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.prr_out = 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 */, false);
597 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
598 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
599 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
601 return (IPPROTO_DONE);
605 return (tcp_input(mp, offp, proto));
610 tcp_input(struct mbuf **mp, int *offp, int proto)
612 struct mbuf *m = *mp;
613 struct tcphdr *th = NULL;
614 struct ip *ip = NULL;
615 struct inpcb *inp = NULL;
616 struct tcpcb *tp = NULL;
617 struct socket *so = NULL;
628 int rstreason = 0; /* For badport_bandlim accounting purposes */
630 struct m_tag *fwd_tag = NULL;
632 struct ip6_hdr *ip6 = NULL;
635 const void *ip6 = NULL;
637 struct tcpopt to; /* options in this segment */
638 char *s = NULL; /* address and port logging */
641 * The size of tcp_saveipgen must be the size of the max ip header,
644 u_char tcp_saveipgen[IP6_HDR_LEN];
645 struct tcphdr tcp_savetcp;
652 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
659 TCPSTAT_INC(tcps_rcvtotal);
663 ip6 = mtod(m, struct ip6_hdr *);
664 th = (struct tcphdr *)((caddr_t)ip6 + off0);
665 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
666 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
667 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
668 th->th_sum = m->m_pkthdr.csum_data;
670 th->th_sum = in6_cksum_pseudo(ip6, tlen,
671 IPPROTO_TCP, m->m_pkthdr.csum_data);
672 th->th_sum ^= 0xffff;
674 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
676 TCPSTAT_INC(tcps_rcvbadsum);
681 * Be proactive about unspecified IPv6 address in source.
682 * As we use all-zero to indicate unbounded/unconnected pcb,
683 * unspecified IPv6 address can be used to confuse us.
685 * Note that packets with unspecified IPv6 destination is
686 * already dropped in ip6_input.
688 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
692 iptos = IPV6_TRAFFIC_CLASS(ip6);
695 #if defined(INET) && defined(INET6)
701 * Get IP and TCP header together in first mbuf.
702 * Note: IP leaves IP header in first mbuf.
704 if (off0 > sizeof (struct ip)) {
706 off0 = sizeof(struct ip);
708 if (m->m_len < sizeof (struct tcpiphdr)) {
709 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
711 TCPSTAT_INC(tcps_rcvshort);
712 return (IPPROTO_DONE);
715 ip = mtod(m, struct ip *);
716 th = (struct tcphdr *)((caddr_t)ip + off0);
717 tlen = ntohs(ip->ip_len) - off0;
720 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
721 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
722 th->th_sum = m->m_pkthdr.csum_data;
724 th->th_sum = in_pseudo(ip->ip_src.s_addr,
726 htonl(m->m_pkthdr.csum_data + tlen +
728 th->th_sum ^= 0xffff;
730 struct ipovly *ipov = (struct ipovly *)ip;
733 * Checksum extended TCP header and data.
737 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
738 ipov->ih_len = htons(tlen);
739 th->th_sum = in_cksum(m, len);
740 /* Reset length for SDT probes. */
741 ip->ip_len = htons(len);
744 /* Re-initialization for later version check */
746 ip->ip_v = IPVERSION;
747 ip->ip_hl = off0 >> 2;
751 TCPSTAT_INC(tcps_rcvbadsum);
758 * Check that TCP offset makes sense,
759 * pull out TCP options and adjust length. XXX
761 off = th->th_off << 2;
762 if (off < sizeof (struct tcphdr) || off > tlen) {
763 TCPSTAT_INC(tcps_rcvbadoff);
766 tlen -= off; /* tlen is used instead of ti->ti_len */
767 if (off > sizeof (struct tcphdr)) {
770 if (m->m_len < off0 + off) {
771 m = m_pullup(m, off0 + off);
773 TCPSTAT_INC(tcps_rcvshort);
774 return (IPPROTO_DONE);
777 ip6 = mtod(m, struct ip6_hdr *);
778 th = (struct tcphdr *)((caddr_t)ip6 + off0);
781 #if defined(INET) && defined(INET6)
786 if (m->m_len < sizeof(struct ip) + off) {
787 if ((m = m_pullup(m, sizeof (struct ip) + off))
789 TCPSTAT_INC(tcps_rcvshort);
790 return (IPPROTO_DONE);
792 ip = mtod(m, struct ip *);
793 th = (struct tcphdr *)((caddr_t)ip + off0);
797 optlen = off - sizeof (struct tcphdr);
798 optp = (u_char *)(th + 1);
800 thflags = th->th_flags;
803 * Convert TCP protocol specific fields to host format.
805 tcp_fields_to_host(th);
808 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
810 drop_hdrlen = off0 + off;
813 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
817 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
819 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
822 #if defined(INET) && !defined(INET6)
823 (m->m_flags & M_IP_NEXTHOP)
826 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
830 if (isipv6 && fwd_tag != NULL) {
831 struct sockaddr_in6 *next_hop6;
833 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
835 * Transparently forwarded. Pretend to be the destination.
836 * Already got one like this?
838 inp = in6_pcblookup_mbuf(&V_tcbinfo,
839 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
840 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
843 * It's new. Try to find the ambushing socket.
844 * Because we've rewritten the destination address,
845 * any hardware-generated hash is ignored.
847 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
848 th->th_sport, &next_hop6->sin6_addr,
849 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
850 th->th_dport, INPLOOKUP_WILDCARD |
851 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
854 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
855 th->th_sport, &ip6->ip6_dst, th->th_dport,
856 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
857 m->m_pkthdr.rcvif, m);
860 #if defined(INET6) && defined(INET)
864 if (fwd_tag != NULL) {
865 struct sockaddr_in *next_hop;
867 next_hop = (struct sockaddr_in *)(fwd_tag+1);
869 * Transparently forwarded. Pretend to be the destination.
870 * already got one like this?
872 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
873 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
874 m->m_pkthdr.rcvif, m);
877 * It's new. Try to find the ambushing socket.
878 * Because we've rewritten the destination address,
879 * any hardware-generated hash is ignored.
881 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
882 th->th_sport, next_hop->sin_addr,
883 next_hop->sin_port ? ntohs(next_hop->sin_port) :
884 th->th_dport, INPLOOKUP_WILDCARD |
885 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
888 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
889 th->th_sport, ip->ip_dst, th->th_dport,
890 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
891 m->m_pkthdr.rcvif, m);
895 * If the INPCB does not exist then all data in the incoming
896 * segment is discarded and an appropriate RST is sent back.
897 * XXX MRT Send RST using which routing table?
901 * Log communication attempts to ports that are not
904 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
905 V_tcp_log_in_vain == 2) {
906 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
907 log(LOG_INFO, "%s; %s: Connection attempt "
908 "to closed port\n", s, __func__);
911 * When blackholing do not respond with a RST but
912 * completely ignore the segment and drop it.
914 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
918 rstreason = BANDLIM_RST_CLOSEDPORT;
921 INP_WLOCK_ASSERT(inp);
923 * While waiting for inp lock during the lookup, another thread
924 * can have dropped the inpcb, in which case we need to loop back
925 * and try to find a new inpcb to deliver to.
927 if (inp->inp_flags & INP_DROPPED) {
932 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
933 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
934 ((inp->inp_socket == NULL) ||
935 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
936 inp->inp_flowid = m->m_pkthdr.flowid;
937 inp->inp_flowtype = M_HASHTYPE_GET(m);
939 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
941 if (isipv6 && IPSEC_ENABLED(ipv6) &&
942 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
950 if (IPSEC_ENABLED(ipv4) &&
951 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
958 * Check the minimum TTL for socket.
960 if (inp->inp_ip_minttl != 0) {
963 if (inp->inp_ip_minttl > ip6->ip6_hlim)
967 if (inp->inp_ip_minttl > ip->ip_ttl)
972 * A previous connection in TIMEWAIT state is supposed to catch stray
973 * or duplicate segments arriving late. If this segment was a
974 * legitimate new connection attempt, the old INPCB gets removed and
975 * we can try again to find a listening socket.
977 * At this point, due to earlier optimism, we may hold only an inpcb
978 * lock, and not the inpcbinfo write lock. If so, we need to try to
979 * acquire it, or if that fails, acquire a reference on the inpcb,
980 * drop all locks, acquire a global write lock, and then re-acquire
981 * the inpcb lock. We may at that point discover that another thread
982 * has tried to free the inpcb, in which case we need to loop back
983 * and try to find a new inpcb to deliver to.
985 * XXXRW: It may be time to rethink timewait locking.
987 if (inp->inp_flags & INP_TIMEWAIT) {
988 tcp_dooptions(&to, optp, optlen,
989 (thflags & TH_SYN) ? TO_SYN : 0);
991 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
993 if (tcp_twcheck(inp, &to, th, m, tlen))
995 return (IPPROTO_DONE);
998 * The TCPCB may no longer exist if the connection is winding
999 * down or it is in the CLOSED state. Either way we drop the
1000 * segment and send an appropriate response.
1002 tp = intotcpcb(inp);
1003 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1004 rstreason = BANDLIM_RST_CLOSEDPORT;
1009 if (tp->t_flags & TF_TOE) {
1010 tcp_offload_input(tp, m);
1011 m = NULL; /* consumed by the TOE driver */
1017 INP_WLOCK_ASSERT(inp);
1018 if (mac_inpcb_check_deliver(inp, m))
1021 so = inp->inp_socket;
1022 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1024 if (so->so_options & SO_DEBUG) {
1025 ostate = tp->t_state;
1028 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1031 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1034 #endif /* TCPDEBUG */
1036 * When the socket is accepting connections (the INPCB is in LISTEN
1037 * state) we look into the SYN cache if this is a new connection
1038 * attempt or the completion of a previous one.
1040 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1041 ("%s: so accepting but tp %p not listening", __func__, tp));
1042 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1043 struct in_conninfo inc;
1045 bzero(&inc, sizeof(inc));
1048 inc.inc_flags |= INC_ISIPV6;
1049 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1050 inc.inc_flags |= INC_IPV6MINMTU;
1051 inc.inc6_faddr = ip6->ip6_src;
1052 inc.inc6_laddr = ip6->ip6_dst;
1056 inc.inc_faddr = ip->ip_src;
1057 inc.inc_laddr = ip->ip_dst;
1059 inc.inc_fport = th->th_sport;
1060 inc.inc_lport = th->th_dport;
1061 inc.inc_fibnum = so->so_fibnum;
1064 * Check for an existing connection attempt in syncache if
1065 * the flag is only ACK. A successful lookup creates a new
1066 * socket appended to the listen queue in SYN_RECEIVED state.
1068 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1070 * Parse the TCP options here because
1071 * syncookies need access to the reflected
1074 tcp_dooptions(&to, optp, optlen, 0);
1076 * NB: syncache_expand() doesn't unlock
1077 * inp and tcpinfo locks.
1079 rstreason = syncache_expand(&inc, &to, th, &so, m);
1080 if (rstreason < 0) {
1082 * A failing TCP MD5 signature comparison
1083 * must result in the segment being dropped
1084 * and must not produce any response back
1088 } else if (rstreason == 0) {
1090 * No syncache entry or ACK was not
1091 * for our SYN/ACK. Send a RST.
1092 * NB: syncache did its own logging
1093 * of the failure cause.
1095 rstreason = BANDLIM_RST_OPENPORT;
1101 * We completed the 3-way handshake
1102 * but could not allocate a socket
1103 * either due to memory shortage,
1104 * listen queue length limits or
1105 * global socket limits. Send RST
1106 * or wait and have the remote end
1107 * retransmit the ACK for another
1110 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1111 log(LOG_DEBUG, "%s; %s: Listen socket: "
1112 "Socket allocation failed due to "
1113 "limits or memory shortage, %s\n",
1115 V_tcp_sc_rst_sock_fail ?
1116 "sending RST" : "try again");
1117 if (V_tcp_sc_rst_sock_fail) {
1118 rstreason = BANDLIM_UNLIMITED;
1124 * Socket is created in state SYN_RECEIVED.
1125 * Unlock the listen socket, lock the newly
1126 * created socket and update the tp variable.
1128 INP_WUNLOCK(inp); /* listen socket */
1129 inp = sotoinpcb(so);
1131 * New connection inpcb is already locked by
1132 * syncache_expand().
1134 INP_WLOCK_ASSERT(inp);
1135 tp = intotcpcb(inp);
1136 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1137 ("%s: ", __func__));
1139 * Process the segment and the data it
1140 * contains. tcp_do_segment() consumes
1141 * the mbuf chain and unlocks the inpcb.
1143 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1144 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1146 return (IPPROTO_DONE);
1149 * Segment flag validation for new connection attempts:
1151 * Our (SYN|ACK) response was rejected.
1152 * Check with syncache and remove entry to prevent
1155 * NB: syncache_chkrst does its own logging of failure
1158 if (thflags & TH_RST) {
1159 syncache_chkrst(&inc, th, m);
1163 * We can't do anything without SYN.
1165 if ((thflags & TH_SYN) == 0) {
1166 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1167 log(LOG_DEBUG, "%s; %s: Listen socket: "
1168 "SYN is missing, segment ignored\n",
1170 TCPSTAT_INC(tcps_badsyn);
1174 * (SYN|ACK) is bogus on a listen socket.
1176 if (thflags & TH_ACK) {
1177 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1178 log(LOG_DEBUG, "%s; %s: Listen socket: "
1179 "SYN|ACK invalid, segment rejected\n",
1181 syncache_badack(&inc); /* XXX: Not needed! */
1182 TCPSTAT_INC(tcps_badsyn);
1183 rstreason = BANDLIM_RST_OPENPORT;
1187 * If the drop_synfin option is enabled, drop all
1188 * segments with both the SYN and FIN bits set.
1189 * This prevents e.g. nmap from identifying the
1191 * XXX: Poor reasoning. nmap has other methods
1192 * and is constantly refining its stack detection
1194 * XXX: This is a violation of the TCP specification
1195 * and was used by RFC1644.
1197 if ((thflags & TH_FIN) && V_drop_synfin) {
1198 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1199 log(LOG_DEBUG, "%s; %s: Listen socket: "
1200 "SYN|FIN segment ignored (based on "
1201 "sysctl setting)\n", s, __func__);
1202 TCPSTAT_INC(tcps_badsyn);
1206 * Segment's flags are (SYN) or (SYN|FIN).
1208 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1209 * as they do not affect the state of the TCP FSM.
1210 * The data pointed to by TH_URG and th_urp is ignored.
1212 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1213 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1214 KASSERT(thflags & (TH_SYN),
1215 ("%s: Listen socket: TH_SYN not set", __func__));
1218 * If deprecated address is forbidden,
1219 * we do not accept SYN to deprecated interface
1220 * address to prevent any new inbound connection from
1221 * getting established.
1222 * When we do not accept SYN, we send a TCP RST,
1223 * with deprecated source address (instead of dropping
1224 * it). We compromise it as it is much better for peer
1225 * to send a RST, and RST will be the final packet
1228 * If we do not forbid deprecated addresses, we accept
1229 * the SYN packet. RFC2462 does not suggest dropping
1231 * If we decipher RFC2462 5.5.4, it says like this:
1232 * 1. use of deprecated addr with existing
1233 * communication is okay - "SHOULD continue to be
1235 * 2. use of it with new communication:
1236 * (2a) "SHOULD NOT be used if alternate address
1237 * with sufficient scope is available"
1238 * (2b) nothing mentioned otherwise.
1239 * Here we fall into (2b) case as we have no choice in
1240 * our source address selection - we must obey the peer.
1242 * The wording in RFC2462 is confusing, and there are
1243 * multiple description text for deprecated address
1244 * handling - worse, they are not exactly the same.
1245 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1247 if (isipv6 && !V_ip6_use_deprecated) {
1248 struct in6_ifaddr *ia6;
1250 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1252 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1253 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1254 log(LOG_DEBUG, "%s; %s: Listen socket: "
1255 "Connection attempt to deprecated "
1256 "IPv6 address rejected\n",
1258 rstreason = BANDLIM_RST_OPENPORT;
1264 * Basic sanity checks on incoming SYN requests:
1265 * Don't respond if the destination is a link layer
1266 * broadcast according to RFC1122 4.2.3.10, p. 104.
1267 * If it is from this socket it must be forged.
1268 * Don't respond if the source or destination is a
1269 * global or subnet broad- or multicast address.
1270 * Note that it is quite possible to receive unicast
1271 * link-layer packets with a broadcast IP address. Use
1272 * in_broadcast() to find them.
1274 if (m->m_flags & (M_BCAST|M_MCAST)) {
1275 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1276 log(LOG_DEBUG, "%s; %s: Listen socket: "
1277 "Connection attempt from broad- or multicast "
1278 "link layer address ignored\n", s, __func__);
1283 if (th->th_dport == th->th_sport &&
1284 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1285 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1286 log(LOG_DEBUG, "%s; %s: Listen socket: "
1287 "Connection attempt to/from self "
1288 "ignored\n", s, __func__);
1291 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1292 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1293 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1294 log(LOG_DEBUG, "%s; %s: Listen socket: "
1295 "Connection attempt from/to multicast "
1296 "address ignored\n", s, __func__);
1301 #if defined(INET) && defined(INET6)
1306 if (th->th_dport == th->th_sport &&
1307 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1308 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1309 log(LOG_DEBUG, "%s; %s: Listen socket: "
1310 "Connection attempt from/to self "
1311 "ignored\n", s, __func__);
1314 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1315 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1316 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1317 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1318 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1319 log(LOG_DEBUG, "%s; %s: Listen socket: "
1320 "Connection attempt from/to broad- "
1321 "or multicast address ignored\n",
1328 * SYN appears to be valid. Create compressed TCP state
1332 if (so->so_options & SO_DEBUG)
1333 tcp_trace(TA_INPUT, ostate, tp,
1334 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1336 TCP_PROBE3(debug__input, tp, th, m);
1337 tcp_dooptions(&to, optp, optlen, TO_SYN);
1338 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos))
1339 goto tfo_socket_result;
1342 * Entry added to syncache and mbuf consumed.
1343 * Only the listen socket is unlocked by syncache_add().
1345 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1346 return (IPPROTO_DONE);
1347 } else if (tp->t_state == TCPS_LISTEN) {
1349 * When a listen socket is torn down the SO_ACCEPTCONN
1350 * flag is removed first while connections are drained
1351 * from the accept queue in a unlock/lock cycle of the
1352 * ACCEPT_LOCK, opening a race condition allowing a SYN
1353 * attempt go through unhandled.
1357 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1358 if (tp->t_flags & TF_SIGNATURE) {
1359 tcp_dooptions(&to, optp, optlen, thflags);
1360 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1361 TCPSTAT_INC(tcps_sig_err_nosigopt);
1364 if (!TCPMD5_ENABLED() ||
1365 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1369 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1372 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1373 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1374 * the inpcb, and unlocks pcbinfo.
1376 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1377 return (IPPROTO_DONE);
1380 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1383 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1386 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1387 m = NULL; /* mbuf chain got consumed. */
1392 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1398 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1403 return (IPPROTO_DONE);
1407 * Automatic sizing of receive socket buffer. Often the send
1408 * buffer size is not optimally adjusted to the actual network
1409 * conditions at hand (delay bandwidth product). Setting the
1410 * buffer size too small limits throughput on links with high
1411 * bandwidth and high delay (eg. trans-continental/oceanic links).
1413 * On the receive side the socket buffer memory is only rarely
1414 * used to any significant extent. This allows us to be much
1415 * more aggressive in scaling the receive socket buffer. For
1416 * the case that the buffer space is actually used to a large
1417 * extent and we run out of kernel memory we can simply drop
1418 * the new segments; TCP on the sender will just retransmit it
1419 * later. Setting the buffer size too big may only consume too
1420 * much kernel memory if the application doesn't read() from
1421 * the socket or packet loss or reordering makes use of the
1424 * The criteria to step up the receive buffer one notch are:
1425 * 1. Application has not set receive buffer size with
1426 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1427 * 2. the number of bytes received during 1/2 of an sRTT
1428 * is at least 3/8 of the current socket buffer size.
1429 * 3. receive buffer size has not hit maximal automatic size;
1431 * If all of the criteria are met we increaset the socket buffer
1432 * by a 1/2 (bounded by the max). This allows us to keep ahead
1433 * of slow-start but also makes it so our peer never gets limited
1434 * by our rwnd which we then open up causing a burst.
1436 * This algorithm does two steps per RTT at most and only if
1437 * we receive a bulk stream w/o packet losses or reorderings.
1438 * Shrinking the buffer during idle times is not necessary as
1439 * it doesn't consume any memory when idle.
1441 * TODO: Only step up if the application is actually serving
1442 * the buffer to better manage the socket buffer resources.
1445 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1446 struct tcpcb *tp, int tlen)
1450 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1451 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1452 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1453 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1454 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1455 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1456 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1458 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1460 /* Start over with next RTT. */
1464 tp->rfbuf_cnt += tlen; /* add up */
1470 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1473 * Since tp might be gone if the session entered
1474 * the TIME_WAIT state before coming here, we need
1475 * to check if the socket is still connected.
1477 if ((so->so_state & SS_ISCONNECTED) == 0)
1479 INP_LOCK_ASSERT(tp->t_inpcb);
1480 if (tp->t_flags & TF_WAKESOR) {
1481 tp->t_flags &= ~TF_WAKESOR;
1482 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
1485 if (tp->t_flags & TF_WAKESOW) {
1486 tp->t_flags &= ~TF_WAKESOW;
1487 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
1493 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1494 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1496 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1497 int rstreason, todrop, win, incforsyn = 0;
1501 struct in_conninfo *inc;
1509 * The size of tcp_saveipgen must be the size of the max ip header,
1512 u_char tcp_saveipgen[IP6_HDR_LEN];
1513 struct tcphdr tcp_savetcp;
1516 thflags = th->th_flags;
1517 inc = &tp->t_inpcb->inp_inc;
1518 tp->sackhint.last_sack_ack = 0;
1520 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1523 INP_WLOCK_ASSERT(tp->t_inpcb);
1524 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1526 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1530 /* Save segment, if requested. */
1531 tcp_pcap_add(th, m, &(tp->t_inpkts));
1533 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1536 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1537 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1538 log(LOG_DEBUG, "%s; %s: "
1539 "SYN|FIN segment ignored (based on "
1540 "sysctl setting)\n", s, __func__);
1547 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1548 * check SEQ.ACK first.
1550 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1551 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1552 rstreason = BANDLIM_UNLIMITED;
1557 * Segment received on connection.
1558 * Reset idle time and keep-alive timer.
1559 * XXX: This should be done after segment
1560 * validation to ignore broken/spoofed segs.
1562 tp->t_rcvtime = ticks;
1565 * Scale up the window into a 32-bit value.
1566 * For the SYN_SENT state the scale is zero.
1568 tiwin = th->th_win << tp->snd_scale;
1570 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1574 * TCP ECN processing.
1576 if (tp->t_flags2 & TF2_ECN_PERMIT) {
1577 if (thflags & TH_CWR) {
1578 tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1579 tp->t_flags |= TF_ACKNOW;
1581 switch (iptos & IPTOS_ECN_MASK) {
1583 tp->t_flags2 |= TF2_ECN_SND_ECE;
1584 TCPSTAT_INC(tcps_ecn_ce);
1586 case IPTOS_ECN_ECT0:
1587 TCPSTAT_INC(tcps_ecn_ect0);
1589 case IPTOS_ECN_ECT1:
1590 TCPSTAT_INC(tcps_ecn_ect1);
1594 /* Process a packet differently from RFC3168. */
1595 cc_ecnpkt_handler(tp, th, iptos);
1597 /* Congestion experienced. */
1598 if (thflags & TH_ECE) {
1599 cc_cong_signal(tp, th, CC_ECN);
1604 * Parse options on any incoming segment.
1606 tcp_dooptions(&to, (u_char *)(th + 1),
1607 (th->th_off << 2) - sizeof(struct tcphdr),
1608 (thflags & TH_SYN) ? TO_SYN : 0);
1610 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1611 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1612 (to.to_flags & TOF_SIGNATURE) == 0) {
1613 TCPSTAT_INC(tcps_sig_err_sigopt);
1614 /* XXX: should drop? */
1618 * If echoed timestamp is later than the current time,
1619 * fall back to non RFC1323 RTT calculation. Normalize
1620 * timestamp if syncookies were used when this connection
1623 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1624 to.to_tsecr -= tp->ts_offset;
1625 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1627 else if (tp->t_flags & TF_PREVVALID &&
1628 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1629 cc_cong_signal(tp, th, CC_RTO_ERR);
1632 * Process options only when we get SYN/ACK back. The SYN case
1633 * for incoming connections is handled in tcp_syncache.
1634 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1635 * or <SYN,ACK>) segment itself is never scaled.
1636 * XXX this is traditional behavior, may need to be cleaned up.
1638 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1639 /* Handle parallel SYN for ECN */
1640 if (!(thflags & TH_ACK) &&
1641 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1642 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1643 tp->t_flags2 |= TF2_ECN_PERMIT;
1644 tp->t_flags2 |= TF2_ECN_SND_ECE;
1645 TCPSTAT_INC(tcps_ecn_shs);
1647 if ((to.to_flags & TOF_SCALE) &&
1648 (tp->t_flags & TF_REQ_SCALE) &&
1649 !(tp->t_flags & TF_NOOPT)) {
1650 tp->t_flags |= TF_RCVD_SCALE;
1651 tp->snd_scale = to.to_wscale;
1653 tp->t_flags &= ~TF_REQ_SCALE;
1655 * Initial send window. It will be updated with
1656 * the next incoming segment to the scaled value.
1658 tp->snd_wnd = th->th_win;
1659 if ((to.to_flags & TOF_TS) &&
1660 (tp->t_flags & TF_REQ_TSTMP) &&
1661 !(tp->t_flags & TF_NOOPT)) {
1662 tp->t_flags |= TF_RCVD_TSTMP;
1663 tp->ts_recent = to.to_tsval;
1664 tp->ts_recent_age = tcp_ts_getticks();
1666 tp->t_flags &= ~TF_REQ_TSTMP;
1667 if (to.to_flags & TOF_MSS)
1668 tcp_mss(tp, to.to_mss);
1669 if ((tp->t_flags & TF_SACK_PERMIT) &&
1670 (!(to.to_flags & TOF_SACKPERM) ||
1671 (tp->t_flags & TF_NOOPT)))
1672 tp->t_flags &= ~TF_SACK_PERMIT;
1673 if (IS_FASTOPEN(tp->t_flags)) {
1674 if ((to.to_flags & TOF_FASTOPEN) &&
1675 !(tp->t_flags & TF_NOOPT)) {
1678 if (to.to_flags & TOF_MSS)
1681 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1685 tcp_fastopen_update_cache(tp, mss,
1686 to.to_tfo_len, to.to_tfo_cookie);
1688 tcp_fastopen_disable_path(tp);
1693 * If timestamps were negotiated during SYN/ACK and a
1694 * segment without a timestamp is received, silently drop
1695 * the segment, unless it is a RST segment or missing timestamps are
1697 * See section 3.2 of RFC 7323.
1699 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1700 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1701 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1702 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1703 "segment processed normally\n",
1708 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1709 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1710 "segment silently dropped\n", s, __func__);
1717 * If timestamps were not negotiated during SYN/ACK and a
1718 * segment with a timestamp is received, ignore the
1719 * timestamp and process the packet normally.
1720 * See section 3.2 of RFC 7323.
1722 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1723 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1724 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1725 "segment processed normally\n", s, __func__);
1731 * Header prediction: check for the two common cases
1732 * of a uni-directional data xfer. If the packet has
1733 * no control flags, is in-sequence, the window didn't
1734 * change and we're not retransmitting, it's a
1735 * candidate. If the length is zero and the ack moved
1736 * forward, we're the sender side of the xfer. Just
1737 * free the data acked & wake any higher level process
1738 * that was blocked waiting for space. If the length
1739 * is non-zero and the ack didn't move, we're the
1740 * receiver side. If we're getting packets in-order
1741 * (the reassembly queue is empty), add the data to
1742 * the socket buffer and note that we need a delayed ack.
1743 * Make sure that the hidden state-flags are also off.
1744 * Since we check for TCPS_ESTABLISHED first, it can only
1747 if (tp->t_state == TCPS_ESTABLISHED &&
1748 th->th_seq == tp->rcv_nxt &&
1749 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1750 tp->snd_nxt == tp->snd_max &&
1751 tiwin && tiwin == tp->snd_wnd &&
1752 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1754 ((to.to_flags & TOF_TS) == 0 ||
1755 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1757 * If last ACK falls within this segment's sequence numbers,
1758 * record the timestamp.
1759 * NOTE that the test is modified according to the latest
1760 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1762 if ((to.to_flags & TOF_TS) != 0 &&
1763 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1764 tp->ts_recent_age = tcp_ts_getticks();
1765 tp->ts_recent = to.to_tsval;
1769 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1770 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1771 !IN_RECOVERY(tp->t_flags) &&
1772 (to.to_flags & TOF_SACK) == 0 &&
1773 TAILQ_EMPTY(&tp->snd_holes)) {
1775 * This is a pure ack for outstanding data.
1777 TCPSTAT_INC(tcps_predack);
1780 * "bad retransmit" recovery without timestamps.
1782 if ((to.to_flags & TOF_TS) == 0 &&
1783 tp->t_rxtshift == 1 &&
1784 tp->t_flags & TF_PREVVALID &&
1785 (int)(ticks - tp->t_badrxtwin) < 0) {
1786 cc_cong_signal(tp, th, CC_RTO_ERR);
1790 * Recalculate the transmit timer / rtt.
1792 * Some boxes send broken timestamp replies
1793 * during the SYN+ACK phase, ignore
1794 * timestamps of 0 or we could calculate a
1795 * huge RTT and blow up the retransmit timer.
1797 if ((to.to_flags & TOF_TS) != 0 &&
1801 t = tcp_ts_getticks() - to.to_tsecr;
1802 if (!tp->t_rttlow || tp->t_rttlow > t)
1805 TCP_TS_TO_TICKS(t) + 1);
1806 } else if (tp->t_rtttime &&
1807 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1808 if (!tp->t_rttlow ||
1809 tp->t_rttlow > ticks - tp->t_rtttime)
1810 tp->t_rttlow = ticks - tp->t_rtttime;
1812 ticks - tp->t_rtttime);
1814 acked = BYTES_THIS_ACK(tp, th);
1817 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1818 hhook_run_tcp_est_in(tp, th, &to);
1821 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1822 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1823 sbdrop(&so->so_snd, acked);
1824 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1825 SEQ_LEQ(th->th_ack, tp->snd_recover))
1826 tp->snd_recover = th->th_ack - 1;
1829 * Let the congestion control algorithm update
1830 * congestion control related information. This
1831 * typically means increasing the congestion
1834 cc_ack_received(tp, th, nsegs, CC_ACK);
1836 tp->snd_una = th->th_ack;
1838 * Pull snd_wl2 up to prevent seq wrap relative
1841 tp->snd_wl2 = th->th_ack;
1846 * If all outstanding data are acked, stop
1847 * retransmit timer, otherwise restart timer
1848 * using current (possibly backed-off) value.
1849 * If process is waiting for space,
1850 * wakeup/selwakeup/signal. If data
1851 * are ready to send, let tcp_output
1852 * decide between more output or persist.
1855 if (so->so_options & SO_DEBUG)
1856 tcp_trace(TA_INPUT, ostate, tp,
1857 (void *)tcp_saveipgen,
1860 TCP_PROBE3(debug__input, tp, th, m);
1861 if (tp->snd_una == tp->snd_max)
1862 tcp_timer_activate(tp, TT_REXMT, 0);
1863 else if (!tcp_timer_active(tp, TT_PERSIST))
1864 tcp_timer_activate(tp, TT_REXMT,
1866 tp->t_flags |= TF_WAKESOW;
1867 if (sbavail(&so->so_snd))
1868 (void) tp->t_fb->tfb_tcp_output(tp);
1871 } else if (th->th_ack == tp->snd_una &&
1872 tlen <= sbspace(&so->so_rcv)) {
1873 int newsize = 0; /* automatic sockbuf scaling */
1876 * This is a pure, in-sequence data packet with
1877 * nothing on the reassembly queue and we have enough
1878 * buffer space to take it.
1880 /* Clean receiver SACK report if present */
1881 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1882 tcp_clean_sackreport(tp);
1883 TCPSTAT_INC(tcps_preddat);
1884 tp->rcv_nxt += tlen;
1886 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1887 (tp->t_fbyte_in == 0)) {
1888 tp->t_fbyte_in = ticks;
1889 if (tp->t_fbyte_in == 0)
1891 if (tp->t_fbyte_out && tp->t_fbyte_in)
1892 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1895 * Pull snd_wl1 up to prevent seq wrap relative to
1898 tp->snd_wl1 = th->th_seq;
1900 * Pull rcv_up up to prevent seq wrap relative to
1903 tp->rcv_up = tp->rcv_nxt;
1904 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1905 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1907 if (so->so_options & SO_DEBUG)
1908 tcp_trace(TA_INPUT, ostate, tp,
1909 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1911 TCP_PROBE3(debug__input, tp, th, m);
1913 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1915 /* Add data to socket buffer. */
1916 SOCKBUF_LOCK(&so->so_rcv);
1917 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1921 * Set new socket buffer size.
1922 * Give up when limit is reached.
1925 if (!sbreserve_locked(&so->so_rcv,
1927 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1928 m_adj(m, drop_hdrlen); /* delayed header drop */
1929 sbappendstream_locked(&so->so_rcv, m, 0);
1931 SOCKBUF_UNLOCK(&so->so_rcv);
1932 tp->t_flags |= TF_WAKESOR;
1933 if (DELAY_ACK(tp, tlen)) {
1934 tp->t_flags |= TF_DELACK;
1936 tp->t_flags |= TF_ACKNOW;
1937 tp->t_fb->tfb_tcp_output(tp);
1944 * Calculate amount of space in receive window,
1945 * and then do TCP input processing.
1946 * Receive window is amount of space in rcv queue,
1947 * but not less than advertised window.
1949 win = sbspace(&so->so_rcv);
1952 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1954 switch (tp->t_state) {
1956 * If the state is SYN_RECEIVED:
1957 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1959 case TCPS_SYN_RECEIVED:
1960 if ((thflags & TH_ACK) &&
1961 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1962 SEQ_GT(th->th_ack, tp->snd_max))) {
1963 rstreason = BANDLIM_RST_OPENPORT;
1966 if (IS_FASTOPEN(tp->t_flags)) {
1968 * When a TFO connection is in SYN_RECEIVED, the
1969 * only valid packets are the initial SYN, a
1970 * retransmit/copy of the initial SYN (possibly with
1971 * a subset of the original data), a valid ACK, a
1974 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1975 rstreason = BANDLIM_RST_OPENPORT;
1977 } else if (thflags & TH_SYN) {
1978 /* non-initial SYN is ignored */
1979 if ((tcp_timer_active(tp, TT_DELACK) ||
1980 tcp_timer_active(tp, TT_REXMT)))
1982 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1989 * If the state is SYN_SENT:
1990 * if seg contains a RST with valid ACK (SEQ.ACK has already
1991 * been verified), then drop the connection.
1992 * if seg contains a RST without an ACK, drop the seg.
1993 * if seg does not contain SYN, then drop the seg.
1994 * Otherwise this is an acceptable SYN segment
1995 * initialize tp->rcv_nxt and tp->irs
1996 * if seg contains ack then advance tp->snd_una
1997 * if seg contains an ECE and ECN support is enabled, the stream
1999 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2000 * arrange for segment to be acked (eventually)
2001 * continue processing rest of data/controls, beginning with URG
2004 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2005 TCP_PROBE5(connect__refused, NULL, tp,
2007 tp = tcp_drop(tp, ECONNREFUSED);
2009 if (thflags & TH_RST)
2011 if (!(thflags & TH_SYN))
2014 tp->irs = th->th_seq;
2016 if (thflags & TH_ACK) {
2017 int tfo_partial_ack = 0;
2019 TCPSTAT_INC(tcps_connects);
2022 mac_socketpeer_set_from_mbuf(m, so);
2024 /* Do window scaling on this connection? */
2025 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2026 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2027 tp->rcv_scale = tp->request_r_scale;
2029 tp->rcv_adv += min(tp->rcv_wnd,
2030 TCP_MAXWIN << tp->rcv_scale);
2031 tp->snd_una++; /* SYN is acked */
2033 * If not all the data that was sent in the TFO SYN
2034 * has been acked, resend the remainder right away.
2036 if (IS_FASTOPEN(tp->t_flags) &&
2037 (tp->snd_una != tp->snd_max)) {
2038 tp->snd_nxt = th->th_ack;
2039 tfo_partial_ack = 1;
2042 * If there's data, delay ACK; if there's also a FIN
2043 * ACKNOW will be turned on later.
2045 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2046 tcp_timer_activate(tp, TT_DELACK,
2049 tp->t_flags |= TF_ACKNOW;
2051 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2052 (V_tcp_do_ecn == 1)) {
2053 tp->t_flags2 |= TF2_ECN_PERMIT;
2054 TCPSTAT_INC(tcps_ecn_shs);
2058 * Received <SYN,ACK> in SYN_SENT[*] state.
2060 * SYN_SENT --> ESTABLISHED
2061 * SYN_SENT* --> FIN_WAIT_1
2063 tp->t_starttime = ticks;
2064 if (tp->t_flags & TF_NEEDFIN) {
2065 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2066 tp->t_flags &= ~TF_NEEDFIN;
2069 tcp_state_change(tp, TCPS_ESTABLISHED);
2070 TCP_PROBE5(connect__established, NULL, tp,
2073 tcp_timer_activate(tp, TT_KEEP,
2078 * Received initial SYN in SYN-SENT[*] state =>
2079 * simultaneous open.
2080 * If it succeeds, connection is * half-synchronized.
2081 * Otherwise, do 3-way handshake:
2082 * SYN-SENT -> SYN-RECEIVED
2083 * SYN-SENT* -> SYN-RECEIVED*
2085 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2086 tcp_timer_activate(tp, TT_REXMT, 0);
2087 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2090 INP_WLOCK_ASSERT(tp->t_inpcb);
2093 * Advance th->th_seq to correspond to first data byte.
2094 * If data, trim to stay within window,
2095 * dropping FIN if necessary.
2098 if (tlen > tp->rcv_wnd) {
2099 todrop = tlen - tp->rcv_wnd;
2103 TCPSTAT_INC(tcps_rcvpackafterwin);
2104 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2106 tp->snd_wl1 = th->th_seq - 1;
2107 tp->rcv_up = th->th_seq;
2109 * Client side of transaction: already sent SYN and data.
2110 * If the remote host used T/TCP to validate the SYN,
2111 * our data will be ACK'd; if so, enter normal data segment
2112 * processing in the middle of step 5, ack processing.
2113 * Otherwise, goto step 6.
2115 if (thflags & TH_ACK)
2121 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2122 * do normal processing.
2124 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2128 break; /* continue normal processing */
2132 * States other than LISTEN or SYN_SENT.
2133 * First check the RST flag and sequence number since reset segments
2134 * are exempt from the timestamp and connection count tests. This
2135 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2136 * below which allowed reset segments in half the sequence space
2137 * to fall though and be processed (which gives forged reset
2138 * segments with a random sequence number a 50 percent chance of
2139 * killing a connection).
2140 * Then check timestamp, if present.
2141 * Then check the connection count, if present.
2142 * Then check that at least some bytes of segment are within
2143 * receive window. If segment begins before rcv_nxt,
2144 * drop leading data (and SYN); if nothing left, just ack.
2146 if (thflags & TH_RST) {
2148 * RFC5961 Section 3.2
2150 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2151 * - If RST is in window, we send challenge ACK.
2153 * Note: to take into account delayed ACKs, we should
2154 * test against last_ack_sent instead of rcv_nxt.
2155 * Note 2: we handle special case of closed window, not
2156 * covered by the RFC.
2158 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2159 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2160 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2161 KASSERT(tp->t_state != TCPS_SYN_SENT,
2162 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2165 if (V_tcp_insecure_rst ||
2166 tp->last_ack_sent == th->th_seq) {
2167 TCPSTAT_INC(tcps_drops);
2168 /* Drop the connection. */
2169 switch (tp->t_state) {
2170 case TCPS_SYN_RECEIVED:
2171 so->so_error = ECONNREFUSED;
2173 case TCPS_ESTABLISHED:
2174 case TCPS_FIN_WAIT_1:
2175 case TCPS_FIN_WAIT_2:
2176 case TCPS_CLOSE_WAIT:
2179 so->so_error = ECONNRESET;
2186 TCPSTAT_INC(tcps_badrst);
2187 /* Send challenge ACK. */
2188 tcp_respond(tp, mtod(m, void *), th, m,
2189 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2190 tp->last_ack_sent = tp->rcv_nxt;
2198 * RFC5961 Section 4.2
2199 * Send challenge ACK for any SYN in synchronized state.
2201 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2202 tp->t_state != TCPS_SYN_RECEIVED) {
2203 TCPSTAT_INC(tcps_badsyn);
2204 if (V_tcp_insecure_syn &&
2205 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2206 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2207 tp = tcp_drop(tp, ECONNRESET);
2208 rstreason = BANDLIM_UNLIMITED;
2210 /* Send challenge ACK. */
2211 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2212 tp->snd_nxt, TH_ACK);
2213 tp->last_ack_sent = tp->rcv_nxt;
2220 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2221 * and it's less than ts_recent, drop it.
2223 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2224 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2225 /* Check to see if ts_recent is over 24 days old. */
2226 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2228 * Invalidate ts_recent. If this segment updates
2229 * ts_recent, the age will be reset later and ts_recent
2230 * will get a valid value. If it does not, setting
2231 * ts_recent to zero will at least satisfy the
2232 * requirement that zero be placed in the timestamp
2233 * echo reply when ts_recent isn't valid. The
2234 * age isn't reset until we get a valid ts_recent
2235 * because we don't want out-of-order segments to be
2236 * dropped when ts_recent is old.
2240 TCPSTAT_INC(tcps_rcvduppack);
2241 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2242 TCPSTAT_INC(tcps_pawsdrop);
2250 * In the SYN-RECEIVED state, validate that the packet belongs to
2251 * this connection before trimming the data to fit the receive
2252 * window. Check the sequence number versus IRS since we know
2253 * the sequence numbers haven't wrapped. This is a partial fix
2254 * for the "LAND" DoS attack.
2256 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2257 rstreason = BANDLIM_RST_OPENPORT;
2261 todrop = tp->rcv_nxt - th->th_seq;
2263 if (thflags & TH_SYN) {
2273 * Following if statement from Stevens, vol. 2, p. 960.
2276 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2278 * Any valid FIN must be to the left of the window.
2279 * At this point the FIN must be a duplicate or out
2280 * of sequence; drop it.
2285 * Send an ACK to resynchronize and drop any data.
2286 * But keep on processing for RST or ACK.
2288 tp->t_flags |= TF_ACKNOW;
2290 TCPSTAT_INC(tcps_rcvduppack);
2291 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2293 TCPSTAT_INC(tcps_rcvpartduppack);
2294 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2297 * DSACK - add SACK block for dropped range
2299 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2300 tcp_update_sack_list(tp, th->th_seq,
2301 th->th_seq + todrop);
2303 * ACK now, as the next in-sequence segment
2304 * will clear the DSACK block again
2306 tp->t_flags |= TF_ACKNOW;
2308 drop_hdrlen += todrop; /* drop from the top afterwards */
2309 th->th_seq += todrop;
2311 if (th->th_urp > todrop)
2312 th->th_urp -= todrop;
2320 * If new data are received on a connection after the
2321 * user processes are gone, then RST the other end.
2323 if ((so->so_state & SS_NOFDREF) &&
2324 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2325 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2326 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2327 "after socket was closed, "
2328 "sending RST and removing tcpcb\n",
2329 s, __func__, tcpstates[tp->t_state], tlen);
2333 TCPSTAT_INC(tcps_rcvafterclose);
2334 rstreason = BANDLIM_UNLIMITED;
2339 * If segment ends after window, drop trailing data
2340 * (and PUSH and FIN); if nothing left, just ACK.
2342 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2344 TCPSTAT_INC(tcps_rcvpackafterwin);
2345 if (todrop >= tlen) {
2346 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2348 * If window is closed can only take segments at
2349 * window edge, and have to drop data and PUSH from
2350 * incoming segments. Continue processing, but
2351 * remember to ack. Otherwise, drop segment
2354 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2355 tp->t_flags |= TF_ACKNOW;
2356 TCPSTAT_INC(tcps_rcvwinprobe);
2360 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2363 thflags &= ~(TH_PUSH|TH_FIN);
2367 * If last ACK falls within this segment's sequence numbers,
2368 * record its timestamp.
2370 * 1) That the test incorporates suggestions from the latest
2371 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2372 * 2) That updating only on newer timestamps interferes with
2373 * our earlier PAWS tests, so this check should be solely
2374 * predicated on the sequence space of this segment.
2375 * 3) That we modify the segment boundary check to be
2376 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2377 * instead of RFC1323's
2378 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2379 * This modified check allows us to overcome RFC1323's
2380 * limitations as described in Stevens TCP/IP Illustrated
2381 * Vol. 2 p.869. In such cases, we can still calculate the
2382 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2384 if ((to.to_flags & TOF_TS) != 0 &&
2385 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2386 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2387 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2388 tp->ts_recent_age = tcp_ts_getticks();
2389 tp->ts_recent = to.to_tsval;
2393 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2394 * flag is on (half-synchronized state), then queue data for
2395 * later processing; else drop segment and return.
2397 if ((thflags & TH_ACK) == 0) {
2398 if (tp->t_state == TCPS_SYN_RECEIVED ||
2399 (tp->t_flags & TF_NEEDSYN)) {
2400 if (tp->t_state == TCPS_SYN_RECEIVED &&
2401 IS_FASTOPEN(tp->t_flags)) {
2402 tp->snd_wnd = tiwin;
2406 } else if (tp->t_flags & TF_ACKNOW)
2415 switch (tp->t_state) {
2417 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2418 * ESTABLISHED state and continue processing.
2419 * The ACK was checked above.
2421 case TCPS_SYN_RECEIVED:
2423 TCPSTAT_INC(tcps_connects);
2425 /* Do window scaling? */
2426 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2427 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2428 tp->rcv_scale = tp->request_r_scale;
2430 tp->snd_wnd = tiwin;
2433 * SYN-RECEIVED -> ESTABLISHED
2434 * SYN-RECEIVED* -> FIN-WAIT-1
2436 tp->t_starttime = ticks;
2437 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2438 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2439 tp->t_tfo_pending = NULL;
2441 if (tp->t_flags & TF_NEEDFIN) {
2442 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2443 tp->t_flags &= ~TF_NEEDFIN;
2445 tcp_state_change(tp, TCPS_ESTABLISHED);
2446 TCP_PROBE5(accept__established, NULL, tp,
2449 * TFO connections call cc_conn_init() during SYN
2450 * processing. Calling it again here for such
2451 * connections is not harmless as it would undo the
2452 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2455 if (!IS_FASTOPEN(tp->t_flags))
2457 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2460 * Account for the ACK of our SYN prior to
2461 * regular ACK processing below, except for
2462 * simultaneous SYN, which is handled later.
2464 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2467 * If segment contains data or ACK, will call tcp_reass()
2468 * later; if not, do so now to pass queued data to user.
2470 if (tlen == 0 && (thflags & TH_FIN) == 0)
2471 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2473 tp->snd_wl1 = th->th_seq - 1;
2477 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2478 * ACKs. If the ack is in the range
2479 * tp->snd_una < th->th_ack <= tp->snd_max
2480 * then advance tp->snd_una to th->th_ack and drop
2481 * data from the retransmission queue. If this ACK reflects
2482 * more up to date window information we update our window information.
2484 case TCPS_ESTABLISHED:
2485 case TCPS_FIN_WAIT_1:
2486 case TCPS_FIN_WAIT_2:
2487 case TCPS_CLOSE_WAIT:
2490 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2491 TCPSTAT_INC(tcps_rcvacktoomuch);
2494 if ((tp->t_flags & TF_SACK_PERMIT) &&
2495 ((to.to_flags & TOF_SACK) ||
2496 !TAILQ_EMPTY(&tp->snd_holes)))
2497 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2500 * Reset the value so that previous (valid) value
2501 * from the last ack with SACK doesn't get used.
2503 tp->sackhint.sacked_bytes = 0;
2506 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2507 hhook_run_tcp_est_in(tp, th, &to);
2510 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2511 maxseg = tcp_maxseg(tp);
2513 (tiwin == tp->snd_wnd ||
2514 (tp->t_flags & TF_SACK_PERMIT))) {
2516 * If this is the first time we've seen a
2517 * FIN from the remote, this is not a
2518 * duplicate and it needs to be processed
2519 * normally. This happens during a
2520 * simultaneous close.
2522 if ((thflags & TH_FIN) &&
2523 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2527 TCPSTAT_INC(tcps_rcvdupack);
2529 * If we have outstanding data (other than
2530 * a window probe), this is a completely
2531 * duplicate ack (ie, window info didn't
2532 * change and FIN isn't set),
2533 * the ack is the biggest we've
2534 * seen and we've seen exactly our rexmt
2535 * threshold of them, assume a packet
2536 * has been dropped and retransmit it.
2537 * Kludge snd_nxt & the congestion
2538 * window so we send only this one
2541 * We know we're losing at the current
2542 * window size so do congestion avoidance
2543 * (set ssthresh to half the current window
2544 * and pull our congestion window back to
2545 * the new ssthresh).
2547 * Dup acks mean that packets have left the
2548 * network (they're now cached at the receiver)
2549 * so bump cwnd by the amount in the receiver
2550 * to keep a constant cwnd packets in the
2553 * When using TCP ECN, notify the peer that
2554 * we reduced the cwnd.
2557 * Following 2 kinds of acks should not affect
2560 * 2) Acks with SACK but without any new SACK
2561 * information in them. These could result from
2562 * any anomaly in the network like a switch
2563 * duplicating packets or a possible DoS attack.
2565 if (th->th_ack != tp->snd_una ||
2566 ((tp->t_flags & TF_SACK_PERMIT) &&
2567 (to.to_flags & TOF_SACK) &&
2570 else if (!tcp_timer_active(tp, TT_REXMT))
2572 else if (++tp->t_dupacks > tcprexmtthresh ||
2573 IN_FASTRECOVERY(tp->t_flags)) {
2574 cc_ack_received(tp, th, nsegs,
2577 IN_FASTRECOVERY(tp->t_flags) &&
2578 (tp->t_flags & TF_SACK_PERMIT)) {
2579 tcp_do_prr_ack(tp, th);
2580 } else if ((tp->t_flags & TF_SACK_PERMIT) &&
2581 (to.to_flags & TOF_SACK) &&
2582 IN_FASTRECOVERY(tp->t_flags)) {
2586 * Compute the amount of data in flight first.
2587 * We can inject new data into the pipe iff
2588 * we have less than 1/2 the original window's
2589 * worth of data in flight.
2591 if (V_tcp_do_rfc6675_pipe)
2592 awnd = tcp_compute_pipe(tp);
2594 awnd = (tp->snd_nxt - tp->snd_fack) +
2595 tp->sackhint.sack_bytes_rexmit;
2597 if (awnd < tp->snd_ssthresh) {
2598 tp->snd_cwnd += maxseg;
2599 if (tp->snd_cwnd > tp->snd_ssthresh)
2600 tp->snd_cwnd = tp->snd_ssthresh;
2603 tp->snd_cwnd += maxseg;
2604 (void) tp->t_fb->tfb_tcp_output(tp);
2606 } else if (tp->t_dupacks == tcprexmtthresh ||
2607 (tp->t_flags & TF_SACK_PERMIT &&
2608 V_tcp_do_rfc6675_pipe &&
2609 tp->sackhint.sacked_bytes >
2610 (tcprexmtthresh - 1) * maxseg)) {
2613 * Above is the RFC6675 trigger condition of
2614 * more than (dupthresh-1)*maxseg sacked data.
2615 * If the count of holes in the
2616 * scoreboard is >= dupthresh, we could
2617 * also enter loss recovery, but don't
2618 * have that value readily available.
2620 tp->t_dupacks = tcprexmtthresh;
2621 tcp_seq onxt = tp->snd_nxt;
2624 * If we're doing sack, or prr, check
2625 * to see if we're already in sack
2626 * recovery. If we're not doing sack,
2627 * check to see if we're in newreno
2631 (tp->t_flags & TF_SACK_PERMIT)) {
2632 if (IN_FASTRECOVERY(tp->t_flags)) {
2637 if (SEQ_LEQ(th->th_ack,
2643 /* Congestion signal before ack. */
2644 cc_cong_signal(tp, th, CC_NDUPACK);
2645 cc_ack_received(tp, th, nsegs,
2647 tcp_timer_activate(tp, TT_REXMT, 0);
2651 * snd_ssthresh is already updated by
2654 tp->sackhint.prr_delivered =
2655 tp->sackhint.sacked_bytes;
2656 tp->sackhint.recover_fs = max(1,
2657 tp->snd_nxt - tp->snd_una);
2659 if ((tp->t_flags & TF_SACK_PERMIT) &&
2660 (to.to_flags & TOF_SACK)) {
2662 tcps_sack_recovery_episode);
2663 tp->snd_recover = tp->snd_nxt;
2664 tp->snd_cwnd = maxseg;
2665 (void) tp->t_fb->tfb_tcp_output(tp);
2666 if (SEQ_GT(th->th_ack, tp->snd_una))
2667 goto resume_partialack;
2670 tp->snd_nxt = th->th_ack;
2671 tp->snd_cwnd = maxseg;
2672 (void) tp->t_fb->tfb_tcp_output(tp);
2673 KASSERT(tp->snd_limited <= 2,
2674 ("%s: tp->snd_limited too big",
2676 tp->snd_cwnd = tp->snd_ssthresh +
2678 (tp->t_dupacks - tp->snd_limited);
2679 if (SEQ_GT(onxt, tp->snd_nxt))
2682 } else if (V_tcp_do_rfc3042) {
2684 * Process first and second duplicate
2685 * ACKs. Each indicates a segment
2686 * leaving the network, creating room
2687 * for more. Make sure we can send a
2688 * packet on reception of each duplicate
2689 * ACK by increasing snd_cwnd by one
2690 * segment. Restore the original
2691 * snd_cwnd after packet transmission.
2693 cc_ack_received(tp, th, nsegs,
2695 uint32_t oldcwnd = tp->snd_cwnd;
2696 tcp_seq oldsndmax = tp->snd_max;
2700 KASSERT(tp->t_dupacks == 1 ||
2702 ("%s: dupacks not 1 or 2",
2704 if (tp->t_dupacks == 1)
2705 tp->snd_limited = 0;
2707 (tp->snd_nxt - tp->snd_una) +
2708 (tp->t_dupacks - tp->snd_limited) *
2711 * Only call tcp_output when there
2712 * is new data available to be sent.
2713 * Otherwise we would send pure ACKs.
2715 SOCKBUF_LOCK(&so->so_snd);
2716 avail = sbavail(&so->so_snd) -
2717 (tp->snd_nxt - tp->snd_una);
2718 SOCKBUF_UNLOCK(&so->so_snd);
2720 (void) tp->t_fb->tfb_tcp_output(tp);
2721 sent = tp->snd_max - oldsndmax;
2722 if (sent > maxseg) {
2723 KASSERT((tp->t_dupacks == 2 &&
2724 tp->snd_limited == 0) ||
2725 (sent == maxseg + 1 &&
2726 tp->t_flags & TF_SENTFIN),
2727 ("%s: sent too much",
2729 tp->snd_limited = 2;
2730 } else if (sent > 0)
2732 tp->snd_cwnd = oldcwnd;
2739 * This ack is advancing the left edge, reset the
2744 * If this ack also has new SACK info, increment the
2745 * counter as per rfc6675. The variable
2746 * sack_changed tracks all changes to the SACK
2747 * scoreboard, including when partial ACKs without
2748 * SACK options are received, and clear the scoreboard
2749 * from the left side. Such partial ACKs should not be
2750 * counted as dupacks here.
2752 if ((tp->t_flags & TF_SACK_PERMIT) &&
2753 (to.to_flags & TOF_SACK) &&
2756 /* limit overhead by setting maxseg last */
2757 if (!IN_FASTRECOVERY(tp->t_flags) &&
2758 (tp->sackhint.sacked_bytes >
2759 ((tcprexmtthresh - 1) *
2760 (maxseg = tcp_maxseg(tp))))) {
2761 goto enter_recovery;
2767 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2768 ("%s: th_ack <= snd_una", __func__));
2771 * If the congestion window was inflated to account
2772 * for the other side's cached packets, retract it.
2774 if (IN_FASTRECOVERY(tp->t_flags)) {
2775 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2776 if (tp->t_flags & TF_SACK_PERMIT)
2777 if (V_tcp_do_prr && to.to_flags & TOF_SACK) {
2778 tcp_timer_activate(tp, TT_REXMT, 0);
2780 tcp_do_prr_ack(tp, th);
2781 tp->t_flags |= TF_ACKNOW;
2782 (void) tcp_output(tp);
2784 tcp_sack_partialack(tp, th);
2786 tcp_newreno_partial_ack(tp, th);
2788 cc_post_recovery(tp, th);
2789 } else if (IN_CONGRECOVERY(tp->t_flags)) {
2790 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2792 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th);
2793 tp->snd_fack = th->th_ack;
2794 tcp_do_prr_ack(tp, th);
2795 (void) tcp_output(tp);
2798 cc_post_recovery(tp, th);
2801 * If we reach this point, ACK is not a duplicate,
2802 * i.e., it ACKs something we sent.
2804 if (tp->t_flags & TF_NEEDSYN) {
2806 * T/TCP: Connection was half-synchronized, and our
2807 * SYN has been ACK'd (so connection is now fully
2808 * synchronized). Go to non-starred state,
2809 * increment snd_una for ACK of SYN, and check if
2810 * we can do window scaling.
2812 tp->t_flags &= ~TF_NEEDSYN;
2814 /* Do window scaling? */
2815 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2816 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2817 tp->rcv_scale = tp->request_r_scale;
2818 /* Send window already scaled. */
2823 INP_WLOCK_ASSERT(tp->t_inpcb);
2826 * Adjust for the SYN bit in sequence space,
2827 * but don't account for it in cwnd calculations.
2828 * This is for the SYN_RECEIVED, non-simultaneous
2829 * SYN case. SYN_SENT and simultaneous SYN are
2830 * treated elsewhere.
2834 acked = BYTES_THIS_ACK(tp, th);
2835 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2836 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2837 tp->snd_una, th->th_ack, tp, m));
2838 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2839 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2842 * If we just performed our first retransmit, and the ACK
2843 * arrives within our recovery window, then it was a mistake
2844 * to do the retransmit in the first place. Recover our
2845 * original cwnd and ssthresh, and proceed to transmit where
2848 if (tp->t_rxtshift == 1 &&
2849 tp->t_flags & TF_PREVVALID &&
2851 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2852 cc_cong_signal(tp, th, CC_RTO_ERR);
2855 * If we have a timestamp reply, update smoothed
2856 * round trip time. If no timestamp is present but
2857 * transmit timer is running and timed sequence
2858 * number was acked, update smoothed round trip time.
2859 * Since we now have an rtt measurement, cancel the
2860 * timer backoff (cf., Phil Karn's retransmit alg.).
2861 * Recompute the initial retransmit timer.
2863 * Some boxes send broken timestamp replies
2864 * during the SYN+ACK phase, ignore
2865 * timestamps of 0 or we could calculate a
2866 * huge RTT and blow up the retransmit timer.
2868 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2871 t = tcp_ts_getticks() - to.to_tsecr;
2872 if (!tp->t_rttlow || tp->t_rttlow > t)
2874 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2875 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2876 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2877 tp->t_rttlow = ticks - tp->t_rtttime;
2878 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2882 * If all outstanding data is acked, stop retransmit
2883 * timer and remember to restart (more output or persist).
2884 * If there is more data to be acked, restart retransmit
2885 * timer, using current (possibly backed-off) value.
2887 if (th->th_ack == tp->snd_max) {
2888 tcp_timer_activate(tp, TT_REXMT, 0);
2890 } else if (!tcp_timer_active(tp, TT_PERSIST))
2891 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2894 * If no data (only SYN) was ACK'd,
2895 * skip rest of ACK processing.
2901 * Let the congestion control algorithm update congestion
2902 * control related information. This typically means increasing
2903 * the congestion window.
2905 cc_ack_received(tp, th, nsegs, CC_ACK);
2907 SOCKBUF_LOCK(&so->so_snd);
2908 if (acked > sbavail(&so->so_snd)) {
2909 if (tp->snd_wnd >= sbavail(&so->so_snd))
2910 tp->snd_wnd -= sbavail(&so->so_snd);
2913 mfree = sbcut_locked(&so->so_snd,
2914 (int)sbavail(&so->so_snd));
2917 mfree = sbcut_locked(&so->so_snd, acked);
2918 if (tp->snd_wnd >= (uint32_t) acked)
2919 tp->snd_wnd -= acked;
2924 SOCKBUF_UNLOCK(&so->so_snd);
2925 tp->t_flags |= TF_WAKESOW;
2927 /* Detect una wraparound. */
2928 if (!IN_RECOVERY(tp->t_flags) &&
2929 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2930 SEQ_LEQ(th->th_ack, tp->snd_recover))
2931 tp->snd_recover = th->th_ack - 1;
2932 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2933 if (IN_RECOVERY(tp->t_flags) &&
2934 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2935 EXIT_RECOVERY(tp->t_flags);
2937 tp->snd_una = th->th_ack;
2938 if (tp->t_flags & TF_SACK_PERMIT) {
2939 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2940 tp->snd_recover = tp->snd_una;
2942 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2943 tp->snd_nxt = tp->snd_una;
2945 switch (tp->t_state) {
2947 * In FIN_WAIT_1 STATE in addition to the processing
2948 * for the ESTABLISHED state if our FIN is now acknowledged
2949 * then enter FIN_WAIT_2.
2951 case TCPS_FIN_WAIT_1:
2952 if (ourfinisacked) {
2954 * If we can't receive any more
2955 * data, then closing user can proceed.
2956 * Starting the timer is contrary to the
2957 * specification, but if we don't get a FIN
2958 * we'll hang forever.
2961 * we should release the tp also, and use a
2964 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2965 soisdisconnected(so);
2966 tcp_timer_activate(tp, TT_2MSL,
2967 (tcp_fast_finwait2_recycle ?
2968 tcp_finwait2_timeout :
2971 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2976 * In CLOSING STATE in addition to the processing for
2977 * the ESTABLISHED state if the ACK acknowledges our FIN
2978 * then enter the TIME-WAIT state, otherwise ignore
2982 if (ourfinisacked) {
2990 * In LAST_ACK, we may still be waiting for data to drain
2991 * and/or to be acked, as well as for the ack of our FIN.
2992 * If our FIN is now acknowledged, delete the TCB,
2993 * enter the closed state and return.
2996 if (ourfinisacked) {
3005 INP_WLOCK_ASSERT(tp->t_inpcb);
3008 * Update window information.
3009 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3011 if ((thflags & TH_ACK) &&
3012 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3013 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3014 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3015 /* keep track of pure window updates */
3017 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3018 TCPSTAT_INC(tcps_rcvwinupd);
3019 tp->snd_wnd = tiwin;
3020 tp->snd_wl1 = th->th_seq;
3021 tp->snd_wl2 = th->th_ack;
3022 if (tp->snd_wnd > tp->max_sndwnd)
3023 tp->max_sndwnd = tp->snd_wnd;
3028 * Process segments with URG.
3030 if ((thflags & TH_URG) && th->th_urp &&
3031 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3033 * This is a kludge, but if we receive and accept
3034 * random urgent pointers, we'll crash in
3035 * soreceive. It's hard to imagine someone
3036 * actually wanting to send this much urgent data.
3038 SOCKBUF_LOCK(&so->so_rcv);
3039 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3040 th->th_urp = 0; /* XXX */
3041 thflags &= ~TH_URG; /* XXX */
3042 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3043 goto dodata; /* XXX */
3046 * If this segment advances the known urgent pointer,
3047 * then mark the data stream. This should not happen
3048 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3049 * a FIN has been received from the remote side.
3050 * In these states we ignore the URG.
3052 * According to RFC961 (Assigned Protocols),
3053 * the urgent pointer points to the last octet
3054 * of urgent data. We continue, however,
3055 * to consider it to indicate the first octet
3056 * of data past the urgent section as the original
3057 * spec states (in one of two places).
3059 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3060 tp->rcv_up = th->th_seq + th->th_urp;
3061 so->so_oobmark = sbavail(&so->so_rcv) +
3062 (tp->rcv_up - tp->rcv_nxt) - 1;
3063 if (so->so_oobmark == 0)
3064 so->so_rcv.sb_state |= SBS_RCVATMARK;
3066 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3068 SOCKBUF_UNLOCK(&so->so_rcv);
3070 * Remove out of band data so doesn't get presented to user.
3071 * This can happen independent of advancing the URG pointer,
3072 * but if two URG's are pending at once, some out-of-band
3073 * data may creep in... ick.
3075 if (th->th_urp <= (uint32_t)tlen &&
3076 !(so->so_options & SO_OOBINLINE)) {
3077 /* hdr drop is delayed */
3078 tcp_pulloutofband(so, th, m, drop_hdrlen);
3082 * If no out of band data is expected,
3083 * pull receive urgent pointer along
3084 * with the receive window.
3086 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3087 tp->rcv_up = tp->rcv_nxt;
3090 INP_WLOCK_ASSERT(tp->t_inpcb);
3093 * Process the segment text, merging it into the TCP sequencing queue,
3094 * and arranging for acknowledgment of receipt if necessary.
3095 * This process logically involves adjusting tp->rcv_wnd as data
3096 * is presented to the user (this happens in tcp_usrreq.c,
3097 * case PRU_RCVD). If a FIN has already been received on this
3098 * connection then we just ignore the text.
3100 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3101 IS_FASTOPEN(tp->t_flags));
3102 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3103 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3104 tcp_seq save_start = th->th_seq;
3105 tcp_seq save_rnxt = tp->rcv_nxt;
3106 int save_tlen = tlen;
3107 m_adj(m, drop_hdrlen); /* delayed header drop */
3109 * Insert segment which includes th into TCP reassembly queue
3110 * with control block tp. Set thflags to whether reassembly now
3111 * includes a segment with FIN. This handles the common case
3112 * inline (segment is the next to be received on an established
3113 * connection, and the queue is empty), avoiding linkage into
3114 * and removal from the queue and repetition of various
3116 * Set DELACK for segments received in order, but ack
3117 * immediately when segments are out of order (so
3118 * fast retransmit can work).
3120 if (th->th_seq == tp->rcv_nxt &&
3122 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3124 if (DELAY_ACK(tp, tlen) || tfo_syn)
3125 tp->t_flags |= TF_DELACK;
3127 tp->t_flags |= TF_ACKNOW;
3128 tp->rcv_nxt += tlen;
3130 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3131 (tp->t_fbyte_in == 0)) {
3132 tp->t_fbyte_in = ticks;
3133 if (tp->t_fbyte_in == 0)
3135 if (tp->t_fbyte_out && tp->t_fbyte_in)
3136 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3138 thflags = th->th_flags & TH_FIN;
3139 TCPSTAT_INC(tcps_rcvpack);
3140 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3141 SOCKBUF_LOCK(&so->so_rcv);
3142 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3145 sbappendstream_locked(&so->so_rcv, m, 0);
3146 SOCKBUF_UNLOCK(&so->so_rcv);
3147 tp->t_flags |= TF_WAKESOR;
3150 * XXX: Due to the header drop above "th" is
3151 * theoretically invalid by now. Fortunately
3152 * m_adj() doesn't actually frees any mbufs
3153 * when trimming from the head.
3155 tcp_seq temp = save_start;
3156 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3157 tp->t_flags |= TF_ACKNOW;
3159 if ((tp->t_flags & TF_SACK_PERMIT) &&
3161 TCPS_HAVEESTABLISHED(tp->t_state)) {
3162 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3164 * DSACK actually handled in the fastpath
3167 tcp_update_sack_list(tp, save_start,
3168 save_start + save_tlen);
3169 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3170 if ((tp->rcv_numsacks >= 1) &&
3171 (tp->sackblks[0].end == save_start)) {
3173 * Partial overlap, recorded at todrop
3176 tcp_update_sack_list(tp,
3177 tp->sackblks[0].start,
3178 tp->sackblks[0].end);
3180 tcp_update_dsack_list(tp, save_start,
3181 save_start + save_tlen);
3183 } else if (tlen >= save_tlen) {
3184 /* Update of sackblks. */
3185 tcp_update_dsack_list(tp, save_start,
3186 save_start + save_tlen);
3187 } else if (tlen > 0) {
3188 tcp_update_dsack_list(tp, save_start,
3194 * Note the amount of data that peer has sent into
3195 * our window, in order to estimate the sender's
3199 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3200 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3202 len = so->so_rcv.sb_hiwat;
3210 * If FIN is received ACK the FIN and let the user know
3211 * that the connection is closing.
3213 if (thflags & TH_FIN) {
3214 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3216 /* The socket upcall is handled by socantrcvmore. */
3217 tp->t_flags &= ~TF_WAKESOR;
3219 * If connection is half-synchronized
3220 * (ie NEEDSYN flag on) then delay ACK,
3221 * so it may be piggybacked when SYN is sent.
3222 * Otherwise, since we received a FIN then no
3223 * more input can be expected, send ACK now.
3225 if (tp->t_flags & TF_NEEDSYN)
3226 tp->t_flags |= TF_DELACK;
3228 tp->t_flags |= TF_ACKNOW;
3231 switch (tp->t_state) {
3233 * In SYN_RECEIVED and ESTABLISHED STATES
3234 * enter the CLOSE_WAIT state.
3236 case TCPS_SYN_RECEIVED:
3237 tp->t_starttime = ticks;
3239 case TCPS_ESTABLISHED:
3240 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3244 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3245 * enter the CLOSING state.
3247 case TCPS_FIN_WAIT_1:
3248 tcp_state_change(tp, TCPS_CLOSING);
3252 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3253 * starting the time-wait timer, turning off the other
3256 case TCPS_FIN_WAIT_2:
3262 if (so->so_options & SO_DEBUG)
3263 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3266 TCP_PROBE3(debug__input, tp, th, m);
3269 * Return any desired output.
3271 if (needoutput || (tp->t_flags & TF_ACKNOW))
3272 (void) tp->t_fb->tfb_tcp_output(tp);
3275 INP_WLOCK_ASSERT(tp->t_inpcb);
3277 if (tp->t_flags & TF_DELACK) {
3278 tp->t_flags &= ~TF_DELACK;
3279 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3281 tcp_handle_wakeup(tp, so);
3282 INP_WUNLOCK(tp->t_inpcb);
3287 * Generate an ACK dropping incoming segment if it occupies
3288 * sequence space, where the ACK reflects our state.
3290 * We can now skip the test for the RST flag since all
3291 * paths to this code happen after packets containing
3292 * RST have been dropped.
3294 * In the SYN-RECEIVED state, don't send an ACK unless the
3295 * segment we received passes the SYN-RECEIVED ACK test.
3296 * If it fails send a RST. This breaks the loop in the
3297 * "LAND" DoS attack, and also prevents an ACK storm
3298 * between two listening ports that have been sent forged
3299 * SYN segments, each with the source address of the other.
3301 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3302 (SEQ_GT(tp->snd_una, th->th_ack) ||
3303 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3304 rstreason = BANDLIM_RST_OPENPORT;
3308 if (so->so_options & SO_DEBUG)
3309 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3312 TCP_PROBE3(debug__input, tp, th, m);
3313 tp->t_flags |= TF_ACKNOW;
3314 (void) tp->t_fb->tfb_tcp_output(tp);
3315 tcp_handle_wakeup(tp, so);
3316 INP_WUNLOCK(tp->t_inpcb);
3322 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3323 tcp_handle_wakeup(tp, so);
3324 INP_WUNLOCK(tp->t_inpcb);
3326 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3331 * Drop space held by incoming segment and return.
3334 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3335 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3338 TCP_PROBE3(debug__input, tp, th, m);
3340 tcp_handle_wakeup(tp, so);
3341 INP_WUNLOCK(tp->t_inpcb);
3347 * Issue RST and make ACK acceptable to originator of segment.
3348 * The mbuf must still include the original packet header.
3352 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3353 int tlen, int rstreason)
3359 struct ip6_hdr *ip6;
3363 INP_WLOCK_ASSERT(tp->t_inpcb);
3366 /* Don't bother if destination was broadcast/multicast. */
3367 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3370 if (mtod(m, struct ip *)->ip_v == 6) {
3371 ip6 = mtod(m, struct ip6_hdr *);
3372 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3373 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3375 /* IPv6 anycast check is done at tcp6_input() */
3378 #if defined(INET) && defined(INET6)
3383 ip = mtod(m, struct ip *);
3384 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3385 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3386 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3387 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3392 /* Perform bandwidth limiting. */
3393 if (badport_bandlim(rstreason) < 0)
3396 /* tcp_respond consumes the mbuf chain. */
3397 if (th->th_flags & TH_ACK) {
3398 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3399 th->th_ack, TH_RST);
3401 if (th->th_flags & TH_SYN)
3403 if (th->th_flags & TH_FIN)
3405 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3406 (tcp_seq)0, TH_RST|TH_ACK);
3414 * Parse TCP options and place in tcpopt.
3417 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3422 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3424 if (opt == TCPOPT_EOL)
3426 if (opt == TCPOPT_NOP)
3432 if (optlen < 2 || optlen > cnt)
3437 if (optlen != TCPOLEN_MAXSEG)
3439 if (!(flags & TO_SYN))
3441 to->to_flags |= TOF_MSS;
3442 bcopy((char *)cp + 2,
3443 (char *)&to->to_mss, sizeof(to->to_mss));
3444 to->to_mss = ntohs(to->to_mss);
3447 if (optlen != TCPOLEN_WINDOW)
3449 if (!(flags & TO_SYN))
3451 to->to_flags |= TOF_SCALE;
3452 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3454 case TCPOPT_TIMESTAMP:
3455 if (optlen != TCPOLEN_TIMESTAMP)
3457 to->to_flags |= TOF_TS;
3458 bcopy((char *)cp + 2,
3459 (char *)&to->to_tsval, sizeof(to->to_tsval));
3460 to->to_tsval = ntohl(to->to_tsval);
3461 bcopy((char *)cp + 6,
3462 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3463 to->to_tsecr = ntohl(to->to_tsecr);
3465 case TCPOPT_SIGNATURE:
3467 * In order to reply to a host which has set the
3468 * TCP_SIGNATURE option in its initial SYN, we have
3469 * to record the fact that the option was observed
3470 * here for the syncache code to perform the correct
3473 if (optlen != TCPOLEN_SIGNATURE)
3475 to->to_flags |= TOF_SIGNATURE;
3476 to->to_signature = cp + 2;
3478 case TCPOPT_SACK_PERMITTED:
3479 if (optlen != TCPOLEN_SACK_PERMITTED)
3481 if (!(flags & TO_SYN))
3485 to->to_flags |= TOF_SACKPERM;
3488 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3492 to->to_flags |= TOF_SACK;
3493 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3494 to->to_sacks = cp + 2;
3495 TCPSTAT_INC(tcps_sack_rcv_blocks);
3497 case TCPOPT_FAST_OPEN:
3499 * Cookie length validation is performed by the
3500 * server side cookie checking code or the client
3501 * side cookie cache update code.
3503 if (!(flags & TO_SYN))
3505 if (!V_tcp_fastopen_client_enable &&
3506 !V_tcp_fastopen_server_enable)
3508 to->to_flags |= TOF_FASTOPEN;
3509 to->to_tfo_len = optlen - 2;
3510 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3519 * Pull out of band byte out of a segment so
3520 * it doesn't appear in the user's data queue.
3521 * It is still reflected in the segment length for
3522 * sequencing purposes.
3525 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3528 int cnt = off + th->th_urp - 1;
3531 if (m->m_len > cnt) {
3532 char *cp = mtod(m, caddr_t) + cnt;
3533 struct tcpcb *tp = sototcpcb(so);
3535 INP_WLOCK_ASSERT(tp->t_inpcb);
3538 tp->t_oobflags |= TCPOOB_HAVEDATA;
3539 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3541 if (m->m_flags & M_PKTHDR)
3550 panic("tcp_pulloutofband");
3554 * Collect new round-trip time estimate
3555 * and update averages and current timeout.
3558 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3562 INP_WLOCK_ASSERT(tp->t_inpcb);
3564 TCPSTAT_INC(tcps_rttupdated);
3567 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3568 imax(0, rtt * 1000 / hz));
3570 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3572 * srtt is stored as fixed point with 5 bits after the
3573 * binary point (i.e., scaled by 8). The following magic
3574 * is equivalent to the smoothing algorithm in rfc793 with
3575 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3576 * point). Adjust rtt to origin 0.
3578 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3579 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3581 if ((tp->t_srtt += delta) <= 0)
3585 * We accumulate a smoothed rtt variance (actually, a
3586 * smoothed mean difference), then set the retransmit
3587 * timer to smoothed rtt + 4 times the smoothed variance.
3588 * rttvar is stored as fixed point with 4 bits after the
3589 * binary point (scaled by 16). The following is
3590 * equivalent to rfc793 smoothing with an alpha of .75
3591 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3592 * rfc793's wired-in beta.
3596 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3597 if ((tp->t_rttvar += delta) <= 0)
3599 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3600 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3603 * No rtt measurement yet - use the unsmoothed rtt.
3604 * Set the variance to half the rtt (so our first
3605 * retransmit happens at 3*rtt).
3607 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3608 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3609 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3615 * the retransmit should happen at rtt + 4 * rttvar.
3616 * Because of the way we do the smoothing, srtt and rttvar
3617 * will each average +1/2 tick of bias. When we compute
3618 * the retransmit timer, we want 1/2 tick of rounding and
3619 * 1 extra tick because of +-1/2 tick uncertainty in the
3620 * firing of the timer. The bias will give us exactly the
3621 * 1.5 tick we need. But, because the bias is
3622 * statistical, we have to test that we don't drop below
3623 * the minimum feasible timer (which is 2 ticks).
3625 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3626 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3629 * We received an ack for a packet that wasn't retransmitted;
3630 * it is probably safe to discard any error indications we've
3631 * received recently. This isn't quite right, but close enough
3632 * for now (a route might have failed after we sent a segment,
3633 * and the return path might not be symmetrical).
3635 tp->t_softerror = 0;
3639 * Determine a reasonable value for maxseg size.
3640 * If the route is known, check route for mtu.
3641 * If none, use an mss that can be handled on the outgoing interface
3642 * without forcing IP to fragment. If no route is found, route has no mtu,
3643 * or the destination isn't local, use a default, hopefully conservative
3644 * size (usually 512 or the default IP max size, but no more than the mtu
3645 * of the interface), as we can't discover anything about intervening
3646 * gateways or networks. We also initialize the congestion/slow start
3647 * window to be a single segment if the destination isn't local.
3648 * While looking at the routing entry, we also initialize other path-dependent
3649 * parameters from pre-set or cached values in the routing entry.
3651 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3652 * IP options, e.g. IPSEC data, since length of this data may vary, and
3653 * thus it is calculated for every segment separately in tcp_output().
3655 * NOTE that this routine is only called when we process an incoming
3656 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3657 * settings are handled in tcp_mssopt().
3660 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3661 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3664 uint32_t maxmtu = 0;
3665 struct inpcb *inp = tp->t_inpcb;
3666 struct hc_metrics_lite metrics;
3668 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3669 size_t min_protoh = isipv6 ?
3670 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3671 sizeof (struct tcpiphdr);
3673 const size_t min_protoh = sizeof(struct tcpiphdr);
3676 INP_WLOCK_ASSERT(tp->t_inpcb);
3678 if (mtuoffer != -1) {
3679 KASSERT(offer == -1, ("%s: conflict", __func__));
3680 offer = mtuoffer - min_protoh;
3686 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3687 tp->t_maxseg = V_tcp_v6mssdflt;
3690 #if defined(INET) && defined(INET6)
3695 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3696 tp->t_maxseg = V_tcp_mssdflt;
3701 * No route to sender, stay with default mss and return.
3705 * In case we return early we need to initialize metrics
3706 * to a defined state as tcp_hc_get() would do for us
3707 * if there was no cache hit.
3709 if (metricptr != NULL)
3710 bzero(metricptr, sizeof(struct hc_metrics_lite));
3714 /* What have we got? */
3718 * Offer == 0 means that there was no MSS on the SYN
3719 * segment, in this case we use tcp_mssdflt as
3720 * already assigned to t_maxseg above.
3722 offer = tp->t_maxseg;
3727 * Offer == -1 means that we didn't receive SYN yet.
3733 * Prevent DoS attack with too small MSS. Round up
3734 * to at least minmss.
3736 offer = max(offer, V_tcp_minmss);
3740 * rmx information is now retrieved from tcp_hostcache.
3742 tcp_hc_get(&inp->inp_inc, &metrics);
3743 if (metricptr != NULL)
3744 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3747 * If there's a discovered mtu in tcp hostcache, use it.
3748 * Else, use the link mtu.
3750 if (metrics.rmx_mtu)
3751 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3755 mss = maxmtu - min_protoh;
3756 if (!V_path_mtu_discovery &&
3757 !in6_localaddr(&inp->in6p_faddr))
3758 mss = min(mss, V_tcp_v6mssdflt);
3761 #if defined(INET) && defined(INET6)
3766 mss = maxmtu - min_protoh;
3767 if (!V_path_mtu_discovery &&
3768 !in_localaddr(inp->inp_faddr))
3769 mss = min(mss, V_tcp_mssdflt);
3773 * XXX - The above conditional (mss = maxmtu - min_protoh)
3774 * probably violates the TCP spec.
3775 * The problem is that, since we don't know the
3776 * other end's MSS, we are supposed to use a conservative
3777 * default. But, if we do that, then MTU discovery will
3778 * never actually take place, because the conservative
3779 * default is much less than the MTUs typically seen
3780 * on the Internet today. For the moment, we'll sweep
3781 * this under the carpet.
3783 * The conservative default might not actually be a problem
3784 * if the only case this occurs is when sending an initial
3785 * SYN with options and data to a host we've never talked
3786 * to before. Then, they will reply with an MSS value which
3787 * will get recorded and the new parameters should get
3788 * recomputed. For Further Study.
3791 mss = min(mss, offer);
3794 * Sanity check: make sure that maxseg will be large
3795 * enough to allow some data on segments even if the
3796 * all the option space is used (40bytes). Otherwise
3797 * funny things may happen in tcp_output.
3799 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3807 tcp_mss(struct tcpcb *tp, int offer)
3813 struct hc_metrics_lite metrics;
3814 struct tcp_ifcap cap;
3816 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3818 bzero(&cap, sizeof(cap));
3819 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3825 * If there's a pipesize, change the socket buffer to that size,
3826 * don't change if sb_hiwat is different than default (then it
3827 * has been changed on purpose with setsockopt).
3828 * Make the socket buffers an integral number of mss units;
3829 * if the mss is larger than the socket buffer, decrease the mss.
3831 so = inp->inp_socket;
3832 SOCKBUF_LOCK(&so->so_snd);
3833 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3834 bufsize = metrics.rmx_sendpipe;
3836 bufsize = so->so_snd.sb_hiwat;
3840 bufsize = roundup(bufsize, mss);
3841 if (bufsize > sb_max)
3843 if (bufsize > so->so_snd.sb_hiwat)
3844 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3846 SOCKBUF_UNLOCK(&so->so_snd);
3848 * Sanity check: make sure that maxseg will be large
3849 * enough to allow some data on segments even if the
3850 * all the option space is used (40bytes). Otherwise
3851 * funny things may happen in tcp_output.
3853 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3855 tp->t_maxseg = max(mss, 64);
3857 SOCKBUF_LOCK(&so->so_rcv);
3858 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3859 bufsize = metrics.rmx_recvpipe;
3861 bufsize = so->so_rcv.sb_hiwat;
3862 if (bufsize > mss) {
3863 bufsize = roundup(bufsize, mss);
3864 if (bufsize > sb_max)
3866 if (bufsize > so->so_rcv.sb_hiwat)
3867 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3869 SOCKBUF_UNLOCK(&so->so_rcv);
3871 /* Check the interface for TSO capabilities. */
3872 if (cap.ifcap & CSUM_TSO) {
3873 tp->t_flags |= TF_TSO;
3874 tp->t_tsomax = cap.tsomax;
3875 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3876 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3881 * Determine the MSS option to send on an outgoing SYN.
3884 tcp_mssopt(struct in_conninfo *inc)
3887 uint32_t thcmtu = 0;
3888 uint32_t maxmtu = 0;
3891 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3894 if (inc->inc_flags & INC_ISIPV6) {
3895 mss = V_tcp_v6mssdflt;
3896 maxmtu = tcp_maxmtu6(inc, NULL);
3897 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3900 #if defined(INET) && defined(INET6)
3905 mss = V_tcp_mssdflt;
3906 maxmtu = tcp_maxmtu(inc, NULL);
3907 min_protoh = sizeof(struct tcpiphdr);
3910 #if defined(INET6) || defined(INET)
3911 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3914 if (maxmtu && thcmtu)
3915 mss = min(maxmtu, thcmtu) - min_protoh;
3916 else if (maxmtu || thcmtu)
3917 mss = max(maxmtu, thcmtu) - min_protoh;
3923 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th)
3925 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3926 int maxseg = tcp_maxseg(tp);
3928 INP_WLOCK_ASSERT(tp->t_inpcb);
3931 * Compute the amount of data that this ACK is indicating
3932 * (del_data) and an estimate of how many bytes are in the
3935 del_data = tp->sackhint.delivered_data;
3936 if (V_tcp_do_rfc6675_pipe)
3937 pipe = tcp_compute_pipe(tp);
3939 pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit;
3940 tp->sackhint.prr_delivered += del_data;
3942 * Proportional Rate Reduction
3944 if (pipe >= tp->snd_ssthresh) {
3945 if (tp->sackhint.recover_fs == 0)
3946 tp->sackhint.recover_fs =
3947 imax(1, tp->snd_nxt - tp->snd_una);
3948 snd_cnt = howmany((long)tp->sackhint.prr_delivered *
3949 tp->snd_ssthresh, tp->sackhint.recover_fs) -
3950 tp->sackhint.prr_out;
3952 if (V_tcp_do_prr_conservative)
3953 limit = tp->sackhint.prr_delivered -
3954 tp->sackhint.prr_out;
3956 limit = imax(tp->sackhint.prr_delivered -
3957 tp->sackhint.prr_out, del_data) +
3959 snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
3961 snd_cnt = imax(snd_cnt, 0) / maxseg;
3963 * Send snd_cnt new data into the network in response to this ack.
3964 * If there is going to be a SACK retransmission, adjust snd_cwnd
3967 if (IN_FASTRECOVERY(tp->t_flags)) {
3968 tp->snd_cwnd = imax(maxseg, tp->snd_nxt - tp->snd_recover +
3969 tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg));
3970 } else if (IN_CONGRECOVERY(tp->t_flags))
3971 tp->snd_cwnd = imax(maxseg, pipe - del_data +
3972 (snd_cnt * maxseg));
3976 * On a partial ack arrives, force the retransmission of the
3977 * next unacknowledged segment. Do not clear tp->t_dupacks.
3978 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3982 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3984 tcp_seq onxt = tp->snd_nxt;
3985 uint32_t ocwnd = tp->snd_cwnd;
3986 u_int maxseg = tcp_maxseg(tp);
3988 INP_WLOCK_ASSERT(tp->t_inpcb);
3990 tcp_timer_activate(tp, TT_REXMT, 0);
3992 tp->snd_nxt = th->th_ack;
3994 * Set snd_cwnd to one segment beyond acknowledged offset.
3995 * (tp->snd_una has not yet been updated when this function is called.)
3997 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3998 tp->t_flags |= TF_ACKNOW;
3999 (void) tp->t_fb->tfb_tcp_output(tp);
4000 tp->snd_cwnd = ocwnd;
4001 if (SEQ_GT(onxt, tp->snd_nxt))
4004 * Partial window deflation. Relies on fact that tp->snd_una
4007 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4008 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4011 tp->snd_cwnd += maxseg;
4015 tcp_compute_pipe(struct tcpcb *tp)
4017 return (tp->snd_max - tp->snd_una +
4018 tp->sackhint.sack_bytes_rexmit -
4019 tp->sackhint.sacked_bytes);
4023 tcp_compute_initwnd(uint32_t maxseg)
4026 * Calculate the Initial Window, also used as Restart Window
4028 * RFC5681 Section 3.1 specifies the default conservative values.
4029 * RFC3390 specifies slightly more aggressive values.
4030 * RFC6928 increases it to ten segments.
4031 * Support for user specified value for initial flight size.
4033 if (V_tcp_initcwnd_segments)
4034 return min(V_tcp_initcwnd_segments * maxseg,
4035 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4036 else if (V_tcp_do_rfc3390)
4037 return min(4 * maxseg, max(2 * maxseg, 4380));
4039 /* Per RFC5681 Section 3.1 */
4041 return (2 * maxseg);
4042 else if (maxseg > 1095)
4043 return (3 * maxseg);
4045 return (4 * maxseg);