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>
54 #include "opt_inet6.h"
55 #include "opt_ipsec.h"
58 #include <sys/param.h>
60 #include <sys/kernel.h>
62 #include <sys/hhook.h>
64 #include <sys/malloc.h>
66 #include <sys/proc.h> /* for proc0 declaration */
67 #include <sys/protosw.h>
68 #include <sys/qmath.h>
70 #include <sys/signalvar.h>
71 #include <sys/socket.h>
72 #include <sys/socketvar.h>
73 #include <sys/sysctl.h>
74 #include <sys/syslog.h>
75 #include <sys/systm.h>
76 #include <sys/stats.h>
78 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
83 #include <net/if_var.h>
84 #include <net/route.h>
85 #include <net/rss_config.h>
88 #define TCPSTATES /* for logging */
90 #include <netinet/in.h>
91 #include <netinet/in_kdtrace.h>
92 #include <netinet/in_pcb.h>
93 #include <netinet/in_rss.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_rss.h>
104 #include <netinet6/in6_var.h>
105 #include <netinet6/ip6_var.h>
106 #include <netinet6/nd6.h>
107 #include <netinet/tcp.h>
108 #include <netinet/tcp_fsm.h>
109 #include <netinet/tcp_seq.h>
110 #include <netinet/tcp_timer.h>
111 #include <netinet/tcp_var.h>
112 #include <netinet/tcp_log_buf.h>
113 #include <netinet6/tcp6_var.h>
114 #include <netinet/tcpip.h>
115 #include <netinet/cc/cc.h>
116 #include <netinet/tcp_fastopen.h>
118 #include <netinet/tcp_pcap.h>
120 #include <netinet/tcp_syncache.h>
122 #include <netinet/tcp_offload.h>
124 #include <netinet/tcp_ecn.h>
125 #include <netinet/udp.h>
127 #include <netipsec/ipsec_support.h>
129 #include <machine/in_cksum.h>
131 #include <security/mac/mac_framework.h>
133 const int tcprexmtthresh = 3;
135 VNET_DEFINE(int, tcp_log_in_vain) = 0;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
137 &VNET_NAME(tcp_log_in_vain), 0,
138 "Log all incoming TCP segments to closed ports");
140 VNET_DEFINE(int, blackhole) = 0;
141 #define V_blackhole VNET(blackhole)
142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
143 &VNET_NAME(blackhole), 0,
144 "Do not send RST on segments to closed ports");
146 VNET_DEFINE(bool, blackhole_local) = false;
147 #define V_blackhole_local VNET(blackhole_local)
148 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET |
149 CTLFLAG_RW, &VNET_NAME(blackhole_local), false,
150 "Enforce net.inet.tcp.blackhole for locally originated packets");
152 VNET_DEFINE(int, tcp_delack_enabled) = 1;
153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
154 &VNET_NAME(tcp_delack_enabled), 0,
155 "Delay ACK to try and piggyback it onto a data packet");
157 VNET_DEFINE(int, drop_synfin) = 0;
158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
159 &VNET_NAME(drop_synfin), 0,
160 "Drop TCP packets with SYN+FIN set");
162 VNET_DEFINE(int, tcp_do_prr_conservative) = 0;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW,
164 &VNET_NAME(tcp_do_prr_conservative), 0,
165 "Do conservative Proportional Rate Reduction");
167 VNET_DEFINE(int, tcp_do_prr) = 1;
168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW,
169 &VNET_NAME(tcp_do_prr), 1,
170 "Enable Proportional Rate Reduction per RFC 6937");
172 VNET_DEFINE(int, tcp_do_lrd) = 0;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW,
174 &VNET_NAME(tcp_do_lrd), 1,
175 "Perform Lost Retransmission Detection");
177 VNET_DEFINE(int, tcp_do_newcwv) = 0;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
179 &VNET_NAME(tcp_do_newcwv), 0,
180 "Enable New Congestion Window Validation per RFC7661");
182 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
184 &VNET_NAME(tcp_do_rfc3042), 0,
185 "Enable RFC 3042 (Limited Transmit)");
187 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
189 &VNET_NAME(tcp_do_rfc3390), 0,
190 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
192 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
194 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
195 "Slow-start flight size (initial congestion window) in number of segments");
197 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
199 &VNET_NAME(tcp_do_rfc3465), 0,
200 "Enable RFC 3465 (Appropriate Byte Counting)");
202 VNET_DEFINE(int, tcp_abc_l_var) = 2;
203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
204 &VNET_NAME(tcp_abc_l_var), 2,
205 "Cap the max cwnd increment during slow-start to this number of segments");
207 VNET_DEFINE(int, tcp_insecure_syn) = 0;
208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
209 &VNET_NAME(tcp_insecure_syn), 0,
210 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
212 VNET_DEFINE(int, tcp_insecure_rst) = 0;
213 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
214 &VNET_NAME(tcp_insecure_rst), 0,
215 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
217 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
218 #define V_tcp_recvspace VNET(tcp_recvspace)
219 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
220 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
222 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
224 &VNET_NAME(tcp_do_autorcvbuf), 0,
225 "Enable automatic receive buffer sizing");
227 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
229 &VNET_NAME(tcp_autorcvbuf_max), 0,
230 "Max size of automatic receive buffer");
232 VNET_DEFINE(struct inpcbinfo, tcbinfo);
235 * TCP statistics are stored in an array of counter(9)s, which size matches
236 * size of struct tcpstat. TCP running connection count is a regular array.
238 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
239 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
240 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
241 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
242 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
243 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
244 "TCP connection counts by TCP state");
247 * Kernel module interface for updating tcpstat. The first argument is an index
248 * into tcpstat treated as an array.
251 kmod_tcpstat_add(int statnum, int val)
254 counter_u64_add(VNET(tcpstat)[statnum], val);
258 * Make sure that we only start a SACK loss recovery when
259 * receiving a duplicate ACK with a SACK block, and also
260 * complete SACK loss recovery in case the other end
264 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to)
266 return ((tp->t_flags & TF_SACK_PERMIT) &&
267 ((to->to_flags & TOF_SACK) ||
268 (!TAILQ_EMPTY(&tp->snd_holes))));
273 * Wrapper for the TCP established input helper hook.
276 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
278 struct tcp_hhook_data hhook_data;
280 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
285 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
292 * CC wrapper hook functions
295 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
302 INP_WLOCK_ASSERT(tptoinpcb(tp));
304 tp->t_ccv.nsegs = nsegs;
305 tp->t_ccv.bytes_this_ack = BYTES_THIS_ACK(tp, th);
306 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
307 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
308 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
309 tp->t_ccv.flags |= CCF_CWND_LIMITED;
311 tp->t_ccv.flags &= ~CCF_CWND_LIMITED;
313 if (type == CC_ACK) {
315 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
316 ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
317 if (!IN_RECOVERY(tp->t_flags))
318 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
319 tp->t_ccv.bytes_this_ack / (tcp_maxseg(tp) * nsegs));
320 if ((tp->t_flags & TF_GPUTINPROG) &&
321 SEQ_GEQ(th->th_ack, tp->gput_ack)) {
323 * Compute goodput in bits per millisecond.
325 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) /
326 max(1, tcp_ts_getticks() - tp->gput_ts);
327 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
330 * XXXLAS: This is a temporary hack, and should be
331 * chained off VOI_TCP_GPUT when stats(9) grows an API
332 * to deal with chained VOIs.
334 if (tp->t_stats_gput_prev > 0)
335 stats_voi_update_abs_s32(tp->t_stats,
337 ((gput - tp->t_stats_gput_prev) * 100) /
338 tp->t_stats_gput_prev);
339 tp->t_flags &= ~TF_GPUTINPROG;
340 tp->t_stats_gput_prev = gput;
343 if (tp->snd_cwnd > tp->snd_ssthresh) {
344 tp->t_bytes_acked += tp->t_ccv.bytes_this_ack;
345 if (tp->t_bytes_acked >= tp->snd_cwnd) {
346 tp->t_bytes_acked -= tp->snd_cwnd;
347 tp->t_ccv.flags |= CCF_ABC_SENTAWND;
350 tp->t_ccv.flags &= ~CCF_ABC_SENTAWND;
351 tp->t_bytes_acked = 0;
355 if (CC_ALGO(tp)->ack_received != NULL) {
356 /* XXXLAS: Find a way to live without this */
357 tp->t_ccv.curack = th->th_ack;
358 CC_ALGO(tp)->ack_received(&tp->t_ccv, type);
361 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
366 cc_conn_init(struct tcpcb *tp)
368 struct hc_metrics_lite metrics;
369 struct inpcb *inp = tptoinpcb(tp);
373 INP_WLOCK_ASSERT(inp);
375 tcp_hc_get(&inp->inp_inc, &metrics);
376 maxseg = tcp_maxseg(tp);
378 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
380 TCPSTAT_INC(tcps_usedrtt);
381 if (metrics.rmx_rttvar) {
382 tp->t_rttvar = metrics.rmx_rttvar;
383 TCPSTAT_INC(tcps_usedrttvar);
385 /* default variation is +- 1 rtt */
387 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
389 TCPT_RANGESET(tp->t_rxtcur,
390 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
391 tp->t_rttmin, TCPTV_REXMTMAX);
393 if (metrics.rmx_ssthresh) {
395 * There's some sort of gateway or interface
396 * buffer limit on the path. Use this to set
397 * the slow start threshold, but set the
398 * threshold to no less than 2*mss.
400 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
401 TCPSTAT_INC(tcps_usedssthresh);
405 * Set the initial slow-start flight size.
407 * If a SYN or SYN/ACK was lost and retransmitted, we have to
408 * reduce the initial CWND to one segment as congestion is likely
409 * requiring us to be cautious.
411 if (tp->snd_cwnd == 1)
412 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
414 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
416 if (CC_ALGO(tp)->conn_init != NULL)
417 CC_ALGO(tp)->conn_init(&tp->t_ccv);
421 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
423 INP_WLOCK_ASSERT(tptoinpcb(tp));
426 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
431 if (!IN_FASTRECOVERY(tp->t_flags)) {
432 tp->snd_recover = tp->snd_max;
433 if (tp->t_flags2 & TF2_ECN_PERMIT)
434 tp->t_flags2 |= TF2_ECN_SND_CWR;
438 if (!IN_CONGRECOVERY(tp->t_flags) ||
440 * Allow ECN reaction on ACK to CWR, if
441 * that data segment was also CE marked.
443 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
444 EXIT_CONGRECOVERY(tp->t_flags);
445 TCPSTAT_INC(tcps_ecn_rcwnd);
446 tp->snd_recover = tp->snd_max + 1;
447 if (tp->t_flags2 & TF2_ECN_PERMIT)
448 tp->t_flags2 |= TF2_ECN_SND_CWR;
453 tp->t_bytes_acked = 0;
454 EXIT_RECOVERY(tp->t_flags);
455 if (tp->t_flags2 & TF2_ECN_PERMIT)
456 tp->t_flags2 |= TF2_ECN_SND_CWR;
459 TCPSTAT_INC(tcps_sndrexmitbad);
460 /* RTO was unnecessary, so reset everything. */
461 tp->snd_cwnd = tp->snd_cwnd_prev;
462 tp->snd_ssthresh = tp->snd_ssthresh_prev;
463 tp->snd_recover = tp->snd_recover_prev;
464 if (tp->t_flags & TF_WASFRECOVERY)
465 ENTER_FASTRECOVERY(tp->t_flags);
466 if (tp->t_flags & TF_WASCRECOVERY)
467 ENTER_CONGRECOVERY(tp->t_flags);
468 tp->snd_nxt = tp->snd_max;
469 tp->t_flags &= ~TF_PREVVALID;
474 if (CC_ALGO(tp)->cong_signal != NULL) {
476 tp->t_ccv.curack = th->th_ack;
477 CC_ALGO(tp)->cong_signal(&tp->t_ccv, type);
482 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
484 INP_WLOCK_ASSERT(tptoinpcb(tp));
486 /* XXXLAS: KASSERT that we're in recovery? */
488 if (CC_ALGO(tp)->post_recovery != NULL) {
489 tp->t_ccv.curack = th->th_ack;
490 CC_ALGO(tp)->post_recovery(&tp->t_ccv);
492 /* XXXLAS: EXIT_RECOVERY ? */
493 tp->t_bytes_acked = 0;
494 tp->sackhint.delivered_data = 0;
495 tp->sackhint.prr_out = 0;
499 * Indicate whether this ack should be delayed. We can delay the ack if
500 * following conditions are met:
501 * - There is no delayed ack timer in progress.
502 * - Our last ack wasn't a 0-sized window. We never want to delay
503 * the ack that opens up a 0-sized window.
504 * - LRO wasn't used for this segment. We make sure by checking that the
505 * segment size is not larger than the MSS.
507 #define DELAY_ACK(tp, tlen) \
508 ((!tcp_timer_active(tp, TT_DELACK) && \
509 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
510 (tlen <= tp->t_maxseg) && \
511 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
514 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos)
516 INP_WLOCK_ASSERT(tptoinpcb(tp));
518 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
519 switch (iptos & IPTOS_ECN_MASK) {
521 tp->t_ccv.flags |= CCF_IPHDR_CE;
527 case IPTOS_ECN_NOTECT:
528 tp->t_ccv.flags &= ~CCF_IPHDR_CE;
533 tp->t_ccv.flags |= CCF_TCPHDR_CWR;
535 tp->t_ccv.flags &= ~CCF_TCPHDR_CWR;
537 CC_ALGO(tp)->ecnpkt_handler(&tp->t_ccv);
539 if (tp->t_ccv.flags & CCF_ACKNOW) {
540 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
541 tp->t_flags |= TF_ACKNOW;
547 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
549 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos);
553 * TCP input handling is split into multiple parts:
554 * tcp6_input is a thin wrapper around tcp_input for the extended
555 * ip6_protox[] call format in ip6_input
556 * tcp_input handles primary segment validation, inpcb lookup and
557 * SYN processing on listen sockets
558 * tcp_do_segment processes the ACK and text of the segment for
559 * establishing, established and closing connections
563 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
566 struct in6_ifaddr *ia6;
570 if (m->m_len < *offp + sizeof(struct tcphdr)) {
571 m = m_pullup(m, *offp + sizeof(struct tcphdr));
574 TCPSTAT_INC(tcps_rcvshort);
575 return (IPPROTO_DONE);
580 * draft-itojun-ipv6-tcp-to-anycast
581 * better place to put this in?
583 ip6 = mtod(m, struct ip6_hdr *);
584 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
585 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
586 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
587 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
589 return (IPPROTO_DONE);
593 return (tcp_input_with_port(mp, offp, proto, port));
597 tcp6_input(struct mbuf **mp, int *offp, int proto)
600 return(tcp6_input_with_port(mp, offp, proto, 0));
605 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
607 struct mbuf *m = *mp;
608 struct tcphdr *th = NULL;
609 struct ip *ip = NULL;
610 struct inpcb *inp = NULL;
611 struct tcpcb *tp = NULL;
612 struct socket *so = NULL;
623 int rstreason = 0; /* For badport_bandlim accounting purposes */
626 struct m_tag *fwd_tag = NULL;
628 struct ip6_hdr *ip6 = NULL;
631 const void *ip6 = NULL;
633 struct tcpopt to; /* options in this segment */
634 char *s = NULL; /* address and port logging */
639 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
646 TCPSTAT_INC(tcps_rcvtotal);
650 ip6 = mtod(m, struct ip6_hdr *);
651 th = (struct tcphdr *)((caddr_t)ip6 + off0);
652 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
655 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
656 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
657 th->th_sum = m->m_pkthdr.csum_data;
659 th->th_sum = in6_cksum_pseudo(ip6, tlen,
660 IPPROTO_TCP, m->m_pkthdr.csum_data);
661 th->th_sum ^= 0xffff;
663 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
665 TCPSTAT_INC(tcps_rcvbadsum);
670 * Be proactive about unspecified IPv6 address in source.
671 * As we use all-zero to indicate unbounded/unconnected pcb,
672 * unspecified IPv6 address can be used to confuse us.
674 * Note that packets with unspecified IPv6 destination is
675 * already dropped in ip6_input.
677 KASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_dst),
678 ("%s: unspecified destination v6 address", __func__));
679 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
680 IP6STAT_INC(ip6s_badscope); /* XXX */
683 iptos = IPV6_TRAFFIC_CLASS(ip6);
686 #if defined(INET) && defined(INET6)
692 * Get IP and TCP header together in first mbuf.
693 * Note: IP leaves IP header in first mbuf.
695 if (off0 > sizeof (struct ip)) {
697 off0 = sizeof(struct ip);
699 if (m->m_len < sizeof (struct tcpiphdr)) {
700 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
702 TCPSTAT_INC(tcps_rcvshort);
703 return (IPPROTO_DONE);
706 ip = mtod(m, struct ip *);
707 th = (struct tcphdr *)((caddr_t)ip + off0);
708 tlen = ntohs(ip->ip_len) - off0;
713 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
714 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
715 th->th_sum = m->m_pkthdr.csum_data;
717 th->th_sum = in_pseudo(ip->ip_src.s_addr,
719 htonl(m->m_pkthdr.csum_data + tlen +
721 th->th_sum ^= 0xffff;
723 struct ipovly *ipov = (struct ipovly *)ip;
726 * Checksum extended TCP header and data.
730 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
731 ipov->ih_len = htons(tlen);
732 th->th_sum = in_cksum(m, len);
733 /* Reset length for SDT probes. */
734 ip->ip_len = htons(len);
737 /* Re-initialization for later version check */
739 ip->ip_v = IPVERSION;
740 ip->ip_hl = off0 >> 2;
743 if (th->th_sum && (port == 0)) {
744 TCPSTAT_INC(tcps_rcvbadsum);
747 KASSERT(ip->ip_dst.s_addr != INADDR_ANY,
748 ("%s: unspecified destination v4 address", __func__));
749 if (__predict_false(ip->ip_src.s_addr == INADDR_ANY)) {
750 IPSTAT_INC(ips_badaddr);
757 * Check that TCP offset makes sense,
758 * pull out TCP options and adjust length. XXX
760 off = th->th_off << 2;
761 if (off < sizeof (struct tcphdr) || off > tlen) {
762 TCPSTAT_INC(tcps_rcvbadoff);
765 tlen -= off; /* tlen is used instead of ti->ti_len */
766 if (off > sizeof (struct tcphdr)) {
769 if (m->m_len < off0 + off) {
770 m = m_pullup(m, off0 + off);
772 TCPSTAT_INC(tcps_rcvshort);
773 return (IPPROTO_DONE);
776 ip6 = mtod(m, struct ip6_hdr *);
777 th = (struct tcphdr *)((caddr_t)ip6 + off0);
780 #if defined(INET) && defined(INET6)
785 if (m->m_len < sizeof(struct ip) + off) {
786 if ((m = m_pullup(m, sizeof (struct ip) + off))
788 TCPSTAT_INC(tcps_rcvshort);
789 return (IPPROTO_DONE);
791 ip = mtod(m, struct ip *);
792 th = (struct tcphdr *)((caddr_t)ip + off0);
796 optlen = off - sizeof (struct tcphdr);
797 optp = (u_char *)(th + 1);
799 thflags = tcp_get_flags(th);
802 * Convert TCP protocol specific fields to host format.
804 tcp_fields_to_host(th);
807 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
809 drop_hdrlen = off0 + off;
812 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
816 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
818 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
821 #if defined(INET) && !defined(INET6)
822 (m->m_flags & M_IP_NEXTHOP)
825 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
828 * For initial SYN packets we don't need write lock on matching
829 * PCB, be it a listening one or a synchronized one. The packet
830 * shall not modify its state.
832 lookupflag = INPLOOKUP_WILDCARD |
833 ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ?
834 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB);
837 if (isipv6 && fwd_tag != NULL) {
838 struct sockaddr_in6 *next_hop6;
840 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
842 * Transparently forwarded. Pretend to be the destination.
843 * Already got one like this?
845 inp = in6_pcblookup_mbuf(&V_tcbinfo,
846 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
847 lookupflag & ~INPLOOKUP_WILDCARD, m->m_pkthdr.rcvif, m);
850 * It's new. Try to find the ambushing socket.
851 * Because we've rewritten the destination address,
852 * any hardware-generated hash is ignored.
854 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
855 th->th_sport, &next_hop6->sin6_addr,
856 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
857 th->th_dport, lookupflag, m->m_pkthdr.rcvif);
860 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
861 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag,
862 m->m_pkthdr.rcvif, m);
865 #if defined(INET6) && defined(INET)
869 if (fwd_tag != NULL) {
870 struct sockaddr_in *next_hop;
872 next_hop = (struct sockaddr_in *)(fwd_tag+1);
874 * Transparently forwarded. Pretend to be the destination.
875 * already got one like this?
877 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
878 ip->ip_dst, th->th_dport, lookupflag & ~INPLOOKUP_WILDCARD,
879 m->m_pkthdr.rcvif, m);
882 * It's new. Try to find the ambushing socket.
883 * Because we've rewritten the destination address,
884 * any hardware-generated hash is ignored.
886 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
887 th->th_sport, next_hop->sin_addr,
888 next_hop->sin_port ? ntohs(next_hop->sin_port) :
889 th->th_dport, lookupflag, m->m_pkthdr.rcvif);
892 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
893 th->th_sport, ip->ip_dst, th->th_dport, lookupflag,
894 m->m_pkthdr.rcvif, m);
898 * If the INPCB does not exist then all data in the incoming
899 * segment is discarded and an appropriate RST is sent back.
900 * XXX MRT Send RST using which routing table?
903 if (rstreason != 0) {
904 /* We came here after second (safety) lookup. */
905 MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0);
909 * Log communication attempts to ports that are not
912 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
913 V_tcp_log_in_vain == 2) {
914 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
915 log(LOG_INFO, "%s; %s: Connection attempt "
916 "to closed port\n", s, __func__);
919 * When blackholing do not respond with a RST but
920 * completely ignore the segment and drop it.
922 if (((V_blackhole == 1 && (thflags & TH_SYN)) ||
923 V_blackhole == 2) && (V_blackhole_local || (
925 isipv6 ? !in6_localaddr(&ip6->ip6_src) :
928 !in_localip(ip->ip_src)
935 rstreason = BANDLIM_RST_CLOSEDPORT;
938 INP_LOCK_ASSERT(inp);
940 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
941 !SOLISTENING(inp->inp_socket)) {
942 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) {
943 inp->inp_flowid = m->m_pkthdr.flowid;
944 inp->inp_flowtype = M_HASHTYPE_GET(m);
947 /* assign flowid by software RSS hash */
950 rss_proto_software_hash_v6(&inp->in6p_faddr,
960 rss_proto_software_hash_v4(inp->inp_faddr,
971 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
973 if (isipv6 && IPSEC_ENABLED(ipv6) &&
974 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
982 if (IPSEC_ENABLED(ipv4) &&
983 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
990 * Check the minimum TTL for socket.
992 if (inp->inp_ip_minttl != 0) {
995 if (inp->inp_ip_minttl > ip6->ip6_hlim)
999 if (inp->inp_ip_minttl > ip->ip_ttl)
1003 tp = intotcpcb(inp);
1004 switch (tp->t_state) {
1005 case TCPS_TIME_WAIT:
1007 * A previous connection in TIMEWAIT state is supposed to catch
1008 * stray or duplicate segments arriving late. If this segment
1009 * was a legitimate new connection attempt, the old INPCB gets
1010 * removed and we can try again to find a listening socket.
1012 tcp_dooptions(&to, optp, optlen,
1013 (thflags & TH_SYN) ? TO_SYN : 0);
1015 * tcp_twcheck unlocks the inp always, and frees the m if fails.
1017 if (tcp_twcheck(inp, &to, th, m, tlen))
1019 return (IPPROTO_DONE);
1022 * The TCPCB may no longer exist if the connection is winding
1023 * down or it is in the CLOSED state. Either way we drop the
1024 * segment and send an appropriate response.
1026 rstreason = BANDLIM_RST_CLOSEDPORT;
1030 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) {
1031 rstreason = BANDLIM_RST_CLOSEDPORT;
1036 if (tp->t_flags & TF_TOE) {
1037 tcp_offload_input(tp, m);
1038 m = NULL; /* consumed by the TOE driver */
1044 if (mac_inpcb_check_deliver(inp, m))
1047 so = inp->inp_socket;
1048 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1050 * When the socket is accepting connections (the INPCB is in LISTEN
1051 * state) we look into the SYN cache if this is a new connection
1052 * attempt or the completion of a previous one.
1054 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so),
1055 ("%s: so accepting but tp %p not listening", __func__, tp));
1056 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) {
1057 struct in_conninfo inc;
1059 bzero(&inc, sizeof(inc));
1062 inc.inc_flags |= INC_ISIPV6;
1063 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1064 inc.inc_flags |= INC_IPV6MINMTU;
1065 inc.inc6_faddr = ip6->ip6_src;
1066 inc.inc6_laddr = ip6->ip6_dst;
1070 inc.inc_faddr = ip->ip_src;
1071 inc.inc_laddr = ip->ip_dst;
1073 inc.inc_fport = th->th_sport;
1074 inc.inc_lport = th->th_dport;
1075 inc.inc_fibnum = so->so_fibnum;
1078 * Check for an existing connection attempt in syncache if
1079 * the flag is only ACK. A successful lookup creates a new
1080 * socket appended to the listen queue in SYN_RECEIVED state.
1082 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1084 * Parse the TCP options here because
1085 * syncookies need access to the reflected
1088 tcp_dooptions(&to, optp, optlen, 0);
1090 * NB: syncache_expand() doesn't unlock inp.
1092 rstreason = syncache_expand(&inc, &to, th, &so, m, port);
1093 if (rstreason < 0) {
1095 * A failing TCP MD5 signature comparison
1096 * must result in the segment being dropped
1097 * and must not produce any response back
1101 } else if (rstreason == 0) {
1103 * No syncache entry, or ACK was not for our
1104 * SYN/ACK. Do our protection against double
1105 * ACK. If peer sent us 2 ACKs, then for the
1106 * first one syncache_expand() successfully
1107 * converted syncache entry into a socket,
1108 * while we were waiting on the inpcb lock. We
1109 * don't want to sent RST for the second ACK,
1110 * so we perform second lookup without wildcard
1111 * match, hoping to find the new socket. If
1112 * the ACK is stray indeed, rstreason would
1113 * hint the above code that the lookup was a
1116 * NB: syncache did its own logging
1117 * of the failure cause.
1120 rstreason = BANDLIM_RST_OPENPORT;
1121 lookupflag &= ~INPLOOKUP_WILDCARD;
1127 * We completed the 3-way handshake
1128 * but could not allocate a socket
1129 * either due to memory shortage,
1130 * listen queue length limits or
1131 * global socket limits. Send RST
1132 * or wait and have the remote end
1133 * retransmit the ACK for another
1136 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1137 log(LOG_DEBUG, "%s; %s: Listen socket: "
1138 "Socket allocation failed due to "
1139 "limits or memory shortage, %s\n",
1141 V_tcp_sc_rst_sock_fail ?
1142 "sending RST" : "try again");
1143 if (V_tcp_sc_rst_sock_fail) {
1144 rstreason = BANDLIM_UNLIMITED;
1150 * Socket is created in state SYN_RECEIVED.
1151 * Unlock the listen socket, lock the newly
1152 * created socket and update the tp variable.
1153 * If we came here via jump to tfo_socket_result,
1154 * then listening socket is read-locked.
1156 INP_UNLOCK(inp); /* listen socket */
1157 inp = sotoinpcb(so);
1159 * New connection inpcb is already locked by
1160 * syncache_expand().
1162 INP_WLOCK_ASSERT(inp);
1163 tp = intotcpcb(inp);
1164 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1165 ("%s: ", __func__));
1167 * Process the segment and the data it
1168 * contains. tcp_do_segment() consumes
1169 * the mbuf chain and unlocks the inpcb.
1171 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1172 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen,
1174 return (IPPROTO_DONE);
1177 * Segment flag validation for new connection attempts:
1179 * Our (SYN|ACK) response was rejected.
1180 * Check with syncache and remove entry to prevent
1183 * NB: syncache_chkrst does its own logging of failure
1186 if (thflags & TH_RST) {
1187 syncache_chkrst(&inc, th, m, port);
1191 * We can't do anything without SYN.
1193 if ((thflags & TH_SYN) == 0) {
1194 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1195 log(LOG_DEBUG, "%s; %s: Listen socket: "
1196 "SYN is missing, segment ignored\n",
1198 TCPSTAT_INC(tcps_badsyn);
1202 * (SYN|ACK) is bogus on a listen socket.
1204 if (thflags & TH_ACK) {
1205 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1206 log(LOG_DEBUG, "%s; %s: Listen socket: "
1207 "SYN|ACK invalid, segment rejected\n",
1209 syncache_badack(&inc, port); /* XXX: Not needed! */
1210 TCPSTAT_INC(tcps_badsyn);
1211 rstreason = BANDLIM_RST_OPENPORT;
1215 * If the drop_synfin option is enabled, drop all
1216 * segments with both the SYN and FIN bits set.
1217 * This prevents e.g. nmap from identifying the
1219 * XXX: Poor reasoning. nmap has other methods
1220 * and is constantly refining its stack detection
1222 * XXX: This is a violation of the TCP specification
1223 * and was used by RFC1644.
1225 if ((thflags & TH_FIN) && V_drop_synfin) {
1226 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1227 log(LOG_DEBUG, "%s; %s: Listen socket: "
1228 "SYN|FIN segment ignored (based on "
1229 "sysctl setting)\n", s, __func__);
1230 TCPSTAT_INC(tcps_badsyn);
1234 * Segment's flags are (SYN) or (SYN|FIN).
1236 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1237 * as they do not affect the state of the TCP FSM.
1238 * The data pointed to by TH_URG and th_urp is ignored.
1240 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1241 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1242 KASSERT(thflags & (TH_SYN),
1243 ("%s: Listen socket: TH_SYN not set", __func__));
1244 INP_RLOCK_ASSERT(inp);
1247 * If deprecated address is forbidden,
1248 * we do not accept SYN to deprecated interface
1249 * address to prevent any new inbound connection from
1250 * getting established.
1251 * When we do not accept SYN, we send a TCP RST,
1252 * with deprecated source address (instead of dropping
1253 * it). We compromise it as it is much better for peer
1254 * to send a RST, and RST will be the final packet
1257 * If we do not forbid deprecated addresses, we accept
1258 * the SYN packet. RFC2462 does not suggest dropping
1260 * If we decipher RFC2462 5.5.4, it says like this:
1261 * 1. use of deprecated addr with existing
1262 * communication is okay - "SHOULD continue to be
1264 * 2. use of it with new communication:
1265 * (2a) "SHOULD NOT be used if alternate address
1266 * with sufficient scope is available"
1267 * (2b) nothing mentioned otherwise.
1268 * Here we fall into (2b) case as we have no choice in
1269 * our source address selection - we must obey the peer.
1271 * The wording in RFC2462 is confusing, and there are
1272 * multiple description text for deprecated address
1273 * handling - worse, they are not exactly the same.
1274 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1276 if (isipv6 && !V_ip6_use_deprecated) {
1277 struct in6_ifaddr *ia6;
1279 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1281 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1282 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1283 log(LOG_DEBUG, "%s; %s: Listen socket: "
1284 "Connection attempt to deprecated "
1285 "IPv6 address rejected\n",
1287 rstreason = BANDLIM_RST_OPENPORT;
1293 * Basic sanity checks on incoming SYN requests:
1294 * Don't respond if the destination is a link layer
1295 * broadcast according to RFC1122 4.2.3.10, p. 104.
1296 * If it is from this socket it must be forged.
1297 * Don't respond if the source or destination is a
1298 * global or subnet broad- or multicast address.
1299 * Note that it is quite possible to receive unicast
1300 * link-layer packets with a broadcast IP address. Use
1301 * in_broadcast() to find them.
1303 if (m->m_flags & (M_BCAST|M_MCAST)) {
1304 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1305 log(LOG_DEBUG, "%s; %s: Listen socket: "
1306 "Connection attempt from broad- or multicast "
1307 "link layer address ignored\n", s, __func__);
1312 if (th->th_dport == th->th_sport &&
1313 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1314 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1315 log(LOG_DEBUG, "%s; %s: Listen socket: "
1316 "Connection attempt to/from self "
1317 "ignored\n", s, __func__);
1320 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1321 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1322 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1323 log(LOG_DEBUG, "%s; %s: Listen socket: "
1324 "Connection attempt from/to multicast "
1325 "address ignored\n", s, __func__);
1330 #if defined(INET) && defined(INET6)
1335 if (th->th_dport == th->th_sport &&
1336 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1337 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1338 log(LOG_DEBUG, "%s; %s: Listen socket: "
1339 "Connection attempt from/to self "
1340 "ignored\n", s, __func__);
1343 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1344 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1345 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1346 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1347 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1348 log(LOG_DEBUG, "%s; %s: Listen socket: "
1349 "Connection attempt from/to broad- "
1350 "or multicast address ignored\n",
1357 * SYN appears to be valid. Create compressed TCP state
1360 TCP_PROBE3(debug__input, tp, th, m);
1361 tcp_dooptions(&to, optp, optlen, TO_SYN);
1362 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL,
1363 iptos, port)) != NULL)
1364 goto tfo_socket_result;
1367 * Entry added to syncache and mbuf consumed.
1368 * Only the listen socket is unlocked by syncache_add().
1370 return (IPPROTO_DONE);
1371 } else if (tp->t_state == TCPS_LISTEN) {
1373 * When a listen socket is torn down the SO_ACCEPTCONN
1374 * flag is removed first while connections are drained
1375 * from the accept queue in a unlock/lock cycle of the
1376 * ACCEPT_LOCK, opening a race condition allowing a SYN
1377 * attempt go through unhandled.
1381 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1382 if (tp->t_flags & TF_SIGNATURE) {
1383 tcp_dooptions(&to, optp, optlen, thflags);
1384 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1385 TCPSTAT_INC(tcps_sig_err_nosigopt);
1388 if (!TCPMD5_ENABLED() ||
1389 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1393 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1396 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1397 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1398 * the inpcb, and unlocks pcbinfo.
1400 * XXXGL: in case of a pure SYN arriving on existing connection
1401 * TCP stacks won't need to modify the PCB, they would either drop
1402 * the segment silently, or send a challenge ACK. However, we try
1403 * to upgrade the lock, because calling convention for stacks is
1404 * write-lock on PCB. If upgrade fails, drop the SYN.
1406 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0)
1409 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, tlen, iptos);
1410 return (IPPROTO_DONE);
1413 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1416 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1419 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1420 m = NULL; /* mbuf chain got consumed. */
1425 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1435 return (IPPROTO_DONE);
1439 * Automatic sizing of receive socket buffer. Often the send
1440 * buffer size is not optimally adjusted to the actual network
1441 * conditions at hand (delay bandwidth product). Setting the
1442 * buffer size too small limits throughput on links with high
1443 * bandwidth and high delay (eg. trans-continental/oceanic links).
1445 * On the receive side the socket buffer memory is only rarely
1446 * used to any significant extent. This allows us to be much
1447 * more aggressive in scaling the receive socket buffer. For
1448 * the case that the buffer space is actually used to a large
1449 * extent and we run out of kernel memory we can simply drop
1450 * the new segments; TCP on the sender will just retransmit it
1451 * later. Setting the buffer size too big may only consume too
1452 * much kernel memory if the application doesn't read() from
1453 * the socket or packet loss or reordering makes use of the
1456 * The criteria to step up the receive buffer one notch are:
1457 * 1. Application has not set receive buffer size with
1458 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1459 * 2. the number of bytes received during 1/2 of an sRTT
1460 * is at least 3/8 of the current socket buffer size.
1461 * 3. receive buffer size has not hit maximal automatic size;
1463 * If all of the criteria are met we increaset the socket buffer
1464 * by a 1/2 (bounded by the max). This allows us to keep ahead
1465 * of slow-start but also makes it so our peer never gets limited
1466 * by our rwnd which we then open up causing a burst.
1468 * This algorithm does two steps per RTT at most and only if
1469 * we receive a bulk stream w/o packet losses or reorderings.
1470 * Shrinking the buffer during idle times is not necessary as
1471 * it doesn't consume any memory when idle.
1473 * TODO: Only step up if the application is actually serving
1474 * the buffer to better manage the socket buffer resources.
1477 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1478 struct tcpcb *tp, int tlen)
1482 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1483 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1484 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1485 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1486 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1487 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1488 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1490 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1492 /* Start over with next RTT. */
1496 tp->rfbuf_cnt += tlen; /* add up */
1502 tcp_input(struct mbuf **mp, int *offp, int proto)
1504 return(tcp_input_with_port(mp, offp, proto, 0));
1508 tcp_handle_wakeup(struct tcpcb *tp)
1511 INP_WLOCK_ASSERT(tptoinpcb(tp));
1513 if (tp->t_flags & TF_WAKESOR) {
1514 struct socket *so = tptosocket(tp);
1516 tp->t_flags &= ~TF_WAKESOR;
1517 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1518 sorwakeup_locked(so);
1523 tcp_do_segment(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th,
1524 int drop_hdrlen, int tlen, uint8_t iptos)
1527 int acked, ourfinisacked, needoutput = 0, sack_changed;
1528 int rstreason, todrop, win, incforsyn = 0;
1532 struct inpcb *inp = tptoinpcb(tp);
1533 struct socket *so = tptosocket(tp);
1534 struct in_conninfo *inc = &inp->inp_inc;
1540 thflags = tcp_get_flags(th);
1541 tp->sackhint.last_sack_ack = 0;
1543 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1546 INP_WLOCK_ASSERT(inp);
1547 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1549 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1553 /* Save segment, if requested. */
1554 tcp_pcap_add(th, m, &(tp->t_inpkts));
1556 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1559 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1560 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1561 log(LOG_DEBUG, "%s; %s: "
1562 "SYN|FIN segment ignored (based on "
1563 "sysctl setting)\n", s, __func__);
1570 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1571 * check SEQ.ACK first.
1573 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1574 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1575 rstreason = BANDLIM_UNLIMITED;
1576 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
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 if (tp->t_idle_reduce &&
1587 (tp->snd_max == tp->snd_una) &&
1588 ((ticks - tp->t_rcvtime) >= tp->t_rxtcur))
1590 tp->t_rcvtime = ticks;
1592 if (thflags & TH_FIN)
1593 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN);
1595 * Scale up the window into a 32-bit value.
1596 * For the SYN_SENT state the scale is zero.
1598 tiwin = th->th_win << tp->snd_scale;
1600 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1604 * TCP ECN processing.
1606 if (tcp_ecn_input_segment(tp, thflags, tlen,
1607 tcp_packets_this_ack(tp, th->th_ack),
1609 cc_cong_signal(tp, th, CC_ECN);
1612 * Parse options on any incoming segment.
1614 tcp_dooptions(&to, (u_char *)(th + 1),
1615 (th->th_off << 2) - sizeof(struct tcphdr),
1616 (thflags & TH_SYN) ? TO_SYN : 0);
1618 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1619 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1620 (to.to_flags & TOF_SIGNATURE) == 0) {
1621 TCPSTAT_INC(tcps_sig_err_sigopt);
1622 /* XXX: should drop? */
1626 * If echoed timestamp is later than the current time,
1627 * fall back to non RFC1323 RTT calculation. Normalize
1628 * timestamp if syncookies were used when this connection
1631 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1632 to.to_tsecr -= tp->ts_offset;
1633 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1635 else if (tp->t_rxtshift == 1 &&
1636 tp->t_flags & TF_PREVVALID &&
1637 tp->t_badrxtwin != 0 &&
1638 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
1639 cc_cong_signal(tp, th, CC_RTO_ERR);
1642 * Process options only when we get SYN/ACK back. The SYN case
1643 * for incoming connections is handled in tcp_syncache.
1644 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1645 * or <SYN,ACK>) segment itself is never scaled.
1646 * XXX this is traditional behavior, may need to be cleaned up.
1648 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1649 /* Handle parallel SYN for ECN */
1650 tcp_ecn_input_parallel_syn(tp, thflags, iptos);
1651 if ((to.to_flags & TOF_SCALE) &&
1652 (tp->t_flags & TF_REQ_SCALE) &&
1653 !(tp->t_flags & TF_NOOPT)) {
1654 tp->t_flags |= TF_RCVD_SCALE;
1655 tp->snd_scale = to.to_wscale;
1657 tp->t_flags &= ~TF_REQ_SCALE;
1659 * Initial send window. It will be updated with
1660 * the next incoming segment to the scaled value.
1662 tp->snd_wnd = th->th_win;
1663 if ((to.to_flags & TOF_TS) &&
1664 (tp->t_flags & TF_REQ_TSTMP) &&
1665 !(tp->t_flags & TF_NOOPT)) {
1666 tp->t_flags |= TF_RCVD_TSTMP;
1667 tp->ts_recent = to.to_tsval;
1668 tp->ts_recent_age = tcp_ts_getticks();
1670 tp->t_flags &= ~TF_REQ_TSTMP;
1671 if (to.to_flags & TOF_MSS)
1672 tcp_mss(tp, to.to_mss);
1673 if ((tp->t_flags & TF_SACK_PERMIT) &&
1674 (!(to.to_flags & TOF_SACKPERM) ||
1675 (tp->t_flags & TF_NOOPT)))
1676 tp->t_flags &= ~TF_SACK_PERMIT;
1677 if (IS_FASTOPEN(tp->t_flags)) {
1678 if ((to.to_flags & TOF_FASTOPEN) &&
1679 !(tp->t_flags & TF_NOOPT)) {
1682 if (to.to_flags & TOF_MSS)
1685 if ((inp->inp_vflag & INP_IPV6) != 0)
1689 tcp_fastopen_update_cache(tp, mss,
1690 to.to_tfo_len, to.to_tfo_cookie);
1692 tcp_fastopen_disable_path(tp);
1697 * If timestamps were negotiated during SYN/ACK and a
1698 * segment without a timestamp is received, silently drop
1699 * the segment, unless it is a RST segment or missing timestamps are
1701 * See section 3.2 of RFC 7323.
1703 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1704 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1705 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1706 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1707 "segment processed normally\n",
1712 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1713 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1714 "segment silently dropped\n", s, __func__);
1721 * If timestamps were not negotiated during SYN/ACK and a
1722 * segment with a timestamp is received, ignore the
1723 * timestamp and process the packet normally.
1724 * See section 3.2 of RFC 7323.
1726 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1727 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1728 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1729 "segment processed normally\n", s, __func__);
1735 * Header prediction: check for the two common cases
1736 * of a uni-directional data xfer. If the packet has
1737 * no control flags, is in-sequence, the window didn't
1738 * change and we're not retransmitting, it's a
1739 * candidate. If the length is zero and the ack moved
1740 * forward, we're the sender side of the xfer. Just
1741 * free the data acked & wake any higher level process
1742 * that was blocked waiting for space. If the length
1743 * is non-zero and the ack didn't move, we're the
1744 * receiver side. If we're getting packets in-order
1745 * (the reassembly queue is empty), add the data to
1746 * the socket buffer and note that we need a delayed ack.
1747 * Make sure that the hidden state-flags are also off.
1748 * Since we check for TCPS_ESTABLISHED first, it can only
1751 if (tp->t_state == TCPS_ESTABLISHED &&
1752 th->th_seq == tp->rcv_nxt &&
1753 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1754 tp->snd_nxt == tp->snd_max &&
1755 tiwin && tiwin == tp->snd_wnd &&
1756 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1758 ((to.to_flags & TOF_TS) == 0 ||
1759 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1761 * If last ACK falls within this segment's sequence numbers,
1762 * record the timestamp.
1763 * NOTE that the test is modified according to the latest
1764 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1766 if ((to.to_flags & TOF_TS) != 0 &&
1767 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1768 tp->ts_recent_age = tcp_ts_getticks();
1769 tp->ts_recent = to.to_tsval;
1773 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1774 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1775 !IN_RECOVERY(tp->t_flags) &&
1776 (to.to_flags & TOF_SACK) == 0 &&
1777 TAILQ_EMPTY(&tp->snd_holes)) {
1779 * This is a pure ack for outstanding data.
1781 TCPSTAT_INC(tcps_predack);
1784 * "bad retransmit" recovery without timestamps.
1786 if ((to.to_flags & TOF_TS) == 0 &&
1787 tp->t_rxtshift == 1 &&
1788 tp->t_flags & TF_PREVVALID &&
1789 tp->t_badrxtwin != 0 &&
1790 TSTMP_LT(ticks, tp->t_badrxtwin)) {
1791 cc_cong_signal(tp, th, CC_RTO_ERR);
1795 * Recalculate the transmit timer / rtt.
1797 * Some boxes send broken timestamp replies
1798 * during the SYN+ACK phase, ignore
1799 * timestamps of 0 or we could calculate a
1800 * huge RTT and blow up the retransmit timer.
1802 if ((to.to_flags & TOF_TS) != 0 &&
1806 t = tcp_ts_getticks() - to.to_tsecr;
1807 if (!tp->t_rttlow || tp->t_rttlow > t)
1810 TCP_TS_TO_TICKS(t) + 1);
1811 } else if (tp->t_rtttime &&
1812 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1813 if (!tp->t_rttlow ||
1814 tp->t_rttlow > ticks - tp->t_rtttime)
1815 tp->t_rttlow = ticks - tp->t_rtttime;
1817 ticks - tp->t_rtttime);
1819 acked = BYTES_THIS_ACK(tp, th);
1822 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1823 hhook_run_tcp_est_in(tp, th, &to);
1826 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1827 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1828 sbdrop(&so->so_snd, acked);
1829 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1830 SEQ_LEQ(th->th_ack, tp->snd_recover))
1831 tp->snd_recover = th->th_ack - 1;
1834 * Let the congestion control algorithm update
1835 * congestion control related information. This
1836 * typically means increasing the congestion
1839 cc_ack_received(tp, th, nsegs, CC_ACK);
1841 tp->snd_una = th->th_ack;
1843 * Pull snd_wl2 up to prevent seq wrap relative
1846 tp->snd_wl2 = th->th_ack;
1851 * If all outstanding data are acked, stop
1852 * retransmit timer, otherwise restart timer
1853 * using current (possibly backed-off) value.
1854 * If process is waiting for space,
1855 * wakeup/selwakeup/signal. If data
1856 * are ready to send, let tcp_output
1857 * decide between more output or persist.
1859 TCP_PROBE3(debug__input, tp, th, m);
1861 * Clear t_acktime if remote side has ACKd
1862 * all data in the socket buffer.
1863 * Otherwise, update t_acktime if we received
1864 * a sufficiently large ACK.
1866 if (sbavail(&so->so_snd) == 0)
1869 tp->t_acktime = ticks;
1870 if (tp->snd_una == tp->snd_max)
1871 tcp_timer_activate(tp, TT_REXMT, 0);
1872 else if (!tcp_timer_active(tp, TT_PERSIST))
1873 tcp_timer_activate(tp, TT_REXMT,
1876 if (sbavail(&so->so_snd))
1877 (void) tcp_output(tp);
1880 } else if (th->th_ack == tp->snd_una &&
1881 tlen <= sbspace(&so->so_rcv)) {
1882 int newsize = 0; /* automatic sockbuf scaling */
1885 * This is a pure, in-sequence data packet with
1886 * nothing on the reassembly queue and we have enough
1887 * buffer space to take it.
1889 /* Clean receiver SACK report if present */
1890 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1891 tcp_clean_sackreport(tp);
1892 TCPSTAT_INC(tcps_preddat);
1893 tp->rcv_nxt += tlen;
1895 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1896 (tp->t_fbyte_in == 0)) {
1897 tp->t_fbyte_in = ticks;
1898 if (tp->t_fbyte_in == 0)
1900 if (tp->t_fbyte_out && tp->t_fbyte_in)
1901 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1904 * Pull snd_wl1 up to prevent seq wrap relative to
1907 tp->snd_wl1 = th->th_seq;
1909 * Pull rcv_up up to prevent seq wrap relative to
1912 tp->rcv_up = tp->rcv_nxt;
1913 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1914 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1915 TCP_PROBE3(debug__input, tp, th, m);
1917 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1919 /* Add data to socket buffer. */
1920 SOCKBUF_LOCK(&so->so_rcv);
1921 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1925 * Set new socket buffer size.
1926 * Give up when limit is reached.
1929 if (!sbreserve_locked(so, SO_RCV,
1931 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1932 m_adj(m, drop_hdrlen); /* delayed header drop */
1933 sbappendstream_locked(&so->so_rcv, m, 0);
1935 /* NB: sorwakeup_locked() does an implicit unlock. */
1936 sorwakeup_locked(so);
1937 if (DELAY_ACK(tp, tlen)) {
1938 tp->t_flags |= TF_DELACK;
1940 tp->t_flags |= TF_ACKNOW;
1948 * Calculate amount of space in receive window,
1949 * and then do TCP input processing.
1950 * Receive window is amount of space in rcv queue,
1951 * but not less than advertised window.
1953 win = sbspace(&so->so_rcv);
1956 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1958 switch (tp->t_state) {
1960 * If the state is SYN_RECEIVED:
1961 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1963 case TCPS_SYN_RECEIVED:
1964 if (thflags & TH_RST) {
1965 /* Handle RST segments later. */
1968 if ((thflags & TH_ACK) &&
1969 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1970 SEQ_GT(th->th_ack, tp->snd_max))) {
1971 rstreason = BANDLIM_RST_OPENPORT;
1972 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
1975 if (IS_FASTOPEN(tp->t_flags)) {
1977 * When a TFO connection is in SYN_RECEIVED, the
1978 * only valid packets are the initial SYN, a
1979 * retransmit/copy of the initial SYN (possibly with
1980 * a subset of the original data), a valid ACK, a
1983 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1984 rstreason = BANDLIM_RST_OPENPORT;
1985 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
1987 } else if (thflags & TH_SYN) {
1988 /* non-initial SYN is ignored */
1989 if ((tcp_timer_active(tp, TT_DELACK) ||
1990 tcp_timer_active(tp, TT_REXMT)))
1992 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1999 * If the state is SYN_SENT:
2000 * if seg contains a RST with valid ACK (SEQ.ACK has already
2001 * been verified), then drop the connection.
2002 * if seg contains a RST without an ACK, drop the seg.
2003 * if seg does not contain SYN, then drop the seg.
2004 * Otherwise this is an acceptable SYN segment
2005 * initialize tp->rcv_nxt and tp->irs
2006 * if seg contains ack then advance tp->snd_una
2007 * if seg contains an ECE and ECN support is enabled, the stream
2009 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2010 * arrange for segment to be acked (eventually)
2011 * continue processing rest of data/controls, beginning with URG
2014 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2015 TCP_PROBE5(connect__refused, NULL, tp,
2017 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2018 tp = tcp_drop(tp, ECONNREFUSED);
2020 if (thflags & TH_RST)
2022 if (!(thflags & TH_SYN))
2025 tp->irs = th->th_seq;
2027 if (thflags & TH_ACK) {
2028 int tfo_partial_ack = 0;
2030 TCPSTAT_INC(tcps_connects);
2033 mac_socketpeer_set_from_mbuf(m, so);
2035 /* Do window scaling on this connection? */
2036 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2037 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2038 tp->rcv_scale = tp->request_r_scale;
2040 tp->rcv_adv += min(tp->rcv_wnd,
2041 TCP_MAXWIN << tp->rcv_scale);
2042 tp->snd_una++; /* SYN is acked */
2044 * If not all the data that was sent in the TFO SYN
2045 * has been acked, resend the remainder right away.
2047 if (IS_FASTOPEN(tp->t_flags) &&
2048 (tp->snd_una != tp->snd_max)) {
2049 tp->snd_nxt = th->th_ack;
2050 tfo_partial_ack = 1;
2053 * If there's data, delay ACK; if there's also a FIN
2054 * ACKNOW will be turned on later.
2056 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2057 tcp_timer_activate(tp, TT_DELACK,
2060 tp->t_flags |= TF_ACKNOW;
2062 tcp_ecn_input_syn_sent(tp, thflags, iptos);
2065 * Received <SYN,ACK> in SYN_SENT[*] state.
2067 * SYN_SENT --> ESTABLISHED
2068 * SYN_SENT* --> FIN_WAIT_1
2070 tp->t_starttime = ticks;
2071 if (tp->t_flags & TF_NEEDFIN) {
2072 tp->t_acktime = ticks;
2073 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2074 tp->t_flags &= ~TF_NEEDFIN;
2077 tcp_state_change(tp, TCPS_ESTABLISHED);
2078 TCP_PROBE5(connect__established, NULL, tp,
2081 tcp_timer_activate(tp, TT_KEEP,
2086 * Received initial SYN in SYN-SENT[*] state =>
2087 * simultaneous open.
2088 * If it succeeds, connection is * half-synchronized.
2089 * Otherwise, do 3-way handshake:
2090 * SYN-SENT -> SYN-RECEIVED
2091 * SYN-SENT* -> SYN-RECEIVED*
2093 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN);
2094 tcp_timer_activate(tp, TT_REXMT, 0);
2095 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2099 * Advance th->th_seq to correspond to first data byte.
2100 * If data, trim to stay within window,
2101 * dropping FIN if necessary.
2104 if (tlen > tp->rcv_wnd) {
2105 todrop = tlen - tp->rcv_wnd;
2109 TCPSTAT_INC(tcps_rcvpackafterwin);
2110 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2112 tp->snd_wl1 = th->th_seq - 1;
2113 tp->rcv_up = th->th_seq;
2115 * Client side of transaction: already sent SYN and data.
2116 * If the remote host used T/TCP to validate the SYN,
2117 * our data will be ACK'd; if so, enter normal data segment
2118 * processing in the middle of step 5, ack processing.
2119 * Otherwise, goto step 6.
2121 if (thflags & TH_ACK)
2128 * States other than LISTEN or SYN_SENT.
2129 * First check the RST flag and sequence number since reset segments
2130 * are exempt from the timestamp and connection count tests. This
2131 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2132 * below which allowed reset segments in half the sequence space
2133 * to fall though and be processed (which gives forged reset
2134 * segments with a random sequence number a 50 percent chance of
2135 * killing a connection).
2136 * Then check timestamp, if present.
2137 * Then check the connection count, if present.
2138 * Then check that at least some bytes of segment are within
2139 * receive window. If segment begins before rcv_nxt,
2140 * drop leading data (and SYN); if nothing left, just ack.
2142 if (thflags & TH_RST) {
2144 * RFC5961 Section 3.2
2146 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2147 * - If RST is in window, we send challenge ACK.
2149 * Note: to take into account delayed ACKs, we should
2150 * test against last_ack_sent instead of rcv_nxt.
2151 * Note 2: we handle special case of closed window, not
2152 * covered by the RFC.
2154 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2155 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2156 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2157 KASSERT(tp->t_state != TCPS_SYN_SENT,
2158 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2161 if (V_tcp_insecure_rst ||
2162 tp->last_ack_sent == th->th_seq) {
2163 TCPSTAT_INC(tcps_drops);
2164 /* Drop the connection. */
2165 switch (tp->t_state) {
2166 case TCPS_SYN_RECEIVED:
2167 so->so_error = ECONNREFUSED;
2169 case TCPS_ESTABLISHED:
2170 case TCPS_FIN_WAIT_1:
2171 case TCPS_FIN_WAIT_2:
2172 case TCPS_CLOSE_WAIT:
2175 so->so_error = ECONNRESET;
2179 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST);
2183 TCPSTAT_INC(tcps_badrst);
2184 /* Send challenge ACK. */
2185 tcp_respond(tp, mtod(m, void *), th, m,
2186 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2187 tp->last_ack_sent = tp->rcv_nxt;
2195 * RFC5961 Section 4.2
2196 * Send challenge ACK for any SYN in synchronized state.
2198 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2199 tp->t_state != TCPS_SYN_RECEIVED) {
2200 TCPSTAT_INC(tcps_badsyn);
2201 if (V_tcp_insecure_syn &&
2202 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2203 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2204 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2205 tp = tcp_drop(tp, ECONNRESET);
2206 rstreason = BANDLIM_UNLIMITED;
2208 tcp_ecn_input_syn_sent(tp, thflags, iptos);
2209 /* Send challenge ACK. */
2210 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2211 tp->snd_nxt, TH_ACK);
2212 tp->last_ack_sent = tp->rcv_nxt;
2219 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2220 * and it's less than ts_recent, drop it.
2222 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2223 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2224 /* Check to see if ts_recent is over 24 days old. */
2225 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2227 * Invalidate ts_recent. If this segment updates
2228 * ts_recent, the age will be reset later and ts_recent
2229 * will get a valid value. If it does not, setting
2230 * ts_recent to zero will at least satisfy the
2231 * requirement that zero be placed in the timestamp
2232 * echo reply when ts_recent isn't valid. The
2233 * age isn't reset until we get a valid ts_recent
2234 * because we don't want out-of-order segments to be
2235 * dropped when ts_recent is old.
2239 TCPSTAT_INC(tcps_rcvduppack);
2240 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2241 TCPSTAT_INC(tcps_pawsdrop);
2249 * In the SYN-RECEIVED state, validate that the packet belongs to
2250 * this connection before trimming the data to fit the receive
2251 * window. Check the sequence number versus IRS since we know
2252 * the sequence numbers haven't wrapped. This is a partial fix
2253 * for the "LAND" DoS attack.
2255 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2256 rstreason = BANDLIM_RST_OPENPORT;
2257 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2261 todrop = tp->rcv_nxt - th->th_seq;
2263 if (thflags & TH_SYN) {
2273 * Following if statement from Stevens, vol. 2, p. 960.
2276 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2278 * Any valid FIN must be to the left of the window.
2279 * At this point the FIN must be a duplicate or out
2280 * of sequence; drop it.
2285 * Send an ACK to resynchronize and drop any data.
2286 * But keep on processing for RST or ACK.
2288 tp->t_flags |= TF_ACKNOW;
2290 TCPSTAT_INC(tcps_rcvduppack);
2291 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2293 TCPSTAT_INC(tcps_rcvpartduppack);
2294 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2297 * DSACK - add SACK block for dropped range
2299 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2300 tcp_update_sack_list(tp, th->th_seq,
2301 th->th_seq + todrop);
2303 * ACK now, as the next in-sequence segment
2304 * will clear the DSACK block again
2306 tp->t_flags |= TF_ACKNOW;
2308 drop_hdrlen += todrop; /* drop from the top afterwards */
2309 th->th_seq += todrop;
2311 if (th->th_urp > todrop)
2312 th->th_urp -= todrop;
2320 * If new data are received on a connection after the
2321 * user processes are gone, then RST the other end.
2323 if ((tp->t_flags & TF_CLOSED) && tlen) {
2324 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2325 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2326 "after socket was closed, "
2327 "sending RST and removing tcpcb\n",
2328 s, __func__, tcpstates[tp->t_state], tlen);
2331 tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE);
2332 /* tcp_close will kill the inp pre-log the Reset */
2333 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST);
2335 TCPSTAT_INC(tcps_rcvafterclose);
2336 rstreason = BANDLIM_UNLIMITED;
2341 * If segment ends after window, drop trailing data
2342 * (and PUSH and FIN); if nothing left, just ACK.
2344 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2346 TCPSTAT_INC(tcps_rcvpackafterwin);
2347 if (todrop >= tlen) {
2348 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2350 * If window is closed can only take segments at
2351 * window edge, and have to drop data and PUSH from
2352 * incoming segments. Continue processing, but
2353 * remember to ack. Otherwise, drop segment
2356 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2357 tp->t_flags |= TF_ACKNOW;
2358 TCPSTAT_INC(tcps_rcvwinprobe);
2362 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2365 thflags &= ~(TH_PUSH|TH_FIN);
2369 * If last ACK falls within this segment's sequence numbers,
2370 * record its timestamp.
2372 * 1) That the test incorporates suggestions from the latest
2373 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2374 * 2) That updating only on newer timestamps interferes with
2375 * our earlier PAWS tests, so this check should be solely
2376 * predicated on the sequence space of this segment.
2377 * 3) That we modify the segment boundary check to be
2378 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2379 * instead of RFC1323's
2380 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2381 * This modified check allows us to overcome RFC1323's
2382 * limitations as described in Stevens TCP/IP Illustrated
2383 * Vol. 2 p.869. In such cases, we can still calculate the
2384 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2386 if ((to.to_flags & TOF_TS) != 0 &&
2387 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2388 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2389 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2390 tp->ts_recent_age = tcp_ts_getticks();
2391 tp->ts_recent = to.to_tsval;
2395 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2396 * flag is on (half-synchronized state), then queue data for
2397 * later processing; else drop segment and return.
2399 if ((thflags & TH_ACK) == 0) {
2400 if (tp->t_state == TCPS_SYN_RECEIVED ||
2401 (tp->t_flags & TF_NEEDSYN)) {
2402 if (tp->t_state == TCPS_SYN_RECEIVED &&
2403 IS_FASTOPEN(tp->t_flags)) {
2404 tp->snd_wnd = tiwin;
2408 } else if (tp->t_flags & TF_ACKNOW)
2417 switch (tp->t_state) {
2419 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2420 * ESTABLISHED state and continue processing.
2421 * The ACK was checked above.
2423 case TCPS_SYN_RECEIVED:
2425 TCPSTAT_INC(tcps_connects);
2426 if (tp->t_flags & TF_SONOTCONN) {
2428 * Usually SYN_RECEIVED had been created from a LISTEN,
2429 * and solisten_enqueue() has already marked the socket
2430 * layer as connected. If it didn't, which can happen
2431 * only with an accept_filter(9), then the tp is marked
2432 * with TF_SONOTCONN. The other reason for this mark
2433 * to be set is a simultaneous open, a SYN_RECEIVED
2434 * that had been created from SYN_SENT.
2436 tp->t_flags &= ~TF_SONOTCONN;
2439 /* Do window scaling? */
2440 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2441 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2442 tp->rcv_scale = tp->request_r_scale;
2444 tp->snd_wnd = tiwin;
2447 * SYN-RECEIVED -> ESTABLISHED
2448 * SYN-RECEIVED* -> FIN-WAIT-1
2450 tp->t_starttime = ticks;
2451 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2452 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2453 tp->t_tfo_pending = NULL;
2455 if (tp->t_flags & TF_NEEDFIN) {
2456 tp->t_acktime = ticks;
2457 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2458 tp->t_flags &= ~TF_NEEDFIN;
2460 tcp_state_change(tp, TCPS_ESTABLISHED);
2461 TCP_PROBE5(accept__established, NULL, tp,
2464 * TFO connections call cc_conn_init() during SYN
2465 * processing. Calling it again here for such
2466 * connections is not harmless as it would undo the
2467 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2470 if (!IS_FASTOPEN(tp->t_flags))
2472 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2475 * Account for the ACK of our SYN prior to
2476 * regular ACK processing below, except for
2477 * simultaneous SYN, which is handled later.
2479 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2482 * If segment contains data or ACK, will call tcp_reass()
2483 * later; if not, do so now to pass queued data to user.
2485 if (tlen == 0 && (thflags & TH_FIN) == 0) {
2486 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2488 tcp_handle_wakeup(tp);
2490 tp->snd_wl1 = th->th_seq - 1;
2494 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2495 * ACKs. If the ack is in the range
2496 * tp->snd_una < th->th_ack <= tp->snd_max
2497 * then advance tp->snd_una to th->th_ack and drop
2498 * data from the retransmission queue. If this ACK reflects
2499 * more up to date window information we update our window information.
2501 case TCPS_ESTABLISHED:
2502 case TCPS_FIN_WAIT_1:
2503 case TCPS_FIN_WAIT_2:
2504 case TCPS_CLOSE_WAIT:
2507 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2508 TCPSTAT_INC(tcps_rcvacktoomuch);
2511 if (tcp_is_sack_recovery(tp, &to)) {
2512 if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) &&
2513 (tp->t_flags & TF_LRD)) {
2514 tcp_sack_lost_retransmission(tp, th);
2518 * Reset the value so that previous (valid) value
2519 * from the last ack with SACK doesn't get used.
2521 tp->sackhint.sacked_bytes = 0;
2524 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2525 hhook_run_tcp_est_in(tp, th, &to);
2528 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2529 maxseg = tcp_maxseg(tp);
2531 (tiwin == tp->snd_wnd ||
2532 (tp->t_flags & TF_SACK_PERMIT))) {
2534 * If this is the first time we've seen a
2535 * FIN from the remote, this is not a
2536 * duplicate and it needs to be processed
2537 * normally. This happens during a
2538 * simultaneous close.
2540 if ((thflags & TH_FIN) &&
2541 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2545 TCPSTAT_INC(tcps_rcvdupack);
2547 * If we have outstanding data (other than
2548 * a window probe), this is a completely
2549 * duplicate ack (ie, window info didn't
2550 * change and FIN isn't set),
2551 * the ack is the biggest we've
2552 * seen and we've seen exactly our rexmt
2553 * threshold of them, assume a packet
2554 * has been dropped and retransmit it.
2555 * Kludge snd_nxt & the congestion
2556 * window so we send only this one
2559 * We know we're losing at the current
2560 * window size so do congestion avoidance
2561 * (set ssthresh to half the current window
2562 * and pull our congestion window back to
2563 * the new ssthresh).
2565 * Dup acks mean that packets have left the
2566 * network (they're now cached at the receiver)
2567 * so bump cwnd by the amount in the receiver
2568 * to keep a constant cwnd packets in the
2571 * When using TCP ECN, notify the peer that
2572 * we reduced the cwnd.
2575 * Following 2 kinds of acks should not affect
2578 * 2) Acks with SACK but without any new SACK
2579 * information in them. These could result from
2580 * any anomaly in the network like a switch
2581 * duplicating packets or a possible DoS attack.
2583 if (th->th_ack != tp->snd_una ||
2584 (tcp_is_sack_recovery(tp, &to) &&
2587 else if (!tcp_timer_active(tp, TT_REXMT))
2589 else if (++tp->t_dupacks > tcprexmtthresh ||
2590 IN_FASTRECOVERY(tp->t_flags)) {
2591 cc_ack_received(tp, th, nsegs,
2594 IN_FASTRECOVERY(tp->t_flags)) {
2595 tcp_do_prr_ack(tp, th, &to);
2596 } else if (tcp_is_sack_recovery(tp, &to) &&
2597 IN_FASTRECOVERY(tp->t_flags)) {
2601 * Compute the amount of data in flight first.
2602 * We can inject new data into the pipe iff
2603 * we have less than 1/2 the original window's
2604 * worth of data in flight.
2606 if (V_tcp_do_newsack)
2607 awnd = tcp_compute_pipe(tp);
2609 awnd = (tp->snd_nxt - tp->snd_fack) +
2610 tp->sackhint.sack_bytes_rexmit;
2612 if (awnd < tp->snd_ssthresh) {
2613 tp->snd_cwnd += maxseg;
2614 if (tp->snd_cwnd > tp->snd_ssthresh)
2615 tp->snd_cwnd = tp->snd_ssthresh;
2618 tp->snd_cwnd += maxseg;
2619 (void) tcp_output(tp);
2621 } else if (tp->t_dupacks == tcprexmtthresh ||
2622 (tp->t_flags & TF_SACK_PERMIT &&
2624 tp->sackhint.sacked_bytes >
2625 (tcprexmtthresh - 1) * maxseg)) {
2628 * Above is the RFC6675 trigger condition of
2629 * more than (dupthresh-1)*maxseg sacked data.
2630 * If the count of holes in the
2631 * scoreboard is >= dupthresh, we could
2632 * also enter loss recovery, but don't
2633 * have that value readily available.
2635 tp->t_dupacks = tcprexmtthresh;
2636 tcp_seq onxt = tp->snd_nxt;
2639 * If we're doing sack, or prr, check
2640 * to see if we're already in sack
2641 * recovery. If we're not doing sack,
2642 * check to see if we're in newreno
2646 (tp->t_flags & TF_SACK_PERMIT)) {
2647 if (IN_FASTRECOVERY(tp->t_flags)) {
2652 if (SEQ_LEQ(th->th_ack,
2658 /* Congestion signal before ack. */
2659 cc_cong_signal(tp, th, CC_NDUPACK);
2660 cc_ack_received(tp, th, nsegs,
2662 tcp_timer_activate(tp, TT_REXMT, 0);
2666 * snd_ssthresh is already updated by
2669 if (tcp_is_sack_recovery(tp, &to)) {
2670 tp->sackhint.prr_delivered =
2671 tp->sackhint.sacked_bytes;
2673 tp->sackhint.prr_delivered =
2674 imin(tp->snd_max - tp->snd_una,
2675 imin(INT_MAX / 65536,
2676 tp->t_dupacks) * maxseg);
2678 tp->sackhint.recover_fs = max(1,
2679 tp->snd_nxt - tp->snd_una);
2681 if (tcp_is_sack_recovery(tp, &to)) {
2683 tcps_sack_recovery_episode);
2684 tp->snd_recover = tp->snd_nxt;
2685 tp->snd_cwnd = maxseg;
2686 (void) tcp_output(tp);
2687 if (SEQ_GT(th->th_ack, tp->snd_una))
2688 goto resume_partialack;
2691 tp->snd_nxt = th->th_ack;
2692 tp->snd_cwnd = maxseg;
2693 (void) tcp_output(tp);
2694 KASSERT(tp->snd_limited <= 2,
2695 ("%s: tp->snd_limited too big",
2697 tp->snd_cwnd = tp->snd_ssthresh +
2699 (tp->t_dupacks - tp->snd_limited);
2700 if (SEQ_GT(onxt, tp->snd_nxt))
2703 } else if (V_tcp_do_rfc3042) {
2705 * Process first and second duplicate
2706 * ACKs. Each indicates a segment
2707 * leaving the network, creating room
2708 * for more. Make sure we can send a
2709 * packet on reception of each duplicate
2710 * ACK by increasing snd_cwnd by one
2711 * segment. Restore the original
2712 * snd_cwnd after packet transmission.
2714 cc_ack_received(tp, th, nsegs,
2716 uint32_t oldcwnd = tp->snd_cwnd;
2717 tcp_seq oldsndmax = tp->snd_max;
2721 KASSERT(tp->t_dupacks == 1 ||
2723 ("%s: dupacks not 1 or 2",
2725 if (tp->t_dupacks == 1)
2726 tp->snd_limited = 0;
2728 (tp->snd_nxt - tp->snd_una) +
2729 (tp->t_dupacks - tp->snd_limited) *
2732 * Only call tcp_output when there
2733 * is new data available to be sent
2734 * or we need to send an ACK.
2736 SOCKBUF_LOCK(&so->so_snd);
2737 avail = sbavail(&so->so_snd) -
2738 (tp->snd_nxt - tp->snd_una);
2739 SOCKBUF_UNLOCK(&so->so_snd);
2740 if (avail > 0 || tp->t_flags & TF_ACKNOW)
2741 (void) tcp_output(tp);
2742 sent = tp->snd_max - oldsndmax;
2743 if (sent > maxseg) {
2744 KASSERT((tp->t_dupacks == 2 &&
2745 tp->snd_limited == 0) ||
2746 (sent == maxseg + 1 &&
2747 tp->t_flags & TF_SENTFIN),
2748 ("%s: sent too much",
2750 tp->snd_limited = 2;
2751 } else if (sent > 0)
2753 tp->snd_cwnd = oldcwnd;
2760 * This ack is advancing the left edge, reset the
2765 * If this ack also has new SACK info, increment the
2766 * counter as per rfc6675. The variable
2767 * sack_changed tracks all changes to the SACK
2768 * scoreboard, including when partial ACKs without
2769 * SACK options are received, and clear the scoreboard
2770 * from the left side. Such partial ACKs should not be
2771 * counted as dupacks here.
2773 if (tcp_is_sack_recovery(tp, &to) &&
2776 /* limit overhead by setting maxseg last */
2777 if (!IN_FASTRECOVERY(tp->t_flags) &&
2778 (tp->sackhint.sacked_bytes >
2779 ((tcprexmtthresh - 1) *
2780 (maxseg = tcp_maxseg(tp))))) {
2781 goto enter_recovery;
2787 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2788 ("%s: th_ack <= snd_una", __func__));
2791 * If the congestion window was inflated to account
2792 * for the other side's cached packets, retract it.
2794 if (IN_FASTRECOVERY(tp->t_flags)) {
2795 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2796 if (tp->t_flags & TF_SACK_PERMIT)
2797 if (V_tcp_do_prr && to.to_flags & TOF_SACK) {
2798 tcp_timer_activate(tp, TT_REXMT, 0);
2800 tcp_do_prr_ack(tp, th, &to);
2801 tp->t_flags |= TF_ACKNOW;
2802 (void) tcp_output(tp);
2804 tcp_sack_partialack(tp, th);
2806 tcp_newreno_partial_ack(tp, th);
2808 cc_post_recovery(tp, th);
2809 } else if (IN_CONGRECOVERY(tp->t_flags)) {
2810 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2812 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th);
2813 tp->snd_fack = th->th_ack;
2814 tcp_do_prr_ack(tp, th, &to);
2815 (void) tcp_output(tp);
2818 cc_post_recovery(tp, th);
2821 * If we reach this point, ACK is not a duplicate,
2822 * i.e., it ACKs something we sent.
2824 if (tp->t_flags & TF_NEEDSYN) {
2826 * T/TCP: Connection was half-synchronized, and our
2827 * SYN has been ACK'd (so connection is now fully
2828 * synchronized). Go to non-starred state,
2829 * increment snd_una for ACK of SYN, and check if
2830 * we can do window scaling.
2832 tp->t_flags &= ~TF_NEEDSYN;
2834 /* Do window scaling? */
2835 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2836 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2837 tp->rcv_scale = tp->request_r_scale;
2838 /* Send window already scaled. */
2843 INP_WLOCK_ASSERT(inp);
2846 * Adjust for the SYN bit in sequence space,
2847 * but don't account for it in cwnd calculations.
2848 * This is for the SYN_RECEIVED, non-simultaneous
2849 * SYN case. SYN_SENT and simultaneous SYN are
2850 * treated elsewhere.
2854 acked = BYTES_THIS_ACK(tp, th);
2855 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2856 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2857 tp->snd_una, th->th_ack, tp, m));
2858 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2859 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2862 * If we just performed our first retransmit, and the ACK
2863 * arrives within our recovery window, then it was a mistake
2864 * to do the retransmit in the first place. Recover our
2865 * original cwnd and ssthresh, and proceed to transmit where
2868 if (tp->t_rxtshift == 1 &&
2869 tp->t_flags & TF_PREVVALID &&
2870 tp->t_badrxtwin != 0 &&
2871 to.to_flags & TOF_TS &&
2873 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
2874 cc_cong_signal(tp, th, CC_RTO_ERR);
2877 * If we have a timestamp reply, update smoothed
2878 * round trip time. If no timestamp is present but
2879 * transmit timer is running and timed sequence
2880 * number was acked, update smoothed round trip time.
2881 * Since we now have an rtt measurement, cancel the
2882 * timer backoff (cf., Phil Karn's retransmit alg.).
2883 * Recompute the initial retransmit timer.
2885 * Some boxes send broken timestamp replies
2886 * during the SYN+ACK phase, ignore
2887 * timestamps of 0 or we could calculate a
2888 * huge RTT and blow up the retransmit timer.
2890 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2893 t = tcp_ts_getticks() - to.to_tsecr;
2894 if (!tp->t_rttlow || tp->t_rttlow > t)
2896 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2897 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2898 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2899 tp->t_rttlow = ticks - tp->t_rtttime;
2900 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2903 SOCKBUF_LOCK(&so->so_snd);
2905 * Clear t_acktime if remote side has ACKd all data in the
2906 * socket buffer and FIN (if applicable).
2907 * Otherwise, update t_acktime if we received a sufficiently
2910 if ((tp->t_state <= TCPS_CLOSE_WAIT &&
2911 acked == sbavail(&so->so_snd)) ||
2912 acked > sbavail(&so->so_snd))
2915 tp->t_acktime = ticks;
2918 * If all outstanding data is acked, stop retransmit
2919 * timer and remember to restart (more output or persist).
2920 * If there is more data to be acked, restart retransmit
2921 * timer, using current (possibly backed-off) value.
2923 if (th->th_ack == tp->snd_max) {
2924 tcp_timer_activate(tp, TT_REXMT, 0);
2926 } else if (!tcp_timer_active(tp, TT_PERSIST))
2927 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp));
2930 * If no data (only SYN) was ACK'd,
2931 * skip rest of ACK processing.
2934 SOCKBUF_UNLOCK(&so->so_snd);
2939 * Let the congestion control algorithm update congestion
2940 * control related information. This typically means increasing
2941 * the congestion window.
2943 cc_ack_received(tp, th, nsegs, CC_ACK);
2945 if (acked > sbavail(&so->so_snd)) {
2946 if (tp->snd_wnd >= sbavail(&so->so_snd))
2947 tp->snd_wnd -= sbavail(&so->so_snd);
2950 mfree = sbcut_locked(&so->so_snd,
2951 (int)sbavail(&so->so_snd));
2954 mfree = sbcut_locked(&so->so_snd, acked);
2955 if (tp->snd_wnd >= (uint32_t) acked)
2956 tp->snd_wnd -= acked;
2961 /* NB: sowwakeup_locked() does an implicit unlock. */
2962 sowwakeup_locked(so);
2964 /* Detect una wraparound. */
2965 if (!IN_RECOVERY(tp->t_flags) &&
2966 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2967 SEQ_LEQ(th->th_ack, tp->snd_recover))
2968 tp->snd_recover = th->th_ack - 1;
2969 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2970 if (IN_RECOVERY(tp->t_flags) &&
2971 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2972 EXIT_RECOVERY(tp->t_flags);
2974 tp->snd_una = th->th_ack;
2975 if (tp->t_flags & TF_SACK_PERMIT) {
2976 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2977 tp->snd_recover = tp->snd_una;
2979 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2980 tp->snd_nxt = tp->snd_una;
2982 switch (tp->t_state) {
2984 * In FIN_WAIT_1 STATE in addition to the processing
2985 * for the ESTABLISHED state if our FIN is now acknowledged
2986 * then enter FIN_WAIT_2.
2988 case TCPS_FIN_WAIT_1:
2989 if (ourfinisacked) {
2991 * If we can't receive any more
2992 * data, then closing user can proceed.
2993 * Starting the timer is contrary to the
2994 * specification, but if we don't get a FIN
2995 * we'll hang forever.
2997 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2998 soisdisconnected(so);
2999 tcp_timer_activate(tp, TT_2MSL,
3000 (tcp_fast_finwait2_recycle ?
3001 tcp_finwait2_timeout :
3004 tcp_state_change(tp, TCPS_FIN_WAIT_2);
3009 * In CLOSING STATE in addition to the processing for
3010 * the ESTABLISHED state if the ACK acknowledges our FIN
3011 * then enter the TIME-WAIT state, otherwise ignore
3015 if (ourfinisacked) {
3023 * In LAST_ACK, we may still be waiting for data to drain
3024 * and/or to be acked, as well as for the ack of our FIN.
3025 * If our FIN is now acknowledged, delete the TCB,
3026 * enter the closed state and return.
3029 if (ourfinisacked) {
3038 INP_WLOCK_ASSERT(inp);
3041 * Update window information.
3042 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3044 if ((thflags & TH_ACK) &&
3045 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3046 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3047 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3048 /* keep track of pure window updates */
3050 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3051 TCPSTAT_INC(tcps_rcvwinupd);
3052 tp->snd_wnd = tiwin;
3053 tp->snd_wl1 = th->th_seq;
3054 tp->snd_wl2 = th->th_ack;
3055 if (tp->snd_wnd > tp->max_sndwnd)
3056 tp->max_sndwnd = tp->snd_wnd;
3061 * Process segments with URG.
3063 if ((thflags & TH_URG) && th->th_urp &&
3064 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3066 * This is a kludge, but if we receive and accept
3067 * random urgent pointers, we'll crash in
3068 * soreceive. It's hard to imagine someone
3069 * actually wanting to send this much urgent data.
3071 SOCKBUF_LOCK(&so->so_rcv);
3072 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3073 th->th_urp = 0; /* XXX */
3074 thflags &= ~TH_URG; /* XXX */
3075 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3076 goto dodata; /* XXX */
3079 * If this segment advances the known urgent pointer,
3080 * then mark the data stream. This should not happen
3081 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3082 * a FIN has been received from the remote side.
3083 * In these states we ignore the URG.
3085 * According to RFC961 (Assigned Protocols),
3086 * the urgent pointer points to the last octet
3087 * of urgent data. We continue, however,
3088 * to consider it to indicate the first octet
3089 * of data past the urgent section as the original
3090 * spec states (in one of two places).
3092 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3093 tp->rcv_up = th->th_seq + th->th_urp;
3094 so->so_oobmark = sbavail(&so->so_rcv) +
3095 (tp->rcv_up - tp->rcv_nxt) - 1;
3096 if (so->so_oobmark == 0)
3097 so->so_rcv.sb_state |= SBS_RCVATMARK;
3099 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3101 SOCKBUF_UNLOCK(&so->so_rcv);
3103 * Remove out of band data so doesn't get presented to user.
3104 * This can happen independent of advancing the URG pointer,
3105 * but if two URG's are pending at once, some out-of-band
3106 * data may creep in... ick.
3108 if (th->th_urp <= (uint32_t)tlen &&
3109 !(so->so_options & SO_OOBINLINE)) {
3110 /* hdr drop is delayed */
3111 tcp_pulloutofband(so, th, m, drop_hdrlen);
3115 * If no out of band data is expected,
3116 * pull receive urgent pointer along
3117 * with the receive window.
3119 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3120 tp->rcv_up = tp->rcv_nxt;
3123 INP_WLOCK_ASSERT(inp);
3126 * Process the segment text, merging it into the TCP sequencing queue,
3127 * and arranging for acknowledgment of receipt if necessary.
3128 * This process logically involves adjusting tp->rcv_wnd as data
3129 * is presented to the user (this happens in tcp_usrreq.c,
3130 * case PRU_RCVD). If a FIN has already been received on this
3131 * connection then we just ignore the text.
3133 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3134 IS_FASTOPEN(tp->t_flags));
3135 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3136 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3137 tcp_seq save_start = th->th_seq;
3138 tcp_seq save_rnxt = tp->rcv_nxt;
3139 int save_tlen = tlen;
3140 m_adj(m, drop_hdrlen); /* delayed header drop */
3142 * Insert segment which includes th into TCP reassembly queue
3143 * with control block tp. Set thflags to whether reassembly now
3144 * includes a segment with FIN. This handles the common case
3145 * inline (segment is the next to be received on an established
3146 * connection, and the queue is empty), avoiding linkage into
3147 * and removal from the queue and repetition of various
3149 * Set DELACK for segments received in order, but ack
3150 * immediately when segments are out of order (so
3151 * fast retransmit can work).
3153 if (th->th_seq == tp->rcv_nxt &&
3155 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3157 if (DELAY_ACK(tp, tlen) || tfo_syn)
3158 tp->t_flags |= TF_DELACK;
3160 tp->t_flags |= TF_ACKNOW;
3161 tp->rcv_nxt += tlen;
3163 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3164 (tp->t_fbyte_in == 0)) {
3165 tp->t_fbyte_in = ticks;
3166 if (tp->t_fbyte_in == 0)
3168 if (tp->t_fbyte_out && tp->t_fbyte_in)
3169 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3171 thflags = tcp_get_flags(th) & TH_FIN;
3172 TCPSTAT_INC(tcps_rcvpack);
3173 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3174 SOCKBUF_LOCK(&so->so_rcv);
3175 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3178 sbappendstream_locked(&so->so_rcv, m, 0);
3179 tp->t_flags |= TF_WAKESOR;
3182 * XXX: Due to the header drop above "th" is
3183 * theoretically invalid by now. Fortunately
3184 * m_adj() doesn't actually frees any mbufs
3185 * when trimming from the head.
3187 tcp_seq temp = save_start;
3189 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3190 tp->t_flags |= TF_ACKNOW;
3192 if ((tp->t_flags & TF_SACK_PERMIT) &&
3194 TCPS_HAVEESTABLISHED(tp->t_state)) {
3195 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3197 * DSACK actually handled in the fastpath
3200 tcp_update_sack_list(tp, save_start,
3201 save_start + save_tlen);
3202 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3203 if ((tp->rcv_numsacks >= 1) &&
3204 (tp->sackblks[0].end == save_start)) {
3206 * Partial overlap, recorded at todrop
3209 tcp_update_sack_list(tp,
3210 tp->sackblks[0].start,
3211 tp->sackblks[0].end);
3213 tcp_update_dsack_list(tp, save_start,
3214 save_start + save_tlen);
3216 } else if (tlen >= save_tlen) {
3217 /* Update of sackblks. */
3218 tcp_update_dsack_list(tp, save_start,
3219 save_start + save_tlen);
3220 } else if (tlen > 0) {
3221 tcp_update_dsack_list(tp, save_start,
3225 tcp_handle_wakeup(tp);
3228 * Note the amount of data that peer has sent into
3229 * our window, in order to estimate the sender's
3233 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3234 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3236 len = so->so_rcv.sb_hiwat;
3244 * If FIN is received ACK the FIN and let the user know
3245 * that the connection is closing.
3247 if (thflags & TH_FIN) {
3248 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3249 /* The socket upcall is handled by socantrcvmore. */
3252 * If connection is half-synchronized
3253 * (ie NEEDSYN flag on) then delay ACK,
3254 * so it may be piggybacked when SYN is sent.
3255 * Otherwise, since we received a FIN then no
3256 * more input can be expected, send ACK now.
3258 if (tp->t_flags & TF_NEEDSYN)
3259 tp->t_flags |= TF_DELACK;
3261 tp->t_flags |= TF_ACKNOW;
3264 switch (tp->t_state) {
3266 * In SYN_RECEIVED and ESTABLISHED STATES
3267 * enter the CLOSE_WAIT state.
3269 case TCPS_SYN_RECEIVED:
3270 tp->t_starttime = ticks;
3272 case TCPS_ESTABLISHED:
3273 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3277 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3278 * enter the CLOSING state.
3280 case TCPS_FIN_WAIT_1:
3281 tcp_state_change(tp, TCPS_CLOSING);
3285 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3286 * starting the time-wait timer, turning off the other
3289 case TCPS_FIN_WAIT_2:
3294 TCP_PROBE3(debug__input, tp, th, m);
3297 * Return any desired output.
3299 if (needoutput || (tp->t_flags & TF_ACKNOW))
3300 (void) tcp_output(tp);
3303 INP_WLOCK_ASSERT(inp);
3305 if (tp->t_flags & TF_DELACK) {
3306 tp->t_flags &= ~TF_DELACK;
3307 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3314 * Generate an ACK dropping incoming segment if it occupies
3315 * sequence space, where the ACK reflects our state.
3317 * We can now skip the test for the RST flag since all
3318 * paths to this code happen after packets containing
3319 * RST have been dropped.
3321 * In the SYN-RECEIVED state, don't send an ACK unless the
3322 * segment we received passes the SYN-RECEIVED ACK test.
3323 * If it fails send a RST. This breaks the loop in the
3324 * "LAND" DoS attack, and also prevents an ACK storm
3325 * between two listening ports that have been sent forged
3326 * SYN segments, each with the source address of the other.
3328 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3329 (SEQ_GT(tp->snd_una, th->th_ack) ||
3330 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3331 rstreason = BANDLIM_RST_OPENPORT;
3332 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
3335 TCP_PROBE3(debug__input, tp, th, m);
3336 tp->t_flags |= TF_ACKNOW;
3337 (void) tcp_output(tp);
3344 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3347 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3352 * Drop space held by incoming segment and return.
3354 TCP_PROBE3(debug__input, tp, th, m);
3362 * Issue RST and make ACK acceptable to originator of segment.
3363 * The mbuf must still include the original packet header.
3367 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3368 int tlen, int rstreason)
3374 struct ip6_hdr *ip6;
3378 INP_LOCK_ASSERT(tptoinpcb(tp));
3381 /* Don't bother if destination was broadcast/multicast. */
3382 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3385 if (mtod(m, struct ip *)->ip_v == 6) {
3386 ip6 = mtod(m, struct ip6_hdr *);
3387 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3388 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3390 /* IPv6 anycast check is done at tcp6_input() */
3393 #if defined(INET) && defined(INET6)
3398 ip = mtod(m, struct ip *);
3399 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3400 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3401 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3402 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3407 /* Perform bandwidth limiting. */
3408 if (badport_bandlim(rstreason) < 0)
3411 /* tcp_respond consumes the mbuf chain. */
3412 if (tcp_get_flags(th) & TH_ACK) {
3413 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3414 th->th_ack, TH_RST);
3416 if (tcp_get_flags(th) & TH_SYN)
3418 if (tcp_get_flags(th) & TH_FIN)
3420 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3421 (tcp_seq)0, TH_RST|TH_ACK);
3429 * Parse TCP options and place in tcpopt.
3432 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3437 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3439 if (opt == TCPOPT_EOL)
3441 if (opt == TCPOPT_NOP)
3447 if (optlen < 2 || optlen > cnt)
3452 if (optlen != TCPOLEN_MAXSEG)
3454 if (!(flags & TO_SYN))
3456 to->to_flags |= TOF_MSS;
3457 bcopy((char *)cp + 2,
3458 (char *)&to->to_mss, sizeof(to->to_mss));
3459 to->to_mss = ntohs(to->to_mss);
3462 if (optlen != TCPOLEN_WINDOW)
3464 if (!(flags & TO_SYN))
3466 to->to_flags |= TOF_SCALE;
3467 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3469 case TCPOPT_TIMESTAMP:
3470 if (optlen != TCPOLEN_TIMESTAMP)
3472 to->to_flags |= TOF_TS;
3473 bcopy((char *)cp + 2,
3474 (char *)&to->to_tsval, sizeof(to->to_tsval));
3475 to->to_tsval = ntohl(to->to_tsval);
3476 bcopy((char *)cp + 6,
3477 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3478 to->to_tsecr = ntohl(to->to_tsecr);
3480 case TCPOPT_SIGNATURE:
3482 * In order to reply to a host which has set the
3483 * TCP_SIGNATURE option in its initial SYN, we have
3484 * to record the fact that the option was observed
3485 * here for the syncache code to perform the correct
3488 if (optlen != TCPOLEN_SIGNATURE)
3490 to->to_flags |= TOF_SIGNATURE;
3491 to->to_signature = cp + 2;
3493 case TCPOPT_SACK_PERMITTED:
3494 if (optlen != TCPOLEN_SACK_PERMITTED)
3496 if (!(flags & TO_SYN))
3500 to->to_flags |= TOF_SACKPERM;
3503 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3507 to->to_flags |= TOF_SACK;
3508 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3509 to->to_sacks = cp + 2;
3510 TCPSTAT_INC(tcps_sack_rcv_blocks);
3512 case TCPOPT_FAST_OPEN:
3514 * Cookie length validation is performed by the
3515 * server side cookie checking code or the client
3516 * side cookie cache update code.
3518 if (!(flags & TO_SYN))
3520 if (!V_tcp_fastopen_client_enable &&
3521 !V_tcp_fastopen_server_enable)
3523 to->to_flags |= TOF_FASTOPEN;
3524 to->to_tfo_len = optlen - 2;
3525 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3534 * Pull out of band byte out of a segment so
3535 * it doesn't appear in the user's data queue.
3536 * It is still reflected in the segment length for
3537 * sequencing purposes.
3540 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3543 int cnt = off + th->th_urp - 1;
3546 if (m->m_len > cnt) {
3547 char *cp = mtod(m, caddr_t) + cnt;
3548 struct tcpcb *tp = sototcpcb(so);
3550 INP_WLOCK_ASSERT(tptoinpcb(tp));
3553 tp->t_oobflags |= TCPOOB_HAVEDATA;
3554 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3556 if (m->m_flags & M_PKTHDR)
3565 panic("tcp_pulloutofband");
3569 * Collect new round-trip time estimate
3570 * and update averages and current timeout.
3573 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3577 INP_WLOCK_ASSERT(tptoinpcb(tp));
3579 TCPSTAT_INC(tcps_rttupdated);
3580 if (tp->t_rttupdated < UCHAR_MAX)
3583 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3584 imax(0, rtt * 1000 / hz));
3586 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3588 * srtt is stored as fixed point with 5 bits after the
3589 * binary point (i.e., scaled by 8). The following magic
3590 * is equivalent to the smoothing algorithm in rfc793 with
3591 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3592 * point). Adjust rtt to origin 0.
3594 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3595 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3597 if ((tp->t_srtt += delta) <= 0)
3601 * We accumulate a smoothed rtt variance (actually, a
3602 * smoothed mean difference), then set the retransmit
3603 * timer to smoothed rtt + 4 times the smoothed variance.
3604 * rttvar is stored as fixed point with 4 bits after the
3605 * binary point (scaled by 16). The following is
3606 * equivalent to rfc793 smoothing with an alpha of .75
3607 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3608 * rfc793's wired-in beta.
3612 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3613 if ((tp->t_rttvar += delta) <= 0)
3617 * No rtt measurement yet - use the unsmoothed rtt.
3618 * Set the variance to half the rtt (so our first
3619 * retransmit happens at 3*rtt).
3621 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3622 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3628 * the retransmit should happen at rtt + 4 * rttvar.
3629 * Because of the way we do the smoothing, srtt and rttvar
3630 * will each average +1/2 tick of bias. When we compute
3631 * the retransmit timer, we want 1/2 tick of rounding and
3632 * 1 extra tick because of +-1/2 tick uncertainty in the
3633 * firing of the timer. The bias will give us exactly the
3634 * 1.5 tick we need. But, because the bias is
3635 * statistical, we have to test that we don't drop below
3636 * the minimum feasible timer (which is 2 ticks).
3638 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3639 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3642 * We received an ack for a packet that wasn't retransmitted;
3643 * it is probably safe to discard any error indications we've
3644 * received recently. This isn't quite right, but close enough
3645 * for now (a route might have failed after we sent a segment,
3646 * and the return path might not be symmetrical).
3648 tp->t_softerror = 0;
3652 * Determine a reasonable value for maxseg size.
3653 * If the route is known, check route for mtu.
3654 * If none, use an mss that can be handled on the outgoing interface
3655 * without forcing IP to fragment. If no route is found, route has no mtu,
3656 * or the destination isn't local, use a default, hopefully conservative
3657 * size (usually 512 or the default IP max size, but no more than the mtu
3658 * of the interface), as we can't discover anything about intervening
3659 * gateways or networks. We also initialize the congestion/slow start
3660 * window to be a single segment if the destination isn't local.
3661 * While looking at the routing entry, we also initialize other path-dependent
3662 * parameters from pre-set or cached values in the routing entry.
3664 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3665 * IP options, e.g. IPSEC data, since length of this data may vary, and
3666 * thus it is calculated for every segment separately in tcp_output().
3668 * NOTE that this routine is only called when we process an incoming
3669 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3670 * settings are handled in tcp_mssopt().
3673 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3674 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3677 uint32_t maxmtu = 0;
3678 struct inpcb *inp = tptoinpcb(tp);
3679 struct hc_metrics_lite metrics;
3681 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3682 size_t min_protoh = isipv6 ?
3683 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3684 sizeof (struct tcpiphdr);
3686 size_t min_protoh = sizeof(struct tcpiphdr);
3689 INP_WLOCK_ASSERT(inp);
3692 min_protoh += V_tcp_udp_tunneling_overhead;
3693 if (mtuoffer != -1) {
3694 KASSERT(offer == -1, ("%s: conflict", __func__));
3695 offer = mtuoffer - min_protoh;
3701 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3702 tp->t_maxseg = V_tcp_v6mssdflt;
3705 #if defined(INET) && defined(INET6)
3710 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3711 tp->t_maxseg = V_tcp_mssdflt;
3716 * No route to sender, stay with default mss and return.
3720 * In case we return early we need to initialize metrics
3721 * to a defined state as tcp_hc_get() would do for us
3722 * if there was no cache hit.
3724 if (metricptr != NULL)
3725 bzero(metricptr, sizeof(struct hc_metrics_lite));
3729 /* What have we got? */
3733 * Offer == 0 means that there was no MSS on the SYN
3734 * segment, in this case we use tcp_mssdflt as
3735 * already assigned to t_maxseg above.
3737 offer = tp->t_maxseg;
3742 * Offer == -1 means that we didn't receive SYN yet.
3748 * Prevent DoS attack with too small MSS. Round up
3749 * to at least minmss.
3751 offer = max(offer, V_tcp_minmss);
3755 * rmx information is now retrieved from tcp_hostcache.
3757 tcp_hc_get(&inp->inp_inc, &metrics);
3758 if (metricptr != NULL)
3759 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3762 * If there's a discovered mtu in tcp hostcache, use it.
3763 * Else, use the link mtu.
3765 if (metrics.rmx_mtu)
3766 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3770 mss = maxmtu - min_protoh;
3771 if (!V_path_mtu_discovery &&
3772 !in6_localaddr(&inp->in6p_faddr))
3773 mss = min(mss, V_tcp_v6mssdflt);
3776 #if defined(INET) && defined(INET6)
3781 mss = maxmtu - min_protoh;
3782 if (!V_path_mtu_discovery &&
3783 !in_localaddr(inp->inp_faddr))
3784 mss = min(mss, V_tcp_mssdflt);
3788 * XXX - The above conditional (mss = maxmtu - min_protoh)
3789 * probably violates the TCP spec.
3790 * The problem is that, since we don't know the
3791 * other end's MSS, we are supposed to use a conservative
3792 * default. But, if we do that, then MTU discovery will
3793 * never actually take place, because the conservative
3794 * default is much less than the MTUs typically seen
3795 * on the Internet today. For the moment, we'll sweep
3796 * this under the carpet.
3798 * The conservative default might not actually be a problem
3799 * if the only case this occurs is when sending an initial
3800 * SYN with options and data to a host we've never talked
3801 * to before. Then, they will reply with an MSS value which
3802 * will get recorded and the new parameters should get
3803 * recomputed. For Further Study.
3806 mss = min(mss, offer);
3809 * Sanity check: make sure that maxseg will be large
3810 * enough to allow some data on segments even if the
3811 * all the option space is used (40bytes). Otherwise
3812 * funny things may happen in tcp_output.
3814 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3822 tcp_mss(struct tcpcb *tp, int offer)
3826 struct inpcb *inp = tptoinpcb(tp);
3828 struct hc_metrics_lite metrics;
3829 struct tcp_ifcap cap;
3831 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3833 bzero(&cap, sizeof(cap));
3834 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3839 * If there's a pipesize, change the socket buffer to that size,
3840 * don't change if sb_hiwat is different than default (then it
3841 * has been changed on purpose with setsockopt).
3842 * Make the socket buffers an integral number of mss units;
3843 * if the mss is larger than the socket buffer, decrease the mss.
3845 so = inp->inp_socket;
3846 SOCKBUF_LOCK(&so->so_snd);
3847 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3848 bufsize = metrics.rmx_sendpipe;
3850 bufsize = so->so_snd.sb_hiwat;
3854 bufsize = roundup(bufsize, mss);
3855 if (bufsize > sb_max)
3857 if (bufsize > so->so_snd.sb_hiwat)
3858 (void)sbreserve_locked(so, SO_SND, bufsize, NULL);
3860 SOCKBUF_UNLOCK(&so->so_snd);
3862 * Sanity check: make sure that maxseg will be large
3863 * enough to allow some data on segments even if the
3864 * all the option space is used (40bytes). Otherwise
3865 * funny things may happen in tcp_output.
3867 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3869 tp->t_maxseg = max(mss, 64);
3871 SOCKBUF_LOCK(&so->so_rcv);
3872 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3873 bufsize = metrics.rmx_recvpipe;
3875 bufsize = so->so_rcv.sb_hiwat;
3876 if (bufsize > mss) {
3877 bufsize = roundup(bufsize, mss);
3878 if (bufsize > sb_max)
3880 if (bufsize > so->so_rcv.sb_hiwat)
3881 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL);
3883 SOCKBUF_UNLOCK(&so->so_rcv);
3885 /* Check the interface for TSO capabilities. */
3886 if (cap.ifcap & CSUM_TSO) {
3887 tp->t_flags |= TF_TSO;
3888 tp->t_tsomax = cap.tsomax;
3889 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3890 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3895 * Determine the MSS option to send on an outgoing SYN.
3898 tcp_mssopt(struct in_conninfo *inc)
3901 uint32_t thcmtu = 0;
3902 uint32_t maxmtu = 0;
3905 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3908 if (inc->inc_flags & INC_ISIPV6) {
3909 mss = V_tcp_v6mssdflt;
3910 maxmtu = tcp_maxmtu6(inc, NULL);
3911 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3914 #if defined(INET) && defined(INET6)
3919 mss = V_tcp_mssdflt;
3920 maxmtu = tcp_maxmtu(inc, NULL);
3921 min_protoh = sizeof(struct tcpiphdr);
3924 #if defined(INET6) || defined(INET)
3925 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3928 if (maxmtu && thcmtu)
3929 mss = min(maxmtu, thcmtu) - min_protoh;
3930 else if (maxmtu || thcmtu)
3931 mss = max(maxmtu, thcmtu) - min_protoh;
3937 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
3939 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
3940 int maxseg = tcp_maxseg(tp);
3942 INP_WLOCK_ASSERT(tptoinpcb(tp));
3945 * Compute the amount of data that this ACK is indicating
3946 * (del_data) and an estimate of how many bytes are in the
3949 if (tcp_is_sack_recovery(tp, to) ||
3950 (IN_CONGRECOVERY(tp->t_flags) &&
3951 !IN_FASTRECOVERY(tp->t_flags))) {
3952 del_data = tp->sackhint.delivered_data;
3953 if (V_tcp_do_newsack)
3954 pipe = tcp_compute_pipe(tp);
3956 pipe = (tp->snd_nxt - tp->snd_fack) +
3957 tp->sackhint.sack_bytes_rexmit;
3959 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg +
3960 tp->snd_recover - tp->snd_una))
3962 pipe = imax(0, tp->snd_max - tp->snd_una -
3963 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg);
3965 tp->sackhint.prr_delivered += del_data;
3967 * Proportional Rate Reduction
3969 if (pipe >= tp->snd_ssthresh) {
3970 if (tp->sackhint.recover_fs == 0)
3971 tp->sackhint.recover_fs =
3972 imax(1, tp->snd_nxt - tp->snd_una);
3973 snd_cnt = howmany((long)tp->sackhint.prr_delivered *
3974 tp->snd_ssthresh, tp->sackhint.recover_fs) -
3975 tp->sackhint.prr_out;
3977 if (V_tcp_do_prr_conservative || (del_data == 0))
3978 limit = tp->sackhint.prr_delivered -
3979 tp->sackhint.prr_out;
3981 limit = imax(tp->sackhint.prr_delivered -
3982 tp->sackhint.prr_out, del_data) +
3984 snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
3986 snd_cnt = imax(snd_cnt, 0) / maxseg;
3988 * Send snd_cnt new data into the network in response to this ack.
3989 * If there is going to be a SACK retransmission, adjust snd_cwnd
3992 if (IN_FASTRECOVERY(tp->t_flags)) {
3993 if (tcp_is_sack_recovery(tp, to)) {
3994 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
3995 tp->sackhint.sack_bytes_rexmit +
3998 tp->snd_cwnd = (tp->snd_max - tp->snd_una) +
4001 } else if (IN_CONGRECOVERY(tp->t_flags))
4002 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg);
4003 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd);
4007 * On a partial ack arrives, force the retransmission of the
4008 * next unacknowledged segment. Do not clear tp->t_dupacks.
4009 * By setting snd_nxt to ti_ack, this forces retransmission timer to
4013 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4015 tcp_seq onxt = tp->snd_nxt;
4016 uint32_t ocwnd = tp->snd_cwnd;
4017 u_int maxseg = tcp_maxseg(tp);
4019 INP_WLOCK_ASSERT(tptoinpcb(tp));
4021 tcp_timer_activate(tp, TT_REXMT, 0);
4023 tp->snd_nxt = th->th_ack;
4025 * Set snd_cwnd to one segment beyond acknowledged offset.
4026 * (tp->snd_una has not yet been updated when this function is called.)
4028 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4029 tp->t_flags |= TF_ACKNOW;
4030 (void) tcp_output(tp);
4031 tp->snd_cwnd = ocwnd;
4032 if (SEQ_GT(onxt, tp->snd_nxt))
4035 * Partial window deflation. Relies on fact that tp->snd_una
4038 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4039 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4042 tp->snd_cwnd += maxseg;
4046 tcp_compute_pipe(struct tcpcb *tp)
4048 if (tp->t_fb->tfb_compute_pipe == NULL) {
4049 return (tp->snd_max - tp->snd_una +
4050 tp->sackhint.sack_bytes_rexmit -
4051 tp->sackhint.sacked_bytes);
4053 return((*tp->t_fb->tfb_compute_pipe)(tp));
4058 tcp_compute_initwnd(uint32_t maxseg)
4061 * Calculate the Initial Window, also used as Restart Window
4063 * RFC5681 Section 3.1 specifies the default conservative values.
4064 * RFC3390 specifies slightly more aggressive values.
4065 * RFC6928 increases it to ten segments.
4066 * Support for user specified value for initial flight size.
4068 if (V_tcp_initcwnd_segments)
4069 return min(V_tcp_initcwnd_segments * maxseg,
4070 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4071 else if (V_tcp_do_rfc3390)
4072 return min(4 * maxseg, max(2 * maxseg, 4380));
4074 /* Per RFC5681 Section 3.1 */
4076 return (2 * maxseg);
4077 else if (maxseg > 1095)
4078 return (3 * maxseg);
4080 return (4 * maxseg);