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>
104 #include <netinet/tcp_fastopen.h>
106 #include <netinet/tcp.h>
107 #include <netinet/tcp_fsm.h>
108 #include <netinet/tcp_seq.h>
109 #include <netinet/tcp_timer.h>
110 #include <netinet/tcp_var.h>
111 #include <netinet6/tcp6_var.h>
112 #include <netinet/tcpip.h>
113 #include <netinet/cc/cc.h>
115 #include <netinet/tcp_pcap.h>
117 #include <netinet/tcp_syncache.h>
119 #include <netinet/tcp_debug.h>
120 #endif /* TCPDEBUG */
122 #include <netinet/tcp_offload.h>
125 #include <netipsec/ipsec_support.h>
127 #include <machine/in_cksum.h>
129 #include <security/mac/mac_framework.h>
131 const int tcprexmtthresh = 3;
133 int tcp_log_in_vain = 0;
134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
136 "Log all incoming TCP segments to closed ports");
138 VNET_DEFINE(int, blackhole) = 0;
139 #define V_blackhole VNET(blackhole)
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
141 &VNET_NAME(blackhole), 0,
142 "Do not send RST on segments to closed ports");
144 VNET_DEFINE(int, tcp_delack_enabled) = 1;
145 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
146 &VNET_NAME(tcp_delack_enabled), 0,
147 "Delay ACK to try and piggyback it onto a data packet");
149 VNET_DEFINE(int, drop_synfin) = 0;
150 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
151 &VNET_NAME(drop_synfin), 0,
152 "Drop TCP packets with SYN+FIN set");
154 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
155 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
156 &VNET_NAME(tcp_do_rfc6675_pipe), 0,
157 "Use calculated pipe/in-flight bytes per RFC 6675");
159 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
161 &VNET_NAME(tcp_do_rfc3042), 0,
162 "Enable RFC 3042 (Limited Transmit)");
164 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
166 &VNET_NAME(tcp_do_rfc3390), 0,
167 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
169 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
171 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
172 "Slow-start flight size (initial congestion window) in number of segments");
174 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
176 &VNET_NAME(tcp_do_rfc3465), 0,
177 "Enable RFC 3465 (Appropriate Byte Counting)");
179 VNET_DEFINE(int, tcp_abc_l_var) = 2;
180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
181 &VNET_NAME(tcp_abc_l_var), 2,
182 "Cap the max cwnd increment during slow-start to this number of segments");
184 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
186 VNET_DEFINE(int, tcp_do_ecn) = 2;
187 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
188 &VNET_NAME(tcp_do_ecn), 0,
191 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
192 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
193 &VNET_NAME(tcp_ecn_maxretries), 0,
194 "Max retries before giving up on ECN");
196 VNET_DEFINE(int, tcp_insecure_syn) = 0;
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
198 &VNET_NAME(tcp_insecure_syn), 0,
199 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
201 VNET_DEFINE(int, tcp_insecure_rst) = 0;
202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
203 &VNET_NAME(tcp_insecure_rst), 0,
204 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
206 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
207 #define V_tcp_recvspace VNET(tcp_recvspace)
208 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
209 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
211 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
212 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
213 &VNET_NAME(tcp_do_autorcvbuf), 0,
214 "Enable automatic receive buffer sizing");
216 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW,
218 &VNET_NAME(tcp_autorcvbuf_inc), 0,
219 "Incrementor step size of automatic receive buffer");
221 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
223 &VNET_NAME(tcp_autorcvbuf_max), 0,
224 "Max size of automatic receive buffer");
226 VNET_DEFINE(struct inpcbhead, tcb);
227 #define tcb6 tcb /* for KAME src sync over BSD*'s */
228 VNET_DEFINE(struct inpcbinfo, tcbinfo);
231 * TCP statistics are stored in an array of counter(9)s, which size matches
232 * size of struct tcpstat. TCP running connection count is a regular array.
234 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
235 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
236 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
237 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
238 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
239 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
240 "TCP connection counts by TCP state");
243 tcp_vnet_init(const void *unused)
246 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
247 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
249 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
250 tcp_vnet_init, NULL);
254 tcp_vnet_uninit(const void *unused)
257 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
258 VNET_PCPUSTAT_FREE(tcpstat);
260 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
261 tcp_vnet_uninit, NULL);
265 * Kernel module interface for updating tcpstat. The argument is an index
266 * into tcpstat treated as an array.
269 kmod_tcpstat_inc(int statnum)
272 counter_u64_add(VNET(tcpstat)[statnum], 1);
277 * Wrapper for the TCP established input helper hook.
280 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
282 struct tcp_hhook_data hhook_data;
284 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
289 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
296 * CC wrapper hook functions
299 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
302 INP_WLOCK_ASSERT(tp->t_inpcb);
304 tp->ccv->nsegs = nsegs;
305 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
306 if (tp->snd_cwnd <= tp->snd_wnd)
307 tp->ccv->flags |= CCF_CWND_LIMITED;
309 tp->ccv->flags &= ~CCF_CWND_LIMITED;
311 if (type == CC_ACK) {
312 if (tp->snd_cwnd > tp->snd_ssthresh) {
313 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
314 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp));
315 if (tp->t_bytes_acked >= tp->snd_cwnd) {
316 tp->t_bytes_acked -= tp->snd_cwnd;
317 tp->ccv->flags |= CCF_ABC_SENTAWND;
320 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
321 tp->t_bytes_acked = 0;
325 if (CC_ALGO(tp)->ack_received != NULL) {
326 /* XXXLAS: Find a way to live without this */
327 tp->ccv->curack = th->th_ack;
328 CC_ALGO(tp)->ack_received(tp->ccv, type);
333 cc_conn_init(struct tcpcb *tp)
335 struct hc_metrics_lite metrics;
336 struct inpcb *inp = tp->t_inpcb;
340 INP_WLOCK_ASSERT(tp->t_inpcb);
342 tcp_hc_get(&inp->inp_inc, &metrics);
343 maxseg = tcp_maxseg(tp);
345 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
347 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
348 TCPSTAT_INC(tcps_usedrtt);
349 if (metrics.rmx_rttvar) {
350 tp->t_rttvar = metrics.rmx_rttvar;
351 TCPSTAT_INC(tcps_usedrttvar);
353 /* default variation is +- 1 rtt */
355 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
357 TCPT_RANGESET(tp->t_rxtcur,
358 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
359 tp->t_rttmin, TCPTV_REXMTMAX);
361 if (metrics.rmx_ssthresh) {
363 * There's some sort of gateway or interface
364 * buffer limit on the path. Use this to set
365 * the slow start threshold, but set the
366 * threshold to no less than 2*mss.
368 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
369 TCPSTAT_INC(tcps_usedssthresh);
373 * Set the initial slow-start flight size.
375 * RFC5681 Section 3.1 specifies the default conservative values.
376 * RFC3390 specifies slightly more aggressive values.
377 * RFC6928 increases it to ten segments.
378 * Support for user specified value for initial flight size.
380 * If a SYN or SYN/ACK was lost and retransmitted, we have to
381 * reduce the initial CWND to one segment as congestion is likely
382 * requiring us to be cautious.
384 if (tp->snd_cwnd == 1)
385 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
386 else if (V_tcp_initcwnd_segments)
387 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg,
388 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
389 else if (V_tcp_do_rfc3390)
390 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380));
392 /* Per RFC5681 Section 3.1 */
394 tp->snd_cwnd = 2 * maxseg;
395 else if (maxseg > 1095)
396 tp->snd_cwnd = 3 * maxseg;
398 tp->snd_cwnd = 4 * maxseg;
401 if (CC_ALGO(tp)->conn_init != NULL)
402 CC_ALGO(tp)->conn_init(tp->ccv);
406 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
410 INP_WLOCK_ASSERT(tp->t_inpcb);
414 if (!IN_FASTRECOVERY(tp->t_flags)) {
415 tp->snd_recover = tp->snd_max;
416 if (tp->t_flags & TF_ECN_PERMIT)
417 tp->t_flags |= TF_ECN_SND_CWR;
421 if (!IN_CONGRECOVERY(tp->t_flags)) {
422 TCPSTAT_INC(tcps_ecn_rcwnd);
423 tp->snd_recover = tp->snd_max;
424 if (tp->t_flags & TF_ECN_PERMIT)
425 tp->t_flags |= TF_ECN_SND_CWR;
429 maxseg = tcp_maxseg(tp);
431 tp->t_bytes_acked = 0;
432 EXIT_RECOVERY(tp->t_flags);
433 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
435 tp->snd_cwnd = maxseg;
438 TCPSTAT_INC(tcps_sndrexmitbad);
439 /* RTO was unnecessary, so reset everything. */
440 tp->snd_cwnd = tp->snd_cwnd_prev;
441 tp->snd_ssthresh = tp->snd_ssthresh_prev;
442 tp->snd_recover = tp->snd_recover_prev;
443 if (tp->t_flags & TF_WASFRECOVERY)
444 ENTER_FASTRECOVERY(tp->t_flags);
445 if (tp->t_flags & TF_WASCRECOVERY)
446 ENTER_CONGRECOVERY(tp->t_flags);
447 tp->snd_nxt = tp->snd_max;
448 tp->t_flags &= ~TF_PREVVALID;
453 if (CC_ALGO(tp)->cong_signal != NULL) {
455 tp->ccv->curack = th->th_ack;
456 CC_ALGO(tp)->cong_signal(tp->ccv, type);
461 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
463 INP_WLOCK_ASSERT(tp->t_inpcb);
465 /* XXXLAS: KASSERT that we're in recovery? */
467 if (CC_ALGO(tp)->post_recovery != NULL) {
468 tp->ccv->curack = th->th_ack;
469 CC_ALGO(tp)->post_recovery(tp->ccv);
471 /* XXXLAS: EXIT_RECOVERY ? */
472 tp->t_bytes_acked = 0;
476 * Indicate whether this ack should be delayed. We can delay the ack if
477 * following conditions are met:
478 * - There is no delayed ack timer in progress.
479 * - Our last ack wasn't a 0-sized window. We never want to delay
480 * the ack that opens up a 0-sized window.
481 * - LRO wasn't used for this segment. We make sure by checking that the
482 * segment size is not larger than the MSS.
484 #define DELAY_ACK(tp, tlen) \
485 ((!tcp_timer_active(tp, TT_DELACK) && \
486 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
487 (tlen <= tp->t_maxseg) && \
488 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
491 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
493 INP_WLOCK_ASSERT(tp->t_inpcb);
495 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
496 switch (iptos & IPTOS_ECN_MASK) {
498 tp->ccv->flags |= CCF_IPHDR_CE;
501 tp->ccv->flags &= ~CCF_IPHDR_CE;
504 tp->ccv->flags &= ~CCF_IPHDR_CE;
508 if (th->th_flags & TH_CWR)
509 tp->ccv->flags |= CCF_TCPHDR_CWR;
511 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
513 if (tp->t_flags & TF_DELACK)
514 tp->ccv->flags |= CCF_DELACK;
516 tp->ccv->flags &= ~CCF_DELACK;
518 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
520 if (tp->ccv->flags & CCF_ACKNOW)
521 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
526 * TCP input handling is split into multiple parts:
527 * tcp6_input is a thin wrapper around tcp_input for the extended
528 * ip6_protox[] call format in ip6_input
529 * tcp_input handles primary segment validation, inpcb lookup and
530 * SYN processing on listen sockets
531 * tcp_do_segment processes the ACK and text of the segment for
532 * establishing, established and closing connections
536 tcp6_input(struct mbuf **mp, int *offp, int proto)
538 struct mbuf *m = *mp;
539 struct in6_ifaddr *ia6;
542 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
545 * draft-itojun-ipv6-tcp-to-anycast
546 * better place to put this in?
548 ip6 = mtod(m, struct ip6_hdr *);
549 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
550 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
553 ifa_free(&ia6->ia_ifa);
554 ip6 = mtod(m, struct ip6_hdr *);
555 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
556 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
557 return (IPPROTO_DONE);
560 ifa_free(&ia6->ia_ifa);
562 return (tcp_input(mp, offp, proto));
567 tcp_input(struct mbuf **mp, int *offp, int proto)
569 struct mbuf *m = *mp;
570 struct tcphdr *th = NULL;
571 struct ip *ip = NULL;
572 struct inpcb *inp = NULL;
573 struct tcpcb *tp = NULL;
574 struct socket *so = NULL;
584 int rstreason = 0; /* For badport_bandlim accounting purposes */
586 struct m_tag *fwd_tag = NULL;
588 struct ip6_hdr *ip6 = NULL;
591 const void *ip6 = NULL;
593 struct tcpopt to; /* options in this segment */
594 char *s = NULL; /* address and port logging */
598 * The size of tcp_saveipgen must be the size of the max ip header,
601 u_char tcp_saveipgen[IP6_HDR_LEN];
602 struct tcphdr tcp_savetcp;
607 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
614 TCPSTAT_INC(tcps_rcvtotal);
618 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
620 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
621 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
623 TCPSTAT_INC(tcps_rcvshort);
624 return (IPPROTO_DONE);
628 ip6 = mtod(m, struct ip6_hdr *);
629 th = (struct tcphdr *)((caddr_t)ip6 + off0);
630 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
631 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
632 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
633 th->th_sum = m->m_pkthdr.csum_data;
635 th->th_sum = in6_cksum_pseudo(ip6, tlen,
636 IPPROTO_TCP, m->m_pkthdr.csum_data);
637 th->th_sum ^= 0xffff;
639 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
641 TCPSTAT_INC(tcps_rcvbadsum);
646 * Be proactive about unspecified IPv6 address in source.
647 * As we use all-zero to indicate unbounded/unconnected pcb,
648 * unspecified IPv6 address can be used to confuse us.
650 * Note that packets with unspecified IPv6 destination is
651 * already dropped in ip6_input.
653 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
657 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
660 #if defined(INET) && defined(INET6)
666 * Get IP and TCP header together in first mbuf.
667 * Note: IP leaves IP header in first mbuf.
669 if (off0 > sizeof (struct ip)) {
671 off0 = sizeof(struct ip);
673 if (m->m_len < sizeof (struct tcpiphdr)) {
674 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
676 TCPSTAT_INC(tcps_rcvshort);
677 return (IPPROTO_DONE);
680 ip = mtod(m, struct ip *);
681 th = (struct tcphdr *)((caddr_t)ip + off0);
682 tlen = ntohs(ip->ip_len) - off0;
685 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
686 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
687 th->th_sum = m->m_pkthdr.csum_data;
689 th->th_sum = in_pseudo(ip->ip_src.s_addr,
691 htonl(m->m_pkthdr.csum_data + tlen +
693 th->th_sum ^= 0xffff;
695 struct ipovly *ipov = (struct ipovly *)ip;
698 * Checksum extended TCP header and data.
701 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
702 ipov->ih_len = htons(tlen);
703 th->th_sum = in_cksum(m, len);
704 /* Reset length for SDT probes. */
705 ip->ip_len = htons(len);
708 /* Re-initialization for later version check */
709 ip->ip_v = IPVERSION;
713 TCPSTAT_INC(tcps_rcvbadsum);
720 * Check that TCP offset makes sense,
721 * pull out TCP options and adjust length. XXX
723 off = th->th_off << 2;
724 if (off < sizeof (struct tcphdr) || off > tlen) {
725 TCPSTAT_INC(tcps_rcvbadoff);
728 tlen -= off; /* tlen is used instead of ti->ti_len */
729 if (off > sizeof (struct tcphdr)) {
732 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
733 ip6 = mtod(m, struct ip6_hdr *);
734 th = (struct tcphdr *)((caddr_t)ip6 + off0);
737 #if defined(INET) && defined(INET6)
742 if (m->m_len < sizeof(struct ip) + off) {
743 if ((m = m_pullup(m, sizeof (struct ip) + off))
745 TCPSTAT_INC(tcps_rcvshort);
746 return (IPPROTO_DONE);
748 ip = mtod(m, struct ip *);
749 th = (struct tcphdr *)((caddr_t)ip + off0);
753 optlen = off - sizeof (struct tcphdr);
754 optp = (u_char *)(th + 1);
756 thflags = th->th_flags;
759 * Convert TCP protocol specific fields to host format.
761 tcp_fields_to_host(th);
764 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
766 drop_hdrlen = off0 + off;
769 * Locate pcb for segment; if we're likely to add or remove a
770 * connection then first acquire pcbinfo lock. There are three cases
771 * where we might discover later we need a write lock despite the
772 * flags: ACKs moving a connection out of the syncache, ACKs for a
773 * connection in TIMEWAIT and SYNs not targeting a listening socket.
775 if ((thflags & (TH_FIN | TH_RST)) != 0) {
776 INP_INFO_RLOCK(&V_tcbinfo);
777 ti_locked = TI_RLOCKED;
779 ti_locked = TI_UNLOCKED;
782 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
786 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
788 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
791 #if defined(INET) && !defined(INET6)
792 (m->m_flags & M_IP_NEXTHOP)
795 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
799 if (ti_locked == TI_RLOCKED) {
800 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
802 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
806 if (isipv6 && fwd_tag != NULL) {
807 struct sockaddr_in6 *next_hop6;
809 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
811 * Transparently forwarded. Pretend to be the destination.
812 * Already got one like this?
814 inp = in6_pcblookup_mbuf(&V_tcbinfo,
815 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
816 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
819 * It's new. Try to find the ambushing socket.
820 * Because we've rewritten the destination address,
821 * any hardware-generated hash is ignored.
823 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
824 th->th_sport, &next_hop6->sin6_addr,
825 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
826 th->th_dport, INPLOOKUP_WILDCARD |
827 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
830 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
831 th->th_sport, &ip6->ip6_dst, th->th_dport,
832 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
833 m->m_pkthdr.rcvif, m);
836 #if defined(INET6) && defined(INET)
840 if (fwd_tag != NULL) {
841 struct sockaddr_in *next_hop;
843 next_hop = (struct sockaddr_in *)(fwd_tag+1);
845 * Transparently forwarded. Pretend to be the destination.
846 * already got one like this?
848 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
849 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
850 m->m_pkthdr.rcvif, m);
853 * It's new. Try to find the ambushing socket.
854 * Because we've rewritten the destination address,
855 * any hardware-generated hash is ignored.
857 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
858 th->th_sport, next_hop->sin_addr,
859 next_hop->sin_port ? ntohs(next_hop->sin_port) :
860 th->th_dport, INPLOOKUP_WILDCARD |
861 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
864 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
865 th->th_sport, ip->ip_dst, th->th_dport,
866 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
867 m->m_pkthdr.rcvif, m);
871 * If the INPCB does not exist then all data in the incoming
872 * segment is discarded and an appropriate RST is sent back.
873 * XXX MRT Send RST using which routing table?
877 * Log communication attempts to ports that are not
880 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
881 tcp_log_in_vain == 2) {
882 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
883 log(LOG_INFO, "%s; %s: Connection attempt "
884 "to closed port\n", s, __func__);
887 * When blackholing do not respond with a RST but
888 * completely ignore the segment and drop it.
890 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
894 rstreason = BANDLIM_RST_CLOSEDPORT;
897 INP_WLOCK_ASSERT(inp);
899 * While waiting for inp lock during the lookup, another thread
900 * can have dropped the inpcb, in which case we need to loop back
901 * and try to find a new inpcb to deliver to.
903 if (inp->inp_flags & INP_DROPPED) {
908 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
909 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
910 ((inp->inp_socket == NULL) ||
911 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
912 inp->inp_flowid = m->m_pkthdr.flowid;
913 inp->inp_flowtype = M_HASHTYPE_GET(m);
915 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
917 if (isipv6 && IPSEC_ENABLED(ipv6) &&
918 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
926 if (IPSEC_ENABLED(ipv4) &&
927 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
934 * Check the minimum TTL for socket.
936 if (inp->inp_ip_minttl != 0) {
939 if (inp->inp_ip_minttl > ip6->ip6_hlim)
943 if (inp->inp_ip_minttl > ip->ip_ttl)
948 * A previous connection in TIMEWAIT state is supposed to catch stray
949 * or duplicate segments arriving late. If this segment was a
950 * legitimate new connection attempt, the old INPCB gets removed and
951 * we can try again to find a listening socket.
953 * At this point, due to earlier optimism, we may hold only an inpcb
954 * lock, and not the inpcbinfo write lock. If so, we need to try to
955 * acquire it, or if that fails, acquire a reference on the inpcb,
956 * drop all locks, acquire a global write lock, and then re-acquire
957 * the inpcb lock. We may at that point discover that another thread
958 * has tried to free the inpcb, in which case we need to loop back
959 * and try to find a new inpcb to deliver to.
961 * XXXRW: It may be time to rethink timewait locking.
964 if (inp->inp_flags & INP_TIMEWAIT) {
965 if (ti_locked == TI_UNLOCKED) {
966 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
969 INP_INFO_RLOCK(&V_tcbinfo);
970 ti_locked = TI_RLOCKED;
972 if (in_pcbrele_wlocked(inp)) {
975 } else if (inp->inp_flags & INP_DROPPED) {
981 ti_locked = TI_RLOCKED;
983 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
985 if (thflags & TH_SYN)
986 tcp_dooptions(&to, optp, optlen, TO_SYN);
988 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
990 if (tcp_twcheck(inp, &to, th, m, tlen))
992 INP_INFO_RUNLOCK(&V_tcbinfo);
993 return (IPPROTO_DONE);
996 * The TCPCB may no longer exist if the connection is winding
997 * down or it is in the CLOSED state. Either way we drop the
998 * segment and send an appropriate response.
1000 tp = intotcpcb(inp);
1001 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1002 rstreason = BANDLIM_RST_CLOSEDPORT;
1007 if (tp->t_flags & TF_TOE) {
1008 tcp_offload_input(tp, m);
1009 m = NULL; /* consumed by the TOE driver */
1015 * We've identified a valid inpcb, but it could be that we need an
1016 * inpcbinfo write lock but don't hold it. In this case, attempt to
1017 * acquire using the same strategy as the TIMEWAIT case above. If we
1018 * relock, we have to jump back to 'relocked' as the connection might
1019 * now be in TIMEWAIT.
1022 if ((thflags & (TH_FIN | TH_RST)) != 0)
1023 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1025 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) ||
1026 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) &&
1027 !IS_FASTOPEN(tp->t_flags)))) {
1028 if (ti_locked == TI_UNLOCKED) {
1029 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
1032 INP_INFO_RLOCK(&V_tcbinfo);
1033 ti_locked = TI_RLOCKED;
1035 if (in_pcbrele_wlocked(inp)) {
1038 } else if (inp->inp_flags & INP_DROPPED) {
1045 ti_locked = TI_RLOCKED;
1047 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1051 INP_WLOCK_ASSERT(inp);
1052 if (mac_inpcb_check_deliver(inp, m))
1055 so = inp->inp_socket;
1056 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1058 if (so->so_options & SO_DEBUG) {
1059 ostate = tp->t_state;
1062 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1065 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1068 #endif /* TCPDEBUG */
1070 * When the socket is accepting connections (the INPCB is in LISTEN
1071 * state) we look into the SYN cache if this is a new connection
1072 * attempt or the completion of a previous one.
1074 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1075 ("%s: so accepting but tp %p not listening", __func__, tp));
1076 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1077 struct in_conninfo inc;
1079 bzero(&inc, sizeof(inc));
1082 inc.inc_flags |= INC_ISIPV6;
1083 inc.inc6_faddr = ip6->ip6_src;
1084 inc.inc6_laddr = ip6->ip6_dst;
1088 inc.inc_faddr = ip->ip_src;
1089 inc.inc_laddr = ip->ip_dst;
1091 inc.inc_fport = th->th_sport;
1092 inc.inc_lport = th->th_dport;
1093 inc.inc_fibnum = so->so_fibnum;
1096 * Check for an existing connection attempt in syncache if
1097 * the flag is only ACK. A successful lookup creates a new
1098 * socket appended to the listen queue in SYN_RECEIVED state.
1100 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1102 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1104 * Parse the TCP options here because
1105 * syncookies need access to the reflected
1108 tcp_dooptions(&to, optp, optlen, 0);
1110 * NB: syncache_expand() doesn't unlock
1111 * inp and tcpinfo locks.
1113 rstreason = syncache_expand(&inc, &to, th, &so, m);
1114 if (rstreason < 0) {
1116 * A failing TCP MD5 signature comparison
1117 * must result in the segment being dropped
1118 * and must not produce any response back
1122 } else if (rstreason == 0) {
1124 * No syncache entry or ACK was not
1125 * for our SYN/ACK. Send a RST.
1126 * NB: syncache did its own logging
1127 * of the failure cause.
1129 rstreason = BANDLIM_RST_OPENPORT;
1137 * We completed the 3-way handshake
1138 * but could not allocate a socket
1139 * either due to memory shortage,
1140 * listen queue length limits or
1141 * global socket limits. Send RST
1142 * or wait and have the remote end
1143 * retransmit the ACK for another
1146 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1147 log(LOG_DEBUG, "%s; %s: Listen socket: "
1148 "Socket allocation failed due to "
1149 "limits or memory shortage, %s\n",
1151 V_tcp_sc_rst_sock_fail ?
1152 "sending RST" : "try again");
1153 if (V_tcp_sc_rst_sock_fail) {
1154 rstreason = BANDLIM_UNLIMITED;
1160 * Socket is created in state SYN_RECEIVED.
1161 * Unlock the listen socket, lock the newly
1162 * created socket and update the tp variable.
1164 INP_WUNLOCK(inp); /* listen socket */
1165 inp = sotoinpcb(so);
1167 * New connection inpcb is already locked by
1168 * syncache_expand().
1170 INP_WLOCK_ASSERT(inp);
1171 tp = intotcpcb(inp);
1172 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1173 ("%s: ", __func__));
1175 * Process the segment and the data it
1176 * contains. tcp_do_segment() consumes
1177 * the mbuf chain and unlocks the inpcb.
1179 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1181 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1182 return (IPPROTO_DONE);
1185 * Segment flag validation for new connection attempts:
1187 * Our (SYN|ACK) response was rejected.
1188 * Check with syncache and remove entry to prevent
1191 * NB: syncache_chkrst does its own logging of failure
1194 if (thflags & TH_RST) {
1195 syncache_chkrst(&inc, th);
1199 * We can't do anything without SYN.
1201 if ((thflags & TH_SYN) == 0) {
1202 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1203 log(LOG_DEBUG, "%s; %s: Listen socket: "
1204 "SYN is missing, segment ignored\n",
1206 TCPSTAT_INC(tcps_badsyn);
1210 * (SYN|ACK) is bogus on a listen socket.
1212 if (thflags & TH_ACK) {
1213 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1214 log(LOG_DEBUG, "%s; %s: Listen socket: "
1215 "SYN|ACK invalid, segment rejected\n",
1217 syncache_badack(&inc); /* XXX: Not needed! */
1218 TCPSTAT_INC(tcps_badsyn);
1219 rstreason = BANDLIM_RST_OPENPORT;
1223 * If the drop_synfin option is enabled, drop all
1224 * segments with both the SYN and FIN bits set.
1225 * This prevents e.g. nmap from identifying the
1227 * XXX: Poor reasoning. nmap has other methods
1228 * and is constantly refining its stack detection
1230 * XXX: This is a violation of the TCP specification
1231 * and was used by RFC1644.
1233 if ((thflags & TH_FIN) && V_drop_synfin) {
1234 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1235 log(LOG_DEBUG, "%s; %s: Listen socket: "
1236 "SYN|FIN segment ignored (based on "
1237 "sysctl setting)\n", s, __func__);
1238 TCPSTAT_INC(tcps_badsyn);
1242 * Segment's flags are (SYN) or (SYN|FIN).
1244 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1245 * as they do not affect the state of the TCP FSM.
1246 * The data pointed to by TH_URG and th_urp is ignored.
1248 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1249 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1250 KASSERT(thflags & (TH_SYN),
1251 ("%s: Listen socket: TH_SYN not set", __func__));
1254 * If deprecated address is forbidden,
1255 * we do not accept SYN to deprecated interface
1256 * address to prevent any new inbound connection from
1257 * getting established.
1258 * When we do not accept SYN, we send a TCP RST,
1259 * with deprecated source address (instead of dropping
1260 * it). We compromise it as it is much better for peer
1261 * to send a RST, and RST will be the final packet
1264 * If we do not forbid deprecated addresses, we accept
1265 * the SYN packet. RFC2462 does not suggest dropping
1267 * If we decipher RFC2462 5.5.4, it says like this:
1268 * 1. use of deprecated addr with existing
1269 * communication is okay - "SHOULD continue to be
1271 * 2. use of it with new communication:
1272 * (2a) "SHOULD NOT be used if alternate address
1273 * with sufficient scope is available"
1274 * (2b) nothing mentioned otherwise.
1275 * Here we fall into (2b) case as we have no choice in
1276 * our source address selection - we must obey the peer.
1278 * The wording in RFC2462 is confusing, and there are
1279 * multiple description text for deprecated address
1280 * handling - worse, they are not exactly the same.
1281 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1283 if (isipv6 && !V_ip6_use_deprecated) {
1284 struct in6_ifaddr *ia6;
1286 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1288 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1289 ifa_free(&ia6->ia_ifa);
1290 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1291 log(LOG_DEBUG, "%s; %s: Listen socket: "
1292 "Connection attempt to deprecated "
1293 "IPv6 address rejected\n",
1295 rstreason = BANDLIM_RST_OPENPORT;
1299 ifa_free(&ia6->ia_ifa);
1303 * Basic sanity checks on incoming SYN requests:
1304 * Don't respond if the destination is a link layer
1305 * broadcast according to RFC1122 4.2.3.10, p. 104.
1306 * If it is from this socket it must be forged.
1307 * Don't respond if the source or destination is a
1308 * global or subnet broad- or multicast address.
1309 * Note that it is quite possible to receive unicast
1310 * link-layer packets with a broadcast IP address. Use
1311 * in_broadcast() to find them.
1313 if (m->m_flags & (M_BCAST|M_MCAST)) {
1314 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1315 log(LOG_DEBUG, "%s; %s: Listen socket: "
1316 "Connection attempt from broad- or multicast "
1317 "link layer address ignored\n", s, __func__);
1322 if (th->th_dport == th->th_sport &&
1323 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1324 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1325 log(LOG_DEBUG, "%s; %s: Listen socket: "
1326 "Connection attempt to/from self "
1327 "ignored\n", s, __func__);
1330 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1331 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1332 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1333 log(LOG_DEBUG, "%s; %s: Listen socket: "
1334 "Connection attempt from/to multicast "
1335 "address ignored\n", s, __func__);
1340 #if defined(INET) && defined(INET6)
1345 if (th->th_dport == th->th_sport &&
1346 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1347 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1348 log(LOG_DEBUG, "%s; %s: Listen socket: "
1349 "Connection attempt from/to self "
1350 "ignored\n", s, __func__);
1353 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1354 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1355 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1356 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1357 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1358 log(LOG_DEBUG, "%s; %s: Listen socket: "
1359 "Connection attempt from/to broad- "
1360 "or multicast address ignored\n",
1367 * SYN appears to be valid. Create compressed TCP state
1371 if (so->so_options & SO_DEBUG)
1372 tcp_trace(TA_INPUT, ostate, tp,
1373 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1375 TCP_PROBE3(debug__input, tp, th, m);
1376 tcp_dooptions(&to, optp, optlen, TO_SYN);
1378 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL))
1379 goto tfo_socket_result;
1381 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1384 * Entry added to syncache and mbuf consumed.
1385 * Only the listen socket is unlocked by syncache_add().
1387 if (ti_locked == TI_RLOCKED) {
1388 INP_INFO_RUNLOCK(&V_tcbinfo);
1389 ti_locked = TI_UNLOCKED;
1391 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1392 return (IPPROTO_DONE);
1393 } else if (tp->t_state == TCPS_LISTEN) {
1395 * When a listen socket is torn down the SO_ACCEPTCONN
1396 * flag is removed first while connections are drained
1397 * from the accept queue in a unlock/lock cycle of the
1398 * ACCEPT_LOCK, opening a race condition allowing a SYN
1399 * attempt go through unhandled.
1403 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1404 if (tp->t_flags & TF_SIGNATURE) {
1405 tcp_dooptions(&to, optp, optlen, thflags);
1406 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1407 TCPSTAT_INC(tcps_sig_err_nosigopt);
1410 if (!TCPMD5_ENABLED() ||
1411 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1415 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1418 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1419 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1420 * the inpcb, and unlocks pcbinfo.
1422 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1423 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1424 return (IPPROTO_DONE);
1427 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1429 if (ti_locked == TI_RLOCKED) {
1430 INP_INFO_RUNLOCK(&V_tcbinfo);
1431 ti_locked = TI_UNLOCKED;
1435 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1436 "ti_locked: %d", __func__, ti_locked));
1437 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1442 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1445 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1446 m = NULL; /* mbuf chain got consumed. */
1451 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1453 if (ti_locked == TI_RLOCKED) {
1454 INP_INFO_RUNLOCK(&V_tcbinfo);
1455 ti_locked = TI_UNLOCKED;
1459 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1460 "ti_locked: %d", __func__, ti_locked));
1461 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1469 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1474 return (IPPROTO_DONE);
1478 * Automatic sizing of receive socket buffer. Often the send
1479 * buffer size is not optimally adjusted to the actual network
1480 * conditions at hand (delay bandwidth product). Setting the
1481 * buffer size too small limits throughput on links with high
1482 * bandwidth and high delay (eg. trans-continental/oceanic links).
1484 * On the receive side the socket buffer memory is only rarely
1485 * used to any significant extent. This allows us to be much
1486 * more aggressive in scaling the receive socket buffer. For
1487 * the case that the buffer space is actually used to a large
1488 * extent and we run out of kernel memory we can simply drop
1489 * the new segments; TCP on the sender will just retransmit it
1490 * later. Setting the buffer size too big may only consume too
1491 * much kernel memory if the application doesn't read() from
1492 * the socket or packet loss or reordering makes use of the
1495 * The criteria to step up the receive buffer one notch are:
1496 * 1. Application has not set receive buffer size with
1497 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1498 * 2. the number of bytes received during the time it takes
1499 * one timestamp to be reflected back to us (the RTT);
1500 * 3. received bytes per RTT is within seven eighth of the
1501 * current socket buffer size;
1502 * 4. receive buffer size has not hit maximal automatic size;
1504 * This algorithm does one step per RTT at most and only if
1505 * we receive a bulk stream w/o packet losses or reorderings.
1506 * Shrinking the buffer during idle times is not necessary as
1507 * it doesn't consume any memory when idle.
1509 * TODO: Only step up if the application is actually serving
1510 * the buffer to better manage the socket buffer resources.
1513 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1514 struct tcpcb *tp, int tlen)
1518 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1519 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1520 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1521 (tp->t_srtt >> TCP_RTT_SHIFT)) {
1522 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) &&
1523 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1524 newsize = min(so->so_rcv.sb_hiwat +
1525 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max);
1527 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1529 /* Start over with next RTT. */
1533 tp->rfbuf_cnt += tlen; /* add up */
1540 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1541 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1544 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1545 int rstreason, todrop, win;
1549 struct in_conninfo *inc;
1558 * The size of tcp_saveipgen must be the size of the max ip header,
1561 u_char tcp_saveipgen[IP6_HDR_LEN];
1562 struct tcphdr tcp_savetcp;
1565 thflags = th->th_flags;
1566 inc = &tp->t_inpcb->inp_inc;
1567 tp->sackhint.last_sack_ack = 0;
1569 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1572 * If this is either a state-changing packet or current state isn't
1573 * established, we require a write lock on tcbinfo. Otherwise, we
1574 * allow the tcbinfo to be in either alocked or unlocked, as the
1575 * caller may have unnecessarily acquired a write lock due to a race.
1577 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1578 tp->t_state != TCPS_ESTABLISHED) {
1579 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for "
1580 "SYN/FIN/RST/!EST", __func__, ti_locked));
1581 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1584 if (ti_locked == TI_RLOCKED)
1585 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1587 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1588 "ti_locked: %d", __func__, ti_locked));
1589 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1593 INP_WLOCK_ASSERT(tp->t_inpcb);
1594 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1596 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1600 /* Save segment, if requested. */
1601 tcp_pcap_add(th, m, &(tp->t_inpkts));
1604 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1605 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1606 log(LOG_DEBUG, "%s; %s: "
1607 "SYN|FIN segment ignored (based on "
1608 "sysctl setting)\n", s, __func__);
1615 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1616 * check SEQ.ACK first.
1618 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1619 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1620 rstreason = BANDLIM_UNLIMITED;
1625 * Segment received on connection.
1626 * Reset idle time and keep-alive timer.
1627 * XXX: This should be done after segment
1628 * validation to ignore broken/spoofed segs.
1630 tp->t_rcvtime = ticks;
1633 * Scale up the window into a 32-bit value.
1634 * For the SYN_SENT state the scale is zero.
1636 tiwin = th->th_win << tp->snd_scale;
1639 * TCP ECN processing.
1641 if (tp->t_flags & TF_ECN_PERMIT) {
1642 if (thflags & TH_CWR)
1643 tp->t_flags &= ~TF_ECN_SND_ECE;
1644 switch (iptos & IPTOS_ECN_MASK) {
1646 tp->t_flags |= TF_ECN_SND_ECE;
1647 TCPSTAT_INC(tcps_ecn_ce);
1649 case IPTOS_ECN_ECT0:
1650 TCPSTAT_INC(tcps_ecn_ect0);
1652 case IPTOS_ECN_ECT1:
1653 TCPSTAT_INC(tcps_ecn_ect1);
1657 /* Process a packet differently from RFC3168. */
1658 cc_ecnpkt_handler(tp, th, iptos);
1660 /* Congestion experienced. */
1661 if (thflags & TH_ECE) {
1662 cc_cong_signal(tp, th, CC_ECN);
1667 * Parse options on any incoming segment.
1669 tcp_dooptions(&to, (u_char *)(th + 1),
1670 (th->th_off << 2) - sizeof(struct tcphdr),
1671 (thflags & TH_SYN) ? TO_SYN : 0);
1673 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1674 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1675 (to.to_flags & TOF_SIGNATURE) == 0) {
1676 TCPSTAT_INC(tcps_sig_err_sigopt);
1677 /* XXX: should drop? */
1681 * If echoed timestamp is later than the current time,
1682 * fall back to non RFC1323 RTT calculation. Normalize
1683 * timestamp if syncookies were used when this connection
1686 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1687 to.to_tsecr -= tp->ts_offset;
1688 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1692 * Process options only when we get SYN/ACK back. The SYN case
1693 * for incoming connections is handled in tcp_syncache.
1694 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1695 * or <SYN,ACK>) segment itself is never scaled.
1696 * XXX this is traditional behavior, may need to be cleaned up.
1698 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1699 if ((to.to_flags & TOF_SCALE) &&
1700 (tp->t_flags & TF_REQ_SCALE)) {
1701 tp->t_flags |= TF_RCVD_SCALE;
1702 tp->snd_scale = to.to_wscale;
1705 * Initial send window. It will be updated with
1706 * the next incoming segment to the scaled value.
1708 tp->snd_wnd = th->th_win;
1709 if (to.to_flags & TOF_TS) {
1710 tp->t_flags |= TF_RCVD_TSTMP;
1711 tp->ts_recent = to.to_tsval;
1712 tp->ts_recent_age = tcp_ts_getticks();
1714 if (to.to_flags & TOF_MSS)
1715 tcp_mss(tp, to.to_mss);
1716 if ((tp->t_flags & TF_SACK_PERMIT) &&
1717 (to.to_flags & TOF_SACKPERM) == 0)
1718 tp->t_flags &= ~TF_SACK_PERMIT;
1722 * If timestamps were negotiated during SYN/ACK they should
1723 * appear on every segment during this session and vice versa.
1725 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1726 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1727 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1728 "no action\n", s, __func__);
1732 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1733 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1734 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1735 "no action\n", s, __func__);
1741 * Header prediction: check for the two common cases
1742 * of a uni-directional data xfer. If the packet has
1743 * no control flags, is in-sequence, the window didn't
1744 * change and we're not retransmitting, it's a
1745 * candidate. If the length is zero and the ack moved
1746 * forward, we're the sender side of the xfer. Just
1747 * free the data acked & wake any higher level process
1748 * that was blocked waiting for space. If the length
1749 * is non-zero and the ack didn't move, we're the
1750 * receiver side. If we're getting packets in-order
1751 * (the reassembly queue is empty), add the data to
1752 * the socket buffer and note that we need a delayed ack.
1753 * Make sure that the hidden state-flags are also off.
1754 * Since we check for TCPS_ESTABLISHED first, it can only
1757 if (tp->t_state == TCPS_ESTABLISHED &&
1758 th->th_seq == tp->rcv_nxt &&
1759 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1760 tp->snd_nxt == tp->snd_max &&
1761 tiwin && tiwin == tp->snd_wnd &&
1762 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1763 LIST_EMPTY(&tp->t_segq) &&
1764 ((to.to_flags & TOF_TS) == 0 ||
1765 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1768 * If last ACK falls within this segment's sequence numbers,
1769 * record the timestamp.
1770 * NOTE that the test is modified according to the latest
1771 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1773 if ((to.to_flags & TOF_TS) != 0 &&
1774 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1775 tp->ts_recent_age = tcp_ts_getticks();
1776 tp->ts_recent = to.to_tsval;
1780 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1781 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1782 !IN_RECOVERY(tp->t_flags) &&
1783 (to.to_flags & TOF_SACK) == 0 &&
1784 TAILQ_EMPTY(&tp->snd_holes)) {
1786 * This is a pure ack for outstanding data.
1788 if (ti_locked == TI_RLOCKED)
1789 INP_INFO_RUNLOCK(&V_tcbinfo);
1790 ti_locked = TI_UNLOCKED;
1792 TCPSTAT_INC(tcps_predack);
1795 * "bad retransmit" recovery.
1797 if (tp->t_rxtshift == 1 &&
1798 tp->t_flags & TF_PREVVALID &&
1799 (int)(ticks - tp->t_badrxtwin) < 0) {
1800 cc_cong_signal(tp, th, CC_RTO_ERR);
1804 * Recalculate the transmit timer / rtt.
1806 * Some boxes send broken timestamp replies
1807 * during the SYN+ACK phase, ignore
1808 * timestamps of 0 or we could calculate a
1809 * huge RTT and blow up the retransmit timer.
1811 if ((to.to_flags & TOF_TS) != 0 &&
1815 t = tcp_ts_getticks() - to.to_tsecr;
1816 if (!tp->t_rttlow || tp->t_rttlow > t)
1819 TCP_TS_TO_TICKS(t) + 1);
1820 } else if (tp->t_rtttime &&
1821 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1822 if (!tp->t_rttlow ||
1823 tp->t_rttlow > ticks - tp->t_rtttime)
1824 tp->t_rttlow = ticks - tp->t_rtttime;
1826 ticks - tp->t_rtttime);
1828 acked = BYTES_THIS_ACK(tp, th);
1831 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1832 hhook_run_tcp_est_in(tp, th, &to);
1835 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1836 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1837 sbdrop(&so->so_snd, acked);
1838 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1839 SEQ_LEQ(th->th_ack, tp->snd_recover))
1840 tp->snd_recover = th->th_ack - 1;
1843 * Let the congestion control algorithm update
1844 * congestion control related information. This
1845 * typically means increasing the congestion
1848 cc_ack_received(tp, th, nsegs, CC_ACK);
1850 tp->snd_una = th->th_ack;
1852 * Pull snd_wl2 up to prevent seq wrap relative
1855 tp->snd_wl2 = th->th_ack;
1860 * If all outstanding data are acked, stop
1861 * retransmit timer, otherwise restart timer
1862 * using current (possibly backed-off) value.
1863 * If process is waiting for space,
1864 * wakeup/selwakeup/signal. If data
1865 * are ready to send, let tcp_output
1866 * decide between more output or persist.
1869 if (so->so_options & SO_DEBUG)
1870 tcp_trace(TA_INPUT, ostate, tp,
1871 (void *)tcp_saveipgen,
1874 TCP_PROBE3(debug__input, tp, th, m);
1875 if (tp->snd_una == tp->snd_max)
1876 tcp_timer_activate(tp, TT_REXMT, 0);
1877 else if (!tcp_timer_active(tp, TT_PERSIST))
1878 tcp_timer_activate(tp, TT_REXMT,
1881 if (sbavail(&so->so_snd))
1882 (void) tp->t_fb->tfb_tcp_output(tp);
1885 } else if (th->th_ack == tp->snd_una &&
1886 tlen <= sbspace(&so->so_rcv)) {
1887 int newsize = 0; /* automatic sockbuf scaling */
1890 * This is a pure, in-sequence data packet with
1891 * nothing on the reassembly queue and we have enough
1892 * buffer space to take it.
1894 if (ti_locked == TI_RLOCKED)
1895 INP_INFO_RUNLOCK(&V_tcbinfo);
1896 ti_locked = TI_UNLOCKED;
1898 /* Clean receiver SACK report if present */
1899 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1900 tcp_clean_sackreport(tp);
1901 TCPSTAT_INC(tcps_preddat);
1902 tp->rcv_nxt += tlen;
1904 * Pull snd_wl1 up to prevent seq wrap relative to
1907 tp->snd_wl1 = th->th_seq;
1909 * Pull rcv_up up to prevent seq wrap relative to
1912 tp->rcv_up = tp->rcv_nxt;
1913 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1914 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1916 if (so->so_options & SO_DEBUG)
1917 tcp_trace(TA_INPUT, ostate, tp,
1918 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1920 TCP_PROBE3(debug__input, tp, th, m);
1922 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1924 /* Add data to socket buffer. */
1925 SOCKBUF_LOCK(&so->so_rcv);
1926 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1930 * Set new socket buffer size.
1931 * Give up when limit is reached.
1934 if (!sbreserve_locked(&so->so_rcv,
1936 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1937 m_adj(m, drop_hdrlen); /* delayed header drop */
1938 sbappendstream_locked(&so->so_rcv, m, 0);
1940 /* NB: sorwakeup_locked() does an implicit unlock. */
1941 sorwakeup_locked(so);
1942 if (DELAY_ACK(tp, tlen)) {
1943 tp->t_flags |= TF_DELACK;
1945 tp->t_flags |= TF_ACKNOW;
1946 tp->t_fb->tfb_tcp_output(tp);
1953 * Calculate amount of space in receive window,
1954 * and then do TCP input processing.
1955 * Receive window is amount of space in rcv queue,
1956 * but not less than advertised window.
1958 win = sbspace(&so->so_rcv);
1961 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1963 switch (tp->t_state) {
1966 * If the state is SYN_RECEIVED:
1967 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1969 case TCPS_SYN_RECEIVED:
1970 if ((thflags & TH_ACK) &&
1971 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1972 SEQ_GT(th->th_ack, tp->snd_max))) {
1973 rstreason = BANDLIM_RST_OPENPORT;
1977 if (IS_FASTOPEN(tp->t_flags)) {
1979 * When a TFO connection is in SYN_RECEIVED, the
1980 * only valid packets are the initial SYN, a
1981 * retransmit/copy of the initial SYN (possibly with
1982 * a subset of the original data), a valid ACK, a
1985 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1986 rstreason = BANDLIM_RST_OPENPORT;
1988 } else if (thflags & TH_SYN) {
1989 /* non-initial SYN is ignored */
1990 if ((tcp_timer_active(tp, TT_DELACK) ||
1991 tcp_timer_active(tp, TT_REXMT)))
1993 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2001 * If the state is SYN_SENT:
2002 * if seg contains a RST with valid ACK (SEQ.ACK has already
2003 * been verified), then drop the connection.
2004 * if seg contains a RST without an ACK, drop the seg.
2005 * if seg does not contain SYN, then drop the seg.
2006 * Otherwise this is an acceptable SYN segment
2007 * initialize tp->rcv_nxt and tp->irs
2008 * if seg contains ack then advance tp->snd_una
2009 * if seg contains an ECE and ECN support is enabled, the stream
2011 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2012 * arrange for segment to be acked (eventually)
2013 * continue processing rest of data/controls, beginning with URG
2016 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2017 TCP_PROBE5(connect__refused, NULL, tp,
2019 tp = tcp_drop(tp, ECONNREFUSED);
2021 if (thflags & TH_RST)
2023 if (!(thflags & TH_SYN))
2026 tp->irs = th->th_seq;
2028 if (thflags & TH_ACK) {
2029 TCPSTAT_INC(tcps_connects);
2032 mac_socketpeer_set_from_mbuf(m, so);
2034 /* Do window scaling on this connection? */
2035 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2036 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2037 tp->rcv_scale = tp->request_r_scale;
2039 tp->rcv_adv += min(tp->rcv_wnd,
2040 TCP_MAXWIN << tp->rcv_scale);
2041 tp->snd_una++; /* SYN is acked */
2043 * If there's data, delay ACK; if there's also a FIN
2044 * ACKNOW will be turned on later.
2046 if (DELAY_ACK(tp, tlen) && tlen != 0)
2047 tcp_timer_activate(tp, TT_DELACK,
2050 tp->t_flags |= TF_ACKNOW;
2052 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
2053 tp->t_flags |= TF_ECN_PERMIT;
2054 TCPSTAT_INC(tcps_ecn_shs);
2058 * Received <SYN,ACK> in SYN_SENT[*] state.
2060 * SYN_SENT --> ESTABLISHED
2061 * SYN_SENT* --> FIN_WAIT_1
2063 tp->t_starttime = ticks;
2064 if (tp->t_flags & TF_NEEDFIN) {
2065 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2066 tp->t_flags &= ~TF_NEEDFIN;
2069 tcp_state_change(tp, TCPS_ESTABLISHED);
2070 TCP_PROBE5(connect__established, NULL, tp,
2073 tcp_timer_activate(tp, TT_KEEP,
2078 * Received initial SYN in SYN-SENT[*] state =>
2079 * simultaneous open.
2080 * If it succeeds, connection is * half-synchronized.
2081 * Otherwise, do 3-way handshake:
2082 * SYN-SENT -> SYN-RECEIVED
2083 * SYN-SENT* -> SYN-RECEIVED*
2085 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2086 tcp_timer_activate(tp, TT_REXMT, 0);
2087 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2090 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: "
2091 "ti_locked %d", __func__, ti_locked));
2092 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2093 INP_WLOCK_ASSERT(tp->t_inpcb);
2096 * Advance th->th_seq to correspond to first data byte.
2097 * If data, trim to stay within window,
2098 * dropping FIN if necessary.
2101 if (tlen > tp->rcv_wnd) {
2102 todrop = tlen - tp->rcv_wnd;
2106 TCPSTAT_INC(tcps_rcvpackafterwin);
2107 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2109 tp->snd_wl1 = th->th_seq - 1;
2110 tp->rcv_up = th->th_seq;
2112 * Client side of transaction: already sent SYN and data.
2113 * If the remote host used T/TCP to validate the SYN,
2114 * our data will be ACK'd; if so, enter normal data segment
2115 * processing in the middle of step 5, ack processing.
2116 * Otherwise, goto step 6.
2118 if (thflags & TH_ACK)
2124 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2125 * do normal processing.
2127 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2131 break; /* continue normal processing */
2135 * States other than LISTEN or SYN_SENT.
2136 * First check the RST flag and sequence number since reset segments
2137 * are exempt from the timestamp and connection count tests. This
2138 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2139 * below which allowed reset segments in half the sequence space
2140 * to fall though and be processed (which gives forged reset
2141 * segments with a random sequence number a 50 percent chance of
2142 * killing a connection).
2143 * Then check timestamp, if present.
2144 * Then check the connection count, if present.
2145 * Then check that at least some bytes of segment are within
2146 * receive window. If segment begins before rcv_nxt,
2147 * drop leading data (and SYN); if nothing left, just ack.
2149 if (thflags & TH_RST) {
2151 * RFC5961 Section 3.2
2153 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2154 * - If RST is in window, we send challenge ACK.
2156 * Note: to take into account delayed ACKs, we should
2157 * test against last_ack_sent instead of rcv_nxt.
2158 * Note 2: we handle special case of closed window, not
2159 * covered by the RFC.
2161 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2162 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2163 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2165 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2166 KASSERT(ti_locked == TI_RLOCKED,
2167 ("%s: TH_RST ti_locked %d, th %p tp %p",
2168 __func__, ti_locked, th, tp));
2169 KASSERT(tp->t_state != TCPS_SYN_SENT,
2170 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2173 if (V_tcp_insecure_rst ||
2174 tp->last_ack_sent == th->th_seq) {
2175 TCPSTAT_INC(tcps_drops);
2176 /* Drop the connection. */
2177 switch (tp->t_state) {
2178 case TCPS_SYN_RECEIVED:
2179 so->so_error = ECONNREFUSED;
2181 case TCPS_ESTABLISHED:
2182 case TCPS_FIN_WAIT_1:
2183 case TCPS_FIN_WAIT_2:
2184 case TCPS_CLOSE_WAIT:
2187 so->so_error = ECONNRESET;
2194 TCPSTAT_INC(tcps_badrst);
2195 /* Send challenge ACK. */
2196 tcp_respond(tp, mtod(m, void *), th, m,
2197 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2198 tp->last_ack_sent = tp->rcv_nxt;
2206 * RFC5961 Section 4.2
2207 * Send challenge ACK for any SYN in synchronized state.
2209 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2210 tp->t_state != TCPS_SYN_RECEIVED) {
2211 KASSERT(ti_locked == TI_RLOCKED,
2212 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2213 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2215 TCPSTAT_INC(tcps_badsyn);
2216 if (V_tcp_insecure_syn &&
2217 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2218 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2219 tp = tcp_drop(tp, ECONNRESET);
2220 rstreason = BANDLIM_UNLIMITED;
2222 /* Send challenge ACK. */
2223 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2224 tp->snd_nxt, TH_ACK);
2225 tp->last_ack_sent = tp->rcv_nxt;
2232 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2233 * and it's less than ts_recent, drop it.
2235 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2236 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2238 /* Check to see if ts_recent is over 24 days old. */
2239 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2241 * Invalidate ts_recent. If this segment updates
2242 * ts_recent, the age will be reset later and ts_recent
2243 * will get a valid value. If it does not, setting
2244 * ts_recent to zero will at least satisfy the
2245 * requirement that zero be placed in the timestamp
2246 * echo reply when ts_recent isn't valid. The
2247 * age isn't reset until we get a valid ts_recent
2248 * because we don't want out-of-order segments to be
2249 * dropped when ts_recent is old.
2253 TCPSTAT_INC(tcps_rcvduppack);
2254 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2255 TCPSTAT_INC(tcps_pawsdrop);
2263 * In the SYN-RECEIVED state, validate that the packet belongs to
2264 * this connection before trimming the data to fit the receive
2265 * window. Check the sequence number versus IRS since we know
2266 * the sequence numbers haven't wrapped. This is a partial fix
2267 * for the "LAND" DoS attack.
2269 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2270 rstreason = BANDLIM_RST_OPENPORT;
2274 todrop = tp->rcv_nxt - th->th_seq;
2276 if (thflags & TH_SYN) {
2286 * Following if statement from Stevens, vol. 2, p. 960.
2289 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2291 * Any valid FIN must be to the left of the window.
2292 * At this point the FIN must be a duplicate or out
2293 * of sequence; drop it.
2298 * Send an ACK to resynchronize and drop any data.
2299 * But keep on processing for RST or ACK.
2301 tp->t_flags |= TF_ACKNOW;
2303 TCPSTAT_INC(tcps_rcvduppack);
2304 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2306 TCPSTAT_INC(tcps_rcvpartduppack);
2307 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2309 drop_hdrlen += todrop; /* drop from the top afterwards */
2310 th->th_seq += todrop;
2312 if (th->th_urp > todrop)
2313 th->th_urp -= todrop;
2321 * If new data are received on a connection after the
2322 * user processes are gone, then RST the other end.
2324 if ((so->so_state & SS_NOFDREF) &&
2325 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2326 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && "
2327 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2328 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2330 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2331 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2332 "after socket was closed, "
2333 "sending RST and removing tcpcb\n",
2334 s, __func__, tcpstates[tp->t_state], tlen);
2338 TCPSTAT_INC(tcps_rcvafterclose);
2339 rstreason = BANDLIM_UNLIMITED;
2344 * If segment ends after window, drop trailing data
2345 * (and PUSH and FIN); if nothing left, just ACK.
2347 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2349 TCPSTAT_INC(tcps_rcvpackafterwin);
2350 if (todrop >= tlen) {
2351 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2353 * If window is closed can only take segments at
2354 * window edge, and have to drop data and PUSH from
2355 * incoming segments. Continue processing, but
2356 * remember to ack. Otherwise, drop segment
2359 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2360 tp->t_flags |= TF_ACKNOW;
2361 TCPSTAT_INC(tcps_rcvwinprobe);
2365 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2368 thflags &= ~(TH_PUSH|TH_FIN);
2372 * If last ACK falls within this segment's sequence numbers,
2373 * record its timestamp.
2375 * 1) That the test incorporates suggestions from the latest
2376 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2377 * 2) That updating only on newer timestamps interferes with
2378 * our earlier PAWS tests, so this check should be solely
2379 * predicated on the sequence space of this segment.
2380 * 3) That we modify the segment boundary check to be
2381 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2382 * instead of RFC1323's
2383 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2384 * This modified check allows us to overcome RFC1323's
2385 * limitations as described in Stevens TCP/IP Illustrated
2386 * Vol. 2 p.869. In such cases, we can still calculate the
2387 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2389 if ((to.to_flags & TOF_TS) != 0 &&
2390 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2391 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2392 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2393 tp->ts_recent_age = tcp_ts_getticks();
2394 tp->ts_recent = to.to_tsval;
2398 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2399 * flag is on (half-synchronized state), then queue data for
2400 * later processing; else drop segment and return.
2402 if ((thflags & TH_ACK) == 0) {
2403 if (tp->t_state == TCPS_SYN_RECEIVED ||
2404 (tp->t_flags & TF_NEEDSYN)) {
2406 if (tp->t_state == TCPS_SYN_RECEIVED &&
2407 IS_FASTOPEN(tp->t_flags)) {
2408 tp->snd_wnd = tiwin;
2413 } else if (tp->t_flags & TF_ACKNOW)
2422 switch (tp->t_state) {
2425 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2426 * ESTABLISHED state and continue processing.
2427 * The ACK was checked above.
2429 case TCPS_SYN_RECEIVED:
2431 TCPSTAT_INC(tcps_connects);
2433 /* Do window scaling? */
2434 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2435 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2436 tp->rcv_scale = tp->request_r_scale;
2437 tp->snd_wnd = tiwin;
2441 * SYN-RECEIVED -> ESTABLISHED
2442 * SYN-RECEIVED* -> FIN-WAIT-1
2444 tp->t_starttime = ticks;
2445 if (tp->t_flags & TF_NEEDFIN) {
2446 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2447 tp->t_flags &= ~TF_NEEDFIN;
2449 tcp_state_change(tp, TCPS_ESTABLISHED);
2450 TCP_PROBE5(accept__established, NULL, tp,
2453 if (tp->t_tfo_pending) {
2454 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2455 tp->t_tfo_pending = NULL;
2458 * Account for the ACK of our SYN prior to
2459 * regular ACK processing below.
2464 * TFO connections call cc_conn_init() during SYN
2465 * processing. Calling it again here for such
2466 * connections is not harmless as it would undo the
2467 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2470 if (!IS_FASTOPEN(tp->t_flags))
2473 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2476 * If segment contains data or ACK, will call tcp_reass()
2477 * later; if not, do so now to pass queued data to user.
2479 if (tlen == 0 && (thflags & TH_FIN) == 0)
2480 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2482 tp->snd_wl1 = th->th_seq - 1;
2486 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2487 * ACKs. If the ack is in the range
2488 * tp->snd_una < th->th_ack <= tp->snd_max
2489 * then advance tp->snd_una to th->th_ack and drop
2490 * data from the retransmission queue. If this ACK reflects
2491 * more up to date window information we update our window information.
2493 case TCPS_ESTABLISHED:
2494 case TCPS_FIN_WAIT_1:
2495 case TCPS_FIN_WAIT_2:
2496 case TCPS_CLOSE_WAIT:
2499 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2500 TCPSTAT_INC(tcps_rcvacktoomuch);
2503 if ((tp->t_flags & TF_SACK_PERMIT) &&
2504 ((to.to_flags & TOF_SACK) ||
2505 !TAILQ_EMPTY(&tp->snd_holes)))
2506 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2509 * Reset the value so that previous (valid) value
2510 * from the last ack with SACK doesn't get used.
2512 tp->sackhint.sacked_bytes = 0;
2515 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2516 hhook_run_tcp_est_in(tp, th, &to);
2519 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2522 maxseg = tcp_maxseg(tp);
2524 (tiwin == tp->snd_wnd ||
2525 (tp->t_flags & TF_SACK_PERMIT))) {
2527 * If this is the first time we've seen a
2528 * FIN from the remote, this is not a
2529 * duplicate and it needs to be processed
2530 * normally. This happens during a
2531 * simultaneous close.
2533 if ((thflags & TH_FIN) &&
2534 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2538 TCPSTAT_INC(tcps_rcvdupack);
2540 * If we have outstanding data (other than
2541 * a window probe), this is a completely
2542 * duplicate ack (ie, window info didn't
2543 * change and FIN isn't set),
2544 * the ack is the biggest we've
2545 * seen and we've seen exactly our rexmt
2546 * threshold of them, assume a packet
2547 * has been dropped and retransmit it.
2548 * Kludge snd_nxt & the congestion
2549 * window so we send only this one
2552 * We know we're losing at the current
2553 * window size so do congestion avoidance
2554 * (set ssthresh to half the current window
2555 * and pull our congestion window back to
2556 * the new ssthresh).
2558 * Dup acks mean that packets have left the
2559 * network (they're now cached at the receiver)
2560 * so bump cwnd by the amount in the receiver
2561 * to keep a constant cwnd packets in the
2564 * When using TCP ECN, notify the peer that
2565 * we reduced the cwnd.
2568 * Following 2 kinds of acks should not affect
2571 * 2) Acks with SACK but without any new SACK
2572 * information in them. These could result from
2573 * any anomaly in the network like a switch
2574 * duplicating packets or a possible DoS attack.
2576 if (th->th_ack != tp->snd_una ||
2577 ((tp->t_flags & TF_SACK_PERMIT) &&
2580 else if (!tcp_timer_active(tp, TT_REXMT))
2582 else if (++tp->t_dupacks > tcprexmtthresh ||
2583 IN_FASTRECOVERY(tp->t_flags)) {
2584 cc_ack_received(tp, th, nsegs,
2586 if ((tp->t_flags & TF_SACK_PERMIT) &&
2587 IN_FASTRECOVERY(tp->t_flags)) {
2591 * Compute the amount of data in flight first.
2592 * We can inject new data into the pipe iff
2593 * we have less than 1/2 the original window's
2594 * worth of data in flight.
2596 if (V_tcp_do_rfc6675_pipe)
2597 awnd = tcp_compute_pipe(tp);
2599 awnd = (tp->snd_nxt - tp->snd_fack) +
2600 tp->sackhint.sack_bytes_rexmit;
2602 if (awnd < tp->snd_ssthresh) {
2603 tp->snd_cwnd += maxseg;
2604 if (tp->snd_cwnd > tp->snd_ssthresh)
2605 tp->snd_cwnd = tp->snd_ssthresh;
2608 tp->snd_cwnd += maxseg;
2609 (void) tp->t_fb->tfb_tcp_output(tp);
2611 } else if (tp->t_dupacks == tcprexmtthresh) {
2612 tcp_seq onxt = tp->snd_nxt;
2615 * If we're doing sack, check to
2616 * see if we're already in sack
2617 * recovery. If we're not doing sack,
2618 * check to see if we're in newreno
2621 if (tp->t_flags & TF_SACK_PERMIT) {
2622 if (IN_FASTRECOVERY(tp->t_flags)) {
2627 if (SEQ_LEQ(th->th_ack,
2633 /* Congestion signal before ack. */
2634 cc_cong_signal(tp, th, CC_NDUPACK);
2635 cc_ack_received(tp, th, nsegs,
2637 tcp_timer_activate(tp, TT_REXMT, 0);
2639 if (tp->t_flags & TF_SACK_PERMIT) {
2641 tcps_sack_recovery_episode);
2642 tp->sack_newdata = tp->snd_nxt;
2643 tp->snd_cwnd = maxseg;
2644 (void) tp->t_fb->tfb_tcp_output(tp);
2647 tp->snd_nxt = th->th_ack;
2648 tp->snd_cwnd = maxseg;
2649 (void) tp->t_fb->tfb_tcp_output(tp);
2650 KASSERT(tp->snd_limited <= 2,
2651 ("%s: tp->snd_limited too big",
2653 tp->snd_cwnd = tp->snd_ssthresh +
2655 (tp->t_dupacks - tp->snd_limited);
2656 if (SEQ_GT(onxt, tp->snd_nxt))
2659 } else if (V_tcp_do_rfc3042) {
2661 * Process first and second duplicate
2662 * ACKs. Each indicates a segment
2663 * leaving the network, creating room
2664 * for more. Make sure we can send a
2665 * packet on reception of each duplicate
2666 * ACK by increasing snd_cwnd by one
2667 * segment. Restore the original
2668 * snd_cwnd after packet transmission.
2670 cc_ack_received(tp, th, nsegs,
2672 uint32_t oldcwnd = tp->snd_cwnd;
2673 tcp_seq oldsndmax = tp->snd_max;
2677 KASSERT(tp->t_dupacks == 1 ||
2679 ("%s: dupacks not 1 or 2",
2681 if (tp->t_dupacks == 1)
2682 tp->snd_limited = 0;
2684 (tp->snd_nxt - tp->snd_una) +
2685 (tp->t_dupacks - tp->snd_limited) *
2688 * Only call tcp_output when there
2689 * is new data available to be sent.
2690 * Otherwise we would send pure ACKs.
2692 SOCKBUF_LOCK(&so->so_snd);
2693 avail = sbavail(&so->so_snd) -
2694 (tp->snd_nxt - tp->snd_una);
2695 SOCKBUF_UNLOCK(&so->so_snd);
2697 (void) tp->t_fb->tfb_tcp_output(tp);
2698 sent = tp->snd_max - oldsndmax;
2699 if (sent > maxseg) {
2700 KASSERT((tp->t_dupacks == 2 &&
2701 tp->snd_limited == 0) ||
2702 (sent == maxseg + 1 &&
2703 tp->t_flags & TF_SENTFIN),
2704 ("%s: sent too much",
2706 tp->snd_limited = 2;
2707 } else if (sent > 0)
2709 tp->snd_cwnd = oldcwnd;
2716 * This ack is advancing the left edge, reset the
2721 * If this ack also has new SACK info, increment the
2722 * counter as per rfc6675.
2724 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed)
2728 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2729 ("%s: th_ack <= snd_una", __func__));
2732 * If the congestion window was inflated to account
2733 * for the other side's cached packets, retract it.
2735 if (IN_FASTRECOVERY(tp->t_flags)) {
2736 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2737 if (tp->t_flags & TF_SACK_PERMIT)
2738 tcp_sack_partialack(tp, th);
2740 tcp_newreno_partial_ack(tp, th);
2742 cc_post_recovery(tp, th);
2745 * If we reach this point, ACK is not a duplicate,
2746 * i.e., it ACKs something we sent.
2748 if (tp->t_flags & TF_NEEDSYN) {
2750 * T/TCP: Connection was half-synchronized, and our
2751 * SYN has been ACK'd (so connection is now fully
2752 * synchronized). Go to non-starred state,
2753 * increment snd_una for ACK of SYN, and check if
2754 * we can do window scaling.
2756 tp->t_flags &= ~TF_NEEDSYN;
2758 /* Do window scaling? */
2759 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2760 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2761 tp->rcv_scale = tp->request_r_scale;
2762 /* Send window already scaled. */
2767 INP_WLOCK_ASSERT(tp->t_inpcb);
2769 acked = BYTES_THIS_ACK(tp, th);
2770 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2771 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2772 tp->snd_una, th->th_ack, tp, m));
2773 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2774 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2777 * If we just performed our first retransmit, and the ACK
2778 * arrives within our recovery window, then it was a mistake
2779 * to do the retransmit in the first place. Recover our
2780 * original cwnd and ssthresh, and proceed to transmit where
2783 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2784 (int)(ticks - tp->t_badrxtwin) < 0)
2785 cc_cong_signal(tp, th, CC_RTO_ERR);
2788 * If we have a timestamp reply, update smoothed
2789 * round trip time. If no timestamp is present but
2790 * transmit timer is running and timed sequence
2791 * number was acked, update smoothed round trip time.
2792 * Since we now have an rtt measurement, cancel the
2793 * timer backoff (cf., Phil Karn's retransmit alg.).
2794 * Recompute the initial retransmit timer.
2796 * Some boxes send broken timestamp replies
2797 * during the SYN+ACK phase, ignore
2798 * timestamps of 0 or we could calculate a
2799 * huge RTT and blow up the retransmit timer.
2801 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2804 t = tcp_ts_getticks() - to.to_tsecr;
2805 if (!tp->t_rttlow || tp->t_rttlow > t)
2807 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2808 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2809 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2810 tp->t_rttlow = ticks - tp->t_rtttime;
2811 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2815 * If all outstanding data is acked, stop retransmit
2816 * timer and remember to restart (more output or persist).
2817 * If there is more data to be acked, restart retransmit
2818 * timer, using current (possibly backed-off) value.
2820 if (th->th_ack == tp->snd_max) {
2821 tcp_timer_activate(tp, TT_REXMT, 0);
2823 } else if (!tcp_timer_active(tp, TT_PERSIST))
2824 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2827 * If no data (only SYN) was ACK'd,
2828 * skip rest of ACK processing.
2834 * Let the congestion control algorithm update congestion
2835 * control related information. This typically means increasing
2836 * the congestion window.
2838 cc_ack_received(tp, th, nsegs, CC_ACK);
2840 SOCKBUF_LOCK(&so->so_snd);
2841 if (acked > sbavail(&so->so_snd)) {
2842 if (tp->snd_wnd >= sbavail(&so->so_snd))
2843 tp->snd_wnd -= sbavail(&so->so_snd);
2846 mfree = sbcut_locked(&so->so_snd,
2847 (int)sbavail(&so->so_snd));
2850 mfree = sbcut_locked(&so->so_snd, acked);
2851 if (tp->snd_wnd >= (uint32_t) acked)
2852 tp->snd_wnd -= acked;
2857 /* NB: sowwakeup_locked() does an implicit unlock. */
2858 sowwakeup_locked(so);
2860 /* Detect una wraparound. */
2861 if (!IN_RECOVERY(tp->t_flags) &&
2862 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2863 SEQ_LEQ(th->th_ack, tp->snd_recover))
2864 tp->snd_recover = th->th_ack - 1;
2865 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2866 if (IN_RECOVERY(tp->t_flags) &&
2867 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2868 EXIT_RECOVERY(tp->t_flags);
2870 tp->snd_una = th->th_ack;
2871 if (tp->t_flags & TF_SACK_PERMIT) {
2872 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2873 tp->snd_recover = tp->snd_una;
2875 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2876 tp->snd_nxt = tp->snd_una;
2878 switch (tp->t_state) {
2881 * In FIN_WAIT_1 STATE in addition to the processing
2882 * for the ESTABLISHED state if our FIN is now acknowledged
2883 * then enter FIN_WAIT_2.
2885 case TCPS_FIN_WAIT_1:
2886 if (ourfinisacked) {
2888 * If we can't receive any more
2889 * data, then closing user can proceed.
2890 * Starting the timer is contrary to the
2891 * specification, but if we don't get a FIN
2892 * we'll hang forever.
2895 * we should release the tp also, and use a
2898 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2899 soisdisconnected(so);
2900 tcp_timer_activate(tp, TT_2MSL,
2901 (tcp_fast_finwait2_recycle ?
2902 tcp_finwait2_timeout :
2905 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2910 * In CLOSING STATE in addition to the processing for
2911 * the ESTABLISHED state if the ACK acknowledges our FIN
2912 * then enter the TIME-WAIT state, otherwise ignore
2916 if (ourfinisacked) {
2917 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2919 INP_INFO_RUNLOCK(&V_tcbinfo);
2926 * In LAST_ACK, we may still be waiting for data to drain
2927 * and/or to be acked, as well as for the ack of our FIN.
2928 * If our FIN is now acknowledged, delete the TCB,
2929 * enter the closed state and return.
2932 if (ourfinisacked) {
2933 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2942 INP_WLOCK_ASSERT(tp->t_inpcb);
2945 * Update window information.
2946 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2948 if ((thflags & TH_ACK) &&
2949 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2950 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2951 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2952 /* keep track of pure window updates */
2954 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2955 TCPSTAT_INC(tcps_rcvwinupd);
2956 tp->snd_wnd = tiwin;
2957 tp->snd_wl1 = th->th_seq;
2958 tp->snd_wl2 = th->th_ack;
2959 if (tp->snd_wnd > tp->max_sndwnd)
2960 tp->max_sndwnd = tp->snd_wnd;
2965 * Process segments with URG.
2967 if ((thflags & TH_URG) && th->th_urp &&
2968 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2970 * This is a kludge, but if we receive and accept
2971 * random urgent pointers, we'll crash in
2972 * soreceive. It's hard to imagine someone
2973 * actually wanting to send this much urgent data.
2975 SOCKBUF_LOCK(&so->so_rcv);
2976 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2977 th->th_urp = 0; /* XXX */
2978 thflags &= ~TH_URG; /* XXX */
2979 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2980 goto dodata; /* XXX */
2983 * If this segment advances the known urgent pointer,
2984 * then mark the data stream. This should not happen
2985 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2986 * a FIN has been received from the remote side.
2987 * In these states we ignore the URG.
2989 * According to RFC961 (Assigned Protocols),
2990 * the urgent pointer points to the last octet
2991 * of urgent data. We continue, however,
2992 * to consider it to indicate the first octet
2993 * of data past the urgent section as the original
2994 * spec states (in one of two places).
2996 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2997 tp->rcv_up = th->th_seq + th->th_urp;
2998 so->so_oobmark = sbavail(&so->so_rcv) +
2999 (tp->rcv_up - tp->rcv_nxt) - 1;
3000 if (so->so_oobmark == 0)
3001 so->so_rcv.sb_state |= SBS_RCVATMARK;
3003 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3005 SOCKBUF_UNLOCK(&so->so_rcv);
3007 * Remove out of band data so doesn't get presented to user.
3008 * This can happen independent of advancing the URG pointer,
3009 * but if two URG's are pending at once, some out-of-band
3010 * data may creep in... ick.
3012 if (th->th_urp <= (uint32_t)tlen &&
3013 !(so->so_options & SO_OOBINLINE)) {
3014 /* hdr drop is delayed */
3015 tcp_pulloutofband(so, th, m, drop_hdrlen);
3019 * If no out of band data is expected,
3020 * pull receive urgent pointer along
3021 * with the receive window.
3023 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3024 tp->rcv_up = tp->rcv_nxt;
3027 INP_WLOCK_ASSERT(tp->t_inpcb);
3030 * Process the segment text, merging it into the TCP sequencing queue,
3031 * and arranging for acknowledgment of receipt if necessary.
3032 * This process logically involves adjusting tp->rcv_wnd as data
3033 * is presented to the user (this happens in tcp_usrreq.c,
3034 * case PRU_RCVD). If a FIN has already been received on this
3035 * connection then we just ignore the text.
3038 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3039 IS_FASTOPEN(tp->t_flags));
3041 #define tfo_syn (false)
3043 if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
3044 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3045 tcp_seq save_start = th->th_seq;
3046 m_adj(m, drop_hdrlen); /* delayed header drop */
3048 * Insert segment which includes th into TCP reassembly queue
3049 * with control block tp. Set thflags to whether reassembly now
3050 * includes a segment with FIN. This handles the common case
3051 * inline (segment is the next to be received on an established
3052 * connection, and the queue is empty), avoiding linkage into
3053 * and removal from the queue and repetition of various
3055 * Set DELACK for segments received in order, but ack
3056 * immediately when segments are out of order (so
3057 * fast retransmit can work).
3059 if (th->th_seq == tp->rcv_nxt &&
3060 LIST_EMPTY(&tp->t_segq) &&
3061 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3063 if (DELAY_ACK(tp, tlen) || tfo_syn)
3064 tp->t_flags |= TF_DELACK;
3066 tp->t_flags |= TF_ACKNOW;
3067 tp->rcv_nxt += tlen;
3068 thflags = th->th_flags & TH_FIN;
3069 TCPSTAT_INC(tcps_rcvpack);
3070 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3071 SOCKBUF_LOCK(&so->so_rcv);
3072 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3075 sbappendstream_locked(&so->so_rcv, m, 0);
3076 /* NB: sorwakeup_locked() does an implicit unlock. */
3077 sorwakeup_locked(so);
3080 * XXX: Due to the header drop above "th" is
3081 * theoretically invalid by now. Fortunately
3082 * m_adj() doesn't actually frees any mbufs
3083 * when trimming from the head.
3085 thflags = tcp_reass(tp, th, &tlen, m);
3086 tp->t_flags |= TF_ACKNOW;
3088 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
3089 tcp_update_sack_list(tp, save_start, save_start + tlen);
3092 * Note the amount of data that peer has sent into
3093 * our window, in order to estimate the sender's
3097 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3098 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3100 len = so->so_rcv.sb_hiwat;
3108 * If FIN is received ACK the FIN and let the user know
3109 * that the connection is closing.
3111 if (thflags & TH_FIN) {
3112 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3115 * If connection is half-synchronized
3116 * (ie NEEDSYN flag on) then delay ACK,
3117 * so it may be piggybacked when SYN is sent.
3118 * Otherwise, since we received a FIN then no
3119 * more input can be expected, send ACK now.
3121 if (tp->t_flags & TF_NEEDSYN)
3122 tp->t_flags |= TF_DELACK;
3124 tp->t_flags |= TF_ACKNOW;
3127 switch (tp->t_state) {
3130 * In SYN_RECEIVED and ESTABLISHED STATES
3131 * enter the CLOSE_WAIT state.
3133 case TCPS_SYN_RECEIVED:
3134 tp->t_starttime = ticks;
3136 case TCPS_ESTABLISHED:
3137 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3141 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3142 * enter the CLOSING state.
3144 case TCPS_FIN_WAIT_1:
3145 tcp_state_change(tp, TCPS_CLOSING);
3149 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3150 * starting the time-wait timer, turning off the other
3153 case TCPS_FIN_WAIT_2:
3154 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
3155 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata "
3156 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
3160 INP_INFO_RUNLOCK(&V_tcbinfo);
3164 if (ti_locked == TI_RLOCKED)
3165 INP_INFO_RUNLOCK(&V_tcbinfo);
3166 ti_locked = TI_UNLOCKED;
3169 if (so->so_options & SO_DEBUG)
3170 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3173 TCP_PROBE3(debug__input, tp, th, m);
3176 * Return any desired output.
3178 if (needoutput || (tp->t_flags & TF_ACKNOW))
3179 (void) tp->t_fb->tfb_tcp_output(tp);
3182 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3183 __func__, ti_locked));
3184 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3185 INP_WLOCK_ASSERT(tp->t_inpcb);
3187 if (tp->t_flags & TF_DELACK) {
3188 tp->t_flags &= ~TF_DELACK;
3189 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3191 INP_WUNLOCK(tp->t_inpcb);
3196 * Generate an ACK dropping incoming segment if it occupies
3197 * sequence space, where the ACK reflects our state.
3199 * We can now skip the test for the RST flag since all
3200 * paths to this code happen after packets containing
3201 * RST have been dropped.
3203 * In the SYN-RECEIVED state, don't send an ACK unless the
3204 * segment we received passes the SYN-RECEIVED ACK test.
3205 * If it fails send a RST. This breaks the loop in the
3206 * "LAND" DoS attack, and also prevents an ACK storm
3207 * between two listening ports that have been sent forged
3208 * SYN segments, each with the source address of the other.
3210 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3211 (SEQ_GT(tp->snd_una, th->th_ack) ||
3212 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3213 rstreason = BANDLIM_RST_OPENPORT;
3217 if (so->so_options & SO_DEBUG)
3218 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3221 TCP_PROBE3(debug__input, tp, th, m);
3222 if (ti_locked == TI_RLOCKED)
3223 INP_INFO_RUNLOCK(&V_tcbinfo);
3224 ti_locked = TI_UNLOCKED;
3226 tp->t_flags |= TF_ACKNOW;
3227 (void) tp->t_fb->tfb_tcp_output(tp);
3228 INP_WUNLOCK(tp->t_inpcb);
3233 if (ti_locked == TI_RLOCKED)
3234 INP_INFO_RUNLOCK(&V_tcbinfo);
3235 ti_locked = TI_UNLOCKED;
3238 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3239 INP_WUNLOCK(tp->t_inpcb);
3241 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3245 if (ti_locked == TI_RLOCKED) {
3246 INP_INFO_RUNLOCK(&V_tcbinfo);
3247 ti_locked = TI_UNLOCKED;
3251 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3255 * Drop space held by incoming segment and return.
3258 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3259 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3262 TCP_PROBE3(debug__input, tp, th, m);
3264 INP_WUNLOCK(tp->t_inpcb);
3272 * Issue RST and make ACK acceptable to originator of segment.
3273 * The mbuf must still include the original packet header.
3277 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3278 int tlen, int rstreason)
3284 struct ip6_hdr *ip6;
3288 INP_WLOCK_ASSERT(tp->t_inpcb);
3291 /* Don't bother if destination was broadcast/multicast. */
3292 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3295 if (mtod(m, struct ip *)->ip_v == 6) {
3296 ip6 = mtod(m, struct ip6_hdr *);
3297 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3298 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3300 /* IPv6 anycast check is done at tcp6_input() */
3303 #if defined(INET) && defined(INET6)
3308 ip = mtod(m, struct ip *);
3309 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3310 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3311 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3312 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3317 /* Perform bandwidth limiting. */
3318 if (badport_bandlim(rstreason) < 0)
3321 /* tcp_respond consumes the mbuf chain. */
3322 if (th->th_flags & TH_ACK) {
3323 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3324 th->th_ack, TH_RST);
3326 if (th->th_flags & TH_SYN)
3328 if (th->th_flags & TH_FIN)
3330 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3331 (tcp_seq)0, TH_RST|TH_ACK);
3339 * Parse TCP options and place in tcpopt.
3342 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3347 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3349 if (opt == TCPOPT_EOL)
3351 if (opt == TCPOPT_NOP)
3357 if (optlen < 2 || optlen > cnt)
3362 if (optlen != TCPOLEN_MAXSEG)
3364 if (!(flags & TO_SYN))
3366 to->to_flags |= TOF_MSS;
3367 bcopy((char *)cp + 2,
3368 (char *)&to->to_mss, sizeof(to->to_mss));
3369 to->to_mss = ntohs(to->to_mss);
3372 if (optlen != TCPOLEN_WINDOW)
3374 if (!(flags & TO_SYN))
3376 to->to_flags |= TOF_SCALE;
3377 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3379 case TCPOPT_TIMESTAMP:
3380 if (optlen != TCPOLEN_TIMESTAMP)
3382 to->to_flags |= TOF_TS;
3383 bcopy((char *)cp + 2,
3384 (char *)&to->to_tsval, sizeof(to->to_tsval));
3385 to->to_tsval = ntohl(to->to_tsval);
3386 bcopy((char *)cp + 6,
3387 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3388 to->to_tsecr = ntohl(to->to_tsecr);
3390 case TCPOPT_SIGNATURE:
3392 * In order to reply to a host which has set the
3393 * TCP_SIGNATURE option in its initial SYN, we have
3394 * to record the fact that the option was observed
3395 * here for the syncache code to perform the correct
3398 if (optlen != TCPOLEN_SIGNATURE)
3400 to->to_flags |= TOF_SIGNATURE;
3401 to->to_signature = cp + 2;
3403 case TCPOPT_SACK_PERMITTED:
3404 if (optlen != TCPOLEN_SACK_PERMITTED)
3406 if (!(flags & TO_SYN))
3410 to->to_flags |= TOF_SACKPERM;
3413 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3417 to->to_flags |= TOF_SACK;
3418 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3419 to->to_sacks = cp + 2;
3420 TCPSTAT_INC(tcps_sack_rcv_blocks);
3423 case TCPOPT_FAST_OPEN:
3424 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) &&
3425 (optlen < TCPOLEN_FAST_OPEN_MIN) &&
3426 (optlen > TCPOLEN_FAST_OPEN_MAX))
3428 if (!(flags & TO_SYN))
3430 if (!V_tcp_fastopen_enabled)
3432 to->to_flags |= TOF_FASTOPEN;
3433 to->to_tfo_len = optlen - 2;
3434 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3444 * Pull out of band byte out of a segment so
3445 * it doesn't appear in the user's data queue.
3446 * It is still reflected in the segment length for
3447 * sequencing purposes.
3450 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3453 int cnt = off + th->th_urp - 1;
3456 if (m->m_len > cnt) {
3457 char *cp = mtod(m, caddr_t) + cnt;
3458 struct tcpcb *tp = sototcpcb(so);
3460 INP_WLOCK_ASSERT(tp->t_inpcb);
3463 tp->t_oobflags |= TCPOOB_HAVEDATA;
3464 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3466 if (m->m_flags & M_PKTHDR)
3475 panic("tcp_pulloutofband");
3479 * Collect new round-trip time estimate
3480 * and update averages and current timeout.
3483 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3487 INP_WLOCK_ASSERT(tp->t_inpcb);
3489 TCPSTAT_INC(tcps_rttupdated);
3491 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3493 * srtt is stored as fixed point with 5 bits after the
3494 * binary point (i.e., scaled by 8). The following magic
3495 * is equivalent to the smoothing algorithm in rfc793 with
3496 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3497 * point). Adjust rtt to origin 0.
3499 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3500 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3502 if ((tp->t_srtt += delta) <= 0)
3506 * We accumulate a smoothed rtt variance (actually, a
3507 * smoothed mean difference), then set the retransmit
3508 * timer to smoothed rtt + 4 times the smoothed variance.
3509 * rttvar is stored as fixed point with 4 bits after the
3510 * binary point (scaled by 16). The following is
3511 * equivalent to rfc793 smoothing with an alpha of .75
3512 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3513 * rfc793's wired-in beta.
3517 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3518 if ((tp->t_rttvar += delta) <= 0)
3520 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3521 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3524 * No rtt measurement yet - use the unsmoothed rtt.
3525 * Set the variance to half the rtt (so our first
3526 * retransmit happens at 3*rtt).
3528 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3529 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3530 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3536 * the retransmit should happen at rtt + 4 * rttvar.
3537 * Because of the way we do the smoothing, srtt and rttvar
3538 * will each average +1/2 tick of bias. When we compute
3539 * the retransmit timer, we want 1/2 tick of rounding and
3540 * 1 extra tick because of +-1/2 tick uncertainty in the
3541 * firing of the timer. The bias will give us exactly the
3542 * 1.5 tick we need. But, because the bias is
3543 * statistical, we have to test that we don't drop below
3544 * the minimum feasible timer (which is 2 ticks).
3546 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3547 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3550 * We received an ack for a packet that wasn't retransmitted;
3551 * it is probably safe to discard any error indications we've
3552 * received recently. This isn't quite right, but close enough
3553 * for now (a route might have failed after we sent a segment,
3554 * and the return path might not be symmetrical).
3556 tp->t_softerror = 0;
3560 * Determine a reasonable value for maxseg size.
3561 * If the route is known, check route for mtu.
3562 * If none, use an mss that can be handled on the outgoing interface
3563 * without forcing IP to fragment. If no route is found, route has no mtu,
3564 * or the destination isn't local, use a default, hopefully conservative
3565 * size (usually 512 or the default IP max size, but no more than the mtu
3566 * of the interface), as we can't discover anything about intervening
3567 * gateways or networks. We also initialize the congestion/slow start
3568 * window to be a single segment if the destination isn't local.
3569 * While looking at the routing entry, we also initialize other path-dependent
3570 * parameters from pre-set or cached values in the routing entry.
3572 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3573 * IP options, e.g. IPSEC data, since length of this data may vary, and
3574 * thus it is calculated for every segment separately in tcp_output().
3576 * NOTE that this routine is only called when we process an incoming
3577 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3578 * settings are handled in tcp_mssopt().
3581 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3582 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3585 uint32_t maxmtu = 0;
3586 struct inpcb *inp = tp->t_inpcb;
3587 struct hc_metrics_lite metrics;
3589 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3590 size_t min_protoh = isipv6 ?
3591 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3592 sizeof (struct tcpiphdr);
3594 const size_t min_protoh = sizeof(struct tcpiphdr);
3597 INP_WLOCK_ASSERT(tp->t_inpcb);
3599 if (mtuoffer != -1) {
3600 KASSERT(offer == -1, ("%s: conflict", __func__));
3601 offer = mtuoffer - min_protoh;
3607 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3608 tp->t_maxseg = V_tcp_v6mssdflt;
3611 #if defined(INET) && defined(INET6)
3616 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3617 tp->t_maxseg = V_tcp_mssdflt;
3622 * No route to sender, stay with default mss and return.
3626 * In case we return early we need to initialize metrics
3627 * to a defined state as tcp_hc_get() would do for us
3628 * if there was no cache hit.
3630 if (metricptr != NULL)
3631 bzero(metricptr, sizeof(struct hc_metrics_lite));
3635 /* What have we got? */
3639 * Offer == 0 means that there was no MSS on the SYN
3640 * segment, in this case we use tcp_mssdflt as
3641 * already assigned to t_maxseg above.
3643 offer = tp->t_maxseg;
3648 * Offer == -1 means that we didn't receive SYN yet.
3654 * Prevent DoS attack with too small MSS. Round up
3655 * to at least minmss.
3657 offer = max(offer, V_tcp_minmss);
3661 * rmx information is now retrieved from tcp_hostcache.
3663 tcp_hc_get(&inp->inp_inc, &metrics);
3664 if (metricptr != NULL)
3665 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3668 * If there's a discovered mtu in tcp hostcache, use it.
3669 * Else, use the link mtu.
3671 if (metrics.rmx_mtu)
3672 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3676 mss = maxmtu - min_protoh;
3677 if (!V_path_mtu_discovery &&
3678 !in6_localaddr(&inp->in6p_faddr))
3679 mss = min(mss, V_tcp_v6mssdflt);
3682 #if defined(INET) && defined(INET6)
3687 mss = maxmtu - min_protoh;
3688 if (!V_path_mtu_discovery &&
3689 !in_localaddr(inp->inp_faddr))
3690 mss = min(mss, V_tcp_mssdflt);
3694 * XXX - The above conditional (mss = maxmtu - min_protoh)
3695 * probably violates the TCP spec.
3696 * The problem is that, since we don't know the
3697 * other end's MSS, we are supposed to use a conservative
3698 * default. But, if we do that, then MTU discovery will
3699 * never actually take place, because the conservative
3700 * default is much less than the MTUs typically seen
3701 * on the Internet today. For the moment, we'll sweep
3702 * this under the carpet.
3704 * The conservative default might not actually be a problem
3705 * if the only case this occurs is when sending an initial
3706 * SYN with options and data to a host we've never talked
3707 * to before. Then, they will reply with an MSS value which
3708 * will get recorded and the new parameters should get
3709 * recomputed. For Further Study.
3712 mss = min(mss, offer);
3715 * Sanity check: make sure that maxseg will be large
3716 * enough to allow some data on segments even if the
3717 * all the option space is used (40bytes). Otherwise
3718 * funny things may happen in tcp_output.
3720 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3728 tcp_mss(struct tcpcb *tp, int offer)
3734 struct hc_metrics_lite metrics;
3735 struct tcp_ifcap cap;
3737 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3739 bzero(&cap, sizeof(cap));
3740 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3746 * If there's a pipesize, change the socket buffer to that size,
3747 * don't change if sb_hiwat is different than default (then it
3748 * has been changed on purpose with setsockopt).
3749 * Make the socket buffers an integral number of mss units;
3750 * if the mss is larger than the socket buffer, decrease the mss.
3752 so = inp->inp_socket;
3753 SOCKBUF_LOCK(&so->so_snd);
3754 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3755 bufsize = metrics.rmx_sendpipe;
3757 bufsize = so->so_snd.sb_hiwat;
3761 bufsize = roundup(bufsize, mss);
3762 if (bufsize > sb_max)
3764 if (bufsize > so->so_snd.sb_hiwat)
3765 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3767 SOCKBUF_UNLOCK(&so->so_snd);
3769 * Sanity check: make sure that maxseg will be large
3770 * enough to allow some data on segments even if the
3771 * all the option space is used (40bytes). Otherwise
3772 * funny things may happen in tcp_output.
3774 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3776 tp->t_maxseg = max(mss, 64);
3778 SOCKBUF_LOCK(&so->so_rcv);
3779 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3780 bufsize = metrics.rmx_recvpipe;
3782 bufsize = so->so_rcv.sb_hiwat;
3783 if (bufsize > mss) {
3784 bufsize = roundup(bufsize, mss);
3785 if (bufsize > sb_max)
3787 if (bufsize > so->so_rcv.sb_hiwat)
3788 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3790 SOCKBUF_UNLOCK(&so->so_rcv);
3792 /* Check the interface for TSO capabilities. */
3793 if (cap.ifcap & CSUM_TSO) {
3794 tp->t_flags |= TF_TSO;
3795 tp->t_tsomax = cap.tsomax;
3796 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3797 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3802 * Determine the MSS option to send on an outgoing SYN.
3805 tcp_mssopt(struct in_conninfo *inc)
3808 uint32_t thcmtu = 0;
3809 uint32_t maxmtu = 0;
3812 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3815 if (inc->inc_flags & INC_ISIPV6) {
3816 mss = V_tcp_v6mssdflt;
3817 maxmtu = tcp_maxmtu6(inc, NULL);
3818 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3821 #if defined(INET) && defined(INET6)
3826 mss = V_tcp_mssdflt;
3827 maxmtu = tcp_maxmtu(inc, NULL);
3828 min_protoh = sizeof(struct tcpiphdr);
3831 #if defined(INET6) || defined(INET)
3832 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3835 if (maxmtu && thcmtu)
3836 mss = min(maxmtu, thcmtu) - min_protoh;
3837 else if (maxmtu || thcmtu)
3838 mss = max(maxmtu, thcmtu) - min_protoh;
3845 * On a partial ack arrives, force the retransmission of the
3846 * next unacknowledged segment. Do not clear tp->t_dupacks.
3847 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3851 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3853 tcp_seq onxt = tp->snd_nxt;
3854 uint32_t ocwnd = tp->snd_cwnd;
3855 u_int maxseg = tcp_maxseg(tp);
3857 INP_WLOCK_ASSERT(tp->t_inpcb);
3859 tcp_timer_activate(tp, TT_REXMT, 0);
3861 tp->snd_nxt = th->th_ack;
3863 * Set snd_cwnd to one segment beyond acknowledged offset.
3864 * (tp->snd_una has not yet been updated when this function is called.)
3866 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3867 tp->t_flags |= TF_ACKNOW;
3868 (void) tp->t_fb->tfb_tcp_output(tp);
3869 tp->snd_cwnd = ocwnd;
3870 if (SEQ_GT(onxt, tp->snd_nxt))
3873 * Partial window deflation. Relies on fact that tp->snd_una
3876 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3877 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3880 tp->snd_cwnd += maxseg;
3884 tcp_compute_pipe(struct tcpcb *tp)
3886 return (tp->snd_max - tp->snd_una +
3887 tp->sackhint.sack_bytes_rexmit -
3888 tp->sackhint.sacked_bytes);