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>
120 #include <netinet/tcp_debug.h>
122 #include <netinet/tcp_offload.h>
124 #include <netinet/tcp_ecn.h>
125 #include <netinet/udp.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(bool, blackhole_local) = false;
147 #define V_blackhole_local VNET(blackhole_local)
148 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET |
149 CTLFLAG_RW, &VNET_NAME(blackhole_local), false,
150 "Enforce net.inet.tcp.blackhole for locally originated packets");
152 VNET_DEFINE(int, tcp_delack_enabled) = 1;
153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
154 &VNET_NAME(tcp_delack_enabled), 0,
155 "Delay ACK to try and piggyback it onto a data packet");
157 VNET_DEFINE(int, drop_synfin) = 0;
158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
159 &VNET_NAME(drop_synfin), 0,
160 "Drop TCP packets with SYN+FIN set");
162 VNET_DEFINE(int, tcp_do_prr_conservative) = 0;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW,
164 &VNET_NAME(tcp_do_prr_conservative), 0,
165 "Do conservative Proportional Rate Reduction");
167 VNET_DEFINE(int, tcp_do_prr) = 1;
168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW,
169 &VNET_NAME(tcp_do_prr), 1,
170 "Enable Proportional Rate Reduction per RFC 6937");
172 VNET_DEFINE(int, tcp_do_lrd) = 0;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW,
174 &VNET_NAME(tcp_do_lrd), 1,
175 "Perform Lost Retransmission Detection");
177 VNET_DEFINE(int, tcp_do_newcwv) = 0;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
179 &VNET_NAME(tcp_do_newcwv), 0,
180 "Enable New Congestion Window Validation per RFC7661");
182 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
184 &VNET_NAME(tcp_do_rfc3042), 0,
185 "Enable RFC 3042 (Limited Transmit)");
187 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
189 &VNET_NAME(tcp_do_rfc3390), 0,
190 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
192 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
194 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
195 "Slow-start flight size (initial congestion window) in number of segments");
197 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
199 &VNET_NAME(tcp_do_rfc3465), 0,
200 "Enable RFC 3465 (Appropriate Byte Counting)");
202 VNET_DEFINE(int, tcp_abc_l_var) = 2;
203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
204 &VNET_NAME(tcp_abc_l_var), 2,
205 "Cap the max cwnd increment during slow-start to this number of segments");
207 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn,
208 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
211 VNET_DEFINE(int, tcp_do_ecn) = 2;
212 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
213 &VNET_NAME(tcp_do_ecn), 0,
216 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
217 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
218 &VNET_NAME(tcp_ecn_maxretries), 0,
219 "Max retries before giving up on ECN");
221 VNET_DEFINE(int, tcp_insecure_syn) = 0;
222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
223 &VNET_NAME(tcp_insecure_syn), 0,
224 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
226 VNET_DEFINE(int, tcp_insecure_rst) = 0;
227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
228 &VNET_NAME(tcp_insecure_rst), 0,
229 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
231 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
232 #define V_tcp_recvspace VNET(tcp_recvspace)
233 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
234 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
236 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
237 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
238 &VNET_NAME(tcp_do_autorcvbuf), 0,
239 "Enable automatic receive buffer sizing");
241 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
242 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
243 &VNET_NAME(tcp_autorcvbuf_max), 0,
244 "Max size of automatic receive buffer");
246 VNET_DEFINE(struct inpcbinfo, tcbinfo);
249 * TCP statistics are stored in an array of counter(9)s, which size matches
250 * size of struct tcpstat. TCP running connection count is a regular array.
252 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
253 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
254 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
255 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
256 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
257 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
258 "TCP connection counts by TCP state");
261 * Kernel module interface for updating tcpstat. The first argument is an index
262 * into tcpstat treated as an array.
265 kmod_tcpstat_add(int statnum, int val)
268 counter_u64_add(VNET(tcpstat)[statnum], val);
272 * Make sure that we only start a SACK loss recovery when
273 * receiving a duplicate ACK with a SACK block, and also
274 * complete SACK loss recovery in case the other end
278 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to)
280 return ((tp->t_flags & TF_SACK_PERMIT) &&
281 ((to->to_flags & TOF_SACK) ||
282 (!TAILQ_EMPTY(&tp->snd_holes))));
287 * Wrapper for the TCP established input helper hook.
290 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
292 struct tcp_hhook_data hhook_data;
294 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
299 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
306 * CC wrapper hook functions
309 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
316 INP_WLOCK_ASSERT(tp->t_inpcb);
318 tp->ccv->nsegs = nsegs;
319 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
320 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
321 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
322 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
323 tp->ccv->flags |= CCF_CWND_LIMITED;
325 tp->ccv->flags &= ~CCF_CWND_LIMITED;
327 if (type == CC_ACK) {
329 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
330 ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
331 if (!IN_RECOVERY(tp->t_flags))
332 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
333 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
334 if ((tp->t_flags & TF_GPUTINPROG) &&
335 SEQ_GEQ(th->th_ack, tp->gput_ack)) {
337 * Compute goodput in bits per millisecond.
339 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) /
340 max(1, tcp_ts_getticks() - tp->gput_ts);
341 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
344 * XXXLAS: This is a temporary hack, and should be
345 * chained off VOI_TCP_GPUT when stats(9) grows an API
346 * to deal with chained VOIs.
348 if (tp->t_stats_gput_prev > 0)
349 stats_voi_update_abs_s32(tp->t_stats,
351 ((gput - tp->t_stats_gput_prev) * 100) /
352 tp->t_stats_gput_prev);
353 tp->t_flags &= ~TF_GPUTINPROG;
354 tp->t_stats_gput_prev = gput;
357 if (tp->snd_cwnd > tp->snd_ssthresh) {
358 tp->t_bytes_acked += tp->ccv->bytes_this_ack;
359 if (tp->t_bytes_acked >= tp->snd_cwnd) {
360 tp->t_bytes_acked -= tp->snd_cwnd;
361 tp->ccv->flags |= CCF_ABC_SENTAWND;
364 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
365 tp->t_bytes_acked = 0;
369 if (CC_ALGO(tp)->ack_received != NULL) {
370 /* XXXLAS: Find a way to live without this */
371 tp->ccv->curack = th->th_ack;
372 CC_ALGO(tp)->ack_received(tp->ccv, type);
375 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
380 cc_conn_init(struct tcpcb *tp)
382 struct hc_metrics_lite metrics;
383 struct inpcb *inp = tp->t_inpcb;
387 INP_WLOCK_ASSERT(tp->t_inpcb);
389 tcp_hc_get(&inp->inp_inc, &metrics);
390 maxseg = tcp_maxseg(tp);
392 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
394 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
395 TCPSTAT_INC(tcps_usedrtt);
396 if (metrics.rmx_rttvar) {
397 tp->t_rttvar = metrics.rmx_rttvar;
398 TCPSTAT_INC(tcps_usedrttvar);
400 /* default variation is +- 1 rtt */
402 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
404 TCPT_RANGESET(tp->t_rxtcur,
405 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
406 tp->t_rttmin, TCPTV_REXMTMAX);
408 if (metrics.rmx_ssthresh) {
410 * There's some sort of gateway or interface
411 * buffer limit on the path. Use this to set
412 * the slow start threshold, but set the
413 * threshold to no less than 2*mss.
415 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
416 TCPSTAT_INC(tcps_usedssthresh);
420 * Set the initial slow-start flight size.
422 * If a SYN or SYN/ACK was lost and retransmitted, we have to
423 * reduce the initial CWND to one segment as congestion is likely
424 * requiring us to be cautious.
426 if (tp->snd_cwnd == 1)
427 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
429 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
431 if (CC_ALGO(tp)->conn_init != NULL)
432 CC_ALGO(tp)->conn_init(tp->ccv);
436 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
438 INP_WLOCK_ASSERT(tp->t_inpcb);
441 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
446 if (!IN_FASTRECOVERY(tp->t_flags)) {
447 tp->snd_recover = tp->snd_max;
448 if (tp->t_flags2 & TF2_ECN_PERMIT)
449 tp->t_flags2 |= TF2_ECN_SND_CWR;
453 if (!IN_CONGRECOVERY(tp->t_flags) ||
455 * Allow ECN reaction on ACK to CWR, if
456 * that data segment was also CE marked.
458 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
459 EXIT_CONGRECOVERY(tp->t_flags);
460 TCPSTAT_INC(tcps_ecn_rcwnd);
461 tp->snd_recover = tp->snd_max + 1;
462 if (tp->t_flags2 & TF2_ECN_PERMIT)
463 tp->t_flags2 |= TF2_ECN_SND_CWR;
468 tp->t_bytes_acked = 0;
469 EXIT_RECOVERY(tp->t_flags);
470 if (tp->t_flags2 & TF2_ECN_PERMIT)
471 tp->t_flags2 |= TF2_ECN_SND_CWR;
474 TCPSTAT_INC(tcps_sndrexmitbad);
475 /* RTO was unnecessary, so reset everything. */
476 tp->snd_cwnd = tp->snd_cwnd_prev;
477 tp->snd_ssthresh = tp->snd_ssthresh_prev;
478 tp->snd_recover = tp->snd_recover_prev;
479 if (tp->t_flags & TF_WASFRECOVERY)
480 ENTER_FASTRECOVERY(tp->t_flags);
481 if (tp->t_flags & TF_WASCRECOVERY)
482 ENTER_CONGRECOVERY(tp->t_flags);
483 tp->snd_nxt = tp->snd_max;
484 tp->t_flags &= ~TF_PREVVALID;
489 if (CC_ALGO(tp)->cong_signal != NULL) {
491 tp->ccv->curack = th->th_ack;
492 CC_ALGO(tp)->cong_signal(tp->ccv, type);
497 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
499 INP_WLOCK_ASSERT(tp->t_inpcb);
501 /* XXXLAS: KASSERT that we're in recovery? */
503 if (CC_ALGO(tp)->post_recovery != NULL) {
504 tp->ccv->curack = th->th_ack;
505 CC_ALGO(tp)->post_recovery(tp->ccv);
507 /* XXXLAS: EXIT_RECOVERY ? */
508 tp->t_bytes_acked = 0;
509 tp->sackhint.delivered_data = 0;
510 tp->sackhint.prr_out = 0;
514 * Indicate whether this ack should be delayed. We can delay the ack if
515 * following conditions are met:
516 * - There is no delayed ack timer in progress.
517 * - Our last ack wasn't a 0-sized window. We never want to delay
518 * the ack that opens up a 0-sized window.
519 * - LRO wasn't used for this segment. We make sure by checking that the
520 * segment size is not larger than the MSS.
522 #define DELAY_ACK(tp, tlen) \
523 ((!tcp_timer_active(tp, TT_DELACK) && \
524 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
525 (tlen <= tp->t_maxseg) && \
526 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
529 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos)
531 INP_WLOCK_ASSERT(tp->t_inpcb);
533 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
534 switch (iptos & IPTOS_ECN_MASK) {
536 tp->ccv->flags |= CCF_IPHDR_CE;
542 case IPTOS_ECN_NOTECT:
543 tp->ccv->flags &= ~CCF_IPHDR_CE;
548 tp->ccv->flags |= CCF_TCPHDR_CWR;
550 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
552 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
554 if (tp->ccv->flags & CCF_ACKNOW) {
555 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
556 tp->t_flags |= TF_ACKNOW;
562 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
564 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos);
568 * TCP input handling is split into multiple parts:
569 * tcp6_input is a thin wrapper around tcp_input for the extended
570 * ip6_protox[] call format in ip6_input
571 * tcp_input handles primary segment validation, inpcb lookup and
572 * SYN processing on listen sockets
573 * tcp_do_segment processes the ACK and text of the segment for
574 * establishing, established and closing connections
578 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
581 struct in6_ifaddr *ia6;
585 if (m->m_len < *offp + sizeof(struct tcphdr)) {
586 m = m_pullup(m, *offp + sizeof(struct tcphdr));
589 TCPSTAT_INC(tcps_rcvshort);
590 return (IPPROTO_DONE);
595 * draft-itojun-ipv6-tcp-to-anycast
596 * better place to put this in?
598 ip6 = mtod(m, struct ip6_hdr *);
599 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
600 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
601 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
602 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
604 return (IPPROTO_DONE);
608 return (tcp_input_with_port(mp, offp, proto, port));
612 tcp6_input(struct mbuf **mp, int *offp, int proto)
615 return(tcp6_input_with_port(mp, offp, proto, 0));
620 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
622 struct mbuf *m = *mp;
623 struct tcphdr *th = NULL;
624 struct ip *ip = NULL;
625 struct inpcb *inp = NULL;
626 struct tcpcb *tp = NULL;
627 struct socket *so = NULL;
638 int rstreason = 0; /* For badport_bandlim accounting purposes */
641 struct m_tag *fwd_tag = NULL;
643 struct ip6_hdr *ip6 = NULL;
646 const void *ip6 = NULL;
648 struct tcpopt to; /* options in this segment */
649 char *s = NULL; /* address and port logging */
652 * The size of tcp_saveipgen must be the size of the max ip header,
655 u_char tcp_saveipgen[IP6_HDR_LEN];
656 struct tcphdr tcp_savetcp;
663 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
670 TCPSTAT_INC(tcps_rcvtotal);
674 ip6 = mtod(m, struct ip6_hdr *);
675 th = (struct tcphdr *)((caddr_t)ip6 + off0);
676 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
679 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
680 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
681 th->th_sum = m->m_pkthdr.csum_data;
683 th->th_sum = in6_cksum_pseudo(ip6, tlen,
684 IPPROTO_TCP, m->m_pkthdr.csum_data);
685 th->th_sum ^= 0xffff;
687 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
689 TCPSTAT_INC(tcps_rcvbadsum);
694 * Be proactive about unspecified IPv6 address in source.
695 * As we use all-zero to indicate unbounded/unconnected pcb,
696 * unspecified IPv6 address can be used to confuse us.
698 * Note that packets with unspecified IPv6 destination is
699 * already dropped in ip6_input.
701 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
705 iptos = IPV6_TRAFFIC_CLASS(ip6);
708 #if defined(INET) && defined(INET6)
714 * Get IP and TCP header together in first mbuf.
715 * Note: IP leaves IP header in first mbuf.
717 if (off0 > sizeof (struct ip)) {
719 off0 = sizeof(struct ip);
721 if (m->m_len < sizeof (struct tcpiphdr)) {
722 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
724 TCPSTAT_INC(tcps_rcvshort);
725 return (IPPROTO_DONE);
728 ip = mtod(m, struct ip *);
729 th = (struct tcphdr *)((caddr_t)ip + off0);
730 tlen = ntohs(ip->ip_len) - off0;
735 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
736 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
737 th->th_sum = m->m_pkthdr.csum_data;
739 th->th_sum = in_pseudo(ip->ip_src.s_addr,
741 htonl(m->m_pkthdr.csum_data + tlen +
743 th->th_sum ^= 0xffff;
745 struct ipovly *ipov = (struct ipovly *)ip;
748 * Checksum extended TCP header and data.
752 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
753 ipov->ih_len = htons(tlen);
754 th->th_sum = in_cksum(m, len);
755 /* Reset length for SDT probes. */
756 ip->ip_len = htons(len);
759 /* Re-initialization for later version check */
761 ip->ip_v = IPVERSION;
762 ip->ip_hl = off0 >> 2;
765 if (th->th_sum && (port == 0)) {
766 TCPSTAT_INC(tcps_rcvbadsum);
773 * Check that TCP offset makes sense,
774 * pull out TCP options and adjust length. XXX
776 off = th->th_off << 2;
777 if (off < sizeof (struct tcphdr) || off > tlen) {
778 TCPSTAT_INC(tcps_rcvbadoff);
781 tlen -= off; /* tlen is used instead of ti->ti_len */
782 if (off > sizeof (struct tcphdr)) {
785 if (m->m_len < off0 + off) {
786 m = m_pullup(m, off0 + off);
788 TCPSTAT_INC(tcps_rcvshort);
789 return (IPPROTO_DONE);
792 ip6 = mtod(m, struct ip6_hdr *);
793 th = (struct tcphdr *)((caddr_t)ip6 + off0);
796 #if defined(INET) && defined(INET6)
801 if (m->m_len < sizeof(struct ip) + off) {
802 if ((m = m_pullup(m, sizeof (struct ip) + off))
804 TCPSTAT_INC(tcps_rcvshort);
805 return (IPPROTO_DONE);
807 ip = mtod(m, struct ip *);
808 th = (struct tcphdr *)((caddr_t)ip + off0);
812 optlen = off - sizeof (struct tcphdr);
813 optp = (u_char *)(th + 1);
815 thflags = tcp_get_flags(th);
818 * Convert TCP protocol specific fields to host format.
820 tcp_fields_to_host(th);
823 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
825 drop_hdrlen = off0 + off;
828 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
832 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
834 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
837 #if defined(INET) && !defined(INET6)
838 (m->m_flags & M_IP_NEXTHOP)
841 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
844 * For initial SYN packets we don't need write lock on matching
845 * PCB, be it a listening one or a synchronized one. The packet
846 * shall not modify its state.
848 lookupflag = INPLOOKUP_WILDCARD |
849 ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ?
850 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB);
853 if (isipv6 && fwd_tag != NULL) {
854 struct sockaddr_in6 *next_hop6;
856 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
858 * Transparently forwarded. Pretend to be the destination.
859 * Already got one like this?
861 inp = in6_pcblookup_mbuf(&V_tcbinfo,
862 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
863 lookupflag, m->m_pkthdr.rcvif, m);
866 * It's new. Try to find the ambushing socket.
867 * Because we've rewritten the destination address,
868 * any hardware-generated hash is ignored.
870 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
871 th->th_sport, &next_hop6->sin6_addr,
872 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
873 th->th_dport, lookupflag, m->m_pkthdr.rcvif);
876 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
877 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag,
878 m->m_pkthdr.rcvif, m);
881 #if defined(INET6) && defined(INET)
885 if (fwd_tag != NULL) {
886 struct sockaddr_in *next_hop;
888 next_hop = (struct sockaddr_in *)(fwd_tag+1);
890 * Transparently forwarded. Pretend to be the destination.
891 * already got one like this?
893 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
894 ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m);
897 * It's new. Try to find the ambushing socket.
898 * Because we've rewritten the destination address,
899 * any hardware-generated hash is ignored.
901 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
902 th->th_sport, next_hop->sin_addr,
903 next_hop->sin_port ? ntohs(next_hop->sin_port) :
904 th->th_dport, lookupflag, m->m_pkthdr.rcvif);
907 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
908 th->th_sport, ip->ip_dst, th->th_dport, lookupflag,
909 m->m_pkthdr.rcvif, m);
913 * If the INPCB does not exist then all data in the incoming
914 * segment is discarded and an appropriate RST is sent back.
915 * XXX MRT Send RST using which routing table?
918 if (rstreason != 0) {
919 /* We came here after second (safety) lookup. */
920 MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0);
924 * Log communication attempts to ports that are not
927 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
928 V_tcp_log_in_vain == 2) {
929 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
930 log(LOG_INFO, "%s; %s: Connection attempt "
931 "to closed port\n", s, __func__);
934 * When blackholing do not respond with a RST but
935 * completely ignore the segment and drop it.
937 if (((V_blackhole == 1 && (thflags & TH_SYN)) ||
938 V_blackhole == 2) && (V_blackhole_local || (
940 isipv6 ? !in6_localaddr(&ip6->ip6_src) :
943 !in_localip(ip->ip_src)
950 rstreason = BANDLIM_RST_CLOSEDPORT;
953 INP_LOCK_ASSERT(inp);
955 * While waiting for inp lock during the lookup, another thread
956 * can have dropped the inpcb, in which case we need to loop back
957 * and try to find a new inpcb to deliver to.
959 if (inp->inp_flags & INP_DROPPED) {
964 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
965 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
966 ((inp->inp_socket == NULL) || !SOLISTENING(inp->inp_socket))) {
967 inp->inp_flowid = m->m_pkthdr.flowid;
968 inp->inp_flowtype = M_HASHTYPE_GET(m);
970 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
972 if (isipv6 && IPSEC_ENABLED(ipv6) &&
973 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
981 if (IPSEC_ENABLED(ipv4) &&
982 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
989 * Check the minimum TTL for socket.
991 if (inp->inp_ip_minttl != 0) {
994 if (inp->inp_ip_minttl > ip6->ip6_hlim)
998 if (inp->inp_ip_minttl > ip->ip_ttl)
1003 * A previous connection in TIMEWAIT state is supposed to catch stray
1004 * or duplicate segments arriving late. If this segment was a
1005 * legitimate new connection attempt, the old INPCB gets removed and
1006 * we can try again to find a listening socket.
1008 if (inp->inp_flags & INP_TIMEWAIT) {
1009 tcp_dooptions(&to, optp, optlen,
1010 (thflags & TH_SYN) ? TO_SYN : 0);
1012 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
1014 if (tcp_twcheck(inp, &to, th, m, tlen))
1016 return (IPPROTO_DONE);
1019 * The TCPCB may no longer exist if the connection is winding
1020 * down or it is in the CLOSED state. Either way we drop the
1021 * segment and send an appropriate response.
1023 tp = intotcpcb(inp);
1024 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1025 rstreason = BANDLIM_RST_CLOSEDPORT;
1029 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) {
1030 rstreason = BANDLIM_RST_CLOSEDPORT;
1035 if (tp->t_flags & TF_TOE) {
1036 tcp_offload_input(tp, m);
1037 m = NULL; /* consumed by the TOE driver */
1043 if (mac_inpcb_check_deliver(inp, m))
1046 so = inp->inp_socket;
1047 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1049 if (so->so_options & SO_DEBUG) {
1050 ostate = tp->t_state;
1053 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1056 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1059 #endif /* TCPDEBUG */
1061 * When the socket is accepting connections (the INPCB is in LISTEN
1062 * state) we look into the SYN cache if this is a new connection
1063 * attempt or the completion of a previous one.
1065 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so),
1066 ("%s: so accepting but tp %p not listening", __func__, tp));
1067 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) {
1068 struct in_conninfo inc;
1070 bzero(&inc, sizeof(inc));
1073 inc.inc_flags |= INC_ISIPV6;
1074 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1075 inc.inc_flags |= INC_IPV6MINMTU;
1076 inc.inc6_faddr = ip6->ip6_src;
1077 inc.inc6_laddr = ip6->ip6_dst;
1081 inc.inc_faddr = ip->ip_src;
1082 inc.inc_laddr = ip->ip_dst;
1084 inc.inc_fport = th->th_sport;
1085 inc.inc_lport = th->th_dport;
1086 inc.inc_fibnum = so->so_fibnum;
1089 * Check for an existing connection attempt in syncache if
1090 * the flag is only ACK. A successful lookup creates a new
1091 * socket appended to the listen queue in SYN_RECEIVED state.
1093 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1095 * Parse the TCP options here because
1096 * syncookies need access to the reflected
1099 tcp_dooptions(&to, optp, optlen, 0);
1101 * NB: syncache_expand() doesn't unlock inp.
1103 rstreason = syncache_expand(&inc, &to, th, &so, m, port);
1104 if (rstreason < 0) {
1106 * A failing TCP MD5 signature comparison
1107 * must result in the segment being dropped
1108 * and must not produce any response back
1112 } else if (rstreason == 0) {
1114 * No syncache entry, or ACK was not for our
1115 * SYN/ACK. Do our protection against double
1116 * ACK. If peer sent us 2 ACKs, then for the
1117 * first one syncache_expand() successfully
1118 * converted syncache entry into a socket,
1119 * while we were waiting on the inpcb lock. We
1120 * don't want to sent RST for the second ACK,
1121 * so we perform second lookup without wildcard
1122 * match, hoping to find the new socket. If
1123 * the ACK is stray indeed, rstreason would
1124 * hint the above code that the lookup was a
1127 * NB: syncache did its own logging
1128 * of the failure cause.
1131 rstreason = BANDLIM_RST_OPENPORT;
1132 lookupflag &= ~INPLOOKUP_WILDCARD;
1138 * We completed the 3-way handshake
1139 * but could not allocate a socket
1140 * either due to memory shortage,
1141 * listen queue length limits or
1142 * global socket limits. Send RST
1143 * or wait and have the remote end
1144 * retransmit the ACK for another
1147 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1148 log(LOG_DEBUG, "%s; %s: Listen socket: "
1149 "Socket allocation failed due to "
1150 "limits or memory shortage, %s\n",
1152 V_tcp_sc_rst_sock_fail ?
1153 "sending RST" : "try again");
1154 if (V_tcp_sc_rst_sock_fail) {
1155 rstreason = BANDLIM_UNLIMITED;
1161 * Socket is created in state SYN_RECEIVED.
1162 * Unlock the listen socket, lock the newly
1163 * created socket and update the tp variable.
1164 * If we came here via jump to tfo_socket_result,
1165 * then listening socket is read-locked.
1167 INP_UNLOCK(inp); /* listen socket */
1168 inp = sotoinpcb(so);
1170 * New connection inpcb is already locked by
1171 * syncache_expand().
1173 INP_WLOCK_ASSERT(inp);
1174 tp = intotcpcb(inp);
1175 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1176 ("%s: ", __func__));
1178 * Process the segment and the data it
1179 * contains. tcp_do_segment() consumes
1180 * the mbuf chain and unlocks the inpcb.
1182 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1183 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1185 return (IPPROTO_DONE);
1188 * Segment flag validation for new connection attempts:
1190 * Our (SYN|ACK) response was rejected.
1191 * Check with syncache and remove entry to prevent
1194 * NB: syncache_chkrst does its own logging of failure
1197 if (thflags & TH_RST) {
1198 syncache_chkrst(&inc, th, m, port);
1202 * We can't do anything without SYN.
1204 if ((thflags & TH_SYN) == 0) {
1205 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1206 log(LOG_DEBUG, "%s; %s: Listen socket: "
1207 "SYN is missing, segment ignored\n",
1209 TCPSTAT_INC(tcps_badsyn);
1213 * (SYN|ACK) is bogus on a listen socket.
1215 if (thflags & TH_ACK) {
1216 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1217 log(LOG_DEBUG, "%s; %s: Listen socket: "
1218 "SYN|ACK invalid, segment rejected\n",
1220 syncache_badack(&inc, port); /* XXX: Not needed! */
1221 TCPSTAT_INC(tcps_badsyn);
1222 rstreason = BANDLIM_RST_OPENPORT;
1226 * If the drop_synfin option is enabled, drop all
1227 * segments with both the SYN and FIN bits set.
1228 * This prevents e.g. nmap from identifying the
1230 * XXX: Poor reasoning. nmap has other methods
1231 * and is constantly refining its stack detection
1233 * XXX: This is a violation of the TCP specification
1234 * and was used by RFC1644.
1236 if ((thflags & TH_FIN) && V_drop_synfin) {
1237 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1238 log(LOG_DEBUG, "%s; %s: Listen socket: "
1239 "SYN|FIN segment ignored (based on "
1240 "sysctl setting)\n", s, __func__);
1241 TCPSTAT_INC(tcps_badsyn);
1245 * Segment's flags are (SYN) or (SYN|FIN).
1247 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1248 * as they do not affect the state of the TCP FSM.
1249 * The data pointed to by TH_URG and th_urp is ignored.
1251 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1252 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1253 KASSERT(thflags & (TH_SYN),
1254 ("%s: Listen socket: TH_SYN not set", __func__));
1255 INP_RLOCK_ASSERT(inp);
1258 * If deprecated address is forbidden,
1259 * we do not accept SYN to deprecated interface
1260 * address to prevent any new inbound connection from
1261 * getting established.
1262 * When we do not accept SYN, we send a TCP RST,
1263 * with deprecated source address (instead of dropping
1264 * it). We compromise it as it is much better for peer
1265 * to send a RST, and RST will be the final packet
1268 * If we do not forbid deprecated addresses, we accept
1269 * the SYN packet. RFC2462 does not suggest dropping
1271 * If we decipher RFC2462 5.5.4, it says like this:
1272 * 1. use of deprecated addr with existing
1273 * communication is okay - "SHOULD continue to be
1275 * 2. use of it with new communication:
1276 * (2a) "SHOULD NOT be used if alternate address
1277 * with sufficient scope is available"
1278 * (2b) nothing mentioned otherwise.
1279 * Here we fall into (2b) case as we have no choice in
1280 * our source address selection - we must obey the peer.
1282 * The wording in RFC2462 is confusing, and there are
1283 * multiple description text for deprecated address
1284 * handling - worse, they are not exactly the same.
1285 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1287 if (isipv6 && !V_ip6_use_deprecated) {
1288 struct in6_ifaddr *ia6;
1290 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1292 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1293 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1294 log(LOG_DEBUG, "%s; %s: Listen socket: "
1295 "Connection attempt to deprecated "
1296 "IPv6 address rejected\n",
1298 rstreason = BANDLIM_RST_OPENPORT;
1304 * Basic sanity checks on incoming SYN requests:
1305 * Don't respond if the destination is a link layer
1306 * broadcast according to RFC1122 4.2.3.10, p. 104.
1307 * If it is from this socket it must be forged.
1308 * Don't respond if the source or destination is a
1309 * global or subnet broad- or multicast address.
1310 * Note that it is quite possible to receive unicast
1311 * link-layer packets with a broadcast IP address. Use
1312 * in_broadcast() to find them.
1314 if (m->m_flags & (M_BCAST|M_MCAST)) {
1315 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1316 log(LOG_DEBUG, "%s; %s: Listen socket: "
1317 "Connection attempt from broad- or multicast "
1318 "link layer address ignored\n", s, __func__);
1323 if (th->th_dport == th->th_sport &&
1324 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1325 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1326 log(LOG_DEBUG, "%s; %s: Listen socket: "
1327 "Connection attempt to/from self "
1328 "ignored\n", s, __func__);
1331 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1332 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1333 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1334 log(LOG_DEBUG, "%s; %s: Listen socket: "
1335 "Connection attempt from/to multicast "
1336 "address ignored\n", s, __func__);
1341 #if defined(INET) && defined(INET6)
1346 if (th->th_dport == th->th_sport &&
1347 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1348 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1349 log(LOG_DEBUG, "%s; %s: Listen socket: "
1350 "Connection attempt from/to self "
1351 "ignored\n", s, __func__);
1354 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1355 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1356 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1357 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1358 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1359 log(LOG_DEBUG, "%s; %s: Listen socket: "
1360 "Connection attempt from/to broad- "
1361 "or multicast address ignored\n",
1368 * SYN appears to be valid. Create compressed TCP state
1372 if (so->so_options & SO_DEBUG)
1373 tcp_trace(TA_INPUT, ostate, tp,
1374 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1376 TCP_PROBE3(debug__input, tp, th, m);
1377 tcp_dooptions(&to, optp, optlen, TO_SYN);
1378 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL,
1379 iptos, port)) != NULL)
1380 goto tfo_socket_result;
1383 * Entry added to syncache and mbuf consumed.
1384 * Only the listen socket is unlocked by syncache_add().
1386 return (IPPROTO_DONE);
1387 } else if (tp->t_state == TCPS_LISTEN) {
1389 * When a listen socket is torn down the SO_ACCEPTCONN
1390 * flag is removed first while connections are drained
1391 * from the accept queue in a unlock/lock cycle of the
1392 * ACCEPT_LOCK, opening a race condition allowing a SYN
1393 * attempt go through unhandled.
1397 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1398 if (tp->t_flags & TF_SIGNATURE) {
1399 tcp_dooptions(&to, optp, optlen, thflags);
1400 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1401 TCPSTAT_INC(tcps_sig_err_nosigopt);
1404 if (!TCPMD5_ENABLED() ||
1405 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1409 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1412 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1413 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1414 * the inpcb, and unlocks pcbinfo.
1416 * XXXGL: in case of a pure SYN arriving on existing connection
1417 * TCP stacks won't need to modify the PCB, they would either drop
1418 * the segment silently, or send a challenge ACK. However, we try
1419 * to upgrade the lock, because calling convention for stacks is
1420 * write-lock on PCB. If upgrade fails, drop the SYN.
1422 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0)
1425 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1426 return (IPPROTO_DONE);
1429 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1432 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1435 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1436 m = NULL; /* mbuf chain got consumed. */
1441 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1451 return (IPPROTO_DONE);
1455 * Automatic sizing of receive socket buffer. Often the send
1456 * buffer size is not optimally adjusted to the actual network
1457 * conditions at hand (delay bandwidth product). Setting the
1458 * buffer size too small limits throughput on links with high
1459 * bandwidth and high delay (eg. trans-continental/oceanic links).
1461 * On the receive side the socket buffer memory is only rarely
1462 * used to any significant extent. This allows us to be much
1463 * more aggressive in scaling the receive socket buffer. For
1464 * the case that the buffer space is actually used to a large
1465 * extent and we run out of kernel memory we can simply drop
1466 * the new segments; TCP on the sender will just retransmit it
1467 * later. Setting the buffer size too big may only consume too
1468 * much kernel memory if the application doesn't read() from
1469 * the socket or packet loss or reordering makes use of the
1472 * The criteria to step up the receive buffer one notch are:
1473 * 1. Application has not set receive buffer size with
1474 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1475 * 2. the number of bytes received during 1/2 of an sRTT
1476 * is at least 3/8 of the current socket buffer size.
1477 * 3. receive buffer size has not hit maximal automatic size;
1479 * If all of the criteria are met we increaset the socket buffer
1480 * by a 1/2 (bounded by the max). This allows us to keep ahead
1481 * of slow-start but also makes it so our peer never gets limited
1482 * by our rwnd which we then open up causing a burst.
1484 * This algorithm does two steps per RTT at most and only if
1485 * we receive a bulk stream w/o packet losses or reorderings.
1486 * Shrinking the buffer during idle times is not necessary as
1487 * it doesn't consume any memory when idle.
1489 * TODO: Only step up if the application is actually serving
1490 * the buffer to better manage the socket buffer resources.
1493 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1494 struct tcpcb *tp, int tlen)
1498 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1499 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1500 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1501 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1502 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1503 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1504 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1506 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1508 /* Start over with next RTT. */
1512 tp->rfbuf_cnt += tlen; /* add up */
1518 tcp_input(struct mbuf **mp, int *offp, int proto)
1520 return(tcp_input_with_port(mp, offp, proto, 0));
1524 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1527 * Since tp might be gone if the session entered
1528 * the TIME_WAIT state before coming here, we need
1529 * to check if the socket is still connected.
1537 INP_LOCK_ASSERT(tp->t_inpcb);
1538 if (tp->t_flags & TF_WAKESOR) {
1539 tp->t_flags &= ~TF_WAKESOR;
1540 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1541 sorwakeup_locked(so);
1546 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1547 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1550 int acked, ourfinisacked, needoutput = 0, sack_changed;
1551 int rstreason, todrop, win, incforsyn = 0;
1555 struct in_conninfo *inc;
1563 * The size of tcp_saveipgen must be the size of the max ip header,
1566 u_char tcp_saveipgen[IP6_HDR_LEN];
1567 struct tcphdr tcp_savetcp;
1570 thflags = tcp_get_flags(th);
1571 inc = &tp->t_inpcb->inp_inc;
1572 tp->sackhint.last_sack_ack = 0;
1574 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1577 INP_WLOCK_ASSERT(tp->t_inpcb);
1578 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1580 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1584 /* Save segment, if requested. */
1585 tcp_pcap_add(th, m, &(tp->t_inpkts));
1587 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1590 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1591 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1592 log(LOG_DEBUG, "%s; %s: "
1593 "SYN|FIN segment ignored (based on "
1594 "sysctl setting)\n", s, __func__);
1601 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1602 * check SEQ.ACK first.
1604 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1605 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1606 rstreason = BANDLIM_UNLIMITED;
1607 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
1612 * Segment received on connection.
1613 * Reset idle time and keep-alive timer.
1614 * XXX: This should be done after segment
1615 * validation to ignore broken/spoofed segs.
1617 if (tp->t_idle_reduce &&
1618 (tp->snd_max == tp->snd_una) &&
1619 ((ticks - tp->t_rcvtime) >= tp->t_rxtcur))
1621 tp->t_rcvtime = ticks;
1623 if (thflags & TH_FIN)
1624 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN);
1626 * Scale up the window into a 32-bit value.
1627 * For the SYN_SENT state the scale is zero.
1629 tiwin = th->th_win << tp->snd_scale;
1631 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1635 * TCP ECN processing.
1637 if (tcp_ecn_input_segment(tp, thflags, iptos))
1638 cc_cong_signal(tp, th, CC_ECN);
1641 * Parse options on any incoming segment.
1643 tcp_dooptions(&to, (u_char *)(th + 1),
1644 (th->th_off << 2) - sizeof(struct tcphdr),
1645 (thflags & TH_SYN) ? TO_SYN : 0);
1647 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1648 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1649 (to.to_flags & TOF_SIGNATURE) == 0) {
1650 TCPSTAT_INC(tcps_sig_err_sigopt);
1651 /* XXX: should drop? */
1655 * If echoed timestamp is later than the current time,
1656 * fall back to non RFC1323 RTT calculation. Normalize
1657 * timestamp if syncookies were used when this connection
1660 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1661 to.to_tsecr -= tp->ts_offset;
1662 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1664 else if (tp->t_rxtshift == 1 &&
1665 tp->t_flags & TF_PREVVALID &&
1666 tp->t_badrxtwin != 0 &&
1667 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
1668 cc_cong_signal(tp, th, CC_RTO_ERR);
1671 * Process options only when we get SYN/ACK back. The SYN case
1672 * for incoming connections is handled in tcp_syncache.
1673 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1674 * or <SYN,ACK>) segment itself is never scaled.
1675 * XXX this is traditional behavior, may need to be cleaned up.
1677 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1678 /* Handle parallel SYN for ECN */
1679 tcp_ecn_input_parallel_syn(tp, thflags, iptos);
1680 if ((to.to_flags & TOF_SCALE) &&
1681 (tp->t_flags & TF_REQ_SCALE) &&
1682 !(tp->t_flags & TF_NOOPT)) {
1683 tp->t_flags |= TF_RCVD_SCALE;
1684 tp->snd_scale = to.to_wscale;
1686 tp->t_flags &= ~TF_REQ_SCALE;
1688 * Initial send window. It will be updated with
1689 * the next incoming segment to the scaled value.
1691 tp->snd_wnd = th->th_win;
1692 if ((to.to_flags & TOF_TS) &&
1693 (tp->t_flags & TF_REQ_TSTMP) &&
1694 !(tp->t_flags & TF_NOOPT)) {
1695 tp->t_flags |= TF_RCVD_TSTMP;
1696 tp->ts_recent = to.to_tsval;
1697 tp->ts_recent_age = tcp_ts_getticks();
1699 tp->t_flags &= ~TF_REQ_TSTMP;
1700 if (to.to_flags & TOF_MSS)
1701 tcp_mss(tp, to.to_mss);
1702 if ((tp->t_flags & TF_SACK_PERMIT) &&
1703 (!(to.to_flags & TOF_SACKPERM) ||
1704 (tp->t_flags & TF_NOOPT)))
1705 tp->t_flags &= ~TF_SACK_PERMIT;
1706 if (IS_FASTOPEN(tp->t_flags)) {
1707 if ((to.to_flags & TOF_FASTOPEN) &&
1708 !(tp->t_flags & TF_NOOPT)) {
1711 if (to.to_flags & TOF_MSS)
1714 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1718 tcp_fastopen_update_cache(tp, mss,
1719 to.to_tfo_len, to.to_tfo_cookie);
1721 tcp_fastopen_disable_path(tp);
1726 * If timestamps were negotiated during SYN/ACK and a
1727 * segment without a timestamp is received, silently drop
1728 * the segment, unless it is a RST segment or missing timestamps are
1730 * See section 3.2 of RFC 7323.
1732 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1733 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1734 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1735 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1736 "segment processed normally\n",
1741 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1742 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1743 "segment silently dropped\n", s, __func__);
1750 * If timestamps were not negotiated during SYN/ACK and a
1751 * segment with a timestamp is received, ignore the
1752 * timestamp and process the packet normally.
1753 * See section 3.2 of RFC 7323.
1755 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1756 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1757 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1758 "segment processed normally\n", s, __func__);
1764 * Header prediction: check for the two common cases
1765 * of a uni-directional data xfer. If the packet has
1766 * no control flags, is in-sequence, the window didn't
1767 * change and we're not retransmitting, it's a
1768 * candidate. If the length is zero and the ack moved
1769 * forward, we're the sender side of the xfer. Just
1770 * free the data acked & wake any higher level process
1771 * that was blocked waiting for space. If the length
1772 * is non-zero and the ack didn't move, we're the
1773 * receiver side. If we're getting packets in-order
1774 * (the reassembly queue is empty), add the data to
1775 * the socket buffer and note that we need a delayed ack.
1776 * Make sure that the hidden state-flags are also off.
1777 * Since we check for TCPS_ESTABLISHED first, it can only
1780 if (tp->t_state == TCPS_ESTABLISHED &&
1781 th->th_seq == tp->rcv_nxt &&
1782 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1783 tp->snd_nxt == tp->snd_max &&
1784 tiwin && tiwin == tp->snd_wnd &&
1785 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1787 ((to.to_flags & TOF_TS) == 0 ||
1788 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1790 * If last ACK falls within this segment's sequence numbers,
1791 * record the timestamp.
1792 * NOTE that the test is modified according to the latest
1793 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1795 if ((to.to_flags & TOF_TS) != 0 &&
1796 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1797 tp->ts_recent_age = tcp_ts_getticks();
1798 tp->ts_recent = to.to_tsval;
1802 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1803 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1804 !IN_RECOVERY(tp->t_flags) &&
1805 (to.to_flags & TOF_SACK) == 0 &&
1806 TAILQ_EMPTY(&tp->snd_holes)) {
1808 * This is a pure ack for outstanding data.
1810 TCPSTAT_INC(tcps_predack);
1813 * "bad retransmit" recovery without timestamps.
1815 if ((to.to_flags & TOF_TS) == 0 &&
1816 tp->t_rxtshift == 1 &&
1817 tp->t_flags & TF_PREVVALID &&
1818 tp->t_badrxtwin != 0 &&
1819 TSTMP_LT(ticks, tp->t_badrxtwin)) {
1820 cc_cong_signal(tp, th, CC_RTO_ERR);
1824 * Recalculate the transmit timer / rtt.
1826 * Some boxes send broken timestamp replies
1827 * during the SYN+ACK phase, ignore
1828 * timestamps of 0 or we could calculate a
1829 * huge RTT and blow up the retransmit timer.
1831 if ((to.to_flags & TOF_TS) != 0 &&
1835 t = tcp_ts_getticks() - to.to_tsecr;
1836 if (!tp->t_rttlow || tp->t_rttlow > t)
1839 TCP_TS_TO_TICKS(t) + 1);
1840 } else if (tp->t_rtttime &&
1841 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1842 if (!tp->t_rttlow ||
1843 tp->t_rttlow > ticks - tp->t_rtttime)
1844 tp->t_rttlow = ticks - tp->t_rtttime;
1846 ticks - tp->t_rtttime);
1848 acked = BYTES_THIS_ACK(tp, th);
1851 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1852 hhook_run_tcp_est_in(tp, th, &to);
1855 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1856 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1857 sbdrop(&so->so_snd, acked);
1858 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1859 SEQ_LEQ(th->th_ack, tp->snd_recover))
1860 tp->snd_recover = th->th_ack - 1;
1863 * Let the congestion control algorithm update
1864 * congestion control related information. This
1865 * typically means increasing the congestion
1868 cc_ack_received(tp, th, nsegs, CC_ACK);
1870 tp->snd_una = th->th_ack;
1872 * Pull snd_wl2 up to prevent seq wrap relative
1875 tp->snd_wl2 = th->th_ack;
1880 * If all outstanding data are acked, stop
1881 * retransmit timer, otherwise restart timer
1882 * using current (possibly backed-off) value.
1883 * If process is waiting for space,
1884 * wakeup/selwakeup/signal. If data
1885 * are ready to send, let tcp_output
1886 * decide between more output or persist.
1889 if (so->so_options & SO_DEBUG)
1890 tcp_trace(TA_INPUT, ostate, tp,
1891 (void *)tcp_saveipgen,
1894 TCP_PROBE3(debug__input, tp, th, m);
1896 * Clear t_acktime if remote side has ACKd
1897 * all data in the socket buffer.
1898 * Otherwise, update t_acktime if we received
1899 * a sufficiently large ACK.
1901 if (sbavail(&so->so_snd) == 0)
1904 tp->t_acktime = ticks;
1905 if (tp->snd_una == tp->snd_max)
1906 tcp_timer_activate(tp, TT_REXMT, 0);
1907 else if (!tcp_timer_active(tp, TT_PERSIST))
1908 tcp_timer_activate(tp, TT_REXMT,
1911 if (sbavail(&so->so_snd))
1912 (void) tcp_output(tp);
1915 } else if (th->th_ack == tp->snd_una &&
1916 tlen <= sbspace(&so->so_rcv)) {
1917 int newsize = 0; /* automatic sockbuf scaling */
1920 * This is a pure, in-sequence data packet with
1921 * nothing on the reassembly queue and we have enough
1922 * buffer space to take it.
1924 /* Clean receiver SACK report if present */
1925 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1926 tcp_clean_sackreport(tp);
1927 TCPSTAT_INC(tcps_preddat);
1928 tp->rcv_nxt += tlen;
1930 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1931 (tp->t_fbyte_in == 0)) {
1932 tp->t_fbyte_in = ticks;
1933 if (tp->t_fbyte_in == 0)
1935 if (tp->t_fbyte_out && tp->t_fbyte_in)
1936 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1939 * Pull snd_wl1 up to prevent seq wrap relative to
1942 tp->snd_wl1 = th->th_seq;
1944 * Pull rcv_up up to prevent seq wrap relative to
1947 tp->rcv_up = tp->rcv_nxt;
1948 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1949 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1951 if (so->so_options & SO_DEBUG)
1952 tcp_trace(TA_INPUT, ostate, tp,
1953 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1955 TCP_PROBE3(debug__input, tp, th, m);
1957 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1959 /* Add data to socket buffer. */
1960 SOCKBUF_LOCK(&so->so_rcv);
1961 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1965 * Set new socket buffer size.
1966 * Give up when limit is reached.
1969 if (!sbreserve_locked(so, SO_RCV,
1971 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1972 m_adj(m, drop_hdrlen); /* delayed header drop */
1973 sbappendstream_locked(&so->so_rcv, m, 0);
1975 /* NB: sorwakeup_locked() does an implicit unlock. */
1976 sorwakeup_locked(so);
1977 if (DELAY_ACK(tp, tlen)) {
1978 tp->t_flags |= TF_DELACK;
1980 tp->t_flags |= TF_ACKNOW;
1988 * Calculate amount of space in receive window,
1989 * and then do TCP input processing.
1990 * Receive window is amount of space in rcv queue,
1991 * but not less than advertised window.
1993 win = sbspace(&so->so_rcv);
1996 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1998 switch (tp->t_state) {
2000 * If the state is SYN_RECEIVED:
2001 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
2003 case TCPS_SYN_RECEIVED:
2004 if ((thflags & TH_ACK) &&
2005 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
2006 SEQ_GT(th->th_ack, tp->snd_max))) {
2007 rstreason = BANDLIM_RST_OPENPORT;
2008 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2011 if (IS_FASTOPEN(tp->t_flags)) {
2013 * When a TFO connection is in SYN_RECEIVED, the
2014 * only valid packets are the initial SYN, a
2015 * retransmit/copy of the initial SYN (possibly with
2016 * a subset of the original data), a valid ACK, a
2019 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
2020 rstreason = BANDLIM_RST_OPENPORT;
2021 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2023 } else if (thflags & TH_SYN) {
2024 /* non-initial SYN is ignored */
2025 if ((tcp_timer_active(tp, TT_DELACK) ||
2026 tcp_timer_active(tp, TT_REXMT)))
2028 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2035 * If the state is SYN_SENT:
2036 * if seg contains a RST with valid ACK (SEQ.ACK has already
2037 * been verified), then drop the connection.
2038 * if seg contains a RST without an ACK, drop the seg.
2039 * if seg does not contain SYN, then drop the seg.
2040 * Otherwise this is an acceptable SYN segment
2041 * initialize tp->rcv_nxt and tp->irs
2042 * if seg contains ack then advance tp->snd_una
2043 * if seg contains an ECE and ECN support is enabled, the stream
2045 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2046 * arrange for segment to be acked (eventually)
2047 * continue processing rest of data/controls, beginning with URG
2050 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2051 TCP_PROBE5(connect__refused, NULL, tp,
2053 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2054 tp = tcp_drop(tp, ECONNREFUSED);
2056 if (thflags & TH_RST)
2058 if (!(thflags & TH_SYN))
2061 tp->irs = th->th_seq;
2063 if (thflags & TH_ACK) {
2064 int tfo_partial_ack = 0;
2066 TCPSTAT_INC(tcps_connects);
2069 mac_socketpeer_set_from_mbuf(m, so);
2071 /* Do window scaling on this connection? */
2072 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2073 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2074 tp->rcv_scale = tp->request_r_scale;
2076 tp->rcv_adv += min(tp->rcv_wnd,
2077 TCP_MAXWIN << tp->rcv_scale);
2078 tp->snd_una++; /* SYN is acked */
2080 * If not all the data that was sent in the TFO SYN
2081 * has been acked, resend the remainder right away.
2083 if (IS_FASTOPEN(tp->t_flags) &&
2084 (tp->snd_una != tp->snd_max)) {
2085 tp->snd_nxt = th->th_ack;
2086 tfo_partial_ack = 1;
2089 * If there's data, delay ACK; if there's also a FIN
2090 * ACKNOW will be turned on later.
2092 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2093 tcp_timer_activate(tp, TT_DELACK,
2096 tp->t_flags |= TF_ACKNOW;
2098 tcp_ecn_input_syn_sent(tp, thflags, iptos);
2101 * Received <SYN,ACK> in SYN_SENT[*] state.
2103 * SYN_SENT --> ESTABLISHED
2104 * SYN_SENT* --> FIN_WAIT_1
2106 tp->t_starttime = ticks;
2107 if (tp->t_flags & TF_NEEDFIN) {
2108 tp->t_acktime = ticks;
2109 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2110 tp->t_flags &= ~TF_NEEDFIN;
2113 tcp_state_change(tp, TCPS_ESTABLISHED);
2114 TCP_PROBE5(connect__established, NULL, tp,
2117 tcp_timer_activate(tp, TT_KEEP,
2122 * Received initial SYN in SYN-SENT[*] state =>
2123 * simultaneous open.
2124 * If it succeeds, connection is * half-synchronized.
2125 * Otherwise, do 3-way handshake:
2126 * SYN-SENT -> SYN-RECEIVED
2127 * SYN-SENT* -> SYN-RECEIVED*
2129 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN);
2130 tcp_timer_activate(tp, TT_REXMT, 0);
2131 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2134 INP_WLOCK_ASSERT(tp->t_inpcb);
2137 * Advance th->th_seq to correspond to first data byte.
2138 * If data, trim to stay within window,
2139 * dropping FIN if necessary.
2142 if (tlen > tp->rcv_wnd) {
2143 todrop = tlen - tp->rcv_wnd;
2147 TCPSTAT_INC(tcps_rcvpackafterwin);
2148 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2150 tp->snd_wl1 = th->th_seq - 1;
2151 tp->rcv_up = th->th_seq;
2153 * Client side of transaction: already sent SYN and data.
2154 * If the remote host used T/TCP to validate the SYN,
2155 * our data will be ACK'd; if so, enter normal data segment
2156 * processing in the middle of step 5, ack processing.
2157 * Otherwise, goto step 6.
2159 if (thflags & TH_ACK)
2166 * States other than LISTEN or SYN_SENT.
2167 * First check the RST flag and sequence number since reset segments
2168 * are exempt from the timestamp and connection count tests. This
2169 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2170 * below which allowed reset segments in half the sequence space
2171 * to fall though and be processed (which gives forged reset
2172 * segments with a random sequence number a 50 percent chance of
2173 * killing a connection).
2174 * Then check timestamp, if present.
2175 * Then check the connection count, if present.
2176 * Then check that at least some bytes of segment are within
2177 * receive window. If segment begins before rcv_nxt,
2178 * drop leading data (and SYN); if nothing left, just ack.
2180 if (thflags & TH_RST) {
2182 * RFC5961 Section 3.2
2184 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2185 * - If RST is in window, we send challenge ACK.
2187 * Note: to take into account delayed ACKs, we should
2188 * test against last_ack_sent instead of rcv_nxt.
2189 * Note 2: we handle special case of closed window, not
2190 * covered by the RFC.
2192 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2193 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2194 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2195 KASSERT(tp->t_state != TCPS_SYN_SENT,
2196 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2199 if (V_tcp_insecure_rst ||
2200 tp->last_ack_sent == th->th_seq) {
2201 TCPSTAT_INC(tcps_drops);
2202 /* Drop the connection. */
2203 switch (tp->t_state) {
2204 case TCPS_SYN_RECEIVED:
2205 so->so_error = ECONNREFUSED;
2207 case TCPS_ESTABLISHED:
2208 case TCPS_FIN_WAIT_1:
2209 case TCPS_FIN_WAIT_2:
2210 case TCPS_CLOSE_WAIT:
2213 so->so_error = ECONNRESET;
2217 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST);
2221 TCPSTAT_INC(tcps_badrst);
2222 /* Send challenge ACK. */
2223 tcp_respond(tp, mtod(m, void *), th, m,
2224 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2225 tp->last_ack_sent = tp->rcv_nxt;
2233 * RFC5961 Section 4.2
2234 * Send challenge ACK for any SYN in synchronized state.
2236 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2237 tp->t_state != TCPS_SYN_RECEIVED) {
2238 TCPSTAT_INC(tcps_badsyn);
2239 if (V_tcp_insecure_syn &&
2240 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2241 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2242 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2243 tp = tcp_drop(tp, ECONNRESET);
2244 rstreason = BANDLIM_UNLIMITED;
2246 /* Send challenge ACK. */
2247 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2248 tp->snd_nxt, TH_ACK);
2249 tp->last_ack_sent = tp->rcv_nxt;
2256 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2257 * and it's less than ts_recent, drop it.
2259 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2260 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2261 /* Check to see if ts_recent is over 24 days old. */
2262 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2264 * Invalidate ts_recent. If this segment updates
2265 * ts_recent, the age will be reset later and ts_recent
2266 * will get a valid value. If it does not, setting
2267 * ts_recent to zero will at least satisfy the
2268 * requirement that zero be placed in the timestamp
2269 * echo reply when ts_recent isn't valid. The
2270 * age isn't reset until we get a valid ts_recent
2271 * because we don't want out-of-order segments to be
2272 * dropped when ts_recent is old.
2276 TCPSTAT_INC(tcps_rcvduppack);
2277 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2278 TCPSTAT_INC(tcps_pawsdrop);
2286 * In the SYN-RECEIVED state, validate that the packet belongs to
2287 * this connection before trimming the data to fit the receive
2288 * window. Check the sequence number versus IRS since we know
2289 * the sequence numbers haven't wrapped. This is a partial fix
2290 * for the "LAND" DoS attack.
2292 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2293 rstreason = BANDLIM_RST_OPENPORT;
2294 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2298 todrop = tp->rcv_nxt - th->th_seq;
2300 if (thflags & TH_SYN) {
2310 * Following if statement from Stevens, vol. 2, p. 960.
2313 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2315 * Any valid FIN must be to the left of the window.
2316 * At this point the FIN must be a duplicate or out
2317 * of sequence; drop it.
2322 * Send an ACK to resynchronize and drop any data.
2323 * But keep on processing for RST or ACK.
2325 tp->t_flags |= TF_ACKNOW;
2327 TCPSTAT_INC(tcps_rcvduppack);
2328 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2330 TCPSTAT_INC(tcps_rcvpartduppack);
2331 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2334 * DSACK - add SACK block for dropped range
2336 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2337 tcp_update_sack_list(tp, th->th_seq,
2338 th->th_seq + todrop);
2340 * ACK now, as the next in-sequence segment
2341 * will clear the DSACK block again
2343 tp->t_flags |= TF_ACKNOW;
2345 drop_hdrlen += todrop; /* drop from the top afterwards */
2346 th->th_seq += todrop;
2348 if (th->th_urp > todrop)
2349 th->th_urp -= todrop;
2357 * If new data are received on a connection after the
2358 * user processes are gone, then RST the other end.
2360 if ((tp->t_flags & TF_CLOSED) && tlen) {
2361 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2362 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2363 "after socket was closed, "
2364 "sending RST and removing tcpcb\n",
2365 s, __func__, tcpstates[tp->t_state], tlen);
2368 tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE);
2369 /* tcp_close will kill the inp pre-log the Reset */
2370 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST);
2372 TCPSTAT_INC(tcps_rcvafterclose);
2373 rstreason = BANDLIM_UNLIMITED;
2378 * If segment ends after window, drop trailing data
2379 * (and PUSH and FIN); if nothing left, just ACK.
2381 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2383 TCPSTAT_INC(tcps_rcvpackafterwin);
2384 if (todrop >= tlen) {
2385 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2387 * If window is closed can only take segments at
2388 * window edge, and have to drop data and PUSH from
2389 * incoming segments. Continue processing, but
2390 * remember to ack. Otherwise, drop segment
2393 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2394 tp->t_flags |= TF_ACKNOW;
2395 TCPSTAT_INC(tcps_rcvwinprobe);
2399 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2402 thflags &= ~(TH_PUSH|TH_FIN);
2406 * If last ACK falls within this segment's sequence numbers,
2407 * record its timestamp.
2409 * 1) That the test incorporates suggestions from the latest
2410 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2411 * 2) That updating only on newer timestamps interferes with
2412 * our earlier PAWS tests, so this check should be solely
2413 * predicated on the sequence space of this segment.
2414 * 3) That we modify the segment boundary check to be
2415 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2416 * instead of RFC1323's
2417 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2418 * This modified check allows us to overcome RFC1323's
2419 * limitations as described in Stevens TCP/IP Illustrated
2420 * Vol. 2 p.869. In such cases, we can still calculate the
2421 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2423 if ((to.to_flags & TOF_TS) != 0 &&
2424 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2425 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2426 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2427 tp->ts_recent_age = tcp_ts_getticks();
2428 tp->ts_recent = to.to_tsval;
2432 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2433 * flag is on (half-synchronized state), then queue data for
2434 * later processing; else drop segment and return.
2436 if ((thflags & TH_ACK) == 0) {
2437 if (tp->t_state == TCPS_SYN_RECEIVED ||
2438 (tp->t_flags & TF_NEEDSYN)) {
2439 if (tp->t_state == TCPS_SYN_RECEIVED &&
2440 IS_FASTOPEN(tp->t_flags)) {
2441 tp->snd_wnd = tiwin;
2445 } else if (tp->t_flags & TF_ACKNOW)
2454 switch (tp->t_state) {
2456 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2457 * ESTABLISHED state and continue processing.
2458 * The ACK was checked above.
2460 case TCPS_SYN_RECEIVED:
2462 TCPSTAT_INC(tcps_connects);
2463 if (tp->t_flags & TF_SONOTCONN) {
2465 * Usually SYN_RECEIVED had been created from a LISTEN,
2466 * and solisten_enqueue() has already marked the socket
2467 * layer as connected. If it didn't, which can happen
2468 * only with an accept_filter(9), then the tp is marked
2469 * with TF_SONOTCONN. The other reason for this mark
2470 * to be set is a simultaneous open, a SYN_RECEIVED
2471 * that had been created from SYN_SENT.
2473 tp->t_flags &= ~TF_SONOTCONN;
2476 /* Do window scaling? */
2477 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2478 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2479 tp->rcv_scale = tp->request_r_scale;
2481 tp->snd_wnd = tiwin;
2484 * SYN-RECEIVED -> ESTABLISHED
2485 * SYN-RECEIVED* -> FIN-WAIT-1
2487 tp->t_starttime = ticks;
2488 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2489 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2490 tp->t_tfo_pending = NULL;
2492 if (tp->t_flags & TF_NEEDFIN) {
2493 tp->t_acktime = ticks;
2494 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2495 tp->t_flags &= ~TF_NEEDFIN;
2497 tcp_state_change(tp, TCPS_ESTABLISHED);
2498 TCP_PROBE5(accept__established, NULL, tp,
2501 * TFO connections call cc_conn_init() during SYN
2502 * processing. Calling it again here for such
2503 * connections is not harmless as it would undo the
2504 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2507 if (!IS_FASTOPEN(tp->t_flags))
2509 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2512 * Account for the ACK of our SYN prior to
2513 * regular ACK processing below, except for
2514 * simultaneous SYN, which is handled later.
2516 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2519 * If segment contains data or ACK, will call tcp_reass()
2520 * later; if not, do so now to pass queued data to user.
2522 if (tlen == 0 && (thflags & TH_FIN) == 0) {
2523 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2525 tcp_handle_wakeup(tp, so);
2527 tp->snd_wl1 = th->th_seq - 1;
2531 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2532 * ACKs. If the ack is in the range
2533 * tp->snd_una < th->th_ack <= tp->snd_max
2534 * then advance tp->snd_una to th->th_ack and drop
2535 * data from the retransmission queue. If this ACK reflects
2536 * more up to date window information we update our window information.
2538 case TCPS_ESTABLISHED:
2539 case TCPS_FIN_WAIT_1:
2540 case TCPS_FIN_WAIT_2:
2541 case TCPS_CLOSE_WAIT:
2544 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2545 TCPSTAT_INC(tcps_rcvacktoomuch);
2548 if (tcp_is_sack_recovery(tp, &to)) {
2549 if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) &&
2550 (tp->t_flags & TF_LRD)) {
2551 tcp_sack_lost_retransmission(tp, th);
2555 * Reset the value so that previous (valid) value
2556 * from the last ack with SACK doesn't get used.
2558 tp->sackhint.sacked_bytes = 0;
2561 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2562 hhook_run_tcp_est_in(tp, th, &to);
2565 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2566 maxseg = tcp_maxseg(tp);
2568 (tiwin == tp->snd_wnd ||
2569 (tp->t_flags & TF_SACK_PERMIT))) {
2571 * If this is the first time we've seen a
2572 * FIN from the remote, this is not a
2573 * duplicate and it needs to be processed
2574 * normally. This happens during a
2575 * simultaneous close.
2577 if ((thflags & TH_FIN) &&
2578 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2582 TCPSTAT_INC(tcps_rcvdupack);
2584 * If we have outstanding data (other than
2585 * a window probe), this is a completely
2586 * duplicate ack (ie, window info didn't
2587 * change and FIN isn't set),
2588 * the ack is the biggest we've
2589 * seen and we've seen exactly our rexmt
2590 * threshold of them, assume a packet
2591 * has been dropped and retransmit it.
2592 * Kludge snd_nxt & the congestion
2593 * window so we send only this one
2596 * We know we're losing at the current
2597 * window size so do congestion avoidance
2598 * (set ssthresh to half the current window
2599 * and pull our congestion window back to
2600 * the new ssthresh).
2602 * Dup acks mean that packets have left the
2603 * network (they're now cached at the receiver)
2604 * so bump cwnd by the amount in the receiver
2605 * to keep a constant cwnd packets in the
2608 * When using TCP ECN, notify the peer that
2609 * we reduced the cwnd.
2612 * Following 2 kinds of acks should not affect
2615 * 2) Acks with SACK but without any new SACK
2616 * information in them. These could result from
2617 * any anomaly in the network like a switch
2618 * duplicating packets or a possible DoS attack.
2620 if (th->th_ack != tp->snd_una ||
2621 (tcp_is_sack_recovery(tp, &to) &&
2624 else if (!tcp_timer_active(tp, TT_REXMT))
2626 else if (++tp->t_dupacks > tcprexmtthresh ||
2627 IN_FASTRECOVERY(tp->t_flags)) {
2628 cc_ack_received(tp, th, nsegs,
2631 IN_FASTRECOVERY(tp->t_flags)) {
2632 tcp_do_prr_ack(tp, th, &to);
2633 } else if (tcp_is_sack_recovery(tp, &to) &&
2634 IN_FASTRECOVERY(tp->t_flags)) {
2638 * Compute the amount of data in flight first.
2639 * We can inject new data into the pipe iff
2640 * we have less than 1/2 the original window's
2641 * worth of data in flight.
2643 if (V_tcp_do_newsack)
2644 awnd = tcp_compute_pipe(tp);
2646 awnd = (tp->snd_nxt - tp->snd_fack) +
2647 tp->sackhint.sack_bytes_rexmit;
2649 if (awnd < tp->snd_ssthresh) {
2650 tp->snd_cwnd += maxseg;
2651 if (tp->snd_cwnd > tp->snd_ssthresh)
2652 tp->snd_cwnd = tp->snd_ssthresh;
2655 tp->snd_cwnd += maxseg;
2656 (void) tcp_output(tp);
2658 } else if (tp->t_dupacks == tcprexmtthresh ||
2659 (tp->t_flags & TF_SACK_PERMIT &&
2661 tp->sackhint.sacked_bytes >
2662 (tcprexmtthresh - 1) * maxseg)) {
2665 * Above is the RFC6675 trigger condition of
2666 * more than (dupthresh-1)*maxseg sacked data.
2667 * If the count of holes in the
2668 * scoreboard is >= dupthresh, we could
2669 * also enter loss recovery, but don't
2670 * have that value readily available.
2672 tp->t_dupacks = tcprexmtthresh;
2673 tcp_seq onxt = tp->snd_nxt;
2676 * If we're doing sack, or prr, check
2677 * to see if we're already in sack
2678 * recovery. If we're not doing sack,
2679 * check to see if we're in newreno
2683 (tp->t_flags & TF_SACK_PERMIT)) {
2684 if (IN_FASTRECOVERY(tp->t_flags)) {
2689 if (SEQ_LEQ(th->th_ack,
2695 /* Congestion signal before ack. */
2696 cc_cong_signal(tp, th, CC_NDUPACK);
2697 cc_ack_received(tp, th, nsegs,
2699 tcp_timer_activate(tp, TT_REXMT, 0);
2703 * snd_ssthresh is already updated by
2706 if (tcp_is_sack_recovery(tp, &to)) {
2707 tp->sackhint.prr_delivered =
2708 tp->sackhint.sacked_bytes;
2710 tp->sackhint.prr_delivered =
2711 imin(tp->snd_max - tp->snd_una,
2712 imin(INT_MAX / 65536,
2713 tp->t_dupacks) * maxseg);
2715 tp->sackhint.recover_fs = max(1,
2716 tp->snd_nxt - tp->snd_una);
2718 if (tcp_is_sack_recovery(tp, &to)) {
2720 tcps_sack_recovery_episode);
2721 tp->snd_recover = tp->snd_nxt;
2722 tp->snd_cwnd = maxseg;
2723 (void) tcp_output(tp);
2724 if (SEQ_GT(th->th_ack, tp->snd_una))
2725 goto resume_partialack;
2728 tp->snd_nxt = th->th_ack;
2729 tp->snd_cwnd = maxseg;
2730 (void) tcp_output(tp);
2731 KASSERT(tp->snd_limited <= 2,
2732 ("%s: tp->snd_limited too big",
2734 tp->snd_cwnd = tp->snd_ssthresh +
2736 (tp->t_dupacks - tp->snd_limited);
2737 if (SEQ_GT(onxt, tp->snd_nxt))
2740 } else if (V_tcp_do_rfc3042) {
2742 * Process first and second duplicate
2743 * ACKs. Each indicates a segment
2744 * leaving the network, creating room
2745 * for more. Make sure we can send a
2746 * packet on reception of each duplicate
2747 * ACK by increasing snd_cwnd by one
2748 * segment. Restore the original
2749 * snd_cwnd after packet transmission.
2751 cc_ack_received(tp, th, nsegs,
2753 uint32_t oldcwnd = tp->snd_cwnd;
2754 tcp_seq oldsndmax = tp->snd_max;
2758 KASSERT(tp->t_dupacks == 1 ||
2760 ("%s: dupacks not 1 or 2",
2762 if (tp->t_dupacks == 1)
2763 tp->snd_limited = 0;
2765 (tp->snd_nxt - tp->snd_una) +
2766 (tp->t_dupacks - tp->snd_limited) *
2769 * Only call tcp_output when there
2770 * is new data available to be sent
2771 * or we need to send an ACK.
2773 SOCKBUF_LOCK(&so->so_snd);
2774 avail = sbavail(&so->so_snd) -
2775 (tp->snd_nxt - tp->snd_una);
2776 SOCKBUF_UNLOCK(&so->so_snd);
2777 if (avail > 0 || tp->t_flags & TF_ACKNOW)
2778 (void) tcp_output(tp);
2779 sent = tp->snd_max - oldsndmax;
2780 if (sent > maxseg) {
2781 KASSERT((tp->t_dupacks == 2 &&
2782 tp->snd_limited == 0) ||
2783 (sent == maxseg + 1 &&
2784 tp->t_flags & TF_SENTFIN),
2785 ("%s: sent too much",
2787 tp->snd_limited = 2;
2788 } else if (sent > 0)
2790 tp->snd_cwnd = oldcwnd;
2797 * This ack is advancing the left edge, reset the
2802 * If this ack also has new SACK info, increment the
2803 * counter as per rfc6675. The variable
2804 * sack_changed tracks all changes to the SACK
2805 * scoreboard, including when partial ACKs without
2806 * SACK options are received, and clear the scoreboard
2807 * from the left side. Such partial ACKs should not be
2808 * counted as dupacks here.
2810 if (tcp_is_sack_recovery(tp, &to) &&
2813 /* limit overhead by setting maxseg last */
2814 if (!IN_FASTRECOVERY(tp->t_flags) &&
2815 (tp->sackhint.sacked_bytes >
2816 ((tcprexmtthresh - 1) *
2817 (maxseg = tcp_maxseg(tp))))) {
2818 goto enter_recovery;
2824 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2825 ("%s: th_ack <= snd_una", __func__));
2828 * If the congestion window was inflated to account
2829 * for the other side's cached packets, retract it.
2831 if (IN_FASTRECOVERY(tp->t_flags)) {
2832 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2833 if (tp->t_flags & TF_SACK_PERMIT)
2834 if (V_tcp_do_prr && to.to_flags & TOF_SACK) {
2835 tcp_timer_activate(tp, TT_REXMT, 0);
2837 tcp_do_prr_ack(tp, th, &to);
2838 tp->t_flags |= TF_ACKNOW;
2839 (void) tcp_output(tp);
2841 tcp_sack_partialack(tp, th);
2843 tcp_newreno_partial_ack(tp, th);
2845 cc_post_recovery(tp, th);
2846 } else if (IN_CONGRECOVERY(tp->t_flags)) {
2847 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2849 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th);
2850 tp->snd_fack = th->th_ack;
2851 tcp_do_prr_ack(tp, th, &to);
2852 (void) tcp_output(tp);
2855 cc_post_recovery(tp, th);
2858 * If we reach this point, ACK is not a duplicate,
2859 * i.e., it ACKs something we sent.
2861 if (tp->t_flags & TF_NEEDSYN) {
2863 * T/TCP: Connection was half-synchronized, and our
2864 * SYN has been ACK'd (so connection is now fully
2865 * synchronized). Go to non-starred state,
2866 * increment snd_una for ACK of SYN, and check if
2867 * we can do window scaling.
2869 tp->t_flags &= ~TF_NEEDSYN;
2871 /* Do window scaling? */
2872 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2873 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2874 tp->rcv_scale = tp->request_r_scale;
2875 /* Send window already scaled. */
2880 INP_WLOCK_ASSERT(tp->t_inpcb);
2883 * Adjust for the SYN bit in sequence space,
2884 * but don't account for it in cwnd calculations.
2885 * This is for the SYN_RECEIVED, non-simultaneous
2886 * SYN case. SYN_SENT and simultaneous SYN are
2887 * treated elsewhere.
2891 acked = BYTES_THIS_ACK(tp, th);
2892 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2893 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2894 tp->snd_una, th->th_ack, tp, m));
2895 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2896 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2899 * If we just performed our first retransmit, and the ACK
2900 * arrives within our recovery window, then it was a mistake
2901 * to do the retransmit in the first place. Recover our
2902 * original cwnd and ssthresh, and proceed to transmit where
2905 if (tp->t_rxtshift == 1 &&
2906 tp->t_flags & TF_PREVVALID &&
2907 tp->t_badrxtwin != 0 &&
2908 to.to_flags & TOF_TS &&
2910 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
2911 cc_cong_signal(tp, th, CC_RTO_ERR);
2914 * If we have a timestamp reply, update smoothed
2915 * round trip time. If no timestamp is present but
2916 * transmit timer is running and timed sequence
2917 * number was acked, update smoothed round trip time.
2918 * Since we now have an rtt measurement, cancel the
2919 * timer backoff (cf., Phil Karn's retransmit alg.).
2920 * Recompute the initial retransmit timer.
2922 * Some boxes send broken timestamp replies
2923 * during the SYN+ACK phase, ignore
2924 * timestamps of 0 or we could calculate a
2925 * huge RTT and blow up the retransmit timer.
2927 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2930 t = tcp_ts_getticks() - to.to_tsecr;
2931 if (!tp->t_rttlow || tp->t_rttlow > t)
2933 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2934 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2935 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2936 tp->t_rttlow = ticks - tp->t_rtttime;
2937 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2940 SOCKBUF_LOCK(&so->so_snd);
2942 * Clear t_acktime if remote side has ACKd all data in the
2943 * socket buffer and FIN (if applicable).
2944 * Otherwise, update t_acktime if we received a sufficiently
2947 if ((tp->t_state <= TCPS_CLOSE_WAIT &&
2948 acked == sbavail(&so->so_snd)) ||
2949 acked > sbavail(&so->so_snd))
2952 tp->t_acktime = ticks;
2955 * If all outstanding data is acked, stop retransmit
2956 * timer and remember to restart (more output or persist).
2957 * If there is more data to be acked, restart retransmit
2958 * timer, using current (possibly backed-off) value.
2960 if (th->th_ack == tp->snd_max) {
2961 tcp_timer_activate(tp, TT_REXMT, 0);
2963 } else if (!tcp_timer_active(tp, TT_PERSIST))
2964 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp));
2967 * If no data (only SYN) was ACK'd,
2968 * skip rest of ACK processing.
2971 SOCKBUF_UNLOCK(&so->so_snd);
2976 * Let the congestion control algorithm update congestion
2977 * control related information. This typically means increasing
2978 * the congestion window.
2980 cc_ack_received(tp, th, nsegs, CC_ACK);
2982 if (acked > sbavail(&so->so_snd)) {
2983 if (tp->snd_wnd >= sbavail(&so->so_snd))
2984 tp->snd_wnd -= sbavail(&so->so_snd);
2987 mfree = sbcut_locked(&so->so_snd,
2988 (int)sbavail(&so->so_snd));
2991 mfree = sbcut_locked(&so->so_snd, acked);
2992 if (tp->snd_wnd >= (uint32_t) acked)
2993 tp->snd_wnd -= acked;
2998 /* NB: sowwakeup_locked() does an implicit unlock. */
2999 sowwakeup_locked(so);
3001 /* Detect una wraparound. */
3002 if (!IN_RECOVERY(tp->t_flags) &&
3003 SEQ_GT(tp->snd_una, tp->snd_recover) &&
3004 SEQ_LEQ(th->th_ack, tp->snd_recover))
3005 tp->snd_recover = th->th_ack - 1;
3006 /* XXXLAS: Can this be moved up into cc_post_recovery? */
3007 if (IN_RECOVERY(tp->t_flags) &&
3008 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
3009 EXIT_RECOVERY(tp->t_flags);
3011 tp->snd_una = th->th_ack;
3012 if (tp->t_flags & TF_SACK_PERMIT) {
3013 if (SEQ_GT(tp->snd_una, tp->snd_recover))
3014 tp->snd_recover = tp->snd_una;
3016 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
3017 tp->snd_nxt = tp->snd_una;
3019 switch (tp->t_state) {
3021 * In FIN_WAIT_1 STATE in addition to the processing
3022 * for the ESTABLISHED state if our FIN is now acknowledged
3023 * then enter FIN_WAIT_2.
3025 case TCPS_FIN_WAIT_1:
3026 if (ourfinisacked) {
3028 * If we can't receive any more
3029 * data, then closing user can proceed.
3030 * Starting the timer is contrary to the
3031 * specification, but if we don't get a FIN
3032 * we'll hang forever.
3035 * we should release the tp also, and use a
3038 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3039 soisdisconnected(so);
3040 tcp_timer_activate(tp, TT_2MSL,
3041 (tcp_fast_finwait2_recycle ?
3042 tcp_finwait2_timeout :
3045 tcp_state_change(tp, TCPS_FIN_WAIT_2);
3050 * In CLOSING STATE in addition to the processing for
3051 * the ESTABLISHED state if the ACK acknowledges our FIN
3052 * then enter the TIME-WAIT state, otherwise ignore
3056 if (ourfinisacked) {
3064 * In LAST_ACK, we may still be waiting for data to drain
3065 * and/or to be acked, as well as for the ack of our FIN.
3066 * If our FIN is now acknowledged, delete the TCB,
3067 * enter the closed state and return.
3070 if (ourfinisacked) {
3079 INP_WLOCK_ASSERT(tp->t_inpcb);
3082 * Update window information.
3083 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3085 if ((thflags & TH_ACK) &&
3086 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3087 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3088 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3089 /* keep track of pure window updates */
3091 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3092 TCPSTAT_INC(tcps_rcvwinupd);
3093 tp->snd_wnd = tiwin;
3094 tp->snd_wl1 = th->th_seq;
3095 tp->snd_wl2 = th->th_ack;
3096 if (tp->snd_wnd > tp->max_sndwnd)
3097 tp->max_sndwnd = tp->snd_wnd;
3102 * Process segments with URG.
3104 if ((thflags & TH_URG) && th->th_urp &&
3105 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3107 * This is a kludge, but if we receive and accept
3108 * random urgent pointers, we'll crash in
3109 * soreceive. It's hard to imagine someone
3110 * actually wanting to send this much urgent data.
3112 SOCKBUF_LOCK(&so->so_rcv);
3113 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3114 th->th_urp = 0; /* XXX */
3115 thflags &= ~TH_URG; /* XXX */
3116 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3117 goto dodata; /* XXX */
3120 * If this segment advances the known urgent pointer,
3121 * then mark the data stream. This should not happen
3122 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3123 * a FIN has been received from the remote side.
3124 * In these states we ignore the URG.
3126 * According to RFC961 (Assigned Protocols),
3127 * the urgent pointer points to the last octet
3128 * of urgent data. We continue, however,
3129 * to consider it to indicate the first octet
3130 * of data past the urgent section as the original
3131 * spec states (in one of two places).
3133 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3134 tp->rcv_up = th->th_seq + th->th_urp;
3135 so->so_oobmark = sbavail(&so->so_rcv) +
3136 (tp->rcv_up - tp->rcv_nxt) - 1;
3137 if (so->so_oobmark == 0)
3138 so->so_rcv.sb_state |= SBS_RCVATMARK;
3140 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3142 SOCKBUF_UNLOCK(&so->so_rcv);
3144 * Remove out of band data so doesn't get presented to user.
3145 * This can happen independent of advancing the URG pointer,
3146 * but if two URG's are pending at once, some out-of-band
3147 * data may creep in... ick.
3149 if (th->th_urp <= (uint32_t)tlen &&
3150 !(so->so_options & SO_OOBINLINE)) {
3151 /* hdr drop is delayed */
3152 tcp_pulloutofband(so, th, m, drop_hdrlen);
3156 * If no out of band data is expected,
3157 * pull receive urgent pointer along
3158 * with the receive window.
3160 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3161 tp->rcv_up = tp->rcv_nxt;
3164 INP_WLOCK_ASSERT(tp->t_inpcb);
3167 * Process the segment text, merging it into the TCP sequencing queue,
3168 * and arranging for acknowledgment of receipt if necessary.
3169 * This process logically involves adjusting tp->rcv_wnd as data
3170 * is presented to the user (this happens in tcp_usrreq.c,
3171 * case PRU_RCVD). If a FIN has already been received on this
3172 * connection then we just ignore the text.
3174 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3175 IS_FASTOPEN(tp->t_flags));
3176 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3177 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3178 tcp_seq save_start = th->th_seq;
3179 tcp_seq save_rnxt = tp->rcv_nxt;
3180 int save_tlen = tlen;
3181 m_adj(m, drop_hdrlen); /* delayed header drop */
3183 * Insert segment which includes th into TCP reassembly queue
3184 * with control block tp. Set thflags to whether reassembly now
3185 * includes a segment with FIN. This handles the common case
3186 * inline (segment is the next to be received on an established
3187 * connection, and the queue is empty), avoiding linkage into
3188 * and removal from the queue and repetition of various
3190 * Set DELACK for segments received in order, but ack
3191 * immediately when segments are out of order (so
3192 * fast retransmit can work).
3194 if (th->th_seq == tp->rcv_nxt &&
3196 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3198 if (DELAY_ACK(tp, tlen) || tfo_syn)
3199 tp->t_flags |= TF_DELACK;
3201 tp->t_flags |= TF_ACKNOW;
3202 tp->rcv_nxt += tlen;
3204 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3205 (tp->t_fbyte_in == 0)) {
3206 tp->t_fbyte_in = ticks;
3207 if (tp->t_fbyte_in == 0)
3209 if (tp->t_fbyte_out && tp->t_fbyte_in)
3210 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3212 thflags = tcp_get_flags(th) & TH_FIN;
3213 TCPSTAT_INC(tcps_rcvpack);
3214 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3215 SOCKBUF_LOCK(&so->so_rcv);
3216 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3219 sbappendstream_locked(&so->so_rcv, m, 0);
3220 tp->t_flags |= TF_WAKESOR;
3223 * XXX: Due to the header drop above "th" is
3224 * theoretically invalid by now. Fortunately
3225 * m_adj() doesn't actually frees any mbufs
3226 * when trimming from the head.
3228 tcp_seq temp = save_start;
3230 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3231 tp->t_flags |= TF_ACKNOW;
3233 if ((tp->t_flags & TF_SACK_PERMIT) &&
3235 TCPS_HAVEESTABLISHED(tp->t_state)) {
3236 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3238 * DSACK actually handled in the fastpath
3241 tcp_update_sack_list(tp, save_start,
3242 save_start + save_tlen);
3243 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3244 if ((tp->rcv_numsacks >= 1) &&
3245 (tp->sackblks[0].end == save_start)) {
3247 * Partial overlap, recorded at todrop
3250 tcp_update_sack_list(tp,
3251 tp->sackblks[0].start,
3252 tp->sackblks[0].end);
3254 tcp_update_dsack_list(tp, save_start,
3255 save_start + save_tlen);
3257 } else if (tlen >= save_tlen) {
3258 /* Update of sackblks. */
3259 tcp_update_dsack_list(tp, save_start,
3260 save_start + save_tlen);
3261 } else if (tlen > 0) {
3262 tcp_update_dsack_list(tp, save_start,
3266 tcp_handle_wakeup(tp, so);
3269 * Note the amount of data that peer has sent into
3270 * our window, in order to estimate the sender's
3274 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3275 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3277 len = so->so_rcv.sb_hiwat;
3285 * If FIN is received ACK the FIN and let the user know
3286 * that the connection is closing.
3288 if (thflags & TH_FIN) {
3289 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3290 /* The socket upcall is handled by socantrcvmore. */
3293 * If connection is half-synchronized
3294 * (ie NEEDSYN flag on) then delay ACK,
3295 * so it may be piggybacked when SYN is sent.
3296 * Otherwise, since we received a FIN then no
3297 * more input can be expected, send ACK now.
3299 if (tp->t_flags & TF_NEEDSYN)
3300 tp->t_flags |= TF_DELACK;
3302 tp->t_flags |= TF_ACKNOW;
3305 switch (tp->t_state) {
3307 * In SYN_RECEIVED and ESTABLISHED STATES
3308 * enter the CLOSE_WAIT state.
3310 case TCPS_SYN_RECEIVED:
3311 tp->t_starttime = ticks;
3313 case TCPS_ESTABLISHED:
3314 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3318 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3319 * enter the CLOSING state.
3321 case TCPS_FIN_WAIT_1:
3322 tcp_state_change(tp, TCPS_CLOSING);
3326 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3327 * starting the time-wait timer, turning off the other
3330 case TCPS_FIN_WAIT_2:
3336 if (so->so_options & SO_DEBUG)
3337 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3340 TCP_PROBE3(debug__input, tp, th, m);
3343 * Return any desired output.
3345 if (needoutput || (tp->t_flags & TF_ACKNOW))
3346 (void) tcp_output(tp);
3349 INP_WLOCK_ASSERT(tp->t_inpcb);
3351 if (tp->t_flags & TF_DELACK) {
3352 tp->t_flags &= ~TF_DELACK;
3353 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3355 INP_WUNLOCK(tp->t_inpcb);
3360 * Generate an ACK dropping incoming segment if it occupies
3361 * sequence space, where the ACK reflects our state.
3363 * We can now skip the test for the RST flag since all
3364 * paths to this code happen after packets containing
3365 * RST have been dropped.
3367 * In the SYN-RECEIVED state, don't send an ACK unless the
3368 * segment we received passes the SYN-RECEIVED ACK test.
3369 * If it fails send a RST. This breaks the loop in the
3370 * "LAND" DoS attack, and also prevents an ACK storm
3371 * between two listening ports that have been sent forged
3372 * SYN segments, each with the source address of the other.
3374 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3375 (SEQ_GT(tp->snd_una, th->th_ack) ||
3376 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3377 rstreason = BANDLIM_RST_OPENPORT;
3378 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
3382 if (so->so_options & SO_DEBUG)
3383 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3386 TCP_PROBE3(debug__input, tp, th, m);
3387 tp->t_flags |= TF_ACKNOW;
3388 (void) tcp_output(tp);
3389 INP_WUNLOCK(tp->t_inpcb);
3395 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3396 INP_WUNLOCK(tp->t_inpcb);
3398 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3403 * Drop space held by incoming segment and return.
3406 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3407 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3410 TCP_PROBE3(debug__input, tp, th, m);
3412 INP_WUNLOCK(tp->t_inpcb);
3418 * Issue RST and make ACK acceptable to originator of segment.
3419 * The mbuf must still include the original packet header.
3423 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3424 int tlen, int rstreason)
3430 struct ip6_hdr *ip6;
3434 INP_LOCK_ASSERT(tp->t_inpcb);
3437 /* Don't bother if destination was broadcast/multicast. */
3438 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3441 if (mtod(m, struct ip *)->ip_v == 6) {
3442 ip6 = mtod(m, struct ip6_hdr *);
3443 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3444 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3446 /* IPv6 anycast check is done at tcp6_input() */
3449 #if defined(INET) && defined(INET6)
3454 ip = mtod(m, struct ip *);
3455 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3456 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3457 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3458 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3463 /* Perform bandwidth limiting. */
3464 if (badport_bandlim(rstreason) < 0)
3467 /* tcp_respond consumes the mbuf chain. */
3468 if (tcp_get_flags(th) & TH_ACK) {
3469 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3470 th->th_ack, TH_RST);
3472 if (tcp_get_flags(th) & TH_SYN)
3474 if (tcp_get_flags(th) & TH_FIN)
3476 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3477 (tcp_seq)0, TH_RST|TH_ACK);
3485 * Parse TCP options and place in tcpopt.
3488 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3493 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3495 if (opt == TCPOPT_EOL)
3497 if (opt == TCPOPT_NOP)
3503 if (optlen < 2 || optlen > cnt)
3508 if (optlen != TCPOLEN_MAXSEG)
3510 if (!(flags & TO_SYN))
3512 to->to_flags |= TOF_MSS;
3513 bcopy((char *)cp + 2,
3514 (char *)&to->to_mss, sizeof(to->to_mss));
3515 to->to_mss = ntohs(to->to_mss);
3518 if (optlen != TCPOLEN_WINDOW)
3520 if (!(flags & TO_SYN))
3522 to->to_flags |= TOF_SCALE;
3523 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3525 case TCPOPT_TIMESTAMP:
3526 if (optlen != TCPOLEN_TIMESTAMP)
3528 to->to_flags |= TOF_TS;
3529 bcopy((char *)cp + 2,
3530 (char *)&to->to_tsval, sizeof(to->to_tsval));
3531 to->to_tsval = ntohl(to->to_tsval);
3532 bcopy((char *)cp + 6,
3533 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3534 to->to_tsecr = ntohl(to->to_tsecr);
3536 case TCPOPT_SIGNATURE:
3538 * In order to reply to a host which has set the
3539 * TCP_SIGNATURE option in its initial SYN, we have
3540 * to record the fact that the option was observed
3541 * here for the syncache code to perform the correct
3544 if (optlen != TCPOLEN_SIGNATURE)
3546 to->to_flags |= TOF_SIGNATURE;
3547 to->to_signature = cp + 2;
3549 case TCPOPT_SACK_PERMITTED:
3550 if (optlen != TCPOLEN_SACK_PERMITTED)
3552 if (!(flags & TO_SYN))
3556 to->to_flags |= TOF_SACKPERM;
3559 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3563 to->to_flags |= TOF_SACK;
3564 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3565 to->to_sacks = cp + 2;
3566 TCPSTAT_INC(tcps_sack_rcv_blocks);
3568 case TCPOPT_FAST_OPEN:
3570 * Cookie length validation is performed by the
3571 * server side cookie checking code or the client
3572 * side cookie cache update code.
3574 if (!(flags & TO_SYN))
3576 if (!V_tcp_fastopen_client_enable &&
3577 !V_tcp_fastopen_server_enable)
3579 to->to_flags |= TOF_FASTOPEN;
3580 to->to_tfo_len = optlen - 2;
3581 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3590 * Pull out of band byte out of a segment so
3591 * it doesn't appear in the user's data queue.
3592 * It is still reflected in the segment length for
3593 * sequencing purposes.
3596 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3599 int cnt = off + th->th_urp - 1;
3602 if (m->m_len > cnt) {
3603 char *cp = mtod(m, caddr_t) + cnt;
3604 struct tcpcb *tp = sototcpcb(so);
3606 INP_WLOCK_ASSERT(tp->t_inpcb);
3609 tp->t_oobflags |= TCPOOB_HAVEDATA;
3610 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3612 if (m->m_flags & M_PKTHDR)
3621 panic("tcp_pulloutofband");
3625 * Collect new round-trip time estimate
3626 * and update averages and current timeout.
3629 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3633 INP_WLOCK_ASSERT(tp->t_inpcb);
3635 TCPSTAT_INC(tcps_rttupdated);
3638 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3639 imax(0, rtt * 1000 / hz));
3641 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3643 * srtt is stored as fixed point with 5 bits after the
3644 * binary point (i.e., scaled by 8). The following magic
3645 * is equivalent to the smoothing algorithm in rfc793 with
3646 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3647 * point). Adjust rtt to origin 0.
3649 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3650 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3652 if ((tp->t_srtt += delta) <= 0)
3656 * We accumulate a smoothed rtt variance (actually, a
3657 * smoothed mean difference), then set the retransmit
3658 * timer to smoothed rtt + 4 times the smoothed variance.
3659 * rttvar is stored as fixed point with 4 bits after the
3660 * binary point (scaled by 16). The following is
3661 * equivalent to rfc793 smoothing with an alpha of .75
3662 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3663 * rfc793's wired-in beta.
3667 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3668 if ((tp->t_rttvar += delta) <= 0)
3670 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3671 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3674 * No rtt measurement yet - use the unsmoothed rtt.
3675 * Set the variance to half the rtt (so our first
3676 * retransmit happens at 3*rtt).
3678 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3679 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3680 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3686 * the retransmit should happen at rtt + 4 * rttvar.
3687 * Because of the way we do the smoothing, srtt and rttvar
3688 * will each average +1/2 tick of bias. When we compute
3689 * the retransmit timer, we want 1/2 tick of rounding and
3690 * 1 extra tick because of +-1/2 tick uncertainty in the
3691 * firing of the timer. The bias will give us exactly the
3692 * 1.5 tick we need. But, because the bias is
3693 * statistical, we have to test that we don't drop below
3694 * the minimum feasible timer (which is 2 ticks).
3696 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3697 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3700 * We received an ack for a packet that wasn't retransmitted;
3701 * it is probably safe to discard any error indications we've
3702 * received recently. This isn't quite right, but close enough
3703 * for now (a route might have failed after we sent a segment,
3704 * and the return path might not be symmetrical).
3706 tp->t_softerror = 0;
3710 * Determine a reasonable value for maxseg size.
3711 * If the route is known, check route for mtu.
3712 * If none, use an mss that can be handled on the outgoing interface
3713 * without forcing IP to fragment. If no route is found, route has no mtu,
3714 * or the destination isn't local, use a default, hopefully conservative
3715 * size (usually 512 or the default IP max size, but no more than the mtu
3716 * of the interface), as we can't discover anything about intervening
3717 * gateways or networks. We also initialize the congestion/slow start
3718 * window to be a single segment if the destination isn't local.
3719 * While looking at the routing entry, we also initialize other path-dependent
3720 * parameters from pre-set or cached values in the routing entry.
3722 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3723 * IP options, e.g. IPSEC data, since length of this data may vary, and
3724 * thus it is calculated for every segment separately in tcp_output().
3726 * NOTE that this routine is only called when we process an incoming
3727 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3728 * settings are handled in tcp_mssopt().
3731 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3732 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3735 uint32_t maxmtu = 0;
3736 struct inpcb *inp = tp->t_inpcb;
3737 struct hc_metrics_lite metrics;
3739 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3740 size_t min_protoh = isipv6 ?
3741 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3742 sizeof (struct tcpiphdr);
3744 size_t min_protoh = sizeof(struct tcpiphdr);
3747 INP_WLOCK_ASSERT(tp->t_inpcb);
3750 min_protoh += V_tcp_udp_tunneling_overhead;
3751 if (mtuoffer != -1) {
3752 KASSERT(offer == -1, ("%s: conflict", __func__));
3753 offer = mtuoffer - min_protoh;
3759 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3760 tp->t_maxseg = V_tcp_v6mssdflt;
3763 #if defined(INET) && defined(INET6)
3768 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3769 tp->t_maxseg = V_tcp_mssdflt;
3774 * No route to sender, stay with default mss and return.
3778 * In case we return early we need to initialize metrics
3779 * to a defined state as tcp_hc_get() would do for us
3780 * if there was no cache hit.
3782 if (metricptr != NULL)
3783 bzero(metricptr, sizeof(struct hc_metrics_lite));
3787 /* What have we got? */
3791 * Offer == 0 means that there was no MSS on the SYN
3792 * segment, in this case we use tcp_mssdflt as
3793 * already assigned to t_maxseg above.
3795 offer = tp->t_maxseg;
3800 * Offer == -1 means that we didn't receive SYN yet.
3806 * Prevent DoS attack with too small MSS. Round up
3807 * to at least minmss.
3809 offer = max(offer, V_tcp_minmss);
3813 * rmx information is now retrieved from tcp_hostcache.
3815 tcp_hc_get(&inp->inp_inc, &metrics);
3816 if (metricptr != NULL)
3817 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3820 * If there's a discovered mtu in tcp hostcache, use it.
3821 * Else, use the link mtu.
3823 if (metrics.rmx_mtu)
3824 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3828 mss = maxmtu - min_protoh;
3829 if (!V_path_mtu_discovery &&
3830 !in6_localaddr(&inp->in6p_faddr))
3831 mss = min(mss, V_tcp_v6mssdflt);
3834 #if defined(INET) && defined(INET6)
3839 mss = maxmtu - min_protoh;
3840 if (!V_path_mtu_discovery &&
3841 !in_localaddr(inp->inp_faddr))
3842 mss = min(mss, V_tcp_mssdflt);
3846 * XXX - The above conditional (mss = maxmtu - min_protoh)
3847 * probably violates the TCP spec.
3848 * The problem is that, since we don't know the
3849 * other end's MSS, we are supposed to use a conservative
3850 * default. But, if we do that, then MTU discovery will
3851 * never actually take place, because the conservative
3852 * default is much less than the MTUs typically seen
3853 * on the Internet today. For the moment, we'll sweep
3854 * this under the carpet.
3856 * The conservative default might not actually be a problem
3857 * if the only case this occurs is when sending an initial
3858 * SYN with options and data to a host we've never talked
3859 * to before. Then, they will reply with an MSS value which
3860 * will get recorded and the new parameters should get
3861 * recomputed. For Further Study.
3864 mss = min(mss, offer);
3867 * Sanity check: make sure that maxseg will be large
3868 * enough to allow some data on segments even if the
3869 * all the option space is used (40bytes). Otherwise
3870 * funny things may happen in tcp_output.
3872 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3880 tcp_mss(struct tcpcb *tp, int offer)
3886 struct hc_metrics_lite metrics;
3887 struct tcp_ifcap cap;
3889 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3891 bzero(&cap, sizeof(cap));
3892 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3898 * If there's a pipesize, change the socket buffer to that size,
3899 * don't change if sb_hiwat is different than default (then it
3900 * has been changed on purpose with setsockopt).
3901 * Make the socket buffers an integral number of mss units;
3902 * if the mss is larger than the socket buffer, decrease the mss.
3904 so = inp->inp_socket;
3905 SOCKBUF_LOCK(&so->so_snd);
3906 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3907 bufsize = metrics.rmx_sendpipe;
3909 bufsize = so->so_snd.sb_hiwat;
3913 bufsize = roundup(bufsize, mss);
3914 if (bufsize > sb_max)
3916 if (bufsize > so->so_snd.sb_hiwat)
3917 (void)sbreserve_locked(so, SO_SND, bufsize, NULL);
3919 SOCKBUF_UNLOCK(&so->so_snd);
3921 * Sanity check: make sure that maxseg will be large
3922 * enough to allow some data on segments even if the
3923 * all the option space is used (40bytes). Otherwise
3924 * funny things may happen in tcp_output.
3926 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3928 tp->t_maxseg = max(mss, 64);
3930 SOCKBUF_LOCK(&so->so_rcv);
3931 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3932 bufsize = metrics.rmx_recvpipe;
3934 bufsize = so->so_rcv.sb_hiwat;
3935 if (bufsize > mss) {
3936 bufsize = roundup(bufsize, mss);
3937 if (bufsize > sb_max)
3939 if (bufsize > so->so_rcv.sb_hiwat)
3940 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL);
3942 SOCKBUF_UNLOCK(&so->so_rcv);
3944 /* Check the interface for TSO capabilities. */
3945 if (cap.ifcap & CSUM_TSO) {
3946 tp->t_flags |= TF_TSO;
3947 tp->t_tsomax = cap.tsomax;
3948 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3949 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3954 * Determine the MSS option to send on an outgoing SYN.
3957 tcp_mssopt(struct in_conninfo *inc)
3960 uint32_t thcmtu = 0;
3961 uint32_t maxmtu = 0;
3964 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3967 if (inc->inc_flags & INC_ISIPV6) {
3968 mss = V_tcp_v6mssdflt;
3969 maxmtu = tcp_maxmtu6(inc, NULL);
3970 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3973 #if defined(INET) && defined(INET6)
3978 mss = V_tcp_mssdflt;
3979 maxmtu = tcp_maxmtu(inc, NULL);
3980 min_protoh = sizeof(struct tcpiphdr);
3983 #if defined(INET6) || defined(INET)
3984 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3987 if (maxmtu && thcmtu)
3988 mss = min(maxmtu, thcmtu) - min_protoh;
3989 else if (maxmtu || thcmtu)
3990 mss = max(maxmtu, thcmtu) - min_protoh;
3996 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
3998 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3999 int maxseg = tcp_maxseg(tp);
4001 INP_WLOCK_ASSERT(tp->t_inpcb);
4004 * Compute the amount of data that this ACK is indicating
4005 * (del_data) and an estimate of how many bytes are in the
4008 if (tcp_is_sack_recovery(tp, to) ||
4009 (IN_CONGRECOVERY(tp->t_flags) &&
4010 !IN_FASTRECOVERY(tp->t_flags))) {
4011 del_data = tp->sackhint.delivered_data;
4012 if (V_tcp_do_newsack)
4013 pipe = tcp_compute_pipe(tp);
4015 pipe = (tp->snd_nxt - tp->snd_fack) +
4016 tp->sackhint.sack_bytes_rexmit;
4018 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg +
4019 tp->snd_recover - tp->snd_una))
4021 pipe = imax(0, tp->snd_max - tp->snd_una -
4022 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg);
4024 tp->sackhint.prr_delivered += del_data;
4026 * Proportional Rate Reduction
4028 if (pipe >= tp->snd_ssthresh) {
4029 if (tp->sackhint.recover_fs == 0)
4030 tp->sackhint.recover_fs =
4031 imax(1, tp->snd_nxt - tp->snd_una);
4032 snd_cnt = howmany((long)tp->sackhint.prr_delivered *
4033 tp->snd_ssthresh, tp->sackhint.recover_fs) -
4034 tp->sackhint.prr_out;
4036 if (V_tcp_do_prr_conservative || (del_data == 0))
4037 limit = tp->sackhint.prr_delivered -
4038 tp->sackhint.prr_out;
4040 limit = imax(tp->sackhint.prr_delivered -
4041 tp->sackhint.prr_out, del_data) +
4043 snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
4045 snd_cnt = imax(snd_cnt, 0) / maxseg;
4047 * Send snd_cnt new data into the network in response to this ack.
4048 * If there is going to be a SACK retransmission, adjust snd_cwnd
4051 if (IN_FASTRECOVERY(tp->t_flags)) {
4052 if (tcp_is_sack_recovery(tp, to)) {
4053 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
4054 tp->sackhint.sack_bytes_rexmit +
4057 tp->snd_cwnd = (tp->snd_max - tp->snd_una) +
4060 } else if (IN_CONGRECOVERY(tp->t_flags))
4061 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg);
4062 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd);
4066 * On a partial ack arrives, force the retransmission of the
4067 * next unacknowledged segment. Do not clear tp->t_dupacks.
4068 * By setting snd_nxt to ti_ack, this forces retransmission timer to
4072 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4074 tcp_seq onxt = tp->snd_nxt;
4075 uint32_t ocwnd = tp->snd_cwnd;
4076 u_int maxseg = tcp_maxseg(tp);
4078 INP_WLOCK_ASSERT(tp->t_inpcb);
4080 tcp_timer_activate(tp, TT_REXMT, 0);
4082 tp->snd_nxt = th->th_ack;
4084 * Set snd_cwnd to one segment beyond acknowledged offset.
4085 * (tp->snd_una has not yet been updated when this function is called.)
4087 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4088 tp->t_flags |= TF_ACKNOW;
4089 (void) tcp_output(tp);
4090 tp->snd_cwnd = ocwnd;
4091 if (SEQ_GT(onxt, tp->snd_nxt))
4094 * Partial window deflation. Relies on fact that tp->snd_una
4097 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4098 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4101 tp->snd_cwnd += maxseg;
4105 tcp_compute_pipe(struct tcpcb *tp)
4107 if (tp->t_fb->tfb_compute_pipe == NULL) {
4108 return (tp->snd_max - tp->snd_una +
4109 tp->sackhint.sack_bytes_rexmit -
4110 tp->sackhint.sacked_bytes);
4112 return((*tp->t_fb->tfb_compute_pipe)(tp));
4117 tcp_compute_initwnd(uint32_t maxseg)
4120 * Calculate the Initial Window, also used as Restart Window
4122 * RFC5681 Section 3.1 specifies the default conservative values.
4123 * RFC3390 specifies slightly more aggressive values.
4124 * RFC6928 increases it to ten segments.
4125 * Support for user specified value for initial flight size.
4127 if (V_tcp_initcwnd_segments)
4128 return min(V_tcp_initcwnd_segments * maxseg,
4129 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4130 else if (V_tcp_do_rfc3390)
4131 return min(4 * maxseg, max(2 * maxseg, 4380));
4133 /* Per RFC5681 Section 3.1 */
4135 return (2 * maxseg);
4136 else if (maxseg > 1095)
4137 return (3 * maxseg);
4139 return (4 * maxseg);