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>
62 #include <sys/kernel.h>
64 #include <sys/hhook.h>
66 #include <sys/malloc.h>
68 #include <sys/proc.h> /* for proc0 declaration */
69 #include <sys/protosw.h>
70 #include <sys/qmath.h>
72 #include <sys/signalvar.h>
73 #include <sys/socket.h>
74 #include <sys/socketvar.h>
75 #include <sys/sysctl.h>
76 #include <sys/syslog.h>
77 #include <sys/systm.h>
78 #include <sys/stats.h>
80 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
85 #include <net/if_var.h>
86 #include <net/route.h>
89 #define TCPSTATES /* for logging */
91 #include <netinet/in.h>
92 #include <netinet/in_kdtrace.h>
93 #include <netinet/in_pcb.h>
94 #include <netinet/in_systm.h>
95 #include <netinet/ip.h>
96 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
97 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
98 #include <netinet/ip_var.h>
99 #include <netinet/ip_options.h>
100 #include <netinet/ip6.h>
101 #include <netinet/icmp6.h>
102 #include <netinet6/in6_pcb.h>
103 #include <netinet6/in6_var.h>
104 #include <netinet6/ip6_var.h>
105 #include <netinet6/nd6.h>
106 #include <netinet/tcp.h>
107 #include <netinet/tcp_fsm.h>
108 #include <netinet/tcp_log_buf.h>
109 #include <netinet/tcp_seq.h>
110 #include <netinet/tcp_timer.h>
111 #include <netinet/tcp_var.h>
112 #include <netinet6/tcp6_var.h>
113 #include <netinet/tcpip.h>
114 #include <netinet/cc/cc.h>
115 #include <netinet/tcp_fastopen.h>
117 #include <netinet/tcp_pcap.h>
119 #include <netinet/tcp_syncache.h>
121 #include <netinet/tcp_debug.h>
122 #endif /* TCPDEBUG */
124 #include <netinet/tcp_offload.h>
127 #include <netipsec/ipsec_support.h>
129 #include <machine/in_cksum.h>
131 #include <security/mac/mac_framework.h>
133 const int tcprexmtthresh = 3;
135 VNET_DEFINE(int, tcp_log_in_vain) = 0;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
137 &VNET_NAME(tcp_log_in_vain), 0,
138 "Log all incoming TCP segments to closed ports");
140 VNET_DEFINE(int, blackhole) = 0;
141 #define V_blackhole VNET(blackhole)
142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
143 &VNET_NAME(blackhole), 0,
144 "Do not send RST on segments to closed ports");
146 VNET_DEFINE(int, tcp_delack_enabled) = 1;
147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
148 &VNET_NAME(tcp_delack_enabled), 0,
149 "Delay ACK to try and piggyback it onto a data packet");
151 VNET_DEFINE(int, drop_synfin) = 0;
152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
153 &VNET_NAME(drop_synfin), 0,
154 "Drop TCP packets with SYN+FIN set");
156 VNET_DEFINE(int, tcp_do_newcwv) = 0;
157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
158 &VNET_NAME(tcp_do_newcwv), 0,
159 "Enable New Congestion Window Validation per RFC7661");
161 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
163 &VNET_NAME(tcp_do_rfc6675_pipe), 0,
164 "Use calculated pipe/in-flight bytes per RFC 6675");
166 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
168 &VNET_NAME(tcp_do_rfc3042), 0,
169 "Enable RFC 3042 (Limited Transmit)");
171 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
173 &VNET_NAME(tcp_do_rfc3390), 0,
174 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
176 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
178 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
179 "Slow-start flight size (initial congestion window) in number of segments");
181 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
183 &VNET_NAME(tcp_do_rfc3465), 0,
184 "Enable RFC 3465 (Appropriate Byte Counting)");
186 VNET_DEFINE(int, tcp_abc_l_var) = 2;
187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
188 &VNET_NAME(tcp_abc_l_var), 2,
189 "Cap the max cwnd increment during slow-start to this number of segments");
191 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn,
192 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
195 VNET_DEFINE(int, tcp_do_ecn) = 2;
196 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
197 &VNET_NAME(tcp_do_ecn), 0,
200 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
201 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
202 &VNET_NAME(tcp_ecn_maxretries), 0,
203 "Max retries before giving up on ECN");
205 VNET_DEFINE(int, tcp_insecure_syn) = 0;
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
207 &VNET_NAME(tcp_insecure_syn), 0,
208 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
210 VNET_DEFINE(int, tcp_insecure_rst) = 0;
211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
212 &VNET_NAME(tcp_insecure_rst), 0,
213 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
215 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
216 #define V_tcp_recvspace VNET(tcp_recvspace)
217 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
218 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
220 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
222 &VNET_NAME(tcp_do_autorcvbuf), 0,
223 "Enable automatic receive buffer sizing");
225 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
226 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
227 &VNET_NAME(tcp_autorcvbuf_max), 0,
228 "Max size of automatic receive buffer");
230 VNET_DEFINE(struct inpcbhead, tcb);
231 #define tcb6 tcb /* for KAME src sync over BSD*'s */
232 VNET_DEFINE(struct inpcbinfo, tcbinfo);
235 * TCP statistics are stored in an array of counter(9)s, which size matches
236 * size of struct tcpstat. TCP running connection count is a regular array.
238 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
239 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
240 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
241 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
242 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
243 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
244 "TCP connection counts by TCP state");
247 tcp_vnet_init(const void *unused)
250 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
251 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
253 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
254 tcp_vnet_init, NULL);
258 tcp_vnet_uninit(const void *unused)
261 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
262 VNET_PCPUSTAT_FREE(tcpstat);
264 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
265 tcp_vnet_uninit, NULL);
269 * Kernel module interface for updating tcpstat. The first argument is an index
270 * into tcpstat treated as an array.
273 kmod_tcpstat_add(int statnum, int val)
276 counter_u64_add(VNET(tcpstat)[statnum], val);
281 * Wrapper for the TCP established input helper hook.
284 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
286 struct tcp_hhook_data hhook_data;
288 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
293 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
300 * CC wrapper hook functions
303 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
310 INP_WLOCK_ASSERT(tp->t_inpcb);
312 tp->ccv->nsegs = nsegs;
313 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
314 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
315 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
316 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
317 tp->ccv->flags |= CCF_CWND_LIMITED;
319 tp->ccv->flags &= ~CCF_CWND_LIMITED;
321 if (type == CC_ACK) {
323 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
324 ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
325 if (!IN_RECOVERY(tp->t_flags))
326 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
327 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
328 if ((tp->t_flags & TF_GPUTINPROG) &&
329 SEQ_GEQ(th->th_ack, tp->gput_ack)) {
331 * Compute goodput in bits per millisecond.
333 gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) /
334 max(1, tcp_ts_getticks() - tp->gput_ts);
335 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
338 * XXXLAS: This is a temporary hack, and should be
339 * chained off VOI_TCP_GPUT when stats(9) grows an API
340 * to deal with chained VOIs.
342 if (tp->t_stats_gput_prev > 0)
343 stats_voi_update_abs_s32(tp->t_stats,
345 ((gput - tp->t_stats_gput_prev) * 100) /
346 tp->t_stats_gput_prev);
347 tp->t_flags &= ~TF_GPUTINPROG;
348 tp->t_stats_gput_prev = gput;
351 if (tp->snd_cwnd > tp->snd_ssthresh) {
352 tp->t_bytes_acked += tp->ccv->bytes_this_ack;
353 if (tp->t_bytes_acked >= tp->snd_cwnd) {
354 tp->t_bytes_acked -= tp->snd_cwnd;
355 tp->ccv->flags |= CCF_ABC_SENTAWND;
358 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
359 tp->t_bytes_acked = 0;
363 if (CC_ALGO(tp)->ack_received != NULL) {
364 /* XXXLAS: Find a way to live without this */
365 tp->ccv->curack = th->th_ack;
366 CC_ALGO(tp)->ack_received(tp->ccv, type);
369 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
374 cc_conn_init(struct tcpcb *tp)
376 struct hc_metrics_lite metrics;
377 struct inpcb *inp = tp->t_inpcb;
381 INP_WLOCK_ASSERT(tp->t_inpcb);
383 tcp_hc_get(&inp->inp_inc, &metrics);
384 maxseg = tcp_maxseg(tp);
386 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
388 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
389 TCPSTAT_INC(tcps_usedrtt);
390 if (metrics.rmx_rttvar) {
391 tp->t_rttvar = metrics.rmx_rttvar;
392 TCPSTAT_INC(tcps_usedrttvar);
394 /* default variation is +- 1 rtt */
396 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
398 TCPT_RANGESET(tp->t_rxtcur,
399 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
400 tp->t_rttmin, TCPTV_REXMTMAX);
402 if (metrics.rmx_ssthresh) {
404 * There's some sort of gateway or interface
405 * buffer limit on the path. Use this to set
406 * the slow start threshold, but set the
407 * threshold to no less than 2*mss.
409 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
410 TCPSTAT_INC(tcps_usedssthresh);
414 * Set the initial slow-start flight size.
416 * If a SYN or SYN/ACK was lost and retransmitted, we have to
417 * reduce the initial CWND to one segment as congestion is likely
418 * requiring us to be cautious.
420 if (tp->snd_cwnd == 1)
421 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
423 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
425 if (CC_ALGO(tp)->conn_init != NULL)
426 CC_ALGO(tp)->conn_init(tp->ccv);
430 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
434 INP_WLOCK_ASSERT(tp->t_inpcb);
437 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
442 if (!IN_FASTRECOVERY(tp->t_flags)) {
443 tp->snd_recover = tp->snd_max;
444 if (tp->t_flags2 & TF2_ECN_PERMIT)
445 tp->t_flags2 |= TF2_ECN_SND_CWR;
449 if (!IN_CONGRECOVERY(tp->t_flags) ||
451 * Allow ECN reaction on ACK to CWR, if
452 * that data segment was also CE marked.
454 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
455 EXIT_CONGRECOVERY(tp->t_flags);
456 TCPSTAT_INC(tcps_ecn_rcwnd);
457 tp->snd_recover = tp->snd_max + 1;
458 if (tp->t_flags2 & TF2_ECN_PERMIT)
459 tp->t_flags2 |= TF2_ECN_SND_CWR;
463 maxseg = tcp_maxseg(tp);
465 tp->t_bytes_acked = 0;
466 EXIT_RECOVERY(tp->t_flags);
467 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
469 tp->snd_cwnd = maxseg;
470 if (tp->t_flags2 & TF2_ECN_PERMIT)
471 tp->t_flags2 |= TF2_ECN_SND_CWR;
474 TCPSTAT_INC(tcps_sndrexmitbad);
475 /* RTO was unnecessary, so reset everything. */
476 tp->snd_cwnd = tp->snd_cwnd_prev;
477 tp->snd_ssthresh = tp->snd_ssthresh_prev;
478 tp->snd_recover = tp->snd_recover_prev;
479 if (tp->t_flags & TF_WASFRECOVERY)
480 ENTER_FASTRECOVERY(tp->t_flags);
481 if (tp->t_flags & TF_WASCRECOVERY)
482 ENTER_CONGRECOVERY(tp->t_flags);
483 tp->snd_nxt = tp->snd_max;
484 tp->t_flags &= ~TF_PREVVALID;
489 if (CC_ALGO(tp)->cong_signal != NULL) {
491 tp->ccv->curack = th->th_ack;
492 CC_ALGO(tp)->cong_signal(tp->ccv, type);
497 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
499 INP_WLOCK_ASSERT(tp->t_inpcb);
501 /* XXXLAS: KASSERT that we're in recovery? */
503 if (CC_ALGO(tp)->post_recovery != NULL) {
504 tp->ccv->curack = th->th_ack;
505 CC_ALGO(tp)->post_recovery(tp->ccv);
507 /* XXXLAS: EXIT_RECOVERY ? */
508 tp->t_bytes_acked = 0;
512 * Indicate whether this ack should be delayed. We can delay the ack if
513 * following conditions are met:
514 * - There is no delayed ack timer in progress.
515 * - Our last ack wasn't a 0-sized window. We never want to delay
516 * the ack that opens up a 0-sized window.
517 * - LRO wasn't used for this segment. We make sure by checking that the
518 * segment size is not larger than the MSS.
520 #define DELAY_ACK(tp, tlen) \
521 ((!tcp_timer_active(tp, TT_DELACK) && \
522 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
523 (tlen <= tp->t_maxseg) && \
524 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
527 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
529 INP_WLOCK_ASSERT(tp->t_inpcb);
531 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
532 switch (iptos & IPTOS_ECN_MASK) {
534 tp->ccv->flags |= CCF_IPHDR_CE;
540 case IPTOS_ECN_NOTECT:
541 tp->ccv->flags &= ~CCF_IPHDR_CE;
545 if (th->th_flags & TH_CWR)
546 tp->ccv->flags |= CCF_TCPHDR_CWR;
548 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
550 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
552 if (tp->ccv->flags & CCF_ACKNOW) {
553 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
554 tp->t_flags |= TF_ACKNOW;
560 * TCP input handling is split into multiple parts:
561 * tcp6_input is a thin wrapper around tcp_input for the extended
562 * ip6_protox[] call format in ip6_input
563 * tcp_input handles primary segment validation, inpcb lookup and
564 * SYN processing on listen sockets
565 * tcp_do_segment processes the ACK and text of the segment for
566 * establishing, established and closing connections
570 tcp6_input(struct mbuf **mp, int *offp, int proto)
573 struct in6_ifaddr *ia6;
577 if (m->m_len < *offp + sizeof(struct tcphdr)) {
578 m = m_pullup(m, *offp + sizeof(struct tcphdr));
581 TCPSTAT_INC(tcps_rcvshort);
582 return (IPPROTO_DONE);
587 * draft-itojun-ipv6-tcp-to-anycast
588 * better place to put this in?
590 ip6 = mtod(m, struct ip6_hdr *);
591 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
592 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
594 ifa_free(&ia6->ia_ifa);
595 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
596 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
598 return (IPPROTO_DONE);
601 ifa_free(&ia6->ia_ifa);
604 return (tcp_input(mp, offp, proto));
609 tcp_input(struct mbuf **mp, int *offp, int proto)
611 struct mbuf *m = *mp;
612 struct tcphdr *th = NULL;
613 struct ip *ip = NULL;
614 struct inpcb *inp = NULL;
615 struct tcpcb *tp = NULL;
616 struct socket *so = NULL;
627 int rstreason = 0; /* For badport_bandlim accounting purposes */
629 struct m_tag *fwd_tag = NULL;
631 struct ip6_hdr *ip6 = NULL;
634 const void *ip6 = NULL;
636 struct tcpopt to; /* options in this segment */
637 char *s = NULL; /* address and port logging */
640 * The size of tcp_saveipgen must be the size of the max ip header,
643 u_char tcp_saveipgen[IP6_HDR_LEN];
644 struct tcphdr tcp_savetcp;
651 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
658 TCPSTAT_INC(tcps_rcvtotal);
663 ip6 = mtod(m, struct ip6_hdr *);
664 th = (struct tcphdr *)((caddr_t)ip6 + off0);
665 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
666 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
667 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
668 th->th_sum = m->m_pkthdr.csum_data;
670 th->th_sum = in6_cksum_pseudo(ip6, tlen,
671 IPPROTO_TCP, m->m_pkthdr.csum_data);
672 th->th_sum ^= 0xffff;
674 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
676 TCPSTAT_INC(tcps_rcvbadsum);
681 * Be proactive about unspecified IPv6 address in source.
682 * As we use all-zero to indicate unbounded/unconnected pcb,
683 * unspecified IPv6 address can be used to confuse us.
685 * Note that packets with unspecified IPv6 destination is
686 * already dropped in ip6_input.
688 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
692 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
695 #if defined(INET) && defined(INET6)
701 * Get IP and TCP header together in first mbuf.
702 * Note: IP leaves IP header in first mbuf.
704 if (off0 > sizeof (struct ip)) {
706 off0 = sizeof(struct ip);
708 if (m->m_len < sizeof (struct tcpiphdr)) {
709 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
711 TCPSTAT_INC(tcps_rcvshort);
712 return (IPPROTO_DONE);
715 ip = mtod(m, struct ip *);
716 th = (struct tcphdr *)((caddr_t)ip + off0);
717 tlen = ntohs(ip->ip_len) - off0;
720 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
721 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
722 th->th_sum = m->m_pkthdr.csum_data;
724 th->th_sum = in_pseudo(ip->ip_src.s_addr,
726 htonl(m->m_pkthdr.csum_data + tlen +
728 th->th_sum ^= 0xffff;
730 struct ipovly *ipov = (struct ipovly *)ip;
733 * Checksum extended TCP header and data.
737 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
738 ipov->ih_len = htons(tlen);
739 th->th_sum = in_cksum(m, len);
740 /* Reset length for SDT probes. */
741 ip->ip_len = htons(len);
744 /* Re-initialization for later version check */
746 ip->ip_v = IPVERSION;
747 ip->ip_hl = off0 >> 2;
751 TCPSTAT_INC(tcps_rcvbadsum);
758 * Check that TCP offset makes sense,
759 * pull out TCP options and adjust length. XXX
761 off = th->th_off << 2;
762 if (off < sizeof (struct tcphdr) || off > tlen) {
763 TCPSTAT_INC(tcps_rcvbadoff);
766 tlen -= off; /* tlen is used instead of ti->ti_len */
767 if (off > sizeof (struct tcphdr)) {
770 if (m->m_len < off0 + off) {
771 m = m_pullup(m, off0 + off);
773 TCPSTAT_INC(tcps_rcvshort);
774 return (IPPROTO_DONE);
777 ip6 = mtod(m, struct ip6_hdr *);
778 th = (struct tcphdr *)((caddr_t)ip6 + off0);
781 #if defined(INET) && defined(INET6)
786 if (m->m_len < sizeof(struct ip) + off) {
787 if ((m = m_pullup(m, sizeof (struct ip) + off))
789 TCPSTAT_INC(tcps_rcvshort);
790 return (IPPROTO_DONE);
792 ip = mtod(m, struct ip *);
793 th = (struct tcphdr *)((caddr_t)ip + off0);
797 optlen = off - sizeof (struct tcphdr);
798 optp = (u_char *)(th + 1);
800 thflags = th->th_flags;
803 * Convert TCP protocol specific fields to host format.
805 tcp_fields_to_host(th);
808 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
810 drop_hdrlen = off0 + off;
813 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
817 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
819 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
822 #if defined(INET) && !defined(INET6)
823 (m->m_flags & M_IP_NEXTHOP)
826 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
830 if (isipv6 && fwd_tag != NULL) {
831 struct sockaddr_in6 *next_hop6;
833 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
835 * Transparently forwarded. Pretend to be the destination.
836 * Already got one like this?
838 inp = in6_pcblookup_mbuf(&V_tcbinfo,
839 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
840 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
843 * It's new. Try to find the ambushing socket.
844 * Because we've rewritten the destination address,
845 * any hardware-generated hash is ignored.
847 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
848 th->th_sport, &next_hop6->sin6_addr,
849 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
850 th->th_dport, INPLOOKUP_WILDCARD |
851 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
854 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
855 th->th_sport, &ip6->ip6_dst, th->th_dport,
856 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
857 m->m_pkthdr.rcvif, m);
860 #if defined(INET6) && defined(INET)
864 if (fwd_tag != NULL) {
865 struct sockaddr_in *next_hop;
867 next_hop = (struct sockaddr_in *)(fwd_tag+1);
869 * Transparently forwarded. Pretend to be the destination.
870 * already got one like this?
872 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
873 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
874 m->m_pkthdr.rcvif, m);
877 * It's new. Try to find the ambushing socket.
878 * Because we've rewritten the destination address,
879 * any hardware-generated hash is ignored.
881 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
882 th->th_sport, next_hop->sin_addr,
883 next_hop->sin_port ? ntohs(next_hop->sin_port) :
884 th->th_dport, INPLOOKUP_WILDCARD |
885 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
888 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
889 th->th_sport, ip->ip_dst, th->th_dport,
890 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
891 m->m_pkthdr.rcvif, m);
895 * If the INPCB does not exist then all data in the incoming
896 * segment is discarded and an appropriate RST is sent back.
897 * XXX MRT Send RST using which routing table?
901 * Log communication attempts to ports that are not
904 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
905 V_tcp_log_in_vain == 2) {
906 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
907 log(LOG_INFO, "%s; %s: Connection attempt "
908 "to closed port\n", s, __func__);
911 * When blackholing do not respond with a RST but
912 * completely ignore the segment and drop it.
914 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
918 rstreason = BANDLIM_RST_CLOSEDPORT;
921 INP_WLOCK_ASSERT(inp);
923 * While waiting for inp lock during the lookup, another thread
924 * can have dropped the inpcb, in which case we need to loop back
925 * and try to find a new inpcb to deliver to.
927 if (inp->inp_flags & INP_DROPPED) {
932 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
933 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
934 ((inp->inp_socket == NULL) ||
935 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
936 inp->inp_flowid = m->m_pkthdr.flowid;
937 inp->inp_flowtype = M_HASHTYPE_GET(m);
939 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
941 if (isipv6 && IPSEC_ENABLED(ipv6) &&
942 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
950 if (IPSEC_ENABLED(ipv4) &&
951 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
958 * Check the minimum TTL for socket.
960 if (inp->inp_ip_minttl != 0) {
963 if (inp->inp_ip_minttl > ip6->ip6_hlim)
967 if (inp->inp_ip_minttl > ip->ip_ttl)
972 * A previous connection in TIMEWAIT state is supposed to catch stray
973 * or duplicate segments arriving late. If this segment was a
974 * legitimate new connection attempt, the old INPCB gets removed and
975 * we can try again to find a listening socket.
977 * At this point, due to earlier optimism, we may hold only an inpcb
978 * lock, and not the inpcbinfo write lock. If so, we need to try to
979 * acquire it, or if that fails, acquire a reference on the inpcb,
980 * drop all locks, acquire a global write lock, and then re-acquire
981 * the inpcb lock. We may at that point discover that another thread
982 * has tried to free the inpcb, in which case we need to loop back
983 * and try to find a new inpcb to deliver to.
985 * XXXRW: It may be time to rethink timewait locking.
987 if (inp->inp_flags & INP_TIMEWAIT) {
988 if (thflags & TH_SYN)
989 tcp_dooptions(&to, optp, optlen, TO_SYN);
991 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
993 if (tcp_twcheck(inp, &to, th, m, tlen))
995 return (IPPROTO_DONE);
998 * The TCPCB may no longer exist if the connection is winding
999 * down or it is in the CLOSED state. Either way we drop the
1000 * segment and send an appropriate response.
1002 tp = intotcpcb(inp);
1003 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1004 rstreason = BANDLIM_RST_CLOSEDPORT;
1009 if (tp->t_flags & TF_TOE) {
1010 tcp_offload_input(tp, m);
1011 m = NULL; /* consumed by the TOE driver */
1017 INP_WLOCK_ASSERT(inp);
1018 if (mac_inpcb_check_deliver(inp, m))
1021 so = inp->inp_socket;
1022 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1024 if (so->so_options & SO_DEBUG) {
1025 ostate = tp->t_state;
1028 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1031 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1034 #endif /* TCPDEBUG */
1036 * When the socket is accepting connections (the INPCB is in LISTEN
1037 * state) we look into the SYN cache if this is a new connection
1038 * attempt or the completion of a previous one.
1040 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1041 ("%s: so accepting but tp %p not listening", __func__, tp));
1042 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1043 struct in_conninfo inc;
1045 bzero(&inc, sizeof(inc));
1048 inc.inc_flags |= INC_ISIPV6;
1049 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1050 inc.inc_flags |= INC_IPV6MINMTU;
1051 inc.inc6_faddr = ip6->ip6_src;
1052 inc.inc6_laddr = ip6->ip6_dst;
1056 inc.inc_faddr = ip->ip_src;
1057 inc.inc_laddr = ip->ip_dst;
1059 inc.inc_fport = th->th_sport;
1060 inc.inc_lport = th->th_dport;
1061 inc.inc_fibnum = so->so_fibnum;
1064 * Check for an existing connection attempt in syncache if
1065 * the flag is only ACK. A successful lookup creates a new
1066 * socket appended to the listen queue in SYN_RECEIVED state.
1068 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1071 * Parse the TCP options here because
1072 * syncookies need access to the reflected
1075 tcp_dooptions(&to, optp, optlen, 0);
1077 * NB: syncache_expand() doesn't unlock
1078 * inp and tcpinfo locks.
1080 rstreason = syncache_expand(&inc, &to, th, &so, m);
1081 if (rstreason < 0) {
1083 * A failing TCP MD5 signature comparison
1084 * must result in the segment being dropped
1085 * and must not produce any response back
1089 } else if (rstreason == 0) {
1091 * No syncache entry or ACK was not
1092 * for our SYN/ACK. Send a RST.
1093 * NB: syncache did its own logging
1094 * of the failure cause.
1096 rstreason = BANDLIM_RST_OPENPORT;
1102 * We completed the 3-way handshake
1103 * but could not allocate a socket
1104 * either due to memory shortage,
1105 * listen queue length limits or
1106 * global socket limits. Send RST
1107 * or wait and have the remote end
1108 * retransmit the ACK for another
1111 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1112 log(LOG_DEBUG, "%s; %s: Listen socket: "
1113 "Socket allocation failed due to "
1114 "limits or memory shortage, %s\n",
1116 V_tcp_sc_rst_sock_fail ?
1117 "sending RST" : "try again");
1118 if (V_tcp_sc_rst_sock_fail) {
1119 rstreason = BANDLIM_UNLIMITED;
1125 * Socket is created in state SYN_RECEIVED.
1126 * Unlock the listen socket, lock the newly
1127 * created socket and update the tp variable.
1129 INP_WUNLOCK(inp); /* listen socket */
1130 inp = sotoinpcb(so);
1132 * New connection inpcb is already locked by
1133 * syncache_expand().
1135 INP_WLOCK_ASSERT(inp);
1136 tp = intotcpcb(inp);
1137 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1138 ("%s: ", __func__));
1140 * Process the segment and the data it
1141 * contains. tcp_do_segment() consumes
1142 * the mbuf chain and unlocks the inpcb.
1144 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1145 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1147 return (IPPROTO_DONE);
1150 * Segment flag validation for new connection attempts:
1152 * Our (SYN|ACK) response was rejected.
1153 * Check with syncache and remove entry to prevent
1156 * NB: syncache_chkrst does its own logging of failure
1159 if (thflags & TH_RST) {
1160 syncache_chkrst(&inc, th, m);
1164 * We can't do anything without SYN.
1166 if ((thflags & TH_SYN) == 0) {
1167 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1168 log(LOG_DEBUG, "%s; %s: Listen socket: "
1169 "SYN is missing, segment ignored\n",
1171 TCPSTAT_INC(tcps_badsyn);
1175 * (SYN|ACK) is bogus on a listen socket.
1177 if (thflags & TH_ACK) {
1178 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1179 log(LOG_DEBUG, "%s; %s: Listen socket: "
1180 "SYN|ACK invalid, segment rejected\n",
1182 syncache_badack(&inc); /* XXX: Not needed! */
1183 TCPSTAT_INC(tcps_badsyn);
1184 rstreason = BANDLIM_RST_OPENPORT;
1188 * If the drop_synfin option is enabled, drop all
1189 * segments with both the SYN and FIN bits set.
1190 * This prevents e.g. nmap from identifying the
1192 * XXX: Poor reasoning. nmap has other methods
1193 * and is constantly refining its stack detection
1195 * XXX: This is a violation of the TCP specification
1196 * and was used by RFC1644.
1198 if ((thflags & TH_FIN) && V_drop_synfin) {
1199 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1200 log(LOG_DEBUG, "%s; %s: Listen socket: "
1201 "SYN|FIN segment ignored (based on "
1202 "sysctl setting)\n", s, __func__);
1203 TCPSTAT_INC(tcps_badsyn);
1207 * Segment's flags are (SYN) or (SYN|FIN).
1209 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1210 * as they do not affect the state of the TCP FSM.
1211 * The data pointed to by TH_URG and th_urp is ignored.
1213 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1214 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1215 KASSERT(thflags & (TH_SYN),
1216 ("%s: Listen socket: TH_SYN not set", __func__));
1219 * If deprecated address is forbidden,
1220 * we do not accept SYN to deprecated interface
1221 * address to prevent any new inbound connection from
1222 * getting established.
1223 * When we do not accept SYN, we send a TCP RST,
1224 * with deprecated source address (instead of dropping
1225 * it). We compromise it as it is much better for peer
1226 * to send a RST, and RST will be the final packet
1229 * If we do not forbid deprecated addresses, we accept
1230 * the SYN packet. RFC2462 does not suggest dropping
1232 * If we decipher RFC2462 5.5.4, it says like this:
1233 * 1. use of deprecated addr with existing
1234 * communication is okay - "SHOULD continue to be
1236 * 2. use of it with new communication:
1237 * (2a) "SHOULD NOT be used if alternate address
1238 * with sufficient scope is available"
1239 * (2b) nothing mentioned otherwise.
1240 * Here we fall into (2b) case as we have no choice in
1241 * our source address selection - we must obey the peer.
1243 * The wording in RFC2462 is confusing, and there are
1244 * multiple description text for deprecated address
1245 * handling - worse, they are not exactly the same.
1246 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1248 if (isipv6 && !V_ip6_use_deprecated) {
1249 struct in6_ifaddr *ia6;
1251 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1253 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1254 ifa_free(&ia6->ia_ifa);
1255 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1256 log(LOG_DEBUG, "%s; %s: Listen socket: "
1257 "Connection attempt to deprecated "
1258 "IPv6 address rejected\n",
1260 rstreason = BANDLIM_RST_OPENPORT;
1264 ifa_free(&ia6->ia_ifa);
1268 * Basic sanity checks on incoming SYN requests:
1269 * Don't respond if the destination is a link layer
1270 * broadcast according to RFC1122 4.2.3.10, p. 104.
1271 * If it is from this socket it must be forged.
1272 * Don't respond if the source or destination is a
1273 * global or subnet broad- or multicast address.
1274 * Note that it is quite possible to receive unicast
1275 * link-layer packets with a broadcast IP address. Use
1276 * in_broadcast() to find them.
1278 if (m->m_flags & (M_BCAST|M_MCAST)) {
1279 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1280 log(LOG_DEBUG, "%s; %s: Listen socket: "
1281 "Connection attempt from broad- or multicast "
1282 "link layer address ignored\n", s, __func__);
1287 if (th->th_dport == th->th_sport &&
1288 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1289 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1290 log(LOG_DEBUG, "%s; %s: Listen socket: "
1291 "Connection attempt to/from self "
1292 "ignored\n", s, __func__);
1295 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1296 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1297 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1298 log(LOG_DEBUG, "%s; %s: Listen socket: "
1299 "Connection attempt from/to multicast "
1300 "address ignored\n", s, __func__);
1305 #if defined(INET) && defined(INET6)
1310 if (th->th_dport == th->th_sport &&
1311 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1312 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1313 log(LOG_DEBUG, "%s; %s: Listen socket: "
1314 "Connection attempt from/to self "
1315 "ignored\n", s, __func__);
1318 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1319 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1320 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1321 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1322 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1323 log(LOG_DEBUG, "%s; %s: Listen socket: "
1324 "Connection attempt from/to broad- "
1325 "or multicast address ignored\n",
1332 * SYN appears to be valid. Create compressed TCP state
1336 if (so->so_options & SO_DEBUG)
1337 tcp_trace(TA_INPUT, ostate, tp,
1338 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1340 TCP_PROBE3(debug__input, tp, th, m);
1341 tcp_dooptions(&to, optp, optlen, TO_SYN);
1342 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos))
1343 goto tfo_socket_result;
1346 * Entry added to syncache and mbuf consumed.
1347 * Only the listen socket is unlocked by syncache_add().
1349 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1350 return (IPPROTO_DONE);
1351 } else if (tp->t_state == TCPS_LISTEN) {
1353 * When a listen socket is torn down the SO_ACCEPTCONN
1354 * flag is removed first while connections are drained
1355 * from the accept queue in a unlock/lock cycle of the
1356 * ACCEPT_LOCK, opening a race condition allowing a SYN
1357 * attempt go through unhandled.
1361 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1362 if (tp->t_flags & TF_SIGNATURE) {
1363 tcp_dooptions(&to, optp, optlen, thflags);
1364 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1365 TCPSTAT_INC(tcps_sig_err_nosigopt);
1368 if (!TCPMD5_ENABLED() ||
1369 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1373 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1376 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1377 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1378 * the inpcb, and unlocks pcbinfo.
1380 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1381 return (IPPROTO_DONE);
1384 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1387 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1390 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1391 m = NULL; /* mbuf chain got consumed. */
1396 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1402 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1407 return (IPPROTO_DONE);
1411 * Automatic sizing of receive socket buffer. Often the send
1412 * buffer size is not optimally adjusted to the actual network
1413 * conditions at hand (delay bandwidth product). Setting the
1414 * buffer size too small limits throughput on links with high
1415 * bandwidth and high delay (eg. trans-continental/oceanic links).
1417 * On the receive side the socket buffer memory is only rarely
1418 * used to any significant extent. This allows us to be much
1419 * more aggressive in scaling the receive socket buffer. For
1420 * the case that the buffer space is actually used to a large
1421 * extent and we run out of kernel memory we can simply drop
1422 * the new segments; TCP on the sender will just retransmit it
1423 * later. Setting the buffer size too big may only consume too
1424 * much kernel memory if the application doesn't read() from
1425 * the socket or packet loss or reordering makes use of the
1428 * The criteria to step up the receive buffer one notch are:
1429 * 1. Application has not set receive buffer size with
1430 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1431 * 2. the number of bytes received during 1/2 of an sRTT
1432 * is at least 3/8 of the current socket buffer size.
1433 * 3. receive buffer size has not hit maximal automatic size;
1435 * If all of the criteria are met we increaset the socket buffer
1436 * by a 1/2 (bounded by the max). This allows us to keep ahead
1437 * of slow-start but also makes it so our peer never gets limited
1438 * by our rwnd which we then open up causing a burst.
1440 * This algorithm does two steps per RTT at most and only if
1441 * we receive a bulk stream w/o packet losses or reorderings.
1442 * Shrinking the buffer during idle times is not necessary as
1443 * it doesn't consume any memory when idle.
1445 * TODO: Only step up if the application is actually serving
1446 * the buffer to better manage the socket buffer resources.
1449 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1450 struct tcpcb *tp, int tlen)
1454 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1455 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1456 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1457 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1458 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1459 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1460 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1462 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1464 /* Start over with next RTT. */
1468 tp->rfbuf_cnt += tlen; /* add up */
1474 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1475 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1477 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1478 int rstreason, todrop, win, incforsyn = 0;
1482 struct in_conninfo *inc;
1489 * The size of tcp_saveipgen must be the size of the max ip header,
1492 u_char tcp_saveipgen[IP6_HDR_LEN];
1493 struct tcphdr tcp_savetcp;
1496 thflags = th->th_flags;
1497 inc = &tp->t_inpcb->inp_inc;
1498 tp->sackhint.last_sack_ack = 0;
1500 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1503 INP_WLOCK_ASSERT(tp->t_inpcb);
1504 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1506 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1510 /* Save segment, if requested. */
1511 tcp_pcap_add(th, m, &(tp->t_inpkts));
1513 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1516 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1517 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1518 log(LOG_DEBUG, "%s; %s: "
1519 "SYN|FIN segment ignored (based on "
1520 "sysctl setting)\n", s, __func__);
1527 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1528 * check SEQ.ACK first.
1530 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1531 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1532 rstreason = BANDLIM_UNLIMITED;
1537 * Segment received on connection.
1538 * Reset idle time and keep-alive timer.
1539 * XXX: This should be done after segment
1540 * validation to ignore broken/spoofed segs.
1542 tp->t_rcvtime = ticks;
1545 * Scale up the window into a 32-bit value.
1546 * For the SYN_SENT state the scale is zero.
1548 tiwin = th->th_win << tp->snd_scale;
1550 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1554 * TCP ECN processing.
1556 if (tp->t_flags2 & TF2_ECN_PERMIT) {
1557 if (thflags & TH_CWR) {
1558 tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1559 tp->t_flags |= TF_ACKNOW;
1561 switch (iptos & IPTOS_ECN_MASK) {
1563 tp->t_flags2 |= TF2_ECN_SND_ECE;
1564 TCPSTAT_INC(tcps_ecn_ce);
1566 case IPTOS_ECN_ECT0:
1567 TCPSTAT_INC(tcps_ecn_ect0);
1569 case IPTOS_ECN_ECT1:
1570 TCPSTAT_INC(tcps_ecn_ect1);
1574 /* Process a packet differently from RFC3168. */
1575 cc_ecnpkt_handler(tp, th, iptos);
1577 /* Congestion experienced. */
1578 if (thflags & TH_ECE) {
1579 cc_cong_signal(tp, th, CC_ECN);
1584 * Parse options on any incoming segment.
1586 tcp_dooptions(&to, (u_char *)(th + 1),
1587 (th->th_off << 2) - sizeof(struct tcphdr),
1588 (thflags & TH_SYN) ? TO_SYN : 0);
1590 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1591 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1592 (to.to_flags & TOF_SIGNATURE) == 0) {
1593 TCPSTAT_INC(tcps_sig_err_sigopt);
1594 /* XXX: should drop? */
1598 * If echoed timestamp is later than the current time,
1599 * fall back to non RFC1323 RTT calculation. Normalize
1600 * timestamp if syncookies were used when this connection
1603 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1604 to.to_tsecr -= tp->ts_offset;
1605 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1607 else if (tp->t_flags & TF_PREVVALID &&
1608 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1609 cc_cong_signal(tp, th, CC_RTO_ERR);
1612 * Process options only when we get SYN/ACK back. The SYN case
1613 * for incoming connections is handled in tcp_syncache.
1614 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1615 * or <SYN,ACK>) segment itself is never scaled.
1616 * XXX this is traditional behavior, may need to be cleaned up.
1618 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1619 /* Handle parallel SYN for ECN */
1620 if (!(thflags & TH_ACK) &&
1621 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1622 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1623 tp->t_flags2 |= TF2_ECN_PERMIT;
1624 tp->t_flags2 |= TF2_ECN_SND_ECE;
1625 TCPSTAT_INC(tcps_ecn_shs);
1627 if ((to.to_flags & TOF_SCALE) &&
1628 (tp->t_flags & TF_REQ_SCALE)) {
1629 tp->t_flags |= TF_RCVD_SCALE;
1630 tp->snd_scale = to.to_wscale;
1632 tp->t_flags &= ~TF_REQ_SCALE;
1634 * Initial send window. It will be updated with
1635 * the next incoming segment to the scaled value.
1637 tp->snd_wnd = th->th_win;
1638 if ((to.to_flags & TOF_TS) &&
1639 (tp->t_flags & TF_REQ_TSTMP)) {
1640 tp->t_flags |= TF_RCVD_TSTMP;
1641 tp->ts_recent = to.to_tsval;
1642 tp->ts_recent_age = tcp_ts_getticks();
1644 tp->t_flags &= ~TF_REQ_TSTMP;
1645 if (to.to_flags & TOF_MSS)
1646 tcp_mss(tp, to.to_mss);
1647 if ((tp->t_flags & TF_SACK_PERMIT) &&
1648 (to.to_flags & TOF_SACKPERM) == 0)
1649 tp->t_flags &= ~TF_SACK_PERMIT;
1650 if (IS_FASTOPEN(tp->t_flags)) {
1651 if (to.to_flags & TOF_FASTOPEN) {
1654 if (to.to_flags & TOF_MSS)
1657 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1661 tcp_fastopen_update_cache(tp, mss,
1662 to.to_tfo_len, to.to_tfo_cookie);
1664 tcp_fastopen_disable_path(tp);
1669 * If timestamps were negotiated during SYN/ACK they should
1670 * appear on every segment during this session and vice versa.
1672 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1673 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1674 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1675 "no action\n", s, __func__);
1679 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1680 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1681 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1682 "no action\n", s, __func__);
1688 * Header prediction: check for the two common cases
1689 * of a uni-directional data xfer. If the packet has
1690 * no control flags, is in-sequence, the window didn't
1691 * change and we're not retransmitting, it's a
1692 * candidate. If the length is zero and the ack moved
1693 * forward, we're the sender side of the xfer. Just
1694 * free the data acked & wake any higher level process
1695 * that was blocked waiting for space. If the length
1696 * is non-zero and the ack didn't move, we're the
1697 * receiver side. If we're getting packets in-order
1698 * (the reassembly queue is empty), add the data to
1699 * the socket buffer and note that we need a delayed ack.
1700 * Make sure that the hidden state-flags are also off.
1701 * Since we check for TCPS_ESTABLISHED first, it can only
1704 if (tp->t_state == TCPS_ESTABLISHED &&
1705 th->th_seq == tp->rcv_nxt &&
1706 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1707 tp->snd_nxt == tp->snd_max &&
1708 tiwin && tiwin == tp->snd_wnd &&
1709 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1711 ((to.to_flags & TOF_TS) == 0 ||
1712 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1715 * If last ACK falls within this segment's sequence numbers,
1716 * record the timestamp.
1717 * NOTE that the test is modified according to the latest
1718 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1720 if ((to.to_flags & TOF_TS) != 0 &&
1721 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1722 tp->ts_recent_age = tcp_ts_getticks();
1723 tp->ts_recent = to.to_tsval;
1727 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1728 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1729 !IN_RECOVERY(tp->t_flags) &&
1730 (to.to_flags & TOF_SACK) == 0 &&
1731 TAILQ_EMPTY(&tp->snd_holes)) {
1733 * This is a pure ack for outstanding data.
1735 TCPSTAT_INC(tcps_predack);
1738 * "bad retransmit" recovery without timestamps.
1740 if ((to.to_flags & TOF_TS) == 0 &&
1741 tp->t_rxtshift == 1 &&
1742 tp->t_flags & TF_PREVVALID &&
1743 (int)(ticks - tp->t_badrxtwin) < 0) {
1744 cc_cong_signal(tp, th, CC_RTO_ERR);
1748 * Recalculate the transmit timer / rtt.
1750 * Some boxes send broken timestamp replies
1751 * during the SYN+ACK phase, ignore
1752 * timestamps of 0 or we could calculate a
1753 * huge RTT and blow up the retransmit timer.
1755 if ((to.to_flags & TOF_TS) != 0 &&
1759 t = tcp_ts_getticks() - to.to_tsecr;
1760 if (!tp->t_rttlow || tp->t_rttlow > t)
1763 TCP_TS_TO_TICKS(t) + 1);
1764 } else if (tp->t_rtttime &&
1765 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1766 if (!tp->t_rttlow ||
1767 tp->t_rttlow > ticks - tp->t_rtttime)
1768 tp->t_rttlow = ticks - tp->t_rtttime;
1770 ticks - tp->t_rtttime);
1772 acked = BYTES_THIS_ACK(tp, th);
1775 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1776 hhook_run_tcp_est_in(tp, th, &to);
1779 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1780 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1781 sbdrop(&so->so_snd, acked);
1782 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1783 SEQ_LEQ(th->th_ack, tp->snd_recover))
1784 tp->snd_recover = th->th_ack - 1;
1787 * Let the congestion control algorithm update
1788 * congestion control related information. This
1789 * typically means increasing the congestion
1792 cc_ack_received(tp, th, nsegs, CC_ACK);
1794 tp->snd_una = th->th_ack;
1796 * Pull snd_wl2 up to prevent seq wrap relative
1799 tp->snd_wl2 = th->th_ack;
1804 * If all outstanding data are acked, stop
1805 * retransmit timer, otherwise restart timer
1806 * using current (possibly backed-off) value.
1807 * If process is waiting for space,
1808 * wakeup/selwakeup/signal. If data
1809 * are ready to send, let tcp_output
1810 * decide between more output or persist.
1813 if (so->so_options & SO_DEBUG)
1814 tcp_trace(TA_INPUT, ostate, tp,
1815 (void *)tcp_saveipgen,
1818 TCP_PROBE3(debug__input, tp, th, m);
1819 if (tp->snd_una == tp->snd_max)
1820 tcp_timer_activate(tp, TT_REXMT, 0);
1821 else if (!tcp_timer_active(tp, TT_PERSIST))
1822 tcp_timer_activate(tp, TT_REXMT,
1825 if (sbavail(&so->so_snd))
1826 (void) tp->t_fb->tfb_tcp_output(tp);
1829 } else if (th->th_ack == tp->snd_una &&
1830 tlen <= sbspace(&so->so_rcv)) {
1831 int newsize = 0; /* automatic sockbuf scaling */
1834 * This is a pure, in-sequence data packet with
1835 * nothing on the reassembly queue and we have enough
1836 * buffer space to take it.
1838 /* Clean receiver SACK report if present */
1839 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1840 tcp_clean_sackreport(tp);
1841 TCPSTAT_INC(tcps_preddat);
1842 tp->rcv_nxt += tlen;
1844 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1845 (tp->t_fbyte_in == 0)) {
1846 tp->t_fbyte_in = ticks;
1847 if (tp->t_fbyte_in == 0)
1849 if (tp->t_fbyte_out && tp->t_fbyte_in)
1850 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1853 * Pull snd_wl1 up to prevent seq wrap relative to
1856 tp->snd_wl1 = th->th_seq;
1858 * Pull rcv_up up to prevent seq wrap relative to
1861 tp->rcv_up = tp->rcv_nxt;
1862 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1863 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1865 if (so->so_options & SO_DEBUG)
1866 tcp_trace(TA_INPUT, ostate, tp,
1867 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1869 TCP_PROBE3(debug__input, tp, th, m);
1871 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1873 /* Add data to socket buffer. */
1874 SOCKBUF_LOCK(&so->so_rcv);
1875 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1879 * Set new socket buffer size.
1880 * Give up when limit is reached.
1883 if (!sbreserve_locked(&so->so_rcv,
1885 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1886 m_adj(m, drop_hdrlen); /* delayed header drop */
1887 sbappendstream_locked(&so->so_rcv, m, 0);
1889 /* NB: sorwakeup_locked() does an implicit unlock. */
1890 sorwakeup_locked(so);
1891 if (DELAY_ACK(tp, tlen)) {
1892 tp->t_flags |= TF_DELACK;
1894 tp->t_flags |= TF_ACKNOW;
1895 tp->t_fb->tfb_tcp_output(tp);
1902 * Calculate amount of space in receive window,
1903 * and then do TCP input processing.
1904 * Receive window is amount of space in rcv queue,
1905 * but not less than advertised window.
1907 win = sbspace(&so->so_rcv);
1910 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1912 switch (tp->t_state) {
1915 * If the state is SYN_RECEIVED:
1916 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1918 case TCPS_SYN_RECEIVED:
1919 if ((thflags & TH_ACK) &&
1920 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1921 SEQ_GT(th->th_ack, tp->snd_max))) {
1922 rstreason = BANDLIM_RST_OPENPORT;
1925 if (IS_FASTOPEN(tp->t_flags)) {
1927 * When a TFO connection is in SYN_RECEIVED, the
1928 * only valid packets are the initial SYN, a
1929 * retransmit/copy of the initial SYN (possibly with
1930 * a subset of the original data), a valid ACK, a
1933 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1934 rstreason = BANDLIM_RST_OPENPORT;
1936 } else if (thflags & TH_SYN) {
1937 /* non-initial SYN is ignored */
1938 if ((tcp_timer_active(tp, TT_DELACK) ||
1939 tcp_timer_active(tp, TT_REXMT)))
1941 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1948 * If the state is SYN_SENT:
1949 * if seg contains a RST with valid ACK (SEQ.ACK has already
1950 * been verified), then drop the connection.
1951 * if seg contains a RST without an ACK, drop the seg.
1952 * if seg does not contain SYN, then drop the seg.
1953 * Otherwise this is an acceptable SYN segment
1954 * initialize tp->rcv_nxt and tp->irs
1955 * if seg contains ack then advance tp->snd_una
1956 * if seg contains an ECE and ECN support is enabled, the stream
1958 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1959 * arrange for segment to be acked (eventually)
1960 * continue processing rest of data/controls, beginning with URG
1963 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1964 TCP_PROBE5(connect__refused, NULL, tp,
1966 tp = tcp_drop(tp, ECONNREFUSED);
1968 if (thflags & TH_RST)
1970 if (!(thflags & TH_SYN))
1973 tp->irs = th->th_seq;
1975 if (thflags & TH_ACK) {
1976 int tfo_partial_ack = 0;
1978 TCPSTAT_INC(tcps_connects);
1981 mac_socketpeer_set_from_mbuf(m, so);
1983 /* Do window scaling on this connection? */
1984 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1985 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1986 tp->rcv_scale = tp->request_r_scale;
1988 tp->rcv_adv += min(tp->rcv_wnd,
1989 TCP_MAXWIN << tp->rcv_scale);
1990 tp->snd_una++; /* SYN is acked */
1992 * If not all the data that was sent in the TFO SYN
1993 * has been acked, resend the remainder right away.
1995 if (IS_FASTOPEN(tp->t_flags) &&
1996 (tp->snd_una != tp->snd_max)) {
1997 tp->snd_nxt = th->th_ack;
1998 tfo_partial_ack = 1;
2001 * If there's data, delay ACK; if there's also a FIN
2002 * ACKNOW will be turned on later.
2004 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2005 tcp_timer_activate(tp, TT_DELACK,
2008 tp->t_flags |= TF_ACKNOW;
2010 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2011 (V_tcp_do_ecn == 1)) {
2012 tp->t_flags2 |= TF2_ECN_PERMIT;
2013 TCPSTAT_INC(tcps_ecn_shs);
2017 * Received <SYN,ACK> in SYN_SENT[*] state.
2019 * SYN_SENT --> ESTABLISHED
2020 * SYN_SENT* --> FIN_WAIT_1
2022 tp->t_starttime = ticks;
2023 if (tp->t_flags & TF_NEEDFIN) {
2024 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2025 tp->t_flags &= ~TF_NEEDFIN;
2028 tcp_state_change(tp, TCPS_ESTABLISHED);
2029 TCP_PROBE5(connect__established, NULL, tp,
2032 tcp_timer_activate(tp, TT_KEEP,
2037 * Received initial SYN in SYN-SENT[*] state =>
2038 * simultaneous open.
2039 * If it succeeds, connection is * half-synchronized.
2040 * Otherwise, do 3-way handshake:
2041 * SYN-SENT -> SYN-RECEIVED
2042 * SYN-SENT* -> SYN-RECEIVED*
2044 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2045 tcp_timer_activate(tp, TT_REXMT, 0);
2046 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2049 INP_WLOCK_ASSERT(tp->t_inpcb);
2052 * Advance th->th_seq to correspond to first data byte.
2053 * If data, trim to stay within window,
2054 * dropping FIN if necessary.
2057 if (tlen > tp->rcv_wnd) {
2058 todrop = tlen - tp->rcv_wnd;
2062 TCPSTAT_INC(tcps_rcvpackafterwin);
2063 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2065 tp->snd_wl1 = th->th_seq - 1;
2066 tp->rcv_up = th->th_seq;
2068 * Client side of transaction: already sent SYN and data.
2069 * If the remote host used T/TCP to validate the SYN,
2070 * our data will be ACK'd; if so, enter normal data segment
2071 * processing in the middle of step 5, ack processing.
2072 * Otherwise, goto step 6.
2074 if (thflags & TH_ACK)
2080 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2081 * do normal processing.
2083 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2087 break; /* continue normal processing */
2091 * States other than LISTEN or SYN_SENT.
2092 * First check the RST flag and sequence number since reset segments
2093 * are exempt from the timestamp and connection count tests. This
2094 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2095 * below which allowed reset segments in half the sequence space
2096 * to fall though and be processed (which gives forged reset
2097 * segments with a random sequence number a 50 percent chance of
2098 * killing a connection).
2099 * Then check timestamp, if present.
2100 * Then check the connection count, if present.
2101 * Then check that at least some bytes of segment are within
2102 * receive window. If segment begins before rcv_nxt,
2103 * drop leading data (and SYN); if nothing left, just ack.
2105 if (thflags & TH_RST) {
2107 * RFC5961 Section 3.2
2109 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2110 * - If RST is in window, we send challenge ACK.
2112 * Note: to take into account delayed ACKs, we should
2113 * test against last_ack_sent instead of rcv_nxt.
2114 * Note 2: we handle special case of closed window, not
2115 * covered by the RFC.
2117 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2118 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2119 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2121 KASSERT(tp->t_state != TCPS_SYN_SENT,
2122 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2125 if (V_tcp_insecure_rst ||
2126 tp->last_ack_sent == th->th_seq) {
2127 TCPSTAT_INC(tcps_drops);
2128 /* Drop the connection. */
2129 switch (tp->t_state) {
2130 case TCPS_SYN_RECEIVED:
2131 so->so_error = ECONNREFUSED;
2133 case TCPS_ESTABLISHED:
2134 case TCPS_FIN_WAIT_1:
2135 case TCPS_FIN_WAIT_2:
2136 case TCPS_CLOSE_WAIT:
2139 so->so_error = ECONNRESET;
2146 TCPSTAT_INC(tcps_badrst);
2147 /* Send challenge ACK. */
2148 tcp_respond(tp, mtod(m, void *), th, m,
2149 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2150 tp->last_ack_sent = tp->rcv_nxt;
2158 * RFC5961 Section 4.2
2159 * Send challenge ACK for any SYN in synchronized state.
2161 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2162 tp->t_state != TCPS_SYN_RECEIVED) {
2163 TCPSTAT_INC(tcps_badsyn);
2164 if (V_tcp_insecure_syn &&
2165 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2166 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2167 tp = tcp_drop(tp, ECONNRESET);
2168 rstreason = BANDLIM_UNLIMITED;
2170 /* Send challenge ACK. */
2171 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2172 tp->snd_nxt, TH_ACK);
2173 tp->last_ack_sent = tp->rcv_nxt;
2180 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2181 * and it's less than ts_recent, drop it.
2183 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2184 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2186 /* Check to see if ts_recent is over 24 days old. */
2187 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2189 * Invalidate ts_recent. If this segment updates
2190 * ts_recent, the age will be reset later and ts_recent
2191 * will get a valid value. If it does not, setting
2192 * ts_recent to zero will at least satisfy the
2193 * requirement that zero be placed in the timestamp
2194 * echo reply when ts_recent isn't valid. The
2195 * age isn't reset until we get a valid ts_recent
2196 * because we don't want out-of-order segments to be
2197 * dropped when ts_recent is old.
2201 TCPSTAT_INC(tcps_rcvduppack);
2202 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2203 TCPSTAT_INC(tcps_pawsdrop);
2211 * In the SYN-RECEIVED state, validate that the packet belongs to
2212 * this connection before trimming the data to fit the receive
2213 * window. Check the sequence number versus IRS since we know
2214 * the sequence numbers haven't wrapped. This is a partial fix
2215 * for the "LAND" DoS attack.
2217 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2218 rstreason = BANDLIM_RST_OPENPORT;
2222 todrop = tp->rcv_nxt - th->th_seq;
2224 if (thflags & TH_SYN) {
2234 * Following if statement from Stevens, vol. 2, p. 960.
2237 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2239 * Any valid FIN must be to the left of the window.
2240 * At this point the FIN must be a duplicate or out
2241 * of sequence; drop it.
2246 * Send an ACK to resynchronize and drop any data.
2247 * But keep on processing for RST or ACK.
2249 tp->t_flags |= TF_ACKNOW;
2251 TCPSTAT_INC(tcps_rcvduppack);
2252 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2254 TCPSTAT_INC(tcps_rcvpartduppack);
2255 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2258 * DSACK - add SACK block for dropped range
2260 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2261 tcp_update_sack_list(tp, th->th_seq,
2262 th->th_seq + todrop);
2264 * ACK now, as the next in-sequence segment
2265 * will clear the DSACK block again
2267 tp->t_flags |= TF_ACKNOW;
2269 drop_hdrlen += todrop; /* drop from the top afterwards */
2270 th->th_seq += todrop;
2272 if (th->th_urp > todrop)
2273 th->th_urp -= todrop;
2281 * If new data are received on a connection after the
2282 * user processes are gone, then RST the other end.
2284 if ((so->so_state & SS_NOFDREF) &&
2285 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2286 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2287 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2288 "after socket was closed, "
2289 "sending RST and removing tcpcb\n",
2290 s, __func__, tcpstates[tp->t_state], tlen);
2294 TCPSTAT_INC(tcps_rcvafterclose);
2295 rstreason = BANDLIM_UNLIMITED;
2300 * If segment ends after window, drop trailing data
2301 * (and PUSH and FIN); if nothing left, just ACK.
2303 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2305 TCPSTAT_INC(tcps_rcvpackafterwin);
2306 if (todrop >= tlen) {
2307 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2309 * If window is closed can only take segments at
2310 * window edge, and have to drop data and PUSH from
2311 * incoming segments. Continue processing, but
2312 * remember to ack. Otherwise, drop segment
2315 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2316 tp->t_flags |= TF_ACKNOW;
2317 TCPSTAT_INC(tcps_rcvwinprobe);
2321 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2324 thflags &= ~(TH_PUSH|TH_FIN);
2328 * If last ACK falls within this segment's sequence numbers,
2329 * record its timestamp.
2331 * 1) That the test incorporates suggestions from the latest
2332 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2333 * 2) That updating only on newer timestamps interferes with
2334 * our earlier PAWS tests, so this check should be solely
2335 * predicated on the sequence space of this segment.
2336 * 3) That we modify the segment boundary check to be
2337 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2338 * instead of RFC1323's
2339 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2340 * This modified check allows us to overcome RFC1323's
2341 * limitations as described in Stevens TCP/IP Illustrated
2342 * Vol. 2 p.869. In such cases, we can still calculate the
2343 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2345 if ((to.to_flags & TOF_TS) != 0 &&
2346 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2347 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2348 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2349 tp->ts_recent_age = tcp_ts_getticks();
2350 tp->ts_recent = to.to_tsval;
2354 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2355 * flag is on (half-synchronized state), then queue data for
2356 * later processing; else drop segment and return.
2358 if ((thflags & TH_ACK) == 0) {
2359 if (tp->t_state == TCPS_SYN_RECEIVED ||
2360 (tp->t_flags & TF_NEEDSYN)) {
2361 if (tp->t_state == TCPS_SYN_RECEIVED &&
2362 IS_FASTOPEN(tp->t_flags)) {
2363 tp->snd_wnd = tiwin;
2367 } else if (tp->t_flags & TF_ACKNOW)
2376 switch (tp->t_state) {
2379 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2380 * ESTABLISHED state and continue processing.
2381 * The ACK was checked above.
2383 case TCPS_SYN_RECEIVED:
2385 TCPSTAT_INC(tcps_connects);
2387 /* Do window scaling? */
2388 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2389 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2390 tp->rcv_scale = tp->request_r_scale;
2392 tp->snd_wnd = tiwin;
2395 * SYN-RECEIVED -> ESTABLISHED
2396 * SYN-RECEIVED* -> FIN-WAIT-1
2398 tp->t_starttime = ticks;
2399 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2400 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2401 tp->t_tfo_pending = NULL;
2403 if (tp->t_flags & TF_NEEDFIN) {
2404 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2405 tp->t_flags &= ~TF_NEEDFIN;
2407 tcp_state_change(tp, TCPS_ESTABLISHED);
2408 TCP_PROBE5(accept__established, NULL, tp,
2411 * TFO connections call cc_conn_init() during SYN
2412 * processing. Calling it again here for such
2413 * connections is not harmless as it would undo the
2414 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2417 if (!IS_FASTOPEN(tp->t_flags))
2419 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2422 * Account for the ACK of our SYN prior to
2423 * regular ACK processing below, except for
2424 * simultaneous SYN, which is handled later.
2426 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2429 * If segment contains data or ACK, will call tcp_reass()
2430 * later; if not, do so now to pass queued data to user.
2432 if (tlen == 0 && (thflags & TH_FIN) == 0)
2433 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2435 tp->snd_wl1 = th->th_seq - 1;
2439 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2440 * ACKs. If the ack is in the range
2441 * tp->snd_una < th->th_ack <= tp->snd_max
2442 * then advance tp->snd_una to th->th_ack and drop
2443 * data from the retransmission queue. If this ACK reflects
2444 * more up to date window information we update our window information.
2446 case TCPS_ESTABLISHED:
2447 case TCPS_FIN_WAIT_1:
2448 case TCPS_FIN_WAIT_2:
2449 case TCPS_CLOSE_WAIT:
2452 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2453 TCPSTAT_INC(tcps_rcvacktoomuch);
2456 if ((tp->t_flags & TF_SACK_PERMIT) &&
2457 ((to.to_flags & TOF_SACK) ||
2458 !TAILQ_EMPTY(&tp->snd_holes)))
2459 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2462 * Reset the value so that previous (valid) value
2463 * from the last ack with SACK doesn't get used.
2465 tp->sackhint.sacked_bytes = 0;
2468 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2469 hhook_run_tcp_est_in(tp, th, &to);
2472 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2475 maxseg = tcp_maxseg(tp);
2477 (tiwin == tp->snd_wnd ||
2478 (tp->t_flags & TF_SACK_PERMIT))) {
2480 * If this is the first time we've seen a
2481 * FIN from the remote, this is not a
2482 * duplicate and it needs to be processed
2483 * normally. This happens during a
2484 * simultaneous close.
2486 if ((thflags & TH_FIN) &&
2487 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2491 TCPSTAT_INC(tcps_rcvdupack);
2493 * If we have outstanding data (other than
2494 * a window probe), this is a completely
2495 * duplicate ack (ie, window info didn't
2496 * change and FIN isn't set),
2497 * the ack is the biggest we've
2498 * seen and we've seen exactly our rexmt
2499 * threshold of them, assume a packet
2500 * has been dropped and retransmit it.
2501 * Kludge snd_nxt & the congestion
2502 * window so we send only this one
2505 * We know we're losing at the current
2506 * window size so do congestion avoidance
2507 * (set ssthresh to half the current window
2508 * and pull our congestion window back to
2509 * the new ssthresh).
2511 * Dup acks mean that packets have left the
2512 * network (they're now cached at the receiver)
2513 * so bump cwnd by the amount in the receiver
2514 * to keep a constant cwnd packets in the
2517 * When using TCP ECN, notify the peer that
2518 * we reduced the cwnd.
2521 * Following 2 kinds of acks should not affect
2524 * 2) Acks with SACK but without any new SACK
2525 * information in them. These could result from
2526 * any anomaly in the network like a switch
2527 * duplicating packets or a possible DoS attack.
2529 if (th->th_ack != tp->snd_una ||
2530 ((tp->t_flags & TF_SACK_PERMIT) &&
2533 else if (!tcp_timer_active(tp, TT_REXMT))
2535 else if (++tp->t_dupacks > tcprexmtthresh ||
2536 IN_FASTRECOVERY(tp->t_flags)) {
2537 cc_ack_received(tp, th, nsegs,
2539 if ((tp->t_flags & TF_SACK_PERMIT) &&
2540 IN_FASTRECOVERY(tp->t_flags)) {
2544 * Compute the amount of data in flight first.
2545 * We can inject new data into the pipe iff
2546 * we have less than 1/2 the original window's
2547 * worth of data in flight.
2549 if (V_tcp_do_rfc6675_pipe)
2550 awnd = tcp_compute_pipe(tp);
2552 awnd = (tp->snd_nxt - tp->snd_fack) +
2553 tp->sackhint.sack_bytes_rexmit;
2555 if (awnd < tp->snd_ssthresh) {
2556 tp->snd_cwnd += maxseg;
2557 if (tp->snd_cwnd > tp->snd_ssthresh)
2558 tp->snd_cwnd = tp->snd_ssthresh;
2561 tp->snd_cwnd += maxseg;
2562 (void) tp->t_fb->tfb_tcp_output(tp);
2564 } else if (tp->t_dupacks == tcprexmtthresh) {
2565 tcp_seq onxt = tp->snd_nxt;
2568 * If we're doing sack, check to
2569 * see if we're already in sack
2570 * recovery. If we're not doing sack,
2571 * check to see if we're in newreno
2574 if (tp->t_flags & TF_SACK_PERMIT) {
2575 if (IN_FASTRECOVERY(tp->t_flags)) {
2580 if (SEQ_LEQ(th->th_ack,
2586 /* Congestion signal before ack. */
2587 cc_cong_signal(tp, th, CC_NDUPACK);
2588 cc_ack_received(tp, th, nsegs,
2590 tcp_timer_activate(tp, TT_REXMT, 0);
2592 if (tp->t_flags & TF_SACK_PERMIT) {
2594 tcps_sack_recovery_episode);
2595 tp->snd_recover = tp->snd_nxt;
2596 tp->snd_cwnd = maxseg;
2597 (void) tp->t_fb->tfb_tcp_output(tp);
2600 tp->snd_nxt = th->th_ack;
2601 tp->snd_cwnd = maxseg;
2602 (void) tp->t_fb->tfb_tcp_output(tp);
2603 KASSERT(tp->snd_limited <= 2,
2604 ("%s: tp->snd_limited too big",
2606 tp->snd_cwnd = tp->snd_ssthresh +
2608 (tp->t_dupacks - tp->snd_limited);
2609 if (SEQ_GT(onxt, tp->snd_nxt))
2612 } else if (V_tcp_do_rfc3042) {
2614 * Process first and second duplicate
2615 * ACKs. Each indicates a segment
2616 * leaving the network, creating room
2617 * for more. Make sure we can send a
2618 * packet on reception of each duplicate
2619 * ACK by increasing snd_cwnd by one
2620 * segment. Restore the original
2621 * snd_cwnd after packet transmission.
2623 cc_ack_received(tp, th, nsegs,
2625 uint32_t oldcwnd = tp->snd_cwnd;
2626 tcp_seq oldsndmax = tp->snd_max;
2630 KASSERT(tp->t_dupacks == 1 ||
2632 ("%s: dupacks not 1 or 2",
2634 if (tp->t_dupacks == 1)
2635 tp->snd_limited = 0;
2637 (tp->snd_nxt - tp->snd_una) +
2638 (tp->t_dupacks - tp->snd_limited) *
2641 * Only call tcp_output when there
2642 * is new data available to be sent.
2643 * Otherwise we would send pure ACKs.
2645 SOCKBUF_LOCK(&so->so_snd);
2646 avail = sbavail(&so->so_snd) -
2647 (tp->snd_nxt - tp->snd_una);
2648 SOCKBUF_UNLOCK(&so->so_snd);
2650 (void) tp->t_fb->tfb_tcp_output(tp);
2651 sent = tp->snd_max - oldsndmax;
2652 if (sent > maxseg) {
2653 KASSERT((tp->t_dupacks == 2 &&
2654 tp->snd_limited == 0) ||
2655 (sent == maxseg + 1 &&
2656 tp->t_flags & TF_SENTFIN),
2657 ("%s: sent too much",
2659 tp->snd_limited = 2;
2660 } else if (sent > 0)
2662 tp->snd_cwnd = oldcwnd;
2669 * This ack is advancing the left edge, reset the
2674 * If this ack also has new SACK info, increment the
2675 * counter as per rfc6675. The variable
2676 * sack_changed tracks all changes to the SACK
2677 * scoreboard, including when partial ACKs without
2678 * SACK options are received, and clear the scoreboard
2679 * from the left side. Such partial ACKs should not be
2680 * counted as dupacks here.
2682 if ((tp->t_flags & TF_SACK_PERMIT) &&
2683 (to.to_flags & TOF_SACK) &&
2688 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2689 ("%s: th_ack <= snd_una", __func__));
2692 * If the congestion window was inflated to account
2693 * for the other side's cached packets, retract it.
2695 if (IN_FASTRECOVERY(tp->t_flags)) {
2696 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2697 if (tp->t_flags & TF_SACK_PERMIT)
2698 tcp_sack_partialack(tp, th);
2700 tcp_newreno_partial_ack(tp, th);
2702 cc_post_recovery(tp, th);
2705 * If we reach this point, ACK is not a duplicate,
2706 * i.e., it ACKs something we sent.
2708 if (tp->t_flags & TF_NEEDSYN) {
2710 * T/TCP: Connection was half-synchronized, and our
2711 * SYN has been ACK'd (so connection is now fully
2712 * synchronized). Go to non-starred state,
2713 * increment snd_una for ACK of SYN, and check if
2714 * we can do window scaling.
2716 tp->t_flags &= ~TF_NEEDSYN;
2718 /* Do window scaling? */
2719 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2720 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2721 tp->rcv_scale = tp->request_r_scale;
2722 /* Send window already scaled. */
2727 INP_WLOCK_ASSERT(tp->t_inpcb);
2730 * Adjust for the SYN bit in sequence space,
2731 * but don't account for it in cwnd calculations.
2732 * This is for the SYN_RECEIVED, non-simultaneous
2733 * SYN case. SYN_SENT and simultaneous SYN are
2734 * treated elsewhere.
2738 acked = BYTES_THIS_ACK(tp, th);
2739 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2740 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2741 tp->snd_una, th->th_ack, tp, m));
2742 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2743 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2746 * If we just performed our first retransmit, and the ACK
2747 * arrives within our recovery window, then it was a mistake
2748 * to do the retransmit in the first place. Recover our
2749 * original cwnd and ssthresh, and proceed to transmit where
2752 if (tp->t_rxtshift == 1 &&
2753 tp->t_flags & TF_PREVVALID &&
2755 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2756 cc_cong_signal(tp, th, CC_RTO_ERR);
2759 * If we have a timestamp reply, update smoothed
2760 * round trip time. If no timestamp is present but
2761 * transmit timer is running and timed sequence
2762 * number was acked, update smoothed round trip time.
2763 * Since we now have an rtt measurement, cancel the
2764 * timer backoff (cf., Phil Karn's retransmit alg.).
2765 * Recompute the initial retransmit timer.
2767 * Some boxes send broken timestamp replies
2768 * during the SYN+ACK phase, ignore
2769 * timestamps of 0 or we could calculate a
2770 * huge RTT and blow up the retransmit timer.
2772 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2775 t = tcp_ts_getticks() - to.to_tsecr;
2776 if (!tp->t_rttlow || tp->t_rttlow > t)
2778 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2779 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2780 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2781 tp->t_rttlow = ticks - tp->t_rtttime;
2782 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2786 * If all outstanding data is acked, stop retransmit
2787 * timer and remember to restart (more output or persist).
2788 * If there is more data to be acked, restart retransmit
2789 * timer, using current (possibly backed-off) value.
2791 if (th->th_ack == tp->snd_max) {
2792 tcp_timer_activate(tp, TT_REXMT, 0);
2794 } else if (!tcp_timer_active(tp, TT_PERSIST))
2795 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2798 * If no data (only SYN) was ACK'd,
2799 * skip rest of ACK processing.
2805 * Let the congestion control algorithm update congestion
2806 * control related information. This typically means increasing
2807 * the congestion window.
2809 cc_ack_received(tp, th, nsegs, CC_ACK);
2811 SOCKBUF_LOCK(&so->so_snd);
2812 if (acked > sbavail(&so->so_snd)) {
2813 if (tp->snd_wnd >= sbavail(&so->so_snd))
2814 tp->snd_wnd -= sbavail(&so->so_snd);
2817 mfree = sbcut_locked(&so->so_snd,
2818 (int)sbavail(&so->so_snd));
2821 mfree = sbcut_locked(&so->so_snd, acked);
2822 if (tp->snd_wnd >= (uint32_t) acked)
2823 tp->snd_wnd -= acked;
2828 /* NB: sowwakeup_locked() does an implicit unlock. */
2829 sowwakeup_locked(so);
2831 /* Detect una wraparound. */
2832 if (!IN_RECOVERY(tp->t_flags) &&
2833 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2834 SEQ_LEQ(th->th_ack, tp->snd_recover))
2835 tp->snd_recover = th->th_ack - 1;
2836 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2837 if (IN_RECOVERY(tp->t_flags) &&
2838 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2839 EXIT_RECOVERY(tp->t_flags);
2841 tp->snd_una = th->th_ack;
2842 if (tp->t_flags & TF_SACK_PERMIT) {
2843 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2844 tp->snd_recover = tp->snd_una;
2846 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2847 tp->snd_nxt = tp->snd_una;
2849 switch (tp->t_state) {
2852 * In FIN_WAIT_1 STATE in addition to the processing
2853 * for the ESTABLISHED state if our FIN is now acknowledged
2854 * then enter FIN_WAIT_2.
2856 case TCPS_FIN_WAIT_1:
2857 if (ourfinisacked) {
2859 * If we can't receive any more
2860 * data, then closing user can proceed.
2861 * Starting the timer is contrary to the
2862 * specification, but if we don't get a FIN
2863 * we'll hang forever.
2866 * we should release the tp also, and use a
2869 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2870 soisdisconnected(so);
2871 tcp_timer_activate(tp, TT_2MSL,
2872 (tcp_fast_finwait2_recycle ?
2873 tcp_finwait2_timeout :
2876 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2881 * In CLOSING STATE in addition to the processing for
2882 * the ESTABLISHED state if the ACK acknowledges our FIN
2883 * then enter the TIME-WAIT state, otherwise ignore
2887 if (ourfinisacked) {
2895 * In LAST_ACK, we may still be waiting for data to drain
2896 * and/or to be acked, as well as for the ack of our FIN.
2897 * If our FIN is now acknowledged, delete the TCB,
2898 * enter the closed state and return.
2901 if (ourfinisacked) {
2910 INP_WLOCK_ASSERT(tp->t_inpcb);
2913 * Update window information.
2914 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2916 if ((thflags & TH_ACK) &&
2917 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2918 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2919 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2920 /* keep track of pure window updates */
2922 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2923 TCPSTAT_INC(tcps_rcvwinupd);
2924 tp->snd_wnd = tiwin;
2925 tp->snd_wl1 = th->th_seq;
2926 tp->snd_wl2 = th->th_ack;
2927 if (tp->snd_wnd > tp->max_sndwnd)
2928 tp->max_sndwnd = tp->snd_wnd;
2933 * Process segments with URG.
2935 if ((thflags & TH_URG) && th->th_urp &&
2936 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2938 * This is a kludge, but if we receive and accept
2939 * random urgent pointers, we'll crash in
2940 * soreceive. It's hard to imagine someone
2941 * actually wanting to send this much urgent data.
2943 SOCKBUF_LOCK(&so->so_rcv);
2944 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2945 th->th_urp = 0; /* XXX */
2946 thflags &= ~TH_URG; /* XXX */
2947 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2948 goto dodata; /* XXX */
2951 * If this segment advances the known urgent pointer,
2952 * then mark the data stream. This should not happen
2953 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2954 * a FIN has been received from the remote side.
2955 * In these states we ignore the URG.
2957 * According to RFC961 (Assigned Protocols),
2958 * the urgent pointer points to the last octet
2959 * of urgent data. We continue, however,
2960 * to consider it to indicate the first octet
2961 * of data past the urgent section as the original
2962 * spec states (in one of two places).
2964 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2965 tp->rcv_up = th->th_seq + th->th_urp;
2966 so->so_oobmark = sbavail(&so->so_rcv) +
2967 (tp->rcv_up - tp->rcv_nxt) - 1;
2968 if (so->so_oobmark == 0)
2969 so->so_rcv.sb_state |= SBS_RCVATMARK;
2971 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2973 SOCKBUF_UNLOCK(&so->so_rcv);
2975 * Remove out of band data so doesn't get presented to user.
2976 * This can happen independent of advancing the URG pointer,
2977 * but if two URG's are pending at once, some out-of-band
2978 * data may creep in... ick.
2980 if (th->th_urp <= (uint32_t)tlen &&
2981 !(so->so_options & SO_OOBINLINE)) {
2982 /* hdr drop is delayed */
2983 tcp_pulloutofband(so, th, m, drop_hdrlen);
2987 * If no out of band data is expected,
2988 * pull receive urgent pointer along
2989 * with the receive window.
2991 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2992 tp->rcv_up = tp->rcv_nxt;
2995 INP_WLOCK_ASSERT(tp->t_inpcb);
2998 * Process the segment text, merging it into the TCP sequencing queue,
2999 * and arranging for acknowledgment of receipt if necessary.
3000 * This process logically involves adjusting tp->rcv_wnd as data
3001 * is presented to the user (this happens in tcp_usrreq.c,
3002 * case PRU_RCVD). If a FIN has already been received on this
3003 * connection then we just ignore the text.
3005 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3006 IS_FASTOPEN(tp->t_flags));
3007 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3008 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3009 tcp_seq save_start = th->th_seq;
3010 tcp_seq save_rnxt = tp->rcv_nxt;
3011 int save_tlen = tlen;
3012 m_adj(m, drop_hdrlen); /* delayed header drop */
3014 * Insert segment which includes th into TCP reassembly queue
3015 * with control block tp. Set thflags to whether reassembly now
3016 * includes a segment with FIN. This handles the common case
3017 * inline (segment is the next to be received on an established
3018 * connection, and the queue is empty), avoiding linkage into
3019 * and removal from the queue and repetition of various
3021 * Set DELACK for segments received in order, but ack
3022 * immediately when segments are out of order (so
3023 * fast retransmit can work).
3025 if (th->th_seq == tp->rcv_nxt &&
3027 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3029 if (DELAY_ACK(tp, tlen) || tfo_syn)
3030 tp->t_flags |= TF_DELACK;
3032 tp->t_flags |= TF_ACKNOW;
3033 tp->rcv_nxt += tlen;
3035 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3036 (tp->t_fbyte_in == 0)) {
3037 tp->t_fbyte_in = ticks;
3038 if (tp->t_fbyte_in == 0)
3040 if (tp->t_fbyte_out && tp->t_fbyte_in)
3041 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3043 thflags = th->th_flags & TH_FIN;
3044 TCPSTAT_INC(tcps_rcvpack);
3045 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3046 SOCKBUF_LOCK(&so->so_rcv);
3047 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3050 sbappendstream_locked(&so->so_rcv, m, 0);
3051 /* NB: sorwakeup_locked() does an implicit unlock. */
3052 sorwakeup_locked(so);
3055 * XXX: Due to the header drop above "th" is
3056 * theoretically invalid by now. Fortunately
3057 * m_adj() doesn't actually frees any mbufs
3058 * when trimming from the head.
3060 tcp_seq temp = save_start;
3061 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3062 tp->t_flags |= TF_ACKNOW;
3064 if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) {
3065 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3067 * DSACK actually handled in the fastpath
3070 tcp_update_sack_list(tp, save_start,
3071 save_start + save_tlen);
3072 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3073 if ((tp->rcv_numsacks >= 1) &&
3074 (tp->sackblks[0].end == save_start)) {
3076 * Partial overlap, recorded at todrop
3079 tcp_update_sack_list(tp,
3080 tp->sackblks[0].start,
3081 tp->sackblks[0].end);
3083 tcp_update_dsack_list(tp, save_start,
3084 save_start + save_tlen);
3086 } else if (tlen >= save_tlen) {
3087 /* Update of sackblks. */
3088 tcp_update_dsack_list(tp, save_start,
3089 save_start + save_tlen);
3090 } else if (tlen > 0) {
3091 tcp_update_dsack_list(tp, save_start,
3097 * Note the amount of data that peer has sent into
3098 * our window, in order to estimate the sender's
3102 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3103 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3105 len = so->so_rcv.sb_hiwat;
3113 * If FIN is received ACK the FIN and let the user know
3114 * that the connection is closing.
3116 if (thflags & TH_FIN) {
3117 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3120 * If connection is half-synchronized
3121 * (ie NEEDSYN flag on) then delay ACK,
3122 * so it may be piggybacked when SYN is sent.
3123 * Otherwise, since we received a FIN then no
3124 * more input can be expected, send ACK now.
3126 if (tp->t_flags & TF_NEEDSYN)
3127 tp->t_flags |= TF_DELACK;
3129 tp->t_flags |= TF_ACKNOW;
3132 switch (tp->t_state) {
3135 * In SYN_RECEIVED and ESTABLISHED STATES
3136 * enter the CLOSE_WAIT state.
3138 case TCPS_SYN_RECEIVED:
3139 tp->t_starttime = ticks;
3141 case TCPS_ESTABLISHED:
3142 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3146 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3147 * enter the CLOSING state.
3149 case TCPS_FIN_WAIT_1:
3150 tcp_state_change(tp, TCPS_CLOSING);
3154 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3155 * starting the time-wait timer, turning off the other
3158 case TCPS_FIN_WAIT_2:
3164 if (so->so_options & SO_DEBUG)
3165 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3168 TCP_PROBE3(debug__input, tp, th, m);
3171 * Return any desired output.
3173 if (needoutput || (tp->t_flags & TF_ACKNOW))
3174 (void) tp->t_fb->tfb_tcp_output(tp);
3177 INP_WLOCK_ASSERT(tp->t_inpcb);
3179 if (tp->t_flags & TF_DELACK) {
3180 tp->t_flags &= ~TF_DELACK;
3181 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3183 INP_WUNLOCK(tp->t_inpcb);
3188 * Generate an ACK dropping incoming segment if it occupies
3189 * sequence space, where the ACK reflects our state.
3191 * We can now skip the test for the RST flag since all
3192 * paths to this code happen after packets containing
3193 * RST have been dropped.
3195 * In the SYN-RECEIVED state, don't send an ACK unless the
3196 * segment we received passes the SYN-RECEIVED ACK test.
3197 * If it fails send a RST. This breaks the loop in the
3198 * "LAND" DoS attack, and also prevents an ACK storm
3199 * between two listening ports that have been sent forged
3200 * SYN segments, each with the source address of the other.
3202 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3203 (SEQ_GT(tp->snd_una, th->th_ack) ||
3204 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3205 rstreason = BANDLIM_RST_OPENPORT;
3209 if (so->so_options & SO_DEBUG)
3210 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3213 TCP_PROBE3(debug__input, tp, th, m);
3214 tp->t_flags |= TF_ACKNOW;
3215 (void) tp->t_fb->tfb_tcp_output(tp);
3216 INP_WUNLOCK(tp->t_inpcb);
3222 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3223 INP_WUNLOCK(tp->t_inpcb);
3225 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3230 * Drop space held by incoming segment and return.
3233 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3234 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3237 TCP_PROBE3(debug__input, tp, th, m);
3239 INP_WUNLOCK(tp->t_inpcb);
3244 * Issue RST and make ACK acceptable to originator of segment.
3245 * The mbuf must still include the original packet header.
3249 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3250 int tlen, int rstreason)
3256 struct ip6_hdr *ip6;
3260 INP_WLOCK_ASSERT(tp->t_inpcb);
3263 /* Don't bother if destination was broadcast/multicast. */
3264 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3267 if (mtod(m, struct ip *)->ip_v == 6) {
3268 ip6 = mtod(m, struct ip6_hdr *);
3269 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3270 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3272 /* IPv6 anycast check is done at tcp6_input() */
3275 #if defined(INET) && defined(INET6)
3280 ip = mtod(m, struct ip *);
3281 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3282 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3283 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3284 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3289 /* Perform bandwidth limiting. */
3290 if (badport_bandlim(rstreason) < 0)
3293 /* tcp_respond consumes the mbuf chain. */
3294 if (th->th_flags & TH_ACK) {
3295 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3296 th->th_ack, TH_RST);
3298 if (th->th_flags & TH_SYN)
3300 if (th->th_flags & TH_FIN)
3302 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3303 (tcp_seq)0, TH_RST|TH_ACK);
3311 * Parse TCP options and place in tcpopt.
3314 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3319 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3321 if (opt == TCPOPT_EOL)
3323 if (opt == TCPOPT_NOP)
3329 if (optlen < 2 || optlen > cnt)
3334 if (optlen != TCPOLEN_MAXSEG)
3336 if (!(flags & TO_SYN))
3338 to->to_flags |= TOF_MSS;
3339 bcopy((char *)cp + 2,
3340 (char *)&to->to_mss, sizeof(to->to_mss));
3341 to->to_mss = ntohs(to->to_mss);
3344 if (optlen != TCPOLEN_WINDOW)
3346 if (!(flags & TO_SYN))
3348 to->to_flags |= TOF_SCALE;
3349 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3351 case TCPOPT_TIMESTAMP:
3352 if (optlen != TCPOLEN_TIMESTAMP)
3354 to->to_flags |= TOF_TS;
3355 bcopy((char *)cp + 2,
3356 (char *)&to->to_tsval, sizeof(to->to_tsval));
3357 to->to_tsval = ntohl(to->to_tsval);
3358 bcopy((char *)cp + 6,
3359 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3360 to->to_tsecr = ntohl(to->to_tsecr);
3362 case TCPOPT_SIGNATURE:
3364 * In order to reply to a host which has set the
3365 * TCP_SIGNATURE option in its initial SYN, we have
3366 * to record the fact that the option was observed
3367 * here for the syncache code to perform the correct
3370 if (optlen != TCPOLEN_SIGNATURE)
3372 to->to_flags |= TOF_SIGNATURE;
3373 to->to_signature = cp + 2;
3375 case TCPOPT_SACK_PERMITTED:
3376 if (optlen != TCPOLEN_SACK_PERMITTED)
3378 if (!(flags & TO_SYN))
3382 to->to_flags |= TOF_SACKPERM;
3385 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3389 to->to_flags |= TOF_SACK;
3390 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3391 to->to_sacks = cp + 2;
3392 TCPSTAT_INC(tcps_sack_rcv_blocks);
3394 case TCPOPT_FAST_OPEN:
3396 * Cookie length validation is performed by the
3397 * server side cookie checking code or the client
3398 * side cookie cache update code.
3400 if (!(flags & TO_SYN))
3402 if (!V_tcp_fastopen_client_enable &&
3403 !V_tcp_fastopen_server_enable)
3405 to->to_flags |= TOF_FASTOPEN;
3406 to->to_tfo_len = optlen - 2;
3407 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3416 * Pull out of band byte out of a segment so
3417 * it doesn't appear in the user's data queue.
3418 * It is still reflected in the segment length for
3419 * sequencing purposes.
3422 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3425 int cnt = off + th->th_urp - 1;
3428 if (m->m_len > cnt) {
3429 char *cp = mtod(m, caddr_t) + cnt;
3430 struct tcpcb *tp = sototcpcb(so);
3432 INP_WLOCK_ASSERT(tp->t_inpcb);
3435 tp->t_oobflags |= TCPOOB_HAVEDATA;
3436 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3438 if (m->m_flags & M_PKTHDR)
3447 panic("tcp_pulloutofband");
3451 * Collect new round-trip time estimate
3452 * and update averages and current timeout.
3455 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3459 INP_WLOCK_ASSERT(tp->t_inpcb);
3461 TCPSTAT_INC(tcps_rttupdated);
3464 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3465 imax(0, rtt * 1000 / hz));
3467 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3469 * srtt is stored as fixed point with 5 bits after the
3470 * binary point (i.e., scaled by 8). The following magic
3471 * is equivalent to the smoothing algorithm in rfc793 with
3472 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3473 * point). Adjust rtt to origin 0.
3475 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3476 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3478 if ((tp->t_srtt += delta) <= 0)
3482 * We accumulate a smoothed rtt variance (actually, a
3483 * smoothed mean difference), then set the retransmit
3484 * timer to smoothed rtt + 4 times the smoothed variance.
3485 * rttvar is stored as fixed point with 4 bits after the
3486 * binary point (scaled by 16). The following is
3487 * equivalent to rfc793 smoothing with an alpha of .75
3488 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3489 * rfc793's wired-in beta.
3493 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3494 if ((tp->t_rttvar += delta) <= 0)
3496 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3497 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3500 * No rtt measurement yet - use the unsmoothed rtt.
3501 * Set the variance to half the rtt (so our first
3502 * retransmit happens at 3*rtt).
3504 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3505 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3506 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3512 * the retransmit should happen at rtt + 4 * rttvar.
3513 * Because of the way we do the smoothing, srtt and rttvar
3514 * will each average +1/2 tick of bias. When we compute
3515 * the retransmit timer, we want 1/2 tick of rounding and
3516 * 1 extra tick because of +-1/2 tick uncertainty in the
3517 * firing of the timer. The bias will give us exactly the
3518 * 1.5 tick we need. But, because the bias is
3519 * statistical, we have to test that we don't drop below
3520 * the minimum feasible timer (which is 2 ticks).
3522 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3523 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3526 * We received an ack for a packet that wasn't retransmitted;
3527 * it is probably safe to discard any error indications we've
3528 * received recently. This isn't quite right, but close enough
3529 * for now (a route might have failed after we sent a segment,
3530 * and the return path might not be symmetrical).
3532 tp->t_softerror = 0;
3536 * Determine a reasonable value for maxseg size.
3537 * If the route is known, check route for mtu.
3538 * If none, use an mss that can be handled on the outgoing interface
3539 * without forcing IP to fragment. If no route is found, route has no mtu,
3540 * or the destination isn't local, use a default, hopefully conservative
3541 * size (usually 512 or the default IP max size, but no more than the mtu
3542 * of the interface), as we can't discover anything about intervening
3543 * gateways or networks. We also initialize the congestion/slow start
3544 * window to be a single segment if the destination isn't local.
3545 * While looking at the routing entry, we also initialize other path-dependent
3546 * parameters from pre-set or cached values in the routing entry.
3548 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3549 * IP options, e.g. IPSEC data, since length of this data may vary, and
3550 * thus it is calculated for every segment separately in tcp_output().
3552 * NOTE that this routine is only called when we process an incoming
3553 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3554 * settings are handled in tcp_mssopt().
3557 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3558 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3561 uint32_t maxmtu = 0;
3562 struct inpcb *inp = tp->t_inpcb;
3563 struct hc_metrics_lite metrics;
3565 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3566 size_t min_protoh = isipv6 ?
3567 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3568 sizeof (struct tcpiphdr);
3570 const size_t min_protoh = sizeof(struct tcpiphdr);
3573 INP_WLOCK_ASSERT(tp->t_inpcb);
3575 if (mtuoffer != -1) {
3576 KASSERT(offer == -1, ("%s: conflict", __func__));
3577 offer = mtuoffer - min_protoh;
3583 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3584 tp->t_maxseg = V_tcp_v6mssdflt;
3587 #if defined(INET) && defined(INET6)
3592 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3593 tp->t_maxseg = V_tcp_mssdflt;
3598 * No route to sender, stay with default mss and return.
3602 * In case we return early we need to initialize metrics
3603 * to a defined state as tcp_hc_get() would do for us
3604 * if there was no cache hit.
3606 if (metricptr != NULL)
3607 bzero(metricptr, sizeof(struct hc_metrics_lite));
3611 /* What have we got? */
3615 * Offer == 0 means that there was no MSS on the SYN
3616 * segment, in this case we use tcp_mssdflt as
3617 * already assigned to t_maxseg above.
3619 offer = tp->t_maxseg;
3624 * Offer == -1 means that we didn't receive SYN yet.
3630 * Prevent DoS attack with too small MSS. Round up
3631 * to at least minmss.
3633 offer = max(offer, V_tcp_minmss);
3637 * rmx information is now retrieved from tcp_hostcache.
3639 tcp_hc_get(&inp->inp_inc, &metrics);
3640 if (metricptr != NULL)
3641 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3644 * If there's a discovered mtu in tcp hostcache, use it.
3645 * Else, use the link mtu.
3647 if (metrics.rmx_mtu)
3648 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3652 mss = maxmtu - min_protoh;
3653 if (!V_path_mtu_discovery &&
3654 !in6_localaddr(&inp->in6p_faddr))
3655 mss = min(mss, V_tcp_v6mssdflt);
3658 #if defined(INET) && defined(INET6)
3663 mss = maxmtu - min_protoh;
3664 if (!V_path_mtu_discovery &&
3665 !in_localaddr(inp->inp_faddr))
3666 mss = min(mss, V_tcp_mssdflt);
3670 * XXX - The above conditional (mss = maxmtu - min_protoh)
3671 * probably violates the TCP spec.
3672 * The problem is that, since we don't know the
3673 * other end's MSS, we are supposed to use a conservative
3674 * default. But, if we do that, then MTU discovery will
3675 * never actually take place, because the conservative
3676 * default is much less than the MTUs typically seen
3677 * on the Internet today. For the moment, we'll sweep
3678 * this under the carpet.
3680 * The conservative default might not actually be a problem
3681 * if the only case this occurs is when sending an initial
3682 * SYN with options and data to a host we've never talked
3683 * to before. Then, they will reply with an MSS value which
3684 * will get recorded and the new parameters should get
3685 * recomputed. For Further Study.
3688 mss = min(mss, offer);
3691 * Sanity check: make sure that maxseg will be large
3692 * enough to allow some data on segments even if the
3693 * all the option space is used (40bytes). Otherwise
3694 * funny things may happen in tcp_output.
3696 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3704 tcp_mss(struct tcpcb *tp, int offer)
3710 struct hc_metrics_lite metrics;
3711 struct tcp_ifcap cap;
3713 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3715 bzero(&cap, sizeof(cap));
3716 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3722 * If there's a pipesize, change the socket buffer to that size,
3723 * don't change if sb_hiwat is different than default (then it
3724 * has been changed on purpose with setsockopt).
3725 * Make the socket buffers an integral number of mss units;
3726 * if the mss is larger than the socket buffer, decrease the mss.
3728 so = inp->inp_socket;
3729 SOCKBUF_LOCK(&so->so_snd);
3730 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3731 bufsize = metrics.rmx_sendpipe;
3733 bufsize = so->so_snd.sb_hiwat;
3737 bufsize = roundup(bufsize, mss);
3738 if (bufsize > sb_max)
3740 if (bufsize > so->so_snd.sb_hiwat)
3741 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3743 SOCKBUF_UNLOCK(&so->so_snd);
3745 * Sanity check: make sure that maxseg will be large
3746 * enough to allow some data on segments even if the
3747 * all the option space is used (40bytes). Otherwise
3748 * funny things may happen in tcp_output.
3750 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3752 tp->t_maxseg = max(mss, 64);
3754 SOCKBUF_LOCK(&so->so_rcv);
3755 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3756 bufsize = metrics.rmx_recvpipe;
3758 bufsize = so->so_rcv.sb_hiwat;
3759 if (bufsize > mss) {
3760 bufsize = roundup(bufsize, mss);
3761 if (bufsize > sb_max)
3763 if (bufsize > so->so_rcv.sb_hiwat)
3764 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3766 SOCKBUF_UNLOCK(&so->so_rcv);
3768 /* Check the interface for TSO capabilities. */
3769 if (cap.ifcap & CSUM_TSO) {
3770 tp->t_flags |= TF_TSO;
3771 tp->t_tsomax = cap.tsomax;
3772 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3773 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3778 * Determine the MSS option to send on an outgoing SYN.
3781 tcp_mssopt(struct in_conninfo *inc)
3784 uint32_t thcmtu = 0;
3785 uint32_t maxmtu = 0;
3788 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3791 if (inc->inc_flags & INC_ISIPV6) {
3792 mss = V_tcp_v6mssdflt;
3793 maxmtu = tcp_maxmtu6(inc, NULL);
3794 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3797 #if defined(INET) && defined(INET6)
3802 mss = V_tcp_mssdflt;
3803 maxmtu = tcp_maxmtu(inc, NULL);
3804 min_protoh = sizeof(struct tcpiphdr);
3807 #if defined(INET6) || defined(INET)
3808 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3811 if (maxmtu && thcmtu)
3812 mss = min(maxmtu, thcmtu) - min_protoh;
3813 else if (maxmtu || thcmtu)
3814 mss = max(maxmtu, thcmtu) - min_protoh;
3821 * On a partial ack arrives, force the retransmission of the
3822 * next unacknowledged segment. Do not clear tp->t_dupacks.
3823 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3827 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3829 tcp_seq onxt = tp->snd_nxt;
3830 uint32_t ocwnd = tp->snd_cwnd;
3831 u_int maxseg = tcp_maxseg(tp);
3833 INP_WLOCK_ASSERT(tp->t_inpcb);
3835 tcp_timer_activate(tp, TT_REXMT, 0);
3837 tp->snd_nxt = th->th_ack;
3839 * Set snd_cwnd to one segment beyond acknowledged offset.
3840 * (tp->snd_una has not yet been updated when this function is called.)
3842 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3843 tp->t_flags |= TF_ACKNOW;
3844 (void) tp->t_fb->tfb_tcp_output(tp);
3845 tp->snd_cwnd = ocwnd;
3846 if (SEQ_GT(onxt, tp->snd_nxt))
3849 * Partial window deflation. Relies on fact that tp->snd_una
3852 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3853 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3856 tp->snd_cwnd += maxseg;
3860 tcp_compute_pipe(struct tcpcb *tp)
3862 return (tp->snd_max - tp->snd_una +
3863 tp->sackhint.sack_bytes_rexmit -
3864 tp->sackhint.sacked_bytes);
3868 tcp_compute_initwnd(uint32_t maxseg)
3871 * Calculate the Initial Window, also used as Restart Window
3873 * RFC5681 Section 3.1 specifies the default conservative values.
3874 * RFC3390 specifies slightly more aggressive values.
3875 * RFC6928 increases it to ten segments.
3876 * Support for user specified value for initial flight size.
3878 if (V_tcp_initcwnd_segments)
3879 return min(V_tcp_initcwnd_segments * maxseg,
3880 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
3881 else if (V_tcp_do_rfc3390)
3882 return min(4 * maxseg, max(2 * maxseg, 4380));
3884 /* Per RFC5681 Section 3.1 */
3886 return (2 * maxseg);
3887 else if (maxseg > 1095)
3888 return (3 * maxseg);
3890 return (4 * maxseg);