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_prr_conservative) = 0;
157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW,
158 &VNET_NAME(tcp_do_prr_conservative), 0,
159 "Do conservative Proportional Rate Reduction");
161 VNET_DEFINE(int, tcp_do_prr) = 1;
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW,
163 &VNET_NAME(tcp_do_prr), 1,
164 "Enable Proportional Rate Reduction per RFC 6937");
166 VNET_DEFINE(int, tcp_do_newcwv) = 0;
167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
168 &VNET_NAME(tcp_do_newcwv), 0,
169 "Enable New Congestion Window Validation per RFC7661");
171 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
173 &VNET_NAME(tcp_do_rfc6675_pipe), 0,
174 "Use calculated pipe/in-flight bytes per RFC 6675");
176 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
178 &VNET_NAME(tcp_do_rfc3042), 0,
179 "Enable RFC 3042 (Limited Transmit)");
181 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
183 &VNET_NAME(tcp_do_rfc3390), 0,
184 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
186 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
188 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
189 "Slow-start flight size (initial congestion window) in number of segments");
191 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
193 &VNET_NAME(tcp_do_rfc3465), 0,
194 "Enable RFC 3465 (Appropriate Byte Counting)");
196 VNET_DEFINE(int, tcp_abc_l_var) = 2;
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
198 &VNET_NAME(tcp_abc_l_var), 2,
199 "Cap the max cwnd increment during slow-start to this number of segments");
201 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn,
202 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
205 VNET_DEFINE(int, tcp_do_ecn) = 2;
206 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
207 &VNET_NAME(tcp_do_ecn), 0,
210 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
211 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
212 &VNET_NAME(tcp_ecn_maxretries), 0,
213 "Max retries before giving up on ECN");
215 VNET_DEFINE(int, tcp_insecure_syn) = 0;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
217 &VNET_NAME(tcp_insecure_syn), 0,
218 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
220 VNET_DEFINE(int, tcp_insecure_rst) = 0;
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
222 &VNET_NAME(tcp_insecure_rst), 0,
223 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
225 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
226 #define V_tcp_recvspace VNET(tcp_recvspace)
227 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
228 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
230 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
232 &VNET_NAME(tcp_do_autorcvbuf), 0,
233 "Enable automatic receive buffer sizing");
235 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
236 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
237 &VNET_NAME(tcp_autorcvbuf_max), 0,
238 "Max size of automatic receive buffer");
240 VNET_DEFINE(struct inpcbhead, tcb);
241 #define tcb6 tcb /* for KAME src sync over BSD*'s */
242 VNET_DEFINE(struct inpcbinfo, tcbinfo);
245 * TCP statistics are stored in an array of counter(9)s, which size matches
246 * size of struct tcpstat. TCP running connection count is a regular array.
248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
249 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
250 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
251 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
252 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
253 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
254 "TCP connection counts by TCP state");
257 tcp_vnet_init(const void *unused)
260 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
261 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
263 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
264 tcp_vnet_init, NULL);
268 tcp_vnet_uninit(const void *unused)
271 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
272 VNET_PCPUSTAT_FREE(tcpstat);
274 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
275 tcp_vnet_uninit, NULL);
279 * Kernel module interface for updating tcpstat. The first argument is an index
280 * into tcpstat treated as an array.
283 kmod_tcpstat_add(int statnum, int val)
286 counter_u64_add(VNET(tcpstat)[statnum], val);
291 * Wrapper for the TCP established input helper hook.
294 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
296 struct tcp_hhook_data hhook_data;
298 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
303 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
310 * CC wrapper hook functions
313 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
320 INP_WLOCK_ASSERT(tp->t_inpcb);
322 tp->ccv->nsegs = nsegs;
323 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
324 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
325 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
326 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
327 tp->ccv->flags |= CCF_CWND_LIMITED;
329 tp->ccv->flags &= ~CCF_CWND_LIMITED;
331 if (type == CC_ACK) {
333 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
334 ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
335 if (!IN_RECOVERY(tp->t_flags))
336 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
337 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
338 if ((tp->t_flags & TF_GPUTINPROG) &&
339 SEQ_GEQ(th->th_ack, tp->gput_ack)) {
341 * Compute goodput in bits per millisecond.
343 gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) /
344 max(1, tcp_ts_getticks() - tp->gput_ts);
345 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
348 * XXXLAS: This is a temporary hack, and should be
349 * chained off VOI_TCP_GPUT when stats(9) grows an API
350 * to deal with chained VOIs.
352 if (tp->t_stats_gput_prev > 0)
353 stats_voi_update_abs_s32(tp->t_stats,
355 ((gput - tp->t_stats_gput_prev) * 100) /
356 tp->t_stats_gput_prev);
357 tp->t_flags &= ~TF_GPUTINPROG;
358 tp->t_stats_gput_prev = gput;
361 if (tp->snd_cwnd > tp->snd_ssthresh) {
362 tp->t_bytes_acked += tp->ccv->bytes_this_ack;
363 if (tp->t_bytes_acked >= tp->snd_cwnd) {
364 tp->t_bytes_acked -= tp->snd_cwnd;
365 tp->ccv->flags |= CCF_ABC_SENTAWND;
368 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
369 tp->t_bytes_acked = 0;
373 if (CC_ALGO(tp)->ack_received != NULL) {
374 /* XXXLAS: Find a way to live without this */
375 tp->ccv->curack = th->th_ack;
376 CC_ALGO(tp)->ack_received(tp->ccv, type);
379 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
384 cc_conn_init(struct tcpcb *tp)
386 struct hc_metrics_lite metrics;
387 struct inpcb *inp = tp->t_inpcb;
391 INP_WLOCK_ASSERT(tp->t_inpcb);
393 tcp_hc_get(&inp->inp_inc, &metrics);
394 maxseg = tcp_maxseg(tp);
396 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
398 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
399 TCPSTAT_INC(tcps_usedrtt);
400 if (metrics.rmx_rttvar) {
401 tp->t_rttvar = metrics.rmx_rttvar;
402 TCPSTAT_INC(tcps_usedrttvar);
404 /* default variation is +- 1 rtt */
406 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
408 TCPT_RANGESET(tp->t_rxtcur,
409 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
410 tp->t_rttmin, TCPTV_REXMTMAX);
412 if (metrics.rmx_ssthresh) {
414 * There's some sort of gateway or interface
415 * buffer limit on the path. Use this to set
416 * the slow start threshold, but set the
417 * threshold to no less than 2*mss.
419 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
420 TCPSTAT_INC(tcps_usedssthresh);
424 * Set the initial slow-start flight size.
426 * If a SYN or SYN/ACK was lost and retransmitted, we have to
427 * reduce the initial CWND to one segment as congestion is likely
428 * requiring us to be cautious.
430 if (tp->snd_cwnd == 1)
431 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
433 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
435 if (CC_ALGO(tp)->conn_init != NULL)
436 CC_ALGO(tp)->conn_init(tp->ccv);
440 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
442 INP_WLOCK_ASSERT(tp->t_inpcb);
445 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
450 if (!IN_FASTRECOVERY(tp->t_flags)) {
451 tp->snd_recover = tp->snd_max;
452 if (tp->t_flags2 & TF2_ECN_PERMIT)
453 tp->t_flags2 |= TF2_ECN_SND_CWR;
457 if (!IN_CONGRECOVERY(tp->t_flags) ||
459 * Allow ECN reaction on ACK to CWR, if
460 * that data segment was also CE marked.
462 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
463 EXIT_CONGRECOVERY(tp->t_flags);
464 TCPSTAT_INC(tcps_ecn_rcwnd);
465 tp->snd_recover = tp->snd_max + 1;
466 if (tp->t_flags2 & TF2_ECN_PERMIT)
467 tp->t_flags2 |= TF2_ECN_SND_CWR;
472 tp->t_bytes_acked = 0;
473 EXIT_RECOVERY(tp->t_flags);
474 if (tp->t_flags2 & TF2_ECN_PERMIT)
475 tp->t_flags2 |= TF2_ECN_SND_CWR;
478 TCPSTAT_INC(tcps_sndrexmitbad);
479 /* RTO was unnecessary, so reset everything. */
480 tp->snd_cwnd = tp->snd_cwnd_prev;
481 tp->snd_ssthresh = tp->snd_ssthresh_prev;
482 tp->snd_recover = tp->snd_recover_prev;
483 if (tp->t_flags & TF_WASFRECOVERY)
484 ENTER_FASTRECOVERY(tp->t_flags);
485 if (tp->t_flags & TF_WASCRECOVERY)
486 ENTER_CONGRECOVERY(tp->t_flags);
487 tp->snd_nxt = tp->snd_max;
488 tp->t_flags &= ~TF_PREVVALID;
493 if (CC_ALGO(tp)->cong_signal != NULL) {
495 tp->ccv->curack = th->th_ack;
496 CC_ALGO(tp)->cong_signal(tp->ccv, type);
501 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
503 INP_WLOCK_ASSERT(tp->t_inpcb);
505 /* XXXLAS: KASSERT that we're in recovery? */
507 if (CC_ALGO(tp)->post_recovery != NULL) {
508 tp->ccv->curack = th->th_ack;
509 CC_ALGO(tp)->post_recovery(tp->ccv);
511 /* XXXLAS: EXIT_RECOVERY ? */
512 tp->t_bytes_acked = 0;
513 tp->sackhint.recover_fs = 0;
517 * Indicate whether this ack should be delayed. We can delay the ack if
518 * following conditions are met:
519 * - There is no delayed ack timer in progress.
520 * - Our last ack wasn't a 0-sized window. We never want to delay
521 * the ack that opens up a 0-sized window.
522 * - LRO wasn't used for this segment. We make sure by checking that the
523 * segment size is not larger than the MSS.
525 #define DELAY_ACK(tp, tlen) \
526 ((!tcp_timer_active(tp, TT_DELACK) && \
527 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
528 (tlen <= tp->t_maxseg) && \
529 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
532 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
534 INP_WLOCK_ASSERT(tp->t_inpcb);
536 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
537 switch (iptos & IPTOS_ECN_MASK) {
539 tp->ccv->flags |= CCF_IPHDR_CE;
545 case IPTOS_ECN_NOTECT:
546 tp->ccv->flags &= ~CCF_IPHDR_CE;
550 if (th->th_flags & TH_CWR)
551 tp->ccv->flags |= CCF_TCPHDR_CWR;
553 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
555 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
557 if (tp->ccv->flags & CCF_ACKNOW) {
558 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
559 tp->t_flags |= TF_ACKNOW;
565 * TCP input handling is split into multiple parts:
566 * tcp6_input is a thin wrapper around tcp_input for the extended
567 * ip6_protox[] call format in ip6_input
568 * tcp_input handles primary segment validation, inpcb lookup and
569 * SYN processing on listen sockets
570 * tcp_do_segment processes the ACK and text of the segment for
571 * establishing, established and closing connections
575 tcp6_input(struct mbuf **mp, int *offp, int proto)
578 struct in6_ifaddr *ia6;
582 if (m->m_len < *offp + sizeof(struct tcphdr)) {
583 m = m_pullup(m, *offp + sizeof(struct tcphdr));
586 TCPSTAT_INC(tcps_rcvshort);
587 return (IPPROTO_DONE);
592 * draft-itojun-ipv6-tcp-to-anycast
593 * better place to put this in?
595 ip6 = mtod(m, struct ip6_hdr *);
596 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
597 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
598 ifa_free(&ia6->ia_ifa);
599 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
600 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
602 return (IPPROTO_DONE);
605 ifa_free(&ia6->ia_ifa);
608 return (tcp_input(mp, offp, proto));
613 tcp_input(struct mbuf **mp, int *offp, int proto)
615 struct mbuf *m = *mp;
616 struct tcphdr *th = NULL;
617 struct ip *ip = NULL;
618 struct inpcb *inp = NULL;
619 struct tcpcb *tp = NULL;
620 struct socket *so = NULL;
631 int rstreason = 0; /* For badport_bandlim accounting purposes */
633 struct m_tag *fwd_tag = NULL;
635 struct ip6_hdr *ip6 = NULL;
638 const void *ip6 = NULL;
640 struct tcpopt to; /* options in this segment */
641 char *s = NULL; /* address and port logging */
644 * The size of tcp_saveipgen must be the size of the max ip header,
647 u_char tcp_saveipgen[IP6_HDR_LEN];
648 struct tcphdr tcp_savetcp;
655 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
662 TCPSTAT_INC(tcps_rcvtotal);
666 ip6 = mtod(m, struct ip6_hdr *);
667 th = (struct tcphdr *)((caddr_t)ip6 + off0);
668 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
669 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
670 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
671 th->th_sum = m->m_pkthdr.csum_data;
673 th->th_sum = in6_cksum_pseudo(ip6, tlen,
674 IPPROTO_TCP, m->m_pkthdr.csum_data);
675 th->th_sum ^= 0xffff;
677 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
679 TCPSTAT_INC(tcps_rcvbadsum);
684 * Be proactive about unspecified IPv6 address in source.
685 * As we use all-zero to indicate unbounded/unconnected pcb,
686 * unspecified IPv6 address can be used to confuse us.
688 * Note that packets with unspecified IPv6 destination is
689 * already dropped in ip6_input.
691 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
695 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
698 #if defined(INET) && defined(INET6)
704 * Get IP and TCP header together in first mbuf.
705 * Note: IP leaves IP header in first mbuf.
707 if (off0 > sizeof (struct ip)) {
709 off0 = sizeof(struct ip);
711 if (m->m_len < sizeof (struct tcpiphdr)) {
712 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
714 TCPSTAT_INC(tcps_rcvshort);
715 return (IPPROTO_DONE);
718 ip = mtod(m, struct ip *);
719 th = (struct tcphdr *)((caddr_t)ip + off0);
720 tlen = ntohs(ip->ip_len) - off0;
723 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
724 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
725 th->th_sum = m->m_pkthdr.csum_data;
727 th->th_sum = in_pseudo(ip->ip_src.s_addr,
729 htonl(m->m_pkthdr.csum_data + tlen +
731 th->th_sum ^= 0xffff;
733 struct ipovly *ipov = (struct ipovly *)ip;
736 * Checksum extended TCP header and data.
740 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
741 ipov->ih_len = htons(tlen);
742 th->th_sum = in_cksum(m, len);
743 /* Reset length for SDT probes. */
744 ip->ip_len = htons(len);
747 /* Re-initialization for later version check */
749 ip->ip_v = IPVERSION;
750 ip->ip_hl = off0 >> 2;
754 TCPSTAT_INC(tcps_rcvbadsum);
761 * Check that TCP offset makes sense,
762 * pull out TCP options and adjust length. XXX
764 off = th->th_off << 2;
765 if (off < sizeof (struct tcphdr) || off > tlen) {
766 TCPSTAT_INC(tcps_rcvbadoff);
769 tlen -= off; /* tlen is used instead of ti->ti_len */
770 if (off > sizeof (struct tcphdr)) {
773 if (m->m_len < off0 + off) {
774 m = m_pullup(m, off0 + off);
776 TCPSTAT_INC(tcps_rcvshort);
777 return (IPPROTO_DONE);
780 ip6 = mtod(m, struct ip6_hdr *);
781 th = (struct tcphdr *)((caddr_t)ip6 + off0);
784 #if defined(INET) && defined(INET6)
789 if (m->m_len < sizeof(struct ip) + off) {
790 if ((m = m_pullup(m, sizeof (struct ip) + off))
792 TCPSTAT_INC(tcps_rcvshort);
793 return (IPPROTO_DONE);
795 ip = mtod(m, struct ip *);
796 th = (struct tcphdr *)((caddr_t)ip + off0);
800 optlen = off - sizeof (struct tcphdr);
801 optp = (u_char *)(th + 1);
803 thflags = th->th_flags;
806 * Convert TCP protocol specific fields to host format.
808 tcp_fields_to_host(th);
811 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
813 drop_hdrlen = off0 + off;
816 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
820 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
822 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
825 #if defined(INET) && !defined(INET6)
826 (m->m_flags & M_IP_NEXTHOP)
829 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
833 if (isipv6 && fwd_tag != NULL) {
834 struct sockaddr_in6 *next_hop6;
836 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
838 * Transparently forwarded. Pretend to be the destination.
839 * Already got one like this?
841 inp = in6_pcblookup_mbuf(&V_tcbinfo,
842 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
843 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
846 * It's new. Try to find the ambushing socket.
847 * Because we've rewritten the destination address,
848 * any hardware-generated hash is ignored.
850 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
851 th->th_sport, &next_hop6->sin6_addr,
852 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
853 th->th_dport, INPLOOKUP_WILDCARD |
854 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
857 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
858 th->th_sport, &ip6->ip6_dst, th->th_dport,
859 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
860 m->m_pkthdr.rcvif, m);
863 #if defined(INET6) && defined(INET)
867 if (fwd_tag != NULL) {
868 struct sockaddr_in *next_hop;
870 next_hop = (struct sockaddr_in *)(fwd_tag+1);
872 * Transparently forwarded. Pretend to be the destination.
873 * already got one like this?
875 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
876 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
877 m->m_pkthdr.rcvif, m);
880 * It's new. Try to find the ambushing socket.
881 * Because we've rewritten the destination address,
882 * any hardware-generated hash is ignored.
884 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
885 th->th_sport, next_hop->sin_addr,
886 next_hop->sin_port ? ntohs(next_hop->sin_port) :
887 th->th_dport, INPLOOKUP_WILDCARD |
888 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
891 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
892 th->th_sport, ip->ip_dst, th->th_dport,
893 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
894 m->m_pkthdr.rcvif, m);
898 * If the INPCB does not exist then all data in the incoming
899 * segment is discarded and an appropriate RST is sent back.
900 * XXX MRT Send RST using which routing table?
904 * Log communication attempts to ports that are not
907 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
908 V_tcp_log_in_vain == 2) {
909 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
910 log(LOG_INFO, "%s; %s: Connection attempt "
911 "to closed port\n", s, __func__);
914 * When blackholing do not respond with a RST but
915 * completely ignore the segment and drop it.
917 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
921 rstreason = BANDLIM_RST_CLOSEDPORT;
924 INP_WLOCK_ASSERT(inp);
926 * While waiting for inp lock during the lookup, another thread
927 * can have dropped the inpcb, in which case we need to loop back
928 * and try to find a new inpcb to deliver to.
930 if (inp->inp_flags & INP_DROPPED) {
935 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
936 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
937 ((inp->inp_socket == NULL) ||
938 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
939 inp->inp_flowid = m->m_pkthdr.flowid;
940 inp->inp_flowtype = M_HASHTYPE_GET(m);
942 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
944 if (isipv6 && IPSEC_ENABLED(ipv6) &&
945 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
953 if (IPSEC_ENABLED(ipv4) &&
954 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
961 * Check the minimum TTL for socket.
963 if (inp->inp_ip_minttl != 0) {
966 if (inp->inp_ip_minttl > ip6->ip6_hlim)
970 if (inp->inp_ip_minttl > ip->ip_ttl)
975 * A previous connection in TIMEWAIT state is supposed to catch stray
976 * or duplicate segments arriving late. If this segment was a
977 * legitimate new connection attempt, the old INPCB gets removed and
978 * we can try again to find a listening socket.
980 * At this point, due to earlier optimism, we may hold only an inpcb
981 * lock, and not the inpcbinfo write lock. If so, we need to try to
982 * acquire it, or if that fails, acquire a reference on the inpcb,
983 * drop all locks, acquire a global write lock, and then re-acquire
984 * the inpcb lock. We may at that point discover that another thread
985 * has tried to free the inpcb, in which case we need to loop back
986 * and try to find a new inpcb to deliver to.
988 * XXXRW: It may be time to rethink timewait locking.
990 if (inp->inp_flags & INP_TIMEWAIT) {
991 tcp_dooptions(&to, optp, optlen,
992 (thflags & TH_SYN) ? TO_SYN : 0);
994 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
996 if (tcp_twcheck(inp, &to, th, m, tlen))
998 return (IPPROTO_DONE);
1001 * The TCPCB may no longer exist if the connection is winding
1002 * down or it is in the CLOSED state. Either way we drop the
1003 * segment and send an appropriate response.
1005 tp = intotcpcb(inp);
1006 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1007 rstreason = BANDLIM_RST_CLOSEDPORT;
1012 if (tp->t_flags & TF_TOE) {
1013 tcp_offload_input(tp, m);
1014 m = NULL; /* consumed by the TOE driver */
1020 INP_WLOCK_ASSERT(inp);
1021 if (mac_inpcb_check_deliver(inp, m))
1024 so = inp->inp_socket;
1025 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1027 if (so->so_options & SO_DEBUG) {
1028 ostate = tp->t_state;
1031 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1034 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1037 #endif /* TCPDEBUG */
1039 * When the socket is accepting connections (the INPCB is in LISTEN
1040 * state) we look into the SYN cache if this is a new connection
1041 * attempt or the completion of a previous one.
1043 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1044 ("%s: so accepting but tp %p not listening", __func__, tp));
1045 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1046 struct in_conninfo inc;
1048 bzero(&inc, sizeof(inc));
1051 inc.inc_flags |= INC_ISIPV6;
1052 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1053 inc.inc_flags |= INC_IPV6MINMTU;
1054 inc.inc6_faddr = ip6->ip6_src;
1055 inc.inc6_laddr = ip6->ip6_dst;
1059 inc.inc_faddr = ip->ip_src;
1060 inc.inc_laddr = ip->ip_dst;
1062 inc.inc_fport = th->th_sport;
1063 inc.inc_lport = th->th_dport;
1064 inc.inc_fibnum = so->so_fibnum;
1067 * Check for an existing connection attempt in syncache if
1068 * the flag is only ACK. A successful lookup creates a new
1069 * socket appended to the listen queue in SYN_RECEIVED state.
1071 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1073 * Parse the TCP options here because
1074 * syncookies need access to the reflected
1077 tcp_dooptions(&to, optp, optlen, 0);
1079 * NB: syncache_expand() doesn't unlock
1080 * inp and tcpinfo locks.
1082 rstreason = syncache_expand(&inc, &to, th, &so, m);
1083 if (rstreason < 0) {
1085 * A failing TCP MD5 signature comparison
1086 * must result in the segment being dropped
1087 * and must not produce any response back
1091 } else if (rstreason == 0) {
1093 * No syncache entry or ACK was not
1094 * for our SYN/ACK. Send a RST.
1095 * NB: syncache did its own logging
1096 * of the failure cause.
1098 rstreason = BANDLIM_RST_OPENPORT;
1104 * We completed the 3-way handshake
1105 * but could not allocate a socket
1106 * either due to memory shortage,
1107 * listen queue length limits or
1108 * global socket limits. Send RST
1109 * or wait and have the remote end
1110 * retransmit the ACK for another
1113 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1114 log(LOG_DEBUG, "%s; %s: Listen socket: "
1115 "Socket allocation failed due to "
1116 "limits or memory shortage, %s\n",
1118 V_tcp_sc_rst_sock_fail ?
1119 "sending RST" : "try again");
1120 if (V_tcp_sc_rst_sock_fail) {
1121 rstreason = BANDLIM_UNLIMITED;
1127 * Socket is created in state SYN_RECEIVED.
1128 * Unlock the listen socket, lock the newly
1129 * created socket and update the tp variable.
1131 INP_WUNLOCK(inp); /* listen socket */
1132 inp = sotoinpcb(so);
1134 * New connection inpcb is already locked by
1135 * syncache_expand().
1137 INP_WLOCK_ASSERT(inp);
1138 tp = intotcpcb(inp);
1139 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1140 ("%s: ", __func__));
1142 * Process the segment and the data it
1143 * contains. tcp_do_segment() consumes
1144 * the mbuf chain and unlocks the inpcb.
1146 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1147 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1149 return (IPPROTO_DONE);
1152 * Segment flag validation for new connection attempts:
1154 * Our (SYN|ACK) response was rejected.
1155 * Check with syncache and remove entry to prevent
1158 * NB: syncache_chkrst does its own logging of failure
1161 if (thflags & TH_RST) {
1162 syncache_chkrst(&inc, th, m);
1166 * We can't do anything without SYN.
1168 if ((thflags & TH_SYN) == 0) {
1169 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1170 log(LOG_DEBUG, "%s; %s: Listen socket: "
1171 "SYN is missing, segment ignored\n",
1173 TCPSTAT_INC(tcps_badsyn);
1177 * (SYN|ACK) is bogus on a listen socket.
1179 if (thflags & TH_ACK) {
1180 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1181 log(LOG_DEBUG, "%s; %s: Listen socket: "
1182 "SYN|ACK invalid, segment rejected\n",
1184 syncache_badack(&inc); /* XXX: Not needed! */
1185 TCPSTAT_INC(tcps_badsyn);
1186 rstreason = BANDLIM_RST_OPENPORT;
1190 * If the drop_synfin option is enabled, drop all
1191 * segments with both the SYN and FIN bits set.
1192 * This prevents e.g. nmap from identifying the
1194 * XXX: Poor reasoning. nmap has other methods
1195 * and is constantly refining its stack detection
1197 * XXX: This is a violation of the TCP specification
1198 * and was used by RFC1644.
1200 if ((thflags & TH_FIN) && V_drop_synfin) {
1201 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1202 log(LOG_DEBUG, "%s; %s: Listen socket: "
1203 "SYN|FIN segment ignored (based on "
1204 "sysctl setting)\n", s, __func__);
1205 TCPSTAT_INC(tcps_badsyn);
1209 * Segment's flags are (SYN) or (SYN|FIN).
1211 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1212 * as they do not affect the state of the TCP FSM.
1213 * The data pointed to by TH_URG and th_urp is ignored.
1215 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1216 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1217 KASSERT(thflags & (TH_SYN),
1218 ("%s: Listen socket: TH_SYN not set", __func__));
1221 * If deprecated address is forbidden,
1222 * we do not accept SYN to deprecated interface
1223 * address to prevent any new inbound connection from
1224 * getting established.
1225 * When we do not accept SYN, we send a TCP RST,
1226 * with deprecated source address (instead of dropping
1227 * it). We compromise it as it is much better for peer
1228 * to send a RST, and RST will be the final packet
1231 * If we do not forbid deprecated addresses, we accept
1232 * the SYN packet. RFC2462 does not suggest dropping
1234 * If we decipher RFC2462 5.5.4, it says like this:
1235 * 1. use of deprecated addr with existing
1236 * communication is okay - "SHOULD continue to be
1238 * 2. use of it with new communication:
1239 * (2a) "SHOULD NOT be used if alternate address
1240 * with sufficient scope is available"
1241 * (2b) nothing mentioned otherwise.
1242 * Here we fall into (2b) case as we have no choice in
1243 * our source address selection - we must obey the peer.
1245 * The wording in RFC2462 is confusing, and there are
1246 * multiple description text for deprecated address
1247 * handling - worse, they are not exactly the same.
1248 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1250 if (isipv6 && !V_ip6_use_deprecated) {
1251 struct in6_ifaddr *ia6;
1253 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1255 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1256 ifa_free(&ia6->ia_ifa);
1257 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1258 log(LOG_DEBUG, "%s; %s: Listen socket: "
1259 "Connection attempt to deprecated "
1260 "IPv6 address rejected\n",
1262 rstreason = BANDLIM_RST_OPENPORT;
1266 ifa_free(&ia6->ia_ifa);
1270 * Basic sanity checks on incoming SYN requests:
1271 * Don't respond if the destination is a link layer
1272 * broadcast according to RFC1122 4.2.3.10, p. 104.
1273 * If it is from this socket it must be forged.
1274 * Don't respond if the source or destination is a
1275 * global or subnet broad- or multicast address.
1276 * Note that it is quite possible to receive unicast
1277 * link-layer packets with a broadcast IP address. Use
1278 * in_broadcast() to find them.
1280 if (m->m_flags & (M_BCAST|M_MCAST)) {
1281 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1282 log(LOG_DEBUG, "%s; %s: Listen socket: "
1283 "Connection attempt from broad- or multicast "
1284 "link layer address ignored\n", s, __func__);
1289 if (th->th_dport == th->th_sport &&
1290 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1291 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1292 log(LOG_DEBUG, "%s; %s: Listen socket: "
1293 "Connection attempt to/from self "
1294 "ignored\n", s, __func__);
1297 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1298 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1299 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1300 log(LOG_DEBUG, "%s; %s: Listen socket: "
1301 "Connection attempt from/to multicast "
1302 "address ignored\n", s, __func__);
1307 #if defined(INET) && defined(INET6)
1312 if (th->th_dport == th->th_sport &&
1313 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1314 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1315 log(LOG_DEBUG, "%s; %s: Listen socket: "
1316 "Connection attempt from/to self "
1317 "ignored\n", s, __func__);
1320 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1321 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1322 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1323 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1324 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1325 log(LOG_DEBUG, "%s; %s: Listen socket: "
1326 "Connection attempt from/to broad- "
1327 "or multicast address ignored\n",
1334 * SYN appears to be valid. Create compressed TCP state
1338 if (so->so_options & SO_DEBUG)
1339 tcp_trace(TA_INPUT, ostate, tp,
1340 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1342 TCP_PROBE3(debug__input, tp, th, m);
1343 tcp_dooptions(&to, optp, optlen, TO_SYN);
1344 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos))
1345 goto tfo_socket_result;
1348 * Entry added to syncache and mbuf consumed.
1349 * Only the listen socket is unlocked by syncache_add().
1351 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1352 return (IPPROTO_DONE);
1353 } else if (tp->t_state == TCPS_LISTEN) {
1355 * When a listen socket is torn down the SO_ACCEPTCONN
1356 * flag is removed first while connections are drained
1357 * from the accept queue in a unlock/lock cycle of the
1358 * ACCEPT_LOCK, opening a race condition allowing a SYN
1359 * attempt go through unhandled.
1363 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1364 if (tp->t_flags & TF_SIGNATURE) {
1365 tcp_dooptions(&to, optp, optlen, thflags);
1366 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1367 TCPSTAT_INC(tcps_sig_err_nosigopt);
1370 if (!TCPMD5_ENABLED() ||
1371 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1375 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1378 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1379 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1380 * the inpcb, and unlocks pcbinfo.
1382 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1383 return (IPPROTO_DONE);
1386 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1389 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1392 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1393 m = NULL; /* mbuf chain got consumed. */
1398 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1404 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1409 return (IPPROTO_DONE);
1413 * Automatic sizing of receive socket buffer. Often the send
1414 * buffer size is not optimally adjusted to the actual network
1415 * conditions at hand (delay bandwidth product). Setting the
1416 * buffer size too small limits throughput on links with high
1417 * bandwidth and high delay (eg. trans-continental/oceanic links).
1419 * On the receive side the socket buffer memory is only rarely
1420 * used to any significant extent. This allows us to be much
1421 * more aggressive in scaling the receive socket buffer. For
1422 * the case that the buffer space is actually used to a large
1423 * extent and we run out of kernel memory we can simply drop
1424 * the new segments; TCP on the sender will just retransmit it
1425 * later. Setting the buffer size too big may only consume too
1426 * much kernel memory if the application doesn't read() from
1427 * the socket or packet loss or reordering makes use of the
1430 * The criteria to step up the receive buffer one notch are:
1431 * 1. Application has not set receive buffer size with
1432 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1433 * 2. the number of bytes received during 1/2 of an sRTT
1434 * is at least 3/8 of the current socket buffer size.
1435 * 3. receive buffer size has not hit maximal automatic size;
1437 * If all of the criteria are met we increaset the socket buffer
1438 * by a 1/2 (bounded by the max). This allows us to keep ahead
1439 * of slow-start but also makes it so our peer never gets limited
1440 * by our rwnd which we then open up causing a burst.
1442 * This algorithm does two steps per RTT at most and only if
1443 * we receive a bulk stream w/o packet losses or reorderings.
1444 * Shrinking the buffer during idle times is not necessary as
1445 * it doesn't consume any memory when idle.
1447 * TODO: Only step up if the application is actually serving
1448 * the buffer to better manage the socket buffer resources.
1451 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1452 struct tcpcb *tp, int tlen)
1456 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1457 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1458 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1459 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1460 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1461 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1462 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1464 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1466 /* Start over with next RTT. */
1470 tp->rfbuf_cnt += tlen; /* add up */
1476 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1479 * Since tp might be gone if the session entered
1480 * the TIME_WAIT state before coming here, we need
1481 * to check if the socket is still connected.
1483 if ((so->so_state & SS_ISCONNECTED) == 0)
1485 INP_LOCK_ASSERT(tp->t_inpcb);
1486 if (tp->t_flags & TF_WAKESOR) {
1487 tp->t_flags &= ~TF_WAKESOR;
1488 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
1491 if (tp->t_flags & TF_WAKESOW) {
1492 tp->t_flags &= ~TF_WAKESOW;
1493 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
1499 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1500 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1502 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1503 int rstreason, todrop, win, incforsyn = 0;
1507 struct in_conninfo *inc;
1514 * The size of tcp_saveipgen must be the size of the max ip header,
1517 u_char tcp_saveipgen[IP6_HDR_LEN];
1518 struct tcphdr tcp_savetcp;
1521 thflags = th->th_flags;
1522 inc = &tp->t_inpcb->inp_inc;
1523 tp->sackhint.last_sack_ack = 0;
1525 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1528 INP_WLOCK_ASSERT(tp->t_inpcb);
1529 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1531 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1535 /* Save segment, if requested. */
1536 tcp_pcap_add(th, m, &(tp->t_inpkts));
1538 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1541 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1542 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1543 log(LOG_DEBUG, "%s; %s: "
1544 "SYN|FIN segment ignored (based on "
1545 "sysctl setting)\n", s, __func__);
1552 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1553 * check SEQ.ACK first.
1555 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1556 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1557 rstreason = BANDLIM_UNLIMITED;
1562 * Segment received on connection.
1563 * Reset idle time and keep-alive timer.
1564 * XXX: This should be done after segment
1565 * validation to ignore broken/spoofed segs.
1567 tp->t_rcvtime = ticks;
1570 * Scale up the window into a 32-bit value.
1571 * For the SYN_SENT state the scale is zero.
1573 tiwin = th->th_win << tp->snd_scale;
1575 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1579 * TCP ECN processing.
1581 if (tp->t_flags2 & TF2_ECN_PERMIT) {
1582 if (thflags & TH_CWR) {
1583 tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1584 tp->t_flags |= TF_ACKNOW;
1586 switch (iptos & IPTOS_ECN_MASK) {
1588 tp->t_flags2 |= TF2_ECN_SND_ECE;
1589 TCPSTAT_INC(tcps_ecn_ce);
1591 case IPTOS_ECN_ECT0:
1592 TCPSTAT_INC(tcps_ecn_ect0);
1594 case IPTOS_ECN_ECT1:
1595 TCPSTAT_INC(tcps_ecn_ect1);
1599 /* Process a packet differently from RFC3168. */
1600 cc_ecnpkt_handler(tp, th, iptos);
1602 /* Congestion experienced. */
1603 if (thflags & TH_ECE) {
1604 cc_cong_signal(tp, th, CC_ECN);
1609 * Parse options on any incoming segment.
1611 tcp_dooptions(&to, (u_char *)(th + 1),
1612 (th->th_off << 2) - sizeof(struct tcphdr),
1613 (thflags & TH_SYN) ? TO_SYN : 0);
1615 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1616 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1617 (to.to_flags & TOF_SIGNATURE) == 0) {
1618 TCPSTAT_INC(tcps_sig_err_sigopt);
1619 /* XXX: should drop? */
1623 * If echoed timestamp is later than the current time,
1624 * fall back to non RFC1323 RTT calculation. Normalize
1625 * timestamp if syncookies were used when this connection
1628 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1629 to.to_tsecr -= tp->ts_offset;
1630 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1632 else if (tp->t_flags & TF_PREVVALID &&
1633 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1634 cc_cong_signal(tp, th, CC_RTO_ERR);
1637 * Process options only when we get SYN/ACK back. The SYN case
1638 * for incoming connections is handled in tcp_syncache.
1639 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1640 * or <SYN,ACK>) segment itself is never scaled.
1641 * XXX this is traditional behavior, may need to be cleaned up.
1643 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1644 /* Handle parallel SYN for ECN */
1645 if (!(thflags & TH_ACK) &&
1646 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1647 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1648 tp->t_flags2 |= TF2_ECN_PERMIT;
1649 tp->t_flags2 |= TF2_ECN_SND_ECE;
1650 TCPSTAT_INC(tcps_ecn_shs);
1652 if ((to.to_flags & TOF_SCALE) &&
1653 (tp->t_flags & TF_REQ_SCALE)) {
1654 tp->t_flags |= TF_RCVD_SCALE;
1655 tp->snd_scale = to.to_wscale;
1657 tp->t_flags &= ~TF_REQ_SCALE;
1659 * Initial send window. It will be updated with
1660 * the next incoming segment to the scaled value.
1662 tp->snd_wnd = th->th_win;
1663 if ((to.to_flags & TOF_TS) &&
1664 (tp->t_flags & TF_REQ_TSTMP)) {
1665 tp->t_flags |= TF_RCVD_TSTMP;
1666 tp->ts_recent = to.to_tsval;
1667 tp->ts_recent_age = tcp_ts_getticks();
1669 tp->t_flags &= ~TF_REQ_TSTMP;
1670 if (to.to_flags & TOF_MSS)
1671 tcp_mss(tp, to.to_mss);
1672 if ((tp->t_flags & TF_SACK_PERMIT) &&
1673 (to.to_flags & TOF_SACKPERM) == 0)
1674 tp->t_flags &= ~TF_SACK_PERMIT;
1675 if (IS_FASTOPEN(tp->t_flags)) {
1676 if (to.to_flags & TOF_FASTOPEN) {
1679 if (to.to_flags & TOF_MSS)
1682 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1686 tcp_fastopen_update_cache(tp, mss,
1687 to.to_tfo_len, to.to_tfo_cookie);
1689 tcp_fastopen_disable_path(tp);
1694 * If timestamps were negotiated during SYN/ACK and a
1695 * segment without a timestamp is received, silently drop
1696 * the segment, unless it is a RST segment or missing timestamps are
1698 * See section 3.2 of RFC 7323.
1700 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1701 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1702 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1703 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1704 "segment processed normally\n",
1709 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1710 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1711 "segment silently dropped\n", s, __func__);
1718 * If timestamps were not negotiated during SYN/ACK and a
1719 * segment with a timestamp is received, ignore the
1720 * timestamp and process the packet normally.
1721 * See section 3.2 of RFC 7323.
1723 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1724 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1725 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1726 "segment processed normally\n", s, __func__);
1732 * Header prediction: check for the two common cases
1733 * of a uni-directional data xfer. If the packet has
1734 * no control flags, is in-sequence, the window didn't
1735 * change and we're not retransmitting, it's a
1736 * candidate. If the length is zero and the ack moved
1737 * forward, we're the sender side of the xfer. Just
1738 * free the data acked & wake any higher level process
1739 * that was blocked waiting for space. If the length
1740 * is non-zero and the ack didn't move, we're the
1741 * receiver side. If we're getting packets in-order
1742 * (the reassembly queue is empty), add the data to
1743 * the socket buffer and note that we need a delayed ack.
1744 * Make sure that the hidden state-flags are also off.
1745 * Since we check for TCPS_ESTABLISHED first, it can only
1748 if (tp->t_state == TCPS_ESTABLISHED &&
1749 th->th_seq == tp->rcv_nxt &&
1750 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1751 tp->snd_nxt == tp->snd_max &&
1752 tiwin && tiwin == tp->snd_wnd &&
1753 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1755 ((to.to_flags & TOF_TS) == 0 ||
1756 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1758 * If last ACK falls within this segment's sequence numbers,
1759 * record the timestamp.
1760 * NOTE that the test is modified according to the latest
1761 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1763 if ((to.to_flags & TOF_TS) != 0 &&
1764 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1765 tp->ts_recent_age = tcp_ts_getticks();
1766 tp->ts_recent = to.to_tsval;
1770 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1771 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1772 !IN_RECOVERY(tp->t_flags) &&
1773 (to.to_flags & TOF_SACK) == 0 &&
1774 TAILQ_EMPTY(&tp->snd_holes)) {
1776 * This is a pure ack for outstanding data.
1778 TCPSTAT_INC(tcps_predack);
1781 * "bad retransmit" recovery without timestamps.
1783 if ((to.to_flags & TOF_TS) == 0 &&
1784 tp->t_rxtshift == 1 &&
1785 tp->t_flags & TF_PREVVALID &&
1786 (int)(ticks - tp->t_badrxtwin) < 0) {
1787 cc_cong_signal(tp, th, CC_RTO_ERR);
1791 * Recalculate the transmit timer / rtt.
1793 * Some boxes send broken timestamp replies
1794 * during the SYN+ACK phase, ignore
1795 * timestamps of 0 or we could calculate a
1796 * huge RTT and blow up the retransmit timer.
1798 if ((to.to_flags & TOF_TS) != 0 &&
1802 t = tcp_ts_getticks() - to.to_tsecr;
1803 if (!tp->t_rttlow || tp->t_rttlow > t)
1806 TCP_TS_TO_TICKS(t) + 1);
1807 } else if (tp->t_rtttime &&
1808 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1809 if (!tp->t_rttlow ||
1810 tp->t_rttlow > ticks - tp->t_rtttime)
1811 tp->t_rttlow = ticks - tp->t_rtttime;
1813 ticks - tp->t_rtttime);
1815 acked = BYTES_THIS_ACK(tp, th);
1818 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1819 hhook_run_tcp_est_in(tp, th, &to);
1822 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1823 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1824 sbdrop(&so->so_snd, acked);
1825 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1826 SEQ_LEQ(th->th_ack, tp->snd_recover))
1827 tp->snd_recover = th->th_ack - 1;
1830 * Let the congestion control algorithm update
1831 * congestion control related information. This
1832 * typically means increasing the congestion
1835 cc_ack_received(tp, th, nsegs, CC_ACK);
1837 tp->snd_una = th->th_ack;
1839 * Pull snd_wl2 up to prevent seq wrap relative
1842 tp->snd_wl2 = th->th_ack;
1847 * If all outstanding data are acked, stop
1848 * retransmit timer, otherwise restart timer
1849 * using current (possibly backed-off) value.
1850 * If process is waiting for space,
1851 * wakeup/selwakeup/signal. If data
1852 * are ready to send, let tcp_output
1853 * decide between more output or persist.
1856 if (so->so_options & SO_DEBUG)
1857 tcp_trace(TA_INPUT, ostate, tp,
1858 (void *)tcp_saveipgen,
1861 TCP_PROBE3(debug__input, tp, th, m);
1862 if (tp->snd_una == tp->snd_max)
1863 tcp_timer_activate(tp, TT_REXMT, 0);
1864 else if (!tcp_timer_active(tp, TT_PERSIST))
1865 tcp_timer_activate(tp, TT_REXMT,
1867 tp->t_flags |= TF_WAKESOW;
1868 if (sbavail(&so->so_snd))
1869 (void) tp->t_fb->tfb_tcp_output(tp);
1872 } else if (th->th_ack == tp->snd_una &&
1873 tlen <= sbspace(&so->so_rcv)) {
1874 int newsize = 0; /* automatic sockbuf scaling */
1877 * This is a pure, in-sequence data packet with
1878 * nothing on the reassembly queue and we have enough
1879 * buffer space to take it.
1881 /* Clean receiver SACK report if present */
1882 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1883 tcp_clean_sackreport(tp);
1884 TCPSTAT_INC(tcps_preddat);
1885 tp->rcv_nxt += tlen;
1887 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1888 (tp->t_fbyte_in == 0)) {
1889 tp->t_fbyte_in = ticks;
1890 if (tp->t_fbyte_in == 0)
1892 if (tp->t_fbyte_out && tp->t_fbyte_in)
1893 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1896 * Pull snd_wl1 up to prevent seq wrap relative to
1899 tp->snd_wl1 = th->th_seq;
1901 * Pull rcv_up up to prevent seq wrap relative to
1904 tp->rcv_up = tp->rcv_nxt;
1905 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1906 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1908 if (so->so_options & SO_DEBUG)
1909 tcp_trace(TA_INPUT, ostate, tp,
1910 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1912 TCP_PROBE3(debug__input, tp, th, m);
1914 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1916 /* Add data to socket buffer. */
1917 SOCKBUF_LOCK(&so->so_rcv);
1918 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1922 * Set new socket buffer size.
1923 * Give up when limit is reached.
1926 if (!sbreserve_locked(&so->so_rcv,
1928 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1929 m_adj(m, drop_hdrlen); /* delayed header drop */
1930 sbappendstream_locked(&so->so_rcv, m, 0);
1932 SOCKBUF_UNLOCK(&so->so_rcv);
1933 tp->t_flags |= TF_WAKESOR;
1934 if (DELAY_ACK(tp, tlen)) {
1935 tp->t_flags |= TF_DELACK;
1937 tp->t_flags |= TF_ACKNOW;
1938 tp->t_fb->tfb_tcp_output(tp);
1945 * Calculate amount of space in receive window,
1946 * and then do TCP input processing.
1947 * Receive window is amount of space in rcv queue,
1948 * but not less than advertised window.
1950 win = sbspace(&so->so_rcv);
1953 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1955 switch (tp->t_state) {
1957 * If the state is SYN_RECEIVED:
1958 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1960 case TCPS_SYN_RECEIVED:
1961 if ((thflags & TH_ACK) &&
1962 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1963 SEQ_GT(th->th_ack, tp->snd_max))) {
1964 rstreason = BANDLIM_RST_OPENPORT;
1967 if (IS_FASTOPEN(tp->t_flags)) {
1969 * When a TFO connection is in SYN_RECEIVED, the
1970 * only valid packets are the initial SYN, a
1971 * retransmit/copy of the initial SYN (possibly with
1972 * a subset of the original data), a valid ACK, a
1975 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1976 rstreason = BANDLIM_RST_OPENPORT;
1978 } else if (thflags & TH_SYN) {
1979 /* non-initial SYN is ignored */
1980 if ((tcp_timer_active(tp, TT_DELACK) ||
1981 tcp_timer_active(tp, TT_REXMT)))
1983 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1990 * If the state is SYN_SENT:
1991 * if seg contains a RST with valid ACK (SEQ.ACK has already
1992 * been verified), then drop the connection.
1993 * if seg contains a RST without an ACK, drop the seg.
1994 * if seg does not contain SYN, then drop the seg.
1995 * Otherwise this is an acceptable SYN segment
1996 * initialize tp->rcv_nxt and tp->irs
1997 * if seg contains ack then advance tp->snd_una
1998 * if seg contains an ECE and ECN support is enabled, the stream
2000 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2001 * arrange for segment to be acked (eventually)
2002 * continue processing rest of data/controls, beginning with URG
2005 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2006 TCP_PROBE5(connect__refused, NULL, tp,
2008 tp = tcp_drop(tp, ECONNREFUSED);
2010 if (thflags & TH_RST)
2012 if (!(thflags & TH_SYN))
2015 tp->irs = th->th_seq;
2017 if (thflags & TH_ACK) {
2018 int tfo_partial_ack = 0;
2020 TCPSTAT_INC(tcps_connects);
2023 mac_socketpeer_set_from_mbuf(m, so);
2025 /* Do window scaling on this connection? */
2026 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2027 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2028 tp->rcv_scale = tp->request_r_scale;
2030 tp->rcv_adv += min(tp->rcv_wnd,
2031 TCP_MAXWIN << tp->rcv_scale);
2032 tp->snd_una++; /* SYN is acked */
2034 * If not all the data that was sent in the TFO SYN
2035 * has been acked, resend the remainder right away.
2037 if (IS_FASTOPEN(tp->t_flags) &&
2038 (tp->snd_una != tp->snd_max)) {
2039 tp->snd_nxt = th->th_ack;
2040 tfo_partial_ack = 1;
2043 * If there's data, delay ACK; if there's also a FIN
2044 * ACKNOW will be turned on later.
2046 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2047 tcp_timer_activate(tp, TT_DELACK,
2050 tp->t_flags |= TF_ACKNOW;
2052 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2053 (V_tcp_do_ecn == 1)) {
2054 tp->t_flags2 |= TF2_ECN_PERMIT;
2055 TCPSTAT_INC(tcps_ecn_shs);
2059 * Received <SYN,ACK> in SYN_SENT[*] state.
2061 * SYN_SENT --> ESTABLISHED
2062 * SYN_SENT* --> FIN_WAIT_1
2064 tp->t_starttime = ticks;
2065 if (tp->t_flags & TF_NEEDFIN) {
2066 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2067 tp->t_flags &= ~TF_NEEDFIN;
2070 tcp_state_change(tp, TCPS_ESTABLISHED);
2071 TCP_PROBE5(connect__established, NULL, tp,
2074 tcp_timer_activate(tp, TT_KEEP,
2079 * Received initial SYN in SYN-SENT[*] state =>
2080 * simultaneous open.
2081 * If it succeeds, connection is * half-synchronized.
2082 * Otherwise, do 3-way handshake:
2083 * SYN-SENT -> SYN-RECEIVED
2084 * SYN-SENT* -> SYN-RECEIVED*
2086 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2087 tcp_timer_activate(tp, TT_REXMT, 0);
2088 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2091 INP_WLOCK_ASSERT(tp->t_inpcb);
2094 * Advance th->th_seq to correspond to first data byte.
2095 * If data, trim to stay within window,
2096 * dropping FIN if necessary.
2099 if (tlen > tp->rcv_wnd) {
2100 todrop = tlen - tp->rcv_wnd;
2104 TCPSTAT_INC(tcps_rcvpackafterwin);
2105 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2107 tp->snd_wl1 = th->th_seq - 1;
2108 tp->rcv_up = th->th_seq;
2110 * Client side of transaction: already sent SYN and data.
2111 * If the remote host used T/TCP to validate the SYN,
2112 * our data will be ACK'd; if so, enter normal data segment
2113 * processing in the middle of step 5, ack processing.
2114 * Otherwise, goto step 6.
2116 if (thflags & TH_ACK)
2122 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2123 * do normal processing.
2125 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2129 break; /* continue normal processing */
2133 * States other than LISTEN or SYN_SENT.
2134 * First check the RST flag and sequence number since reset segments
2135 * are exempt from the timestamp and connection count tests. This
2136 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2137 * below which allowed reset segments in half the sequence space
2138 * to fall though and be processed (which gives forged reset
2139 * segments with a random sequence number a 50 percent chance of
2140 * killing a connection).
2141 * Then check timestamp, if present.
2142 * Then check the connection count, if present.
2143 * Then check that at least some bytes of segment are within
2144 * receive window. If segment begins before rcv_nxt,
2145 * drop leading data (and SYN); if nothing left, just ack.
2147 if (thflags & TH_RST) {
2149 * RFC5961 Section 3.2
2151 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2152 * - If RST is in window, we send challenge ACK.
2154 * Note: to take into account delayed ACKs, we should
2155 * test against last_ack_sent instead of rcv_nxt.
2156 * Note 2: we handle special case of closed window, not
2157 * covered by the RFC.
2159 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2160 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2161 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2162 KASSERT(tp->t_state != TCPS_SYN_SENT,
2163 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2166 if (V_tcp_insecure_rst ||
2167 tp->last_ack_sent == th->th_seq) {
2168 TCPSTAT_INC(tcps_drops);
2169 /* Drop the connection. */
2170 switch (tp->t_state) {
2171 case TCPS_SYN_RECEIVED:
2172 so->so_error = ECONNREFUSED;
2174 case TCPS_ESTABLISHED:
2175 case TCPS_FIN_WAIT_1:
2176 case TCPS_FIN_WAIT_2:
2177 case TCPS_CLOSE_WAIT:
2180 so->so_error = ECONNRESET;
2187 TCPSTAT_INC(tcps_badrst);
2188 /* Send challenge ACK. */
2189 tcp_respond(tp, mtod(m, void *), th, m,
2190 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2191 tp->last_ack_sent = tp->rcv_nxt;
2199 * RFC5961 Section 4.2
2200 * Send challenge ACK for any SYN in synchronized state.
2202 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2203 tp->t_state != TCPS_SYN_RECEIVED) {
2204 TCPSTAT_INC(tcps_badsyn);
2205 if (V_tcp_insecure_syn &&
2206 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2207 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2208 tp = tcp_drop(tp, ECONNRESET);
2209 rstreason = BANDLIM_UNLIMITED;
2211 /* Send challenge ACK. */
2212 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2213 tp->snd_nxt, TH_ACK);
2214 tp->last_ack_sent = tp->rcv_nxt;
2221 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2222 * and it's less than ts_recent, drop it.
2224 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2225 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2226 /* Check to see if ts_recent is over 24 days old. */
2227 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2229 * Invalidate ts_recent. If this segment updates
2230 * ts_recent, the age will be reset later and ts_recent
2231 * will get a valid value. If it does not, setting
2232 * ts_recent to zero will at least satisfy the
2233 * requirement that zero be placed in the timestamp
2234 * echo reply when ts_recent isn't valid. The
2235 * age isn't reset until we get a valid ts_recent
2236 * because we don't want out-of-order segments to be
2237 * dropped when ts_recent is old.
2241 TCPSTAT_INC(tcps_rcvduppack);
2242 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2243 TCPSTAT_INC(tcps_pawsdrop);
2251 * In the SYN-RECEIVED state, validate that the packet belongs to
2252 * this connection before trimming the data to fit the receive
2253 * window. Check the sequence number versus IRS since we know
2254 * the sequence numbers haven't wrapped. This is a partial fix
2255 * for the "LAND" DoS attack.
2257 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2258 rstreason = BANDLIM_RST_OPENPORT;
2262 todrop = tp->rcv_nxt - th->th_seq;
2264 if (thflags & TH_SYN) {
2274 * Following if statement from Stevens, vol. 2, p. 960.
2277 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2279 * Any valid FIN must be to the left of the window.
2280 * At this point the FIN must be a duplicate or out
2281 * of sequence; drop it.
2286 * Send an ACK to resynchronize and drop any data.
2287 * But keep on processing for RST or ACK.
2289 tp->t_flags |= TF_ACKNOW;
2291 TCPSTAT_INC(tcps_rcvduppack);
2292 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2294 TCPSTAT_INC(tcps_rcvpartduppack);
2295 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2298 * DSACK - add SACK block for dropped range
2300 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2301 tcp_update_sack_list(tp, th->th_seq,
2302 th->th_seq + todrop);
2304 * ACK now, as the next in-sequence segment
2305 * will clear the DSACK block again
2307 tp->t_flags |= TF_ACKNOW;
2309 drop_hdrlen += todrop; /* drop from the top afterwards */
2310 th->th_seq += todrop;
2312 if (th->th_urp > todrop)
2313 th->th_urp -= todrop;
2321 * If new data are received on a connection after the
2322 * user processes are gone, then RST the other end.
2324 if ((so->so_state & SS_NOFDREF) &&
2325 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2326 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2327 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2328 "after socket was closed, "
2329 "sending RST and removing tcpcb\n",
2330 s, __func__, tcpstates[tp->t_state], tlen);
2334 TCPSTAT_INC(tcps_rcvafterclose);
2335 rstreason = BANDLIM_UNLIMITED;
2340 * If segment ends after window, drop trailing data
2341 * (and PUSH and FIN); if nothing left, just ACK.
2343 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2345 TCPSTAT_INC(tcps_rcvpackafterwin);
2346 if (todrop >= tlen) {
2347 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2349 * If window is closed can only take segments at
2350 * window edge, and have to drop data and PUSH from
2351 * incoming segments. Continue processing, but
2352 * remember to ack. Otherwise, drop segment
2355 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2356 tp->t_flags |= TF_ACKNOW;
2357 TCPSTAT_INC(tcps_rcvwinprobe);
2361 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2364 thflags &= ~(TH_PUSH|TH_FIN);
2368 * If last ACK falls within this segment's sequence numbers,
2369 * record its timestamp.
2371 * 1) That the test incorporates suggestions from the latest
2372 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2373 * 2) That updating only on newer timestamps interferes with
2374 * our earlier PAWS tests, so this check should be solely
2375 * predicated on the sequence space of this segment.
2376 * 3) That we modify the segment boundary check to be
2377 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2378 * instead of RFC1323's
2379 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2380 * This modified check allows us to overcome RFC1323's
2381 * limitations as described in Stevens TCP/IP Illustrated
2382 * Vol. 2 p.869. In such cases, we can still calculate the
2383 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2385 if ((to.to_flags & TOF_TS) != 0 &&
2386 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2387 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2388 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2389 tp->ts_recent_age = tcp_ts_getticks();
2390 tp->ts_recent = to.to_tsval;
2394 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2395 * flag is on (half-synchronized state), then queue data for
2396 * later processing; else drop segment and return.
2398 if ((thflags & TH_ACK) == 0) {
2399 if (tp->t_state == TCPS_SYN_RECEIVED ||
2400 (tp->t_flags & TF_NEEDSYN)) {
2401 if (tp->t_state == TCPS_SYN_RECEIVED &&
2402 IS_FASTOPEN(tp->t_flags)) {
2403 tp->snd_wnd = tiwin;
2407 } else if (tp->t_flags & TF_ACKNOW)
2416 switch (tp->t_state) {
2418 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2419 * ESTABLISHED state and continue processing.
2420 * The ACK was checked above.
2422 case TCPS_SYN_RECEIVED:
2424 TCPSTAT_INC(tcps_connects);
2426 /* Do window scaling? */
2427 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2428 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2429 tp->rcv_scale = tp->request_r_scale;
2431 tp->snd_wnd = tiwin;
2434 * SYN-RECEIVED -> ESTABLISHED
2435 * SYN-RECEIVED* -> FIN-WAIT-1
2437 tp->t_starttime = ticks;
2438 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2439 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2440 tp->t_tfo_pending = NULL;
2442 if (tp->t_flags & TF_NEEDFIN) {
2443 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2444 tp->t_flags &= ~TF_NEEDFIN;
2446 tcp_state_change(tp, TCPS_ESTABLISHED);
2447 TCP_PROBE5(accept__established, NULL, tp,
2450 * TFO connections call cc_conn_init() during SYN
2451 * processing. Calling it again here for such
2452 * connections is not harmless as it would undo the
2453 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2456 if (!IS_FASTOPEN(tp->t_flags))
2458 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2461 * Account for the ACK of our SYN prior to
2462 * regular ACK processing below, except for
2463 * simultaneous SYN, which is handled later.
2465 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2468 * If segment contains data or ACK, will call tcp_reass()
2469 * later; if not, do so now to pass queued data to user.
2471 if (tlen == 0 && (thflags & TH_FIN) == 0)
2472 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2474 tp->snd_wl1 = th->th_seq - 1;
2478 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2479 * ACKs. If the ack is in the range
2480 * tp->snd_una < th->th_ack <= tp->snd_max
2481 * then advance tp->snd_una to th->th_ack and drop
2482 * data from the retransmission queue. If this ACK reflects
2483 * more up to date window information we update our window information.
2485 case TCPS_ESTABLISHED:
2486 case TCPS_FIN_WAIT_1:
2487 case TCPS_FIN_WAIT_2:
2488 case TCPS_CLOSE_WAIT:
2491 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2492 TCPSTAT_INC(tcps_rcvacktoomuch);
2495 if ((tp->t_flags & TF_SACK_PERMIT) &&
2496 ((to.to_flags & TOF_SACK) ||
2497 !TAILQ_EMPTY(&tp->snd_holes)))
2498 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2501 * Reset the value so that previous (valid) value
2502 * from the last ack with SACK doesn't get used.
2504 tp->sackhint.sacked_bytes = 0;
2507 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2508 hhook_run_tcp_est_in(tp, th, &to);
2511 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2514 maxseg = tcp_maxseg(tp);
2516 (tiwin == tp->snd_wnd ||
2517 (tp->t_flags & TF_SACK_PERMIT))) {
2519 * If this is the first time we've seen a
2520 * FIN from the remote, this is not a
2521 * duplicate and it needs to be processed
2522 * normally. This happens during a
2523 * simultaneous close.
2525 if ((thflags & TH_FIN) &&
2526 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2530 TCPSTAT_INC(tcps_rcvdupack);
2532 * If we have outstanding data (other than
2533 * a window probe), this is a completely
2534 * duplicate ack (ie, window info didn't
2535 * change and FIN isn't set),
2536 * the ack is the biggest we've
2537 * seen and we've seen exactly our rexmt
2538 * threshold of them, assume a packet
2539 * has been dropped and retransmit it.
2540 * Kludge snd_nxt & the congestion
2541 * window so we send only this one
2544 * We know we're losing at the current
2545 * window size so do congestion avoidance
2546 * (set ssthresh to half the current window
2547 * and pull our congestion window back to
2548 * the new ssthresh).
2550 * Dup acks mean that packets have left the
2551 * network (they're now cached at the receiver)
2552 * so bump cwnd by the amount in the receiver
2553 * to keep a constant cwnd packets in the
2556 * When using TCP ECN, notify the peer that
2557 * we reduced the cwnd.
2560 * Following 2 kinds of acks should not affect
2563 * 2) Acks with SACK but without any new SACK
2564 * information in them. These could result from
2565 * any anomaly in the network like a switch
2566 * duplicating packets or a possible DoS attack.
2568 if (th->th_ack != tp->snd_una ||
2569 ((tp->t_flags & TF_SACK_PERMIT) &&
2572 else if (!tcp_timer_active(tp, TT_REXMT))
2574 else if (++tp->t_dupacks > tcprexmtthresh ||
2575 IN_FASTRECOVERY(tp->t_flags)) {
2576 cc_ack_received(tp, th, nsegs,
2579 IN_FASTRECOVERY(tp->t_flags) &&
2580 (tp->t_flags & TF_SACK_PERMIT)) {
2581 long snd_cnt = 0, limit = 0;
2582 long del_data = 0, pipe = 0;
2584 * In a duplicate ACK del_data is only the
2585 * diff_in_sack. If no SACK is used del_data
2586 * will be 0. Pipe is the amount of data we
2587 * estimate to be in the network.
2589 del_data = tp->sackhint.delivered_data;
2590 pipe = (tp->snd_nxt - tp->snd_fack) +
2591 tp->sackhint.sack_bytes_rexmit;
2592 tp->sackhint.prr_delivered += del_data;
2593 if (pipe > tp->snd_ssthresh) {
2594 if (tp->sackhint.recover_fs == 0)
2595 tp->sackhint.recover_fs =
2596 max(1, tp->snd_nxt - tp->snd_una);
2597 snd_cnt = (tp->sackhint.prr_delivered *
2599 tp->sackhint.recover_fs) +
2600 1 - tp->sackhint.sack_bytes_rexmit;
2602 if (V_tcp_do_prr_conservative)
2603 limit = tp->sackhint.prr_delivered -
2604 tp->sackhint.sack_bytes_rexmit;
2606 if ((tp->sackhint.prr_delivered -
2607 tp->sackhint.sack_bytes_rexmit) >
2609 limit = tp->sackhint.prr_delivered -
2610 tp->sackhint.sack_bytes_rexmit +
2613 limit = del_data + maxseg;
2614 if ((tp->snd_ssthresh - pipe) < limit)
2615 snd_cnt = tp->snd_ssthresh - pipe;
2619 snd_cnt = max((snd_cnt / maxseg), 0);
2621 * Send snd_cnt new data into the network in
2622 * response to this ACK. If there is a going
2623 * to be a SACK retransmission, adjust snd_cwnd
2626 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
2627 tp->sackhint.sack_bytes_rexmit +
2629 } else if ((tp->t_flags & TF_SACK_PERMIT) &&
2630 IN_FASTRECOVERY(tp->t_flags)) {
2634 * Compute the amount of data in flight first.
2635 * We can inject new data into the pipe iff
2636 * we have less than 1/2 the original window's
2637 * worth of data in flight.
2639 if (V_tcp_do_rfc6675_pipe)
2640 awnd = tcp_compute_pipe(tp);
2642 awnd = (tp->snd_nxt - tp->snd_fack) +
2643 tp->sackhint.sack_bytes_rexmit;
2645 if (awnd < tp->snd_ssthresh) {
2646 tp->snd_cwnd += maxseg;
2647 if (tp->snd_cwnd > tp->snd_ssthresh)
2648 tp->snd_cwnd = tp->snd_ssthresh;
2651 tp->snd_cwnd += maxseg;
2652 (void) tp->t_fb->tfb_tcp_output(tp);
2654 } else if (tp->t_dupacks == tcprexmtthresh) {
2655 tcp_seq onxt = tp->snd_nxt;
2658 * If we're doing sack, or prr, check
2659 * to see if we're already in sack
2660 * recovery. If we're not doing sack,
2661 * check to see if we're in newreno
2665 (tp->t_flags & TF_SACK_PERMIT)) {
2666 if (IN_FASTRECOVERY(tp->t_flags)) {
2671 if (SEQ_LEQ(th->th_ack,
2677 /* Congestion signal before ack. */
2678 cc_cong_signal(tp, th, CC_NDUPACK);
2679 cc_ack_received(tp, th, nsegs,
2681 tcp_timer_activate(tp, TT_REXMT, 0);
2685 * snd_ssthresh is already updated by
2688 tp->sackhint.prr_delivered = 0;
2689 tp->sackhint.sack_bytes_rexmit = 0;
2690 tp->sackhint.recover_fs = max(1,
2691 tp->snd_nxt - tp->snd_una);
2693 if (tp->t_flags & TF_SACK_PERMIT) {
2695 tcps_sack_recovery_episode);
2696 tp->snd_recover = tp->snd_nxt;
2697 tp->snd_cwnd = maxseg;
2698 (void) tp->t_fb->tfb_tcp_output(tp);
2701 tp->snd_nxt = th->th_ack;
2702 tp->snd_cwnd = maxseg;
2703 (void) tp->t_fb->tfb_tcp_output(tp);
2704 KASSERT(tp->snd_limited <= 2,
2705 ("%s: tp->snd_limited too big",
2707 tp->snd_cwnd = tp->snd_ssthresh +
2709 (tp->t_dupacks - tp->snd_limited);
2710 if (SEQ_GT(onxt, tp->snd_nxt))
2713 } else if (V_tcp_do_rfc3042) {
2715 * Process first and second duplicate
2716 * ACKs. Each indicates a segment
2717 * leaving the network, creating room
2718 * for more. Make sure we can send a
2719 * packet on reception of each duplicate
2720 * ACK by increasing snd_cwnd by one
2721 * segment. Restore the original
2722 * snd_cwnd after packet transmission.
2724 cc_ack_received(tp, th, nsegs,
2726 uint32_t oldcwnd = tp->snd_cwnd;
2727 tcp_seq oldsndmax = tp->snd_max;
2731 KASSERT(tp->t_dupacks == 1 ||
2733 ("%s: dupacks not 1 or 2",
2735 if (tp->t_dupacks == 1)
2736 tp->snd_limited = 0;
2738 (tp->snd_nxt - tp->snd_una) +
2739 (tp->t_dupacks - tp->snd_limited) *
2742 * Only call tcp_output when there
2743 * is new data available to be sent.
2744 * Otherwise we would send pure ACKs.
2746 SOCKBUF_LOCK(&so->so_snd);
2747 avail = sbavail(&so->so_snd) -
2748 (tp->snd_nxt - tp->snd_una);
2749 SOCKBUF_UNLOCK(&so->so_snd);
2751 (void) tp->t_fb->tfb_tcp_output(tp);
2752 sent = tp->snd_max - oldsndmax;
2753 if (sent > maxseg) {
2754 KASSERT((tp->t_dupacks == 2 &&
2755 tp->snd_limited == 0) ||
2756 (sent == maxseg + 1 &&
2757 tp->t_flags & TF_SENTFIN),
2758 ("%s: sent too much",
2760 tp->snd_limited = 2;
2761 } else if (sent > 0)
2763 tp->snd_cwnd = oldcwnd;
2770 * This ack is advancing the left edge, reset the
2775 * If this ack also has new SACK info, increment the
2776 * counter as per rfc6675. The variable
2777 * sack_changed tracks all changes to the SACK
2778 * scoreboard, including when partial ACKs without
2779 * SACK options are received, and clear the scoreboard
2780 * from the left side. Such partial ACKs should not be
2781 * counted as dupacks here.
2783 if ((tp->t_flags & TF_SACK_PERMIT) &&
2784 (to.to_flags & TOF_SACK) &&
2789 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2790 ("%s: th_ack <= snd_una", __func__));
2793 * If the congestion window was inflated to account
2794 * for the other side's cached packets, retract it.
2796 if (IN_FASTRECOVERY(tp->t_flags)) {
2797 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2798 if (tp->t_flags & TF_SACK_PERMIT)
2800 tcp_prr_partialack(tp, th);
2802 tcp_sack_partialack(tp, th);
2804 tcp_newreno_partial_ack(tp, th);
2806 cc_post_recovery(tp, th);
2809 * If we reach this point, ACK is not a duplicate,
2810 * i.e., it ACKs something we sent.
2812 if (tp->t_flags & TF_NEEDSYN) {
2814 * T/TCP: Connection was half-synchronized, and our
2815 * SYN has been ACK'd (so connection is now fully
2816 * synchronized). Go to non-starred state,
2817 * increment snd_una for ACK of SYN, and check if
2818 * we can do window scaling.
2820 tp->t_flags &= ~TF_NEEDSYN;
2822 /* Do window scaling? */
2823 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2824 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2825 tp->rcv_scale = tp->request_r_scale;
2826 /* Send window already scaled. */
2831 INP_WLOCK_ASSERT(tp->t_inpcb);
2834 * Adjust for the SYN bit in sequence space,
2835 * but don't account for it in cwnd calculations.
2836 * This is for the SYN_RECEIVED, non-simultaneous
2837 * SYN case. SYN_SENT and simultaneous SYN are
2838 * treated elsewhere.
2842 acked = BYTES_THIS_ACK(tp, th);
2843 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2844 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2845 tp->snd_una, th->th_ack, tp, m));
2846 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2847 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2850 * If we just performed our first retransmit, and the ACK
2851 * arrives within our recovery window, then it was a mistake
2852 * to do the retransmit in the first place. Recover our
2853 * original cwnd and ssthresh, and proceed to transmit where
2856 if (tp->t_rxtshift == 1 &&
2857 tp->t_flags & TF_PREVVALID &&
2859 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2860 cc_cong_signal(tp, th, CC_RTO_ERR);
2863 * If we have a timestamp reply, update smoothed
2864 * round trip time. If no timestamp is present but
2865 * transmit timer is running and timed sequence
2866 * number was acked, update smoothed round trip time.
2867 * Since we now have an rtt measurement, cancel the
2868 * timer backoff (cf., Phil Karn's retransmit alg.).
2869 * Recompute the initial retransmit timer.
2871 * Some boxes send broken timestamp replies
2872 * during the SYN+ACK phase, ignore
2873 * timestamps of 0 or we could calculate a
2874 * huge RTT and blow up the retransmit timer.
2876 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2879 t = tcp_ts_getticks() - to.to_tsecr;
2880 if (!tp->t_rttlow || tp->t_rttlow > t)
2882 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2883 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2884 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2885 tp->t_rttlow = ticks - tp->t_rtttime;
2886 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2890 * If all outstanding data is acked, stop retransmit
2891 * timer and remember to restart (more output or persist).
2892 * If there is more data to be acked, restart retransmit
2893 * timer, using current (possibly backed-off) value.
2895 if (th->th_ack == tp->snd_max) {
2896 tcp_timer_activate(tp, TT_REXMT, 0);
2898 } else if (!tcp_timer_active(tp, TT_PERSIST))
2899 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2902 * If no data (only SYN) was ACK'd,
2903 * skip rest of ACK processing.
2909 * Let the congestion control algorithm update congestion
2910 * control related information. This typically means increasing
2911 * the congestion window.
2913 cc_ack_received(tp, th, nsegs, CC_ACK);
2915 SOCKBUF_LOCK(&so->so_snd);
2916 if (acked > sbavail(&so->so_snd)) {
2917 if (tp->snd_wnd >= sbavail(&so->so_snd))
2918 tp->snd_wnd -= sbavail(&so->so_snd);
2921 mfree = sbcut_locked(&so->so_snd,
2922 (int)sbavail(&so->so_snd));
2925 mfree = sbcut_locked(&so->so_snd, acked);
2926 if (tp->snd_wnd >= (uint32_t) acked)
2927 tp->snd_wnd -= acked;
2932 SOCKBUF_UNLOCK(&so->so_snd);
2933 tp->t_flags |= TF_WAKESOW;
2935 /* Detect una wraparound. */
2936 if (!IN_RECOVERY(tp->t_flags) &&
2937 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2938 SEQ_LEQ(th->th_ack, tp->snd_recover))
2939 tp->snd_recover = th->th_ack - 1;
2940 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2941 if (IN_RECOVERY(tp->t_flags) &&
2942 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2943 EXIT_RECOVERY(tp->t_flags);
2945 tp->snd_una = th->th_ack;
2946 if (tp->t_flags & TF_SACK_PERMIT) {
2947 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2948 tp->snd_recover = tp->snd_una;
2950 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2951 tp->snd_nxt = tp->snd_una;
2953 switch (tp->t_state) {
2955 * In FIN_WAIT_1 STATE in addition to the processing
2956 * for the ESTABLISHED state if our FIN is now acknowledged
2957 * then enter FIN_WAIT_2.
2959 case TCPS_FIN_WAIT_1:
2960 if (ourfinisacked) {
2962 * If we can't receive any more
2963 * data, then closing user can proceed.
2964 * Starting the timer is contrary to the
2965 * specification, but if we don't get a FIN
2966 * we'll hang forever.
2969 * we should release the tp also, and use a
2972 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2973 soisdisconnected(so);
2974 tcp_timer_activate(tp, TT_2MSL,
2975 (tcp_fast_finwait2_recycle ?
2976 tcp_finwait2_timeout :
2979 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2984 * In CLOSING STATE in addition to the processing for
2985 * the ESTABLISHED state if the ACK acknowledges our FIN
2986 * then enter the TIME-WAIT state, otherwise ignore
2990 if (ourfinisacked) {
2998 * In LAST_ACK, we may still be waiting for data to drain
2999 * and/or to be acked, as well as for the ack of our FIN.
3000 * If our FIN is now acknowledged, delete the TCB,
3001 * enter the closed state and return.
3004 if (ourfinisacked) {
3013 INP_WLOCK_ASSERT(tp->t_inpcb);
3016 * Update window information.
3017 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3019 if ((thflags & TH_ACK) &&
3020 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3021 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3022 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3023 /* keep track of pure window updates */
3025 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3026 TCPSTAT_INC(tcps_rcvwinupd);
3027 tp->snd_wnd = tiwin;
3028 tp->snd_wl1 = th->th_seq;
3029 tp->snd_wl2 = th->th_ack;
3030 if (tp->snd_wnd > tp->max_sndwnd)
3031 tp->max_sndwnd = tp->snd_wnd;
3036 * Process segments with URG.
3038 if ((thflags & TH_URG) && th->th_urp &&
3039 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3041 * This is a kludge, but if we receive and accept
3042 * random urgent pointers, we'll crash in
3043 * soreceive. It's hard to imagine someone
3044 * actually wanting to send this much urgent data.
3046 SOCKBUF_LOCK(&so->so_rcv);
3047 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3048 th->th_urp = 0; /* XXX */
3049 thflags &= ~TH_URG; /* XXX */
3050 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3051 goto dodata; /* XXX */
3054 * If this segment advances the known urgent pointer,
3055 * then mark the data stream. This should not happen
3056 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3057 * a FIN has been received from the remote side.
3058 * In these states we ignore the URG.
3060 * According to RFC961 (Assigned Protocols),
3061 * the urgent pointer points to the last octet
3062 * of urgent data. We continue, however,
3063 * to consider it to indicate the first octet
3064 * of data past the urgent section as the original
3065 * spec states (in one of two places).
3067 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3068 tp->rcv_up = th->th_seq + th->th_urp;
3069 so->so_oobmark = sbavail(&so->so_rcv) +
3070 (tp->rcv_up - tp->rcv_nxt) - 1;
3071 if (so->so_oobmark == 0)
3072 so->so_rcv.sb_state |= SBS_RCVATMARK;
3074 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3076 SOCKBUF_UNLOCK(&so->so_rcv);
3078 * Remove out of band data so doesn't get presented to user.
3079 * This can happen independent of advancing the URG pointer,
3080 * but if two URG's are pending at once, some out-of-band
3081 * data may creep in... ick.
3083 if (th->th_urp <= (uint32_t)tlen &&
3084 !(so->so_options & SO_OOBINLINE)) {
3085 /* hdr drop is delayed */
3086 tcp_pulloutofband(so, th, m, drop_hdrlen);
3090 * If no out of band data is expected,
3091 * pull receive urgent pointer along
3092 * with the receive window.
3094 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3095 tp->rcv_up = tp->rcv_nxt;
3098 INP_WLOCK_ASSERT(tp->t_inpcb);
3101 * Process the segment text, merging it into the TCP sequencing queue,
3102 * and arranging for acknowledgment of receipt if necessary.
3103 * This process logically involves adjusting tp->rcv_wnd as data
3104 * is presented to the user (this happens in tcp_usrreq.c,
3105 * case PRU_RCVD). If a FIN has already been received on this
3106 * connection then we just ignore the text.
3108 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3109 IS_FASTOPEN(tp->t_flags));
3110 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3111 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3112 tcp_seq save_start = th->th_seq;
3113 tcp_seq save_rnxt = tp->rcv_nxt;
3114 int save_tlen = tlen;
3115 m_adj(m, drop_hdrlen); /* delayed header drop */
3117 * Insert segment which includes th into TCP reassembly queue
3118 * with control block tp. Set thflags to whether reassembly now
3119 * includes a segment with FIN. This handles the common case
3120 * inline (segment is the next to be received on an established
3121 * connection, and the queue is empty), avoiding linkage into
3122 * and removal from the queue and repetition of various
3124 * Set DELACK for segments received in order, but ack
3125 * immediately when segments are out of order (so
3126 * fast retransmit can work).
3128 if (th->th_seq == tp->rcv_nxt &&
3130 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3132 if (DELAY_ACK(tp, tlen) || tfo_syn)
3133 tp->t_flags |= TF_DELACK;
3135 tp->t_flags |= TF_ACKNOW;
3136 tp->rcv_nxt += tlen;
3138 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3139 (tp->t_fbyte_in == 0)) {
3140 tp->t_fbyte_in = ticks;
3141 if (tp->t_fbyte_in == 0)
3143 if (tp->t_fbyte_out && tp->t_fbyte_in)
3144 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3146 thflags = th->th_flags & TH_FIN;
3147 TCPSTAT_INC(tcps_rcvpack);
3148 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3149 SOCKBUF_LOCK(&so->so_rcv);
3150 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3153 sbappendstream_locked(&so->so_rcv, m, 0);
3154 SOCKBUF_UNLOCK(&so->so_rcv);
3155 tp->t_flags |= TF_WAKESOR;
3158 * XXX: Due to the header drop above "th" is
3159 * theoretically invalid by now. Fortunately
3160 * m_adj() doesn't actually frees any mbufs
3161 * when trimming from the head.
3163 tcp_seq temp = save_start;
3164 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3165 tp->t_flags |= TF_ACKNOW;
3167 if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) {
3168 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3170 * DSACK actually handled in the fastpath
3173 tcp_update_sack_list(tp, save_start,
3174 save_start + save_tlen);
3175 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3176 if ((tp->rcv_numsacks >= 1) &&
3177 (tp->sackblks[0].end == save_start)) {
3179 * Partial overlap, recorded at todrop
3182 tcp_update_sack_list(tp,
3183 tp->sackblks[0].start,
3184 tp->sackblks[0].end);
3186 tcp_update_dsack_list(tp, save_start,
3187 save_start + save_tlen);
3189 } else if (tlen >= save_tlen) {
3190 /* Update of sackblks. */
3191 tcp_update_dsack_list(tp, save_start,
3192 save_start + save_tlen);
3193 } else if (tlen > 0) {
3194 tcp_update_dsack_list(tp, save_start,
3200 * Note the amount of data that peer has sent into
3201 * our window, in order to estimate the sender's
3205 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3206 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3208 len = so->so_rcv.sb_hiwat;
3216 * If FIN is received ACK the FIN and let the user know
3217 * that the connection is closing.
3219 if (thflags & TH_FIN) {
3220 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3222 /* The socket upcall is handled by socantrcvmore. */
3223 tp->t_flags &= ~TF_WAKESOR;
3225 * If connection is half-synchronized
3226 * (ie NEEDSYN flag on) then delay ACK,
3227 * so it may be piggybacked when SYN is sent.
3228 * Otherwise, since we received a FIN then no
3229 * more input can be expected, send ACK now.
3231 if (tp->t_flags & TF_NEEDSYN)
3232 tp->t_flags |= TF_DELACK;
3234 tp->t_flags |= TF_ACKNOW;
3237 switch (tp->t_state) {
3239 * In SYN_RECEIVED and ESTABLISHED STATES
3240 * enter the CLOSE_WAIT state.
3242 case TCPS_SYN_RECEIVED:
3243 tp->t_starttime = ticks;
3245 case TCPS_ESTABLISHED:
3246 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3250 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3251 * enter the CLOSING state.
3253 case TCPS_FIN_WAIT_1:
3254 tcp_state_change(tp, TCPS_CLOSING);
3258 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3259 * starting the time-wait timer, turning off the other
3262 case TCPS_FIN_WAIT_2:
3268 if (so->so_options & SO_DEBUG)
3269 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3272 TCP_PROBE3(debug__input, tp, th, m);
3275 * Return any desired output.
3277 if (needoutput || (tp->t_flags & TF_ACKNOW))
3278 (void) tp->t_fb->tfb_tcp_output(tp);
3281 INP_WLOCK_ASSERT(tp->t_inpcb);
3283 if (tp->t_flags & TF_DELACK) {
3284 tp->t_flags &= ~TF_DELACK;
3285 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3287 tcp_handle_wakeup(tp, so);
3288 INP_WUNLOCK(tp->t_inpcb);
3293 * Generate an ACK dropping incoming segment if it occupies
3294 * sequence space, where the ACK reflects our state.
3296 * We can now skip the test for the RST flag since all
3297 * paths to this code happen after packets containing
3298 * RST have been dropped.
3300 * In the SYN-RECEIVED state, don't send an ACK unless the
3301 * segment we received passes the SYN-RECEIVED ACK test.
3302 * If it fails send a RST. This breaks the loop in the
3303 * "LAND" DoS attack, and also prevents an ACK storm
3304 * between two listening ports that have been sent forged
3305 * SYN segments, each with the source address of the other.
3307 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3308 (SEQ_GT(tp->snd_una, th->th_ack) ||
3309 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3310 rstreason = BANDLIM_RST_OPENPORT;
3314 if (so->so_options & SO_DEBUG)
3315 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3318 TCP_PROBE3(debug__input, tp, th, m);
3319 tp->t_flags |= TF_ACKNOW;
3320 (void) tp->t_fb->tfb_tcp_output(tp);
3321 tcp_handle_wakeup(tp, so);
3322 INP_WUNLOCK(tp->t_inpcb);
3328 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3329 tcp_handle_wakeup(tp, so);
3330 INP_WUNLOCK(tp->t_inpcb);
3332 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3337 * Drop space held by incoming segment and return.
3340 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3341 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3344 TCP_PROBE3(debug__input, tp, th, m);
3346 tcp_handle_wakeup(tp, so);
3347 INP_WUNLOCK(tp->t_inpcb);
3353 * Issue RST and make ACK acceptable to originator of segment.
3354 * The mbuf must still include the original packet header.
3358 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3359 int tlen, int rstreason)
3365 struct ip6_hdr *ip6;
3369 INP_WLOCK_ASSERT(tp->t_inpcb);
3372 /* Don't bother if destination was broadcast/multicast. */
3373 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3376 if (mtod(m, struct ip *)->ip_v == 6) {
3377 ip6 = mtod(m, struct ip6_hdr *);
3378 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3379 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3381 /* IPv6 anycast check is done at tcp6_input() */
3384 #if defined(INET) && defined(INET6)
3389 ip = mtod(m, struct ip *);
3390 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3391 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3392 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3393 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3398 /* Perform bandwidth limiting. */
3399 if (badport_bandlim(rstreason) < 0)
3402 /* tcp_respond consumes the mbuf chain. */
3403 if (th->th_flags & TH_ACK) {
3404 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3405 th->th_ack, TH_RST);
3407 if (th->th_flags & TH_SYN)
3409 if (th->th_flags & TH_FIN)
3411 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3412 (tcp_seq)0, TH_RST|TH_ACK);
3420 * Parse TCP options and place in tcpopt.
3423 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3428 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3430 if (opt == TCPOPT_EOL)
3432 if (opt == TCPOPT_NOP)
3438 if (optlen < 2 || optlen > cnt)
3443 if (optlen != TCPOLEN_MAXSEG)
3445 if (!(flags & TO_SYN))
3447 to->to_flags |= TOF_MSS;
3448 bcopy((char *)cp + 2,
3449 (char *)&to->to_mss, sizeof(to->to_mss));
3450 to->to_mss = ntohs(to->to_mss);
3453 if (optlen != TCPOLEN_WINDOW)
3455 if (!(flags & TO_SYN))
3457 to->to_flags |= TOF_SCALE;
3458 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3460 case TCPOPT_TIMESTAMP:
3461 if (optlen != TCPOLEN_TIMESTAMP)
3463 to->to_flags |= TOF_TS;
3464 bcopy((char *)cp + 2,
3465 (char *)&to->to_tsval, sizeof(to->to_tsval));
3466 to->to_tsval = ntohl(to->to_tsval);
3467 bcopy((char *)cp + 6,
3468 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3469 to->to_tsecr = ntohl(to->to_tsecr);
3471 case TCPOPT_SIGNATURE:
3473 * In order to reply to a host which has set the
3474 * TCP_SIGNATURE option in its initial SYN, we have
3475 * to record the fact that the option was observed
3476 * here for the syncache code to perform the correct
3479 if (optlen != TCPOLEN_SIGNATURE)
3481 to->to_flags |= TOF_SIGNATURE;
3482 to->to_signature = cp + 2;
3484 case TCPOPT_SACK_PERMITTED:
3485 if (optlen != TCPOLEN_SACK_PERMITTED)
3487 if (!(flags & TO_SYN))
3491 to->to_flags |= TOF_SACKPERM;
3494 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3498 to->to_flags |= TOF_SACK;
3499 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3500 to->to_sacks = cp + 2;
3501 TCPSTAT_INC(tcps_sack_rcv_blocks);
3503 case TCPOPT_FAST_OPEN:
3505 * Cookie length validation is performed by the
3506 * server side cookie checking code or the client
3507 * side cookie cache update code.
3509 if (!(flags & TO_SYN))
3511 if (!V_tcp_fastopen_client_enable &&
3512 !V_tcp_fastopen_server_enable)
3514 to->to_flags |= TOF_FASTOPEN;
3515 to->to_tfo_len = optlen - 2;
3516 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3525 * Pull out of band byte out of a segment so
3526 * it doesn't appear in the user's data queue.
3527 * It is still reflected in the segment length for
3528 * sequencing purposes.
3531 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3534 int cnt = off + th->th_urp - 1;
3537 if (m->m_len > cnt) {
3538 char *cp = mtod(m, caddr_t) + cnt;
3539 struct tcpcb *tp = sototcpcb(so);
3541 INP_WLOCK_ASSERT(tp->t_inpcb);
3544 tp->t_oobflags |= TCPOOB_HAVEDATA;
3545 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3547 if (m->m_flags & M_PKTHDR)
3556 panic("tcp_pulloutofband");
3560 * Collect new round-trip time estimate
3561 * and update averages and current timeout.
3564 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3568 INP_WLOCK_ASSERT(tp->t_inpcb);
3570 TCPSTAT_INC(tcps_rttupdated);
3573 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3574 imax(0, rtt * 1000 / hz));
3576 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3578 * srtt is stored as fixed point with 5 bits after the
3579 * binary point (i.e., scaled by 8). The following magic
3580 * is equivalent to the smoothing algorithm in rfc793 with
3581 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3582 * point). Adjust rtt to origin 0.
3584 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3585 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3587 if ((tp->t_srtt += delta) <= 0)
3591 * We accumulate a smoothed rtt variance (actually, a
3592 * smoothed mean difference), then set the retransmit
3593 * timer to smoothed rtt + 4 times the smoothed variance.
3594 * rttvar is stored as fixed point with 4 bits after the
3595 * binary point (scaled by 16). The following is
3596 * equivalent to rfc793 smoothing with an alpha of .75
3597 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3598 * rfc793's wired-in beta.
3602 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3603 if ((tp->t_rttvar += delta) <= 0)
3605 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3606 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3609 * No rtt measurement yet - use the unsmoothed rtt.
3610 * Set the variance to half the rtt (so our first
3611 * retransmit happens at 3*rtt).
3613 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3614 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3615 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3621 * the retransmit should happen at rtt + 4 * rttvar.
3622 * Because of the way we do the smoothing, srtt and rttvar
3623 * will each average +1/2 tick of bias. When we compute
3624 * the retransmit timer, we want 1/2 tick of rounding and
3625 * 1 extra tick because of +-1/2 tick uncertainty in the
3626 * firing of the timer. The bias will give us exactly the
3627 * 1.5 tick we need. But, because the bias is
3628 * statistical, we have to test that we don't drop below
3629 * the minimum feasible timer (which is 2 ticks).
3631 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3632 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3635 * We received an ack for a packet that wasn't retransmitted;
3636 * it is probably safe to discard any error indications we've
3637 * received recently. This isn't quite right, but close enough
3638 * for now (a route might have failed after we sent a segment,
3639 * and the return path might not be symmetrical).
3641 tp->t_softerror = 0;
3645 * Determine a reasonable value for maxseg size.
3646 * If the route is known, check route for mtu.
3647 * If none, use an mss that can be handled on the outgoing interface
3648 * without forcing IP to fragment. If no route is found, route has no mtu,
3649 * or the destination isn't local, use a default, hopefully conservative
3650 * size (usually 512 or the default IP max size, but no more than the mtu
3651 * of the interface), as we can't discover anything about intervening
3652 * gateways or networks. We also initialize the congestion/slow start
3653 * window to be a single segment if the destination isn't local.
3654 * While looking at the routing entry, we also initialize other path-dependent
3655 * parameters from pre-set or cached values in the routing entry.
3657 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3658 * IP options, e.g. IPSEC data, since length of this data may vary, and
3659 * thus it is calculated for every segment separately in tcp_output().
3661 * NOTE that this routine is only called when we process an incoming
3662 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3663 * settings are handled in tcp_mssopt().
3666 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3667 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3670 uint32_t maxmtu = 0;
3671 struct inpcb *inp = tp->t_inpcb;
3672 struct hc_metrics_lite metrics;
3674 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3675 size_t min_protoh = isipv6 ?
3676 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3677 sizeof (struct tcpiphdr);
3679 const size_t min_protoh = sizeof(struct tcpiphdr);
3682 INP_WLOCK_ASSERT(tp->t_inpcb);
3684 if (mtuoffer != -1) {
3685 KASSERT(offer == -1, ("%s: conflict", __func__));
3686 offer = mtuoffer - min_protoh;
3692 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3693 tp->t_maxseg = V_tcp_v6mssdflt;
3696 #if defined(INET) && defined(INET6)
3701 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3702 tp->t_maxseg = V_tcp_mssdflt;
3707 * No route to sender, stay with default mss and return.
3711 * In case we return early we need to initialize metrics
3712 * to a defined state as tcp_hc_get() would do for us
3713 * if there was no cache hit.
3715 if (metricptr != NULL)
3716 bzero(metricptr, sizeof(struct hc_metrics_lite));
3720 /* What have we got? */
3724 * Offer == 0 means that there was no MSS on the SYN
3725 * segment, in this case we use tcp_mssdflt as
3726 * already assigned to t_maxseg above.
3728 offer = tp->t_maxseg;
3733 * Offer == -1 means that we didn't receive SYN yet.
3739 * Prevent DoS attack with too small MSS. Round up
3740 * to at least minmss.
3742 offer = max(offer, V_tcp_minmss);
3746 * rmx information is now retrieved from tcp_hostcache.
3748 tcp_hc_get(&inp->inp_inc, &metrics);
3749 if (metricptr != NULL)
3750 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3753 * If there's a discovered mtu in tcp hostcache, use it.
3754 * Else, use the link mtu.
3756 if (metrics.rmx_mtu)
3757 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3761 mss = maxmtu - min_protoh;
3762 if (!V_path_mtu_discovery &&
3763 !in6_localaddr(&inp->in6p_faddr))
3764 mss = min(mss, V_tcp_v6mssdflt);
3767 #if defined(INET) && defined(INET6)
3772 mss = maxmtu - min_protoh;
3773 if (!V_path_mtu_discovery &&
3774 !in_localaddr(inp->inp_faddr))
3775 mss = min(mss, V_tcp_mssdflt);
3779 * XXX - The above conditional (mss = maxmtu - min_protoh)
3780 * probably violates the TCP spec.
3781 * The problem is that, since we don't know the
3782 * other end's MSS, we are supposed to use a conservative
3783 * default. But, if we do that, then MTU discovery will
3784 * never actually take place, because the conservative
3785 * default is much less than the MTUs typically seen
3786 * on the Internet today. For the moment, we'll sweep
3787 * this under the carpet.
3789 * The conservative default might not actually be a problem
3790 * if the only case this occurs is when sending an initial
3791 * SYN with options and data to a host we've never talked
3792 * to before. Then, they will reply with an MSS value which
3793 * will get recorded and the new parameters should get
3794 * recomputed. For Further Study.
3797 mss = min(mss, offer);
3800 * Sanity check: make sure that maxseg will be large
3801 * enough to allow some data on segments even if the
3802 * all the option space is used (40bytes). Otherwise
3803 * funny things may happen in tcp_output.
3805 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3813 tcp_mss(struct tcpcb *tp, int offer)
3819 struct hc_metrics_lite metrics;
3820 struct tcp_ifcap cap;
3822 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3824 bzero(&cap, sizeof(cap));
3825 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3831 * If there's a pipesize, change the socket buffer to that size,
3832 * don't change if sb_hiwat is different than default (then it
3833 * has been changed on purpose with setsockopt).
3834 * Make the socket buffers an integral number of mss units;
3835 * if the mss is larger than the socket buffer, decrease the mss.
3837 so = inp->inp_socket;
3838 SOCKBUF_LOCK(&so->so_snd);
3839 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3840 bufsize = metrics.rmx_sendpipe;
3842 bufsize = so->so_snd.sb_hiwat;
3846 bufsize = roundup(bufsize, mss);
3847 if (bufsize > sb_max)
3849 if (bufsize > so->so_snd.sb_hiwat)
3850 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3852 SOCKBUF_UNLOCK(&so->so_snd);
3854 * Sanity check: make sure that maxseg will be large
3855 * enough to allow some data on segments even if the
3856 * all the option space is used (40bytes). Otherwise
3857 * funny things may happen in tcp_output.
3859 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3861 tp->t_maxseg = max(mss, 64);
3863 SOCKBUF_LOCK(&so->so_rcv);
3864 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3865 bufsize = metrics.rmx_recvpipe;
3867 bufsize = so->so_rcv.sb_hiwat;
3868 if (bufsize > mss) {
3869 bufsize = roundup(bufsize, mss);
3870 if (bufsize > sb_max)
3872 if (bufsize > so->so_rcv.sb_hiwat)
3873 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3875 SOCKBUF_UNLOCK(&so->so_rcv);
3877 /* Check the interface for TSO capabilities. */
3878 if (cap.ifcap & CSUM_TSO) {
3879 tp->t_flags |= TF_TSO;
3880 tp->t_tsomax = cap.tsomax;
3881 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3882 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3887 * Determine the MSS option to send on an outgoing SYN.
3890 tcp_mssopt(struct in_conninfo *inc)
3893 uint32_t thcmtu = 0;
3894 uint32_t maxmtu = 0;
3897 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3900 if (inc->inc_flags & INC_ISIPV6) {
3901 mss = V_tcp_v6mssdflt;
3902 maxmtu = tcp_maxmtu6(inc, NULL);
3903 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3906 #if defined(INET) && defined(INET6)
3911 mss = V_tcp_mssdflt;
3912 maxmtu = tcp_maxmtu(inc, NULL);
3913 min_protoh = sizeof(struct tcpiphdr);
3916 #if defined(INET6) || defined(INET)
3917 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3920 if (maxmtu && thcmtu)
3921 mss = min(maxmtu, thcmtu) - min_protoh;
3922 else if (maxmtu || thcmtu)
3923 mss = max(maxmtu, thcmtu) - min_protoh;
3929 tcp_prr_partialack(struct tcpcb *tp, struct tcphdr *th)
3931 long snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3932 int maxseg = tcp_maxseg(tp);
3934 INP_WLOCK_ASSERT(tp->t_inpcb);
3936 tcp_timer_activate(tp, TT_REXMT, 0);
3939 * Compute the amount of data that this ACK is indicating
3940 * (del_data) and an estimate of how many bytes are in the
3943 if (SEQ_GEQ(th->th_ack, tp->snd_una))
3944 del_data = BYTES_THIS_ACK(tp, th);
3945 del_data += tp->sackhint.delivered_data;
3946 pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit;
3947 tp->sackhint.prr_delivered += del_data;
3949 * Proportional Rate Reduction
3951 if (pipe > tp->snd_ssthresh)
3952 snd_cnt = (tp->sackhint.prr_delivered * tp->snd_ssthresh /
3953 tp->sackhint.recover_fs) - tp->sackhint.sack_bytes_rexmit;
3955 if (V_tcp_do_prr_conservative)
3956 limit = tp->sackhint.prr_delivered -
3957 tp->sackhint.sack_bytes_rexmit;
3959 if ((tp->sackhint.prr_delivered -
3960 tp->sackhint.sack_bytes_rexmit) > del_data)
3961 limit = tp->sackhint.prr_delivered -
3962 tp->sackhint.sack_bytes_rexmit + maxseg;
3964 limit = del_data + maxseg;
3965 snd_cnt = min((tp->snd_ssthresh - pipe), limit);
3967 snd_cnt = max((snd_cnt / maxseg), 0);
3969 * Send snd_cnt new data into the network in response to this ack.
3970 * If there is going to be a SACK retransmission, adjust snd_cwnd
3973 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
3974 tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg);
3975 tp->t_flags |= TF_ACKNOW;
3976 (void) tcp_output(tp);
3980 * On a partial ack arrives, force the retransmission of the
3981 * next unacknowledged segment. Do not clear tp->t_dupacks.
3982 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3986 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3988 tcp_seq onxt = tp->snd_nxt;
3989 uint32_t ocwnd = tp->snd_cwnd;
3990 u_int maxseg = tcp_maxseg(tp);
3992 INP_WLOCK_ASSERT(tp->t_inpcb);
3994 tcp_timer_activate(tp, TT_REXMT, 0);
3996 tp->snd_nxt = th->th_ack;
3998 * Set snd_cwnd to one segment beyond acknowledged offset.
3999 * (tp->snd_una has not yet been updated when this function is called.)
4001 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4002 tp->t_flags |= TF_ACKNOW;
4003 (void) tp->t_fb->tfb_tcp_output(tp);
4004 tp->snd_cwnd = ocwnd;
4005 if (SEQ_GT(onxt, tp->snd_nxt))
4008 * Partial window deflation. Relies on fact that tp->snd_una
4011 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4012 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4015 tp->snd_cwnd += maxseg;
4019 tcp_compute_pipe(struct tcpcb *tp)
4021 return (tp->snd_max - tp->snd_una +
4022 tp->sackhint.sack_bytes_rexmit -
4023 tp->sackhint.sacked_bytes);
4027 tcp_compute_initwnd(uint32_t maxseg)
4030 * Calculate the Initial Window, also used as Restart Window
4032 * RFC5681 Section 3.1 specifies the default conservative values.
4033 * RFC3390 specifies slightly more aggressive values.
4034 * RFC6928 increases it to ten segments.
4035 * Support for user specified value for initial flight size.
4037 if (V_tcp_initcwnd_segments)
4038 return min(V_tcp_initcwnd_segments * maxseg,
4039 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4040 else if (V_tcp_do_rfc3390)
4041 return min(4 * maxseg, max(2 * maxseg, 4380));
4043 /* Per RFC5681 Section 3.1 */
4045 return (2 * maxseg);
4046 else if (maxseg > 1095)
4047 return (3 * maxseg);
4049 return (4 * maxseg);