2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
5 * The Regents of the University of California. All rights reserved.
6 * Copyright (c) 2007-2008,2010
7 * Swinburne University of Technology, Melbourne, Australia.
8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
9 * Copyright (c) 2010 The FreeBSD Foundation
10 * Copyright (c) 2010-2011 Juniper Networks, Inc.
11 * All rights reserved.
13 * Portions of this software were developed at the Centre for Advanced Internet
14 * Architectures, Swinburne University of Technology, by Lawrence Stewart,
15 * James Healy and David Hayes, made possible in part by a grant from the Cisco
16 * University Research Program Fund at Community Foundation Silicon Valley.
18 * Portions of this software were developed at the Centre for Advanced
19 * Internet Architectures, Swinburne University of Technology, Melbourne,
20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
22 * Portions of this software were developed by Robert N. M. Watson under
23 * contract to Juniper Networks, Inc.
25 * Redistribution and use in source and binary forms, with or without
26 * modification, are permitted provided that the following conditions
28 * 1. Redistributions of source code must retain the above copyright
29 * notice, this list of conditions and the following disclaimer.
30 * 2. Redistributions in binary form must reproduce the above copyright
31 * notice, this list of conditions and the following disclaimer in the
32 * documentation and/or other materials provided with the distribution.
33 * 3. Neither the name of the University nor the names of its contributors
34 * may be used to endorse or promote products derived from this software
35 * without specific prior written permission.
37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
49 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
52 #include <sys/cdefs.h>
53 __FBSDID("$FreeBSD$");
56 #include "opt_inet6.h"
57 #include "opt_ipsec.h"
58 #include "opt_tcpdebug.h"
60 #include <sys/param.h>
61 #include <sys/kernel.h>
63 #include <sys/hhook.h>
65 #include <sys/malloc.h>
67 #include <sys/proc.h> /* for proc0 declaration */
68 #include <sys/protosw.h>
70 #include <sys/signalvar.h>
71 #include <sys/socket.h>
72 #include <sys/socketvar.h>
73 #include <sys/sysctl.h>
74 #include <sys/syslog.h>
75 #include <sys/systm.h>
77 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
82 #include <net/if_var.h>
83 #include <net/route.h>
86 #define TCPSTATES /* for logging */
88 #include <netinet/in.h>
89 #include <netinet/in_kdtrace.h>
90 #include <netinet/in_pcb.h>
91 #include <netinet/in_systm.h>
92 #include <netinet/ip.h>
93 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
94 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
95 #include <netinet/ip_var.h>
96 #include <netinet/ip_options.h>
97 #include <netinet/ip6.h>
98 #include <netinet/icmp6.h>
99 #include <netinet6/in6_pcb.h>
100 #include <netinet6/in6_var.h>
101 #include <netinet6/ip6_var.h>
102 #include <netinet6/nd6.h>
103 #include <netinet/tcp.h>
104 #include <netinet/tcp_fsm.h>
105 #include <netinet/tcp_log_buf.h>
106 #include <netinet/tcp_seq.h>
107 #include <netinet/tcp_timer.h>
108 #include <netinet/tcp_var.h>
109 #include <netinet6/tcp6_var.h>
110 #include <netinet/tcpip.h>
111 #include <netinet/cc/cc.h>
112 #include <netinet/tcp_fastopen.h>
114 #include <netinet/tcp_pcap.h>
116 #include <netinet/tcp_syncache.h>
118 #include <netinet/tcp_debug.h>
119 #endif /* TCPDEBUG */
121 #include <netinet/tcp_offload.h>
124 #include <netipsec/ipsec_support.h>
126 #include <machine/in_cksum.h>
128 #include <security/mac/mac_framework.h>
130 const int tcprexmtthresh = 3;
132 VNET_DEFINE(int, tcp_log_in_vain) = 0;
133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
134 &VNET_NAME(tcp_log_in_vain), 0,
135 "Log all incoming TCP segments to closed ports");
137 VNET_DEFINE(int, blackhole) = 0;
138 #define V_blackhole VNET(blackhole)
139 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
140 &VNET_NAME(blackhole), 0,
141 "Do not send RST on segments to closed ports");
143 VNET_DEFINE(int, tcp_delack_enabled) = 1;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
145 &VNET_NAME(tcp_delack_enabled), 0,
146 "Delay ACK to try and piggyback it onto a data packet");
148 VNET_DEFINE(int, drop_synfin) = 0;
149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
150 &VNET_NAME(drop_synfin), 0,
151 "Drop TCP packets with SYN+FIN set");
153 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
155 &VNET_NAME(tcp_do_rfc6675_pipe), 0,
156 "Use calculated pipe/in-flight bytes per RFC 6675");
158 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
160 &VNET_NAME(tcp_do_rfc3042), 0,
161 "Enable RFC 3042 (Limited Transmit)");
163 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
165 &VNET_NAME(tcp_do_rfc3390), 0,
166 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
168 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
170 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
171 "Slow-start flight size (initial congestion window) in number of segments");
173 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
175 &VNET_NAME(tcp_do_rfc3465), 0,
176 "Enable RFC 3465 (Appropriate Byte Counting)");
178 VNET_DEFINE(int, tcp_abc_l_var) = 2;
179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
180 &VNET_NAME(tcp_abc_l_var), 2,
181 "Cap the max cwnd increment during slow-start to this number of segments");
183 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
185 VNET_DEFINE(int, tcp_do_ecn) = 2;
186 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
187 &VNET_NAME(tcp_do_ecn), 0,
190 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
191 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
192 &VNET_NAME(tcp_ecn_maxretries), 0,
193 "Max retries before giving up on ECN");
195 VNET_DEFINE(int, tcp_insecure_syn) = 0;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
197 &VNET_NAME(tcp_insecure_syn), 0,
198 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
200 VNET_DEFINE(int, tcp_insecure_rst) = 0;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
202 &VNET_NAME(tcp_insecure_rst), 0,
203 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
205 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
206 #define V_tcp_recvspace VNET(tcp_recvspace)
207 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
208 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
210 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
212 &VNET_NAME(tcp_do_autorcvbuf), 0,
213 "Enable automatic receive buffer sizing");
215 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW,
217 &VNET_NAME(tcp_autorcvbuf_inc), 0,
218 "Incrementor step size of automatic receive buffer");
220 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
222 &VNET_NAME(tcp_autorcvbuf_max), 0,
223 "Max size of automatic receive buffer");
225 VNET_DEFINE(struct inpcbhead, tcb);
226 #define tcb6 tcb /* for KAME src sync over BSD*'s */
227 VNET_DEFINE(struct inpcbinfo, tcbinfo);
230 * TCP statistics are stored in an array of counter(9)s, which size matches
231 * size of struct tcpstat. TCP running connection count is a regular array.
233 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
234 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
235 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
236 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
237 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
238 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
239 "TCP connection counts by TCP state");
242 tcp_vnet_init(const void *unused)
245 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
246 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
248 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
249 tcp_vnet_init, NULL);
253 tcp_vnet_uninit(const void *unused)
256 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
257 VNET_PCPUSTAT_FREE(tcpstat);
259 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
260 tcp_vnet_uninit, NULL);
264 * Kernel module interface for updating tcpstat. The argument is an index
265 * into tcpstat treated as an array.
268 kmod_tcpstat_inc(int statnum)
271 counter_u64_add(VNET(tcpstat)[statnum], 1);
276 * Wrapper for the TCP established input helper hook.
279 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
281 struct tcp_hhook_data hhook_data;
283 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
288 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
295 * CC wrapper hook functions
298 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
301 INP_WLOCK_ASSERT(tp->t_inpcb);
303 tp->ccv->nsegs = nsegs;
304 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
305 if (tp->snd_cwnd <= tp->snd_wnd)
306 tp->ccv->flags |= CCF_CWND_LIMITED;
308 tp->ccv->flags &= ~CCF_CWND_LIMITED;
310 if (type == CC_ACK) {
311 if (tp->snd_cwnd > tp->snd_ssthresh) {
312 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
313 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp));
314 if (tp->t_bytes_acked >= tp->snd_cwnd) {
315 tp->t_bytes_acked -= tp->snd_cwnd;
316 tp->ccv->flags |= CCF_ABC_SENTAWND;
319 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
320 tp->t_bytes_acked = 0;
324 if (CC_ALGO(tp)->ack_received != NULL) {
325 /* XXXLAS: Find a way to live without this */
326 tp->ccv->curack = th->th_ack;
327 CC_ALGO(tp)->ack_received(tp->ccv, type);
332 cc_conn_init(struct tcpcb *tp)
334 struct hc_metrics_lite metrics;
335 struct inpcb *inp = tp->t_inpcb;
339 INP_WLOCK_ASSERT(tp->t_inpcb);
341 tcp_hc_get(&inp->inp_inc, &metrics);
342 maxseg = tcp_maxseg(tp);
344 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
346 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
347 TCPSTAT_INC(tcps_usedrtt);
348 if (metrics.rmx_rttvar) {
349 tp->t_rttvar = metrics.rmx_rttvar;
350 TCPSTAT_INC(tcps_usedrttvar);
352 /* default variation is +- 1 rtt */
354 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
356 TCPT_RANGESET(tp->t_rxtcur,
357 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
358 tp->t_rttmin, TCPTV_REXMTMAX);
360 if (metrics.rmx_ssthresh) {
362 * There's some sort of gateway or interface
363 * buffer limit on the path. Use this to set
364 * the slow start threshold, but set the
365 * threshold to no less than 2*mss.
367 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
368 TCPSTAT_INC(tcps_usedssthresh);
372 * Set the initial slow-start flight size.
374 * RFC5681 Section 3.1 specifies the default conservative values.
375 * RFC3390 specifies slightly more aggressive values.
376 * RFC6928 increases it to ten segments.
377 * Support for user specified value for initial flight size.
379 * If a SYN or SYN/ACK was lost and retransmitted, we have to
380 * reduce the initial CWND to one segment as congestion is likely
381 * requiring us to be cautious.
383 if (tp->snd_cwnd == 1)
384 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
385 else if (V_tcp_initcwnd_segments)
386 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg,
387 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
388 else if (V_tcp_do_rfc3390)
389 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380));
391 /* Per RFC5681 Section 3.1 */
393 tp->snd_cwnd = 2 * maxseg;
394 else if (maxseg > 1095)
395 tp->snd_cwnd = 3 * maxseg;
397 tp->snd_cwnd = 4 * maxseg;
400 if (CC_ALGO(tp)->conn_init != NULL)
401 CC_ALGO(tp)->conn_init(tp->ccv);
405 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
409 INP_WLOCK_ASSERT(tp->t_inpcb);
413 if (!IN_FASTRECOVERY(tp->t_flags)) {
414 tp->snd_recover = tp->snd_max;
415 if (tp->t_flags & TF_ECN_PERMIT)
416 tp->t_flags |= TF_ECN_SND_CWR;
420 if (!IN_CONGRECOVERY(tp->t_flags) ||
422 * Allow ECN reaction on ACK to CWR, if
423 * that data segment was also CE marked.
425 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
426 EXIT_CONGRECOVERY(tp->t_flags);
427 TCPSTAT_INC(tcps_ecn_rcwnd);
428 tp->snd_recover = tp->snd_max + 1;
429 if (tp->t_flags & TF_ECN_PERMIT)
430 tp->t_flags |= TF_ECN_SND_CWR;
434 maxseg = tcp_maxseg(tp);
436 tp->t_bytes_acked = 0;
437 EXIT_RECOVERY(tp->t_flags);
438 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
440 tp->snd_cwnd = maxseg;
443 TCPSTAT_INC(tcps_sndrexmitbad);
444 /* RTO was unnecessary, so reset everything. */
445 tp->snd_cwnd = tp->snd_cwnd_prev;
446 tp->snd_ssthresh = tp->snd_ssthresh_prev;
447 tp->snd_recover = tp->snd_recover_prev;
448 if (tp->t_flags & TF_WASFRECOVERY)
449 ENTER_FASTRECOVERY(tp->t_flags);
450 if (tp->t_flags & TF_WASCRECOVERY)
451 ENTER_CONGRECOVERY(tp->t_flags);
452 tp->snd_nxt = tp->snd_max;
453 tp->t_flags &= ~TF_PREVVALID;
458 if (CC_ALGO(tp)->cong_signal != NULL) {
460 tp->ccv->curack = th->th_ack;
461 CC_ALGO(tp)->cong_signal(tp->ccv, type);
466 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
468 INP_WLOCK_ASSERT(tp->t_inpcb);
470 /* XXXLAS: KASSERT that we're in recovery? */
472 if (CC_ALGO(tp)->post_recovery != NULL) {
473 tp->ccv->curack = th->th_ack;
474 CC_ALGO(tp)->post_recovery(tp->ccv);
476 /* XXXLAS: EXIT_RECOVERY ? */
477 tp->t_bytes_acked = 0;
481 * Indicate whether this ack should be delayed. We can delay the ack if
482 * following conditions are met:
483 * - There is no delayed ack timer in progress.
484 * - Our last ack wasn't a 0-sized window. We never want to delay
485 * the ack that opens up a 0-sized window.
486 * - LRO wasn't used for this segment. We make sure by checking that the
487 * segment size is not larger than the MSS.
489 #define DELAY_ACK(tp, tlen) \
490 ((!tcp_timer_active(tp, TT_DELACK) && \
491 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
492 (tlen <= tp->t_maxseg) && \
493 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
496 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
498 INP_WLOCK_ASSERT(tp->t_inpcb);
500 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
501 switch (iptos & IPTOS_ECN_MASK) {
503 tp->ccv->flags |= CCF_IPHDR_CE;
506 tp->ccv->flags &= ~CCF_IPHDR_CE;
509 tp->ccv->flags &= ~CCF_IPHDR_CE;
513 if (th->th_flags & TH_CWR)
514 tp->ccv->flags |= CCF_TCPHDR_CWR;
516 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
518 if (tp->t_flags & TF_DELACK)
519 tp->ccv->flags |= CCF_DELACK;
521 tp->ccv->flags &= ~CCF_DELACK;
523 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
525 if (tp->ccv->flags & CCF_ACKNOW)
526 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
531 * TCP input handling is split into multiple parts:
532 * tcp6_input is a thin wrapper around tcp_input for the extended
533 * ip6_protox[] call format in ip6_input
534 * tcp_input handles primary segment validation, inpcb lookup and
535 * SYN processing on listen sockets
536 * tcp_do_segment processes the ACK and text of the segment for
537 * establishing, established and closing connections
541 tcp6_input(struct mbuf **mp, int *offp, int proto)
544 struct in6_ifaddr *ia6;
548 if (m->m_len < *offp + sizeof(struct tcphdr)) {
549 m = m_pullup(m, *offp + sizeof(struct tcphdr));
552 TCPSTAT_INC(tcps_rcvshort);
553 return (IPPROTO_DONE);
558 * draft-itojun-ipv6-tcp-to-anycast
559 * better place to put this in?
561 ip6 = mtod(m, struct ip6_hdr *);
562 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
563 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
565 ifa_free(&ia6->ia_ifa);
566 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
567 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
569 return (IPPROTO_DONE);
572 ifa_free(&ia6->ia_ifa);
575 return (tcp_input(mp, offp, proto));
580 tcp_input(struct mbuf **mp, int *offp, int proto)
582 struct mbuf *m = *mp;
583 struct tcphdr *th = NULL;
584 struct ip *ip = NULL;
585 struct inpcb *inp = NULL;
586 struct tcpcb *tp = NULL;
587 struct socket *so = NULL;
598 int rstreason = 0; /* For badport_bandlim accounting purposes */
600 struct m_tag *fwd_tag = NULL;
601 struct epoch_tracker et;
603 struct ip6_hdr *ip6 = NULL;
606 const void *ip6 = NULL;
608 struct tcpopt to; /* options in this segment */
609 char *s = NULL; /* address and port logging */
613 * The size of tcp_saveipgen must be the size of the max ip header,
616 u_char tcp_saveipgen[IP6_HDR_LEN];
617 struct tcphdr tcp_savetcp;
622 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
629 TCPSTAT_INC(tcps_rcvtotal);
634 ip6 = mtod(m, struct ip6_hdr *);
635 th = (struct tcphdr *)((caddr_t)ip6 + off0);
636 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
637 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
638 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
639 th->th_sum = m->m_pkthdr.csum_data;
641 th->th_sum = in6_cksum_pseudo(ip6, tlen,
642 IPPROTO_TCP, m->m_pkthdr.csum_data);
643 th->th_sum ^= 0xffff;
645 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
647 TCPSTAT_INC(tcps_rcvbadsum);
652 * Be proactive about unspecified IPv6 address in source.
653 * As we use all-zero to indicate unbounded/unconnected pcb,
654 * unspecified IPv6 address can be used to confuse us.
656 * Note that packets with unspecified IPv6 destination is
657 * already dropped in ip6_input.
659 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
663 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
666 #if defined(INET) && defined(INET6)
672 * Get IP and TCP header together in first mbuf.
673 * Note: IP leaves IP header in first mbuf.
675 if (off0 > sizeof (struct ip)) {
677 off0 = sizeof(struct ip);
679 if (m->m_len < sizeof (struct tcpiphdr)) {
680 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
682 TCPSTAT_INC(tcps_rcvshort);
683 return (IPPROTO_DONE);
686 ip = mtod(m, struct ip *);
687 th = (struct tcphdr *)((caddr_t)ip + off0);
688 tlen = ntohs(ip->ip_len) - off0;
691 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
692 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
693 th->th_sum = m->m_pkthdr.csum_data;
695 th->th_sum = in_pseudo(ip->ip_src.s_addr,
697 htonl(m->m_pkthdr.csum_data + tlen +
699 th->th_sum ^= 0xffff;
701 struct ipovly *ipov = (struct ipovly *)ip;
704 * Checksum extended TCP header and data.
708 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
709 ipov->ih_len = htons(tlen);
710 th->th_sum = in_cksum(m, len);
711 /* Reset length for SDT probes. */
712 ip->ip_len = htons(len);
715 /* Re-initialization for later version check */
717 ip->ip_v = IPVERSION;
718 ip->ip_hl = off0 >> 2;
722 TCPSTAT_INC(tcps_rcvbadsum);
729 * Check that TCP offset makes sense,
730 * pull out TCP options and adjust length. XXX
732 off = th->th_off << 2;
733 if (off < sizeof (struct tcphdr) || off > tlen) {
734 TCPSTAT_INC(tcps_rcvbadoff);
737 tlen -= off; /* tlen is used instead of ti->ti_len */
738 if (off > sizeof (struct tcphdr)) {
741 if (m->m_len < off0 + off) {
742 m = m_pullup(m, off0 + off);
744 TCPSTAT_INC(tcps_rcvshort);
745 return (IPPROTO_DONE);
748 ip6 = mtod(m, struct ip6_hdr *);
749 th = (struct tcphdr *)((caddr_t)ip6 + off0);
752 #if defined(INET) && defined(INET6)
757 if (m->m_len < sizeof(struct ip) + off) {
758 if ((m = m_pullup(m, sizeof (struct ip) + off))
760 TCPSTAT_INC(tcps_rcvshort);
761 return (IPPROTO_DONE);
763 ip = mtod(m, struct ip *);
764 th = (struct tcphdr *)((caddr_t)ip + off0);
768 optlen = off - sizeof (struct tcphdr);
769 optp = (u_char *)(th + 1);
771 thflags = th->th_flags;
774 * Convert TCP protocol specific fields to host format.
776 tcp_fields_to_host(th);
779 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
781 drop_hdrlen = off0 + off;
784 * Locate pcb for segment; if we're likely to add or remove a
785 * connection then first acquire pcbinfo lock. There are three cases
786 * where we might discover later we need a write lock despite the
787 * flags: ACKs moving a connection out of the syncache, ACKs for a
788 * connection in TIMEWAIT and SYNs not targeting a listening socket.
790 if ((thflags & (TH_FIN | TH_RST)) != 0) {
791 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
792 ti_locked = TI_RLOCKED;
794 ti_locked = TI_UNLOCKED;
797 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
801 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
803 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
806 #if defined(INET) && !defined(INET6)
807 (m->m_flags & M_IP_NEXTHOP)
810 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
814 if (ti_locked == TI_RLOCKED) {
815 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
817 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
821 if (isipv6 && fwd_tag != NULL) {
822 struct sockaddr_in6 *next_hop6;
824 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
826 * Transparently forwarded. Pretend to be the destination.
827 * Already got one like this?
829 inp = in6_pcblookup_mbuf(&V_tcbinfo,
830 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
831 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
834 * It's new. Try to find the ambushing socket.
835 * Because we've rewritten the destination address,
836 * any hardware-generated hash is ignored.
838 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
839 th->th_sport, &next_hop6->sin6_addr,
840 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
841 th->th_dport, INPLOOKUP_WILDCARD |
842 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
845 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
846 th->th_sport, &ip6->ip6_dst, th->th_dport,
847 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
848 m->m_pkthdr.rcvif, m);
851 #if defined(INET6) && defined(INET)
855 if (fwd_tag != NULL) {
856 struct sockaddr_in *next_hop;
858 next_hop = (struct sockaddr_in *)(fwd_tag+1);
860 * Transparently forwarded. Pretend to be the destination.
861 * already got one like this?
863 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
864 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
865 m->m_pkthdr.rcvif, m);
868 * It's new. Try to find the ambushing socket.
869 * Because we've rewritten the destination address,
870 * any hardware-generated hash is ignored.
872 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
873 th->th_sport, next_hop->sin_addr,
874 next_hop->sin_port ? ntohs(next_hop->sin_port) :
875 th->th_dport, INPLOOKUP_WILDCARD |
876 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
879 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
880 th->th_sport, ip->ip_dst, th->th_dport,
881 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
882 m->m_pkthdr.rcvif, m);
886 * If the INPCB does not exist then all data in the incoming
887 * segment is discarded and an appropriate RST is sent back.
888 * XXX MRT Send RST using which routing table?
892 * Log communication attempts to ports that are not
895 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
896 V_tcp_log_in_vain == 2) {
897 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
898 log(LOG_INFO, "%s; %s: Connection attempt "
899 "to closed port\n", s, __func__);
902 * When blackholing do not respond with a RST but
903 * completely ignore the segment and drop it.
905 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
909 rstreason = BANDLIM_RST_CLOSEDPORT;
912 INP_WLOCK_ASSERT(inp);
914 * While waiting for inp lock during the lookup, another thread
915 * can have dropped the inpcb, in which case we need to loop back
916 * and try to find a new inpcb to deliver to.
918 if (inp->inp_flags & INP_DROPPED) {
923 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
924 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
925 ((inp->inp_socket == NULL) ||
926 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
927 inp->inp_flowid = m->m_pkthdr.flowid;
928 inp->inp_flowtype = M_HASHTYPE_GET(m);
930 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
932 if (isipv6 && IPSEC_ENABLED(ipv6) &&
933 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
941 if (IPSEC_ENABLED(ipv4) &&
942 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
949 * Check the minimum TTL for socket.
951 if (inp->inp_ip_minttl != 0) {
954 if (inp->inp_ip_minttl > ip6->ip6_hlim)
958 if (inp->inp_ip_minttl > ip->ip_ttl)
963 * A previous connection in TIMEWAIT state is supposed to catch stray
964 * or duplicate segments arriving late. If this segment was a
965 * legitimate new connection attempt, the old INPCB gets removed and
966 * we can try again to find a listening socket.
968 * At this point, due to earlier optimism, we may hold only an inpcb
969 * lock, and not the inpcbinfo write lock. If so, we need to try to
970 * acquire it, or if that fails, acquire a reference on the inpcb,
971 * drop all locks, acquire a global write lock, and then re-acquire
972 * the inpcb lock. We may at that point discover that another thread
973 * has tried to free the inpcb, in which case we need to loop back
974 * and try to find a new inpcb to deliver to.
976 * XXXRW: It may be time to rethink timewait locking.
978 if (inp->inp_flags & INP_TIMEWAIT) {
979 if (ti_locked == TI_UNLOCKED) {
980 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
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_ET(&V_tcbinfo, et);
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 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1030 ti_locked = TI_RLOCKED;
1032 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1036 INP_WLOCK_ASSERT(inp);
1037 if (mac_inpcb_check_deliver(inp, m))
1040 so = inp->inp_socket;
1041 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1043 if (so->so_options & SO_DEBUG) {
1044 ostate = tp->t_state;
1047 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1050 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1053 #endif /* TCPDEBUG */
1055 * When the socket is accepting connections (the INPCB is in LISTEN
1056 * state) we look into the SYN cache if this is a new connection
1057 * attempt or the completion of a previous one.
1059 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1060 ("%s: so accepting but tp %p not listening", __func__, tp));
1061 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1062 struct in_conninfo inc;
1064 bzero(&inc, sizeof(inc));
1067 inc.inc_flags |= INC_ISIPV6;
1068 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1069 inc.inc_flags |= INC_IPV6MINMTU;
1070 inc.inc6_faddr = ip6->ip6_src;
1071 inc.inc6_laddr = ip6->ip6_dst;
1075 inc.inc_faddr = ip->ip_src;
1076 inc.inc_laddr = ip->ip_dst;
1078 inc.inc_fport = th->th_sport;
1079 inc.inc_lport = th->th_dport;
1080 inc.inc_fibnum = so->so_fibnum;
1083 * Check for an existing connection attempt in syncache if
1084 * the flag is only ACK. A successful lookup creates a new
1085 * socket appended to the listen queue in SYN_RECEIVED state.
1087 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1089 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1091 * Parse the TCP options here because
1092 * syncookies need access to the reflected
1095 tcp_dooptions(&to, optp, optlen, 0);
1097 * NB: syncache_expand() doesn't unlock
1098 * inp and tcpinfo locks.
1100 rstreason = syncache_expand(&inc, &to, th, &so, m);
1101 if (rstreason < 0) {
1103 * A failing TCP MD5 signature comparison
1104 * must result in the segment being dropped
1105 * and must not produce any response back
1109 } else if (rstreason == 0) {
1111 * No syncache entry or ACK was not
1112 * for our SYN/ACK. Send a RST.
1113 * NB: syncache did its own logging
1114 * of the failure cause.
1116 rstreason = BANDLIM_RST_OPENPORT;
1122 * We completed the 3-way handshake
1123 * but could not allocate a socket
1124 * either due to memory shortage,
1125 * listen queue length limits or
1126 * global socket limits. Send RST
1127 * or wait and have the remote end
1128 * retransmit the ACK for another
1131 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1132 log(LOG_DEBUG, "%s; %s: Listen socket: "
1133 "Socket allocation failed due to "
1134 "limits or memory shortage, %s\n",
1136 V_tcp_sc_rst_sock_fail ?
1137 "sending RST" : "try again");
1138 if (V_tcp_sc_rst_sock_fail) {
1139 rstreason = BANDLIM_UNLIMITED;
1145 * Socket is created in state SYN_RECEIVED.
1146 * Unlock the listen socket, lock the newly
1147 * created socket and update the tp variable.
1149 INP_WUNLOCK(inp); /* listen socket */
1150 inp = sotoinpcb(so);
1152 * New connection inpcb is already locked by
1153 * syncache_expand().
1155 INP_WLOCK_ASSERT(inp);
1156 tp = intotcpcb(inp);
1157 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1158 ("%s: ", __func__));
1160 * Process the segment and the data it
1161 * contains. tcp_do_segment() consumes
1162 * the mbuf chain and unlocks the inpcb.
1164 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1165 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1167 if (ti_locked == TI_RLOCKED)
1168 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1169 return (IPPROTO_DONE);
1172 * Segment flag validation for new connection attempts:
1174 * Our (SYN|ACK) response was rejected.
1175 * Check with syncache and remove entry to prevent
1178 * NB: syncache_chkrst does its own logging of failure
1181 if (thflags & TH_RST) {
1182 syncache_chkrst(&inc, th, m);
1186 * We can't do anything without SYN.
1188 if ((thflags & TH_SYN) == 0) {
1189 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1190 log(LOG_DEBUG, "%s; %s: Listen socket: "
1191 "SYN is missing, segment ignored\n",
1193 TCPSTAT_INC(tcps_badsyn);
1197 * (SYN|ACK) is bogus on a listen socket.
1199 if (thflags & TH_ACK) {
1200 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1201 log(LOG_DEBUG, "%s; %s: Listen socket: "
1202 "SYN|ACK invalid, segment rejected\n",
1204 syncache_badack(&inc); /* XXX: Not needed! */
1205 TCPSTAT_INC(tcps_badsyn);
1206 rstreason = BANDLIM_RST_OPENPORT;
1210 * If the drop_synfin option is enabled, drop all
1211 * segments with both the SYN and FIN bits set.
1212 * This prevents e.g. nmap from identifying the
1214 * XXX: Poor reasoning. nmap has other methods
1215 * and is constantly refining its stack detection
1217 * XXX: This is a violation of the TCP specification
1218 * and was used by RFC1644.
1220 if ((thflags & TH_FIN) && V_drop_synfin) {
1221 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1222 log(LOG_DEBUG, "%s; %s: Listen socket: "
1223 "SYN|FIN segment ignored (based on "
1224 "sysctl setting)\n", s, __func__);
1225 TCPSTAT_INC(tcps_badsyn);
1229 * Segment's flags are (SYN) or (SYN|FIN).
1231 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1232 * as they do not affect the state of the TCP FSM.
1233 * The data pointed to by TH_URG and th_urp is ignored.
1235 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1236 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1237 KASSERT(thflags & (TH_SYN),
1238 ("%s: Listen socket: TH_SYN not set", __func__));
1241 * If deprecated address is forbidden,
1242 * we do not accept SYN to deprecated interface
1243 * address to prevent any new inbound connection from
1244 * getting established.
1245 * When we do not accept SYN, we send a TCP RST,
1246 * with deprecated source address (instead of dropping
1247 * it). We compromise it as it is much better for peer
1248 * to send a RST, and RST will be the final packet
1251 * If we do not forbid deprecated addresses, we accept
1252 * the SYN packet. RFC2462 does not suggest dropping
1254 * If we decipher RFC2462 5.5.4, it says like this:
1255 * 1. use of deprecated addr with existing
1256 * communication is okay - "SHOULD continue to be
1258 * 2. use of it with new communication:
1259 * (2a) "SHOULD NOT be used if alternate address
1260 * with sufficient scope is available"
1261 * (2b) nothing mentioned otherwise.
1262 * Here we fall into (2b) case as we have no choice in
1263 * our source address selection - we must obey the peer.
1265 * The wording in RFC2462 is confusing, and there are
1266 * multiple description text for deprecated address
1267 * handling - worse, they are not exactly the same.
1268 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1270 if (isipv6 && !V_ip6_use_deprecated) {
1271 struct in6_ifaddr *ia6;
1273 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1275 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1276 ifa_free(&ia6->ia_ifa);
1277 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1278 log(LOG_DEBUG, "%s; %s: Listen socket: "
1279 "Connection attempt to deprecated "
1280 "IPv6 address rejected\n",
1282 rstreason = BANDLIM_RST_OPENPORT;
1286 ifa_free(&ia6->ia_ifa);
1290 * Basic sanity checks on incoming SYN requests:
1291 * Don't respond if the destination is a link layer
1292 * broadcast according to RFC1122 4.2.3.10, p. 104.
1293 * If it is from this socket it must be forged.
1294 * Don't respond if the source or destination is a
1295 * global or subnet broad- or multicast address.
1296 * Note that it is quite possible to receive unicast
1297 * link-layer packets with a broadcast IP address. Use
1298 * in_broadcast() to find them.
1300 if (m->m_flags & (M_BCAST|M_MCAST)) {
1301 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1302 log(LOG_DEBUG, "%s; %s: Listen socket: "
1303 "Connection attempt from broad- or multicast "
1304 "link layer address ignored\n", s, __func__);
1309 if (th->th_dport == th->th_sport &&
1310 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1311 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1312 log(LOG_DEBUG, "%s; %s: Listen socket: "
1313 "Connection attempt to/from self "
1314 "ignored\n", s, __func__);
1317 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1318 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1319 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1320 log(LOG_DEBUG, "%s; %s: Listen socket: "
1321 "Connection attempt from/to multicast "
1322 "address ignored\n", s, __func__);
1327 #if defined(INET) && defined(INET6)
1332 if (th->th_dport == th->th_sport &&
1333 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1334 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1335 log(LOG_DEBUG, "%s; %s: Listen socket: "
1336 "Connection attempt from/to self "
1337 "ignored\n", s, __func__);
1340 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1341 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1342 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1343 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1344 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1345 log(LOG_DEBUG, "%s; %s: Listen socket: "
1346 "Connection attempt from/to broad- "
1347 "or multicast address ignored\n",
1354 * SYN appears to be valid. Create compressed TCP state
1358 if (so->so_options & SO_DEBUG)
1359 tcp_trace(TA_INPUT, ostate, tp,
1360 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1362 TCP_PROBE3(debug__input, tp, th, m);
1363 tcp_dooptions(&to, optp, optlen, TO_SYN);
1364 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos))
1365 goto tfo_socket_result;
1368 * Entry added to syncache and mbuf consumed.
1369 * Only the listen socket is unlocked by syncache_add().
1371 if (ti_locked == TI_RLOCKED) {
1372 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1373 ti_locked = TI_UNLOCKED;
1375 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1376 return (IPPROTO_DONE);
1377 } else if (tp->t_state == TCPS_LISTEN) {
1379 * When a listen socket is torn down the SO_ACCEPTCONN
1380 * flag is removed first while connections are drained
1381 * from the accept queue in a unlock/lock cycle of the
1382 * ACCEPT_LOCK, opening a race condition allowing a SYN
1383 * attempt go through unhandled.
1387 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1388 if (tp->t_flags & TF_SIGNATURE) {
1389 tcp_dooptions(&to, optp, optlen, thflags);
1390 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1391 TCPSTAT_INC(tcps_sig_err_nosigopt);
1394 if (!TCPMD5_ENABLED() ||
1395 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1399 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1402 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1403 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1404 * the inpcb, and unlocks pcbinfo.
1406 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1407 if (ti_locked == TI_RLOCKED)
1408 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1409 return (IPPROTO_DONE);
1412 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1414 if (ti_locked == TI_RLOCKED) {
1415 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1416 ti_locked = TI_UNLOCKED;
1420 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1421 "ti_locked: %d", __func__, ti_locked));
1422 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1427 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1430 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1431 m = NULL; /* mbuf chain got consumed. */
1436 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1438 if (ti_locked == TI_RLOCKED) {
1439 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1440 ti_locked = TI_UNLOCKED;
1444 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1445 "ti_locked: %d", __func__, ti_locked));
1446 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1454 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1459 return (IPPROTO_DONE);
1463 * Automatic sizing of receive socket buffer. Often the send
1464 * buffer size is not optimally adjusted to the actual network
1465 * conditions at hand (delay bandwidth product). Setting the
1466 * buffer size too small limits throughput on links with high
1467 * bandwidth and high delay (eg. trans-continental/oceanic links).
1469 * On the receive side the socket buffer memory is only rarely
1470 * used to any significant extent. This allows us to be much
1471 * more aggressive in scaling the receive socket buffer. For
1472 * the case that the buffer space is actually used to a large
1473 * extent and we run out of kernel memory we can simply drop
1474 * the new segments; TCP on the sender will just retransmit it
1475 * later. Setting the buffer size too big may only consume too
1476 * much kernel memory if the application doesn't read() from
1477 * the socket or packet loss or reordering makes use of the
1480 * The criteria to step up the receive buffer one notch are:
1481 * 1. Application has not set receive buffer size with
1482 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1483 * 2. the number of bytes received during the time it takes
1484 * one timestamp to be reflected back to us (the RTT);
1485 * 3. received bytes per RTT is within seven eighth of the
1486 * current socket buffer size;
1487 * 4. receive buffer size has not hit maximal automatic size;
1489 * This algorithm does one step per RTT at most and only if
1490 * we receive a bulk stream w/o packet losses or reorderings.
1491 * Shrinking the buffer during idle times is not necessary as
1492 * it doesn't consume any memory when idle.
1494 * TODO: Only step up if the application is actually serving
1495 * the buffer to better manage the socket buffer resources.
1498 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1499 struct tcpcb *tp, int tlen)
1503 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1504 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1505 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1506 (tp->t_srtt >> TCP_RTT_SHIFT)) {
1507 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) &&
1508 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1509 newsize = min(so->so_rcv.sb_hiwat +
1510 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max);
1512 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1514 /* Start over with next RTT. */
1518 tp->rfbuf_cnt += tlen; /* add up */
1524 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1525 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1527 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1528 int rstreason, todrop, win, incforsyn = 0;
1532 struct in_conninfo *inc;
1539 * The size of tcp_saveipgen must be the size of the max ip header,
1542 u_char tcp_saveipgen[IP6_HDR_LEN];
1543 struct tcphdr tcp_savetcp;
1546 thflags = th->th_flags;
1547 inc = &tp->t_inpcb->inp_inc;
1548 tp->sackhint.last_sack_ack = 0;
1550 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1552 * If this is either a state-changing packet or current state isn't
1553 * established, we require a write lock on tcbinfo. Otherwise, we
1554 * allow the tcbinfo to be in either alocked or unlocked, as the
1555 * caller may have unnecessarily acquired a write lock due to a race.
1557 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1558 tp->t_state != TCPS_ESTABLISHED) {
1559 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1561 INP_WLOCK_ASSERT(tp->t_inpcb);
1562 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1564 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1568 /* Save segment, if requested. */
1569 tcp_pcap_add(th, m, &(tp->t_inpkts));
1571 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1574 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1575 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1576 log(LOG_DEBUG, "%s; %s: "
1577 "SYN|FIN segment ignored (based on "
1578 "sysctl setting)\n", s, __func__);
1585 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1586 * check SEQ.ACK first.
1588 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1589 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1590 rstreason = BANDLIM_UNLIMITED;
1595 * Segment received on connection.
1596 * Reset idle time and keep-alive timer.
1597 * XXX: This should be done after segment
1598 * validation to ignore broken/spoofed segs.
1600 tp->t_rcvtime = ticks;
1603 * Scale up the window into a 32-bit value.
1604 * For the SYN_SENT state the scale is zero.
1606 tiwin = th->th_win << tp->snd_scale;
1609 * TCP ECN processing.
1611 if (tp->t_flags & TF_ECN_PERMIT) {
1612 if (thflags & TH_CWR)
1613 tp->t_flags &= ~TF_ECN_SND_ECE;
1614 switch (iptos & IPTOS_ECN_MASK) {
1616 tp->t_flags |= TF_ECN_SND_ECE;
1617 TCPSTAT_INC(tcps_ecn_ce);
1619 case IPTOS_ECN_ECT0:
1620 TCPSTAT_INC(tcps_ecn_ect0);
1622 case IPTOS_ECN_ECT1:
1623 TCPSTAT_INC(tcps_ecn_ect1);
1627 /* Process a packet differently from RFC3168. */
1628 cc_ecnpkt_handler(tp, th, iptos);
1630 /* Congestion experienced. */
1631 if (thflags & TH_ECE) {
1632 cc_cong_signal(tp, th, CC_ECN);
1637 * Parse options on any incoming segment.
1639 tcp_dooptions(&to, (u_char *)(th + 1),
1640 (th->th_off << 2) - sizeof(struct tcphdr),
1641 (thflags & TH_SYN) ? TO_SYN : 0);
1643 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1644 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1645 (to.to_flags & TOF_SIGNATURE) == 0) {
1646 TCPSTAT_INC(tcps_sig_err_sigopt);
1647 /* XXX: should drop? */
1651 * If echoed timestamp is later than the current time,
1652 * fall back to non RFC1323 RTT calculation. Normalize
1653 * timestamp if syncookies were used when this connection
1656 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1657 to.to_tsecr -= tp->ts_offset;
1658 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1660 else if (tp->t_flags & TF_PREVVALID &&
1661 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1662 cc_cong_signal(tp, th, CC_RTO_ERR);
1665 * Process options only when we get SYN/ACK back. The SYN case
1666 * for incoming connections is handled in tcp_syncache.
1667 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1668 * or <SYN,ACK>) segment itself is never scaled.
1669 * XXX this is traditional behavior, may need to be cleaned up.
1671 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1672 if ((to.to_flags & TOF_SCALE) &&
1673 (tp->t_flags & TF_REQ_SCALE)) {
1674 tp->t_flags |= TF_RCVD_SCALE;
1675 tp->snd_scale = to.to_wscale;
1677 tp->t_flags &= ~TF_REQ_SCALE;
1679 * Initial send window. It will be updated with
1680 * the next incoming segment to the scaled value.
1682 tp->snd_wnd = th->th_win;
1683 if ((to.to_flags & TOF_TS) &&
1684 (tp->t_flags & TF_REQ_TSTMP)) {
1685 tp->t_flags |= TF_RCVD_TSTMP;
1686 tp->ts_recent = to.to_tsval;
1687 tp->ts_recent_age = tcp_ts_getticks();
1689 tp->t_flags &= ~TF_REQ_TSTMP;
1690 if (to.to_flags & TOF_MSS)
1691 tcp_mss(tp, to.to_mss);
1692 if ((tp->t_flags & TF_SACK_PERMIT) &&
1693 (to.to_flags & TOF_SACKPERM) == 0)
1694 tp->t_flags &= ~TF_SACK_PERMIT;
1695 if (IS_FASTOPEN(tp->t_flags)) {
1696 if (to.to_flags & TOF_FASTOPEN) {
1699 if (to.to_flags & TOF_MSS)
1702 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1706 tcp_fastopen_update_cache(tp, mss,
1707 to.to_tfo_len, to.to_tfo_cookie);
1709 tcp_fastopen_disable_path(tp);
1714 * If timestamps were negotiated during SYN/ACK they should
1715 * appear on every segment during this session and vice versa.
1717 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1718 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1719 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1720 "no action\n", s, __func__);
1724 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1725 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1726 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1727 "no action\n", s, __func__);
1733 * Header prediction: check for the two common cases
1734 * of a uni-directional data xfer. If the packet has
1735 * no control flags, is in-sequence, the window didn't
1736 * change and we're not retransmitting, it's a
1737 * candidate. If the length is zero and the ack moved
1738 * forward, we're the sender side of the xfer. Just
1739 * free the data acked & wake any higher level process
1740 * that was blocked waiting for space. If the length
1741 * is non-zero and the ack didn't move, we're the
1742 * receiver side. If we're getting packets in-order
1743 * (the reassembly queue is empty), add the data to
1744 * the socket buffer and note that we need a delayed ack.
1745 * Make sure that the hidden state-flags are also off.
1746 * Since we check for TCPS_ESTABLISHED first, it can only
1749 if (tp->t_state == TCPS_ESTABLISHED &&
1750 th->th_seq == tp->rcv_nxt &&
1751 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1752 tp->snd_nxt == tp->snd_max &&
1753 tiwin && tiwin == tp->snd_wnd &&
1754 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1756 ((to.to_flags & TOF_TS) == 0 ||
1757 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1760 * If last ACK falls within this segment's sequence numbers,
1761 * record the timestamp.
1762 * NOTE that the test is modified according to the latest
1763 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1765 if ((to.to_flags & TOF_TS) != 0 &&
1766 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1767 tp->ts_recent_age = tcp_ts_getticks();
1768 tp->ts_recent = to.to_tsval;
1772 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1773 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1774 !IN_RECOVERY(tp->t_flags) &&
1775 (to.to_flags & TOF_SACK) == 0 &&
1776 TAILQ_EMPTY(&tp->snd_holes)) {
1778 * This is a pure ack for outstanding data.
1780 TCPSTAT_INC(tcps_predack);
1783 * "bad retransmit" recovery without timestamps.
1785 if ((to.to_flags & TOF_TS) == 0 &&
1786 tp->t_rxtshift == 1 &&
1787 tp->t_flags & TF_PREVVALID &&
1788 (int)(ticks - tp->t_badrxtwin) < 0) {
1789 cc_cong_signal(tp, th, CC_RTO_ERR);
1793 * Recalculate the transmit timer / rtt.
1795 * Some boxes send broken timestamp replies
1796 * during the SYN+ACK phase, ignore
1797 * timestamps of 0 or we could calculate a
1798 * huge RTT and blow up the retransmit timer.
1800 if ((to.to_flags & TOF_TS) != 0 &&
1804 t = tcp_ts_getticks() - to.to_tsecr;
1805 if (!tp->t_rttlow || tp->t_rttlow > t)
1808 TCP_TS_TO_TICKS(t) + 1);
1809 } else if (tp->t_rtttime &&
1810 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1811 if (!tp->t_rttlow ||
1812 tp->t_rttlow > ticks - tp->t_rtttime)
1813 tp->t_rttlow = ticks - tp->t_rtttime;
1815 ticks - tp->t_rtttime);
1817 acked = BYTES_THIS_ACK(tp, th);
1820 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1821 hhook_run_tcp_est_in(tp, th, &to);
1824 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1825 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1826 sbdrop(&so->so_snd, acked);
1827 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1828 SEQ_LEQ(th->th_ack, tp->snd_recover))
1829 tp->snd_recover = th->th_ack - 1;
1832 * Let the congestion control algorithm update
1833 * congestion control related information. This
1834 * typically means increasing the congestion
1837 cc_ack_received(tp, th, nsegs, CC_ACK);
1839 tp->snd_una = th->th_ack;
1841 * Pull snd_wl2 up to prevent seq wrap relative
1844 tp->snd_wl2 = th->th_ack;
1849 * If all outstanding data are acked, stop
1850 * retransmit timer, otherwise restart timer
1851 * using current (possibly backed-off) value.
1852 * If process is waiting for space,
1853 * wakeup/selwakeup/signal. If data
1854 * are ready to send, let tcp_output
1855 * decide between more output or persist.
1858 if (so->so_options & SO_DEBUG)
1859 tcp_trace(TA_INPUT, ostate, tp,
1860 (void *)tcp_saveipgen,
1863 TCP_PROBE3(debug__input, tp, th, m);
1864 if (tp->snd_una == tp->snd_max)
1865 tcp_timer_activate(tp, TT_REXMT, 0);
1866 else if (!tcp_timer_active(tp, TT_PERSIST))
1867 tcp_timer_activate(tp, TT_REXMT,
1870 if (sbavail(&so->so_snd))
1871 (void) tp->t_fb->tfb_tcp_output(tp);
1874 } else if (th->th_ack == tp->snd_una &&
1875 tlen <= sbspace(&so->so_rcv)) {
1876 int newsize = 0; /* automatic sockbuf scaling */
1879 * This is a pure, in-sequence data packet with
1880 * nothing on the reassembly queue and we have enough
1881 * buffer space to take it.
1883 /* Clean receiver SACK report if present */
1884 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1885 tcp_clean_sackreport(tp);
1886 TCPSTAT_INC(tcps_preddat);
1887 tp->rcv_nxt += tlen;
1889 * Pull snd_wl1 up to prevent seq wrap relative to
1892 tp->snd_wl1 = th->th_seq;
1894 * Pull rcv_up up to prevent seq wrap relative to
1897 tp->rcv_up = tp->rcv_nxt;
1898 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1899 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1901 if (so->so_options & SO_DEBUG)
1902 tcp_trace(TA_INPUT, ostate, tp,
1903 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1905 TCP_PROBE3(debug__input, tp, th, m);
1907 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1909 /* Add data to socket buffer. */
1910 SOCKBUF_LOCK(&so->so_rcv);
1911 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1915 * Set new socket buffer size.
1916 * Give up when limit is reached.
1919 if (!sbreserve_locked(&so->so_rcv,
1921 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1922 m_adj(m, drop_hdrlen); /* delayed header drop */
1923 sbappendstream_locked(&so->so_rcv, m, 0);
1925 /* NB: sorwakeup_locked() does an implicit unlock. */
1926 sorwakeup_locked(so);
1927 if (DELAY_ACK(tp, tlen)) {
1928 tp->t_flags |= TF_DELACK;
1930 tp->t_flags |= TF_ACKNOW;
1931 tp->t_fb->tfb_tcp_output(tp);
1938 * Calculate amount of space in receive window,
1939 * and then do TCP input processing.
1940 * Receive window is amount of space in rcv queue,
1941 * but not less than advertised window.
1943 win = sbspace(&so->so_rcv);
1946 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1948 switch (tp->t_state) {
1951 * If the state is SYN_RECEIVED:
1952 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1954 case TCPS_SYN_RECEIVED:
1955 if ((thflags & TH_ACK) &&
1956 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1957 SEQ_GT(th->th_ack, tp->snd_max))) {
1958 rstreason = BANDLIM_RST_OPENPORT;
1961 if (IS_FASTOPEN(tp->t_flags)) {
1963 * When a TFO connection is in SYN_RECEIVED, the
1964 * only valid packets are the initial SYN, a
1965 * retransmit/copy of the initial SYN (possibly with
1966 * a subset of the original data), a valid ACK, a
1969 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1970 rstreason = BANDLIM_RST_OPENPORT;
1972 } else if (thflags & TH_SYN) {
1973 /* non-initial SYN is ignored */
1974 if ((tcp_timer_active(tp, TT_DELACK) ||
1975 tcp_timer_active(tp, TT_REXMT)))
1977 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1984 * If the state is SYN_SENT:
1985 * if seg contains a RST with valid ACK (SEQ.ACK has already
1986 * been verified), then drop the connection.
1987 * if seg contains a RST without an ACK, drop the seg.
1988 * if seg does not contain SYN, then drop the seg.
1989 * Otherwise this is an acceptable SYN segment
1990 * initialize tp->rcv_nxt and tp->irs
1991 * if seg contains ack then advance tp->snd_una
1992 * if seg contains an ECE and ECN support is enabled, the stream
1994 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1995 * arrange for segment to be acked (eventually)
1996 * continue processing rest of data/controls, beginning with URG
1999 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2000 TCP_PROBE5(connect__refused, NULL, tp,
2002 tp = tcp_drop(tp, ECONNREFUSED);
2004 if (thflags & TH_RST)
2006 if (!(thflags & TH_SYN))
2009 tp->irs = th->th_seq;
2011 if (thflags & TH_ACK) {
2012 int tfo_partial_ack = 0;
2014 TCPSTAT_INC(tcps_connects);
2017 mac_socketpeer_set_from_mbuf(m, so);
2019 /* Do window scaling on this connection? */
2020 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2021 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2022 tp->rcv_scale = tp->request_r_scale;
2024 tp->rcv_adv += min(tp->rcv_wnd,
2025 TCP_MAXWIN << tp->rcv_scale);
2026 tp->snd_una++; /* SYN is acked */
2028 * If not all the data that was sent in the TFO SYN
2029 * has been acked, resend the remainder right away.
2031 if (IS_FASTOPEN(tp->t_flags) &&
2032 (tp->snd_una != tp->snd_max)) {
2033 tp->snd_nxt = th->th_ack;
2034 tfo_partial_ack = 1;
2037 * If there's data, delay ACK; if there's also a FIN
2038 * ACKNOW will be turned on later.
2040 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2041 tcp_timer_activate(tp, TT_DELACK,
2044 tp->t_flags |= TF_ACKNOW;
2046 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2048 tp->t_flags |= TF_ECN_PERMIT;
2049 TCPSTAT_INC(tcps_ecn_shs);
2053 * Received <SYN,ACK> in SYN_SENT[*] state.
2055 * SYN_SENT --> ESTABLISHED
2056 * SYN_SENT* --> FIN_WAIT_1
2058 tp->t_starttime = ticks;
2059 if (tp->t_flags & TF_NEEDFIN) {
2060 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2061 tp->t_flags &= ~TF_NEEDFIN;
2064 tcp_state_change(tp, TCPS_ESTABLISHED);
2065 TCP_PROBE5(connect__established, NULL, tp,
2068 tcp_timer_activate(tp, TT_KEEP,
2073 * Received initial SYN in SYN-SENT[*] state =>
2074 * simultaneous open.
2075 * If it succeeds, connection is * half-synchronized.
2076 * Otherwise, do 3-way handshake:
2077 * SYN-SENT -> SYN-RECEIVED
2078 * SYN-SENT* -> SYN-RECEIVED*
2080 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2081 tcp_timer_activate(tp, TT_REXMT, 0);
2082 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2085 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2086 INP_WLOCK_ASSERT(tp->t_inpcb);
2089 * Advance th->th_seq to correspond to first data byte.
2090 * If data, trim to stay within window,
2091 * dropping FIN if necessary.
2094 if (tlen > tp->rcv_wnd) {
2095 todrop = tlen - tp->rcv_wnd;
2099 TCPSTAT_INC(tcps_rcvpackafterwin);
2100 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2102 tp->snd_wl1 = th->th_seq - 1;
2103 tp->rcv_up = th->th_seq;
2105 * Client side of transaction: already sent SYN and data.
2106 * If the remote host used T/TCP to validate the SYN,
2107 * our data will be ACK'd; if so, enter normal data segment
2108 * processing in the middle of step 5, ack processing.
2109 * Otherwise, goto step 6.
2111 if (thflags & TH_ACK)
2117 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2118 * do normal processing.
2120 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2124 break; /* continue normal processing */
2128 * States other than LISTEN or SYN_SENT.
2129 * First check the RST flag and sequence number since reset segments
2130 * are exempt from the timestamp and connection count tests. This
2131 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2132 * below which allowed reset segments in half the sequence space
2133 * to fall though and be processed (which gives forged reset
2134 * segments with a random sequence number a 50 percent chance of
2135 * killing a connection).
2136 * Then check timestamp, if present.
2137 * Then check the connection count, if present.
2138 * Then check that at least some bytes of segment are within
2139 * receive window. If segment begins before rcv_nxt,
2140 * drop leading data (and SYN); if nothing left, just ack.
2142 if (thflags & TH_RST) {
2144 * RFC5961 Section 3.2
2146 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2147 * - If RST is in window, we send challenge ACK.
2149 * Note: to take into account delayed ACKs, we should
2150 * test against last_ack_sent instead of rcv_nxt.
2151 * Note 2: we handle special case of closed window, not
2152 * covered by the RFC.
2154 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2155 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2156 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2158 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2159 KASSERT(tp->t_state != TCPS_SYN_SENT,
2160 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2163 if (V_tcp_insecure_rst ||
2164 tp->last_ack_sent == th->th_seq) {
2165 TCPSTAT_INC(tcps_drops);
2166 /* Drop the connection. */
2167 switch (tp->t_state) {
2168 case TCPS_SYN_RECEIVED:
2169 so->so_error = ECONNREFUSED;
2171 case TCPS_ESTABLISHED:
2172 case TCPS_FIN_WAIT_1:
2173 case TCPS_FIN_WAIT_2:
2174 case TCPS_CLOSE_WAIT:
2177 so->so_error = ECONNRESET;
2184 TCPSTAT_INC(tcps_badrst);
2185 /* Send challenge ACK. */
2186 tcp_respond(tp, mtod(m, void *), th, m,
2187 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2188 tp->last_ack_sent = tp->rcv_nxt;
2196 * RFC5961 Section 4.2
2197 * Send challenge ACK for any SYN in synchronized state.
2199 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2200 tp->t_state != TCPS_SYN_RECEIVED) {
2201 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2203 TCPSTAT_INC(tcps_badsyn);
2204 if (V_tcp_insecure_syn &&
2205 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2206 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2207 tp = tcp_drop(tp, ECONNRESET);
2208 rstreason = BANDLIM_UNLIMITED;
2210 /* Send challenge ACK. */
2211 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2212 tp->snd_nxt, TH_ACK);
2213 tp->last_ack_sent = tp->rcv_nxt;
2220 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2221 * and it's less than ts_recent, drop it.
2223 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2224 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2226 /* Check to see if ts_recent is over 24 days old. */
2227 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2229 * Invalidate ts_recent. If this segment updates
2230 * ts_recent, the age will be reset later and ts_recent
2231 * will get a valid value. If it does not, setting
2232 * ts_recent to zero will at least satisfy the
2233 * requirement that zero be placed in the timestamp
2234 * echo reply when ts_recent isn't valid. The
2235 * age isn't reset until we get a valid ts_recent
2236 * because we don't want out-of-order segments to be
2237 * dropped when ts_recent is old.
2241 TCPSTAT_INC(tcps_rcvduppack);
2242 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2243 TCPSTAT_INC(tcps_pawsdrop);
2251 * In the SYN-RECEIVED state, validate that the packet belongs to
2252 * this connection before trimming the data to fit the receive
2253 * window. Check the sequence number versus IRS since we know
2254 * the sequence numbers haven't wrapped. This is a partial fix
2255 * for the "LAND" DoS attack.
2257 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2258 rstreason = BANDLIM_RST_OPENPORT;
2262 todrop = tp->rcv_nxt - th->th_seq;
2264 if (thflags & TH_SYN) {
2274 * Following if statement from Stevens, vol. 2, p. 960.
2277 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2279 * Any valid FIN must be to the left of the window.
2280 * At this point the FIN must be a duplicate or out
2281 * of sequence; drop it.
2286 * Send an ACK to resynchronize and drop any data.
2287 * But keep on processing for RST or ACK.
2289 tp->t_flags |= TF_ACKNOW;
2291 TCPSTAT_INC(tcps_rcvduppack);
2292 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2294 TCPSTAT_INC(tcps_rcvpartduppack);
2295 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2298 * DSACK - add SACK block for dropped range
2300 if (tp->t_flags & TF_SACK_PERMIT) {
2301 tcp_update_sack_list(tp, th->th_seq,
2302 th->th_seq + todrop);
2304 * ACK now, as the next in-sequence segment
2305 * will clear the DSACK block again
2307 tp->t_flags |= TF_ACKNOW;
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 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2328 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2329 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2330 "after socket was closed, "
2331 "sending RST and removing tcpcb\n",
2332 s, __func__, tcpstates[tp->t_state], tlen);
2336 TCPSTAT_INC(tcps_rcvafterclose);
2337 rstreason = BANDLIM_UNLIMITED;
2342 * If segment ends after window, drop trailing data
2343 * (and PUSH and FIN); if nothing left, just ACK.
2345 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2347 TCPSTAT_INC(tcps_rcvpackafterwin);
2348 if (todrop >= tlen) {
2349 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2351 * If window is closed can only take segments at
2352 * window edge, and have to drop data and PUSH from
2353 * incoming segments. Continue processing, but
2354 * remember to ack. Otherwise, drop segment
2357 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2358 tp->t_flags |= TF_ACKNOW;
2359 TCPSTAT_INC(tcps_rcvwinprobe);
2363 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2366 thflags &= ~(TH_PUSH|TH_FIN);
2370 * If last ACK falls within this segment's sequence numbers,
2371 * record its timestamp.
2373 * 1) That the test incorporates suggestions from the latest
2374 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2375 * 2) That updating only on newer timestamps interferes with
2376 * our earlier PAWS tests, so this check should be solely
2377 * predicated on the sequence space of this segment.
2378 * 3) That we modify the segment boundary check to be
2379 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2380 * instead of RFC1323's
2381 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2382 * This modified check allows us to overcome RFC1323's
2383 * limitations as described in Stevens TCP/IP Illustrated
2384 * Vol. 2 p.869. In such cases, we can still calculate the
2385 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2387 if ((to.to_flags & TOF_TS) != 0 &&
2388 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2389 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2390 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2391 tp->ts_recent_age = tcp_ts_getticks();
2392 tp->ts_recent = to.to_tsval;
2396 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2397 * flag is on (half-synchronized state), then queue data for
2398 * later processing; else drop segment and return.
2400 if ((thflags & TH_ACK) == 0) {
2401 if (tp->t_state == TCPS_SYN_RECEIVED ||
2402 (tp->t_flags & TF_NEEDSYN)) {
2403 if (tp->t_state == TCPS_SYN_RECEIVED &&
2404 IS_FASTOPEN(tp->t_flags)) {
2405 tp->snd_wnd = tiwin;
2409 } else if (tp->t_flags & TF_ACKNOW)
2418 switch (tp->t_state) {
2421 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2422 * ESTABLISHED state and continue processing.
2423 * The ACK was checked above.
2425 case TCPS_SYN_RECEIVED:
2427 TCPSTAT_INC(tcps_connects);
2429 /* Do window scaling? */
2430 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2431 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2432 tp->rcv_scale = tp->request_r_scale;
2434 tp->snd_wnd = tiwin;
2437 * SYN-RECEIVED -> ESTABLISHED
2438 * SYN-RECEIVED* -> FIN-WAIT-1
2440 tp->t_starttime = ticks;
2441 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2442 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2443 tp->t_tfo_pending = NULL;
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 * TFO connections call cc_conn_init() during SYN
2454 * processing. Calling it again here for such
2455 * connections is not harmless as it would undo the
2456 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2459 if (!IS_FASTOPEN(tp->t_flags))
2461 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2464 * Account for the ACK of our SYN prior to
2465 * regular ACK processing below, except for
2466 * simultaneous SYN, which is handled later.
2468 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2471 * If segment contains data or ACK, will call tcp_reass()
2472 * later; if not, do so now to pass queued data to user.
2474 if (tlen == 0 && (thflags & TH_FIN) == 0)
2475 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2477 tp->snd_wl1 = th->th_seq - 1;
2481 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2482 * ACKs. If the ack is in the range
2483 * tp->snd_una < th->th_ack <= tp->snd_max
2484 * then advance tp->snd_una to th->th_ack and drop
2485 * data from the retransmission queue. If this ACK reflects
2486 * more up to date window information we update our window information.
2488 case TCPS_ESTABLISHED:
2489 case TCPS_FIN_WAIT_1:
2490 case TCPS_FIN_WAIT_2:
2491 case TCPS_CLOSE_WAIT:
2494 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2495 TCPSTAT_INC(tcps_rcvacktoomuch);
2498 if ((tp->t_flags & TF_SACK_PERMIT) &&
2499 ((to.to_flags & TOF_SACK) ||
2500 !TAILQ_EMPTY(&tp->snd_holes)))
2501 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2504 * Reset the value so that previous (valid) value
2505 * from the last ack with SACK doesn't get used.
2507 tp->sackhint.sacked_bytes = 0;
2510 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2511 hhook_run_tcp_est_in(tp, th, &to);
2514 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2517 maxseg = tcp_maxseg(tp);
2519 (tiwin == tp->snd_wnd ||
2520 (tp->t_flags & TF_SACK_PERMIT))) {
2522 * If this is the first time we've seen a
2523 * FIN from the remote, this is not a
2524 * duplicate and it needs to be processed
2525 * normally. This happens during a
2526 * simultaneous close.
2528 if ((thflags & TH_FIN) &&
2529 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2533 TCPSTAT_INC(tcps_rcvdupack);
2535 * If we have outstanding data (other than
2536 * a window probe), this is a completely
2537 * duplicate ack (ie, window info didn't
2538 * change and FIN isn't set),
2539 * the ack is the biggest we've
2540 * seen and we've seen exactly our rexmt
2541 * threshold of them, assume a packet
2542 * has been dropped and retransmit it.
2543 * Kludge snd_nxt & the congestion
2544 * window so we send only this one
2547 * We know we're losing at the current
2548 * window size so do congestion avoidance
2549 * (set ssthresh to half the current window
2550 * and pull our congestion window back to
2551 * the new ssthresh).
2553 * Dup acks mean that packets have left the
2554 * network (they're now cached at the receiver)
2555 * so bump cwnd by the amount in the receiver
2556 * to keep a constant cwnd packets in the
2559 * When using TCP ECN, notify the peer that
2560 * we reduced the cwnd.
2563 * Following 2 kinds of acks should not affect
2566 * 2) Acks with SACK but without any new SACK
2567 * information in them. These could result from
2568 * any anomaly in the network like a switch
2569 * duplicating packets or a possible DoS attack.
2571 if (th->th_ack != tp->snd_una ||
2572 ((tp->t_flags & TF_SACK_PERMIT) &&
2575 else if (!tcp_timer_active(tp, TT_REXMT))
2577 else if (++tp->t_dupacks > tcprexmtthresh ||
2578 IN_FASTRECOVERY(tp->t_flags)) {
2579 cc_ack_received(tp, th, nsegs,
2581 if ((tp->t_flags & TF_SACK_PERMIT) &&
2582 IN_FASTRECOVERY(tp->t_flags)) {
2586 * Compute the amount of data in flight first.
2587 * We can inject new data into the pipe iff
2588 * we have less than 1/2 the original window's
2589 * worth of data in flight.
2591 if (V_tcp_do_rfc6675_pipe)
2592 awnd = tcp_compute_pipe(tp);
2594 awnd = (tp->snd_nxt - tp->snd_fack) +
2595 tp->sackhint.sack_bytes_rexmit;
2597 if (awnd < tp->snd_ssthresh) {
2598 tp->snd_cwnd += maxseg;
2599 if (tp->snd_cwnd > tp->snd_ssthresh)
2600 tp->snd_cwnd = tp->snd_ssthresh;
2603 tp->snd_cwnd += maxseg;
2604 (void) tp->t_fb->tfb_tcp_output(tp);
2606 } else if (tp->t_dupacks == tcprexmtthresh) {
2607 tcp_seq onxt = tp->snd_nxt;
2610 * If we're doing sack, check to
2611 * see if we're already in sack
2612 * recovery. If we're not doing sack,
2613 * check to see if we're in newreno
2616 if (tp->t_flags & TF_SACK_PERMIT) {
2617 if (IN_FASTRECOVERY(tp->t_flags)) {
2622 if (SEQ_LEQ(th->th_ack,
2628 /* Congestion signal before ack. */
2629 cc_cong_signal(tp, th, CC_NDUPACK);
2630 cc_ack_received(tp, th, nsegs,
2632 tcp_timer_activate(tp, TT_REXMT, 0);
2634 if (tp->t_flags & TF_SACK_PERMIT) {
2636 tcps_sack_recovery_episode);
2637 tp->sack_newdata = tp->snd_nxt;
2638 tp->snd_cwnd = maxseg;
2639 (void) tp->t_fb->tfb_tcp_output(tp);
2642 tp->snd_nxt = th->th_ack;
2643 tp->snd_cwnd = maxseg;
2644 (void) tp->t_fb->tfb_tcp_output(tp);
2645 KASSERT(tp->snd_limited <= 2,
2646 ("%s: tp->snd_limited too big",
2648 tp->snd_cwnd = tp->snd_ssthresh +
2650 (tp->t_dupacks - tp->snd_limited);
2651 if (SEQ_GT(onxt, tp->snd_nxt))
2654 } else if (V_tcp_do_rfc3042) {
2656 * Process first and second duplicate
2657 * ACKs. Each indicates a segment
2658 * leaving the network, creating room
2659 * for more. Make sure we can send a
2660 * packet on reception of each duplicate
2661 * ACK by increasing snd_cwnd by one
2662 * segment. Restore the original
2663 * snd_cwnd after packet transmission.
2665 cc_ack_received(tp, th, nsegs,
2667 uint32_t oldcwnd = tp->snd_cwnd;
2668 tcp_seq oldsndmax = tp->snd_max;
2672 KASSERT(tp->t_dupacks == 1 ||
2674 ("%s: dupacks not 1 or 2",
2676 if (tp->t_dupacks == 1)
2677 tp->snd_limited = 0;
2679 (tp->snd_nxt - tp->snd_una) +
2680 (tp->t_dupacks - tp->snd_limited) *
2683 * Only call tcp_output when there
2684 * is new data available to be sent.
2685 * Otherwise we would send pure ACKs.
2687 SOCKBUF_LOCK(&so->so_snd);
2688 avail = sbavail(&so->so_snd) -
2689 (tp->snd_nxt - tp->snd_una);
2690 SOCKBUF_UNLOCK(&so->so_snd);
2692 (void) tp->t_fb->tfb_tcp_output(tp);
2693 sent = tp->snd_max - oldsndmax;
2694 if (sent > maxseg) {
2695 KASSERT((tp->t_dupacks == 2 &&
2696 tp->snd_limited == 0) ||
2697 (sent == maxseg + 1 &&
2698 tp->t_flags & TF_SENTFIN),
2699 ("%s: sent too much",
2701 tp->snd_limited = 2;
2702 } else if (sent > 0)
2704 tp->snd_cwnd = oldcwnd;
2711 * This ack is advancing the left edge, reset the
2716 * If this ack also has new SACK info, increment the
2717 * counter as per rfc6675.
2719 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed)
2723 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2724 ("%s: th_ack <= snd_una", __func__));
2727 * If the congestion window was inflated to account
2728 * for the other side's cached packets, retract it.
2730 if (IN_FASTRECOVERY(tp->t_flags)) {
2731 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2732 if (tp->t_flags & TF_SACK_PERMIT)
2733 tcp_sack_partialack(tp, th);
2735 tcp_newreno_partial_ack(tp, th);
2737 cc_post_recovery(tp, th);
2740 * If we reach this point, ACK is not a duplicate,
2741 * i.e., it ACKs something we sent.
2743 if (tp->t_flags & TF_NEEDSYN) {
2745 * T/TCP: Connection was half-synchronized, and our
2746 * SYN has been ACK'd (so connection is now fully
2747 * synchronized). Go to non-starred state,
2748 * increment snd_una for ACK of SYN, and check if
2749 * we can do window scaling.
2751 tp->t_flags &= ~TF_NEEDSYN;
2753 /* Do window scaling? */
2754 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2755 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2756 tp->rcv_scale = tp->request_r_scale;
2757 /* Send window already scaled. */
2762 INP_WLOCK_ASSERT(tp->t_inpcb);
2765 * Adjust for the SYN bit in sequence space,
2766 * but don't account for it in cwnd calculations.
2767 * This is for the SYN_RECEIVED, non-simultaneous
2768 * SYN case. SYN_SENT and simultaneous SYN are
2769 * treated elsewhere.
2773 acked = BYTES_THIS_ACK(tp, th);
2774 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2775 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2776 tp->snd_una, th->th_ack, tp, m));
2777 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2778 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2781 * If we just performed our first retransmit, and the ACK
2782 * arrives within our recovery window, then it was a mistake
2783 * to do the retransmit in the first place. Recover our
2784 * original cwnd and ssthresh, and proceed to transmit where
2787 if (tp->t_rxtshift == 1 &&
2788 tp->t_flags & TF_PREVVALID &&
2790 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2791 cc_cong_signal(tp, th, CC_RTO_ERR);
2794 * If we have a timestamp reply, update smoothed
2795 * round trip time. If no timestamp is present but
2796 * transmit timer is running and timed sequence
2797 * number was acked, update smoothed round trip time.
2798 * Since we now have an rtt measurement, cancel the
2799 * timer backoff (cf., Phil Karn's retransmit alg.).
2800 * Recompute the initial retransmit timer.
2802 * Some boxes send broken timestamp replies
2803 * during the SYN+ACK phase, ignore
2804 * timestamps of 0 or we could calculate a
2805 * huge RTT and blow up the retransmit timer.
2807 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2810 t = tcp_ts_getticks() - to.to_tsecr;
2811 if (!tp->t_rttlow || tp->t_rttlow > t)
2813 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2814 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2815 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2816 tp->t_rttlow = ticks - tp->t_rtttime;
2817 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2821 * If all outstanding data is acked, stop retransmit
2822 * timer and remember to restart (more output or persist).
2823 * If there is more data to be acked, restart retransmit
2824 * timer, using current (possibly backed-off) value.
2826 if (th->th_ack == tp->snd_max) {
2827 tcp_timer_activate(tp, TT_REXMT, 0);
2829 } else if (!tcp_timer_active(tp, TT_PERSIST))
2830 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2833 * If no data (only SYN) was ACK'd,
2834 * skip rest of ACK processing.
2840 * Let the congestion control algorithm update congestion
2841 * control related information. This typically means increasing
2842 * the congestion window.
2844 cc_ack_received(tp, th, nsegs, CC_ACK);
2846 SOCKBUF_LOCK(&so->so_snd);
2847 if (acked > sbavail(&so->so_snd)) {
2848 if (tp->snd_wnd >= sbavail(&so->so_snd))
2849 tp->snd_wnd -= sbavail(&so->so_snd);
2852 mfree = sbcut_locked(&so->so_snd,
2853 (int)sbavail(&so->so_snd));
2856 mfree = sbcut_locked(&so->so_snd, acked);
2857 if (tp->snd_wnd >= (uint32_t) acked)
2858 tp->snd_wnd -= acked;
2863 /* NB: sowwakeup_locked() does an implicit unlock. */
2864 sowwakeup_locked(so);
2866 /* Detect una wraparound. */
2867 if (!IN_RECOVERY(tp->t_flags) &&
2868 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2869 SEQ_LEQ(th->th_ack, tp->snd_recover))
2870 tp->snd_recover = th->th_ack - 1;
2871 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2872 if (IN_RECOVERY(tp->t_flags) &&
2873 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2874 EXIT_RECOVERY(tp->t_flags);
2876 tp->snd_una = th->th_ack;
2877 if (tp->t_flags & TF_SACK_PERMIT) {
2878 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2879 tp->snd_recover = tp->snd_una;
2881 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2882 tp->snd_nxt = tp->snd_una;
2884 switch (tp->t_state) {
2887 * In FIN_WAIT_1 STATE in addition to the processing
2888 * for the ESTABLISHED state if our FIN is now acknowledged
2889 * then enter FIN_WAIT_2.
2891 case TCPS_FIN_WAIT_1:
2892 if (ourfinisacked) {
2894 * If we can't receive any more
2895 * data, then closing user can proceed.
2896 * Starting the timer is contrary to the
2897 * specification, but if we don't get a FIN
2898 * we'll hang forever.
2901 * we should release the tp also, and use a
2904 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2905 soisdisconnected(so);
2906 tcp_timer_activate(tp, TT_2MSL,
2907 (tcp_fast_finwait2_recycle ?
2908 tcp_finwait2_timeout :
2911 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2916 * In CLOSING STATE in addition to the processing for
2917 * the ESTABLISHED state if the ACK acknowledges our FIN
2918 * then enter the TIME-WAIT state, otherwise ignore
2922 if (ourfinisacked) {
2923 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2931 * In LAST_ACK, we may still be waiting for data to drain
2932 * and/or to be acked, as well as for the ack of our FIN.
2933 * If our FIN is now acknowledged, delete the TCB,
2934 * enter the closed state and return.
2937 if (ourfinisacked) {
2938 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2947 INP_WLOCK_ASSERT(tp->t_inpcb);
2950 * Update window information.
2951 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2953 if ((thflags & TH_ACK) &&
2954 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2955 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2956 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2957 /* keep track of pure window updates */
2959 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2960 TCPSTAT_INC(tcps_rcvwinupd);
2961 tp->snd_wnd = tiwin;
2962 tp->snd_wl1 = th->th_seq;
2963 tp->snd_wl2 = th->th_ack;
2964 if (tp->snd_wnd > tp->max_sndwnd)
2965 tp->max_sndwnd = tp->snd_wnd;
2970 * Process segments with URG.
2972 if ((thflags & TH_URG) && th->th_urp &&
2973 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2975 * This is a kludge, but if we receive and accept
2976 * random urgent pointers, we'll crash in
2977 * soreceive. It's hard to imagine someone
2978 * actually wanting to send this much urgent data.
2980 SOCKBUF_LOCK(&so->so_rcv);
2981 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2982 th->th_urp = 0; /* XXX */
2983 thflags &= ~TH_URG; /* XXX */
2984 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2985 goto dodata; /* XXX */
2988 * If this segment advances the known urgent pointer,
2989 * then mark the data stream. This should not happen
2990 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2991 * a FIN has been received from the remote side.
2992 * In these states we ignore the URG.
2994 * According to RFC961 (Assigned Protocols),
2995 * the urgent pointer points to the last octet
2996 * of urgent data. We continue, however,
2997 * to consider it to indicate the first octet
2998 * of data past the urgent section as the original
2999 * spec states (in one of two places).
3001 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3002 tp->rcv_up = th->th_seq + th->th_urp;
3003 so->so_oobmark = sbavail(&so->so_rcv) +
3004 (tp->rcv_up - tp->rcv_nxt) - 1;
3005 if (so->so_oobmark == 0)
3006 so->so_rcv.sb_state |= SBS_RCVATMARK;
3008 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3010 SOCKBUF_UNLOCK(&so->so_rcv);
3012 * Remove out of band data so doesn't get presented to user.
3013 * This can happen independent of advancing the URG pointer,
3014 * but if two URG's are pending at once, some out-of-band
3015 * data may creep in... ick.
3017 if (th->th_urp <= (uint32_t)tlen &&
3018 !(so->so_options & SO_OOBINLINE)) {
3019 /* hdr drop is delayed */
3020 tcp_pulloutofband(so, th, m, drop_hdrlen);
3024 * If no out of band data is expected,
3025 * pull receive urgent pointer along
3026 * with the receive window.
3028 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3029 tp->rcv_up = tp->rcv_nxt;
3032 INP_WLOCK_ASSERT(tp->t_inpcb);
3035 * Process the segment text, merging it into the TCP sequencing queue,
3036 * and arranging for acknowledgment of receipt if necessary.
3037 * This process logically involves adjusting tp->rcv_wnd as data
3038 * is presented to the user (this happens in tcp_usrreq.c,
3039 * case PRU_RCVD). If a FIN has already been received on this
3040 * connection then we just ignore the text.
3042 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3043 IS_FASTOPEN(tp->t_flags));
3044 if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
3045 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3046 tcp_seq save_start = th->th_seq;
3047 tcp_seq save_rnxt = tp->rcv_nxt;
3048 int save_tlen = tlen;
3049 m_adj(m, drop_hdrlen); /* delayed header drop */
3051 * Insert segment which includes th into TCP reassembly queue
3052 * with control block tp. Set thflags to whether reassembly now
3053 * includes a segment with FIN. This handles the common case
3054 * inline (segment is the next to be received on an established
3055 * connection, and the queue is empty), avoiding linkage into
3056 * and removal from the queue and repetition of various
3058 * Set DELACK for segments received in order, but ack
3059 * immediately when segments are out of order (so
3060 * fast retransmit can work).
3062 if (th->th_seq == tp->rcv_nxt &&
3064 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3066 if (DELAY_ACK(tp, tlen) || tfo_syn)
3067 tp->t_flags |= TF_DELACK;
3069 tp->t_flags |= TF_ACKNOW;
3070 tp->rcv_nxt += tlen;
3071 thflags = th->th_flags & TH_FIN;
3072 TCPSTAT_INC(tcps_rcvpack);
3073 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3074 SOCKBUF_LOCK(&so->so_rcv);
3075 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3078 sbappendstream_locked(&so->so_rcv, m, 0);
3079 /* NB: sorwakeup_locked() does an implicit unlock. */
3080 sorwakeup_locked(so);
3083 * XXX: Due to the header drop above "th" is
3084 * theoretically invalid by now. Fortunately
3085 * m_adj() doesn't actually frees any mbufs
3086 * when trimming from the head.
3088 tcp_seq temp = save_start;
3089 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3090 tp->t_flags |= TF_ACKNOW;
3092 if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) {
3093 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3095 * DSACK actually handled in the fastpath
3098 tcp_update_sack_list(tp, save_start,
3099 save_start + save_tlen);
3100 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3101 if ((tp->rcv_numsacks >= 1) &&
3102 (tp->sackblks[0].end == save_start)) {
3104 * Partial overlap, recorded at todrop
3107 tcp_update_sack_list(tp,
3108 tp->sackblks[0].start,
3109 tp->sackblks[0].end);
3111 tcp_update_dsack_list(tp, save_start,
3112 save_start + save_tlen);
3114 } else if (tlen >= save_tlen) {
3115 /* Update of sackblks. */
3116 tcp_update_dsack_list(tp, save_start,
3117 save_start + save_tlen);
3118 } else if (tlen > 0) {
3119 tcp_update_dsack_list(tp, save_start,
3125 * Note the amount of data that peer has sent into
3126 * our window, in order to estimate the sender's
3130 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3131 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3133 len = so->so_rcv.sb_hiwat;
3141 * If FIN is received ACK the FIN and let the user know
3142 * that the connection is closing.
3144 if (thflags & TH_FIN) {
3145 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3148 * If connection is half-synchronized
3149 * (ie NEEDSYN flag on) then delay ACK,
3150 * so it may be piggybacked when SYN is sent.
3151 * Otherwise, since we received a FIN then no
3152 * more input can be expected, send ACK now.
3154 if (tp->t_flags & TF_NEEDSYN)
3155 tp->t_flags |= TF_DELACK;
3157 tp->t_flags |= TF_ACKNOW;
3160 switch (tp->t_state) {
3163 * In SYN_RECEIVED and ESTABLISHED STATES
3164 * enter the CLOSE_WAIT state.
3166 case TCPS_SYN_RECEIVED:
3167 tp->t_starttime = ticks;
3169 case TCPS_ESTABLISHED:
3170 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3174 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3175 * enter the CLOSING state.
3177 case TCPS_FIN_WAIT_1:
3178 tcp_state_change(tp, TCPS_CLOSING);
3182 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3183 * starting the time-wait timer, turning off the other
3186 case TCPS_FIN_WAIT_2:
3187 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
3194 if (so->so_options & SO_DEBUG)
3195 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3198 TCP_PROBE3(debug__input, tp, th, m);
3201 * Return any desired output.
3203 if (needoutput || (tp->t_flags & TF_ACKNOW))
3204 (void) tp->t_fb->tfb_tcp_output(tp);
3207 INP_WLOCK_ASSERT(tp->t_inpcb);
3209 if (tp->t_flags & TF_DELACK) {
3210 tp->t_flags &= ~TF_DELACK;
3211 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3213 INP_WUNLOCK(tp->t_inpcb);
3218 * Generate an ACK dropping incoming segment if it occupies
3219 * sequence space, where the ACK reflects our state.
3221 * We can now skip the test for the RST flag since all
3222 * paths to this code happen after packets containing
3223 * RST have been dropped.
3225 * In the SYN-RECEIVED state, don't send an ACK unless the
3226 * segment we received passes the SYN-RECEIVED ACK test.
3227 * If it fails send a RST. This breaks the loop in the
3228 * "LAND" DoS attack, and also prevents an ACK storm
3229 * between two listening ports that have been sent forged
3230 * SYN segments, each with the source address of the other.
3232 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3233 (SEQ_GT(tp->snd_una, th->th_ack) ||
3234 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3235 rstreason = BANDLIM_RST_OPENPORT;
3239 if (so->so_options & SO_DEBUG)
3240 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3243 TCP_PROBE3(debug__input, tp, th, m);
3244 tp->t_flags |= TF_ACKNOW;
3245 (void) tp->t_fb->tfb_tcp_output(tp);
3246 INP_WUNLOCK(tp->t_inpcb);
3252 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3253 INP_WUNLOCK(tp->t_inpcb);
3255 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3260 * Drop space held by incoming segment and return.
3263 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3264 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3267 TCP_PROBE3(debug__input, tp, th, m);
3269 INP_WUNLOCK(tp->t_inpcb);
3274 * Issue RST and make ACK acceptable to originator of segment.
3275 * The mbuf must still include the original packet header.
3279 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3280 int tlen, int rstreason)
3286 struct ip6_hdr *ip6;
3290 INP_WLOCK_ASSERT(tp->t_inpcb);
3293 /* Don't bother if destination was broadcast/multicast. */
3294 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3297 if (mtod(m, struct ip *)->ip_v == 6) {
3298 ip6 = mtod(m, struct ip6_hdr *);
3299 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3300 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3302 /* IPv6 anycast check is done at tcp6_input() */
3305 #if defined(INET) && defined(INET6)
3310 ip = mtod(m, struct ip *);
3311 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3312 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3313 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3314 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3319 /* Perform bandwidth limiting. */
3320 if (badport_bandlim(rstreason) < 0)
3323 /* tcp_respond consumes the mbuf chain. */
3324 if (th->th_flags & TH_ACK) {
3325 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3326 th->th_ack, TH_RST);
3328 if (th->th_flags & TH_SYN)
3330 if (th->th_flags & TH_FIN)
3332 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3333 (tcp_seq)0, TH_RST|TH_ACK);
3341 * Parse TCP options and place in tcpopt.
3344 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3349 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3351 if (opt == TCPOPT_EOL)
3353 if (opt == TCPOPT_NOP)
3359 if (optlen < 2 || optlen > cnt)
3364 if (optlen != TCPOLEN_MAXSEG)
3366 if (!(flags & TO_SYN))
3368 to->to_flags |= TOF_MSS;
3369 bcopy((char *)cp + 2,
3370 (char *)&to->to_mss, sizeof(to->to_mss));
3371 to->to_mss = ntohs(to->to_mss);
3374 if (optlen != TCPOLEN_WINDOW)
3376 if (!(flags & TO_SYN))
3378 to->to_flags |= TOF_SCALE;
3379 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3381 case TCPOPT_TIMESTAMP:
3382 if (optlen != TCPOLEN_TIMESTAMP)
3384 to->to_flags |= TOF_TS;
3385 bcopy((char *)cp + 2,
3386 (char *)&to->to_tsval, sizeof(to->to_tsval));
3387 to->to_tsval = ntohl(to->to_tsval);
3388 bcopy((char *)cp + 6,
3389 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3390 to->to_tsecr = ntohl(to->to_tsecr);
3392 case TCPOPT_SIGNATURE:
3394 * In order to reply to a host which has set the
3395 * TCP_SIGNATURE option in its initial SYN, we have
3396 * to record the fact that the option was observed
3397 * here for the syncache code to perform the correct
3400 if (optlen != TCPOLEN_SIGNATURE)
3402 to->to_flags |= TOF_SIGNATURE;
3403 to->to_signature = cp + 2;
3405 case TCPOPT_SACK_PERMITTED:
3406 if (optlen != TCPOLEN_SACK_PERMITTED)
3408 if (!(flags & TO_SYN))
3412 to->to_flags |= TOF_SACKPERM;
3415 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3419 to->to_flags |= TOF_SACK;
3420 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3421 to->to_sacks = cp + 2;
3422 TCPSTAT_INC(tcps_sack_rcv_blocks);
3424 case TCPOPT_FAST_OPEN:
3426 * Cookie length validation is performed by the
3427 * server side cookie checking code or the client
3428 * side cookie cache update code.
3430 if (!(flags & TO_SYN))
3432 if (!V_tcp_fastopen_client_enable &&
3433 !V_tcp_fastopen_server_enable)
3435 to->to_flags |= TOF_FASTOPEN;
3436 to->to_tfo_len = optlen - 2;
3437 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3446 * Pull out of band byte out of a segment so
3447 * it doesn't appear in the user's data queue.
3448 * It is still reflected in the segment length for
3449 * sequencing purposes.
3452 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3455 int cnt = off + th->th_urp - 1;
3458 if (m->m_len > cnt) {
3459 char *cp = mtod(m, caddr_t) + cnt;
3460 struct tcpcb *tp = sototcpcb(so);
3462 INP_WLOCK_ASSERT(tp->t_inpcb);
3465 tp->t_oobflags |= TCPOOB_HAVEDATA;
3466 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3468 if (m->m_flags & M_PKTHDR)
3477 panic("tcp_pulloutofband");
3481 * Collect new round-trip time estimate
3482 * and update averages and current timeout.
3485 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3489 INP_WLOCK_ASSERT(tp->t_inpcb);
3491 TCPSTAT_INC(tcps_rttupdated);
3493 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3495 * srtt is stored as fixed point with 5 bits after the
3496 * binary point (i.e., scaled by 8). The following magic
3497 * is equivalent to the smoothing algorithm in rfc793 with
3498 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3499 * point). Adjust rtt to origin 0.
3501 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3502 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3504 if ((tp->t_srtt += delta) <= 0)
3508 * We accumulate a smoothed rtt variance (actually, a
3509 * smoothed mean difference), then set the retransmit
3510 * timer to smoothed rtt + 4 times the smoothed variance.
3511 * rttvar is stored as fixed point with 4 bits after the
3512 * binary point (scaled by 16). The following is
3513 * equivalent to rfc793 smoothing with an alpha of .75
3514 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3515 * rfc793's wired-in beta.
3519 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3520 if ((tp->t_rttvar += delta) <= 0)
3522 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3523 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3526 * No rtt measurement yet - use the unsmoothed rtt.
3527 * Set the variance to half the rtt (so our first
3528 * retransmit happens at 3*rtt).
3530 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3531 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3532 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3538 * the retransmit should happen at rtt + 4 * rttvar.
3539 * Because of the way we do the smoothing, srtt and rttvar
3540 * will each average +1/2 tick of bias. When we compute
3541 * the retransmit timer, we want 1/2 tick of rounding and
3542 * 1 extra tick because of +-1/2 tick uncertainty in the
3543 * firing of the timer. The bias will give us exactly the
3544 * 1.5 tick we need. But, because the bias is
3545 * statistical, we have to test that we don't drop below
3546 * the minimum feasible timer (which is 2 ticks).
3548 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3549 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3552 * We received an ack for a packet that wasn't retransmitted;
3553 * it is probably safe to discard any error indications we've
3554 * received recently. This isn't quite right, but close enough
3555 * for now (a route might have failed after we sent a segment,
3556 * and the return path might not be symmetrical).
3558 tp->t_softerror = 0;
3562 * Determine a reasonable value for maxseg size.
3563 * If the route is known, check route for mtu.
3564 * If none, use an mss that can be handled on the outgoing interface
3565 * without forcing IP to fragment. If no route is found, route has no mtu,
3566 * or the destination isn't local, use a default, hopefully conservative
3567 * size (usually 512 or the default IP max size, but no more than the mtu
3568 * of the interface), as we can't discover anything about intervening
3569 * gateways or networks. We also initialize the congestion/slow start
3570 * window to be a single segment if the destination isn't local.
3571 * While looking at the routing entry, we also initialize other path-dependent
3572 * parameters from pre-set or cached values in the routing entry.
3574 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3575 * IP options, e.g. IPSEC data, since length of this data may vary, and
3576 * thus it is calculated for every segment separately in tcp_output().
3578 * NOTE that this routine is only called when we process an incoming
3579 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3580 * settings are handled in tcp_mssopt().
3583 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3584 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3587 uint32_t maxmtu = 0;
3588 struct inpcb *inp = tp->t_inpcb;
3589 struct hc_metrics_lite metrics;
3591 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3592 size_t min_protoh = isipv6 ?
3593 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3594 sizeof (struct tcpiphdr);
3596 const size_t min_protoh = sizeof(struct tcpiphdr);
3599 INP_WLOCK_ASSERT(tp->t_inpcb);
3601 if (mtuoffer != -1) {
3602 KASSERT(offer == -1, ("%s: conflict", __func__));
3603 offer = mtuoffer - min_protoh;
3609 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3610 tp->t_maxseg = V_tcp_v6mssdflt;
3613 #if defined(INET) && defined(INET6)
3618 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3619 tp->t_maxseg = V_tcp_mssdflt;
3624 * No route to sender, stay with default mss and return.
3628 * In case we return early we need to initialize metrics
3629 * to a defined state as tcp_hc_get() would do for us
3630 * if there was no cache hit.
3632 if (metricptr != NULL)
3633 bzero(metricptr, sizeof(struct hc_metrics_lite));
3637 /* What have we got? */
3641 * Offer == 0 means that there was no MSS on the SYN
3642 * segment, in this case we use tcp_mssdflt as
3643 * already assigned to t_maxseg above.
3645 offer = tp->t_maxseg;
3650 * Offer == -1 means that we didn't receive SYN yet.
3656 * Prevent DoS attack with too small MSS. Round up
3657 * to at least minmss.
3659 offer = max(offer, V_tcp_minmss);
3663 * rmx information is now retrieved from tcp_hostcache.
3665 tcp_hc_get(&inp->inp_inc, &metrics);
3666 if (metricptr != NULL)
3667 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3670 * If there's a discovered mtu in tcp hostcache, use it.
3671 * Else, use the link mtu.
3673 if (metrics.rmx_mtu)
3674 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3678 mss = maxmtu - min_protoh;
3679 if (!V_path_mtu_discovery &&
3680 !in6_localaddr(&inp->in6p_faddr))
3681 mss = min(mss, V_tcp_v6mssdflt);
3684 #if defined(INET) && defined(INET6)
3689 mss = maxmtu - min_protoh;
3690 if (!V_path_mtu_discovery &&
3691 !in_localaddr(inp->inp_faddr))
3692 mss = min(mss, V_tcp_mssdflt);
3696 * XXX - The above conditional (mss = maxmtu - min_protoh)
3697 * probably violates the TCP spec.
3698 * The problem is that, since we don't know the
3699 * other end's MSS, we are supposed to use a conservative
3700 * default. But, if we do that, then MTU discovery will
3701 * never actually take place, because the conservative
3702 * default is much less than the MTUs typically seen
3703 * on the Internet today. For the moment, we'll sweep
3704 * this under the carpet.
3706 * The conservative default might not actually be a problem
3707 * if the only case this occurs is when sending an initial
3708 * SYN with options and data to a host we've never talked
3709 * to before. Then, they will reply with an MSS value which
3710 * will get recorded and the new parameters should get
3711 * recomputed. For Further Study.
3714 mss = min(mss, offer);
3717 * Sanity check: make sure that maxseg will be large
3718 * enough to allow some data on segments even if the
3719 * all the option space is used (40bytes). Otherwise
3720 * funny things may happen in tcp_output.
3722 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3730 tcp_mss(struct tcpcb *tp, int offer)
3736 struct hc_metrics_lite metrics;
3737 struct tcp_ifcap cap;
3739 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3741 bzero(&cap, sizeof(cap));
3742 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3748 * If there's a pipesize, change the socket buffer to that size,
3749 * don't change if sb_hiwat is different than default (then it
3750 * has been changed on purpose with setsockopt).
3751 * Make the socket buffers an integral number of mss units;
3752 * if the mss is larger than the socket buffer, decrease the mss.
3754 so = inp->inp_socket;
3755 SOCKBUF_LOCK(&so->so_snd);
3756 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3757 bufsize = metrics.rmx_sendpipe;
3759 bufsize = so->so_snd.sb_hiwat;
3763 bufsize = roundup(bufsize, mss);
3764 if (bufsize > sb_max)
3766 if (bufsize > so->so_snd.sb_hiwat)
3767 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3769 SOCKBUF_UNLOCK(&so->so_snd);
3771 * Sanity check: make sure that maxseg will be large
3772 * enough to allow some data on segments even if the
3773 * all the option space is used (40bytes). Otherwise
3774 * funny things may happen in tcp_output.
3776 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3778 tp->t_maxseg = max(mss, 64);
3780 SOCKBUF_LOCK(&so->so_rcv);
3781 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3782 bufsize = metrics.rmx_recvpipe;
3784 bufsize = so->so_rcv.sb_hiwat;
3785 if (bufsize > mss) {
3786 bufsize = roundup(bufsize, mss);
3787 if (bufsize > sb_max)
3789 if (bufsize > so->so_rcv.sb_hiwat)
3790 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3792 SOCKBUF_UNLOCK(&so->so_rcv);
3794 /* Check the interface for TSO capabilities. */
3795 if (cap.ifcap & CSUM_TSO) {
3796 tp->t_flags |= TF_TSO;
3797 tp->t_tsomax = cap.tsomax;
3798 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3799 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3804 * Determine the MSS option to send on an outgoing SYN.
3807 tcp_mssopt(struct in_conninfo *inc)
3810 uint32_t thcmtu = 0;
3811 uint32_t maxmtu = 0;
3814 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3817 if (inc->inc_flags & INC_ISIPV6) {
3818 mss = V_tcp_v6mssdflt;
3819 maxmtu = tcp_maxmtu6(inc, NULL);
3820 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3823 #if defined(INET) && defined(INET6)
3828 mss = V_tcp_mssdflt;
3829 maxmtu = tcp_maxmtu(inc, NULL);
3830 min_protoh = sizeof(struct tcpiphdr);
3833 #if defined(INET6) || defined(INET)
3834 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3837 if (maxmtu && thcmtu)
3838 mss = min(maxmtu, thcmtu) - min_protoh;
3839 else if (maxmtu || thcmtu)
3840 mss = max(maxmtu, thcmtu) - min_protoh;
3847 * On a partial ack arrives, force the retransmission of the
3848 * next unacknowledged segment. Do not clear tp->t_dupacks.
3849 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3853 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3855 tcp_seq onxt = tp->snd_nxt;
3856 uint32_t ocwnd = tp->snd_cwnd;
3857 u_int maxseg = tcp_maxseg(tp);
3859 INP_WLOCK_ASSERT(tp->t_inpcb);
3861 tcp_timer_activate(tp, TT_REXMT, 0);
3863 tp->snd_nxt = th->th_ack;
3865 * Set snd_cwnd to one segment beyond acknowledged offset.
3866 * (tp->snd_una has not yet been updated when this function is called.)
3868 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3869 tp->t_flags |= TF_ACKNOW;
3870 (void) tp->t_fb->tfb_tcp_output(tp);
3871 tp->snd_cwnd = ocwnd;
3872 if (SEQ_GT(onxt, tp->snd_nxt))
3875 * Partial window deflation. Relies on fact that tp->snd_una
3878 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3879 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3882 tp->snd_cwnd += maxseg;
3886 tcp_compute_pipe(struct tcpcb *tp)
3888 return (tp->snd_max - tp->snd_una +
3889 tp->sackhint.sack_bytes_rexmit -
3890 tp->sackhint.sacked_bytes);