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)) {
593 ifa_free(&ia6->ia_ifa);
594 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
595 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
597 return (IPPROTO_DONE);
600 ifa_free(&ia6->ia_ifa);
603 return (tcp_input(mp, offp, proto));
608 tcp_input(struct mbuf **mp, int *offp, int proto)
610 struct mbuf *m = *mp;
611 struct tcphdr *th = NULL;
612 struct ip *ip = NULL;
613 struct inpcb *inp = NULL;
614 struct tcpcb *tp = NULL;
615 struct socket *so = NULL;
626 int rstreason = 0; /* For badport_bandlim accounting purposes */
628 struct m_tag *fwd_tag = NULL;
630 struct ip6_hdr *ip6 = NULL;
633 const void *ip6 = NULL;
635 struct tcpopt to; /* options in this segment */
636 char *s = NULL; /* address and port logging */
639 * The size of tcp_saveipgen must be the size of the max ip header,
642 u_char tcp_saveipgen[IP6_HDR_LEN];
643 struct tcphdr tcp_savetcp;
650 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
657 TCPSTAT_INC(tcps_rcvtotal);
661 ip6 = mtod(m, struct ip6_hdr *);
662 th = (struct tcphdr *)((caddr_t)ip6 + off0);
663 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
664 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
665 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
666 th->th_sum = m->m_pkthdr.csum_data;
668 th->th_sum = in6_cksum_pseudo(ip6, tlen,
669 IPPROTO_TCP, m->m_pkthdr.csum_data);
670 th->th_sum ^= 0xffff;
672 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
674 TCPSTAT_INC(tcps_rcvbadsum);
679 * Be proactive about unspecified IPv6 address in source.
680 * As we use all-zero to indicate unbounded/unconnected pcb,
681 * unspecified IPv6 address can be used to confuse us.
683 * Note that packets with unspecified IPv6 destination is
684 * already dropped in ip6_input.
686 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
690 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
693 #if defined(INET) && defined(INET6)
699 * Get IP and TCP header together in first mbuf.
700 * Note: IP leaves IP header in first mbuf.
702 if (off0 > sizeof (struct ip)) {
704 off0 = sizeof(struct ip);
706 if (m->m_len < sizeof (struct tcpiphdr)) {
707 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
709 TCPSTAT_INC(tcps_rcvshort);
710 return (IPPROTO_DONE);
713 ip = mtod(m, struct ip *);
714 th = (struct tcphdr *)((caddr_t)ip + off0);
715 tlen = ntohs(ip->ip_len) - off0;
718 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
719 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
720 th->th_sum = m->m_pkthdr.csum_data;
722 th->th_sum = in_pseudo(ip->ip_src.s_addr,
724 htonl(m->m_pkthdr.csum_data + tlen +
726 th->th_sum ^= 0xffff;
728 struct ipovly *ipov = (struct ipovly *)ip;
731 * Checksum extended TCP header and data.
735 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
736 ipov->ih_len = htons(tlen);
737 th->th_sum = in_cksum(m, len);
738 /* Reset length for SDT probes. */
739 ip->ip_len = htons(len);
742 /* Re-initialization for later version check */
744 ip->ip_v = IPVERSION;
745 ip->ip_hl = off0 >> 2;
749 TCPSTAT_INC(tcps_rcvbadsum);
756 * Check that TCP offset makes sense,
757 * pull out TCP options and adjust length. XXX
759 off = th->th_off << 2;
760 if (off < sizeof (struct tcphdr) || off > tlen) {
761 TCPSTAT_INC(tcps_rcvbadoff);
764 tlen -= off; /* tlen is used instead of ti->ti_len */
765 if (off > sizeof (struct tcphdr)) {
768 if (m->m_len < off0 + off) {
769 m = m_pullup(m, off0 + off);
771 TCPSTAT_INC(tcps_rcvshort);
772 return (IPPROTO_DONE);
775 ip6 = mtod(m, struct ip6_hdr *);
776 th = (struct tcphdr *)((caddr_t)ip6 + off0);
779 #if defined(INET) && defined(INET6)
784 if (m->m_len < sizeof(struct ip) + off) {
785 if ((m = m_pullup(m, sizeof (struct ip) + off))
787 TCPSTAT_INC(tcps_rcvshort);
788 return (IPPROTO_DONE);
790 ip = mtod(m, struct ip *);
791 th = (struct tcphdr *)((caddr_t)ip + off0);
795 optlen = off - sizeof (struct tcphdr);
796 optp = (u_char *)(th + 1);
798 thflags = th->th_flags;
801 * Convert TCP protocol specific fields to host format.
803 tcp_fields_to_host(th);
806 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
808 drop_hdrlen = off0 + off;
811 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
815 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
817 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
820 #if defined(INET) && !defined(INET6)
821 (m->m_flags & M_IP_NEXTHOP)
824 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
828 if (isipv6 && fwd_tag != NULL) {
829 struct sockaddr_in6 *next_hop6;
831 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
833 * Transparently forwarded. Pretend to be the destination.
834 * Already got one like this?
836 inp = in6_pcblookup_mbuf(&V_tcbinfo,
837 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
838 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
841 * It's new. Try to find the ambushing socket.
842 * Because we've rewritten the destination address,
843 * any hardware-generated hash is ignored.
845 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
846 th->th_sport, &next_hop6->sin6_addr,
847 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
848 th->th_dport, INPLOOKUP_WILDCARD |
849 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
852 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
853 th->th_sport, &ip6->ip6_dst, th->th_dport,
854 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
855 m->m_pkthdr.rcvif, m);
858 #if defined(INET6) && defined(INET)
862 if (fwd_tag != NULL) {
863 struct sockaddr_in *next_hop;
865 next_hop = (struct sockaddr_in *)(fwd_tag+1);
867 * Transparently forwarded. Pretend to be the destination.
868 * already got one like this?
870 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
871 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
872 m->m_pkthdr.rcvif, m);
875 * It's new. Try to find the ambushing socket.
876 * Because we've rewritten the destination address,
877 * any hardware-generated hash is ignored.
879 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
880 th->th_sport, next_hop->sin_addr,
881 next_hop->sin_port ? ntohs(next_hop->sin_port) :
882 th->th_dport, INPLOOKUP_WILDCARD |
883 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
886 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
887 th->th_sport, ip->ip_dst, th->th_dport,
888 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
889 m->m_pkthdr.rcvif, m);
893 * If the INPCB does not exist then all data in the incoming
894 * segment is discarded and an appropriate RST is sent back.
895 * XXX MRT Send RST using which routing table?
899 * Log communication attempts to ports that are not
902 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
903 V_tcp_log_in_vain == 2) {
904 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
905 log(LOG_INFO, "%s; %s: Connection attempt "
906 "to closed port\n", s, __func__);
909 * When blackholing do not respond with a RST but
910 * completely ignore the segment and drop it.
912 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
916 rstreason = BANDLIM_RST_CLOSEDPORT;
919 INP_WLOCK_ASSERT(inp);
921 * While waiting for inp lock during the lookup, another thread
922 * can have dropped the inpcb, in which case we need to loop back
923 * and try to find a new inpcb to deliver to.
925 if (inp->inp_flags & INP_DROPPED) {
930 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
931 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
932 ((inp->inp_socket == NULL) ||
933 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
934 inp->inp_flowid = m->m_pkthdr.flowid;
935 inp->inp_flowtype = M_HASHTYPE_GET(m);
937 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
939 if (isipv6 && IPSEC_ENABLED(ipv6) &&
940 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
948 if (IPSEC_ENABLED(ipv4) &&
949 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
956 * Check the minimum TTL for socket.
958 if (inp->inp_ip_minttl != 0) {
961 if (inp->inp_ip_minttl > ip6->ip6_hlim)
965 if (inp->inp_ip_minttl > ip->ip_ttl)
970 * A previous connection in TIMEWAIT state is supposed to catch stray
971 * or duplicate segments arriving late. If this segment was a
972 * legitimate new connection attempt, the old INPCB gets removed and
973 * we can try again to find a listening socket.
975 * At this point, due to earlier optimism, we may hold only an inpcb
976 * lock, and not the inpcbinfo write lock. If so, we need to try to
977 * acquire it, or if that fails, acquire a reference on the inpcb,
978 * drop all locks, acquire a global write lock, and then re-acquire
979 * the inpcb lock. We may at that point discover that another thread
980 * has tried to free the inpcb, in which case we need to loop back
981 * and try to find a new inpcb to deliver to.
983 * XXXRW: It may be time to rethink timewait locking.
985 if (inp->inp_flags & INP_TIMEWAIT) {
986 if (thflags & TH_SYN)
987 tcp_dooptions(&to, optp, optlen, TO_SYN);
989 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
991 if (tcp_twcheck(inp, &to, th, m, tlen))
993 return (IPPROTO_DONE);
996 * The TCPCB may no longer exist if the connection is winding
997 * down or it is in the CLOSED state. Either way we drop the
998 * segment and send an appropriate response.
1000 tp = intotcpcb(inp);
1001 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1002 rstreason = BANDLIM_RST_CLOSEDPORT;
1007 if (tp->t_flags & TF_TOE) {
1008 tcp_offload_input(tp, m);
1009 m = NULL; /* consumed by the TOE driver */
1015 INP_WLOCK_ASSERT(inp);
1016 if (mac_inpcb_check_deliver(inp, m))
1019 so = inp->inp_socket;
1020 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1022 if (so->so_options & SO_DEBUG) {
1023 ostate = tp->t_state;
1026 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1029 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1032 #endif /* TCPDEBUG */
1034 * When the socket is accepting connections (the INPCB is in LISTEN
1035 * state) we look into the SYN cache if this is a new connection
1036 * attempt or the completion of a previous one.
1038 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1039 ("%s: so accepting but tp %p not listening", __func__, tp));
1040 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1041 struct in_conninfo inc;
1043 bzero(&inc, sizeof(inc));
1046 inc.inc_flags |= INC_ISIPV6;
1047 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1048 inc.inc_flags |= INC_IPV6MINMTU;
1049 inc.inc6_faddr = ip6->ip6_src;
1050 inc.inc6_laddr = ip6->ip6_dst;
1054 inc.inc_faddr = ip->ip_src;
1055 inc.inc_laddr = ip->ip_dst;
1057 inc.inc_fport = th->th_sport;
1058 inc.inc_lport = th->th_dport;
1059 inc.inc_fibnum = so->so_fibnum;
1062 * Check for an existing connection attempt in syncache if
1063 * the flag is only ACK. A successful lookup creates a new
1064 * socket appended to the listen queue in SYN_RECEIVED state.
1066 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1068 * Parse the TCP options here because
1069 * syncookies need access to the reflected
1072 tcp_dooptions(&to, optp, optlen, 0);
1074 * NB: syncache_expand() doesn't unlock
1075 * inp and tcpinfo locks.
1077 rstreason = syncache_expand(&inc, &to, th, &so, m);
1078 if (rstreason < 0) {
1080 * A failing TCP MD5 signature comparison
1081 * must result in the segment being dropped
1082 * and must not produce any response back
1086 } else if (rstreason == 0) {
1088 * No syncache entry or ACK was not
1089 * for our SYN/ACK. Send a RST.
1090 * NB: syncache did its own logging
1091 * of the failure cause.
1093 rstreason = BANDLIM_RST_OPENPORT;
1099 * We completed the 3-way handshake
1100 * but could not allocate a socket
1101 * either due to memory shortage,
1102 * listen queue length limits or
1103 * global socket limits. Send RST
1104 * or wait and have the remote end
1105 * retransmit the ACK for another
1108 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1109 log(LOG_DEBUG, "%s; %s: Listen socket: "
1110 "Socket allocation failed due to "
1111 "limits or memory shortage, %s\n",
1113 V_tcp_sc_rst_sock_fail ?
1114 "sending RST" : "try again");
1115 if (V_tcp_sc_rst_sock_fail) {
1116 rstreason = BANDLIM_UNLIMITED;
1122 * Socket is created in state SYN_RECEIVED.
1123 * Unlock the listen socket, lock the newly
1124 * created socket and update the tp variable.
1126 INP_WUNLOCK(inp); /* listen socket */
1127 inp = sotoinpcb(so);
1129 * New connection inpcb is already locked by
1130 * syncache_expand().
1132 INP_WLOCK_ASSERT(inp);
1133 tp = intotcpcb(inp);
1134 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1135 ("%s: ", __func__));
1137 * Process the segment and the data it
1138 * contains. tcp_do_segment() consumes
1139 * the mbuf chain and unlocks the inpcb.
1141 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1142 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1144 return (IPPROTO_DONE);
1147 * Segment flag validation for new connection attempts:
1149 * Our (SYN|ACK) response was rejected.
1150 * Check with syncache and remove entry to prevent
1153 * NB: syncache_chkrst does its own logging of failure
1156 if (thflags & TH_RST) {
1157 syncache_chkrst(&inc, th, m);
1161 * We can't do anything without SYN.
1163 if ((thflags & TH_SYN) == 0) {
1164 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1165 log(LOG_DEBUG, "%s; %s: Listen socket: "
1166 "SYN is missing, segment ignored\n",
1168 TCPSTAT_INC(tcps_badsyn);
1172 * (SYN|ACK) is bogus on a listen socket.
1174 if (thflags & TH_ACK) {
1175 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1176 log(LOG_DEBUG, "%s; %s: Listen socket: "
1177 "SYN|ACK invalid, segment rejected\n",
1179 syncache_badack(&inc); /* XXX: Not needed! */
1180 TCPSTAT_INC(tcps_badsyn);
1181 rstreason = BANDLIM_RST_OPENPORT;
1185 * If the drop_synfin option is enabled, drop all
1186 * segments with both the SYN and FIN bits set.
1187 * This prevents e.g. nmap from identifying the
1189 * XXX: Poor reasoning. nmap has other methods
1190 * and is constantly refining its stack detection
1192 * XXX: This is a violation of the TCP specification
1193 * and was used by RFC1644.
1195 if ((thflags & TH_FIN) && V_drop_synfin) {
1196 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1197 log(LOG_DEBUG, "%s; %s: Listen socket: "
1198 "SYN|FIN segment ignored (based on "
1199 "sysctl setting)\n", s, __func__);
1200 TCPSTAT_INC(tcps_badsyn);
1204 * Segment's flags are (SYN) or (SYN|FIN).
1206 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1207 * as they do not affect the state of the TCP FSM.
1208 * The data pointed to by TH_URG and th_urp is ignored.
1210 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1211 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1212 KASSERT(thflags & (TH_SYN),
1213 ("%s: Listen socket: TH_SYN not set", __func__));
1216 * If deprecated address is forbidden,
1217 * we do not accept SYN to deprecated interface
1218 * address to prevent any new inbound connection from
1219 * getting established.
1220 * When we do not accept SYN, we send a TCP RST,
1221 * with deprecated source address (instead of dropping
1222 * it). We compromise it as it is much better for peer
1223 * to send a RST, and RST will be the final packet
1226 * If we do not forbid deprecated addresses, we accept
1227 * the SYN packet. RFC2462 does not suggest dropping
1229 * If we decipher RFC2462 5.5.4, it says like this:
1230 * 1. use of deprecated addr with existing
1231 * communication is okay - "SHOULD continue to be
1233 * 2. use of it with new communication:
1234 * (2a) "SHOULD NOT be used if alternate address
1235 * with sufficient scope is available"
1236 * (2b) nothing mentioned otherwise.
1237 * Here we fall into (2b) case as we have no choice in
1238 * our source address selection - we must obey the peer.
1240 * The wording in RFC2462 is confusing, and there are
1241 * multiple description text for deprecated address
1242 * handling - worse, they are not exactly the same.
1243 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1245 if (isipv6 && !V_ip6_use_deprecated) {
1246 struct in6_ifaddr *ia6;
1248 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1250 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1251 ifa_free(&ia6->ia_ifa);
1252 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1253 log(LOG_DEBUG, "%s; %s: Listen socket: "
1254 "Connection attempt to deprecated "
1255 "IPv6 address rejected\n",
1257 rstreason = BANDLIM_RST_OPENPORT;
1261 ifa_free(&ia6->ia_ifa);
1265 * Basic sanity checks on incoming SYN requests:
1266 * Don't respond if the destination is a link layer
1267 * broadcast according to RFC1122 4.2.3.10, p. 104.
1268 * If it is from this socket it must be forged.
1269 * Don't respond if the source or destination is a
1270 * global or subnet broad- or multicast address.
1271 * Note that it is quite possible to receive unicast
1272 * link-layer packets with a broadcast IP address. Use
1273 * in_broadcast() to find them.
1275 if (m->m_flags & (M_BCAST|M_MCAST)) {
1276 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1277 log(LOG_DEBUG, "%s; %s: Listen socket: "
1278 "Connection attempt from broad- or multicast "
1279 "link layer address ignored\n", s, __func__);
1284 if (th->th_dport == th->th_sport &&
1285 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1286 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1287 log(LOG_DEBUG, "%s; %s: Listen socket: "
1288 "Connection attempt to/from self "
1289 "ignored\n", s, __func__);
1292 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1293 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1294 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1295 log(LOG_DEBUG, "%s; %s: Listen socket: "
1296 "Connection attempt from/to multicast "
1297 "address ignored\n", s, __func__);
1302 #if defined(INET) && defined(INET6)
1307 if (th->th_dport == th->th_sport &&
1308 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1309 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1310 log(LOG_DEBUG, "%s; %s: Listen socket: "
1311 "Connection attempt from/to self "
1312 "ignored\n", s, __func__);
1315 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1316 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1317 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1318 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1319 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1320 log(LOG_DEBUG, "%s; %s: Listen socket: "
1321 "Connection attempt from/to broad- "
1322 "or multicast address ignored\n",
1329 * SYN appears to be valid. Create compressed TCP state
1333 if (so->so_options & SO_DEBUG)
1334 tcp_trace(TA_INPUT, ostate, tp,
1335 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1337 TCP_PROBE3(debug__input, tp, th, m);
1338 tcp_dooptions(&to, optp, optlen, TO_SYN);
1339 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos))
1340 goto tfo_socket_result;
1343 * Entry added to syncache and mbuf consumed.
1344 * Only the listen socket is unlocked by syncache_add().
1346 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1347 return (IPPROTO_DONE);
1348 } else if (tp->t_state == TCPS_LISTEN) {
1350 * When a listen socket is torn down the SO_ACCEPTCONN
1351 * flag is removed first while connections are drained
1352 * from the accept queue in a unlock/lock cycle of the
1353 * ACCEPT_LOCK, opening a race condition allowing a SYN
1354 * attempt go through unhandled.
1358 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1359 if (tp->t_flags & TF_SIGNATURE) {
1360 tcp_dooptions(&to, optp, optlen, thflags);
1361 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1362 TCPSTAT_INC(tcps_sig_err_nosigopt);
1365 if (!TCPMD5_ENABLED() ||
1366 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1370 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1373 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1374 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1375 * the inpcb, and unlocks pcbinfo.
1377 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1378 return (IPPROTO_DONE);
1381 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1384 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1387 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1388 m = NULL; /* mbuf chain got consumed. */
1393 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1399 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1404 return (IPPROTO_DONE);
1408 * Automatic sizing of receive socket buffer. Often the send
1409 * buffer size is not optimally adjusted to the actual network
1410 * conditions at hand (delay bandwidth product). Setting the
1411 * buffer size too small limits throughput on links with high
1412 * bandwidth and high delay (eg. trans-continental/oceanic links).
1414 * On the receive side the socket buffer memory is only rarely
1415 * used to any significant extent. This allows us to be much
1416 * more aggressive in scaling the receive socket buffer. For
1417 * the case that the buffer space is actually used to a large
1418 * extent and we run out of kernel memory we can simply drop
1419 * the new segments; TCP on the sender will just retransmit it
1420 * later. Setting the buffer size too big may only consume too
1421 * much kernel memory if the application doesn't read() from
1422 * the socket or packet loss or reordering makes use of the
1425 * The criteria to step up the receive buffer one notch are:
1426 * 1. Application has not set receive buffer size with
1427 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1428 * 2. the number of bytes received during 1/2 of an sRTT
1429 * is at least 3/8 of the current socket buffer size.
1430 * 3. receive buffer size has not hit maximal automatic size;
1432 * If all of the criteria are met we increaset the socket buffer
1433 * by a 1/2 (bounded by the max). This allows us to keep ahead
1434 * of slow-start but also makes it so our peer never gets limited
1435 * by our rwnd which we then open up causing a burst.
1437 * This algorithm does two steps per RTT at most and only if
1438 * we receive a bulk stream w/o packet losses or reorderings.
1439 * Shrinking the buffer during idle times is not necessary as
1440 * it doesn't consume any memory when idle.
1442 * TODO: Only step up if the application is actually serving
1443 * the buffer to better manage the socket buffer resources.
1446 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1447 struct tcpcb *tp, int tlen)
1451 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1452 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1453 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1454 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1455 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1456 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1457 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1459 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1461 /* Start over with next RTT. */
1465 tp->rfbuf_cnt += tlen; /* add up */
1471 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1472 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1474 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1475 int rstreason, todrop, win, incforsyn = 0;
1479 struct in_conninfo *inc;
1486 * The size of tcp_saveipgen must be the size of the max ip header,
1489 u_char tcp_saveipgen[IP6_HDR_LEN];
1490 struct tcphdr tcp_savetcp;
1493 thflags = th->th_flags;
1494 inc = &tp->t_inpcb->inp_inc;
1495 tp->sackhint.last_sack_ack = 0;
1497 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1500 INP_WLOCK_ASSERT(tp->t_inpcb);
1501 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1503 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1507 /* Save segment, if requested. */
1508 tcp_pcap_add(th, m, &(tp->t_inpkts));
1510 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1513 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1514 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1515 log(LOG_DEBUG, "%s; %s: "
1516 "SYN|FIN segment ignored (based on "
1517 "sysctl setting)\n", s, __func__);
1524 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1525 * check SEQ.ACK first.
1527 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1528 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1529 rstreason = BANDLIM_UNLIMITED;
1534 * Segment received on connection.
1535 * Reset idle time and keep-alive timer.
1536 * XXX: This should be done after segment
1537 * validation to ignore broken/spoofed segs.
1539 tp->t_rcvtime = ticks;
1542 * Scale up the window into a 32-bit value.
1543 * For the SYN_SENT state the scale is zero.
1545 tiwin = th->th_win << tp->snd_scale;
1547 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1551 * TCP ECN processing.
1553 if (tp->t_flags2 & TF2_ECN_PERMIT) {
1554 if (thflags & TH_CWR) {
1555 tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1556 tp->t_flags |= TF_ACKNOW;
1558 switch (iptos & IPTOS_ECN_MASK) {
1560 tp->t_flags2 |= TF2_ECN_SND_ECE;
1561 TCPSTAT_INC(tcps_ecn_ce);
1563 case IPTOS_ECN_ECT0:
1564 TCPSTAT_INC(tcps_ecn_ect0);
1566 case IPTOS_ECN_ECT1:
1567 TCPSTAT_INC(tcps_ecn_ect1);
1571 /* Process a packet differently from RFC3168. */
1572 cc_ecnpkt_handler(tp, th, iptos);
1574 /* Congestion experienced. */
1575 if (thflags & TH_ECE) {
1576 cc_cong_signal(tp, th, CC_ECN);
1581 * Parse options on any incoming segment.
1583 tcp_dooptions(&to, (u_char *)(th + 1),
1584 (th->th_off << 2) - sizeof(struct tcphdr),
1585 (thflags & TH_SYN) ? TO_SYN : 0);
1587 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1588 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1589 (to.to_flags & TOF_SIGNATURE) == 0) {
1590 TCPSTAT_INC(tcps_sig_err_sigopt);
1591 /* XXX: should drop? */
1595 * If echoed timestamp is later than the current time,
1596 * fall back to non RFC1323 RTT calculation. Normalize
1597 * timestamp if syncookies were used when this connection
1600 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1601 to.to_tsecr -= tp->ts_offset;
1602 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1604 else if (tp->t_flags & TF_PREVVALID &&
1605 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1606 cc_cong_signal(tp, th, CC_RTO_ERR);
1609 * Process options only when we get SYN/ACK back. The SYN case
1610 * for incoming connections is handled in tcp_syncache.
1611 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1612 * or <SYN,ACK>) segment itself is never scaled.
1613 * XXX this is traditional behavior, may need to be cleaned up.
1615 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1616 /* Handle parallel SYN for ECN */
1617 if (!(thflags & TH_ACK) &&
1618 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1619 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1620 tp->t_flags2 |= TF2_ECN_PERMIT;
1621 tp->t_flags2 |= TF2_ECN_SND_ECE;
1622 TCPSTAT_INC(tcps_ecn_shs);
1624 if ((to.to_flags & TOF_SCALE) &&
1625 (tp->t_flags & TF_REQ_SCALE)) {
1626 tp->t_flags |= TF_RCVD_SCALE;
1627 tp->snd_scale = to.to_wscale;
1629 tp->t_flags &= ~TF_REQ_SCALE;
1631 * Initial send window. It will be updated with
1632 * the next incoming segment to the scaled value.
1634 tp->snd_wnd = th->th_win;
1635 if ((to.to_flags & TOF_TS) &&
1636 (tp->t_flags & TF_REQ_TSTMP)) {
1637 tp->t_flags |= TF_RCVD_TSTMP;
1638 tp->ts_recent = to.to_tsval;
1639 tp->ts_recent_age = tcp_ts_getticks();
1641 tp->t_flags &= ~TF_REQ_TSTMP;
1642 if (to.to_flags & TOF_MSS)
1643 tcp_mss(tp, to.to_mss);
1644 if ((tp->t_flags & TF_SACK_PERMIT) &&
1645 (to.to_flags & TOF_SACKPERM) == 0)
1646 tp->t_flags &= ~TF_SACK_PERMIT;
1647 if (IS_FASTOPEN(tp->t_flags)) {
1648 if (to.to_flags & TOF_FASTOPEN) {
1651 if (to.to_flags & TOF_MSS)
1654 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1658 tcp_fastopen_update_cache(tp, mss,
1659 to.to_tfo_len, to.to_tfo_cookie);
1661 tcp_fastopen_disable_path(tp);
1666 * If timestamps were negotiated during SYN/ACK they should
1667 * appear on every segment during this session and vice versa.
1669 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1670 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1671 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1672 "no action\n", s, __func__);
1676 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1677 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1678 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1679 "no action\n", s, __func__);
1685 * Header prediction: check for the two common cases
1686 * of a uni-directional data xfer. If the packet has
1687 * no control flags, is in-sequence, the window didn't
1688 * change and we're not retransmitting, it's a
1689 * candidate. If the length is zero and the ack moved
1690 * forward, we're the sender side of the xfer. Just
1691 * free the data acked & wake any higher level process
1692 * that was blocked waiting for space. If the length
1693 * is non-zero and the ack didn't move, we're the
1694 * receiver side. If we're getting packets in-order
1695 * (the reassembly queue is empty), add the data to
1696 * the socket buffer and note that we need a delayed ack.
1697 * Make sure that the hidden state-flags are also off.
1698 * Since we check for TCPS_ESTABLISHED first, it can only
1701 if (tp->t_state == TCPS_ESTABLISHED &&
1702 th->th_seq == tp->rcv_nxt &&
1703 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1704 tp->snd_nxt == tp->snd_max &&
1705 tiwin && tiwin == tp->snd_wnd &&
1706 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1708 ((to.to_flags & TOF_TS) == 0 ||
1709 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1711 * If last ACK falls within this segment's sequence numbers,
1712 * record the timestamp.
1713 * NOTE that the test is modified according to the latest
1714 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1716 if ((to.to_flags & TOF_TS) != 0 &&
1717 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1718 tp->ts_recent_age = tcp_ts_getticks();
1719 tp->ts_recent = to.to_tsval;
1723 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1724 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1725 !IN_RECOVERY(tp->t_flags) &&
1726 (to.to_flags & TOF_SACK) == 0 &&
1727 TAILQ_EMPTY(&tp->snd_holes)) {
1729 * This is a pure ack for outstanding data.
1731 TCPSTAT_INC(tcps_predack);
1734 * "bad retransmit" recovery without timestamps.
1736 if ((to.to_flags & TOF_TS) == 0 &&
1737 tp->t_rxtshift == 1 &&
1738 tp->t_flags & TF_PREVVALID &&
1739 (int)(ticks - tp->t_badrxtwin) < 0) {
1740 cc_cong_signal(tp, th, CC_RTO_ERR);
1744 * Recalculate the transmit timer / rtt.
1746 * Some boxes send broken timestamp replies
1747 * during the SYN+ACK phase, ignore
1748 * timestamps of 0 or we could calculate a
1749 * huge RTT and blow up the retransmit timer.
1751 if ((to.to_flags & TOF_TS) != 0 &&
1755 t = tcp_ts_getticks() - to.to_tsecr;
1756 if (!tp->t_rttlow || tp->t_rttlow > t)
1759 TCP_TS_TO_TICKS(t) + 1);
1760 } else if (tp->t_rtttime &&
1761 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1762 if (!tp->t_rttlow ||
1763 tp->t_rttlow > ticks - tp->t_rtttime)
1764 tp->t_rttlow = ticks - tp->t_rtttime;
1766 ticks - tp->t_rtttime);
1768 acked = BYTES_THIS_ACK(tp, th);
1771 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1772 hhook_run_tcp_est_in(tp, th, &to);
1775 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1776 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1777 sbdrop(&so->so_snd, acked);
1778 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1779 SEQ_LEQ(th->th_ack, tp->snd_recover))
1780 tp->snd_recover = th->th_ack - 1;
1783 * Let the congestion control algorithm update
1784 * congestion control related information. This
1785 * typically means increasing the congestion
1788 cc_ack_received(tp, th, nsegs, CC_ACK);
1790 tp->snd_una = th->th_ack;
1792 * Pull snd_wl2 up to prevent seq wrap relative
1795 tp->snd_wl2 = th->th_ack;
1800 * If all outstanding data are acked, stop
1801 * retransmit timer, otherwise restart timer
1802 * using current (possibly backed-off) value.
1803 * If process is waiting for space,
1804 * wakeup/selwakeup/signal. If data
1805 * are ready to send, let tcp_output
1806 * decide between more output or persist.
1809 if (so->so_options & SO_DEBUG)
1810 tcp_trace(TA_INPUT, ostate, tp,
1811 (void *)tcp_saveipgen,
1814 TCP_PROBE3(debug__input, tp, th, m);
1815 if (tp->snd_una == tp->snd_max)
1816 tcp_timer_activate(tp, TT_REXMT, 0);
1817 else if (!tcp_timer_active(tp, TT_PERSIST))
1818 tcp_timer_activate(tp, TT_REXMT,
1821 if (sbavail(&so->so_snd))
1822 (void) tp->t_fb->tfb_tcp_output(tp);
1825 } else if (th->th_ack == tp->snd_una &&
1826 tlen <= sbspace(&so->so_rcv)) {
1827 int newsize = 0; /* automatic sockbuf scaling */
1830 * This is a pure, in-sequence data packet with
1831 * nothing on the reassembly queue and we have enough
1832 * buffer space to take it.
1834 /* Clean receiver SACK report if present */
1835 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1836 tcp_clean_sackreport(tp);
1837 TCPSTAT_INC(tcps_preddat);
1838 tp->rcv_nxt += tlen;
1840 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1841 (tp->t_fbyte_in == 0)) {
1842 tp->t_fbyte_in = ticks;
1843 if (tp->t_fbyte_in == 0)
1845 if (tp->t_fbyte_out && tp->t_fbyte_in)
1846 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1849 * Pull snd_wl1 up to prevent seq wrap relative to
1852 tp->snd_wl1 = th->th_seq;
1854 * Pull rcv_up up to prevent seq wrap relative to
1857 tp->rcv_up = tp->rcv_nxt;
1858 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1859 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1861 if (so->so_options & SO_DEBUG)
1862 tcp_trace(TA_INPUT, ostate, tp,
1863 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1865 TCP_PROBE3(debug__input, tp, th, m);
1867 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1869 /* Add data to socket buffer. */
1870 SOCKBUF_LOCK(&so->so_rcv);
1871 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1875 * Set new socket buffer size.
1876 * Give up when limit is reached.
1879 if (!sbreserve_locked(&so->so_rcv,
1881 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1882 m_adj(m, drop_hdrlen); /* delayed header drop */
1883 sbappendstream_locked(&so->so_rcv, m, 0);
1885 /* NB: sorwakeup_locked() does an implicit unlock. */
1886 sorwakeup_locked(so);
1887 if (DELAY_ACK(tp, tlen)) {
1888 tp->t_flags |= TF_DELACK;
1890 tp->t_flags |= TF_ACKNOW;
1891 tp->t_fb->tfb_tcp_output(tp);
1898 * Calculate amount of space in receive window,
1899 * and then do TCP input processing.
1900 * Receive window is amount of space in rcv queue,
1901 * but not less than advertised window.
1903 win = sbspace(&so->so_rcv);
1906 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1908 switch (tp->t_state) {
1910 * If the state is SYN_RECEIVED:
1911 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1913 case TCPS_SYN_RECEIVED:
1914 if ((thflags & TH_ACK) &&
1915 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1916 SEQ_GT(th->th_ack, tp->snd_max))) {
1917 rstreason = BANDLIM_RST_OPENPORT;
1920 if (IS_FASTOPEN(tp->t_flags)) {
1922 * When a TFO connection is in SYN_RECEIVED, the
1923 * only valid packets are the initial SYN, a
1924 * retransmit/copy of the initial SYN (possibly with
1925 * a subset of the original data), a valid ACK, a
1928 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1929 rstreason = BANDLIM_RST_OPENPORT;
1931 } else if (thflags & TH_SYN) {
1932 /* non-initial SYN is ignored */
1933 if ((tcp_timer_active(tp, TT_DELACK) ||
1934 tcp_timer_active(tp, TT_REXMT)))
1936 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1943 * If the state is SYN_SENT:
1944 * if seg contains a RST with valid ACK (SEQ.ACK has already
1945 * been verified), then drop the connection.
1946 * if seg contains a RST without an ACK, drop the seg.
1947 * if seg does not contain SYN, then drop the seg.
1948 * Otherwise this is an acceptable SYN segment
1949 * initialize tp->rcv_nxt and tp->irs
1950 * if seg contains ack then advance tp->snd_una
1951 * if seg contains an ECE and ECN support is enabled, the stream
1953 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1954 * arrange for segment to be acked (eventually)
1955 * continue processing rest of data/controls, beginning with URG
1958 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1959 TCP_PROBE5(connect__refused, NULL, tp,
1961 tp = tcp_drop(tp, ECONNREFUSED);
1963 if (thflags & TH_RST)
1965 if (!(thflags & TH_SYN))
1968 tp->irs = th->th_seq;
1970 if (thflags & TH_ACK) {
1971 int tfo_partial_ack = 0;
1973 TCPSTAT_INC(tcps_connects);
1976 mac_socketpeer_set_from_mbuf(m, so);
1978 /* Do window scaling on this connection? */
1979 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1980 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1981 tp->rcv_scale = tp->request_r_scale;
1983 tp->rcv_adv += min(tp->rcv_wnd,
1984 TCP_MAXWIN << tp->rcv_scale);
1985 tp->snd_una++; /* SYN is acked */
1987 * If not all the data that was sent in the TFO SYN
1988 * has been acked, resend the remainder right away.
1990 if (IS_FASTOPEN(tp->t_flags) &&
1991 (tp->snd_una != tp->snd_max)) {
1992 tp->snd_nxt = th->th_ack;
1993 tfo_partial_ack = 1;
1996 * If there's data, delay ACK; if there's also a FIN
1997 * ACKNOW will be turned on later.
1999 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2000 tcp_timer_activate(tp, TT_DELACK,
2003 tp->t_flags |= TF_ACKNOW;
2005 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2006 (V_tcp_do_ecn == 1)) {
2007 tp->t_flags2 |= TF2_ECN_PERMIT;
2008 TCPSTAT_INC(tcps_ecn_shs);
2012 * Received <SYN,ACK> in SYN_SENT[*] state.
2014 * SYN_SENT --> ESTABLISHED
2015 * SYN_SENT* --> FIN_WAIT_1
2017 tp->t_starttime = ticks;
2018 if (tp->t_flags & TF_NEEDFIN) {
2019 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2020 tp->t_flags &= ~TF_NEEDFIN;
2023 tcp_state_change(tp, TCPS_ESTABLISHED);
2024 TCP_PROBE5(connect__established, NULL, tp,
2027 tcp_timer_activate(tp, TT_KEEP,
2032 * Received initial SYN in SYN-SENT[*] state =>
2033 * simultaneous open.
2034 * If it succeeds, connection is * half-synchronized.
2035 * Otherwise, do 3-way handshake:
2036 * SYN-SENT -> SYN-RECEIVED
2037 * SYN-SENT* -> SYN-RECEIVED*
2039 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2040 tcp_timer_activate(tp, TT_REXMT, 0);
2041 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2044 INP_WLOCK_ASSERT(tp->t_inpcb);
2047 * Advance th->th_seq to correspond to first data byte.
2048 * If data, trim to stay within window,
2049 * dropping FIN if necessary.
2052 if (tlen > tp->rcv_wnd) {
2053 todrop = tlen - tp->rcv_wnd;
2057 TCPSTAT_INC(tcps_rcvpackafterwin);
2058 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2060 tp->snd_wl1 = th->th_seq - 1;
2061 tp->rcv_up = th->th_seq;
2063 * Client side of transaction: already sent SYN and data.
2064 * If the remote host used T/TCP to validate the SYN,
2065 * our data will be ACK'd; if so, enter normal data segment
2066 * processing in the middle of step 5, ack processing.
2067 * Otherwise, goto step 6.
2069 if (thflags & TH_ACK)
2075 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2076 * do normal processing.
2078 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2082 break; /* continue normal processing */
2086 * States other than LISTEN or SYN_SENT.
2087 * First check the RST flag and sequence number since reset segments
2088 * are exempt from the timestamp and connection count tests. This
2089 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2090 * below which allowed reset segments in half the sequence space
2091 * to fall though and be processed (which gives forged reset
2092 * segments with a random sequence number a 50 percent chance of
2093 * killing a connection).
2094 * Then check timestamp, if present.
2095 * Then check the connection count, if present.
2096 * Then check that at least some bytes of segment are within
2097 * receive window. If segment begins before rcv_nxt,
2098 * drop leading data (and SYN); if nothing left, just ack.
2100 if (thflags & TH_RST) {
2102 * RFC5961 Section 3.2
2104 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2105 * - If RST is in window, we send challenge ACK.
2107 * Note: to take into account delayed ACKs, we should
2108 * test against last_ack_sent instead of rcv_nxt.
2109 * Note 2: we handle special case of closed window, not
2110 * covered by the RFC.
2112 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2113 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2114 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2115 KASSERT(tp->t_state != TCPS_SYN_SENT,
2116 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2119 if (V_tcp_insecure_rst ||
2120 tp->last_ack_sent == th->th_seq) {
2121 TCPSTAT_INC(tcps_drops);
2122 /* Drop the connection. */
2123 switch (tp->t_state) {
2124 case TCPS_SYN_RECEIVED:
2125 so->so_error = ECONNREFUSED;
2127 case TCPS_ESTABLISHED:
2128 case TCPS_FIN_WAIT_1:
2129 case TCPS_FIN_WAIT_2:
2130 case TCPS_CLOSE_WAIT:
2133 so->so_error = ECONNRESET;
2140 TCPSTAT_INC(tcps_badrst);
2141 /* Send challenge ACK. */
2142 tcp_respond(tp, mtod(m, void *), th, m,
2143 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2144 tp->last_ack_sent = tp->rcv_nxt;
2152 * RFC5961 Section 4.2
2153 * Send challenge ACK for any SYN in synchronized state.
2155 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2156 tp->t_state != TCPS_SYN_RECEIVED) {
2157 TCPSTAT_INC(tcps_badsyn);
2158 if (V_tcp_insecure_syn &&
2159 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2160 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2161 tp = tcp_drop(tp, ECONNRESET);
2162 rstreason = BANDLIM_UNLIMITED;
2164 /* Send challenge ACK. */
2165 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2166 tp->snd_nxt, TH_ACK);
2167 tp->last_ack_sent = tp->rcv_nxt;
2174 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2175 * and it's less than ts_recent, drop it.
2177 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2178 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2179 /* Check to see if ts_recent is over 24 days old. */
2180 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2182 * Invalidate ts_recent. If this segment updates
2183 * ts_recent, the age will be reset later and ts_recent
2184 * will get a valid value. If it does not, setting
2185 * ts_recent to zero will at least satisfy the
2186 * requirement that zero be placed in the timestamp
2187 * echo reply when ts_recent isn't valid. The
2188 * age isn't reset until we get a valid ts_recent
2189 * because we don't want out-of-order segments to be
2190 * dropped when ts_recent is old.
2194 TCPSTAT_INC(tcps_rcvduppack);
2195 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2196 TCPSTAT_INC(tcps_pawsdrop);
2204 * In the SYN-RECEIVED state, validate that the packet belongs to
2205 * this connection before trimming the data to fit the receive
2206 * window. Check the sequence number versus IRS since we know
2207 * the sequence numbers haven't wrapped. This is a partial fix
2208 * for the "LAND" DoS attack.
2210 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2211 rstreason = BANDLIM_RST_OPENPORT;
2215 todrop = tp->rcv_nxt - th->th_seq;
2217 if (thflags & TH_SYN) {
2227 * Following if statement from Stevens, vol. 2, p. 960.
2230 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2232 * Any valid FIN must be to the left of the window.
2233 * At this point the FIN must be a duplicate or out
2234 * of sequence; drop it.
2239 * Send an ACK to resynchronize and drop any data.
2240 * But keep on processing for RST or ACK.
2242 tp->t_flags |= TF_ACKNOW;
2244 TCPSTAT_INC(tcps_rcvduppack);
2245 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2247 TCPSTAT_INC(tcps_rcvpartduppack);
2248 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2251 * DSACK - add SACK block for dropped range
2253 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2254 tcp_update_sack_list(tp, th->th_seq,
2255 th->th_seq + todrop);
2257 * ACK now, as the next in-sequence segment
2258 * will clear the DSACK block again
2260 tp->t_flags |= TF_ACKNOW;
2262 drop_hdrlen += todrop; /* drop from the top afterwards */
2263 th->th_seq += todrop;
2265 if (th->th_urp > todrop)
2266 th->th_urp -= todrop;
2274 * If new data are received on a connection after the
2275 * user processes are gone, then RST the other end.
2277 if ((so->so_state & SS_NOFDREF) &&
2278 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2279 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2280 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2281 "after socket was closed, "
2282 "sending RST and removing tcpcb\n",
2283 s, __func__, tcpstates[tp->t_state], tlen);
2287 TCPSTAT_INC(tcps_rcvafterclose);
2288 rstreason = BANDLIM_UNLIMITED;
2293 * If segment ends after window, drop trailing data
2294 * (and PUSH and FIN); if nothing left, just ACK.
2296 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2298 TCPSTAT_INC(tcps_rcvpackafterwin);
2299 if (todrop >= tlen) {
2300 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2302 * If window is closed can only take segments at
2303 * window edge, and have to drop data and PUSH from
2304 * incoming segments. Continue processing, but
2305 * remember to ack. Otherwise, drop segment
2308 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2309 tp->t_flags |= TF_ACKNOW;
2310 TCPSTAT_INC(tcps_rcvwinprobe);
2314 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2317 thflags &= ~(TH_PUSH|TH_FIN);
2321 * If last ACK falls within this segment's sequence numbers,
2322 * record its timestamp.
2324 * 1) That the test incorporates suggestions from the latest
2325 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2326 * 2) That updating only on newer timestamps interferes with
2327 * our earlier PAWS tests, so this check should be solely
2328 * predicated on the sequence space of this segment.
2329 * 3) That we modify the segment boundary check to be
2330 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2331 * instead of RFC1323's
2332 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2333 * This modified check allows us to overcome RFC1323's
2334 * limitations as described in Stevens TCP/IP Illustrated
2335 * Vol. 2 p.869. In such cases, we can still calculate the
2336 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2338 if ((to.to_flags & TOF_TS) != 0 &&
2339 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2340 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2341 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2342 tp->ts_recent_age = tcp_ts_getticks();
2343 tp->ts_recent = to.to_tsval;
2347 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2348 * flag is on (half-synchronized state), then queue data for
2349 * later processing; else drop segment and return.
2351 if ((thflags & TH_ACK) == 0) {
2352 if (tp->t_state == TCPS_SYN_RECEIVED ||
2353 (tp->t_flags & TF_NEEDSYN)) {
2354 if (tp->t_state == TCPS_SYN_RECEIVED &&
2355 IS_FASTOPEN(tp->t_flags)) {
2356 tp->snd_wnd = tiwin;
2360 } else if (tp->t_flags & TF_ACKNOW)
2369 switch (tp->t_state) {
2371 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2372 * ESTABLISHED state and continue processing.
2373 * The ACK was checked above.
2375 case TCPS_SYN_RECEIVED:
2377 TCPSTAT_INC(tcps_connects);
2379 /* Do window scaling? */
2380 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2381 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2382 tp->rcv_scale = tp->request_r_scale;
2384 tp->snd_wnd = tiwin;
2387 * SYN-RECEIVED -> ESTABLISHED
2388 * SYN-RECEIVED* -> FIN-WAIT-1
2390 tp->t_starttime = ticks;
2391 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2392 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2393 tp->t_tfo_pending = NULL;
2395 if (tp->t_flags & TF_NEEDFIN) {
2396 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2397 tp->t_flags &= ~TF_NEEDFIN;
2399 tcp_state_change(tp, TCPS_ESTABLISHED);
2400 TCP_PROBE5(accept__established, NULL, tp,
2403 * TFO connections call cc_conn_init() during SYN
2404 * processing. Calling it again here for such
2405 * connections is not harmless as it would undo the
2406 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2409 if (!IS_FASTOPEN(tp->t_flags))
2411 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2414 * Account for the ACK of our SYN prior to
2415 * regular ACK processing below, except for
2416 * simultaneous SYN, which is handled later.
2418 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2421 * If segment contains data or ACK, will call tcp_reass()
2422 * later; if not, do so now to pass queued data to user.
2424 if (tlen == 0 && (thflags & TH_FIN) == 0)
2425 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2427 tp->snd_wl1 = th->th_seq - 1;
2431 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2432 * ACKs. If the ack is in the range
2433 * tp->snd_una < th->th_ack <= tp->snd_max
2434 * then advance tp->snd_una to th->th_ack and drop
2435 * data from the retransmission queue. If this ACK reflects
2436 * more up to date window information we update our window information.
2438 case TCPS_ESTABLISHED:
2439 case TCPS_FIN_WAIT_1:
2440 case TCPS_FIN_WAIT_2:
2441 case TCPS_CLOSE_WAIT:
2444 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2445 TCPSTAT_INC(tcps_rcvacktoomuch);
2448 if ((tp->t_flags & TF_SACK_PERMIT) &&
2449 ((to.to_flags & TOF_SACK) ||
2450 !TAILQ_EMPTY(&tp->snd_holes)))
2451 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2454 * Reset the value so that previous (valid) value
2455 * from the last ack with SACK doesn't get used.
2457 tp->sackhint.sacked_bytes = 0;
2460 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2461 hhook_run_tcp_est_in(tp, th, &to);
2464 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2467 maxseg = tcp_maxseg(tp);
2469 (tiwin == tp->snd_wnd ||
2470 (tp->t_flags & TF_SACK_PERMIT))) {
2472 * If this is the first time we've seen a
2473 * FIN from the remote, this is not a
2474 * duplicate and it needs to be processed
2475 * normally. This happens during a
2476 * simultaneous close.
2478 if ((thflags & TH_FIN) &&
2479 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2483 TCPSTAT_INC(tcps_rcvdupack);
2485 * If we have outstanding data (other than
2486 * a window probe), this is a completely
2487 * duplicate ack (ie, window info didn't
2488 * change and FIN isn't set),
2489 * the ack is the biggest we've
2490 * seen and we've seen exactly our rexmt
2491 * threshold of them, assume a packet
2492 * has been dropped and retransmit it.
2493 * Kludge snd_nxt & the congestion
2494 * window so we send only this one
2497 * We know we're losing at the current
2498 * window size so do congestion avoidance
2499 * (set ssthresh to half the current window
2500 * and pull our congestion window back to
2501 * the new ssthresh).
2503 * Dup acks mean that packets have left the
2504 * network (they're now cached at the receiver)
2505 * so bump cwnd by the amount in the receiver
2506 * to keep a constant cwnd packets in the
2509 * When using TCP ECN, notify the peer that
2510 * we reduced the cwnd.
2513 * Following 2 kinds of acks should not affect
2516 * 2) Acks with SACK but without any new SACK
2517 * information in them. These could result from
2518 * any anomaly in the network like a switch
2519 * duplicating packets or a possible DoS attack.
2521 if (th->th_ack != tp->snd_una ||
2522 ((tp->t_flags & TF_SACK_PERMIT) &&
2525 else if (!tcp_timer_active(tp, TT_REXMT))
2527 else if (++tp->t_dupacks > tcprexmtthresh ||
2528 IN_FASTRECOVERY(tp->t_flags)) {
2529 cc_ack_received(tp, th, nsegs,
2531 if ((tp->t_flags & TF_SACK_PERMIT) &&
2532 IN_FASTRECOVERY(tp->t_flags)) {
2536 * Compute the amount of data in flight first.
2537 * We can inject new data into the pipe iff
2538 * we have less than 1/2 the original window's
2539 * worth of data in flight.
2541 if (V_tcp_do_rfc6675_pipe)
2542 awnd = tcp_compute_pipe(tp);
2544 awnd = (tp->snd_nxt - tp->snd_fack) +
2545 tp->sackhint.sack_bytes_rexmit;
2547 if (awnd < tp->snd_ssthresh) {
2548 tp->snd_cwnd += maxseg;
2549 if (tp->snd_cwnd > tp->snd_ssthresh)
2550 tp->snd_cwnd = tp->snd_ssthresh;
2553 tp->snd_cwnd += maxseg;
2554 (void) tp->t_fb->tfb_tcp_output(tp);
2556 } else if (tp->t_dupacks == tcprexmtthresh) {
2557 tcp_seq onxt = tp->snd_nxt;
2560 * If we're doing sack, check to
2561 * see if we're already in sack
2562 * recovery. If we're not doing sack,
2563 * check to see if we're in newreno
2566 if (tp->t_flags & TF_SACK_PERMIT) {
2567 if (IN_FASTRECOVERY(tp->t_flags)) {
2572 if (SEQ_LEQ(th->th_ack,
2578 /* Congestion signal before ack. */
2579 cc_cong_signal(tp, th, CC_NDUPACK);
2580 cc_ack_received(tp, th, nsegs,
2582 tcp_timer_activate(tp, TT_REXMT, 0);
2584 if (tp->t_flags & TF_SACK_PERMIT) {
2586 tcps_sack_recovery_episode);
2587 tp->snd_recover = tp->snd_nxt;
2588 tp->snd_cwnd = maxseg;
2589 (void) tp->t_fb->tfb_tcp_output(tp);
2592 tp->snd_nxt = th->th_ack;
2593 tp->snd_cwnd = maxseg;
2594 (void) tp->t_fb->tfb_tcp_output(tp);
2595 KASSERT(tp->snd_limited <= 2,
2596 ("%s: tp->snd_limited too big",
2598 tp->snd_cwnd = tp->snd_ssthresh +
2600 (tp->t_dupacks - tp->snd_limited);
2601 if (SEQ_GT(onxt, tp->snd_nxt))
2604 } else if (V_tcp_do_rfc3042) {
2606 * Process first and second duplicate
2607 * ACKs. Each indicates a segment
2608 * leaving the network, creating room
2609 * for more. Make sure we can send a
2610 * packet on reception of each duplicate
2611 * ACK by increasing snd_cwnd by one
2612 * segment. Restore the original
2613 * snd_cwnd after packet transmission.
2615 cc_ack_received(tp, th, nsegs,
2617 uint32_t oldcwnd = tp->snd_cwnd;
2618 tcp_seq oldsndmax = tp->snd_max;
2622 KASSERT(tp->t_dupacks == 1 ||
2624 ("%s: dupacks not 1 or 2",
2626 if (tp->t_dupacks == 1)
2627 tp->snd_limited = 0;
2629 (tp->snd_nxt - tp->snd_una) +
2630 (tp->t_dupacks - tp->snd_limited) *
2633 * Only call tcp_output when there
2634 * is new data available to be sent.
2635 * Otherwise we would send pure ACKs.
2637 SOCKBUF_LOCK(&so->so_snd);
2638 avail = sbavail(&so->so_snd) -
2639 (tp->snd_nxt - tp->snd_una);
2640 SOCKBUF_UNLOCK(&so->so_snd);
2642 (void) tp->t_fb->tfb_tcp_output(tp);
2643 sent = tp->snd_max - oldsndmax;
2644 if (sent > maxseg) {
2645 KASSERT((tp->t_dupacks == 2 &&
2646 tp->snd_limited == 0) ||
2647 (sent == maxseg + 1 &&
2648 tp->t_flags & TF_SENTFIN),
2649 ("%s: sent too much",
2651 tp->snd_limited = 2;
2652 } else if (sent > 0)
2654 tp->snd_cwnd = oldcwnd;
2661 * This ack is advancing the left edge, reset the
2666 * If this ack also has new SACK info, increment the
2667 * counter as per rfc6675. The variable
2668 * sack_changed tracks all changes to the SACK
2669 * scoreboard, including when partial ACKs without
2670 * SACK options are received, and clear the scoreboard
2671 * from the left side. Such partial ACKs should not be
2672 * counted as dupacks here.
2674 if ((tp->t_flags & TF_SACK_PERMIT) &&
2675 (to.to_flags & TOF_SACK) &&
2680 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2681 ("%s: th_ack <= snd_una", __func__));
2684 * If the congestion window was inflated to account
2685 * for the other side's cached packets, retract it.
2687 if (IN_FASTRECOVERY(tp->t_flags)) {
2688 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2689 if (tp->t_flags & TF_SACK_PERMIT)
2690 tcp_sack_partialack(tp, th);
2692 tcp_newreno_partial_ack(tp, th);
2694 cc_post_recovery(tp, th);
2697 * If we reach this point, ACK is not a duplicate,
2698 * i.e., it ACKs something we sent.
2700 if (tp->t_flags & TF_NEEDSYN) {
2702 * T/TCP: Connection was half-synchronized, and our
2703 * SYN has been ACK'd (so connection is now fully
2704 * synchronized). Go to non-starred state,
2705 * increment snd_una for ACK of SYN, and check if
2706 * we can do window scaling.
2708 tp->t_flags &= ~TF_NEEDSYN;
2710 /* Do window scaling? */
2711 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2712 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2713 tp->rcv_scale = tp->request_r_scale;
2714 /* Send window already scaled. */
2719 INP_WLOCK_ASSERT(tp->t_inpcb);
2722 * Adjust for the SYN bit in sequence space,
2723 * but don't account for it in cwnd calculations.
2724 * This is for the SYN_RECEIVED, non-simultaneous
2725 * SYN case. SYN_SENT and simultaneous SYN are
2726 * treated elsewhere.
2730 acked = BYTES_THIS_ACK(tp, th);
2731 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2732 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2733 tp->snd_una, th->th_ack, tp, m));
2734 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2735 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2738 * If we just performed our first retransmit, and the ACK
2739 * arrives within our recovery window, then it was a mistake
2740 * to do the retransmit in the first place. Recover our
2741 * original cwnd and ssthresh, and proceed to transmit where
2744 if (tp->t_rxtshift == 1 &&
2745 tp->t_flags & TF_PREVVALID &&
2747 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2748 cc_cong_signal(tp, th, CC_RTO_ERR);
2751 * If we have a timestamp reply, update smoothed
2752 * round trip time. If no timestamp is present but
2753 * transmit timer is running and timed sequence
2754 * number was acked, update smoothed round trip time.
2755 * Since we now have an rtt measurement, cancel the
2756 * timer backoff (cf., Phil Karn's retransmit alg.).
2757 * Recompute the initial retransmit timer.
2759 * Some boxes send broken timestamp replies
2760 * during the SYN+ACK phase, ignore
2761 * timestamps of 0 or we could calculate a
2762 * huge RTT and blow up the retransmit timer.
2764 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2767 t = tcp_ts_getticks() - to.to_tsecr;
2768 if (!tp->t_rttlow || tp->t_rttlow > t)
2770 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2771 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2772 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2773 tp->t_rttlow = ticks - tp->t_rtttime;
2774 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2778 * If all outstanding data is acked, stop retransmit
2779 * timer and remember to restart (more output or persist).
2780 * If there is more data to be acked, restart retransmit
2781 * timer, using current (possibly backed-off) value.
2783 if (th->th_ack == tp->snd_max) {
2784 tcp_timer_activate(tp, TT_REXMT, 0);
2786 } else if (!tcp_timer_active(tp, TT_PERSIST))
2787 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2790 * If no data (only SYN) was ACK'd,
2791 * skip rest of ACK processing.
2797 * Let the congestion control algorithm update congestion
2798 * control related information. This typically means increasing
2799 * the congestion window.
2801 cc_ack_received(tp, th, nsegs, CC_ACK);
2803 SOCKBUF_LOCK(&so->so_snd);
2804 if (acked > sbavail(&so->so_snd)) {
2805 if (tp->snd_wnd >= sbavail(&so->so_snd))
2806 tp->snd_wnd -= sbavail(&so->so_snd);
2809 mfree = sbcut_locked(&so->so_snd,
2810 (int)sbavail(&so->so_snd));
2813 mfree = sbcut_locked(&so->so_snd, acked);
2814 if (tp->snd_wnd >= (uint32_t) acked)
2815 tp->snd_wnd -= acked;
2820 /* NB: sowwakeup_locked() does an implicit unlock. */
2821 sowwakeup_locked(so);
2823 /* Detect una wraparound. */
2824 if (!IN_RECOVERY(tp->t_flags) &&
2825 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2826 SEQ_LEQ(th->th_ack, tp->snd_recover))
2827 tp->snd_recover = th->th_ack - 1;
2828 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2829 if (IN_RECOVERY(tp->t_flags) &&
2830 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2831 EXIT_RECOVERY(tp->t_flags);
2833 tp->snd_una = th->th_ack;
2834 if (tp->t_flags & TF_SACK_PERMIT) {
2835 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2836 tp->snd_recover = tp->snd_una;
2838 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2839 tp->snd_nxt = tp->snd_una;
2841 switch (tp->t_state) {
2843 * In FIN_WAIT_1 STATE in addition to the processing
2844 * for the ESTABLISHED state if our FIN is now acknowledged
2845 * then enter FIN_WAIT_2.
2847 case TCPS_FIN_WAIT_1:
2848 if (ourfinisacked) {
2850 * If we can't receive any more
2851 * data, then closing user can proceed.
2852 * Starting the timer is contrary to the
2853 * specification, but if we don't get a FIN
2854 * we'll hang forever.
2857 * we should release the tp also, and use a
2860 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2861 soisdisconnected(so);
2862 tcp_timer_activate(tp, TT_2MSL,
2863 (tcp_fast_finwait2_recycle ?
2864 tcp_finwait2_timeout :
2867 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2872 * In CLOSING STATE in addition to the processing for
2873 * the ESTABLISHED state if the ACK acknowledges our FIN
2874 * then enter the TIME-WAIT state, otherwise ignore
2878 if (ourfinisacked) {
2886 * In LAST_ACK, we may still be waiting for data to drain
2887 * and/or to be acked, as well as for the ack of our FIN.
2888 * If our FIN is now acknowledged, delete the TCB,
2889 * enter the closed state and return.
2892 if (ourfinisacked) {
2901 INP_WLOCK_ASSERT(tp->t_inpcb);
2904 * Update window information.
2905 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2907 if ((thflags & TH_ACK) &&
2908 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2909 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2910 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2911 /* keep track of pure window updates */
2913 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2914 TCPSTAT_INC(tcps_rcvwinupd);
2915 tp->snd_wnd = tiwin;
2916 tp->snd_wl1 = th->th_seq;
2917 tp->snd_wl2 = th->th_ack;
2918 if (tp->snd_wnd > tp->max_sndwnd)
2919 tp->max_sndwnd = tp->snd_wnd;
2924 * Process segments with URG.
2926 if ((thflags & TH_URG) && th->th_urp &&
2927 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2929 * This is a kludge, but if we receive and accept
2930 * random urgent pointers, we'll crash in
2931 * soreceive. It's hard to imagine someone
2932 * actually wanting to send this much urgent data.
2934 SOCKBUF_LOCK(&so->so_rcv);
2935 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2936 th->th_urp = 0; /* XXX */
2937 thflags &= ~TH_URG; /* XXX */
2938 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2939 goto dodata; /* XXX */
2942 * If this segment advances the known urgent pointer,
2943 * then mark the data stream. This should not happen
2944 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2945 * a FIN has been received from the remote side.
2946 * In these states we ignore the URG.
2948 * According to RFC961 (Assigned Protocols),
2949 * the urgent pointer points to the last octet
2950 * of urgent data. We continue, however,
2951 * to consider it to indicate the first octet
2952 * of data past the urgent section as the original
2953 * spec states (in one of two places).
2955 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2956 tp->rcv_up = th->th_seq + th->th_urp;
2957 so->so_oobmark = sbavail(&so->so_rcv) +
2958 (tp->rcv_up - tp->rcv_nxt) - 1;
2959 if (so->so_oobmark == 0)
2960 so->so_rcv.sb_state |= SBS_RCVATMARK;
2962 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2964 SOCKBUF_UNLOCK(&so->so_rcv);
2966 * Remove out of band data so doesn't get presented to user.
2967 * This can happen independent of advancing the URG pointer,
2968 * but if two URG's are pending at once, some out-of-band
2969 * data may creep in... ick.
2971 if (th->th_urp <= (uint32_t)tlen &&
2972 !(so->so_options & SO_OOBINLINE)) {
2973 /* hdr drop is delayed */
2974 tcp_pulloutofband(so, th, m, drop_hdrlen);
2978 * If no out of band data is expected,
2979 * pull receive urgent pointer along
2980 * with the receive window.
2982 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2983 tp->rcv_up = tp->rcv_nxt;
2986 INP_WLOCK_ASSERT(tp->t_inpcb);
2989 * Process the segment text, merging it into the TCP sequencing queue,
2990 * and arranging for acknowledgment of receipt if necessary.
2991 * This process logically involves adjusting tp->rcv_wnd as data
2992 * is presented to the user (this happens in tcp_usrreq.c,
2993 * case PRU_RCVD). If a FIN has already been received on this
2994 * connection then we just ignore the text.
2996 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
2997 IS_FASTOPEN(tp->t_flags));
2998 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
2999 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3000 tcp_seq save_start = th->th_seq;
3001 tcp_seq save_rnxt = tp->rcv_nxt;
3002 int save_tlen = tlen;
3003 m_adj(m, drop_hdrlen); /* delayed header drop */
3005 * Insert segment which includes th into TCP reassembly queue
3006 * with control block tp. Set thflags to whether reassembly now
3007 * includes a segment with FIN. This handles the common case
3008 * inline (segment is the next to be received on an established
3009 * connection, and the queue is empty), avoiding linkage into
3010 * and removal from the queue and repetition of various
3012 * Set DELACK for segments received in order, but ack
3013 * immediately when segments are out of order (so
3014 * fast retransmit can work).
3016 if (th->th_seq == tp->rcv_nxt &&
3018 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3020 if (DELAY_ACK(tp, tlen) || tfo_syn)
3021 tp->t_flags |= TF_DELACK;
3023 tp->t_flags |= TF_ACKNOW;
3024 tp->rcv_nxt += tlen;
3026 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3027 (tp->t_fbyte_in == 0)) {
3028 tp->t_fbyte_in = ticks;
3029 if (tp->t_fbyte_in == 0)
3031 if (tp->t_fbyte_out && tp->t_fbyte_in)
3032 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3034 thflags = th->th_flags & TH_FIN;
3035 TCPSTAT_INC(tcps_rcvpack);
3036 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3037 SOCKBUF_LOCK(&so->so_rcv);
3038 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3041 sbappendstream_locked(&so->so_rcv, m, 0);
3042 /* NB: sorwakeup_locked() does an implicit unlock. */
3043 sorwakeup_locked(so);
3046 * XXX: Due to the header drop above "th" is
3047 * theoretically invalid by now. Fortunately
3048 * m_adj() doesn't actually frees any mbufs
3049 * when trimming from the head.
3051 tcp_seq temp = save_start;
3052 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3053 tp->t_flags |= TF_ACKNOW;
3055 if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) {
3056 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3058 * DSACK actually handled in the fastpath
3061 tcp_update_sack_list(tp, save_start,
3062 save_start + save_tlen);
3063 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3064 if ((tp->rcv_numsacks >= 1) &&
3065 (tp->sackblks[0].end == save_start)) {
3067 * Partial overlap, recorded at todrop
3070 tcp_update_sack_list(tp,
3071 tp->sackblks[0].start,
3072 tp->sackblks[0].end);
3074 tcp_update_dsack_list(tp, save_start,
3075 save_start + save_tlen);
3077 } else if (tlen >= save_tlen) {
3078 /* Update of sackblks. */
3079 tcp_update_dsack_list(tp, save_start,
3080 save_start + save_tlen);
3081 } else if (tlen > 0) {
3082 tcp_update_dsack_list(tp, save_start,
3088 * Note the amount of data that peer has sent into
3089 * our window, in order to estimate the sender's
3093 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3094 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3096 len = so->so_rcv.sb_hiwat;
3104 * If FIN is received ACK the FIN and let the user know
3105 * that the connection is closing.
3107 if (thflags & TH_FIN) {
3108 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3111 * If connection is half-synchronized
3112 * (ie NEEDSYN flag on) then delay ACK,
3113 * so it may be piggybacked when SYN is sent.
3114 * Otherwise, since we received a FIN then no
3115 * more input can be expected, send ACK now.
3117 if (tp->t_flags & TF_NEEDSYN)
3118 tp->t_flags |= TF_DELACK;
3120 tp->t_flags |= TF_ACKNOW;
3123 switch (tp->t_state) {
3125 * In SYN_RECEIVED and ESTABLISHED STATES
3126 * enter the CLOSE_WAIT state.
3128 case TCPS_SYN_RECEIVED:
3129 tp->t_starttime = ticks;
3131 case TCPS_ESTABLISHED:
3132 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3136 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3137 * enter the CLOSING state.
3139 case TCPS_FIN_WAIT_1:
3140 tcp_state_change(tp, TCPS_CLOSING);
3144 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3145 * starting the time-wait timer, turning off the other
3148 case TCPS_FIN_WAIT_2:
3154 if (so->so_options & SO_DEBUG)
3155 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3158 TCP_PROBE3(debug__input, tp, th, m);
3161 * Return any desired output.
3163 if (needoutput || (tp->t_flags & TF_ACKNOW))
3164 (void) tp->t_fb->tfb_tcp_output(tp);
3167 INP_WLOCK_ASSERT(tp->t_inpcb);
3169 if (tp->t_flags & TF_DELACK) {
3170 tp->t_flags &= ~TF_DELACK;
3171 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3173 INP_WUNLOCK(tp->t_inpcb);
3178 * Generate an ACK dropping incoming segment if it occupies
3179 * sequence space, where the ACK reflects our state.
3181 * We can now skip the test for the RST flag since all
3182 * paths to this code happen after packets containing
3183 * RST have been dropped.
3185 * In the SYN-RECEIVED state, don't send an ACK unless the
3186 * segment we received passes the SYN-RECEIVED ACK test.
3187 * If it fails send a RST. This breaks the loop in the
3188 * "LAND" DoS attack, and also prevents an ACK storm
3189 * between two listening ports that have been sent forged
3190 * SYN segments, each with the source address of the other.
3192 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3193 (SEQ_GT(tp->snd_una, th->th_ack) ||
3194 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3195 rstreason = BANDLIM_RST_OPENPORT;
3199 if (so->so_options & SO_DEBUG)
3200 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3203 TCP_PROBE3(debug__input, tp, th, m);
3204 tp->t_flags |= TF_ACKNOW;
3205 (void) tp->t_fb->tfb_tcp_output(tp);
3206 INP_WUNLOCK(tp->t_inpcb);
3212 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3213 INP_WUNLOCK(tp->t_inpcb);
3215 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3220 * Drop space held by incoming segment and return.
3223 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3224 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3227 TCP_PROBE3(debug__input, tp, th, m);
3229 INP_WUNLOCK(tp->t_inpcb);
3234 * Issue RST and make ACK acceptable to originator of segment.
3235 * The mbuf must still include the original packet header.
3239 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3240 int tlen, int rstreason)
3246 struct ip6_hdr *ip6;
3250 INP_WLOCK_ASSERT(tp->t_inpcb);
3253 /* Don't bother if destination was broadcast/multicast. */
3254 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3257 if (mtod(m, struct ip *)->ip_v == 6) {
3258 ip6 = mtod(m, struct ip6_hdr *);
3259 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3260 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3262 /* IPv6 anycast check is done at tcp6_input() */
3265 #if defined(INET) && defined(INET6)
3270 ip = mtod(m, struct ip *);
3271 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3272 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3273 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3274 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3279 /* Perform bandwidth limiting. */
3280 if (badport_bandlim(rstreason) < 0)
3283 /* tcp_respond consumes the mbuf chain. */
3284 if (th->th_flags & TH_ACK) {
3285 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3286 th->th_ack, TH_RST);
3288 if (th->th_flags & TH_SYN)
3290 if (th->th_flags & TH_FIN)
3292 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3293 (tcp_seq)0, TH_RST|TH_ACK);
3301 * Parse TCP options and place in tcpopt.
3304 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3309 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3311 if (opt == TCPOPT_EOL)
3313 if (opt == TCPOPT_NOP)
3319 if (optlen < 2 || optlen > cnt)
3324 if (optlen != TCPOLEN_MAXSEG)
3326 if (!(flags & TO_SYN))
3328 to->to_flags |= TOF_MSS;
3329 bcopy((char *)cp + 2,
3330 (char *)&to->to_mss, sizeof(to->to_mss));
3331 to->to_mss = ntohs(to->to_mss);
3334 if (optlen != TCPOLEN_WINDOW)
3336 if (!(flags & TO_SYN))
3338 to->to_flags |= TOF_SCALE;
3339 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3341 case TCPOPT_TIMESTAMP:
3342 if (optlen != TCPOLEN_TIMESTAMP)
3344 to->to_flags |= TOF_TS;
3345 bcopy((char *)cp + 2,
3346 (char *)&to->to_tsval, sizeof(to->to_tsval));
3347 to->to_tsval = ntohl(to->to_tsval);
3348 bcopy((char *)cp + 6,
3349 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3350 to->to_tsecr = ntohl(to->to_tsecr);
3352 case TCPOPT_SIGNATURE:
3354 * In order to reply to a host which has set the
3355 * TCP_SIGNATURE option in its initial SYN, we have
3356 * to record the fact that the option was observed
3357 * here for the syncache code to perform the correct
3360 if (optlen != TCPOLEN_SIGNATURE)
3362 to->to_flags |= TOF_SIGNATURE;
3363 to->to_signature = cp + 2;
3365 case TCPOPT_SACK_PERMITTED:
3366 if (optlen != TCPOLEN_SACK_PERMITTED)
3368 if (!(flags & TO_SYN))
3372 to->to_flags |= TOF_SACKPERM;
3375 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3379 to->to_flags |= TOF_SACK;
3380 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3381 to->to_sacks = cp + 2;
3382 TCPSTAT_INC(tcps_sack_rcv_blocks);
3384 case TCPOPT_FAST_OPEN:
3386 * Cookie length validation is performed by the
3387 * server side cookie checking code or the client
3388 * side cookie cache update code.
3390 if (!(flags & TO_SYN))
3392 if (!V_tcp_fastopen_client_enable &&
3393 !V_tcp_fastopen_server_enable)
3395 to->to_flags |= TOF_FASTOPEN;
3396 to->to_tfo_len = optlen - 2;
3397 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3406 * Pull out of band byte out of a segment so
3407 * it doesn't appear in the user's data queue.
3408 * It is still reflected in the segment length for
3409 * sequencing purposes.
3412 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3415 int cnt = off + th->th_urp - 1;
3418 if (m->m_len > cnt) {
3419 char *cp = mtod(m, caddr_t) + cnt;
3420 struct tcpcb *tp = sototcpcb(so);
3422 INP_WLOCK_ASSERT(tp->t_inpcb);
3425 tp->t_oobflags |= TCPOOB_HAVEDATA;
3426 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3428 if (m->m_flags & M_PKTHDR)
3437 panic("tcp_pulloutofband");
3441 * Collect new round-trip time estimate
3442 * and update averages and current timeout.
3445 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3449 INP_WLOCK_ASSERT(tp->t_inpcb);
3451 TCPSTAT_INC(tcps_rttupdated);
3454 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3455 imax(0, rtt * 1000 / hz));
3457 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3459 * srtt is stored as fixed point with 5 bits after the
3460 * binary point (i.e., scaled by 8). The following magic
3461 * is equivalent to the smoothing algorithm in rfc793 with
3462 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3463 * point). Adjust rtt to origin 0.
3465 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3466 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3468 if ((tp->t_srtt += delta) <= 0)
3472 * We accumulate a smoothed rtt variance (actually, a
3473 * smoothed mean difference), then set the retransmit
3474 * timer to smoothed rtt + 4 times the smoothed variance.
3475 * rttvar is stored as fixed point with 4 bits after the
3476 * binary point (scaled by 16). The following is
3477 * equivalent to rfc793 smoothing with an alpha of .75
3478 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3479 * rfc793's wired-in beta.
3483 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3484 if ((tp->t_rttvar += delta) <= 0)
3486 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3487 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3490 * No rtt measurement yet - use the unsmoothed rtt.
3491 * Set the variance to half the rtt (so our first
3492 * retransmit happens at 3*rtt).
3494 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3495 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3496 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3502 * the retransmit should happen at rtt + 4 * rttvar.
3503 * Because of the way we do the smoothing, srtt and rttvar
3504 * will each average +1/2 tick of bias. When we compute
3505 * the retransmit timer, we want 1/2 tick of rounding and
3506 * 1 extra tick because of +-1/2 tick uncertainty in the
3507 * firing of the timer. The bias will give us exactly the
3508 * 1.5 tick we need. But, because the bias is
3509 * statistical, we have to test that we don't drop below
3510 * the minimum feasible timer (which is 2 ticks).
3512 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3513 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3516 * We received an ack for a packet that wasn't retransmitted;
3517 * it is probably safe to discard any error indications we've
3518 * received recently. This isn't quite right, but close enough
3519 * for now (a route might have failed after we sent a segment,
3520 * and the return path might not be symmetrical).
3522 tp->t_softerror = 0;
3526 * Determine a reasonable value for maxseg size.
3527 * If the route is known, check route for mtu.
3528 * If none, use an mss that can be handled on the outgoing interface
3529 * without forcing IP to fragment. If no route is found, route has no mtu,
3530 * or the destination isn't local, use a default, hopefully conservative
3531 * size (usually 512 or the default IP max size, but no more than the mtu
3532 * of the interface), as we can't discover anything about intervening
3533 * gateways or networks. We also initialize the congestion/slow start
3534 * window to be a single segment if the destination isn't local.
3535 * While looking at the routing entry, we also initialize other path-dependent
3536 * parameters from pre-set or cached values in the routing entry.
3538 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3539 * IP options, e.g. IPSEC data, since length of this data may vary, and
3540 * thus it is calculated for every segment separately in tcp_output().
3542 * NOTE that this routine is only called when we process an incoming
3543 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3544 * settings are handled in tcp_mssopt().
3547 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3548 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3551 uint32_t maxmtu = 0;
3552 struct inpcb *inp = tp->t_inpcb;
3553 struct hc_metrics_lite metrics;
3555 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3556 size_t min_protoh = isipv6 ?
3557 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3558 sizeof (struct tcpiphdr);
3560 const size_t min_protoh = sizeof(struct tcpiphdr);
3563 INP_WLOCK_ASSERT(tp->t_inpcb);
3565 if (mtuoffer != -1) {
3566 KASSERT(offer == -1, ("%s: conflict", __func__));
3567 offer = mtuoffer - min_protoh;
3573 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3574 tp->t_maxseg = V_tcp_v6mssdflt;
3577 #if defined(INET) && defined(INET6)
3582 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3583 tp->t_maxseg = V_tcp_mssdflt;
3588 * No route to sender, stay with default mss and return.
3592 * In case we return early we need to initialize metrics
3593 * to a defined state as tcp_hc_get() would do for us
3594 * if there was no cache hit.
3596 if (metricptr != NULL)
3597 bzero(metricptr, sizeof(struct hc_metrics_lite));
3601 /* What have we got? */
3605 * Offer == 0 means that there was no MSS on the SYN
3606 * segment, in this case we use tcp_mssdflt as
3607 * already assigned to t_maxseg above.
3609 offer = tp->t_maxseg;
3614 * Offer == -1 means that we didn't receive SYN yet.
3620 * Prevent DoS attack with too small MSS. Round up
3621 * to at least minmss.
3623 offer = max(offer, V_tcp_minmss);
3627 * rmx information is now retrieved from tcp_hostcache.
3629 tcp_hc_get(&inp->inp_inc, &metrics);
3630 if (metricptr != NULL)
3631 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3634 * If there's a discovered mtu in tcp hostcache, use it.
3635 * Else, use the link mtu.
3637 if (metrics.rmx_mtu)
3638 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3642 mss = maxmtu - min_protoh;
3643 if (!V_path_mtu_discovery &&
3644 !in6_localaddr(&inp->in6p_faddr))
3645 mss = min(mss, V_tcp_v6mssdflt);
3648 #if defined(INET) && defined(INET6)
3653 mss = maxmtu - min_protoh;
3654 if (!V_path_mtu_discovery &&
3655 !in_localaddr(inp->inp_faddr))
3656 mss = min(mss, V_tcp_mssdflt);
3660 * XXX - The above conditional (mss = maxmtu - min_protoh)
3661 * probably violates the TCP spec.
3662 * The problem is that, since we don't know the
3663 * other end's MSS, we are supposed to use a conservative
3664 * default. But, if we do that, then MTU discovery will
3665 * never actually take place, because the conservative
3666 * default is much less than the MTUs typically seen
3667 * on the Internet today. For the moment, we'll sweep
3668 * this under the carpet.
3670 * The conservative default might not actually be a problem
3671 * if the only case this occurs is when sending an initial
3672 * SYN with options and data to a host we've never talked
3673 * to before. Then, they will reply with an MSS value which
3674 * will get recorded and the new parameters should get
3675 * recomputed. For Further Study.
3678 mss = min(mss, offer);
3681 * Sanity check: make sure that maxseg will be large
3682 * enough to allow some data on segments even if the
3683 * all the option space is used (40bytes). Otherwise
3684 * funny things may happen in tcp_output.
3686 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3694 tcp_mss(struct tcpcb *tp, int offer)
3700 struct hc_metrics_lite metrics;
3701 struct tcp_ifcap cap;
3703 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3705 bzero(&cap, sizeof(cap));
3706 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3712 * If there's a pipesize, change the socket buffer to that size,
3713 * don't change if sb_hiwat is different than default (then it
3714 * has been changed on purpose with setsockopt).
3715 * Make the socket buffers an integral number of mss units;
3716 * if the mss is larger than the socket buffer, decrease the mss.
3718 so = inp->inp_socket;
3719 SOCKBUF_LOCK(&so->so_snd);
3720 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3721 bufsize = metrics.rmx_sendpipe;
3723 bufsize = so->so_snd.sb_hiwat;
3727 bufsize = roundup(bufsize, mss);
3728 if (bufsize > sb_max)
3730 if (bufsize > so->so_snd.sb_hiwat)
3731 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3733 SOCKBUF_UNLOCK(&so->so_snd);
3735 * Sanity check: make sure that maxseg will be large
3736 * enough to allow some data on segments even if the
3737 * all the option space is used (40bytes). Otherwise
3738 * funny things may happen in tcp_output.
3740 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3742 tp->t_maxseg = max(mss, 64);
3744 SOCKBUF_LOCK(&so->so_rcv);
3745 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3746 bufsize = metrics.rmx_recvpipe;
3748 bufsize = so->so_rcv.sb_hiwat;
3749 if (bufsize > mss) {
3750 bufsize = roundup(bufsize, mss);
3751 if (bufsize > sb_max)
3753 if (bufsize > so->so_rcv.sb_hiwat)
3754 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3756 SOCKBUF_UNLOCK(&so->so_rcv);
3758 /* Check the interface for TSO capabilities. */
3759 if (cap.ifcap & CSUM_TSO) {
3760 tp->t_flags |= TF_TSO;
3761 tp->t_tsomax = cap.tsomax;
3762 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3763 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3768 * Determine the MSS option to send on an outgoing SYN.
3771 tcp_mssopt(struct in_conninfo *inc)
3774 uint32_t thcmtu = 0;
3775 uint32_t maxmtu = 0;
3778 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3781 if (inc->inc_flags & INC_ISIPV6) {
3782 mss = V_tcp_v6mssdflt;
3783 maxmtu = tcp_maxmtu6(inc, NULL);
3784 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3787 #if defined(INET) && defined(INET6)
3792 mss = V_tcp_mssdflt;
3793 maxmtu = tcp_maxmtu(inc, NULL);
3794 min_protoh = sizeof(struct tcpiphdr);
3797 #if defined(INET6) || defined(INET)
3798 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3801 if (maxmtu && thcmtu)
3802 mss = min(maxmtu, thcmtu) - min_protoh;
3803 else if (maxmtu || thcmtu)
3804 mss = max(maxmtu, thcmtu) - min_protoh;
3810 * On a partial ack arrives, force the retransmission of the
3811 * next unacknowledged segment. Do not clear tp->t_dupacks.
3812 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3816 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3818 tcp_seq onxt = tp->snd_nxt;
3819 uint32_t ocwnd = tp->snd_cwnd;
3820 u_int maxseg = tcp_maxseg(tp);
3822 INP_WLOCK_ASSERT(tp->t_inpcb);
3824 tcp_timer_activate(tp, TT_REXMT, 0);
3826 tp->snd_nxt = th->th_ack;
3828 * Set snd_cwnd to one segment beyond acknowledged offset.
3829 * (tp->snd_una has not yet been updated when this function is called.)
3831 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3832 tp->t_flags |= TF_ACKNOW;
3833 (void) tp->t_fb->tfb_tcp_output(tp);
3834 tp->snd_cwnd = ocwnd;
3835 if (SEQ_GT(onxt, tp->snd_nxt))
3838 * Partial window deflation. Relies on fact that tp->snd_una
3841 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3842 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3845 tp->snd_cwnd += maxseg;
3849 tcp_compute_pipe(struct tcpcb *tp)
3851 return (tp->snd_max - tp->snd_una +
3852 tp->sackhint.sack_bytes_rexmit -
3853 tp->sackhint.sacked_bytes);
3857 tcp_compute_initwnd(uint32_t maxseg)
3860 * Calculate the Initial Window, also used as Restart Window
3862 * RFC5681 Section 3.1 specifies the default conservative values.
3863 * RFC3390 specifies slightly more aggressive values.
3864 * RFC6928 increases it to ten segments.
3865 * Support for user specified value for initial flight size.
3867 if (V_tcp_initcwnd_segments)
3868 return min(V_tcp_initcwnd_segments * maxseg,
3869 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
3870 else if (V_tcp_do_rfc3390)
3871 return min(4 * maxseg, max(2 * maxseg, 4380));
3873 /* Per RFC5681 Section 3.1 */
3875 return (2 * maxseg);
3876 else if (maxseg > 1095)
3877 return (3 * maxseg);
3879 return (4 * maxseg);