2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
5 * The Regents of the University of California. All rights reserved.
6 * Copyright (c) 2007-2008,2010
7 * Swinburne University of Technology, Melbourne, Australia.
8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
9 * Copyright (c) 2010 The FreeBSD Foundation
10 * Copyright (c) 2010-2011 Juniper Networks, Inc.
11 * All rights reserved.
13 * Portions of this software were developed at the Centre for Advanced Internet
14 * Architectures, Swinburne University of Technology, by Lawrence Stewart,
15 * James Healy and David Hayes, made possible in part by a grant from the Cisco
16 * University Research Program Fund at Community Foundation Silicon Valley.
18 * Portions of this software were developed at the Centre for Advanced
19 * Internet Architectures, Swinburne University of Technology, Melbourne,
20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
22 * Portions of this software were developed by Robert N. M. Watson under
23 * contract to Juniper Networks, Inc.
25 * Redistribution and use in source and binary forms, with or without
26 * modification, are permitted provided that the following conditions
28 * 1. Redistributions of source code must retain the above copyright
29 * notice, this list of conditions and the following disclaimer.
30 * 2. Redistributions in binary form must reproduce the above copyright
31 * notice, this list of conditions and the following disclaimer in the
32 * documentation and/or other materials provided with the distribution.
33 * 3. Neither the name of the University nor the names of its contributors
34 * may be used to endorse or promote products derived from this software
35 * without specific prior written permission.
37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
49 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
52 #include <sys/cdefs.h>
53 __FBSDID("$FreeBSD$");
56 #include "opt_inet6.h"
57 #include "opt_ipsec.h"
58 #include "opt_tcpdebug.h"
60 #include <sys/param.h>
61 #include <sys/kernel.h>
63 #include <sys/hhook.h>
65 #include <sys/malloc.h>
67 #include <sys/proc.h> /* for proc0 declaration */
68 #include <sys/protosw.h>
70 #include <sys/signalvar.h>
71 #include <sys/socket.h>
72 #include <sys/socketvar.h>
73 #include <sys/sysctl.h>
74 #include <sys/syslog.h>
75 #include <sys/systm.h>
77 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
82 #include <net/if_var.h>
83 #include <net/route.h>
86 #define TCPSTATES /* for logging */
88 #include <netinet/in.h>
89 #include <netinet/in_kdtrace.h>
90 #include <netinet/in_pcb.h>
91 #include <netinet/in_systm.h>
92 #include <netinet/ip.h>
93 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
94 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
95 #include <netinet/ip_var.h>
96 #include <netinet/ip_options.h>
97 #include <netinet/ip6.h>
98 #include <netinet/icmp6.h>
99 #include <netinet6/in6_pcb.h>
100 #include <netinet6/in6_var.h>
101 #include <netinet6/ip6_var.h>
102 #include <netinet6/nd6.h>
103 #include <netinet/tcp.h>
104 #include <netinet/tcp_fsm.h>
105 #include <netinet/tcp_log_buf.h>
106 #include <netinet/tcp_seq.h>
107 #include <netinet/tcp_timer.h>
108 #include <netinet/tcp_var.h>
109 #include <netinet6/tcp6_var.h>
110 #include <netinet/tcpip.h>
111 #include <netinet/cc/cc.h>
112 #include <netinet/tcp_fastopen.h>
114 #include <netinet/tcp_pcap.h>
116 #include <netinet/tcp_syncache.h>
118 #include <netinet/tcp_debug.h>
119 #endif /* TCPDEBUG */
121 #include <netinet/tcp_offload.h>
124 #include <netipsec/ipsec_support.h>
126 #include <machine/in_cksum.h>
128 #include <security/mac/mac_framework.h>
130 const int tcprexmtthresh = 3;
132 int tcp_log_in_vain = 0;
133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
135 "Log all incoming TCP segments to closed ports");
137 VNET_DEFINE(int, blackhole) = 0;
138 #define V_blackhole VNET(blackhole)
139 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
140 &VNET_NAME(blackhole), 0,
141 "Do not send RST on segments to closed ports");
143 VNET_DEFINE(int, tcp_delack_enabled) = 1;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
145 &VNET_NAME(tcp_delack_enabled), 0,
146 "Delay ACK to try and piggyback it onto a data packet");
148 VNET_DEFINE(int, drop_synfin) = 0;
149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
150 &VNET_NAME(drop_synfin), 0,
151 "Drop TCP packets with SYN+FIN set");
153 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
155 &VNET_NAME(tcp_do_rfc6675_pipe), 0,
156 "Use calculated pipe/in-flight bytes per RFC 6675");
158 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
160 &VNET_NAME(tcp_do_rfc3042), 0,
161 "Enable RFC 3042 (Limited Transmit)");
163 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
165 &VNET_NAME(tcp_do_rfc3390), 0,
166 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
168 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
170 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
171 "Slow-start flight size (initial congestion window) in number of segments");
173 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
175 &VNET_NAME(tcp_do_rfc3465), 0,
176 "Enable RFC 3465 (Appropriate Byte Counting)");
178 VNET_DEFINE(int, tcp_abc_l_var) = 2;
179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
180 &VNET_NAME(tcp_abc_l_var), 2,
181 "Cap the max cwnd increment during slow-start to this number of segments");
183 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
185 VNET_DEFINE(int, tcp_do_ecn) = 2;
186 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
187 &VNET_NAME(tcp_do_ecn), 0,
190 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
191 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
192 &VNET_NAME(tcp_ecn_maxretries), 0,
193 "Max retries before giving up on ECN");
195 VNET_DEFINE(int, tcp_insecure_syn) = 0;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
197 &VNET_NAME(tcp_insecure_syn), 0,
198 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
200 VNET_DEFINE(int, tcp_insecure_rst) = 0;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
202 &VNET_NAME(tcp_insecure_rst), 0,
203 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
205 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
206 #define V_tcp_recvspace VNET(tcp_recvspace)
207 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
208 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
210 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
212 &VNET_NAME(tcp_do_autorcvbuf), 0,
213 "Enable automatic receive buffer sizing");
215 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW,
217 &VNET_NAME(tcp_autorcvbuf_inc), 0,
218 "Incrementor step size of automatic receive buffer");
220 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
222 &VNET_NAME(tcp_autorcvbuf_max), 0,
223 "Max size of automatic receive buffer");
225 VNET_DEFINE(struct inpcbhead, tcb);
226 #define tcb6 tcb /* for KAME src sync over BSD*'s */
227 VNET_DEFINE(struct inpcbinfo, tcbinfo);
230 * TCP statistics are stored in an array of counter(9)s, which size matches
231 * size of struct tcpstat. TCP running connection count is a regular array.
233 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
234 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
235 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
236 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
237 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
238 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
239 "TCP connection counts by TCP state");
242 tcp_vnet_init(const void *unused)
245 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
246 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
248 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
249 tcp_vnet_init, NULL);
253 tcp_vnet_uninit(const void *unused)
256 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
257 VNET_PCPUSTAT_FREE(tcpstat);
259 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
260 tcp_vnet_uninit, NULL);
264 * Kernel module interface for updating tcpstat. The argument is an index
265 * into tcpstat treated as an array.
268 kmod_tcpstat_inc(int statnum)
271 counter_u64_add(VNET(tcpstat)[statnum], 1);
276 * Wrapper for the TCP established input helper hook.
279 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
281 struct tcp_hhook_data hhook_data;
283 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
288 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
295 * CC wrapper hook functions
298 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
301 INP_WLOCK_ASSERT(tp->t_inpcb);
303 tp->ccv->nsegs = nsegs;
304 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
305 if (tp->snd_cwnd <= tp->snd_wnd)
306 tp->ccv->flags |= CCF_CWND_LIMITED;
308 tp->ccv->flags &= ~CCF_CWND_LIMITED;
310 if (type == CC_ACK) {
311 if (tp->snd_cwnd > tp->snd_ssthresh) {
312 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
313 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp));
314 if (tp->t_bytes_acked >= tp->snd_cwnd) {
315 tp->t_bytes_acked -= tp->snd_cwnd;
316 tp->ccv->flags |= CCF_ABC_SENTAWND;
319 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
320 tp->t_bytes_acked = 0;
324 if (CC_ALGO(tp)->ack_received != NULL) {
325 /* XXXLAS: Find a way to live without this */
326 tp->ccv->curack = th->th_ack;
327 CC_ALGO(tp)->ack_received(tp->ccv, type);
332 cc_conn_init(struct tcpcb *tp)
334 struct hc_metrics_lite metrics;
335 struct inpcb *inp = tp->t_inpcb;
339 INP_WLOCK_ASSERT(tp->t_inpcb);
341 tcp_hc_get(&inp->inp_inc, &metrics);
342 maxseg = tcp_maxseg(tp);
344 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
346 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
347 TCPSTAT_INC(tcps_usedrtt);
348 if (metrics.rmx_rttvar) {
349 tp->t_rttvar = metrics.rmx_rttvar;
350 TCPSTAT_INC(tcps_usedrttvar);
352 /* default variation is +- 1 rtt */
354 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
356 TCPT_RANGESET(tp->t_rxtcur,
357 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
358 tp->t_rttmin, TCPTV_REXMTMAX);
360 if (metrics.rmx_ssthresh) {
362 * There's some sort of gateway or interface
363 * buffer limit on the path. Use this to set
364 * the slow start threshold, but set the
365 * threshold to no less than 2*mss.
367 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
368 TCPSTAT_INC(tcps_usedssthresh);
372 * Set the initial slow-start flight size.
374 * 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)) {
421 TCPSTAT_INC(tcps_ecn_rcwnd);
422 tp->snd_recover = tp->snd_max;
423 if (tp->t_flags & TF_ECN_PERMIT)
424 tp->t_flags |= TF_ECN_SND_CWR;
428 maxseg = tcp_maxseg(tp);
430 tp->t_bytes_acked = 0;
431 EXIT_RECOVERY(tp->t_flags);
432 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
434 tp->snd_cwnd = maxseg;
437 TCPSTAT_INC(tcps_sndrexmitbad);
438 /* RTO was unnecessary, so reset everything. */
439 tp->snd_cwnd = tp->snd_cwnd_prev;
440 tp->snd_ssthresh = tp->snd_ssthresh_prev;
441 tp->snd_recover = tp->snd_recover_prev;
442 if (tp->t_flags & TF_WASFRECOVERY)
443 ENTER_FASTRECOVERY(tp->t_flags);
444 if (tp->t_flags & TF_WASCRECOVERY)
445 ENTER_CONGRECOVERY(tp->t_flags);
446 tp->snd_nxt = tp->snd_max;
447 tp->t_flags &= ~TF_PREVVALID;
452 if (CC_ALGO(tp)->cong_signal != NULL) {
454 tp->ccv->curack = th->th_ack;
455 CC_ALGO(tp)->cong_signal(tp->ccv, type);
460 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
462 INP_WLOCK_ASSERT(tp->t_inpcb);
464 /* XXXLAS: KASSERT that we're in recovery? */
466 if (CC_ALGO(tp)->post_recovery != NULL) {
467 tp->ccv->curack = th->th_ack;
468 CC_ALGO(tp)->post_recovery(tp->ccv);
470 /* XXXLAS: EXIT_RECOVERY ? */
471 tp->t_bytes_acked = 0;
475 * Indicate whether this ack should be delayed. We can delay the ack if
476 * following conditions are met:
477 * - There is no delayed ack timer in progress.
478 * - Our last ack wasn't a 0-sized window. We never want to delay
479 * the ack that opens up a 0-sized window.
480 * - LRO wasn't used for this segment. We make sure by checking that the
481 * segment size is not larger than the MSS.
483 #define DELAY_ACK(tp, tlen) \
484 ((!tcp_timer_active(tp, TT_DELACK) && \
485 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
486 (tlen <= tp->t_maxseg) && \
487 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
490 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
492 INP_WLOCK_ASSERT(tp->t_inpcb);
494 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
495 switch (iptos & IPTOS_ECN_MASK) {
497 tp->ccv->flags |= CCF_IPHDR_CE;
500 tp->ccv->flags &= ~CCF_IPHDR_CE;
503 tp->ccv->flags &= ~CCF_IPHDR_CE;
507 if (th->th_flags & TH_CWR)
508 tp->ccv->flags |= CCF_TCPHDR_CWR;
510 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
512 if (tp->t_flags & TF_DELACK)
513 tp->ccv->flags |= CCF_DELACK;
515 tp->ccv->flags &= ~CCF_DELACK;
517 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
519 if (tp->ccv->flags & CCF_ACKNOW)
520 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
525 * TCP input handling is split into multiple parts:
526 * tcp6_input is a thin wrapper around tcp_input for the extended
527 * ip6_protox[] call format in ip6_input
528 * tcp_input handles primary segment validation, inpcb lookup and
529 * SYN processing on listen sockets
530 * tcp_do_segment processes the ACK and text of the segment for
531 * establishing, established and closing connections
535 tcp6_input(struct mbuf **mp, int *offp, int proto)
537 struct mbuf *m = *mp;
538 struct in6_ifaddr *ia6;
541 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
544 * draft-itojun-ipv6-tcp-to-anycast
545 * better place to put this in?
547 ip6 = mtod(m, struct ip6_hdr *);
548 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
549 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
552 ifa_free(&ia6->ia_ifa);
553 ip6 = mtod(m, struct ip6_hdr *);
554 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
555 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
556 return (IPPROTO_DONE);
559 ifa_free(&ia6->ia_ifa);
561 return (tcp_input(mp, offp, proto));
566 tcp_input(struct mbuf **mp, int *offp, int proto)
568 struct mbuf *m = *mp;
569 struct tcphdr *th = NULL;
570 struct ip *ip = NULL;
571 struct inpcb *inp = NULL;
572 struct tcpcb *tp = NULL;
573 struct socket *so = NULL;
583 int rstreason = 0; /* For badport_bandlim accounting purposes */
585 struct m_tag *fwd_tag = NULL;
586 struct epoch_tracker et;
588 struct ip6_hdr *ip6 = NULL;
591 const void *ip6 = NULL;
593 struct tcpopt to; /* options in this segment */
594 char *s = NULL; /* address and port logging */
598 * The size of tcp_saveipgen must be the size of the max ip header,
601 u_char tcp_saveipgen[IP6_HDR_LEN];
602 struct tcphdr tcp_savetcp;
607 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
614 TCPSTAT_INC(tcps_rcvtotal);
618 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
620 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
621 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
623 TCPSTAT_INC(tcps_rcvshort);
624 return (IPPROTO_DONE);
628 ip6 = mtod(m, struct ip6_hdr *);
629 th = (struct tcphdr *)((caddr_t)ip6 + off0);
630 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
631 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
632 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
633 th->th_sum = m->m_pkthdr.csum_data;
635 th->th_sum = in6_cksum_pseudo(ip6, tlen,
636 IPPROTO_TCP, m->m_pkthdr.csum_data);
637 th->th_sum ^= 0xffff;
639 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
641 TCPSTAT_INC(tcps_rcvbadsum);
646 * Be proactive about unspecified IPv6 address in source.
647 * As we use all-zero to indicate unbounded/unconnected pcb,
648 * unspecified IPv6 address can be used to confuse us.
650 * Note that packets with unspecified IPv6 destination is
651 * already dropped in ip6_input.
653 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
657 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
660 #if defined(INET) && defined(INET6)
666 * Get IP and TCP header together in first mbuf.
667 * Note: IP leaves IP header in first mbuf.
669 if (off0 > sizeof (struct ip)) {
671 off0 = sizeof(struct ip);
673 if (m->m_len < sizeof (struct tcpiphdr)) {
674 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
676 TCPSTAT_INC(tcps_rcvshort);
677 return (IPPROTO_DONE);
680 ip = mtod(m, struct ip *);
681 th = (struct tcphdr *)((caddr_t)ip + off0);
682 tlen = ntohs(ip->ip_len) - off0;
685 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
686 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
687 th->th_sum = m->m_pkthdr.csum_data;
689 th->th_sum = in_pseudo(ip->ip_src.s_addr,
691 htonl(m->m_pkthdr.csum_data + tlen +
693 th->th_sum ^= 0xffff;
695 struct ipovly *ipov = (struct ipovly *)ip;
698 * Checksum extended TCP header and data.
701 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
702 ipov->ih_len = htons(tlen);
703 th->th_sum = in_cksum(m, len);
704 /* Reset length for SDT probes. */
705 ip->ip_len = htons(len);
708 /* Re-initialization for later version check */
709 ip->ip_v = IPVERSION;
710 ip->ip_hl = off0 >> 2;
714 TCPSTAT_INC(tcps_rcvbadsum);
721 * Check that TCP offset makes sense,
722 * pull out TCP options and adjust length. XXX
724 off = th->th_off << 2;
725 if (off < sizeof (struct tcphdr) || off > tlen) {
726 TCPSTAT_INC(tcps_rcvbadoff);
729 tlen -= off; /* tlen is used instead of ti->ti_len */
730 if (off > sizeof (struct tcphdr)) {
733 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
734 ip6 = mtod(m, struct ip6_hdr *);
735 th = (struct tcphdr *)((caddr_t)ip6 + off0);
738 #if defined(INET) && defined(INET6)
743 if (m->m_len < sizeof(struct ip) + off) {
744 if ((m = m_pullup(m, sizeof (struct ip) + off))
746 TCPSTAT_INC(tcps_rcvshort);
747 return (IPPROTO_DONE);
749 ip = mtod(m, struct ip *);
750 th = (struct tcphdr *)((caddr_t)ip + off0);
754 optlen = off - sizeof (struct tcphdr);
755 optp = (u_char *)(th + 1);
757 thflags = th->th_flags;
760 * Convert TCP protocol specific fields to host format.
762 tcp_fields_to_host(th);
765 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
767 drop_hdrlen = off0 + off;
770 * Locate pcb for segment; if we're likely to add or remove a
771 * connection then first acquire pcbinfo lock. There are three cases
772 * where we might discover later we need a write lock despite the
773 * flags: ACKs moving a connection out of the syncache, ACKs for a
774 * connection in TIMEWAIT and SYNs not targeting a listening socket.
776 if ((thflags & (TH_FIN | TH_RST)) != 0) {
777 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
778 ti_locked = TI_RLOCKED;
780 ti_locked = TI_UNLOCKED;
783 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
787 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
789 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
792 #if defined(INET) && !defined(INET6)
793 (m->m_flags & M_IP_NEXTHOP)
796 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
800 if (ti_locked == TI_RLOCKED) {
801 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
803 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
807 if (isipv6 && fwd_tag != NULL) {
808 struct sockaddr_in6 *next_hop6;
810 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
812 * Transparently forwarded. Pretend to be the destination.
813 * Already got one like this?
815 inp = in6_pcblookup_mbuf(&V_tcbinfo,
816 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
817 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
820 * It's new. Try to find the ambushing socket.
821 * Because we've rewritten the destination address,
822 * any hardware-generated hash is ignored.
824 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
825 th->th_sport, &next_hop6->sin6_addr,
826 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
827 th->th_dport, INPLOOKUP_WILDCARD |
828 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
831 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
832 th->th_sport, &ip6->ip6_dst, th->th_dport,
833 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
834 m->m_pkthdr.rcvif, m);
837 #if defined(INET6) && defined(INET)
841 if (fwd_tag != NULL) {
842 struct sockaddr_in *next_hop;
844 next_hop = (struct sockaddr_in *)(fwd_tag+1);
846 * Transparently forwarded. Pretend to be the destination.
847 * already got one like this?
849 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
850 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
851 m->m_pkthdr.rcvif, m);
854 * It's new. Try to find the ambushing socket.
855 * Because we've rewritten the destination address,
856 * any hardware-generated hash is ignored.
858 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
859 th->th_sport, next_hop->sin_addr,
860 next_hop->sin_port ? ntohs(next_hop->sin_port) :
861 th->th_dport, INPLOOKUP_WILDCARD |
862 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
865 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
866 th->th_sport, ip->ip_dst, th->th_dport,
867 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
868 m->m_pkthdr.rcvif, m);
872 * If the INPCB does not exist then all data in the incoming
873 * segment is discarded and an appropriate RST is sent back.
874 * XXX MRT Send RST using which routing table?
878 * Log communication attempts to ports that are not
881 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
882 tcp_log_in_vain == 2) {
883 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
884 log(LOG_INFO, "%s; %s: Connection attempt "
885 "to closed port\n", s, __func__);
888 * When blackholing do not respond with a RST but
889 * completely ignore the segment and drop it.
891 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
895 rstreason = BANDLIM_RST_CLOSEDPORT;
898 INP_WLOCK_ASSERT(inp);
900 * While waiting for inp lock during the lookup, another thread
901 * can have dropped the inpcb, in which case we need to loop back
902 * and try to find a new inpcb to deliver to.
904 if (inp->inp_flags & INP_DROPPED) {
909 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
910 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
911 ((inp->inp_socket == NULL) ||
912 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
913 inp->inp_flowid = m->m_pkthdr.flowid;
914 inp->inp_flowtype = M_HASHTYPE_GET(m);
916 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
918 if (isipv6 && IPSEC_ENABLED(ipv6) &&
919 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
927 if (IPSEC_ENABLED(ipv4) &&
928 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
935 * Check the minimum TTL for socket.
937 if (inp->inp_ip_minttl != 0) {
940 if (inp->inp_ip_minttl > ip6->ip6_hlim)
944 if (inp->inp_ip_minttl > ip->ip_ttl)
949 * A previous connection in TIMEWAIT state is supposed to catch stray
950 * or duplicate segments arriving late. If this segment was a
951 * legitimate new connection attempt, the old INPCB gets removed and
952 * we can try again to find a listening socket.
954 * At this point, due to earlier optimism, we may hold only an inpcb
955 * lock, and not the inpcbinfo write lock. If so, we need to try to
956 * acquire it, or if that fails, acquire a reference on the inpcb,
957 * drop all locks, acquire a global write lock, and then re-acquire
958 * the inpcb lock. We may at that point discover that another thread
959 * has tried to free the inpcb, in which case we need to loop back
960 * and try to find a new inpcb to deliver to.
962 * XXXRW: It may be time to rethink timewait locking.
964 if (inp->inp_flags & INP_TIMEWAIT) {
965 if (ti_locked == TI_UNLOCKED) {
966 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
967 ti_locked = TI_RLOCKED;
969 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
971 if (thflags & TH_SYN)
972 tcp_dooptions(&to, optp, optlen, TO_SYN);
974 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
976 if (tcp_twcheck(inp, &to, th, m, tlen))
978 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
979 return (IPPROTO_DONE);
982 * The TCPCB may no longer exist if the connection is winding
983 * down or it is in the CLOSED state. Either way we drop the
984 * segment and send an appropriate response.
987 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
988 rstreason = BANDLIM_RST_CLOSEDPORT;
993 if (tp->t_flags & TF_TOE) {
994 tcp_offload_input(tp, m);
995 m = NULL; /* consumed by the TOE driver */
1001 * We've identified a valid inpcb, but it could be that we need an
1002 * inpcbinfo write lock but don't hold it. In this case, attempt to
1003 * acquire using the same strategy as the TIMEWAIT case above. If we
1004 * relock, we have to jump back to 'relocked' as the connection might
1005 * now be in TIMEWAIT.
1008 if ((thflags & (TH_FIN | TH_RST)) != 0)
1009 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1011 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) ||
1012 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) &&
1013 !IS_FASTOPEN(tp->t_flags)))) {
1014 if (ti_locked == TI_UNLOCKED) {
1015 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1016 ti_locked = TI_RLOCKED;
1018 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1022 INP_WLOCK_ASSERT(inp);
1023 if (mac_inpcb_check_deliver(inp, m))
1026 so = inp->inp_socket;
1027 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1029 if (so->so_options & SO_DEBUG) {
1030 ostate = tp->t_state;
1033 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1036 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1039 #endif /* TCPDEBUG */
1041 * When the socket is accepting connections (the INPCB is in LISTEN
1042 * state) we look into the SYN cache if this is a new connection
1043 * attempt or the completion of a previous one.
1045 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1046 ("%s: so accepting but tp %p not listening", __func__, tp));
1047 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1048 struct in_conninfo inc;
1050 bzero(&inc, sizeof(inc));
1053 inc.inc_flags |= INC_ISIPV6;
1054 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1055 inc.inc_flags |= INC_IPV6MINMTU;
1056 inc.inc6_faddr = ip6->ip6_src;
1057 inc.inc6_laddr = ip6->ip6_dst;
1061 inc.inc_faddr = ip->ip_src;
1062 inc.inc_laddr = ip->ip_dst;
1064 inc.inc_fport = th->th_sport;
1065 inc.inc_lport = th->th_dport;
1066 inc.inc_fibnum = so->so_fibnum;
1069 * Check for an existing connection attempt in syncache if
1070 * the flag is only ACK. A successful lookup creates a new
1071 * socket appended to the listen queue in SYN_RECEIVED state.
1073 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1075 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1077 * Parse the TCP options here because
1078 * syncookies need access to the reflected
1081 tcp_dooptions(&to, optp, optlen, 0);
1083 * NB: syncache_expand() doesn't unlock
1084 * inp and tcpinfo locks.
1086 rstreason = syncache_expand(&inc, &to, th, &so, m);
1087 if (rstreason < 0) {
1089 * A failing TCP MD5 signature comparison
1090 * must result in the segment being dropped
1091 * and must not produce any response back
1095 } else if (rstreason == 0) {
1097 * No syncache entry or ACK was not
1098 * for our SYN/ACK. Send a RST.
1099 * NB: syncache did its own logging
1100 * of the failure cause.
1102 rstreason = BANDLIM_RST_OPENPORT;
1108 * We completed the 3-way handshake
1109 * but could not allocate a socket
1110 * either due to memory shortage,
1111 * listen queue length limits or
1112 * global socket limits. Send RST
1113 * or wait and have the remote end
1114 * retransmit the ACK for another
1117 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1118 log(LOG_DEBUG, "%s; %s: Listen socket: "
1119 "Socket allocation failed due to "
1120 "limits or memory shortage, %s\n",
1122 V_tcp_sc_rst_sock_fail ?
1123 "sending RST" : "try again");
1124 if (V_tcp_sc_rst_sock_fail) {
1125 rstreason = BANDLIM_UNLIMITED;
1131 * Socket is created in state SYN_RECEIVED.
1132 * Unlock the listen socket, lock the newly
1133 * created socket and update the tp variable.
1135 INP_WUNLOCK(inp); /* listen socket */
1136 inp = sotoinpcb(so);
1138 * New connection inpcb is already locked by
1139 * syncache_expand().
1141 INP_WLOCK_ASSERT(inp);
1142 tp = intotcpcb(inp);
1143 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1144 ("%s: ", __func__));
1146 * Process the segment and the data it
1147 * contains. tcp_do_segment() consumes
1148 * the mbuf chain and unlocks the inpcb.
1150 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1151 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1153 if (ti_locked == TI_RLOCKED)
1154 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1155 return (IPPROTO_DONE);
1158 * Segment flag validation for new connection attempts:
1160 * Our (SYN|ACK) response was rejected.
1161 * Check with syncache and remove entry to prevent
1164 * NB: syncache_chkrst does its own logging of failure
1167 if (thflags & TH_RST) {
1168 syncache_chkrst(&inc, th, m);
1172 * We can't do anything without SYN.
1174 if ((thflags & TH_SYN) == 0) {
1175 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1176 log(LOG_DEBUG, "%s; %s: Listen socket: "
1177 "SYN is missing, segment ignored\n",
1179 TCPSTAT_INC(tcps_badsyn);
1183 * (SYN|ACK) is bogus on a listen socket.
1185 if (thflags & TH_ACK) {
1186 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1187 log(LOG_DEBUG, "%s; %s: Listen socket: "
1188 "SYN|ACK invalid, segment rejected\n",
1190 syncache_badack(&inc); /* XXX: Not needed! */
1191 TCPSTAT_INC(tcps_badsyn);
1192 rstreason = BANDLIM_RST_OPENPORT;
1196 * If the drop_synfin option is enabled, drop all
1197 * segments with both the SYN and FIN bits set.
1198 * This prevents e.g. nmap from identifying the
1200 * XXX: Poor reasoning. nmap has other methods
1201 * and is constantly refining its stack detection
1203 * XXX: This is a violation of the TCP specification
1204 * and was used by RFC1644.
1206 if ((thflags & TH_FIN) && V_drop_synfin) {
1207 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1208 log(LOG_DEBUG, "%s; %s: Listen socket: "
1209 "SYN|FIN segment ignored (based on "
1210 "sysctl setting)\n", s, __func__);
1211 TCPSTAT_INC(tcps_badsyn);
1215 * Segment's flags are (SYN) or (SYN|FIN).
1217 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1218 * as they do not affect the state of the TCP FSM.
1219 * The data pointed to by TH_URG and th_urp is ignored.
1221 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1222 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1223 KASSERT(thflags & (TH_SYN),
1224 ("%s: Listen socket: TH_SYN not set", __func__));
1227 * If deprecated address is forbidden,
1228 * we do not accept SYN to deprecated interface
1229 * address to prevent any new inbound connection from
1230 * getting established.
1231 * When we do not accept SYN, we send a TCP RST,
1232 * with deprecated source address (instead of dropping
1233 * it). We compromise it as it is much better for peer
1234 * to send a RST, and RST will be the final packet
1237 * If we do not forbid deprecated addresses, we accept
1238 * the SYN packet. RFC2462 does not suggest dropping
1240 * If we decipher RFC2462 5.5.4, it says like this:
1241 * 1. use of deprecated addr with existing
1242 * communication is okay - "SHOULD continue to be
1244 * 2. use of it with new communication:
1245 * (2a) "SHOULD NOT be used if alternate address
1246 * with sufficient scope is available"
1247 * (2b) nothing mentioned otherwise.
1248 * Here we fall into (2b) case as we have no choice in
1249 * our source address selection - we must obey the peer.
1251 * The wording in RFC2462 is confusing, and there are
1252 * multiple description text for deprecated address
1253 * handling - worse, they are not exactly the same.
1254 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1256 if (isipv6 && !V_ip6_use_deprecated) {
1257 struct in6_ifaddr *ia6;
1259 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1261 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1262 ifa_free(&ia6->ia_ifa);
1263 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1264 log(LOG_DEBUG, "%s; %s: Listen socket: "
1265 "Connection attempt to deprecated "
1266 "IPv6 address rejected\n",
1268 rstreason = BANDLIM_RST_OPENPORT;
1272 ifa_free(&ia6->ia_ifa);
1276 * Basic sanity checks on incoming SYN requests:
1277 * Don't respond if the destination is a link layer
1278 * broadcast according to RFC1122 4.2.3.10, p. 104.
1279 * If it is from this socket it must be forged.
1280 * Don't respond if the source or destination is a
1281 * global or subnet broad- or multicast address.
1282 * Note that it is quite possible to receive unicast
1283 * link-layer packets with a broadcast IP address. Use
1284 * in_broadcast() to find them.
1286 if (m->m_flags & (M_BCAST|M_MCAST)) {
1287 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1288 log(LOG_DEBUG, "%s; %s: Listen socket: "
1289 "Connection attempt from broad- or multicast "
1290 "link layer address ignored\n", s, __func__);
1295 if (th->th_dport == th->th_sport &&
1296 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1297 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1298 log(LOG_DEBUG, "%s; %s: Listen socket: "
1299 "Connection attempt to/from self "
1300 "ignored\n", s, __func__);
1303 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1304 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1305 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1306 log(LOG_DEBUG, "%s; %s: Listen socket: "
1307 "Connection attempt from/to multicast "
1308 "address ignored\n", s, __func__);
1313 #if defined(INET) && defined(INET6)
1318 if (th->th_dport == th->th_sport &&
1319 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1320 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1321 log(LOG_DEBUG, "%s; %s: Listen socket: "
1322 "Connection attempt from/to self "
1323 "ignored\n", s, __func__);
1326 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1327 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1328 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1329 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1330 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1331 log(LOG_DEBUG, "%s; %s: Listen socket: "
1332 "Connection attempt from/to broad- "
1333 "or multicast address ignored\n",
1340 * SYN appears to be valid. Create compressed TCP state
1344 if (so->so_options & SO_DEBUG)
1345 tcp_trace(TA_INPUT, ostate, tp,
1346 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1348 TCP_PROBE3(debug__input, tp, th, m);
1349 tcp_dooptions(&to, optp, optlen, TO_SYN);
1350 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL))
1351 goto tfo_socket_result;
1354 * Entry added to syncache and mbuf consumed.
1355 * Only the listen socket is unlocked by syncache_add().
1357 if (ti_locked == TI_RLOCKED) {
1358 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1359 ti_locked = TI_UNLOCKED;
1361 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1362 return (IPPROTO_DONE);
1363 } else if (tp->t_state == TCPS_LISTEN) {
1365 * When a listen socket is torn down the SO_ACCEPTCONN
1366 * flag is removed first while connections are drained
1367 * from the accept queue in a unlock/lock cycle of the
1368 * ACCEPT_LOCK, opening a race condition allowing a SYN
1369 * attempt go through unhandled.
1373 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1374 if (tp->t_flags & TF_SIGNATURE) {
1375 tcp_dooptions(&to, optp, optlen, thflags);
1376 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1377 TCPSTAT_INC(tcps_sig_err_nosigopt);
1380 if (!TCPMD5_ENABLED() ||
1381 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1385 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1388 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1389 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1390 * the inpcb, and unlocks pcbinfo.
1392 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1393 if (ti_locked == TI_RLOCKED)
1394 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1395 return (IPPROTO_DONE);
1398 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1400 if (ti_locked == TI_RLOCKED) {
1401 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1402 ti_locked = TI_UNLOCKED;
1406 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1407 "ti_locked: %d", __func__, ti_locked));
1408 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1413 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1416 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1417 m = NULL; /* mbuf chain got consumed. */
1422 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1424 if (ti_locked == TI_RLOCKED) {
1425 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1426 ti_locked = TI_UNLOCKED;
1430 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1431 "ti_locked: %d", __func__, ti_locked));
1432 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1440 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1445 return (IPPROTO_DONE);
1449 * Automatic sizing of receive socket buffer. Often the send
1450 * buffer size is not optimally adjusted to the actual network
1451 * conditions at hand (delay bandwidth product). Setting the
1452 * buffer size too small limits throughput on links with high
1453 * bandwidth and high delay (eg. trans-continental/oceanic links).
1455 * On the receive side the socket buffer memory is only rarely
1456 * used to any significant extent. This allows us to be much
1457 * more aggressive in scaling the receive socket buffer. For
1458 * the case that the buffer space is actually used to a large
1459 * extent and we run out of kernel memory we can simply drop
1460 * the new segments; TCP on the sender will just retransmit it
1461 * later. Setting the buffer size too big may only consume too
1462 * much kernel memory if the application doesn't read() from
1463 * the socket or packet loss or reordering makes use of the
1466 * The criteria to step up the receive buffer one notch are:
1467 * 1. Application has not set receive buffer size with
1468 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1469 * 2. the number of bytes received during the time it takes
1470 * one timestamp to be reflected back to us (the RTT);
1471 * 3. received bytes per RTT is within seven eighth of the
1472 * current socket buffer size;
1473 * 4. receive buffer size has not hit maximal automatic size;
1475 * This algorithm does one step per RTT at most and only if
1476 * we receive a bulk stream w/o packet losses or reorderings.
1477 * Shrinking the buffer during idle times is not necessary as
1478 * it doesn't consume any memory when idle.
1480 * TODO: Only step up if the application is actually serving
1481 * the buffer to better manage the socket buffer resources.
1484 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1485 struct tcpcb *tp, int tlen)
1489 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1490 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1491 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1492 (tp->t_srtt >> TCP_RTT_SHIFT)) {
1493 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) &&
1494 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1495 newsize = min(so->so_rcv.sb_hiwat +
1496 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max);
1498 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1500 /* Start over with next RTT. */
1504 tp->rfbuf_cnt += tlen; /* add up */
1511 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1512 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1514 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1515 int rstreason, todrop, win;
1519 struct in_conninfo *inc;
1526 * The size of tcp_saveipgen must be the size of the max ip header,
1529 u_char tcp_saveipgen[IP6_HDR_LEN];
1530 struct tcphdr tcp_savetcp;
1533 thflags = th->th_flags;
1534 inc = &tp->t_inpcb->inp_inc;
1535 tp->sackhint.last_sack_ack = 0;
1537 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1539 * If this is either a state-changing packet or current state isn't
1540 * established, we require a write lock on tcbinfo. Otherwise, we
1541 * allow the tcbinfo to be in either alocked or unlocked, as the
1542 * caller may have unnecessarily acquired a write lock due to a race.
1544 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1545 tp->t_state != TCPS_ESTABLISHED) {
1546 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1548 INP_WLOCK_ASSERT(tp->t_inpcb);
1549 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1551 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1555 /* Save segment, if requested. */
1556 tcp_pcap_add(th, m, &(tp->t_inpkts));
1558 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1561 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1562 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1563 log(LOG_DEBUG, "%s; %s: "
1564 "SYN|FIN segment ignored (based on "
1565 "sysctl setting)\n", s, __func__);
1572 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1573 * check SEQ.ACK first.
1575 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1576 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1577 rstreason = BANDLIM_UNLIMITED;
1582 * Segment received on connection.
1583 * Reset idle time and keep-alive timer.
1584 * XXX: This should be done after segment
1585 * validation to ignore broken/spoofed segs.
1587 tp->t_rcvtime = ticks;
1590 * Scale up the window into a 32-bit value.
1591 * For the SYN_SENT state the scale is zero.
1593 tiwin = th->th_win << tp->snd_scale;
1596 * TCP ECN processing.
1598 if (tp->t_flags & TF_ECN_PERMIT) {
1599 if (thflags & TH_CWR)
1600 tp->t_flags &= ~TF_ECN_SND_ECE;
1601 switch (iptos & IPTOS_ECN_MASK) {
1603 tp->t_flags |= TF_ECN_SND_ECE;
1604 TCPSTAT_INC(tcps_ecn_ce);
1606 case IPTOS_ECN_ECT0:
1607 TCPSTAT_INC(tcps_ecn_ect0);
1609 case IPTOS_ECN_ECT1:
1610 TCPSTAT_INC(tcps_ecn_ect1);
1614 /* Process a packet differently from RFC3168. */
1615 cc_ecnpkt_handler(tp, th, iptos);
1617 /* Congestion experienced. */
1618 if (thflags & TH_ECE) {
1619 cc_cong_signal(tp, th, CC_ECN);
1624 * Parse options on any incoming segment.
1626 tcp_dooptions(&to, (u_char *)(th + 1),
1627 (th->th_off << 2) - sizeof(struct tcphdr),
1628 (thflags & TH_SYN) ? TO_SYN : 0);
1630 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1631 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1632 (to.to_flags & TOF_SIGNATURE) == 0) {
1633 TCPSTAT_INC(tcps_sig_err_sigopt);
1634 /* XXX: should drop? */
1638 * If echoed timestamp is later than the current time,
1639 * fall back to non RFC1323 RTT calculation. Normalize
1640 * timestamp if syncookies were used when this connection
1643 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1644 to.to_tsecr -= tp->ts_offset;
1645 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1647 else if (tp->t_flags & TF_PREVVALID &&
1648 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1649 cc_cong_signal(tp, th, CC_RTO_ERR);
1652 * Process options only when we get SYN/ACK back. The SYN case
1653 * for incoming connections is handled in tcp_syncache.
1654 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1655 * or <SYN,ACK>) segment itself is never scaled.
1656 * XXX this is traditional behavior, may need to be cleaned up.
1658 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1659 if ((to.to_flags & TOF_SCALE) &&
1660 (tp->t_flags & TF_REQ_SCALE)) {
1661 tp->t_flags |= TF_RCVD_SCALE;
1662 tp->snd_scale = to.to_wscale;
1665 * Initial send window. It will be updated with
1666 * the next incoming segment to the scaled value.
1668 tp->snd_wnd = th->th_win;
1669 if (to.to_flags & TOF_TS) {
1670 tp->t_flags |= TF_RCVD_TSTMP;
1671 tp->ts_recent = to.to_tsval;
1672 tp->ts_recent_age = tcp_ts_getticks();
1674 if (to.to_flags & TOF_MSS)
1675 tcp_mss(tp, to.to_mss);
1676 if ((tp->t_flags & TF_SACK_PERMIT) &&
1677 (to.to_flags & TOF_SACKPERM) == 0)
1678 tp->t_flags &= ~TF_SACK_PERMIT;
1679 if (IS_FASTOPEN(tp->t_flags)) {
1680 if (to.to_flags & TOF_FASTOPEN) {
1683 if (to.to_flags & TOF_MSS)
1686 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1690 tcp_fastopen_update_cache(tp, mss,
1691 to.to_tfo_len, to.to_tfo_cookie);
1693 tcp_fastopen_disable_path(tp);
1698 * If timestamps were negotiated during SYN/ACK they should
1699 * appear on every segment during this session and vice versa.
1701 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1702 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1703 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1704 "no action\n", s, __func__);
1708 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1709 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1710 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1711 "no action\n", s, __func__);
1717 * Header prediction: check for the two common cases
1718 * of a uni-directional data xfer. If the packet has
1719 * no control flags, is in-sequence, the window didn't
1720 * change and we're not retransmitting, it's a
1721 * candidate. If the length is zero and the ack moved
1722 * forward, we're the sender side of the xfer. Just
1723 * free the data acked & wake any higher level process
1724 * that was blocked waiting for space. If the length
1725 * is non-zero and the ack didn't move, we're the
1726 * receiver side. If we're getting packets in-order
1727 * (the reassembly queue is empty), add the data to
1728 * the socket buffer and note that we need a delayed ack.
1729 * Make sure that the hidden state-flags are also off.
1730 * Since we check for TCPS_ESTABLISHED first, it can only
1733 if (tp->t_state == TCPS_ESTABLISHED &&
1734 th->th_seq == tp->rcv_nxt &&
1735 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1736 tp->snd_nxt == tp->snd_max &&
1737 tiwin && tiwin == tp->snd_wnd &&
1738 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1740 ((to.to_flags & TOF_TS) == 0 ||
1741 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1744 * If last ACK falls within this segment's sequence numbers,
1745 * record the timestamp.
1746 * NOTE that the test is modified according to the latest
1747 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1749 if ((to.to_flags & TOF_TS) != 0 &&
1750 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1751 tp->ts_recent_age = tcp_ts_getticks();
1752 tp->ts_recent = to.to_tsval;
1756 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1757 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1758 !IN_RECOVERY(tp->t_flags) &&
1759 (to.to_flags & TOF_SACK) == 0 &&
1760 TAILQ_EMPTY(&tp->snd_holes)) {
1762 * This is a pure ack for outstanding data.
1764 TCPSTAT_INC(tcps_predack);
1767 * "bad retransmit" recovery without timestamps.
1769 if ((to.to_flags & TOF_TS) == 0 &&
1770 tp->t_rxtshift == 1 &&
1771 tp->t_flags & TF_PREVVALID &&
1772 (int)(ticks - tp->t_badrxtwin) < 0) {
1773 cc_cong_signal(tp, th, CC_RTO_ERR);
1777 * Recalculate the transmit timer / rtt.
1779 * Some boxes send broken timestamp replies
1780 * during the SYN+ACK phase, ignore
1781 * timestamps of 0 or we could calculate a
1782 * huge RTT and blow up the retransmit timer.
1784 if ((to.to_flags & TOF_TS) != 0 &&
1788 t = tcp_ts_getticks() - to.to_tsecr;
1789 if (!tp->t_rttlow || tp->t_rttlow > t)
1792 TCP_TS_TO_TICKS(t) + 1);
1793 } else if (tp->t_rtttime &&
1794 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1795 if (!tp->t_rttlow ||
1796 tp->t_rttlow > ticks - tp->t_rtttime)
1797 tp->t_rttlow = ticks - tp->t_rtttime;
1799 ticks - tp->t_rtttime);
1801 acked = BYTES_THIS_ACK(tp, th);
1804 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1805 hhook_run_tcp_est_in(tp, th, &to);
1808 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1809 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1810 sbdrop(&so->so_snd, acked);
1811 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1812 SEQ_LEQ(th->th_ack, tp->snd_recover))
1813 tp->snd_recover = th->th_ack - 1;
1816 * Let the congestion control algorithm update
1817 * congestion control related information. This
1818 * typically means increasing the congestion
1821 cc_ack_received(tp, th, nsegs, CC_ACK);
1823 tp->snd_una = th->th_ack;
1825 * Pull snd_wl2 up to prevent seq wrap relative
1828 tp->snd_wl2 = th->th_ack;
1833 * If all outstanding data are acked, stop
1834 * retransmit timer, otherwise restart timer
1835 * using current (possibly backed-off) value.
1836 * If process is waiting for space,
1837 * wakeup/selwakeup/signal. If data
1838 * are ready to send, let tcp_output
1839 * decide between more output or persist.
1842 if (so->so_options & SO_DEBUG)
1843 tcp_trace(TA_INPUT, ostate, tp,
1844 (void *)tcp_saveipgen,
1847 TCP_PROBE3(debug__input, tp, th, m);
1848 if (tp->snd_una == tp->snd_max)
1849 tcp_timer_activate(tp, TT_REXMT, 0);
1850 else if (!tcp_timer_active(tp, TT_PERSIST))
1851 tcp_timer_activate(tp, TT_REXMT,
1854 if (sbavail(&so->so_snd))
1855 (void) tp->t_fb->tfb_tcp_output(tp);
1858 } else if (th->th_ack == tp->snd_una &&
1859 tlen <= sbspace(&so->so_rcv)) {
1860 int newsize = 0; /* automatic sockbuf scaling */
1863 * This is a pure, in-sequence data packet with
1864 * nothing on the reassembly queue and we have enough
1865 * buffer space to take it.
1867 /* Clean receiver SACK report if present */
1868 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1869 tcp_clean_sackreport(tp);
1870 TCPSTAT_INC(tcps_preddat);
1871 tp->rcv_nxt += tlen;
1873 * Pull snd_wl1 up to prevent seq wrap relative to
1876 tp->snd_wl1 = th->th_seq;
1878 * Pull rcv_up up to prevent seq wrap relative to
1881 tp->rcv_up = tp->rcv_nxt;
1882 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1883 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1885 if (so->so_options & SO_DEBUG)
1886 tcp_trace(TA_INPUT, ostate, tp,
1887 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1889 TCP_PROBE3(debug__input, tp, th, m);
1891 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1893 /* Add data to socket buffer. */
1894 SOCKBUF_LOCK(&so->so_rcv);
1895 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1899 * Set new socket buffer size.
1900 * Give up when limit is reached.
1903 if (!sbreserve_locked(&so->so_rcv,
1905 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1906 m_adj(m, drop_hdrlen); /* delayed header drop */
1907 sbappendstream_locked(&so->so_rcv, m, 0);
1909 /* NB: sorwakeup_locked() does an implicit unlock. */
1910 sorwakeup_locked(so);
1911 if (DELAY_ACK(tp, tlen)) {
1912 tp->t_flags |= TF_DELACK;
1914 tp->t_flags |= TF_ACKNOW;
1915 tp->t_fb->tfb_tcp_output(tp);
1922 * Calculate amount of space in receive window,
1923 * and then do TCP input processing.
1924 * Receive window is amount of space in rcv queue,
1925 * but not less than advertised window.
1927 win = sbspace(&so->so_rcv);
1930 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1932 switch (tp->t_state) {
1935 * If the state is SYN_RECEIVED:
1936 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1938 case TCPS_SYN_RECEIVED:
1939 if ((thflags & TH_ACK) &&
1940 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1941 SEQ_GT(th->th_ack, tp->snd_max))) {
1942 rstreason = BANDLIM_RST_OPENPORT;
1945 if (IS_FASTOPEN(tp->t_flags)) {
1947 * When a TFO connection is in SYN_RECEIVED, the
1948 * only valid packets are the initial SYN, a
1949 * retransmit/copy of the initial SYN (possibly with
1950 * a subset of the original data), a valid ACK, a
1953 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1954 rstreason = BANDLIM_RST_OPENPORT;
1956 } else if (thflags & TH_SYN) {
1957 /* non-initial SYN is ignored */
1958 if ((tcp_timer_active(tp, TT_DELACK) ||
1959 tcp_timer_active(tp, TT_REXMT)))
1961 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1968 * If the state is SYN_SENT:
1969 * if seg contains a RST with valid ACK (SEQ.ACK has already
1970 * been verified), then drop the connection.
1971 * if seg contains a RST without an ACK, drop the seg.
1972 * if seg does not contain SYN, then drop the seg.
1973 * Otherwise this is an acceptable SYN segment
1974 * initialize tp->rcv_nxt and tp->irs
1975 * if seg contains ack then advance tp->snd_una
1976 * if seg contains an ECE and ECN support is enabled, the stream
1978 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1979 * arrange for segment to be acked (eventually)
1980 * continue processing rest of data/controls, beginning with URG
1983 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1984 TCP_PROBE5(connect__refused, NULL, tp,
1986 tp = tcp_drop(tp, ECONNREFUSED);
1988 if (thflags & TH_RST)
1990 if (!(thflags & TH_SYN))
1993 tp->irs = th->th_seq;
1995 if (thflags & TH_ACK) {
1996 int tfo_partial_ack = 0;
1998 TCPSTAT_INC(tcps_connects);
2001 mac_socketpeer_set_from_mbuf(m, so);
2003 /* Do window scaling on this connection? */
2004 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2005 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2006 tp->rcv_scale = tp->request_r_scale;
2008 tp->rcv_adv += min(tp->rcv_wnd,
2009 TCP_MAXWIN << tp->rcv_scale);
2010 tp->snd_una++; /* SYN is acked */
2012 * If not all the data that was sent in the TFO SYN
2013 * has been acked, resend the remainder right away.
2015 if (IS_FASTOPEN(tp->t_flags) &&
2016 (tp->snd_una != tp->snd_max)) {
2017 tp->snd_nxt = th->th_ack;
2018 tfo_partial_ack = 1;
2021 * If there's data, delay ACK; if there's also a FIN
2022 * ACKNOW will be turned on later.
2024 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2025 tcp_timer_activate(tp, TT_DELACK,
2028 tp->t_flags |= TF_ACKNOW;
2030 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2032 tp->t_flags |= TF_ECN_PERMIT;
2033 TCPSTAT_INC(tcps_ecn_shs);
2037 * Received <SYN,ACK> in SYN_SENT[*] state.
2039 * SYN_SENT --> ESTABLISHED
2040 * SYN_SENT* --> FIN_WAIT_1
2042 tp->t_starttime = ticks;
2043 if (tp->t_flags & TF_NEEDFIN) {
2044 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2045 tp->t_flags &= ~TF_NEEDFIN;
2048 tcp_state_change(tp, TCPS_ESTABLISHED);
2049 TCP_PROBE5(connect__established, NULL, tp,
2052 tcp_timer_activate(tp, TT_KEEP,
2057 * Received initial SYN in SYN-SENT[*] state =>
2058 * simultaneous open.
2059 * If it succeeds, connection is * half-synchronized.
2060 * Otherwise, do 3-way handshake:
2061 * SYN-SENT -> SYN-RECEIVED
2062 * SYN-SENT* -> SYN-RECEIVED*
2064 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2065 tcp_timer_activate(tp, TT_REXMT, 0);
2066 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2069 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2070 INP_WLOCK_ASSERT(tp->t_inpcb);
2073 * Advance th->th_seq to correspond to first data byte.
2074 * If data, trim to stay within window,
2075 * dropping FIN if necessary.
2078 if (tlen > tp->rcv_wnd) {
2079 todrop = tlen - tp->rcv_wnd;
2083 TCPSTAT_INC(tcps_rcvpackafterwin);
2084 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2086 tp->snd_wl1 = th->th_seq - 1;
2087 tp->rcv_up = th->th_seq;
2089 * Client side of transaction: already sent SYN and data.
2090 * If the remote host used T/TCP to validate the SYN,
2091 * our data will be ACK'd; if so, enter normal data segment
2092 * processing in the middle of step 5, ack processing.
2093 * Otherwise, goto step 6.
2095 if (thflags & TH_ACK)
2101 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2102 * do normal processing.
2104 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2108 break; /* continue normal processing */
2112 * States other than LISTEN or SYN_SENT.
2113 * First check the RST flag and sequence number since reset segments
2114 * are exempt from the timestamp and connection count tests. This
2115 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2116 * below which allowed reset segments in half the sequence space
2117 * to fall though and be processed (which gives forged reset
2118 * segments with a random sequence number a 50 percent chance of
2119 * killing a connection).
2120 * Then check timestamp, if present.
2121 * Then check the connection count, if present.
2122 * Then check that at least some bytes of segment are within
2123 * receive window. If segment begins before rcv_nxt,
2124 * drop leading data (and SYN); if nothing left, just ack.
2126 if (thflags & TH_RST) {
2128 * RFC5961 Section 3.2
2130 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2131 * - If RST is in window, we send challenge ACK.
2133 * Note: to take into account delayed ACKs, we should
2134 * test against last_ack_sent instead of rcv_nxt.
2135 * Note 2: we handle special case of closed window, not
2136 * covered by the RFC.
2138 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2139 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2140 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2142 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2143 KASSERT(tp->t_state != TCPS_SYN_SENT,
2144 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2147 if (V_tcp_insecure_rst ||
2148 tp->last_ack_sent == th->th_seq) {
2149 TCPSTAT_INC(tcps_drops);
2150 /* Drop the connection. */
2151 switch (tp->t_state) {
2152 case TCPS_SYN_RECEIVED:
2153 so->so_error = ECONNREFUSED;
2155 case TCPS_ESTABLISHED:
2156 case TCPS_FIN_WAIT_1:
2157 case TCPS_FIN_WAIT_2:
2158 case TCPS_CLOSE_WAIT:
2161 so->so_error = ECONNRESET;
2168 TCPSTAT_INC(tcps_badrst);
2169 /* Send challenge ACK. */
2170 tcp_respond(tp, mtod(m, void *), th, m,
2171 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2172 tp->last_ack_sent = tp->rcv_nxt;
2180 * RFC5961 Section 4.2
2181 * Send challenge ACK for any SYN in synchronized state.
2183 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2184 tp->t_state != TCPS_SYN_RECEIVED) {
2185 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2187 TCPSTAT_INC(tcps_badsyn);
2188 if (V_tcp_insecure_syn &&
2189 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2190 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2191 tp = tcp_drop(tp, ECONNRESET);
2192 rstreason = BANDLIM_UNLIMITED;
2194 /* Send challenge ACK. */
2195 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2196 tp->snd_nxt, TH_ACK);
2197 tp->last_ack_sent = tp->rcv_nxt;
2204 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2205 * and it's less than ts_recent, drop it.
2207 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2208 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2210 /* Check to see if ts_recent is over 24 days old. */
2211 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2213 * Invalidate ts_recent. If this segment updates
2214 * ts_recent, the age will be reset later and ts_recent
2215 * will get a valid value. If it does not, setting
2216 * ts_recent to zero will at least satisfy the
2217 * requirement that zero be placed in the timestamp
2218 * echo reply when ts_recent isn't valid. The
2219 * age isn't reset until we get a valid ts_recent
2220 * because we don't want out-of-order segments to be
2221 * dropped when ts_recent is old.
2225 TCPSTAT_INC(tcps_rcvduppack);
2226 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2227 TCPSTAT_INC(tcps_pawsdrop);
2235 * In the SYN-RECEIVED state, validate that the packet belongs to
2236 * this connection before trimming the data to fit the receive
2237 * window. Check the sequence number versus IRS since we know
2238 * the sequence numbers haven't wrapped. This is a partial fix
2239 * for the "LAND" DoS attack.
2241 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2242 rstreason = BANDLIM_RST_OPENPORT;
2246 todrop = tp->rcv_nxt - th->th_seq;
2248 if (thflags & TH_SYN) {
2258 * Following if statement from Stevens, vol. 2, p. 960.
2261 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2263 * Any valid FIN must be to the left of the window.
2264 * At this point the FIN must be a duplicate or out
2265 * of sequence; drop it.
2270 * Send an ACK to resynchronize and drop any data.
2271 * But keep on processing for RST or ACK.
2273 tp->t_flags |= TF_ACKNOW;
2275 TCPSTAT_INC(tcps_rcvduppack);
2276 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2278 TCPSTAT_INC(tcps_rcvpartduppack);
2279 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2282 * DSACK - add SACK block for dropped range
2284 if (tp->t_flags & TF_SACK_PERMIT) {
2285 tcp_update_sack_list(tp, th->th_seq, th->th_seq+tlen);
2287 * ACK now, as the next in-sequence segment
2288 * will clear the DSACK block again
2290 tp->t_flags |= TF_ACKNOW;
2292 drop_hdrlen += todrop; /* drop from the top afterwards */
2293 th->th_seq += todrop;
2295 if (th->th_urp > todrop)
2296 th->th_urp -= todrop;
2304 * If new data are received on a connection after the
2305 * user processes are gone, then RST the other end.
2307 if ((so->so_state & SS_NOFDREF) &&
2308 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2309 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2311 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2312 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2313 "after socket was closed, "
2314 "sending RST and removing tcpcb\n",
2315 s, __func__, tcpstates[tp->t_state], tlen);
2319 TCPSTAT_INC(tcps_rcvafterclose);
2320 rstreason = BANDLIM_UNLIMITED;
2325 * If segment ends after window, drop trailing data
2326 * (and PUSH and FIN); if nothing left, just ACK.
2328 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2330 TCPSTAT_INC(tcps_rcvpackafterwin);
2331 if (todrop >= tlen) {
2332 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2334 * If window is closed can only take segments at
2335 * window edge, and have to drop data and PUSH from
2336 * incoming segments. Continue processing, but
2337 * remember to ack. Otherwise, drop segment
2340 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2341 tp->t_flags |= TF_ACKNOW;
2342 TCPSTAT_INC(tcps_rcvwinprobe);
2346 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2349 thflags &= ~(TH_PUSH|TH_FIN);
2353 * If last ACK falls within this segment's sequence numbers,
2354 * record its timestamp.
2356 * 1) That the test incorporates suggestions from the latest
2357 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2358 * 2) That updating only on newer timestamps interferes with
2359 * our earlier PAWS tests, so this check should be solely
2360 * predicated on the sequence space of this segment.
2361 * 3) That we modify the segment boundary check to be
2362 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2363 * instead of RFC1323's
2364 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2365 * This modified check allows us to overcome RFC1323's
2366 * limitations as described in Stevens TCP/IP Illustrated
2367 * Vol. 2 p.869. In such cases, we can still calculate the
2368 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2370 if ((to.to_flags & TOF_TS) != 0 &&
2371 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2372 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2373 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2374 tp->ts_recent_age = tcp_ts_getticks();
2375 tp->ts_recent = to.to_tsval;
2379 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2380 * flag is on (half-synchronized state), then queue data for
2381 * later processing; else drop segment and return.
2383 if ((thflags & TH_ACK) == 0) {
2384 if (tp->t_state == TCPS_SYN_RECEIVED ||
2385 (tp->t_flags & TF_NEEDSYN)) {
2386 if (tp->t_state == TCPS_SYN_RECEIVED &&
2387 IS_FASTOPEN(tp->t_flags)) {
2388 tp->snd_wnd = tiwin;
2392 } else if (tp->t_flags & TF_ACKNOW)
2401 switch (tp->t_state) {
2404 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2405 * ESTABLISHED state and continue processing.
2406 * The ACK was checked above.
2408 case TCPS_SYN_RECEIVED:
2410 TCPSTAT_INC(tcps_connects);
2412 /* Do window scaling? */
2413 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2414 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2415 tp->rcv_scale = tp->request_r_scale;
2417 tp->snd_wnd = tiwin;
2420 * SYN-RECEIVED -> ESTABLISHED
2421 * SYN-RECEIVED* -> FIN-WAIT-1
2423 tp->t_starttime = ticks;
2424 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2425 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2426 tp->t_tfo_pending = NULL;
2429 * Account for the ACK of our SYN prior to
2430 * regular ACK processing below.
2434 if (tp->t_flags & TF_NEEDFIN) {
2435 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2436 tp->t_flags &= ~TF_NEEDFIN;
2438 tcp_state_change(tp, TCPS_ESTABLISHED);
2439 TCP_PROBE5(accept__established, NULL, tp,
2442 * TFO connections call cc_conn_init() during SYN
2443 * processing. Calling it again here for such
2444 * connections is not harmless as it would undo the
2445 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2448 if (!IS_FASTOPEN(tp->t_flags))
2450 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2453 * If segment contains data or ACK, will call tcp_reass()
2454 * later; if not, do so now to pass queued data to user.
2456 if (tlen == 0 && (thflags & TH_FIN) == 0)
2457 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2459 tp->snd_wl1 = th->th_seq - 1;
2463 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2464 * ACKs. If the ack is in the range
2465 * tp->snd_una < th->th_ack <= tp->snd_max
2466 * then advance tp->snd_una to th->th_ack and drop
2467 * data from the retransmission queue. If this ACK reflects
2468 * more up to date window information we update our window information.
2470 case TCPS_ESTABLISHED:
2471 case TCPS_FIN_WAIT_1:
2472 case TCPS_FIN_WAIT_2:
2473 case TCPS_CLOSE_WAIT:
2476 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2477 TCPSTAT_INC(tcps_rcvacktoomuch);
2480 if ((tp->t_flags & TF_SACK_PERMIT) &&
2481 ((to.to_flags & TOF_SACK) ||
2482 !TAILQ_EMPTY(&tp->snd_holes)))
2483 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2486 * Reset the value so that previous (valid) value
2487 * from the last ack with SACK doesn't get used.
2489 tp->sackhint.sacked_bytes = 0;
2492 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2493 hhook_run_tcp_est_in(tp, th, &to);
2496 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2499 maxseg = tcp_maxseg(tp);
2501 (tiwin == tp->snd_wnd ||
2502 (tp->t_flags & TF_SACK_PERMIT))) {
2504 * If this is the first time we've seen a
2505 * FIN from the remote, this is not a
2506 * duplicate and it needs to be processed
2507 * normally. This happens during a
2508 * simultaneous close.
2510 if ((thflags & TH_FIN) &&
2511 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2515 TCPSTAT_INC(tcps_rcvdupack);
2517 * If we have outstanding data (other than
2518 * a window probe), this is a completely
2519 * duplicate ack (ie, window info didn't
2520 * change and FIN isn't set),
2521 * the ack is the biggest we've
2522 * seen and we've seen exactly our rexmt
2523 * threshold of them, assume a packet
2524 * has been dropped and retransmit it.
2525 * Kludge snd_nxt & the congestion
2526 * window so we send only this one
2529 * We know we're losing at the current
2530 * window size so do congestion avoidance
2531 * (set ssthresh to half the current window
2532 * and pull our congestion window back to
2533 * the new ssthresh).
2535 * Dup acks mean that packets have left the
2536 * network (they're now cached at the receiver)
2537 * so bump cwnd by the amount in the receiver
2538 * to keep a constant cwnd packets in the
2541 * When using TCP ECN, notify the peer that
2542 * we reduced the cwnd.
2545 * Following 2 kinds of acks should not affect
2548 * 2) Acks with SACK but without any new SACK
2549 * information in them. These could result from
2550 * any anomaly in the network like a switch
2551 * duplicating packets or a possible DoS attack.
2553 if (th->th_ack != tp->snd_una ||
2554 ((tp->t_flags & TF_SACK_PERMIT) &&
2557 else if (!tcp_timer_active(tp, TT_REXMT))
2559 else if (++tp->t_dupacks > tcprexmtthresh ||
2560 IN_FASTRECOVERY(tp->t_flags)) {
2561 cc_ack_received(tp, th, nsegs,
2563 if ((tp->t_flags & TF_SACK_PERMIT) &&
2564 IN_FASTRECOVERY(tp->t_flags)) {
2568 * Compute the amount of data in flight first.
2569 * We can inject new data into the pipe iff
2570 * we have less than 1/2 the original window's
2571 * worth of data in flight.
2573 if (V_tcp_do_rfc6675_pipe)
2574 awnd = tcp_compute_pipe(tp);
2576 awnd = (tp->snd_nxt - tp->snd_fack) +
2577 tp->sackhint.sack_bytes_rexmit;
2579 if (awnd < tp->snd_ssthresh) {
2580 tp->snd_cwnd += maxseg;
2581 if (tp->snd_cwnd > tp->snd_ssthresh)
2582 tp->snd_cwnd = tp->snd_ssthresh;
2585 tp->snd_cwnd += maxseg;
2586 (void) tp->t_fb->tfb_tcp_output(tp);
2588 } else if (tp->t_dupacks == tcprexmtthresh) {
2589 tcp_seq onxt = tp->snd_nxt;
2592 * If we're doing sack, check to
2593 * see if we're already in sack
2594 * recovery. If we're not doing sack,
2595 * check to see if we're in newreno
2598 if (tp->t_flags & TF_SACK_PERMIT) {
2599 if (IN_FASTRECOVERY(tp->t_flags)) {
2604 if (SEQ_LEQ(th->th_ack,
2610 /* Congestion signal before ack. */
2611 cc_cong_signal(tp, th, CC_NDUPACK);
2612 cc_ack_received(tp, th, nsegs,
2614 tcp_timer_activate(tp, TT_REXMT, 0);
2616 if (tp->t_flags & TF_SACK_PERMIT) {
2618 tcps_sack_recovery_episode);
2619 tp->sack_newdata = tp->snd_nxt;
2620 tp->snd_cwnd = maxseg;
2621 (void) tp->t_fb->tfb_tcp_output(tp);
2624 tp->snd_nxt = th->th_ack;
2625 tp->snd_cwnd = maxseg;
2626 (void) tp->t_fb->tfb_tcp_output(tp);
2627 KASSERT(tp->snd_limited <= 2,
2628 ("%s: tp->snd_limited too big",
2630 tp->snd_cwnd = tp->snd_ssthresh +
2632 (tp->t_dupacks - tp->snd_limited);
2633 if (SEQ_GT(onxt, tp->snd_nxt))
2636 } else if (V_tcp_do_rfc3042) {
2638 * Process first and second duplicate
2639 * ACKs. Each indicates a segment
2640 * leaving the network, creating room
2641 * for more. Make sure we can send a
2642 * packet on reception of each duplicate
2643 * ACK by increasing snd_cwnd by one
2644 * segment. Restore the original
2645 * snd_cwnd after packet transmission.
2647 cc_ack_received(tp, th, nsegs,
2649 uint32_t oldcwnd = tp->snd_cwnd;
2650 tcp_seq oldsndmax = tp->snd_max;
2654 KASSERT(tp->t_dupacks == 1 ||
2656 ("%s: dupacks not 1 or 2",
2658 if (tp->t_dupacks == 1)
2659 tp->snd_limited = 0;
2661 (tp->snd_nxt - tp->snd_una) +
2662 (tp->t_dupacks - tp->snd_limited) *
2665 * Only call tcp_output when there
2666 * is new data available to be sent.
2667 * Otherwise we would send pure ACKs.
2669 SOCKBUF_LOCK(&so->so_snd);
2670 avail = sbavail(&so->so_snd) -
2671 (tp->snd_nxt - tp->snd_una);
2672 SOCKBUF_UNLOCK(&so->so_snd);
2674 (void) tp->t_fb->tfb_tcp_output(tp);
2675 sent = tp->snd_max - oldsndmax;
2676 if (sent > maxseg) {
2677 KASSERT((tp->t_dupacks == 2 &&
2678 tp->snd_limited == 0) ||
2679 (sent == maxseg + 1 &&
2680 tp->t_flags & TF_SENTFIN),
2681 ("%s: sent too much",
2683 tp->snd_limited = 2;
2684 } else if (sent > 0)
2686 tp->snd_cwnd = oldcwnd;
2693 * This ack is advancing the left edge, reset the
2698 * If this ack also has new SACK info, increment the
2699 * counter as per rfc6675.
2701 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed)
2705 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2706 ("%s: th_ack <= snd_una", __func__));
2709 * If the congestion window was inflated to account
2710 * for the other side's cached packets, retract it.
2712 if (IN_FASTRECOVERY(tp->t_flags)) {
2713 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2714 if (tp->t_flags & TF_SACK_PERMIT)
2715 tcp_sack_partialack(tp, th);
2717 tcp_newreno_partial_ack(tp, th);
2719 cc_post_recovery(tp, th);
2722 * If we reach this point, ACK is not a duplicate,
2723 * i.e., it ACKs something we sent.
2725 if (tp->t_flags & TF_NEEDSYN) {
2727 * T/TCP: Connection was half-synchronized, and our
2728 * SYN has been ACK'd (so connection is now fully
2729 * synchronized). Go to non-starred state,
2730 * increment snd_una for ACK of SYN, and check if
2731 * we can do window scaling.
2733 tp->t_flags &= ~TF_NEEDSYN;
2735 /* Do window scaling? */
2736 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2737 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2738 tp->rcv_scale = tp->request_r_scale;
2739 /* Send window already scaled. */
2744 INP_WLOCK_ASSERT(tp->t_inpcb);
2746 acked = BYTES_THIS_ACK(tp, th);
2747 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2748 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2749 tp->snd_una, th->th_ack, tp, m));
2750 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2751 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2754 * If we just performed our first retransmit, and the ACK
2755 * arrives within our recovery window, then it was a mistake
2756 * to do the retransmit in the first place. Recover our
2757 * original cwnd and ssthresh, and proceed to transmit where
2760 if (tp->t_rxtshift == 1 &&
2761 tp->t_flags & TF_PREVVALID &&
2763 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2764 cc_cong_signal(tp, th, CC_RTO_ERR);
2767 * If we have a timestamp reply, update smoothed
2768 * round trip time. If no timestamp is present but
2769 * transmit timer is running and timed sequence
2770 * number was acked, update smoothed round trip time.
2771 * Since we now have an rtt measurement, cancel the
2772 * timer backoff (cf., Phil Karn's retransmit alg.).
2773 * Recompute the initial retransmit timer.
2775 * Some boxes send broken timestamp replies
2776 * during the SYN+ACK phase, ignore
2777 * timestamps of 0 or we could calculate a
2778 * huge RTT and blow up the retransmit timer.
2780 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2783 t = tcp_ts_getticks() - to.to_tsecr;
2784 if (!tp->t_rttlow || tp->t_rttlow > t)
2786 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2787 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2788 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2789 tp->t_rttlow = ticks - tp->t_rtttime;
2790 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2794 * If all outstanding data is acked, stop retransmit
2795 * timer and remember to restart (more output or persist).
2796 * If there is more data to be acked, restart retransmit
2797 * timer, using current (possibly backed-off) value.
2799 if (th->th_ack == tp->snd_max) {
2800 tcp_timer_activate(tp, TT_REXMT, 0);
2802 } else if (!tcp_timer_active(tp, TT_PERSIST))
2803 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2806 * If no data (only SYN) was ACK'd,
2807 * skip rest of ACK processing.
2813 * Let the congestion control algorithm update congestion
2814 * control related information. This typically means increasing
2815 * the congestion window.
2817 cc_ack_received(tp, th, nsegs, CC_ACK);
2819 SOCKBUF_LOCK(&so->so_snd);
2820 if (acked > sbavail(&so->so_snd)) {
2821 if (tp->snd_wnd >= sbavail(&so->so_snd))
2822 tp->snd_wnd -= sbavail(&so->so_snd);
2825 mfree = sbcut_locked(&so->so_snd,
2826 (int)sbavail(&so->so_snd));
2829 mfree = sbcut_locked(&so->so_snd, acked);
2830 if (tp->snd_wnd >= (uint32_t) acked)
2831 tp->snd_wnd -= acked;
2836 /* NB: sowwakeup_locked() does an implicit unlock. */
2837 sowwakeup_locked(so);
2839 /* Detect una wraparound. */
2840 if (!IN_RECOVERY(tp->t_flags) &&
2841 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2842 SEQ_LEQ(th->th_ack, tp->snd_recover))
2843 tp->snd_recover = th->th_ack - 1;
2844 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2845 if (IN_RECOVERY(tp->t_flags) &&
2846 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2847 EXIT_RECOVERY(tp->t_flags);
2849 tp->snd_una = th->th_ack;
2850 if (tp->t_flags & TF_SACK_PERMIT) {
2851 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2852 tp->snd_recover = tp->snd_una;
2854 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2855 tp->snd_nxt = tp->snd_una;
2857 switch (tp->t_state) {
2860 * In FIN_WAIT_1 STATE in addition to the processing
2861 * for the ESTABLISHED state if our FIN is now acknowledged
2862 * then enter FIN_WAIT_2.
2864 case TCPS_FIN_WAIT_1:
2865 if (ourfinisacked) {
2867 * If we can't receive any more
2868 * data, then closing user can proceed.
2869 * Starting the timer is contrary to the
2870 * specification, but if we don't get a FIN
2871 * we'll hang forever.
2874 * we should release the tp also, and use a
2877 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2878 soisdisconnected(so);
2879 tcp_timer_activate(tp, TT_2MSL,
2880 (tcp_fast_finwait2_recycle ?
2881 tcp_finwait2_timeout :
2884 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2889 * In CLOSING STATE in addition to the processing for
2890 * the ESTABLISHED state if the ACK acknowledges our FIN
2891 * then enter the TIME-WAIT state, otherwise ignore
2895 if (ourfinisacked) {
2896 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2904 * In LAST_ACK, we may still be waiting for data to drain
2905 * and/or to be acked, as well as for the ack of our FIN.
2906 * If our FIN is now acknowledged, delete the TCB,
2907 * enter the closed state and return.
2910 if (ourfinisacked) {
2911 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2920 INP_WLOCK_ASSERT(tp->t_inpcb);
2923 * Update window information.
2924 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2926 if ((thflags & TH_ACK) &&
2927 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2928 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2929 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2930 /* keep track of pure window updates */
2932 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2933 TCPSTAT_INC(tcps_rcvwinupd);
2934 tp->snd_wnd = tiwin;
2935 tp->snd_wl1 = th->th_seq;
2936 tp->snd_wl2 = th->th_ack;
2937 if (tp->snd_wnd > tp->max_sndwnd)
2938 tp->max_sndwnd = tp->snd_wnd;
2943 * Process segments with URG.
2945 if ((thflags & TH_URG) && th->th_urp &&
2946 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2948 * This is a kludge, but if we receive and accept
2949 * random urgent pointers, we'll crash in
2950 * soreceive. It's hard to imagine someone
2951 * actually wanting to send this much urgent data.
2953 SOCKBUF_LOCK(&so->so_rcv);
2954 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2955 th->th_urp = 0; /* XXX */
2956 thflags &= ~TH_URG; /* XXX */
2957 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2958 goto dodata; /* XXX */
2961 * If this segment advances the known urgent pointer,
2962 * then mark the data stream. This should not happen
2963 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2964 * a FIN has been received from the remote side.
2965 * In these states we ignore the URG.
2967 * According to RFC961 (Assigned Protocols),
2968 * the urgent pointer points to the last octet
2969 * of urgent data. We continue, however,
2970 * to consider it to indicate the first octet
2971 * of data past the urgent section as the original
2972 * spec states (in one of two places).
2974 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2975 tp->rcv_up = th->th_seq + th->th_urp;
2976 so->so_oobmark = sbavail(&so->so_rcv) +
2977 (tp->rcv_up - tp->rcv_nxt) - 1;
2978 if (so->so_oobmark == 0)
2979 so->so_rcv.sb_state |= SBS_RCVATMARK;
2981 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2983 SOCKBUF_UNLOCK(&so->so_rcv);
2985 * Remove out of band data so doesn't get presented to user.
2986 * This can happen independent of advancing the URG pointer,
2987 * but if two URG's are pending at once, some out-of-band
2988 * data may creep in... ick.
2990 if (th->th_urp <= (uint32_t)tlen &&
2991 !(so->so_options & SO_OOBINLINE)) {
2992 /* hdr drop is delayed */
2993 tcp_pulloutofband(so, th, m, drop_hdrlen);
2997 * If no out of band data is expected,
2998 * pull receive urgent pointer along
2999 * with the receive window.
3001 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3002 tp->rcv_up = tp->rcv_nxt;
3005 INP_WLOCK_ASSERT(tp->t_inpcb);
3008 * Process the segment text, merging it into the TCP sequencing queue,
3009 * and arranging for acknowledgment of receipt if necessary.
3010 * This process logically involves adjusting tp->rcv_wnd as data
3011 * is presented to the user (this happens in tcp_usrreq.c,
3012 * case PRU_RCVD). If a FIN has already been received on this
3013 * connection then we just ignore the text.
3015 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3016 IS_FASTOPEN(tp->t_flags));
3017 if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
3018 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3019 tcp_seq save_start = th->th_seq;
3020 tcp_seq save_rnxt = tp->rcv_nxt;
3021 int save_tlen = tlen;
3022 m_adj(m, drop_hdrlen); /* delayed header drop */
3024 * Insert segment which includes th into TCP reassembly queue
3025 * with control block tp. Set thflags to whether reassembly now
3026 * includes a segment with FIN. This handles the common case
3027 * inline (segment is the next to be received on an established
3028 * connection, and the queue is empty), avoiding linkage into
3029 * and removal from the queue and repetition of various
3031 * Set DELACK for segments received in order, but ack
3032 * immediately when segments are out of order (so
3033 * fast retransmit can work).
3035 if (th->th_seq == tp->rcv_nxt &&
3037 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3039 if (DELAY_ACK(tp, tlen) || tfo_syn)
3040 tp->t_flags |= TF_DELACK;
3042 tp->t_flags |= TF_ACKNOW;
3043 tp->rcv_nxt += tlen;
3044 thflags = th->th_flags & TH_FIN;
3045 TCPSTAT_INC(tcps_rcvpack);
3046 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3047 SOCKBUF_LOCK(&so->so_rcv);
3048 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3051 sbappendstream_locked(&so->so_rcv, m, 0);
3052 /* NB: sorwakeup_locked() does an implicit unlock. */
3053 sorwakeup_locked(so);
3056 * XXX: Due to the header drop above "th" is
3057 * theoretically invalid by now. Fortunately
3058 * m_adj() doesn't actually frees any mbufs
3059 * when trimming from the head.
3061 tcp_seq temp = save_start;
3062 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3063 tp->t_flags |= TF_ACKNOW;
3065 if (tp->t_flags & TF_SACK_PERMIT) {
3066 if (((tlen == 0) && (save_tlen > 0) &&
3067 (SEQ_LT(save_start, save_rnxt)))) {
3069 * DSACK actually handled in the fastpath
3072 tcp_update_sack_list(tp, save_start, save_start + save_tlen);
3074 if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3076 * Cleaning sackblks by using zero length
3079 tcp_update_sack_list(tp, save_start, save_start);
3081 if ((tlen > 0) && (tlen >= save_tlen)) {
3082 /* Update of sackblks. */
3083 tcp_update_sack_list(tp, save_start, save_start + save_tlen);
3086 tcp_update_sack_list(tp, save_start, save_start+tlen);
3091 * Note the amount of data that peer has sent into
3092 * our window, in order to estimate the sender's
3096 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3097 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3099 len = so->so_rcv.sb_hiwat;
3107 * If FIN is received ACK the FIN and let the user know
3108 * that the connection is closing.
3110 if (thflags & TH_FIN) {
3111 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3114 * If connection is half-synchronized
3115 * (ie NEEDSYN flag on) then delay ACK,
3116 * so it may be piggybacked when SYN is sent.
3117 * Otherwise, since we received a FIN then no
3118 * more input can be expected, send ACK now.
3120 if (tp->t_flags & TF_NEEDSYN)
3121 tp->t_flags |= TF_DELACK;
3123 tp->t_flags |= TF_ACKNOW;
3126 switch (tp->t_state) {
3129 * In SYN_RECEIVED and ESTABLISHED STATES
3130 * enter the CLOSE_WAIT state.
3132 case TCPS_SYN_RECEIVED:
3133 tp->t_starttime = ticks;
3135 case TCPS_ESTABLISHED:
3136 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3140 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3141 * enter the CLOSING state.
3143 case TCPS_FIN_WAIT_1:
3144 tcp_state_change(tp, TCPS_CLOSING);
3148 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3149 * starting the time-wait timer, turning off the other
3152 case TCPS_FIN_WAIT_2:
3153 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
3160 if (so->so_options & SO_DEBUG)
3161 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3164 TCP_PROBE3(debug__input, tp, th, m);
3167 * Return any desired output.
3169 if (needoutput || (tp->t_flags & TF_ACKNOW))
3170 (void) tp->t_fb->tfb_tcp_output(tp);
3173 INP_WLOCK_ASSERT(tp->t_inpcb);
3175 if (tp->t_flags & TF_DELACK) {
3176 tp->t_flags &= ~TF_DELACK;
3177 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3179 INP_WUNLOCK(tp->t_inpcb);
3184 * Generate an ACK dropping incoming segment if it occupies
3185 * sequence space, where the ACK reflects our state.
3187 * We can now skip the test for the RST flag since all
3188 * paths to this code happen after packets containing
3189 * RST have been dropped.
3191 * In the SYN-RECEIVED state, don't send an ACK unless the
3192 * segment we received passes the SYN-RECEIVED ACK test.
3193 * If it fails send a RST. This breaks the loop in the
3194 * "LAND" DoS attack, and also prevents an ACK storm
3195 * between two listening ports that have been sent forged
3196 * SYN segments, each with the source address of the other.
3198 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3199 (SEQ_GT(tp->snd_una, th->th_ack) ||
3200 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3201 rstreason = BANDLIM_RST_OPENPORT;
3205 if (so->so_options & SO_DEBUG)
3206 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3209 TCP_PROBE3(debug__input, tp, th, m);
3210 tp->t_flags |= TF_ACKNOW;
3211 (void) tp->t_fb->tfb_tcp_output(tp);
3212 INP_WUNLOCK(tp->t_inpcb);
3218 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3219 INP_WUNLOCK(tp->t_inpcb);
3221 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3226 * Drop space held by incoming segment and return.
3229 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3230 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3233 TCP_PROBE3(debug__input, tp, th, m);
3235 INP_WUNLOCK(tp->t_inpcb);
3240 * Issue RST and make ACK acceptable to originator of segment.
3241 * The mbuf must still include the original packet header.
3245 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3246 int tlen, int rstreason)
3252 struct ip6_hdr *ip6;
3256 INP_WLOCK_ASSERT(tp->t_inpcb);
3259 /* Don't bother if destination was broadcast/multicast. */
3260 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3263 if (mtod(m, struct ip *)->ip_v == 6) {
3264 ip6 = mtod(m, struct ip6_hdr *);
3265 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3266 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3268 /* IPv6 anycast check is done at tcp6_input() */
3271 #if defined(INET) && defined(INET6)
3276 ip = mtod(m, struct ip *);
3277 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3278 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3279 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3280 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3285 /* Perform bandwidth limiting. */
3286 if (badport_bandlim(rstreason) < 0)
3289 /* tcp_respond consumes the mbuf chain. */
3290 if (th->th_flags & TH_ACK) {
3291 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3292 th->th_ack, TH_RST);
3294 if (th->th_flags & TH_SYN)
3296 if (th->th_flags & TH_FIN)
3298 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3299 (tcp_seq)0, TH_RST|TH_ACK);
3307 * Parse TCP options and place in tcpopt.
3310 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3315 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3317 if (opt == TCPOPT_EOL)
3319 if (opt == TCPOPT_NOP)
3325 if (optlen < 2 || optlen > cnt)
3330 if (optlen != TCPOLEN_MAXSEG)
3332 if (!(flags & TO_SYN))
3334 to->to_flags |= TOF_MSS;
3335 bcopy((char *)cp + 2,
3336 (char *)&to->to_mss, sizeof(to->to_mss));
3337 to->to_mss = ntohs(to->to_mss);
3340 if (optlen != TCPOLEN_WINDOW)
3342 if (!(flags & TO_SYN))
3344 to->to_flags |= TOF_SCALE;
3345 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3347 case TCPOPT_TIMESTAMP:
3348 if (optlen != TCPOLEN_TIMESTAMP)
3350 to->to_flags |= TOF_TS;
3351 bcopy((char *)cp + 2,
3352 (char *)&to->to_tsval, sizeof(to->to_tsval));
3353 to->to_tsval = ntohl(to->to_tsval);
3354 bcopy((char *)cp + 6,
3355 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3356 to->to_tsecr = ntohl(to->to_tsecr);
3358 case TCPOPT_SIGNATURE:
3360 * In order to reply to a host which has set the
3361 * TCP_SIGNATURE option in its initial SYN, we have
3362 * to record the fact that the option was observed
3363 * here for the syncache code to perform the correct
3366 if (optlen != TCPOLEN_SIGNATURE)
3368 to->to_flags |= TOF_SIGNATURE;
3369 to->to_signature = cp + 2;
3371 case TCPOPT_SACK_PERMITTED:
3372 if (optlen != TCPOLEN_SACK_PERMITTED)
3374 if (!(flags & TO_SYN))
3378 to->to_flags |= TOF_SACKPERM;
3381 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3385 to->to_flags |= TOF_SACK;
3386 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3387 to->to_sacks = cp + 2;
3388 TCPSTAT_INC(tcps_sack_rcv_blocks);
3390 case TCPOPT_FAST_OPEN:
3392 * Cookie length validation is performed by the
3393 * server side cookie checking code or the client
3394 * side cookie cache update code.
3396 if (!(flags & TO_SYN))
3398 if (!V_tcp_fastopen_client_enable &&
3399 !V_tcp_fastopen_server_enable)
3401 to->to_flags |= TOF_FASTOPEN;
3402 to->to_tfo_len = optlen - 2;
3403 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3412 * Pull out of band byte out of a segment so
3413 * it doesn't appear in the user's data queue.
3414 * It is still reflected in the segment length for
3415 * sequencing purposes.
3418 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3421 int cnt = off + th->th_urp - 1;
3424 if (m->m_len > cnt) {
3425 char *cp = mtod(m, caddr_t) + cnt;
3426 struct tcpcb *tp = sototcpcb(so);
3428 INP_WLOCK_ASSERT(tp->t_inpcb);
3431 tp->t_oobflags |= TCPOOB_HAVEDATA;
3432 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3434 if (m->m_flags & M_PKTHDR)
3443 panic("tcp_pulloutofband");
3447 * Collect new round-trip time estimate
3448 * and update averages and current timeout.
3451 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3455 INP_WLOCK_ASSERT(tp->t_inpcb);
3457 TCPSTAT_INC(tcps_rttupdated);
3459 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3461 * srtt is stored as fixed point with 5 bits after the
3462 * binary point (i.e., scaled by 8). The following magic
3463 * is equivalent to the smoothing algorithm in rfc793 with
3464 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3465 * point). Adjust rtt to origin 0.
3467 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3468 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3470 if ((tp->t_srtt += delta) <= 0)
3474 * We accumulate a smoothed rtt variance (actually, a
3475 * smoothed mean difference), then set the retransmit
3476 * timer to smoothed rtt + 4 times the smoothed variance.
3477 * rttvar is stored as fixed point with 4 bits after the
3478 * binary point (scaled by 16). The following is
3479 * equivalent to rfc793 smoothing with an alpha of .75
3480 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3481 * rfc793's wired-in beta.
3485 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3486 if ((tp->t_rttvar += delta) <= 0)
3488 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3489 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3492 * No rtt measurement yet - use the unsmoothed rtt.
3493 * Set the variance to half the rtt (so our first
3494 * retransmit happens at 3*rtt).
3496 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3497 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3498 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3504 * the retransmit should happen at rtt + 4 * rttvar.
3505 * Because of the way we do the smoothing, srtt and rttvar
3506 * will each average +1/2 tick of bias. When we compute
3507 * the retransmit timer, we want 1/2 tick of rounding and
3508 * 1 extra tick because of +-1/2 tick uncertainty in the
3509 * firing of the timer. The bias will give us exactly the
3510 * 1.5 tick we need. But, because the bias is
3511 * statistical, we have to test that we don't drop below
3512 * the minimum feasible timer (which is 2 ticks).
3514 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3515 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3518 * We received an ack for a packet that wasn't retransmitted;
3519 * it is probably safe to discard any error indications we've
3520 * received recently. This isn't quite right, but close enough
3521 * for now (a route might have failed after we sent a segment,
3522 * and the return path might not be symmetrical).
3524 tp->t_softerror = 0;
3528 * Determine a reasonable value for maxseg size.
3529 * If the route is known, check route for mtu.
3530 * If none, use an mss that can be handled on the outgoing interface
3531 * without forcing IP to fragment. If no route is found, route has no mtu,
3532 * or the destination isn't local, use a default, hopefully conservative
3533 * size (usually 512 or the default IP max size, but no more than the mtu
3534 * of the interface), as we can't discover anything about intervening
3535 * gateways or networks. We also initialize the congestion/slow start
3536 * window to be a single segment if the destination isn't local.
3537 * While looking at the routing entry, we also initialize other path-dependent
3538 * parameters from pre-set or cached values in the routing entry.
3540 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3541 * IP options, e.g. IPSEC data, since length of this data may vary, and
3542 * thus it is calculated for every segment separately in tcp_output().
3544 * NOTE that this routine is only called when we process an incoming
3545 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3546 * settings are handled in tcp_mssopt().
3549 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3550 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3553 uint32_t maxmtu = 0;
3554 struct inpcb *inp = tp->t_inpcb;
3555 struct hc_metrics_lite metrics;
3557 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3558 size_t min_protoh = isipv6 ?
3559 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3560 sizeof (struct tcpiphdr);
3562 const size_t min_protoh = sizeof(struct tcpiphdr);
3565 INP_WLOCK_ASSERT(tp->t_inpcb);
3567 if (mtuoffer != -1) {
3568 KASSERT(offer == -1, ("%s: conflict", __func__));
3569 offer = mtuoffer - min_protoh;
3575 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3576 tp->t_maxseg = V_tcp_v6mssdflt;
3579 #if defined(INET) && defined(INET6)
3584 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3585 tp->t_maxseg = V_tcp_mssdflt;
3590 * No route to sender, stay with default mss and return.
3594 * In case we return early we need to initialize metrics
3595 * to a defined state as tcp_hc_get() would do for us
3596 * if there was no cache hit.
3598 if (metricptr != NULL)
3599 bzero(metricptr, sizeof(struct hc_metrics_lite));
3603 /* What have we got? */
3607 * Offer == 0 means that there was no MSS on the SYN
3608 * segment, in this case we use tcp_mssdflt as
3609 * already assigned to t_maxseg above.
3611 offer = tp->t_maxseg;
3616 * Offer == -1 means that we didn't receive SYN yet.
3622 * Prevent DoS attack with too small MSS. Round up
3623 * to at least minmss.
3625 offer = max(offer, V_tcp_minmss);
3629 * rmx information is now retrieved from tcp_hostcache.
3631 tcp_hc_get(&inp->inp_inc, &metrics);
3632 if (metricptr != NULL)
3633 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3636 * If there's a discovered mtu in tcp hostcache, use it.
3637 * Else, use the link mtu.
3639 if (metrics.rmx_mtu)
3640 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3644 mss = maxmtu - min_protoh;
3645 if (!V_path_mtu_discovery &&
3646 !in6_localaddr(&inp->in6p_faddr))
3647 mss = min(mss, V_tcp_v6mssdflt);
3650 #if defined(INET) && defined(INET6)
3655 mss = maxmtu - min_protoh;
3656 if (!V_path_mtu_discovery &&
3657 !in_localaddr(inp->inp_faddr))
3658 mss = min(mss, V_tcp_mssdflt);
3662 * XXX - The above conditional (mss = maxmtu - min_protoh)
3663 * probably violates the TCP spec.
3664 * The problem is that, since we don't know the
3665 * other end's MSS, we are supposed to use a conservative
3666 * default. But, if we do that, then MTU discovery will
3667 * never actually take place, because the conservative
3668 * default is much less than the MTUs typically seen
3669 * on the Internet today. For the moment, we'll sweep
3670 * this under the carpet.
3672 * The conservative default might not actually be a problem
3673 * if the only case this occurs is when sending an initial
3674 * SYN with options and data to a host we've never talked
3675 * to before. Then, they will reply with an MSS value which
3676 * will get recorded and the new parameters should get
3677 * recomputed. For Further Study.
3680 mss = min(mss, offer);
3683 * Sanity check: make sure that maxseg will be large
3684 * enough to allow some data on segments even if the
3685 * all the option space is used (40bytes). Otherwise
3686 * funny things may happen in tcp_output.
3688 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3696 tcp_mss(struct tcpcb *tp, int offer)
3702 struct hc_metrics_lite metrics;
3703 struct tcp_ifcap cap;
3705 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3707 bzero(&cap, sizeof(cap));
3708 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3714 * If there's a pipesize, change the socket buffer to that size,
3715 * don't change if sb_hiwat is different than default (then it
3716 * has been changed on purpose with setsockopt).
3717 * Make the socket buffers an integral number of mss units;
3718 * if the mss is larger than the socket buffer, decrease the mss.
3720 so = inp->inp_socket;
3721 SOCKBUF_LOCK(&so->so_snd);
3722 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3723 bufsize = metrics.rmx_sendpipe;
3725 bufsize = so->so_snd.sb_hiwat;
3729 bufsize = roundup(bufsize, mss);
3730 if (bufsize > sb_max)
3732 if (bufsize > so->so_snd.sb_hiwat)
3733 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3735 SOCKBUF_UNLOCK(&so->so_snd);
3737 * Sanity check: make sure that maxseg will be large
3738 * enough to allow some data on segments even if the
3739 * all the option space is used (40bytes). Otherwise
3740 * funny things may happen in tcp_output.
3742 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3744 tp->t_maxseg = max(mss, 64);
3746 SOCKBUF_LOCK(&so->so_rcv);
3747 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3748 bufsize = metrics.rmx_recvpipe;
3750 bufsize = so->so_rcv.sb_hiwat;
3751 if (bufsize > mss) {
3752 bufsize = roundup(bufsize, mss);
3753 if (bufsize > sb_max)
3755 if (bufsize > so->so_rcv.sb_hiwat)
3756 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3758 SOCKBUF_UNLOCK(&so->so_rcv);
3760 /* Check the interface for TSO capabilities. */
3761 if (cap.ifcap & CSUM_TSO) {
3762 tp->t_flags |= TF_TSO;
3763 tp->t_tsomax = cap.tsomax;
3764 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3765 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3770 * Determine the MSS option to send on an outgoing SYN.
3773 tcp_mssopt(struct in_conninfo *inc)
3776 uint32_t thcmtu = 0;
3777 uint32_t maxmtu = 0;
3780 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3783 if (inc->inc_flags & INC_ISIPV6) {
3784 mss = V_tcp_v6mssdflt;
3785 maxmtu = tcp_maxmtu6(inc, NULL);
3786 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3789 #if defined(INET) && defined(INET6)
3794 mss = V_tcp_mssdflt;
3795 maxmtu = tcp_maxmtu(inc, NULL);
3796 min_protoh = sizeof(struct tcpiphdr);
3799 #if defined(INET6) || defined(INET)
3800 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3803 if (maxmtu && thcmtu)
3804 mss = min(maxmtu, thcmtu) - min_protoh;
3805 else if (maxmtu || thcmtu)
3806 mss = max(maxmtu, thcmtu) - min_protoh;
3813 * On a partial ack arrives, force the retransmission of the
3814 * next unacknowledged segment. Do not clear tp->t_dupacks.
3815 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3819 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3821 tcp_seq onxt = tp->snd_nxt;
3822 uint32_t ocwnd = tp->snd_cwnd;
3823 u_int maxseg = tcp_maxseg(tp);
3825 INP_WLOCK_ASSERT(tp->t_inpcb);
3827 tcp_timer_activate(tp, TT_REXMT, 0);
3829 tp->snd_nxt = th->th_ack;
3831 * Set snd_cwnd to one segment beyond acknowledged offset.
3832 * (tp->snd_una has not yet been updated when this function is called.)
3834 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3835 tp->t_flags |= TF_ACKNOW;
3836 (void) tp->t_fb->tfb_tcp_output(tp);
3837 tp->snd_cwnd = ocwnd;
3838 if (SEQ_GT(onxt, tp->snd_nxt))
3841 * Partial window deflation. Relies on fact that tp->snd_una
3844 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3845 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3848 tp->snd_cwnd += maxseg;
3852 tcp_compute_pipe(struct tcpcb *tp)
3854 return (tp->snd_max - tp->snd_una +
3855 tp->sackhint.sack_bytes_rexmit -
3856 tp->sackhint.sacked_bytes);