 ti_locked = TI_UNLOCKED;
1389 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1390 "ti_locked: %d", __func__, ti_locked));
1391 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1396 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1399 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1400 m = NULL; /* mbuf chain got consumed. */
1405 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1407 if (ti_locked == TI_RLOCKED) {
1408 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1409 ti_locked = TI_UNLOCKED;
1413 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1414 "ti_locked: %d", __func__, ti_locked));
1415 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1423 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1428 return (IPPROTO_DONE);
1432 * Automatic sizing of receive socket buffer. Often the send
1433 * buffer size is not optimally adjusted to the actual network
1434 * conditions at hand (delay bandwidth product). Setting the
1435 * buffer size too small limits throughput on links with high
1436 * bandwidth and high delay (eg. trans-continental/oceanic links).
1438 * On the receive side the socket buffer memory is only rarely
1439 * used to any significant extent. This allows us to be much
1440 * more aggressive in scaling the receive socket buffer. For
1441 * the case that the buffer space is actually used to a large
1442 * extent and we run out of kernel memory we can simply drop
1443 * the new segments; TCP on the sender will just retransmit it
1444 * later. Setting the buffer size too big may only consume too
1445 * much kernel memory if the application doesn't read() from
1446 * the socket or packet loss or reordering makes use of the
1449 * The criteria to step up the receive buffer one notch are:
1450 * 1. Application has not set receive buffer size with
1451 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1452 * 2. the number of bytes received during the time it takes
1453 * one timestamp to be reflected back to us (the RTT);
1454 * 3. received bytes per RTT is within seven eighth of the
1455 * current socket buffer size;
1456 * 4. receive buffer size has not hit maximal automatic size;
1458 * This algorithm does one step per RTT at most and only if
1459 * we receive a bulk stream w/o packet losses or reorderings.
1460 * Shrinking the buffer during idle times is not necessary as
1461 * it doesn't consume any memory when idle.
1463 * TODO: Only step up if the application is actually serving
1464 * the buffer to better manage the socket buffer resources.
1467 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1468 struct tcpcb *tp, int tlen)
1472 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1473 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1474 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1475 (tp->t_srtt >> TCP_RTT_SHIFT)) {
1476 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) &&
1477 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1478 newsize = min(so->so_rcv.sb_hiwat +
1479 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max);
1481 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1483 /* Start over with next RTT. */
1487 tp->rfbuf_cnt += tlen; /* add up */
1494 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1495 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1497 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1498 int rstreason, todrop, win;
1502 struct in_conninfo *inc;
1509 * The size of tcp_saveipgen must be the size of the max ip header,
1512 u_char tcp_saveipgen[IP6_HDR_LEN];
1513 struct tcphdr tcp_savetcp;
1516 thflags = th->th_flags;
1517 inc = &tp->t_inpcb->inp_inc;
1518 tp->sackhint.last_sack_ack = 0;
1520 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1522 * If this is either a state-changing packet or current state isn't
1523 * established, we require a write lock on tcbinfo. Otherwise, we
1524 * allow the tcbinfo to be in either alocked or unlocked, as the
1525 * caller may have unnecessarily acquired a write lock due to a race.
1527 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1528 tp->t_state != TCPS_ESTABLISHED) {
1529 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1531 INP_WLOCK_ASSERT(tp->t_inpcb);
1532 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1534 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1538 /* Save segment, if requested. */
1539 tcp_pcap_add(th, m, &(tp->t_inpkts));
1541 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1544 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1545 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1546 log(LOG_DEBUG, "%s; %s: "
1547 "SYN|FIN segment ignored (based on "
1548 "sysctl setting)\n", s, __func__);
1555 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1556 * check SEQ.ACK first.
1558 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1559 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1560 rstreason = BANDLIM_UNLIMITED;
1565 * Segment received on connection.
1566 * Reset idle time and keep-alive timer.
1567 * XXX: This should be done after segment
1568 * validation to ignore broken/spoofed segs.
1570 tp->t_rcvtime = ticks;
1573 * Scale up the window into a 32-bit value.
1574 * For the SYN_SENT state the scale is zero.
1576 tiwin = th->th_win << tp->snd_scale;
1579 * TCP ECN processing.
1581 if (tp->t_flags & TF_ECN_PERMIT) {
1582 if (thflags & TH_CWR)
1583 tp->t_flags &= ~TF_ECN_SND_ECE;
1584 switch (iptos & IPTOS_ECN_MASK) {
1586 tp->t_flags |= TF_ECN_SND_ECE;
1587 TCPSTAT_INC(tcps_ecn_ce);
1589 case IPTOS_ECN_ECT0:
1590 TCPSTAT_INC(tcps_ecn_ect0);
1592 case IPTOS_ECN_ECT1:
1593 TCPSTAT_INC(tcps_ecn_ect1);
1597 /* Process a packet differently from RFC3168. */
1598 cc_ecnpkt_handler(tp, th, iptos);
1600 /* Congestion experienced. */
1601 if (thflags & TH_ECE) {
1602 cc_cong_signal(tp, th, CC_ECN);
1607 * Parse options on any incoming segment.
1609 tcp_dooptions(&to, (u_char *)(th + 1),
1610 (th->th_off << 2) - sizeof(struct tcphdr),
1611 (thflags & TH_SYN) ? TO_SYN : 0);
1613 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1614 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1615 (to.to_flags & TOF_SIGNATURE) == 0) {
1616 TCPSTAT_INC(tcps_sig_err_sigopt);
1617 /* XXX: should drop? */
1621 * If echoed timestamp is later than the current time,
1622 * fall back to non RFC1323 RTT calculation. Normalize
1623 * timestamp if syncookies were used when this connection
1626 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1627 to.to_tsecr -= tp->ts_offset;
1628 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1630 else if (tp->t_flags & TF_PREVVALID &&
1631 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1632 cc_cong_signal(tp, th, CC_RTO_ERR);
1635 * Process options only when we get SYN/ACK back. The SYN case
1636 * for incoming connections is handled in tcp_syncache.
1637 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1638 * or <SYN,ACK>) segment itself is never scaled.
1639 * XXX this is traditional behavior, may need to be cleaned up.
1641 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1642 if ((to.to_flags & TOF_SCALE) &&
1643 (tp->t_flags & TF_REQ_SCALE)) {
1644 tp->t_flags |= TF_RCVD_SCALE;
1645 tp->snd_scale = to.to_wscale;
1648 * Initial send window. It will be updated with
1649 * the next incoming segment to the scaled value.
1651 tp->snd_wnd = th->th_win;
1652 if (to.to_flags & TOF_TS) {
1653 tp->t_flags |= TF_RCVD_TSTMP;
1654 tp->ts_recent = to.to_tsval;
1655 tp->ts_recent_age = tcp_ts_getticks();
1657 if (to.to_flags & TOF_MSS)
1658 tcp_mss(tp, to.to_mss);
1659 if ((tp->t_flags & TF_SACK_PERMIT) &&
1660 (to.to_flags & TOF_SACKPERM) == 0)
1661 tp->t_flags &= ~TF_SACK_PERMIT;
1662 if (IS_FASTOPEN(tp->t_flags)) {
1663 if (to.to_flags & TOF_FASTOPEN) {
1666 if (to.to_flags & TOF_MSS)
1669 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1673 tcp_fastopen_update_cache(tp, mss,
1674 to.to_tfo_len, to.to_tfo_cookie);
1676 tcp_fastopen_disable_path(tp);
1681 * If timestamps were negotiated during SYN/ACK they should
1682 * appear on every segment during this session and vice versa.
1684 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1685 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1686 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1687 "no action\n", s, __func__);
1691 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1692 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1693 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1694 "no action\n", s, __func__);
1700 * Header prediction: check for the two common cases
1701 * of a uni-directional data xfer. If the packet has
1702 * no control flags, is in-sequence, the window didn't
1703 * change and we're not retransmitting, it's a
1704 * candidate. If the length is zero and the ack moved
1705 * forward, we're the sender side of the xfer. Just
1706 * free the data acked & wake any higher level process
1707 * that was blocked waiting for space. If the length
1708 * is non-zero and the ack didn't move, we're the
1709 * receiver side. If we're getting packets in-order
1710 * (the reassembly queue is empty), add the data to
1711 * the socket buffer and note that we need a delayed ack.
1712 * Make sure that the hidden state-flags are also off.
1713 * Since we check for TCPS_ESTABLISHED first, it can only
1716 if (tp->t_state == TCPS_ESTABLISHED &&
1717 th->th_seq == tp->rcv_nxt &&
1718 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1719 tp->snd_nxt == tp->snd_max &&
1720 tiwin && tiwin == tp->snd_wnd &&
1721 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1723 ((to.to_flags & TOF_TS) == 0 ||
1724 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1727 * If last ACK falls within this segment's sequence numbers,
1728 * record the timestamp.
1729 * NOTE that the test is modified according to the latest
1730 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1732 if ((to.to_flags & TOF_TS) != 0 &&
1733 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1734 tp->ts_recent_age = tcp_ts_getticks();
1735 tp->ts_recent = to.to_tsval;
1739 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1740 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1741 !IN_RECOVERY(tp->t_flags) &&
1742 (to.to_flags & TOF_SACK) == 0 &&
1743 TAILQ_EMPTY(&tp->snd_holes)) {
1745 * This is a pure ack for outstanding data.
1747 TCPSTAT_INC(tcps_predack);
1750 * "bad retransmit" recovery without timestamps.
1752 if ((to.to_flags & TOF_TS) == 0 &&
1753 tp->t_rxtshift == 1 &&
1754 tp->t_flags & TF_PREVVALID &&
1755 (int)(ticks - tp->t_badrxtwin) < 0) {
1756 cc_cong_signal(tp, th, CC_RTO_ERR);
1760 * Recalculate the transmit timer / rtt.
1762 * Some boxes send broken timestamp replies
1763 * during the SYN+ACK phase, ignore
1764 * timestamps of 0 or we could calculate a
1765 * huge RTT and blow up the retransmit timer.
1767 if ((to.to_flags & TOF_TS) != 0 &&
1771 t = tcp_ts_getticks() - to.to_tsecr;
1772 if (!tp->t_rttlow || tp->t_rttlow > t)
1775 TCP_TS_TO_TICKS(t) + 1);
1776 } else if (tp->t_rtttime &&
1777 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1778 if (!tp->t_rttlow ||
1779 tp->t_rttlow > ticks - tp->t_rtttime)
1780 tp->t_rttlow = ticks - tp->t_rtttime;
1782 ticks - tp->t_rtttime);
1784 acked = BYTES_THIS_ACK(tp, th);
1787 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1788 hhook_run_tcp_est_in(tp, th, &to);
1791 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1792 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1793 sbdrop(&so->so_snd, acked);
1794 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1795 SEQ_LEQ(th->th_ack, tp->snd_recover))
1796 tp->snd_recover = th->th_ack - 1;
1799 * Let the congestion control algorithm update
1800 * congestion control related information. This
1801 * typically means increasing the congestion
1804 cc_ack_received(tp, th, nsegs, CC_ACK);
1806 tp->snd_una = th->th_ack;
1808 * Pull snd_wl2 up to prevent seq wrap relative
1811 tp->snd_wl2 = th->th_ack;
1816 * If all outstanding data are acked, stop
1817 * retransmit timer, otherwise restart timer
1818 * using current (possibly backed-off) value.
1819 * If process is waiting for space,
1820 * wakeup/selwakeup/signal. If data
1821 * are ready to send, let tcp_output
1822 * decide between more output or persist.
1825 if (so->so_options & SO_DEBUG)
1826 tcp_trace(TA_INPUT, ostate, tp,
1827 (void *)tcp_saveipgen,
1830 TCP_PROBE3(debug__input, tp, th, m);
1831 if (tp->snd_una == tp->snd_max)
1832 tcp_timer_activate(tp, TT_REXMT, 0);
1833 else if (!tcp_timer_active(tp, TT_PERSIST))
1834 tcp_timer_activate(tp, TT_REXMT,
1837 if (sbavail(&so->so_snd))
1838 (void) tp->t_fb->tfb_tcp_output(tp);
1841 } else if (th->th_ack == tp->snd_una &&
1842 tlen <= sbspace(&so->so_rcv)) {
1843 int newsize = 0; /* automatic sockbuf scaling */
1846 * This is a pure, in-sequence data packet with
1847 * nothing on the reassembly queue and we have enough
1848 * buffer space to take it.
1850 /* Clean receiver SACK report if present */
1851 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1852 tcp_clean_sackreport(tp);
1853 TCPSTAT_INC(tcps_preddat);
1854 tp->rcv_nxt += tlen;
1856 * Pull snd_wl1 up to prevent seq wrap relative to
1859 tp->snd_wl1 = th->th_seq;
1861 * Pull rcv_up up to prevent seq wrap relative to
1864 tp->rcv_up = tp->rcv_nxt;
1865 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1866 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1868 if (so->so_options & SO_DEBUG)
1869 tcp_trace(TA_INPUT, ostate, tp,
1870 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1872 TCP_PROBE3(debug__input, tp, th, m);
1874 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1876 /* Add data to socket buffer. */
1877 SOCKBUF_LOCK(&so->so_rcv);
1878 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1882 * Set new socket buffer size.
1883 * Give up when limit is reached.
1886 if (!sbreserve_locked(&so->so_rcv,
1888 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1889 m_adj(m, drop_hdrlen); /* delayed header drop */
1890 sbappendstream_locked(&so->so_rcv, m, 0);
1892 /* NB: sorwakeup_locked() does an implicit unlock. */
1893 sorwakeup_locked(so);
1894 if (DELAY_ACK(tp, tlen)) {
1895 tp->t_flags |= TF_DELACK;
1897 tp->t_flags |= TF_ACKNOW;
1898 tp->t_fb->tfb_tcp_output(tp);
1905 * Calculate amount of space in receive window,
1906 * and then do TCP input processing.
1907 * Receive window is amount of space in rcv queue,
1908 * but not less than advertised window.
1910 win = sbspace(&so->so_rcv);
1913 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1915 switch (tp->t_state) {
1918 * If the state is SYN_RECEIVED:
1919 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1921 case TCPS_SYN_RECEIVED:
1922 if ((thflags & TH_ACK) &&
1923 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1924 SEQ_GT(th->th_ack, tp->snd_max))) {
1925 rstreason = BANDLIM_RST_OPENPORT;
1928 if (IS_FASTOPEN(tp->t_flags)) {
1930 * When a TFO connection is in SYN_RECEIVED, the
1931 * only valid packets are the initial SYN, a
1932 * retransmit/copy of the initial SYN (possibly with
1933 * a subset of the original data), a valid ACK, a
1936 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1937 rstreason = BANDLIM_RST_OPENPORT;
1939 } else if (thflags & TH_SYN) {
1940 /* non-initial SYN is ignored */
1941 if ((tcp_timer_active(tp, TT_DELACK) ||
1942 tcp_timer_active(tp, TT_REXMT)))
1944 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1951 * If the state is SYN_SENT:
1952 * if seg contains a RST with valid ACK (SEQ.ACK has already
1953 * been verified), then drop the connection.
1954 * if seg contains a RST without an ACK, drop the seg.
1955 * if seg does not contain SYN, then drop the seg.
1956 * Otherwise this is an acceptable SYN segment
1957 * initialize tp->rcv_nxt and tp->irs
1958 * if seg contains ack then advance tp->snd_una
1959 * if seg contains an ECE and ECN support is enabled, the stream
1961 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1962 * arrange for segment to be acked (eventually)
1963 * continue processing rest of data/controls, beginning with URG
1966 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1967 TCP_PROBE5(connect__refused, NULL, tp,
1969 tp = tcp_drop(tp, ECONNREFUSED);
1971 if (thflags & TH_RST)
1973 if (!(thflags & TH_SYN))
1976 tp->irs = th->th_seq;
1978 if (thflags & TH_ACK) {
1979 int tfo_partial_ack = 0;
1981 TCPSTAT_INC(tcps_connects);
1984 mac_socketpeer_set_from_mbuf(m, so);
1986 /* Do window scaling on this connection? */
1987 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1988 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1989 tp->rcv_scale = tp->request_r_scale;
1991 tp->rcv_adv += min(tp->rcv_wnd,
1992 TCP_MAXWIN << tp->rcv_scale);
1993 tp->snd_una++; /* SYN is acked */
1995 * If not all the data that was sent in the TFO SYN
1996 * has been acked, resend the remainder right away.
1998 if (IS_FASTOPEN(tp->t_flags) &&
1999 (tp->snd_una != tp->snd_max)) {
2000 tp->snd_nxt = th->th_ack;
2001 tfo_partial_ack = 1;
2004 * If there's data, delay ACK; if there's also a FIN
2005 * ACKNOW will be turned on later.
2007 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2008 tcp_timer_activate(tp, TT_DELACK,
2011 tp->t_flags |= TF_ACKNOW;
2013 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2015 tp->t_flags |= TF_ECN_PERMIT;
2016 TCPSTAT_INC(tcps_ecn_shs);
2020 * Received <SYN,ACK> in SYN_SENT[*] state.
2022 * SYN_SENT --> ESTABLISHED
2023 * SYN_SENT* --> FIN_WAIT_1
2025 tp->t_starttime = ticks;
2026 if (tp->t_flags & TF_NEEDFIN) {
2027 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2028 tp->t_flags &= ~TF_NEEDFIN;
2031 tcp_state_change(tp, TCPS_ESTABLISHED);
2032 TCP_PROBE5(connect__established, NULL, tp,
2035 tcp_timer_activate(tp, TT_KEEP,
2040 * Received initial SYN in SYN-SENT[*] state =>
2041 * simultaneous open.
2042 * If it succeeds, connection is * half-synchronized.
2043 * Otherwise, do 3-way handshake:
2044 * SYN-SENT -> SYN-RECEIVED
2045 * SYN-SENT* -> SYN-RECEIVED*
2047 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2048 tcp_timer_activate(tp, TT_REXMT, 0);
2049 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2052 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2053 INP_WLOCK_ASSERT(tp->t_inpcb);
2056 * Advance th->th_seq to correspond to first data byte.
2057 * If data, trim to stay within window,
2058 * dropping FIN if necessary.
2061 if (tlen > tp->rcv_wnd) {
2062 todrop = tlen - tp->rcv_wnd;
2066 TCPSTAT_INC(tcps_rcvpackafterwin);
2067 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2069 tp->snd_wl1 = th->th_seq - 1;
2070 tp->rcv_up = th->th_seq;
2072 * Client side of transaction: already sent SYN and data.
2073 * If the remote host used T/TCP to validate the SYN,
2074 * our data will be ACK'd; if so, enter normal data segment
2075 * processing in the middle of step 5, ack processing.
2076 * Otherwise, goto step 6.
2078 if (thflags & TH_ACK)
2084 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2085 * do normal processing.
2087 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2091 break; /* continue normal processing */
2095 * States other than LISTEN or SYN_SENT.
2096 * First check the RST flag and sequence number since reset segments
2097 * are exempt from the timestamp and connection count tests. This
2098 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2099 * below which allowed reset segments in half the sequence space
2100 * to fall though and be processed (which gives forged reset
2101 * segments with a random sequence number a 50 percent chance of
2102 * killing a connection).
2103 * Then check timestamp, if present.
2104 * Then check the connection count, if present.
2105 * Then check that at least some bytes of segment are within
2106 * receive window. If segment begins before rcv_nxt,
2107 * drop leading data (and SYN); if nothing left, just ack.
2109 if (thflags & TH_RST) {
2111 * RFC5961 Section 3.2
2113 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2114 * - If RST is in window, we send challenge ACK.
2116 * Note: to take into account delayed ACKs, we should
2117 * test against last_ack_sent instead of rcv_nxt.
2118 * Note 2: we handle special case of closed window, not
2119 * covered by the RFC.
2121 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2122 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2123 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2125 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2126 KASSERT(tp->t_state != TCPS_SYN_SENT,
2127 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2130 if (V_tcp_insecure_rst ||
2131 tp->last_ack_sent == th->th_seq) {
2132 TCPSTAT_INC(tcps_drops);
2133 /* Drop the connection. */
2134 switch (tp->t_state) {
2135 case TCPS_SYN_RECEIVED:
2136 so->so_error = ECONNREFUSED;
2138 case TCPS_ESTABLISHED:
2139 case TCPS_FIN_WAIT_1:
2140 case TCPS_FIN_WAIT_2:
2141 case TCPS_CLOSE_WAIT:
2144 so->so_error = ECONNRESET;
2151 TCPSTAT_INC(tcps_badrst);
2152 /* Send challenge ACK. */
2153 tcp_respond(tp, mtod(m, void *), th, m,
2154 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2155 tp->last_ack_sent = tp->rcv_nxt;
2163 * RFC5961 Section 4.2
2164 * Send challenge ACK for any SYN in synchronized state.
2166 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2167 tp->t_state != TCPS_SYN_RECEIVED) {
2168 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2170 TCPSTAT_INC(tcps_badsyn);
2171 if (V_tcp_insecure_syn &&
2172 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2173 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2174 tp = tcp_drop(tp, ECONNRESET);
2175 rstreason = BANDLIM_UNLIMITED;
2177 /* Send challenge ACK. */
2178 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2179 tp->snd_nxt, TH_ACK);
2180 tp->last_ack_sent = tp->rcv_nxt;
2187 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2188 * and it's less than ts_recent, drop it.
2190 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2191 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2193 /* Check to see if ts_recent is over 24 days old. */
2194 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2196 * Invalidate ts_recent. If this segment updates
2197 * ts_recent, the age will be reset later and ts_recent
2198 * will get a valid value. If it does not, setting
2199 * ts_recent to zero will at least satisfy the
2200 * requirement that zero be placed in the timestamp
2201 * echo reply when ts_recent isn't valid. The
2202 * age isn't reset until we get a valid ts_recent
2203 * because we don't want out-of-order segments to be
2204 * dropped when ts_recent is old.
2208 TCPSTAT_INC(tcps_rcvduppack);
2209 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2210 TCPSTAT_INC(tcps_pawsdrop);
2218 * In the SYN-RECEIVED state, validate that the packet belongs to
2219 * this connection before trimming the data to fit the receive
2220 * window. Check the sequence number versus IRS since we know
2221 * the sequence numbers haven't wrapped. This is a partial fix
2222 * for the "LAND" DoS attack.
2224 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2225 rstreason = BANDLIM_RST_OPENPORT;
2229 todrop = tp->rcv_nxt - th->th_seq;
2231 if (thflags & TH_SYN) {
2241 * Following if statement from Stevens, vol. 2, p. 960.
2244 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2246 * Any valid FIN must be to the left of the window.
2247 * At this point the FIN must be a duplicate or out
2248 * of sequence; drop it.
2253 * Send an ACK to resynchronize and drop any data.
2254 * But keep on processing for RST or ACK.
2256 tp->t_flags |= TF_ACKNOW;
2258 TCPSTAT_INC(tcps_rcvduppack);
2259 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2261 TCPSTAT_INC(tcps_rcvpartduppack);
2262 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2264 drop_hdrlen += todrop; /* drop from the top afterwards */
2265 th->th_seq += todrop;
2267 if (th->th_urp > todrop)
2268 th->th_urp -= todrop;
2276 * If new data are received on a connection after the
2277 * user processes are gone, then RST the other end.
2279 if ((so->so_state & SS_NOFDREF) &&
2280 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2281 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2283 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2284 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2285 "after socket was closed, "
2286 "sending RST and removing tcpcb\n",
2287 s, __func__, tcpstates[tp->t_state], tlen);
2291 TCPSTAT_INC(tcps_rcvafterclose);
2292 rstreason = BANDLIM_UNLIMITED;
2297 * If segment ends after window, drop trailing data
2298 * (and PUSH and FIN); if nothing left, just ACK.
2300 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2302 TCPSTAT_INC(tcps_rcvpackafterwin);
2303 if (todrop >= tlen) {
2304 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2306 * If window is closed can only take segments at
2307 * window edge, and have to drop data and PUSH from
2308 * incoming segments. Continue processing, but
2309 * remember to ack. Otherwise, drop segment
2312 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2313 tp->t_flags |= TF_ACKNOW;
2314 TCPSTAT_INC(tcps_rcvwinprobe);
2318 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2321 thflags &= ~(TH_PUSH|TH_FIN);
2325 * If last ACK falls within this segment's sequence numbers,
2326 * record its timestamp.
2328 * 1) That the test incorporates suggestions from the latest
2329 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2330 * 2) That updating only on newer timestamps interferes with
2331 * our earlier PAWS tests, so this check should be solely
2332 * predicated on the sequence space of this segment.
2333 * 3) That we modify the segment boundary check to be
2334 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2335 * instead of RFC1323's
2336 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2337 * This modified check allows us to overcome RFC1323's
2338 * limitations as described in Stevens TCP/IP Illustrated
2339 * Vol. 2 p.869. In such cases, we can still calculate the
2340 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2342 if ((to.to_flags & TOF_TS) != 0 &&
2343 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2344 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2345 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2346 tp->ts_recent_age = tcp_ts_getticks();
2347 tp->ts_recent = to.to_tsval;
2351 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2352 * flag is on (half-synchronized state), then queue data for
2353 * later processing; else drop segment and return.
2355 if ((thflags & TH_ACK) == 0) {
2356 if (tp->t_state == TCPS_SYN_RECEIVED ||
2357 (tp->t_flags & TF_NEEDSYN)) {
2358 if (tp->t_state == TCPS_SYN_RECEIVED &&
2359 IS_FASTOPEN(tp->t_flags)) {
2360 tp->snd_wnd = tiwin;
2364 } else if (tp->t_flags & TF_ACKNOW)
2373 switch (tp->t_state) {
2376 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2377 * ESTABLISHED state and continue processing.
2378 * The ACK was checked above.
2380 case TCPS_SYN_RECEIVED:
2382 TCPSTAT_INC(tcps_connects);
2384 /* Do window scaling? */
2385 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2386 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2387 tp->rcv_scale = tp->request_r_scale;
2389 tp->snd_wnd = tiwin;
2392 * SYN-RECEIVED -> ESTABLISHED
2393 * SYN-RECEIVED* -> FIN-WAIT-1
2395 tp->t_starttime = ticks;
2396 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2397 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2398 tp->t_tfo_pending = NULL;
2401 * Account for the ACK of our SYN prior to
2402 * regular ACK processing below.
2406 if (tp->t_flags & TF_NEEDFIN) {
2407 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2408 tp->t_flags &= ~TF_NEEDFIN;
2410 tcp_state_change(tp, TCPS_ESTABLISHED);
2411 TCP_PROBE5(accept__established, NULL, tp,
2414 * TFO connections call cc_conn_init() during SYN
2415 * processing. Calling it again here for such
2416 * connections is not harmless as it would undo the
2417 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2420 if (!IS_FASTOPEN(tp->t_flags))
2422 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2425 * If segment contains data or ACK, will call tcp_reass()
2426 * later; if not, do so now to pass queued data to user.
2428 if (tlen == 0 && (thflags & TH_FIN) == 0)
2429 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2431 tp->snd_wl1 = th->th_seq - 1;
2435 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2436 * ACKs. If the ack is in the range
2437 * tp->snd_una < th->th_ack <= tp->snd_max
2438 * then advance tp->snd_una to th->th_ack and drop
2439 * data from the retransmission queue. If this ACK reflects
2440 * more up to date window information we update our window information.
2442 case TCPS_ESTABLISHED:
2443 case TCPS_FIN_WAIT_1:
2444 case TCPS_FIN_WAIT_2:
2445 case TCPS_CLOSE_WAIT:
2448 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2449 TCPSTAT_INC(tcps_rcvacktoomuch);
2452 if ((tp->t_flags & TF_SACK_PERMIT) &&
2453 ((to.to_flags & TOF_SACK) ||
2454 !TAILQ_EMPTY(&tp->snd_holes)))
2455 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2458 * Reset the value so that previous (valid) value
2459 * from the last ack with SACK doesn't get used.
2461 tp->sackhint.sacked_bytes = 0;
2464 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2465 hhook_run_tcp_est_in(tp, th, &to);
2468 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2471 maxseg = tcp_maxseg(tp);
2473 (tiwin == tp->snd_wnd ||
2474 (tp->t_flags & TF_SACK_PERMIT))) {
2476 * If this is the first time we've seen a
2477 * FIN from the remote, this is not a
2478 * duplicate and it needs to be processed
2479 * normally. This happens during a
2480 * simultaneous close.
2482 if ((thflags & TH_FIN) &&
2483 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2487 TCPSTAT_INC(tcps_rcvdupack);
2489 * If we have outstanding data (other than
2490 * a window probe), this is a completely
2491 * duplicate ack (ie, window info didn't
2492 * change and FIN isn't set),
2493 * the ack is the biggest we've
2494 * seen and we've seen exactly our rexmt
2495 * threshold of them, assume a packet
2496 * has been dropped and retransmit it.
2497 * Kludge snd_nxt & the congestion
2498 * window so we send only this one
2501 * We know we're losing at the current
2502 * window size so do congestion avoidance
2503 * (set ssthresh to half the current window
2504 * and pull our congestion window back to
2505 * the new ssthresh).
2507 * Dup acks mean that packets have left the
2508 * network (they're now cached at the receiver)
2509 * so bump cwnd by the amount in the receiver
2510 * to keep a constant cwnd packets in the
2513 * When using TCP ECN, notify the peer that
2514 * we reduced the cwnd.
2517 * Following 2 kinds of acks should not affect
2520 * 2) Acks with SACK but without any new SACK
2521 * information in them. These could result from
2522 * any anomaly in the network like a switch
2523 * duplicating packets or a possible DoS attack.
2525 if (th->th_ack != tp->snd_una ||
2526 ((tp->t_flags & TF_SACK_PERMIT) &&
2529 else if (!tcp_timer_active(tp, TT_REXMT))
2531 else if (++tp->t_dupacks > tcprexmtthresh ||
2532 IN_FASTRECOVERY(tp->t_flags)) {
2533 cc_ack_received(tp, th, nsegs,
2535 if ((tp->t_flags & TF_SACK_PERMIT) &&
2536 IN_FASTRECOVERY(tp->t_flags)) {
2540 * Compute the amount of data in flight first.
2541 * We can inject new data into the pipe iff
2542 * we have less than 1/2 the original window's
2543 * worth of data in flight.
2545 if (V_tcp_do_rfc6675_pipe)
2546 awnd = tcp_compute_pipe(tp);
2548 awnd = (tp->snd_nxt - tp->snd_fack) +
2549 tp->sackhint.sack_bytes_rexmit;
2551 if (awnd < tp->snd_ssthresh) {
2552 tp->snd_cwnd += maxseg;
2553 if (tp->snd_cwnd > tp->snd_ssthresh)
2554 tp->snd_cwnd = tp->snd_ssthresh;
2557 tp->snd_cwnd += maxseg;
2558 (void) tp->t_fb->tfb_tcp_output(tp);
2560 } else if (tp->t_dupacks == tcprexmtthresh) {
2561 tcp_seq onxt = tp->snd_nxt;
2564 * If we're doing sack, check to
2565 * see if we're already in sack
2566 * recovery. If we're not doing sack,
2567 * check to see if we're in newreno
2570 if (tp->t_flags & TF_SACK_PERMIT) {
2571 if (IN_FASTRECOVERY(tp->t_flags)) {
2576 if (SEQ_LEQ(th->th_ack,
2582 /* Congestion signal before ack. */
2583 cc_cong_signal(tp, th, CC_NDUPACK);
2584 cc_ack_received(tp, th, nsegs,
2586 tcp_timer_activate(tp, TT_REXMT, 0);
2588 if (tp->t_flags & TF_SACK_PERMIT) {
2590 tcps_sack_recovery_episode);
2591 tp->sack_newdata = tp->snd_nxt;
2592 tp->snd_cwnd = maxseg;
2593 (void) tp->t_fb->tfb_tcp_output(tp);
2596 tp->snd_nxt = th->th_ack;
2597 tp->snd_cwnd = maxseg;
2598 (void) tp->t_fb->tfb_tcp_output(tp);
2599 KASSERT(tp->snd_limited <= 2,
2600 ("%s: tp->snd_limited too big",
2602 tp->snd_cwnd = tp->snd_ssthresh +
2604 (tp->t_dupacks - tp->snd_limited);
2605 if (SEQ_GT(onxt, tp->snd_nxt))
2608 } else if (V_tcp_do_rfc3042) {
2610 * Process first and second duplicate
2611 * ACKs. Each indicates a segment
2612 * leaving the network, creating room
2613 * for more. Make sure we can send a
2614 * packet on reception of each duplicate
2615 * ACK by increasing snd_cwnd by one
2616 * segment. Restore the original
2617 * snd_cwnd after packet transmission.
2619 cc_ack_received(tp, th, nsegs,
2621 uint32_t oldcwnd = tp->snd_cwnd;
2622 tcp_seq oldsndmax = tp->snd_max;
2626 KASSERT(tp->t_dupacks == 1 ||
2628 ("%s: dupacks not 1 or 2",
2630 if (tp->t_dupacks == 1)
2631 tp->snd_limited = 0;
2633 (tp->snd_nxt - tp->snd_una) +
2634 (tp->t_dupacks - tp->snd_limited) *
2637 * Only call tcp_output when there
2638 * is new data available to be sent.
2639 * Otherwise we would send pure ACKs.
2641 SOCKBUF_LOCK(&so->so_snd);
2642 avail = sbavail(&so->so_snd) -
2643 (tp->snd_nxt - tp->snd_una);
2644 SOCKBUF_UNLOCK(&so->so_snd);
2646 (void) tp->t_fb->tfb_tcp_output(tp);
2647 sent = tp->snd_max - oldsndmax;
2648 if (sent > maxseg) {
2649 KASSERT((tp->t_dupacks == 2 &&
2650 tp->snd_limited == 0) ||
2651 (sent == maxseg + 1 &&
2652 tp->t_flags & TF_SENTFIN),
2653 ("%s: sent too much",
2655 tp->snd_limited = 2;
2656 } else if (sent > 0)
2658 tp->snd_cwnd = oldcwnd;
2665 * This ack is advancing the left edge, reset the
2670 * If this ack also has new SACK info, increment the
2671 * counter as per rfc6675.
2673 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed)
2677 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2678 ("%s: th_ack <= snd_una", __func__));
2681 * If the congestion window was inflated to account
2682 * for the other side's cached packets, retract it.
2684 if (IN_FASTRECOVERY(tp->t_flags)) {
2685 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2686 if (tp->t_flags & TF_SACK_PERMIT)
2687 tcp_sack_partialack(tp, th);
2689 tcp_newreno_partial_ack(tp, th);
2691 cc_post_recovery(tp, th);
2694 * If we reach this point, ACK is not a duplicate,
2695 * i.e., it ACKs something we sent.
2697 if (tp->t_flags & TF_NEEDSYN) {
2699 * T/TCP: Connection was half-synchronized, and our
2700 * SYN has been ACK'd (so connection is now fully
2701 * synchronized). Go to non-starred state,
2702 * increment snd_una for ACK of SYN, and check if
2703 * we can do window scaling.
2705 tp->t_flags &= ~TF_NEEDSYN;
2707 /* Do window scaling? */
2708 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2709 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2710 tp->rcv_scale = tp->request_r_scale;
2711 /* Send window already scaled. */
2716 INP_WLOCK_ASSERT(tp->t_inpcb);
2718 acked = BYTES_THIS_ACK(tp, th);
2719 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2720 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2721 tp->snd_una, th->th_ack, tp, m));
2722 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2723 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2726 * If we just performed our first retransmit, and the ACK
2727 * arrives within our recovery window, then it was a mistake
2728 * to do the retransmit in the first place. Recover our
2729 * original cwnd and ssthresh, and proceed to transmit where
2732 if (tp->t_rxtshift == 1 &&
2733 tp->t_flags & TF_PREVVALID &&
2735 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2736 cc_cong_signal(tp, th, CC_RTO_ERR);
2739 * If we have a timestamp reply, update smoothed
2740 * round trip time. If no timestamp is present but
2741 * transmit timer is running and timed sequence
2742 * number was acked, update smoothed round trip time.
2743 * Since we now have an rtt measurement, cancel the
2744 * timer backoff (cf., Phil Karn's retransmit alg.).
2745 * Recompute the initial retransmit timer.
2747 * Some boxes send broken timestamp replies
2748 * during the SYN+ACK phase, ignore
2749 * timestamps of 0 or we could calculate a
2750 * huge RTT and blow up the retransmit timer.
2752 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2755 t = tcp_ts_getticks() - to.to_tsecr;
2756 if (!tp->t_rttlow || tp->t_rttlow > t)
2758 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2759 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2760 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2761 tp->t_rttlow = ticks - tp->t_rtttime;
2762 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2766 * If all outstanding data is acked, stop retransmit
2767 * timer and remember to restart (more output or persist).
2768 * If there is more data to be acked, restart retransmit
2769 * timer, using current (possibly backed-off) value.
2771 if (th->th_ack == tp->snd_max) {
2772 tcp_timer_activate(tp, TT_REXMT, 0);
2774 } else if (!tcp_timer_active(tp, TT_PERSIST))
2775 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2778 * If no data (only SYN) was ACK'd,
2779 * skip rest of ACK processing.
2785 * Let the congestion control algorithm update congestion
2786 * control related information. This typically means increasing
2787 * the congestion window.
2789 cc_ack_received(tp, th, nsegs, CC_ACK);
2791 SOCKBUF_LOCK(&so->so_snd);
2792 if (acked > sbavail(&so->so_snd)) {
2793 if (tp->snd_wnd >= sbavail(&so->so_snd))
2794 tp->snd_wnd -= sbavail(&so->so_snd);
2797 mfree = sbcut_locked(&so->so_snd,
2798 (int)sbavail(&so->so_snd));
2801 mfree = sbcut_locked(&so->so_snd, acked);
2802 if (tp->snd_wnd >= (uint32_t) acked)
2803 tp->snd_wnd -= acked;
2808 /* NB: sowwakeup_locked() does an implicit unlock. */
2809 sowwakeup_locked(so);
2811 /* Detect una wraparound. */
2812 if (!IN_RECOVERY(tp->t_flags) &&
2813 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2814 SEQ_LEQ(th->th_ack, tp->snd_recover))
2815 tp->snd_recover = th->th_ack - 1;
2816 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2817 if (IN_RECOVERY(tp->t_flags) &&
2818 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2819 EXIT_RECOVERY(tp->t_flags);
2821 tp->snd_una = th->th_ack;
2822 if (tp->t_flags & TF_SACK_PERMIT) {
2823 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2824 tp->snd_recover = tp->snd_una;
2826 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2827 tp->snd_nxt = tp->snd_una;
2829 switch (tp->t_state) {
2832 * In FIN_WAIT_1 STATE in addition to the processing
2833 * for the ESTABLISHED state if our FIN is now acknowledged
2834 * then enter FIN_WAIT_2.
2836 case TCPS_FIN_WAIT_1:
2837 if (ourfinisacked) {
2839 * If we can't receive any more
2840 * data, then closing user can proceed.
2841 * Starting the timer is contrary to the
2842 * specification, but if we don't get a FIN
2843 * we'll hang forever.
2846 * we should release the tp also, and use a
2849 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2850 soisdisconnected(so);
2851 tcp_timer_activate(tp, TT_2MSL,
2852 (tcp_fast_finwait2_recycle ?
2853 tcp_finwait2_timeout :
2856 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2861 * In CLOSING STATE in addition to the processing for
2862 * the ESTABLISHED state if the ACK acknowledges our FIN
2863 * then enter the TIME-WAIT state, otherwise ignore
2867 if (ourfinisacked) {
2868 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2876 * In LAST_ACK, we may still be waiting for data to drain
2877 * and/or to be acked, as well as for the ack of our FIN.
2878 * If our FIN is now acknowledged, delete the TCB,
2879 * enter the closed state and return.
2882 if (ourfinisacked) {
2883 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2892 INP_WLOCK_ASSERT(tp->t_inpcb);
2895 * Update window information.
2896 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2898 if ((thflags & TH_ACK) &&
2899 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2900 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2901 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2902 /* keep track of pure window updates */
2904 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2905 TCPSTAT_INC(tcps_rcvwinupd);
2906 tp->snd_wnd = tiwin;
2907 tp->snd_wl1 = th->th_seq;
2908 tp->snd_wl2 = th->th_ack;
2909 if (tp->snd_wnd > tp->max_sndwnd)
2910 tp->max_sndwnd = tp->snd_wnd;
2915 * Process segments with URG.
2917 if ((thflags & TH_URG) && th->th_urp &&
2918 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2920 * This is a kludge, but if we receive and accept
2921 * random urgent pointers, we'll crash in
2922 * soreceive. It's hard to imagine someone
2923 * actually wanting to send this much urgent data.
2925 SOCKBUF_LOCK(&so->so_rcv);
2926 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2927 th->th_urp = 0; /* XXX */
2928 thflags &= ~TH_URG; /* XXX */
2929 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2930 goto dodata; /* XXX */
2933 * If this segment advances the known urgent pointer,
2934 * then mark the data stream. This should not happen
2935 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2936 * a FIN has been received from the remote side.
2937 * In these states we ignore the URG.
2939 * According to RFC961 (Assigned Protocols),
2940 * the urgent pointer points to the last octet
2941 * of urgent data. We continue, however,
2942 * to consider it to indicate the first octet
2943 * of data past the urgent section as the original
2944 * spec states (in one of two places).
2946 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2947 tp->rcv_up = th->th_seq + th->th_urp;
2948 so->so_oobmark = sbavail(&so->so_rcv) +
2949 (tp->rcv_up - tp->rcv_nxt) - 1;
2950 if (so->so_oobmark == 0)
2951 so->so_rcv.sb_state |= SBS_RCVATMARK;
2953 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2955 SOCKBUF_UNLOCK(&so->so_rcv);
2957 * Remove out of band data so doesn't get presented to user.
2958 * This can happen independent of advancing the URG pointer,
2959 * but if two URG's are pending at once, some out-of-band
2960 * data may creep in... ick.
2962 if (th->th_urp <= (uint32_t)tlen &&
2963 !(so->so_options & SO_OOBINLINE)) {
2964 /* hdr drop is delayed */
2965 tcp_pulloutofband(so, th, m, drop_hdrlen);
2969 * If no out of band data is expected,
2970 * pull receive urgent pointer along
2971 * with the receive window.
2973 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2974 tp->rcv_up = tp->rcv_nxt;
2977 INP_WLOCK_ASSERT(tp->t_inpcb);
2980 * Process the segment text, merging it into the TCP sequencing queue,
2981 * and arranging for acknowledgment of receipt if necessary.
2982 * This process logically involves adjusting tp->rcv_wnd as data
2983 * is presented to the user (this happens in tcp_usrreq.c,
2984 * case PRU_RCVD). If a FIN has already been received on this
2985 * connection then we just ignore the text.
2987 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
2988 IS_FASTOPEN(tp->t_flags));
2989 if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
2990 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2991 tcp_seq save_start = th->th_seq;
2992 m_adj(m, drop_hdrlen); /* delayed header drop */
2994 * Insert segment which includes th into TCP reassembly queue
2995 * with control block tp. Set thflags to whether reassembly now
2996 * includes a segment with FIN. This handles the common case
2997 * inline (segment is the next to be received on an established
2998 * connection, and the queue is empty), avoiding linkage into
2999 * and removal from the queue and repetition of various
3001 * Set DELACK for segments received in order, but ack
3002 * immediately when segments are out of order (so
3003 * fast retransmit can work).
3005 if (th->th_seq == tp->rcv_nxt &&
3007 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3009 if (DELAY_ACK(tp, tlen) || tfo_syn)
3010 tp->t_flags |= TF_DELACK;
3012 tp->t_flags |= TF_ACKNOW;
3013 tp->rcv_nxt += tlen;
3014 thflags = th->th_flags & TH_FIN;
3015 TCPSTAT_INC(tcps_rcvpack);
3016 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3017 SOCKBUF_LOCK(&so->so_rcv);
3018 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3021 sbappendstream_locked(&so->so_rcv, m, 0);
3022 /* NB: sorwakeup_locked() does an implicit unlock. */
3023 sorwakeup_locked(so);
3026 * XXX: Due to the header drop above "th" is
3027 * theoretically invalid by now. Fortunately
3028 * m_adj() doesn't actually frees any mbufs
3029 * when trimming from the head.
3031 thflags = tcp_reass(tp, th, &save_start, &tlen, m);
3032 tp->t_flags |= TF_ACKNOW;
3034 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
3035 tcp_update_sack_list(tp, save_start, save_start + tlen);
3038 * Note the amount of data that peer has sent into
3039 * our window, in order to estimate the sender's
3043 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3044 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3046 len = so->so_rcv.sb_hiwat;
3054 * If FIN is received ACK the FIN and let the user know
3055 * that the connection is closing.
3057 if (thflags & TH_FIN) {
3058 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3061 * If connection is half-synchronized
3062 * (ie NEEDSYN flag on) then delay ACK,
3063 * so it may be piggybacked when SYN is sent.
3064 * Otherwise, since we received a FIN then no
3065 * more input can be expected, send ACK now.
3067 if (tp->t_flags & TF_NEEDSYN)
3068 tp->t_flags |= TF_DELACK;
3070 tp->t_flags |= TF_ACKNOW;
3073 switch (tp->t_state) {
3076 * In SYN_RECEIVED and ESTABLISHED STATES
3077 * enter the CLOSE_WAIT state.
3079 case TCPS_SYN_RECEIVED:
3080 tp->t_starttime = ticks;
3082 case TCPS_ESTABLISHED:
3083 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3087 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3088 * enter the CLOSING state.
3090 case TCPS_FIN_WAIT_1:
3091 tcp_state_change(tp, TCPS_CLOSING);
3095 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3096 * starting the time-wait timer, turning off the other
3099 case TCPS_FIN_WAIT_2:
3100 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
3107 if (so->so_options & SO_DEBUG)
3108 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3111 TCP_PROBE3(debug__input, tp, th, m);
3114 * Return any desired output.
3116 if (needoutput || (tp->t_flags & TF_ACKNOW))
3117 (void) tp->t_fb->tfb_tcp_output(tp);
3120 INP_WLOCK_ASSERT(tp->t_inpcb);
3122 if (tp->t_flags & TF_DELACK) {
3123 tp->t_flags &= ~TF_DELACK;
3124 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3126 INP_WUNLOCK(tp->t_inpcb);
3131 * Generate an ACK dropping incoming segment if it occupies
3132 * sequence space, where the ACK reflects our state.
3134 * We can now skip the test for the RST flag since all
3135 * paths to this code happen after packets containing
3136 * RST have been dropped.
3138 * In the SYN-RECEIVED state, don't send an ACK unless the
3139 * segment we received passes the SYN-RECEIVED ACK test.
3140 * If it fails send a RST. This breaks the loop in the
3141 * "LAND" DoS attack, and also prevents an ACK storm
3142 * between two listening ports that have been sent forged
3143 * SYN segments, each with the source address of the other.
3145 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3146 (SEQ_GT(tp->snd_una, th->th_ack) ||
3147 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3148 rstreason = BANDLIM_RST_OPENPORT;
3152 if (so->so_options & SO_DEBUG)
3153 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3156 TCP_PROBE3(debug__input, tp, th, m);
3157 tp->t_flags |= TF_ACKNOW;
3158 (void) tp->t_fb->tfb_tcp_output(tp);
3159 INP_WUNLOCK(tp->t_inpcb);
3165 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3166 INP_WUNLOCK(tp->t_inpcb);
3168 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3173 * Drop space held by incoming segment and return.
3176 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3177 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3180 TCP_PROBE3(debug__input, tp, th, m);
3182 INP_WUNLOCK(tp->t_inpcb);
3187 * Issue RST and make ACK acceptable to originator of segment.
3188 * The mbuf must still include the original packet header.
3192 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3193 int tlen, int rstreason)
3199 struct ip6_hdr *ip6;
3203 INP_WLOCK_ASSERT(tp->t_inpcb);
3206 /* Don't bother if destination was broadcast/multicast. */
3207 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3210 if (mtod(m, struct ip *)->ip_v == 6) {
3211 ip6 = mtod(m, struct ip6_hdr *);
3212 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3213 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3215 /* IPv6 anycast check is done at tcp6_input() */
3218 #if defined(INET) && defined(INET6)
3223 ip = mtod(m, struct ip *);
3224 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3225 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3226 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3227 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3232 /* Perform bandwidth limiting. */
3233 if (badport_bandlim(rstreason) < 0)
3236 /* tcp_respond consumes the mbuf chain. */
3237 if (th->th_flags & TH_ACK) {
3238 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3239 th->th_ack, TH_RST);
3241 if (th->th_flags & TH_SYN)
3243 if (th->th_flags & TH_FIN)
3245 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3246 (tcp_seq)0, TH_RST|TH_ACK);
3254 * Parse TCP options and place in tcpopt.
3257 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3262 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3264 if (opt == TCPOPT_EOL)
3266 if (opt == TCPOPT_NOP)
3272 if (optlen < 2 || optlen > cnt)
3277 if (optlen != TCPOLEN_MAXSEG)
3279 if (!(flags & TO_SYN))
3281 to->to_flags |= TOF_MSS;
3282 bcopy((char *)cp + 2,
3283 (char *)&to->to_mss, sizeof(to->to_mss));
3284 to->to_mss = ntohs(to->to_mss);
3287 if (optlen != TCPOLEN_WINDOW)
3289 if (!(flags & TO_SYN))
3291 to->to_flags |= TOF_SCALE;
3292 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3294 case TCPOPT_TIMESTAMP:
3295 if (optlen != TCPOLEN_TIMESTAMP)
3297 to->to_flags |= TOF_TS;
3298 bcopy((char *)cp + 2,
3299 (char *)&to->to_tsval, sizeof(to->to_tsval));
3300 to->to_tsval = ntohl(to->to_tsval);
3301 bcopy((char *)cp + 6,
3302 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3303 to->to_tsecr = ntohl(to->to_tsecr);
3305 case TCPOPT_SIGNATURE:
3307 * In order to reply to a host which has set the
3308 * TCP_SIGNATURE option in its initial SYN, we have
3309 * to record the fact that the option was observed
3310 * here for the syncache code to perform the correct
3313 if (optlen != TCPOLEN_SIGNATURE)
3315 to->to_flags |= TOF_SIGNATURE;
3316 to->to_signature = cp + 2;
3318 case TCPOPT_SACK_PERMITTED:
3319 if (optlen != TCPOLEN_SACK_PERMITTED)
3321 if (!(flags & TO_SYN))
3325 to->to_flags |= TOF_SACKPERM;
3328 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3332 to->to_flags |= TOF_SACK;
3333 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3334 to->to_sacks = cp + 2;
3335 TCPSTAT_INC(tcps_sack_rcv_blocks);
3337 case TCPOPT_FAST_OPEN:
3339 * Cookie length validation is performed by the
3340 * server side cookie checking code or the client
3341 * side cookie cache update code.
3343 if (!(flags & TO_SYN))
3345 if (!V_tcp_fastopen_client_enable &&
3346 !V_tcp_fastopen_server_enable)
3348 to->to_flags |= TOF_FASTOPEN;
3349 to->to_tfo_len = optlen - 2;
3350 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3359 * Pull out of band byte out of a segment so
3360 * it doesn't appear in the user's data queue.
3361 * It is still reflected in the segment length for
3362 * sequencing purposes.
3365 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3368 int cnt = off + th->th_urp - 1;
3371 if (m->m_len > cnt) {
3372 char *cp = mtod(m, caddr_t) + cnt;
3373 struct tcpcb *tp = sototcpcb(so);
3375 INP_WLOCK_ASSERT(tp->t_inpcb);
3378 tp->t_oobflags |= TCPOOB_HAVEDATA;
3379 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3381 if (m->m_flags & M_PKTHDR)
3390 panic("tcp_pulloutofband");
3394 * Collect new round-trip time estimate
3395 * and update averages and current timeout.
3398 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3402 INP_WLOCK_ASSERT(tp->t_inpcb);
3404 TCPSTAT_INC(tcps_rttupdated);
3406 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3408 * srtt is stored as fixed point with 5 bits after the
3409 * binary point (i.e., scaled by 8). The following magic
3410 * is equivalent to the smoothing algorithm in rfc793 with
3411 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3412 * point). Adjust rtt to origin 0.
3414 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3415 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3417 if ((tp->t_srtt += delta) <= 0)
3421 * We accumulate a smoothed rtt variance (actually, a
3422 * smoothed mean difference), then set the retransmit
3423 * timer to smoothed rtt + 4 times the smoothed variance.
3424 * rttvar is stored as fixed point with 4 bits after the
3425 * binary point (scaled by 16). The following is
3426 * equivalent to rfc793 smoothing with an alpha of .75
3427 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3428 * rfc793's wired-in beta.
3432 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3433 if ((tp->t_rttvar += delta) <= 0)
3435 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3436 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3439 * No rtt measurement yet - use the unsmoothed rtt.
3440 * Set the variance to half the rtt (so our first
3441 * retransmit happens at 3*rtt).
3443 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3444 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3445 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3451 * the retransmit should happen at rtt + 4 * rttvar.
3452 * Because of the way we do the smoothing, srtt and rttvar
3453 * will each average +1/2 tick of bias. When we compute
3454 * the retransmit timer, we want 1/2 tick of rounding and
3455 * 1 extra tick because of +-1/2 tick uncertainty in the
3456 * firing of the timer. The bias will give us exactly the
3457 * 1.5 tick we need. But, because the bias is
3458 * statistical, we have to test that we don't drop below
3459 * the minimum feasible timer (which is 2 ticks).
3461 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3462 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3465 * We received an ack for a packet that wasn't retransmitted;
3466 * it is probably safe to discard any error indications we've
3467 * received recently. This isn't quite right, but close enough
3468 * for now (a route might have failed after we sent a segment,
3469 * and the return path might not be symmetrical).
3471 tp->t_softerror = 0;
3475 * Determine a reasonable value for maxseg size.
3476 * If the route is known, check route for mtu.
3477 * If none, use an mss that can be handled on the outgoing interface
3478 * without forcing IP to fragment. If no route is found, route has no mtu,
3479 * or the destination isn't local, use a default, hopefully conservative
3480 * size (usually 512 or the default IP max size, but no more than the mtu
3481 * of the interface), as we can't discover anything about intervening
3482 * gateways or networks. We also initialize the congestion/slow start
3483 * window to be a single segment if the destination isn't local.
3484 * While looking at the routing entry, we also initialize other path-dependent
3485 * parameters from pre-set or cached values in the routing entry.
3487 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3488 * IP options, e.g. IPSEC data, since length of this data may vary, and
3489 * thus it is calculated for every segment separately in tcp_output().
3491 * NOTE that this routine is only called when we process an incoming
3492 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3493 * settings are handled in tcp_mssopt().
3496 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3497 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3500 uint32_t maxmtu = 0;
3501 struct inpcb *inp = tp->t_inpcb;
3502 struct hc_metrics_lite metrics;
3504 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3505 size_t min_protoh = isipv6 ?
3506 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3507 sizeof (struct tcpiphdr);
3509 const size_t min_protoh = sizeof(struct tcpiphdr);
3512 INP_WLOCK_ASSERT(tp->t_inpcb);
3514 if (mtuoffer != -1) {
3515 KASSERT(offer == -1, ("%s: conflict", __func__));
3516 offer = mtuoffer - min_protoh;
3522 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3523 tp->t_maxseg = V_tcp_v6mssdflt;
3526 #if defined(INET) && defined(INET6)
3531 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3532 tp->t_maxseg = V_tcp_mssdflt;
3537 * No route to sender, stay with default mss and return.
3541 * In case we return early we need to initialize metrics
3542 * to a defined state as tcp_hc_get() would do for us
3543 * if there was no cache hit.
3545 if (metricptr != NULL)
3546 bzero(metricptr, sizeof(struct hc_metrics_lite));
3550 /* What have we got? */
3554 * Offer == 0 means that there was no MSS on the SYN
3555 * segment, in this case we use tcp_mssdflt as
3556 * already assigned to t_maxseg above.
3558 offer = tp->t_maxseg;
3563 * Offer == -1 means that we didn't receive SYN yet.
3569 * Prevent DoS attack with too small MSS. Round up
3570 * to at least minmss.
3572 offer = max(offer, V_tcp_minmss);
3576 * rmx information is now retrieved from tcp_hostcache.
3578 tcp_hc_get(&inp->inp_inc, &metrics);
3579 if (metricptr != NULL)
3580 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3583 * If there's a discovered mtu in tcp hostcache, use it.
3584 * Else, use the link mtu.
3586 if (metrics.rmx_mtu)
3587 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3591 mss = maxmtu - min_protoh;
3592 if (!V_path_mtu_discovery &&
3593 !in6_localaddr(&inp->in6p_faddr))
3594 mss = min(mss, V_tcp_v6mssdflt);
3597 #if defined(INET) && defined(INET6)
3602 mss = maxmtu - min_protoh;
3603 if (!V_path_mtu_discovery &&
3604 !in_localaddr(inp->inp_faddr))
3605 mss = min(mss, V_tcp_mssdflt);
3609 * XXX - The above conditional (mss = maxmtu - min_protoh)
3610 * probably violates the TCP spec.
3611 * The problem is that, since we don't know the
3612 * other end's MSS, we are supposed to use a conservative
3613 * default. But, if we do that, then MTU discovery will
3614 * never actually take place, because the conservative
3615 * default is much less than the MTUs typically seen
3616 * on the Internet today. For the moment, we'll sweep
3617 * this under the carpet.
3619 * The conservative default might not actually be a problem
3620 * if the only case this occurs is when sending an initial
3621 * SYN with options and data to a host we've never talked
3622 * to before. Then, they will reply with an MSS value which
3623 * will get recorded and the new parameters should get
3624 * recomputed. For Further Study.
3627 mss = min(mss, offer);
3630 * Sanity check: make sure that maxseg will be large
3631 * enough to allow some data on segments even if the
3632 * all the option space is used (40bytes). Otherwise
3633 * funny things may happen in tcp_output.
3635 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3643 tcp_mss(struct tcpcb *tp, int offer)
3649 struct hc_metrics_lite metrics;
3650 struct tcp_ifcap cap;
3652 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3654 bzero(&cap, sizeof(cap));
3655 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3661 * If there's a pipesize, change the socket buffer to that size,
3662 * don't change if sb_hiwat is different than default (then it
3663 * has been changed on purpose with setsockopt).
3664 * Make the socket buffers an integral number of mss units;
3665 * if the mss is larger than the socket buffer, decrease the mss.
3667 so = inp->inp_socket;
3668 SOCKBUF_LOCK(&so->so_snd);
3669 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3670 bufsize = metrics.rmx_sendpipe;
3672 bufsize = so->so_snd.sb_hiwat;
3676 bufsize = roundup(bufsize, mss);
3677 if (bufsize > sb_max)
3679 if (bufsize > so->so_snd.sb_hiwat)
3680 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3682 SOCKBUF_UNLOCK(&so->so_snd);
3684 * Sanity check: make sure that maxseg will be large
3685 * enough to allow some data on segments even if the
3686 * all the option space is used (40bytes). Otherwise
3687 * funny things may happen in tcp_output.
3689 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3691 tp->t_maxseg = max(mss, 64);
3693 SOCKBUF_LOCK(&so->so_rcv);
3694 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3695 bufsize = metrics.rmx_recvpipe;
3697 bufsize = so->so_rcv.sb_hiwat;
3698 if (bufsize > mss) {
3699 bufsize = roundup(bufsize, mss);
3700 if (bufsize > sb_max)
3702 if (bufsize > so->so_rcv.sb_hiwat)
3703 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3705 SOCKBUF_UNLOCK(&so->so_rcv);
3707 /* Check the interface for TSO capabilities. */
3708 if (cap.ifcap & CSUM_TSO) {
3709 tp->t_flags |= TF_TSO;
3710 tp->t_tsomax = cap.tsomax;
3711 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3712 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3717 * Determine the MSS option to send on an outgoing SYN.
3720 tcp_mssopt(struct in_conninfo *inc)
3723 uint32_t thcmtu = 0;
3724 uint32_t maxmtu = 0;
3727 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3730 if (inc->inc_flags & INC_ISIPV6) {
3731 mss = V_tcp_v6mssdflt;
3732 maxmtu = tcp_maxmtu6(inc, NULL);
3733 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3736 #if defined(INET) && defined(INET6)
3741 mss = V_tcp_mssdflt;
3742 maxmtu = tcp_maxmtu(inc, NULL);
3743 min_protoh = sizeof(struct tcpiphdr);
3746 #if defined(INET6) || defined(INET)
3747 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3750 if (maxmtu && thcmtu)
3751 mss = min(maxmtu, thcmtu) - min_protoh;
3752 else if (maxmtu || thcmtu)
3753 mss = max(maxmtu, thcmtu) - min_protoh;
3760 * On a partial ack arrives, force the retransmission of the
3761 * next unacknowledged segment. Do not clear tp->t_dupacks.
3762 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3766 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3768 tcp_seq onxt = tp->snd_nxt;
3769 uint32_t ocwnd = tp->snd_cwnd;
3770 u_int maxseg = tcp_maxseg(tp);
3772 INP_WLOCK_ASSERT(tp->t_inpcb);
3774 tcp_timer_activate(tp, TT_REXMT, 0);
3776 tp->snd_nxt = th->th_ack;
3778 * Set snd_cwnd to one segment beyond acknowledged offset.
3779 * (tp->snd_una has not yet been updated when this function is called.)
3781 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3782 tp->t_flags |= TF_ACKNOW;
3783 (void) tp->t_fb->tfb_tcp_output(tp);
3784 tp->snd_cwnd = ocwnd;
3785 if (SEQ_GT(onxt, tp->snd_nxt))
3788 * Partial window deflation. Relies on fact that tp->snd_una
3791 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3792 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3795 tp->snd_cwnd += maxseg;
3799 tcp_compute_pipe(struct tcpcb *tp)
3801 return (tp->snd_max - tp->snd_una +
3802 tp->sackhint.sack_bytes_rexmit -
3803 tp->sackhint.sacked_bytes);
3807 tcp_compute_initwnd(uint32_t maxseg)
3810 * Calculate the Initial Window, also used as Restart Window
3812 * RFC5681 Section 3.1 specifies the default conservative values.
3813 * RFC3390 specifies slightly more aggressive values.
3814 * RFC6928 increases it to ten segments.
3815 * Support for user specified value for initial flight size.
3817 if (V_tcp_initcwnd_segments)
3818 return min(V_tcp_initcwnd_segments * maxseg,
3819 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
3820 else if (V_tcp_do_rfc3390)
3821 return min(4 * maxseg, max(2 * maxseg, 4380));
3823 /* Per RFC5681 Section 3.1 */
3825 return (2 * maxseg);
3826 else if (maxseg > 1095)
3827 return (3 * maxseg);
3829 return (4 * maxseg);