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 int tcp_log_in_vain = 0;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
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, CTLFLAG_RW, 0, "TCP ECN");
193 VNET_DEFINE(int, tcp_do_ecn) = 2;
194 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
195 &VNET_NAME(tcp_do_ecn), 0,
198 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
199 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
200 &VNET_NAME(tcp_ecn_maxretries), 0,
201 "Max retries before giving up on ECN");
203 VNET_DEFINE(int, tcp_insecure_syn) = 0;
204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
205 &VNET_NAME(tcp_insecure_syn), 0,
206 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
208 VNET_DEFINE(int, tcp_insecure_rst) = 0;
209 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
210 &VNET_NAME(tcp_insecure_rst), 0,
211 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
213 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
214 #define V_tcp_recvspace VNET(tcp_recvspace)
215 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
216 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
218 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
219 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
220 &VNET_NAME(tcp_do_autorcvbuf), 0,
221 "Enable automatic receive buffer sizing");
223 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
225 &VNET_NAME(tcp_autorcvbuf_max), 0,
226 "Max size of automatic receive buffer");
228 VNET_DEFINE(struct inpcbhead, tcb);
229 #define tcb6 tcb /* for KAME src sync over BSD*'s */
230 VNET_DEFINE(struct inpcbinfo, tcbinfo);
233 * TCP statistics are stored in an array of counter(9)s, which size matches
234 * size of struct tcpstat. TCP running connection count is a regular array.
236 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
237 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
238 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
239 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
240 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
241 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
242 "TCP connection counts by TCP state");
245 tcp_vnet_init(const void *unused)
248 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
249 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
251 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
252 tcp_vnet_init, NULL);
256 tcp_vnet_uninit(const void *unused)
259 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
260 VNET_PCPUSTAT_FREE(tcpstat);
262 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
263 tcp_vnet_uninit, NULL);
267 * Kernel module interface for updating tcpstat. The argument is an index
268 * into tcpstat treated as an array.
271 kmod_tcpstat_inc(int statnum)
274 counter_u64_add(VNET(tcpstat)[statnum], 1);
279 * Wrapper for the TCP established input helper hook.
282 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
284 struct tcp_hhook_data hhook_data;
286 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
291 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
298 * CC wrapper hook functions
301 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
308 INP_WLOCK_ASSERT(tp->t_inpcb);
310 tp->ccv->nsegs = nsegs;
311 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
312 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
313 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
314 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
315 tp->ccv->flags |= CCF_CWND_LIMITED;
317 tp->ccv->flags &= ~CCF_CWND_LIMITED;
319 if (type == CC_ACK) {
321 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
322 ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
323 if (!IN_RECOVERY(tp->t_flags))
324 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
325 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs));
326 if ((tp->t_flags & TF_GPUTINPROG) &&
327 SEQ_GEQ(th->th_ack, tp->gput_ack)) {
329 * Compute goodput in bits per millisecond.
331 gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) /
332 max(1, tcp_ts_getticks() - tp->gput_ts);
333 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
336 * XXXLAS: This is a temporary hack, and should be
337 * chained off VOI_TCP_GPUT when stats(9) grows an API
338 * to deal with chained VOIs.
340 if (tp->t_stats_gput_prev > 0)
341 stats_voi_update_abs_s32(tp->t_stats,
343 ((gput - tp->t_stats_gput_prev) * 100) /
344 tp->t_stats_gput_prev);
345 tp->t_flags &= ~TF_GPUTINPROG;
346 tp->t_stats_gput_prev = gput;
349 if (tp->snd_cwnd > tp->snd_ssthresh) {
350 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
351 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp));
352 if (tp->t_bytes_acked >= tp->snd_cwnd) {
353 tp->t_bytes_acked -= tp->snd_cwnd;
354 tp->ccv->flags |= CCF_ABC_SENTAWND;
357 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
358 tp->t_bytes_acked = 0;
362 if (CC_ALGO(tp)->ack_received != NULL) {
363 /* XXXLAS: Find a way to live without this */
364 tp->ccv->curack = th->th_ack;
365 CC_ALGO(tp)->ack_received(tp->ccv, type);
368 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
373 cc_conn_init(struct tcpcb *tp)
375 struct hc_metrics_lite metrics;
376 struct inpcb *inp = tp->t_inpcb;
380 INP_WLOCK_ASSERT(tp->t_inpcb);
382 tcp_hc_get(&inp->inp_inc, &metrics);
383 maxseg = tcp_maxseg(tp);
385 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
387 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
388 TCPSTAT_INC(tcps_usedrtt);
389 if (metrics.rmx_rttvar) {
390 tp->t_rttvar = metrics.rmx_rttvar;
391 TCPSTAT_INC(tcps_usedrttvar);
393 /* default variation is +- 1 rtt */
395 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
397 TCPT_RANGESET(tp->t_rxtcur,
398 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
399 tp->t_rttmin, TCPTV_REXMTMAX);
401 if (metrics.rmx_ssthresh) {
403 * There's some sort of gateway or interface
404 * buffer limit on the path. Use this to set
405 * the slow start threshold, but set the
406 * threshold to no less than 2*mss.
408 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
409 TCPSTAT_INC(tcps_usedssthresh);
413 * Set the initial slow-start flight size.
415 * If a SYN or SYN/ACK was lost and retransmitted, we have to
416 * reduce the initial CWND to one segment as congestion is likely
417 * requiring us to be cautious.
419 if (tp->snd_cwnd == 1)
420 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
422 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
424 if (CC_ALGO(tp)->conn_init != NULL)
425 CC_ALGO(tp)->conn_init(tp->ccv);
429 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
433 INP_WLOCK_ASSERT(tp->t_inpcb);
436 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
441 if (!IN_FASTRECOVERY(tp->t_flags)) {
442 tp->snd_recover = tp->snd_max;
443 if (tp->t_flags2 & TF2_ECN_PERMIT)
444 tp->t_flags2 |= TF2_ECN_SND_CWR;
448 if (!IN_CONGRECOVERY(tp->t_flags)) {
449 TCPSTAT_INC(tcps_ecn_rcwnd);
450 tp->snd_recover = tp->snd_max;
451 if (tp->t_flags2 & TF2_ECN_PERMIT)
452 tp->t_flags2 |= TF2_ECN_SND_CWR;
456 maxseg = tcp_maxseg(tp);
458 tp->t_bytes_acked = 0;
459 EXIT_RECOVERY(tp->t_flags);
460 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
462 tp->snd_cwnd = maxseg;
465 TCPSTAT_INC(tcps_sndrexmitbad);
466 /* RTO was unnecessary, so reset everything. */
467 tp->snd_cwnd = tp->snd_cwnd_prev;
468 tp->snd_ssthresh = tp->snd_ssthresh_prev;
469 tp->snd_recover = tp->snd_recover_prev;
470 if (tp->t_flags & TF_WASFRECOVERY)
471 ENTER_FASTRECOVERY(tp->t_flags);
472 if (tp->t_flags & TF_WASCRECOVERY)
473 ENTER_CONGRECOVERY(tp->t_flags);
474 tp->snd_nxt = tp->snd_max;
475 tp->t_flags &= ~TF_PREVVALID;
480 if (CC_ALGO(tp)->cong_signal != NULL) {
482 tp->ccv->curack = th->th_ack;
483 CC_ALGO(tp)->cong_signal(tp->ccv, type);
488 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
490 INP_WLOCK_ASSERT(tp->t_inpcb);
492 /* XXXLAS: KASSERT that we're in recovery? */
494 if (CC_ALGO(tp)->post_recovery != NULL) {
495 tp->ccv->curack = th->th_ack;
496 CC_ALGO(tp)->post_recovery(tp->ccv);
498 /* XXXLAS: EXIT_RECOVERY ? */
499 tp->t_bytes_acked = 0;
503 * Indicate whether this ack should be delayed. We can delay the ack if
504 * following conditions are met:
505 * - There is no delayed ack timer in progress.
506 * - Our last ack wasn't a 0-sized window. We never want to delay
507 * the ack that opens up a 0-sized window.
508 * - LRO wasn't used for this segment. We make sure by checking that the
509 * segment size is not larger than the MSS.
511 #define DELAY_ACK(tp, tlen) \
512 ((!tcp_timer_active(tp, TT_DELACK) && \
513 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
514 (tlen <= tp->t_maxseg) && \
515 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
518 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
520 INP_WLOCK_ASSERT(tp->t_inpcb);
522 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
523 switch (iptos & IPTOS_ECN_MASK) {
525 tp->ccv->flags |= CCF_IPHDR_CE;
531 case IPTOS_ECN_NOTECT:
532 tp->ccv->flags &= ~CCF_IPHDR_CE;
536 if (th->th_flags & TH_CWR)
537 tp->ccv->flags |= CCF_TCPHDR_CWR;
539 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
541 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
543 if (tp->ccv->flags & CCF_ACKNOW) {
544 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
545 tp->t_flags |= TF_ACKNOW;
551 * TCP input handling is split into multiple parts:
552 * tcp6_input is a thin wrapper around tcp_input for the extended
553 * ip6_protox[] call format in ip6_input
554 * tcp_input handles primary segment validation, inpcb lookup and
555 * SYN processing on listen sockets
556 * tcp_do_segment processes the ACK and text of the segment for
557 * establishing, established and closing connections
561 tcp6_input(struct mbuf **mp, int *offp, int proto)
564 struct in6_ifaddr *ia6;
568 if (m->m_len < *offp + sizeof(struct tcphdr)) {
569 m = m_pullup(m, *offp + sizeof(struct tcphdr));
572 TCPSTAT_INC(tcps_rcvshort);
573 return (IPPROTO_DONE);
578 * draft-itojun-ipv6-tcp-to-anycast
579 * better place to put this in?
581 ip6 = mtod(m, struct ip6_hdr *);
582 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
583 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
585 ifa_free(&ia6->ia_ifa);
586 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
587 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
589 return (IPPROTO_DONE);
592 ifa_free(&ia6->ia_ifa);
595 return (tcp_input(mp, offp, proto));
600 tcp_input(struct mbuf **mp, int *offp, int proto)
602 struct mbuf *m = *mp;
603 struct tcphdr *th = NULL;
604 struct ip *ip = NULL;
605 struct inpcb *inp = NULL;
606 struct tcpcb *tp = NULL;
607 struct socket *so = NULL;
618 int rstreason = 0; /* For badport_bandlim accounting purposes */
620 struct m_tag *fwd_tag = NULL;
622 struct ip6_hdr *ip6 = NULL;
625 const void *ip6 = NULL;
627 struct tcpopt to; /* options in this segment */
628 char *s = NULL; /* address and port logging */
631 * The size of tcp_saveipgen must be the size of the max ip header,
634 u_char tcp_saveipgen[IP6_HDR_LEN];
635 struct tcphdr tcp_savetcp;
642 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
649 TCPSTAT_INC(tcps_rcvtotal);
654 ip6 = mtod(m, struct ip6_hdr *);
655 th = (struct tcphdr *)((caddr_t)ip6 + off0);
656 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
657 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
658 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
659 th->th_sum = m->m_pkthdr.csum_data;
661 th->th_sum = in6_cksum_pseudo(ip6, tlen,
662 IPPROTO_TCP, m->m_pkthdr.csum_data);
663 th->th_sum ^= 0xffff;
665 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
667 TCPSTAT_INC(tcps_rcvbadsum);
672 * Be proactive about unspecified IPv6 address in source.
673 * As we use all-zero to indicate unbounded/unconnected pcb,
674 * unspecified IPv6 address can be used to confuse us.
676 * Note that packets with unspecified IPv6 destination is
677 * already dropped in ip6_input.
679 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
683 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
686 #if defined(INET) && defined(INET6)
692 * Get IP and TCP header together in first mbuf.
693 * Note: IP leaves IP header in first mbuf.
695 if (off0 > sizeof (struct ip)) {
697 off0 = sizeof(struct ip);
699 if (m->m_len < sizeof (struct tcpiphdr)) {
700 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
702 TCPSTAT_INC(tcps_rcvshort);
703 return (IPPROTO_DONE);
706 ip = mtod(m, struct ip *);
707 th = (struct tcphdr *)((caddr_t)ip + off0);
708 tlen = ntohs(ip->ip_len) - off0;
711 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
712 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
713 th->th_sum = m->m_pkthdr.csum_data;
715 th->th_sum = in_pseudo(ip->ip_src.s_addr,
717 htonl(m->m_pkthdr.csum_data + tlen +
719 th->th_sum ^= 0xffff;
721 struct ipovly *ipov = (struct ipovly *)ip;
724 * Checksum extended TCP header and data.
728 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
729 ipov->ih_len = htons(tlen);
730 th->th_sum = in_cksum(m, len);
731 /* Reset length for SDT probes. */
732 ip->ip_len = htons(len);
735 /* Re-initialization for later version check */
737 ip->ip_v = IPVERSION;
738 ip->ip_hl = off0 >> 2;
742 TCPSTAT_INC(tcps_rcvbadsum);
749 * Check that TCP offset makes sense,
750 * pull out TCP options and adjust length. XXX
752 off = th->th_off << 2;
753 if (off < sizeof (struct tcphdr) || off > tlen) {
754 TCPSTAT_INC(tcps_rcvbadoff);
757 tlen -= off; /* tlen is used instead of ti->ti_len */
758 if (off > sizeof (struct tcphdr)) {
761 if (m->m_len < off0 + off) {
762 m = m_pullup(m, off0 + off);
764 TCPSTAT_INC(tcps_rcvshort);
765 return (IPPROTO_DONE);
768 ip6 = mtod(m, struct ip6_hdr *);
769 th = (struct tcphdr *)((caddr_t)ip6 + off0);
772 #if defined(INET) && defined(INET6)
777 if (m->m_len < sizeof(struct ip) + off) {
778 if ((m = m_pullup(m, sizeof (struct ip) + off))
780 TCPSTAT_INC(tcps_rcvshort);
781 return (IPPROTO_DONE);
783 ip = mtod(m, struct ip *);
784 th = (struct tcphdr *)((caddr_t)ip + off0);
788 optlen = off - sizeof (struct tcphdr);
789 optp = (u_char *)(th + 1);
791 thflags = th->th_flags;
794 * Convert TCP protocol specific fields to host format.
796 tcp_fields_to_host(th);
799 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
801 drop_hdrlen = off0 + off;
804 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
808 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
810 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
813 #if defined(INET) && !defined(INET6)
814 (m->m_flags & M_IP_NEXTHOP)
817 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
821 if (isipv6 && fwd_tag != NULL) {
822 struct sockaddr_in6 *next_hop6;
824 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
826 * Transparently forwarded. Pretend to be the destination.
827 * Already got one like this?
829 inp = in6_pcblookup_mbuf(&V_tcbinfo,
830 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
831 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
834 * It's new. Try to find the ambushing socket.
835 * Because we've rewritten the destination address,
836 * any hardware-generated hash is ignored.
838 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
839 th->th_sport, &next_hop6->sin6_addr,
840 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
841 th->th_dport, INPLOOKUP_WILDCARD |
842 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
845 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
846 th->th_sport, &ip6->ip6_dst, th->th_dport,
847 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
848 m->m_pkthdr.rcvif, m);
851 #if defined(INET6) && defined(INET)
855 if (fwd_tag != NULL) {
856 struct sockaddr_in *next_hop;
858 next_hop = (struct sockaddr_in *)(fwd_tag+1);
860 * Transparently forwarded. Pretend to be the destination.
861 * already got one like this?
863 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
864 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
865 m->m_pkthdr.rcvif, m);
868 * It's new. Try to find the ambushing socket.
869 * Because we've rewritten the destination address,
870 * any hardware-generated hash is ignored.
872 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
873 th->th_sport, next_hop->sin_addr,
874 next_hop->sin_port ? ntohs(next_hop->sin_port) :
875 th->th_dport, INPLOOKUP_WILDCARD |
876 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
879 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
880 th->th_sport, ip->ip_dst, th->th_dport,
881 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
882 m->m_pkthdr.rcvif, m);
886 * If the INPCB does not exist then all data in the incoming
887 * segment is discarded and an appropriate RST is sent back.
888 * XXX MRT Send RST using which routing table?
892 * Log communication attempts to ports that are not
895 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
896 tcp_log_in_vain == 2) {
897 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
898 log(LOG_INFO, "%s; %s: Connection attempt "
899 "to closed port\n", s, __func__);
902 * When blackholing do not respond with a RST but
903 * completely ignore the segment and drop it.
905 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
909 rstreason = BANDLIM_RST_CLOSEDPORT;
912 INP_WLOCK_ASSERT(inp);
914 * While waiting for inp lock during the lookup, another thread
915 * can have dropped the inpcb, in which case we need to loop back
916 * and try to find a new inpcb to deliver to.
918 if (inp->inp_flags & INP_DROPPED) {
923 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
924 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
925 ((inp->inp_socket == NULL) ||
926 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
927 inp->inp_flowid = m->m_pkthdr.flowid;
928 inp->inp_flowtype = M_HASHTYPE_GET(m);
930 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
932 if (isipv6 && IPSEC_ENABLED(ipv6) &&
933 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
941 if (IPSEC_ENABLED(ipv4) &&
942 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
949 * Check the minimum TTL for socket.
951 if (inp->inp_ip_minttl != 0) {
954 if (inp->inp_ip_minttl > ip6->ip6_hlim)
958 if (inp->inp_ip_minttl > ip->ip_ttl)
963 * A previous connection in TIMEWAIT state is supposed to catch stray
964 * or duplicate segments arriving late. If this segment was a
965 * legitimate new connection attempt, the old INPCB gets removed and
966 * we can try again to find a listening socket.
968 * At this point, due to earlier optimism, we may hold only an inpcb
969 * lock, and not the inpcbinfo write lock. If so, we need to try to
970 * acquire it, or if that fails, acquire a reference on the inpcb,
971 * drop all locks, acquire a global write lock, and then re-acquire
972 * the inpcb lock. We may at that point discover that another thread
973 * has tried to free the inpcb, in which case we need to loop back
974 * and try to find a new inpcb to deliver to.
976 * XXXRW: It may be time to rethink timewait locking.
978 if (inp->inp_flags & INP_TIMEWAIT) {
979 if (thflags & TH_SYN)
980 tcp_dooptions(&to, optp, optlen, TO_SYN);
982 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
984 if (tcp_twcheck(inp, &to, th, m, tlen))
986 return (IPPROTO_DONE);
989 * The TCPCB may no longer exist if the connection is winding
990 * down or it is in the CLOSED state. Either way we drop the
991 * segment and send an appropriate response.
994 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
995 rstreason = BANDLIM_RST_CLOSEDPORT;
1000 if (tp->t_flags & TF_TOE) {
1001 tcp_offload_input(tp, m);
1002 m = NULL; /* consumed by the TOE driver */
1008 INP_WLOCK_ASSERT(inp);
1009 if (mac_inpcb_check_deliver(inp, m))
1012 so = inp->inp_socket;
1013 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1015 if (so->so_options & SO_DEBUG) {
1016 ostate = tp->t_state;
1019 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1022 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1025 #endif /* TCPDEBUG */
1027 * When the socket is accepting connections (the INPCB is in LISTEN
1028 * state) we look into the SYN cache if this is a new connection
1029 * attempt or the completion of a previous one.
1031 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1032 ("%s: so accepting but tp %p not listening", __func__, tp));
1033 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1034 struct in_conninfo inc;
1036 bzero(&inc, sizeof(inc));
1039 inc.inc_flags |= INC_ISIPV6;
1040 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1041 inc.inc_flags |= INC_IPV6MINMTU;
1042 inc.inc6_faddr = ip6->ip6_src;
1043 inc.inc6_laddr = ip6->ip6_dst;
1047 inc.inc_faddr = ip->ip_src;
1048 inc.inc_laddr = ip->ip_dst;
1050 inc.inc_fport = th->th_sport;
1051 inc.inc_lport = th->th_dport;
1052 inc.inc_fibnum = so->so_fibnum;
1055 * Check for an existing connection attempt in syncache if
1056 * the flag is only ACK. A successful lookup creates a new
1057 * socket appended to the listen queue in SYN_RECEIVED state.
1059 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_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1466 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1468 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1469 int rstreason, todrop, win;
1473 struct in_conninfo *inc;
1480 * The size of tcp_saveipgen must be the size of the max ip header,
1483 u_char tcp_saveipgen[IP6_HDR_LEN];
1484 struct tcphdr tcp_savetcp;
1487 thflags = th->th_flags;
1488 inc = &tp->t_inpcb->inp_inc;
1489 tp->sackhint.last_sack_ack = 0;
1491 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1494 INP_WLOCK_ASSERT(tp->t_inpcb);
1495 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1497 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1501 /* Save segment, if requested. */
1502 tcp_pcap_add(th, m, &(tp->t_inpkts));
1504 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1507 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1508 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1509 log(LOG_DEBUG, "%s; %s: "
1510 "SYN|FIN segment ignored (based on "
1511 "sysctl setting)\n", s, __func__);
1518 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1519 * check SEQ.ACK first.
1521 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1522 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1523 rstreason = BANDLIM_UNLIMITED;
1528 * Segment received on connection.
1529 * Reset idle time and keep-alive timer.
1530 * XXX: This should be done after segment
1531 * validation to ignore broken/spoofed segs.
1533 tp->t_rcvtime = ticks;
1536 * Scale up the window into a 32-bit value.
1537 * For the SYN_SENT state the scale is zero.
1539 tiwin = th->th_win << tp->snd_scale;
1541 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1545 * TCP ECN processing.
1547 if (tp->t_flags2 & TF2_ECN_PERMIT) {
1548 if (thflags & TH_CWR)
1549 tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1550 switch (iptos & IPTOS_ECN_MASK) {
1552 tp->t_flags2 |= TF2_ECN_SND_ECE;
1553 TCPSTAT_INC(tcps_ecn_ce);
1555 case IPTOS_ECN_ECT0:
1556 TCPSTAT_INC(tcps_ecn_ect0);
1558 case IPTOS_ECN_ECT1:
1559 TCPSTAT_INC(tcps_ecn_ect1);
1563 /* Process a packet differently from RFC3168. */
1564 cc_ecnpkt_handler(tp, th, iptos);
1566 /* Congestion experienced. */
1567 if (thflags & TH_ECE) {
1568 cc_cong_signal(tp, th, CC_ECN);
1573 * Parse options on any incoming segment.
1575 tcp_dooptions(&to, (u_char *)(th + 1),
1576 (th->th_off << 2) - sizeof(struct tcphdr),
1577 (thflags & TH_SYN) ? TO_SYN : 0);
1579 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1580 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1581 (to.to_flags & TOF_SIGNATURE) == 0) {
1582 TCPSTAT_INC(tcps_sig_err_sigopt);
1583 /* XXX: should drop? */
1587 * If echoed timestamp is later than the current time,
1588 * fall back to non RFC1323 RTT calculation. Normalize
1589 * timestamp if syncookies were used when this connection
1592 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1593 to.to_tsecr -= tp->ts_offset;
1594 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1596 else if (tp->t_flags & TF_PREVVALID &&
1597 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1598 cc_cong_signal(tp, th, CC_RTO_ERR);
1601 * Process options only when we get SYN/ACK back. The SYN case
1602 * for incoming connections is handled in tcp_syncache.
1603 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1604 * or <SYN,ACK>) segment itself is never scaled.
1605 * XXX this is traditional behavior, may need to be cleaned up.
1607 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1608 if ((to.to_flags & TOF_SCALE) &&
1609 (tp->t_flags & TF_REQ_SCALE)) {
1610 tp->t_flags |= TF_RCVD_SCALE;
1611 tp->snd_scale = to.to_wscale;
1614 * Initial send window. It will be updated with
1615 * the next incoming segment to the scaled value.
1617 tp->snd_wnd = th->th_win;
1618 if (to.to_flags & TOF_TS) {
1619 tp->t_flags |= TF_RCVD_TSTMP;
1620 tp->ts_recent = to.to_tsval;
1621 tp->ts_recent_age = tcp_ts_getticks();
1623 if (to.to_flags & TOF_MSS)
1624 tcp_mss(tp, to.to_mss);
1625 if ((tp->t_flags & TF_SACK_PERMIT) &&
1626 (to.to_flags & TOF_SACKPERM) == 0)
1627 tp->t_flags &= ~TF_SACK_PERMIT;
1628 if (IS_FASTOPEN(tp->t_flags)) {
1629 if (to.to_flags & TOF_FASTOPEN) {
1632 if (to.to_flags & TOF_MSS)
1635 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1639 tcp_fastopen_update_cache(tp, mss,
1640 to.to_tfo_len, to.to_tfo_cookie);
1642 tcp_fastopen_disable_path(tp);
1647 * If timestamps were negotiated during SYN/ACK they should
1648 * appear on every segment during this session and vice versa.
1650 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1651 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1652 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1653 "no action\n", s, __func__);
1657 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1658 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1659 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1660 "no action\n", s, __func__);
1666 * Header prediction: check for the two common cases
1667 * of a uni-directional data xfer. If the packet has
1668 * no control flags, is in-sequence, the window didn't
1669 * change and we're not retransmitting, it's a
1670 * candidate. If the length is zero and the ack moved
1671 * forward, we're the sender side of the xfer. Just
1672 * free the data acked & wake any higher level process
1673 * that was blocked waiting for space. If the length
1674 * is non-zero and the ack didn't move, we're the
1675 * receiver side. If we're getting packets in-order
1676 * (the reassembly queue is empty), add the data to
1677 * the socket buffer and note that we need a delayed ack.
1678 * Make sure that the hidden state-flags are also off.
1679 * Since we check for TCPS_ESTABLISHED first, it can only
1682 if (tp->t_state == TCPS_ESTABLISHED &&
1683 th->th_seq == tp->rcv_nxt &&
1684 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1685 tp->snd_nxt == tp->snd_max &&
1686 tiwin && tiwin == tp->snd_wnd &&
1687 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1689 ((to.to_flags & TOF_TS) == 0 ||
1690 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1693 * If last ACK falls within this segment's sequence numbers,
1694 * record the timestamp.
1695 * NOTE that the test is modified according to the latest
1696 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1698 if ((to.to_flags & TOF_TS) != 0 &&
1699 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1700 tp->ts_recent_age = tcp_ts_getticks();
1701 tp->ts_recent = to.to_tsval;
1705 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1706 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1707 !IN_RECOVERY(tp->t_flags) &&
1708 (to.to_flags & TOF_SACK) == 0 &&
1709 TAILQ_EMPTY(&tp->snd_holes)) {
1711 * This is a pure ack for outstanding data.
1713 TCPSTAT_INC(tcps_predack);
1716 * "bad retransmit" recovery without timestamps.
1718 if ((to.to_flags & TOF_TS) == 0 &&
1719 tp->t_rxtshift == 1 &&
1720 tp->t_flags & TF_PREVVALID &&
1721 (int)(ticks - tp->t_badrxtwin) < 0) {
1722 cc_cong_signal(tp, th, CC_RTO_ERR);
1726 * Recalculate the transmit timer / rtt.
1728 * Some boxes send broken timestamp replies
1729 * during the SYN+ACK phase, ignore
1730 * timestamps of 0 or we could calculate a
1731 * huge RTT and blow up the retransmit timer.
1733 if ((to.to_flags & TOF_TS) != 0 &&
1737 t = tcp_ts_getticks() - to.to_tsecr;
1738 if (!tp->t_rttlow || tp->t_rttlow > t)
1741 TCP_TS_TO_TICKS(t) + 1);
1742 } else if (tp->t_rtttime &&
1743 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1744 if (!tp->t_rttlow ||
1745 tp->t_rttlow > ticks - tp->t_rtttime)
1746 tp->t_rttlow = ticks - tp->t_rtttime;
1748 ticks - tp->t_rtttime);
1750 acked = BYTES_THIS_ACK(tp, th);
1753 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1754 hhook_run_tcp_est_in(tp, th, &to);
1757 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1758 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1759 sbdrop(&so->so_snd, acked);
1760 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1761 SEQ_LEQ(th->th_ack, tp->snd_recover))
1762 tp->snd_recover = th->th_ack - 1;
1765 * Let the congestion control algorithm update
1766 * congestion control related information. This
1767 * typically means increasing the congestion
1770 cc_ack_received(tp, th, nsegs, CC_ACK);
1772 tp->snd_una = th->th_ack;
1774 * Pull snd_wl2 up to prevent seq wrap relative
1777 tp->snd_wl2 = th->th_ack;
1782 * If all outstanding data are acked, stop
1783 * retransmit timer, otherwise restart timer
1784 * using current (possibly backed-off) value.
1785 * If process is waiting for space,
1786 * wakeup/selwakeup/signal. If data
1787 * are ready to send, let tcp_output
1788 * decide between more output or persist.
1791 if (so->so_options & SO_DEBUG)
1792 tcp_trace(TA_INPUT, ostate, tp,
1793 (void *)tcp_saveipgen,
1796 TCP_PROBE3(debug__input, tp, th, m);
1797 if (tp->snd_una == tp->snd_max)
1798 tcp_timer_activate(tp, TT_REXMT, 0);
1799 else if (!tcp_timer_active(tp, TT_PERSIST))
1800 tcp_timer_activate(tp, TT_REXMT,
1803 if (sbavail(&so->so_snd))
1804 (void) tp->t_fb->tfb_tcp_output(tp);
1807 } else if (th->th_ack == tp->snd_una &&
1808 tlen <= sbspace(&so->so_rcv)) {
1809 int newsize = 0; /* automatic sockbuf scaling */
1812 * This is a pure, in-sequence data packet with
1813 * nothing on the reassembly queue and we have enough
1814 * buffer space to take it.
1816 /* Clean receiver SACK report if present */
1817 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1818 tcp_clean_sackreport(tp);
1819 TCPSTAT_INC(tcps_preddat);
1820 tp->rcv_nxt += tlen;
1822 * Pull snd_wl1 up to prevent seq wrap relative to
1825 tp->snd_wl1 = th->th_seq;
1827 * Pull rcv_up up to prevent seq wrap relative to
1830 tp->rcv_up = tp->rcv_nxt;
1831 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1832 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1834 if (so->so_options & SO_DEBUG)
1835 tcp_trace(TA_INPUT, ostate, tp,
1836 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1838 TCP_PROBE3(debug__input, tp, th, m);
1840 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1842 /* Add data to socket buffer. */
1843 SOCKBUF_LOCK(&so->so_rcv);
1844 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1848 * Set new socket buffer size.
1849 * Give up when limit is reached.
1852 if (!sbreserve_locked(&so->so_rcv,
1854 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1855 m_adj(m, drop_hdrlen); /* delayed header drop */
1856 sbappendstream_locked(&so->so_rcv, m, 0);
1858 /* NB: sorwakeup_locked() does an implicit unlock. */
1859 sorwakeup_locked(so);
1860 if (DELAY_ACK(tp, tlen)) {
1861 tp->t_flags |= TF_DELACK;
1863 tp->t_flags |= TF_ACKNOW;
1864 tp->t_fb->tfb_tcp_output(tp);
1871 * Calculate amount of space in receive window,
1872 * and then do TCP input processing.
1873 * Receive window is amount of space in rcv queue,
1874 * but not less than advertised window.
1876 win = sbspace(&so->so_rcv);
1879 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1881 switch (tp->t_state) {
1884 * If the state is SYN_RECEIVED:
1885 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1887 case TCPS_SYN_RECEIVED:
1888 if ((thflags & TH_ACK) &&
1889 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1890 SEQ_GT(th->th_ack, tp->snd_max))) {
1891 rstreason = BANDLIM_RST_OPENPORT;
1894 if (IS_FASTOPEN(tp->t_flags)) {
1896 * When a TFO connection is in SYN_RECEIVED, the
1897 * only valid packets are the initial SYN, a
1898 * retransmit/copy of the initial SYN (possibly with
1899 * a subset of the original data), a valid ACK, a
1902 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1903 rstreason = BANDLIM_RST_OPENPORT;
1905 } else if (thflags & TH_SYN) {
1906 /* non-initial SYN is ignored */
1907 if ((tcp_timer_active(tp, TT_DELACK) ||
1908 tcp_timer_active(tp, TT_REXMT)))
1910 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1917 * If the state is SYN_SENT:
1918 * if seg contains a RST with valid ACK (SEQ.ACK has already
1919 * been verified), then drop the connection.
1920 * if seg contains a RST without an ACK, drop the seg.
1921 * if seg does not contain SYN, then drop the seg.
1922 * Otherwise this is an acceptable SYN segment
1923 * initialize tp->rcv_nxt and tp->irs
1924 * if seg contains ack then advance tp->snd_una
1925 * if seg contains an ECE and ECN support is enabled, the stream
1927 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1928 * arrange for segment to be acked (eventually)
1929 * continue processing rest of data/controls, beginning with URG
1932 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1933 TCP_PROBE5(connect__refused, NULL, tp,
1935 tp = tcp_drop(tp, ECONNREFUSED);
1937 if (thflags & TH_RST)
1939 if (!(thflags & TH_SYN))
1942 tp->irs = th->th_seq;
1944 if (thflags & TH_ACK) {
1945 int tfo_partial_ack = 0;
1947 TCPSTAT_INC(tcps_connects);
1950 mac_socketpeer_set_from_mbuf(m, so);
1952 /* Do window scaling on this connection? */
1953 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1954 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1955 tp->rcv_scale = tp->request_r_scale;
1957 tp->rcv_adv += min(tp->rcv_wnd,
1958 TCP_MAXWIN << tp->rcv_scale);
1959 tp->snd_una++; /* SYN is acked */
1961 * If not all the data that was sent in the TFO SYN
1962 * has been acked, resend the remainder right away.
1964 if (IS_FASTOPEN(tp->t_flags) &&
1965 (tp->snd_una != tp->snd_max)) {
1966 tp->snd_nxt = th->th_ack;
1967 tfo_partial_ack = 1;
1970 * If there's data, delay ACK; if there's also a FIN
1971 * ACKNOW will be turned on later.
1973 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
1974 tcp_timer_activate(tp, TT_DELACK,
1977 tp->t_flags |= TF_ACKNOW;
1979 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
1981 tp->t_flags2 |= TF2_ECN_PERMIT;
1982 TCPSTAT_INC(tcps_ecn_shs);
1986 * Received <SYN,ACK> in SYN_SENT[*] state.
1988 * SYN_SENT --> ESTABLISHED
1989 * SYN_SENT* --> FIN_WAIT_1
1991 tp->t_starttime = ticks;
1992 if (tp->t_flags & TF_NEEDFIN) {
1993 tcp_state_change(tp, TCPS_FIN_WAIT_1);
1994 tp->t_flags &= ~TF_NEEDFIN;
1997 tcp_state_change(tp, TCPS_ESTABLISHED);
1998 TCP_PROBE5(connect__established, NULL, tp,
2001 tcp_timer_activate(tp, TT_KEEP,
2006 * Received initial SYN in SYN-SENT[*] state =>
2007 * simultaneous open.
2008 * If it succeeds, connection is * half-synchronized.
2009 * Otherwise, do 3-way handshake:
2010 * SYN-SENT -> SYN-RECEIVED
2011 * SYN-SENT* -> SYN-RECEIVED*
2013 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2014 tcp_timer_activate(tp, TT_REXMT, 0);
2015 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2018 INP_WLOCK_ASSERT(tp->t_inpcb);
2021 * Advance th->th_seq to correspond to first data byte.
2022 * If data, trim to stay within window,
2023 * dropping FIN if necessary.
2026 if (tlen > tp->rcv_wnd) {
2027 todrop = tlen - tp->rcv_wnd;
2031 TCPSTAT_INC(tcps_rcvpackafterwin);
2032 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2034 tp->snd_wl1 = th->th_seq - 1;
2035 tp->rcv_up = th->th_seq;
2037 * Client side of transaction: already sent SYN and data.
2038 * If the remote host used T/TCP to validate the SYN,
2039 * our data will be ACK'd; if so, enter normal data segment
2040 * processing in the middle of step 5, ack processing.
2041 * Otherwise, goto step 6.
2043 if (thflags & TH_ACK)
2049 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2050 * do normal processing.
2052 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2056 break; /* continue normal processing */
2060 * States other than LISTEN or SYN_SENT.
2061 * First check the RST flag and sequence number since reset segments
2062 * are exempt from the timestamp and connection count tests. This
2063 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2064 * below which allowed reset segments in half the sequence space
2065 * to fall though and be processed (which gives forged reset
2066 * segments with a random sequence number a 50 percent chance of
2067 * killing a connection).
2068 * Then check timestamp, if present.
2069 * Then check the connection count, if present.
2070 * Then check that at least some bytes of segment are within
2071 * receive window. If segment begins before rcv_nxt,
2072 * drop leading data (and SYN); if nothing left, just ack.
2074 if (thflags & TH_RST) {
2076 * RFC5961 Section 3.2
2078 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2079 * - If RST is in window, we send challenge ACK.
2081 * Note: to take into account delayed ACKs, we should
2082 * test against last_ack_sent instead of rcv_nxt.
2083 * Note 2: we handle special case of closed window, not
2084 * covered by the RFC.
2086 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2087 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2088 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2090 KASSERT(tp->t_state != TCPS_SYN_SENT,
2091 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2094 if (V_tcp_insecure_rst ||
2095 tp->last_ack_sent == th->th_seq) {
2096 TCPSTAT_INC(tcps_drops);
2097 /* Drop the connection. */
2098 switch (tp->t_state) {
2099 case TCPS_SYN_RECEIVED:
2100 so->so_error = ECONNREFUSED;
2102 case TCPS_ESTABLISHED:
2103 case TCPS_FIN_WAIT_1:
2104 case TCPS_FIN_WAIT_2:
2105 case TCPS_CLOSE_WAIT:
2108 so->so_error = ECONNRESET;
2115 TCPSTAT_INC(tcps_badrst);
2116 /* Send challenge ACK. */
2117 tcp_respond(tp, mtod(m, void *), th, m,
2118 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2119 tp->last_ack_sent = tp->rcv_nxt;
2127 * RFC5961 Section 4.2
2128 * Send challenge ACK for any SYN in synchronized state.
2130 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2131 tp->t_state != TCPS_SYN_RECEIVED) {
2132 TCPSTAT_INC(tcps_badsyn);
2133 if (V_tcp_insecure_syn &&
2134 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2135 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2136 tp = tcp_drop(tp, ECONNRESET);
2137 rstreason = BANDLIM_UNLIMITED;
2139 /* Send challenge ACK. */
2140 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2141 tp->snd_nxt, TH_ACK);
2142 tp->last_ack_sent = tp->rcv_nxt;
2149 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2150 * and it's less than ts_recent, drop it.
2152 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2153 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2155 /* Check to see if ts_recent is over 24 days old. */
2156 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2158 * Invalidate ts_recent. If this segment updates
2159 * ts_recent, the age will be reset later and ts_recent
2160 * will get a valid value. If it does not, setting
2161 * ts_recent to zero will at least satisfy the
2162 * requirement that zero be placed in the timestamp
2163 * echo reply when ts_recent isn't valid. The
2164 * age isn't reset until we get a valid ts_recent
2165 * because we don't want out-of-order segments to be
2166 * dropped when ts_recent is old.
2170 TCPSTAT_INC(tcps_rcvduppack);
2171 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2172 TCPSTAT_INC(tcps_pawsdrop);
2180 * In the SYN-RECEIVED state, validate that the packet belongs to
2181 * this connection before trimming the data to fit the receive
2182 * window. Check the sequence number versus IRS since we know
2183 * the sequence numbers haven't wrapped. This is a partial fix
2184 * for the "LAND" DoS attack.
2186 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2187 rstreason = BANDLIM_RST_OPENPORT;
2191 todrop = tp->rcv_nxt - th->th_seq;
2193 if (thflags & TH_SYN) {
2203 * Following if statement from Stevens, vol. 2, p. 960.
2206 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2208 * Any valid FIN must be to the left of the window.
2209 * At this point the FIN must be a duplicate or out
2210 * of sequence; drop it.
2215 * Send an ACK to resynchronize and drop any data.
2216 * But keep on processing for RST or ACK.
2218 tp->t_flags |= TF_ACKNOW;
2220 TCPSTAT_INC(tcps_rcvduppack);
2221 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2223 TCPSTAT_INC(tcps_rcvpartduppack);
2224 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2227 * DSACK - add SACK block for dropped range
2229 if (tp->t_flags & TF_SACK_PERMIT) {
2230 tcp_update_sack_list(tp, th->th_seq,
2231 th->th_seq + todrop);
2233 * ACK now, as the next in-sequence segment
2234 * will clear the DSACK block again
2236 tp->t_flags |= TF_ACKNOW;
2238 drop_hdrlen += todrop; /* drop from the top afterwards */
2239 th->th_seq += todrop;
2241 if (th->th_urp > todrop)
2242 th->th_urp -= todrop;
2250 * If new data are received on a connection after the
2251 * user processes are gone, then RST the other end.
2253 if ((so->so_state & SS_NOFDREF) &&
2254 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2255 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2256 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2257 "after socket was closed, "
2258 "sending RST and removing tcpcb\n",
2259 s, __func__, tcpstates[tp->t_state], tlen);
2263 TCPSTAT_INC(tcps_rcvafterclose);
2264 rstreason = BANDLIM_UNLIMITED;
2269 * If segment ends after window, drop trailing data
2270 * (and PUSH and FIN); if nothing left, just ACK.
2272 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2274 TCPSTAT_INC(tcps_rcvpackafterwin);
2275 if (todrop >= tlen) {
2276 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2278 * If window is closed can only take segments at
2279 * window edge, and have to drop data and PUSH from
2280 * incoming segments. Continue processing, but
2281 * remember to ack. Otherwise, drop segment
2284 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2285 tp->t_flags |= TF_ACKNOW;
2286 TCPSTAT_INC(tcps_rcvwinprobe);
2290 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2293 thflags &= ~(TH_PUSH|TH_FIN);
2297 * If last ACK falls within this segment's sequence numbers,
2298 * record its timestamp.
2300 * 1) That the test incorporates suggestions from the latest
2301 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2302 * 2) That updating only on newer timestamps interferes with
2303 * our earlier PAWS tests, so this check should be solely
2304 * predicated on the sequence space of this segment.
2305 * 3) That we modify the segment boundary check to be
2306 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2307 * instead of RFC1323's
2308 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2309 * This modified check allows us to overcome RFC1323's
2310 * limitations as described in Stevens TCP/IP Illustrated
2311 * Vol. 2 p.869. In such cases, we can still calculate the
2312 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2314 if ((to.to_flags & TOF_TS) != 0 &&
2315 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2316 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2317 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2318 tp->ts_recent_age = tcp_ts_getticks();
2319 tp->ts_recent = to.to_tsval;
2323 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2324 * flag is on (half-synchronized state), then queue data for
2325 * later processing; else drop segment and return.
2327 if ((thflags & TH_ACK) == 0) {
2328 if (tp->t_state == TCPS_SYN_RECEIVED ||
2329 (tp->t_flags & TF_NEEDSYN)) {
2330 if (tp->t_state == TCPS_SYN_RECEIVED &&
2331 IS_FASTOPEN(tp->t_flags)) {
2332 tp->snd_wnd = tiwin;
2336 } else if (tp->t_flags & TF_ACKNOW)
2345 switch (tp->t_state) {
2348 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2349 * ESTABLISHED state and continue processing.
2350 * The ACK was checked above.
2352 case TCPS_SYN_RECEIVED:
2354 TCPSTAT_INC(tcps_connects);
2356 /* Do window scaling? */
2357 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2358 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2359 tp->rcv_scale = tp->request_r_scale;
2361 tp->snd_wnd = tiwin;
2364 * SYN-RECEIVED -> ESTABLISHED
2365 * SYN-RECEIVED* -> FIN-WAIT-1
2367 tp->t_starttime = ticks;
2368 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2369 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2370 tp->t_tfo_pending = NULL;
2373 * Account for the ACK of our SYN prior to
2374 * regular ACK processing below.
2378 if (tp->t_flags & TF_NEEDFIN) {
2379 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2380 tp->t_flags &= ~TF_NEEDFIN;
2382 tcp_state_change(tp, TCPS_ESTABLISHED);
2383 TCP_PROBE5(accept__established, NULL, tp,
2386 * TFO connections call cc_conn_init() during SYN
2387 * processing. Calling it again here for such
2388 * connections is not harmless as it would undo the
2389 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2392 if (!IS_FASTOPEN(tp->t_flags))
2394 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2397 * If segment contains data or ACK, will call tcp_reass()
2398 * later; if not, do so now to pass queued data to user.
2400 if (tlen == 0 && (thflags & TH_FIN) == 0)
2401 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2403 tp->snd_wl1 = th->th_seq - 1;
2407 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2408 * ACKs. If the ack is in the range
2409 * tp->snd_una < th->th_ack <= tp->snd_max
2410 * then advance tp->snd_una to th->th_ack and drop
2411 * data from the retransmission queue. If this ACK reflects
2412 * more up to date window information we update our window information.
2414 case TCPS_ESTABLISHED:
2415 case TCPS_FIN_WAIT_1:
2416 case TCPS_FIN_WAIT_2:
2417 case TCPS_CLOSE_WAIT:
2420 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2421 TCPSTAT_INC(tcps_rcvacktoomuch);
2424 if ((tp->t_flags & TF_SACK_PERMIT) &&
2425 ((to.to_flags & TOF_SACK) ||
2426 !TAILQ_EMPTY(&tp->snd_holes)))
2427 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2430 * Reset the value so that previous (valid) value
2431 * from the last ack with SACK doesn't get used.
2433 tp->sackhint.sacked_bytes = 0;
2436 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2437 hhook_run_tcp_est_in(tp, th, &to);
2440 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2443 maxseg = tcp_maxseg(tp);
2445 (tiwin == tp->snd_wnd ||
2446 (tp->t_flags & TF_SACK_PERMIT))) {
2448 * If this is the first time we've seen a
2449 * FIN from the remote, this is not a
2450 * duplicate and it needs to be processed
2451 * normally. This happens during a
2452 * simultaneous close.
2454 if ((thflags & TH_FIN) &&
2455 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2459 TCPSTAT_INC(tcps_rcvdupack);
2461 * If we have outstanding data (other than
2462 * a window probe), this is a completely
2463 * duplicate ack (ie, window info didn't
2464 * change and FIN isn't set),
2465 * the ack is the biggest we've
2466 * seen and we've seen exactly our rexmt
2467 * threshold of them, assume a packet
2468 * has been dropped and retransmit it.
2469 * Kludge snd_nxt & the congestion
2470 * window so we send only this one
2473 * We know we're losing at the current
2474 * window size so do congestion avoidance
2475 * (set ssthresh to half the current window
2476 * and pull our congestion window back to
2477 * the new ssthresh).
2479 * Dup acks mean that packets have left the
2480 * network (they're now cached at the receiver)
2481 * so bump cwnd by the amount in the receiver
2482 * to keep a constant cwnd packets in the
2485 * When using TCP ECN, notify the peer that
2486 * we reduced the cwnd.
2489 * Following 2 kinds of acks should not affect
2492 * 2) Acks with SACK but without any new SACK
2493 * information in them. These could result from
2494 * any anomaly in the network like a switch
2495 * duplicating packets or a possible DoS attack.
2497 if (th->th_ack != tp->snd_una ||
2498 ((tp->t_flags & TF_SACK_PERMIT) &&
2501 else if (!tcp_timer_active(tp, TT_REXMT))
2503 else if (++tp->t_dupacks > tcprexmtthresh ||
2504 IN_FASTRECOVERY(tp->t_flags)) {
2505 cc_ack_received(tp, th, nsegs,
2507 if ((tp->t_flags & TF_SACK_PERMIT) &&
2508 IN_FASTRECOVERY(tp->t_flags)) {
2512 * Compute the amount of data in flight first.
2513 * We can inject new data into the pipe iff
2514 * we have less than 1/2 the original window's
2515 * worth of data in flight.
2517 if (V_tcp_do_rfc6675_pipe)
2518 awnd = tcp_compute_pipe(tp);
2520 awnd = (tp->snd_nxt - tp->snd_fack) +
2521 tp->sackhint.sack_bytes_rexmit;
2523 if (awnd < tp->snd_ssthresh) {
2524 tp->snd_cwnd += maxseg;
2525 if (tp->snd_cwnd > tp->snd_ssthresh)
2526 tp->snd_cwnd = tp->snd_ssthresh;
2529 tp->snd_cwnd += maxseg;
2530 (void) tp->t_fb->tfb_tcp_output(tp);
2532 } else if (tp->t_dupacks == tcprexmtthresh) {
2533 tcp_seq onxt = tp->snd_nxt;
2536 * If we're doing sack, check to
2537 * see if we're already in sack
2538 * recovery. If we're not doing sack,
2539 * check to see if we're in newreno
2542 if (tp->t_flags & TF_SACK_PERMIT) {
2543 if (IN_FASTRECOVERY(tp->t_flags)) {
2548 if (SEQ_LEQ(th->th_ack,
2554 /* Congestion signal before ack. */
2555 cc_cong_signal(tp, th, CC_NDUPACK);
2556 cc_ack_received(tp, th, nsegs,
2558 tcp_timer_activate(tp, TT_REXMT, 0);
2560 if (tp->t_flags & TF_SACK_PERMIT) {
2562 tcps_sack_recovery_episode);
2563 tp->sack_newdata = tp->snd_nxt;
2564 tp->snd_cwnd = maxseg;
2565 (void) tp->t_fb->tfb_tcp_output(tp);
2568 tp->snd_nxt = th->th_ack;
2569 tp->snd_cwnd = maxseg;
2570 (void) tp->t_fb->tfb_tcp_output(tp);
2571 KASSERT(tp->snd_limited <= 2,
2572 ("%s: tp->snd_limited too big",
2574 tp->snd_cwnd = tp->snd_ssthresh +
2576 (tp->t_dupacks - tp->snd_limited);
2577 if (SEQ_GT(onxt, tp->snd_nxt))
2580 } else if (V_tcp_do_rfc3042) {
2582 * Process first and second duplicate
2583 * ACKs. Each indicates a segment
2584 * leaving the network, creating room
2585 * for more. Make sure we can send a
2586 * packet on reception of each duplicate
2587 * ACK by increasing snd_cwnd by one
2588 * segment. Restore the original
2589 * snd_cwnd after packet transmission.
2591 cc_ack_received(tp, th, nsegs,
2593 uint32_t oldcwnd = tp->snd_cwnd;
2594 tcp_seq oldsndmax = tp->snd_max;
2598 KASSERT(tp->t_dupacks == 1 ||
2600 ("%s: dupacks not 1 or 2",
2602 if (tp->t_dupacks == 1)
2603 tp->snd_limited = 0;
2605 (tp->snd_nxt - tp->snd_una) +
2606 (tp->t_dupacks - tp->snd_limited) *
2609 * Only call tcp_output when there
2610 * is new data available to be sent.
2611 * Otherwise we would send pure ACKs.
2613 SOCKBUF_LOCK(&so->so_snd);
2614 avail = sbavail(&so->so_snd) -
2615 (tp->snd_nxt - tp->snd_una);
2616 SOCKBUF_UNLOCK(&so->so_snd);
2618 (void) tp->t_fb->tfb_tcp_output(tp);
2619 sent = tp->snd_max - oldsndmax;
2620 if (sent > maxseg) {
2621 KASSERT((tp->t_dupacks == 2 &&
2622 tp->snd_limited == 0) ||
2623 (sent == maxseg + 1 &&
2624 tp->t_flags & TF_SENTFIN),
2625 ("%s: sent too much",
2627 tp->snd_limited = 2;
2628 } else if (sent > 0)
2630 tp->snd_cwnd = oldcwnd;
2637 * This ack is advancing the left edge, reset the
2642 * If this ack also has new SACK info, increment the
2643 * counter as per rfc6675.
2645 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed)
2649 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2650 ("%s: th_ack <= snd_una", __func__));
2653 * If the congestion window was inflated to account
2654 * for the other side's cached packets, retract it.
2656 if (IN_FASTRECOVERY(tp->t_flags)) {
2657 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2658 if (tp->t_flags & TF_SACK_PERMIT)
2659 tcp_sack_partialack(tp, th);
2661 tcp_newreno_partial_ack(tp, th);
2663 cc_post_recovery(tp, th);
2666 * If we reach this point, ACK is not a duplicate,
2667 * i.e., it ACKs something we sent.
2669 if (tp->t_flags & TF_NEEDSYN) {
2671 * T/TCP: Connection was half-synchronized, and our
2672 * SYN has been ACK'd (so connection is now fully
2673 * synchronized). Go to non-starred state,
2674 * increment snd_una for ACK of SYN, and check if
2675 * we can do window scaling.
2677 tp->t_flags &= ~TF_NEEDSYN;
2679 /* Do window scaling? */
2680 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2681 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2682 tp->rcv_scale = tp->request_r_scale;
2683 /* Send window already scaled. */
2688 INP_WLOCK_ASSERT(tp->t_inpcb);
2690 acked = BYTES_THIS_ACK(tp, th);
2691 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2692 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2693 tp->snd_una, th->th_ack, tp, m));
2694 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2695 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2698 * If we just performed our first retransmit, and the ACK
2699 * arrives within our recovery window, then it was a mistake
2700 * to do the retransmit in the first place. Recover our
2701 * original cwnd and ssthresh, and proceed to transmit where
2704 if (tp->t_rxtshift == 1 &&
2705 tp->t_flags & TF_PREVVALID &&
2707 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2708 cc_cong_signal(tp, th, CC_RTO_ERR);
2711 * If we have a timestamp reply, update smoothed
2712 * round trip time. If no timestamp is present but
2713 * transmit timer is running and timed sequence
2714 * number was acked, update smoothed round trip time.
2715 * Since we now have an rtt measurement, cancel the
2716 * timer backoff (cf., Phil Karn's retransmit alg.).
2717 * Recompute the initial retransmit timer.
2719 * Some boxes send broken timestamp replies
2720 * during the SYN+ACK phase, ignore
2721 * timestamps of 0 or we could calculate a
2722 * huge RTT and blow up the retransmit timer.
2724 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2727 t = tcp_ts_getticks() - to.to_tsecr;
2728 if (!tp->t_rttlow || tp->t_rttlow > t)
2730 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2731 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2732 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2733 tp->t_rttlow = ticks - tp->t_rtttime;
2734 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2738 * If all outstanding data is acked, stop retransmit
2739 * timer and remember to restart (more output or persist).
2740 * If there is more data to be acked, restart retransmit
2741 * timer, using current (possibly backed-off) value.
2743 if (th->th_ack == tp->snd_max) {
2744 tcp_timer_activate(tp, TT_REXMT, 0);
2746 } else if (!tcp_timer_active(tp, TT_PERSIST))
2747 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2750 * If no data (only SYN) was ACK'd,
2751 * skip rest of ACK processing.
2757 * Let the congestion control algorithm update congestion
2758 * control related information. This typically means increasing
2759 * the congestion window.
2761 cc_ack_received(tp, th, nsegs, CC_ACK);
2763 SOCKBUF_LOCK(&so->so_snd);
2764 if (acked > sbavail(&so->so_snd)) {
2765 if (tp->snd_wnd >= sbavail(&so->so_snd))
2766 tp->snd_wnd -= sbavail(&so->so_snd);
2769 mfree = sbcut_locked(&so->so_snd,
2770 (int)sbavail(&so->so_snd));
2773 mfree = sbcut_locked(&so->so_snd, acked);
2774 if (tp->snd_wnd >= (uint32_t) acked)
2775 tp->snd_wnd -= acked;
2780 /* NB: sowwakeup_locked() does an implicit unlock. */
2781 sowwakeup_locked(so);
2783 /* Detect una wraparound. */
2784 if (!IN_RECOVERY(tp->t_flags) &&
2785 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2786 SEQ_LEQ(th->th_ack, tp->snd_recover))
2787 tp->snd_recover = th->th_ack - 1;
2788 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2789 if (IN_RECOVERY(tp->t_flags) &&
2790 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2791 EXIT_RECOVERY(tp->t_flags);
2793 tp->snd_una = th->th_ack;
2794 if (tp->t_flags & TF_SACK_PERMIT) {
2795 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2796 tp->snd_recover = tp->snd_una;
2798 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2799 tp->snd_nxt = tp->snd_una;
2801 switch (tp->t_state) {
2804 * In FIN_WAIT_1 STATE in addition to the processing
2805 * for the ESTABLISHED state if our FIN is now acknowledged
2806 * then enter FIN_WAIT_2.
2808 case TCPS_FIN_WAIT_1:
2809 if (ourfinisacked) {
2811 * If we can't receive any more
2812 * data, then closing user can proceed.
2813 * Starting the timer is contrary to the
2814 * specification, but if we don't get a FIN
2815 * we'll hang forever.
2818 * we should release the tp also, and use a
2821 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2822 soisdisconnected(so);
2823 tcp_timer_activate(tp, TT_2MSL,
2824 (tcp_fast_finwait2_recycle ?
2825 tcp_finwait2_timeout :
2828 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2833 * In CLOSING STATE in addition to the processing for
2834 * the ESTABLISHED state if the ACK acknowledges our FIN
2835 * then enter the TIME-WAIT state, otherwise ignore
2839 if (ourfinisacked) {
2847 * In LAST_ACK, we may still be waiting for data to drain
2848 * and/or to be acked, as well as for the ack of our FIN.
2849 * If our FIN is now acknowledged, delete the TCB,
2850 * enter the closed state and return.
2853 if (ourfinisacked) {
2862 INP_WLOCK_ASSERT(tp->t_inpcb);
2865 * Update window information.
2866 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2868 if ((thflags & TH_ACK) &&
2869 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2870 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2871 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2872 /* keep track of pure window updates */
2874 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2875 TCPSTAT_INC(tcps_rcvwinupd);
2876 tp->snd_wnd = tiwin;
2877 tp->snd_wl1 = th->th_seq;
2878 tp->snd_wl2 = th->th_ack;
2879 if (tp->snd_wnd > tp->max_sndwnd)
2880 tp->max_sndwnd = tp->snd_wnd;
2885 * Process segments with URG.
2887 if ((thflags & TH_URG) && th->th_urp &&
2888 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2890 * This is a kludge, but if we receive and accept
2891 * random urgent pointers, we'll crash in
2892 * soreceive. It's hard to imagine someone
2893 * actually wanting to send this much urgent data.
2895 SOCKBUF_LOCK(&so->so_rcv);
2896 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2897 th->th_urp = 0; /* XXX */
2898 thflags &= ~TH_URG; /* XXX */
2899 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2900 goto dodata; /* XXX */
2903 * If this segment advances the known urgent pointer,
2904 * then mark the data stream. This should not happen
2905 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2906 * a FIN has been received from the remote side.
2907 * In these states we ignore the URG.
2909 * According to RFC961 (Assigned Protocols),
2910 * the urgent pointer points to the last octet
2911 * of urgent data. We continue, however,
2912 * to consider it to indicate the first octet
2913 * of data past the urgent section as the original
2914 * spec states (in one of two places).
2916 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2917 tp->rcv_up = th->th_seq + th->th_urp;
2918 so->so_oobmark = sbavail(&so->so_rcv) +
2919 (tp->rcv_up - tp->rcv_nxt) - 1;
2920 if (so->so_oobmark == 0)
2921 so->so_rcv.sb_state |= SBS_RCVATMARK;
2923 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2925 SOCKBUF_UNLOCK(&so->so_rcv);
2927 * Remove out of band data so doesn't get presented to user.
2928 * This can happen independent of advancing the URG pointer,
2929 * but if two URG's are pending at once, some out-of-band
2930 * data may creep in... ick.
2932 if (th->th_urp <= (uint32_t)tlen &&
2933 !(so->so_options & SO_OOBINLINE)) {
2934 /* hdr drop is delayed */
2935 tcp_pulloutofband(so, th, m, drop_hdrlen);
2939 * If no out of band data is expected,
2940 * pull receive urgent pointer along
2941 * with the receive window.
2943 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2944 tp->rcv_up = tp->rcv_nxt;
2947 INP_WLOCK_ASSERT(tp->t_inpcb);
2950 * Process the segment text, merging it into the TCP sequencing queue,
2951 * and arranging for acknowledgment of receipt if necessary.
2952 * This process logically involves adjusting tp->rcv_wnd as data
2953 * is presented to the user (this happens in tcp_usrreq.c,
2954 * case PRU_RCVD). If a FIN has already been received on this
2955 * connection then we just ignore the text.
2957 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
2958 IS_FASTOPEN(tp->t_flags));
2959 if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
2960 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2961 tcp_seq save_start = th->th_seq;
2962 tcp_seq save_rnxt = tp->rcv_nxt;
2963 int save_tlen = tlen;
2964 m_adj(m, drop_hdrlen); /* delayed header drop */
2966 * Insert segment which includes th into TCP reassembly queue
2967 * with control block tp. Set thflags to whether reassembly now
2968 * includes a segment with FIN. This handles the common case
2969 * inline (segment is the next to be received on an established
2970 * connection, and the queue is empty), avoiding linkage into
2971 * and removal from the queue and repetition of various
2973 * Set DELACK for segments received in order, but ack
2974 * immediately when segments are out of order (so
2975 * fast retransmit can work).
2977 if (th->th_seq == tp->rcv_nxt &&
2979 (TCPS_HAVEESTABLISHED(tp->t_state) ||
2981 if (DELAY_ACK(tp, tlen) || tfo_syn)
2982 tp->t_flags |= TF_DELACK;
2984 tp->t_flags |= TF_ACKNOW;
2985 tp->rcv_nxt += tlen;
2986 thflags = th->th_flags & TH_FIN;
2987 TCPSTAT_INC(tcps_rcvpack);
2988 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2989 SOCKBUF_LOCK(&so->so_rcv);
2990 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2993 sbappendstream_locked(&so->so_rcv, m, 0);
2994 /* NB: sorwakeup_locked() does an implicit unlock. */
2995 sorwakeup_locked(so);
2998 * XXX: Due to the header drop above "th" is
2999 * theoretically invalid by now. Fortunately
3000 * m_adj() doesn't actually frees any mbufs
3001 * when trimming from the head.
3003 tcp_seq temp = save_start;
3004 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3005 tp->t_flags |= TF_ACKNOW;
3007 if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) {
3008 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3010 * DSACK actually handled in the fastpath
3013 tcp_update_sack_list(tp, save_start,
3014 save_start + save_tlen);
3015 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3016 if ((tp->rcv_numsacks >= 1) &&
3017 (tp->sackblks[0].end == save_start)) {
3019 * Partial overlap, recorded at todrop
3022 tcp_update_sack_list(tp,
3023 tp->sackblks[0].start,
3024 tp->sackblks[0].end);
3026 tcp_update_dsack_list(tp, save_start,
3027 save_start + save_tlen);
3029 } else if (tlen >= save_tlen) {
3030 /* Update of sackblks. */
3031 tcp_update_dsack_list(tp, save_start,
3032 save_start + save_tlen);
3033 } else if (tlen > 0) {
3034 tcp_update_dsack_list(tp, save_start,
3040 * Note the amount of data that peer has sent into
3041 * our window, in order to estimate the sender's
3045 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3046 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3048 len = so->so_rcv.sb_hiwat;
3056 * If FIN is received ACK the FIN and let the user know
3057 * that the connection is closing.
3059 if (thflags & TH_FIN) {
3060 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3063 * If connection is half-synchronized
3064 * (ie NEEDSYN flag on) then delay ACK,
3065 * so it may be piggybacked when SYN is sent.
3066 * Otherwise, since we received a FIN then no
3067 * more input can be expected, send ACK now.
3069 if (tp->t_flags & TF_NEEDSYN)
3070 tp->t_flags |= TF_DELACK;
3072 tp->t_flags |= TF_ACKNOW;
3075 switch (tp->t_state) {
3078 * In SYN_RECEIVED and ESTABLISHED STATES
3079 * enter the CLOSE_WAIT state.
3081 case TCPS_SYN_RECEIVED:
3082 tp->t_starttime = ticks;
3084 case TCPS_ESTABLISHED:
3085 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3089 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3090 * enter the CLOSING state.
3092 case TCPS_FIN_WAIT_1:
3093 tcp_state_change(tp, TCPS_CLOSING);
3097 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3098 * starting the time-wait timer, turning off the other
3101 case TCPS_FIN_WAIT_2:
3107 if (so->so_options & SO_DEBUG)
3108 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3111 TCP_PROBE3(debug__input, tp, th, m);
3114 * Return any desired output.
3116 if (needoutput || (tp->t_flags & TF_ACKNOW))
3117 (void) tp->t_fb->tfb_tcp_output(tp);
3120 INP_WLOCK_ASSERT(tp->t_inpcb);
3122 if (tp->t_flags & TF_DELACK) {
3123 tp->t_flags &= ~TF_DELACK;
3124 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3126 INP_WUNLOCK(tp->t_inpcb);
3131 * Generate an ACK dropping incoming segment if it occupies
3132 * sequence space, where the ACK reflects our state.
3134 * We can now skip the test for the RST flag since all
3135 * paths to this code happen after packets containing
3136 * RST have been dropped.
3138 * In the SYN-RECEIVED state, don't send an ACK unless the
3139 * segment we received passes the SYN-RECEIVED ACK test.
3140 * If it fails send a RST. This breaks the loop in the
3141 * "LAND" DoS attack, and also prevents an ACK storm
3142 * between two listening ports that have been sent forged
3143 * SYN segments, each with the source address of the other.
3145 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3146 (SEQ_GT(tp->snd_una, th->th_ack) ||
3147 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3148 rstreason = BANDLIM_RST_OPENPORT;
3152 if (so->so_options & SO_DEBUG)
3153 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3156 TCP_PROBE3(debug__input, tp, th, m);
3157 tp->t_flags |= TF_ACKNOW;
3158 (void) tp->t_fb->tfb_tcp_output(tp);
3159 INP_WUNLOCK(tp->t_inpcb);
3165 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3166 INP_WUNLOCK(tp->t_inpcb);
3168 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3173 * Drop space held by incoming segment and return.
3176 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3177 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3180 TCP_PROBE3(debug__input, tp, th, m);
3182 INP_WUNLOCK(tp->t_inpcb);
3187 * Issue RST and make ACK acceptable to originator of segment.
3188 * The mbuf must still include the original packet header.
3192 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3193 int tlen, int rstreason)
3199 struct ip6_hdr *ip6;
3203 INP_WLOCK_ASSERT(tp->t_inpcb);
3206 /* Don't bother if destination was broadcast/multicast. */
3207 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3210 if (mtod(m, struct ip *)->ip_v == 6) {
3211 ip6 = mtod(m, struct ip6_hdr *);
3212 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3213 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3215 /* IPv6 anycast check is done at tcp6_input() */
3218 #if defined(INET) && defined(INET6)
3223 ip = mtod(m, struct ip *);
3224 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3225 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3226 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3227 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3232 /* Perform bandwidth limiting. */
3233 if (badport_bandlim(rstreason) < 0)
3236 /* tcp_respond consumes the mbuf chain. */
3237 if (th->th_flags & TH_ACK) {
3238 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3239 th->th_ack, TH_RST);
3241 if (th->th_flags & TH_SYN)
3243 if (th->th_flags & TH_FIN)
3245 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3246 (tcp_seq)0, TH_RST|TH_ACK);
3254 * Parse TCP options and place in tcpopt.
3257 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3262 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3264 if (opt == TCPOPT_EOL)
3266 if (opt == TCPOPT_NOP)
3272 if (optlen < 2 || optlen > cnt)
3277 if (optlen != TCPOLEN_MAXSEG)
3279 if (!(flags & TO_SYN))
3281 to->to_flags |= TOF_MSS;
3282 bcopy((char *)cp + 2,
3283 (char *)&to->to_mss, sizeof(to->to_mss));
3284 to->to_mss = ntohs(to->to_mss);
3287 if (optlen != TCPOLEN_WINDOW)
3289 if (!(flags & TO_SYN))
3291 to->to_flags |= TOF_SCALE;
3292 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3294 case TCPOPT_TIMESTAMP:
3295 if (optlen != TCPOLEN_TIMESTAMP)
3297 to->to_flags |= TOF_TS;
3298 bcopy((char *)cp + 2,
3299 (char *)&to->to_tsval, sizeof(to->to_tsval));
3300 to->to_tsval = ntohl(to->to_tsval);
3301 bcopy((char *)cp + 6,
3302 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3303 to->to_tsecr = ntohl(to->to_tsecr);
3305 case TCPOPT_SIGNATURE:
3307 * In order to reply to a host which has set the
3308 * TCP_SIGNATURE option in its initial SYN, we have
3309 * to record the fact that the option was observed
3310 * here for the syncache code to perform the correct
3313 if (optlen != TCPOLEN_SIGNATURE)
3315 to->to_flags |= TOF_SIGNATURE;
3316 to->to_signature = cp + 2;
3318 case TCPOPT_SACK_PERMITTED:
3319 if (optlen != TCPOLEN_SACK_PERMITTED)
3321 if (!(flags & TO_SYN))
3325 to->to_flags |= TOF_SACKPERM;
3328 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3332 to->to_flags |= TOF_SACK;
3333 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3334 to->to_sacks = cp + 2;
3335 TCPSTAT_INC(tcps_sack_rcv_blocks);
3337 case TCPOPT_FAST_OPEN:
3339 * Cookie length validation is performed by the
3340 * server side cookie checking code or the client
3341 * side cookie cache update code.
3343 if (!(flags & TO_SYN))
3345 if (!V_tcp_fastopen_client_enable &&
3346 !V_tcp_fastopen_server_enable)
3348 to->to_flags |= TOF_FASTOPEN;
3349 to->to_tfo_len = optlen - 2;
3350 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3359 * Pull out of band byte out of a segment so
3360 * it doesn't appear in the user's data queue.
3361 * It is still reflected in the segment length for
3362 * sequencing purposes.
3365 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3368 int cnt = off + th->th_urp - 1;
3371 if (m->m_len > cnt) {
3372 char *cp = mtod(m, caddr_t) + cnt;
3373 struct tcpcb *tp = sototcpcb(so);
3375 INP_WLOCK_ASSERT(tp->t_inpcb);
3378 tp->t_oobflags |= TCPOOB_HAVEDATA;
3379 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3381 if (m->m_flags & M_PKTHDR)
3390 panic("tcp_pulloutofband");
3394 * Collect new round-trip time estimate
3395 * and update averages and current timeout.
3398 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3402 INP_WLOCK_ASSERT(tp->t_inpcb);
3404 TCPSTAT_INC(tcps_rttupdated);
3407 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3408 imax(0, rtt * 1000 / hz));
3410 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3412 * srtt is stored as fixed point with 5 bits after the
3413 * binary point (i.e., scaled by 8). The following magic
3414 * is equivalent to the smoothing algorithm in rfc793 with
3415 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3416 * point). Adjust rtt to origin 0.
3418 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3419 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3421 if ((tp->t_srtt += delta) <= 0)
3425 * We accumulate a smoothed rtt variance (actually, a
3426 * smoothed mean difference), then set the retransmit
3427 * timer to smoothed rtt + 4 times the smoothed variance.
3428 * rttvar is stored as fixed point with 4 bits after the
3429 * binary point (scaled by 16). The following is
3430 * equivalent to rfc793 smoothing with an alpha of .75
3431 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3432 * rfc793's wired-in beta.
3436 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3437 if ((tp->t_rttvar += delta) <= 0)
3439 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3440 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3443 * No rtt measurement yet - use the unsmoothed rtt.
3444 * Set the variance to half the rtt (so our first
3445 * retransmit happens at 3*rtt).
3447 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3448 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3449 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3455 * the retransmit should happen at rtt + 4 * rttvar.
3456 * Because of the way we do the smoothing, srtt and rttvar
3457 * will each average +1/2 tick of bias. When we compute
3458 * the retransmit timer, we want 1/2 tick of rounding and
3459 * 1 extra tick because of +-1/2 tick uncertainty in the
3460 * firing of the timer. The bias will give us exactly the
3461 * 1.5 tick we need. But, because the bias is
3462 * statistical, we have to test that we don't drop below
3463 * the minimum feasible timer (which is 2 ticks).
3465 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3466 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3469 * We received an ack for a packet that wasn't retransmitted;
3470 * it is probably safe to discard any error indications we've
3471 * received recently. This isn't quite right, but close enough
3472 * for now (a route might have failed after we sent a segment,
3473 * and the return path might not be symmetrical).
3475 tp->t_softerror = 0;
3479 * Determine a reasonable value for maxseg size.
3480 * If the route is known, check route for mtu.
3481 * If none, use an mss that can be handled on the outgoing interface
3482 * without forcing IP to fragment. If no route is found, route has no mtu,
3483 * or the destination isn't local, use a default, hopefully conservative
3484 * size (usually 512 or the default IP max size, but no more than the mtu
3485 * of the interface), as we can't discover anything about intervening
3486 * gateways or networks. We also initialize the congestion/slow start
3487 * window to be a single segment if the destination isn't local.
3488 * While looking at the routing entry, we also initialize other path-dependent
3489 * parameters from pre-set or cached values in the routing entry.
3491 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3492 * IP options, e.g. IPSEC data, since length of this data may vary, and
3493 * thus it is calculated for every segment separately in tcp_output().
3495 * NOTE that this routine is only called when we process an incoming
3496 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3497 * settings are handled in tcp_mssopt().
3500 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3501 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3504 uint32_t maxmtu = 0;
3505 struct inpcb *inp = tp->t_inpcb;
3506 struct hc_metrics_lite metrics;
3508 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3509 size_t min_protoh = isipv6 ?
3510 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3511 sizeof (struct tcpiphdr);
3513 const size_t min_protoh = sizeof(struct tcpiphdr);
3516 INP_WLOCK_ASSERT(tp->t_inpcb);
3518 if (mtuoffer != -1) {
3519 KASSERT(offer == -1, ("%s: conflict", __func__));
3520 offer = mtuoffer - min_protoh;
3526 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3527 tp->t_maxseg = V_tcp_v6mssdflt;
3530 #if defined(INET) && defined(INET6)
3535 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3536 tp->t_maxseg = V_tcp_mssdflt;
3541 * No route to sender, stay with default mss and return.
3545 * In case we return early we need to initialize metrics
3546 * to a defined state as tcp_hc_get() would do for us
3547 * if there was no cache hit.
3549 if (metricptr != NULL)
3550 bzero(metricptr, sizeof(struct hc_metrics_lite));
3554 /* What have we got? */
3558 * Offer == 0 means that there was no MSS on the SYN
3559 * segment, in this case we use tcp_mssdflt as
3560 * already assigned to t_maxseg above.
3562 offer = tp->t_maxseg;
3567 * Offer == -1 means that we didn't receive SYN yet.
3573 * Prevent DoS attack with too small MSS. Round up
3574 * to at least minmss.
3576 offer = max(offer, V_tcp_minmss);
3580 * rmx information is now retrieved from tcp_hostcache.
3582 tcp_hc_get(&inp->inp_inc, &metrics);
3583 if (metricptr != NULL)
3584 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3587 * If there's a discovered mtu in tcp hostcache, use it.
3588 * Else, use the link mtu.
3590 if (metrics.rmx_mtu)
3591 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3595 mss = maxmtu - min_protoh;
3596 if (!V_path_mtu_discovery &&
3597 !in6_localaddr(&inp->in6p_faddr))
3598 mss = min(mss, V_tcp_v6mssdflt);
3601 #if defined(INET) && defined(INET6)
3606 mss = maxmtu - min_protoh;
3607 if (!V_path_mtu_discovery &&
3608 !in_localaddr(inp->inp_faddr))
3609 mss = min(mss, V_tcp_mssdflt);
3613 * XXX - The above conditional (mss = maxmtu - min_protoh)
3614 * probably violates the TCP spec.
3615 * The problem is that, since we don't know the
3616 * other end's MSS, we are supposed to use a conservative
3617 * default. But, if we do that, then MTU discovery will
3618 * never actually take place, because the conservative
3619 * default is much less than the MTUs typically seen
3620 * on the Internet today. For the moment, we'll sweep
3621 * this under the carpet.
3623 * The conservative default might not actually be a problem
3624 * if the only case this occurs is when sending an initial
3625 * SYN with options and data to a host we've never talked
3626 * to before. Then, they will reply with an MSS value which
3627 * will get recorded and the new parameters should get
3628 * recomputed. For Further Study.
3631 mss = min(mss, offer);
3634 * Sanity check: make sure that maxseg will be large
3635 * enough to allow some data on segments even if the
3636 * all the option space is used (40bytes). Otherwise
3637 * funny things may happen in tcp_output.
3639 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3647 tcp_mss(struct tcpcb *tp, int offer)
3653 struct hc_metrics_lite metrics;
3654 struct tcp_ifcap cap;
3656 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3658 bzero(&cap, sizeof(cap));
3659 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3665 * If there's a pipesize, change the socket buffer to that size,
3666 * don't change if sb_hiwat is different than default (then it
3667 * has been changed on purpose with setsockopt).
3668 * Make the socket buffers an integral number of mss units;
3669 * if the mss is larger than the socket buffer, decrease the mss.
3671 so = inp->inp_socket;
3672 SOCKBUF_LOCK(&so->so_snd);
3673 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3674 bufsize = metrics.rmx_sendpipe;
3676 bufsize = so->so_snd.sb_hiwat;
3680 bufsize = roundup(bufsize, mss);
3681 if (bufsize > sb_max)
3683 if (bufsize > so->so_snd.sb_hiwat)
3684 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3686 SOCKBUF_UNLOCK(&so->so_snd);
3688 * Sanity check: make sure that maxseg will be large
3689 * enough to allow some data on segments even if the
3690 * all the option space is used (40bytes). Otherwise
3691 * funny things may happen in tcp_output.
3693 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3695 tp->t_maxseg = max(mss, 64);
3697 SOCKBUF_LOCK(&so->so_rcv);
3698 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3699 bufsize = metrics.rmx_recvpipe;
3701 bufsize = so->so_rcv.sb_hiwat;
3702 if (bufsize > mss) {
3703 bufsize = roundup(bufsize, mss);
3704 if (bufsize > sb_max)
3706 if (bufsize > so->so_rcv.sb_hiwat)
3707 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3709 SOCKBUF_UNLOCK(&so->so_rcv);
3711 /* Check the interface for TSO capabilities. */
3712 if (cap.ifcap & CSUM_TSO) {
3713 tp->t_flags |= TF_TSO;
3714 tp->t_tsomax = cap.tsomax;
3715 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3716 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3721 * Determine the MSS option to send on an outgoing SYN.
3724 tcp_mssopt(struct in_conninfo *inc)
3727 uint32_t thcmtu = 0;
3728 uint32_t maxmtu = 0;
3731 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3734 if (inc->inc_flags & INC_ISIPV6) {
3735 mss = V_tcp_v6mssdflt;
3736 maxmtu = tcp_maxmtu6(inc, NULL);
3737 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3740 #if defined(INET) && defined(INET6)
3745 mss = V_tcp_mssdflt;
3746 maxmtu = tcp_maxmtu(inc, NULL);
3747 min_protoh = sizeof(struct tcpiphdr);
3750 #if defined(INET6) || defined(INET)
3751 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3754 if (maxmtu && thcmtu)
3755 mss = min(maxmtu, thcmtu) - min_protoh;
3756 else if (maxmtu || thcmtu)
3757 mss = max(maxmtu, thcmtu) - min_protoh;
3764 * On a partial ack arrives, force the retransmission of the
3765 * next unacknowledged segment. Do not clear tp->t_dupacks.
3766 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3770 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3772 tcp_seq onxt = tp->snd_nxt;
3773 uint32_t ocwnd = tp->snd_cwnd;
3774 u_int maxseg = tcp_maxseg(tp);
3776 INP_WLOCK_ASSERT(tp->t_inpcb);
3778 tcp_timer_activate(tp, TT_REXMT, 0);
3780 tp->snd_nxt = th->th_ack;
3782 * Set snd_cwnd to one segment beyond acknowledged offset.
3783 * (tp->snd_una has not yet been updated when this function is called.)
3785 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3786 tp->t_flags |= TF_ACKNOW;
3787 (void) tp->t_fb->tfb_tcp_output(tp);
3788 tp->snd_cwnd = ocwnd;
3789 if (SEQ_GT(onxt, tp->snd_nxt))
3792 * Partial window deflation. Relies on fact that tp->snd_una
3795 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3796 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3799 tp->snd_cwnd += maxseg;
3803 tcp_compute_pipe(struct tcpcb *tp)
3805 return (tp->snd_max - tp->snd_una +
3806 tp->sackhint.sack_bytes_rexmit -
3807 tp->sackhint.sacked_bytes);
3811 tcp_compute_initwnd(uint32_t maxseg)
3814 * Calculate the Initial Window, also used as Restart Window
3816 * RFC5681 Section 3.1 specifies the default conservative values.
3817 * RFC3390 specifies slightly more aggressive values.
3818 * RFC6928 increases it to ten segments.
3819 * Support for user specified value for initial flight size.
3821 if (V_tcp_initcwnd_segments)
3822 return min(V_tcp_initcwnd_segments * maxseg,
3823 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
3824 else if (V_tcp_do_rfc3390)
3825 return min(4 * maxseg, max(2 * maxseg, 4380));
3827 /* Per RFC5681 Section 3.1 */
3829 return (2 * maxseg);
3830 else if (maxseg > 1095)
3831 return (3 * maxseg);
3833 return (4 * maxseg);