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 * 3. 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>
123 #include <netipsec/ipsec_support.h>
125 #include <machine/in_cksum.h>
127 #include <security/mac/mac_framework.h>
129 const int tcprexmtthresh = 3;
131 int tcp_log_in_vain = 0;
132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
134 "Log all incoming TCP segments to closed ports");
136 VNET_DEFINE(int, blackhole) = 0;
137 #define V_blackhole VNET(blackhole)
138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
139 &VNET_NAME(blackhole), 0,
140 "Do not send RST on segments to closed ports");
142 VNET_DEFINE(int, tcp_delack_enabled) = 1;
143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
144 &VNET_NAME(tcp_delack_enabled), 0,
145 "Delay ACK to try and piggyback it onto a data packet");
147 VNET_DEFINE(int, drop_synfin) = 0;
148 #define V_drop_synfin VNET(drop_synfin)
149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
150 &VNET_NAME(drop_synfin), 0,
151 "Drop TCP packets with SYN+FIN set");
153 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
155 &VNET_NAME(tcp_do_rfc6675_pipe), 0,
156 "Use calculated pipe/in-flight bytes per RFC 6675");
158 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
159 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
161 &VNET_NAME(tcp_do_rfc3042), 0,
162 "Enable RFC 3042 (Limited Transmit)");
164 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
166 &VNET_NAME(tcp_do_rfc3390), 0,
167 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
169 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
171 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
172 "Slow-start flight size (initial congestion window) in number of segments");
174 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
176 &VNET_NAME(tcp_do_rfc3465), 0,
177 "Enable RFC 3465 (Appropriate Byte Counting)");
179 VNET_DEFINE(int, tcp_abc_l_var) = 2;
180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
181 &VNET_NAME(tcp_abc_l_var), 2,
182 "Cap the max cwnd increment during slow-start to this number of segments");
184 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
186 VNET_DEFINE(int, tcp_do_ecn) = 2;
187 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
188 &VNET_NAME(tcp_do_ecn), 0,
191 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
192 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
193 &VNET_NAME(tcp_ecn_maxretries), 0,
194 "Max retries before giving up on ECN");
196 VNET_DEFINE(int, tcp_insecure_syn) = 0;
197 #define V_tcp_insecure_syn VNET(tcp_insecure_syn)
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
199 &VNET_NAME(tcp_insecure_syn), 0,
200 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
202 VNET_DEFINE(int, tcp_insecure_rst) = 0;
203 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
205 &VNET_NAME(tcp_insecure_rst), 0,
206 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
208 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
209 #define V_tcp_recvspace VNET(tcp_recvspace)
210 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
211 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
213 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
214 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
216 &VNET_NAME(tcp_do_autorcvbuf), 0,
217 "Enable automatic receive buffer sizing");
219 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
220 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW,
222 &VNET_NAME(tcp_autorcvbuf_inc), 0,
223 "Incrementor step size of automatic receive buffer");
225 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
226 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
228 &VNET_NAME(tcp_autorcvbuf_max), 0,
229 "Max size of automatic receive buffer");
231 VNET_DEFINE(struct inpcbhead, tcb);
232 #define tcb6 tcb /* for KAME src sync over BSD*'s */
233 VNET_DEFINE(struct inpcbinfo, tcbinfo);
236 * TCP statistics are stored in an array of counter(9)s, which size matches
237 * size of struct tcpstat. TCP running connection count is a regular array.
239 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
240 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
241 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
242 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
243 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
244 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
245 "TCP connection counts by TCP state");
248 tcp_vnet_init(const void *unused)
251 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
252 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
254 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
255 tcp_vnet_init, NULL);
259 tcp_vnet_uninit(const void *unused)
262 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
263 VNET_PCPUSTAT_FREE(tcpstat);
265 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
266 tcp_vnet_uninit, NULL);
270 * Kernel module interface for updating tcpstat. The argument is an index
271 * into tcpstat treated as an array.
274 kmod_tcpstat_inc(int statnum)
277 counter_u64_add(VNET(tcpstat)[statnum], 1);
282 * Wrapper for the TCP established input helper hook.
285 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
287 struct tcp_hhook_data hhook_data;
289 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
294 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
301 * CC wrapper hook functions
304 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
307 INP_WLOCK_ASSERT(tp->t_inpcb);
309 tp->ccv->nsegs = nsegs;
310 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
311 if (tp->snd_cwnd <= tp->snd_wnd)
312 tp->ccv->flags |= CCF_CWND_LIMITED;
314 tp->ccv->flags &= ~CCF_CWND_LIMITED;
316 if (type == CC_ACK) {
317 if (tp->snd_cwnd > tp->snd_ssthresh) {
318 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
319 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp));
320 if (tp->t_bytes_acked >= tp->snd_cwnd) {
321 tp->t_bytes_acked -= tp->snd_cwnd;
322 tp->ccv->flags |= CCF_ABC_SENTAWND;
325 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
326 tp->t_bytes_acked = 0;
330 if (CC_ALGO(tp)->ack_received != NULL) {
331 /* XXXLAS: Find a way to live without this */
332 tp->ccv->curack = th->th_ack;
333 CC_ALGO(tp)->ack_received(tp->ccv, type);
338 cc_conn_init(struct tcpcb *tp)
340 struct hc_metrics_lite metrics;
341 struct inpcb *inp = tp->t_inpcb;
345 INP_WLOCK_ASSERT(tp->t_inpcb);
347 tcp_hc_get(&inp->inp_inc, &metrics);
348 maxseg = tcp_maxseg(tp);
350 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
352 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
353 TCPSTAT_INC(tcps_usedrtt);
354 if (metrics.rmx_rttvar) {
355 tp->t_rttvar = metrics.rmx_rttvar;
356 TCPSTAT_INC(tcps_usedrttvar);
358 /* default variation is +- 1 rtt */
360 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
362 TCPT_RANGESET(tp->t_rxtcur,
363 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
364 tp->t_rttmin, TCPTV_REXMTMAX);
366 if (metrics.rmx_ssthresh) {
368 * There's some sort of gateway or interface
369 * buffer limit on the path. Use this to set
370 * the slow start threshold, but set the
371 * threshold to no less than 2*mss.
373 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
374 TCPSTAT_INC(tcps_usedssthresh);
378 * Set the initial slow-start flight size.
380 * RFC5681 Section 3.1 specifies the default conservative values.
381 * RFC3390 specifies slightly more aggressive values.
382 * RFC6928 increases it to ten segments.
383 * Support for user specified value for initial flight size.
385 * If a SYN or SYN/ACK was lost and retransmitted, we have to
386 * reduce the initial CWND to one segment as congestion is likely
387 * requiring us to be cautious.
389 if (tp->snd_cwnd == 1)
390 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
391 else if (V_tcp_initcwnd_segments)
392 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg,
393 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
394 else if (V_tcp_do_rfc3390)
395 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380));
397 /* Per RFC5681 Section 3.1 */
399 tp->snd_cwnd = 2 * maxseg;
400 else if (maxseg > 1095)
401 tp->snd_cwnd = 3 * maxseg;
403 tp->snd_cwnd = 4 * maxseg;
406 if (CC_ALGO(tp)->conn_init != NULL)
407 CC_ALGO(tp)->conn_init(tp->ccv);
411 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
415 INP_WLOCK_ASSERT(tp->t_inpcb);
419 if (!IN_FASTRECOVERY(tp->t_flags)) {
420 tp->snd_recover = tp->snd_max;
421 if (tp->t_flags & TF_ECN_PERMIT)
422 tp->t_flags |= TF_ECN_SND_CWR;
426 if (!IN_CONGRECOVERY(tp->t_flags)) {
427 TCPSTAT_INC(tcps_ecn_rcwnd);
428 tp->snd_recover = tp->snd_max;
429 if (tp->t_flags & TF_ECN_PERMIT)
430 tp->t_flags |= TF_ECN_SND_CWR;
434 maxseg = tcp_maxseg(tp);
436 tp->t_bytes_acked = 0;
437 EXIT_RECOVERY(tp->t_flags);
438 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
440 tp->snd_cwnd = maxseg;
443 TCPSTAT_INC(tcps_sndrexmitbad);
444 /* RTO was unnecessary, so reset everything. */
445 tp->snd_cwnd = tp->snd_cwnd_prev;
446 tp->snd_ssthresh = tp->snd_ssthresh_prev;
447 tp->snd_recover = tp->snd_recover_prev;
448 if (tp->t_flags & TF_WASFRECOVERY)
449 ENTER_FASTRECOVERY(tp->t_flags);
450 if (tp->t_flags & TF_WASCRECOVERY)
451 ENTER_CONGRECOVERY(tp->t_flags);
452 tp->snd_nxt = tp->snd_max;
453 tp->t_flags &= ~TF_PREVVALID;
458 if (CC_ALGO(tp)->cong_signal != NULL) {
460 tp->ccv->curack = th->th_ack;
461 CC_ALGO(tp)->cong_signal(tp->ccv, type);
466 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
468 INP_WLOCK_ASSERT(tp->t_inpcb);
470 /* XXXLAS: KASSERT that we're in recovery? */
472 if (CC_ALGO(tp)->post_recovery != NULL) {
473 tp->ccv->curack = th->th_ack;
474 CC_ALGO(tp)->post_recovery(tp->ccv);
476 /* XXXLAS: EXIT_RECOVERY ? */
477 tp->t_bytes_acked = 0;
481 * Indicate whether this ack should be delayed. We can delay the ack if
482 * following conditions are met:
483 * - There is no delayed ack timer in progress.
484 * - Our last ack wasn't a 0-sized window. We never want to delay
485 * the ack that opens up a 0-sized window.
486 * - LRO wasn't used for this segment. We make sure by checking that the
487 * segment size is not larger than the MSS.
489 #define DELAY_ACK(tp, tlen) \
490 ((!tcp_timer_active(tp, TT_DELACK) && \
491 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
492 (tlen <= tp->t_maxseg) && \
493 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
496 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
498 INP_WLOCK_ASSERT(tp->t_inpcb);
500 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
501 switch (iptos & IPTOS_ECN_MASK) {
503 tp->ccv->flags |= CCF_IPHDR_CE;
506 tp->ccv->flags &= ~CCF_IPHDR_CE;
509 tp->ccv->flags &= ~CCF_IPHDR_CE;
513 if (th->th_flags & TH_CWR)
514 tp->ccv->flags |= CCF_TCPHDR_CWR;
516 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
518 if (tp->t_flags & TF_DELACK)
519 tp->ccv->flags |= CCF_DELACK;
521 tp->ccv->flags &= ~CCF_DELACK;
523 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
525 if (tp->ccv->flags & CCF_ACKNOW)
526 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
531 * TCP input handling is split into multiple parts:
532 * tcp6_input is a thin wrapper around tcp_input for the extended
533 * ip6_protox[] call format in ip6_input
534 * tcp_input handles primary segment validation, inpcb lookup and
535 * SYN processing on listen sockets
536 * tcp_do_segment processes the ACK and text of the segment for
537 * establishing, established and closing connections
541 tcp6_input(struct mbuf **mp, int *offp, int proto)
543 struct mbuf *m = *mp;
544 struct in6_ifaddr *ia6;
547 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
550 * draft-itojun-ipv6-tcp-to-anycast
551 * better place to put this in?
553 ip6 = mtod(m, struct ip6_hdr *);
554 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
555 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
558 ifa_free(&ia6->ia_ifa);
559 ip6 = mtod(m, struct ip6_hdr *);
560 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
561 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
562 return (IPPROTO_DONE);
565 ifa_free(&ia6->ia_ifa);
567 return (tcp_input(mp, offp, proto));
572 tcp_input(struct mbuf **mp, int *offp, int proto)
574 struct mbuf *m = *mp;
575 struct tcphdr *th = NULL;
576 struct ip *ip = NULL;
577 struct inpcb *inp = NULL;
578 struct tcpcb *tp = NULL;
579 struct socket *so = NULL;
589 int rstreason = 0; /* For badport_bandlim accounting purposes */
591 struct m_tag *fwd_tag = NULL;
593 struct ip6_hdr *ip6 = NULL;
596 const void *ip6 = NULL;
598 struct tcpopt to; /* options in this segment */
599 char *s = NULL; /* address and port logging */
603 * The size of tcp_saveipgen must be the size of the max ip header,
606 u_char tcp_saveipgen[IP6_HDR_LEN];
607 struct tcphdr tcp_savetcp;
612 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
619 TCPSTAT_INC(tcps_rcvtotal);
623 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
625 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
626 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
628 TCPSTAT_INC(tcps_rcvshort);
629 return (IPPROTO_DONE);
633 ip6 = mtod(m, struct ip6_hdr *);
634 th = (struct tcphdr *)((caddr_t)ip6 + off0);
635 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
636 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
637 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
638 th->th_sum = m->m_pkthdr.csum_data;
640 th->th_sum = in6_cksum_pseudo(ip6, tlen,
641 IPPROTO_TCP, m->m_pkthdr.csum_data);
642 th->th_sum ^= 0xffff;
644 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
646 TCPSTAT_INC(tcps_rcvbadsum);
651 * Be proactive about unspecified IPv6 address in source.
652 * As we use all-zero to indicate unbounded/unconnected pcb,
653 * unspecified IPv6 address can be used to confuse us.
655 * Note that packets with unspecified IPv6 destination is
656 * already dropped in ip6_input.
658 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
662 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
665 #if defined(INET) && defined(INET6)
671 * Get IP and TCP header together in first mbuf.
672 * Note: IP leaves IP header in first mbuf.
674 if (off0 > sizeof (struct ip)) {
676 off0 = sizeof(struct ip);
678 if (m->m_len < sizeof (struct tcpiphdr)) {
679 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
681 TCPSTAT_INC(tcps_rcvshort);
682 return (IPPROTO_DONE);
685 ip = mtod(m, struct ip *);
686 th = (struct tcphdr *)((caddr_t)ip + off0);
687 tlen = ntohs(ip->ip_len) - off0;
690 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
691 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
692 th->th_sum = m->m_pkthdr.csum_data;
694 th->th_sum = in_pseudo(ip->ip_src.s_addr,
696 htonl(m->m_pkthdr.csum_data + tlen +
698 th->th_sum ^= 0xffff;
700 struct ipovly *ipov = (struct ipovly *)ip;
703 * Checksum extended TCP header and data.
706 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
707 ipov->ih_len = htons(tlen);
708 th->th_sum = in_cksum(m, len);
709 /* Reset length for SDT probes. */
710 ip->ip_len = htons(len);
713 /* Re-initialization for later version check */
714 ip->ip_v = IPVERSION;
718 TCPSTAT_INC(tcps_rcvbadsum);
725 * Check that TCP offset makes sense,
726 * pull out TCP options and adjust length. XXX
728 off = th->th_off << 2;
729 if (off < sizeof (struct tcphdr) || off > tlen) {
730 TCPSTAT_INC(tcps_rcvbadoff);
733 tlen -= off; /* tlen is used instead of ti->ti_len */
734 if (off > sizeof (struct tcphdr)) {
737 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
738 ip6 = mtod(m, struct ip6_hdr *);
739 th = (struct tcphdr *)((caddr_t)ip6 + off0);
742 #if defined(INET) && defined(INET6)
747 if (m->m_len < sizeof(struct ip) + off) {
748 if ((m = m_pullup(m, sizeof (struct ip) + off))
750 TCPSTAT_INC(tcps_rcvshort);
751 return (IPPROTO_DONE);
753 ip = mtod(m, struct ip *);
754 th = (struct tcphdr *)((caddr_t)ip + off0);
758 optlen = off - sizeof (struct tcphdr);
759 optp = (u_char *)(th + 1);
761 thflags = th->th_flags;
764 * Convert TCP protocol specific fields to host format.
766 tcp_fields_to_host(th);
769 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
771 drop_hdrlen = off0 + off;
774 * Locate pcb for segment; if we're likely to add or remove a
775 * connection then first acquire pcbinfo lock. There are three cases
776 * where we might discover later we need a write lock despite the
777 * flags: ACKs moving a connection out of the syncache, ACKs for a
778 * connection in TIMEWAIT and SYNs not targeting a listening socket.
780 if ((thflags & (TH_FIN | TH_RST)) != 0) {
781 INP_INFO_RLOCK(&V_tcbinfo);
782 ti_locked = TI_RLOCKED;
784 ti_locked = TI_UNLOCKED;
787 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
791 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
793 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
796 #if defined(INET) && !defined(INET6)
797 (m->m_flags & M_IP_NEXTHOP)
800 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
804 if (ti_locked == TI_RLOCKED) {
805 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
807 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
811 if (isipv6 && fwd_tag != NULL) {
812 struct sockaddr_in6 *next_hop6;
814 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
816 * Transparently forwarded. Pretend to be the destination.
817 * Already got one like this?
819 inp = in6_pcblookup_mbuf(&V_tcbinfo,
820 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
821 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
824 * It's new. Try to find the ambushing socket.
825 * Because we've rewritten the destination address,
826 * any hardware-generated hash is ignored.
828 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
829 th->th_sport, &next_hop6->sin6_addr,
830 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
831 th->th_dport, INPLOOKUP_WILDCARD |
832 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
835 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
836 th->th_sport, &ip6->ip6_dst, th->th_dport,
837 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
838 m->m_pkthdr.rcvif, m);
841 #if defined(INET6) && defined(INET)
845 if (fwd_tag != NULL) {
846 struct sockaddr_in *next_hop;
848 next_hop = (struct sockaddr_in *)(fwd_tag+1);
850 * Transparently forwarded. Pretend to be the destination.
851 * already got one like this?
853 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
854 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
855 m->m_pkthdr.rcvif, m);
858 * It's new. Try to find the ambushing socket.
859 * Because we've rewritten the destination address,
860 * any hardware-generated hash is ignored.
862 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
863 th->th_sport, next_hop->sin_addr,
864 next_hop->sin_port ? ntohs(next_hop->sin_port) :
865 th->th_dport, INPLOOKUP_WILDCARD |
866 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
869 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
870 th->th_sport, ip->ip_dst, th->th_dport,
871 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
872 m->m_pkthdr.rcvif, m);
876 * If the INPCB does not exist then all data in the incoming
877 * segment is discarded and an appropriate RST is sent back.
878 * XXX MRT Send RST using which routing table?
882 * Log communication attempts to ports that are not
885 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
886 tcp_log_in_vain == 2) {
887 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
888 log(LOG_INFO, "%s; %s: Connection attempt "
889 "to closed port\n", s, __func__);
892 * When blackholing do not respond with a RST but
893 * completely ignore the segment and drop it.
895 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
899 rstreason = BANDLIM_RST_CLOSEDPORT;
902 INP_WLOCK_ASSERT(inp);
904 * While waiting for inp lock during the lookup, another thread
905 * can have dropped the inpcb, in which case we need to loop back
906 * and try to find a new inpcb to deliver to.
908 if (inp->inp_flags & INP_DROPPED) {
913 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
914 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
915 ((inp->inp_socket == NULL) ||
916 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
917 inp->inp_flowid = m->m_pkthdr.flowid;
918 inp->inp_flowtype = M_HASHTYPE_GET(m);
920 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
922 if (isipv6 && IPSEC_ENABLED(ipv6) &&
923 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
931 if (IPSEC_ENABLED(ipv4) &&
932 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
939 * Check the minimum TTL for socket.
941 if (inp->inp_ip_minttl != 0) {
944 if (inp->inp_ip_minttl > ip6->ip6_hlim)
948 if (inp->inp_ip_minttl > ip->ip_ttl)
953 * A previous connection in TIMEWAIT state is supposed to catch stray
954 * or duplicate segments arriving late. If this segment was a
955 * legitimate new connection attempt, the old INPCB gets removed and
956 * we can try again to find a listening socket.
958 * At this point, due to earlier optimism, we may hold only an inpcb
959 * lock, and not the inpcbinfo write lock. If so, we need to try to
960 * acquire it, or if that fails, acquire a reference on the inpcb,
961 * drop all locks, acquire a global write lock, and then re-acquire
962 * the inpcb lock. We may at that point discover that another thread
963 * has tried to free the inpcb, in which case we need to loop back
964 * and try to find a new inpcb to deliver to.
966 * XXXRW: It may be time to rethink timewait locking.
969 if (inp->inp_flags & INP_TIMEWAIT) {
970 if (ti_locked == TI_UNLOCKED) {
971 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
974 INP_INFO_RLOCK(&V_tcbinfo);
975 ti_locked = TI_RLOCKED;
977 if (in_pcbrele_wlocked(inp)) {
980 } else if (inp->inp_flags & INP_DROPPED) {
986 ti_locked = TI_RLOCKED;
988 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
990 if (thflags & TH_SYN)
991 tcp_dooptions(&to, optp, optlen, TO_SYN);
993 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
995 if (tcp_twcheck(inp, &to, th, m, tlen))
997 INP_INFO_RUNLOCK(&V_tcbinfo);
998 return (IPPROTO_DONE);
1001 * The TCPCB may no longer exist if the connection is winding
1002 * down or it is in the CLOSED state. Either way we drop the
1003 * segment and send an appropriate response.
1005 tp = intotcpcb(inp);
1006 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1007 rstreason = BANDLIM_RST_CLOSEDPORT;
1012 if (tp->t_flags & TF_TOE) {
1013 tcp_offload_input(tp, m);
1014 m = NULL; /* consumed by the TOE driver */
1020 * We've identified a valid inpcb, but it could be that we need an
1021 * inpcbinfo write lock but don't hold it. In this case, attempt to
1022 * acquire using the same strategy as the TIMEWAIT case above. If we
1023 * relock, we have to jump back to 'relocked' as the connection might
1024 * now be in TIMEWAIT.
1027 if ((thflags & (TH_FIN | TH_RST)) != 0)
1028 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1030 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) ||
1031 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) &&
1032 !IS_FASTOPEN(tp->t_flags)))) {
1033 if (ti_locked == TI_UNLOCKED) {
1034 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
1037 INP_INFO_RLOCK(&V_tcbinfo);
1038 ti_locked = TI_RLOCKED;
1040 if (in_pcbrele_wlocked(inp)) {
1043 } else if (inp->inp_flags & INP_DROPPED) {
1050 ti_locked = TI_RLOCKED;
1052 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1056 INP_WLOCK_ASSERT(inp);
1057 if (mac_inpcb_check_deliver(inp, m))
1060 so = inp->inp_socket;
1061 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1063 if (so->so_options & SO_DEBUG) {
1064 ostate = tp->t_state;
1067 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1070 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1073 #endif /* TCPDEBUG */
1075 * When the socket is accepting connections (the INPCB is in LISTEN
1076 * state) we look into the SYN cache if this is a new connection
1077 * attempt or the completion of a previous one.
1079 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1080 ("%s: so accepting but tp %p not listening", __func__, tp));
1081 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1082 struct in_conninfo inc;
1084 bzero(&inc, sizeof(inc));
1087 inc.inc_flags |= INC_ISIPV6;
1088 inc.inc6_faddr = ip6->ip6_src;
1089 inc.inc6_laddr = ip6->ip6_dst;
1093 inc.inc_faddr = ip->ip_src;
1094 inc.inc_laddr = ip->ip_dst;
1096 inc.inc_fport = th->th_sport;
1097 inc.inc_lport = th->th_dport;
1098 inc.inc_fibnum = so->so_fibnum;
1101 * Check for an existing connection attempt in syncache if
1102 * the flag is only ACK. A successful lookup creates a new
1103 * socket appended to the listen queue in SYN_RECEIVED state.
1105 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1107 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1109 * Parse the TCP options here because
1110 * syncookies need access to the reflected
1113 tcp_dooptions(&to, optp, optlen, 0);
1115 * NB: syncache_expand() doesn't unlock
1116 * inp and tcpinfo locks.
1118 rstreason = syncache_expand(&inc, &to, th, &so, m);
1119 if (rstreason < 0) {
1121 * A failing TCP MD5 signature comparison
1122 * must result in the segment being dropped
1123 * and must not produce any response back
1127 } else if (rstreason == 0) {
1129 * No syncache entry or ACK was not
1130 * for our SYN/ACK. Send a RST.
1131 * NB: syncache did its own logging
1132 * of the failure cause.
1134 rstreason = BANDLIM_RST_OPENPORT;
1142 * We completed the 3-way handshake
1143 * but could not allocate a socket
1144 * either due to memory shortage,
1145 * listen queue length limits or
1146 * global socket limits. Send RST
1147 * or wait and have the remote end
1148 * retransmit the ACK for another
1151 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1152 log(LOG_DEBUG, "%s; %s: Listen socket: "
1153 "Socket allocation failed due to "
1154 "limits or memory shortage, %s\n",
1156 V_tcp_sc_rst_sock_fail ?
1157 "sending RST" : "try again");
1158 if (V_tcp_sc_rst_sock_fail) {
1159 rstreason = BANDLIM_UNLIMITED;
1165 * Socket is created in state SYN_RECEIVED.
1166 * Unlock the listen socket, lock the newly
1167 * created socket and update the tp variable.
1169 INP_WUNLOCK(inp); /* listen socket */
1170 inp = sotoinpcb(so);
1172 * New connection inpcb is already locked by
1173 * syncache_expand().
1175 INP_WLOCK_ASSERT(inp);
1176 tp = intotcpcb(inp);
1177 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1178 ("%s: ", __func__));
1180 * Process the segment and the data it
1181 * contains. tcp_do_segment() consumes
1182 * the mbuf chain and unlocks the inpcb.
1184 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1186 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1187 return (IPPROTO_DONE);
1190 * Segment flag validation for new connection attempts:
1192 * Our (SYN|ACK) response was rejected.
1193 * Check with syncache and remove entry to prevent
1196 * NB: syncache_chkrst does its own logging of failure
1199 if (thflags & TH_RST) {
1200 syncache_chkrst(&inc, th);
1204 * We can't do anything without SYN.
1206 if ((thflags & TH_SYN) == 0) {
1207 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1208 log(LOG_DEBUG, "%s; %s: Listen socket: "
1209 "SYN is missing, segment ignored\n",
1211 TCPSTAT_INC(tcps_badsyn);
1215 * (SYN|ACK) is bogus on a listen socket.
1217 if (thflags & TH_ACK) {
1218 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1219 log(LOG_DEBUG, "%s; %s: Listen socket: "
1220 "SYN|ACK invalid, segment rejected\n",
1222 syncache_badack(&inc); /* XXX: Not needed! */
1223 TCPSTAT_INC(tcps_badsyn);
1224 rstreason = BANDLIM_RST_OPENPORT;
1228 * If the drop_synfin option is enabled, drop all
1229 * segments with both the SYN and FIN bits set.
1230 * This prevents e.g. nmap from identifying the
1232 * XXX: Poor reasoning. nmap has other methods
1233 * and is constantly refining its stack detection
1235 * XXX: This is a violation of the TCP specification
1236 * and was used by RFC1644.
1238 if ((thflags & TH_FIN) && V_drop_synfin) {
1239 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1240 log(LOG_DEBUG, "%s; %s: Listen socket: "
1241 "SYN|FIN segment ignored (based on "
1242 "sysctl setting)\n", s, __func__);
1243 TCPSTAT_INC(tcps_badsyn);
1247 * Segment's flags are (SYN) or (SYN|FIN).
1249 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1250 * as they do not affect the state of the TCP FSM.
1251 * The data pointed to by TH_URG and th_urp is ignored.
1253 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1254 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1255 KASSERT(thflags & (TH_SYN),
1256 ("%s: Listen socket: TH_SYN not set", __func__));
1259 * If deprecated address is forbidden,
1260 * we do not accept SYN to deprecated interface
1261 * address to prevent any new inbound connection from
1262 * getting established.
1263 * When we do not accept SYN, we send a TCP RST,
1264 * with deprecated source address (instead of dropping
1265 * it). We compromise it as it is much better for peer
1266 * to send a RST, and RST will be the final packet
1269 * If we do not forbid deprecated addresses, we accept
1270 * the SYN packet. RFC2462 does not suggest dropping
1272 * If we decipher RFC2462 5.5.4, it says like this:
1273 * 1. use of deprecated addr with existing
1274 * communication is okay - "SHOULD continue to be
1276 * 2. use of it with new communication:
1277 * (2a) "SHOULD NOT be used if alternate address
1278 * with sufficient scope is available"
1279 * (2b) nothing mentioned otherwise.
1280 * Here we fall into (2b) case as we have no choice in
1281 * our source address selection - we must obey the peer.
1283 * The wording in RFC2462 is confusing, and there are
1284 * multiple description text for deprecated address
1285 * handling - worse, they are not exactly the same.
1286 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1288 if (isipv6 && !V_ip6_use_deprecated) {
1289 struct in6_ifaddr *ia6;
1291 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1293 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1294 ifa_free(&ia6->ia_ifa);
1295 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1296 log(LOG_DEBUG, "%s; %s: Listen socket: "
1297 "Connection attempt to deprecated "
1298 "IPv6 address rejected\n",
1300 rstreason = BANDLIM_RST_OPENPORT;
1304 ifa_free(&ia6->ia_ifa);
1308 * Basic sanity checks on incoming SYN requests:
1309 * Don't respond if the destination is a link layer
1310 * broadcast according to RFC1122 4.2.3.10, p. 104.
1311 * If it is from this socket it must be forged.
1312 * Don't respond if the source or destination is a
1313 * global or subnet broad- or multicast address.
1314 * Note that it is quite possible to receive unicast
1315 * link-layer packets with a broadcast IP address. Use
1316 * in_broadcast() to find them.
1318 if (m->m_flags & (M_BCAST|M_MCAST)) {
1319 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1320 log(LOG_DEBUG, "%s; %s: Listen socket: "
1321 "Connection attempt from broad- or multicast "
1322 "link layer address ignored\n", s, __func__);
1327 if (th->th_dport == th->th_sport &&
1328 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1329 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1330 log(LOG_DEBUG, "%s; %s: Listen socket: "
1331 "Connection attempt to/from self "
1332 "ignored\n", s, __func__);
1335 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1336 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1337 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1338 log(LOG_DEBUG, "%s; %s: Listen socket: "
1339 "Connection attempt from/to multicast "
1340 "address ignored\n", s, __func__);
1345 #if defined(INET) && defined(INET6)
1350 if (th->th_dport == th->th_sport &&
1351 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1352 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1353 log(LOG_DEBUG, "%s; %s: Listen socket: "
1354 "Connection attempt from/to self "
1355 "ignored\n", s, __func__);
1358 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1359 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1360 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1361 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1362 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1363 log(LOG_DEBUG, "%s; %s: Listen socket: "
1364 "Connection attempt from/to broad- "
1365 "or multicast address ignored\n",
1372 * SYN appears to be valid. Create compressed TCP state
1376 if (so->so_options & SO_DEBUG)
1377 tcp_trace(TA_INPUT, ostate, tp,
1378 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1380 TCP_PROBE3(debug__input, tp, th, m);
1381 tcp_dooptions(&to, optp, optlen, TO_SYN);
1383 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL))
1384 goto tfo_socket_result;
1386 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1389 * Entry added to syncache and mbuf consumed.
1390 * Only the listen socket is unlocked by syncache_add().
1392 if (ti_locked == TI_RLOCKED) {
1393 INP_INFO_RUNLOCK(&V_tcbinfo);
1394 ti_locked = TI_UNLOCKED;
1396 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1397 return (IPPROTO_DONE);
1398 } else if (tp->t_state == TCPS_LISTEN) {
1400 * When a listen socket is torn down the SO_ACCEPTCONN
1401 * flag is removed first while connections are drained
1402 * from the accept queue in a unlock/lock cycle of the
1403 * ACCEPT_LOCK, opening a race condition allowing a SYN
1404 * attempt go through unhandled.
1408 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1409 if (tp->t_flags & TF_SIGNATURE) {
1410 tcp_dooptions(&to, optp, optlen, thflags);
1411 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1412 TCPSTAT_INC(tcps_sig_err_nosigopt);
1415 if (!TCPMD5_ENABLED() ||
1416 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1420 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1423 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1424 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1425 * the inpcb, and unlocks pcbinfo.
1427 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1428 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1429 return (IPPROTO_DONE);
1432 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1434 if (ti_locked == TI_RLOCKED) {
1435 INP_INFO_RUNLOCK(&V_tcbinfo);
1436 ti_locked = TI_UNLOCKED;
1440 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1441 "ti_locked: %d", __func__, ti_locked));
1442 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1447 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1450 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1451 m = NULL; /* mbuf chain got consumed. */
1456 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1458 if (ti_locked == TI_RLOCKED) {
1459 INP_INFO_RUNLOCK(&V_tcbinfo);
1460 ti_locked = TI_UNLOCKED;
1464 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1465 "ti_locked: %d", __func__, ti_locked));
1466 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1474 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1479 return (IPPROTO_DONE);
1483 * Automatic sizing of receive socket buffer. Often the send
1484 * buffer size is not optimally adjusted to the actual network
1485 * conditions at hand (delay bandwidth product). Setting the
1486 * buffer size too small limits throughput on links with high
1487 * bandwidth and high delay (eg. trans-continental/oceanic links).
1489 * On the receive side the socket buffer memory is only rarely
1490 * used to any significant extent. This allows us to be much
1491 * more aggressive in scaling the receive socket buffer. For
1492 * the case that the buffer space is actually used to a large
1493 * extent and we run out of kernel memory we can simply drop
1494 * the new segments; TCP on the sender will just retransmit it
1495 * later. Setting the buffer size too big may only consume too
1496 * much kernel memory if the application doesn't read() from
1497 * the socket or packet loss or reordering makes use of the
1500 * The criteria to step up the receive buffer one notch are:
1501 * 1. Application has not set receive buffer size with
1502 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1503 * 2. the number of bytes received during the time it takes
1504 * one timestamp to be reflected back to us (the RTT);
1505 * 3. received bytes per RTT is within seven eighth of the
1506 * current socket buffer size;
1507 * 4. receive buffer size has not hit maximal automatic size;
1509 * This algorithm does one step per RTT at most and only if
1510 * we receive a bulk stream w/o packet losses or reorderings.
1511 * Shrinking the buffer during idle times is not necessary as
1512 * it doesn't consume any memory when idle.
1514 * TODO: Only step up if the application is actually serving
1515 * the buffer to better manage the socket buffer resources.
1518 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1519 struct tcpcb *tp, int tlen)
1523 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1524 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1525 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1526 (tp->t_srtt >> TCP_RTT_SHIFT)) {
1527 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) &&
1528 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1529 newsize = min(so->so_rcv.sb_hiwat +
1530 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max);
1532 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1534 /* Start over with next RTT. */
1538 tp->rfbuf_cnt += tlen; /* add up */
1545 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1546 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1549 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1550 int rstreason, todrop, win;
1554 struct in_conninfo *inc;
1563 * The size of tcp_saveipgen must be the size of the max ip header,
1566 u_char tcp_saveipgen[IP6_HDR_LEN];
1567 struct tcphdr tcp_savetcp;
1570 thflags = th->th_flags;
1571 inc = &tp->t_inpcb->inp_inc;
1572 tp->sackhint.last_sack_ack = 0;
1574 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1577 * If this is either a state-changing packet or current state isn't
1578 * established, we require a write lock on tcbinfo. Otherwise, we
1579 * allow the tcbinfo to be in either alocked or unlocked, as the
1580 * caller may have unnecessarily acquired a write lock due to a race.
1582 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1583 tp->t_state != TCPS_ESTABLISHED) {
1584 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for "
1585 "SYN/FIN/RST/!EST", __func__, ti_locked));
1586 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1589 if (ti_locked == TI_RLOCKED)
1590 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1592 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1593 "ti_locked: %d", __func__, ti_locked));
1594 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1598 INP_WLOCK_ASSERT(tp->t_inpcb);
1599 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1601 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1605 /* Save segment, if requested. */
1606 tcp_pcap_add(th, m, &(tp->t_inpkts));
1609 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1610 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1611 log(LOG_DEBUG, "%s; %s: "
1612 "SYN|FIN segment ignored (based on "
1613 "sysctl setting)\n", s, __func__);
1620 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1621 * check SEQ.ACK first.
1623 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1624 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1625 rstreason = BANDLIM_UNLIMITED;
1630 * Segment received on connection.
1631 * Reset idle time and keep-alive timer.
1632 * XXX: This should be done after segment
1633 * validation to ignore broken/spoofed segs.
1635 tp->t_rcvtime = ticks;
1638 * Scale up the window into a 32-bit value.
1639 * For the SYN_SENT state the scale is zero.
1641 tiwin = th->th_win << tp->snd_scale;
1644 * TCP ECN processing.
1646 if (tp->t_flags & TF_ECN_PERMIT) {
1647 if (thflags & TH_CWR)
1648 tp->t_flags &= ~TF_ECN_SND_ECE;
1649 switch (iptos & IPTOS_ECN_MASK) {
1651 tp->t_flags |= TF_ECN_SND_ECE;
1652 TCPSTAT_INC(tcps_ecn_ce);
1654 case IPTOS_ECN_ECT0:
1655 TCPSTAT_INC(tcps_ecn_ect0);
1657 case IPTOS_ECN_ECT1:
1658 TCPSTAT_INC(tcps_ecn_ect1);
1662 /* Process a packet differently from RFC3168. */
1663 cc_ecnpkt_handler(tp, th, iptos);
1665 /* Congestion experienced. */
1666 if (thflags & TH_ECE) {
1667 cc_cong_signal(tp, th, CC_ECN);
1672 * Parse options on any incoming segment.
1674 tcp_dooptions(&to, (u_char *)(th + 1),
1675 (th->th_off << 2) - sizeof(struct tcphdr),
1676 (thflags & TH_SYN) ? TO_SYN : 0);
1678 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1679 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1680 (to.to_flags & TOF_SIGNATURE) == 0) {
1681 TCPSTAT_INC(tcps_sig_err_sigopt);
1682 /* XXX: should drop? */
1686 * If echoed timestamp is later than the current time,
1687 * fall back to non RFC1323 RTT calculation. Normalize
1688 * timestamp if syncookies were used when this connection
1691 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1692 to.to_tsecr -= tp->ts_offset;
1693 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1697 * Process options only when we get SYN/ACK back. The SYN case
1698 * for incoming connections is handled in tcp_syncache.
1699 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1700 * or <SYN,ACK>) segment itself is never scaled.
1701 * XXX this is traditional behavior, may need to be cleaned up.
1703 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1704 if ((to.to_flags & TOF_SCALE) &&
1705 (tp->t_flags & TF_REQ_SCALE)) {
1706 tp->t_flags |= TF_RCVD_SCALE;
1707 tp->snd_scale = to.to_wscale;
1710 * Initial send window. It will be updated with
1711 * the next incoming segment to the scaled value.
1713 tp->snd_wnd = th->th_win;
1714 if (to.to_flags & TOF_TS) {
1715 tp->t_flags |= TF_RCVD_TSTMP;
1716 tp->ts_recent = to.to_tsval;
1717 tp->ts_recent_age = tcp_ts_getticks();
1719 if (to.to_flags & TOF_MSS)
1720 tcp_mss(tp, to.to_mss);
1721 if ((tp->t_flags & TF_SACK_PERMIT) &&
1722 (to.to_flags & TOF_SACKPERM) == 0)
1723 tp->t_flags &= ~TF_SACK_PERMIT;
1727 * If timestamps were negotiated during SYN/ACK they should
1728 * appear on every segment during this session and vice versa.
1730 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1731 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1732 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1733 "no action\n", s, __func__);
1737 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1738 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1739 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1740 "no action\n", s, __func__);
1746 * Header prediction: check for the two common cases
1747 * of a uni-directional data xfer. If the packet has
1748 * no control flags, is in-sequence, the window didn't
1749 * change and we're not retransmitting, it's a
1750 * candidate. If the length is zero and the ack moved
1751 * forward, we're the sender side of the xfer. Just
1752 * free the data acked & wake any higher level process
1753 * that was blocked waiting for space. If the length
1754 * is non-zero and the ack didn't move, we're the
1755 * receiver side. If we're getting packets in-order
1756 * (the reassembly queue is empty), add the data to
1757 * the socket buffer and note that we need a delayed ack.
1758 * Make sure that the hidden state-flags are also off.
1759 * Since we check for TCPS_ESTABLISHED first, it can only
1762 if (tp->t_state == TCPS_ESTABLISHED &&
1763 th->th_seq == tp->rcv_nxt &&
1764 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1765 tp->snd_nxt == tp->snd_max &&
1766 tiwin && tiwin == tp->snd_wnd &&
1767 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1768 LIST_EMPTY(&tp->t_segq) &&
1769 ((to.to_flags & TOF_TS) == 0 ||
1770 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1773 * If last ACK falls within this segment's sequence numbers,
1774 * record the timestamp.
1775 * NOTE that the test is modified according to the latest
1776 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1778 if ((to.to_flags & TOF_TS) != 0 &&
1779 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1780 tp->ts_recent_age = tcp_ts_getticks();
1781 tp->ts_recent = to.to_tsval;
1785 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1786 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1787 !IN_RECOVERY(tp->t_flags) &&
1788 (to.to_flags & TOF_SACK) == 0 &&
1789 TAILQ_EMPTY(&tp->snd_holes)) {
1791 * This is a pure ack for outstanding data.
1793 if (ti_locked == TI_RLOCKED)
1794 INP_INFO_RUNLOCK(&V_tcbinfo);
1795 ti_locked = TI_UNLOCKED;
1797 TCPSTAT_INC(tcps_predack);
1800 * "bad retransmit" recovery.
1802 if (tp->t_rxtshift == 1 &&
1803 tp->t_flags & TF_PREVVALID &&
1804 (int)(ticks - tp->t_badrxtwin) < 0) {
1805 cc_cong_signal(tp, th, CC_RTO_ERR);
1809 * Recalculate the transmit timer / rtt.
1811 * Some boxes send broken timestamp replies
1812 * during the SYN+ACK phase, ignore
1813 * timestamps of 0 or we could calculate a
1814 * huge RTT and blow up the retransmit timer.
1816 if ((to.to_flags & TOF_TS) != 0 &&
1820 t = tcp_ts_getticks() - to.to_tsecr;
1821 if (!tp->t_rttlow || tp->t_rttlow > t)
1824 TCP_TS_TO_TICKS(t) + 1);
1825 } else if (tp->t_rtttime &&
1826 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1827 if (!tp->t_rttlow ||
1828 tp->t_rttlow > ticks - tp->t_rtttime)
1829 tp->t_rttlow = ticks - tp->t_rtttime;
1831 ticks - tp->t_rtttime);
1833 acked = BYTES_THIS_ACK(tp, th);
1836 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1837 hhook_run_tcp_est_in(tp, th, &to);
1840 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1841 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1842 sbdrop(&so->so_snd, acked);
1843 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1844 SEQ_LEQ(th->th_ack, tp->snd_recover))
1845 tp->snd_recover = th->th_ack - 1;
1848 * Let the congestion control algorithm update
1849 * congestion control related information. This
1850 * typically means increasing the congestion
1853 cc_ack_received(tp, th, nsegs, CC_ACK);
1855 tp->snd_una = th->th_ack;
1857 * Pull snd_wl2 up to prevent seq wrap relative
1860 tp->snd_wl2 = th->th_ack;
1865 * If all outstanding data are acked, stop
1866 * retransmit timer, otherwise restart timer
1867 * using current (possibly backed-off) value.
1868 * If process is waiting for space,
1869 * wakeup/selwakeup/signal. If data
1870 * are ready to send, let tcp_output
1871 * decide between more output or persist.
1874 if (so->so_options & SO_DEBUG)
1875 tcp_trace(TA_INPUT, ostate, tp,
1876 (void *)tcp_saveipgen,
1879 TCP_PROBE3(debug__input, tp, th, m);
1880 if (tp->snd_una == tp->snd_max)
1881 tcp_timer_activate(tp, TT_REXMT, 0);
1882 else if (!tcp_timer_active(tp, TT_PERSIST))
1883 tcp_timer_activate(tp, TT_REXMT,
1886 if (sbavail(&so->so_snd))
1887 (void) tp->t_fb->tfb_tcp_output(tp);
1890 } else if (th->th_ack == tp->snd_una &&
1891 tlen <= sbspace(&so->so_rcv)) {
1892 int newsize = 0; /* automatic sockbuf scaling */
1895 * This is a pure, in-sequence data packet with
1896 * nothing on the reassembly queue and we have enough
1897 * buffer space to take it.
1899 if (ti_locked == TI_RLOCKED)
1900 INP_INFO_RUNLOCK(&V_tcbinfo);
1901 ti_locked = TI_UNLOCKED;
1903 /* Clean receiver SACK report if present */
1904 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1905 tcp_clean_sackreport(tp);
1906 TCPSTAT_INC(tcps_preddat);
1907 tp->rcv_nxt += tlen;
1909 * Pull snd_wl1 up to prevent seq wrap relative to
1912 tp->snd_wl1 = th->th_seq;
1914 * Pull rcv_up up to prevent seq wrap relative to
1917 tp->rcv_up = tp->rcv_nxt;
1918 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1919 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1921 if (so->so_options & SO_DEBUG)
1922 tcp_trace(TA_INPUT, ostate, tp,
1923 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1925 TCP_PROBE3(debug__input, tp, th, m);
1927 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1929 /* Add data to socket buffer. */
1930 SOCKBUF_LOCK(&so->so_rcv);
1931 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1935 * Set new socket buffer size.
1936 * Give up when limit is reached.
1939 if (!sbreserve_locked(&so->so_rcv,
1941 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1942 m_adj(m, drop_hdrlen); /* delayed header drop */
1943 sbappendstream_locked(&so->so_rcv, m, 0);
1945 /* NB: sorwakeup_locked() does an implicit unlock. */
1946 sorwakeup_locked(so);
1947 if (DELAY_ACK(tp, tlen)) {
1948 tp->t_flags |= TF_DELACK;
1950 tp->t_flags |= TF_ACKNOW;
1951 tp->t_fb->tfb_tcp_output(tp);
1958 * Calculate amount of space in receive window,
1959 * and then do TCP input processing.
1960 * Receive window is amount of space in rcv queue,
1961 * but not less than advertised window.
1963 win = sbspace(&so->so_rcv);
1966 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1968 switch (tp->t_state) {
1971 * If the state is SYN_RECEIVED:
1972 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1974 case TCPS_SYN_RECEIVED:
1975 if ((thflags & TH_ACK) &&
1976 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1977 SEQ_GT(th->th_ack, tp->snd_max))) {
1978 rstreason = BANDLIM_RST_OPENPORT;
1982 if (IS_FASTOPEN(tp->t_flags)) {
1984 * When a TFO connection is in SYN_RECEIVED, the
1985 * only valid packets are the initial SYN, a
1986 * retransmit/copy of the initial SYN (possibly with
1987 * a subset of the original data), a valid ACK, a
1990 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1991 rstreason = BANDLIM_RST_OPENPORT;
1993 } else if (thflags & TH_SYN) {
1994 /* non-initial SYN is ignored */
1995 if ((tcp_timer_active(tp, TT_DELACK) ||
1996 tcp_timer_active(tp, TT_REXMT)))
1998 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2006 * If the state is SYN_SENT:
2007 * if seg contains a RST with valid ACK (SEQ.ACK has already
2008 * been verified), then drop the connection.
2009 * if seg contains a RST without an ACK, drop the seg.
2010 * if seg does not contain SYN, then drop the seg.
2011 * Otherwise this is an acceptable SYN segment
2012 * initialize tp->rcv_nxt and tp->irs
2013 * if seg contains ack then advance tp->snd_una
2014 * if seg contains an ECE and ECN support is enabled, the stream
2016 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2017 * arrange for segment to be acked (eventually)
2018 * continue processing rest of data/controls, beginning with URG
2021 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2022 TCP_PROBE5(connect__refused, NULL, tp,
2024 tp = tcp_drop(tp, ECONNREFUSED);
2026 if (thflags & TH_RST)
2028 if (!(thflags & TH_SYN))
2031 tp->irs = th->th_seq;
2033 if (thflags & TH_ACK) {
2034 TCPSTAT_INC(tcps_connects);
2037 mac_socketpeer_set_from_mbuf(m, so);
2039 /* Do window scaling on this connection? */
2040 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2041 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2042 tp->rcv_scale = tp->request_r_scale;
2044 tp->rcv_adv += min(tp->rcv_wnd,
2045 TCP_MAXWIN << tp->rcv_scale);
2046 tp->snd_una++; /* SYN is acked */
2048 * If there's data, delay ACK; if there's also a FIN
2049 * ACKNOW will be turned on later.
2051 if (DELAY_ACK(tp, tlen) && tlen != 0)
2052 tcp_timer_activate(tp, TT_DELACK,
2055 tp->t_flags |= TF_ACKNOW;
2057 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
2058 tp->t_flags |= TF_ECN_PERMIT;
2059 TCPSTAT_INC(tcps_ecn_shs);
2063 * Received <SYN,ACK> in SYN_SENT[*] state.
2065 * SYN_SENT --> ESTABLISHED
2066 * SYN_SENT* --> FIN_WAIT_1
2068 tp->t_starttime = ticks;
2069 if (tp->t_flags & TF_NEEDFIN) {
2070 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2071 tp->t_flags &= ~TF_NEEDFIN;
2074 tcp_state_change(tp, TCPS_ESTABLISHED);
2075 TCP_PROBE5(connect__established, NULL, tp,
2078 tcp_timer_activate(tp, TT_KEEP,
2083 * Received initial SYN in SYN-SENT[*] state =>
2084 * simultaneous open.
2085 * If it succeeds, connection is * half-synchronized.
2086 * Otherwise, do 3-way handshake:
2087 * SYN-SENT -> SYN-RECEIVED
2088 * SYN-SENT* -> SYN-RECEIVED*
2090 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2091 tcp_timer_activate(tp, TT_REXMT, 0);
2092 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2095 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: "
2096 "ti_locked %d", __func__, ti_locked));
2097 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2098 INP_WLOCK_ASSERT(tp->t_inpcb);
2101 * Advance th->th_seq to correspond to first data byte.
2102 * If data, trim to stay within window,
2103 * dropping FIN if necessary.
2106 if (tlen > tp->rcv_wnd) {
2107 todrop = tlen - tp->rcv_wnd;
2111 TCPSTAT_INC(tcps_rcvpackafterwin);
2112 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2114 tp->snd_wl1 = th->th_seq - 1;
2115 tp->rcv_up = th->th_seq;
2117 * Client side of transaction: already sent SYN and data.
2118 * If the remote host used T/TCP to validate the SYN,
2119 * our data will be ACK'd; if so, enter normal data segment
2120 * processing in the middle of step 5, ack processing.
2121 * Otherwise, goto step 6.
2123 if (thflags & TH_ACK)
2129 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2130 * do normal processing.
2132 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2136 break; /* continue normal processing */
2140 * States other than LISTEN or SYN_SENT.
2141 * First check the RST flag and sequence number since reset segments
2142 * are exempt from the timestamp and connection count tests. This
2143 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2144 * below which allowed reset segments in half the sequence space
2145 * to fall though and be processed (which gives forged reset
2146 * segments with a random sequence number a 50 percent chance of
2147 * killing a connection).
2148 * Then check timestamp, if present.
2149 * Then check the connection count, if present.
2150 * Then check that at least some bytes of segment are within
2151 * receive window. If segment begins before rcv_nxt,
2152 * drop leading data (and SYN); if nothing left, just ack.
2154 if (thflags & TH_RST) {
2156 * RFC5961 Section 3.2
2158 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2159 * - If RST is in window, we send challenge ACK.
2161 * Note: to take into account delayed ACKs, we should
2162 * test against last_ack_sent instead of rcv_nxt.
2163 * Note 2: we handle special case of closed window, not
2164 * covered by the RFC.
2166 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2167 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2168 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2170 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2171 KASSERT(ti_locked == TI_RLOCKED,
2172 ("%s: TH_RST ti_locked %d, th %p tp %p",
2173 __func__, ti_locked, th, tp));
2174 KASSERT(tp->t_state != TCPS_SYN_SENT,
2175 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2178 if (V_tcp_insecure_rst ||
2179 tp->last_ack_sent == th->th_seq) {
2180 TCPSTAT_INC(tcps_drops);
2181 /* Drop the connection. */
2182 switch (tp->t_state) {
2183 case TCPS_SYN_RECEIVED:
2184 so->so_error = ECONNREFUSED;
2186 case TCPS_ESTABLISHED:
2187 case TCPS_FIN_WAIT_1:
2188 case TCPS_FIN_WAIT_2:
2189 case TCPS_CLOSE_WAIT:
2192 so->so_error = ECONNRESET;
2199 TCPSTAT_INC(tcps_badrst);
2200 /* Send challenge ACK. */
2201 tcp_respond(tp, mtod(m, void *), th, m,
2202 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2203 tp->last_ack_sent = tp->rcv_nxt;
2211 * RFC5961 Section 4.2
2212 * Send challenge ACK for any SYN in synchronized state.
2214 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2215 tp->t_state != TCPS_SYN_RECEIVED) {
2216 KASSERT(ti_locked == TI_RLOCKED,
2217 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2218 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2220 TCPSTAT_INC(tcps_badsyn);
2221 if (V_tcp_insecure_syn &&
2222 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2223 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2224 tp = tcp_drop(tp, ECONNRESET);
2225 rstreason = BANDLIM_UNLIMITED;
2227 /* Send challenge ACK. */
2228 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2229 tp->snd_nxt, TH_ACK);
2230 tp->last_ack_sent = tp->rcv_nxt;
2237 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2238 * and it's less than ts_recent, drop it.
2240 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2241 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2243 /* Check to see if ts_recent is over 24 days old. */
2244 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2246 * Invalidate ts_recent. If this segment updates
2247 * ts_recent, the age will be reset later and ts_recent
2248 * will get a valid value. If it does not, setting
2249 * ts_recent to zero will at least satisfy the
2250 * requirement that zero be placed in the timestamp
2251 * echo reply when ts_recent isn't valid. The
2252 * age isn't reset until we get a valid ts_recent
2253 * because we don't want out-of-order segments to be
2254 * dropped when ts_recent is old.
2258 TCPSTAT_INC(tcps_rcvduppack);
2259 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2260 TCPSTAT_INC(tcps_pawsdrop);
2268 * In the SYN-RECEIVED state, validate that the packet belongs to
2269 * this connection before trimming the data to fit the receive
2270 * window. Check the sequence number versus IRS since we know
2271 * the sequence numbers haven't wrapped. This is a partial fix
2272 * for the "LAND" DoS attack.
2274 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2275 rstreason = BANDLIM_RST_OPENPORT;
2279 todrop = tp->rcv_nxt - th->th_seq;
2281 if (thflags & TH_SYN) {
2291 * Following if statement from Stevens, vol. 2, p. 960.
2294 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2296 * Any valid FIN must be to the left of the window.
2297 * At this point the FIN must be a duplicate or out
2298 * of sequence; drop it.
2303 * Send an ACK to resynchronize and drop any data.
2304 * But keep on processing for RST or ACK.
2306 tp->t_flags |= TF_ACKNOW;
2308 TCPSTAT_INC(tcps_rcvduppack);
2309 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2311 TCPSTAT_INC(tcps_rcvpartduppack);
2312 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2314 drop_hdrlen += todrop; /* drop from the top afterwards */
2315 th->th_seq += todrop;
2317 if (th->th_urp > todrop)
2318 th->th_urp -= todrop;
2326 * If new data are received on a connection after the
2327 * user processes are gone, then RST the other end.
2329 if ((so->so_state & SS_NOFDREF) &&
2330 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2331 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && "
2332 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2333 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2335 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2336 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2337 "after socket was closed, "
2338 "sending RST and removing tcpcb\n",
2339 s, __func__, tcpstates[tp->t_state], tlen);
2343 TCPSTAT_INC(tcps_rcvafterclose);
2344 rstreason = BANDLIM_UNLIMITED;
2349 * If segment ends after window, drop trailing data
2350 * (and PUSH and FIN); if nothing left, just ACK.
2352 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2354 TCPSTAT_INC(tcps_rcvpackafterwin);
2355 if (todrop >= tlen) {
2356 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2358 * If window is closed can only take segments at
2359 * window edge, and have to drop data and PUSH from
2360 * incoming segments. Continue processing, but
2361 * remember to ack. Otherwise, drop segment
2364 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2365 tp->t_flags |= TF_ACKNOW;
2366 TCPSTAT_INC(tcps_rcvwinprobe);
2370 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2373 thflags &= ~(TH_PUSH|TH_FIN);
2377 * If last ACK falls within this segment's sequence numbers,
2378 * record its timestamp.
2380 * 1) That the test incorporates suggestions from the latest
2381 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2382 * 2) That updating only on newer timestamps interferes with
2383 * our earlier PAWS tests, so this check should be solely
2384 * predicated on the sequence space of this segment.
2385 * 3) That we modify the segment boundary check to be
2386 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2387 * instead of RFC1323's
2388 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2389 * This modified check allows us to overcome RFC1323's
2390 * limitations as described in Stevens TCP/IP Illustrated
2391 * Vol. 2 p.869. In such cases, we can still calculate the
2392 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2394 if ((to.to_flags & TOF_TS) != 0 &&
2395 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2396 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2397 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2398 tp->ts_recent_age = tcp_ts_getticks();
2399 tp->ts_recent = to.to_tsval;
2403 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2404 * flag is on (half-synchronized state), then queue data for
2405 * later processing; else drop segment and return.
2407 if ((thflags & TH_ACK) == 0) {
2408 if (tp->t_state == TCPS_SYN_RECEIVED ||
2409 (tp->t_flags & TF_NEEDSYN)) {
2411 if (tp->t_state == TCPS_SYN_RECEIVED &&
2412 IS_FASTOPEN(tp->t_flags)) {
2413 tp->snd_wnd = tiwin;
2418 } else if (tp->t_flags & TF_ACKNOW)
2427 switch (tp->t_state) {
2430 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2431 * ESTABLISHED state and continue processing.
2432 * The ACK was checked above.
2434 case TCPS_SYN_RECEIVED:
2436 TCPSTAT_INC(tcps_connects);
2438 /* Do window scaling? */
2439 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2440 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2441 tp->rcv_scale = tp->request_r_scale;
2442 tp->snd_wnd = tiwin;
2446 * SYN-RECEIVED -> ESTABLISHED
2447 * SYN-RECEIVED* -> FIN-WAIT-1
2449 tp->t_starttime = ticks;
2450 if (tp->t_flags & TF_NEEDFIN) {
2451 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2452 tp->t_flags &= ~TF_NEEDFIN;
2454 tcp_state_change(tp, TCPS_ESTABLISHED);
2455 TCP_PROBE5(accept__established, NULL, tp,
2458 if (tp->t_tfo_pending) {
2459 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2460 tp->t_tfo_pending = NULL;
2463 * Account for the ACK of our SYN prior to
2464 * regular ACK processing below.
2469 * TFO connections call cc_conn_init() during SYN
2470 * processing. Calling it again here for such
2471 * connections is not harmless as it would undo the
2472 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2475 if (!IS_FASTOPEN(tp->t_flags))
2478 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2481 * If segment contains data or ACK, will call tcp_reass()
2482 * later; if not, do so now to pass queued data to user.
2484 if (tlen == 0 && (thflags & TH_FIN) == 0)
2485 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2487 tp->snd_wl1 = th->th_seq - 1;
2491 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2492 * ACKs. If the ack is in the range
2493 * tp->snd_una < th->th_ack <= tp->snd_max
2494 * then advance tp->snd_una to th->th_ack and drop
2495 * data from the retransmission queue. If this ACK reflects
2496 * more up to date window information we update our window information.
2498 case TCPS_ESTABLISHED:
2499 case TCPS_FIN_WAIT_1:
2500 case TCPS_FIN_WAIT_2:
2501 case TCPS_CLOSE_WAIT:
2504 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2505 TCPSTAT_INC(tcps_rcvacktoomuch);
2508 if ((tp->t_flags & TF_SACK_PERMIT) &&
2509 ((to.to_flags & TOF_SACK) ||
2510 !TAILQ_EMPTY(&tp->snd_holes)))
2511 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2514 * Reset the value so that previous (valid) value
2515 * from the last ack with SACK doesn't get used.
2517 tp->sackhint.sacked_bytes = 0;
2520 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2521 hhook_run_tcp_est_in(tp, th, &to);
2524 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2527 maxseg = tcp_maxseg(tp);
2529 (tiwin == tp->snd_wnd ||
2530 (tp->t_flags & TF_SACK_PERMIT))) {
2532 * If this is the first time we've seen a
2533 * FIN from the remote, this is not a
2534 * duplicate and it needs to be processed
2535 * normally. This happens during a
2536 * simultaneous close.
2538 if ((thflags & TH_FIN) &&
2539 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2543 TCPSTAT_INC(tcps_rcvdupack);
2545 * If we have outstanding data (other than
2546 * a window probe), this is a completely
2547 * duplicate ack (ie, window info didn't
2548 * change and FIN isn't set),
2549 * the ack is the biggest we've
2550 * seen and we've seen exactly our rexmt
2551 * threshold of them, assume a packet
2552 * has been dropped and retransmit it.
2553 * Kludge snd_nxt & the congestion
2554 * window so we send only this one
2557 * We know we're losing at the current
2558 * window size so do congestion avoidance
2559 * (set ssthresh to half the current window
2560 * and pull our congestion window back to
2561 * the new ssthresh).
2563 * Dup acks mean that packets have left the
2564 * network (they're now cached at the receiver)
2565 * so bump cwnd by the amount in the receiver
2566 * to keep a constant cwnd packets in the
2569 * When using TCP ECN, notify the peer that
2570 * we reduced the cwnd.
2573 * Following 2 kinds of acks should not affect
2576 * 2) Acks with SACK but without any new SACK
2577 * information in them. These could result from
2578 * any anomaly in the network like a switch
2579 * duplicating packets or a possible DoS attack.
2581 if (th->th_ack != tp->snd_una ||
2582 ((tp->t_flags & TF_SACK_PERMIT) &&
2585 else if (!tcp_timer_active(tp, TT_REXMT))
2587 else if (++tp->t_dupacks > tcprexmtthresh ||
2588 IN_FASTRECOVERY(tp->t_flags)) {
2589 cc_ack_received(tp, th, nsegs,
2591 if ((tp->t_flags & TF_SACK_PERMIT) &&
2592 IN_FASTRECOVERY(tp->t_flags)) {
2596 * Compute the amount of data in flight first.
2597 * We can inject new data into the pipe iff
2598 * we have less than 1/2 the original window's
2599 * worth of data in flight.
2601 if (V_tcp_do_rfc6675_pipe)
2602 awnd = tcp_compute_pipe(tp);
2604 awnd = (tp->snd_nxt - tp->snd_fack) +
2605 tp->sackhint.sack_bytes_rexmit;
2607 if (awnd < tp->snd_ssthresh) {
2608 tp->snd_cwnd += maxseg;
2609 if (tp->snd_cwnd > tp->snd_ssthresh)
2610 tp->snd_cwnd = tp->snd_ssthresh;
2613 tp->snd_cwnd += maxseg;
2614 (void) tp->t_fb->tfb_tcp_output(tp);
2616 } else if (tp->t_dupacks == tcprexmtthresh) {
2617 tcp_seq onxt = tp->snd_nxt;
2620 * If we're doing sack, check to
2621 * see if we're already in sack
2622 * recovery. If we're not doing sack,
2623 * check to see if we're in newreno
2626 if (tp->t_flags & TF_SACK_PERMIT) {
2627 if (IN_FASTRECOVERY(tp->t_flags)) {
2632 if (SEQ_LEQ(th->th_ack,
2638 /* Congestion signal before ack. */
2639 cc_cong_signal(tp, th, CC_NDUPACK);
2640 cc_ack_received(tp, th, nsegs,
2642 tcp_timer_activate(tp, TT_REXMT, 0);
2644 if (tp->t_flags & TF_SACK_PERMIT) {
2646 tcps_sack_recovery_episode);
2647 tp->sack_newdata = tp->snd_nxt;
2648 tp->snd_cwnd = maxseg;
2649 (void) tp->t_fb->tfb_tcp_output(tp);
2652 tp->snd_nxt = th->th_ack;
2653 tp->snd_cwnd = maxseg;
2654 (void) tp->t_fb->tfb_tcp_output(tp);
2655 KASSERT(tp->snd_limited <= 2,
2656 ("%s: tp->snd_limited too big",
2658 tp->snd_cwnd = tp->snd_ssthresh +
2660 (tp->t_dupacks - tp->snd_limited);
2661 if (SEQ_GT(onxt, tp->snd_nxt))
2664 } else if (V_tcp_do_rfc3042) {
2666 * Process first and second duplicate
2667 * ACKs. Each indicates a segment
2668 * leaving the network, creating room
2669 * for more. Make sure we can send a
2670 * packet on reception of each duplicate
2671 * ACK by increasing snd_cwnd by one
2672 * segment. Restore the original
2673 * snd_cwnd after packet transmission.
2675 cc_ack_received(tp, th, nsegs,
2677 uint32_t oldcwnd = tp->snd_cwnd;
2678 tcp_seq oldsndmax = tp->snd_max;
2682 KASSERT(tp->t_dupacks == 1 ||
2684 ("%s: dupacks not 1 or 2",
2686 if (tp->t_dupacks == 1)
2687 tp->snd_limited = 0;
2689 (tp->snd_nxt - tp->snd_una) +
2690 (tp->t_dupacks - tp->snd_limited) *
2693 * Only call tcp_output when there
2694 * is new data available to be sent.
2695 * Otherwise we would send pure ACKs.
2697 SOCKBUF_LOCK(&so->so_snd);
2698 avail = sbavail(&so->so_snd) -
2699 (tp->snd_nxt - tp->snd_una);
2700 SOCKBUF_UNLOCK(&so->so_snd);
2702 (void) tp->t_fb->tfb_tcp_output(tp);
2703 sent = tp->snd_max - oldsndmax;
2704 if (sent > maxseg) {
2705 KASSERT((tp->t_dupacks == 2 &&
2706 tp->snd_limited == 0) ||
2707 (sent == maxseg + 1 &&
2708 tp->t_flags & TF_SENTFIN),
2709 ("%s: sent too much",
2711 tp->snd_limited = 2;
2712 } else if (sent > 0)
2714 tp->snd_cwnd = oldcwnd;
2721 * This ack is advancing the left edge, reset the
2726 * If this ack also has new SACK info, increment the
2727 * counter as per rfc6675.
2729 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed)
2733 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2734 ("%s: th_ack <= snd_una", __func__));
2737 * If the congestion window was inflated to account
2738 * for the other side's cached packets, retract it.
2740 if (IN_FASTRECOVERY(tp->t_flags)) {
2741 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2742 if (tp->t_flags & TF_SACK_PERMIT)
2743 tcp_sack_partialack(tp, th);
2745 tcp_newreno_partial_ack(tp, th);
2747 cc_post_recovery(tp, th);
2750 * If we reach this point, ACK is not a duplicate,
2751 * i.e., it ACKs something we sent.
2753 if (tp->t_flags & TF_NEEDSYN) {
2755 * T/TCP: Connection was half-synchronized, and our
2756 * SYN has been ACK'd (so connection is now fully
2757 * synchronized). Go to non-starred state,
2758 * increment snd_una for ACK of SYN, and check if
2759 * we can do window scaling.
2761 tp->t_flags &= ~TF_NEEDSYN;
2763 /* Do window scaling? */
2764 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2765 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2766 tp->rcv_scale = tp->request_r_scale;
2767 /* Send window already scaled. */
2772 INP_WLOCK_ASSERT(tp->t_inpcb);
2774 acked = BYTES_THIS_ACK(tp, th);
2775 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2776 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2777 tp->snd_una, th->th_ack, tp, m));
2778 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2779 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2782 * If we just performed our first retransmit, and the ACK
2783 * arrives within our recovery window, then it was a mistake
2784 * to do the retransmit in the first place. Recover our
2785 * original cwnd and ssthresh, and proceed to transmit where
2788 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2789 (int)(ticks - tp->t_badrxtwin) < 0)
2790 cc_cong_signal(tp, th, CC_RTO_ERR);
2793 * If we have a timestamp reply, update smoothed
2794 * round trip time. If no timestamp is present but
2795 * transmit timer is running and timed sequence
2796 * number was acked, update smoothed round trip time.
2797 * Since we now have an rtt measurement, cancel the
2798 * timer backoff (cf., Phil Karn's retransmit alg.).
2799 * Recompute the initial retransmit timer.
2801 * Some boxes send broken timestamp replies
2802 * during the SYN+ACK phase, ignore
2803 * timestamps of 0 or we could calculate a
2804 * huge RTT and blow up the retransmit timer.
2806 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2809 t = tcp_ts_getticks() - to.to_tsecr;
2810 if (!tp->t_rttlow || tp->t_rttlow > t)
2812 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2813 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2814 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2815 tp->t_rttlow = ticks - tp->t_rtttime;
2816 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2820 * If all outstanding data is acked, stop retransmit
2821 * timer and remember to restart (more output or persist).
2822 * If there is more data to be acked, restart retransmit
2823 * timer, using current (possibly backed-off) value.
2825 if (th->th_ack == tp->snd_max) {
2826 tcp_timer_activate(tp, TT_REXMT, 0);
2828 } else if (!tcp_timer_active(tp, TT_PERSIST))
2829 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2832 * If no data (only SYN) was ACK'd,
2833 * skip rest of ACK processing.
2839 * Let the congestion control algorithm update congestion
2840 * control related information. This typically means increasing
2841 * the congestion window.
2843 cc_ack_received(tp, th, nsegs, CC_ACK);
2845 SOCKBUF_LOCK(&so->so_snd);
2846 if (acked > sbavail(&so->so_snd)) {
2847 if (tp->snd_wnd >= sbavail(&so->so_snd))
2848 tp->snd_wnd -= sbavail(&so->so_snd);
2851 mfree = sbcut_locked(&so->so_snd,
2852 (int)sbavail(&so->so_snd));
2855 mfree = sbcut_locked(&so->so_snd, acked);
2856 if (tp->snd_wnd >= (uint32_t) acked)
2857 tp->snd_wnd -= acked;
2862 /* NB: sowwakeup_locked() does an implicit unlock. */
2863 sowwakeup_locked(so);
2865 /* Detect una wraparound. */
2866 if (!IN_RECOVERY(tp->t_flags) &&
2867 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2868 SEQ_LEQ(th->th_ack, tp->snd_recover))
2869 tp->snd_recover = th->th_ack - 1;
2870 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2871 if (IN_RECOVERY(tp->t_flags) &&
2872 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2873 EXIT_RECOVERY(tp->t_flags);
2875 tp->snd_una = th->th_ack;
2876 if (tp->t_flags & TF_SACK_PERMIT) {
2877 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2878 tp->snd_recover = tp->snd_una;
2880 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2881 tp->snd_nxt = tp->snd_una;
2883 switch (tp->t_state) {
2886 * In FIN_WAIT_1 STATE in addition to the processing
2887 * for the ESTABLISHED state if our FIN is now acknowledged
2888 * then enter FIN_WAIT_2.
2890 case TCPS_FIN_WAIT_1:
2891 if (ourfinisacked) {
2893 * If we can't receive any more
2894 * data, then closing user can proceed.
2895 * Starting the timer is contrary to the
2896 * specification, but if we don't get a FIN
2897 * we'll hang forever.
2900 * we should release the tp also, and use a
2903 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2904 soisdisconnected(so);
2905 tcp_timer_activate(tp, TT_2MSL,
2906 (tcp_fast_finwait2_recycle ?
2907 tcp_finwait2_timeout :
2910 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2915 * In CLOSING STATE in addition to the processing for
2916 * the ESTABLISHED state if the ACK acknowledges our FIN
2917 * then enter the TIME-WAIT state, otherwise ignore
2921 if (ourfinisacked) {
2922 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2924 INP_INFO_RUNLOCK(&V_tcbinfo);
2931 * In LAST_ACK, we may still be waiting for data to drain
2932 * and/or to be acked, as well as for the ack of our FIN.
2933 * If our FIN is now acknowledged, delete the TCB,
2934 * enter the closed state and return.
2937 if (ourfinisacked) {
2938 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2947 INP_WLOCK_ASSERT(tp->t_inpcb);
2950 * Update window information.
2951 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2953 if ((thflags & TH_ACK) &&
2954 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2955 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2956 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2957 /* keep track of pure window updates */
2959 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2960 TCPSTAT_INC(tcps_rcvwinupd);
2961 tp->snd_wnd = tiwin;
2962 tp->snd_wl1 = th->th_seq;
2963 tp->snd_wl2 = th->th_ack;
2964 if (tp->snd_wnd > tp->max_sndwnd)
2965 tp->max_sndwnd = tp->snd_wnd;
2970 * Process segments with URG.
2972 if ((thflags & TH_URG) && th->th_urp &&
2973 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2975 * This is a kludge, but if we receive and accept
2976 * random urgent pointers, we'll crash in
2977 * soreceive. It's hard to imagine someone
2978 * actually wanting to send this much urgent data.
2980 SOCKBUF_LOCK(&so->so_rcv);
2981 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2982 th->th_urp = 0; /* XXX */
2983 thflags &= ~TH_URG; /* XXX */
2984 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2985 goto dodata; /* XXX */
2988 * If this segment advances the known urgent pointer,
2989 * then mark the data stream. This should not happen
2990 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2991 * a FIN has been received from the remote side.
2992 * In these states we ignore the URG.
2994 * According to RFC961 (Assigned Protocols),
2995 * the urgent pointer points to the last octet
2996 * of urgent data. We continue, however,
2997 * to consider it to indicate the first octet
2998 * of data past the urgent section as the original
2999 * spec states (in one of two places).
3001 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3002 tp->rcv_up = th->th_seq + th->th_urp;
3003 so->so_oobmark = sbavail(&so->so_rcv) +
3004 (tp->rcv_up - tp->rcv_nxt) - 1;
3005 if (so->so_oobmark == 0)
3006 so->so_rcv.sb_state |= SBS_RCVATMARK;
3008 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3010 SOCKBUF_UNLOCK(&so->so_rcv);
3012 * Remove out of band data so doesn't get presented to user.
3013 * This can happen independent of advancing the URG pointer,
3014 * but if two URG's are pending at once, some out-of-band
3015 * data may creep in... ick.
3017 if (th->th_urp <= (uint32_t)tlen &&
3018 !(so->so_options & SO_OOBINLINE)) {
3019 /* hdr drop is delayed */
3020 tcp_pulloutofband(so, th, m, drop_hdrlen);
3024 * If no out of band data is expected,
3025 * pull receive urgent pointer along
3026 * with the receive window.
3028 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3029 tp->rcv_up = tp->rcv_nxt;
3032 INP_WLOCK_ASSERT(tp->t_inpcb);
3035 * Process the segment text, merging it into the TCP sequencing queue,
3036 * and arranging for acknowledgment of receipt if necessary.
3037 * This process logically involves adjusting tp->rcv_wnd as data
3038 * is presented to the user (this happens in tcp_usrreq.c,
3039 * case PRU_RCVD). If a FIN has already been received on this
3040 * connection then we just ignore the text.
3043 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3044 IS_FASTOPEN(tp->t_flags));
3046 #define tfo_syn (false)
3048 if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
3049 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3050 tcp_seq save_start = th->th_seq;
3051 m_adj(m, drop_hdrlen); /* delayed header drop */
3053 * Insert segment which includes th into TCP reassembly queue
3054 * with control block tp. Set thflags to whether reassembly now
3055 * includes a segment with FIN. This handles the common case
3056 * inline (segment is the next to be received on an established
3057 * connection, and the queue is empty), avoiding linkage into
3058 * and removal from the queue and repetition of various
3060 * Set DELACK for segments received in order, but ack
3061 * immediately when segments are out of order (so
3062 * fast retransmit can work).
3064 if (th->th_seq == tp->rcv_nxt &&
3065 LIST_EMPTY(&tp->t_segq) &&
3066 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3068 if (DELAY_ACK(tp, tlen) || tfo_syn)
3069 tp->t_flags |= TF_DELACK;
3071 tp->t_flags |= TF_ACKNOW;
3072 tp->rcv_nxt += tlen;
3073 thflags = th->th_flags & TH_FIN;
3074 TCPSTAT_INC(tcps_rcvpack);
3075 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3076 SOCKBUF_LOCK(&so->so_rcv);
3077 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3080 sbappendstream_locked(&so->so_rcv, m, 0);
3081 /* NB: sorwakeup_locked() does an implicit unlock. */
3082 sorwakeup_locked(so);
3085 * XXX: Due to the header drop above "th" is
3086 * theoretically invalid by now. Fortunately
3087 * m_adj() doesn't actually frees any mbufs
3088 * when trimming from the head.
3090 thflags = tcp_reass(tp, th, &tlen, m);
3091 tp->t_flags |= TF_ACKNOW;
3093 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
3094 tcp_update_sack_list(tp, save_start, save_start + tlen);
3097 * Note the amount of data that peer has sent into
3098 * our window, in order to estimate the sender's
3102 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3103 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3105 len = so->so_rcv.sb_hiwat;
3113 * If FIN is received ACK the FIN and let the user know
3114 * that the connection is closing.
3116 if (thflags & TH_FIN) {
3117 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3120 * If connection is half-synchronized
3121 * (ie NEEDSYN flag on) then delay ACK,
3122 * so it may be piggybacked when SYN is sent.
3123 * Otherwise, since we received a FIN then no
3124 * more input can be expected, send ACK now.
3126 if (tp->t_flags & TF_NEEDSYN)
3127 tp->t_flags |= TF_DELACK;
3129 tp->t_flags |= TF_ACKNOW;
3132 switch (tp->t_state) {
3135 * In SYN_RECEIVED and ESTABLISHED STATES
3136 * enter the CLOSE_WAIT state.
3138 case TCPS_SYN_RECEIVED:
3139 tp->t_starttime = ticks;
3141 case TCPS_ESTABLISHED:
3142 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3146 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3147 * enter the CLOSING state.
3149 case TCPS_FIN_WAIT_1:
3150 tcp_state_change(tp, TCPS_CLOSING);
3154 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3155 * starting the time-wait timer, turning off the other
3158 case TCPS_FIN_WAIT_2:
3159 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
3160 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata "
3161 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
3165 INP_INFO_RUNLOCK(&V_tcbinfo);
3169 if (ti_locked == TI_RLOCKED)
3170 INP_INFO_RUNLOCK(&V_tcbinfo);
3171 ti_locked = TI_UNLOCKED;
3174 if (so->so_options & SO_DEBUG)
3175 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3178 TCP_PROBE3(debug__input, tp, th, m);
3181 * Return any desired output.
3183 if (needoutput || (tp->t_flags & TF_ACKNOW))
3184 (void) tp->t_fb->tfb_tcp_output(tp);
3187 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3188 __func__, ti_locked));
3189 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3190 INP_WLOCK_ASSERT(tp->t_inpcb);
3192 if (tp->t_flags & TF_DELACK) {
3193 tp->t_flags &= ~TF_DELACK;
3194 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3196 INP_WUNLOCK(tp->t_inpcb);
3201 * Generate an ACK dropping incoming segment if it occupies
3202 * sequence space, where the ACK reflects our state.
3204 * We can now skip the test for the RST flag since all
3205 * paths to this code happen after packets containing
3206 * RST have been dropped.
3208 * In the SYN-RECEIVED state, don't send an ACK unless the
3209 * segment we received passes the SYN-RECEIVED ACK test.
3210 * If it fails send a RST. This breaks the loop in the
3211 * "LAND" DoS attack, and also prevents an ACK storm
3212 * between two listening ports that have been sent forged
3213 * SYN segments, each with the source address of the other.
3215 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3216 (SEQ_GT(tp->snd_una, th->th_ack) ||
3217 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3218 rstreason = BANDLIM_RST_OPENPORT;
3222 if (so->so_options & SO_DEBUG)
3223 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3226 TCP_PROBE3(debug__input, tp, th, m);
3227 if (ti_locked == TI_RLOCKED)
3228 INP_INFO_RUNLOCK(&V_tcbinfo);
3229 ti_locked = TI_UNLOCKED;
3231 tp->t_flags |= TF_ACKNOW;
3232 (void) tp->t_fb->tfb_tcp_output(tp);
3233 INP_WUNLOCK(tp->t_inpcb);
3238 if (ti_locked == TI_RLOCKED)
3239 INP_INFO_RUNLOCK(&V_tcbinfo);
3240 ti_locked = TI_UNLOCKED;
3243 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3244 INP_WUNLOCK(tp->t_inpcb);
3246 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3250 if (ti_locked == TI_RLOCKED) {
3251 INP_INFO_RUNLOCK(&V_tcbinfo);
3252 ti_locked = TI_UNLOCKED;
3256 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3260 * Drop space held by incoming segment and return.
3263 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3264 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3267 TCP_PROBE3(debug__input, tp, th, m);
3269 INP_WUNLOCK(tp->t_inpcb);
3277 * Issue RST and make ACK acceptable to originator of segment.
3278 * The mbuf must still include the original packet header.
3282 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3283 int tlen, int rstreason)
3289 struct ip6_hdr *ip6;
3293 INP_WLOCK_ASSERT(tp->t_inpcb);
3296 /* Don't bother if destination was broadcast/multicast. */
3297 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3300 if (mtod(m, struct ip *)->ip_v == 6) {
3301 ip6 = mtod(m, struct ip6_hdr *);
3302 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3303 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3305 /* IPv6 anycast check is done at tcp6_input() */
3308 #if defined(INET) && defined(INET6)
3313 ip = mtod(m, struct ip *);
3314 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3315 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3316 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3317 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3322 /* Perform bandwidth limiting. */
3323 if (badport_bandlim(rstreason) < 0)
3326 /* tcp_respond consumes the mbuf chain. */
3327 if (th->th_flags & TH_ACK) {
3328 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3329 th->th_ack, TH_RST);
3331 if (th->th_flags & TH_SYN)
3333 if (th->th_flags & TH_FIN)
3335 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3336 (tcp_seq)0, TH_RST|TH_ACK);
3344 * Parse TCP options and place in tcpopt.
3347 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3352 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3354 if (opt == TCPOPT_EOL)
3356 if (opt == TCPOPT_NOP)
3362 if (optlen < 2 || optlen > cnt)
3367 if (optlen != TCPOLEN_MAXSEG)
3369 if (!(flags & TO_SYN))
3371 to->to_flags |= TOF_MSS;
3372 bcopy((char *)cp + 2,
3373 (char *)&to->to_mss, sizeof(to->to_mss));
3374 to->to_mss = ntohs(to->to_mss);
3377 if (optlen != TCPOLEN_WINDOW)
3379 if (!(flags & TO_SYN))
3381 to->to_flags |= TOF_SCALE;
3382 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3384 case TCPOPT_TIMESTAMP:
3385 if (optlen != TCPOLEN_TIMESTAMP)
3387 to->to_flags |= TOF_TS;
3388 bcopy((char *)cp + 2,
3389 (char *)&to->to_tsval, sizeof(to->to_tsval));
3390 to->to_tsval = ntohl(to->to_tsval);
3391 bcopy((char *)cp + 6,
3392 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3393 to->to_tsecr = ntohl(to->to_tsecr);
3395 case TCPOPT_SIGNATURE:
3397 * In order to reply to a host which has set the
3398 * TCP_SIGNATURE option in its initial SYN, we have
3399 * to record the fact that the option was observed
3400 * here for the syncache code to perform the correct
3403 if (optlen != TCPOLEN_SIGNATURE)
3405 to->to_flags |= TOF_SIGNATURE;
3406 to->to_signature = cp + 2;
3408 case TCPOPT_SACK_PERMITTED:
3409 if (optlen != TCPOLEN_SACK_PERMITTED)
3411 if (!(flags & TO_SYN))
3415 to->to_flags |= TOF_SACKPERM;
3418 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3422 to->to_flags |= TOF_SACK;
3423 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3424 to->to_sacks = cp + 2;
3425 TCPSTAT_INC(tcps_sack_rcv_blocks);
3428 case TCPOPT_FAST_OPEN:
3429 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) &&
3430 (optlen < TCPOLEN_FAST_OPEN_MIN) &&
3431 (optlen > TCPOLEN_FAST_OPEN_MAX))
3433 if (!(flags & TO_SYN))
3435 if (!V_tcp_fastopen_enabled)
3437 to->to_flags |= TOF_FASTOPEN;
3438 to->to_tfo_len = optlen - 2;
3439 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3449 * Pull out of band byte out of a segment so
3450 * it doesn't appear in the user's data queue.
3451 * It is still reflected in the segment length for
3452 * sequencing purposes.
3455 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3458 int cnt = off + th->th_urp - 1;
3461 if (m->m_len > cnt) {
3462 char *cp = mtod(m, caddr_t) + cnt;
3463 struct tcpcb *tp = sototcpcb(so);
3465 INP_WLOCK_ASSERT(tp->t_inpcb);
3468 tp->t_oobflags |= TCPOOB_HAVEDATA;
3469 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3471 if (m->m_flags & M_PKTHDR)
3480 panic("tcp_pulloutofband");
3484 * Collect new round-trip time estimate
3485 * and update averages and current timeout.
3488 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3492 INP_WLOCK_ASSERT(tp->t_inpcb);
3494 TCPSTAT_INC(tcps_rttupdated);
3496 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3498 * srtt is stored as fixed point with 5 bits after the
3499 * binary point (i.e., scaled by 8). The following magic
3500 * is equivalent to the smoothing algorithm in rfc793 with
3501 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3502 * point). Adjust rtt to origin 0.
3504 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3505 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3507 if ((tp->t_srtt += delta) <= 0)
3511 * We accumulate a smoothed rtt variance (actually, a
3512 * smoothed mean difference), then set the retransmit
3513 * timer to smoothed rtt + 4 times the smoothed variance.
3514 * rttvar is stored as fixed point with 4 bits after the
3515 * binary point (scaled by 16). The following is
3516 * equivalent to rfc793 smoothing with an alpha of .75
3517 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3518 * rfc793's wired-in beta.
3522 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3523 if ((tp->t_rttvar += delta) <= 0)
3525 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3526 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3529 * No rtt measurement yet - use the unsmoothed rtt.
3530 * Set the variance to half the rtt (so our first
3531 * retransmit happens at 3*rtt).
3533 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3534 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3535 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3541 * the retransmit should happen at rtt + 4 * rttvar.
3542 * Because of the way we do the smoothing, srtt and rttvar
3543 * will each average +1/2 tick of bias. When we compute
3544 * the retransmit timer, we want 1/2 tick of rounding and
3545 * 1 extra tick because of +-1/2 tick uncertainty in the
3546 * firing of the timer. The bias will give us exactly the
3547 * 1.5 tick we need. But, because the bias is
3548 * statistical, we have to test that we don't drop below
3549 * the minimum feasible timer (which is 2 ticks).
3551 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3552 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3555 * We received an ack for a packet that wasn't retransmitted;
3556 * it is probably safe to discard any error indications we've
3557 * received recently. This isn't quite right, but close enough
3558 * for now (a route might have failed after we sent a segment,
3559 * and the return path might not be symmetrical).
3561 tp->t_softerror = 0;
3565 * Determine a reasonable value for maxseg size.
3566 * If the route is known, check route for mtu.
3567 * If none, use an mss that can be handled on the outgoing interface
3568 * without forcing IP to fragment. If no route is found, route has no mtu,
3569 * or the destination isn't local, use a default, hopefully conservative
3570 * size (usually 512 or the default IP max size, but no more than the mtu
3571 * of the interface), as we can't discover anything about intervening
3572 * gateways or networks. We also initialize the congestion/slow start
3573 * window to be a single segment if the destination isn't local.
3574 * While looking at the routing entry, we also initialize other path-dependent
3575 * parameters from pre-set or cached values in the routing entry.
3577 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3578 * IP options, e.g. IPSEC data, since length of this data may vary, and
3579 * thus it is calculated for every segment separately in tcp_output().
3581 * NOTE that this routine is only called when we process an incoming
3582 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3583 * settings are handled in tcp_mssopt().
3586 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3587 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3590 uint32_t maxmtu = 0;
3591 struct inpcb *inp = tp->t_inpcb;
3592 struct hc_metrics_lite metrics;
3594 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3595 size_t min_protoh = isipv6 ?
3596 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3597 sizeof (struct tcpiphdr);
3599 const size_t min_protoh = sizeof(struct tcpiphdr);
3602 INP_WLOCK_ASSERT(tp->t_inpcb);
3604 if (mtuoffer != -1) {
3605 KASSERT(offer == -1, ("%s: conflict", __func__));
3606 offer = mtuoffer - min_protoh;
3612 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3613 tp->t_maxseg = V_tcp_v6mssdflt;
3616 #if defined(INET) && defined(INET6)
3621 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3622 tp->t_maxseg = V_tcp_mssdflt;
3627 * No route to sender, stay with default mss and return.
3631 * In case we return early we need to initialize metrics
3632 * to a defined state as tcp_hc_get() would do for us
3633 * if there was no cache hit.
3635 if (metricptr != NULL)
3636 bzero(metricptr, sizeof(struct hc_metrics_lite));
3640 /* What have we got? */
3644 * Offer == 0 means that there was no MSS on the SYN
3645 * segment, in this case we use tcp_mssdflt as
3646 * already assigned to t_maxseg above.
3648 offer = tp->t_maxseg;
3653 * Offer == -1 means that we didn't receive SYN yet.
3659 * Prevent DoS attack with too small MSS. Round up
3660 * to at least minmss.
3662 offer = max(offer, V_tcp_minmss);
3666 * rmx information is now retrieved from tcp_hostcache.
3668 tcp_hc_get(&inp->inp_inc, &metrics);
3669 if (metricptr != NULL)
3670 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3673 * If there's a discovered mtu in tcp hostcache, use it.
3674 * Else, use the link mtu.
3676 if (metrics.rmx_mtu)
3677 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3681 mss = maxmtu - min_protoh;
3682 if (!V_path_mtu_discovery &&
3683 !in6_localaddr(&inp->in6p_faddr))
3684 mss = min(mss, V_tcp_v6mssdflt);
3687 #if defined(INET) && defined(INET6)
3692 mss = maxmtu - min_protoh;
3693 if (!V_path_mtu_discovery &&
3694 !in_localaddr(inp->inp_faddr))
3695 mss = min(mss, V_tcp_mssdflt);
3699 * XXX - The above conditional (mss = maxmtu - min_protoh)
3700 * probably violates the TCP spec.
3701 * The problem is that, since we don't know the
3702 * other end's MSS, we are supposed to use a conservative
3703 * default. But, if we do that, then MTU discovery will
3704 * never actually take place, because the conservative
3705 * default is much less than the MTUs typically seen
3706 * on the Internet today. For the moment, we'll sweep
3707 * this under the carpet.
3709 * The conservative default might not actually be a problem
3710 * if the only case this occurs is when sending an initial
3711 * SYN with options and data to a host we've never talked
3712 * to before. Then, they will reply with an MSS value which
3713 * will get recorded and the new parameters should get
3714 * recomputed. For Further Study.
3717 mss = min(mss, offer);
3720 * Sanity check: make sure that maxseg will be large
3721 * enough to allow some data on segments even if the
3722 * all the option space is used (40bytes). Otherwise
3723 * funny things may happen in tcp_output.
3725 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3733 tcp_mss(struct tcpcb *tp, int offer)
3739 struct hc_metrics_lite metrics;
3740 struct tcp_ifcap cap;
3742 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3744 bzero(&cap, sizeof(cap));
3745 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3751 * If there's a pipesize, change the socket buffer to that size,
3752 * don't change if sb_hiwat is different than default (then it
3753 * has been changed on purpose with setsockopt).
3754 * Make the socket buffers an integral number of mss units;
3755 * if the mss is larger than the socket buffer, decrease the mss.
3757 so = inp->inp_socket;
3758 SOCKBUF_LOCK(&so->so_snd);
3759 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3760 bufsize = metrics.rmx_sendpipe;
3762 bufsize = so->so_snd.sb_hiwat;
3766 bufsize = roundup(bufsize, mss);
3767 if (bufsize > sb_max)
3769 if (bufsize > so->so_snd.sb_hiwat)
3770 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3772 SOCKBUF_UNLOCK(&so->so_snd);
3774 * Sanity check: make sure that maxseg will be large
3775 * enough to allow some data on segments even if the
3776 * all the option space is used (40bytes). Otherwise
3777 * funny things may happen in tcp_output.
3779 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3781 tp->t_maxseg = max(mss, 64);
3783 SOCKBUF_LOCK(&so->so_rcv);
3784 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3785 bufsize = metrics.rmx_recvpipe;
3787 bufsize = so->so_rcv.sb_hiwat;
3788 if (bufsize > mss) {
3789 bufsize = roundup(bufsize, mss);
3790 if (bufsize > sb_max)
3792 if (bufsize > so->so_rcv.sb_hiwat)
3793 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3795 SOCKBUF_UNLOCK(&so->so_rcv);
3797 /* Check the interface for TSO capabilities. */
3798 if (cap.ifcap & CSUM_TSO) {
3799 tp->t_flags |= TF_TSO;
3800 tp->t_tsomax = cap.tsomax;
3801 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3802 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3807 * Determine the MSS option to send on an outgoing SYN.
3810 tcp_mssopt(struct in_conninfo *inc)
3813 uint32_t thcmtu = 0;
3814 uint32_t maxmtu = 0;
3817 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3820 if (inc->inc_flags & INC_ISIPV6) {
3821 mss = V_tcp_v6mssdflt;
3822 maxmtu = tcp_maxmtu6(inc, NULL);
3823 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3826 #if defined(INET) && defined(INET6)
3831 mss = V_tcp_mssdflt;
3832 maxmtu = tcp_maxmtu(inc, NULL);
3833 min_protoh = sizeof(struct tcpiphdr);
3836 #if defined(INET6) || defined(INET)
3837 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3840 if (maxmtu && thcmtu)
3841 mss = min(maxmtu, thcmtu) - min_protoh;
3842 else if (maxmtu || thcmtu)
3843 mss = max(maxmtu, thcmtu) - min_protoh;
3850 * On a partial ack arrives, force the retransmission of the
3851 * next unacknowledged segment. Do not clear tp->t_dupacks.
3852 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3856 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3858 tcp_seq onxt = tp->snd_nxt;
3859 uint32_t ocwnd = tp->snd_cwnd;
3860 u_int maxseg = tcp_maxseg(tp);
3862 INP_WLOCK_ASSERT(tp->t_inpcb);
3864 tcp_timer_activate(tp, TT_REXMT, 0);
3866 tp->snd_nxt = th->th_ack;
3868 * Set snd_cwnd to one segment beyond acknowledged offset.
3869 * (tp->snd_una has not yet been updated when this function is called.)
3871 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3872 tp->t_flags |= TF_ACKNOW;
3873 (void) tp->t_fb->tfb_tcp_output(tp);
3874 tp->snd_cwnd = ocwnd;
3875 if (SEQ_GT(onxt, tp->snd_nxt))
3878 * Partial window deflation. Relies on fact that tp->snd_una
3881 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3882 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3885 tp->snd_cwnd += maxseg;
3889 tcp_compute_pipe(struct tcpcb *tp)
3891 return (tp->snd_max - tp->snd_una +
3892 tp->sackhint.sack_bytes_rexmit -
3893 tp->sackhint.sacked_bytes);