2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
5 * The Regents of the University of California. All rights reserved.
6 * Copyright (c) 2007-2008,2010
7 * Swinburne University of Technology, Melbourne, Australia.
8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
9 * Copyright (c) 2010 The FreeBSD Foundation
10 * Copyright (c) 2010-2011 Juniper Networks, Inc.
11 * All rights reserved.
13 * Portions of this software were developed at the Centre for Advanced Internet
14 * Architectures, Swinburne University of Technology, by Lawrence Stewart,
15 * James Healy and David Hayes, made possible in part by a grant from the Cisco
16 * University Research Program Fund at Community Foundation Silicon Valley.
18 * Portions of this software were developed at the Centre for Advanced
19 * Internet Architectures, Swinburne University of Technology, Melbourne,
20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
22 * Portions of this software were developed by Robert N. M. Watson under
23 * contract to Juniper Networks, Inc.
25 * Redistribution and use in source and binary forms, with or without
26 * modification, are permitted provided that the following conditions
28 * 1. Redistributions of source code must retain the above copyright
29 * notice, this list of conditions and the following disclaimer.
30 * 2. Redistributions in binary form must reproduce the above copyright
31 * notice, this list of conditions and the following disclaimer in the
32 * documentation and/or other materials provided with the distribution.
33 * 3. Neither the name of the University nor the names of its contributors
34 * may be used to endorse or promote products derived from this software
35 * without specific prior written permission.
37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
49 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
52 #include <sys/cdefs.h>
53 __FBSDID("$FreeBSD$");
56 #include "opt_inet6.h"
57 #include "opt_ipsec.h"
58 #include "opt_tcpdebug.h"
60 #include <sys/param.h>
62 #include <sys/kernel.h>
64 #include <sys/hhook.h>
66 #include <sys/malloc.h>
68 #include <sys/proc.h> /* for proc0 declaration */
69 #include <sys/protosw.h>
70 #include <sys/qmath.h>
72 #include <sys/signalvar.h>
73 #include <sys/socket.h>
74 #include <sys/socketvar.h>
75 #include <sys/sysctl.h>
76 #include <sys/syslog.h>
77 #include <sys/systm.h>
78 #include <sys/stats.h>
80 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
85 #include <net/if_var.h>
86 #include <net/route.h>
89 #define TCPSTATES /* for logging */
91 #include <netinet/in.h>
92 #include <netinet/in_kdtrace.h>
93 #include <netinet/in_pcb.h>
94 #include <netinet/in_systm.h>
95 #include <netinet/ip.h>
96 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
97 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
98 #include <netinet/ip_var.h>
99 #include <netinet/ip_options.h>
100 #include <netinet/ip6.h>
101 #include <netinet/icmp6.h>
102 #include <netinet6/in6_pcb.h>
103 #include <netinet6/in6_var.h>
104 #include <netinet6/ip6_var.h>
105 #include <netinet6/nd6.h>
106 #include <netinet/tcp.h>
107 #include <netinet/tcp_fsm.h>
108 #include <netinet/tcp_log_buf.h>
109 #include <netinet/tcp_seq.h>
110 #include <netinet/tcp_timer.h>
111 #include <netinet/tcp_var.h>
112 #include <netinet6/tcp6_var.h>
113 #include <netinet/tcpip.h>
114 #include <netinet/cc/cc.h>
115 #include <netinet/tcp_fastopen.h>
117 #include <netinet/tcp_pcap.h>
119 #include <netinet/tcp_syncache.h>
121 #include <netinet/tcp_debug.h>
122 #endif /* TCPDEBUG */
124 #include <netinet/tcp_offload.h>
127 #include <netipsec/ipsec_support.h>
129 #include <machine/in_cksum.h>
131 #include <security/mac/mac_framework.h>
133 const int tcprexmtthresh = 3;
135 VNET_DEFINE(int, tcp_log_in_vain) = 0;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
137 &VNET_NAME(tcp_log_in_vain), 0,
138 "Log all incoming TCP segments to closed ports");
140 VNET_DEFINE(int, blackhole) = 0;
141 #define V_blackhole VNET(blackhole)
142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
143 &VNET_NAME(blackhole), 0,
144 "Do not send RST on segments to closed ports");
146 VNET_DEFINE(int, tcp_delack_enabled) = 1;
147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
148 &VNET_NAME(tcp_delack_enabled), 0,
149 "Delay ACK to try and piggyback it onto a data packet");
151 VNET_DEFINE(int, drop_synfin) = 0;
152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
153 &VNET_NAME(drop_synfin), 0,
154 "Drop TCP packets with SYN+FIN set");
156 VNET_DEFINE(int, tcp_do_newcwv) = 0;
157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
158 &VNET_NAME(tcp_do_newcwv), 0,
159 "Enable New Congestion Window Validation per RFC7661");
161 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
163 &VNET_NAME(tcp_do_rfc6675_pipe), 0,
164 "Use calculated pipe/in-flight bytes per RFC 6675");
166 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
168 &VNET_NAME(tcp_do_rfc3042), 0,
169 "Enable RFC 3042 (Limited Transmit)");
171 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
173 &VNET_NAME(tcp_do_rfc3390), 0,
174 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
176 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
178 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
179 "Slow-start flight size (initial congestion window) in number of segments");
181 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
183 &VNET_NAME(tcp_do_rfc3465), 0,
184 "Enable RFC 3465 (Appropriate Byte Counting)");
186 VNET_DEFINE(int, tcp_abc_l_var) = 2;
187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
188 &VNET_NAME(tcp_abc_l_var), 2,
189 "Cap the max cwnd increment during slow-start to this number of segments");
191 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn,
192 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
195 VNET_DEFINE(int, tcp_do_ecn) = 2;
196 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
197 &VNET_NAME(tcp_do_ecn), 0,
200 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
201 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
202 &VNET_NAME(tcp_ecn_maxretries), 0,
203 "Max retries before giving up on ECN");
205 VNET_DEFINE(int, tcp_insecure_syn) = 0;
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
207 &VNET_NAME(tcp_insecure_syn), 0,
208 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
210 VNET_DEFINE(int, tcp_insecure_rst) = 0;
211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
212 &VNET_NAME(tcp_insecure_rst), 0,
213 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
215 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
216 #define V_tcp_recvspace VNET(tcp_recvspace)
217 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
218 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
220 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
222 &VNET_NAME(tcp_do_autorcvbuf), 0,
223 "Enable automatic receive buffer sizing");
225 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
226 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
227 &VNET_NAME(tcp_autorcvbuf_max), 0,
228 "Max size of automatic receive buffer");
230 VNET_DEFINE(struct inpcbhead, tcb);
231 #define tcb6 tcb /* for KAME src sync over BSD*'s */
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 tcp_vnet_init(const void *unused)
250 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
251 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
253 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
254 tcp_vnet_init, NULL);
258 tcp_vnet_uninit(const void *unused)
261 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
262 VNET_PCPUSTAT_FREE(tcpstat);
264 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
265 tcp_vnet_uninit, NULL);
269 * Kernel module interface for updating tcpstat. The first argument is an index
270 * into tcpstat treated as an array.
273 kmod_tcpstat_add(int statnum, int val)
276 counter_u64_add(VNET(tcpstat)[statnum], val);
281 * Wrapper for the TCP established input helper hook.
284 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
286 struct tcp_hhook_data hhook_data;
288 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
293 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
300 * CC wrapper hook functions
303 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
310 INP_WLOCK_ASSERT(tp->t_inpcb);
312 tp->ccv->nsegs = nsegs;
313 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
314 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
315 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
316 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
317 tp->ccv->flags |= CCF_CWND_LIMITED;
319 tp->ccv->flags &= ~CCF_CWND_LIMITED;
321 if (type == CC_ACK) {
323 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
324 ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
325 if (!IN_RECOVERY(tp->t_flags))
326 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
327 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
328 if ((tp->t_flags & TF_GPUTINPROG) &&
329 SEQ_GEQ(th->th_ack, tp->gput_ack)) {
331 * Compute goodput in bits per millisecond.
333 gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) /
334 max(1, tcp_ts_getticks() - tp->gput_ts);
335 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
338 * XXXLAS: This is a temporary hack, and should be
339 * chained off VOI_TCP_GPUT when stats(9) grows an API
340 * to deal with chained VOIs.
342 if (tp->t_stats_gput_prev > 0)
343 stats_voi_update_abs_s32(tp->t_stats,
345 ((gput - tp->t_stats_gput_prev) * 100) /
346 tp->t_stats_gput_prev);
347 tp->t_flags &= ~TF_GPUTINPROG;
348 tp->t_stats_gput_prev = gput;
351 if (tp->snd_cwnd > tp->snd_ssthresh) {
352 tp->t_bytes_acked += tp->ccv->bytes_this_ack;
353 if (tp->t_bytes_acked >= tp->snd_cwnd) {
354 tp->t_bytes_acked -= tp->snd_cwnd;
355 tp->ccv->flags |= CCF_ABC_SENTAWND;
358 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
359 tp->t_bytes_acked = 0;
363 if (CC_ALGO(tp)->ack_received != NULL) {
364 /* XXXLAS: Find a way to live without this */
365 tp->ccv->curack = th->th_ack;
366 CC_ALGO(tp)->ack_received(tp->ccv, type);
369 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
374 cc_conn_init(struct tcpcb *tp)
376 struct hc_metrics_lite metrics;
377 struct inpcb *inp = tp->t_inpcb;
381 INP_WLOCK_ASSERT(tp->t_inpcb);
383 tcp_hc_get(&inp->inp_inc, &metrics);
384 maxseg = tcp_maxseg(tp);
386 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
388 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
389 TCPSTAT_INC(tcps_usedrtt);
390 if (metrics.rmx_rttvar) {
391 tp->t_rttvar = metrics.rmx_rttvar;
392 TCPSTAT_INC(tcps_usedrttvar);
394 /* default variation is +- 1 rtt */
396 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
398 TCPT_RANGESET(tp->t_rxtcur,
399 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
400 tp->t_rttmin, TCPTV_REXMTMAX);
402 if (metrics.rmx_ssthresh) {
404 * There's some sort of gateway or interface
405 * buffer limit on the path. Use this to set
406 * the slow start threshold, but set the
407 * threshold to no less than 2*mss.
409 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
410 TCPSTAT_INC(tcps_usedssthresh);
414 * Set the initial slow-start flight size.
416 * If a SYN or SYN/ACK was lost and retransmitted, we have to
417 * reduce the initial CWND to one segment as congestion is likely
418 * requiring us to be cautious.
420 if (tp->snd_cwnd == 1)
421 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
423 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
425 if (CC_ALGO(tp)->conn_init != NULL)
426 CC_ALGO(tp)->conn_init(tp->ccv);
430 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
432 INP_WLOCK_ASSERT(tp->t_inpcb);
435 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
440 if (!IN_FASTRECOVERY(tp->t_flags)) {
441 tp->snd_recover = tp->snd_max;
442 if (tp->t_flags2 & TF2_ECN_PERMIT)
443 tp->t_flags2 |= TF2_ECN_SND_CWR;
447 if (!IN_CONGRECOVERY(tp->t_flags) ||
449 * Allow ECN reaction on ACK to CWR, if
450 * that data segment was also CE marked.
452 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
453 EXIT_CONGRECOVERY(tp->t_flags);
454 TCPSTAT_INC(tcps_ecn_rcwnd);
455 tp->snd_recover = tp->snd_max + 1;
456 if (tp->t_flags2 & TF2_ECN_PERMIT)
457 tp->t_flags2 |= TF2_ECN_SND_CWR;
462 tp->t_bytes_acked = 0;
463 EXIT_RECOVERY(tp->t_flags);
464 if (tp->t_flags2 & TF2_ECN_PERMIT)
465 tp->t_flags2 |= TF2_ECN_SND_CWR;
468 TCPSTAT_INC(tcps_sndrexmitbad);
469 /* RTO was unnecessary, so reset everything. */
470 tp->snd_cwnd = tp->snd_cwnd_prev;
471 tp->snd_ssthresh = tp->snd_ssthresh_prev;
472 tp->snd_recover = tp->snd_recover_prev;
473 if (tp->t_flags & TF_WASFRECOVERY)
474 ENTER_FASTRECOVERY(tp->t_flags);
475 if (tp->t_flags & TF_WASCRECOVERY)
476 ENTER_CONGRECOVERY(tp->t_flags);
477 tp->snd_nxt = tp->snd_max;
478 tp->t_flags &= ~TF_PREVVALID;
483 if (CC_ALGO(tp)->cong_signal != NULL) {
485 tp->ccv->curack = th->th_ack;
486 CC_ALGO(tp)->cong_signal(tp->ccv, type);
491 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
493 INP_WLOCK_ASSERT(tp->t_inpcb);
495 /* XXXLAS: KASSERT that we're in recovery? */
497 if (CC_ALGO(tp)->post_recovery != NULL) {
498 tp->ccv->curack = th->th_ack;
499 CC_ALGO(tp)->post_recovery(tp->ccv);
501 /* XXXLAS: EXIT_RECOVERY ? */
502 tp->t_bytes_acked = 0;
506 * Indicate whether this ack should be delayed. We can delay the ack if
507 * following conditions are met:
508 * - There is no delayed ack timer in progress.
509 * - Our last ack wasn't a 0-sized window. We never want to delay
510 * the ack that opens up a 0-sized window.
511 * - LRO wasn't used for this segment. We make sure by checking that the
512 * segment size is not larger than the MSS.
514 #define DELAY_ACK(tp, tlen) \
515 ((!tcp_timer_active(tp, TT_DELACK) && \
516 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
517 (tlen <= tp->t_maxseg) && \
518 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
521 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
523 INP_WLOCK_ASSERT(tp->t_inpcb);
525 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
526 switch (iptos & IPTOS_ECN_MASK) {
528 tp->ccv->flags |= CCF_IPHDR_CE;
534 case IPTOS_ECN_NOTECT:
535 tp->ccv->flags &= ~CCF_IPHDR_CE;
539 if (th->th_flags & TH_CWR)
540 tp->ccv->flags |= CCF_TCPHDR_CWR;
542 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
544 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
546 if (tp->ccv->flags & CCF_ACKNOW) {
547 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
548 tp->t_flags |= TF_ACKNOW;
554 * TCP input handling is split into multiple parts:
555 * tcp6_input is a thin wrapper around tcp_input for the extended
556 * ip6_protox[] call format in ip6_input
557 * tcp_input handles primary segment validation, inpcb lookup and
558 * SYN processing on listen sockets
559 * tcp_do_segment processes the ACK and text of the segment for
560 * establishing, established and closing connections
564 tcp6_input(struct mbuf **mp, int *offp, int proto)
567 struct in6_ifaddr *ia6;
571 if (m->m_len < *offp + sizeof(struct tcphdr)) {
572 m = m_pullup(m, *offp + sizeof(struct tcphdr));
575 TCPSTAT_INC(tcps_rcvshort);
576 return (IPPROTO_DONE);
581 * draft-itojun-ipv6-tcp-to-anycast
582 * better place to put this in?
584 ip6 = mtod(m, struct ip6_hdr *);
585 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
586 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
587 ifa_free(&ia6->ia_ifa);
588 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
589 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
591 return (IPPROTO_DONE);
594 ifa_free(&ia6->ia_ifa);
597 return (tcp_input(mp, offp, proto));
602 tcp_input(struct mbuf **mp, int *offp, int proto)
604 struct mbuf *m = *mp;
605 struct tcphdr *th = NULL;
606 struct ip *ip = NULL;
607 struct inpcb *inp = NULL;
608 struct tcpcb *tp = NULL;
609 struct socket *so = NULL;
620 int rstreason = 0; /* For badport_bandlim accounting purposes */
622 struct m_tag *fwd_tag = NULL;
624 struct ip6_hdr *ip6 = NULL;
627 const void *ip6 = NULL;
629 struct tcpopt to; /* options in this segment */
630 char *s = NULL; /* address and port logging */
633 * The size of tcp_saveipgen must be the size of the max ip header,
636 u_char tcp_saveipgen[IP6_HDR_LEN];
637 struct tcphdr tcp_savetcp;
644 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
651 TCPSTAT_INC(tcps_rcvtotal);
655 ip6 = mtod(m, struct ip6_hdr *);
656 th = (struct tcphdr *)((caddr_t)ip6 + off0);
657 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
658 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
659 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
660 th->th_sum = m->m_pkthdr.csum_data;
662 th->th_sum = in6_cksum_pseudo(ip6, tlen,
663 IPPROTO_TCP, m->m_pkthdr.csum_data);
664 th->th_sum ^= 0xffff;
666 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
668 TCPSTAT_INC(tcps_rcvbadsum);
673 * Be proactive about unspecified IPv6 address in source.
674 * As we use all-zero to indicate unbounded/unconnected pcb,
675 * unspecified IPv6 address can be used to confuse us.
677 * Note that packets with unspecified IPv6 destination is
678 * already dropped in ip6_input.
680 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
684 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
687 #if defined(INET) && defined(INET6)
693 * Get IP and TCP header together in first mbuf.
694 * Note: IP leaves IP header in first mbuf.
696 if (off0 > sizeof (struct ip)) {
698 off0 = sizeof(struct ip);
700 if (m->m_len < sizeof (struct tcpiphdr)) {
701 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
703 TCPSTAT_INC(tcps_rcvshort);
704 return (IPPROTO_DONE);
707 ip = mtod(m, struct ip *);
708 th = (struct tcphdr *)((caddr_t)ip + off0);
709 tlen = ntohs(ip->ip_len) - off0;
712 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
713 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
714 th->th_sum = m->m_pkthdr.csum_data;
716 th->th_sum = in_pseudo(ip->ip_src.s_addr,
718 htonl(m->m_pkthdr.csum_data + tlen +
720 th->th_sum ^= 0xffff;
722 struct ipovly *ipov = (struct ipovly *)ip;
725 * Checksum extended TCP header and data.
729 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
730 ipov->ih_len = htons(tlen);
731 th->th_sum = in_cksum(m, len);
732 /* Reset length for SDT probes. */
733 ip->ip_len = htons(len);
736 /* Re-initialization for later version check */
738 ip->ip_v = IPVERSION;
739 ip->ip_hl = off0 >> 2;
743 TCPSTAT_INC(tcps_rcvbadsum);
750 * Check that TCP offset makes sense,
751 * pull out TCP options and adjust length. XXX
753 off = th->th_off << 2;
754 if (off < sizeof (struct tcphdr) || off > tlen) {
755 TCPSTAT_INC(tcps_rcvbadoff);
758 tlen -= off; /* tlen is used instead of ti->ti_len */
759 if (off > sizeof (struct tcphdr)) {
762 if (m->m_len < off0 + off) {
763 m = m_pullup(m, off0 + off);
765 TCPSTAT_INC(tcps_rcvshort);
766 return (IPPROTO_DONE);
769 ip6 = mtod(m, struct ip6_hdr *);
770 th = (struct tcphdr *)((caddr_t)ip6 + off0);
773 #if defined(INET) && defined(INET6)
778 if (m->m_len < sizeof(struct ip) + off) {
779 if ((m = m_pullup(m, sizeof (struct ip) + off))
781 TCPSTAT_INC(tcps_rcvshort);
782 return (IPPROTO_DONE);
784 ip = mtod(m, struct ip *);
785 th = (struct tcphdr *)((caddr_t)ip + off0);
789 optlen = off - sizeof (struct tcphdr);
790 optp = (u_char *)(th + 1);
792 thflags = th->th_flags;
795 * Convert TCP protocol specific fields to host format.
797 tcp_fields_to_host(th);
800 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
802 drop_hdrlen = off0 + off;
805 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
809 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
811 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
814 #if defined(INET) && !defined(INET6)
815 (m->m_flags & M_IP_NEXTHOP)
818 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
822 if (isipv6 && fwd_tag != NULL) {
823 struct sockaddr_in6 *next_hop6;
825 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
827 * Transparently forwarded. Pretend to be the destination.
828 * Already got one like this?
830 inp = in6_pcblookup_mbuf(&V_tcbinfo,
831 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
832 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
835 * It's new. Try to find the ambushing socket.
836 * Because we've rewritten the destination address,
837 * any hardware-generated hash is ignored.
839 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
840 th->th_sport, &next_hop6->sin6_addr,
841 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
842 th->th_dport, INPLOOKUP_WILDCARD |
843 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
846 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
847 th->th_sport, &ip6->ip6_dst, th->th_dport,
848 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
849 m->m_pkthdr.rcvif, m);
852 #if defined(INET6) && defined(INET)
856 if (fwd_tag != NULL) {
857 struct sockaddr_in *next_hop;
859 next_hop = (struct sockaddr_in *)(fwd_tag+1);
861 * Transparently forwarded. Pretend to be the destination.
862 * already got one like this?
864 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
865 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
866 m->m_pkthdr.rcvif, m);
869 * It's new. Try to find the ambushing socket.
870 * Because we've rewritten the destination address,
871 * any hardware-generated hash is ignored.
873 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
874 th->th_sport, next_hop->sin_addr,
875 next_hop->sin_port ? ntohs(next_hop->sin_port) :
876 th->th_dport, INPLOOKUP_WILDCARD |
877 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
880 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
881 th->th_sport, ip->ip_dst, th->th_dport,
882 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
883 m->m_pkthdr.rcvif, m);
887 * If the INPCB does not exist then all data in the incoming
888 * segment is discarded and an appropriate RST is sent back.
889 * XXX MRT Send RST using which routing table?
893 * Log communication attempts to ports that are not
896 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
897 V_tcp_log_in_vain == 2) {
898 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
899 log(LOG_INFO, "%s; %s: Connection attempt "
900 "to closed port\n", s, __func__);
903 * When blackholing do not respond with a RST but
904 * completely ignore the segment and drop it.
906 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
910 rstreason = BANDLIM_RST_CLOSEDPORT;
913 INP_WLOCK_ASSERT(inp);
915 * While waiting for inp lock during the lookup, another thread
916 * can have dropped the inpcb, in which case we need to loop back
917 * and try to find a new inpcb to deliver to.
919 if (inp->inp_flags & INP_DROPPED) {
924 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
925 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
926 ((inp->inp_socket == NULL) ||
927 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
928 inp->inp_flowid = m->m_pkthdr.flowid;
929 inp->inp_flowtype = M_HASHTYPE_GET(m);
931 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
933 if (isipv6 && IPSEC_ENABLED(ipv6) &&
934 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
942 if (IPSEC_ENABLED(ipv4) &&
943 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
950 * Check the minimum TTL for socket.
952 if (inp->inp_ip_minttl != 0) {
955 if (inp->inp_ip_minttl > ip6->ip6_hlim)
959 if (inp->inp_ip_minttl > ip->ip_ttl)
964 * A previous connection in TIMEWAIT state is supposed to catch stray
965 * or duplicate segments arriving late. If this segment was a
966 * legitimate new connection attempt, the old INPCB gets removed and
967 * we can try again to find a listening socket.
969 * At this point, due to earlier optimism, we may hold only an inpcb
970 * lock, and not the inpcbinfo write lock. If so, we need to try to
971 * acquire it, or if that fails, acquire a reference on the inpcb,
972 * drop all locks, acquire a global write lock, and then re-acquire
973 * the inpcb lock. We may at that point discover that another thread
974 * has tried to free the inpcb, in which case we need to loop back
975 * and try to find a new inpcb to deliver to.
977 * XXXRW: It may be time to rethink timewait locking.
979 if (inp->inp_flags & INP_TIMEWAIT) {
980 tcp_dooptions(&to, optp, optlen,
981 (thflags & TH_SYN) ? TO_SYN : 0);
983 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
985 if (tcp_twcheck(inp, &to, th, m, tlen))
987 return (IPPROTO_DONE);
990 * The TCPCB may no longer exist if the connection is winding
991 * down or it is in the CLOSED state. Either way we drop the
992 * segment and send an appropriate response.
995 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
996 rstreason = BANDLIM_RST_CLOSEDPORT;
1001 if (tp->t_flags & TF_TOE) {
1002 tcp_offload_input(tp, m);
1003 m = NULL; /* consumed by the TOE driver */
1009 INP_WLOCK_ASSERT(inp);
1010 if (mac_inpcb_check_deliver(inp, m))
1013 so = inp->inp_socket;
1014 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1016 if (so->so_options & SO_DEBUG) {
1017 ostate = tp->t_state;
1020 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1023 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1026 #endif /* TCPDEBUG */
1028 * When the socket is accepting connections (the INPCB is in LISTEN
1029 * state) we look into the SYN cache if this is a new connection
1030 * attempt or the completion of a previous one.
1032 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1033 ("%s: so accepting but tp %p not listening", __func__, tp));
1034 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1035 struct in_conninfo inc;
1037 bzero(&inc, sizeof(inc));
1040 inc.inc_flags |= INC_ISIPV6;
1041 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1042 inc.inc_flags |= INC_IPV6MINMTU;
1043 inc.inc6_faddr = ip6->ip6_src;
1044 inc.inc6_laddr = ip6->ip6_dst;
1048 inc.inc_faddr = ip->ip_src;
1049 inc.inc_laddr = ip->ip_dst;
1051 inc.inc_fport = th->th_sport;
1052 inc.inc_lport = th->th_dport;
1053 inc.inc_fibnum = so->so_fibnum;
1056 * Check for an existing connection attempt in syncache if
1057 * the flag is only ACK. A successful lookup creates a new
1058 * socket appended to the listen queue in SYN_RECEIVED state.
1060 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1062 * Parse the TCP options here because
1063 * syncookies need access to the reflected
1066 tcp_dooptions(&to, optp, optlen, 0);
1068 * NB: syncache_expand() doesn't unlock
1069 * inp and tcpinfo locks.
1071 rstreason = syncache_expand(&inc, &to, th, &so, m);
1072 if (rstreason < 0) {
1074 * A failing TCP MD5 signature comparison
1075 * must result in the segment being dropped
1076 * and must not produce any response back
1080 } else if (rstreason == 0) {
1082 * No syncache entry or ACK was not
1083 * for our SYN/ACK. Send a RST.
1084 * NB: syncache did its own logging
1085 * of the failure cause.
1087 rstreason = BANDLIM_RST_OPENPORT;
1093 * We completed the 3-way handshake
1094 * but could not allocate a socket
1095 * either due to memory shortage,
1096 * listen queue length limits or
1097 * global socket limits. Send RST
1098 * or wait and have the remote end
1099 * retransmit the ACK for another
1102 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1103 log(LOG_DEBUG, "%s; %s: Listen socket: "
1104 "Socket allocation failed due to "
1105 "limits or memory shortage, %s\n",
1107 V_tcp_sc_rst_sock_fail ?
1108 "sending RST" : "try again");
1109 if (V_tcp_sc_rst_sock_fail) {
1110 rstreason = BANDLIM_UNLIMITED;
1116 * Socket is created in state SYN_RECEIVED.
1117 * Unlock the listen socket, lock the newly
1118 * created socket and update the tp variable.
1120 INP_WUNLOCK(inp); /* listen socket */
1121 inp = sotoinpcb(so);
1123 * New connection inpcb is already locked by
1124 * syncache_expand().
1126 INP_WLOCK_ASSERT(inp);
1127 tp = intotcpcb(inp);
1128 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1129 ("%s: ", __func__));
1131 * Process the segment and the data it
1132 * contains. tcp_do_segment() consumes
1133 * the mbuf chain and unlocks the inpcb.
1135 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1136 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1138 return (IPPROTO_DONE);
1141 * Segment flag validation for new connection attempts:
1143 * Our (SYN|ACK) response was rejected.
1144 * Check with syncache and remove entry to prevent
1147 * NB: syncache_chkrst does its own logging of failure
1150 if (thflags & TH_RST) {
1151 syncache_chkrst(&inc, th, m);
1155 * We can't do anything without SYN.
1157 if ((thflags & TH_SYN) == 0) {
1158 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1159 log(LOG_DEBUG, "%s; %s: Listen socket: "
1160 "SYN is missing, segment ignored\n",
1162 TCPSTAT_INC(tcps_badsyn);
1166 * (SYN|ACK) is bogus on a listen socket.
1168 if (thflags & TH_ACK) {
1169 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1170 log(LOG_DEBUG, "%s; %s: Listen socket: "
1171 "SYN|ACK invalid, segment rejected\n",
1173 syncache_badack(&inc); /* XXX: Not needed! */
1174 TCPSTAT_INC(tcps_badsyn);
1175 rstreason = BANDLIM_RST_OPENPORT;
1179 * If the drop_synfin option is enabled, drop all
1180 * segments with both the SYN and FIN bits set.
1181 * This prevents e.g. nmap from identifying the
1183 * XXX: Poor reasoning. nmap has other methods
1184 * and is constantly refining its stack detection
1186 * XXX: This is a violation of the TCP specification
1187 * and was used by RFC1644.
1189 if ((thflags & TH_FIN) && V_drop_synfin) {
1190 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1191 log(LOG_DEBUG, "%s; %s: Listen socket: "
1192 "SYN|FIN segment ignored (based on "
1193 "sysctl setting)\n", s, __func__);
1194 TCPSTAT_INC(tcps_badsyn);
1198 * Segment's flags are (SYN) or (SYN|FIN).
1200 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1201 * as they do not affect the state of the TCP FSM.
1202 * The data pointed to by TH_URG and th_urp is ignored.
1204 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1205 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1206 KASSERT(thflags & (TH_SYN),
1207 ("%s: Listen socket: TH_SYN not set", __func__));
1210 * If deprecated address is forbidden,
1211 * we do not accept SYN to deprecated interface
1212 * address to prevent any new inbound connection from
1213 * getting established.
1214 * When we do not accept SYN, we send a TCP RST,
1215 * with deprecated source address (instead of dropping
1216 * it). We compromise it as it is much better for peer
1217 * to send a RST, and RST will be the final packet
1220 * If we do not forbid deprecated addresses, we accept
1221 * the SYN packet. RFC2462 does not suggest dropping
1223 * If we decipher RFC2462 5.5.4, it says like this:
1224 * 1. use of deprecated addr with existing
1225 * communication is okay - "SHOULD continue to be
1227 * 2. use of it with new communication:
1228 * (2a) "SHOULD NOT be used if alternate address
1229 * with sufficient scope is available"
1230 * (2b) nothing mentioned otherwise.
1231 * Here we fall into (2b) case as we have no choice in
1232 * our source address selection - we must obey the peer.
1234 * The wording in RFC2462 is confusing, and there are
1235 * multiple description text for deprecated address
1236 * handling - worse, they are not exactly the same.
1237 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1239 if (isipv6 && !V_ip6_use_deprecated) {
1240 struct in6_ifaddr *ia6;
1242 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1244 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1245 ifa_free(&ia6->ia_ifa);
1246 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1247 log(LOG_DEBUG, "%s; %s: Listen socket: "
1248 "Connection attempt to deprecated "
1249 "IPv6 address rejected\n",
1251 rstreason = BANDLIM_RST_OPENPORT;
1255 ifa_free(&ia6->ia_ifa);
1259 * Basic sanity checks on incoming SYN requests:
1260 * Don't respond if the destination is a link layer
1261 * broadcast according to RFC1122 4.2.3.10, p. 104.
1262 * If it is from this socket it must be forged.
1263 * Don't respond if the source or destination is a
1264 * global or subnet broad- or multicast address.
1265 * Note that it is quite possible to receive unicast
1266 * link-layer packets with a broadcast IP address. Use
1267 * in_broadcast() to find them.
1269 if (m->m_flags & (M_BCAST|M_MCAST)) {
1270 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1271 log(LOG_DEBUG, "%s; %s: Listen socket: "
1272 "Connection attempt from broad- or multicast "
1273 "link layer address ignored\n", s, __func__);
1278 if (th->th_dport == th->th_sport &&
1279 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1280 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1281 log(LOG_DEBUG, "%s; %s: Listen socket: "
1282 "Connection attempt to/from self "
1283 "ignored\n", s, __func__);
1286 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1287 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1288 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1289 log(LOG_DEBUG, "%s; %s: Listen socket: "
1290 "Connection attempt from/to multicast "
1291 "address ignored\n", s, __func__);
1296 #if defined(INET) && defined(INET6)
1301 if (th->th_dport == th->th_sport &&
1302 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1303 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1304 log(LOG_DEBUG, "%s; %s: Listen socket: "
1305 "Connection attempt from/to self "
1306 "ignored\n", s, __func__);
1309 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1310 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1311 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1312 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1313 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1314 log(LOG_DEBUG, "%s; %s: Listen socket: "
1315 "Connection attempt from/to broad- "
1316 "or multicast address ignored\n",
1323 * SYN appears to be valid. Create compressed TCP state
1327 if (so->so_options & SO_DEBUG)
1328 tcp_trace(TA_INPUT, ostate, tp,
1329 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1331 TCP_PROBE3(debug__input, tp, th, m);
1332 tcp_dooptions(&to, optp, optlen, TO_SYN);
1333 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos))
1334 goto tfo_socket_result;
1337 * Entry added to syncache and mbuf consumed.
1338 * Only the listen socket is unlocked by syncache_add().
1340 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1341 return (IPPROTO_DONE);
1342 } else if (tp->t_state == TCPS_LISTEN) {
1344 * When a listen socket is torn down the SO_ACCEPTCONN
1345 * flag is removed first while connections are drained
1346 * from the accept queue in a unlock/lock cycle of the
1347 * ACCEPT_LOCK, opening a race condition allowing a SYN
1348 * attempt go through unhandled.
1352 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1353 if (tp->t_flags & TF_SIGNATURE) {
1354 tcp_dooptions(&to, optp, optlen, thflags);
1355 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1356 TCPSTAT_INC(tcps_sig_err_nosigopt);
1359 if (!TCPMD5_ENABLED() ||
1360 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1364 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1367 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1368 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1369 * the inpcb, and unlocks pcbinfo.
1371 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1372 return (IPPROTO_DONE);
1375 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1378 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1381 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1382 m = NULL; /* mbuf chain got consumed. */
1387 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1393 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1398 return (IPPROTO_DONE);
1402 * Automatic sizing of receive socket buffer. Often the send
1403 * buffer size is not optimally adjusted to the actual network
1404 * conditions at hand (delay bandwidth product). Setting the
1405 * buffer size too small limits throughput on links with high
1406 * bandwidth and high delay (eg. trans-continental/oceanic links).
1408 * On the receive side the socket buffer memory is only rarely
1409 * used to any significant extent. This allows us to be much
1410 * more aggressive in scaling the receive socket buffer. For
1411 * the case that the buffer space is actually used to a large
1412 * extent and we run out of kernel memory we can simply drop
1413 * the new segments; TCP on the sender will just retransmit it
1414 * later. Setting the buffer size too big may only consume too
1415 * much kernel memory if the application doesn't read() from
1416 * the socket or packet loss or reordering makes use of the
1419 * The criteria to step up the receive buffer one notch are:
1420 * 1. Application has not set receive buffer size with
1421 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1422 * 2. the number of bytes received during 1/2 of an sRTT
1423 * is at least 3/8 of the current socket buffer size.
1424 * 3. receive buffer size has not hit maximal automatic size;
1426 * If all of the criteria are met we increaset the socket buffer
1427 * by a 1/2 (bounded by the max). This allows us to keep ahead
1428 * of slow-start but also makes it so our peer never gets limited
1429 * by our rwnd which we then open up causing a burst.
1431 * This algorithm does two steps per RTT at most and only if
1432 * we receive a bulk stream w/o packet losses or reorderings.
1433 * Shrinking the buffer during idle times is not necessary as
1434 * it doesn't consume any memory when idle.
1436 * TODO: Only step up if the application is actually serving
1437 * the buffer to better manage the socket buffer resources.
1440 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1441 struct tcpcb *tp, int tlen)
1445 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1446 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1447 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1448 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1449 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1450 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1451 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1453 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1455 /* Start over with next RTT. */
1459 tp->rfbuf_cnt += tlen; /* add up */
1465 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so)
1468 * Since tp might be gone if the session entered
1469 * the TIME_WAIT state before coming here, we need
1470 * to check if the socket is still connected.
1472 if ((so->so_state & SS_ISCONNECTED) == 0)
1474 INP_LOCK_ASSERT(tp->t_inpcb);
1475 if (tp->t_flags & TF_WAKESOR) {
1476 tp->t_flags &= ~TF_WAKESOR;
1477 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
1480 if (tp->t_flags & TF_WAKESOW) {
1481 tp->t_flags &= ~TF_WAKESOW;
1482 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
1488 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1489 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1491 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1492 int rstreason, todrop, win, incforsyn = 0;
1496 struct in_conninfo *inc;
1503 * The size of tcp_saveipgen must be the size of the max ip header,
1506 u_char tcp_saveipgen[IP6_HDR_LEN];
1507 struct tcphdr tcp_savetcp;
1510 thflags = th->th_flags;
1511 inc = &tp->t_inpcb->inp_inc;
1512 tp->sackhint.last_sack_ack = 0;
1514 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1517 INP_WLOCK_ASSERT(tp->t_inpcb);
1518 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1520 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1524 /* Save segment, if requested. */
1525 tcp_pcap_add(th, m, &(tp->t_inpkts));
1527 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1530 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1531 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1532 log(LOG_DEBUG, "%s; %s: "
1533 "SYN|FIN segment ignored (based on "
1534 "sysctl setting)\n", s, __func__);
1541 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1542 * check SEQ.ACK first.
1544 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1545 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1546 rstreason = BANDLIM_UNLIMITED;
1551 * Segment received on connection.
1552 * Reset idle time and keep-alive timer.
1553 * XXX: This should be done after segment
1554 * validation to ignore broken/spoofed segs.
1556 tp->t_rcvtime = ticks;
1559 * Scale up the window into a 32-bit value.
1560 * For the SYN_SENT state the scale is zero.
1562 tiwin = th->th_win << tp->snd_scale;
1564 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1568 * TCP ECN processing.
1570 if (tp->t_flags2 & TF2_ECN_PERMIT) {
1571 if (thflags & TH_CWR) {
1572 tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1573 tp->t_flags |= TF_ACKNOW;
1575 switch (iptos & IPTOS_ECN_MASK) {
1577 tp->t_flags2 |= TF2_ECN_SND_ECE;
1578 TCPSTAT_INC(tcps_ecn_ce);
1580 case IPTOS_ECN_ECT0:
1581 TCPSTAT_INC(tcps_ecn_ect0);
1583 case IPTOS_ECN_ECT1:
1584 TCPSTAT_INC(tcps_ecn_ect1);
1588 /* Process a packet differently from RFC3168. */
1589 cc_ecnpkt_handler(tp, th, iptos);
1591 /* Congestion experienced. */
1592 if (thflags & TH_ECE) {
1593 cc_cong_signal(tp, th, CC_ECN);
1598 * Parse options on any incoming segment.
1600 tcp_dooptions(&to, (u_char *)(th + 1),
1601 (th->th_off << 2) - sizeof(struct tcphdr),
1602 (thflags & TH_SYN) ? TO_SYN : 0);
1604 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1605 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1606 (to.to_flags & TOF_SIGNATURE) == 0) {
1607 TCPSTAT_INC(tcps_sig_err_sigopt);
1608 /* XXX: should drop? */
1612 * If echoed timestamp is later than the current time,
1613 * fall back to non RFC1323 RTT calculation. Normalize
1614 * timestamp if syncookies were used when this connection
1617 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1618 to.to_tsecr -= tp->ts_offset;
1619 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1621 else if (tp->t_flags & TF_PREVVALID &&
1622 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1623 cc_cong_signal(tp, th, CC_RTO_ERR);
1626 * Process options only when we get SYN/ACK back. The SYN case
1627 * for incoming connections is handled in tcp_syncache.
1628 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1629 * or <SYN,ACK>) segment itself is never scaled.
1630 * XXX this is traditional behavior, may need to be cleaned up.
1632 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1633 /* Handle parallel SYN for ECN */
1634 if (!(thflags & TH_ACK) &&
1635 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) &&
1636 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) {
1637 tp->t_flags2 |= TF2_ECN_PERMIT;
1638 tp->t_flags2 |= TF2_ECN_SND_ECE;
1639 TCPSTAT_INC(tcps_ecn_shs);
1641 if ((to.to_flags & TOF_SCALE) &&
1642 (tp->t_flags & TF_REQ_SCALE)) {
1643 tp->t_flags |= TF_RCVD_SCALE;
1644 tp->snd_scale = to.to_wscale;
1646 tp->t_flags &= ~TF_REQ_SCALE;
1648 * Initial send window. It will be updated with
1649 * the next incoming segment to the scaled value.
1651 tp->snd_wnd = th->th_win;
1652 if ((to.to_flags & TOF_TS) &&
1653 (tp->t_flags & TF_REQ_TSTMP)) {
1654 tp->t_flags |= TF_RCVD_TSTMP;
1655 tp->ts_recent = to.to_tsval;
1656 tp->ts_recent_age = tcp_ts_getticks();
1658 tp->t_flags &= ~TF_REQ_TSTMP;
1659 if (to.to_flags & TOF_MSS)
1660 tcp_mss(tp, to.to_mss);
1661 if ((tp->t_flags & TF_SACK_PERMIT) &&
1662 (to.to_flags & TOF_SACKPERM) == 0)
1663 tp->t_flags &= ~TF_SACK_PERMIT;
1664 if (IS_FASTOPEN(tp->t_flags)) {
1665 if (to.to_flags & TOF_FASTOPEN) {
1668 if (to.to_flags & TOF_MSS)
1671 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1675 tcp_fastopen_update_cache(tp, mss,
1676 to.to_tfo_len, to.to_tfo_cookie);
1678 tcp_fastopen_disable_path(tp);
1683 * If timestamps were negotiated during SYN/ACK and a
1684 * segment without a timestamp is received, silently drop
1686 * See section 3.2 of RFC 7323.
1688 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1689 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1690 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1691 "segment silently dropped\n", s, __func__);
1697 * If timestamps were not negotiated during SYN/ACK and a
1698 * segment with a timestamp is received, ignore the
1699 * timestamp and process the packet normally.
1700 * See section 3.2 of RFC 7323.
1702 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1703 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1704 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1705 "segment processed normally\n", s, __func__);
1711 * Header prediction: check for the two common cases
1712 * of a uni-directional data xfer. If the packet has
1713 * no control flags, is in-sequence, the window didn't
1714 * change and we're not retransmitting, it's a
1715 * candidate. If the length is zero and the ack moved
1716 * forward, we're the sender side of the xfer. Just
1717 * free the data acked & wake any higher level process
1718 * that was blocked waiting for space. If the length
1719 * is non-zero and the ack didn't move, we're the
1720 * receiver side. If we're getting packets in-order
1721 * (the reassembly queue is empty), add the data to
1722 * the socket buffer and note that we need a delayed ack.
1723 * Make sure that the hidden state-flags are also off.
1724 * Since we check for TCPS_ESTABLISHED first, it can only
1727 if (tp->t_state == TCPS_ESTABLISHED &&
1728 th->th_seq == tp->rcv_nxt &&
1729 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1730 tp->snd_nxt == tp->snd_max &&
1731 tiwin && tiwin == tp->snd_wnd &&
1732 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1734 ((to.to_flags & TOF_TS) == 0 ||
1735 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1737 * If last ACK falls within this segment's sequence numbers,
1738 * record the timestamp.
1739 * NOTE that the test is modified according to the latest
1740 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1742 if ((to.to_flags & TOF_TS) != 0 &&
1743 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1744 tp->ts_recent_age = tcp_ts_getticks();
1745 tp->ts_recent = to.to_tsval;
1749 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1750 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1751 !IN_RECOVERY(tp->t_flags) &&
1752 (to.to_flags & TOF_SACK) == 0 &&
1753 TAILQ_EMPTY(&tp->snd_holes)) {
1755 * This is a pure ack for outstanding data.
1757 TCPSTAT_INC(tcps_predack);
1760 * "bad retransmit" recovery without timestamps.
1762 if ((to.to_flags & TOF_TS) == 0 &&
1763 tp->t_rxtshift == 1 &&
1764 tp->t_flags & TF_PREVVALID &&
1765 (int)(ticks - tp->t_badrxtwin) < 0) {
1766 cc_cong_signal(tp, th, CC_RTO_ERR);
1770 * Recalculate the transmit timer / rtt.
1772 * Some boxes send broken timestamp replies
1773 * during the SYN+ACK phase, ignore
1774 * timestamps of 0 or we could calculate a
1775 * huge RTT and blow up the retransmit timer.
1777 if ((to.to_flags & TOF_TS) != 0 &&
1781 t = tcp_ts_getticks() - to.to_tsecr;
1782 if (!tp->t_rttlow || tp->t_rttlow > t)
1785 TCP_TS_TO_TICKS(t) + 1);
1786 } else if (tp->t_rtttime &&
1787 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1788 if (!tp->t_rttlow ||
1789 tp->t_rttlow > ticks - tp->t_rtttime)
1790 tp->t_rttlow = ticks - tp->t_rtttime;
1792 ticks - tp->t_rtttime);
1794 acked = BYTES_THIS_ACK(tp, th);
1797 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1798 hhook_run_tcp_est_in(tp, th, &to);
1801 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1802 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1803 sbdrop(&so->so_snd, acked);
1804 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1805 SEQ_LEQ(th->th_ack, tp->snd_recover))
1806 tp->snd_recover = th->th_ack - 1;
1809 * Let the congestion control algorithm update
1810 * congestion control related information. This
1811 * typically means increasing the congestion
1814 cc_ack_received(tp, th, nsegs, CC_ACK);
1816 tp->snd_una = th->th_ack;
1818 * Pull snd_wl2 up to prevent seq wrap relative
1821 tp->snd_wl2 = th->th_ack;
1826 * If all outstanding data are acked, stop
1827 * retransmit timer, otherwise restart timer
1828 * using current (possibly backed-off) value.
1829 * If process is waiting for space,
1830 * wakeup/selwakeup/signal. If data
1831 * are ready to send, let tcp_output
1832 * decide between more output or persist.
1835 if (so->so_options & SO_DEBUG)
1836 tcp_trace(TA_INPUT, ostate, tp,
1837 (void *)tcp_saveipgen,
1840 TCP_PROBE3(debug__input, tp, th, m);
1841 if (tp->snd_una == tp->snd_max)
1842 tcp_timer_activate(tp, TT_REXMT, 0);
1843 else if (!tcp_timer_active(tp, TT_PERSIST))
1844 tcp_timer_activate(tp, TT_REXMT,
1846 tp->t_flags |= TF_WAKESOW;
1847 if (sbavail(&so->so_snd))
1848 (void) tp->t_fb->tfb_tcp_output(tp);
1851 } else if (th->th_ack == tp->snd_una &&
1852 tlen <= sbspace(&so->so_rcv)) {
1853 int newsize = 0; /* automatic sockbuf scaling */
1856 * This is a pure, in-sequence data packet with
1857 * nothing on the reassembly queue and we have enough
1858 * buffer space to take it.
1860 /* Clean receiver SACK report if present */
1861 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1862 tcp_clean_sackreport(tp);
1863 TCPSTAT_INC(tcps_preddat);
1864 tp->rcv_nxt += tlen;
1866 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1867 (tp->t_fbyte_in == 0)) {
1868 tp->t_fbyte_in = ticks;
1869 if (tp->t_fbyte_in == 0)
1871 if (tp->t_fbyte_out && tp->t_fbyte_in)
1872 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1875 * Pull snd_wl1 up to prevent seq wrap relative to
1878 tp->snd_wl1 = th->th_seq;
1880 * Pull rcv_up up to prevent seq wrap relative to
1883 tp->rcv_up = tp->rcv_nxt;
1884 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1885 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1887 if (so->so_options & SO_DEBUG)
1888 tcp_trace(TA_INPUT, ostate, tp,
1889 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1891 TCP_PROBE3(debug__input, tp, th, m);
1893 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1895 /* Add data to socket buffer. */
1896 SOCKBUF_LOCK(&so->so_rcv);
1897 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1901 * Set new socket buffer size.
1902 * Give up when limit is reached.
1905 if (!sbreserve_locked(&so->so_rcv,
1907 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1908 m_adj(m, drop_hdrlen); /* delayed header drop */
1909 sbappendstream_locked(&so->so_rcv, m, 0);
1911 SOCKBUF_UNLOCK(&so->so_rcv);
1912 tp->t_flags |= TF_WAKESOR;
1913 if (DELAY_ACK(tp, tlen)) {
1914 tp->t_flags |= TF_DELACK;
1916 tp->t_flags |= TF_ACKNOW;
1917 tp->t_fb->tfb_tcp_output(tp);
1924 * Calculate amount of space in receive window,
1925 * and then do TCP input processing.
1926 * Receive window is amount of space in rcv queue,
1927 * but not less than advertised window.
1929 win = sbspace(&so->so_rcv);
1932 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1934 switch (tp->t_state) {
1936 * If the state is SYN_RECEIVED:
1937 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1939 case TCPS_SYN_RECEIVED:
1940 if ((thflags & TH_ACK) &&
1941 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1942 SEQ_GT(th->th_ack, tp->snd_max))) {
1943 rstreason = BANDLIM_RST_OPENPORT;
1946 if (IS_FASTOPEN(tp->t_flags)) {
1948 * When a TFO connection is in SYN_RECEIVED, the
1949 * only valid packets are the initial SYN, a
1950 * retransmit/copy of the initial SYN (possibly with
1951 * a subset of the original data), a valid ACK, a
1954 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1955 rstreason = BANDLIM_RST_OPENPORT;
1957 } else if (thflags & TH_SYN) {
1958 /* non-initial SYN is ignored */
1959 if ((tcp_timer_active(tp, TT_DELACK) ||
1960 tcp_timer_active(tp, TT_REXMT)))
1962 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1969 * If the state is SYN_SENT:
1970 * if seg contains a RST with valid ACK (SEQ.ACK has already
1971 * been verified), then drop the connection.
1972 * if seg contains a RST without an ACK, drop the seg.
1973 * if seg does not contain SYN, then drop the seg.
1974 * Otherwise this is an acceptable SYN segment
1975 * initialize tp->rcv_nxt and tp->irs
1976 * if seg contains ack then advance tp->snd_una
1977 * if seg contains an ECE and ECN support is enabled, the stream
1979 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1980 * arrange for segment to be acked (eventually)
1981 * continue processing rest of data/controls, beginning with URG
1984 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1985 TCP_PROBE5(connect__refused, NULL, tp,
1987 tp = tcp_drop(tp, ECONNREFUSED);
1989 if (thflags & TH_RST)
1991 if (!(thflags & TH_SYN))
1994 tp->irs = th->th_seq;
1996 if (thflags & TH_ACK) {
1997 int tfo_partial_ack = 0;
1999 TCPSTAT_INC(tcps_connects);
2002 mac_socketpeer_set_from_mbuf(m, so);
2004 /* Do window scaling on this connection? */
2005 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2006 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2007 tp->rcv_scale = tp->request_r_scale;
2009 tp->rcv_adv += min(tp->rcv_wnd,
2010 TCP_MAXWIN << tp->rcv_scale);
2011 tp->snd_una++; /* SYN is acked */
2013 * If not all the data that was sent in the TFO SYN
2014 * has been acked, resend the remainder right away.
2016 if (IS_FASTOPEN(tp->t_flags) &&
2017 (tp->snd_una != tp->snd_max)) {
2018 tp->snd_nxt = th->th_ack;
2019 tfo_partial_ack = 1;
2022 * If there's data, delay ACK; if there's also a FIN
2023 * ACKNOW will be turned on later.
2025 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2026 tcp_timer_activate(tp, TT_DELACK,
2029 tp->t_flags |= TF_ACKNOW;
2031 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
2032 (V_tcp_do_ecn == 1)) {
2033 tp->t_flags2 |= TF2_ECN_PERMIT;
2034 TCPSTAT_INC(tcps_ecn_shs);
2038 * Received <SYN,ACK> in SYN_SENT[*] state.
2040 * SYN_SENT --> ESTABLISHED
2041 * SYN_SENT* --> FIN_WAIT_1
2043 tp->t_starttime = ticks;
2044 if (tp->t_flags & TF_NEEDFIN) {
2045 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2046 tp->t_flags &= ~TF_NEEDFIN;
2049 tcp_state_change(tp, TCPS_ESTABLISHED);
2050 TCP_PROBE5(connect__established, NULL, tp,
2053 tcp_timer_activate(tp, TT_KEEP,
2058 * Received initial SYN in SYN-SENT[*] state =>
2059 * simultaneous open.
2060 * If it succeeds, connection is * half-synchronized.
2061 * Otherwise, do 3-way handshake:
2062 * SYN-SENT -> SYN-RECEIVED
2063 * SYN-SENT* -> SYN-RECEIVED*
2065 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2066 tcp_timer_activate(tp, TT_REXMT, 0);
2067 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2070 INP_WLOCK_ASSERT(tp->t_inpcb);
2073 * Advance th->th_seq to correspond to first data byte.
2074 * If data, trim to stay within window,
2075 * dropping FIN if necessary.
2078 if (tlen > tp->rcv_wnd) {
2079 todrop = tlen - tp->rcv_wnd;
2083 TCPSTAT_INC(tcps_rcvpackafterwin);
2084 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2086 tp->snd_wl1 = th->th_seq - 1;
2087 tp->rcv_up = th->th_seq;
2089 * Client side of transaction: already sent SYN and data.
2090 * If the remote host used T/TCP to validate the SYN,
2091 * our data will be ACK'd; if so, enter normal data segment
2092 * processing in the middle of step 5, ack processing.
2093 * Otherwise, goto step 6.
2095 if (thflags & TH_ACK)
2101 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2102 * do normal processing.
2104 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2108 break; /* continue normal processing */
2112 * States other than LISTEN or SYN_SENT.
2113 * First check the RST flag and sequence number since reset segments
2114 * are exempt from the timestamp and connection count tests. This
2115 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2116 * below which allowed reset segments in half the sequence space
2117 * to fall though and be processed (which gives forged reset
2118 * segments with a random sequence number a 50 percent chance of
2119 * killing a connection).
2120 * Then check timestamp, if present.
2121 * Then check the connection count, if present.
2122 * Then check that at least some bytes of segment are within
2123 * receive window. If segment begins before rcv_nxt,
2124 * drop leading data (and SYN); if nothing left, just ack.
2126 if (thflags & TH_RST) {
2128 * RFC5961 Section 3.2
2130 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2131 * - If RST is in window, we send challenge ACK.
2133 * Note: to take into account delayed ACKs, we should
2134 * test against last_ack_sent instead of rcv_nxt.
2135 * Note 2: we handle special case of closed window, not
2136 * covered by the RFC.
2138 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2139 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2140 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2141 KASSERT(tp->t_state != TCPS_SYN_SENT,
2142 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2145 if (V_tcp_insecure_rst ||
2146 tp->last_ack_sent == th->th_seq) {
2147 TCPSTAT_INC(tcps_drops);
2148 /* Drop the connection. */
2149 switch (tp->t_state) {
2150 case TCPS_SYN_RECEIVED:
2151 so->so_error = ECONNREFUSED;
2153 case TCPS_ESTABLISHED:
2154 case TCPS_FIN_WAIT_1:
2155 case TCPS_FIN_WAIT_2:
2156 case TCPS_CLOSE_WAIT:
2159 so->so_error = ECONNRESET;
2166 TCPSTAT_INC(tcps_badrst);
2167 /* Send challenge ACK. */
2168 tcp_respond(tp, mtod(m, void *), th, m,
2169 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2170 tp->last_ack_sent = tp->rcv_nxt;
2178 * RFC5961 Section 4.2
2179 * Send challenge ACK for any SYN in synchronized state.
2181 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2182 tp->t_state != TCPS_SYN_RECEIVED) {
2183 TCPSTAT_INC(tcps_badsyn);
2184 if (V_tcp_insecure_syn &&
2185 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2186 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2187 tp = tcp_drop(tp, ECONNRESET);
2188 rstreason = BANDLIM_UNLIMITED;
2190 /* Send challenge ACK. */
2191 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2192 tp->snd_nxt, TH_ACK);
2193 tp->last_ack_sent = tp->rcv_nxt;
2200 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2201 * and it's less than ts_recent, drop it.
2203 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2204 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2205 /* Check to see if ts_recent is over 24 days old. */
2206 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2208 * Invalidate ts_recent. If this segment updates
2209 * ts_recent, the age will be reset later and ts_recent
2210 * will get a valid value. If it does not, setting
2211 * ts_recent to zero will at least satisfy the
2212 * requirement that zero be placed in the timestamp
2213 * echo reply when ts_recent isn't valid. The
2214 * age isn't reset until we get a valid ts_recent
2215 * because we don't want out-of-order segments to be
2216 * dropped when ts_recent is old.
2220 TCPSTAT_INC(tcps_rcvduppack);
2221 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2222 TCPSTAT_INC(tcps_pawsdrop);
2230 * In the SYN-RECEIVED state, validate that the packet belongs to
2231 * this connection before trimming the data to fit the receive
2232 * window. Check the sequence number versus IRS since we know
2233 * the sequence numbers haven't wrapped. This is a partial fix
2234 * for the "LAND" DoS attack.
2236 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2237 rstreason = BANDLIM_RST_OPENPORT;
2241 todrop = tp->rcv_nxt - th->th_seq;
2243 if (thflags & TH_SYN) {
2253 * Following if statement from Stevens, vol. 2, p. 960.
2256 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2258 * Any valid FIN must be to the left of the window.
2259 * At this point the FIN must be a duplicate or out
2260 * of sequence; drop it.
2265 * Send an ACK to resynchronize and drop any data.
2266 * But keep on processing for RST or ACK.
2268 tp->t_flags |= TF_ACKNOW;
2270 TCPSTAT_INC(tcps_rcvduppack);
2271 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2273 TCPSTAT_INC(tcps_rcvpartduppack);
2274 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2277 * DSACK - add SACK block for dropped range
2279 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2280 tcp_update_sack_list(tp, th->th_seq,
2281 th->th_seq + todrop);
2283 * ACK now, as the next in-sequence segment
2284 * will clear the DSACK block again
2286 tp->t_flags |= TF_ACKNOW;
2288 drop_hdrlen += todrop; /* drop from the top afterwards */
2289 th->th_seq += todrop;
2291 if (th->th_urp > todrop)
2292 th->th_urp -= todrop;
2300 * If new data are received on a connection after the
2301 * user processes are gone, then RST the other end.
2303 if ((so->so_state & SS_NOFDREF) &&
2304 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2305 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2306 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2307 "after socket was closed, "
2308 "sending RST and removing tcpcb\n",
2309 s, __func__, tcpstates[tp->t_state], tlen);
2313 TCPSTAT_INC(tcps_rcvafterclose);
2314 rstreason = BANDLIM_UNLIMITED;
2319 * If segment ends after window, drop trailing data
2320 * (and PUSH and FIN); if nothing left, just ACK.
2322 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2324 TCPSTAT_INC(tcps_rcvpackafterwin);
2325 if (todrop >= tlen) {
2326 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2328 * If window is closed can only take segments at
2329 * window edge, and have to drop data and PUSH from
2330 * incoming segments. Continue processing, but
2331 * remember to ack. Otherwise, drop segment
2334 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2335 tp->t_flags |= TF_ACKNOW;
2336 TCPSTAT_INC(tcps_rcvwinprobe);
2340 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2343 thflags &= ~(TH_PUSH|TH_FIN);
2347 * If last ACK falls within this segment's sequence numbers,
2348 * record its timestamp.
2350 * 1) That the test incorporates suggestions from the latest
2351 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2352 * 2) That updating only on newer timestamps interferes with
2353 * our earlier PAWS tests, so this check should be solely
2354 * predicated on the sequence space of this segment.
2355 * 3) That we modify the segment boundary check to be
2356 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2357 * instead of RFC1323's
2358 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2359 * This modified check allows us to overcome RFC1323's
2360 * limitations as described in Stevens TCP/IP Illustrated
2361 * Vol. 2 p.869. In such cases, we can still calculate the
2362 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2364 if ((to.to_flags & TOF_TS) != 0 &&
2365 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2366 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2367 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2368 tp->ts_recent_age = tcp_ts_getticks();
2369 tp->ts_recent = to.to_tsval;
2373 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2374 * flag is on (half-synchronized state), then queue data for
2375 * later processing; else drop segment and return.
2377 if ((thflags & TH_ACK) == 0) {
2378 if (tp->t_state == TCPS_SYN_RECEIVED ||
2379 (tp->t_flags & TF_NEEDSYN)) {
2380 if (tp->t_state == TCPS_SYN_RECEIVED &&
2381 IS_FASTOPEN(tp->t_flags)) {
2382 tp->snd_wnd = tiwin;
2386 } else if (tp->t_flags & TF_ACKNOW)
2395 switch (tp->t_state) {
2397 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2398 * ESTABLISHED state and continue processing.
2399 * The ACK was checked above.
2401 case TCPS_SYN_RECEIVED:
2403 TCPSTAT_INC(tcps_connects);
2405 /* Do window scaling? */
2406 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2407 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2408 tp->rcv_scale = tp->request_r_scale;
2410 tp->snd_wnd = tiwin;
2413 * SYN-RECEIVED -> ESTABLISHED
2414 * SYN-RECEIVED* -> FIN-WAIT-1
2416 tp->t_starttime = ticks;
2417 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2418 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2419 tp->t_tfo_pending = NULL;
2421 if (tp->t_flags & TF_NEEDFIN) {
2422 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2423 tp->t_flags &= ~TF_NEEDFIN;
2425 tcp_state_change(tp, TCPS_ESTABLISHED);
2426 TCP_PROBE5(accept__established, NULL, tp,
2429 * TFO connections call cc_conn_init() during SYN
2430 * processing. Calling it again here for such
2431 * connections is not harmless as it would undo the
2432 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2435 if (!IS_FASTOPEN(tp->t_flags))
2437 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2440 * Account for the ACK of our SYN prior to
2441 * regular ACK processing below, except for
2442 * simultaneous SYN, which is handled later.
2444 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2447 * If segment contains data or ACK, will call tcp_reass()
2448 * later; if not, do so now to pass queued data to user.
2450 if (tlen == 0 && (thflags & TH_FIN) == 0)
2451 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2453 tp->snd_wl1 = th->th_seq - 1;
2457 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2458 * ACKs. If the ack is in the range
2459 * tp->snd_una < th->th_ack <= tp->snd_max
2460 * then advance tp->snd_una to th->th_ack and drop
2461 * data from the retransmission queue. If this ACK reflects
2462 * more up to date window information we update our window information.
2464 case TCPS_ESTABLISHED:
2465 case TCPS_FIN_WAIT_1:
2466 case TCPS_FIN_WAIT_2:
2467 case TCPS_CLOSE_WAIT:
2470 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2471 TCPSTAT_INC(tcps_rcvacktoomuch);
2474 if ((tp->t_flags & TF_SACK_PERMIT) &&
2475 ((to.to_flags & TOF_SACK) ||
2476 !TAILQ_EMPTY(&tp->snd_holes)))
2477 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2480 * Reset the value so that previous (valid) value
2481 * from the last ack with SACK doesn't get used.
2483 tp->sackhint.sacked_bytes = 0;
2486 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2487 hhook_run_tcp_est_in(tp, th, &to);
2490 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2493 maxseg = tcp_maxseg(tp);
2495 (tiwin == tp->snd_wnd ||
2496 (tp->t_flags & TF_SACK_PERMIT))) {
2498 * If this is the first time we've seen a
2499 * FIN from the remote, this is not a
2500 * duplicate and it needs to be processed
2501 * normally. This happens during a
2502 * simultaneous close.
2504 if ((thflags & TH_FIN) &&
2505 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2509 TCPSTAT_INC(tcps_rcvdupack);
2511 * If we have outstanding data (other than
2512 * a window probe), this is a completely
2513 * duplicate ack (ie, window info didn't
2514 * change and FIN isn't set),
2515 * the ack is the biggest we've
2516 * seen and we've seen exactly our rexmt
2517 * threshold of them, assume a packet
2518 * has been dropped and retransmit it.
2519 * Kludge snd_nxt & the congestion
2520 * window so we send only this one
2523 * We know we're losing at the current
2524 * window size so do congestion avoidance
2525 * (set ssthresh to half the current window
2526 * and pull our congestion window back to
2527 * the new ssthresh).
2529 * Dup acks mean that packets have left the
2530 * network (they're now cached at the receiver)
2531 * so bump cwnd by the amount in the receiver
2532 * to keep a constant cwnd packets in the
2535 * When using TCP ECN, notify the peer that
2536 * we reduced the cwnd.
2539 * Following 2 kinds of acks should not affect
2542 * 2) Acks with SACK but without any new SACK
2543 * information in them. These could result from
2544 * any anomaly in the network like a switch
2545 * duplicating packets or a possible DoS attack.
2547 if (th->th_ack != tp->snd_una ||
2548 ((tp->t_flags & TF_SACK_PERMIT) &&
2551 else if (!tcp_timer_active(tp, TT_REXMT))
2553 else if (++tp->t_dupacks > tcprexmtthresh ||
2554 IN_FASTRECOVERY(tp->t_flags)) {
2555 cc_ack_received(tp, th, nsegs,
2557 if ((tp->t_flags & TF_SACK_PERMIT) &&
2558 IN_FASTRECOVERY(tp->t_flags)) {
2562 * Compute the amount of data in flight first.
2563 * We can inject new data into the pipe iff
2564 * we have less than 1/2 the original window's
2565 * worth of data in flight.
2567 if (V_tcp_do_rfc6675_pipe)
2568 awnd = tcp_compute_pipe(tp);
2570 awnd = (tp->snd_nxt - tp->snd_fack) +
2571 tp->sackhint.sack_bytes_rexmit;
2573 if (awnd < tp->snd_ssthresh) {
2574 tp->snd_cwnd += maxseg;
2575 if (tp->snd_cwnd > tp->snd_ssthresh)
2576 tp->snd_cwnd = tp->snd_ssthresh;
2579 tp->snd_cwnd += maxseg;
2580 (void) tp->t_fb->tfb_tcp_output(tp);
2582 } else if (tp->t_dupacks == tcprexmtthresh) {
2583 tcp_seq onxt = tp->snd_nxt;
2586 * If we're doing sack, check to
2587 * see if we're already in sack
2588 * recovery. If we're not doing sack,
2589 * check to see if we're in newreno
2592 if (tp->t_flags & TF_SACK_PERMIT) {
2593 if (IN_FASTRECOVERY(tp->t_flags)) {
2598 if (SEQ_LEQ(th->th_ack,
2604 /* Congestion signal before ack. */
2605 cc_cong_signal(tp, th, CC_NDUPACK);
2606 cc_ack_received(tp, th, nsegs,
2608 tcp_timer_activate(tp, TT_REXMT, 0);
2610 if (tp->t_flags & TF_SACK_PERMIT) {
2612 tcps_sack_recovery_episode);
2613 tp->snd_recover = tp->snd_nxt;
2614 tp->snd_cwnd = maxseg;
2615 (void) tp->t_fb->tfb_tcp_output(tp);
2618 tp->snd_nxt = th->th_ack;
2619 tp->snd_cwnd = maxseg;
2620 (void) tp->t_fb->tfb_tcp_output(tp);
2621 KASSERT(tp->snd_limited <= 2,
2622 ("%s: tp->snd_limited too big",
2624 tp->snd_cwnd = tp->snd_ssthresh +
2626 (tp->t_dupacks - tp->snd_limited);
2627 if (SEQ_GT(onxt, tp->snd_nxt))
2630 } else if (V_tcp_do_rfc3042) {
2632 * Process first and second duplicate
2633 * ACKs. Each indicates a segment
2634 * leaving the network, creating room
2635 * for more. Make sure we can send a
2636 * packet on reception of each duplicate
2637 * ACK by increasing snd_cwnd by one
2638 * segment. Restore the original
2639 * snd_cwnd after packet transmission.
2641 cc_ack_received(tp, th, nsegs,
2643 uint32_t oldcwnd = tp->snd_cwnd;
2644 tcp_seq oldsndmax = tp->snd_max;
2648 KASSERT(tp->t_dupacks == 1 ||
2650 ("%s: dupacks not 1 or 2",
2652 if (tp->t_dupacks == 1)
2653 tp->snd_limited = 0;
2655 (tp->snd_nxt - tp->snd_una) +
2656 (tp->t_dupacks - tp->snd_limited) *
2659 * Only call tcp_output when there
2660 * is new data available to be sent.
2661 * Otherwise we would send pure ACKs.
2663 SOCKBUF_LOCK(&so->so_snd);
2664 avail = sbavail(&so->so_snd) -
2665 (tp->snd_nxt - tp->snd_una);
2666 SOCKBUF_UNLOCK(&so->so_snd);
2668 (void) tp->t_fb->tfb_tcp_output(tp);
2669 sent = tp->snd_max - oldsndmax;
2670 if (sent > maxseg) {
2671 KASSERT((tp->t_dupacks == 2 &&
2672 tp->snd_limited == 0) ||
2673 (sent == maxseg + 1 &&
2674 tp->t_flags & TF_SENTFIN),
2675 ("%s: sent too much",
2677 tp->snd_limited = 2;
2678 } else if (sent > 0)
2680 tp->snd_cwnd = oldcwnd;
2687 * This ack is advancing the left edge, reset the
2692 * If this ack also has new SACK info, increment the
2693 * counter as per rfc6675. The variable
2694 * sack_changed tracks all changes to the SACK
2695 * scoreboard, including when partial ACKs without
2696 * SACK options are received, and clear the scoreboard
2697 * from the left side. Such partial ACKs should not be
2698 * counted as dupacks here.
2700 if ((tp->t_flags & TF_SACK_PERMIT) &&
2701 (to.to_flags & TOF_SACK) &&
2706 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2707 ("%s: th_ack <= snd_una", __func__));
2710 * If the congestion window was inflated to account
2711 * for the other side's cached packets, retract it.
2713 if (IN_FASTRECOVERY(tp->t_flags)) {
2714 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2715 if (tp->t_flags & TF_SACK_PERMIT)
2716 tcp_sack_partialack(tp, th);
2718 tcp_newreno_partial_ack(tp, th);
2720 cc_post_recovery(tp, th);
2723 * If we reach this point, ACK is not a duplicate,
2724 * i.e., it ACKs something we sent.
2726 if (tp->t_flags & TF_NEEDSYN) {
2728 * T/TCP: Connection was half-synchronized, and our
2729 * SYN has been ACK'd (so connection is now fully
2730 * synchronized). Go to non-starred state,
2731 * increment snd_una for ACK of SYN, and check if
2732 * we can do window scaling.
2734 tp->t_flags &= ~TF_NEEDSYN;
2736 /* Do window scaling? */
2737 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2738 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2739 tp->rcv_scale = tp->request_r_scale;
2740 /* Send window already scaled. */
2745 INP_WLOCK_ASSERT(tp->t_inpcb);
2748 * Adjust for the SYN bit in sequence space,
2749 * but don't account for it in cwnd calculations.
2750 * This is for the SYN_RECEIVED, non-simultaneous
2751 * SYN case. SYN_SENT and simultaneous SYN are
2752 * treated elsewhere.
2756 acked = BYTES_THIS_ACK(tp, th);
2757 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2758 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2759 tp->snd_una, th->th_ack, tp, m));
2760 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2761 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2764 * If we just performed our first retransmit, and the ACK
2765 * arrives within our recovery window, then it was a mistake
2766 * to do the retransmit in the first place. Recover our
2767 * original cwnd and ssthresh, and proceed to transmit where
2770 if (tp->t_rxtshift == 1 &&
2771 tp->t_flags & TF_PREVVALID &&
2773 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2774 cc_cong_signal(tp, th, CC_RTO_ERR);
2777 * If we have a timestamp reply, update smoothed
2778 * round trip time. If no timestamp is present but
2779 * transmit timer is running and timed sequence
2780 * number was acked, update smoothed round trip time.
2781 * Since we now have an rtt measurement, cancel the
2782 * timer backoff (cf., Phil Karn's retransmit alg.).
2783 * Recompute the initial retransmit timer.
2785 * Some boxes send broken timestamp replies
2786 * during the SYN+ACK phase, ignore
2787 * timestamps of 0 or we could calculate a
2788 * huge RTT and blow up the retransmit timer.
2790 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2793 t = tcp_ts_getticks() - to.to_tsecr;
2794 if (!tp->t_rttlow || tp->t_rttlow > t)
2796 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2797 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2798 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2799 tp->t_rttlow = ticks - tp->t_rtttime;
2800 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2804 * If all outstanding data is acked, stop retransmit
2805 * timer and remember to restart (more output or persist).
2806 * If there is more data to be acked, restart retransmit
2807 * timer, using current (possibly backed-off) value.
2809 if (th->th_ack == tp->snd_max) {
2810 tcp_timer_activate(tp, TT_REXMT, 0);
2812 } else if (!tcp_timer_active(tp, TT_PERSIST))
2813 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2816 * If no data (only SYN) was ACK'd,
2817 * skip rest of ACK processing.
2823 * Let the congestion control algorithm update congestion
2824 * control related information. This typically means increasing
2825 * the congestion window.
2827 cc_ack_received(tp, th, nsegs, CC_ACK);
2829 SOCKBUF_LOCK(&so->so_snd);
2830 if (acked > sbavail(&so->so_snd)) {
2831 if (tp->snd_wnd >= sbavail(&so->so_snd))
2832 tp->snd_wnd -= sbavail(&so->so_snd);
2835 mfree = sbcut_locked(&so->so_snd,
2836 (int)sbavail(&so->so_snd));
2839 mfree = sbcut_locked(&so->so_snd, acked);
2840 if (tp->snd_wnd >= (uint32_t) acked)
2841 tp->snd_wnd -= acked;
2846 SOCKBUF_UNLOCK(&so->so_snd);
2847 tp->t_flags |= TF_WAKESOW;
2849 /* Detect una wraparound. */
2850 if (!IN_RECOVERY(tp->t_flags) &&
2851 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2852 SEQ_LEQ(th->th_ack, tp->snd_recover))
2853 tp->snd_recover = th->th_ack - 1;
2854 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2855 if (IN_RECOVERY(tp->t_flags) &&
2856 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2857 EXIT_RECOVERY(tp->t_flags);
2859 tp->snd_una = th->th_ack;
2860 if (tp->t_flags & TF_SACK_PERMIT) {
2861 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2862 tp->snd_recover = tp->snd_una;
2864 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2865 tp->snd_nxt = tp->snd_una;
2867 switch (tp->t_state) {
2869 * In FIN_WAIT_1 STATE in addition to the processing
2870 * for the ESTABLISHED state if our FIN is now acknowledged
2871 * then enter FIN_WAIT_2.
2873 case TCPS_FIN_WAIT_1:
2874 if (ourfinisacked) {
2876 * If we can't receive any more
2877 * data, then closing user can proceed.
2878 * Starting the timer is contrary to the
2879 * specification, but if we don't get a FIN
2880 * we'll hang forever.
2883 * we should release the tp also, and use a
2886 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2887 soisdisconnected(so);
2888 tcp_timer_activate(tp, TT_2MSL,
2889 (tcp_fast_finwait2_recycle ?
2890 tcp_finwait2_timeout :
2893 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2898 * In CLOSING STATE in addition to the processing for
2899 * the ESTABLISHED state if the ACK acknowledges our FIN
2900 * then enter the TIME-WAIT state, otherwise ignore
2904 if (ourfinisacked) {
2912 * In LAST_ACK, we may still be waiting for data to drain
2913 * and/or to be acked, as well as for the ack of our FIN.
2914 * If our FIN is now acknowledged, delete the TCB,
2915 * enter the closed state and return.
2918 if (ourfinisacked) {
2927 INP_WLOCK_ASSERT(tp->t_inpcb);
2930 * Update window information.
2931 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2933 if ((thflags & TH_ACK) &&
2934 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2935 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2936 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2937 /* keep track of pure window updates */
2939 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2940 TCPSTAT_INC(tcps_rcvwinupd);
2941 tp->snd_wnd = tiwin;
2942 tp->snd_wl1 = th->th_seq;
2943 tp->snd_wl2 = th->th_ack;
2944 if (tp->snd_wnd > tp->max_sndwnd)
2945 tp->max_sndwnd = tp->snd_wnd;
2950 * Process segments with URG.
2952 if ((thflags & TH_URG) && th->th_urp &&
2953 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2955 * This is a kludge, but if we receive and accept
2956 * random urgent pointers, we'll crash in
2957 * soreceive. It's hard to imagine someone
2958 * actually wanting to send this much urgent data.
2960 SOCKBUF_LOCK(&so->so_rcv);
2961 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2962 th->th_urp = 0; /* XXX */
2963 thflags &= ~TH_URG; /* XXX */
2964 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2965 goto dodata; /* XXX */
2968 * If this segment advances the known urgent pointer,
2969 * then mark the data stream. This should not happen
2970 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2971 * a FIN has been received from the remote side.
2972 * In these states we ignore the URG.
2974 * According to RFC961 (Assigned Protocols),
2975 * the urgent pointer points to the last octet
2976 * of urgent data. We continue, however,
2977 * to consider it to indicate the first octet
2978 * of data past the urgent section as the original
2979 * spec states (in one of two places).
2981 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2982 tp->rcv_up = th->th_seq + th->th_urp;
2983 so->so_oobmark = sbavail(&so->so_rcv) +
2984 (tp->rcv_up - tp->rcv_nxt) - 1;
2985 if (so->so_oobmark == 0)
2986 so->so_rcv.sb_state |= SBS_RCVATMARK;
2988 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2990 SOCKBUF_UNLOCK(&so->so_rcv);
2992 * Remove out of band data so doesn't get presented to user.
2993 * This can happen independent of advancing the URG pointer,
2994 * but if two URG's are pending at once, some out-of-band
2995 * data may creep in... ick.
2997 if (th->th_urp <= (uint32_t)tlen &&
2998 !(so->so_options & SO_OOBINLINE)) {
2999 /* hdr drop is delayed */
3000 tcp_pulloutofband(so, th, m, drop_hdrlen);
3004 * If no out of band data is expected,
3005 * pull receive urgent pointer along
3006 * with the receive window.
3008 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3009 tp->rcv_up = tp->rcv_nxt;
3012 INP_WLOCK_ASSERT(tp->t_inpcb);
3015 * Process the segment text, merging it into the TCP sequencing queue,
3016 * and arranging for acknowledgment of receipt if necessary.
3017 * This process logically involves adjusting tp->rcv_wnd as data
3018 * is presented to the user (this happens in tcp_usrreq.c,
3019 * case PRU_RCVD). If a FIN has already been received on this
3020 * connection then we just ignore the text.
3022 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3023 IS_FASTOPEN(tp->t_flags));
3024 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3025 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3026 tcp_seq save_start = th->th_seq;
3027 tcp_seq save_rnxt = tp->rcv_nxt;
3028 int save_tlen = tlen;
3029 m_adj(m, drop_hdrlen); /* delayed header drop */
3031 * Insert segment which includes th into TCP reassembly queue
3032 * with control block tp. Set thflags to whether reassembly now
3033 * includes a segment with FIN. This handles the common case
3034 * inline (segment is the next to be received on an established
3035 * connection, and the queue is empty), avoiding linkage into
3036 * and removal from the queue and repetition of various
3038 * Set DELACK for segments received in order, but ack
3039 * immediately when segments are out of order (so
3040 * fast retransmit can work).
3042 if (th->th_seq == tp->rcv_nxt &&
3044 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3046 if (DELAY_ACK(tp, tlen) || tfo_syn)
3047 tp->t_flags |= TF_DELACK;
3049 tp->t_flags |= TF_ACKNOW;
3050 tp->rcv_nxt += tlen;
3052 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3053 (tp->t_fbyte_in == 0)) {
3054 tp->t_fbyte_in = ticks;
3055 if (tp->t_fbyte_in == 0)
3057 if (tp->t_fbyte_out && tp->t_fbyte_in)
3058 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3060 thflags = th->th_flags & TH_FIN;
3061 TCPSTAT_INC(tcps_rcvpack);
3062 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3063 SOCKBUF_LOCK(&so->so_rcv);
3064 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3067 sbappendstream_locked(&so->so_rcv, m, 0);
3068 SOCKBUF_UNLOCK(&so->so_rcv);
3069 tp->t_flags |= TF_WAKESOR;
3072 * XXX: Due to the header drop above "th" is
3073 * theoretically invalid by now. Fortunately
3074 * m_adj() doesn't actually frees any mbufs
3075 * when trimming from the head.
3077 tcp_seq temp = save_start;
3078 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3079 tp->t_flags |= TF_ACKNOW;
3081 if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) {
3082 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3084 * DSACK actually handled in the fastpath
3087 tcp_update_sack_list(tp, save_start,
3088 save_start + save_tlen);
3089 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3090 if ((tp->rcv_numsacks >= 1) &&
3091 (tp->sackblks[0].end == save_start)) {
3093 * Partial overlap, recorded at todrop
3096 tcp_update_sack_list(tp,
3097 tp->sackblks[0].start,
3098 tp->sackblks[0].end);
3100 tcp_update_dsack_list(tp, save_start,
3101 save_start + save_tlen);
3103 } else if (tlen >= save_tlen) {
3104 /* Update of sackblks. */
3105 tcp_update_dsack_list(tp, save_start,
3106 save_start + save_tlen);
3107 } else if (tlen > 0) {
3108 tcp_update_dsack_list(tp, save_start,
3114 * Note the amount of data that peer has sent into
3115 * our window, in order to estimate the sender's
3119 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3120 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3122 len = so->so_rcv.sb_hiwat;
3130 * If FIN is received ACK the FIN and let the user know
3131 * that the connection is closing.
3133 if (thflags & TH_FIN) {
3134 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3136 /* The socket upcall is handled by socantrcvmore. */
3137 tp->t_flags &= ~TF_WAKESOR;
3139 * If connection is half-synchronized
3140 * (ie NEEDSYN flag on) then delay ACK,
3141 * so it may be piggybacked when SYN is sent.
3142 * Otherwise, since we received a FIN then no
3143 * more input can be expected, send ACK now.
3145 if (tp->t_flags & TF_NEEDSYN)
3146 tp->t_flags |= TF_DELACK;
3148 tp->t_flags |= TF_ACKNOW;
3151 switch (tp->t_state) {
3153 * In SYN_RECEIVED and ESTABLISHED STATES
3154 * enter the CLOSE_WAIT state.
3156 case TCPS_SYN_RECEIVED:
3157 tp->t_starttime = ticks;
3159 case TCPS_ESTABLISHED:
3160 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3164 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3165 * enter the CLOSING state.
3167 case TCPS_FIN_WAIT_1:
3168 tcp_state_change(tp, TCPS_CLOSING);
3172 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3173 * starting the time-wait timer, turning off the other
3176 case TCPS_FIN_WAIT_2:
3182 if (so->so_options & SO_DEBUG)
3183 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3186 TCP_PROBE3(debug__input, tp, th, m);
3189 * Return any desired output.
3191 if (needoutput || (tp->t_flags & TF_ACKNOW))
3192 (void) tp->t_fb->tfb_tcp_output(tp);
3195 INP_WLOCK_ASSERT(tp->t_inpcb);
3197 if (tp->t_flags & TF_DELACK) {
3198 tp->t_flags &= ~TF_DELACK;
3199 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3201 tcp_handle_wakeup(tp, so);
3202 INP_WUNLOCK(tp->t_inpcb);
3207 * Generate an ACK dropping incoming segment if it occupies
3208 * sequence space, where the ACK reflects our state.
3210 * We can now skip the test for the RST flag since all
3211 * paths to this code happen after packets containing
3212 * RST have been dropped.
3214 * In the SYN-RECEIVED state, don't send an ACK unless the
3215 * segment we received passes the SYN-RECEIVED ACK test.
3216 * If it fails send a RST. This breaks the loop in the
3217 * "LAND" DoS attack, and also prevents an ACK storm
3218 * between two listening ports that have been sent forged
3219 * SYN segments, each with the source address of the other.
3221 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3222 (SEQ_GT(tp->snd_una, th->th_ack) ||
3223 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3224 rstreason = BANDLIM_RST_OPENPORT;
3228 if (so->so_options & SO_DEBUG)
3229 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3232 TCP_PROBE3(debug__input, tp, th, m);
3233 tp->t_flags |= TF_ACKNOW;
3234 (void) tp->t_fb->tfb_tcp_output(tp);
3235 tcp_handle_wakeup(tp, so);
3236 INP_WUNLOCK(tp->t_inpcb);
3242 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3243 tcp_handle_wakeup(tp, so);
3244 INP_WUNLOCK(tp->t_inpcb);
3246 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3251 * Drop space held by incoming segment and return.
3254 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3255 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3258 TCP_PROBE3(debug__input, tp, th, m);
3260 tcp_handle_wakeup(tp, so);
3261 INP_WUNLOCK(tp->t_inpcb);
3267 * Issue RST and make ACK acceptable to originator of segment.
3268 * The mbuf must still include the original packet header.
3272 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3273 int tlen, int rstreason)
3279 struct ip6_hdr *ip6;
3283 INP_WLOCK_ASSERT(tp->t_inpcb);
3286 /* Don't bother if destination was broadcast/multicast. */
3287 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3290 if (mtod(m, struct ip *)->ip_v == 6) {
3291 ip6 = mtod(m, struct ip6_hdr *);
3292 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3293 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3295 /* IPv6 anycast check is done at tcp6_input() */
3298 #if defined(INET) && defined(INET6)
3303 ip = mtod(m, struct ip *);
3304 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3305 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3306 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3307 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3312 /* Perform bandwidth limiting. */
3313 if (badport_bandlim(rstreason) < 0)
3316 /* tcp_respond consumes the mbuf chain. */
3317 if (th->th_flags & TH_ACK) {
3318 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3319 th->th_ack, TH_RST);
3321 if (th->th_flags & TH_SYN)
3323 if (th->th_flags & TH_FIN)
3325 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3326 (tcp_seq)0, TH_RST|TH_ACK);
3334 * Parse TCP options and place in tcpopt.
3337 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3342 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3344 if (opt == TCPOPT_EOL)
3346 if (opt == TCPOPT_NOP)
3352 if (optlen < 2 || optlen > cnt)
3357 if (optlen != TCPOLEN_MAXSEG)
3359 if (!(flags & TO_SYN))
3361 to->to_flags |= TOF_MSS;
3362 bcopy((char *)cp + 2,
3363 (char *)&to->to_mss, sizeof(to->to_mss));
3364 to->to_mss = ntohs(to->to_mss);
3367 if (optlen != TCPOLEN_WINDOW)
3369 if (!(flags & TO_SYN))
3371 to->to_flags |= TOF_SCALE;
3372 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3374 case TCPOPT_TIMESTAMP:
3375 if (optlen != TCPOLEN_TIMESTAMP)
3377 to->to_flags |= TOF_TS;
3378 bcopy((char *)cp + 2,
3379 (char *)&to->to_tsval, sizeof(to->to_tsval));
3380 to->to_tsval = ntohl(to->to_tsval);
3381 bcopy((char *)cp + 6,
3382 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3383 to->to_tsecr = ntohl(to->to_tsecr);
3385 case TCPOPT_SIGNATURE:
3387 * In order to reply to a host which has set the
3388 * TCP_SIGNATURE option in its initial SYN, we have
3389 * to record the fact that the option was observed
3390 * here for the syncache code to perform the correct
3393 if (optlen != TCPOLEN_SIGNATURE)
3395 to->to_flags |= TOF_SIGNATURE;
3396 to->to_signature = cp + 2;
3398 case TCPOPT_SACK_PERMITTED:
3399 if (optlen != TCPOLEN_SACK_PERMITTED)
3401 if (!(flags & TO_SYN))
3405 to->to_flags |= TOF_SACKPERM;
3408 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3412 to->to_flags |= TOF_SACK;
3413 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3414 to->to_sacks = cp + 2;
3415 TCPSTAT_INC(tcps_sack_rcv_blocks);
3417 case TCPOPT_FAST_OPEN:
3419 * Cookie length validation is performed by the
3420 * server side cookie checking code or the client
3421 * side cookie cache update code.
3423 if (!(flags & TO_SYN))
3425 if (!V_tcp_fastopen_client_enable &&
3426 !V_tcp_fastopen_server_enable)
3428 to->to_flags |= TOF_FASTOPEN;
3429 to->to_tfo_len = optlen - 2;
3430 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3439 * Pull out of band byte out of a segment so
3440 * it doesn't appear in the user's data queue.
3441 * It is still reflected in the segment length for
3442 * sequencing purposes.
3445 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3448 int cnt = off + th->th_urp - 1;
3451 if (m->m_len > cnt) {
3452 char *cp = mtod(m, caddr_t) + cnt;
3453 struct tcpcb *tp = sototcpcb(so);
3455 INP_WLOCK_ASSERT(tp->t_inpcb);
3458 tp->t_oobflags |= TCPOOB_HAVEDATA;
3459 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3461 if (m->m_flags & M_PKTHDR)
3470 panic("tcp_pulloutofband");
3474 * Collect new round-trip time estimate
3475 * and update averages and current timeout.
3478 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3482 INP_WLOCK_ASSERT(tp->t_inpcb);
3484 TCPSTAT_INC(tcps_rttupdated);
3487 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3488 imax(0, rtt * 1000 / hz));
3490 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3492 * srtt is stored as fixed point with 5 bits after the
3493 * binary point (i.e., scaled by 8). The following magic
3494 * is equivalent to the smoothing algorithm in rfc793 with
3495 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3496 * point). Adjust rtt to origin 0.
3498 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3499 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3501 if ((tp->t_srtt += delta) <= 0)
3505 * We accumulate a smoothed rtt variance (actually, a
3506 * smoothed mean difference), then set the retransmit
3507 * timer to smoothed rtt + 4 times the smoothed variance.
3508 * rttvar is stored as fixed point with 4 bits after the
3509 * binary point (scaled by 16). The following is
3510 * equivalent to rfc793 smoothing with an alpha of .75
3511 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3512 * rfc793's wired-in beta.
3516 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3517 if ((tp->t_rttvar += delta) <= 0)
3519 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3520 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3523 * No rtt measurement yet - use the unsmoothed rtt.
3524 * Set the variance to half the rtt (so our first
3525 * retransmit happens at 3*rtt).
3527 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3528 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3529 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3535 * the retransmit should happen at rtt + 4 * rttvar.
3536 * Because of the way we do the smoothing, srtt and rttvar
3537 * will each average +1/2 tick of bias. When we compute
3538 * the retransmit timer, we want 1/2 tick of rounding and
3539 * 1 extra tick because of +-1/2 tick uncertainty in the
3540 * firing of the timer. The bias will give us exactly the
3541 * 1.5 tick we need. But, because the bias is
3542 * statistical, we have to test that we don't drop below
3543 * the minimum feasible timer (which is 2 ticks).
3545 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3546 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3549 * We received an ack for a packet that wasn't retransmitted;
3550 * it is probably safe to discard any error indications we've
3551 * received recently. This isn't quite right, but close enough
3552 * for now (a route might have failed after we sent a segment,
3553 * and the return path might not be symmetrical).
3555 tp->t_softerror = 0;
3559 * Determine a reasonable value for maxseg size.
3560 * If the route is known, check route for mtu.
3561 * If none, use an mss that can be handled on the outgoing interface
3562 * without forcing IP to fragment. If no route is found, route has no mtu,
3563 * or the destination isn't local, use a default, hopefully conservative
3564 * size (usually 512 or the default IP max size, but no more than the mtu
3565 * of the interface), as we can't discover anything about intervening
3566 * gateways or networks. We also initialize the congestion/slow start
3567 * window to be a single segment if the destination isn't local.
3568 * While looking at the routing entry, we also initialize other path-dependent
3569 * parameters from pre-set or cached values in the routing entry.
3571 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3572 * IP options, e.g. IPSEC data, since length of this data may vary, and
3573 * thus it is calculated for every segment separately in tcp_output().
3575 * NOTE that this routine is only called when we process an incoming
3576 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3577 * settings are handled in tcp_mssopt().
3580 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3581 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3584 uint32_t maxmtu = 0;
3585 struct inpcb *inp = tp->t_inpcb;
3586 struct hc_metrics_lite metrics;
3588 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3589 size_t min_protoh = isipv6 ?
3590 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3591 sizeof (struct tcpiphdr);
3593 const size_t min_protoh = sizeof(struct tcpiphdr);
3596 INP_WLOCK_ASSERT(tp->t_inpcb);
3598 if (mtuoffer != -1) {
3599 KASSERT(offer == -1, ("%s: conflict", __func__));
3600 offer = mtuoffer - min_protoh;
3606 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3607 tp->t_maxseg = V_tcp_v6mssdflt;
3610 #if defined(INET) && defined(INET6)
3615 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3616 tp->t_maxseg = V_tcp_mssdflt;
3621 * No route to sender, stay with default mss and return.
3625 * In case we return early we need to initialize metrics
3626 * to a defined state as tcp_hc_get() would do for us
3627 * if there was no cache hit.
3629 if (metricptr != NULL)
3630 bzero(metricptr, sizeof(struct hc_metrics_lite));
3634 /* What have we got? */
3638 * Offer == 0 means that there was no MSS on the SYN
3639 * segment, in this case we use tcp_mssdflt as
3640 * already assigned to t_maxseg above.
3642 offer = tp->t_maxseg;
3647 * Offer == -1 means that we didn't receive SYN yet.
3653 * Prevent DoS attack with too small MSS. Round up
3654 * to at least minmss.
3656 offer = max(offer, V_tcp_minmss);
3660 * rmx information is now retrieved from tcp_hostcache.
3662 tcp_hc_get(&inp->inp_inc, &metrics);
3663 if (metricptr != NULL)
3664 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3667 * If there's a discovered mtu in tcp hostcache, use it.
3668 * Else, use the link mtu.
3670 if (metrics.rmx_mtu)
3671 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3675 mss = maxmtu - min_protoh;
3676 if (!V_path_mtu_discovery &&
3677 !in6_localaddr(&inp->in6p_faddr))
3678 mss = min(mss, V_tcp_v6mssdflt);
3681 #if defined(INET) && defined(INET6)
3686 mss = maxmtu - min_protoh;
3687 if (!V_path_mtu_discovery &&
3688 !in_localaddr(inp->inp_faddr))
3689 mss = min(mss, V_tcp_mssdflt);
3693 * XXX - The above conditional (mss = maxmtu - min_protoh)
3694 * probably violates the TCP spec.
3695 * The problem is that, since we don't know the
3696 * other end's MSS, we are supposed to use a conservative
3697 * default. But, if we do that, then MTU discovery will
3698 * never actually take place, because the conservative
3699 * default is much less than the MTUs typically seen
3700 * on the Internet today. For the moment, we'll sweep
3701 * this under the carpet.
3703 * The conservative default might not actually be a problem
3704 * if the only case this occurs is when sending an initial
3705 * SYN with options and data to a host we've never talked
3706 * to before. Then, they will reply with an MSS value which
3707 * will get recorded and the new parameters should get
3708 * recomputed. For Further Study.
3711 mss = min(mss, offer);
3714 * Sanity check: make sure that maxseg will be large
3715 * enough to allow some data on segments even if the
3716 * all the option space is used (40bytes). Otherwise
3717 * funny things may happen in tcp_output.
3719 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3727 tcp_mss(struct tcpcb *tp, int offer)
3733 struct hc_metrics_lite metrics;
3734 struct tcp_ifcap cap;
3736 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3738 bzero(&cap, sizeof(cap));
3739 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3745 * If there's a pipesize, change the socket buffer to that size,
3746 * don't change if sb_hiwat is different than default (then it
3747 * has been changed on purpose with setsockopt).
3748 * Make the socket buffers an integral number of mss units;
3749 * if the mss is larger than the socket buffer, decrease the mss.
3751 so = inp->inp_socket;
3752 SOCKBUF_LOCK(&so->so_snd);
3753 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3754 bufsize = metrics.rmx_sendpipe;
3756 bufsize = so->so_snd.sb_hiwat;
3760 bufsize = roundup(bufsize, mss);
3761 if (bufsize > sb_max)
3763 if (bufsize > so->so_snd.sb_hiwat)
3764 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3766 SOCKBUF_UNLOCK(&so->so_snd);
3768 * Sanity check: make sure that maxseg will be large
3769 * enough to allow some data on segments even if the
3770 * all the option space is used (40bytes). Otherwise
3771 * funny things may happen in tcp_output.
3773 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3775 tp->t_maxseg = max(mss, 64);
3777 SOCKBUF_LOCK(&so->so_rcv);
3778 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3779 bufsize = metrics.rmx_recvpipe;
3781 bufsize = so->so_rcv.sb_hiwat;
3782 if (bufsize > mss) {
3783 bufsize = roundup(bufsize, mss);
3784 if (bufsize > sb_max)
3786 if (bufsize > so->so_rcv.sb_hiwat)
3787 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3789 SOCKBUF_UNLOCK(&so->so_rcv);
3791 /* Check the interface for TSO capabilities. */
3792 if (cap.ifcap & CSUM_TSO) {
3793 tp->t_flags |= TF_TSO;
3794 tp->t_tsomax = cap.tsomax;
3795 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3796 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3801 * Determine the MSS option to send on an outgoing SYN.
3804 tcp_mssopt(struct in_conninfo *inc)
3807 uint32_t thcmtu = 0;
3808 uint32_t maxmtu = 0;
3811 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3814 if (inc->inc_flags & INC_ISIPV6) {
3815 mss = V_tcp_v6mssdflt;
3816 maxmtu = tcp_maxmtu6(inc, NULL);
3817 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3820 #if defined(INET) && defined(INET6)
3825 mss = V_tcp_mssdflt;
3826 maxmtu = tcp_maxmtu(inc, NULL);
3827 min_protoh = sizeof(struct tcpiphdr);
3830 #if defined(INET6) || defined(INET)
3831 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3834 if (maxmtu && thcmtu)
3835 mss = min(maxmtu, thcmtu) - min_protoh;
3836 else if (maxmtu || thcmtu)
3837 mss = max(maxmtu, thcmtu) - min_protoh;
3843 * On a partial ack arrives, force the retransmission of the
3844 * next unacknowledged segment. Do not clear tp->t_dupacks.
3845 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3849 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3851 tcp_seq onxt = tp->snd_nxt;
3852 uint32_t ocwnd = tp->snd_cwnd;
3853 u_int maxseg = tcp_maxseg(tp);
3855 INP_WLOCK_ASSERT(tp->t_inpcb);
3857 tcp_timer_activate(tp, TT_REXMT, 0);
3859 tp->snd_nxt = th->th_ack;
3861 * Set snd_cwnd to one segment beyond acknowledged offset.
3862 * (tp->snd_una has not yet been updated when this function is called.)
3864 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3865 tp->t_flags |= TF_ACKNOW;
3866 (void) tp->t_fb->tfb_tcp_output(tp);
3867 tp->snd_cwnd = ocwnd;
3868 if (SEQ_GT(onxt, tp->snd_nxt))
3871 * Partial window deflation. Relies on fact that tp->snd_una
3874 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3875 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3878 tp->snd_cwnd += maxseg;
3882 tcp_compute_pipe(struct tcpcb *tp)
3884 return (tp->snd_max - tp->snd_una +
3885 tp->sackhint.sack_bytes_rexmit -
3886 tp->sackhint.sacked_bytes);
3890 tcp_compute_initwnd(uint32_t maxseg)
3893 * Calculate the Initial Window, also used as Restart Window
3895 * RFC5681 Section 3.1 specifies the default conservative values.
3896 * RFC3390 specifies slightly more aggressive values.
3897 * RFC6928 increases it to ten segments.
3898 * Support for user specified value for initial flight size.
3900 if (V_tcp_initcwnd_segments)
3901 return min(V_tcp_initcwnd_segments * maxseg,
3902 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
3903 else if (V_tcp_do_rfc3390)
3904 return min(4 * maxseg, max(2 * maxseg, 4380));
3906 /* Per RFC5681 Section 3.1 */
3908 return (2 * maxseg);
3909 else if (maxseg > 1095)
3910 return (3 * maxseg);
3912 return (4 * maxseg);