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;
528 tp->ccv->flags &= ~CCF_IPHDR_CE;
531 tp->ccv->flags &= ~CCF_IPHDR_CE;
535 if (th->th_flags & TH_CWR)
536 tp->ccv->flags |= CCF_TCPHDR_CWR;
538 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
540 if (tp->t_flags & TF_DELACK)
541 tp->ccv->flags |= CCF_DELACK;
543 tp->ccv->flags &= ~CCF_DELACK;
545 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
547 if (tp->ccv->flags & CCF_ACKNOW)
548 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
553 * TCP input handling is split into multiple parts:
554 * tcp6_input is a thin wrapper around tcp_input for the extended
555 * ip6_protox[] call format in ip6_input
556 * tcp_input handles primary segment validation, inpcb lookup and
557 * SYN processing on listen sockets
558 * tcp_do_segment processes the ACK and text of the segment for
559 * establishing, established and closing connections
563 tcp6_input(struct mbuf **mp, int *offp, int proto)
566 struct in6_ifaddr *ia6;
570 if (m->m_len < *offp + sizeof(struct tcphdr)) {
571 m = m_pullup(m, *offp + sizeof(struct tcphdr));
574 TCPSTAT_INC(tcps_rcvshort);
575 return (IPPROTO_DONE);
580 * draft-itojun-ipv6-tcp-to-anycast
581 * better place to put this in?
583 ip6 = mtod(m, struct ip6_hdr *);
584 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
585 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
587 ifa_free(&ia6->ia_ifa);
588 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
589 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
591 return (IPPROTO_DONE);
594 ifa_free(&ia6->ia_ifa);
597 return (tcp_input(mp, offp, proto));
602 tcp_input(struct mbuf **mp, int *offp, int proto)
604 struct mbuf *m = *mp;
605 struct tcphdr *th = NULL;
606 struct ip *ip = NULL;
607 struct inpcb *inp = NULL;
608 struct tcpcb *tp = NULL;
609 struct socket *so = NULL;
620 int rstreason = 0; /* For badport_bandlim accounting purposes */
622 struct m_tag *fwd_tag = NULL;
624 struct ip6_hdr *ip6 = NULL;
627 const void *ip6 = NULL;
629 struct tcpopt to; /* options in this segment */
630 char *s = NULL; /* address and port logging */
633 * The size of tcp_saveipgen must be the size of the max ip header,
636 u_char tcp_saveipgen[IP6_HDR_LEN];
637 struct tcphdr tcp_savetcp;
644 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
651 TCPSTAT_INC(tcps_rcvtotal);
656 ip6 = mtod(m, struct ip6_hdr *);
657 th = (struct tcphdr *)((caddr_t)ip6 + off0);
658 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
659 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
660 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
661 th->th_sum = m->m_pkthdr.csum_data;
663 th->th_sum = in6_cksum_pseudo(ip6, tlen,
664 IPPROTO_TCP, m->m_pkthdr.csum_data);
665 th->th_sum ^= 0xffff;
667 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
669 TCPSTAT_INC(tcps_rcvbadsum);
674 * Be proactive about unspecified IPv6 address in source.
675 * As we use all-zero to indicate unbounded/unconnected pcb,
676 * unspecified IPv6 address can be used to confuse us.
678 * Note that packets with unspecified IPv6 destination is
679 * already dropped in ip6_input.
681 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
685 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
688 #if defined(INET) && defined(INET6)
694 * Get IP and TCP header together in first mbuf.
695 * Note: IP leaves IP header in first mbuf.
697 if (off0 > sizeof (struct ip)) {
699 off0 = sizeof(struct ip);
701 if (m->m_len < sizeof (struct tcpiphdr)) {
702 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
704 TCPSTAT_INC(tcps_rcvshort);
705 return (IPPROTO_DONE);
708 ip = mtod(m, struct ip *);
709 th = (struct tcphdr *)((caddr_t)ip + off0);
710 tlen = ntohs(ip->ip_len) - off0;
713 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
714 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
715 th->th_sum = m->m_pkthdr.csum_data;
717 th->th_sum = in_pseudo(ip->ip_src.s_addr,
719 htonl(m->m_pkthdr.csum_data + tlen +
721 th->th_sum ^= 0xffff;
723 struct ipovly *ipov = (struct ipovly *)ip;
726 * Checksum extended TCP header and data.
730 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
731 ipov->ih_len = htons(tlen);
732 th->th_sum = in_cksum(m, len);
733 /* Reset length for SDT probes. */
734 ip->ip_len = htons(len);
737 /* Re-initialization for later version check */
739 ip->ip_v = IPVERSION;
740 ip->ip_hl = off0 >> 2;
744 TCPSTAT_INC(tcps_rcvbadsum);
751 * Check that TCP offset makes sense,
752 * pull out TCP options and adjust length. XXX
754 off = th->th_off << 2;
755 if (off < sizeof (struct tcphdr) || off > tlen) {
756 TCPSTAT_INC(tcps_rcvbadoff);
759 tlen -= off; /* tlen is used instead of ti->ti_len */
760 if (off > sizeof (struct tcphdr)) {
763 if (m->m_len < off0 + off) {
764 m = m_pullup(m, off0 + off);
766 TCPSTAT_INC(tcps_rcvshort);
767 return (IPPROTO_DONE);
770 ip6 = mtod(m, struct ip6_hdr *);
771 th = (struct tcphdr *)((caddr_t)ip6 + off0);
774 #if defined(INET) && defined(INET6)
779 if (m->m_len < sizeof(struct ip) + off) {
780 if ((m = m_pullup(m, sizeof (struct ip) + off))
782 TCPSTAT_INC(tcps_rcvshort);
783 return (IPPROTO_DONE);
785 ip = mtod(m, struct ip *);
786 th = (struct tcphdr *)((caddr_t)ip + off0);
790 optlen = off - sizeof (struct tcphdr);
791 optp = (u_char *)(th + 1);
793 thflags = th->th_flags;
796 * Convert TCP protocol specific fields to host format.
798 tcp_fields_to_host(th);
801 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
803 drop_hdrlen = off0 + off;
806 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
810 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
812 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
815 #if defined(INET) && !defined(INET6)
816 (m->m_flags & M_IP_NEXTHOP)
819 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
823 if (isipv6 && fwd_tag != NULL) {
824 struct sockaddr_in6 *next_hop6;
826 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
828 * Transparently forwarded. Pretend to be the destination.
829 * Already got one like this?
831 inp = in6_pcblookup_mbuf(&V_tcbinfo,
832 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
833 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
836 * It's new. Try to find the ambushing socket.
837 * Because we've rewritten the destination address,
838 * any hardware-generated hash is ignored.
840 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
841 th->th_sport, &next_hop6->sin6_addr,
842 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
843 th->th_dport, INPLOOKUP_WILDCARD |
844 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
847 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
848 th->th_sport, &ip6->ip6_dst, th->th_dport,
849 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
850 m->m_pkthdr.rcvif, m);
853 #if defined(INET6) && defined(INET)
857 if (fwd_tag != NULL) {
858 struct sockaddr_in *next_hop;
860 next_hop = (struct sockaddr_in *)(fwd_tag+1);
862 * Transparently forwarded. Pretend to be the destination.
863 * already got one like this?
865 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
866 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
867 m->m_pkthdr.rcvif, m);
870 * It's new. Try to find the ambushing socket.
871 * Because we've rewritten the destination address,
872 * any hardware-generated hash is ignored.
874 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
875 th->th_sport, next_hop->sin_addr,
876 next_hop->sin_port ? ntohs(next_hop->sin_port) :
877 th->th_dport, INPLOOKUP_WILDCARD |
878 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
881 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
882 th->th_sport, ip->ip_dst, th->th_dport,
883 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
884 m->m_pkthdr.rcvif, m);
888 * If the INPCB does not exist then all data in the incoming
889 * segment is discarded and an appropriate RST is sent back.
890 * XXX MRT Send RST using which routing table?
894 * Log communication attempts to ports that are not
897 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
898 tcp_log_in_vain == 2) {
899 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
900 log(LOG_INFO, "%s; %s: Connection attempt "
901 "to closed port\n", s, __func__);
904 * When blackholing do not respond with a RST but
905 * completely ignore the segment and drop it.
907 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
911 rstreason = BANDLIM_RST_CLOSEDPORT;
914 INP_WLOCK_ASSERT(inp);
916 * While waiting for inp lock during the lookup, another thread
917 * can have dropped the inpcb, in which case we need to loop back
918 * and try to find a new inpcb to deliver to.
920 if (inp->inp_flags & INP_DROPPED) {
925 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
926 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
927 ((inp->inp_socket == NULL) ||
928 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
929 inp->inp_flowid = m->m_pkthdr.flowid;
930 inp->inp_flowtype = M_HASHTYPE_GET(m);
932 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
934 if (isipv6 && IPSEC_ENABLED(ipv6) &&
935 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
943 if (IPSEC_ENABLED(ipv4) &&
944 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
951 * Check the minimum TTL for socket.
953 if (inp->inp_ip_minttl != 0) {
956 if (inp->inp_ip_minttl > ip6->ip6_hlim)
960 if (inp->inp_ip_minttl > ip->ip_ttl)
965 * A previous connection in TIMEWAIT state is supposed to catch stray
966 * or duplicate segments arriving late. If this segment was a
967 * legitimate new connection attempt, the old INPCB gets removed and
968 * we can try again to find a listening socket.
970 * At this point, due to earlier optimism, we may hold only an inpcb
971 * lock, and not the inpcbinfo write lock. If so, we need to try to
972 * acquire it, or if that fails, acquire a reference on the inpcb,
973 * drop all locks, acquire a global write lock, and then re-acquire
974 * the inpcb lock. We may at that point discover that another thread
975 * has tried to free the inpcb, in which case we need to loop back
976 * and try to find a new inpcb to deliver to.
978 * XXXRW: It may be time to rethink timewait locking.
980 if (inp->inp_flags & INP_TIMEWAIT) {
981 if (thflags & TH_SYN)
982 tcp_dooptions(&to, optp, optlen, TO_SYN);
984 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
986 if (tcp_twcheck(inp, &to, th, m, tlen))
988 return (IPPROTO_DONE);
991 * The TCPCB may no longer exist if the connection is winding
992 * down or it is in the CLOSED state. Either way we drop the
993 * segment and send an appropriate response.
996 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
997 rstreason = BANDLIM_RST_CLOSEDPORT;
1002 if (tp->t_flags & TF_TOE) {
1003 tcp_offload_input(tp, m);
1004 m = NULL; /* consumed by the TOE driver */
1010 INP_WLOCK_ASSERT(inp);
1011 if (mac_inpcb_check_deliver(inp, m))
1014 so = inp->inp_socket;
1015 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1017 if (so->so_options & SO_DEBUG) {
1018 ostate = tp->t_state;
1021 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1024 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1027 #endif /* TCPDEBUG */
1029 * When the socket is accepting connections (the INPCB is in LISTEN
1030 * state) we look into the SYN cache if this is a new connection
1031 * attempt or the completion of a previous one.
1033 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1034 ("%s: so accepting but tp %p not listening", __func__, tp));
1035 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1036 struct in_conninfo inc;
1038 bzero(&inc, sizeof(inc));
1041 inc.inc_flags |= INC_ISIPV6;
1042 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1043 inc.inc_flags |= INC_IPV6MINMTU;
1044 inc.inc6_faddr = ip6->ip6_src;
1045 inc.inc6_laddr = ip6->ip6_dst;
1049 inc.inc_faddr = ip->ip_src;
1050 inc.inc_laddr = ip->ip_dst;
1052 inc.inc_fport = th->th_sport;
1053 inc.inc_lport = th->th_dport;
1054 inc.inc_fibnum = so->so_fibnum;
1057 * Check for an existing connection attempt in syncache if
1058 * the flag is only ACK. A successful lookup creates a new
1059 * socket appended to the listen queue in SYN_RECEIVED state.
1061 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1064 * Parse the TCP options here because
1065 * syncookies need access to the reflected
1068 tcp_dooptions(&to, optp, optlen, 0);
1070 * NB: syncache_expand() doesn't unlock
1071 * inp and tcpinfo locks.
1073 rstreason = syncache_expand(&inc, &to, th, &so, m);
1074 if (rstreason < 0) {
1076 * A failing TCP MD5 signature comparison
1077 * must result in the segment being dropped
1078 * and must not produce any response back
1082 } else if (rstreason == 0) {
1084 * No syncache entry or ACK was not
1085 * for our SYN/ACK. Send a RST.
1086 * NB: syncache did its own logging
1087 * of the failure cause.
1089 rstreason = BANDLIM_RST_OPENPORT;
1095 * We completed the 3-way handshake
1096 * but could not allocate a socket
1097 * either due to memory shortage,
1098 * listen queue length limits or
1099 * global socket limits. Send RST
1100 * or wait and have the remote end
1101 * retransmit the ACK for another
1104 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1105 log(LOG_DEBUG, "%s; %s: Listen socket: "
1106 "Socket allocation failed due to "
1107 "limits or memory shortage, %s\n",
1109 V_tcp_sc_rst_sock_fail ?
1110 "sending RST" : "try again");
1111 if (V_tcp_sc_rst_sock_fail) {
1112 rstreason = BANDLIM_UNLIMITED;
1118 * Socket is created in state SYN_RECEIVED.
1119 * Unlock the listen socket, lock the newly
1120 * created socket and update the tp variable.
1122 INP_WUNLOCK(inp); /* listen socket */
1123 inp = sotoinpcb(so);
1125 * New connection inpcb is already locked by
1126 * syncache_expand().
1128 INP_WLOCK_ASSERT(inp);
1129 tp = intotcpcb(inp);
1130 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1131 ("%s: ", __func__));
1133 * Process the segment and the data it
1134 * contains. tcp_do_segment() consumes
1135 * the mbuf chain and unlocks the inpcb.
1137 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1138 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1140 return (IPPROTO_DONE);
1143 * Segment flag validation for new connection attempts:
1145 * Our (SYN|ACK) response was rejected.
1146 * Check with syncache and remove entry to prevent
1149 * NB: syncache_chkrst does its own logging of failure
1152 if (thflags & TH_RST) {
1153 syncache_chkrst(&inc, th, m);
1157 * We can't do anything without SYN.
1159 if ((thflags & TH_SYN) == 0) {
1160 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1161 log(LOG_DEBUG, "%s; %s: Listen socket: "
1162 "SYN is missing, segment ignored\n",
1164 TCPSTAT_INC(tcps_badsyn);
1168 * (SYN|ACK) is bogus on a listen socket.
1170 if (thflags & TH_ACK) {
1171 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1172 log(LOG_DEBUG, "%s; %s: Listen socket: "
1173 "SYN|ACK invalid, segment rejected\n",
1175 syncache_badack(&inc); /* XXX: Not needed! */
1176 TCPSTAT_INC(tcps_badsyn);
1177 rstreason = BANDLIM_RST_OPENPORT;
1181 * If the drop_synfin option is enabled, drop all
1182 * segments with both the SYN and FIN bits set.
1183 * This prevents e.g. nmap from identifying the
1185 * XXX: Poor reasoning. nmap has other methods
1186 * and is constantly refining its stack detection
1188 * XXX: This is a violation of the TCP specification
1189 * and was used by RFC1644.
1191 if ((thflags & TH_FIN) && V_drop_synfin) {
1192 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1193 log(LOG_DEBUG, "%s; %s: Listen socket: "
1194 "SYN|FIN segment ignored (based on "
1195 "sysctl setting)\n", s, __func__);
1196 TCPSTAT_INC(tcps_badsyn);
1200 * Segment's flags are (SYN) or (SYN|FIN).
1202 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1203 * as they do not affect the state of the TCP FSM.
1204 * The data pointed to by TH_URG and th_urp is ignored.
1206 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1207 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1208 KASSERT(thflags & (TH_SYN),
1209 ("%s: Listen socket: TH_SYN not set", __func__));
1212 * If deprecated address is forbidden,
1213 * we do not accept SYN to deprecated interface
1214 * address to prevent any new inbound connection from
1215 * getting established.
1216 * When we do not accept SYN, we send a TCP RST,
1217 * with deprecated source address (instead of dropping
1218 * it). We compromise it as it is much better for peer
1219 * to send a RST, and RST will be the final packet
1222 * If we do not forbid deprecated addresses, we accept
1223 * the SYN packet. RFC2462 does not suggest dropping
1225 * If we decipher RFC2462 5.5.4, it says like this:
1226 * 1. use of deprecated addr with existing
1227 * communication is okay - "SHOULD continue to be
1229 * 2. use of it with new communication:
1230 * (2a) "SHOULD NOT be used if alternate address
1231 * with sufficient scope is available"
1232 * (2b) nothing mentioned otherwise.
1233 * Here we fall into (2b) case as we have no choice in
1234 * our source address selection - we must obey the peer.
1236 * The wording in RFC2462 is confusing, and there are
1237 * multiple description text for deprecated address
1238 * handling - worse, they are not exactly the same.
1239 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1241 if (isipv6 && !V_ip6_use_deprecated) {
1242 struct in6_ifaddr *ia6;
1244 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1246 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1247 ifa_free(&ia6->ia_ifa);
1248 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1249 log(LOG_DEBUG, "%s; %s: Listen socket: "
1250 "Connection attempt to deprecated "
1251 "IPv6 address rejected\n",
1253 rstreason = BANDLIM_RST_OPENPORT;
1257 ifa_free(&ia6->ia_ifa);
1261 * Basic sanity checks on incoming SYN requests:
1262 * Don't respond if the destination is a link layer
1263 * broadcast according to RFC1122 4.2.3.10, p. 104.
1264 * If it is from this socket it must be forged.
1265 * Don't respond if the source or destination is a
1266 * global or subnet broad- or multicast address.
1267 * Note that it is quite possible to receive unicast
1268 * link-layer packets with a broadcast IP address. Use
1269 * in_broadcast() to find them.
1271 if (m->m_flags & (M_BCAST|M_MCAST)) {
1272 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1273 log(LOG_DEBUG, "%s; %s: Listen socket: "
1274 "Connection attempt from broad- or multicast "
1275 "link layer address ignored\n", s, __func__);
1280 if (th->th_dport == th->th_sport &&
1281 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1282 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1283 log(LOG_DEBUG, "%s; %s: Listen socket: "
1284 "Connection attempt to/from self "
1285 "ignored\n", s, __func__);
1288 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1289 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1290 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1291 log(LOG_DEBUG, "%s; %s: Listen socket: "
1292 "Connection attempt from/to multicast "
1293 "address ignored\n", s, __func__);
1298 #if defined(INET) && defined(INET6)
1303 if (th->th_dport == th->th_sport &&
1304 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1305 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1306 log(LOG_DEBUG, "%s; %s: Listen socket: "
1307 "Connection attempt from/to self "
1308 "ignored\n", s, __func__);
1311 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1312 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1313 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1314 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1315 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1316 log(LOG_DEBUG, "%s; %s: Listen socket: "
1317 "Connection attempt from/to broad- "
1318 "or multicast address ignored\n",
1325 * SYN appears to be valid. Create compressed TCP state
1329 if (so->so_options & SO_DEBUG)
1330 tcp_trace(TA_INPUT, ostate, tp,
1331 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1333 TCP_PROBE3(debug__input, tp, th, m);
1334 tcp_dooptions(&to, optp, optlen, TO_SYN);
1335 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos))
1336 goto tfo_socket_result;
1339 * Entry added to syncache and mbuf consumed.
1340 * Only the listen socket is unlocked by syncache_add().
1342 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1343 return (IPPROTO_DONE);
1344 } else if (tp->t_state == TCPS_LISTEN) {
1346 * When a listen socket is torn down the SO_ACCEPTCONN
1347 * flag is removed first while connections are drained
1348 * from the accept queue in a unlock/lock cycle of the
1349 * ACCEPT_LOCK, opening a race condition allowing a SYN
1350 * attempt go through unhandled.
1354 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1355 if (tp->t_flags & TF_SIGNATURE) {
1356 tcp_dooptions(&to, optp, optlen, thflags);
1357 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1358 TCPSTAT_INC(tcps_sig_err_nosigopt);
1361 if (!TCPMD5_ENABLED() ||
1362 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1366 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1369 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1370 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1371 * the inpcb, and unlocks pcbinfo.
1373 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos);
1374 return (IPPROTO_DONE);
1377 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1380 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1383 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1384 m = NULL; /* mbuf chain got consumed. */
1389 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1395 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo);
1400 return (IPPROTO_DONE);
1404 * Automatic sizing of receive socket buffer. Often the send
1405 * buffer size is not optimally adjusted to the actual network
1406 * conditions at hand (delay bandwidth product). Setting the
1407 * buffer size too small limits throughput on links with high
1408 * bandwidth and high delay (eg. trans-continental/oceanic links).
1410 * On the receive side the socket buffer memory is only rarely
1411 * used to any significant extent. This allows us to be much
1412 * more aggressive in scaling the receive socket buffer. For
1413 * the case that the buffer space is actually used to a large
1414 * extent and we run out of kernel memory we can simply drop
1415 * the new segments; TCP on the sender will just retransmit it
1416 * later. Setting the buffer size too big may only consume too
1417 * much kernel memory if the application doesn't read() from
1418 * the socket or packet loss or reordering makes use of the
1421 * The criteria to step up the receive buffer one notch are:
1422 * 1. Application has not set receive buffer size with
1423 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1424 * 2. the number of bytes received during 1/2 of an sRTT
1425 * is at least 3/8 of the current socket buffer size.
1426 * 3. receive buffer size has not hit maximal automatic size;
1428 * If all of the criteria are met we increaset the socket buffer
1429 * by a 1/2 (bounded by the max). This allows us to keep ahead
1430 * of slow-start but also makes it so our peer never gets limited
1431 * by our rwnd which we then open up causing a burst.
1433 * This algorithm does two steps per RTT at most and only if
1434 * we receive a bulk stream w/o packet losses or reorderings.
1435 * Shrinking the buffer during idle times is not necessary as
1436 * it doesn't consume any memory when idle.
1438 * TODO: Only step up if the application is actually serving
1439 * the buffer to better manage the socket buffer resources.
1442 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1443 struct tcpcb *tp, int tlen)
1447 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1448 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1449 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1450 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1451 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1452 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1453 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1455 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1457 /* Start over with next RTT. */
1461 tp->rfbuf_cnt += tlen; /* add up */
1467 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1468 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos)
1470 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1471 int rstreason, todrop, win;
1475 struct in_conninfo *inc;
1482 * The size of tcp_saveipgen must be the size of the max ip header,
1485 u_char tcp_saveipgen[IP6_HDR_LEN];
1486 struct tcphdr tcp_savetcp;
1489 thflags = th->th_flags;
1490 inc = &tp->t_inpcb->inp_inc;
1491 tp->sackhint.last_sack_ack = 0;
1493 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1496 INP_WLOCK_ASSERT(tp->t_inpcb);
1497 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1499 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1503 /* Save segment, if requested. */
1504 tcp_pcap_add(th, m, &(tp->t_inpkts));
1506 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1509 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1510 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1511 log(LOG_DEBUG, "%s; %s: "
1512 "SYN|FIN segment ignored (based on "
1513 "sysctl setting)\n", s, __func__);
1520 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1521 * check SEQ.ACK first.
1523 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1524 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1525 rstreason = BANDLIM_UNLIMITED;
1530 * Segment received on connection.
1531 * Reset idle time and keep-alive timer.
1532 * XXX: This should be done after segment
1533 * validation to ignore broken/spoofed segs.
1535 tp->t_rcvtime = ticks;
1538 * Scale up the window into a 32-bit value.
1539 * For the SYN_SENT state the scale is zero.
1541 tiwin = th->th_win << tp->snd_scale;
1543 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1547 * TCP ECN processing.
1549 if (tp->t_flags2 & TF2_ECN_PERMIT) {
1550 if (thflags & TH_CWR)
1551 tp->t_flags2 &= ~TF2_ECN_SND_ECE;
1552 switch (iptos & IPTOS_ECN_MASK) {
1554 tp->t_flags2 |= TF2_ECN_SND_ECE;
1555 TCPSTAT_INC(tcps_ecn_ce);
1557 case IPTOS_ECN_ECT0:
1558 TCPSTAT_INC(tcps_ecn_ect0);
1560 case IPTOS_ECN_ECT1:
1561 TCPSTAT_INC(tcps_ecn_ect1);
1565 /* Process a packet differently from RFC3168. */
1566 cc_ecnpkt_handler(tp, th, iptos);
1568 /* Congestion experienced. */
1569 if (thflags & TH_ECE) {
1570 cc_cong_signal(tp, th, CC_ECN);
1575 * Parse options on any incoming segment.
1577 tcp_dooptions(&to, (u_char *)(th + 1),
1578 (th->th_off << 2) - sizeof(struct tcphdr),
1579 (thflags & TH_SYN) ? TO_SYN : 0);
1581 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1582 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1583 (to.to_flags & TOF_SIGNATURE) == 0) {
1584 TCPSTAT_INC(tcps_sig_err_sigopt);
1585 /* XXX: should drop? */
1589 * If echoed timestamp is later than the current time,
1590 * fall back to non RFC1323 RTT calculation. Normalize
1591 * timestamp if syncookies were used when this connection
1594 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1595 to.to_tsecr -= tp->ts_offset;
1596 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1598 else if (tp->t_flags & TF_PREVVALID &&
1599 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
1600 cc_cong_signal(tp, th, CC_RTO_ERR);
1603 * Process options only when we get SYN/ACK back. The SYN case
1604 * for incoming connections is handled in tcp_syncache.
1605 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1606 * or <SYN,ACK>) segment itself is never scaled.
1607 * XXX this is traditional behavior, may need to be cleaned up.
1609 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1610 if ((to.to_flags & TOF_SCALE) &&
1611 (tp->t_flags & TF_REQ_SCALE)) {
1612 tp->t_flags |= TF_RCVD_SCALE;
1613 tp->snd_scale = to.to_wscale;
1616 * Initial send window. It will be updated with
1617 * the next incoming segment to the scaled value.
1619 tp->snd_wnd = th->th_win;
1620 if (to.to_flags & TOF_TS) {
1621 tp->t_flags |= TF_RCVD_TSTMP;
1622 tp->ts_recent = to.to_tsval;
1623 tp->ts_recent_age = tcp_ts_getticks();
1625 if (to.to_flags & TOF_MSS)
1626 tcp_mss(tp, to.to_mss);
1627 if ((tp->t_flags & TF_SACK_PERMIT) &&
1628 (to.to_flags & TOF_SACKPERM) == 0)
1629 tp->t_flags &= ~TF_SACK_PERMIT;
1630 if (IS_FASTOPEN(tp->t_flags)) {
1631 if (to.to_flags & TOF_FASTOPEN) {
1634 if (to.to_flags & TOF_MSS)
1637 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
1641 tcp_fastopen_update_cache(tp, mss,
1642 to.to_tfo_len, to.to_tfo_cookie);
1644 tcp_fastopen_disable_path(tp);
1649 * If timestamps were negotiated during SYN/ACK they should
1650 * appear on every segment during this session and vice versa.
1652 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1653 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1654 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1655 "no action\n", s, __func__);
1659 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1660 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1661 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1662 "no action\n", s, __func__);
1668 * Header prediction: check for the two common cases
1669 * of a uni-directional data xfer. If the packet has
1670 * no control flags, is in-sequence, the window didn't
1671 * change and we're not retransmitting, it's a
1672 * candidate. If the length is zero and the ack moved
1673 * forward, we're the sender side of the xfer. Just
1674 * free the data acked & wake any higher level process
1675 * that was blocked waiting for space. If the length
1676 * is non-zero and the ack didn't move, we're the
1677 * receiver side. If we're getting packets in-order
1678 * (the reassembly queue is empty), add the data to
1679 * the socket buffer and note that we need a delayed ack.
1680 * Make sure that the hidden state-flags are also off.
1681 * Since we check for TCPS_ESTABLISHED first, it can only
1684 if (tp->t_state == TCPS_ESTABLISHED &&
1685 th->th_seq == tp->rcv_nxt &&
1686 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1687 tp->snd_nxt == tp->snd_max &&
1688 tiwin && tiwin == tp->snd_wnd &&
1689 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1691 ((to.to_flags & TOF_TS) == 0 ||
1692 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1695 * If last ACK falls within this segment's sequence numbers,
1696 * record the timestamp.
1697 * NOTE that the test is modified according to the latest
1698 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1700 if ((to.to_flags & TOF_TS) != 0 &&
1701 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1702 tp->ts_recent_age = tcp_ts_getticks();
1703 tp->ts_recent = to.to_tsval;
1707 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1708 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1709 !IN_RECOVERY(tp->t_flags) &&
1710 (to.to_flags & TOF_SACK) == 0 &&
1711 TAILQ_EMPTY(&tp->snd_holes)) {
1713 * This is a pure ack for outstanding data.
1715 TCPSTAT_INC(tcps_predack);
1718 * "bad retransmit" recovery without timestamps.
1720 if ((to.to_flags & TOF_TS) == 0 &&
1721 tp->t_rxtshift == 1 &&
1722 tp->t_flags & TF_PREVVALID &&
1723 (int)(ticks - tp->t_badrxtwin) < 0) {
1724 cc_cong_signal(tp, th, CC_RTO_ERR);
1728 * Recalculate the transmit timer / rtt.
1730 * Some boxes send broken timestamp replies
1731 * during the SYN+ACK phase, ignore
1732 * timestamps of 0 or we could calculate a
1733 * huge RTT and blow up the retransmit timer.
1735 if ((to.to_flags & TOF_TS) != 0 &&
1739 t = tcp_ts_getticks() - to.to_tsecr;
1740 if (!tp->t_rttlow || tp->t_rttlow > t)
1743 TCP_TS_TO_TICKS(t) + 1);
1744 } else if (tp->t_rtttime &&
1745 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1746 if (!tp->t_rttlow ||
1747 tp->t_rttlow > ticks - tp->t_rtttime)
1748 tp->t_rttlow = ticks - tp->t_rtttime;
1750 ticks - tp->t_rtttime);
1752 acked = BYTES_THIS_ACK(tp, th);
1755 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1756 hhook_run_tcp_est_in(tp, th, &to);
1759 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1760 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1761 sbdrop(&so->so_snd, acked);
1762 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1763 SEQ_LEQ(th->th_ack, tp->snd_recover))
1764 tp->snd_recover = th->th_ack - 1;
1767 * Let the congestion control algorithm update
1768 * congestion control related information. This
1769 * typically means increasing the congestion
1772 cc_ack_received(tp, th, nsegs, CC_ACK);
1774 tp->snd_una = th->th_ack;
1776 * Pull snd_wl2 up to prevent seq wrap relative
1779 tp->snd_wl2 = th->th_ack;
1784 * If all outstanding data are acked, stop
1785 * retransmit timer, otherwise restart timer
1786 * using current (possibly backed-off) value.
1787 * If process is waiting for space,
1788 * wakeup/selwakeup/signal. If data
1789 * are ready to send, let tcp_output
1790 * decide between more output or persist.
1793 if (so->so_options & SO_DEBUG)
1794 tcp_trace(TA_INPUT, ostate, tp,
1795 (void *)tcp_saveipgen,
1798 TCP_PROBE3(debug__input, tp, th, m);
1799 if (tp->snd_una == tp->snd_max)
1800 tcp_timer_activate(tp, TT_REXMT, 0);
1801 else if (!tcp_timer_active(tp, TT_PERSIST))
1802 tcp_timer_activate(tp, TT_REXMT,
1805 if (sbavail(&so->so_snd))
1806 (void) tp->t_fb->tfb_tcp_output(tp);
1809 } else if (th->th_ack == tp->snd_una &&
1810 tlen <= sbspace(&so->so_rcv)) {
1811 int newsize = 0; /* automatic sockbuf scaling */
1814 * This is a pure, in-sequence data packet with
1815 * nothing on the reassembly queue and we have enough
1816 * buffer space to take it.
1818 /* Clean receiver SACK report if present */
1819 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1820 tcp_clean_sackreport(tp);
1821 TCPSTAT_INC(tcps_preddat);
1822 tp->rcv_nxt += tlen;
1824 * Pull snd_wl1 up to prevent seq wrap relative to
1827 tp->snd_wl1 = th->th_seq;
1829 * Pull rcv_up up to prevent seq wrap relative to
1832 tp->rcv_up = tp->rcv_nxt;
1833 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1834 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1836 if (so->so_options & SO_DEBUG)
1837 tcp_trace(TA_INPUT, ostate, tp,
1838 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1840 TCP_PROBE3(debug__input, tp, th, m);
1842 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1844 /* Add data to socket buffer. */
1845 SOCKBUF_LOCK(&so->so_rcv);
1846 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1850 * Set new socket buffer size.
1851 * Give up when limit is reached.
1854 if (!sbreserve_locked(&so->so_rcv,
1856 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1857 m_adj(m, drop_hdrlen); /* delayed header drop */
1858 sbappendstream_locked(&so->so_rcv, m, 0);
1860 /* NB: sorwakeup_locked() does an implicit unlock. */
1861 sorwakeup_locked(so);
1862 if (DELAY_ACK(tp, tlen)) {
1863 tp->t_flags |= TF_DELACK;
1865 tp->t_flags |= TF_ACKNOW;
1866 tp->t_fb->tfb_tcp_output(tp);
1873 * Calculate amount of space in receive window,
1874 * and then do TCP input processing.
1875 * Receive window is amount of space in rcv queue,
1876 * but not less than advertised window.
1878 win = sbspace(&so->so_rcv);
1881 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1883 switch (tp->t_state) {
1886 * If the state is SYN_RECEIVED:
1887 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1889 case TCPS_SYN_RECEIVED:
1890 if ((thflags & TH_ACK) &&
1891 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1892 SEQ_GT(th->th_ack, tp->snd_max))) {
1893 rstreason = BANDLIM_RST_OPENPORT;
1896 if (IS_FASTOPEN(tp->t_flags)) {
1898 * When a TFO connection is in SYN_RECEIVED, the
1899 * only valid packets are the initial SYN, a
1900 * retransmit/copy of the initial SYN (possibly with
1901 * a subset of the original data), a valid ACK, a
1904 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1905 rstreason = BANDLIM_RST_OPENPORT;
1907 } else if (thflags & TH_SYN) {
1908 /* non-initial SYN is ignored */
1909 if ((tcp_timer_active(tp, TT_DELACK) ||
1910 tcp_timer_active(tp, TT_REXMT)))
1912 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1919 * If the state is SYN_SENT:
1920 * if seg contains a RST with valid ACK (SEQ.ACK has already
1921 * been verified), then drop the connection.
1922 * if seg contains a RST without an ACK, drop the seg.
1923 * if seg does not contain SYN, then drop the seg.
1924 * Otherwise this is an acceptable SYN segment
1925 * initialize tp->rcv_nxt and tp->irs
1926 * if seg contains ack then advance tp->snd_una
1927 * if seg contains an ECE and ECN support is enabled, the stream
1929 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1930 * arrange for segment to be acked (eventually)
1931 * continue processing rest of data/controls, beginning with URG
1934 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1935 TCP_PROBE5(connect__refused, NULL, tp,
1937 tp = tcp_drop(tp, ECONNREFUSED);
1939 if (thflags & TH_RST)
1941 if (!(thflags & TH_SYN))
1944 tp->irs = th->th_seq;
1946 if (thflags & TH_ACK) {
1947 int tfo_partial_ack = 0;
1949 TCPSTAT_INC(tcps_connects);
1952 mac_socketpeer_set_from_mbuf(m, so);
1954 /* Do window scaling on this connection? */
1955 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1956 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1957 tp->rcv_scale = tp->request_r_scale;
1959 tp->rcv_adv += min(tp->rcv_wnd,
1960 TCP_MAXWIN << tp->rcv_scale);
1961 tp->snd_una++; /* SYN is acked */
1963 * If not all the data that was sent in the TFO SYN
1964 * has been acked, resend the remainder right away.
1966 if (IS_FASTOPEN(tp->t_flags) &&
1967 (tp->snd_una != tp->snd_max)) {
1968 tp->snd_nxt = th->th_ack;
1969 tfo_partial_ack = 1;
1972 * If there's data, delay ACK; if there's also a FIN
1973 * ACKNOW will be turned on later.
1975 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
1976 tcp_timer_activate(tp, TT_DELACK,
1979 tp->t_flags |= TF_ACKNOW;
1981 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) &&
1983 tp->t_flags2 |= TF2_ECN_PERMIT;
1984 TCPSTAT_INC(tcps_ecn_shs);
1988 * Received <SYN,ACK> in SYN_SENT[*] state.
1990 * SYN_SENT --> ESTABLISHED
1991 * SYN_SENT* --> FIN_WAIT_1
1993 tp->t_starttime = ticks;
1994 if (tp->t_flags & TF_NEEDFIN) {
1995 tcp_state_change(tp, TCPS_FIN_WAIT_1);
1996 tp->t_flags &= ~TF_NEEDFIN;
1999 tcp_state_change(tp, TCPS_ESTABLISHED);
2000 TCP_PROBE5(connect__established, NULL, tp,
2003 tcp_timer_activate(tp, TT_KEEP,
2008 * Received initial SYN in SYN-SENT[*] state =>
2009 * simultaneous open.
2010 * If it succeeds, connection is * half-synchronized.
2011 * Otherwise, do 3-way handshake:
2012 * SYN-SENT -> SYN-RECEIVED
2013 * SYN-SENT* -> SYN-RECEIVED*
2015 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2016 tcp_timer_activate(tp, TT_REXMT, 0);
2017 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2020 INP_WLOCK_ASSERT(tp->t_inpcb);
2023 * Advance th->th_seq to correspond to first data byte.
2024 * If data, trim to stay within window,
2025 * dropping FIN if necessary.
2028 if (tlen > tp->rcv_wnd) {
2029 todrop = tlen - tp->rcv_wnd;
2033 TCPSTAT_INC(tcps_rcvpackafterwin);
2034 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2036 tp->snd_wl1 = th->th_seq - 1;
2037 tp->rcv_up = th->th_seq;
2039 * Client side of transaction: already sent SYN and data.
2040 * If the remote host used T/TCP to validate the SYN,
2041 * our data will be ACK'd; if so, enter normal data segment
2042 * processing in the middle of step 5, ack processing.
2043 * Otherwise, goto step 6.
2045 if (thflags & TH_ACK)
2051 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2052 * do normal processing.
2054 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2058 break; /* continue normal processing */
2062 * States other than LISTEN or SYN_SENT.
2063 * First check the RST flag and sequence number since reset segments
2064 * are exempt from the timestamp and connection count tests. This
2065 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2066 * below which allowed reset segments in half the sequence space
2067 * to fall though and be processed (which gives forged reset
2068 * segments with a random sequence number a 50 percent chance of
2069 * killing a connection).
2070 * Then check timestamp, if present.
2071 * Then check the connection count, if present.
2072 * Then check that at least some bytes of segment are within
2073 * receive window. If segment begins before rcv_nxt,
2074 * drop leading data (and SYN); if nothing left, just ack.
2076 if (thflags & TH_RST) {
2078 * RFC5961 Section 3.2
2080 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2081 * - If RST is in window, we send challenge ACK.
2083 * Note: to take into account delayed ACKs, we should
2084 * test against last_ack_sent instead of rcv_nxt.
2085 * Note 2: we handle special case of closed window, not
2086 * covered by the RFC.
2088 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2089 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2090 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2092 KASSERT(tp->t_state != TCPS_SYN_SENT,
2093 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2096 if (V_tcp_insecure_rst ||
2097 tp->last_ack_sent == th->th_seq) {
2098 TCPSTAT_INC(tcps_drops);
2099 /* Drop the connection. */
2100 switch (tp->t_state) {
2101 case TCPS_SYN_RECEIVED:
2102 so->so_error = ECONNREFUSED;
2104 case TCPS_ESTABLISHED:
2105 case TCPS_FIN_WAIT_1:
2106 case TCPS_FIN_WAIT_2:
2107 case TCPS_CLOSE_WAIT:
2110 so->so_error = ECONNRESET;
2117 TCPSTAT_INC(tcps_badrst);
2118 /* Send challenge ACK. */
2119 tcp_respond(tp, mtod(m, void *), th, m,
2120 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2121 tp->last_ack_sent = tp->rcv_nxt;
2129 * RFC5961 Section 4.2
2130 * Send challenge ACK for any SYN in synchronized state.
2132 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2133 tp->t_state != TCPS_SYN_RECEIVED) {
2134 TCPSTAT_INC(tcps_badsyn);
2135 if (V_tcp_insecure_syn &&
2136 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2137 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2138 tp = tcp_drop(tp, ECONNRESET);
2139 rstreason = BANDLIM_UNLIMITED;
2141 /* Send challenge ACK. */
2142 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2143 tp->snd_nxt, TH_ACK);
2144 tp->last_ack_sent = tp->rcv_nxt;
2151 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2152 * and it's less than ts_recent, drop it.
2154 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2155 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2157 /* Check to see if ts_recent is over 24 days old. */
2158 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2160 * Invalidate ts_recent. If this segment updates
2161 * ts_recent, the age will be reset later and ts_recent
2162 * will get a valid value. If it does not, setting
2163 * ts_recent to zero will at least satisfy the
2164 * requirement that zero be placed in the timestamp
2165 * echo reply when ts_recent isn't valid. The
2166 * age isn't reset until we get a valid ts_recent
2167 * because we don't want out-of-order segments to be
2168 * dropped when ts_recent is old.
2172 TCPSTAT_INC(tcps_rcvduppack);
2173 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2174 TCPSTAT_INC(tcps_pawsdrop);
2182 * In the SYN-RECEIVED state, validate that the packet belongs to
2183 * this connection before trimming the data to fit the receive
2184 * window. Check the sequence number versus IRS since we know
2185 * the sequence numbers haven't wrapped. This is a partial fix
2186 * for the "LAND" DoS attack.
2188 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2189 rstreason = BANDLIM_RST_OPENPORT;
2193 todrop = tp->rcv_nxt - th->th_seq;
2195 if (thflags & TH_SYN) {
2205 * Following if statement from Stevens, vol. 2, p. 960.
2208 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2210 * Any valid FIN must be to the left of the window.
2211 * At this point the FIN must be a duplicate or out
2212 * of sequence; drop it.
2217 * Send an ACK to resynchronize and drop any data.
2218 * But keep on processing for RST or ACK.
2220 tp->t_flags |= TF_ACKNOW;
2222 TCPSTAT_INC(tcps_rcvduppack);
2223 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2225 TCPSTAT_INC(tcps_rcvpartduppack);
2226 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2229 * DSACK - add SACK block for dropped range
2231 if (tp->t_flags & TF_SACK_PERMIT) {
2232 tcp_update_sack_list(tp, th->th_seq,
2233 th->th_seq + todrop);
2235 * ACK now, as the next in-sequence segment
2236 * will clear the DSACK block again
2238 tp->t_flags |= TF_ACKNOW;
2240 drop_hdrlen += todrop; /* drop from the top afterwards */
2241 th->th_seq += todrop;
2243 if (th->th_urp > todrop)
2244 th->th_urp -= todrop;
2252 * If new data are received on a connection after the
2253 * user processes are gone, then RST the other end.
2255 if ((so->so_state & SS_NOFDREF) &&
2256 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2257 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2258 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2259 "after socket was closed, "
2260 "sending RST and removing tcpcb\n",
2261 s, __func__, tcpstates[tp->t_state], tlen);
2265 TCPSTAT_INC(tcps_rcvafterclose);
2266 rstreason = BANDLIM_UNLIMITED;
2271 * If segment ends after window, drop trailing data
2272 * (and PUSH and FIN); if nothing left, just ACK.
2274 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2276 TCPSTAT_INC(tcps_rcvpackafterwin);
2277 if (todrop >= tlen) {
2278 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2280 * If window is closed can only take segments at
2281 * window edge, and have to drop data and PUSH from
2282 * incoming segments. Continue processing, but
2283 * remember to ack. Otherwise, drop segment
2286 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2287 tp->t_flags |= TF_ACKNOW;
2288 TCPSTAT_INC(tcps_rcvwinprobe);
2292 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2295 thflags &= ~(TH_PUSH|TH_FIN);
2299 * If last ACK falls within this segment's sequence numbers,
2300 * record its timestamp.
2302 * 1) That the test incorporates suggestions from the latest
2303 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2304 * 2) That updating only on newer timestamps interferes with
2305 * our earlier PAWS tests, so this check should be solely
2306 * predicated on the sequence space of this segment.
2307 * 3) That we modify the segment boundary check to be
2308 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2309 * instead of RFC1323's
2310 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2311 * This modified check allows us to overcome RFC1323's
2312 * limitations as described in Stevens TCP/IP Illustrated
2313 * Vol. 2 p.869. In such cases, we can still calculate the
2314 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2316 if ((to.to_flags & TOF_TS) != 0 &&
2317 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2318 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2319 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2320 tp->ts_recent_age = tcp_ts_getticks();
2321 tp->ts_recent = to.to_tsval;
2325 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2326 * flag is on (half-synchronized state), then queue data for
2327 * later processing; else drop segment and return.
2329 if ((thflags & TH_ACK) == 0) {
2330 if (tp->t_state == TCPS_SYN_RECEIVED ||
2331 (tp->t_flags & TF_NEEDSYN)) {
2332 if (tp->t_state == TCPS_SYN_RECEIVED &&
2333 IS_FASTOPEN(tp->t_flags)) {
2334 tp->snd_wnd = tiwin;
2338 } else if (tp->t_flags & TF_ACKNOW)
2347 switch (tp->t_state) {
2350 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2351 * ESTABLISHED state and continue processing.
2352 * The ACK was checked above.
2354 case TCPS_SYN_RECEIVED:
2356 TCPSTAT_INC(tcps_connects);
2358 /* Do window scaling? */
2359 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2360 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2361 tp->rcv_scale = tp->request_r_scale;
2363 tp->snd_wnd = tiwin;
2366 * SYN-RECEIVED -> ESTABLISHED
2367 * SYN-RECEIVED* -> FIN-WAIT-1
2369 tp->t_starttime = ticks;
2370 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
2371 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2372 tp->t_tfo_pending = NULL;
2375 * Account for the ACK of our SYN prior to
2376 * regular ACK processing below.
2380 if (tp->t_flags & TF_NEEDFIN) {
2381 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2382 tp->t_flags &= ~TF_NEEDFIN;
2384 tcp_state_change(tp, TCPS_ESTABLISHED);
2385 TCP_PROBE5(accept__established, NULL, tp,
2388 * TFO connections call cc_conn_init() during SYN
2389 * processing. Calling it again here for such
2390 * connections is not harmless as it would undo the
2391 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2394 if (!IS_FASTOPEN(tp->t_flags))
2396 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2399 * If segment contains data or ACK, will call tcp_reass()
2400 * later; if not, do so now to pass queued data to user.
2402 if (tlen == 0 && (thflags & TH_FIN) == 0)
2403 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2405 tp->snd_wl1 = th->th_seq - 1;
2409 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2410 * ACKs. If the ack is in the range
2411 * tp->snd_una < th->th_ack <= tp->snd_max
2412 * then advance tp->snd_una to th->th_ack and drop
2413 * data from the retransmission queue. If this ACK reflects
2414 * more up to date window information we update our window information.
2416 case TCPS_ESTABLISHED:
2417 case TCPS_FIN_WAIT_1:
2418 case TCPS_FIN_WAIT_2:
2419 case TCPS_CLOSE_WAIT:
2422 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2423 TCPSTAT_INC(tcps_rcvacktoomuch);
2426 if ((tp->t_flags & TF_SACK_PERMIT) &&
2427 ((to.to_flags & TOF_SACK) ||
2428 !TAILQ_EMPTY(&tp->snd_holes)))
2429 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2432 * Reset the value so that previous (valid) value
2433 * from the last ack with SACK doesn't get used.
2435 tp->sackhint.sacked_bytes = 0;
2438 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2439 hhook_run_tcp_est_in(tp, th, &to);
2442 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2445 maxseg = tcp_maxseg(tp);
2447 (tiwin == tp->snd_wnd ||
2448 (tp->t_flags & TF_SACK_PERMIT))) {
2450 * If this is the first time we've seen a
2451 * FIN from the remote, this is not a
2452 * duplicate and it needs to be processed
2453 * normally. This happens during a
2454 * simultaneous close.
2456 if ((thflags & TH_FIN) &&
2457 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2461 TCPSTAT_INC(tcps_rcvdupack);
2463 * If we have outstanding data (other than
2464 * a window probe), this is a completely
2465 * duplicate ack (ie, window info didn't
2466 * change and FIN isn't set),
2467 * the ack is the biggest we've
2468 * seen and we've seen exactly our rexmt
2469 * threshold of them, assume a packet
2470 * has been dropped and retransmit it.
2471 * Kludge snd_nxt & the congestion
2472 * window so we send only this one
2475 * We know we're losing at the current
2476 * window size so do congestion avoidance
2477 * (set ssthresh to half the current window
2478 * and pull our congestion window back to
2479 * the new ssthresh).
2481 * Dup acks mean that packets have left the
2482 * network (they're now cached at the receiver)
2483 * so bump cwnd by the amount in the receiver
2484 * to keep a constant cwnd packets in the
2487 * When using TCP ECN, notify the peer that
2488 * we reduced the cwnd.
2491 * Following 2 kinds of acks should not affect
2494 * 2) Acks with SACK but without any new SACK
2495 * information in them. These could result from
2496 * any anomaly in the network like a switch
2497 * duplicating packets or a possible DoS attack.
2499 if (th->th_ack != tp->snd_una ||
2500 ((tp->t_flags & TF_SACK_PERMIT) &&
2503 else if (!tcp_timer_active(tp, TT_REXMT))
2505 else if (++tp->t_dupacks > tcprexmtthresh ||
2506 IN_FASTRECOVERY(tp->t_flags)) {
2507 cc_ack_received(tp, th, nsegs,
2509 if ((tp->t_flags & TF_SACK_PERMIT) &&
2510 IN_FASTRECOVERY(tp->t_flags)) {
2514 * Compute the amount of data in flight first.
2515 * We can inject new data into the pipe iff
2516 * we have less than 1/2 the original window's
2517 * worth of data in flight.
2519 if (V_tcp_do_rfc6675_pipe)
2520 awnd = tcp_compute_pipe(tp);
2522 awnd = (tp->snd_nxt - tp->snd_fack) +
2523 tp->sackhint.sack_bytes_rexmit;
2525 if (awnd < tp->snd_ssthresh) {
2526 tp->snd_cwnd += maxseg;
2527 if (tp->snd_cwnd > tp->snd_ssthresh)
2528 tp->snd_cwnd = tp->snd_ssthresh;
2531 tp->snd_cwnd += maxseg;
2532 (void) tp->t_fb->tfb_tcp_output(tp);
2534 } else if (tp->t_dupacks == tcprexmtthresh) {
2535 tcp_seq onxt = tp->snd_nxt;
2538 * If we're doing sack, check to
2539 * see if we're already in sack
2540 * recovery. If we're not doing sack,
2541 * check to see if we're in newreno
2544 if (tp->t_flags & TF_SACK_PERMIT) {
2545 if (IN_FASTRECOVERY(tp->t_flags)) {
2550 if (SEQ_LEQ(th->th_ack,
2556 /* Congestion signal before ack. */
2557 cc_cong_signal(tp, th, CC_NDUPACK);
2558 cc_ack_received(tp, th, nsegs,
2560 tcp_timer_activate(tp, TT_REXMT, 0);
2562 if (tp->t_flags & TF_SACK_PERMIT) {
2564 tcps_sack_recovery_episode);
2565 tp->sack_newdata = tp->snd_nxt;
2566 tp->snd_cwnd = maxseg;
2567 (void) tp->t_fb->tfb_tcp_output(tp);
2570 tp->snd_nxt = th->th_ack;
2571 tp->snd_cwnd = maxseg;
2572 (void) tp->t_fb->tfb_tcp_output(tp);
2573 KASSERT(tp->snd_limited <= 2,
2574 ("%s: tp->snd_limited too big",
2576 tp->snd_cwnd = tp->snd_ssthresh +
2578 (tp->t_dupacks - tp->snd_limited);
2579 if (SEQ_GT(onxt, tp->snd_nxt))
2582 } else if (V_tcp_do_rfc3042) {
2584 * Process first and second duplicate
2585 * ACKs. Each indicates a segment
2586 * leaving the network, creating room
2587 * for more. Make sure we can send a
2588 * packet on reception of each duplicate
2589 * ACK by increasing snd_cwnd by one
2590 * segment. Restore the original
2591 * snd_cwnd after packet transmission.
2593 cc_ack_received(tp, th, nsegs,
2595 uint32_t oldcwnd = tp->snd_cwnd;
2596 tcp_seq oldsndmax = tp->snd_max;
2600 KASSERT(tp->t_dupacks == 1 ||
2602 ("%s: dupacks not 1 or 2",
2604 if (tp->t_dupacks == 1)
2605 tp->snd_limited = 0;
2607 (tp->snd_nxt - tp->snd_una) +
2608 (tp->t_dupacks - tp->snd_limited) *
2611 * Only call tcp_output when there
2612 * is new data available to be sent.
2613 * Otherwise we would send pure ACKs.
2615 SOCKBUF_LOCK(&so->so_snd);
2616 avail = sbavail(&so->so_snd) -
2617 (tp->snd_nxt - tp->snd_una);
2618 SOCKBUF_UNLOCK(&so->so_snd);
2620 (void) tp->t_fb->tfb_tcp_output(tp);
2621 sent = tp->snd_max - oldsndmax;
2622 if (sent > maxseg) {
2623 KASSERT((tp->t_dupacks == 2 &&
2624 tp->snd_limited == 0) ||
2625 (sent == maxseg + 1 &&
2626 tp->t_flags & TF_SENTFIN),
2627 ("%s: sent too much",
2629 tp->snd_limited = 2;
2630 } else if (sent > 0)
2632 tp->snd_cwnd = oldcwnd;
2639 * This ack is advancing the left edge, reset the
2644 * If this ack also has new SACK info, increment the
2645 * counter as per rfc6675.
2647 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed)
2651 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2652 ("%s: th_ack <= snd_una", __func__));
2655 * If the congestion window was inflated to account
2656 * for the other side's cached packets, retract it.
2658 if (IN_FASTRECOVERY(tp->t_flags)) {
2659 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2660 if (tp->t_flags & TF_SACK_PERMIT)
2661 tcp_sack_partialack(tp, th);
2663 tcp_newreno_partial_ack(tp, th);
2665 cc_post_recovery(tp, th);
2668 * If we reach this point, ACK is not a duplicate,
2669 * i.e., it ACKs something we sent.
2671 if (tp->t_flags & TF_NEEDSYN) {
2673 * T/TCP: Connection was half-synchronized, and our
2674 * SYN has been ACK'd (so connection is now fully
2675 * synchronized). Go to non-starred state,
2676 * increment snd_una for ACK of SYN, and check if
2677 * we can do window scaling.
2679 tp->t_flags &= ~TF_NEEDSYN;
2681 /* Do window scaling? */
2682 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2683 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2684 tp->rcv_scale = tp->request_r_scale;
2685 /* Send window already scaled. */
2690 INP_WLOCK_ASSERT(tp->t_inpcb);
2692 acked = BYTES_THIS_ACK(tp, th);
2693 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2694 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2695 tp->snd_una, th->th_ack, tp, m));
2696 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2697 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2700 * If we just performed our first retransmit, and the ACK
2701 * arrives within our recovery window, then it was a mistake
2702 * to do the retransmit in the first place. Recover our
2703 * original cwnd and ssthresh, and proceed to transmit where
2706 if (tp->t_rxtshift == 1 &&
2707 tp->t_flags & TF_PREVVALID &&
2709 SEQ_LT(to.to_tsecr, tp->t_badrxtwin))
2710 cc_cong_signal(tp, th, CC_RTO_ERR);
2713 * If we have a timestamp reply, update smoothed
2714 * round trip time. If no timestamp is present but
2715 * transmit timer is running and timed sequence
2716 * number was acked, update smoothed round trip time.
2717 * Since we now have an rtt measurement, cancel the
2718 * timer backoff (cf., Phil Karn's retransmit alg.).
2719 * Recompute the initial retransmit timer.
2721 * Some boxes send broken timestamp replies
2722 * during the SYN+ACK phase, ignore
2723 * timestamps of 0 or we could calculate a
2724 * huge RTT and blow up the retransmit timer.
2726 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2729 t = tcp_ts_getticks() - to.to_tsecr;
2730 if (!tp->t_rttlow || tp->t_rttlow > t)
2732 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2733 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2734 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2735 tp->t_rttlow = ticks - tp->t_rtttime;
2736 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2740 * If all outstanding data is acked, stop retransmit
2741 * timer and remember to restart (more output or persist).
2742 * If there is more data to be acked, restart retransmit
2743 * timer, using current (possibly backed-off) value.
2745 if (th->th_ack == tp->snd_max) {
2746 tcp_timer_activate(tp, TT_REXMT, 0);
2748 } else if (!tcp_timer_active(tp, TT_PERSIST))
2749 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2752 * If no data (only SYN) was ACK'd,
2753 * skip rest of ACK processing.
2759 * Let the congestion control algorithm update congestion
2760 * control related information. This typically means increasing
2761 * the congestion window.
2763 cc_ack_received(tp, th, nsegs, CC_ACK);
2765 SOCKBUF_LOCK(&so->so_snd);
2766 if (acked > sbavail(&so->so_snd)) {
2767 if (tp->snd_wnd >= sbavail(&so->so_snd))
2768 tp->snd_wnd -= sbavail(&so->so_snd);
2771 mfree = sbcut_locked(&so->so_snd,
2772 (int)sbavail(&so->so_snd));
2775 mfree = sbcut_locked(&so->so_snd, acked);
2776 if (tp->snd_wnd >= (uint32_t) acked)
2777 tp->snd_wnd -= acked;
2782 /* NB: sowwakeup_locked() does an implicit unlock. */
2783 sowwakeup_locked(so);
2785 /* Detect una wraparound. */
2786 if (!IN_RECOVERY(tp->t_flags) &&
2787 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2788 SEQ_LEQ(th->th_ack, tp->snd_recover))
2789 tp->snd_recover = th->th_ack - 1;
2790 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2791 if (IN_RECOVERY(tp->t_flags) &&
2792 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2793 EXIT_RECOVERY(tp->t_flags);
2795 tp->snd_una = th->th_ack;
2796 if (tp->t_flags & TF_SACK_PERMIT) {
2797 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2798 tp->snd_recover = tp->snd_una;
2800 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2801 tp->snd_nxt = tp->snd_una;
2803 switch (tp->t_state) {
2806 * In FIN_WAIT_1 STATE in addition to the processing
2807 * for the ESTABLISHED state if our FIN is now acknowledged
2808 * then enter FIN_WAIT_2.
2810 case TCPS_FIN_WAIT_1:
2811 if (ourfinisacked) {
2813 * If we can't receive any more
2814 * data, then closing user can proceed.
2815 * Starting the timer is contrary to the
2816 * specification, but if we don't get a FIN
2817 * we'll hang forever.
2820 * we should release the tp also, and use a
2823 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2824 soisdisconnected(so);
2825 tcp_timer_activate(tp, TT_2MSL,
2826 (tcp_fast_finwait2_recycle ?
2827 tcp_finwait2_timeout :
2830 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2835 * In CLOSING STATE in addition to the processing for
2836 * the ESTABLISHED state if the ACK acknowledges our FIN
2837 * then enter the TIME-WAIT state, otherwise ignore
2841 if (ourfinisacked) {
2849 * In LAST_ACK, we may still be waiting for data to drain
2850 * and/or to be acked, as well as for the ack of our FIN.
2851 * If our FIN is now acknowledged, delete the TCB,
2852 * enter the closed state and return.
2855 if (ourfinisacked) {
2864 INP_WLOCK_ASSERT(tp->t_inpcb);
2867 * Update window information.
2868 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2870 if ((thflags & TH_ACK) &&
2871 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2872 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2873 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2874 /* keep track of pure window updates */
2876 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2877 TCPSTAT_INC(tcps_rcvwinupd);
2878 tp->snd_wnd = tiwin;
2879 tp->snd_wl1 = th->th_seq;
2880 tp->snd_wl2 = th->th_ack;
2881 if (tp->snd_wnd > tp->max_sndwnd)
2882 tp->max_sndwnd = tp->snd_wnd;
2887 * Process segments with URG.
2889 if ((thflags & TH_URG) && th->th_urp &&
2890 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2892 * This is a kludge, but if we receive and accept
2893 * random urgent pointers, we'll crash in
2894 * soreceive. It's hard to imagine someone
2895 * actually wanting to send this much urgent data.
2897 SOCKBUF_LOCK(&so->so_rcv);
2898 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2899 th->th_urp = 0; /* XXX */
2900 thflags &= ~TH_URG; /* XXX */
2901 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2902 goto dodata; /* XXX */
2905 * If this segment advances the known urgent pointer,
2906 * then mark the data stream. This should not happen
2907 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2908 * a FIN has been received from the remote side.
2909 * In these states we ignore the URG.
2911 * According to RFC961 (Assigned Protocols),
2912 * the urgent pointer points to the last octet
2913 * of urgent data. We continue, however,
2914 * to consider it to indicate the first octet
2915 * of data past the urgent section as the original
2916 * spec states (in one of two places).
2918 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2919 tp->rcv_up = th->th_seq + th->th_urp;
2920 so->so_oobmark = sbavail(&so->so_rcv) +
2921 (tp->rcv_up - tp->rcv_nxt) - 1;
2922 if (so->so_oobmark == 0)
2923 so->so_rcv.sb_state |= SBS_RCVATMARK;
2925 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2927 SOCKBUF_UNLOCK(&so->so_rcv);
2929 * Remove out of band data so doesn't get presented to user.
2930 * This can happen independent of advancing the URG pointer,
2931 * but if two URG's are pending at once, some out-of-band
2932 * data may creep in... ick.
2934 if (th->th_urp <= (uint32_t)tlen &&
2935 !(so->so_options & SO_OOBINLINE)) {
2936 /* hdr drop is delayed */
2937 tcp_pulloutofband(so, th, m, drop_hdrlen);
2941 * If no out of band data is expected,
2942 * pull receive urgent pointer along
2943 * with the receive window.
2945 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2946 tp->rcv_up = tp->rcv_nxt;
2949 INP_WLOCK_ASSERT(tp->t_inpcb);
2952 * Process the segment text, merging it into the TCP sequencing queue,
2953 * and arranging for acknowledgment of receipt if necessary.
2954 * This process logically involves adjusting tp->rcv_wnd as data
2955 * is presented to the user (this happens in tcp_usrreq.c,
2956 * case PRU_RCVD). If a FIN has already been received on this
2957 * connection then we just ignore the text.
2959 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
2960 IS_FASTOPEN(tp->t_flags));
2961 if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
2962 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2963 tcp_seq save_start = th->th_seq;
2964 tcp_seq save_rnxt = tp->rcv_nxt;
2965 int save_tlen = tlen;
2966 m_adj(m, drop_hdrlen); /* delayed header drop */
2968 * Insert segment which includes th into TCP reassembly queue
2969 * with control block tp. Set thflags to whether reassembly now
2970 * includes a segment with FIN. This handles the common case
2971 * inline (segment is the next to be received on an established
2972 * connection, and the queue is empty), avoiding linkage into
2973 * and removal from the queue and repetition of various
2975 * Set DELACK for segments received in order, but ack
2976 * immediately when segments are out of order (so
2977 * fast retransmit can work).
2979 if (th->th_seq == tp->rcv_nxt &&
2981 (TCPS_HAVEESTABLISHED(tp->t_state) ||
2983 if (DELAY_ACK(tp, tlen) || tfo_syn)
2984 tp->t_flags |= TF_DELACK;
2986 tp->t_flags |= TF_ACKNOW;
2987 tp->rcv_nxt += tlen;
2988 thflags = th->th_flags & TH_FIN;
2989 TCPSTAT_INC(tcps_rcvpack);
2990 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2991 SOCKBUF_LOCK(&so->so_rcv);
2992 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2995 sbappendstream_locked(&so->so_rcv, m, 0);
2996 /* NB: sorwakeup_locked() does an implicit unlock. */
2997 sorwakeup_locked(so);
3000 * XXX: Due to the header drop above "th" is
3001 * theoretically invalid by now. Fortunately
3002 * m_adj() doesn't actually frees any mbufs
3003 * when trimming from the head.
3005 tcp_seq temp = save_start;
3006 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3007 tp->t_flags |= TF_ACKNOW;
3009 if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) {
3010 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3012 * DSACK actually handled in the fastpath
3015 tcp_update_sack_list(tp, save_start,
3016 save_start + save_tlen);
3017 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3018 if ((tp->rcv_numsacks >= 1) &&
3019 (tp->sackblks[0].end == save_start)) {
3021 * Partial overlap, recorded at todrop
3024 tcp_update_sack_list(tp,
3025 tp->sackblks[0].start,
3026 tp->sackblks[0].end);
3028 tcp_update_dsack_list(tp, save_start,
3029 save_start + save_tlen);
3031 } else if (tlen >= save_tlen) {
3032 /* Update of sackblks. */
3033 tcp_update_dsack_list(tp, save_start,
3034 save_start + save_tlen);
3035 } else if (tlen > 0) {
3036 tcp_update_dsack_list(tp, save_start,
3042 * Note the amount of data that peer has sent into
3043 * our window, in order to estimate the sender's
3047 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3048 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3050 len = so->so_rcv.sb_hiwat;
3058 * If FIN is received ACK the FIN and let the user know
3059 * that the connection is closing.
3061 if (thflags & TH_FIN) {
3062 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3065 * If connection is half-synchronized
3066 * (ie NEEDSYN flag on) then delay ACK,
3067 * so it may be piggybacked when SYN is sent.
3068 * Otherwise, since we received a FIN then no
3069 * more input can be expected, send ACK now.
3071 if (tp->t_flags & TF_NEEDSYN)
3072 tp->t_flags |= TF_DELACK;
3074 tp->t_flags |= TF_ACKNOW;
3077 switch (tp->t_state) {
3080 * In SYN_RECEIVED and ESTABLISHED STATES
3081 * enter the CLOSE_WAIT state.
3083 case TCPS_SYN_RECEIVED:
3084 tp->t_starttime = ticks;
3086 case TCPS_ESTABLISHED:
3087 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3091 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3092 * enter the CLOSING state.
3094 case TCPS_FIN_WAIT_1:
3095 tcp_state_change(tp, TCPS_CLOSING);
3099 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3100 * starting the time-wait timer, turning off the other
3103 case TCPS_FIN_WAIT_2:
3109 if (so->so_options & SO_DEBUG)
3110 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3113 TCP_PROBE3(debug__input, tp, th, m);
3116 * Return any desired output.
3118 if (needoutput || (tp->t_flags & TF_ACKNOW))
3119 (void) tp->t_fb->tfb_tcp_output(tp);
3122 INP_WLOCK_ASSERT(tp->t_inpcb);
3124 if (tp->t_flags & TF_DELACK) {
3125 tp->t_flags &= ~TF_DELACK;
3126 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3128 INP_WUNLOCK(tp->t_inpcb);
3133 * Generate an ACK dropping incoming segment if it occupies
3134 * sequence space, where the ACK reflects our state.
3136 * We can now skip the test for the RST flag since all
3137 * paths to this code happen after packets containing
3138 * RST have been dropped.
3140 * In the SYN-RECEIVED state, don't send an ACK unless the
3141 * segment we received passes the SYN-RECEIVED ACK test.
3142 * If it fails send a RST. This breaks the loop in the
3143 * "LAND" DoS attack, and also prevents an ACK storm
3144 * between two listening ports that have been sent forged
3145 * SYN segments, each with the source address of the other.
3147 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3148 (SEQ_GT(tp->snd_una, th->th_ack) ||
3149 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3150 rstreason = BANDLIM_RST_OPENPORT;
3154 if (so->so_options & SO_DEBUG)
3155 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3158 TCP_PROBE3(debug__input, tp, th, m);
3159 tp->t_flags |= TF_ACKNOW;
3160 (void) tp->t_fb->tfb_tcp_output(tp);
3161 INP_WUNLOCK(tp->t_inpcb);
3167 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3168 INP_WUNLOCK(tp->t_inpcb);
3170 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3175 * Drop space held by incoming segment and return.
3178 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3179 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3182 TCP_PROBE3(debug__input, tp, th, m);
3184 INP_WUNLOCK(tp->t_inpcb);
3189 * Issue RST and make ACK acceptable to originator of segment.
3190 * The mbuf must still include the original packet header.
3194 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3195 int tlen, int rstreason)
3201 struct ip6_hdr *ip6;
3205 INP_WLOCK_ASSERT(tp->t_inpcb);
3208 /* Don't bother if destination was broadcast/multicast. */
3209 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3212 if (mtod(m, struct ip *)->ip_v == 6) {
3213 ip6 = mtod(m, struct ip6_hdr *);
3214 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3215 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3217 /* IPv6 anycast check is done at tcp6_input() */
3220 #if defined(INET) && defined(INET6)
3225 ip = mtod(m, struct ip *);
3226 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3227 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3228 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3229 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3234 /* Perform bandwidth limiting. */
3235 if (badport_bandlim(rstreason) < 0)
3238 /* tcp_respond consumes the mbuf chain. */
3239 if (th->th_flags & TH_ACK) {
3240 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3241 th->th_ack, TH_RST);
3243 if (th->th_flags & TH_SYN)
3245 if (th->th_flags & TH_FIN)
3247 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3248 (tcp_seq)0, TH_RST|TH_ACK);
3256 * Parse TCP options and place in tcpopt.
3259 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3264 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3266 if (opt == TCPOPT_EOL)
3268 if (opt == TCPOPT_NOP)
3274 if (optlen < 2 || optlen > cnt)
3279 if (optlen != TCPOLEN_MAXSEG)
3281 if (!(flags & TO_SYN))
3283 to->to_flags |= TOF_MSS;
3284 bcopy((char *)cp + 2,
3285 (char *)&to->to_mss, sizeof(to->to_mss));
3286 to->to_mss = ntohs(to->to_mss);
3289 if (optlen != TCPOLEN_WINDOW)
3291 if (!(flags & TO_SYN))
3293 to->to_flags |= TOF_SCALE;
3294 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3296 case TCPOPT_TIMESTAMP:
3297 if (optlen != TCPOLEN_TIMESTAMP)
3299 to->to_flags |= TOF_TS;
3300 bcopy((char *)cp + 2,
3301 (char *)&to->to_tsval, sizeof(to->to_tsval));
3302 to->to_tsval = ntohl(to->to_tsval);
3303 bcopy((char *)cp + 6,
3304 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3305 to->to_tsecr = ntohl(to->to_tsecr);
3307 case TCPOPT_SIGNATURE:
3309 * In order to reply to a host which has set the
3310 * TCP_SIGNATURE option in its initial SYN, we have
3311 * to record the fact that the option was observed
3312 * here for the syncache code to perform the correct
3315 if (optlen != TCPOLEN_SIGNATURE)
3317 to->to_flags |= TOF_SIGNATURE;
3318 to->to_signature = cp + 2;
3320 case TCPOPT_SACK_PERMITTED:
3321 if (optlen != TCPOLEN_SACK_PERMITTED)
3323 if (!(flags & TO_SYN))
3327 to->to_flags |= TOF_SACKPERM;
3330 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3334 to->to_flags |= TOF_SACK;
3335 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3336 to->to_sacks = cp + 2;
3337 TCPSTAT_INC(tcps_sack_rcv_blocks);
3339 case TCPOPT_FAST_OPEN:
3341 * Cookie length validation is performed by the
3342 * server side cookie checking code or the client
3343 * side cookie cache update code.
3345 if (!(flags & TO_SYN))
3347 if (!V_tcp_fastopen_client_enable &&
3348 !V_tcp_fastopen_server_enable)
3350 to->to_flags |= TOF_FASTOPEN;
3351 to->to_tfo_len = optlen - 2;
3352 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3361 * Pull out of band byte out of a segment so
3362 * it doesn't appear in the user's data queue.
3363 * It is still reflected in the segment length for
3364 * sequencing purposes.
3367 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3370 int cnt = off + th->th_urp - 1;
3373 if (m->m_len > cnt) {
3374 char *cp = mtod(m, caddr_t) + cnt;
3375 struct tcpcb *tp = sototcpcb(so);
3377 INP_WLOCK_ASSERT(tp->t_inpcb);
3380 tp->t_oobflags |= TCPOOB_HAVEDATA;
3381 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3383 if (m->m_flags & M_PKTHDR)
3392 panic("tcp_pulloutofband");
3396 * Collect new round-trip time estimate
3397 * and update averages and current timeout.
3400 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3404 INP_WLOCK_ASSERT(tp->t_inpcb);
3406 TCPSTAT_INC(tcps_rttupdated);
3409 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3410 imax(0, rtt * 1000 / hz));
3412 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3414 * srtt is stored as fixed point with 5 bits after the
3415 * binary point (i.e., scaled by 8). The following magic
3416 * is equivalent to the smoothing algorithm in rfc793 with
3417 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3418 * point). Adjust rtt to origin 0.
3420 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3421 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3423 if ((tp->t_srtt += delta) <= 0)
3427 * We accumulate a smoothed rtt variance (actually, a
3428 * smoothed mean difference), then set the retransmit
3429 * timer to smoothed rtt + 4 times the smoothed variance.
3430 * rttvar is stored as fixed point with 4 bits after the
3431 * binary point (scaled by 16). The following is
3432 * equivalent to rfc793 smoothing with an alpha of .75
3433 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3434 * rfc793's wired-in beta.
3438 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3439 if ((tp->t_rttvar += delta) <= 0)
3441 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3442 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3445 * No rtt measurement yet - use the unsmoothed rtt.
3446 * Set the variance to half the rtt (so our first
3447 * retransmit happens at 3*rtt).
3449 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3450 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3451 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3457 * the retransmit should happen at rtt + 4 * rttvar.
3458 * Because of the way we do the smoothing, srtt and rttvar
3459 * will each average +1/2 tick of bias. When we compute
3460 * the retransmit timer, we want 1/2 tick of rounding and
3461 * 1 extra tick because of +-1/2 tick uncertainty in the
3462 * firing of the timer. The bias will give us exactly the
3463 * 1.5 tick we need. But, because the bias is
3464 * statistical, we have to test that we don't drop below
3465 * the minimum feasible timer (which is 2 ticks).
3467 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3468 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3471 * We received an ack for a packet that wasn't retransmitted;
3472 * it is probably safe to discard any error indications we've
3473 * received recently. This isn't quite right, but close enough
3474 * for now (a route might have failed after we sent a segment,
3475 * and the return path might not be symmetrical).
3477 tp->t_softerror = 0;
3481 * Determine a reasonable value for maxseg size.
3482 * If the route is known, check route for mtu.
3483 * If none, use an mss that can be handled on the outgoing interface
3484 * without forcing IP to fragment. If no route is found, route has no mtu,
3485 * or the destination isn't local, use a default, hopefully conservative
3486 * size (usually 512 or the default IP max size, but no more than the mtu
3487 * of the interface), as we can't discover anything about intervening
3488 * gateways or networks. We also initialize the congestion/slow start
3489 * window to be a single segment if the destination isn't local.
3490 * While looking at the routing entry, we also initialize other path-dependent
3491 * parameters from pre-set or cached values in the routing entry.
3493 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3494 * IP options, e.g. IPSEC data, since length of this data may vary, and
3495 * thus it is calculated for every segment separately in tcp_output().
3497 * NOTE that this routine is only called when we process an incoming
3498 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3499 * settings are handled in tcp_mssopt().
3502 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3503 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3506 uint32_t maxmtu = 0;
3507 struct inpcb *inp = tp->t_inpcb;
3508 struct hc_metrics_lite metrics;
3510 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3511 size_t min_protoh = isipv6 ?
3512 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3513 sizeof (struct tcpiphdr);
3515 const size_t min_protoh = sizeof(struct tcpiphdr);
3518 INP_WLOCK_ASSERT(tp->t_inpcb);
3520 if (mtuoffer != -1) {
3521 KASSERT(offer == -1, ("%s: conflict", __func__));
3522 offer = mtuoffer - min_protoh;
3528 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3529 tp->t_maxseg = V_tcp_v6mssdflt;
3532 #if defined(INET) && defined(INET6)
3537 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3538 tp->t_maxseg = V_tcp_mssdflt;
3543 * No route to sender, stay with default mss and return.
3547 * In case we return early we need to initialize metrics
3548 * to a defined state as tcp_hc_get() would do for us
3549 * if there was no cache hit.
3551 if (metricptr != NULL)
3552 bzero(metricptr, sizeof(struct hc_metrics_lite));
3556 /* What have we got? */
3560 * Offer == 0 means that there was no MSS on the SYN
3561 * segment, in this case we use tcp_mssdflt as
3562 * already assigned to t_maxseg above.
3564 offer = tp->t_maxseg;
3569 * Offer == -1 means that we didn't receive SYN yet.
3575 * Prevent DoS attack with too small MSS. Round up
3576 * to at least minmss.
3578 offer = max(offer, V_tcp_minmss);
3582 * rmx information is now retrieved from tcp_hostcache.
3584 tcp_hc_get(&inp->inp_inc, &metrics);
3585 if (metricptr != NULL)
3586 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3589 * If there's a discovered mtu in tcp hostcache, use it.
3590 * Else, use the link mtu.
3592 if (metrics.rmx_mtu)
3593 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3597 mss = maxmtu - min_protoh;
3598 if (!V_path_mtu_discovery &&
3599 !in6_localaddr(&inp->in6p_faddr))
3600 mss = min(mss, V_tcp_v6mssdflt);
3603 #if defined(INET) && defined(INET6)
3608 mss = maxmtu - min_protoh;
3609 if (!V_path_mtu_discovery &&
3610 !in_localaddr(inp->inp_faddr))
3611 mss = min(mss, V_tcp_mssdflt);
3615 * XXX - The above conditional (mss = maxmtu - min_protoh)
3616 * probably violates the TCP spec.
3617 * The problem is that, since we don't know the
3618 * other end's MSS, we are supposed to use a conservative
3619 * default. But, if we do that, then MTU discovery will
3620 * never actually take place, because the conservative
3621 * default is much less than the MTUs typically seen
3622 * on the Internet today. For the moment, we'll sweep
3623 * this under the carpet.
3625 * The conservative default might not actually be a problem
3626 * if the only case this occurs is when sending an initial
3627 * SYN with options and data to a host we've never talked
3628 * to before. Then, they will reply with an MSS value which
3629 * will get recorded and the new parameters should get
3630 * recomputed. For Further Study.
3633 mss = min(mss, offer);
3636 * Sanity check: make sure that maxseg will be large
3637 * enough to allow some data on segments even if the
3638 * all the option space is used (40bytes). Otherwise
3639 * funny things may happen in tcp_output.
3641 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3649 tcp_mss(struct tcpcb *tp, int offer)
3655 struct hc_metrics_lite metrics;
3656 struct tcp_ifcap cap;
3658 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3660 bzero(&cap, sizeof(cap));
3661 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3667 * If there's a pipesize, change the socket buffer to that size,
3668 * don't change if sb_hiwat is different than default (then it
3669 * has been changed on purpose with setsockopt).
3670 * Make the socket buffers an integral number of mss units;
3671 * if the mss is larger than the socket buffer, decrease the mss.
3673 so = inp->inp_socket;
3674 SOCKBUF_LOCK(&so->so_snd);
3675 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3676 bufsize = metrics.rmx_sendpipe;
3678 bufsize = so->so_snd.sb_hiwat;
3682 bufsize = roundup(bufsize, mss);
3683 if (bufsize > sb_max)
3685 if (bufsize > so->so_snd.sb_hiwat)
3686 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3688 SOCKBUF_UNLOCK(&so->so_snd);
3690 * Sanity check: make sure that maxseg will be large
3691 * enough to allow some data on segments even if the
3692 * all the option space is used (40bytes). Otherwise
3693 * funny things may happen in tcp_output.
3695 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3697 tp->t_maxseg = max(mss, 64);
3699 SOCKBUF_LOCK(&so->so_rcv);
3700 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3701 bufsize = metrics.rmx_recvpipe;
3703 bufsize = so->so_rcv.sb_hiwat;
3704 if (bufsize > mss) {
3705 bufsize = roundup(bufsize, mss);
3706 if (bufsize > sb_max)
3708 if (bufsize > so->so_rcv.sb_hiwat)
3709 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3711 SOCKBUF_UNLOCK(&so->so_rcv);
3713 /* Check the interface for TSO capabilities. */
3714 if (cap.ifcap & CSUM_TSO) {
3715 tp->t_flags |= TF_TSO;
3716 tp->t_tsomax = cap.tsomax;
3717 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3718 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3723 * Determine the MSS option to send on an outgoing SYN.
3726 tcp_mssopt(struct in_conninfo *inc)
3729 uint32_t thcmtu = 0;
3730 uint32_t maxmtu = 0;
3733 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3736 if (inc->inc_flags & INC_ISIPV6) {
3737 mss = V_tcp_v6mssdflt;
3738 maxmtu = tcp_maxmtu6(inc, NULL);
3739 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3742 #if defined(INET) && defined(INET6)
3747 mss = V_tcp_mssdflt;
3748 maxmtu = tcp_maxmtu(inc, NULL);
3749 min_protoh = sizeof(struct tcpiphdr);
3752 #if defined(INET6) || defined(INET)
3753 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3756 if (maxmtu && thcmtu)
3757 mss = min(maxmtu, thcmtu) - min_protoh;
3758 else if (maxmtu || thcmtu)
3759 mss = max(maxmtu, thcmtu) - min_protoh;
3766 * On a partial ack arrives, force the retransmission of the
3767 * next unacknowledged segment. Do not clear tp->t_dupacks.
3768 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3772 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3774 tcp_seq onxt = tp->snd_nxt;
3775 uint32_t ocwnd = tp->snd_cwnd;
3776 u_int maxseg = tcp_maxseg(tp);
3778 INP_WLOCK_ASSERT(tp->t_inpcb);
3780 tcp_timer_activate(tp, TT_REXMT, 0);
3782 tp->snd_nxt = th->th_ack;
3784 * Set snd_cwnd to one segment beyond acknowledged offset.
3785 * (tp->snd_una has not yet been updated when this function is called.)
3787 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3788 tp->t_flags |= TF_ACKNOW;
3789 (void) tp->t_fb->tfb_tcp_output(tp);
3790 tp->snd_cwnd = ocwnd;
3791 if (SEQ_GT(onxt, tp->snd_nxt))
3794 * Partial window deflation. Relies on fact that tp->snd_una
3797 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3798 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3801 tp->snd_cwnd += maxseg;
3805 tcp_compute_pipe(struct tcpcb *tp)
3807 return (tp->snd_max - tp->snd_una +
3808 tp->sackhint.sack_bytes_rexmit -
3809 tp->sackhint.sacked_bytes);
3813 tcp_compute_initwnd(uint32_t maxseg)
3816 * Calculate the Initial Window, also used as Restart Window
3818 * RFC5681 Section 3.1 specifies the default conservative values.
3819 * RFC3390 specifies slightly more aggressive values.
3820 * RFC6928 increases it to ten segments.
3821 * Support for user specified value for initial flight size.
3823 if (V_tcp_initcwnd_segments)
3824 return min(V_tcp_initcwnd_segments * maxseg,
3825 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
3826 else if (V_tcp_do_rfc3390)
3827 return min(4 * maxseg, max(2 * maxseg, 4380));
3829 /* Per RFC5681 Section 3.1 */
3831 return (2 * maxseg);
3832 else if (maxseg > 1095)
3833 return (3 * maxseg);
3835 return (4 * maxseg);