2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (c) 2007-2008,2010
5 * Swinburne University of Technology, Melbourne, Australia.
6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
7 * Copyright (c) 2010 The FreeBSD Foundation
8 * Copyright (c) 2010-2011 Juniper Networks, Inc.
11 * Portions of this software were developed at the Centre for Advanced Internet
12 * Architectures, Swinburne University of Technology, by Lawrence Stewart,
13 * James Healy and David Hayes, made possible in part by a grant from the Cisco
14 * University Research Program Fund at Community Foundation Silicon Valley.
16 * Portions of this software were developed at the Centre for Advanced
17 * Internet Architectures, Swinburne University of Technology, Melbourne,
18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
20 * Portions of this software were developed by Robert N. M. Watson under
21 * contract to Juniper Networks, Inc.
23 * Redistribution and use in source and binary forms, with or without
24 * modification, are permitted provided that the following conditions
26 * 1. Redistributions of source code must retain the above copyright
27 * notice, this list of conditions and the following disclaimer.
28 * 2. Redistributions in binary form must reproduce the above copyright
29 * notice, this list of conditions and the following disclaimer in the
30 * documentation and/or other materials provided with the distribution.
31 * 4. Neither the name of the University nor the names of its contributors
32 * may be used to endorse or promote products derived from this software
33 * without specific prior written permission.
35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD$");
54 #include "opt_inet6.h"
55 #include "opt_ipsec.h"
56 #include "opt_tcpdebug.h"
58 #include <sys/param.h>
59 #include <sys/kernel.h>
61 #include <sys/hhook.h>
63 #include <sys/malloc.h>
65 #include <sys/proc.h> /* for proc0 declaration */
66 #include <sys/protosw.h>
68 #include <sys/signalvar.h>
69 #include <sys/socket.h>
70 #include <sys/socketvar.h>
71 #include <sys/sysctl.h>
72 #include <sys/syslog.h>
73 #include <sys/systm.h>
75 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
80 #include <net/if_var.h>
81 #include <net/route.h>
84 #define TCPSTATES /* for logging */
86 #include <netinet/in.h>
87 #include <netinet/in_kdtrace.h>
88 #include <netinet/in_pcb.h>
89 #include <netinet/in_systm.h>
90 #include <netinet/ip.h>
91 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
92 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
93 #include <netinet/ip_var.h>
94 #include <netinet/ip_options.h>
95 #include <netinet/ip6.h>
96 #include <netinet/icmp6.h>
97 #include <netinet6/in6_pcb.h>
98 #include <netinet6/in6_var.h>
99 #include <netinet6/ip6_var.h>
100 #include <netinet6/nd6.h>
102 #include <netinet/tcp_fastopen.h>
104 #include <netinet/tcp.h>
105 #include <netinet/tcp_fsm.h>
106 #include <netinet/tcp_seq.h>
107 #include <netinet/tcp_timer.h>
108 #include <netinet/tcp_var.h>
109 #include <netinet6/tcp6_var.h>
110 #include <netinet/tcpip.h>
111 #include <netinet/cc/cc.h>
113 #include <netinet/tcp_pcap.h>
115 #include <netinet/tcp_syncache.h>
117 #include <netinet/tcp_debug.h>
118 #endif /* TCPDEBUG */
120 #include <netinet/tcp_offload.h>
124 #include <netipsec/ipsec.h>
125 #include <netipsec/ipsec6.h>
128 #include <machine/in_cksum.h>
130 #include <security/mac/mac_framework.h>
132 const int tcprexmtthresh = 3;
134 int tcp_log_in_vain = 0;
135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
137 "Log all incoming TCP segments to closed ports");
139 VNET_DEFINE(int, blackhole) = 0;
140 #define V_blackhole VNET(blackhole)
141 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
142 &VNET_NAME(blackhole), 0,
143 "Do not send RST on segments to closed ports");
145 VNET_DEFINE(int, tcp_delack_enabled) = 1;
146 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
147 &VNET_NAME(tcp_delack_enabled), 0,
148 "Delay ACK to try and piggyback it onto a data packet");
150 VNET_DEFINE(int, drop_synfin) = 0;
151 #define V_drop_synfin VNET(drop_synfin)
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_rfc6675_pipe) = 0;
157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
158 &VNET_NAME(tcp_do_rfc6675_pipe), 0,
159 "Use calculated pipe/in-flight bytes per RFC 6675");
161 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
162 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
164 &VNET_NAME(tcp_do_rfc3042), 0,
165 "Enable RFC 3042 (Limited Transmit)");
167 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
169 &VNET_NAME(tcp_do_rfc3390), 0,
170 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
172 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
174 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
175 "Slow-start flight size (initial congestion window) in number of segments");
177 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
179 &VNET_NAME(tcp_do_rfc3465), 0,
180 "Enable RFC 3465 (Appropriate Byte Counting)");
182 VNET_DEFINE(int, tcp_abc_l_var) = 2;
183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
184 &VNET_NAME(tcp_abc_l_var), 2,
185 "Cap the max cwnd increment during slow-start to this number of segments");
187 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
189 VNET_DEFINE(int, tcp_do_ecn) = 2;
190 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
191 &VNET_NAME(tcp_do_ecn), 0,
194 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
195 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
196 &VNET_NAME(tcp_ecn_maxretries), 0,
197 "Max retries before giving up on ECN");
199 VNET_DEFINE(int, tcp_insecure_syn) = 0;
200 #define V_tcp_insecure_syn VNET(tcp_insecure_syn)
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
202 &VNET_NAME(tcp_insecure_syn), 0,
203 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
205 VNET_DEFINE(int, tcp_insecure_rst) = 0;
206 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
207 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
208 &VNET_NAME(tcp_insecure_rst), 0,
209 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
211 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
212 #define V_tcp_recvspace VNET(tcp_recvspace)
213 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
214 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
216 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
217 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
218 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
219 &VNET_NAME(tcp_do_autorcvbuf), 0,
220 "Enable automatic receive buffer sizing");
222 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
223 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW,
225 &VNET_NAME(tcp_autorcvbuf_inc), 0,
226 "Incrementor step size of automatic receive buffer");
228 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
229 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
230 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
231 &VNET_NAME(tcp_autorcvbuf_max), 0,
232 "Max size of automatic receive buffer");
234 VNET_DEFINE(struct inpcbhead, tcb);
235 #define tcb6 tcb /* for KAME src sync over BSD*'s */
236 VNET_DEFINE(struct inpcbinfo, tcbinfo);
239 * TCP statistics are stored in an array of counter(9)s, which size matches
240 * size of struct tcpstat. TCP running connection count is a regular array.
242 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
243 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
244 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
245 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
246 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
247 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
248 "TCP connection counts by TCP state");
251 tcp_vnet_init(const void *unused)
254 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
255 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
257 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
258 tcp_vnet_init, NULL);
262 tcp_vnet_uninit(const void *unused)
265 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
266 VNET_PCPUSTAT_FREE(tcpstat);
268 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
269 tcp_vnet_uninit, NULL);
273 * Kernel module interface for updating tcpstat. The argument is an index
274 * into tcpstat treated as an array.
277 kmod_tcpstat_inc(int statnum)
280 counter_u64_add(VNET(tcpstat)[statnum], 1);
285 * Wrapper for the TCP established input helper hook.
288 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
290 struct tcp_hhook_data hhook_data;
292 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
297 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
304 * CC wrapper hook functions
307 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
310 INP_WLOCK_ASSERT(tp->t_inpcb);
312 tp->ccv->nsegs = nsegs;
313 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
314 if (tp->snd_cwnd <= tp->snd_wnd)
315 tp->ccv->flags |= CCF_CWND_LIMITED;
317 tp->ccv->flags &= ~CCF_CWND_LIMITED;
319 if (type == CC_ACK) {
320 if (tp->snd_cwnd > tp->snd_ssthresh) {
321 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
322 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp));
323 if (tp->t_bytes_acked >= tp->snd_cwnd) {
324 tp->t_bytes_acked -= tp->snd_cwnd;
325 tp->ccv->flags |= CCF_ABC_SENTAWND;
328 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
329 tp->t_bytes_acked = 0;
333 if (CC_ALGO(tp)->ack_received != NULL) {
334 /* XXXLAS: Find a way to live without this */
335 tp->ccv->curack = th->th_ack;
336 CC_ALGO(tp)->ack_received(tp->ccv, type);
341 cc_conn_init(struct tcpcb *tp)
343 struct hc_metrics_lite metrics;
344 struct inpcb *inp = tp->t_inpcb;
348 INP_WLOCK_ASSERT(tp->t_inpcb);
350 tcp_hc_get(&inp->inp_inc, &metrics);
351 maxseg = tcp_maxseg(tp);
353 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
355 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
356 TCPSTAT_INC(tcps_usedrtt);
357 if (metrics.rmx_rttvar) {
358 tp->t_rttvar = metrics.rmx_rttvar;
359 TCPSTAT_INC(tcps_usedrttvar);
361 /* default variation is +- 1 rtt */
363 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
365 TCPT_RANGESET(tp->t_rxtcur,
366 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
367 tp->t_rttmin, TCPTV_REXMTMAX);
369 if (metrics.rmx_ssthresh) {
371 * There's some sort of gateway or interface
372 * buffer limit on the path. Use this to set
373 * the slow start threshold, but set the
374 * threshold to no less than 2*mss.
376 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
377 TCPSTAT_INC(tcps_usedssthresh);
381 * Set the initial slow-start flight size.
383 * RFC5681 Section 3.1 specifies the default conservative values.
384 * RFC3390 specifies slightly more aggressive values.
385 * RFC6928 increases it to ten segments.
386 * Support for user specified value for initial flight size.
388 * If a SYN or SYN/ACK was lost and retransmitted, we have to
389 * reduce the initial CWND to one segment as congestion is likely
390 * requiring us to be cautious.
392 if (tp->snd_cwnd == 1)
393 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
394 else if (V_tcp_initcwnd_segments)
395 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg,
396 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
397 else if (V_tcp_do_rfc3390)
398 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380));
400 /* Per RFC5681 Section 3.1 */
402 tp->snd_cwnd = 2 * maxseg;
403 else if (maxseg > 1095)
404 tp->snd_cwnd = 3 * maxseg;
406 tp->snd_cwnd = 4 * maxseg;
409 if (CC_ALGO(tp)->conn_init != NULL)
410 CC_ALGO(tp)->conn_init(tp->ccv);
414 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
418 INP_WLOCK_ASSERT(tp->t_inpcb);
422 if (!IN_FASTRECOVERY(tp->t_flags)) {
423 tp->snd_recover = tp->snd_max;
424 if (tp->t_flags & TF_ECN_PERMIT)
425 tp->t_flags |= TF_ECN_SND_CWR;
429 if (!IN_CONGRECOVERY(tp->t_flags)) {
430 TCPSTAT_INC(tcps_ecn_rcwnd);
431 tp->snd_recover = tp->snd_max;
432 if (tp->t_flags & TF_ECN_PERMIT)
433 tp->t_flags |= TF_ECN_SND_CWR;
437 maxseg = tcp_maxseg(tp);
439 tp->t_bytes_acked = 0;
440 EXIT_RECOVERY(tp->t_flags);
441 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
443 tp->snd_cwnd = maxseg;
446 TCPSTAT_INC(tcps_sndrexmitbad);
447 /* RTO was unnecessary, so reset everything. */
448 tp->snd_cwnd = tp->snd_cwnd_prev;
449 tp->snd_ssthresh = tp->snd_ssthresh_prev;
450 tp->snd_recover = tp->snd_recover_prev;
451 if (tp->t_flags & TF_WASFRECOVERY)
452 ENTER_FASTRECOVERY(tp->t_flags);
453 if (tp->t_flags & TF_WASCRECOVERY)
454 ENTER_CONGRECOVERY(tp->t_flags);
455 tp->snd_nxt = tp->snd_max;
456 tp->t_flags &= ~TF_PREVVALID;
461 if (CC_ALGO(tp)->cong_signal != NULL) {
463 tp->ccv->curack = th->th_ack;
464 CC_ALGO(tp)->cong_signal(tp->ccv, type);
469 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
471 INP_WLOCK_ASSERT(tp->t_inpcb);
473 /* XXXLAS: KASSERT that we're in recovery? */
475 if (CC_ALGO(tp)->post_recovery != NULL) {
476 tp->ccv->curack = th->th_ack;
477 CC_ALGO(tp)->post_recovery(tp->ccv);
479 /* XXXLAS: EXIT_RECOVERY ? */
480 tp->t_bytes_acked = 0;
485 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
486 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
490 tcp_fields_to_net(th);
491 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
492 tcp_fields_to_host(th);
498 * Indicate whether this ack should be delayed. We can delay the ack if
499 * following conditions are met:
500 * - There is no delayed ack timer in progress.
501 * - Our last ack wasn't a 0-sized window. We never want to delay
502 * the ack that opens up a 0-sized window.
503 * - LRO wasn't used for this segment. We make sure by checking that the
504 * segment size is not larger than the MSS.
506 #define DELAY_ACK(tp, tlen) \
507 ((!tcp_timer_active(tp, TT_DELACK) && \
508 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
509 (tlen <= tp->t_maxseg) && \
510 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
513 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
515 INP_WLOCK_ASSERT(tp->t_inpcb);
517 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
518 switch (iptos & IPTOS_ECN_MASK) {
520 tp->ccv->flags |= CCF_IPHDR_CE;
523 tp->ccv->flags &= ~CCF_IPHDR_CE;
526 tp->ccv->flags &= ~CCF_IPHDR_CE;
530 if (th->th_flags & TH_CWR)
531 tp->ccv->flags |= CCF_TCPHDR_CWR;
533 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
535 if (tp->t_flags & TF_DELACK)
536 tp->ccv->flags |= CCF_DELACK;
538 tp->ccv->flags &= ~CCF_DELACK;
540 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
542 if (tp->ccv->flags & CCF_ACKNOW)
543 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
548 * TCP input handling is split into multiple parts:
549 * tcp6_input is a thin wrapper around tcp_input for the extended
550 * ip6_protox[] call format in ip6_input
551 * tcp_input handles primary segment validation, inpcb lookup and
552 * SYN processing on listen sockets
553 * tcp_do_segment processes the ACK and text of the segment for
554 * establishing, established and closing connections
558 tcp6_input(struct mbuf **mp, int *offp, int proto)
560 struct mbuf *m = *mp;
561 struct in6_ifaddr *ia6;
564 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
567 * draft-itojun-ipv6-tcp-to-anycast
568 * better place to put this in?
570 ip6 = mtod(m, struct ip6_hdr *);
571 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
572 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
575 ifa_free(&ia6->ia_ifa);
576 ip6 = mtod(m, struct ip6_hdr *);
577 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
578 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
579 return (IPPROTO_DONE);
582 ifa_free(&ia6->ia_ifa);
584 return (tcp_input(mp, offp, proto));
589 tcp_input(struct mbuf **mp, int *offp, int proto)
591 struct mbuf *m = *mp;
592 struct tcphdr *th = NULL;
593 struct ip *ip = NULL;
594 struct inpcb *inp = NULL;
595 struct tcpcb *tp = NULL;
596 struct socket *so = NULL;
606 int rstreason = 0; /* For badport_bandlim accounting purposes */
608 uint8_t sig_checked = 0;
611 struct m_tag *fwd_tag = NULL;
613 struct ip6_hdr *ip6 = NULL;
616 const void *ip6 = NULL;
618 struct tcpopt to; /* options in this segment */
619 char *s = NULL; /* address and port logging */
623 * The size of tcp_saveipgen must be the size of the max ip header,
626 u_char tcp_saveipgen[IP6_HDR_LEN];
627 struct tcphdr tcp_savetcp;
632 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
639 TCPSTAT_INC(tcps_rcvtotal);
643 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
645 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
646 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
648 TCPSTAT_INC(tcps_rcvshort);
649 return (IPPROTO_DONE);
653 ip6 = mtod(m, struct ip6_hdr *);
654 th = (struct tcphdr *)((caddr_t)ip6 + off0);
655 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
656 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
657 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
658 th->th_sum = m->m_pkthdr.csum_data;
660 th->th_sum = in6_cksum_pseudo(ip6, tlen,
661 IPPROTO_TCP, m->m_pkthdr.csum_data);
662 th->th_sum ^= 0xffff;
664 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
666 TCPSTAT_INC(tcps_rcvbadsum);
671 * Be proactive about unspecified IPv6 address in source.
672 * As we use all-zero to indicate unbounded/unconnected pcb,
673 * unspecified IPv6 address can be used to confuse us.
675 * Note that packets with unspecified IPv6 destination is
676 * already dropped in ip6_input.
678 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
682 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
685 #if defined(INET) && defined(INET6)
691 * Get IP and TCP header together in first mbuf.
692 * Note: IP leaves IP header in first mbuf.
694 if (off0 > sizeof (struct ip)) {
696 off0 = sizeof(struct ip);
698 if (m->m_len < sizeof (struct tcpiphdr)) {
699 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
701 TCPSTAT_INC(tcps_rcvshort);
702 return (IPPROTO_DONE);
705 ip = mtod(m, struct ip *);
706 th = (struct tcphdr *)((caddr_t)ip + off0);
707 tlen = ntohs(ip->ip_len) - off0;
710 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
711 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
712 th->th_sum = m->m_pkthdr.csum_data;
714 th->th_sum = in_pseudo(ip->ip_src.s_addr,
716 htonl(m->m_pkthdr.csum_data + tlen +
718 th->th_sum ^= 0xffff;
720 struct ipovly *ipov = (struct ipovly *)ip;
723 * Checksum extended TCP header and data.
726 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
727 ipov->ih_len = htons(tlen);
728 th->th_sum = in_cksum(m, len);
729 /* Reset length for SDT probes. */
730 ip->ip_len = htons(len);
733 /* Re-initialization for later version check */
734 ip->ip_v = IPVERSION;
738 TCPSTAT_INC(tcps_rcvbadsum);
745 * Check that TCP offset makes sense,
746 * pull out TCP options and adjust length. XXX
748 off = th->th_off << 2;
749 if (off < sizeof (struct tcphdr) || off > tlen) {
750 TCPSTAT_INC(tcps_rcvbadoff);
753 tlen -= off; /* tlen is used instead of ti->ti_len */
754 if (off > sizeof (struct tcphdr)) {
757 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
758 ip6 = mtod(m, struct ip6_hdr *);
759 th = (struct tcphdr *)((caddr_t)ip6 + off0);
762 #if defined(INET) && defined(INET6)
767 if (m->m_len < sizeof(struct ip) + off) {
768 if ((m = m_pullup(m, sizeof (struct ip) + off))
770 TCPSTAT_INC(tcps_rcvshort);
771 return (IPPROTO_DONE);
773 ip = mtod(m, struct ip *);
774 th = (struct tcphdr *)((caddr_t)ip + off0);
778 optlen = off - sizeof (struct tcphdr);
779 optp = (u_char *)(th + 1);
781 thflags = th->th_flags;
784 * Convert TCP protocol specific fields to host format.
786 tcp_fields_to_host(th);
789 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
791 drop_hdrlen = off0 + off;
794 * Locate pcb for segment; if we're likely to add or remove a
795 * connection then first acquire pcbinfo lock. There are three cases
796 * where we might discover later we need a write lock despite the
797 * flags: ACKs moving a connection out of the syncache, ACKs for a
798 * connection in TIMEWAIT and SYNs not targeting a listening socket.
800 if ((thflags & (TH_FIN | TH_RST)) != 0) {
801 INP_INFO_RLOCK(&V_tcbinfo);
802 ti_locked = TI_RLOCKED;
804 ti_locked = TI_UNLOCKED;
807 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
811 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
813 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
816 #if defined(INET) && !defined(INET6)
817 (m->m_flags & M_IP_NEXTHOP)
820 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
824 if (ti_locked == TI_RLOCKED) {
825 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
827 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
831 if (isipv6 && fwd_tag != NULL) {
832 struct sockaddr_in6 *next_hop6;
834 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
836 * Transparently forwarded. Pretend to be the destination.
837 * Already got one like this?
839 inp = in6_pcblookup_mbuf(&V_tcbinfo,
840 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
841 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
844 * It's new. Try to find the ambushing socket.
845 * Because we've rewritten the destination address,
846 * any hardware-generated hash is ignored.
848 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
849 th->th_sport, &next_hop6->sin6_addr,
850 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
851 th->th_dport, INPLOOKUP_WILDCARD |
852 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
855 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
856 th->th_sport, &ip6->ip6_dst, th->th_dport,
857 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
858 m->m_pkthdr.rcvif, m);
861 #if defined(INET6) && defined(INET)
865 if (fwd_tag != NULL) {
866 struct sockaddr_in *next_hop;
868 next_hop = (struct sockaddr_in *)(fwd_tag+1);
870 * Transparently forwarded. Pretend to be the destination.
871 * already got one like this?
873 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
874 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
875 m->m_pkthdr.rcvif, m);
878 * It's new. Try to find the ambushing socket.
879 * Because we've rewritten the destination address,
880 * any hardware-generated hash is ignored.
882 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
883 th->th_sport, next_hop->sin_addr,
884 next_hop->sin_port ? ntohs(next_hop->sin_port) :
885 th->th_dport, INPLOOKUP_WILDCARD |
886 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
889 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
890 th->th_sport, ip->ip_dst, th->th_dport,
891 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
892 m->m_pkthdr.rcvif, m);
896 * If the INPCB does not exist then all data in the incoming
897 * segment is discarded and an appropriate RST is sent back.
898 * XXX MRT Send RST using which routing table?
902 * Log communication attempts to ports that are not
905 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
906 tcp_log_in_vain == 2) {
907 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
908 log(LOG_INFO, "%s; %s: Connection attempt "
909 "to closed port\n", s, __func__);
912 * When blackholing do not respond with a RST but
913 * completely ignore the segment and drop it.
915 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
919 rstreason = BANDLIM_RST_CLOSEDPORT;
922 INP_WLOCK_ASSERT(inp);
923 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
924 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
925 ((inp->inp_socket == NULL) ||
926 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
927 inp->inp_flowid = m->m_pkthdr.flowid;
928 inp->inp_flowtype = M_HASHTYPE_GET(m);
932 if (isipv6 && ipsec6_in_reject(m, inp)) {
936 if (ipsec4_in_reject(m, inp) != 0) {
942 * Check the minimum TTL for socket.
944 if (inp->inp_ip_minttl != 0) {
947 if (inp->inp_ip_minttl > ip6->ip6_hlim)
951 if (inp->inp_ip_minttl > ip->ip_ttl)
956 * A previous connection in TIMEWAIT state is supposed to catch stray
957 * or duplicate segments arriving late. If this segment was a
958 * legitimate new connection attempt, the old INPCB gets removed and
959 * we can try again to find a listening socket.
961 * At this point, due to earlier optimism, we may hold only an inpcb
962 * lock, and not the inpcbinfo write lock. If so, we need to try to
963 * acquire it, or if that fails, acquire a reference on the inpcb,
964 * drop all locks, acquire a global write lock, and then re-acquire
965 * the inpcb lock. We may at that point discover that another thread
966 * has tried to free the inpcb, in which case we need to loop back
967 * and try to find a new inpcb to deliver to.
969 * XXXRW: It may be time to rethink timewait locking.
972 if (inp->inp_flags & INP_TIMEWAIT) {
973 if (ti_locked == TI_UNLOCKED) {
974 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
977 INP_INFO_RLOCK(&V_tcbinfo);
978 ti_locked = TI_RLOCKED;
980 if (in_pcbrele_wlocked(inp)) {
985 ti_locked = TI_RLOCKED;
987 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
989 if (thflags & TH_SYN)
990 tcp_dooptions(&to, optp, optlen, TO_SYN);
992 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
994 if (tcp_twcheck(inp, &to, th, m, tlen))
996 INP_INFO_RUNLOCK(&V_tcbinfo);
997 return (IPPROTO_DONE);
1000 * The TCPCB may no longer exist if the connection is winding
1001 * down or it is in the CLOSED state. Either way we drop the
1002 * segment and send an appropriate response.
1004 tp = intotcpcb(inp);
1005 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1006 rstreason = BANDLIM_RST_CLOSEDPORT;
1011 if (tp->t_flags & TF_TOE) {
1012 tcp_offload_input(tp, m);
1013 m = NULL; /* consumed by the TOE driver */
1019 * We've identified a valid inpcb, but it could be that we need an
1020 * inpcbinfo write lock but don't hold it. In this case, attempt to
1021 * acquire using the same strategy as the TIMEWAIT case above. If we
1022 * relock, we have to jump back to 'relocked' as the connection might
1023 * now be in TIMEWAIT.
1026 if ((thflags & (TH_FIN | TH_RST)) != 0)
1027 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1029 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) ||
1030 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) &&
1031 !IS_FASTOPEN(tp->t_flags)))) {
1032 if (ti_locked == TI_UNLOCKED) {
1033 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
1036 INP_INFO_RLOCK(&V_tcbinfo);
1037 ti_locked = TI_RLOCKED;
1039 if (in_pcbrele_wlocked(inp)) {
1045 ti_locked = TI_RLOCKED;
1047 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1051 INP_WLOCK_ASSERT(inp);
1052 if (mac_inpcb_check_deliver(inp, m))
1055 so = inp->inp_socket;
1056 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1058 if (so->so_options & SO_DEBUG) {
1059 ostate = tp->t_state;
1062 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1065 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1068 #endif /* TCPDEBUG */
1070 * When the socket is accepting connections (the INPCB is in LISTEN
1071 * state) we look into the SYN cache if this is a new connection
1072 * attempt or the completion of a previous one.
1074 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1075 ("%s: so accepting but tp %p not listening", __func__, tp));
1076 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1077 struct in_conninfo inc;
1079 bzero(&inc, sizeof(inc));
1082 inc.inc_flags |= INC_ISIPV6;
1083 inc.inc6_faddr = ip6->ip6_src;
1084 inc.inc6_laddr = ip6->ip6_dst;
1088 inc.inc_faddr = ip->ip_src;
1089 inc.inc_laddr = ip->ip_dst;
1091 inc.inc_fport = th->th_sport;
1092 inc.inc_lport = th->th_dport;
1093 inc.inc_fibnum = so->so_fibnum;
1096 * Check for an existing connection attempt in syncache if
1097 * the flag is only ACK. A successful lookup creates a new
1098 * socket appended to the listen queue in SYN_RECEIVED state.
1100 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1102 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1104 * Parse the TCP options here because
1105 * syncookies need access to the reflected
1108 tcp_dooptions(&to, optp, optlen, 0);
1110 * NB: syncache_expand() doesn't unlock
1111 * inp and tcpinfo locks.
1113 if (!syncache_expand(&inc, &to, th, &so, m)) {
1115 * No syncache entry or ACK was not
1116 * for our SYN/ACK. Send a RST.
1117 * NB: syncache did its own logging
1118 * of the failure cause.
1120 rstreason = BANDLIM_RST_OPENPORT;
1128 * We completed the 3-way handshake
1129 * but could not allocate a socket
1130 * either due to memory shortage,
1131 * listen queue length limits or
1132 * global socket limits. Send RST
1133 * or wait and have the remote end
1134 * retransmit the ACK for another
1137 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1138 log(LOG_DEBUG, "%s; %s: Listen socket: "
1139 "Socket allocation failed due to "
1140 "limits or memory shortage, %s\n",
1142 V_tcp_sc_rst_sock_fail ?
1143 "sending RST" : "try again");
1144 if (V_tcp_sc_rst_sock_fail) {
1145 rstreason = BANDLIM_UNLIMITED;
1151 * Socket is created in state SYN_RECEIVED.
1152 * Unlock the listen socket, lock the newly
1153 * created socket and update the tp variable.
1155 INP_WUNLOCK(inp); /* listen socket */
1156 inp = sotoinpcb(so);
1158 * New connection inpcb is already locked by
1159 * syncache_expand().
1161 INP_WLOCK_ASSERT(inp);
1162 tp = intotcpcb(inp);
1163 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1164 ("%s: ", __func__));
1165 #ifdef TCP_SIGNATURE
1166 if (sig_checked == 0) {
1167 tcp_dooptions(&to, optp, optlen,
1168 (thflags & TH_SYN) ? TO_SYN : 0);
1169 if (!tcp_signature_verify_input(m, off0, tlen,
1170 optlen, &to, th, tp->t_flags)) {
1173 * In SYN_SENT state if it receives an
1174 * RST, it is allowed for further
1177 if ((thflags & TH_RST) == 0 ||
1178 (tp->t_state == TCPS_SYN_SENT) == 0)
1186 * Process the segment and the data it
1187 * contains. tcp_do_segment() consumes
1188 * the mbuf chain and unlocks the inpcb.
1190 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1192 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1193 return (IPPROTO_DONE);
1196 * Segment flag validation for new connection attempts:
1198 * Our (SYN|ACK) response was rejected.
1199 * Check with syncache and remove entry to prevent
1202 * NB: syncache_chkrst does its own logging of failure
1205 if (thflags & TH_RST) {
1206 syncache_chkrst(&inc, th);
1210 * We can't do anything without SYN.
1212 if ((thflags & TH_SYN) == 0) {
1213 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1214 log(LOG_DEBUG, "%s; %s: Listen socket: "
1215 "SYN is missing, segment ignored\n",
1217 TCPSTAT_INC(tcps_badsyn);
1221 * (SYN|ACK) is bogus on a listen socket.
1223 if (thflags & TH_ACK) {
1224 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1225 log(LOG_DEBUG, "%s; %s: Listen socket: "
1226 "SYN|ACK invalid, segment rejected\n",
1228 syncache_badack(&inc); /* XXX: Not needed! */
1229 TCPSTAT_INC(tcps_badsyn);
1230 rstreason = BANDLIM_RST_OPENPORT;
1234 * If the drop_synfin option is enabled, drop all
1235 * segments with both the SYN and FIN bits set.
1236 * This prevents e.g. nmap from identifying the
1238 * XXX: Poor reasoning. nmap has other methods
1239 * and is constantly refining its stack detection
1241 * XXX: This is a violation of the TCP specification
1242 * and was used by RFC1644.
1244 if ((thflags & TH_FIN) && V_drop_synfin) {
1245 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1246 log(LOG_DEBUG, "%s; %s: Listen socket: "
1247 "SYN|FIN segment ignored (based on "
1248 "sysctl setting)\n", s, __func__);
1249 TCPSTAT_INC(tcps_badsyn);
1253 * Segment's flags are (SYN) or (SYN|FIN).
1255 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1256 * as they do not affect the state of the TCP FSM.
1257 * The data pointed to by TH_URG and th_urp is ignored.
1259 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1260 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1261 KASSERT(thflags & (TH_SYN),
1262 ("%s: Listen socket: TH_SYN not set", __func__));
1265 * If deprecated address is forbidden,
1266 * we do not accept SYN to deprecated interface
1267 * address to prevent any new inbound connection from
1268 * getting established.
1269 * When we do not accept SYN, we send a TCP RST,
1270 * with deprecated source address (instead of dropping
1271 * it). We compromise it as it is much better for peer
1272 * to send a RST, and RST will be the final packet
1275 * If we do not forbid deprecated addresses, we accept
1276 * the SYN packet. RFC2462 does not suggest dropping
1278 * If we decipher RFC2462 5.5.4, it says like this:
1279 * 1. use of deprecated addr with existing
1280 * communication is okay - "SHOULD continue to be
1282 * 2. use of it with new communication:
1283 * (2a) "SHOULD NOT be used if alternate address
1284 * with sufficient scope is available"
1285 * (2b) nothing mentioned otherwise.
1286 * Here we fall into (2b) case as we have no choice in
1287 * our source address selection - we must obey the peer.
1289 * The wording in RFC2462 is confusing, and there are
1290 * multiple description text for deprecated address
1291 * handling - worse, they are not exactly the same.
1292 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1294 if (isipv6 && !V_ip6_use_deprecated) {
1295 struct in6_ifaddr *ia6;
1297 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1299 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1300 ifa_free(&ia6->ia_ifa);
1301 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1302 log(LOG_DEBUG, "%s; %s: Listen socket: "
1303 "Connection attempt to deprecated "
1304 "IPv6 address rejected\n",
1306 rstreason = BANDLIM_RST_OPENPORT;
1310 ifa_free(&ia6->ia_ifa);
1314 * Basic sanity checks on incoming SYN requests:
1315 * Don't respond if the destination is a link layer
1316 * broadcast according to RFC1122 4.2.3.10, p. 104.
1317 * If it is from this socket it must be forged.
1318 * Don't respond if the source or destination is a
1319 * global or subnet broad- or multicast address.
1320 * Note that it is quite possible to receive unicast
1321 * link-layer packets with a broadcast IP address. Use
1322 * in_broadcast() to find them.
1324 if (m->m_flags & (M_BCAST|M_MCAST)) {
1325 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1326 log(LOG_DEBUG, "%s; %s: Listen socket: "
1327 "Connection attempt from broad- or multicast "
1328 "link layer address ignored\n", s, __func__);
1333 if (th->th_dport == th->th_sport &&
1334 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1335 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1336 log(LOG_DEBUG, "%s; %s: Listen socket: "
1337 "Connection attempt to/from self "
1338 "ignored\n", s, __func__);
1341 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1342 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1343 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1344 log(LOG_DEBUG, "%s; %s: Listen socket: "
1345 "Connection attempt from/to multicast "
1346 "address ignored\n", s, __func__);
1351 #if defined(INET) && defined(INET6)
1356 if (th->th_dport == th->th_sport &&
1357 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1358 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1359 log(LOG_DEBUG, "%s; %s: Listen socket: "
1360 "Connection attempt from/to self "
1361 "ignored\n", s, __func__);
1364 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1365 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1366 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1367 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1368 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1369 log(LOG_DEBUG, "%s; %s: Listen socket: "
1370 "Connection attempt from/to broad- "
1371 "or multicast address ignored\n",
1378 * SYN appears to be valid. Create compressed TCP state
1382 if (so->so_options & SO_DEBUG)
1383 tcp_trace(TA_INPUT, ostate, tp,
1384 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1386 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1387 tcp_dooptions(&to, optp, optlen, TO_SYN);
1389 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL))
1390 goto tfo_socket_result;
1392 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1395 * Entry added to syncache and mbuf consumed.
1396 * Only the listen socket is unlocked by syncache_add().
1398 if (ti_locked == TI_RLOCKED) {
1399 INP_INFO_RUNLOCK(&V_tcbinfo);
1400 ti_locked = TI_UNLOCKED;
1402 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1403 return (IPPROTO_DONE);
1404 } else if (tp->t_state == TCPS_LISTEN) {
1406 * When a listen socket is torn down the SO_ACCEPTCONN
1407 * flag is removed first while connections are drained
1408 * from the accept queue in a unlock/lock cycle of the
1409 * ACCEPT_LOCK, opening a race condition allowing a SYN
1410 * attempt go through unhandled.
1415 #ifdef TCP_SIGNATURE
1416 if (sig_checked == 0) {
1417 tcp_dooptions(&to, optp, optlen,
1418 (thflags & TH_SYN) ? TO_SYN : 0);
1419 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1423 * In SYN_SENT state if it receives an RST, it is
1424 * allowed for further processing.
1426 if ((thflags & TH_RST) == 0 ||
1427 (tp->t_state == TCPS_SYN_SENT) == 0)
1434 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1437 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1438 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1439 * the inpcb, and unlocks pcbinfo.
1441 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1442 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1443 return (IPPROTO_DONE);
1446 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1448 if (ti_locked == TI_RLOCKED) {
1449 INP_INFO_RUNLOCK(&V_tcbinfo);
1450 ti_locked = TI_UNLOCKED;
1454 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1455 "ti_locked: %d", __func__, ti_locked));
1456 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1461 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1464 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1465 m = NULL; /* mbuf chain got consumed. */
1470 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1472 if (ti_locked == TI_RLOCKED) {
1473 INP_INFO_RUNLOCK(&V_tcbinfo);
1474 ti_locked = TI_UNLOCKED;
1478 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1479 "ti_locked: %d", __func__, ti_locked));
1480 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1488 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1493 return (IPPROTO_DONE);
1497 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1498 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1501 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1502 int rstreason, todrop, win;
1506 struct in_conninfo *inc;
1515 * The size of tcp_saveipgen must be the size of the max ip header,
1518 u_char tcp_saveipgen[IP6_HDR_LEN];
1519 struct tcphdr tcp_savetcp;
1522 thflags = th->th_flags;
1523 inc = &tp->t_inpcb->inp_inc;
1524 tp->sackhint.last_sack_ack = 0;
1526 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1529 * If this is either a state-changing packet or current state isn't
1530 * established, we require a write lock on tcbinfo. Otherwise, we
1531 * allow the tcbinfo to be in either alocked or unlocked, as the
1532 * caller may have unnecessarily acquired a write lock due to a race.
1534 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1535 tp->t_state != TCPS_ESTABLISHED) {
1536 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for "
1537 "SYN/FIN/RST/!EST", __func__, ti_locked));
1538 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1541 if (ti_locked == TI_RLOCKED)
1542 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1544 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1545 "ti_locked: %d", __func__, ti_locked));
1546 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1550 INP_WLOCK_ASSERT(tp->t_inpcb);
1551 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1553 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1557 /* Save segment, if requested. */
1558 tcp_pcap_add(th, m, &(tp->t_inpkts));
1562 * Segment received on connection.
1563 * Reset idle time and keep-alive timer.
1564 * XXX: This should be done after segment
1565 * validation to ignore broken/spoofed segs.
1567 tp->t_rcvtime = ticks;
1570 * Scale up the window into a 32-bit value.
1571 * For the SYN_SENT state the scale is zero.
1573 tiwin = th->th_win << tp->snd_scale;
1576 * TCP ECN processing.
1578 if (tp->t_flags & TF_ECN_PERMIT) {
1579 if (thflags & TH_CWR)
1580 tp->t_flags &= ~TF_ECN_SND_ECE;
1581 switch (iptos & IPTOS_ECN_MASK) {
1583 tp->t_flags |= TF_ECN_SND_ECE;
1584 TCPSTAT_INC(tcps_ecn_ce);
1586 case IPTOS_ECN_ECT0:
1587 TCPSTAT_INC(tcps_ecn_ect0);
1589 case IPTOS_ECN_ECT1:
1590 TCPSTAT_INC(tcps_ecn_ect1);
1594 /* Process a packet differently from RFC3168. */
1595 cc_ecnpkt_handler(tp, th, iptos);
1597 /* Congestion experienced. */
1598 if (thflags & TH_ECE) {
1599 cc_cong_signal(tp, th, CC_ECN);
1604 * Parse options on any incoming segment.
1606 tcp_dooptions(&to, (u_char *)(th + 1),
1607 (th->th_off << 2) - sizeof(struct tcphdr),
1608 (thflags & TH_SYN) ? TO_SYN : 0);
1611 * If echoed timestamp is later than the current time,
1612 * fall back to non RFC1323 RTT calculation. Normalize
1613 * timestamp if syncookies were used when this connection
1616 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1617 to.to_tsecr -= tp->ts_offset;
1618 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1622 * If timestamps were negotiated during SYN/ACK they should
1623 * appear on every segment during this session and vice versa.
1625 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1626 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1627 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1628 "no action\n", s, __func__);
1632 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1633 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1634 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1635 "no action\n", s, __func__);
1641 * Process options only when we get SYN/ACK back. The SYN case
1642 * for incoming connections is handled in tcp_syncache.
1643 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1644 * or <SYN,ACK>) segment itself is never scaled.
1645 * XXX this is traditional behavior, may need to be cleaned up.
1647 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1648 if ((to.to_flags & TOF_SCALE) &&
1649 (tp->t_flags & TF_REQ_SCALE)) {
1650 tp->t_flags |= TF_RCVD_SCALE;
1651 tp->snd_scale = to.to_wscale;
1654 * Initial send window. It will be updated with
1655 * the next incoming segment to the scaled value.
1657 tp->snd_wnd = th->th_win;
1658 if (to.to_flags & TOF_TS) {
1659 tp->t_flags |= TF_RCVD_TSTMP;
1660 tp->ts_recent = to.to_tsval;
1661 tp->ts_recent_age = tcp_ts_getticks();
1663 if (to.to_flags & TOF_MSS)
1664 tcp_mss(tp, to.to_mss);
1665 if ((tp->t_flags & TF_SACK_PERMIT) &&
1666 (to.to_flags & TOF_SACKPERM) == 0)
1667 tp->t_flags &= ~TF_SACK_PERMIT;
1671 * Header prediction: check for the two common cases
1672 * of a uni-directional data xfer. If the packet has
1673 * no control flags, is in-sequence, the window didn't
1674 * change and we're not retransmitting, it's a
1675 * candidate. If the length is zero and the ack moved
1676 * forward, we're the sender side of the xfer. Just
1677 * free the data acked & wake any higher level process
1678 * that was blocked waiting for space. If the length
1679 * is non-zero and the ack didn't move, we're the
1680 * receiver side. If we're getting packets in-order
1681 * (the reassembly queue is empty), add the data to
1682 * the socket buffer and note that we need a delayed ack.
1683 * Make sure that the hidden state-flags are also off.
1684 * Since we check for TCPS_ESTABLISHED first, it can only
1687 if (tp->t_state == TCPS_ESTABLISHED &&
1688 th->th_seq == tp->rcv_nxt &&
1689 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1690 tp->snd_nxt == tp->snd_max &&
1691 tiwin && tiwin == tp->snd_wnd &&
1692 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1693 LIST_EMPTY(&tp->t_segq) &&
1694 ((to.to_flags & TOF_TS) == 0 ||
1695 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1698 * If last ACK falls within this segment's sequence numbers,
1699 * record the timestamp.
1700 * NOTE that the test is modified according to the latest
1701 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1703 if ((to.to_flags & TOF_TS) != 0 &&
1704 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1705 tp->ts_recent_age = tcp_ts_getticks();
1706 tp->ts_recent = to.to_tsval;
1710 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1711 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1712 !IN_RECOVERY(tp->t_flags) &&
1713 (to.to_flags & TOF_SACK) == 0 &&
1714 TAILQ_EMPTY(&tp->snd_holes)) {
1716 * This is a pure ack for outstanding data.
1718 if (ti_locked == TI_RLOCKED)
1719 INP_INFO_RUNLOCK(&V_tcbinfo);
1720 ti_locked = TI_UNLOCKED;
1722 TCPSTAT_INC(tcps_predack);
1725 * "bad retransmit" recovery.
1727 if (tp->t_rxtshift == 1 &&
1728 tp->t_flags & TF_PREVVALID &&
1729 (int)(ticks - tp->t_badrxtwin) < 0) {
1730 cc_cong_signal(tp, th, CC_RTO_ERR);
1734 * Recalculate the transmit timer / rtt.
1736 * Some boxes send broken timestamp replies
1737 * during the SYN+ACK phase, ignore
1738 * timestamps of 0 or we could calculate a
1739 * huge RTT and blow up the retransmit timer.
1741 if ((to.to_flags & TOF_TS) != 0 &&
1745 t = tcp_ts_getticks() - to.to_tsecr;
1746 if (!tp->t_rttlow || tp->t_rttlow > t)
1749 TCP_TS_TO_TICKS(t) + 1);
1750 } else if (tp->t_rtttime &&
1751 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1752 if (!tp->t_rttlow ||
1753 tp->t_rttlow > ticks - tp->t_rtttime)
1754 tp->t_rttlow = ticks - tp->t_rtttime;
1756 ticks - tp->t_rtttime);
1758 acked = BYTES_THIS_ACK(tp, th);
1761 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1762 hhook_run_tcp_est_in(tp, th, &to);
1765 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1766 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1767 sbdrop(&so->so_snd, acked);
1768 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1769 SEQ_LEQ(th->th_ack, tp->snd_recover))
1770 tp->snd_recover = th->th_ack - 1;
1773 * Let the congestion control algorithm update
1774 * congestion control related information. This
1775 * typically means increasing the congestion
1778 cc_ack_received(tp, th, nsegs, CC_ACK);
1780 tp->snd_una = th->th_ack;
1782 * Pull snd_wl2 up to prevent seq wrap relative
1785 tp->snd_wl2 = th->th_ack;
1790 * If all outstanding data are acked, stop
1791 * retransmit timer, otherwise restart timer
1792 * using current (possibly backed-off) value.
1793 * If process is waiting for space,
1794 * wakeup/selwakeup/signal. If data
1795 * are ready to send, let tcp_output
1796 * decide between more output or persist.
1799 if (so->so_options & SO_DEBUG)
1800 tcp_trace(TA_INPUT, ostate, tp,
1801 (void *)tcp_saveipgen,
1804 TCP_PROBE3(debug__input, tp, th,
1805 mtod(m, const char *));
1806 if (tp->snd_una == tp->snd_max)
1807 tcp_timer_activate(tp, TT_REXMT, 0);
1808 else if (!tcp_timer_active(tp, TT_PERSIST))
1809 tcp_timer_activate(tp, TT_REXMT,
1812 if (sbavail(&so->so_snd))
1813 (void) tp->t_fb->tfb_tcp_output(tp);
1816 } else if (th->th_ack == tp->snd_una &&
1817 tlen <= sbspace(&so->so_rcv)) {
1818 int newsize = 0; /* automatic sockbuf scaling */
1821 * This is a pure, in-sequence data packet with
1822 * nothing on the reassembly queue and we have enough
1823 * buffer space to take it.
1825 if (ti_locked == TI_RLOCKED)
1826 INP_INFO_RUNLOCK(&V_tcbinfo);
1827 ti_locked = TI_UNLOCKED;
1829 /* Clean receiver SACK report if present */
1830 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1831 tcp_clean_sackreport(tp);
1832 TCPSTAT_INC(tcps_preddat);
1833 tp->rcv_nxt += tlen;
1835 * Pull snd_wl1 up to prevent seq wrap relative to
1838 tp->snd_wl1 = th->th_seq;
1840 * Pull rcv_up up to prevent seq wrap relative to
1843 tp->rcv_up = tp->rcv_nxt;
1844 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1845 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1847 if (so->so_options & SO_DEBUG)
1848 tcp_trace(TA_INPUT, ostate, tp,
1849 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1851 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1854 * Automatic sizing of receive socket buffer. Often the send
1855 * buffer size is not optimally adjusted to the actual network
1856 * conditions at hand (delay bandwidth product). Setting the
1857 * buffer size too small limits throughput on links with high
1858 * bandwidth and high delay (eg. trans-continental/oceanic links).
1860 * On the receive side the socket buffer memory is only rarely
1861 * used to any significant extent. This allows us to be much
1862 * more aggressive in scaling the receive socket buffer. For
1863 * the case that the buffer space is actually used to a large
1864 * extent and we run out of kernel memory we can simply drop
1865 * the new segments; TCP on the sender will just retransmit it
1866 * later. Setting the buffer size too big may only consume too
1867 * much kernel memory if the application doesn't read() from
1868 * the socket or packet loss or reordering makes use of the
1871 * The criteria to step up the receive buffer one notch are:
1872 * 1. Application has not set receive buffer size with
1873 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1874 * 2. the number of bytes received during the time it takes
1875 * one timestamp to be reflected back to us (the RTT);
1876 * 3. received bytes per RTT is within seven eighth of the
1877 * current socket buffer size;
1878 * 4. receive buffer size has not hit maximal automatic size;
1880 * This algorithm does one step per RTT at most and only if
1881 * we receive a bulk stream w/o packet losses or reorderings.
1882 * Shrinking the buffer during idle times is not necessary as
1883 * it doesn't consume any memory when idle.
1885 * TODO: Only step up if the application is actually serving
1886 * the buffer to better manage the socket buffer resources.
1888 if (V_tcp_do_autorcvbuf &&
1889 (to.to_flags & TOF_TS) &&
1891 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1892 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1893 to.to_tsecr - tp->rfbuf_ts < hz) {
1895 (so->so_rcv.sb_hiwat / 8 * 7) &&
1896 so->so_rcv.sb_hiwat <
1897 V_tcp_autorcvbuf_max) {
1899 min(so->so_rcv.sb_hiwat +
1900 V_tcp_autorcvbuf_inc,
1901 V_tcp_autorcvbuf_max);
1903 /* Start over with next RTT. */
1907 tp->rfbuf_cnt += tlen; /* add up */
1910 /* Add data to socket buffer. */
1911 SOCKBUF_LOCK(&so->so_rcv);
1912 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1916 * Set new socket buffer size.
1917 * Give up when limit is reached.
1920 if (!sbreserve_locked(&so->so_rcv,
1922 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1923 m_adj(m, drop_hdrlen); /* delayed header drop */
1924 sbappendstream_locked(&so->so_rcv, m, 0);
1926 /* NB: sorwakeup_locked() does an implicit unlock. */
1927 sorwakeup_locked(so);
1928 if (DELAY_ACK(tp, tlen)) {
1929 tp->t_flags |= TF_DELACK;
1931 tp->t_flags |= TF_ACKNOW;
1932 tp->t_fb->tfb_tcp_output(tp);
1939 * Calculate amount of space in receive window,
1940 * and then do TCP input processing.
1941 * Receive window is amount of space in rcv queue,
1942 * but not less than advertised window.
1944 win = sbspace(&so->so_rcv);
1947 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1949 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1953 switch (tp->t_state) {
1956 * If the state is SYN_RECEIVED:
1957 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1959 case TCPS_SYN_RECEIVED:
1960 if ((thflags & TH_ACK) &&
1961 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1962 SEQ_GT(th->th_ack, tp->snd_max))) {
1963 rstreason = BANDLIM_RST_OPENPORT;
1967 if (IS_FASTOPEN(tp->t_flags)) {
1969 * When a TFO connection is in SYN_RECEIVED, the
1970 * only valid packets are the initial SYN, a
1971 * retransmit/copy of the initial SYN (possibly with
1972 * a subset of the original data), a valid ACK, a
1975 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1976 rstreason = BANDLIM_RST_OPENPORT;
1978 } else if (thflags & TH_SYN) {
1979 /* non-initial SYN is ignored */
1980 if ((tcp_timer_active(tp, TT_DELACK) ||
1981 tcp_timer_active(tp, TT_REXMT)))
1983 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1991 * If the state is SYN_SENT:
1992 * if seg contains an ACK, but not for our SYN, drop the input.
1993 * if seg contains a RST, then drop the connection.
1994 * if seg does not contain SYN, then drop it.
1995 * Otherwise this is an acceptable SYN segment
1996 * initialize tp->rcv_nxt and tp->irs
1997 * if seg contains ack then advance tp->snd_una
1998 * if seg contains an ECE and ECN support is enabled, the stream
2000 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2001 * arrange for segment to be acked (eventually)
2002 * continue processing rest of data/controls, beginning with URG
2005 if ((thflags & TH_ACK) &&
2006 (SEQ_LEQ(th->th_ack, tp->iss) ||
2007 SEQ_GT(th->th_ack, tp->snd_max))) {
2008 rstreason = BANDLIM_UNLIMITED;
2011 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2012 TCP_PROBE5(connect__refused, NULL, tp,
2013 mtod(m, const char *), tp, th);
2014 tp = tcp_drop(tp, ECONNREFUSED);
2016 if (thflags & TH_RST)
2018 if (!(thflags & TH_SYN))
2021 tp->irs = th->th_seq;
2023 if (thflags & TH_ACK) {
2024 TCPSTAT_INC(tcps_connects);
2027 mac_socketpeer_set_from_mbuf(m, so);
2029 /* Do window scaling on this connection? */
2030 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2031 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2032 tp->rcv_scale = tp->request_r_scale;
2034 tp->rcv_adv += min(tp->rcv_wnd,
2035 TCP_MAXWIN << tp->rcv_scale);
2036 tp->snd_una++; /* SYN is acked */
2038 * If there's data, delay ACK; if there's also a FIN
2039 * ACKNOW will be turned on later.
2041 if (DELAY_ACK(tp, tlen) && tlen != 0)
2042 tcp_timer_activate(tp, TT_DELACK,
2045 tp->t_flags |= TF_ACKNOW;
2047 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
2048 tp->t_flags |= TF_ECN_PERMIT;
2049 TCPSTAT_INC(tcps_ecn_shs);
2053 * Received <SYN,ACK> in SYN_SENT[*] state.
2055 * SYN_SENT --> ESTABLISHED
2056 * SYN_SENT* --> FIN_WAIT_1
2058 tp->t_starttime = ticks;
2059 if (tp->t_flags & TF_NEEDFIN) {
2060 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2061 tp->t_flags &= ~TF_NEEDFIN;
2064 tcp_state_change(tp, TCPS_ESTABLISHED);
2065 TCP_PROBE5(connect__established, NULL, tp,
2066 mtod(m, const char *), tp, th);
2068 tcp_timer_activate(tp, TT_KEEP,
2073 * Received initial SYN in SYN-SENT[*] state =>
2074 * simultaneous open.
2075 * If it succeeds, connection is * half-synchronized.
2076 * Otherwise, do 3-way handshake:
2077 * SYN-SENT -> SYN-RECEIVED
2078 * SYN-SENT* -> SYN-RECEIVED*
2080 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2081 tcp_timer_activate(tp, TT_REXMT, 0);
2082 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2085 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: "
2086 "ti_locked %d", __func__, ti_locked));
2087 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2088 INP_WLOCK_ASSERT(tp->t_inpcb);
2091 * Advance th->th_seq to correspond to first data byte.
2092 * If data, trim to stay within window,
2093 * dropping FIN if necessary.
2096 if (tlen > tp->rcv_wnd) {
2097 todrop = tlen - tp->rcv_wnd;
2101 TCPSTAT_INC(tcps_rcvpackafterwin);
2102 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2104 tp->snd_wl1 = th->th_seq - 1;
2105 tp->rcv_up = th->th_seq;
2107 * Client side of transaction: already sent SYN and data.
2108 * If the remote host used T/TCP to validate the SYN,
2109 * our data will be ACK'd; if so, enter normal data segment
2110 * processing in the middle of step 5, ack processing.
2111 * Otherwise, goto step 6.
2113 if (thflags & TH_ACK)
2119 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2120 * do normal processing.
2122 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2126 break; /* continue normal processing */
2130 * States other than LISTEN or SYN_SENT.
2131 * First check the RST flag and sequence number since reset segments
2132 * are exempt from the timestamp and connection count tests. This
2133 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2134 * below which allowed reset segments in half the sequence space
2135 * to fall though and be processed (which gives forged reset
2136 * segments with a random sequence number a 50 percent chance of
2137 * killing a connection).
2138 * Then check timestamp, if present.
2139 * Then check the connection count, if present.
2140 * Then check that at least some bytes of segment are within
2141 * receive window. If segment begins before rcv_nxt,
2142 * drop leading data (and SYN); if nothing left, just ack.
2144 if (thflags & TH_RST) {
2146 * RFC5961 Section 3.2
2148 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2149 * - If RST is in window, we send challenge ACK.
2151 * Note: to take into account delayed ACKs, we should
2152 * test against last_ack_sent instead of rcv_nxt.
2153 * Note 2: we handle special case of closed window, not
2154 * covered by the RFC.
2156 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2157 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2158 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2160 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2161 KASSERT(ti_locked == TI_RLOCKED,
2162 ("%s: TH_RST ti_locked %d, th %p tp %p",
2163 __func__, ti_locked, th, tp));
2164 KASSERT(tp->t_state != TCPS_SYN_SENT,
2165 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2168 if (V_tcp_insecure_rst ||
2169 tp->last_ack_sent == th->th_seq) {
2170 TCPSTAT_INC(tcps_drops);
2171 /* Drop the connection. */
2172 switch (tp->t_state) {
2173 case TCPS_SYN_RECEIVED:
2174 so->so_error = ECONNREFUSED;
2176 case TCPS_ESTABLISHED:
2177 case TCPS_FIN_WAIT_1:
2178 case TCPS_FIN_WAIT_2:
2179 case TCPS_CLOSE_WAIT:
2180 so->so_error = ECONNRESET;
2182 tcp_state_change(tp, TCPS_CLOSED);
2188 TCPSTAT_INC(tcps_badrst);
2189 /* Send challenge ACK. */
2190 tcp_respond(tp, mtod(m, void *), th, m,
2191 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2192 tp->last_ack_sent = tp->rcv_nxt;
2200 * RFC5961 Section 4.2
2201 * Send challenge ACK for any SYN in synchronized state.
2203 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2204 tp->t_state != TCPS_SYN_RECEIVED) {
2205 KASSERT(ti_locked == TI_RLOCKED,
2206 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2207 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2209 TCPSTAT_INC(tcps_badsyn);
2210 if (V_tcp_insecure_syn &&
2211 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2212 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2213 tp = tcp_drop(tp, ECONNRESET);
2214 rstreason = BANDLIM_UNLIMITED;
2216 /* Send challenge ACK. */
2217 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2218 tp->snd_nxt, TH_ACK);
2219 tp->last_ack_sent = tp->rcv_nxt;
2226 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2227 * and it's less than ts_recent, drop it.
2229 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2230 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2232 /* Check to see if ts_recent is over 24 days old. */
2233 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2235 * Invalidate ts_recent. If this segment updates
2236 * ts_recent, the age will be reset later and ts_recent
2237 * will get a valid value. If it does not, setting
2238 * ts_recent to zero will at least satisfy the
2239 * requirement that zero be placed in the timestamp
2240 * echo reply when ts_recent isn't valid. The
2241 * age isn't reset until we get a valid ts_recent
2242 * because we don't want out-of-order segments to be
2243 * dropped when ts_recent is old.
2247 TCPSTAT_INC(tcps_rcvduppack);
2248 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2249 TCPSTAT_INC(tcps_pawsdrop);
2257 * In the SYN-RECEIVED state, validate that the packet belongs to
2258 * this connection before trimming the data to fit the receive
2259 * window. Check the sequence number versus IRS since we know
2260 * the sequence numbers haven't wrapped. This is a partial fix
2261 * for the "LAND" DoS attack.
2263 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2264 rstreason = BANDLIM_RST_OPENPORT;
2268 todrop = tp->rcv_nxt - th->th_seq;
2270 if (thflags & TH_SYN) {
2280 * Following if statement from Stevens, vol. 2, p. 960.
2283 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2285 * Any valid FIN must be to the left of the window.
2286 * At this point the FIN must be a duplicate or out
2287 * of sequence; drop it.
2292 * Send an ACK to resynchronize and drop any data.
2293 * But keep on processing for RST or ACK.
2295 tp->t_flags |= TF_ACKNOW;
2297 TCPSTAT_INC(tcps_rcvduppack);
2298 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2300 TCPSTAT_INC(tcps_rcvpartduppack);
2301 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2303 drop_hdrlen += todrop; /* drop from the top afterwards */
2304 th->th_seq += todrop;
2306 if (th->th_urp > todrop)
2307 th->th_urp -= todrop;
2315 * If new data are received on a connection after the
2316 * user processes are gone, then RST the other end.
2318 if ((so->so_state & SS_NOFDREF) &&
2319 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2320 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && "
2321 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2322 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2324 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2325 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2326 "after socket was closed, "
2327 "sending RST and removing tcpcb\n",
2328 s, __func__, tcpstates[tp->t_state], tlen);
2332 TCPSTAT_INC(tcps_rcvafterclose);
2333 rstreason = BANDLIM_UNLIMITED;
2338 * If segment ends after window, drop trailing data
2339 * (and PUSH and FIN); if nothing left, just ACK.
2341 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2343 TCPSTAT_INC(tcps_rcvpackafterwin);
2344 if (todrop >= tlen) {
2345 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2347 * If window is closed can only take segments at
2348 * window edge, and have to drop data and PUSH from
2349 * incoming segments. Continue processing, but
2350 * remember to ack. Otherwise, drop segment
2353 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2354 tp->t_flags |= TF_ACKNOW;
2355 TCPSTAT_INC(tcps_rcvwinprobe);
2359 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2362 thflags &= ~(TH_PUSH|TH_FIN);
2366 * If last ACK falls within this segment's sequence numbers,
2367 * record its timestamp.
2369 * 1) That the test incorporates suggestions from the latest
2370 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2371 * 2) That updating only on newer timestamps interferes with
2372 * our earlier PAWS tests, so this check should be solely
2373 * predicated on the sequence space of this segment.
2374 * 3) That we modify the segment boundary check to be
2375 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2376 * instead of RFC1323's
2377 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2378 * This modified check allows us to overcome RFC1323's
2379 * limitations as described in Stevens TCP/IP Illustrated
2380 * Vol. 2 p.869. In such cases, we can still calculate the
2381 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2383 if ((to.to_flags & TOF_TS) != 0 &&
2384 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2385 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2386 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2387 tp->ts_recent_age = tcp_ts_getticks();
2388 tp->ts_recent = to.to_tsval;
2392 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2393 * flag is on (half-synchronized state), then queue data for
2394 * later processing; else drop segment and return.
2396 if ((thflags & TH_ACK) == 0) {
2397 if (tp->t_state == TCPS_SYN_RECEIVED ||
2398 (tp->t_flags & TF_NEEDSYN)) {
2400 if (tp->t_state == TCPS_SYN_RECEIVED &&
2401 IS_FASTOPEN(tp->t_flags)) {
2402 tp->snd_wnd = tiwin;
2407 } else if (tp->t_flags & TF_ACKNOW)
2416 switch (tp->t_state) {
2419 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2420 * ESTABLISHED state and continue processing.
2421 * The ACK was checked above.
2423 case TCPS_SYN_RECEIVED:
2425 TCPSTAT_INC(tcps_connects);
2427 /* Do window scaling? */
2428 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2429 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2430 tp->rcv_scale = tp->request_r_scale;
2431 tp->snd_wnd = tiwin;
2435 * SYN-RECEIVED -> ESTABLISHED
2436 * SYN-RECEIVED* -> FIN-WAIT-1
2438 tp->t_starttime = ticks;
2439 if (tp->t_flags & TF_NEEDFIN) {
2440 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2441 tp->t_flags &= ~TF_NEEDFIN;
2443 tcp_state_change(tp, TCPS_ESTABLISHED);
2444 TCP_PROBE5(accept__established, NULL, tp,
2445 mtod(m, const char *), tp, th);
2447 if (tp->t_tfo_pending) {
2448 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2449 tp->t_tfo_pending = NULL;
2452 * Account for the ACK of our SYN prior to
2453 * regular ACK processing below.
2458 * TFO connections call cc_conn_init() during SYN
2459 * processing. Calling it again here for such
2460 * connections is not harmless as it would undo the
2461 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2464 if (!IS_FASTOPEN(tp->t_flags))
2467 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2470 * If segment contains data or ACK, will call tcp_reass()
2471 * later; if not, do so now to pass queued data to user.
2473 if (tlen == 0 && (thflags & TH_FIN) == 0)
2474 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2476 tp->snd_wl1 = th->th_seq - 1;
2480 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2481 * ACKs. If the ack is in the range
2482 * tp->snd_una < th->th_ack <= tp->snd_max
2483 * then advance tp->snd_una to th->th_ack and drop
2484 * data from the retransmission queue. If this ACK reflects
2485 * more up to date window information we update our window information.
2487 case TCPS_ESTABLISHED:
2488 case TCPS_FIN_WAIT_1:
2489 case TCPS_FIN_WAIT_2:
2490 case TCPS_CLOSE_WAIT:
2493 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2494 TCPSTAT_INC(tcps_rcvacktoomuch);
2497 if ((tp->t_flags & TF_SACK_PERMIT) &&
2498 ((to.to_flags & TOF_SACK) ||
2499 !TAILQ_EMPTY(&tp->snd_holes)))
2500 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2503 * Reset the value so that previous (valid) value
2504 * from the last ack with SACK doesn't get used.
2506 tp->sackhint.sacked_bytes = 0;
2509 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2510 hhook_run_tcp_est_in(tp, th, &to);
2513 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2516 maxseg = tcp_maxseg(tp);
2518 (tiwin == tp->snd_wnd ||
2519 (tp->t_flags & TF_SACK_PERMIT))) {
2521 * If this is the first time we've seen a
2522 * FIN from the remote, this is not a
2523 * duplicate and it needs to be processed
2524 * normally. This happens during a
2525 * simultaneous close.
2527 if ((thflags & TH_FIN) &&
2528 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2532 TCPSTAT_INC(tcps_rcvdupack);
2534 * If we have outstanding data (other than
2535 * a window probe), this is a completely
2536 * duplicate ack (ie, window info didn't
2537 * change and FIN isn't set),
2538 * the ack is the biggest we've
2539 * seen and we've seen exactly our rexmt
2540 * threshold of them, assume a packet
2541 * has been dropped and retransmit it.
2542 * Kludge snd_nxt & the congestion
2543 * window so we send only this one
2546 * We know we're losing at the current
2547 * window size so do congestion avoidance
2548 * (set ssthresh to half the current window
2549 * and pull our congestion window back to
2550 * the new ssthresh).
2552 * Dup acks mean that packets have left the
2553 * network (they're now cached at the receiver)
2554 * so bump cwnd by the amount in the receiver
2555 * to keep a constant cwnd packets in the
2558 * When using TCP ECN, notify the peer that
2559 * we reduced the cwnd.
2562 * Following 2 kinds of acks should not affect
2565 * 2) Acks with SACK but without any new SACK
2566 * information in them. These could result from
2567 * any anomaly in the network like a switch
2568 * duplicating packets or a possible DoS attack.
2570 if (th->th_ack != tp->snd_una ||
2571 ((tp->t_flags & TF_SACK_PERMIT) &&
2574 else if (!tcp_timer_active(tp, TT_REXMT))
2576 else if (++tp->t_dupacks > tcprexmtthresh ||
2577 IN_FASTRECOVERY(tp->t_flags)) {
2578 cc_ack_received(tp, th, nsegs,
2580 if ((tp->t_flags & TF_SACK_PERMIT) &&
2581 IN_FASTRECOVERY(tp->t_flags)) {
2585 * Compute the amount of data in flight first.
2586 * We can inject new data into the pipe iff
2587 * we have less than 1/2 the original window's
2588 * worth of data in flight.
2590 if (V_tcp_do_rfc6675_pipe)
2591 awnd = tcp_compute_pipe(tp);
2593 awnd = (tp->snd_nxt - tp->snd_fack) +
2594 tp->sackhint.sack_bytes_rexmit;
2596 if (awnd < tp->snd_ssthresh) {
2597 tp->snd_cwnd += maxseg;
2598 if (tp->snd_cwnd > tp->snd_ssthresh)
2599 tp->snd_cwnd = tp->snd_ssthresh;
2602 tp->snd_cwnd += maxseg;
2603 (void) tp->t_fb->tfb_tcp_output(tp);
2605 } else if (tp->t_dupacks == tcprexmtthresh) {
2606 tcp_seq onxt = tp->snd_nxt;
2609 * If we're doing sack, check to
2610 * see if we're already in sack
2611 * recovery. If we're not doing sack,
2612 * check to see if we're in newreno
2615 if (tp->t_flags & TF_SACK_PERMIT) {
2616 if (IN_FASTRECOVERY(tp->t_flags)) {
2621 if (SEQ_LEQ(th->th_ack,
2627 /* Congestion signal before ack. */
2628 cc_cong_signal(tp, th, CC_NDUPACK);
2629 cc_ack_received(tp, th, nsegs,
2631 tcp_timer_activate(tp, TT_REXMT, 0);
2633 if (tp->t_flags & TF_SACK_PERMIT) {
2635 tcps_sack_recovery_episode);
2636 tp->sack_newdata = tp->snd_nxt;
2637 tp->snd_cwnd = maxseg;
2638 (void) tp->t_fb->tfb_tcp_output(tp);
2641 tp->snd_nxt = th->th_ack;
2642 tp->snd_cwnd = maxseg;
2643 (void) tp->t_fb->tfb_tcp_output(tp);
2644 KASSERT(tp->snd_limited <= 2,
2645 ("%s: tp->snd_limited too big",
2647 tp->snd_cwnd = tp->snd_ssthresh +
2649 (tp->t_dupacks - tp->snd_limited);
2650 if (SEQ_GT(onxt, tp->snd_nxt))
2653 } else if (V_tcp_do_rfc3042) {
2655 * Process first and second duplicate
2656 * ACKs. Each indicates a segment
2657 * leaving the network, creating room
2658 * for more. Make sure we can send a
2659 * packet on reception of each duplicate
2660 * ACK by increasing snd_cwnd by one
2661 * segment. Restore the original
2662 * snd_cwnd after packet transmission.
2664 cc_ack_received(tp, th, nsegs,
2666 uint32_t oldcwnd = tp->snd_cwnd;
2667 tcp_seq oldsndmax = tp->snd_max;
2671 KASSERT(tp->t_dupacks == 1 ||
2673 ("%s: dupacks not 1 or 2",
2675 if (tp->t_dupacks == 1)
2676 tp->snd_limited = 0;
2678 (tp->snd_nxt - tp->snd_una) +
2679 (tp->t_dupacks - tp->snd_limited) *
2682 * Only call tcp_output when there
2683 * is new data available to be sent.
2684 * Otherwise we would send pure ACKs.
2686 SOCKBUF_LOCK(&so->so_snd);
2687 avail = sbavail(&so->so_snd) -
2688 (tp->snd_nxt - tp->snd_una);
2689 SOCKBUF_UNLOCK(&so->so_snd);
2691 (void) tp->t_fb->tfb_tcp_output(tp);
2692 sent = tp->snd_max - oldsndmax;
2693 if (sent > maxseg) {
2694 KASSERT((tp->t_dupacks == 2 &&
2695 tp->snd_limited == 0) ||
2696 (sent == maxseg + 1 &&
2697 tp->t_flags & TF_SENTFIN),
2698 ("%s: sent too much",
2700 tp->snd_limited = 2;
2701 } else if (sent > 0)
2703 tp->snd_cwnd = oldcwnd;
2710 * This ack is advancing the left edge, reset the
2715 * If this ack also has new SACK info, increment the
2716 * counter as per rfc6675.
2718 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed)
2722 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2723 ("%s: th_ack <= snd_una", __func__));
2726 * If the congestion window was inflated to account
2727 * for the other side's cached packets, retract it.
2729 if (IN_FASTRECOVERY(tp->t_flags)) {
2730 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2731 if (tp->t_flags & TF_SACK_PERMIT)
2732 tcp_sack_partialack(tp, th);
2734 tcp_newreno_partial_ack(tp, th);
2736 cc_post_recovery(tp, th);
2739 * If we reach this point, ACK is not a duplicate,
2740 * i.e., it ACKs something we sent.
2742 if (tp->t_flags & TF_NEEDSYN) {
2744 * T/TCP: Connection was half-synchronized, and our
2745 * SYN has been ACK'd (so connection is now fully
2746 * synchronized). Go to non-starred state,
2747 * increment snd_una for ACK of SYN, and check if
2748 * we can do window scaling.
2750 tp->t_flags &= ~TF_NEEDSYN;
2752 /* Do window scaling? */
2753 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2754 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2755 tp->rcv_scale = tp->request_r_scale;
2756 /* Send window already scaled. */
2761 INP_WLOCK_ASSERT(tp->t_inpcb);
2763 acked = BYTES_THIS_ACK(tp, th);
2764 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2765 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2766 tp->snd_una, th->th_ack, tp, m));
2767 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2768 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2771 * If we just performed our first retransmit, and the ACK
2772 * arrives within our recovery window, then it was a mistake
2773 * to do the retransmit in the first place. Recover our
2774 * original cwnd and ssthresh, and proceed to transmit where
2777 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2778 (int)(ticks - tp->t_badrxtwin) < 0)
2779 cc_cong_signal(tp, th, CC_RTO_ERR);
2782 * If we have a timestamp reply, update smoothed
2783 * round trip time. If no timestamp is present but
2784 * transmit timer is running and timed sequence
2785 * number was acked, update smoothed round trip time.
2786 * Since we now have an rtt measurement, cancel the
2787 * timer backoff (cf., Phil Karn's retransmit alg.).
2788 * Recompute the initial retransmit timer.
2790 * Some boxes send broken timestamp replies
2791 * during the SYN+ACK phase, ignore
2792 * timestamps of 0 or we could calculate a
2793 * huge RTT and blow up the retransmit timer.
2795 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2798 t = tcp_ts_getticks() - to.to_tsecr;
2799 if (!tp->t_rttlow || tp->t_rttlow > t)
2801 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2802 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2803 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2804 tp->t_rttlow = ticks - tp->t_rtttime;
2805 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2809 * If all outstanding data is acked, stop retransmit
2810 * timer and remember to restart (more output or persist).
2811 * If there is more data to be acked, restart retransmit
2812 * timer, using current (possibly backed-off) value.
2814 if (th->th_ack == tp->snd_max) {
2815 tcp_timer_activate(tp, TT_REXMT, 0);
2817 } else if (!tcp_timer_active(tp, TT_PERSIST))
2818 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2821 * If no data (only SYN) was ACK'd,
2822 * skip rest of ACK processing.
2828 * Let the congestion control algorithm update congestion
2829 * control related information. This typically means increasing
2830 * the congestion window.
2832 cc_ack_received(tp, th, nsegs, CC_ACK);
2834 SOCKBUF_LOCK(&so->so_snd);
2835 if (acked > sbavail(&so->so_snd)) {
2836 if (tp->snd_wnd >= sbavail(&so->so_snd))
2837 tp->snd_wnd -= sbavail(&so->so_snd);
2840 mfree = sbcut_locked(&so->so_snd,
2841 (int)sbavail(&so->so_snd));
2844 mfree = sbcut_locked(&so->so_snd, acked);
2845 if (tp->snd_wnd >= (uint32_t) acked)
2846 tp->snd_wnd -= acked;
2851 /* NB: sowwakeup_locked() does an implicit unlock. */
2852 sowwakeup_locked(so);
2854 /* Detect una wraparound. */
2855 if (!IN_RECOVERY(tp->t_flags) &&
2856 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2857 SEQ_LEQ(th->th_ack, tp->snd_recover))
2858 tp->snd_recover = th->th_ack - 1;
2859 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2860 if (IN_RECOVERY(tp->t_flags) &&
2861 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2862 EXIT_RECOVERY(tp->t_flags);
2864 tp->snd_una = th->th_ack;
2865 if (tp->t_flags & TF_SACK_PERMIT) {
2866 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2867 tp->snd_recover = tp->snd_una;
2869 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2870 tp->snd_nxt = tp->snd_una;
2872 switch (tp->t_state) {
2875 * In FIN_WAIT_1 STATE in addition to the processing
2876 * for the ESTABLISHED state if our FIN is now acknowledged
2877 * then enter FIN_WAIT_2.
2879 case TCPS_FIN_WAIT_1:
2880 if (ourfinisacked) {
2882 * If we can't receive any more
2883 * data, then closing user can proceed.
2884 * Starting the timer is contrary to the
2885 * specification, but if we don't get a FIN
2886 * we'll hang forever.
2889 * we should release the tp also, and use a
2892 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2893 soisdisconnected(so);
2894 tcp_timer_activate(tp, TT_2MSL,
2895 (tcp_fast_finwait2_recycle ?
2896 tcp_finwait2_timeout :
2899 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2904 * In CLOSING STATE in addition to the processing for
2905 * the ESTABLISHED state if the ACK acknowledges our FIN
2906 * then enter the TIME-WAIT state, otherwise ignore
2910 if (ourfinisacked) {
2911 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2913 INP_INFO_RUNLOCK(&V_tcbinfo);
2920 * In LAST_ACK, we may still be waiting for data to drain
2921 * and/or to be acked, as well as for the ack of our FIN.
2922 * If our FIN is now acknowledged, delete the TCB,
2923 * enter the closed state and return.
2926 if (ourfinisacked) {
2927 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2936 INP_WLOCK_ASSERT(tp->t_inpcb);
2939 * Update window information.
2940 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2942 if ((thflags & TH_ACK) &&
2943 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2944 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2945 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2946 /* keep track of pure window updates */
2948 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2949 TCPSTAT_INC(tcps_rcvwinupd);
2950 tp->snd_wnd = tiwin;
2951 tp->snd_wl1 = th->th_seq;
2952 tp->snd_wl2 = th->th_ack;
2953 if (tp->snd_wnd > tp->max_sndwnd)
2954 tp->max_sndwnd = tp->snd_wnd;
2959 * Process segments with URG.
2961 if ((thflags & TH_URG) && th->th_urp &&
2962 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2964 * This is a kludge, but if we receive and accept
2965 * random urgent pointers, we'll crash in
2966 * soreceive. It's hard to imagine someone
2967 * actually wanting to send this much urgent data.
2969 SOCKBUF_LOCK(&so->so_rcv);
2970 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2971 th->th_urp = 0; /* XXX */
2972 thflags &= ~TH_URG; /* XXX */
2973 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2974 goto dodata; /* XXX */
2977 * If this segment advances the known urgent pointer,
2978 * then mark the data stream. This should not happen
2979 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2980 * a FIN has been received from the remote side.
2981 * In these states we ignore the URG.
2983 * According to RFC961 (Assigned Protocols),
2984 * the urgent pointer points to the last octet
2985 * of urgent data. We continue, however,
2986 * to consider it to indicate the first octet
2987 * of data past the urgent section as the original
2988 * spec states (in one of two places).
2990 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2991 tp->rcv_up = th->th_seq + th->th_urp;
2992 so->so_oobmark = sbavail(&so->so_rcv) +
2993 (tp->rcv_up - tp->rcv_nxt) - 1;
2994 if (so->so_oobmark == 0)
2995 so->so_rcv.sb_state |= SBS_RCVATMARK;
2997 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2999 SOCKBUF_UNLOCK(&so->so_rcv);
3001 * Remove out of band data so doesn't get presented to user.
3002 * This can happen independent of advancing the URG pointer,
3003 * but if two URG's are pending at once, some out-of-band
3004 * data may creep in... ick.
3006 if (th->th_urp <= (uint32_t)tlen &&
3007 !(so->so_options & SO_OOBINLINE)) {
3008 /* hdr drop is delayed */
3009 tcp_pulloutofband(so, th, m, drop_hdrlen);
3013 * If no out of band data is expected,
3014 * pull receive urgent pointer along
3015 * with the receive window.
3017 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3018 tp->rcv_up = tp->rcv_nxt;
3021 INP_WLOCK_ASSERT(tp->t_inpcb);
3024 * Process the segment text, merging it into the TCP sequencing queue,
3025 * and arranging for acknowledgment of receipt if necessary.
3026 * This process logically involves adjusting tp->rcv_wnd as data
3027 * is presented to the user (this happens in tcp_usrreq.c,
3028 * case PRU_RCVD). If a FIN has already been received on this
3029 * connection then we just ignore the text.
3032 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3033 IS_FASTOPEN(tp->t_flags));
3035 #define tfo_syn (false)
3037 if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
3038 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3039 tcp_seq save_start = th->th_seq;
3040 m_adj(m, drop_hdrlen); /* delayed header drop */
3042 * Insert segment which includes th into TCP reassembly queue
3043 * with control block tp. Set thflags to whether reassembly now
3044 * includes a segment with FIN. This handles the common case
3045 * inline (segment is the next to be received on an established
3046 * connection, and the queue is empty), avoiding linkage into
3047 * and removal from the queue and repetition of various
3049 * Set DELACK for segments received in order, but ack
3050 * immediately when segments are out of order (so
3051 * fast retransmit can work).
3053 if (th->th_seq == tp->rcv_nxt &&
3054 LIST_EMPTY(&tp->t_segq) &&
3055 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3057 if (DELAY_ACK(tp, tlen) || tfo_syn)
3058 tp->t_flags |= TF_DELACK;
3060 tp->t_flags |= TF_ACKNOW;
3061 tp->rcv_nxt += tlen;
3062 thflags = th->th_flags & TH_FIN;
3063 TCPSTAT_INC(tcps_rcvpack);
3064 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3065 SOCKBUF_LOCK(&so->so_rcv);
3066 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3069 sbappendstream_locked(&so->so_rcv, m, 0);
3070 /* NB: sorwakeup_locked() does an implicit unlock. */
3071 sorwakeup_locked(so);
3074 * XXX: Due to the header drop above "th" is
3075 * theoretically invalid by now. Fortunately
3076 * m_adj() doesn't actually frees any mbufs
3077 * when trimming from the head.
3079 thflags = tcp_reass(tp, th, &tlen, m);
3080 tp->t_flags |= TF_ACKNOW;
3082 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
3083 tcp_update_sack_list(tp, save_start, save_start + tlen);
3086 * Note the amount of data that peer has sent into
3087 * our window, in order to estimate the sender's
3091 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3092 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3094 len = so->so_rcv.sb_hiwat;
3102 * If FIN is received ACK the FIN and let the user know
3103 * that the connection is closing.
3105 if (thflags & TH_FIN) {
3106 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3109 * If connection is half-synchronized
3110 * (ie NEEDSYN flag on) then delay ACK,
3111 * so it may be piggybacked when SYN is sent.
3112 * Otherwise, since we received a FIN then no
3113 * more input can be expected, send ACK now.
3115 if (tp->t_flags & TF_NEEDSYN)
3116 tp->t_flags |= TF_DELACK;
3118 tp->t_flags |= TF_ACKNOW;
3121 switch (tp->t_state) {
3124 * In SYN_RECEIVED and ESTABLISHED STATES
3125 * enter the CLOSE_WAIT state.
3127 case TCPS_SYN_RECEIVED:
3128 tp->t_starttime = ticks;
3130 case TCPS_ESTABLISHED:
3131 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3135 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3136 * enter the CLOSING state.
3138 case TCPS_FIN_WAIT_1:
3139 tcp_state_change(tp, TCPS_CLOSING);
3143 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3144 * starting the time-wait timer, turning off the other
3147 case TCPS_FIN_WAIT_2:
3148 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
3149 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata "
3150 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
3154 INP_INFO_RUNLOCK(&V_tcbinfo);
3158 if (ti_locked == TI_RLOCKED)
3159 INP_INFO_RUNLOCK(&V_tcbinfo);
3160 ti_locked = TI_UNLOCKED;
3163 if (so->so_options & SO_DEBUG)
3164 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3167 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
3170 * Return any desired output.
3172 if (needoutput || (tp->t_flags & TF_ACKNOW))
3173 (void) tp->t_fb->tfb_tcp_output(tp);
3176 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3177 __func__, ti_locked));
3178 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3179 INP_WLOCK_ASSERT(tp->t_inpcb);
3181 if (tp->t_flags & TF_DELACK) {
3182 tp->t_flags &= ~TF_DELACK;
3183 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3185 INP_WUNLOCK(tp->t_inpcb);
3190 * Generate an ACK dropping incoming segment if it occupies
3191 * sequence space, where the ACK reflects our state.
3193 * We can now skip the test for the RST flag since all
3194 * paths to this code happen after packets containing
3195 * RST have been dropped.
3197 * In the SYN-RECEIVED state, don't send an ACK unless the
3198 * segment we received passes the SYN-RECEIVED ACK test.
3199 * If it fails send a RST. This breaks the loop in the
3200 * "LAND" DoS attack, and also prevents an ACK storm
3201 * between two listening ports that have been sent forged
3202 * SYN segments, each with the source address of the other.
3204 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3205 (SEQ_GT(tp->snd_una, th->th_ack) ||
3206 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3207 rstreason = BANDLIM_RST_OPENPORT;
3211 if (so->so_options & SO_DEBUG)
3212 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3215 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
3216 if (ti_locked == TI_RLOCKED)
3217 INP_INFO_RUNLOCK(&V_tcbinfo);
3218 ti_locked = TI_UNLOCKED;
3220 tp->t_flags |= TF_ACKNOW;
3221 (void) tp->t_fb->tfb_tcp_output(tp);
3222 INP_WUNLOCK(tp->t_inpcb);
3227 if (ti_locked == TI_RLOCKED)
3228 INP_INFO_RUNLOCK(&V_tcbinfo);
3229 ti_locked = TI_UNLOCKED;
3232 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3233 INP_WUNLOCK(tp->t_inpcb);
3235 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3239 if (ti_locked == TI_RLOCKED) {
3240 INP_INFO_RUNLOCK(&V_tcbinfo);
3241 ti_locked = TI_UNLOCKED;
3245 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3249 * Drop space held by incoming segment and return.
3252 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3253 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3256 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
3258 INP_WUNLOCK(tp->t_inpcb);
3266 * Issue RST and make ACK acceptable to originator of segment.
3267 * The mbuf must still include the original packet header.
3271 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3272 int tlen, int rstreason)
3278 struct ip6_hdr *ip6;
3282 INP_WLOCK_ASSERT(tp->t_inpcb);
3285 /* Don't bother if destination was broadcast/multicast. */
3286 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3289 if (mtod(m, struct ip *)->ip_v == 6) {
3290 ip6 = mtod(m, struct ip6_hdr *);
3291 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3292 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3294 /* IPv6 anycast check is done at tcp6_input() */
3297 #if defined(INET) && defined(INET6)
3302 ip = mtod(m, struct ip *);
3303 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3304 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3305 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3306 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3311 /* Perform bandwidth limiting. */
3312 if (badport_bandlim(rstreason) < 0)
3315 /* tcp_respond consumes the mbuf chain. */
3316 if (th->th_flags & TH_ACK) {
3317 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3318 th->th_ack, TH_RST);
3320 if (th->th_flags & TH_SYN)
3322 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3323 (tcp_seq)0, TH_RST|TH_ACK);
3331 * Parse TCP options and place in tcpopt.
3334 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3339 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3341 if (opt == TCPOPT_EOL)
3343 if (opt == TCPOPT_NOP)
3349 if (optlen < 2 || optlen > cnt)
3354 if (optlen != TCPOLEN_MAXSEG)
3356 if (!(flags & TO_SYN))
3358 to->to_flags |= TOF_MSS;
3359 bcopy((char *)cp + 2,
3360 (char *)&to->to_mss, sizeof(to->to_mss));
3361 to->to_mss = ntohs(to->to_mss);
3364 if (optlen != TCPOLEN_WINDOW)
3366 if (!(flags & TO_SYN))
3368 to->to_flags |= TOF_SCALE;
3369 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3371 case TCPOPT_TIMESTAMP:
3372 if (optlen != TCPOLEN_TIMESTAMP)
3374 to->to_flags |= TOF_TS;
3375 bcopy((char *)cp + 2,
3376 (char *)&to->to_tsval, sizeof(to->to_tsval));
3377 to->to_tsval = ntohl(to->to_tsval);
3378 bcopy((char *)cp + 6,
3379 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3380 to->to_tsecr = ntohl(to->to_tsecr);
3382 #ifdef TCP_SIGNATURE
3384 * XXX In order to reply to a host which has set the
3385 * TCP_SIGNATURE option in its initial SYN, we have to
3386 * record the fact that the option was observed here
3387 * for the syncache code to perform the correct response.
3389 case TCPOPT_SIGNATURE:
3390 if (optlen != TCPOLEN_SIGNATURE)
3392 to->to_flags |= TOF_SIGNATURE;
3393 to->to_signature = cp + 2;
3396 case TCPOPT_SACK_PERMITTED:
3397 if (optlen != TCPOLEN_SACK_PERMITTED)
3399 if (!(flags & TO_SYN))
3403 to->to_flags |= TOF_SACKPERM;
3406 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3410 to->to_flags |= TOF_SACK;
3411 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3412 to->to_sacks = cp + 2;
3413 TCPSTAT_INC(tcps_sack_rcv_blocks);
3416 case TCPOPT_FAST_OPEN:
3417 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) &&
3418 (optlen < TCPOLEN_FAST_OPEN_MIN) &&
3419 (optlen > TCPOLEN_FAST_OPEN_MAX))
3421 if (!(flags & TO_SYN))
3423 if (!V_tcp_fastopen_enabled)
3425 to->to_flags |= TOF_FASTOPEN;
3426 to->to_tfo_len = optlen - 2;
3427 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3437 * Pull out of band byte out of a segment so
3438 * it doesn't appear in the user's data queue.
3439 * It is still reflected in the segment length for
3440 * sequencing purposes.
3443 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3446 int cnt = off + th->th_urp - 1;
3449 if (m->m_len > cnt) {
3450 char *cp = mtod(m, caddr_t) + cnt;
3451 struct tcpcb *tp = sototcpcb(so);
3453 INP_WLOCK_ASSERT(tp->t_inpcb);
3456 tp->t_oobflags |= TCPOOB_HAVEDATA;
3457 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3459 if (m->m_flags & M_PKTHDR)
3468 panic("tcp_pulloutofband");
3472 * Collect new round-trip time estimate
3473 * and update averages and current timeout.
3476 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3480 INP_WLOCK_ASSERT(tp->t_inpcb);
3482 TCPSTAT_INC(tcps_rttupdated);
3484 if (tp->t_srtt != 0) {
3486 * srtt is stored as fixed point with 5 bits after the
3487 * binary point (i.e., scaled by 8). The following magic
3488 * is equivalent to the smoothing algorithm in rfc793 with
3489 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3490 * point). Adjust rtt to origin 0.
3492 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3493 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3495 if ((tp->t_srtt += delta) <= 0)
3499 * We accumulate a smoothed rtt variance (actually, a
3500 * smoothed mean difference), then set the retransmit
3501 * timer to smoothed rtt + 4 times the smoothed variance.
3502 * rttvar is stored as fixed point with 4 bits after the
3503 * binary point (scaled by 16). The following is
3504 * equivalent to rfc793 smoothing with an alpha of .75
3505 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3506 * rfc793's wired-in beta.
3510 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3511 if ((tp->t_rttvar += delta) <= 0)
3513 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3514 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3517 * No rtt measurement yet - use the unsmoothed rtt.
3518 * Set the variance to half the rtt (so our first
3519 * retransmit happens at 3*rtt).
3521 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3522 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3523 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3529 * the retransmit should happen at rtt + 4 * rttvar.
3530 * Because of the way we do the smoothing, srtt and rttvar
3531 * will each average +1/2 tick of bias. When we compute
3532 * the retransmit timer, we want 1/2 tick of rounding and
3533 * 1 extra tick because of +-1/2 tick uncertainty in the
3534 * firing of the timer. The bias will give us exactly the
3535 * 1.5 tick we need. But, because the bias is
3536 * statistical, we have to test that we don't drop below
3537 * the minimum feasible timer (which is 2 ticks).
3539 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3540 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3543 * We received an ack for a packet that wasn't retransmitted;
3544 * it is probably safe to discard any error indications we've
3545 * received recently. This isn't quite right, but close enough
3546 * for now (a route might have failed after we sent a segment,
3547 * and the return path might not be symmetrical).
3549 tp->t_softerror = 0;
3553 * Determine a reasonable value for maxseg size.
3554 * If the route is known, check route for mtu.
3555 * If none, use an mss that can be handled on the outgoing interface
3556 * without forcing IP to fragment. If no route is found, route has no mtu,
3557 * or the destination isn't local, use a default, hopefully conservative
3558 * size (usually 512 or the default IP max size, but no more than the mtu
3559 * of the interface), as we can't discover anything about intervening
3560 * gateways or networks. We also initialize the congestion/slow start
3561 * window to be a single segment if the destination isn't local.
3562 * While looking at the routing entry, we also initialize other path-dependent
3563 * parameters from pre-set or cached values in the routing entry.
3565 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3566 * IP options, e.g. IPSEC data, since length of this data may vary, and
3567 * thus it is calculated for every segment separately in tcp_output().
3569 * NOTE that this routine is only called when we process an incoming
3570 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3571 * settings are handled in tcp_mssopt().
3574 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3575 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3578 uint32_t maxmtu = 0;
3579 struct inpcb *inp = tp->t_inpcb;
3580 struct hc_metrics_lite metrics;
3582 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3583 size_t min_protoh = isipv6 ?
3584 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3585 sizeof (struct tcpiphdr);
3587 const size_t min_protoh = sizeof(struct tcpiphdr);
3590 INP_WLOCK_ASSERT(tp->t_inpcb);
3592 if (mtuoffer != -1) {
3593 KASSERT(offer == -1, ("%s: conflict", __func__));
3594 offer = mtuoffer - min_protoh;
3600 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3601 tp->t_maxseg = V_tcp_v6mssdflt;
3604 #if defined(INET) && defined(INET6)
3609 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3610 tp->t_maxseg = V_tcp_mssdflt;
3615 * No route to sender, stay with default mss and return.
3619 * In case we return early we need to initialize metrics
3620 * to a defined state as tcp_hc_get() would do for us
3621 * if there was no cache hit.
3623 if (metricptr != NULL)
3624 bzero(metricptr, sizeof(struct hc_metrics_lite));
3628 /* What have we got? */
3632 * Offer == 0 means that there was no MSS on the SYN
3633 * segment, in this case we use tcp_mssdflt as
3634 * already assigned to t_maxseg above.
3636 offer = tp->t_maxseg;
3641 * Offer == -1 means that we didn't receive SYN yet.
3647 * Prevent DoS attack with too small MSS. Round up
3648 * to at least minmss.
3650 offer = max(offer, V_tcp_minmss);
3654 * rmx information is now retrieved from tcp_hostcache.
3656 tcp_hc_get(&inp->inp_inc, &metrics);
3657 if (metricptr != NULL)
3658 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3661 * If there's a discovered mtu in tcp hostcache, use it.
3662 * Else, use the link mtu.
3664 if (metrics.rmx_mtu)
3665 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3669 mss = maxmtu - min_protoh;
3670 if (!V_path_mtu_discovery &&
3671 !in6_localaddr(&inp->in6p_faddr))
3672 mss = min(mss, V_tcp_v6mssdflt);
3675 #if defined(INET) && defined(INET6)
3680 mss = maxmtu - min_protoh;
3681 if (!V_path_mtu_discovery &&
3682 !in_localaddr(inp->inp_faddr))
3683 mss = min(mss, V_tcp_mssdflt);
3687 * XXX - The above conditional (mss = maxmtu - min_protoh)
3688 * probably violates the TCP spec.
3689 * The problem is that, since we don't know the
3690 * other end's MSS, we are supposed to use a conservative
3691 * default. But, if we do that, then MTU discovery will
3692 * never actually take place, because the conservative
3693 * default is much less than the MTUs typically seen
3694 * on the Internet today. For the moment, we'll sweep
3695 * this under the carpet.
3697 * The conservative default might not actually be a problem
3698 * if the only case this occurs is when sending an initial
3699 * SYN with options and data to a host we've never talked
3700 * to before. Then, they will reply with an MSS value which
3701 * will get recorded and the new parameters should get
3702 * recomputed. For Further Study.
3705 mss = min(mss, offer);
3708 * Sanity check: make sure that maxseg will be large
3709 * enough to allow some data on segments even if the
3710 * all the option space is used (40bytes). Otherwise
3711 * funny things may happen in tcp_output.
3713 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3721 tcp_mss(struct tcpcb *tp, int offer)
3727 struct hc_metrics_lite metrics;
3728 struct tcp_ifcap cap;
3730 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3732 bzero(&cap, sizeof(cap));
3733 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3739 * If there's a pipesize, change the socket buffer to that size,
3740 * don't change if sb_hiwat is different than default (then it
3741 * has been changed on purpose with setsockopt).
3742 * Make the socket buffers an integral number of mss units;
3743 * if the mss is larger than the socket buffer, decrease the mss.
3745 so = inp->inp_socket;
3746 SOCKBUF_LOCK(&so->so_snd);
3747 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3748 bufsize = metrics.rmx_sendpipe;
3750 bufsize = so->so_snd.sb_hiwat;
3754 bufsize = roundup(bufsize, mss);
3755 if (bufsize > sb_max)
3757 if (bufsize > so->so_snd.sb_hiwat)
3758 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3760 SOCKBUF_UNLOCK(&so->so_snd);
3763 SOCKBUF_LOCK(&so->so_rcv);
3764 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3765 bufsize = metrics.rmx_recvpipe;
3767 bufsize = so->so_rcv.sb_hiwat;
3768 if (bufsize > mss) {
3769 bufsize = roundup(bufsize, mss);
3770 if (bufsize > sb_max)
3772 if (bufsize > so->so_rcv.sb_hiwat)
3773 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3775 SOCKBUF_UNLOCK(&so->so_rcv);
3777 /* Check the interface for TSO capabilities. */
3778 if (cap.ifcap & CSUM_TSO) {
3779 tp->t_flags |= TF_TSO;
3780 tp->t_tsomax = cap.tsomax;
3781 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3782 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3787 * Determine the MSS option to send on an outgoing SYN.
3790 tcp_mssopt(struct in_conninfo *inc)
3793 uint32_t thcmtu = 0;
3794 uint32_t maxmtu = 0;
3797 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3800 if (inc->inc_flags & INC_ISIPV6) {
3801 mss = V_tcp_v6mssdflt;
3802 maxmtu = tcp_maxmtu6(inc, NULL);
3803 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3806 #if defined(INET) && defined(INET6)
3811 mss = V_tcp_mssdflt;
3812 maxmtu = tcp_maxmtu(inc, NULL);
3813 min_protoh = sizeof(struct tcpiphdr);
3816 #if defined(INET6) || defined(INET)
3817 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3820 if (maxmtu && thcmtu)
3821 mss = min(maxmtu, thcmtu) - min_protoh;
3822 else if (maxmtu || thcmtu)
3823 mss = max(maxmtu, thcmtu) - min_protoh;
3830 * On a partial ack arrives, force the retransmission of the
3831 * next unacknowledged segment. Do not clear tp->t_dupacks.
3832 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3836 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3838 tcp_seq onxt = tp->snd_nxt;
3839 uint32_t ocwnd = tp->snd_cwnd;
3840 u_int maxseg = tcp_maxseg(tp);
3842 INP_WLOCK_ASSERT(tp->t_inpcb);
3844 tcp_timer_activate(tp, TT_REXMT, 0);
3846 tp->snd_nxt = th->th_ack;
3848 * Set snd_cwnd to one segment beyond acknowledged offset.
3849 * (tp->snd_una has not yet been updated when this function is called.)
3851 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3852 tp->t_flags |= TF_ACKNOW;
3853 (void) tp->t_fb->tfb_tcp_output(tp);
3854 tp->snd_cwnd = ocwnd;
3855 if (SEQ_GT(onxt, tp->snd_nxt))
3858 * Partial window deflation. Relies on fact that tp->snd_una
3861 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3862 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3865 tp->snd_cwnd += maxseg;
3869 tcp_compute_pipe(struct tcpcb *tp)
3871 return (tp->snd_max - tp->snd_una +
3872 tp->sackhint.sack_bytes_rexmit -
3873 tp->sackhint.sacked_bytes);