2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
5 * The Regents of the University of California. All rights reserved.
6 * Copyright (c) 2007-2008,2010
7 * Swinburne University of Technology, Melbourne, Australia.
8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
9 * Copyright (c) 2010 The FreeBSD Foundation
10 * Copyright (c) 2010-2011 Juniper Networks, Inc.
11 * All rights reserved.
13 * Portions of this software were developed at the Centre for Advanced Internet
14 * Architectures, Swinburne University of Technology, by Lawrence Stewart,
15 * James Healy and David Hayes, made possible in part by a grant from the Cisco
16 * University Research Program Fund at Community Foundation Silicon Valley.
18 * Portions of this software were developed at the Centre for Advanced
19 * Internet Architectures, Swinburne University of Technology, Melbourne,
20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
22 * Portions of this software were developed by Robert N. M. Watson under
23 * contract to Juniper Networks, Inc.
25 * Redistribution and use in source and binary forms, with or without
26 * modification, are permitted provided that the following conditions
28 * 1. Redistributions of source code must retain the above copyright
29 * notice, this list of conditions and the following disclaimer.
30 * 2. Redistributions in binary form must reproduce the above copyright
31 * notice, this list of conditions and the following disclaimer in the
32 * documentation and/or other materials provided with the distribution.
33 * 3. Neither the name of the University nor the names of its contributors
34 * may be used to endorse or promote products derived from this software
35 * without specific prior written permission.
37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
49 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
52 #include <sys/cdefs.h>
53 __FBSDID("$FreeBSD$");
56 #include "opt_inet6.h"
57 #include "opt_ipsec.h"
58 #include "opt_tcpdebug.h"
60 #include <sys/param.h>
62 #include <sys/kernel.h>
64 #include <sys/hhook.h>
66 #include <sys/malloc.h>
68 #include <sys/proc.h> /* for proc0 declaration */
69 #include <sys/protosw.h>
70 #include <sys/qmath.h>
72 #include <sys/signalvar.h>
73 #include <sys/socket.h>
74 #include <sys/socketvar.h>
75 #include <sys/sysctl.h>
76 #include <sys/syslog.h>
77 #include <sys/systm.h>
78 #include <sys/stats.h>
80 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
85 #include <net/if_var.h>
86 #include <net/route.h>
89 #define TCPSTATES /* for logging */
91 #include <netinet/in.h>
92 #include <netinet/in_kdtrace.h>
93 #include <netinet/in_pcb.h>
94 #include <netinet/in_systm.h>
95 #include <netinet/ip.h>
96 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
97 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
98 #include <netinet/ip_var.h>
99 #include <netinet/ip_options.h>
100 #include <netinet/ip6.h>
101 #include <netinet/icmp6.h>
102 #include <netinet6/in6_pcb.h>
103 #include <netinet6/in6_var.h>
104 #include <netinet6/ip6_var.h>
105 #include <netinet6/nd6.h>
106 #include <netinet/tcp.h>
107 #include <netinet/tcp_fsm.h>
108 #include <netinet/tcp_log_buf.h>
109 #include <netinet/tcp_seq.h>
110 #include <netinet/tcp_timer.h>
111 #include <netinet/tcp_var.h>
112 #include <netinet6/tcp6_var.h>
113 #include <netinet/tcpip.h>
114 #include <netinet/cc/cc.h>
115 #include <netinet/tcp_fastopen.h>
117 #include <netinet/tcp_pcap.h>
119 #include <netinet/tcp_syncache.h>
121 #include <netinet/tcp_debug.h>
122 #endif /* TCPDEBUG */
124 #include <netinet/tcp_offload.h>
126 #include <netinet/udp.h>
128 #include <netipsec/ipsec_support.h>
130 #include <machine/in_cksum.h>
132 #include <security/mac/mac_framework.h>
134 const int tcprexmtthresh = 3;
136 VNET_DEFINE(int, tcp_log_in_vain) = 0;
137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
138 &VNET_NAME(tcp_log_in_vain), 0,
139 "Log all incoming TCP segments to closed ports");
141 VNET_DEFINE(int, blackhole) = 0;
142 #define V_blackhole VNET(blackhole)
143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
144 &VNET_NAME(blackhole), 0,
145 "Do not send RST on segments to closed ports");
147 VNET_DEFINE(bool, blackhole_local) = false;
148 #define V_blackhole_local VNET(blackhole_local)
149 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET |
150 CTLFLAG_RW, &VNET_NAME(blackhole_local), false,
151 "Enforce net.inet.tcp.blackhole for locally originated packets");
153 VNET_DEFINE(int, tcp_delack_enabled) = 1;
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
155 &VNET_NAME(tcp_delack_enabled), 0,
156 "Delay ACK to try and piggyback it onto a data packet");
158 VNET_DEFINE(int, drop_synfin) = 0;
159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
160 &VNET_NAME(drop_synfin), 0,
161 "Drop TCP packets with SYN+FIN set");
163 VNET_DEFINE(int, tcp_do_prr_conservative) = 0;
164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW,
165 &VNET_NAME(tcp_do_prr_conservative), 0,
166 "Do conservative Proportional Rate Reduction");
168 VNET_DEFINE(int, tcp_do_prr) = 1;
169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW,
170 &VNET_NAME(tcp_do_prr), 1,
171 "Enable Proportional Rate Reduction per RFC 6937");
173 VNET_DEFINE(int, tcp_do_lrd) = 0;
174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW,
175 &VNET_NAME(tcp_do_lrd), 1,
176 "Perform Lost Retransmission Detection");
178 VNET_DEFINE(int, tcp_do_newcwv) = 0;
179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
180 &VNET_NAME(tcp_do_newcwv), 0,
181 "Enable New Congestion Window Validation per RFC7661");
183 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
184 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
185 &VNET_NAME(tcp_do_rfc3042), 0,
186 "Enable RFC 3042 (Limited Transmit)");
188 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
190 &VNET_NAME(tcp_do_rfc3390), 0,
191 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
193 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
195 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
196 "Slow-start flight size (initial congestion window) in number of segments");
198 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
200 &VNET_NAME(tcp_do_rfc3465), 0,
201 "Enable RFC 3465 (Appropriate Byte Counting)");
203 VNET_DEFINE(int, tcp_abc_l_var) = 2;
204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
205 &VNET_NAME(tcp_abc_l_var), 2,
206 "Cap the max cwnd increment during slow-start to this number of segments");
208 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn,
209 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
212 VNET_DEFINE(int, tcp_do_ecn) = 2;
213 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
214 &VNET_NAME(tcp_do_ecn), 0,
217 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
218 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
219 &VNET_NAME(tcp_ecn_maxretries), 0,
220 "Max retries before giving up on ECN");
222 VNET_DEFINE(int, tcp_insecure_syn) = 0;
223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
224 &VNET_NAME(tcp_insecure_syn), 0,
225 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
227 VNET_DEFINE(int, tcp_insecure_rst) = 0;
228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
229 &VNET_NAME(tcp_insecure_rst), 0,
230 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
232 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
233 #define V_tcp_recvspace VNET(tcp_recvspace)
234 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
235 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
237 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
238 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
239 &VNET_NAME(tcp_do_autorcvbuf), 0,
240 "Enable automatic receive buffer sizing");
242 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
243 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
244 &VNET_NAME(tcp_autorcvbuf_max), 0,
245 "Max size of automatic receive buffer");
247 VNET_DEFINE(struct inpcbinfo, tcbinfo);
250 * TCP statistics are stored in an array of counter(9)s, which size matches
251 * size of struct tcpstat. TCP running connection count is a regular array.
253 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
254 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
255 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
256 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
257 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
258 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
259 "TCP connection counts by TCP state");
262 * Kernel module interface for updating tcpstat. The first argument is an index
263 * into tcpstat treated as an array.
266 kmod_tcpstat_add(int statnum, int val)
269 counter_u64_add(VNET(tcpstat)[statnum], val);
274 * Wrapper for the TCP established input helper hook.
277 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
279 struct tcp_hhook_data hhook_data;
281 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
286 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
293 * CC wrapper hook functions
296 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
303 INP_WLOCK_ASSERT(tp->t_inpcb);
305 tp->ccv->nsegs = nsegs;
306 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
307 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
308 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
309 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
310 tp->ccv->flags |= CCF_CWND_LIMITED;
312 tp->ccv->flags &= ~CCF_CWND_LIMITED;
314 if (type == CC_ACK) {
316 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
317 ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
318 if (!IN_RECOVERY(tp->t_flags))
319 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
320 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
321 if ((tp->t_flags & TF_GPUTINPROG) &&
322 SEQ_GEQ(th->th_ack, tp->gput_ack)) {
324 * Compute goodput in bits per millisecond.
326 gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) /
327 max(1, tcp_ts_getticks() - tp->gput_ts);
328 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
331 * XXXLAS: This is a temporary hack, and should be
332 * chained off VOI_TCP_GPUT when stats(9) grows an API
333 * to deal with chained VOIs.
335 if (tp->t_stats_gput_prev > 0)
336 stats_voi_update_abs_s32(tp->t_stats,
338 ((gput - tp->t_stats_gput_prev) * 100) /
339 tp->t_stats_gput_prev);
340 tp->t_flags &= ~TF_GPUTINPROG;
341 tp->t_stats_gput_prev = gput;
344 if (tp->snd_cwnd > tp->snd_ssthresh) {
345 tp->t_bytes_acked += tp->ccv->bytes_this_ack;
346 if (tp->t_bytes_acked >= tp->snd_cwnd) {
347 tp->t_bytes_acked -= tp->snd_cwnd;
348 tp->ccv->flags |= CCF_ABC_SENTAWND;
351 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
352 tp->t_bytes_acked = 0;
356 if (CC_ALGO(tp)->ack_received != NULL) {
357 /* XXXLAS: Find a way to live without this */
358 tp->ccv->curack = th->th_ack;
359 CC_ALGO(tp)->ack_received(tp->ccv, type);
362 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
367 cc_conn_init(struct tcpcb *tp)
369 struct hc_metrics_lite metrics;
370 struct inpcb *inp = tp->t_inpcb;
374 INP_WLOCK_ASSERT(tp->t_inpcb);
376 tcp_hc_get(&inp->inp_inc, &metrics);
377 maxseg = tcp_maxseg(tp);
379 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
381 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
382 TCPSTAT_INC(tcps_usedrtt);
383 if (metrics.rmx_rttvar) {
384 tp->t_rttvar = metrics.rmx_rttvar;
385 TCPSTAT_INC(tcps_usedrttvar);
387 /* default variation is +- 1 rtt */
389 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
391 TCPT_RANGESET(tp->t_rxtcur,
392 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
393 tp->t_rttmin, TCPTV_REXMTMAX);
395 if (metrics.rmx_ssthresh) {
397 * There's some sort of gateway or interface
398 * buffer limit on the path. Use this to set
399 * the slow start threshold, but set the
400 * threshold to no less than 2*mss.
402 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
403 TCPSTAT_INC(tcps_usedssthresh);
407 * Set the initial slow-start flight size.
409 * If a SYN or SYN/ACK was lost and retransmitted, we have to
410 * reduce the initial CWND to one segment as congestion is likely
411 * requiring us to be cautious.
413 if (tp->snd_cwnd == 1)
414 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
416 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
418 if (CC_ALGO(tp)->conn_init != NULL)
419 CC_ALGO(tp)->conn_init(tp->ccv);
423 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
425 INP_WLOCK_ASSERT(tp->t_inpcb);
428 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
433 if (!IN_FASTRECOVERY(tp->t_flags)) {
434 tp->snd_recover = tp->snd_max;
435 if (tp->t_flags2 & TF2_ECN_PERMIT)
436 tp->t_flags2 |= TF2_ECN_SND_CWR;
440 if (!IN_CONGRECOVERY(tp->t_flags) ||
442 * Allow ECN reaction on ACK to CWR, if
443 * that data segment was also CE marked.
445 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
446 EXIT_CONGRECOVERY(tp->t_flags);
447 TCPSTAT_INC(tcps_ecn_rcwnd);
448 tp->snd_recover = tp->snd_max + 1;
449 if (tp->t_flags2 & TF2_ECN_PERMIT)
450 tp->t_flags2 |= TF2_ECN_SND_CWR;
455 tp->t_bytes_acked = 0;
456 EXIT_RECOVERY(tp->t_flags);
457 if (tp->t_flags2 & TF2_ECN_PERMIT)
458 tp->t_flags2 |= TF2_ECN_SND_CWR;
461 TCPSTAT_INC(tcps_sndrexmitbad);
462 /* RTO was unnecessary, so reset everything. */
463 tp->snd_cwnd = tp->snd_cwnd_prev;
464 tp->snd_ssthresh = tp->snd_ssthresh_prev;
465 tp->snd_recover = tp->snd_recover_prev;
466 if (tp->t_flags & TF_WASFRECOVERY)
467 ENTER_FASTRECOVERY(tp->t_flags);
468 if (tp->t_flags & TF_WASCRECOVERY)
469 ENTER_CONGRECOVERY(tp->t_flags);
470 tp->snd_nxt = tp->snd_max;
471 tp->t_flags &= ~TF_PREVVALID;
476 if (CC_ALGO(tp)->cong_signal != NULL) {
478 tp->ccv->curack = th->th_ack;
479 CC_ALGO(tp)->cong_signal(tp->ccv, type);
484 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
486 INP_WLOCK_ASSERT(tp->t_inpcb);
488 /* XXXLAS: KASSERT that we're in recovery? */
490 if (CC_ALGO(tp)->post_recovery != NULL) {
491 tp->ccv->curack = th->th_ack;
492 CC_ALGO(tp)->post_recovery(tp->ccv);
494 /* XXXLAS: EXIT_RECOVERY ? */
495 tp->t_bytes_acked = 0;
496 tp->sackhint.delivered_data = 0;
497 tp->sackhint.prr_out = 0;
501 * Indicate whether this ack should be delayed. We can delay the ack if
502 * following conditions are met:
503 * - There is no delayed ack timer in progress.
504 * - Our last ack wasn't a 0-sized window. We never want to delay
505 * the ack that opens up a 0-sized window.
506 * - LRO wasn't used for this segment. We make sure by checking that the
507 * segment size is not larger than the MSS.
509 #define DELAY_ACK(tp, tlen) \
510 ((!tcp_timer_active(tp, TT_DELACK) && \
511 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
512 (tlen <= tp->t_maxseg) && \
513 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
516 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos)
518 INP_WLOCK_ASSERT(tp->t_inpcb);
520 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
521 switch (iptos & IPTOS_ECN_MASK) {
523 tp->ccv->flags |= CCF_IPHDR_CE;
529 case IPTOS_ECN_NOTECT:
530 tp->ccv->flags &= ~CCF_IPHDR_CE;
535 tp->ccv->flags |= CCF_TCPHDR_CWR;
537 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
539 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
541 if (tp->ccv->flags & CCF_ACKNOW) {
542 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
543 tp->t_flags |= TF_ACKNOW;
549 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
551 cc_ecnpkt_handler_flags(tp, th->th_flags, iptos);
555 * TCP input handling is split into multiple parts:
556 * tcp6_input is a thin wrapper around tcp_input for the extended
557 * ip6_protox[] call format in ip6_input
558 * tcp_input handles primary segment validation, inpcb lookup and
559 * SYN processing on listen sockets
560 * tcp_do_segment processes the ACK and text of the segment for
561 * establishing, established and closing connections
565 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
568 struct in6_ifaddr *ia6;
572 if (m->m_len < *offp + sizeof(struct tcphdr)) {
573 m = m_pullup(m, *offp + sizeof(struct tcphdr));
576 TCPSTAT_INC(tcps_rcvshort);
577 return (IPPROTO_DONE);
582 * draft-itojun-ipv6-tcp-to-anycast
583 * better place to put this in?
585 ip6 = mtod(m, struct ip6_hdr *);
586 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
587 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
588 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
589 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
591 return (IPPROTO_DONE);
595 return (tcp_input_with_port(mp, offp, proto, port));
599 tcp6_input(struct mbuf **mp, int *offp, int proto)
602 return(tcp6_input_with_port(mp, offp, proto, 0));
607 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
609 struct mbuf *m = *mp;
610 struct tcphdr *th = NULL;
611 struct ip *ip = NULL;
612 struct inpcb *inp = NULL;
613 struct tcpcb *tp = NULL;
614 struct socket *so = NULL;
625 int rstreason = 0; /* For badport_bandlim accounting purposes */
628 struct m_tag *fwd_tag = NULL;
630 struct ip6_hdr *ip6 = NULL;
633 const void *ip6 = NULL;
635 struct tcpopt to; /* options in this segment */
636 char *s = NULL; /* address and port logging */
639 * The size of tcp_saveipgen must be the size of the max ip header,
642 u_char tcp_saveipgen[IP6_HDR_LEN];
643 struct tcphdr tcp_savetcp;
650 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
657 TCPSTAT_INC(tcps_rcvtotal);
661 ip6 = mtod(m, struct ip6_hdr *);
662 th = (struct tcphdr *)((caddr_t)ip6 + off0);
663 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
666 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
667 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
668 th->th_sum = m->m_pkthdr.csum_data;
670 th->th_sum = in6_cksum_pseudo(ip6, tlen,
671 IPPROTO_TCP, m->m_pkthdr.csum_data);
672 th->th_sum ^= 0xffff;
674 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
676 TCPSTAT_INC(tcps_rcvbadsum);
681 * Be proactive about unspecified IPv6 address in source.
682 * As we use all-zero to indicate unbounded/unconnected pcb,
683 * unspecified IPv6 address can be used to confuse us.
685 * Note that packets with unspecified IPv6 destination is
686 * already dropped in ip6_input.
688 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
692 iptos = IPV6_TRAFFIC_CLASS(ip6);
695 #if defined(INET) && defined(INET6)
701 * Get IP and TCP header together in first mbuf.
702 * Note: IP leaves IP header in first mbuf.
704 if (off0 > sizeof (struct ip)) {
706 off0 = sizeof(struct ip);
708 if (m->m_len < sizeof (struct tcpiphdr)) {
709 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
711 TCPSTAT_INC(tcps_rcvshort);
712 return (IPPROTO_DONE);
715 ip = mtod(m, struct ip *);
716 th = (struct tcphdr *)((caddr_t)ip + off0);
717 tlen = ntohs(ip->ip_len) - off0;
722 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
723 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
724 th->th_sum = m->m_pkthdr.csum_data;
726 th->th_sum = in_pseudo(ip->ip_src.s_addr,
728 htonl(m->m_pkthdr.csum_data + tlen +
730 th->th_sum ^= 0xffff;
732 struct ipovly *ipov = (struct ipovly *)ip;
735 * Checksum extended TCP header and data.
739 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
740 ipov->ih_len = htons(tlen);
741 th->th_sum = in_cksum(m, len);
742 /* Reset length for SDT probes. */
743 ip->ip_len = htons(len);
746 /* Re-initialization for later version check */
748 ip->ip_v = IPVERSION;
749 ip->ip_hl = off0 >> 2;
752 if (th->th_sum && (port == 0)) {
753 TCPSTAT_INC(tcps_rcvbadsum);
760 * Check that TCP offset makes sense,
761 * pull out TCP options and adjust length. XXX
763 off = th->th_off << 2;
764 if (off < sizeof (struct tcphdr) || off > tlen) {
765 TCPSTAT_INC(tcps_rcvbadoff);
768 tlen -= off; /* tlen is used instead of ti->ti_len */
769 if (off > sizeof (struct tcphdr)) {
772 if (m->m_len < off0 + off) {
773 m = m_pullup(m, off0 + off);
775 TCPSTAT_INC(tcps_rcvshort);
776 return (IPPROTO_DONE);
779 ip6 = mtod(m, struct ip6_hdr *);
780 th = (struct tcphdr *)((caddr_t)ip6 + off0);
783 #if defined(INET) && defined(INET6)
788 if (m->m_len < sizeof(struct ip) + off) {
789 if ((m = m_pullup(m, sizeof (struct ip) + off))
791 TCPSTAT_INC(tcps_rcvshort);
792 return (IPPROTO_DONE);
794 ip = mtod(m, struct ip *);
795 th = (struct tcphdr *)((caddr_t)ip + off0);
799 optlen = off - sizeof (struct tcphdr);
800 optp = (u_char *)(th + 1);
802 thflags = th->th_flags;
805 * Convert TCP protocol specific fields to host format.
807 tcp_fields_to_host(th);
810 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
812 drop_hdrlen = off0 + off;
815 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
819 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
821 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
824 #if defined(INET) && !defined(INET6)
825 (m->m_flags & M_IP_NEXTHOP)
828 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
831 * For initial SYN packets we don't need write lock on matching
832 * PCB, be it a listening one or a synchronized one. The packet
833 * shall not modify its state.
835 lookupflag = (thflags & (TH_ACK|TH_SYN)) == TH_SYN ?
836 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB;
839 if (isipv6 && fwd_tag != NULL) {
840 struct sockaddr_in6 *next_hop6;
842 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
844 * Transparently forwarded. Pretend to be the destination.
845 * Already got one like this?
847 inp = in6_pcblookup_mbuf(&V_tcbinfo,
848 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
849 lookupflag, m->m_pkthdr.rcvif, m);
852 * It's new. Try to find the ambushing socket.
853 * Because we've rewritten the destination address,
854 * any hardware-generated hash is ignored.
856 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
857 th->th_sport, &next_hop6->sin6_addr,
858 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
859 th->th_dport, INPLOOKUP_WILDCARD | lookupflag,
863 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
864 th->th_sport, &ip6->ip6_dst, th->th_dport,
865 INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m);
868 #if defined(INET6) && defined(INET)
872 if (fwd_tag != NULL) {
873 struct sockaddr_in *next_hop;
875 next_hop = (struct sockaddr_in *)(fwd_tag+1);
877 * Transparently forwarded. Pretend to be the destination.
878 * already got one like this?
880 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
881 ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m);
884 * It's new. Try to find the ambushing socket.
885 * Because we've rewritten the destination address,
886 * any hardware-generated hash is ignored.
888 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
889 th->th_sport, next_hop->sin_addr,
890 next_hop->sin_port ? ntohs(next_hop->sin_port) :
891 th->th_dport, INPLOOKUP_WILDCARD | lookupflag,
895 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
896 th->th_sport, ip->ip_dst, th->th_dport,
897 INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m);
901 * If the INPCB does not exist then all data in the incoming
902 * segment is discarded and an appropriate RST is sent back.
903 * XXX MRT Send RST using which routing table?
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 * While waiting for inp lock during the lookup, another thread
939 * can have dropped the inpcb, in which case we need to loop back
940 * and try to find a new inpcb to deliver to.
942 if (inp->inp_flags & INP_DROPPED) {
947 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
948 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
949 ((inp->inp_socket == NULL) || !SOLISTENING(inp->inp_socket))) {
950 inp->inp_flowid = m->m_pkthdr.flowid;
951 inp->inp_flowtype = M_HASHTYPE_GET(m);
953 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
955 if (isipv6 && IPSEC_ENABLED(ipv6) &&
956 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
964 if (IPSEC_ENABLED(ipv4) &&
965 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
972 * Check the minimum TTL for socket.
974 if (inp->inp_ip_minttl != 0) {
977 if (inp->inp_ip_minttl > ip6->ip6_hlim)
981 if (inp->inp_ip_minttl > ip->ip_ttl)
986 * A previous connection in TIMEWAIT state is supposed to catch stray
987 * or duplicate segments arriving late. If this segment was a
988 * legitimate new connection attempt, the old INPCB gets removed and
989 * we can try again to find a listening socket.
991 if (inp->inp_flags & INP_TIMEWAIT) {
992 tcp_dooptions(&to, optp, optlen,
993 (thflags & TH_SYN) ? TO_SYN : 0);
995 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
997 if (tcp_twcheck(inp, &to, th, m, tlen))
999 return (IPPROTO_DONE);
1002 * The TCPCB may no longer exist if the connection is winding
1003 * down or it is in the CLOSED state. Either way we drop the
1004 * segment and send an appropriate response.
1006 tp = intotcpcb(inp);
1007 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1008 rstreason = BANDLIM_RST_CLOSEDPORT;
1012 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) {
1013 rstreason = BANDLIM_RST_CLOSEDPORT;
1018 if (tp->t_flags & TF_TOE) {
1019 tcp_offload_input(tp, m);
1020 m = NULL; /* consumed by the TOE driver */
1026 if (mac_inpcb_check_deliver(inp, m))
1029 so = inp->inp_socket;
1030 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1032 if (so->so_options & SO_DEBUG) {
1033 ostate = tp->t_state;
1036 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1039 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1042 #endif /* TCPDEBUG */
1044 * When the socket is accepting connections (the INPCB is in LISTEN
1045 * state) we look into the SYN cache if this is a new connection
1046 * attempt or the completion of a previous one.
1048 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so),
1049 ("%s: so accepting but tp %p not listening", __func__, tp));
1050 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) {
1051 struct in_conninfo inc;
1053 bzero(&inc, sizeof(inc));
1056 inc.inc_flags |= INC_ISIPV6;
1057 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1058 inc.inc_flags |= INC_IPV6MINMTU;
1059 inc.inc6_faddr = ip6->ip6_src;
1060 inc.inc6_laddr = ip6->ip6_dst;
1064 inc.inc_faddr = ip->ip_src;
1065 inc.inc_laddr = ip->ip_dst;
1067 inc.inc_fport = th->th_sport;
1068 inc.inc_lport = th->th_dport;
1069 inc.inc_fibnum = so->so_fibnum;
1072 * Check for an existing connection attempt in syncache if
1073 * the flag is only ACK. A successful lookup creates a new
1074 * socket appended to the listen queue in SYN_RECEIVED state.
1076 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1078 * Parse the TCP options here because
1079 * syncookies need access to the reflected
1082 tcp_dooptions(&to, optp, optlen, 0);
1084 * NB: syncache_expand() doesn't unlock
1085 * inp and tcpinfo locks.
1087 rstreason = syncache_expand(&inc, &to, th, &so, m, port);
1088 if (rstreason < 0) {
1090 * A failing TCP MD5 signature comparison
1091 * must result in the segment being dropped
1092 * and must not produce any response back
1096 } else if (rstreason == 0) {
1098 * No syncache entry or ACK was not
1099 * for our SYN/ACK. Send a RST.
1100 * NB: syncache did its own logging
1101 * of the failure cause.
1103 rstreason = BANDLIM_RST_OPENPORT;
1109 * We completed the 3-way handshake
1110 * but could not allocate a socket
1111 * either due to memory shortage,
1112 * listen queue length limits or
1113 * global socket limits. Send RST
1114 * or wait and have the remote end
1115 * retransmit the ACK for another
1118 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1119 log(LOG_DEBUG, "%s; %s: Listen socket: "
1120 "Socket allocation failed due to "
1121 "limits or memory shortage, %s\n",
1123 V_tcp_sc_rst_sock_fail ?
1124 "sending RST" : "try again");
1125 if (V_tcp_sc_rst_sock_fail) {
1126 rstreason = BANDLIM_UNLIMITED;
1132 * Socket is created in state SYN_RECEIVED.
1133 * Unlock the listen socket, lock the newly
1134 * created socket and update the tp variable.
1135 * If we came here via jump to tfo_socket_result,
1136 * then listening socket is read-locked.
1138 INP_UNLOCK(inp); /* listen socket */
1139 inp = sotoinpcb(so);
1141 * New connection inpcb is already locked by
1142 * syncache_expand().
1144 INP_WLOCK_ASSERT(inp);
1145 tp = intotcpcb(inp);
1146 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1147 ("%s: ", __func__));
1149 * Process the segment and the data it
1150 * contains. tcp_do_segment() consumes
1151 * the mbuf chain and unlocks the inpcb.
1153 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1154 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1156 return (IPPROTO_DONE);
1159 * Segment flag validation for new connection attempts:
1161 * Our (SYN|ACK) response was rejected.
1162 * Check with syncache and remove entry to prevent
1165 * NB: syncache_chkrst does its own logging of failure
1168 if (thflags & TH_RST) {
1169 syncache_chkrst(&inc, th, m, port);
1173 * We can't do anything without SYN.
1175 if ((thflags & TH_SYN) == 0) {
1176 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1177 log(LOG_DEBUG, "%s; %s: Listen socket: "
1178 "SYN is missing, segment ignored\n",
1180 TCPSTAT_INC(tcps_badsyn);
1184 * (SYN|ACK) is bogus on a listen socket.
1186 if (thflags & TH_ACK) {
1187 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1188 log(LOG_DEBUG, "%s; %s: Listen socket: "
1189 "SYN|ACK invalid, segment rejected\n",
1191 syncache_badack(&inc, port); /* XXX: Not needed! */
1192 TCPSTAT_INC(tcps_badsyn);
1193 rstreason = BANDLIM_RST_OPENPORT;
1197 * If the drop_synfin option is enabled, drop all
1198 * segments with both the SYN and FIN bits set.
1199 * This prevents e.g. nmap from identifying the
1201 * XXX: Poor reasoning. nmap has other methods
1202 * and is constantly refining its stack detection
1204 * XXX: This is a violation of the TCP specification
1205 * and was used by RFC1644.
1207 if ((thflags & TH_FIN) && V_drop_synfin) {
1208 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1209 log(LOG_DEBUG, "%s; %s: Listen socket: "
1210 "SYN|FIN segment ignored (based on "
1211 "sysctl setting)\n", s, __func__);
1212 TCPSTAT_INC(tcps_badsyn);
1216 * Segment's flags are (SYN) or (SYN|FIN).
1218 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1219 * as they do not affect the state of the TCP FSM.
1220 * The data pointed to by TH_URG and th_urp is ignored.
1222 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1223 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1224 KASSERT(thflags & (TH_SYN),
1225 ("%s: Listen socket: TH_SYN not set", __func__));
1226 INP_RLOCK_ASSERT(inp);
1229 * If deprecated address is forbidden,
1230 * we do not accept SYN to deprecated interface
1231 * address to prevent any new inbound connection from
1232 * getting established.
1233 * When we do not accept SYN, we send a TCP RST,
1234 * with deprecated source address (instead of dropping
1235 * it). We compromise it as it is much better for peer
1236 * to send a RST, and RST will be the final packet
1239 * If we do not forbid deprecated addresses, we accept
1240 * the SYN packet. RFC2462 does not suggest dropping
1242 * If we decipher RFC2462 5.5.4, it says like this:
1243 * 1. use of deprecated addr with existing
1244 * communication is okay - "SHOULD continue to be
1246 * 2. use of it with new communication:
1247 * (2a) "SHOULD NOT be used if alternate address
1248 * with sufficient scope is available"
1249 * (2b) nothing mentioned otherwise.
1250 * Here we fall into (2b) case as we have no choice in
1251 * our source address selection - we must obey the peer.
1253 * The wording in RFC2462 is confusing, and there are
1254 * multiple description text for deprecated address
1255 * handling - worse, they are not exactly the same.
1256 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1258 if (isipv6 && !V_ip6_use_deprecated) {
1259 struct in6_ifaddr *ia6;
1261 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1263 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1264 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1265 log(LOG_DEBUG, "%s; %s: Listen socket: "
1266 "Connection attempt to deprecated "
1267 "IPv6 address rejected\n",
1269 rstreason = BANDLIM_RST_OPENPORT;
1275 * Basic sanity checks on incoming SYN requests:
1276 * Don't respond if the destination is a link layer
1277 * broadcast according to RFC1122 4.2.3.10, p. 104.
1278 * If it is from this socket it must be forged.
1279 * Don't respond if the source or destination is a
1280 * global or subnet broad- or multicast address.
1281 * Note that it is quite possible to receive unicast
1282 * link-layer packets with a broadcast IP address. Use
1283 * in_broadcast() to find them.
1285 if (m->m_flags & (M_BCAST|M_MCAST)) {
1286 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1287 log(LOG_DEBUG, "%s; %s: Listen socket: "
1288 "Connection attempt from broad- or multicast "
1289 "link layer address ignored\n", s, __func__);
1294 if (th->th_dport == th->th_sport &&
1295 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1296 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1297 log(LOG_DEBUG, "%s; %s: Listen socket: "
1298 "Connection attempt to/from self "
1299 "ignored\n", s, __func__);
1302 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1303 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1304 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1305 log(LOG_DEBUG, "%s; %s: Listen socket: "
1306 "Connection attempt from/to multicast "
1307 "address ignored\n", s, __func__);
1312 #if defined(INET) && defined(INET6)
1317 if (th->th_dport == th->th_sport &&
1318 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1319 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1320 log(LOG_DEBUG, "%s; %s: Listen socket: "
1321 "Connection attempt from/to self "
1322 "ignored\n", s, __func__);
1325 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1326 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1327 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1328 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1329 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1330 log(LOG_DEBUG, "%s; %s: Listen socket: "
1331 "Connection attempt from/to broad- "
1332 "or multicast address ignored\n",
1339 * SYN appears to be valid. Create compressed TCP state
1343 if (so->so_options & SO_DEBUG)
1344 tcp_trace(TA_INPUT, ostate, tp,
1345 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1347 TCP_PROBE3(debug__input, tp, th, m);
1348 tcp_dooptions(&to, optp, optlen, TO_SYN);
1349 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL,
1350 iptos, port)) != NULL)
1351 goto tfo_socket_result;
1354 * Entry added to syncache and mbuf consumed.
1355 * Only the listen socket is unlocked by syncache_add().
1357 return (IPPROTO_DONE);
1358 } else if (tp->t_state == TCPS_LISTEN) {
1360 * When a listen socket is torn down the SO_ACCEPTCONN
1361 * flag is removed first while connections are drained
1362 * from the accept queue in a unlock/lock cycle of the
1363 * ACCEPT_LOCK, opening a race condition allowing a SYN
1364 * attempt go through unhandled.
1368 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1369 if (tp->t_flags & TF_SIGNATURE) {
1370 tcp_dooptions(&to, optp, optlen, thflags);
1371 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1372 TCPSTAT_INC(tcps_sig_err_nosigopt);
1375 if (!TCPMD5_ENABLED() ||
1376 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1380 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1383 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1384 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1385 * the inpcb, and unlocks pcbinfo.
1387 * XXXGL: in case of a pure SYN arriving on existing connection
1388 * TCP stacks won't need to modify the PCB, they would either drop
1389 * the segment silently, or send a challenge ACK. However, we try
1390 * to upgrade the lock, because calling convention for stacks is
1391 * write-lock on PCB. If upgrade fails, drop the SYN.
1393 if (lookupflag == INPLOOKUP_RLOCKPCB && INP_TRY_UPGRADE(inp) == 0)
1396 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1397 return (IPPROTO_DONE);
1400 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1403 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1406 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1407 m = NULL; /* mbuf chain got consumed. */
1412 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1422 return (IPPROTO_DONE);
1426 * Automatic sizing of receive socket buffer. Often the send
1427 * buffer size is not optimally adjusted to the actual network
1428 * conditions at hand (delay bandwidth product). Setting the
1429 * buffer size too small limits throughput on links with high
1430 * bandwidth and high delay (eg. trans-continental/oceanic links).
1432 * On the receive side the socket buffer memory is only rarely
1433 * used to any significant extent. This allows us to be much
1434 * more aggressive in scaling the receive socket buffer. For
1435 * the case that the buffer space is actually used to a large
1436 * extent and we run out of kernel memory we can simply drop
1437 * the new segments; TCP on the sender will just retransmit it
1438 * later. Setting the buffer size too big may only consume too
1439 * much kernel memory if the application doesn't read() from
1440 * the socket or packet loss or reordering makes use of the
1443 * The criteria to step up the receive buffer one notch are:
1444 * 1. Application has not set receive buffer size with
1445 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1446 * 2. the number of bytes received during 1/2 of an sRTT
1447 * is at least 3/8 of the current socket buffer size.
1448 * 3. receive buffer size has not hit maximal automatic size;
1450 * If all of the criteria are met we increaset the socket buffer
1451 * by a 1/2 (bounded by the max). This allows us to keep ahead
1452 * of slow-start but also makes it so our peer never gets limited
1453 * by our rwnd which we then open up causing a burst.
1455 * This algorithm does two steps per RTT at most and only if
1456 * we receive a bulk stream w/o packet losses or reorderings.
1457 * Shrinking the buffer during idle times is not necessary as
1458 * it doesn't consume any memory when idle.
1460 * TODO: Only step up if the application is actually serving
1461 * the buffer to better manage the socket buffer resources.
1464 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1465 struct tcpcb *tp, int tlen)
1469 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1470 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1471 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1472 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1473 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1474 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1475 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1477 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1479 /* Start over with next RTT. */
1483 tp->rfbuf_cnt += tlen; /* add up */
1489 tcp_input(struct mbuf **mp, int *offp, int proto)
1491 return(tcp_input_with_port(mp, offp, proto, 0));
1495 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1498 * Since tp might be gone if the session entered
1499 * the TIME_WAIT state before coming here, we need
1500 * to check if the socket is still connected.
1508 INP_LOCK_ASSERT(tp->t_inpcb);
1509 if (tp->t_flags & TF_WAKESOR) {
1510 tp->t_flags &= ~TF_WAKESOR;
1511 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1512 sorwakeup_locked(so);
1517 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1518 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1520 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1521 int rstreason, todrop, win, incforsyn = 0;
1525 struct in_conninfo *inc;
1533 * The size of tcp_saveipgen must be the size of the max ip header,
1536 u_char tcp_saveipgen[IP6_HDR_LEN];
1537 struct tcphdr tcp_savetcp;
1540 thflags = th->th_flags;
1541 inc = &tp->t_inpcb->inp_inc;
1542 tp->sackhint.last_sack_ack = 0;
1544 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1547 INP_WLOCK_ASSERT(tp->t_inpcb);
1548 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1550 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1554 /* Save segment, if requested. */
1555 tcp_pcap_add(th, m, &(tp->t_inpkts));
1557 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1560 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1561 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1562 log(LOG_DEBUG, "%s; %s: "
1563 "SYN|FIN segment ignored (based on "
1564 "sysctl setting)\n", s, __func__);
1571 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1572 * check SEQ.ACK first.
1574 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1575 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1576 rstreason = BANDLIM_UNLIMITED;
1581 * Segment received on connection.
1582 * Reset idle time and keep-alive timer.
1583 * XXX: This should be done after segment
1584 * validation to ignore broken/spoofed segs.
1586 tp->t_rcvtime = ticks;
1589 * Scale up the window into a 32-bit value.
1590 * For the SYN_SENT state the scale is zero.
1592 tiwin = th->th_win << tp->snd_scale;
1594 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1598 * TCP ECN processing.
1600 if (tp->t_flags2 & TF2_ECN_PERMIT) {
1601 if (thflags & TH_CWR) {
1602 tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1603 tp->t_flags |= TF_ACKNOW;
1605 switch (iptos & IPTOS_ECN_MASK) {
1607 tp->t_flags2 |= TF2_ECN_SND_ECE;
1608 TCPSTAT_INC(tcps_ecn_ce);
1610 case IPTOS_ECN_ECT0:
1611 TCPSTAT_INC(tcps_ecn_ect0);
1613 case IPTOS_ECN_ECT1:
1614 TCPSTAT_INC(tcps_ecn_ect1);
1618 /* Process a packet differently from RFC3168. */
1619 cc_ecnpkt_handler(tp, th, iptos);
1621 /* Congestion experienced. */
1622 if (thflags & TH_ECE) {
1623 cc_cong_signal(tp, th, CC_ECN);
1628 * Parse options on any incoming segment.
1630 tcp_dooptions(&to, (u_char *)(th + 1),
1631 (th->th_off << 2) - sizeof(struct tcphdr),
1632 (thflags & TH_SYN) ? TO_SYN : 0);
1634 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1635 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1636 (to.to_flags & TOF_SIGNATURE) == 0) {
1637 TCPSTAT_INC(tcps_sig_err_sigopt);
1638 /* XXX: should drop? */
1642 * If echoed timestamp is later than the current time,
1643 * fall back to non RFC1323 RTT calculation. Normalize
1644 * timestamp if syncookies were used when this connection
1647 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1648 to.to_tsecr -= tp->ts_offset;
1649 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1651 else if (tp->t_rxtshift == 1 &&
1652 tp->t_flags & TF_PREVVALID &&
1653 tp->t_badrxtwin != 0 &&
1654 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
1655 cc_cong_signal(tp, th, CC_RTO_ERR);
1658 * Process options only when we get SYN/ACK back. The SYN case
1659 * for incoming connections is handled in tcp_syncache.
1660 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1661 * or <SYN,ACK>) segment itself is never scaled.
1662 * XXX this is traditional behavior, may need to be cleaned up.
1664 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1665 /* Handle parallel SYN for ECN */
1666 if (!(thflags & TH_ACK) &&
1667 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1668 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1669 tp->t_flags2 |= TF2_ECN_PERMIT;
1670 tp->t_flags2 |= TF2_ECN_SND_ECE;
1671 TCPSTAT_INC(tcps_ecn_shs);
1673 if ((to.to_flags & TOF_SCALE) &&
1674 (tp->t_flags & TF_REQ_SCALE) &&
1675 !(tp->t_flags & TF_NOOPT)) {
1676 tp->t_flags |= TF_RCVD_SCALE;
1677 tp->snd_scale = to.to_wscale;
1679 tp->t_flags &= ~TF_REQ_SCALE;
1681 * Initial send window. It will be updated with
1682 * the next incoming segment to the scaled value.
1684 tp->snd_wnd = th->th_win;
1685 if ((to.to_flags & TOF_TS) &&
1686 (tp->t_flags & TF_REQ_TSTMP) &&
1687 !(tp->t_flags & TF_NOOPT)) {
1688 tp->t_flags |= TF_RCVD_TSTMP;
1689 tp->ts_recent = to.to_tsval;
1690 tp->ts_recent_age = tcp_ts_getticks();
1692 tp->t_flags &= ~TF_REQ_TSTMP;
1693 if (to.to_flags & TOF_MSS)
1694 tcp_mss(tp, to.to_mss);
1695 if ((tp->t_flags & TF_SACK_PERMIT) &&
1696 (!(to.to_flags & TOF_SACKPERM) ||
1697 (tp->t_flags & TF_NOOPT)))
1698 tp->t_flags &= ~TF_SACK_PERMIT;
1699 if (IS_FASTOPEN(tp->t_flags)) {
1700 if ((to.to_flags & TOF_FASTOPEN) &&
1701 !(tp->t_flags & TF_NOOPT)) {
1704 if (to.to_flags & TOF_MSS)
1707 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1711 tcp_fastopen_update_cache(tp, mss,
1712 to.to_tfo_len, to.to_tfo_cookie);
1714 tcp_fastopen_disable_path(tp);
1719 * If timestamps were negotiated during SYN/ACK and a
1720 * segment without a timestamp is received, silently drop
1721 * the segment, unless it is a RST segment or missing timestamps are
1723 * See section 3.2 of RFC 7323.
1725 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1726 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1727 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1728 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1729 "segment processed normally\n",
1734 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1735 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1736 "segment silently dropped\n", s, __func__);
1743 * If timestamps were not negotiated during SYN/ACK and a
1744 * segment with a timestamp is received, ignore the
1745 * timestamp and process the packet normally.
1746 * See section 3.2 of RFC 7323.
1748 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1749 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1750 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1751 "segment processed normally\n", s, __func__);
1757 * Header prediction: check for the two common cases
1758 * of a uni-directional data xfer. If the packet has
1759 * no control flags, is in-sequence, the window didn't
1760 * change and we're not retransmitting, it's a
1761 * candidate. If the length is zero and the ack moved
1762 * forward, we're the sender side of the xfer. Just
1763 * free the data acked & wake any higher level process
1764 * that was blocked waiting for space. If the length
1765 * is non-zero and the ack didn't move, we're the
1766 * receiver side. If we're getting packets in-order
1767 * (the reassembly queue is empty), add the data to
1768 * the socket buffer and note that we need a delayed ack.
1769 * Make sure that the hidden state-flags are also off.
1770 * Since we check for TCPS_ESTABLISHED first, it can only
1773 if (tp->t_state == TCPS_ESTABLISHED &&
1774 th->th_seq == tp->rcv_nxt &&
1775 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1776 tp->snd_nxt == tp->snd_max &&
1777 tiwin && tiwin == tp->snd_wnd &&
1778 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1780 ((to.to_flags & TOF_TS) == 0 ||
1781 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1783 * If last ACK falls within this segment's sequence numbers,
1784 * record the timestamp.
1785 * NOTE that the test is modified according to the latest
1786 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1788 if ((to.to_flags & TOF_TS) != 0 &&
1789 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1790 tp->ts_recent_age = tcp_ts_getticks();
1791 tp->ts_recent = to.to_tsval;
1795 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1796 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1797 !IN_RECOVERY(tp->t_flags) &&
1798 (to.to_flags & TOF_SACK) == 0 &&
1799 TAILQ_EMPTY(&tp->snd_holes)) {
1801 * This is a pure ack for outstanding data.
1803 TCPSTAT_INC(tcps_predack);
1806 * "bad retransmit" recovery without timestamps.
1808 if ((to.to_flags & TOF_TS) == 0 &&
1809 tp->t_rxtshift == 1 &&
1810 tp->t_flags & TF_PREVVALID &&
1811 tp->t_badrxtwin != 0 &&
1812 TSTMP_LT(ticks, tp->t_badrxtwin)) {
1813 cc_cong_signal(tp, th, CC_RTO_ERR);
1817 * Recalculate the transmit timer / rtt.
1819 * Some boxes send broken timestamp replies
1820 * during the SYN+ACK phase, ignore
1821 * timestamps of 0 or we could calculate a
1822 * huge RTT and blow up the retransmit timer.
1824 if ((to.to_flags & TOF_TS) != 0 &&
1828 t = tcp_ts_getticks() - to.to_tsecr;
1829 if (!tp->t_rttlow || tp->t_rttlow > t)
1832 TCP_TS_TO_TICKS(t) + 1);
1833 } else if (tp->t_rtttime &&
1834 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1835 if (!tp->t_rttlow ||
1836 tp->t_rttlow > ticks - tp->t_rtttime)
1837 tp->t_rttlow = ticks - tp->t_rtttime;
1839 ticks - tp->t_rtttime);
1841 acked = BYTES_THIS_ACK(tp, th);
1844 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1845 hhook_run_tcp_est_in(tp, th, &to);
1848 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1849 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1850 sbdrop(&so->so_snd, acked);
1851 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1852 SEQ_LEQ(th->th_ack, tp->snd_recover))
1853 tp->snd_recover = th->th_ack - 1;
1856 * Let the congestion control algorithm update
1857 * congestion control related information. This
1858 * typically means increasing the congestion
1861 cc_ack_received(tp, th, nsegs, CC_ACK);
1863 tp->snd_una = th->th_ack;
1865 * Pull snd_wl2 up to prevent seq wrap relative
1868 tp->snd_wl2 = th->th_ack;
1873 * If all outstanding data are acked, stop
1874 * retransmit timer, otherwise restart timer
1875 * using current (possibly backed-off) value.
1876 * If process is waiting for space,
1877 * wakeup/selwakeup/signal. If data
1878 * are ready to send, let tcp_output
1879 * decide between more output or persist.
1882 if (so->so_options & SO_DEBUG)
1883 tcp_trace(TA_INPUT, ostate, tp,
1884 (void *)tcp_saveipgen,
1887 TCP_PROBE3(debug__input, tp, th, m);
1888 if (tp->snd_una == tp->snd_max)
1889 tcp_timer_activate(tp, TT_REXMT, 0);
1890 else if (!tcp_timer_active(tp, TT_PERSIST))
1891 tcp_timer_activate(tp, TT_REXMT,
1894 if (sbavail(&so->so_snd))
1895 (void) tcp_output(tp);
1898 } else if (th->th_ack == tp->snd_una &&
1899 tlen <= sbspace(&so->so_rcv)) {
1900 int newsize = 0; /* automatic sockbuf scaling */
1903 * This is a pure, in-sequence data packet with
1904 * nothing on the reassembly queue and we have enough
1905 * buffer space to take it.
1907 /* Clean receiver SACK report if present */
1908 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1909 tcp_clean_sackreport(tp);
1910 TCPSTAT_INC(tcps_preddat);
1911 tp->rcv_nxt += tlen;
1913 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1914 (tp->t_fbyte_in == 0)) {
1915 tp->t_fbyte_in = ticks;
1916 if (tp->t_fbyte_in == 0)
1918 if (tp->t_fbyte_out && tp->t_fbyte_in)
1919 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1922 * Pull snd_wl1 up to prevent seq wrap relative to
1925 tp->snd_wl1 = th->th_seq;
1927 * Pull rcv_up up to prevent seq wrap relative to
1930 tp->rcv_up = tp->rcv_nxt;
1931 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1932 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1934 if (so->so_options & SO_DEBUG)
1935 tcp_trace(TA_INPUT, ostate, tp,
1936 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1938 TCP_PROBE3(debug__input, tp, th, m);
1940 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1942 /* Add data to socket buffer. */
1943 SOCKBUF_LOCK(&so->so_rcv);
1944 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1948 * Set new socket buffer size.
1949 * Give up when limit is reached.
1952 if (!sbreserve_locked(&so->so_rcv,
1954 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1955 m_adj(m, drop_hdrlen); /* delayed header drop */
1956 sbappendstream_locked(&so->so_rcv, m, 0);
1958 /* NB: sorwakeup_locked() does an implicit unlock. */
1959 sorwakeup_locked(so);
1960 if (DELAY_ACK(tp, tlen)) {
1961 tp->t_flags |= TF_DELACK;
1963 tp->t_flags |= TF_ACKNOW;
1971 * Calculate amount of space in receive window,
1972 * and then do TCP input processing.
1973 * Receive window is amount of space in rcv queue,
1974 * but not less than advertised window.
1976 win = sbspace(&so->so_rcv);
1979 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1981 switch (tp->t_state) {
1983 * If the state is SYN_RECEIVED:
1984 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1986 case TCPS_SYN_RECEIVED:
1987 if ((thflags & TH_ACK) &&
1988 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1989 SEQ_GT(th->th_ack, tp->snd_max))) {
1990 rstreason = BANDLIM_RST_OPENPORT;
1993 if (IS_FASTOPEN(tp->t_flags)) {
1995 * When a TFO connection is in SYN_RECEIVED, the
1996 * only valid packets are the initial SYN, a
1997 * retransmit/copy of the initial SYN (possibly with
1998 * a subset of the original data), a valid ACK, a
2001 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
2002 rstreason = BANDLIM_RST_OPENPORT;
2004 } else if (thflags & TH_SYN) {
2005 /* non-initial SYN is ignored */
2006 if ((tcp_timer_active(tp, TT_DELACK) ||
2007 tcp_timer_active(tp, TT_REXMT)))
2009 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2016 * If the state is SYN_SENT:
2017 * if seg contains a RST with valid ACK (SEQ.ACK has already
2018 * been verified), then drop the connection.
2019 * if seg contains a RST without an ACK, drop the seg.
2020 * if seg does not contain SYN, then drop the seg.
2021 * Otherwise this is an acceptable SYN segment
2022 * initialize tp->rcv_nxt and tp->irs
2023 * if seg contains ack then advance tp->snd_una
2024 * if seg contains an ECE and ECN support is enabled, the stream
2026 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2027 * arrange for segment to be acked (eventually)
2028 * continue processing rest of data/controls, beginning with URG
2031 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2032 TCP_PROBE5(connect__refused, NULL, tp,
2034 tp = tcp_drop(tp, ECONNREFUSED);
2036 if (thflags & TH_RST)
2038 if (!(thflags & TH_SYN))
2041 tp->irs = th->th_seq;
2043 if (thflags & TH_ACK) {
2044 int tfo_partial_ack = 0;
2046 TCPSTAT_INC(tcps_connects);
2049 mac_socketpeer_set_from_mbuf(m, so);
2051 /* Do window scaling on this connection? */
2052 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2053 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2054 tp->rcv_scale = tp->request_r_scale;
2056 tp->rcv_adv += min(tp->rcv_wnd,
2057 TCP_MAXWIN << tp->rcv_scale);
2058 tp->snd_una++; /* SYN is acked */
2060 * If not all the data that was sent in the TFO SYN
2061 * has been acked, resend the remainder right away.
2063 if (IS_FASTOPEN(tp->t_flags) &&
2064 (tp->snd_una != tp->snd_max)) {
2065 tp->snd_nxt = th->th_ack;
2066 tfo_partial_ack = 1;
2069 * If there's data, delay ACK; if there's also a FIN
2070 * ACKNOW will be turned on later.
2072 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2073 tcp_timer_activate(tp, TT_DELACK,
2076 tp->t_flags |= TF_ACKNOW;
2078 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2079 (V_tcp_do_ecn == 1)) {
2080 tp->t_flags2 |= TF2_ECN_PERMIT;
2081 TCPSTAT_INC(tcps_ecn_shs);
2085 * Received <SYN,ACK> in SYN_SENT[*] state.
2087 * SYN_SENT --> ESTABLISHED
2088 * SYN_SENT* --> FIN_WAIT_1
2090 tp->t_starttime = ticks;
2091 if (tp->t_flags & TF_NEEDFIN) {
2092 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2093 tp->t_flags &= ~TF_NEEDFIN;
2096 tcp_state_change(tp, TCPS_ESTABLISHED);
2097 TCP_PROBE5(connect__established, NULL, tp,
2100 tcp_timer_activate(tp, TT_KEEP,
2105 * Received initial SYN in SYN-SENT[*] state =>
2106 * simultaneous open.
2107 * If it succeeds, connection is * half-synchronized.
2108 * Otherwise, do 3-way handshake:
2109 * SYN-SENT -> SYN-RECEIVED
2110 * SYN-SENT* -> SYN-RECEIVED*
2112 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2113 tcp_timer_activate(tp, TT_REXMT, 0);
2114 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2117 INP_WLOCK_ASSERT(tp->t_inpcb);
2120 * Advance th->th_seq to correspond to first data byte.
2121 * If data, trim to stay within window,
2122 * dropping FIN if necessary.
2125 if (tlen > tp->rcv_wnd) {
2126 todrop = tlen - tp->rcv_wnd;
2130 TCPSTAT_INC(tcps_rcvpackafterwin);
2131 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2133 tp->snd_wl1 = th->th_seq - 1;
2134 tp->rcv_up = th->th_seq;
2136 * Client side of transaction: already sent SYN and data.
2137 * If the remote host used T/TCP to validate the SYN,
2138 * our data will be ACK'd; if so, enter normal data segment
2139 * processing in the middle of step 5, ack processing.
2140 * Otherwise, goto step 6.
2142 if (thflags & TH_ACK)
2148 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2149 * do normal processing.
2151 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2155 break; /* continue normal processing */
2159 * States other than LISTEN or SYN_SENT.
2160 * First check the RST flag and sequence number since reset segments
2161 * are exempt from the timestamp and connection count tests. This
2162 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2163 * below which allowed reset segments in half the sequence space
2164 * to fall though and be processed (which gives forged reset
2165 * segments with a random sequence number a 50 percent chance of
2166 * killing a connection).
2167 * Then check timestamp, if present.
2168 * Then check the connection count, if present.
2169 * Then check that at least some bytes of segment are within
2170 * receive window. If segment begins before rcv_nxt,
2171 * drop leading data (and SYN); if nothing left, just ack.
2173 if (thflags & TH_RST) {
2175 * RFC5961 Section 3.2
2177 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2178 * - If RST is in window, we send challenge ACK.
2180 * Note: to take into account delayed ACKs, we should
2181 * test against last_ack_sent instead of rcv_nxt.
2182 * Note 2: we handle special case of closed window, not
2183 * covered by the RFC.
2185 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2186 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2187 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2188 KASSERT(tp->t_state != TCPS_SYN_SENT,
2189 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2192 if (V_tcp_insecure_rst ||
2193 tp->last_ack_sent == th->th_seq) {
2194 TCPSTAT_INC(tcps_drops);
2195 /* Drop the connection. */
2196 switch (tp->t_state) {
2197 case TCPS_SYN_RECEIVED:
2198 so->so_error = ECONNREFUSED;
2200 case TCPS_ESTABLISHED:
2201 case TCPS_FIN_WAIT_1:
2202 case TCPS_FIN_WAIT_2:
2203 case TCPS_CLOSE_WAIT:
2206 so->so_error = ECONNRESET;
2213 TCPSTAT_INC(tcps_badrst);
2214 /* Send challenge ACK. */
2215 tcp_respond(tp, mtod(m, void *), th, m,
2216 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2217 tp->last_ack_sent = tp->rcv_nxt;
2225 * RFC5961 Section 4.2
2226 * Send challenge ACK for any SYN in synchronized state.
2228 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2229 tp->t_state != TCPS_SYN_RECEIVED) {
2230 TCPSTAT_INC(tcps_badsyn);
2231 if (V_tcp_insecure_syn &&
2232 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2233 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2234 tp = tcp_drop(tp, ECONNRESET);
2235 rstreason = BANDLIM_UNLIMITED;
2237 /* Send challenge ACK. */
2238 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2239 tp->snd_nxt, TH_ACK);
2240 tp->last_ack_sent = tp->rcv_nxt;
2247 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2248 * and it's less than ts_recent, drop it.
2250 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2251 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2252 /* Check to see if ts_recent is over 24 days old. */
2253 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2255 * Invalidate ts_recent. If this segment updates
2256 * ts_recent, the age will be reset later and ts_recent
2257 * will get a valid value. If it does not, setting
2258 * ts_recent to zero will at least satisfy the
2259 * requirement that zero be placed in the timestamp
2260 * echo reply when ts_recent isn't valid. The
2261 * age isn't reset until we get a valid ts_recent
2262 * because we don't want out-of-order segments to be
2263 * dropped when ts_recent is old.
2267 TCPSTAT_INC(tcps_rcvduppack);
2268 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2269 TCPSTAT_INC(tcps_pawsdrop);
2277 * In the SYN-RECEIVED state, validate that the packet belongs to
2278 * this connection before trimming the data to fit the receive
2279 * window. Check the sequence number versus IRS since we know
2280 * the sequence numbers haven't wrapped. This is a partial fix
2281 * for the "LAND" DoS attack.
2283 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2284 rstreason = BANDLIM_RST_OPENPORT;
2288 todrop = tp->rcv_nxt - th->th_seq;
2290 if (thflags & TH_SYN) {
2300 * Following if statement from Stevens, vol. 2, p. 960.
2303 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2305 * Any valid FIN must be to the left of the window.
2306 * At this point the FIN must be a duplicate or out
2307 * of sequence; drop it.
2312 * Send an ACK to resynchronize and drop any data.
2313 * But keep on processing for RST or ACK.
2315 tp->t_flags |= TF_ACKNOW;
2317 TCPSTAT_INC(tcps_rcvduppack);
2318 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2320 TCPSTAT_INC(tcps_rcvpartduppack);
2321 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2324 * DSACK - add SACK block for dropped range
2326 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2327 tcp_update_sack_list(tp, th->th_seq,
2328 th->th_seq + todrop);
2330 * ACK now, as the next in-sequence segment
2331 * will clear the DSACK block again
2333 tp->t_flags |= TF_ACKNOW;
2335 drop_hdrlen += todrop; /* drop from the top afterwards */
2336 th->th_seq += todrop;
2338 if (th->th_urp > todrop)
2339 th->th_urp -= todrop;
2347 * If new data are received on a connection after the
2348 * user processes are gone, then RST the other end.
2350 if ((so->so_state & SS_NOFDREF) &&
2351 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2352 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2353 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2354 "after socket was closed, "
2355 "sending RST and removing tcpcb\n",
2356 s, __func__, tcpstates[tp->t_state], tlen);
2360 TCPSTAT_INC(tcps_rcvafterclose);
2361 rstreason = BANDLIM_UNLIMITED;
2366 * If segment ends after window, drop trailing data
2367 * (and PUSH and FIN); if nothing left, just ACK.
2369 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2371 TCPSTAT_INC(tcps_rcvpackafterwin);
2372 if (todrop >= tlen) {
2373 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2375 * If window is closed can only take segments at
2376 * window edge, and have to drop data and PUSH from
2377 * incoming segments. Continue processing, but
2378 * remember to ack. Otherwise, drop segment
2381 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2382 tp->t_flags |= TF_ACKNOW;
2383 TCPSTAT_INC(tcps_rcvwinprobe);
2387 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2390 thflags &= ~(TH_PUSH|TH_FIN);
2394 * If last ACK falls within this segment's sequence numbers,
2395 * record its timestamp.
2397 * 1) That the test incorporates suggestions from the latest
2398 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2399 * 2) That updating only on newer timestamps interferes with
2400 * our earlier PAWS tests, so this check should be solely
2401 * predicated on the sequence space of this segment.
2402 * 3) That we modify the segment boundary check to be
2403 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2404 * instead of RFC1323's
2405 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2406 * This modified check allows us to overcome RFC1323's
2407 * limitations as described in Stevens TCP/IP Illustrated
2408 * Vol. 2 p.869. In such cases, we can still calculate the
2409 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2411 if ((to.to_flags & TOF_TS) != 0 &&
2412 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2413 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2414 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2415 tp->ts_recent_age = tcp_ts_getticks();
2416 tp->ts_recent = to.to_tsval;
2420 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2421 * flag is on (half-synchronized state), then queue data for
2422 * later processing; else drop segment and return.
2424 if ((thflags & TH_ACK) == 0) {
2425 if (tp->t_state == TCPS_SYN_RECEIVED ||
2426 (tp->t_flags & TF_NEEDSYN)) {
2427 if (tp->t_state == TCPS_SYN_RECEIVED &&
2428 IS_FASTOPEN(tp->t_flags)) {
2429 tp->snd_wnd = tiwin;
2433 } else if (tp->t_flags & TF_ACKNOW)
2442 switch (tp->t_state) {
2444 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2445 * ESTABLISHED state and continue processing.
2446 * The ACK was checked above.
2448 case TCPS_SYN_RECEIVED:
2450 TCPSTAT_INC(tcps_connects);
2452 /* Do window scaling? */
2453 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2454 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2455 tp->rcv_scale = tp->request_r_scale;
2457 tp->snd_wnd = tiwin;
2460 * SYN-RECEIVED -> ESTABLISHED
2461 * SYN-RECEIVED* -> FIN-WAIT-1
2463 tp->t_starttime = ticks;
2464 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2465 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2466 tp->t_tfo_pending = NULL;
2468 if (tp->t_flags & TF_NEEDFIN) {
2469 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2470 tp->t_flags &= ~TF_NEEDFIN;
2472 tcp_state_change(tp, TCPS_ESTABLISHED);
2473 TCP_PROBE5(accept__established, NULL, tp,
2476 * TFO connections call cc_conn_init() during SYN
2477 * processing. Calling it again here for such
2478 * connections is not harmless as it would undo the
2479 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2482 if (!IS_FASTOPEN(tp->t_flags))
2484 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2487 * Account for the ACK of our SYN prior to
2488 * regular ACK processing below, except for
2489 * simultaneous SYN, which is handled later.
2491 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2494 * If segment contains data or ACK, will call tcp_reass()
2495 * later; if not, do so now to pass queued data to user.
2497 if (tlen == 0 && (thflags & TH_FIN) == 0) {
2498 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2500 tcp_handle_wakeup(tp, so);
2502 tp->snd_wl1 = th->th_seq - 1;
2506 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2507 * ACKs. If the ack is in the range
2508 * tp->snd_una < th->th_ack <= tp->snd_max
2509 * then advance tp->snd_una to th->th_ack and drop
2510 * data from the retransmission queue. If this ACK reflects
2511 * more up to date window information we update our window information.
2513 case TCPS_ESTABLISHED:
2514 case TCPS_FIN_WAIT_1:
2515 case TCPS_FIN_WAIT_2:
2516 case TCPS_CLOSE_WAIT:
2519 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2520 TCPSTAT_INC(tcps_rcvacktoomuch);
2523 if ((tp->t_flags & TF_SACK_PERMIT) &&
2524 ((to.to_flags & TOF_SACK) ||
2525 !TAILQ_EMPTY(&tp->snd_holes))) {
2526 if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) &&
2527 (tp->t_flags & TF_LRD)) {
2528 tcp_sack_lost_retransmission(tp, th);
2532 * Reset the value so that previous (valid) value
2533 * from the last ack with SACK doesn't get used.
2535 tp->sackhint.sacked_bytes = 0;
2538 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2539 hhook_run_tcp_est_in(tp, th, &to);
2542 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2543 maxseg = tcp_maxseg(tp);
2545 (tiwin == tp->snd_wnd ||
2546 (tp->t_flags & TF_SACK_PERMIT))) {
2548 * If this is the first time we've seen a
2549 * FIN from the remote, this is not a
2550 * duplicate and it needs to be processed
2551 * normally. This happens during a
2552 * simultaneous close.
2554 if ((thflags & TH_FIN) &&
2555 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2559 TCPSTAT_INC(tcps_rcvdupack);
2561 * If we have outstanding data (other than
2562 * a window probe), this is a completely
2563 * duplicate ack (ie, window info didn't
2564 * change and FIN isn't set),
2565 * the ack is the biggest we've
2566 * seen and we've seen exactly our rexmt
2567 * threshold of them, assume a packet
2568 * has been dropped and retransmit it.
2569 * Kludge snd_nxt & the congestion
2570 * window so we send only this one
2573 * We know we're losing at the current
2574 * window size so do congestion avoidance
2575 * (set ssthresh to half the current window
2576 * and pull our congestion window back to
2577 * the new ssthresh).
2579 * Dup acks mean that packets have left the
2580 * network (they're now cached at the receiver)
2581 * so bump cwnd by the amount in the receiver
2582 * to keep a constant cwnd packets in the
2585 * When using TCP ECN, notify the peer that
2586 * we reduced the cwnd.
2589 * Following 2 kinds of acks should not affect
2592 * 2) Acks with SACK but without any new SACK
2593 * information in them. These could result from
2594 * any anomaly in the network like a switch
2595 * duplicating packets or a possible DoS attack.
2597 if (th->th_ack != tp->snd_una ||
2598 ((tp->t_flags & TF_SACK_PERMIT) &&
2599 (to.to_flags & TOF_SACK) &&
2602 else if (!tcp_timer_active(tp, TT_REXMT))
2604 else if (++tp->t_dupacks > tcprexmtthresh ||
2605 IN_FASTRECOVERY(tp->t_flags)) {
2606 cc_ack_received(tp, th, nsegs,
2609 IN_FASTRECOVERY(tp->t_flags)) {
2610 tcp_do_prr_ack(tp, th, &to);
2611 } else if ((tp->t_flags & TF_SACK_PERMIT) &&
2612 (to.to_flags & TOF_SACK) &&
2613 IN_FASTRECOVERY(tp->t_flags)) {
2617 * Compute the amount of data in flight first.
2618 * We can inject new data into the pipe iff
2619 * we have less than 1/2 the original window's
2620 * worth of data in flight.
2622 if (V_tcp_do_newsack)
2623 awnd = tcp_compute_pipe(tp);
2625 awnd = (tp->snd_nxt - tp->snd_fack) +
2626 tp->sackhint.sack_bytes_rexmit;
2628 if (awnd < tp->snd_ssthresh) {
2629 tp->snd_cwnd += maxseg;
2630 if (tp->snd_cwnd > tp->snd_ssthresh)
2631 tp->snd_cwnd = tp->snd_ssthresh;
2634 tp->snd_cwnd += maxseg;
2635 (void) tcp_output(tp);
2637 } else if (tp->t_dupacks == tcprexmtthresh ||
2638 (tp->t_flags & TF_SACK_PERMIT &&
2640 tp->sackhint.sacked_bytes >
2641 (tcprexmtthresh - 1) * maxseg)) {
2644 * Above is the RFC6675 trigger condition of
2645 * more than (dupthresh-1)*maxseg sacked data.
2646 * If the count of holes in the
2647 * scoreboard is >= dupthresh, we could
2648 * also enter loss recovery, but don't
2649 * have that value readily available.
2651 tp->t_dupacks = tcprexmtthresh;
2652 tcp_seq onxt = tp->snd_nxt;
2655 * If we're doing sack, or prr, check
2656 * to see if we're already in sack
2657 * recovery. If we're not doing sack,
2658 * check to see if we're in newreno
2662 (tp->t_flags & TF_SACK_PERMIT)) {
2663 if (IN_FASTRECOVERY(tp->t_flags)) {
2668 if (SEQ_LEQ(th->th_ack,
2674 /* Congestion signal before ack. */
2675 cc_cong_signal(tp, th, CC_NDUPACK);
2676 cc_ack_received(tp, th, nsegs,
2678 tcp_timer_activate(tp, TT_REXMT, 0);
2682 * snd_ssthresh is already updated by
2685 if ((tp->t_flags & TF_SACK_PERMIT) &&
2686 (to.to_flags & TOF_SACK)) {
2687 tp->sackhint.prr_delivered =
2688 tp->sackhint.sacked_bytes;
2690 tp->sackhint.prr_delivered =
2691 imin(tp->snd_max - tp->snd_una,
2692 imin(INT_MAX / 65536,
2693 tp->t_dupacks) * maxseg);
2695 tp->sackhint.recover_fs = max(1,
2696 tp->snd_nxt - tp->snd_una);
2698 if ((tp->t_flags & TF_SACK_PERMIT) &&
2699 (to.to_flags & TOF_SACK)) {
2701 tcps_sack_recovery_episode);
2702 tp->snd_recover = tp->snd_nxt;
2703 tp->snd_cwnd = maxseg;
2704 (void) tcp_output(tp);
2705 if (SEQ_GT(th->th_ack, tp->snd_una))
2706 goto resume_partialack;
2709 tp->snd_nxt = th->th_ack;
2710 tp->snd_cwnd = maxseg;
2711 (void) tcp_output(tp);
2712 KASSERT(tp->snd_limited <= 2,
2713 ("%s: tp->snd_limited too big",
2715 tp->snd_cwnd = tp->snd_ssthresh +
2717 (tp->t_dupacks - tp->snd_limited);
2718 if (SEQ_GT(onxt, tp->snd_nxt))
2721 } else if (V_tcp_do_rfc3042) {
2723 * Process first and second duplicate
2724 * ACKs. Each indicates a segment
2725 * leaving the network, creating room
2726 * for more. Make sure we can send a
2727 * packet on reception of each duplicate
2728 * ACK by increasing snd_cwnd by one
2729 * segment. Restore the original
2730 * snd_cwnd after packet transmission.
2732 cc_ack_received(tp, th, nsegs,
2734 uint32_t oldcwnd = tp->snd_cwnd;
2735 tcp_seq oldsndmax = tp->snd_max;
2739 KASSERT(tp->t_dupacks == 1 ||
2741 ("%s: dupacks not 1 or 2",
2743 if (tp->t_dupacks == 1)
2744 tp->snd_limited = 0;
2746 (tp->snd_nxt - tp->snd_una) +
2747 (tp->t_dupacks - tp->snd_limited) *
2750 * Only call tcp_output when there
2751 * is new data available to be sent.
2752 * Otherwise we would send pure ACKs.
2754 SOCKBUF_LOCK(&so->so_snd);
2755 avail = sbavail(&so->so_snd) -
2756 (tp->snd_nxt - tp->snd_una);
2757 SOCKBUF_UNLOCK(&so->so_snd);
2759 (void) tcp_output(tp);
2760 sent = tp->snd_max - oldsndmax;
2761 if (sent > maxseg) {
2762 KASSERT((tp->t_dupacks == 2 &&
2763 tp->snd_limited == 0) ||
2764 (sent == maxseg + 1 &&
2765 tp->t_flags & TF_SENTFIN),
2766 ("%s: sent too much",
2768 tp->snd_limited = 2;
2769 } else if (sent > 0)
2771 tp->snd_cwnd = oldcwnd;
2778 * This ack is advancing the left edge, reset the
2783 * If this ack also has new SACK info, increment the
2784 * counter as per rfc6675. The variable
2785 * sack_changed tracks all changes to the SACK
2786 * scoreboard, including when partial ACKs without
2787 * SACK options are received, and clear the scoreboard
2788 * from the left side. Such partial ACKs should not be
2789 * counted as dupacks here.
2791 if ((tp->t_flags & TF_SACK_PERMIT) &&
2792 (to.to_flags & TOF_SACK) &&
2795 /* limit overhead by setting maxseg last */
2796 if (!IN_FASTRECOVERY(tp->t_flags) &&
2797 (tp->sackhint.sacked_bytes >
2798 ((tcprexmtthresh - 1) *
2799 (maxseg = tcp_maxseg(tp))))) {
2800 goto enter_recovery;
2806 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2807 ("%s: th_ack <= snd_una", __func__));
2810 * If the congestion window was inflated to account
2811 * for the other side's cached packets, retract it.
2813 if (IN_FASTRECOVERY(tp->t_flags)) {
2814 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2815 if (tp->t_flags & TF_SACK_PERMIT)
2816 if (V_tcp_do_prr && to.to_flags & TOF_SACK) {
2817 tcp_timer_activate(tp, TT_REXMT, 0);
2819 tcp_do_prr_ack(tp, th, &to);
2820 tp->t_flags |= TF_ACKNOW;
2821 (void) tcp_output(tp);
2823 tcp_sack_partialack(tp, th);
2825 tcp_newreno_partial_ack(tp, th);
2827 cc_post_recovery(tp, th);
2828 } else if (IN_CONGRECOVERY(tp->t_flags)) {
2829 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2831 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th);
2832 tp->snd_fack = th->th_ack;
2833 tcp_do_prr_ack(tp, th, &to);
2834 (void) tcp_output(tp);
2837 cc_post_recovery(tp, th);
2840 * If we reach this point, ACK is not a duplicate,
2841 * i.e., it ACKs something we sent.
2843 if (tp->t_flags & TF_NEEDSYN) {
2845 * T/TCP: Connection was half-synchronized, and our
2846 * SYN has been ACK'd (so connection is now fully
2847 * synchronized). Go to non-starred state,
2848 * increment snd_una for ACK of SYN, and check if
2849 * we can do window scaling.
2851 tp->t_flags &= ~TF_NEEDSYN;
2853 /* Do window scaling? */
2854 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2855 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2856 tp->rcv_scale = tp->request_r_scale;
2857 /* Send window already scaled. */
2862 INP_WLOCK_ASSERT(tp->t_inpcb);
2865 * Adjust for the SYN bit in sequence space,
2866 * but don't account for it in cwnd calculations.
2867 * This is for the SYN_RECEIVED, non-simultaneous
2868 * SYN case. SYN_SENT and simultaneous SYN are
2869 * treated elsewhere.
2873 acked = BYTES_THIS_ACK(tp, th);
2874 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2875 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2876 tp->snd_una, th->th_ack, tp, m));
2877 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2878 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2881 * If we just performed our first retransmit, and the ACK
2882 * arrives within our recovery window, then it was a mistake
2883 * to do the retransmit in the first place. Recover our
2884 * original cwnd and ssthresh, and proceed to transmit where
2887 if (tp->t_rxtshift == 1 &&
2888 tp->t_flags & TF_PREVVALID &&
2889 tp->t_badrxtwin != 0 &&
2890 to.to_flags & TOF_TS &&
2892 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
2893 cc_cong_signal(tp, th, CC_RTO_ERR);
2896 * If we have a timestamp reply, update smoothed
2897 * round trip time. If no timestamp is present but
2898 * transmit timer is running and timed sequence
2899 * number was acked, update smoothed round trip time.
2900 * Since we now have an rtt measurement, cancel the
2901 * timer backoff (cf., Phil Karn's retransmit alg.).
2902 * Recompute the initial retransmit timer.
2904 * Some boxes send broken timestamp replies
2905 * during the SYN+ACK phase, ignore
2906 * timestamps of 0 or we could calculate a
2907 * huge RTT and blow up the retransmit timer.
2909 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2912 t = tcp_ts_getticks() - to.to_tsecr;
2913 if (!tp->t_rttlow || tp->t_rttlow > t)
2915 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2916 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2917 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2918 tp->t_rttlow = ticks - tp->t_rtttime;
2919 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2923 * If all outstanding data is acked, stop retransmit
2924 * timer and remember to restart (more output or persist).
2925 * If there is more data to be acked, restart retransmit
2926 * timer, using current (possibly backed-off) value.
2928 if (th->th_ack == tp->snd_max) {
2929 tcp_timer_activate(tp, TT_REXMT, 0);
2931 } else if (!tcp_timer_active(tp, TT_PERSIST))
2932 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2935 * If no data (only SYN) was ACK'd,
2936 * skip rest of ACK processing.
2942 * Let the congestion control algorithm update congestion
2943 * control related information. This typically means increasing
2944 * the congestion window.
2946 cc_ack_received(tp, th, nsegs, CC_ACK);
2948 SOCKBUF_LOCK(&so->so_snd);
2949 if (acked > sbavail(&so->so_snd)) {
2950 if (tp->snd_wnd >= sbavail(&so->so_snd))
2951 tp->snd_wnd -= sbavail(&so->so_snd);
2954 mfree = sbcut_locked(&so->so_snd,
2955 (int)sbavail(&so->so_snd));
2958 mfree = sbcut_locked(&so->so_snd, acked);
2959 if (tp->snd_wnd >= (uint32_t) acked)
2960 tp->snd_wnd -= acked;
2965 /* NB: sowwakeup_locked() does an implicit unlock. */
2966 sowwakeup_locked(so);
2968 /* Detect una wraparound. */
2969 if (!IN_RECOVERY(tp->t_flags) &&
2970 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2971 SEQ_LEQ(th->th_ack, tp->snd_recover))
2972 tp->snd_recover = th->th_ack - 1;
2973 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2974 if (IN_RECOVERY(tp->t_flags) &&
2975 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2976 EXIT_RECOVERY(tp->t_flags);
2978 tp->snd_una = th->th_ack;
2979 if (tp->t_flags & TF_SACK_PERMIT) {
2980 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2981 tp->snd_recover = tp->snd_una;
2983 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2984 tp->snd_nxt = tp->snd_una;
2986 switch (tp->t_state) {
2988 * In FIN_WAIT_1 STATE in addition to the processing
2989 * for the ESTABLISHED state if our FIN is now acknowledged
2990 * then enter FIN_WAIT_2.
2992 case TCPS_FIN_WAIT_1:
2993 if (ourfinisacked) {
2995 * If we can't receive any more
2996 * data, then closing user can proceed.
2997 * Starting the timer is contrary to the
2998 * specification, but if we don't get a FIN
2999 * we'll hang forever.
3002 * we should release the tp also, and use a
3005 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3006 soisdisconnected(so);
3007 tcp_timer_activate(tp, TT_2MSL,
3008 (tcp_fast_finwait2_recycle ?
3009 tcp_finwait2_timeout :
3012 tcp_state_change(tp, TCPS_FIN_WAIT_2);
3017 * In CLOSING STATE in addition to the processing for
3018 * the ESTABLISHED state if the ACK acknowledges our FIN
3019 * then enter the TIME-WAIT state, otherwise ignore
3023 if (ourfinisacked) {
3031 * In LAST_ACK, we may still be waiting for data to drain
3032 * and/or to be acked, as well as for the ack of our FIN.
3033 * If our FIN is now acknowledged, delete the TCB,
3034 * enter the closed state and return.
3037 if (ourfinisacked) {
3046 INP_WLOCK_ASSERT(tp->t_inpcb);
3049 * Update window information.
3050 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3052 if ((thflags & TH_ACK) &&
3053 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3054 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3055 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3056 /* keep track of pure window updates */
3058 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3059 TCPSTAT_INC(tcps_rcvwinupd);
3060 tp->snd_wnd = tiwin;
3061 tp->snd_wl1 = th->th_seq;
3062 tp->snd_wl2 = th->th_ack;
3063 if (tp->snd_wnd > tp->max_sndwnd)
3064 tp->max_sndwnd = tp->snd_wnd;
3069 * Process segments with URG.
3071 if ((thflags & TH_URG) && th->th_urp &&
3072 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3074 * This is a kludge, but if we receive and accept
3075 * random urgent pointers, we'll crash in
3076 * soreceive. It's hard to imagine someone
3077 * actually wanting to send this much urgent data.
3079 SOCKBUF_LOCK(&so->so_rcv);
3080 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3081 th->th_urp = 0; /* XXX */
3082 thflags &= ~TH_URG; /* XXX */
3083 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3084 goto dodata; /* XXX */
3087 * If this segment advances the known urgent pointer,
3088 * then mark the data stream. This should not happen
3089 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3090 * a FIN has been received from the remote side.
3091 * In these states we ignore the URG.
3093 * According to RFC961 (Assigned Protocols),
3094 * the urgent pointer points to the last octet
3095 * of urgent data. We continue, however,
3096 * to consider it to indicate the first octet
3097 * of data past the urgent section as the original
3098 * spec states (in one of two places).
3100 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3101 tp->rcv_up = th->th_seq + th->th_urp;
3102 so->so_oobmark = sbavail(&so->so_rcv) +
3103 (tp->rcv_up - tp->rcv_nxt) - 1;
3104 if (so->so_oobmark == 0)
3105 so->so_rcv.sb_state |= SBS_RCVATMARK;
3107 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3109 SOCKBUF_UNLOCK(&so->so_rcv);
3111 * Remove out of band data so doesn't get presented to user.
3112 * This can happen independent of advancing the URG pointer,
3113 * but if two URG's are pending at once, some out-of-band
3114 * data may creep in... ick.
3116 if (th->th_urp <= (uint32_t)tlen &&
3117 !(so->so_options & SO_OOBINLINE)) {
3118 /* hdr drop is delayed */
3119 tcp_pulloutofband(so, th, m, drop_hdrlen);
3123 * If no out of band data is expected,
3124 * pull receive urgent pointer along
3125 * with the receive window.
3127 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3128 tp->rcv_up = tp->rcv_nxt;
3131 INP_WLOCK_ASSERT(tp->t_inpcb);
3134 * Process the segment text, merging it into the TCP sequencing queue,
3135 * and arranging for acknowledgment of receipt if necessary.
3136 * This process logically involves adjusting tp->rcv_wnd as data
3137 * is presented to the user (this happens in tcp_usrreq.c,
3138 * case PRU_RCVD). If a FIN has already been received on this
3139 * connection then we just ignore the text.
3141 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3142 IS_FASTOPEN(tp->t_flags));
3143 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3144 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3145 tcp_seq save_start = th->th_seq;
3146 tcp_seq save_rnxt = tp->rcv_nxt;
3147 int save_tlen = tlen;
3148 m_adj(m, drop_hdrlen); /* delayed header drop */
3150 * Insert segment which includes th into TCP reassembly queue
3151 * with control block tp. Set thflags to whether reassembly now
3152 * includes a segment with FIN. This handles the common case
3153 * inline (segment is the next to be received on an established
3154 * connection, and the queue is empty), avoiding linkage into
3155 * and removal from the queue and repetition of various
3157 * Set DELACK for segments received in order, but ack
3158 * immediately when segments are out of order (so
3159 * fast retransmit can work).
3161 if (th->th_seq == tp->rcv_nxt &&
3163 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3165 if (DELAY_ACK(tp, tlen) || tfo_syn)
3166 tp->t_flags |= TF_DELACK;
3168 tp->t_flags |= TF_ACKNOW;
3169 tp->rcv_nxt += tlen;
3171 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3172 (tp->t_fbyte_in == 0)) {
3173 tp->t_fbyte_in = ticks;
3174 if (tp->t_fbyte_in == 0)
3176 if (tp->t_fbyte_out && tp->t_fbyte_in)
3177 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3179 thflags = th->th_flags & TH_FIN;
3180 TCPSTAT_INC(tcps_rcvpack);
3181 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3182 SOCKBUF_LOCK(&so->so_rcv);
3183 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3186 sbappendstream_locked(&so->so_rcv, m, 0);
3187 tp->t_flags |= TF_WAKESOR;
3190 * XXX: Due to the header drop above "th" is
3191 * theoretically invalid by now. Fortunately
3192 * m_adj() doesn't actually frees any mbufs
3193 * when trimming from the head.
3195 tcp_seq temp = save_start;
3197 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3198 tp->t_flags |= TF_ACKNOW;
3200 if ((tp->t_flags & TF_SACK_PERMIT) &&
3202 TCPS_HAVEESTABLISHED(tp->t_state)) {
3203 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3205 * DSACK actually handled in the fastpath
3208 tcp_update_sack_list(tp, save_start,
3209 save_start + save_tlen);
3210 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3211 if ((tp->rcv_numsacks >= 1) &&
3212 (tp->sackblks[0].end == save_start)) {
3214 * Partial overlap, recorded at todrop
3217 tcp_update_sack_list(tp,
3218 tp->sackblks[0].start,
3219 tp->sackblks[0].end);
3221 tcp_update_dsack_list(tp, save_start,
3222 save_start + save_tlen);
3224 } else if (tlen >= save_tlen) {
3225 /* Update of sackblks. */
3226 tcp_update_dsack_list(tp, save_start,
3227 save_start + save_tlen);
3228 } else if (tlen > 0) {
3229 tcp_update_dsack_list(tp, save_start,
3233 tcp_handle_wakeup(tp, so);
3236 * Note the amount of data that peer has sent into
3237 * our window, in order to estimate the sender's
3241 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3242 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3244 len = so->so_rcv.sb_hiwat;
3252 * If FIN is received ACK the FIN and let the user know
3253 * that the connection is closing.
3255 if (thflags & TH_FIN) {
3256 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3257 /* The socket upcall is handled by socantrcvmore. */
3260 * If connection is half-synchronized
3261 * (ie NEEDSYN flag on) then delay ACK,
3262 * so it may be piggybacked when SYN is sent.
3263 * Otherwise, since we received a FIN then no
3264 * more input can be expected, send ACK now.
3266 if (tp->t_flags & TF_NEEDSYN)
3267 tp->t_flags |= TF_DELACK;
3269 tp->t_flags |= TF_ACKNOW;
3272 switch (tp->t_state) {
3274 * In SYN_RECEIVED and ESTABLISHED STATES
3275 * enter the CLOSE_WAIT state.
3277 case TCPS_SYN_RECEIVED:
3278 tp->t_starttime = ticks;
3280 case TCPS_ESTABLISHED:
3281 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3285 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3286 * enter the CLOSING state.
3288 case TCPS_FIN_WAIT_1:
3289 tcp_state_change(tp, TCPS_CLOSING);
3293 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3294 * starting the time-wait timer, turning off the other
3297 case TCPS_FIN_WAIT_2:
3303 if (so->so_options & SO_DEBUG)
3304 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3307 TCP_PROBE3(debug__input, tp, th, m);
3310 * Return any desired output.
3312 if (needoutput || (tp->t_flags & TF_ACKNOW))
3313 (void) tcp_output(tp);
3316 INP_WLOCK_ASSERT(tp->t_inpcb);
3318 if (tp->t_flags & TF_DELACK) {
3319 tp->t_flags &= ~TF_DELACK;
3320 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3322 INP_WUNLOCK(tp->t_inpcb);
3327 * Generate an ACK dropping incoming segment if it occupies
3328 * sequence space, where the ACK reflects our state.
3330 * We can now skip the test for the RST flag since all
3331 * paths to this code happen after packets containing
3332 * RST have been dropped.
3334 * In the SYN-RECEIVED state, don't send an ACK unless the
3335 * segment we received passes the SYN-RECEIVED ACK test.
3336 * If it fails send a RST. This breaks the loop in the
3337 * "LAND" DoS attack, and also prevents an ACK storm
3338 * between two listening ports that have been sent forged
3339 * SYN segments, each with the source address of the other.
3341 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3342 (SEQ_GT(tp->snd_una, th->th_ack) ||
3343 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3344 rstreason = BANDLIM_RST_OPENPORT;
3348 if (so->so_options & SO_DEBUG)
3349 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3352 TCP_PROBE3(debug__input, tp, th, m);
3353 tp->t_flags |= TF_ACKNOW;
3354 (void) tcp_output(tp);
3355 INP_WUNLOCK(tp->t_inpcb);
3361 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3362 INP_WUNLOCK(tp->t_inpcb);
3364 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3369 * Drop space held by incoming segment and return.
3372 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3373 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3376 TCP_PROBE3(debug__input, tp, th, m);
3378 INP_WUNLOCK(tp->t_inpcb);
3384 * Issue RST and make ACK acceptable to originator of segment.
3385 * The mbuf must still include the original packet header.
3389 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3390 int tlen, int rstreason)
3396 struct ip6_hdr *ip6;
3400 INP_LOCK_ASSERT(tp->t_inpcb);
3403 /* Don't bother if destination was broadcast/multicast. */
3404 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3407 if (mtod(m, struct ip *)->ip_v == 6) {
3408 ip6 = mtod(m, struct ip6_hdr *);
3409 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3410 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3412 /* IPv6 anycast check is done at tcp6_input() */
3415 #if defined(INET) && defined(INET6)
3420 ip = mtod(m, struct ip *);
3421 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3422 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3423 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3424 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3429 /* Perform bandwidth limiting. */
3430 if (badport_bandlim(rstreason) < 0)
3433 /* tcp_respond consumes the mbuf chain. */
3434 if (th->th_flags & TH_ACK) {
3435 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3436 th->th_ack, TH_RST);
3438 if (th->th_flags & TH_SYN)
3440 if (th->th_flags & TH_FIN)
3442 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3443 (tcp_seq)0, TH_RST|TH_ACK);
3451 * Parse TCP options and place in tcpopt.
3454 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3459 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3461 if (opt == TCPOPT_EOL)
3463 if (opt == TCPOPT_NOP)
3469 if (optlen < 2 || optlen > cnt)
3474 if (optlen != TCPOLEN_MAXSEG)
3476 if (!(flags & TO_SYN))
3478 to->to_flags |= TOF_MSS;
3479 bcopy((char *)cp + 2,
3480 (char *)&to->to_mss, sizeof(to->to_mss));
3481 to->to_mss = ntohs(to->to_mss);
3484 if (optlen != TCPOLEN_WINDOW)
3486 if (!(flags & TO_SYN))
3488 to->to_flags |= TOF_SCALE;
3489 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3491 case TCPOPT_TIMESTAMP:
3492 if (optlen != TCPOLEN_TIMESTAMP)
3494 to->to_flags |= TOF_TS;
3495 bcopy((char *)cp + 2,
3496 (char *)&to->to_tsval, sizeof(to->to_tsval));
3497 to->to_tsval = ntohl(to->to_tsval);
3498 bcopy((char *)cp + 6,
3499 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3500 to->to_tsecr = ntohl(to->to_tsecr);
3502 case TCPOPT_SIGNATURE:
3504 * In order to reply to a host which has set the
3505 * TCP_SIGNATURE option in its initial SYN, we have
3506 * to record the fact that the option was observed
3507 * here for the syncache code to perform the correct
3510 if (optlen != TCPOLEN_SIGNATURE)
3512 to->to_flags |= TOF_SIGNATURE;
3513 to->to_signature = cp + 2;
3515 case TCPOPT_SACK_PERMITTED:
3516 if (optlen != TCPOLEN_SACK_PERMITTED)
3518 if (!(flags & TO_SYN))
3522 to->to_flags |= TOF_SACKPERM;
3525 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3529 to->to_flags |= TOF_SACK;
3530 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3531 to->to_sacks = cp + 2;
3532 TCPSTAT_INC(tcps_sack_rcv_blocks);
3534 case TCPOPT_FAST_OPEN:
3536 * Cookie length validation is performed by the
3537 * server side cookie checking code or the client
3538 * side cookie cache update code.
3540 if (!(flags & TO_SYN))
3542 if (!V_tcp_fastopen_client_enable &&
3543 !V_tcp_fastopen_server_enable)
3545 to->to_flags |= TOF_FASTOPEN;
3546 to->to_tfo_len = optlen - 2;
3547 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3556 * Pull out of band byte out of a segment so
3557 * it doesn't appear in the user's data queue.
3558 * It is still reflected in the segment length for
3559 * sequencing purposes.
3562 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3565 int cnt = off + th->th_urp - 1;
3568 if (m->m_len > cnt) {
3569 char *cp = mtod(m, caddr_t) + cnt;
3570 struct tcpcb *tp = sototcpcb(so);
3572 INP_WLOCK_ASSERT(tp->t_inpcb);
3575 tp->t_oobflags |= TCPOOB_HAVEDATA;
3576 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3578 if (m->m_flags & M_PKTHDR)
3587 panic("tcp_pulloutofband");
3591 * Collect new round-trip time estimate
3592 * and update averages and current timeout.
3595 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3599 INP_WLOCK_ASSERT(tp->t_inpcb);
3601 TCPSTAT_INC(tcps_rttupdated);
3604 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3605 imax(0, rtt * 1000 / hz));
3607 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3609 * srtt is stored as fixed point with 5 bits after the
3610 * binary point (i.e., scaled by 8). The following magic
3611 * is equivalent to the smoothing algorithm in rfc793 with
3612 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3613 * point). Adjust rtt to origin 0.
3615 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3616 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3618 if ((tp->t_srtt += delta) <= 0)
3622 * We accumulate a smoothed rtt variance (actually, a
3623 * smoothed mean difference), then set the retransmit
3624 * timer to smoothed rtt + 4 times the smoothed variance.
3625 * rttvar is stored as fixed point with 4 bits after the
3626 * binary point (scaled by 16). The following is
3627 * equivalent to rfc793 smoothing with an alpha of .75
3628 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3629 * rfc793's wired-in beta.
3633 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3634 if ((tp->t_rttvar += delta) <= 0)
3636 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3637 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3640 * No rtt measurement yet - use the unsmoothed rtt.
3641 * Set the variance to half the rtt (so our first
3642 * retransmit happens at 3*rtt).
3644 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3645 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3646 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3652 * the retransmit should happen at rtt + 4 * rttvar.
3653 * Because of the way we do the smoothing, srtt and rttvar
3654 * will each average +1/2 tick of bias. When we compute
3655 * the retransmit timer, we want 1/2 tick of rounding and
3656 * 1 extra tick because of +-1/2 tick uncertainty in the
3657 * firing of the timer. The bias will give us exactly the
3658 * 1.5 tick we need. But, because the bias is
3659 * statistical, we have to test that we don't drop below
3660 * the minimum feasible timer (which is 2 ticks).
3662 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3663 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3666 * We received an ack for a packet that wasn't retransmitted;
3667 * it is probably safe to discard any error indications we've
3668 * received recently. This isn't quite right, but close enough
3669 * for now (a route might have failed after we sent a segment,
3670 * and the return path might not be symmetrical).
3672 tp->t_softerror = 0;
3676 * Determine a reasonable value for maxseg size.
3677 * If the route is known, check route for mtu.
3678 * If none, use an mss that can be handled on the outgoing interface
3679 * without forcing IP to fragment. If no route is found, route has no mtu,
3680 * or the destination isn't local, use a default, hopefully conservative
3681 * size (usually 512 or the default IP max size, but no more than the mtu
3682 * of the interface), as we can't discover anything about intervening
3683 * gateways or networks. We also initialize the congestion/slow start
3684 * window to be a single segment if the destination isn't local.
3685 * While looking at the routing entry, we also initialize other path-dependent
3686 * parameters from pre-set or cached values in the routing entry.
3688 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3689 * IP options, e.g. IPSEC data, since length of this data may vary, and
3690 * thus it is calculated for every segment separately in tcp_output().
3692 * NOTE that this routine is only called when we process an incoming
3693 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3694 * settings are handled in tcp_mssopt().
3697 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3698 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3701 uint32_t maxmtu = 0;
3702 struct inpcb *inp = tp->t_inpcb;
3703 struct hc_metrics_lite metrics;
3705 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3706 size_t min_protoh = isipv6 ?
3707 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3708 sizeof (struct tcpiphdr);
3710 size_t min_protoh = sizeof(struct tcpiphdr);
3713 INP_WLOCK_ASSERT(tp->t_inpcb);
3716 min_protoh += V_tcp_udp_tunneling_overhead;
3717 if (mtuoffer != -1) {
3718 KASSERT(offer == -1, ("%s: conflict", __func__));
3719 offer = mtuoffer - min_protoh;
3725 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3726 tp->t_maxseg = V_tcp_v6mssdflt;
3729 #if defined(INET) && defined(INET6)
3734 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3735 tp->t_maxseg = V_tcp_mssdflt;
3740 * No route to sender, stay with default mss and return.
3744 * In case we return early we need to initialize metrics
3745 * to a defined state as tcp_hc_get() would do for us
3746 * if there was no cache hit.
3748 if (metricptr != NULL)
3749 bzero(metricptr, sizeof(struct hc_metrics_lite));
3753 /* What have we got? */
3757 * Offer == 0 means that there was no MSS on the SYN
3758 * segment, in this case we use tcp_mssdflt as
3759 * already assigned to t_maxseg above.
3761 offer = tp->t_maxseg;
3766 * Offer == -1 means that we didn't receive SYN yet.
3772 * Prevent DoS attack with too small MSS. Round up
3773 * to at least minmss.
3775 offer = max(offer, V_tcp_minmss);
3779 * rmx information is now retrieved from tcp_hostcache.
3781 tcp_hc_get(&inp->inp_inc, &metrics);
3782 if (metricptr != NULL)
3783 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3786 * If there's a discovered mtu in tcp hostcache, use it.
3787 * Else, use the link mtu.
3789 if (metrics.rmx_mtu)
3790 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3794 mss = maxmtu - min_protoh;
3795 if (!V_path_mtu_discovery &&
3796 !in6_localaddr(&inp->in6p_faddr))
3797 mss = min(mss, V_tcp_v6mssdflt);
3800 #if defined(INET) && defined(INET6)
3805 mss = maxmtu - min_protoh;
3806 if (!V_path_mtu_discovery &&
3807 !in_localaddr(inp->inp_faddr))
3808 mss = min(mss, V_tcp_mssdflt);
3812 * XXX - The above conditional (mss = maxmtu - min_protoh)
3813 * probably violates the TCP spec.
3814 * The problem is that, since we don't know the
3815 * other end's MSS, we are supposed to use a conservative
3816 * default. But, if we do that, then MTU discovery will
3817 * never actually take place, because the conservative
3818 * default is much less than the MTUs typically seen
3819 * on the Internet today. For the moment, we'll sweep
3820 * this under the carpet.
3822 * The conservative default might not actually be a problem
3823 * if the only case this occurs is when sending an initial
3824 * SYN with options and data to a host we've never talked
3825 * to before. Then, they will reply with an MSS value which
3826 * will get recorded and the new parameters should get
3827 * recomputed. For Further Study.
3830 mss = min(mss, offer);
3833 * Sanity check: make sure that maxseg will be large
3834 * enough to allow some data on segments even if the
3835 * all the option space is used (40bytes). Otherwise
3836 * funny things may happen in tcp_output.
3838 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3846 tcp_mss(struct tcpcb *tp, int offer)
3852 struct hc_metrics_lite metrics;
3853 struct tcp_ifcap cap;
3855 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3857 bzero(&cap, sizeof(cap));
3858 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3864 * If there's a pipesize, change the socket buffer to that size,
3865 * don't change if sb_hiwat is different than default (then it
3866 * has been changed on purpose with setsockopt).
3867 * Make the socket buffers an integral number of mss units;
3868 * if the mss is larger than the socket buffer, decrease the mss.
3870 so = inp->inp_socket;
3871 SOCKBUF_LOCK(&so->so_snd);
3872 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3873 bufsize = metrics.rmx_sendpipe;
3875 bufsize = so->so_snd.sb_hiwat;
3879 bufsize = roundup(bufsize, mss);
3880 if (bufsize > sb_max)
3882 if (bufsize > so->so_snd.sb_hiwat)
3883 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3885 SOCKBUF_UNLOCK(&so->so_snd);
3887 * Sanity check: make sure that maxseg will be large
3888 * enough to allow some data on segments even if the
3889 * all the option space is used (40bytes). Otherwise
3890 * funny things may happen in tcp_output.
3892 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3894 tp->t_maxseg = max(mss, 64);
3896 SOCKBUF_LOCK(&so->so_rcv);
3897 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3898 bufsize = metrics.rmx_recvpipe;
3900 bufsize = so->so_rcv.sb_hiwat;
3901 if (bufsize > mss) {
3902 bufsize = roundup(bufsize, mss);
3903 if (bufsize > sb_max)
3905 if (bufsize > so->so_rcv.sb_hiwat)
3906 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3908 SOCKBUF_UNLOCK(&so->so_rcv);
3910 /* Check the interface for TSO capabilities. */
3911 if (cap.ifcap & CSUM_TSO) {
3912 tp->t_flags |= TF_TSO;
3913 tp->t_tsomax = cap.tsomax;
3914 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3915 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3920 * Determine the MSS option to send on an outgoing SYN.
3923 tcp_mssopt(struct in_conninfo *inc)
3926 uint32_t thcmtu = 0;
3927 uint32_t maxmtu = 0;
3930 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3933 if (inc->inc_flags & INC_ISIPV6) {
3934 mss = V_tcp_v6mssdflt;
3935 maxmtu = tcp_maxmtu6(inc, NULL);
3936 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3939 #if defined(INET) && defined(INET6)
3944 mss = V_tcp_mssdflt;
3945 maxmtu = tcp_maxmtu(inc, NULL);
3946 min_protoh = sizeof(struct tcpiphdr);
3949 #if defined(INET6) || defined(INET)
3950 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3953 if (maxmtu && thcmtu)
3954 mss = min(maxmtu, thcmtu) - min_protoh;
3955 else if (maxmtu || thcmtu)
3956 mss = max(maxmtu, thcmtu) - min_protoh;
3962 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
3964 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3965 int maxseg = tcp_maxseg(tp);
3967 INP_WLOCK_ASSERT(tp->t_inpcb);
3970 * Compute the amount of data that this ACK is indicating
3971 * (del_data) and an estimate of how many bytes are in the
3974 if (((tp->t_flags & TF_SACK_PERMIT) &&
3975 (to->to_flags & TOF_SACK)) ||
3976 (IN_CONGRECOVERY(tp->t_flags) &&
3977 !IN_FASTRECOVERY(tp->t_flags))) {
3978 del_data = tp->sackhint.delivered_data;
3979 if (V_tcp_do_newsack)
3980 pipe = tcp_compute_pipe(tp);
3982 pipe = (tp->snd_nxt - tp->snd_fack) +
3983 tp->sackhint.sack_bytes_rexmit;
3985 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg +
3986 tp->snd_recover - tp->snd_una))
3988 pipe = imax(0, tp->snd_max - tp->snd_una -
3989 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg);
3991 tp->sackhint.prr_delivered += del_data;
3993 * Proportional Rate Reduction
3995 if (pipe >= tp->snd_ssthresh) {
3996 if (tp->sackhint.recover_fs == 0)
3997 tp->sackhint.recover_fs =
3998 imax(1, tp->snd_nxt - tp->snd_una);
3999 snd_cnt = howmany((long)tp->sackhint.prr_delivered *
4000 tp->snd_ssthresh, tp->sackhint.recover_fs) -
4001 tp->sackhint.prr_out;
4003 if (V_tcp_do_prr_conservative || (del_data == 0))
4004 limit = tp->sackhint.prr_delivered -
4005 tp->sackhint.prr_out;
4007 limit = imax(tp->sackhint.prr_delivered -
4008 tp->sackhint.prr_out, del_data) +
4010 snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
4012 snd_cnt = imax(snd_cnt, 0) / maxseg;
4014 * Send snd_cnt new data into the network in response to this ack.
4015 * If there is going to be a SACK retransmission, adjust snd_cwnd
4018 if (IN_FASTRECOVERY(tp->t_flags)) {
4019 if ((tp->t_flags & TF_SACK_PERMIT) &&
4020 (to->to_flags & TOF_SACK)) {
4021 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
4022 tp->sackhint.sack_bytes_rexmit +
4025 tp->snd_cwnd = (tp->snd_max - tp->snd_una) +
4028 } else if (IN_CONGRECOVERY(tp->t_flags))
4029 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg);
4030 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd);
4034 * On a partial ack arrives, force the retransmission of the
4035 * next unacknowledged segment. Do not clear tp->t_dupacks.
4036 * By setting snd_nxt to ti_ack, this forces retransmission timer to
4040 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4042 tcp_seq onxt = tp->snd_nxt;
4043 uint32_t ocwnd = tp->snd_cwnd;
4044 u_int maxseg = tcp_maxseg(tp);
4046 INP_WLOCK_ASSERT(tp->t_inpcb);
4048 tcp_timer_activate(tp, TT_REXMT, 0);
4050 tp->snd_nxt = th->th_ack;
4052 * Set snd_cwnd to one segment beyond acknowledged offset.
4053 * (tp->snd_una has not yet been updated when this function is called.)
4055 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4056 tp->t_flags |= TF_ACKNOW;
4057 (void) tcp_output(tp);
4058 tp->snd_cwnd = ocwnd;
4059 if (SEQ_GT(onxt, tp->snd_nxt))
4062 * Partial window deflation. Relies on fact that tp->snd_una
4065 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4066 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4069 tp->snd_cwnd += maxseg;
4073 tcp_compute_pipe(struct tcpcb *tp)
4075 return (tp->snd_max - tp->snd_una +
4076 tp->sackhint.sack_bytes_rexmit -
4077 tp->sackhint.sacked_bytes);
4081 tcp_compute_initwnd(uint32_t maxseg)
4084 * Calculate the Initial Window, also used as Restart Window
4086 * RFC5681 Section 3.1 specifies the default conservative values.
4087 * RFC3390 specifies slightly more aggressive values.
4088 * RFC6928 increases it to ten segments.
4089 * Support for user specified value for initial flight size.
4091 if (V_tcp_initcwnd_segments)
4092 return min(V_tcp_initcwnd_segments * maxseg,
4093 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4094 else if (V_tcp_do_rfc3390)
4095 return min(4 * maxseg, max(2 * maxseg, 4380));
4097 /* Per RFC5681 Section 3.1 */
4099 return (2 * maxseg);
4100 else if (maxseg > 1095)
4101 return (3 * maxseg);
4103 return (4 * maxseg);