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;
516 * Indicate whether this ack should be delayed. We can delay the ack if
517 * following conditions are met:
518 * - There is no delayed ack timer in progress.
519 * - Our last ack wasn't a 0-sized window. We never want to delay
520 * the ack that opens up a 0-sized window.
521 * - LRO wasn't used for this segment. We make sure by checking that the
522 * segment size is not larger than the MSS.
524 #define DELAY_ACK(tp, tlen) \
525 ((!tcp_timer_active(tp, TT_DELACK) && \
526 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
527 (tlen <= tp->t_maxseg) && \
528 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
531 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
533 INP_WLOCK_ASSERT(tp->t_inpcb);
535 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
536 switch (iptos & IPTOS_ECN_MASK) {
538 tp->ccv->flags |= CCF_IPHDR_CE;
544 case IPTOS_ECN_NOTECT:
545 tp->ccv->flags &= ~CCF_IPHDR_CE;
549 if (th->th_flags & TH_CWR)
550 tp->ccv->flags |= CCF_TCPHDR_CWR;
552 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
554 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
556 if (tp->ccv->flags & CCF_ACKNOW) {
557 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
558 tp->t_flags |= TF_ACKNOW;
564 * TCP input handling is split into multiple parts:
565 * tcp6_input is a thin wrapper around tcp_input for the extended
566 * ip6_protox[] call format in ip6_input
567 * tcp_input handles primary segment validation, inpcb lookup and
568 * SYN processing on listen sockets
569 * tcp_do_segment processes the ACK and text of the segment for
570 * establishing, established and closing connections
574 tcp6_input(struct mbuf **mp, int *offp, int proto)
577 struct in6_ifaddr *ia6;
581 if (m->m_len < *offp + sizeof(struct tcphdr)) {
582 m = m_pullup(m, *offp + sizeof(struct tcphdr));
585 TCPSTAT_INC(tcps_rcvshort);
586 return (IPPROTO_DONE);
591 * draft-itojun-ipv6-tcp-to-anycast
592 * better place to put this in?
594 ip6 = mtod(m, struct ip6_hdr *);
595 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
596 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
597 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
598 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
600 return (IPPROTO_DONE);
604 return (tcp_input(mp, offp, proto));
609 tcp_input(struct mbuf **mp, int *offp, int proto)
611 struct mbuf *m = *mp;
612 struct tcphdr *th = NULL;
613 struct ip *ip = NULL;
614 struct inpcb *inp = NULL;
615 struct tcpcb *tp = NULL;
616 struct socket *so = NULL;
627 int rstreason = 0; /* For badport_bandlim accounting purposes */
629 struct m_tag *fwd_tag = NULL;
631 struct ip6_hdr *ip6 = NULL;
634 const void *ip6 = NULL;
636 struct tcpopt to; /* options in this segment */
637 char *s = NULL; /* address and port logging */
640 * The size of tcp_saveipgen must be the size of the max ip header,
643 u_char tcp_saveipgen[IP6_HDR_LEN];
644 struct tcphdr tcp_savetcp;
651 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
658 TCPSTAT_INC(tcps_rcvtotal);
662 ip6 = mtod(m, struct ip6_hdr *);
663 th = (struct tcphdr *)((caddr_t)ip6 + off0);
664 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
665 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
666 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
667 th->th_sum = m->m_pkthdr.csum_data;
669 th->th_sum = in6_cksum_pseudo(ip6, tlen,
670 IPPROTO_TCP, m->m_pkthdr.csum_data);
671 th->th_sum ^= 0xffff;
673 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
675 TCPSTAT_INC(tcps_rcvbadsum);
680 * Be proactive about unspecified IPv6 address in source.
681 * As we use all-zero to indicate unbounded/unconnected pcb,
682 * unspecified IPv6 address can be used to confuse us.
684 * Note that packets with unspecified IPv6 destination is
685 * already dropped in ip6_input.
687 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
691 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
694 #if defined(INET) && defined(INET6)
700 * Get IP and TCP header together in first mbuf.
701 * Note: IP leaves IP header in first mbuf.
703 if (off0 > sizeof (struct ip)) {
705 off0 = sizeof(struct ip);
707 if (m->m_len < sizeof (struct tcpiphdr)) {
708 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
710 TCPSTAT_INC(tcps_rcvshort);
711 return (IPPROTO_DONE);
714 ip = mtod(m, struct ip *);
715 th = (struct tcphdr *)((caddr_t)ip + off0);
716 tlen = ntohs(ip->ip_len) - off0;
719 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
720 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
721 th->th_sum = m->m_pkthdr.csum_data;
723 th->th_sum = in_pseudo(ip->ip_src.s_addr,
725 htonl(m->m_pkthdr.csum_data + tlen +
727 th->th_sum ^= 0xffff;
729 struct ipovly *ipov = (struct ipovly *)ip;
732 * Checksum extended TCP header and data.
736 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
737 ipov->ih_len = htons(tlen);
738 th->th_sum = in_cksum(m, len);
739 /* Reset length for SDT probes. */
740 ip->ip_len = htons(len);
743 /* Re-initialization for later version check */
745 ip->ip_v = IPVERSION;
746 ip->ip_hl = off0 >> 2;
750 TCPSTAT_INC(tcps_rcvbadsum);
757 * Check that TCP offset makes sense,
758 * pull out TCP options and adjust length. XXX
760 off = th->th_off << 2;
761 if (off < sizeof (struct tcphdr) || off > tlen) {
762 TCPSTAT_INC(tcps_rcvbadoff);
765 tlen -= off; /* tlen is used instead of ti->ti_len */
766 if (off > sizeof (struct tcphdr)) {
769 if (m->m_len < off0 + off) {
770 m = m_pullup(m, off0 + off);
772 TCPSTAT_INC(tcps_rcvshort);
773 return (IPPROTO_DONE);
776 ip6 = mtod(m, struct ip6_hdr *);
777 th = (struct tcphdr *)((caddr_t)ip6 + off0);
780 #if defined(INET) && defined(INET6)
785 if (m->m_len < sizeof(struct ip) + off) {
786 if ((m = m_pullup(m, sizeof (struct ip) + off))
788 TCPSTAT_INC(tcps_rcvshort);
789 return (IPPROTO_DONE);
791 ip = mtod(m, struct ip *);
792 th = (struct tcphdr *)((caddr_t)ip + off0);
796 optlen = off - sizeof (struct tcphdr);
797 optp = (u_char *)(th + 1);
799 thflags = th->th_flags;
802 * Convert TCP protocol specific fields to host format.
804 tcp_fields_to_host(th);
807 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
809 drop_hdrlen = off0 + off;
812 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
816 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
818 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
821 #if defined(INET) && !defined(INET6)
822 (m->m_flags & M_IP_NEXTHOP)
825 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
829 if (isipv6 && fwd_tag != NULL) {
830 struct sockaddr_in6 *next_hop6;
832 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
834 * Transparently forwarded. Pretend to be the destination.
835 * Already got one like this?
837 inp = in6_pcblookup_mbuf(&V_tcbinfo,
838 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
839 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
842 * It's new. Try to find the ambushing socket.
843 * Because we've rewritten the destination address,
844 * any hardware-generated hash is ignored.
846 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
847 th->th_sport, &next_hop6->sin6_addr,
848 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
849 th->th_dport, INPLOOKUP_WILDCARD |
850 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
853 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
854 th->th_sport, &ip6->ip6_dst, th->th_dport,
855 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
856 m->m_pkthdr.rcvif, m);
859 #if defined(INET6) && defined(INET)
863 if (fwd_tag != NULL) {
864 struct sockaddr_in *next_hop;
866 next_hop = (struct sockaddr_in *)(fwd_tag+1);
868 * Transparently forwarded. Pretend to be the destination.
869 * already got one like this?
871 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
872 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
873 m->m_pkthdr.rcvif, m);
876 * It's new. Try to find the ambushing socket.
877 * Because we've rewritten the destination address,
878 * any hardware-generated hash is ignored.
880 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
881 th->th_sport, next_hop->sin_addr,
882 next_hop->sin_port ? ntohs(next_hop->sin_port) :
883 th->th_dport, INPLOOKUP_WILDCARD |
884 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
887 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
888 th->th_sport, ip->ip_dst, th->th_dport,
889 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
890 m->m_pkthdr.rcvif, m);
894 * If the INPCB does not exist then all data in the incoming
895 * segment is discarded and an appropriate RST is sent back.
896 * XXX MRT Send RST using which routing table?
900 * Log communication attempts to ports that are not
903 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
904 V_tcp_log_in_vain == 2) {
905 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
906 log(LOG_INFO, "%s; %s: Connection attempt "
907 "to closed port\n", s, __func__);
910 * When blackholing do not respond with a RST but
911 * completely ignore the segment and drop it.
913 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
917 rstreason = BANDLIM_RST_CLOSEDPORT;
920 INP_WLOCK_ASSERT(inp);
922 * While waiting for inp lock during the lookup, another thread
923 * can have dropped the inpcb, in which case we need to loop back
924 * and try to find a new inpcb to deliver to.
926 if (inp->inp_flags & INP_DROPPED) {
931 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
932 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
933 ((inp->inp_socket == NULL) ||
934 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
935 inp->inp_flowid = m->m_pkthdr.flowid;
936 inp->inp_flowtype = M_HASHTYPE_GET(m);
938 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
940 if (isipv6 && IPSEC_ENABLED(ipv6) &&
941 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
949 if (IPSEC_ENABLED(ipv4) &&
950 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
957 * Check the minimum TTL for socket.
959 if (inp->inp_ip_minttl != 0) {
962 if (inp->inp_ip_minttl > ip6->ip6_hlim)
966 if (inp->inp_ip_minttl > ip->ip_ttl)
971 * A previous connection in TIMEWAIT state is supposed to catch stray
972 * or duplicate segments arriving late. If this segment was a
973 * legitimate new connection attempt, the old INPCB gets removed and
974 * we can try again to find a listening socket.
976 * At this point, due to earlier optimism, we may hold only an inpcb
977 * lock, and not the inpcbinfo write lock. If so, we need to try to
978 * acquire it, or if that fails, acquire a reference on the inpcb,
979 * drop all locks, acquire a global write lock, and then re-acquire
980 * the inpcb lock. We may at that point discover that another thread
981 * has tried to free the inpcb, in which case we need to loop back
982 * and try to find a new inpcb to deliver to.
984 * XXXRW: It may be time to rethink timewait locking.
986 if (inp->inp_flags & INP_TIMEWAIT) {
987 tcp_dooptions(&to, optp, optlen,
988 (thflags & TH_SYN) ? TO_SYN : 0);
990 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
992 if (tcp_twcheck(inp, &to, th, m, tlen))
994 return (IPPROTO_DONE);
997 * The TCPCB may no longer exist if the connection is winding
998 * down or it is in the CLOSED state. Either way we drop the
999 * segment and send an appropriate response.
1001 tp = intotcpcb(inp);
1002 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1003 rstreason = BANDLIM_RST_CLOSEDPORT;
1008 if (tp->t_flags & TF_TOE) {
1009 tcp_offload_input(tp, m);
1010 m = NULL; /* consumed by the TOE driver */
1016 INP_WLOCK_ASSERT(inp);
1017 if (mac_inpcb_check_deliver(inp, m))
1020 so = inp->inp_socket;
1021 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1023 if (so->so_options & SO_DEBUG) {
1024 ostate = tp->t_state;
1027 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1030 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1033 #endif /* TCPDEBUG */
1035 * When the socket is accepting connections (the INPCB is in LISTEN
1036 * state) we look into the SYN cache if this is a new connection
1037 * attempt or the completion of a previous one.
1039 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1040 ("%s: so accepting but tp %p not listening", __func__, tp));
1041 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1042 struct in_conninfo inc;
1044 bzero(&inc, sizeof(inc));
1047 inc.inc_flags |= INC_ISIPV6;
1048 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1049 inc.inc_flags |= INC_IPV6MINMTU;
1050 inc.inc6_faddr = ip6->ip6_src;
1051 inc.inc6_laddr = ip6->ip6_dst;
1055 inc.inc_faddr = ip->ip_src;
1056 inc.inc_laddr = ip->ip_dst;
1058 inc.inc_fport = th->th_sport;
1059 inc.inc_lport = th->th_dport;
1060 inc.inc_fibnum = so->so_fibnum;
1063 * Check for an existing connection attempt in syncache if
1064 * the flag is only ACK. A successful lookup creates a new
1065 * socket appended to the listen queue in SYN_RECEIVED state.
1067 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1069 * Parse the TCP options here because
1070 * syncookies need access to the reflected
1073 tcp_dooptions(&to, optp, optlen, 0);
1075 * NB: syncache_expand() doesn't unlock
1076 * inp and tcpinfo locks.
1078 rstreason = syncache_expand(&inc, &to, th, &so, m);
1079 if (rstreason < 0) {
1081 * A failing TCP MD5 signature comparison
1082 * must result in the segment being dropped
1083 * and must not produce any response back
1087 } else if (rstreason == 0) {
1089 * No syncache entry or ACK was not
1090 * for our SYN/ACK. Send a RST.
1091 * NB: syncache did its own logging
1092 * of the failure cause.
1094 rstreason = BANDLIM_RST_OPENPORT;
1100 * We completed the 3-way handshake
1101 * but could not allocate a socket
1102 * either due to memory shortage,
1103 * listen queue length limits or
1104 * global socket limits. Send RST
1105 * or wait and have the remote end
1106 * retransmit the ACK for another
1109 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1110 log(LOG_DEBUG, "%s; %s: Listen socket: "
1111 "Socket allocation failed due to "
1112 "limits or memory shortage, %s\n",
1114 V_tcp_sc_rst_sock_fail ?
1115 "sending RST" : "try again");
1116 if (V_tcp_sc_rst_sock_fail) {
1117 rstreason = BANDLIM_UNLIMITED;
1123 * Socket is created in state SYN_RECEIVED.
1124 * Unlock the listen socket, lock the newly
1125 * created socket and update the tp variable.
1127 INP_WUNLOCK(inp); /* listen socket */
1128 inp = sotoinpcb(so);
1130 * New connection inpcb is already locked by
1131 * syncache_expand().
1133 INP_WLOCK_ASSERT(inp);
1134 tp = intotcpcb(inp);
1135 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1136 ("%s: ", __func__));
1138 * Process the segment and the data it
1139 * contains. tcp_do_segment() consumes
1140 * the mbuf chain and unlocks the inpcb.
1142 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1143 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1145 return (IPPROTO_DONE);
1148 * Segment flag validation for new connection attempts:
1150 * Our (SYN|ACK) response was rejected.
1151 * Check with syncache and remove entry to prevent
1154 * NB: syncache_chkrst does its own logging of failure
1157 if (thflags & TH_RST) {
1158 syncache_chkrst(&inc, th, m);
1162 * We can't do anything without SYN.
1164 if ((thflags & TH_SYN) == 0) {
1165 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1166 log(LOG_DEBUG, "%s; %s: Listen socket: "
1167 "SYN is missing, segment ignored\n",
1169 TCPSTAT_INC(tcps_badsyn);
1173 * (SYN|ACK) is bogus on a listen socket.
1175 if (thflags & TH_ACK) {
1176 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1177 log(LOG_DEBUG, "%s; %s: Listen socket: "
1178 "SYN|ACK invalid, segment rejected\n",
1180 syncache_badack(&inc); /* XXX: Not needed! */
1181 TCPSTAT_INC(tcps_badsyn);
1182 rstreason = BANDLIM_RST_OPENPORT;
1186 * If the drop_synfin option is enabled, drop all
1187 * segments with both the SYN and FIN bits set.
1188 * This prevents e.g. nmap from identifying the
1190 * XXX: Poor reasoning. nmap has other methods
1191 * and is constantly refining its stack detection
1193 * XXX: This is a violation of the TCP specification
1194 * and was used by RFC1644.
1196 if ((thflags & TH_FIN) && V_drop_synfin) {
1197 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1198 log(LOG_DEBUG, "%s; %s: Listen socket: "
1199 "SYN|FIN segment ignored (based on "
1200 "sysctl setting)\n", s, __func__);
1201 TCPSTAT_INC(tcps_badsyn);
1205 * Segment's flags are (SYN) or (SYN|FIN).
1207 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1208 * as they do not affect the state of the TCP FSM.
1209 * The data pointed to by TH_URG and th_urp is ignored.
1211 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1212 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1213 KASSERT(thflags & (TH_SYN),
1214 ("%s: Listen socket: TH_SYN not set", __func__));
1217 * If deprecated address is forbidden,
1218 * we do not accept SYN to deprecated interface
1219 * address to prevent any new inbound connection from
1220 * getting established.
1221 * When we do not accept SYN, we send a TCP RST,
1222 * with deprecated source address (instead of dropping
1223 * it). We compromise it as it is much better for peer
1224 * to send a RST, and RST will be the final packet
1227 * If we do not forbid deprecated addresses, we accept
1228 * the SYN packet. RFC2462 does not suggest dropping
1230 * If we decipher RFC2462 5.5.4, it says like this:
1231 * 1. use of deprecated addr with existing
1232 * communication is okay - "SHOULD continue to be
1234 * 2. use of it with new communication:
1235 * (2a) "SHOULD NOT be used if alternate address
1236 * with sufficient scope is available"
1237 * (2b) nothing mentioned otherwise.
1238 * Here we fall into (2b) case as we have no choice in
1239 * our source address selection - we must obey the peer.
1241 * The wording in RFC2462 is confusing, and there are
1242 * multiple description text for deprecated address
1243 * handling - worse, they are not exactly the same.
1244 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1246 if (isipv6 && !V_ip6_use_deprecated) {
1247 struct in6_ifaddr *ia6;
1249 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1251 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1252 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1253 log(LOG_DEBUG, "%s; %s: Listen socket: "
1254 "Connection attempt to deprecated "
1255 "IPv6 address rejected\n",
1257 rstreason = BANDLIM_RST_OPENPORT;
1263 * Basic sanity checks on incoming SYN requests:
1264 * Don't respond if the destination is a link layer
1265 * broadcast according to RFC1122 4.2.3.10, p. 104.
1266 * If it is from this socket it must be forged.
1267 * Don't respond if the source or destination is a
1268 * global or subnet broad- or multicast address.
1269 * Note that it is quite possible to receive unicast
1270 * link-layer packets with a broadcast IP address. Use
1271 * in_broadcast() to find them.
1273 if (m->m_flags & (M_BCAST|M_MCAST)) {
1274 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1275 log(LOG_DEBUG, "%s; %s: Listen socket: "
1276 "Connection attempt from broad- or multicast "
1277 "link layer address ignored\n", s, __func__);
1282 if (th->th_dport == th->th_sport &&
1283 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1284 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1285 log(LOG_DEBUG, "%s; %s: Listen socket: "
1286 "Connection attempt to/from self "
1287 "ignored\n", s, __func__);
1290 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1291 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1292 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1293 log(LOG_DEBUG, "%s; %s: Listen socket: "
1294 "Connection attempt from/to multicast "
1295 "address ignored\n", s, __func__);
1300 #if defined(INET) && defined(INET6)
1305 if (th->th_dport == th->th_sport &&
1306 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1307 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1308 log(LOG_DEBUG, "%s; %s: Listen socket: "
1309 "Connection attempt from/to self "
1310 "ignored\n", s, __func__);
1313 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1314 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1315 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1316 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1317 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1318 log(LOG_DEBUG, "%s; %s: Listen socket: "
1319 "Connection attempt from/to broad- "
1320 "or multicast address ignored\n",
1327 * SYN appears to be valid. Create compressed TCP state
1331 if (so->so_options & SO_DEBUG)
1332 tcp_trace(TA_INPUT, ostate, tp,
1333 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1335 TCP_PROBE3(debug__input, tp, th, m);
1336 tcp_dooptions(&to, optp, optlen, TO_SYN);
1337 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos))
1338 goto tfo_socket_result;
1341 * Entry added to syncache and mbuf consumed.
1342 * Only the listen socket is unlocked by syncache_add().
1344 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1345 return (IPPROTO_DONE);
1346 } else if (tp->t_state == TCPS_LISTEN) {
1348 * When a listen socket is torn down the SO_ACCEPTCONN
1349 * flag is removed first while connections are drained
1350 * from the accept queue in a unlock/lock cycle of the
1351 * ACCEPT_LOCK, opening a race condition allowing a SYN
1352 * attempt go through unhandled.
1356 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1357 if (tp->t_flags & TF_SIGNATURE) {
1358 tcp_dooptions(&to, optp, optlen, thflags);
1359 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1360 TCPSTAT_INC(tcps_sig_err_nosigopt);
1363 if (!TCPMD5_ENABLED() ||
1364 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1368 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1371 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1372 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1373 * the inpcb, and unlocks pcbinfo.
1375 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1376 return (IPPROTO_DONE);
1379 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1382 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1385 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1386 m = NULL; /* mbuf chain got consumed. */
1391 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1397 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1402 return (IPPROTO_DONE);
1406 * Automatic sizing of receive socket buffer. Often the send
1407 * buffer size is not optimally adjusted to the actual network
1408 * conditions at hand (delay bandwidth product). Setting the
1409 * buffer size too small limits throughput on links with high
1410 * bandwidth and high delay (eg. trans-continental/oceanic links).
1412 * On the receive side the socket buffer memory is only rarely
1413 * used to any significant extent. This allows us to be much
1414 * more aggressive in scaling the receive socket buffer. For
1415 * the case that the buffer space is actually used to a large
1416 * extent and we run out of kernel memory we can simply drop
1417 * the new segments; TCP on the sender will just retransmit it
1418 * later. Setting the buffer size too big may only consume too
1419 * much kernel memory if the application doesn't read() from
1420 * the socket or packet loss or reordering makes use of the
1423 * The criteria to step up the receive buffer one notch are:
1424 * 1. Application has not set receive buffer size with
1425 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1426 * 2. the number of bytes received during 1/2 of an sRTT
1427 * is at least 3/8 of the current socket buffer size.
1428 * 3. receive buffer size has not hit maximal automatic size;
1430 * If all of the criteria are met we increaset the socket buffer
1431 * by a 1/2 (bounded by the max). This allows us to keep ahead
1432 * of slow-start but also makes it so our peer never gets limited
1433 * by our rwnd which we then open up causing a burst.
1435 * This algorithm does two steps per RTT at most and only if
1436 * we receive a bulk stream w/o packet losses or reorderings.
1437 * Shrinking the buffer during idle times is not necessary as
1438 * it doesn't consume any memory when idle.
1440 * TODO: Only step up if the application is actually serving
1441 * the buffer to better manage the socket buffer resources.
1444 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1445 struct tcpcb *tp, int tlen)
1449 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1450 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1451 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1452 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1453 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1454 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1455 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1457 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1459 /* Start over with next RTT. */
1463 tp->rfbuf_cnt += tlen; /* add up */
1469 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1472 * Since tp might be gone if the session entered
1473 * the TIME_WAIT state before coming here, we need
1474 * to check if the socket is still connected.
1476 if ((so->so_state & SS_ISCONNECTED) == 0)
1478 INP_LOCK_ASSERT(tp->t_inpcb);
1479 if (tp->t_flags & TF_WAKESOR) {
1480 tp->t_flags &= ~TF_WAKESOR;
1481 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
1484 if (tp->t_flags & TF_WAKESOW) {
1485 tp->t_flags &= ~TF_WAKESOW;
1486 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
1492 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1493 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1495 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1496 int rstreason, todrop, win, incforsyn = 0;
1500 struct in_conninfo *inc;
1508 * The size of tcp_saveipgen must be the size of the max ip header,
1511 u_char tcp_saveipgen[IP6_HDR_LEN];
1512 struct tcphdr tcp_savetcp;
1515 thflags = th->th_flags;
1516 inc = &tp->t_inpcb->inp_inc;
1517 tp->sackhint.last_sack_ack = 0;
1519 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1522 INP_WLOCK_ASSERT(tp->t_inpcb);
1523 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1525 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1529 /* Save segment, if requested. */
1530 tcp_pcap_add(th, m, &(tp->t_inpkts));
1532 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1535 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1536 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1537 log(LOG_DEBUG, "%s; %s: "
1538 "SYN|FIN segment ignored (based on "
1539 "sysctl setting)\n", s, __func__);
1546 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1547 * check SEQ.ACK first.
1549 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1550 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1551 rstreason = BANDLIM_UNLIMITED;
1556 * Segment received on connection.
1557 * Reset idle time and keep-alive timer.
1558 * XXX: This should be done after segment
1559 * validation to ignore broken/spoofed segs.
1561 tp->t_rcvtime = ticks;
1564 * Scale up the window into a 32-bit value.
1565 * For the SYN_SENT state the scale is zero.
1567 tiwin = th->th_win << tp->snd_scale;
1569 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1573 * TCP ECN processing.
1575 if (tp->t_flags2 & TF2_ECN_PERMIT) {
1576 if (thflags & TH_CWR) {
1577 tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1578 tp->t_flags |= TF_ACKNOW;
1580 switch (iptos & IPTOS_ECN_MASK) {
1582 tp->t_flags2 |= TF2_ECN_SND_ECE;
1583 TCPSTAT_INC(tcps_ecn_ce);
1585 case IPTOS_ECN_ECT0:
1586 TCPSTAT_INC(tcps_ecn_ect0);
1588 case IPTOS_ECN_ECT1:
1589 TCPSTAT_INC(tcps_ecn_ect1);
1593 /* Process a packet differently from RFC3168. */
1594 cc_ecnpkt_handler(tp, th, iptos);
1596 /* Congestion experienced. */
1597 if (thflags & TH_ECE) {
1598 cc_cong_signal(tp, th, CC_ECN);
1603 * Parse options on any incoming segment.
1605 tcp_dooptions(&to, (u_char *)(th + 1),
1606 (th->th_off << 2) - sizeof(struct tcphdr),
1607 (thflags & TH_SYN) ? TO_SYN : 0);
1609 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1610 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1611 (to.to_flags & TOF_SIGNATURE) == 0) {
1612 TCPSTAT_INC(tcps_sig_err_sigopt);
1613 /* XXX: should drop? */
1617 * If echoed timestamp is later than the current time,
1618 * fall back to non RFC1323 RTT calculation. Normalize
1619 * timestamp if syncookies were used when this connection
1622 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1623 to.to_tsecr -= tp->ts_offset;
1624 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1626 else if (tp->t_flags & TF_PREVVALID &&
1627 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1628 cc_cong_signal(tp, th, CC_RTO_ERR);
1631 * Process options only when we get SYN/ACK back. The SYN case
1632 * for incoming connections is handled in tcp_syncache.
1633 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1634 * or <SYN,ACK>) segment itself is never scaled.
1635 * XXX this is traditional behavior, may need to be cleaned up.
1637 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1638 /* Handle parallel SYN for ECN */
1639 if (!(thflags & TH_ACK) &&
1640 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1641 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1642 tp->t_flags2 |= TF2_ECN_PERMIT;
1643 tp->t_flags2 |= TF2_ECN_SND_ECE;
1644 TCPSTAT_INC(tcps_ecn_shs);
1646 if ((to.to_flags & TOF_SCALE) &&
1647 (tp->t_flags & TF_REQ_SCALE)) {
1648 tp->t_flags |= TF_RCVD_SCALE;
1649 tp->snd_scale = to.to_wscale;
1651 tp->t_flags &= ~TF_REQ_SCALE;
1653 * Initial send window. It will be updated with
1654 * the next incoming segment to the scaled value.
1656 tp->snd_wnd = th->th_win;
1657 if ((to.to_flags & TOF_TS) &&
1658 (tp->t_flags & TF_REQ_TSTMP)) {
1659 tp->t_flags |= TF_RCVD_TSTMP;
1660 tp->ts_recent = to.to_tsval;
1661 tp->ts_recent_age = tcp_ts_getticks();
1663 tp->t_flags &= ~TF_REQ_TSTMP;
1664 if (to.to_flags & TOF_MSS)
1665 tcp_mss(tp, to.to_mss);
1666 if ((tp->t_flags & TF_SACK_PERMIT) &&
1667 (to.to_flags & TOF_SACKPERM) == 0)
1668 tp->t_flags &= ~TF_SACK_PERMIT;
1669 if (IS_FASTOPEN(tp->t_flags)) {
1670 if (to.to_flags & TOF_FASTOPEN) {
1673 if (to.to_flags & TOF_MSS)
1676 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1680 tcp_fastopen_update_cache(tp, mss,
1681 to.to_tfo_len, to.to_tfo_cookie);
1683 tcp_fastopen_disable_path(tp);
1688 * If timestamps were negotiated during SYN/ACK and a
1689 * segment without a timestamp is received, silently drop
1690 * the segment, unless it is a RST segment or missing timestamps are
1692 * See section 3.2 of RFC 7323.
1694 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1695 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1696 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1697 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1698 "segment processed normally\n",
1703 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1704 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1705 "segment silently dropped\n", s, __func__);
1712 * If timestamps were not negotiated during SYN/ACK and a
1713 * segment with a timestamp is received, ignore the
1714 * timestamp and process the packet normally.
1715 * See section 3.2 of RFC 7323.
1717 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1718 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1719 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1720 "segment processed normally\n", s, __func__);
1726 * Header prediction: check for the two common cases
1727 * of a uni-directional data xfer. If the packet has
1728 * no control flags, is in-sequence, the window didn't
1729 * change and we're not retransmitting, it's a
1730 * candidate. If the length is zero and the ack moved
1731 * forward, we're the sender side of the xfer. Just
1732 * free the data acked & wake any higher level process
1733 * that was blocked waiting for space. If the length
1734 * is non-zero and the ack didn't move, we're the
1735 * receiver side. If we're getting packets in-order
1736 * (the reassembly queue is empty), add the data to
1737 * the socket buffer and note that we need a delayed ack.
1738 * Make sure that the hidden state-flags are also off.
1739 * Since we check for TCPS_ESTABLISHED first, it can only
1742 if (tp->t_state == TCPS_ESTABLISHED &&
1743 th->th_seq == tp->rcv_nxt &&
1744 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1745 tp->snd_nxt == tp->snd_max &&
1746 tiwin && tiwin == tp->snd_wnd &&
1747 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1749 ((to.to_flags & TOF_TS) == 0 ||
1750 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1752 * If last ACK falls within this segment's sequence numbers,
1753 * record the timestamp.
1754 * NOTE that the test is modified according to the latest
1755 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1757 if ((to.to_flags & TOF_TS) != 0 &&
1758 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1759 tp->ts_recent_age = tcp_ts_getticks();
1760 tp->ts_recent = to.to_tsval;
1764 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1765 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1766 !IN_RECOVERY(tp->t_flags) &&
1767 (to.to_flags & TOF_SACK) == 0 &&
1768 TAILQ_EMPTY(&tp->snd_holes)) {
1770 * This is a pure ack for outstanding data.
1772 TCPSTAT_INC(tcps_predack);
1775 * "bad retransmit" recovery without timestamps.
1777 if ((to.to_flags & TOF_TS) == 0 &&
1778 tp->t_rxtshift == 1 &&
1779 tp->t_flags & TF_PREVVALID &&
1780 (int)(ticks - tp->t_badrxtwin) < 0) {
1781 cc_cong_signal(tp, th, CC_RTO_ERR);
1785 * Recalculate the transmit timer / rtt.
1787 * Some boxes send broken timestamp replies
1788 * during the SYN+ACK phase, ignore
1789 * timestamps of 0 or we could calculate a
1790 * huge RTT and blow up the retransmit timer.
1792 if ((to.to_flags & TOF_TS) != 0 &&
1796 t = tcp_ts_getticks() - to.to_tsecr;
1797 if (!tp->t_rttlow || tp->t_rttlow > t)
1800 TCP_TS_TO_TICKS(t) + 1);
1801 } else if (tp->t_rtttime &&
1802 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1803 if (!tp->t_rttlow ||
1804 tp->t_rttlow > ticks - tp->t_rtttime)
1805 tp->t_rttlow = ticks - tp->t_rtttime;
1807 ticks - tp->t_rtttime);
1809 acked = BYTES_THIS_ACK(tp, th);
1812 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1813 hhook_run_tcp_est_in(tp, th, &to);
1816 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1817 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1818 sbdrop(&so->so_snd, acked);
1819 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1820 SEQ_LEQ(th->th_ack, tp->snd_recover))
1821 tp->snd_recover = th->th_ack - 1;
1824 * Let the congestion control algorithm update
1825 * congestion control related information. This
1826 * typically means increasing the congestion
1829 cc_ack_received(tp, th, nsegs, CC_ACK);
1831 tp->snd_una = th->th_ack;
1833 * Pull snd_wl2 up to prevent seq wrap relative
1836 tp->snd_wl2 = th->th_ack;
1841 * If all outstanding data are acked, stop
1842 * retransmit timer, otherwise restart timer
1843 * using current (possibly backed-off) value.
1844 * If process is waiting for space,
1845 * wakeup/selwakeup/signal. If data
1846 * are ready to send, let tcp_output
1847 * decide between more output or persist.
1850 if (so->so_options & SO_DEBUG)
1851 tcp_trace(TA_INPUT, ostate, tp,
1852 (void *)tcp_saveipgen,
1855 TCP_PROBE3(debug__input, tp, th, m);
1856 if (tp->snd_una == tp->snd_max)
1857 tcp_timer_activate(tp, TT_REXMT, 0);
1858 else if (!tcp_timer_active(tp, TT_PERSIST))
1859 tcp_timer_activate(tp, TT_REXMT,
1861 tp->t_flags |= TF_WAKESOW;
1862 if (sbavail(&so->so_snd))
1863 (void) tp->t_fb->tfb_tcp_output(tp);
1866 } else if (th->th_ack == tp->snd_una &&
1867 tlen <= sbspace(&so->so_rcv)) {
1868 int newsize = 0; /* automatic sockbuf scaling */
1871 * This is a pure, in-sequence data packet with
1872 * nothing on the reassembly queue and we have enough
1873 * buffer space to take it.
1875 /* Clean receiver SACK report if present */
1876 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1877 tcp_clean_sackreport(tp);
1878 TCPSTAT_INC(tcps_preddat);
1879 tp->rcv_nxt += tlen;
1881 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1882 (tp->t_fbyte_in == 0)) {
1883 tp->t_fbyte_in = ticks;
1884 if (tp->t_fbyte_in == 0)
1886 if (tp->t_fbyte_out && tp->t_fbyte_in)
1887 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1890 * Pull snd_wl1 up to prevent seq wrap relative to
1893 tp->snd_wl1 = th->th_seq;
1895 * Pull rcv_up up to prevent seq wrap relative to
1898 tp->rcv_up = tp->rcv_nxt;
1899 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1900 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1902 if (so->so_options & SO_DEBUG)
1903 tcp_trace(TA_INPUT, ostate, tp,
1904 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1906 TCP_PROBE3(debug__input, tp, th, m);
1908 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1910 /* Add data to socket buffer. */
1911 SOCKBUF_LOCK(&so->so_rcv);
1912 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1916 * Set new socket buffer size.
1917 * Give up when limit is reached.
1920 if (!sbreserve_locked(&so->so_rcv,
1922 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1923 m_adj(m, drop_hdrlen); /* delayed header drop */
1924 sbappendstream_locked(&so->so_rcv, m, 0);
1926 SOCKBUF_UNLOCK(&so->so_rcv);
1927 tp->t_flags |= TF_WAKESOR;
1928 if (DELAY_ACK(tp, tlen)) {
1929 tp->t_flags |= TF_DELACK;
1931 tp->t_flags |= TF_ACKNOW;
1932 tp->t_fb->tfb_tcp_output(tp);
1939 * Calculate amount of space in receive window,
1940 * and then do TCP input processing.
1941 * Receive window is amount of space in rcv queue,
1942 * but not less than advertised window.
1944 win = sbspace(&so->so_rcv);
1947 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1949 switch (tp->t_state) {
1951 * If the state is SYN_RECEIVED:
1952 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1954 case TCPS_SYN_RECEIVED:
1955 if ((thflags & TH_ACK) &&
1956 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1957 SEQ_GT(th->th_ack, tp->snd_max))) {
1958 rstreason = BANDLIM_RST_OPENPORT;
1961 if (IS_FASTOPEN(tp->t_flags)) {
1963 * When a TFO connection is in SYN_RECEIVED, the
1964 * only valid packets are the initial SYN, a
1965 * retransmit/copy of the initial SYN (possibly with
1966 * a subset of the original data), a valid ACK, a
1969 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1970 rstreason = BANDLIM_RST_OPENPORT;
1972 } else if (thflags & TH_SYN) {
1973 /* non-initial SYN is ignored */
1974 if ((tcp_timer_active(tp, TT_DELACK) ||
1975 tcp_timer_active(tp, TT_REXMT)))
1977 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1984 * If the state is SYN_SENT:
1985 * if seg contains a RST with valid ACK (SEQ.ACK has already
1986 * been verified), then drop the connection.
1987 * if seg contains a RST without an ACK, drop the seg.
1988 * if seg does not contain SYN, then drop the seg.
1989 * Otherwise this is an acceptable SYN segment
1990 * initialize tp->rcv_nxt and tp->irs
1991 * if seg contains ack then advance tp->snd_una
1992 * if seg contains an ECE and ECN support is enabled, the stream
1994 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1995 * arrange for segment to be acked (eventually)
1996 * continue processing rest of data/controls, beginning with URG
1999 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2000 TCP_PROBE5(connect__refused, NULL, tp,
2002 tp = tcp_drop(tp, ECONNREFUSED);
2004 if (thflags & TH_RST)
2006 if (!(thflags & TH_SYN))
2009 tp->irs = th->th_seq;
2011 if (thflags & TH_ACK) {
2012 int tfo_partial_ack = 0;
2014 TCPSTAT_INC(tcps_connects);
2017 mac_socketpeer_set_from_mbuf(m, so);
2019 /* Do window scaling on this connection? */
2020 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2021 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2022 tp->rcv_scale = tp->request_r_scale;
2024 tp->rcv_adv += min(tp->rcv_wnd,
2025 TCP_MAXWIN << tp->rcv_scale);
2026 tp->snd_una++; /* SYN is acked */
2028 * If not all the data that was sent in the TFO SYN
2029 * has been acked, resend the remainder right away.
2031 if (IS_FASTOPEN(tp->t_flags) &&
2032 (tp->snd_una != tp->snd_max)) {
2033 tp->snd_nxt = th->th_ack;
2034 tfo_partial_ack = 1;
2037 * If there's data, delay ACK; if there's also a FIN
2038 * ACKNOW will be turned on later.
2040 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2041 tcp_timer_activate(tp, TT_DELACK,
2044 tp->t_flags |= TF_ACKNOW;
2046 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2047 (V_tcp_do_ecn == 1)) {
2048 tp->t_flags2 |= TF2_ECN_PERMIT;
2049 TCPSTAT_INC(tcps_ecn_shs);
2053 * Received <SYN,ACK> in SYN_SENT[*] state.
2055 * SYN_SENT --> ESTABLISHED
2056 * SYN_SENT* --> FIN_WAIT_1
2058 tp->t_starttime = ticks;
2059 if (tp->t_flags & TF_NEEDFIN) {
2060 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2061 tp->t_flags &= ~TF_NEEDFIN;
2064 tcp_state_change(tp, TCPS_ESTABLISHED);
2065 TCP_PROBE5(connect__established, NULL, tp,
2068 tcp_timer_activate(tp, TT_KEEP,
2073 * Received initial SYN in SYN-SENT[*] state =>
2074 * simultaneous open.
2075 * If it succeeds, connection is * half-synchronized.
2076 * Otherwise, do 3-way handshake:
2077 * SYN-SENT -> SYN-RECEIVED
2078 * SYN-SENT* -> SYN-RECEIVED*
2080 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2081 tcp_timer_activate(tp, TT_REXMT, 0);
2082 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2085 INP_WLOCK_ASSERT(tp->t_inpcb);
2088 * Advance th->th_seq to correspond to first data byte.
2089 * If data, trim to stay within window,
2090 * dropping FIN if necessary.
2093 if (tlen > tp->rcv_wnd) {
2094 todrop = tlen - tp->rcv_wnd;
2098 TCPSTAT_INC(tcps_rcvpackafterwin);
2099 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2101 tp->snd_wl1 = th->th_seq - 1;
2102 tp->rcv_up = th->th_seq;
2104 * Client side of transaction: already sent SYN and data.
2105 * If the remote host used T/TCP to validate the SYN,
2106 * our data will be ACK'd; if so, enter normal data segment
2107 * processing in the middle of step 5, ack processing.
2108 * Otherwise, goto step 6.
2110 if (thflags & TH_ACK)
2116 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2117 * do normal processing.
2119 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2123 break; /* continue normal processing */
2127 * States other than LISTEN or SYN_SENT.
2128 * First check the RST flag and sequence number since reset segments
2129 * are exempt from the timestamp and connection count tests. This
2130 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2131 * below which allowed reset segments in half the sequence space
2132 * to fall though and be processed (which gives forged reset
2133 * segments with a random sequence number a 50 percent chance of
2134 * killing a connection).
2135 * Then check timestamp, if present.
2136 * Then check the connection count, if present.
2137 * Then check that at least some bytes of segment are within
2138 * receive window. If segment begins before rcv_nxt,
2139 * drop leading data (and SYN); if nothing left, just ack.
2141 if (thflags & TH_RST) {
2143 * RFC5961 Section 3.2
2145 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2146 * - If RST is in window, we send challenge ACK.
2148 * Note: to take into account delayed ACKs, we should
2149 * test against last_ack_sent instead of rcv_nxt.
2150 * Note 2: we handle special case of closed window, not
2151 * covered by the RFC.
2153 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2154 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2155 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2156 KASSERT(tp->t_state != TCPS_SYN_SENT,
2157 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2160 if (V_tcp_insecure_rst ||
2161 tp->last_ack_sent == th->th_seq) {
2162 TCPSTAT_INC(tcps_drops);
2163 /* Drop the connection. */
2164 switch (tp->t_state) {
2165 case TCPS_SYN_RECEIVED:
2166 so->so_error = ECONNREFUSED;
2168 case TCPS_ESTABLISHED:
2169 case TCPS_FIN_WAIT_1:
2170 case TCPS_FIN_WAIT_2:
2171 case TCPS_CLOSE_WAIT:
2174 so->so_error = ECONNRESET;
2181 TCPSTAT_INC(tcps_badrst);
2182 /* Send challenge ACK. */
2183 tcp_respond(tp, mtod(m, void *), th, m,
2184 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2185 tp->last_ack_sent = tp->rcv_nxt;
2193 * RFC5961 Section 4.2
2194 * Send challenge ACK for any SYN in synchronized state.
2196 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2197 tp->t_state != TCPS_SYN_RECEIVED) {
2198 TCPSTAT_INC(tcps_badsyn);
2199 if (V_tcp_insecure_syn &&
2200 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2201 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2202 tp = tcp_drop(tp, ECONNRESET);
2203 rstreason = BANDLIM_UNLIMITED;
2205 /* Send challenge ACK. */
2206 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2207 tp->snd_nxt, TH_ACK);
2208 tp->last_ack_sent = tp->rcv_nxt;
2215 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2216 * and it's less than ts_recent, drop it.
2218 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2219 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2220 /* Check to see if ts_recent is over 24 days old. */
2221 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2223 * Invalidate ts_recent. If this segment updates
2224 * ts_recent, the age will be reset later and ts_recent
2225 * will get a valid value. If it does not, setting
2226 * ts_recent to zero will at least satisfy the
2227 * requirement that zero be placed in the timestamp
2228 * echo reply when ts_recent isn't valid. The
2229 * age isn't reset until we get a valid ts_recent
2230 * because we don't want out-of-order segments to be
2231 * dropped when ts_recent is old.
2235 TCPSTAT_INC(tcps_rcvduppack);
2236 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2237 TCPSTAT_INC(tcps_pawsdrop);
2245 * In the SYN-RECEIVED state, validate that the packet belongs to
2246 * this connection before trimming the data to fit the receive
2247 * window. Check the sequence number versus IRS since we know
2248 * the sequence numbers haven't wrapped. This is a partial fix
2249 * for the "LAND" DoS attack.
2251 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2252 rstreason = BANDLIM_RST_OPENPORT;
2256 todrop = tp->rcv_nxt - th->th_seq;
2258 if (thflags & TH_SYN) {
2268 * Following if statement from Stevens, vol. 2, p. 960.
2271 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2273 * Any valid FIN must be to the left of the window.
2274 * At this point the FIN must be a duplicate or out
2275 * of sequence; drop it.
2280 * Send an ACK to resynchronize and drop any data.
2281 * But keep on processing for RST or ACK.
2283 tp->t_flags |= TF_ACKNOW;
2285 TCPSTAT_INC(tcps_rcvduppack);
2286 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2288 TCPSTAT_INC(tcps_rcvpartduppack);
2289 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2292 * DSACK - add SACK block for dropped range
2294 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2295 tcp_update_sack_list(tp, th->th_seq,
2296 th->th_seq + todrop);
2298 * ACK now, as the next in-sequence segment
2299 * will clear the DSACK block again
2301 tp->t_flags |= TF_ACKNOW;
2303 drop_hdrlen += todrop; /* drop from the top afterwards */
2304 th->th_seq += todrop;
2306 if (th->th_urp > todrop)
2307 th->th_urp -= todrop;
2315 * If new data are received on a connection after the
2316 * user processes are gone, then RST the other end.
2318 if ((so->so_state & SS_NOFDREF) &&
2319 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2320 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2321 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2322 "after socket was closed, "
2323 "sending RST and removing tcpcb\n",
2324 s, __func__, tcpstates[tp->t_state], tlen);
2328 TCPSTAT_INC(tcps_rcvafterclose);
2329 rstreason = BANDLIM_UNLIMITED;
2334 * If segment ends after window, drop trailing data
2335 * (and PUSH and FIN); if nothing left, just ACK.
2337 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2339 TCPSTAT_INC(tcps_rcvpackafterwin);
2340 if (todrop >= tlen) {
2341 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2343 * If window is closed can only take segments at
2344 * window edge, and have to drop data and PUSH from
2345 * incoming segments. Continue processing, but
2346 * remember to ack. Otherwise, drop segment
2349 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2350 tp->t_flags |= TF_ACKNOW;
2351 TCPSTAT_INC(tcps_rcvwinprobe);
2355 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2358 thflags &= ~(TH_PUSH|TH_FIN);
2362 * If last ACK falls within this segment's sequence numbers,
2363 * record its timestamp.
2365 * 1) That the test incorporates suggestions from the latest
2366 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2367 * 2) That updating only on newer timestamps interferes with
2368 * our earlier PAWS tests, so this check should be solely
2369 * predicated on the sequence space of this segment.
2370 * 3) That we modify the segment boundary check to be
2371 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2372 * instead of RFC1323's
2373 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2374 * This modified check allows us to overcome RFC1323's
2375 * limitations as described in Stevens TCP/IP Illustrated
2376 * Vol. 2 p.869. In such cases, we can still calculate the
2377 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2379 if ((to.to_flags & TOF_TS) != 0 &&
2380 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2381 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2382 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2383 tp->ts_recent_age = tcp_ts_getticks();
2384 tp->ts_recent = to.to_tsval;
2388 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2389 * flag is on (half-synchronized state), then queue data for
2390 * later processing; else drop segment and return.
2392 if ((thflags & TH_ACK) == 0) {
2393 if (tp->t_state == TCPS_SYN_RECEIVED ||
2394 (tp->t_flags & TF_NEEDSYN)) {
2395 if (tp->t_state == TCPS_SYN_RECEIVED &&
2396 IS_FASTOPEN(tp->t_flags)) {
2397 tp->snd_wnd = tiwin;
2401 } else if (tp->t_flags & TF_ACKNOW)
2410 switch (tp->t_state) {
2412 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2413 * ESTABLISHED state and continue processing.
2414 * The ACK was checked above.
2416 case TCPS_SYN_RECEIVED:
2418 TCPSTAT_INC(tcps_connects);
2420 /* Do window scaling? */
2421 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2422 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2423 tp->rcv_scale = tp->request_r_scale;
2425 tp->snd_wnd = tiwin;
2428 * SYN-RECEIVED -> ESTABLISHED
2429 * SYN-RECEIVED* -> FIN-WAIT-1
2431 tp->t_starttime = ticks;
2432 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2433 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2434 tp->t_tfo_pending = NULL;
2436 if (tp->t_flags & TF_NEEDFIN) {
2437 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2438 tp->t_flags &= ~TF_NEEDFIN;
2440 tcp_state_change(tp, TCPS_ESTABLISHED);
2441 TCP_PROBE5(accept__established, NULL, tp,
2444 * TFO connections call cc_conn_init() during SYN
2445 * processing. Calling it again here for such
2446 * connections is not harmless as it would undo the
2447 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2450 if (!IS_FASTOPEN(tp->t_flags))
2452 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2455 * Account for the ACK of our SYN prior to
2456 * regular ACK processing below, except for
2457 * simultaneous SYN, which is handled later.
2459 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2462 * If segment contains data or ACK, will call tcp_reass()
2463 * later; if not, do so now to pass queued data to user.
2465 if (tlen == 0 && (thflags & TH_FIN) == 0)
2466 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2468 tp->snd_wl1 = th->th_seq - 1;
2472 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2473 * ACKs. If the ack is in the range
2474 * tp->snd_una < th->th_ack <= tp->snd_max
2475 * then advance tp->snd_una to th->th_ack and drop
2476 * data from the retransmission queue. If this ACK reflects
2477 * more up to date window information we update our window information.
2479 case TCPS_ESTABLISHED:
2480 case TCPS_FIN_WAIT_1:
2481 case TCPS_FIN_WAIT_2:
2482 case TCPS_CLOSE_WAIT:
2485 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2486 TCPSTAT_INC(tcps_rcvacktoomuch);
2489 if ((tp->t_flags & TF_SACK_PERMIT) &&
2490 ((to.to_flags & TOF_SACK) ||
2491 !TAILQ_EMPTY(&tp->snd_holes)))
2492 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2495 * Reset the value so that previous (valid) value
2496 * from the last ack with SACK doesn't get used.
2498 tp->sackhint.sacked_bytes = 0;
2501 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2502 hhook_run_tcp_est_in(tp, th, &to);
2505 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2506 maxseg = tcp_maxseg(tp);
2508 (tiwin == tp->snd_wnd ||
2509 (tp->t_flags & TF_SACK_PERMIT))) {
2511 * If this is the first time we've seen a
2512 * FIN from the remote, this is not a
2513 * duplicate and it needs to be processed
2514 * normally. This happens during a
2515 * simultaneous close.
2517 if ((thflags & TH_FIN) &&
2518 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2522 TCPSTAT_INC(tcps_rcvdupack);
2524 * If we have outstanding data (other than
2525 * a window probe), this is a completely
2526 * duplicate ack (ie, window info didn't
2527 * change and FIN isn't set),
2528 * the ack is the biggest we've
2529 * seen and we've seen exactly our rexmt
2530 * threshold of them, assume a packet
2531 * has been dropped and retransmit it.
2532 * Kludge snd_nxt & the congestion
2533 * window so we send only this one
2536 * We know we're losing at the current
2537 * window size so do congestion avoidance
2538 * (set ssthresh to half the current window
2539 * and pull our congestion window back to
2540 * the new ssthresh).
2542 * Dup acks mean that packets have left the
2543 * network (they're now cached at the receiver)
2544 * so bump cwnd by the amount in the receiver
2545 * to keep a constant cwnd packets in the
2548 * When using TCP ECN, notify the peer that
2549 * we reduced the cwnd.
2552 * Following 2 kinds of acks should not affect
2555 * 2) Acks with SACK but without any new SACK
2556 * information in them. These could result from
2557 * any anomaly in the network like a switch
2558 * duplicating packets or a possible DoS attack.
2560 if (th->th_ack != tp->snd_una ||
2561 ((tp->t_flags & TF_SACK_PERMIT) &&
2564 else if (!tcp_timer_active(tp, TT_REXMT))
2566 else if (++tp->t_dupacks > tcprexmtthresh ||
2567 IN_FASTRECOVERY(tp->t_flags)) {
2568 cc_ack_received(tp, th, nsegs,
2571 IN_FASTRECOVERY(tp->t_flags) &&
2572 (tp->t_flags & TF_SACK_PERMIT)) {
2573 long snd_cnt = 0, limit = 0;
2574 long del_data = 0, pipe = 0;
2576 * In a duplicate ACK del_data is only the
2577 * diff_in_sack. If no SACK is used del_data
2578 * will be 0. Pipe is the amount of data we
2579 * estimate to be in the network.
2581 del_data = tp->sackhint.delivered_data;
2582 if (V_tcp_do_rfc6675_pipe)
2583 pipe = tcp_compute_pipe(tp);
2585 pipe = (tp->snd_nxt - tp->snd_fack) +
2586 tp->sackhint.sack_bytes_rexmit;
2587 tp->sackhint.prr_delivered += del_data;
2588 if (pipe > tp->snd_ssthresh) {
2589 if (tp->sackhint.recover_fs == 0)
2590 tp->sackhint.recover_fs =
2591 max(1, tp->snd_nxt - tp->snd_una);
2592 snd_cnt = (tp->sackhint.prr_delivered *
2594 tp->sackhint.recover_fs) +
2595 1 - tp->sackhint.sack_bytes_rexmit;
2597 if (V_tcp_do_prr_conservative)
2598 limit = tp->sackhint.prr_delivered -
2599 tp->sackhint.sack_bytes_rexmit;
2601 if ((tp->sackhint.prr_delivered -
2602 tp->sackhint.sack_bytes_rexmit) >
2604 limit = tp->sackhint.prr_delivered -
2605 tp->sackhint.sack_bytes_rexmit +
2608 limit = del_data + maxseg;
2609 if ((tp->snd_ssthresh - pipe) < limit)
2610 snd_cnt = tp->snd_ssthresh - pipe;
2614 snd_cnt = max((snd_cnt / maxseg), 0);
2616 * Send snd_cnt new data into the network in
2617 * response to this ACK. If there is a going
2618 * to be a SACK retransmission, adjust snd_cwnd
2621 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
2622 tp->sackhint.sack_bytes_rexmit +
2624 } else if ((tp->t_flags & TF_SACK_PERMIT) &&
2625 IN_FASTRECOVERY(tp->t_flags)) {
2629 * Compute the amount of data in flight first.
2630 * We can inject new data into the pipe iff
2631 * we have less than 1/2 the original window's
2632 * worth of data in flight.
2634 if (V_tcp_do_rfc6675_pipe)
2635 awnd = tcp_compute_pipe(tp);
2637 awnd = (tp->snd_nxt - tp->snd_fack) +
2638 tp->sackhint.sack_bytes_rexmit;
2640 if (awnd < tp->snd_ssthresh) {
2641 tp->snd_cwnd += maxseg;
2642 if (tp->snd_cwnd > tp->snd_ssthresh)
2643 tp->snd_cwnd = tp->snd_ssthresh;
2646 tp->snd_cwnd += maxseg;
2647 (void) tp->t_fb->tfb_tcp_output(tp);
2649 } else if (tp->t_dupacks == tcprexmtthresh ||
2650 (tp->t_flags & TF_SACK_PERMIT &&
2651 V_tcp_do_rfc6675_pipe &&
2652 tp->sackhint.sacked_bytes >
2653 (tcprexmtthresh - 1) * maxseg)) {
2656 * Above is the RFC6675 trigger condition of
2657 * more than (dupthresh-1)*maxseg sacked data.
2658 * If the count of holes in the
2659 * scoreboard is >= dupthresh, we could
2660 * also enter loss recovery, but don't
2661 * have that value readily available.
2663 tp->t_dupacks = tcprexmtthresh;
2664 tcp_seq onxt = tp->snd_nxt;
2667 * If we're doing sack, or prr, check
2668 * to see if we're already in sack
2669 * recovery. If we're not doing sack,
2670 * check to see if we're in newreno
2674 (tp->t_flags & TF_SACK_PERMIT)) {
2675 if (IN_FASTRECOVERY(tp->t_flags)) {
2680 if (SEQ_LEQ(th->th_ack,
2686 /* Congestion signal before ack. */
2687 cc_cong_signal(tp, th, CC_NDUPACK);
2688 cc_ack_received(tp, th, nsegs,
2690 tcp_timer_activate(tp, TT_REXMT, 0);
2694 * snd_ssthresh is already updated by
2697 tp->sackhint.prr_delivered = 0;
2698 tp->sackhint.sack_bytes_rexmit = 0;
2699 tp->sackhint.recover_fs = max(1,
2700 tp->snd_nxt - tp->snd_una);
2702 if (tp->t_flags & TF_SACK_PERMIT) {
2704 tcps_sack_recovery_episode);
2705 tp->snd_recover = tp->snd_nxt;
2706 tp->snd_cwnd = maxseg;
2707 (void) tp->t_fb->tfb_tcp_output(tp);
2708 if (SEQ_GT(th->th_ack, tp->snd_una))
2709 goto resume_partialack;
2712 tp->snd_nxt = th->th_ack;
2713 tp->snd_cwnd = maxseg;
2714 (void) tp->t_fb->tfb_tcp_output(tp);
2715 KASSERT(tp->snd_limited <= 2,
2716 ("%s: tp->snd_limited too big",
2718 tp->snd_cwnd = tp->snd_ssthresh +
2720 (tp->t_dupacks - tp->snd_limited);
2721 if (SEQ_GT(onxt, tp->snd_nxt))
2724 } else if (V_tcp_do_rfc3042) {
2726 * Process first and second duplicate
2727 * ACKs. Each indicates a segment
2728 * leaving the network, creating room
2729 * for more. Make sure we can send a
2730 * packet on reception of each duplicate
2731 * ACK by increasing snd_cwnd by one
2732 * segment. Restore the original
2733 * snd_cwnd after packet transmission.
2735 cc_ack_received(tp, th, nsegs,
2737 uint32_t oldcwnd = tp->snd_cwnd;
2738 tcp_seq oldsndmax = tp->snd_max;
2742 KASSERT(tp->t_dupacks == 1 ||
2744 ("%s: dupacks not 1 or 2",
2746 if (tp->t_dupacks == 1)
2747 tp->snd_limited = 0;
2749 (tp->snd_nxt - tp->snd_una) +
2750 (tp->t_dupacks - tp->snd_limited) *
2753 * Only call tcp_output when there
2754 * is new data available to be sent.
2755 * Otherwise we would send pure ACKs.
2757 SOCKBUF_LOCK(&so->so_snd);
2758 avail = sbavail(&so->so_snd) -
2759 (tp->snd_nxt - tp->snd_una);
2760 SOCKBUF_UNLOCK(&so->so_snd);
2762 (void) tp->t_fb->tfb_tcp_output(tp);
2763 sent = tp->snd_max - oldsndmax;
2764 if (sent > maxseg) {
2765 KASSERT((tp->t_dupacks == 2 &&
2766 tp->snd_limited == 0) ||
2767 (sent == maxseg + 1 &&
2768 tp->t_flags & TF_SENTFIN),
2769 ("%s: sent too much",
2771 tp->snd_limited = 2;
2772 } else if (sent > 0)
2774 tp->snd_cwnd = oldcwnd;
2781 * This ack is advancing the left edge, reset the
2786 * If this ack also has new SACK info, increment the
2787 * counter as per rfc6675. The variable
2788 * sack_changed tracks all changes to the SACK
2789 * scoreboard, including when partial ACKs without
2790 * SACK options are received, and clear the scoreboard
2791 * from the left side. Such partial ACKs should not be
2792 * counted as dupacks here.
2794 if ((tp->t_flags & TF_SACK_PERMIT) &&
2795 (to.to_flags & TOF_SACK) &&
2798 /* limit overhead by setting maxseg last */
2799 if (!IN_FASTRECOVERY(tp->t_flags) &&
2800 (tp->sackhint.sacked_bytes >
2801 ((tcprexmtthresh - 1) *
2802 (maxseg = tcp_maxseg(tp))))) {
2803 goto enter_recovery;
2809 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2810 ("%s: th_ack <= snd_una", __func__));
2813 * If the congestion window was inflated to account
2814 * for the other side's cached packets, retract it.
2816 if (IN_FASTRECOVERY(tp->t_flags)) {
2817 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2818 if (tp->t_flags & TF_SACK_PERMIT)
2819 if (V_tcp_do_prr && to.to_flags & TOF_SACK)
2820 tcp_prr_partialack(tp, th);
2822 tcp_sack_partialack(tp, th);
2824 tcp_newreno_partial_ack(tp, th);
2826 cc_post_recovery(tp, th);
2829 * If we reach this point, ACK is not a duplicate,
2830 * i.e., it ACKs something we sent.
2832 if (tp->t_flags & TF_NEEDSYN) {
2834 * T/TCP: Connection was half-synchronized, and our
2835 * SYN has been ACK'd (so connection is now fully
2836 * synchronized). Go to non-starred state,
2837 * increment snd_una for ACK of SYN, and check if
2838 * we can do window scaling.
2840 tp->t_flags &= ~TF_NEEDSYN;
2842 /* Do window scaling? */
2843 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2844 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2845 tp->rcv_scale = tp->request_r_scale;
2846 /* Send window already scaled. */
2851 INP_WLOCK_ASSERT(tp->t_inpcb);
2854 * Adjust for the SYN bit in sequence space,
2855 * but don't account for it in cwnd calculations.
2856 * This is for the SYN_RECEIVED, non-simultaneous
2857 * SYN case. SYN_SENT and simultaneous SYN are
2858 * treated elsewhere.
2862 acked = BYTES_THIS_ACK(tp, th);
2863 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2864 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2865 tp->snd_una, th->th_ack, tp, m));
2866 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2867 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2870 * If we just performed our first retransmit, and the ACK
2871 * arrives within our recovery window, then it was a mistake
2872 * to do the retransmit in the first place. Recover our
2873 * original cwnd and ssthresh, and proceed to transmit where
2876 if (tp->t_rxtshift == 1 &&
2877 tp->t_flags & TF_PREVVALID &&
2879 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2880 cc_cong_signal(tp, th, CC_RTO_ERR);
2883 * If we have a timestamp reply, update smoothed
2884 * round trip time. If no timestamp is present but
2885 * transmit timer is running and timed sequence
2886 * number was acked, update smoothed round trip time.
2887 * Since we now have an rtt measurement, cancel the
2888 * timer backoff (cf., Phil Karn's retransmit alg.).
2889 * Recompute the initial retransmit timer.
2891 * Some boxes send broken timestamp replies
2892 * during the SYN+ACK phase, ignore
2893 * timestamps of 0 or we could calculate a
2894 * huge RTT and blow up the retransmit timer.
2896 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2899 t = tcp_ts_getticks() - to.to_tsecr;
2900 if (!tp->t_rttlow || tp->t_rttlow > t)
2902 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2903 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2904 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2905 tp->t_rttlow = ticks - tp->t_rtttime;
2906 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2910 * If all outstanding data is acked, stop retransmit
2911 * timer and remember to restart (more output or persist).
2912 * If there is more data to be acked, restart retransmit
2913 * timer, using current (possibly backed-off) value.
2915 if (th->th_ack == tp->snd_max) {
2916 tcp_timer_activate(tp, TT_REXMT, 0);
2918 } else if (!tcp_timer_active(tp, TT_PERSIST))
2919 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2922 * If no data (only SYN) was ACK'd,
2923 * skip rest of ACK processing.
2929 * Let the congestion control algorithm update congestion
2930 * control related information. This typically means increasing
2931 * the congestion window.
2933 cc_ack_received(tp, th, nsegs, CC_ACK);
2935 SOCKBUF_LOCK(&so->so_snd);
2936 if (acked > sbavail(&so->so_snd)) {
2937 if (tp->snd_wnd >= sbavail(&so->so_snd))
2938 tp->snd_wnd -= sbavail(&so->so_snd);
2941 mfree = sbcut_locked(&so->so_snd,
2942 (int)sbavail(&so->so_snd));
2945 mfree = sbcut_locked(&so->so_snd, acked);
2946 if (tp->snd_wnd >= (uint32_t) acked)
2947 tp->snd_wnd -= acked;
2952 SOCKBUF_UNLOCK(&so->so_snd);
2953 tp->t_flags |= TF_WAKESOW;
2955 /* Detect una wraparound. */
2956 if (!IN_RECOVERY(tp->t_flags) &&
2957 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2958 SEQ_LEQ(th->th_ack, tp->snd_recover))
2959 tp->snd_recover = th->th_ack - 1;
2960 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2961 if (IN_RECOVERY(tp->t_flags) &&
2962 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2963 EXIT_RECOVERY(tp->t_flags);
2965 tp->snd_una = th->th_ack;
2966 if (tp->t_flags & TF_SACK_PERMIT) {
2967 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2968 tp->snd_recover = tp->snd_una;
2970 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2971 tp->snd_nxt = tp->snd_una;
2973 switch (tp->t_state) {
2975 * In FIN_WAIT_1 STATE in addition to the processing
2976 * for the ESTABLISHED state if our FIN is now acknowledged
2977 * then enter FIN_WAIT_2.
2979 case TCPS_FIN_WAIT_1:
2980 if (ourfinisacked) {
2982 * If we can't receive any more
2983 * data, then closing user can proceed.
2984 * Starting the timer is contrary to the
2985 * specification, but if we don't get a FIN
2986 * we'll hang forever.
2989 * we should release the tp also, and use a
2992 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2993 soisdisconnected(so);
2994 tcp_timer_activate(tp, TT_2MSL,
2995 (tcp_fast_finwait2_recycle ?
2996 tcp_finwait2_timeout :
2999 tcp_state_change(tp, TCPS_FIN_WAIT_2);
3004 * In CLOSING STATE in addition to the processing for
3005 * the ESTABLISHED state if the ACK acknowledges our FIN
3006 * then enter the TIME-WAIT state, otherwise ignore
3010 if (ourfinisacked) {
3018 * In LAST_ACK, we may still be waiting for data to drain
3019 * and/or to be acked, as well as for the ack of our FIN.
3020 * If our FIN is now acknowledged, delete the TCB,
3021 * enter the closed state and return.
3024 if (ourfinisacked) {
3033 INP_WLOCK_ASSERT(tp->t_inpcb);
3036 * Update window information.
3037 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3039 if ((thflags & TH_ACK) &&
3040 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3041 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3042 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3043 /* keep track of pure window updates */
3045 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3046 TCPSTAT_INC(tcps_rcvwinupd);
3047 tp->snd_wnd = tiwin;
3048 tp->snd_wl1 = th->th_seq;
3049 tp->snd_wl2 = th->th_ack;
3050 if (tp->snd_wnd > tp->max_sndwnd)
3051 tp->max_sndwnd = tp->snd_wnd;
3056 * Process segments with URG.
3058 if ((thflags & TH_URG) && th->th_urp &&
3059 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3061 * This is a kludge, but if we receive and accept
3062 * random urgent pointers, we'll crash in
3063 * soreceive. It's hard to imagine someone
3064 * actually wanting to send this much urgent data.
3066 SOCKBUF_LOCK(&so->so_rcv);
3067 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3068 th->th_urp = 0; /* XXX */
3069 thflags &= ~TH_URG; /* XXX */
3070 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3071 goto dodata; /* XXX */
3074 * If this segment advances the known urgent pointer,
3075 * then mark the data stream. This should not happen
3076 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3077 * a FIN has been received from the remote side.
3078 * In these states we ignore the URG.
3080 * According to RFC961 (Assigned Protocols),
3081 * the urgent pointer points to the last octet
3082 * of urgent data. We continue, however,
3083 * to consider it to indicate the first octet
3084 * of data past the urgent section as the original
3085 * spec states (in one of two places).
3087 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3088 tp->rcv_up = th->th_seq + th->th_urp;
3089 so->so_oobmark = sbavail(&so->so_rcv) +
3090 (tp->rcv_up - tp->rcv_nxt) - 1;
3091 if (so->so_oobmark == 0)
3092 so->so_rcv.sb_state |= SBS_RCVATMARK;
3094 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3096 SOCKBUF_UNLOCK(&so->so_rcv);
3098 * Remove out of band data so doesn't get presented to user.
3099 * This can happen independent of advancing the URG pointer,
3100 * but if two URG's are pending at once, some out-of-band
3101 * data may creep in... ick.
3103 if (th->th_urp <= (uint32_t)tlen &&
3104 !(so->so_options & SO_OOBINLINE)) {
3105 /* hdr drop is delayed */
3106 tcp_pulloutofband(so, th, m, drop_hdrlen);
3110 * If no out of band data is expected,
3111 * pull receive urgent pointer along
3112 * with the receive window.
3114 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3115 tp->rcv_up = tp->rcv_nxt;
3118 INP_WLOCK_ASSERT(tp->t_inpcb);
3121 * Process the segment text, merging it into the TCP sequencing queue,
3122 * and arranging for acknowledgment of receipt if necessary.
3123 * This process logically involves adjusting tp->rcv_wnd as data
3124 * is presented to the user (this happens in tcp_usrreq.c,
3125 * case PRU_RCVD). If a FIN has already been received on this
3126 * connection then we just ignore the text.
3128 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3129 IS_FASTOPEN(tp->t_flags));
3130 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3131 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3132 tcp_seq save_start = th->th_seq;
3133 tcp_seq save_rnxt = tp->rcv_nxt;
3134 int save_tlen = tlen;
3135 m_adj(m, drop_hdrlen); /* delayed header drop */
3137 * Insert segment which includes th into TCP reassembly queue
3138 * with control block tp. Set thflags to whether reassembly now
3139 * includes a segment with FIN. This handles the common case
3140 * inline (segment is the next to be received on an established
3141 * connection, and the queue is empty), avoiding linkage into
3142 * and removal from the queue and repetition of various
3144 * Set DELACK for segments received in order, but ack
3145 * immediately when segments are out of order (so
3146 * fast retransmit can work).
3148 if (th->th_seq == tp->rcv_nxt &&
3150 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3152 if (DELAY_ACK(tp, tlen) || tfo_syn)
3153 tp->t_flags |= TF_DELACK;
3155 tp->t_flags |= TF_ACKNOW;
3156 tp->rcv_nxt += tlen;
3158 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3159 (tp->t_fbyte_in == 0)) {
3160 tp->t_fbyte_in = ticks;
3161 if (tp->t_fbyte_in == 0)
3163 if (tp->t_fbyte_out && tp->t_fbyte_in)
3164 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3166 thflags = th->th_flags & TH_FIN;
3167 TCPSTAT_INC(tcps_rcvpack);
3168 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3169 SOCKBUF_LOCK(&so->so_rcv);
3170 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3173 sbappendstream_locked(&so->so_rcv, m, 0);
3174 SOCKBUF_UNLOCK(&so->so_rcv);
3175 tp->t_flags |= TF_WAKESOR;
3178 * XXX: Due to the header drop above "th" is
3179 * theoretically invalid by now. Fortunately
3180 * m_adj() doesn't actually frees any mbufs
3181 * when trimming from the head.
3183 tcp_seq temp = save_start;
3184 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3185 tp->t_flags |= TF_ACKNOW;
3187 if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) {
3188 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3190 * DSACK actually handled in the fastpath
3193 tcp_update_sack_list(tp, save_start,
3194 save_start + save_tlen);
3195 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3196 if ((tp->rcv_numsacks >= 1) &&
3197 (tp->sackblks[0].end == save_start)) {
3199 * Partial overlap, recorded at todrop
3202 tcp_update_sack_list(tp,
3203 tp->sackblks[0].start,
3204 tp->sackblks[0].end);
3206 tcp_update_dsack_list(tp, save_start,
3207 save_start + save_tlen);
3209 } else if (tlen >= save_tlen) {
3210 /* Update of sackblks. */
3211 tcp_update_dsack_list(tp, save_start,
3212 save_start + save_tlen);
3213 } else if (tlen > 0) {
3214 tcp_update_dsack_list(tp, save_start,
3220 * Note the amount of data that peer has sent into
3221 * our window, in order to estimate the sender's
3225 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3226 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3228 len = so->so_rcv.sb_hiwat;
3236 * If FIN is received ACK the FIN and let the user know
3237 * that the connection is closing.
3239 if (thflags & TH_FIN) {
3240 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3242 /* The socket upcall is handled by socantrcvmore. */
3243 tp->t_flags &= ~TF_WAKESOR;
3245 * If connection is half-synchronized
3246 * (ie NEEDSYN flag on) then delay ACK,
3247 * so it may be piggybacked when SYN is sent.
3248 * Otherwise, since we received a FIN then no
3249 * more input can be expected, send ACK now.
3251 if (tp->t_flags & TF_NEEDSYN)
3252 tp->t_flags |= TF_DELACK;
3254 tp->t_flags |= TF_ACKNOW;
3257 switch (tp->t_state) {
3259 * In SYN_RECEIVED and ESTABLISHED STATES
3260 * enter the CLOSE_WAIT state.
3262 case TCPS_SYN_RECEIVED:
3263 tp->t_starttime = ticks;
3265 case TCPS_ESTABLISHED:
3266 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3270 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3271 * enter the CLOSING state.
3273 case TCPS_FIN_WAIT_1:
3274 tcp_state_change(tp, TCPS_CLOSING);
3278 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3279 * starting the time-wait timer, turning off the other
3282 case TCPS_FIN_WAIT_2:
3288 if (so->so_options & SO_DEBUG)
3289 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3292 TCP_PROBE3(debug__input, tp, th, m);
3295 * Return any desired output.
3297 if (needoutput || (tp->t_flags & TF_ACKNOW))
3298 (void) tp->t_fb->tfb_tcp_output(tp);
3301 INP_WLOCK_ASSERT(tp->t_inpcb);
3303 if (tp->t_flags & TF_DELACK) {
3304 tp->t_flags &= ~TF_DELACK;
3305 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3307 tcp_handle_wakeup(tp, so);
3308 INP_WUNLOCK(tp->t_inpcb);
3313 * Generate an ACK dropping incoming segment if it occupies
3314 * sequence space, where the ACK reflects our state.
3316 * We can now skip the test for the RST flag since all
3317 * paths to this code happen after packets containing
3318 * RST have been dropped.
3320 * In the SYN-RECEIVED state, don't send an ACK unless the
3321 * segment we received passes the SYN-RECEIVED ACK test.
3322 * If it fails send a RST. This breaks the loop in the
3323 * "LAND" DoS attack, and also prevents an ACK storm
3324 * between two listening ports that have been sent forged
3325 * SYN segments, each with the source address of the other.
3327 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3328 (SEQ_GT(tp->snd_una, th->th_ack) ||
3329 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3330 rstreason = BANDLIM_RST_OPENPORT;
3334 if (so->so_options & SO_DEBUG)
3335 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3338 TCP_PROBE3(debug__input, tp, th, m);
3339 tp->t_flags |= TF_ACKNOW;
3340 (void) tp->t_fb->tfb_tcp_output(tp);
3341 tcp_handle_wakeup(tp, so);
3342 INP_WUNLOCK(tp->t_inpcb);
3348 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3349 tcp_handle_wakeup(tp, so);
3350 INP_WUNLOCK(tp->t_inpcb);
3352 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3357 * Drop space held by incoming segment and return.
3360 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3361 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3364 TCP_PROBE3(debug__input, tp, th, m);
3366 tcp_handle_wakeup(tp, so);
3367 INP_WUNLOCK(tp->t_inpcb);
3373 * Issue RST and make ACK acceptable to originator of segment.
3374 * The mbuf must still include the original packet header.
3378 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3379 int tlen, int rstreason)
3385 struct ip6_hdr *ip6;
3389 INP_WLOCK_ASSERT(tp->t_inpcb);
3392 /* Don't bother if destination was broadcast/multicast. */
3393 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3396 if (mtod(m, struct ip *)->ip_v == 6) {
3397 ip6 = mtod(m, struct ip6_hdr *);
3398 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3399 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3401 /* IPv6 anycast check is done at tcp6_input() */
3404 #if defined(INET) && defined(INET6)
3409 ip = mtod(m, struct ip *);
3410 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3411 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3412 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3413 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3418 /* Perform bandwidth limiting. */
3419 if (badport_bandlim(rstreason) < 0)
3422 /* tcp_respond consumes the mbuf chain. */
3423 if (th->th_flags & TH_ACK) {
3424 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3425 th->th_ack, TH_RST);
3427 if (th->th_flags & TH_SYN)
3429 if (th->th_flags & TH_FIN)
3431 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3432 (tcp_seq)0, TH_RST|TH_ACK);
3440 * Parse TCP options and place in tcpopt.
3443 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3448 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3450 if (opt == TCPOPT_EOL)
3452 if (opt == TCPOPT_NOP)
3458 if (optlen < 2 || optlen > cnt)
3463 if (optlen != TCPOLEN_MAXSEG)
3465 if (!(flags & TO_SYN))
3467 to->to_flags |= TOF_MSS;
3468 bcopy((char *)cp + 2,
3469 (char *)&to->to_mss, sizeof(to->to_mss));
3470 to->to_mss = ntohs(to->to_mss);
3473 if (optlen != TCPOLEN_WINDOW)
3475 if (!(flags & TO_SYN))
3477 to->to_flags |= TOF_SCALE;
3478 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3480 case TCPOPT_TIMESTAMP:
3481 if (optlen != TCPOLEN_TIMESTAMP)
3483 to->to_flags |= TOF_TS;
3484 bcopy((char *)cp + 2,
3485 (char *)&to->to_tsval, sizeof(to->to_tsval));
3486 to->to_tsval = ntohl(to->to_tsval);
3487 bcopy((char *)cp + 6,
3488 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3489 to->to_tsecr = ntohl(to->to_tsecr);
3491 case TCPOPT_SIGNATURE:
3493 * In order to reply to a host which has set the
3494 * TCP_SIGNATURE option in its initial SYN, we have
3495 * to record the fact that the option was observed
3496 * here for the syncache code to perform the correct
3499 if (optlen != TCPOLEN_SIGNATURE)
3501 to->to_flags |= TOF_SIGNATURE;
3502 to->to_signature = cp + 2;
3504 case TCPOPT_SACK_PERMITTED:
3505 if (optlen != TCPOLEN_SACK_PERMITTED)
3507 if (!(flags & TO_SYN))
3511 to->to_flags |= TOF_SACKPERM;
3514 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3518 to->to_flags |= TOF_SACK;
3519 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3520 to->to_sacks = cp + 2;
3521 TCPSTAT_INC(tcps_sack_rcv_blocks);
3523 case TCPOPT_FAST_OPEN:
3525 * Cookie length validation is performed by the
3526 * server side cookie checking code or the client
3527 * side cookie cache update code.
3529 if (!(flags & TO_SYN))
3531 if (!V_tcp_fastopen_client_enable &&
3532 !V_tcp_fastopen_server_enable)
3534 to->to_flags |= TOF_FASTOPEN;
3535 to->to_tfo_len = optlen - 2;
3536 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3545 * Pull out of band byte out of a segment so
3546 * it doesn't appear in the user's data queue.
3547 * It is still reflected in the segment length for
3548 * sequencing purposes.
3551 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3554 int cnt = off + th->th_urp - 1;
3557 if (m->m_len > cnt) {
3558 char *cp = mtod(m, caddr_t) + cnt;
3559 struct tcpcb *tp = sototcpcb(so);
3561 INP_WLOCK_ASSERT(tp->t_inpcb);
3564 tp->t_oobflags |= TCPOOB_HAVEDATA;
3565 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3567 if (m->m_flags & M_PKTHDR)
3576 panic("tcp_pulloutofband");
3580 * Collect new round-trip time estimate
3581 * and update averages and current timeout.
3584 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3588 INP_WLOCK_ASSERT(tp->t_inpcb);
3590 TCPSTAT_INC(tcps_rttupdated);
3593 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3594 imax(0, rtt * 1000 / hz));
3596 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3598 * srtt is stored as fixed point with 5 bits after the
3599 * binary point (i.e., scaled by 8). The following magic
3600 * is equivalent to the smoothing algorithm in rfc793 with
3601 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3602 * point). Adjust rtt to origin 0.
3604 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3605 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3607 if ((tp->t_srtt += delta) <= 0)
3611 * We accumulate a smoothed rtt variance (actually, a
3612 * smoothed mean difference), then set the retransmit
3613 * timer to smoothed rtt + 4 times the smoothed variance.
3614 * rttvar is stored as fixed point with 4 bits after the
3615 * binary point (scaled by 16). The following is
3616 * equivalent to rfc793 smoothing with an alpha of .75
3617 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3618 * rfc793's wired-in beta.
3622 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3623 if ((tp->t_rttvar += delta) <= 0)
3625 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3626 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3629 * No rtt measurement yet - use the unsmoothed rtt.
3630 * Set the variance to half the rtt (so our first
3631 * retransmit happens at 3*rtt).
3633 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3634 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3635 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3641 * the retransmit should happen at rtt + 4 * rttvar.
3642 * Because of the way we do the smoothing, srtt and rttvar
3643 * will each average +1/2 tick of bias. When we compute
3644 * the retransmit timer, we want 1/2 tick of rounding and
3645 * 1 extra tick because of +-1/2 tick uncertainty in the
3646 * firing of the timer. The bias will give us exactly the
3647 * 1.5 tick we need. But, because the bias is
3648 * statistical, we have to test that we don't drop below
3649 * the minimum feasible timer (which is 2 ticks).
3651 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3652 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3655 * We received an ack for a packet that wasn't retransmitted;
3656 * it is probably safe to discard any error indications we've
3657 * received recently. This isn't quite right, but close enough
3658 * for now (a route might have failed after we sent a segment,
3659 * and the return path might not be symmetrical).
3661 tp->t_softerror = 0;
3665 * Determine a reasonable value for maxseg size.
3666 * If the route is known, check route for mtu.
3667 * If none, use an mss that can be handled on the outgoing interface
3668 * without forcing IP to fragment. If no route is found, route has no mtu,
3669 * or the destination isn't local, use a default, hopefully conservative
3670 * size (usually 512 or the default IP max size, but no more than the mtu
3671 * of the interface), as we can't discover anything about intervening
3672 * gateways or networks. We also initialize the congestion/slow start
3673 * window to be a single segment if the destination isn't local.
3674 * While looking at the routing entry, we also initialize other path-dependent
3675 * parameters from pre-set or cached values in the routing entry.
3677 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3678 * IP options, e.g. IPSEC data, since length of this data may vary, and
3679 * thus it is calculated for every segment separately in tcp_output().
3681 * NOTE that this routine is only called when we process an incoming
3682 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3683 * settings are handled in tcp_mssopt().
3686 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3687 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3690 uint32_t maxmtu = 0;
3691 struct inpcb *inp = tp->t_inpcb;
3692 struct hc_metrics_lite metrics;
3694 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3695 size_t min_protoh = isipv6 ?
3696 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3697 sizeof (struct tcpiphdr);
3699 const size_t min_protoh = sizeof(struct tcpiphdr);
3702 INP_WLOCK_ASSERT(tp->t_inpcb);
3704 if (mtuoffer != -1) {
3705 KASSERT(offer == -1, ("%s: conflict", __func__));
3706 offer = mtuoffer - min_protoh;
3712 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3713 tp->t_maxseg = V_tcp_v6mssdflt;
3716 #if defined(INET) && defined(INET6)
3721 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3722 tp->t_maxseg = V_tcp_mssdflt;
3727 * No route to sender, stay with default mss and return.
3731 * In case we return early we need to initialize metrics
3732 * to a defined state as tcp_hc_get() would do for us
3733 * if there was no cache hit.
3735 if (metricptr != NULL)
3736 bzero(metricptr, sizeof(struct hc_metrics_lite));
3740 /* What have we got? */
3744 * Offer == 0 means that there was no MSS on the SYN
3745 * segment, in this case we use tcp_mssdflt as
3746 * already assigned to t_maxseg above.
3748 offer = tp->t_maxseg;
3753 * Offer == -1 means that we didn't receive SYN yet.
3759 * Prevent DoS attack with too small MSS. Round up
3760 * to at least minmss.
3762 offer = max(offer, V_tcp_minmss);
3766 * rmx information is now retrieved from tcp_hostcache.
3768 tcp_hc_get(&inp->inp_inc, &metrics);
3769 if (metricptr != NULL)
3770 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3773 * If there's a discovered mtu in tcp hostcache, use it.
3774 * Else, use the link mtu.
3776 if (metrics.rmx_mtu)
3777 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3781 mss = maxmtu - min_protoh;
3782 if (!V_path_mtu_discovery &&
3783 !in6_localaddr(&inp->in6p_faddr))
3784 mss = min(mss, V_tcp_v6mssdflt);
3787 #if defined(INET) && defined(INET6)
3792 mss = maxmtu - min_protoh;
3793 if (!V_path_mtu_discovery &&
3794 !in_localaddr(inp->inp_faddr))
3795 mss = min(mss, V_tcp_mssdflt);
3799 * XXX - The above conditional (mss = maxmtu - min_protoh)
3800 * probably violates the TCP spec.
3801 * The problem is that, since we don't know the
3802 * other end's MSS, we are supposed to use a conservative
3803 * default. But, if we do that, then MTU discovery will
3804 * never actually take place, because the conservative
3805 * default is much less than the MTUs typically seen
3806 * on the Internet today. For the moment, we'll sweep
3807 * this under the carpet.
3809 * The conservative default might not actually be a problem
3810 * if the only case this occurs is when sending an initial
3811 * SYN with options and data to a host we've never talked
3812 * to before. Then, they will reply with an MSS value which
3813 * will get recorded and the new parameters should get
3814 * recomputed. For Further Study.
3817 mss = min(mss, offer);
3820 * Sanity check: make sure that maxseg will be large
3821 * enough to allow some data on segments even if the
3822 * all the option space is used (40bytes). Otherwise
3823 * funny things may happen in tcp_output.
3825 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3833 tcp_mss(struct tcpcb *tp, int offer)
3839 struct hc_metrics_lite metrics;
3840 struct tcp_ifcap cap;
3842 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3844 bzero(&cap, sizeof(cap));
3845 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3851 * If there's a pipesize, change the socket buffer to that size,
3852 * don't change if sb_hiwat is different than default (then it
3853 * has been changed on purpose with setsockopt).
3854 * Make the socket buffers an integral number of mss units;
3855 * if the mss is larger than the socket buffer, decrease the mss.
3857 so = inp->inp_socket;
3858 SOCKBUF_LOCK(&so->so_snd);
3859 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3860 bufsize = metrics.rmx_sendpipe;
3862 bufsize = so->so_snd.sb_hiwat;
3866 bufsize = roundup(bufsize, mss);
3867 if (bufsize > sb_max)
3869 if (bufsize > so->so_snd.sb_hiwat)
3870 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3872 SOCKBUF_UNLOCK(&so->so_snd);
3874 * Sanity check: make sure that maxseg will be large
3875 * enough to allow some data on segments even if the
3876 * all the option space is used (40bytes). Otherwise
3877 * funny things may happen in tcp_output.
3879 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3881 tp->t_maxseg = max(mss, 64);
3883 SOCKBUF_LOCK(&so->so_rcv);
3884 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3885 bufsize = metrics.rmx_recvpipe;
3887 bufsize = so->so_rcv.sb_hiwat;
3888 if (bufsize > mss) {
3889 bufsize = roundup(bufsize, mss);
3890 if (bufsize > sb_max)
3892 if (bufsize > so->so_rcv.sb_hiwat)
3893 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3895 SOCKBUF_UNLOCK(&so->so_rcv);
3897 /* Check the interface for TSO capabilities. */
3898 if (cap.ifcap & CSUM_TSO) {
3899 tp->t_flags |= TF_TSO;
3900 tp->t_tsomax = cap.tsomax;
3901 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3902 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3907 * Determine the MSS option to send on an outgoing SYN.
3910 tcp_mssopt(struct in_conninfo *inc)
3913 uint32_t thcmtu = 0;
3914 uint32_t maxmtu = 0;
3917 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3920 if (inc->inc_flags & INC_ISIPV6) {
3921 mss = V_tcp_v6mssdflt;
3922 maxmtu = tcp_maxmtu6(inc, NULL);
3923 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3926 #if defined(INET) && defined(INET6)
3931 mss = V_tcp_mssdflt;
3932 maxmtu = tcp_maxmtu(inc, NULL);
3933 min_protoh = sizeof(struct tcpiphdr);
3936 #if defined(INET6) || defined(INET)
3937 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3940 if (maxmtu && thcmtu)
3941 mss = min(maxmtu, thcmtu) - min_protoh;
3942 else if (maxmtu || thcmtu)
3943 mss = max(maxmtu, thcmtu) - min_protoh;
3949 tcp_prr_partialack(struct tcpcb *tp, struct tcphdr *th)
3951 long snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3952 int maxseg = tcp_maxseg(tp);
3954 INP_WLOCK_ASSERT(tp->t_inpcb);
3956 tcp_timer_activate(tp, TT_REXMT, 0);
3959 * Compute the amount of data that this ACK is indicating
3960 * (del_data) and an estimate of how many bytes are in the
3963 if (SEQ_GEQ(th->th_ack, tp->snd_una))
3964 del_data = BYTES_THIS_ACK(tp, th);
3965 del_data += tp->sackhint.delivered_data;
3966 if (V_tcp_do_rfc6675_pipe)
3967 pipe = tcp_compute_pipe(tp);
3969 pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit;
3970 tp->sackhint.prr_delivered += del_data;
3972 * Proportional Rate Reduction
3974 if (pipe > tp->snd_ssthresh) {
3975 if (tp->sackhint.recover_fs == 0)
3976 tp->sackhint.recover_fs =
3977 max(1, tp->snd_nxt - tp->snd_una);
3978 snd_cnt = (tp->sackhint.prr_delivered * tp->snd_ssthresh /
3979 tp->sackhint.recover_fs) - tp->sackhint.sack_bytes_rexmit;
3981 if (V_tcp_do_prr_conservative)
3982 limit = tp->sackhint.prr_delivered -
3983 tp->sackhint.sack_bytes_rexmit;
3985 if ((tp->sackhint.prr_delivered -
3986 tp->sackhint.sack_bytes_rexmit) > del_data)
3987 limit = tp->sackhint.prr_delivered -
3988 tp->sackhint.sack_bytes_rexmit + maxseg;
3990 limit = del_data + maxseg;
3991 snd_cnt = min((tp->snd_ssthresh - pipe), limit);
3993 snd_cnt = max((snd_cnt / maxseg), 0);
3995 * Send snd_cnt new data into the network in response to this ack.
3996 * If there is going to be a SACK retransmission, adjust snd_cwnd
3999 tp->snd_cwnd = max(maxseg, (int64_t)tp->snd_nxt - tp->snd_recover +
4000 tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg));
4001 tp->t_flags |= TF_ACKNOW;
4002 (void) tcp_output(tp);
4006 * On a partial ack arrives, force the retransmission of the
4007 * next unacknowledged segment. Do not clear tp->t_dupacks.
4008 * By setting snd_nxt to ti_ack, this forces retransmission timer to
4012 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4014 tcp_seq onxt = tp->snd_nxt;
4015 uint32_t ocwnd = tp->snd_cwnd;
4016 u_int maxseg = tcp_maxseg(tp);
4018 INP_WLOCK_ASSERT(tp->t_inpcb);
4020 tcp_timer_activate(tp, TT_REXMT, 0);
4022 tp->snd_nxt = th->th_ack;
4024 * Set snd_cwnd to one segment beyond acknowledged offset.
4025 * (tp->snd_una has not yet been updated when this function is called.)
4027 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4028 tp->t_flags |= TF_ACKNOW;
4029 (void) tp->t_fb->tfb_tcp_output(tp);
4030 tp->snd_cwnd = ocwnd;
4031 if (SEQ_GT(onxt, tp->snd_nxt))
4034 * Partial window deflation. Relies on fact that tp->snd_una
4037 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4038 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4041 tp->snd_cwnd += maxseg;
4045 tcp_compute_pipe(struct tcpcb *tp)
4047 return (tp->snd_max - tp->snd_una +
4048 tp->sackhint.sack_bytes_rexmit -
4049 tp->sackhint.sacked_bytes);
4053 tcp_compute_initwnd(uint32_t maxseg)
4056 * Calculate the Initial Window, also used as Restart Window
4058 * RFC5681 Section 3.1 specifies the default conservative values.
4059 * RFC3390 specifies slightly more aggressive values.
4060 * RFC6928 increases it to ten segments.
4061 * Support for user specified value for initial flight size.
4063 if (V_tcp_initcwnd_segments)
4064 return min(V_tcp_initcwnd_segments * maxseg,
4065 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4066 else if (V_tcp_do_rfc3390)
4067 return min(4 * maxseg, max(2 * maxseg, 4380));
4069 /* Per RFC5681 Section 3.1 */
4071 return (2 * maxseg);
4072 else if (maxseg > 1095)
4073 return (3 * maxseg);
4075 return (4 * maxseg);