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;
562 int rstreason = 0; /* For badport_bandlim accounting purposes */
564 struct m_tag *fwd_tag = NULL;
565 struct epoch_tracker et;
567 struct ip6_hdr *ip6 = NULL;
570 const void *ip6 = NULL;
572 struct tcpopt to; /* options in this segment */
573 char *s = NULL; /* address and port logging */
577 * The size of tcp_saveipgen must be the size of the max ip header,
580 u_char tcp_saveipgen[IP6_HDR_LEN];
581 struct tcphdr tcp_savetcp;
586 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
593 TCPSTAT_INC(tcps_rcvtotal);
597 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
599 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
600 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
602 TCPSTAT_INC(tcps_rcvshort);
603 return (IPPROTO_DONE);
607 ip6 = mtod(m, struct ip6_hdr *);
608 th = (struct tcphdr *)((caddr_t)ip6 + off0);
609 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
610 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
611 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
612 th->th_sum = m->m_pkthdr.csum_data;
614 th->th_sum = in6_cksum_pseudo(ip6, tlen,
615 IPPROTO_TCP, m->m_pkthdr.csum_data);
616 th->th_sum ^= 0xffff;
618 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
620 TCPSTAT_INC(tcps_rcvbadsum);
625 * Be proactive about unspecified IPv6 address in source.
626 * As we use all-zero to indicate unbounded/unconnected pcb,
627 * unspecified IPv6 address can be used to confuse us.
629 * Note that packets with unspecified IPv6 destination is
630 * already dropped in ip6_input.
632 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
636 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
639 #if defined(INET) && defined(INET6)
645 * Get IP and TCP header together in first mbuf.
646 * Note: IP leaves IP header in first mbuf.
648 if (off0 > sizeof (struct ip)) {
650 off0 = sizeof(struct ip);
652 if (m->m_len < sizeof (struct tcpiphdr)) {
653 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
655 TCPSTAT_INC(tcps_rcvshort);
656 return (IPPROTO_DONE);
659 ip = mtod(m, struct ip *);
660 th = (struct tcphdr *)((caddr_t)ip + off0);
661 tlen = ntohs(ip->ip_len) - off0;
664 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
665 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
666 th->th_sum = m->m_pkthdr.csum_data;
668 th->th_sum = in_pseudo(ip->ip_src.s_addr,
670 htonl(m->m_pkthdr.csum_data + tlen +
672 th->th_sum ^= 0xffff;
674 struct ipovly *ipov = (struct ipovly *)ip;
677 * Checksum extended TCP header and data.
681 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
682 ipov->ih_len = htons(tlen);
683 th->th_sum = in_cksum(m, len);
684 /* Reset length for SDT probes. */
685 ip->ip_len = htons(len);
688 /* Re-initialization for later version check */
690 ip->ip_v = IPVERSION;
691 ip->ip_hl = off0 >> 2;
695 TCPSTAT_INC(tcps_rcvbadsum);
702 * Check that TCP offset makes sense,
703 * pull out TCP options and adjust length. XXX
705 off = th->th_off << 2;
706 if (off < sizeof (struct tcphdr) || off > tlen) {
707 TCPSTAT_INC(tcps_rcvbadoff);
710 tlen -= off; /* tlen is used instead of ti->ti_len */
711 if (off > sizeof (struct tcphdr)) {
714 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
715 ip6 = mtod(m, struct ip6_hdr *);
716 th = (struct tcphdr *)((caddr_t)ip6 + off0);
719 #if defined(INET) && defined(INET6)
724 if (m->m_len < sizeof(struct ip) + off) {
725 if ((m = m_pullup(m, sizeof (struct ip) + off))
727 TCPSTAT_INC(tcps_rcvshort);
728 return (IPPROTO_DONE);
730 ip = mtod(m, struct ip *);
731 th = (struct tcphdr *)((caddr_t)ip + off0);
735 optlen = off - sizeof (struct tcphdr);
736 optp = (u_char *)(th + 1);
738 thflags = th->th_flags;
741 * Convert TCP protocol specific fields to host format.
743 tcp_fields_to_host(th);
746 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
748 drop_hdrlen = off0 + off;
751 * Locate pcb for segment; if we're likely to add or remove a
752 * connection then first acquire pcbinfo lock. There are three cases
753 * where we might discover later we need a write lock despite the
754 * flags: ACKs moving a connection out of the syncache, ACKs for a
755 * connection in TIMEWAIT and SYNs not targeting a listening socket.
757 if ((thflags & (TH_FIN | TH_RST)) != 0) {
758 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
759 ti_locked = TI_RLOCKED;
761 ti_locked = TI_UNLOCKED;
764 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
768 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
770 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
773 #if defined(INET) && !defined(INET6)
774 (m->m_flags & M_IP_NEXTHOP)
777 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
781 if (ti_locked == TI_RLOCKED) {
782 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
784 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
788 if (isipv6 && fwd_tag != NULL) {
789 struct sockaddr_in6 *next_hop6;
791 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
793 * Transparently forwarded. Pretend to be the destination.
794 * Already got one like this?
796 inp = in6_pcblookup_mbuf(&V_tcbinfo,
797 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
798 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
801 * It's new. Try to find the ambushing socket.
802 * Because we've rewritten the destination address,
803 * any hardware-generated hash is ignored.
805 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
806 th->th_sport, &next_hop6->sin6_addr,
807 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
808 th->th_dport, INPLOOKUP_WILDCARD |
809 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
812 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
813 th->th_sport, &ip6->ip6_dst, th->th_dport,
814 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
815 m->m_pkthdr.rcvif, m);
818 #if defined(INET6) && defined(INET)
822 if (fwd_tag != NULL) {
823 struct sockaddr_in *next_hop;
825 next_hop = (struct sockaddr_in *)(fwd_tag+1);
827 * Transparently forwarded. Pretend to be the destination.
828 * already got one like this?
830 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
831 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
832 m->m_pkthdr.rcvif, m);
835 * It's new. Try to find the ambushing socket.
836 * Because we've rewritten the destination address,
837 * any hardware-generated hash is ignored.
839 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
840 th->th_sport, next_hop->sin_addr,
841 next_hop->sin_port ? ntohs(next_hop->sin_port) :
842 th->th_dport, INPLOOKUP_WILDCARD |
843 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
846 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
847 th->th_sport, ip->ip_dst, th->th_dport,
848 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
849 m->m_pkthdr.rcvif, m);
853 * If the INPCB does not exist then all data in the incoming
854 * segment is discarded and an appropriate RST is sent back.
855 * XXX MRT Send RST using which routing table?
859 * Log communication attempts to ports that are not
862 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
863 tcp_log_in_vain == 2) {
864 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
865 log(LOG_INFO, "%s; %s: Connection attempt "
866 "to closed port\n", s, __func__);
869 * When blackholing do not respond with a RST but
870 * completely ignore the segment and drop it.
872 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
876 rstreason = BANDLIM_RST_CLOSEDPORT;
879 INP_WLOCK_ASSERT(inp);
881 * While waiting for inp lock during the lookup, another thread
882 * can have dropped the inpcb, in which case we need to loop back
883 * and try to find a new inpcb to deliver to.
885 if (inp->inp_flags & INP_DROPPED) {
890 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
891 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
892 ((inp->inp_socket == NULL) ||
893 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
894 inp->inp_flowid = m->m_pkthdr.flowid;
895 inp->inp_flowtype = M_HASHTYPE_GET(m);
897 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
899 if (isipv6 && IPSEC_ENABLED(ipv6) &&
900 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
908 if (IPSEC_ENABLED(ipv4) &&
909 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
916 * Check the minimum TTL for socket.
918 if (inp->inp_ip_minttl != 0) {
921 if (inp->inp_ip_minttl > ip6->ip6_hlim)
925 if (inp->inp_ip_minttl > ip->ip_ttl)
930 * A previous connection in TIMEWAIT state is supposed to catch stray
931 * or duplicate segments arriving late. If this segment was a
932 * legitimate new connection attempt, the old INPCB gets removed and
933 * we can try again to find a listening socket.
935 * At this point, due to earlier optimism, we may hold only an inpcb
936 * lock, and not the inpcbinfo write lock. If so, we need to try to
937 * acquire it, or if that fails, acquire a reference on the inpcb,
938 * drop all locks, acquire a global write lock, and then re-acquire
939 * the inpcb lock. We may at that point discover that another thread
940 * has tried to free the inpcb, in which case we need to loop back
941 * and try to find a new inpcb to deliver to.
943 * XXXRW: It may be time to rethink timewait locking.
945 if (inp->inp_flags & INP_TIMEWAIT) {
946 if (ti_locked == TI_UNLOCKED) {
947 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
948 ti_locked = TI_RLOCKED;
950 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
952 if (thflags & TH_SYN)
953 tcp_dooptions(&to, optp, optlen, TO_SYN);
955 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
957 if (tcp_twcheck(inp, &to, th, m, tlen))
959 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
960 return (IPPROTO_DONE);
963 * The TCPCB may no longer exist if the connection is winding
964 * down or it is in the CLOSED state. Either way we drop the
965 * segment and send an appropriate response.
968 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
969 rstreason = BANDLIM_RST_CLOSEDPORT;
974 if (tp->t_flags & TF_TOE) {
975 tcp_offload_input(tp, m);
976 m = NULL; /* consumed by the TOE driver */
982 * We've identified a valid inpcb, but it could be that we need an
983 * inpcbinfo write lock but don't hold it. In this case, attempt to
984 * acquire using the same strategy as the TIMEWAIT case above. If we
985 * relock, we have to jump back to 'relocked' as the connection might
986 * now be in TIMEWAIT.
989 if ((thflags & (TH_FIN | TH_RST)) != 0)
990 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
992 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) ||
993 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) &&
994 !IS_FASTOPEN(tp->t_flags)))) {
995 if (ti_locked == TI_UNLOCKED) {
996 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
997 ti_locked = TI_RLOCKED;
999 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1003 INP_WLOCK_ASSERT(inp);
1004 if (mac_inpcb_check_deliver(inp, m))
1007 so = inp->inp_socket;
1008 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1010 if (so->so_options & SO_DEBUG) {
1011 ostate = tp->t_state;
1014 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1017 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1020 #endif /* TCPDEBUG */
1022 * When the socket is accepting connections (the INPCB is in LISTEN
1023 * state) we look into the SYN cache if this is a new connection
1024 * attempt or the completion of a previous one.
1026 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1027 ("%s: so accepting but tp %p not listening", __func__, tp));
1028 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1029 struct in_conninfo inc;
1031 bzero(&inc, sizeof(inc));
1034 inc.inc_flags |= INC_ISIPV6;
1035 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1036 inc.inc_flags |= INC_IPV6MINMTU;
1037 inc.inc6_faddr = ip6->ip6_src;
1038 inc.inc6_laddr = ip6->ip6_dst;
1042 inc.inc_faddr = ip->ip_src;
1043 inc.inc_laddr = ip->ip_dst;
1045 inc.inc_fport = th->th_sport;
1046 inc.inc_lport = th->th_dport;
1047 inc.inc_fibnum = so->so_fibnum;
1050 * Check for an existing connection attempt in syncache if
1051 * the flag is only ACK. A successful lookup creates a new
1052 * socket appended to the listen queue in SYN_RECEIVED state.
1054 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1056 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1058 * Parse the TCP options here because
1059 * syncookies need access to the reflected
1062 tcp_dooptions(&to, optp, optlen, 0);
1064 * NB: syncache_expand() doesn't unlock
1065 * inp and tcpinfo locks.
1067 rstreason = syncache_expand(&inc, &to, th, &so, m);
1068 if (rstreason < 0) {
1070 * A failing TCP MD5 signature comparison
1071 * must result in the segment being dropped
1072 * and must not produce any response back
1076 } else if (rstreason == 0) {
1078 * No syncache entry or ACK was not
1079 * for our SYN/ACK. Send a RST.
1080 * NB: syncache did its own logging
1081 * of the failure cause.
1083 rstreason = BANDLIM_RST_OPENPORT;
1089 * We completed the 3-way handshake
1090 * but could not allocate a socket
1091 * either due to memory shortage,
1092 * listen queue length limits or
1093 * global socket limits. Send RST
1094 * or wait and have the remote end
1095 * retransmit the ACK for another
1098 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1099 log(LOG_DEBUG, "%s; %s: Listen socket: "
1100 "Socket allocation failed due to "
1101 "limits or memory shortage, %s\n",
1103 V_tcp_sc_rst_sock_fail ?
1104 "sending RST" : "try again");
1105 if (V_tcp_sc_rst_sock_fail) {
1106 rstreason = BANDLIM_UNLIMITED;
1112 * Socket is created in state SYN_RECEIVED.
1113 * Unlock the listen socket, lock the newly
1114 * created socket and update the tp variable.
1116 INP_WUNLOCK(inp); /* listen socket */
1117 inp = sotoinpcb(so);
1119 * New connection inpcb is already locked by
1120 * syncache_expand().
1122 INP_WLOCK_ASSERT(inp);
1123 tp = intotcpcb(inp);
1124 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1125 ("%s: ", __func__));
1127 * Process the segment and the data it
1128 * contains. tcp_do_segment() consumes
1129 * the mbuf chain and unlocks the inpcb.
1131 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1132 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1134 if (ti_locked == TI_RLOCKED)
1135 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1136 return (IPPROTO_DONE);
1139 * Segment flag validation for new connection attempts:
1141 * Our (SYN|ACK) response was rejected.
1142 * Check with syncache and remove entry to prevent
1145 * NB: syncache_chkrst does its own logging of failure
1148 if (thflags & TH_RST) {
1149 syncache_chkrst(&inc, th, m);
1153 * We can't do anything without SYN.
1155 if ((thflags & TH_SYN) == 0) {
1156 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1157 log(LOG_DEBUG, "%s; %s: Listen socket: "
1158 "SYN is missing, segment ignored\n",
1160 TCPSTAT_INC(tcps_badsyn);
1164 * (SYN|ACK) is bogus on a listen socket.
1166 if (thflags & TH_ACK) {
1167 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1168 log(LOG_DEBUG, "%s; %s: Listen socket: "
1169 "SYN|ACK invalid, segment rejected\n",
1171 syncache_badack(&inc); /* XXX: Not needed! */
1172 TCPSTAT_INC(tcps_badsyn);
1173 rstreason = BANDLIM_RST_OPENPORT;
1177 * If the drop_synfin option is enabled, drop all
1178 * segments with both the SYN and FIN bits set.
1179 * This prevents e.g. nmap from identifying the
1181 * XXX: Poor reasoning. nmap has other methods
1182 * and is constantly refining its stack detection
1184 * XXX: This is a violation of the TCP specification
1185 * and was used by RFC1644.
1187 if ((thflags & TH_FIN) && V_drop_synfin) {
1188 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1189 log(LOG_DEBUG, "%s; %s: Listen socket: "
1190 "SYN|FIN segment ignored (based on "
1191 "sysctl setting)\n", s, __func__);
1192 TCPSTAT_INC(tcps_badsyn);
1196 * Segment's flags are (SYN) or (SYN|FIN).
1198 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1199 * as they do not affect the state of the TCP FSM.
1200 * The data pointed to by TH_URG and th_urp is ignored.
1202 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1203 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1204 KASSERT(thflags & (TH_SYN),
1205 ("%s: Listen socket: TH_SYN not set", __func__));
1208 * If deprecated address is forbidden,
1209 * we do not accept SYN to deprecated interface
1210 * address to prevent any new inbound connection from
1211 * getting established.
1212 * When we do not accept SYN, we send a TCP RST,
1213 * with deprecated source address (instead of dropping
1214 * it). We compromise it as it is much better for peer
1215 * to send a RST, and RST will be the final packet
1218 * If we do not forbid deprecated addresses, we accept
1219 * the SYN packet. RFC2462 does not suggest dropping
1221 * If we decipher RFC2462 5.5.4, it says like this:
1222 * 1. use of deprecated addr with existing
1223 * communication is okay - "SHOULD continue to be
1225 * 2. use of it with new communication:
1226 * (2a) "SHOULD NOT be used if alternate address
1227 * with sufficient scope is available"
1228 * (2b) nothing mentioned otherwise.
1229 * Here we fall into (2b) case as we have no choice in
1230 * our source address selection - we must obey the peer.
1232 * The wording in RFC2462 is confusing, and there are
1233 * multiple description text for deprecated address
1234 * handling - worse, they are not exactly the same.
1235 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1237 if (isipv6 && !V_ip6_use_deprecated) {
1238 struct in6_ifaddr *ia6;
1240 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1242 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1243 ifa_free(&ia6->ia_ifa);
1244 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1245 log(LOG_DEBUG, "%s; %s: Listen socket: "
1246 "Connection attempt to deprecated "
1247 "IPv6 address rejected\n",
1249 rstreason = BANDLIM_RST_OPENPORT;
1253 ifa_free(&ia6->ia_ifa);
1257 * Basic sanity checks on incoming SYN requests:
1258 * Don't respond if the destination is a link layer
1259 * broadcast according to RFC1122 4.2.3.10, p. 104.
1260 * If it is from this socket it must be forged.
1261 * Don't respond if the source or destination is a
1262 * global or subnet broad- or multicast address.
1263 * Note that it is quite possible to receive unicast
1264 * link-layer packets with a broadcast IP address. Use
1265 * in_broadcast() to find them.
1267 if (m->m_flags & (M_BCAST|M_MCAST)) {
1268 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1269 log(LOG_DEBUG, "%s; %s: Listen socket: "
1270 "Connection attempt from broad- or multicast "
1271 "link layer address ignored\n", s, __func__);
1276 if (th->th_dport == th->th_sport &&
1277 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1278 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1279 log(LOG_DEBUG, "%s; %s: Listen socket: "
1280 "Connection attempt to/from self "
1281 "ignored\n", s, __func__);
1284 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1285 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1286 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1287 log(LOG_DEBUG, "%s; %s: Listen socket: "
1288 "Connection attempt from/to multicast "
1289 "address ignored\n", s, __func__);
1294 #if defined(INET) && defined(INET6)
1299 if (th->th_dport == th->th_sport &&
1300 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1301 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1302 log(LOG_DEBUG, "%s; %s: Listen socket: "
1303 "Connection attempt from/to self "
1304 "ignored\n", s, __func__);
1307 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1308 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1309 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1310 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1311 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1312 log(LOG_DEBUG, "%s; %s: Listen socket: "
1313 "Connection attempt from/to broad- "
1314 "or multicast address ignored\n",
1321 * SYN appears to be valid. Create compressed TCP state
1325 if (so->so_options & SO_DEBUG)
1326 tcp_trace(TA_INPUT, ostate, tp,
1327 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1329 TCP_PROBE3(debug__input, tp, th, m);
1330 tcp_dooptions(&to, optp, optlen, TO_SYN);
1331 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL))
1332 goto tfo_socket_result;
1335 * Entry added to syncache and mbuf consumed.
1336 * Only the listen socket is unlocked by syncache_add().
1338 if (ti_locked == TI_RLOCKED) {
1339 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1340 ti_locked = TI_UNLOCKED;
1342 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1343 return (IPPROTO_DONE);
1344 } else if (tp->t_state == TCPS_LISTEN) {
1346 * When a listen socket is torn down the SO_ACCEPTCONN
1347 * flag is removed first while connections are drained
1348 * from the accept queue in a unlock/lock cycle of the
1349 * ACCEPT_LOCK, opening a race condition allowing a SYN
1350 * attempt go through unhandled.
1354 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1355 if (tp->t_flags & TF_SIGNATURE) {
1356 tcp_dooptions(&to, optp, optlen, thflags);
1357 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1358 TCPSTAT_INC(tcps_sig_err_nosigopt);
1361 if (!TCPMD5_ENABLED() ||
1362 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1366 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1369 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1370 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1371 * the inpcb, and unlocks pcbinfo.
1373 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1374 if (ti_locked == TI_RLOCKED)
1375 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1376 return (IPPROTO_DONE);
1379 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1381 if (ti_locked == TI_RLOCKED) {
1382 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1383 ti_locked = TI_UNLOCKED;
1387 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1388 "ti_locked: %d", __func__, ti_locked));
1389 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1394 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1397 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1398 m = NULL; /* mbuf chain got consumed. */
1403 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1405 if (ti_locked == TI_RLOCKED) {
1406 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1407 ti_locked = TI_UNLOCKED;
1411 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1412 "ti_locked: %d", __func__, ti_locked));
1413 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1421 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1426 return (IPPROTO_DONE);
1430 * Automatic sizing of receive socket buffer. Often the send
1431 * buffer size is not optimally adjusted to the actual network
1432 * conditions at hand (delay bandwidth product). Setting the
1433 * buffer size too small limits throughput on links with high
1434 * bandwidth and high delay (eg. trans-continental/oceanic links).
1436 * On the receive side the socket buffer memory is only rarely
1437 * used to any significant extent. This allows us to be much
1438 * more aggressive in scaling the receive socket buffer. For
1439 * the case that the buffer space is actually used to a large
1440 * extent and we run out of kernel memory we can simply drop
1441 * the new segments; TCP on the sender will just retransmit it
1442 * later. Setting the buffer size too big may only consume too
1443 * much kernel memory if the application doesn't read() from
1444 * the socket or packet loss or reordering makes use of the
1447 * The criteria to step up the receive buffer one notch are:
1448 * 1. Application has not set receive buffer size with
1449 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1450 * 2. the number of bytes received during 1/2 of an sRTT
1451 * is at least 3/8 of the current socket buffer size.
1452 * 3. receive buffer size has not hit maximal automatic size;
1454 * If all of the criteria are met we increaset the socket buffer
1455 * by a 1/2 (bounded by the max). This allows us to keep ahead
1456 * of slow-start but also makes it so our peer never gets limited
1457 * by our rwnd which we then open up causing a burst.
1459 * This algorithm does two steps per RTT at most and only if
1460 * we receive a bulk stream w/o packet losses or reorderings.
1461 * Shrinking the buffer during idle times is not necessary as
1462 * it doesn't consume any memory when idle.
1464 * TODO: Only step up if the application is actually serving
1465 * the buffer to better manage the socket buffer resources.
1468 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1469 struct tcpcb *tp, int tlen)
1473 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1474 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1475 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1476 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1477 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1478 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1479 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1481 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1483 /* Start over with next RTT. */
1487 tp->rfbuf_cnt += tlen; /* add up */
1494 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1495 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1497 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1498 int rstreason, todrop, win;
1502 struct in_conninfo *inc;
1509 * The size of tcp_saveipgen must be the size of the max ip header,
1512 u_char tcp_saveipgen[IP6_HDR_LEN];
1513 struct tcphdr tcp_savetcp;
1516 thflags = th->th_flags;
1517 inc = &tp->t_inpcb->inp_inc;
1518 tp->sackhint.last_sack_ack = 0;
1520 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1522 * If this is either a state-changing packet or current state isn't
1523 * established, we require a write lock on tcbinfo. Otherwise, we
1524 * allow the tcbinfo to be in either alocked or unlocked, as the
1525 * caller may have unnecessarily acquired a write lock due to a race.
1527 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1528 tp->t_state != TCPS_ESTABLISHED) {
1529 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1531 INP_WLOCK_ASSERT(tp->t_inpcb);
1532 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1534 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1538 /* Save segment, if requested. */
1539 tcp_pcap_add(th, m, &(tp->t_inpkts));
1541 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1544 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1545 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1546 log(LOG_DEBUG, "%s; %s: "
1547 "SYN|FIN segment ignored (based on "
1548 "sysctl setting)\n", s, __func__);
1555 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1556 * check SEQ.ACK first.
1558 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1559 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1560 rstreason = BANDLIM_UNLIMITED;
1565 * Segment received on connection.
1566 * Reset idle time and keep-alive timer.
1567 * XXX: This should be done after segment
1568 * validation to ignore broken/spoofed segs.
1570 tp->t_rcvtime = ticks;
1573 * Scale up the window into a 32-bit value.
1574 * For the SYN_SENT state the scale is zero.
1576 tiwin = th->th_win << tp->snd_scale;
1579 * TCP ECN processing.
1581 if (tp->t_flags & TF_ECN_PERMIT) {
1582 if (thflags & TH_CWR)
1583 tp->t_flags &= ~TF_ECN_SND_ECE;
1584 switch (iptos & IPTOS_ECN_MASK) {
1586 tp->t_flags |= TF_ECN_SND_ECE;
1587 TCPSTAT_INC(tcps_ecn_ce);
1589 case IPTOS_ECN_ECT0:
1590 TCPSTAT_INC(tcps_ecn_ect0);
1592 case IPTOS_ECN_ECT1:
1593 TCPSTAT_INC(tcps_ecn_ect1);
1597 /* Process a packet differently from RFC3168. */
1598 cc_ecnpkt_handler(tp, th, iptos);
1600 /* Congestion experienced. */
1601 if (thflags & TH_ECE) {
1602 cc_cong_signal(tp, th, CC_ECN);
1607 * Parse options on any incoming segment.
1609 tcp_dooptions(&to, (u_char *)(th + 1),
1610 (th->th_off << 2) - sizeof(struct tcphdr),
1611 (thflags & TH_SYN) ? TO_SYN : 0);
1613 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1614 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1615 (to.to_flags & TOF_SIGNATURE) == 0) {
1616 TCPSTAT_INC(tcps_sig_err_sigopt);
1617 /* XXX: should drop? */
1621 * If echoed timestamp is later than the current time,
1622 * fall back to non RFC1323 RTT calculation. Normalize
1623 * timestamp if syncookies were used when this connection
1626 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1627 to.to_tsecr -= tp->ts_offset;
1628 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1630 else if (tp->t_flags & TF_PREVVALID &&
1631 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1632 cc_cong_signal(tp, th, CC_RTO_ERR);
1635 * Process options only when we get SYN/ACK back. The SYN case
1636 * for incoming connections is handled in tcp_syncache.
1637 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1638 * or <SYN,ACK>) segment itself is never scaled.
1639 * XXX this is traditional behavior, may need to be cleaned up.
1641 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1642 if ((to.to_flags & TOF_SCALE) &&
1643 (tp->t_flags & TF_REQ_SCALE)) {
1644 tp->t_flags |= TF_RCVD_SCALE;
1645 tp->snd_scale = to.to_wscale;
1648 * Initial send window. It will be updated with
1649 * the next incoming segment to the scaled value.
1651 tp->snd_wnd = th->th_win;
1652 if (to.to_flags & TOF_TS) {
1653 tp->t_flags |= TF_RCVD_TSTMP;
1654 tp->ts_recent = to.to_tsval;
1655 tp->ts_recent_age = tcp_ts_getticks();
1657 if (to.to_flags & TOF_MSS)
1658 tcp_mss(tp, to.to_mss);
1659 if ((tp->t_flags & TF_SACK_PERMIT) &&
1660 (to.to_flags & TOF_SACKPERM) == 0)
1661 tp->t_flags &= ~TF_SACK_PERMIT;
1662 if (IS_FASTOPEN(tp->t_flags)) {
1663 if (to.to_flags & TOF_FASTOPEN) {
1666 if (to.to_flags & TOF_MSS)
1669 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1673 tcp_fastopen_update_cache(tp, mss,
1674 to.to_tfo_len, to.to_tfo_cookie);
1676 tcp_fastopen_disable_path(tp);
1681 * If timestamps were negotiated during SYN/ACK they should
1682 * appear on every segment during this session and vice versa.
1684 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1685 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1686 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1687 "no action\n", s, __func__);
1691 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1692 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1693 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1694 "no action\n", s, __func__);
1700 * Header prediction: check for the two common cases
1701 * of a uni-directional data xfer. If the packet has
1702 * no control flags, is in-sequence, the window didn't
1703 * change and we're not retransmitting, it's a
1704 * candidate. If the length is zero and the ack moved
1705 * forward, we're the sender side of the xfer. Just
1706 * free the data acked & wake any higher level process
1707 * that was blocked waiting for space. If the length
1708 * is non-zero and the ack didn't move, we're the
1709 * receiver side. If we're getting packets in-order
1710 * (the reassembly queue is empty), add the data to
1711 * the socket buffer and note that we need a delayed ack.
1712 * Make sure that the hidden state-flags are also off.
1713 * Since we check for TCPS_ESTABLISHED first, it can only
1716 if (tp->t_state == TCPS_ESTABLISHED &&
1717 th->th_seq == tp->rcv_nxt &&
1718 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1719 tp->snd_nxt == tp->snd_max &&
1720 tiwin && tiwin == tp->snd_wnd &&
1721 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1723 ((to.to_flags & TOF_TS) == 0 ||
1724 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1727 * If last ACK falls within this segment's sequence numbers,
1728 * record the timestamp.
1729 * NOTE that the test is modified according to the latest
1730 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1732 if ((to.to_flags & TOF_TS) != 0 &&
1733 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1734 tp->ts_recent_age = tcp_ts_getticks();
1735 tp->ts_recent = to.to_tsval;
1739 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1740 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1741 !IN_RECOVERY(tp->t_flags) &&
1742 (to.to_flags & TOF_SACK) == 0 &&
1743 TAILQ_EMPTY(&tp->snd_holes)) {
1745 * This is a pure ack for outstanding data.
1747 TCPSTAT_INC(tcps_predack);
1750 * "bad retransmit" recovery without timestamps.
1752 if ((to.to_flags & TOF_TS) == 0 &&
1753 tp->t_rxtshift == 1 &&
1754 tp->t_flags & TF_PREVVALID &&
1755 (int)(ticks - tp->t_badrxtwin) < 0) {
1756 cc_cong_signal(tp, th, CC_RTO_ERR);
1760 * Recalculate the transmit timer / rtt.
1762 * Some boxes send broken timestamp replies
1763 * during the SYN+ACK phase, ignore
1764 * timestamps of 0 or we could calculate a
1765 * huge RTT and blow up the retransmit timer.
1767 if ((to.to_flags & TOF_TS) != 0 &&
1771 t = tcp_ts_getticks() - to.to_tsecr;
1772 if (!tp->t_rttlow || tp->t_rttlow > t)
1775 TCP_TS_TO_TICKS(t) + 1);
1776 } else if (tp->t_rtttime &&
1777 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1778 if (!tp->t_rttlow ||
1779 tp->t_rttlow > ticks - tp->t_rtttime)
1780 tp->t_rttlow = ticks - tp->t_rtttime;
1782 ticks - tp->t_rtttime);
1784 acked = BYTES_THIS_ACK(tp, th);
1787 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1788 hhook_run_tcp_est_in(tp, th, &to);
1791 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1792 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1793 sbdrop(&so->so_snd, acked);
1794 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1795 SEQ_LEQ(th->th_ack, tp->snd_recover))
1796 tp->snd_recover = th->th_ack - 1;
1799 * Let the congestion control algorithm update
1800 * congestion control related information. This
1801 * typically means increasing the congestion
1804 cc_ack_received(tp, th, nsegs, CC_ACK);
1806 tp->snd_una = th->th_ack;
1808 * Pull snd_wl2 up to prevent seq wrap relative
1811 tp->snd_wl2 = th->th_ack;
1816 * If all outstanding data are acked, stop
1817 * retransmit timer, otherwise restart timer
1818 * using current (possibly backed-off) value.
1819 * If process is waiting for space,
1820 * wakeup/selwakeup/signal. If data
1821 * are ready to send, let tcp_output
1822 * decide between more output or persist.
1825 if (so->so_options & SO_DEBUG)
1826 tcp_trace(TA_INPUT, ostate, tp,
1827 (void *)tcp_saveipgen,
1830 TCP_PROBE3(debug__input, tp, th, m);
1831 if (tp->snd_una == tp->snd_max)
1832 tcp_timer_activate(tp, TT_REXMT, 0);
1833 else if (!tcp_timer_active(tp, TT_PERSIST))
1834 tcp_timer_activate(tp, TT_REXMT,
1837 if (sbavail(&so->so_snd))
1838 (void) tp->t_fb->tfb_tcp_output(tp);
1841 } else if (th->th_ack == tp->snd_una &&
1842 tlen <= sbspace(&so->so_rcv)) {
1843 int newsize = 0; /* automatic sockbuf scaling */
1846 * This is a pure, in-sequence data packet with
1847 * nothing on the reassembly queue and we have enough
1848 * buffer space to take it.
1850 /* Clean receiver SACK report if present */
1851 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1852 tcp_clean_sackreport(tp);
1853 TCPSTAT_INC(tcps_preddat);
1854 tp->rcv_nxt += tlen;
1856 * Pull snd_wl1 up to prevent seq wrap relative to
1859 tp->snd_wl1 = th->th_seq;
1861 * Pull rcv_up up to prevent seq wrap relative to
1864 tp->rcv_up = tp->rcv_nxt;
1865 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1866 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1868 if (so->so_options & SO_DEBUG)
1869 tcp_trace(TA_INPUT, ostate, tp,
1870 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1872 TCP_PROBE3(debug__input, tp, th, m);
1874 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1876 /* Add data to socket buffer. */
1877 SOCKBUF_LOCK(&so->so_rcv);
1878 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1882 * Set new socket buffer size.
1883 * Give up when limit is reached.
1886 if (!sbreserve_locked(&so->so_rcv,
1888 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1889 m_adj(m, drop_hdrlen); /* delayed header drop */
1890 sbappendstream_locked(&so->so_rcv, m, 0);
1892 /* NB: sorwakeup_locked() does an implicit unlock. */
1893 sorwakeup_locked(so);
1894 if (DELAY_ACK(tp, tlen)) {
1895 tp->t_flags |= TF_DELACK;
1897 tp->t_flags |= TF_ACKNOW;
1898 tp->t_fb->tfb_tcp_output(tp);
1905 * Calculate amount of space in receive window,
1906 * and then do TCP input processing.
1907 * Receive window is amount of space in rcv queue,
1908 * but not less than advertised window.
1910 win = sbspace(&so->so_rcv);
1913 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1915 switch (tp->t_state) {
1918 * If the state is SYN_RECEIVED:
1919 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1921 case TCPS_SYN_RECEIVED:
1922 if ((thflags & TH_ACK) &&
1923 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1924 SEQ_GT(th->th_ack, tp->snd_max))) {
1925 rstreason = BANDLIM_RST_OPENPORT;
1928 if (IS_FASTOPEN(tp->t_flags)) {
1930 * When a TFO connection is in SYN_RECEIVED, the
1931 * only valid packets are the initial SYN, a
1932 * retransmit/copy of the initial SYN (possibly with
1933 * a subset of the original data), a valid ACK, a
1936 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1937 rstreason = BANDLIM_RST_OPENPORT;
1939 } else if (thflags & TH_SYN) {
1940 /* non-initial SYN is ignored */
1941 if ((tcp_timer_active(tp, TT_DELACK) ||
1942 tcp_timer_active(tp, TT_REXMT)))
1944 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1951 * If the state is SYN_SENT:
1952 * if seg contains a RST with valid ACK (SEQ.ACK has already
1953 * been verified), then drop the connection.
1954 * if seg contains a RST without an ACK, drop the seg.
1955 * if seg does not contain SYN, then drop the seg.
1956 * Otherwise this is an acceptable SYN segment
1957 * initialize tp->rcv_nxt and tp->irs
1958 * if seg contains ack then advance tp->snd_una
1959 * if seg contains an ECE and ECN support is enabled, the stream
1961 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1962 * arrange for segment to be acked (eventually)
1963 * continue processing rest of data/controls, beginning with URG
1966 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1967 TCP_PROBE5(connect__refused, NULL, tp,
1969 tp = tcp_drop(tp, ECONNREFUSED);
1971 if (thflags & TH_RST)
1973 if (!(thflags & TH_SYN))
1976 tp->irs = th->th_seq;
1978 if (thflags & TH_ACK) {
1979 int tfo_partial_ack = 0;
1981 TCPSTAT_INC(tcps_connects);
1984 mac_socketpeer_set_from_mbuf(m, so);
1986 /* Do window scaling on this connection? */
1987 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1988 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1989 tp->rcv_scale = tp->request_r_scale;
1991 tp->rcv_adv += min(tp->rcv_wnd,
1992 TCP_MAXWIN << tp->rcv_scale);
1993 tp->snd_una++; /* SYN is acked */
1995 * If not all the data that was sent in the TFO SYN
1996 * has been acked, resend the remainder right away.
1998 if (IS_FASTOPEN(tp->t_flags) &&
1999 (tp->snd_una != tp->snd_max)) {
2000 tp->snd_nxt = th->th_ack;
2001 tfo_partial_ack = 1;
2004 * If there's data, delay ACK; if there's also a FIN
2005 * ACKNOW will be turned on later.
2007 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2008 tcp_timer_activate(tp, TT_DELACK,
2011 tp->t_flags |= TF_ACKNOW;
2013 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2015 tp->t_flags |= TF_ECN_PERMIT;
2016 TCPSTAT_INC(tcps_ecn_shs);
2020 * Received <SYN,ACK> in SYN_SENT[*] state.
2022 * SYN_SENT --> ESTABLISHED
2023 * SYN_SENT* --> FIN_WAIT_1
2025 tp->t_starttime = ticks;
2026 if (tp->t_flags & TF_NEEDFIN) {
2027 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2028 tp->t_flags &= ~TF_NEEDFIN;
2031 tcp_state_change(tp, TCPS_ESTABLISHED);
2032 TCP_PROBE5(connect__established, NULL, tp,
2035 tcp_timer_activate(tp, TT_KEEP,
2040 * Received initial SYN in SYN-SENT[*] state =>
2041 * simultaneous open.
2042 * If it succeeds, connection is * half-synchronized.
2043 * Otherwise, do 3-way handshake:
2044 * SYN-SENT -> SYN-RECEIVED
2045 * SYN-SENT* -> SYN-RECEIVED*
2047 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2048 tcp_timer_activate(tp, TT_REXMT, 0);
2049 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2052 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2053 INP_WLOCK_ASSERT(tp->t_inpcb);
2056 * Advance th->th_seq to correspond to first data byte.
2057 * If data, trim to stay within window,
2058 * dropping FIN if necessary.
2061 if (tlen > tp->rcv_wnd) {
2062 todrop = tlen - tp->rcv_wnd;
2066 TCPSTAT_INC(tcps_rcvpackafterwin);
2067 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2069 tp->snd_wl1 = th->th_seq - 1;
2070 tp->rcv_up = th->th_seq;
2072 * Client side of transaction: already sent SYN and data.
2073 * If the remote host used T/TCP to validate the SYN,
2074 * our data will be ACK'd; if so, enter normal data segment
2075 * processing in the middle of step 5, ack processing.
2076 * Otherwise, goto step 6.
2078 if (thflags & TH_ACK)
2084 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2085 * do normal processing.
2087 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2091 break; /* continue normal processing */
2095 * States other than LISTEN or SYN_SENT.
2096 * First check the RST flag and sequence number since reset segments
2097 * are exempt from the timestamp and connection count tests. This
2098 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2099 * below which allowed reset segments in half the sequence space
2100 * to fall though and be processed (which gives forged reset
2101 * segments with a random sequence number a 50 percent chance of
2102 * killing a connection).
2103 * Then check timestamp, if present.
2104 * Then check the connection count, if present.
2105 * Then check that at least some bytes of segment are within
2106 * receive window. If segment begins before rcv_nxt,
2107 * drop leading data (and SYN); if nothing left, just ack.
2109 if (thflags & TH_RST) {
2111 * RFC5961 Section 3.2
2113 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2114 * - If RST is in window, we send challenge ACK.
2116 * Note: to take into account delayed ACKs, we should
2117 * test against last_ack_sent instead of rcv_nxt.
2118 * Note 2: we handle special case of closed window, not
2119 * covered by the RFC.
2121 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2122 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2123 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2125 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2126 KASSERT(tp->t_state != TCPS_SYN_SENT,
2127 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2130 if (V_tcp_insecure_rst ||
2131 tp->last_ack_sent == th->th_seq) {
2132 TCPSTAT_INC(tcps_drops);
2133 /* Drop the connection. */
2134 switch (tp->t_state) {
2135 case TCPS_SYN_RECEIVED:
2136 so->so_error = ECONNREFUSED;
2138 case TCPS_ESTABLISHED:
2139 case TCPS_FIN_WAIT_1:
2140 case TCPS_FIN_WAIT_2:
2141 case TCPS_CLOSE_WAIT:
2144 so->so_error = ECONNRESET;
2151 TCPSTAT_INC(tcps_badrst);
2152 /* Send challenge ACK. */
2153 tcp_respond(tp, mtod(m, void *), th, m,
2154 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2155 tp->last_ack_sent = tp->rcv_nxt;
2163 * RFC5961 Section 4.2
2164 * Send challenge ACK for any SYN in synchronized state.
2166 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2167 tp->t_state != TCPS_SYN_RECEIVED) {
2168 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2170 TCPSTAT_INC(tcps_badsyn);
2171 if (V_tcp_insecure_syn &&
2172 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2173 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2174 tp = tcp_drop(tp, ECONNRESET);
2175 rstreason = BANDLIM_UNLIMITED;
2177 /* Send challenge ACK. */
2178 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2179 tp->snd_nxt, TH_ACK);
2180 tp->last_ack_sent = tp->rcv_nxt;
2187 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2188 * and it's less than ts_recent, drop it.
2190 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2191 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2193 /* Check to see if ts_recent is over 24 days old. */
2194 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2196 * Invalidate ts_recent. If this segment updates
2197 * ts_recent, the age will be reset later and ts_recent
2198 * will get a valid value. If it does not, setting
2199 * ts_recent to zero will at least satisfy the
2200 * requirement that zero be placed in the timestamp
2201 * echo reply when ts_recent isn't valid. The
2202 * age isn't reset until we get a valid ts_recent
2203 * because we don't want out-of-order segments to be
2204 * dropped when ts_recent is old.
2208 TCPSTAT_INC(tcps_rcvduppack);
2209 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2210 TCPSTAT_INC(tcps_pawsdrop);
2218 * In the SYN-RECEIVED state, validate that the packet belongs to
2219 * this connection before trimming the data to fit the receive
2220 * window. Check the sequence number versus IRS since we know
2221 * the sequence numbers haven't wrapped. This is a partial fix
2222 * for the "LAND" DoS attack.
2224 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2225 rstreason = BANDLIM_RST_OPENPORT;
2229 todrop = tp->rcv_nxt - th->th_seq;
2231 if (thflags & TH_SYN) {
2241 * Following if statement from Stevens, vol. 2, p. 960.
2244 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2246 * Any valid FIN must be to the left of the window.
2247 * At this point the FIN must be a duplicate or out
2248 * of sequence; drop it.
2253 * Send an ACK to resynchronize and drop any data.
2254 * But keep on processing for RST or ACK.
2256 tp->t_flags |= TF_ACKNOW;
2258 TCPSTAT_INC(tcps_rcvduppack);
2259 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2261 TCPSTAT_INC(tcps_rcvpartduppack);
2262 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2265 * DSACK - add SACK block for dropped range
2267 if (tp->t_flags & TF_SACK_PERMIT) {
2268 tcp_update_sack_list(tp, th->th_seq, th->th_seq+tlen);
2270 * ACK now, as the next in-sequence segment
2271 * will clear the DSACK block again
2273 tp->t_flags |= TF_ACKNOW;
2275 drop_hdrlen += todrop; /* drop from the top afterwards */
2276 th->th_seq += todrop;
2278 if (th->th_urp > todrop)
2279 th->th_urp -= todrop;
2287 * If new data are received on a connection after the
2288 * user processes are gone, then RST the other end.
2290 if ((so->so_state & SS_NOFDREF) &&
2291 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2292 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2294 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2295 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2296 "after socket was closed, "
2297 "sending RST and removing tcpcb\n",
2298 s, __func__, tcpstates[tp->t_state], tlen);
2302 TCPSTAT_INC(tcps_rcvafterclose);
2303 rstreason = BANDLIM_UNLIMITED;
2308 * If segment ends after window, drop trailing data
2309 * (and PUSH and FIN); if nothing left, just ACK.
2311 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2313 TCPSTAT_INC(tcps_rcvpackafterwin);
2314 if (todrop >= tlen) {
2315 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2317 * If window is closed can only take segments at
2318 * window edge, and have to drop data and PUSH from
2319 * incoming segments. Continue processing, but
2320 * remember to ack. Otherwise, drop segment
2323 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2324 tp->t_flags |= TF_ACKNOW;
2325 TCPSTAT_INC(tcps_rcvwinprobe);
2329 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2332 thflags &= ~(TH_PUSH|TH_FIN);
2336 * If last ACK falls within this segment's sequence numbers,
2337 * record its timestamp.
2339 * 1) That the test incorporates suggestions from the latest
2340 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2341 * 2) That updating only on newer timestamps interferes with
2342 * our earlier PAWS tests, so this check should be solely
2343 * predicated on the sequence space of this segment.
2344 * 3) That we modify the segment boundary check to be
2345 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2346 * instead of RFC1323's
2347 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2348 * This modified check allows us to overcome RFC1323's
2349 * limitations as described in Stevens TCP/IP Illustrated
2350 * Vol. 2 p.869. In such cases, we can still calculate the
2351 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2353 if ((to.to_flags & TOF_TS) != 0 &&
2354 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2355 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2356 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2357 tp->ts_recent_age = tcp_ts_getticks();
2358 tp->ts_recent = to.to_tsval;
2362 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2363 * flag is on (half-synchronized state), then queue data for
2364 * later processing; else drop segment and return.
2366 if ((thflags & TH_ACK) == 0) {
2367 if (tp->t_state == TCPS_SYN_RECEIVED ||
2368 (tp->t_flags & TF_NEEDSYN)) {
2369 if (tp->t_state == TCPS_SYN_RECEIVED &&
2370 IS_FASTOPEN(tp->t_flags)) {
2371 tp->snd_wnd = tiwin;
2375 } else if (tp->t_flags & TF_ACKNOW)
2384 switch (tp->t_state) {
2387 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2388 * ESTABLISHED state and continue processing.
2389 * The ACK was checked above.
2391 case TCPS_SYN_RECEIVED:
2393 TCPSTAT_INC(tcps_connects);
2395 /* Do window scaling? */
2396 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2397 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2398 tp->rcv_scale = tp->request_r_scale;
2400 tp->snd_wnd = tiwin;
2403 * SYN-RECEIVED -> ESTABLISHED
2404 * SYN-RECEIVED* -> FIN-WAIT-1
2406 tp->t_starttime = ticks;
2407 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2408 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2409 tp->t_tfo_pending = NULL;
2412 * Account for the ACK of our SYN prior to
2413 * regular ACK processing below.
2417 if (tp->t_flags & TF_NEEDFIN) {
2418 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2419 tp->t_flags &= ~TF_NEEDFIN;
2421 tcp_state_change(tp, TCPS_ESTABLISHED);
2422 TCP_PROBE5(accept__established, NULL, tp,
2425 * TFO connections call cc_conn_init() during SYN
2426 * processing. Calling it again here for such
2427 * connections is not harmless as it would undo the
2428 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2431 if (!IS_FASTOPEN(tp->t_flags))
2433 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2436 * If segment contains data or ACK, will call tcp_reass()
2437 * later; if not, do so now to pass queued data to user.
2439 if (tlen == 0 && (thflags & TH_FIN) == 0)
2440 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2442 tp->snd_wl1 = th->th_seq - 1;
2446 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2447 * ACKs. If the ack is in the range
2448 * tp->snd_una < th->th_ack <= tp->snd_max
2449 * then advance tp->snd_una to th->th_ack and drop
2450 * data from the retransmission queue. If this ACK reflects
2451 * more up to date window information we update our window information.
2453 case TCPS_ESTABLISHED:
2454 case TCPS_FIN_WAIT_1:
2455 case TCPS_FIN_WAIT_2:
2456 case TCPS_CLOSE_WAIT:
2459 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2460 TCPSTAT_INC(tcps_rcvacktoomuch);
2463 if ((tp->t_flags & TF_SACK_PERMIT) &&
2464 ((to.to_flags & TOF_SACK) ||
2465 !TAILQ_EMPTY(&tp->snd_holes)))
2466 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2469 * Reset the value so that previous (valid) value
2470 * from the last ack with SACK doesn't get used.
2472 tp->sackhint.sacked_bytes = 0;
2475 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2476 hhook_run_tcp_est_in(tp, th, &to);
2479 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2482 maxseg = tcp_maxseg(tp);
2484 (tiwin == tp->snd_wnd ||
2485 (tp->t_flags & TF_SACK_PERMIT))) {
2487 * If this is the first time we've seen a
2488 * FIN from the remote, this is not a
2489 * duplicate and it needs to be processed
2490 * normally. This happens during a
2491 * simultaneous close.
2493 if ((thflags & TH_FIN) &&
2494 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2498 TCPSTAT_INC(tcps_rcvdupack);
2500 * If we have outstanding data (other than
2501 * a window probe), this is a completely
2502 * duplicate ack (ie, window info didn't
2503 * change and FIN isn't set),
2504 * the ack is the biggest we've
2505 * seen and we've seen exactly our rexmt
2506 * threshold of them, assume a packet
2507 * has been dropped and retransmit it.
2508 * Kludge snd_nxt & the congestion
2509 * window so we send only this one
2512 * We know we're losing at the current
2513 * window size so do congestion avoidance
2514 * (set ssthresh to half the current window
2515 * and pull our congestion window back to
2516 * the new ssthresh).
2518 * Dup acks mean that packets have left the
2519 * network (they're now cached at the receiver)
2520 * so bump cwnd by the amount in the receiver
2521 * to keep a constant cwnd packets in the
2524 * When using TCP ECN, notify the peer that
2525 * we reduced the cwnd.
2528 * Following 2 kinds of acks should not affect
2531 * 2) Acks with SACK but without any new SACK
2532 * information in them. These could result from
2533 * any anomaly in the network like a switch
2534 * duplicating packets or a possible DoS attack.
2536 if (th->th_ack != tp->snd_una ||
2537 ((tp->t_flags & TF_SACK_PERMIT) &&
2540 else if (!tcp_timer_active(tp, TT_REXMT))
2542 else if (++tp->t_dupacks > tcprexmtthresh ||
2543 IN_FASTRECOVERY(tp->t_flags)) {
2544 cc_ack_received(tp, th, nsegs,
2546 if ((tp->t_flags & TF_SACK_PERMIT) &&
2547 IN_FASTRECOVERY(tp->t_flags)) {
2551 * Compute the amount of data in flight first.
2552 * We can inject new data into the pipe iff
2553 * we have less than 1/2 the original window's
2554 * worth of data in flight.
2556 if (V_tcp_do_rfc6675_pipe)
2557 awnd = tcp_compute_pipe(tp);
2559 awnd = (tp->snd_nxt - tp->snd_fack) +
2560 tp->sackhint.sack_bytes_rexmit;
2562 if (awnd < tp->snd_ssthresh) {
2563 tp->snd_cwnd += maxseg;
2564 if (tp->snd_cwnd > tp->snd_ssthresh)
2565 tp->snd_cwnd = tp->snd_ssthresh;
2568 tp->snd_cwnd += maxseg;
2569 (void) tp->t_fb->tfb_tcp_output(tp);
2571 } else if (tp->t_dupacks == tcprexmtthresh) {
2572 tcp_seq onxt = tp->snd_nxt;
2575 * If we're doing sack, check to
2576 * see if we're already in sack
2577 * recovery. If we're not doing sack,
2578 * check to see if we're in newreno
2581 if (tp->t_flags & TF_SACK_PERMIT) {
2582 if (IN_FASTRECOVERY(tp->t_flags)) {
2587 if (SEQ_LEQ(th->th_ack,
2593 /* Congestion signal before ack. */
2594 cc_cong_signal(tp, th, CC_NDUPACK);
2595 cc_ack_received(tp, th, nsegs,
2597 tcp_timer_activate(tp, TT_REXMT, 0);
2599 if (tp->t_flags & TF_SACK_PERMIT) {
2601 tcps_sack_recovery_episode);
2602 tp->sack_newdata = tp->snd_nxt;
2603 tp->snd_cwnd = maxseg;
2604 (void) tp->t_fb->tfb_tcp_output(tp);
2607 tp->snd_nxt = th->th_ack;
2608 tp->snd_cwnd = maxseg;
2609 (void) tp->t_fb->tfb_tcp_output(tp);
2610 KASSERT(tp->snd_limited <= 2,
2611 ("%s: tp->snd_limited too big",
2613 tp->snd_cwnd = tp->snd_ssthresh +
2615 (tp->t_dupacks - tp->snd_limited);
2616 if (SEQ_GT(onxt, tp->snd_nxt))
2619 } else if (V_tcp_do_rfc3042) {
2621 * Process first and second duplicate
2622 * ACKs. Each indicates a segment
2623 * leaving the network, creating room
2624 * for more. Make sure we can send a
2625 * packet on reception of each duplicate
2626 * ACK by increasing snd_cwnd by one
2627 * segment. Restore the original
2628 * snd_cwnd after packet transmission.
2630 cc_ack_received(tp, th, nsegs,
2632 uint32_t oldcwnd = tp->snd_cwnd;
2633 tcp_seq oldsndmax = tp->snd_max;
2637 KASSERT(tp->t_dupacks == 1 ||
2639 ("%s: dupacks not 1 or 2",
2641 if (tp->t_dupacks == 1)
2642 tp->snd_limited = 0;
2644 (tp->snd_nxt - tp->snd_una) +
2645 (tp->t_dupacks - tp->snd_limited) *
2648 * Only call tcp_output when there
2649 * is new data available to be sent.
2650 * Otherwise we would send pure ACKs.
2652 SOCKBUF_LOCK(&so->so_snd);
2653 avail = sbavail(&so->so_snd) -
2654 (tp->snd_nxt - tp->snd_una);
2655 SOCKBUF_UNLOCK(&so->so_snd);
2657 (void) tp->t_fb->tfb_tcp_output(tp);
2658 sent = tp->snd_max - oldsndmax;
2659 if (sent > maxseg) {
2660 KASSERT((tp->t_dupacks == 2 &&
2661 tp->snd_limited == 0) ||
2662 (sent == maxseg + 1 &&
2663 tp->t_flags & TF_SENTFIN),
2664 ("%s: sent too much",
2666 tp->snd_limited = 2;
2667 } else if (sent > 0)
2669 tp->snd_cwnd = oldcwnd;
2676 * This ack is advancing the left edge, reset the
2681 * If this ack also has new SACK info, increment the
2682 * counter as per rfc6675.
2684 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed)
2688 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2689 ("%s: th_ack <= snd_una", __func__));
2692 * If the congestion window was inflated to account
2693 * for the other side's cached packets, retract it.
2695 if (IN_FASTRECOVERY(tp->t_flags)) {
2696 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2697 if (tp->t_flags & TF_SACK_PERMIT)
2698 tcp_sack_partialack(tp, th);
2700 tcp_newreno_partial_ack(tp, th);
2702 cc_post_recovery(tp, th);
2705 * If we reach this point, ACK is not a duplicate,
2706 * i.e., it ACKs something we sent.
2708 if (tp->t_flags & TF_NEEDSYN) {
2710 * T/TCP: Connection was half-synchronized, and our
2711 * SYN has been ACK'd (so connection is now fully
2712 * synchronized). Go to non-starred state,
2713 * increment snd_una for ACK of SYN, and check if
2714 * we can do window scaling.
2716 tp->t_flags &= ~TF_NEEDSYN;
2718 /* Do window scaling? */
2719 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2720 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2721 tp->rcv_scale = tp->request_r_scale;
2722 /* Send window already scaled. */
2727 INP_WLOCK_ASSERT(tp->t_inpcb);
2729 acked = BYTES_THIS_ACK(tp, th);
2730 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2731 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2732 tp->snd_una, th->th_ack, tp, m));
2733 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2734 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2737 * If we just performed our first retransmit, and the ACK
2738 * arrives within our recovery window, then it was a mistake
2739 * to do the retransmit in the first place. Recover our
2740 * original cwnd and ssthresh, and proceed to transmit where
2743 if (tp->t_rxtshift == 1 &&
2744 tp->t_flags & TF_PREVVALID &&
2746 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2747 cc_cong_signal(tp, th, CC_RTO_ERR);
2750 * If we have a timestamp reply, update smoothed
2751 * round trip time. If no timestamp is present but
2752 * transmit timer is running and timed sequence
2753 * number was acked, update smoothed round trip time.
2754 * Since we now have an rtt measurement, cancel the
2755 * timer backoff (cf., Phil Karn's retransmit alg.).
2756 * Recompute the initial retransmit timer.
2758 * Some boxes send broken timestamp replies
2759 * during the SYN+ACK phase, ignore
2760 * timestamps of 0 or we could calculate a
2761 * huge RTT and blow up the retransmit timer.
2763 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2766 t = tcp_ts_getticks() - to.to_tsecr;
2767 if (!tp->t_rttlow || tp->t_rttlow > t)
2769 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2770 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2771 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2772 tp->t_rttlow = ticks - tp->t_rtttime;
2773 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2777 * If all outstanding data is acked, stop retransmit
2778 * timer and remember to restart (more output or persist).
2779 * If there is more data to be acked, restart retransmit
2780 * timer, using current (possibly backed-off) value.
2782 if (th->th_ack == tp->snd_max) {
2783 tcp_timer_activate(tp, TT_REXMT, 0);
2785 } else if (!tcp_timer_active(tp, TT_PERSIST))
2786 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2789 * If no data (only SYN) was ACK'd,
2790 * skip rest of ACK processing.
2796 * Let the congestion control algorithm update congestion
2797 * control related information. This typically means increasing
2798 * the congestion window.
2800 cc_ack_received(tp, th, nsegs, CC_ACK);
2802 SOCKBUF_LOCK(&so->so_snd);
2803 if (acked > sbavail(&so->so_snd)) {
2804 if (tp->snd_wnd >= sbavail(&so->so_snd))
2805 tp->snd_wnd -= sbavail(&so->so_snd);
2808 mfree = sbcut_locked(&so->so_snd,
2809 (int)sbavail(&so->so_snd));
2812 mfree = sbcut_locked(&so->so_snd, acked);
2813 if (tp->snd_wnd >= (uint32_t) acked)
2814 tp->snd_wnd -= acked;
2819 /* NB: sowwakeup_locked() does an implicit unlock. */
2820 sowwakeup_locked(so);
2822 /* Detect una wraparound. */
2823 if (!IN_RECOVERY(tp->t_flags) &&
2824 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2825 SEQ_LEQ(th->th_ack, tp->snd_recover))
2826 tp->snd_recover = th->th_ack - 1;
2827 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2828 if (IN_RECOVERY(tp->t_flags) &&
2829 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2830 EXIT_RECOVERY(tp->t_flags);
2832 tp->snd_una = th->th_ack;
2833 if (tp->t_flags & TF_SACK_PERMIT) {
2834 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2835 tp->snd_recover = tp->snd_una;
2837 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2838 tp->snd_nxt = tp->snd_una;
2840 switch (tp->t_state) {
2843 * In FIN_WAIT_1 STATE in addition to the processing
2844 * for the ESTABLISHED state if our FIN is now acknowledged
2845 * then enter FIN_WAIT_2.
2847 case TCPS_FIN_WAIT_1:
2848 if (ourfinisacked) {
2850 * If we can't receive any more
2851 * data, then closing user can proceed.
2852 * Starting the timer is contrary to the
2853 * specification, but if we don't get a FIN
2854 * we'll hang forever.
2857 * we should release the tp also, and use a
2860 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2861 soisdisconnected(so);
2862 tcp_timer_activate(tp, TT_2MSL,
2863 (tcp_fast_finwait2_recycle ?
2864 tcp_finwait2_timeout :
2867 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2872 * In CLOSING STATE in addition to the processing for
2873 * the ESTABLISHED state if the ACK acknowledges our FIN
2874 * then enter the TIME-WAIT state, otherwise ignore
2878 if (ourfinisacked) {
2879 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2887 * In LAST_ACK, we may still be waiting for data to drain
2888 * and/or to be acked, as well as for the ack of our FIN.
2889 * If our FIN is now acknowledged, delete the TCB,
2890 * enter the closed state and return.
2893 if (ourfinisacked) {
2894 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2903 INP_WLOCK_ASSERT(tp->t_inpcb);
2906 * Update window information.
2907 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2909 if ((thflags & TH_ACK) &&
2910 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2911 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2912 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2913 /* keep track of pure window updates */
2915 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2916 TCPSTAT_INC(tcps_rcvwinupd);
2917 tp->snd_wnd = tiwin;
2918 tp->snd_wl1 = th->th_seq;
2919 tp->snd_wl2 = th->th_ack;
2920 if (tp->snd_wnd > tp->max_sndwnd)
2921 tp->max_sndwnd = tp->snd_wnd;
2926 * Process segments with URG.
2928 if ((thflags & TH_URG) && th->th_urp &&
2929 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2931 * This is a kludge, but if we receive and accept
2932 * random urgent pointers, we'll crash in
2933 * soreceive. It's hard to imagine someone
2934 * actually wanting to send this much urgent data.
2936 SOCKBUF_LOCK(&so->so_rcv);
2937 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2938 th->th_urp = 0; /* XXX */
2939 thflags &= ~TH_URG; /* XXX */
2940 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2941 goto dodata; /* XXX */
2944 * If this segment advances the known urgent pointer,
2945 * then mark the data stream. This should not happen
2946 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2947 * a FIN has been received from the remote side.
2948 * In these states we ignore the URG.
2950 * According to RFC961 (Assigned Protocols),
2951 * the urgent pointer points to the last octet
2952 * of urgent data. We continue, however,
2953 * to consider it to indicate the first octet
2954 * of data past the urgent section as the original
2955 * spec states (in one of two places).
2957 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2958 tp->rcv_up = th->th_seq + th->th_urp;
2959 so->so_oobmark = sbavail(&so->so_rcv) +
2960 (tp->rcv_up - tp->rcv_nxt) - 1;
2961 if (so->so_oobmark == 0)
2962 so->so_rcv.sb_state |= SBS_RCVATMARK;
2964 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2966 SOCKBUF_UNLOCK(&so->so_rcv);
2968 * Remove out of band data so doesn't get presented to user.
2969 * This can happen independent of advancing the URG pointer,
2970 * but if two URG's are pending at once, some out-of-band
2971 * data may creep in... ick.
2973 if (th->th_urp <= (uint32_t)tlen &&
2974 !(so->so_options & SO_OOBINLINE)) {
2975 /* hdr drop is delayed */
2976 tcp_pulloutofband(so, th, m, drop_hdrlen);
2980 * If no out of band data is expected,
2981 * pull receive urgent pointer along
2982 * with the receive window.
2984 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2985 tp->rcv_up = tp->rcv_nxt;
2988 INP_WLOCK_ASSERT(tp->t_inpcb);
2991 * Process the segment text, merging it into the TCP sequencing queue,
2992 * and arranging for acknowledgment of receipt if necessary.
2993 * This process logically involves adjusting tp->rcv_wnd as data
2994 * is presented to the user (this happens in tcp_usrreq.c,
2995 * case PRU_RCVD). If a FIN has already been received on this
2996 * connection then we just ignore the text.
2998 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
2999 IS_FASTOPEN(tp->t_flags));
3000 if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
3001 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3002 tcp_seq save_start = th->th_seq;
3003 tcp_seq save_rnxt = tp->rcv_nxt;
3004 int save_tlen = tlen;
3005 m_adj(m, drop_hdrlen); /* delayed header drop */
3007 * Insert segment which includes th into TCP reassembly queue
3008 * with control block tp. Set thflags to whether reassembly now
3009 * includes a segment with FIN. This handles the common case
3010 * inline (segment is the next to be received on an established
3011 * connection, and the queue is empty), avoiding linkage into
3012 * and removal from the queue and repetition of various
3014 * Set DELACK for segments received in order, but ack
3015 * immediately when segments are out of order (so
3016 * fast retransmit can work).
3018 if (th->th_seq == tp->rcv_nxt &&
3020 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3022 if (DELAY_ACK(tp, tlen) || tfo_syn)
3023 tp->t_flags |= TF_DELACK;
3025 tp->t_flags |= TF_ACKNOW;
3026 tp->rcv_nxt += tlen;
3027 thflags = th->th_flags & TH_FIN;
3028 TCPSTAT_INC(tcps_rcvpack);
3029 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3030 SOCKBUF_LOCK(&so->so_rcv);
3031 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3034 sbappendstream_locked(&so->so_rcv, m, 0);
3035 /* NB: sorwakeup_locked() does an implicit unlock. */
3036 sorwakeup_locked(so);
3039 * XXX: Due to the header drop above "th" is
3040 * theoretically invalid by now. Fortunately
3041 * m_adj() doesn't actually frees any mbufs
3042 * when trimming from the head.
3044 tcp_seq temp = save_start;
3045 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3046 tp->t_flags |= TF_ACKNOW;
3048 if (tp->t_flags & TF_SACK_PERMIT) {
3049 if (((tlen == 0) && (save_tlen > 0) &&
3050 (SEQ_LT(save_start, save_rnxt)))) {
3052 * DSACK actually handled in the fastpath
3055 tcp_update_sack_list(tp, save_start, save_start + save_tlen);
3057 if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3059 * Cleaning sackblks by using zero length
3062 tcp_update_sack_list(tp, save_start, save_start);
3064 if ((tlen > 0) && (tlen >= save_tlen)) {
3065 /* Update of sackblks. */
3066 tcp_update_sack_list(tp, save_start, save_start + save_tlen);
3069 tcp_update_sack_list(tp, save_start, save_start+tlen);
3074 * Note the amount of data that peer has sent into
3075 * our window, in order to estimate the sender's
3079 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3080 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3082 len = so->so_rcv.sb_hiwat;
3090 * If FIN is received ACK the FIN and let the user know
3091 * that the connection is closing.
3093 if (thflags & TH_FIN) {
3094 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3097 * If connection is half-synchronized
3098 * (ie NEEDSYN flag on) then delay ACK,
3099 * so it may be piggybacked when SYN is sent.
3100 * Otherwise, since we received a FIN then no
3101 * more input can be expected, send ACK now.
3103 if (tp->t_flags & TF_NEEDSYN)
3104 tp->t_flags |= TF_DELACK;
3106 tp->t_flags |= TF_ACKNOW;
3109 switch (tp->t_state) {
3112 * In SYN_RECEIVED and ESTABLISHED STATES
3113 * enter the CLOSE_WAIT state.
3115 case TCPS_SYN_RECEIVED:
3116 tp->t_starttime = ticks;
3118 case TCPS_ESTABLISHED:
3119 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3123 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3124 * enter the CLOSING state.
3126 case TCPS_FIN_WAIT_1:
3127 tcp_state_change(tp, TCPS_CLOSING);
3131 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3132 * starting the time-wait timer, turning off the other
3135 case TCPS_FIN_WAIT_2:
3136 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
3143 if (so->so_options & SO_DEBUG)
3144 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3147 TCP_PROBE3(debug__input, tp, th, m);
3150 * Return any desired output.
3152 if (needoutput || (tp->t_flags & TF_ACKNOW))
3153 (void) tp->t_fb->tfb_tcp_output(tp);
3156 INP_WLOCK_ASSERT(tp->t_inpcb);
3158 if (tp->t_flags & TF_DELACK) {
3159 tp->t_flags &= ~TF_DELACK;
3160 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3162 INP_WUNLOCK(tp->t_inpcb);
3167 * Generate an ACK dropping incoming segment if it occupies
3168 * sequence space, where the ACK reflects our state.
3170 * We can now skip the test for the RST flag since all
3171 * paths to this code happen after packets containing
3172 * RST have been dropped.
3174 * In the SYN-RECEIVED state, don't send an ACK unless the
3175 * segment we received passes the SYN-RECEIVED ACK test.
3176 * If it fails send a RST. This breaks the loop in the
3177 * "LAND" DoS attack, and also prevents an ACK storm
3178 * between two listening ports that have been sent forged
3179 * SYN segments, each with the source address of the other.
3181 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3182 (SEQ_GT(tp->snd_una, th->th_ack) ||
3183 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3184 rstreason = BANDLIM_RST_OPENPORT;
3188 if (so->so_options & SO_DEBUG)
3189 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3192 TCP_PROBE3(debug__input, tp, th, m);
3193 tp->t_flags |= TF_ACKNOW;
3194 (void) tp->t_fb->tfb_tcp_output(tp);
3195 INP_WUNLOCK(tp->t_inpcb);
3201 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3202 INP_WUNLOCK(tp->t_inpcb);
3204 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3209 * Drop space held by incoming segment and return.
3212 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3213 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3216 TCP_PROBE3(debug__input, tp, th, m);
3218 INP_WUNLOCK(tp->t_inpcb);
3223 * Issue RST and make ACK acceptable to originator of segment.
3224 * The mbuf must still include the original packet header.
3228 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3229 int tlen, int rstreason)
3235 struct ip6_hdr *ip6;
3239 INP_WLOCK_ASSERT(tp->t_inpcb);
3242 /* Don't bother if destination was broadcast/multicast. */
3243 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3246 if (mtod(m, struct ip *)->ip_v == 6) {
3247 ip6 = mtod(m, struct ip6_hdr *);
3248 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3249 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3251 /* IPv6 anycast check is done at tcp6_input() */
3254 #if defined(INET) && defined(INET6)
3259 ip = mtod(m, struct ip *);
3260 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3261 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3262 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3263 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3268 /* Perform bandwidth limiting. */
3269 if (badport_bandlim(rstreason) < 0)
3272 /* tcp_respond consumes the mbuf chain. */
3273 if (th->th_flags & TH_ACK) {
3274 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3275 th->th_ack, TH_RST);
3277 if (th->th_flags & TH_SYN)
3279 if (th->th_flags & TH_FIN)
3281 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3282 (tcp_seq)0, TH_RST|TH_ACK);
3290 * Parse TCP options and place in tcpopt.
3293 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3298 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3300 if (opt == TCPOPT_EOL)
3302 if (opt == TCPOPT_NOP)
3308 if (optlen < 2 || optlen > cnt)
3313 if (optlen != TCPOLEN_MAXSEG)
3315 if (!(flags & TO_SYN))
3317 to->to_flags |= TOF_MSS;
3318 bcopy((char *)cp + 2,
3319 (char *)&to->to_mss, sizeof(to->to_mss));
3320 to->to_mss = ntohs(to->to_mss);
3323 if (optlen != TCPOLEN_WINDOW)
3325 if (!(flags & TO_SYN))
3327 to->to_flags |= TOF_SCALE;
3328 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3330 case TCPOPT_TIMESTAMP:
3331 if (optlen != TCPOLEN_TIMESTAMP)
3333 to->to_flags |= TOF_TS;
3334 bcopy((char *)cp + 2,
3335 (char *)&to->to_tsval, sizeof(to->to_tsval));
3336 to->to_tsval = ntohl(to->to_tsval);
3337 bcopy((char *)cp + 6,
3338 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3339 to->to_tsecr = ntohl(to->to_tsecr);
3341 case TCPOPT_SIGNATURE:
3343 * In order to reply to a host which has set the
3344 * TCP_SIGNATURE option in its initial SYN, we have
3345 * to record the fact that the option was observed
3346 * here for the syncache code to perform the correct
3349 if (optlen != TCPOLEN_SIGNATURE)
3351 to->to_flags |= TOF_SIGNATURE;
3352 to->to_signature = cp + 2;
3354 case TCPOPT_SACK_PERMITTED:
3355 if (optlen != TCPOLEN_SACK_PERMITTED)
3357 if (!(flags & TO_SYN))
3361 to->to_flags |= TOF_SACKPERM;
3364 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3368 to->to_flags |= TOF_SACK;
3369 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3370 to->to_sacks = cp + 2;
3371 TCPSTAT_INC(tcps_sack_rcv_blocks);
3373 case TCPOPT_FAST_OPEN:
3375 * Cookie length validation is performed by the
3376 * server side cookie checking code or the client
3377 * side cookie cache update code.
3379 if (!(flags & TO_SYN))
3381 if (!V_tcp_fastopen_client_enable &&
3382 !V_tcp_fastopen_server_enable)
3384 to->to_flags |= TOF_FASTOPEN;
3385 to->to_tfo_len = optlen - 2;
3386 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3395 * Pull out of band byte out of a segment so
3396 * it doesn't appear in the user's data queue.
3397 * It is still reflected in the segment length for
3398 * sequencing purposes.
3401 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3404 int cnt = off + th->th_urp - 1;
3407 if (m->m_len > cnt) {
3408 char *cp = mtod(m, caddr_t) + cnt;
3409 struct tcpcb *tp = sototcpcb(so);
3411 INP_WLOCK_ASSERT(tp->t_inpcb);
3414 tp->t_oobflags |= TCPOOB_HAVEDATA;
3415 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3417 if (m->m_flags & M_PKTHDR)
3426 panic("tcp_pulloutofband");
3430 * Collect new round-trip time estimate
3431 * and update averages and current timeout.
3434 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3438 INP_WLOCK_ASSERT(tp->t_inpcb);
3440 TCPSTAT_INC(tcps_rttupdated);
3442 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3444 * srtt is stored as fixed point with 5 bits after the
3445 * binary point (i.e., scaled by 8). The following magic
3446 * is equivalent to the smoothing algorithm in rfc793 with
3447 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3448 * point). Adjust rtt to origin 0.
3450 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3451 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3453 if ((tp->t_srtt += delta) <= 0)
3457 * We accumulate a smoothed rtt variance (actually, a
3458 * smoothed mean difference), then set the retransmit
3459 * timer to smoothed rtt + 4 times the smoothed variance.
3460 * rttvar is stored as fixed point with 4 bits after the
3461 * binary point (scaled by 16). The following is
3462 * equivalent to rfc793 smoothing with an alpha of .75
3463 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3464 * rfc793's wired-in beta.
3468 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3469 if ((tp->t_rttvar += delta) <= 0)
3471 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3472 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3475 * No rtt measurement yet - use the unsmoothed rtt.
3476 * Set the variance to half the rtt (so our first
3477 * retransmit happens at 3*rtt).
3479 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3480 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3481 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3487 * the retransmit should happen at rtt + 4 * rttvar.
3488 * Because of the way we do the smoothing, srtt and rttvar
3489 * will each average +1/2 tick of bias. When we compute
3490 * the retransmit timer, we want 1/2 tick of rounding and
3491 * 1 extra tick because of +-1/2 tick uncertainty in the
3492 * firing of the timer. The bias will give us exactly the
3493 * 1.5 tick we need. But, because the bias is
3494 * statistical, we have to test that we don't drop below
3495 * the minimum feasible timer (which is 2 ticks).
3497 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3498 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3501 * We received an ack for a packet that wasn't retransmitted;
3502 * it is probably safe to discard any error indications we've
3503 * received recently. This isn't quite right, but close enough
3504 * for now (a route might have failed after we sent a segment,
3505 * and the return path might not be symmetrical).
3507 tp->t_softerror = 0;
3511 * Determine a reasonable value for maxseg size.
3512 * If the route is known, check route for mtu.
3513 * If none, use an mss that can be handled on the outgoing interface
3514 * without forcing IP to fragment. If no route is found, route has no mtu,
3515 * or the destination isn't local, use a default, hopefully conservative
3516 * size (usually 512 or the default IP max size, but no more than the mtu
3517 * of the interface), as we can't discover anything about intervening
3518 * gateways or networks. We also initialize the congestion/slow start
3519 * window to be a single segment if the destination isn't local.
3520 * While looking at the routing entry, we also initialize other path-dependent
3521 * parameters from pre-set or cached values in the routing entry.
3523 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3524 * IP options, e.g. IPSEC data, since length of this data may vary, and
3525 * thus it is calculated for every segment separately in tcp_output().
3527 * NOTE that this routine is only called when we process an incoming
3528 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3529 * settings are handled in tcp_mssopt().
3532 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3533 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3536 uint32_t maxmtu = 0;
3537 struct inpcb *inp = tp->t_inpcb;
3538 struct hc_metrics_lite metrics;
3540 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3541 size_t min_protoh = isipv6 ?
3542 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3543 sizeof (struct tcpiphdr);
3545 const size_t min_protoh = sizeof(struct tcpiphdr);
3548 INP_WLOCK_ASSERT(tp->t_inpcb);
3550 if (mtuoffer != -1) {
3551 KASSERT(offer == -1, ("%s: conflict", __func__));
3552 offer = mtuoffer - min_protoh;
3558 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3559 tp->t_maxseg = V_tcp_v6mssdflt;
3562 #if defined(INET) && defined(INET6)
3567 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3568 tp->t_maxseg = V_tcp_mssdflt;
3573 * No route to sender, stay with default mss and return.
3577 * In case we return early we need to initialize metrics
3578 * to a defined state as tcp_hc_get() would do for us
3579 * if there was no cache hit.
3581 if (metricptr != NULL)
3582 bzero(metricptr, sizeof(struct hc_metrics_lite));
3586 /* What have we got? */
3590 * Offer == 0 means that there was no MSS on the SYN
3591 * segment, in this case we use tcp_mssdflt as
3592 * already assigned to t_maxseg above.
3594 offer = tp->t_maxseg;
3599 * Offer == -1 means that we didn't receive SYN yet.
3605 * Prevent DoS attack with too small MSS. Round up
3606 * to at least minmss.
3608 offer = max(offer, V_tcp_minmss);
3612 * rmx information is now retrieved from tcp_hostcache.
3614 tcp_hc_get(&inp->inp_inc, &metrics);
3615 if (metricptr != NULL)
3616 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3619 * If there's a discovered mtu in tcp hostcache, use it.
3620 * Else, use the link mtu.
3622 if (metrics.rmx_mtu)
3623 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3627 mss = maxmtu - min_protoh;
3628 if (!V_path_mtu_discovery &&
3629 !in6_localaddr(&inp->in6p_faddr))
3630 mss = min(mss, V_tcp_v6mssdflt);
3633 #if defined(INET) && defined(INET6)
3638 mss = maxmtu - min_protoh;
3639 if (!V_path_mtu_discovery &&
3640 !in_localaddr(inp->inp_faddr))
3641 mss = min(mss, V_tcp_mssdflt);
3645 * XXX - The above conditional (mss = maxmtu - min_protoh)
3646 * probably violates the TCP spec.
3647 * The problem is that, since we don't know the
3648 * other end's MSS, we are supposed to use a conservative
3649 * default. But, if we do that, then MTU discovery will
3650 * never actually take place, because the conservative
3651 * default is much less than the MTUs typically seen
3652 * on the Internet today. For the moment, we'll sweep
3653 * this under the carpet.
3655 * The conservative default might not actually be a problem
3656 * if the only case this occurs is when sending an initial
3657 * SYN with options and data to a host we've never talked
3658 * to before. Then, they will reply with an MSS value which
3659 * will get recorded and the new parameters should get
3660 * recomputed. For Further Study.
3663 mss = min(mss, offer);
3666 * Sanity check: make sure that maxseg will be large
3667 * enough to allow some data on segments even if the
3668 * all the option space is used (40bytes). Otherwise
3669 * funny things may happen in tcp_output.
3671 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3679 tcp_mss(struct tcpcb *tp, int offer)
3685 struct hc_metrics_lite metrics;
3686 struct tcp_ifcap cap;
3688 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3690 bzero(&cap, sizeof(cap));
3691 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3697 * If there's a pipesize, change the socket buffer to that size,
3698 * don't change if sb_hiwat is different than default (then it
3699 * has been changed on purpose with setsockopt).
3700 * Make the socket buffers an integral number of mss units;
3701 * if the mss is larger than the socket buffer, decrease the mss.
3703 so = inp->inp_socket;
3704 SOCKBUF_LOCK(&so->so_snd);
3705 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3706 bufsize = metrics.rmx_sendpipe;
3708 bufsize = so->so_snd.sb_hiwat;
3712 bufsize = roundup(bufsize, mss);
3713 if (bufsize > sb_max)
3715 if (bufsize > so->so_snd.sb_hiwat)
3716 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3718 SOCKBUF_UNLOCK(&so->so_snd);
3720 * Sanity check: make sure that maxseg will be large
3721 * enough to allow some data on segments even if the
3722 * all the option space is used (40bytes). Otherwise
3723 * funny things may happen in tcp_output.
3725 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3727 tp->t_maxseg = max(mss, 64);
3729 SOCKBUF_LOCK(&so->so_rcv);
3730 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3731 bufsize = metrics.rmx_recvpipe;
3733 bufsize = so->so_rcv.sb_hiwat;
3734 if (bufsize > mss) {
3735 bufsize = roundup(bufsize, mss);
3736 if (bufsize > sb_max)
3738 if (bufsize > so->so_rcv.sb_hiwat)
3739 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3741 SOCKBUF_UNLOCK(&so->so_rcv);
3743 /* Check the interface for TSO capabilities. */
3744 if (cap.ifcap & CSUM_TSO) {
3745 tp->t_flags |= TF_TSO;
3746 tp->t_tsomax = cap.tsomax;
3747 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3748 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3753 * Determine the MSS option to send on an outgoing SYN.
3756 tcp_mssopt(struct in_conninfo *inc)
3759 uint32_t thcmtu = 0;
3760 uint32_t maxmtu = 0;
3763 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3766 if (inc->inc_flags & INC_ISIPV6) {
3767 mss = V_tcp_v6mssdflt;
3768 maxmtu = tcp_maxmtu6(inc, NULL);
3769 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3772 #if defined(INET) && defined(INET6)
3777 mss = V_tcp_mssdflt;
3778 maxmtu = tcp_maxmtu(inc, NULL);
3779 min_protoh = sizeof(struct tcpiphdr);
3782 #if defined(INET6) || defined(INET)
3783 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3786 if (maxmtu && thcmtu)
3787 mss = min(maxmtu, thcmtu) - min_protoh;
3788 else if (maxmtu || thcmtu)
3789 mss = max(maxmtu, thcmtu) - min_protoh;
3796 * On a partial ack arrives, force the retransmission of the
3797 * next unacknowledged segment. Do not clear tp->t_dupacks.
3798 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3802 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3804 tcp_seq onxt = tp->snd_nxt;
3805 uint32_t ocwnd = tp->snd_cwnd;
3806 u_int maxseg = tcp_maxseg(tp);
3808 INP_WLOCK_ASSERT(tp->t_inpcb);
3810 tcp_timer_activate(tp, TT_REXMT, 0);
3812 tp->snd_nxt = th->th_ack;
3814 * Set snd_cwnd to one segment beyond acknowledged offset.
3815 * (tp->snd_una has not yet been updated when this function is called.)
3817 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3818 tp->t_flags |= TF_ACKNOW;
3819 (void) tp->t_fb->tfb_tcp_output(tp);
3820 tp->snd_cwnd = ocwnd;
3821 if (SEQ_GT(onxt, tp->snd_nxt))
3824 * Partial window deflation. Relies on fact that tp->snd_una
3827 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3828 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3831 tp->snd_cwnd += maxseg;
3835 tcp_compute_pipe(struct tcpcb *tp)
3837 return (tp->snd_max - tp->snd_una +
3838 tp->sackhint.sack_bytes_rexmit -
3839 tp->sackhint.sacked_bytes);
3843 tcp_compute_initwnd(uint32_t maxseg)
3846 * Calculate the Initial Window, also used as Restart Window
3848 * RFC5681 Section 3.1 specifies the default conservative values.
3849 * RFC3390 specifies slightly more aggressive values.
3850 * RFC6928 increases it to ten segments.
3851 * Support for user specified value for initial flight size.
3853 if (V_tcp_initcwnd_segments)
3854 return min(V_tcp_initcwnd_segments * maxseg,
3855 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
3856 else if (V_tcp_do_rfc3390)
3857 return min(4 * maxseg, max(2 * maxseg, 4380));
3859 /* Per RFC5681 Section 3.1 */
3861 return (2 * maxseg);
3862 else if (maxseg > 1095)
3863 return (3 * maxseg);
3865 return (4 * maxseg);