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 int tcp_log_in_vain = 0;
124 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
126 "Log all incoming TCP segments to closed ports");
128 VNET_DEFINE(int, blackhole) = 0;
129 #define V_blackhole VNET(blackhole)
130 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
131 &VNET_NAME(blackhole), 0,
132 "Do not send RST on segments to closed ports");
134 VNET_DEFINE(int, tcp_delack_enabled) = 1;
135 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
136 &VNET_NAME(tcp_delack_enabled), 0,
137 "Delay ACK to try and piggyback it onto a data packet");
139 VNET_DEFINE(int, drop_synfin) = 0;
140 #define V_drop_synfin VNET(drop_synfin)
141 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
142 &VNET_NAME(drop_synfin), 0,
143 "Drop TCP packets with SYN+FIN set");
145 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
146 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
147 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
148 &VNET_NAME(tcp_do_rfc3042), 0,
149 "Enable RFC 3042 (Limited Transmit)");
151 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
152 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
153 &VNET_NAME(tcp_do_rfc3390), 0,
154 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
156 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0,
157 "Experimental TCP extensions");
159 VNET_DEFINE(int, tcp_do_initcwnd10) = 1;
160 SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW,
161 &VNET_NAME(tcp_do_initcwnd10), 0,
162 "Enable draft-ietf-tcpm-initcwnd-05 (Increasing initial CWND to 10)");
164 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
165 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW,
166 &VNET_NAME(tcp_do_rfc3465), 0,
167 "Enable RFC 3465 (Appropriate Byte Counting)");
169 VNET_DEFINE(int, tcp_abc_l_var) = 2;
170 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW,
171 &VNET_NAME(tcp_abc_l_var), 2,
172 "Cap the max cwnd increment during slow-start to this number of segments");
174 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
176 VNET_DEFINE(int, tcp_do_ecn) = 0;
177 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW,
178 &VNET_NAME(tcp_do_ecn), 0,
181 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
182 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW,
183 &VNET_NAME(tcp_ecn_maxretries), 0,
184 "Max retries before giving up on ECN");
186 VNET_DEFINE(int, tcp_insecure_rst) = 0;
187 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
188 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
189 &VNET_NAME(tcp_insecure_rst), 0,
190 "Follow the old (insecure) criteria for accepting RST packets");
192 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
193 #define V_tcp_recvspace VNET(tcp_recvspace)
194 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
195 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
197 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
198 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
199 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
200 &VNET_NAME(tcp_do_autorcvbuf), 0,
201 "Enable automatic receive buffer sizing");
203 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
204 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
205 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
206 &VNET_NAME(tcp_autorcvbuf_inc), 0,
207 "Incrementor step size of automatic receive buffer");
209 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
210 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
211 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
212 &VNET_NAME(tcp_autorcvbuf_max), 0,
213 "Max size of automatic receive buffer");
215 VNET_DEFINE(struct inpcbhead, tcb);
216 #define tcb6 tcb /* for KAME src sync over BSD*'s */
217 VNET_DEFINE(struct inpcbinfo, tcbinfo);
219 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
220 static void tcp_do_segment(struct mbuf *, struct tcphdr *,
221 struct socket *, struct tcpcb *, int, int, uint8_t,
223 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
224 struct tcpcb *, int, int);
225 static void tcp_pulloutofband(struct socket *,
226 struct tcphdr *, struct mbuf *, int);
227 static void tcp_xmit_timer(struct tcpcb *, int);
228 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
229 static void inline tcp_fields_to_host(struct tcphdr *);
231 static void inline tcp_fields_to_net(struct tcphdr *);
232 static int inline tcp_signature_verify_input(struct mbuf *, int, int,
233 int, struct tcpopt *, struct tcphdr *, u_int);
235 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th,
237 static void inline cc_conn_init(struct tcpcb *tp);
238 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th);
239 static void inline hhook_run_tcp_est_in(struct tcpcb *tp,
240 struct tcphdr *th, struct tcpopt *to);
243 * TCP statistics are stored in an "array" of counter(9)s.
245 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
246 VNET_PCPUSTAT_SYSINIT(tcpstat);
247 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
248 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
251 VNET_PCPUSTAT_SYSUNINIT(tcpstat);
254 * Kernel module interface for updating tcpstat. The argument is an index
255 * into tcpstat treated as an array.
258 kmod_tcpstat_inc(int statnum)
261 counter_u64_add(VNET(tcpstat)[statnum], 1);
265 * Wrapper for the TCP established input helper hook.
268 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
270 struct tcp_hhook_data hhook_data;
272 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
277 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
283 * CC wrapper hook functions
286 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type)
288 INP_WLOCK_ASSERT(tp->t_inpcb);
290 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
291 if (tp->snd_cwnd <= tp->snd_wnd)
292 tp->ccv->flags |= CCF_CWND_LIMITED;
294 tp->ccv->flags &= ~CCF_CWND_LIMITED;
296 if (type == CC_ACK) {
297 if (tp->snd_cwnd > tp->snd_ssthresh) {
298 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
299 V_tcp_abc_l_var * tp->t_maxseg);
300 if (tp->t_bytes_acked >= tp->snd_cwnd) {
301 tp->t_bytes_acked -= tp->snd_cwnd;
302 tp->ccv->flags |= CCF_ABC_SENTAWND;
305 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
306 tp->t_bytes_acked = 0;
310 if (CC_ALGO(tp)->ack_received != NULL) {
311 /* XXXLAS: Find a way to live without this */
312 tp->ccv->curack = th->th_ack;
313 CC_ALGO(tp)->ack_received(tp->ccv, type);
318 cc_conn_init(struct tcpcb *tp)
320 struct hc_metrics_lite metrics;
321 struct inpcb *inp = tp->t_inpcb;
324 INP_WLOCK_ASSERT(tp->t_inpcb);
326 tcp_hc_get(&inp->inp_inc, &metrics);
328 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
330 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
331 TCPSTAT_INC(tcps_usedrtt);
332 if (metrics.rmx_rttvar) {
333 tp->t_rttvar = metrics.rmx_rttvar;
334 TCPSTAT_INC(tcps_usedrttvar);
336 /* default variation is +- 1 rtt */
338 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
340 TCPT_RANGESET(tp->t_rxtcur,
341 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
342 tp->t_rttmin, TCPTV_REXMTMAX);
344 if (metrics.rmx_ssthresh) {
346 * There's some sort of gateway or interface
347 * buffer limit on the path. Use this to set
348 * the slow start threshhold, but set the
349 * threshold to no less than 2*mss.
351 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh);
352 TCPSTAT_INC(tcps_usedssthresh);
356 * Set the initial slow-start flight size.
358 * RFC5681 Section 3.1 specifies the default conservative values.
359 * RFC3390 specifies slightly more aggressive values.
360 * Draft-ietf-tcpm-initcwnd-05 increases it to ten segments.
362 * If a SYN or SYN/ACK was lost and retransmitted, we have to
363 * reduce the initial CWND to one segment as congestion is likely
364 * requiring us to be cautious.
366 if (tp->snd_cwnd == 1)
367 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */
368 else if (V_tcp_do_initcwnd10)
369 tp->snd_cwnd = min(10 * tp->t_maxseg,
370 max(2 * tp->t_maxseg, 14600));
371 else if (V_tcp_do_rfc3390)
372 tp->snd_cwnd = min(4 * tp->t_maxseg,
373 max(2 * tp->t_maxseg, 4380));
375 /* Per RFC5681 Section 3.1 */
376 if (tp->t_maxseg > 2190)
377 tp->snd_cwnd = 2 * tp->t_maxseg;
378 else if (tp->t_maxseg > 1095)
379 tp->snd_cwnd = 3 * tp->t_maxseg;
381 tp->snd_cwnd = 4 * tp->t_maxseg;
384 if (CC_ALGO(tp)->conn_init != NULL)
385 CC_ALGO(tp)->conn_init(tp->ccv);
389 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
391 INP_WLOCK_ASSERT(tp->t_inpcb);
395 if (!IN_FASTRECOVERY(tp->t_flags)) {
396 tp->snd_recover = tp->snd_max;
397 if (tp->t_flags & TF_ECN_PERMIT)
398 tp->t_flags |= TF_ECN_SND_CWR;
402 if (!IN_CONGRECOVERY(tp->t_flags)) {
403 TCPSTAT_INC(tcps_ecn_rcwnd);
404 tp->snd_recover = tp->snd_max;
405 if (tp->t_flags & TF_ECN_PERMIT)
406 tp->t_flags |= TF_ECN_SND_CWR;
411 tp->t_bytes_acked = 0;
412 EXIT_RECOVERY(tp->t_flags);
413 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
414 tp->t_maxseg) * tp->t_maxseg;
415 tp->snd_cwnd = tp->t_maxseg;
418 TCPSTAT_INC(tcps_sndrexmitbad);
419 /* RTO was unnecessary, so reset everything. */
420 tp->snd_cwnd = tp->snd_cwnd_prev;
421 tp->snd_ssthresh = tp->snd_ssthresh_prev;
422 tp->snd_recover = tp->snd_recover_prev;
423 if (tp->t_flags & TF_WASFRECOVERY)
424 ENTER_FASTRECOVERY(tp->t_flags);
425 if (tp->t_flags & TF_WASCRECOVERY)
426 ENTER_CONGRECOVERY(tp->t_flags);
427 tp->snd_nxt = tp->snd_max;
428 tp->t_flags &= ~TF_PREVVALID;
433 if (CC_ALGO(tp)->cong_signal != NULL) {
435 tp->ccv->curack = th->th_ack;
436 CC_ALGO(tp)->cong_signal(tp->ccv, type);
441 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
443 INP_WLOCK_ASSERT(tp->t_inpcb);
445 /* XXXLAS: KASSERT that we're in recovery? */
447 if (CC_ALGO(tp)->post_recovery != NULL) {
448 tp->ccv->curack = th->th_ack;
449 CC_ALGO(tp)->post_recovery(tp->ccv);
451 /* XXXLAS: EXIT_RECOVERY ? */
452 tp->t_bytes_acked = 0;
456 tcp_fields_to_host(struct tcphdr *th)
459 th->th_seq = ntohl(th->th_seq);
460 th->th_ack = ntohl(th->th_ack);
461 th->th_win = ntohs(th->th_win);
462 th->th_urp = ntohs(th->th_urp);
467 tcp_fields_to_net(struct tcphdr *th)
470 th->th_seq = htonl(th->th_seq);
471 th->th_ack = htonl(th->th_ack);
472 th->th_win = htons(th->th_win);
473 th->th_urp = htons(th->th_urp);
477 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
478 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
482 tcp_fields_to_net(th);
483 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
484 tcp_fields_to_host(th);
489 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
491 #define ND6_HINT(tp) \
493 if ((tp) && (tp)->t_inpcb && \
494 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
495 nd6_nud_hint(NULL, NULL, 0); \
502 * Indicate whether this ack should be delayed. We can delay the ack if
503 * - there is no delayed ack timer in progress and
504 * - our last ack wasn't a 0-sized window. We never want to delay
505 * the ack that opens up a 0-sized window and
506 * - delayed acks are enabled or
507 * - this is a half-synchronized T/TCP connection.
509 #define DELAY_ACK(tp) \
510 ((!tcp_timer_active(tp, TT_DELACK) && \
511 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
512 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
515 * TCP input handling is split into multiple parts:
516 * tcp6_input is a thin wrapper around tcp_input for the extended
517 * ip6_protox[] call format in ip6_input
518 * tcp_input handles primary segment validation, inpcb lookup and
519 * SYN processing on listen sockets
520 * tcp_do_segment processes the ACK and text of the segment for
521 * establishing, established and closing connections
525 tcp6_input(struct mbuf **mp, int *offp, int proto)
527 struct mbuf *m = *mp;
528 struct in6_ifaddr *ia6;
530 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
533 * draft-itojun-ipv6-tcp-to-anycast
534 * better place to put this in?
536 ia6 = ip6_getdstifaddr(m);
537 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
540 ifa_free(&ia6->ia_ifa);
541 ip6 = mtod(m, struct ip6_hdr *);
542 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
543 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
547 ifa_free(&ia6->ia_ifa);
555 tcp_input(struct mbuf *m, int off0)
557 struct tcphdr *th = NULL;
558 struct ip *ip = NULL;
559 struct inpcb *inp = NULL;
560 struct tcpcb *tp = NULL;
561 struct socket *so = NULL;
570 int rstreason = 0; /* For badport_bandlim accounting purposes */
572 uint8_t sig_checked = 0;
575 struct m_tag *fwd_tag = NULL;
577 struct ip6_hdr *ip6 = NULL;
580 const void *ip6 = NULL;
582 struct tcpopt to; /* options in this segment */
583 char *s = NULL; /* address and port logging */
585 #define TI_UNLOCKED 1
590 * The size of tcp_saveipgen must be the size of the max ip header,
593 u_char tcp_saveipgen[IP6_HDR_LEN];
594 struct tcphdr tcp_savetcp;
599 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
603 TCPSTAT_INC(tcps_rcvtotal);
607 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
609 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
610 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
612 TCPSTAT_INC(tcps_rcvshort);
617 ip6 = mtod(m, struct ip6_hdr *);
618 th = (struct tcphdr *)((caddr_t)ip6 + off0);
619 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
620 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
621 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
622 th->th_sum = m->m_pkthdr.csum_data;
624 th->th_sum = in6_cksum_pseudo(ip6, tlen,
625 IPPROTO_TCP, m->m_pkthdr.csum_data);
626 th->th_sum ^= 0xffff;
628 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
630 TCPSTAT_INC(tcps_rcvbadsum);
635 * Be proactive about unspecified IPv6 address in source.
636 * As we use all-zero to indicate unbounded/unconnected pcb,
637 * unspecified IPv6 address can be used to confuse us.
639 * Note that packets with unspecified IPv6 destination is
640 * already dropped in ip6_input.
642 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
648 #if defined(INET) && defined(INET6)
654 * Get IP and TCP header together in first mbuf.
655 * Note: IP leaves IP header in first mbuf.
657 if (off0 > sizeof (struct ip)) {
659 off0 = sizeof(struct ip);
661 if (m->m_len < sizeof (struct tcpiphdr)) {
662 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
664 TCPSTAT_INC(tcps_rcvshort);
668 ip = mtod(m, struct ip *);
669 th = (struct tcphdr *)((caddr_t)ip + off0);
670 tlen = ntohs(ip->ip_len) - off0;
672 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
673 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
674 th->th_sum = m->m_pkthdr.csum_data;
676 th->th_sum = in_pseudo(ip->ip_src.s_addr,
678 htonl(m->m_pkthdr.csum_data + tlen +
680 th->th_sum ^= 0xffff;
682 struct ipovly *ipov = (struct ipovly *)ip;
685 * Checksum extended TCP header and data.
688 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
689 ipov->ih_len = htons(tlen);
690 th->th_sum = in_cksum(m, len);
693 TCPSTAT_INC(tcps_rcvbadsum);
696 /* Re-initialization for later version check */
697 ip->ip_v = IPVERSION;
703 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
705 #if defined(INET) && defined(INET6)
713 * Check that TCP offset makes sense,
714 * pull out TCP options and adjust length. XXX
716 off = th->th_off << 2;
717 if (off < sizeof (struct tcphdr) || off > tlen) {
718 TCPSTAT_INC(tcps_rcvbadoff);
721 tlen -= off; /* tlen is used instead of ti->ti_len */
722 if (off > sizeof (struct tcphdr)) {
725 IP6_EXTHDR_CHECK(m, off0, off, );
726 ip6 = mtod(m, struct ip6_hdr *);
727 th = (struct tcphdr *)((caddr_t)ip6 + off0);
730 #if defined(INET) && defined(INET6)
735 if (m->m_len < sizeof(struct ip) + off) {
736 if ((m = m_pullup(m, sizeof (struct ip) + off))
738 TCPSTAT_INC(tcps_rcvshort);
741 ip = mtod(m, struct ip *);
742 th = (struct tcphdr *)((caddr_t)ip + off0);
746 optlen = off - sizeof (struct tcphdr);
747 optp = (u_char *)(th + 1);
749 thflags = th->th_flags;
752 * Convert TCP protocol specific fields to host format.
754 tcp_fields_to_host(th);
757 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
759 drop_hdrlen = off0 + off;
762 * Locate pcb for segment; if we're likely to add or remove a
763 * connection then first acquire pcbinfo lock. There are two cases
764 * where we might discover later we need a write lock despite the
765 * flags: ACKs moving a connection out of the syncache, and ACKs for
766 * a connection in TIMEWAIT.
768 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) {
769 INP_INFO_WLOCK(&V_tcbinfo);
770 ti_locked = TI_WLOCKED;
772 ti_locked = TI_UNLOCKED;
775 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
779 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
781 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
784 #if defined(INET) && !defined(INET6)
785 (m->m_flags & M_IP_NEXTHOP)
788 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
792 if (ti_locked == TI_WLOCKED) {
793 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
795 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
799 if (isipv6 && fwd_tag != NULL) {
800 struct sockaddr_in6 *next_hop6;
802 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
804 * Transparently forwarded. Pretend to be the destination.
805 * Already got one like this?
807 inp = in6_pcblookup_mbuf(&V_tcbinfo,
808 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
809 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
812 * It's new. Try to find the ambushing socket.
813 * Because we've rewritten the destination address,
814 * any hardware-generated hash is ignored.
816 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
817 th->th_sport, &next_hop6->sin6_addr,
818 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
819 th->th_dport, INPLOOKUP_WILDCARD |
820 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
823 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
824 th->th_sport, &ip6->ip6_dst, th->th_dport,
825 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
826 m->m_pkthdr.rcvif, m);
829 #if defined(INET6) && defined(INET)
833 if (fwd_tag != NULL) {
834 struct sockaddr_in *next_hop;
836 next_hop = (struct sockaddr_in *)(fwd_tag+1);
838 * Transparently forwarded. Pretend to be the destination.
839 * already got one like this?
841 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
842 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
843 m->m_pkthdr.rcvif, m);
846 * It's new. Try to find the ambushing socket.
847 * Because we've rewritten the destination address,
848 * any hardware-generated hash is ignored.
850 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
851 th->th_sport, next_hop->sin_addr,
852 next_hop->sin_port ? ntohs(next_hop->sin_port) :
853 th->th_dport, INPLOOKUP_WILDCARD |
854 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
857 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
858 th->th_sport, ip->ip_dst, th->th_dport,
859 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
860 m->m_pkthdr.rcvif, m);
864 * If the INPCB does not exist then all data in the incoming
865 * segment is discarded and an appropriate RST is sent back.
866 * XXX MRT Send RST using which routing table?
870 * Log communication attempts to ports that are not
873 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
874 tcp_log_in_vain == 2) {
875 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
876 log(LOG_INFO, "%s; %s: Connection attempt "
877 "to closed port\n", s, __func__);
880 * When blackholing do not respond with a RST but
881 * completely ignore the segment and drop it.
883 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
887 rstreason = BANDLIM_RST_CLOSEDPORT;
890 INP_WLOCK_ASSERT(inp);
891 if (!(inp->inp_flags & INP_HW_FLOWID)
892 && (m->m_flags & M_FLOWID)
893 && ((inp->inp_socket == NULL)
894 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) {
895 inp->inp_flags |= INP_HW_FLOWID;
896 inp->inp_flags &= ~INP_SW_FLOWID;
897 inp->inp_flowid = m->m_pkthdr.flowid;
901 if (isipv6 && ipsec6_in_reject(m, inp)) {
902 IPSEC6STAT_INC(ips_in_polvio);
906 if (ipsec4_in_reject(m, inp) != 0) {
907 IPSECSTAT_INC(ips_in_polvio);
913 * Check the minimum TTL for socket.
915 if (inp->inp_ip_minttl != 0) {
917 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
921 if (inp->inp_ip_minttl > ip->ip_ttl)
926 * A previous connection in TIMEWAIT state is supposed to catch stray
927 * or duplicate segments arriving late. If this segment was a
928 * legitimate new connection attempt, the old INPCB gets removed and
929 * we can try again to find a listening socket.
931 * At this point, due to earlier optimism, we may hold only an inpcb
932 * lock, and not the inpcbinfo write lock. If so, we need to try to
933 * acquire it, or if that fails, acquire a reference on the inpcb,
934 * drop all locks, acquire a global write lock, and then re-acquire
935 * the inpcb lock. We may at that point discover that another thread
936 * has tried to free the inpcb, in which case we need to loop back
937 * and try to find a new inpcb to deliver to.
939 * XXXRW: It may be time to rethink timewait locking.
942 if (inp->inp_flags & INP_TIMEWAIT) {
943 if (ti_locked == TI_UNLOCKED) {
944 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
947 INP_INFO_WLOCK(&V_tcbinfo);
948 ti_locked = TI_WLOCKED;
950 if (in_pcbrele_wlocked(inp)) {
955 ti_locked = TI_WLOCKED;
957 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
959 if (thflags & TH_SYN)
960 tcp_dooptions(&to, optp, optlen, TO_SYN);
962 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
964 if (tcp_twcheck(inp, &to, th, m, tlen))
966 INP_INFO_WUNLOCK(&V_tcbinfo);
970 * The TCPCB may no longer exist if the connection is winding
971 * down or it is in the CLOSED state. Either way we drop the
972 * segment and send an appropriate response.
975 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
976 rstreason = BANDLIM_RST_CLOSEDPORT;
981 if (tp->t_flags & TF_TOE) {
982 tcp_offload_input(tp, m);
983 m = NULL; /* consumed by the TOE driver */
989 * We've identified a valid inpcb, but it could be that we need an
990 * inpcbinfo write lock but don't hold it. In this case, attempt to
991 * acquire using the same strategy as the TIMEWAIT case above. If we
992 * relock, we have to jump back to 'relocked' as the connection might
993 * now be in TIMEWAIT.
996 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0)
997 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
999 if (tp->t_state != TCPS_ESTABLISHED) {
1000 if (ti_locked == TI_UNLOCKED) {
1001 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
1004 INP_INFO_WLOCK(&V_tcbinfo);
1005 ti_locked = TI_WLOCKED;
1007 if (in_pcbrele_wlocked(inp)) {
1013 ti_locked = TI_WLOCKED;
1015 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1019 INP_WLOCK_ASSERT(inp);
1020 if (mac_inpcb_check_deliver(inp, m))
1023 so = inp->inp_socket;
1024 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1026 if (so->so_options & SO_DEBUG) {
1027 ostate = tp->t_state;
1030 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1033 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1036 #endif /* TCPDEBUG */
1038 * When the socket is accepting connections (the INPCB is in LISTEN
1039 * state) we look into the SYN cache if this is a new connection
1040 * attempt or the completion of a previous one. Because listen
1041 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be
1042 * held in this case.
1044 if (so->so_options & SO_ACCEPTCONN) {
1045 struct in_conninfo inc;
1047 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
1048 "tp not listening", __func__));
1049 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1051 bzero(&inc, sizeof(inc));
1054 inc.inc_flags |= INC_ISIPV6;
1055 inc.inc6_faddr = ip6->ip6_src;
1056 inc.inc6_laddr = ip6->ip6_dst;
1060 inc.inc_faddr = ip->ip_src;
1061 inc.inc_laddr = ip->ip_dst;
1063 inc.inc_fport = th->th_sport;
1064 inc.inc_lport = th->th_dport;
1065 inc.inc_fibnum = so->so_fibnum;
1068 * Check for an existing connection attempt in syncache if
1069 * the flag is only ACK. A successful lookup creates a new
1070 * socket appended to the listen queue in SYN_RECEIVED state.
1072 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1074 * Parse the TCP options here because
1075 * syncookies need access to the reflected
1078 tcp_dooptions(&to, optp, optlen, 0);
1080 * NB: syncache_expand() doesn't unlock
1081 * inp and tcpinfo locks.
1083 if (!syncache_expand(&inc, &to, th, &so, m)) {
1085 * No syncache entry or ACK was not
1086 * for our SYN/ACK. Send a RST.
1087 * NB: syncache did its own logging
1088 * of the failure cause.
1090 rstreason = BANDLIM_RST_OPENPORT;
1095 * We completed the 3-way handshake
1096 * but could not allocate a socket
1097 * either due to memory shortage,
1098 * listen queue length limits or
1099 * global socket limits. Send RST
1100 * or wait and have the remote end
1101 * retransmit the ACK for another
1104 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1105 log(LOG_DEBUG, "%s; %s: Listen socket: "
1106 "Socket allocation failed due to "
1107 "limits or memory shortage, %s\n",
1109 V_tcp_sc_rst_sock_fail ?
1110 "sending RST" : "try again");
1111 if (V_tcp_sc_rst_sock_fail) {
1112 rstreason = BANDLIM_UNLIMITED;
1118 * Socket is created in state SYN_RECEIVED.
1119 * Unlock the listen socket, lock the newly
1120 * created socket and update the tp variable.
1122 INP_WUNLOCK(inp); /* listen socket */
1123 inp = sotoinpcb(so);
1124 INP_WLOCK(inp); /* new connection */
1125 tp = intotcpcb(inp);
1126 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1127 ("%s: ", __func__));
1128 #ifdef TCP_SIGNATURE
1129 if (sig_checked == 0) {
1130 tcp_dooptions(&to, optp, optlen,
1131 (thflags & TH_SYN) ? TO_SYN : 0);
1132 if (!tcp_signature_verify_input(m, off0, tlen,
1133 optlen, &to, th, tp->t_flags)) {
1136 * In SYN_SENT state if it receives an
1137 * RST, it is allowed for further
1140 if ((thflags & TH_RST) == 0 ||
1141 (tp->t_state == TCPS_SYN_SENT) == 0)
1149 * Process the segment and the data it
1150 * contains. tcp_do_segment() consumes
1151 * the mbuf chain and unlocks the inpcb.
1153 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1155 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1159 * Segment flag validation for new connection attempts:
1161 * Our (SYN|ACK) response was rejected.
1162 * Check with syncache and remove entry to prevent
1165 * NB: syncache_chkrst does its own logging of failure
1168 if (thflags & TH_RST) {
1169 syncache_chkrst(&inc, th);
1173 * We can't do anything without SYN.
1175 if ((thflags & TH_SYN) == 0) {
1176 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1177 log(LOG_DEBUG, "%s; %s: Listen socket: "
1178 "SYN is missing, segment ignored\n",
1180 TCPSTAT_INC(tcps_badsyn);
1184 * (SYN|ACK) is bogus on a listen socket.
1186 if (thflags & TH_ACK) {
1187 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1188 log(LOG_DEBUG, "%s; %s: Listen socket: "
1189 "SYN|ACK invalid, segment rejected\n",
1191 syncache_badack(&inc); /* XXX: Not needed! */
1192 TCPSTAT_INC(tcps_badsyn);
1193 rstreason = BANDLIM_RST_OPENPORT;
1197 * If the drop_synfin option is enabled, drop all
1198 * segments with both the SYN and FIN bits set.
1199 * This prevents e.g. nmap from identifying the
1201 * XXX: Poor reasoning. nmap has other methods
1202 * and is constantly refining its stack detection
1204 * XXX: This is a violation of the TCP specification
1205 * and was used by RFC1644.
1207 if ((thflags & TH_FIN) && V_drop_synfin) {
1208 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1209 log(LOG_DEBUG, "%s; %s: Listen socket: "
1210 "SYN|FIN segment ignored (based on "
1211 "sysctl setting)\n", s, __func__);
1212 TCPSTAT_INC(tcps_badsyn);
1216 * Segment's flags are (SYN) or (SYN|FIN).
1218 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1219 * as they do not affect the state of the TCP FSM.
1220 * The data pointed to by TH_URG and th_urp is ignored.
1222 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1223 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1224 KASSERT(thflags & (TH_SYN),
1225 ("%s: Listen socket: TH_SYN not set", __func__));
1228 * If deprecated address is forbidden,
1229 * we do not accept SYN to deprecated interface
1230 * address to prevent any new inbound connection from
1231 * getting established.
1232 * When we do not accept SYN, we send a TCP RST,
1233 * with deprecated source address (instead of dropping
1234 * it). We compromise it as it is much better for peer
1235 * to send a RST, and RST will be the final packet
1238 * If we do not forbid deprecated addresses, we accept
1239 * the SYN packet. RFC2462 does not suggest dropping
1241 * If we decipher RFC2462 5.5.4, it says like this:
1242 * 1. use of deprecated addr with existing
1243 * communication is okay - "SHOULD continue to be
1245 * 2. use of it with new communication:
1246 * (2a) "SHOULD NOT be used if alternate address
1247 * with sufficient scope is available"
1248 * (2b) nothing mentioned otherwise.
1249 * Here we fall into (2b) case as we have no choice in
1250 * our source address selection - we must obey the peer.
1252 * The wording in RFC2462 is confusing, and there are
1253 * multiple description text for deprecated address
1254 * handling - worse, they are not exactly the same.
1255 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1257 if (isipv6 && !V_ip6_use_deprecated) {
1258 struct in6_ifaddr *ia6;
1260 ia6 = ip6_getdstifaddr(m);
1262 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1263 ifa_free(&ia6->ia_ifa);
1264 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1265 log(LOG_DEBUG, "%s; %s: Listen socket: "
1266 "Connection attempt to deprecated "
1267 "IPv6 address rejected\n",
1269 rstreason = BANDLIM_RST_OPENPORT;
1273 ifa_free(&ia6->ia_ifa);
1277 * Basic sanity checks on incoming SYN requests:
1278 * Don't respond if the destination is a link layer
1279 * broadcast according to RFC1122 4.2.3.10, p. 104.
1280 * If it is from this socket it must be forged.
1281 * Don't respond if the source or destination is a
1282 * global or subnet broad- or multicast address.
1283 * Note that it is quite possible to receive unicast
1284 * link-layer packets with a broadcast IP address. Use
1285 * in_broadcast() to find them.
1287 if (m->m_flags & (M_BCAST|M_MCAST)) {
1288 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1289 log(LOG_DEBUG, "%s; %s: Listen socket: "
1290 "Connection attempt from broad- or multicast "
1291 "link layer address ignored\n", s, __func__);
1296 if (th->th_dport == th->th_sport &&
1297 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1298 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1299 log(LOG_DEBUG, "%s; %s: Listen socket: "
1300 "Connection attempt to/from self "
1301 "ignored\n", s, __func__);
1304 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1305 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1306 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1307 log(LOG_DEBUG, "%s; %s: Listen socket: "
1308 "Connection attempt from/to multicast "
1309 "address ignored\n", s, __func__);
1314 #if defined(INET) && defined(INET6)
1319 if (th->th_dport == th->th_sport &&
1320 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1321 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1322 log(LOG_DEBUG, "%s; %s: Listen socket: "
1323 "Connection attempt from/to self "
1324 "ignored\n", s, __func__);
1327 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1328 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1329 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1330 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1331 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1332 log(LOG_DEBUG, "%s; %s: Listen socket: "
1333 "Connection attempt from/to broad- "
1334 "or multicast address ignored\n",
1341 * SYN appears to be valid. Create compressed TCP state
1345 if (so->so_options & SO_DEBUG)
1346 tcp_trace(TA_INPUT, ostate, tp,
1347 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1349 tcp_dooptions(&to, optp, optlen, TO_SYN);
1350 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1352 * Entry added to syncache and mbuf consumed.
1353 * Everything already unlocked by syncache_add().
1355 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1357 } else if (tp->t_state == TCPS_LISTEN) {
1359 * When a listen socket is torn down the SO_ACCEPTCONN
1360 * flag is removed first while connections are drained
1361 * from the accept queue in a unlock/lock cycle of the
1362 * ACCEPT_LOCK, opening a race condition allowing a SYN
1363 * attempt go through unhandled.
1368 #ifdef TCP_SIGNATURE
1369 if (sig_checked == 0) {
1370 tcp_dooptions(&to, optp, optlen,
1371 (thflags & TH_SYN) ? TO_SYN : 0);
1372 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1376 * In SYN_SENT state if it receives an RST, it is
1377 * allowed for further processing.
1379 if ((thflags & TH_RST) == 0 ||
1380 (tp->t_state == TCPS_SYN_SENT) == 0)
1388 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1389 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1390 * the inpcb, and unlocks pcbinfo.
1392 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1393 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1397 if (ti_locked == TI_WLOCKED) {
1398 INP_INFO_WUNLOCK(&V_tcbinfo);
1399 ti_locked = TI_UNLOCKED;
1403 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1404 "ti_locked: %d", __func__, ti_locked));
1405 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1410 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1413 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1414 m = NULL; /* mbuf chain got consumed. */
1418 if (ti_locked == TI_WLOCKED) {
1419 INP_INFO_WUNLOCK(&V_tcbinfo);
1420 ti_locked = TI_UNLOCKED;
1424 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1425 "ti_locked: %d", __func__, ti_locked));
1426 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1434 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1442 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1443 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1446 int thflags, acked, ourfinisacked, needoutput = 0;
1447 int rstreason, todrop, win;
1450 struct in_conninfo *inc;
1455 * The size of tcp_saveipgen must be the size of the max ip header,
1458 u_char tcp_saveipgen[IP6_HDR_LEN];
1459 struct tcphdr tcp_savetcp;
1462 thflags = th->th_flags;
1463 inc = &tp->t_inpcb->inp_inc;
1464 tp->sackhint.last_sack_ack = 0;
1467 * If this is either a state-changing packet or current state isn't
1468 * established, we require a write lock on tcbinfo. Otherwise, we
1469 * allow the tcbinfo to be in either alocked or unlocked, as the
1470 * caller may have unnecessarily acquired a write lock due to a race.
1472 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1473 tp->t_state != TCPS_ESTABLISHED) {
1474 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for "
1475 "SYN/FIN/RST/!EST", __func__, ti_locked));
1476 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1479 if (ti_locked == TI_WLOCKED)
1480 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1482 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1483 "ti_locked: %d", __func__, ti_locked));
1484 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1488 INP_WLOCK_ASSERT(tp->t_inpcb);
1489 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1491 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1495 * Segment received on connection.
1496 * Reset idle time and keep-alive timer.
1497 * XXX: This should be done after segment
1498 * validation to ignore broken/spoofed segs.
1500 tp->t_rcvtime = ticks;
1501 if (TCPS_HAVEESTABLISHED(tp->t_state))
1502 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
1505 * Unscale the window into a 32-bit value.
1506 * For the SYN_SENT state the scale is zero.
1508 tiwin = th->th_win << tp->snd_scale;
1511 * TCP ECN processing.
1513 if (tp->t_flags & TF_ECN_PERMIT) {
1514 if (thflags & TH_CWR)
1515 tp->t_flags &= ~TF_ECN_SND_ECE;
1516 switch (iptos & IPTOS_ECN_MASK) {
1518 tp->t_flags |= TF_ECN_SND_ECE;
1519 TCPSTAT_INC(tcps_ecn_ce);
1521 case IPTOS_ECN_ECT0:
1522 TCPSTAT_INC(tcps_ecn_ect0);
1524 case IPTOS_ECN_ECT1:
1525 TCPSTAT_INC(tcps_ecn_ect1);
1528 /* Congestion experienced. */
1529 if (thflags & TH_ECE) {
1530 cc_cong_signal(tp, th, CC_ECN);
1535 * Parse options on any incoming segment.
1537 tcp_dooptions(&to, (u_char *)(th + 1),
1538 (th->th_off << 2) - sizeof(struct tcphdr),
1539 (thflags & TH_SYN) ? TO_SYN : 0);
1542 * If echoed timestamp is later than the current time,
1543 * fall back to non RFC1323 RTT calculation. Normalize
1544 * timestamp if syncookies were used when this connection
1547 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1548 to.to_tsecr -= tp->ts_offset;
1549 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1553 * If timestamps were negotiated during SYN/ACK they should
1554 * appear on every segment during this session and vice versa.
1556 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1557 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1558 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1559 "no action\n", s, __func__);
1563 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1564 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1565 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1566 "no action\n", s, __func__);
1572 * Process options only when we get SYN/ACK back. The SYN case
1573 * for incoming connections is handled in tcp_syncache.
1574 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1575 * or <SYN,ACK>) segment itself is never scaled.
1576 * XXX this is traditional behavior, may need to be cleaned up.
1578 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1579 if ((to.to_flags & TOF_SCALE) &&
1580 (tp->t_flags & TF_REQ_SCALE)) {
1581 tp->t_flags |= TF_RCVD_SCALE;
1582 tp->snd_scale = to.to_wscale;
1585 * Initial send window. It will be updated with
1586 * the next incoming segment to the scaled value.
1588 tp->snd_wnd = th->th_win;
1589 if (to.to_flags & TOF_TS) {
1590 tp->t_flags |= TF_RCVD_TSTMP;
1591 tp->ts_recent = to.to_tsval;
1592 tp->ts_recent_age = tcp_ts_getticks();
1594 if (to.to_flags & TOF_MSS)
1595 tcp_mss(tp, to.to_mss);
1596 if ((tp->t_flags & TF_SACK_PERMIT) &&
1597 (to.to_flags & TOF_SACKPERM) == 0)
1598 tp->t_flags &= ~TF_SACK_PERMIT;
1602 * Header prediction: check for the two common cases
1603 * of a uni-directional data xfer. If the packet has
1604 * no control flags, is in-sequence, the window didn't
1605 * change and we're not retransmitting, it's a
1606 * candidate. If the length is zero and the ack moved
1607 * forward, we're the sender side of the xfer. Just
1608 * free the data acked & wake any higher level process
1609 * that was blocked waiting for space. If the length
1610 * is non-zero and the ack didn't move, we're the
1611 * receiver side. If we're getting packets in-order
1612 * (the reassembly queue is empty), add the data to
1613 * the socket buffer and note that we need a delayed ack.
1614 * Make sure that the hidden state-flags are also off.
1615 * Since we check for TCPS_ESTABLISHED first, it can only
1618 if (tp->t_state == TCPS_ESTABLISHED &&
1619 th->th_seq == tp->rcv_nxt &&
1620 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1621 tp->snd_nxt == tp->snd_max &&
1622 tiwin && tiwin == tp->snd_wnd &&
1623 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1624 LIST_EMPTY(&tp->t_segq) &&
1625 ((to.to_flags & TOF_TS) == 0 ||
1626 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1629 * If last ACK falls within this segment's sequence numbers,
1630 * record the timestamp.
1631 * NOTE that the test is modified according to the latest
1632 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1634 if ((to.to_flags & TOF_TS) != 0 &&
1635 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1636 tp->ts_recent_age = tcp_ts_getticks();
1637 tp->ts_recent = to.to_tsval;
1641 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1642 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1643 !IN_RECOVERY(tp->t_flags) &&
1644 (to.to_flags & TOF_SACK) == 0 &&
1645 TAILQ_EMPTY(&tp->snd_holes)) {
1647 * This is a pure ack for outstanding data.
1649 if (ti_locked == TI_WLOCKED)
1650 INP_INFO_WUNLOCK(&V_tcbinfo);
1651 ti_locked = TI_UNLOCKED;
1653 TCPSTAT_INC(tcps_predack);
1656 * "bad retransmit" recovery.
1658 if (tp->t_rxtshift == 1 &&
1659 tp->t_flags & TF_PREVVALID &&
1660 (int)(ticks - tp->t_badrxtwin) < 0) {
1661 cc_cong_signal(tp, th, CC_RTO_ERR);
1665 * Recalculate the transmit timer / rtt.
1667 * Some boxes send broken timestamp replies
1668 * during the SYN+ACK phase, ignore
1669 * timestamps of 0 or we could calculate a
1670 * huge RTT and blow up the retransmit timer.
1672 if ((to.to_flags & TOF_TS) != 0 &&
1676 t = tcp_ts_getticks() - to.to_tsecr;
1677 if (!tp->t_rttlow || tp->t_rttlow > t)
1680 TCP_TS_TO_TICKS(t) + 1);
1681 } else if (tp->t_rtttime &&
1682 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1683 if (!tp->t_rttlow ||
1684 tp->t_rttlow > ticks - tp->t_rtttime)
1685 tp->t_rttlow = ticks - tp->t_rtttime;
1687 ticks - tp->t_rtttime);
1689 acked = BYTES_THIS_ACK(tp, th);
1691 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1692 hhook_run_tcp_est_in(tp, th, &to);
1694 TCPSTAT_INC(tcps_rcvackpack);
1695 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1696 sbdrop(&so->so_snd, acked);
1697 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1698 SEQ_LEQ(th->th_ack, tp->snd_recover))
1699 tp->snd_recover = th->th_ack - 1;
1702 * Let the congestion control algorithm update
1703 * congestion control related information. This
1704 * typically means increasing the congestion
1707 cc_ack_received(tp, th, CC_ACK);
1709 tp->snd_una = th->th_ack;
1711 * Pull snd_wl2 up to prevent seq wrap relative
1714 tp->snd_wl2 = th->th_ack;
1717 ND6_HINT(tp); /* Some progress has been made. */
1720 * If all outstanding data are acked, stop
1721 * retransmit timer, otherwise restart timer
1722 * using current (possibly backed-off) value.
1723 * If process is waiting for space,
1724 * wakeup/selwakeup/signal. If data
1725 * are ready to send, let tcp_output
1726 * decide between more output or persist.
1729 if (so->so_options & SO_DEBUG)
1730 tcp_trace(TA_INPUT, ostate, tp,
1731 (void *)tcp_saveipgen,
1734 if (tp->snd_una == tp->snd_max)
1735 tcp_timer_activate(tp, TT_REXMT, 0);
1736 else if (!tcp_timer_active(tp, TT_PERSIST))
1737 tcp_timer_activate(tp, TT_REXMT,
1740 if (so->so_snd.sb_cc)
1741 (void) tcp_output(tp);
1744 } else if (th->th_ack == tp->snd_una &&
1745 tlen <= sbspace(&so->so_rcv)) {
1746 int newsize = 0; /* automatic sockbuf scaling */
1749 * This is a pure, in-sequence data packet with
1750 * nothing on the reassembly queue and we have enough
1751 * buffer space to take it.
1753 if (ti_locked == TI_WLOCKED)
1754 INP_INFO_WUNLOCK(&V_tcbinfo);
1755 ti_locked = TI_UNLOCKED;
1757 /* Clean receiver SACK report if present */
1758 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1759 tcp_clean_sackreport(tp);
1760 TCPSTAT_INC(tcps_preddat);
1761 tp->rcv_nxt += tlen;
1763 * Pull snd_wl1 up to prevent seq wrap relative to
1766 tp->snd_wl1 = th->th_seq;
1768 * Pull rcv_up up to prevent seq wrap relative to
1771 tp->rcv_up = tp->rcv_nxt;
1772 TCPSTAT_INC(tcps_rcvpack);
1773 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1774 ND6_HINT(tp); /* Some progress has been made */
1776 if (so->so_options & SO_DEBUG)
1777 tcp_trace(TA_INPUT, ostate, tp,
1778 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1781 * Automatic sizing of receive socket buffer. Often the send
1782 * buffer size is not optimally adjusted to the actual network
1783 * conditions at hand (delay bandwidth product). Setting the
1784 * buffer size too small limits throughput on links with high
1785 * bandwidth and high delay (eg. trans-continental/oceanic links).
1787 * On the receive side the socket buffer memory is only rarely
1788 * used to any significant extent. This allows us to be much
1789 * more aggressive in scaling the receive socket buffer. For
1790 * the case that the buffer space is actually used to a large
1791 * extent and we run out of kernel memory we can simply drop
1792 * the new segments; TCP on the sender will just retransmit it
1793 * later. Setting the buffer size too big may only consume too
1794 * much kernel memory if the application doesn't read() from
1795 * the socket or packet loss or reordering makes use of the
1798 * The criteria to step up the receive buffer one notch are:
1799 * 1. the number of bytes received during the time it takes
1800 * one timestamp to be reflected back to us (the RTT);
1801 * 2. received bytes per RTT is within seven eighth of the
1802 * current socket buffer size;
1803 * 3. receive buffer size has not hit maximal automatic size;
1805 * This algorithm does one step per RTT at most and only if
1806 * we receive a bulk stream w/o packet losses or reorderings.
1807 * Shrinking the buffer during idle times is not necessary as
1808 * it doesn't consume any memory when idle.
1810 * TODO: Only step up if the application is actually serving
1811 * the buffer to better manage the socket buffer resources.
1813 if (V_tcp_do_autorcvbuf &&
1815 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1816 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1817 to.to_tsecr - tp->rfbuf_ts < hz) {
1819 (so->so_rcv.sb_hiwat / 8 * 7) &&
1820 so->so_rcv.sb_hiwat <
1821 V_tcp_autorcvbuf_max) {
1823 min(so->so_rcv.sb_hiwat +
1824 V_tcp_autorcvbuf_inc,
1825 V_tcp_autorcvbuf_max);
1827 /* Start over with next RTT. */
1831 tp->rfbuf_cnt += tlen; /* add up */
1834 /* Add data to socket buffer. */
1835 SOCKBUF_LOCK(&so->so_rcv);
1836 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1840 * Set new socket buffer size.
1841 * Give up when limit is reached.
1844 if (!sbreserve_locked(&so->so_rcv,
1846 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1847 m_adj(m, drop_hdrlen); /* delayed header drop */
1848 sbappendstream_locked(&so->so_rcv, m);
1850 /* NB: sorwakeup_locked() does an implicit unlock. */
1851 sorwakeup_locked(so);
1852 if (DELAY_ACK(tp)) {
1853 tp->t_flags |= TF_DELACK;
1855 tp->t_flags |= TF_ACKNOW;
1863 * Calculate amount of space in receive window,
1864 * and then do TCP input processing.
1865 * Receive window is amount of space in rcv queue,
1866 * but not less than advertised window.
1868 win = sbspace(&so->so_rcv);
1871 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1873 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1877 switch (tp->t_state) {
1880 * If the state is SYN_RECEIVED:
1881 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1883 case TCPS_SYN_RECEIVED:
1884 if ((thflags & TH_ACK) &&
1885 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1886 SEQ_GT(th->th_ack, tp->snd_max))) {
1887 rstreason = BANDLIM_RST_OPENPORT;
1893 * If the state is SYN_SENT:
1894 * if seg contains an ACK, but not for our SYN, drop the input.
1895 * if seg contains a RST, then drop the connection.
1896 * if seg does not contain SYN, then drop it.
1897 * Otherwise this is an acceptable SYN segment
1898 * initialize tp->rcv_nxt and tp->irs
1899 * if seg contains ack then advance tp->snd_una
1900 * if seg contains an ECE and ECN support is enabled, the stream
1902 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1903 * arrange for segment to be acked (eventually)
1904 * continue processing rest of data/controls, beginning with URG
1907 if ((thflags & TH_ACK) &&
1908 (SEQ_LEQ(th->th_ack, tp->iss) ||
1909 SEQ_GT(th->th_ack, tp->snd_max))) {
1910 rstreason = BANDLIM_UNLIMITED;
1913 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST))
1914 tp = tcp_drop(tp, ECONNREFUSED);
1915 if (thflags & TH_RST)
1917 if (!(thflags & TH_SYN))
1920 tp->irs = th->th_seq;
1922 if (thflags & TH_ACK) {
1923 TCPSTAT_INC(tcps_connects);
1926 mac_socketpeer_set_from_mbuf(m, so);
1928 /* Do window scaling on this connection? */
1929 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1930 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1931 tp->rcv_scale = tp->request_r_scale;
1933 tp->rcv_adv += imin(tp->rcv_wnd,
1934 TCP_MAXWIN << tp->rcv_scale);
1935 tp->snd_una++; /* SYN is acked */
1937 * If there's data, delay ACK; if there's also a FIN
1938 * ACKNOW will be turned on later.
1940 if (DELAY_ACK(tp) && tlen != 0)
1941 tcp_timer_activate(tp, TT_DELACK,
1944 tp->t_flags |= TF_ACKNOW;
1946 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1947 tp->t_flags |= TF_ECN_PERMIT;
1948 TCPSTAT_INC(tcps_ecn_shs);
1952 * Received <SYN,ACK> in SYN_SENT[*] state.
1954 * SYN_SENT --> ESTABLISHED
1955 * SYN_SENT* --> FIN_WAIT_1
1957 tp->t_starttime = ticks;
1958 if (tp->t_flags & TF_NEEDFIN) {
1959 tp->t_state = TCPS_FIN_WAIT_1;
1960 tp->t_flags &= ~TF_NEEDFIN;
1963 tp->t_state = TCPS_ESTABLISHED;
1965 tcp_timer_activate(tp, TT_KEEP,
1970 * Received initial SYN in SYN-SENT[*] state =>
1971 * simultaneous open. If segment contains CC option
1972 * and there is a cached CC, apply TAO test.
1973 * If it succeeds, connection is * half-synchronized.
1974 * Otherwise, do 3-way handshake:
1975 * SYN-SENT -> SYN-RECEIVED
1976 * SYN-SENT* -> SYN-RECEIVED*
1977 * If there was no CC option, clear cached CC value.
1979 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1980 tcp_timer_activate(tp, TT_REXMT, 0);
1981 tp->t_state = TCPS_SYN_RECEIVED;
1984 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: "
1985 "ti_locked %d", __func__, ti_locked));
1986 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1987 INP_WLOCK_ASSERT(tp->t_inpcb);
1990 * Advance th->th_seq to correspond to first data byte.
1991 * If data, trim to stay within window,
1992 * dropping FIN if necessary.
1995 if (tlen > tp->rcv_wnd) {
1996 todrop = tlen - tp->rcv_wnd;
2000 TCPSTAT_INC(tcps_rcvpackafterwin);
2001 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2003 tp->snd_wl1 = th->th_seq - 1;
2004 tp->rcv_up = th->th_seq;
2006 * Client side of transaction: already sent SYN and data.
2007 * If the remote host used T/TCP to validate the SYN,
2008 * our data will be ACK'd; if so, enter normal data segment
2009 * processing in the middle of step 5, ack processing.
2010 * Otherwise, goto step 6.
2012 if (thflags & TH_ACK)
2018 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2019 * do normal processing.
2021 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2025 break; /* continue normal processing */
2029 * States other than LISTEN or SYN_SENT.
2030 * First check the RST flag and sequence number since reset segments
2031 * are exempt from the timestamp and connection count tests. This
2032 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2033 * below which allowed reset segments in half the sequence space
2034 * to fall though and be processed (which gives forged reset
2035 * segments with a random sequence number a 50 percent chance of
2036 * killing a connection).
2037 * Then check timestamp, if present.
2038 * Then check the connection count, if present.
2039 * Then check that at least some bytes of segment are within
2040 * receive window. If segment begins before rcv_nxt,
2041 * drop leading data (and SYN); if nothing left, just ack.
2044 * If the RST bit is set, check the sequence number to see
2045 * if this is a valid reset segment.
2047 * In all states except SYN-SENT, all reset (RST) segments
2048 * are validated by checking their SEQ-fields. A reset is
2049 * valid if its sequence number is in the window.
2050 * Note: this does not take into account delayed ACKs, so
2051 * we should test against last_ack_sent instead of rcv_nxt.
2052 * The sequence number in the reset segment is normally an
2053 * echo of our outgoing acknowlegement numbers, but some hosts
2054 * send a reset with the sequence number at the rightmost edge
2055 * of our receive window, and we have to handle this case.
2056 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
2057 * that brute force RST attacks are possible. To combat this,
2058 * we use a much stricter check while in the ESTABLISHED state,
2059 * only accepting RSTs where the sequence number is equal to
2060 * last_ack_sent. In all other states (the states in which a
2061 * RST is more likely), the more permissive check is used.
2062 * If we have multiple segments in flight, the initial reset
2063 * segment sequence numbers will be to the left of last_ack_sent,
2064 * but they will eventually catch up.
2065 * In any case, it never made sense to trim reset segments to
2066 * fit the receive window since RFC 1122 says:
2067 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
2069 * A TCP SHOULD allow a received RST segment to include data.
2072 * It has been suggested that a RST segment could contain
2073 * ASCII text that encoded and explained the cause of the
2074 * RST. No standard has yet been established for such
2077 * If the reset segment passes the sequence number test examine
2079 * SYN_RECEIVED STATE:
2080 * If passive open, return to LISTEN state.
2081 * If active open, inform user that connection was refused.
2082 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
2083 * Inform user that connection was reset, and close tcb.
2084 * CLOSING, LAST_ACK STATES:
2087 * Drop the segment - see Stevens, vol. 2, p. 964 and
2090 if (thflags & TH_RST) {
2091 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2092 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2093 switch (tp->t_state) {
2095 case TCPS_SYN_RECEIVED:
2096 so->so_error = ECONNREFUSED;
2099 case TCPS_ESTABLISHED:
2100 if (V_tcp_insecure_rst == 0 &&
2101 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
2102 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
2103 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2104 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
2105 TCPSTAT_INC(tcps_badrst);
2109 case TCPS_FIN_WAIT_1:
2110 case TCPS_FIN_WAIT_2:
2111 case TCPS_CLOSE_WAIT:
2112 so->so_error = ECONNRESET;
2114 KASSERT(ti_locked == TI_WLOCKED,
2115 ("tcp_do_segment: TH_RST 1 ti_locked %d",
2117 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2119 tp->t_state = TCPS_CLOSED;
2120 TCPSTAT_INC(tcps_drops);
2126 KASSERT(ti_locked == TI_WLOCKED,
2127 ("tcp_do_segment: TH_RST 2 ti_locked %d",
2129 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2139 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2140 * and it's less than ts_recent, drop it.
2142 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2143 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2145 /* Check to see if ts_recent is over 24 days old. */
2146 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2148 * Invalidate ts_recent. If this segment updates
2149 * ts_recent, the age will be reset later and ts_recent
2150 * will get a valid value. If it does not, setting
2151 * ts_recent to zero will at least satisfy the
2152 * requirement that zero be placed in the timestamp
2153 * echo reply when ts_recent isn't valid. The
2154 * age isn't reset until we get a valid ts_recent
2155 * because we don't want out-of-order segments to be
2156 * dropped when ts_recent is old.
2160 TCPSTAT_INC(tcps_rcvduppack);
2161 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2162 TCPSTAT_INC(tcps_pawsdrop);
2170 * In the SYN-RECEIVED state, validate that the packet belongs to
2171 * this connection before trimming the data to fit the receive
2172 * window. Check the sequence number versus IRS since we know
2173 * the sequence numbers haven't wrapped. This is a partial fix
2174 * for the "LAND" DoS attack.
2176 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2177 rstreason = BANDLIM_RST_OPENPORT;
2181 todrop = tp->rcv_nxt - th->th_seq;
2184 * If this is a duplicate SYN for our current connection,
2185 * advance over it and pretend and it's not a SYN.
2187 if (thflags & TH_SYN && th->th_seq == tp->irs) {
2197 * Following if statement from Stevens, vol. 2, p. 960.
2200 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2202 * Any valid FIN must be to the left of the window.
2203 * At this point the FIN must be a duplicate or out
2204 * of sequence; drop it.
2209 * Send an ACK to resynchronize and drop any data.
2210 * But keep on processing for RST or ACK.
2212 tp->t_flags |= TF_ACKNOW;
2214 TCPSTAT_INC(tcps_rcvduppack);
2215 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2217 TCPSTAT_INC(tcps_rcvpartduppack);
2218 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2220 drop_hdrlen += todrop; /* drop from the top afterwards */
2221 th->th_seq += todrop;
2223 if (th->th_urp > todrop)
2224 th->th_urp -= todrop;
2232 * If new data are received on a connection after the
2233 * user processes are gone, then RST the other end.
2235 if ((so->so_state & SS_NOFDREF) &&
2236 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2237 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && "
2238 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2239 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2241 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2242 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2243 "after socket was closed, "
2244 "sending RST and removing tcpcb\n",
2245 s, __func__, tcpstates[tp->t_state], tlen);
2249 TCPSTAT_INC(tcps_rcvafterclose);
2250 rstreason = BANDLIM_UNLIMITED;
2255 * If segment ends after window, drop trailing data
2256 * (and PUSH and FIN); if nothing left, just ACK.
2258 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2260 TCPSTAT_INC(tcps_rcvpackafterwin);
2261 if (todrop >= tlen) {
2262 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2264 * If window is closed can only take segments at
2265 * window edge, and have to drop data and PUSH from
2266 * incoming segments. Continue processing, but
2267 * remember to ack. Otherwise, drop segment
2270 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2271 tp->t_flags |= TF_ACKNOW;
2272 TCPSTAT_INC(tcps_rcvwinprobe);
2276 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2279 thflags &= ~(TH_PUSH|TH_FIN);
2283 * If last ACK falls within this segment's sequence numbers,
2284 * record its timestamp.
2286 * 1) That the test incorporates suggestions from the latest
2287 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2288 * 2) That updating only on newer timestamps interferes with
2289 * our earlier PAWS tests, so this check should be solely
2290 * predicated on the sequence space of this segment.
2291 * 3) That we modify the segment boundary check to be
2292 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2293 * instead of RFC1323's
2294 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2295 * This modified check allows us to overcome RFC1323's
2296 * limitations as described in Stevens TCP/IP Illustrated
2297 * Vol. 2 p.869. In such cases, we can still calculate the
2298 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2300 if ((to.to_flags & TOF_TS) != 0 &&
2301 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2302 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2303 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2304 tp->ts_recent_age = tcp_ts_getticks();
2305 tp->ts_recent = to.to_tsval;
2309 * If a SYN is in the window, then this is an
2310 * error and we send an RST and drop the connection.
2312 if (thflags & TH_SYN) {
2313 KASSERT(ti_locked == TI_WLOCKED,
2314 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2315 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2317 tp = tcp_drop(tp, ECONNRESET);
2318 rstreason = BANDLIM_UNLIMITED;
2323 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2324 * flag is on (half-synchronized state), then queue data for
2325 * later processing; else drop segment and return.
2327 if ((thflags & TH_ACK) == 0) {
2328 if (tp->t_state == TCPS_SYN_RECEIVED ||
2329 (tp->t_flags & TF_NEEDSYN))
2331 else if (tp->t_flags & TF_ACKNOW)
2340 switch (tp->t_state) {
2343 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2344 * ESTABLISHED state and continue processing.
2345 * The ACK was checked above.
2347 case TCPS_SYN_RECEIVED:
2349 TCPSTAT_INC(tcps_connects);
2351 /* Do window scaling? */
2352 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2353 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2354 tp->rcv_scale = tp->request_r_scale;
2355 tp->snd_wnd = tiwin;
2359 * SYN-RECEIVED -> ESTABLISHED
2360 * SYN-RECEIVED* -> FIN-WAIT-1
2362 tp->t_starttime = ticks;
2363 if (tp->t_flags & TF_NEEDFIN) {
2364 tp->t_state = TCPS_FIN_WAIT_1;
2365 tp->t_flags &= ~TF_NEEDFIN;
2367 tp->t_state = TCPS_ESTABLISHED;
2369 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2372 * If segment contains data or ACK, will call tcp_reass()
2373 * later; if not, do so now to pass queued data to user.
2375 if (tlen == 0 && (thflags & TH_FIN) == 0)
2376 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2378 tp->snd_wl1 = th->th_seq - 1;
2382 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2383 * ACKs. If the ack is in the range
2384 * tp->snd_una < th->th_ack <= tp->snd_max
2385 * then advance tp->snd_una to th->th_ack and drop
2386 * data from the retransmission queue. If this ACK reflects
2387 * more up to date window information we update our window information.
2389 case TCPS_ESTABLISHED:
2390 case TCPS_FIN_WAIT_1:
2391 case TCPS_FIN_WAIT_2:
2392 case TCPS_CLOSE_WAIT:
2395 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2396 TCPSTAT_INC(tcps_rcvacktoomuch);
2399 if ((tp->t_flags & TF_SACK_PERMIT) &&
2400 ((to.to_flags & TOF_SACK) ||
2401 !TAILQ_EMPTY(&tp->snd_holes)))
2402 tcp_sack_doack(tp, &to, th->th_ack);
2404 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2405 hhook_run_tcp_est_in(tp, th, &to);
2407 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2408 if (tlen == 0 && tiwin == tp->snd_wnd) {
2409 TCPSTAT_INC(tcps_rcvdupack);
2411 * If we have outstanding data (other than
2412 * a window probe), this is a completely
2413 * duplicate ack (ie, window info didn't
2414 * change), the ack is the biggest we've
2415 * seen and we've seen exactly our rexmt
2416 * threshhold of them, assume a packet
2417 * has been dropped and retransmit it.
2418 * Kludge snd_nxt & the congestion
2419 * window so we send only this one
2422 * We know we're losing at the current
2423 * window size so do congestion avoidance
2424 * (set ssthresh to half the current window
2425 * and pull our congestion window back to
2426 * the new ssthresh).
2428 * Dup acks mean that packets have left the
2429 * network (they're now cached at the receiver)
2430 * so bump cwnd by the amount in the receiver
2431 * to keep a constant cwnd packets in the
2434 * When using TCP ECN, notify the peer that
2435 * we reduced the cwnd.
2437 if (!tcp_timer_active(tp, TT_REXMT) ||
2438 th->th_ack != tp->snd_una)
2440 else if (++tp->t_dupacks > tcprexmtthresh ||
2441 IN_FASTRECOVERY(tp->t_flags)) {
2442 cc_ack_received(tp, th, CC_DUPACK);
2443 if ((tp->t_flags & TF_SACK_PERMIT) &&
2444 IN_FASTRECOVERY(tp->t_flags)) {
2448 * Compute the amount of data in flight first.
2449 * We can inject new data into the pipe iff
2450 * we have less than 1/2 the original window's
2451 * worth of data in flight.
2453 awnd = (tp->snd_nxt - tp->snd_fack) +
2454 tp->sackhint.sack_bytes_rexmit;
2455 if (awnd < tp->snd_ssthresh) {
2456 tp->snd_cwnd += tp->t_maxseg;
2457 if (tp->snd_cwnd > tp->snd_ssthresh)
2458 tp->snd_cwnd = tp->snd_ssthresh;
2461 tp->snd_cwnd += tp->t_maxseg;
2462 if ((thflags & TH_FIN) &&
2463 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2465 * If its a fin we need to process
2466 * it to avoid a race where both
2467 * sides enter FIN-WAIT and send FIN|ACK
2472 (void) tcp_output(tp);
2474 } else if (tp->t_dupacks == tcprexmtthresh) {
2475 tcp_seq onxt = tp->snd_nxt;
2478 * If we're doing sack, check to
2479 * see if we're already in sack
2480 * recovery. If we're not doing sack,
2481 * check to see if we're in newreno
2484 if (tp->t_flags & TF_SACK_PERMIT) {
2485 if (IN_FASTRECOVERY(tp->t_flags)) {
2490 if (SEQ_LEQ(th->th_ack,
2496 /* Congestion signal before ack. */
2497 cc_cong_signal(tp, th, CC_NDUPACK);
2498 cc_ack_received(tp, th, CC_DUPACK);
2499 tcp_timer_activate(tp, TT_REXMT, 0);
2501 if (tp->t_flags & TF_SACK_PERMIT) {
2503 tcps_sack_recovery_episode);
2504 tp->sack_newdata = tp->snd_nxt;
2505 tp->snd_cwnd = tp->t_maxseg;
2506 (void) tcp_output(tp);
2509 tp->snd_nxt = th->th_ack;
2510 tp->snd_cwnd = tp->t_maxseg;
2511 if ((thflags & TH_FIN) &&
2512 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2514 * If its a fin we need to process
2515 * it to avoid a race where both
2516 * sides enter FIN-WAIT and send FIN|ACK
2521 (void) tcp_output(tp);
2522 KASSERT(tp->snd_limited <= 2,
2523 ("%s: tp->snd_limited too big",
2525 tp->snd_cwnd = tp->snd_ssthresh +
2527 (tp->t_dupacks - tp->snd_limited);
2528 if (SEQ_GT(onxt, tp->snd_nxt))
2531 } else if (V_tcp_do_rfc3042) {
2532 cc_ack_received(tp, th, CC_DUPACK);
2533 u_long oldcwnd = tp->snd_cwnd;
2534 tcp_seq oldsndmax = tp->snd_max;
2538 KASSERT(tp->t_dupacks == 1 ||
2540 ("%s: dupacks not 1 or 2",
2542 if (tp->t_dupacks == 1)
2543 tp->snd_limited = 0;
2545 (tp->snd_nxt - tp->snd_una) +
2546 (tp->t_dupacks - tp->snd_limited) *
2548 if ((thflags & TH_FIN) &&
2549 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2551 * If its a fin we need to process
2552 * it to avoid a race where both
2553 * sides enter FIN-WAIT and send FIN|ACK
2559 * Only call tcp_output when there
2560 * is new data available to be sent.
2561 * Otherwise we would send pure ACKs.
2563 SOCKBUF_LOCK(&so->so_snd);
2564 avail = so->so_snd.sb_cc -
2565 (tp->snd_nxt - tp->snd_una);
2566 SOCKBUF_UNLOCK(&so->so_snd);
2568 (void) tcp_output(tp);
2569 sent = tp->snd_max - oldsndmax;
2570 if (sent > tp->t_maxseg) {
2571 KASSERT((tp->t_dupacks == 2 &&
2572 tp->snd_limited == 0) ||
2573 (sent == tp->t_maxseg + 1 &&
2574 tp->t_flags & TF_SENTFIN),
2575 ("%s: sent too much",
2577 tp->snd_limited = 2;
2578 } else if (sent > 0)
2580 tp->snd_cwnd = oldcwnd;
2588 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2589 ("%s: th_ack <= snd_una", __func__));
2592 * If the congestion window was inflated to account
2593 * for the other side's cached packets, retract it.
2595 if (IN_FASTRECOVERY(tp->t_flags)) {
2596 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2597 if (tp->t_flags & TF_SACK_PERMIT)
2598 tcp_sack_partialack(tp, th);
2600 tcp_newreno_partial_ack(tp, th);
2602 cc_post_recovery(tp, th);
2606 * If we reach this point, ACK is not a duplicate,
2607 * i.e., it ACKs something we sent.
2609 if (tp->t_flags & TF_NEEDSYN) {
2611 * T/TCP: Connection was half-synchronized, and our
2612 * SYN has been ACK'd (so connection is now fully
2613 * synchronized). Go to non-starred state,
2614 * increment snd_una for ACK of SYN, and check if
2615 * we can do window scaling.
2617 tp->t_flags &= ~TF_NEEDSYN;
2619 /* Do window scaling? */
2620 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2621 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2622 tp->rcv_scale = tp->request_r_scale;
2623 /* Send window already scaled. */
2628 INP_WLOCK_ASSERT(tp->t_inpcb);
2630 acked = BYTES_THIS_ACK(tp, th);
2631 TCPSTAT_INC(tcps_rcvackpack);
2632 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2635 * If we just performed our first retransmit, and the ACK
2636 * arrives within our recovery window, then it was a mistake
2637 * to do the retransmit in the first place. Recover our
2638 * original cwnd and ssthresh, and proceed to transmit where
2641 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2642 (int)(ticks - tp->t_badrxtwin) < 0)
2643 cc_cong_signal(tp, th, CC_RTO_ERR);
2646 * If we have a timestamp reply, update smoothed
2647 * round trip time. If no timestamp is present but
2648 * transmit timer is running and timed sequence
2649 * number was acked, update smoothed round trip time.
2650 * Since we now have an rtt measurement, cancel the
2651 * timer backoff (cf., Phil Karn's retransmit alg.).
2652 * Recompute the initial retransmit timer.
2654 * Some boxes send broken timestamp replies
2655 * during the SYN+ACK phase, ignore
2656 * timestamps of 0 or we could calculate a
2657 * huge RTT and blow up the retransmit timer.
2659 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2662 t = tcp_ts_getticks() - to.to_tsecr;
2663 if (!tp->t_rttlow || tp->t_rttlow > t)
2665 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2666 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2667 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2668 tp->t_rttlow = ticks - tp->t_rtttime;
2669 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2673 * If all outstanding data is acked, stop retransmit
2674 * timer and remember to restart (more output or persist).
2675 * If there is more data to be acked, restart retransmit
2676 * timer, using current (possibly backed-off) value.
2678 if (th->th_ack == tp->snd_max) {
2679 tcp_timer_activate(tp, TT_REXMT, 0);
2681 } else if (!tcp_timer_active(tp, TT_PERSIST))
2682 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2685 * If no data (only SYN) was ACK'd,
2686 * skip rest of ACK processing.
2692 * Let the congestion control algorithm update congestion
2693 * control related information. This typically means increasing
2694 * the congestion window.
2696 cc_ack_received(tp, th, CC_ACK);
2698 SOCKBUF_LOCK(&so->so_snd);
2699 if (acked > so->so_snd.sb_cc) {
2700 tp->snd_wnd -= so->so_snd.sb_cc;
2701 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2704 sbdrop_locked(&so->so_snd, acked);
2705 tp->snd_wnd -= acked;
2708 /* NB: sowwakeup_locked() does an implicit unlock. */
2709 sowwakeup_locked(so);
2710 /* Detect una wraparound. */
2711 if (!IN_RECOVERY(tp->t_flags) &&
2712 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2713 SEQ_LEQ(th->th_ack, tp->snd_recover))
2714 tp->snd_recover = th->th_ack - 1;
2715 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2716 if (IN_RECOVERY(tp->t_flags) &&
2717 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2718 EXIT_RECOVERY(tp->t_flags);
2720 tp->snd_una = th->th_ack;
2721 if (tp->t_flags & TF_SACK_PERMIT) {
2722 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2723 tp->snd_recover = tp->snd_una;
2725 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2726 tp->snd_nxt = tp->snd_una;
2728 switch (tp->t_state) {
2731 * In FIN_WAIT_1 STATE in addition to the processing
2732 * for the ESTABLISHED state if our FIN is now acknowledged
2733 * then enter FIN_WAIT_2.
2735 case TCPS_FIN_WAIT_1:
2736 if (ourfinisacked) {
2738 * If we can't receive any more
2739 * data, then closing user can proceed.
2740 * Starting the timer is contrary to the
2741 * specification, but if we don't get a FIN
2742 * we'll hang forever.
2745 * we should release the tp also, and use a
2748 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2749 soisdisconnected(so);
2750 tcp_timer_activate(tp, TT_2MSL,
2751 (tcp_fast_finwait2_recycle ?
2752 tcp_finwait2_timeout :
2755 tp->t_state = TCPS_FIN_WAIT_2;
2760 * In CLOSING STATE in addition to the processing for
2761 * the ESTABLISHED state if the ACK acknowledges our FIN
2762 * then enter the TIME-WAIT state, otherwise ignore
2766 if (ourfinisacked) {
2767 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2769 INP_INFO_WUNLOCK(&V_tcbinfo);
2776 * In LAST_ACK, we may still be waiting for data to drain
2777 * and/or to be acked, as well as for the ack of our FIN.
2778 * If our FIN is now acknowledged, delete the TCB,
2779 * enter the closed state and return.
2782 if (ourfinisacked) {
2783 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2792 INP_WLOCK_ASSERT(tp->t_inpcb);
2795 * Update window information.
2796 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2798 if ((thflags & TH_ACK) &&
2799 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2800 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2801 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2802 /* keep track of pure window updates */
2804 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2805 TCPSTAT_INC(tcps_rcvwinupd);
2806 tp->snd_wnd = tiwin;
2807 tp->snd_wl1 = th->th_seq;
2808 tp->snd_wl2 = th->th_ack;
2809 if (tp->snd_wnd > tp->max_sndwnd)
2810 tp->max_sndwnd = tp->snd_wnd;
2815 * Process segments with URG.
2817 if ((thflags & TH_URG) && th->th_urp &&
2818 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2820 * This is a kludge, but if we receive and accept
2821 * random urgent pointers, we'll crash in
2822 * soreceive. It's hard to imagine someone
2823 * actually wanting to send this much urgent data.
2825 SOCKBUF_LOCK(&so->so_rcv);
2826 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2827 th->th_urp = 0; /* XXX */
2828 thflags &= ~TH_URG; /* XXX */
2829 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2830 goto dodata; /* XXX */
2833 * If this segment advances the known urgent pointer,
2834 * then mark the data stream. This should not happen
2835 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2836 * a FIN has been received from the remote side.
2837 * In these states we ignore the URG.
2839 * According to RFC961 (Assigned Protocols),
2840 * the urgent pointer points to the last octet
2841 * of urgent data. We continue, however,
2842 * to consider it to indicate the first octet
2843 * of data past the urgent section as the original
2844 * spec states (in one of two places).
2846 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2847 tp->rcv_up = th->th_seq + th->th_urp;
2848 so->so_oobmark = so->so_rcv.sb_cc +
2849 (tp->rcv_up - tp->rcv_nxt) - 1;
2850 if (so->so_oobmark == 0)
2851 so->so_rcv.sb_state |= SBS_RCVATMARK;
2853 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2855 SOCKBUF_UNLOCK(&so->so_rcv);
2857 * Remove out of band data so doesn't get presented to user.
2858 * This can happen independent of advancing the URG pointer,
2859 * but if two URG's are pending at once, some out-of-band
2860 * data may creep in... ick.
2862 if (th->th_urp <= (u_long)tlen &&
2863 !(so->so_options & SO_OOBINLINE)) {
2864 /* hdr drop is delayed */
2865 tcp_pulloutofband(so, th, m, drop_hdrlen);
2869 * If no out of band data is expected,
2870 * pull receive urgent pointer along
2871 * with the receive window.
2873 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2874 tp->rcv_up = tp->rcv_nxt;
2877 INP_WLOCK_ASSERT(tp->t_inpcb);
2880 * Process the segment text, merging it into the TCP sequencing queue,
2881 * and arranging for acknowledgment of receipt if necessary.
2882 * This process logically involves adjusting tp->rcv_wnd as data
2883 * is presented to the user (this happens in tcp_usrreq.c,
2884 * case PRU_RCVD). If a FIN has already been received on this
2885 * connection then we just ignore the text.
2887 if ((tlen || (thflags & TH_FIN)) &&
2888 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2889 tcp_seq save_start = th->th_seq;
2890 m_adj(m, drop_hdrlen); /* delayed header drop */
2892 * Insert segment which includes th into TCP reassembly queue
2893 * with control block tp. Set thflags to whether reassembly now
2894 * includes a segment with FIN. This handles the common case
2895 * inline (segment is the next to be received on an established
2896 * connection, and the queue is empty), avoiding linkage into
2897 * and removal from the queue and repetition of various
2899 * Set DELACK for segments received in order, but ack
2900 * immediately when segments are out of order (so
2901 * fast retransmit can work).
2903 if (th->th_seq == tp->rcv_nxt &&
2904 LIST_EMPTY(&tp->t_segq) &&
2905 TCPS_HAVEESTABLISHED(tp->t_state)) {
2907 tp->t_flags |= TF_DELACK;
2909 tp->t_flags |= TF_ACKNOW;
2910 tp->rcv_nxt += tlen;
2911 thflags = th->th_flags & TH_FIN;
2912 TCPSTAT_INC(tcps_rcvpack);
2913 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2915 SOCKBUF_LOCK(&so->so_rcv);
2916 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2919 sbappendstream_locked(&so->so_rcv, m);
2920 /* NB: sorwakeup_locked() does an implicit unlock. */
2921 sorwakeup_locked(so);
2924 * XXX: Due to the header drop above "th" is
2925 * theoretically invalid by now. Fortunately
2926 * m_adj() doesn't actually frees any mbufs
2927 * when trimming from the head.
2929 thflags = tcp_reass(tp, th, &tlen, m);
2930 tp->t_flags |= TF_ACKNOW;
2932 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2933 tcp_update_sack_list(tp, save_start, save_start + tlen);
2936 * Note the amount of data that peer has sent into
2937 * our window, in order to estimate the sender's
2941 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2942 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2944 len = so->so_rcv.sb_hiwat;
2952 * If FIN is received ACK the FIN and let the user know
2953 * that the connection is closing.
2955 if (thflags & TH_FIN) {
2956 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2959 * If connection is half-synchronized
2960 * (ie NEEDSYN flag on) then delay ACK,
2961 * so it may be piggybacked when SYN is sent.
2962 * Otherwise, since we received a FIN then no
2963 * more input can be expected, send ACK now.
2965 if (tp->t_flags & TF_NEEDSYN)
2966 tp->t_flags |= TF_DELACK;
2968 tp->t_flags |= TF_ACKNOW;
2971 switch (tp->t_state) {
2974 * In SYN_RECEIVED and ESTABLISHED STATES
2975 * enter the CLOSE_WAIT state.
2977 case TCPS_SYN_RECEIVED:
2978 tp->t_starttime = ticks;
2980 case TCPS_ESTABLISHED:
2981 tp->t_state = TCPS_CLOSE_WAIT;
2985 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2986 * enter the CLOSING state.
2988 case TCPS_FIN_WAIT_1:
2989 tp->t_state = TCPS_CLOSING;
2993 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2994 * starting the time-wait timer, turning off the other
2997 case TCPS_FIN_WAIT_2:
2998 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2999 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata "
3000 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
3004 INP_INFO_WUNLOCK(&V_tcbinfo);
3008 if (ti_locked == TI_WLOCKED)
3009 INP_INFO_WUNLOCK(&V_tcbinfo);
3010 ti_locked = TI_UNLOCKED;
3013 if (so->so_options & SO_DEBUG)
3014 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3019 * Return any desired output.
3021 if (needoutput || (tp->t_flags & TF_ACKNOW))
3022 (void) tcp_output(tp);
3025 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3026 __func__, ti_locked));
3027 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3028 INP_WLOCK_ASSERT(tp->t_inpcb);
3030 if (tp->t_flags & TF_DELACK) {
3031 tp->t_flags &= ~TF_DELACK;
3032 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3034 INP_WUNLOCK(tp->t_inpcb);
3039 * Generate an ACK dropping incoming segment if it occupies
3040 * sequence space, where the ACK reflects our state.
3042 * We can now skip the test for the RST flag since all
3043 * paths to this code happen after packets containing
3044 * RST have been dropped.
3046 * In the SYN-RECEIVED state, don't send an ACK unless the
3047 * segment we received passes the SYN-RECEIVED ACK test.
3048 * If it fails send a RST. This breaks the loop in the
3049 * "LAND" DoS attack, and also prevents an ACK storm
3050 * between two listening ports that have been sent forged
3051 * SYN segments, each with the source address of the other.
3053 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3054 (SEQ_GT(tp->snd_una, th->th_ack) ||
3055 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3056 rstreason = BANDLIM_RST_OPENPORT;
3060 if (so->so_options & SO_DEBUG)
3061 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3064 if (ti_locked == TI_WLOCKED)
3065 INP_INFO_WUNLOCK(&V_tcbinfo);
3066 ti_locked = TI_UNLOCKED;
3068 tp->t_flags |= TF_ACKNOW;
3069 (void) tcp_output(tp);
3070 INP_WUNLOCK(tp->t_inpcb);
3075 if (ti_locked == TI_WLOCKED)
3076 INP_INFO_WUNLOCK(&V_tcbinfo);
3077 ti_locked = TI_UNLOCKED;
3080 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3081 INP_WUNLOCK(tp->t_inpcb);
3083 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3087 if (ti_locked == TI_WLOCKED) {
3088 INP_INFO_WUNLOCK(&V_tcbinfo);
3089 ti_locked = TI_UNLOCKED;
3093 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3097 * Drop space held by incoming segment and return.
3100 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3101 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3105 INP_WUNLOCK(tp->t_inpcb);
3110 * Issue RST and make ACK acceptable to originator of segment.
3111 * The mbuf must still include the original packet header.
3115 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3116 int tlen, int rstreason)
3122 struct ip6_hdr *ip6;
3126 INP_WLOCK_ASSERT(tp->t_inpcb);
3129 /* Don't bother if destination was broadcast/multicast. */
3130 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3133 if (mtod(m, struct ip *)->ip_v == 6) {
3134 ip6 = mtod(m, struct ip6_hdr *);
3135 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3136 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3138 /* IPv6 anycast check is done at tcp6_input() */
3141 #if defined(INET) && defined(INET6)
3146 ip = mtod(m, struct ip *);
3147 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3148 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3149 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3150 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3155 /* Perform bandwidth limiting. */
3156 if (badport_bandlim(rstreason) < 0)
3159 /* tcp_respond consumes the mbuf chain. */
3160 if (th->th_flags & TH_ACK) {
3161 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3162 th->th_ack, TH_RST);
3164 if (th->th_flags & TH_SYN)
3166 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3167 (tcp_seq)0, TH_RST|TH_ACK);
3175 * Parse TCP options and place in tcpopt.
3178 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3183 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3185 if (opt == TCPOPT_EOL)
3187 if (opt == TCPOPT_NOP)
3193 if (optlen < 2 || optlen > cnt)
3198 if (optlen != TCPOLEN_MAXSEG)
3200 if (!(flags & TO_SYN))
3202 to->to_flags |= TOF_MSS;
3203 bcopy((char *)cp + 2,
3204 (char *)&to->to_mss, sizeof(to->to_mss));
3205 to->to_mss = ntohs(to->to_mss);
3208 if (optlen != TCPOLEN_WINDOW)
3210 if (!(flags & TO_SYN))
3212 to->to_flags |= TOF_SCALE;
3213 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3215 case TCPOPT_TIMESTAMP:
3216 if (optlen != TCPOLEN_TIMESTAMP)
3218 to->to_flags |= TOF_TS;
3219 bcopy((char *)cp + 2,
3220 (char *)&to->to_tsval, sizeof(to->to_tsval));
3221 to->to_tsval = ntohl(to->to_tsval);
3222 bcopy((char *)cp + 6,
3223 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3224 to->to_tsecr = ntohl(to->to_tsecr);
3226 #ifdef TCP_SIGNATURE
3228 * XXX In order to reply to a host which has set the
3229 * TCP_SIGNATURE option in its initial SYN, we have to
3230 * record the fact that the option was observed here
3231 * for the syncache code to perform the correct response.
3233 case TCPOPT_SIGNATURE:
3234 if (optlen != TCPOLEN_SIGNATURE)
3236 to->to_flags |= TOF_SIGNATURE;
3237 to->to_signature = cp + 2;
3240 case TCPOPT_SACK_PERMITTED:
3241 if (optlen != TCPOLEN_SACK_PERMITTED)
3243 if (!(flags & TO_SYN))
3247 to->to_flags |= TOF_SACKPERM;
3250 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3254 to->to_flags |= TOF_SACK;
3255 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3256 to->to_sacks = cp + 2;
3257 TCPSTAT_INC(tcps_sack_rcv_blocks);
3266 * Pull out of band byte out of a segment so
3267 * it doesn't appear in the user's data queue.
3268 * It is still reflected in the segment length for
3269 * sequencing purposes.
3272 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3275 int cnt = off + th->th_urp - 1;
3278 if (m->m_len > cnt) {
3279 char *cp = mtod(m, caddr_t) + cnt;
3280 struct tcpcb *tp = sototcpcb(so);
3282 INP_WLOCK_ASSERT(tp->t_inpcb);
3285 tp->t_oobflags |= TCPOOB_HAVEDATA;
3286 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3288 if (m->m_flags & M_PKTHDR)
3297 panic("tcp_pulloutofband");
3301 * Collect new round-trip time estimate
3302 * and update averages and current timeout.
3305 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3309 INP_WLOCK_ASSERT(tp->t_inpcb);
3311 TCPSTAT_INC(tcps_rttupdated);
3313 if (tp->t_srtt != 0) {
3315 * srtt is stored as fixed point with 5 bits after the
3316 * binary point (i.e., scaled by 8). The following magic
3317 * is equivalent to the smoothing algorithm in rfc793 with
3318 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3319 * point). Adjust rtt to origin 0.
3321 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3322 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3324 if ((tp->t_srtt += delta) <= 0)
3328 * We accumulate a smoothed rtt variance (actually, a
3329 * smoothed mean difference), then set the retransmit
3330 * timer to smoothed rtt + 4 times the smoothed variance.
3331 * rttvar is stored as fixed point with 4 bits after the
3332 * binary point (scaled by 16). The following is
3333 * equivalent to rfc793 smoothing with an alpha of .75
3334 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3335 * rfc793's wired-in beta.
3339 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3340 if ((tp->t_rttvar += delta) <= 0)
3342 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3343 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3346 * No rtt measurement yet - use the unsmoothed rtt.
3347 * Set the variance to half the rtt (so our first
3348 * retransmit happens at 3*rtt).
3350 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3351 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3352 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3358 * the retransmit should happen at rtt + 4 * rttvar.
3359 * Because of the way we do the smoothing, srtt and rttvar
3360 * will each average +1/2 tick of bias. When we compute
3361 * the retransmit timer, we want 1/2 tick of rounding and
3362 * 1 extra tick because of +-1/2 tick uncertainty in the
3363 * firing of the timer. The bias will give us exactly the
3364 * 1.5 tick we need. But, because the bias is
3365 * statistical, we have to test that we don't drop below
3366 * the minimum feasible timer (which is 2 ticks).
3368 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3369 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3372 * We received an ack for a packet that wasn't retransmitted;
3373 * it is probably safe to discard any error indications we've
3374 * received recently. This isn't quite right, but close enough
3375 * for now (a route might have failed after we sent a segment,
3376 * and the return path might not be symmetrical).
3378 tp->t_softerror = 0;
3382 * Determine a reasonable value for maxseg size.
3383 * If the route is known, check route for mtu.
3384 * If none, use an mss that can be handled on the outgoing interface
3385 * without forcing IP to fragment. If no route is found, route has no mtu,
3386 * or the destination isn't local, use a default, hopefully conservative
3387 * size (usually 512 or the default IP max size, but no more than the mtu
3388 * of the interface), as we can't discover anything about intervening
3389 * gateways or networks. We also initialize the congestion/slow start
3390 * window to be a single segment if the destination isn't local.
3391 * While looking at the routing entry, we also initialize other path-dependent
3392 * parameters from pre-set or cached values in the routing entry.
3394 * Also take into account the space needed for options that we
3395 * send regularly. Make maxseg shorter by that amount to assure
3396 * that we can send maxseg amount of data even when the options
3397 * are present. Store the upper limit of the length of options plus
3400 * NOTE that this routine is only called when we process an incoming
3401 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3402 * settings are handled in tcp_mssopt().
3405 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3406 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3410 struct inpcb *inp = tp->t_inpcb;
3411 struct hc_metrics_lite metrics;
3414 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3415 size_t min_protoh = isipv6 ?
3416 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3417 sizeof (struct tcpiphdr);
3419 const size_t min_protoh = sizeof(struct tcpiphdr);
3422 INP_WLOCK_ASSERT(tp->t_inpcb);
3424 if (mtuoffer != -1) {
3425 KASSERT(offer == -1, ("%s: conflict", __func__));
3426 offer = mtuoffer - min_protoh;
3433 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3434 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3437 #if defined(INET) && defined(INET6)
3442 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3443 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3448 * No route to sender, stay with default mss and return.
3452 * In case we return early we need to initialize metrics
3453 * to a defined state as tcp_hc_get() would do for us
3454 * if there was no cache hit.
3456 if (metricptr != NULL)
3457 bzero(metricptr, sizeof(struct hc_metrics_lite));
3461 /* What have we got? */
3465 * Offer == 0 means that there was no MSS on the SYN
3466 * segment, in this case we use tcp_mssdflt as
3467 * already assigned to t_maxopd above.
3469 offer = tp->t_maxopd;
3474 * Offer == -1 means that we didn't receive SYN yet.
3480 * Prevent DoS attack with too small MSS. Round up
3481 * to at least minmss.
3483 offer = max(offer, V_tcp_minmss);
3487 * rmx information is now retrieved from tcp_hostcache.
3489 tcp_hc_get(&inp->inp_inc, &metrics);
3490 if (metricptr != NULL)
3491 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3494 * If there's a discovered mtu int tcp hostcache, use it
3495 * else, use the link mtu.
3497 if (metrics.rmx_mtu)
3498 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3502 mss = maxmtu - min_protoh;
3503 if (!V_path_mtu_discovery &&
3504 !in6_localaddr(&inp->in6p_faddr))
3505 mss = min(mss, V_tcp_v6mssdflt);
3508 #if defined(INET) && defined(INET6)
3513 mss = maxmtu - min_protoh;
3514 if (!V_path_mtu_discovery &&
3515 !in_localaddr(inp->inp_faddr))
3516 mss = min(mss, V_tcp_mssdflt);
3520 * XXX - The above conditional (mss = maxmtu - min_protoh)
3521 * probably violates the TCP spec.
3522 * The problem is that, since we don't know the
3523 * other end's MSS, we are supposed to use a conservative
3524 * default. But, if we do that, then MTU discovery will
3525 * never actually take place, because the conservative
3526 * default is much less than the MTUs typically seen
3527 * on the Internet today. For the moment, we'll sweep
3528 * this under the carpet.
3530 * The conservative default might not actually be a problem
3531 * if the only case this occurs is when sending an initial
3532 * SYN with options and data to a host we've never talked
3533 * to before. Then, they will reply with an MSS value which
3534 * will get recorded and the new parameters should get
3535 * recomputed. For Further Study.
3538 mss = min(mss, offer);
3541 * Sanity check: make sure that maxopd will be large
3542 * enough to allow some data on segments even if the
3543 * all the option space is used (40bytes). Otherwise
3544 * funny things may happen in tcp_output.
3549 * maxopd stores the maximum length of data AND options
3550 * in a segment; maxseg is the amount of data in a normal
3551 * segment. We need to store this value (maxopd) apart
3552 * from maxseg, because now every segment carries options
3553 * and thus we normally have somewhat less data in segments.
3558 * origoffer==-1 indicates that no segments were received yet.
3559 * In this case we just guess.
3561 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3563 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3564 mss -= TCPOLEN_TSTAMP_APPA;
3570 tcp_mss(struct tcpcb *tp, int offer)
3576 struct hc_metrics_lite metrics;
3577 struct tcp_ifcap cap;
3579 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3581 bzero(&cap, sizeof(cap));
3582 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3588 * If there's a pipesize, change the socket buffer to that size,
3589 * don't change if sb_hiwat is different than default (then it
3590 * has been changed on purpose with setsockopt).
3591 * Make the socket buffers an integral number of mss units;
3592 * if the mss is larger than the socket buffer, decrease the mss.
3594 so = inp->inp_socket;
3595 SOCKBUF_LOCK(&so->so_snd);
3596 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3597 bufsize = metrics.rmx_sendpipe;
3599 bufsize = so->so_snd.sb_hiwat;
3603 bufsize = roundup(bufsize, mss);
3604 if (bufsize > sb_max)
3606 if (bufsize > so->so_snd.sb_hiwat)
3607 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3609 SOCKBUF_UNLOCK(&so->so_snd);
3612 SOCKBUF_LOCK(&so->so_rcv);
3613 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3614 bufsize = metrics.rmx_recvpipe;
3616 bufsize = so->so_rcv.sb_hiwat;
3617 if (bufsize > mss) {
3618 bufsize = roundup(bufsize, mss);
3619 if (bufsize > sb_max)
3621 if (bufsize > so->so_rcv.sb_hiwat)
3622 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3624 SOCKBUF_UNLOCK(&so->so_rcv);
3626 /* Check the interface for TSO capabilities. */
3627 if (cap.ifcap & CSUM_TSO) {
3628 tp->t_flags |= TF_TSO;
3629 tp->t_tsomax = cap.tsomax;
3634 * Determine the MSS option to send on an outgoing SYN.
3637 tcp_mssopt(struct in_conninfo *inc)
3644 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3647 if (inc->inc_flags & INC_ISIPV6) {
3648 mss = V_tcp_v6mssdflt;
3649 maxmtu = tcp_maxmtu6(inc, NULL);
3650 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3653 #if defined(INET) && defined(INET6)
3658 mss = V_tcp_mssdflt;
3659 maxmtu = tcp_maxmtu(inc, NULL);
3660 min_protoh = sizeof(struct tcpiphdr);
3663 #if defined(INET6) || defined(INET)
3664 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3667 if (maxmtu && thcmtu)
3668 mss = min(maxmtu, thcmtu) - min_protoh;
3669 else if (maxmtu || thcmtu)
3670 mss = max(maxmtu, thcmtu) - min_protoh;
3677 * On a partial ack arrives, force the retransmission of the
3678 * next unacknowledged segment. Do not clear tp->t_dupacks.
3679 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3683 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3685 tcp_seq onxt = tp->snd_nxt;
3686 u_long ocwnd = tp->snd_cwnd;
3688 INP_WLOCK_ASSERT(tp->t_inpcb);
3690 tcp_timer_activate(tp, TT_REXMT, 0);
3692 tp->snd_nxt = th->th_ack;
3694 * Set snd_cwnd to one segment beyond acknowledged offset.
3695 * (tp->snd_una has not yet been updated when this function is called.)
3697 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3698 tp->t_flags |= TF_ACKNOW;
3699 (void) tcp_output(tp);
3700 tp->snd_cwnd = ocwnd;
3701 if (SEQ_GT(onxt, tp->snd_nxt))
3704 * Partial window deflation. Relies on fact that tp->snd_una
3707 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3708 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3711 tp->snd_cwnd += tp->t_maxseg;