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_max) = 2*1024*1024;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
217 &VNET_NAME(tcp_autorcvbuf_max), 0,
218 "Max size of automatic receive buffer");
220 VNET_DEFINE(struct inpcbhead, tcb);
221 #define tcb6 tcb /* for KAME src sync over BSD*'s */
222 VNET_DEFINE(struct inpcbinfo, tcbinfo);
225 * TCP statistics are stored in an array of counter(9)s, which size matches
226 * size of struct tcpstat. TCP running connection count is a regular array.
228 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
229 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
230 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
231 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
232 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
233 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
234 "TCP connection counts by TCP state");
237 tcp_vnet_init(const void *unused)
240 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
241 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
243 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
244 tcp_vnet_init, NULL);
248 tcp_vnet_uninit(const void *unused)
251 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
252 VNET_PCPUSTAT_FREE(tcpstat);
254 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
255 tcp_vnet_uninit, NULL);
259 * Kernel module interface for updating tcpstat. The argument is an index
260 * into tcpstat treated as an array.
263 kmod_tcpstat_inc(int statnum)
266 counter_u64_add(VNET(tcpstat)[statnum], 1);
271 * Wrapper for the TCP established input helper hook.
274 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
276 struct tcp_hhook_data hhook_data;
278 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
283 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
290 * CC wrapper hook functions
293 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
296 INP_WLOCK_ASSERT(tp->t_inpcb);
298 tp->ccv->nsegs = nsegs;
299 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
300 if (tp->snd_cwnd <= tp->snd_wnd)
301 tp->ccv->flags |= CCF_CWND_LIMITED;
303 tp->ccv->flags &= ~CCF_CWND_LIMITED;
305 if (type == CC_ACK) {
306 if (tp->snd_cwnd > tp->snd_ssthresh) {
307 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
308 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp));
309 if (tp->t_bytes_acked >= tp->snd_cwnd) {
310 tp->t_bytes_acked -= tp->snd_cwnd;
311 tp->ccv->flags |= CCF_ABC_SENTAWND;
314 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
315 tp->t_bytes_acked = 0;
319 if (CC_ALGO(tp)->ack_received != NULL) {
320 /* XXXLAS: Find a way to live without this */
321 tp->ccv->curack = th->th_ack;
322 CC_ALGO(tp)->ack_received(tp->ccv, type);
327 cc_conn_init(struct tcpcb *tp)
329 struct hc_metrics_lite metrics;
330 struct inpcb *inp = tp->t_inpcb;
334 INP_WLOCK_ASSERT(tp->t_inpcb);
336 tcp_hc_get(&inp->inp_inc, &metrics);
337 maxseg = tcp_maxseg(tp);
339 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
341 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
342 TCPSTAT_INC(tcps_usedrtt);
343 if (metrics.rmx_rttvar) {
344 tp->t_rttvar = metrics.rmx_rttvar;
345 TCPSTAT_INC(tcps_usedrttvar);
347 /* default variation is +- 1 rtt */
349 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
351 TCPT_RANGESET(tp->t_rxtcur,
352 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
353 tp->t_rttmin, TCPTV_REXMTMAX);
355 if (metrics.rmx_ssthresh) {
357 * There's some sort of gateway or interface
358 * buffer limit on the path. Use this to set
359 * the slow start threshold, but set the
360 * threshold to no less than 2*mss.
362 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
363 TCPSTAT_INC(tcps_usedssthresh);
367 * Set the initial slow-start flight size.
369 * If a SYN or SYN/ACK was lost and retransmitted, we have to
370 * reduce the initial CWND to one segment as congestion is likely
371 * requiring us to be cautious.
373 if (tp->snd_cwnd == 1)
374 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
376 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
378 if (CC_ALGO(tp)->conn_init != NULL)
379 CC_ALGO(tp)->conn_init(tp->ccv);
383 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
387 INP_WLOCK_ASSERT(tp->t_inpcb);
391 if (!IN_FASTRECOVERY(tp->t_flags)) {
392 tp->snd_recover = tp->snd_max;
393 if (tp->t_flags & TF_ECN_PERMIT)
394 tp->t_flags |= TF_ECN_SND_CWR;
398 if (!IN_CONGRECOVERY(tp->t_flags)) {
399 TCPSTAT_INC(tcps_ecn_rcwnd);
400 tp->snd_recover = tp->snd_max;
401 if (tp->t_flags & TF_ECN_PERMIT)
402 tp->t_flags |= TF_ECN_SND_CWR;
406 maxseg = tcp_maxseg(tp);
408 tp->t_bytes_acked = 0;
409 EXIT_RECOVERY(tp->t_flags);
410 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
412 tp->snd_cwnd = maxseg;
415 TCPSTAT_INC(tcps_sndrexmitbad);
416 /* RTO was unnecessary, so reset everything. */
417 tp->snd_cwnd = tp->snd_cwnd_prev;
418 tp->snd_ssthresh = tp->snd_ssthresh_prev;
419 tp->snd_recover = tp->snd_recover_prev;
420 if (tp->t_flags & TF_WASFRECOVERY)
421 ENTER_FASTRECOVERY(tp->t_flags);
422 if (tp->t_flags & TF_WASCRECOVERY)
423 ENTER_CONGRECOVERY(tp->t_flags);
424 tp->snd_nxt = tp->snd_max;
425 tp->t_flags &= ~TF_PREVVALID;
430 if (CC_ALGO(tp)->cong_signal != NULL) {
432 tp->ccv->curack = th->th_ack;
433 CC_ALGO(tp)->cong_signal(tp->ccv, type);
438 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
440 INP_WLOCK_ASSERT(tp->t_inpcb);
442 /* XXXLAS: KASSERT that we're in recovery? */
444 if (CC_ALGO(tp)->post_recovery != NULL) {
445 tp->ccv->curack = th->th_ack;
446 CC_ALGO(tp)->post_recovery(tp->ccv);
448 /* XXXLAS: EXIT_RECOVERY ? */
449 tp->t_bytes_acked = 0;
453 * Indicate whether this ack should be delayed. We can delay the ack if
454 * following conditions are met:
455 * - There is no delayed ack timer in progress.
456 * - Our last ack wasn't a 0-sized window. We never want to delay
457 * the ack that opens up a 0-sized window.
458 * - LRO wasn't used for this segment. We make sure by checking that the
459 * segment size is not larger than the MSS.
461 #define DELAY_ACK(tp, tlen) \
462 ((!tcp_timer_active(tp, TT_DELACK) && \
463 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
464 (tlen <= tp->t_maxseg) && \
465 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
468 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
470 INP_WLOCK_ASSERT(tp->t_inpcb);
472 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
473 switch (iptos & IPTOS_ECN_MASK) {
475 tp->ccv->flags |= CCF_IPHDR_CE;
478 tp->ccv->flags &= ~CCF_IPHDR_CE;
481 tp->ccv->flags &= ~CCF_IPHDR_CE;
485 if (th->th_flags & TH_CWR)
486 tp->ccv->flags |= CCF_TCPHDR_CWR;
488 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
490 if (tp->t_flags & TF_DELACK)
491 tp->ccv->flags |= CCF_DELACK;
493 tp->ccv->flags &= ~CCF_DELACK;
495 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
497 if (tp->ccv->flags & CCF_ACKNOW)
498 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
503 * TCP input handling is split into multiple parts:
504 * tcp6_input is a thin wrapper around tcp_input for the extended
505 * ip6_protox[] call format in ip6_input
506 * tcp_input handles primary segment validation, inpcb lookup and
507 * SYN processing on listen sockets
508 * tcp_do_segment processes the ACK and text of the segment for
509 * establishing, established and closing connections
513 tcp6_input(struct mbuf **mp, int *offp, int proto)
515 struct mbuf *m = *mp;
516 struct in6_ifaddr *ia6;
519 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
522 * draft-itojun-ipv6-tcp-to-anycast
523 * better place to put this in?
525 ip6 = mtod(m, struct ip6_hdr *);
526 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
527 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
530 ifa_free(&ia6->ia_ifa);
531 ip6 = mtod(m, struct ip6_hdr *);
532 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
533 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
534 return (IPPROTO_DONE);
537 ifa_free(&ia6->ia_ifa);
539 return (tcp_input(mp, offp, proto));
544 tcp_input(struct mbuf **mp, int *offp, int proto)
546 struct mbuf *m = *mp;
547 struct tcphdr *th = NULL;
548 struct ip *ip = NULL;
549 struct inpcb *inp = NULL;
550 struct tcpcb *tp = NULL;
551 struct socket *so = NULL;
561 int rstreason = 0; /* For badport_bandlim accounting purposes */
563 struct m_tag *fwd_tag = NULL;
564 struct epoch_tracker et;
566 struct ip6_hdr *ip6 = NULL;
569 const void *ip6 = NULL;
571 struct tcpopt to; /* options in this segment */
572 char *s = NULL; /* address and port logging */
576 * The size of tcp_saveipgen must be the size of the max ip header,
579 u_char tcp_saveipgen[IP6_HDR_LEN];
580 struct tcphdr tcp_savetcp;
585 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
592 TCPSTAT_INC(tcps_rcvtotal);
596 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
598 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
599 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
601 TCPSTAT_INC(tcps_rcvshort);
602 return (IPPROTO_DONE);
606 ip6 = mtod(m, struct ip6_hdr *);
607 th = (struct tcphdr *)((caddr_t)ip6 + off0);
608 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
609 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
610 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
611 th->th_sum = m->m_pkthdr.csum_data;
613 th->th_sum = in6_cksum_pseudo(ip6, tlen,
614 IPPROTO_TCP, m->m_pkthdr.csum_data);
615 th->th_sum ^= 0xffff;
617 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
619 TCPSTAT_INC(tcps_rcvbadsum);
624 * Be proactive about unspecified IPv6 address in source.
625 * As we use all-zero to indicate unbounded/unconnected pcb,
626 * unspecified IPv6 address can be used to confuse us.
628 * Note that packets with unspecified IPv6 destination is
629 * already dropped in ip6_input.
631 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
635 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
638 #if defined(INET) && defined(INET6)
644 * Get IP and TCP header together in first mbuf.
645 * Note: IP leaves IP header in first mbuf.
647 if (off0 > sizeof (struct ip)) {
649 off0 = sizeof(struct ip);
651 if (m->m_len < sizeof (struct tcpiphdr)) {
652 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
654 TCPSTAT_INC(tcps_rcvshort);
655 return (IPPROTO_DONE);
658 ip = mtod(m, struct ip *);
659 th = (struct tcphdr *)((caddr_t)ip + off0);
660 tlen = ntohs(ip->ip_len) - off0;
663 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
664 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
665 th->th_sum = m->m_pkthdr.csum_data;
667 th->th_sum = in_pseudo(ip->ip_src.s_addr,
669 htonl(m->m_pkthdr.csum_data + tlen +
671 th->th_sum ^= 0xffff;
673 struct ipovly *ipov = (struct ipovly *)ip;
676 * Checksum extended TCP header and data.
679 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
680 ipov->ih_len = htons(tlen);
681 th->th_sum = in_cksum(m, len);
682 /* Reset length for SDT probes. */
683 ip->ip_len = htons(len);
686 /* Re-initialization for later version check */
687 ip->ip_v = IPVERSION;
688 ip->ip_hl = off0 >> 2;
692 TCPSTAT_INC(tcps_rcvbadsum);
699 * Check that TCP offset makes sense,
700 * pull out TCP options and adjust length. XXX
702 off = th->th_off << 2;
703 if (off < sizeof (struct tcphdr) || off > tlen) {
704 TCPSTAT_INC(tcps_rcvbadoff);
707 tlen -= off; /* tlen is used instead of ti->ti_len */
708 if (off > sizeof (struct tcphdr)) {
711 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
712 ip6 = mtod(m, struct ip6_hdr *);
713 th = (struct tcphdr *)((caddr_t)ip6 + off0);
716 #if defined(INET) && defined(INET6)
721 if (m->m_len < sizeof(struct ip) + off) {
722 if ((m = m_pullup(m, sizeof (struct ip) + off))
724 TCPSTAT_INC(tcps_rcvshort);
725 return (IPPROTO_DONE);
727 ip = mtod(m, struct ip *);
728 th = (struct tcphdr *)((caddr_t)ip + off0);
732 optlen = off - sizeof (struct tcphdr);
733 optp = (u_char *)(th + 1);
735 thflags = th->th_flags;
738 * Convert TCP protocol specific fields to host format.
740 tcp_fields_to_host(th);
743 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
745 drop_hdrlen = off0 + off;
748 * Locate pcb for segment; if we're likely to add or remove a
749 * connection then first acquire pcbinfo lock. There are three cases
750 * where we might discover later we need a write lock despite the
751 * flags: ACKs moving a connection out of the syncache, ACKs for a
752 * connection in TIMEWAIT and SYNs not targeting a listening socket.
754 if ((thflags & (TH_FIN | TH_RST)) != 0) {
755 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
756 ti_locked = TI_RLOCKED;
758 ti_locked = TI_UNLOCKED;
761 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
765 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
767 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
770 #if defined(INET) && !defined(INET6)
771 (m->m_flags & M_IP_NEXTHOP)
774 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
778 if (ti_locked == TI_RLOCKED) {
779 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
781 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
785 if (isipv6 && fwd_tag != NULL) {
786 struct sockaddr_in6 *next_hop6;
788 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
790 * Transparently forwarded. Pretend to be the destination.
791 * Already got one like this?
793 inp = in6_pcblookup_mbuf(&V_tcbinfo,
794 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
795 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
798 * It's new. Try to find the ambushing socket.
799 * Because we've rewritten the destination address,
800 * any hardware-generated hash is ignored.
802 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
803 th->th_sport, &next_hop6->sin6_addr,
804 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
805 th->th_dport, INPLOOKUP_WILDCARD |
806 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
809 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
810 th->th_sport, &ip6->ip6_dst, th->th_dport,
811 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
812 m->m_pkthdr.rcvif, m);
815 #if defined(INET6) && defined(INET)
819 if (fwd_tag != NULL) {
820 struct sockaddr_in *next_hop;
822 next_hop = (struct sockaddr_in *)(fwd_tag+1);
824 * Transparently forwarded. Pretend to be the destination.
825 * already got one like this?
827 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
828 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
829 m->m_pkthdr.rcvif, m);
832 * It's new. Try to find the ambushing socket.
833 * Because we've rewritten the destination address,
834 * any hardware-generated hash is ignored.
836 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
837 th->th_sport, next_hop->sin_addr,
838 next_hop->sin_port ? ntohs(next_hop->sin_port) :
839 th->th_dport, INPLOOKUP_WILDCARD |
840 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
843 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
844 th->th_sport, ip->ip_dst, th->th_dport,
845 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
846 m->m_pkthdr.rcvif, m);
850 * If the INPCB does not exist then all data in the incoming
851 * segment is discarded and an appropriate RST is sent back.
852 * XXX MRT Send RST using which routing table?
856 * Log communication attempts to ports that are not
859 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
860 tcp_log_in_vain == 2) {
861 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
862 log(LOG_INFO, "%s; %s: Connection attempt "
863 "to closed port\n", s, __func__);
866 * When blackholing do not respond with a RST but
867 * completely ignore the segment and drop it.
869 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
873 rstreason = BANDLIM_RST_CLOSEDPORT;
876 INP_WLOCK_ASSERT(inp);
878 * While waiting for inp lock during the lookup, another thread
879 * can have dropped the inpcb, in which case we need to loop back
880 * and try to find a new inpcb to deliver to.
882 if (inp->inp_flags & INP_DROPPED) {
887 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
888 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
889 ((inp->inp_socket == NULL) ||
890 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
891 inp->inp_flowid = m->m_pkthdr.flowid;
892 inp->inp_flowtype = M_HASHTYPE_GET(m);
894 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
896 if (isipv6 && IPSEC_ENABLED(ipv6) &&
897 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
905 if (IPSEC_ENABLED(ipv4) &&
906 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
913 * Check the minimum TTL for socket.
915 if (inp->inp_ip_minttl != 0) {
918 if (inp->inp_ip_minttl > ip6->ip6_hlim)
922 if (inp->inp_ip_minttl > ip->ip_ttl)
927 * A previous connection in TIMEWAIT state is supposed to catch stray
928 * or duplicate segments arriving late. If this segment was a
929 * legitimate new connection attempt, the old INPCB gets removed and
930 * we can try again to find a listening socket.
932 * At this point, due to earlier optimism, we may hold only an inpcb
933 * lock, and not the inpcbinfo write lock. If so, we need to try to
934 * acquire it, or if that fails, acquire a reference on the inpcb,
935 * drop all locks, acquire a global write lock, and then re-acquire
936 * the inpcb lock. We may at that point discover that another thread
937 * has tried to free the inpcb, in which case we need to loop back
938 * and try to find a new inpcb to deliver to.
940 * XXXRW: It may be time to rethink timewait locking.
942 if (inp->inp_flags & INP_TIMEWAIT) {
943 if (ti_locked == TI_UNLOCKED) {
944 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
945 ti_locked = TI_RLOCKED;
947 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
949 if (thflags & TH_SYN)
950 tcp_dooptions(&to, optp, optlen, TO_SYN);
952 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
954 if (tcp_twcheck(inp, &to, th, m, tlen))
956 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
957 return (IPPROTO_DONE);
960 * The TCPCB may no longer exist if the connection is winding
961 * down or it is in the CLOSED state. Either way we drop the
962 * segment and send an appropriate response.
965 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
966 rstreason = BANDLIM_RST_CLOSEDPORT;
971 if (tp->t_flags & TF_TOE) {
972 tcp_offload_input(tp, m);
973 m = NULL; /* consumed by the TOE driver */
979 * We've identified a valid inpcb, but it could be that we need an
980 * inpcbinfo write lock but don't hold it. In this case, attempt to
981 * acquire using the same strategy as the TIMEWAIT case above. If we
982 * relock, we have to jump back to 'relocked' as the connection might
983 * now be in TIMEWAIT.
986 if ((thflags & (TH_FIN | TH_RST)) != 0)
987 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
989 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) ||
990 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) &&
991 !IS_FASTOPEN(tp->t_flags)))) {
992 if (ti_locked == TI_UNLOCKED) {
993 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
994 ti_locked = TI_RLOCKED;
996 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1000 INP_WLOCK_ASSERT(inp);
1001 if (mac_inpcb_check_deliver(inp, m))
1004 so = inp->inp_socket;
1005 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1007 if (so->so_options & SO_DEBUG) {
1008 ostate = tp->t_state;
1011 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1014 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1017 #endif /* TCPDEBUG */
1019 * When the socket is accepting connections (the INPCB is in LISTEN
1020 * state) we look into the SYN cache if this is a new connection
1021 * attempt or the completion of a previous one.
1023 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1024 ("%s: so accepting but tp %p not listening", __func__, tp));
1025 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1026 struct in_conninfo inc;
1028 bzero(&inc, sizeof(inc));
1031 inc.inc_flags |= INC_ISIPV6;
1032 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1033 inc.inc_flags |= INC_IPV6MINMTU;
1034 inc.inc6_faddr = ip6->ip6_src;
1035 inc.inc6_laddr = ip6->ip6_dst;
1039 inc.inc_faddr = ip->ip_src;
1040 inc.inc_laddr = ip->ip_dst;
1042 inc.inc_fport = th->th_sport;
1043 inc.inc_lport = th->th_dport;
1044 inc.inc_fibnum = so->so_fibnum;
1047 * Check for an existing connection attempt in syncache if
1048 * the flag is only ACK. A successful lookup creates a new
1049 * socket appended to the listen queue in SYN_RECEIVED state.
1051 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1053 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1055 * Parse the TCP options here because
1056 * syncookies need access to the reflected
1059 tcp_dooptions(&to, optp, optlen, 0);
1061 * NB: syncache_expand() doesn't unlock
1062 * inp and tcpinfo locks.
1064 rstreason = syncache_expand(&inc, &to, th, &so, m);
1065 if (rstreason < 0) {
1067 * A failing TCP MD5 signature comparison
1068 * must result in the segment being dropped
1069 * and must not produce any response back
1073 } else if (rstreason == 0) {
1075 * No syncache entry or ACK was not
1076 * for our SYN/ACK. Send a RST.
1077 * NB: syncache did its own logging
1078 * of the failure cause.
1080 rstreason = BANDLIM_RST_OPENPORT;
1086 * We completed the 3-way handshake
1087 * but could not allocate a socket
1088 * either due to memory shortage,
1089 * listen queue length limits or
1090 * global socket limits. Send RST
1091 * or wait and have the remote end
1092 * retransmit the ACK for another
1095 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1096 log(LOG_DEBUG, "%s; %s: Listen socket: "
1097 "Socket allocation failed due to "
1098 "limits or memory shortage, %s\n",
1100 V_tcp_sc_rst_sock_fail ?
1101 "sending RST" : "try again");
1102 if (V_tcp_sc_rst_sock_fail) {
1103 rstreason = BANDLIM_UNLIMITED;
1109 * Socket is created in state SYN_RECEIVED.
1110 * Unlock the listen socket, lock the newly
1111 * created socket and update the tp variable.
1113 INP_WUNLOCK(inp); /* listen socket */
1114 inp = sotoinpcb(so);
1116 * New connection inpcb is already locked by
1117 * syncache_expand().
1119 INP_WLOCK_ASSERT(inp);
1120 tp = intotcpcb(inp);
1121 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1122 ("%s: ", __func__));
1124 * Process the segment and the data it
1125 * contains. tcp_do_segment() consumes
1126 * the mbuf chain and unlocks the inpcb.
1128 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1129 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1131 if (ti_locked == TI_RLOCKED)
1132 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1133 return (IPPROTO_DONE);
1136 * Segment flag validation for new connection attempts:
1138 * Our (SYN|ACK) response was rejected.
1139 * Check with syncache and remove entry to prevent
1142 * NB: syncache_chkrst does its own logging of failure
1145 if (thflags & TH_RST) {
1146 syncache_chkrst(&inc, th, m);
1150 * We can't do anything without SYN.
1152 if ((thflags & TH_SYN) == 0) {
1153 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1154 log(LOG_DEBUG, "%s; %s: Listen socket: "
1155 "SYN is missing, segment ignored\n",
1157 TCPSTAT_INC(tcps_badsyn);
1161 * (SYN|ACK) is bogus on a listen socket.
1163 if (thflags & TH_ACK) {
1164 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1165 log(LOG_DEBUG, "%s; %s: Listen socket: "
1166 "SYN|ACK invalid, segment rejected\n",
1168 syncache_badack(&inc); /* XXX: Not needed! */
1169 TCPSTAT_INC(tcps_badsyn);
1170 rstreason = BANDLIM_RST_OPENPORT;
1174 * If the drop_synfin option is enabled, drop all
1175 * segments with both the SYN and FIN bits set.
1176 * This prevents e.g. nmap from identifying the
1178 * XXX: Poor reasoning. nmap has other methods
1179 * and is constantly refining its stack detection
1181 * XXX: This is a violation of the TCP specification
1182 * and was used by RFC1644.
1184 if ((thflags & TH_FIN) && V_drop_synfin) {
1185 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1186 log(LOG_DEBUG, "%s; %s: Listen socket: "
1187 "SYN|FIN segment ignored (based on "
1188 "sysctl setting)\n", s, __func__);
1189 TCPSTAT_INC(tcps_badsyn);
1193 * Segment's flags are (SYN) or (SYN|FIN).
1195 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1196 * as they do not affect the state of the TCP FSM.
1197 * The data pointed to by TH_URG and th_urp is ignored.
1199 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1200 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1201 KASSERT(thflags & (TH_SYN),
1202 ("%s: Listen socket: TH_SYN not set", __func__));
1205 * If deprecated address is forbidden,
1206 * we do not accept SYN to deprecated interface
1207 * address to prevent any new inbound connection from
1208 * getting established.
1209 * When we do not accept SYN, we send a TCP RST,
1210 * with deprecated source address (instead of dropping
1211 * it). We compromise it as it is much better for peer
1212 * to send a RST, and RST will be the final packet
1215 * If we do not forbid deprecated addresses, we accept
1216 * the SYN packet. RFC2462 does not suggest dropping
1218 * If we decipher RFC2462 5.5.4, it says like this:
1219 * 1. use of deprecated addr with existing
1220 * communication is okay - "SHOULD continue to be
1222 * 2. use of it with new communication:
1223 * (2a) "SHOULD NOT be used if alternate address
1224 * with sufficient scope is available"
1225 * (2b) nothing mentioned otherwise.
1226 * Here we fall into (2b) case as we have no choice in
1227 * our source address selection - we must obey the peer.
1229 * The wording in RFC2462 is confusing, and there are
1230 * multiple description text for deprecated address
1231 * handling - worse, they are not exactly the same.
1232 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1234 if (isipv6 && !V_ip6_use_deprecated) {
1235 struct in6_ifaddr *ia6;
1237 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1239 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1240 ifa_free(&ia6->ia_ifa);
1241 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1242 log(LOG_DEBUG, "%s; %s: Listen socket: "
1243 "Connection attempt to deprecated "
1244 "IPv6 address rejected\n",
1246 rstreason = BANDLIM_RST_OPENPORT;
1250 ifa_free(&ia6->ia_ifa);
1254 * Basic sanity checks on incoming SYN requests:
1255 * Don't respond if the destination is a link layer
1256 * broadcast according to RFC1122 4.2.3.10, p. 104.
1257 * If it is from this socket it must be forged.
1258 * Don't respond if the source or destination is a
1259 * global or subnet broad- or multicast address.
1260 * Note that it is quite possible to receive unicast
1261 * link-layer packets with a broadcast IP address. Use
1262 * in_broadcast() to find them.
1264 if (m->m_flags & (M_BCAST|M_MCAST)) {
1265 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1266 log(LOG_DEBUG, "%s; %s: Listen socket: "
1267 "Connection attempt from broad- or multicast "
1268 "link layer address ignored\n", s, __func__);
1273 if (th->th_dport == th->th_sport &&
1274 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1275 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1276 log(LOG_DEBUG, "%s; %s: Listen socket: "
1277 "Connection attempt to/from self "
1278 "ignored\n", s, __func__);
1281 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1282 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1283 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1284 log(LOG_DEBUG, "%s; %s: Listen socket: "
1285 "Connection attempt from/to multicast "
1286 "address ignored\n", s, __func__);
1291 #if defined(INET) && defined(INET6)
1296 if (th->th_dport == th->th_sport &&
1297 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1298 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1299 log(LOG_DEBUG, "%s; %s: Listen socket: "
1300 "Connection attempt from/to self "
1301 "ignored\n", s, __func__);
1304 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1305 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1306 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1307 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1308 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1309 log(LOG_DEBUG, "%s; %s: Listen socket: "
1310 "Connection attempt from/to broad- "
1311 "or multicast address ignored\n",
1318 * SYN appears to be valid. Create compressed TCP state
1322 if (so->so_options & SO_DEBUG)
1323 tcp_trace(TA_INPUT, ostate, tp,
1324 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1326 TCP_PROBE3(debug__input, tp, th, m);
1327 tcp_dooptions(&to, optp, optlen, TO_SYN);
1328 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL))
1329 goto tfo_socket_result;
1332 * Entry added to syncache and mbuf consumed.
1333 * Only the listen socket is unlocked by syncache_add().
1335 if (ti_locked == TI_RLOCKED) {
1336 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1337 ti_locked = TI_UNLOCKED;
1339 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1340 return (IPPROTO_DONE);
1341 } else if (tp->t_state == TCPS_LISTEN) {
1343 * When a listen socket is torn down the SO_ACCEPTCONN
1344 * flag is removed first while connections are drained
1345 * from the accept queue in a unlock/lock cycle of the
1346 * ACCEPT_LOCK, opening a race condition allowing a SYN
1347 * attempt go through unhandled.
1351 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1352 if (tp->t_flags & TF_SIGNATURE) {
1353 tcp_dooptions(&to, optp, optlen, thflags);
1354 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1355 TCPSTAT_INC(tcps_sig_err_nosigopt);
1358 if (!TCPMD5_ENABLED() ||
1359 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1363 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1366 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1367 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1368 * the inpcb, and unlocks pcbinfo.
1370 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1371 if (ti_locked == TI_RLOCKED)
1372 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1373 return (IPPROTO_DONE);
1376 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1378 if (ti_locked == TI_RLOCKED) {
1379 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1380 ti_locked = TI_UNLOCKED;
1384 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1385 "ti_locked: %d", __func__, ti_locked));
1386 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1391 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1394 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1395 m = NULL; /* mbuf chain got consumed. */
1400 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1402 if (ti_locked == TI_RLOCKED) {
1403 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1404 ti_locked = TI_UNLOCKED;
1408 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1409 "ti_locked: %d", __func__, ti_locked));
1410 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1418 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1423 return (IPPROTO_DONE);
1427 * Automatic sizing of receive socket buffer. Often the send
1428 * buffer size is not optimally adjusted to the actual network
1429 * conditions at hand (delay bandwidth product). Setting the
1430 * buffer size too small limits throughput on links with high
1431 * bandwidth and high delay (eg. trans-continental/oceanic links).
1433 * On the receive side the socket buffer memory is only rarely
1434 * used to any significant extent. This allows us to be much
1435 * more aggressive in scaling the receive socket buffer. For
1436 * the case that the buffer space is actually used to a large
1437 * extent and we run out of kernel memory we can simply drop
1438 * the new segments; TCP on the sender will just retransmit it
1439 * later. Setting the buffer size too big may only consume too
1440 * much kernel memory if the application doesn't read() from
1441 * the socket or packet loss or reordering makes use of the
1444 * The criteria to step up the receive buffer one notch are:
1445 * 1. Application has not set receive buffer size with
1446 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1447 * 2. the number of bytes received during 1/2 of an sRTT
1448 * is at least 3/8 of the current socket buffer size.
1449 * 3. receive buffer size has not hit maximal automatic size;
1451 * If all of the criteria are met we increaset the socket buffer
1452 * by a 1/2 (bounded by the max). This allows us to keep ahead
1453 * of slow-start but also makes it so our peer never gets limited
1454 * by our rwnd which we then open up causing a burst.
1456 * This algorithm does two steps per RTT at most and only if
1457 * we receive a bulk stream w/o packet losses or reorderings.
1458 * Shrinking the buffer during idle times is not necessary as
1459 * it doesn't consume any memory when idle.
1461 * TODO: Only step up if the application is actually serving
1462 * the buffer to better manage the socket buffer resources.
1465 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1466 struct tcpcb *tp, int tlen)
1470 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1471 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1472 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1473 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1474 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1475 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1476 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1478 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1480 /* Start over with next RTT. */
1484 tp->rfbuf_cnt += tlen; /* add up */
1491 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1492 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1494 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1495 int rstreason, todrop, win;
1499 struct in_conninfo *inc;
1506 * The size of tcp_saveipgen must be the size of the max ip header,
1509 u_char tcp_saveipgen[IP6_HDR_LEN];
1510 struct tcphdr tcp_savetcp;
1513 thflags = th->th_flags;
1514 inc = &tp->t_inpcb->inp_inc;
1515 tp->sackhint.last_sack_ack = 0;
1517 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1519 * If this is either a state-changing packet or current state isn't
1520 * established, we require a write lock on tcbinfo. Otherwise, we
1521 * allow the tcbinfo to be in either alocked or unlocked, as the
1522 * caller may have unnecessarily acquired a write lock due to a race.
1524 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1525 tp->t_state != TCPS_ESTABLISHED) {
1526 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1528 INP_WLOCK_ASSERT(tp->t_inpcb);
1529 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1531 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1535 /* Save segment, if requested. */
1536 tcp_pcap_add(th, m, &(tp->t_inpkts));
1538 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1541 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1542 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1543 log(LOG_DEBUG, "%s; %s: "
1544 "SYN|FIN segment ignored (based on "
1545 "sysctl setting)\n", s, __func__);
1552 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1553 * check SEQ.ACK first.
1555 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1556 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1557 rstreason = BANDLIM_UNLIMITED;
1562 * Segment received on connection.
1563 * Reset idle time and keep-alive timer.
1564 * XXX: This should be done after segment
1565 * validation to ignore broken/spoofed segs.
1567 tp->t_rcvtime = ticks;
1570 * Scale up the window into a 32-bit value.
1571 * For the SYN_SENT state the scale is zero.
1573 tiwin = th->th_win << tp->snd_scale;
1576 * TCP ECN processing.
1578 if (tp->t_flags & TF_ECN_PERMIT) {
1579 if (thflags & TH_CWR)
1580 tp->t_flags &= ~TF_ECN_SND_ECE;
1581 switch (iptos & IPTOS_ECN_MASK) {
1583 tp->t_flags |= TF_ECN_SND_ECE;
1584 TCPSTAT_INC(tcps_ecn_ce);
1586 case IPTOS_ECN_ECT0:
1587 TCPSTAT_INC(tcps_ecn_ect0);
1589 case IPTOS_ECN_ECT1:
1590 TCPSTAT_INC(tcps_ecn_ect1);
1594 /* Process a packet differently from RFC3168. */
1595 cc_ecnpkt_handler(tp, th, iptos);
1597 /* Congestion experienced. */
1598 if (thflags & TH_ECE) {
1599 cc_cong_signal(tp, th, CC_ECN);
1604 * Parse options on any incoming segment.
1606 tcp_dooptions(&to, (u_char *)(th + 1),
1607 (th->th_off << 2) - sizeof(struct tcphdr),
1608 (thflags & TH_SYN) ? TO_SYN : 0);
1610 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1611 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1612 (to.to_flags & TOF_SIGNATURE) == 0) {
1613 TCPSTAT_INC(tcps_sig_err_sigopt);
1614 /* XXX: should drop? */
1618 * If echoed timestamp is later than the current time,
1619 * fall back to non RFC1323 RTT calculation. Normalize
1620 * timestamp if syncookies were used when this connection
1623 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1624 to.to_tsecr -= tp->ts_offset;
1625 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1627 else if (tp->t_flags & TF_PREVVALID &&
1628 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1629 cc_cong_signal(tp, th, CC_RTO_ERR);
1632 * Process options only when we get SYN/ACK back. The SYN case
1633 * for incoming connections is handled in tcp_syncache.
1634 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1635 * or <SYN,ACK>) segment itself is never scaled.
1636 * XXX this is traditional behavior, may need to be cleaned up.
1638 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1639 if ((to.to_flags & TOF_SCALE) &&
1640 (tp->t_flags & TF_REQ_SCALE)) {
1641 tp->t_flags |= TF_RCVD_SCALE;
1642 tp->snd_scale = to.to_wscale;
1645 * Initial send window. It will be updated with
1646 * the next incoming segment to the scaled value.
1648 tp->snd_wnd = th->th_win;
1649 if (to.to_flags & TOF_TS) {
1650 tp->t_flags |= TF_RCVD_TSTMP;
1651 tp->ts_recent = to.to_tsval;
1652 tp->ts_recent_age = tcp_ts_getticks();
1654 if (to.to_flags & TOF_MSS)
1655 tcp_mss(tp, to.to_mss);
1656 if ((tp->t_flags & TF_SACK_PERMIT) &&
1657 (to.to_flags & TOF_SACKPERM) == 0)
1658 tp->t_flags &= ~TF_SACK_PERMIT;
1659 if (IS_FASTOPEN(tp->t_flags)) {
1660 if (to.to_flags & TOF_FASTOPEN) {
1663 if (to.to_flags & TOF_MSS)
1666 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1670 tcp_fastopen_update_cache(tp, mss,
1671 to.to_tfo_len, to.to_tfo_cookie);
1673 tcp_fastopen_disable_path(tp);
1678 * If timestamps were negotiated during SYN/ACK they should
1679 * appear on every segment during this session and vice versa.
1681 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1682 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1683 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1684 "no action\n", s, __func__);
1688 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1689 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1690 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1691 "no action\n", s, __func__);
1697 * Header prediction: check for the two common cases
1698 * of a uni-directional data xfer. If the packet has
1699 * no control flags, is in-sequence, the window didn't
1700 * change and we're not retransmitting, it's a
1701 * candidate. If the length is zero and the ack moved
1702 * forward, we're the sender side of the xfer. Just
1703 * free the data acked & wake any higher level process
1704 * that was blocked waiting for space. If the length
1705 * is non-zero and the ack didn't move, we're the
1706 * receiver side. If we're getting packets in-order
1707 * (the reassembly queue is empty), add the data to
1708 * the socket buffer and note that we need a delayed ack.
1709 * Make sure that the hidden state-flags are also off.
1710 * Since we check for TCPS_ESTABLISHED first, it can only
1713 if (tp->t_state == TCPS_ESTABLISHED &&
1714 th->th_seq == tp->rcv_nxt &&
1715 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1716 tp->snd_nxt == tp->snd_max &&
1717 tiwin && tiwin == tp->snd_wnd &&
1718 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1720 ((to.to_flags & TOF_TS) == 0 ||
1721 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1724 * If last ACK falls within this segment's sequence numbers,
1725 * record the timestamp.
1726 * NOTE that the test is modified according to the latest
1727 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1729 if ((to.to_flags & TOF_TS) != 0 &&
1730 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1731 tp->ts_recent_age = tcp_ts_getticks();
1732 tp->ts_recent = to.to_tsval;
1736 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1737 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1738 !IN_RECOVERY(tp->t_flags) &&
1739 (to.to_flags & TOF_SACK) == 0 &&
1740 TAILQ_EMPTY(&tp->snd_holes)) {
1742 * This is a pure ack for outstanding data.
1744 TCPSTAT_INC(tcps_predack);
1747 * "bad retransmit" recovery without timestamps.
1749 if ((to.to_flags & TOF_TS) == 0 &&
1750 tp->t_rxtshift == 1 &&
1751 tp->t_flags & TF_PREVVALID &&
1752 (int)(ticks - tp->t_badrxtwin) < 0) {
1753 cc_cong_signal(tp, th, CC_RTO_ERR);
1757 * Recalculate the transmit timer / rtt.
1759 * Some boxes send broken timestamp replies
1760 * during the SYN+ACK phase, ignore
1761 * timestamps of 0 or we could calculate a
1762 * huge RTT and blow up the retransmit timer.
1764 if ((to.to_flags & TOF_TS) != 0 &&
1768 t = tcp_ts_getticks() - to.to_tsecr;
1769 if (!tp->t_rttlow || tp->t_rttlow > t)
1772 TCP_TS_TO_TICKS(t) + 1);
1773 } else if (tp->t_rtttime &&
1774 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1775 if (!tp->t_rttlow ||
1776 tp->t_rttlow > ticks - tp->t_rtttime)
1777 tp->t_rttlow = ticks - tp->t_rtttime;
1779 ticks - tp->t_rtttime);
1781 acked = BYTES_THIS_ACK(tp, th);
1784 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1785 hhook_run_tcp_est_in(tp, th, &to);
1788 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1789 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1790 sbdrop(&so->so_snd, acked);
1791 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1792 SEQ_LEQ(th->th_ack, tp->snd_recover))
1793 tp->snd_recover = th->th_ack - 1;
1796 * Let the congestion control algorithm update
1797 * congestion control related information. This
1798 * typically means increasing the congestion
1801 cc_ack_received(tp, th, nsegs, CC_ACK);
1803 tp->snd_una = th->th_ack;
1805 * Pull snd_wl2 up to prevent seq wrap relative
1808 tp->snd_wl2 = th->th_ack;
1813 * If all outstanding data are acked, stop
1814 * retransmit timer, otherwise restart timer
1815 * using current (possibly backed-off) value.
1816 * If process is waiting for space,
1817 * wakeup/selwakeup/signal. If data
1818 * are ready to send, let tcp_output
1819 * decide between more output or persist.
1822 if (so->so_options & SO_DEBUG)
1823 tcp_trace(TA_INPUT, ostate, tp,
1824 (void *)tcp_saveipgen,
1827 TCP_PROBE3(debug__input, tp, th, m);
1828 if (tp->snd_una == tp->snd_max)
1829 tcp_timer_activate(tp, TT_REXMT, 0);
1830 else if (!tcp_timer_active(tp, TT_PERSIST))
1831 tcp_timer_activate(tp, TT_REXMT,
1834 if (sbavail(&so->so_snd))
1835 (void) tp->t_fb->tfb_tcp_output(tp);
1838 } else if (th->th_ack == tp->snd_una &&
1839 tlen <= sbspace(&so->so_rcv)) {
1840 int newsize = 0; /* automatic sockbuf scaling */
1843 * This is a pure, in-sequence data packet with
1844 * nothing on the reassembly queue and we have enough
1845 * buffer space to take it.
1847 /* Clean receiver SACK report if present */
1848 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1849 tcp_clean_sackreport(tp);
1850 TCPSTAT_INC(tcps_preddat);
1851 tp->rcv_nxt += tlen;
1853 * Pull snd_wl1 up to prevent seq wrap relative to
1856 tp->snd_wl1 = th->th_seq;
1858 * Pull rcv_up up to prevent seq wrap relative to
1861 tp->rcv_up = tp->rcv_nxt;
1862 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1863 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1865 if (so->so_options & SO_DEBUG)
1866 tcp_trace(TA_INPUT, ostate, tp,
1867 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1869 TCP_PROBE3(debug__input, tp, th, m);
1871 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1873 /* Add data to socket buffer. */
1874 SOCKBUF_LOCK(&so->so_rcv);
1875 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1879 * Set new socket buffer size.
1880 * Give up when limit is reached.
1883 if (!sbreserve_locked(&so->so_rcv,
1885 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1886 m_adj(m, drop_hdrlen); /* delayed header drop */
1887 sbappendstream_locked(&so->so_rcv, m, 0);
1889 /* NB: sorwakeup_locked() does an implicit unlock. */
1890 sorwakeup_locked(so);
1891 if (DELAY_ACK(tp, tlen)) {
1892 tp->t_flags |= TF_DELACK;
1894 tp->t_flags |= TF_ACKNOW;
1895 tp->t_fb->tfb_tcp_output(tp);
1902 * Calculate amount of space in receive window,
1903 * and then do TCP input processing.
1904 * Receive window is amount of space in rcv queue,
1905 * but not less than advertised window.
1907 win = sbspace(&so->so_rcv);
1910 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1912 switch (tp->t_state) {
1915 * If the state is SYN_RECEIVED:
1916 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1918 case TCPS_SYN_RECEIVED:
1919 if ((thflags & TH_ACK) &&
1920 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1921 SEQ_GT(th->th_ack, tp->snd_max))) {
1922 rstreason = BANDLIM_RST_OPENPORT;
1925 if (IS_FASTOPEN(tp->t_flags)) {
1927 * When a TFO connection is in SYN_RECEIVED, the
1928 * only valid packets are the initial SYN, a
1929 * retransmit/copy of the initial SYN (possibly with
1930 * a subset of the original data), a valid ACK, a
1933 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1934 rstreason = BANDLIM_RST_OPENPORT;
1936 } else if (thflags & TH_SYN) {
1937 /* non-initial SYN is ignored */
1938 if ((tcp_timer_active(tp, TT_DELACK) ||
1939 tcp_timer_active(tp, TT_REXMT)))
1941 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1948 * If the state is SYN_SENT:
1949 * if seg contains a RST with valid ACK (SEQ.ACK has already
1950 * been verified), then drop the connection.
1951 * if seg contains a RST without an ACK, drop the seg.
1952 * if seg does not contain SYN, then drop the seg.
1953 * Otherwise this is an acceptable SYN segment
1954 * initialize tp->rcv_nxt and tp->irs
1955 * if seg contains ack then advance tp->snd_una
1956 * if seg contains an ECE and ECN support is enabled, the stream
1958 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1959 * arrange for segment to be acked (eventually)
1960 * continue processing rest of data/controls, beginning with URG
1963 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1964 TCP_PROBE5(connect__refused, NULL, tp,
1966 tp = tcp_drop(tp, ECONNREFUSED);
1968 if (thflags & TH_RST)
1970 if (!(thflags & TH_SYN))
1973 tp->irs = th->th_seq;
1975 if (thflags & TH_ACK) {
1976 int tfo_partial_ack = 0;
1978 TCPSTAT_INC(tcps_connects);
1981 mac_socketpeer_set_from_mbuf(m, so);
1983 /* Do window scaling on this connection? */
1984 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1985 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1986 tp->rcv_scale = tp->request_r_scale;
1988 tp->rcv_adv += min(tp->rcv_wnd,
1989 TCP_MAXWIN << tp->rcv_scale);
1990 tp->snd_una++; /* SYN is acked */
1992 * If not all the data that was sent in the TFO SYN
1993 * has been acked, resend the remainder right away.
1995 if (IS_FASTOPEN(tp->t_flags) &&
1996 (tp->snd_una != tp->snd_max)) {
1997 tp->snd_nxt = th->th_ack;
1998 tfo_partial_ack = 1;
2001 * If there's data, delay ACK; if there's also a FIN
2002 * ACKNOW will be turned on later.
2004 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2005 tcp_timer_activate(tp, TT_DELACK,
2008 tp->t_flags |= TF_ACKNOW;
2010 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2012 tp->t_flags |= TF_ECN_PERMIT;
2013 TCPSTAT_INC(tcps_ecn_shs);
2017 * Received <SYN,ACK> in SYN_SENT[*] state.
2019 * SYN_SENT --> ESTABLISHED
2020 * SYN_SENT* --> FIN_WAIT_1
2022 tp->t_starttime = ticks;
2023 if (tp->t_flags & TF_NEEDFIN) {
2024 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2025 tp->t_flags &= ~TF_NEEDFIN;
2028 tcp_state_change(tp, TCPS_ESTABLISHED);
2029 TCP_PROBE5(connect__established, NULL, tp,
2032 tcp_timer_activate(tp, TT_KEEP,
2037 * Received initial SYN in SYN-SENT[*] state =>
2038 * simultaneous open.
2039 * If it succeeds, connection is * half-synchronized.
2040 * Otherwise, do 3-way handshake:
2041 * SYN-SENT -> SYN-RECEIVED
2042 * SYN-SENT* -> SYN-RECEIVED*
2044 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2045 tcp_timer_activate(tp, TT_REXMT, 0);
2046 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2049 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2050 INP_WLOCK_ASSERT(tp->t_inpcb);
2053 * Advance th->th_seq to correspond to first data byte.
2054 * If data, trim to stay within window,
2055 * dropping FIN if necessary.
2058 if (tlen > tp->rcv_wnd) {
2059 todrop = tlen - tp->rcv_wnd;
2063 TCPSTAT_INC(tcps_rcvpackafterwin);
2064 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2066 tp->snd_wl1 = th->th_seq - 1;
2067 tp->rcv_up = th->th_seq;
2069 * Client side of transaction: already sent SYN and data.
2070 * If the remote host used T/TCP to validate the SYN,
2071 * our data will be ACK'd; if so, enter normal data segment
2072 * processing in the middle of step 5, ack processing.
2073 * Otherwise, goto step 6.
2075 if (thflags & TH_ACK)
2081 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2082 * do normal processing.
2084 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2088 break; /* continue normal processing */
2092 * States other than LISTEN or SYN_SENT.
2093 * First check the RST flag and sequence number since reset segments
2094 * are exempt from the timestamp and connection count tests. This
2095 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2096 * below which allowed reset segments in half the sequence space
2097 * to fall though and be processed (which gives forged reset
2098 * segments with a random sequence number a 50 percent chance of
2099 * killing a connection).
2100 * Then check timestamp, if present.
2101 * Then check the connection count, if present.
2102 * Then check that at least some bytes of segment are within
2103 * receive window. If segment begins before rcv_nxt,
2104 * drop leading data (and SYN); if nothing left, just ack.
2106 if (thflags & TH_RST) {
2108 * RFC5961 Section 3.2
2110 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2111 * - If RST is in window, we send challenge ACK.
2113 * Note: to take into account delayed ACKs, we should
2114 * test against last_ack_sent instead of rcv_nxt.
2115 * Note 2: we handle special case of closed window, not
2116 * covered by the RFC.
2118 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2119 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2120 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2122 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2123 KASSERT(tp->t_state != TCPS_SYN_SENT,
2124 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2127 if (V_tcp_insecure_rst ||
2128 tp->last_ack_sent == th->th_seq) {
2129 TCPSTAT_INC(tcps_drops);
2130 /* Drop the connection. */
2131 switch (tp->t_state) {
2132 case TCPS_SYN_RECEIVED:
2133 so->so_error = ECONNREFUSED;
2135 case TCPS_ESTABLISHED:
2136 case TCPS_FIN_WAIT_1:
2137 case TCPS_FIN_WAIT_2:
2138 case TCPS_CLOSE_WAIT:
2141 so->so_error = ECONNRESET;
2148 TCPSTAT_INC(tcps_badrst);
2149 /* Send challenge ACK. */
2150 tcp_respond(tp, mtod(m, void *), th, m,
2151 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2152 tp->last_ack_sent = tp->rcv_nxt;
2160 * RFC5961 Section 4.2
2161 * Send challenge ACK for any SYN in synchronized state.
2163 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2164 tp->t_state != TCPS_SYN_RECEIVED) {
2165 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2167 TCPSTAT_INC(tcps_badsyn);
2168 if (V_tcp_insecure_syn &&
2169 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2170 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2171 tp = tcp_drop(tp, ECONNRESET);
2172 rstreason = BANDLIM_UNLIMITED;
2174 /* Send challenge ACK. */
2175 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2176 tp->snd_nxt, TH_ACK);
2177 tp->last_ack_sent = tp->rcv_nxt;
2184 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2185 * and it's less than ts_recent, drop it.
2187 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2188 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2190 /* Check to see if ts_recent is over 24 days old. */
2191 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2193 * Invalidate ts_recent. If this segment updates
2194 * ts_recent, the age will be reset later and ts_recent
2195 * will get a valid value. If it does not, setting
2196 * ts_recent to zero will at least satisfy the
2197 * requirement that zero be placed in the timestamp
2198 * echo reply when ts_recent isn't valid. The
2199 * age isn't reset until we get a valid ts_recent
2200 * because we don't want out-of-order segments to be
2201 * dropped when ts_recent is old.
2205 TCPSTAT_INC(tcps_rcvduppack);
2206 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2207 TCPSTAT_INC(tcps_pawsdrop);
2215 * In the SYN-RECEIVED state, validate that the packet belongs to
2216 * this connection before trimming the data to fit the receive
2217 * window. Check the sequence number versus IRS since we know
2218 * the sequence numbers haven't wrapped. This is a partial fix
2219 * for the "LAND" DoS attack.
2221 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2222 rstreason = BANDLIM_RST_OPENPORT;
2226 todrop = tp->rcv_nxt - th->th_seq;
2228 if (thflags & TH_SYN) {
2238 * Following if statement from Stevens, vol. 2, p. 960.
2241 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2243 * Any valid FIN must be to the left of the window.
2244 * At this point the FIN must be a duplicate or out
2245 * of sequence; drop it.
2250 * Send an ACK to resynchronize and drop any data.
2251 * But keep on processing for RST or ACK.
2253 tp->t_flags |= TF_ACKNOW;
2255 TCPSTAT_INC(tcps_rcvduppack);
2256 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2258 TCPSTAT_INC(tcps_rcvpartduppack);
2259 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2261 drop_hdrlen += todrop; /* drop from the top afterwards */
2262 th->th_seq += todrop;
2264 if (th->th_urp > todrop)
2265 th->th_urp -= todrop;
2273 * If new data are received on a connection after the
2274 * user processes are gone, then RST the other end.
2276 if ((so->so_state & SS_NOFDREF) &&
2277 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2278 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2280 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2281 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2282 "after socket was closed, "
2283 "sending RST and removing tcpcb\n",
2284 s, __func__, tcpstates[tp->t_state], tlen);
2288 TCPSTAT_INC(tcps_rcvafterclose);
2289 rstreason = BANDLIM_UNLIMITED;
2294 * If segment ends after window, drop trailing data
2295 * (and PUSH and FIN); if nothing left, just ACK.
2297 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2299 TCPSTAT_INC(tcps_rcvpackafterwin);
2300 if (todrop >= tlen) {
2301 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2303 * If window is closed can only take segments at
2304 * window edge, and have to drop data and PUSH from
2305 * incoming segments. Continue processing, but
2306 * remember to ack. Otherwise, drop segment
2309 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2310 tp->t_flags |= TF_ACKNOW;
2311 TCPSTAT_INC(tcps_rcvwinprobe);
2315 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2318 thflags &= ~(TH_PUSH|TH_FIN);
2322 * If last ACK falls within this segment's sequence numbers,
2323 * record its timestamp.
2325 * 1) That the test incorporates suggestions from the latest
2326 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2327 * 2) That updating only on newer timestamps interferes with
2328 * our earlier PAWS tests, so this check should be solely
2329 * predicated on the sequence space of this segment.
2330 * 3) That we modify the segment boundary check to be
2331 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2332 * instead of RFC1323's
2333 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2334 * This modified check allows us to overcome RFC1323's
2335 * limitations as described in Stevens TCP/IP Illustrated
2336 * Vol. 2 p.869. In such cases, we can still calculate the
2337 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2339 if ((to.to_flags & TOF_TS) != 0 &&
2340 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2341 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2342 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2343 tp->ts_recent_age = tcp_ts_getticks();
2344 tp->ts_recent = to.to_tsval;
2348 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2349 * flag is on (half-synchronized state), then queue data for
2350 * later processing; else drop segment and return.
2352 if ((thflags & TH_ACK) == 0) {
2353 if (tp->t_state == TCPS_SYN_RECEIVED ||
2354 (tp->t_flags & TF_NEEDSYN)) {
2355 if (tp->t_state == TCPS_SYN_RECEIVED &&
2356 IS_FASTOPEN(tp->t_flags)) {
2357 tp->snd_wnd = tiwin;
2361 } else if (tp->t_flags & TF_ACKNOW)
2370 switch (tp->t_state) {
2373 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2374 * ESTABLISHED state and continue processing.
2375 * The ACK was checked above.
2377 case TCPS_SYN_RECEIVED:
2379 TCPSTAT_INC(tcps_connects);
2381 /* Do window scaling? */
2382 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2383 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2384 tp->rcv_scale = tp->request_r_scale;
2386 tp->snd_wnd = tiwin;
2389 * SYN-RECEIVED -> ESTABLISHED
2390 * SYN-RECEIVED* -> FIN-WAIT-1
2392 tp->t_starttime = ticks;
2393 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2394 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2395 tp->t_tfo_pending = NULL;
2398 * Account for the ACK of our SYN prior to
2399 * regular ACK processing below.
2403 if (tp->t_flags & TF_NEEDFIN) {
2404 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2405 tp->t_flags &= ~TF_NEEDFIN;
2407 tcp_state_change(tp, TCPS_ESTABLISHED);
2408 TCP_PROBE5(accept__established, NULL, tp,
2411 * TFO connections call cc_conn_init() during SYN
2412 * processing. Calling it again here for such
2413 * connections is not harmless as it would undo the
2414 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2417 if (!IS_FASTOPEN(tp->t_flags))
2419 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2422 * If segment contains data or ACK, will call tcp_reass()
2423 * later; if not, do so now to pass queued data to user.
2425 if (tlen == 0 && (thflags & TH_FIN) == 0)
2426 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2428 tp->snd_wl1 = th->th_seq - 1;
2432 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2433 * ACKs. If the ack is in the range
2434 * tp->snd_una < th->th_ack <= tp->snd_max
2435 * then advance tp->snd_una to th->th_ack and drop
2436 * data from the retransmission queue. If this ACK reflects
2437 * more up to date window information we update our window information.
2439 case TCPS_ESTABLISHED:
2440 case TCPS_FIN_WAIT_1:
2441 case TCPS_FIN_WAIT_2:
2442 case TCPS_CLOSE_WAIT:
2445 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2446 TCPSTAT_INC(tcps_rcvacktoomuch);
2449 if ((tp->t_flags & TF_SACK_PERMIT) &&
2450 ((to.to_flags & TOF_SACK) ||
2451 !TAILQ_EMPTY(&tp->snd_holes)))
2452 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2455 * Reset the value so that previous (valid) value
2456 * from the last ack with SACK doesn't get used.
2458 tp->sackhint.sacked_bytes = 0;
2461 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2462 hhook_run_tcp_est_in(tp, th, &to);
2465 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2468 maxseg = tcp_maxseg(tp);
2470 (tiwin == tp->snd_wnd ||
2471 (tp->t_flags & TF_SACK_PERMIT))) {
2473 * If this is the first time we've seen a
2474 * FIN from the remote, this is not a
2475 * duplicate and it needs to be processed
2476 * normally. This happens during a
2477 * simultaneous close.
2479 if ((thflags & TH_FIN) &&
2480 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2484 TCPSTAT_INC(tcps_rcvdupack);
2486 * If we have outstanding data (other than
2487 * a window probe), this is a completely
2488 * duplicate ack (ie, window info didn't
2489 * change and FIN isn't set),
2490 * the ack is the biggest we've
2491 * seen and we've seen exactly our rexmt
2492 * threshold of them, assume a packet
2493 * has been dropped and retransmit it.
2494 * Kludge snd_nxt & the congestion
2495 * window so we send only this one
2498 * We know we're losing at the current
2499 * window size so do congestion avoidance
2500 * (set ssthresh to half the current window
2501 * and pull our congestion window back to
2502 * the new ssthresh).
2504 * Dup acks mean that packets have left the
2505 * network (they're now cached at the receiver)
2506 * so bump cwnd by the amount in the receiver
2507 * to keep a constant cwnd packets in the
2510 * When using TCP ECN, notify the peer that
2511 * we reduced the cwnd.
2514 * Following 2 kinds of acks should not affect
2517 * 2) Acks with SACK but without any new SACK
2518 * information in them. These could result from
2519 * any anomaly in the network like a switch
2520 * duplicating packets or a possible DoS attack.
2522 if (th->th_ack != tp->snd_una ||
2523 ((tp->t_flags & TF_SACK_PERMIT) &&
2526 else if (!tcp_timer_active(tp, TT_REXMT))
2528 else if (++tp->t_dupacks > tcprexmtthresh ||
2529 IN_FASTRECOVERY(tp->t_flags)) {
2530 cc_ack_received(tp, th, nsegs,
2532 if ((tp->t_flags & TF_SACK_PERMIT) &&
2533 IN_FASTRECOVERY(tp->t_flags)) {
2537 * Compute the amount of data in flight first.
2538 * We can inject new data into the pipe iff
2539 * we have less than 1/2 the original window's
2540 * worth of data in flight.
2542 if (V_tcp_do_rfc6675_pipe)
2543 awnd = tcp_compute_pipe(tp);
2545 awnd = (tp->snd_nxt - tp->snd_fack) +
2546 tp->sackhint.sack_bytes_rexmit;
2548 if (awnd < tp->snd_ssthresh) {
2549 tp->snd_cwnd += maxseg;
2550 if (tp->snd_cwnd > tp->snd_ssthresh)
2551 tp->snd_cwnd = tp->snd_ssthresh;
2554 tp->snd_cwnd += maxseg;
2555 (void) tp->t_fb->tfb_tcp_output(tp);
2557 } else if (tp->t_dupacks == tcprexmtthresh) {
2558 tcp_seq onxt = tp->snd_nxt;
2561 * If we're doing sack, check to
2562 * see if we're already in sack
2563 * recovery. If we're not doing sack,
2564 * check to see if we're in newreno
2567 if (tp->t_flags & TF_SACK_PERMIT) {
2568 if (IN_FASTRECOVERY(tp->t_flags)) {
2573 if (SEQ_LEQ(th->th_ack,
2579 /* Congestion signal before ack. */
2580 cc_cong_signal(tp, th, CC_NDUPACK);
2581 cc_ack_received(tp, th, nsegs,
2583 tcp_timer_activate(tp, TT_REXMT, 0);
2585 if (tp->t_flags & TF_SACK_PERMIT) {
2587 tcps_sack_recovery_episode);
2588 tp->sack_newdata = tp->snd_nxt;
2589 tp->snd_cwnd = maxseg;
2590 (void) tp->t_fb->tfb_tcp_output(tp);
2593 tp->snd_nxt = th->th_ack;
2594 tp->snd_cwnd = maxseg;
2595 (void) tp->t_fb->tfb_tcp_output(tp);
2596 KASSERT(tp->snd_limited <= 2,
2597 ("%s: tp->snd_limited too big",
2599 tp->snd_cwnd = tp->snd_ssthresh +
2601 (tp->t_dupacks - tp->snd_limited);
2602 if (SEQ_GT(onxt, tp->snd_nxt))
2605 } else if (V_tcp_do_rfc3042) {
2607 * Process first and second duplicate
2608 * ACKs. Each indicates a segment
2609 * leaving the network, creating room
2610 * for more. Make sure we can send a
2611 * packet on reception of each duplicate
2612 * ACK by increasing snd_cwnd by one
2613 * segment. Restore the original
2614 * snd_cwnd after packet transmission.
2616 cc_ack_received(tp, th, nsegs,
2618 uint32_t oldcwnd = tp->snd_cwnd;
2619 tcp_seq oldsndmax = tp->snd_max;
2623 KASSERT(tp->t_dupacks == 1 ||
2625 ("%s: dupacks not 1 or 2",
2627 if (tp->t_dupacks == 1)
2628 tp->snd_limited = 0;
2630 (tp->snd_nxt - tp->snd_una) +
2631 (tp->t_dupacks - tp->snd_limited) *
2634 * Only call tcp_output when there
2635 * is new data available to be sent.
2636 * Otherwise we would send pure ACKs.
2638 SOCKBUF_LOCK(&so->so_snd);
2639 avail = sbavail(&so->so_snd) -
2640 (tp->snd_nxt - tp->snd_una);
2641 SOCKBUF_UNLOCK(&so->so_snd);
2643 (void) tp->t_fb->tfb_tcp_output(tp);
2644 sent = tp->snd_max - oldsndmax;
2645 if (sent > maxseg) {
2646 KASSERT((tp->t_dupacks == 2 &&
2647 tp->snd_limited == 0) ||
2648 (sent == maxseg + 1 &&
2649 tp->t_flags & TF_SENTFIN),
2650 ("%s: sent too much",
2652 tp->snd_limited = 2;
2653 } else if (sent > 0)
2655 tp->snd_cwnd = oldcwnd;
2662 * This ack is advancing the left edge, reset the
2667 * If this ack also has new SACK info, increment the
2668 * counter as per rfc6675.
2670 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed)
2674 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2675 ("%s: th_ack <= snd_una", __func__));
2678 * If the congestion window was inflated to account
2679 * for the other side's cached packets, retract it.
2681 if (IN_FASTRECOVERY(tp->t_flags)) {
2682 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2683 if (tp->t_flags & TF_SACK_PERMIT)
2684 tcp_sack_partialack(tp, th);
2686 tcp_newreno_partial_ack(tp, th);
2688 cc_post_recovery(tp, th);
2691 * If we reach this point, ACK is not a duplicate,
2692 * i.e., it ACKs something we sent.
2694 if (tp->t_flags & TF_NEEDSYN) {
2696 * T/TCP: Connection was half-synchronized, and our
2697 * SYN has been ACK'd (so connection is now fully
2698 * synchronized). Go to non-starred state,
2699 * increment snd_una for ACK of SYN, and check if
2700 * we can do window scaling.
2702 tp->t_flags &= ~TF_NEEDSYN;
2704 /* Do window scaling? */
2705 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2706 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2707 tp->rcv_scale = tp->request_r_scale;
2708 /* Send window already scaled. */
2713 INP_WLOCK_ASSERT(tp->t_inpcb);
2715 acked = BYTES_THIS_ACK(tp, th);
2716 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2717 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2718 tp->snd_una, th->th_ack, tp, m));
2719 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2720 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2723 * If we just performed our first retransmit, and the ACK
2724 * arrives within our recovery window, then it was a mistake
2725 * to do the retransmit in the first place. Recover our
2726 * original cwnd and ssthresh, and proceed to transmit where
2729 if (tp->t_rxtshift == 1 &&
2730 tp->t_flags & TF_PREVVALID &&
2732 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2733 cc_cong_signal(tp, th, CC_RTO_ERR);
2736 * If we have a timestamp reply, update smoothed
2737 * round trip time. If no timestamp is present but
2738 * transmit timer is running and timed sequence
2739 * number was acked, update smoothed round trip time.
2740 * Since we now have an rtt measurement, cancel the
2741 * timer backoff (cf., Phil Karn's retransmit alg.).
2742 * Recompute the initial retransmit timer.
2744 * Some boxes send broken timestamp replies
2745 * during the SYN+ACK phase, ignore
2746 * timestamps of 0 or we could calculate a
2747 * huge RTT and blow up the retransmit timer.
2749 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2752 t = tcp_ts_getticks() - to.to_tsecr;
2753 if (!tp->t_rttlow || tp->t_rttlow > t)
2755 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2756 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2757 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2758 tp->t_rttlow = ticks - tp->t_rtttime;
2759 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2763 * If all outstanding data is acked, stop retransmit
2764 * timer and remember to restart (more output or persist).
2765 * If there is more data to be acked, restart retransmit
2766 * timer, using current (possibly backed-off) value.
2768 if (th->th_ack == tp->snd_max) {
2769 tcp_timer_activate(tp, TT_REXMT, 0);
2771 } else if (!tcp_timer_active(tp, TT_PERSIST))
2772 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2775 * If no data (only SYN) was ACK'd,
2776 * skip rest of ACK processing.
2782 * Let the congestion control algorithm update congestion
2783 * control related information. This typically means increasing
2784 * the congestion window.
2786 cc_ack_received(tp, th, nsegs, CC_ACK);
2788 SOCKBUF_LOCK(&so->so_snd);
2789 if (acked > sbavail(&so->so_snd)) {
2790 if (tp->snd_wnd >= sbavail(&so->so_snd))
2791 tp->snd_wnd -= sbavail(&so->so_snd);
2794 mfree = sbcut_locked(&so->so_snd,
2795 (int)sbavail(&so->so_snd));
2798 mfree = sbcut_locked(&so->so_snd, acked);
2799 if (tp->snd_wnd >= (uint32_t) acked)
2800 tp->snd_wnd -= acked;
2805 /* NB: sowwakeup_locked() does an implicit unlock. */
2806 sowwakeup_locked(so);
2808 /* Detect una wraparound. */
2809 if (!IN_RECOVERY(tp->t_flags) &&
2810 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2811 SEQ_LEQ(th->th_ack, tp->snd_recover))
2812 tp->snd_recover = th->th_ack - 1;
2813 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2814 if (IN_RECOVERY(tp->t_flags) &&
2815 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2816 EXIT_RECOVERY(tp->t_flags);
2818 tp->snd_una = th->th_ack;
2819 if (tp->t_flags & TF_SACK_PERMIT) {
2820 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2821 tp->snd_recover = tp->snd_una;
2823 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2824 tp->snd_nxt = tp->snd_una;
2826 switch (tp->t_state) {
2829 * In FIN_WAIT_1 STATE in addition to the processing
2830 * for the ESTABLISHED state if our FIN is now acknowledged
2831 * then enter FIN_WAIT_2.
2833 case TCPS_FIN_WAIT_1:
2834 if (ourfinisacked) {
2836 * If we can't receive any more
2837 * data, then closing user can proceed.
2838 * Starting the timer is contrary to the
2839 * specification, but if we don't get a FIN
2840 * we'll hang forever.
2843 * we should release the tp also, and use a
2846 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2847 soisdisconnected(so);
2848 tcp_timer_activate(tp, TT_2MSL,
2849 (tcp_fast_finwait2_recycle ?
2850 tcp_finwait2_timeout :
2853 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2858 * In CLOSING STATE in addition to the processing for
2859 * the ESTABLISHED state if the ACK acknowledges our FIN
2860 * then enter the TIME-WAIT state, otherwise ignore
2864 if (ourfinisacked) {
2865 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2873 * In LAST_ACK, we may still be waiting for data to drain
2874 * and/or to be acked, as well as for the ack of our FIN.
2875 * If our FIN is now acknowledged, delete the TCB,
2876 * enter the closed state and return.
2879 if (ourfinisacked) {
2880 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2889 INP_WLOCK_ASSERT(tp->t_inpcb);
2892 * Update window information.
2893 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2895 if ((thflags & TH_ACK) &&
2896 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2897 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2898 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2899 /* keep track of pure window updates */
2901 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2902 TCPSTAT_INC(tcps_rcvwinupd);
2903 tp->snd_wnd = tiwin;
2904 tp->snd_wl1 = th->th_seq;
2905 tp->snd_wl2 = th->th_ack;
2906 if (tp->snd_wnd > tp->max_sndwnd)
2907 tp->max_sndwnd = tp->snd_wnd;
2912 * Process segments with URG.
2914 if ((thflags & TH_URG) && th->th_urp &&
2915 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2917 * This is a kludge, but if we receive and accept
2918 * random urgent pointers, we'll crash in
2919 * soreceive. It's hard to imagine someone
2920 * actually wanting to send this much urgent data.
2922 SOCKBUF_LOCK(&so->so_rcv);
2923 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2924 th->th_urp = 0; /* XXX */
2925 thflags &= ~TH_URG; /* XXX */
2926 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2927 goto dodata; /* XXX */
2930 * If this segment advances the known urgent pointer,
2931 * then mark the data stream. This should not happen
2932 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2933 * a FIN has been received from the remote side.
2934 * In these states we ignore the URG.
2936 * According to RFC961 (Assigned Protocols),
2937 * the urgent pointer points to the last octet
2938 * of urgent data. We continue, however,
2939 * to consider it to indicate the first octet
2940 * of data past the urgent section as the original
2941 * spec states (in one of two places).
2943 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2944 tp->rcv_up = th->th_seq + th->th_urp;
2945 so->so_oobmark = sbavail(&so->so_rcv) +
2946 (tp->rcv_up - tp->rcv_nxt) - 1;
2947 if (so->so_oobmark == 0)
2948 so->so_rcv.sb_state |= SBS_RCVATMARK;
2950 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2952 SOCKBUF_UNLOCK(&so->so_rcv);
2954 * Remove out of band data so doesn't get presented to user.
2955 * This can happen independent of advancing the URG pointer,
2956 * but if two URG's are pending at once, some out-of-band
2957 * data may creep in... ick.
2959 if (th->th_urp <= (uint32_t)tlen &&
2960 !(so->so_options & SO_OOBINLINE)) {
2961 /* hdr drop is delayed */
2962 tcp_pulloutofband(so, th, m, drop_hdrlen);
2966 * If no out of band data is expected,
2967 * pull receive urgent pointer along
2968 * with the receive window.
2970 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2971 tp->rcv_up = tp->rcv_nxt;
2974 INP_WLOCK_ASSERT(tp->t_inpcb);
2977 * Process the segment text, merging it into the TCP sequencing queue,
2978 * and arranging for acknowledgment of receipt if necessary.
2979 * This process logically involves adjusting tp->rcv_wnd as data
2980 * is presented to the user (this happens in tcp_usrreq.c,
2981 * case PRU_RCVD). If a FIN has already been received on this
2982 * connection then we just ignore the text.
2984 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
2985 IS_FASTOPEN(tp->t_flags));
2986 if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
2987 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2988 tcp_seq save_start = th->th_seq;
2989 m_adj(m, drop_hdrlen); /* delayed header drop */
2991 * Insert segment which includes th into TCP reassembly queue
2992 * with control block tp. Set thflags to whether reassembly now
2993 * includes a segment with FIN. This handles the common case
2994 * inline (segment is the next to be received on an established
2995 * connection, and the queue is empty), avoiding linkage into
2996 * and removal from the queue and repetition of various
2998 * Set DELACK for segments received in order, but ack
2999 * immediately when segments are out of order (so
3000 * fast retransmit can work).
3002 if (th->th_seq == tp->rcv_nxt &&
3004 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3006 if (DELAY_ACK(tp, tlen) || tfo_syn)
3007 tp->t_flags |= TF_DELACK;
3009 tp->t_flags |= TF_ACKNOW;
3010 tp->rcv_nxt += tlen;
3011 thflags = th->th_flags & TH_FIN;
3012 TCPSTAT_INC(tcps_rcvpack);
3013 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3014 SOCKBUF_LOCK(&so->so_rcv);
3015 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3018 sbappendstream_locked(&so->so_rcv, m, 0);
3019 /* NB: sorwakeup_locked() does an implicit unlock. */
3020 sorwakeup_locked(so);
3023 * XXX: Due to the header drop above "th" is
3024 * theoretically invalid by now. Fortunately
3025 * m_adj() doesn't actually frees any mbufs
3026 * when trimming from the head.
3028 thflags = tcp_reass(tp, th, &save_start, &tlen, m);
3029 tp->t_flags |= TF_ACKNOW;
3031 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
3032 tcp_update_sack_list(tp, save_start, save_start + tlen);
3035 * Note the amount of data that peer has sent into
3036 * our window, in order to estimate the sender's
3040 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3041 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3043 len = so->so_rcv.sb_hiwat;
3051 * If FIN is received ACK the FIN and let the user know
3052 * that the connection is closing.
3054 if (thflags & TH_FIN) {
3055 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3058 * If connection is half-synchronized
3059 * (ie NEEDSYN flag on) then delay ACK,
3060 * so it may be piggybacked when SYN is sent.
3061 * Otherwise, since we received a FIN then no
3062 * more input can be expected, send ACK now.
3064 if (tp->t_flags & TF_NEEDSYN)
3065 tp->t_flags |= TF_DELACK;
3067 tp->t_flags |= TF_ACKNOW;
3070 switch (tp->t_state) {
3073 * In SYN_RECEIVED and ESTABLISHED STATES
3074 * enter the CLOSE_WAIT state.
3076 case TCPS_SYN_RECEIVED:
3077 tp->t_starttime = ticks;
3079 case TCPS_ESTABLISHED:
3080 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3084 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3085 * enter the CLOSING state.
3087 case TCPS_FIN_WAIT_1:
3088 tcp_state_change(tp, TCPS_CLOSING);
3092 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3093 * starting the time-wait timer, turning off the other
3096 case TCPS_FIN_WAIT_2:
3097 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
3104 if (so->so_options & SO_DEBUG)
3105 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3108 TCP_PROBE3(debug__input, tp, th, m);
3111 * Return any desired output.
3113 if (needoutput || (tp->t_flags & TF_ACKNOW))
3114 (void) tp->t_fb->tfb_tcp_output(tp);
3117 INP_WLOCK_ASSERT(tp->t_inpcb);
3119 if (tp->t_flags & TF_DELACK) {
3120 tp->t_flags &= ~TF_DELACK;
3121 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3123 INP_WUNLOCK(tp->t_inpcb);
3128 * Generate an ACK dropping incoming segment if it occupies
3129 * sequence space, where the ACK reflects our state.
3131 * We can now skip the test for the RST flag since all
3132 * paths to this code happen after packets containing
3133 * RST have been dropped.
3135 * In the SYN-RECEIVED state, don't send an ACK unless the
3136 * segment we received passes the SYN-RECEIVED ACK test.
3137 * If it fails send a RST. This breaks the loop in the
3138 * "LAND" DoS attack, and also prevents an ACK storm
3139 * between two listening ports that have been sent forged
3140 * SYN segments, each with the source address of the other.
3142 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3143 (SEQ_GT(tp->snd_una, th->th_ack) ||
3144 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3145 rstreason = BANDLIM_RST_OPENPORT;
3149 if (so->so_options & SO_DEBUG)
3150 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3153 TCP_PROBE3(debug__input, tp, th, m);
3154 tp->t_flags |= TF_ACKNOW;
3155 (void) tp->t_fb->tfb_tcp_output(tp);
3156 INP_WUNLOCK(tp->t_inpcb);
3162 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3163 INP_WUNLOCK(tp->t_inpcb);
3165 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3170 * Drop space held by incoming segment and return.
3173 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3174 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3177 TCP_PROBE3(debug__input, tp, th, m);
3179 INP_WUNLOCK(tp->t_inpcb);
3184 * Issue RST and make ACK acceptable to originator of segment.
3185 * The mbuf must still include the original packet header.
3189 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3190 int tlen, int rstreason)
3196 struct ip6_hdr *ip6;
3200 INP_WLOCK_ASSERT(tp->t_inpcb);
3203 /* Don't bother if destination was broadcast/multicast. */
3204 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3207 if (mtod(m, struct ip *)->ip_v == 6) {
3208 ip6 = mtod(m, struct ip6_hdr *);
3209 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3210 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3212 /* IPv6 anycast check is done at tcp6_input() */
3215 #if defined(INET) && defined(INET6)
3220 ip = mtod(m, struct ip *);
3221 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3222 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3223 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3224 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3229 /* Perform bandwidth limiting. */
3230 if (badport_bandlim(rstreason) < 0)
3233 /* tcp_respond consumes the mbuf chain. */
3234 if (th->th_flags & TH_ACK) {
3235 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3236 th->th_ack, TH_RST);
3238 if (th->th_flags & TH_SYN)
3240 if (th->th_flags & TH_FIN)
3242 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3243 (tcp_seq)0, TH_RST|TH_ACK);
3251 * Parse TCP options and place in tcpopt.
3254 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3259 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3261 if (opt == TCPOPT_EOL)
3263 if (opt == TCPOPT_NOP)
3269 if (optlen < 2 || optlen > cnt)
3274 if (optlen != TCPOLEN_MAXSEG)
3276 if (!(flags & TO_SYN))
3278 to->to_flags |= TOF_MSS;
3279 bcopy((char *)cp + 2,
3280 (char *)&to->to_mss, sizeof(to->to_mss));
3281 to->to_mss = ntohs(to->to_mss);
3284 if (optlen != TCPOLEN_WINDOW)
3286 if (!(flags & TO_SYN))
3288 to->to_flags |= TOF_SCALE;
3289 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3291 case TCPOPT_TIMESTAMP:
3292 if (optlen != TCPOLEN_TIMESTAMP)
3294 to->to_flags |= TOF_TS;
3295 bcopy((char *)cp + 2,
3296 (char *)&to->to_tsval, sizeof(to->to_tsval));
3297 to->to_tsval = ntohl(to->to_tsval);
3298 bcopy((char *)cp + 6,
3299 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3300 to->to_tsecr = ntohl(to->to_tsecr);
3302 case TCPOPT_SIGNATURE:
3304 * In order to reply to a host which has set the
3305 * TCP_SIGNATURE option in its initial SYN, we have
3306 * to record the fact that the option was observed
3307 * here for the syncache code to perform the correct
3310 if (optlen != TCPOLEN_SIGNATURE)
3312 to->to_flags |= TOF_SIGNATURE;
3313 to->to_signature = cp + 2;
3315 case TCPOPT_SACK_PERMITTED:
3316 if (optlen != TCPOLEN_SACK_PERMITTED)
3318 if (!(flags & TO_SYN))
3322 to->to_flags |= TOF_SACKPERM;
3325 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3329 to->to_flags |= TOF_SACK;
3330 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3331 to->to_sacks = cp + 2;
3332 TCPSTAT_INC(tcps_sack_rcv_blocks);
3334 case TCPOPT_FAST_OPEN:
3336 * Cookie length validation is performed by the
3337 * server side cookie checking code or the client
3338 * side cookie cache update code.
3340 if (!(flags & TO_SYN))
3342 if (!V_tcp_fastopen_client_enable &&
3343 !V_tcp_fastopen_server_enable)
3345 to->to_flags |= TOF_FASTOPEN;
3346 to->to_tfo_len = optlen - 2;
3347 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3356 * Pull out of band byte out of a segment so
3357 * it doesn't appear in the user's data queue.
3358 * It is still reflected in the segment length for
3359 * sequencing purposes.
3362 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3365 int cnt = off + th->th_urp - 1;
3368 if (m->m_len > cnt) {
3369 char *cp = mtod(m, caddr_t) + cnt;
3370 struct tcpcb *tp = sototcpcb(so);
3372 INP_WLOCK_ASSERT(tp->t_inpcb);
3375 tp->t_oobflags |= TCPOOB_HAVEDATA;
3376 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3378 if (m->m_flags & M_PKTHDR)
3387 panic("tcp_pulloutofband");
3391 * Collect new round-trip time estimate
3392 * and update averages and current timeout.
3395 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3399 INP_WLOCK_ASSERT(tp->t_inpcb);
3401 TCPSTAT_INC(tcps_rttupdated);
3403 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3405 * srtt is stored as fixed point with 5 bits after the
3406 * binary point (i.e., scaled by 8). The following magic
3407 * is equivalent to the smoothing algorithm in rfc793 with
3408 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3409 * point). Adjust rtt to origin 0.
3411 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3412 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3414 if ((tp->t_srtt += delta) <= 0)
3418 * We accumulate a smoothed rtt variance (actually, a
3419 * smoothed mean difference), then set the retransmit
3420 * timer to smoothed rtt + 4 times the smoothed variance.
3421 * rttvar is stored as fixed point with 4 bits after the
3422 * binary point (scaled by 16). The following is
3423 * equivalent to rfc793 smoothing with an alpha of .75
3424 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3425 * rfc793's wired-in beta.
3429 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3430 if ((tp->t_rttvar += delta) <= 0)
3432 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3433 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3436 * No rtt measurement yet - use the unsmoothed rtt.
3437 * Set the variance to half the rtt (so our first
3438 * retransmit happens at 3*rtt).
3440 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3441 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3442 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3448 * the retransmit should happen at rtt + 4 * rttvar.
3449 * Because of the way we do the smoothing, srtt and rttvar
3450 * will each average +1/2 tick of bias. When we compute
3451 * the retransmit timer, we want 1/2 tick of rounding and
3452 * 1 extra tick because of +-1/2 tick uncertainty in the
3453 * firing of the timer. The bias will give us exactly the
3454 * 1.5 tick we need. But, because the bias is
3455 * statistical, we have to test that we don't drop below
3456 * the minimum feasible timer (which is 2 ticks).
3458 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3459 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3462 * We received an ack for a packet that wasn't retransmitted;
3463 * it is probably safe to discard any error indications we've
3464 * received recently. This isn't quite right, but close enough
3465 * for now (a route might have failed after we sent a segment,
3466 * and the return path might not be symmetrical).
3468 tp->t_softerror = 0;
3472 * Determine a reasonable value for maxseg size.
3473 * If the route is known, check route for mtu.
3474 * If none, use an mss that can be handled on the outgoing interface
3475 * without forcing IP to fragment. If no route is found, route has no mtu,
3476 * or the destination isn't local, use a default, hopefully conservative
3477 * size (usually 512 or the default IP max size, but no more than the mtu
3478 * of the interface), as we can't discover anything about intervening
3479 * gateways or networks. We also initialize the congestion/slow start
3480 * window to be a single segment if the destination isn't local.
3481 * While looking at the routing entry, we also initialize other path-dependent
3482 * parameters from pre-set or cached values in the routing entry.
3484 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3485 * IP options, e.g. IPSEC data, since length of this data may vary, and
3486 * thus it is calculated for every segment separately in tcp_output().
3488 * NOTE that this routine is only called when we process an incoming
3489 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3490 * settings are handled in tcp_mssopt().
3493 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3494 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3497 uint32_t maxmtu = 0;
3498 struct inpcb *inp = tp->t_inpcb;
3499 struct hc_metrics_lite metrics;
3501 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3502 size_t min_protoh = isipv6 ?
3503 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3504 sizeof (struct tcpiphdr);
3506 const size_t min_protoh = sizeof(struct tcpiphdr);
3509 INP_WLOCK_ASSERT(tp->t_inpcb);
3511 if (mtuoffer != -1) {
3512 KASSERT(offer == -1, ("%s: conflict", __func__));
3513 offer = mtuoffer - min_protoh;
3519 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3520 tp->t_maxseg = V_tcp_v6mssdflt;
3523 #if defined(INET) && defined(INET6)
3528 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3529 tp->t_maxseg = V_tcp_mssdflt;
3534 * No route to sender, stay with default mss and return.
3538 * In case we return early we need to initialize metrics
3539 * to a defined state as tcp_hc_get() would do for us
3540 * if there was no cache hit.
3542 if (metricptr != NULL)
3543 bzero(metricptr, sizeof(struct hc_metrics_lite));
3547 /* What have we got? */
3551 * Offer == 0 means that there was no MSS on the SYN
3552 * segment, in this case we use tcp_mssdflt as
3553 * already assigned to t_maxseg above.
3555 offer = tp->t_maxseg;
3560 * Offer == -1 means that we didn't receive SYN yet.
3566 * Prevent DoS attack with too small MSS. Round up
3567 * to at least minmss.
3569 offer = max(offer, V_tcp_minmss);
3573 * rmx information is now retrieved from tcp_hostcache.
3575 tcp_hc_get(&inp->inp_inc, &metrics);
3576 if (metricptr != NULL)
3577 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3580 * If there's a discovered mtu in tcp hostcache, use it.
3581 * Else, use the link mtu.
3583 if (metrics.rmx_mtu)
3584 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3588 mss = maxmtu - min_protoh;
3589 if (!V_path_mtu_discovery &&
3590 !in6_localaddr(&inp->in6p_faddr))
3591 mss = min(mss, V_tcp_v6mssdflt);
3594 #if defined(INET) && defined(INET6)
3599 mss = maxmtu - min_protoh;
3600 if (!V_path_mtu_discovery &&
3601 !in_localaddr(inp->inp_faddr))
3602 mss = min(mss, V_tcp_mssdflt);
3606 * XXX - The above conditional (mss = maxmtu - min_protoh)
3607 * probably violates the TCP spec.
3608 * The problem is that, since we don't know the
3609 * other end's MSS, we are supposed to use a conservative
3610 * default. But, if we do that, then MTU discovery will
3611 * never actually take place, because the conservative
3612 * default is much less than the MTUs typically seen
3613 * on the Internet today. For the moment, we'll sweep
3614 * this under the carpet.
3616 * The conservative default might not actually be a problem
3617 * if the only case this occurs is when sending an initial
3618 * SYN with options and data to a host we've never talked
3619 * to before. Then, they will reply with an MSS value which
3620 * will get recorded and the new parameters should get
3621 * recomputed. For Further Study.
3624 mss = min(mss, offer);
3627 * Sanity check: make sure that maxseg will be large
3628 * enough to allow some data on segments even if the
3629 * all the option space is used (40bytes). Otherwise
3630 * funny things may happen in tcp_output.
3632 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3640 tcp_mss(struct tcpcb *tp, int offer)
3646 struct hc_metrics_lite metrics;
3647 struct tcp_ifcap cap;
3649 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3651 bzero(&cap, sizeof(cap));
3652 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3658 * If there's a pipesize, change the socket buffer to that size,
3659 * don't change if sb_hiwat is different than default (then it
3660 * has been changed on purpose with setsockopt).
3661 * Make the socket buffers an integral number of mss units;
3662 * if the mss is larger than the socket buffer, decrease the mss.
3664 so = inp->inp_socket;
3665 SOCKBUF_LOCK(&so->so_snd);
3666 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3667 bufsize = metrics.rmx_sendpipe;
3669 bufsize = so->so_snd.sb_hiwat;
3673 bufsize = roundup(bufsize, mss);
3674 if (bufsize > sb_max)
3676 if (bufsize > so->so_snd.sb_hiwat)
3677 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3679 SOCKBUF_UNLOCK(&so->so_snd);
3681 * Sanity check: make sure that maxseg will be large
3682 * enough to allow some data on segments even if the
3683 * all the option space is used (40bytes). Otherwise
3684 * funny things may happen in tcp_output.
3686 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3688 tp->t_maxseg = max(mss, 64);
3690 SOCKBUF_LOCK(&so->so_rcv);
3691 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3692 bufsize = metrics.rmx_recvpipe;
3694 bufsize = so->so_rcv.sb_hiwat;
3695 if (bufsize > mss) {
3696 bufsize = roundup(bufsize, mss);
3697 if (bufsize > sb_max)
3699 if (bufsize > so->so_rcv.sb_hiwat)
3700 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3702 SOCKBUF_UNLOCK(&so->so_rcv);
3704 /* Check the interface for TSO capabilities. */
3705 if (cap.ifcap & CSUM_TSO) {
3706 tp->t_flags |= TF_TSO;
3707 tp->t_tsomax = cap.tsomax;
3708 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3709 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3714 * Determine the MSS option to send on an outgoing SYN.
3717 tcp_mssopt(struct in_conninfo *inc)
3720 uint32_t thcmtu = 0;
3721 uint32_t maxmtu = 0;
3724 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3727 if (inc->inc_flags & INC_ISIPV6) {
3728 mss = V_tcp_v6mssdflt;
3729 maxmtu = tcp_maxmtu6(inc, NULL);
3730 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3733 #if defined(INET) && defined(INET6)
3738 mss = V_tcp_mssdflt;
3739 maxmtu = tcp_maxmtu(inc, NULL);
3740 min_protoh = sizeof(struct tcpiphdr);
3743 #if defined(INET6) || defined(INET)
3744 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3747 if (maxmtu && thcmtu)
3748 mss = min(maxmtu, thcmtu) - min_protoh;
3749 else if (maxmtu || thcmtu)
3750 mss = max(maxmtu, thcmtu) - min_protoh;
3757 * On a partial ack arrives, force the retransmission of the
3758 * next unacknowledged segment. Do not clear tp->t_dupacks.
3759 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3763 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3765 tcp_seq onxt = tp->snd_nxt;
3766 uint32_t ocwnd = tp->snd_cwnd;
3767 u_int maxseg = tcp_maxseg(tp);
3769 INP_WLOCK_ASSERT(tp->t_inpcb);
3771 tcp_timer_activate(tp, TT_REXMT, 0);
3773 tp->snd_nxt = th->th_ack;
3775 * Set snd_cwnd to one segment beyond acknowledged offset.
3776 * (tp->snd_una has not yet been updated when this function is called.)
3778 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3779 tp->t_flags |= TF_ACKNOW;
3780 (void) tp->t_fb->tfb_tcp_output(tp);
3781 tp->snd_cwnd = ocwnd;
3782 if (SEQ_GT(onxt, tp->snd_nxt))
3785 * Partial window deflation. Relies on fact that tp->snd_una
3788 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3789 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3792 tp->snd_cwnd += maxseg;
3796 tcp_compute_pipe(struct tcpcb *tp)
3798 return (tp->snd_max - tp->snd_una +
3799 tp->sackhint.sack_bytes_rexmit -
3800 tp->sackhint.sacked_bytes);
3804 tcp_compute_initwnd(uint32_t maxseg)
3807 * Calculate the Initial Window, also used as Restart Window
3809 * RFC5681 Section 3.1 specifies the default conservative values.
3810 * RFC3390 specifies slightly more aggressive values.
3811 * RFC6928 increases it to ten segments.
3812 * Support for user specified value for initial flight size.
3814 if (V_tcp_initcwnd_segments)
3815 return min(V_tcp_initcwnd_segments * maxseg,
3816 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
3817 else if (V_tcp_do_rfc3390)
3818 return min(4 * maxseg, max(2 * maxseg, 4380));
3820 /* Per RFC5681 Section 3.1 */
3822 return (2 * maxseg);
3823 else if (maxseg > 1095)
3824 return (3 * maxseg);
3826 return (4 * maxseg);