2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (c) 2007-2008,2010
5 * Swinburne University of Technology, Melbourne, Australia.
6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
7 * Copyright (c) 2010 The FreeBSD Foundation
8 * Copyright (c) 2010-2011 Juniper Networks, Inc.
11 * Portions of this software were developed at the Centre for Advanced Internet
12 * Architectures, Swinburne University of Technology, by Lawrence Stewart,
13 * James Healy and David Hayes, made possible in part by a grant from the Cisco
14 * University Research Program Fund at Community Foundation Silicon Valley.
16 * Portions of this software were developed at the Centre for Advanced
17 * Internet Architectures, Swinburne University of Technology, Melbourne,
18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
20 * Portions of this software were developed by Robert N. M. Watson under
21 * contract to Juniper Networks, Inc.
23 * Redistribution and use in source and binary forms, with or without
24 * modification, are permitted provided that the following conditions
26 * 1. Redistributions of source code must retain the above copyright
27 * notice, this list of conditions and the following disclaimer.
28 * 2. Redistributions in binary form must reproduce the above copyright
29 * notice, this list of conditions and the following disclaimer in the
30 * documentation and/or other materials provided with the distribution.
31 * 4. Neither the name of the University nor the names of its contributors
32 * may be used to endorse or promote products derived from this software
33 * without specific prior written permission.
35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD$");
54 #include "opt_inet6.h"
55 #include "opt_ipsec.h"
56 #include "opt_tcpdebug.h"
58 #include <sys/param.h>
59 #include <sys/kernel.h>
61 #include <sys/hhook.h>
63 #include <sys/malloc.h>
65 #include <sys/proc.h> /* for proc0 declaration */
66 #include <sys/protosw.h>
68 #include <sys/signalvar.h>
69 #include <sys/socket.h>
70 #include <sys/socketvar.h>
71 #include <sys/sysctl.h>
72 #include <sys/syslog.h>
73 #include <sys/systm.h>
75 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
80 #include <net/if_var.h>
81 #include <net/route.h>
84 #define TCPSTATES /* for logging */
86 #include <netinet/in.h>
87 #include <netinet/in_kdtrace.h>
88 #include <netinet/in_pcb.h>
89 #include <netinet/in_systm.h>
90 #include <netinet/ip.h>
91 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
92 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
93 #include <netinet/ip_var.h>
94 #include <netinet/ip_options.h>
95 #include <netinet/ip6.h>
96 #include <netinet/icmp6.h>
97 #include <netinet6/in6_pcb.h>
98 #include <netinet6/in6_var.h>
99 #include <netinet6/ip6_var.h>
100 #include <netinet6/nd6.h>
102 #include <netinet/tcp_fastopen.h>
104 #include <netinet/tcp.h>
105 #include <netinet/tcp_fsm.h>
106 #include <netinet/tcp_seq.h>
107 #include <netinet/tcp_timer.h>
108 #include <netinet/tcp_var.h>
109 #include <netinet6/tcp6_var.h>
110 #include <netinet/tcpip.h>
111 #include <netinet/cc/cc.h>
113 #include <netinet/tcp_pcap.h>
115 #include <netinet/tcp_syncache.h>
117 #include <netinet/tcp_debug.h>
118 #endif /* TCPDEBUG */
120 #include <netinet/tcp_offload.h>
124 #include <netipsec/ipsec.h>
125 #include <netipsec/ipsec6.h>
128 #include <machine/in_cksum.h>
130 #include <security/mac/mac_framework.h>
132 const int tcprexmtthresh = 3;
134 int tcp_log_in_vain = 0;
135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
137 "Log all incoming TCP segments to closed ports");
139 VNET_DEFINE(int, blackhole) = 0;
140 #define V_blackhole VNET(blackhole)
141 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
142 &VNET_NAME(blackhole), 0,
143 "Do not send RST on segments to closed ports");
145 VNET_DEFINE(int, tcp_delack_enabled) = 1;
146 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
147 &VNET_NAME(tcp_delack_enabled), 0,
148 "Delay ACK to try and piggyback it onto a data packet");
150 VNET_DEFINE(int, drop_synfin) = 0;
151 #define V_drop_synfin VNET(drop_synfin)
152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
153 &VNET_NAME(drop_synfin), 0,
154 "Drop TCP packets with SYN+FIN set");
156 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
158 &VNET_NAME(tcp_do_rfc6675_pipe), 0,
159 "Use calculated pipe/in-flight bytes per RFC 6675");
161 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
162 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
164 &VNET_NAME(tcp_do_rfc3042), 0,
165 "Enable RFC 3042 (Limited Transmit)");
167 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
169 &VNET_NAME(tcp_do_rfc3390), 0,
170 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
172 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
174 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
175 "Slow-start flight size (initial congestion window) in number of segments");
177 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
179 &VNET_NAME(tcp_do_rfc3465), 0,
180 "Enable RFC 3465 (Appropriate Byte Counting)");
182 VNET_DEFINE(int, tcp_abc_l_var) = 2;
183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
184 &VNET_NAME(tcp_abc_l_var), 2,
185 "Cap the max cwnd increment during slow-start to this number of segments");
187 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
189 VNET_DEFINE(int, tcp_do_ecn) = 2;
190 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
191 &VNET_NAME(tcp_do_ecn), 0,
194 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
195 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
196 &VNET_NAME(tcp_ecn_maxretries), 0,
197 "Max retries before giving up on ECN");
199 VNET_DEFINE(int, tcp_insecure_syn) = 0;
200 #define V_tcp_insecure_syn VNET(tcp_insecure_syn)
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
202 &VNET_NAME(tcp_insecure_syn), 0,
203 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
205 VNET_DEFINE(int, tcp_insecure_rst) = 0;
206 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
207 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
208 &VNET_NAME(tcp_insecure_rst), 0,
209 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
211 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
212 #define V_tcp_recvspace VNET(tcp_recvspace)
213 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
214 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
216 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
217 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
218 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
219 &VNET_NAME(tcp_do_autorcvbuf), 0,
220 "Enable automatic receive buffer sizing");
222 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
223 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW,
225 &VNET_NAME(tcp_autorcvbuf_inc), 0,
226 "Incrementor step size of automatic receive buffer");
228 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
229 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
230 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
231 &VNET_NAME(tcp_autorcvbuf_max), 0,
232 "Max size of automatic receive buffer");
234 VNET_DEFINE(struct inpcbhead, tcb);
235 #define tcb6 tcb /* for KAME src sync over BSD*'s */
236 VNET_DEFINE(struct inpcbinfo, tcbinfo);
239 * TCP statistics are stored in an array of counter(9)s, which size matches
240 * size of struct tcpstat. TCP running connection count is a regular array.
242 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
243 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
244 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
245 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
246 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
247 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
248 "TCP connection counts by TCP state");
251 tcp_vnet_init(const void *unused)
254 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
255 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
257 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
258 tcp_vnet_init, NULL);
262 tcp_vnet_uninit(const void *unused)
265 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
266 VNET_PCPUSTAT_FREE(tcpstat);
268 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
269 tcp_vnet_uninit, NULL);
273 * Kernel module interface for updating tcpstat. The argument is an index
274 * into tcpstat treated as an array.
277 kmod_tcpstat_inc(int statnum)
280 counter_u64_add(VNET(tcpstat)[statnum], 1);
285 * Wrapper for the TCP established input helper hook.
288 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
290 struct tcp_hhook_data hhook_data;
292 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
297 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
304 * CC wrapper hook functions
307 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
310 INP_WLOCK_ASSERT(tp->t_inpcb);
312 tp->ccv->nsegs = nsegs;
313 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
314 if (tp->snd_cwnd <= tp->snd_wnd)
315 tp->ccv->flags |= CCF_CWND_LIMITED;
317 tp->ccv->flags &= ~CCF_CWND_LIMITED;
319 if (type == CC_ACK) {
320 if (tp->snd_cwnd > tp->snd_ssthresh) {
321 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
322 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp));
323 if (tp->t_bytes_acked >= tp->snd_cwnd) {
324 tp->t_bytes_acked -= tp->snd_cwnd;
325 tp->ccv->flags |= CCF_ABC_SENTAWND;
328 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
329 tp->t_bytes_acked = 0;
333 if (CC_ALGO(tp)->ack_received != NULL) {
334 /* XXXLAS: Find a way to live without this */
335 tp->ccv->curack = th->th_ack;
336 CC_ALGO(tp)->ack_received(tp->ccv, type);
341 cc_conn_init(struct tcpcb *tp)
343 struct hc_metrics_lite metrics;
344 struct inpcb *inp = tp->t_inpcb;
348 INP_WLOCK_ASSERT(tp->t_inpcb);
350 tcp_hc_get(&inp->inp_inc, &metrics);
351 maxseg = tcp_maxseg(tp);
353 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
355 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
356 TCPSTAT_INC(tcps_usedrtt);
357 if (metrics.rmx_rttvar) {
358 tp->t_rttvar = metrics.rmx_rttvar;
359 TCPSTAT_INC(tcps_usedrttvar);
361 /* default variation is +- 1 rtt */
363 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
365 TCPT_RANGESET(tp->t_rxtcur,
366 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
367 tp->t_rttmin, TCPTV_REXMTMAX);
369 if (metrics.rmx_ssthresh) {
371 * There's some sort of gateway or interface
372 * buffer limit on the path. Use this to set
373 * the slow start threshold, but set the
374 * threshold to no less than 2*mss.
376 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
377 TCPSTAT_INC(tcps_usedssthresh);
381 * Set the initial slow-start flight size.
383 * RFC5681 Section 3.1 specifies the default conservative values.
384 * RFC3390 specifies slightly more aggressive values.
385 * RFC6928 increases it to ten segments.
386 * Support for user specified value for initial flight size.
388 * If a SYN or SYN/ACK was lost and retransmitted, we have to
389 * reduce the initial CWND to one segment as congestion is likely
390 * requiring us to be cautious.
392 if (tp->snd_cwnd == 1)
393 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
394 else if (V_tcp_initcwnd_segments)
395 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg,
396 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
397 else if (V_tcp_do_rfc3390)
398 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380));
400 /* Per RFC5681 Section 3.1 */
402 tp->snd_cwnd = 2 * maxseg;
403 else if (maxseg > 1095)
404 tp->snd_cwnd = 3 * maxseg;
406 tp->snd_cwnd = 4 * maxseg;
409 if (CC_ALGO(tp)->conn_init != NULL)
410 CC_ALGO(tp)->conn_init(tp->ccv);
414 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
418 INP_WLOCK_ASSERT(tp->t_inpcb);
422 if (!IN_FASTRECOVERY(tp->t_flags)) {
423 tp->snd_recover = tp->snd_max;
424 if (tp->t_flags & TF_ECN_PERMIT)
425 tp->t_flags |= TF_ECN_SND_CWR;
429 if (!IN_CONGRECOVERY(tp->t_flags)) {
430 TCPSTAT_INC(tcps_ecn_rcwnd);
431 tp->snd_recover = tp->snd_max;
432 if (tp->t_flags & TF_ECN_PERMIT)
433 tp->t_flags |= TF_ECN_SND_CWR;
437 maxseg = tcp_maxseg(tp);
439 tp->t_bytes_acked = 0;
440 EXIT_RECOVERY(tp->t_flags);
441 if (CC_ALGO(tp)->cong_signal == NULL) {
443 * RFC5681 Section 3.1
444 * ssthresh = max (FlightSize / 2, 2*SMSS) eq (4)
447 max((tp->snd_max - tp->snd_una) / 2, 2 * maxseg);
448 tp->snd_cwnd = maxseg;
452 TCPSTAT_INC(tcps_sndrexmitbad);
453 /* RTO was unnecessary, so reset everything. */
454 tp->snd_cwnd = tp->snd_cwnd_prev;
455 tp->snd_ssthresh = tp->snd_ssthresh_prev;
456 tp->snd_recover = tp->snd_recover_prev;
457 if (tp->t_flags & TF_WASFRECOVERY)
458 ENTER_FASTRECOVERY(tp->t_flags);
459 if (tp->t_flags & TF_WASCRECOVERY)
460 ENTER_CONGRECOVERY(tp->t_flags);
461 tp->snd_nxt = tp->snd_max;
462 tp->t_flags &= ~TF_PREVVALID;
467 if (CC_ALGO(tp)->cong_signal != NULL) {
469 tp->ccv->curack = th->th_ack;
470 CC_ALGO(tp)->cong_signal(tp->ccv, type);
475 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
477 INP_WLOCK_ASSERT(tp->t_inpcb);
479 /* XXXLAS: KASSERT that we're in recovery? */
481 if (CC_ALGO(tp)->post_recovery != NULL) {
482 tp->ccv->curack = th->th_ack;
483 CC_ALGO(tp)->post_recovery(tp->ccv);
485 /* XXXLAS: EXIT_RECOVERY ? */
486 tp->t_bytes_acked = 0;
491 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
492 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
496 tcp_fields_to_net(th);
497 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
498 tcp_fields_to_host(th);
504 * Indicate whether this ack should be delayed. We can delay the ack if
505 * following conditions are met:
506 * - There is no delayed ack timer in progress.
507 * - Our last ack wasn't a 0-sized window. We never want to delay
508 * the ack that opens up a 0-sized window.
509 * - LRO wasn't used for this segment. We make sure by checking that the
510 * segment size is not larger than the MSS.
512 #define DELAY_ACK(tp, tlen) \
513 ((!tcp_timer_active(tp, TT_DELACK) && \
514 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
515 (tlen <= tp->t_maxseg) && \
516 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
519 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
521 INP_WLOCK_ASSERT(tp->t_inpcb);
523 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
524 switch (iptos & IPTOS_ECN_MASK) {
526 tp->ccv->flags |= CCF_IPHDR_CE;
529 tp->ccv->flags &= ~CCF_IPHDR_CE;
532 tp->ccv->flags &= ~CCF_IPHDR_CE;
536 if (th->th_flags & TH_CWR)
537 tp->ccv->flags |= CCF_TCPHDR_CWR;
539 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
541 if (tp->t_flags & TF_DELACK)
542 tp->ccv->flags |= CCF_DELACK;
544 tp->ccv->flags &= ~CCF_DELACK;
546 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
548 if (tp->ccv->flags & CCF_ACKNOW)
549 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
554 * TCP input handling is split into multiple parts:
555 * tcp6_input is a thin wrapper around tcp_input for the extended
556 * ip6_protox[] call format in ip6_input
557 * tcp_input handles primary segment validation, inpcb lookup and
558 * SYN processing on listen sockets
559 * tcp_do_segment processes the ACK and text of the segment for
560 * establishing, established and closing connections
564 tcp6_input(struct mbuf **mp, int *offp, int proto)
566 struct mbuf *m = *mp;
567 struct in6_ifaddr *ia6;
570 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
573 * draft-itojun-ipv6-tcp-to-anycast
574 * better place to put this in?
576 ip6 = mtod(m, struct ip6_hdr *);
577 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
578 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
581 ifa_free(&ia6->ia_ifa);
582 ip6 = mtod(m, struct ip6_hdr *);
583 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
584 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
585 return (IPPROTO_DONE);
588 ifa_free(&ia6->ia_ifa);
590 return (tcp_input(mp, offp, proto));
595 tcp_input(struct mbuf **mp, int *offp, int proto)
597 struct mbuf *m = *mp;
598 struct tcphdr *th = NULL;
599 struct ip *ip = NULL;
600 struct inpcb *inp = NULL;
601 struct tcpcb *tp = NULL;
602 struct socket *so = NULL;
612 int rstreason = 0; /* For badport_bandlim accounting purposes */
614 uint8_t sig_checked = 0;
617 struct m_tag *fwd_tag = NULL;
619 struct ip6_hdr *ip6 = NULL;
622 const void *ip6 = NULL;
624 struct tcpopt to; /* options in this segment */
625 char *s = NULL; /* address and port logging */
629 * The size of tcp_saveipgen must be the size of the max ip header,
632 u_char tcp_saveipgen[IP6_HDR_LEN];
633 struct tcphdr tcp_savetcp;
638 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
645 TCPSTAT_INC(tcps_rcvtotal);
649 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
651 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
652 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
654 TCPSTAT_INC(tcps_rcvshort);
655 return (IPPROTO_DONE);
659 ip6 = mtod(m, struct ip6_hdr *);
660 th = (struct tcphdr *)((caddr_t)ip6 + off0);
661 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
662 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
663 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
664 th->th_sum = m->m_pkthdr.csum_data;
666 th->th_sum = in6_cksum_pseudo(ip6, tlen,
667 IPPROTO_TCP, m->m_pkthdr.csum_data);
668 th->th_sum ^= 0xffff;
670 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
672 TCPSTAT_INC(tcps_rcvbadsum);
677 * Be proactive about unspecified IPv6 address in source.
678 * As we use all-zero to indicate unbounded/unconnected pcb,
679 * unspecified IPv6 address can be used to confuse us.
681 * Note that packets with unspecified IPv6 destination is
682 * already dropped in ip6_input.
684 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
688 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
691 #if defined(INET) && defined(INET6)
697 * Get IP and TCP header together in first mbuf.
698 * Note: IP leaves IP header in first mbuf.
700 if (off0 > sizeof (struct ip)) {
702 off0 = sizeof(struct ip);
704 if (m->m_len < sizeof (struct tcpiphdr)) {
705 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
707 TCPSTAT_INC(tcps_rcvshort);
708 return (IPPROTO_DONE);
711 ip = mtod(m, struct ip *);
712 th = (struct tcphdr *)((caddr_t)ip + off0);
713 tlen = ntohs(ip->ip_len) - off0;
716 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
717 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
718 th->th_sum = m->m_pkthdr.csum_data;
720 th->th_sum = in_pseudo(ip->ip_src.s_addr,
722 htonl(m->m_pkthdr.csum_data + tlen +
724 th->th_sum ^= 0xffff;
726 struct ipovly *ipov = (struct ipovly *)ip;
729 * Checksum extended TCP header and data.
732 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
733 ipov->ih_len = htons(tlen);
734 th->th_sum = in_cksum(m, len);
735 /* Reset length for SDT probes. */
736 ip->ip_len = htons(len);
739 /* Re-initialization for later version check */
740 ip->ip_v = IPVERSION;
744 TCPSTAT_INC(tcps_rcvbadsum);
751 * Check that TCP offset makes sense,
752 * pull out TCP options and adjust length. XXX
754 off = th->th_off << 2;
755 if (off < sizeof (struct tcphdr) || off > tlen) {
756 TCPSTAT_INC(tcps_rcvbadoff);
759 tlen -= off; /* tlen is used instead of ti->ti_len */
760 if (off > sizeof (struct tcphdr)) {
763 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
764 ip6 = mtod(m, struct ip6_hdr *);
765 th = (struct tcphdr *)((caddr_t)ip6 + off0);
768 #if defined(INET) && defined(INET6)
773 if (m->m_len < sizeof(struct ip) + off) {
774 if ((m = m_pullup(m, sizeof (struct ip) + off))
776 TCPSTAT_INC(tcps_rcvshort);
777 return (IPPROTO_DONE);
779 ip = mtod(m, struct ip *);
780 th = (struct tcphdr *)((caddr_t)ip + off0);
784 optlen = off - sizeof (struct tcphdr);
785 optp = (u_char *)(th + 1);
787 thflags = th->th_flags;
790 * Convert TCP protocol specific fields to host format.
792 tcp_fields_to_host(th);
795 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
797 drop_hdrlen = off0 + off;
800 * Locate pcb for segment; if we're likely to add or remove a
801 * connection then first acquire pcbinfo lock. There are three cases
802 * where we might discover later we need a write lock despite the
803 * flags: ACKs moving a connection out of the syncache, ACKs for a
804 * connection in TIMEWAIT and SYNs not targeting a listening socket.
806 if ((thflags & (TH_FIN | TH_RST)) != 0) {
807 INP_INFO_RLOCK(&V_tcbinfo);
808 ti_locked = TI_RLOCKED;
810 ti_locked = TI_UNLOCKED;
813 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
817 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
819 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
822 #if defined(INET) && !defined(INET6)
823 (m->m_flags & M_IP_NEXTHOP)
826 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
830 if (ti_locked == TI_RLOCKED) {
831 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
833 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
837 if (isipv6 && fwd_tag != NULL) {
838 struct sockaddr_in6 *next_hop6;
840 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
842 * Transparently forwarded. Pretend to be the destination.
843 * Already got one like this?
845 inp = in6_pcblookup_mbuf(&V_tcbinfo,
846 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
847 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
850 * It's new. Try to find the ambushing socket.
851 * Because we've rewritten the destination address,
852 * any hardware-generated hash is ignored.
854 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
855 th->th_sport, &next_hop6->sin6_addr,
856 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
857 th->th_dport, INPLOOKUP_WILDCARD |
858 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
861 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
862 th->th_sport, &ip6->ip6_dst, th->th_dport,
863 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
864 m->m_pkthdr.rcvif, m);
867 #if defined(INET6) && defined(INET)
871 if (fwd_tag != NULL) {
872 struct sockaddr_in *next_hop;
874 next_hop = (struct sockaddr_in *)(fwd_tag+1);
876 * Transparently forwarded. Pretend to be the destination.
877 * already got one like this?
879 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
880 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
881 m->m_pkthdr.rcvif, m);
884 * It's new. Try to find the ambushing socket.
885 * Because we've rewritten the destination address,
886 * any hardware-generated hash is ignored.
888 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
889 th->th_sport, next_hop->sin_addr,
890 next_hop->sin_port ? ntohs(next_hop->sin_port) :
891 th->th_dport, INPLOOKUP_WILDCARD |
892 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
895 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
896 th->th_sport, ip->ip_dst, th->th_dport,
897 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
898 m->m_pkthdr.rcvif, m);
902 * If the INPCB does not exist then all data in the incoming
903 * segment is discarded and an appropriate RST is sent back.
904 * XXX MRT Send RST using which routing table?
908 * Log communication attempts to ports that are not
911 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
912 tcp_log_in_vain == 2) {
913 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
914 log(LOG_INFO, "%s; %s: Connection attempt "
915 "to closed port\n", s, __func__);
918 * When blackholing do not respond with a RST but
919 * completely ignore the segment and drop it.
921 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
925 rstreason = BANDLIM_RST_CLOSEDPORT;
928 INP_WLOCK_ASSERT(inp);
930 * While waiting for inp lock during the lookup, another thread
931 * can have dropped the inpcb, in which case we need to loop back
932 * and try to find a new inpcb to deliver to.
934 if (inp->inp_flags & INP_DROPPED) {
939 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
940 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
941 ((inp->inp_socket == NULL) ||
942 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
943 inp->inp_flowid = m->m_pkthdr.flowid;
944 inp->inp_flowtype = M_HASHTYPE_GET(m);
948 if (isipv6 && ipsec6_in_reject(m, inp)) {
952 if (ipsec4_in_reject(m, inp) != 0) {
958 * Check the minimum TTL for socket.
960 if (inp->inp_ip_minttl != 0) {
963 if (inp->inp_ip_minttl > ip6->ip6_hlim)
967 if (inp->inp_ip_minttl > ip->ip_ttl)
972 * A previous connection in TIMEWAIT state is supposed to catch stray
973 * or duplicate segments arriving late. If this segment was a
974 * legitimate new connection attempt, the old INPCB gets removed and
975 * we can try again to find a listening socket.
977 * At this point, due to earlier optimism, we may hold only an inpcb
978 * lock, and not the inpcbinfo write lock. If so, we need to try to
979 * acquire it, or if that fails, acquire a reference on the inpcb,
980 * drop all locks, acquire a global write lock, and then re-acquire
981 * the inpcb lock. We may at that point discover that another thread
982 * has tried to free the inpcb, in which case we need to loop back
983 * and try to find a new inpcb to deliver to.
985 * XXXRW: It may be time to rethink timewait locking.
988 if (inp->inp_flags & INP_TIMEWAIT) {
989 if (ti_locked == TI_UNLOCKED) {
990 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
993 INP_INFO_RLOCK(&V_tcbinfo);
994 ti_locked = TI_RLOCKED;
996 if (in_pcbrele_wlocked(inp)) {
999 } else if (inp->inp_flags & INP_DROPPED) {
1005 ti_locked = TI_RLOCKED;
1007 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1009 if (thflags & TH_SYN)
1010 tcp_dooptions(&to, optp, optlen, TO_SYN);
1012 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
1014 if (tcp_twcheck(inp, &to, th, m, tlen))
1016 INP_INFO_RUNLOCK(&V_tcbinfo);
1017 return (IPPROTO_DONE);
1020 * The TCPCB may no longer exist if the connection is winding
1021 * down or it is in the CLOSED state. Either way we drop the
1022 * segment and send an appropriate response.
1024 tp = intotcpcb(inp);
1025 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1026 rstreason = BANDLIM_RST_CLOSEDPORT;
1031 if (tp->t_flags & TF_TOE) {
1032 tcp_offload_input(tp, m);
1033 m = NULL; /* consumed by the TOE driver */
1039 * We've identified a valid inpcb, but it could be that we need an
1040 * inpcbinfo write lock but don't hold it. In this case, attempt to
1041 * acquire using the same strategy as the TIMEWAIT case above. If we
1042 * relock, we have to jump back to 'relocked' as the connection might
1043 * now be in TIMEWAIT.
1046 if ((thflags & (TH_FIN | TH_RST)) != 0)
1047 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1049 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) ||
1050 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) &&
1051 !IS_FASTOPEN(tp->t_flags)))) {
1052 if (ti_locked == TI_UNLOCKED) {
1053 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
1056 INP_INFO_RLOCK(&V_tcbinfo);
1057 ti_locked = TI_RLOCKED;
1059 if (in_pcbrele_wlocked(inp)) {
1062 } else if (inp->inp_flags & INP_DROPPED) {
1069 ti_locked = TI_RLOCKED;
1071 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1075 INP_WLOCK_ASSERT(inp);
1076 if (mac_inpcb_check_deliver(inp, m))
1079 so = inp->inp_socket;
1080 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1082 if (so->so_options & SO_DEBUG) {
1083 ostate = tp->t_state;
1086 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1089 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1092 #endif /* TCPDEBUG */
1094 * When the socket is accepting connections (the INPCB is in LISTEN
1095 * state) we look into the SYN cache if this is a new connection
1096 * attempt or the completion of a previous one.
1098 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1099 ("%s: so accepting but tp %p not listening", __func__, tp));
1100 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1101 struct in_conninfo inc;
1103 bzero(&inc, sizeof(inc));
1106 inc.inc_flags |= INC_ISIPV6;
1107 inc.inc6_faddr = ip6->ip6_src;
1108 inc.inc6_laddr = ip6->ip6_dst;
1112 inc.inc_faddr = ip->ip_src;
1113 inc.inc_laddr = ip->ip_dst;
1115 inc.inc_fport = th->th_sport;
1116 inc.inc_lport = th->th_dport;
1117 inc.inc_fibnum = so->so_fibnum;
1120 * Check for an existing connection attempt in syncache if
1121 * the flag is only ACK. A successful lookup creates a new
1122 * socket appended to the listen queue in SYN_RECEIVED state.
1124 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1126 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1128 * Parse the TCP options here because
1129 * syncookies need access to the reflected
1132 tcp_dooptions(&to, optp, optlen, 0);
1134 * NB: syncache_expand() doesn't unlock
1135 * inp and tcpinfo locks.
1137 if (!syncache_expand(&inc, &to, th, &so, m)) {
1139 * No syncache entry or ACK was not
1140 * for our SYN/ACK. Send a RST.
1141 * NB: syncache did its own logging
1142 * of the failure cause.
1144 rstreason = BANDLIM_RST_OPENPORT;
1152 * We completed the 3-way handshake
1153 * but could not allocate a socket
1154 * either due to memory shortage,
1155 * listen queue length limits or
1156 * global socket limits. Send RST
1157 * or wait and have the remote end
1158 * retransmit the ACK for another
1161 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1162 log(LOG_DEBUG, "%s; %s: Listen socket: "
1163 "Socket allocation failed due to "
1164 "limits or memory shortage, %s\n",
1166 V_tcp_sc_rst_sock_fail ?
1167 "sending RST" : "try again");
1168 if (V_tcp_sc_rst_sock_fail) {
1169 rstreason = BANDLIM_UNLIMITED;
1175 * Socket is created in state SYN_RECEIVED.
1176 * Unlock the listen socket, lock the newly
1177 * created socket and update the tp variable.
1179 INP_WUNLOCK(inp); /* listen socket */
1180 inp = sotoinpcb(so);
1182 * New connection inpcb is already locked by
1183 * syncache_expand().
1185 INP_WLOCK_ASSERT(inp);
1186 tp = intotcpcb(inp);
1187 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1188 ("%s: ", __func__));
1189 #ifdef TCP_SIGNATURE
1190 if (sig_checked == 0) {
1191 tcp_dooptions(&to, optp, optlen,
1192 (thflags & TH_SYN) ? TO_SYN : 0);
1193 if (!tcp_signature_verify_input(m, off0, tlen,
1194 optlen, &to, th, tp->t_flags)) {
1197 * In SYN_SENT state if it receives an
1198 * RST, it is allowed for further
1201 if ((thflags & TH_RST) == 0 ||
1202 (tp->t_state == TCPS_SYN_SENT) == 0)
1210 * Process the segment and the data it
1211 * contains. tcp_do_segment() consumes
1212 * the mbuf chain and unlocks the inpcb.
1214 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1216 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1217 return (IPPROTO_DONE);
1220 * Segment flag validation for new connection attempts:
1222 * Our (SYN|ACK) response was rejected.
1223 * Check with syncache and remove entry to prevent
1226 * NB: syncache_chkrst does its own logging of failure
1229 if (thflags & TH_RST) {
1230 syncache_chkrst(&inc, th);
1234 * We can't do anything without SYN.
1236 if ((thflags & TH_SYN) == 0) {
1237 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1238 log(LOG_DEBUG, "%s; %s: Listen socket: "
1239 "SYN is missing, segment ignored\n",
1241 TCPSTAT_INC(tcps_badsyn);
1245 * (SYN|ACK) is bogus on a listen socket.
1247 if (thflags & TH_ACK) {
1248 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1249 log(LOG_DEBUG, "%s; %s: Listen socket: "
1250 "SYN|ACK invalid, segment rejected\n",
1252 syncache_badack(&inc); /* XXX: Not needed! */
1253 TCPSTAT_INC(tcps_badsyn);
1254 rstreason = BANDLIM_RST_OPENPORT;
1258 * If the drop_synfin option is enabled, drop all
1259 * segments with both the SYN and FIN bits set.
1260 * This prevents e.g. nmap from identifying the
1262 * XXX: Poor reasoning. nmap has other methods
1263 * and is constantly refining its stack detection
1265 * XXX: This is a violation of the TCP specification
1266 * and was used by RFC1644.
1268 if ((thflags & TH_FIN) && V_drop_synfin) {
1269 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1270 log(LOG_DEBUG, "%s; %s: Listen socket: "
1271 "SYN|FIN segment ignored (based on "
1272 "sysctl setting)\n", s, __func__);
1273 TCPSTAT_INC(tcps_badsyn);
1277 * Segment's flags are (SYN) or (SYN|FIN).
1279 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1280 * as they do not affect the state of the TCP FSM.
1281 * The data pointed to by TH_URG and th_urp is ignored.
1283 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1284 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1285 KASSERT(thflags & (TH_SYN),
1286 ("%s: Listen socket: TH_SYN not set", __func__));
1289 * If deprecated address is forbidden,
1290 * we do not accept SYN to deprecated interface
1291 * address to prevent any new inbound connection from
1292 * getting established.
1293 * When we do not accept SYN, we send a TCP RST,
1294 * with deprecated source address (instead of dropping
1295 * it). We compromise it as it is much better for peer
1296 * to send a RST, and RST will be the final packet
1299 * If we do not forbid deprecated addresses, we accept
1300 * the SYN packet. RFC2462 does not suggest dropping
1302 * If we decipher RFC2462 5.5.4, it says like this:
1303 * 1. use of deprecated addr with existing
1304 * communication is okay - "SHOULD continue to be
1306 * 2. use of it with new communication:
1307 * (2a) "SHOULD NOT be used if alternate address
1308 * with sufficient scope is available"
1309 * (2b) nothing mentioned otherwise.
1310 * Here we fall into (2b) case as we have no choice in
1311 * our source address selection - we must obey the peer.
1313 * The wording in RFC2462 is confusing, and there are
1314 * multiple description text for deprecated address
1315 * handling - worse, they are not exactly the same.
1316 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1318 if (isipv6 && !V_ip6_use_deprecated) {
1319 struct in6_ifaddr *ia6;
1321 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1323 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1324 ifa_free(&ia6->ia_ifa);
1325 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1326 log(LOG_DEBUG, "%s; %s: Listen socket: "
1327 "Connection attempt to deprecated "
1328 "IPv6 address rejected\n",
1330 rstreason = BANDLIM_RST_OPENPORT;
1334 ifa_free(&ia6->ia_ifa);
1338 * Basic sanity checks on incoming SYN requests:
1339 * Don't respond if the destination is a link layer
1340 * broadcast according to RFC1122 4.2.3.10, p. 104.
1341 * If it is from this socket it must be forged.
1342 * Don't respond if the source or destination is a
1343 * global or subnet broad- or multicast address.
1344 * Note that it is quite possible to receive unicast
1345 * link-layer packets with a broadcast IP address. Use
1346 * in_broadcast() to find them.
1348 if (m->m_flags & (M_BCAST|M_MCAST)) {
1349 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1350 log(LOG_DEBUG, "%s; %s: Listen socket: "
1351 "Connection attempt from broad- or multicast "
1352 "link layer address ignored\n", s, __func__);
1357 if (th->th_dport == th->th_sport &&
1358 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1359 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1360 log(LOG_DEBUG, "%s; %s: Listen socket: "
1361 "Connection attempt to/from self "
1362 "ignored\n", s, __func__);
1365 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1366 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1367 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1368 log(LOG_DEBUG, "%s; %s: Listen socket: "
1369 "Connection attempt from/to multicast "
1370 "address ignored\n", s, __func__);
1375 #if defined(INET) && defined(INET6)
1380 if (th->th_dport == th->th_sport &&
1381 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1382 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1383 log(LOG_DEBUG, "%s; %s: Listen socket: "
1384 "Connection attempt from/to self "
1385 "ignored\n", s, __func__);
1388 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1389 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1390 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1391 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1392 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1393 log(LOG_DEBUG, "%s; %s: Listen socket: "
1394 "Connection attempt from/to broad- "
1395 "or multicast address ignored\n",
1402 * SYN appears to be valid. Create compressed TCP state
1406 if (so->so_options & SO_DEBUG)
1407 tcp_trace(TA_INPUT, ostate, tp,
1408 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1410 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1411 tcp_dooptions(&to, optp, optlen, TO_SYN);
1413 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL))
1414 goto tfo_socket_result;
1416 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1419 * Entry added to syncache and mbuf consumed.
1420 * Only the listen socket is unlocked by syncache_add().
1422 if (ti_locked == TI_RLOCKED) {
1423 INP_INFO_RUNLOCK(&V_tcbinfo);
1424 ti_locked = TI_UNLOCKED;
1426 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1427 return (IPPROTO_DONE);
1428 } else if (tp->t_state == TCPS_LISTEN) {
1430 * When a listen socket is torn down the SO_ACCEPTCONN
1431 * flag is removed first while connections are drained
1432 * from the accept queue in a unlock/lock cycle of the
1433 * ACCEPT_LOCK, opening a race condition allowing a SYN
1434 * attempt go through unhandled.
1439 #ifdef TCP_SIGNATURE
1440 if (sig_checked == 0) {
1441 tcp_dooptions(&to, optp, optlen,
1442 (thflags & TH_SYN) ? TO_SYN : 0);
1443 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1447 * In SYN_SENT state if it receives an RST, it is
1448 * allowed for further processing.
1450 if ((thflags & TH_RST) == 0 ||
1451 (tp->t_state == TCPS_SYN_SENT) == 0)
1458 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1461 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1462 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1463 * the inpcb, and unlocks pcbinfo.
1465 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1466 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1467 return (IPPROTO_DONE);
1470 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1472 if (ti_locked == TI_RLOCKED) {
1473 INP_INFO_RUNLOCK(&V_tcbinfo);
1474 ti_locked = TI_UNLOCKED;
1478 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1479 "ti_locked: %d", __func__, ti_locked));
1480 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1485 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1488 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1489 m = NULL; /* mbuf chain got consumed. */
1494 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1496 if (ti_locked == TI_RLOCKED) {
1497 INP_INFO_RUNLOCK(&V_tcbinfo);
1498 ti_locked = TI_UNLOCKED;
1502 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1503 "ti_locked: %d", __func__, ti_locked));
1504 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1512 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1517 return (IPPROTO_DONE);
1521 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1522 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1525 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1526 int rstreason, todrop, win;
1530 struct in_conninfo *inc;
1539 * The size of tcp_saveipgen must be the size of the max ip header,
1542 u_char tcp_saveipgen[IP6_HDR_LEN];
1543 struct tcphdr tcp_savetcp;
1546 thflags = th->th_flags;
1547 inc = &tp->t_inpcb->inp_inc;
1548 tp->sackhint.last_sack_ack = 0;
1550 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1553 * If this is either a state-changing packet or current state isn't
1554 * established, we require a write lock on tcbinfo. Otherwise, we
1555 * allow the tcbinfo to be in either alocked or unlocked, as the
1556 * caller may have unnecessarily acquired a write lock due to a race.
1558 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1559 tp->t_state != TCPS_ESTABLISHED) {
1560 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for "
1561 "SYN/FIN/RST/!EST", __func__, ti_locked));
1562 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1565 if (ti_locked == TI_RLOCKED)
1566 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1568 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1569 "ti_locked: %d", __func__, ti_locked));
1570 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1574 INP_WLOCK_ASSERT(tp->t_inpcb);
1575 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1577 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1581 /* Save segment, if requested. */
1582 tcp_pcap_add(th, m, &(tp->t_inpkts));
1586 * Segment received on connection.
1587 * Reset idle time and keep-alive timer.
1588 * XXX: This should be done after segment
1589 * validation to ignore broken/spoofed segs.
1591 tp->t_rcvtime = ticks;
1594 * Scale up the window into a 32-bit value.
1595 * For the SYN_SENT state the scale is zero.
1597 tiwin = th->th_win << tp->snd_scale;
1600 * TCP ECN processing.
1602 if (tp->t_flags & TF_ECN_PERMIT) {
1603 if (thflags & TH_CWR)
1604 tp->t_flags &= ~TF_ECN_SND_ECE;
1605 switch (iptos & IPTOS_ECN_MASK) {
1607 tp->t_flags |= TF_ECN_SND_ECE;
1608 TCPSTAT_INC(tcps_ecn_ce);
1610 case IPTOS_ECN_ECT0:
1611 TCPSTAT_INC(tcps_ecn_ect0);
1613 case IPTOS_ECN_ECT1:
1614 TCPSTAT_INC(tcps_ecn_ect1);
1618 /* Process a packet differently from RFC3168. */
1619 cc_ecnpkt_handler(tp, th, iptos);
1621 /* Congestion experienced. */
1622 if (thflags & TH_ECE) {
1623 cc_cong_signal(tp, th, CC_ECN);
1628 * Parse options on any incoming segment.
1630 tcp_dooptions(&to, (u_char *)(th + 1),
1631 (th->th_off << 2) - sizeof(struct tcphdr),
1632 (thflags & TH_SYN) ? TO_SYN : 0);
1635 * If echoed timestamp is later than the current time,
1636 * fall back to non RFC1323 RTT calculation. Normalize
1637 * timestamp if syncookies were used when this connection
1640 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1641 to.to_tsecr -= tp->ts_offset;
1642 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1646 * If timestamps were negotiated during SYN/ACK they should
1647 * appear on every segment during this session and vice versa.
1649 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1650 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1651 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1652 "no action\n", s, __func__);
1656 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1657 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1658 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1659 "no action\n", s, __func__);
1665 * Process options only when we get SYN/ACK back. The SYN case
1666 * for incoming connections is handled in tcp_syncache.
1667 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1668 * or <SYN,ACK>) segment itself is never scaled.
1669 * XXX this is traditional behavior, may need to be cleaned up.
1671 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1672 if ((to.to_flags & TOF_SCALE) &&
1673 (tp->t_flags & TF_REQ_SCALE)) {
1674 tp->t_flags |= TF_RCVD_SCALE;
1675 tp->snd_scale = to.to_wscale;
1678 * Initial send window. It will be updated with
1679 * the next incoming segment to the scaled value.
1681 tp->snd_wnd = th->th_win;
1682 if (to.to_flags & TOF_TS) {
1683 tp->t_flags |= TF_RCVD_TSTMP;
1684 tp->ts_recent = to.to_tsval;
1685 tp->ts_recent_age = tcp_ts_getticks();
1687 if (to.to_flags & TOF_MSS)
1688 tcp_mss(tp, to.to_mss);
1689 if ((tp->t_flags & TF_SACK_PERMIT) &&
1690 (to.to_flags & TOF_SACKPERM) == 0)
1691 tp->t_flags &= ~TF_SACK_PERMIT;
1695 * Header prediction: check for the two common cases
1696 * of a uni-directional data xfer. If the packet has
1697 * no control flags, is in-sequence, the window didn't
1698 * change and we're not retransmitting, it's a
1699 * candidate. If the length is zero and the ack moved
1700 * forward, we're the sender side of the xfer. Just
1701 * free the data acked & wake any higher level process
1702 * that was blocked waiting for space. If the length
1703 * is non-zero and the ack didn't move, we're the
1704 * receiver side. If we're getting packets in-order
1705 * (the reassembly queue is empty), add the data to
1706 * the socket buffer and note that we need a delayed ack.
1707 * Make sure that the hidden state-flags are also off.
1708 * Since we check for TCPS_ESTABLISHED first, it can only
1711 if (tp->t_state == TCPS_ESTABLISHED &&
1712 th->th_seq == tp->rcv_nxt &&
1713 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1714 tp->snd_nxt == tp->snd_max &&
1715 tiwin && tiwin == tp->snd_wnd &&
1716 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1717 LIST_EMPTY(&tp->t_segq) &&
1718 ((to.to_flags & TOF_TS) == 0 ||
1719 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1722 * If last ACK falls within this segment's sequence numbers,
1723 * record the timestamp.
1724 * NOTE that the test is modified according to the latest
1725 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1727 if ((to.to_flags & TOF_TS) != 0 &&
1728 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1729 tp->ts_recent_age = tcp_ts_getticks();
1730 tp->ts_recent = to.to_tsval;
1734 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1735 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1736 !IN_RECOVERY(tp->t_flags) &&
1737 (to.to_flags & TOF_SACK) == 0 &&
1738 TAILQ_EMPTY(&tp->snd_holes)) {
1740 * This is a pure ack for outstanding data.
1742 if (ti_locked == TI_RLOCKED)
1743 INP_INFO_RUNLOCK(&V_tcbinfo);
1744 ti_locked = TI_UNLOCKED;
1746 TCPSTAT_INC(tcps_predack);
1749 * "bad retransmit" recovery.
1751 if (tp->t_rxtshift == 1 &&
1752 tp->t_flags & TF_PREVVALID &&
1753 (int)(ticks - tp->t_badrxtwin) < 0) {
1754 cc_cong_signal(tp, th, CC_RTO_ERR);
1758 * Recalculate the transmit timer / rtt.
1760 * Some boxes send broken timestamp replies
1761 * during the SYN+ACK phase, ignore
1762 * timestamps of 0 or we could calculate a
1763 * huge RTT and blow up the retransmit timer.
1765 if ((to.to_flags & TOF_TS) != 0 &&
1769 t = tcp_ts_getticks() - to.to_tsecr;
1770 if (!tp->t_rttlow || tp->t_rttlow > t)
1773 TCP_TS_TO_TICKS(t) + 1);
1774 } else if (tp->t_rtttime &&
1775 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1776 if (!tp->t_rttlow ||
1777 tp->t_rttlow > ticks - tp->t_rtttime)
1778 tp->t_rttlow = ticks - tp->t_rtttime;
1780 ticks - tp->t_rtttime);
1782 acked = BYTES_THIS_ACK(tp, th);
1785 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1786 hhook_run_tcp_est_in(tp, th, &to);
1789 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1790 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1791 sbdrop(&so->so_snd, acked);
1792 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1793 SEQ_LEQ(th->th_ack, tp->snd_recover))
1794 tp->snd_recover = th->th_ack - 1;
1797 * Let the congestion control algorithm update
1798 * congestion control related information. This
1799 * typically means increasing the congestion
1802 cc_ack_received(tp, th, nsegs, CC_ACK);
1804 tp->snd_una = th->th_ack;
1806 * Pull snd_wl2 up to prevent seq wrap relative
1809 tp->snd_wl2 = th->th_ack;
1814 * If all outstanding data are acked, stop
1815 * retransmit timer, otherwise restart timer
1816 * using current (possibly backed-off) value.
1817 * If process is waiting for space,
1818 * wakeup/selwakeup/signal. If data
1819 * are ready to send, let tcp_output
1820 * decide between more output or persist.
1823 if (so->so_options & SO_DEBUG)
1824 tcp_trace(TA_INPUT, ostate, tp,
1825 (void *)tcp_saveipgen,
1828 TCP_PROBE3(debug__input, tp, th,
1829 mtod(m, const char *));
1830 if (tp->snd_una == tp->snd_max)
1831 tcp_timer_activate(tp, TT_REXMT, 0);
1832 else if (!tcp_timer_active(tp, TT_PERSIST))
1833 tcp_timer_activate(tp, TT_REXMT,
1836 if (sbavail(&so->so_snd))
1837 (void) tp->t_fb->tfb_tcp_output(tp);
1840 } else if (th->th_ack == tp->snd_una &&
1841 tlen <= sbspace(&so->so_rcv)) {
1842 int newsize = 0; /* automatic sockbuf scaling */
1845 * This is a pure, in-sequence data packet with
1846 * nothing on the reassembly queue and we have enough
1847 * buffer space to take it.
1849 if (ti_locked == TI_RLOCKED)
1850 INP_INFO_RUNLOCK(&V_tcbinfo);
1851 ti_locked = TI_UNLOCKED;
1853 /* Clean receiver SACK report if present */
1854 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1855 tcp_clean_sackreport(tp);
1856 TCPSTAT_INC(tcps_preddat);
1857 tp->rcv_nxt += tlen;
1859 * Pull snd_wl1 up to prevent seq wrap relative to
1862 tp->snd_wl1 = th->th_seq;
1864 * Pull rcv_up up to prevent seq wrap relative to
1867 tp->rcv_up = tp->rcv_nxt;
1868 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1869 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1871 if (so->so_options & SO_DEBUG)
1872 tcp_trace(TA_INPUT, ostate, tp,
1873 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1875 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1878 * Automatic sizing of receive socket buffer. Often the send
1879 * buffer size is not optimally adjusted to the actual network
1880 * conditions at hand (delay bandwidth product). Setting the
1881 * buffer size too small limits throughput on links with high
1882 * bandwidth and high delay (eg. trans-continental/oceanic links).
1884 * On the receive side the socket buffer memory is only rarely
1885 * used to any significant extent. This allows us to be much
1886 * more aggressive in scaling the receive socket buffer. For
1887 * the case that the buffer space is actually used to a large
1888 * extent and we run out of kernel memory we can simply drop
1889 * the new segments; TCP on the sender will just retransmit it
1890 * later. Setting the buffer size too big may only consume too
1891 * much kernel memory if the application doesn't read() from
1892 * the socket or packet loss or reordering makes use of the
1895 * The criteria to step up the receive buffer one notch are:
1896 * 1. Application has not set receive buffer size with
1897 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1898 * 2. the number of bytes received during the time it takes
1899 * one timestamp to be reflected back to us (the RTT);
1900 * 3. received bytes per RTT is within seven eighth of the
1901 * current socket buffer size;
1902 * 4. receive buffer size has not hit maximal automatic size;
1904 * This algorithm does one step per RTT at most and only if
1905 * we receive a bulk stream w/o packet losses or reorderings.
1906 * Shrinking the buffer during idle times is not necessary as
1907 * it doesn't consume any memory when idle.
1909 * TODO: Only step up if the application is actually serving
1910 * the buffer to better manage the socket buffer resources.
1912 if (V_tcp_do_autorcvbuf &&
1913 (to.to_flags & TOF_TS) &&
1915 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1916 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1917 to.to_tsecr - tp->rfbuf_ts < hz) {
1919 (so->so_rcv.sb_hiwat / 8 * 7) &&
1920 so->so_rcv.sb_hiwat <
1921 V_tcp_autorcvbuf_max) {
1923 min(so->so_rcv.sb_hiwat +
1924 V_tcp_autorcvbuf_inc,
1925 V_tcp_autorcvbuf_max);
1927 /* Start over with next RTT. */
1931 tp->rfbuf_cnt += tlen; /* add up */
1934 /* Add data to socket buffer. */
1935 SOCKBUF_LOCK(&so->so_rcv);
1936 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1940 * Set new socket buffer size.
1941 * Give up when limit is reached.
1944 if (!sbreserve_locked(&so->so_rcv,
1946 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1947 m_adj(m, drop_hdrlen); /* delayed header drop */
1948 sbappendstream_locked(&so->so_rcv, m, 0);
1950 /* NB: sorwakeup_locked() does an implicit unlock. */
1951 sorwakeup_locked(so);
1952 if (DELAY_ACK(tp, tlen)) {
1953 tp->t_flags |= TF_DELACK;
1955 tp->t_flags |= TF_ACKNOW;
1956 tp->t_fb->tfb_tcp_output(tp);
1963 * Calculate amount of space in receive window,
1964 * and then do TCP input processing.
1965 * Receive window is amount of space in rcv queue,
1966 * but not less than advertised window.
1968 win = sbspace(&so->so_rcv);
1971 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1973 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1977 switch (tp->t_state) {
1980 * If the state is SYN_RECEIVED:
1981 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1983 case TCPS_SYN_RECEIVED:
1984 if ((thflags & TH_ACK) &&
1985 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1986 SEQ_GT(th->th_ack, tp->snd_max))) {
1987 rstreason = BANDLIM_RST_OPENPORT;
1991 if (IS_FASTOPEN(tp->t_flags)) {
1993 * When a TFO connection is in SYN_RECEIVED, the
1994 * only valid packets are the initial SYN, a
1995 * retransmit/copy of the initial SYN (possibly with
1996 * a subset of the original data), a valid ACK, a
1999 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
2000 rstreason = BANDLIM_RST_OPENPORT;
2002 } else if (thflags & TH_SYN) {
2003 /* non-initial SYN is ignored */
2004 if ((tcp_timer_active(tp, TT_DELACK) ||
2005 tcp_timer_active(tp, TT_REXMT)))
2007 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2015 * If the state is SYN_SENT:
2016 * if seg contains an ACK, but not for our SYN, drop the input.
2017 * if seg contains a RST, then drop the connection.
2018 * if seg does not contain SYN, then drop it.
2019 * Otherwise this is an acceptable SYN segment
2020 * initialize tp->rcv_nxt and tp->irs
2021 * if seg contains ack then advance tp->snd_una
2022 * if seg contains an ECE and ECN support is enabled, the stream
2024 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2025 * arrange for segment to be acked (eventually)
2026 * continue processing rest of data/controls, beginning with URG
2029 if ((thflags & TH_ACK) &&
2030 (SEQ_LEQ(th->th_ack, tp->iss) ||
2031 SEQ_GT(th->th_ack, tp->snd_max))) {
2032 rstreason = BANDLIM_UNLIMITED;
2035 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2036 TCP_PROBE5(connect__refused, NULL, tp,
2037 mtod(m, const char *), tp, th);
2038 tp = tcp_drop(tp, ECONNREFUSED);
2040 if (thflags & TH_RST)
2042 if (!(thflags & TH_SYN))
2045 tp->irs = th->th_seq;
2047 if (thflags & TH_ACK) {
2048 TCPSTAT_INC(tcps_connects);
2051 mac_socketpeer_set_from_mbuf(m, so);
2053 /* Do window scaling on this connection? */
2054 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2055 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2056 tp->rcv_scale = tp->request_r_scale;
2058 tp->rcv_adv += min(tp->rcv_wnd,
2059 TCP_MAXWIN << tp->rcv_scale);
2060 tp->snd_una++; /* SYN is acked */
2062 * If there's data, delay ACK; if there's also a FIN
2063 * ACKNOW will be turned on later.
2065 if (DELAY_ACK(tp, tlen) && tlen != 0)
2066 tcp_timer_activate(tp, TT_DELACK,
2069 tp->t_flags |= TF_ACKNOW;
2071 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
2072 tp->t_flags |= TF_ECN_PERMIT;
2073 TCPSTAT_INC(tcps_ecn_shs);
2077 * Received <SYN,ACK> in SYN_SENT[*] state.
2079 * SYN_SENT --> ESTABLISHED
2080 * SYN_SENT* --> FIN_WAIT_1
2082 tp->t_starttime = ticks;
2083 if (tp->t_flags & TF_NEEDFIN) {
2084 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2085 tp->t_flags &= ~TF_NEEDFIN;
2088 tcp_state_change(tp, TCPS_ESTABLISHED);
2089 TCP_PROBE5(connect__established, NULL, tp,
2090 mtod(m, const char *), tp, th);
2092 tcp_timer_activate(tp, TT_KEEP,
2097 * Received initial SYN in SYN-SENT[*] state =>
2098 * simultaneous open.
2099 * If it succeeds, connection is * half-synchronized.
2100 * Otherwise, do 3-way handshake:
2101 * SYN-SENT -> SYN-RECEIVED
2102 * SYN-SENT* -> SYN-RECEIVED*
2104 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2105 tcp_timer_activate(tp, TT_REXMT, 0);
2106 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2109 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: "
2110 "ti_locked %d", __func__, ti_locked));
2111 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2112 INP_WLOCK_ASSERT(tp->t_inpcb);
2115 * Advance th->th_seq to correspond to first data byte.
2116 * If data, trim to stay within window,
2117 * dropping FIN if necessary.
2120 if (tlen > tp->rcv_wnd) {
2121 todrop = tlen - tp->rcv_wnd;
2125 TCPSTAT_INC(tcps_rcvpackafterwin);
2126 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2128 tp->snd_wl1 = th->th_seq - 1;
2129 tp->rcv_up = th->th_seq;
2131 * Client side of transaction: already sent SYN and data.
2132 * If the remote host used T/TCP to validate the SYN,
2133 * our data will be ACK'd; if so, enter normal data segment
2134 * processing in the middle of step 5, ack processing.
2135 * Otherwise, goto step 6.
2137 if (thflags & TH_ACK)
2143 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2144 * do normal processing.
2146 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2150 break; /* continue normal processing */
2154 * States other than LISTEN or SYN_SENT.
2155 * First check the RST flag and sequence number since reset segments
2156 * are exempt from the timestamp and connection count tests. This
2157 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2158 * below which allowed reset segments in half the sequence space
2159 * to fall though and be processed (which gives forged reset
2160 * segments with a random sequence number a 50 percent chance of
2161 * killing a connection).
2162 * Then check timestamp, if present.
2163 * Then check the connection count, if present.
2164 * Then check that at least some bytes of segment are within
2165 * receive window. If segment begins before rcv_nxt,
2166 * drop leading data (and SYN); if nothing left, just ack.
2168 if (thflags & TH_RST) {
2170 * RFC5961 Section 3.2
2172 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2173 * - If RST is in window, we send challenge ACK.
2175 * Note: to take into account delayed ACKs, we should
2176 * test against last_ack_sent instead of rcv_nxt.
2177 * Note 2: we handle special case of closed window, not
2178 * covered by the RFC.
2180 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2181 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2182 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2184 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2185 KASSERT(ti_locked == TI_RLOCKED,
2186 ("%s: TH_RST ti_locked %d, th %p tp %p",
2187 __func__, ti_locked, th, tp));
2188 KASSERT(tp->t_state != TCPS_SYN_SENT,
2189 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2192 if (V_tcp_insecure_rst ||
2193 tp->last_ack_sent == th->th_seq) {
2194 TCPSTAT_INC(tcps_drops);
2195 /* Drop the connection. */
2196 switch (tp->t_state) {
2197 case TCPS_SYN_RECEIVED:
2198 so->so_error = ECONNREFUSED;
2200 case TCPS_ESTABLISHED:
2201 case TCPS_FIN_WAIT_1:
2202 case TCPS_FIN_WAIT_2:
2203 case TCPS_CLOSE_WAIT:
2204 so->so_error = ECONNRESET;
2206 tcp_state_change(tp, TCPS_CLOSED);
2212 TCPSTAT_INC(tcps_badrst);
2213 /* Send challenge ACK. */
2214 tcp_respond(tp, mtod(m, void *), th, m,
2215 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2216 tp->last_ack_sent = tp->rcv_nxt;
2224 * RFC5961 Section 4.2
2225 * Send challenge ACK for any SYN in synchronized state.
2227 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2228 tp->t_state != TCPS_SYN_RECEIVED) {
2229 KASSERT(ti_locked == TI_RLOCKED,
2230 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2231 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2233 TCPSTAT_INC(tcps_badsyn);
2234 if (V_tcp_insecure_syn &&
2235 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2236 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2237 tp = tcp_drop(tp, ECONNRESET);
2238 rstreason = BANDLIM_UNLIMITED;
2240 /* Send challenge ACK. */
2241 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2242 tp->snd_nxt, TH_ACK);
2243 tp->last_ack_sent = tp->rcv_nxt;
2250 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2251 * and it's less than ts_recent, drop it.
2253 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2254 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2256 /* Check to see if ts_recent is over 24 days old. */
2257 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2259 * Invalidate ts_recent. If this segment updates
2260 * ts_recent, the age will be reset later and ts_recent
2261 * will get a valid value. If it does not, setting
2262 * ts_recent to zero will at least satisfy the
2263 * requirement that zero be placed in the timestamp
2264 * echo reply when ts_recent isn't valid. The
2265 * age isn't reset until we get a valid ts_recent
2266 * because we don't want out-of-order segments to be
2267 * dropped when ts_recent is old.
2271 TCPSTAT_INC(tcps_rcvduppack);
2272 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2273 TCPSTAT_INC(tcps_pawsdrop);
2281 * In the SYN-RECEIVED state, validate that the packet belongs to
2282 * this connection before trimming the data to fit the receive
2283 * window. Check the sequence number versus IRS since we know
2284 * the sequence numbers haven't wrapped. This is a partial fix
2285 * for the "LAND" DoS attack.
2287 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2288 rstreason = BANDLIM_RST_OPENPORT;
2292 todrop = tp->rcv_nxt - th->th_seq;
2294 if (thflags & TH_SYN) {
2304 * Following if statement from Stevens, vol. 2, p. 960.
2307 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2309 * Any valid FIN must be to the left of the window.
2310 * At this point the FIN must be a duplicate or out
2311 * of sequence; drop it.
2316 * Send an ACK to resynchronize and drop any data.
2317 * But keep on processing for RST or ACK.
2319 tp->t_flags |= TF_ACKNOW;
2321 TCPSTAT_INC(tcps_rcvduppack);
2322 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2324 TCPSTAT_INC(tcps_rcvpartduppack);
2325 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2327 drop_hdrlen += todrop; /* drop from the top afterwards */
2328 th->th_seq += todrop;
2330 if (th->th_urp > todrop)
2331 th->th_urp -= todrop;
2339 * If new data are received on a connection after the
2340 * user processes are gone, then RST the other end.
2342 if ((so->so_state & SS_NOFDREF) &&
2343 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2344 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && "
2345 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2346 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2348 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2349 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2350 "after socket was closed, "
2351 "sending RST and removing tcpcb\n",
2352 s, __func__, tcpstates[tp->t_state], tlen);
2356 TCPSTAT_INC(tcps_rcvafterclose);
2357 rstreason = BANDLIM_UNLIMITED;
2362 * If segment ends after window, drop trailing data
2363 * (and PUSH and FIN); if nothing left, just ACK.
2365 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2367 TCPSTAT_INC(tcps_rcvpackafterwin);
2368 if (todrop >= tlen) {
2369 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2371 * If window is closed can only take segments at
2372 * window edge, and have to drop data and PUSH from
2373 * incoming segments. Continue processing, but
2374 * remember to ack. Otherwise, drop segment
2377 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2378 tp->t_flags |= TF_ACKNOW;
2379 TCPSTAT_INC(tcps_rcvwinprobe);
2383 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2386 thflags &= ~(TH_PUSH|TH_FIN);
2390 * If last ACK falls within this segment's sequence numbers,
2391 * record its timestamp.
2393 * 1) That the test incorporates suggestions from the latest
2394 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2395 * 2) That updating only on newer timestamps interferes with
2396 * our earlier PAWS tests, so this check should be solely
2397 * predicated on the sequence space of this segment.
2398 * 3) That we modify the segment boundary check to be
2399 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2400 * instead of RFC1323's
2401 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2402 * This modified check allows us to overcome RFC1323's
2403 * limitations as described in Stevens TCP/IP Illustrated
2404 * Vol. 2 p.869. In such cases, we can still calculate the
2405 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2407 if ((to.to_flags & TOF_TS) != 0 &&
2408 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2409 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2410 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2411 tp->ts_recent_age = tcp_ts_getticks();
2412 tp->ts_recent = to.to_tsval;
2416 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2417 * flag is on (half-synchronized state), then queue data for
2418 * later processing; else drop segment and return.
2420 if ((thflags & TH_ACK) == 0) {
2421 if (tp->t_state == TCPS_SYN_RECEIVED ||
2422 (tp->t_flags & TF_NEEDSYN)) {
2424 if (tp->t_state == TCPS_SYN_RECEIVED &&
2425 IS_FASTOPEN(tp->t_flags)) {
2426 tp->snd_wnd = tiwin;
2431 } else if (tp->t_flags & TF_ACKNOW)
2440 switch (tp->t_state) {
2443 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2444 * ESTABLISHED state and continue processing.
2445 * The ACK was checked above.
2447 case TCPS_SYN_RECEIVED:
2449 TCPSTAT_INC(tcps_connects);
2451 /* Do window scaling? */
2452 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2453 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2454 tp->rcv_scale = tp->request_r_scale;
2455 tp->snd_wnd = tiwin;
2459 * SYN-RECEIVED -> ESTABLISHED
2460 * SYN-RECEIVED* -> FIN-WAIT-1
2462 tp->t_starttime = ticks;
2463 if (tp->t_flags & TF_NEEDFIN) {
2464 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2465 tp->t_flags &= ~TF_NEEDFIN;
2467 tcp_state_change(tp, TCPS_ESTABLISHED);
2468 TCP_PROBE5(accept__established, NULL, tp,
2469 mtod(m, const char *), tp, th);
2471 if (tp->t_tfo_pending) {
2472 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2473 tp->t_tfo_pending = NULL;
2476 * Account for the ACK of our SYN prior to
2477 * regular ACK processing below.
2482 * TFO connections call cc_conn_init() during SYN
2483 * processing. Calling it again here for such
2484 * connections is not harmless as it would undo the
2485 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2488 if (!IS_FASTOPEN(tp->t_flags))
2491 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2494 * If segment contains data or ACK, will call tcp_reass()
2495 * later; if not, do so now to pass queued data to user.
2497 if (tlen == 0 && (thflags & TH_FIN) == 0)
2498 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2500 tp->snd_wl1 = th->th_seq - 1;
2504 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2505 * ACKs. If the ack is in the range
2506 * tp->snd_una < th->th_ack <= tp->snd_max
2507 * then advance tp->snd_una to th->th_ack and drop
2508 * data from the retransmission queue. If this ACK reflects
2509 * more up to date window information we update our window information.
2511 case TCPS_ESTABLISHED:
2512 case TCPS_FIN_WAIT_1:
2513 case TCPS_FIN_WAIT_2:
2514 case TCPS_CLOSE_WAIT:
2517 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2518 TCPSTAT_INC(tcps_rcvacktoomuch);
2521 if ((tp->t_flags & TF_SACK_PERMIT) &&
2522 ((to.to_flags & TOF_SACK) ||
2523 !TAILQ_EMPTY(&tp->snd_holes)))
2524 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2527 * Reset the value so that previous (valid) value
2528 * from the last ack with SACK doesn't get used.
2530 tp->sackhint.sacked_bytes = 0;
2533 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2534 hhook_run_tcp_est_in(tp, th, &to);
2537 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2540 maxseg = tcp_maxseg(tp);
2542 (tiwin == tp->snd_wnd ||
2543 (tp->t_flags & TF_SACK_PERMIT))) {
2545 * If this is the first time we've seen a
2546 * FIN from the remote, this is not a
2547 * duplicate and it needs to be processed
2548 * normally. This happens during a
2549 * simultaneous close.
2551 if ((thflags & TH_FIN) &&
2552 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2556 TCPSTAT_INC(tcps_rcvdupack);
2558 * If we have outstanding data (other than
2559 * a window probe), this is a completely
2560 * duplicate ack (ie, window info didn't
2561 * change and FIN isn't set),
2562 * the ack is the biggest we've
2563 * seen and we've seen exactly our rexmt
2564 * threshold of them, assume a packet
2565 * has been dropped and retransmit it.
2566 * Kludge snd_nxt & the congestion
2567 * window so we send only this one
2570 * We know we're losing at the current
2571 * window size so do congestion avoidance
2572 * (set ssthresh to half the current window
2573 * and pull our congestion window back to
2574 * the new ssthresh).
2576 * Dup acks mean that packets have left the
2577 * network (they're now cached at the receiver)
2578 * so bump cwnd by the amount in the receiver
2579 * to keep a constant cwnd packets in the
2582 * When using TCP ECN, notify the peer that
2583 * we reduced the cwnd.
2586 * Following 2 kinds of acks should not affect
2589 * 2) Acks with SACK but without any new SACK
2590 * information in them. These could result from
2591 * any anomaly in the network like a switch
2592 * duplicating packets or a possible DoS attack.
2594 if (th->th_ack != tp->snd_una ||
2595 ((tp->t_flags & TF_SACK_PERMIT) &&
2598 else if (!tcp_timer_active(tp, TT_REXMT))
2600 else if (++tp->t_dupacks > tcprexmtthresh ||
2601 IN_FASTRECOVERY(tp->t_flags)) {
2602 cc_ack_received(tp, th, nsegs,
2604 if ((tp->t_flags & TF_SACK_PERMIT) &&
2605 IN_FASTRECOVERY(tp->t_flags)) {
2609 * Compute the amount of data in flight first.
2610 * We can inject new data into the pipe iff
2611 * we have less than 1/2 the original window's
2612 * worth of data in flight.
2614 if (V_tcp_do_rfc6675_pipe)
2615 awnd = tcp_compute_pipe(tp);
2617 awnd = (tp->snd_nxt - tp->snd_fack) +
2618 tp->sackhint.sack_bytes_rexmit;
2620 if (awnd < tp->snd_ssthresh) {
2621 tp->snd_cwnd += maxseg;
2623 * RFC5681 Section 3.2 talks about cwnd
2624 * inflation on additional dupacks and
2625 * deflation on recovering from loss.
2627 * We keep cwnd into check so that
2628 * we don't have to 'deflate' it when we
2629 * get out of recovery.
2631 if (tp->snd_cwnd > tp->snd_ssthresh)
2632 tp->snd_cwnd = tp->snd_ssthresh;
2635 tp->snd_cwnd += maxseg;
2636 (void) tp->t_fb->tfb_tcp_output(tp);
2638 } else if (tp->t_dupacks == tcprexmtthresh) {
2639 tcp_seq onxt = tp->snd_nxt;
2642 * If we're doing sack, check to
2643 * see if we're already in sack
2644 * recovery. If we're not doing sack,
2645 * check to see if we're in newreno
2648 if (tp->t_flags & TF_SACK_PERMIT) {
2649 if (IN_FASTRECOVERY(tp->t_flags)) {
2654 if (SEQ_LEQ(th->th_ack,
2660 /* Congestion signal before ack. */
2661 cc_cong_signal(tp, th, CC_NDUPACK);
2662 cc_ack_received(tp, th, nsegs,
2664 tcp_timer_activate(tp, TT_REXMT, 0);
2666 if (tp->t_flags & TF_SACK_PERMIT) {
2668 tcps_sack_recovery_episode);
2669 tp->sack_newdata = tp->snd_nxt;
2670 if (CC_ALGO(tp)->cong_signal == NULL)
2671 tp->snd_cwnd = maxseg;
2672 (void) tp->t_fb->tfb_tcp_output(tp);
2675 tp->snd_nxt = th->th_ack;
2676 if (CC_ALGO(tp)->cong_signal == NULL)
2677 tp->snd_cwnd = maxseg;
2678 (void) tp->t_fb->tfb_tcp_output(tp);
2679 KASSERT(tp->snd_limited <= 2,
2680 ("%s: tp->snd_limited too big",
2682 if (CC_ALGO(tp)->cong_signal == NULL)
2683 tp->snd_cwnd = tp->snd_ssthresh +
2685 (tp->t_dupacks - tp->snd_limited);
2686 if (SEQ_GT(onxt, tp->snd_nxt))
2689 } else if (V_tcp_do_rfc3042) {
2691 * Process first and second duplicate
2692 * ACKs. Each indicates a segment
2693 * leaving the network, creating room
2694 * for more. Make sure we can send a
2695 * packet on reception of each duplicate
2696 * ACK by increasing snd_cwnd by one
2697 * segment. Restore the original
2698 * snd_cwnd after packet transmission.
2700 cc_ack_received(tp, th, nsegs,
2702 uint32_t oldcwnd = tp->snd_cwnd;
2703 tcp_seq oldsndmax = tp->snd_max;
2707 KASSERT(tp->t_dupacks == 1 ||
2709 ("%s: dupacks not 1 or 2",
2711 if (tp->t_dupacks == 1)
2712 tp->snd_limited = 0;
2714 (tp->snd_nxt - tp->snd_una) +
2715 (tp->t_dupacks - tp->snd_limited) *
2718 * Only call tcp_output when there
2719 * is new data available to be sent.
2720 * Otherwise we would send pure ACKs.
2722 SOCKBUF_LOCK(&so->so_snd);
2723 avail = sbavail(&so->so_snd) -
2724 (tp->snd_nxt - tp->snd_una);
2725 SOCKBUF_UNLOCK(&so->so_snd);
2727 (void) tp->t_fb->tfb_tcp_output(tp);
2728 sent = tp->snd_max - oldsndmax;
2729 if (sent > maxseg) {
2730 KASSERT((tp->t_dupacks == 2 &&
2731 tp->snd_limited == 0) ||
2732 (sent == maxseg + 1 &&
2733 tp->t_flags & TF_SENTFIN),
2734 ("%s: sent too much",
2736 tp->snd_limited = 2;
2737 } else if (sent > 0)
2739 tp->snd_cwnd = oldcwnd;
2746 * This ack is advancing the left edge, reset the
2751 * If this ack also has new SACK info, increment the
2752 * counter as per rfc6675.
2754 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed)
2758 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2759 ("%s: th_ack <= snd_una", __func__));
2762 * If the congestion window was inflated to account
2763 * for the other side's cached packets, retract it.
2765 if (IN_FASTRECOVERY(tp->t_flags)) {
2766 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2767 if (tp->t_flags & TF_SACK_PERMIT)
2768 tcp_sack_partialack(tp, th);
2770 tcp_newreno_partial_ack(tp, th);
2772 cc_post_recovery(tp, th);
2775 * If we reach this point, ACK is not a duplicate,
2776 * i.e., it ACKs something we sent.
2778 if (tp->t_flags & TF_NEEDSYN) {
2780 * T/TCP: Connection was half-synchronized, and our
2781 * SYN has been ACK'd (so connection is now fully
2782 * synchronized). Go to non-starred state,
2783 * increment snd_una for ACK of SYN, and check if
2784 * we can do window scaling.
2786 tp->t_flags &= ~TF_NEEDSYN;
2788 /* Do window scaling? */
2789 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2790 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2791 tp->rcv_scale = tp->request_r_scale;
2792 /* Send window already scaled. */
2797 INP_WLOCK_ASSERT(tp->t_inpcb);
2799 acked = BYTES_THIS_ACK(tp, th);
2800 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2801 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2802 tp->snd_una, th->th_ack, tp, m));
2803 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2804 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2807 * If we just performed our first retransmit, and the ACK
2808 * arrives within our recovery window, then it was a mistake
2809 * to do the retransmit in the first place. Recover our
2810 * original cwnd and ssthresh, and proceed to transmit where
2813 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2814 (int)(ticks - tp->t_badrxtwin) < 0)
2815 cc_cong_signal(tp, th, CC_RTO_ERR);
2818 * If we have a timestamp reply, update smoothed
2819 * round trip time. If no timestamp is present but
2820 * transmit timer is running and timed sequence
2821 * number was acked, update smoothed round trip time.
2822 * Since we now have an rtt measurement, cancel the
2823 * timer backoff (cf., Phil Karn's retransmit alg.).
2824 * Recompute the initial retransmit timer.
2826 * Some boxes send broken timestamp replies
2827 * during the SYN+ACK phase, ignore
2828 * timestamps of 0 or we could calculate a
2829 * huge RTT and blow up the retransmit timer.
2831 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2834 t = tcp_ts_getticks() - to.to_tsecr;
2835 if (!tp->t_rttlow || tp->t_rttlow > t)
2837 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2838 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2839 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2840 tp->t_rttlow = ticks - tp->t_rtttime;
2841 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2845 * If all outstanding data is acked, stop retransmit
2846 * timer and remember to restart (more output or persist).
2847 * If there is more data to be acked, restart retransmit
2848 * timer, using current (possibly backed-off) value.
2850 if (th->th_ack == tp->snd_max) {
2851 tcp_timer_activate(tp, TT_REXMT, 0);
2853 } else if (!tcp_timer_active(tp, TT_PERSIST))
2854 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2857 * If no data (only SYN) was ACK'd,
2858 * skip rest of ACK processing.
2864 * Let the congestion control algorithm update congestion
2865 * control related information. This typically means increasing
2866 * the congestion window.
2868 cc_ack_received(tp, th, nsegs, CC_ACK);
2870 SOCKBUF_LOCK(&so->so_snd);
2871 if (acked > sbavail(&so->so_snd)) {
2872 if (tp->snd_wnd >= sbavail(&so->so_snd))
2873 tp->snd_wnd -= sbavail(&so->so_snd);
2876 mfree = sbcut_locked(&so->so_snd,
2877 (int)sbavail(&so->so_snd));
2880 mfree = sbcut_locked(&so->so_snd, acked);
2881 if (tp->snd_wnd >= (uint32_t) acked)
2882 tp->snd_wnd -= acked;
2887 /* NB: sowwakeup_locked() does an implicit unlock. */
2888 sowwakeup_locked(so);
2890 /* Detect una wraparound. */
2891 if (!IN_RECOVERY(tp->t_flags) &&
2892 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2893 SEQ_LEQ(th->th_ack, tp->snd_recover))
2894 tp->snd_recover = th->th_ack - 1;
2895 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2896 if (IN_RECOVERY(tp->t_flags) &&
2897 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2898 EXIT_RECOVERY(tp->t_flags);
2900 tp->snd_una = th->th_ack;
2901 if (tp->t_flags & TF_SACK_PERMIT) {
2902 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2903 tp->snd_recover = tp->snd_una;
2905 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2906 tp->snd_nxt = tp->snd_una;
2908 switch (tp->t_state) {
2911 * In FIN_WAIT_1 STATE in addition to the processing
2912 * for the ESTABLISHED state if our FIN is now acknowledged
2913 * then enter FIN_WAIT_2.
2915 case TCPS_FIN_WAIT_1:
2916 if (ourfinisacked) {
2918 * If we can't receive any more
2919 * data, then closing user can proceed.
2920 * Starting the timer is contrary to the
2921 * specification, but if we don't get a FIN
2922 * we'll hang forever.
2925 * we should release the tp also, and use a
2928 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2929 soisdisconnected(so);
2930 tcp_timer_activate(tp, TT_2MSL,
2931 (tcp_fast_finwait2_recycle ?
2932 tcp_finwait2_timeout :
2935 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2940 * In CLOSING STATE in addition to the processing for
2941 * the ESTABLISHED state if the ACK acknowledges our FIN
2942 * then enter the TIME-WAIT state, otherwise ignore
2946 if (ourfinisacked) {
2947 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2949 INP_INFO_RUNLOCK(&V_tcbinfo);
2956 * In LAST_ACK, we may still be waiting for data to drain
2957 * and/or to be acked, as well as for the ack of our FIN.
2958 * If our FIN is now acknowledged, delete the TCB,
2959 * enter the closed state and return.
2962 if (ourfinisacked) {
2963 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2972 INP_WLOCK_ASSERT(tp->t_inpcb);
2975 * Update window information.
2976 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2978 if ((thflags & TH_ACK) &&
2979 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2980 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2981 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2982 /* keep track of pure window updates */
2984 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2985 TCPSTAT_INC(tcps_rcvwinupd);
2986 tp->snd_wnd = tiwin;
2987 tp->snd_wl1 = th->th_seq;
2988 tp->snd_wl2 = th->th_ack;
2989 if (tp->snd_wnd > tp->max_sndwnd)
2990 tp->max_sndwnd = tp->snd_wnd;
2995 * Process segments with URG.
2997 if ((thflags & TH_URG) && th->th_urp &&
2998 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3000 * This is a kludge, but if we receive and accept
3001 * random urgent pointers, we'll crash in
3002 * soreceive. It's hard to imagine someone
3003 * actually wanting to send this much urgent data.
3005 SOCKBUF_LOCK(&so->so_rcv);
3006 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3007 th->th_urp = 0; /* XXX */
3008 thflags &= ~TH_URG; /* XXX */
3009 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3010 goto dodata; /* XXX */
3013 * If this segment advances the known urgent pointer,
3014 * then mark the data stream. This should not happen
3015 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3016 * a FIN has been received from the remote side.
3017 * In these states we ignore the URG.
3019 * According to RFC961 (Assigned Protocols),
3020 * the urgent pointer points to the last octet
3021 * of urgent data. We continue, however,
3022 * to consider it to indicate the first octet
3023 * of data past the urgent section as the original
3024 * spec states (in one of two places).
3026 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3027 tp->rcv_up = th->th_seq + th->th_urp;
3028 so->so_oobmark = sbavail(&so->so_rcv) +
3029 (tp->rcv_up - tp->rcv_nxt) - 1;
3030 if (so->so_oobmark == 0)
3031 so->so_rcv.sb_state |= SBS_RCVATMARK;
3033 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3035 SOCKBUF_UNLOCK(&so->so_rcv);
3037 * Remove out of band data so doesn't get presented to user.
3038 * This can happen independent of advancing the URG pointer,
3039 * but if two URG's are pending at once, some out-of-band
3040 * data may creep in... ick.
3042 if (th->th_urp <= (uint32_t)tlen &&
3043 !(so->so_options & SO_OOBINLINE)) {
3044 /* hdr drop is delayed */
3045 tcp_pulloutofband(so, th, m, drop_hdrlen);
3049 * If no out of band data is expected,
3050 * pull receive urgent pointer along
3051 * with the receive window.
3053 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3054 tp->rcv_up = tp->rcv_nxt;
3057 INP_WLOCK_ASSERT(tp->t_inpcb);
3060 * Process the segment text, merging it into the TCP sequencing queue,
3061 * and arranging for acknowledgment of receipt if necessary.
3062 * This process logically involves adjusting tp->rcv_wnd as data
3063 * is presented to the user (this happens in tcp_usrreq.c,
3064 * case PRU_RCVD). If a FIN has already been received on this
3065 * connection then we just ignore the text.
3068 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3069 IS_FASTOPEN(tp->t_flags));
3071 #define tfo_syn (false)
3073 if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
3074 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3075 tcp_seq save_start = th->th_seq;
3076 m_adj(m, drop_hdrlen); /* delayed header drop */
3078 * Insert segment which includes th into TCP reassembly queue
3079 * with control block tp. Set thflags to whether reassembly now
3080 * includes a segment with FIN. This handles the common case
3081 * inline (segment is the next to be received on an established
3082 * connection, and the queue is empty), avoiding linkage into
3083 * and removal from the queue and repetition of various
3085 * Set DELACK for segments received in order, but ack
3086 * immediately when segments are out of order (so
3087 * fast retransmit can work).
3089 if (th->th_seq == tp->rcv_nxt &&
3090 LIST_EMPTY(&tp->t_segq) &&
3091 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3093 if (DELAY_ACK(tp, tlen) || tfo_syn)
3094 tp->t_flags |= TF_DELACK;
3096 tp->t_flags |= TF_ACKNOW;
3097 tp->rcv_nxt += tlen;
3098 thflags = th->th_flags & TH_FIN;
3099 TCPSTAT_INC(tcps_rcvpack);
3100 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3101 SOCKBUF_LOCK(&so->so_rcv);
3102 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3105 sbappendstream_locked(&so->so_rcv, m, 0);
3106 /* NB: sorwakeup_locked() does an implicit unlock. */
3107 sorwakeup_locked(so);
3110 * XXX: Due to the header drop above "th" is
3111 * theoretically invalid by now. Fortunately
3112 * m_adj() doesn't actually frees any mbufs
3113 * when trimming from the head.
3115 thflags = tcp_reass(tp, th, &tlen, m);
3116 tp->t_flags |= TF_ACKNOW;
3118 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
3119 tcp_update_sack_list(tp, save_start, save_start + tlen);
3122 * Note the amount of data that peer has sent into
3123 * our window, in order to estimate the sender's
3127 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3128 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3130 len = so->so_rcv.sb_hiwat;
3138 * If FIN is received ACK the FIN and let the user know
3139 * that the connection is closing.
3141 if (thflags & TH_FIN) {
3142 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3145 * If connection is half-synchronized
3146 * (ie NEEDSYN flag on) then delay ACK,
3147 * so it may be piggybacked when SYN is sent.
3148 * Otherwise, since we received a FIN then no
3149 * more input can be expected, send ACK now.
3151 if (tp->t_flags & TF_NEEDSYN)
3152 tp->t_flags |= TF_DELACK;
3154 tp->t_flags |= TF_ACKNOW;
3157 switch (tp->t_state) {
3160 * In SYN_RECEIVED and ESTABLISHED STATES
3161 * enter the CLOSE_WAIT state.
3163 case TCPS_SYN_RECEIVED:
3164 tp->t_starttime = ticks;
3166 case TCPS_ESTABLISHED:
3167 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3171 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3172 * enter the CLOSING state.
3174 case TCPS_FIN_WAIT_1:
3175 tcp_state_change(tp, TCPS_CLOSING);
3179 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3180 * starting the time-wait timer, turning off the other
3183 case TCPS_FIN_WAIT_2:
3184 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
3185 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata "
3186 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
3190 INP_INFO_RUNLOCK(&V_tcbinfo);
3194 if (ti_locked == TI_RLOCKED)
3195 INP_INFO_RUNLOCK(&V_tcbinfo);
3196 ti_locked = TI_UNLOCKED;
3199 if (so->so_options & SO_DEBUG)
3200 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3203 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
3206 * Return any desired output.
3208 if (needoutput || (tp->t_flags & TF_ACKNOW))
3209 (void) tp->t_fb->tfb_tcp_output(tp);
3212 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3213 __func__, ti_locked));
3214 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3215 INP_WLOCK_ASSERT(tp->t_inpcb);
3217 if (tp->t_flags & TF_DELACK) {
3218 tp->t_flags &= ~TF_DELACK;
3219 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3221 INP_WUNLOCK(tp->t_inpcb);
3226 * Generate an ACK dropping incoming segment if it occupies
3227 * sequence space, where the ACK reflects our state.
3229 * We can now skip the test for the RST flag since all
3230 * paths to this code happen after packets containing
3231 * RST have been dropped.
3233 * In the SYN-RECEIVED state, don't send an ACK unless the
3234 * segment we received passes the SYN-RECEIVED ACK test.
3235 * If it fails send a RST. This breaks the loop in the
3236 * "LAND" DoS attack, and also prevents an ACK storm
3237 * between two listening ports that have been sent forged
3238 * SYN segments, each with the source address of the other.
3240 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3241 (SEQ_GT(tp->snd_una, th->th_ack) ||
3242 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3243 rstreason = BANDLIM_RST_OPENPORT;
3247 if (so->so_options & SO_DEBUG)
3248 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3251 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
3252 if (ti_locked == TI_RLOCKED)
3253 INP_INFO_RUNLOCK(&V_tcbinfo);
3254 ti_locked = TI_UNLOCKED;
3256 tp->t_flags |= TF_ACKNOW;
3257 (void) tp->t_fb->tfb_tcp_output(tp);
3258 INP_WUNLOCK(tp->t_inpcb);
3263 if (ti_locked == TI_RLOCKED)
3264 INP_INFO_RUNLOCK(&V_tcbinfo);
3265 ti_locked = TI_UNLOCKED;
3268 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3269 INP_WUNLOCK(tp->t_inpcb);
3271 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3275 if (ti_locked == TI_RLOCKED) {
3276 INP_INFO_RUNLOCK(&V_tcbinfo);
3277 ti_locked = TI_UNLOCKED;
3281 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3285 * Drop space held by incoming segment and return.
3288 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3289 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3292 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
3294 INP_WUNLOCK(tp->t_inpcb);
3302 * Issue RST and make ACK acceptable to originator of segment.
3303 * The mbuf must still include the original packet header.
3307 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3308 int tlen, int rstreason)
3314 struct ip6_hdr *ip6;
3318 INP_WLOCK_ASSERT(tp->t_inpcb);
3321 /* Don't bother if destination was broadcast/multicast. */
3322 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3325 if (mtod(m, struct ip *)->ip_v == 6) {
3326 ip6 = mtod(m, struct ip6_hdr *);
3327 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3328 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3330 /* IPv6 anycast check is done at tcp6_input() */
3333 #if defined(INET) && defined(INET6)
3338 ip = mtod(m, struct ip *);
3339 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3340 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3341 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3342 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3347 /* Perform bandwidth limiting. */
3348 if (badport_bandlim(rstreason) < 0)
3351 /* tcp_respond consumes the mbuf chain. */
3352 if (th->th_flags & TH_ACK) {
3353 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3354 th->th_ack, TH_RST);
3356 if (th->th_flags & TH_SYN)
3358 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3359 (tcp_seq)0, TH_RST|TH_ACK);
3367 * Parse TCP options and place in tcpopt.
3370 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3375 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3377 if (opt == TCPOPT_EOL)
3379 if (opt == TCPOPT_NOP)
3385 if (optlen < 2 || optlen > cnt)
3390 if (optlen != TCPOLEN_MAXSEG)
3392 if (!(flags & TO_SYN))
3394 to->to_flags |= TOF_MSS;
3395 bcopy((char *)cp + 2,
3396 (char *)&to->to_mss, sizeof(to->to_mss));
3397 to->to_mss = ntohs(to->to_mss);
3400 if (optlen != TCPOLEN_WINDOW)
3402 if (!(flags & TO_SYN))
3404 to->to_flags |= TOF_SCALE;
3405 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3407 case TCPOPT_TIMESTAMP:
3408 if (optlen != TCPOLEN_TIMESTAMP)
3410 to->to_flags |= TOF_TS;
3411 bcopy((char *)cp + 2,
3412 (char *)&to->to_tsval, sizeof(to->to_tsval));
3413 to->to_tsval = ntohl(to->to_tsval);
3414 bcopy((char *)cp + 6,
3415 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3416 to->to_tsecr = ntohl(to->to_tsecr);
3418 #ifdef TCP_SIGNATURE
3420 * XXX In order to reply to a host which has set the
3421 * TCP_SIGNATURE option in its initial SYN, we have to
3422 * record the fact that the option was observed here
3423 * for the syncache code to perform the correct response.
3425 case TCPOPT_SIGNATURE:
3426 if (optlen != TCPOLEN_SIGNATURE)
3428 to->to_flags |= TOF_SIGNATURE;
3429 to->to_signature = cp + 2;
3432 case TCPOPT_SACK_PERMITTED:
3433 if (optlen != TCPOLEN_SACK_PERMITTED)
3435 if (!(flags & TO_SYN))
3439 to->to_flags |= TOF_SACKPERM;
3442 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3446 to->to_flags |= TOF_SACK;
3447 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3448 to->to_sacks = cp + 2;
3449 TCPSTAT_INC(tcps_sack_rcv_blocks);
3452 case TCPOPT_FAST_OPEN:
3453 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) &&
3454 (optlen < TCPOLEN_FAST_OPEN_MIN) &&
3455 (optlen > TCPOLEN_FAST_OPEN_MAX))
3457 if (!(flags & TO_SYN))
3459 if (!V_tcp_fastopen_enabled)
3461 to->to_flags |= TOF_FASTOPEN;
3462 to->to_tfo_len = optlen - 2;
3463 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3473 * Pull out of band byte out of a segment so
3474 * it doesn't appear in the user's data queue.
3475 * It is still reflected in the segment length for
3476 * sequencing purposes.
3479 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3482 int cnt = off + th->th_urp - 1;
3485 if (m->m_len > cnt) {
3486 char *cp = mtod(m, caddr_t) + cnt;
3487 struct tcpcb *tp = sototcpcb(so);
3489 INP_WLOCK_ASSERT(tp->t_inpcb);
3492 tp->t_oobflags |= TCPOOB_HAVEDATA;
3493 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3495 if (m->m_flags & M_PKTHDR)
3504 panic("tcp_pulloutofband");
3508 * Collect new round-trip time estimate
3509 * and update averages and current timeout.
3512 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3516 INP_WLOCK_ASSERT(tp->t_inpcb);
3518 TCPSTAT_INC(tcps_rttupdated);
3520 if (tp->t_srtt != 0) {
3522 * srtt is stored as fixed point with 5 bits after the
3523 * binary point (i.e., scaled by 8). The following magic
3524 * is equivalent to the smoothing algorithm in rfc793 with
3525 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3526 * point). Adjust rtt to origin 0.
3528 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3529 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3531 if ((tp->t_srtt += delta) <= 0)
3535 * We accumulate a smoothed rtt variance (actually, a
3536 * smoothed mean difference), then set the retransmit
3537 * timer to smoothed rtt + 4 times the smoothed variance.
3538 * rttvar is stored as fixed point with 4 bits after the
3539 * binary point (scaled by 16). The following is
3540 * equivalent to rfc793 smoothing with an alpha of .75
3541 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3542 * rfc793's wired-in beta.
3546 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3547 if ((tp->t_rttvar += delta) <= 0)
3549 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3550 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3553 * No rtt measurement yet - use the unsmoothed rtt.
3554 * Set the variance to half the rtt (so our first
3555 * retransmit happens at 3*rtt).
3557 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3558 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3559 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3565 * the retransmit should happen at rtt + 4 * rttvar.
3566 * Because of the way we do the smoothing, srtt and rttvar
3567 * will each average +1/2 tick of bias. When we compute
3568 * the retransmit timer, we want 1/2 tick of rounding and
3569 * 1 extra tick because of +-1/2 tick uncertainty in the
3570 * firing of the timer. The bias will give us exactly the
3571 * 1.5 tick we need. But, because the bias is
3572 * statistical, we have to test that we don't drop below
3573 * the minimum feasible timer (which is 2 ticks).
3575 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3576 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3579 * We received an ack for a packet that wasn't retransmitted;
3580 * it is probably safe to discard any error indications we've
3581 * received recently. This isn't quite right, but close enough
3582 * for now (a route might have failed after we sent a segment,
3583 * and the return path might not be symmetrical).
3585 tp->t_softerror = 0;
3589 * Determine a reasonable value for maxseg size.
3590 * If the route is known, check route for mtu.
3591 * If none, use an mss that can be handled on the outgoing interface
3592 * without forcing IP to fragment. If no route is found, route has no mtu,
3593 * or the destination isn't local, use a default, hopefully conservative
3594 * size (usually 512 or the default IP max size, but no more than the mtu
3595 * of the interface), as we can't discover anything about intervening
3596 * gateways or networks. We also initialize the congestion/slow start
3597 * window to be a single segment if the destination isn't local.
3598 * While looking at the routing entry, we also initialize other path-dependent
3599 * parameters from pre-set or cached values in the routing entry.
3601 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3602 * IP options, e.g. IPSEC data, since length of this data may vary, and
3603 * thus it is calculated for every segment separately in tcp_output().
3605 * NOTE that this routine is only called when we process an incoming
3606 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3607 * settings are handled in tcp_mssopt().
3610 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3611 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3614 uint32_t maxmtu = 0;
3615 struct inpcb *inp = tp->t_inpcb;
3616 struct hc_metrics_lite metrics;
3618 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3619 size_t min_protoh = isipv6 ?
3620 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3621 sizeof (struct tcpiphdr);
3623 const size_t min_protoh = sizeof(struct tcpiphdr);
3626 INP_WLOCK_ASSERT(tp->t_inpcb);
3628 if (mtuoffer != -1) {
3629 KASSERT(offer == -1, ("%s: conflict", __func__));
3630 offer = mtuoffer - min_protoh;
3636 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3637 tp->t_maxseg = V_tcp_v6mssdflt;
3640 #if defined(INET) && defined(INET6)
3645 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3646 tp->t_maxseg = V_tcp_mssdflt;
3651 * No route to sender, stay with default mss and return.
3655 * In case we return early we need to initialize metrics
3656 * to a defined state as tcp_hc_get() would do for us
3657 * if there was no cache hit.
3659 if (metricptr != NULL)
3660 bzero(metricptr, sizeof(struct hc_metrics_lite));
3664 /* What have we got? */
3668 * Offer == 0 means that there was no MSS on the SYN
3669 * segment, in this case we use tcp_mssdflt as
3670 * already assigned to t_maxseg above.
3672 offer = tp->t_maxseg;
3677 * Offer == -1 means that we didn't receive SYN yet.
3683 * Prevent DoS attack with too small MSS. Round up
3684 * to at least minmss.
3686 offer = max(offer, V_tcp_minmss);
3690 * rmx information is now retrieved from tcp_hostcache.
3692 tcp_hc_get(&inp->inp_inc, &metrics);
3693 if (metricptr != NULL)
3694 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3697 * If there's a discovered mtu in tcp hostcache, use it.
3698 * Else, use the link mtu.
3700 if (metrics.rmx_mtu)
3701 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3705 mss = maxmtu - min_protoh;
3706 if (!V_path_mtu_discovery &&
3707 !in6_localaddr(&inp->in6p_faddr))
3708 mss = min(mss, V_tcp_v6mssdflt);
3711 #if defined(INET) && defined(INET6)
3716 mss = maxmtu - min_protoh;
3717 if (!V_path_mtu_discovery &&
3718 !in_localaddr(inp->inp_faddr))
3719 mss = min(mss, V_tcp_mssdflt);
3723 * XXX - The above conditional (mss = maxmtu - min_protoh)
3724 * probably violates the TCP spec.
3725 * The problem is that, since we don't know the
3726 * other end's MSS, we are supposed to use a conservative
3727 * default. But, if we do that, then MTU discovery will
3728 * never actually take place, because the conservative
3729 * default is much less than the MTUs typically seen
3730 * on the Internet today. For the moment, we'll sweep
3731 * this under the carpet.
3733 * The conservative default might not actually be a problem
3734 * if the only case this occurs is when sending an initial
3735 * SYN with options and data to a host we've never talked
3736 * to before. Then, they will reply with an MSS value which
3737 * will get recorded and the new parameters should get
3738 * recomputed. For Further Study.
3741 mss = min(mss, offer);
3744 * Sanity check: make sure that maxseg will be large
3745 * enough to allow some data on segments even if the
3746 * all the option space is used (40bytes). Otherwise
3747 * funny things may happen in tcp_output.
3749 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3757 tcp_mss(struct tcpcb *tp, int offer)
3763 struct hc_metrics_lite metrics;
3764 struct tcp_ifcap cap;
3766 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3768 bzero(&cap, sizeof(cap));
3769 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3775 * If there's a pipesize, change the socket buffer to that size,
3776 * don't change if sb_hiwat is different than default (then it
3777 * has been changed on purpose with setsockopt).
3778 * Make the socket buffers an integral number of mss units;
3779 * if the mss is larger than the socket buffer, decrease the mss.
3781 so = inp->inp_socket;
3782 SOCKBUF_LOCK(&so->so_snd);
3783 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3784 bufsize = metrics.rmx_sendpipe;
3786 bufsize = so->so_snd.sb_hiwat;
3790 bufsize = roundup(bufsize, mss);
3791 if (bufsize > sb_max)
3793 if (bufsize > so->so_snd.sb_hiwat)
3794 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3796 SOCKBUF_UNLOCK(&so->so_snd);
3798 * Sanity check: make sure that maxseg will be large
3799 * enough to allow some data on segments even if the
3800 * all the option space is used (40bytes). Otherwise
3801 * funny things may happen in tcp_output.
3803 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3805 tp->t_maxseg = max(mss, 64);
3807 SOCKBUF_LOCK(&so->so_rcv);
3808 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3809 bufsize = metrics.rmx_recvpipe;
3811 bufsize = so->so_rcv.sb_hiwat;
3812 if (bufsize > mss) {
3813 bufsize = roundup(bufsize, mss);
3814 if (bufsize > sb_max)
3816 if (bufsize > so->so_rcv.sb_hiwat)
3817 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3819 SOCKBUF_UNLOCK(&so->so_rcv);
3821 /* Check the interface for TSO capabilities. */
3822 if (cap.ifcap & CSUM_TSO) {
3823 tp->t_flags |= TF_TSO;
3824 tp->t_tsomax = cap.tsomax;
3825 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3826 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3831 * Determine the MSS option to send on an outgoing SYN.
3834 tcp_mssopt(struct in_conninfo *inc)
3837 uint32_t thcmtu = 0;
3838 uint32_t maxmtu = 0;
3841 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3844 if (inc->inc_flags & INC_ISIPV6) {
3845 mss = V_tcp_v6mssdflt;
3846 maxmtu = tcp_maxmtu6(inc, NULL);
3847 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3850 #if defined(INET) && defined(INET6)
3855 mss = V_tcp_mssdflt;
3856 maxmtu = tcp_maxmtu(inc, NULL);
3857 min_protoh = sizeof(struct tcpiphdr);
3860 #if defined(INET6) || defined(INET)
3861 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3864 if (maxmtu && thcmtu)
3865 mss = min(maxmtu, thcmtu) - min_protoh;
3866 else if (maxmtu || thcmtu)
3867 mss = max(maxmtu, thcmtu) - min_protoh;
3874 * On a partial ack arrives, force the retransmission of the
3875 * next unacknowledged segment. Do not clear tp->t_dupacks.
3876 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3880 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3882 tcp_seq onxt = tp->snd_nxt;
3883 uint32_t ocwnd = tp->snd_cwnd;
3884 u_int maxseg = tcp_maxseg(tp);
3886 INP_WLOCK_ASSERT(tp->t_inpcb);
3888 tcp_timer_activate(tp, TT_REXMT, 0);
3890 tp->snd_nxt = th->th_ack;
3892 * Set snd_cwnd to one segment beyond acknowledged offset.
3893 * (tp->snd_una has not yet been updated when this function is called.)
3895 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3896 tp->t_flags |= TF_ACKNOW;
3897 (void) tp->t_fb->tfb_tcp_output(tp);
3898 tp->snd_cwnd = ocwnd;
3899 if (SEQ_GT(onxt, tp->snd_nxt))
3902 * Partial window deflation. Relies on fact that tp->snd_una
3905 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3906 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3909 tp->snd_cwnd += maxseg;
3913 tcp_compute_pipe(struct tcpcb *tp)
3915 return (tp->snd_max - tp->snd_una +
3916 tp->sackhint.sack_bytes_rexmit -
3917 tp->sackhint.sacked_bytes);