2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (c) 2007-2008,2010
5 * Swinburne University of Technology, Melbourne, Australia.
6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
7 * Copyright (c) 2010 The FreeBSD Foundation
8 * Copyright (c) 2010-2011 Juniper Networks, Inc.
11 * Portions of this software were developed at the Centre for Advanced Internet
12 * Architectures, Swinburne University of Technology, by Lawrence Stewart,
13 * James Healy and David Hayes, made possible in part by a grant from the Cisco
14 * University Research Program Fund at Community Foundation Silicon Valley.
16 * Portions of this software were developed at the Centre for Advanced
17 * Internet Architectures, Swinburne University of Technology, Melbourne,
18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
20 * Portions of this software were developed by Robert N. M. Watson under
21 * contract to Juniper Networks, Inc.
23 * Redistribution and use in source and binary forms, with or without
24 * modification, are permitted provided that the following conditions
26 * 1. Redistributions of source code must retain the above copyright
27 * notice, this list of conditions and the following disclaimer.
28 * 2. Redistributions in binary form must reproduce the above copyright
29 * notice, this list of conditions and the following disclaimer in the
30 * documentation and/or other materials provided with the distribution.
31 * 4. Neither the name of the University nor the names of its contributors
32 * may be used to endorse or promote products derived from this software
33 * without specific prior written permission.
35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD$");
54 #include "opt_inet6.h"
55 #include "opt_ipsec.h"
56 #include "opt_tcpdebug.h"
58 #include <sys/param.h>
59 #include <sys/kernel.h>
60 #include <sys/hhook.h>
61 #include <sys/malloc.h>
63 #include <sys/proc.h> /* for proc0 declaration */
64 #include <sys/protosw.h>
66 #include <sys/signalvar.h>
67 #include <sys/socket.h>
68 #include <sys/socketvar.h>
69 #include <sys/sysctl.h>
70 #include <sys/syslog.h>
71 #include <sys/systm.h>
73 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
78 #include <net/if_var.h>
79 #include <net/route.h>
82 #define TCPSTATES /* for logging */
84 #include <netinet/in.h>
85 #include <netinet/in_kdtrace.h>
86 #include <netinet/in_pcb.h>
87 #include <netinet/in_systm.h>
88 #include <netinet/ip.h>
89 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
90 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
91 #include <netinet/ip_var.h>
92 #include <netinet/ip_options.h>
93 #include <netinet/ip6.h>
94 #include <netinet/icmp6.h>
95 #include <netinet6/in6_pcb.h>
96 #include <netinet6/in6_var.h>
97 #include <netinet6/ip6_var.h>
98 #include <netinet6/nd6.h>
100 #include <netinet/tcp_fastopen.h>
102 #include <netinet/tcp.h>
103 #include <netinet/tcp_fsm.h>
104 #include <netinet/tcp_seq.h>
105 #include <netinet/tcp_timer.h>
106 #include <netinet/tcp_var.h>
107 #include <netinet6/tcp6_var.h>
108 #include <netinet/tcpip.h>
109 #include <netinet/cc/cc.h>
111 #include <netinet/tcp_pcap.h>
113 #include <netinet/tcp_syncache.h>
115 #include <netinet/tcp_debug.h>
116 #endif /* TCPDEBUG */
118 #include <netinet/tcp_offload.h>
121 #include <netipsec/ipsec_support.h>
123 #include <machine/in_cksum.h>
125 #include <security/mac/mac_framework.h>
127 const int tcprexmtthresh = 3;
129 int tcp_log_in_vain = 0;
130 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
132 "Log all incoming TCP segments to closed ports");
134 VNET_DEFINE(int, blackhole) = 0;
135 #define V_blackhole VNET(blackhole)
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
137 &VNET_NAME(blackhole), 0,
138 "Do not send RST on segments to closed ports");
140 VNET_DEFINE(int, tcp_delack_enabled) = 1;
141 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
142 &VNET_NAME(tcp_delack_enabled), 0,
143 "Delay ACK to try and piggyback it onto a data packet");
145 VNET_DEFINE(int, drop_synfin) = 0;
146 #define V_drop_synfin VNET(drop_synfin)
147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
148 &VNET_NAME(drop_synfin), 0,
149 "Drop TCP packets with SYN+FIN set");
151 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
153 &VNET_NAME(tcp_do_rfc6675_pipe), 0,
154 "Use calculated pipe/in-flight bytes per RFC 6675");
156 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
157 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
159 &VNET_NAME(tcp_do_rfc3042), 0,
160 "Enable RFC 3042 (Limited Transmit)");
162 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
164 &VNET_NAME(tcp_do_rfc3390), 0,
165 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
167 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
169 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
170 "Slow-start flight size (initial congestion window) in number of segments");
172 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
174 &VNET_NAME(tcp_do_rfc3465), 0,
175 "Enable RFC 3465 (Appropriate Byte Counting)");
177 VNET_DEFINE(int, tcp_abc_l_var) = 2;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
179 &VNET_NAME(tcp_abc_l_var), 2,
180 "Cap the max cwnd increment during slow-start to this number of segments");
182 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
184 VNET_DEFINE(int, tcp_do_ecn) = 2;
185 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
186 &VNET_NAME(tcp_do_ecn), 0,
189 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
190 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
191 &VNET_NAME(tcp_ecn_maxretries), 0,
192 "Max retries before giving up on ECN");
194 VNET_DEFINE(int, tcp_insecure_syn) = 0;
195 #define V_tcp_insecure_syn VNET(tcp_insecure_syn)
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
197 &VNET_NAME(tcp_insecure_syn), 0,
198 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
200 VNET_DEFINE(int, tcp_insecure_rst) = 0;
201 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
203 &VNET_NAME(tcp_insecure_rst), 0,
204 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
206 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
207 #define V_tcp_recvspace VNET(tcp_recvspace)
208 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
209 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
211 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
212 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
213 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
214 &VNET_NAME(tcp_do_autorcvbuf), 0,
215 "Enable automatic receive buffer sizing");
217 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
218 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
219 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW,
220 &VNET_NAME(tcp_autorcvbuf_inc), 0,
221 "Incrementor step size of automatic receive buffer");
223 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
224 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
225 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
226 &VNET_NAME(tcp_autorcvbuf_max), 0,
227 "Max size of automatic receive buffer");
229 VNET_DEFINE(struct inpcbhead, tcb);
230 #define tcb6 tcb /* for KAME src sync over BSD*'s */
231 VNET_DEFINE(struct inpcbinfo, tcbinfo);
234 * TCP statistics are stored in an array of counter(9)s, which size matches
235 * size of struct tcpstat. TCP running connection count is a regular array.
237 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
238 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
239 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
240 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
241 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
242 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
243 "TCP connection counts by TCP state");
246 tcp_vnet_init(const void *unused)
249 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
250 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
252 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
253 tcp_vnet_init, NULL);
257 tcp_vnet_uninit(const void *unused)
260 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
261 VNET_PCPUSTAT_FREE(tcpstat);
263 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
264 tcp_vnet_uninit, NULL);
268 * Kernel module interface for updating tcpstat. The argument is an index
269 * into tcpstat treated as an array.
272 kmod_tcpstat_inc(int statnum)
275 counter_u64_add(VNET(tcpstat)[statnum], 1);
279 * Wrapper for the TCP established input helper hook.
282 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
284 struct tcp_hhook_data hhook_data;
286 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
291 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
297 * CC wrapper hook functions
300 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type)
302 INP_WLOCK_ASSERT(tp->t_inpcb);
304 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
305 if (tp->snd_cwnd <= tp->snd_wnd)
306 tp->ccv->flags |= CCF_CWND_LIMITED;
308 tp->ccv->flags &= ~CCF_CWND_LIMITED;
310 if (type == CC_ACK) {
311 if (tp->snd_cwnd > tp->snd_ssthresh) {
312 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
313 V_tcp_abc_l_var * tcp_maxseg(tp));
314 if (tp->t_bytes_acked >= tp->snd_cwnd) {
315 tp->t_bytes_acked -= tp->snd_cwnd;
316 tp->ccv->flags |= CCF_ABC_SENTAWND;
319 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
320 tp->t_bytes_acked = 0;
324 if (CC_ALGO(tp)->ack_received != NULL) {
325 /* XXXLAS: Find a way to live without this */
326 tp->ccv->curack = th->th_ack;
327 CC_ALGO(tp)->ack_received(tp->ccv, type);
332 cc_conn_init(struct tcpcb *tp)
334 struct hc_metrics_lite metrics;
335 struct inpcb *inp = tp->t_inpcb;
339 INP_WLOCK_ASSERT(tp->t_inpcb);
341 tcp_hc_get(&inp->inp_inc, &metrics);
342 maxseg = tcp_maxseg(tp);
344 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
346 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
347 TCPSTAT_INC(tcps_usedrtt);
348 if (metrics.rmx_rttvar) {
349 tp->t_rttvar = metrics.rmx_rttvar;
350 TCPSTAT_INC(tcps_usedrttvar);
352 /* default variation is +- 1 rtt */
354 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
356 TCPT_RANGESET(tp->t_rxtcur,
357 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
358 tp->t_rttmin, TCPTV_REXMTMAX);
360 if (metrics.rmx_ssthresh) {
362 * There's some sort of gateway or interface
363 * buffer limit on the path. Use this to set
364 * the slow start threshold, but set the
365 * threshold to no less than 2*mss.
367 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
368 TCPSTAT_INC(tcps_usedssthresh);
372 * Set the initial slow-start flight size.
374 * RFC5681 Section 3.1 specifies the default conservative values.
375 * RFC3390 specifies slightly more aggressive values.
376 * RFC6928 increases it to ten segments.
377 * Support for user specified value for initial flight size.
379 * If a SYN or SYN/ACK was lost and retransmitted, we have to
380 * reduce the initial CWND to one segment as congestion is likely
381 * requiring us to be cautious.
383 if (tp->snd_cwnd == 1)
384 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
385 else if (V_tcp_initcwnd_segments)
386 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg,
387 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
388 else if (V_tcp_do_rfc3390)
389 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380));
391 /* Per RFC5681 Section 3.1 */
393 tp->snd_cwnd = 2 * maxseg;
394 else if (maxseg > 1095)
395 tp->snd_cwnd = 3 * maxseg;
397 tp->snd_cwnd = 4 * maxseg;
400 if (CC_ALGO(tp)->conn_init != NULL)
401 CC_ALGO(tp)->conn_init(tp->ccv);
405 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
409 INP_WLOCK_ASSERT(tp->t_inpcb);
413 if (!IN_FASTRECOVERY(tp->t_flags)) {
414 tp->snd_recover = tp->snd_max;
415 if (tp->t_flags & TF_ECN_PERMIT)
416 tp->t_flags |= TF_ECN_SND_CWR;
420 if (!IN_CONGRECOVERY(tp->t_flags)) {
421 TCPSTAT_INC(tcps_ecn_rcwnd);
422 tp->snd_recover = tp->snd_max;
423 if (tp->t_flags & TF_ECN_PERMIT)
424 tp->t_flags |= TF_ECN_SND_CWR;
428 maxseg = tcp_maxseg(tp);
430 tp->t_bytes_acked = 0;
431 EXIT_RECOVERY(tp->t_flags);
432 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
434 tp->snd_cwnd = maxseg;
437 TCPSTAT_INC(tcps_sndrexmitbad);
438 /* RTO was unnecessary, so reset everything. */
439 tp->snd_cwnd = tp->snd_cwnd_prev;
440 tp->snd_ssthresh = tp->snd_ssthresh_prev;
441 tp->snd_recover = tp->snd_recover_prev;
442 if (tp->t_flags & TF_WASFRECOVERY)
443 ENTER_FASTRECOVERY(tp->t_flags);
444 if (tp->t_flags & TF_WASCRECOVERY)
445 ENTER_CONGRECOVERY(tp->t_flags);
446 tp->snd_nxt = tp->snd_max;
447 tp->t_flags &= ~TF_PREVVALID;
452 if (CC_ALGO(tp)->cong_signal != NULL) {
454 tp->ccv->curack = th->th_ack;
455 CC_ALGO(tp)->cong_signal(tp->ccv, type);
460 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
462 INP_WLOCK_ASSERT(tp->t_inpcb);
464 /* XXXLAS: KASSERT that we're in recovery? */
466 if (CC_ALGO(tp)->post_recovery != NULL) {
467 tp->ccv->curack = th->th_ack;
468 CC_ALGO(tp)->post_recovery(tp->ccv);
470 /* XXXLAS: EXIT_RECOVERY ? */
471 tp->t_bytes_acked = 0;
475 * Indicate whether this ack should be delayed. We can delay the ack if
476 * following conditions are met:
477 * - There is no delayed ack timer in progress.
478 * - Our last ack wasn't a 0-sized window. We never want to delay
479 * the ack that opens up a 0-sized window.
480 * - LRO wasn't used for this segment. We make sure by checking that the
481 * segment size is not larger than the MSS.
483 #define DELAY_ACK(tp, tlen) \
484 ((!tcp_timer_active(tp, TT_DELACK) && \
485 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
486 (tlen <= tp->t_maxseg) && \
487 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
490 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
492 INP_WLOCK_ASSERT(tp->t_inpcb);
494 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
495 switch (iptos & IPTOS_ECN_MASK) {
497 tp->ccv->flags |= CCF_IPHDR_CE;
500 tp->ccv->flags &= ~CCF_IPHDR_CE;
503 tp->ccv->flags &= ~CCF_IPHDR_CE;
507 if (th->th_flags & TH_CWR)
508 tp->ccv->flags |= CCF_TCPHDR_CWR;
510 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
512 if (tp->t_flags & TF_DELACK)
513 tp->ccv->flags |= CCF_DELACK;
515 tp->ccv->flags &= ~CCF_DELACK;
517 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
519 if (tp->ccv->flags & CCF_ACKNOW)
520 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
525 * TCP input handling is split into multiple parts:
526 * tcp6_input is a thin wrapper around tcp_input for the extended
527 * ip6_protox[] call format in ip6_input
528 * tcp_input handles primary segment validation, inpcb lookup and
529 * SYN processing on listen sockets
530 * tcp_do_segment processes the ACK and text of the segment for
531 * establishing, established and closing connections
535 tcp6_input(struct mbuf **mp, int *offp, int proto)
537 struct mbuf *m = *mp;
538 struct in6_ifaddr *ia6;
541 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
544 * draft-itojun-ipv6-tcp-to-anycast
545 * better place to put this in?
547 ip6 = mtod(m, struct ip6_hdr *);
548 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
549 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
552 ifa_free(&ia6->ia_ifa);
553 ip6 = mtod(m, struct ip6_hdr *);
554 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
555 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
556 return (IPPROTO_DONE);
559 ifa_free(&ia6->ia_ifa);
561 return (tcp_input(mp, offp, proto));
566 tcp_input(struct mbuf **mp, int *offp, int proto)
568 struct mbuf *m = *mp;
569 struct tcphdr *th = NULL;
570 struct ip *ip = NULL;
571 struct inpcb *inp = NULL;
572 struct tcpcb *tp = NULL;
573 struct socket *so = NULL;
583 int rstreason = 0; /* For badport_bandlim accounting purposes */
585 struct m_tag *fwd_tag = NULL;
587 struct ip6_hdr *ip6 = NULL;
590 const void *ip6 = NULL;
592 struct tcpopt to; /* options in this segment */
593 char *s = NULL; /* address and port logging */
597 * The size of tcp_saveipgen must be the size of the max ip header,
600 u_char tcp_saveipgen[IP6_HDR_LEN];
601 struct tcphdr tcp_savetcp;
606 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
613 TCPSTAT_INC(tcps_rcvtotal);
617 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
619 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
620 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
622 TCPSTAT_INC(tcps_rcvshort);
623 return (IPPROTO_DONE);
627 ip6 = mtod(m, struct ip6_hdr *);
628 th = (struct tcphdr *)((caddr_t)ip6 + off0);
629 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
630 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
631 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
632 th->th_sum = m->m_pkthdr.csum_data;
634 th->th_sum = in6_cksum_pseudo(ip6, tlen,
635 IPPROTO_TCP, m->m_pkthdr.csum_data);
636 th->th_sum ^= 0xffff;
638 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
640 TCPSTAT_INC(tcps_rcvbadsum);
645 * Be proactive about unspecified IPv6 address in source.
646 * As we use all-zero to indicate unbounded/unconnected pcb,
647 * unspecified IPv6 address can be used to confuse us.
649 * Note that packets with unspecified IPv6 destination is
650 * already dropped in ip6_input.
652 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
656 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
659 #if defined(INET) && defined(INET6)
665 * Get IP and TCP header together in first mbuf.
666 * Note: IP leaves IP header in first mbuf.
668 if (off0 > sizeof (struct ip)) {
670 off0 = sizeof(struct ip);
672 if (m->m_len < sizeof (struct tcpiphdr)) {
673 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
675 TCPSTAT_INC(tcps_rcvshort);
676 return (IPPROTO_DONE);
679 ip = mtod(m, struct ip *);
680 th = (struct tcphdr *)((caddr_t)ip + off0);
681 tlen = ntohs(ip->ip_len) - off0;
684 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
685 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
686 th->th_sum = m->m_pkthdr.csum_data;
688 th->th_sum = in_pseudo(ip->ip_src.s_addr,
690 htonl(m->m_pkthdr.csum_data + tlen +
692 th->th_sum ^= 0xffff;
694 struct ipovly *ipov = (struct ipovly *)ip;
697 * Checksum extended TCP header and data.
700 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
701 ipov->ih_len = htons(tlen);
702 th->th_sum = in_cksum(m, len);
703 /* Reset length for SDT probes. */
704 ip->ip_len = htons(len);
707 /* Re-initialization for later version check */
708 ip->ip_v = IPVERSION;
709 ip->ip_hl = off0 >> 2;
713 TCPSTAT_INC(tcps_rcvbadsum);
720 * Check that TCP offset makes sense,
721 * pull out TCP options and adjust length. XXX
723 off = th->th_off << 2;
724 if (off < sizeof (struct tcphdr) || off > tlen) {
725 TCPSTAT_INC(tcps_rcvbadoff);
728 tlen -= off; /* tlen is used instead of ti->ti_len */
729 if (off > sizeof (struct tcphdr)) {
732 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
733 ip6 = mtod(m, struct ip6_hdr *);
734 th = (struct tcphdr *)((caddr_t)ip6 + off0);
737 #if defined(INET) && defined(INET6)
742 if (m->m_len < sizeof(struct ip) + off) {
743 if ((m = m_pullup(m, sizeof (struct ip) + off))
745 TCPSTAT_INC(tcps_rcvshort);
746 return (IPPROTO_DONE);
748 ip = mtod(m, struct ip *);
749 th = (struct tcphdr *)((caddr_t)ip + off0);
753 optlen = off - sizeof (struct tcphdr);
754 optp = (u_char *)(th + 1);
756 thflags = th->th_flags;
759 * Convert TCP protocol specific fields to host format.
761 tcp_fields_to_host(th);
764 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
766 drop_hdrlen = off0 + off;
769 * Locate pcb for segment; if we're likely to add or remove a
770 * connection then first acquire pcbinfo lock. There are three cases
771 * where we might discover later we need a write lock despite the
772 * flags: ACKs moving a connection out of the syncache, ACKs for a
773 * connection in TIMEWAIT and SYNs not targeting a listening socket.
775 if ((thflags & (TH_FIN | TH_RST)) != 0) {
776 INP_INFO_RLOCK(&V_tcbinfo);
777 ti_locked = TI_RLOCKED;
779 ti_locked = TI_UNLOCKED;
782 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
786 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
788 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
791 #if defined(INET) && !defined(INET6)
792 (m->m_flags & M_IP_NEXTHOP)
795 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
799 if (ti_locked == TI_RLOCKED) {
800 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
802 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
806 if (isipv6 && fwd_tag != NULL) {
807 struct sockaddr_in6 *next_hop6;
809 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
811 * Transparently forwarded. Pretend to be the destination.
812 * Already got one like this?
814 inp = in6_pcblookup_mbuf(&V_tcbinfo,
815 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
816 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
819 * It's new. Try to find the ambushing socket.
820 * Because we've rewritten the destination address,
821 * any hardware-generated hash is ignored.
823 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
824 th->th_sport, &next_hop6->sin6_addr,
825 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
826 th->th_dport, INPLOOKUP_WILDCARD |
827 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
830 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
831 th->th_sport, &ip6->ip6_dst, th->th_dport,
832 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
833 m->m_pkthdr.rcvif, m);
836 #if defined(INET6) && defined(INET)
840 if (fwd_tag != NULL) {
841 struct sockaddr_in *next_hop;
843 next_hop = (struct sockaddr_in *)(fwd_tag+1);
845 * Transparently forwarded. Pretend to be the destination.
846 * already got one like this?
848 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
849 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
850 m->m_pkthdr.rcvif, m);
853 * It's new. Try to find the ambushing socket.
854 * Because we've rewritten the destination address,
855 * any hardware-generated hash is ignored.
857 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
858 th->th_sport, next_hop->sin_addr,
859 next_hop->sin_port ? ntohs(next_hop->sin_port) :
860 th->th_dport, INPLOOKUP_WILDCARD |
861 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
864 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
865 th->th_sport, ip->ip_dst, th->th_dport,
866 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
867 m->m_pkthdr.rcvif, m);
871 * If the INPCB does not exist then all data in the incoming
872 * segment is discarded and an appropriate RST is sent back.
873 * XXX MRT Send RST using which routing table?
877 * Log communication attempts to ports that are not
880 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
881 tcp_log_in_vain == 2) {
882 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
883 log(LOG_INFO, "%s; %s: Connection attempt "
884 "to closed port\n", s, __func__);
887 * When blackholing do not respond with a RST but
888 * completely ignore the segment and drop it.
890 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
894 rstreason = BANDLIM_RST_CLOSEDPORT;
897 INP_WLOCK_ASSERT(inp);
899 * While waiting for inp lock during the lookup, another thread
900 * can have dropped the inpcb, in which case we need to loop back
901 * and try to find a new inpcb to deliver to.
903 if (inp->inp_flags & INP_DROPPED) {
908 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
909 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
910 ((inp->inp_socket == NULL) ||
911 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
912 inp->inp_flowid = m->m_pkthdr.flowid;
913 inp->inp_flowtype = M_HASHTYPE_GET(m);
915 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
917 if (isipv6 && IPSEC_ENABLED(ipv6) &&
918 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
926 if (IPSEC_ENABLED(ipv4) &&
927 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
934 * Check the minimum TTL for socket.
936 if (inp->inp_ip_minttl != 0) {
939 if (inp->inp_ip_minttl > ip6->ip6_hlim)
943 if (inp->inp_ip_minttl > ip->ip_ttl)
948 * A previous connection in TIMEWAIT state is supposed to catch stray
949 * or duplicate segments arriving late. If this segment was a
950 * legitimate new connection attempt, the old INPCB gets removed and
951 * we can try again to find a listening socket.
953 * At this point, due to earlier optimism, we may hold only an inpcb
954 * lock, and not the inpcbinfo write lock. If so, we need to try to
955 * acquire it, or if that fails, acquire a reference on the inpcb,
956 * drop all locks, acquire a global write lock, and then re-acquire
957 * the inpcb lock. We may at that point discover that another thread
958 * has tried to free the inpcb, in which case we need to loop back
959 * and try to find a new inpcb to deliver to.
961 * XXXRW: It may be time to rethink timewait locking.
964 if (inp->inp_flags & INP_TIMEWAIT) {
965 if (ti_locked == TI_UNLOCKED) {
966 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
969 INP_INFO_RLOCK(&V_tcbinfo);
970 ti_locked = TI_RLOCKED;
972 if (in_pcbrele_wlocked(inp)) {
975 } else if (inp->inp_flags & INP_DROPPED) {
981 ti_locked = TI_RLOCKED;
983 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
985 if (thflags & TH_SYN)
986 tcp_dooptions(&to, optp, optlen, TO_SYN);
988 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
990 if (tcp_twcheck(inp, &to, th, m, tlen))
992 INP_INFO_RUNLOCK(&V_tcbinfo);
993 return (IPPROTO_DONE);
996 * The TCPCB may no longer exist if the connection is winding
997 * down or it is in the CLOSED state. Either way we drop the
998 * segment and send an appropriate response.
1000 tp = intotcpcb(inp);
1001 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1002 rstreason = BANDLIM_RST_CLOSEDPORT;
1007 if (tp->t_flags & TF_TOE) {
1008 tcp_offload_input(tp, m);
1009 m = NULL; /* consumed by the TOE driver */
1015 * We've identified a valid inpcb, but it could be that we need an
1016 * inpcbinfo write lock but don't hold it. In this case, attempt to
1017 * acquire using the same strategy as the TIMEWAIT case above. If we
1018 * relock, we have to jump back to 'relocked' as the connection might
1019 * now be in TIMEWAIT.
1022 if ((thflags & (TH_FIN | TH_RST)) != 0)
1023 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1025 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) ||
1026 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) &&
1027 !(tp->t_flags & TF_FASTOPEN)))) {
1028 if (ti_locked == TI_UNLOCKED) {
1029 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
1032 INP_INFO_RLOCK(&V_tcbinfo);
1033 ti_locked = TI_RLOCKED;
1035 if (in_pcbrele_wlocked(inp)) {
1038 } else if (inp->inp_flags & INP_DROPPED) {
1045 ti_locked = TI_RLOCKED;
1047 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1051 INP_WLOCK_ASSERT(inp);
1052 if (mac_inpcb_check_deliver(inp, m))
1055 so = inp->inp_socket;
1056 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1058 if (so->so_options & SO_DEBUG) {
1059 ostate = tp->t_state;
1062 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1065 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1068 #endif /* TCPDEBUG */
1070 * When the socket is accepting connections (the INPCB is in LISTEN
1071 * state) we look into the SYN cache if this is a new connection
1072 * attempt or the completion of a previous one.
1074 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1075 ("%s: so accepting but tp %p not listening", __func__, tp));
1076 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1077 struct in_conninfo inc;
1079 bzero(&inc, sizeof(inc));
1082 inc.inc_flags |= INC_ISIPV6;
1083 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1084 inc.inc_flags |= INC_IPV6MINMTU;
1085 inc.inc6_faddr = ip6->ip6_src;
1086 inc.inc6_laddr = ip6->ip6_dst;
1090 inc.inc_faddr = ip->ip_src;
1091 inc.inc_laddr = ip->ip_dst;
1093 inc.inc_fport = th->th_sport;
1094 inc.inc_lport = th->th_dport;
1095 inc.inc_fibnum = so->so_fibnum;
1098 * Check for an existing connection attempt in syncache if
1099 * the flag is only ACK. A successful lookup creates a new
1100 * socket appended to the listen queue in SYN_RECEIVED state.
1102 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1104 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1106 * Parse the TCP options here because
1107 * syncookies need access to the reflected
1110 tcp_dooptions(&to, optp, optlen, 0);
1112 * NB: syncache_expand() doesn't unlock
1113 * inp and tcpinfo locks.
1115 rstreason = syncache_expand(&inc, &to, th, &so, m);
1116 if (rstreason < 0) {
1118 * A failing TCP MD5 signature comparison
1119 * must result in the segment being dropped
1120 * and must not produce any response back
1124 } else if (rstreason == 0) {
1126 * No syncache entry or ACK was not
1127 * for our SYN/ACK. Send a RST.
1128 * NB: syncache did its own logging
1129 * of the failure cause.
1131 rstreason = BANDLIM_RST_OPENPORT;
1139 * We completed the 3-way handshake
1140 * but could not allocate a socket
1141 * either due to memory shortage,
1142 * listen queue length limits or
1143 * global socket limits. Send RST
1144 * or wait and have the remote end
1145 * retransmit the ACK for another
1148 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1149 log(LOG_DEBUG, "%s; %s: Listen socket: "
1150 "Socket allocation failed due to "
1151 "limits or memory shortage, %s\n",
1153 V_tcp_sc_rst_sock_fail ?
1154 "sending RST" : "try again");
1155 if (V_tcp_sc_rst_sock_fail) {
1156 rstreason = BANDLIM_UNLIMITED;
1162 * Socket is created in state SYN_RECEIVED.
1163 * Unlock the listen socket, lock the newly
1164 * created socket and update the tp variable.
1166 INP_WUNLOCK(inp); /* listen socket */
1167 inp = sotoinpcb(so);
1169 * New connection inpcb is already locked by
1170 * syncache_expand().
1172 INP_WLOCK_ASSERT(inp);
1173 tp = intotcpcb(inp);
1174 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1175 ("%s: ", __func__));
1177 * Process the segment and the data it
1178 * contains. tcp_do_segment() consumes
1179 * the mbuf chain and unlocks the inpcb.
1181 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1182 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1184 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1185 return (IPPROTO_DONE);
1188 * Segment flag validation for new connection attempts:
1190 * Our (SYN|ACK) response was rejected.
1191 * Check with syncache and remove entry to prevent
1194 * NB: syncache_chkrst does its own logging of failure
1197 if (thflags & TH_RST) {
1198 syncache_chkrst(&inc, th);
1202 * We can't do anything without SYN.
1204 if ((thflags & TH_SYN) == 0) {
1205 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1206 log(LOG_DEBUG, "%s; %s: Listen socket: "
1207 "SYN is missing, segment ignored\n",
1209 TCPSTAT_INC(tcps_badsyn);
1213 * (SYN|ACK) is bogus on a listen socket.
1215 if (thflags & TH_ACK) {
1216 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1217 log(LOG_DEBUG, "%s; %s: Listen socket: "
1218 "SYN|ACK invalid, segment rejected\n",
1220 syncache_badack(&inc); /* XXX: Not needed! */
1221 TCPSTAT_INC(tcps_badsyn);
1222 rstreason = BANDLIM_RST_OPENPORT;
1226 * If the drop_synfin option is enabled, drop all
1227 * segments with both the SYN and FIN bits set.
1228 * This prevents e.g. nmap from identifying the
1230 * XXX: Poor reasoning. nmap has other methods
1231 * and is constantly refining its stack detection
1233 * XXX: This is a violation of the TCP specification
1234 * and was used by RFC1644.
1236 if ((thflags & TH_FIN) && V_drop_synfin) {
1237 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1238 log(LOG_DEBUG, "%s; %s: Listen socket: "
1239 "SYN|FIN segment ignored (based on "
1240 "sysctl setting)\n", s, __func__);
1241 TCPSTAT_INC(tcps_badsyn);
1245 * Segment's flags are (SYN) or (SYN|FIN).
1247 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1248 * as they do not affect the state of the TCP FSM.
1249 * The data pointed to by TH_URG and th_urp is ignored.
1251 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1252 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1253 KASSERT(thflags & (TH_SYN),
1254 ("%s: Listen socket: TH_SYN not set", __func__));
1257 * If deprecated address is forbidden,
1258 * we do not accept SYN to deprecated interface
1259 * address to prevent any new inbound connection from
1260 * getting established.
1261 * When we do not accept SYN, we send a TCP RST,
1262 * with deprecated source address (instead of dropping
1263 * it). We compromise it as it is much better for peer
1264 * to send a RST, and RST will be the final packet
1267 * If we do not forbid deprecated addresses, we accept
1268 * the SYN packet. RFC2462 does not suggest dropping
1270 * If we decipher RFC2462 5.5.4, it says like this:
1271 * 1. use of deprecated addr with existing
1272 * communication is okay - "SHOULD continue to be
1274 * 2. use of it with new communication:
1275 * (2a) "SHOULD NOT be used if alternate address
1276 * with sufficient scope is available"
1277 * (2b) nothing mentioned otherwise.
1278 * Here we fall into (2b) case as we have no choice in
1279 * our source address selection - we must obey the peer.
1281 * The wording in RFC2462 is confusing, and there are
1282 * multiple description text for deprecated address
1283 * handling - worse, they are not exactly the same.
1284 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1286 if (isipv6 && !V_ip6_use_deprecated) {
1287 struct in6_ifaddr *ia6;
1289 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1291 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1292 ifa_free(&ia6->ia_ifa);
1293 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1294 log(LOG_DEBUG, "%s; %s: Listen socket: "
1295 "Connection attempt to deprecated "
1296 "IPv6 address rejected\n",
1298 rstreason = BANDLIM_RST_OPENPORT;
1302 ifa_free(&ia6->ia_ifa);
1306 * Basic sanity checks on incoming SYN requests:
1307 * Don't respond if the destination is a link layer
1308 * broadcast according to RFC1122 4.2.3.10, p. 104.
1309 * If it is from this socket it must be forged.
1310 * Don't respond if the source or destination is a
1311 * global or subnet broad- or multicast address.
1312 * Note that it is quite possible to receive unicast
1313 * link-layer packets with a broadcast IP address. Use
1314 * in_broadcast() to find them.
1316 if (m->m_flags & (M_BCAST|M_MCAST)) {
1317 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1318 log(LOG_DEBUG, "%s; %s: Listen socket: "
1319 "Connection attempt from broad- or multicast "
1320 "link layer address ignored\n", s, __func__);
1325 if (th->th_dport == th->th_sport &&
1326 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1327 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1328 log(LOG_DEBUG, "%s; %s: Listen socket: "
1329 "Connection attempt to/from self "
1330 "ignored\n", s, __func__);
1333 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1334 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1335 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1336 log(LOG_DEBUG, "%s; %s: Listen socket: "
1337 "Connection attempt from/to multicast "
1338 "address ignored\n", s, __func__);
1343 #if defined(INET) && defined(INET6)
1348 if (th->th_dport == th->th_sport &&
1349 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1350 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1351 log(LOG_DEBUG, "%s; %s: Listen socket: "
1352 "Connection attempt from/to self "
1353 "ignored\n", s, __func__);
1356 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1357 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1358 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1359 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1360 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1361 log(LOG_DEBUG, "%s; %s: Listen socket: "
1362 "Connection attempt from/to broad- "
1363 "or multicast address ignored\n",
1370 * SYN appears to be valid. Create compressed TCP state
1374 if (so->so_options & SO_DEBUG)
1375 tcp_trace(TA_INPUT, ostate, tp,
1376 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1378 TCP_PROBE3(debug__input, tp, th, m);
1379 tcp_dooptions(&to, optp, optlen, TO_SYN);
1381 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL))
1382 goto new_tfo_socket;
1384 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1387 * Entry added to syncache and mbuf consumed.
1388 * Only the listen socket is unlocked by syncache_add().
1390 if (ti_locked == TI_RLOCKED) {
1391 INP_INFO_RUNLOCK(&V_tcbinfo);
1392 ti_locked = TI_UNLOCKED;
1394 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1395 return (IPPROTO_DONE);
1396 } else if (tp->t_state == TCPS_LISTEN) {
1398 * When a listen socket is torn down the SO_ACCEPTCONN
1399 * flag is removed first while connections are drained
1400 * from the accept queue in a unlock/lock cycle of the
1401 * ACCEPT_LOCK, opening a race condition allowing a SYN
1402 * attempt go through unhandled.
1406 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1407 if (tp->t_flags & TF_SIGNATURE) {
1408 tcp_dooptions(&to, optp, optlen, thflags);
1409 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1410 TCPSTAT_INC(tcps_sig_err_nosigopt);
1413 if (!TCPMD5_ENABLED() ||
1414 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1418 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1421 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1422 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1423 * the inpcb, and unlocks pcbinfo.
1425 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1426 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1427 return (IPPROTO_DONE);
1430 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1432 if (ti_locked == TI_RLOCKED) {
1433 INP_INFO_RUNLOCK(&V_tcbinfo);
1434 ti_locked = TI_UNLOCKED;
1438 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1439 "ti_locked: %d", __func__, ti_locked));
1440 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1445 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1448 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1449 m = NULL; /* mbuf chain got consumed. */
1454 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1456 if (ti_locked == TI_RLOCKED) {
1457 INP_INFO_RUNLOCK(&V_tcbinfo);
1458 ti_locked = TI_UNLOCKED;
1462 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1463 "ti_locked: %d", __func__, ti_locked));
1464 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1472 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1477 return (IPPROTO_DONE);
1481 * Automatic sizing of receive socket buffer. Often the send
1482 * buffer size is not optimally adjusted to the actual network
1483 * conditions at hand (delay bandwidth product). Setting the
1484 * buffer size too small limits throughput on links with high
1485 * bandwidth and high delay (eg. trans-continental/oceanic links).
1487 * On the receive side the socket buffer memory is only rarely
1488 * used to any significant extent. This allows us to be much
1489 * more aggressive in scaling the receive socket buffer. For
1490 * the case that the buffer space is actually used to a large
1491 * extent and we run out of kernel memory we can simply drop
1492 * the new segments; TCP on the sender will just retransmit it
1493 * later. Setting the buffer size too big may only consume too
1494 * much kernel memory if the application doesn't read() from
1495 * the socket or packet loss or reordering makes use of the
1498 * The criteria to step up the receive buffer one notch are:
1499 * 1. Application has not set receive buffer size with
1500 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1501 * 2. the number of bytes received during the time it takes
1502 * one timestamp to be reflected back to us (the RTT);
1503 * 3. received bytes per RTT is within seven eighth of the
1504 * current socket buffer size;
1505 * 4. receive buffer size has not hit maximal automatic size;
1507 * This algorithm does one step per RTT at most and only if
1508 * we receive a bulk stream w/o packet losses or reorderings.
1509 * Shrinking the buffer during idle times is not necessary as
1510 * it doesn't consume any memory when idle.
1512 * TODO: Only step up if the application is actually serving
1513 * the buffer to better manage the socket buffer resources.
1516 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1517 struct tcpcb *tp, int tlen)
1521 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1522 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1523 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1524 (tp->t_srtt >> TCP_RTT_SHIFT)) {
1525 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) &&
1526 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1527 newsize = min(so->so_rcv.sb_hiwat +
1528 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max);
1530 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1532 /* Start over with next RTT. */
1536 tp->rfbuf_cnt += tlen; /* add up */
1543 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1544 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1547 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1548 int rstreason, todrop, win;
1551 struct in_conninfo *inc;
1558 * The size of tcp_saveipgen must be the size of the max ip header,
1561 u_char tcp_saveipgen[IP6_HDR_LEN];
1562 struct tcphdr tcp_savetcp;
1565 thflags = th->th_flags;
1566 inc = &tp->t_inpcb->inp_inc;
1567 tp->sackhint.last_sack_ack = 0;
1571 * If this is either a state-changing packet or current state isn't
1572 * established, we require a write lock on tcbinfo. Otherwise, we
1573 * allow the tcbinfo to be in either alocked or unlocked, as the
1574 * caller may have unnecessarily acquired a write lock due to a race.
1576 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1577 tp->t_state != TCPS_ESTABLISHED) {
1578 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for "
1579 "SYN/FIN/RST/!EST", __func__, ti_locked));
1580 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1583 if (ti_locked == TI_RLOCKED)
1584 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1586 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1587 "ti_locked: %d", __func__, ti_locked));
1588 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1592 INP_WLOCK_ASSERT(tp->t_inpcb);
1593 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1595 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1599 /* Save segment, if requested. */
1600 tcp_pcap_add(th, m, &(tp->t_inpkts));
1603 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1604 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1605 log(LOG_DEBUG, "%s; %s: "
1606 "SYN|FIN segment ignored (based on "
1607 "sysctl setting)\n", s, __func__);
1614 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1615 * check SEQ.ACK first.
1617 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1618 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1619 rstreason = BANDLIM_UNLIMITED;
1624 * Segment received on connection.
1625 * Reset idle time and keep-alive timer.
1626 * XXX: This should be done after segment
1627 * validation to ignore broken/spoofed segs.
1629 tp->t_rcvtime = ticks;
1630 if (TCPS_HAVEESTABLISHED(tp->t_state))
1631 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
1634 * Scale up the window into a 32-bit value.
1635 * For the SYN_SENT state the scale is zero.
1637 tiwin = th->th_win << tp->snd_scale;
1640 * TCP ECN processing.
1642 if (tp->t_flags & TF_ECN_PERMIT) {
1643 if (thflags & TH_CWR)
1644 tp->t_flags &= ~TF_ECN_SND_ECE;
1645 switch (iptos & IPTOS_ECN_MASK) {
1647 tp->t_flags |= TF_ECN_SND_ECE;
1648 TCPSTAT_INC(tcps_ecn_ce);
1650 case IPTOS_ECN_ECT0:
1651 TCPSTAT_INC(tcps_ecn_ect0);
1653 case IPTOS_ECN_ECT1:
1654 TCPSTAT_INC(tcps_ecn_ect1);
1658 /* Process a packet differently from RFC3168. */
1659 cc_ecnpkt_handler(tp, th, iptos);
1661 /* Congestion experienced. */
1662 if (thflags & TH_ECE) {
1663 cc_cong_signal(tp, th, CC_ECN);
1668 * Parse options on any incoming segment.
1670 tcp_dooptions(&to, (u_char *)(th + 1),
1671 (th->th_off << 2) - sizeof(struct tcphdr),
1672 (thflags & TH_SYN) ? TO_SYN : 0);
1674 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1675 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1676 (to.to_flags & TOF_SIGNATURE) == 0) {
1677 TCPSTAT_INC(tcps_sig_err_sigopt);
1678 /* XXX: should drop? */
1682 * If echoed timestamp is later than the current time,
1683 * fall back to non RFC1323 RTT calculation. Normalize
1684 * timestamp if syncookies were used when this connection
1687 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1688 to.to_tsecr -= tp->ts_offset;
1689 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1693 * Process options only when we get SYN/ACK back. The SYN case
1694 * for incoming connections is handled in tcp_syncache.
1695 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1696 * or <SYN,ACK>) segment itself is never scaled.
1697 * XXX this is traditional behavior, may need to be cleaned up.
1699 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1700 if ((to.to_flags & TOF_SCALE) &&
1701 (tp->t_flags & TF_REQ_SCALE)) {
1702 tp->t_flags |= TF_RCVD_SCALE;
1703 tp->snd_scale = to.to_wscale;
1706 * Initial send window. It will be updated with
1707 * the next incoming segment to the scaled value.
1709 tp->snd_wnd = th->th_win;
1710 if (to.to_flags & TOF_TS) {
1711 tp->t_flags |= TF_RCVD_TSTMP;
1712 tp->ts_recent = to.to_tsval;
1713 tp->ts_recent_age = tcp_ts_getticks();
1715 if (to.to_flags & TOF_MSS)
1716 tcp_mss(tp, to.to_mss);
1717 if ((tp->t_flags & TF_SACK_PERMIT) &&
1718 (to.to_flags & TOF_SACKPERM) == 0)
1719 tp->t_flags &= ~TF_SACK_PERMIT;
1723 * If timestamps were negotiated during SYN/ACK they should
1724 * appear on every segment during this session and vice versa.
1726 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1727 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1728 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1729 "no action\n", s, __func__);
1733 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1734 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1735 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1736 "no action\n", s, __func__);
1742 * Header prediction: check for the two common cases
1743 * of a uni-directional data xfer. If the packet has
1744 * no control flags, is in-sequence, the window didn't
1745 * change and we're not retransmitting, it's a
1746 * candidate. If the length is zero and the ack moved
1747 * forward, we're the sender side of the xfer. Just
1748 * free the data acked & wake any higher level process
1749 * that was blocked waiting for space. If the length
1750 * is non-zero and the ack didn't move, we're the
1751 * receiver side. If we're getting packets in-order
1752 * (the reassembly queue is empty), add the data to
1753 * the socket buffer and note that we need a delayed ack.
1754 * Make sure that the hidden state-flags are also off.
1755 * Since we check for TCPS_ESTABLISHED first, it can only
1758 if (tp->t_state == TCPS_ESTABLISHED &&
1759 th->th_seq == tp->rcv_nxt &&
1760 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1761 tp->snd_nxt == tp->snd_max &&
1762 tiwin && tiwin == tp->snd_wnd &&
1763 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1765 ((to.to_flags & TOF_TS) == 0 ||
1766 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1769 * If last ACK falls within this segment's sequence numbers,
1770 * record the timestamp.
1771 * NOTE that the test is modified according to the latest
1772 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1774 if ((to.to_flags & TOF_TS) != 0 &&
1775 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1776 tp->ts_recent_age = tcp_ts_getticks();
1777 tp->ts_recent = to.to_tsval;
1781 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1782 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1783 !IN_RECOVERY(tp->t_flags) &&
1784 (to.to_flags & TOF_SACK) == 0 &&
1785 TAILQ_EMPTY(&tp->snd_holes)) {
1787 * This is a pure ack for outstanding data.
1789 if (ti_locked == TI_RLOCKED)
1790 INP_INFO_RUNLOCK(&V_tcbinfo);
1791 ti_locked = TI_UNLOCKED;
1793 TCPSTAT_INC(tcps_predack);
1796 * "bad retransmit" recovery.
1798 if (tp->t_rxtshift == 1 &&
1799 tp->t_flags & TF_PREVVALID &&
1800 (int)(ticks - tp->t_badrxtwin) < 0) {
1801 cc_cong_signal(tp, th, CC_RTO_ERR);
1805 * Recalculate the transmit timer / rtt.
1807 * Some boxes send broken timestamp replies
1808 * during the SYN+ACK phase, ignore
1809 * timestamps of 0 or we could calculate a
1810 * huge RTT and blow up the retransmit timer.
1812 if ((to.to_flags & TOF_TS) != 0 &&
1816 t = tcp_ts_getticks() - to.to_tsecr;
1817 if (!tp->t_rttlow || tp->t_rttlow > t)
1820 TCP_TS_TO_TICKS(t) + 1);
1821 } else if (tp->t_rtttime &&
1822 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1823 if (!tp->t_rttlow ||
1824 tp->t_rttlow > ticks - tp->t_rtttime)
1825 tp->t_rttlow = ticks - tp->t_rtttime;
1827 ticks - tp->t_rtttime);
1829 acked = BYTES_THIS_ACK(tp, th);
1831 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1832 hhook_run_tcp_est_in(tp, th, &to);
1834 TCPSTAT_INC(tcps_rcvackpack);
1835 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1836 sbdrop(&so->so_snd, acked);
1837 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1838 SEQ_LEQ(th->th_ack, tp->snd_recover))
1839 tp->snd_recover = th->th_ack - 1;
1842 * Let the congestion control algorithm update
1843 * congestion control related information. This
1844 * typically means increasing the congestion
1847 cc_ack_received(tp, th, CC_ACK);
1849 tp->snd_una = th->th_ack;
1851 * Pull snd_wl2 up to prevent seq wrap relative
1854 tp->snd_wl2 = th->th_ack;
1859 * If all outstanding data are acked, stop
1860 * retransmit timer, otherwise restart timer
1861 * using current (possibly backed-off) value.
1862 * If process is waiting for space,
1863 * wakeup/selwakeup/signal. If data
1864 * are ready to send, let tcp_output
1865 * decide between more output or persist.
1868 if (so->so_options & SO_DEBUG)
1869 tcp_trace(TA_INPUT, ostate, tp,
1870 (void *)tcp_saveipgen,
1873 TCP_PROBE3(debug__input, tp, th, m);
1874 if (tp->snd_una == tp->snd_max)
1875 tcp_timer_activate(tp, TT_REXMT, 0);
1876 else if (!tcp_timer_active(tp, TT_PERSIST))
1877 tcp_timer_activate(tp, TT_REXMT,
1880 if (sbavail(&so->so_snd))
1881 (void) tp->t_fb->tfb_tcp_output(tp);
1884 } else if (th->th_ack == tp->snd_una &&
1885 tlen <= sbspace(&so->so_rcv)) {
1886 int newsize = 0; /* automatic sockbuf scaling */
1889 * This is a pure, in-sequence data packet with
1890 * nothing on the reassembly queue and we have enough
1891 * buffer space to take it.
1893 if (ti_locked == TI_RLOCKED)
1894 INP_INFO_RUNLOCK(&V_tcbinfo);
1895 ti_locked = TI_UNLOCKED;
1897 /* Clean receiver SACK report if present */
1898 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1899 tcp_clean_sackreport(tp);
1900 TCPSTAT_INC(tcps_preddat);
1901 tp->rcv_nxt += tlen;
1903 * Pull snd_wl1 up to prevent seq wrap relative to
1906 tp->snd_wl1 = th->th_seq;
1908 * Pull rcv_up up to prevent seq wrap relative to
1911 tp->rcv_up = tp->rcv_nxt;
1912 TCPSTAT_INC(tcps_rcvpack);
1913 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1915 if (so->so_options & SO_DEBUG)
1916 tcp_trace(TA_INPUT, ostate, tp,
1917 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1919 TCP_PROBE3(debug__input, tp, th, m);
1921 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1923 /* Add data to socket buffer. */
1924 SOCKBUF_LOCK(&so->so_rcv);
1925 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1929 * Set new socket buffer size.
1930 * Give up when limit is reached.
1933 if (!sbreserve_locked(&so->so_rcv,
1935 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1936 m_adj(m, drop_hdrlen); /* delayed header drop */
1937 sbappendstream_locked(&so->so_rcv, m, 0);
1939 /* NB: sorwakeup_locked() does an implicit unlock. */
1940 sorwakeup_locked(so);
1941 if (DELAY_ACK(tp, tlen)) {
1942 tp->t_flags |= TF_DELACK;
1944 tp->t_flags |= TF_ACKNOW;
1945 tp->t_fb->tfb_tcp_output(tp);
1952 * Calculate amount of space in receive window,
1953 * and then do TCP input processing.
1954 * Receive window is amount of space in rcv queue,
1955 * but not less than advertised window.
1957 win = sbspace(&so->so_rcv);
1960 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1962 switch (tp->t_state) {
1965 * If the state is SYN_RECEIVED:
1966 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1968 case TCPS_SYN_RECEIVED:
1969 if ((thflags & TH_ACK) &&
1970 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1971 SEQ_GT(th->th_ack, tp->snd_max))) {
1972 rstreason = BANDLIM_RST_OPENPORT;
1976 if (tp->t_flags & TF_FASTOPEN) {
1978 * When a TFO connection is in SYN_RECEIVED, the
1979 * only valid packets are the initial SYN, a
1980 * retransmit/copy of the initial SYN (possibly with
1981 * a subset of the original data), a valid ACK, a
1984 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1985 rstreason = BANDLIM_RST_OPENPORT;
1987 } else if (thflags & TH_SYN) {
1988 /* non-initial SYN is ignored */
1989 if ((tcp_timer_active(tp, TT_DELACK) ||
1990 tcp_timer_active(tp, TT_REXMT)))
1992 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2000 * If the state is SYN_SENT:
2001 * if seg contains a RST, then drop the connection.
2002 * if seg does not contain SYN, then drop it.
2003 * Otherwise this is an acceptable SYN segment
2004 * initialize tp->rcv_nxt and tp->irs
2005 * if seg contains ack then advance tp->snd_una
2006 * if seg contains an ECE and ECN support is enabled, the stream
2008 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2009 * arrange for segment to be acked (eventually)
2010 * continue processing rest of data/controls, beginning with URG
2013 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2014 TCP_PROBE5(connect__refused, NULL, tp,
2016 tp = tcp_drop(tp, ECONNREFUSED);
2018 if (thflags & TH_RST)
2020 if (!(thflags & TH_SYN))
2023 tp->irs = th->th_seq;
2025 if (thflags & TH_ACK) {
2026 TCPSTAT_INC(tcps_connects);
2029 mac_socketpeer_set_from_mbuf(m, so);
2031 /* Do window scaling on this connection? */
2032 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2033 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2034 tp->rcv_scale = tp->request_r_scale;
2036 tp->rcv_adv += imin(tp->rcv_wnd,
2037 TCP_MAXWIN << tp->rcv_scale);
2038 tp->snd_una++; /* SYN is acked */
2040 * If there's data, delay ACK; if there's also a FIN
2041 * ACKNOW will be turned on later.
2043 if (DELAY_ACK(tp, tlen) && tlen != 0)
2044 tcp_timer_activate(tp, TT_DELACK,
2047 tp->t_flags |= TF_ACKNOW;
2049 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
2050 tp->t_flags |= TF_ECN_PERMIT;
2051 TCPSTAT_INC(tcps_ecn_shs);
2055 * Received <SYN,ACK> in SYN_SENT[*] state.
2057 * SYN_SENT --> ESTABLISHED
2058 * SYN_SENT* --> FIN_WAIT_1
2060 tp->t_starttime = ticks;
2061 if (tp->t_flags & TF_NEEDFIN) {
2062 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2063 tp->t_flags &= ~TF_NEEDFIN;
2066 tcp_state_change(tp, TCPS_ESTABLISHED);
2067 TCP_PROBE5(connect__established, NULL, tp,
2070 tcp_timer_activate(tp, TT_KEEP,
2075 * Received initial SYN in SYN-SENT[*] state =>
2076 * simultaneous open.
2077 * If it succeeds, connection is * half-synchronized.
2078 * Otherwise, do 3-way handshake:
2079 * SYN-SENT -> SYN-RECEIVED
2080 * SYN-SENT* -> SYN-RECEIVED*
2082 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2083 tcp_timer_activate(tp, TT_REXMT, 0);
2084 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2087 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: "
2088 "ti_locked %d", __func__, ti_locked));
2089 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2090 INP_WLOCK_ASSERT(tp->t_inpcb);
2093 * Advance th->th_seq to correspond to first data byte.
2094 * If data, trim to stay within window,
2095 * dropping FIN if necessary.
2098 if (tlen > tp->rcv_wnd) {
2099 todrop = tlen - tp->rcv_wnd;
2103 TCPSTAT_INC(tcps_rcvpackafterwin);
2104 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2106 tp->snd_wl1 = th->th_seq - 1;
2107 tp->rcv_up = th->th_seq;
2109 * Client side of transaction: already sent SYN and data.
2110 * If the remote host used T/TCP to validate the SYN,
2111 * our data will be ACK'd; if so, enter normal data segment
2112 * processing in the middle of step 5, ack processing.
2113 * Otherwise, goto step 6.
2115 if (thflags & TH_ACK)
2121 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2122 * do normal processing.
2124 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2128 break; /* continue normal processing */
2132 * States other than LISTEN or SYN_SENT.
2133 * First check the RST flag and sequence number since reset segments
2134 * are exempt from the timestamp and connection count tests. This
2135 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2136 * below which allowed reset segments in half the sequence space
2137 * to fall though and be processed (which gives forged reset
2138 * segments with a random sequence number a 50 percent chance of
2139 * killing a connection).
2140 * Then check timestamp, if present.
2141 * Then check the connection count, if present.
2142 * Then check that at least some bytes of segment are within
2143 * receive window. If segment begins before rcv_nxt,
2144 * drop leading data (and SYN); if nothing left, just ack.
2146 if (thflags & TH_RST) {
2148 * RFC5961 Section 3.2
2150 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2151 * - If RST is in window, we send challenge ACK.
2153 * Note: to take into account delayed ACKs, we should
2154 * test against last_ack_sent instead of rcv_nxt.
2155 * Note 2: we handle special case of closed window, not
2156 * covered by the RFC.
2158 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2159 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2160 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2162 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2163 KASSERT(ti_locked == TI_RLOCKED,
2164 ("%s: TH_RST ti_locked %d, th %p tp %p",
2165 __func__, ti_locked, th, tp));
2166 KASSERT(tp->t_state != TCPS_SYN_SENT,
2167 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2170 if (V_tcp_insecure_rst ||
2171 tp->last_ack_sent == th->th_seq) {
2172 TCPSTAT_INC(tcps_drops);
2173 /* Drop the connection. */
2174 switch (tp->t_state) {
2175 case TCPS_SYN_RECEIVED:
2176 so->so_error = ECONNREFUSED;
2178 case TCPS_ESTABLISHED:
2179 case TCPS_FIN_WAIT_1:
2180 case TCPS_FIN_WAIT_2:
2181 case TCPS_CLOSE_WAIT:
2184 so->so_error = ECONNRESET;
2191 TCPSTAT_INC(tcps_badrst);
2192 /* Send challenge ACK. */
2193 tcp_respond(tp, mtod(m, void *), th, m,
2194 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2195 tp->last_ack_sent = tp->rcv_nxt;
2203 * RFC5961 Section 4.2
2204 * Send challenge ACK for any SYN in synchronized state.
2206 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2207 tp->t_state != TCPS_SYN_RECEIVED) {
2208 KASSERT(ti_locked == TI_RLOCKED,
2209 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2210 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2212 TCPSTAT_INC(tcps_badsyn);
2213 if (V_tcp_insecure_syn &&
2214 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2215 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2216 tp = tcp_drop(tp, ECONNRESET);
2217 rstreason = BANDLIM_UNLIMITED;
2219 /* Send challenge ACK. */
2220 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2221 tp->snd_nxt, TH_ACK);
2222 tp->last_ack_sent = tp->rcv_nxt;
2229 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2230 * and it's less than ts_recent, drop it.
2232 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2233 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2235 /* Check to see if ts_recent is over 24 days old. */
2236 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2238 * Invalidate ts_recent. If this segment updates
2239 * ts_recent, the age will be reset later and ts_recent
2240 * will get a valid value. If it does not, setting
2241 * ts_recent to zero will at least satisfy the
2242 * requirement that zero be placed in the timestamp
2243 * echo reply when ts_recent isn't valid. The
2244 * age isn't reset until we get a valid ts_recent
2245 * because we don't want out-of-order segments to be
2246 * dropped when ts_recent is old.
2250 TCPSTAT_INC(tcps_rcvduppack);
2251 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2252 TCPSTAT_INC(tcps_pawsdrop);
2260 * In the SYN-RECEIVED state, validate that the packet belongs to
2261 * this connection before trimming the data to fit the receive
2262 * window. Check the sequence number versus IRS since we know
2263 * the sequence numbers haven't wrapped. This is a partial fix
2264 * for the "LAND" DoS attack.
2266 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2267 rstreason = BANDLIM_RST_OPENPORT;
2271 todrop = tp->rcv_nxt - th->th_seq;
2273 if (thflags & TH_SYN) {
2283 * Following if statement from Stevens, vol. 2, p. 960.
2286 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2288 * Any valid FIN must be to the left of the window.
2289 * At this point the FIN must be a duplicate or out
2290 * of sequence; drop it.
2295 * Send an ACK to resynchronize and drop any data.
2296 * But keep on processing for RST or ACK.
2298 tp->t_flags |= TF_ACKNOW;
2300 TCPSTAT_INC(tcps_rcvduppack);
2301 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2303 TCPSTAT_INC(tcps_rcvpartduppack);
2304 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2306 drop_hdrlen += todrop; /* drop from the top afterwards */
2307 th->th_seq += todrop;
2309 if (th->th_urp > todrop)
2310 th->th_urp -= todrop;
2318 * If new data are received on a connection after the
2319 * user processes are gone, then RST the other end.
2321 if ((so->so_state & SS_NOFDREF) &&
2322 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2323 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && "
2324 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2325 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2327 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2328 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2329 "after socket was closed, "
2330 "sending RST and removing tcpcb\n",
2331 s, __func__, tcpstates[tp->t_state], tlen);
2335 TCPSTAT_INC(tcps_rcvafterclose);
2336 rstreason = BANDLIM_UNLIMITED;
2341 * If segment ends after window, drop trailing data
2342 * (and PUSH and FIN); if nothing left, just ACK.
2344 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2346 TCPSTAT_INC(tcps_rcvpackafterwin);
2347 if (todrop >= tlen) {
2348 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2350 * If window is closed can only take segments at
2351 * window edge, and have to drop data and PUSH from
2352 * incoming segments. Continue processing, but
2353 * remember to ack. Otherwise, drop segment
2356 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2357 tp->t_flags |= TF_ACKNOW;
2358 TCPSTAT_INC(tcps_rcvwinprobe);
2362 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2365 thflags &= ~(TH_PUSH|TH_FIN);
2369 * If last ACK falls within this segment's sequence numbers,
2370 * record its timestamp.
2372 * 1) That the test incorporates suggestions from the latest
2373 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2374 * 2) That updating only on newer timestamps interferes with
2375 * our earlier PAWS tests, so this check should be solely
2376 * predicated on the sequence space of this segment.
2377 * 3) That we modify the segment boundary check to be
2378 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2379 * instead of RFC1323's
2380 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2381 * This modified check allows us to overcome RFC1323's
2382 * limitations as described in Stevens TCP/IP Illustrated
2383 * Vol. 2 p.869. In such cases, we can still calculate the
2384 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2386 if ((to.to_flags & TOF_TS) != 0 &&
2387 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2388 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2389 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2390 tp->ts_recent_age = tcp_ts_getticks();
2391 tp->ts_recent = to.to_tsval;
2395 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2396 * flag is on (half-synchronized state), then queue data for
2397 * later processing; else drop segment and return.
2399 if ((thflags & TH_ACK) == 0) {
2400 if (tp->t_state == TCPS_SYN_RECEIVED ||
2401 (tp->t_flags & TF_NEEDSYN)) {
2403 if (tp->t_state == TCPS_SYN_RECEIVED &&
2404 tp->t_flags & TF_FASTOPEN) {
2405 tp->snd_wnd = tiwin;
2410 } else if (tp->t_flags & TF_ACKNOW)
2419 switch (tp->t_state) {
2422 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2423 * ESTABLISHED state and continue processing.
2424 * The ACK was checked above.
2426 case TCPS_SYN_RECEIVED:
2428 TCPSTAT_INC(tcps_connects);
2430 /* Do window scaling? */
2431 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2432 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2433 tp->rcv_scale = tp->request_r_scale;
2434 tp->snd_wnd = tiwin;
2438 * SYN-RECEIVED -> ESTABLISHED
2439 * SYN-RECEIVED* -> FIN-WAIT-1
2441 tp->t_starttime = ticks;
2442 if (tp->t_flags & TF_NEEDFIN) {
2443 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2444 tp->t_flags &= ~TF_NEEDFIN;
2446 tcp_state_change(tp, TCPS_ESTABLISHED);
2447 TCP_PROBE5(accept__established, NULL, tp,
2450 if (tp->t_tfo_pending) {
2451 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2452 tp->t_tfo_pending = NULL;
2455 * Account for the ACK of our SYN prior to
2456 * regular ACK processing below.
2461 * TFO connections call cc_conn_init() during SYN
2462 * processing. Calling it again here for such
2463 * connections is not harmless as it would undo the
2464 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2467 if (!(tp->t_flags & TF_FASTOPEN))
2470 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2473 * If segment contains data or ACK, will call tcp_reass()
2474 * later; if not, do so now to pass queued data to user.
2476 if (tlen == 0 && (thflags & TH_FIN) == 0)
2477 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2479 tp->snd_wl1 = th->th_seq - 1;
2483 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2484 * ACKs. If the ack is in the range
2485 * tp->snd_una < th->th_ack <= tp->snd_max
2486 * then advance tp->snd_una to th->th_ack and drop
2487 * data from the retransmission queue. If this ACK reflects
2488 * more up to date window information we update our window information.
2490 case TCPS_ESTABLISHED:
2491 case TCPS_FIN_WAIT_1:
2492 case TCPS_FIN_WAIT_2:
2493 case TCPS_CLOSE_WAIT:
2496 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2497 TCPSTAT_INC(tcps_rcvacktoomuch);
2500 if ((tp->t_flags & TF_SACK_PERMIT) &&
2501 ((to.to_flags & TOF_SACK) ||
2502 !TAILQ_EMPTY(&tp->snd_holes)))
2503 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2506 * Reset the value so that previous (valid) value
2507 * from the last ack with SACK doesn't get used.
2509 tp->sackhint.sacked_bytes = 0;
2511 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2512 hhook_run_tcp_est_in(tp, th, &to);
2514 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2517 maxseg = tcp_maxseg(tp);
2519 (tiwin == tp->snd_wnd ||
2520 (tp->t_flags & TF_SACK_PERMIT))) {
2522 * If this is the first time we've seen a
2523 * FIN from the remote, this is not a
2524 * duplicate and it needs to be processed
2525 * normally. This happens during a
2526 * simultaneous close.
2528 if ((thflags & TH_FIN) &&
2529 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2533 TCPSTAT_INC(tcps_rcvdupack);
2535 * If we have outstanding data (other than
2536 * a window probe), this is a completely
2537 * duplicate ack (ie, window info didn't
2538 * change and FIN isn't set),
2539 * the ack is the biggest we've
2540 * seen and we've seen exactly our rexmt
2541 * threshold of them, assume a packet
2542 * has been dropped and retransmit it.
2543 * Kludge snd_nxt & the congestion
2544 * window so we send only this one
2547 * We know we're losing at the current
2548 * window size so do congestion avoidance
2549 * (set ssthresh to half the current window
2550 * and pull our congestion window back to
2551 * the new ssthresh).
2553 * Dup acks mean that packets have left the
2554 * network (they're now cached at the receiver)
2555 * so bump cwnd by the amount in the receiver
2556 * to keep a constant cwnd packets in the
2559 * When using TCP ECN, notify the peer that
2560 * we reduced the cwnd.
2563 * Following 2 kinds of acks should not affect
2566 * 2) Acks with SACK but without any new SACK
2567 * information in them. These could result from
2568 * any anomaly in the network like a switch
2569 * duplicating packets or a possible DoS attack.
2571 if (th->th_ack != tp->snd_una ||
2572 ((tp->t_flags & TF_SACK_PERMIT) &&
2575 else if (!tcp_timer_active(tp, TT_REXMT))
2577 else if (++tp->t_dupacks > tcprexmtthresh ||
2578 IN_FASTRECOVERY(tp->t_flags)) {
2579 cc_ack_received(tp, th, CC_DUPACK);
2580 if ((tp->t_flags & TF_SACK_PERMIT) &&
2581 IN_FASTRECOVERY(tp->t_flags)) {
2585 * Compute the amount of data in flight first.
2586 * We can inject new data into the pipe iff
2587 * we have less than 1/2 the original window's
2588 * worth of data in flight.
2590 if (V_tcp_do_rfc6675_pipe)
2591 awnd = tcp_compute_pipe(tp);
2593 awnd = (tp->snd_nxt - tp->snd_fack) +
2594 tp->sackhint.sack_bytes_rexmit;
2596 if (awnd < tp->snd_ssthresh) {
2597 tp->snd_cwnd += maxseg;
2598 if (tp->snd_cwnd > tp->snd_ssthresh)
2599 tp->snd_cwnd = tp->snd_ssthresh;
2602 tp->snd_cwnd += maxseg;
2603 (void) tp->t_fb->tfb_tcp_output(tp);
2605 } else if (tp->t_dupacks == tcprexmtthresh) {
2606 tcp_seq onxt = tp->snd_nxt;
2609 * If we're doing sack, check to
2610 * see if we're already in sack
2611 * recovery. If we're not doing sack,
2612 * check to see if we're in newreno
2615 if (tp->t_flags & TF_SACK_PERMIT) {
2616 if (IN_FASTRECOVERY(tp->t_flags)) {
2621 if (SEQ_LEQ(th->th_ack,
2627 /* Congestion signal before ack. */
2628 cc_cong_signal(tp, th, CC_NDUPACK);
2629 cc_ack_received(tp, th, CC_DUPACK);
2630 tcp_timer_activate(tp, TT_REXMT, 0);
2632 if (tp->t_flags & TF_SACK_PERMIT) {
2634 tcps_sack_recovery_episode);
2635 tp->sack_newdata = tp->snd_nxt;
2636 tp->snd_cwnd = maxseg;
2637 (void) tp->t_fb->tfb_tcp_output(tp);
2640 tp->snd_nxt = th->th_ack;
2641 tp->snd_cwnd = maxseg;
2642 (void) tp->t_fb->tfb_tcp_output(tp);
2643 KASSERT(tp->snd_limited <= 2,
2644 ("%s: tp->snd_limited too big",
2646 tp->snd_cwnd = tp->snd_ssthresh +
2648 (tp->t_dupacks - tp->snd_limited);
2649 if (SEQ_GT(onxt, tp->snd_nxt))
2652 } else if (V_tcp_do_rfc3042) {
2654 * Process first and second duplicate
2655 * ACKs. Each indicates a segment
2656 * leaving the network, creating room
2657 * for more. Make sure we can send a
2658 * packet on reception of each duplicate
2659 * ACK by increasing snd_cwnd by one
2660 * segment. Restore the original
2661 * snd_cwnd after packet transmission.
2663 cc_ack_received(tp, th, CC_DUPACK);
2664 u_long oldcwnd = tp->snd_cwnd;
2665 tcp_seq oldsndmax = tp->snd_max;
2669 KASSERT(tp->t_dupacks == 1 ||
2671 ("%s: dupacks not 1 or 2",
2673 if (tp->t_dupacks == 1)
2674 tp->snd_limited = 0;
2676 (tp->snd_nxt - tp->snd_una) +
2677 (tp->t_dupacks - tp->snd_limited) *
2680 * Only call tcp_output when there
2681 * is new data available to be sent.
2682 * Otherwise we would send pure ACKs.
2684 SOCKBUF_LOCK(&so->so_snd);
2685 avail = sbavail(&so->so_snd) -
2686 (tp->snd_nxt - tp->snd_una);
2687 SOCKBUF_UNLOCK(&so->so_snd);
2689 (void) tp->t_fb->tfb_tcp_output(tp);
2690 sent = tp->snd_max - oldsndmax;
2691 if (sent > maxseg) {
2692 KASSERT((tp->t_dupacks == 2 &&
2693 tp->snd_limited == 0) ||
2694 (sent == maxseg + 1 &&
2695 tp->t_flags & TF_SENTFIN),
2696 ("%s: sent too much",
2698 tp->snd_limited = 2;
2699 } else if (sent > 0)
2701 tp->snd_cwnd = oldcwnd;
2708 * This ack is advancing the left edge, reset the
2713 * If this ack also has new SACK info, increment the
2714 * counter as per rfc6675.
2716 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed)
2720 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2721 ("%s: th_ack <= snd_una", __func__));
2724 * If the congestion window was inflated to account
2725 * for the other side's cached packets, retract it.
2727 if (IN_FASTRECOVERY(tp->t_flags)) {
2728 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2729 if (tp->t_flags & TF_SACK_PERMIT)
2730 tcp_sack_partialack(tp, th);
2732 tcp_newreno_partial_ack(tp, th);
2734 cc_post_recovery(tp, th);
2737 * If we reach this point, ACK is not a duplicate,
2738 * i.e., it ACKs something we sent.
2740 if (tp->t_flags & TF_NEEDSYN) {
2742 * T/TCP: Connection was half-synchronized, and our
2743 * SYN has been ACK'd (so connection is now fully
2744 * synchronized). Go to non-starred state,
2745 * increment snd_una for ACK of SYN, and check if
2746 * we can do window scaling.
2748 tp->t_flags &= ~TF_NEEDSYN;
2750 /* Do window scaling? */
2751 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2752 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2753 tp->rcv_scale = tp->request_r_scale;
2754 /* Send window already scaled. */
2759 INP_WLOCK_ASSERT(tp->t_inpcb);
2761 acked = BYTES_THIS_ACK(tp, th);
2762 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2763 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2764 tp->snd_una, th->th_ack, tp, m));
2765 TCPSTAT_INC(tcps_rcvackpack);
2766 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2769 * If we just performed our first retransmit, and the ACK
2770 * arrives within our recovery window, then it was a mistake
2771 * to do the retransmit in the first place. Recover our
2772 * original cwnd and ssthresh, and proceed to transmit where
2775 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2776 (int)(ticks - tp->t_badrxtwin) < 0)
2777 cc_cong_signal(tp, th, CC_RTO_ERR);
2780 * If we have a timestamp reply, update smoothed
2781 * round trip time. If no timestamp is present but
2782 * transmit timer is running and timed sequence
2783 * number was acked, update smoothed round trip time.
2784 * Since we now have an rtt measurement, cancel the
2785 * timer backoff (cf., Phil Karn's retransmit alg.).
2786 * Recompute the initial retransmit timer.
2788 * Some boxes send broken timestamp replies
2789 * during the SYN+ACK phase, ignore
2790 * timestamps of 0 or we could calculate a
2791 * huge RTT and blow up the retransmit timer.
2793 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2796 t = tcp_ts_getticks() - to.to_tsecr;
2797 if (!tp->t_rttlow || tp->t_rttlow > t)
2799 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2800 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2801 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2802 tp->t_rttlow = ticks - tp->t_rtttime;
2803 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2807 * If all outstanding data is acked, stop retransmit
2808 * timer and remember to restart (more output or persist).
2809 * If there is more data to be acked, restart retransmit
2810 * timer, using current (possibly backed-off) value.
2812 if (th->th_ack == tp->snd_max) {
2813 tcp_timer_activate(tp, TT_REXMT, 0);
2815 } else if (!tcp_timer_active(tp, TT_PERSIST))
2816 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2819 * If no data (only SYN) was ACK'd,
2820 * skip rest of ACK processing.
2826 * Let the congestion control algorithm update congestion
2827 * control related information. This typically means increasing
2828 * the congestion window.
2830 cc_ack_received(tp, th, CC_ACK);
2832 SOCKBUF_LOCK(&so->so_snd);
2833 if (acked > sbavail(&so->so_snd)) {
2834 if (tp->snd_wnd >= sbavail(&so->so_snd))
2835 tp->snd_wnd -= sbavail(&so->so_snd);
2838 mfree = sbcut_locked(&so->so_snd,
2839 (int)sbavail(&so->so_snd));
2842 mfree = sbcut_locked(&so->so_snd, acked);
2843 if (tp->snd_wnd >= (u_long) acked)
2844 tp->snd_wnd -= acked;
2849 /* NB: sowwakeup_locked() does an implicit unlock. */
2850 sowwakeup_locked(so);
2852 /* Detect una wraparound. */
2853 if (!IN_RECOVERY(tp->t_flags) &&
2854 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2855 SEQ_LEQ(th->th_ack, tp->snd_recover))
2856 tp->snd_recover = th->th_ack - 1;
2857 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2858 if (IN_RECOVERY(tp->t_flags) &&
2859 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2860 EXIT_RECOVERY(tp->t_flags);
2862 tp->snd_una = th->th_ack;
2863 if (tp->t_flags & TF_SACK_PERMIT) {
2864 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2865 tp->snd_recover = tp->snd_una;
2867 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2868 tp->snd_nxt = tp->snd_una;
2870 switch (tp->t_state) {
2873 * In FIN_WAIT_1 STATE in addition to the processing
2874 * for the ESTABLISHED state if our FIN is now acknowledged
2875 * then enter FIN_WAIT_2.
2877 case TCPS_FIN_WAIT_1:
2878 if (ourfinisacked) {
2880 * If we can't receive any more
2881 * data, then closing user can proceed.
2882 * Starting the timer is contrary to the
2883 * specification, but if we don't get a FIN
2884 * we'll hang forever.
2887 * we should release the tp also, and use a
2890 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2891 soisdisconnected(so);
2892 tcp_timer_activate(tp, TT_2MSL,
2893 (tcp_fast_finwait2_recycle ?
2894 tcp_finwait2_timeout :
2897 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2902 * In CLOSING STATE in addition to the processing for
2903 * the ESTABLISHED state if the ACK acknowledges our FIN
2904 * then enter the TIME-WAIT state, otherwise ignore
2908 if (ourfinisacked) {
2909 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2911 INP_INFO_RUNLOCK(&V_tcbinfo);
2918 * In LAST_ACK, we may still be waiting for data to drain
2919 * and/or to be acked, as well as for the ack of our FIN.
2920 * If our FIN is now acknowledged, delete the TCB,
2921 * enter the closed state and return.
2924 if (ourfinisacked) {
2925 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2934 INP_WLOCK_ASSERT(tp->t_inpcb);
2937 * Update window information.
2938 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2940 if ((thflags & TH_ACK) &&
2941 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2942 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2943 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2944 /* keep track of pure window updates */
2946 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2947 TCPSTAT_INC(tcps_rcvwinupd);
2948 tp->snd_wnd = tiwin;
2949 tp->snd_wl1 = th->th_seq;
2950 tp->snd_wl2 = th->th_ack;
2951 if (tp->snd_wnd > tp->max_sndwnd)
2952 tp->max_sndwnd = tp->snd_wnd;
2957 * Process segments with URG.
2959 if ((thflags & TH_URG) && th->th_urp &&
2960 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2962 * This is a kludge, but if we receive and accept
2963 * random urgent pointers, we'll crash in
2964 * soreceive. It's hard to imagine someone
2965 * actually wanting to send this much urgent data.
2967 SOCKBUF_LOCK(&so->so_rcv);
2968 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2969 th->th_urp = 0; /* XXX */
2970 thflags &= ~TH_URG; /* XXX */
2971 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2972 goto dodata; /* XXX */
2975 * If this segment advances the known urgent pointer,
2976 * then mark the data stream. This should not happen
2977 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2978 * a FIN has been received from the remote side.
2979 * In these states we ignore the URG.
2981 * According to RFC961 (Assigned Protocols),
2982 * the urgent pointer points to the last octet
2983 * of urgent data. We continue, however,
2984 * to consider it to indicate the first octet
2985 * of data past the urgent section as the original
2986 * spec states (in one of two places).
2988 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2989 tp->rcv_up = th->th_seq + th->th_urp;
2990 so->so_oobmark = sbavail(&so->so_rcv) +
2991 (tp->rcv_up - tp->rcv_nxt) - 1;
2992 if (so->so_oobmark == 0)
2993 so->so_rcv.sb_state |= SBS_RCVATMARK;
2995 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2997 SOCKBUF_UNLOCK(&so->so_rcv);
2999 * Remove out of band data so doesn't get presented to user.
3000 * This can happen independent of advancing the URG pointer,
3001 * but if two URG's are pending at once, some out-of-band
3002 * data may creep in... ick.
3004 if (th->th_urp <= (u_long)tlen &&
3005 !(so->so_options & SO_OOBINLINE)) {
3006 /* hdr drop is delayed */
3007 tcp_pulloutofband(so, th, m, drop_hdrlen);
3011 * If no out of band data is expected,
3012 * pull receive urgent pointer along
3013 * with the receive window.
3015 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3016 tp->rcv_up = tp->rcv_nxt;
3019 INP_WLOCK_ASSERT(tp->t_inpcb);
3022 * Process the segment text, merging it into the TCP sequencing queue,
3023 * and arranging for acknowledgment of receipt if necessary.
3024 * This process logically involves adjusting tp->rcv_wnd as data
3025 * is presented to the user (this happens in tcp_usrreq.c,
3026 * case PRU_RCVD). If a FIN has already been received on this
3027 * connection then we just ignore the text.
3029 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3030 (tp->t_flags & TF_FASTOPEN));
3031 if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
3032 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3033 tcp_seq save_start = th->th_seq;
3034 m_adj(m, drop_hdrlen); /* delayed header drop */
3036 * Insert segment which includes th into TCP reassembly queue
3037 * with control block tp. Set thflags to whether reassembly now
3038 * includes a segment with FIN. This handles the common case
3039 * inline (segment is the next to be received on an established
3040 * connection, and the queue is empty), avoiding linkage into
3041 * and removal from the queue and repetition of various
3043 * Set DELACK for segments received in order, but ack
3044 * immediately when segments are out of order (so
3045 * fast retransmit can work).
3047 if (th->th_seq == tp->rcv_nxt &&
3049 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3051 if (DELAY_ACK(tp, tlen) || tfo_syn)
3052 tp->t_flags |= TF_DELACK;
3054 tp->t_flags |= TF_ACKNOW;
3055 tp->rcv_nxt += tlen;
3056 thflags = th->th_flags & TH_FIN;
3057 TCPSTAT_INC(tcps_rcvpack);
3058 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3059 SOCKBUF_LOCK(&so->so_rcv);
3060 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3063 sbappendstream_locked(&so->so_rcv, m, 0);
3064 /* NB: sorwakeup_locked() does an implicit unlock. */
3065 sorwakeup_locked(so);
3068 * XXX: Due to the header drop above "th" is
3069 * theoretically invalid by now. Fortunately
3070 * m_adj() doesn't actually frees any mbufs
3071 * when trimming from the head.
3073 thflags = tcp_reass(tp, th, &save_start, &tlen, m);
3074 tp->t_flags |= TF_ACKNOW;
3076 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
3077 tcp_update_sack_list(tp, save_start, save_start + tlen);
3080 * Note the amount of data that peer has sent into
3081 * our window, in order to estimate the sender's
3085 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3086 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3088 len = so->so_rcv.sb_hiwat;
3096 * If FIN is received ACK the FIN and let the user know
3097 * that the connection is closing.
3099 if (thflags & TH_FIN) {
3100 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3103 * If connection is half-synchronized
3104 * (ie NEEDSYN flag on) then delay ACK,
3105 * so it may be piggybacked when SYN is sent.
3106 * Otherwise, since we received a FIN then no
3107 * more input can be expected, send ACK now.
3109 if (tp->t_flags & TF_NEEDSYN)
3110 tp->t_flags |= TF_DELACK;
3112 tp->t_flags |= TF_ACKNOW;
3115 switch (tp->t_state) {
3118 * In SYN_RECEIVED and ESTABLISHED STATES
3119 * enter the CLOSE_WAIT state.
3121 case TCPS_SYN_RECEIVED:
3122 tp->t_starttime = ticks;
3124 case TCPS_ESTABLISHED:
3125 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3129 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3130 * enter the CLOSING state.
3132 case TCPS_FIN_WAIT_1:
3133 tcp_state_change(tp, TCPS_CLOSING);
3137 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3138 * starting the time-wait timer, turning off the other
3141 case TCPS_FIN_WAIT_2:
3142 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
3143 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata "
3144 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
3148 INP_INFO_RUNLOCK(&V_tcbinfo);
3152 if (ti_locked == TI_RLOCKED)
3153 INP_INFO_RUNLOCK(&V_tcbinfo);
3154 ti_locked = TI_UNLOCKED;
3157 if (so->so_options & SO_DEBUG)
3158 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3161 TCP_PROBE3(debug__input, tp, th, m);
3164 * Return any desired output.
3166 if (needoutput || (tp->t_flags & TF_ACKNOW))
3167 (void) tp->t_fb->tfb_tcp_output(tp);
3170 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3171 __func__, ti_locked));
3172 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3173 INP_WLOCK_ASSERT(tp->t_inpcb);
3175 if (tp->t_flags & TF_DELACK) {
3176 tp->t_flags &= ~TF_DELACK;
3177 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3179 INP_WUNLOCK(tp->t_inpcb);
3184 * Generate an ACK dropping incoming segment if it occupies
3185 * sequence space, where the ACK reflects our state.
3187 * We can now skip the test for the RST flag since all
3188 * paths to this code happen after packets containing
3189 * RST have been dropped.
3191 * In the SYN-RECEIVED state, don't send an ACK unless the
3192 * segment we received passes the SYN-RECEIVED ACK test.
3193 * If it fails send a RST. This breaks the loop in the
3194 * "LAND" DoS attack, and also prevents an ACK storm
3195 * between two listening ports that have been sent forged
3196 * SYN segments, each with the source address of the other.
3198 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3199 (SEQ_GT(tp->snd_una, th->th_ack) ||
3200 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3201 rstreason = BANDLIM_RST_OPENPORT;
3205 if (so->so_options & SO_DEBUG)
3206 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3209 TCP_PROBE3(debug__input, tp, th, m);
3210 if (ti_locked == TI_RLOCKED)
3211 INP_INFO_RUNLOCK(&V_tcbinfo);
3212 ti_locked = TI_UNLOCKED;
3214 tp->t_flags |= TF_ACKNOW;
3215 (void) tp->t_fb->tfb_tcp_output(tp);
3216 INP_WUNLOCK(tp->t_inpcb);
3221 if (ti_locked == TI_RLOCKED)
3222 INP_INFO_RUNLOCK(&V_tcbinfo);
3223 ti_locked = TI_UNLOCKED;
3226 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3227 INP_WUNLOCK(tp->t_inpcb);
3229 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3233 if (ti_locked == TI_RLOCKED) {
3234 INP_INFO_RUNLOCK(&V_tcbinfo);
3235 ti_locked = TI_UNLOCKED;
3239 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3243 * Drop space held by incoming segment and return.
3246 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3247 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3250 TCP_PROBE3(debug__input, tp, th, m);
3252 INP_WUNLOCK(tp->t_inpcb);
3257 * Issue RST and make ACK acceptable to originator of segment.
3258 * The mbuf must still include the original packet header.
3262 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3263 int tlen, int rstreason)
3269 struct ip6_hdr *ip6;
3273 INP_WLOCK_ASSERT(tp->t_inpcb);
3276 /* Don't bother if destination was broadcast/multicast. */
3277 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3280 if (mtod(m, struct ip *)->ip_v == 6) {
3281 ip6 = mtod(m, struct ip6_hdr *);
3282 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3283 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3285 /* IPv6 anycast check is done at tcp6_input() */
3288 #if defined(INET) && defined(INET6)
3293 ip = mtod(m, struct ip *);
3294 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3295 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3296 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3297 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3302 /* Perform bandwidth limiting. */
3303 if (badport_bandlim(rstreason) < 0)
3306 /* tcp_respond consumes the mbuf chain. */
3307 if (th->th_flags & TH_ACK) {
3308 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3309 th->th_ack, TH_RST);
3311 if (th->th_flags & TH_SYN)
3313 if (th->th_flags & TH_FIN)
3315 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3316 (tcp_seq)0, TH_RST|TH_ACK);
3324 * Parse TCP options and place in tcpopt.
3327 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3332 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3334 if (opt == TCPOPT_EOL)
3336 if (opt == TCPOPT_NOP)
3342 if (optlen < 2 || optlen > cnt)
3347 if (optlen != TCPOLEN_MAXSEG)
3349 if (!(flags & TO_SYN))
3351 to->to_flags |= TOF_MSS;
3352 bcopy((char *)cp + 2,
3353 (char *)&to->to_mss, sizeof(to->to_mss));
3354 to->to_mss = ntohs(to->to_mss);
3357 if (optlen != TCPOLEN_WINDOW)
3359 if (!(flags & TO_SYN))
3361 to->to_flags |= TOF_SCALE;
3362 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3364 case TCPOPT_TIMESTAMP:
3365 if (optlen != TCPOLEN_TIMESTAMP)
3367 to->to_flags |= TOF_TS;
3368 bcopy((char *)cp + 2,
3369 (char *)&to->to_tsval, sizeof(to->to_tsval));
3370 to->to_tsval = ntohl(to->to_tsval);
3371 bcopy((char *)cp + 6,
3372 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3373 to->to_tsecr = ntohl(to->to_tsecr);
3375 case TCPOPT_SIGNATURE:
3377 * In order to reply to a host which has set the
3378 * TCP_SIGNATURE option in its initial SYN, we have
3379 * to record the fact that the option was observed
3380 * here for the syncache code to perform the correct
3383 if (optlen != TCPOLEN_SIGNATURE)
3385 to->to_flags |= TOF_SIGNATURE;
3386 to->to_signature = cp + 2;
3388 case TCPOPT_SACK_PERMITTED:
3389 if (optlen != TCPOLEN_SACK_PERMITTED)
3391 if (!(flags & TO_SYN))
3395 to->to_flags |= TOF_SACKPERM;
3398 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3402 to->to_flags |= TOF_SACK;
3403 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3404 to->to_sacks = cp + 2;
3405 TCPSTAT_INC(tcps_sack_rcv_blocks);
3408 case TCPOPT_FAST_OPEN:
3409 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) &&
3410 (optlen < TCPOLEN_FAST_OPEN_MIN) &&
3411 (optlen > TCPOLEN_FAST_OPEN_MAX))
3413 if (!(flags & TO_SYN))
3415 if (!V_tcp_fastopen_enabled)
3417 to->to_flags |= TOF_FASTOPEN;
3418 to->to_tfo_len = optlen - 2;
3419 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3429 * Pull out of band byte out of a segment so
3430 * it doesn't appear in the user's data queue.
3431 * It is still reflected in the segment length for
3432 * sequencing purposes.
3435 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3438 int cnt = off + th->th_urp - 1;
3441 if (m->m_len > cnt) {
3442 char *cp = mtod(m, caddr_t) + cnt;
3443 struct tcpcb *tp = sototcpcb(so);
3445 INP_WLOCK_ASSERT(tp->t_inpcb);
3448 tp->t_oobflags |= TCPOOB_HAVEDATA;
3449 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3451 if (m->m_flags & M_PKTHDR)
3460 panic("tcp_pulloutofband");
3464 * Collect new round-trip time estimate
3465 * and update averages and current timeout.
3468 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3472 INP_WLOCK_ASSERT(tp->t_inpcb);
3474 TCPSTAT_INC(tcps_rttupdated);
3476 if (tp->t_srtt != 0) {
3478 * srtt is stored as fixed point with 5 bits after the
3479 * binary point (i.e., scaled by 8). The following magic
3480 * is equivalent to the smoothing algorithm in rfc793 with
3481 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3482 * point). Adjust rtt to origin 0.
3484 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3485 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3487 if ((tp->t_srtt += delta) <= 0)
3491 * We accumulate a smoothed rtt variance (actually, a
3492 * smoothed mean difference), then set the retransmit
3493 * timer to smoothed rtt + 4 times the smoothed variance.
3494 * rttvar is stored as fixed point with 4 bits after the
3495 * binary point (scaled by 16). The following is
3496 * equivalent to rfc793 smoothing with an alpha of .75
3497 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3498 * rfc793's wired-in beta.
3502 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3503 if ((tp->t_rttvar += delta) <= 0)
3505 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3506 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3509 * No rtt measurement yet - use the unsmoothed rtt.
3510 * Set the variance to half the rtt (so our first
3511 * retransmit happens at 3*rtt).
3513 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3514 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3515 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3521 * the retransmit should happen at rtt + 4 * rttvar.
3522 * Because of the way we do the smoothing, srtt and rttvar
3523 * will each average +1/2 tick of bias. When we compute
3524 * the retransmit timer, we want 1/2 tick of rounding and
3525 * 1 extra tick because of +-1/2 tick uncertainty in the
3526 * firing of the timer. The bias will give us exactly the
3527 * 1.5 tick we need. But, because the bias is
3528 * statistical, we have to test that we don't drop below
3529 * the minimum feasible timer (which is 2 ticks).
3531 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3532 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3535 * We received an ack for a packet that wasn't retransmitted;
3536 * it is probably safe to discard any error indications we've
3537 * received recently. This isn't quite right, but close enough
3538 * for now (a route might have failed after we sent a segment,
3539 * and the return path might not be symmetrical).
3541 tp->t_softerror = 0;
3545 * Determine a reasonable value for maxseg size.
3546 * If the route is known, check route for mtu.
3547 * If none, use an mss that can be handled on the outgoing interface
3548 * without forcing IP to fragment. If no route is found, route has no mtu,
3549 * or the destination isn't local, use a default, hopefully conservative
3550 * size (usually 512 or the default IP max size, but no more than the mtu
3551 * of the interface), as we can't discover anything about intervening
3552 * gateways or networks. We also initialize the congestion/slow start
3553 * window to be a single segment if the destination isn't local.
3554 * While looking at the routing entry, we also initialize other path-dependent
3555 * parameters from pre-set or cached values in the routing entry.
3557 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3558 * IP options, e.g. IPSEC data, since length of this data may vary, and
3559 * thus it is calculated for every segment separately in tcp_output().
3561 * NOTE that this routine is only called when we process an incoming
3562 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3563 * settings are handled in tcp_mssopt().
3566 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3567 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3571 struct inpcb *inp = tp->t_inpcb;
3572 struct hc_metrics_lite metrics;
3574 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3575 size_t min_protoh = isipv6 ?
3576 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3577 sizeof (struct tcpiphdr);
3579 const size_t min_protoh = sizeof(struct tcpiphdr);
3582 INP_WLOCK_ASSERT(tp->t_inpcb);
3584 if (mtuoffer != -1) {
3585 KASSERT(offer == -1, ("%s: conflict", __func__));
3586 offer = mtuoffer - min_protoh;
3592 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3593 tp->t_maxseg = V_tcp_v6mssdflt;
3596 #if defined(INET) && defined(INET6)
3601 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3602 tp->t_maxseg = V_tcp_mssdflt;
3607 * No route to sender, stay with default mss and return.
3611 * In case we return early we need to initialize metrics
3612 * to a defined state as tcp_hc_get() would do for us
3613 * if there was no cache hit.
3615 if (metricptr != NULL)
3616 bzero(metricptr, sizeof(struct hc_metrics_lite));
3620 /* What have we got? */
3624 * Offer == 0 means that there was no MSS on the SYN
3625 * segment, in this case we use tcp_mssdflt as
3626 * already assigned to t_maxseg above.
3628 offer = tp->t_maxseg;
3633 * Offer == -1 means that we didn't receive SYN yet.
3639 * Prevent DoS attack with too small MSS. Round up
3640 * to at least minmss.
3642 offer = max(offer, V_tcp_minmss);
3646 * rmx information is now retrieved from tcp_hostcache.
3648 tcp_hc_get(&inp->inp_inc, &metrics);
3649 if (metricptr != NULL)
3650 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3653 * If there's a discovered mtu in tcp hostcache, use it.
3654 * Else, use the link mtu.
3656 if (metrics.rmx_mtu)
3657 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3661 mss = maxmtu - min_protoh;
3662 if (!V_path_mtu_discovery &&
3663 !in6_localaddr(&inp->in6p_faddr))
3664 mss = min(mss, V_tcp_v6mssdflt);
3667 #if defined(INET) && defined(INET6)
3672 mss = maxmtu - min_protoh;
3673 if (!V_path_mtu_discovery &&
3674 !in_localaddr(inp->inp_faddr))
3675 mss = min(mss, V_tcp_mssdflt);
3679 * XXX - The above conditional (mss = maxmtu - min_protoh)
3680 * probably violates the TCP spec.
3681 * The problem is that, since we don't know the
3682 * other end's MSS, we are supposed to use a conservative
3683 * default. But, if we do that, then MTU discovery will
3684 * never actually take place, because the conservative
3685 * default is much less than the MTUs typically seen
3686 * on the Internet today. For the moment, we'll sweep
3687 * this under the carpet.
3689 * The conservative default might not actually be a problem
3690 * if the only case this occurs is when sending an initial
3691 * SYN with options and data to a host we've never talked
3692 * to before. Then, they will reply with an MSS value which
3693 * will get recorded and the new parameters should get
3694 * recomputed. For Further Study.
3697 mss = min(mss, offer);
3700 * Sanity check: make sure that maxseg will be large
3701 * enough to allow some data on segments even if the
3702 * all the option space is used (40bytes). Otherwise
3703 * funny things may happen in tcp_output.
3705 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3713 tcp_mss(struct tcpcb *tp, int offer)
3719 struct hc_metrics_lite metrics;
3720 struct tcp_ifcap cap;
3722 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3724 bzero(&cap, sizeof(cap));
3725 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3731 * If there's a pipesize, change the socket buffer to that size,
3732 * don't change if sb_hiwat is different than default (then it
3733 * has been changed on purpose with setsockopt).
3734 * Make the socket buffers an integral number of mss units;
3735 * if the mss is larger than the socket buffer, decrease the mss.
3737 so = inp->inp_socket;
3738 SOCKBUF_LOCK(&so->so_snd);
3739 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3740 bufsize = metrics.rmx_sendpipe;
3742 bufsize = so->so_snd.sb_hiwat;
3746 bufsize = roundup(bufsize, mss);
3747 if (bufsize > sb_max)
3749 if (bufsize > so->so_snd.sb_hiwat)
3750 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3752 SOCKBUF_UNLOCK(&so->so_snd);
3754 * Sanity check: make sure that maxseg will be large
3755 * enough to allow some data on segments even if the
3756 * all the option space is used (40bytes). Otherwise
3757 * funny things may happen in tcp_output.
3759 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3761 tp->t_maxseg = max(mss, 64);
3763 SOCKBUF_LOCK(&so->so_rcv);
3764 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3765 bufsize = metrics.rmx_recvpipe;
3767 bufsize = so->so_rcv.sb_hiwat;
3768 if (bufsize > mss) {
3769 bufsize = roundup(bufsize, mss);
3770 if (bufsize > sb_max)
3772 if (bufsize > so->so_rcv.sb_hiwat)
3773 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3775 SOCKBUF_UNLOCK(&so->so_rcv);
3777 /* Check the interface for TSO capabilities. */
3778 if (cap.ifcap & CSUM_TSO) {
3779 tp->t_flags |= TF_TSO;
3780 tp->t_tsomax = cap.tsomax;
3781 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3782 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3787 * Determine the MSS option to send on an outgoing SYN.
3790 tcp_mssopt(struct in_conninfo *inc)
3797 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3800 if (inc->inc_flags & INC_ISIPV6) {
3801 mss = V_tcp_v6mssdflt;
3802 maxmtu = tcp_maxmtu6(inc, NULL);
3803 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3806 #if defined(INET) && defined(INET6)
3811 mss = V_tcp_mssdflt;
3812 maxmtu = tcp_maxmtu(inc, NULL);
3813 min_protoh = sizeof(struct tcpiphdr);
3816 #if defined(INET6) || defined(INET)
3817 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3820 if (maxmtu && thcmtu)
3821 mss = min(maxmtu, thcmtu) - min_protoh;
3822 else if (maxmtu || thcmtu)
3823 mss = max(maxmtu, thcmtu) - min_protoh;
3830 * On a partial ack arrives, force the retransmission of the
3831 * next unacknowledged segment. Do not clear tp->t_dupacks.
3832 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3836 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3838 tcp_seq onxt = tp->snd_nxt;
3839 u_long ocwnd = tp->snd_cwnd;
3840 u_int maxseg = tcp_maxseg(tp);
3842 INP_WLOCK_ASSERT(tp->t_inpcb);
3844 tcp_timer_activate(tp, TT_REXMT, 0);
3846 tp->snd_nxt = th->th_ack;
3848 * Set snd_cwnd to one segment beyond acknowledged offset.
3849 * (tp->snd_una has not yet been updated when this function is called.)
3851 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3852 tp->t_flags |= TF_ACKNOW;
3853 (void) tp->t_fb->tfb_tcp_output(tp);
3854 tp->snd_cwnd = ocwnd;
3855 if (SEQ_GT(onxt, tp->snd_nxt))
3858 * Partial window deflation. Relies on fact that tp->snd_una
3861 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3862 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3865 tp->snd_cwnd += maxseg;
3869 tcp_compute_pipe(struct tcpcb *tp)
3871 return (tp->snd_max - tp->snd_una +
3872 tp->sackhint.sack_bytes_rexmit -
3873 tp->sackhint.sacked_bytes);