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) = 1;
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;
556 struct inpcb *inp = NULL;
557 struct tcpcb *tp = NULL;
558 struct socket *so = NULL;
567 int rstreason = 0; /* For badport_bandlim accounting purposes */
569 uint8_t sig_checked = 0;
572 struct m_tag *fwd_tag = NULL;
574 struct ip6_hdr *ip6 = NULL;
577 const void *ip6 = NULL;
579 struct tcpopt to; /* options in this segment */
580 char *s = NULL; /* address and port logging */
582 #define TI_UNLOCKED 1
587 * The size of tcp_saveipgen must be the size of the max ip header,
590 u_char tcp_saveipgen[IP6_HDR_LEN];
591 struct tcphdr tcp_savetcp;
596 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
600 TCPSTAT_INC(tcps_rcvtotal);
604 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
606 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
607 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
609 TCPSTAT_INC(tcps_rcvshort);
614 ip6 = mtod(m, struct ip6_hdr *);
615 th = (struct tcphdr *)((caddr_t)ip6 + off0);
616 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
617 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
618 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
619 th->th_sum = m->m_pkthdr.csum_data;
621 th->th_sum = in6_cksum_pseudo(ip6, tlen,
622 IPPROTO_TCP, m->m_pkthdr.csum_data);
623 th->th_sum ^= 0xffff;
625 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
627 TCPSTAT_INC(tcps_rcvbadsum);
632 * Be proactive about unspecified IPv6 address in source.
633 * As we use all-zero to indicate unbounded/unconnected pcb,
634 * unspecified IPv6 address can be used to confuse us.
636 * Note that packets with unspecified IPv6 destination is
637 * already dropped in ip6_input.
639 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
645 #if defined(INET) && defined(INET6)
651 * Get IP and TCP header together in first mbuf.
652 * Note: IP leaves IP header in first mbuf.
654 if (off0 > sizeof (struct ip)) {
656 off0 = sizeof(struct ip);
658 if (m->m_len < sizeof (struct tcpiphdr)) {
659 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
661 TCPSTAT_INC(tcps_rcvshort);
665 ip = mtod(m, struct ip *);
666 th = (struct tcphdr *)((caddr_t)ip + off0);
667 tlen = ntohs(ip->ip_len) - off0;
669 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
670 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
671 th->th_sum = m->m_pkthdr.csum_data;
673 th->th_sum = in_pseudo(ip->ip_src.s_addr,
675 htonl(m->m_pkthdr.csum_data + tlen +
677 th->th_sum ^= 0xffff;
679 struct ipovly *ipov = (struct ipovly *)ip;
682 * Checksum extended TCP header and data.
685 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
686 ipov->ih_len = htons(tlen);
687 th->th_sum = in_cksum(m, len);
690 TCPSTAT_INC(tcps_rcvbadsum);
693 /* Re-initialization for later version check */
694 ip->ip_v = IPVERSION;
700 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
702 #if defined(INET) && defined(INET6)
710 * Check that TCP offset makes sense,
711 * pull out TCP options and adjust length. XXX
713 off = th->th_off << 2;
714 if (off < sizeof (struct tcphdr) || off > tlen) {
715 TCPSTAT_INC(tcps_rcvbadoff);
718 tlen -= off; /* tlen is used instead of ti->ti_len */
719 if (off > sizeof (struct tcphdr)) {
722 IP6_EXTHDR_CHECK(m, off0, off, );
723 ip6 = mtod(m, struct ip6_hdr *);
724 th = (struct tcphdr *)((caddr_t)ip6 + off0);
727 #if defined(INET) && defined(INET6)
732 if (m->m_len < sizeof(struct ip) + off) {
733 if ((m = m_pullup(m, sizeof (struct ip) + off))
735 TCPSTAT_INC(tcps_rcvshort);
738 ip = mtod(m, struct ip *);
739 th = (struct tcphdr *)((caddr_t)ip + off0);
743 optlen = off - sizeof (struct tcphdr);
744 optp = (u_char *)(th + 1);
746 thflags = th->th_flags;
749 * Convert TCP protocol specific fields to host format.
751 tcp_fields_to_host(th);
754 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
756 drop_hdrlen = off0 + off;
759 * Locate pcb for segment; if we're likely to add or remove a
760 * connection then first acquire pcbinfo lock. There are two cases
761 * where we might discover later we need a write lock despite the
762 * flags: ACKs moving a connection out of the syncache, and ACKs for
763 * a connection in TIMEWAIT.
765 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) {
766 INP_INFO_WLOCK(&V_tcbinfo);
767 ti_locked = TI_WLOCKED;
769 ti_locked = TI_UNLOCKED;
773 if (ti_locked == TI_WLOCKED) {
774 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
776 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
781 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
785 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
787 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
790 #if defined(INET) && !defined(INET6)
791 (m->m_flags & M_IP_NEXTHOP)
794 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
797 if (isipv6 && fwd_tag != NULL) {
798 struct sockaddr_in6 *next_hop6;
800 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
802 * Transparently forwarded. Pretend to be the destination.
803 * Already got one like this?
805 inp = in6_pcblookup_mbuf(&V_tcbinfo,
806 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
807 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
810 * It's new. Try to find the ambushing socket.
811 * Because we've rewritten the destination address,
812 * any hardware-generated hash is ignored.
814 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
815 th->th_sport, &next_hop6->sin6_addr,
816 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
817 th->th_dport, INPLOOKUP_WILDCARD |
818 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
820 /* Remove the tag from the packet. We don't need it anymore. */
821 m_tag_delete(m, fwd_tag);
822 m->m_flags &= ~M_IP6_NEXTHOP;
825 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
826 th->th_sport, &ip6->ip6_dst, th->th_dport,
827 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
828 m->m_pkthdr.rcvif, m);
831 #if defined(INET6) && defined(INET)
835 if (fwd_tag != NULL) {
836 struct sockaddr_in *next_hop;
838 next_hop = (struct sockaddr_in *)(fwd_tag+1);
840 * Transparently forwarded. Pretend to be the destination.
841 * already got one like this?
843 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
844 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
845 m->m_pkthdr.rcvif, m);
848 * It's new. Try to find the ambushing socket.
849 * Because we've rewritten the destination address,
850 * any hardware-generated hash is ignored.
852 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
853 th->th_sport, next_hop->sin_addr,
854 next_hop->sin_port ? ntohs(next_hop->sin_port) :
855 th->th_dport, INPLOOKUP_WILDCARD |
856 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
858 /* Remove the tag from the packet. We don't need it anymore. */
859 m_tag_delete(m, fwd_tag);
860 m->m_flags &= ~M_IP_NEXTHOP;
863 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
864 th->th_sport, ip->ip_dst, th->th_dport,
865 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
866 m->m_pkthdr.rcvif, m);
870 * If the INPCB does not exist then all data in the incoming
871 * segment is discarded and an appropriate RST is sent back.
872 * XXX MRT Send RST using which routing table?
876 * Log communication attempts to ports that are not
879 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
880 tcp_log_in_vain == 2) {
881 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
882 log(LOG_INFO, "%s; %s: Connection attempt "
883 "to closed port\n", s, __func__);
886 * When blackholing do not respond with a RST but
887 * completely ignore the segment and drop it.
889 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
893 rstreason = BANDLIM_RST_CLOSEDPORT;
896 INP_WLOCK_ASSERT(inp);
897 if (!(inp->inp_flags & INP_HW_FLOWID)
898 && (m->m_flags & M_FLOWID)
899 && ((inp->inp_socket == NULL)
900 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) {
901 inp->inp_flags |= INP_HW_FLOWID;
902 inp->inp_flags &= ~INP_SW_FLOWID;
903 inp->inp_flowid = m->m_pkthdr.flowid;
907 if (isipv6 && ipsec6_in_reject(m, inp)) {
908 V_ipsec6stat.in_polvio++;
912 if (ipsec4_in_reject(m, inp) != 0) {
913 V_ipsec4stat.in_polvio++;
919 * Check the minimum TTL for socket.
921 if (inp->inp_ip_minttl != 0) {
923 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
927 if (inp->inp_ip_minttl > ip->ip_ttl)
932 * A previous connection in TIMEWAIT state is supposed to catch stray
933 * or duplicate segments arriving late. If this segment was a
934 * legitimate new connection attempt, the old INPCB gets removed and
935 * we can try again to find a listening socket.
937 * At this point, due to earlier optimism, we may hold only an inpcb
938 * lock, and not the inpcbinfo write lock. If so, we need to try to
939 * acquire it, or if that fails, acquire a reference on the inpcb,
940 * drop all locks, acquire a global write lock, and then re-acquire
941 * the inpcb lock. We may at that point discover that another thread
942 * has tried to free the inpcb, in which case we need to loop back
943 * and try to find a new inpcb to deliver to.
945 * XXXRW: It may be time to rethink timewait locking.
948 if (inp->inp_flags & INP_TIMEWAIT) {
949 if (ti_locked == TI_UNLOCKED) {
950 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
953 INP_INFO_WLOCK(&V_tcbinfo);
954 ti_locked = TI_WLOCKED;
956 if (in_pcbrele_wlocked(inp)) {
961 ti_locked = TI_WLOCKED;
963 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
965 if (thflags & TH_SYN)
966 tcp_dooptions(&to, optp, optlen, TO_SYN);
968 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
970 if (tcp_twcheck(inp, &to, th, m, tlen))
972 INP_INFO_WUNLOCK(&V_tcbinfo);
976 * The TCPCB may no longer exist if the connection is winding
977 * down or it is in the CLOSED state. Either way we drop the
978 * segment and send an appropriate response.
981 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
982 rstreason = BANDLIM_RST_CLOSEDPORT;
987 if (tp->t_flags & TF_TOE) {
988 tcp_offload_input(tp, m);
989 m = NULL; /* consumed by the TOE driver */
995 * We've identified a valid inpcb, but it could be that we need an
996 * inpcbinfo write lock but don't hold it. In this case, attempt to
997 * acquire using the same strategy as the TIMEWAIT case above. If we
998 * relock, we have to jump back to 'relocked' as the connection might
999 * now be in TIMEWAIT.
1002 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0)
1003 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1005 if (tp->t_state != TCPS_ESTABLISHED) {
1006 if (ti_locked == TI_UNLOCKED) {
1007 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
1010 INP_INFO_WLOCK(&V_tcbinfo);
1011 ti_locked = TI_WLOCKED;
1013 if (in_pcbrele_wlocked(inp)) {
1019 ti_locked = TI_WLOCKED;
1021 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1025 INP_WLOCK_ASSERT(inp);
1026 if (mac_inpcb_check_deliver(inp, m))
1029 so = inp->inp_socket;
1030 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1032 if (so->so_options & SO_DEBUG) {
1033 ostate = tp->t_state;
1036 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1039 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1042 #endif /* TCPDEBUG */
1044 * When the socket is accepting connections (the INPCB is in LISTEN
1045 * state) we look into the SYN cache if this is a new connection
1046 * attempt or the completion of a previous one. Because listen
1047 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be
1048 * held in this case.
1050 if (so->so_options & SO_ACCEPTCONN) {
1051 struct in_conninfo inc;
1053 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
1054 "tp not listening", __func__));
1055 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1057 bzero(&inc, sizeof(inc));
1060 inc.inc_flags |= INC_ISIPV6;
1061 inc.inc6_faddr = ip6->ip6_src;
1062 inc.inc6_laddr = ip6->ip6_dst;
1066 inc.inc_faddr = ip->ip_src;
1067 inc.inc_laddr = ip->ip_dst;
1069 inc.inc_fport = th->th_sport;
1070 inc.inc_lport = th->th_dport;
1071 inc.inc_fibnum = so->so_fibnum;
1074 * Check for an existing connection attempt in syncache if
1075 * the flag is only ACK. A successful lookup creates a new
1076 * socket appended to the listen queue in SYN_RECEIVED state.
1078 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1080 * Parse the TCP options here because
1081 * syncookies need access to the reflected
1084 tcp_dooptions(&to, optp, optlen, 0);
1086 * NB: syncache_expand() doesn't unlock
1087 * inp and tcpinfo locks.
1089 if (!syncache_expand(&inc, &to, th, &so, m)) {
1091 * No syncache entry or ACK was not
1092 * for our SYN/ACK. Send a RST.
1093 * NB: syncache did its own logging
1094 * of the failure cause.
1096 rstreason = BANDLIM_RST_OPENPORT;
1101 * We completed the 3-way handshake
1102 * but could not allocate a socket
1103 * either due to memory shortage,
1104 * listen queue length limits or
1105 * global socket limits. Send RST
1106 * or wait and have the remote end
1107 * retransmit the ACK for another
1110 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1111 log(LOG_DEBUG, "%s; %s: Listen socket: "
1112 "Socket allocation failed due to "
1113 "limits or memory shortage, %s\n",
1115 V_tcp_sc_rst_sock_fail ?
1116 "sending RST" : "try again");
1117 if (V_tcp_sc_rst_sock_fail) {
1118 rstreason = BANDLIM_UNLIMITED;
1124 * Socket is created in state SYN_RECEIVED.
1125 * Unlock the listen socket, lock the newly
1126 * created socket and update the tp variable.
1128 INP_WUNLOCK(inp); /* listen socket */
1129 inp = sotoinpcb(so);
1130 INP_WLOCK(inp); /* new connection */
1131 tp = intotcpcb(inp);
1132 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1133 ("%s: ", __func__));
1134 #ifdef TCP_SIGNATURE
1135 if (sig_checked == 0) {
1136 tcp_dooptions(&to, optp, optlen,
1137 (thflags & TH_SYN) ? TO_SYN : 0);
1138 if (!tcp_signature_verify_input(m, off0, tlen,
1139 optlen, &to, th, tp->t_flags)) {
1142 * In SYN_SENT state if it receives an
1143 * RST, it is allowed for further
1146 if ((thflags & TH_RST) == 0 ||
1147 (tp->t_state == TCPS_SYN_SENT) == 0)
1155 * Process the segment and the data it
1156 * contains. tcp_do_segment() consumes
1157 * the mbuf chain and unlocks the inpcb.
1159 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1161 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1165 * Segment flag validation for new connection attempts:
1167 * Our (SYN|ACK) response was rejected.
1168 * Check with syncache and remove entry to prevent
1171 * NB: syncache_chkrst does its own logging of failure
1174 if (thflags & TH_RST) {
1175 syncache_chkrst(&inc, th);
1179 * We can't do anything without SYN.
1181 if ((thflags & TH_SYN) == 0) {
1182 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1183 log(LOG_DEBUG, "%s; %s: Listen socket: "
1184 "SYN is missing, segment ignored\n",
1186 TCPSTAT_INC(tcps_badsyn);
1190 * (SYN|ACK) is bogus on a listen socket.
1192 if (thflags & TH_ACK) {
1193 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1194 log(LOG_DEBUG, "%s; %s: Listen socket: "
1195 "SYN|ACK invalid, segment rejected\n",
1197 syncache_badack(&inc); /* XXX: Not needed! */
1198 TCPSTAT_INC(tcps_badsyn);
1199 rstreason = BANDLIM_RST_OPENPORT;
1203 * If the drop_synfin option is enabled, drop all
1204 * segments with both the SYN and FIN bits set.
1205 * This prevents e.g. nmap from identifying the
1207 * XXX: Poor reasoning. nmap has other methods
1208 * and is constantly refining its stack detection
1210 * XXX: This is a violation of the TCP specification
1211 * and was used by RFC1644.
1213 if ((thflags & TH_FIN) && V_drop_synfin) {
1214 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1215 log(LOG_DEBUG, "%s; %s: Listen socket: "
1216 "SYN|FIN segment ignored (based on "
1217 "sysctl setting)\n", s, __func__);
1218 TCPSTAT_INC(tcps_badsyn);
1222 * Segment's flags are (SYN) or (SYN|FIN).
1224 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1225 * as they do not affect the state of the TCP FSM.
1226 * The data pointed to by TH_URG and th_urp is ignored.
1228 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1229 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1230 KASSERT(thflags & (TH_SYN),
1231 ("%s: Listen socket: TH_SYN not set", __func__));
1234 * If deprecated address is forbidden,
1235 * we do not accept SYN to deprecated interface
1236 * address to prevent any new inbound connection from
1237 * getting established.
1238 * When we do not accept SYN, we send a TCP RST,
1239 * with deprecated source address (instead of dropping
1240 * it). We compromise it as it is much better for peer
1241 * to send a RST, and RST will be the final packet
1244 * If we do not forbid deprecated addresses, we accept
1245 * the SYN packet. RFC2462 does not suggest dropping
1247 * If we decipher RFC2462 5.5.4, it says like this:
1248 * 1. use of deprecated addr with existing
1249 * communication is okay - "SHOULD continue to be
1251 * 2. use of it with new communication:
1252 * (2a) "SHOULD NOT be used if alternate address
1253 * with sufficient scope is available"
1254 * (2b) nothing mentioned otherwise.
1255 * Here we fall into (2b) case as we have no choice in
1256 * our source address selection - we must obey the peer.
1258 * The wording in RFC2462 is confusing, and there are
1259 * multiple description text for deprecated address
1260 * handling - worse, they are not exactly the same.
1261 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1263 if (isipv6 && !V_ip6_use_deprecated) {
1264 struct in6_ifaddr *ia6;
1266 ia6 = ip6_getdstifaddr(m);
1268 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1269 ifa_free(&ia6->ia_ifa);
1270 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1271 log(LOG_DEBUG, "%s; %s: Listen socket: "
1272 "Connection attempt to deprecated "
1273 "IPv6 address rejected\n",
1275 rstreason = BANDLIM_RST_OPENPORT;
1279 ifa_free(&ia6->ia_ifa);
1283 * Basic sanity checks on incoming SYN requests:
1284 * Don't respond if the destination is a link layer
1285 * broadcast according to RFC1122 4.2.3.10, p. 104.
1286 * If it is from this socket it must be forged.
1287 * Don't respond if the source or destination is a
1288 * global or subnet broad- or multicast address.
1289 * Note that it is quite possible to receive unicast
1290 * link-layer packets with a broadcast IP address. Use
1291 * in_broadcast() to find them.
1293 if (m->m_flags & (M_BCAST|M_MCAST)) {
1294 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1295 log(LOG_DEBUG, "%s; %s: Listen socket: "
1296 "Connection attempt from broad- or multicast "
1297 "link layer address ignored\n", s, __func__);
1302 if (th->th_dport == th->th_sport &&
1303 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1304 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1305 log(LOG_DEBUG, "%s; %s: Listen socket: "
1306 "Connection attempt to/from self "
1307 "ignored\n", s, __func__);
1310 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1311 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1312 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1313 log(LOG_DEBUG, "%s; %s: Listen socket: "
1314 "Connection attempt from/to multicast "
1315 "address ignored\n", s, __func__);
1320 #if defined(INET) && defined(INET6)
1325 if (th->th_dport == th->th_sport &&
1326 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1327 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1328 log(LOG_DEBUG, "%s; %s: Listen socket: "
1329 "Connection attempt from/to self "
1330 "ignored\n", s, __func__);
1333 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1334 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1335 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1336 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1337 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1338 log(LOG_DEBUG, "%s; %s: Listen socket: "
1339 "Connection attempt from/to broad- "
1340 "or multicast address ignored\n",
1347 * SYN appears to be valid. Create compressed TCP state
1351 if (so->so_options & SO_DEBUG)
1352 tcp_trace(TA_INPUT, ostate, tp,
1353 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1355 tcp_dooptions(&to, optp, optlen, TO_SYN);
1356 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1358 * Entry added to syncache and mbuf consumed.
1359 * Everything already unlocked by syncache_add().
1361 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1365 #ifdef TCP_SIGNATURE
1366 if (sig_checked == 0) {
1367 tcp_dooptions(&to, optp, optlen,
1368 (thflags & TH_SYN) ? TO_SYN : 0);
1369 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1373 * In SYN_SENT state if it receives an RST, it is
1374 * allowed for further processing.
1376 if ((thflags & TH_RST) == 0 ||
1377 (tp->t_state == TCPS_SYN_SENT) == 0)
1385 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1386 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1387 * the inpcb, and unlocks pcbinfo.
1389 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1390 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1394 if (ti_locked == TI_WLOCKED) {
1395 INP_INFO_WUNLOCK(&V_tcbinfo);
1396 ti_locked = TI_UNLOCKED;
1400 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1401 "ti_locked: %d", __func__, ti_locked));
1402 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1407 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1410 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1411 m = NULL; /* mbuf chain got consumed. */
1415 if (ti_locked == TI_WLOCKED) {
1416 INP_INFO_WUNLOCK(&V_tcbinfo);
1417 ti_locked = TI_UNLOCKED;
1421 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1422 "ti_locked: %d", __func__, ti_locked));
1423 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1431 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1439 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1440 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1443 int thflags, acked, ourfinisacked, needoutput = 0;
1444 int rstreason, todrop, win;
1450 * The size of tcp_saveipgen must be the size of the max ip header,
1453 u_char tcp_saveipgen[IP6_HDR_LEN];
1454 struct tcphdr tcp_savetcp;
1457 thflags = th->th_flags;
1458 tp->sackhint.last_sack_ack = 0;
1461 * If this is either a state-changing packet or current state isn't
1462 * established, we require a write lock on tcbinfo. Otherwise, we
1463 * allow the tcbinfo to be in either alocked or unlocked, as the
1464 * caller may have unnecessarily acquired a write lock due to a race.
1466 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1467 tp->t_state != TCPS_ESTABLISHED) {
1468 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for "
1469 "SYN/FIN/RST/!EST", __func__, ti_locked));
1470 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1473 if (ti_locked == TI_WLOCKED)
1474 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1476 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1477 "ti_locked: %d", __func__, ti_locked));
1478 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1482 INP_WLOCK_ASSERT(tp->t_inpcb);
1483 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1485 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1489 * Segment received on connection.
1490 * Reset idle time and keep-alive timer.
1491 * XXX: This should be done after segment
1492 * validation to ignore broken/spoofed segs.
1494 tp->t_rcvtime = ticks;
1495 if (TCPS_HAVEESTABLISHED(tp->t_state))
1496 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
1499 * Unscale the window into a 32-bit value.
1500 * For the SYN_SENT state the scale is zero.
1502 tiwin = th->th_win << tp->snd_scale;
1505 * TCP ECN processing.
1507 if (tp->t_flags & TF_ECN_PERMIT) {
1508 if (thflags & TH_CWR)
1509 tp->t_flags &= ~TF_ECN_SND_ECE;
1510 switch (iptos & IPTOS_ECN_MASK) {
1512 tp->t_flags |= TF_ECN_SND_ECE;
1513 TCPSTAT_INC(tcps_ecn_ce);
1515 case IPTOS_ECN_ECT0:
1516 TCPSTAT_INC(tcps_ecn_ect0);
1518 case IPTOS_ECN_ECT1:
1519 TCPSTAT_INC(tcps_ecn_ect1);
1522 /* Congestion experienced. */
1523 if (thflags & TH_ECE) {
1524 cc_cong_signal(tp, th, CC_ECN);
1529 * Parse options on any incoming segment.
1531 tcp_dooptions(&to, (u_char *)(th + 1),
1532 (th->th_off << 2) - sizeof(struct tcphdr),
1533 (thflags & TH_SYN) ? TO_SYN : 0);
1536 * If echoed timestamp is later than the current time,
1537 * fall back to non RFC1323 RTT calculation. Normalize
1538 * timestamp if syncookies were used when this connection
1541 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1542 to.to_tsecr -= tp->ts_offset;
1543 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1548 * Process options only when we get SYN/ACK back. The SYN case
1549 * for incoming connections is handled in tcp_syncache.
1550 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1551 * or <SYN,ACK>) segment itself is never scaled.
1552 * XXX this is traditional behavior, may need to be cleaned up.
1554 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1555 if ((to.to_flags & TOF_SCALE) &&
1556 (tp->t_flags & TF_REQ_SCALE)) {
1557 tp->t_flags |= TF_RCVD_SCALE;
1558 tp->snd_scale = to.to_wscale;
1561 * Initial send window. It will be updated with
1562 * the next incoming segment to the scaled value.
1564 tp->snd_wnd = th->th_win;
1565 if (to.to_flags & TOF_TS) {
1566 tp->t_flags |= TF_RCVD_TSTMP;
1567 tp->ts_recent = to.to_tsval;
1568 tp->ts_recent_age = tcp_ts_getticks();
1570 if (to.to_flags & TOF_MSS)
1571 tcp_mss(tp, to.to_mss);
1572 if ((tp->t_flags & TF_SACK_PERMIT) &&
1573 (to.to_flags & TOF_SACKPERM) == 0)
1574 tp->t_flags &= ~TF_SACK_PERMIT;
1578 * Header prediction: check for the two common cases
1579 * of a uni-directional data xfer. If the packet has
1580 * no control flags, is in-sequence, the window didn't
1581 * change and we're not retransmitting, it's a
1582 * candidate. If the length is zero and the ack moved
1583 * forward, we're the sender side of the xfer. Just
1584 * free the data acked & wake any higher level process
1585 * that was blocked waiting for space. If the length
1586 * is non-zero and the ack didn't move, we're the
1587 * receiver side. If we're getting packets in-order
1588 * (the reassembly queue is empty), add the data to
1589 * the socket buffer and note that we need a delayed ack.
1590 * Make sure that the hidden state-flags are also off.
1591 * Since we check for TCPS_ESTABLISHED first, it can only
1594 if (tp->t_state == TCPS_ESTABLISHED &&
1595 th->th_seq == tp->rcv_nxt &&
1596 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1597 tp->snd_nxt == tp->snd_max &&
1598 tiwin && tiwin == tp->snd_wnd &&
1599 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1600 LIST_EMPTY(&tp->t_segq) &&
1601 ((to.to_flags & TOF_TS) == 0 ||
1602 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1605 * If last ACK falls within this segment's sequence numbers,
1606 * record the timestamp.
1607 * NOTE that the test is modified according to the latest
1608 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1610 if ((to.to_flags & TOF_TS) != 0 &&
1611 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1612 tp->ts_recent_age = tcp_ts_getticks();
1613 tp->ts_recent = to.to_tsval;
1617 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1618 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1619 !IN_RECOVERY(tp->t_flags) &&
1620 (to.to_flags & TOF_SACK) == 0 &&
1621 TAILQ_EMPTY(&tp->snd_holes)) {
1623 * This is a pure ack for outstanding data.
1625 if (ti_locked == TI_WLOCKED)
1626 INP_INFO_WUNLOCK(&V_tcbinfo);
1627 ti_locked = TI_UNLOCKED;
1629 TCPSTAT_INC(tcps_predack);
1632 * "bad retransmit" recovery.
1634 if (tp->t_rxtshift == 1 &&
1635 tp->t_flags & TF_PREVVALID &&
1636 (int)(ticks - tp->t_badrxtwin) < 0) {
1637 cc_cong_signal(tp, th, CC_RTO_ERR);
1641 * Recalculate the transmit timer / rtt.
1643 * Some boxes send broken timestamp replies
1644 * during the SYN+ACK phase, ignore
1645 * timestamps of 0 or we could calculate a
1646 * huge RTT and blow up the retransmit timer.
1648 if ((to.to_flags & TOF_TS) != 0 &&
1652 t = tcp_ts_getticks() - to.to_tsecr;
1653 if (!tp->t_rttlow || tp->t_rttlow > t)
1656 TCP_TS_TO_TICKS(t) + 1);
1657 } else if (tp->t_rtttime &&
1658 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1659 if (!tp->t_rttlow ||
1660 tp->t_rttlow > ticks - tp->t_rtttime)
1661 tp->t_rttlow = ticks - tp->t_rtttime;
1663 ticks - tp->t_rtttime);
1665 acked = BYTES_THIS_ACK(tp, th);
1667 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1668 hhook_run_tcp_est_in(tp, th, &to);
1670 TCPSTAT_INC(tcps_rcvackpack);
1671 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1672 sbdrop(&so->so_snd, acked);
1673 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1674 SEQ_LEQ(th->th_ack, tp->snd_recover))
1675 tp->snd_recover = th->th_ack - 1;
1678 * Let the congestion control algorithm update
1679 * congestion control related information. This
1680 * typically means increasing the congestion
1683 cc_ack_received(tp, th, CC_ACK);
1685 tp->snd_una = th->th_ack;
1687 * Pull snd_wl2 up to prevent seq wrap relative
1690 tp->snd_wl2 = th->th_ack;
1693 ND6_HINT(tp); /* Some progress has been made. */
1696 * If all outstanding data are acked, stop
1697 * retransmit timer, otherwise restart timer
1698 * using current (possibly backed-off) value.
1699 * If process is waiting for space,
1700 * wakeup/selwakeup/signal. If data
1701 * are ready to send, let tcp_output
1702 * decide between more output or persist.
1705 if (so->so_options & SO_DEBUG)
1706 tcp_trace(TA_INPUT, ostate, tp,
1707 (void *)tcp_saveipgen,
1710 if (tp->snd_una == tp->snd_max)
1711 tcp_timer_activate(tp, TT_REXMT, 0);
1712 else if (!tcp_timer_active(tp, TT_PERSIST))
1713 tcp_timer_activate(tp, TT_REXMT,
1716 if (so->so_snd.sb_cc)
1717 (void) tcp_output(tp);
1720 } else if (th->th_ack == tp->snd_una &&
1721 tlen <= sbspace(&so->so_rcv)) {
1722 int newsize = 0; /* automatic sockbuf scaling */
1725 * This is a pure, in-sequence data packet with
1726 * nothing on the reassembly queue and we have enough
1727 * buffer space to take it.
1729 if (ti_locked == TI_WLOCKED)
1730 INP_INFO_WUNLOCK(&V_tcbinfo);
1731 ti_locked = TI_UNLOCKED;
1733 /* Clean receiver SACK report if present */
1734 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1735 tcp_clean_sackreport(tp);
1736 TCPSTAT_INC(tcps_preddat);
1737 tp->rcv_nxt += tlen;
1739 * Pull snd_wl1 up to prevent seq wrap relative to
1742 tp->snd_wl1 = th->th_seq;
1744 * Pull rcv_up up to prevent seq wrap relative to
1747 tp->rcv_up = tp->rcv_nxt;
1748 TCPSTAT_INC(tcps_rcvpack);
1749 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1750 ND6_HINT(tp); /* Some progress has been made */
1752 if (so->so_options & SO_DEBUG)
1753 tcp_trace(TA_INPUT, ostate, tp,
1754 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1757 * Automatic sizing of receive socket buffer. Often the send
1758 * buffer size is not optimally adjusted to the actual network
1759 * conditions at hand (delay bandwidth product). Setting the
1760 * buffer size too small limits throughput on links with high
1761 * bandwidth and high delay (eg. trans-continental/oceanic links).
1763 * On the receive side the socket buffer memory is only rarely
1764 * used to any significant extent. This allows us to be much
1765 * more aggressive in scaling the receive socket buffer. For
1766 * the case that the buffer space is actually used to a large
1767 * extent and we run out of kernel memory we can simply drop
1768 * the new segments; TCP on the sender will just retransmit it
1769 * later. Setting the buffer size too big may only consume too
1770 * much kernel memory if the application doesn't read() from
1771 * the socket or packet loss or reordering makes use of the
1774 * The criteria to step up the receive buffer one notch are:
1775 * 1. the number of bytes received during the time it takes
1776 * one timestamp to be reflected back to us (the RTT);
1777 * 2. received bytes per RTT is within seven eighth of the
1778 * current socket buffer size;
1779 * 3. receive buffer size has not hit maximal automatic size;
1781 * This algorithm does one step per RTT at most and only if
1782 * we receive a bulk stream w/o packet losses or reorderings.
1783 * Shrinking the buffer during idle times is not necessary as
1784 * it doesn't consume any memory when idle.
1786 * TODO: Only step up if the application is actually serving
1787 * the buffer to better manage the socket buffer resources.
1789 if (V_tcp_do_autorcvbuf &&
1791 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1792 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1793 to.to_tsecr - tp->rfbuf_ts < hz) {
1795 (so->so_rcv.sb_hiwat / 8 * 7) &&
1796 so->so_rcv.sb_hiwat <
1797 V_tcp_autorcvbuf_max) {
1799 min(so->so_rcv.sb_hiwat +
1800 V_tcp_autorcvbuf_inc,
1801 V_tcp_autorcvbuf_max);
1803 /* Start over with next RTT. */
1807 tp->rfbuf_cnt += tlen; /* add up */
1810 /* Add data to socket buffer. */
1811 SOCKBUF_LOCK(&so->so_rcv);
1812 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1816 * Set new socket buffer size.
1817 * Give up when limit is reached.
1820 if (!sbreserve_locked(&so->so_rcv,
1822 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1823 m_adj(m, drop_hdrlen); /* delayed header drop */
1824 sbappendstream_locked(&so->so_rcv, m);
1826 /* NB: sorwakeup_locked() does an implicit unlock. */
1827 sorwakeup_locked(so);
1828 if (DELAY_ACK(tp)) {
1829 tp->t_flags |= TF_DELACK;
1831 tp->t_flags |= TF_ACKNOW;
1839 * Calculate amount of space in receive window,
1840 * and then do TCP input processing.
1841 * Receive window is amount of space in rcv queue,
1842 * but not less than advertised window.
1844 win = sbspace(&so->so_rcv);
1847 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1849 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1853 switch (tp->t_state) {
1856 * If the state is SYN_RECEIVED:
1857 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1859 case TCPS_SYN_RECEIVED:
1860 if ((thflags & TH_ACK) &&
1861 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1862 SEQ_GT(th->th_ack, tp->snd_max))) {
1863 rstreason = BANDLIM_RST_OPENPORT;
1869 * If the state is SYN_SENT:
1870 * if seg contains an ACK, but not for our SYN, drop the input.
1871 * if seg contains a RST, then drop the connection.
1872 * if seg does not contain SYN, then drop it.
1873 * Otherwise this is an acceptable SYN segment
1874 * initialize tp->rcv_nxt and tp->irs
1875 * if seg contains ack then advance tp->snd_una
1876 * if seg contains an ECE and ECN support is enabled, the stream
1878 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1879 * arrange for segment to be acked (eventually)
1880 * continue processing rest of data/controls, beginning with URG
1883 if ((thflags & TH_ACK) &&
1884 (SEQ_LEQ(th->th_ack, tp->iss) ||
1885 SEQ_GT(th->th_ack, tp->snd_max))) {
1886 rstreason = BANDLIM_UNLIMITED;
1889 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST))
1890 tp = tcp_drop(tp, ECONNREFUSED);
1891 if (thflags & TH_RST)
1893 if (!(thflags & TH_SYN))
1896 tp->irs = th->th_seq;
1898 if (thflags & TH_ACK) {
1899 TCPSTAT_INC(tcps_connects);
1902 mac_socketpeer_set_from_mbuf(m, so);
1904 /* Do window scaling on this connection? */
1905 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1906 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1907 tp->rcv_scale = tp->request_r_scale;
1909 tp->rcv_adv += imin(tp->rcv_wnd,
1910 TCP_MAXWIN << tp->rcv_scale);
1911 tp->snd_una++; /* SYN is acked */
1913 * If there's data, delay ACK; if there's also a FIN
1914 * ACKNOW will be turned on later.
1916 if (DELAY_ACK(tp) && tlen != 0)
1917 tcp_timer_activate(tp, TT_DELACK,
1920 tp->t_flags |= TF_ACKNOW;
1922 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1923 tp->t_flags |= TF_ECN_PERMIT;
1924 TCPSTAT_INC(tcps_ecn_shs);
1928 * Received <SYN,ACK> in SYN_SENT[*] state.
1930 * SYN_SENT --> ESTABLISHED
1931 * SYN_SENT* --> FIN_WAIT_1
1933 tp->t_starttime = ticks;
1934 if (tp->t_flags & TF_NEEDFIN) {
1935 tp->t_state = TCPS_FIN_WAIT_1;
1936 tp->t_flags &= ~TF_NEEDFIN;
1939 tp->t_state = TCPS_ESTABLISHED;
1941 tcp_timer_activate(tp, TT_KEEP,
1946 * Received initial SYN in SYN-SENT[*] state =>
1947 * simultaneous open. If segment contains CC option
1948 * and there is a cached CC, apply TAO test.
1949 * If it succeeds, connection is * half-synchronized.
1950 * Otherwise, do 3-way handshake:
1951 * SYN-SENT -> SYN-RECEIVED
1952 * SYN-SENT* -> SYN-RECEIVED*
1953 * If there was no CC option, clear cached CC value.
1955 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1956 tcp_timer_activate(tp, TT_REXMT, 0);
1957 tp->t_state = TCPS_SYN_RECEIVED;
1960 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: "
1961 "ti_locked %d", __func__, ti_locked));
1962 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1963 INP_WLOCK_ASSERT(tp->t_inpcb);
1966 * Advance th->th_seq to correspond to first data byte.
1967 * If data, trim to stay within window,
1968 * dropping FIN if necessary.
1971 if (tlen > tp->rcv_wnd) {
1972 todrop = tlen - tp->rcv_wnd;
1976 TCPSTAT_INC(tcps_rcvpackafterwin);
1977 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
1979 tp->snd_wl1 = th->th_seq - 1;
1980 tp->rcv_up = th->th_seq;
1982 * Client side of transaction: already sent SYN and data.
1983 * If the remote host used T/TCP to validate the SYN,
1984 * our data will be ACK'd; if so, enter normal data segment
1985 * processing in the middle of step 5, ack processing.
1986 * Otherwise, goto step 6.
1988 if (thflags & TH_ACK)
1994 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1995 * do normal processing.
1997 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2001 break; /* continue normal processing */
2005 * States other than LISTEN or SYN_SENT.
2006 * First check the RST flag and sequence number since reset segments
2007 * are exempt from the timestamp and connection count tests. This
2008 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2009 * below which allowed reset segments in half the sequence space
2010 * to fall though and be processed (which gives forged reset
2011 * segments with a random sequence number a 50 percent chance of
2012 * killing a connection).
2013 * Then check timestamp, if present.
2014 * Then check the connection count, if present.
2015 * Then check that at least some bytes of segment are within
2016 * receive window. If segment begins before rcv_nxt,
2017 * drop leading data (and SYN); if nothing left, just ack.
2020 * If the RST bit is set, check the sequence number to see
2021 * if this is a valid reset segment.
2023 * In all states except SYN-SENT, all reset (RST) segments
2024 * are validated by checking their SEQ-fields. A reset is
2025 * valid if its sequence number is in the window.
2026 * Note: this does not take into account delayed ACKs, so
2027 * we should test against last_ack_sent instead of rcv_nxt.
2028 * The sequence number in the reset segment is normally an
2029 * echo of our outgoing acknowlegement numbers, but some hosts
2030 * send a reset with the sequence number at the rightmost edge
2031 * of our receive window, and we have to handle this case.
2032 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
2033 * that brute force RST attacks are possible. To combat this,
2034 * we use a much stricter check while in the ESTABLISHED state,
2035 * only accepting RSTs where the sequence number is equal to
2036 * last_ack_sent. In all other states (the states in which a
2037 * RST is more likely), the more permissive check is used.
2038 * If we have multiple segments in flight, the initial reset
2039 * segment sequence numbers will be to the left of last_ack_sent,
2040 * but they will eventually catch up.
2041 * In any case, it never made sense to trim reset segments to
2042 * fit the receive window since RFC 1122 says:
2043 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
2045 * A TCP SHOULD allow a received RST segment to include data.
2048 * It has been suggested that a RST segment could contain
2049 * ASCII text that encoded and explained the cause of the
2050 * RST. No standard has yet been established for such
2053 * If the reset segment passes the sequence number test examine
2055 * SYN_RECEIVED STATE:
2056 * If passive open, return to LISTEN state.
2057 * If active open, inform user that connection was refused.
2058 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
2059 * Inform user that connection was reset, and close tcb.
2060 * CLOSING, LAST_ACK STATES:
2063 * Drop the segment - see Stevens, vol. 2, p. 964 and
2066 if (thflags & TH_RST) {
2067 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2068 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2069 switch (tp->t_state) {
2071 case TCPS_SYN_RECEIVED:
2072 so->so_error = ECONNREFUSED;
2075 case TCPS_ESTABLISHED:
2076 if (V_tcp_insecure_rst == 0 &&
2077 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
2078 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
2079 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2080 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
2081 TCPSTAT_INC(tcps_badrst);
2085 case TCPS_FIN_WAIT_1:
2086 case TCPS_FIN_WAIT_2:
2087 case TCPS_CLOSE_WAIT:
2088 so->so_error = ECONNRESET;
2090 KASSERT(ti_locked == TI_WLOCKED,
2091 ("tcp_do_segment: TH_RST 1 ti_locked %d",
2093 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2095 tp->t_state = TCPS_CLOSED;
2096 TCPSTAT_INC(tcps_drops);
2102 KASSERT(ti_locked == TI_WLOCKED,
2103 ("tcp_do_segment: TH_RST 2 ti_locked %d",
2105 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2115 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2116 * and it's less than ts_recent, drop it.
2118 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2119 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2121 /* Check to see if ts_recent is over 24 days old. */
2122 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2124 * Invalidate ts_recent. If this segment updates
2125 * ts_recent, the age will be reset later and ts_recent
2126 * will get a valid value. If it does not, setting
2127 * ts_recent to zero will at least satisfy the
2128 * requirement that zero be placed in the timestamp
2129 * echo reply when ts_recent isn't valid. The
2130 * age isn't reset until we get a valid ts_recent
2131 * because we don't want out-of-order segments to be
2132 * dropped when ts_recent is old.
2136 TCPSTAT_INC(tcps_rcvduppack);
2137 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2138 TCPSTAT_INC(tcps_pawsdrop);
2146 * In the SYN-RECEIVED state, validate that the packet belongs to
2147 * this connection before trimming the data to fit the receive
2148 * window. Check the sequence number versus IRS since we know
2149 * the sequence numbers haven't wrapped. This is a partial fix
2150 * for the "LAND" DoS attack.
2152 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2153 rstreason = BANDLIM_RST_OPENPORT;
2157 todrop = tp->rcv_nxt - th->th_seq;
2160 * If this is a duplicate SYN for our current connection,
2161 * advance over it and pretend and it's not a SYN.
2163 if (thflags & TH_SYN && th->th_seq == tp->irs) {
2173 * Following if statement from Stevens, vol. 2, p. 960.
2176 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2178 * Any valid FIN must be to the left of the window.
2179 * At this point the FIN must be a duplicate or out
2180 * of sequence; drop it.
2185 * Send an ACK to resynchronize and drop any data.
2186 * But keep on processing for RST or ACK.
2188 tp->t_flags |= TF_ACKNOW;
2190 TCPSTAT_INC(tcps_rcvduppack);
2191 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2193 TCPSTAT_INC(tcps_rcvpartduppack);
2194 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2196 drop_hdrlen += todrop; /* drop from the top afterwards */
2197 th->th_seq += todrop;
2199 if (th->th_urp > todrop)
2200 th->th_urp -= todrop;
2208 * If new data are received on a connection after the
2209 * user processes are gone, then RST the other end.
2211 if ((so->so_state & SS_NOFDREF) &&
2212 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2215 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && "
2216 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2217 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2219 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
2220 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket "
2221 "was closed, sending RST and removing tcpcb\n",
2222 s, __func__, tcpstates[tp->t_state], tlen);
2226 TCPSTAT_INC(tcps_rcvafterclose);
2227 rstreason = BANDLIM_UNLIMITED;
2232 * If segment ends after window, drop trailing data
2233 * (and PUSH and FIN); if nothing left, just ACK.
2235 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2237 TCPSTAT_INC(tcps_rcvpackafterwin);
2238 if (todrop >= tlen) {
2239 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2241 * If window is closed can only take segments at
2242 * window edge, and have to drop data and PUSH from
2243 * incoming segments. Continue processing, but
2244 * remember to ack. Otherwise, drop segment
2247 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2248 tp->t_flags |= TF_ACKNOW;
2249 TCPSTAT_INC(tcps_rcvwinprobe);
2253 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2256 thflags &= ~(TH_PUSH|TH_FIN);
2260 * If last ACK falls within this segment's sequence numbers,
2261 * record its timestamp.
2263 * 1) That the test incorporates suggestions from the latest
2264 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2265 * 2) That updating only on newer timestamps interferes with
2266 * our earlier PAWS tests, so this check should be solely
2267 * predicated on the sequence space of this segment.
2268 * 3) That we modify the segment boundary check to be
2269 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2270 * instead of RFC1323's
2271 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2272 * This modified check allows us to overcome RFC1323's
2273 * limitations as described in Stevens TCP/IP Illustrated
2274 * Vol. 2 p.869. In such cases, we can still calculate the
2275 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2277 if ((to.to_flags & TOF_TS) != 0 &&
2278 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2279 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2280 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2281 tp->ts_recent_age = tcp_ts_getticks();
2282 tp->ts_recent = to.to_tsval;
2286 * If a SYN is in the window, then this is an
2287 * error and we send an RST and drop the connection.
2289 if (thflags & TH_SYN) {
2290 KASSERT(ti_locked == TI_WLOCKED,
2291 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2292 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2294 tp = tcp_drop(tp, ECONNRESET);
2295 rstreason = BANDLIM_UNLIMITED;
2300 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2301 * flag is on (half-synchronized state), then queue data for
2302 * later processing; else drop segment and return.
2304 if ((thflags & TH_ACK) == 0) {
2305 if (tp->t_state == TCPS_SYN_RECEIVED ||
2306 (tp->t_flags & TF_NEEDSYN))
2308 else if (tp->t_flags & TF_ACKNOW)
2317 switch (tp->t_state) {
2320 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2321 * ESTABLISHED state and continue processing.
2322 * The ACK was checked above.
2324 case TCPS_SYN_RECEIVED:
2326 TCPSTAT_INC(tcps_connects);
2328 /* Do window scaling? */
2329 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2330 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2331 tp->rcv_scale = tp->request_r_scale;
2332 tp->snd_wnd = tiwin;
2336 * SYN-RECEIVED -> ESTABLISHED
2337 * SYN-RECEIVED* -> FIN-WAIT-1
2339 tp->t_starttime = ticks;
2340 if (tp->t_flags & TF_NEEDFIN) {
2341 tp->t_state = TCPS_FIN_WAIT_1;
2342 tp->t_flags &= ~TF_NEEDFIN;
2344 tp->t_state = TCPS_ESTABLISHED;
2346 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2349 * If segment contains data or ACK, will call tcp_reass()
2350 * later; if not, do so now to pass queued data to user.
2352 if (tlen == 0 && (thflags & TH_FIN) == 0)
2353 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2355 tp->snd_wl1 = th->th_seq - 1;
2359 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2360 * ACKs. If the ack is in the range
2361 * tp->snd_una < th->th_ack <= tp->snd_max
2362 * then advance tp->snd_una to th->th_ack and drop
2363 * data from the retransmission queue. If this ACK reflects
2364 * more up to date window information we update our window information.
2366 case TCPS_ESTABLISHED:
2367 case TCPS_FIN_WAIT_1:
2368 case TCPS_FIN_WAIT_2:
2369 case TCPS_CLOSE_WAIT:
2372 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2373 TCPSTAT_INC(tcps_rcvacktoomuch);
2376 if ((tp->t_flags & TF_SACK_PERMIT) &&
2377 ((to.to_flags & TOF_SACK) ||
2378 !TAILQ_EMPTY(&tp->snd_holes)))
2379 tcp_sack_doack(tp, &to, th->th_ack);
2381 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2382 hhook_run_tcp_est_in(tp, th, &to);
2384 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2385 if (tlen == 0 && tiwin == tp->snd_wnd) {
2386 TCPSTAT_INC(tcps_rcvdupack);
2388 * If we have outstanding data (other than
2389 * a window probe), this is a completely
2390 * duplicate ack (ie, window info didn't
2391 * change), the ack is the biggest we've
2392 * seen and we've seen exactly our rexmt
2393 * threshhold of them, assume a packet
2394 * has been dropped and retransmit it.
2395 * Kludge snd_nxt & the congestion
2396 * window so we send only this one
2399 * We know we're losing at the current
2400 * window size so do congestion avoidance
2401 * (set ssthresh to half the current window
2402 * and pull our congestion window back to
2403 * the new ssthresh).
2405 * Dup acks mean that packets have left the
2406 * network (they're now cached at the receiver)
2407 * so bump cwnd by the amount in the receiver
2408 * to keep a constant cwnd packets in the
2411 * When using TCP ECN, notify the peer that
2412 * we reduced the cwnd.
2414 if (!tcp_timer_active(tp, TT_REXMT) ||
2415 th->th_ack != tp->snd_una)
2417 else if (++tp->t_dupacks > tcprexmtthresh ||
2418 IN_FASTRECOVERY(tp->t_flags)) {
2419 cc_ack_received(tp, th, CC_DUPACK);
2420 if ((tp->t_flags & TF_SACK_PERMIT) &&
2421 IN_FASTRECOVERY(tp->t_flags)) {
2425 * Compute the amount of data in flight first.
2426 * We can inject new data into the pipe iff
2427 * we have less than 1/2 the original window's
2428 * worth of data in flight.
2430 awnd = (tp->snd_nxt - tp->snd_fack) +
2431 tp->sackhint.sack_bytes_rexmit;
2432 if (awnd < tp->snd_ssthresh) {
2433 tp->snd_cwnd += tp->t_maxseg;
2434 if (tp->snd_cwnd > tp->snd_ssthresh)
2435 tp->snd_cwnd = tp->snd_ssthresh;
2438 tp->snd_cwnd += tp->t_maxseg;
2439 if ((thflags & TH_FIN) &&
2440 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2442 * If its a fin we need to process
2443 * it to avoid a race where both
2444 * sides enter FIN-WAIT and send FIN|ACK
2449 (void) tcp_output(tp);
2451 } else if (tp->t_dupacks == tcprexmtthresh) {
2452 tcp_seq onxt = tp->snd_nxt;
2455 * If we're doing sack, check to
2456 * see if we're already in sack
2457 * recovery. If we're not doing sack,
2458 * check to see if we're in newreno
2461 if (tp->t_flags & TF_SACK_PERMIT) {
2462 if (IN_FASTRECOVERY(tp->t_flags)) {
2467 if (SEQ_LEQ(th->th_ack,
2473 /* Congestion signal before ack. */
2474 cc_cong_signal(tp, th, CC_NDUPACK);
2475 cc_ack_received(tp, th, CC_DUPACK);
2476 tcp_timer_activate(tp, TT_REXMT, 0);
2478 if (tp->t_flags & TF_SACK_PERMIT) {
2480 tcps_sack_recovery_episode);
2481 tp->sack_newdata = tp->snd_nxt;
2482 tp->snd_cwnd = tp->t_maxseg;
2483 (void) tcp_output(tp);
2486 tp->snd_nxt = th->th_ack;
2487 tp->snd_cwnd = tp->t_maxseg;
2488 if ((thflags & TH_FIN) &&
2489 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2491 * If its a fin we need to process
2492 * it to avoid a race where both
2493 * sides enter FIN-WAIT and send FIN|ACK
2498 (void) tcp_output(tp);
2499 KASSERT(tp->snd_limited <= 2,
2500 ("%s: tp->snd_limited too big",
2502 tp->snd_cwnd = tp->snd_ssthresh +
2504 (tp->t_dupacks - tp->snd_limited);
2505 if (SEQ_GT(onxt, tp->snd_nxt))
2508 } else if (V_tcp_do_rfc3042) {
2509 cc_ack_received(tp, th, CC_DUPACK);
2510 u_long oldcwnd = tp->snd_cwnd;
2511 tcp_seq oldsndmax = tp->snd_max;
2514 KASSERT(tp->t_dupacks == 1 ||
2516 ("%s: dupacks not 1 or 2",
2518 if (tp->t_dupacks == 1)
2519 tp->snd_limited = 0;
2521 (tp->snd_nxt - tp->snd_una) +
2522 (tp->t_dupacks - tp->snd_limited) *
2524 if ((thflags & TH_FIN) &&
2525 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2527 * If its a fin we need to process
2528 * it to avoid a race where both
2529 * sides enter FIN-WAIT and send FIN|ACK
2534 (void) tcp_output(tp);
2535 sent = tp->snd_max - oldsndmax;
2536 if (sent > tp->t_maxseg) {
2537 KASSERT((tp->t_dupacks == 2 &&
2538 tp->snd_limited == 0) ||
2539 (sent == tp->t_maxseg + 1 &&
2540 tp->t_flags & TF_SENTFIN),
2541 ("%s: sent too much",
2543 tp->snd_limited = 2;
2544 } else if (sent > 0)
2546 tp->snd_cwnd = oldcwnd;
2554 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2555 ("%s: th_ack <= snd_una", __func__));
2558 * If the congestion window was inflated to account
2559 * for the other side's cached packets, retract it.
2561 if (IN_FASTRECOVERY(tp->t_flags)) {
2562 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2563 if (tp->t_flags & TF_SACK_PERMIT)
2564 tcp_sack_partialack(tp, th);
2566 tcp_newreno_partial_ack(tp, th);
2568 cc_post_recovery(tp, th);
2572 * If we reach this point, ACK is not a duplicate,
2573 * i.e., it ACKs something we sent.
2575 if (tp->t_flags & TF_NEEDSYN) {
2577 * T/TCP: Connection was half-synchronized, and our
2578 * SYN has been ACK'd (so connection is now fully
2579 * synchronized). Go to non-starred state,
2580 * increment snd_una for ACK of SYN, and check if
2581 * we can do window scaling.
2583 tp->t_flags &= ~TF_NEEDSYN;
2585 /* Do window scaling? */
2586 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2587 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2588 tp->rcv_scale = tp->request_r_scale;
2589 /* Send window already scaled. */
2594 INP_WLOCK_ASSERT(tp->t_inpcb);
2596 acked = BYTES_THIS_ACK(tp, th);
2597 TCPSTAT_INC(tcps_rcvackpack);
2598 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2601 * If we just performed our first retransmit, and the ACK
2602 * arrives within our recovery window, then it was a mistake
2603 * to do the retransmit in the first place. Recover our
2604 * original cwnd and ssthresh, and proceed to transmit where
2607 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2608 (int)(ticks - tp->t_badrxtwin) < 0)
2609 cc_cong_signal(tp, th, CC_RTO_ERR);
2612 * If we have a timestamp reply, update smoothed
2613 * round trip time. If no timestamp is present but
2614 * transmit timer is running and timed sequence
2615 * number was acked, update smoothed round trip time.
2616 * Since we now have an rtt measurement, cancel the
2617 * timer backoff (cf., Phil Karn's retransmit alg.).
2618 * Recompute the initial retransmit timer.
2620 * Some boxes send broken timestamp replies
2621 * during the SYN+ACK phase, ignore
2622 * timestamps of 0 or we could calculate a
2623 * huge RTT and blow up the retransmit timer.
2625 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2628 t = tcp_ts_getticks() - to.to_tsecr;
2629 if (!tp->t_rttlow || tp->t_rttlow > t)
2631 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2632 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2633 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2634 tp->t_rttlow = ticks - tp->t_rtttime;
2635 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2639 * If all outstanding data is acked, stop retransmit
2640 * timer and remember to restart (more output or persist).
2641 * If there is more data to be acked, restart retransmit
2642 * timer, using current (possibly backed-off) value.
2644 if (th->th_ack == tp->snd_max) {
2645 tcp_timer_activate(tp, TT_REXMT, 0);
2647 } else if (!tcp_timer_active(tp, TT_PERSIST))
2648 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2651 * If no data (only SYN) was ACK'd,
2652 * skip rest of ACK processing.
2658 * Let the congestion control algorithm update congestion
2659 * control related information. This typically means increasing
2660 * the congestion window.
2662 cc_ack_received(tp, th, CC_ACK);
2664 SOCKBUF_LOCK(&so->so_snd);
2665 if (acked > so->so_snd.sb_cc) {
2666 tp->snd_wnd -= so->so_snd.sb_cc;
2667 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2670 sbdrop_locked(&so->so_snd, acked);
2671 tp->snd_wnd -= acked;
2674 /* NB: sowwakeup_locked() does an implicit unlock. */
2675 sowwakeup_locked(so);
2676 /* Detect una wraparound. */
2677 if (!IN_RECOVERY(tp->t_flags) &&
2678 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2679 SEQ_LEQ(th->th_ack, tp->snd_recover))
2680 tp->snd_recover = th->th_ack - 1;
2681 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2682 if (IN_RECOVERY(tp->t_flags) &&
2683 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2684 EXIT_RECOVERY(tp->t_flags);
2686 tp->snd_una = th->th_ack;
2687 if (tp->t_flags & TF_SACK_PERMIT) {
2688 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2689 tp->snd_recover = tp->snd_una;
2691 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2692 tp->snd_nxt = tp->snd_una;
2694 switch (tp->t_state) {
2697 * In FIN_WAIT_1 STATE in addition to the processing
2698 * for the ESTABLISHED state if our FIN is now acknowledged
2699 * then enter FIN_WAIT_2.
2701 case TCPS_FIN_WAIT_1:
2702 if (ourfinisacked) {
2704 * If we can't receive any more
2705 * data, then closing user can proceed.
2706 * Starting the timer is contrary to the
2707 * specification, but if we don't get a FIN
2708 * we'll hang forever.
2711 * we should release the tp also, and use a
2714 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2715 soisdisconnected(so);
2716 tcp_timer_activate(tp, TT_2MSL,
2717 (tcp_fast_finwait2_recycle ?
2718 tcp_finwait2_timeout :
2721 tp->t_state = TCPS_FIN_WAIT_2;
2726 * In CLOSING STATE in addition to the processing for
2727 * the ESTABLISHED state if the ACK acknowledges our FIN
2728 * then enter the TIME-WAIT state, otherwise ignore
2732 if (ourfinisacked) {
2733 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2735 INP_INFO_WUNLOCK(&V_tcbinfo);
2742 * In LAST_ACK, we may still be waiting for data to drain
2743 * and/or to be acked, as well as for the ack of our FIN.
2744 * If our FIN is now acknowledged, delete the TCB,
2745 * enter the closed state and return.
2748 if (ourfinisacked) {
2749 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2758 INP_WLOCK_ASSERT(tp->t_inpcb);
2761 * Update window information.
2762 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2764 if ((thflags & TH_ACK) &&
2765 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2766 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2767 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2768 /* keep track of pure window updates */
2770 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2771 TCPSTAT_INC(tcps_rcvwinupd);
2772 tp->snd_wnd = tiwin;
2773 tp->snd_wl1 = th->th_seq;
2774 tp->snd_wl2 = th->th_ack;
2775 if (tp->snd_wnd > tp->max_sndwnd)
2776 tp->max_sndwnd = tp->snd_wnd;
2781 * Process segments with URG.
2783 if ((thflags & TH_URG) && th->th_urp &&
2784 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2786 * This is a kludge, but if we receive and accept
2787 * random urgent pointers, we'll crash in
2788 * soreceive. It's hard to imagine someone
2789 * actually wanting to send this much urgent data.
2791 SOCKBUF_LOCK(&so->so_rcv);
2792 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2793 th->th_urp = 0; /* XXX */
2794 thflags &= ~TH_URG; /* XXX */
2795 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2796 goto dodata; /* XXX */
2799 * If this segment advances the known urgent pointer,
2800 * then mark the data stream. This should not happen
2801 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2802 * a FIN has been received from the remote side.
2803 * In these states we ignore the URG.
2805 * According to RFC961 (Assigned Protocols),
2806 * the urgent pointer points to the last octet
2807 * of urgent data. We continue, however,
2808 * to consider it to indicate the first octet
2809 * of data past the urgent section as the original
2810 * spec states (in one of two places).
2812 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2813 tp->rcv_up = th->th_seq + th->th_urp;
2814 so->so_oobmark = so->so_rcv.sb_cc +
2815 (tp->rcv_up - tp->rcv_nxt) - 1;
2816 if (so->so_oobmark == 0)
2817 so->so_rcv.sb_state |= SBS_RCVATMARK;
2819 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2821 SOCKBUF_UNLOCK(&so->so_rcv);
2823 * Remove out of band data so doesn't get presented to user.
2824 * This can happen independent of advancing the URG pointer,
2825 * but if two URG's are pending at once, some out-of-band
2826 * data may creep in... ick.
2828 if (th->th_urp <= (u_long)tlen &&
2829 !(so->so_options & SO_OOBINLINE)) {
2830 /* hdr drop is delayed */
2831 tcp_pulloutofband(so, th, m, drop_hdrlen);
2835 * If no out of band data is expected,
2836 * pull receive urgent pointer along
2837 * with the receive window.
2839 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2840 tp->rcv_up = tp->rcv_nxt;
2843 INP_WLOCK_ASSERT(tp->t_inpcb);
2846 * Process the segment text, merging it into the TCP sequencing queue,
2847 * and arranging for acknowledgment of receipt if necessary.
2848 * This process logically involves adjusting tp->rcv_wnd as data
2849 * is presented to the user (this happens in tcp_usrreq.c,
2850 * case PRU_RCVD). If a FIN has already been received on this
2851 * connection then we just ignore the text.
2853 if ((tlen || (thflags & TH_FIN)) &&
2854 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2855 tcp_seq save_start = th->th_seq;
2856 m_adj(m, drop_hdrlen); /* delayed header drop */
2858 * Insert segment which includes th into TCP reassembly queue
2859 * with control block tp. Set thflags to whether reassembly now
2860 * includes a segment with FIN. This handles the common case
2861 * inline (segment is the next to be received on an established
2862 * connection, and the queue is empty), avoiding linkage into
2863 * and removal from the queue and repetition of various
2865 * Set DELACK for segments received in order, but ack
2866 * immediately when segments are out of order (so
2867 * fast retransmit can work).
2869 if (th->th_seq == tp->rcv_nxt &&
2870 LIST_EMPTY(&tp->t_segq) &&
2871 TCPS_HAVEESTABLISHED(tp->t_state)) {
2873 tp->t_flags |= TF_DELACK;
2875 tp->t_flags |= TF_ACKNOW;
2876 tp->rcv_nxt += tlen;
2877 thflags = th->th_flags & TH_FIN;
2878 TCPSTAT_INC(tcps_rcvpack);
2879 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2881 SOCKBUF_LOCK(&so->so_rcv);
2882 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2885 sbappendstream_locked(&so->so_rcv, m);
2886 /* NB: sorwakeup_locked() does an implicit unlock. */
2887 sorwakeup_locked(so);
2890 * XXX: Due to the header drop above "th" is
2891 * theoretically invalid by now. Fortunately
2892 * m_adj() doesn't actually frees any mbufs
2893 * when trimming from the head.
2895 thflags = tcp_reass(tp, th, &tlen, m);
2896 tp->t_flags |= TF_ACKNOW;
2898 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2899 tcp_update_sack_list(tp, save_start, save_start + tlen);
2902 * Note the amount of data that peer has sent into
2903 * our window, in order to estimate the sender's
2907 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2908 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2910 len = so->so_rcv.sb_hiwat;
2918 * If FIN is received ACK the FIN and let the user know
2919 * that the connection is closing.
2921 if (thflags & TH_FIN) {
2922 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2925 * If connection is half-synchronized
2926 * (ie NEEDSYN flag on) then delay ACK,
2927 * so it may be piggybacked when SYN is sent.
2928 * Otherwise, since we received a FIN then no
2929 * more input can be expected, send ACK now.
2931 if (tp->t_flags & TF_NEEDSYN)
2932 tp->t_flags |= TF_DELACK;
2934 tp->t_flags |= TF_ACKNOW;
2937 switch (tp->t_state) {
2940 * In SYN_RECEIVED and ESTABLISHED STATES
2941 * enter the CLOSE_WAIT state.
2943 case TCPS_SYN_RECEIVED:
2944 tp->t_starttime = ticks;
2946 case TCPS_ESTABLISHED:
2947 tp->t_state = TCPS_CLOSE_WAIT;
2951 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2952 * enter the CLOSING state.
2954 case TCPS_FIN_WAIT_1:
2955 tp->t_state = TCPS_CLOSING;
2959 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2960 * starting the time-wait timer, turning off the other
2963 case TCPS_FIN_WAIT_2:
2964 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2965 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata "
2966 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
2970 INP_INFO_WUNLOCK(&V_tcbinfo);
2974 if (ti_locked == TI_WLOCKED)
2975 INP_INFO_WUNLOCK(&V_tcbinfo);
2976 ti_locked = TI_UNLOCKED;
2979 if (so->so_options & SO_DEBUG)
2980 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2985 * Return any desired output.
2987 if (needoutput || (tp->t_flags & TF_ACKNOW))
2988 (void) tcp_output(tp);
2991 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
2992 __func__, ti_locked));
2993 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
2994 INP_WLOCK_ASSERT(tp->t_inpcb);
2996 if (tp->t_flags & TF_DELACK) {
2997 tp->t_flags &= ~TF_DELACK;
2998 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3000 INP_WUNLOCK(tp->t_inpcb);
3005 * Generate an ACK dropping incoming segment if it occupies
3006 * sequence space, where the ACK reflects our state.
3008 * We can now skip the test for the RST flag since all
3009 * paths to this code happen after packets containing
3010 * RST have been dropped.
3012 * In the SYN-RECEIVED state, don't send an ACK unless the
3013 * segment we received passes the SYN-RECEIVED ACK test.
3014 * If it fails send a RST. This breaks the loop in the
3015 * "LAND" DoS attack, and also prevents an ACK storm
3016 * between two listening ports that have been sent forged
3017 * SYN segments, each with the source address of the other.
3019 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3020 (SEQ_GT(tp->snd_una, th->th_ack) ||
3021 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3022 rstreason = BANDLIM_RST_OPENPORT;
3026 if (so->so_options & SO_DEBUG)
3027 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3030 if (ti_locked == TI_WLOCKED)
3031 INP_INFO_WUNLOCK(&V_tcbinfo);
3032 ti_locked = TI_UNLOCKED;
3034 tp->t_flags |= TF_ACKNOW;
3035 (void) tcp_output(tp);
3036 INP_WUNLOCK(tp->t_inpcb);
3041 if (ti_locked == TI_WLOCKED)
3042 INP_INFO_WUNLOCK(&V_tcbinfo);
3043 ti_locked = TI_UNLOCKED;
3046 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3047 INP_WUNLOCK(tp->t_inpcb);
3049 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3053 if (ti_locked == TI_WLOCKED) {
3054 INP_INFO_WUNLOCK(&V_tcbinfo);
3055 ti_locked = TI_UNLOCKED;
3059 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3063 * Drop space held by incoming segment and return.
3066 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3067 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3071 INP_WUNLOCK(tp->t_inpcb);
3076 * Issue RST and make ACK acceptable to originator of segment.
3077 * The mbuf must still include the original packet header.
3081 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3082 int tlen, int rstreason)
3088 struct ip6_hdr *ip6;
3092 INP_WLOCK_ASSERT(tp->t_inpcb);
3095 /* Don't bother if destination was broadcast/multicast. */
3096 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3099 if (mtod(m, struct ip *)->ip_v == 6) {
3100 ip6 = mtod(m, struct ip6_hdr *);
3101 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3102 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3104 /* IPv6 anycast check is done at tcp6_input() */
3107 #if defined(INET) && defined(INET6)
3112 ip = mtod(m, struct ip *);
3113 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3114 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3115 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3116 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3121 /* Perform bandwidth limiting. */
3122 if (badport_bandlim(rstreason) < 0)
3125 /* tcp_respond consumes the mbuf chain. */
3126 if (th->th_flags & TH_ACK) {
3127 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3128 th->th_ack, TH_RST);
3130 if (th->th_flags & TH_SYN)
3132 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3133 (tcp_seq)0, TH_RST|TH_ACK);
3141 * Parse TCP options and place in tcpopt.
3144 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3149 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3151 if (opt == TCPOPT_EOL)
3153 if (opt == TCPOPT_NOP)
3159 if (optlen < 2 || optlen > cnt)
3164 if (optlen != TCPOLEN_MAXSEG)
3166 if (!(flags & TO_SYN))
3168 to->to_flags |= TOF_MSS;
3169 bcopy((char *)cp + 2,
3170 (char *)&to->to_mss, sizeof(to->to_mss));
3171 to->to_mss = ntohs(to->to_mss);
3174 if (optlen != TCPOLEN_WINDOW)
3176 if (!(flags & TO_SYN))
3178 to->to_flags |= TOF_SCALE;
3179 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3181 case TCPOPT_TIMESTAMP:
3182 if (optlen != TCPOLEN_TIMESTAMP)
3184 to->to_flags |= TOF_TS;
3185 bcopy((char *)cp + 2,
3186 (char *)&to->to_tsval, sizeof(to->to_tsval));
3187 to->to_tsval = ntohl(to->to_tsval);
3188 bcopy((char *)cp + 6,
3189 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3190 to->to_tsecr = ntohl(to->to_tsecr);
3192 #ifdef TCP_SIGNATURE
3194 * XXX In order to reply to a host which has set the
3195 * TCP_SIGNATURE option in its initial SYN, we have to
3196 * record the fact that the option was observed here
3197 * for the syncache code to perform the correct response.
3199 case TCPOPT_SIGNATURE:
3200 if (optlen != TCPOLEN_SIGNATURE)
3202 to->to_flags |= TOF_SIGNATURE;
3203 to->to_signature = cp + 2;
3206 case TCPOPT_SACK_PERMITTED:
3207 if (optlen != TCPOLEN_SACK_PERMITTED)
3209 if (!(flags & TO_SYN))
3213 to->to_flags |= TOF_SACKPERM;
3216 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3220 to->to_flags |= TOF_SACK;
3221 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3222 to->to_sacks = cp + 2;
3223 TCPSTAT_INC(tcps_sack_rcv_blocks);
3232 * Pull out of band byte out of a segment so
3233 * it doesn't appear in the user's data queue.
3234 * It is still reflected in the segment length for
3235 * sequencing purposes.
3238 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3241 int cnt = off + th->th_urp - 1;
3244 if (m->m_len > cnt) {
3245 char *cp = mtod(m, caddr_t) + cnt;
3246 struct tcpcb *tp = sototcpcb(so);
3248 INP_WLOCK_ASSERT(tp->t_inpcb);
3251 tp->t_oobflags |= TCPOOB_HAVEDATA;
3252 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3254 if (m->m_flags & M_PKTHDR)
3263 panic("tcp_pulloutofband");
3267 * Collect new round-trip time estimate
3268 * and update averages and current timeout.
3271 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3275 INP_WLOCK_ASSERT(tp->t_inpcb);
3277 TCPSTAT_INC(tcps_rttupdated);
3279 if (tp->t_srtt != 0) {
3281 * srtt is stored as fixed point with 5 bits after the
3282 * binary point (i.e., scaled by 8). The following magic
3283 * is equivalent to the smoothing algorithm in rfc793 with
3284 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3285 * point). Adjust rtt to origin 0.
3287 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3288 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3290 if ((tp->t_srtt += delta) <= 0)
3294 * We accumulate a smoothed rtt variance (actually, a
3295 * smoothed mean difference), then set the retransmit
3296 * timer to smoothed rtt + 4 times the smoothed variance.
3297 * rttvar is stored as fixed point with 4 bits after the
3298 * binary point (scaled by 16). The following is
3299 * equivalent to rfc793 smoothing with an alpha of .75
3300 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3301 * rfc793's wired-in beta.
3305 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3306 if ((tp->t_rttvar += delta) <= 0)
3308 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3309 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3312 * No rtt measurement yet - use the unsmoothed rtt.
3313 * Set the variance to half the rtt (so our first
3314 * retransmit happens at 3*rtt).
3316 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3317 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3318 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3324 * the retransmit should happen at rtt + 4 * rttvar.
3325 * Because of the way we do the smoothing, srtt and rttvar
3326 * will each average +1/2 tick of bias. When we compute
3327 * the retransmit timer, we want 1/2 tick of rounding and
3328 * 1 extra tick because of +-1/2 tick uncertainty in the
3329 * firing of the timer. The bias will give us exactly the
3330 * 1.5 tick we need. But, because the bias is
3331 * statistical, we have to test that we don't drop below
3332 * the minimum feasible timer (which is 2 ticks).
3334 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3335 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3338 * We received an ack for a packet that wasn't retransmitted;
3339 * it is probably safe to discard any error indications we've
3340 * received recently. This isn't quite right, but close enough
3341 * for now (a route might have failed after we sent a segment,
3342 * and the return path might not be symmetrical).
3344 tp->t_softerror = 0;
3348 * Determine a reasonable value for maxseg size.
3349 * If the route is known, check route for mtu.
3350 * If none, use an mss that can be handled on the outgoing interface
3351 * without forcing IP to fragment. If no route is found, route has no mtu,
3352 * or the destination isn't local, use a default, hopefully conservative
3353 * size (usually 512 or the default IP max size, but no more than the mtu
3354 * of the interface), as we can't discover anything about intervening
3355 * gateways or networks. We also initialize the congestion/slow start
3356 * window to be a single segment if the destination isn't local.
3357 * While looking at the routing entry, we also initialize other path-dependent
3358 * parameters from pre-set or cached values in the routing entry.
3360 * Also take into account the space needed for options that we
3361 * send regularly. Make maxseg shorter by that amount to assure
3362 * that we can send maxseg amount of data even when the options
3363 * are present. Store the upper limit of the length of options plus
3366 * NOTE that this routine is only called when we process an incoming
3367 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3368 * settings are handled in tcp_mssopt().
3371 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3372 struct hc_metrics_lite *metricptr, int *mtuflags)
3376 struct inpcb *inp = tp->t_inpcb;
3377 struct hc_metrics_lite metrics;
3380 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3381 size_t min_protoh = isipv6 ?
3382 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3383 sizeof (struct tcpiphdr);
3385 const size_t min_protoh = sizeof(struct tcpiphdr);
3388 INP_WLOCK_ASSERT(tp->t_inpcb);
3390 if (mtuoffer != -1) {
3391 KASSERT(offer == -1, ("%s: conflict", __func__));
3392 offer = mtuoffer - min_protoh;
3399 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags);
3400 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3403 #if defined(INET) && defined(INET6)
3408 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags);
3409 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3414 * No route to sender, stay with default mss and return.
3418 * In case we return early we need to initialize metrics
3419 * to a defined state as tcp_hc_get() would do for us
3420 * if there was no cache hit.
3422 if (metricptr != NULL)
3423 bzero(metricptr, sizeof(struct hc_metrics_lite));
3427 /* What have we got? */
3431 * Offer == 0 means that there was no MSS on the SYN
3432 * segment, in this case we use tcp_mssdflt as
3433 * already assigned to t_maxopd above.
3435 offer = tp->t_maxopd;
3440 * Offer == -1 means that we didn't receive SYN yet.
3446 * Prevent DoS attack with too small MSS. Round up
3447 * to at least minmss.
3449 offer = max(offer, V_tcp_minmss);
3453 * rmx information is now retrieved from tcp_hostcache.
3455 tcp_hc_get(&inp->inp_inc, &metrics);
3456 if (metricptr != NULL)
3457 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3460 * If there's a discovered mtu int tcp hostcache, use it
3461 * else, use the link mtu.
3463 if (metrics.rmx_mtu)
3464 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3468 mss = maxmtu - min_protoh;
3469 if (!V_path_mtu_discovery &&
3470 !in6_localaddr(&inp->in6p_faddr))
3471 mss = min(mss, V_tcp_v6mssdflt);
3474 #if defined(INET) && defined(INET6)
3479 mss = maxmtu - min_protoh;
3480 if (!V_path_mtu_discovery &&
3481 !in_localaddr(inp->inp_faddr))
3482 mss = min(mss, V_tcp_mssdflt);
3486 * XXX - The above conditional (mss = maxmtu - min_protoh)
3487 * probably violates the TCP spec.
3488 * The problem is that, since we don't know the
3489 * other end's MSS, we are supposed to use a conservative
3490 * default. But, if we do that, then MTU discovery will
3491 * never actually take place, because the conservative
3492 * default is much less than the MTUs typically seen
3493 * on the Internet today. For the moment, we'll sweep
3494 * this under the carpet.
3496 * The conservative default might not actually be a problem
3497 * if the only case this occurs is when sending an initial
3498 * SYN with options and data to a host we've never talked
3499 * to before. Then, they will reply with an MSS value which
3500 * will get recorded and the new parameters should get
3501 * recomputed. For Further Study.
3504 mss = min(mss, offer);
3507 * Sanity check: make sure that maxopd will be large
3508 * enough to allow some data on segments even if the
3509 * all the option space is used (40bytes). Otherwise
3510 * funny things may happen in tcp_output.
3515 * maxopd stores the maximum length of data AND options
3516 * in a segment; maxseg is the amount of data in a normal
3517 * segment. We need to store this value (maxopd) apart
3518 * from maxseg, because now every segment carries options
3519 * and thus we normally have somewhat less data in segments.
3524 * origoffer==-1 indicates that no segments were received yet.
3525 * In this case we just guess.
3527 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3529 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3530 mss -= TCPOLEN_TSTAMP_APPA;
3536 tcp_mss(struct tcpcb *tp, int offer)
3542 struct hc_metrics_lite metrics;
3545 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3547 tcp_mss_update(tp, offer, -1, &metrics, &mtuflags);
3553 * If there's a pipesize, change the socket buffer to that size,
3554 * don't change if sb_hiwat is different than default (then it
3555 * has been changed on purpose with setsockopt).
3556 * Make the socket buffers an integral number of mss units;
3557 * if the mss is larger than the socket buffer, decrease the mss.
3559 so = inp->inp_socket;
3560 SOCKBUF_LOCK(&so->so_snd);
3561 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3562 bufsize = metrics.rmx_sendpipe;
3564 bufsize = so->so_snd.sb_hiwat;
3568 bufsize = roundup(bufsize, mss);
3569 if (bufsize > sb_max)
3571 if (bufsize > so->so_snd.sb_hiwat)
3572 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3574 SOCKBUF_UNLOCK(&so->so_snd);
3577 SOCKBUF_LOCK(&so->so_rcv);
3578 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3579 bufsize = metrics.rmx_recvpipe;
3581 bufsize = so->so_rcv.sb_hiwat;
3582 if (bufsize > mss) {
3583 bufsize = roundup(bufsize, mss);
3584 if (bufsize > sb_max)
3586 if (bufsize > so->so_rcv.sb_hiwat)
3587 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3589 SOCKBUF_UNLOCK(&so->so_rcv);
3591 /* Check the interface for TSO capabilities. */
3592 if (mtuflags & CSUM_TSO)
3593 tp->t_flags |= TF_TSO;
3597 * Determine the MSS option to send on an outgoing SYN.
3600 tcp_mssopt(struct in_conninfo *inc)
3607 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3610 if (inc->inc_flags & INC_ISIPV6) {
3611 mss = V_tcp_v6mssdflt;
3612 maxmtu = tcp_maxmtu6(inc, NULL);
3613 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3616 #if defined(INET) && defined(INET6)
3621 mss = V_tcp_mssdflt;
3622 maxmtu = tcp_maxmtu(inc, NULL);
3623 min_protoh = sizeof(struct tcpiphdr);
3626 #if defined(INET6) || defined(INET)
3627 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3630 if (maxmtu && thcmtu)
3631 mss = min(maxmtu, thcmtu) - min_protoh;
3632 else if (maxmtu || thcmtu)
3633 mss = max(maxmtu, thcmtu) - min_protoh;
3640 * On a partial ack arrives, force the retransmission of the
3641 * next unacknowledged segment. Do not clear tp->t_dupacks.
3642 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3646 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3648 tcp_seq onxt = tp->snd_nxt;
3649 u_long ocwnd = tp->snd_cwnd;
3651 INP_WLOCK_ASSERT(tp->t_inpcb);
3653 tcp_timer_activate(tp, TT_REXMT, 0);
3655 tp->snd_nxt = th->th_ack;
3657 * Set snd_cwnd to one segment beyond acknowledged offset.
3658 * (tp->snd_una has not yet been updated when this function is called.)
3660 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3661 tp->t_flags |= TF_ACKNOW;
3662 (void) tcp_output(tp);
3663 tp->snd_cwnd = ocwnd;
3664 if (SEQ_GT(onxt, tp->snd_nxt))
3667 * Partial window deflation. Relies on fact that tp->snd_una
3670 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3671 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3674 tp->snd_cwnd += tp->t_maxseg;