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("$FreeBSD$");
53 #include "opt_ipfw.h" /* for ipfw_fwd */
55 #include "opt_inet6.h"
56 #include "opt_ipsec.h"
57 #include "opt_tcpdebug.h"
59 #include <sys/param.h>
60 #include <sys/kernel.h>
61 #include <sys/hhook.h>
62 #include <sys/malloc.h>
64 #include <sys/proc.h> /* for proc0 declaration */
65 #include <sys/protosw.h>
66 #include <sys/signalvar.h>
67 #include <sys/socket.h>
68 #include <sys/socketvar.h>
69 #include <sys/sysctl.h>
70 #include <sys/syslog.h>
71 #include <sys/systm.h>
73 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
78 #include <net/route.h>
81 #define TCPSTATES /* for logging */
83 #include <netinet/cc.h>
84 #include <netinet/in.h>
85 #include <netinet/in_pcb.h>
86 #include <netinet/in_systm.h>
87 #include <netinet/in_var.h>
88 #include <netinet/ip.h>
89 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
90 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
91 #include <netinet/ip_var.h>
92 #include <netinet/ip_options.h>
93 #include <netinet/ip6.h>
94 #include <netinet/icmp6.h>
95 #include <netinet6/in6_pcb.h>
96 #include <netinet6/ip6_var.h>
97 #include <netinet6/nd6.h>
98 #include <netinet/tcp_fsm.h>
99 #include <netinet/tcp_seq.h>
100 #include <netinet/tcp_timer.h>
101 #include <netinet/tcp_var.h>
102 #include <netinet6/tcp6_var.h>
103 #include <netinet/tcpip.h>
104 #include <netinet/tcp_syncache.h>
106 #include <netinet/tcp_debug.h>
107 #endif /* TCPDEBUG */
109 #include <netinet/tcp_offload.h>
113 #include <netipsec/ipsec.h>
114 #include <netipsec/ipsec6.h>
117 #include <machine/in_cksum.h>
119 #include <security/mac/mac_framework.h>
121 const int tcprexmtthresh = 3;
123 VNET_DEFINE(struct tcpstat, tcpstat);
124 SYSCTL_VNET_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
125 &VNET_NAME(tcpstat), tcpstat,
126 "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
128 int tcp_log_in_vain = 0;
129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
131 "Log all incoming TCP segments to closed ports");
133 VNET_DEFINE(int, blackhole) = 0;
134 #define V_blackhole VNET(blackhole)
135 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
136 &VNET_NAME(blackhole), 0,
137 "Do not send RST on segments to closed ports");
139 VNET_DEFINE(int, tcp_delack_enabled) = 1;
140 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
141 &VNET_NAME(tcp_delack_enabled), 0,
142 "Delay ACK to try and piggyback it onto a data packet");
144 VNET_DEFINE(int, drop_synfin) = 0;
145 #define V_drop_synfin VNET(drop_synfin)
146 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
147 &VNET_NAME(drop_synfin), 0,
148 "Drop TCP packets with SYN+FIN set");
150 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
151 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
152 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
153 &VNET_NAME(tcp_do_rfc3042), 0,
154 "Enable RFC 3042 (Limited Transmit)");
156 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
157 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
158 &VNET_NAME(tcp_do_rfc3390), 0,
159 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
161 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0,
162 "Experimental TCP extensions");
164 VNET_DEFINE(int, tcp_do_initcwnd10) = 0;
165 SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW,
166 &VNET_NAME(tcp_do_initcwnd10), 0,
167 "Enable draft-ietf-tcpm-initcwnd-05 (Increasing initial CWND to 10)");
169 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
170 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW,
171 &VNET_NAME(tcp_do_rfc3465), 0,
172 "Enable RFC 3465 (Appropriate Byte Counting)");
174 VNET_DEFINE(int, tcp_abc_l_var) = 2;
175 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW,
176 &VNET_NAME(tcp_abc_l_var), 2,
177 "Cap the max cwnd increment during slow-start to this number of segments");
179 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
181 VNET_DEFINE(int, tcp_do_ecn) = 0;
182 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW,
183 &VNET_NAME(tcp_do_ecn), 0,
186 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
187 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW,
188 &VNET_NAME(tcp_ecn_maxretries), 0,
189 "Max retries before giving up on ECN");
191 VNET_DEFINE(int, tcp_insecure_rst) = 0;
192 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
193 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
194 &VNET_NAME(tcp_insecure_rst), 0,
195 "Follow the old (insecure) criteria for accepting RST packets");
197 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
198 #define V_tcp_recvspace VNET(tcp_recvspace)
199 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
200 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
202 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
203 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
204 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
205 &VNET_NAME(tcp_do_autorcvbuf), 0,
206 "Enable automatic receive buffer sizing");
208 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
209 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
210 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
211 &VNET_NAME(tcp_autorcvbuf_inc), 0,
212 "Incrementor step size of automatic receive buffer");
214 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
215 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
216 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
217 &VNET_NAME(tcp_autorcvbuf_max), 0,
218 "Max size of automatic receive buffer");
220 VNET_DEFINE(struct inpcbhead, tcb);
221 #define tcb6 tcb /* for KAME src sync over BSD*'s */
222 VNET_DEFINE(struct inpcbinfo, tcbinfo);
224 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
225 static void tcp_do_segment(struct mbuf *, struct tcphdr *,
226 struct socket *, struct tcpcb *, int, int, uint8_t,
228 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
229 struct tcpcb *, int, int);
230 static void tcp_pulloutofband(struct socket *,
231 struct tcphdr *, struct mbuf *, int);
232 static void tcp_xmit_timer(struct tcpcb *, int);
233 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
234 static void inline tcp_fields_to_host(struct tcphdr *);
236 static void inline tcp_fields_to_net(struct tcphdr *);
237 static int inline tcp_signature_verify_input(struct mbuf *, int, int,
238 int, struct tcpopt *, struct tcphdr *, u_int);
240 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th,
242 static void inline cc_conn_init(struct tcpcb *tp);
243 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th);
244 static void inline hhook_run_tcp_est_in(struct tcpcb *tp,
245 struct tcphdr *th, struct tcpopt *to);
248 * Kernel module interface for updating tcpstat. The argument is an index
249 * into tcpstat treated as an array of u_long. While this encodes the
250 * general layout of tcpstat into the caller, it doesn't encode its location,
251 * so that future changes to add, for example, per-CPU stats support won't
252 * cause binary compatibility problems for kernel modules.
255 kmod_tcpstat_inc(int statnum)
258 (*((u_long *)&V_tcpstat + statnum))++;
262 * Wrapper for the TCP established input helper hook.
265 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
267 struct tcp_hhook_data hhook_data;
269 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
274 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
280 * CC wrapper hook functions
283 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type)
285 INP_WLOCK_ASSERT(tp->t_inpcb);
287 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
288 if (tp->snd_cwnd <= tp->snd_wnd)
289 tp->ccv->flags |= CCF_CWND_LIMITED;
291 tp->ccv->flags &= ~CCF_CWND_LIMITED;
293 if (type == CC_ACK) {
294 if (tp->snd_cwnd > tp->snd_ssthresh) {
295 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
296 V_tcp_abc_l_var * tp->t_maxseg);
297 if (tp->t_bytes_acked >= tp->snd_cwnd) {
298 tp->t_bytes_acked -= tp->snd_cwnd;
299 tp->ccv->flags |= CCF_ABC_SENTAWND;
302 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
303 tp->t_bytes_acked = 0;
307 if (CC_ALGO(tp)->ack_received != NULL) {
308 /* XXXLAS: Find a way to live without this */
309 tp->ccv->curack = th->th_ack;
310 CC_ALGO(tp)->ack_received(tp->ccv, type);
315 cc_conn_init(struct tcpcb *tp)
317 struct hc_metrics_lite metrics;
318 struct inpcb *inp = tp->t_inpcb;
321 INP_WLOCK_ASSERT(tp->t_inpcb);
323 tcp_hc_get(&inp->inp_inc, &metrics);
325 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
327 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
328 TCPSTAT_INC(tcps_usedrtt);
329 if (metrics.rmx_rttvar) {
330 tp->t_rttvar = metrics.rmx_rttvar;
331 TCPSTAT_INC(tcps_usedrttvar);
333 /* default variation is +- 1 rtt */
335 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
337 TCPT_RANGESET(tp->t_rxtcur,
338 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
339 tp->t_rttmin, TCPTV_REXMTMAX);
341 if (metrics.rmx_ssthresh) {
343 * There's some sort of gateway or interface
344 * buffer limit on the path. Use this to set
345 * the slow start threshhold, but set the
346 * threshold to no less than 2*mss.
348 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh);
349 TCPSTAT_INC(tcps_usedssthresh);
353 * Set the initial slow-start flight size.
355 * RFC5681 Section 3.1 specifies the default conservative values.
356 * RFC3390 specifies slightly more aggressive values.
357 * Draft-ietf-tcpm-initcwnd-05 increases it to ten segments.
359 * If a SYN or SYN/ACK was lost and retransmitted, we have to
360 * reduce the initial CWND to one segment as congestion is likely
361 * requiring us to be cautious.
363 if (tp->snd_cwnd == 1)
364 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */
365 else if (V_tcp_do_initcwnd10)
366 tp->snd_cwnd = min(10 * tp->t_maxseg,
367 max(2 * tp->t_maxseg, 14600));
368 else if (V_tcp_do_rfc3390)
369 tp->snd_cwnd = min(4 * tp->t_maxseg,
370 max(2 * tp->t_maxseg, 4380));
372 /* Per RFC5681 Section 3.1 */
373 if (tp->t_maxseg > 2190)
374 tp->snd_cwnd = 2 * tp->t_maxseg;
375 else if (tp->t_maxseg > 1095)
376 tp->snd_cwnd = 3 * tp->t_maxseg;
378 tp->snd_cwnd = 4 * tp->t_maxseg;
381 if (CC_ALGO(tp)->conn_init != NULL)
382 CC_ALGO(tp)->conn_init(tp->ccv);
386 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
388 INP_WLOCK_ASSERT(tp->t_inpcb);
392 if (!IN_FASTRECOVERY(tp->t_flags)) {
393 tp->snd_recover = tp->snd_max;
394 if (tp->t_flags & TF_ECN_PERMIT)
395 tp->t_flags |= TF_ECN_SND_CWR;
399 if (!IN_CONGRECOVERY(tp->t_flags)) {
400 TCPSTAT_INC(tcps_ecn_rcwnd);
401 tp->snd_recover = tp->snd_max;
402 if (tp->t_flags & TF_ECN_PERMIT)
403 tp->t_flags |= TF_ECN_SND_CWR;
408 tp->t_bytes_acked = 0;
409 EXIT_RECOVERY(tp->t_flags);
410 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
411 tp->t_maxseg) * tp->t_maxseg;
412 tp->snd_cwnd = tp->t_maxseg;
415 TCPSTAT_INC(tcps_sndrexmitbad);
416 /* RTO was unnecessary, so reset everything. */
417 tp->snd_cwnd = tp->snd_cwnd_prev;
418 tp->snd_ssthresh = tp->snd_ssthresh_prev;
419 tp->snd_recover = tp->snd_recover_prev;
420 if (tp->t_flags & TF_WASFRECOVERY)
421 ENTER_FASTRECOVERY(tp->t_flags);
422 if (tp->t_flags & TF_WASCRECOVERY)
423 ENTER_CONGRECOVERY(tp->t_flags);
424 tp->snd_nxt = tp->snd_max;
425 tp->t_flags &= ~TF_PREVVALID;
430 if (CC_ALGO(tp)->cong_signal != NULL) {
432 tp->ccv->curack = th->th_ack;
433 CC_ALGO(tp)->cong_signal(tp->ccv, type);
438 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
440 INP_WLOCK_ASSERT(tp->t_inpcb);
442 /* XXXLAS: KASSERT that we're in recovery? */
444 if (CC_ALGO(tp)->post_recovery != NULL) {
445 tp->ccv->curack = th->th_ack;
446 CC_ALGO(tp)->post_recovery(tp->ccv);
448 /* XXXLAS: EXIT_RECOVERY ? */
449 tp->t_bytes_acked = 0;
453 tcp_fields_to_host(struct tcphdr *th)
456 th->th_seq = ntohl(th->th_seq);
457 th->th_ack = ntohl(th->th_ack);
458 th->th_win = ntohs(th->th_win);
459 th->th_urp = ntohs(th->th_urp);
464 tcp_fields_to_net(struct tcphdr *th)
467 th->th_seq = htonl(th->th_seq);
468 th->th_ack = htonl(th->th_ack);
469 th->th_win = htons(th->th_win);
470 th->th_urp = htons(th->th_urp);
474 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
475 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
479 tcp_fields_to_net(th);
480 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
481 tcp_fields_to_host(th);
486 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
488 #define ND6_HINT(tp) \
490 if ((tp) && (tp)->t_inpcb && \
491 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
492 nd6_nud_hint(NULL, NULL, 0); \
499 * Indicate whether this ack should be delayed. We can delay the ack if
500 * - there is no delayed ack timer in progress and
501 * - our last ack wasn't a 0-sized window. We never want to delay
502 * the ack that opens up a 0-sized window and
503 * - delayed acks are enabled or
504 * - this is a half-synchronized T/TCP connection.
506 #define DELAY_ACK(tp) \
507 ((!tcp_timer_active(tp, TT_DELACK) && \
508 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
509 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
512 * TCP input handling is split into multiple parts:
513 * tcp6_input is a thin wrapper around tcp_input for the extended
514 * ip6_protox[] call format in ip6_input
515 * tcp_input handles primary segment validation, inpcb lookup and
516 * SYN processing on listen sockets
517 * tcp_do_segment processes the ACK and text of the segment for
518 * establishing, established and closing connections
522 tcp6_input(struct mbuf **mp, int *offp, int proto)
524 struct mbuf *m = *mp;
525 struct in6_ifaddr *ia6;
527 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
530 * draft-itojun-ipv6-tcp-to-anycast
531 * better place to put this in?
533 ia6 = ip6_getdstifaddr(m);
534 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
537 ifa_free(&ia6->ia_ifa);
538 ip6 = mtod(m, struct ip6_hdr *);
539 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
540 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
544 ifa_free(&ia6->ia_ifa);
552 tcp_input(struct mbuf *m, int off0)
554 struct tcphdr *th = NULL;
555 struct ip *ip = NULL;
559 struct inpcb *inp = NULL;
560 struct tcpcb *tp = NULL;
561 struct socket *so = NULL;
570 int rstreason = 0; /* For badport_bandlim accounting purposes */
572 uint8_t sig_checked = 0;
575 struct m_tag *fwd_tag = NULL;
577 struct ip6_hdr *ip6 = NULL;
580 const void *ip6 = NULL;
582 struct tcpopt to; /* options in this segment */
583 char *s = NULL; /* address and port logging */
585 #define TI_UNLOCKED 1
590 * The size of tcp_saveipgen must be the size of the max ip header,
593 u_char tcp_saveipgen[IP6_HDR_LEN];
594 struct tcphdr tcp_savetcp;
599 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
603 TCPSTAT_INC(tcps_rcvtotal);
607 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
609 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
610 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
612 TCPSTAT_INC(tcps_rcvshort);
617 ip6 = mtod(m, struct ip6_hdr *);
618 th = (struct tcphdr *)((caddr_t)ip6 + off0);
619 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
620 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
621 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
622 th->th_sum = m->m_pkthdr.csum_data;
624 th->th_sum = in6_cksum_pseudo(ip6, tlen,
625 IPPROTO_TCP, m->m_pkthdr.csum_data);
626 th->th_sum ^= 0xffff;
628 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
630 TCPSTAT_INC(tcps_rcvbadsum);
635 * Be proactive about unspecified IPv6 address in source.
636 * As we use all-zero to indicate unbounded/unconnected pcb,
637 * unspecified IPv6 address can be used to confuse us.
639 * Note that packets with unspecified IPv6 destination is
640 * already dropped in ip6_input.
642 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
648 #if defined(INET) && defined(INET6)
654 * Get IP and TCP header together in first mbuf.
655 * Note: IP leaves IP header in first mbuf.
657 if (off0 > sizeof (struct ip)) {
658 ip_stripoptions(m, (struct mbuf *)0);
659 off0 = sizeof(struct ip);
661 if (m->m_len < sizeof (struct tcpiphdr)) {
662 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
664 TCPSTAT_INC(tcps_rcvshort);
668 ip = mtod(m, struct ip *);
669 ipov = (struct ipovly *)ip;
670 th = (struct tcphdr *)((caddr_t)ip + off0);
673 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
674 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
675 th->th_sum = m->m_pkthdr.csum_data;
677 th->th_sum = in_pseudo(ip->ip_src.s_addr,
679 htonl(m->m_pkthdr.csum_data +
682 th->th_sum ^= 0xffff;
684 ipov->ih_len = (u_short)tlen;
685 ipov->ih_len = htons(ipov->ih_len);
689 * Checksum extended TCP header and data.
691 len = sizeof (struct ip) + tlen;
692 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
693 ipov->ih_len = (u_short)tlen;
694 ipov->ih_len = htons(ipov->ih_len);
695 th->th_sum = in_cksum(m, len);
698 TCPSTAT_INC(tcps_rcvbadsum);
701 /* Re-initialization for later version check */
702 ip->ip_v = IPVERSION;
708 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
710 #if defined(INET) && defined(INET6)
718 * Check that TCP offset makes sense,
719 * pull out TCP options and adjust length. XXX
721 off = th->th_off << 2;
722 if (off < sizeof (struct tcphdr) || off > tlen) {
723 TCPSTAT_INC(tcps_rcvbadoff);
726 tlen -= off; /* tlen is used instead of ti->ti_len */
727 if (off > sizeof (struct tcphdr)) {
730 IP6_EXTHDR_CHECK(m, off0, off, );
731 ip6 = mtod(m, struct ip6_hdr *);
732 th = (struct tcphdr *)((caddr_t)ip6 + off0);
735 #if defined(INET) && defined(INET6)
740 if (m->m_len < sizeof(struct ip) + off) {
741 if ((m = m_pullup(m, sizeof (struct ip) + off))
743 TCPSTAT_INC(tcps_rcvshort);
746 ip = mtod(m, struct ip *);
747 ipov = (struct ipovly *)ip;
748 th = (struct tcphdr *)((caddr_t)ip + off0);
752 optlen = off - sizeof (struct tcphdr);
753 optp = (u_char *)(th + 1);
755 thflags = th->th_flags;
758 * Convert TCP protocol specific fields to host format.
760 tcp_fields_to_host(th);
763 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
765 drop_hdrlen = off0 + off;
768 * Locate pcb for segment; if we're likely to add or remove a
769 * connection then first acquire pcbinfo lock. There are two cases
770 * where we might discover later we need a write lock despite the
771 * flags: ACKs moving a connection out of the syncache, and ACKs for
772 * a connection in TIMEWAIT.
774 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) {
775 INP_INFO_WLOCK(&V_tcbinfo);
776 ti_locked = TI_WLOCKED;
778 ti_locked = TI_UNLOCKED;
782 if (ti_locked == TI_WLOCKED) {
783 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
785 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
790 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
794 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
796 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
799 #if defined(INET) && !defined(INET6)
800 (m->m_flags & M_IP_NEXTHOP)
803 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
806 if (isipv6 && fwd_tag != NULL) {
807 struct sockaddr_in6 *next_hop6;
809 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
811 * Transparently forwarded. Pretend to be the destination.
812 * Already got one like this?
814 inp = in6_pcblookup_mbuf(&V_tcbinfo,
815 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
816 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
819 * It's new. Try to find the ambushing socket.
820 * Because we've rewritten the destination address,
821 * any hardware-generated hash is ignored.
823 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
824 th->th_sport, &next_hop6->sin6_addr,
825 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
826 th->th_dport, INPLOOKUP_WILDCARD |
827 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
829 /* Remove the tag from the packet. We don't need it anymore. */
830 m_tag_delete(m, fwd_tag);
831 m->m_flags &= ~M_IP6_NEXTHOP;
834 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
835 th->th_sport, &ip6->ip6_dst, th->th_dport,
836 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
837 m->m_pkthdr.rcvif, m);
840 #if defined(INET6) && defined(INET)
844 if (fwd_tag != NULL) {
845 struct sockaddr_in *next_hop;
847 next_hop = (struct sockaddr_in *)(fwd_tag+1);
849 * Transparently forwarded. Pretend to be the destination.
850 * already got one like this?
852 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
853 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
854 m->m_pkthdr.rcvif, m);
857 * It's new. Try to find the ambushing socket.
858 * Because we've rewritten the destination address,
859 * any hardware-generated hash is ignored.
861 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
862 th->th_sport, next_hop->sin_addr,
863 next_hop->sin_port ? ntohs(next_hop->sin_port) :
864 th->th_dport, INPLOOKUP_WILDCARD |
865 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
867 /* Remove the tag from the packet. We don't need it anymore. */
868 m_tag_delete(m, fwd_tag);
869 m->m_flags &= ~M_IP_NEXTHOP;
872 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
873 th->th_sport, ip->ip_dst, th->th_dport,
874 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
875 m->m_pkthdr.rcvif, m);
879 * If the INPCB does not exist then all data in the incoming
880 * segment is discarded and an appropriate RST is sent back.
881 * XXX MRT Send RST using which routing table?
885 * Log communication attempts to ports that are not
888 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
889 tcp_log_in_vain == 2) {
890 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
891 log(LOG_INFO, "%s; %s: Connection attempt "
892 "to closed port\n", s, __func__);
895 * When blackholing do not respond with a RST but
896 * completely ignore the segment and drop it.
898 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
902 rstreason = BANDLIM_RST_CLOSEDPORT;
905 INP_WLOCK_ASSERT(inp);
906 if (!(inp->inp_flags & INP_HW_FLOWID)
907 && (m->m_flags & M_FLOWID)
908 && ((inp->inp_socket == NULL)
909 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) {
910 inp->inp_flags |= INP_HW_FLOWID;
911 inp->inp_flags &= ~INP_SW_FLOWID;
912 inp->inp_flowid = m->m_pkthdr.flowid;
916 if (isipv6 && ipsec6_in_reject(m, inp)) {
917 IPSEC6STAT_INC(in_polvio);
921 if (ipsec4_in_reject(m, inp) != 0) {
922 IPSECSTAT_INC(in_polvio);
928 * Check the minimum TTL for socket.
930 if (inp->inp_ip_minttl != 0) {
932 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
936 if (inp->inp_ip_minttl > ip->ip_ttl)
941 * A previous connection in TIMEWAIT state is supposed to catch stray
942 * or duplicate segments arriving late. If this segment was a
943 * legitimate new connection attempt the old INPCB gets removed and
944 * we can try again to find a listening socket.
946 * At this point, due to earlier optimism, we may hold only an inpcb
947 * lock, and not the inpcbinfo write lock. If so, we need to try to
948 * acquire it, or if that fails, acquire a reference on the inpcb,
949 * drop all locks, acquire a global write lock, and then re-acquire
950 * the inpcb lock. We may at that point discover that another thread
951 * has tried to free the inpcb, in which case we need to loop back
952 * and try to find a new inpcb to deliver to.
954 * XXXRW: It may be time to rethink timewait locking.
957 if (inp->inp_flags & INP_TIMEWAIT) {
958 if (ti_locked == TI_UNLOCKED) {
959 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
962 INP_INFO_WLOCK(&V_tcbinfo);
963 ti_locked = TI_WLOCKED;
965 if (in_pcbrele_wlocked(inp)) {
970 ti_locked = TI_WLOCKED;
972 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
974 if (thflags & TH_SYN)
975 tcp_dooptions(&to, optp, optlen, TO_SYN);
977 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
979 if (tcp_twcheck(inp, &to, th, m, tlen))
981 INP_INFO_WUNLOCK(&V_tcbinfo);
985 * The TCPCB may no longer exist if the connection is winding
986 * down or it is in the CLOSED state. Either way we drop the
987 * segment and send an appropriate response.
990 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
991 rstreason = BANDLIM_RST_CLOSEDPORT;
996 if (tp->t_flags & TF_TOE) {
997 tcp_offload_input(tp, m);
998 m = NULL; /* consumed by the TOE driver */
1004 * We've identified a valid inpcb, but it could be that we need an
1005 * inpcbinfo write lock but don't hold it. In this case, attempt to
1006 * acquire using the same strategy as the TIMEWAIT case above. If we
1007 * relock, we have to jump back to 'relocked' as the connection might
1008 * now be in TIMEWAIT.
1011 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0)
1012 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1014 if (tp->t_state != TCPS_ESTABLISHED) {
1015 if (ti_locked == TI_UNLOCKED) {
1016 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
1019 INP_INFO_WLOCK(&V_tcbinfo);
1020 ti_locked = TI_WLOCKED;
1022 if (in_pcbrele_wlocked(inp)) {
1028 ti_locked = TI_WLOCKED;
1030 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1034 INP_WLOCK_ASSERT(inp);
1035 if (mac_inpcb_check_deliver(inp, m))
1038 so = inp->inp_socket;
1039 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1041 if (so->so_options & SO_DEBUG) {
1042 ostate = tp->t_state;
1045 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1048 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1051 #endif /* TCPDEBUG */
1053 * When the socket is accepting connections (the INPCB is in LISTEN
1054 * state) we look into the SYN cache if this is a new connection
1055 * attempt or the completion of a previous one. Because listen
1056 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be
1057 * held in this case.
1059 if (so->so_options & SO_ACCEPTCONN) {
1060 struct in_conninfo inc;
1062 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
1063 "tp not listening", __func__));
1064 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1066 bzero(&inc, sizeof(inc));
1069 inc.inc_flags |= INC_ISIPV6;
1070 inc.inc6_faddr = ip6->ip6_src;
1071 inc.inc6_laddr = ip6->ip6_dst;
1075 inc.inc_faddr = ip->ip_src;
1076 inc.inc_laddr = ip->ip_dst;
1078 inc.inc_fport = th->th_sport;
1079 inc.inc_lport = th->th_dport;
1080 inc.inc_fibnum = so->so_fibnum;
1083 * Check for an existing connection attempt in syncache if
1084 * the flag is only ACK. A successful lookup creates a new
1085 * socket appended to the listen queue in SYN_RECEIVED state.
1087 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1089 * Parse the TCP options here because
1090 * syncookies need access to the reflected
1093 tcp_dooptions(&to, optp, optlen, 0);
1095 * NB: syncache_expand() doesn't unlock
1096 * inp and tcpinfo locks.
1098 if (!syncache_expand(&inc, &to, th, &so, m)) {
1100 * No syncache entry or ACK was not
1101 * for our SYN/ACK. Send a RST.
1102 * NB: syncache did its own logging
1103 * of the failure cause.
1105 rstreason = BANDLIM_RST_OPENPORT;
1110 * We completed the 3-way handshake
1111 * but could not allocate a socket
1112 * either due to memory shortage,
1113 * listen queue length limits or
1114 * global socket limits. Send RST
1115 * or wait and have the remote end
1116 * retransmit the ACK for another
1119 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1120 log(LOG_DEBUG, "%s; %s: Listen socket: "
1121 "Socket allocation failed due to "
1122 "limits or memory shortage, %s\n",
1124 V_tcp_sc_rst_sock_fail ?
1125 "sending RST" : "try again");
1126 if (V_tcp_sc_rst_sock_fail) {
1127 rstreason = BANDLIM_UNLIMITED;
1133 * Socket is created in state SYN_RECEIVED.
1134 * Unlock the listen socket, lock the newly
1135 * created socket and update the tp variable.
1137 INP_WUNLOCK(inp); /* listen socket */
1138 inp = sotoinpcb(so);
1139 INP_WLOCK(inp); /* new connection */
1140 tp = intotcpcb(inp);
1141 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1142 ("%s: ", __func__));
1143 #ifdef TCP_SIGNATURE
1144 if (sig_checked == 0) {
1145 tcp_dooptions(&to, optp, optlen,
1146 (thflags & TH_SYN) ? TO_SYN : 0);
1147 if (!tcp_signature_verify_input(m, off0, tlen,
1148 optlen, &to, th, tp->t_flags)) {
1151 * In SYN_SENT state if it receives an
1152 * RST, it is allowed for further
1155 if ((thflags & TH_RST) == 0 ||
1156 (tp->t_state == TCPS_SYN_SENT) == 0)
1164 * Process the segment and the data it
1165 * contains. tcp_do_segment() consumes
1166 * the mbuf chain and unlocks the inpcb.
1168 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1170 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1174 * Segment flag validation for new connection attempts:
1176 * Our (SYN|ACK) response was rejected.
1177 * Check with syncache and remove entry to prevent
1180 * NB: syncache_chkrst does its own logging of failure
1183 if (thflags & TH_RST) {
1184 syncache_chkrst(&inc, th);
1188 * We can't do anything without SYN.
1190 if ((thflags & TH_SYN) == 0) {
1191 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1192 log(LOG_DEBUG, "%s; %s: Listen socket: "
1193 "SYN is missing, segment ignored\n",
1195 TCPSTAT_INC(tcps_badsyn);
1199 * (SYN|ACK) is bogus on a listen socket.
1201 if (thflags & TH_ACK) {
1202 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1203 log(LOG_DEBUG, "%s; %s: Listen socket: "
1204 "SYN|ACK invalid, segment rejected\n",
1206 syncache_badack(&inc); /* XXX: Not needed! */
1207 TCPSTAT_INC(tcps_badsyn);
1208 rstreason = BANDLIM_RST_OPENPORT;
1212 * If the drop_synfin option is enabled, drop all
1213 * segments with both the SYN and FIN bits set.
1214 * This prevents e.g. nmap from identifying the
1216 * XXX: Poor reasoning. nmap has other methods
1217 * and is constantly refining its stack detection
1219 * XXX: This is a violation of the TCP specification
1220 * and was used by RFC1644.
1222 if ((thflags & TH_FIN) && V_drop_synfin) {
1223 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1224 log(LOG_DEBUG, "%s; %s: Listen socket: "
1225 "SYN|FIN segment ignored (based on "
1226 "sysctl setting)\n", s, __func__);
1227 TCPSTAT_INC(tcps_badsyn);
1231 * Segment's flags are (SYN) or (SYN|FIN).
1233 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1234 * as they do not affect the state of the TCP FSM.
1235 * The data pointed to by TH_URG and th_urp is ignored.
1237 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1238 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1239 KASSERT(thflags & (TH_SYN),
1240 ("%s: Listen socket: TH_SYN not set", __func__));
1243 * If deprecated address is forbidden,
1244 * we do not accept SYN to deprecated interface
1245 * address to prevent any new inbound connection from
1246 * getting established.
1247 * When we do not accept SYN, we send a TCP RST,
1248 * with deprecated source address (instead of dropping
1249 * it). We compromise it as it is much better for peer
1250 * to send a RST, and RST will be the final packet
1253 * If we do not forbid deprecated addresses, we accept
1254 * the SYN packet. RFC2462 does not suggest dropping
1256 * If we decipher RFC2462 5.5.4, it says like this:
1257 * 1. use of deprecated addr with existing
1258 * communication is okay - "SHOULD continue to be
1260 * 2. use of it with new communication:
1261 * (2a) "SHOULD NOT be used if alternate address
1262 * with sufficient scope is available"
1263 * (2b) nothing mentioned otherwise.
1264 * Here we fall into (2b) case as we have no choice in
1265 * our source address selection - we must obey the peer.
1267 * The wording in RFC2462 is confusing, and there are
1268 * multiple description text for deprecated address
1269 * handling - worse, they are not exactly the same.
1270 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1272 if (isipv6 && !V_ip6_use_deprecated) {
1273 struct in6_ifaddr *ia6;
1275 ia6 = ip6_getdstifaddr(m);
1277 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1278 ifa_free(&ia6->ia_ifa);
1279 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1280 log(LOG_DEBUG, "%s; %s: Listen socket: "
1281 "Connection attempt to deprecated "
1282 "IPv6 address rejected\n",
1284 rstreason = BANDLIM_RST_OPENPORT;
1288 ifa_free(&ia6->ia_ifa);
1292 * Basic sanity checks on incoming SYN requests:
1293 * Don't respond if the destination is a link layer
1294 * broadcast according to RFC1122 4.2.3.10, p. 104.
1295 * If it is from this socket it must be forged.
1296 * Don't respond if the source or destination is a
1297 * global or subnet broad- or multicast address.
1298 * Note that it is quite possible to receive unicast
1299 * link-layer packets with a broadcast IP address. Use
1300 * in_broadcast() to find them.
1302 if (m->m_flags & (M_BCAST|M_MCAST)) {
1303 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1304 log(LOG_DEBUG, "%s; %s: Listen socket: "
1305 "Connection attempt from broad- or multicast "
1306 "link layer address ignored\n", s, __func__);
1311 if (th->th_dport == th->th_sport &&
1312 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1313 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1314 log(LOG_DEBUG, "%s; %s: Listen socket: "
1315 "Connection attempt to/from self "
1316 "ignored\n", s, __func__);
1319 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1320 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1321 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1322 log(LOG_DEBUG, "%s; %s: Listen socket: "
1323 "Connection attempt from/to multicast "
1324 "address ignored\n", s, __func__);
1329 #if defined(INET) && defined(INET6)
1334 if (th->th_dport == th->th_sport &&
1335 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1336 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1337 log(LOG_DEBUG, "%s; %s: Listen socket: "
1338 "Connection attempt from/to self "
1339 "ignored\n", s, __func__);
1342 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1343 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1344 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1345 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1346 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1347 log(LOG_DEBUG, "%s; %s: Listen socket: "
1348 "Connection attempt from/to broad- "
1349 "or multicast address ignored\n",
1356 * SYN appears to be valid. Create compressed TCP state
1360 if (so->so_options & SO_DEBUG)
1361 tcp_trace(TA_INPUT, ostate, tp,
1362 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1364 tcp_dooptions(&to, optp, optlen, TO_SYN);
1365 syncache_add(&inc, &to, th, inp, &so, m);
1367 * Entry added to syncache and mbuf consumed.
1368 * Everything already unlocked by syncache_add().
1370 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1372 } else if (tp->t_state == TCPS_LISTEN) {
1374 * When a listen socket is torn down the SO_ACCEPTCONN
1375 * flag is removed first while connections are drained
1376 * from the accept queue in a unlock/lock cycle of the
1377 * ACCEPT_LOCK, opening a race condition allowing a SYN
1378 * attempt go through unhandled.
1383 #ifdef TCP_SIGNATURE
1384 if (sig_checked == 0) {
1385 tcp_dooptions(&to, optp, optlen,
1386 (thflags & TH_SYN) ? TO_SYN : 0);
1387 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1391 * In SYN_SENT state if it receives an RST, it is
1392 * allowed for further processing.
1394 if ((thflags & TH_RST) == 0 ||
1395 (tp->t_state == TCPS_SYN_SENT) == 0)
1403 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1404 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1405 * the inpcb, and unlocks pcbinfo.
1407 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1408 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1412 if (ti_locked == TI_WLOCKED) {
1413 INP_INFO_WUNLOCK(&V_tcbinfo);
1414 ti_locked = TI_UNLOCKED;
1418 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1419 "ti_locked: %d", __func__, ti_locked));
1420 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1425 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1428 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1429 m = NULL; /* mbuf chain got consumed. */
1433 if (ti_locked == TI_WLOCKED) {
1434 INP_INFO_WUNLOCK(&V_tcbinfo);
1435 ti_locked = TI_UNLOCKED;
1439 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1440 "ti_locked: %d", __func__, ti_locked));
1441 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1449 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1457 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1458 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1461 int thflags, acked, ourfinisacked, needoutput = 0;
1462 int rstreason, todrop, win;
1468 * The size of tcp_saveipgen must be the size of the max ip header,
1471 u_char tcp_saveipgen[IP6_HDR_LEN];
1472 struct tcphdr tcp_savetcp;
1475 thflags = th->th_flags;
1476 tp->sackhint.last_sack_ack = 0;
1479 * If this is either a state-changing packet or current state isn't
1480 * established, we require a write lock on tcbinfo. Otherwise, we
1481 * allow either a read lock or a write lock, as we may have acquired
1482 * a write lock due to a race.
1484 * Require a global write lock for SYN/FIN/RST segments or
1485 * non-established connections; otherwise accept either a read or
1486 * write lock, as we may have conservatively acquired a write lock in
1487 * certain cases in tcp_input() (is this still true?). Currently we
1488 * will never enter with no lock, so we try to drop it quickly in the
1489 * common pure ack/pure data cases.
1491 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1492 tp->t_state != TCPS_ESTABLISHED) {
1493 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for "
1494 "SYN/FIN/RST/!EST", __func__, ti_locked));
1495 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1498 if (ti_locked == TI_WLOCKED)
1499 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1501 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1502 "ti_locked: %d", __func__, ti_locked));
1503 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1507 INP_WLOCK_ASSERT(tp->t_inpcb);
1508 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1510 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1514 * Segment received on connection.
1515 * Reset idle time and keep-alive timer.
1516 * XXX: This should be done after segment
1517 * validation to ignore broken/spoofed segs.
1519 tp->t_rcvtime = ticks;
1520 if (TCPS_HAVEESTABLISHED(tp->t_state))
1521 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
1524 * Unscale the window into a 32-bit value.
1525 * For the SYN_SENT state the scale is zero.
1527 tiwin = th->th_win << tp->snd_scale;
1530 * TCP ECN processing.
1532 if (tp->t_flags & TF_ECN_PERMIT) {
1533 if (thflags & TH_CWR)
1534 tp->t_flags &= ~TF_ECN_SND_ECE;
1535 switch (iptos & IPTOS_ECN_MASK) {
1537 tp->t_flags |= TF_ECN_SND_ECE;
1538 TCPSTAT_INC(tcps_ecn_ce);
1540 case IPTOS_ECN_ECT0:
1541 TCPSTAT_INC(tcps_ecn_ect0);
1543 case IPTOS_ECN_ECT1:
1544 TCPSTAT_INC(tcps_ecn_ect1);
1547 /* Congestion experienced. */
1548 if (thflags & TH_ECE) {
1549 cc_cong_signal(tp, th, CC_ECN);
1554 * Parse options on any incoming segment.
1556 tcp_dooptions(&to, (u_char *)(th + 1),
1557 (th->th_off << 2) - sizeof(struct tcphdr),
1558 (thflags & TH_SYN) ? TO_SYN : 0);
1561 * If echoed timestamp is later than the current time,
1562 * fall back to non RFC1323 RTT calculation. Normalize
1563 * timestamp if syncookies were used when this connection
1566 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1567 to.to_tsecr -= tp->ts_offset;
1568 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1573 * Process options only when we get SYN/ACK back. The SYN case
1574 * for incoming connections is handled in tcp_syncache.
1575 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1576 * or <SYN,ACK>) segment itself is never scaled.
1577 * XXX this is traditional behavior, may need to be cleaned up.
1579 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1580 if ((to.to_flags & TOF_SCALE) &&
1581 (tp->t_flags & TF_REQ_SCALE)) {
1582 tp->t_flags |= TF_RCVD_SCALE;
1583 tp->snd_scale = to.to_wscale;
1586 * Initial send window. It will be updated with
1587 * the next incoming segment to the scaled value.
1589 tp->snd_wnd = th->th_win;
1590 if (to.to_flags & TOF_TS) {
1591 tp->t_flags |= TF_RCVD_TSTMP;
1592 tp->ts_recent = to.to_tsval;
1593 tp->ts_recent_age = tcp_ts_getticks();
1595 if (to.to_flags & TOF_MSS)
1596 tcp_mss(tp, to.to_mss);
1597 if ((tp->t_flags & TF_SACK_PERMIT) &&
1598 (to.to_flags & TOF_SACKPERM) == 0)
1599 tp->t_flags &= ~TF_SACK_PERMIT;
1603 * Header prediction: check for the two common cases
1604 * of a uni-directional data xfer. If the packet has
1605 * no control flags, is in-sequence, the window didn't
1606 * change and we're not retransmitting, it's a
1607 * candidate. If the length is zero and the ack moved
1608 * forward, we're the sender side of the xfer. Just
1609 * free the data acked & wake any higher level process
1610 * that was blocked waiting for space. If the length
1611 * is non-zero and the ack didn't move, we're the
1612 * receiver side. If we're getting packets in-order
1613 * (the reassembly queue is empty), add the data to
1614 * the socket buffer and note that we need a delayed ack.
1615 * Make sure that the hidden state-flags are also off.
1616 * Since we check for TCPS_ESTABLISHED first, it can only
1619 if (tp->t_state == TCPS_ESTABLISHED &&
1620 th->th_seq == tp->rcv_nxt &&
1621 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1622 tp->snd_nxt == tp->snd_max &&
1623 tiwin && tiwin == tp->snd_wnd &&
1624 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1625 LIST_EMPTY(&tp->t_segq) &&
1626 ((to.to_flags & TOF_TS) == 0 ||
1627 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1630 * If last ACK falls within this segment's sequence numbers,
1631 * record the timestamp.
1632 * NOTE that the test is modified according to the latest
1633 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1635 if ((to.to_flags & TOF_TS) != 0 &&
1636 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1637 tp->ts_recent_age = tcp_ts_getticks();
1638 tp->ts_recent = to.to_tsval;
1642 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1643 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1644 !IN_RECOVERY(tp->t_flags) &&
1645 (to.to_flags & TOF_SACK) == 0 &&
1646 TAILQ_EMPTY(&tp->snd_holes)) {
1648 * This is a pure ack for outstanding data.
1650 if (ti_locked == TI_WLOCKED)
1651 INP_INFO_WUNLOCK(&V_tcbinfo);
1652 ti_locked = TI_UNLOCKED;
1654 TCPSTAT_INC(tcps_predack);
1657 * "bad retransmit" recovery.
1659 if (tp->t_rxtshift == 1 &&
1660 tp->t_flags & TF_PREVVALID &&
1661 (int)(ticks - tp->t_badrxtwin) < 0) {
1662 cc_cong_signal(tp, th, CC_RTO_ERR);
1666 * Recalculate the transmit timer / rtt.
1668 * Some boxes send broken timestamp replies
1669 * during the SYN+ACK phase, ignore
1670 * timestamps of 0 or we could calculate a
1671 * huge RTT and blow up the retransmit timer.
1673 if ((to.to_flags & TOF_TS) != 0 &&
1677 t = tcp_ts_getticks() - to.to_tsecr;
1678 if (!tp->t_rttlow || tp->t_rttlow > t)
1681 TCP_TS_TO_TICKS(t) + 1);
1682 } else if (tp->t_rtttime &&
1683 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1684 if (!tp->t_rttlow ||
1685 tp->t_rttlow > ticks - tp->t_rtttime)
1686 tp->t_rttlow = ticks - tp->t_rtttime;
1688 ticks - tp->t_rtttime);
1690 acked = BYTES_THIS_ACK(tp, th);
1692 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1693 hhook_run_tcp_est_in(tp, th, &to);
1695 TCPSTAT_INC(tcps_rcvackpack);
1696 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1697 sbdrop(&so->so_snd, acked);
1698 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1699 SEQ_LEQ(th->th_ack, tp->snd_recover))
1700 tp->snd_recover = th->th_ack - 1;
1703 * Let the congestion control algorithm update
1704 * congestion control related information. This
1705 * typically means increasing the congestion
1708 cc_ack_received(tp, th, CC_ACK);
1710 tp->snd_una = th->th_ack;
1712 * Pull snd_wl2 up to prevent seq wrap relative
1715 tp->snd_wl2 = th->th_ack;
1718 ND6_HINT(tp); /* Some progress has been made. */
1721 * If all outstanding data are acked, stop
1722 * retransmit timer, otherwise restart timer
1723 * using current (possibly backed-off) value.
1724 * If process is waiting for space,
1725 * wakeup/selwakeup/signal. If data
1726 * are ready to send, let tcp_output
1727 * decide between more output or persist.
1730 if (so->so_options & SO_DEBUG)
1731 tcp_trace(TA_INPUT, ostate, tp,
1732 (void *)tcp_saveipgen,
1735 if (tp->snd_una == tp->snd_max)
1736 tcp_timer_activate(tp, TT_REXMT, 0);
1737 else if (!tcp_timer_active(tp, TT_PERSIST))
1738 tcp_timer_activate(tp, TT_REXMT,
1741 if (so->so_snd.sb_cc)
1742 (void) tcp_output(tp);
1745 } else if (th->th_ack == tp->snd_una &&
1746 tlen <= sbspace(&so->so_rcv)) {
1747 int newsize = 0; /* automatic sockbuf scaling */
1750 * This is a pure, in-sequence data packet with
1751 * nothing on the reassembly queue and we have enough
1752 * buffer space to take it.
1754 if (ti_locked == TI_WLOCKED)
1755 INP_INFO_WUNLOCK(&V_tcbinfo);
1756 ti_locked = TI_UNLOCKED;
1758 /* Clean receiver SACK report if present */
1759 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1760 tcp_clean_sackreport(tp);
1761 TCPSTAT_INC(tcps_preddat);
1762 tp->rcv_nxt += tlen;
1764 * Pull snd_wl1 up to prevent seq wrap relative to
1767 tp->snd_wl1 = th->th_seq;
1769 * Pull rcv_up up to prevent seq wrap relative to
1772 tp->rcv_up = tp->rcv_nxt;
1773 TCPSTAT_INC(tcps_rcvpack);
1774 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1775 ND6_HINT(tp); /* Some progress has been made */
1777 if (so->so_options & SO_DEBUG)
1778 tcp_trace(TA_INPUT, ostate, tp,
1779 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1782 * Automatic sizing of receive socket buffer. Often the send
1783 * buffer size is not optimally adjusted to the actual network
1784 * conditions at hand (delay bandwidth product). Setting the
1785 * buffer size too small limits throughput on links with high
1786 * bandwidth and high delay (eg. trans-continental/oceanic links).
1788 * On the receive side the socket buffer memory is only rarely
1789 * used to any significant extent. This allows us to be much
1790 * more aggressive in scaling the receive socket buffer. For
1791 * the case that the buffer space is actually used to a large
1792 * extent and we run out of kernel memory we can simply drop
1793 * the new segments; TCP on the sender will just retransmit it
1794 * later. Setting the buffer size too big may only consume too
1795 * much kernel memory if the application doesn't read() from
1796 * the socket or packet loss or reordering makes use of the
1799 * The criteria to step up the receive buffer one notch are:
1800 * 1. the number of bytes received during the time it takes
1801 * one timestamp to be reflected back to us (the RTT);
1802 * 2. received bytes per RTT is within seven eighth of the
1803 * current socket buffer size;
1804 * 3. receive buffer size has not hit maximal automatic size;
1806 * This algorithm does one step per RTT at most and only if
1807 * we receive a bulk stream w/o packet losses or reorderings.
1808 * Shrinking the buffer during idle times is not necessary as
1809 * it doesn't consume any memory when idle.
1811 * TODO: Only step up if the application is actually serving
1812 * the buffer to better manage the socket buffer resources.
1814 if (V_tcp_do_autorcvbuf &&
1816 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1817 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1818 to.to_tsecr - tp->rfbuf_ts < hz) {
1820 (so->so_rcv.sb_hiwat / 8 * 7) &&
1821 so->so_rcv.sb_hiwat <
1822 V_tcp_autorcvbuf_max) {
1824 min(so->so_rcv.sb_hiwat +
1825 V_tcp_autorcvbuf_inc,
1826 V_tcp_autorcvbuf_max);
1828 /* Start over with next RTT. */
1832 tp->rfbuf_cnt += tlen; /* add up */
1835 /* Add data to socket buffer. */
1836 SOCKBUF_LOCK(&so->so_rcv);
1837 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1841 * Set new socket buffer size.
1842 * Give up when limit is reached.
1845 if (!sbreserve_locked(&so->so_rcv,
1847 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1848 m_adj(m, drop_hdrlen); /* delayed header drop */
1849 sbappendstream_locked(&so->so_rcv, m);
1851 /* NB: sorwakeup_locked() does an implicit unlock. */
1852 sorwakeup_locked(so);
1853 if (DELAY_ACK(tp)) {
1854 tp->t_flags |= TF_DELACK;
1856 tp->t_flags |= TF_ACKNOW;
1864 * Calculate amount of space in receive window,
1865 * and then do TCP input processing.
1866 * Receive window is amount of space in rcv queue,
1867 * but not less than advertised window.
1869 win = sbspace(&so->so_rcv);
1872 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1874 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1878 switch (tp->t_state) {
1881 * If the state is SYN_RECEIVED:
1882 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1884 case TCPS_SYN_RECEIVED:
1885 if ((thflags & TH_ACK) &&
1886 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1887 SEQ_GT(th->th_ack, tp->snd_max))) {
1888 rstreason = BANDLIM_RST_OPENPORT;
1894 * If the state is SYN_SENT:
1895 * if seg contains an ACK, but not for our SYN, drop the input.
1896 * if seg contains a RST, then drop the connection.
1897 * if seg does not contain SYN, then drop it.
1898 * Otherwise this is an acceptable SYN segment
1899 * initialize tp->rcv_nxt and tp->irs
1900 * if seg contains ack then advance tp->snd_una
1901 * if seg contains an ECE and ECN support is enabled, the stream
1903 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1904 * arrange for segment to be acked (eventually)
1905 * continue processing rest of data/controls, beginning with URG
1908 if ((thflags & TH_ACK) &&
1909 (SEQ_LEQ(th->th_ack, tp->iss) ||
1910 SEQ_GT(th->th_ack, tp->snd_max))) {
1911 rstreason = BANDLIM_UNLIMITED;
1914 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST))
1915 tp = tcp_drop(tp, ECONNREFUSED);
1916 if (thflags & TH_RST)
1918 if (!(thflags & TH_SYN))
1921 tp->irs = th->th_seq;
1923 if (thflags & TH_ACK) {
1924 TCPSTAT_INC(tcps_connects);
1927 mac_socketpeer_set_from_mbuf(m, so);
1929 /* Do window scaling on this connection? */
1930 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1931 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1932 tp->rcv_scale = tp->request_r_scale;
1934 tp->rcv_adv += imin(tp->rcv_wnd,
1935 TCP_MAXWIN << tp->rcv_scale);
1936 tp->snd_una++; /* SYN is acked */
1938 * If there's data, delay ACK; if there's also a FIN
1939 * ACKNOW will be turned on later.
1941 if (DELAY_ACK(tp) && tlen != 0)
1942 tcp_timer_activate(tp, TT_DELACK,
1945 tp->t_flags |= TF_ACKNOW;
1947 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1948 tp->t_flags |= TF_ECN_PERMIT;
1949 TCPSTAT_INC(tcps_ecn_shs);
1953 * Received <SYN,ACK> in SYN_SENT[*] state.
1955 * SYN_SENT --> ESTABLISHED
1956 * SYN_SENT* --> FIN_WAIT_1
1958 tp->t_starttime = ticks;
1959 if (tp->t_flags & TF_NEEDFIN) {
1960 tp->t_state = TCPS_FIN_WAIT_1;
1961 tp->t_flags &= ~TF_NEEDFIN;
1964 tp->t_state = TCPS_ESTABLISHED;
1966 tcp_timer_activate(tp, TT_KEEP,
1971 * Received initial SYN in SYN-SENT[*] state =>
1972 * simultaneous open. If segment contains CC option
1973 * and there is a cached CC, apply TAO test.
1974 * If it succeeds, connection is * half-synchronized.
1975 * Otherwise, do 3-way handshake:
1976 * SYN-SENT -> SYN-RECEIVED
1977 * SYN-SENT* -> SYN-RECEIVED*
1978 * If there was no CC option, clear cached CC value.
1980 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1981 tcp_timer_activate(tp, TT_REXMT, 0);
1982 tp->t_state = TCPS_SYN_RECEIVED;
1985 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: "
1986 "ti_locked %d", __func__, ti_locked));
1987 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1988 INP_WLOCK_ASSERT(tp->t_inpcb);
1991 * Advance th->th_seq to correspond to first data byte.
1992 * If data, trim to stay within window,
1993 * dropping FIN if necessary.
1996 if (tlen > tp->rcv_wnd) {
1997 todrop = tlen - tp->rcv_wnd;
2001 TCPSTAT_INC(tcps_rcvpackafterwin);
2002 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2004 tp->snd_wl1 = th->th_seq - 1;
2005 tp->rcv_up = th->th_seq;
2007 * Client side of transaction: already sent SYN and data.
2008 * If the remote host used T/TCP to validate the SYN,
2009 * our data will be ACK'd; if so, enter normal data segment
2010 * processing in the middle of step 5, ack processing.
2011 * Otherwise, goto step 6.
2013 if (thflags & TH_ACK)
2019 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2020 * do normal processing.
2022 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2026 break; /* continue normal processing */
2030 * States other than LISTEN or SYN_SENT.
2031 * First check the RST flag and sequence number since reset segments
2032 * are exempt from the timestamp and connection count tests. This
2033 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2034 * below which allowed reset segments in half the sequence space
2035 * to fall though and be processed (which gives forged reset
2036 * segments with a random sequence number a 50 percent chance of
2037 * killing a connection).
2038 * Then check timestamp, if present.
2039 * Then check the connection count, if present.
2040 * Then check that at least some bytes of segment are within
2041 * receive window. If segment begins before rcv_nxt,
2042 * drop leading data (and SYN); if nothing left, just ack.
2045 * If the RST bit is set, check the sequence number to see
2046 * if this is a valid reset segment.
2048 * In all states except SYN-SENT, all reset (RST) segments
2049 * are validated by checking their SEQ-fields. A reset is
2050 * valid if its sequence number is in the window.
2051 * Note: this does not take into account delayed ACKs, so
2052 * we should test against last_ack_sent instead of rcv_nxt.
2053 * The sequence number in the reset segment is normally an
2054 * echo of our outgoing acknowlegement numbers, but some hosts
2055 * send a reset with the sequence number at the rightmost edge
2056 * of our receive window, and we have to handle this case.
2057 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
2058 * that brute force RST attacks are possible. To combat this,
2059 * we use a much stricter check while in the ESTABLISHED state,
2060 * only accepting RSTs where the sequence number is equal to
2061 * last_ack_sent. In all other states (the states in which a
2062 * RST is more likely), the more permissive check is used.
2063 * If we have multiple segments in flight, the initial reset
2064 * segment sequence numbers will be to the left of last_ack_sent,
2065 * but they will eventually catch up.
2066 * In any case, it never made sense to trim reset segments to
2067 * fit the receive window since RFC 1122 says:
2068 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
2070 * A TCP SHOULD allow a received RST segment to include data.
2073 * It has been suggested that a RST segment could contain
2074 * ASCII text that encoded and explained the cause of the
2075 * RST. No standard has yet been established for such
2078 * If the reset segment passes the sequence number test examine
2080 * SYN_RECEIVED STATE:
2081 * If passive open, return to LISTEN state.
2082 * If active open, inform user that connection was refused.
2083 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
2084 * Inform user that connection was reset, and close tcb.
2085 * CLOSING, LAST_ACK STATES:
2088 * Drop the segment - see Stevens, vol. 2, p. 964 and
2091 if (thflags & TH_RST) {
2092 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2093 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2094 switch (tp->t_state) {
2096 case TCPS_SYN_RECEIVED:
2097 so->so_error = ECONNREFUSED;
2100 case TCPS_ESTABLISHED:
2101 if (V_tcp_insecure_rst == 0 &&
2102 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
2103 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
2104 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2105 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
2106 TCPSTAT_INC(tcps_badrst);
2110 case TCPS_FIN_WAIT_1:
2111 case TCPS_FIN_WAIT_2:
2112 case TCPS_CLOSE_WAIT:
2113 so->so_error = ECONNRESET;
2115 KASSERT(ti_locked == TI_WLOCKED,
2116 ("tcp_do_segment: TH_RST 1 ti_locked %d",
2118 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2120 tp->t_state = TCPS_CLOSED;
2121 TCPSTAT_INC(tcps_drops);
2127 KASSERT(ti_locked == TI_WLOCKED,
2128 ("tcp_do_segment: TH_RST 2 ti_locked %d",
2130 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2140 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2141 * and it's less than ts_recent, drop it.
2143 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2144 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2146 /* Check to see if ts_recent is over 24 days old. */
2147 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2149 * Invalidate ts_recent. If this segment updates
2150 * ts_recent, the age will be reset later and ts_recent
2151 * will get a valid value. If it does not, setting
2152 * ts_recent to zero will at least satisfy the
2153 * requirement that zero be placed in the timestamp
2154 * echo reply when ts_recent isn't valid. The
2155 * age isn't reset until we get a valid ts_recent
2156 * because we don't want out-of-order segments to be
2157 * dropped when ts_recent is old.
2161 TCPSTAT_INC(tcps_rcvduppack);
2162 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2163 TCPSTAT_INC(tcps_pawsdrop);
2171 * In the SYN-RECEIVED state, validate that the packet belongs to
2172 * this connection before trimming the data to fit the receive
2173 * window. Check the sequence number versus IRS since we know
2174 * the sequence numbers haven't wrapped. This is a partial fix
2175 * for the "LAND" DoS attack.
2177 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2178 rstreason = BANDLIM_RST_OPENPORT;
2182 todrop = tp->rcv_nxt - th->th_seq;
2185 * If this is a duplicate SYN for our current connection,
2186 * advance over it and pretend and it's not a SYN.
2188 if (thflags & TH_SYN && th->th_seq == tp->irs) {
2198 * Following if statement from Stevens, vol. 2, p. 960.
2201 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2203 * Any valid FIN must be to the left of the window.
2204 * At this point the FIN must be a duplicate or out
2205 * of sequence; drop it.
2210 * Send an ACK to resynchronize and drop any data.
2211 * But keep on processing for RST or ACK.
2213 tp->t_flags |= TF_ACKNOW;
2215 TCPSTAT_INC(tcps_rcvduppack);
2216 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2218 TCPSTAT_INC(tcps_rcvpartduppack);
2219 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2221 drop_hdrlen += todrop; /* drop from the top afterwards */
2222 th->th_seq += todrop;
2224 if (th->th_urp > todrop)
2225 th->th_urp -= todrop;
2233 * If new data are received on a connection after the
2234 * user processes are gone, then RST the other end.
2236 if ((so->so_state & SS_NOFDREF) &&
2237 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2240 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && "
2241 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2242 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2244 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
2245 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket "
2246 "was closed, sending RST and removing tcpcb\n",
2247 s, __func__, tcpstates[tp->t_state], tlen);
2251 TCPSTAT_INC(tcps_rcvafterclose);
2252 rstreason = BANDLIM_UNLIMITED;
2257 * If segment ends after window, drop trailing data
2258 * (and PUSH and FIN); if nothing left, just ACK.
2260 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2262 TCPSTAT_INC(tcps_rcvpackafterwin);
2263 if (todrop >= tlen) {
2264 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2266 * If window is closed can only take segments at
2267 * window edge, and have to drop data and PUSH from
2268 * incoming segments. Continue processing, but
2269 * remember to ack. Otherwise, drop segment
2272 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2273 tp->t_flags |= TF_ACKNOW;
2274 TCPSTAT_INC(tcps_rcvwinprobe);
2278 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2281 thflags &= ~(TH_PUSH|TH_FIN);
2285 * If last ACK falls within this segment's sequence numbers,
2286 * record its timestamp.
2288 * 1) That the test incorporates suggestions from the latest
2289 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2290 * 2) That updating only on newer timestamps interferes with
2291 * our earlier PAWS tests, so this check should be solely
2292 * predicated on the sequence space of this segment.
2293 * 3) That we modify the segment boundary check to be
2294 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2295 * instead of RFC1323's
2296 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2297 * This modified check allows us to overcome RFC1323's
2298 * limitations as described in Stevens TCP/IP Illustrated
2299 * Vol. 2 p.869. In such cases, we can still calculate the
2300 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2302 if ((to.to_flags & TOF_TS) != 0 &&
2303 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2304 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2305 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2306 tp->ts_recent_age = tcp_ts_getticks();
2307 tp->ts_recent = to.to_tsval;
2311 * If a SYN is in the window, then this is an
2312 * error and we send an RST and drop the connection.
2314 if (thflags & TH_SYN) {
2315 KASSERT(ti_locked == TI_WLOCKED,
2316 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2317 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2319 tp = tcp_drop(tp, ECONNRESET);
2320 rstreason = BANDLIM_UNLIMITED;
2325 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2326 * flag is on (half-synchronized state), then queue data for
2327 * later processing; else drop segment and return.
2329 if ((thflags & TH_ACK) == 0) {
2330 if (tp->t_state == TCPS_SYN_RECEIVED ||
2331 (tp->t_flags & TF_NEEDSYN))
2333 else if (tp->t_flags & TF_ACKNOW)
2342 switch (tp->t_state) {
2345 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2346 * ESTABLISHED state and continue processing.
2347 * The ACK was checked above.
2349 case TCPS_SYN_RECEIVED:
2351 TCPSTAT_INC(tcps_connects);
2353 /* Do window scaling? */
2354 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2355 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2356 tp->rcv_scale = tp->request_r_scale;
2357 tp->snd_wnd = tiwin;
2361 * SYN-RECEIVED -> ESTABLISHED
2362 * SYN-RECEIVED* -> FIN-WAIT-1
2364 tp->t_starttime = ticks;
2365 if (tp->t_flags & TF_NEEDFIN) {
2366 tp->t_state = TCPS_FIN_WAIT_1;
2367 tp->t_flags &= ~TF_NEEDFIN;
2369 tp->t_state = TCPS_ESTABLISHED;
2371 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2374 * If segment contains data or ACK, will call tcp_reass()
2375 * later; if not, do so now to pass queued data to user.
2377 if (tlen == 0 && (thflags & TH_FIN) == 0)
2378 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2380 tp->snd_wl1 = th->th_seq - 1;
2384 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2385 * ACKs. If the ack is in the range
2386 * tp->snd_una < th->th_ack <= tp->snd_max
2387 * then advance tp->snd_una to th->th_ack and drop
2388 * data from the retransmission queue. If this ACK reflects
2389 * more up to date window information we update our window information.
2391 case TCPS_ESTABLISHED:
2392 case TCPS_FIN_WAIT_1:
2393 case TCPS_FIN_WAIT_2:
2394 case TCPS_CLOSE_WAIT:
2397 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2398 TCPSTAT_INC(tcps_rcvacktoomuch);
2401 if ((tp->t_flags & TF_SACK_PERMIT) &&
2402 ((to.to_flags & TOF_SACK) ||
2403 !TAILQ_EMPTY(&tp->snd_holes)))
2404 tcp_sack_doack(tp, &to, th->th_ack);
2406 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2407 hhook_run_tcp_est_in(tp, th, &to);
2409 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2410 if (tlen == 0 && tiwin == tp->snd_wnd) {
2411 TCPSTAT_INC(tcps_rcvdupack);
2413 * If we have outstanding data (other than
2414 * a window probe), this is a completely
2415 * duplicate ack (ie, window info didn't
2416 * change), the ack is the biggest we've
2417 * seen and we've seen exactly our rexmt
2418 * threshhold of them, assume a packet
2419 * has been dropped and retransmit it.
2420 * Kludge snd_nxt & the congestion
2421 * window so we send only this one
2424 * We know we're losing at the current
2425 * window size so do congestion avoidance
2426 * (set ssthresh to half the current window
2427 * and pull our congestion window back to
2428 * the new ssthresh).
2430 * Dup acks mean that packets have left the
2431 * network (they're now cached at the receiver)
2432 * so bump cwnd by the amount in the receiver
2433 * to keep a constant cwnd packets in the
2436 * When using TCP ECN, notify the peer that
2437 * we reduced the cwnd.
2439 if (!tcp_timer_active(tp, TT_REXMT) ||
2440 th->th_ack != tp->snd_una)
2442 else if (++tp->t_dupacks > tcprexmtthresh ||
2443 IN_FASTRECOVERY(tp->t_flags)) {
2444 cc_ack_received(tp, th, CC_DUPACK);
2445 if ((tp->t_flags & TF_SACK_PERMIT) &&
2446 IN_FASTRECOVERY(tp->t_flags)) {
2450 * Compute the amount of data in flight first.
2451 * We can inject new data into the pipe iff
2452 * we have less than 1/2 the original window's
2453 * worth of data in flight.
2455 awnd = (tp->snd_nxt - tp->snd_fack) +
2456 tp->sackhint.sack_bytes_rexmit;
2457 if (awnd < tp->snd_ssthresh) {
2458 tp->snd_cwnd += tp->t_maxseg;
2459 if (tp->snd_cwnd > tp->snd_ssthresh)
2460 tp->snd_cwnd = tp->snd_ssthresh;
2463 tp->snd_cwnd += tp->t_maxseg;
2464 if ((thflags & TH_FIN) &&
2465 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2467 * If its a fin we need to process
2468 * it to avoid a race where both
2469 * sides enter FIN-WAIT and send FIN|ACK
2474 (void) tcp_output(tp);
2476 } else if (tp->t_dupacks == tcprexmtthresh) {
2477 tcp_seq onxt = tp->snd_nxt;
2480 * If we're doing sack, check to
2481 * see if we're already in sack
2482 * recovery. If we're not doing sack,
2483 * check to see if we're in newreno
2486 if (tp->t_flags & TF_SACK_PERMIT) {
2487 if (IN_FASTRECOVERY(tp->t_flags)) {
2492 if (SEQ_LEQ(th->th_ack,
2498 /* Congestion signal before ack. */
2499 cc_cong_signal(tp, th, CC_NDUPACK);
2500 cc_ack_received(tp, th, CC_DUPACK);
2501 tcp_timer_activate(tp, TT_REXMT, 0);
2503 if (tp->t_flags & TF_SACK_PERMIT) {
2505 tcps_sack_recovery_episode);
2506 tp->sack_newdata = tp->snd_nxt;
2507 tp->snd_cwnd = tp->t_maxseg;
2508 (void) tcp_output(tp);
2511 tp->snd_nxt = th->th_ack;
2512 tp->snd_cwnd = tp->t_maxseg;
2513 if ((thflags & TH_FIN) &&
2514 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2516 * If its a fin we need to process
2517 * it to avoid a race where both
2518 * sides enter FIN-WAIT and send FIN|ACK
2523 (void) tcp_output(tp);
2524 KASSERT(tp->snd_limited <= 2,
2525 ("%s: tp->snd_limited too big",
2527 tp->snd_cwnd = tp->snd_ssthresh +
2529 (tp->t_dupacks - tp->snd_limited);
2530 if (SEQ_GT(onxt, tp->snd_nxt))
2533 } else if (V_tcp_do_rfc3042) {
2534 cc_ack_received(tp, th, CC_DUPACK);
2535 u_long oldcwnd = tp->snd_cwnd;
2536 tcp_seq oldsndmax = tp->snd_max;
2540 KASSERT(tp->t_dupacks == 1 ||
2542 ("%s: dupacks not 1 or 2",
2544 if (tp->t_dupacks == 1)
2545 tp->snd_limited = 0;
2547 (tp->snd_nxt - tp->snd_una) +
2548 (tp->t_dupacks - tp->snd_limited) *
2550 if ((thflags & TH_FIN) &&
2551 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2553 * If its a fin we need to process
2554 * it to avoid a race where both
2555 * sides enter FIN-WAIT and send FIN|ACK
2561 * Only call tcp_output when there
2562 * is new data available to be sent.
2563 * Otherwise we would send pure ACKs.
2565 SOCKBUF_LOCK(&so->so_snd);
2566 avail = so->so_snd.sb_cc -
2567 (tp->snd_nxt - tp->snd_una);
2568 SOCKBUF_UNLOCK(&so->so_snd);
2570 (void) tcp_output(tp);
2571 sent = tp->snd_max - oldsndmax;
2572 if (sent > tp->t_maxseg) {
2573 KASSERT((tp->t_dupacks == 2 &&
2574 tp->snd_limited == 0) ||
2575 (sent == tp->t_maxseg + 1 &&
2576 tp->t_flags & TF_SENTFIN),
2577 ("%s: sent too much",
2579 tp->snd_limited = 2;
2580 } else if (sent > 0)
2582 tp->snd_cwnd = oldcwnd;
2590 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2591 ("%s: th_ack <= snd_una", __func__));
2594 * If the congestion window was inflated to account
2595 * for the other side's cached packets, retract it.
2597 if (IN_FASTRECOVERY(tp->t_flags)) {
2598 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2599 if (tp->t_flags & TF_SACK_PERMIT)
2600 tcp_sack_partialack(tp, th);
2602 tcp_newreno_partial_ack(tp, th);
2604 cc_post_recovery(tp, th);
2608 * If we reach this point, ACK is not a duplicate,
2609 * i.e., it ACKs something we sent.
2611 if (tp->t_flags & TF_NEEDSYN) {
2613 * T/TCP: Connection was half-synchronized, and our
2614 * SYN has been ACK'd (so connection is now fully
2615 * synchronized). Go to non-starred state,
2616 * increment snd_una for ACK of SYN, and check if
2617 * we can do window scaling.
2619 tp->t_flags &= ~TF_NEEDSYN;
2621 /* Do window scaling? */
2622 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2623 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2624 tp->rcv_scale = tp->request_r_scale;
2625 /* Send window already scaled. */
2630 INP_WLOCK_ASSERT(tp->t_inpcb);
2632 acked = BYTES_THIS_ACK(tp, th);
2633 TCPSTAT_INC(tcps_rcvackpack);
2634 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2637 * If we just performed our first retransmit, and the ACK
2638 * arrives within our recovery window, then it was a mistake
2639 * to do the retransmit in the first place. Recover our
2640 * original cwnd and ssthresh, and proceed to transmit where
2643 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2644 (int)(ticks - tp->t_badrxtwin) < 0)
2645 cc_cong_signal(tp, th, CC_RTO_ERR);
2648 * If we have a timestamp reply, update smoothed
2649 * round trip time. If no timestamp is present but
2650 * transmit timer is running and timed sequence
2651 * number was acked, update smoothed round trip time.
2652 * Since we now have an rtt measurement, cancel the
2653 * timer backoff (cf., Phil Karn's retransmit alg.).
2654 * Recompute the initial retransmit timer.
2656 * Some boxes send broken timestamp replies
2657 * during the SYN+ACK phase, ignore
2658 * timestamps of 0 or we could calculate a
2659 * huge RTT and blow up the retransmit timer.
2661 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2664 t = tcp_ts_getticks() - to.to_tsecr;
2665 if (!tp->t_rttlow || tp->t_rttlow > t)
2667 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2668 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2669 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2670 tp->t_rttlow = ticks - tp->t_rtttime;
2671 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2675 * If all outstanding data is acked, stop retransmit
2676 * timer and remember to restart (more output or persist).
2677 * If there is more data to be acked, restart retransmit
2678 * timer, using current (possibly backed-off) value.
2680 if (th->th_ack == tp->snd_max) {
2681 tcp_timer_activate(tp, TT_REXMT, 0);
2683 } else if (!tcp_timer_active(tp, TT_PERSIST))
2684 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2687 * If no data (only SYN) was ACK'd,
2688 * skip rest of ACK processing.
2694 * Let the congestion control algorithm update congestion
2695 * control related information. This typically means increasing
2696 * the congestion window.
2698 cc_ack_received(tp, th, CC_ACK);
2700 SOCKBUF_LOCK(&so->so_snd);
2701 if (acked > so->so_snd.sb_cc) {
2702 tp->snd_wnd -= so->so_snd.sb_cc;
2703 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2706 sbdrop_locked(&so->so_snd, acked);
2707 tp->snd_wnd -= acked;
2710 /* NB: sowwakeup_locked() does an implicit unlock. */
2711 sowwakeup_locked(so);
2712 /* Detect una wraparound. */
2713 if (!IN_RECOVERY(tp->t_flags) &&
2714 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2715 SEQ_LEQ(th->th_ack, tp->snd_recover))
2716 tp->snd_recover = th->th_ack - 1;
2717 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2718 if (IN_RECOVERY(tp->t_flags) &&
2719 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2720 EXIT_RECOVERY(tp->t_flags);
2722 tp->snd_una = th->th_ack;
2723 if (tp->t_flags & TF_SACK_PERMIT) {
2724 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2725 tp->snd_recover = tp->snd_una;
2727 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2728 tp->snd_nxt = tp->snd_una;
2730 switch (tp->t_state) {
2733 * In FIN_WAIT_1 STATE in addition to the processing
2734 * for the ESTABLISHED state if our FIN is now acknowledged
2735 * then enter FIN_WAIT_2.
2737 case TCPS_FIN_WAIT_1:
2738 if (ourfinisacked) {
2740 * If we can't receive any more
2741 * data, then closing user can proceed.
2742 * Starting the timer is contrary to the
2743 * specification, but if we don't get a FIN
2744 * we'll hang forever.
2747 * we should release the tp also, and use a
2750 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2751 soisdisconnected(so);
2752 tcp_timer_activate(tp, TT_2MSL,
2753 (tcp_fast_finwait2_recycle ?
2754 tcp_finwait2_timeout :
2757 tp->t_state = TCPS_FIN_WAIT_2;
2762 * In CLOSING STATE in addition to the processing for
2763 * the ESTABLISHED state if the ACK acknowledges our FIN
2764 * then enter the TIME-WAIT state, otherwise ignore
2768 if (ourfinisacked) {
2769 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2771 INP_INFO_WUNLOCK(&V_tcbinfo);
2778 * In LAST_ACK, we may still be waiting for data to drain
2779 * and/or to be acked, as well as for the ack of our FIN.
2780 * If our FIN is now acknowledged, delete the TCB,
2781 * enter the closed state and return.
2784 if (ourfinisacked) {
2785 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2794 INP_WLOCK_ASSERT(tp->t_inpcb);
2797 * Update window information.
2798 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2800 if ((thflags & TH_ACK) &&
2801 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2802 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2803 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2804 /* keep track of pure window updates */
2806 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2807 TCPSTAT_INC(tcps_rcvwinupd);
2808 tp->snd_wnd = tiwin;
2809 tp->snd_wl1 = th->th_seq;
2810 tp->snd_wl2 = th->th_ack;
2811 if (tp->snd_wnd > tp->max_sndwnd)
2812 tp->max_sndwnd = tp->snd_wnd;
2817 * Process segments with URG.
2819 if ((thflags & TH_URG) && th->th_urp &&
2820 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2822 * This is a kludge, but if we receive and accept
2823 * random urgent pointers, we'll crash in
2824 * soreceive. It's hard to imagine someone
2825 * actually wanting to send this much urgent data.
2827 SOCKBUF_LOCK(&so->so_rcv);
2828 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2829 th->th_urp = 0; /* XXX */
2830 thflags &= ~TH_URG; /* XXX */
2831 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2832 goto dodata; /* XXX */
2835 * If this segment advances the known urgent pointer,
2836 * then mark the data stream. This should not happen
2837 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2838 * a FIN has been received from the remote side.
2839 * In these states we ignore the URG.
2841 * According to RFC961 (Assigned Protocols),
2842 * the urgent pointer points to the last octet
2843 * of urgent data. We continue, however,
2844 * to consider it to indicate the first octet
2845 * of data past the urgent section as the original
2846 * spec states (in one of two places).
2848 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2849 tp->rcv_up = th->th_seq + th->th_urp;
2850 so->so_oobmark = so->so_rcv.sb_cc +
2851 (tp->rcv_up - tp->rcv_nxt) - 1;
2852 if (so->so_oobmark == 0)
2853 so->so_rcv.sb_state |= SBS_RCVATMARK;
2855 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2857 SOCKBUF_UNLOCK(&so->so_rcv);
2859 * Remove out of band data so doesn't get presented to user.
2860 * This can happen independent of advancing the URG pointer,
2861 * but if two URG's are pending at once, some out-of-band
2862 * data may creep in... ick.
2864 if (th->th_urp <= (u_long)tlen &&
2865 !(so->so_options & SO_OOBINLINE)) {
2866 /* hdr drop is delayed */
2867 tcp_pulloutofband(so, th, m, drop_hdrlen);
2871 * If no out of band data is expected,
2872 * pull receive urgent pointer along
2873 * with the receive window.
2875 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2876 tp->rcv_up = tp->rcv_nxt;
2879 INP_WLOCK_ASSERT(tp->t_inpcb);
2882 * Process the segment text, merging it into the TCP sequencing queue,
2883 * and arranging for acknowledgment of receipt if necessary.
2884 * This process logically involves adjusting tp->rcv_wnd as data
2885 * is presented to the user (this happens in tcp_usrreq.c,
2886 * case PRU_RCVD). If a FIN has already been received on this
2887 * connection then we just ignore the text.
2889 if ((tlen || (thflags & TH_FIN)) &&
2890 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2891 tcp_seq save_start = th->th_seq;
2892 m_adj(m, drop_hdrlen); /* delayed header drop */
2894 * Insert segment which includes th into TCP reassembly queue
2895 * with control block tp. Set thflags to whether reassembly now
2896 * includes a segment with FIN. This handles the common case
2897 * inline (segment is the next to be received on an established
2898 * connection, and the queue is empty), avoiding linkage into
2899 * and removal from the queue and repetition of various
2901 * Set DELACK for segments received in order, but ack
2902 * immediately when segments are out of order (so
2903 * fast retransmit can work).
2905 if (th->th_seq == tp->rcv_nxt &&
2906 LIST_EMPTY(&tp->t_segq) &&
2907 TCPS_HAVEESTABLISHED(tp->t_state)) {
2909 tp->t_flags |= TF_DELACK;
2911 tp->t_flags |= TF_ACKNOW;
2912 tp->rcv_nxt += tlen;
2913 thflags = th->th_flags & TH_FIN;
2914 TCPSTAT_INC(tcps_rcvpack);
2915 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2917 SOCKBUF_LOCK(&so->so_rcv);
2918 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2921 sbappendstream_locked(&so->so_rcv, m);
2922 /* NB: sorwakeup_locked() does an implicit unlock. */
2923 sorwakeup_locked(so);
2926 * XXX: Due to the header drop above "th" is
2927 * theoretically invalid by now. Fortunately
2928 * m_adj() doesn't actually frees any mbufs
2929 * when trimming from the head.
2931 thflags = tcp_reass(tp, th, &tlen, m);
2932 tp->t_flags |= TF_ACKNOW;
2934 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2935 tcp_update_sack_list(tp, save_start, save_start + tlen);
2938 * Note the amount of data that peer has sent into
2939 * our window, in order to estimate the sender's
2943 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2944 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2946 len = so->so_rcv.sb_hiwat;
2954 * If FIN is received ACK the FIN and let the user know
2955 * that the connection is closing.
2957 if (thflags & TH_FIN) {
2958 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2961 * If connection is half-synchronized
2962 * (ie NEEDSYN flag on) then delay ACK,
2963 * so it may be piggybacked when SYN is sent.
2964 * Otherwise, since we received a FIN then no
2965 * more input can be expected, send ACK now.
2967 if (tp->t_flags & TF_NEEDSYN)
2968 tp->t_flags |= TF_DELACK;
2970 tp->t_flags |= TF_ACKNOW;
2973 switch (tp->t_state) {
2976 * In SYN_RECEIVED and ESTABLISHED STATES
2977 * enter the CLOSE_WAIT state.
2979 case TCPS_SYN_RECEIVED:
2980 tp->t_starttime = ticks;
2982 case TCPS_ESTABLISHED:
2983 tp->t_state = TCPS_CLOSE_WAIT;
2987 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2988 * enter the CLOSING state.
2990 case TCPS_FIN_WAIT_1:
2991 tp->t_state = TCPS_CLOSING;
2995 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2996 * starting the time-wait timer, turning off the other
2999 case TCPS_FIN_WAIT_2:
3000 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
3001 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata "
3002 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
3006 INP_INFO_WUNLOCK(&V_tcbinfo);
3010 if (ti_locked == TI_WLOCKED)
3011 INP_INFO_WUNLOCK(&V_tcbinfo);
3012 ti_locked = TI_UNLOCKED;
3015 if (so->so_options & SO_DEBUG)
3016 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3021 * Return any desired output.
3023 if (needoutput || (tp->t_flags & TF_ACKNOW))
3024 (void) tcp_output(tp);
3027 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3028 __func__, ti_locked));
3029 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3030 INP_WLOCK_ASSERT(tp->t_inpcb);
3032 if (tp->t_flags & TF_DELACK) {
3033 tp->t_flags &= ~TF_DELACK;
3034 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3036 INP_WUNLOCK(tp->t_inpcb);
3041 * Generate an ACK dropping incoming segment if it occupies
3042 * sequence space, where the ACK reflects our state.
3044 * We can now skip the test for the RST flag since all
3045 * paths to this code happen after packets containing
3046 * RST have been dropped.
3048 * In the SYN-RECEIVED state, don't send an ACK unless the
3049 * segment we received passes the SYN-RECEIVED ACK test.
3050 * If it fails send a RST. This breaks the loop in the
3051 * "LAND" DoS attack, and also prevents an ACK storm
3052 * between two listening ports that have been sent forged
3053 * SYN segments, each with the source address of the other.
3055 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3056 (SEQ_GT(tp->snd_una, th->th_ack) ||
3057 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3058 rstreason = BANDLIM_RST_OPENPORT;
3062 if (so->so_options & SO_DEBUG)
3063 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3066 if (ti_locked == TI_WLOCKED)
3067 INP_INFO_WUNLOCK(&V_tcbinfo);
3068 ti_locked = TI_UNLOCKED;
3070 tp->t_flags |= TF_ACKNOW;
3071 (void) tcp_output(tp);
3072 INP_WUNLOCK(tp->t_inpcb);
3077 if (ti_locked == TI_WLOCKED)
3078 INP_INFO_WUNLOCK(&V_tcbinfo);
3079 ti_locked = TI_UNLOCKED;
3082 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3083 INP_WUNLOCK(tp->t_inpcb);
3085 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3089 if (ti_locked == TI_WLOCKED) {
3090 INP_INFO_WUNLOCK(&V_tcbinfo);
3091 ti_locked = TI_UNLOCKED;
3095 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3099 * Drop space held by incoming segment and return.
3102 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3103 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3107 INP_WUNLOCK(tp->t_inpcb);
3112 * Issue RST and make ACK acceptable to originator of segment.
3113 * The mbuf must still include the original packet header.
3117 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3118 int tlen, int rstreason)
3124 struct ip6_hdr *ip6;
3128 INP_WLOCK_ASSERT(tp->t_inpcb);
3131 /* Don't bother if destination was broadcast/multicast. */
3132 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3135 if (mtod(m, struct ip *)->ip_v == 6) {
3136 ip6 = mtod(m, struct ip6_hdr *);
3137 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3138 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3140 /* IPv6 anycast check is done at tcp6_input() */
3143 #if defined(INET) && defined(INET6)
3148 ip = mtod(m, struct ip *);
3149 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3150 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3151 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3152 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3157 /* Perform bandwidth limiting. */
3158 if (badport_bandlim(rstreason) < 0)
3161 /* tcp_respond consumes the mbuf chain. */
3162 if (th->th_flags & TH_ACK) {
3163 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3164 th->th_ack, TH_RST);
3166 if (th->th_flags & TH_SYN)
3168 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3169 (tcp_seq)0, TH_RST|TH_ACK);
3177 * Parse TCP options and place in tcpopt.
3180 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3185 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3187 if (opt == TCPOPT_EOL)
3189 if (opt == TCPOPT_NOP)
3195 if (optlen < 2 || optlen > cnt)
3200 if (optlen != TCPOLEN_MAXSEG)
3202 if (!(flags & TO_SYN))
3204 to->to_flags |= TOF_MSS;
3205 bcopy((char *)cp + 2,
3206 (char *)&to->to_mss, sizeof(to->to_mss));
3207 to->to_mss = ntohs(to->to_mss);
3210 if (optlen != TCPOLEN_WINDOW)
3212 if (!(flags & TO_SYN))
3214 to->to_flags |= TOF_SCALE;
3215 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3217 case TCPOPT_TIMESTAMP:
3218 if (optlen != TCPOLEN_TIMESTAMP)
3220 to->to_flags |= TOF_TS;
3221 bcopy((char *)cp + 2,
3222 (char *)&to->to_tsval, sizeof(to->to_tsval));
3223 to->to_tsval = ntohl(to->to_tsval);
3224 bcopy((char *)cp + 6,
3225 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3226 to->to_tsecr = ntohl(to->to_tsecr);
3228 #ifdef TCP_SIGNATURE
3230 * XXX In order to reply to a host which has set the
3231 * TCP_SIGNATURE option in its initial SYN, we have to
3232 * record the fact that the option was observed here
3233 * for the syncache code to perform the correct response.
3235 case TCPOPT_SIGNATURE:
3236 if (optlen != TCPOLEN_SIGNATURE)
3238 to->to_flags |= TOF_SIGNATURE;
3239 to->to_signature = cp + 2;
3242 case TCPOPT_SACK_PERMITTED:
3243 if (optlen != TCPOLEN_SACK_PERMITTED)
3245 if (!(flags & TO_SYN))
3249 to->to_flags |= TOF_SACKPERM;
3252 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3256 to->to_flags |= TOF_SACK;
3257 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3258 to->to_sacks = cp + 2;
3259 TCPSTAT_INC(tcps_sack_rcv_blocks);
3268 * Pull out of band byte out of a segment so
3269 * it doesn't appear in the user's data queue.
3270 * It is still reflected in the segment length for
3271 * sequencing purposes.
3274 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3277 int cnt = off + th->th_urp - 1;
3280 if (m->m_len > cnt) {
3281 char *cp = mtod(m, caddr_t) + cnt;
3282 struct tcpcb *tp = sototcpcb(so);
3284 INP_WLOCK_ASSERT(tp->t_inpcb);
3287 tp->t_oobflags |= TCPOOB_HAVEDATA;
3288 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3290 if (m->m_flags & M_PKTHDR)
3299 panic("tcp_pulloutofband");
3303 * Collect new round-trip time estimate
3304 * and update averages and current timeout.
3307 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3311 INP_WLOCK_ASSERT(tp->t_inpcb);
3313 TCPSTAT_INC(tcps_rttupdated);
3315 if (tp->t_srtt != 0) {
3317 * srtt is stored as fixed point with 5 bits after the
3318 * binary point (i.e., scaled by 8). The following magic
3319 * is equivalent to the smoothing algorithm in rfc793 with
3320 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3321 * point). Adjust rtt to origin 0.
3323 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3324 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3326 if ((tp->t_srtt += delta) <= 0)
3330 * We accumulate a smoothed rtt variance (actually, a
3331 * smoothed mean difference), then set the retransmit
3332 * timer to smoothed rtt + 4 times the smoothed variance.
3333 * rttvar is stored as fixed point with 4 bits after the
3334 * binary point (scaled by 16). The following is
3335 * equivalent to rfc793 smoothing with an alpha of .75
3336 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3337 * rfc793's wired-in beta.
3341 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3342 if ((tp->t_rttvar += delta) <= 0)
3344 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3345 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3348 * No rtt measurement yet - use the unsmoothed rtt.
3349 * Set the variance to half the rtt (so our first
3350 * retransmit happens at 3*rtt).
3352 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3353 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3354 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3360 * the retransmit should happen at rtt + 4 * rttvar.
3361 * Because of the way we do the smoothing, srtt and rttvar
3362 * will each average +1/2 tick of bias. When we compute
3363 * the retransmit timer, we want 1/2 tick of rounding and
3364 * 1 extra tick because of +-1/2 tick uncertainty in the
3365 * firing of the timer. The bias will give us exactly the
3366 * 1.5 tick we need. But, because the bias is
3367 * statistical, we have to test that we don't drop below
3368 * the minimum feasible timer (which is 2 ticks).
3370 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3371 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3374 * We received an ack for a packet that wasn't retransmitted;
3375 * it is probably safe to discard any error indications we've
3376 * received recently. This isn't quite right, but close enough
3377 * for now (a route might have failed after we sent a segment,
3378 * and the return path might not be symmetrical).
3380 tp->t_softerror = 0;
3384 * Determine a reasonable value for maxseg size.
3385 * If the route is known, check route for mtu.
3386 * If none, use an mss that can be handled on the outgoing interface
3387 * without forcing IP to fragment. If no route is found, route has no mtu,
3388 * or the destination isn't local, use a default, hopefully conservative
3389 * size (usually 512 or the default IP max size, but no more than the mtu
3390 * of the interface), as we can't discover anything about intervening
3391 * gateways or networks. We also initialize the congestion/slow start
3392 * window to be a single segment if the destination isn't local.
3393 * While looking at the routing entry, we also initialize other path-dependent
3394 * parameters from pre-set or cached values in the routing entry.
3396 * Also take into account the space needed for options that we
3397 * send regularly. Make maxseg shorter by that amount to assure
3398 * that we can send maxseg amount of data even when the options
3399 * are present. Store the upper limit of the length of options plus
3402 * NOTE that this routine is only called when we process an incoming
3403 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3404 * settings are handled in tcp_mssopt().
3407 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3408 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3412 struct inpcb *inp = tp->t_inpcb;
3413 struct hc_metrics_lite metrics;
3416 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3417 size_t min_protoh = isipv6 ?
3418 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3419 sizeof (struct tcpiphdr);
3421 const size_t min_protoh = sizeof(struct tcpiphdr);
3424 INP_WLOCK_ASSERT(tp->t_inpcb);
3426 if (mtuoffer != -1) {
3427 KASSERT(offer == -1, ("%s: conflict", __func__));
3428 offer = mtuoffer - min_protoh;
3435 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3436 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3439 #if defined(INET) && defined(INET6)
3444 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3445 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3450 * No route to sender, stay with default mss and return.
3454 * In case we return early we need to initialize metrics
3455 * to a defined state as tcp_hc_get() would do for us
3456 * if there was no cache hit.
3458 if (metricptr != NULL)
3459 bzero(metricptr, sizeof(struct hc_metrics_lite));
3463 /* What have we got? */
3467 * Offer == 0 means that there was no MSS on the SYN
3468 * segment, in this case we use tcp_mssdflt as
3469 * already assigned to t_maxopd above.
3471 offer = tp->t_maxopd;
3476 * Offer == -1 means that we didn't receive SYN yet.
3482 * Prevent DoS attack with too small MSS. Round up
3483 * to at least minmss.
3485 offer = max(offer, V_tcp_minmss);
3489 * rmx information is now retrieved from tcp_hostcache.
3491 tcp_hc_get(&inp->inp_inc, &metrics);
3492 if (metricptr != NULL)
3493 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3496 * If there's a discovered mtu int tcp hostcache, use it
3497 * else, use the link mtu.
3499 if (metrics.rmx_mtu)
3500 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3504 mss = maxmtu - min_protoh;
3505 if (!V_path_mtu_discovery &&
3506 !in6_localaddr(&inp->in6p_faddr))
3507 mss = min(mss, V_tcp_v6mssdflt);
3510 #if defined(INET) && defined(INET6)
3515 mss = maxmtu - min_protoh;
3516 if (!V_path_mtu_discovery &&
3517 !in_localaddr(inp->inp_faddr))
3518 mss = min(mss, V_tcp_mssdflt);
3522 * XXX - The above conditional (mss = maxmtu - min_protoh)
3523 * probably violates the TCP spec.
3524 * The problem is that, since we don't know the
3525 * other end's MSS, we are supposed to use a conservative
3526 * default. But, if we do that, then MTU discovery will
3527 * never actually take place, because the conservative
3528 * default is much less than the MTUs typically seen
3529 * on the Internet today. For the moment, we'll sweep
3530 * this under the carpet.
3532 * The conservative default might not actually be a problem
3533 * if the only case this occurs is when sending an initial
3534 * SYN with options and data to a host we've never talked
3535 * to before. Then, they will reply with an MSS value which
3536 * will get recorded and the new parameters should get
3537 * recomputed. For Further Study.
3540 mss = min(mss, offer);
3543 * Sanity check: make sure that maxopd will be large
3544 * enough to allow some data on segments even if the
3545 * all the option space is used (40bytes). Otherwise
3546 * funny things may happen in tcp_output.
3551 * maxopd stores the maximum length of data AND options
3552 * in a segment; maxseg is the amount of data in a normal
3553 * segment. We need to store this value (maxopd) apart
3554 * from maxseg, because now every segment carries options
3555 * and thus we normally have somewhat less data in segments.
3560 * origoffer==-1 indicates that no segments were received yet.
3561 * In this case we just guess.
3563 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3565 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3566 mss -= TCPOLEN_TSTAMP_APPA;
3572 tcp_mss(struct tcpcb *tp, int offer)
3578 struct hc_metrics_lite metrics;
3579 struct tcp_ifcap cap;
3581 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3583 bzero(&cap, sizeof(cap));
3584 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3590 * If there's a pipesize, change the socket buffer to that size,
3591 * don't change if sb_hiwat is different than default (then it
3592 * has been changed on purpose with setsockopt).
3593 * Make the socket buffers an integral number of mss units;
3594 * if the mss is larger than the socket buffer, decrease the mss.
3596 so = inp->inp_socket;
3597 SOCKBUF_LOCK(&so->so_snd);
3598 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3599 bufsize = metrics.rmx_sendpipe;
3601 bufsize = so->so_snd.sb_hiwat;
3605 bufsize = roundup(bufsize, mss);
3606 if (bufsize > sb_max)
3608 if (bufsize > so->so_snd.sb_hiwat)
3609 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3611 SOCKBUF_UNLOCK(&so->so_snd);
3614 SOCKBUF_LOCK(&so->so_rcv);
3615 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3616 bufsize = metrics.rmx_recvpipe;
3618 bufsize = so->so_rcv.sb_hiwat;
3619 if (bufsize > mss) {
3620 bufsize = roundup(bufsize, mss);
3621 if (bufsize > sb_max)
3623 if (bufsize > so->so_rcv.sb_hiwat)
3624 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3626 SOCKBUF_UNLOCK(&so->so_rcv);
3628 /* Check the interface for TSO capabilities. */
3629 if (cap.ifcap & CSUM_TSO) {
3630 tp->t_flags |= TF_TSO;
3631 tp->t_tsomax = cap.tsomax;
3636 * Determine the MSS option to send on an outgoing SYN.
3639 tcp_mssopt(struct in_conninfo *inc)
3646 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3649 if (inc->inc_flags & INC_ISIPV6) {
3650 mss = V_tcp_v6mssdflt;
3651 maxmtu = tcp_maxmtu6(inc, NULL);
3652 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3655 #if defined(INET) && defined(INET6)
3660 mss = V_tcp_mssdflt;
3661 maxmtu = tcp_maxmtu(inc, NULL);
3662 min_protoh = sizeof(struct tcpiphdr);
3665 #if defined(INET6) || defined(INET)
3666 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3669 if (maxmtu && thcmtu)
3670 mss = min(maxmtu, thcmtu) - min_protoh;
3671 else if (maxmtu || thcmtu)
3672 mss = max(maxmtu, thcmtu) - min_protoh;
3679 * On a partial ack arrives, force the retransmission of the
3680 * next unacknowledged segment. Do not clear tp->t_dupacks.
3681 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3685 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3687 tcp_seq onxt = tp->snd_nxt;
3688 u_long ocwnd = tp->snd_cwnd;
3690 INP_WLOCK_ASSERT(tp->t_inpcb);
3692 tcp_timer_activate(tp, TT_REXMT, 0);
3694 tp->snd_nxt = th->th_ack;
3696 * Set snd_cwnd to one segment beyond acknowledged offset.
3697 * (tp->snd_una has not yet been updated when this function is called.)
3699 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3700 tp->t_flags |= TF_ACKNOW;
3701 (void) tcp_output(tp);
3702 tp->snd_cwnd = ocwnd;
3703 if (SEQ_GT(onxt, tp->snd_nxt))
3706 * Partial window deflation. Relies on fact that tp->snd_una
3709 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3710 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3713 tp->snd_cwnd += tp->t_maxseg;