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"
59 #include <sys/param.h>
61 #include <sys/kernel.h>
63 #include <sys/hhook.h>
65 #include <sys/malloc.h>
67 #include <sys/proc.h> /* for proc0 declaration */
68 #include <sys/protosw.h>
69 #include <sys/qmath.h>
71 #include <sys/signalvar.h>
72 #include <sys/socket.h>
73 #include <sys/socketvar.h>
74 #include <sys/sysctl.h>
75 #include <sys/syslog.h>
76 #include <sys/systm.h>
77 #include <sys/stats.h>
79 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
84 #include <net/if_var.h>
85 #include <net/route.h>
88 #define TCPSTATES /* for logging */
90 #include <netinet/in.h>
91 #include <netinet/in_kdtrace.h>
92 #include <netinet/in_pcb.h>
93 #include <netinet/in_systm.h>
94 #include <netinet/ip.h>
95 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
96 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
97 #include <netinet/ip_var.h>
98 #include <netinet/ip_options.h>
99 #include <netinet/ip6.h>
100 #include <netinet/icmp6.h>
101 #include <netinet6/in6_pcb.h>
102 #include <netinet6/in6_var.h>
103 #include <netinet6/ip6_var.h>
104 #include <netinet6/nd6.h>
105 #include <netinet/tcp.h>
106 #include <netinet/tcp_fsm.h>
107 #include <netinet/tcp_seq.h>
108 #include <netinet/tcp_timer.h>
109 #include <netinet/tcp_var.h>
110 #include <netinet/tcp_log_buf.h>
111 #include <netinet6/tcp6_var.h>
112 #include <netinet/tcpip.h>
113 #include <netinet/cc/cc.h>
114 #include <netinet/tcp_fastopen.h>
116 #include <netinet/tcp_pcap.h>
118 #include <netinet/tcp_syncache.h>
120 #include <netinet/tcp_offload.h>
122 #include <netinet/tcp_ecn.h>
123 #include <netinet/udp.h>
125 #include <netipsec/ipsec_support.h>
127 #include <machine/in_cksum.h>
129 #include <security/mac/mac_framework.h>
131 const int tcprexmtthresh = 3;
133 VNET_DEFINE(int, tcp_log_in_vain) = 0;
134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
135 &VNET_NAME(tcp_log_in_vain), 0,
136 "Log all incoming TCP segments to closed ports");
138 VNET_DEFINE(int, blackhole) = 0;
139 #define V_blackhole VNET(blackhole)
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
141 &VNET_NAME(blackhole), 0,
142 "Do not send RST on segments to closed ports");
144 VNET_DEFINE(bool, blackhole_local) = false;
145 #define V_blackhole_local VNET(blackhole_local)
146 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET |
147 CTLFLAG_RW, &VNET_NAME(blackhole_local), false,
148 "Enforce net.inet.tcp.blackhole for locally originated packets");
150 VNET_DEFINE(int, tcp_delack_enabled) = 1;
151 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
152 &VNET_NAME(tcp_delack_enabled), 0,
153 "Delay ACK to try and piggyback it onto a data packet");
155 VNET_DEFINE(int, drop_synfin) = 0;
156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
157 &VNET_NAME(drop_synfin), 0,
158 "Drop TCP packets with SYN+FIN set");
160 VNET_DEFINE(int, tcp_do_prr_conservative) = 0;
161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW,
162 &VNET_NAME(tcp_do_prr_conservative), 0,
163 "Do conservative Proportional Rate Reduction");
165 VNET_DEFINE(int, tcp_do_prr) = 1;
166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW,
167 &VNET_NAME(tcp_do_prr), 1,
168 "Enable Proportional Rate Reduction per RFC 6937");
170 VNET_DEFINE(int, tcp_do_lrd) = 0;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW,
172 &VNET_NAME(tcp_do_lrd), 1,
173 "Perform Lost Retransmission Detection");
175 VNET_DEFINE(int, tcp_do_newcwv) = 0;
176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
177 &VNET_NAME(tcp_do_newcwv), 0,
178 "Enable New Congestion Window Validation per RFC7661");
180 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
181 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
182 &VNET_NAME(tcp_do_rfc3042), 0,
183 "Enable RFC 3042 (Limited Transmit)");
185 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
186 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
187 &VNET_NAME(tcp_do_rfc3390), 0,
188 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
190 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
192 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
193 "Slow-start flight size (initial congestion window) in number of segments");
195 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
197 &VNET_NAME(tcp_do_rfc3465), 0,
198 "Enable RFC 3465 (Appropriate Byte Counting)");
200 VNET_DEFINE(int, tcp_abc_l_var) = 2;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
202 &VNET_NAME(tcp_abc_l_var), 2,
203 "Cap the max cwnd increment during slow-start to this number of segments");
205 VNET_DEFINE(int, tcp_insecure_syn) = 0;
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
207 &VNET_NAME(tcp_insecure_syn), 0,
208 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
210 VNET_DEFINE(int, tcp_insecure_rst) = 0;
211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
212 &VNET_NAME(tcp_insecure_rst), 0,
213 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
215 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
216 #define V_tcp_recvspace VNET(tcp_recvspace)
217 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
218 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
220 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
222 &VNET_NAME(tcp_do_autorcvbuf), 0,
223 "Enable automatic receive buffer sizing");
225 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
226 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
227 &VNET_NAME(tcp_autorcvbuf_max), 0,
228 "Max size of automatic receive buffer");
230 VNET_DEFINE(struct inpcbinfo, tcbinfo);
233 * TCP statistics are stored in an array of counter(9)s, which size matches
234 * size of struct tcpstat. TCP running connection count is a regular array.
236 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
237 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
238 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
239 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
240 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
241 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
242 "TCP connection counts by TCP state");
245 * Kernel module interface for updating tcpstat. The first argument is an index
246 * into tcpstat treated as an array.
249 kmod_tcpstat_add(int statnum, int val)
252 counter_u64_add(VNET(tcpstat)[statnum], val);
256 * Make sure that we only start a SACK loss recovery when
257 * receiving a duplicate ACK with a SACK block, and also
258 * complete SACK loss recovery in case the other end
262 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to)
264 return ((tp->t_flags & TF_SACK_PERMIT) &&
265 ((to->to_flags & TOF_SACK) ||
266 (!TAILQ_EMPTY(&tp->snd_holes))));
271 * Wrapper for the TCP established input helper hook.
274 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
276 struct tcp_hhook_data hhook_data;
278 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
283 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
290 * CC wrapper hook functions
293 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
300 INP_WLOCK_ASSERT(tptoinpcb(tp));
302 tp->t_ccv.nsegs = nsegs;
303 tp->t_ccv.bytes_this_ack = BYTES_THIS_ACK(tp, th);
304 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
305 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
306 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
307 tp->t_ccv.flags |= CCF_CWND_LIMITED;
309 tp->t_ccv.flags &= ~CCF_CWND_LIMITED;
311 if (type == CC_ACK) {
313 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
314 ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
315 if (!IN_RECOVERY(tp->t_flags))
316 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
317 tp->t_ccv.bytes_this_ack / (tcp_maxseg(tp) * nsegs));
318 if ((tp->t_flags & TF_GPUTINPROG) &&
319 SEQ_GEQ(th->th_ack, tp->gput_ack)) {
321 * Compute goodput in bits per millisecond.
323 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) /
324 max(1, tcp_ts_getticks() - tp->gput_ts);
325 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
328 * XXXLAS: This is a temporary hack, and should be
329 * chained off VOI_TCP_GPUT when stats(9) grows an API
330 * to deal with chained VOIs.
332 if (tp->t_stats_gput_prev > 0)
333 stats_voi_update_abs_s32(tp->t_stats,
335 ((gput - tp->t_stats_gput_prev) * 100) /
336 tp->t_stats_gput_prev);
337 tp->t_flags &= ~TF_GPUTINPROG;
338 tp->t_stats_gput_prev = gput;
341 if (tp->snd_cwnd > tp->snd_ssthresh) {
342 tp->t_bytes_acked += tp->t_ccv.bytes_this_ack;
343 if (tp->t_bytes_acked >= tp->snd_cwnd) {
344 tp->t_bytes_acked -= tp->snd_cwnd;
345 tp->t_ccv.flags |= CCF_ABC_SENTAWND;
348 tp->t_ccv.flags &= ~CCF_ABC_SENTAWND;
349 tp->t_bytes_acked = 0;
353 if (CC_ALGO(tp)->ack_received != NULL) {
354 /* XXXLAS: Find a way to live without this */
355 tp->t_ccv.curack = th->th_ack;
356 CC_ALGO(tp)->ack_received(&tp->t_ccv, type);
359 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
364 cc_conn_init(struct tcpcb *tp)
366 struct hc_metrics_lite metrics;
367 struct inpcb *inp = tptoinpcb(tp);
371 INP_WLOCK_ASSERT(inp);
373 tcp_hc_get(&inp->inp_inc, &metrics);
374 maxseg = tcp_maxseg(tp);
376 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
378 TCPSTAT_INC(tcps_usedrtt);
379 if (metrics.rmx_rttvar) {
380 tp->t_rttvar = metrics.rmx_rttvar;
381 TCPSTAT_INC(tcps_usedrttvar);
383 /* default variation is +- 1 rtt */
385 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
387 TCPT_RANGESET(tp->t_rxtcur,
388 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
389 tp->t_rttmin, TCPTV_REXMTMAX);
391 if (metrics.rmx_ssthresh) {
393 * There's some sort of gateway or interface
394 * buffer limit on the path. Use this to set
395 * the slow start threshold, but set the
396 * threshold to no less than 2*mss.
398 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
399 TCPSTAT_INC(tcps_usedssthresh);
403 * Set the initial slow-start flight size.
405 * If a SYN or SYN/ACK was lost and retransmitted, we have to
406 * reduce the initial CWND to one segment as congestion is likely
407 * requiring us to be cautious.
409 if (tp->snd_cwnd == 1)
410 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
412 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
414 if (CC_ALGO(tp)->conn_init != NULL)
415 CC_ALGO(tp)->conn_init(&tp->t_ccv);
419 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
421 INP_WLOCK_ASSERT(tptoinpcb(tp));
424 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
429 if (!IN_FASTRECOVERY(tp->t_flags)) {
430 tp->snd_recover = tp->snd_max;
431 if (tp->t_flags2 & TF2_ECN_PERMIT)
432 tp->t_flags2 |= TF2_ECN_SND_CWR;
436 if (!IN_CONGRECOVERY(tp->t_flags) ||
438 * Allow ECN reaction on ACK to CWR, if
439 * that data segment was also CE marked.
441 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
442 EXIT_CONGRECOVERY(tp->t_flags);
443 TCPSTAT_INC(tcps_ecn_rcwnd);
444 tp->snd_recover = tp->snd_max + 1;
445 if (tp->t_flags2 & TF2_ECN_PERMIT)
446 tp->t_flags2 |= TF2_ECN_SND_CWR;
451 tp->t_bytes_acked = 0;
452 EXIT_RECOVERY(tp->t_flags);
453 if (tp->t_flags2 & TF2_ECN_PERMIT)
454 tp->t_flags2 |= TF2_ECN_SND_CWR;
457 TCPSTAT_INC(tcps_sndrexmitbad);
458 /* RTO was unnecessary, so reset everything. */
459 tp->snd_cwnd = tp->snd_cwnd_prev;
460 tp->snd_ssthresh = tp->snd_ssthresh_prev;
461 tp->snd_recover = tp->snd_recover_prev;
462 if (tp->t_flags & TF_WASFRECOVERY)
463 ENTER_FASTRECOVERY(tp->t_flags);
464 if (tp->t_flags & TF_WASCRECOVERY)
465 ENTER_CONGRECOVERY(tp->t_flags);
466 tp->snd_nxt = tp->snd_max;
467 tp->t_flags &= ~TF_PREVVALID;
472 if (CC_ALGO(tp)->cong_signal != NULL) {
474 tp->t_ccv.curack = th->th_ack;
475 CC_ALGO(tp)->cong_signal(&tp->t_ccv, type);
480 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
482 INP_WLOCK_ASSERT(tptoinpcb(tp));
484 /* XXXLAS: KASSERT that we're in recovery? */
486 if (CC_ALGO(tp)->post_recovery != NULL) {
487 tp->t_ccv.curack = th->th_ack;
488 CC_ALGO(tp)->post_recovery(&tp->t_ccv);
490 /* XXXLAS: EXIT_RECOVERY ? */
491 tp->t_bytes_acked = 0;
492 tp->sackhint.delivered_data = 0;
493 tp->sackhint.prr_out = 0;
497 * Indicate whether this ack should be delayed. We can delay the ack if
498 * following conditions are met:
499 * - There is no delayed ack timer in progress.
500 * - Our last ack wasn't a 0-sized window. We never want to delay
501 * the ack that opens up a 0-sized window.
502 * - LRO wasn't used for this segment. We make sure by checking that the
503 * segment size is not larger than the MSS.
505 #define DELAY_ACK(tp, tlen) \
506 ((!tcp_timer_active(tp, TT_DELACK) && \
507 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
508 (tlen <= tp->t_maxseg) && \
509 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
512 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos)
514 INP_WLOCK_ASSERT(tptoinpcb(tp));
516 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
517 switch (iptos & IPTOS_ECN_MASK) {
519 tp->t_ccv.flags |= CCF_IPHDR_CE;
525 case IPTOS_ECN_NOTECT:
526 tp->t_ccv.flags &= ~CCF_IPHDR_CE;
531 tp->t_ccv.flags |= CCF_TCPHDR_CWR;
533 tp->t_ccv.flags &= ~CCF_TCPHDR_CWR;
535 CC_ALGO(tp)->ecnpkt_handler(&tp->t_ccv);
537 if (tp->t_ccv.flags & CCF_ACKNOW) {
538 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
539 tp->t_flags |= TF_ACKNOW;
545 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
547 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos);
551 * TCP input handling is split into multiple parts:
552 * tcp6_input is a thin wrapper around tcp_input for the extended
553 * ip6_protox[] call format in ip6_input
554 * tcp_input handles primary segment validation, inpcb lookup and
555 * SYN processing on listen sockets
556 * tcp_do_segment processes the ACK and text of the segment for
557 * establishing, established and closing connections
561 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
564 struct in6_ifaddr *ia6;
568 if (m->m_len < *offp + sizeof(struct tcphdr)) {
569 m = m_pullup(m, *offp + sizeof(struct tcphdr));
572 TCPSTAT_INC(tcps_rcvshort);
573 return (IPPROTO_DONE);
578 * draft-itojun-ipv6-tcp-to-anycast
579 * better place to put this in?
581 ip6 = mtod(m, struct ip6_hdr *);
582 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
583 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
584 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
585 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
587 return (IPPROTO_DONE);
591 return (tcp_input_with_port(mp, offp, proto, port));
595 tcp6_input(struct mbuf **mp, int *offp, int proto)
598 return(tcp6_input_with_port(mp, offp, proto, 0));
603 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
605 struct mbuf *m = *mp;
606 struct tcphdr *th = NULL;
607 struct ip *ip = NULL;
608 struct inpcb *inp = NULL;
609 struct tcpcb *tp = NULL;
610 struct socket *so = NULL;
621 int rstreason = 0; /* For badport_bandlim accounting purposes */
624 struct m_tag *fwd_tag = NULL;
626 struct ip6_hdr *ip6 = NULL;
629 const void *ip6 = NULL;
631 struct tcpopt to; /* options in this segment */
632 char *s = NULL; /* address and port logging */
637 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
644 TCPSTAT_INC(tcps_rcvtotal);
648 ip6 = mtod(m, struct ip6_hdr *);
649 th = (struct tcphdr *)((caddr_t)ip6 + off0);
650 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
653 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
654 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
655 th->th_sum = m->m_pkthdr.csum_data;
657 th->th_sum = in6_cksum_pseudo(ip6, tlen,
658 IPPROTO_TCP, m->m_pkthdr.csum_data);
659 th->th_sum ^= 0xffff;
661 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
663 TCPSTAT_INC(tcps_rcvbadsum);
668 * Be proactive about unspecified IPv6 address in source.
669 * As we use all-zero to indicate unbounded/unconnected pcb,
670 * unspecified IPv6 address can be used to confuse us.
672 * Note that packets with unspecified IPv6 destination is
673 * already dropped in ip6_input.
675 KASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_dst),
676 ("%s: unspecified destination v6 address", __func__));
677 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
678 IP6STAT_INC(ip6s_badscope); /* XXX */
681 iptos = IPV6_TRAFFIC_CLASS(ip6);
684 #if defined(INET) && defined(INET6)
690 * Get IP and TCP header together in first mbuf.
691 * Note: IP leaves IP header in first mbuf.
693 if (off0 > sizeof (struct ip)) {
695 off0 = sizeof(struct ip);
697 if (m->m_len < sizeof (struct tcpiphdr)) {
698 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
700 TCPSTAT_INC(tcps_rcvshort);
701 return (IPPROTO_DONE);
704 ip = mtod(m, struct ip *);
705 th = (struct tcphdr *)((caddr_t)ip + off0);
706 tlen = ntohs(ip->ip_len) - off0;
711 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
712 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
713 th->th_sum = m->m_pkthdr.csum_data;
715 th->th_sum = in_pseudo(ip->ip_src.s_addr,
717 htonl(m->m_pkthdr.csum_data + tlen +
719 th->th_sum ^= 0xffff;
721 struct ipovly *ipov = (struct ipovly *)ip;
724 * Checksum extended TCP header and data.
728 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
729 ipov->ih_len = htons(tlen);
730 th->th_sum = in_cksum(m, len);
731 /* Reset length for SDT probes. */
732 ip->ip_len = htons(len);
735 /* Re-initialization for later version check */
737 ip->ip_v = IPVERSION;
738 ip->ip_hl = off0 >> 2;
741 if (th->th_sum && (port == 0)) {
742 TCPSTAT_INC(tcps_rcvbadsum);
745 KASSERT(ip->ip_dst.s_addr != INADDR_ANY,
746 ("%s: unspecified destination v4 address", __func__));
747 if (__predict_false(ip->ip_src.s_addr == INADDR_ANY)) {
748 IPSTAT_INC(ips_badaddr);
755 * Check that TCP offset makes sense,
756 * pull out TCP options and adjust length. XXX
758 off = th->th_off << 2;
759 if (off < sizeof (struct tcphdr) || off > tlen) {
760 TCPSTAT_INC(tcps_rcvbadoff);
763 tlen -= off; /* tlen is used instead of ti->ti_len */
764 if (off > sizeof (struct tcphdr)) {
767 if (m->m_len < off0 + off) {
768 m = m_pullup(m, off0 + off);
770 TCPSTAT_INC(tcps_rcvshort);
771 return (IPPROTO_DONE);
774 ip6 = mtod(m, struct ip6_hdr *);
775 th = (struct tcphdr *)((caddr_t)ip6 + off0);
778 #if defined(INET) && defined(INET6)
783 if (m->m_len < sizeof(struct ip) + off) {
784 if ((m = m_pullup(m, sizeof (struct ip) + off))
786 TCPSTAT_INC(tcps_rcvshort);
787 return (IPPROTO_DONE);
789 ip = mtod(m, struct ip *);
790 th = (struct tcphdr *)((caddr_t)ip + off0);
794 optlen = off - sizeof (struct tcphdr);
795 optp = (u_char *)(th + 1);
797 thflags = tcp_get_flags(th);
800 * Convert TCP protocol specific fields to host format.
802 tcp_fields_to_host(th);
805 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
807 drop_hdrlen = off0 + off;
810 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
814 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
816 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
819 #if defined(INET) && !defined(INET6)
820 (m->m_flags & M_IP_NEXTHOP)
823 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
826 * For initial SYN packets we don't need write lock on matching
827 * PCB, be it a listening one or a synchronized one. The packet
828 * shall not modify its state.
830 lookupflag = INPLOOKUP_WILDCARD |
831 ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ?
832 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB);
835 if (isipv6 && fwd_tag != NULL) {
836 struct sockaddr_in6 *next_hop6;
838 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
840 * Transparently forwarded. Pretend to be the destination.
841 * Already got one like this?
843 inp = in6_pcblookup_mbuf(&V_tcbinfo,
844 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
845 lookupflag & ~INPLOOKUP_WILDCARD, m->m_pkthdr.rcvif, m);
848 * It's new. Try to find the ambushing socket.
849 * Because we've rewritten the destination address,
850 * any hardware-generated hash is ignored.
852 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
853 th->th_sport, &next_hop6->sin6_addr,
854 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
855 th->th_dport, lookupflag, m->m_pkthdr.rcvif);
858 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
859 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag,
860 m->m_pkthdr.rcvif, m);
863 #if defined(INET6) && defined(INET)
867 if (fwd_tag != NULL) {
868 struct sockaddr_in *next_hop;
870 next_hop = (struct sockaddr_in *)(fwd_tag+1);
872 * Transparently forwarded. Pretend to be the destination.
873 * already got one like this?
875 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
876 ip->ip_dst, th->th_dport, lookupflag & ~INPLOOKUP_WILDCARD,
877 m->m_pkthdr.rcvif, m);
880 * It's new. Try to find the ambushing socket.
881 * Because we've rewritten the destination address,
882 * any hardware-generated hash is ignored.
884 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
885 th->th_sport, next_hop->sin_addr,
886 next_hop->sin_port ? ntohs(next_hop->sin_port) :
887 th->th_dport, lookupflag, m->m_pkthdr.rcvif);
890 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
891 th->th_sport, ip->ip_dst, th->th_dport, lookupflag,
892 m->m_pkthdr.rcvif, m);
896 * If the INPCB does not exist then all data in the incoming
897 * segment is discarded and an appropriate RST is sent back.
898 * XXX MRT Send RST using which routing table?
901 if (rstreason != 0) {
902 /* We came here after second (safety) lookup. */
903 MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0);
907 * Log communication attempts to ports that are not
910 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
911 V_tcp_log_in_vain == 2) {
912 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
913 log(LOG_INFO, "%s; %s: Connection attempt "
914 "to closed port\n", s, __func__);
917 * When blackholing do not respond with a RST but
918 * completely ignore the segment and drop it.
920 if (((V_blackhole == 1 && (thflags & TH_SYN)) ||
921 V_blackhole == 2) && (V_blackhole_local || (
923 isipv6 ? !in6_localaddr(&ip6->ip6_src) :
926 !in_localip(ip->ip_src)
933 rstreason = BANDLIM_RST_CLOSEDPORT;
936 INP_LOCK_ASSERT(inp);
938 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
939 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
940 !SOLISTENING(inp->inp_socket)) {
941 inp->inp_flowid = m->m_pkthdr.flowid;
942 inp->inp_flowtype = M_HASHTYPE_GET(m);
944 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
946 if (isipv6 && IPSEC_ENABLED(ipv6) &&
947 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
955 if (IPSEC_ENABLED(ipv4) &&
956 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
963 * Check the minimum TTL for socket.
965 if (inp->inp_ip_minttl != 0) {
968 if (inp->inp_ip_minttl > ip6->ip6_hlim)
972 if (inp->inp_ip_minttl > ip->ip_ttl)
977 switch (tp->t_state) {
980 * A previous connection in TIMEWAIT state is supposed to catch
981 * stray or duplicate segments arriving late. If this segment
982 * was a legitimate new connection attempt, the old INPCB gets
983 * removed and we can try again to find a listening socket.
985 tcp_dooptions(&to, optp, optlen,
986 (thflags & TH_SYN) ? TO_SYN : 0);
988 * tcp_twcheck unlocks the inp always, and frees the m if fails.
990 if (tcp_twcheck(inp, &to, th, m, tlen))
992 return (IPPROTO_DONE);
995 * The TCPCB may no longer exist if the connection is winding
996 * down or it is in the CLOSED state. Either way we drop the
997 * segment and send an appropriate response.
999 rstreason = BANDLIM_RST_CLOSEDPORT;
1003 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) {
1004 rstreason = BANDLIM_RST_CLOSEDPORT;
1009 if (tp->t_flags & TF_TOE) {
1010 tcp_offload_input(tp, m);
1011 m = NULL; /* consumed by the TOE driver */
1017 if (mac_inpcb_check_deliver(inp, m))
1020 so = inp->inp_socket;
1021 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1023 * When the socket is accepting connections (the INPCB is in LISTEN
1024 * state) we look into the SYN cache if this is a new connection
1025 * attempt or the completion of a previous one.
1027 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so),
1028 ("%s: so accepting but tp %p not listening", __func__, tp));
1029 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) {
1030 struct in_conninfo inc;
1032 bzero(&inc, sizeof(inc));
1035 inc.inc_flags |= INC_ISIPV6;
1036 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1037 inc.inc_flags |= INC_IPV6MINMTU;
1038 inc.inc6_faddr = ip6->ip6_src;
1039 inc.inc6_laddr = ip6->ip6_dst;
1043 inc.inc_faddr = ip->ip_src;
1044 inc.inc_laddr = ip->ip_dst;
1046 inc.inc_fport = th->th_sport;
1047 inc.inc_lport = th->th_dport;
1048 inc.inc_fibnum = so->so_fibnum;
1051 * Check for an existing connection attempt in syncache if
1052 * the flag is only ACK. A successful lookup creates a new
1053 * socket appended to the listen queue in SYN_RECEIVED state.
1055 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1057 * Parse the TCP options here because
1058 * syncookies need access to the reflected
1061 tcp_dooptions(&to, optp, optlen, 0);
1063 * NB: syncache_expand() doesn't unlock inp.
1065 rstreason = syncache_expand(&inc, &to, th, &so, m, port);
1066 if (rstreason < 0) {
1068 * A failing TCP MD5 signature comparison
1069 * must result in the segment being dropped
1070 * and must not produce any response back
1074 } else if (rstreason == 0) {
1076 * No syncache entry, or ACK was not for our
1077 * SYN/ACK. Do our protection against double
1078 * ACK. If peer sent us 2 ACKs, then for the
1079 * first one syncache_expand() successfully
1080 * converted syncache entry into a socket,
1081 * while we were waiting on the inpcb lock. We
1082 * don't want to sent RST for the second ACK,
1083 * so we perform second lookup without wildcard
1084 * match, hoping to find the new socket. If
1085 * the ACK is stray indeed, rstreason would
1086 * hint the above code that the lookup was a
1089 * NB: syncache did its own logging
1090 * of the failure cause.
1093 rstreason = BANDLIM_RST_OPENPORT;
1094 lookupflag &= ~INPLOOKUP_WILDCARD;
1100 * We completed the 3-way handshake
1101 * but could not allocate a socket
1102 * either due to memory shortage,
1103 * listen queue length limits or
1104 * global socket limits. Send RST
1105 * or wait and have the remote end
1106 * retransmit the ACK for another
1109 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1110 log(LOG_DEBUG, "%s; %s: Listen socket: "
1111 "Socket allocation failed due to "
1112 "limits or memory shortage, %s\n",
1114 V_tcp_sc_rst_sock_fail ?
1115 "sending RST" : "try again");
1116 if (V_tcp_sc_rst_sock_fail) {
1117 rstreason = BANDLIM_UNLIMITED;
1123 * Socket is created in state SYN_RECEIVED.
1124 * Unlock the listen socket, lock the newly
1125 * created socket and update the tp variable.
1126 * If we came here via jump to tfo_socket_result,
1127 * then listening socket is read-locked.
1129 INP_UNLOCK(inp); /* listen socket */
1130 inp = sotoinpcb(so);
1132 * New connection inpcb is already locked by
1133 * syncache_expand().
1135 INP_WLOCK_ASSERT(inp);
1136 tp = intotcpcb(inp);
1137 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1138 ("%s: ", __func__));
1140 * Process the segment and the data it
1141 * contains. tcp_do_segment() consumes
1142 * the mbuf chain and unlocks the inpcb.
1144 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1145 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen,
1147 return (IPPROTO_DONE);
1150 * Segment flag validation for new connection attempts:
1152 * Our (SYN|ACK) response was rejected.
1153 * Check with syncache and remove entry to prevent
1156 * NB: syncache_chkrst does its own logging of failure
1159 if (thflags & TH_RST) {
1160 syncache_chkrst(&inc, th, m, port);
1164 * We can't do anything without SYN.
1166 if ((thflags & TH_SYN) == 0) {
1167 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1168 log(LOG_DEBUG, "%s; %s: Listen socket: "
1169 "SYN is missing, segment ignored\n",
1171 TCPSTAT_INC(tcps_badsyn);
1175 * (SYN|ACK) is bogus on a listen socket.
1177 if (thflags & TH_ACK) {
1178 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1179 log(LOG_DEBUG, "%s; %s: Listen socket: "
1180 "SYN|ACK invalid, segment rejected\n",
1182 syncache_badack(&inc, port); /* XXX: Not needed! */
1183 TCPSTAT_INC(tcps_badsyn);
1184 rstreason = BANDLIM_RST_OPENPORT;
1188 * If the drop_synfin option is enabled, drop all
1189 * segments with both the SYN and FIN bits set.
1190 * This prevents e.g. nmap from identifying the
1192 * XXX: Poor reasoning. nmap has other methods
1193 * and is constantly refining its stack detection
1195 * XXX: This is a violation of the TCP specification
1196 * and was used by RFC1644.
1198 if ((thflags & TH_FIN) && V_drop_synfin) {
1199 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1200 log(LOG_DEBUG, "%s; %s: Listen socket: "
1201 "SYN|FIN segment ignored (based on "
1202 "sysctl setting)\n", s, __func__);
1203 TCPSTAT_INC(tcps_badsyn);
1207 * Segment's flags are (SYN) or (SYN|FIN).
1209 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1210 * as they do not affect the state of the TCP FSM.
1211 * The data pointed to by TH_URG and th_urp is ignored.
1213 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1214 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1215 KASSERT(thflags & (TH_SYN),
1216 ("%s: Listen socket: TH_SYN not set", __func__));
1217 INP_RLOCK_ASSERT(inp);
1220 * If deprecated address is forbidden,
1221 * we do not accept SYN to deprecated interface
1222 * address to prevent any new inbound connection from
1223 * getting established.
1224 * When we do not accept SYN, we send a TCP RST,
1225 * with deprecated source address (instead of dropping
1226 * it). We compromise it as it is much better for peer
1227 * to send a RST, and RST will be the final packet
1230 * If we do not forbid deprecated addresses, we accept
1231 * the SYN packet. RFC2462 does not suggest dropping
1233 * If we decipher RFC2462 5.5.4, it says like this:
1234 * 1. use of deprecated addr with existing
1235 * communication is okay - "SHOULD continue to be
1237 * 2. use of it with new communication:
1238 * (2a) "SHOULD NOT be used if alternate address
1239 * with sufficient scope is available"
1240 * (2b) nothing mentioned otherwise.
1241 * Here we fall into (2b) case as we have no choice in
1242 * our source address selection - we must obey the peer.
1244 * The wording in RFC2462 is confusing, and there are
1245 * multiple description text for deprecated address
1246 * handling - worse, they are not exactly the same.
1247 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1249 if (isipv6 && !V_ip6_use_deprecated) {
1250 struct in6_ifaddr *ia6;
1252 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1254 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1255 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1256 log(LOG_DEBUG, "%s; %s: Listen socket: "
1257 "Connection attempt to deprecated "
1258 "IPv6 address rejected\n",
1260 rstreason = BANDLIM_RST_OPENPORT;
1266 * Basic sanity checks on incoming SYN requests:
1267 * Don't respond if the destination is a link layer
1268 * broadcast according to RFC1122 4.2.3.10, p. 104.
1269 * If it is from this socket it must be forged.
1270 * Don't respond if the source or destination is a
1271 * global or subnet broad- or multicast address.
1272 * Note that it is quite possible to receive unicast
1273 * link-layer packets with a broadcast IP address. Use
1274 * in_broadcast() to find them.
1276 if (m->m_flags & (M_BCAST|M_MCAST)) {
1277 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1278 log(LOG_DEBUG, "%s; %s: Listen socket: "
1279 "Connection attempt from broad- or multicast "
1280 "link layer address ignored\n", s, __func__);
1285 if (th->th_dport == th->th_sport &&
1286 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1287 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1288 log(LOG_DEBUG, "%s; %s: Listen socket: "
1289 "Connection attempt to/from self "
1290 "ignored\n", s, __func__);
1293 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1294 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1295 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1296 log(LOG_DEBUG, "%s; %s: Listen socket: "
1297 "Connection attempt from/to multicast "
1298 "address ignored\n", s, __func__);
1303 #if defined(INET) && defined(INET6)
1308 if (th->th_dport == th->th_sport &&
1309 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1310 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1311 log(LOG_DEBUG, "%s; %s: Listen socket: "
1312 "Connection attempt from/to self "
1313 "ignored\n", s, __func__);
1316 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1317 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1318 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1319 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1320 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1321 log(LOG_DEBUG, "%s; %s: Listen socket: "
1322 "Connection attempt from/to broad- "
1323 "or multicast address ignored\n",
1330 * SYN appears to be valid. Create compressed TCP state
1333 TCP_PROBE3(debug__input, tp, th, m);
1334 tcp_dooptions(&to, optp, optlen, TO_SYN);
1335 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL,
1336 iptos, port)) != NULL)
1337 goto tfo_socket_result;
1340 * Entry added to syncache and mbuf consumed.
1341 * Only the listen socket is unlocked by syncache_add().
1343 return (IPPROTO_DONE);
1344 } else if (tp->t_state == TCPS_LISTEN) {
1346 * When a listen socket is torn down the SO_ACCEPTCONN
1347 * flag is removed first while connections are drained
1348 * from the accept queue in a unlock/lock cycle of the
1349 * ACCEPT_LOCK, opening a race condition allowing a SYN
1350 * attempt go through unhandled.
1354 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1355 if (tp->t_flags & TF_SIGNATURE) {
1356 tcp_dooptions(&to, optp, optlen, thflags);
1357 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1358 TCPSTAT_INC(tcps_sig_err_nosigopt);
1361 if (!TCPMD5_ENABLED() ||
1362 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1366 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1369 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1370 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1371 * the inpcb, and unlocks pcbinfo.
1373 * XXXGL: in case of a pure SYN arriving on existing connection
1374 * TCP stacks won't need to modify the PCB, they would either drop
1375 * the segment silently, or send a challenge ACK. However, we try
1376 * to upgrade the lock, because calling convention for stacks is
1377 * write-lock on PCB. If upgrade fails, drop the SYN.
1379 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0)
1382 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, tlen, iptos);
1383 return (IPPROTO_DONE);
1386 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1389 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1392 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1393 m = NULL; /* mbuf chain got consumed. */
1398 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1408 return (IPPROTO_DONE);
1412 * Automatic sizing of receive socket buffer. Often the send
1413 * buffer size is not optimally adjusted to the actual network
1414 * conditions at hand (delay bandwidth product). Setting the
1415 * buffer size too small limits throughput on links with high
1416 * bandwidth and high delay (eg. trans-continental/oceanic links).
1418 * On the receive side the socket buffer memory is only rarely
1419 * used to any significant extent. This allows us to be much
1420 * more aggressive in scaling the receive socket buffer. For
1421 * the case that the buffer space is actually used to a large
1422 * extent and we run out of kernel memory we can simply drop
1423 * the new segments; TCP on the sender will just retransmit it
1424 * later. Setting the buffer size too big may only consume too
1425 * much kernel memory if the application doesn't read() from
1426 * the socket or packet loss or reordering makes use of the
1429 * The criteria to step up the receive buffer one notch are:
1430 * 1. Application has not set receive buffer size with
1431 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1432 * 2. the number of bytes received during 1/2 of an sRTT
1433 * is at least 3/8 of the current socket buffer size.
1434 * 3. receive buffer size has not hit maximal automatic size;
1436 * If all of the criteria are met we increaset the socket buffer
1437 * by a 1/2 (bounded by the max). This allows us to keep ahead
1438 * of slow-start but also makes it so our peer never gets limited
1439 * by our rwnd which we then open up causing a burst.
1441 * This algorithm does two steps per RTT at most and only if
1442 * we receive a bulk stream w/o packet losses or reorderings.
1443 * Shrinking the buffer during idle times is not necessary as
1444 * it doesn't consume any memory when idle.
1446 * TODO: Only step up if the application is actually serving
1447 * the buffer to better manage the socket buffer resources.
1450 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1451 struct tcpcb *tp, int tlen)
1455 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1456 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1457 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1458 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1459 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1460 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1461 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1463 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1465 /* Start over with next RTT. */
1469 tp->rfbuf_cnt += tlen; /* add up */
1475 tcp_input(struct mbuf **mp, int *offp, int proto)
1477 return(tcp_input_with_port(mp, offp, proto, 0));
1481 tcp_handle_wakeup(struct tcpcb *tp)
1484 INP_WLOCK_ASSERT(tptoinpcb(tp));
1486 if (tp->t_flags & TF_WAKESOR) {
1487 struct socket *so = tptosocket(tp);
1489 tp->t_flags &= ~TF_WAKESOR;
1490 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1491 sorwakeup_locked(so);
1496 tcp_do_segment(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th,
1497 int drop_hdrlen, int tlen, uint8_t iptos)
1500 int acked, ourfinisacked, needoutput = 0, sack_changed;
1501 int rstreason, todrop, win, incforsyn = 0;
1505 struct inpcb *inp = tptoinpcb(tp);
1506 struct socket *so = tptosocket(tp);
1507 struct in_conninfo *inc = &inp->inp_inc;
1513 thflags = tcp_get_flags(th);
1514 tp->sackhint.last_sack_ack = 0;
1516 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1519 INP_WLOCK_ASSERT(inp);
1520 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1522 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1526 /* Save segment, if requested. */
1527 tcp_pcap_add(th, m, &(tp->t_inpkts));
1529 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1532 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1533 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1534 log(LOG_DEBUG, "%s; %s: "
1535 "SYN|FIN segment ignored (based on "
1536 "sysctl setting)\n", s, __func__);
1543 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1544 * check SEQ.ACK first.
1546 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1547 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1548 rstreason = BANDLIM_UNLIMITED;
1549 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
1554 * Segment received on connection.
1555 * Reset idle time and keep-alive timer.
1556 * XXX: This should be done after segment
1557 * validation to ignore broken/spoofed segs.
1559 if (tp->t_idle_reduce &&
1560 (tp->snd_max == tp->snd_una) &&
1561 ((ticks - tp->t_rcvtime) >= tp->t_rxtcur))
1563 tp->t_rcvtime = ticks;
1565 if (thflags & TH_FIN)
1566 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN);
1568 * Scale up the window into a 32-bit value.
1569 * For the SYN_SENT state the scale is zero.
1571 tiwin = th->th_win << tp->snd_scale;
1573 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1577 * TCP ECN processing.
1579 if (tcp_ecn_input_segment(tp, thflags, tlen,
1580 tcp_packets_this_ack(tp, th->th_ack),
1582 cc_cong_signal(tp, th, CC_ECN);
1585 * Parse options on any incoming segment.
1587 tcp_dooptions(&to, (u_char *)(th + 1),
1588 (th->th_off << 2) - sizeof(struct tcphdr),
1589 (thflags & TH_SYN) ? TO_SYN : 0);
1591 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1592 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1593 (to.to_flags & TOF_SIGNATURE) == 0) {
1594 TCPSTAT_INC(tcps_sig_err_sigopt);
1595 /* XXX: should drop? */
1599 * If echoed timestamp is later than the current time,
1600 * fall back to non RFC1323 RTT calculation. Normalize
1601 * timestamp if syncookies were used when this connection
1604 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1605 to.to_tsecr -= tp->ts_offset;
1606 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1608 else if (tp->t_rxtshift == 1 &&
1609 tp->t_flags & TF_PREVVALID &&
1610 tp->t_badrxtwin != 0 &&
1611 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
1612 cc_cong_signal(tp, th, CC_RTO_ERR);
1615 * Process options only when we get SYN/ACK back. The SYN case
1616 * for incoming connections is handled in tcp_syncache.
1617 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1618 * or <SYN,ACK>) segment itself is never scaled.
1619 * XXX this is traditional behavior, may need to be cleaned up.
1621 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1622 /* Handle parallel SYN for ECN */
1623 tcp_ecn_input_parallel_syn(tp, thflags, iptos);
1624 if ((to.to_flags & TOF_SCALE) &&
1625 (tp->t_flags & TF_REQ_SCALE) &&
1626 !(tp->t_flags & TF_NOOPT)) {
1627 tp->t_flags |= TF_RCVD_SCALE;
1628 tp->snd_scale = to.to_wscale;
1630 tp->t_flags &= ~TF_REQ_SCALE;
1632 * Initial send window. It will be updated with
1633 * the next incoming segment to the scaled value.
1635 tp->snd_wnd = th->th_win;
1636 if ((to.to_flags & TOF_TS) &&
1637 (tp->t_flags & TF_REQ_TSTMP) &&
1638 !(tp->t_flags & TF_NOOPT)) {
1639 tp->t_flags |= TF_RCVD_TSTMP;
1640 tp->ts_recent = to.to_tsval;
1641 tp->ts_recent_age = tcp_ts_getticks();
1643 tp->t_flags &= ~TF_REQ_TSTMP;
1644 if (to.to_flags & TOF_MSS)
1645 tcp_mss(tp, to.to_mss);
1646 if ((tp->t_flags & TF_SACK_PERMIT) &&
1647 (!(to.to_flags & TOF_SACKPERM) ||
1648 (tp->t_flags & TF_NOOPT)))
1649 tp->t_flags &= ~TF_SACK_PERMIT;
1650 if (IS_FASTOPEN(tp->t_flags)) {
1651 if ((to.to_flags & TOF_FASTOPEN) &&
1652 !(tp->t_flags & TF_NOOPT)) {
1655 if (to.to_flags & TOF_MSS)
1658 if ((inp->inp_vflag & INP_IPV6) != 0)
1662 tcp_fastopen_update_cache(tp, mss,
1663 to.to_tfo_len, to.to_tfo_cookie);
1665 tcp_fastopen_disable_path(tp);
1670 * If timestamps were negotiated during SYN/ACK and a
1671 * segment without a timestamp is received, silently drop
1672 * the segment, unless it is a RST segment or missing timestamps are
1674 * See section 3.2 of RFC 7323.
1676 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1677 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1678 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1679 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1680 "segment processed normally\n",
1685 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1686 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1687 "segment silently dropped\n", s, __func__);
1694 * If timestamps were not negotiated during SYN/ACK and a
1695 * segment with a timestamp is received, ignore the
1696 * timestamp and process the packet normally.
1697 * See section 3.2 of RFC 7323.
1699 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1700 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1701 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1702 "segment processed normally\n", s, __func__);
1708 * Header prediction: check for the two common cases
1709 * of a uni-directional data xfer. If the packet has
1710 * no control flags, is in-sequence, the window didn't
1711 * change and we're not retransmitting, it's a
1712 * candidate. If the length is zero and the ack moved
1713 * forward, we're the sender side of the xfer. Just
1714 * free the data acked & wake any higher level process
1715 * that was blocked waiting for space. If the length
1716 * is non-zero and the ack didn't move, we're the
1717 * receiver side. If we're getting packets in-order
1718 * (the reassembly queue is empty), add the data to
1719 * the socket buffer and note that we need a delayed ack.
1720 * Make sure that the hidden state-flags are also off.
1721 * Since we check for TCPS_ESTABLISHED first, it can only
1724 if (tp->t_state == TCPS_ESTABLISHED &&
1725 th->th_seq == tp->rcv_nxt &&
1726 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1727 tp->snd_nxt == tp->snd_max &&
1728 tiwin && tiwin == tp->snd_wnd &&
1729 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1731 ((to.to_flags & TOF_TS) == 0 ||
1732 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1734 * If last ACK falls within this segment's sequence numbers,
1735 * record the timestamp.
1736 * NOTE that the test is modified according to the latest
1737 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1739 if ((to.to_flags & TOF_TS) != 0 &&
1740 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1741 tp->ts_recent_age = tcp_ts_getticks();
1742 tp->ts_recent = to.to_tsval;
1746 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1747 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1748 !IN_RECOVERY(tp->t_flags) &&
1749 (to.to_flags & TOF_SACK) == 0 &&
1750 TAILQ_EMPTY(&tp->snd_holes)) {
1752 * This is a pure ack for outstanding data.
1754 TCPSTAT_INC(tcps_predack);
1757 * "bad retransmit" recovery without timestamps.
1759 if ((to.to_flags & TOF_TS) == 0 &&
1760 tp->t_rxtshift == 1 &&
1761 tp->t_flags & TF_PREVVALID &&
1762 tp->t_badrxtwin != 0 &&
1763 TSTMP_LT(ticks, tp->t_badrxtwin)) {
1764 cc_cong_signal(tp, th, CC_RTO_ERR);
1768 * Recalculate the transmit timer / rtt.
1770 * Some boxes send broken timestamp replies
1771 * during the SYN+ACK phase, ignore
1772 * timestamps of 0 or we could calculate a
1773 * huge RTT and blow up the retransmit timer.
1775 if ((to.to_flags & TOF_TS) != 0 &&
1779 t = tcp_ts_getticks() - to.to_tsecr;
1780 if (!tp->t_rttlow || tp->t_rttlow > t)
1783 TCP_TS_TO_TICKS(t) + 1);
1784 } else if (tp->t_rtttime &&
1785 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1786 if (!tp->t_rttlow ||
1787 tp->t_rttlow > ticks - tp->t_rtttime)
1788 tp->t_rttlow = ticks - tp->t_rtttime;
1790 ticks - tp->t_rtttime);
1792 acked = BYTES_THIS_ACK(tp, th);
1795 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1796 hhook_run_tcp_est_in(tp, th, &to);
1799 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1800 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1801 sbdrop(&so->so_snd, acked);
1802 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1803 SEQ_LEQ(th->th_ack, tp->snd_recover))
1804 tp->snd_recover = th->th_ack - 1;
1807 * Let the congestion control algorithm update
1808 * congestion control related information. This
1809 * typically means increasing the congestion
1812 cc_ack_received(tp, th, nsegs, CC_ACK);
1814 tp->snd_una = th->th_ack;
1816 * Pull snd_wl2 up to prevent seq wrap relative
1819 tp->snd_wl2 = th->th_ack;
1824 * If all outstanding data are acked, stop
1825 * retransmit timer, otherwise restart timer
1826 * using current (possibly backed-off) value.
1827 * If process is waiting for space,
1828 * wakeup/selwakeup/signal. If data
1829 * are ready to send, let tcp_output
1830 * decide between more output or persist.
1832 TCP_PROBE3(debug__input, tp, th, m);
1834 * Clear t_acktime if remote side has ACKd
1835 * all data in the socket buffer.
1836 * Otherwise, update t_acktime if we received
1837 * a sufficiently large ACK.
1839 if (sbavail(&so->so_snd) == 0)
1842 tp->t_acktime = ticks;
1843 if (tp->snd_una == tp->snd_max)
1844 tcp_timer_activate(tp, TT_REXMT, 0);
1845 else if (!tcp_timer_active(tp, TT_PERSIST))
1846 tcp_timer_activate(tp, TT_REXMT,
1849 if (sbavail(&so->so_snd))
1850 (void) tcp_output(tp);
1853 } else if (th->th_ack == tp->snd_una &&
1854 tlen <= sbspace(&so->so_rcv)) {
1855 int newsize = 0; /* automatic sockbuf scaling */
1858 * This is a pure, in-sequence data packet with
1859 * nothing on the reassembly queue and we have enough
1860 * buffer space to take it.
1862 /* Clean receiver SACK report if present */
1863 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1864 tcp_clean_sackreport(tp);
1865 TCPSTAT_INC(tcps_preddat);
1866 tp->rcv_nxt += tlen;
1868 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1869 (tp->t_fbyte_in == 0)) {
1870 tp->t_fbyte_in = ticks;
1871 if (tp->t_fbyte_in == 0)
1873 if (tp->t_fbyte_out && tp->t_fbyte_in)
1874 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1877 * Pull snd_wl1 up to prevent seq wrap relative to
1880 tp->snd_wl1 = th->th_seq;
1882 * Pull rcv_up up to prevent seq wrap relative to
1885 tp->rcv_up = tp->rcv_nxt;
1886 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1887 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1888 TCP_PROBE3(debug__input, tp, th, m);
1890 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1892 /* Add data to socket buffer. */
1893 SOCKBUF_LOCK(&so->so_rcv);
1894 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1898 * Set new socket buffer size.
1899 * Give up when limit is reached.
1902 if (!sbreserve_locked(so, SO_RCV,
1904 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1905 m_adj(m, drop_hdrlen); /* delayed header drop */
1906 sbappendstream_locked(&so->so_rcv, m, 0);
1908 /* NB: sorwakeup_locked() does an implicit unlock. */
1909 sorwakeup_locked(so);
1910 if (DELAY_ACK(tp, tlen)) {
1911 tp->t_flags |= TF_DELACK;
1913 tp->t_flags |= TF_ACKNOW;
1921 * Calculate amount of space in receive window,
1922 * and then do TCP input processing.
1923 * Receive window is amount of space in rcv queue,
1924 * but not less than advertised window.
1926 win = sbspace(&so->so_rcv);
1929 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1931 switch (tp->t_state) {
1933 * If the state is SYN_RECEIVED:
1934 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1936 case TCPS_SYN_RECEIVED:
1937 if ((thflags & TH_ACK) &&
1938 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1939 SEQ_GT(th->th_ack, tp->snd_max))) {
1940 rstreason = BANDLIM_RST_OPENPORT;
1941 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
1944 if (IS_FASTOPEN(tp->t_flags)) {
1946 * When a TFO connection is in SYN_RECEIVED, the
1947 * only valid packets are the initial SYN, a
1948 * retransmit/copy of the initial SYN (possibly with
1949 * a subset of the original data), a valid ACK, a
1952 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1953 rstreason = BANDLIM_RST_OPENPORT;
1954 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
1956 } else if (thflags & TH_SYN) {
1957 /* non-initial SYN is ignored */
1958 if ((tcp_timer_active(tp, TT_DELACK) ||
1959 tcp_timer_active(tp, TT_REXMT)))
1961 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1968 * If the state is SYN_SENT:
1969 * if seg contains a RST with valid ACK (SEQ.ACK has already
1970 * been verified), then drop the connection.
1971 * if seg contains a RST without an ACK, drop the seg.
1972 * if seg does not contain SYN, then drop the seg.
1973 * Otherwise this is an acceptable SYN segment
1974 * initialize tp->rcv_nxt and tp->irs
1975 * if seg contains ack then advance tp->snd_una
1976 * if seg contains an ECE and ECN support is enabled, the stream
1978 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1979 * arrange for segment to be acked (eventually)
1980 * continue processing rest of data/controls, beginning with URG
1983 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1984 TCP_PROBE5(connect__refused, NULL, tp,
1986 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
1987 tp = tcp_drop(tp, ECONNREFUSED);
1989 if (thflags & TH_RST)
1991 if (!(thflags & TH_SYN))
1994 tp->irs = th->th_seq;
1996 if (thflags & TH_ACK) {
1997 int tfo_partial_ack = 0;
1999 TCPSTAT_INC(tcps_connects);
2002 mac_socketpeer_set_from_mbuf(m, so);
2004 /* Do window scaling on this connection? */
2005 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2006 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2007 tp->rcv_scale = tp->request_r_scale;
2009 tp->rcv_adv += min(tp->rcv_wnd,
2010 TCP_MAXWIN << tp->rcv_scale);
2011 tp->snd_una++; /* SYN is acked */
2013 * If not all the data that was sent in the TFO SYN
2014 * has been acked, resend the remainder right away.
2016 if (IS_FASTOPEN(tp->t_flags) &&
2017 (tp->snd_una != tp->snd_max)) {
2018 tp->snd_nxt = th->th_ack;
2019 tfo_partial_ack = 1;
2022 * If there's data, delay ACK; if there's also a FIN
2023 * ACKNOW will be turned on later.
2025 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2026 tcp_timer_activate(tp, TT_DELACK,
2029 tp->t_flags |= TF_ACKNOW;
2031 tcp_ecn_input_syn_sent(tp, thflags, iptos);
2034 * Received <SYN,ACK> in SYN_SENT[*] state.
2036 * SYN_SENT --> ESTABLISHED
2037 * SYN_SENT* --> FIN_WAIT_1
2039 tp->t_starttime = ticks;
2040 if (tp->t_flags & TF_NEEDFIN) {
2041 tp->t_acktime = ticks;
2042 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2043 tp->t_flags &= ~TF_NEEDFIN;
2046 tcp_state_change(tp, TCPS_ESTABLISHED);
2047 TCP_PROBE5(connect__established, NULL, tp,
2050 tcp_timer_activate(tp, TT_KEEP,
2055 * Received initial SYN in SYN-SENT[*] state =>
2056 * simultaneous open.
2057 * If it succeeds, connection is * half-synchronized.
2058 * Otherwise, do 3-way handshake:
2059 * SYN-SENT -> SYN-RECEIVED
2060 * SYN-SENT* -> SYN-RECEIVED*
2062 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN);
2063 tcp_timer_activate(tp, TT_REXMT, 0);
2064 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2068 * Advance th->th_seq to correspond to first data byte.
2069 * If data, trim to stay within window,
2070 * dropping FIN if necessary.
2073 if (tlen > tp->rcv_wnd) {
2074 todrop = tlen - tp->rcv_wnd;
2078 TCPSTAT_INC(tcps_rcvpackafterwin);
2079 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2081 tp->snd_wl1 = th->th_seq - 1;
2082 tp->rcv_up = th->th_seq;
2084 * Client side of transaction: already sent SYN and data.
2085 * If the remote host used T/TCP to validate the SYN,
2086 * our data will be ACK'd; if so, enter normal data segment
2087 * processing in the middle of step 5, ack processing.
2088 * Otherwise, goto step 6.
2090 if (thflags & TH_ACK)
2097 * States other than LISTEN or SYN_SENT.
2098 * First check the RST flag and sequence number since reset segments
2099 * are exempt from the timestamp and connection count tests. This
2100 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2101 * below which allowed reset segments in half the sequence space
2102 * to fall though and be processed (which gives forged reset
2103 * segments with a random sequence number a 50 percent chance of
2104 * killing a connection).
2105 * Then check timestamp, if present.
2106 * Then check the connection count, if present.
2107 * Then check that at least some bytes of segment are within
2108 * receive window. If segment begins before rcv_nxt,
2109 * drop leading data (and SYN); if nothing left, just ack.
2111 if (thflags & TH_RST) {
2113 * RFC5961 Section 3.2
2115 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2116 * - If RST is in window, we send challenge ACK.
2118 * Note: to take into account delayed ACKs, we should
2119 * test against last_ack_sent instead of rcv_nxt.
2120 * Note 2: we handle special case of closed window, not
2121 * covered by the RFC.
2123 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2124 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2125 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2126 KASSERT(tp->t_state != TCPS_SYN_SENT,
2127 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2130 if (V_tcp_insecure_rst ||
2131 tp->last_ack_sent == th->th_seq) {
2132 TCPSTAT_INC(tcps_drops);
2133 /* Drop the connection. */
2134 switch (tp->t_state) {
2135 case TCPS_SYN_RECEIVED:
2136 so->so_error = ECONNREFUSED;
2138 case TCPS_ESTABLISHED:
2139 case TCPS_FIN_WAIT_1:
2140 case TCPS_FIN_WAIT_2:
2141 case TCPS_CLOSE_WAIT:
2144 so->so_error = ECONNRESET;
2148 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST);
2152 TCPSTAT_INC(tcps_badrst);
2153 /* Send challenge ACK. */
2154 tcp_respond(tp, mtod(m, void *), th, m,
2155 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2156 tp->last_ack_sent = tp->rcv_nxt;
2164 * RFC5961 Section 4.2
2165 * Send challenge ACK for any SYN in synchronized state.
2167 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2168 tp->t_state != TCPS_SYN_RECEIVED) {
2169 TCPSTAT_INC(tcps_badsyn);
2170 if (V_tcp_insecure_syn &&
2171 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2172 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2173 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2174 tp = tcp_drop(tp, ECONNRESET);
2175 rstreason = BANDLIM_UNLIMITED;
2177 tcp_ecn_input_syn_sent(tp, thflags, iptos);
2178 /* Send challenge ACK. */
2179 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2180 tp->snd_nxt, TH_ACK);
2181 tp->last_ack_sent = tp->rcv_nxt;
2188 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2189 * and it's less than ts_recent, drop it.
2191 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2192 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2193 /* Check to see if ts_recent is over 24 days old. */
2194 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2196 * Invalidate ts_recent. If this segment updates
2197 * ts_recent, the age will be reset later and ts_recent
2198 * will get a valid value. If it does not, setting
2199 * ts_recent to zero will at least satisfy the
2200 * requirement that zero be placed in the timestamp
2201 * echo reply when ts_recent isn't valid. The
2202 * age isn't reset until we get a valid ts_recent
2203 * because we don't want out-of-order segments to be
2204 * dropped when ts_recent is old.
2208 TCPSTAT_INC(tcps_rcvduppack);
2209 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2210 TCPSTAT_INC(tcps_pawsdrop);
2218 * In the SYN-RECEIVED state, validate that the packet belongs to
2219 * this connection before trimming the data to fit the receive
2220 * window. Check the sequence number versus IRS since we know
2221 * the sequence numbers haven't wrapped. This is a partial fix
2222 * for the "LAND" DoS attack.
2224 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2225 rstreason = BANDLIM_RST_OPENPORT;
2226 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2230 todrop = tp->rcv_nxt - th->th_seq;
2232 if (thflags & TH_SYN) {
2242 * Following if statement from Stevens, vol. 2, p. 960.
2245 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2247 * Any valid FIN must be to the left of the window.
2248 * At this point the FIN must be a duplicate or out
2249 * of sequence; drop it.
2254 * Send an ACK to resynchronize and drop any data.
2255 * But keep on processing for RST or ACK.
2257 tp->t_flags |= TF_ACKNOW;
2259 TCPSTAT_INC(tcps_rcvduppack);
2260 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2262 TCPSTAT_INC(tcps_rcvpartduppack);
2263 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2266 * DSACK - add SACK block for dropped range
2268 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2269 tcp_update_sack_list(tp, th->th_seq,
2270 th->th_seq + todrop);
2272 * ACK now, as the next in-sequence segment
2273 * will clear the DSACK block again
2275 tp->t_flags |= TF_ACKNOW;
2277 drop_hdrlen += todrop; /* drop from the top afterwards */
2278 th->th_seq += todrop;
2280 if (th->th_urp > todrop)
2281 th->th_urp -= todrop;
2289 * If new data are received on a connection after the
2290 * user processes are gone, then RST the other end.
2292 if ((tp->t_flags & TF_CLOSED) && tlen) {
2293 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2294 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2295 "after socket was closed, "
2296 "sending RST and removing tcpcb\n",
2297 s, __func__, tcpstates[tp->t_state], tlen);
2300 tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE);
2301 /* tcp_close will kill the inp pre-log the Reset */
2302 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST);
2304 TCPSTAT_INC(tcps_rcvafterclose);
2305 rstreason = BANDLIM_UNLIMITED;
2310 * If segment ends after window, drop trailing data
2311 * (and PUSH and FIN); if nothing left, just ACK.
2313 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2315 TCPSTAT_INC(tcps_rcvpackafterwin);
2316 if (todrop >= tlen) {
2317 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2319 * If window is closed can only take segments at
2320 * window edge, and have to drop data and PUSH from
2321 * incoming segments. Continue processing, but
2322 * remember to ack. Otherwise, drop segment
2325 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2326 tp->t_flags |= TF_ACKNOW;
2327 TCPSTAT_INC(tcps_rcvwinprobe);
2331 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2334 thflags &= ~(TH_PUSH|TH_FIN);
2338 * If last ACK falls within this segment's sequence numbers,
2339 * record its timestamp.
2341 * 1) That the test incorporates suggestions from the latest
2342 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2343 * 2) That updating only on newer timestamps interferes with
2344 * our earlier PAWS tests, so this check should be solely
2345 * predicated on the sequence space of this segment.
2346 * 3) That we modify the segment boundary check to be
2347 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2348 * instead of RFC1323's
2349 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2350 * This modified check allows us to overcome RFC1323's
2351 * limitations as described in Stevens TCP/IP Illustrated
2352 * Vol. 2 p.869. In such cases, we can still calculate the
2353 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2355 if ((to.to_flags & TOF_TS) != 0 &&
2356 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2357 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2358 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2359 tp->ts_recent_age = tcp_ts_getticks();
2360 tp->ts_recent = to.to_tsval;
2364 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2365 * flag is on (half-synchronized state), then queue data for
2366 * later processing; else drop segment and return.
2368 if ((thflags & TH_ACK) == 0) {
2369 if (tp->t_state == TCPS_SYN_RECEIVED ||
2370 (tp->t_flags & TF_NEEDSYN)) {
2371 if (tp->t_state == TCPS_SYN_RECEIVED &&
2372 IS_FASTOPEN(tp->t_flags)) {
2373 tp->snd_wnd = tiwin;
2377 } else if (tp->t_flags & TF_ACKNOW)
2386 switch (tp->t_state) {
2388 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2389 * ESTABLISHED state and continue processing.
2390 * The ACK was checked above.
2392 case TCPS_SYN_RECEIVED:
2394 TCPSTAT_INC(tcps_connects);
2395 if (tp->t_flags & TF_SONOTCONN) {
2397 * Usually SYN_RECEIVED had been created from a LISTEN,
2398 * and solisten_enqueue() has already marked the socket
2399 * layer as connected. If it didn't, which can happen
2400 * only with an accept_filter(9), then the tp is marked
2401 * with TF_SONOTCONN. The other reason for this mark
2402 * to be set is a simultaneous open, a SYN_RECEIVED
2403 * that had been created from SYN_SENT.
2405 tp->t_flags &= ~TF_SONOTCONN;
2408 /* Do window scaling? */
2409 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2410 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2411 tp->rcv_scale = tp->request_r_scale;
2413 tp->snd_wnd = tiwin;
2416 * SYN-RECEIVED -> ESTABLISHED
2417 * SYN-RECEIVED* -> FIN-WAIT-1
2419 tp->t_starttime = ticks;
2420 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2421 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2422 tp->t_tfo_pending = NULL;
2424 if (tp->t_flags & TF_NEEDFIN) {
2425 tp->t_acktime = ticks;
2426 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2427 tp->t_flags &= ~TF_NEEDFIN;
2429 tcp_state_change(tp, TCPS_ESTABLISHED);
2430 TCP_PROBE5(accept__established, NULL, tp,
2433 * TFO connections call cc_conn_init() during SYN
2434 * processing. Calling it again here for such
2435 * connections is not harmless as it would undo the
2436 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2439 if (!IS_FASTOPEN(tp->t_flags))
2441 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2444 * Account for the ACK of our SYN prior to
2445 * regular ACK processing below, except for
2446 * simultaneous SYN, which is handled later.
2448 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2451 * If segment contains data or ACK, will call tcp_reass()
2452 * later; if not, do so now to pass queued data to user.
2454 if (tlen == 0 && (thflags & TH_FIN) == 0) {
2455 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2457 tcp_handle_wakeup(tp);
2459 tp->snd_wl1 = th->th_seq - 1;
2463 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2464 * ACKs. If the ack is in the range
2465 * tp->snd_una < th->th_ack <= tp->snd_max
2466 * then advance tp->snd_una to th->th_ack and drop
2467 * data from the retransmission queue. If this ACK reflects
2468 * more up to date window information we update our window information.
2470 case TCPS_ESTABLISHED:
2471 case TCPS_FIN_WAIT_1:
2472 case TCPS_FIN_WAIT_2:
2473 case TCPS_CLOSE_WAIT:
2476 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2477 TCPSTAT_INC(tcps_rcvacktoomuch);
2480 if (tcp_is_sack_recovery(tp, &to)) {
2481 if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) &&
2482 (tp->t_flags & TF_LRD)) {
2483 tcp_sack_lost_retransmission(tp, th);
2487 * Reset the value so that previous (valid) value
2488 * from the last ack with SACK doesn't get used.
2490 tp->sackhint.sacked_bytes = 0;
2493 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2494 hhook_run_tcp_est_in(tp, th, &to);
2497 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2498 maxseg = tcp_maxseg(tp);
2500 (tiwin == tp->snd_wnd ||
2501 (tp->t_flags & TF_SACK_PERMIT))) {
2503 * If this is the first time we've seen a
2504 * FIN from the remote, this is not a
2505 * duplicate and it needs to be processed
2506 * normally. This happens during a
2507 * simultaneous close.
2509 if ((thflags & TH_FIN) &&
2510 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2514 TCPSTAT_INC(tcps_rcvdupack);
2516 * If we have outstanding data (other than
2517 * a window probe), this is a completely
2518 * duplicate ack (ie, window info didn't
2519 * change and FIN isn't set),
2520 * the ack is the biggest we've
2521 * seen and we've seen exactly our rexmt
2522 * threshold of them, assume a packet
2523 * has been dropped and retransmit it.
2524 * Kludge snd_nxt & the congestion
2525 * window so we send only this one
2528 * We know we're losing at the current
2529 * window size so do congestion avoidance
2530 * (set ssthresh to half the current window
2531 * and pull our congestion window back to
2532 * the new ssthresh).
2534 * Dup acks mean that packets have left the
2535 * network (they're now cached at the receiver)
2536 * so bump cwnd by the amount in the receiver
2537 * to keep a constant cwnd packets in the
2540 * When using TCP ECN, notify the peer that
2541 * we reduced the cwnd.
2544 * Following 2 kinds of acks should not affect
2547 * 2) Acks with SACK but without any new SACK
2548 * information in them. These could result from
2549 * any anomaly in the network like a switch
2550 * duplicating packets or a possible DoS attack.
2552 if (th->th_ack != tp->snd_una ||
2553 (tcp_is_sack_recovery(tp, &to) &&
2556 else if (!tcp_timer_active(tp, TT_REXMT))
2558 else if (++tp->t_dupacks > tcprexmtthresh ||
2559 IN_FASTRECOVERY(tp->t_flags)) {
2560 cc_ack_received(tp, th, nsegs,
2563 IN_FASTRECOVERY(tp->t_flags)) {
2564 tcp_do_prr_ack(tp, th, &to);
2565 } else if (tcp_is_sack_recovery(tp, &to) &&
2566 IN_FASTRECOVERY(tp->t_flags)) {
2570 * Compute the amount of data in flight first.
2571 * We can inject new data into the pipe iff
2572 * we have less than 1/2 the original window's
2573 * worth of data in flight.
2575 if (V_tcp_do_newsack)
2576 awnd = tcp_compute_pipe(tp);
2578 awnd = (tp->snd_nxt - tp->snd_fack) +
2579 tp->sackhint.sack_bytes_rexmit;
2581 if (awnd < tp->snd_ssthresh) {
2582 tp->snd_cwnd += maxseg;
2583 if (tp->snd_cwnd > tp->snd_ssthresh)
2584 tp->snd_cwnd = tp->snd_ssthresh;
2587 tp->snd_cwnd += maxseg;
2588 (void) tcp_output(tp);
2590 } else if (tp->t_dupacks == tcprexmtthresh ||
2591 (tp->t_flags & TF_SACK_PERMIT &&
2593 tp->sackhint.sacked_bytes >
2594 (tcprexmtthresh - 1) * maxseg)) {
2597 * Above is the RFC6675 trigger condition of
2598 * more than (dupthresh-1)*maxseg sacked data.
2599 * If the count of holes in the
2600 * scoreboard is >= dupthresh, we could
2601 * also enter loss recovery, but don't
2602 * have that value readily available.
2604 tp->t_dupacks = tcprexmtthresh;
2605 tcp_seq onxt = tp->snd_nxt;
2608 * If we're doing sack, or prr, check
2609 * to see if we're already in sack
2610 * recovery. If we're not doing sack,
2611 * check to see if we're in newreno
2615 (tp->t_flags & TF_SACK_PERMIT)) {
2616 if (IN_FASTRECOVERY(tp->t_flags)) {
2621 if (SEQ_LEQ(th->th_ack,
2627 /* Congestion signal before ack. */
2628 cc_cong_signal(tp, th, CC_NDUPACK);
2629 cc_ack_received(tp, th, nsegs,
2631 tcp_timer_activate(tp, TT_REXMT, 0);
2635 * snd_ssthresh is already updated by
2638 if (tcp_is_sack_recovery(tp, &to)) {
2639 tp->sackhint.prr_delivered =
2640 tp->sackhint.sacked_bytes;
2642 tp->sackhint.prr_delivered =
2643 imin(tp->snd_max - tp->snd_una,
2644 imin(INT_MAX / 65536,
2645 tp->t_dupacks) * maxseg);
2647 tp->sackhint.recover_fs = max(1,
2648 tp->snd_nxt - tp->snd_una);
2650 if (tcp_is_sack_recovery(tp, &to)) {
2652 tcps_sack_recovery_episode);
2653 tp->snd_recover = tp->snd_nxt;
2654 tp->snd_cwnd = maxseg;
2655 (void) tcp_output(tp);
2656 if (SEQ_GT(th->th_ack, tp->snd_una))
2657 goto resume_partialack;
2660 tp->snd_nxt = th->th_ack;
2661 tp->snd_cwnd = maxseg;
2662 (void) tcp_output(tp);
2663 KASSERT(tp->snd_limited <= 2,
2664 ("%s: tp->snd_limited too big",
2666 tp->snd_cwnd = tp->snd_ssthresh +
2668 (tp->t_dupacks - tp->snd_limited);
2669 if (SEQ_GT(onxt, tp->snd_nxt))
2672 } else if (V_tcp_do_rfc3042) {
2674 * Process first and second duplicate
2675 * ACKs. Each indicates a segment
2676 * leaving the network, creating room
2677 * for more. Make sure we can send a
2678 * packet on reception of each duplicate
2679 * ACK by increasing snd_cwnd by one
2680 * segment. Restore the original
2681 * snd_cwnd after packet transmission.
2683 cc_ack_received(tp, th, nsegs,
2685 uint32_t oldcwnd = tp->snd_cwnd;
2686 tcp_seq oldsndmax = tp->snd_max;
2690 KASSERT(tp->t_dupacks == 1 ||
2692 ("%s: dupacks not 1 or 2",
2694 if (tp->t_dupacks == 1)
2695 tp->snd_limited = 0;
2697 (tp->snd_nxt - tp->snd_una) +
2698 (tp->t_dupacks - tp->snd_limited) *
2701 * Only call tcp_output when there
2702 * is new data available to be sent
2703 * or we need to send an ACK.
2705 SOCKBUF_LOCK(&so->so_snd);
2706 avail = sbavail(&so->so_snd) -
2707 (tp->snd_nxt - tp->snd_una);
2708 SOCKBUF_UNLOCK(&so->so_snd);
2709 if (avail > 0 || tp->t_flags & TF_ACKNOW)
2710 (void) tcp_output(tp);
2711 sent = tp->snd_max - oldsndmax;
2712 if (sent > maxseg) {
2713 KASSERT((tp->t_dupacks == 2 &&
2714 tp->snd_limited == 0) ||
2715 (sent == maxseg + 1 &&
2716 tp->t_flags & TF_SENTFIN),
2717 ("%s: sent too much",
2719 tp->snd_limited = 2;
2720 } else if (sent > 0)
2722 tp->snd_cwnd = oldcwnd;
2729 * This ack is advancing the left edge, reset the
2734 * If this ack also has new SACK info, increment the
2735 * counter as per rfc6675. The variable
2736 * sack_changed tracks all changes to the SACK
2737 * scoreboard, including when partial ACKs without
2738 * SACK options are received, and clear the scoreboard
2739 * from the left side. Such partial ACKs should not be
2740 * counted as dupacks here.
2742 if (tcp_is_sack_recovery(tp, &to) &&
2745 /* limit overhead by setting maxseg last */
2746 if (!IN_FASTRECOVERY(tp->t_flags) &&
2747 (tp->sackhint.sacked_bytes >
2748 ((tcprexmtthresh - 1) *
2749 (maxseg = tcp_maxseg(tp))))) {
2750 goto enter_recovery;
2756 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2757 ("%s: th_ack <= snd_una", __func__));
2760 * If the congestion window was inflated to account
2761 * for the other side's cached packets, retract it.
2763 if (IN_FASTRECOVERY(tp->t_flags)) {
2764 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2765 if (tp->t_flags & TF_SACK_PERMIT)
2766 if (V_tcp_do_prr && to.to_flags & TOF_SACK) {
2767 tcp_timer_activate(tp, TT_REXMT, 0);
2769 tcp_do_prr_ack(tp, th, &to);
2770 tp->t_flags |= TF_ACKNOW;
2771 (void) tcp_output(tp);
2773 tcp_sack_partialack(tp, th);
2775 tcp_newreno_partial_ack(tp, th);
2777 cc_post_recovery(tp, th);
2778 } else if (IN_CONGRECOVERY(tp->t_flags)) {
2779 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2781 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th);
2782 tp->snd_fack = th->th_ack;
2783 tcp_do_prr_ack(tp, th, &to);
2784 (void) tcp_output(tp);
2787 cc_post_recovery(tp, th);
2790 * If we reach this point, ACK is not a duplicate,
2791 * i.e., it ACKs something we sent.
2793 if (tp->t_flags & TF_NEEDSYN) {
2795 * T/TCP: Connection was half-synchronized, and our
2796 * SYN has been ACK'd (so connection is now fully
2797 * synchronized). Go to non-starred state,
2798 * increment snd_una for ACK of SYN, and check if
2799 * we can do window scaling.
2801 tp->t_flags &= ~TF_NEEDSYN;
2803 /* Do window scaling? */
2804 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2805 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2806 tp->rcv_scale = tp->request_r_scale;
2807 /* Send window already scaled. */
2812 INP_WLOCK_ASSERT(inp);
2815 * Adjust for the SYN bit in sequence space,
2816 * but don't account for it in cwnd calculations.
2817 * This is for the SYN_RECEIVED, non-simultaneous
2818 * SYN case. SYN_SENT and simultaneous SYN are
2819 * treated elsewhere.
2823 acked = BYTES_THIS_ACK(tp, th);
2824 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2825 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2826 tp->snd_una, th->th_ack, tp, m));
2827 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2828 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2831 * If we just performed our first retransmit, and the ACK
2832 * arrives within our recovery window, then it was a mistake
2833 * to do the retransmit in the first place. Recover our
2834 * original cwnd and ssthresh, and proceed to transmit where
2837 if (tp->t_rxtshift == 1 &&
2838 tp->t_flags & TF_PREVVALID &&
2839 tp->t_badrxtwin != 0 &&
2840 to.to_flags & TOF_TS &&
2842 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
2843 cc_cong_signal(tp, th, CC_RTO_ERR);
2846 * If we have a timestamp reply, update smoothed
2847 * round trip time. If no timestamp is present but
2848 * transmit timer is running and timed sequence
2849 * number was acked, update smoothed round trip time.
2850 * Since we now have an rtt measurement, cancel the
2851 * timer backoff (cf., Phil Karn's retransmit alg.).
2852 * Recompute the initial retransmit timer.
2854 * Some boxes send broken timestamp replies
2855 * during the SYN+ACK phase, ignore
2856 * timestamps of 0 or we could calculate a
2857 * huge RTT and blow up the retransmit timer.
2859 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2862 t = tcp_ts_getticks() - to.to_tsecr;
2863 if (!tp->t_rttlow || tp->t_rttlow > t)
2865 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2866 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2867 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2868 tp->t_rttlow = ticks - tp->t_rtttime;
2869 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2872 SOCKBUF_LOCK(&so->so_snd);
2874 * Clear t_acktime if remote side has ACKd all data in the
2875 * socket buffer and FIN (if applicable).
2876 * Otherwise, update t_acktime if we received a sufficiently
2879 if ((tp->t_state <= TCPS_CLOSE_WAIT &&
2880 acked == sbavail(&so->so_snd)) ||
2881 acked > sbavail(&so->so_snd))
2884 tp->t_acktime = ticks;
2887 * If all outstanding data is acked, stop retransmit
2888 * timer and remember to restart (more output or persist).
2889 * If there is more data to be acked, restart retransmit
2890 * timer, using current (possibly backed-off) value.
2892 if (th->th_ack == tp->snd_max) {
2893 tcp_timer_activate(tp, TT_REXMT, 0);
2895 } else if (!tcp_timer_active(tp, TT_PERSIST))
2896 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp));
2899 * If no data (only SYN) was ACK'd,
2900 * skip rest of ACK processing.
2903 SOCKBUF_UNLOCK(&so->so_snd);
2908 * Let the congestion control algorithm update congestion
2909 * control related information. This typically means increasing
2910 * the congestion window.
2912 cc_ack_received(tp, th, nsegs, CC_ACK);
2914 if (acked > sbavail(&so->so_snd)) {
2915 if (tp->snd_wnd >= sbavail(&so->so_snd))
2916 tp->snd_wnd -= sbavail(&so->so_snd);
2919 mfree = sbcut_locked(&so->so_snd,
2920 (int)sbavail(&so->so_snd));
2923 mfree = sbcut_locked(&so->so_snd, acked);
2924 if (tp->snd_wnd >= (uint32_t) acked)
2925 tp->snd_wnd -= acked;
2930 /* NB: sowwakeup_locked() does an implicit unlock. */
2931 sowwakeup_locked(so);
2933 /* Detect una wraparound. */
2934 if (!IN_RECOVERY(tp->t_flags) &&
2935 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2936 SEQ_LEQ(th->th_ack, tp->snd_recover))
2937 tp->snd_recover = th->th_ack - 1;
2938 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2939 if (IN_RECOVERY(tp->t_flags) &&
2940 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2941 EXIT_RECOVERY(tp->t_flags);
2943 tp->snd_una = th->th_ack;
2944 if (tp->t_flags & TF_SACK_PERMIT) {
2945 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2946 tp->snd_recover = tp->snd_una;
2948 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2949 tp->snd_nxt = tp->snd_una;
2951 switch (tp->t_state) {
2953 * In FIN_WAIT_1 STATE in addition to the processing
2954 * for the ESTABLISHED state if our FIN is now acknowledged
2955 * then enter FIN_WAIT_2.
2957 case TCPS_FIN_WAIT_1:
2958 if (ourfinisacked) {
2960 * If we can't receive any more
2961 * data, then closing user can proceed.
2962 * Starting the timer is contrary to the
2963 * specification, but if we don't get a FIN
2964 * we'll hang forever.
2966 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2967 soisdisconnected(so);
2968 tcp_timer_activate(tp, TT_2MSL,
2969 (tcp_fast_finwait2_recycle ?
2970 tcp_finwait2_timeout :
2973 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2978 * In CLOSING STATE in addition to the processing for
2979 * the ESTABLISHED state if the ACK acknowledges our FIN
2980 * then enter the TIME-WAIT state, otherwise ignore
2984 if (ourfinisacked) {
2992 * In LAST_ACK, we may still be waiting for data to drain
2993 * and/or to be acked, as well as for the ack of our FIN.
2994 * If our FIN is now acknowledged, delete the TCB,
2995 * enter the closed state and return.
2998 if (ourfinisacked) {
3007 INP_WLOCK_ASSERT(inp);
3010 * Update window information.
3011 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3013 if ((thflags & TH_ACK) &&
3014 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3015 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3016 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3017 /* keep track of pure window updates */
3019 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3020 TCPSTAT_INC(tcps_rcvwinupd);
3021 tp->snd_wnd = tiwin;
3022 tp->snd_wl1 = th->th_seq;
3023 tp->snd_wl2 = th->th_ack;
3024 if (tp->snd_wnd > tp->max_sndwnd)
3025 tp->max_sndwnd = tp->snd_wnd;
3030 * Process segments with URG.
3032 if ((thflags & TH_URG) && th->th_urp &&
3033 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3035 * This is a kludge, but if we receive and accept
3036 * random urgent pointers, we'll crash in
3037 * soreceive. It's hard to imagine someone
3038 * actually wanting to send this much urgent data.
3040 SOCKBUF_LOCK(&so->so_rcv);
3041 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3042 th->th_urp = 0; /* XXX */
3043 thflags &= ~TH_URG; /* XXX */
3044 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3045 goto dodata; /* XXX */
3048 * If this segment advances the known urgent pointer,
3049 * then mark the data stream. This should not happen
3050 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3051 * a FIN has been received from the remote side.
3052 * In these states we ignore the URG.
3054 * According to RFC961 (Assigned Protocols),
3055 * the urgent pointer points to the last octet
3056 * of urgent data. We continue, however,
3057 * to consider it to indicate the first octet
3058 * of data past the urgent section as the original
3059 * spec states (in one of two places).
3061 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3062 tp->rcv_up = th->th_seq + th->th_urp;
3063 so->so_oobmark = sbavail(&so->so_rcv) +
3064 (tp->rcv_up - tp->rcv_nxt) - 1;
3065 if (so->so_oobmark == 0)
3066 so->so_rcv.sb_state |= SBS_RCVATMARK;
3068 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3070 SOCKBUF_UNLOCK(&so->so_rcv);
3072 * Remove out of band data so doesn't get presented to user.
3073 * This can happen independent of advancing the URG pointer,
3074 * but if two URG's are pending at once, some out-of-band
3075 * data may creep in... ick.
3077 if (th->th_urp <= (uint32_t)tlen &&
3078 !(so->so_options & SO_OOBINLINE)) {
3079 /* hdr drop is delayed */
3080 tcp_pulloutofband(so, th, m, drop_hdrlen);
3084 * If no out of band data is expected,
3085 * pull receive urgent pointer along
3086 * with the receive window.
3088 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3089 tp->rcv_up = tp->rcv_nxt;
3092 INP_WLOCK_ASSERT(inp);
3095 * Process the segment text, merging it into the TCP sequencing queue,
3096 * and arranging for acknowledgment of receipt if necessary.
3097 * This process logically involves adjusting tp->rcv_wnd as data
3098 * is presented to the user (this happens in tcp_usrreq.c,
3099 * case PRU_RCVD). If a FIN has already been received on this
3100 * connection then we just ignore the text.
3102 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3103 IS_FASTOPEN(tp->t_flags));
3104 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3105 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3106 tcp_seq save_start = th->th_seq;
3107 tcp_seq save_rnxt = tp->rcv_nxt;
3108 int save_tlen = tlen;
3109 m_adj(m, drop_hdrlen); /* delayed header drop */
3111 * Insert segment which includes th into TCP reassembly queue
3112 * with control block tp. Set thflags to whether reassembly now
3113 * includes a segment with FIN. This handles the common case
3114 * inline (segment is the next to be received on an established
3115 * connection, and the queue is empty), avoiding linkage into
3116 * and removal from the queue and repetition of various
3118 * Set DELACK for segments received in order, but ack
3119 * immediately when segments are out of order (so
3120 * fast retransmit can work).
3122 if (th->th_seq == tp->rcv_nxt &&
3124 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3126 if (DELAY_ACK(tp, tlen) || tfo_syn)
3127 tp->t_flags |= TF_DELACK;
3129 tp->t_flags |= TF_ACKNOW;
3130 tp->rcv_nxt += tlen;
3132 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3133 (tp->t_fbyte_in == 0)) {
3134 tp->t_fbyte_in = ticks;
3135 if (tp->t_fbyte_in == 0)
3137 if (tp->t_fbyte_out && tp->t_fbyte_in)
3138 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3140 thflags = tcp_get_flags(th) & TH_FIN;
3141 TCPSTAT_INC(tcps_rcvpack);
3142 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3143 SOCKBUF_LOCK(&so->so_rcv);
3144 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3147 sbappendstream_locked(&so->so_rcv, m, 0);
3148 tp->t_flags |= TF_WAKESOR;
3151 * XXX: Due to the header drop above "th" is
3152 * theoretically invalid by now. Fortunately
3153 * m_adj() doesn't actually frees any mbufs
3154 * when trimming from the head.
3156 tcp_seq temp = save_start;
3158 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3159 tp->t_flags |= TF_ACKNOW;
3161 if ((tp->t_flags & TF_SACK_PERMIT) &&
3163 TCPS_HAVEESTABLISHED(tp->t_state)) {
3164 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3166 * DSACK actually handled in the fastpath
3169 tcp_update_sack_list(tp, save_start,
3170 save_start + save_tlen);
3171 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3172 if ((tp->rcv_numsacks >= 1) &&
3173 (tp->sackblks[0].end == save_start)) {
3175 * Partial overlap, recorded at todrop
3178 tcp_update_sack_list(tp,
3179 tp->sackblks[0].start,
3180 tp->sackblks[0].end);
3182 tcp_update_dsack_list(tp, save_start,
3183 save_start + save_tlen);
3185 } else if (tlen >= save_tlen) {
3186 /* Update of sackblks. */
3187 tcp_update_dsack_list(tp, save_start,
3188 save_start + save_tlen);
3189 } else if (tlen > 0) {
3190 tcp_update_dsack_list(tp, save_start,
3194 tcp_handle_wakeup(tp);
3197 * Note the amount of data that peer has sent into
3198 * our window, in order to estimate the sender's
3202 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3203 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3205 len = so->so_rcv.sb_hiwat;
3213 * If FIN is received ACK the FIN and let the user know
3214 * that the connection is closing.
3216 if (thflags & TH_FIN) {
3217 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3218 /* The socket upcall is handled by socantrcvmore. */
3221 * If connection is half-synchronized
3222 * (ie NEEDSYN flag on) then delay ACK,
3223 * so it may be piggybacked when SYN is sent.
3224 * Otherwise, since we received a FIN then no
3225 * more input can be expected, send ACK now.
3227 if (tp->t_flags & TF_NEEDSYN)
3228 tp->t_flags |= TF_DELACK;
3230 tp->t_flags |= TF_ACKNOW;
3233 switch (tp->t_state) {
3235 * In SYN_RECEIVED and ESTABLISHED STATES
3236 * enter the CLOSE_WAIT state.
3238 case TCPS_SYN_RECEIVED:
3239 tp->t_starttime = ticks;
3241 case TCPS_ESTABLISHED:
3242 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3246 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3247 * enter the CLOSING state.
3249 case TCPS_FIN_WAIT_1:
3250 tcp_state_change(tp, TCPS_CLOSING);
3254 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3255 * starting the time-wait timer, turning off the other
3258 case TCPS_FIN_WAIT_2:
3263 TCP_PROBE3(debug__input, tp, th, m);
3266 * Return any desired output.
3268 if (needoutput || (tp->t_flags & TF_ACKNOW))
3269 (void) tcp_output(tp);
3272 INP_WLOCK_ASSERT(inp);
3274 if (tp->t_flags & TF_DELACK) {
3275 tp->t_flags &= ~TF_DELACK;
3276 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3283 * Generate an ACK dropping incoming segment if it occupies
3284 * sequence space, where the ACK reflects our state.
3286 * We can now skip the test for the RST flag since all
3287 * paths to this code happen after packets containing
3288 * RST have been dropped.
3290 * In the SYN-RECEIVED state, don't send an ACK unless the
3291 * segment we received passes the SYN-RECEIVED ACK test.
3292 * If it fails send a RST. This breaks the loop in the
3293 * "LAND" DoS attack, and also prevents an ACK storm
3294 * between two listening ports that have been sent forged
3295 * SYN segments, each with the source address of the other.
3297 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3298 (SEQ_GT(tp->snd_una, th->th_ack) ||
3299 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3300 rstreason = BANDLIM_RST_OPENPORT;
3301 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
3304 TCP_PROBE3(debug__input, tp, th, m);
3305 tp->t_flags |= TF_ACKNOW;
3306 (void) tcp_output(tp);
3313 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3316 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3321 * Drop space held by incoming segment and return.
3323 TCP_PROBE3(debug__input, tp, th, m);
3331 * Issue RST and make ACK acceptable to originator of segment.
3332 * The mbuf must still include the original packet header.
3336 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3337 int tlen, int rstreason)
3343 struct ip6_hdr *ip6;
3347 INP_LOCK_ASSERT(tptoinpcb(tp));
3350 /* Don't bother if destination was broadcast/multicast. */
3351 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3354 if (mtod(m, struct ip *)->ip_v == 6) {
3355 ip6 = mtod(m, struct ip6_hdr *);
3356 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3357 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3359 /* IPv6 anycast check is done at tcp6_input() */
3362 #if defined(INET) && defined(INET6)
3367 ip = mtod(m, struct ip *);
3368 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3369 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3370 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3371 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3376 /* Perform bandwidth limiting. */
3377 if (badport_bandlim(rstreason) < 0)
3380 /* tcp_respond consumes the mbuf chain. */
3381 if (tcp_get_flags(th) & TH_ACK) {
3382 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3383 th->th_ack, TH_RST);
3385 if (tcp_get_flags(th) & TH_SYN)
3387 if (tcp_get_flags(th) & TH_FIN)
3389 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3390 (tcp_seq)0, TH_RST|TH_ACK);
3398 * Parse TCP options and place in tcpopt.
3401 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3406 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3408 if (opt == TCPOPT_EOL)
3410 if (opt == TCPOPT_NOP)
3416 if (optlen < 2 || optlen > cnt)
3421 if (optlen != TCPOLEN_MAXSEG)
3423 if (!(flags & TO_SYN))
3425 to->to_flags |= TOF_MSS;
3426 bcopy((char *)cp + 2,
3427 (char *)&to->to_mss, sizeof(to->to_mss));
3428 to->to_mss = ntohs(to->to_mss);
3431 if (optlen != TCPOLEN_WINDOW)
3433 if (!(flags & TO_SYN))
3435 to->to_flags |= TOF_SCALE;
3436 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3438 case TCPOPT_TIMESTAMP:
3439 if (optlen != TCPOLEN_TIMESTAMP)
3441 to->to_flags |= TOF_TS;
3442 bcopy((char *)cp + 2,
3443 (char *)&to->to_tsval, sizeof(to->to_tsval));
3444 to->to_tsval = ntohl(to->to_tsval);
3445 bcopy((char *)cp + 6,
3446 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3447 to->to_tsecr = ntohl(to->to_tsecr);
3449 case TCPOPT_SIGNATURE:
3451 * In order to reply to a host which has set the
3452 * TCP_SIGNATURE option in its initial SYN, we have
3453 * to record the fact that the option was observed
3454 * here for the syncache code to perform the correct
3457 if (optlen != TCPOLEN_SIGNATURE)
3459 to->to_flags |= TOF_SIGNATURE;
3460 to->to_signature = cp + 2;
3462 case TCPOPT_SACK_PERMITTED:
3463 if (optlen != TCPOLEN_SACK_PERMITTED)
3465 if (!(flags & TO_SYN))
3469 to->to_flags |= TOF_SACKPERM;
3472 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3476 to->to_flags |= TOF_SACK;
3477 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3478 to->to_sacks = cp + 2;
3479 TCPSTAT_INC(tcps_sack_rcv_blocks);
3481 case TCPOPT_FAST_OPEN:
3483 * Cookie length validation is performed by the
3484 * server side cookie checking code or the client
3485 * side cookie cache update code.
3487 if (!(flags & TO_SYN))
3489 if (!V_tcp_fastopen_client_enable &&
3490 !V_tcp_fastopen_server_enable)
3492 to->to_flags |= TOF_FASTOPEN;
3493 to->to_tfo_len = optlen - 2;
3494 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3503 * Pull out of band byte out of a segment so
3504 * it doesn't appear in the user's data queue.
3505 * It is still reflected in the segment length for
3506 * sequencing purposes.
3509 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3512 int cnt = off + th->th_urp - 1;
3515 if (m->m_len > cnt) {
3516 char *cp = mtod(m, caddr_t) + cnt;
3517 struct tcpcb *tp = sototcpcb(so);
3519 INP_WLOCK_ASSERT(tptoinpcb(tp));
3522 tp->t_oobflags |= TCPOOB_HAVEDATA;
3523 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3525 if (m->m_flags & M_PKTHDR)
3534 panic("tcp_pulloutofband");
3538 * Collect new round-trip time estimate
3539 * and update averages and current timeout.
3542 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3546 INP_WLOCK_ASSERT(tptoinpcb(tp));
3548 TCPSTAT_INC(tcps_rttupdated);
3549 if (tp->t_rttupdated < UCHAR_MAX)
3552 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3553 imax(0, rtt * 1000 / hz));
3555 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3557 * srtt is stored as fixed point with 5 bits after the
3558 * binary point (i.e., scaled by 8). The following magic
3559 * is equivalent to the smoothing algorithm in rfc793 with
3560 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3561 * point). Adjust rtt to origin 0.
3563 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3564 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3566 if ((tp->t_srtt += delta) <= 0)
3570 * We accumulate a smoothed rtt variance (actually, a
3571 * smoothed mean difference), then set the retransmit
3572 * timer to smoothed rtt + 4 times the smoothed variance.
3573 * rttvar is stored as fixed point with 4 bits after the
3574 * binary point (scaled by 16). The following is
3575 * equivalent to rfc793 smoothing with an alpha of .75
3576 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3577 * rfc793's wired-in beta.
3581 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3582 if ((tp->t_rttvar += delta) <= 0)
3586 * No rtt measurement yet - use the unsmoothed rtt.
3587 * Set the variance to half the rtt (so our first
3588 * retransmit happens at 3*rtt).
3590 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3591 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3597 * the retransmit should happen at rtt + 4 * rttvar.
3598 * Because of the way we do the smoothing, srtt and rttvar
3599 * will each average +1/2 tick of bias. When we compute
3600 * the retransmit timer, we want 1/2 tick of rounding and
3601 * 1 extra tick because of +-1/2 tick uncertainty in the
3602 * firing of the timer. The bias will give us exactly the
3603 * 1.5 tick we need. But, because the bias is
3604 * statistical, we have to test that we don't drop below
3605 * the minimum feasible timer (which is 2 ticks).
3607 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3608 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3611 * We received an ack for a packet that wasn't retransmitted;
3612 * it is probably safe to discard any error indications we've
3613 * received recently. This isn't quite right, but close enough
3614 * for now (a route might have failed after we sent a segment,
3615 * and the return path might not be symmetrical).
3617 tp->t_softerror = 0;
3621 * Determine a reasonable value for maxseg size.
3622 * If the route is known, check route for mtu.
3623 * If none, use an mss that can be handled on the outgoing interface
3624 * without forcing IP to fragment. If no route is found, route has no mtu,
3625 * or the destination isn't local, use a default, hopefully conservative
3626 * size (usually 512 or the default IP max size, but no more than the mtu
3627 * of the interface), as we can't discover anything about intervening
3628 * gateways or networks. We also initialize the congestion/slow start
3629 * window to be a single segment if the destination isn't local.
3630 * While looking at the routing entry, we also initialize other path-dependent
3631 * parameters from pre-set or cached values in the routing entry.
3633 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3634 * IP options, e.g. IPSEC data, since length of this data may vary, and
3635 * thus it is calculated for every segment separately in tcp_output().
3637 * NOTE that this routine is only called when we process an incoming
3638 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3639 * settings are handled in tcp_mssopt().
3642 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3643 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3646 uint32_t maxmtu = 0;
3647 struct inpcb *inp = tptoinpcb(tp);
3648 struct hc_metrics_lite metrics;
3650 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3651 size_t min_protoh = isipv6 ?
3652 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3653 sizeof (struct tcpiphdr);
3655 size_t min_protoh = sizeof(struct tcpiphdr);
3658 INP_WLOCK_ASSERT(inp);
3661 min_protoh += V_tcp_udp_tunneling_overhead;
3662 if (mtuoffer != -1) {
3663 KASSERT(offer == -1, ("%s: conflict", __func__));
3664 offer = mtuoffer - min_protoh;
3670 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3671 tp->t_maxseg = V_tcp_v6mssdflt;
3674 #if defined(INET) && defined(INET6)
3679 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3680 tp->t_maxseg = V_tcp_mssdflt;
3685 * No route to sender, stay with default mss and return.
3689 * In case we return early we need to initialize metrics
3690 * to a defined state as tcp_hc_get() would do for us
3691 * if there was no cache hit.
3693 if (metricptr != NULL)
3694 bzero(metricptr, sizeof(struct hc_metrics_lite));
3698 /* What have we got? */
3702 * Offer == 0 means that there was no MSS on the SYN
3703 * segment, in this case we use tcp_mssdflt as
3704 * already assigned to t_maxseg above.
3706 offer = tp->t_maxseg;
3711 * Offer == -1 means that we didn't receive SYN yet.
3717 * Prevent DoS attack with too small MSS. Round up
3718 * to at least minmss.
3720 offer = max(offer, V_tcp_minmss);
3724 * rmx information is now retrieved from tcp_hostcache.
3726 tcp_hc_get(&inp->inp_inc, &metrics);
3727 if (metricptr != NULL)
3728 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3731 * If there's a discovered mtu in tcp hostcache, use it.
3732 * Else, use the link mtu.
3734 if (metrics.rmx_mtu)
3735 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3739 mss = maxmtu - min_protoh;
3740 if (!V_path_mtu_discovery &&
3741 !in6_localaddr(&inp->in6p_faddr))
3742 mss = min(mss, V_tcp_v6mssdflt);
3745 #if defined(INET) && defined(INET6)
3750 mss = maxmtu - min_protoh;
3751 if (!V_path_mtu_discovery &&
3752 !in_localaddr(inp->inp_faddr))
3753 mss = min(mss, V_tcp_mssdflt);
3757 * XXX - The above conditional (mss = maxmtu - min_protoh)
3758 * probably violates the TCP spec.
3759 * The problem is that, since we don't know the
3760 * other end's MSS, we are supposed to use a conservative
3761 * default. But, if we do that, then MTU discovery will
3762 * never actually take place, because the conservative
3763 * default is much less than the MTUs typically seen
3764 * on the Internet today. For the moment, we'll sweep
3765 * this under the carpet.
3767 * The conservative default might not actually be a problem
3768 * if the only case this occurs is when sending an initial
3769 * SYN with options and data to a host we've never talked
3770 * to before. Then, they will reply with an MSS value which
3771 * will get recorded and the new parameters should get
3772 * recomputed. For Further Study.
3775 mss = min(mss, offer);
3778 * Sanity check: make sure that maxseg will be large
3779 * enough to allow some data on segments even if the
3780 * all the option space is used (40bytes). Otherwise
3781 * funny things may happen in tcp_output.
3783 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3791 tcp_mss(struct tcpcb *tp, int offer)
3795 struct inpcb *inp = tptoinpcb(tp);
3797 struct hc_metrics_lite metrics;
3798 struct tcp_ifcap cap;
3800 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3802 bzero(&cap, sizeof(cap));
3803 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3808 * If there's a pipesize, change the socket buffer to that size,
3809 * don't change if sb_hiwat is different than default (then it
3810 * has been changed on purpose with setsockopt).
3811 * Make the socket buffers an integral number of mss units;
3812 * if the mss is larger than the socket buffer, decrease the mss.
3814 so = inp->inp_socket;
3815 SOCKBUF_LOCK(&so->so_snd);
3816 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3817 bufsize = metrics.rmx_sendpipe;
3819 bufsize = so->so_snd.sb_hiwat;
3823 bufsize = roundup(bufsize, mss);
3824 if (bufsize > sb_max)
3826 if (bufsize > so->so_snd.sb_hiwat)
3827 (void)sbreserve_locked(so, SO_SND, bufsize, NULL);
3829 SOCKBUF_UNLOCK(&so->so_snd);
3831 * Sanity check: make sure that maxseg will be large
3832 * enough to allow some data on segments even if the
3833 * all the option space is used (40bytes). Otherwise
3834 * funny things may happen in tcp_output.
3836 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3838 tp->t_maxseg = max(mss, 64);
3840 SOCKBUF_LOCK(&so->so_rcv);
3841 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3842 bufsize = metrics.rmx_recvpipe;
3844 bufsize = so->so_rcv.sb_hiwat;
3845 if (bufsize > mss) {
3846 bufsize = roundup(bufsize, mss);
3847 if (bufsize > sb_max)
3849 if (bufsize > so->so_rcv.sb_hiwat)
3850 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL);
3852 SOCKBUF_UNLOCK(&so->so_rcv);
3854 /* Check the interface for TSO capabilities. */
3855 if (cap.ifcap & CSUM_TSO) {
3856 tp->t_flags |= TF_TSO;
3857 tp->t_tsomax = cap.tsomax;
3858 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3859 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3864 * Determine the MSS option to send on an outgoing SYN.
3867 tcp_mssopt(struct in_conninfo *inc)
3870 uint32_t thcmtu = 0;
3871 uint32_t maxmtu = 0;
3874 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3877 if (inc->inc_flags & INC_ISIPV6) {
3878 mss = V_tcp_v6mssdflt;
3879 maxmtu = tcp_maxmtu6(inc, NULL);
3880 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3883 #if defined(INET) && defined(INET6)
3888 mss = V_tcp_mssdflt;
3889 maxmtu = tcp_maxmtu(inc, NULL);
3890 min_protoh = sizeof(struct tcpiphdr);
3893 #if defined(INET6) || defined(INET)
3894 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3897 if (maxmtu && thcmtu)
3898 mss = min(maxmtu, thcmtu) - min_protoh;
3899 else if (maxmtu || thcmtu)
3900 mss = max(maxmtu, thcmtu) - min_protoh;
3906 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
3908 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3909 int maxseg = tcp_maxseg(tp);
3911 INP_WLOCK_ASSERT(tptoinpcb(tp));
3914 * Compute the amount of data that this ACK is indicating
3915 * (del_data) and an estimate of how many bytes are in the
3918 if (tcp_is_sack_recovery(tp, to) ||
3919 (IN_CONGRECOVERY(tp->t_flags) &&
3920 !IN_FASTRECOVERY(tp->t_flags))) {
3921 del_data = tp->sackhint.delivered_data;
3922 if (V_tcp_do_newsack)
3923 pipe = tcp_compute_pipe(tp);
3925 pipe = (tp->snd_nxt - tp->snd_fack) +
3926 tp->sackhint.sack_bytes_rexmit;
3928 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg +
3929 tp->snd_recover - tp->snd_una))
3931 pipe = imax(0, tp->snd_max - tp->snd_una -
3932 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg);
3934 tp->sackhint.prr_delivered += del_data;
3936 * Proportional Rate Reduction
3938 if (pipe >= tp->snd_ssthresh) {
3939 if (tp->sackhint.recover_fs == 0)
3940 tp->sackhint.recover_fs =
3941 imax(1, tp->snd_nxt - tp->snd_una);
3942 snd_cnt = howmany((long)tp->sackhint.prr_delivered *
3943 tp->snd_ssthresh, tp->sackhint.recover_fs) -
3944 tp->sackhint.prr_out;
3946 if (V_tcp_do_prr_conservative || (del_data == 0))
3947 limit = tp->sackhint.prr_delivered -
3948 tp->sackhint.prr_out;
3950 limit = imax(tp->sackhint.prr_delivered -
3951 tp->sackhint.prr_out, del_data) +
3953 snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
3955 snd_cnt = imax(snd_cnt, 0) / maxseg;
3957 * Send snd_cnt new data into the network in response to this ack.
3958 * If there is going to be a SACK retransmission, adjust snd_cwnd
3961 if (IN_FASTRECOVERY(tp->t_flags)) {
3962 if (tcp_is_sack_recovery(tp, to)) {
3963 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
3964 tp->sackhint.sack_bytes_rexmit +
3967 tp->snd_cwnd = (tp->snd_max - tp->snd_una) +
3970 } else if (IN_CONGRECOVERY(tp->t_flags))
3971 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg);
3972 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd);
3976 * On a partial ack arrives, force the retransmission of the
3977 * next unacknowledged segment. Do not clear tp->t_dupacks.
3978 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3982 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3984 tcp_seq onxt = tp->snd_nxt;
3985 uint32_t ocwnd = tp->snd_cwnd;
3986 u_int maxseg = tcp_maxseg(tp);
3988 INP_WLOCK_ASSERT(tptoinpcb(tp));
3990 tcp_timer_activate(tp, TT_REXMT, 0);
3992 tp->snd_nxt = th->th_ack;
3994 * Set snd_cwnd to one segment beyond acknowledged offset.
3995 * (tp->snd_una has not yet been updated when this function is called.)
3997 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3998 tp->t_flags |= TF_ACKNOW;
3999 (void) tcp_output(tp);
4000 tp->snd_cwnd = ocwnd;
4001 if (SEQ_GT(onxt, tp->snd_nxt))
4004 * Partial window deflation. Relies on fact that tp->snd_una
4007 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4008 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4011 tp->snd_cwnd += maxseg;
4015 tcp_compute_pipe(struct tcpcb *tp)
4017 if (tp->t_fb->tfb_compute_pipe == NULL) {
4018 return (tp->snd_max - tp->snd_una +
4019 tp->sackhint.sack_bytes_rexmit -
4020 tp->sackhint.sacked_bytes);
4022 return((*tp->t_fb->tfb_compute_pipe)(tp));
4027 tcp_compute_initwnd(uint32_t maxseg)
4030 * Calculate the Initial Window, also used as Restart Window
4032 * RFC5681 Section 3.1 specifies the default conservative values.
4033 * RFC3390 specifies slightly more aggressive values.
4034 * RFC6928 increases it to ten segments.
4035 * Support for user specified value for initial flight size.
4037 if (V_tcp_initcwnd_segments)
4038 return min(V_tcp_initcwnd_segments * maxseg,
4039 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4040 else if (V_tcp_do_rfc3390)
4041 return min(4 * maxseg, max(2 * maxseg, 4380));
4043 /* Per RFC5681 Section 3.1 */
4045 return (2 * maxseg);
4046 else if (maxseg > 1095)
4047 return (3 * maxseg);
4049 return (4 * maxseg);